mke2fs.c revision 1.3
1/* $NetBSD: mke2fs.c,v 1.3 2007/11/19 15:11:50 tsutsui Exp $ */ 2 3/* 4 * Copyright (c) 2007 Izumi Tsutsui. 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 * 3. The name of the author may not be used to endorse or promote products 15 * derived from this software without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29/* 30 * Copyright (c) 1980, 1989, 1993 31 * The Regents of the University of California. All rights reserved. 32 * 33 * Redistribution and use in source and binary forms, with or without 34 * modification, are permitted provided that the following conditions 35 * are met: 36 * 1. Redistributions of source code must retain the above copyright 37 * notice, this list of conditions and the following disclaimer. 38 * 2. Redistributions in binary form must reproduce the above copyright 39 * notice, this list of conditions and the following disclaimer in the 40 * documentation and/or other materials provided with the distribution. 41 * 3. Neither the name of the University nor the names of its contributors 42 * may be used to endorse or promote products derived from this software 43 * without specific prior written permission. 44 * 45 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 48 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 55 * SUCH DAMAGE. 56 */ 57 58/* 59 * Copyright (c) 1997 Manuel Bouyer. 60 * 61 * Redistribution and use in source and binary forms, with or without 62 * modification, are permitted provided that the following conditions 63 * are met: 64 * 1. Redistributions of source code must retain the above copyright 65 * notice, this list of conditions and the following disclaimer. 66 * 2. Redistributions in binary form must reproduce the above copyright 67 * notice, this list of conditions and the following disclaimer in the 68 * documentation and/or other materials provided with the distribution. 69 * 3. All advertising materials mentioning features or use of this software 70 * must display the following acknowledgement: 71 * This product includes software developed by Manuel Bouyer. 72 * 4. The name of the author may not be used to endorse or promote products 73 * derived from this software without specific prior written permission. 74 * 75 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 76 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 77 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 78 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 79 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 80 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 81 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 82 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 83 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 84 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 85 */ 86 87/* 88 * mke2fs.c: "re-invent (dumb but non-GPLed) wheel as a fun project" 89 * 90 * In spite of this name, there is no piece of code 91 * derived from GPLed e2fsprogs written for Linux. 92 * I referred them only to see how each structure 93 * member should be initialized. 94 * 95 * Reference: 96 * - All NetBSD sources under src/sys/ufs/ext2fs and src/sbin/fsck_ext2fs 97 * - Ext2fs Home Page 98 * http://e2fsprogs.sourceforge.net/ext2.html 99 * - Design and Implementation of the Second Extended Filesystem 100 * http://e2fsprogs.sourceforge.net/ext2intro.html 101 * - Linux Documentation "The Second Extended Filesystem" 102 * src/linux/Documentation/filesystems/ext2.txt 103 * in the Linux kernel distribution 104 */ 105 106#include <sys/cdefs.h> 107#ifndef lint 108#if 0 109static char sccsid[] = "@(#)mkfs.c 8.11 (Berkeley) 5/3/95"; 110#else 111__RCSID("$NetBSD: mke2fs.c,v 1.3 2007/11/19 15:11:50 tsutsui Exp $"); 112#endif 113#endif /* not lint */ 114 115#include <sys/param.h> 116#include <sys/mman.h> 117#include <sys/time.h> 118#include <ufs/ext2fs/ext2fs_dinode.h> 119#include <ufs/ext2fs/ext2fs_dir.h> 120#include <ufs/ext2fs/ext2fs.h> 121#include <sys/ioctl.h> 122 123#include <err.h> 124#include <errno.h> 125#include <string.h> 126#include <unistd.h> 127#include <stdlib.h> 128#include <stddef.h> 129#include <stdio.h> 130#include <uuid.h> 131 132#include "extern.h" 133 134static void initcg(uint); 135static void zap_old_sblock(daddr_t); 136static uint cgoverhead(uint); 137static int fsinit(const struct timeval *); 138static int makedir(struct ext2fs_direct *, int); 139static void copy_dir(struct ext2fs_direct *, struct ext2fs_direct *); 140static void init_resizeino(const struct timeval *); 141static uint32_t alloc(uint32_t, uint16_t); 142static void iput(struct ext2fs_dinode *, ino_t); 143static void rdfs(daddr_t, int, void *); 144static void wtfs(daddr_t, int, void *); 145static int ilog2(uint); 146static int skpc(int, size_t, uint8_t *); 147 148/* XXX: some of these macro should be into <ufs/ext2fs/ext2fs.h>? */ 149#define EXT2_DEF_MAX_MNT_COUNT 20 150#define EXT2_DEF_FSCKINTV (180 * 24 * 60 * 60) /* 180 days */ 151#define EXT2_RESERVED_INODES (EXT2_FIRSTINO - 1) 152#define EXT2_UMASK 0755 153 154#define EXT2_INO_INDEX(ino) ((ino) - 1) /* no inode zero */ 155 156#define EXT2_LOSTFOUNDSIZE 16384 157#define EXT2_LOSTFOUNDINO EXT2_FIRSTINO /* XXX: not quite */ 158#define EXT2_LOSTFOUNDUMASK 0700 159 160#define EXT2_RESIZEINOUMASK 0600 161 162#define NBLOCK_SUPERBLOCK 1 163#define NBLOCK_BLOCK_BITMAP 1 164#define NBLOCK_INODE_BITMAP 1 165 166#define cgbase(fs, c) \ 167 ((fs)->e2fs.e2fs_first_dblock + (fs)->e2fs.e2fs_bpg * (c)) 168 169 170/* 171 * ext2fs super block and group descriptor structures 172 * 173 * We don't have to use or setup whole in-memory m_ext2fs structure, 174 * but prepare it to use several macro defined in kernel headers. 175 */ 176union { 177 struct m_ext2fs m_ext2fs; 178 char pad[SBSIZE]; 179} ext2fsun; 180#define sblock ext2fsun.m_ext2fs 181#define gd ext2fsun.m_ext2fs.e2fs_gd 182 183static uint8_t *iobuf; /* for superblock and group descriptors */ 184static int iobufsize; 185 186static uint8_t buf[MAXBSIZE]; /* for initcg() and makedir() ops */ 187 188static int fsi, fso; 189 190void 191mke2fs(const char *fsys, int fi, int fo) 192{ 193 struct timeval tv; 194 int64_t minfssize; 195 uint bcount, fbcount, ficount; 196 uint blocks_gd, blocks_per_cg, inodes_per_cg, iblocks_per_cg; 197 uint minblocks_per_cg, blocks_lastcg; 198 uint ncg, cylno, sboff; 199 uuid_t uuid; 200 uint32_t uustat; 201 int i, len, col, delta, fld_width, max_cols; 202 struct winsize winsize; 203 204 gettimeofday(&tv, NULL); 205 fsi = fi; 206 fso = fo; 207 208 /* 209 * collect and verify the block and fragment sizes 210 */ 211 if (!powerof2(bsize)) { 212 errx(EXIT_FAILURE, 213 "block size must be a power of 2, not %u\n", 214 bsize); 215 } 216 if (!powerof2(fsize)) { 217 errx(EXIT_FAILURE, 218 "fragment size must be a power of 2, not %u\n", 219 fsize); 220 } 221 if (fsize < sectorsize) { 222 errx(EXIT_FAILURE, 223 "fragment size %u is too small, minimum is %u\n", 224 fsize, sectorsize); 225 } 226 if (bsize < MINBSIZE) { 227 errx(EXIT_FAILURE, 228 "block size %u is too small, minimum is %u\n", 229 bsize, MINBSIZE); 230 } 231 if (bsize > MAXBSIZE) { 232 errx(EXIT_FAILURE, 233 "block size %u is too large, maximum is %u\n", 234 bsize, MAXBSIZE); 235 } 236 if (bsize != fsize) { 237 /* 238 * There is no fragment support on current ext2fs (yet?), 239 * but some kernel code refers fsize or fpg as bsize or bpg 240 * and Linux seems to set the same values to them. 241 */ 242 errx(EXIT_FAILURE, 243 "block size (%u) can't be diffrent from " 244 "fragment size (%u)\n", 245 bsize, fsize); 246 } 247 248 sblock.e2fs.e2fs_log_bsize = ilog2(bsize) - LOG_MINBSIZE; 249 /* Umm, why not e2fs_log_fsize? */ 250 sblock.e2fs.e2fs_fsize = ilog2(fsize) - LOG_MINBSIZE; 251 252 sblock.e2fs_bsize = bsize; 253 sblock.e2fs_bshift = sblock.e2fs.e2fs_log_bsize + LOG_MINBSIZE; 254 sblock.e2fs_qbmask = sblock.e2fs_bsize - 1; 255 sblock.e2fs_bmask = ~sblock.e2fs_qbmask; 256 sblock.e2fs_fsbtodb = ilog2(sblock.e2fs_bsize) - ilog2(sectorsize); 257 258 /* 259 * Ext2fs preseves BBSIZE (1024 bytes) space at the top for 260 * bootloader (though it is not enough at all for our bootloader). 261 * If bsize == BBSIZE we have to preserve one block. 262 * If bsize > BBSIZE, the first block already contains BBSIZE space 263 * before superblock because superblock is allocated at SBOFF and 264 * bsize is a power of two (i.e. 2048 bytes or more). 265 */ 266 sblock.e2fs.e2fs_first_dblock = (sblock.e2fs_bsize > BBSIZE) ? 0 : 1; 267 minfssize = fsbtodb(&sblock, 268 sblock.e2fs.e2fs_first_dblock + 269 NBLOCK_SUPERBLOCK + 270 1 /* at least one group descriptor */ + 271 NBLOCK_BLOCK_BITMAP + 272 NBLOCK_INODE_BITMAP + 273 1 /* at least one inode table block */ + 274 1 /* at least one data block for rootdir */ + 275 1 /* at least one data block for data */ 276 ); /* XXX and more? */ 277 278 if (fssize < minfssize) 279 errx(EXIT_FAILURE, "Filesystem size %" PRId64 280 " < minimum size of %" PRId64 "\n", fssize, minfssize); 281 282 bcount = dbtofsb(&sblock, fssize); 283 284 /* 285 * While many people claim that ext2fs is a (bad) clone of ufs/ffs, 286 * it isn't actual ffs so maybe we should call it "block group" 287 * as their native name rather than ffs delived "cylinder group." 288 * But we'll use the latter here since other kernel sources use it. 289 * (I also agree "cylinder" based allocation is obsolete though) 290 */ 291 292 /* maybe "simple is the best" */ 293 blocks_per_cg = sblock.e2fs_bsize * NBBY; 294 295 ncg = howmany(bcount - sblock.e2fs.e2fs_first_dblock, blocks_per_cg); 296 blocks_gd = howmany(sizeof(struct ext2_gd) * ncg, bsize); 297 298 /* check range of inode number */ 299 if (num_inodes < EXT2_FIRSTINO) 300 num_inodes = EXT2_FIRSTINO; /* needs reserved inodes + 1 */ 301 if (num_inodes > UINT16_MAX * ncg) 302 num_inodes = UINT16_MAX * ncg; /* ext2bgd_nifree is uint16_t */ 303 304 inodes_per_cg = num_inodes / ncg; 305 iblocks_per_cg = howmany(EXT2_DINODE_SIZE * inodes_per_cg, bsize); 306 307 /* Check that the last cylinder group has enough space for inodes */ 308 minblocks_per_cg = 309 NBLOCK_BLOCK_BITMAP + 310 NBLOCK_INODE_BITMAP + 311 iblocks_per_cg + 312 1; /* at least one data block */ 313 if (Oflag == 0 || cg_has_sb(ncg - 1) != 0) 314 minblocks_per_cg += NBLOCK_SUPERBLOCK + blocks_gd; 315 316 blocks_lastcg = bcount - sblock.e2fs.e2fs_first_dblock - 317 blocks_per_cg * (ncg - 1); 318 if (blocks_lastcg < minblocks_per_cg) { 319 /* 320 * Since we make all the cylinder groups the same size, the 321 * last will only be small if there are more than one 322 * cylinder groups. If the last one is too small to store 323 * filesystem data, just kill it. 324 * 325 * XXX: Does fsck_ext2fs(8) properly handle this case? 326 */ 327 bcount -= blocks_lastcg; 328 ncg--; 329 blocks_lastcg = blocks_per_cg; 330 blocks_gd = howmany(sizeof(struct ext2_gd) * ncg, bsize); 331 inodes_per_cg = num_inodes / ncg; 332 iblocks_per_cg = 333 howmany(EXT2_DINODE_SIZE * inodes_per_cg, bsize); 334 } 335 /* inodes_per_cg should be a multiple of 8 for bitmap ops */ 336 inodes_per_cg = rounddown(inodes_per_cg, NBBY); 337 num_inodes = inodes_per_cg * ncg; 338 339 /* XXX: probably we should check these adjusted values again */ 340 341 sblock.e2fs.e2fs_bcount = bcount; 342 sblock.e2fs.e2fs_icount = num_inodes; 343 344 sblock.e2fs_ncg = ncg; 345 sblock.e2fs_ngdb = blocks_gd; 346 sblock.e2fs_ipb = sblock.e2fs_bsize / EXT2_DINODE_SIZE; 347 sblock.e2fs_itpg = iblocks_per_cg; 348 349 sblock.e2fs.e2fs_rbcount = sblock.e2fs.e2fs_bcount * minfree / 100; 350 /* e2fs_fbcount will be accounted later */ 351 /* e2fs_ficount will be accounted later */ 352 353 sblock.e2fs.e2fs_bpg = blocks_per_cg; 354 sblock.e2fs.e2fs_fpg = blocks_per_cg; 355 356 sblock.e2fs.e2fs_ipg = inodes_per_cg; 357 358 sblock.e2fs.e2fs_mtime = 0; 359 sblock.e2fs.e2fs_wtime = tv.tv_sec; 360 sblock.e2fs.e2fs_mnt_count = 0; 361 /* XXX: should add some entropy to avoid checking all fs at once? */ 362 sblock.e2fs.e2fs_max_mnt_count = EXT2_DEF_MAX_MNT_COUNT; 363 364 sblock.e2fs.e2fs_magic = E2FS_MAGIC; 365 sblock.e2fs.e2fs_state = E2FS_ISCLEAN; 366 sblock.e2fs.e2fs_beh = E2FS_BEH_DEFAULT; 367 sblock.e2fs.e2fs_minrev = 0; 368 sblock.e2fs.e2fs_lastfsck = tv.tv_sec; 369 sblock.e2fs.e2fs_fsckintv = EXT2_DEF_FSCKINTV; 370 371 /* 372 * Maybe we can use E2FS_OS_FREEBSD here and it would be more proper, 373 * but the purpose of this newfs_ext2fs(8) command is to provide 374 * a filesystem which can be recognized by firmwares on some 375 * Linux based appliances that can load bootstrap files only from 376 * (their native) ext2fs, and anyway we will (and should) try to 377 * act like them as much as possible. 378 * 379 * Anyway, I hope that all newer such boxes will keep their support 380 * for the "GOOD_OLD_REV" ext2fs. 381 */ 382 sblock.e2fs.e2fs_creator = E2FS_OS_LINUX; 383 384 if (Oflag == 0) { 385 sblock.e2fs.e2fs_rev = E2FS_REV0; 386 sblock.e2fs.e2fs_features_compat = 0; 387 sblock.e2fs.e2fs_features_incompat = 0; 388 sblock.e2fs.e2fs_features_rocompat = 0; 389 } else { 390 sblock.e2fs.e2fs_rev = E2FS_REV1; 391 /* 392 * e2fsprogs say "REV1" is "dynamic" so 393 * it isn't quite a version and maybe it means 394 * "extended from REV0 so check compat features." 395 * 396 * XXX: We don't have any native tool to activate 397 * the EXT2F_COMPAT_RESIZE feature and 398 * fsck_ext2fs(8) might not fix structures for it. 399 */ 400 sblock.e2fs.e2fs_features_compat = EXT2F_COMPAT_RESIZE; 401 sblock.e2fs.e2fs_features_incompat = EXT2F_INCOMPAT_FTYPE; 402 sblock.e2fs.e2fs_features_rocompat = 403 EXT2F_ROCOMPAT_SPARSESUPER | EXT2F_ROCOMPAT_LARGEFILE; 404 } 405 406 sblock.e2fs.e2fs_ruid = geteuid(); 407 sblock.e2fs.e2fs_rgid = getegid(); 408 409 sblock.e2fs.e2fs_first_ino = EXT2_FIRSTINO; 410 sblock.e2fs.e2fs_inode_size = EXT2_DINODE_SIZE; 411 412 /* e2fs_block_group_nr is set on writing superblock to each group */ 413 414 uuid_create(&uuid, &uustat); 415 if (uustat != uuid_s_ok) 416 errx(EXIT_FAILURE, "Failed to generate uuid\n"); 417 uuid_enc_le(sblock.e2fs.e2fs_uuid, &uuid); 418 if (volname != NULL) { 419 if (strlen(volname) > sizeof(sblock.e2fs.e2fs_vname)) 420 errx(EXIT_FAILURE, "Volume name is too long"); 421 strlcpy(sblock.e2fs.e2fs_vname, volname, 422 sizeof(sblock.e2fs.e2fs_vname)); 423 } 424 425 sblock.e2fs.e2fs_fsmnt[0] = '\0'; 426 sblock.e2fs_fsmnt[0] = '\0'; 427 428 sblock.e2fs.e2fs_algo = 0; /* XXX unsupported? */ 429 sblock.e2fs.e2fs_prealloc = 0; /* XXX unsupported? */ 430 sblock.e2fs.e2fs_dir_prealloc = 0; /* XXX unsupported? */ 431 432 /* calculate blocks for reserved group descriptors for resize */ 433 sblock.e2fs.e2fs_reserved_ngdb = 0; 434 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 435 (sblock.e2fs.e2fs_features_compat & EXT2F_COMPAT_RESIZE) != 0) { 436 uint64_t target_blocks; 437 uint target_ncg, target_ngdb, reserved_ngdb; 438 439 /* reserve descriptors for size as 1024 times as current */ 440 target_blocks = 441 (sblock.e2fs.e2fs_bcount - sblock.e2fs.e2fs_first_dblock) 442 * 1024ULL; 443 /* number of blocks must be in uint32_t */ 444 if (target_blocks > UINT32_MAX) 445 target_blocks = UINT32_MAX; 446 target_ncg = howmany(target_blocks, sblock.e2fs.e2fs_bpg); 447 target_ngdb = howmany(sizeof(struct ext2_gd) * target_ncg, 448 sblock.e2fs_bsize); 449 /* 450 * Reserved group descriptor blocks are preserved as 451 * the second level double indirect reference blocks in 452 * the EXT2_RESIZEINO inode, so the maximum number of 453 * the blocks is NINDIR(fs). 454 * (see also descriptions in init_resizeino() function) 455 * 456 * We check a number including current e2fs_ngdb here 457 * because they will be moved into reserved gdb on 458 * possible future size shrink, though e2fsprogs don't 459 * seem to care about it. 460 */ 461 if (target_ngdb > NINDIR(&sblock)) 462 target_ngdb = NINDIR(&sblock); 463 464 reserved_ngdb = target_ngdb - sblock.e2fs_ngdb; 465 466 /* make sure reserved_ngdb fits in the last cg */ 467 if (reserved_ngdb >= blocks_lastcg - cgoverhead(ncg - 1)) 468 reserved_ngdb = blocks_lastcg - cgoverhead(ncg - 1); 469 if (reserved_ngdb == 0) { 470 /* if no space for reserved gdb, disable the feature */ 471 sblock.e2fs.e2fs_features_compat &= 472 ~EXT2F_COMPAT_RESIZE; 473 } 474 sblock.e2fs.e2fs_reserved_ngdb = reserved_ngdb; 475 } 476 477 /* 478 * Initialize group descriptors 479 */ 480 gd = malloc(sblock.e2fs_ngdb * bsize); 481 if (gd == NULL) 482 errx(EXIT_FAILURE, "Can't allocate descriptors buffer"); 483 memset(gd, 0, sblock.e2fs_ngdb * bsize); 484 485 fbcount = 0; 486 ficount = 0; 487 for (cylno = 0; cylno < ncg; cylno++) { 488 uint boffset; 489 490 boffset = cgbase(&sblock, cylno); 491 if (sblock.e2fs.e2fs_rev == E2FS_REV0 || 492 (sblock.e2fs.e2fs_features_rocompat & 493 EXT2F_ROCOMPAT_SPARSESUPER) == 0 || 494 cg_has_sb(cylno)) { 495 boffset += NBLOCK_SUPERBLOCK + sblock.e2fs_ngdb; 496 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 497 (sblock.e2fs.e2fs_features_compat & 498 EXT2F_COMPAT_RESIZE) != 0) 499 boffset += sblock.e2fs.e2fs_reserved_ngdb; 500 } 501 gd[cylno].ext2bgd_b_bitmap = boffset; 502 boffset += NBLOCK_BLOCK_BITMAP; 503 gd[cylno].ext2bgd_i_bitmap = boffset; 504 boffset += NBLOCK_INODE_BITMAP; 505 gd[cylno].ext2bgd_i_tables = boffset; 506 if (cylno == (ncg - 1)) 507 gd[cylno].ext2bgd_nbfree = 508 blocks_lastcg - cgoverhead(cylno); 509 else 510 gd[cylno].ext2bgd_nbfree = 511 sblock.e2fs.e2fs_bpg - cgoverhead(cylno); 512 fbcount += gd[cylno].ext2bgd_nbfree; 513 gd[cylno].ext2bgd_nifree = sblock.e2fs.e2fs_ipg; 514 if (cylno == 0) { 515 /* take reserved inodes off nifree */ 516 gd[cylno].ext2bgd_nifree -= EXT2_RESERVED_INODES; 517 } 518 ficount += 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\n"); 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[] = 558 { 1024, 2048, 4096, 8192, 16384, 0 }; 559 uint pblock, epblock; 560 /* 561 * Validate the given file system size. 562 * Verify that its last block can actually be accessed. 563 * Convert to file system fragment sized units. 564 */ 565 if (fssize <= 0) 566 errx(EXIT_FAILURE, "Preposterous size %" PRId64 "\n", 567 fssize); 568 wtfs(fssize - 1, sectorsize, iobuf); 569 570 /* 571 * Ensure there is nothing that looks like a filesystem 572 * superbock anywhere other than where ours will be. 573 * If fsck_ext2fs finds the wrong one all hell breaks loose! 574 * 575 * XXX: needs to check how fsck_ext2fs programs even 576 * on other OSes determine alternate superblocks 577 */ 578 for (i = 0; pbsize[i] != 0; i++) { 579 epblock = (uint64_t)bcount * bsize / pbsize[i]; 580 for (pblock = ((pbsize[i] == SBSIZE) ? 1 : 0); 581 pblock < epblock; 582 pblock += pbsize[i] * NBBY /* bpg */) 583 zap_old_sblock((daddr_t)pblock * 584 pbsize[i] / sectorsize); 585 } 586 } 587 588 if (verbosity >= 3) 589 printf("super-block backups (for fsck_ext2fs -b #) at:\n"); 590 /* If we are printing more than one line of numbers, line up columns */ 591 fld_width = verbosity < 4 ? 1 : snprintf(NULL, 0, "%" PRIu64, 592 (uint64_t)cgbase(&sblock, ncg - 1)); 593 /* Get terminal width */ 594 if (ioctl(fileno(stdout), TIOCGWINSZ, &winsize) == 0) 595 max_cols = winsize.ws_col; 596 else 597 max_cols = 80; 598 if (Nflag && verbosity == 3) 599 /* Leave space to add " ..." after one row of numbers */ 600 max_cols -= 4; 601#define BASE 0x10000 /* For some fixed-point maths */ 602 col = 0; 603 delta = verbosity > 2 ? 0 : max_cols * BASE / ncg; 604 for (cylno = 0; cylno < ncg; cylno++) { 605 fflush(stdout); 606 initcg(cylno); 607 if (verbosity < 2) 608 continue; 609 /* the first one is a master, not backup */ 610 if (cylno == 0) 611 continue; 612 /* skip if this cylinder doesn't have a backup */ 613 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 614 (sblock.e2fs.e2fs_features_rocompat & 615 EXT2F_ROCOMPAT_SPARSESUPER) != 0 && 616 cg_has_sb(cylno) == 0) 617 continue; 618 619 if (delta > 0) { 620 if (Nflag) 621 /* No point doing dots for -N */ 622 break; 623 /* Print dots scaled to end near RH margin */ 624 for (col += delta; col > BASE; col -= BASE) 625 printf("."); 626 continue; 627 } 628 /* Print superblock numbers */ 629 len = printf(" %*" PRIu64 "," + !col, fld_width, 630 (uint64_t)cgbase(&sblock, cylno)); 631 col += len; 632 if (col + len < max_cols) 633 /* Next number fits */ 634 continue; 635 /* Next number won't fit, need a newline */ 636 if (verbosity <= 3) { 637 /* Print dots for subsequent cylinder groups */ 638 delta = sblock.e2fs_ncg - cylno - 1; 639 if (delta != 0) { 640 if (Nflag) { 641 printf(" ..."); 642 break; 643 } 644 delta = max_cols * BASE / delta; 645 } 646 } 647 col = 0; 648 printf("\n"); 649 } 650#undef BASE 651 if (col > 0) 652 printf("\n"); 653 if (Nflag) 654 return; 655 656 /* 657 * Now construct the initial file system, 658 */ 659 if (fsinit(&tv) == 0) 660 errx(EXIT_FAILURE, "Error making filesystem"); 661 /* 662 * Write out the superblock and group descriptors 663 */ 664 sblock.e2fs.e2fs_block_group_nr = 0; 665 sboff = 0; 666 if (cgbase(&sblock, 0) == 0) { 667 /* 668 * If the first block contains the boot block sectors, 669 * (i.e. in case of sblock.e2fs.e2fs_bsize > BBSIZE) 670 * we have to preserve data in it. 671 */ 672 sboff = SBOFF; 673 } 674 e2fs_sbsave(&sblock.e2fs, (struct ext2fs *)(iobuf + sboff)); 675 e2fs_cgsave(gd, (struct ext2_gd *)(iobuf + sblock.e2fs_bsize), 676 sizeof(struct ext2_gd) * sblock.e2fs_ncg); 677 wtfs(fsbtodb(&sblock, cgbase(&sblock, 0)) + sboff / sectorsize, 678 iobufsize - sboff, iobuf + sboff); 679 680 munmap(iobuf, iobufsize); 681} 682 683/* 684 * Initialize a cylinder (block) group. 685 */ 686void 687initcg(uint cylno) 688{ 689 uint nblcg, i, j, sboff; 690 struct ext2fs_dinode *dp; 691 692 /* 693 * Make a copy of the superblock and group descriptors. 694 */ 695 if (sblock.e2fs.e2fs_rev == E2FS_REV0 || 696 (sblock.e2fs.e2fs_features_rocompat & 697 EXT2F_ROCOMPAT_SPARSESUPER) == 0 || 698 cg_has_sb(cylno)) { 699 sblock.e2fs.e2fs_block_group_nr = cylno; 700 sboff = 0; 701 if (cgbase(&sblock, cylno) == 0) { 702 /* preserve data in bootblock in cg0 */ 703 sboff = SBOFF; 704 } 705 e2fs_sbsave(&sblock.e2fs, (struct ext2fs *)(iobuf + sboff)); 706 e2fs_cgsave(gd, (struct ext2_gd *)(iobuf + 707 sblock.e2fs_bsize * NBLOCK_SUPERBLOCK), 708 sizeof(struct ext2_gd) * sblock.e2fs_ncg); 709 /* write superblock and group descriptor backups */ 710 wtfs(fsbtodb(&sblock, cgbase(&sblock, cylno)) + 711 sboff / sectorsize, iobufsize - sboff, iobuf + sboff); 712 } 713 714 /* 715 * Initialize block bitmap. 716 */ 717 memset(buf, 0, sblock.e2fs_bsize); 718 if (cylno == (sblock.e2fs_ncg - 1)) { 719 /* The last group could have less blocks than e2fs_bpg. */ 720 nblcg = sblock.e2fs.e2fs_bcount - 721 cgbase(&sblock, sblock.e2fs_ncg - 1); 722 for (i = nblcg; i < roundup(nblcg, NBBY); i++) 723 setbit(buf, i); 724 memset(&buf[i / NBBY], ~0U, sblock.e2fs.e2fs_bpg - i); 725 } 726 /* set overhead (superblock, group descriptor etc.) blocks used */ 727 for (i = 0; i < cgoverhead(cylno) / NBBY; i++) 728 buf[i] = ~0; 729 i = i * NBBY; 730 for (; i < cgoverhead(cylno); i++) 731 setbit(buf, i); 732 wtfs(fsbtodb(&sblock, gd[cylno].ext2bgd_b_bitmap), sblock.e2fs_bsize, 733 buf); 734 735 /* 736 * Initialize inode bitmap. 737 * 738 * Assume e2fs_ipg is a multiple of NBBY (as we did above). 739 * Note even (possibly smaller) the last group has the same e2fs_ipg. 740 */ 741 i = sblock.e2fs.e2fs_ipg / NBBY; 742 memset(buf, 0, i); 743 memset(buf + i, ~0U, sblock.e2fs_bsize - i); 744 if (cylno == 0) { 745 /* mark reserved inodes */ 746 for (i = 1; i < EXT2_FIRSTINO; i++) 747 setbit(buf, EXT2_INO_INDEX(i)); 748 } 749 wtfs(fsbtodb(&sblock, gd[cylno].ext2bgd_i_bitmap), sblock.e2fs_bsize, 750 buf); 751 752 /* 753 * Initialize inode tables. 754 * 755 * Just initialize generation numbers for NFS security. 756 * XXX: sys/ufs/ext2fs/ext2fs_alloc.c:ext2fs_valloc() seems 757 * to override these generated numbers. 758 */ 759 memset(buf, 0, sblock.e2fs_bsize); 760 dp = (struct ext2fs_dinode *)buf; 761 for (i = 0; i < sblock.e2fs_itpg; i++) { 762 for (j = 0; j < sblock.e2fs_ipb; j++) { 763 /* h2fs32() just for consistency */ 764 dp[j].e2di_gen = h2fs32(arc4random()); 765 } 766 wtfs(fsbtodb(&sblock, gd[cylno].ext2bgd_i_tables + i), 767 sblock.e2fs_bsize, buf); 768 } 769} 770 771/* 772 * Zap possible lingering old superblock data 773 */ 774static void 775zap_old_sblock(daddr_t sec) 776{ 777 static daddr_t cg0_data; 778 uint32_t oldfs[SBSIZE / sizeof(uint32_t)]; 779 static const struct fsm { 780 uint32_t offset; 781 uint32_t magic; 782 uint32_t mask; 783 } fs_magics[] = { 784 {offsetof(struct ext2fs, e2fs_magic) / 4, E2FS_MAGIC, 0xffff}, 785 {offsetof(struct ext2fs, e2fs_magic) / 4, 786 E2FS_MAGIC << 16, 0xffff0000}, 787 {14, 0xef530000, 0xffff0000}, /* EXT2FS (big) */ 788 {0x55c / 4, 0x00011954, ~0U}, /* FS_UFS1_MAGIC */ 789 {0x55c / 4, 0x19540119, ~0U}, /* FS_UFS2_MAGIC */ 790 {0, 0x70162, ~0U}, /* LFS_MAGIC */ 791 {.offset = ~0U}, 792 }; 793 const struct fsm *fsm; 794 795 if (Nflag) 796 return; 797 798 /* don't override data before superblock */ 799 if (sec < SBOFF / sectorsize) 800 return; 801 802 if (cg0_data == 0) { 803 cg0_data = 804 ((daddr_t)sblock.e2fs.e2fs_first_dblock + cgoverhead(0)) * 805 sblock.e2fs_bsize / sectorsize; 806 } 807 808 /* Ignore anything that is beyond our filesystem */ 809 if (sec >= fssize) 810 return; 811 /* Zero anything inside our filesystem... */ 812 if (sec >= sblock.e2fs.e2fs_first_dblock * bsize / sectorsize) { 813 /* ...unless we will write that area anyway */ 814 if (sec >= cg0_data) 815 /* assume iobuf is zero'ed here */ 816 wtfs(sec, roundup(SBSIZE, sectorsize), iobuf); 817 return; 818 } 819 820 /* 821 * The sector might contain boot code, so we must validate it 822 * 823 * XXX: ext2fs won't preserve data after SBOFF, 824 * but first_dblock could have a differnt value. 825 */ 826 rdfs(sec, sizeof(oldfs), &oldfs); 827 for (fsm = fs_magics;; fsm++) { 828 uint32_t v; 829 if (fsm->mask == 0) 830 return; 831 v = oldfs[fsm->offset]; 832 if ((v & fsm->mask) == fsm->magic || 833 (bswap32(v) & fsm->mask) == fsm->magic) 834 break; 835 } 836 837 /* Just zap the magic number */ 838 oldfs[fsm->offset] = 0; 839 wtfs(sec, sizeof(oldfs), &oldfs); 840} 841 842/* 843 * uint cgoverhead(uint c) 844 * 845 * Return a number of reserved blocks on the specified group. 846 * XXX: should be shared with src/sbin/newfs_ext2fs/setup.c 847 */ 848uint 849cgoverhead(uint c) 850{ 851 uint overh; 852 853 overh = NBLOCK_BLOCK_BITMAP + NBLOCK_INODE_BITMAP + sblock.e2fs_itpg; 854 855 if (sblock.e2fs.e2fs_rev == E2FS_REV0 || 856 (sblock.e2fs.e2fs_features_rocompat & 857 EXT2F_ROCOMPAT_SPARSESUPER) == 0 || 858 cg_has_sb(c) != 0) { 859 overh += NBLOCK_SUPERBLOCK + sblock.e2fs_ngdb; 860 861 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 862 (sblock.e2fs.e2fs_features_compat & 863 EXT2F_COMPAT_RESIZE) != 0) 864 overh += sblock.e2fs.e2fs_reserved_ngdb; 865 } 866 867 return overh; 868} 869 870/* 871 * Initialize the file system 872 */ 873 874#define LOSTDIR /* e2fsck complains if there is no lost+found */ 875 876#define PREDEFDIR 2 877 878#ifdef LOSTDIR 879#define PREDEFROOTDIR (PREDEFDIR + 1) 880#else 881#define PREDEFROOTDIR PREDEFDIR 882#endif 883 884struct ext2fs_direct root_dir[] = { 885 { EXT2_ROOTINO, 0, 1, 0, "." }, 886 { EXT2_ROOTINO, 0, 2, 0, ".." }, 887#ifdef LOSTDIR 888 { EXT2_LOSTFOUNDINO, 0, 10, 0, "lost+found" }, 889#endif 890}; 891 892#ifdef LOSTDIR 893struct ext2fs_direct lost_found_dir[] = { 894 { EXT2_LOSTFOUNDINO, 0, 1, 0, "." }, 895 { EXT2_ROOTINO, 0, 2, 0, ".." }, 896}; 897struct ext2fs_direct pad_dir = { 0, sizeof(struct ext2fs_direct), 0, 0, "" }; 898#endif 899 900int 901fsinit(const struct timeval *tv) 902{ 903 struct ext2fs_dinode node; 904#ifdef LOSTDIR 905 uint i, nblks_lostfound, blk; 906#endif 907 908 /* 909 * Initialize the inode for the resizefs feature 910 */ 911 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 912 (sblock.e2fs.e2fs_features_compat & EXT2F_COMPAT_RESIZE) != 0) 913 init_resizeino(tv); 914 915 /* 916 * Initialize the node 917 */ 918 919#ifdef LOSTDIR 920 /* 921 * Create the lost+found directory 922 */ 923 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 924 sblock.e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE) { 925 lost_found_dir[0].e2d_type = EXT2_FT_DIR; 926 lost_found_dir[1].e2d_type = EXT2_FT_DIR; 927 } 928 (void)makedir(lost_found_dir, __arraycount(lost_found_dir)); 929 930 /* prepare a bit large directory for preserved files */ 931 nblks_lostfound = EXT2_LOSTFOUNDSIZE / sblock.e2fs_bsize; 932 /* ...but only with direct blocks */ 933 if (nblks_lostfound > NDADDR) 934 nblks_lostfound = NDADDR; 935 936 memset(&node, 0, sizeof(node)); 937 node.e2di_mode = EXT2_IFDIR | EXT2_LOSTFOUNDUMASK; 938 node.e2di_uid = geteuid(); 939 node.e2di_size = sblock.e2fs_bsize * nblks_lostfound; 940 node.e2di_atime = tv->tv_sec; 941 node.e2di_ctime = tv->tv_sec; 942 node.e2di_mtime = tv->tv_sec; 943 node.e2di_gid = getegid(); 944 node.e2di_nlink = PREDEFDIR; 945 /* e2di_nblock is a number of disk block, not ext2fs block */ 946 node.e2di_nblock = fsbtodb(&sblock, nblks_lostfound); 947 node.e2di_blocks[0] = alloc(sblock.e2fs_bsize, node.e2di_mode); 948 if (node.e2di_blocks[0] == 0) { 949 printf("%s: can't allocate block for lost+found\n", __func__); 950 return 0; 951 } 952 for (i = 1; i < nblks_lostfound; i++) { 953 blk = alloc(sblock.e2fs_bsize, 0); 954 if (blk == 0) { 955 printf("%s: can't allocate blocks for lost+found\n", 956 __func__); 957 return 0; 958 } 959 node.e2di_blocks[i] = blk; 960 } 961 wtfs(fsbtodb(&sblock, node.e2di_blocks[0]), sblock.e2fs_bsize, buf); 962 pad_dir.e2d_reclen = sblock.e2fs_bsize; 963 for (i = 1; i < nblks_lostfound; i++) { 964 memset(buf, 0, sblock.e2fs_bsize); 965 copy_dir(&pad_dir, (struct ext2fs_direct *)buf); 966 wtfs(fsbtodb(&sblock, node.e2di_blocks[i]), sblock.e2fs_bsize, 967 buf); 968 } 969 iput(&node, EXT2_LOSTFOUNDINO); 970#endif 971 /* 972 * create the root directory 973 */ 974 memset(&node, 0, sizeof(node)); 975 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 976 sblock.e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE) { 977 root_dir[0].e2d_type = EXT2_FT_DIR; 978 root_dir[1].e2d_type = EXT2_FT_DIR; 979#ifdef LOSTDIR 980 root_dir[2].e2d_type = EXT2_FT_DIR; 981#endif 982 } 983 node.e2di_mode = EXT2_IFDIR | EXT2_UMASK; 984 node.e2di_uid = geteuid(); 985 node.e2di_size = makedir(root_dir, __arraycount(root_dir)); 986 node.e2di_atime = tv->tv_sec; 987 node.e2di_ctime = tv->tv_sec; 988 node.e2di_mtime = tv->tv_sec; 989 node.e2di_gid = getegid(); 990 node.e2di_nlink = PREDEFROOTDIR; 991 /* e2di_nblock is a number of disk block, not ext2fs block */ 992 node.e2di_nblock = fsbtodb(&sblock, 1); 993 node.e2di_blocks[0] = alloc(node.e2di_size, node.e2di_mode); 994 if (node.e2di_blocks[0] == 0) { 995 printf("%s: can't allocate block for root dir\n", __func__); 996 return 0; 997 } 998 wtfs(fsbtodb(&sblock, node.e2di_blocks[0]), sblock.e2fs_bsize, buf); 999 iput(&node, EXT2_ROOTINO); 1000 return 1; 1001} 1002 1003/* 1004 * Construct a set of directory entries in "buf". 1005 * return size of directory. 1006 */ 1007int 1008makedir(struct ext2fs_direct *protodir, int entries) 1009{ 1010 uint8_t *cp; 1011 uint i, spcleft; 1012 uint dirblksiz; 1013 1014 dirblksiz = sblock.e2fs_bsize; 1015 memset(buf, 0, dirblksiz); 1016 spcleft = dirblksiz; 1017 for (cp = buf, i = 0; i < entries - 1; i++) { 1018 protodir[i].e2d_reclen = EXT2FS_DIRSIZ(protodir[i].e2d_namlen); 1019 copy_dir(&protodir[i], (struct ext2fs_direct *)cp); 1020 cp += protodir[i].e2d_reclen; 1021 spcleft -= protodir[i].e2d_reclen; 1022 } 1023 protodir[i].e2d_reclen = spcleft; 1024 copy_dir(&protodir[i], (struct ext2fs_direct *)cp); 1025 return dirblksiz; 1026} 1027 1028/* 1029 * Copy a direntry to a buffer, in fs byte order 1030 */ 1031static void 1032copy_dir(struct ext2fs_direct *dir, struct ext2fs_direct *dbuf) 1033{ 1034 1035 memcpy(dbuf, dir, EXT2FS_DIRSIZ(dir->e2d_namlen)); 1036 dbuf->e2d_ino = h2fs32(dir->e2d_ino); 1037 dbuf->e2d_reclen = h2fs16(dir->e2d_reclen); 1038} 1039 1040/* 1041 * void init_resizeino(const struct timeval *tv); 1042 * 1043 * Initialize the EXT2_RESEIZE_INO inode to prereserve 1044 * reserved group descriptor blocks for future growth of this ext2fs. 1045 */ 1046void 1047init_resizeino(const struct timeval *tv) 1048{ 1049 struct ext2fs_dinode node; 1050 uint64_t isize; 1051 uint32_t *dindir_block, *reserved_gdb; 1052 uint nblock, i, cylno, n; 1053 1054 memset(&node, 0, sizeof(node)); 1055 1056 /* 1057 * Note this function only prepares required structures for 1058 * future resize. It's a quite different work to implement 1059 * a utility like resize_ext2fs(8) which handles actual 1060 * resize ops even on offline. 1061 * 1062 * Anyway, I'm not sure if there is any documentation about 1063 * this resize ext2fs feature and related data structures, 1064 * and I've written this function based on things what I see 1065 * on some existing implementation and real file system data 1066 * created by existing tools. To be honest, they are not 1067 * so easy to read, so I will try to implement it here without 1068 * any dumb optimization for people who would eventually 1069 * work on "yet another wheel" like resize_ext2fs(8). 1070 */ 1071 1072 /* 1073 * I'm not sure what type is appropriate for this inode. 1074 * The release notes of e2fsprogs says they changed e2fsck to allow 1075 * IFREG for RESIZEINO since a certain resize tool used it. Hmm. 1076 */ 1077 node.e2di_mode = EXT2_IFREG | EXT2_RESIZEINOUMASK; 1078 node.e2di_uid = geteuid(); 1079 node.e2di_atime = tv->tv_sec; 1080 node.e2di_ctime = tv->tv_sec; 1081 node.e2di_mtime = tv->tv_sec; 1082 node.e2di_gid = getegid(); 1083 node.e2di_nlink = 1; 1084 1085 /* 1086 * To preserve the reserved group descriptor blocks, 1087 * EXT2_RESIZEINO uses only double indirect reference 1088 * blocks in its inode entries. 1089 * 1090 * All entries for direct, single indirect and triple 1091 * indirect references are left zero'ed. Maybe it's safe 1092 * because no write operation will happen with this inode. 1093 * 1094 * We have to allocate a block for the first level double 1095 * indirect reference block. Indexes of inode entries in 1096 * this first level dindirect block are corresponding to 1097 * indexes of group descriptors including both used (e2fs_ngdb) 1098 * and reserved (e2fs_reserved_ngdb) group descriptor blocks. 1099 * 1100 * Inode entries of indexes for used (e2fs_ngdb) descriptors are 1101 * left zero'ed. Entries for reserved (e2fs_reserved_ngdb) ones 1102 * have block numbers of actual reserved group descriptors 1103 * allocated at block group zero. This means e2fs_reserved_ngdb 1104 * blocks are reserved as the second level dindirect reference 1105 * blocks, and they acutually contain block numbers of indirect 1106 * references. It may be safe since they don't have to keep any 1107 * data yet. 1108 * 1109 * Each these second dindirect blocks (i.e. reserved group 1110 * descriptor blocks in the first block group) should have 1111 * block numbers of its backups in all other block groups. 1112 * I.e. reserved_ngdb[0] block in block group 0 contains block 1113 * numbers of resreved_ngdb[0] from group 1 through (e2fs_ncg - 1). 1114 * The number of backups can be determined by the 1115 * EXT2_ROCOMPAT_SPARSESUPER feature and cg_has_sb() macro 1116 * as done in the above initcg() function. 1117 */ 1118 1119 /* set e2di_size which occupies whole blocks through DINDIR blocks */ 1120 isize = (uint64_t)sblock.e2fs_bsize * NDADDR + 1121 (uint64_t)sblock.e2fs_bsize * NINDIR(&sblock) + 1122 (uint64_t)sblock.e2fs_bsize * NINDIR(&sblock) * NINDIR(&sblock); 1123 if (isize > UINT32_MAX && 1124 (sblock.e2fs.e2fs_features_rocompat & 1125 EXT2F_ROCOMPAT_LARGEFILE) == 0) { 1126 /* XXX should enable it here and update all backups? */ 1127 errx(EXIT_FAILURE, "%s: large_file rocompat feature is " 1128 "required to enable resize feature for this filesystem\n", 1129 __func__); 1130 } 1131 /* upper 32bit is stored into e2di_dacl on REV1 feature */ 1132 node.e2di_size = isize & UINT32_MAX; 1133 node.e2di_dacl = isize >> 32; 1134 1135#define SINGLE 0 /* index of single indirect block */ 1136#define DOUBLE 1 /* index of double indirect block */ 1137#define TRIPLE 2 /* index of triple indirect block */ 1138 1139 /* zero out entries for direct references */ 1140 for (i = 0; i < NDADDR; i++) 1141 node.e2di_blocks[i] = 0; 1142 /* also zero out entries for single and triple indirect references */ 1143 node.e2di_blocks[NDADDR + SINGLE] = 0; 1144 node.e2di_blocks[NDADDR + TRIPLE] = 0; 1145 1146 /* allocate a block for the first level double indirect reference */ 1147 node.e2di_blocks[NDADDR + DOUBLE] = 1148 alloc(sblock.e2fs_bsize, node.e2di_mode); 1149 if (node.e2di_blocks[NDADDR + DOUBLE] == 0) 1150 errx(EXIT_FAILURE, "%s: Can't allocate a dindirect block", 1151 __func__); 1152 1153 /* account this first block */ 1154 nblock = fsbtodb(&sblock, 1); 1155 1156 /* allocate buffer to set data in the dindirect block */ 1157 dindir_block = malloc(sblock.e2fs_bsize); 1158 if (dindir_block == NULL) 1159 errx(EXIT_FAILURE, 1160 "%s: Can't allocate buffer for a dindirect block", 1161 __func__); 1162 1163 /* allocate buffer to set data in the group descriptor blocks */ 1164 reserved_gdb = malloc(sblock.e2fs_bsize); 1165 if (reserved_gdb == NULL) 1166 errx(EXIT_FAILURE, 1167 "%s: Can't allocate buffer for group descriptor blocks", 1168 __func__); 1169 1170 /* 1171 * Setup block entries in the first level dindirect blocks 1172 */ 1173 for (i = 0; i < sblock.e2fs_ngdb; i++) { 1174 /* no need to handle used group descriptor blocks */ 1175 dindir_block[i] = 0; 1176 } 1177 for (; i < sblock.e2fs_ngdb + sblock.e2fs.e2fs_reserved_ngdb; i++) { 1178 /* 1179 * point reserved group descriptor block in the first 1180 * (i.e. master) block group 1181 * 1182 * XXX: e2fsprogs seem to use "(i % NINDIR(&sblock))" here 1183 * to store maximum NINDIR(&sblock) reserved gdbs. 1184 * I'm not sure what will be done on future filesystem 1185 * shrink in that case on their way. 1186 */ 1187 if (i >= NINDIR(&sblock)) 1188 errx(EXIT_FAILURE, "%s: too many reserved " 1189 "group descriptors (%u) for resize inode", 1190 __func__, sblock.e2fs.e2fs_reserved_ngdb); 1191 dindir_block[i] = 1192 h2fs32(cgbase(&sblock, 0) + NBLOCK_SUPERBLOCK + i); 1193 1194 /* 1195 * Setup block entries in the second dindirect blocks 1196 * (which are primary reserved group descriptor blocks) 1197 * to point their backups. 1198 */ 1199 for (n = 0, cylno = 1; cylno < sblock.e2fs_ncg; cylno++) { 1200 /* skip block groups without backup */ 1201 if ((sblock.e2fs.e2fs_features_rocompat & 1202 EXT2F_ROCOMPAT_SPARSESUPER) != 0 && 1203 cg_has_sb(cylno) == 0) 1204 continue; 1205 1206 if (n >= NINDIR(&sblock)) 1207 errx(EXIT_FAILURE, "%s: too many block groups " 1208 "for the resize feature", __func__); 1209 /* 1210 * These blocks are already reserved in 1211 * initcg() so no need to use alloc() here. 1212 */ 1213 reserved_gdb[n++] = h2fs32(cgbase(&sblock, cylno) + 1214 NBLOCK_SUPERBLOCK + i); 1215 nblock += fsbtodb(&sblock, 1); 1216 } 1217 for (; n < NINDIR(&sblock); n++) 1218 reserved_gdb[n] = 0; 1219 1220 /* write group descriptor block as the second dindirect refs */ 1221 wtfs(fsbtodb(&sblock, fs2h32(dindir_block[i])), 1222 sblock.e2fs_bsize, reserved_gdb); 1223 nblock += fsbtodb(&sblock, 1); 1224 } 1225 for (; i < NINDIR(&sblock); i++) { 1226 /* leave trailing entries unallocated */ 1227 dindir_block[i] = 0; 1228 } 1229 free(reserved_gdb); 1230 1231 /* finally write the first level dindirect block */ 1232 wtfs(fsbtodb(&sblock, node.e2di_blocks[NDADDR + DOUBLE]), 1233 sblock.e2fs_bsize, dindir_block); 1234 free(dindir_block); 1235 1236 node.e2di_nblock = nblock; 1237 iput(&node, EXT2_RESIZEINO); 1238} 1239 1240/* 1241 * uint32_t alloc(uint32_t size, uint16_t mode) 1242 * 1243 * Allocate a block (from cylinder group 0) 1244 * Reference: src/sys/ufs/ext2fs/ext2fs_alloc.c:ext2fs_alloccg() 1245 */ 1246uint32_t 1247alloc(uint32_t size, uint16_t mode) 1248{ 1249 uint32_t loc, bno; 1250 uint8_t *bbp; 1251 uint len, map, i; 1252 1253 if (gd[0].ext2bgd_nbfree == 0) 1254 return 0; 1255 1256 if (size > sblock.e2fs_bsize) 1257 return 0; 1258 1259 bbp = malloc(sblock.e2fs_bsize); 1260 if (bbp == NULL) 1261 return 0; 1262 rdfs(fsbtodb(&sblock, gd[0].ext2bgd_b_bitmap), sblock.e2fs_bsize, bbp); 1263 1264 /* XXX: kernel uses e2fs_fpg here */ 1265 len = sblock.e2fs.e2fs_bpg / NBBY; 1266 1267#if 0 /* no need block allocation for root or lost+found dir */ 1268 for (loc = 0; loc < len; loc++) { 1269 if (bbp[loc] == 0) { 1270 bno = loc * NBBY; 1271 goto gotit; 1272 } 1273 } 1274#endif 1275 1276 loc = skpc(~0U, len, bbp); 1277 if (loc == 0) 1278 return 0; 1279 loc = len - loc; 1280 map = bbp[loc]; 1281 bno = loc * NBBY; 1282 for (i = 0; i < NBBY; i++, bno++) { 1283 if ((map & (1 << i)) == 0) 1284 goto gotit; 1285 } 1286 return 0; 1287 1288 gotit: 1289 if (isset(bbp, bno)) 1290 errx(EXIT_FAILURE, "%s: inconsistent bitmap\n", __func__); 1291 1292 setbit(bbp, bno); 1293 wtfs(fsbtodb(&sblock, gd[0].ext2bgd_b_bitmap), sblock.e2fs_bsize, bbp); 1294 free(bbp); 1295 /* XXX: modified group descriptors won't be written into backups */ 1296 gd[0].ext2bgd_nbfree--; 1297 if ((mode & EXT2_IFDIR) != 0) 1298 gd[0].ext2bgd_ndirs++; 1299 sblock.e2fs.e2fs_fbcount--; 1300 1301 return sblock.e2fs.e2fs_first_dblock + bno; 1302} 1303 1304/* 1305 * void iput(struct ext2fs_dinode *ip, ino_t ino) 1306 * 1307 * Put an inode entry into the corresponding table. 1308 */ 1309static void 1310iput(struct ext2fs_dinode *ip, ino_t ino) 1311{ 1312 daddr_t d; 1313 uint c, i; 1314 struct ext2fs_dinode *dp; 1315 uint8_t *bp; 1316 1317 bp = malloc(sblock.e2fs_bsize); 1318 if (bp == NULL) 1319 errx(EXIT_FAILURE, "%s: can't allocate buffer for inode\n", 1320 __func__); 1321 1322 /* 1323 * Reserved inodes are allocated and accounted in initcg() 1324 * so skip checks of the bitmap and allocation for them. 1325 */ 1326 if (ino >= EXT2_FIRSTINO) { 1327 c = ino_to_cg(&sblock, ino); 1328 1329 /* sanity check */ 1330 if (gd[c].ext2bgd_nifree == 0) 1331 errx(EXIT_FAILURE, 1332 "%s: no free inode %" PRIu64 " in block group %u\n", 1333 __func__, (uint64_t)ino, c); 1334 1335 /* update inode bitmap */ 1336 rdfs(fsbtodb(&sblock, gd[0].ext2bgd_i_bitmap), 1337 sblock.e2fs_bsize, bp); 1338 1339 /* more sanity */ 1340 if (isset(bp, EXT2_INO_INDEX(ino))) 1341 errx(EXIT_FAILURE, "%s: inode %" PRIu64 1342 " already in use\n", __func__, (uint64_t)ino); 1343 setbit(bp, EXT2_INO_INDEX(ino)); 1344 wtfs(fsbtodb(&sblock, gd[0].ext2bgd_i_bitmap), 1345 sblock.e2fs_bsize, bp); 1346 gd[c].ext2bgd_nifree--; 1347 sblock.e2fs.e2fs_ficount--; 1348 } 1349 1350 if (ino >= sblock.e2fs.e2fs_ipg * sblock.e2fs_ncg) 1351 errx(EXIT_FAILURE, "%s: inode value out of range (%" PRIu64 1352 ").\n", __func__, (uint64_t)ino); 1353 1354 /* update an inode entry in the table */ 1355 d = fsbtodb(&sblock, ino_to_fsba(&sblock, ino)); 1356 rdfs(d, sblock.e2fs_bsize, bp); 1357 1358 dp = (struct ext2fs_dinode *)bp; 1359 dp += ino_to_fsbo(&sblock, ino); 1360 e2fs_isave(ip, dp); 1361 /* e2fs_i_bswap() doesn't swap e2di_blocks addrs */ 1362 if ((ip->e2di_mode & EXT2_IFMT) != EXT2_IFLNK) { 1363 for (i = 0; i < NDADDR + NIADDR; i++) 1364 dp->e2di_blocks[i] = h2fs32(ip->e2di_blocks[i]); 1365 } 1366 /* h2fs32() just for consistency */ 1367 dp->e2di_gen = h2fs32(arc4random()); 1368 1369 wtfs(d, sblock.e2fs_bsize, bp); 1370 free(bp); 1371} 1372 1373/* 1374 * Read a block from the file system 1375 */ 1376void 1377rdfs(daddr_t bno, int size, void *bf) 1378{ 1379 int n; 1380 off_t offset; 1381 1382 offset = bno; 1383 n = pread(fsi, bf, size, offset * sectorsize); 1384 if (n != size) 1385 err(EXIT_FAILURE, "%s: read error for sector %" PRId64, 1386 __func__, (int64_t)bno); 1387} 1388 1389/* 1390 * Write a block to the file system 1391 */ 1392void 1393wtfs(daddr_t bno, int size, void *bf) 1394{ 1395 int n; 1396 off_t offset; 1397 1398 if (Nflag) 1399 return; 1400 offset = bno; 1401 n = pwrite(fso, bf, size, offset * sectorsize); 1402 if (n != size) 1403 err(EXIT_FAILURE, "%s: write error for sector %" PRId64, 1404 __func__, (int64_t)bno); 1405} 1406 1407int 1408ilog2(uint val) 1409{ 1410 1411 if (val == 0 || !powerof2(val)) 1412 errx(EXIT_FAILURE, "%s: %u is not a power of 2\n", 1413 __func__, val); 1414 1415 return ffs(val) - 1; 1416} 1417 1418/* 1419 * int skpc(int mask, size_t size, uint8_t *cp) 1420 * 1421 * Locate an unsigned character of value mask inside cp[]. 1422 * (from src/sys/lib/libkern/skpc.c) 1423 */ 1424int 1425skpc(int mask, size_t size, uint8_t *cp) 1426{ 1427 uint8_t *end; 1428 1429 end = &cp[size]; 1430 while (cp < end && *cp == (uint8_t)mask) 1431 cp++; 1432 1433 return end - cp; 1434} 1435