xref: /freebsd/sbin/growfs/growfs.c (revision 66df505066f51e6d8411b966765d828817f88971)
1 /*-
2  * SPDX-License-Identifier: BSD-4-Clause
3  *
4  * Copyright (c) 1980, 1989, 1993 The Regents of the University of California.
5  * Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz
6  * Copyright (c) 2012 The FreeBSD Foundation
7  * All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * Christoph Herrmann and Thomas-Henning von Kamptz, Munich and Frankfurt.
11  *
12  * Portions of this software were developed by Edward Tomasz Napierala
13  * under sponsorship from the FreeBSD Foundation.
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions
17  * are met:
18  * 1. Redistributions of source code must retain the above copyright
19  *    notice, this list of conditions and the following disclaimer.
20  * 2. Redistributions in binary form must reproduce the above copyright
21  *    notice, this list of conditions and the following disclaimer in the
22  *    documentation and/or other materials provided with the distribution.
23  * 3. All advertising materials mentioning features or use of this software
24  *    must display the following acknowledgment:
25  *      This product includes software developed by the University of
26  *      California, Berkeley and its contributors, as well as Christoph
27  *      Herrmann and Thomas-Henning von Kamptz.
28  * 4. Neither the name of the University nor the names of its contributors
29  *    may be used to endorse or promote products derived from this software
30  *    without specific prior written permission.
31  *
32  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
33  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
34  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
35  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
36  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
37  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
38  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
39  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
40  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
41  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
42  * SUCH DAMAGE.
43  *
44  * $TSHeader: src/sbin/growfs/growfs.c,v 1.5 2000/12/12 19:31:00 tomsoft Exp $
45  *
46  */
47 
48 #ifndef lint
49 static const char copyright[] =
50 "@(#) Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz\n\
51 Copyright (c) 1980, 1989, 1993 The Regents of the University of California.\n\
52 All rights reserved.\n";
53 #endif /* not lint */
54 
55 #include <sys/cdefs.h>
56 __FBSDID("$FreeBSD$");
57 
58 #include <sys/param.h>
59 #include <sys/ioctl.h>
60 #include <sys/stat.h>
61 #include <sys/disk.h>
62 #include <sys/ucred.h>
63 #include <sys/mount.h>
64 
65 #include <stdio.h>
66 #include <paths.h>
67 #include <ctype.h>
68 #include <err.h>
69 #include <errno.h>
70 #include <fcntl.h>
71 #include <fstab.h>
72 #include <inttypes.h>
73 #include <limits.h>
74 #include <mntopts.h>
75 #include <paths.h>
76 #include <stdlib.h>
77 #include <stdint.h>
78 #include <string.h>
79 #include <time.h>
80 #include <unistd.h>
81 #include <ufs/ufs/dinode.h>
82 #include <ufs/ffs/fs.h>
83 #include <libutil.h>
84 #include <libufs.h>
85 
86 #include "debug.h"
87 
88 #ifdef FS_DEBUG
89 int	_dbg_lvl_ = (DL_INFO);	/* DL_TRC */
90 #endif /* FS_DEBUG */
91 
92 static union {
93 	struct fs	fs;
94 	char		pad[SBLOCKSIZE];
95 } fsun1, fsun2;
96 #define	sblock	fsun1.fs	/* the new superblock */
97 #define	osblock	fsun2.fs	/* the old superblock */
98 
99 static union {
100 	struct cg	cg;
101 	char		pad[MAXBSIZE];
102 } cgun1, cgun2;
103 #define	acg	cgun1.cg	/* a cylinder cgroup (new) */
104 #define	aocg	cgun2.cg	/* an old cylinder group */
105 
106 static struct csum	*fscs;	/* cylinder summary */
107 
108 static void	growfs(int, int, unsigned int);
109 static void	rdfs(ufs2_daddr_t, size_t, void *, int);
110 static void	wtfs(ufs2_daddr_t, size_t, void *, int, unsigned int);
111 static int	charsperline(void);
112 static void	usage(void);
113 static int	isblock(struct fs *, unsigned char *, int);
114 static void	clrblock(struct fs *, unsigned char *, int);
115 static void	setblock(struct fs *, unsigned char *, int);
116 static void	initcg(int, time_t, int, unsigned int);
117 static void	updjcg(int, time_t, int, int, unsigned int);
118 static void	updcsloc(time_t, int, int, unsigned int);
119 static void	frag_adjust(ufs2_daddr_t, int);
120 static void	updclst(int);
121 static void	mount_reload(const struct statfs *stfs);
122 static void	cgckhash(struct cg *);
123 
124 /*
125  * Here we actually start growing the file system. We basically read the
126  * cylinder summary from the first cylinder group as we want to update
127  * this on the fly during our various operations. First we handle the
128  * changes in the former last cylinder group. Afterwards we create all new
129  * cylinder groups.  Now we handle the cylinder group containing the
130  * cylinder summary which might result in a relocation of the whole
131  * structure.  In the end we write back the updated cylinder summary, the
132  * new superblock, and slightly patched versions of the super block
133  * copies.
134  */
135 static void
136 growfs(int fsi, int fso, unsigned int Nflag)
137 {
138 	DBG_FUNC("growfs")
139 	time_t modtime;
140 	uint cylno;
141 	int i, j, width;
142 	char tmpbuf[100];
143 
144 	DBG_ENTER;
145 
146 	time(&modtime);
147 
148 	/*
149 	 * Get the cylinder summary into the memory.
150 	 */
151 	fscs = (struct csum *)calloc((size_t)1, (size_t)sblock.fs_cssize);
152 	if (fscs == NULL)
153 		errx(1, "calloc failed");
154 	memcpy(fscs, osblock.fs_csp, osblock.fs_cssize);
155 	free(osblock.fs_csp);
156 	osblock.fs_csp = NULL;
157 	sblock.fs_csp = fscs;
158 
159 #ifdef FS_DEBUG
160 	{
161 		struct csum *dbg_csp;
162 		u_int32_t dbg_csc;
163 		char dbg_line[80];
164 
165 		dbg_csp = fscs;
166 
167 		for (dbg_csc = 0; dbg_csc < osblock.fs_ncg; dbg_csc++) {
168 			snprintf(dbg_line, sizeof(dbg_line),
169 			    "%d. old csum in old location", dbg_csc);
170 			DBG_DUMP_CSUM(&osblock, dbg_line, dbg_csp++);
171 		}
172 	}
173 #endif /* FS_DEBUG */
174 	DBG_PRINT0("fscs read\n");
175 
176 	/*
177 	 * Do all needed changes in the former last cylinder group.
178 	 */
179 	updjcg(osblock.fs_ncg - 1, modtime, fsi, fso, Nflag);
180 
181 	/*
182 	 * Dump out summary information about file system.
183 	 */
184 #ifdef FS_DEBUG
185 #define B2MBFACTOR (1 / (1024.0 * 1024.0))
186 	printf("growfs: %.1fMB (%jd sectors) block size %d, fragment size %d\n",
187 	    (float)sblock.fs_size * sblock.fs_fsize * B2MBFACTOR,
188 	    (intmax_t)fsbtodb(&sblock, sblock.fs_size), sblock.fs_bsize,
189 	    sblock.fs_fsize);
190 	printf("\tusing %d cylinder groups of %.2fMB, %d blks, %d inodes.\n",
191 	    sblock.fs_ncg, (float)sblock.fs_fpg * sblock.fs_fsize * B2MBFACTOR,
192 	    sblock.fs_fpg / sblock.fs_frag, sblock.fs_ipg);
193 	if (sblock.fs_flags & FS_DOSOFTDEP)
194 		printf("\twith soft updates\n");
195 #undef B2MBFACTOR
196 #endif /* FS_DEBUG */
197 
198 	/*
199 	 * Now build the cylinders group blocks and
200 	 * then print out indices of cylinder groups.
201 	 */
202 	printf("super-block backups (for fsck_ffs -b #) at:\n");
203 	i = 0;
204 	width = charsperline();
205 
206 	/*
207 	 * Iterate for only the new cylinder groups.
208 	 */
209 	for (cylno = osblock.fs_ncg; cylno < sblock.fs_ncg; cylno++) {
210 		initcg(cylno, modtime, fso, Nflag);
211 		j = sprintf(tmpbuf, " %jd%s",
212 		    (intmax_t)fsbtodb(&sblock, cgsblock(&sblock, cylno)),
213 		    cylno < (sblock.fs_ncg - 1) ? "," : "" );
214 		if (i + j >= width) {
215 			printf("\n");
216 			i = 0;
217 		}
218 		i += j;
219 		printf("%s", tmpbuf);
220 		fflush(stdout);
221 	}
222 	printf("\n");
223 
224 	/*
225 	 * Do all needed changes in the first cylinder group.
226 	 * allocate blocks in new location
227 	 */
228 	updcsloc(modtime, fsi, fso, Nflag);
229 
230 	/*
231 	 * Clean up the dynamic fields in our superblock.
232 	 *
233 	 * XXX
234 	 * The following fields are currently distributed from the superblock
235 	 * to the copies:
236 	 *     fs_minfree
237 	 *     fs_rotdelay
238 	 *     fs_maxcontig
239 	 *     fs_maxbpg
240 	 *     fs_minfree,
241 	 *     fs_optim
242 	 *     fs_flags
243 	 *
244 	 * We probably should rather change the summary for the cylinder group
245 	 * statistics here to the value of what would be in there, if the file
246 	 * system were created initially with the new size. Therefor we still
247 	 * need to find an easy way of calculating that.
248 	 * Possibly we can try to read the first superblock copy and apply the
249 	 * "diffed" stats between the old and new superblock by still copying
250 	 * certain parameters onto that.
251 	 */
252 	sblock.fs_time = modtime;
253 	sblock.fs_fmod = 0;
254 	sblock.fs_clean = 1;
255 	sblock.fs_ronly = 0;
256 	sblock.fs_cgrotor = 0;
257 	sblock.fs_state = 0;
258 	memset((void *)&sblock.fs_fsmnt, 0, sizeof(sblock.fs_fsmnt));
259 
260 	/*
261 	 * Now write the new superblock, its summary information,
262 	 * and all the alternates back to disk.
263 	 */
264 	if (!Nflag && sbput(fso, &sblock, sblock.fs_ncg) != 0)
265 		errc(2, EIO, "could not write updated superblock");
266 	DBG_PRINT0("fscs written\n");
267 
268 #ifdef FS_DEBUG
269 	{
270 		struct csum	*dbg_csp;
271 		u_int32_t	dbg_csc;
272 		char	dbg_line[80];
273 
274 		dbg_csp = fscs;
275 		for (dbg_csc = 0; dbg_csc < sblock.fs_ncg; dbg_csc++) {
276 			snprintf(dbg_line, sizeof(dbg_line),
277 			    "%d. new csum in new location", dbg_csc);
278 			DBG_DUMP_CSUM(&sblock, dbg_line, dbg_csp++);
279 		}
280 	}
281 #endif /* FS_DEBUG */
282 
283 	DBG_PRINT0("sblock written\n");
284 	DBG_DUMP_FS(&sblock, "new initial sblock");
285 
286 	DBG_PRINT0("sblock copies written\n");
287 	DBG_DUMP_FS(&sblock, "new other sblocks");
288 
289 	DBG_LEAVE;
290 	return;
291 }
292 
293 /*
294  * This creates a new cylinder group structure, for more details please see
295  * the source of newfs(8), as this function is taken over almost unchanged.
296  * As this is never called for the first cylinder group, the special
297  * provisions for that case are removed here.
298  */
299 static void
300 initcg(int cylno, time_t modtime, int fso, unsigned int Nflag)
301 {
302 	DBG_FUNC("initcg")
303 	static caddr_t iobuf;
304 	static long iobufsize;
305 	long blkno, start;
306 	ino_t ino;
307 	ufs2_daddr_t i, cbase, dmax;
308 	struct ufs1_dinode *dp1;
309 	struct ufs2_dinode *dp2;
310 	struct csum *cs;
311 	uint j, d, dupper, dlower;
312 
313 	if (iobuf == NULL) {
314 		iobufsize = 2 * sblock.fs_bsize;
315 		if ((iobuf = malloc(iobufsize)) == NULL)
316 			errx(37, "panic: cannot allocate I/O buffer");
317 		memset(iobuf, '\0', iobufsize);
318 	}
319 	/*
320 	 * Determine block bounds for cylinder group.
321 	 * Allow space for super block summary information in first
322 	 * cylinder group.
323 	 */
324 	cbase = cgbase(&sblock, cylno);
325 	dmax = cbase + sblock.fs_fpg;
326 	if (dmax > sblock.fs_size)
327 		dmax = sblock.fs_size;
328 	dlower = cgsblock(&sblock, cylno) - cbase;
329 	dupper = cgdmin(&sblock, cylno) - cbase;
330 	if (cylno == 0)	/* XXX fscs may be relocated */
331 		dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
332 	cs = &fscs[cylno];
333 	memset(&acg, 0, sblock.fs_cgsize);
334 	acg.cg_time = modtime;
335 	acg.cg_magic = CG_MAGIC;
336 	acg.cg_cgx = cylno;
337 	acg.cg_niblk = sblock.fs_ipg;
338 	acg.cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
339 	acg.cg_ndblk = dmax - cbase;
340 	if (sblock.fs_contigsumsize > 0)
341 		acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
342 	start = &acg.cg_space[0] - (u_char *)(&acg.cg_firstfield);
343 	if (sblock.fs_magic == FS_UFS2_MAGIC) {
344 		acg.cg_iusedoff = start;
345 	} else {
346 		acg.cg_old_ncyl = sblock.fs_old_cpg;
347 		acg.cg_old_time = acg.cg_time;
348 		acg.cg_time = 0;
349 		acg.cg_old_niblk = acg.cg_niblk;
350 		acg.cg_niblk = 0;
351 		acg.cg_initediblk = 0;
352 		acg.cg_old_btotoff = start;
353 		acg.cg_old_boff = acg.cg_old_btotoff +
354 		    sblock.fs_old_cpg * sizeof(int32_t);
355 		acg.cg_iusedoff = acg.cg_old_boff +
356 		    sblock.fs_old_cpg * sizeof(u_int16_t);
357 	}
358 	acg.cg_freeoff = acg.cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
359 	acg.cg_nextfreeoff = acg.cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT);
360 	if (sblock.fs_contigsumsize > 0) {
361 		acg.cg_clustersumoff =
362 		    roundup(acg.cg_nextfreeoff, sizeof(u_int32_t));
363 		acg.cg_clustersumoff -= sizeof(u_int32_t);
364 		acg.cg_clusteroff = acg.cg_clustersumoff +
365 		    (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
366 		acg.cg_nextfreeoff = acg.cg_clusteroff +
367 		    howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
368 	}
369 	if (acg.cg_nextfreeoff > (unsigned)sblock.fs_cgsize) {
370 		/*
371 		 * This should never happen as we would have had that panic
372 		 * already on file system creation
373 		 */
374 		errx(37, "panic: cylinder group too big");
375 	}
376 	acg.cg_cs.cs_nifree += sblock.fs_ipg;
377 	if (cylno == 0)
378 		for (ino = 0; ino < UFS_ROOTINO; ino++) {
379 			setbit(cg_inosused(&acg), ino);
380 			acg.cg_cs.cs_nifree--;
381 		}
382 	/*
383 	 * Initialize the initial inode blocks.
384 	 */
385 	dp1 = (struct ufs1_dinode *)(void *)iobuf;
386 	dp2 = (struct ufs2_dinode *)(void *)iobuf;
387 	for (i = 0; i < acg.cg_initediblk; i++) {
388 		if (sblock.fs_magic == FS_UFS1_MAGIC) {
389 			dp1->di_gen = arc4random();
390 			dp1++;
391 		} else {
392 			dp2->di_gen = arc4random();
393 			dp2++;
394 		}
395 	}
396 	wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno)), iobufsize, iobuf,
397 	    fso, Nflag);
398 	/*
399 	 * For the old file system, we have to initialize all the inodes.
400 	 */
401 	if (sblock.fs_magic == FS_UFS1_MAGIC &&
402 	    sblock.fs_ipg > 2 * INOPB(&sblock)) {
403 		for (i = 2 * sblock.fs_frag;
404 		     i < sblock.fs_ipg / INOPF(&sblock);
405 		     i += sblock.fs_frag) {
406 			dp1 = (struct ufs1_dinode *)(void *)iobuf;
407 			for (j = 0; j < INOPB(&sblock); j++) {
408 				dp1->di_gen = arc4random();
409 				dp1++;
410 			}
411 			wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno) + i),
412 			    sblock.fs_bsize, iobuf, fso, Nflag);
413 		}
414 	}
415 	if (cylno > 0) {
416 		/*
417 		 * In cylno 0, beginning space is reserved
418 		 * for boot and super blocks.
419 		 */
420 		for (d = 0; d < dlower; d += sblock.fs_frag) {
421 			blkno = d / sblock.fs_frag;
422 			setblock(&sblock, cg_blksfree(&acg), blkno);
423 			if (sblock.fs_contigsumsize > 0)
424 				setbit(cg_clustersfree(&acg), blkno);
425 			acg.cg_cs.cs_nbfree++;
426 		}
427 		sblock.fs_dsize += dlower;
428 	}
429 	sblock.fs_dsize += acg.cg_ndblk - dupper;
430 	if ((i = dupper % sblock.fs_frag)) {
431 		acg.cg_frsum[sblock.fs_frag - i]++;
432 		for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
433 			setbit(cg_blksfree(&acg), dupper);
434 			acg.cg_cs.cs_nffree++;
435 		}
436 	}
437 	for (d = dupper; d + sblock.fs_frag <= acg.cg_ndblk;
438 	    d += sblock.fs_frag) {
439 		blkno = d / sblock.fs_frag;
440 		setblock(&sblock, cg_blksfree(&acg), blkno);
441 		if (sblock.fs_contigsumsize > 0)
442 			setbit(cg_clustersfree(&acg), blkno);
443 		acg.cg_cs.cs_nbfree++;
444 	}
445 	if (d < acg.cg_ndblk) {
446 		acg.cg_frsum[acg.cg_ndblk - d]++;
447 		for (; d < acg.cg_ndblk; d++) {
448 			setbit(cg_blksfree(&acg), d);
449 			acg.cg_cs.cs_nffree++;
450 		}
451 	}
452 	if (sblock.fs_contigsumsize > 0) {
453 		int32_t *sump = cg_clustersum(&acg);
454 		u_char *mapp = cg_clustersfree(&acg);
455 		int map = *mapp++;
456 		int bit = 1;
457 		int run = 0;
458 
459 		for (i = 0; i < acg.cg_nclusterblks; i++) {
460 			if ((map & bit) != 0)
461 				run++;
462 			else if (run != 0) {
463 				if (run > sblock.fs_contigsumsize)
464 					run = sblock.fs_contigsumsize;
465 				sump[run]++;
466 				run = 0;
467 			}
468 			if ((i & (CHAR_BIT - 1)) != CHAR_BIT - 1)
469 				bit <<= 1;
470 			else {
471 				map = *mapp++;
472 				bit = 1;
473 			}
474 		}
475 		if (run != 0) {
476 			if (run > sblock.fs_contigsumsize)
477 				run = sblock.fs_contigsumsize;
478 			sump[run]++;
479 		}
480 	}
481 	sblock.fs_cstotal.cs_ndir += acg.cg_cs.cs_ndir;
482 	sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
483 	sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
484 	sblock.fs_cstotal.cs_nifree += acg.cg_cs.cs_nifree;
485 	*cs = acg.cg_cs;
486 
487 	cgckhash(&acg);
488 	wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), sblock.fs_cgsize, &acg,
489 	    fso, Nflag);
490 	DBG_DUMP_CG(&sblock, "new cg", &acg);
491 
492 	DBG_LEAVE;
493 	return;
494 }
495 
496 /*
497  * Here we add or subtract (sign +1/-1) the available fragments in a given
498  * block to or from the fragment statistics. By subtracting before and adding
499  * after an operation on the free frag map we can easy update the fragment
500  * statistic, which seems to be otherwise a rather complex operation.
501  */
502 static void
503 frag_adjust(ufs2_daddr_t frag, int sign)
504 {
505 	DBG_FUNC("frag_adjust")
506 	int fragsize;
507 	int f;
508 
509 	DBG_ENTER;
510 
511 	fragsize = 0;
512 	/*
513 	 * Here frag only needs to point to any fragment in the block we want
514 	 * to examine.
515 	 */
516 	for (f = rounddown(frag, sblock.fs_frag);
517 	    f < roundup(frag + 1, sblock.fs_frag); f++) {
518 		/*
519 		 * Count contiguous free fragments.
520 		 */
521 		if (isset(cg_blksfree(&acg), f)) {
522 			fragsize++;
523 		} else {
524 			if (fragsize && fragsize < sblock.fs_frag) {
525 				/*
526 				 * We found something in between.
527 				 */
528 				acg.cg_frsum[fragsize] += sign;
529 				DBG_PRINT2("frag_adjust [%d]+=%d\n",
530 				    fragsize, sign);
531 			}
532 			fragsize = 0;
533 		}
534 	}
535 	if (fragsize && fragsize < sblock.fs_frag) {
536 		/*
537 		 * We found something.
538 		 */
539 		acg.cg_frsum[fragsize] += sign;
540 		DBG_PRINT2("frag_adjust [%d]+=%d\n", fragsize, sign);
541 	}
542 	DBG_PRINT2("frag_adjust [[%d]]+=%d\n", fragsize, sign);
543 
544 	DBG_LEAVE;
545 	return;
546 }
547 
548 /*
549  * Here we do all needed work for the former last cylinder group. It has to be
550  * changed in any case, even if the file system ended exactly on the end of
551  * this group, as there is some slightly inconsistent handling of the number
552  * of cylinders in the cylinder group. We start again by reading the cylinder
553  * group from disk. If the last block was not fully available, we first handle
554  * the missing fragments, then we handle all new full blocks in that file
555  * system and finally we handle the new last fragmented block in the file
556  * system.  We again have to handle the fragment statistics rotational layout
557  * tables and cluster summary during all those operations.
558  */
559 static void
560 updjcg(int cylno, time_t modtime, int fsi, int fso, unsigned int Nflag)
561 {
562 	DBG_FUNC("updjcg")
563 	ufs2_daddr_t cbase, dmax, dupper;
564 	struct csum *cs;
565 	int i, k;
566 	int j = 0;
567 
568 	DBG_ENTER;
569 
570 	/*
571 	 * Read the former last (joining) cylinder group from disk, and make
572 	 * a copy.
573 	 */
574 	rdfs(fsbtodb(&osblock, cgtod(&osblock, cylno)),
575 	    (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
576 	DBG_PRINT0("jcg read\n");
577 	DBG_DUMP_CG(&sblock, "old joining cg", &aocg);
578 
579 	memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
580 
581 	/*
582 	 * If the cylinder group had already its new final size almost
583 	 * nothing is to be done ... except:
584 	 * For some reason the value of cg_ncyl in the last cylinder group has
585 	 * to be zero instead of fs_cpg. As this is now no longer the last
586 	 * cylinder group we have to change that value now to fs_cpg.
587 	 */
588 
589 	if (cgbase(&osblock, cylno + 1) == osblock.fs_size) {
590 		if (sblock.fs_magic == FS_UFS1_MAGIC)
591 			acg.cg_old_ncyl = sblock.fs_old_cpg;
592 
593 		cgckhash(&acg);
594 		wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
595 		    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
596 		DBG_PRINT0("jcg written\n");
597 		DBG_DUMP_CG(&sblock, "new joining cg", &acg);
598 
599 		DBG_LEAVE;
600 		return;
601 	}
602 
603 	/*
604 	 * Set up some variables needed later.
605 	 */
606 	cbase = cgbase(&sblock, cylno);
607 	dmax = cbase + sblock.fs_fpg;
608 	if (dmax > sblock.fs_size)
609 		dmax = sblock.fs_size;
610 	dupper = cgdmin(&sblock, cylno) - cbase;
611 	if (cylno == 0) /* XXX fscs may be relocated */
612 		dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
613 
614 	/*
615 	 * Set pointer to the cylinder summary for our cylinder group.
616 	 */
617 	cs = fscs + cylno;
618 
619 	/*
620 	 * Touch the cylinder group, update all fields in the cylinder group as
621 	 * needed, update the free space in the superblock.
622 	 */
623 	acg.cg_time = modtime;
624 	if ((unsigned)cylno == sblock.fs_ncg - 1) {
625 		/*
626 		 * This is still the last cylinder group.
627 		 */
628 		if (sblock.fs_magic == FS_UFS1_MAGIC)
629 			acg.cg_old_ncyl =
630 			    sblock.fs_old_ncyl % sblock.fs_old_cpg;
631 	} else {
632 		acg.cg_old_ncyl = sblock.fs_old_cpg;
633 	}
634 	DBG_PRINT2("jcg dbg: %d %u", cylno, sblock.fs_ncg);
635 #ifdef FS_DEBUG
636 	if (sblock.fs_magic == FS_UFS1_MAGIC)
637 		DBG_PRINT2("%d %u", acg.cg_old_ncyl, sblock.fs_old_cpg);
638 #endif
639 	DBG_PRINT0("\n");
640 	acg.cg_ndblk = dmax - cbase;
641 	sblock.fs_dsize += acg.cg_ndblk - aocg.cg_ndblk;
642 	if (sblock.fs_contigsumsize > 0)
643 		acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
644 
645 	/*
646 	 * Now we have to update the free fragment bitmap for our new free
647 	 * space.  There again we have to handle the fragmentation and also
648 	 * the rotational layout tables and the cluster summary.  This is
649 	 * also done per fragment for the first new block if the old file
650 	 * system end was not on a block boundary, per fragment for the new
651 	 * last block if the new file system end is not on a block boundary,
652 	 * and per block for all space in between.
653 	 *
654 	 * Handle the first new block here if it was partially available
655 	 * before.
656 	 */
657 	if (osblock.fs_size % sblock.fs_frag) {
658 		if (roundup(osblock.fs_size, sblock.fs_frag) <=
659 		    sblock.fs_size) {
660 			/*
661 			 * The new space is enough to fill at least this
662 			 * block
663 			 */
664 			j = 0;
665 			for (i = roundup(osblock.fs_size - cbase,
666 			    sblock.fs_frag) - 1; i >= osblock.fs_size - cbase;
667 			    i--) {
668 				setbit(cg_blksfree(&acg), i);
669 				acg.cg_cs.cs_nffree++;
670 				j++;
671 			}
672 
673 			/*
674 			 * Check if the fragment just created could join an
675 			 * already existing fragment at the former end of the
676 			 * file system.
677 			 */
678 			if (isblock(&sblock, cg_blksfree(&acg),
679 			    ((osblock.fs_size - cgbase(&sblock, cylno)) /
680 			     sblock.fs_frag))) {
681 				/*
682 				 * The block is now completely available.
683 				 */
684 				DBG_PRINT0("block was\n");
685 				acg.cg_frsum[osblock.fs_size % sblock.fs_frag]--;
686 				acg.cg_cs.cs_nbfree++;
687 				acg.cg_cs.cs_nffree -= sblock.fs_frag;
688 				k = rounddown(osblock.fs_size - cbase,
689 				    sblock.fs_frag);
690 				updclst((osblock.fs_size - cbase) /
691 				    sblock.fs_frag);
692 			} else {
693 				/*
694 				 * Lets rejoin a possible partially growed
695 				 * fragment.
696 				 */
697 				k = 0;
698 				while (isset(cg_blksfree(&acg), i) &&
699 				    (i >= rounddown(osblock.fs_size - cbase,
700 				    sblock.fs_frag))) {
701 					i--;
702 					k++;
703 				}
704 				if (k)
705 					acg.cg_frsum[k]--;
706 				acg.cg_frsum[k + j]++;
707 			}
708 		} else {
709 			/*
710 			 * We only grow by some fragments within this last
711 			 * block.
712 			 */
713 			for (i = sblock.fs_size - cbase - 1;
714 			    i >= osblock.fs_size - cbase; i--) {
715 				setbit(cg_blksfree(&acg), i);
716 				acg.cg_cs.cs_nffree++;
717 				j++;
718 			}
719 			/*
720 			 * Lets rejoin a possible partially growed fragment.
721 			 */
722 			k = 0;
723 			while (isset(cg_blksfree(&acg), i) &&
724 			    (i >= rounddown(osblock.fs_size - cbase,
725 			    sblock.fs_frag))) {
726 				i--;
727 				k++;
728 			}
729 			if (k)
730 				acg.cg_frsum[k]--;
731 			acg.cg_frsum[k + j]++;
732 		}
733 	}
734 
735 	/*
736 	 * Handle all new complete blocks here.
737 	 */
738 	for (i = roundup(osblock.fs_size - cbase, sblock.fs_frag);
739 	    i + sblock.fs_frag <= dmax - cbase;	/* XXX <= or only < ? */
740 	    i += sblock.fs_frag) {
741 		j = i / sblock.fs_frag;
742 		setblock(&sblock, cg_blksfree(&acg), j);
743 		updclst(j);
744 		acg.cg_cs.cs_nbfree++;
745 	}
746 
747 	/*
748 	 * Handle the last new block if there are stll some new fragments left.
749 	 * Here we don't have to bother about the cluster summary or the even
750 	 * the rotational layout table.
751 	 */
752 	if (i < (dmax - cbase)) {
753 		acg.cg_frsum[dmax - cbase - i]++;
754 		for (; i < dmax - cbase; i++) {
755 			setbit(cg_blksfree(&acg), i);
756 			acg.cg_cs.cs_nffree++;
757 		}
758 	}
759 
760 	sblock.fs_cstotal.cs_nffree +=
761 	    (acg.cg_cs.cs_nffree - aocg.cg_cs.cs_nffree);
762 	sblock.fs_cstotal.cs_nbfree +=
763 	    (acg.cg_cs.cs_nbfree - aocg.cg_cs.cs_nbfree);
764 	/*
765 	 * The following statistics are not changed here:
766 	 *     sblock.fs_cstotal.cs_ndir
767 	 *     sblock.fs_cstotal.cs_nifree
768 	 * As the statistics for this cylinder group are ready, copy it to
769 	 * the summary information array.
770 	 */
771 	*cs = acg.cg_cs;
772 
773 	/*
774 	 * Write the updated "joining" cylinder group back to disk.
775 	 */
776 	cgckhash(&acg);
777 	wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize,
778 	    (void *)&acg, fso, Nflag);
779 	DBG_PRINT0("jcg written\n");
780 	DBG_DUMP_CG(&sblock, "new joining cg", &acg);
781 
782 	DBG_LEAVE;
783 	return;
784 }
785 
786 /*
787  * Here we update the location of the cylinder summary. We have two possible
788  * ways of growing the cylinder summary:
789  * (1)	We can try to grow the summary in the current location, and relocate
790  *	possibly used blocks within the current cylinder group.
791  * (2)	Alternatively we can relocate the whole cylinder summary to the first
792  *	new completely empty cylinder group. Once the cylinder summary is no
793  *	longer in the beginning of the first cylinder group you should never
794  *	use a version of fsck which is not aware of the possibility to have
795  *	this structure in a non standard place.
796  * Option (2) is considered to be less intrusive to the structure of the file-
797  * system, so that's the one being used.
798  */
799 static void
800 updcsloc(time_t modtime, int fsi, int fso, unsigned int Nflag)
801 {
802 	DBG_FUNC("updcsloc")
803 	struct csum *cs;
804 	int ocscg, ncscg;
805 	ufs2_daddr_t d;
806 	int lcs = 0;
807 	int block;
808 
809 	DBG_ENTER;
810 
811 	if (howmany(sblock.fs_cssize, sblock.fs_fsize) ==
812 	    howmany(osblock.fs_cssize, osblock.fs_fsize)) {
813 		/*
814 		 * No new fragment needed.
815 		 */
816 		DBG_LEAVE;
817 		return;
818 	}
819 	ocscg = dtog(&osblock, osblock.fs_csaddr);
820 	cs = fscs + ocscg;
821 
822 	/*
823 	 * Read original cylinder group from disk, and make a copy.
824 	 * XXX	If Nflag is set in some very rare cases we now miss
825 	 *	some changes done in updjcg by reading the unmodified
826 	 *	block from disk.
827 	 */
828 	rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)),
829 	    (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
830 	DBG_PRINT0("oscg read\n");
831 	DBG_DUMP_CG(&sblock, "old summary cg", &aocg);
832 
833 	memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
834 
835 	/*
836 	 * Touch the cylinder group, set up local variables needed later
837 	 * and update the superblock.
838 	 */
839 	acg.cg_time = modtime;
840 
841 	/*
842 	 * XXX	In the case of having active snapshots we may need much more
843 	 *	blocks for the copy on write. We need each block twice, and
844 	 *	also up to 8*3 blocks for indirect blocks for all possible
845 	 *	references.
846 	 */
847 	/*
848 	 * There is not enough space in the old cylinder group to
849 	 * relocate all blocks as needed, so we relocate the whole
850 	 * cylinder group summary to a new group. We try to use the
851 	 * first complete new cylinder group just created. Within the
852 	 * cylinder group we align the area immediately after the
853 	 * cylinder group information location in order to be as
854 	 * close as possible to the original implementation of ffs.
855 	 *
856 	 * First we have to make sure we'll find enough space in the
857 	 * new cylinder group. If not, then we currently give up.
858 	 * We start with freeing everything which was used by the
859 	 * fragments of the old cylinder summary in the current group.
860 	 * Now we write back the group meta data, read in the needed
861 	 * meta data from the new cylinder group, and start allocating
862 	 * within that group. Here we can assume, the group to be
863 	 * completely empty. Which makes the handling of fragments and
864 	 * clusters a lot easier.
865 	 */
866 	DBG_TRC;
867 	if (sblock.fs_ncg - osblock.fs_ncg < 2)
868 		errx(2, "panic: not enough space");
869 
870 	/*
871 	 * Point "d" to the first fragment not used by the cylinder
872 	 * summary.
873 	 */
874 	d = osblock.fs_csaddr + (osblock.fs_cssize / osblock.fs_fsize);
875 
876 	/*
877 	 * Set up last cluster size ("lcs") already here. Calculate
878 	 * the size for the trailing cluster just behind where "d"
879 	 * points to.
880 	 */
881 	if (sblock.fs_contigsumsize > 0) {
882 		for (block = howmany(d % sblock.fs_fpg, sblock.fs_frag),
883 		    lcs = 0; lcs < sblock.fs_contigsumsize; block++, lcs++) {
884 			if (isclr(cg_clustersfree(&acg), block))
885 				break;
886 		}
887 	}
888 
889 	/*
890 	 * Point "d" to the last frag used by the cylinder summary.
891 	 */
892 	d--;
893 
894 	DBG_PRINT1("d=%jd\n", (intmax_t)d);
895 	if ((d + 1) % sblock.fs_frag) {
896 		/*
897 		 * The end of the cylinder summary is not a complete
898 		 * block.
899 		 */
900 		DBG_TRC;
901 		frag_adjust(d % sblock.fs_fpg, -1);
902 		for (; (d + 1) % sblock.fs_frag; d--) {
903 			DBG_PRINT1("d=%jd\n", (intmax_t)d);
904 			setbit(cg_blksfree(&acg), d % sblock.fs_fpg);
905 			acg.cg_cs.cs_nffree++;
906 			sblock.fs_cstotal.cs_nffree++;
907 		}
908 		/*
909 		 * Point "d" to the last fragment of the last
910 		 * (incomplete) block of the cylinder summary.
911 		 */
912 		d++;
913 		frag_adjust(d % sblock.fs_fpg, 1);
914 
915 		if (isblock(&sblock, cg_blksfree(&acg),
916 		    (d % sblock.fs_fpg) / sblock.fs_frag)) {
917 			DBG_PRINT1("d=%jd\n", (intmax_t)d);
918 			acg.cg_cs.cs_nffree -= sblock.fs_frag;
919 			acg.cg_cs.cs_nbfree++;
920 			sblock.fs_cstotal.cs_nffree -= sblock.fs_frag;
921 			sblock.fs_cstotal.cs_nbfree++;
922 			if (sblock.fs_contigsumsize > 0) {
923 				setbit(cg_clustersfree(&acg),
924 				    (d % sblock.fs_fpg) / sblock.fs_frag);
925 				if (lcs < sblock.fs_contigsumsize) {
926 					if (lcs)
927 						cg_clustersum(&acg)[lcs]--;
928 					lcs++;
929 					cg_clustersum(&acg)[lcs]++;
930 				}
931 			}
932 		}
933 		/*
934 		 * Point "d" to the first fragment of the block before
935 		 * the last incomplete block.
936 		 */
937 		d--;
938 	}
939 
940 	DBG_PRINT1("d=%jd\n", (intmax_t)d);
941 	for (d = rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr;
942 	    d -= sblock.fs_frag) {
943 		DBG_TRC;
944 		DBG_PRINT1("d=%jd\n", (intmax_t)d);
945 		setblock(&sblock, cg_blksfree(&acg),
946 		    (d % sblock.fs_fpg) / sblock.fs_frag);
947 		acg.cg_cs.cs_nbfree++;
948 		sblock.fs_cstotal.cs_nbfree++;
949 		if (sblock.fs_contigsumsize > 0) {
950 			setbit(cg_clustersfree(&acg),
951 			    (d % sblock.fs_fpg) / sblock.fs_frag);
952 			/*
953 			 * The last cluster size is already set up.
954 			 */
955 			if (lcs < sblock.fs_contigsumsize) {
956 				if (lcs)
957 					cg_clustersum(&acg)[lcs]--;
958 				lcs++;
959 				cg_clustersum(&acg)[lcs]++;
960 			}
961 		}
962 	}
963 	*cs = acg.cg_cs;
964 
965 	/*
966 	 * Now write the former cylinder group containing the cylinder
967 	 * summary back to disk.
968 	 */
969 	cgckhash(&acg);
970 	wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)),
971 	    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
972 	DBG_PRINT0("oscg written\n");
973 	DBG_DUMP_CG(&sblock, "old summary cg", &acg);
974 
975 	/*
976 	 * Find the beginning of the new cylinder group containing the
977 	 * cylinder summary.
978 	 */
979 	sblock.fs_csaddr = cgdmin(&sblock, osblock.fs_ncg);
980 	ncscg = dtog(&sblock, sblock.fs_csaddr);
981 	cs = fscs + ncscg;
982 
983 	/*
984 	 * If Nflag is specified, we would now read random data instead
985 	 * of an empty cg structure from disk. So we can't simulate that
986 	 * part for now.
987 	 */
988 	if (Nflag) {
989 		DBG_PRINT0("nscg update skipped\n");
990 		DBG_LEAVE;
991 		return;
992 	}
993 
994 	/*
995 	 * Read the future cylinder group containing the cylinder
996 	 * summary from disk, and make a copy.
997 	 */
998 	rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
999 	    (size_t)sblock.fs_cgsize, (void *)&aocg, fsi);
1000 	DBG_PRINT0("nscg read\n");
1001 	DBG_DUMP_CG(&sblock, "new summary cg", &aocg);
1002 
1003 	memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
1004 
1005 	/*
1006 	 * Allocate all complete blocks used by the new cylinder
1007 	 * summary.
1008 	 */
1009 	for (d = sblock.fs_csaddr; d + sblock.fs_frag <=
1010 	    sblock.fs_csaddr + (sblock.fs_cssize / sblock.fs_fsize);
1011 	    d += sblock.fs_frag) {
1012 		clrblock(&sblock, cg_blksfree(&acg),
1013 		    (d % sblock.fs_fpg) / sblock.fs_frag);
1014 		acg.cg_cs.cs_nbfree--;
1015 		sblock.fs_cstotal.cs_nbfree--;
1016 		if (sblock.fs_contigsumsize > 0) {
1017 			clrbit(cg_clustersfree(&acg),
1018 			    (d % sblock.fs_fpg) / sblock.fs_frag);
1019 		}
1020 	}
1021 
1022 	/*
1023 	 * Allocate all fragments used by the cylinder summary in the
1024 	 * last block.
1025 	 */
1026 	if (d < sblock.fs_csaddr + (sblock.fs_cssize / sblock.fs_fsize)) {
1027 		for (; d - sblock.fs_csaddr <
1028 		    sblock.fs_cssize/sblock.fs_fsize; d++) {
1029 			clrbit(cg_blksfree(&acg), d % sblock.fs_fpg);
1030 			acg.cg_cs.cs_nffree--;
1031 			sblock.fs_cstotal.cs_nffree--;
1032 		}
1033 		acg.cg_cs.cs_nbfree--;
1034 		acg.cg_cs.cs_nffree += sblock.fs_frag;
1035 		sblock.fs_cstotal.cs_nbfree--;
1036 		sblock.fs_cstotal.cs_nffree += sblock.fs_frag;
1037 		if (sblock.fs_contigsumsize > 0)
1038 			clrbit(cg_clustersfree(&acg),
1039 			    (d % sblock.fs_fpg) / sblock.fs_frag);
1040 
1041 		frag_adjust(d % sblock.fs_fpg, 1);
1042 	}
1043 	/*
1044 	 * XXX	Handle the cluster statistics here in the case this
1045 	 *	cylinder group is now almost full, and the remaining
1046 	 *	space is less then the maximum cluster size. This is
1047 	 *	probably not needed, as you would hardly find a file
1048 	 *	system which has only MAXCSBUFS+FS_MAXCONTIG of free
1049 	 *	space right behind the cylinder group information in
1050 	 *	any new cylinder group.
1051 	 */
1052 
1053 	/*
1054 	 * Update our statistics in the cylinder summary.
1055 	 */
1056 	*cs = acg.cg_cs;
1057 
1058 	/*
1059 	 * Write the new cylinder group containing the cylinder summary
1060 	 * back to disk.
1061 	 */
1062 	cgckhash(&acg);
1063 	wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1064 	    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1065 	DBG_PRINT0("nscg written\n");
1066 	DBG_DUMP_CG(&sblock, "new summary cg", &acg);
1067 
1068 	DBG_LEAVE;
1069 	return;
1070 }
1071 
1072 /*
1073  * Here we read some block(s) from disk.
1074  */
1075 static void
1076 rdfs(ufs2_daddr_t bno, size_t size, void *bf, int fsi)
1077 {
1078 	DBG_FUNC("rdfs")
1079 	ssize_t	n;
1080 
1081 	DBG_ENTER;
1082 
1083 	if (bno < 0)
1084 		err(32, "rdfs: attempting to read negative block number");
1085 	if (lseek(fsi, (off_t)bno * DEV_BSIZE, 0) < 0)
1086 		err(33, "rdfs: seek error: %jd", (intmax_t)bno);
1087 	n = read(fsi, bf, size);
1088 	if (n != (ssize_t)size)
1089 		err(34, "rdfs: read error: %jd", (intmax_t)bno);
1090 
1091 	DBG_LEAVE;
1092 	return;
1093 }
1094 
1095 /*
1096  * Here we write some block(s) to disk.
1097  */
1098 static void
1099 wtfs(ufs2_daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag)
1100 {
1101 	DBG_FUNC("wtfs")
1102 	ssize_t	n;
1103 
1104 	DBG_ENTER;
1105 
1106 	if (Nflag) {
1107 		DBG_LEAVE;
1108 		return;
1109 	}
1110 	if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0)
1111 		err(35, "wtfs: seek error: %ld", (long)bno);
1112 	n = write(fso, bf, size);
1113 	if (n != (ssize_t)size)
1114 		err(36, "wtfs: write error: %ld", (long)bno);
1115 
1116 	DBG_LEAVE;
1117 	return;
1118 }
1119 
1120 /*
1121  * Here we check if all frags of a block are free. For more details again
1122  * please see the source of newfs(8), as this function is taken over almost
1123  * unchanged.
1124  */
1125 static int
1126 isblock(struct fs *fs, unsigned char *cp, int h)
1127 {
1128 	DBG_FUNC("isblock")
1129 	unsigned char mask;
1130 
1131 	DBG_ENTER;
1132 
1133 	switch (fs->fs_frag) {
1134 	case 8:
1135 		DBG_LEAVE;
1136 		return (cp[h] == 0xff);
1137 	case 4:
1138 		mask = 0x0f << ((h & 0x1) << 2);
1139 		DBG_LEAVE;
1140 		return ((cp[h >> 1] & mask) == mask);
1141 	case 2:
1142 		mask = 0x03 << ((h & 0x3) << 1);
1143 		DBG_LEAVE;
1144 		return ((cp[h >> 2] & mask) == mask);
1145 	case 1:
1146 		mask = 0x01 << (h & 0x7);
1147 		DBG_LEAVE;
1148 		return ((cp[h >> 3] & mask) == mask);
1149 	default:
1150 		fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
1151 		DBG_LEAVE;
1152 		return (0);
1153 	}
1154 }
1155 
1156 /*
1157  * Here we allocate a complete block in the block map. For more details again
1158  * please see the source of newfs(8), as this function is taken over almost
1159  * unchanged.
1160  */
1161 static void
1162 clrblock(struct fs *fs, unsigned char *cp, int h)
1163 {
1164 	DBG_FUNC("clrblock")
1165 
1166 	DBG_ENTER;
1167 
1168 	switch ((fs)->fs_frag) {
1169 	case 8:
1170 		cp[h] = 0;
1171 		break;
1172 	case 4:
1173 		cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1174 		break;
1175 	case 2:
1176 		cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1177 		break;
1178 	case 1:
1179 		cp[h >> 3] &= ~(0x01 << (h & 0x7));
1180 		break;
1181 	default:
1182 		warnx("clrblock bad fs_frag %d", fs->fs_frag);
1183 		break;
1184 	}
1185 
1186 	DBG_LEAVE;
1187 	return;
1188 }
1189 
1190 /*
1191  * Here we free a complete block in the free block map. For more details again
1192  * please see the source of newfs(8), as this function is taken over almost
1193  * unchanged.
1194  */
1195 static void
1196 setblock(struct fs *fs, unsigned char *cp, int h)
1197 {
1198 	DBG_FUNC("setblock")
1199 
1200 	DBG_ENTER;
1201 
1202 	switch (fs->fs_frag) {
1203 	case 8:
1204 		cp[h] = 0xff;
1205 		break;
1206 	case 4:
1207 		cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1208 		break;
1209 	case 2:
1210 		cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1211 		break;
1212 	case 1:
1213 		cp[h >> 3] |= (0x01 << (h & 0x7));
1214 		break;
1215 	default:
1216 		warnx("setblock bad fs_frag %d", fs->fs_frag);
1217 		break;
1218 	}
1219 
1220 	DBG_LEAVE;
1221 	return;
1222 }
1223 
1224 /*
1225  * Figure out how many lines our current terminal has. For more details again
1226  * please see the source of newfs(8), as this function is taken over almost
1227  * unchanged.
1228  */
1229 static int
1230 charsperline(void)
1231 {
1232 	DBG_FUNC("charsperline")
1233 	int columns;
1234 	char *cp;
1235 	struct winsize ws;
1236 
1237 	DBG_ENTER;
1238 
1239 	columns = 0;
1240 	if (ioctl(0, TIOCGWINSZ, &ws) != -1)
1241 		columns = ws.ws_col;
1242 	if (columns == 0 && (cp = getenv("COLUMNS")))
1243 		columns = atoi(cp);
1244 	if (columns == 0)
1245 		columns = 80;	/* last resort */
1246 
1247 	DBG_LEAVE;
1248 	return (columns);
1249 }
1250 
1251 static int
1252 is_dev(const char *name)
1253 {
1254 	struct stat devstat;
1255 
1256 	if (stat(name, &devstat) != 0)
1257 		return (0);
1258 	if (!S_ISCHR(devstat.st_mode))
1259 		return (0);
1260 	return (1);
1261 }
1262 
1263 /*
1264  * Return mountpoint on which the device is currently mounted.
1265  */
1266 static const struct statfs *
1267 dev_to_statfs(const char *dev)
1268 {
1269 	struct stat devstat, mntdevstat;
1270 	struct statfs *mntbuf, *statfsp;
1271 	char device[MAXPATHLEN];
1272 	char *mntdevname;
1273 	int i, mntsize;
1274 
1275 	/*
1276 	 * First check the mounted filesystems.
1277 	 */
1278 	if (stat(dev, &devstat) != 0)
1279 		return (NULL);
1280 	if (!S_ISCHR(devstat.st_mode) && !S_ISBLK(devstat.st_mode))
1281 		return (NULL);
1282 
1283 	mntsize = getmntinfo(&mntbuf, MNT_NOWAIT);
1284 	for (i = 0; i < mntsize; i++) {
1285 		statfsp = &mntbuf[i];
1286 		mntdevname = statfsp->f_mntfromname;
1287 		if (*mntdevname != '/') {
1288 			strcpy(device, _PATH_DEV);
1289 			strcat(device, mntdevname);
1290 			mntdevname = device;
1291 		}
1292 		if (stat(mntdevname, &mntdevstat) == 0 &&
1293 		    mntdevstat.st_rdev == devstat.st_rdev)
1294 			return (statfsp);
1295 	}
1296 
1297 	return (NULL);
1298 }
1299 
1300 static const char *
1301 mountpoint_to_dev(const char *mountpoint)
1302 {
1303 	struct statfs *mntbuf, *statfsp;
1304 	struct fstab *fs;
1305 	int i, mntsize;
1306 
1307 	/*
1308 	 * First check the mounted filesystems.
1309 	 */
1310 	mntsize = getmntinfo(&mntbuf, MNT_NOWAIT);
1311 	for (i = 0; i < mntsize; i++) {
1312 		statfsp = &mntbuf[i];
1313 
1314 		if (strcmp(statfsp->f_mntonname, mountpoint) == 0)
1315 			return (statfsp->f_mntfromname);
1316 	}
1317 
1318 	/*
1319 	 * Check the fstab.
1320 	 */
1321 	fs = getfsfile(mountpoint);
1322 	if (fs != NULL)
1323 		return (fs->fs_spec);
1324 
1325 	return (NULL);
1326 }
1327 
1328 static const char *
1329 getdev(const char *name)
1330 {
1331 	static char device[MAXPATHLEN];
1332 	const char *cp, *dev;
1333 
1334 	if (is_dev(name))
1335 		return (name);
1336 
1337 	cp = strrchr(name, '/');
1338 	if (cp == NULL) {
1339 		snprintf(device, sizeof(device), "%s%s", _PATH_DEV, name);
1340 		if (is_dev(device))
1341 			return (device);
1342 	}
1343 
1344 	dev = mountpoint_to_dev(name);
1345 	if (dev != NULL && is_dev(dev))
1346 		return (dev);
1347 
1348 	return (NULL);
1349 }
1350 
1351 /*
1352  * growfs(8) is a utility which allows to increase the size of an existing
1353  * ufs file system. Currently this can only be done on unmounted file system.
1354  * It recognizes some command line options to specify the new desired size,
1355  * and it does some basic checkings. The old file system size is determined
1356  * and after some more checks like we can really access the new last block
1357  * on the disk etc. we calculate the new parameters for the superblock. After
1358  * having done this we just call growfs() which will do the work.
1359  * We still have to provide support for snapshots. Therefore we first have to
1360  * understand what data structures are always replicated in the snapshot on
1361  * creation, for all other blocks we touch during our procedure, we have to
1362  * keep the old blocks unchanged somewhere available for the snapshots. If we
1363  * are lucky, then we only have to handle our blocks to be relocated in that
1364  * way.
1365  * Also we have to consider in what order we actually update the critical
1366  * data structures of the file system to make sure, that in case of a disaster
1367  * fsck(8) is still able to restore any lost data.
1368  * The foreseen last step then will be to provide for growing even mounted
1369  * file systems. There we have to extend the mount() system call to provide
1370  * userland access to the file system locking facility.
1371  */
1372 int
1373 main(int argc, char **argv)
1374 {
1375 	DBG_FUNC("main")
1376 	struct fs *fs;
1377 	const char *device;
1378 	const struct statfs *statfsp;
1379 	uint64_t size = 0;
1380 	off_t mediasize;
1381 	int error, j, fsi, fso, ch, ret, Nflag = 0, yflag = 0;
1382 	char *p, reply[5], oldsizebuf[6], newsizebuf[6];
1383 	void *testbuf;
1384 
1385 	DBG_ENTER;
1386 
1387 	while ((ch = getopt(argc, argv, "Ns:vy")) != -1) {
1388 		switch(ch) {
1389 		case 'N':
1390 			Nflag = 1;
1391 			break;
1392 		case 's':
1393 			size = (off_t)strtoumax(optarg, &p, 0);
1394 			if (p == NULL || *p == '\0')
1395 				size *= DEV_BSIZE;
1396 			else if (*p == 'b' || *p == 'B')
1397 				; /* do nothing */
1398 			else if (*p == 'k' || *p == 'K')
1399 				size <<= 10;
1400 			else if (*p == 'm' || *p == 'M')
1401 				size <<= 20;
1402 			else if (*p == 'g' || *p == 'G')
1403 				size <<= 30;
1404 			else if (*p == 't' || *p == 'T') {
1405 				size <<= 30;
1406 				size <<= 10;
1407 			} else
1408 				errx(1, "unknown suffix on -s argument");
1409 			break;
1410 		case 'v': /* for compatibility to newfs */
1411 			break;
1412 		case 'y':
1413 			yflag = 1;
1414 			break;
1415 		case '?':
1416 			/* FALLTHROUGH */
1417 		default:
1418 			usage();
1419 		}
1420 	}
1421 	argc -= optind;
1422 	argv += optind;
1423 
1424 	if (argc != 1)
1425 		usage();
1426 
1427 	/*
1428 	 * Now try to guess the device name.
1429 	 */
1430 	device = getdev(*argv);
1431 	if (device == NULL)
1432 		errx(1, "cannot find special device for %s", *argv);
1433 
1434 	statfsp = dev_to_statfs(device);
1435 
1436 	fsi = open(device, O_RDONLY);
1437 	if (fsi < 0)
1438 		err(1, "%s", device);
1439 
1440 	/*
1441 	 * Try to guess the slice size if not specified.
1442 	 */
1443 	if (ioctl(fsi, DIOCGMEDIASIZE, &mediasize) == -1)
1444 		err(1,"DIOCGMEDIASIZE");
1445 
1446 	/*
1447 	 * Check if that partition is suitable for growing a file system.
1448 	 */
1449 	if (mediasize < 1)
1450 		errx(1, "partition is unavailable");
1451 
1452 	/*
1453 	 * Read the current superblock, and take a backup.
1454 	 */
1455 	if ((ret = sbget(fsi, &fs, STDSB)) != 0) {
1456 		switch (ret) {
1457 		case ENOENT:
1458 			errx(1, "superblock not recognized");
1459 		default:
1460 			errc(1, ret, "unable to read superblock");
1461 		}
1462 	}
1463 	/*
1464 	 * Check for filesystem that was unclean at mount time.
1465 	 */
1466 	if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) != 0)
1467 		errx(1, "%s is not clean - run fsck.\n", *argv);
1468 	memcpy(&osblock, fs, fs->fs_sbsize);
1469 	free(fs);
1470 	memcpy((void *)&fsun1, (void *)&fsun2, osblock.fs_sbsize);
1471 
1472 	DBG_OPEN("/tmp/growfs.debug"); /* already here we need a superblock */
1473 	DBG_DUMP_FS(&sblock, "old sblock");
1474 
1475 	/*
1476 	 * Determine size to grow to. Default to the device size.
1477 	 */
1478 	if (size == 0)
1479 		size = mediasize;
1480 	else {
1481 		if (size > (uint64_t)mediasize) {
1482 			humanize_number(oldsizebuf, sizeof(oldsizebuf), size,
1483 			    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1484 			humanize_number(newsizebuf, sizeof(newsizebuf),
1485 			    mediasize,
1486 			    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1487 
1488 			errx(1, "requested size %s is larger "
1489 			    "than the available %s", oldsizebuf, newsizebuf);
1490 		}
1491 	}
1492 
1493 	/*
1494 	 * Make sure the new size is a multiple of fs_fsize; /dev/ufssuspend
1495 	 * only supports fragment-aligned IO requests.
1496 	 */
1497 	size -= size % osblock.fs_fsize;
1498 
1499 	if (size <= (uint64_t)(osblock.fs_size * osblock.fs_fsize)) {
1500 		humanize_number(oldsizebuf, sizeof(oldsizebuf),
1501 		    osblock.fs_size * osblock.fs_fsize,
1502 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1503 		humanize_number(newsizebuf, sizeof(newsizebuf), size,
1504 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1505 
1506 		if (size == (uint64_t)(osblock.fs_size * osblock.fs_fsize))
1507 			errx(0, "requested size %s is equal to the current "
1508 			    "filesystem size %s", newsizebuf, oldsizebuf);
1509 		errx(1, "requested size %s is smaller than the current "
1510 		   "filesystem size %s", newsizebuf, oldsizebuf);
1511 	}
1512 
1513 	sblock.fs_size = dbtofsb(&osblock, size / DEV_BSIZE);
1514 	sblock.fs_providersize = dbtofsb(&osblock, mediasize / DEV_BSIZE);
1515 
1516 	/*
1517 	 * Are we really growing?
1518 	 */
1519 	if (osblock.fs_size >= sblock.fs_size) {
1520 		errx(1, "we are not growing (%jd->%jd)",
1521 		    (intmax_t)osblock.fs_size, (intmax_t)sblock.fs_size);
1522 	}
1523 
1524 	/*
1525 	 * Check if we find an active snapshot.
1526 	 */
1527 	if (yflag == 0) {
1528 		for (j = 0; j < FSMAXSNAP; j++) {
1529 			if (sblock.fs_snapinum[j]) {
1530 				errx(1, "active snapshot found in file system; "
1531 				    "please remove all snapshots before "
1532 				    "using growfs");
1533 			}
1534 			if (!sblock.fs_snapinum[j]) /* list is dense */
1535 				break;
1536 		}
1537 	}
1538 
1539 	if (yflag == 0 && Nflag == 0) {
1540 		if (statfsp != NULL && (statfsp->f_flags & MNT_RDONLY) == 0)
1541 			printf("Device is mounted read-write; resizing will "
1542 			    "result in temporary write suspension for %s.\n",
1543 			    statfsp->f_mntonname);
1544 		printf("It's strongly recommended to make a backup "
1545 		    "before growing the file system.\n"
1546 		    "OK to grow filesystem on %s", device);
1547 		if (statfsp != NULL)
1548 			printf(", mounted on %s,", statfsp->f_mntonname);
1549 		humanize_number(oldsizebuf, sizeof(oldsizebuf),
1550 		    osblock.fs_size * osblock.fs_fsize,
1551 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1552 		humanize_number(newsizebuf, sizeof(newsizebuf),
1553 		    sblock.fs_size * sblock.fs_fsize,
1554 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1555 		printf(" from %s to %s? [yes/no] ", oldsizebuf, newsizebuf);
1556 		fflush(stdout);
1557 		fgets(reply, (int)sizeof(reply), stdin);
1558 		if (strcasecmp(reply, "yes\n")){
1559 			printf("Response other than \"yes\"; aborting\n");
1560 			exit(0);
1561 		}
1562 	}
1563 
1564 	/*
1565 	 * Try to access our device for writing.  If it's not mounted,
1566 	 * or mounted read-only, simply open it; otherwise, use UFS
1567 	 * suspension mechanism.
1568 	 */
1569 	if (Nflag) {
1570 		fso = -1;
1571 	} else {
1572 		if (statfsp != NULL && (statfsp->f_flags & MNT_RDONLY) == 0) {
1573 			fso = open(_PATH_UFSSUSPEND, O_RDWR);
1574 			if (fso == -1)
1575 				err(1, "unable to open %s", _PATH_UFSSUSPEND);
1576 			error = ioctl(fso, UFSSUSPEND, &statfsp->f_fsid);
1577 			if (error != 0)
1578 				err(1, "UFSSUSPEND");
1579 		} else {
1580 			fso = open(device, O_WRONLY);
1581 			if (fso < 0)
1582 				err(1, "%s", device);
1583 		}
1584 	}
1585 
1586 	/*
1587 	 * Try to access our new last block in the file system.
1588 	 */
1589 	testbuf = malloc(sblock.fs_fsize);
1590 	if (testbuf == NULL)
1591 		err(1, "malloc");
1592 	rdfs((ufs2_daddr_t)((size - sblock.fs_fsize) / DEV_BSIZE),
1593 	    sblock.fs_fsize, testbuf, fsi);
1594 	wtfs((ufs2_daddr_t)((size - sblock.fs_fsize) / DEV_BSIZE),
1595 	    sblock.fs_fsize, testbuf, fso, Nflag);
1596 	free(testbuf);
1597 
1598 	/*
1599 	 * Now calculate new superblock values and check for reasonable
1600 	 * bound for new file system size:
1601 	 *     fs_size:    is derived from user input
1602 	 *     fs_dsize:   should get updated in the routines creating or
1603 	 *                 updating the cylinder groups on the fly
1604 	 *     fs_cstotal: should get updated in the routines creating or
1605 	 *                 updating the cylinder groups
1606 	 */
1607 
1608 	/*
1609 	 * Update the number of cylinders and cylinder groups in the file system.
1610 	 */
1611 	if (sblock.fs_magic == FS_UFS1_MAGIC) {
1612 		sblock.fs_old_ncyl =
1613 		    sblock.fs_size * sblock.fs_old_nspf / sblock.fs_old_spc;
1614 		if (sblock.fs_size * sblock.fs_old_nspf >
1615 		    sblock.fs_old_ncyl * sblock.fs_old_spc)
1616 			sblock.fs_old_ncyl++;
1617 	}
1618 	sblock.fs_ncg = howmany(sblock.fs_size, sblock.fs_fpg);
1619 
1620 	/*
1621 	 * Allocate last cylinder group only if there is enough room
1622 	 * for at least one data block.
1623 	 */
1624 	if (sblock.fs_size % sblock.fs_fpg != 0 &&
1625 	    sblock.fs_size <= cgdmin(&sblock, sblock.fs_ncg - 1)) {
1626 		humanize_number(oldsizebuf, sizeof(oldsizebuf),
1627 		    (sblock.fs_size % sblock.fs_fpg) * sblock.fs_fsize,
1628 		    "B", HN_AUTOSCALE, HN_B | HN_NOSPACE | HN_DECIMAL);
1629 		warnx("no room to allocate last cylinder group; "
1630 		    "leaving %s unused", oldsizebuf);
1631 		sblock.fs_ncg--;
1632 		if (sblock.fs_magic == FS_UFS1_MAGIC)
1633 			sblock.fs_old_ncyl = sblock.fs_ncg * sblock.fs_old_cpg;
1634 		sblock.fs_size = sblock.fs_ncg * sblock.fs_fpg;
1635 	}
1636 
1637 	/*
1638 	 * Update the space for the cylinder group summary information in the
1639 	 * respective cylinder group data area.
1640 	 */
1641 	sblock.fs_cssize =
1642 	    fragroundup(&sblock, sblock.fs_ncg * sizeof(struct csum));
1643 
1644 	if (osblock.fs_size >= sblock.fs_size)
1645 		errx(1, "not enough new space");
1646 
1647 	DBG_PRINT0("sblock calculated\n");
1648 
1649 	/*
1650 	 * Ok, everything prepared, so now let's do the tricks.
1651 	 */
1652 	growfs(fsi, fso, Nflag);
1653 
1654 	close(fsi);
1655 	if (fso > -1) {
1656 		if (statfsp != NULL && (statfsp->f_flags & MNT_RDONLY) == 0) {
1657 			error = ioctl(fso, UFSRESUME);
1658 			if (error != 0)
1659 				err(1, "UFSRESUME");
1660 		}
1661 		error = close(fso);
1662 		if (error != 0)
1663 			err(1, "close");
1664 		if (statfsp != NULL && (statfsp->f_flags & MNT_RDONLY) != 0)
1665 			mount_reload(statfsp);
1666 	}
1667 
1668 	DBG_CLOSE;
1669 
1670 	DBG_LEAVE;
1671 	return (0);
1672 }
1673 
1674 /*
1675  * Dump a line of usage.
1676  */
1677 static void
1678 usage(void)
1679 {
1680 	DBG_FUNC("usage")
1681 
1682 	DBG_ENTER;
1683 
1684 	fprintf(stderr, "usage: growfs [-Ny] [-s size] special | filesystem\n");
1685 
1686 	DBG_LEAVE;
1687 	exit(1);
1688 }
1689 
1690 /*
1691  * This updates most parameters and the bitmap related to cluster. We have to
1692  * assume that sblock, osblock, acg are set up.
1693  */
1694 static void
1695 updclst(int block)
1696 {
1697 	DBG_FUNC("updclst")
1698 	static int lcs = 0;
1699 
1700 	DBG_ENTER;
1701 
1702 	if (sblock.fs_contigsumsize < 1) /* no clustering */
1703 		return;
1704 	/*
1705 	 * update cluster allocation map
1706 	 */
1707 	setbit(cg_clustersfree(&acg), block);
1708 
1709 	/*
1710 	 * update cluster summary table
1711 	 */
1712 	if (!lcs) {
1713 		/*
1714 		 * calculate size for the trailing cluster
1715 		 */
1716 		for (block--; lcs < sblock.fs_contigsumsize; block--, lcs++ ) {
1717 			if (isclr(cg_clustersfree(&acg), block))
1718 				break;
1719 		}
1720 	}
1721 	if (lcs < sblock.fs_contigsumsize) {
1722 		if (lcs)
1723 			cg_clustersum(&acg)[lcs]--;
1724 		lcs++;
1725 		cg_clustersum(&acg)[lcs]++;
1726 	}
1727 
1728 	DBG_LEAVE;
1729 	return;
1730 }
1731 
1732 static void
1733 mount_reload(const struct statfs *stfs)
1734 {
1735 	char errmsg[255];
1736 	struct iovec *iov;
1737 	int iovlen;
1738 
1739 	iov = NULL;
1740 	iovlen = 0;
1741 	*errmsg = '\0';
1742 	build_iovec(&iov, &iovlen, "fstype", __DECONST(char *, "ffs"), 4);
1743 	build_iovec(&iov, &iovlen, "fspath", __DECONST(char *, stfs->f_mntonname), (size_t)-1);
1744 	build_iovec(&iov, &iovlen, "errmsg", errmsg, sizeof(errmsg));
1745 	build_iovec(&iov, &iovlen, "update", NULL, 0);
1746 	build_iovec(&iov, &iovlen, "reload", NULL, 0);
1747 
1748 	if (nmount(iov, iovlen, stfs->f_flags) < 0) {
1749 		errmsg[sizeof(errmsg) - 1] = '\0';
1750 		err(9, "%s: cannot reload filesystem%s%s", stfs->f_mntonname,
1751 		    *errmsg != '\0' ? ": " : "", errmsg);
1752 	}
1753 }
1754 
1755 /*
1756  * Calculate the check-hash of the cylinder group.
1757  */
1758 static void
1759 cgckhash(struct cg *cgp)
1760 {
1761 
1762 	if ((sblock.fs_metackhash & CK_CYLGRP) == 0)
1763 		return;
1764 	cgp->cg_ckhash = 0;
1765 	cgp->cg_ckhash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
1766 }
1767