xref: /freebsd/sbin/growfs/growfs.c (revision 0b3105a37d7adcadcb720112fed4dc4e8040be99)
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 = sblock.fs_ipg < 2 * INOPB(&sblock) ?
352 	    sblock.fs_ipg : 2 * INOPB(&sblock);
353 	acg.cg_ndblk = dmax - cbase;
354 	if (sblock.fs_contigsumsize > 0)
355 		acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
356 	start = &acg.cg_space[0] - (u_char *)(&acg.cg_firstfield);
357 	if (sblock.fs_magic == FS_UFS2_MAGIC) {
358 		acg.cg_iusedoff = start;
359 	} else {
360 		acg.cg_old_ncyl = sblock.fs_old_cpg;
361 		acg.cg_old_time = acg.cg_time;
362 		acg.cg_time = 0;
363 		acg.cg_old_niblk = acg.cg_niblk;
364 		acg.cg_niblk = 0;
365 		acg.cg_initediblk = 0;
366 		acg.cg_old_btotoff = start;
367 		acg.cg_old_boff = acg.cg_old_btotoff +
368 		    sblock.fs_old_cpg * sizeof(int32_t);
369 		acg.cg_iusedoff = acg.cg_old_boff +
370 		    sblock.fs_old_cpg * sizeof(u_int16_t);
371 	}
372 	acg.cg_freeoff = acg.cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
373 	acg.cg_nextfreeoff = acg.cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT);
374 	if (sblock.fs_contigsumsize > 0) {
375 		acg.cg_clustersumoff =
376 		    roundup(acg.cg_nextfreeoff, sizeof(u_int32_t));
377 		acg.cg_clustersumoff -= sizeof(u_int32_t);
378 		acg.cg_clusteroff = acg.cg_clustersumoff +
379 		    (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
380 		acg.cg_nextfreeoff = acg.cg_clusteroff +
381 		    howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
382 	}
383 	if (acg.cg_nextfreeoff > (unsigned)sblock.fs_cgsize) {
384 		/*
385 		 * This should never happen as we would have had that panic
386 		 * already on file system creation
387 		 */
388 		errx(37, "panic: cylinder group too big");
389 	}
390 	acg.cg_cs.cs_nifree += sblock.fs_ipg;
391 	if (cylno == 0)
392 		for (ino = 0; ino < ROOTINO; ino++) {
393 			setbit(cg_inosused(&acg), ino);
394 			acg.cg_cs.cs_nifree--;
395 		}
396 	/*
397 	 * For the old file system, we have to initialize all the inodes.
398 	 */
399 	if (sblock.fs_magic == FS_UFS1_MAGIC) {
400 		bzero(iobuf, sblock.fs_bsize);
401 		for (i = 0; i < sblock.fs_ipg / INOPF(&sblock);
402 		    i += sblock.fs_frag) {
403 			dp1 = (struct ufs1_dinode *)(void *)iobuf;
404 			for (j = 0; j < INOPB(&sblock); j++) {
405 				dp1->di_gen = arc4random();
406 				dp1++;
407 			}
408 			wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno) + i),
409 			    sblock.fs_bsize, iobuf, fso, Nflag);
410 		}
411 	}
412 	if (cylno > 0) {
413 		/*
414 		 * In cylno 0, beginning space is reserved
415 		 * for boot and super blocks.
416 		 */
417 		for (d = 0; d < dlower; d += sblock.fs_frag) {
418 			blkno = d / sblock.fs_frag;
419 			setblock(&sblock, cg_blksfree(&acg), blkno);
420 			if (sblock.fs_contigsumsize > 0)
421 				setbit(cg_clustersfree(&acg), blkno);
422 			acg.cg_cs.cs_nbfree++;
423 		}
424 		sblock.fs_dsize += dlower;
425 	}
426 	sblock.fs_dsize += acg.cg_ndblk - dupper;
427 	if ((i = dupper % sblock.fs_frag)) {
428 		acg.cg_frsum[sblock.fs_frag - i]++;
429 		for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
430 			setbit(cg_blksfree(&acg), dupper);
431 			acg.cg_cs.cs_nffree++;
432 		}
433 	}
434 	for (d = dupper; d + sblock.fs_frag <= acg.cg_ndblk;
435 	    d += sblock.fs_frag) {
436 		blkno = d / sblock.fs_frag;
437 		setblock(&sblock, cg_blksfree(&acg), blkno);
438 		if (sblock.fs_contigsumsize > 0)
439 			setbit(cg_clustersfree(&acg), blkno);
440 		acg.cg_cs.cs_nbfree++;
441 	}
442 	if (d < acg.cg_ndblk) {
443 		acg.cg_frsum[acg.cg_ndblk - d]++;
444 		for (; d < acg.cg_ndblk; d++) {
445 			setbit(cg_blksfree(&acg), d);
446 			acg.cg_cs.cs_nffree++;
447 		}
448 	}
449 	if (sblock.fs_contigsumsize > 0) {
450 		int32_t *sump = cg_clustersum(&acg);
451 		u_char *mapp = cg_clustersfree(&acg);
452 		int map = *mapp++;
453 		int bit = 1;
454 		int run = 0;
455 
456 		for (i = 0; i < acg.cg_nclusterblks; i++) {
457 			if ((map & bit) != 0)
458 				run++;
459 			else if (run != 0) {
460 				if (run > sblock.fs_contigsumsize)
461 					run = sblock.fs_contigsumsize;
462 				sump[run]++;
463 				run = 0;
464 			}
465 			if ((i & (CHAR_BIT - 1)) != CHAR_BIT - 1)
466 				bit <<= 1;
467 			else {
468 				map = *mapp++;
469 				bit = 1;
470 			}
471 		}
472 		if (run != 0) {
473 			if (run > sblock.fs_contigsumsize)
474 				run = sblock.fs_contigsumsize;
475 			sump[run]++;
476 		}
477 	}
478 	sblock.fs_cstotal.cs_ndir += acg.cg_cs.cs_ndir;
479 	sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
480 	sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
481 	sblock.fs_cstotal.cs_nifree += acg.cg_cs.cs_nifree;
482 	*cs = acg.cg_cs;
483 
484 	memcpy(iobuf, &acg, sblock.fs_cgsize);
485 	memset(iobuf + sblock.fs_cgsize, '\0',
486 	    sblock.fs_bsize * 3 - sblock.fs_cgsize);
487 
488 	wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
489 	    sblock.fs_bsize * 3, iobuf, 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 	wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize,
776 	    (void *)&acg, fso, Nflag);
777 	DBG_PRINT0("jcg written\n");
778 	DBG_DUMP_CG(&sblock, "new joining cg", &acg);
779 
780 	DBG_LEAVE;
781 	return;
782 }
783 
784 /*
785  * Here we update the location of the cylinder summary. We have two possible
786  * ways of growing the cylinder summary:
787  * (1)	We can try to grow the summary in the current location, and relocate
788  *	possibly used blocks within the current cylinder group.
789  * (2)	Alternatively we can relocate the whole cylinder summary to the first
790  *	new completely empty cylinder group. Once the cylinder summary is no
791  *	longer in the beginning of the first cylinder group you should never
792  *	use a version of fsck which is not aware of the possibility to have
793  *	this structure in a non standard place.
794  * Option (2) is considered to be less intrusive to the structure of the file-
795  * system, so that's the one being used.
796  */
797 static void
798 updcsloc(time_t modtime, int fsi, int fso, unsigned int Nflag)
799 {
800 	DBG_FUNC("updcsloc")
801 	struct csum *cs;
802 	int ocscg, ncscg;
803 	ufs2_daddr_t d;
804 	int lcs = 0;
805 	int block;
806 
807 	DBG_ENTER;
808 
809 	if (howmany(sblock.fs_cssize, sblock.fs_fsize) ==
810 	    howmany(osblock.fs_cssize, osblock.fs_fsize)) {
811 		/*
812 		 * No new fragment needed.
813 		 */
814 		DBG_LEAVE;
815 		return;
816 	}
817 	ocscg = dtog(&osblock, osblock.fs_csaddr);
818 	cs = fscs + ocscg;
819 
820 	/*
821 	 * Read original cylinder group from disk, and make a copy.
822 	 * XXX	If Nflag is set in some very rare cases we now miss
823 	 *	some changes done in updjcg by reading the unmodified
824 	 *	block from disk.
825 	 */
826 	rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)),
827 	    (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
828 	DBG_PRINT0("oscg read\n");
829 	DBG_DUMP_CG(&sblock, "old summary cg", &aocg);
830 
831 	memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
832 
833 	/*
834 	 * Touch the cylinder group, set up local variables needed later
835 	 * and update the superblock.
836 	 */
837 	acg.cg_time = modtime;
838 
839 	/*
840 	 * XXX	In the case of having active snapshots we may need much more
841 	 *	blocks for the copy on write. We need each block twice, and
842 	 *	also up to 8*3 blocks for indirect blocks for all possible
843 	 *	references.
844 	 */
845 	/*
846 	 * There is not enough space in the old cylinder group to
847 	 * relocate all blocks as needed, so we relocate the whole
848 	 * cylinder group summary to a new group. We try to use the
849 	 * first complete new cylinder group just created. Within the
850 	 * cylinder group we align the area immediately after the
851 	 * cylinder group information location in order to be as
852 	 * close as possible to the original implementation of ffs.
853 	 *
854 	 * First we have to make sure we'll find enough space in the
855 	 * new cylinder group. If not, then we currently give up.
856 	 * We start with freeing everything which was used by the
857 	 * fragments of the old cylinder summary in the current group.
858 	 * Now we write back the group meta data, read in the needed
859 	 * meta data from the new cylinder group, and start allocating
860 	 * within that group. Here we can assume, the group to be
861 	 * completely empty. Which makes the handling of fragments and
862 	 * clusters a lot easier.
863 	 */
864 	DBG_TRC;
865 	if (sblock.fs_ncg - osblock.fs_ncg < 2)
866 		errx(2, "panic: not enough space");
867 
868 	/*
869 	 * Point "d" to the first fragment not used by the cylinder
870 	 * summary.
871 	 */
872 	d = osblock.fs_csaddr + (osblock.fs_cssize / osblock.fs_fsize);
873 
874 	/*
875 	 * Set up last cluster size ("lcs") already here. Calculate
876 	 * the size for the trailing cluster just behind where "d"
877 	 * points to.
878 	 */
879 	if (sblock.fs_contigsumsize > 0) {
880 		for (block = howmany(d % sblock.fs_fpg, sblock.fs_frag),
881 		    lcs = 0; lcs < sblock.fs_contigsumsize; block++, lcs++) {
882 			if (isclr(cg_clustersfree(&acg), block))
883 				break;
884 		}
885 	}
886 
887 	/*
888 	 * Point "d" to the last frag used by the cylinder summary.
889 	 */
890 	d--;
891 
892 	DBG_PRINT1("d=%jd\n", (intmax_t)d);
893 	if ((d + 1) % sblock.fs_frag) {
894 		/*
895 		 * The end of the cylinder summary is not a complete
896 		 * block.
897 		 */
898 		DBG_TRC;
899 		frag_adjust(d % sblock.fs_fpg, -1);
900 		for (; (d + 1) % sblock.fs_frag; d--) {
901 			DBG_PRINT1("d=%jd\n", (intmax_t)d);
902 			setbit(cg_blksfree(&acg), d % sblock.fs_fpg);
903 			acg.cg_cs.cs_nffree++;
904 			sblock.fs_cstotal.cs_nffree++;
905 		}
906 		/*
907 		 * Point "d" to the last fragment of the last
908 		 * (incomplete) block of the cylinder summary.
909 		 */
910 		d++;
911 		frag_adjust(d % sblock.fs_fpg, 1);
912 
913 		if (isblock(&sblock, cg_blksfree(&acg),
914 		    (d % sblock.fs_fpg) / sblock.fs_frag)) {
915 			DBG_PRINT1("d=%jd\n", (intmax_t)d);
916 			acg.cg_cs.cs_nffree -= sblock.fs_frag;
917 			acg.cg_cs.cs_nbfree++;
918 			sblock.fs_cstotal.cs_nffree -= sblock.fs_frag;
919 			sblock.fs_cstotal.cs_nbfree++;
920 			if (sblock.fs_contigsumsize > 0) {
921 				setbit(cg_clustersfree(&acg),
922 				    (d % sblock.fs_fpg) / sblock.fs_frag);
923 				if (lcs < sblock.fs_contigsumsize) {
924 					if (lcs)
925 						cg_clustersum(&acg)[lcs]--;
926 					lcs++;
927 					cg_clustersum(&acg)[lcs]++;
928 				}
929 			}
930 		}
931 		/*
932 		 * Point "d" to the first fragment of the block before
933 		 * the last incomplete block.
934 		 */
935 		d--;
936 	}
937 
938 	DBG_PRINT1("d=%jd\n", (intmax_t)d);
939 	for (d = rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr;
940 	    d -= sblock.fs_frag) {
941 		DBG_TRC;
942 		DBG_PRINT1("d=%jd\n", (intmax_t)d);
943 		setblock(&sblock, cg_blksfree(&acg),
944 		    (d % sblock.fs_fpg) / sblock.fs_frag);
945 		acg.cg_cs.cs_nbfree++;
946 		sblock.fs_cstotal.cs_nbfree++;
947 		if (sblock.fs_contigsumsize > 0) {
948 			setbit(cg_clustersfree(&acg),
949 			    (d % sblock.fs_fpg) / sblock.fs_frag);
950 			/*
951 			 * The last cluster size is already set up.
952 			 */
953 			if (lcs < sblock.fs_contigsumsize) {
954 				if (lcs)
955 					cg_clustersum(&acg)[lcs]--;
956 				lcs++;
957 				cg_clustersum(&acg)[lcs]++;
958 			}
959 		}
960 	}
961 	*cs = acg.cg_cs;
962 
963 	/*
964 	 * Now write the former cylinder group containing the cylinder
965 	 * summary back to disk.
966 	 */
967 	wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)),
968 	    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
969 	DBG_PRINT0("oscg written\n");
970 	DBG_DUMP_CG(&sblock, "old summary cg", &acg);
971 
972 	/*
973 	 * Find the beginning of the new cylinder group containing the
974 	 * cylinder summary.
975 	 */
976 	sblock.fs_csaddr = cgdmin(&sblock, osblock.fs_ncg);
977 	ncscg = dtog(&sblock, sblock.fs_csaddr);
978 	cs = fscs + ncscg;
979 
980 	/*
981 	 * If Nflag is specified, we would now read random data instead
982 	 * of an empty cg structure from disk. So we can't simulate that
983 	 * part for now.
984 	 */
985 	if (Nflag) {
986 		DBG_PRINT0("nscg update skipped\n");
987 		DBG_LEAVE;
988 		return;
989 	}
990 
991 	/*
992 	 * Read the future cylinder group containing the cylinder
993 	 * summary from disk, and make a copy.
994 	 */
995 	rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
996 	    (size_t)sblock.fs_cgsize, (void *)&aocg, fsi);
997 	DBG_PRINT0("nscg read\n");
998 	DBG_DUMP_CG(&sblock, "new summary cg", &aocg);
999 
1000 	memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
1001 
1002 	/*
1003 	 * Allocate all complete blocks used by the new cylinder
1004 	 * summary.
1005 	 */
1006 	for (d = sblock.fs_csaddr; d + sblock.fs_frag <=
1007 	    sblock.fs_csaddr + (sblock.fs_cssize / sblock.fs_fsize);
1008 	    d += sblock.fs_frag) {
1009 		clrblock(&sblock, cg_blksfree(&acg),
1010 		    (d % sblock.fs_fpg) / sblock.fs_frag);
1011 		acg.cg_cs.cs_nbfree--;
1012 		sblock.fs_cstotal.cs_nbfree--;
1013 		if (sblock.fs_contigsumsize > 0) {
1014 			clrbit(cg_clustersfree(&acg),
1015 			    (d % sblock.fs_fpg) / sblock.fs_frag);
1016 		}
1017 	}
1018 
1019 	/*
1020 	 * Allocate all fragments used by the cylinder summary in the
1021 	 * last block.
1022 	 */
1023 	if (d < sblock.fs_csaddr + (sblock.fs_cssize / sblock.fs_fsize)) {
1024 		for (; d - sblock.fs_csaddr <
1025 		    sblock.fs_cssize/sblock.fs_fsize; d++) {
1026 			clrbit(cg_blksfree(&acg), d % sblock.fs_fpg);
1027 			acg.cg_cs.cs_nffree--;
1028 			sblock.fs_cstotal.cs_nffree--;
1029 		}
1030 		acg.cg_cs.cs_nbfree--;
1031 		acg.cg_cs.cs_nffree += sblock.fs_frag;
1032 		sblock.fs_cstotal.cs_nbfree--;
1033 		sblock.fs_cstotal.cs_nffree += sblock.fs_frag;
1034 		if (sblock.fs_contigsumsize > 0)
1035 			clrbit(cg_clustersfree(&acg),
1036 			    (d % sblock.fs_fpg) / sblock.fs_frag);
1037 
1038 		frag_adjust(d % sblock.fs_fpg, 1);
1039 	}
1040 	/*
1041 	 * XXX	Handle the cluster statistics here in the case this
1042 	 *	cylinder group is now almost full, and the remaining
1043 	 *	space is less then the maximum cluster size. This is
1044 	 *	probably not needed, as you would hardly find a file
1045 	 *	system which has only MAXCSBUFS+FS_MAXCONTIG of free
1046 	 *	space right behind the cylinder group information in
1047 	 *	any new cylinder group.
1048 	 */
1049 
1050 	/*
1051 	 * Update our statistics in the cylinder summary.
1052 	 */
1053 	*cs = acg.cg_cs;
1054 
1055 	/*
1056 	 * Write the new cylinder group containing the cylinder summary
1057 	 * back to disk.
1058 	 */
1059 	wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1060 	    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1061 	DBG_PRINT0("nscg written\n");
1062 	DBG_DUMP_CG(&sblock, "new summary cg", &acg);
1063 
1064 	DBG_LEAVE;
1065 	return;
1066 }
1067 
1068 /*
1069  * Here we read some block(s) from disk.
1070  */
1071 static void
1072 rdfs(ufs2_daddr_t bno, size_t size, void *bf, int fsi)
1073 {
1074 	DBG_FUNC("rdfs")
1075 	ssize_t	n;
1076 
1077 	DBG_ENTER;
1078 
1079 	if (bno < 0)
1080 		err(32, "rdfs: attempting to read negative block number");
1081 	if (lseek(fsi, (off_t)bno * DEV_BSIZE, 0) < 0)
1082 		err(33, "rdfs: seek error: %jd", (intmax_t)bno);
1083 	n = read(fsi, bf, size);
1084 	if (n != (ssize_t)size)
1085 		err(34, "rdfs: read error: %jd", (intmax_t)bno);
1086 
1087 	DBG_LEAVE;
1088 	return;
1089 }
1090 
1091 /*
1092  * Here we write some block(s) to disk.
1093  */
1094 static void
1095 wtfs(ufs2_daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag)
1096 {
1097 	DBG_FUNC("wtfs")
1098 	ssize_t	n;
1099 
1100 	DBG_ENTER;
1101 
1102 	if (Nflag) {
1103 		DBG_LEAVE;
1104 		return;
1105 	}
1106 	if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0)
1107 		err(35, "wtfs: seek error: %ld", (long)bno);
1108 	n = write(fso, bf, size);
1109 	if (n != (ssize_t)size)
1110 		err(36, "wtfs: write error: %ld", (long)bno);
1111 
1112 	DBG_LEAVE;
1113 	return;
1114 }
1115 
1116 /*
1117  * Here we check if all frags of a block are free. For more details again
1118  * please see the source of newfs(8), as this function is taken over almost
1119  * unchanged.
1120  */
1121 static int
1122 isblock(struct fs *fs, unsigned char *cp, int h)
1123 {
1124 	DBG_FUNC("isblock")
1125 	unsigned char mask;
1126 
1127 	DBG_ENTER;
1128 
1129 	switch (fs->fs_frag) {
1130 	case 8:
1131 		DBG_LEAVE;
1132 		return (cp[h] == 0xff);
1133 	case 4:
1134 		mask = 0x0f << ((h & 0x1) << 2);
1135 		DBG_LEAVE;
1136 		return ((cp[h >> 1] & mask) == mask);
1137 	case 2:
1138 		mask = 0x03 << ((h & 0x3) << 1);
1139 		DBG_LEAVE;
1140 		return ((cp[h >> 2] & mask) == mask);
1141 	case 1:
1142 		mask = 0x01 << (h & 0x7);
1143 		DBG_LEAVE;
1144 		return ((cp[h >> 3] & mask) == mask);
1145 	default:
1146 		fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
1147 		DBG_LEAVE;
1148 		return (0);
1149 	}
1150 }
1151 
1152 /*
1153  * Here we allocate a complete block in the block map. For more details again
1154  * please see the source of newfs(8), as this function is taken over almost
1155  * unchanged.
1156  */
1157 static void
1158 clrblock(struct fs *fs, unsigned char *cp, int h)
1159 {
1160 	DBG_FUNC("clrblock")
1161 
1162 	DBG_ENTER;
1163 
1164 	switch ((fs)->fs_frag) {
1165 	case 8:
1166 		cp[h] = 0;
1167 		break;
1168 	case 4:
1169 		cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1170 		break;
1171 	case 2:
1172 		cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1173 		break;
1174 	case 1:
1175 		cp[h >> 3] &= ~(0x01 << (h & 0x7));
1176 		break;
1177 	default:
1178 		warnx("clrblock bad fs_frag %d", fs->fs_frag);
1179 		break;
1180 	}
1181 
1182 	DBG_LEAVE;
1183 	return;
1184 }
1185 
1186 /*
1187  * Here we free a complete block in the free block map. For more details again
1188  * please see the source of newfs(8), as this function is taken over almost
1189  * unchanged.
1190  */
1191 static void
1192 setblock(struct fs *fs, unsigned char *cp, int h)
1193 {
1194 	DBG_FUNC("setblock")
1195 
1196 	DBG_ENTER;
1197 
1198 	switch (fs->fs_frag) {
1199 	case 8:
1200 		cp[h] = 0xff;
1201 		break;
1202 	case 4:
1203 		cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1204 		break;
1205 	case 2:
1206 		cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1207 		break;
1208 	case 1:
1209 		cp[h >> 3] |= (0x01 << (h & 0x7));
1210 		break;
1211 	default:
1212 		warnx("setblock bad fs_frag %d", fs->fs_frag);
1213 		break;
1214 	}
1215 
1216 	DBG_LEAVE;
1217 	return;
1218 }
1219 
1220 /*
1221  * Figure out how many lines our current terminal has. For more details again
1222  * please see the source of newfs(8), as this function is taken over almost
1223  * unchanged.
1224  */
1225 static int
1226 charsperline(void)
1227 {
1228 	DBG_FUNC("charsperline")
1229 	int columns;
1230 	char *cp;
1231 	struct winsize ws;
1232 
1233 	DBG_ENTER;
1234 
1235 	columns = 0;
1236 	if (ioctl(0, TIOCGWINSZ, &ws) != -1)
1237 		columns = ws.ws_col;
1238 	if (columns == 0 && (cp = getenv("COLUMNS")))
1239 		columns = atoi(cp);
1240 	if (columns == 0)
1241 		columns = 80;	/* last resort */
1242 
1243 	DBG_LEAVE;
1244 	return (columns);
1245 }
1246 
1247 static int
1248 is_dev(const char *name)
1249 {
1250 	struct stat devstat;
1251 
1252 	if (stat(name, &devstat) != 0)
1253 		return (0);
1254 	if (!S_ISCHR(devstat.st_mode))
1255 		return (0);
1256 	return (1);
1257 }
1258 
1259 /*
1260  * Return mountpoint on which the device is currently mounted.
1261  */
1262 static const struct statfs *
1263 dev_to_statfs(const char *dev)
1264 {
1265 	struct stat devstat, mntdevstat;
1266 	struct statfs *mntbuf, *statfsp;
1267 	char device[MAXPATHLEN];
1268 	char *mntdevname;
1269 	int i, mntsize;
1270 
1271 	/*
1272 	 * First check the mounted filesystems.
1273 	 */
1274 	if (stat(dev, &devstat) != 0)
1275 		return (NULL);
1276 	if (!S_ISCHR(devstat.st_mode) && !S_ISBLK(devstat.st_mode))
1277 		return (NULL);
1278 
1279 	mntsize = getmntinfo(&mntbuf, MNT_NOWAIT);
1280 	for (i = 0; i < mntsize; i++) {
1281 		statfsp = &mntbuf[i];
1282 		mntdevname = statfsp->f_mntfromname;
1283 		if (*mntdevname != '/') {
1284 			strcpy(device, _PATH_DEV);
1285 			strcat(device, mntdevname);
1286 			mntdevname = device;
1287 		}
1288 		if (stat(mntdevname, &mntdevstat) == 0 &&
1289 		    mntdevstat.st_rdev == devstat.st_rdev)
1290 			return (statfsp);
1291 	}
1292 
1293 	return (NULL);
1294 }
1295 
1296 static const char *
1297 mountpoint_to_dev(const char *mountpoint)
1298 {
1299 	struct statfs *mntbuf, *statfsp;
1300 	struct fstab *fs;
1301 	int i, mntsize;
1302 
1303 	/*
1304 	 * First check the mounted filesystems.
1305 	 */
1306 	mntsize = getmntinfo(&mntbuf, MNT_NOWAIT);
1307 	for (i = 0; i < mntsize; i++) {
1308 		statfsp = &mntbuf[i];
1309 
1310 		if (strcmp(statfsp->f_mntonname, mountpoint) == 0)
1311 			return (statfsp->f_mntfromname);
1312 	}
1313 
1314 	/*
1315 	 * Check the fstab.
1316 	 */
1317 	fs = getfsfile(mountpoint);
1318 	if (fs != NULL)
1319 		return (fs->fs_spec);
1320 
1321 	return (NULL);
1322 }
1323 
1324 static const char *
1325 getdev(const char *name)
1326 {
1327 	static char device[MAXPATHLEN];
1328 	const char *cp, *dev;
1329 
1330 	if (is_dev(name))
1331 		return (name);
1332 
1333 	cp = strrchr(name, '/');
1334 	if (cp == 0) {
1335 		snprintf(device, sizeof(device), "%s%s", _PATH_DEV, name);
1336 		if (is_dev(device))
1337 			return (device);
1338 	}
1339 
1340 	dev = mountpoint_to_dev(name);
1341 	if (dev != NULL && is_dev(dev))
1342 		return (dev);
1343 
1344 	return (NULL);
1345 }
1346 
1347 /*
1348  * growfs(8) is a utility which allows to increase the size of an existing
1349  * ufs file system. Currently this can only be done on unmounted file system.
1350  * It recognizes some command line options to specify the new desired size,
1351  * and it does some basic checkings. The old file system size is determined
1352  * and after some more checks like we can really access the new last block
1353  * on the disk etc. we calculate the new parameters for the superblock. After
1354  * having done this we just call growfs() which will do the work.
1355  * We still have to provide support for snapshots. Therefore we first have to
1356  * understand what data structures are always replicated in the snapshot on
1357  * creation, for all other blocks we touch during our procedure, we have to
1358  * keep the old blocks unchanged somewhere available for the snapshots. If we
1359  * are lucky, then we only have to handle our blocks to be relocated in that
1360  * way.
1361  * Also we have to consider in what order we actually update the critical
1362  * data structures of the file system to make sure, that in case of a disaster
1363  * fsck(8) is still able to restore any lost data.
1364  * The foreseen last step then will be to provide for growing even mounted
1365  * file systems. There we have to extend the mount() system call to provide
1366  * userland access to the file system locking facility.
1367  */
1368 int
1369 main(int argc, char **argv)
1370 {
1371 	DBG_FUNC("main")
1372 	const char *device;
1373 	const struct statfs *statfsp;
1374 	uint64_t size = 0;
1375 	off_t mediasize;
1376 	int error, i, j, fsi, fso, ch, Nflag = 0, yflag = 0;
1377 	char *p, reply[5], oldsizebuf[6], newsizebuf[6];
1378 	void *testbuf;
1379 
1380 	DBG_ENTER;
1381 
1382 	while ((ch = getopt(argc, argv, "Ns:vy")) != -1) {
1383 		switch(ch) {
1384 		case 'N':
1385 			Nflag = 1;
1386 			break;
1387 		case 's':
1388 			size = (off_t)strtoumax(optarg, &p, 0);
1389 			if (p == NULL || *p == '\0')
1390 				size *= DEV_BSIZE;
1391 			else if (*p == 'b' || *p == 'B')
1392 				; /* do nothing */
1393 			else if (*p == 'k' || *p == 'K')
1394 				size <<= 10;
1395 			else if (*p == 'm' || *p == 'M')
1396 				size <<= 20;
1397 			else if (*p == 'g' || *p == 'G')
1398 				size <<= 30;
1399 			else if (*p == 't' || *p == 'T') {
1400 				size <<= 30;
1401 				size <<= 10;
1402 			} else
1403 				errx(1, "unknown suffix on -s argument");
1404 			break;
1405 		case 'v': /* for compatibility to newfs */
1406 			break;
1407 		case 'y':
1408 			yflag = 1;
1409 			break;
1410 		case '?':
1411 			/* FALLTHROUGH */
1412 		default:
1413 			usage();
1414 		}
1415 	}
1416 	argc -= optind;
1417 	argv += optind;
1418 
1419 	if (argc != 1)
1420 		usage();
1421 
1422 	/*
1423 	 * Now try to guess the device name.
1424 	 */
1425 	device = getdev(*argv);
1426 	if (device == NULL)
1427 		errx(1, "cannot find special device for %s", *argv);
1428 
1429 	statfsp = dev_to_statfs(device);
1430 
1431 	fsi = open(device, O_RDONLY);
1432 	if (fsi < 0)
1433 		err(1, "%s", device);
1434 
1435 	/*
1436 	 * Try to guess the slice size if not specified.
1437 	 */
1438 	if (ioctl(fsi, DIOCGMEDIASIZE, &mediasize) == -1)
1439 		err(1,"DIOCGMEDIASIZE");
1440 
1441 	/*
1442 	 * Check if that partition is suitable for growing a file system.
1443 	 */
1444 	if (mediasize < 1)
1445 		errx(1, "partition is unavailable");
1446 
1447 	/*
1448 	 * Read the current superblock, and take a backup.
1449 	 */
1450 	for (i = 0; sblock_try[i] != -1; i++) {
1451 		sblockloc = sblock_try[i] / DEV_BSIZE;
1452 		rdfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&(osblock), fsi);
1453 		if ((osblock.fs_magic == FS_UFS1_MAGIC ||
1454 		    (osblock.fs_magic == FS_UFS2_MAGIC &&
1455 		    osblock.fs_sblockloc == sblock_try[i])) &&
1456 		    osblock.fs_bsize <= MAXBSIZE &&
1457 		    osblock.fs_bsize >= (int32_t) sizeof(struct fs))
1458 			break;
1459 	}
1460 	if (sblock_try[i] == -1)
1461 		errx(1, "superblock not recognized");
1462 	memcpy((void *)&fsun1, (void *)&fsun2, sizeof(fsun2));
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("\nNothing done\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