growfs.c revision 1.31
1/*	$OpenBSD: growfs.c,v 1.31 2013/06/11 16:42:04 deraadt Exp $	*/
2/*
3 * Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz
4 * Copyright (c) 1980, 1989, 1993 The Regents of the University of California.
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 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 *    notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 *    notice, this list of conditions and the following disclaimer in the
17 *    documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 *    must display the following acknowledgment:
20 *      This product includes software developed by the University of
21 *      California, Berkeley and its contributors, as well as Christoph
22 *      Herrmann and Thomas-Henning von Kamptz.
23 * 4. Neither the name of the University nor the names of its contributors
24 *    may be used to endorse or promote products derived from this software
25 *    without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * SUCH DAMAGE.
38 *
39 * $TSHeader: src/sbin/growfs/growfs.c,v 1.5 2000/12/12 19:31:00 tomsoft Exp $
40 * $FreeBSD: src/sbin/growfs/growfs.c,v 1.25 2006/07/17 20:48:36 stefanf Exp $
41 *
42 */
43
44/* ********************************************************** INCLUDES ***** */
45#include <sys/param.h>
46#include <sys/disklabel.h>
47#include <sys/ioctl.h>
48#include <sys/dkio.h>
49#include <sys/stat.h>
50
51#include <stdio.h>
52#include <paths.h>
53#include <ctype.h>
54#include <err.h>
55#include <fcntl.h>
56#include <limits.h>
57#include <stdlib.h>
58#include <stdint.h>
59#include <string.h>
60#include <time.h>
61#include <unistd.h>
62#include <util.h>
63
64#include <ufs/ufs/dinode.h>
65#include <ufs/ffs/fs.h>
66
67#include "debug.h"
68
69#define	rounddown(x, y)	(((x)/(y))*(y))
70#define	roundup(x, y)	((((x)+((y)-1))/(y))*(y))
71
72/* *************************************************** GLOBALS & TYPES ***** */
73#ifdef FS_DEBUG
74int	_dbg_lvl_ = (DL_INFO);	/* DL_TRC */
75#endif /* FS_DEBUG */
76
77static int quiet;		/* quiet flag */
78
79static union {
80	struct	fs fs;
81	char	pad[SBLOCKSIZE];
82} fsun1, fsun2;
83#define	sblock	fsun1.fs	/* the new superblock */
84#define	osblock	fsun2.fs	/* the old superblock */
85
86/*
87 * Possible superblock locations ordered from most to least likely.
88 */
89static int sblock_try[] = SBLOCKSEARCH;
90static daddr_t sblockloc;
91
92static union {
93	struct	cg cg;
94	char	pad[MAXBSIZE];
95} cgun1, cgun2;
96#define	acg	cgun1.cg	/* a cylinder cgroup (new) */
97#define	aocg	cgun2.cg	/* an old cylinder group */
98
99static char	ablk[MAXBSIZE];		/* a block */
100
101static struct csum	*fscs;	/* cylinder summary */
102
103union dinode {
104	struct ufs1_dinode dp1;
105	struct ufs2_dinode dp2;
106};
107#define	DIP(dp, field) \
108	((sblock.fs_magic == FS_UFS1_MAGIC) ? \
109	(uint32_t)(dp)->dp1.field : (dp)->dp2.field)
110#define	DIP_SET(dp, field, val) do { \
111	if (sblock.fs_magic == FS_UFS1_MAGIC) \
112		(dp)->dp1.field = (val); \
113	else \
114		(dp)->dp2.field = (val); \
115	} while (0)
116static daddr_t 	inoblk;			/* inode block address */
117static char		inobuf[MAXBSIZE];	/* inode block */
118ino_t			maxino;			/* last valid inode */
119
120/*
121 * An  array of elements of type struct gfs_bpp describes all blocks  to
122 * be relocated in order to free the space needed for the cylinder group
123 * summary for all cylinder groups located in the first cylinder group.
124 */
125struct gfs_bpp {
126	daddr_t		old;		/* old block number */
127	daddr_t		new;		/* new block number */
128#define GFS_FL_FIRST	1
129#define GFS_FL_LAST	2
130	unsigned int	flags;		/* special handling required */
131	int		found;		/* how many references were updated */
132};
133
134/* ******************************************************** PROTOTYPES ***** */
135static void	growfs(int, int, unsigned int);
136static void	rdfs(daddr_t, size_t, void *, int);
137static void	wtfs(daddr_t, size_t, void *, int, unsigned int);
138static daddr_t alloc(void);
139static int	charsperline(void);
140static void	usage(void);
141static int	isblock(struct fs *, unsigned char *, int);
142static void	clrblock(struct fs *, unsigned char *, int);
143static void	setblock(struct fs *, unsigned char *, int);
144static void	initcg(int, time_t, int, unsigned int);
145static void	updjcg(int, time_t, int, int, unsigned int);
146static void	updcsloc(time_t, int, int, unsigned int);
147static struct disklabel	*get_disklabel(int);
148static void	return_disklabel(int, struct disklabel *, unsigned int);
149static union dinode *ginode(ino_t, int, int);
150static void	frag_adjust(daddr_t, int);
151static int	cond_bl_upd(daddr_t *, struct gfs_bpp *, int, int,
152		    unsigned int);
153static void	updclst(int);
154static void	updrefs(int, ino_t, struct gfs_bpp *, int, int, unsigned int);
155static void	indirchk(daddr_t, daddr_t, daddr_t, daddr_t,
156		    struct gfs_bpp *, int, int, unsigned int);
157static void	ffs1_sb_update(struct fs *, daddr_t);
158
159/* ************************************************************ growfs ***** */
160/*
161 * Here  we actually start growing the filesystem. We basically  read  the
162 * cylinder  summary  from the first cylinder group as we want  to  update
163 * this  on  the fly during our various operations. First  we  handle  the
164 * changes in the former last cylinder group. Afterwards we create all new
165 * cylinder  groups.  Now  we handle the  cylinder  group  containing  the
166 * cylinder  summary  which  might result in a  relocation  of  the  whole
167 * structure.  In the end we write back the updated cylinder summary,  the
168 * new superblock, and slightly patched versions of the super block
169 * copies.
170 */
171static void
172growfs(int fsi, int fso, unsigned int Nflag)
173{
174	DBG_FUNC("growfs")
175	int	i;
176	int	cylno, j;
177	time_t	utime;
178	int	width;
179	char	tmpbuf[100];
180
181	DBG_ENTER;
182
183	time(&utime);
184
185	/*
186	 * Get the cylinder summary into the memory.
187	 */
188	fscs = calloc((size_t)1, (size_t)sblock.fs_cssize);
189	if (fscs == NULL)
190		errx(1, "calloc failed");
191	for (i = 0; i < osblock.fs_cssize; i += osblock.fs_bsize) {
192		rdfs(fsbtodb(&osblock, osblock.fs_csaddr +
193		    numfrags(&osblock, i)), (size_t)MIN(osblock.fs_cssize - i,
194		    osblock.fs_bsize), (void *)(((char *)fscs)+i), fsi);
195	}
196
197#ifdef FS_DEBUG
198{
199	struct csum	*dbg_csp;
200	int	dbg_csc;
201	char	dbg_line[80];
202
203	dbg_csp = fscs;
204	for (dbg_csc = 0; dbg_csc < osblock.fs_ncg; dbg_csc++) {
205		snprintf(dbg_line, sizeof(dbg_line),
206		    "%d. old csum in old location", dbg_csc);
207		DBG_DUMP_CSUM(&osblock, dbg_line, dbg_csp++);
208	}
209}
210#endif /* FS_DEBUG */
211	DBG_PRINT0("fscs read\n");
212
213	/*
214	 * Do all needed changes in the former last cylinder group.
215	 */
216	updjcg(osblock.fs_ncg - 1, utime, fsi, fso, Nflag);
217
218	/*
219	 * Dump out summary information about filesystem.
220	 */
221#define B2MBFACTOR (1 / (1024.0 * 1024.0))
222	printf("growfs: %.1fMB (%jd sectors) block size %d, fragment size %d\n",
223	    (float)sblock.fs_size * sblock.fs_fsize * B2MBFACTOR,
224	    (intmax_t)fsbtodb(&sblock, sblock.fs_size), sblock.fs_bsize,
225	    sblock.fs_fsize);
226	printf("\tusing %d cylinder groups of %.2fMB, %d blks, %d inodes.\n",
227	    sblock.fs_ncg, (float)sblock.fs_fpg * sblock.fs_fsize * B2MBFACTOR,
228	    sblock.fs_fpg / sblock.fs_frag, sblock.fs_ipg);
229	if (sblock.fs_flags & FS_DOSOFTDEP)
230		printf("\twith soft updates\n");
231#undef B2MBFACTOR
232
233	/*
234	 * Now build the cylinders group blocks and
235	 * then print out indices of cylinder groups.
236	 */
237	if (!quiet)
238		printf("super-block backups (for fsck -b #) at:\n");
239	i = 0;
240	width = charsperline();
241
242	/*
243	 * Iterate for only the new cylinder groups.
244	 */
245	for (cylno = osblock.fs_ncg; cylno < sblock.fs_ncg; cylno++) {
246		initcg(cylno, utime, fso, Nflag);
247		if (quiet)
248			continue;
249		j = snprintf(tmpbuf, sizeof(tmpbuf), " %lld%s",
250		    fsbtodb(&sblock, cgsblock(&sblock, cylno)),
251		    cylno < (sblock.fs_ncg - 1) ? "," : "");
252		if (j >= sizeof(tmpbuf))
253			j = sizeof(tmpbuf) - 1;
254		if (j == -1 || i + j >= width) {
255			printf("\n");
256			i = 0;
257		}
258		i += j;
259		printf("%s", tmpbuf);
260		fflush(stdout);
261	}
262	if (!quiet)
263		printf("\n");
264
265	/*
266	 * Do all needed changes in the first cylinder group.
267	 * allocate blocks in new location
268	 */
269	updcsloc(utime, fsi, fso, Nflag);
270
271	/*
272	 * Now write the cylinder summary back to disk.
273	 */
274	for (i = 0; i < sblock.fs_cssize; i += sblock.fs_bsize) {
275		wtfs(fsbtodb(&sblock, sblock.fs_csaddr + numfrags(&sblock, i)),
276		    (size_t)MIN(sblock.fs_cssize - i, sblock.fs_bsize),
277		    (void *)(((char *)fscs) + i), fso, Nflag);
278	}
279	DBG_PRINT0("fscs written\n");
280
281#ifdef FS_DEBUG
282{
283	struct csum	*dbg_csp;
284	int	dbg_csc;
285	char	dbg_line[80];
286
287	dbg_csp = fscs;
288	for (dbg_csc = 0; dbg_csc < sblock.fs_ncg; dbg_csc++) {
289		snprintf(dbg_line, sizeof(dbg_line),
290		    "%d. new csum in new location", dbg_csc);
291		DBG_DUMP_CSUM(&sblock, dbg_line, dbg_csp++);
292	}
293}
294#endif /* FS_DEBUG */
295
296	/*
297	 * Now write the new superblock back to disk.
298	 */
299	sblock.fs_time = utime;
300	sblock.fs_clean = 0;
301	if (sblock.fs_magic == FS_UFS1_MAGIC) {
302		sblock.fs_ffs1_time = (int32_t)sblock.fs_time;
303		sblock.fs_ffs1_size = (int32_t)sblock.fs_size;
304		sblock.fs_ffs1_dsize = (int32_t)sblock.fs_dsize;
305		sblock.fs_ffs1_csaddr = (int32_t)sblock.fs_csaddr;
306		sblock.fs_ffs1_cstotal.cs_ndir =
307		    (int32_t)sblock.fs_cstotal.cs_ndir;
308		sblock.fs_ffs1_cstotal.cs_nbfree =
309		    (int32_t)sblock.fs_cstotal.cs_nbfree;
310		sblock.fs_ffs1_cstotal.cs_nifree =
311		    (int32_t)sblock.fs_cstotal.cs_nifree;
312		sblock.fs_ffs1_cstotal.cs_nffree =
313		    (int32_t)sblock.fs_cstotal.cs_nffree;
314	}
315	wtfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
316	DBG_PRINT0("sblock written\n");
317	DBG_DUMP_FS(&sblock, "new initial sblock");
318
319	/*
320	 * Clean up the dynamic fields in our superblock copies.
321	 */
322	sblock.fs_fmod = 0;
323	sblock.fs_clean = 1;
324	sblock.fs_ronly = 0;
325	sblock.fs_cgrotor = 0;
326	sblock.fs_state = 0;
327	memset((void *)&sblock.fs_fsmnt, 0, sizeof(sblock.fs_fsmnt));
328	sblock.fs_flags &= FS_DOSOFTDEP;
329	if (sblock.fs_magic == FS_UFS1_MAGIC)
330		sblock.fs_ffs1_flags &= FS_DOSOFTDEP;
331
332	/*
333	 * XXX
334	 * The following fields are currently distributed from the  superblock
335	 * to the copies:
336	 *     fs_minfree
337	 *     fs_rotdelay
338	 *     fs_maxcontig
339	 *     fs_maxbpg
340	 *     fs_minfree,
341	 *     fs_optim
342	 *     fs_flags regarding SOFTPDATES
343	 *
344	 * We probably should rather change the summary for the cylinder group
345	 * statistics here to the value of what would be in there, if the file
346	 * system were created initially with the new size. Therefor we  still
347	 * need to find an easy way of calculating that.
348	 * Possibly we can try to read the first superblock copy and apply the
349	 * "diffed" stats between the old and new superblock by still  copying
350	 * certain parameters onto that.
351	 */
352
353	/*
354	 * Write out the duplicate superblocks.
355	 */
356	for (cylno = 0; cylno < sblock.fs_ncg; cylno++) {
357		wtfs(fsbtodb(&sblock, cgsblock(&sblock, cylno)),
358		    (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
359	}
360	DBG_PRINT0("sblock copies written\n");
361	DBG_DUMP_FS(&sblock, "new other sblocks");
362
363	DBG_LEAVE;
364}
365
366/* ************************************************************ initcg ***** */
367/*
368 * This creates a new cylinder group structure, for more details please  see
369 * the  source of newfs(8), as this function is taken over almost unchanged.
370 * As  this  is  never called for the  first  cylinder  group,  the  special
371 * provisions for that case are removed here.
372 */
373static void
374initcg(int cylno, time_t utime, int fso, unsigned int Nflag)
375{
376	DBG_FUNC("initcg")
377	static char *iobuf;
378	daddr_t d, dlower, dupper, blkno, start;
379	daddr_t i, cbase, dmax;
380	struct ufs1_dinode *dp1;
381	struct ufs2_dinode *dp2;
382	struct csum *cs;
383	ino_t j;
384	size_t iobufsize;
385
386	if (sblock.fs_bsize < SBLOCKSIZE)
387		iobufsize = SBLOCKSIZE + 3 * sblock.fs_bsize;
388	else
389		iobufsize = 4 * sblock.fs_bsize;
390
391	if (iobuf == NULL && (iobuf = malloc(iobufsize)) == NULL)
392		errx(37, "panic: cannot allocate I/O buffer");
393	bzero(iobuf, iobufsize);
394
395	/*
396	 * Determine block bounds for cylinder group.
397	 * Allow space for super block summary information in first
398	 * cylinder group.
399	 */
400	cbase = cgbase(&sblock, cylno);
401	dmax = cbase + sblock.fs_fpg;
402	if (dmax > sblock.fs_size)
403		dmax = sblock.fs_size;
404	dlower = cgsblock(&sblock, cylno) - cbase;
405	dupper = cgdmin(&sblock, cylno) - cbase;
406	if (cylno == 0) /* XXX fscs may be relocated */
407		dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
408	cs = &fscs[cylno];
409	memset(&acg, 0, sblock.fs_cgsize);
410	acg.cg_ffs2_time = utime;
411	acg.cg_magic = CG_MAGIC;
412	acg.cg_cgx = cylno;
413	acg.cg_ffs2_niblk = sblock.fs_ipg;
414	acg.cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
415	acg.cg_ndblk = dmax - cbase;
416	if (sblock.fs_contigsumsize > 0)
417		acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
418	start = sizeof(struct cg);
419	if (sblock.fs_magic == FS_UFS2_MAGIC) {
420		acg.cg_iusedoff = start;
421	} else {
422		if (cylno == sblock.fs_ncg - 1)
423			acg.cg_ncyl = sblock.fs_ncyl % sblock.fs_cpg;
424		else
425			acg.cg_ncyl = sblock.fs_cpg;
426		acg.cg_time = (int32_t)acg.cg_ffs2_time;
427		acg.cg_ffs2_time = 0;
428		acg.cg_niblk = (int16_t)acg.cg_ffs2_niblk;
429		acg.cg_ffs2_niblk = 0;
430		acg.cg_initediblk = 0;
431		acg.cg_btotoff = start;
432		acg.cg_boff = acg.cg_btotoff +
433		    sblock.fs_cpg * sizeof(int32_t);
434		acg.cg_iusedoff = acg.cg_boff +
435		    sblock.fs_cpg * sizeof(u_int16_t);
436	}
437	acg.cg_freeoff = acg.cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
438	acg.cg_nextfreeoff = acg.cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT);
439	if (sblock.fs_contigsumsize > 0) {
440		acg.cg_clustersumoff =
441		    roundup(acg.cg_nextfreeoff, sizeof(u_int32_t));
442		acg.cg_clustersumoff -= sizeof(u_int32_t);
443		acg.cg_clusteroff = acg.cg_clustersumoff +
444		    (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
445		acg.cg_nextfreeoff = acg.cg_clusteroff +
446		    howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
447	}
448	if (acg.cg_nextfreeoff > sblock.fs_cgsize) {
449		/*
450		 * This should never happen as we would have had that panic
451		 *     already on filesystem creation
452		 */
453		errx(37, "panic: cylinder group too big");
454	}
455	acg.cg_cs.cs_nifree += sblock.fs_ipg;
456	if (cylno == 0) {
457		for (i = 0; i < ROOTINO; i++) {
458			setbit(cg_inosused(&acg), i);
459			acg.cg_cs.cs_nifree--;
460		}
461	}
462	if (cylno > 0) {
463		/*
464		 * In cylno 0, beginning space is reserved
465		 * for boot and super blocks.
466		 */
467		for (d = 0; d < dlower; d += sblock.fs_frag) {
468			blkno = d / sblock.fs_frag;
469			setblock(&sblock, cg_blksfree(&acg), blkno);
470			if (sblock.fs_contigsumsize > 0)
471				setbit(cg_clustersfree(&acg), blkno);
472			acg.cg_cs.cs_nbfree++;
473		}
474		sblock.fs_dsize += dlower;
475	}
476	sblock.fs_dsize += acg.cg_ndblk - dupper;
477	if ((i = dupper % sblock.fs_frag)) {
478		acg.cg_frsum[sblock.fs_frag - i]++;
479		for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
480			setbit(cg_blksfree(&acg), dupper);
481			acg.cg_cs.cs_nffree++;
482		}
483	}
484	for (d = dupper; d + sblock.fs_frag <= acg.cg_ndblk;
485	     d += sblock.fs_frag) {
486		blkno = d / sblock.fs_frag;
487		setblock(&sblock, cg_blksfree(&acg), blkno);
488		if (sblock.fs_contigsumsize > 0)
489			setbit(cg_clustersfree(&acg), blkno);
490		acg.cg_cs.cs_nbfree++;
491	}
492	if (d < acg.cg_ndblk) {
493		acg.cg_frsum[acg.cg_ndblk - d]++;
494		for (; d < acg.cg_ndblk; d++) {
495			setbit(cg_blksfree(&acg), d);
496			acg.cg_cs.cs_nffree++;
497		}
498	}
499	if (sblock.fs_contigsumsize > 0) {
500		int32_t	*sump = cg_clustersum(&acg);
501		u_char	*mapp = cg_clustersfree(&acg);
502		int	map = *mapp++;
503		int	bit = 1;
504		int	run = 0;
505
506		for (i = 0; i < acg.cg_nclusterblks; i++) {
507			if ((map & bit) != 0)
508				run++;
509			else if (run != 0) {
510				if (run > sblock.fs_contigsumsize)
511					run = sblock.fs_contigsumsize;
512				sump[run]++;
513				run = 0;
514			}
515			if ((i & (CHAR_BIT - 1)) != CHAR_BIT - 1)
516				bit <<= 1;
517			else {
518				map = *mapp++;
519				bit = 1;
520			}
521		}
522		if (run != 0) {
523			if (run > sblock.fs_contigsumsize)
524				run = sblock.fs_contigsumsize;
525			sump[run]++;
526		}
527	}
528	sblock.fs_cstotal.cs_ndir += acg.cg_cs.cs_ndir;
529	sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
530	sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
531	sblock.fs_cstotal.cs_nifree += acg.cg_cs.cs_nifree;
532	*cs = acg.cg_cs;
533
534	/*
535	 * Write out the duplicate superblock, the cylinder group map
536	 * and two blocks worth of inodes in a single write.
537	 */
538	bcopy(&sblock, iobuf, SBLOCKSIZE);
539	start = sblock.fs_bsize > SBLOCKSIZE ? sblock.fs_bsize : SBLOCKSIZE;
540	bcopy(&acg, &iobuf[start], sblock.fs_cgsize);
541	start += sblock.fs_bsize;
542	dp1 = (struct ufs1_dinode *)&iobuf[start];
543	dp2 = (struct ufs2_dinode *)&iobuf[start];
544	for (i = MIN(sblock.fs_ipg, 2 * INOPB(&sblock)); i != 0; i--) {
545		if (sblock.fs_magic == FS_UFS1_MAGIC) {
546			dp1->di_gen = arc4random();
547			dp1++;
548		} else {
549			dp2->di_gen = arc4random();
550			dp2++;
551		}
552	}
553	wtfs(fsbtodb(&sblock, cgsblock(&sblock, cylno)), iobufsize,
554	    iobuf, fso, Nflag);
555
556	/* Initialize inodes for FFS1. */
557	if (sblock.fs_magic == FS_UFS1_MAGIC) {
558		for (i = 2 * sblock.fs_frag; i < sblock.fs_ipg / INOPF(&sblock);
559		     i += sblock.fs_frag) {
560			dp1 = (struct ufs1_dinode *)&iobuf[start];
561			for (j = 0; j < INOPB(&sblock); j++) {
562				dp1->di_gen = arc4random();
563				dp1++;
564			}
565			wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno) + i),
566			    (size_t)sblock.fs_bsize, &iobuf[start], fso, Nflag);
567		}
568	}
569
570	DBG_DUMP_CG(&sblock, "new cg", &acg);
571
572	DBG_LEAVE;
573}
574
575/* ******************************************************* frag_adjust ***** */
576/*
577 * Here  we add or subtract (sign +1/-1) the available fragments in  a  given
578 * block to or from the fragment statistics. By subtracting before and adding
579 * after  an operation on the free frag map we can easy update  the  fragment
580 * statistic, which seems to be otherwise a rather complex operation.
581 */
582static void
583frag_adjust(daddr_t frag, int sign)
584{
585	DBG_FUNC("frag_adjust")
586	int fragsize;
587	int f;
588
589	DBG_ENTER;
590
591	fragsize = 0;
592	/*
593	 * Here frag only needs to point to any fragment in the block we want
594	 * to examine.
595	 */
596	for (f = rounddown(frag, sblock.fs_frag);
597	    f < roundup(frag + 1, sblock.fs_frag);
598	    f++) {
599		/*
600		 * Count contiguous free fragments.
601		 */
602		if (isset(cg_blksfree(&acg), f)) {
603			fragsize++;
604		} else {
605			if (fragsize && fragsize < sblock.fs_frag) {
606				/*
607				 * We found something in between.
608				 */
609				acg.cg_frsum[fragsize] += sign;
610				DBG_PRINT2("frag_adjust [%d]+=%d\n", fragsize,
611				    sign);
612			}
613			fragsize = 0;
614		}
615	}
616	if (fragsize && fragsize < sblock.fs_frag) {
617		/*
618		 * We found something.
619		 */
620		acg.cg_frsum[fragsize] += sign;
621		DBG_PRINT2("frag_adjust [%d]+=%d\n", fragsize, sign);
622	}
623	DBG_PRINT2("frag_adjust [[%d]]+=%d\n", fragsize, sign);
624
625	DBG_LEAVE;
626}
627
628/* ******************************************************* cond_bl_upd ***** */
629/*
630 * Here we conditionally update a pointer to a fragment. We check for all
631 * relocated blocks if any of its fragments is referenced by the current
632 * field,  and update the pointer to the respective fragment in  our  new
633 * block.  If  we find a reference we write back the  block  immediately,
634 * as there is no easy way for our general block reading engine to figure
635 * out if a write back operation is needed.
636 */
637static int
638cond_bl_upd(daddr_t *block, struct gfs_bpp *field, int fsi, int fso,
639    unsigned int Nflag)
640{
641	DBG_FUNC("cond_bl_upd")
642	struct gfs_bpp	*f;
643	daddr_t src, dst;
644	int fragnum;
645	void *ibuf;
646
647	DBG_ENTER;
648
649	for (f = field; f->old != 0; f++) {
650		src = *block;
651		if (fragstoblks(&sblock, src) != f->old)
652			continue;
653		/*
654		 * The fragment is part of the block, so update.
655		 */
656		dst = blkstofrags(&sblock, f->new);
657		fragnum = fragnum(&sblock, src);
658		*block = dst + fragnum;
659		f->found++;
660		DBG_PRINT3("scg (%jd->%jd)[%d] reference updated\n",
661		    (intmax_t)f->old, (intmax_t)f->new, fragnum);
662
663		/*
664		 * Copy the block back immediately.
665		 *
666		 * XXX	If src is is from an indirect block we have
667		 *	to implement copy on write here in case of
668		 *	active snapshots.
669		 */
670		ibuf = malloc(sblock.fs_bsize);
671		if (!ibuf)
672			errx(1, "malloc failed");
673		src -= fragnum;
674		rdfs(fsbtodb(&sblock, src), (size_t)sblock.fs_bsize, ibuf, fsi);
675		wtfs(dst, (size_t)sblock.fs_bsize, ibuf, fso, Nflag);
676		free(ibuf);
677		/*
678		 * The same block can't be found again in this loop.
679		 */
680		return (1);
681	}
682
683	DBG_LEAVE;
684	return (0);
685}
686
687/* ************************************************************ updjcg ***** */
688/*
689 * Here we do all needed work for the former last cylinder group. It has to be
690 * changed  in  any case, even if the filesystem ended exactly on the  end  of
691 * this  group, as there is some slightly inconsistent handling of the  number
692 * of cylinders in the cylinder group. We start again by reading the  cylinder
693 * group from disk. If the last block was not fully available, we first handle
694 * the  missing  fragments, then we handle all new full blocks  in  that  file
695 * system  and  finally we handle the new last fragmented block  in  the  file
696 * system.  We again have to handle the fragment statistics rotational  layout
697 * tables and cluster summary during all those operations.
698 */
699static void
700updjcg(int cylno, time_t utime, int fsi, int fso, unsigned int Nflag)
701{
702	DBG_FUNC("updjcg")
703	daddr_t	cbase, dmax, dupper;
704	struct csum	*cs;
705	int	i, k;
706	int	j = 0;
707
708	DBG_ENTER;
709
710	/*
711	 * Read the former last (joining) cylinder group from disk, and make
712	 * a copy.
713	 */
714	rdfs(fsbtodb(&osblock, cgtod(&osblock, cylno)),
715	    (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
716	DBG_PRINT0("jcg read\n");
717	DBG_DUMP_CG(&sblock, "old joining cg", &aocg);
718
719	memcpy(&cgun1, &cgun2, sizeof(cgun2));
720
721	/*
722	 * If the cylinder group had already its new final size almost
723	 * nothing is to be done ... except:
724	 * For some reason the value of cg_ncyl in the last cylinder group has
725	 * to  be  zero instead of fs_cpg. As this is now no longer  the  last
726	 * cylinder group we have to change that value now to fs_cpg.
727	 */
728	if (cgbase(&osblock, cylno+1) == osblock.fs_size) {
729		if (sblock.fs_magic == FS_UFS1_MAGIC)
730			acg.cg_ncyl = sblock.fs_cpg;
731
732		wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
733		    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
734		DBG_PRINT0("jcg written\n");
735		DBG_DUMP_CG(&sblock, "new joining cg", &acg);
736
737		DBG_LEAVE;
738		return;
739	}
740
741	/*
742	 * Set up some variables needed later.
743	 */
744	cbase = cgbase(&sblock, cylno);
745	dmax = cbase + sblock.fs_fpg;
746	if (dmax > sblock.fs_size)
747		dmax = sblock.fs_size;
748	dupper = cgdmin(&sblock, cylno) - cbase;
749	if (cylno == 0)	/* XXX fscs may be relocated */
750		dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
751
752	/*
753	 * Set pointer to the cylinder summary for our cylinder group.
754	 */
755	cs = fscs + cylno;
756
757	/*
758	 * Touch the cylinder group, update all fields in the cylinder group as
759	 * needed, update the free space in the superblock.
760	 */
761	acg.cg_time = utime;
762	if (sblock.fs_magic == FS_UFS1_MAGIC) {
763		if (cylno == sblock.fs_ncg - 1) {
764			/*
765			 * This is still the last cylinder group.
766			 */
767			acg.cg_ncyl = sblock.fs_ncyl % sblock.fs_cpg;
768		} else {
769			acg.cg_ncyl = sblock.fs_cpg;
770		}
771	}
772	DBG_PRINT2("jcg dbg: %d %u", cylno, sblock.fs_ncg);
773#ifdef FS_DEBUG
774	if (sblock.fs_magic == FS_UFS1_MAGIC)
775		DBG_PRINT2("%d %u", acg.cg_ncyl, sblock.fs_cpg);
776#endif
777	DBG_PRINT0("\n");
778	acg.cg_ndblk = dmax - cbase;
779	sblock.fs_dsize += acg.cg_ndblk-aocg.cg_ndblk;
780	if (sblock.fs_contigsumsize > 0)
781		acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
782
783	/*
784	 * Now  we have to update the free fragment bitmap for our new  free
785	 * space.  There again we have to handle the fragmentation and  also
786	 * the  rotational  layout tables and the cluster summary.  This  is
787	 * also  done per fragment for the first new block if the  old  file
788	 * system end was not on a block boundary, per fragment for the  new
789	 * last block if the new filesystem end is not on a block boundary,
790	 * and per block for all space in between.
791	 *
792	 * Handle the first new block here if it was partially available
793	 * before.
794	 */
795	if (osblock.fs_size % sblock.fs_frag) {
796		if (roundup(osblock.fs_size, sblock.fs_frag) <= sblock.fs_size) {
797			/*
798			 * The new space is enough to fill at least this
799			 * block
800			 */
801			j = 0;
802			for (i = roundup(osblock.fs_size-cbase, sblock.fs_frag) - 1;
803			    i >= osblock.fs_size-cbase; i--) {
804				setbit(cg_blksfree(&acg), i);
805				acg.cg_cs.cs_nffree++;
806				j++;
807			}
808
809			/*
810			 * Check  if the fragment just created could join  an
811			 * already existing fragment at the former end of the
812			 * filesystem.
813			 */
814			if (isblock(&sblock, cg_blksfree(&acg),
815			    ((osblock.fs_size - cgbase(&sblock, cylno))/
816			    sblock.fs_frag))) {
817				/*
818				 * The block is now completely available.
819				 */
820				DBG_PRINT0("block was\n");
821				acg.cg_frsum[osblock.fs_size%sblock.fs_frag]--;
822				acg.cg_cs.cs_nbfree++;
823				acg.cg_cs.cs_nffree-=sblock.fs_frag;
824				k = rounddown(osblock.fs_size-cbase,
825				    sblock.fs_frag);
826				updclst((osblock.fs_size-cbase)/sblock.fs_frag);
827			} else {
828				/*
829				 * Lets rejoin a possible partially growed
830				 * fragment.
831				 */
832				k = 0;
833				while (isset(cg_blksfree(&acg), i) &&
834				    (i >= rounddown(osblock.fs_size - cbase,
835				    sblock.fs_frag))) {
836					i--;
837					k++;
838				}
839				if (k)
840					acg.cg_frsum[k]--;
841				acg.cg_frsum[k + j]++;
842			}
843		} else {
844			/*
845			 * We only grow by some fragments within this last
846			 * block.
847			 */
848			for (i = sblock.fs_size-cbase-1;
849			    i >= osblock.fs_size-cbase; i--) {
850				setbit(cg_blksfree(&acg), i);
851				acg.cg_cs.cs_nffree++;
852				j++;
853			}
854			/*
855			 * Lets rejoin a possible partially growed fragment.
856			 */
857			k = 0;
858			while (isset(cg_blksfree(&acg), i) &&
859			    (i >= rounddown(osblock.fs_size - cbase,
860			    sblock.fs_frag))) {
861				i--;
862				k++;
863			}
864			if (k)
865				acg.cg_frsum[k]--;
866			acg.cg_frsum[k + j]++;
867		}
868	}
869
870	/*
871	 * Handle all new complete blocks here.
872	 */
873	for (i = roundup(osblock.fs_size - cbase, sblock.fs_frag);
874	    i + sblock.fs_frag <= dmax-cbase;	/* XXX <= or only < ? */
875	    i += sblock.fs_frag) {
876		j = i / sblock.fs_frag;
877		setblock(&sblock, cg_blksfree(&acg), j);
878		updclst(j);
879		acg.cg_cs.cs_nbfree++;
880	}
881
882	/*
883	 * Handle the last new block if there are stll some new fragments left.
884	 * Here  we don't have to bother about the cluster summary or the  even
885	 * the rotational layout table.
886	 */
887	if (i < (dmax - cbase)) {
888		acg.cg_frsum[dmax - cbase - i]++;
889		for (; i < dmax - cbase; i++) {
890			setbit(cg_blksfree(&acg), i);
891			acg.cg_cs.cs_nffree++;
892		}
893	}
894
895	sblock.fs_cstotal.cs_nffree +=
896	    (acg.cg_cs.cs_nffree - aocg.cg_cs.cs_nffree);
897	sblock.fs_cstotal.cs_nbfree +=
898	    (acg.cg_cs.cs_nbfree - aocg.cg_cs.cs_nbfree);
899	/*
900	 * The following statistics are not changed here:
901	 *     sblock.fs_cstotal.cs_ndir
902	 *     sblock.fs_cstotal.cs_nifree
903	 * As the statistics for this cylinder group are ready, copy it to
904	 * the summary information array.
905	 */
906	*cs = acg.cg_cs;
907
908	/*
909	 * Write the updated "joining" cylinder group back to disk.
910	 */
911	wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize,
912	    (void *)&acg, fso, Nflag);
913	DBG_PRINT0("jcg written\n");
914	DBG_DUMP_CG(&sblock, "new joining cg", &acg);
915
916	DBG_LEAVE;
917}
918
919/* ********************************************************** updcsloc ***** */
920/*
921 * Here  we update the location of the cylinder summary. We have  two  possible
922 * ways of growing the cylinder summary.
923 * (1)	We can try to grow the summary in the current location, and  relocate
924 *	possibly used blocks within the current cylinder group.
925 * (2)	Alternatively we can relocate the whole cylinder summary to the first
926 *	new completely empty cylinder group. Once the cylinder summary is  no
927 *	longer in the beginning of the first cylinder group you should  never
928 *	use  a version of fsck which is not aware of the possibility to  have
929 *	this structure in a non standard place.
930 * Option (1) is considered to be less intrusive to the structure of the  file-
931 * system. So we try to stick to that whenever possible. If there is not enough
932 * space  in the cylinder group containing the cylinder summary we have to  use
933 * method  (2). In case of active snapshots in the filesystem we  probably  can
934 * completely avoid implementing copy on write if we stick to method (2) only.
935 */
936static void
937updcsloc(time_t utime, int fsi, int fso, unsigned int Nflag)
938{
939	DBG_FUNC("updcsloc")
940	struct csum	*cs;
941	int	ocscg, ncscg;
942	int	blocks;
943	daddr_t	cbase, dupper, odupper, d, f, g;
944	int	ind;
945	int	cylno, inc;
946	struct gfs_bpp	*bp;
947	int	i, l;
948	int	lcs = 0;
949	int	block;
950
951	DBG_ENTER;
952
953	if (howmany(sblock.fs_cssize, sblock.fs_fsize) ==
954	    howmany(osblock.fs_cssize, osblock.fs_fsize)) {
955		/*
956		 * No new fragment needed.
957		 */
958		DBG_LEAVE;
959		return;
960	}
961	ocscg = dtog(&osblock, osblock.fs_csaddr);
962	cs = fscs + ocscg;
963	blocks = 1+howmany(sblock.fs_cssize, sblock.fs_bsize)-
964	    howmany(osblock.fs_cssize, osblock.fs_bsize);
965
966	/*
967	 * Read original cylinder group from disk, and make a copy.
968	 * XXX	If Nflag is set in some very rare cases we now miss
969	 *	some changes done in updjcg by reading the unmodified
970	 *	block from disk.
971	 */
972	rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)),
973	    (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
974	DBG_PRINT0("oscg read\n");
975	DBG_DUMP_CG(&sblock, "old summary cg", &aocg);
976
977	memcpy(&cgun1, &cgun2, sizeof(cgun2));
978
979	/*
980	 * Touch the cylinder group, set up local variables needed later
981	 * and update the superblock.
982	 */
983	acg.cg_time = utime;
984
985	/*
986	 * XXX	In the case of having active snapshots we may need much more
987	 *	blocks for the copy on write. We need each block twice,  and
988	 *	also  up to 8*3 blocks for indirect blocks for all  possible
989	 *	references.
990	 */
991	if (/*((int)sblock.fs_time & 0x3) > 0 || */ cs->cs_nbfree < blocks) {
992		/*
993		 * There  is  not enough space in the old cylinder  group  to
994		 * relocate  all blocks as needed, so we relocate  the  whole
995		 * cylinder  group summary to a new group. We try to use  the
996		 * first complete new cylinder group just created. Within the
997		 * cylinder  group we align the area immediately  after  the
998		 * cylinder  group  information location in order  to  be  as
999		 * close as possible to the original implementation of ffs.
1000		 *
1001		 * First  we have to make sure we'll find enough space in  the
1002		 * new  cylinder  group. If not, then we  currently  give  up.
1003		 * We  start  with freeing everything which was  used  by  the
1004		 * fragments of the old cylinder summary in the current group.
1005		 * Now  we write back the group meta data, read in the  needed
1006		 * meta data from the new cylinder group, and start allocating
1007		 * within  that  group. Here we can assume, the  group  to  be
1008		 * completely empty. Which makes the handling of fragments and
1009		 * clusters a lot easier.
1010		 */
1011		DBG_TRC;
1012		if (sblock.fs_ncg-osblock.fs_ncg < 2)
1013			errx(2, "panic: not enough space");
1014
1015		/*
1016		 * Point "d" to the first fragment not used by the cylinder
1017		 * summary.
1018		 */
1019		d = osblock.fs_csaddr + (osblock.fs_cssize / osblock.fs_fsize);
1020
1021		/*
1022		 * Set up last cluster size ("lcs") already here. Calculate
1023		 * the size for the trailing cluster just behind where  "d"
1024		 * points to.
1025		 */
1026		if (sblock.fs_contigsumsize > 0) {
1027			for (block = howmany(d % sblock.fs_fpg, sblock.fs_frag),
1028			    lcs = 0; lcs < sblock.fs_contigsumsize;
1029			    block++, lcs++) {
1030				if (isclr(cg_clustersfree(&acg), block))
1031					break;
1032			}
1033		}
1034
1035		/*
1036		 * Point "d" to the last frag used by the cylinder summary.
1037		 */
1038		d--;
1039
1040		DBG_PRINT1("d=%jd\n", (intmax_t)d);
1041		if ((d + 1) % sblock.fs_frag) {
1042			/*
1043			 * The end of the cylinder summary is not a complete
1044			 * block.
1045			 */
1046			DBG_TRC;
1047			frag_adjust(d % sblock.fs_fpg, -1);
1048			for (; (d + 1) % sblock.fs_frag; d--) {
1049				DBG_PRINT1("d=%jd\n", (intmax_t)d);
1050				setbit(cg_blksfree(&acg), d % sblock.fs_fpg);
1051				acg.cg_cs.cs_nffree++;
1052				sblock.fs_cstotal.cs_nffree++;
1053			}
1054			/*
1055			 * Point  "d" to the last fragment of the  last
1056			 * (incomplete) block of the cylinder summary.
1057			 */
1058			d++;
1059			frag_adjust(d % sblock.fs_fpg, 1);
1060
1061			if (isblock(&sblock, cg_blksfree(&acg),
1062			    (d % sblock.fs_fpg) / sblock.fs_frag)) {
1063				DBG_PRINT1("d=%jd\n", (intmax_t)d);
1064				acg.cg_cs.cs_nffree -= sblock.fs_frag;
1065				acg.cg_cs.cs_nbfree++;
1066				sblock.fs_cstotal.cs_nffree -= sblock.fs_frag;
1067				sblock.fs_cstotal.cs_nbfree++;
1068				if (sblock.fs_contigsumsize > 0) {
1069					setbit(cg_clustersfree(&acg),
1070					    (d % sblock.fs_fpg) / sblock.fs_frag);
1071					if (lcs < sblock.fs_contigsumsize) {
1072						if (lcs) {
1073							cg_clustersum(&acg)
1074							    [lcs]--;
1075						}
1076						lcs++;
1077						cg_clustersum(&acg)[lcs]++;
1078					}
1079				}
1080			}
1081			/*
1082			 * Point "d" to the first fragment of the block before
1083			 * the last incomplete block.
1084			 */
1085			d--;
1086		}
1087
1088		DBG_PRINT1("d=%jd\n", (intmax_t)d);
1089		for (d = rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr;
1090		    d -= sblock.fs_frag) {
1091			DBG_TRC;
1092			DBG_PRINT1("d=%d\n", d);
1093			setblock(&sblock, cg_blksfree(&acg),
1094			    (d % sblock.fs_fpg) / sblock.fs_frag);
1095			acg.cg_cs.cs_nbfree++;
1096			sblock.fs_cstotal.cs_nbfree++;
1097			 if (sblock.fs_contigsumsize > 0) {
1098				setbit(cg_clustersfree(&acg),
1099				    (d % sblock.fs_fpg) / sblock.fs_frag);
1100				/*
1101				 * The last cluster size is already set up.
1102				 */
1103				if (lcs < sblock.fs_contigsumsize) {
1104					if (lcs) {
1105						cg_clustersum(&acg)[lcs]--;
1106					}
1107					lcs++;
1108					cg_clustersum(&acg)[lcs]++;
1109				}
1110			}
1111		}
1112		*cs = acg.cg_cs;
1113
1114		/*
1115		 * Now write the former cylinder group containing the cylinder
1116		 * summary back to disk.
1117		 */
1118		wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)),
1119		    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1120		DBG_PRINT0("oscg written\n");
1121		DBG_DUMP_CG(&sblock, "old summary cg", &acg);
1122
1123		/*
1124		 * Find the beginning of the new cylinder group containing the
1125		 * cylinder summary.
1126		 */
1127		sblock.fs_csaddr = cgdmin(&sblock, osblock.fs_ncg);
1128		ncscg = dtog(&sblock, sblock.fs_csaddr);
1129		cs = fscs + ncscg;
1130
1131
1132		/*
1133		 * If Nflag is specified, we would now read random data instead
1134		 * of an empty cg structure from disk. So we can't simulate that
1135		 * part for now.
1136		 */
1137		if (Nflag) {
1138			DBG_PRINT0("nscg update skipped\n");
1139			DBG_LEAVE;
1140			return;
1141		}
1142
1143		/*
1144		 * Read the future cylinder group containing the cylinder
1145		 * summary from disk, and make a copy.
1146		 */
1147		rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1148		    (size_t)sblock.fs_cgsize, &aocg, fsi);
1149		DBG_PRINT0("nscg read\n");
1150		DBG_DUMP_CG(&sblock, "new summary cg", &aocg);
1151
1152		memcpy(&cgun1, &cgun2, sizeof(cgun2));
1153
1154		/*
1155		 * Allocate all complete blocks used by the new cylinder
1156		 * summary.
1157		 */
1158		for (d = sblock.fs_csaddr; d + sblock.fs_frag <=
1159		    sblock.fs_csaddr + (sblock.fs_cssize / sblock.fs_fsize);
1160		    d += sblock.fs_frag) {
1161			clrblock(&sblock, cg_blksfree(&acg),
1162			    (d%sblock.fs_fpg)/sblock.fs_frag);
1163			acg.cg_cs.cs_nbfree--;
1164			sblock.fs_cstotal.cs_nbfree--;
1165			if (sblock.fs_contigsumsize > 0) {
1166				clrbit(cg_clustersfree(&acg),
1167				    (d % sblock.fs_fpg) / sblock.fs_frag);
1168			}
1169		}
1170
1171		/*
1172		 * Allocate all fragments used by the cylinder summary in the
1173		 * last block.
1174		 */
1175		if (d < sblock.fs_csaddr + (sblock.fs_cssize / sblock.fs_fsize)) {
1176			for (; d - sblock.fs_csaddr <
1177			    sblock.fs_cssize/sblock.fs_fsize;
1178			    d++) {
1179				clrbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1180				acg.cg_cs.cs_nffree--;
1181				sblock.fs_cstotal.cs_nffree--;
1182			}
1183			acg.cg_cs.cs_nbfree--;
1184			acg.cg_cs.cs_nffree += sblock.fs_frag;
1185			sblock.fs_cstotal.cs_nbfree--;
1186			sblock.fs_cstotal.cs_nffree += sblock.fs_frag;
1187			if (sblock.fs_contigsumsize > 0) {
1188				clrbit(cg_clustersfree(&acg),
1189				    (d%sblock.fs_fpg) / sblock.fs_frag);
1190			}
1191
1192			frag_adjust(d % sblock.fs_fpg, 1);
1193		}
1194		/*
1195		 * XXX	Handle the cluster statistics here in the case  this
1196		 *	cylinder group is now almost full, and the remaining
1197		 *	space is less then the maximum cluster size. This is
1198		 *	probably not needed, as you would hardly find a file
1199		 *	system which has only MAXCSBUFS+FS_MAXCONTIG of free
1200		 *	space right behind the cylinder group information in
1201		 *	any new cylinder group.
1202		 */
1203
1204		/*
1205		 * Update our statistics in the cylinder summary.
1206		 */
1207		*cs = acg.cg_cs;
1208
1209		/*
1210		 * Write the new cylinder group containing the cylinder summary
1211		 * back to disk.
1212		 */
1213		wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1214		    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1215		DBG_PRINT0("nscg written\n");
1216		DBG_DUMP_CG(&sblock, "new summary cg", &acg);
1217
1218		DBG_LEAVE;
1219		return;
1220	}
1221	/*
1222	 * We have got enough of space in the current cylinder group, so we
1223	 * can relocate just a few blocks, and let the summary  information
1224	 * grow in place where it is right now.
1225	 */
1226	DBG_TRC;
1227
1228	cbase = cgbase(&osblock, ocscg);	/* old and new are equal */
1229	dupper = sblock.fs_csaddr - cbase +
1230	    howmany(sblock.fs_cssize, sblock.fs_fsize);
1231	odupper = osblock.fs_csaddr - cbase +
1232	    howmany(osblock.fs_cssize, osblock.fs_fsize);
1233
1234	sblock.fs_dsize -= dupper-odupper;
1235
1236	/*
1237	 * Allocate the space for the array of blocks to be relocated.
1238	 */
1239 	bp = (struct gfs_bpp *)calloc(((dupper-odupper) / sblock.fs_frag + 2),
1240	    sizeof(struct gfs_bpp));
1241	if (bp == NULL)
1242		errx(1, "calloc failed");
1243	memset((char *)bp, 0, ((dupper-odupper) / sblock.fs_frag + 2) *
1244	    sizeof(struct gfs_bpp));
1245
1246	/*
1247	 * Lock all new frags needed for the cylinder group summary. This  is
1248	 * done per fragment in the first and last block of the new  required
1249	 * area, and per block for all other blocks.
1250	 *
1251	 * Handle the first new  block here (but only if some fragments where
1252	 * already used for the cylinder summary).
1253	 */
1254	ind = 0;
1255	frag_adjust(odupper, -1);
1256	for (d = odupper; ((d < dupper) && (d % sblock.fs_frag)); d++) {
1257		DBG_PRINT1("scg first frag check loop d=%jd\n", (intmax_t)d);
1258		if (isclr(cg_blksfree(&acg), d)) {
1259			if (!ind) {
1260				bp[ind].old = d / sblock.fs_frag;
1261				bp[ind].flags|=GFS_FL_FIRST;
1262				if (roundup(d, sblock.fs_frag) >= dupper)
1263					bp[ind].flags |= GFS_FL_LAST;
1264				ind++;
1265			}
1266		} else {
1267			clrbit(cg_blksfree(&acg), d);
1268			acg.cg_cs.cs_nffree--;
1269			sblock.fs_cstotal.cs_nffree--;
1270		}
1271		/*
1272		 * No cluster handling is needed here, as there was at least
1273		 * one  fragment in use by the cylinder summary in  the  old
1274		 * filesystem.
1275		 * No block - free counter handling here as this block was not
1276		 * a free block.
1277		 */
1278	}
1279	frag_adjust(odupper, 1);
1280
1281	/*
1282	 * Handle all needed complete blocks here.
1283	 */
1284	for (; d + sblock.fs_frag <= dupper; d += sblock.fs_frag) {
1285		DBG_PRINT1("scg block check loop d=%jd\n", (intmax_t)d);
1286		if (!isblock(&sblock, cg_blksfree(&acg), d / sblock.fs_frag)) {
1287			for (f = d; f < d + sblock.fs_frag; f++) {
1288				if (isset(cg_blksfree(&aocg), f)) {
1289					acg.cg_cs.cs_nffree--;
1290					sblock.fs_cstotal.cs_nffree--;
1291				}
1292			}
1293			clrblock(&sblock, cg_blksfree(&acg), d / sblock.fs_frag);
1294			bp[ind].old = d / sblock.fs_frag;
1295			ind++;
1296		} else {
1297			clrblock(&sblock, cg_blksfree(&acg), d / sblock.fs_frag);
1298			acg.cg_cs.cs_nbfree--;
1299			sblock.fs_cstotal.cs_nbfree--;
1300			if (sblock.fs_contigsumsize > 0) {
1301				clrbit(cg_clustersfree(&acg), d / sblock.fs_frag);
1302				for (lcs = 0, l = (d / sblock.fs_frag) + 1;
1303				    lcs < sblock.fs_contigsumsize;
1304				    l++, lcs++) {
1305					if (isclr(cg_clustersfree(&acg), l))
1306						break;
1307				}
1308				if (lcs < sblock.fs_contigsumsize) {
1309					cg_clustersum(&acg)[lcs + 1]--;
1310					if (lcs)
1311						cg_clustersum(&acg)[lcs]++;
1312				}
1313			}
1314		}
1315		/*
1316		 * No fragment counter handling is needed here, as this finally
1317		 * doesn't change after the relocation.
1318		 */
1319	}
1320
1321	/*
1322	 * Handle all fragments needed in the last new affected block.
1323	 */
1324	if (d < dupper) {
1325		frag_adjust(dupper - 1, -1);
1326
1327		if (isblock(&sblock, cg_blksfree(&acg), d / sblock.fs_frag)) {
1328			acg.cg_cs.cs_nbfree--;
1329			sblock.fs_cstotal.cs_nbfree--;
1330			acg.cg_cs.cs_nffree+=sblock.fs_frag;
1331			sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1332			if (sblock.fs_contigsumsize > 0) {
1333				clrbit(cg_clustersfree(&acg), d / sblock.fs_frag);
1334				for (lcs = 0, l = (d / sblock.fs_frag) + 1;
1335				    lcs < sblock.fs_contigsumsize;
1336				    l++, lcs++) {
1337					if (isclr(cg_clustersfree(&acg), l))
1338						break;
1339				}
1340				if (lcs < sblock.fs_contigsumsize) {
1341					cg_clustersum(&acg)[lcs + 1]--;
1342					if (lcs)
1343						cg_clustersum(&acg)[lcs]++;
1344				}
1345			}
1346		}
1347
1348		for (; d < dupper; d++) {
1349			DBG_PRINT1("scg second frag check loop d=%jd\n",
1350			    (intmax_t)d);
1351			if (isclr(cg_blksfree(&acg), d)) {
1352				bp[ind].old = d / sblock.fs_frag;
1353				bp[ind].flags |= GFS_FL_LAST;
1354			} else {
1355				clrbit(cg_blksfree(&acg), d);
1356				acg.cg_cs.cs_nffree--;
1357				sblock.fs_cstotal.cs_nffree--;
1358			}
1359		}
1360		if (bp[ind].flags & GFS_FL_LAST) /* we have to advance here */
1361			ind++;
1362		frag_adjust(dupper - 1, 1);
1363	}
1364
1365	/*
1366	 * If we found a block to relocate just do so.
1367	 */
1368	if (ind) {
1369		for (i = 0; i < ind; i++) {
1370			if (!bp[i].old) { /* no more blocks listed */
1371				/*
1372				 * XXX	A relative blocknumber should not be
1373				 *	zero,   which  is   not   explicitly
1374				 *	guaranteed by our code.
1375				 */
1376				break;
1377			}
1378			/*
1379			 * Allocate a complete block in the same (current)
1380			 * cylinder group.
1381			 */
1382			bp[i].new = alloc() / sblock.fs_frag;
1383
1384			/*
1385			 * There is no frag_adjust() needed for the new block
1386			 * as it will have no fragments yet :-).
1387			 */
1388			for (f = bp[i].old * sblock.fs_frag,
1389			    g = bp[i].new * sblock.fs_frag;
1390			    f < (bp[i].old + 1) * sblock.fs_frag;
1391			    f++, g++) {
1392				if (isset(cg_blksfree(&aocg), f)) {
1393					setbit(cg_blksfree(&acg), g);
1394					acg.cg_cs.cs_nffree++;
1395					sblock.fs_cstotal.cs_nffree++;
1396				}
1397			}
1398
1399			/*
1400			 * Special handling is required if this was the  first
1401			 * block. We have to consider the fragments which were
1402			 * used by the cylinder summary in the original  block
1403			 * which  re to be free in the copy of our  block.  We
1404			 * have  to be careful if this first block happens  to
1405			 * be also the last block to be relocated.
1406			 */
1407			if (bp[i].flags & GFS_FL_FIRST) {
1408				for (f = bp[i].old * sblock.fs_frag,
1409				    g = bp[i].new * sblock.fs_frag;
1410				    f < odupper;
1411				    f++, g++) {
1412					setbit(cg_blksfree(&acg), g);
1413					acg.cg_cs.cs_nffree++;
1414					sblock.fs_cstotal.cs_nffree++;
1415				}
1416				if (!(bp[i].flags & GFS_FL_LAST))
1417					frag_adjust(bp[i].new * sblock.fs_frag, 1);
1418			}
1419
1420			/*
1421			 * Special handling is required if this is the last
1422			 * block to be relocated.
1423			 */
1424			if (bp[i].flags & GFS_FL_LAST) {
1425				frag_adjust(bp[i].new * sblock.fs_frag, 1);
1426				frag_adjust(bp[i].old * sblock.fs_frag, -1);
1427				for (f = dupper;
1428				    f < roundup(dupper, sblock.fs_frag);
1429				    f++) {
1430					if (isclr(cg_blksfree(&acg), f)) {
1431						setbit(cg_blksfree(&acg), f);
1432						acg.cg_cs.cs_nffree++;
1433						sblock.fs_cstotal.cs_nffree++;
1434					}
1435				}
1436				frag_adjust(bp[i].old * sblock.fs_frag, 1);
1437			}
1438
1439			/*
1440			 * !!! Attach the cylindergroup offset here.
1441			 */
1442			bp[i].old += cbase / sblock.fs_frag;
1443			bp[i].new += cbase / sblock.fs_frag;
1444
1445			/*
1446			 * Copy the content of the block.
1447			 */
1448			/*
1449			 * XXX	Here we will have to implement a copy on write
1450			 *	in the case we have any active snapshots.
1451			 */
1452			rdfs(fsbtodb(&sblock, bp[i].old * sblock.fs_frag),
1453			    (size_t)sblock.fs_bsize, (void *)&ablk, fsi);
1454			wtfs(fsbtodb(&sblock, bp[i].new * sblock.fs_frag),
1455			    (size_t)sblock.fs_bsize, (void *)&ablk, fso, Nflag);
1456			DBG_DUMP_HEX(&sblock, "copied full block",
1457			    (unsigned char *)&ablk);
1458
1459			DBG_PRINT2("scg (%jd->%jd) block relocated\n",
1460			    (intmax_t)bp[i].old, (intmax_t)bp[i].new);
1461		}
1462
1463		/*
1464		 * Now we have to update all references to any fragment which
1465		 * belongs  to any block relocated. We iterate now  over  all
1466		 * cylinder  groups,  within those over all non  zero  length
1467		 * inodes.
1468		 */
1469		for (cylno = 0; cylno < osblock.fs_ncg; cylno++) {
1470			DBG_PRINT1("scg doing cg (%d)\n", cylno);
1471			for (inc = osblock.fs_ipg - 1; inc > 0; inc--) {
1472				updrefs(cylno, (ino_t)inc, bp, fsi, fso, Nflag);
1473			}
1474		}
1475
1476		/*
1477		 * All inodes are checked, now make sure the number of
1478		 * references found make sense.
1479		 */
1480		for (i = 0; i < ind; i++) {
1481			if (!bp[i].found || (bp[i].found > sblock.fs_frag)) {
1482				warnx("error: %jd refs found for block %jd.",
1483				    (intmax_t)bp[i].found, (intmax_t)bp[i].old);
1484			}
1485
1486		}
1487	}
1488	/*
1489	 * The following statistics are not changed here:
1490	 *     sblock.fs_cstotal.cs_ndir
1491	 *     sblock.fs_cstotal.cs_nifree
1492	 * The following statistics were already updated on the fly:
1493	 *     sblock.fs_cstotal.cs_nffree
1494	 *     sblock.fs_cstotal.cs_nbfree
1495	 * As the statistics for this cylinder group are ready, copy it to
1496	 * the summary information array.
1497	 */
1498
1499	*cs = acg.cg_cs;
1500
1501	/*
1502	 * Write summary cylinder group back to disk.
1503	 */
1504	wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)), (size_t)sblock.fs_cgsize,
1505	    (void *)&acg, fso, Nflag);
1506	DBG_PRINT0("scg written\n");
1507	DBG_DUMP_CG(&sblock, "new summary cg", &acg);
1508
1509	DBG_LEAVE;
1510}
1511
1512/* ************************************************************** rdfs ***** */
1513/*
1514 * Here we read some block(s) from disk.
1515 */
1516static void
1517rdfs(daddr_t bno, size_t size, void *bf, int fsi)
1518{
1519	DBG_FUNC("rdfs")
1520	ssize_t	n;
1521
1522
1523	if (bno < 0) {
1524		err(32, "rdfs: attempting to read negative block number");
1525	}
1526	if (lseek(fsi, (off_t)bno * DEV_BSIZE, 0) < 0) {
1527		err(33, "rdfs: seek error: %jd", (intmax_t)bno);
1528	}
1529	n = read(fsi, bf, size);
1530	if (n != (ssize_t)size) {
1531		err(34, "rdfs: read error: %jd", (intmax_t)bno);
1532	}
1533
1534	DBG_LEAVE;
1535}
1536
1537/* ************************************************************** wtfs ***** */
1538/*
1539 * Here we write some block(s) to disk.
1540 */
1541static void
1542wtfs(daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag)
1543{
1544	DBG_FUNC("wtfs")
1545	ssize_t	n;
1546
1547	DBG_ENTER;
1548
1549	if (Nflag) {
1550		DBG_LEAVE;
1551		return;
1552	}
1553	if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0) {
1554		err(35, "wtfs: seek error: %ld", (long)bno);
1555	}
1556	n = write(fso, bf, size);
1557	if (n != (ssize_t)size) {
1558		err(36, "wtfs: write error: %ld", (long)bno);
1559	}
1560
1561	DBG_LEAVE;
1562}
1563
1564/* ************************************************************* alloc ***** */
1565/*
1566 * Here we allocate a free block in the current cylinder group. It is assumed,
1567 * that  acg contains the current cylinder group. As we may take a block  from
1568 * somewhere in the filesystem we have to handle cluster summary here.
1569 */
1570static daddr_t
1571alloc(void)
1572{
1573	DBG_FUNC("alloc")
1574	daddr_t	d, blkno;
1575	int	lcs1, lcs2;
1576	int	l;
1577	int	csmin, csmax;
1578	int	dlower, dupper, dmax;
1579
1580	DBG_ENTER;
1581
1582	if (acg.cg_magic != CG_MAGIC) {
1583		warnx("acg: bad magic number");
1584		DBG_LEAVE;
1585		return (0);
1586	}
1587	if (acg.cg_cs.cs_nbfree == 0) {
1588		warnx("error: cylinder group ran out of space");
1589		DBG_LEAVE;
1590		return (0);
1591	}
1592	/*
1593	 * We start seeking for free blocks only from the space available after
1594	 * the  end of the new grown cylinder summary. Otherwise we allocate  a
1595	 * block here which we have to relocate a couple of seconds later again
1596	 * again, and we are not prepared to to this anyway.
1597	 */
1598	blkno = -1;
1599	dlower = cgsblock(&sblock, acg.cg_cgx) - cgbase(&sblock, acg.cg_cgx);
1600	dupper = cgdmin(&sblock, acg.cg_cgx) - cgbase(&sblock, acg.cg_cgx);
1601	dmax = cgbase(&sblock, acg.cg_cgx) + sblock.fs_fpg;
1602	if (dmax > sblock.fs_size) {
1603		dmax = sblock.fs_size;
1604	}
1605	dmax -= cgbase(&sblock, acg.cg_cgx); /* retransform into cg */
1606	csmin=sblock.fs_csaddr-cgbase(&sblock, acg.cg_cgx);
1607	csmax = csmin + howmany(sblock.fs_cssize, sblock.fs_fsize);
1608	DBG_PRINT3("seek range: dl=%d, du=%d, dm=%d\n", dlower, dupper, dmax);
1609	DBG_PRINT2("range cont: csmin=%d, csmax=%d\n", csmin, csmax);
1610
1611	for (d = 0; (d < dlower && blkno == -1); d += sblock.fs_frag) {
1612		if (d >= csmin && d <= csmax) {
1613			continue;
1614		}
1615		if (isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1616		    d))) {
1617			blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1618			break;
1619		}
1620	}
1621	for (d = dupper; (d < dmax && blkno == -1); d += sblock.fs_frag) {
1622		if (d >= csmin && d <= csmax) {
1623			continue;
1624		}
1625		if (isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1626		    d))) {
1627			blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1628			break;
1629		}
1630	}
1631	if (blkno == -1) {
1632		warnx("internal error: couldn't find promised block in cg");
1633		DBG_LEAVE;
1634		return (0);
1635	}
1636
1637	/*
1638	 * This is needed if the block was found already in the first loop.
1639	 */
1640	d = blkstofrags(&sblock, blkno);
1641
1642	clrblock(&sblock, cg_blksfree(&acg), blkno);
1643	if (sblock.fs_contigsumsize > 0) {
1644		/*
1645		 * Handle the cluster allocation bitmap.
1646		 */
1647		clrbit(cg_clustersfree(&acg), blkno);
1648		/*
1649		 * We  possibly have split a cluster here, so we have  to  do
1650		 * recalculate the sizes of the remaining cluster halves now,
1651		 * and use them for updating the cluster summary information.
1652		 *
1653		 * Lets start with the blocks before our allocated block ...
1654		 */
1655		for (lcs1 = 0, l = blkno - 1; lcs1 < sblock.fs_contigsumsize;
1656		    l--, lcs1++) {
1657			if (isclr(cg_clustersfree(&acg), l))
1658				break;
1659		}
1660		/*
1661		 * ... and continue with the blocks right after our allocated
1662		 * block.
1663		 */
1664		for (lcs2 = 0, l = blkno + 1; lcs2 < sblock.fs_contigsumsize;
1665		    l++, lcs2++) {
1666			if (isclr(cg_clustersfree(&acg), l))
1667				break;
1668		}
1669
1670		/*
1671		 * Now update all counters.
1672		 */
1673		cg_clustersum(&acg)[MIN(lcs1 + lcs2 + 1, sblock.fs_contigsumsize)]--;
1674		if (lcs1)
1675			cg_clustersum(&acg)[lcs1]++;
1676		if (lcs2)
1677			cg_clustersum(&acg)[lcs2]++;
1678	}
1679	/*
1680	 * Update all statistics based on blocks.
1681	 */
1682	acg.cg_cs.cs_nbfree--;
1683	sblock.fs_cstotal.cs_nbfree--;
1684
1685	DBG_LEAVE;
1686	return (d);
1687}
1688
1689/* *********************************************************** isblock ***** */
1690/*
1691 * Here  we check if all frags of a block are free. For more details  again
1692 * please see the source of newfs(8), as this function is taken over almost
1693 * unchanged.
1694 */
1695static int
1696isblock(struct fs *fs, unsigned char *cp, int h)
1697{
1698	DBG_FUNC("isblock")
1699	unsigned char	mask;
1700
1701	DBG_ENTER;
1702
1703	switch (fs->fs_frag) {
1704	case 8:
1705		DBG_LEAVE;
1706		return (cp[h] == 0xff);
1707	case 4:
1708		mask = 0x0f << ((h & 0x1) << 2);
1709		DBG_LEAVE;
1710		return ((cp[h >> 1] & mask) == mask);
1711	case 2:
1712		mask = 0x03 << ((h & 0x3) << 1);
1713		DBG_LEAVE;
1714		return ((cp[h >> 2] & mask) == mask);
1715	case 1:
1716		mask = 0x01 << (h & 0x7);
1717		DBG_LEAVE;
1718		return ((cp[h >> 3] & mask) == mask);
1719	default:
1720		fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
1721		DBG_LEAVE;
1722		return (0);
1723	}
1724}
1725
1726/* ********************************************************** clrblock ***** */
1727/*
1728 * Here we allocate a complete block in the block map. For more details again
1729 * please  see the source of newfs(8), as this function is taken over  almost
1730 * unchanged.
1731 */
1732static void
1733clrblock(struct fs *fs, unsigned char *cp, int h)
1734{
1735	DBG_FUNC("clrblock")
1736
1737	DBG_ENTER;
1738
1739	switch ((fs)->fs_frag) {
1740	case 8:
1741		cp[h] = 0;
1742		break;
1743	case 4:
1744		cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1745		break;
1746	case 2:
1747		cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1748		break;
1749	case 1:
1750		cp[h >> 3] &= ~(0x01 << (h & 0x7));
1751		break;
1752	default:
1753		warnx("clrblock bad fs_frag %d", fs->fs_frag);
1754		break;
1755	}
1756
1757	DBG_LEAVE;
1758}
1759
1760/* ********************************************************** setblock ***** */
1761/*
1762 * Here we free a complete block in the free block map. For more details again
1763 * please  see the source of newfs(8), as this function is taken  over  almost
1764 * unchanged.
1765 */
1766static void
1767setblock(struct fs *fs, unsigned char *cp, int h)
1768{
1769	DBG_FUNC("setblock")
1770
1771	DBG_ENTER;
1772
1773	switch (fs->fs_frag) {
1774	case 8:
1775		cp[h] = 0xff;
1776		break;
1777	case 4:
1778		cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1779		break;
1780	case 2:
1781		cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1782		break;
1783	case 1:
1784		cp[h >> 3] |= (0x01 << (h & 0x7));
1785		break;
1786	default:
1787		warnx("setblock bad fs_frag %d", fs->fs_frag);
1788		break;
1789	}
1790
1791	DBG_LEAVE;
1792}
1793
1794/* ************************************************************ ginode ***** */
1795/*
1796 * This function provides access to an individual inode. We find out in which
1797 * block  the  requested inode is located, read it from disk if  needed,  and
1798 * return  the pointer into that block. We maintain a cache of one  block  to
1799 * not  read the same block again and again if we iterate linearly  over  all
1800 * inodes.
1801 */
1802static union dinode *
1803ginode(ino_t inumber, int fsi, int cg)
1804{
1805	DBG_FUNC("ginode")
1806	static ino_t	startinum = 0;	/* first inode in cached block */
1807
1808	DBG_ENTER;
1809
1810	/*
1811	 * The inumber passed in is relative to the cg, so use it here to see
1812	 * if the inode has been allocated yet.
1813	 */
1814	if (isclr(cg_inosused(&aocg), inumber)) {
1815		DBG_LEAVE;
1816		return NULL;
1817	}
1818	/*
1819	 * Now make the inumber relative to the entire inode space so it can
1820	 * be sanity checked.
1821	 */
1822	inumber += (cg * sblock.fs_ipg);
1823	if (inumber < ROOTINO) {
1824		DBG_LEAVE;
1825		return NULL;
1826	}
1827	if (inumber > maxino)
1828		errx(8, "bad inode number %llu to ginode",
1829		    (unsigned long long)inumber);
1830	if (startinum == 0 ||
1831	    inumber < startinum || inumber >= startinum + INOPB(&sblock)) {
1832		inoblk = fsbtodb(&sblock, ino_to_fsba(&sblock, inumber));
1833		rdfs(inoblk, (size_t)sblock.fs_bsize, inobuf, fsi);
1834		startinum = (inumber / INOPB(&sblock)) * INOPB(&sblock);
1835	}
1836	DBG_LEAVE;
1837	if (sblock.fs_magic == FS_UFS1_MAGIC)
1838		return (union dinode *)((uintptr_t)inobuf +
1839		    (inumber % INOPB(&sblock)) * sizeof(struct ufs1_dinode));
1840	return (union dinode *)((uintptr_t)inobuf +
1841	    (inumber % INOPB(&sblock)) * sizeof(struct ufs2_dinode));
1842}
1843
1844/* ****************************************************** charsperline ***** */
1845/*
1846 * Figure out how many lines our current terminal has. For more details again
1847 * please  see the source of newfs(8), as this function is taken over  almost
1848 * unchanged.
1849 */
1850static int
1851charsperline(void)
1852{
1853	DBG_FUNC("charsperline")
1854	int	columns;
1855	char	*cp;
1856	struct winsize	ws;
1857
1858	DBG_ENTER;
1859
1860	columns = 0;
1861	if (ioctl(0, TIOCGWINSZ, &ws) != -1) {
1862		columns = ws.ws_col;
1863	}
1864	if (columns == 0 && (cp = getenv("COLUMNS"))) {
1865		columns = atoi(cp);
1866	}
1867	if (columns == 0) {
1868		columns = 80;	/* last resort */
1869	}
1870
1871	DBG_LEAVE;
1872	return columns;
1873}
1874
1875/* ************************************************************** main ***** */
1876/*
1877 * growfs(8)  is a utility which allows to increase the size of  an  existing
1878 * ufs filesystem. Currently this can only be done on unmounted file  system.
1879 * It  recognizes some command line options to specify the new desired  size,
1880 * and  it does some basic checkings. The old filesystem size is  determined
1881 * and  after some more checks like we can really access the new  last  block
1882 * on the disk etc. we calculate the new parameters for the superblock. After
1883 * having  done  this we just call growfs() which will do  the  work.  Before
1884 * we finish the only thing left is to update the disklabel.
1885 * We still have to provide support for snapshots. Therefore we first have to
1886 * understand  what data structures are always replicated in the snapshot  on
1887 * creation,  for all other blocks we touch during our procedure, we have  to
1888 * keep the old blocks unchanged somewhere available for the snapshots. If we
1889 * are lucky, then we only have to handle our blocks to be relocated in  that
1890 * way.
1891 * Also  we  have to consider in what order we actually update  the  critical
1892 * data structures of the filesystem to make sure, that in case of a disaster
1893 * fsck(8) is still able to restore any lost data.
1894 * The  foreseen last step then will be to provide for growing  even  mounted
1895 * file  systems. There we have to extend the mount() system call to  provide
1896 * userland access to the filesystem locking facility.
1897 */
1898int
1899main(int argc, char **argv)
1900{
1901	DBG_FUNC("main")
1902	char	*device;
1903	int	ch;
1904	unsigned int	size = 0;
1905	unsigned int	Nflag = 0;
1906	int	ExpertFlag = 0;
1907	struct stat	st;
1908	struct disklabel	*lp;
1909	struct partition	*pp;
1910	int	i,fsi,fso;
1911	char	reply[5];
1912#ifdef FSMAXSNAP
1913	int	j;
1914#endif /* FSMAXSNAP */
1915
1916	DBG_ENTER;
1917
1918	while ((ch = getopt(argc, argv, "Nqs:vy")) != -1) {
1919		switch (ch) {
1920		case 'N':
1921			Nflag = 1;
1922			break;
1923		case 'q':
1924			quiet = 1;
1925			break;
1926		case 's':
1927			size = (size_t)atol(optarg);
1928			if (size < 1)
1929				usage();
1930			break;
1931		case 'v': /* for compatibility to newfs */
1932			break;
1933		case 'y':
1934			ExpertFlag = 1;
1935			break;
1936		case '?':
1937			/* FALLTHROUGH */
1938		default:
1939			usage();
1940		}
1941	}
1942	argc -= optind;
1943	argv += optind;
1944
1945	if (argc != 1)
1946		usage();
1947
1948	/*
1949	 * Rather than guessing, use opendev() to get the device
1950	 * name, which we open for reading.
1951	 */
1952	if ((fsi = opendev(*argv, O_RDONLY, 0, &device)) < 0)
1953		err(1, "%s", *argv);
1954
1955	/*
1956	 * Try to access our devices for writing ...
1957	 */
1958	if (Nflag) {
1959		fso = -1;
1960	} else {
1961		fso = open(device, O_WRONLY);
1962		if (fso < 0)
1963			err(1, "%s", device);
1964	}
1965
1966	/*
1967	 * Now we have a file descriptor for our device, fstat() it to
1968	 * figure out the partition number.
1969	 */
1970	if (fstat(fsi, &st) != 0)
1971		err(1, "%s: fstat()", device);
1972
1973	/*
1974	 * Try to read a label from the disk.  Then get the partition from the
1975	 * device minor number, using DISKPART().  Probably don't need to
1976	 * check against getmaxpartitions().
1977	 */
1978	lp = get_disklabel(fsi);
1979	if (DISKPART(st.st_rdev) < getmaxpartitions())
1980		pp = &lp->d_partitions[DISKPART(st.st_rdev)];
1981	else
1982		errx(1, "%s: invalid partition number %u",
1983		    device, DISKPART(st.st_rdev));
1984
1985	/*
1986	 * Check if that partition is suitable for growing a file system.
1987	 */
1988	if (pp->p_size < 1)
1989		errx(1, "partition is unavailable");
1990	if (pp->p_fstype != FS_BSDFFS)
1991		errx(1, "can only grow ffs partitions");
1992
1993	/*
1994	 * Read the current superblock, and take a backup.
1995	 */
1996	for (i = 0; sblock_try[i] != -1; i++) {
1997		sblockloc = sblock_try[i] / DEV_BSIZE;
1998		rdfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&(osblock), fsi);
1999		if ((osblock.fs_magic == FS_UFS1_MAGIC ||
2000		     (osblock.fs_magic == FS_UFS2_MAGIC &&
2001		      osblock.fs_sblockloc == sblock_try[i])) &&
2002		    osblock.fs_bsize <= MAXBSIZE &&
2003		    osblock.fs_bsize >= (int32_t) sizeof(struct fs))
2004			break;
2005	}
2006	if (sblock_try[i] == -1)
2007		errx(1, "superblock not recognized");
2008	if (osblock.fs_clean == 0)
2009		errx(1, "filesystem not clean - run fsck");
2010	if (sblock.fs_magic == FS_UFS1_MAGIC &&
2011	    (sblock.fs_ffs1_flags & FS_FLAGS_UPDATED) == 0)
2012		ffs1_sb_update(&sblock, sblock_try[i]);
2013	memcpy(&fsun1, &fsun2, sizeof(fsun2));
2014	maxino = sblock.fs_ncg * sblock.fs_ipg;
2015
2016	DBG_OPEN("/tmp/growfs.debug"); /* already here we need a superblock */
2017	DBG_DUMP_FS(&sblock, "old sblock");
2018
2019	/*
2020	 * Determine size to grow to. Default to the full size specified in
2021	 * the disk label.
2022	 */
2023	sblock.fs_size = dbtofsb(&osblock, pp->p_size);
2024	if (size != 0) {
2025		if (size > pp->p_size) {
2026			errx(1, "there is not enough space (%d < %d)",
2027			    pp->p_size, size);
2028		}
2029		sblock.fs_size = dbtofsb(&osblock, size);
2030	}
2031
2032	/*
2033	 * Are we really growing ?
2034	 */
2035	if (osblock.fs_size >= sblock.fs_size) {
2036		errx(1, "we are not growing (%jd->%jd)",
2037		    (intmax_t)osblock.fs_size, (intmax_t)sblock.fs_size);
2038	}
2039
2040
2041#ifdef FSMAXSNAP
2042	/*
2043	 * Check if we find an active snapshot.
2044	 */
2045	if (ExpertFlag == 0) {
2046		for (j = 0; j < FSMAXSNAP; j++) {
2047			if (sblock.fs_snapinum[j]) {
2048				errx(1, "active snapshot found in filesystem\n"
2049				    "	please remove all snapshots before "
2050				    "using growfs");
2051			}
2052			if (!sblock.fs_snapinum[j])	/* list is dense */
2053				break;
2054		}
2055	}
2056#endif
2057
2058	if (ExpertFlag == 0 && Nflag == 0) {
2059		printf("We strongly recommend you to make a backup "
2060		    "before growing the Filesystem\n\n"
2061		    " Did you backup your data (Yes/No) ? ");
2062		if (fgets(reply, (int)sizeof(reply), stdin) == NULL ||
2063		    strncasecmp(reply, "Yes", 3)) {
2064			printf("\n Nothing done \n");
2065			exit (0);
2066		}
2067	}
2068
2069	if (!quiet)
2070		printf("new file systemsize is: %jd frags\n",
2071		    (intmax_t)sblock.fs_size);
2072
2073	/*
2074	 * Try to access our new last block in the filesystem. Even if we
2075	 * later on realize we have to abort our operation, on that block
2076	 * there should be no data, so we can't destroy something yet.
2077	 */
2078	wtfs((daddr_t)pp->p_size-1, (size_t)DEV_BSIZE, (void *)&sblock,
2079	    fso, Nflag);
2080
2081	/*
2082	 * Now calculate new superblock values and check for reasonable
2083	 * bound for new filesystem size:
2084	 *     fs_size:    is derived from label or user input
2085	 *     fs_dsize:   should get updated in the routines creating or
2086	 *                 updating the cylinder groups on the fly
2087	 *     fs_cstotal: should get updated in the routines creating or
2088	 *                 updating the cylinder groups
2089	 */
2090
2091	/*
2092	 * Update the number of cylinders and cylinder groups in the file system.
2093	 */
2094	if (sblock.fs_magic == FS_UFS1_MAGIC) {
2095		sblock.fs_ncyl = sblock.fs_size * NSPF(&sblock) / sblock.fs_spc;
2096		if (sblock.fs_size * NSPF(&sblock) >
2097		    sblock.fs_ncyl * sblock.fs_spc)
2098		sblock.fs_ncyl++;
2099	}
2100	sblock.fs_ncg = howmany(sblock.fs_size, sblock.fs_fpg);
2101	maxino = sblock.fs_ncg * sblock.fs_ipg;
2102
2103	if (sblock.fs_size % sblock.fs_fpg != 0 &&
2104	    sblock.fs_size % sblock.fs_fpg < cgdmin(&sblock, sblock.fs_ncg)) {
2105		/*
2106		 * The space in the new last cylinder group is too small,
2107		 * so revert back.
2108		 */
2109		sblock.fs_ncg--;
2110		if (sblock.fs_magic == FS_UFS1_MAGIC)
2111			sblock.fs_ncyl = sblock.fs_ncg * sblock.fs_cpg;
2112		if (!quiet)
2113			printf("Warning: %jd sector(s) cannot be allocated.\n",
2114			    (intmax_t)fsbtodb(&sblock,
2115			    sblock.fs_size % sblock.fs_fpg));
2116		sblock.fs_size = sblock.fs_ncg * sblock.fs_fpg;
2117	}
2118
2119	/*
2120	 * Update the space for the cylinder group summary information in the
2121	 * respective cylinder group data area.
2122	 */
2123	sblock.fs_cssize =
2124	    fragroundup(&sblock, sblock.fs_ncg * sizeof(struct csum));
2125
2126	if (osblock.fs_size >= sblock.fs_size)
2127		errx(1, "not enough new space");
2128
2129	DBG_PRINT0("sblock calculated\n");
2130
2131	/*
2132	 * Ok, everything prepared, so now let's do the tricks.
2133	 */
2134	growfs(fsi, fso, Nflag);
2135
2136	/*
2137	 * Update the disk label.
2138	 */
2139	pp->p_fragblock =
2140	    DISKLABELV1_FFS_FRAGBLOCK(sblock.fs_fsize, sblock.fs_frag);
2141        pp->p_cpg = sblock.fs_fpg;
2142
2143	return_disklabel(fso, lp, Nflag);
2144	DBG_PRINT0("label rewritten\n");
2145
2146	close(fsi);
2147	if (fso > -1)
2148		close(fso);
2149
2150	DBG_CLOSE;
2151
2152	DBG_LEAVE;
2153	return 0;
2154}
2155
2156/* ************************************************** return_disklabel ***** */
2157/*
2158 * Write the updated disklabel back to disk.
2159 */
2160static void
2161return_disklabel(int fd, struct disklabel *lp, unsigned int Nflag)
2162{
2163	DBG_FUNC("return_disklabel")
2164	u_short	sum;
2165	u_short	*ptr;
2166
2167	DBG_ENTER;
2168
2169	if (!lp) {
2170		DBG_LEAVE;
2171		return;
2172	}
2173	if (!Nflag) {
2174		lp->d_checksum = 0;
2175		sum = 0;
2176		ptr = (u_short *)lp;
2177
2178		/*
2179		 * recalculate checksum
2180		 */
2181		while (ptr < (u_short *)&lp->d_partitions[lp->d_npartitions])
2182			sum ^= *ptr++;
2183		lp->d_checksum = sum;
2184
2185		if (ioctl(fd, DIOCWDINFO, (char *)lp) < 0)
2186			errx(1, "DIOCWDINFO failed");
2187	}
2188	free(lp);
2189
2190	DBG_LEAVE;
2191	return ;
2192}
2193
2194/* ***************************************************** get_disklabel ***** */
2195/*
2196 * Read the disklabel from disk.
2197 */
2198static struct disklabel *
2199get_disklabel(int fd)
2200{
2201	DBG_FUNC("get_disklabel")
2202	static struct	disklabel *lab;
2203
2204	DBG_ENTER;
2205
2206	lab = (struct disklabel *)malloc(sizeof(struct disklabel));
2207	if (!lab)
2208		errx(1, "malloc failed");
2209	if (ioctl(fd, DIOCGDINFO, (char *)lab) != 0)
2210		err(1, "DIOCGDINFO");
2211
2212	DBG_LEAVE;
2213	return (lab);
2214}
2215
2216
2217/* ************************************************************* usage ***** */
2218/*
2219 * Dump a line of usage.
2220 */
2221static void
2222usage(void)
2223{
2224	DBG_FUNC("usage")
2225
2226	DBG_ENTER;
2227
2228	fprintf(stderr, "usage: growfs [-Nqy] [-s size] special\n");
2229
2230	DBG_LEAVE;
2231	exit(1);
2232}
2233
2234/* *********************************************************** updclst ***** */
2235/*
2236 * This updates most parameters and the bitmap related to cluster. We have to
2237 * assume that sblock, osblock, acg are set up.
2238 */
2239static void
2240updclst(int block)
2241{
2242	DBG_FUNC("updclst")
2243	static int	lcs = 0;
2244
2245	DBG_ENTER;
2246
2247	if (sblock.fs_contigsumsize < 1)	/* no clustering */
2248		return;
2249
2250	/*
2251	 * update cluster allocation map
2252	 */
2253	setbit(cg_clustersfree(&acg), block);
2254
2255	/*
2256	 * update cluster summary table
2257	 */
2258	if (!lcs) {
2259		/*
2260		 * calculate size for the trailing cluster
2261		 */
2262		for (block--; lcs < sblock.fs_contigsumsize; block--, lcs++) {
2263			if (isclr(cg_clustersfree(&acg), block))
2264				break;
2265		}
2266	}
2267	if (lcs < sblock.fs_contigsumsize) {
2268		if (lcs)
2269			cg_clustersum(&acg)[lcs]--;
2270		lcs++;
2271		cg_clustersum(&acg)[lcs]++;
2272	}
2273
2274	DBG_LEAVE;
2275}
2276
2277/* *********************************************************** updrefs ***** */
2278/*
2279 * This updates all references to relocated blocks for the given inode.  The
2280 * inode is given as number within the cylinder group, and the number of the
2281 * cylinder group.
2282 */
2283static void
2284updrefs(int cg, ino_t in, struct gfs_bpp *bp, int fsi, int fso, unsigned int
2285    Nflag)
2286{
2287	DBG_FUNC("updrefs")
2288	daddr_t	len, lbn, numblks;
2289	daddr_t	iptr, blksperindir;
2290	union dinode	*ino;
2291	int		i, mode, inodeupdated;
2292
2293	DBG_ENTER;
2294
2295	ino = ginode(in, fsi, cg);
2296	if (ino == NULL) {
2297		DBG_LEAVE;
2298		return;
2299	}
2300	mode = DIP(ino, di_mode) & IFMT;
2301	if (mode != IFDIR && mode != IFREG && mode != IFLNK) {
2302		DBG_LEAVE;
2303		return; /* only check DIR, FILE, LINK */
2304	}
2305	if (mode == IFLNK &&
2306	    DIP(ino, di_size) < (u_int64_t) sblock.fs_maxsymlinklen) {
2307		DBG_LEAVE;
2308		return;	/* skip short symlinks */
2309	}
2310	numblks = howmany(DIP(ino, di_size), sblock.fs_bsize);
2311	if (numblks == 0) {
2312		DBG_LEAVE;
2313		return;	/* skip empty file */
2314	}
2315	if (DIP(ino, di_blocks) == 0) {
2316		DBG_LEAVE;
2317		return;	/* skip empty swiss cheesy file or old fastlink */
2318	}
2319	DBG_PRINT2("scg checking inode (%llu in %d)\n",
2320	    (unsigned long long)in, cg);
2321
2322	/*
2323	 * Check all the blocks.
2324	 */
2325	inodeupdated = 0;
2326	len = numblks < NDADDR ? numblks : NDADDR;
2327	for (i = 0; i < len; i++) {
2328		iptr = DIP(ino, di_db[i]);
2329		if (iptr == 0)
2330			continue;
2331		if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2332			DIP_SET(ino, di_db[i], iptr);
2333			inodeupdated++;
2334		}
2335	}
2336	DBG_PRINT0("~~scg direct blocks checked\n");
2337
2338	blksperindir = 1;
2339	len = numblks - NDADDR;
2340	lbn = NDADDR;
2341	for (i = 0; len > 0 && i < NIADDR; i++) {
2342		iptr = DIP(ino, di_ib[i]);
2343		if (iptr == 0)
2344			continue;
2345		if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2346			DIP_SET(ino, di_ib[i], iptr);
2347			inodeupdated++;
2348		}
2349		indirchk(blksperindir, lbn, iptr, numblks, bp, fsi, fso, Nflag);
2350		blksperindir *= NINDIR(&sblock);
2351		lbn += blksperindir;
2352		len -= blksperindir;
2353		DBG_PRINT1("scg indirect_%d blocks checked\n", i + 1);
2354	}
2355	if (inodeupdated)
2356		wtfs(inoblk, sblock.fs_bsize, inobuf, fso, Nflag);
2357
2358	DBG_LEAVE;
2359}
2360
2361/*
2362 * Recursively check all the indirect blocks.
2363 */
2364static void
2365indirchk(daddr_t blksperindir, daddr_t lbn, daddr_t blkno,
2366    daddr_t lastlbn, struct gfs_bpp *bp, int fsi, int fso, unsigned int Nflag)
2367{
2368	DBG_FUNC("indirchk")
2369	void *ibuf;
2370	int i, last;
2371	daddr_t iptr;
2372
2373	DBG_ENTER;
2374
2375	/* read in the indirect block. */
2376	ibuf = malloc(sblock.fs_bsize);
2377	if (!ibuf)
2378		errx(1, "malloc failed");
2379	rdfs(fsbtodb(&sblock, blkno), (size_t)sblock.fs_bsize, ibuf, fsi);
2380	last = howmany(lastlbn - lbn, blksperindir) < NINDIR(&sblock) ?
2381	    howmany(lastlbn - lbn, blksperindir) : NINDIR(&sblock);
2382	for (i = 0; i < last; i++) {
2383		if (sblock.fs_magic == FS_UFS1_MAGIC)
2384			iptr = ((int32_t *)ibuf)[i];
2385		else
2386			iptr = ((daddr_t *)ibuf)[i];
2387		if (iptr == 0)
2388			continue;
2389		if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2390			if (sblock.fs_magic == FS_UFS1_MAGIC)
2391				((int32_t *)ibuf)[i] = iptr;
2392			else
2393				((daddr_t *)ibuf)[i] = iptr;
2394		}
2395		if (blksperindir == 1)
2396			continue;
2397		indirchk(blksperindir / NINDIR(&sblock), lbn + blksperindir * i,
2398		    iptr, lastlbn, bp, fsi, fso, Nflag);
2399	}
2400	free(ibuf);
2401
2402	DBG_LEAVE;
2403}
2404
2405static void
2406ffs1_sb_update(struct fs *fs, daddr_t sbloc)
2407{
2408	fs->fs_flags = fs->fs_ffs1_flags;
2409	fs->fs_sblockloc = sbloc;
2410	fs->fs_maxbsize = fs->fs_bsize;
2411	fs->fs_time = fs->fs_ffs1_time;
2412	fs->fs_size = fs->fs_ffs1_size;
2413	fs->fs_dsize = fs->fs_ffs1_dsize;
2414	fs->fs_csaddr = fs->fs_ffs1_csaddr;
2415	fs->fs_cstotal.cs_ndir = fs->fs_ffs1_cstotal.cs_ndir;
2416	fs->fs_cstotal.cs_nbfree = fs->fs_ffs1_cstotal.cs_nbfree;
2417	fs->fs_cstotal.cs_nifree = fs->fs_ffs1_cstotal.cs_nifree;
2418	fs->fs_cstotal.cs_nffree = fs->fs_ffs1_cstotal.cs_nffree;
2419	fs->fs_ffs1_flags |= FS_FLAGS_UPDATED;
2420}
2421