xref: /netbsd-src/sbin/newfs_ext2fs/mke2fs.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: mke2fs.c,v 1.7 2007/12/14 13:19:35 tsutsui Exp $	*/
2 
3 /*
4  * Copyright (c) 2007 Izumi Tsutsui.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. The name of the author may not be used to endorse or promote products
15  *    derived from this software without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /*
30  * Copyright (c) 1980, 1989, 1993
31  *	The Regents of the University of California.  All rights reserved.
32  *
33  * Redistribution and use in source and binary forms, with or without
34  * modification, are permitted provided that the following conditions
35  * are met:
36  * 1. Redistributions of source code must retain the above copyright
37  *    notice, this list of conditions and the following disclaimer.
38  * 2. Redistributions in binary form must reproduce the above copyright
39  *    notice, this list of conditions and the following disclaimer in the
40  *    documentation and/or other materials provided with the distribution.
41  * 3. Neither the name of the University nor the names of its contributors
42  *    may be used to endorse or promote products derived from this software
43  *    without specific prior written permission.
44  *
45  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
46  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
49  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55  * SUCH DAMAGE.
56  */
57 
58 /*
59  * Copyright (c) 1997 Manuel Bouyer.
60  *
61  * Redistribution and use in source and binary forms, with or without
62  * modification, are permitted provided that the following conditions
63  * are met:
64  * 1. Redistributions of source code must retain the above copyright
65  *	notice, this list of conditions and the following disclaimer.
66  * 2. Redistributions in binary form must reproduce the above copyright
67  *	notice, this list of conditions and the following disclaimer in the
68  *	documentation and/or other materials provided with the distribution.
69  * 3. All advertising materials mentioning features or use of this software
70  *	must display the following acknowledgement:
71  *	This product includes software developed by Manuel Bouyer.
72  * 4. The name of the author may not be used to endorse or promote products
73  *    derived from this software without specific prior written permission.
74  *
75  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
76  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
77  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
78  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
79  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
80  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
81  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
82  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
83  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
84  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
85  */
86 
87 /*
88  * mke2fs.c: "re-invent (dumb but non-GPLed) wheel as a fun project"
89  *
90  *	In spite of this name, there is no piece of code
91  *	derived from GPLed e2fsprogs written for Linux.
92  *	I referred them only to see how each structure
93  *	member should be initialized.
94  *
95  * Reference:
96  *	- All NetBSD sources under src/sys/ufs/ext2fs and src/sbin/fsck_ext2fs
97  *	- Ext2fs Home Page
98  *		http://e2fsprogs.sourceforge.net/ext2.html
99  *	- Design and Implementation of the Second Extended Filesystem
100  *		http://e2fsprogs.sourceforge.net/ext2intro.html
101  *	- Linux Documentation "The Second Extended Filesystem"
102  *		src/linux/Documentation/filesystems/ext2.txt
103  *		    in the Linux kernel distribution
104  */
105 
106 #include <sys/cdefs.h>
107 #ifndef lint
108 #if 0
109 static char sccsid[] = "@(#)mkfs.c	8.11 (Berkeley) 5/3/95";
110 #else
111 __RCSID("$NetBSD: mke2fs.c,v 1.7 2007/12/14 13:19:35 tsutsui Exp $");
112 #endif
113 #endif /* not lint */
114 
115 #include <sys/param.h>
116 #include <sys/mman.h>
117 #include <sys/time.h>
118 #include <ufs/ext2fs/ext2fs_dinode.h>
119 #include <ufs/ext2fs/ext2fs_dir.h>
120 #include <ufs/ext2fs/ext2fs.h>
121 #include <sys/ioctl.h>
122 
123 #include <err.h>
124 #include <errno.h>
125 #include <string.h>
126 #include <unistd.h>
127 #include <stdlib.h>
128 #include <stddef.h>
129 #include <stdio.h>
130 #include <uuid.h>
131 
132 #include "extern.h"
133 
134 static void initcg(uint);
135 static void zap_old_sblock(daddr_t);
136 static uint cgoverhead(uint);
137 static int fsinit(const struct timeval *);
138 static int makedir(struct ext2fs_direct *, int);
139 static void copy_dir(struct ext2fs_direct *, struct ext2fs_direct *);
140 static void init_resizeino(const struct timeval *);
141 static uint32_t alloc(uint32_t, uint16_t);
142 static void iput(struct ext2fs_dinode *, ino_t);
143 static void rdfs(daddr_t, int, void *);
144 static void wtfs(daddr_t, int, void *);
145 static int ilog2(uint);
146 static int skpc(int, size_t, uint8_t *);
147 
148 /* XXX: some of these macro should be into <ufs/ext2fs/ext2fs.h>? */
149 #define EXT2_DEF_MAX_MNT_COUNT	20
150 #define EXT2_DEF_FSCKINTV	(180 * 24 * 60 * 60)	/* 180 days */
151 #define EXT2_RESERVED_INODES	(EXT2_FIRSTINO - 1)
152 #define EXT2_UMASK		0755
153 
154 #define EXT2_INO_INDEX(ino)	((ino) - 1)		/* no inode zero */
155 
156 #define EXT2_LOSTFOUNDSIZE	16384
157 #define EXT2_LOSTFOUNDINO	EXT2_FIRSTINO		/* XXX: not quite */
158 #define EXT2_LOSTFOUNDUMASK	0700
159 
160 #define EXT2_RESIZEINOUMASK	0600
161 
162 #define NBLOCK_SUPERBLOCK	1
163 #define NBLOCK_BLOCK_BITMAP	1
164 #define NBLOCK_INODE_BITMAP	1
165 
166 #define cgbase(fs, c)	\
167 	((fs)->e2fs.e2fs_first_dblock + (fs)->e2fs.e2fs_bpg * (c))
168 
169 
170 /*
171  * ext2fs super block and group descriptor structures
172  *
173  *   We don't have to use or setup whole in-memory m_ext2fs structure,
174  *   but prepare it to use several macro defined in kernel headers.
175  */
176 union {
177 	struct m_ext2fs m_ext2fs;
178 	char pad[SBSIZE];
179 } ext2fsun;
180 #define sblock	ext2fsun.m_ext2fs
181 #define gd	ext2fsun.m_ext2fs.e2fs_gd
182 
183 static uint8_t *iobuf;		/* for superblock and group descriptors */
184 static int iobufsize;
185 
186 static uint8_t buf[MAXBSIZE];	/* for initcg() and makedir() ops */
187 
188 static int fsi, fso;
189 
190 void
191 mke2fs(const char *fsys, int fi, int fo)
192 {
193 	struct timeval tv;
194 	int64_t minfssize;
195 	uint bcount, fbcount, ficount;
196 	uint blocks_gd, blocks_per_cg, inodes_per_cg, iblocks_per_cg;
197 	uint minblocks_per_cg, blocks_lastcg;
198 	uint ncg, cylno, sboff;
199 	uuid_t uuid;
200 	uint32_t uustat;
201 	int i, len, col, delta, fld_width, max_cols;
202 	struct winsize winsize;
203 
204 	gettimeofday(&tv, NULL);
205 	fsi = fi;
206 	fso = fo;
207 
208 	/*
209 	 * collect and verify the block and fragment sizes
210 	 */
211 	if (!powerof2(bsize)) {
212 		errx(EXIT_FAILURE,
213 		    "block size must be a power of 2, not %u\n",
214 		    bsize);
215 	}
216 	if (!powerof2(fsize)) {
217 		errx(EXIT_FAILURE,
218 		    "fragment size must be a power of 2, not %u\n",
219 		    fsize);
220 	}
221 	if (fsize < sectorsize) {
222 		errx(EXIT_FAILURE,
223 		    "fragment size %u is too small, minimum is %u\n",
224 		    fsize, sectorsize);
225 	}
226 	if (bsize < MINBSIZE) {
227 		errx(EXIT_FAILURE,
228 		    "block size %u is too small, minimum is %u\n",
229 		    bsize, MINBSIZE);
230 	}
231 	if (bsize > EXT2_MAXBSIZE) {
232 		errx(EXIT_FAILURE,
233 		    "block size %u is too large, maximum is %u\n",
234 		    bsize, MAXBSIZE);
235 	}
236 	if (bsize != fsize) {
237 		/*
238 		 * There is no fragment support on current ext2fs (yet?),
239 		 * but some kernel code refers fsize or fpg as bsize or bpg
240 		 * and Linux seems to set the same values to them.
241 		 */
242 		errx(EXIT_FAILURE,
243 		    "block size (%u) can't be diffrent from "
244 		    "fragment size (%u)\n",
245 		    bsize, fsize);
246 	}
247 
248 	sblock.e2fs.e2fs_log_bsize = ilog2(bsize) - LOG_MINBSIZE;
249 	/* Umm, why not e2fs_log_fsize? */
250 	sblock.e2fs.e2fs_fsize = ilog2(fsize) - LOG_MINBSIZE;
251 
252 	sblock.e2fs_bsize = bsize;
253 	sblock.e2fs_bshift = sblock.e2fs.e2fs_log_bsize + LOG_MINBSIZE;
254 	sblock.e2fs_qbmask = sblock.e2fs_bsize - 1;
255 	sblock.e2fs_bmask = ~sblock.e2fs_qbmask;
256 	sblock.e2fs_fsbtodb = ilog2(sblock.e2fs_bsize) - ilog2(sectorsize);
257 	sblock.e2fs_ipb = sblock.e2fs_bsize / EXT2_DINODE_SIZE;
258 
259 	/*
260 	 * Ext2fs preseves BBSIZE (1024 bytes) space at the top for
261 	 * bootloader (though it is not enough at all for our bootloader).
262 	 * If bsize == BBSIZE we have to preserve one block.
263 	 * If bsize > BBSIZE, the first block already contains BBSIZE space
264 	 * before superblock because superblock is allocated at SBOFF and
265 	 * bsize is a power of two (i.e. 2048 bytes or more).
266 	 */
267 	sblock.e2fs.e2fs_first_dblock = (sblock.e2fs_bsize > BBSIZE) ? 0 : 1;
268 	minfssize = fsbtodb(&sblock,
269 	    sblock.e2fs.e2fs_first_dblock +
270 	    NBLOCK_SUPERBLOCK +
271 	    1 /* at least one group descriptor */ +
272 	    NBLOCK_BLOCK_BITMAP	+
273 	    NBLOCK_INODE_BITMAP +
274 	    1 /* at least one inode table block */ +
275 	    1 /* at least one data block for rootdir */ +
276 	    1 /* at least one data block for data */
277 	    );			/* XXX and more? */
278 
279 	if (fssize < minfssize)
280 		errx(EXIT_FAILURE, "Filesystem size %" PRId64
281 		    " < minimum size of %" PRId64 "\n", fssize, minfssize);
282 
283 	bcount = dbtofsb(&sblock, fssize);
284 
285 	/*
286 	 * While many people claim that ext2fs is a (bad) clone of ufs/ffs,
287 	 * it isn't actual ffs so maybe we should call it "block group"
288 	 * as their native name rather than ffs delived "cylinder group."
289 	 * But we'll use the latter here since other kernel sources use it.
290 	 * (I also agree "cylinder" based allocation is obsolete though)
291 	 */
292 
293 	/* maybe "simple is the best" */
294 	blocks_per_cg = sblock.e2fs_bsize * NBBY;
295 
296 	ncg = howmany(bcount - sblock.e2fs.e2fs_first_dblock, blocks_per_cg);
297 	blocks_gd = howmany(sizeof(struct ext2_gd) * ncg, bsize);
298 
299 	/* check range of inode number */
300 	if (num_inodes < EXT2_FIRSTINO)
301 		num_inodes = EXT2_FIRSTINO;	/* needs reserved inodes + 1 */
302 	if (num_inodes > UINT16_MAX * ncg)
303 		num_inodes = UINT16_MAX * ncg;	/* ext2bgd_nifree is uint16_t */
304 
305 	inodes_per_cg = num_inodes / ncg;
306 	iblocks_per_cg = howmany(EXT2_DINODE_SIZE * inodes_per_cg, bsize);
307 
308 	/* Check that the last cylinder group has enough space for inodes */
309 	minblocks_per_cg =
310 	    NBLOCK_BLOCK_BITMAP +
311 	    NBLOCK_INODE_BITMAP +
312 	    iblocks_per_cg +
313 	    1;	/* at least one data block */
314 	if (Oflag == 0 || cg_has_sb(ncg - 1) != 0)
315 		minblocks_per_cg += NBLOCK_SUPERBLOCK + blocks_gd;
316 
317 	blocks_lastcg = bcount - sblock.e2fs.e2fs_first_dblock -
318 	    blocks_per_cg * (ncg - 1);
319 	if (blocks_lastcg < minblocks_per_cg) {
320 		/*
321 		 * Since we make all the cylinder groups the same size, the
322 		 * last will only be small if there are more than one
323 		 * cylinder groups. If the last one is too small to store
324 		 * filesystem data, just kill it.
325 		 *
326 		 * XXX: Does fsck_ext2fs(8) properly handle this case?
327 		 */
328 		bcount -= blocks_lastcg;
329 		ncg--;
330 		blocks_lastcg = blocks_per_cg;
331 		blocks_gd = howmany(sizeof(struct ext2_gd) * ncg, bsize);
332 		inodes_per_cg = num_inodes / ncg;
333 	}
334 	/* roundup inodes_per_cg to make it use whole inode table blocks */
335 	inodes_per_cg = roundup(inodes_per_cg, sblock.e2fs_ipb);
336 	num_inodes = inodes_per_cg * ncg;
337 	iblocks_per_cg = inodes_per_cg / sblock.e2fs_ipb;
338 
339 	/* XXX: probably we should check these adjusted values again */
340 
341 	sblock.e2fs.e2fs_bcount = bcount;
342 	sblock.e2fs.e2fs_icount = num_inodes;
343 
344 	sblock.e2fs_ncg = ncg;
345 	sblock.e2fs_ngdb = blocks_gd;
346 	sblock.e2fs_itpg = iblocks_per_cg;
347 
348 	sblock.e2fs.e2fs_rbcount = sblock.e2fs.e2fs_bcount * minfree / 100;
349 	/* e2fs_fbcount will be accounted later */
350 	/* e2fs_ficount will be accounted later */
351 
352 	sblock.e2fs.e2fs_bpg = blocks_per_cg;
353 	sblock.e2fs.e2fs_fpg = blocks_per_cg;
354 
355 	sblock.e2fs.e2fs_ipg = inodes_per_cg;
356 
357 	sblock.e2fs.e2fs_mtime = 0;
358 	sblock.e2fs.e2fs_wtime = tv.tv_sec;
359 	sblock.e2fs.e2fs_mnt_count = 0;
360 	/* XXX: should add some entropy to avoid checking all fs at once? */
361 	sblock.e2fs.e2fs_max_mnt_count = EXT2_DEF_MAX_MNT_COUNT;
362 
363 	sblock.e2fs.e2fs_magic = E2FS_MAGIC;
364 	sblock.e2fs.e2fs_state = E2FS_ISCLEAN;
365 	sblock.e2fs.e2fs_beh = E2FS_BEH_DEFAULT;
366 	sblock.e2fs.e2fs_minrev = 0;
367 	sblock.e2fs.e2fs_lastfsck = tv.tv_sec;
368 	sblock.e2fs.e2fs_fsckintv = EXT2_DEF_FSCKINTV;
369 
370 	/*
371 	 * Maybe we can use E2FS_OS_FREEBSD here and it would be more proper,
372 	 * but the purpose of this newfs_ext2fs(8) command is to provide
373 	 * a filesystem which can be recognized by firmwares on some
374 	 * Linux based appliances that can load bootstrap files only from
375 	 * (their native) ext2fs, and anyway we will (and should) try to
376 	 * act like them as much as possible.
377 	 *
378 	 * Anyway, I hope that all newer such boxes will keep their support
379 	 * for the "GOOD_OLD_REV" ext2fs.
380 	 */
381 	sblock.e2fs.e2fs_creator = E2FS_OS_LINUX;
382 
383 	if (Oflag == 0) {
384 		sblock.e2fs.e2fs_rev = E2FS_REV0;
385 		sblock.e2fs.e2fs_features_compat   = 0;
386 		sblock.e2fs.e2fs_features_incompat = 0;
387 		sblock.e2fs.e2fs_features_rocompat = 0;
388 	} else {
389 		sblock.e2fs.e2fs_rev = E2FS_REV1;
390 		/*
391 		 * e2fsprogs say "REV1" is "dynamic" so
392 		 * it isn't quite a version and maybe it means
393 		 * "extended from REV0 so check compat features."
394 		 *
395 		 * XXX: We don't have any native tool to activate
396 		 *      the EXT2F_COMPAT_RESIZE feature and
397 		 *      fsck_ext2fs(8) might not fix structures for it.
398 		 */
399 		sblock.e2fs.e2fs_features_compat   = EXT2F_COMPAT_RESIZE;
400 		sblock.e2fs.e2fs_features_incompat = EXT2F_INCOMPAT_FTYPE;
401 		sblock.e2fs.e2fs_features_rocompat =
402 		    EXT2F_ROCOMPAT_SPARSESUPER | EXT2F_ROCOMPAT_LARGEFILE;
403 	}
404 
405 	sblock.e2fs.e2fs_ruid = geteuid();
406 	sblock.e2fs.e2fs_rgid = getegid();
407 
408 	sblock.e2fs.e2fs_first_ino = EXT2_FIRSTINO;
409 	sblock.e2fs.e2fs_inode_size = EXT2_DINODE_SIZE;
410 
411 	/* e2fs_block_group_nr is set on writing superblock to each group */
412 
413 	uuid_create(&uuid, &uustat);
414 	if (uustat != uuid_s_ok)
415 		errx(EXIT_FAILURE, "Failed to generate uuid\n");
416 	uuid_enc_le(sblock.e2fs.e2fs_uuid, &uuid);
417 	if (volname != NULL) {
418 		if (strlen(volname) > sizeof(sblock.e2fs.e2fs_vname))
419 			errx(EXIT_FAILURE, "Volume name is too long");
420 		strlcpy(sblock.e2fs.e2fs_vname, volname,
421 		    sizeof(sblock.e2fs.e2fs_vname));
422 	}
423 
424 	sblock.e2fs.e2fs_fsmnt[0] = '\0';
425 	sblock.e2fs_fsmnt[0] = '\0';
426 
427 	sblock.e2fs.e2fs_algo = 0;		/* XXX unsupported? */
428 	sblock.e2fs.e2fs_prealloc = 0;		/* XXX unsupported? */
429 	sblock.e2fs.e2fs_dir_prealloc = 0;	/* XXX unsupported? */
430 
431 	/* calculate blocks for reserved group descriptors for resize */
432 	sblock.e2fs.e2fs_reserved_ngdb = 0;
433 	if (sblock.e2fs.e2fs_rev > E2FS_REV0 &&
434 	    (sblock.e2fs.e2fs_features_compat & EXT2F_COMPAT_RESIZE) != 0) {
435 		uint64_t target_blocks;
436 		uint target_ncg, target_ngdb, reserved_ngdb;
437 
438 		/* reserve descriptors for size as 1024 times as current */
439 		target_blocks =
440 		    (sblock.e2fs.e2fs_bcount - sblock.e2fs.e2fs_first_dblock)
441 		    * 1024ULL;
442 		/* number of blocks must be in uint32_t */
443 		if (target_blocks > UINT32_MAX)
444 			target_blocks = UINT32_MAX;
445 		target_ncg = howmany(target_blocks, sblock.e2fs.e2fs_bpg);
446 		target_ngdb = howmany(sizeof(struct ext2_gd) * target_ncg,
447 		    sblock.e2fs_bsize);
448 		/*
449 		 * Reserved group descriptor blocks are preserved as
450 		 * the second level double indirect reference blocks in
451 		 * the EXT2_RESIZEINO inode, so the maximum number of
452 		 * the blocks is NINDIR(fs).
453 		 * (see also descriptions in init_resizeino() function)
454 		 *
455 		 * We check a number including current e2fs_ngdb here
456 		 * because they will be moved into reserved gdb on
457 		 * possible future size shrink, though e2fsprogs don't
458 		 * seem to care about it.
459 		 */
460 		if (target_ngdb > NINDIR(&sblock))
461 			target_ngdb = NINDIR(&sblock);
462 
463 		reserved_ngdb = target_ngdb - sblock.e2fs_ngdb;
464 
465 		/* make sure reserved_ngdb fits in the last cg */
466 		if (reserved_ngdb >= blocks_lastcg - cgoverhead(ncg - 1))
467 			reserved_ngdb = blocks_lastcg - cgoverhead(ncg - 1);
468 		if (reserved_ngdb == 0) {
469 			/* if no space for reserved gdb, disable the feature */
470 			sblock.e2fs.e2fs_features_compat &=
471 			    ~EXT2F_COMPAT_RESIZE;
472 		}
473 		sblock.e2fs.e2fs_reserved_ngdb = reserved_ngdb;
474 	}
475 
476 	/*
477 	 * Initialize group descriptors
478 	 */
479 	gd = malloc(sblock.e2fs_ngdb * bsize);
480 	if (gd == NULL)
481 		errx(EXIT_FAILURE, "Can't allocate descriptors buffer");
482 	memset(gd, 0, sblock.e2fs_ngdb * bsize);
483 
484 	fbcount = 0;
485 	ficount = 0;
486 	for (cylno = 0; cylno < ncg; cylno++) {
487 		uint boffset;
488 
489 		boffset = cgbase(&sblock, cylno);
490 		if (sblock.e2fs.e2fs_rev == E2FS_REV0 ||
491 		    (sblock.e2fs.e2fs_features_rocompat &
492 		     EXT2F_ROCOMPAT_SPARSESUPER) == 0 ||
493 		    cg_has_sb(cylno)) {
494 			boffset += NBLOCK_SUPERBLOCK + sblock.e2fs_ngdb;
495 			if (sblock.e2fs.e2fs_rev > E2FS_REV0 &&
496 			    (sblock.e2fs.e2fs_features_compat &
497 			     EXT2F_COMPAT_RESIZE) != 0)
498 				boffset += sblock.e2fs.e2fs_reserved_ngdb;
499 		}
500 		gd[cylno].ext2bgd_b_bitmap = boffset;
501 		boffset += NBLOCK_BLOCK_BITMAP;
502 		gd[cylno].ext2bgd_i_bitmap = boffset;
503 		boffset += NBLOCK_INODE_BITMAP;
504 		gd[cylno].ext2bgd_i_tables = boffset;
505 		if (cylno == (ncg - 1))
506 			gd[cylno].ext2bgd_nbfree =
507 			    blocks_lastcg - cgoverhead(cylno);
508 		else
509 			gd[cylno].ext2bgd_nbfree =
510 			    sblock.e2fs.e2fs_bpg - cgoverhead(cylno);
511 		fbcount += gd[cylno].ext2bgd_nbfree;
512 		gd[cylno].ext2bgd_nifree = sblock.e2fs.e2fs_ipg;
513 		if (cylno == 0) {
514 			/* take reserved inodes off nifree */
515 			gd[cylno].ext2bgd_nifree -= EXT2_RESERVED_INODES;
516 		}
517 		ficount += gd[cylno].ext2bgd_nifree;
518 		gd[cylno].ext2bgd_ndirs = 0;
519 	}
520 	sblock.e2fs.e2fs_fbcount = fbcount;
521 	sblock.e2fs.e2fs_ficount = ficount;
522 
523 	/*
524 	 * Dump out summary information about file system.
525 	 */
526 	if (verbosity > 0) {
527 		printf("%s: %u.%1uMB (%" PRId64 " sectors) "
528 		    "block size %u, fragment size %u\n",
529 		    fsys,
530 		    (uint)(((uint64_t)bcount * bsize) / (1024 * 1024)),
531 		    (uint)((uint64_t)bcount * bsize -
532 		    rounddown((uint64_t)bcount * bsize, 1024 * 1024))
533 		    / 1024 / 100,
534 		    fssize, bsize, fsize);
535 		printf("\tusing %u block groups of %u.0MB, %u blks, "
536 		    "%u inodes.\n",
537 		    ncg, bsize * sblock.e2fs.e2fs_bpg / (1024 * 1024),
538 		    sblock.e2fs.e2fs_bpg, sblock.e2fs.e2fs_ipg);
539 	}
540 
541 	/*
542 	 * allocate space for superblock and group descriptors
543 	 */
544 	iobufsize = (NBLOCK_SUPERBLOCK + sblock.e2fs_ngdb) * sblock.e2fs_bsize;
545 	iobuf = mmap(0, iobufsize, PROT_READ|PROT_WRITE,
546 	    MAP_ANON|MAP_PRIVATE, -1, 0);
547 	if (iobuf == NULL)
548 		errx(EXIT_FAILURE, "Cannot allocate I/O buffer\n");
549 	memset(iobuf, 0, iobufsize);
550 
551 	/*
552 	 * We now start writing to the filesystem
553 	 */
554 
555 	if (!Nflag) {
556 		static const uint pbsize[] = { 1024, 2048, 4096, 0 };
557 		uint pblock, epblock;
558 		/*
559 		 * Validate the given file system size.
560 		 * Verify that its last block can actually be accessed.
561 		 * Convert to file system fragment sized units.
562 		 */
563 		if (fssize <= 0)
564 			errx(EXIT_FAILURE, "Preposterous size %" PRId64 "\n",
565 			    fssize);
566 		wtfs(fssize - 1, sectorsize, iobuf);
567 
568 		/*
569 		 * Ensure there is nothing that looks like a filesystem
570 		 * superbock anywhere other than where ours will be.
571 		 * If fsck_ext2fs finds the wrong one all hell breaks loose!
572 		 *
573 		 * XXX: needs to check how fsck_ext2fs programs even
574 		 *      on other OSes determine alternate superblocks
575 		 */
576 		for (i = 0; pbsize[i] != 0; i++) {
577 			epblock = (uint64_t)bcount * bsize / pbsize[i];
578 			for (pblock = ((pbsize[i] == SBSIZE) ? 1 : 0);
579 			    pblock < epblock;
580 			    pblock += pbsize[i] * NBBY /* bpg */)
581 				zap_old_sblock((daddr_t)pblock *
582 				    pbsize[i] / sectorsize);
583 		}
584 	}
585 
586 	if (verbosity >= 3)
587 		printf("super-block backups (for fsck_ext2fs -b #) at:\n");
588 	/* If we are printing more than one line of numbers, line up columns */
589 	fld_width = verbosity < 4 ? 1 : snprintf(NULL, 0, "%" PRIu64,
590 	    (uint64_t)cgbase(&sblock, ncg - 1));
591 	/* Get terminal width */
592 	if (ioctl(fileno(stdout), TIOCGWINSZ, &winsize) == 0)
593 		max_cols = winsize.ws_col;
594 	else
595 		max_cols = 80;
596 	if (Nflag && verbosity == 3)
597 		/* Leave space to add " ..." after one row of numbers */
598 		max_cols -= 4;
599 #define BASE 0x10000	/* For some fixed-point maths */
600 	col = 0;
601 	delta = verbosity > 2 ? 0 : max_cols * BASE / ncg;
602 	for (cylno = 0; cylno < ncg; cylno++) {
603 		fflush(stdout);
604 		initcg(cylno);
605 		if (verbosity < 2)
606 			continue;
607 		/* the first one is a master, not backup */
608 		if (cylno == 0)
609 			continue;
610 		/* skip if this cylinder doesn't have a backup */
611 		if (sblock.e2fs.e2fs_rev > E2FS_REV0 &&
612 		    (sblock.e2fs.e2fs_features_rocompat &
613 		     EXT2F_ROCOMPAT_SPARSESUPER) != 0 &&
614 		    cg_has_sb(cylno) == 0)
615 			continue;
616 
617 		if (delta > 0) {
618 			if (Nflag)
619 				/* No point doing dots for -N */
620 				break;
621 			/* Print dots scaled to end near RH margin */
622 			for (col += delta; col > BASE; col -= BASE)
623 				printf(".");
624 			continue;
625 		}
626 		/* Print superblock numbers */
627 		len = printf(" %*" PRIu64 "," + !col, fld_width,
628 		    (uint64_t)cgbase(&sblock, cylno));
629 		col += len;
630 		if (col + len < max_cols)
631 			/* Next number fits */
632 			continue;
633 		/* Next number won't fit, need a newline */
634 		if (verbosity <= 3) {
635 			/* Print dots for subsequent cylinder groups */
636 			delta = sblock.e2fs_ncg - cylno - 1;
637 			if (delta != 0) {
638 				if (Nflag) {
639 					printf(" ...");
640 					break;
641 				}
642 				delta = max_cols * BASE / delta;
643 			}
644 		}
645 		col = 0;
646 		printf("\n");
647 	}
648 #undef BASE
649 	if (col > 0)
650 		printf("\n");
651 	if (Nflag)
652 		return;
653 
654 	/*
655 	 * Now construct the initial file system,
656 	 */
657 	if (fsinit(&tv) == 0)
658 		errx(EXIT_FAILURE, "Error making filesystem");
659 	/*
660 	 * Write out the superblock and group descriptors
661 	 */
662 	sblock.e2fs.e2fs_block_group_nr = 0;
663 	sboff = 0;
664 	if (cgbase(&sblock, 0) == 0) {
665 		/*
666 		 * If the first block contains the boot block sectors,
667 		 * (i.e. in case of sblock.e2fs.e2fs_bsize > BBSIZE)
668 		 * we have to preserve data in it.
669 		 */
670 		sboff = SBOFF;
671 	}
672 	e2fs_sbsave(&sblock.e2fs, (struct ext2fs *)(iobuf + sboff));
673 	e2fs_cgsave(gd, (struct ext2_gd *)(iobuf + sblock.e2fs_bsize),
674 	   sizeof(struct ext2_gd) * sblock.e2fs_ncg);
675 	wtfs(fsbtodb(&sblock, cgbase(&sblock, 0)) + sboff / sectorsize,
676 	    iobufsize - sboff, iobuf + sboff);
677 
678 	munmap(iobuf, iobufsize);
679 }
680 
681 /*
682  * Initialize a cylinder (block) group.
683  */
684 void
685 initcg(uint cylno)
686 {
687 	uint nblcg, i, j, sboff;
688 	struct ext2fs_dinode *dp;
689 
690 	/*
691 	 * Make a copy of the superblock and group descriptors.
692 	 */
693 	if (sblock.e2fs.e2fs_rev == E2FS_REV0 ||
694 	    (sblock.e2fs.e2fs_features_rocompat &
695 	     EXT2F_ROCOMPAT_SPARSESUPER) == 0 ||
696 	    cg_has_sb(cylno)) {
697 		sblock.e2fs.e2fs_block_group_nr = cylno;
698 		sboff = 0;
699 		if (cgbase(&sblock, cylno) == 0) {
700 			/* preserve data in bootblock in cg0 */
701 			sboff = SBOFF;
702 		}
703 		e2fs_sbsave(&sblock.e2fs, (struct ext2fs *)(iobuf + sboff));
704 		e2fs_cgsave(gd, (struct ext2_gd *)(iobuf +
705 		    sblock.e2fs_bsize * NBLOCK_SUPERBLOCK),
706 		    sizeof(struct ext2_gd) * sblock.e2fs_ncg);
707 		/* write superblock and group descriptor backups */
708 		wtfs(fsbtodb(&sblock, cgbase(&sblock, cylno)) +
709 		    sboff / sectorsize, iobufsize - sboff, iobuf + sboff);
710 	}
711 
712 	/*
713 	 * Initialize block bitmap.
714 	 */
715 	memset(buf, 0, sblock.e2fs_bsize);
716 	if (cylno == (sblock.e2fs_ncg - 1)) {
717 		/* The last group could have less blocks than e2fs_bpg. */
718 		nblcg = sblock.e2fs.e2fs_bcount -
719 		    cgbase(&sblock, sblock.e2fs_ncg - 1);
720 		for (i = nblcg; i < roundup(nblcg, NBBY); i++)
721 			setbit(buf, i);
722 		memset(&buf[i / NBBY], ~0U, sblock.e2fs.e2fs_bpg - i);
723 	}
724 	/* set overhead (superblock, group descriptor etc.) blocks used */
725 	for (i = 0; i < cgoverhead(cylno) / NBBY; i++)
726 		buf[i] = ~0;
727 	i = i * NBBY;
728 	for (; i < cgoverhead(cylno); i++)
729 		setbit(buf, i);
730 	wtfs(fsbtodb(&sblock, gd[cylno].ext2bgd_b_bitmap), sblock.e2fs_bsize,
731 	    buf);
732 
733 	/*
734 	 * Initialize inode bitmap.
735 	 *
736 	 *  Assume e2fs_ipg is a multiple of NBBY since
737 	 *  it's a multible of e2fs_ipb (as we did above).
738 	 *  Note even (possibly smaller) the last group has the same e2fs_ipg.
739 	 */
740 	i = sblock.e2fs.e2fs_ipg / NBBY;
741 	memset(buf, 0, i);
742 	memset(buf + i, ~0U, sblock.e2fs_bsize - i);
743 	if (cylno == 0) {
744 		/* mark reserved inodes */
745 		for (i = 1; i < EXT2_FIRSTINO; i++)
746 			setbit(buf, EXT2_INO_INDEX(i));
747 	}
748 	wtfs(fsbtodb(&sblock, gd[cylno].ext2bgd_i_bitmap), sblock.e2fs_bsize,
749 	    buf);
750 
751 	/*
752 	 * Initialize inode tables.
753 	 *
754 	 *  Just initialize generation numbers for NFS security.
755 	 *  XXX: sys/ufs/ext2fs/ext2fs_alloc.c:ext2fs_valloc() seems
756 	 *       to override these generated numbers.
757 	 */
758 	memset(buf, 0, sblock.e2fs_bsize);
759 	dp = (struct ext2fs_dinode *)buf;
760 	for (i = 0; i < sblock.e2fs_itpg; i++) {
761 		for (j = 0; j < sblock.e2fs_ipb; j++) {
762 			/* h2fs32() just for consistency */
763 			dp[j].e2di_gen = h2fs32(arc4random());
764 		}
765 		wtfs(fsbtodb(&sblock, gd[cylno].ext2bgd_i_tables + i),
766 		    sblock.e2fs_bsize, buf);
767 	}
768 }
769 
770 /*
771  * Zap possible lingering old superblock data
772  */
773 static void
774 zap_old_sblock(daddr_t sec)
775 {
776 	static daddr_t cg0_data;
777 	uint32_t oldfs[SBSIZE / sizeof(uint32_t)];
778 	static const struct fsm {
779 		uint32_t offset;
780 		uint32_t magic;
781 		uint32_t mask;
782 	} fs_magics[] = {
783 		{offsetof(struct ext2fs, e2fs_magic) / 4, E2FS_MAGIC, 0xffff},
784 		{offsetof(struct ext2fs, e2fs_magic) / 4,
785 		    E2FS_MAGIC << 16, 0xffff0000},
786 		{14, 0xef530000, 0xffff0000},	/* EXT2FS (big) */
787 		{0x55c / 4, 0x00011954, ~0U},	/* FS_UFS1_MAGIC */
788 		{0x55c / 4, 0x19540119, ~0U},	/* FS_UFS2_MAGIC */
789 		{0, 0x70162, ~0U},		/* LFS_MAGIC */
790 		{.offset = ~0U},
791 	};
792 	const struct fsm *fsm;
793 
794 	if (Nflag)
795 		return;
796 
797 	/* don't override data before superblock */
798 	if (sec < SBOFF / sectorsize)
799 		return;
800 
801 	if (cg0_data == 0) {
802 		cg0_data =
803 		    ((daddr_t)sblock.e2fs.e2fs_first_dblock + cgoverhead(0)) *
804 		    sblock.e2fs_bsize / sectorsize;
805 	}
806 
807 	/* Ignore anything that is beyond our filesystem */
808 	if (sec >= fssize)
809 		return;
810 	/* Zero anything inside our filesystem... */
811 	if (sec >= sblock.e2fs.e2fs_first_dblock * bsize / sectorsize) {
812 		/* ...unless we will write that area anyway */
813 		if (sec >= cg0_data)
814 			/* assume iobuf is zero'ed here */
815 			wtfs(sec, roundup(SBSIZE, sectorsize), iobuf);
816 		return;
817 	}
818 
819 	/*
820 	 * The sector might contain boot code, so we must validate it
821 	 *
822 	 * XXX: ext2fs won't preserve data after SBOFF,
823 	 *      but first_dblock could have a differnt value.
824 	 */
825 	rdfs(sec, sizeof(oldfs), &oldfs);
826 	for (fsm = fs_magics;; fsm++) {
827 		uint32_t v;
828 		if (fsm->mask == 0)
829 			return;
830 		v = oldfs[fsm->offset];
831 		if ((v & fsm->mask) == fsm->magic ||
832 		    (bswap32(v) & fsm->mask) == fsm->magic)
833 			break;
834 	}
835 
836 	/* Just zap the magic number */
837 	oldfs[fsm->offset] = 0;
838 	wtfs(sec, sizeof(oldfs), &oldfs);
839 }
840 
841 /*
842  * uint cgoverhead(uint c)
843  *
844  * 	Return a number of reserved blocks on the specified group.
845  * 	XXX: should be shared with src/sbin/fsck_ext2fs/setup.c
846  */
847 uint
848 cgoverhead(uint c)
849 {
850 	uint overh;
851 
852 	overh = NBLOCK_BLOCK_BITMAP + NBLOCK_INODE_BITMAP + sblock.e2fs_itpg;
853 
854 	if (sblock.e2fs.e2fs_rev == E2FS_REV0 ||
855 	    (sblock.e2fs.e2fs_features_rocompat &
856 	     EXT2F_ROCOMPAT_SPARSESUPER) == 0 ||
857 	    cg_has_sb(c) != 0) {
858 		overh += NBLOCK_SUPERBLOCK + sblock.e2fs_ngdb;
859 
860 		if (sblock.e2fs.e2fs_rev > E2FS_REV0 &&
861 		    (sblock.e2fs.e2fs_features_compat &
862 		     EXT2F_COMPAT_RESIZE) != 0)
863 			overh += sblock.e2fs.e2fs_reserved_ngdb;
864 	}
865 
866 	return overh;
867 }
868 
869 /*
870  * Initialize the file system
871  */
872 
873 #define LOSTDIR		/* e2fsck complains if there is no lost+found */
874 
875 #define	PREDEFDIR	2
876 
877 #ifdef LOSTDIR
878 #define	PREDEFROOTDIR	(PREDEFDIR + 1)
879 #else
880 #define	PREDEFROOTDIR	PREDEFDIR
881 #endif
882 
883 struct ext2fs_direct root_dir[] = {
884 	{ EXT2_ROOTINO, 0, 1, 0, "." },
885 	{ EXT2_ROOTINO, 0, 2, 0, ".." },
886 #ifdef LOSTDIR
887 	{ EXT2_LOSTFOUNDINO, 0, 10, 0, "lost+found" },
888 #endif
889 };
890 
891 #ifdef LOSTDIR
892 struct ext2fs_direct lost_found_dir[] = {
893 	{ EXT2_LOSTFOUNDINO, 0, 1, 0, "." },
894 	{ EXT2_ROOTINO, 0, 2, 0, ".." },
895 };
896 struct ext2fs_direct pad_dir = { 0, sizeof(struct ext2fs_direct), 0, 0, "" };
897 #endif
898 
899 int
900 fsinit(const struct timeval *tv)
901 {
902 	struct ext2fs_dinode node;
903 #ifdef LOSTDIR
904 	uint i, nblks_lostfound, blk;
905 #endif
906 
907 	/*
908 	 * Initialize the inode for the resizefs feature
909 	 */
910 	if (sblock.e2fs.e2fs_rev > E2FS_REV0 &&
911 	    (sblock.e2fs.e2fs_features_compat & EXT2F_COMPAT_RESIZE) != 0)
912 		init_resizeino(tv);
913 
914 	/*
915 	 * Initialize the node
916 	 */
917 
918 #ifdef LOSTDIR
919 	/*
920 	 * Create the lost+found directory
921 	 */
922 	if (sblock.e2fs.e2fs_rev > E2FS_REV0 &&
923 	    sblock.e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE) {
924 		lost_found_dir[0].e2d_type = EXT2_FT_DIR;
925 		lost_found_dir[1].e2d_type = EXT2_FT_DIR;
926 	}
927 	(void)makedir(lost_found_dir, __arraycount(lost_found_dir));
928 
929 	/* prepare a bit large directory for preserved files */
930 	nblks_lostfound = EXT2_LOSTFOUNDSIZE / sblock.e2fs_bsize;
931 	/* ...but only with direct blocks */
932 	if (nblks_lostfound > NDADDR)
933 		nblks_lostfound = NDADDR;
934 
935 	memset(&node, 0, sizeof(node));
936 	node.e2di_mode = EXT2_IFDIR | EXT2_LOSTFOUNDUMASK;
937 	node.e2di_uid = geteuid();
938 	node.e2di_size = sblock.e2fs_bsize * nblks_lostfound;
939 	node.e2di_atime = tv->tv_sec;
940 	node.e2di_ctime = tv->tv_sec;
941 	node.e2di_mtime = tv->tv_sec;
942 	node.e2di_gid = getegid();
943 	node.e2di_nlink = PREDEFDIR;
944 	/* e2di_nblock is a number of disk block, not ext2fs block */
945 	node.e2di_nblock = fsbtodb(&sblock, nblks_lostfound);
946 	node.e2di_blocks[0] = alloc(sblock.e2fs_bsize, node.e2di_mode);
947 	if (node.e2di_blocks[0] == 0) {
948 		printf("%s: can't allocate block for lost+found\n", __func__);
949 		return 0;
950 	}
951 	for (i = 1; i < nblks_lostfound; i++) {
952 		blk = alloc(sblock.e2fs_bsize, 0);
953 		if (blk == 0) {
954 			printf("%s: can't allocate blocks for lost+found\n",
955 			    __func__);
956 			return 0;
957 		}
958 		node.e2di_blocks[i] = blk;
959 	}
960 	wtfs(fsbtodb(&sblock, node.e2di_blocks[0]), sblock.e2fs_bsize, buf);
961 	pad_dir.e2d_reclen = sblock.e2fs_bsize;
962 	for (i = 1; i < nblks_lostfound; i++) {
963 		memset(buf, 0, sblock.e2fs_bsize);
964 		copy_dir(&pad_dir, (struct ext2fs_direct *)buf);
965 		wtfs(fsbtodb(&sblock, node.e2di_blocks[i]), sblock.e2fs_bsize,
966 		    buf);
967 	}
968 	iput(&node, EXT2_LOSTFOUNDINO);
969 #endif
970 	/*
971 	 * create the root directory
972 	 */
973 	memset(&node, 0, sizeof(node));
974 	if (sblock.e2fs.e2fs_rev > E2FS_REV0 &&
975 	    sblock.e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE) {
976 		root_dir[0].e2d_type = EXT2_FT_DIR;
977 		root_dir[1].e2d_type = EXT2_FT_DIR;
978 #ifdef LOSTDIR
979 		root_dir[2].e2d_type = EXT2_FT_DIR;
980 #endif
981 	}
982 	node.e2di_mode = EXT2_IFDIR | EXT2_UMASK;
983 	node.e2di_uid = geteuid();
984 	node.e2di_size = makedir(root_dir, __arraycount(root_dir));
985 	node.e2di_atime = tv->tv_sec;
986 	node.e2di_ctime = tv->tv_sec;
987 	node.e2di_mtime = tv->tv_sec;
988 	node.e2di_gid = getegid();
989 	node.e2di_nlink = PREDEFROOTDIR;
990 	/* e2di_nblock is a number of disk block, not ext2fs block */
991 	node.e2di_nblock = fsbtodb(&sblock, 1);
992 	node.e2di_blocks[0] = alloc(node.e2di_size, node.e2di_mode);
993 	if (node.e2di_blocks[0] == 0) {
994 		printf("%s: can't allocate block for root dir\n", __func__);
995 		return 0;
996 	}
997 	wtfs(fsbtodb(&sblock, node.e2di_blocks[0]), sblock.e2fs_bsize, buf);
998 	iput(&node, EXT2_ROOTINO);
999 	return 1;
1000 }
1001 
1002 /*
1003  * Construct a set of directory entries in "buf".
1004  * return size of directory.
1005  */
1006 int
1007 makedir(struct ext2fs_direct *protodir, int entries)
1008 {
1009 	uint8_t *cp;
1010 	uint i, spcleft;
1011 	uint dirblksiz;
1012 
1013 	dirblksiz = sblock.e2fs_bsize;
1014 	memset(buf, 0, dirblksiz);
1015 	spcleft = dirblksiz;
1016 	for (cp = buf, i = 0; i < entries - 1; i++) {
1017 		protodir[i].e2d_reclen = EXT2FS_DIRSIZ(protodir[i].e2d_namlen);
1018 		copy_dir(&protodir[i], (struct ext2fs_direct *)cp);
1019 		cp += protodir[i].e2d_reclen;
1020 		spcleft -= protodir[i].e2d_reclen;
1021 	}
1022 	protodir[i].e2d_reclen = spcleft;
1023 	copy_dir(&protodir[i], (struct ext2fs_direct *)cp);
1024 	return dirblksiz;
1025 }
1026 
1027 /*
1028  * Copy a direntry to a buffer, in fs byte order
1029  */
1030 static void
1031 copy_dir(struct ext2fs_direct *dir, struct ext2fs_direct *dbuf)
1032 {
1033 
1034 	memcpy(dbuf, dir, EXT2FS_DIRSIZ(dir->e2d_namlen));
1035 	dbuf->e2d_ino = h2fs32(dir->e2d_ino);
1036 	dbuf->e2d_reclen = h2fs16(dir->e2d_reclen);
1037 }
1038 
1039 /*
1040  * void init_resizeino(const struct timeval *tv);
1041  *
1042  *	Initialize the EXT2_RESEIZE_INO inode to prereserve
1043  *	reserved group descriptor blocks for future growth of this ext2fs.
1044  */
1045 void
1046 init_resizeino(const struct timeval *tv)
1047 {
1048 	struct ext2fs_dinode node;
1049 	uint64_t isize;
1050 	uint32_t *dindir_block, *reserved_gdb;
1051 	uint nblock, i, cylno, n;
1052 
1053 	memset(&node, 0, sizeof(node));
1054 
1055 	/*
1056 	 * Note this function only prepares required structures for
1057 	 * future resize. It's a quite different work to implement
1058 	 * a utility like resize_ext2fs(8) which handles actual
1059 	 * resize ops even on offline.
1060 	 *
1061 	 * Anyway, I'm not sure if there is any documentation about
1062 	 * this resize ext2fs feature and related data structures,
1063 	 * and I've written this function based on things what I see
1064 	 * on some existing implementation and real file system data
1065 	 * created by existing tools. To be honest, they are not
1066 	 * so easy to read, so I will try to implement it here without
1067 	 * any dumb optimization for people who would eventually
1068 	 * work on "yet another wheel" like resize_ext2fs(8).
1069 	 */
1070 
1071 	/*
1072 	 * I'm not sure what type is appropriate for this inode.
1073 	 * The release notes of e2fsprogs says they changed e2fsck to allow
1074 	 * IFREG for RESIZEINO since a certain resize tool used it. Hmm.
1075 	 */
1076 	node.e2di_mode = EXT2_IFREG | EXT2_RESIZEINOUMASK;
1077 	node.e2di_uid = geteuid();
1078 	node.e2di_atime = tv->tv_sec;
1079 	node.e2di_ctime = tv->tv_sec;
1080 	node.e2di_mtime = tv->tv_sec;
1081 	node.e2di_gid = getegid();
1082 	node.e2di_nlink = 1;
1083 
1084 	/*
1085 	 * To preserve the reserved group descriptor blocks,
1086 	 * EXT2_RESIZEINO uses only double indirect reference
1087 	 * blocks in its inode entries.
1088 	 *
1089 	 * All entries for direct, single indirect and triple
1090 	 * indirect references are left zero'ed. Maybe it's safe
1091 	 * because no write operation will happen with this inode.
1092 	 *
1093 	 * We have to allocate a block for the first level double
1094 	 * indirect reference block. Indexes of inode entries in
1095 	 * this first level dindirect block are corresponding to
1096 	 * indexes of group descriptors including both used (e2fs_ngdb)
1097 	 * and reserved (e2fs_reserved_ngdb) group descriptor blocks.
1098 	 *
1099 	 * Inode entries of indexes for used (e2fs_ngdb) descriptors are
1100 	 * left zero'ed. Entries for reserved (e2fs_reserved_ngdb) ones
1101 	 * have block numbers of actual reserved group descriptors
1102 	 * allocated at block group zero. This means e2fs_reserved_ngdb
1103 	 * blocks are reserved as the second level dindirect reference
1104 	 * blocks, and they acutually contain block numbers of indirect
1105 	 * references. It may be safe since they don't have to keep any
1106 	 * data yet.
1107 	 *
1108 	 * Each these second dindirect blocks (i.e. reserved group
1109 	 * descriptor blocks in the first block group) should have
1110 	 * block numbers of its backups in all other block groups.
1111 	 * I.e. reserved_ngdb[0] block in block group 0 contains block
1112 	 * numbers of resreved_ngdb[0] from group 1 through (e2fs_ncg - 1).
1113 	 * The number of backups can be determined by the
1114 	 * EXT2_ROCOMPAT_SPARSESUPER feature and cg_has_sb() macro
1115 	 * as done in the above initcg() function.
1116 	 */
1117 
1118 	/* set e2di_size which occupies whole blocks through DINDIR blocks */
1119 	isize = (uint64_t)sblock.e2fs_bsize * NDADDR +
1120 	    (uint64_t)sblock.e2fs_bsize * NINDIR(&sblock) +
1121 	    (uint64_t)sblock.e2fs_bsize * NINDIR(&sblock) * NINDIR(&sblock);
1122 	if (isize > UINT32_MAX &&
1123 	    (sblock.e2fs.e2fs_features_rocompat &
1124 	     EXT2F_ROCOMPAT_LARGEFILE) == 0) {
1125 		/* XXX should enable it here and update all backups? */
1126 		errx(EXIT_FAILURE, "%s: large_file rocompat feature is "
1127 		    "required to enable resize feature for this filesystem\n",
1128 		    __func__);
1129 	}
1130 	/* upper 32bit is stored into e2di_dacl on REV1 feature */
1131 	node.e2di_size = isize & UINT32_MAX;
1132 	node.e2di_dacl = isize >> 32;
1133 
1134 #define SINGLE	0	/* index of single indirect block */
1135 #define DOUBLE	1	/* index of double indirect block */
1136 #define TRIPLE	2	/* index of triple indirect block */
1137 
1138 	/* zero out entries for direct references */
1139 	for (i = 0; i < NDADDR; i++)
1140 		node.e2di_blocks[i] = 0;
1141 	/* also zero out entries for single and triple indirect references */
1142 	node.e2di_blocks[NDADDR + SINGLE] = 0;
1143 	node.e2di_blocks[NDADDR + TRIPLE] = 0;
1144 
1145 	/* allocate a block for the first level double indirect reference */
1146 	node.e2di_blocks[NDADDR + DOUBLE] =
1147 	    alloc(sblock.e2fs_bsize, node.e2di_mode);
1148 	if (node.e2di_blocks[NDADDR + DOUBLE] == 0)
1149 		errx(EXIT_FAILURE, "%s: Can't allocate a dindirect block",
1150 		    __func__);
1151 
1152 	/* account this first block */
1153 	nblock = fsbtodb(&sblock, 1);
1154 
1155 	/* allocate buffer to set data in the dindirect block */
1156 	dindir_block = malloc(sblock.e2fs_bsize);
1157 	if (dindir_block == NULL)
1158 		errx(EXIT_FAILURE,
1159 		    "%s: Can't allocate buffer for a dindirect block",
1160 		    __func__);
1161 
1162 	/* allocate buffer to set data in the group descriptor blocks */
1163 	reserved_gdb = malloc(sblock.e2fs_bsize);
1164 	if (reserved_gdb == NULL)
1165 		errx(EXIT_FAILURE,
1166 		    "%s: Can't allocate buffer for group descriptor blocks",
1167 		    __func__);
1168 
1169 	/*
1170 	 * Setup block entries in the first level dindirect blocks
1171 	 */
1172 	for (i = 0; i < sblock.e2fs_ngdb; i++) {
1173 		/* no need to handle used group descriptor blocks */
1174 		dindir_block[i] = 0;
1175 	}
1176 	for (; i < sblock.e2fs_ngdb + sblock.e2fs.e2fs_reserved_ngdb; i++) {
1177 		/*
1178 		 * point reserved group descriptor block in the first
1179 		 * (i.e. master) block group
1180 		 *
1181 		 * XXX: e2fsprogs seem to use "(i % NINDIR(&sblock))" here
1182 		 *      to store maximum NINDIR(&sblock) reserved gdbs.
1183 		 *      I'm not sure what will be done on future filesystem
1184 		 *      shrink in that case on their way.
1185 		 */
1186 		if (i >= NINDIR(&sblock))
1187 			errx(EXIT_FAILURE, "%s: too many reserved "
1188 			    "group descriptors (%u) for resize inode",
1189 			    __func__, sblock.e2fs.e2fs_reserved_ngdb);
1190 		dindir_block[i] =
1191 		    h2fs32(cgbase(&sblock, 0) + NBLOCK_SUPERBLOCK + i);
1192 
1193 		/*
1194 		 * Setup block entries in the second dindirect blocks
1195 		 * (which are primary reserved group descriptor blocks)
1196 		 * to point their backups.
1197 		 */
1198 		for (n = 0, cylno = 1; cylno < sblock.e2fs_ncg; cylno++) {
1199 			/* skip block groups without backup */
1200 			if ((sblock.e2fs.e2fs_features_rocompat &
1201 			     EXT2F_ROCOMPAT_SPARSESUPER) != 0 &&
1202 			    cg_has_sb(cylno) == 0)
1203 				continue;
1204 
1205 			if (n >= NINDIR(&sblock))
1206 				errx(EXIT_FAILURE, "%s: too many block groups "
1207 				    "for the resize feature", __func__);
1208 			/*
1209 			 * These blocks are already reserved in
1210 			 * initcg() so no need to use alloc() here.
1211 			 */
1212 			reserved_gdb[n++] = h2fs32(cgbase(&sblock, cylno) +
1213 			    NBLOCK_SUPERBLOCK + i);
1214 			nblock += fsbtodb(&sblock, 1);
1215 		}
1216 		for (; n < NINDIR(&sblock); n++)
1217 			reserved_gdb[n] = 0;
1218 
1219 		/* write group descriptor block as the second dindirect refs */
1220 		wtfs(fsbtodb(&sblock, fs2h32(dindir_block[i])),
1221 		    sblock.e2fs_bsize, reserved_gdb);
1222 		nblock += fsbtodb(&sblock, 1);
1223 	}
1224 	for (; i < NINDIR(&sblock); i++) {
1225 		/* leave trailing entries unallocated */
1226 		dindir_block[i] = 0;
1227 	}
1228 	free(reserved_gdb);
1229 
1230 	/* finally write the first level dindirect block */
1231 	wtfs(fsbtodb(&sblock, node.e2di_blocks[NDADDR + DOUBLE]),
1232 	    sblock.e2fs_bsize, dindir_block);
1233 	free(dindir_block);
1234 
1235 	node.e2di_nblock = nblock;
1236 	iput(&node, EXT2_RESIZEINO);
1237 }
1238 
1239 /*
1240  * uint32_t alloc(uint32_t size, uint16_t mode)
1241  *
1242  *	Allocate a block (from cylinder group 0)
1243  *	Reference: src/sys/ufs/ext2fs/ext2fs_alloc.c:ext2fs_alloccg()
1244  */
1245 uint32_t
1246 alloc(uint32_t size, uint16_t mode)
1247 {
1248 	uint32_t loc, bno;
1249 	uint8_t *bbp;
1250 	uint len, map, i;
1251 
1252 	if (gd[0].ext2bgd_nbfree == 0)
1253 		return 0;
1254 
1255 	if (size > sblock.e2fs_bsize)
1256 		return 0;
1257 
1258 	bbp = malloc(sblock.e2fs_bsize);
1259 	if (bbp == NULL)
1260 		return 0;
1261 	rdfs(fsbtodb(&sblock, gd[0].ext2bgd_b_bitmap), sblock.e2fs_bsize, bbp);
1262 
1263 	/* XXX: kernel uses e2fs_fpg here */
1264 	len = sblock.e2fs.e2fs_bpg / NBBY;
1265 
1266 #if 0	/* no need block allocation for root or lost+found dir */
1267 	for (loc = 0; loc < len; loc++) {
1268 		if (bbp[loc] == 0) {
1269 			bno = loc * NBBY;
1270 			goto gotit;
1271 		}
1272 	}
1273 #endif
1274 
1275 	loc = skpc(~0U, len, bbp);
1276 	if (loc == 0)
1277 		return 0;
1278 	loc = len - loc;
1279 	map = bbp[loc];
1280 	bno = loc * NBBY;
1281 	for (i = 0; i < NBBY; i++, bno++) {
1282 		if ((map & (1 << i)) == 0)
1283 			goto gotit;
1284 	}
1285 	return 0;
1286 
1287  gotit:
1288 	if (isset(bbp, bno))
1289 		errx(EXIT_FAILURE, "%s: inconsistent bitmap\n", __func__);
1290 
1291 	setbit(bbp, bno);
1292 	wtfs(fsbtodb(&sblock, gd[0].ext2bgd_b_bitmap), sblock.e2fs_bsize, bbp);
1293 	free(bbp);
1294 	/* XXX: modified group descriptors won't be written into backups */
1295 	gd[0].ext2bgd_nbfree--;
1296 	if ((mode & EXT2_IFDIR) != 0)
1297 		gd[0].ext2bgd_ndirs++;
1298 	sblock.e2fs.e2fs_fbcount--;
1299 
1300 	return sblock.e2fs.e2fs_first_dblock + bno;
1301 }
1302 
1303 /*
1304  * void iput(struct ext2fs_dinode *ip, ino_t ino)
1305  *
1306  *	Put an inode entry into the corresponding table.
1307  */
1308 static void
1309 iput(struct ext2fs_dinode *ip, ino_t ino)
1310 {
1311 	daddr_t d;
1312 	uint c, i;
1313 	struct ext2fs_dinode *dp;
1314 	uint8_t *bp;
1315 
1316 	bp = malloc(sblock.e2fs_bsize);
1317 	if (bp == NULL)
1318 		errx(EXIT_FAILURE, "%s: can't allocate buffer for inode\n",
1319 		    __func__);
1320 
1321 	/*
1322 	 * Reserved inodes are allocated and accounted in initcg()
1323 	 * so skip checks of the bitmap and allocation for them.
1324 	 */
1325 	if (ino >= EXT2_FIRSTINO) {
1326 		c = ino_to_cg(&sblock, ino);
1327 
1328 		/* sanity check */
1329 		if (gd[c].ext2bgd_nifree == 0)
1330 			errx(EXIT_FAILURE,
1331 			    "%s: no free inode %" PRIu64 " in block group %u\n",
1332 			    __func__, (uint64_t)ino, c);
1333 
1334 		/* update inode bitmap */
1335 		rdfs(fsbtodb(&sblock, gd[0].ext2bgd_i_bitmap),
1336 		    sblock.e2fs_bsize, bp);
1337 
1338 		/* more sanity */
1339 		if (isset(bp, EXT2_INO_INDEX(ino)))
1340 			errx(EXIT_FAILURE, "%s: inode %" PRIu64
1341 			    " already in use\n", __func__, (uint64_t)ino);
1342 		setbit(bp, EXT2_INO_INDEX(ino));
1343 		wtfs(fsbtodb(&sblock, gd[0].ext2bgd_i_bitmap),
1344 		    sblock.e2fs_bsize, bp);
1345 		gd[c].ext2bgd_nifree--;
1346 		sblock.e2fs.e2fs_ficount--;
1347 	}
1348 
1349 	if (ino >= sblock.e2fs.e2fs_ipg * sblock.e2fs_ncg)
1350 		errx(EXIT_FAILURE, "%s: inode value out of range (%" PRIu64
1351 		    ").\n", __func__, (uint64_t)ino);
1352 
1353 	/* update an inode entry in the table */
1354 	d = fsbtodb(&sblock, ino_to_fsba(&sblock, ino));
1355 	rdfs(d, sblock.e2fs_bsize, bp);
1356 
1357 	dp = (struct ext2fs_dinode *)bp;
1358 	dp += ino_to_fsbo(&sblock, ino);
1359 	e2fs_isave(ip, dp);
1360 	/* e2fs_i_bswap() doesn't swap e2di_blocks addrs */
1361 	if ((ip->e2di_mode & EXT2_IFMT) != EXT2_IFLNK) {
1362 		for (i = 0; i < NDADDR + NIADDR; i++)
1363 			dp->e2di_blocks[i] = h2fs32(ip->e2di_blocks[i]);
1364 	}
1365 	/* h2fs32() just for consistency */
1366 	dp->e2di_gen = h2fs32(arc4random());
1367 
1368 	wtfs(d, sblock.e2fs_bsize, bp);
1369 	free(bp);
1370 }
1371 
1372 /*
1373  * Read a block from the file system
1374  */
1375 void
1376 rdfs(daddr_t bno, int size, void *bf)
1377 {
1378 	int n;
1379 	off_t offset;
1380 
1381 	offset = bno;
1382 	n = pread(fsi, bf, size, offset * sectorsize);
1383 	if (n != size)
1384 		err(EXIT_FAILURE, "%s: read error for sector %" PRId64,
1385 		    __func__, (int64_t)bno);
1386 }
1387 
1388 /*
1389  * Write a block to the file system
1390  */
1391 void
1392 wtfs(daddr_t bno, int size, void *bf)
1393 {
1394 	int n;
1395 	off_t offset;
1396 
1397 	if (Nflag)
1398 		return;
1399 	offset = bno;
1400 	n = pwrite(fso, bf, size, offset * sectorsize);
1401 	if (n != size)
1402 		err(EXIT_FAILURE, "%s: write error for sector %" PRId64,
1403 		    __func__, (int64_t)bno);
1404 }
1405 
1406 int
1407 ilog2(uint val)
1408 {
1409 
1410 	if (val == 0 || !powerof2(val))
1411 		errx(EXIT_FAILURE, "%s: %u is not a power of 2\n",
1412 		    __func__, val);
1413 
1414 	return ffs(val) - 1;
1415 }
1416 
1417 /*
1418  * int skpc(int mask, size_t size, uint8_t *cp)
1419  *
1420  *	Locate an unsigned character of value mask inside cp[].
1421  * 	(from src/sys/lib/libkern/skpc.c)
1422  */
1423 int
1424 skpc(int mask, size_t size, uint8_t *cp)
1425 {
1426 	uint8_t *end;
1427 
1428 	end = &cp[size];
1429 	while (cp < end && *cp == (uint8_t)mask)
1430 		cp++;
1431 
1432 	return end - cp;
1433 }
1434