xref: /freebsd/sbin/growfs/growfs.c (revision 3416500aef140042c64bc149cb1ec6620483bc44)
1 /*
2  * Copyright (c) 1980, 1989, 1993 The Regents of the University of California.
3  * Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz
4  * Copyright (c) 2012 The FreeBSD Foundation
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Christoph Herrmann and Thomas-Henning von Kamptz, Munich and Frankfurt.
9  *
10  * Portions of this software were developed by Edward Tomasz Napierala
11  * under sponsorship from the FreeBSD Foundation.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. All advertising materials mentioning features or use of this software
22  *    must display the following acknowledgment:
23  *      This product includes software developed by the University of
24  *      California, Berkeley and its contributors, as well as Christoph
25  *      Herrmann and Thomas-Henning von Kamptz.
26  * 4. Neither the name of the University nor the names of its contributors
27  *    may be used to endorse or promote products derived from this software
28  *    without specific prior written permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40  * SUCH DAMAGE.
41  *
42  * $TSHeader: src/sbin/growfs/growfs.c,v 1.5 2000/12/12 19:31:00 tomsoft Exp $
43  *
44  */
45 
46 #ifndef lint
47 static const char copyright[] =
48 "@(#) Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz\n\
49 Copyright (c) 1980, 1989, 1993 The Regents of the University of California.\n\
50 All rights reserved.\n";
51 #endif /* not lint */
52 
53 #include <sys/cdefs.h>
54 __FBSDID("$FreeBSD$");
55 
56 #include <sys/param.h>
57 #include <sys/ioctl.h>
58 #include <sys/stat.h>
59 #include <sys/disk.h>
60 #include <sys/ucred.h>
61 #include <sys/mount.h>
62 
63 #include <stdio.h>
64 #include <paths.h>
65 #include <ctype.h>
66 #include <err.h>
67 #include <fcntl.h>
68 #include <fstab.h>
69 #include <inttypes.h>
70 #include <limits.h>
71 #include <mntopts.h>
72 #include <paths.h>
73 #include <stdlib.h>
74 #include <stdint.h>
75 #include <string.h>
76 #include <time.h>
77 #include <unistd.h>
78 #include <ufs/ufs/dinode.h>
79 #include <ufs/ffs/fs.h>
80 #include <libutil.h>
81 
82 #include "debug.h"
83 
84 #ifdef FS_DEBUG
85 int	_dbg_lvl_ = (DL_INFO);	/* DL_TRC */
86 #endif /* FS_DEBUG */
87 
88 static union {
89 	struct fs	fs;
90 	char		pad[SBLOCKSIZE];
91 } fsun1, fsun2;
92 #define	sblock	fsun1.fs	/* the new superblock */
93 #define	osblock	fsun2.fs	/* the old superblock */
94 
95 /*
96  * Possible superblock locations ordered from most to least likely.
97  */
98 static int sblock_try[] = SBLOCKSEARCH;
99 static ufs2_daddr_t sblockloc;
100 
101 static union {
102 	struct cg	cg;
103 	char		pad[MAXBSIZE];
104 } cgun1, cgun2;
105 #define	acg	cgun1.cg	/* a cylinder cgroup (new) */
106 #define	aocg	cgun2.cg	/* an old cylinder group */
107 
108 static struct csum	*fscs;	/* cylinder summary */
109 
110 static void	growfs(int, int, unsigned int);
111 static void	rdfs(ufs2_daddr_t, size_t, void *, int);
112 static void	wtfs(ufs2_daddr_t, size_t, void *, int, unsigned int);
113 static int	charsperline(void);
114 static void	usage(void);
115 static int	isblock(struct fs *, unsigned char *, int);
116 static void	clrblock(struct fs *, unsigned char *, int);
117 static void	setblock(struct fs *, unsigned char *, int);
118 static void	initcg(int, time_t, int, unsigned int);
119 static void	updjcg(int, time_t, int, int, unsigned int);
120 static void	updcsloc(time_t, int, int, unsigned int);
121 static void	frag_adjust(ufs2_daddr_t, int);
122 static void	updclst(int);
123 static void	mount_reload(const struct statfs *stfs);
124 
125 /*
126  * Here we actually start growing the file system. We basically read the
127  * cylinder summary from the first cylinder group as we want to update
128  * this on the fly during our various operations. First we handle the
129  * changes in the former last cylinder group. Afterwards we create all new
130  * cylinder groups.  Now we handle the cylinder group containing the
131  * cylinder summary which might result in a relocation of the whole
132  * structure.  In the end we write back the updated cylinder summary, the
133  * new superblock, and slightly patched versions of the super block
134  * copies.
135  */
136 static void
137 growfs(int fsi, int fso, unsigned int Nflag)
138 {
139 	DBG_FUNC("growfs")
140 	time_t modtime;
141 	uint cylno;
142 	int i, j, width;
143 	char tmpbuf[100];
144 
145 	DBG_ENTER;
146 
147 	time(&modtime);
148 
149 	/*
150 	 * Get the cylinder summary into the memory.
151 	 */
152 	fscs = (struct csum *)calloc((size_t)1, (size_t)sblock.fs_cssize);
153 	if (fscs == NULL)
154 		errx(1, "calloc failed");
155 	for (i = 0; i < osblock.fs_cssize; i += osblock.fs_bsize) {
156 		rdfs(fsbtodb(&osblock, osblock.fs_csaddr +
157 		    numfrags(&osblock, i)), (size_t)MIN(osblock.fs_cssize - i,
158 		    osblock.fs_bsize), (void *)(((char *)fscs) + i), fsi);
159 	}
160 
161 #ifdef FS_DEBUG
162 	{
163 		struct csum *dbg_csp;
164 		u_int32_t dbg_csc;
165 		char dbg_line[80];
166 
167 		dbg_csp = fscs;
168 
169 		for (dbg_csc = 0; dbg_csc < osblock.fs_ncg; dbg_csc++) {
170 			snprintf(dbg_line, sizeof(dbg_line),
171 			    "%d. old csum in old location", dbg_csc);
172 			DBG_DUMP_CSUM(&osblock, dbg_line, dbg_csp++);
173 		}
174 	}
175 #endif /* FS_DEBUG */
176 	DBG_PRINT0("fscs read\n");
177 
178 	/*
179 	 * Do all needed changes in the former last cylinder group.
180 	 */
181 	updjcg(osblock.fs_ncg - 1, modtime, fsi, fso, Nflag);
182 
183 	/*
184 	 * Dump out summary information about file system.
185 	 */
186 #ifdef FS_DEBUG
187 #define B2MBFACTOR (1 / (1024.0 * 1024.0))
188 	printf("growfs: %.1fMB (%jd sectors) block size %d, fragment size %d\n",
189 	    (float)sblock.fs_size * sblock.fs_fsize * B2MBFACTOR,
190 	    (intmax_t)fsbtodb(&sblock, sblock.fs_size), sblock.fs_bsize,
191 	    sblock.fs_fsize);
192 	printf("\tusing %d cylinder groups of %.2fMB, %d blks, %d inodes.\n",
193 	    sblock.fs_ncg, (float)sblock.fs_fpg * sblock.fs_fsize * B2MBFACTOR,
194 	    sblock.fs_fpg / sblock.fs_frag, sblock.fs_ipg);
195 	if (sblock.fs_flags & FS_DOSOFTDEP)
196 		printf("\twith soft updates\n");
197 #undef B2MBFACTOR
198 #endif /* FS_DEBUG */
199 
200 	/*
201 	 * Now build the cylinders group blocks and
202 	 * then print out indices of cylinder groups.
203 	 */
204 	printf("super-block backups (for fsck_ffs -b #) at:\n");
205 	i = 0;
206 	width = charsperline();
207 
208 	/*
209 	 * Iterate for only the new cylinder groups.
210 	 */
211 	for (cylno = osblock.fs_ncg; cylno < sblock.fs_ncg; cylno++) {
212 		initcg(cylno, modtime, fso, Nflag);
213 		j = sprintf(tmpbuf, " %jd%s",
214 		    (intmax_t)fsbtodb(&sblock, cgsblock(&sblock, cylno)),
215 		    cylno < (sblock.fs_ncg - 1) ? "," : "" );
216 		if (i + j >= width) {
217 			printf("\n");
218 			i = 0;
219 		}
220 		i += j;
221 		printf("%s", tmpbuf);
222 		fflush(stdout);
223 	}
224 	printf("\n");
225 
226 	/*
227 	 * Do all needed changes in the first cylinder group.
228 	 * allocate blocks in new location
229 	 */
230 	updcsloc(modtime, fsi, fso, Nflag);
231 
232 	/*
233 	 * Now write the cylinder summary back to disk.
234 	 */
235 	for (i = 0; i < sblock.fs_cssize; i += sblock.fs_bsize) {
236 		wtfs(fsbtodb(&sblock, sblock.fs_csaddr + numfrags(&sblock, i)),
237 		    (size_t)MIN(sblock.fs_cssize - i, sblock.fs_bsize),
238 		    (void *)(((char *)fscs) + i), fso, Nflag);
239 	}
240 	DBG_PRINT0("fscs written\n");
241 
242 #ifdef FS_DEBUG
243 	{
244 		struct csum	*dbg_csp;
245 		u_int32_t	dbg_csc;
246 		char	dbg_line[80];
247 
248 		dbg_csp = fscs;
249 		for (dbg_csc = 0; dbg_csc < sblock.fs_ncg; dbg_csc++) {
250 			snprintf(dbg_line, sizeof(dbg_line),
251 			    "%d. new csum in new location", dbg_csc);
252 			DBG_DUMP_CSUM(&sblock, dbg_line, dbg_csp++);
253 		}
254 	}
255 #endif /* FS_DEBUG */
256 
257 	/*
258 	 * Now write the new superblock back to disk.
259 	 */
260 	sblock.fs_time = modtime;
261 	wtfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
262 	DBG_PRINT0("sblock written\n");
263 	DBG_DUMP_FS(&sblock, "new initial sblock");
264 
265 	/*
266 	 * Clean up the dynamic fields in our superblock copies.
267 	 */
268 	sblock.fs_fmod = 0;
269 	sblock.fs_clean = 1;
270 	sblock.fs_ronly = 0;
271 	sblock.fs_cgrotor = 0;
272 	sblock.fs_state = 0;
273 	memset((void *)&sblock.fs_fsmnt, 0, sizeof(sblock.fs_fsmnt));
274 	sblock.fs_flags &= FS_DOSOFTDEP;
275 
276 	/*
277 	 * XXX
278 	 * The following fields are currently distributed from the superblock
279 	 * to the copies:
280 	 *     fs_minfree
281 	 *     fs_rotdelay
282 	 *     fs_maxcontig
283 	 *     fs_maxbpg
284 	 *     fs_minfree,
285 	 *     fs_optim
286 	 *     fs_flags regarding SOFTPDATES
287 	 *
288 	 * We probably should rather change the summary for the cylinder group
289 	 * statistics here to the value of what would be in there, if the file
290 	 * system were created initially with the new size. Therefor we still
291 	 * need to find an easy way of calculating that.
292 	 * Possibly we can try to read the first superblock copy and apply the
293 	 * "diffed" stats between the old and new superblock by still copying
294 	 * certain parameters onto that.
295 	 */
296 
297 	/*
298 	 * Write out the duplicate super blocks.
299 	 */
300 	for (cylno = 0; cylno < sblock.fs_ncg; cylno++) {
301 		wtfs(fsbtodb(&sblock, cgsblock(&sblock, cylno)),
302 		    (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
303 	}
304 	DBG_PRINT0("sblock copies written\n");
305 	DBG_DUMP_FS(&sblock, "new other sblocks");
306 
307 	DBG_LEAVE;
308 	return;
309 }
310 
311 /*
312  * This creates a new cylinder group structure, for more details please see
313  * the source of newfs(8), as this function is taken over almost unchanged.
314  * As this is never called for the first cylinder group, the special
315  * provisions for that case are removed here.
316  */
317 static void
318 initcg(int cylno, time_t modtime, int fso, unsigned int Nflag)
319 {
320 	DBG_FUNC("initcg")
321 	static caddr_t iobuf;
322 	long blkno, start;
323 	ino_t ino;
324 	ufs2_daddr_t i, cbase, dmax;
325 	struct ufs1_dinode *dp1;
326 	struct csum *cs;
327 	uint j, d, dupper, dlower;
328 
329 	if (iobuf == NULL && (iobuf = malloc(sblock.fs_bsize * 3)) == NULL)
330 		errx(37, "panic: cannot allocate I/O buffer");
331 
332 	/*
333 	 * Determine block bounds for cylinder group.
334 	 * Allow space for super block summary information in first
335 	 * cylinder group.
336 	 */
337 	cbase = cgbase(&sblock, cylno);
338 	dmax = cbase + sblock.fs_fpg;
339 	if (dmax > sblock.fs_size)
340 		dmax = sblock.fs_size;
341 	dlower = cgsblock(&sblock, cylno) - cbase;
342 	dupper = cgdmin(&sblock, cylno) - cbase;
343 	if (cylno == 0)	/* XXX fscs may be relocated */
344 		dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
345 	cs = &fscs[cylno];
346 	memset(&acg, 0, sblock.fs_cgsize);
347 	acg.cg_time = modtime;
348 	acg.cg_magic = CG_MAGIC;
349 	acg.cg_cgx = cylno;
350 	acg.cg_niblk = sblock.fs_ipg;
351 	acg.cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
352 	acg.cg_ndblk = dmax - cbase;
353 	if (sblock.fs_contigsumsize > 0)
354 		acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
355 	start = &acg.cg_space[0] - (u_char *)(&acg.cg_firstfield);
356 	if (sblock.fs_magic == FS_UFS2_MAGIC) {
357 		acg.cg_iusedoff = start;
358 	} else {
359 		acg.cg_old_ncyl = sblock.fs_old_cpg;
360 		acg.cg_old_time = acg.cg_time;
361 		acg.cg_time = 0;
362 		acg.cg_old_niblk = acg.cg_niblk;
363 		acg.cg_niblk = 0;
364 		acg.cg_initediblk = 0;
365 		acg.cg_old_btotoff = start;
366 		acg.cg_old_boff = acg.cg_old_btotoff +
367 		    sblock.fs_old_cpg * sizeof(int32_t);
368 		acg.cg_iusedoff = acg.cg_old_boff +
369 		    sblock.fs_old_cpg * sizeof(u_int16_t);
370 	}
371 	acg.cg_freeoff = acg.cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
372 	acg.cg_nextfreeoff = acg.cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT);
373 	if (sblock.fs_contigsumsize > 0) {
374 		acg.cg_clustersumoff =
375 		    roundup(acg.cg_nextfreeoff, sizeof(u_int32_t));
376 		acg.cg_clustersumoff -= sizeof(u_int32_t);
377 		acg.cg_clusteroff = acg.cg_clustersumoff +
378 		    (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
379 		acg.cg_nextfreeoff = acg.cg_clusteroff +
380 		    howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
381 	}
382 	if (acg.cg_nextfreeoff > (unsigned)sblock.fs_cgsize) {
383 		/*
384 		 * This should never happen as we would have had that panic
385 		 * already on file system creation
386 		 */
387 		errx(37, "panic: cylinder group too big");
388 	}
389 	acg.cg_cs.cs_nifree += sblock.fs_ipg;
390 	if (cylno == 0)
391 		for (ino = 0; ino < UFS_ROOTINO; ino++) {
392 			setbit(cg_inosused(&acg), ino);
393 			acg.cg_cs.cs_nifree--;
394 		}
395 	/*
396 	 * For the old file system, we have to initialize all the inodes.
397 	 */
398 	if (sblock.fs_magic == FS_UFS1_MAGIC) {
399 		bzero(iobuf, sblock.fs_bsize);
400 		for (i = 0; i < sblock.fs_ipg / INOPF(&sblock);
401 		    i += sblock.fs_frag) {
402 			dp1 = (struct ufs1_dinode *)(void *)iobuf;
403 			for (j = 0; j < INOPB(&sblock); j++) {
404 				dp1->di_gen = arc4random();
405 				dp1++;
406 			}
407 			wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno) + i),
408 			    sblock.fs_bsize, iobuf, fso, Nflag);
409 		}
410 	}
411 	if (cylno > 0) {
412 		/*
413 		 * In cylno 0, beginning space is reserved
414 		 * for boot and super blocks.
415 		 */
416 		for (d = 0; d < dlower; d += sblock.fs_frag) {
417 			blkno = d / sblock.fs_frag;
418 			setblock(&sblock, cg_blksfree(&acg), blkno);
419 			if (sblock.fs_contigsumsize > 0)
420 				setbit(cg_clustersfree(&acg), blkno);
421 			acg.cg_cs.cs_nbfree++;
422 		}
423 		sblock.fs_dsize += dlower;
424 	}
425 	sblock.fs_dsize += acg.cg_ndblk - dupper;
426 	if ((i = dupper % sblock.fs_frag)) {
427 		acg.cg_frsum[sblock.fs_frag - i]++;
428 		for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
429 			setbit(cg_blksfree(&acg), dupper);
430 			acg.cg_cs.cs_nffree++;
431 		}
432 	}
433 	for (d = dupper; d + sblock.fs_frag <= acg.cg_ndblk;
434 	    d += sblock.fs_frag) {
435 		blkno = d / sblock.fs_frag;
436 		setblock(&sblock, cg_blksfree(&acg), blkno);
437 		if (sblock.fs_contigsumsize > 0)
438 			setbit(cg_clustersfree(&acg), blkno);
439 		acg.cg_cs.cs_nbfree++;
440 	}
441 	if (d < acg.cg_ndblk) {
442 		acg.cg_frsum[acg.cg_ndblk - d]++;
443 		for (; d < acg.cg_ndblk; d++) {
444 			setbit(cg_blksfree(&acg), d);
445 			acg.cg_cs.cs_nffree++;
446 		}
447 	}
448 	if (sblock.fs_contigsumsize > 0) {
449 		int32_t *sump = cg_clustersum(&acg);
450 		u_char *mapp = cg_clustersfree(&acg);
451 		int map = *mapp++;
452 		int bit = 1;
453 		int run = 0;
454 
455 		for (i = 0; i < acg.cg_nclusterblks; i++) {
456 			if ((map & bit) != 0)
457 				run++;
458 			else if (run != 0) {
459 				if (run > sblock.fs_contigsumsize)
460 					run = sblock.fs_contigsumsize;
461 				sump[run]++;
462 				run = 0;
463 			}
464 			if ((i & (CHAR_BIT - 1)) != CHAR_BIT - 1)
465 				bit <<= 1;
466 			else {
467 				map = *mapp++;
468 				bit = 1;
469 			}
470 		}
471 		if (run != 0) {
472 			if (run > sblock.fs_contigsumsize)
473 				run = sblock.fs_contigsumsize;
474 			sump[run]++;
475 		}
476 	}
477 	sblock.fs_cstotal.cs_ndir += acg.cg_cs.cs_ndir;
478 	sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
479 	sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
480 	sblock.fs_cstotal.cs_nifree += acg.cg_cs.cs_nifree;
481 	*cs = acg.cg_cs;
482 
483 	memcpy(iobuf, &acg, sblock.fs_cgsize);
484 	memset(iobuf + sblock.fs_cgsize, '\0',
485 	    sblock.fs_bsize * 3 - sblock.fs_cgsize);
486 
487 	wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
488 	    sblock.fs_bsize * 3, iobuf, fso, Nflag);
489 	DBG_DUMP_CG(&sblock, "new cg", &acg);
490 
491 	DBG_LEAVE;
492 	return;
493 }
494 
495 /*
496  * Here we add or subtract (sign +1/-1) the available fragments in a given
497  * block to or from the fragment statistics. By subtracting before and adding
498  * after an operation on the free frag map we can easy update the fragment
499  * statistic, which seems to be otherwise a rather complex operation.
500  */
501 static void
502 frag_adjust(ufs2_daddr_t frag, int sign)
503 {
504 	DBG_FUNC("frag_adjust")
505 	int fragsize;
506 	int f;
507 
508 	DBG_ENTER;
509 
510 	fragsize = 0;
511 	/*
512 	 * Here frag only needs to point to any fragment in the block we want
513 	 * to examine.
514 	 */
515 	for (f = rounddown(frag, sblock.fs_frag);
516 	    f < roundup(frag + 1, sblock.fs_frag); f++) {
517 		/*
518 		 * Count contiguous free fragments.
519 		 */
520 		if (isset(cg_blksfree(&acg), f)) {
521 			fragsize++;
522 		} else {
523 			if (fragsize && fragsize < sblock.fs_frag) {
524 				/*
525 				 * We found something in between.
526 				 */
527 				acg.cg_frsum[fragsize] += sign;
528 				DBG_PRINT2("frag_adjust [%d]+=%d\n",
529 				    fragsize, sign);
530 			}
531 			fragsize = 0;
532 		}
533 	}
534 	if (fragsize && fragsize < sblock.fs_frag) {
535 		/*
536 		 * We found something.
537 		 */
538 		acg.cg_frsum[fragsize] += sign;
539 		DBG_PRINT2("frag_adjust [%d]+=%d\n", fragsize, sign);
540 	}
541 	DBG_PRINT2("frag_adjust [[%d]]+=%d\n", fragsize, sign);
542 
543 	DBG_LEAVE;
544 	return;
545 }
546 
547 /*
548  * Here we do all needed work for the former last cylinder group. It has to be
549  * changed in any case, even if the file system ended exactly on the end of
550  * this group, as there is some slightly inconsistent handling of the number
551  * of cylinders in the cylinder group. We start again by reading the cylinder
552  * group from disk. If the last block was not fully available, we first handle
553  * the missing fragments, then we handle all new full blocks in that file
554  * system and finally we handle the new last fragmented block in the file
555  * system.  We again have to handle the fragment statistics rotational layout
556  * tables and cluster summary during all those operations.
557  */
558 static void
559 updjcg(int cylno, time_t modtime, int fsi, int fso, unsigned int Nflag)
560 {
561 	DBG_FUNC("updjcg")
562 	ufs2_daddr_t cbase, dmax, dupper;
563 	struct csum *cs;
564 	int i, k;
565 	int j = 0;
566 
567 	DBG_ENTER;
568 
569 	/*
570 	 * Read the former last (joining) cylinder group from disk, and make
571 	 * a copy.
572 	 */
573 	rdfs(fsbtodb(&osblock, cgtod(&osblock, cylno)),
574 	    (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
575 	DBG_PRINT0("jcg read\n");
576 	DBG_DUMP_CG(&sblock, "old joining cg", &aocg);
577 
578 	memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
579 
580 	/*
581 	 * If the cylinder group had already its new final size almost
582 	 * nothing is to be done ... except:
583 	 * For some reason the value of cg_ncyl in the last cylinder group has
584 	 * to be zero instead of fs_cpg. As this is now no longer the last
585 	 * cylinder group we have to change that value now to fs_cpg.
586 	 */
587 
588 	if (cgbase(&osblock, cylno + 1) == osblock.fs_size) {
589 		if (sblock.fs_magic == FS_UFS1_MAGIC)
590 			acg.cg_old_ncyl = sblock.fs_old_cpg;
591 
592 		wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
593 		    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
594 		DBG_PRINT0("jcg written\n");
595 		DBG_DUMP_CG(&sblock, "new joining cg", &acg);
596 
597 		DBG_LEAVE;
598 		return;
599 	}
600 
601 	/*
602 	 * Set up some variables needed later.
603 	 */
604 	cbase = cgbase(&sblock, cylno);
605 	dmax = cbase + sblock.fs_fpg;
606 	if (dmax > sblock.fs_size)
607 		dmax = sblock.fs_size;
608 	dupper = cgdmin(&sblock, cylno) - cbase;
609 	if (cylno == 0) /* XXX fscs may be relocated */
610 		dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
611 
612 	/*
613 	 * Set pointer to the cylinder summary for our cylinder group.
614 	 */
615 	cs = fscs + cylno;
616 
617 	/*
618 	 * Touch the cylinder group, update all fields in the cylinder group as
619 	 * needed, update the free space in the superblock.
620 	 */
621 	acg.cg_time = modtime;
622 	if ((unsigned)cylno == sblock.fs_ncg - 1) {
623 		/*
624 		 * This is still the last cylinder group.
625 		 */
626 		if (sblock.fs_magic == FS_UFS1_MAGIC)
627 			acg.cg_old_ncyl =
628 			    sblock.fs_old_ncyl % sblock.fs_old_cpg;
629 	} else {
630 		acg.cg_old_ncyl = sblock.fs_old_cpg;
631 	}
632 	DBG_PRINT2("jcg dbg: %d %u", cylno, sblock.fs_ncg);
633 #ifdef FS_DEBUG
634 	if (sblock.fs_magic == FS_UFS1_MAGIC)
635 		DBG_PRINT2("%d %u", acg.cg_old_ncyl, sblock.fs_old_cpg);
636 #endif
637 	DBG_PRINT0("\n");
638 	acg.cg_ndblk = dmax - cbase;
639 	sblock.fs_dsize += acg.cg_ndblk - aocg.cg_ndblk;
640 	if (sblock.fs_contigsumsize > 0)
641 		acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
642 
643 	/*
644 	 * Now we have to update the free fragment bitmap for our new free
645 	 * space.  There again we have to handle the fragmentation and also
646 	 * the rotational layout tables and the cluster summary.  This is
647 	 * also done per fragment for the first new block if the old file
648 	 * system end was not on a block boundary, per fragment for the new
649 	 * last block if the new file system end is not on a block boundary,
650 	 * and per block for all space in between.
651 	 *
652 	 * Handle the first new block here if it was partially available
653 	 * before.
654 	 */
655 	if (osblock.fs_size % sblock.fs_frag) {
656 		if (roundup(osblock.fs_size, sblock.fs_frag) <=
657 		    sblock.fs_size) {
658 			/*
659 			 * The new space is enough to fill at least this
660 			 * block
661 			 */
662 			j = 0;
663 			for (i = roundup(osblock.fs_size - cbase,
664 			    sblock.fs_frag) - 1; i >= osblock.fs_size - cbase;
665 			    i--) {
666 				setbit(cg_blksfree(&acg), i);
667 				acg.cg_cs.cs_nffree++;
668 				j++;
669 			}
670 
671 			/*
672 			 * Check if the fragment just created could join an
673 			 * already existing fragment at the former end of the
674 			 * file system.
675 			 */
676 			if (isblock(&sblock, cg_blksfree(&acg),
677 			    ((osblock.fs_size - cgbase(&sblock, cylno)) /
678 			     sblock.fs_frag))) {
679 				/*
680 				 * The block is now completely available.
681 				 */
682 				DBG_PRINT0("block was\n");
683 				acg.cg_frsum[osblock.fs_size % sblock.fs_frag]--;
684 				acg.cg_cs.cs_nbfree++;
685 				acg.cg_cs.cs_nffree -= sblock.fs_frag;
686 				k = rounddown(osblock.fs_size - cbase,
687 				    sblock.fs_frag);
688 				updclst((osblock.fs_size - cbase) /
689 				    sblock.fs_frag);
690 			} else {
691 				/*
692 				 * Lets rejoin a possible partially growed
693 				 * fragment.
694 				 */
695 				k = 0;
696 				while (isset(cg_blksfree(&acg), i) &&
697 				    (i >= rounddown(osblock.fs_size - cbase,
698 				    sblock.fs_frag))) {
699 					i--;
700 					k++;
701 				}
702 				if (k)
703 					acg.cg_frsum[k]--;
704 				acg.cg_frsum[k + j]++;
705 			}
706 		} else {
707 			/*
708 			 * We only grow by some fragments within this last
709 			 * block.
710 			 */
711 			for (i = sblock.fs_size - cbase - 1;
712 			    i >= osblock.fs_size - cbase; i--) {
713 				setbit(cg_blksfree(&acg), i);
714 				acg.cg_cs.cs_nffree++;
715 				j++;
716 			}
717 			/*
718 			 * Lets rejoin a possible partially growed fragment.
719 			 */
720 			k = 0;
721 			while (isset(cg_blksfree(&acg), i) &&
722 			    (i >= rounddown(osblock.fs_size - cbase,
723 			    sblock.fs_frag))) {
724 				i--;
725 				k++;
726 			}
727 			if (k)
728 				acg.cg_frsum[k]--;
729 			acg.cg_frsum[k + j]++;
730 		}
731 	}
732 
733 	/*
734 	 * Handle all new complete blocks here.
735 	 */
736 	for (i = roundup(osblock.fs_size - cbase, sblock.fs_frag);
737 	    i + sblock.fs_frag <= dmax - cbase;	/* XXX <= or only < ? */
738 	    i += sblock.fs_frag) {
739 		j = i / sblock.fs_frag;
740 		setblock(&sblock, cg_blksfree(&acg), j);
741 		updclst(j);
742 		acg.cg_cs.cs_nbfree++;
743 	}
744 
745 	/*
746 	 * Handle the last new block if there are stll some new fragments left.
747 	 * Here we don't have to bother about the cluster summary or the even
748 	 * the rotational layout table.
749 	 */
750 	if (i < (dmax - cbase)) {
751 		acg.cg_frsum[dmax - cbase - i]++;
752 		for (; i < dmax - cbase; i++) {
753 			setbit(cg_blksfree(&acg), i);
754 			acg.cg_cs.cs_nffree++;
755 		}
756 	}
757 
758 	sblock.fs_cstotal.cs_nffree +=
759 	    (acg.cg_cs.cs_nffree - aocg.cg_cs.cs_nffree);
760 	sblock.fs_cstotal.cs_nbfree +=
761 	    (acg.cg_cs.cs_nbfree - aocg.cg_cs.cs_nbfree);
762 	/*
763 	 * The following statistics are not changed here:
764 	 *     sblock.fs_cstotal.cs_ndir
765 	 *     sblock.fs_cstotal.cs_nifree
766 	 * As the statistics for this cylinder group are ready, copy it to
767 	 * the summary information array.
768 	 */
769 	*cs = acg.cg_cs;
770 
771 	/*
772 	 * Write the updated "joining" cylinder group back to disk.
773 	 */
774 	wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize,
775 	    (void *)&acg, fso, Nflag);
776 	DBG_PRINT0("jcg written\n");
777 	DBG_DUMP_CG(&sblock, "new joining cg", &acg);
778 
779 	DBG_LEAVE;
780 	return;
781 }
782 
783 /*
784  * Here we update the location of the cylinder summary. We have two possible
785  * ways of growing the cylinder summary:
786  * (1)	We can try to grow the summary in the current location, and relocate
787  *	possibly used blocks within the current cylinder group.
788  * (2)	Alternatively we can relocate the whole cylinder summary to the first
789  *	new completely empty cylinder group. Once the cylinder summary is no
790  *	longer in the beginning of the first cylinder group you should never
791  *	use a version of fsck which is not aware of the possibility to have
792  *	this structure in a non standard place.
793  * Option (2) is considered to be less intrusive to the structure of the file-
794  * system, so that's the one being used.
795  */
796 static void
797 updcsloc(time_t modtime, int fsi, int fso, unsigned int Nflag)
798 {
799 	DBG_FUNC("updcsloc")
800 	struct csum *cs;
801 	int ocscg, ncscg;
802 	ufs2_daddr_t d;
803 	int lcs = 0;
804 	int block;
805 
806 	DBG_ENTER;
807 
808 	if (howmany(sblock.fs_cssize, sblock.fs_fsize) ==
809 	    howmany(osblock.fs_cssize, osblock.fs_fsize)) {
810 		/*
811 		 * No new fragment needed.
812 		 */
813 		DBG_LEAVE;
814 		return;
815 	}
816 	ocscg = dtog(&osblock, osblock.fs_csaddr);
817 	cs = fscs + ocscg;
818 
819 	/*
820 	 * Read original cylinder group from disk, and make a copy.
821 	 * XXX	If Nflag is set in some very rare cases we now miss
822 	 *	some changes done in updjcg by reading the unmodified
823 	 *	block from disk.
824 	 */
825 	rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)),
826 	    (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
827 	DBG_PRINT0("oscg read\n");
828 	DBG_DUMP_CG(&sblock, "old summary cg", &aocg);
829 
830 	memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
831 
832 	/*
833 	 * Touch the cylinder group, set up local variables needed later
834 	 * and update the superblock.
835 	 */
836 	acg.cg_time = modtime;
837 
838 	/*
839 	 * XXX	In the case of having active snapshots we may need much more
840 	 *	blocks for the copy on write. We need each block twice, and
841 	 *	also up to 8*3 blocks for indirect blocks for all possible
842 	 *	references.
843 	 */
844 	/*
845 	 * There is not enough space in the old cylinder group to
846 	 * relocate all blocks as needed, so we relocate the whole
847 	 * cylinder group summary to a new group. We try to use the
848 	 * first complete new cylinder group just created. Within the
849 	 * cylinder group we align the area immediately after the
850 	 * cylinder group information location in order to be as
851 	 * close as possible to the original implementation of ffs.
852 	 *
853 	 * First we have to make sure we'll find enough space in the
854 	 * new cylinder group. If not, then we currently give up.
855 	 * We start with freeing everything which was used by the
856 	 * fragments of the old cylinder summary in the current group.
857 	 * Now we write back the group meta data, read in the needed
858 	 * meta data from the new cylinder group, and start allocating
859 	 * within that group. Here we can assume, the group to be
860 	 * completely empty. Which makes the handling of fragments and
861 	 * clusters a lot easier.
862 	 */
863 	DBG_TRC;
864 	if (sblock.fs_ncg - osblock.fs_ncg < 2)
865 		errx(2, "panic: not enough space");
866 
867 	/*
868 	 * Point "d" to the first fragment not used by the cylinder
869 	 * summary.
870 	 */
871 	d = osblock.fs_csaddr + (osblock.fs_cssize / osblock.fs_fsize);
872 
873 	/*
874 	 * Set up last cluster size ("lcs") already here. Calculate
875 	 * the size for the trailing cluster just behind where "d"
876 	 * points to.
877 	 */
878 	if (sblock.fs_contigsumsize > 0) {
879 		for (block = howmany(d % sblock.fs_fpg, sblock.fs_frag),
880 		    lcs = 0; lcs < sblock.fs_contigsumsize; block++, lcs++) {
881 			if (isclr(cg_clustersfree(&acg), block))
882 				break;
883 		}
884 	}
885 
886 	/*
887 	 * Point "d" to the last frag used by the cylinder summary.
888 	 */
889 	d--;
890 
891 	DBG_PRINT1("d=%jd\n", (intmax_t)d);
892 	if ((d + 1) % sblock.fs_frag) {
893 		/*
894 		 * The end of the cylinder summary is not a complete
895 		 * block.
896 		 */
897 		DBG_TRC;
898 		frag_adjust(d % sblock.fs_fpg, -1);
899 		for (; (d + 1) % sblock.fs_frag; d--) {
900 			DBG_PRINT1("d=%jd\n", (intmax_t)d);
901 			setbit(cg_blksfree(&acg), d % sblock.fs_fpg);
902 			acg.cg_cs.cs_nffree++;
903 			sblock.fs_cstotal.cs_nffree++;
904 		}
905 		/*
906 		 * Point "d" to the last fragment of the last
907 		 * (incomplete) block of the cylinder summary.
908 		 */
909 		d++;
910 		frag_adjust(d % sblock.fs_fpg, 1);
911 
912 		if (isblock(&sblock, cg_blksfree(&acg),
913 		    (d % sblock.fs_fpg) / sblock.fs_frag)) {
914 			DBG_PRINT1("d=%jd\n", (intmax_t)d);
915 			acg.cg_cs.cs_nffree -= sblock.fs_frag;
916 			acg.cg_cs.cs_nbfree++;
917 			sblock.fs_cstotal.cs_nffree -= sblock.fs_frag;
918 			sblock.fs_cstotal.cs_nbfree++;
919 			if (sblock.fs_contigsumsize > 0) {
920 				setbit(cg_clustersfree(&acg),
921 				    (d % sblock.fs_fpg) / sblock.fs_frag);
922 				if (lcs < sblock.fs_contigsumsize) {
923 					if (lcs)
924 						cg_clustersum(&acg)[lcs]--;
925 					lcs++;
926 					cg_clustersum(&acg)[lcs]++;
927 				}
928 			}
929 		}
930 		/*
931 		 * Point "d" to the first fragment of the block before
932 		 * the last incomplete block.
933 		 */
934 		d--;
935 	}
936 
937 	DBG_PRINT1("d=%jd\n", (intmax_t)d);
938 	for (d = rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr;
939 	    d -= sblock.fs_frag) {
940 		DBG_TRC;
941 		DBG_PRINT1("d=%jd\n", (intmax_t)d);
942 		setblock(&sblock, cg_blksfree(&acg),
943 		    (d % sblock.fs_fpg) / sblock.fs_frag);
944 		acg.cg_cs.cs_nbfree++;
945 		sblock.fs_cstotal.cs_nbfree++;
946 		if (sblock.fs_contigsumsize > 0) {
947 			setbit(cg_clustersfree(&acg),
948 			    (d % sblock.fs_fpg) / sblock.fs_frag);
949 			/*
950 			 * The last cluster size is already set up.
951 			 */
952 			if (lcs < sblock.fs_contigsumsize) {
953 				if (lcs)
954 					cg_clustersum(&acg)[lcs]--;
955 				lcs++;
956 				cg_clustersum(&acg)[lcs]++;
957 			}
958 		}
959 	}
960 	*cs = acg.cg_cs;
961 
962 	/*
963 	 * Now write the former cylinder group containing the cylinder
964 	 * summary back to disk.
965 	 */
966 	wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)),
967 	    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
968 	DBG_PRINT0("oscg written\n");
969 	DBG_DUMP_CG(&sblock, "old summary cg", &acg);
970 
971 	/*
972 	 * Find the beginning of the new cylinder group containing the
973 	 * cylinder summary.
974 	 */
975 	sblock.fs_csaddr = cgdmin(&sblock, osblock.fs_ncg);
976 	ncscg = dtog(&sblock, sblock.fs_csaddr);
977 	cs = fscs + ncscg;
978 
979 	/*
980 	 * If Nflag is specified, we would now read random data instead
981 	 * of an empty cg structure from disk. So we can't simulate that
982 	 * part for now.
983 	 */
984 	if (Nflag) {
985 		DBG_PRINT0("nscg update skipped\n");
986 		DBG_LEAVE;
987 		return;
988 	}
989 
990 	/*
991 	 * Read the future cylinder group containing the cylinder
992 	 * summary from disk, and make a copy.
993 	 */
994 	rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
995 	    (size_t)sblock.fs_cgsize, (void *)&aocg, fsi);
996 	DBG_PRINT0("nscg read\n");
997 	DBG_DUMP_CG(&sblock, "new summary cg", &aocg);
998 
999 	memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
1000 
1001 	/*
1002 	 * Allocate all complete blocks used by the new cylinder
1003 	 * summary.
1004 	 */
1005 	for (d = sblock.fs_csaddr; d + sblock.fs_frag <=
1006 	    sblock.fs_csaddr + (sblock.fs_cssize / sblock.fs_fsize);
1007 	    d += sblock.fs_frag) {
1008 		clrblock(&sblock, cg_blksfree(&acg),
1009 		    (d % sblock.fs_fpg) / sblock.fs_frag);
1010 		acg.cg_cs.cs_nbfree--;
1011 		sblock.fs_cstotal.cs_nbfree--;
1012 		if (sblock.fs_contigsumsize > 0) {
1013 			clrbit(cg_clustersfree(&acg),
1014 			    (d % sblock.fs_fpg) / sblock.fs_frag);
1015 		}
1016 	}
1017 
1018 	/*
1019 	 * Allocate all fragments used by the cylinder summary in the
1020 	 * last block.
1021 	 */
1022 	if (d < sblock.fs_csaddr + (sblock.fs_cssize / sblock.fs_fsize)) {
1023 		for (; d - sblock.fs_csaddr <
1024 		    sblock.fs_cssize/sblock.fs_fsize; d++) {
1025 			clrbit(cg_blksfree(&acg), d % sblock.fs_fpg);
1026 			acg.cg_cs.cs_nffree--;
1027 			sblock.fs_cstotal.cs_nffree--;
1028 		}
1029 		acg.cg_cs.cs_nbfree--;
1030 		acg.cg_cs.cs_nffree += sblock.fs_frag;
1031 		sblock.fs_cstotal.cs_nbfree--;
1032 		sblock.fs_cstotal.cs_nffree += sblock.fs_frag;
1033 		if (sblock.fs_contigsumsize > 0)
1034 			clrbit(cg_clustersfree(&acg),
1035 			    (d % sblock.fs_fpg) / sblock.fs_frag);
1036 
1037 		frag_adjust(d % sblock.fs_fpg, 1);
1038 	}
1039 	/*
1040 	 * XXX	Handle the cluster statistics here in the case this
1041 	 *	cylinder group is now almost full, and the remaining
1042 	 *	space is less then the maximum cluster size. This is
1043 	 *	probably not needed, as you would hardly find a file
1044 	 *	system which has only MAXCSBUFS+FS_MAXCONTIG of free
1045 	 *	space right behind the cylinder group information in
1046 	 *	any new cylinder group.
1047 	 */
1048 
1049 	/*
1050 	 * Update our statistics in the cylinder summary.
1051 	 */
1052 	*cs = acg.cg_cs;
1053 
1054 	/*
1055 	 * Write the new cylinder group containing the cylinder summary
1056 	 * back to disk.
1057 	 */
1058 	wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1059 	    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1060 	DBG_PRINT0("nscg written\n");
1061 	DBG_DUMP_CG(&sblock, "new summary cg", &acg);
1062 
1063 	DBG_LEAVE;
1064 	return;
1065 }
1066 
1067 /*
1068  * Here we read some block(s) from disk.
1069  */
1070 static void
1071 rdfs(ufs2_daddr_t bno, size_t size, void *bf, int fsi)
1072 {
1073 	DBG_FUNC("rdfs")
1074 	ssize_t	n;
1075 
1076 	DBG_ENTER;
1077 
1078 	if (bno < 0)
1079 		err(32, "rdfs: attempting to read negative block number");
1080 	if (lseek(fsi, (off_t)bno * DEV_BSIZE, 0) < 0)
1081 		err(33, "rdfs: seek error: %jd", (intmax_t)bno);
1082 	n = read(fsi, bf, size);
1083 	if (n != (ssize_t)size)
1084 		err(34, "rdfs: read error: %jd", (intmax_t)bno);
1085 
1086 	DBG_LEAVE;
1087 	return;
1088 }
1089 
1090 /*
1091  * Here we write some block(s) to disk.
1092  */
1093 static void
1094 wtfs(ufs2_daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag)
1095 {
1096 	DBG_FUNC("wtfs")
1097 	ssize_t	n;
1098 
1099 	DBG_ENTER;
1100 
1101 	if (Nflag) {
1102 		DBG_LEAVE;
1103 		return;
1104 	}
1105 	if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0)
1106 		err(35, "wtfs: seek error: %ld", (long)bno);
1107 	n = write(fso, bf, size);
1108 	if (n != (ssize_t)size)
1109 		err(36, "wtfs: write error: %ld", (long)bno);
1110 
1111 	DBG_LEAVE;
1112 	return;
1113 }
1114 
1115 /*
1116  * Here we check if all frags of a block are free. For more details again
1117  * please see the source of newfs(8), as this function is taken over almost
1118  * unchanged.
1119  */
1120 static int
1121 isblock(struct fs *fs, unsigned char *cp, int h)
1122 {
1123 	DBG_FUNC("isblock")
1124 	unsigned char mask;
1125 
1126 	DBG_ENTER;
1127 
1128 	switch (fs->fs_frag) {
1129 	case 8:
1130 		DBG_LEAVE;
1131 		return (cp[h] == 0xff);
1132 	case 4:
1133 		mask = 0x0f << ((h & 0x1) << 2);
1134 		DBG_LEAVE;
1135 		return ((cp[h >> 1] & mask) == mask);
1136 	case 2:
1137 		mask = 0x03 << ((h & 0x3) << 1);
1138 		DBG_LEAVE;
1139 		return ((cp[h >> 2] & mask) == mask);
1140 	case 1:
1141 		mask = 0x01 << (h & 0x7);
1142 		DBG_LEAVE;
1143 		return ((cp[h >> 3] & mask) == mask);
1144 	default:
1145 		fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
1146 		DBG_LEAVE;
1147 		return (0);
1148 	}
1149 }
1150 
1151 /*
1152  * Here we allocate a complete block in the block map. For more details again
1153  * please see the source of newfs(8), as this function is taken over almost
1154  * unchanged.
1155  */
1156 static void
1157 clrblock(struct fs *fs, unsigned char *cp, int h)
1158 {
1159 	DBG_FUNC("clrblock")
1160 
1161 	DBG_ENTER;
1162 
1163 	switch ((fs)->fs_frag) {
1164 	case 8:
1165 		cp[h] = 0;
1166 		break;
1167 	case 4:
1168 		cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1169 		break;
1170 	case 2:
1171 		cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1172 		break;
1173 	case 1:
1174 		cp[h >> 3] &= ~(0x01 << (h & 0x7));
1175 		break;
1176 	default:
1177 		warnx("clrblock bad fs_frag %d", fs->fs_frag);
1178 		break;
1179 	}
1180 
1181 	DBG_LEAVE;
1182 	return;
1183 }
1184 
1185 /*
1186  * Here we free a complete block in the free block map. For more details again
1187  * please see the source of newfs(8), as this function is taken over almost
1188  * unchanged.
1189  */
1190 static void
1191 setblock(struct fs *fs, unsigned char *cp, int h)
1192 {
1193 	DBG_FUNC("setblock")
1194 
1195 	DBG_ENTER;
1196 
1197 	switch (fs->fs_frag) {
1198 	case 8:
1199 		cp[h] = 0xff;
1200 		break;
1201 	case 4:
1202 		cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1203 		break;
1204 	case 2:
1205 		cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1206 		break;
1207 	case 1:
1208 		cp[h >> 3] |= (0x01 << (h & 0x7));
1209 		break;
1210 	default:
1211 		warnx("setblock bad fs_frag %d", fs->fs_frag);
1212 		break;
1213 	}
1214 
1215 	DBG_LEAVE;
1216 	return;
1217 }
1218 
1219 /*
1220  * Figure out how many lines our current terminal has. For more details again
1221  * please see the source of newfs(8), as this function is taken over almost
1222  * unchanged.
1223  */
1224 static int
1225 charsperline(void)
1226 {
1227 	DBG_FUNC("charsperline")
1228 	int columns;
1229 	char *cp;
1230 	struct winsize ws;
1231 
1232 	DBG_ENTER;
1233 
1234 	columns = 0;
1235 	if (ioctl(0, TIOCGWINSZ, &ws) != -1)
1236 		columns = ws.ws_col;
1237 	if (columns == 0 && (cp = getenv("COLUMNS")))
1238 		columns = atoi(cp);
1239 	if (columns == 0)
1240 		columns = 80;	/* last resort */
1241 
1242 	DBG_LEAVE;
1243 	return (columns);
1244 }
1245 
1246 static int
1247 is_dev(const char *name)
1248 {
1249 	struct stat devstat;
1250 
1251 	if (stat(name, &devstat) != 0)
1252 		return (0);
1253 	if (!S_ISCHR(devstat.st_mode))
1254 		return (0);
1255 	return (1);
1256 }
1257 
1258 /*
1259  * Return mountpoint on which the device is currently mounted.
1260  */
1261 static const struct statfs *
1262 dev_to_statfs(const char *dev)
1263 {
1264 	struct stat devstat, mntdevstat;
1265 	struct statfs *mntbuf, *statfsp;
1266 	char device[MAXPATHLEN];
1267 	char *mntdevname;
1268 	int i, mntsize;
1269 
1270 	/*
1271 	 * First check the mounted filesystems.
1272 	 */
1273 	if (stat(dev, &devstat) != 0)
1274 		return (NULL);
1275 	if (!S_ISCHR(devstat.st_mode) && !S_ISBLK(devstat.st_mode))
1276 		return (NULL);
1277 
1278 	mntsize = getmntinfo(&mntbuf, MNT_NOWAIT);
1279 	for (i = 0; i < mntsize; i++) {
1280 		statfsp = &mntbuf[i];
1281 		mntdevname = statfsp->f_mntfromname;
1282 		if (*mntdevname != '/') {
1283 			strcpy(device, _PATH_DEV);
1284 			strcat(device, mntdevname);
1285 			mntdevname = device;
1286 		}
1287 		if (stat(mntdevname, &mntdevstat) == 0 &&
1288 		    mntdevstat.st_rdev == devstat.st_rdev)
1289 			return (statfsp);
1290 	}
1291 
1292 	return (NULL);
1293 }
1294 
1295 static const char *
1296 mountpoint_to_dev(const char *mountpoint)
1297 {
1298 	struct statfs *mntbuf, *statfsp;
1299 	struct fstab *fs;
1300 	int i, mntsize;
1301 
1302 	/*
1303 	 * First check the mounted filesystems.
1304 	 */
1305 	mntsize = getmntinfo(&mntbuf, MNT_NOWAIT);
1306 	for (i = 0; i < mntsize; i++) {
1307 		statfsp = &mntbuf[i];
1308 
1309 		if (strcmp(statfsp->f_mntonname, mountpoint) == 0)
1310 			return (statfsp->f_mntfromname);
1311 	}
1312 
1313 	/*
1314 	 * Check the fstab.
1315 	 */
1316 	fs = getfsfile(mountpoint);
1317 	if (fs != NULL)
1318 		return (fs->fs_spec);
1319 
1320 	return (NULL);
1321 }
1322 
1323 static const char *
1324 getdev(const char *name)
1325 {
1326 	static char device[MAXPATHLEN];
1327 	const char *cp, *dev;
1328 
1329 	if (is_dev(name))
1330 		return (name);
1331 
1332 	cp = strrchr(name, '/');
1333 	if (cp == NULL) {
1334 		snprintf(device, sizeof(device), "%s%s", _PATH_DEV, name);
1335 		if (is_dev(device))
1336 			return (device);
1337 	}
1338 
1339 	dev = mountpoint_to_dev(name);
1340 	if (dev != NULL && is_dev(dev))
1341 		return (dev);
1342 
1343 	return (NULL);
1344 }
1345 
1346 /*
1347  * growfs(8) is a utility which allows to increase the size of an existing
1348  * ufs file system. Currently this can only be done on unmounted file system.
1349  * It recognizes some command line options to specify the new desired size,
1350  * and it does some basic checkings. The old file system size is determined
1351  * and after some more checks like we can really access the new last block
1352  * on the disk etc. we calculate the new parameters for the superblock. After
1353  * having done this we just call growfs() which will do the work.
1354  * We still have to provide support for snapshots. Therefore we first have to
1355  * understand what data structures are always replicated in the snapshot on
1356  * creation, for all other blocks we touch during our procedure, we have to
1357  * keep the old blocks unchanged somewhere available for the snapshots. If we
1358  * are lucky, then we only have to handle our blocks to be relocated in that
1359  * way.
1360  * Also we have to consider in what order we actually update the critical
1361  * data structures of the file system to make sure, that in case of a disaster
1362  * fsck(8) is still able to restore any lost data.
1363  * The foreseen last step then will be to provide for growing even mounted
1364  * file systems. There we have to extend the mount() system call to provide
1365  * userland access to the file system locking facility.
1366  */
1367 int
1368 main(int argc, char **argv)
1369 {
1370 	DBG_FUNC("main")
1371 	const char *device;
1372 	const struct statfs *statfsp;
1373 	uint64_t size = 0;
1374 	off_t mediasize;
1375 	int error, i, j, fsi, fso, ch, Nflag = 0, yflag = 0;
1376 	char *p, reply[5], oldsizebuf[6], newsizebuf[6];
1377 	void *testbuf;
1378 
1379 	DBG_ENTER;
1380 
1381 	while ((ch = getopt(argc, argv, "Ns:vy")) != -1) {
1382 		switch(ch) {
1383 		case 'N':
1384 			Nflag = 1;
1385 			break;
1386 		case 's':
1387 			size = (off_t)strtoumax(optarg, &p, 0);
1388 			if (p == NULL || *p == '\0')
1389 				size *= DEV_BSIZE;
1390 			else if (*p == 'b' || *p == 'B')
1391 				; /* do nothing */
1392 			else if (*p == 'k' || *p == 'K')
1393 				size <<= 10;
1394 			else if (*p == 'm' || *p == 'M')
1395 				size <<= 20;
1396 			else if (*p == 'g' || *p == 'G')
1397 				size <<= 30;
1398 			else if (*p == 't' || *p == 'T') {
1399 				size <<= 30;
1400 				size <<= 10;
1401 			} else
1402 				errx(1, "unknown suffix on -s argument");
1403 			break;
1404 		case 'v': /* for compatibility to newfs */
1405 			break;
1406 		case 'y':
1407 			yflag = 1;
1408 			break;
1409 		case '?':
1410 			/* FALLTHROUGH */
1411 		default:
1412 			usage();
1413 		}
1414 	}
1415 	argc -= optind;
1416 	argv += optind;
1417 
1418 	if (argc != 1)
1419 		usage();
1420 
1421 	/*
1422 	 * Now try to guess the device name.
1423 	 */
1424 	device = getdev(*argv);
1425 	if (device == NULL)
1426 		errx(1, "cannot find special device for %s", *argv);
1427 
1428 	statfsp = dev_to_statfs(device);
1429 
1430 	fsi = open(device, O_RDONLY);
1431 	if (fsi < 0)
1432 		err(1, "%s", device);
1433 
1434 	/*
1435 	 * Try to guess the slice size if not specified.
1436 	 */
1437 	if (ioctl(fsi, DIOCGMEDIASIZE, &mediasize) == -1)
1438 		err(1,"DIOCGMEDIASIZE");
1439 
1440 	/*
1441 	 * Check if that partition is suitable for growing a file system.
1442 	 */
1443 	if (mediasize < 1)
1444 		errx(1, "partition is unavailable");
1445 
1446 	/*
1447 	 * Read the current superblock, and take a backup.
1448 	 */
1449 	for (i = 0; sblock_try[i] != -1; i++) {
1450 		sblockloc = sblock_try[i] / DEV_BSIZE;
1451 		rdfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&(osblock), fsi);
1452 		if ((osblock.fs_magic == FS_UFS1_MAGIC ||
1453 		    (osblock.fs_magic == FS_UFS2_MAGIC &&
1454 		    osblock.fs_sblockloc == sblock_try[i])) &&
1455 		    osblock.fs_bsize <= MAXBSIZE &&
1456 		    osblock.fs_bsize >= (int32_t) sizeof(struct fs))
1457 			break;
1458 	}
1459 	if (sblock_try[i] == -1)
1460 		errx(1, "superblock not recognized");
1461 	memcpy((void *)&fsun1, (void *)&fsun2, sizeof(fsun2));
1462 
1463 	DBG_OPEN("/tmp/growfs.debug"); /* already here we need a superblock */
1464 	DBG_DUMP_FS(&sblock, "old sblock");
1465 
1466 	/*
1467 	 * Determine size to grow to. Default to the device size.
1468 	 */
1469 	if (size == 0)
1470 		size = mediasize;
1471 	else {
1472 		if (size > (uint64_t)mediasize) {
1473 			humanize_number(oldsizebuf, sizeof(oldsizebuf), size,
1474 			    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1475 			humanize_number(newsizebuf, sizeof(newsizebuf),
1476 			    mediasize,
1477 			    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1478 
1479 			errx(1, "requested size %s is larger "
1480 			    "than the available %s", oldsizebuf, newsizebuf);
1481 		}
1482 	}
1483 
1484 	/*
1485 	 * Make sure the new size is a multiple of fs_fsize; /dev/ufssuspend
1486 	 * only supports fragment-aligned IO requests.
1487 	 */
1488 	size -= size % osblock.fs_fsize;
1489 
1490 	if (size <= (uint64_t)(osblock.fs_size * osblock.fs_fsize)) {
1491 		humanize_number(oldsizebuf, sizeof(oldsizebuf),
1492 		    osblock.fs_size * osblock.fs_fsize,
1493 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1494 		humanize_number(newsizebuf, sizeof(newsizebuf), size,
1495 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1496 
1497 		errx(1, "requested size %s is not larger than the current "
1498 		   "filesystem size %s", newsizebuf, oldsizebuf);
1499 	}
1500 
1501 	sblock.fs_size = dbtofsb(&osblock, size / DEV_BSIZE);
1502 	sblock.fs_providersize = dbtofsb(&osblock, mediasize / DEV_BSIZE);
1503 
1504 	/*
1505 	 * Are we really growing?
1506 	 */
1507 	if (osblock.fs_size >= sblock.fs_size) {
1508 		errx(1, "we are not growing (%jd->%jd)",
1509 		    (intmax_t)osblock.fs_size, (intmax_t)sblock.fs_size);
1510 	}
1511 
1512 	/*
1513 	 * Check if we find an active snapshot.
1514 	 */
1515 	if (yflag == 0) {
1516 		for (j = 0; j < FSMAXSNAP; j++) {
1517 			if (sblock.fs_snapinum[j]) {
1518 				errx(1, "active snapshot found in file system; "
1519 				    "please remove all snapshots before "
1520 				    "using growfs");
1521 			}
1522 			if (!sblock.fs_snapinum[j]) /* list is dense */
1523 				break;
1524 		}
1525 	}
1526 
1527 	if (yflag == 0 && Nflag == 0) {
1528 		if (statfsp != NULL && (statfsp->f_flags & MNT_RDONLY) == 0)
1529 			printf("Device is mounted read-write; resizing will "
1530 			    "result in temporary write suspension for %s.\n",
1531 			    statfsp->f_mntonname);
1532 		printf("It's strongly recommended to make a backup "
1533 		    "before growing the file system.\n"
1534 		    "OK to grow filesystem on %s", device);
1535 		if (statfsp != NULL)
1536 			printf(", mounted on %s,", statfsp->f_mntonname);
1537 		humanize_number(oldsizebuf, sizeof(oldsizebuf),
1538 		    osblock.fs_size * osblock.fs_fsize,
1539 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1540 		humanize_number(newsizebuf, sizeof(newsizebuf),
1541 		    sblock.fs_size * sblock.fs_fsize,
1542 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1543 		printf(" from %s to %s? [yes/no] ", oldsizebuf, newsizebuf);
1544 		fflush(stdout);
1545 		fgets(reply, (int)sizeof(reply), stdin);
1546 		if (strcasecmp(reply, "yes\n")){
1547 			printf("Response other than \"yes\"; aborting\n");
1548 			exit(0);
1549 		}
1550 	}
1551 
1552 	/*
1553 	 * Try to access our device for writing.  If it's not mounted,
1554 	 * or mounted read-only, simply open it; otherwise, use UFS
1555 	 * suspension mechanism.
1556 	 */
1557 	if (Nflag) {
1558 		fso = -1;
1559 	} else {
1560 		if (statfsp != NULL && (statfsp->f_flags & MNT_RDONLY) == 0) {
1561 			fso = open(_PATH_UFSSUSPEND, O_RDWR);
1562 			if (fso == -1)
1563 				err(1, "unable to open %s", _PATH_UFSSUSPEND);
1564 			error = ioctl(fso, UFSSUSPEND, &statfsp->f_fsid);
1565 			if (error != 0)
1566 				err(1, "UFSSUSPEND");
1567 		} else {
1568 			fso = open(device, O_WRONLY);
1569 			if (fso < 0)
1570 				err(1, "%s", device);
1571 		}
1572 	}
1573 
1574 	/*
1575 	 * Try to access our new last block in the file system.
1576 	 */
1577 	testbuf = malloc(sblock.fs_fsize);
1578 	if (testbuf == NULL)
1579 		err(1, "malloc");
1580 	rdfs((ufs2_daddr_t)((size - sblock.fs_fsize) / DEV_BSIZE),
1581 	    sblock.fs_fsize, testbuf, fsi);
1582 	wtfs((ufs2_daddr_t)((size - sblock.fs_fsize) / DEV_BSIZE),
1583 	    sblock.fs_fsize, testbuf, fso, Nflag);
1584 	free(testbuf);
1585 
1586 	/*
1587 	 * Now calculate new superblock values and check for reasonable
1588 	 * bound for new file system size:
1589 	 *     fs_size:    is derived from user input
1590 	 *     fs_dsize:   should get updated in the routines creating or
1591 	 *                 updating the cylinder groups on the fly
1592 	 *     fs_cstotal: should get updated in the routines creating or
1593 	 *                 updating the cylinder groups
1594 	 */
1595 
1596 	/*
1597 	 * Update the number of cylinders and cylinder groups in the file system.
1598 	 */
1599 	if (sblock.fs_magic == FS_UFS1_MAGIC) {
1600 		sblock.fs_old_ncyl =
1601 		    sblock.fs_size * sblock.fs_old_nspf / sblock.fs_old_spc;
1602 		if (sblock.fs_size * sblock.fs_old_nspf >
1603 		    sblock.fs_old_ncyl * sblock.fs_old_spc)
1604 			sblock.fs_old_ncyl++;
1605 	}
1606 	sblock.fs_ncg = howmany(sblock.fs_size, sblock.fs_fpg);
1607 
1608 	/*
1609 	 * Allocate last cylinder group only if there is enough room
1610 	 * for at least one data block.
1611 	 */
1612 	if (sblock.fs_size % sblock.fs_fpg != 0 &&
1613 	    sblock.fs_size <= cgdmin(&sblock, sblock.fs_ncg - 1)) {
1614 		humanize_number(oldsizebuf, sizeof(oldsizebuf),
1615 		    (sblock.fs_size % sblock.fs_fpg) * sblock.fs_fsize,
1616 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1617 		warnx("no room to allocate last cylinder group; "
1618 		    "leaving %s unused", oldsizebuf);
1619 		sblock.fs_ncg--;
1620 		if (sblock.fs_magic == FS_UFS1_MAGIC)
1621 			sblock.fs_old_ncyl = sblock.fs_ncg * sblock.fs_old_cpg;
1622 		sblock.fs_size = sblock.fs_ncg * sblock.fs_fpg;
1623 	}
1624 
1625 	/*
1626 	 * Update the space for the cylinder group summary information in the
1627 	 * respective cylinder group data area.
1628 	 */
1629 	sblock.fs_cssize =
1630 	    fragroundup(&sblock, sblock.fs_ncg * sizeof(struct csum));
1631 
1632 	if (osblock.fs_size >= sblock.fs_size)
1633 		errx(1, "not enough new space");
1634 
1635 	DBG_PRINT0("sblock calculated\n");
1636 
1637 	/*
1638 	 * Ok, everything prepared, so now let's do the tricks.
1639 	 */
1640 	growfs(fsi, fso, Nflag);
1641 
1642 	close(fsi);
1643 	if (fso > -1) {
1644 		if (statfsp != NULL && (statfsp->f_flags & MNT_RDONLY) == 0) {
1645 			error = ioctl(fso, UFSRESUME);
1646 			if (error != 0)
1647 				err(1, "UFSRESUME");
1648 		}
1649 		error = close(fso);
1650 		if (error != 0)
1651 			err(1, "close");
1652 		if (statfsp != NULL && (statfsp->f_flags & MNT_RDONLY) != 0)
1653 			mount_reload(statfsp);
1654 	}
1655 
1656 	DBG_CLOSE;
1657 
1658 	DBG_LEAVE;
1659 	return (0);
1660 }
1661 
1662 /*
1663  * Dump a line of usage.
1664  */
1665 static void
1666 usage(void)
1667 {
1668 	DBG_FUNC("usage")
1669 
1670 	DBG_ENTER;
1671 
1672 	fprintf(stderr, "usage: growfs [-Ny] [-s size] special | filesystem\n");
1673 
1674 	DBG_LEAVE;
1675 	exit(1);
1676 }
1677 
1678 /*
1679  * This updates most parameters and the bitmap related to cluster. We have to
1680  * assume that sblock, osblock, acg are set up.
1681  */
1682 static void
1683 updclst(int block)
1684 {
1685 	DBG_FUNC("updclst")
1686 	static int lcs = 0;
1687 
1688 	DBG_ENTER;
1689 
1690 	if (sblock.fs_contigsumsize < 1) /* no clustering */
1691 		return;
1692 	/*
1693 	 * update cluster allocation map
1694 	 */
1695 	setbit(cg_clustersfree(&acg), block);
1696 
1697 	/*
1698 	 * update cluster summary table
1699 	 */
1700 	if (!lcs) {
1701 		/*
1702 		 * calculate size for the trailing cluster
1703 		 */
1704 		for (block--; lcs < sblock.fs_contigsumsize; block--, lcs++ ) {
1705 			if (isclr(cg_clustersfree(&acg), block))
1706 				break;
1707 		}
1708 	}
1709 	if (lcs < sblock.fs_contigsumsize) {
1710 		if (lcs)
1711 			cg_clustersum(&acg)[lcs]--;
1712 		lcs++;
1713 		cg_clustersum(&acg)[lcs]++;
1714 	}
1715 
1716 	DBG_LEAVE;
1717 	return;
1718 }
1719 
1720 static void
1721 mount_reload(const struct statfs *stfs)
1722 {
1723 	char errmsg[255];
1724 	struct iovec *iov;
1725 	int iovlen;
1726 
1727 	iov = NULL;
1728 	iovlen = 0;
1729 	*errmsg = '\0';
1730 	build_iovec(&iov, &iovlen, "fstype", __DECONST(char *, "ffs"), 4);
1731 	build_iovec(&iov, &iovlen, "fspath", __DECONST(char *, stfs->f_mntonname), (size_t)-1);
1732 	build_iovec(&iov, &iovlen, "errmsg", errmsg, sizeof(errmsg));
1733 	build_iovec(&iov, &iovlen, "update", NULL, 0);
1734 	build_iovec(&iov, &iovlen, "reload", NULL, 0);
1735 
1736 	if (nmount(iov, iovlen, stfs->f_flags) < 0) {
1737 		errmsg[sizeof(errmsg) - 1] = '\0';
1738 		err(9, "%s: cannot reload filesystem%s%s", stfs->f_mntonname,
1739 		    *errmsg != '\0' ? ": " : "", errmsg);
1740 	}
1741 }
1742