mke2fs.c revision 1.12
1/* $NetBSD: mke2fs.c,v 1.12 2009/03/02 10:57:03 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 * 3. All advertising materials mentioning features or use of this software 68 * must display the following acknowledgement: 69 * This product includes software developed by Manuel Bouyer. 70 * 4. The name of the author may not be used to endorse or promote products 71 * derived from this software without specific prior written permission. 72 * 73 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 74 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 75 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 76 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 77 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 78 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 79 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 80 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 81 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 82 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 83 */ 84 85/* 86 * mke2fs.c: "re-invent (dumb but non-GPLed) wheel as a fun project" 87 * 88 * In spite of this name, there is no piece of code 89 * derived from GPLed e2fsprogs written for Linux. 90 * I referred them only to see how each structure 91 * member should be initialized. 92 * 93 * Reference: 94 * - All NetBSD sources under src/sys/ufs/ext2fs and src/sbin/fsck_ext2fs 95 * - Ext2fs Home Page 96 * http://e2fsprogs.sourceforge.net/ext2.html 97 * - Design and Implementation of the Second Extended Filesystem 98 * http://e2fsprogs.sourceforge.net/ext2intro.html 99 * - Linux Documentation "The Second Extended Filesystem" 100 * src/linux/Documentation/filesystems/ext2.txt 101 * in the Linux kernel distribution 102 */ 103 104#include <sys/cdefs.h> 105#ifndef lint 106#if 0 107static char sccsid[] = "@(#)mkfs.c 8.11 (Berkeley) 5/3/95"; 108#else 109__RCSID("$NetBSD: mke2fs.c,v 1.12 2009/03/02 10:57:03 tsutsui Exp $"); 110#endif 111#endif /* not lint */ 112 113#include <sys/param.h> 114#include <sys/mman.h> 115#include <sys/time.h> 116#include <ufs/ext2fs/ext2fs_dinode.h> 117#include <ufs/ext2fs/ext2fs_dir.h> 118#include <ufs/ext2fs/ext2fs.h> 119#include <sys/ioctl.h> 120 121#include <err.h> 122#include <errno.h> 123#include <string.h> 124#include <unistd.h> 125#include <stdlib.h> 126#include <stddef.h> 127#include <stdio.h> 128#include <uuid.h> 129 130#include "extern.h" 131 132static void initcg(uint); 133static void zap_old_sblock(daddr_t); 134static uint cgoverhead(uint); 135static int fsinit(const struct timeval *); 136static int makedir(struct ext2fs_direct *, int); 137static void copy_dir(struct ext2fs_direct *, struct ext2fs_direct *); 138static void init_resizeino(const struct timeval *); 139static uint32_t alloc(uint32_t, uint16_t); 140static void iput(struct ext2fs_dinode *, ino_t); 141static void rdfs(daddr_t, int, void *); 142static void wtfs(daddr_t, int, void *); 143static int ilog2(uint); 144static int skpc(int, size_t, uint8_t *); 145 146/* XXX: some of these macro should be into <ufs/ext2fs/ext2fs.h>? */ 147#define EXT2_DEF_MAX_MNT_COUNT 20 148#define EXT2_DEF_FSCKINTV (180 * 24 * 60 * 60) /* 180 days */ 149#define EXT2_RESERVED_INODES (EXT2_FIRSTINO - 1) 150#define EXT2_UMASK 0755 151 152#define EXT2_INO_INDEX(ino) ((ino) - 1) /* no inode zero */ 153 154#define EXT2_LOSTFOUNDSIZE 16384 155#define EXT2_LOSTFOUNDINO EXT2_FIRSTINO /* XXX: not quite */ 156#define EXT2_LOSTFOUNDUMASK 0700 157 158#define EXT2_RESIZEINOUMASK 0600 159 160#define NBLOCK_SUPERBLOCK 1 161#define NBLOCK_BLOCK_BITMAP 1 162#define NBLOCK_INODE_BITMAP 1 163 164#define cgbase(fs, c) \ 165 ((fs)->e2fs.e2fs_first_dblock + (fs)->e2fs.e2fs_bpg * (c)) 166 167 168/* 169 * ext2fs super block and group descriptor structures 170 * 171 * We don't have to use or setup whole in-memory m_ext2fs structure, 172 * but prepare it to use several macro defined in kernel headers. 173 */ 174union { 175 struct m_ext2fs m_ext2fs; 176 char pad[SBSIZE]; 177} ext2fsun; 178#define sblock ext2fsun.m_ext2fs 179#define gd ext2fsun.m_ext2fs.e2fs_gd 180 181static uint8_t *iobuf; /* for superblock and group descriptors */ 182static int iobufsize; 183 184static uint8_t buf[MAXBSIZE]; /* for initcg() and makedir() ops */ 185 186static int fsi, fso; 187 188void 189mke2fs(const char *fsys, int fi, int fo) 190{ 191 struct timeval tv; 192 int64_t minfssize; 193 uint bcount, fbcount, ficount; 194 uint blocks_gd, blocks_per_cg, inodes_per_cg, iblocks_per_cg; 195 uint minblocks_per_cg, blocks_lastcg; 196 uint ncg, cylno, sboff; 197 uuid_t uuid; 198 uint32_t uustat; 199 int i, len, col, delta, fld_width, max_cols; 200 struct winsize winsize; 201 202 gettimeofday(&tv, NULL); 203 fsi = fi; 204 fso = fo; 205 206 /* 207 * collect and verify the block and fragment sizes 208 */ 209 if (!powerof2(bsize)) { 210 errx(EXIT_FAILURE, 211 "block size must be a power of 2, not %u\n", 212 bsize); 213 } 214 if (!powerof2(fsize)) { 215 errx(EXIT_FAILURE, 216 "fragment size must be a power of 2, not %u\n", 217 fsize); 218 } 219 if (fsize < sectorsize) { 220 errx(EXIT_FAILURE, 221 "fragment size %u is too small, minimum is %u\n", 222 fsize, sectorsize); 223 } 224 if (bsize < MINBSIZE) { 225 errx(EXIT_FAILURE, 226 "block size %u is too small, minimum is %u\n", 227 bsize, MINBSIZE); 228 } 229 if (bsize > EXT2_MAXBSIZE) { 230 errx(EXIT_FAILURE, 231 "block size %u is too large, maximum is %u\n", 232 bsize, MAXBSIZE); 233 } 234 if (bsize != fsize) { 235 /* 236 * There is no fragment support on current ext2fs (yet?), 237 * but some kernel code refers fsize or fpg as bsize or bpg 238 * and Linux seems to set the same values to them. 239 */ 240 errx(EXIT_FAILURE, 241 "block size (%u) can't be different from " 242 "fragment size (%u)\n", 243 bsize, fsize); 244 } 245 246 /* variable inodesize is REV1 feature */ 247 if (Oflag == 0 && inodesize != EXT2_REV0_DINODE_SIZE) { 248 errx(EXIT_FAILURE, "GOOD_OLD_REV file system format" 249 " doesn't support %d byte inode\n", inodesize); 250 } 251 252 sblock.e2fs.e2fs_log_bsize = ilog2(bsize) - LOG_MINBSIZE; 253 /* Umm, why not e2fs_log_fsize? */ 254 sblock.e2fs.e2fs_fsize = ilog2(fsize) - LOG_MINBSIZE; 255 256 sblock.e2fs_bsize = bsize; 257 sblock.e2fs_bshift = sblock.e2fs.e2fs_log_bsize + LOG_MINBSIZE; 258 sblock.e2fs_qbmask = sblock.e2fs_bsize - 1; 259 sblock.e2fs_bmask = ~sblock.e2fs_qbmask; 260 sblock.e2fs_fsbtodb = ilog2(sblock.e2fs_bsize) - ilog2(sectorsize); 261 sblock.e2fs_ipb = sblock.e2fs_bsize / inodesize; 262 263 /* 264 * Ext2fs preserves BBSIZE (1024 bytes) space at the top for 265 * bootloader (though it is not enough at all for our bootloader). 266 * If bsize == BBSIZE we have to preserve one block. 267 * If bsize > BBSIZE, the first block already contains BBSIZE space 268 * before superblock because superblock is allocated at SBOFF and 269 * bsize is a power of two (i.e. 2048 bytes or more). 270 */ 271 sblock.e2fs.e2fs_first_dblock = (sblock.e2fs_bsize > BBSIZE) ? 0 : 1; 272 minfssize = fsbtodb(&sblock, 273 sblock.e2fs.e2fs_first_dblock + 274 NBLOCK_SUPERBLOCK + 275 1 /* at least one group descriptor */ + 276 NBLOCK_BLOCK_BITMAP + 277 NBLOCK_INODE_BITMAP + 278 1 /* at least one inode table block */ + 279 1 /* at least one data block for rootdir */ + 280 1 /* at least one data block for data */ 281 ); /* XXX and more? */ 282 283 if (fssize < minfssize) 284 errx(EXIT_FAILURE, "Filesystem size %" PRId64 285 " < minimum size of %" PRId64 "\n", fssize, minfssize); 286 287 bcount = dbtofsb(&sblock, fssize); 288 289 /* 290 * While many people claim that ext2fs is a (bad) clone of ufs/ffs, 291 * it isn't actual ffs so maybe we should call it "block group" 292 * as their native name rather than ffs derived "cylinder group." 293 * But we'll use the latter here since other kernel sources use it. 294 * (I also agree "cylinder" based allocation is obsolete though) 295 */ 296 297 /* maybe "simple is the best" */ 298 blocks_per_cg = sblock.e2fs_bsize * NBBY; 299 300 ncg = howmany(bcount - sblock.e2fs.e2fs_first_dblock, blocks_per_cg); 301 blocks_gd = howmany(sizeof(struct ext2_gd) * ncg, bsize); 302 303 /* check range of inode number */ 304 if (num_inodes < EXT2_FIRSTINO) 305 num_inodes = EXT2_FIRSTINO; /* needs reserved inodes + 1 */ 306 if (num_inodes > UINT16_MAX * ncg) 307 num_inodes = UINT16_MAX * ncg; /* ext2bgd_nifree is uint16_t */ 308 309 inodes_per_cg = num_inodes / ncg; 310 iblocks_per_cg = howmany(inodesize * inodes_per_cg, bsize); 311 312 /* Check that the last cylinder group has enough space for inodes */ 313 minblocks_per_cg = 314 NBLOCK_BLOCK_BITMAP + 315 NBLOCK_INODE_BITMAP + 316 iblocks_per_cg + 317 1; /* at least one data block */ 318 if (Oflag == 0 || cg_has_sb(ncg - 1) != 0) 319 minblocks_per_cg += NBLOCK_SUPERBLOCK + blocks_gd; 320 321 blocks_lastcg = bcount - sblock.e2fs.e2fs_first_dblock - 322 blocks_per_cg * (ncg - 1); 323 if (blocks_lastcg < minblocks_per_cg) { 324 /* 325 * Since we make all the cylinder groups the same size, the 326 * last will only be small if there are more than one 327 * cylinder groups. If the last one is too small to store 328 * filesystem data, just kill it. 329 * 330 * XXX: Does fsck_ext2fs(8) properly handle this case? 331 */ 332 bcount -= blocks_lastcg; 333 ncg--; 334 blocks_lastcg = blocks_per_cg; 335 blocks_gd = howmany(sizeof(struct ext2_gd) * ncg, bsize); 336 inodes_per_cg = num_inodes / ncg; 337 } 338 /* roundup inodes_per_cg to make it use whole inode table blocks */ 339 inodes_per_cg = roundup(inodes_per_cg, sblock.e2fs_ipb); 340 num_inodes = inodes_per_cg * ncg; 341 iblocks_per_cg = inodes_per_cg / sblock.e2fs_ipb; 342 343 /* XXX: probably we should check these adjusted values again */ 344 345 sblock.e2fs.e2fs_bcount = bcount; 346 sblock.e2fs.e2fs_icount = num_inodes; 347 348 sblock.e2fs_ncg = ncg; 349 sblock.e2fs_ngdb = blocks_gd; 350 sblock.e2fs_itpg = iblocks_per_cg; 351 352 sblock.e2fs.e2fs_rbcount = sblock.e2fs.e2fs_bcount * minfree / 100; 353 /* e2fs_fbcount will be accounted later */ 354 /* e2fs_ficount will be accounted later */ 355 356 sblock.e2fs.e2fs_bpg = blocks_per_cg; 357 sblock.e2fs.e2fs_fpg = blocks_per_cg; 358 359 sblock.e2fs.e2fs_ipg = inodes_per_cg; 360 361 sblock.e2fs.e2fs_mtime = 0; 362 sblock.e2fs.e2fs_wtime = tv.tv_sec; 363 sblock.e2fs.e2fs_mnt_count = 0; 364 /* XXX: should add some entropy to avoid checking all fs at once? */ 365 sblock.e2fs.e2fs_max_mnt_count = EXT2_DEF_MAX_MNT_COUNT; 366 367 sblock.e2fs.e2fs_magic = E2FS_MAGIC; 368 sblock.e2fs.e2fs_state = E2FS_ISCLEAN; 369 sblock.e2fs.e2fs_beh = E2FS_BEH_DEFAULT; 370 sblock.e2fs.e2fs_minrev = 0; 371 sblock.e2fs.e2fs_lastfsck = tv.tv_sec; 372 sblock.e2fs.e2fs_fsckintv = EXT2_DEF_FSCKINTV; 373 374 /* 375 * Maybe we can use E2FS_OS_FREEBSD here and it would be more proper, 376 * but the purpose of this newfs_ext2fs(8) command is to provide 377 * a filesystem which can be recognized by firmware on some 378 * Linux based appliances that can load bootstrap files only from 379 * (their native) ext2fs, and anyway we will (and should) try to 380 * act like them as much as possible. 381 * 382 * Anyway, I hope that all newer such boxes will keep their support 383 * for the "GOOD_OLD_REV" ext2fs. 384 */ 385 sblock.e2fs.e2fs_creator = E2FS_OS_LINUX; 386 387 if (Oflag == 0) { 388 sblock.e2fs.e2fs_rev = E2FS_REV0; 389 sblock.e2fs.e2fs_features_compat = 0; 390 sblock.e2fs.e2fs_features_incompat = 0; 391 sblock.e2fs.e2fs_features_rocompat = 0; 392 } else { 393 sblock.e2fs.e2fs_rev = E2FS_REV1; 394 /* 395 * e2fsprogs say "REV1" is "dynamic" so 396 * it isn't quite a version and maybe it means 397 * "extended from REV0 so check compat features." 398 * 399 * XXX: We don't have any native tool to activate 400 * the EXT2F_COMPAT_RESIZE feature and 401 * fsck_ext2fs(8) might not fix structures for it. 402 */ 403 sblock.e2fs.e2fs_features_compat = EXT2F_COMPAT_RESIZE; 404 sblock.e2fs.e2fs_features_incompat = EXT2F_INCOMPAT_FTYPE; 405 sblock.e2fs.e2fs_features_rocompat = 406 EXT2F_ROCOMPAT_SPARSESUPER | EXT2F_ROCOMPAT_LARGEFILE; 407 } 408 409 sblock.e2fs.e2fs_ruid = geteuid(); 410 sblock.e2fs.e2fs_rgid = getegid(); 411 412 sblock.e2fs.e2fs_first_ino = EXT2_FIRSTINO; 413 sblock.e2fs.e2fs_inode_size = inodesize; 414 415 /* e2fs_block_group_nr is set on writing superblock to each group */ 416 417 uuid_create(&uuid, &uustat); 418 if (uustat != uuid_s_ok) 419 errx(EXIT_FAILURE, "Failed to generate uuid\n"); 420 uuid_enc_le(sblock.e2fs.e2fs_uuid, &uuid); 421 if (volname != NULL) { 422 if (strlen(volname) > sizeof(sblock.e2fs.e2fs_vname)) 423 errx(EXIT_FAILURE, "Volume name is too long"); 424 strlcpy(sblock.e2fs.e2fs_vname, volname, 425 sizeof(sblock.e2fs.e2fs_vname)); 426 } 427 428 sblock.e2fs.e2fs_fsmnt[0] = '\0'; 429 sblock.e2fs_fsmnt[0] = '\0'; 430 431 sblock.e2fs.e2fs_algo = 0; /* XXX unsupported? */ 432 sblock.e2fs.e2fs_prealloc = 0; /* XXX unsupported? */ 433 sblock.e2fs.e2fs_dir_prealloc = 0; /* XXX unsupported? */ 434 435 /* calculate blocks for reserved group descriptors for resize */ 436 sblock.e2fs.e2fs_reserved_ngdb = 0; 437 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 438 (sblock.e2fs.e2fs_features_compat & EXT2F_COMPAT_RESIZE) != 0) { 439 uint64_t target_blocks; 440 uint target_ncg, target_ngdb, reserved_ngdb; 441 442 /* reserve descriptors for size as 1024 times as current */ 443 target_blocks = 444 (sblock.e2fs.e2fs_bcount - sblock.e2fs.e2fs_first_dblock) 445 * 1024ULL; 446 /* number of blocks must be in uint32_t */ 447 if (target_blocks > UINT32_MAX) 448 target_blocks = UINT32_MAX; 449 target_ncg = howmany(target_blocks, sblock.e2fs.e2fs_bpg); 450 target_ngdb = howmany(sizeof(struct ext2_gd) * target_ncg, 451 sblock.e2fs_bsize); 452 /* 453 * Reserved group descriptor blocks are preserved as 454 * the second level double indirect reference blocks in 455 * the EXT2_RESIZEINO inode, so the maximum number of 456 * the blocks is NINDIR(fs). 457 * (see also descriptions in init_resizeino() function) 458 * 459 * We check a number including current e2fs_ngdb here 460 * because they will be moved into reserved gdb on 461 * possible future size shrink, though e2fsprogs don't 462 * seem to care about it. 463 */ 464 if (target_ngdb > NINDIR(&sblock)) 465 target_ngdb = NINDIR(&sblock); 466 467 reserved_ngdb = target_ngdb - sblock.e2fs_ngdb; 468 469 /* make sure reserved_ngdb fits in the last cg */ 470 if (reserved_ngdb >= blocks_lastcg - cgoverhead(ncg - 1)) 471 reserved_ngdb = blocks_lastcg - cgoverhead(ncg - 1); 472 if (reserved_ngdb == 0) { 473 /* if no space for reserved gdb, disable the feature */ 474 sblock.e2fs.e2fs_features_compat &= 475 ~EXT2F_COMPAT_RESIZE; 476 } 477 sblock.e2fs.e2fs_reserved_ngdb = reserved_ngdb; 478 } 479 480 /* 481 * Initialize group descriptors 482 */ 483 gd = malloc(sblock.e2fs_ngdb * bsize); 484 if (gd == NULL) 485 errx(EXIT_FAILURE, "Can't allocate descriptors buffer"); 486 memset(gd, 0, sblock.e2fs_ngdb * bsize); 487 488 fbcount = 0; 489 ficount = 0; 490 for (cylno = 0; cylno < ncg; cylno++) { 491 uint boffset; 492 493 boffset = cgbase(&sblock, cylno); 494 if (sblock.e2fs.e2fs_rev == E2FS_REV0 || 495 (sblock.e2fs.e2fs_features_rocompat & 496 EXT2F_ROCOMPAT_SPARSESUPER) == 0 || 497 cg_has_sb(cylno)) { 498 boffset += NBLOCK_SUPERBLOCK + sblock.e2fs_ngdb; 499 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 500 (sblock.e2fs.e2fs_features_compat & 501 EXT2F_COMPAT_RESIZE) != 0) 502 boffset += sblock.e2fs.e2fs_reserved_ngdb; 503 } 504 gd[cylno].ext2bgd_b_bitmap = boffset; 505 boffset += NBLOCK_BLOCK_BITMAP; 506 gd[cylno].ext2bgd_i_bitmap = boffset; 507 boffset += NBLOCK_INODE_BITMAP; 508 gd[cylno].ext2bgd_i_tables = boffset; 509 if (cylno == (ncg - 1)) 510 gd[cylno].ext2bgd_nbfree = 511 blocks_lastcg - cgoverhead(cylno); 512 else 513 gd[cylno].ext2bgd_nbfree = 514 sblock.e2fs.e2fs_bpg - cgoverhead(cylno); 515 fbcount += gd[cylno].ext2bgd_nbfree; 516 gd[cylno].ext2bgd_nifree = sblock.e2fs.e2fs_ipg; 517 if (cylno == 0) { 518 /* take reserved inodes off nifree */ 519 gd[cylno].ext2bgd_nifree -= EXT2_RESERVED_INODES; 520 } 521 ficount += gd[cylno].ext2bgd_nifree; 522 gd[cylno].ext2bgd_ndirs = 0; 523 } 524 sblock.e2fs.e2fs_fbcount = fbcount; 525 sblock.e2fs.e2fs_ficount = ficount; 526 527 /* 528 * Dump out summary information about file system. 529 */ 530 if (verbosity > 0) { 531 printf("%s: %u.%1uMB (%" PRId64 " sectors) " 532 "block size %u, fragment size %u\n", 533 fsys, 534 (uint)(((uint64_t)bcount * bsize) / (1024 * 1024)), 535 (uint)((uint64_t)bcount * bsize - 536 rounddown((uint64_t)bcount * bsize, 1024 * 1024)) 537 / 1024 / 100, 538 fssize, bsize, fsize); 539 printf("\tusing %u block groups of %u.0MB, %u blks, " 540 "%u inodes.\n", 541 ncg, bsize * sblock.e2fs.e2fs_bpg / (1024 * 1024), 542 sblock.e2fs.e2fs_bpg, sblock.e2fs.e2fs_ipg); 543 } 544 545 /* 546 * allocate space for superblock and group descriptors 547 */ 548 iobufsize = (NBLOCK_SUPERBLOCK + sblock.e2fs_ngdb) * sblock.e2fs_bsize; 549 iobuf = mmap(0, iobufsize, PROT_READ|PROT_WRITE, 550 MAP_ANON|MAP_PRIVATE, -1, 0); 551 if (iobuf == NULL) 552 errx(EXIT_FAILURE, "Cannot allocate I/O buffer\n"); 553 memset(iobuf, 0, iobufsize); 554 555 /* 556 * We now start writing to the filesystem 557 */ 558 559 if (!Nflag) { 560 static const uint pbsize[] = { 1024, 2048, 4096, 0 }; 561 uint pblock, epblock; 562 /* 563 * Validate the given file system size. 564 * Verify that its last block can actually be accessed. 565 * Convert to file system fragment sized units. 566 */ 567 if (fssize <= 0) 568 errx(EXIT_FAILURE, "Preposterous size %" PRId64 "\n", 569 fssize); 570 wtfs(fssize - 1, sectorsize, iobuf); 571 572 /* 573 * Ensure there is nothing that looks like a filesystem 574 * superblock anywhere other than where ours will be. 575 * If fsck_ext2fs finds the wrong one all hell breaks loose! 576 * 577 * XXX: needs to check how fsck_ext2fs programs even 578 * on other OSes determine alternate superblocks 579 */ 580 for (i = 0; pbsize[i] != 0; i++) { 581 epblock = (uint64_t)bcount * bsize / pbsize[i]; 582 for (pblock = ((pbsize[i] == SBSIZE) ? 1 : 0); 583 pblock < epblock; 584 pblock += pbsize[i] * NBBY /* bpg */) 585 zap_old_sblock((daddr_t)pblock * 586 pbsize[i] / sectorsize); 587 } 588 } 589 590 if (verbosity >= 3) 591 printf("super-block backups (for fsck_ext2fs -b #) at:\n"); 592 /* If we are printing more than one line of numbers, line up columns */ 593 fld_width = verbosity < 4 ? 1 : snprintf(NULL, 0, "%" PRIu64, 594 (uint64_t)cgbase(&sblock, ncg - 1)); 595 /* Get terminal width */ 596 if (ioctl(fileno(stdout), TIOCGWINSZ, &winsize) == 0) 597 max_cols = winsize.ws_col; 598 else 599 max_cols = 80; 600 if (Nflag && verbosity == 3) 601 /* Leave space to add " ..." after one row of numbers */ 602 max_cols -= 4; 603#define BASE 0x10000 /* For some fixed-point maths */ 604 col = 0; 605 delta = verbosity > 2 ? 0 : max_cols * BASE / ncg; 606 for (cylno = 0; cylno < ncg; cylno++) { 607 fflush(stdout); 608 initcg(cylno); 609 if (verbosity < 2) 610 continue; 611 /* the first one is a master, not backup */ 612 if (cylno == 0) 613 continue; 614 /* skip if this cylinder doesn't have a backup */ 615 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 616 (sblock.e2fs.e2fs_features_rocompat & 617 EXT2F_ROCOMPAT_SPARSESUPER) != 0 && 618 cg_has_sb(cylno) == 0) 619 continue; 620 621 if (delta > 0) { 622 if (Nflag) 623 /* No point doing dots for -N */ 624 break; 625 /* Print dots scaled to end near RH margin */ 626 for (col += delta; col > BASE; col -= BASE) 627 printf("."); 628 continue; 629 } 630 /* Print superblock numbers */ 631 len = printf(" %*" PRIu64 "," + !col, fld_width, 632 (uint64_t)cgbase(&sblock, cylno)); 633 col += len; 634 if (col + len < max_cols) 635 /* Next number fits */ 636 continue; 637 /* Next number won't fit, need a newline */ 638 if (verbosity <= 3) { 639 /* Print dots for subsequent cylinder groups */ 640 delta = sblock.e2fs_ncg - cylno - 1; 641 if (delta != 0) { 642 if (Nflag) { 643 printf(" ..."); 644 break; 645 } 646 delta = max_cols * BASE / delta; 647 } 648 } 649 col = 0; 650 printf("\n"); 651 } 652#undef BASE 653 if (col > 0) 654 printf("\n"); 655 if (Nflag) 656 return; 657 658 /* 659 * Now construct the initial file system, 660 */ 661 if (fsinit(&tv) == 0) 662 errx(EXIT_FAILURE, "Error making filesystem"); 663 /* 664 * Write out the superblock and group descriptors 665 */ 666 sblock.e2fs.e2fs_block_group_nr = 0; 667 sboff = 0; 668 if (cgbase(&sblock, 0) == 0) { 669 /* 670 * If the first block contains the boot block sectors, 671 * (i.e. in case of sblock.e2fs.e2fs_bsize > BBSIZE) 672 * we have to preserve data in it. 673 */ 674 sboff = SBOFF; 675 } 676 e2fs_sbsave(&sblock.e2fs, (struct ext2fs *)(iobuf + sboff)); 677 e2fs_cgsave(gd, (struct ext2_gd *)(iobuf + sblock.e2fs_bsize), 678 sizeof(struct ext2_gd) * sblock.e2fs_ncg); 679 wtfs(fsbtodb(&sblock, cgbase(&sblock, 0)) + sboff / sectorsize, 680 iobufsize - sboff, iobuf + sboff); 681 682 munmap(iobuf, iobufsize); 683} 684 685/* 686 * Initialize a cylinder (block) group. 687 */ 688void 689initcg(uint cylno) 690{ 691 uint nblcg, i, j, sboff; 692 struct ext2fs_dinode *dp; 693 694 /* 695 * Make a copy of the superblock and group descriptors. 696 */ 697 if (sblock.e2fs.e2fs_rev == E2FS_REV0 || 698 (sblock.e2fs.e2fs_features_rocompat & 699 EXT2F_ROCOMPAT_SPARSESUPER) == 0 || 700 cg_has_sb(cylno)) { 701 sblock.e2fs.e2fs_block_group_nr = cylno; 702 sboff = 0; 703 if (cgbase(&sblock, cylno) == 0) { 704 /* preserve data in bootblock in cg0 */ 705 sboff = SBOFF; 706 } 707 e2fs_sbsave(&sblock.e2fs, (struct ext2fs *)(iobuf + sboff)); 708 e2fs_cgsave(gd, (struct ext2_gd *)(iobuf + 709 sblock.e2fs_bsize * NBLOCK_SUPERBLOCK), 710 sizeof(struct ext2_gd) * sblock.e2fs_ncg); 711 /* write superblock and group descriptor backups */ 712 wtfs(fsbtodb(&sblock, cgbase(&sblock, cylno)) + 713 sboff / sectorsize, iobufsize - sboff, iobuf + sboff); 714 } 715 716 /* 717 * Initialize block bitmap. 718 */ 719 memset(buf, 0, sblock.e2fs_bsize); 720 if (cylno == (sblock.e2fs_ncg - 1)) { 721 /* The last group could have less blocks than e2fs_bpg. */ 722 nblcg = sblock.e2fs.e2fs_bcount - 723 cgbase(&sblock, sblock.e2fs_ncg - 1); 724 for (i = nblcg; i < roundup(nblcg, NBBY); i++) 725 setbit(buf, i); 726 memset(&buf[i / NBBY], ~0U, sblock.e2fs.e2fs_bpg - i); 727 } 728 /* set overhead (superblock, group descriptor etc.) blocks used */ 729 for (i = 0; i < cgoverhead(cylno) / NBBY; i++) 730 buf[i] = ~0; 731 i = i * NBBY; 732 for (; i < cgoverhead(cylno); i++) 733 setbit(buf, i); 734 wtfs(fsbtodb(&sblock, gd[cylno].ext2bgd_b_bitmap), sblock.e2fs_bsize, 735 buf); 736 737 /* 738 * Initialize inode bitmap. 739 * 740 * Assume e2fs_ipg is a multiple of NBBY since 741 * it's a multiple of e2fs_ipb (as we did above). 742 * Note even (possibly smaller) the last group has the same e2fs_ipg. 743 */ 744 i = sblock.e2fs.e2fs_ipg / NBBY; 745 memset(buf, 0, i); 746 memset(buf + i, ~0U, sblock.e2fs_bsize - i); 747 if (cylno == 0) { 748 /* mark reserved inodes */ 749 for (i = 1; i < EXT2_FIRSTINO; i++) 750 setbit(buf, EXT2_INO_INDEX(i)); 751 } 752 wtfs(fsbtodb(&sblock, gd[cylno].ext2bgd_i_bitmap), sblock.e2fs_bsize, 753 buf); 754 755 /* 756 * Initialize inode tables. 757 * 758 * Just initialize generation numbers for NFS security. 759 * XXX: sys/ufs/ext2fs/ext2fs_alloc.c:ext2fs_valloc() seems 760 * to override these generated numbers. 761 */ 762 memset(buf, 0, sblock.e2fs_bsize); 763 for (i = 0; i < sblock.e2fs_itpg; i++) { 764 for (j = 0; j < sblock.e2fs_ipb; j++) { 765 dp = (struct ext2fs_dinode *)(buf + inodesize * j); 766 /* h2fs32() just for consistency */ 767 dp->e2di_gen = h2fs32(arc4random()); 768 } 769 wtfs(fsbtodb(&sblock, gd[cylno].ext2bgd_i_tables + i), 770 sblock.e2fs_bsize, buf); 771 } 772} 773 774/* 775 * Zap possible lingering old superblock data 776 */ 777static void 778zap_old_sblock(daddr_t sec) 779{ 780 static daddr_t cg0_data; 781 uint32_t oldfs[SBSIZE / sizeof(uint32_t)]; 782 static const struct fsm { 783 uint32_t offset; 784 uint32_t magic; 785 uint32_t mask; 786 } fs_magics[] = { 787 {offsetof(struct ext2fs, e2fs_magic) / 4, E2FS_MAGIC, 0xffff}, 788 {offsetof(struct ext2fs, e2fs_magic) / 4, 789 E2FS_MAGIC << 16, 0xffff0000}, 790 {14, 0xef530000, 0xffff0000}, /* EXT2FS (big) */ 791 {0x55c / 4, 0x00011954, ~0U}, /* FS_UFS1_MAGIC */ 792 {0x55c / 4, 0x19540119, ~0U}, /* FS_UFS2_MAGIC */ 793 {0, 0x70162, ~0U}, /* LFS_MAGIC */ 794 {.offset = ~0U}, 795 }; 796 const struct fsm *fsm; 797 798 if (Nflag) 799 return; 800 801 /* don't override data before superblock */ 802 if (sec < SBOFF / sectorsize) 803 return; 804 805 if (cg0_data == 0) { 806 cg0_data = 807 ((daddr_t)sblock.e2fs.e2fs_first_dblock + cgoverhead(0)) * 808 sblock.e2fs_bsize / sectorsize; 809 } 810 811 /* Ignore anything that is beyond our filesystem */ 812 if (sec >= fssize) 813 return; 814 /* Zero anything inside our filesystem... */ 815 if (sec >= sblock.e2fs.e2fs_first_dblock * bsize / sectorsize) { 816 /* ...unless we will write that area anyway */ 817 if (sec >= cg0_data) 818 /* assume iobuf is zero'ed here */ 819 wtfs(sec, roundup(SBSIZE, sectorsize), iobuf); 820 return; 821 } 822 823 /* 824 * The sector might contain boot code, so we must validate it 825 * 826 * XXX: ext2fs won't preserve data after SBOFF, 827 * but first_dblock could have a different value. 828 */ 829 rdfs(sec, sizeof(oldfs), &oldfs); 830 for (fsm = fs_magics;; fsm++) { 831 uint32_t v; 832 if (fsm->mask == 0) 833 return; 834 v = oldfs[fsm->offset]; 835 if ((v & fsm->mask) == fsm->magic || 836 (bswap32(v) & fsm->mask) == fsm->magic) 837 break; 838 } 839 840 /* Just zap the magic number */ 841 oldfs[fsm->offset] = 0; 842 wtfs(sec, sizeof(oldfs), &oldfs); 843} 844 845/* 846 * uint cgoverhead(uint c) 847 * 848 * Return a number of reserved blocks on the specified group. 849 * XXX: should be shared with src/sbin/fsck_ext2fs/setup.c 850 */ 851uint 852cgoverhead(uint c) 853{ 854 uint overh; 855 856 overh = NBLOCK_BLOCK_BITMAP + NBLOCK_INODE_BITMAP + sblock.e2fs_itpg; 857 858 if (sblock.e2fs.e2fs_rev == E2FS_REV0 || 859 (sblock.e2fs.e2fs_features_rocompat & 860 EXT2F_ROCOMPAT_SPARSESUPER) == 0 || 861 cg_has_sb(c) != 0) { 862 overh += NBLOCK_SUPERBLOCK + sblock.e2fs_ngdb; 863 864 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 865 (sblock.e2fs.e2fs_features_compat & 866 EXT2F_COMPAT_RESIZE) != 0) 867 overh += sblock.e2fs.e2fs_reserved_ngdb; 868 } 869 870 return overh; 871} 872 873/* 874 * Initialize the file system 875 */ 876 877#define LOSTDIR /* e2fsck complains if there is no lost+found */ 878 879#define PREDEFDIR 2 880 881#ifdef LOSTDIR 882#define PREDEFROOTDIR (PREDEFDIR + 1) 883#else 884#define PREDEFROOTDIR PREDEFDIR 885#endif 886 887struct ext2fs_direct root_dir[] = { 888 { EXT2_ROOTINO, 0, 1, 0, "." }, 889 { EXT2_ROOTINO, 0, 2, 0, ".." }, 890#ifdef LOSTDIR 891 { EXT2_LOSTFOUNDINO, 0, 10, 0, "lost+found" }, 892#endif 893}; 894 895#ifdef LOSTDIR 896struct ext2fs_direct lost_found_dir[] = { 897 { EXT2_LOSTFOUNDINO, 0, 1, 0, "." }, 898 { EXT2_ROOTINO, 0, 2, 0, ".." }, 899}; 900struct ext2fs_direct pad_dir = { 0, sizeof(struct ext2fs_direct), 0, 0, "" }; 901#endif 902 903int 904fsinit(const struct timeval *tv) 905{ 906 struct ext2fs_dinode node; 907#ifdef LOSTDIR 908 uint i, nblks_lostfound, blk; 909#endif 910 911 /* 912 * Initialize the inode for the resizefs feature 913 */ 914 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 915 (sblock.e2fs.e2fs_features_compat & EXT2F_COMPAT_RESIZE) != 0) 916 init_resizeino(tv); 917 918 /* 919 * Initialize the node 920 */ 921 922#ifdef LOSTDIR 923 /* 924 * Create the lost+found directory 925 */ 926 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 927 sblock.e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE) { 928 lost_found_dir[0].e2d_type = EXT2_FT_DIR; 929 lost_found_dir[1].e2d_type = EXT2_FT_DIR; 930 } 931 (void)makedir(lost_found_dir, __arraycount(lost_found_dir)); 932 933 /* prepare a bit large directory for preserved files */ 934 nblks_lostfound = EXT2_LOSTFOUNDSIZE / sblock.e2fs_bsize; 935 /* ...but only with direct blocks */ 936 if (nblks_lostfound > NDADDR) 937 nblks_lostfound = NDADDR; 938 939 memset(&node, 0, sizeof(node)); 940 node.e2di_mode = EXT2_IFDIR | EXT2_LOSTFOUNDUMASK; 941 node.e2di_uid = geteuid(); 942 node.e2di_size = sblock.e2fs_bsize * nblks_lostfound; 943 node.e2di_atime = tv->tv_sec; 944 node.e2di_ctime = tv->tv_sec; 945 node.e2di_mtime = tv->tv_sec; 946 node.e2di_gid = getegid(); 947 node.e2di_nlink = PREDEFDIR; 948 /* e2di_nblock is a number of disk blocks, not ext2fs blocks */ 949 node.e2di_nblock = fsbtodb(&sblock, nblks_lostfound); 950 node.e2di_blocks[0] = alloc(sblock.e2fs_bsize, node.e2di_mode); 951 if (node.e2di_blocks[0] == 0) { 952 printf("%s: can't allocate block for lost+found\n", __func__); 953 return 0; 954 } 955 for (i = 1; i < nblks_lostfound; i++) { 956 blk = alloc(sblock.e2fs_bsize, 0); 957 if (blk == 0) { 958 printf("%s: can't allocate blocks for lost+found\n", 959 __func__); 960 return 0; 961 } 962 node.e2di_blocks[i] = blk; 963 } 964 wtfs(fsbtodb(&sblock, node.e2di_blocks[0]), sblock.e2fs_bsize, buf); 965 pad_dir.e2d_reclen = sblock.e2fs_bsize; 966 for (i = 1; i < nblks_lostfound; i++) { 967 memset(buf, 0, sblock.e2fs_bsize); 968 copy_dir(&pad_dir, (struct ext2fs_direct *)buf); 969 wtfs(fsbtodb(&sblock, node.e2di_blocks[i]), sblock.e2fs_bsize, 970 buf); 971 } 972 iput(&node, EXT2_LOSTFOUNDINO); 973#endif 974 /* 975 * create the root directory 976 */ 977 memset(&node, 0, sizeof(node)); 978 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 979 sblock.e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE) { 980 root_dir[0].e2d_type = EXT2_FT_DIR; 981 root_dir[1].e2d_type = EXT2_FT_DIR; 982#ifdef LOSTDIR 983 root_dir[2].e2d_type = EXT2_FT_DIR; 984#endif 985 } 986 node.e2di_mode = EXT2_IFDIR | EXT2_UMASK; 987 node.e2di_uid = geteuid(); 988 node.e2di_size = makedir(root_dir, __arraycount(root_dir)); 989 node.e2di_atime = tv->tv_sec; 990 node.e2di_ctime = tv->tv_sec; 991 node.e2di_mtime = tv->tv_sec; 992 node.e2di_gid = getegid(); 993 node.e2di_nlink = PREDEFROOTDIR; 994 /* e2di_nblock is a number of disk block, not ext2fs block */ 995 node.e2di_nblock = fsbtodb(&sblock, 1); 996 node.e2di_blocks[0] = alloc(node.e2di_size, node.e2di_mode); 997 if (node.e2di_blocks[0] == 0) { 998 printf("%s: can't allocate block for root dir\n", __func__); 999 return 0; 1000 } 1001 wtfs(fsbtodb(&sblock, node.e2di_blocks[0]), sblock.e2fs_bsize, buf); 1002 iput(&node, EXT2_ROOTINO); 1003 return 1; 1004} 1005 1006/* 1007 * Construct a set of directory entries in "buf". 1008 * return size of directory. 1009 */ 1010int 1011makedir(struct ext2fs_direct *protodir, int entries) 1012{ 1013 uint8_t *cp; 1014 uint i, spcleft; 1015 uint dirblksiz; 1016 1017 dirblksiz = sblock.e2fs_bsize; 1018 memset(buf, 0, dirblksiz); 1019 spcleft = dirblksiz; 1020 for (cp = buf, i = 0; i < entries - 1; i++) { 1021 protodir[i].e2d_reclen = EXT2FS_DIRSIZ(protodir[i].e2d_namlen); 1022 copy_dir(&protodir[i], (struct ext2fs_direct *)cp); 1023 cp += protodir[i].e2d_reclen; 1024 spcleft -= protodir[i].e2d_reclen; 1025 } 1026 protodir[i].e2d_reclen = spcleft; 1027 copy_dir(&protodir[i], (struct ext2fs_direct *)cp); 1028 return dirblksiz; 1029} 1030 1031/* 1032 * Copy a direntry to a buffer, in fs byte order 1033 */ 1034static void 1035copy_dir(struct ext2fs_direct *dir, struct ext2fs_direct *dbuf) 1036{ 1037 1038 memcpy(dbuf, dir, EXT2FS_DIRSIZ(dir->e2d_namlen)); 1039 dbuf->e2d_ino = h2fs32(dir->e2d_ino); 1040 dbuf->e2d_reclen = h2fs16(dir->e2d_reclen); 1041} 1042 1043/* 1044 * void init_resizeino(const struct timeval *tv); 1045 * 1046 * Initialize the EXT2_RESEIZE_INO inode to preserve 1047 * reserved group descriptor blocks for future growth of this ext2fs. 1048 */ 1049void 1050init_resizeino(const struct timeval *tv) 1051{ 1052 struct ext2fs_dinode node; 1053 uint64_t isize; 1054 uint32_t *dindir_block, *reserved_gdb; 1055 uint nblock, i, cylno, n; 1056 1057 memset(&node, 0, sizeof(node)); 1058 1059 /* 1060 * Note this function only prepares required structures for 1061 * future resize. It's a quite different work to implement 1062 * a utility like resize_ext2fs(8) which handles actual 1063 * resize ops even on offline. 1064 * 1065 * Anyway, I'm not sure if there is any documentation about 1066 * this resize ext2fs feature and related data structures, 1067 * and I've written this function based on things what I see 1068 * on some existing implementation and real file system data 1069 * created by existing tools. To be honest, they are not 1070 * so easy to read, so I will try to implement it here without 1071 * any dumb optimization for people who would eventually 1072 * work on "yet another wheel" like resize_ext2fs(8). 1073 */ 1074 1075 /* 1076 * I'm not sure what type is appropriate for this inode. 1077 * The release notes of e2fsprogs says they changed e2fsck to allow 1078 * IFREG for RESIZEINO since a certain resize tool used it. Hmm. 1079 */ 1080 node.e2di_mode = EXT2_IFREG | EXT2_RESIZEINOUMASK; 1081 node.e2di_uid = geteuid(); 1082 node.e2di_atime = tv->tv_sec; 1083 node.e2di_ctime = tv->tv_sec; 1084 node.e2di_mtime = tv->tv_sec; 1085 node.e2di_gid = getegid(); 1086 node.e2di_nlink = 1; 1087 1088 /* 1089 * To preserve the reserved group descriptor blocks, 1090 * EXT2_RESIZEINO uses only double indirect reference 1091 * blocks in its inode entries. 1092 * 1093 * All entries for direct, single indirect and triple 1094 * indirect references are left zero'ed. Maybe it's safe 1095 * because no write operation will happen with this inode. 1096 * 1097 * We have to allocate a block for the first level double 1098 * indirect reference block. Indexes of inode entries in 1099 * this first level dindirect block are corresponding to 1100 * indexes of group descriptors including both used (e2fs_ngdb) 1101 * and reserved (e2fs_reserved_ngdb) group descriptor blocks. 1102 * 1103 * Inode entries of indexes for used (e2fs_ngdb) descriptors are 1104 * left zero'ed. Entries for reserved (e2fs_reserved_ngdb) ones 1105 * have block numbers of actual reserved group descriptors 1106 * allocated at block group zero. This means e2fs_reserved_ngdb 1107 * blocks are reserved as the second level dindirect reference 1108 * blocks, and they actually contain block numbers of indirect 1109 * references. It may be safe since they don't have to keep any 1110 * data yet. 1111 * 1112 * Each these second dindirect blocks (i.e. reserved group 1113 * descriptor blocks in the first block group) should have 1114 * block numbers of its backups in all other block groups. 1115 * I.e. reserved_ngdb[0] block in block group 0 contains block 1116 * numbers of resreved_ngdb[0] from group 1 through (e2fs_ncg - 1). 1117 * The number of backups can be determined by the 1118 * EXT2_ROCOMPAT_SPARSESUPER feature and cg_has_sb() macro 1119 * as done in the above initcg() function. 1120 */ 1121 1122 /* set e2di_size which occupies whole blocks through DINDIR blocks */ 1123 isize = (uint64_t)sblock.e2fs_bsize * NDADDR + 1124 (uint64_t)sblock.e2fs_bsize * NINDIR(&sblock) + 1125 (uint64_t)sblock.e2fs_bsize * NINDIR(&sblock) * NINDIR(&sblock); 1126 if (isize > UINT32_MAX && 1127 (sblock.e2fs.e2fs_features_rocompat & 1128 EXT2F_ROCOMPAT_LARGEFILE) == 0) { 1129 /* XXX should enable it here and update all backups? */ 1130 errx(EXIT_FAILURE, "%s: large_file rocompat feature is " 1131 "required to enable resize feature for this filesystem\n", 1132 __func__); 1133 } 1134 /* upper 32bit is stored into e2di_dacl on REV1 feature */ 1135 node.e2di_size = isize & UINT32_MAX; 1136 node.e2di_dacl = isize >> 32; 1137 1138#define SINGLE 0 /* index of single indirect block */ 1139#define DOUBLE 1 /* index of double indirect block */ 1140#define TRIPLE 2 /* index of triple indirect block */ 1141 1142 /* zero out entries for direct references */ 1143 for (i = 0; i < NDADDR; i++) 1144 node.e2di_blocks[i] = 0; 1145 /* also zero out entries for single and triple indirect references */ 1146 node.e2di_blocks[NDADDR + SINGLE] = 0; 1147 node.e2di_blocks[NDADDR + TRIPLE] = 0; 1148 1149 /* allocate a block for the first level double indirect reference */ 1150 node.e2di_blocks[NDADDR + DOUBLE] = 1151 alloc(sblock.e2fs_bsize, node.e2di_mode); 1152 if (node.e2di_blocks[NDADDR + DOUBLE] == 0) 1153 errx(EXIT_FAILURE, "%s: Can't allocate a dindirect block", 1154 __func__); 1155 1156 /* account this first block */ 1157 nblock = fsbtodb(&sblock, 1); 1158 1159 /* allocate buffer to set data in the dindirect block */ 1160 dindir_block = malloc(sblock.e2fs_bsize); 1161 if (dindir_block == NULL) 1162 errx(EXIT_FAILURE, 1163 "%s: Can't allocate buffer for a dindirect block", 1164 __func__); 1165 1166 /* allocate buffer to set data in the group descriptor blocks */ 1167 reserved_gdb = malloc(sblock.e2fs_bsize); 1168 if (reserved_gdb == NULL) 1169 errx(EXIT_FAILURE, 1170 "%s: Can't allocate buffer for group descriptor blocks", 1171 __func__); 1172 1173 /* 1174 * Setup block entries in the first level dindirect blocks 1175 */ 1176 for (i = 0; i < sblock.e2fs_ngdb; i++) { 1177 /* no need to handle used group descriptor blocks */ 1178 dindir_block[i] = 0; 1179 } 1180 for (; i < sblock.e2fs_ngdb + sblock.e2fs.e2fs_reserved_ngdb; i++) { 1181 /* 1182 * point reserved group descriptor block in the first 1183 * (i.e. master) block group 1184 * 1185 * XXX: e2fsprogs seem to use "(i % NINDIR(&sblock))" here 1186 * to store maximum NINDIR(&sblock) reserved gdbs. 1187 * I'm not sure what will be done on future filesystem 1188 * shrink in that case on their way. 1189 */ 1190 if (i >= NINDIR(&sblock)) 1191 errx(EXIT_FAILURE, "%s: too many reserved " 1192 "group descriptors (%u) for resize inode", 1193 __func__, sblock.e2fs.e2fs_reserved_ngdb); 1194 dindir_block[i] = 1195 h2fs32(cgbase(&sblock, 0) + NBLOCK_SUPERBLOCK + i); 1196 1197 /* 1198 * Setup block entries in the second dindirect blocks 1199 * (which are primary reserved group descriptor blocks) 1200 * to point their backups. 1201 */ 1202 for (n = 0, cylno = 1; cylno < sblock.e2fs_ncg; cylno++) { 1203 /* skip block groups without backup */ 1204 if ((sblock.e2fs.e2fs_features_rocompat & 1205 EXT2F_ROCOMPAT_SPARSESUPER) != 0 && 1206 cg_has_sb(cylno) == 0) 1207 continue; 1208 1209 if (n >= NINDIR(&sblock)) 1210 errx(EXIT_FAILURE, "%s: too many block groups " 1211 "for the resize feature", __func__); 1212 /* 1213 * These blocks are already reserved in 1214 * initcg() so no need to use alloc() here. 1215 */ 1216 reserved_gdb[n++] = h2fs32(cgbase(&sblock, cylno) + 1217 NBLOCK_SUPERBLOCK + i); 1218 nblock += fsbtodb(&sblock, 1); 1219 } 1220 for (; n < NINDIR(&sblock); n++) 1221 reserved_gdb[n] = 0; 1222 1223 /* write group descriptor block as the second dindirect refs */ 1224 wtfs(fsbtodb(&sblock, fs2h32(dindir_block[i])), 1225 sblock.e2fs_bsize, reserved_gdb); 1226 nblock += fsbtodb(&sblock, 1); 1227 } 1228 for (; i < NINDIR(&sblock); i++) { 1229 /* leave trailing entries unallocated */ 1230 dindir_block[i] = 0; 1231 } 1232 free(reserved_gdb); 1233 1234 /* finally write the first level dindirect block */ 1235 wtfs(fsbtodb(&sblock, node.e2di_blocks[NDADDR + DOUBLE]), 1236 sblock.e2fs_bsize, dindir_block); 1237 free(dindir_block); 1238 1239 node.e2di_nblock = nblock; 1240 iput(&node, EXT2_RESIZEINO); 1241} 1242 1243/* 1244 * uint32_t alloc(uint32_t size, uint16_t mode) 1245 * 1246 * Allocate a block (from cylinder group 0) 1247 * Reference: src/sys/ufs/ext2fs/ext2fs_alloc.c:ext2fs_alloccg() 1248 */ 1249uint32_t 1250alloc(uint32_t size, uint16_t mode) 1251{ 1252 uint32_t loc, bno; 1253 uint8_t *bbp; 1254 uint len, map, i; 1255 1256 if (gd[0].ext2bgd_nbfree == 0) 1257 return 0; 1258 1259 if (size > sblock.e2fs_bsize) 1260 return 0; 1261 1262 bbp = malloc(sblock.e2fs_bsize); 1263 if (bbp == NULL) 1264 return 0; 1265 rdfs(fsbtodb(&sblock, gd[0].ext2bgd_b_bitmap), sblock.e2fs_bsize, bbp); 1266 1267 /* XXX: kernel uses e2fs_fpg here */ 1268 len = sblock.e2fs.e2fs_bpg / NBBY; 1269 1270#if 0 /* no need block allocation for root or lost+found dir */ 1271 for (loc = 0; loc < len; loc++) { 1272 if (bbp[loc] == 0) { 1273 bno = loc * NBBY; 1274 goto gotit; 1275 } 1276 } 1277#endif 1278 1279 loc = skpc(~0U, len, bbp); 1280 if (loc == 0) 1281 return 0; 1282 loc = len - loc; 1283 map = bbp[loc]; 1284 bno = loc * NBBY; 1285 for (i = 0; i < NBBY; i++, bno++) { 1286 if ((map & (1 << i)) == 0) 1287 goto gotit; 1288 } 1289 return 0; 1290 1291 gotit: 1292 if (isset(bbp, bno)) 1293 errx(EXIT_FAILURE, "%s: inconsistent bitmap\n", __func__); 1294 1295 setbit(bbp, bno); 1296 wtfs(fsbtodb(&sblock, gd[0].ext2bgd_b_bitmap), sblock.e2fs_bsize, bbp); 1297 free(bbp); 1298 /* XXX: modified group descriptors won't be written into backups */ 1299 gd[0].ext2bgd_nbfree--; 1300 if ((mode & EXT2_IFDIR) != 0) 1301 gd[0].ext2bgd_ndirs++; 1302 sblock.e2fs.e2fs_fbcount--; 1303 1304 return sblock.e2fs.e2fs_first_dblock + bno; 1305} 1306 1307/* 1308 * void iput(struct ext2fs_dinode *ip, ino_t ino) 1309 * 1310 * Put an inode entry into the corresponding table. 1311 */ 1312static void 1313iput(struct ext2fs_dinode *ip, ino_t ino) 1314{ 1315 daddr_t d; 1316 uint c, i; 1317 struct ext2fs_dinode *dp; 1318 uint8_t *bp; 1319 1320 bp = malloc(sblock.e2fs_bsize); 1321 if (bp == NULL) 1322 errx(EXIT_FAILURE, "%s: can't allocate buffer for inode\n", 1323 __func__); 1324 1325 /* 1326 * Reserved inodes are allocated and accounted in initcg() 1327 * so skip checks of the bitmap and allocation for them. 1328 */ 1329 if (ino >= EXT2_FIRSTINO) { 1330 c = ino_to_cg(&sblock, ino); 1331 1332 /* sanity check */ 1333 if (gd[c].ext2bgd_nifree == 0) 1334 errx(EXIT_FAILURE, 1335 "%s: no free inode %" PRIu64 " in block group %u\n", 1336 __func__, (uint64_t)ino, c); 1337 1338 /* update inode bitmap */ 1339 rdfs(fsbtodb(&sblock, gd[0].ext2bgd_i_bitmap), 1340 sblock.e2fs_bsize, bp); 1341 1342 /* more sanity */ 1343 if (isset(bp, EXT2_INO_INDEX(ino))) 1344 errx(EXIT_FAILURE, "%s: inode %" PRIu64 1345 " already in use\n", __func__, (uint64_t)ino); 1346 setbit(bp, EXT2_INO_INDEX(ino)); 1347 wtfs(fsbtodb(&sblock, gd[0].ext2bgd_i_bitmap), 1348 sblock.e2fs_bsize, bp); 1349 gd[c].ext2bgd_nifree--; 1350 sblock.e2fs.e2fs_ficount--; 1351 } 1352 1353 if (ino >= sblock.e2fs.e2fs_ipg * sblock.e2fs_ncg) 1354 errx(EXIT_FAILURE, "%s: inode value out of range (%" PRIu64 1355 ").\n", __func__, (uint64_t)ino); 1356 1357 /* update an inode entry in the table */ 1358 d = fsbtodb(&sblock, ino_to_fsba(&sblock, ino)); 1359 rdfs(d, sblock.e2fs_bsize, bp); 1360 1361 dp = (struct ext2fs_dinode *)(bp + 1362 inodesize * ino_to_fsbo(&sblock, ino)); 1363 e2fs_isave(ip, dp); 1364 /* e2fs_i_bswap() doesn't swap e2di_blocks addrs */ 1365 if ((ip->e2di_mode & EXT2_IFMT) != EXT2_IFLNK) { 1366 for (i = 0; i < NDADDR + NIADDR; i++) 1367 dp->e2di_blocks[i] = h2fs32(ip->e2di_blocks[i]); 1368 } 1369 /* h2fs32() just for consistency */ 1370 dp->e2di_gen = h2fs32(arc4random()); 1371 1372 wtfs(d, sblock.e2fs_bsize, bp); 1373 free(bp); 1374} 1375 1376/* 1377 * Read a block from the file system 1378 */ 1379void 1380rdfs(daddr_t bno, int size, void *bf) 1381{ 1382 int n; 1383 off_t offset; 1384 1385 offset = bno; 1386 n = pread(fsi, bf, size, offset * sectorsize); 1387 if (n != size) 1388 err(EXIT_FAILURE, "%s: read error for sector %" PRId64, 1389 __func__, (int64_t)bno); 1390} 1391 1392/* 1393 * Write a block to the file system 1394 */ 1395void 1396wtfs(daddr_t bno, int size, void *bf) 1397{ 1398 int n; 1399 off_t offset; 1400 1401 if (Nflag) 1402 return; 1403 offset = bno; 1404 n = pwrite(fso, bf, size, offset * sectorsize); 1405 if (n != size) 1406 err(EXIT_FAILURE, "%s: write error for sector %" PRId64, 1407 __func__, (int64_t)bno); 1408} 1409 1410int 1411ilog2(uint val) 1412{ 1413 1414 if (val == 0 || !powerof2(val)) 1415 errx(EXIT_FAILURE, "%s: %u is not a power of 2\n", 1416 __func__, val); 1417 1418 return ffs(val) - 1; 1419} 1420 1421/* 1422 * int skpc(int mask, size_t size, uint8_t *cp) 1423 * 1424 * Locate an unsigned character of value mask inside cp[]. 1425 * (from src/sys/lib/libkern/skpc.c) 1426 */ 1427int 1428skpc(int mask, size_t size, uint8_t *cp) 1429{ 1430 uint8_t *end; 1431 1432 end = &cp[size]; 1433 while (cp < end && *cp == (uint8_t)mask) 1434 cp++; 1435 1436 return end - cp; 1437} 1438