xref: /freebsd/sbin/growfs/growfs.c (revision 0957b409)
1 /*-
2  * SPDX-License-Identifier: BSD-4-Clause
3  *
4  * Copyright (c) 1980, 1989, 1993 The Regents of the University of California.
5  * Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz
6  * Copyright (c) 2012 The FreeBSD Foundation
7  * All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * Christoph Herrmann and Thomas-Henning von Kamptz, Munich and Frankfurt.
11  *
12  * Portions of this software were developed by Edward Tomasz Napierala
13  * under sponsorship from the FreeBSD Foundation.
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions
17  * are met:
18  * 1. Redistributions of source code must retain the above copyright
19  *    notice, this list of conditions and the following disclaimer.
20  * 2. Redistributions in binary form must reproduce the above copyright
21  *    notice, this list of conditions and the following disclaimer in the
22  *    documentation and/or other materials provided with the distribution.
23  * 3. All advertising materials mentioning features or use of this software
24  *    must display the following acknowledgment:
25  *      This product includes software developed by the University of
26  *      California, Berkeley and its contributors, as well as Christoph
27  *      Herrmann and Thomas-Henning von Kamptz.
28  * 4. Neither the name of the University nor the names of its contributors
29  *    may be used to endorse or promote products derived from this software
30  *    without specific prior written permission.
31  *
32  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
33  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
34  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
35  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
36  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
37  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
38  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
39  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
40  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
41  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
42  * SUCH DAMAGE.
43  *
44  * $TSHeader: src/sbin/growfs/growfs.c,v 1.5 2000/12/12 19:31:00 tomsoft Exp $
45  *
46  */
47 
48 #ifndef lint
49 static const char copyright[] =
50 "@(#) Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz\n\
51 Copyright (c) 1980, 1989, 1993 The Regents of the University of California.\n\
52 All rights reserved.\n";
53 #endif /* not lint */
54 
55 #include <sys/cdefs.h>
56 __FBSDID("$FreeBSD$");
57 
58 #include <sys/param.h>
59 #include <sys/ioctl.h>
60 #include <sys/stat.h>
61 #include <sys/disk.h>
62 #include <sys/ucred.h>
63 #include <sys/mount.h>
64 
65 #include <stdio.h>
66 #include <paths.h>
67 #include <ctype.h>
68 #include <err.h>
69 #include <errno.h>
70 #include <fcntl.h>
71 #include <fstab.h>
72 #include <inttypes.h>
73 #include <limits.h>
74 #include <mntopts.h>
75 #include <paths.h>
76 #include <stdlib.h>
77 #include <stdint.h>
78 #include <string.h>
79 #include <time.h>
80 #include <unistd.h>
81 #include <ufs/ufs/dinode.h>
82 #include <ufs/ffs/fs.h>
83 #include <libutil.h>
84 #include <libufs.h>
85 
86 #include "debug.h"
87 
88 #ifdef FS_DEBUG
89 int	_dbg_lvl_ = (DL_INFO);	/* DL_TRC */
90 #endif /* FS_DEBUG */
91 
92 static union {
93 	struct fs	fs;
94 	char		pad[SBLOCKSIZE];
95 } fsun1, fsun2;
96 #define	sblock	fsun1.fs	/* the new superblock */
97 #define	osblock	fsun2.fs	/* the old superblock */
98 
99 static union {
100 	struct cg	cg;
101 	char		pad[MAXBSIZE];
102 } cgun1, cgun2;
103 #define	acg	cgun1.cg	/* a cylinder cgroup (new) */
104 #define	aocg	cgun2.cg	/* an old cylinder group */
105 
106 static struct csum	*fscs;	/* cylinder summary */
107 
108 static void	growfs(int, int, unsigned int);
109 static void	rdfs(ufs2_daddr_t, size_t, void *, int);
110 static void	wtfs(ufs2_daddr_t, size_t, void *, int, unsigned int);
111 static int	charsperline(void);
112 static void	usage(void);
113 static int	isblock(struct fs *, unsigned char *, int);
114 static void	clrblock(struct fs *, unsigned char *, int);
115 static void	setblock(struct fs *, unsigned char *, int);
116 static void	initcg(int, time_t, int, unsigned int);
117 static void	updjcg(int, time_t, int, int, unsigned int);
118 static void	updcsloc(time_t, int, int, unsigned int);
119 static void	frag_adjust(ufs2_daddr_t, int);
120 static void	updclst(int);
121 static void	mount_reload(const struct statfs *stfs);
122 static void	cgckhash(struct cg *);
123 
124 /*
125  * Here we actually start growing the file system. We basically read the
126  * cylinder summary from the first cylinder group as we want to update
127  * this on the fly during our various operations. First we handle the
128  * changes in the former last cylinder group. Afterwards we create all new
129  * cylinder groups.  Now we handle the cylinder group containing the
130  * cylinder summary which might result in a relocation of the whole
131  * structure.  In the end we write back the updated cylinder summary, the
132  * new superblock, and slightly patched versions of the super block
133  * copies.
134  */
135 static void
136 growfs(int fsi, int fso, unsigned int Nflag)
137 {
138 	DBG_FUNC("growfs")
139 	time_t modtime;
140 	uint cylno;
141 	int i, j, width;
142 	char tmpbuf[100];
143 
144 	DBG_ENTER;
145 
146 	time(&modtime);
147 
148 	/*
149 	 * Get the cylinder summary into the memory.
150 	 */
151 	fscs = (struct csum *)calloc((size_t)1, (size_t)sblock.fs_cssize);
152 	if (fscs == NULL)
153 		errx(1, "calloc failed");
154 	memcpy(fscs, osblock.fs_csp, osblock.fs_cssize);
155 	free(osblock.fs_csp);
156 	osblock.fs_csp = NULL;
157 	sblock.fs_csp = fscs;
158 
159 #ifdef FS_DEBUG
160 	{
161 		struct csum *dbg_csp;
162 		u_int32_t dbg_csc;
163 		char dbg_line[80];
164 
165 		dbg_csp = fscs;
166 
167 		for (dbg_csc = 0; dbg_csc < osblock.fs_ncg; dbg_csc++) {
168 			snprintf(dbg_line, sizeof(dbg_line),
169 			    "%d. old csum in old location", dbg_csc);
170 			DBG_DUMP_CSUM(&osblock, dbg_line, dbg_csp++);
171 		}
172 	}
173 #endif /* FS_DEBUG */
174 	DBG_PRINT0("fscs read\n");
175 
176 	/*
177 	 * Do all needed changes in the former last cylinder group.
178 	 */
179 	updjcg(osblock.fs_ncg - 1, modtime, fsi, fso, Nflag);
180 
181 	/*
182 	 * Dump out summary information about file system.
183 	 */
184 #ifdef FS_DEBUG
185 #define B2MBFACTOR (1 / (1024.0 * 1024.0))
186 	printf("growfs: %.1fMB (%jd sectors) block size %d, fragment size %d\n",
187 	    (float)sblock.fs_size * sblock.fs_fsize * B2MBFACTOR,
188 	    (intmax_t)fsbtodb(&sblock, sblock.fs_size), sblock.fs_bsize,
189 	    sblock.fs_fsize);
190 	printf("\tusing %d cylinder groups of %.2fMB, %d blks, %d inodes.\n",
191 	    sblock.fs_ncg, (float)sblock.fs_fpg * sblock.fs_fsize * B2MBFACTOR,
192 	    sblock.fs_fpg / sblock.fs_frag, sblock.fs_ipg);
193 	if (sblock.fs_flags & FS_DOSOFTDEP)
194 		printf("\twith soft updates\n");
195 #undef B2MBFACTOR
196 #endif /* FS_DEBUG */
197 
198 	/*
199 	 * Now build the cylinders group blocks and
200 	 * then print out indices of cylinder groups.
201 	 */
202 	printf("super-block backups (for fsck_ffs -b #) at:\n");
203 	i = 0;
204 	width = charsperline();
205 
206 	/*
207 	 * Iterate for only the new cylinder groups.
208 	 */
209 	for (cylno = osblock.fs_ncg; cylno < sblock.fs_ncg; cylno++) {
210 		initcg(cylno, modtime, fso, Nflag);
211 		j = sprintf(tmpbuf, " %jd%s",
212 		    (intmax_t)fsbtodb(&sblock, cgsblock(&sblock, cylno)),
213 		    cylno < (sblock.fs_ncg - 1) ? "," : "" );
214 		if (i + j >= width) {
215 			printf("\n");
216 			i = 0;
217 		}
218 		i += j;
219 		printf("%s", tmpbuf);
220 		fflush(stdout);
221 	}
222 	printf("\n");
223 
224 	/*
225 	 * Do all needed changes in the first cylinder group.
226 	 * allocate blocks in new location
227 	 */
228 	updcsloc(modtime, fsi, fso, Nflag);
229 
230 	/*
231 	 * Clean up the dynamic fields in our superblock.
232 	 *
233 	 * XXX
234 	 * The following fields are currently distributed from the superblock
235 	 * to the copies:
236 	 *     fs_minfree
237 	 *     fs_rotdelay
238 	 *     fs_maxcontig
239 	 *     fs_maxbpg
240 	 *     fs_minfree,
241 	 *     fs_optim
242 	 *     fs_flags
243 	 *
244 	 * We probably should rather change the summary for the cylinder group
245 	 * statistics here to the value of what would be in there, if the file
246 	 * system were created initially with the new size. Therefor we still
247 	 * need to find an easy way of calculating that.
248 	 * Possibly we can try to read the first superblock copy and apply the
249 	 * "diffed" stats between the old and new superblock by still copying
250 	 * certain parameters onto that.
251 	 */
252 	sblock.fs_time = modtime;
253 	sblock.fs_fmod = 0;
254 	sblock.fs_clean = 1;
255 	sblock.fs_ronly = 0;
256 	sblock.fs_cgrotor = 0;
257 	sblock.fs_state = 0;
258 	memset((void *)&sblock.fs_fsmnt, 0, sizeof(sblock.fs_fsmnt));
259 
260 	/*
261 	 * Now write the new superblock, its summary information,
262 	 * and all the alternates back to disk.
263 	 */
264 	if (!Nflag && sbput(fso, &sblock, sblock.fs_ncg) != 0)
265 		errc(2, EIO, "could not write updated superblock");
266 	DBG_PRINT0("fscs written\n");
267 
268 #ifdef FS_DEBUG
269 	{
270 		struct csum	*dbg_csp;
271 		u_int32_t	dbg_csc;
272 		char	dbg_line[80];
273 
274 		dbg_csp = fscs;
275 		for (dbg_csc = 0; dbg_csc < sblock.fs_ncg; dbg_csc++) {
276 			snprintf(dbg_line, sizeof(dbg_line),
277 			    "%d. new csum in new location", dbg_csc);
278 			DBG_DUMP_CSUM(&sblock, dbg_line, dbg_csp++);
279 		}
280 	}
281 #endif /* FS_DEBUG */
282 
283 	DBG_PRINT0("sblock written\n");
284 	DBG_DUMP_FS(&sblock, "new initial sblock");
285 
286 	DBG_PRINT0("sblock copies written\n");
287 	DBG_DUMP_FS(&sblock, "new other sblocks");
288 
289 	DBG_LEAVE;
290 	return;
291 }
292 
293 /*
294  * This creates a new cylinder group structure, for more details please see
295  * the source of newfs(8), as this function is taken over almost unchanged.
296  * As this is never called for the first cylinder group, the special
297  * provisions for that case are removed here.
298  */
299 static void
300 initcg(int cylno, time_t modtime, int fso, unsigned int Nflag)
301 {
302 	DBG_FUNC("initcg")
303 	static caddr_t iobuf;
304 	static long iobufsize;
305 	long blkno, start;
306 	ino_t ino;
307 	ufs2_daddr_t i, cbase, dmax;
308 	struct ufs1_dinode *dp1;
309 	struct ufs2_dinode *dp2;
310 	struct csum *cs;
311 	uint j, d, dupper, dlower;
312 
313 	if (iobuf == NULL) {
314 		iobufsize = 2 * sblock.fs_bsize;
315 		if ((iobuf = malloc(iobufsize)) == NULL)
316 			errx(37, "panic: cannot allocate I/O buffer");
317 		memset(iobuf, '\0', iobufsize);
318 	}
319 	/*
320 	 * Determine block bounds for cylinder group.
321 	 * Allow space for super block summary information in first
322 	 * cylinder group.
323 	 */
324 	cbase = cgbase(&sblock, cylno);
325 	dmax = cbase + sblock.fs_fpg;
326 	if (dmax > sblock.fs_size)
327 		dmax = sblock.fs_size;
328 	dlower = cgsblock(&sblock, cylno) - cbase;
329 	dupper = cgdmin(&sblock, cylno) - cbase;
330 	if (cylno == 0)	/* XXX fscs may be relocated */
331 		dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
332 	cs = &fscs[cylno];
333 	memset(&acg, 0, sblock.fs_cgsize);
334 	acg.cg_time = modtime;
335 	acg.cg_magic = CG_MAGIC;
336 	acg.cg_cgx = cylno;
337 	acg.cg_niblk = sblock.fs_ipg;
338 	acg.cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
339 	acg.cg_ndblk = dmax - cbase;
340 	if (sblock.fs_contigsumsize > 0)
341 		acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
342 	start = &acg.cg_space[0] - (u_char *)(&acg.cg_firstfield);
343 	if (sblock.fs_magic == FS_UFS2_MAGIC) {
344 		acg.cg_iusedoff = start;
345 	} else {
346 		acg.cg_old_ncyl = sblock.fs_old_cpg;
347 		acg.cg_old_time = acg.cg_time;
348 		acg.cg_time = 0;
349 		acg.cg_old_niblk = acg.cg_niblk;
350 		acg.cg_niblk = 0;
351 		acg.cg_initediblk = 0;
352 		acg.cg_old_btotoff = start;
353 		acg.cg_old_boff = acg.cg_old_btotoff +
354 		    sblock.fs_old_cpg * sizeof(int32_t);
355 		acg.cg_iusedoff = acg.cg_old_boff +
356 		    sblock.fs_old_cpg * sizeof(u_int16_t);
357 	}
358 	acg.cg_freeoff = acg.cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
359 	acg.cg_nextfreeoff = acg.cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT);
360 	if (sblock.fs_contigsumsize > 0) {
361 		acg.cg_clustersumoff =
362 		    roundup(acg.cg_nextfreeoff, sizeof(u_int32_t));
363 		acg.cg_clustersumoff -= sizeof(u_int32_t);
364 		acg.cg_clusteroff = acg.cg_clustersumoff +
365 		    (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
366 		acg.cg_nextfreeoff = acg.cg_clusteroff +
367 		    howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
368 	}
369 	if (acg.cg_nextfreeoff > (unsigned)sblock.fs_cgsize) {
370 		/*
371 		 * This should never happen as we would have had that panic
372 		 * already on file system creation
373 		 */
374 		errx(37, "panic: cylinder group too big");
375 	}
376 	acg.cg_cs.cs_nifree += sblock.fs_ipg;
377 	if (cylno == 0)
378 		for (ino = 0; ino < UFS_ROOTINO; ino++) {
379 			setbit(cg_inosused(&acg), ino);
380 			acg.cg_cs.cs_nifree--;
381 		}
382 	/*
383 	 * Initialize the initial inode blocks.
384 	 */
385 	dp1 = (struct ufs1_dinode *)(void *)iobuf;
386 	dp2 = (struct ufs2_dinode *)(void *)iobuf;
387 	for (i = 0; i < acg.cg_initediblk; i++) {
388 		if (sblock.fs_magic == FS_UFS1_MAGIC) {
389 			dp1->di_gen = arc4random();
390 			dp1++;
391 		} else {
392 			dp2->di_gen = arc4random();
393 			dp2++;
394 		}
395 	}
396 	wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno)), iobufsize, iobuf,
397 	    fso, Nflag);
398 	/*
399 	 * For the old file system, we have to initialize all the inodes.
400 	 */
401 	if (sblock.fs_magic == FS_UFS1_MAGIC &&
402 	    sblock.fs_ipg > 2 * INOPB(&sblock)) {
403 		for (i = 2 * sblock.fs_frag;
404 		     i < sblock.fs_ipg / INOPF(&sblock);
405 		     i += sblock.fs_frag) {
406 			dp1 = (struct ufs1_dinode *)(void *)iobuf;
407 			for (j = 0; j < INOPB(&sblock); j++) {
408 				dp1->di_gen = arc4random();
409 				dp1++;
410 			}
411 			wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno) + i),
412 			    sblock.fs_bsize, iobuf, fso, Nflag);
413 		}
414 	}
415 	if (cylno > 0) {
416 		/*
417 		 * In cylno 0, beginning space is reserved
418 		 * for boot and super blocks.
419 		 */
420 		for (d = 0; d < dlower; d += sblock.fs_frag) {
421 			blkno = d / sblock.fs_frag;
422 			setblock(&sblock, cg_blksfree(&acg), blkno);
423 			if (sblock.fs_contigsumsize > 0)
424 				setbit(cg_clustersfree(&acg), blkno);
425 			acg.cg_cs.cs_nbfree++;
426 		}
427 		sblock.fs_dsize += dlower;
428 	}
429 	sblock.fs_dsize += acg.cg_ndblk - dupper;
430 	if ((i = dupper % sblock.fs_frag)) {
431 		acg.cg_frsum[sblock.fs_frag - i]++;
432 		for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
433 			setbit(cg_blksfree(&acg), dupper);
434 			acg.cg_cs.cs_nffree++;
435 		}
436 	}
437 	for (d = dupper; d + sblock.fs_frag <= acg.cg_ndblk;
438 	    d += sblock.fs_frag) {
439 		blkno = d / sblock.fs_frag;
440 		setblock(&sblock, cg_blksfree(&acg), blkno);
441 		if (sblock.fs_contigsumsize > 0)
442 			setbit(cg_clustersfree(&acg), blkno);
443 		acg.cg_cs.cs_nbfree++;
444 	}
445 	if (d < acg.cg_ndblk) {
446 		acg.cg_frsum[acg.cg_ndblk - d]++;
447 		for (; d < acg.cg_ndblk; d++) {
448 			setbit(cg_blksfree(&acg), d);
449 			acg.cg_cs.cs_nffree++;
450 		}
451 	}
452 	if (sblock.fs_contigsumsize > 0) {
453 		int32_t *sump = cg_clustersum(&acg);
454 		u_char *mapp = cg_clustersfree(&acg);
455 		int map = *mapp++;
456 		int bit = 1;
457 		int run = 0;
458 
459 		for (i = 0; i < acg.cg_nclusterblks; i++) {
460 			if ((map & bit) != 0)
461 				run++;
462 			else if (run != 0) {
463 				if (run > sblock.fs_contigsumsize)
464 					run = sblock.fs_contigsumsize;
465 				sump[run]++;
466 				run = 0;
467 			}
468 			if ((i & (CHAR_BIT - 1)) != CHAR_BIT - 1)
469 				bit <<= 1;
470 			else {
471 				map = *mapp++;
472 				bit = 1;
473 			}
474 		}
475 		if (run != 0) {
476 			if (run > sblock.fs_contigsumsize)
477 				run = sblock.fs_contigsumsize;
478 			sump[run]++;
479 		}
480 	}
481 	sblock.fs_cstotal.cs_ndir += acg.cg_cs.cs_ndir;
482 	sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
483 	sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
484 	sblock.fs_cstotal.cs_nifree += acg.cg_cs.cs_nifree;
485 	*cs = acg.cg_cs;
486 
487 	cgckhash(&acg);
488 	wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), sblock.fs_cgsize, &acg,
489 	    fso, Nflag);
490 	DBG_DUMP_CG(&sblock, "new cg", &acg);
491 
492 	DBG_LEAVE;
493 	return;
494 }
495 
496 /*
497  * Here we add or subtract (sign +1/-1) the available fragments in a given
498  * block to or from the fragment statistics. By subtracting before and adding
499  * after an operation on the free frag map we can easy update the fragment
500  * statistic, which seems to be otherwise a rather complex operation.
501  */
502 static void
503 frag_adjust(ufs2_daddr_t frag, int sign)
504 {
505 	DBG_FUNC("frag_adjust")
506 	int fragsize;
507 	int f;
508 
509 	DBG_ENTER;
510 
511 	fragsize = 0;
512 	/*
513 	 * Here frag only needs to point to any fragment in the block we want
514 	 * to examine.
515 	 */
516 	for (f = rounddown(frag, sblock.fs_frag);
517 	    f < roundup(frag + 1, sblock.fs_frag); f++) {
518 		/*
519 		 * Count contiguous free fragments.
520 		 */
521 		if (isset(cg_blksfree(&acg), f)) {
522 			fragsize++;
523 		} else {
524 			if (fragsize && fragsize < sblock.fs_frag) {
525 				/*
526 				 * We found something in between.
527 				 */
528 				acg.cg_frsum[fragsize] += sign;
529 				DBG_PRINT2("frag_adjust [%d]+=%d\n",
530 				    fragsize, sign);
531 			}
532 			fragsize = 0;
533 		}
534 	}
535 	if (fragsize && fragsize < sblock.fs_frag) {
536 		/*
537 		 * We found something.
538 		 */
539 		acg.cg_frsum[fragsize] += sign;
540 		DBG_PRINT2("frag_adjust [%d]+=%d\n", fragsize, sign);
541 	}
542 	DBG_PRINT2("frag_adjust [[%d]]+=%d\n", fragsize, sign);
543 
544 	DBG_LEAVE;
545 	return;
546 }
547 
548 /*
549  * Here we do all needed work for the former last cylinder group. It has to be
550  * changed in any case, even if the file system ended exactly on the end of
551  * this group, as there is some slightly inconsistent handling of the number
552  * of cylinders in the cylinder group. We start again by reading the cylinder
553  * group from disk. If the last block was not fully available, we first handle
554  * the missing fragments, then we handle all new full blocks in that file
555  * system and finally we handle the new last fragmented block in the file
556  * system.  We again have to handle the fragment statistics rotational layout
557  * tables and cluster summary during all those operations.
558  */
559 static void
560 updjcg(int cylno, time_t modtime, int fsi, int fso, unsigned int Nflag)
561 {
562 	DBG_FUNC("updjcg")
563 	ufs2_daddr_t cbase, dmax, dupper;
564 	struct csum *cs;
565 	int i, k;
566 	int j = 0;
567 
568 	DBG_ENTER;
569 
570 	/*
571 	 * Read the former last (joining) cylinder group from disk, and make
572 	 * a copy.
573 	 */
574 	rdfs(fsbtodb(&osblock, cgtod(&osblock, cylno)),
575 	    (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
576 	DBG_PRINT0("jcg read\n");
577 	DBG_DUMP_CG(&sblock, "old joining cg", &aocg);
578 
579 	memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
580 
581 	/*
582 	 * If the cylinder group had already its new final size almost
583 	 * nothing is to be done ... except:
584 	 * For some reason the value of cg_ncyl in the last cylinder group has
585 	 * to be zero instead of fs_cpg. As this is now no longer the last
586 	 * cylinder group we have to change that value now to fs_cpg.
587 	 */
588 
589 	if (cgbase(&osblock, cylno + 1) == osblock.fs_size) {
590 		if (sblock.fs_magic == FS_UFS1_MAGIC)
591 			acg.cg_old_ncyl = sblock.fs_old_cpg;
592 
593 		wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
594 		    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
595 		DBG_PRINT0("jcg written\n");
596 		DBG_DUMP_CG(&sblock, "new joining cg", &acg);
597 
598 		DBG_LEAVE;
599 		return;
600 	}
601 
602 	/*
603 	 * Set up some variables needed later.
604 	 */
605 	cbase = cgbase(&sblock, cylno);
606 	dmax = cbase + sblock.fs_fpg;
607 	if (dmax > sblock.fs_size)
608 		dmax = sblock.fs_size;
609 	dupper = cgdmin(&sblock, cylno) - cbase;
610 	if (cylno == 0) /* XXX fscs may be relocated */
611 		dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
612 
613 	/*
614 	 * Set pointer to the cylinder summary for our cylinder group.
615 	 */
616 	cs = fscs + cylno;
617 
618 	/*
619 	 * Touch the cylinder group, update all fields in the cylinder group as
620 	 * needed, update the free space in the superblock.
621 	 */
622 	acg.cg_time = modtime;
623 	if ((unsigned)cylno == sblock.fs_ncg - 1) {
624 		/*
625 		 * This is still the last cylinder group.
626 		 */
627 		if (sblock.fs_magic == FS_UFS1_MAGIC)
628 			acg.cg_old_ncyl =
629 			    sblock.fs_old_ncyl % sblock.fs_old_cpg;
630 	} else {
631 		acg.cg_old_ncyl = sblock.fs_old_cpg;
632 	}
633 	DBG_PRINT2("jcg dbg: %d %u", cylno, sblock.fs_ncg);
634 #ifdef FS_DEBUG
635 	if (sblock.fs_magic == FS_UFS1_MAGIC)
636 		DBG_PRINT2("%d %u", acg.cg_old_ncyl, sblock.fs_old_cpg);
637 #endif
638 	DBG_PRINT0("\n");
639 	acg.cg_ndblk = dmax - cbase;
640 	sblock.fs_dsize += acg.cg_ndblk - aocg.cg_ndblk;
641 	if (sblock.fs_contigsumsize > 0)
642 		acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
643 
644 	/*
645 	 * Now we have to update the free fragment bitmap for our new free
646 	 * space.  There again we have to handle the fragmentation and also
647 	 * the rotational layout tables and the cluster summary.  This is
648 	 * also done per fragment for the first new block if the old file
649 	 * system end was not on a block boundary, per fragment for the new
650 	 * last block if the new file system end is not on a block boundary,
651 	 * and per block for all space in between.
652 	 *
653 	 * Handle the first new block here if it was partially available
654 	 * before.
655 	 */
656 	if (osblock.fs_size % sblock.fs_frag) {
657 		if (roundup(osblock.fs_size, sblock.fs_frag) <=
658 		    sblock.fs_size) {
659 			/*
660 			 * The new space is enough to fill at least this
661 			 * block
662 			 */
663 			j = 0;
664 			for (i = roundup(osblock.fs_size - cbase,
665 			    sblock.fs_frag) - 1; i >= osblock.fs_size - cbase;
666 			    i--) {
667 				setbit(cg_blksfree(&acg), i);
668 				acg.cg_cs.cs_nffree++;
669 				j++;
670 			}
671 
672 			/*
673 			 * Check if the fragment just created could join an
674 			 * already existing fragment at the former end of the
675 			 * file system.
676 			 */
677 			if (isblock(&sblock, cg_blksfree(&acg),
678 			    ((osblock.fs_size - cgbase(&sblock, cylno)) /
679 			     sblock.fs_frag))) {
680 				/*
681 				 * The block is now completely available.
682 				 */
683 				DBG_PRINT0("block was\n");
684 				acg.cg_frsum[osblock.fs_size % sblock.fs_frag]--;
685 				acg.cg_cs.cs_nbfree++;
686 				acg.cg_cs.cs_nffree -= sblock.fs_frag;
687 				k = rounddown(osblock.fs_size - cbase,
688 				    sblock.fs_frag);
689 				updclst((osblock.fs_size - cbase) /
690 				    sblock.fs_frag);
691 			} else {
692 				/*
693 				 * Lets rejoin a possible partially growed
694 				 * fragment.
695 				 */
696 				k = 0;
697 				while (isset(cg_blksfree(&acg), i) &&
698 				    (i >= rounddown(osblock.fs_size - cbase,
699 				    sblock.fs_frag))) {
700 					i--;
701 					k++;
702 				}
703 				if (k)
704 					acg.cg_frsum[k]--;
705 				acg.cg_frsum[k + j]++;
706 			}
707 		} else {
708 			/*
709 			 * We only grow by some fragments within this last
710 			 * block.
711 			 */
712 			for (i = sblock.fs_size - cbase - 1;
713 			    i >= osblock.fs_size - cbase; i--) {
714 				setbit(cg_blksfree(&acg), i);
715 				acg.cg_cs.cs_nffree++;
716 				j++;
717 			}
718 			/*
719 			 * Lets rejoin a possible partially growed fragment.
720 			 */
721 			k = 0;
722 			while (isset(cg_blksfree(&acg), i) &&
723 			    (i >= rounddown(osblock.fs_size - cbase,
724 			    sblock.fs_frag))) {
725 				i--;
726 				k++;
727 			}
728 			if (k)
729 				acg.cg_frsum[k]--;
730 			acg.cg_frsum[k + j]++;
731 		}
732 	}
733 
734 	/*
735 	 * Handle all new complete blocks here.
736 	 */
737 	for (i = roundup(osblock.fs_size - cbase, sblock.fs_frag);
738 	    i + sblock.fs_frag <= dmax - cbase;	/* XXX <= or only < ? */
739 	    i += sblock.fs_frag) {
740 		j = i / sblock.fs_frag;
741 		setblock(&sblock, cg_blksfree(&acg), j);
742 		updclst(j);
743 		acg.cg_cs.cs_nbfree++;
744 	}
745 
746 	/*
747 	 * Handle the last new block if there are stll some new fragments left.
748 	 * Here we don't have to bother about the cluster summary or the even
749 	 * the rotational layout table.
750 	 */
751 	if (i < (dmax - cbase)) {
752 		acg.cg_frsum[dmax - cbase - i]++;
753 		for (; i < dmax - cbase; i++) {
754 			setbit(cg_blksfree(&acg), i);
755 			acg.cg_cs.cs_nffree++;
756 		}
757 	}
758 
759 	sblock.fs_cstotal.cs_nffree +=
760 	    (acg.cg_cs.cs_nffree - aocg.cg_cs.cs_nffree);
761 	sblock.fs_cstotal.cs_nbfree +=
762 	    (acg.cg_cs.cs_nbfree - aocg.cg_cs.cs_nbfree);
763 	/*
764 	 * The following statistics are not changed here:
765 	 *     sblock.fs_cstotal.cs_ndir
766 	 *     sblock.fs_cstotal.cs_nifree
767 	 * As the statistics for this cylinder group are ready, copy it to
768 	 * the summary information array.
769 	 */
770 	*cs = acg.cg_cs;
771 
772 	/*
773 	 * Write the updated "joining" cylinder group back to disk.
774 	 */
775 	cgckhash(&acg);
776 	wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize,
777 	    (void *)&acg, fso, Nflag);
778 	DBG_PRINT0("jcg written\n");
779 	DBG_DUMP_CG(&sblock, "new joining cg", &acg);
780 
781 	DBG_LEAVE;
782 	return;
783 }
784 
785 /*
786  * Here we update the location of the cylinder summary. We have two possible
787  * ways of growing the cylinder summary:
788  * (1)	We can try to grow the summary in the current location, and relocate
789  *	possibly used blocks within the current cylinder group.
790  * (2)	Alternatively we can relocate the whole cylinder summary to the first
791  *	new completely empty cylinder group. Once the cylinder summary is no
792  *	longer in the beginning of the first cylinder group you should never
793  *	use a version of fsck which is not aware of the possibility to have
794  *	this structure in a non standard place.
795  * Option (2) is considered to be less intrusive to the structure of the file-
796  * system, so that's the one being used.
797  */
798 static void
799 updcsloc(time_t modtime, int fsi, int fso, unsigned int Nflag)
800 {
801 	DBG_FUNC("updcsloc")
802 	struct csum *cs;
803 	int ocscg, ncscg;
804 	ufs2_daddr_t d;
805 	int lcs = 0;
806 	int block;
807 
808 	DBG_ENTER;
809 
810 	if (howmany(sblock.fs_cssize, sblock.fs_fsize) ==
811 	    howmany(osblock.fs_cssize, osblock.fs_fsize)) {
812 		/*
813 		 * No new fragment needed.
814 		 */
815 		DBG_LEAVE;
816 		return;
817 	}
818 	ocscg = dtog(&osblock, osblock.fs_csaddr);
819 	cs = fscs + ocscg;
820 
821 	/*
822 	 * Read original cylinder group from disk, and make a copy.
823 	 * XXX	If Nflag is set in some very rare cases we now miss
824 	 *	some changes done in updjcg by reading the unmodified
825 	 *	block from disk.
826 	 */
827 	rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)),
828 	    (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
829 	DBG_PRINT0("oscg read\n");
830 	DBG_DUMP_CG(&sblock, "old summary cg", &aocg);
831 
832 	memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
833 
834 	/*
835 	 * Touch the cylinder group, set up local variables needed later
836 	 * and update the superblock.
837 	 */
838 	acg.cg_time = modtime;
839 
840 	/*
841 	 * XXX	In the case of having active snapshots we may need much more
842 	 *	blocks for the copy on write. We need each block twice, and
843 	 *	also up to 8*3 blocks for indirect blocks for all possible
844 	 *	references.
845 	 */
846 	/*
847 	 * There is not enough space in the old cylinder group to
848 	 * relocate all blocks as needed, so we relocate the whole
849 	 * cylinder group summary to a new group. We try to use the
850 	 * first complete new cylinder group just created. Within the
851 	 * cylinder group we align the area immediately after the
852 	 * cylinder group information location in order to be as
853 	 * close as possible to the original implementation of ffs.
854 	 *
855 	 * First we have to make sure we'll find enough space in the
856 	 * new cylinder group. If not, then we currently give up.
857 	 * We start with freeing everything which was used by the
858 	 * fragments of the old cylinder summary in the current group.
859 	 * Now we write back the group meta data, read in the needed
860 	 * meta data from the new cylinder group, and start allocating
861 	 * within that group. Here we can assume, the group to be
862 	 * completely empty. Which makes the handling of fragments and
863 	 * clusters a lot easier.
864 	 */
865 	DBG_TRC;
866 	if (sblock.fs_ncg - osblock.fs_ncg < 2)
867 		errx(2, "panic: not enough space");
868 
869 	/*
870 	 * Point "d" to the first fragment not used by the cylinder
871 	 * summary.
872 	 */
873 	d = osblock.fs_csaddr + (osblock.fs_cssize / osblock.fs_fsize);
874 
875 	/*
876 	 * Set up last cluster size ("lcs") already here. Calculate
877 	 * the size for the trailing cluster just behind where "d"
878 	 * points to.
879 	 */
880 	if (sblock.fs_contigsumsize > 0) {
881 		for (block = howmany(d % sblock.fs_fpg, sblock.fs_frag),
882 		    lcs = 0; lcs < sblock.fs_contigsumsize; block++, lcs++) {
883 			if (isclr(cg_clustersfree(&acg), block))
884 				break;
885 		}
886 	}
887 
888 	/*
889 	 * Point "d" to the last frag used by the cylinder summary.
890 	 */
891 	d--;
892 
893 	DBG_PRINT1("d=%jd\n", (intmax_t)d);
894 	if ((d + 1) % sblock.fs_frag) {
895 		/*
896 		 * The end of the cylinder summary is not a complete
897 		 * block.
898 		 */
899 		DBG_TRC;
900 		frag_adjust(d % sblock.fs_fpg, -1);
901 		for (; (d + 1) % sblock.fs_frag; d--) {
902 			DBG_PRINT1("d=%jd\n", (intmax_t)d);
903 			setbit(cg_blksfree(&acg), d % sblock.fs_fpg);
904 			acg.cg_cs.cs_nffree++;
905 			sblock.fs_cstotal.cs_nffree++;
906 		}
907 		/*
908 		 * Point "d" to the last fragment of the last
909 		 * (incomplete) block of the cylinder summary.
910 		 */
911 		d++;
912 		frag_adjust(d % sblock.fs_fpg, 1);
913 
914 		if (isblock(&sblock, cg_blksfree(&acg),
915 		    (d % sblock.fs_fpg) / sblock.fs_frag)) {
916 			DBG_PRINT1("d=%jd\n", (intmax_t)d);
917 			acg.cg_cs.cs_nffree -= sblock.fs_frag;
918 			acg.cg_cs.cs_nbfree++;
919 			sblock.fs_cstotal.cs_nffree -= sblock.fs_frag;
920 			sblock.fs_cstotal.cs_nbfree++;
921 			if (sblock.fs_contigsumsize > 0) {
922 				setbit(cg_clustersfree(&acg),
923 				    (d % sblock.fs_fpg) / sblock.fs_frag);
924 				if (lcs < sblock.fs_contigsumsize) {
925 					if (lcs)
926 						cg_clustersum(&acg)[lcs]--;
927 					lcs++;
928 					cg_clustersum(&acg)[lcs]++;
929 				}
930 			}
931 		}
932 		/*
933 		 * Point "d" to the first fragment of the block before
934 		 * the last incomplete block.
935 		 */
936 		d--;
937 	}
938 
939 	DBG_PRINT1("d=%jd\n", (intmax_t)d);
940 	for (d = rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr;
941 	    d -= sblock.fs_frag) {
942 		DBG_TRC;
943 		DBG_PRINT1("d=%jd\n", (intmax_t)d);
944 		setblock(&sblock, cg_blksfree(&acg),
945 		    (d % sblock.fs_fpg) / sblock.fs_frag);
946 		acg.cg_cs.cs_nbfree++;
947 		sblock.fs_cstotal.cs_nbfree++;
948 		if (sblock.fs_contigsumsize > 0) {
949 			setbit(cg_clustersfree(&acg),
950 			    (d % sblock.fs_fpg) / sblock.fs_frag);
951 			/*
952 			 * The last cluster size is already set up.
953 			 */
954 			if (lcs < sblock.fs_contigsumsize) {
955 				if (lcs)
956 					cg_clustersum(&acg)[lcs]--;
957 				lcs++;
958 				cg_clustersum(&acg)[lcs]++;
959 			}
960 		}
961 	}
962 	*cs = acg.cg_cs;
963 
964 	/*
965 	 * Now write the former cylinder group containing the cylinder
966 	 * summary back to disk.
967 	 */
968 	wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)),
969 	    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
970 	DBG_PRINT0("oscg written\n");
971 	DBG_DUMP_CG(&sblock, "old summary cg", &acg);
972 
973 	/*
974 	 * Find the beginning of the new cylinder group containing the
975 	 * cylinder summary.
976 	 */
977 	sblock.fs_csaddr = cgdmin(&sblock, osblock.fs_ncg);
978 	ncscg = dtog(&sblock, sblock.fs_csaddr);
979 	cs = fscs + ncscg;
980 
981 	/*
982 	 * If Nflag is specified, we would now read random data instead
983 	 * of an empty cg structure from disk. So we can't simulate that
984 	 * part for now.
985 	 */
986 	if (Nflag) {
987 		DBG_PRINT0("nscg update skipped\n");
988 		DBG_LEAVE;
989 		return;
990 	}
991 
992 	/*
993 	 * Read the future cylinder group containing the cylinder
994 	 * summary from disk, and make a copy.
995 	 */
996 	rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
997 	    (size_t)sblock.fs_cgsize, (void *)&aocg, fsi);
998 	DBG_PRINT0("nscg read\n");
999 	DBG_DUMP_CG(&sblock, "new summary cg", &aocg);
1000 
1001 	memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
1002 
1003 	/*
1004 	 * Allocate all complete blocks used by the new cylinder
1005 	 * summary.
1006 	 */
1007 	for (d = sblock.fs_csaddr; d + sblock.fs_frag <=
1008 	    sblock.fs_csaddr + (sblock.fs_cssize / sblock.fs_fsize);
1009 	    d += sblock.fs_frag) {
1010 		clrblock(&sblock, cg_blksfree(&acg),
1011 		    (d % sblock.fs_fpg) / sblock.fs_frag);
1012 		acg.cg_cs.cs_nbfree--;
1013 		sblock.fs_cstotal.cs_nbfree--;
1014 		if (sblock.fs_contigsumsize > 0) {
1015 			clrbit(cg_clustersfree(&acg),
1016 			    (d % sblock.fs_fpg) / sblock.fs_frag);
1017 		}
1018 	}
1019 
1020 	/*
1021 	 * Allocate all fragments used by the cylinder summary in the
1022 	 * last block.
1023 	 */
1024 	if (d < sblock.fs_csaddr + (sblock.fs_cssize / sblock.fs_fsize)) {
1025 		for (; d - sblock.fs_csaddr <
1026 		    sblock.fs_cssize/sblock.fs_fsize; d++) {
1027 			clrbit(cg_blksfree(&acg), d % sblock.fs_fpg);
1028 			acg.cg_cs.cs_nffree--;
1029 			sblock.fs_cstotal.cs_nffree--;
1030 		}
1031 		acg.cg_cs.cs_nbfree--;
1032 		acg.cg_cs.cs_nffree += sblock.fs_frag;
1033 		sblock.fs_cstotal.cs_nbfree--;
1034 		sblock.fs_cstotal.cs_nffree += sblock.fs_frag;
1035 		if (sblock.fs_contigsumsize > 0)
1036 			clrbit(cg_clustersfree(&acg),
1037 			    (d % sblock.fs_fpg) / sblock.fs_frag);
1038 
1039 		frag_adjust(d % sblock.fs_fpg, 1);
1040 	}
1041 	/*
1042 	 * XXX	Handle the cluster statistics here in the case this
1043 	 *	cylinder group is now almost full, and the remaining
1044 	 *	space is less then the maximum cluster size. This is
1045 	 *	probably not needed, as you would hardly find a file
1046 	 *	system which has only MAXCSBUFS+FS_MAXCONTIG of free
1047 	 *	space right behind the cylinder group information in
1048 	 *	any new cylinder group.
1049 	 */
1050 
1051 	/*
1052 	 * Update our statistics in the cylinder summary.
1053 	 */
1054 	*cs = acg.cg_cs;
1055 
1056 	/*
1057 	 * Write the new cylinder group containing the cylinder summary
1058 	 * back to disk.
1059 	 */
1060 	wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1061 	    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1062 	DBG_PRINT0("nscg written\n");
1063 	DBG_DUMP_CG(&sblock, "new summary cg", &acg);
1064 
1065 	DBG_LEAVE;
1066 	return;
1067 }
1068 
1069 /*
1070  * Here we read some block(s) from disk.
1071  */
1072 static void
1073 rdfs(ufs2_daddr_t bno, size_t size, void *bf, int fsi)
1074 {
1075 	DBG_FUNC("rdfs")
1076 	ssize_t	n;
1077 
1078 	DBG_ENTER;
1079 
1080 	if (bno < 0)
1081 		err(32, "rdfs: attempting to read negative block number");
1082 	if (lseek(fsi, (off_t)bno * DEV_BSIZE, 0) < 0)
1083 		err(33, "rdfs: seek error: %jd", (intmax_t)bno);
1084 	n = read(fsi, bf, size);
1085 	if (n != (ssize_t)size)
1086 		err(34, "rdfs: read error: %jd", (intmax_t)bno);
1087 
1088 	DBG_LEAVE;
1089 	return;
1090 }
1091 
1092 /*
1093  * Here we write some block(s) to disk.
1094  */
1095 static void
1096 wtfs(ufs2_daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag)
1097 {
1098 	DBG_FUNC("wtfs")
1099 	ssize_t	n;
1100 
1101 	DBG_ENTER;
1102 
1103 	if (Nflag) {
1104 		DBG_LEAVE;
1105 		return;
1106 	}
1107 	if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0)
1108 		err(35, "wtfs: seek error: %ld", (long)bno);
1109 	n = write(fso, bf, size);
1110 	if (n != (ssize_t)size)
1111 		err(36, "wtfs: write error: %ld", (long)bno);
1112 
1113 	DBG_LEAVE;
1114 	return;
1115 }
1116 
1117 /*
1118  * Here we check if all frags of a block are free. For more details again
1119  * please see the source of newfs(8), as this function is taken over almost
1120  * unchanged.
1121  */
1122 static int
1123 isblock(struct fs *fs, unsigned char *cp, int h)
1124 {
1125 	DBG_FUNC("isblock")
1126 	unsigned char mask;
1127 
1128 	DBG_ENTER;
1129 
1130 	switch (fs->fs_frag) {
1131 	case 8:
1132 		DBG_LEAVE;
1133 		return (cp[h] == 0xff);
1134 	case 4:
1135 		mask = 0x0f << ((h & 0x1) << 2);
1136 		DBG_LEAVE;
1137 		return ((cp[h >> 1] & mask) == mask);
1138 	case 2:
1139 		mask = 0x03 << ((h & 0x3) << 1);
1140 		DBG_LEAVE;
1141 		return ((cp[h >> 2] & mask) == mask);
1142 	case 1:
1143 		mask = 0x01 << (h & 0x7);
1144 		DBG_LEAVE;
1145 		return ((cp[h >> 3] & mask) == mask);
1146 	default:
1147 		fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
1148 		DBG_LEAVE;
1149 		return (0);
1150 	}
1151 }
1152 
1153 /*
1154  * Here we allocate a complete block in the block map. For more details again
1155  * please see the source of newfs(8), as this function is taken over almost
1156  * unchanged.
1157  */
1158 static void
1159 clrblock(struct fs *fs, unsigned char *cp, int h)
1160 {
1161 	DBG_FUNC("clrblock")
1162 
1163 	DBG_ENTER;
1164 
1165 	switch ((fs)->fs_frag) {
1166 	case 8:
1167 		cp[h] = 0;
1168 		break;
1169 	case 4:
1170 		cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1171 		break;
1172 	case 2:
1173 		cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1174 		break;
1175 	case 1:
1176 		cp[h >> 3] &= ~(0x01 << (h & 0x7));
1177 		break;
1178 	default:
1179 		warnx("clrblock bad fs_frag %d", fs->fs_frag);
1180 		break;
1181 	}
1182 
1183 	DBG_LEAVE;
1184 	return;
1185 }
1186 
1187 /*
1188  * Here we free a complete block in the free block map. For more details again
1189  * please see the source of newfs(8), as this function is taken over almost
1190  * unchanged.
1191  */
1192 static void
1193 setblock(struct fs *fs, unsigned char *cp, int h)
1194 {
1195 	DBG_FUNC("setblock")
1196 
1197 	DBG_ENTER;
1198 
1199 	switch (fs->fs_frag) {
1200 	case 8:
1201 		cp[h] = 0xff;
1202 		break;
1203 	case 4:
1204 		cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1205 		break;
1206 	case 2:
1207 		cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1208 		break;
1209 	case 1:
1210 		cp[h >> 3] |= (0x01 << (h & 0x7));
1211 		break;
1212 	default:
1213 		warnx("setblock bad fs_frag %d", fs->fs_frag);
1214 		break;
1215 	}
1216 
1217 	DBG_LEAVE;
1218 	return;
1219 }
1220 
1221 /*
1222  * Figure out how many lines our current terminal has. For more details again
1223  * please see the source of newfs(8), as this function is taken over almost
1224  * unchanged.
1225  */
1226 static int
1227 charsperline(void)
1228 {
1229 	DBG_FUNC("charsperline")
1230 	int columns;
1231 	char *cp;
1232 	struct winsize ws;
1233 
1234 	DBG_ENTER;
1235 
1236 	columns = 0;
1237 	if (ioctl(0, TIOCGWINSZ, &ws) != -1)
1238 		columns = ws.ws_col;
1239 	if (columns == 0 && (cp = getenv("COLUMNS")))
1240 		columns = atoi(cp);
1241 	if (columns == 0)
1242 		columns = 80;	/* last resort */
1243 
1244 	DBG_LEAVE;
1245 	return (columns);
1246 }
1247 
1248 static int
1249 is_dev(const char *name)
1250 {
1251 	struct stat devstat;
1252 
1253 	if (stat(name, &devstat) != 0)
1254 		return (0);
1255 	if (!S_ISCHR(devstat.st_mode))
1256 		return (0);
1257 	return (1);
1258 }
1259 
1260 /*
1261  * Return mountpoint on which the device is currently mounted.
1262  */
1263 static const struct statfs *
1264 dev_to_statfs(const char *dev)
1265 {
1266 	struct stat devstat, mntdevstat;
1267 	struct statfs *mntbuf, *statfsp;
1268 	char device[MAXPATHLEN];
1269 	char *mntdevname;
1270 	int i, mntsize;
1271 
1272 	/*
1273 	 * First check the mounted filesystems.
1274 	 */
1275 	if (stat(dev, &devstat) != 0)
1276 		return (NULL);
1277 	if (!S_ISCHR(devstat.st_mode) && !S_ISBLK(devstat.st_mode))
1278 		return (NULL);
1279 
1280 	mntsize = getmntinfo(&mntbuf, MNT_NOWAIT);
1281 	for (i = 0; i < mntsize; i++) {
1282 		statfsp = &mntbuf[i];
1283 		mntdevname = statfsp->f_mntfromname;
1284 		if (*mntdevname != '/') {
1285 			strcpy(device, _PATH_DEV);
1286 			strcat(device, mntdevname);
1287 			mntdevname = device;
1288 		}
1289 		if (stat(mntdevname, &mntdevstat) == 0 &&
1290 		    mntdevstat.st_rdev == devstat.st_rdev)
1291 			return (statfsp);
1292 	}
1293 
1294 	return (NULL);
1295 }
1296 
1297 static const char *
1298 mountpoint_to_dev(const char *mountpoint)
1299 {
1300 	struct statfs *mntbuf, *statfsp;
1301 	struct fstab *fs;
1302 	int i, mntsize;
1303 
1304 	/*
1305 	 * First check the mounted filesystems.
1306 	 */
1307 	mntsize = getmntinfo(&mntbuf, MNT_NOWAIT);
1308 	for (i = 0; i < mntsize; i++) {
1309 		statfsp = &mntbuf[i];
1310 
1311 		if (strcmp(statfsp->f_mntonname, mountpoint) == 0)
1312 			return (statfsp->f_mntfromname);
1313 	}
1314 
1315 	/*
1316 	 * Check the fstab.
1317 	 */
1318 	fs = getfsfile(mountpoint);
1319 	if (fs != NULL)
1320 		return (fs->fs_spec);
1321 
1322 	return (NULL);
1323 }
1324 
1325 static const char *
1326 getdev(const char *name)
1327 {
1328 	static char device[MAXPATHLEN];
1329 	const char *cp, *dev;
1330 
1331 	if (is_dev(name))
1332 		return (name);
1333 
1334 	cp = strrchr(name, '/');
1335 	if (cp == NULL) {
1336 		snprintf(device, sizeof(device), "%s%s", _PATH_DEV, name);
1337 		if (is_dev(device))
1338 			return (device);
1339 	}
1340 
1341 	dev = mountpoint_to_dev(name);
1342 	if (dev != NULL && is_dev(dev))
1343 		return (dev);
1344 
1345 	return (NULL);
1346 }
1347 
1348 /*
1349  * growfs(8) is a utility which allows to increase the size of an existing
1350  * ufs file system. Currently this can only be done on unmounted file system.
1351  * It recognizes some command line options to specify the new desired size,
1352  * and it does some basic checkings. The old file system size is determined
1353  * and after some more checks like we can really access the new last block
1354  * on the disk etc. we calculate the new parameters for the superblock. After
1355  * having done this we just call growfs() which will do the work.
1356  * We still have to provide support for snapshots. Therefore we first have to
1357  * understand what data structures are always replicated in the snapshot on
1358  * creation, for all other blocks we touch during our procedure, we have to
1359  * keep the old blocks unchanged somewhere available for the snapshots. If we
1360  * are lucky, then we only have to handle our blocks to be relocated in that
1361  * way.
1362  * Also we have to consider in what order we actually update the critical
1363  * data structures of the file system to make sure, that in case of a disaster
1364  * fsck(8) is still able to restore any lost data.
1365  * The foreseen last step then will be to provide for growing even mounted
1366  * file systems. There we have to extend the mount() system call to provide
1367  * userland access to the file system locking facility.
1368  */
1369 int
1370 main(int argc, char **argv)
1371 {
1372 	DBG_FUNC("main")
1373 	struct fs *fs;
1374 	const char *device;
1375 	const struct statfs *statfsp;
1376 	uint64_t size = 0;
1377 	off_t mediasize;
1378 	int error, j, fsi, fso, ch, ret, Nflag = 0, yflag = 0;
1379 	char *p, reply[5], oldsizebuf[6], newsizebuf[6];
1380 	void *testbuf;
1381 
1382 	DBG_ENTER;
1383 
1384 	while ((ch = getopt(argc, argv, "Ns:vy")) != -1) {
1385 		switch(ch) {
1386 		case 'N':
1387 			Nflag = 1;
1388 			break;
1389 		case 's':
1390 			size = (off_t)strtoumax(optarg, &p, 0);
1391 			if (p == NULL || *p == '\0')
1392 				size *= DEV_BSIZE;
1393 			else if (*p == 'b' || *p == 'B')
1394 				; /* do nothing */
1395 			else if (*p == 'k' || *p == 'K')
1396 				size <<= 10;
1397 			else if (*p == 'm' || *p == 'M')
1398 				size <<= 20;
1399 			else if (*p == 'g' || *p == 'G')
1400 				size <<= 30;
1401 			else if (*p == 't' || *p == 'T') {
1402 				size <<= 30;
1403 				size <<= 10;
1404 			} else
1405 				errx(1, "unknown suffix on -s argument");
1406 			break;
1407 		case 'v': /* for compatibility to newfs */
1408 			break;
1409 		case 'y':
1410 			yflag = 1;
1411 			break;
1412 		case '?':
1413 			/* FALLTHROUGH */
1414 		default:
1415 			usage();
1416 		}
1417 	}
1418 	argc -= optind;
1419 	argv += optind;
1420 
1421 	if (argc != 1)
1422 		usage();
1423 
1424 	/*
1425 	 * Now try to guess the device name.
1426 	 */
1427 	device = getdev(*argv);
1428 	if (device == NULL)
1429 		errx(1, "cannot find special device for %s", *argv);
1430 
1431 	statfsp = dev_to_statfs(device);
1432 
1433 	fsi = open(device, O_RDONLY);
1434 	if (fsi < 0)
1435 		err(1, "%s", device);
1436 
1437 	/*
1438 	 * Try to guess the slice size if not specified.
1439 	 */
1440 	if (ioctl(fsi, DIOCGMEDIASIZE, &mediasize) == -1)
1441 		err(1,"DIOCGMEDIASIZE");
1442 
1443 	/*
1444 	 * Check if that partition is suitable for growing a file system.
1445 	 */
1446 	if (mediasize < 1)
1447 		errx(1, "partition is unavailable");
1448 
1449 	/*
1450 	 * Read the current superblock, and take a backup.
1451 	 */
1452 	if ((ret = sbget(fsi, &fs, STDSB)) != 0) {
1453 		switch (ret) {
1454 		case ENOENT:
1455 			errx(1, "superblock not recognized");
1456 		default:
1457 			errc(1, ret, "unable to read superblock");
1458 		}
1459 	}
1460 	memcpy(&osblock, fs, fs->fs_sbsize);
1461 	free(fs);
1462 	memcpy((void *)&fsun1, (void *)&fsun2, osblock.fs_sbsize);
1463 
1464 	DBG_OPEN("/tmp/growfs.debug"); /* already here we need a superblock */
1465 	DBG_DUMP_FS(&sblock, "old sblock");
1466 
1467 	/*
1468 	 * Determine size to grow to. Default to the device size.
1469 	 */
1470 	if (size == 0)
1471 		size = mediasize;
1472 	else {
1473 		if (size > (uint64_t)mediasize) {
1474 			humanize_number(oldsizebuf, sizeof(oldsizebuf), size,
1475 			    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1476 			humanize_number(newsizebuf, sizeof(newsizebuf),
1477 			    mediasize,
1478 			    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1479 
1480 			errx(1, "requested size %s is larger "
1481 			    "than the available %s", oldsizebuf, newsizebuf);
1482 		}
1483 	}
1484 
1485 	/*
1486 	 * Make sure the new size is a multiple of fs_fsize; /dev/ufssuspend
1487 	 * only supports fragment-aligned IO requests.
1488 	 */
1489 	size -= size % osblock.fs_fsize;
1490 
1491 	if (size <= (uint64_t)(osblock.fs_size * osblock.fs_fsize)) {
1492 		humanize_number(oldsizebuf, sizeof(oldsizebuf),
1493 		    osblock.fs_size * osblock.fs_fsize,
1494 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1495 		humanize_number(newsizebuf, sizeof(newsizebuf), size,
1496 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1497 
1498 		errx(1, "requested size %s is not larger than the current "
1499 		   "filesystem size %s", newsizebuf, oldsizebuf);
1500 	}
1501 
1502 	sblock.fs_size = dbtofsb(&osblock, size / DEV_BSIZE);
1503 	sblock.fs_providersize = dbtofsb(&osblock, mediasize / DEV_BSIZE);
1504 
1505 	/*
1506 	 * Are we really growing?
1507 	 */
1508 	if (osblock.fs_size >= sblock.fs_size) {
1509 		errx(1, "we are not growing (%jd->%jd)",
1510 		    (intmax_t)osblock.fs_size, (intmax_t)sblock.fs_size);
1511 	}
1512 
1513 	/*
1514 	 * Check if we find an active snapshot.
1515 	 */
1516 	if (yflag == 0) {
1517 		for (j = 0; j < FSMAXSNAP; j++) {
1518 			if (sblock.fs_snapinum[j]) {
1519 				errx(1, "active snapshot found in file system; "
1520 				    "please remove all snapshots before "
1521 				    "using growfs");
1522 			}
1523 			if (!sblock.fs_snapinum[j]) /* list is dense */
1524 				break;
1525 		}
1526 	}
1527 
1528 	if (yflag == 0 && Nflag == 0) {
1529 		if (statfsp != NULL && (statfsp->f_flags & MNT_RDONLY) == 0)
1530 			printf("Device is mounted read-write; resizing will "
1531 			    "result in temporary write suspension for %s.\n",
1532 			    statfsp->f_mntonname);
1533 		printf("It's strongly recommended to make a backup "
1534 		    "before growing the file system.\n"
1535 		    "OK to grow filesystem on %s", device);
1536 		if (statfsp != NULL)
1537 			printf(", mounted on %s,", statfsp->f_mntonname);
1538 		humanize_number(oldsizebuf, sizeof(oldsizebuf),
1539 		    osblock.fs_size * osblock.fs_fsize,
1540 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1541 		humanize_number(newsizebuf, sizeof(newsizebuf),
1542 		    sblock.fs_size * sblock.fs_fsize,
1543 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1544 		printf(" from %s to %s? [yes/no] ", oldsizebuf, newsizebuf);
1545 		fflush(stdout);
1546 		fgets(reply, (int)sizeof(reply), stdin);
1547 		if (strcasecmp(reply, "yes\n")){
1548 			printf("Response other than \"yes\"; aborting\n");
1549 			exit(0);
1550 		}
1551 	}
1552 
1553 	/*
1554 	 * Try to access our device for writing.  If it's not mounted,
1555 	 * or mounted read-only, simply open it; otherwise, use UFS
1556 	 * suspension mechanism.
1557 	 */
1558 	if (Nflag) {
1559 		fso = -1;
1560 	} else {
1561 		if (statfsp != NULL && (statfsp->f_flags & MNT_RDONLY) == 0) {
1562 			fso = open(_PATH_UFSSUSPEND, O_RDWR);
1563 			if (fso == -1)
1564 				err(1, "unable to open %s", _PATH_UFSSUSPEND);
1565 			error = ioctl(fso, UFSSUSPEND, &statfsp->f_fsid);
1566 			if (error != 0)
1567 				err(1, "UFSSUSPEND");
1568 		} else {
1569 			fso = open(device, O_WRONLY);
1570 			if (fso < 0)
1571 				err(1, "%s", device);
1572 		}
1573 	}
1574 
1575 	/*
1576 	 * Try to access our new last block in the file system.
1577 	 */
1578 	testbuf = malloc(sblock.fs_fsize);
1579 	if (testbuf == NULL)
1580 		err(1, "malloc");
1581 	rdfs((ufs2_daddr_t)((size - sblock.fs_fsize) / DEV_BSIZE),
1582 	    sblock.fs_fsize, testbuf, fsi);
1583 	wtfs((ufs2_daddr_t)((size - sblock.fs_fsize) / DEV_BSIZE),
1584 	    sblock.fs_fsize, testbuf, fso, Nflag);
1585 	free(testbuf);
1586 
1587 	/*
1588 	 * Now calculate new superblock values and check for reasonable
1589 	 * bound for new file system size:
1590 	 *     fs_size:    is derived from user input
1591 	 *     fs_dsize:   should get updated in the routines creating or
1592 	 *                 updating the cylinder groups on the fly
1593 	 *     fs_cstotal: should get updated in the routines creating or
1594 	 *                 updating the cylinder groups
1595 	 */
1596 
1597 	/*
1598 	 * Update the number of cylinders and cylinder groups in the file system.
1599 	 */
1600 	if (sblock.fs_magic == FS_UFS1_MAGIC) {
1601 		sblock.fs_old_ncyl =
1602 		    sblock.fs_size * sblock.fs_old_nspf / sblock.fs_old_spc;
1603 		if (sblock.fs_size * sblock.fs_old_nspf >
1604 		    sblock.fs_old_ncyl * sblock.fs_old_spc)
1605 			sblock.fs_old_ncyl++;
1606 	}
1607 	sblock.fs_ncg = howmany(sblock.fs_size, sblock.fs_fpg);
1608 
1609 	/*
1610 	 * Allocate last cylinder group only if there is enough room
1611 	 * for at least one data block.
1612 	 */
1613 	if (sblock.fs_size % sblock.fs_fpg != 0 &&
1614 	    sblock.fs_size <= cgdmin(&sblock, sblock.fs_ncg - 1)) {
1615 		humanize_number(oldsizebuf, sizeof(oldsizebuf),
1616 		    (sblock.fs_size % sblock.fs_fpg) * sblock.fs_fsize,
1617 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1618 		warnx("no room to allocate last cylinder group; "
1619 		    "leaving %s unused", oldsizebuf);
1620 		sblock.fs_ncg--;
1621 		if (sblock.fs_magic == FS_UFS1_MAGIC)
1622 			sblock.fs_old_ncyl = sblock.fs_ncg * sblock.fs_old_cpg;
1623 		sblock.fs_size = sblock.fs_ncg * sblock.fs_fpg;
1624 	}
1625 
1626 	/*
1627 	 * Update the space for the cylinder group summary information in the
1628 	 * respective cylinder group data area.
1629 	 */
1630 	sblock.fs_cssize =
1631 	    fragroundup(&sblock, sblock.fs_ncg * sizeof(struct csum));
1632 
1633 	if (osblock.fs_size >= sblock.fs_size)
1634 		errx(1, "not enough new space");
1635 
1636 	DBG_PRINT0("sblock calculated\n");
1637 
1638 	/*
1639 	 * Ok, everything prepared, so now let's do the tricks.
1640 	 */
1641 	growfs(fsi, fso, Nflag);
1642 
1643 	close(fsi);
1644 	if (fso > -1) {
1645 		if (statfsp != NULL && (statfsp->f_flags & MNT_RDONLY) == 0) {
1646 			error = ioctl(fso, UFSRESUME);
1647 			if (error != 0)
1648 				err(1, "UFSRESUME");
1649 		}
1650 		error = close(fso);
1651 		if (error != 0)
1652 			err(1, "close");
1653 		if (statfsp != NULL && (statfsp->f_flags & MNT_RDONLY) != 0)
1654 			mount_reload(statfsp);
1655 	}
1656 
1657 	DBG_CLOSE;
1658 
1659 	DBG_LEAVE;
1660 	return (0);
1661 }
1662 
1663 /*
1664  * Dump a line of usage.
1665  */
1666 static void
1667 usage(void)
1668 {
1669 	DBG_FUNC("usage")
1670 
1671 	DBG_ENTER;
1672 
1673 	fprintf(stderr, "usage: growfs [-Ny] [-s size] special | filesystem\n");
1674 
1675 	DBG_LEAVE;
1676 	exit(1);
1677 }
1678 
1679 /*
1680  * This updates most parameters and the bitmap related to cluster. We have to
1681  * assume that sblock, osblock, acg are set up.
1682  */
1683 static void
1684 updclst(int block)
1685 {
1686 	DBG_FUNC("updclst")
1687 	static int lcs = 0;
1688 
1689 	DBG_ENTER;
1690 
1691 	if (sblock.fs_contigsumsize < 1) /* no clustering */
1692 		return;
1693 	/*
1694 	 * update cluster allocation map
1695 	 */
1696 	setbit(cg_clustersfree(&acg), block);
1697 
1698 	/*
1699 	 * update cluster summary table
1700 	 */
1701 	if (!lcs) {
1702 		/*
1703 		 * calculate size for the trailing cluster
1704 		 */
1705 		for (block--; lcs < sblock.fs_contigsumsize; block--, lcs++ ) {
1706 			if (isclr(cg_clustersfree(&acg), block))
1707 				break;
1708 		}
1709 	}
1710 	if (lcs < sblock.fs_contigsumsize) {
1711 		if (lcs)
1712 			cg_clustersum(&acg)[lcs]--;
1713 		lcs++;
1714 		cg_clustersum(&acg)[lcs]++;
1715 	}
1716 
1717 	DBG_LEAVE;
1718 	return;
1719 }
1720 
1721 static void
1722 mount_reload(const struct statfs *stfs)
1723 {
1724 	char errmsg[255];
1725 	struct iovec *iov;
1726 	int iovlen;
1727 
1728 	iov = NULL;
1729 	iovlen = 0;
1730 	*errmsg = '\0';
1731 	build_iovec(&iov, &iovlen, "fstype", __DECONST(char *, "ffs"), 4);
1732 	build_iovec(&iov, &iovlen, "fspath", __DECONST(char *, stfs->f_mntonname), (size_t)-1);
1733 	build_iovec(&iov, &iovlen, "errmsg", errmsg, sizeof(errmsg));
1734 	build_iovec(&iov, &iovlen, "update", NULL, 0);
1735 	build_iovec(&iov, &iovlen, "reload", NULL, 0);
1736 
1737 	if (nmount(iov, iovlen, stfs->f_flags) < 0) {
1738 		errmsg[sizeof(errmsg) - 1] = '\0';
1739 		err(9, "%s: cannot reload filesystem%s%s", stfs->f_mntonname,
1740 		    *errmsg != '\0' ? ": " : "", errmsg);
1741 	}
1742 }
1743 
1744 /*
1745  * Calculate the check-hash of the cylinder group.
1746  */
1747 static void
1748 cgckhash(struct cg *cgp)
1749 {
1750 
1751 	if ((sblock.fs_metackhash & CK_CYLGRP) == 0)
1752 		return;
1753 	cgp->cg_ckhash = 0;
1754 	cgp->cg_ckhash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
1755 }
1756