1 /* $NetBSD: mke2fs.c,v 1.14 2010/09/10 15:51:20 tsutsui Exp $ */ 2 3 /*- 4 * Copyright (c) 2007 Izumi Tsutsui. All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 */ 26 27 /* 28 * Copyright (c) 1980, 1989, 1993 29 * The Regents of the University of California. All rights reserved. 30 * 31 * Redistribution and use in source and binary forms, with or without 32 * modification, are permitted provided that the following conditions 33 * are met: 34 * 1. Redistributions of source code must retain the above copyright 35 * notice, this list of conditions and the following disclaimer. 36 * 2. Redistributions in binary form must reproduce the above copyright 37 * notice, this list of conditions and the following disclaimer in the 38 * documentation and/or other materials provided with the distribution. 39 * 3. Neither the name of the University nor the names of its contributors 40 * may be used to endorse or promote products derived from this software 41 * without specific prior written permission. 42 * 43 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 44 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 45 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 46 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 47 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 48 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 49 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 50 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 51 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 52 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 53 * SUCH DAMAGE. 54 */ 55 56 /* 57 * Copyright (c) 1997 Manuel Bouyer. 58 * 59 * Redistribution and use in source and binary forms, with or without 60 * modification, are permitted provided that the following conditions 61 * are met: 62 * 1. Redistributions of source code must retain the above copyright 63 * notice, this list of conditions and the following disclaimer. 64 * 2. Redistributions in binary form must reproduce the above copyright 65 * notice, this list of conditions and the following disclaimer in the 66 * documentation and/or other materials provided with the distribution. 67 * 68 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 69 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 70 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 71 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 72 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 73 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 74 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 75 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 76 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 77 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 78 */ 79 80 /* 81 * mke2fs.c: "re-invent (dumb but non-GPLed) wheel as a fun project" 82 * 83 * In spite of this name, there is no piece of code 84 * derived from GPLed e2fsprogs written for Linux. 85 * I referred them only to see how each structure 86 * member should be initialized. 87 * 88 * Reference: 89 * - All NetBSD sources under src/sys/ufs/ext2fs and src/sbin/fsck_ext2fs 90 * - Ext2fs Home Page 91 * http://e2fsprogs.sourceforge.net/ext2.html 92 * - Design and Implementation of the Second Extended Filesystem 93 * http://e2fsprogs.sourceforge.net/ext2intro.html 94 * - Linux Documentation "The Second Extended Filesystem" 95 * http://www.kernel.org/doc/Documentation/filesystems/ext2.txt 96 */ 97 98 #include <sys/cdefs.h> 99 #ifndef lint 100 #if 0 101 static char sccsid[] = "@(#)mkfs.c 8.11 (Berkeley) 5/3/95"; 102 #else 103 __RCSID("$NetBSD: mke2fs.c,v 1.14 2010/09/10 15:51:20 tsutsui Exp $"); 104 #endif 105 #endif /* not lint */ 106 107 #include <sys/param.h> 108 #include <sys/mman.h> 109 #include <sys/time.h> 110 #include <ufs/ext2fs/ext2fs_dinode.h> 111 #include <ufs/ext2fs/ext2fs_dir.h> 112 #include <ufs/ext2fs/ext2fs.h> 113 #include <sys/ioctl.h> 114 115 #include <err.h> 116 #include <errno.h> 117 #include <string.h> 118 #include <unistd.h> 119 #include <stdlib.h> 120 #include <stddef.h> 121 #include <stdio.h> 122 #include <uuid.h> 123 124 #include "extern.h" 125 126 static void initcg(uint); 127 static void zap_old_sblock(daddr_t); 128 static uint cgoverhead(uint); 129 static int fsinit(const struct timeval *); 130 static int makedir(struct ext2fs_direct *, int); 131 static void copy_dir(struct ext2fs_direct *, struct ext2fs_direct *); 132 static void init_resizeino(const struct timeval *); 133 static uint32_t alloc(uint32_t, uint16_t); 134 static void iput(struct ext2fs_dinode *, ino_t); 135 static void rdfs(daddr_t, int, void *); 136 static void wtfs(daddr_t, int, void *); 137 static int ilog2(uint); 138 static int skpc(int, size_t, uint8_t *); 139 140 /* XXX: some of these macro should be into <ufs/ext2fs/ext2fs.h>? */ 141 #define EXT2_DEF_MAX_MNT_COUNT 20 142 #define EXT2_DEF_FSCKINTV (180 * 24 * 60 * 60) /* 180 days */ 143 #define EXT2_RESERVED_INODES (EXT2_FIRSTINO - 1) 144 #define EXT2_UMASK 0755 145 146 #define EXT2_INO_INDEX(ino) ((ino) - 1) /* no inode zero */ 147 148 #define EXT2_LOSTFOUNDSIZE 16384 149 #define EXT2_LOSTFOUNDINO EXT2_FIRSTINO /* XXX: not quite */ 150 #define EXT2_LOSTFOUNDUMASK 0700 151 152 #define EXT2_RESIZEINOUMASK 0600 153 154 #define NBLOCK_SUPERBLOCK 1 155 #define NBLOCK_BLOCK_BITMAP 1 156 #define NBLOCK_INODE_BITMAP 1 157 158 #define cgbase(fs, c) \ 159 ((fs)->e2fs.e2fs_first_dblock + (fs)->e2fs.e2fs_bpg * (c)) 160 161 162 /* 163 * ext2fs super block and group descriptor structures 164 * 165 * We don't have to use or setup whole in-memory m_ext2fs structure, 166 * but prepare it to use several macro defined in kernel headers. 167 */ 168 union { 169 struct m_ext2fs m_ext2fs; 170 char pad[SBSIZE]; 171 } ext2fsun; 172 #define sblock ext2fsun.m_ext2fs 173 #define gd ext2fsun.m_ext2fs.e2fs_gd 174 175 static uint8_t *iobuf; /* for superblock and group descriptors */ 176 static int iobufsize; 177 178 static uint8_t buf[MAXBSIZE]; /* for initcg() and makedir() ops */ 179 180 static int fsi, fso; 181 182 void 183 mke2fs(const char *fsys, int fi, int fo) 184 { 185 struct timeval tv; 186 int64_t minfssize; 187 uint bcount, fbcount, ficount; 188 uint blocks_gd, blocks_per_cg, inodes_per_cg, iblocks_per_cg; 189 uint minblocks_per_cg, blocks_lastcg; 190 uint ncg, cylno, sboff; 191 uuid_t uuid; 192 uint32_t uustat; 193 int i, len, col, delta, fld_width, max_cols; 194 struct winsize winsize; 195 196 gettimeofday(&tv, NULL); 197 fsi = fi; 198 fso = fo; 199 200 /* 201 * collect and verify the block and fragment sizes 202 */ 203 if (!powerof2(bsize)) { 204 errx(EXIT_FAILURE, 205 "block size must be a power of 2, not %u\n", 206 bsize); 207 } 208 if (!powerof2(fsize)) { 209 errx(EXIT_FAILURE, 210 "fragment size must be a power of 2, not %u\n", 211 fsize); 212 } 213 if (fsize < sectorsize) { 214 errx(EXIT_FAILURE, 215 "fragment size %u is too small, minimum is %u\n", 216 fsize, sectorsize); 217 } 218 if (bsize < MINBSIZE) { 219 errx(EXIT_FAILURE, 220 "block size %u is too small, minimum is %u\n", 221 bsize, MINBSIZE); 222 } 223 if (bsize > EXT2_MAXBSIZE) { 224 errx(EXIT_FAILURE, 225 "block size %u is too large, maximum is %u\n", 226 bsize, MAXBSIZE); 227 } 228 if (bsize != fsize) { 229 /* 230 * There is no fragment support on current ext2fs (yet?), 231 * but some kernel code refers fsize or fpg as bsize or bpg 232 * and Linux seems to set the same values to them. 233 */ 234 errx(EXIT_FAILURE, 235 "block size (%u) can't be different from " 236 "fragment size (%u)\n", 237 bsize, fsize); 238 } 239 240 /* variable inodesize is REV1 feature */ 241 if (Oflag == 0 && inodesize != EXT2_REV0_DINODE_SIZE) { 242 errx(EXIT_FAILURE, "GOOD_OLD_REV file system format" 243 " doesn't support %d byte inode\n", inodesize); 244 } 245 246 sblock.e2fs.e2fs_log_bsize = ilog2(bsize) - LOG_MINBSIZE; 247 /* Umm, why not e2fs_log_fsize? */ 248 sblock.e2fs.e2fs_fsize = ilog2(fsize) - LOG_MINBSIZE; 249 250 sblock.e2fs_bsize = bsize; 251 sblock.e2fs_bshift = sblock.e2fs.e2fs_log_bsize + LOG_MINBSIZE; 252 sblock.e2fs_qbmask = sblock.e2fs_bsize - 1; 253 sblock.e2fs_bmask = ~sblock.e2fs_qbmask; 254 sblock.e2fs_fsbtodb = ilog2(sblock.e2fs_bsize) - ilog2(sectorsize); 255 sblock.e2fs_ipb = sblock.e2fs_bsize / inodesize; 256 257 /* 258 * Ext2fs preserves BBSIZE (1024 bytes) space at the top for 259 * bootloader (though it is not enough at all for our bootloader). 260 * If bsize == BBSIZE we have to preserve one block. 261 * If bsize > BBSIZE, the first block already contains BBSIZE space 262 * before superblock because superblock is allocated at SBOFF and 263 * bsize is a power of two (i.e. 2048 bytes or more). 264 */ 265 sblock.e2fs.e2fs_first_dblock = (sblock.e2fs_bsize > BBSIZE) ? 0 : 1; 266 minfssize = fsbtodb(&sblock, 267 sblock.e2fs.e2fs_first_dblock + 268 NBLOCK_SUPERBLOCK + 269 1 /* at least one group descriptor */ + 270 NBLOCK_BLOCK_BITMAP + 271 NBLOCK_INODE_BITMAP + 272 1 /* at least one inode table block */ + 273 1 /* at least one data block for rootdir */ + 274 1 /* at least one data block for data */ 275 ); /* XXX and more? */ 276 277 if (fssize < minfssize) 278 errx(EXIT_FAILURE, "Filesystem size %" PRId64 279 " < minimum size of %" PRId64 "\n", fssize, minfssize); 280 281 bcount = dbtofsb(&sblock, fssize); 282 283 /* 284 * While many people claim that ext2fs is a (bad) clone of ufs/ffs, 285 * it isn't actual ffs so maybe we should call it "block group" 286 * as their native name rather than ffs derived "cylinder group." 287 * But we'll use the latter here since other kernel sources use it. 288 * (I also agree "cylinder" based allocation is obsolete though) 289 */ 290 291 /* maybe "simple is the best" */ 292 blocks_per_cg = sblock.e2fs_bsize * NBBY; 293 294 ncg = howmany(bcount - sblock.e2fs.e2fs_first_dblock, blocks_per_cg); 295 blocks_gd = howmany(sizeof(struct ext2_gd) * ncg, bsize); 296 297 /* check range of inode number */ 298 if (num_inodes < EXT2_FIRSTINO) 299 num_inodes = EXT2_FIRSTINO; /* needs reserved inodes + 1 */ 300 if (num_inodes > UINT16_MAX * ncg) 301 num_inodes = UINT16_MAX * ncg; /* ext2bgd_nifree is uint16_t */ 302 303 inodes_per_cg = num_inodes / ncg; 304 iblocks_per_cg = howmany(inodesize * inodes_per_cg, bsize); 305 306 /* Check that the last cylinder group has enough space for inodes */ 307 minblocks_per_cg = 308 NBLOCK_BLOCK_BITMAP + 309 NBLOCK_INODE_BITMAP + 310 iblocks_per_cg + 311 1; /* at least one data block */ 312 if (Oflag == 0 || cg_has_sb(ncg - 1) != 0) 313 minblocks_per_cg += NBLOCK_SUPERBLOCK + blocks_gd; 314 315 blocks_lastcg = bcount - sblock.e2fs.e2fs_first_dblock - 316 blocks_per_cg * (ncg - 1); 317 if (blocks_lastcg < minblocks_per_cg) { 318 /* 319 * Since we make all the cylinder groups the same size, the 320 * last will only be small if there are more than one 321 * cylinder groups. If the last one is too small to store 322 * filesystem data, just kill it. 323 * 324 * XXX: Does fsck_ext2fs(8) properly handle this case? 325 */ 326 bcount -= blocks_lastcg; 327 ncg--; 328 blocks_lastcg = blocks_per_cg; 329 blocks_gd = howmany(sizeof(struct ext2_gd) * ncg, bsize); 330 inodes_per_cg = num_inodes / ncg; 331 } 332 /* roundup inodes_per_cg to make it use whole inode table blocks */ 333 inodes_per_cg = roundup(inodes_per_cg, sblock.e2fs_ipb); 334 num_inodes = inodes_per_cg * ncg; 335 iblocks_per_cg = inodes_per_cg / sblock.e2fs_ipb; 336 337 /* XXX: probably we should check these adjusted values again */ 338 339 sblock.e2fs.e2fs_bcount = bcount; 340 sblock.e2fs.e2fs_icount = num_inodes; 341 342 sblock.e2fs_ncg = ncg; 343 sblock.e2fs_ngdb = blocks_gd; 344 sblock.e2fs_itpg = iblocks_per_cg; 345 346 sblock.e2fs.e2fs_rbcount = sblock.e2fs.e2fs_bcount * minfree / 100; 347 /* e2fs_fbcount will be accounted later */ 348 /* e2fs_ficount will be accounted later */ 349 350 sblock.e2fs.e2fs_bpg = blocks_per_cg; 351 sblock.e2fs.e2fs_fpg = blocks_per_cg; 352 353 sblock.e2fs.e2fs_ipg = inodes_per_cg; 354 355 sblock.e2fs.e2fs_mtime = 0; 356 sblock.e2fs.e2fs_wtime = tv.tv_sec; 357 sblock.e2fs.e2fs_mnt_count = 0; 358 /* XXX: should add some entropy to avoid checking all fs at once? */ 359 sblock.e2fs.e2fs_max_mnt_count = EXT2_DEF_MAX_MNT_COUNT; 360 361 sblock.e2fs.e2fs_magic = E2FS_MAGIC; 362 sblock.e2fs.e2fs_state = E2FS_ISCLEAN; 363 sblock.e2fs.e2fs_beh = E2FS_BEH_DEFAULT; 364 sblock.e2fs.e2fs_minrev = 0; 365 sblock.e2fs.e2fs_lastfsck = tv.tv_sec; 366 sblock.e2fs.e2fs_fsckintv = EXT2_DEF_FSCKINTV; 367 368 /* 369 * Maybe we can use E2FS_OS_FREEBSD here and it would be more proper, 370 * but the purpose of this newfs_ext2fs(8) command is to provide 371 * a filesystem which can be recognized by firmware on some 372 * Linux based appliances that can load bootstrap files only from 373 * (their native) ext2fs, and anyway we will (and should) try to 374 * act like them as much as possible. 375 * 376 * Anyway, I hope that all newer such boxes will keep their support 377 * for the "GOOD_OLD_REV" ext2fs. 378 */ 379 sblock.e2fs.e2fs_creator = E2FS_OS_LINUX; 380 381 if (Oflag == 0) { 382 sblock.e2fs.e2fs_rev = E2FS_REV0; 383 sblock.e2fs.e2fs_features_compat = 0; 384 sblock.e2fs.e2fs_features_incompat = 0; 385 sblock.e2fs.e2fs_features_rocompat = 0; 386 } else { 387 sblock.e2fs.e2fs_rev = E2FS_REV1; 388 /* 389 * e2fsprogs say "REV1" is "dynamic" so 390 * it isn't quite a version and maybe it means 391 * "extended from REV0 so check compat features." 392 * 393 * XXX: We don't have any native tool to activate 394 * the EXT2F_COMPAT_RESIZE feature and 395 * fsck_ext2fs(8) might not fix structures for it. 396 */ 397 sblock.e2fs.e2fs_features_compat = EXT2F_COMPAT_RESIZE; 398 sblock.e2fs.e2fs_features_incompat = EXT2F_INCOMPAT_FTYPE; 399 sblock.e2fs.e2fs_features_rocompat = 400 EXT2F_ROCOMPAT_SPARSESUPER | EXT2F_ROCOMPAT_LARGEFILE; 401 } 402 403 sblock.e2fs.e2fs_ruid = geteuid(); 404 sblock.e2fs.e2fs_rgid = getegid(); 405 406 sblock.e2fs.e2fs_first_ino = EXT2_FIRSTINO; 407 sblock.e2fs.e2fs_inode_size = inodesize; 408 409 /* e2fs_block_group_nr is set on writing superblock to each group */ 410 411 uuid_create(&uuid, &uustat); 412 if (uustat != uuid_s_ok) 413 errx(EXIT_FAILURE, "Failed to generate uuid\n"); 414 uuid_enc_le(sblock.e2fs.e2fs_uuid, &uuid); 415 if (volname != NULL) { 416 if (strlen(volname) > sizeof(sblock.e2fs.e2fs_vname)) 417 errx(EXIT_FAILURE, "Volume name is too long"); 418 strlcpy(sblock.e2fs.e2fs_vname, volname, 419 sizeof(sblock.e2fs.e2fs_vname)); 420 } 421 422 sblock.e2fs.e2fs_fsmnt[0] = '\0'; 423 sblock.e2fs_fsmnt[0] = '\0'; 424 425 sblock.e2fs.e2fs_algo = 0; /* XXX unsupported? */ 426 sblock.e2fs.e2fs_prealloc = 0; /* XXX unsupported? */ 427 sblock.e2fs.e2fs_dir_prealloc = 0; /* XXX unsupported? */ 428 429 /* calculate blocks for reserved group descriptors for resize */ 430 sblock.e2fs.e2fs_reserved_ngdb = 0; 431 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 432 (sblock.e2fs.e2fs_features_compat & EXT2F_COMPAT_RESIZE) != 0) { 433 uint64_t target_blocks; 434 uint target_ncg, target_ngdb, reserved_ngdb; 435 436 /* reserve descriptors for size as 1024 times as current */ 437 target_blocks = 438 (sblock.e2fs.e2fs_bcount - sblock.e2fs.e2fs_first_dblock) 439 * 1024ULL; 440 /* number of blocks must be in uint32_t */ 441 if (target_blocks > UINT32_MAX) 442 target_blocks = UINT32_MAX; 443 target_ncg = howmany(target_blocks, sblock.e2fs.e2fs_bpg); 444 target_ngdb = howmany(sizeof(struct ext2_gd) * target_ncg, 445 sblock.e2fs_bsize); 446 /* 447 * Reserved group descriptor blocks are preserved as 448 * the second level double indirect reference blocks in 449 * the EXT2_RESIZEINO inode, so the maximum number of 450 * the blocks is NINDIR(fs). 451 * (see also descriptions in init_resizeino() function) 452 * 453 * We check a number including current e2fs_ngdb here 454 * because they will be moved into reserved gdb on 455 * possible future size shrink, though e2fsprogs don't 456 * seem to care about it. 457 */ 458 if (target_ngdb > NINDIR(&sblock)) 459 target_ngdb = NINDIR(&sblock); 460 461 reserved_ngdb = target_ngdb - sblock.e2fs_ngdb; 462 463 /* make sure reserved_ngdb fits in the last cg */ 464 if (reserved_ngdb >= blocks_lastcg - cgoverhead(ncg - 1)) 465 reserved_ngdb = blocks_lastcg - cgoverhead(ncg - 1); 466 if (reserved_ngdb == 0) { 467 /* if no space for reserved gdb, disable the feature */ 468 sblock.e2fs.e2fs_features_compat &= 469 ~EXT2F_COMPAT_RESIZE; 470 } 471 sblock.e2fs.e2fs_reserved_ngdb = reserved_ngdb; 472 } 473 474 /* 475 * Initialize group descriptors 476 */ 477 gd = malloc(sblock.e2fs_ngdb * bsize); 478 if (gd == NULL) 479 errx(EXIT_FAILURE, "Can't allocate descriptors buffer"); 480 memset(gd, 0, sblock.e2fs_ngdb * bsize); 481 482 fbcount = 0; 483 ficount = 0; 484 for (cylno = 0; cylno < ncg; cylno++) { 485 uint boffset; 486 487 boffset = cgbase(&sblock, cylno); 488 if (sblock.e2fs.e2fs_rev == E2FS_REV0 || 489 (sblock.e2fs.e2fs_features_rocompat & 490 EXT2F_ROCOMPAT_SPARSESUPER) == 0 || 491 cg_has_sb(cylno)) { 492 boffset += NBLOCK_SUPERBLOCK + sblock.e2fs_ngdb; 493 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 494 (sblock.e2fs.e2fs_features_compat & 495 EXT2F_COMPAT_RESIZE) != 0) 496 boffset += sblock.e2fs.e2fs_reserved_ngdb; 497 } 498 gd[cylno].ext2bgd_b_bitmap = boffset; 499 boffset += NBLOCK_BLOCK_BITMAP; 500 gd[cylno].ext2bgd_i_bitmap = boffset; 501 boffset += NBLOCK_INODE_BITMAP; 502 gd[cylno].ext2bgd_i_tables = boffset; 503 if (cylno == (ncg - 1)) 504 gd[cylno].ext2bgd_nbfree = 505 blocks_lastcg - cgoverhead(cylno); 506 else 507 gd[cylno].ext2bgd_nbfree = 508 sblock.e2fs.e2fs_bpg - cgoverhead(cylno); 509 fbcount += gd[cylno].ext2bgd_nbfree; 510 gd[cylno].ext2bgd_nifree = sblock.e2fs.e2fs_ipg; 511 if (cylno == 0) { 512 /* take reserved inodes off nifree */ 513 gd[cylno].ext2bgd_nifree -= EXT2_RESERVED_INODES; 514 } 515 ficount += gd[cylno].ext2bgd_nifree; 516 gd[cylno].ext2bgd_ndirs = 0; 517 } 518 sblock.e2fs.e2fs_fbcount = fbcount; 519 sblock.e2fs.e2fs_ficount = ficount; 520 521 /* 522 * Dump out summary information about file system. 523 */ 524 if (verbosity > 0) { 525 printf("%s: %u.%1uMB (%" PRId64 " sectors) " 526 "block size %u, fragment size %u\n", 527 fsys, 528 (uint)(((uint64_t)bcount * bsize) / (1024 * 1024)), 529 (uint)((uint64_t)bcount * bsize - 530 rounddown((uint64_t)bcount * bsize, 1024 * 1024)) 531 / 1024 / 100, 532 fssize, bsize, fsize); 533 printf("\tusing %u block groups of %u.0MB, %u blks, " 534 "%u inodes.\n", 535 ncg, bsize * sblock.e2fs.e2fs_bpg / (1024 * 1024), 536 sblock.e2fs.e2fs_bpg, sblock.e2fs.e2fs_ipg); 537 } 538 539 /* 540 * allocate space for superblock and group descriptors 541 */ 542 iobufsize = (NBLOCK_SUPERBLOCK + sblock.e2fs_ngdb) * sblock.e2fs_bsize; 543 iobuf = mmap(0, iobufsize, PROT_READ|PROT_WRITE, 544 MAP_ANON|MAP_PRIVATE, -1, 0); 545 if (iobuf == NULL) 546 errx(EXIT_FAILURE, "Cannot allocate I/O buffer\n"); 547 memset(iobuf, 0, iobufsize); 548 549 /* 550 * We now start writing to the filesystem 551 */ 552 553 if (!Nflag) { 554 static const uint pbsize[] = { 1024, 2048, 4096, 0 }; 555 uint pblock, epblock; 556 /* 557 * Validate the given file system size. 558 * Verify that its last block can actually be accessed. 559 * Convert to file system fragment sized units. 560 */ 561 if (fssize <= 0) 562 errx(EXIT_FAILURE, "Preposterous size %" PRId64 "\n", 563 fssize); 564 wtfs(fssize - 1, sectorsize, iobuf); 565 566 /* 567 * Ensure there is nothing that looks like a filesystem 568 * superblock anywhere other than where ours will be. 569 * If fsck_ext2fs finds the wrong one all hell breaks loose! 570 * 571 * XXX: needs to check how fsck_ext2fs programs even 572 * on other OSes determine alternate superblocks 573 */ 574 for (i = 0; pbsize[i] != 0; i++) { 575 epblock = (uint64_t)bcount * bsize / pbsize[i]; 576 for (pblock = ((pbsize[i] == SBSIZE) ? 1 : 0); 577 pblock < epblock; 578 pblock += pbsize[i] * NBBY /* bpg */) 579 zap_old_sblock((daddr_t)pblock * 580 pbsize[i] / sectorsize); 581 } 582 } 583 584 if (verbosity >= 3) 585 printf("super-block backups (for fsck_ext2fs -b #) at:\n"); 586 /* If we are printing more than one line of numbers, line up columns */ 587 fld_width = verbosity < 4 ? 1 : snprintf(NULL, 0, "%" PRIu64, 588 (uint64_t)cgbase(&sblock, ncg - 1)); 589 /* Get terminal width */ 590 if (ioctl(fileno(stdout), TIOCGWINSZ, &winsize) == 0) 591 max_cols = winsize.ws_col; 592 else 593 max_cols = 80; 594 if (Nflag && verbosity == 3) 595 /* Leave space to add " ..." after one row of numbers */ 596 max_cols -= 4; 597 #define BASE 0x10000 /* For some fixed-point maths */ 598 col = 0; 599 delta = verbosity > 2 ? 0 : max_cols * BASE / ncg; 600 for (cylno = 0; cylno < ncg; cylno++) { 601 fflush(stdout); 602 initcg(cylno); 603 if (verbosity < 2) 604 continue; 605 /* the first one is a master, not backup */ 606 if (cylno == 0) 607 continue; 608 /* skip if this cylinder doesn't have a backup */ 609 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 610 (sblock.e2fs.e2fs_features_rocompat & 611 EXT2F_ROCOMPAT_SPARSESUPER) != 0 && 612 cg_has_sb(cylno) == 0) 613 continue; 614 615 if (delta > 0) { 616 if (Nflag) 617 /* No point doing dots for -N */ 618 break; 619 /* Print dots scaled to end near RH margin */ 620 for (col += delta; col > BASE; col -= BASE) 621 printf("."); 622 continue; 623 } 624 /* Print superblock numbers */ 625 len = printf(" %*" PRIu64 "," + !col, fld_width, 626 (uint64_t)cgbase(&sblock, cylno)); 627 col += len; 628 if (col + len < max_cols) 629 /* Next number fits */ 630 continue; 631 /* Next number won't fit, need a newline */ 632 if (verbosity <= 3) { 633 /* Print dots for subsequent cylinder groups */ 634 delta = sblock.e2fs_ncg - cylno - 1; 635 if (delta != 0) { 636 if (Nflag) { 637 printf(" ..."); 638 break; 639 } 640 delta = max_cols * BASE / delta; 641 } 642 } 643 col = 0; 644 printf("\n"); 645 } 646 #undef BASE 647 if (col > 0) 648 printf("\n"); 649 if (Nflag) 650 return; 651 652 /* 653 * Now construct the initial file system, 654 */ 655 if (fsinit(&tv) == 0) 656 errx(EXIT_FAILURE, "Error making filesystem"); 657 /* 658 * Write out the superblock and group descriptors 659 */ 660 sblock.e2fs.e2fs_block_group_nr = 0; 661 sboff = 0; 662 if (cgbase(&sblock, 0) == 0) { 663 /* 664 * If the first block contains the boot block sectors, 665 * (i.e. in case of sblock.e2fs.e2fs_bsize > BBSIZE) 666 * we have to preserve data in it. 667 */ 668 sboff = SBOFF; 669 } 670 e2fs_sbsave(&sblock.e2fs, (struct ext2fs *)(iobuf + sboff)); 671 e2fs_cgsave(gd, (struct ext2_gd *)(iobuf + sblock.e2fs_bsize), 672 sizeof(struct ext2_gd) * sblock.e2fs_ncg); 673 wtfs(fsbtodb(&sblock, cgbase(&sblock, 0)) + sboff / sectorsize, 674 iobufsize - sboff, iobuf + sboff); 675 676 munmap(iobuf, iobufsize); 677 } 678 679 /* 680 * Initialize a cylinder (block) group. 681 */ 682 void 683 initcg(uint cylno) 684 { 685 uint nblcg, i, j, sboff; 686 struct ext2fs_dinode *dp; 687 688 /* 689 * Make a copy of the superblock and group descriptors. 690 */ 691 if (sblock.e2fs.e2fs_rev == E2FS_REV0 || 692 (sblock.e2fs.e2fs_features_rocompat & 693 EXT2F_ROCOMPAT_SPARSESUPER) == 0 || 694 cg_has_sb(cylno)) { 695 sblock.e2fs.e2fs_block_group_nr = cylno; 696 sboff = 0; 697 if (cgbase(&sblock, cylno) == 0) { 698 /* preserve data in bootblock in cg0 */ 699 sboff = SBOFF; 700 } 701 e2fs_sbsave(&sblock.e2fs, (struct ext2fs *)(iobuf + sboff)); 702 e2fs_cgsave(gd, (struct ext2_gd *)(iobuf + 703 sblock.e2fs_bsize * NBLOCK_SUPERBLOCK), 704 sizeof(struct ext2_gd) * sblock.e2fs_ncg); 705 /* write superblock and group descriptor backups */ 706 wtfs(fsbtodb(&sblock, cgbase(&sblock, cylno)) + 707 sboff / sectorsize, iobufsize - sboff, iobuf + sboff); 708 } 709 710 /* 711 * Initialize block bitmap. 712 */ 713 memset(buf, 0, sblock.e2fs_bsize); 714 if (cylno == (sblock.e2fs_ncg - 1)) { 715 /* The last group could have less blocks than e2fs_bpg. */ 716 nblcg = sblock.e2fs.e2fs_bcount - 717 cgbase(&sblock, sblock.e2fs_ncg - 1); 718 for (i = nblcg; i < roundup(nblcg, NBBY); i++) 719 setbit(buf, i); 720 memset(&buf[i / NBBY], ~0U, sblock.e2fs.e2fs_bpg - i); 721 } 722 /* set overhead (superblock, group descriptor etc.) blocks used */ 723 for (i = 0; i < cgoverhead(cylno) / NBBY; i++) 724 buf[i] = ~0; 725 i = i * NBBY; 726 for (; i < cgoverhead(cylno); i++) 727 setbit(buf, i); 728 wtfs(fsbtodb(&sblock, gd[cylno].ext2bgd_b_bitmap), sblock.e2fs_bsize, 729 buf); 730 731 /* 732 * Initialize inode bitmap. 733 * 734 * Assume e2fs_ipg is a multiple of NBBY since 735 * it's a multiple of e2fs_ipb (as we did above). 736 * Note even (possibly smaller) the last group has the same e2fs_ipg. 737 */ 738 i = sblock.e2fs.e2fs_ipg / NBBY; 739 memset(buf, 0, i); 740 memset(buf + i, ~0U, sblock.e2fs_bsize - i); 741 if (cylno == 0) { 742 /* mark reserved inodes */ 743 for (i = 1; i < EXT2_FIRSTINO; i++) 744 setbit(buf, EXT2_INO_INDEX(i)); 745 } 746 wtfs(fsbtodb(&sblock, gd[cylno].ext2bgd_i_bitmap), sblock.e2fs_bsize, 747 buf); 748 749 /* 750 * Initialize inode tables. 751 * 752 * Just initialize generation numbers for NFS security. 753 * XXX: sys/ufs/ext2fs/ext2fs_alloc.c:ext2fs_valloc() seems 754 * to override these generated numbers. 755 */ 756 memset(buf, 0, sblock.e2fs_bsize); 757 for (i = 0; i < sblock.e2fs_itpg; i++) { 758 for (j = 0; j < sblock.e2fs_ipb; j++) { 759 dp = (struct ext2fs_dinode *)(buf + inodesize * j); 760 /* h2fs32() just for consistency */ 761 dp->e2di_gen = h2fs32(arc4random()); 762 } 763 wtfs(fsbtodb(&sblock, gd[cylno].ext2bgd_i_tables + i), 764 sblock.e2fs_bsize, buf); 765 } 766 } 767 768 /* 769 * Zap possible lingering old superblock data 770 */ 771 static void 772 zap_old_sblock(daddr_t sec) 773 { 774 static daddr_t cg0_data; 775 uint32_t oldfs[SBSIZE / sizeof(uint32_t)]; 776 static const struct fsm { 777 uint32_t offset; 778 uint32_t magic; 779 uint32_t mask; 780 } fs_magics[] = { 781 {offsetof(struct ext2fs, e2fs_magic) / 4, E2FS_MAGIC, 0xffff}, 782 {offsetof(struct ext2fs, e2fs_magic) / 4, 783 E2FS_MAGIC << 16, 0xffff0000}, 784 {14, 0xef530000, 0xffff0000}, /* EXT2FS (big) */ 785 {0x55c / 4, 0x00011954, ~0U}, /* FS_UFS1_MAGIC */ 786 {0x55c / 4, 0x19540119, ~0U}, /* FS_UFS2_MAGIC */ 787 {0, 0x70162, ~0U}, /* LFS_MAGIC */ 788 {.offset = ~0U}, 789 }; 790 const struct fsm *fsm; 791 792 if (Nflag) 793 return; 794 795 /* don't override data before superblock */ 796 if (sec < SBOFF / sectorsize) 797 return; 798 799 if (cg0_data == 0) { 800 cg0_data = 801 ((daddr_t)sblock.e2fs.e2fs_first_dblock + cgoverhead(0)) * 802 sblock.e2fs_bsize / sectorsize; 803 } 804 805 /* Ignore anything that is beyond our filesystem */ 806 if (sec >= fssize) 807 return; 808 /* Zero anything inside our filesystem... */ 809 if (sec >= sblock.e2fs.e2fs_first_dblock * bsize / sectorsize) { 810 /* ...unless we will write that area anyway */ 811 if (sec >= cg0_data) 812 /* assume iobuf is zero'ed here */ 813 wtfs(sec, roundup(SBSIZE, sectorsize), iobuf); 814 return; 815 } 816 817 /* 818 * The sector might contain boot code, so we must validate it 819 * 820 * XXX: ext2fs won't preserve data after SBOFF, 821 * but first_dblock could have a different value. 822 */ 823 rdfs(sec, sizeof(oldfs), &oldfs); 824 for (fsm = fs_magics;; fsm++) { 825 uint32_t v; 826 if (fsm->mask == 0) 827 return; 828 v = oldfs[fsm->offset]; 829 if ((v & fsm->mask) == fsm->magic || 830 (bswap32(v) & fsm->mask) == fsm->magic) 831 break; 832 } 833 834 /* Just zap the magic number */ 835 oldfs[fsm->offset] = 0; 836 wtfs(sec, sizeof(oldfs), &oldfs); 837 } 838 839 /* 840 * uint cgoverhead(uint c) 841 * 842 * Return a number of reserved blocks on the specified group. 843 * XXX: should be shared with src/sbin/fsck_ext2fs/setup.c 844 */ 845 uint 846 cgoverhead(uint c) 847 { 848 uint overh; 849 850 overh = NBLOCK_BLOCK_BITMAP + NBLOCK_INODE_BITMAP + sblock.e2fs_itpg; 851 852 if (sblock.e2fs.e2fs_rev == E2FS_REV0 || 853 (sblock.e2fs.e2fs_features_rocompat & 854 EXT2F_ROCOMPAT_SPARSESUPER) == 0 || 855 cg_has_sb(c) != 0) { 856 overh += NBLOCK_SUPERBLOCK + sblock.e2fs_ngdb; 857 858 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 859 (sblock.e2fs.e2fs_features_compat & 860 EXT2F_COMPAT_RESIZE) != 0) 861 overh += sblock.e2fs.e2fs_reserved_ngdb; 862 } 863 864 return overh; 865 } 866 867 /* 868 * Initialize the file system 869 */ 870 871 #define LOSTDIR /* e2fsck complains if there is no lost+found */ 872 873 #define PREDEFDIR 2 874 875 #ifdef LOSTDIR 876 #define PREDEFROOTDIR (PREDEFDIR + 1) 877 #else 878 #define PREDEFROOTDIR PREDEFDIR 879 #endif 880 881 struct ext2fs_direct root_dir[] = { 882 { EXT2_ROOTINO, 0, 1, 0, "." }, 883 { EXT2_ROOTINO, 0, 2, 0, ".." }, 884 #ifdef LOSTDIR 885 { EXT2_LOSTFOUNDINO, 0, 10, 0, "lost+found" }, 886 #endif 887 }; 888 889 #ifdef LOSTDIR 890 struct ext2fs_direct lost_found_dir[] = { 891 { EXT2_LOSTFOUNDINO, 0, 1, 0, "." }, 892 { EXT2_ROOTINO, 0, 2, 0, ".." }, 893 }; 894 struct ext2fs_direct pad_dir = { 0, sizeof(struct ext2fs_direct), 0, 0, "" }; 895 #endif 896 897 int 898 fsinit(const struct timeval *tv) 899 { 900 struct ext2fs_dinode node; 901 #ifdef LOSTDIR 902 uint i, nblks_lostfound, blk; 903 #endif 904 905 /* 906 * Initialize the inode for the resizefs feature 907 */ 908 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 909 (sblock.e2fs.e2fs_features_compat & EXT2F_COMPAT_RESIZE) != 0) 910 init_resizeino(tv); 911 912 /* 913 * Initialize the node 914 */ 915 916 #ifdef LOSTDIR 917 /* 918 * Create the lost+found directory 919 */ 920 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 921 sblock.e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE) { 922 lost_found_dir[0].e2d_type = EXT2_FT_DIR; 923 lost_found_dir[1].e2d_type = EXT2_FT_DIR; 924 } 925 (void)makedir(lost_found_dir, __arraycount(lost_found_dir)); 926 927 /* prepare a bit large directory for preserved files */ 928 nblks_lostfound = EXT2_LOSTFOUNDSIZE / sblock.e2fs_bsize; 929 /* ...but only with direct blocks */ 930 if (nblks_lostfound > NDADDR) 931 nblks_lostfound = NDADDR; 932 933 memset(&node, 0, sizeof(node)); 934 node.e2di_mode = EXT2_IFDIR | EXT2_LOSTFOUNDUMASK; 935 node.e2di_uid = geteuid(); 936 node.e2di_size = sblock.e2fs_bsize * nblks_lostfound; 937 node.e2di_atime = tv->tv_sec; 938 node.e2di_ctime = tv->tv_sec; 939 node.e2di_mtime = tv->tv_sec; 940 node.e2di_gid = getegid(); 941 node.e2di_nlink = PREDEFDIR; 942 /* e2di_nblock is a number of disk blocks, not ext2fs blocks */ 943 node.e2di_nblock = fsbtodb(&sblock, nblks_lostfound); 944 node.e2di_blocks[0] = alloc(sblock.e2fs_bsize, node.e2di_mode); 945 if (node.e2di_blocks[0] == 0) { 946 printf("%s: can't allocate block for lost+found\n", __func__); 947 return 0; 948 } 949 for (i = 1; i < nblks_lostfound; i++) { 950 blk = alloc(sblock.e2fs_bsize, 0); 951 if (blk == 0) { 952 printf("%s: can't allocate blocks for lost+found\n", 953 __func__); 954 return 0; 955 } 956 node.e2di_blocks[i] = blk; 957 } 958 wtfs(fsbtodb(&sblock, node.e2di_blocks[0]), sblock.e2fs_bsize, buf); 959 pad_dir.e2d_reclen = sblock.e2fs_bsize; 960 for (i = 1; i < nblks_lostfound; i++) { 961 memset(buf, 0, sblock.e2fs_bsize); 962 copy_dir(&pad_dir, (struct ext2fs_direct *)buf); 963 wtfs(fsbtodb(&sblock, node.e2di_blocks[i]), sblock.e2fs_bsize, 964 buf); 965 } 966 iput(&node, EXT2_LOSTFOUNDINO); 967 #endif 968 /* 969 * create the root directory 970 */ 971 memset(&node, 0, sizeof(node)); 972 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 973 sblock.e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE) { 974 root_dir[0].e2d_type = EXT2_FT_DIR; 975 root_dir[1].e2d_type = EXT2_FT_DIR; 976 #ifdef LOSTDIR 977 root_dir[2].e2d_type = EXT2_FT_DIR; 978 #endif 979 } 980 node.e2di_mode = EXT2_IFDIR | EXT2_UMASK; 981 node.e2di_uid = geteuid(); 982 node.e2di_size = makedir(root_dir, __arraycount(root_dir)); 983 node.e2di_atime = tv->tv_sec; 984 node.e2di_ctime = tv->tv_sec; 985 node.e2di_mtime = tv->tv_sec; 986 node.e2di_gid = getegid(); 987 node.e2di_nlink = PREDEFROOTDIR; 988 /* e2di_nblock is a number of disk block, not ext2fs block */ 989 node.e2di_nblock = fsbtodb(&sblock, 1); 990 node.e2di_blocks[0] = alloc(node.e2di_size, node.e2di_mode); 991 if (node.e2di_blocks[0] == 0) { 992 printf("%s: can't allocate block for root dir\n", __func__); 993 return 0; 994 } 995 wtfs(fsbtodb(&sblock, node.e2di_blocks[0]), sblock.e2fs_bsize, buf); 996 iput(&node, EXT2_ROOTINO); 997 return 1; 998 } 999 1000 /* 1001 * Construct a set of directory entries in "buf". 1002 * return size of directory. 1003 */ 1004 int 1005 makedir(struct ext2fs_direct *protodir, int entries) 1006 { 1007 uint8_t *cp; 1008 uint i, spcleft; 1009 uint dirblksiz; 1010 1011 dirblksiz = sblock.e2fs_bsize; 1012 memset(buf, 0, dirblksiz); 1013 spcleft = dirblksiz; 1014 for (cp = buf, i = 0; i < entries - 1; i++) { 1015 protodir[i].e2d_reclen = EXT2FS_DIRSIZ(protodir[i].e2d_namlen); 1016 copy_dir(&protodir[i], (struct ext2fs_direct *)cp); 1017 cp += protodir[i].e2d_reclen; 1018 spcleft -= protodir[i].e2d_reclen; 1019 } 1020 protodir[i].e2d_reclen = spcleft; 1021 copy_dir(&protodir[i], (struct ext2fs_direct *)cp); 1022 return dirblksiz; 1023 } 1024 1025 /* 1026 * Copy a direntry to a buffer, in fs byte order 1027 */ 1028 static void 1029 copy_dir(struct ext2fs_direct *dir, struct ext2fs_direct *dbuf) 1030 { 1031 1032 memcpy(dbuf, dir, EXT2FS_DIRSIZ(dir->e2d_namlen)); 1033 dbuf->e2d_ino = h2fs32(dir->e2d_ino); 1034 dbuf->e2d_reclen = h2fs16(dir->e2d_reclen); 1035 } 1036 1037 /* 1038 * void init_resizeino(const struct timeval *tv); 1039 * 1040 * Initialize the EXT2_RESEIZE_INO inode to preserve 1041 * reserved group descriptor blocks for future growth of this ext2fs. 1042 */ 1043 void 1044 init_resizeino(const struct timeval *tv) 1045 { 1046 struct ext2fs_dinode node; 1047 uint64_t isize; 1048 uint32_t *dindir_block, *reserved_gdb; 1049 uint nblock, i, cylno, n; 1050 1051 memset(&node, 0, sizeof(node)); 1052 1053 /* 1054 * Note this function only prepares required structures for 1055 * future resize. It's a quite different work to implement 1056 * a utility like resize_ext2fs(8) which handles actual 1057 * resize ops even on offline. 1058 * 1059 * Anyway, I'm not sure if there is any documentation about 1060 * this resize ext2fs feature and related data structures, 1061 * and I've written this function based on things what I see 1062 * on some existing implementation and real file system data 1063 * created by existing tools. To be honest, they are not 1064 * so easy to read, so I will try to implement it here without 1065 * any dumb optimization for people who would eventually 1066 * work on "yet another wheel" like resize_ext2fs(8). 1067 */ 1068 1069 /* 1070 * I'm not sure what type is appropriate for this inode. 1071 * The release notes of e2fsprogs says they changed e2fsck to allow 1072 * IFREG for RESIZEINO since a certain resize tool used it. Hmm. 1073 */ 1074 node.e2di_mode = EXT2_IFREG | EXT2_RESIZEINOUMASK; 1075 node.e2di_uid = geteuid(); 1076 node.e2di_atime = tv->tv_sec; 1077 node.e2di_ctime = tv->tv_sec; 1078 node.e2di_mtime = tv->tv_sec; 1079 node.e2di_gid = getegid(); 1080 node.e2di_nlink = 1; 1081 1082 /* 1083 * To preserve the reserved group descriptor blocks, 1084 * EXT2_RESIZEINO uses only double indirect reference 1085 * blocks in its inode entries. 1086 * 1087 * All entries for direct, single indirect and triple 1088 * indirect references are left zero'ed. Maybe it's safe 1089 * because no write operation will happen with this inode. 1090 * 1091 * We have to allocate a block for the first level double 1092 * indirect reference block. Indexes of inode entries in 1093 * this first level dindirect block are corresponding to 1094 * indexes of group descriptors including both used (e2fs_ngdb) 1095 * and reserved (e2fs_reserved_ngdb) group descriptor blocks. 1096 * 1097 * Inode entries of indexes for used (e2fs_ngdb) descriptors are 1098 * left zero'ed. Entries for reserved (e2fs_reserved_ngdb) ones 1099 * have block numbers of actual reserved group descriptors 1100 * allocated at block group zero. This means e2fs_reserved_ngdb 1101 * blocks are reserved as the second level dindirect reference 1102 * blocks, and they actually contain block numbers of indirect 1103 * references. It may be safe since they don't have to keep any 1104 * data yet. 1105 * 1106 * Each these second dindirect blocks (i.e. reserved group 1107 * descriptor blocks in the first block group) should have 1108 * block numbers of its backups in all other block groups. 1109 * I.e. reserved_ngdb[0] block in block group 0 contains block 1110 * numbers of resreved_ngdb[0] from group 1 through (e2fs_ncg - 1). 1111 * The number of backups can be determined by the 1112 * EXT2_ROCOMPAT_SPARSESUPER feature and cg_has_sb() macro 1113 * as done in the above initcg() function. 1114 */ 1115 1116 /* set e2di_size which occupies whole blocks through DINDIR blocks */ 1117 isize = (uint64_t)sblock.e2fs_bsize * NDADDR + 1118 (uint64_t)sblock.e2fs_bsize * NINDIR(&sblock) + 1119 (uint64_t)sblock.e2fs_bsize * NINDIR(&sblock) * NINDIR(&sblock); 1120 if (isize > UINT32_MAX && 1121 (sblock.e2fs.e2fs_features_rocompat & 1122 EXT2F_ROCOMPAT_LARGEFILE) == 0) { 1123 /* XXX should enable it here and update all backups? */ 1124 errx(EXIT_FAILURE, "%s: large_file rocompat feature is " 1125 "required to enable resize feature for this filesystem\n", 1126 __func__); 1127 } 1128 /* upper 32bit is stored into e2di_dacl on REV1 feature */ 1129 node.e2di_size = isize & UINT32_MAX; 1130 node.e2di_dacl = isize >> 32; 1131 1132 #define SINGLE 0 /* index of single indirect block */ 1133 #define DOUBLE 1 /* index of double indirect block */ 1134 #define TRIPLE 2 /* index of triple indirect block */ 1135 1136 /* zero out entries for direct references */ 1137 for (i = 0; i < NDADDR; i++) 1138 node.e2di_blocks[i] = 0; 1139 /* also zero out entries for single and triple indirect references */ 1140 node.e2di_blocks[NDADDR + SINGLE] = 0; 1141 node.e2di_blocks[NDADDR + TRIPLE] = 0; 1142 1143 /* allocate a block for the first level double indirect reference */ 1144 node.e2di_blocks[NDADDR + DOUBLE] = 1145 alloc(sblock.e2fs_bsize, node.e2di_mode); 1146 if (node.e2di_blocks[NDADDR + DOUBLE] == 0) 1147 errx(EXIT_FAILURE, "%s: Can't allocate a dindirect block", 1148 __func__); 1149 1150 /* account this first block */ 1151 nblock = fsbtodb(&sblock, 1); 1152 1153 /* allocate buffer to set data in the dindirect block */ 1154 dindir_block = malloc(sblock.e2fs_bsize); 1155 if (dindir_block == NULL) 1156 errx(EXIT_FAILURE, 1157 "%s: Can't allocate buffer for a dindirect block", 1158 __func__); 1159 1160 /* allocate buffer to set data in the group descriptor blocks */ 1161 reserved_gdb = malloc(sblock.e2fs_bsize); 1162 if (reserved_gdb == NULL) 1163 errx(EXIT_FAILURE, 1164 "%s: Can't allocate buffer for group descriptor blocks", 1165 __func__); 1166 1167 /* 1168 * Setup block entries in the first level dindirect blocks 1169 */ 1170 for (i = 0; i < sblock.e2fs_ngdb; i++) { 1171 /* no need to handle used group descriptor blocks */ 1172 dindir_block[i] = 0; 1173 } 1174 for (; i < sblock.e2fs_ngdb + sblock.e2fs.e2fs_reserved_ngdb; i++) { 1175 /* 1176 * point reserved group descriptor block in the first 1177 * (i.e. master) block group 1178 * 1179 * XXX: e2fsprogs seem to use "(i % NINDIR(&sblock))" here 1180 * to store maximum NINDIR(&sblock) reserved gdbs. 1181 * I'm not sure what will be done on future filesystem 1182 * shrink in that case on their way. 1183 */ 1184 if (i >= NINDIR(&sblock)) 1185 errx(EXIT_FAILURE, "%s: too many reserved " 1186 "group descriptors (%u) for resize inode", 1187 __func__, sblock.e2fs.e2fs_reserved_ngdb); 1188 dindir_block[i] = 1189 h2fs32(cgbase(&sblock, 0) + NBLOCK_SUPERBLOCK + i); 1190 1191 /* 1192 * Setup block entries in the second dindirect blocks 1193 * (which are primary reserved group descriptor blocks) 1194 * to point their backups. 1195 */ 1196 for (n = 0, cylno = 1; cylno < sblock.e2fs_ncg; cylno++) { 1197 /* skip block groups without backup */ 1198 if ((sblock.e2fs.e2fs_features_rocompat & 1199 EXT2F_ROCOMPAT_SPARSESUPER) != 0 && 1200 cg_has_sb(cylno) == 0) 1201 continue; 1202 1203 if (n >= NINDIR(&sblock)) 1204 errx(EXIT_FAILURE, "%s: too many block groups " 1205 "for the resize feature", __func__); 1206 /* 1207 * These blocks are already reserved in 1208 * initcg() so no need to use alloc() here. 1209 */ 1210 reserved_gdb[n++] = h2fs32(cgbase(&sblock, cylno) + 1211 NBLOCK_SUPERBLOCK + i); 1212 nblock += fsbtodb(&sblock, 1); 1213 } 1214 for (; n < NINDIR(&sblock); n++) 1215 reserved_gdb[n] = 0; 1216 1217 /* write group descriptor block as the second dindirect refs */ 1218 wtfs(fsbtodb(&sblock, fs2h32(dindir_block[i])), 1219 sblock.e2fs_bsize, reserved_gdb); 1220 nblock += fsbtodb(&sblock, 1); 1221 } 1222 for (; i < NINDIR(&sblock); i++) { 1223 /* leave trailing entries unallocated */ 1224 dindir_block[i] = 0; 1225 } 1226 free(reserved_gdb); 1227 1228 /* finally write the first level dindirect block */ 1229 wtfs(fsbtodb(&sblock, node.e2di_blocks[NDADDR + DOUBLE]), 1230 sblock.e2fs_bsize, dindir_block); 1231 free(dindir_block); 1232 1233 node.e2di_nblock = nblock; 1234 iput(&node, EXT2_RESIZEINO); 1235 } 1236 1237 /* 1238 * uint32_t alloc(uint32_t size, uint16_t mode) 1239 * 1240 * Allocate a block (from cylinder group 0) 1241 * Reference: src/sys/ufs/ext2fs/ext2fs_alloc.c:ext2fs_alloccg() 1242 */ 1243 uint32_t 1244 alloc(uint32_t size, uint16_t mode) 1245 { 1246 uint32_t loc, bno; 1247 uint8_t *bbp; 1248 uint len, map, i; 1249 1250 if (gd[0].ext2bgd_nbfree == 0) 1251 return 0; 1252 1253 if (size > sblock.e2fs_bsize) 1254 return 0; 1255 1256 bbp = malloc(sblock.e2fs_bsize); 1257 if (bbp == NULL) 1258 return 0; 1259 rdfs(fsbtodb(&sblock, gd[0].ext2bgd_b_bitmap), sblock.e2fs_bsize, bbp); 1260 1261 /* XXX: kernel uses e2fs_fpg here */ 1262 len = sblock.e2fs.e2fs_bpg / NBBY; 1263 1264 #if 0 /* no need block allocation for root or lost+found dir */ 1265 for (loc = 0; loc < len; loc++) { 1266 if (bbp[loc] == 0) { 1267 bno = loc * NBBY; 1268 goto gotit; 1269 } 1270 } 1271 #endif 1272 1273 loc = skpc(~0U, len, bbp); 1274 if (loc == 0) 1275 return 0; 1276 loc = len - loc; 1277 map = bbp[loc]; 1278 bno = loc * NBBY; 1279 for (i = 0; i < NBBY; i++, bno++) { 1280 if ((map & (1 << i)) == 0) 1281 goto gotit; 1282 } 1283 return 0; 1284 1285 gotit: 1286 if (isset(bbp, bno)) 1287 errx(EXIT_FAILURE, "%s: inconsistent bitmap\n", __func__); 1288 1289 setbit(bbp, bno); 1290 wtfs(fsbtodb(&sblock, gd[0].ext2bgd_b_bitmap), sblock.e2fs_bsize, bbp); 1291 free(bbp); 1292 /* XXX: modified group descriptors won't be written into backups */ 1293 gd[0].ext2bgd_nbfree--; 1294 if ((mode & EXT2_IFDIR) != 0) 1295 gd[0].ext2bgd_ndirs++; 1296 sblock.e2fs.e2fs_fbcount--; 1297 1298 return sblock.e2fs.e2fs_first_dblock + bno; 1299 } 1300 1301 /* 1302 * void iput(struct ext2fs_dinode *ip, ino_t ino) 1303 * 1304 * Put an inode entry into the corresponding table. 1305 */ 1306 static void 1307 iput(struct ext2fs_dinode *ip, ino_t ino) 1308 { 1309 daddr_t d; 1310 uint c, i; 1311 struct ext2fs_dinode *dp; 1312 uint8_t *bp; 1313 1314 bp = malloc(sblock.e2fs_bsize); 1315 if (bp == NULL) 1316 errx(EXIT_FAILURE, "%s: can't allocate buffer for inode\n", 1317 __func__); 1318 1319 /* 1320 * Reserved inodes are allocated and accounted in initcg() 1321 * so skip checks of the bitmap and allocation for them. 1322 */ 1323 if (ino >= EXT2_FIRSTINO) { 1324 c = ino_to_cg(&sblock, ino); 1325 1326 /* sanity check */ 1327 if (gd[c].ext2bgd_nifree == 0) 1328 errx(EXIT_FAILURE, 1329 "%s: no free inode %" PRIu64 " in block group %u\n", 1330 __func__, (uint64_t)ino, c); 1331 1332 /* update inode bitmap */ 1333 rdfs(fsbtodb(&sblock, gd[0].ext2bgd_i_bitmap), 1334 sblock.e2fs_bsize, bp); 1335 1336 /* more sanity */ 1337 if (isset(bp, EXT2_INO_INDEX(ino))) 1338 errx(EXIT_FAILURE, "%s: inode %" PRIu64 1339 " already in use\n", __func__, (uint64_t)ino); 1340 setbit(bp, EXT2_INO_INDEX(ino)); 1341 wtfs(fsbtodb(&sblock, gd[0].ext2bgd_i_bitmap), 1342 sblock.e2fs_bsize, bp); 1343 gd[c].ext2bgd_nifree--; 1344 sblock.e2fs.e2fs_ficount--; 1345 } 1346 1347 if (ino >= sblock.e2fs.e2fs_ipg * sblock.e2fs_ncg) 1348 errx(EXIT_FAILURE, "%s: inode value out of range (%" PRIu64 1349 ").\n", __func__, (uint64_t)ino); 1350 1351 /* update an inode entry in the table */ 1352 d = fsbtodb(&sblock, ino_to_fsba(&sblock, ino)); 1353 rdfs(d, sblock.e2fs_bsize, bp); 1354 1355 dp = (struct ext2fs_dinode *)(bp + 1356 inodesize * ino_to_fsbo(&sblock, ino)); 1357 e2fs_isave(ip, dp); 1358 /* e2fs_i_bswap() doesn't swap e2di_blocks addrs */ 1359 if ((ip->e2di_mode & EXT2_IFMT) != EXT2_IFLNK) { 1360 for (i = 0; i < NDADDR + NIADDR; i++) 1361 dp->e2di_blocks[i] = h2fs32(ip->e2di_blocks[i]); 1362 } 1363 /* h2fs32() just for consistency */ 1364 dp->e2di_gen = h2fs32(arc4random()); 1365 1366 wtfs(d, sblock.e2fs_bsize, bp); 1367 free(bp); 1368 } 1369 1370 /* 1371 * Read a block from the file system 1372 */ 1373 void 1374 rdfs(daddr_t bno, int size, void *bf) 1375 { 1376 int n; 1377 off_t offset; 1378 1379 offset = bno; 1380 n = pread(fsi, bf, size, offset * sectorsize); 1381 if (n != size) 1382 err(EXIT_FAILURE, "%s: read error for sector %" PRId64, 1383 __func__, (int64_t)bno); 1384 } 1385 1386 /* 1387 * Write a block to the file system 1388 */ 1389 void 1390 wtfs(daddr_t bno, int size, void *bf) 1391 { 1392 int n; 1393 off_t offset; 1394 1395 if (Nflag) 1396 return; 1397 offset = bno; 1398 n = pwrite(fso, bf, size, offset * sectorsize); 1399 if (n != size) 1400 err(EXIT_FAILURE, "%s: write error for sector %" PRId64, 1401 __func__, (int64_t)bno); 1402 } 1403 1404 int 1405 ilog2(uint val) 1406 { 1407 1408 if (val == 0 || !powerof2(val)) 1409 errx(EXIT_FAILURE, "%s: %u is not a power of 2\n", 1410 __func__, val); 1411 1412 return ffs(val) - 1; 1413 } 1414 1415 /* 1416 * int skpc(int mask, size_t size, uint8_t *cp) 1417 * 1418 * Locate an unsigned character of value mask inside cp[]. 1419 * (from src/sys/lib/libkern/skpc.c) 1420 */ 1421 int 1422 skpc(int mask, size_t size, uint8_t *cp) 1423 { 1424 uint8_t *end; 1425 1426 end = &cp[size]; 1427 while (cp < end && *cp == (uint8_t)mask) 1428 cp++; 1429 1430 return end - cp; 1431 } 1432