growfs.c revision 1.38
1/*	$OpenBSD: growfs.c,v 1.38 2015/01/20 18:22:21 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>	/* DEV_BSIZE MAXBSIZE setbit isset isclr clrbit */
46#include <sys/types.h>
47#include <sys/disklabel.h>
48#include <sys/ioctl.h>
49#include <sys/dkio.h>
50#include <sys/stat.h>
51
52#include <stdio.h>
53#include <paths.h>
54#include <ctype.h>
55#include <err.h>
56#include <fcntl.h>
57#include <limits.h>
58#include <stdlib.h>
59#include <stdint.h>
60#include <string.h>
61#include <time.h>
62#include <unistd.h>
63#include <util.h>
64
65#include <ufs/ufs/dinode.h>
66#include <ufs/ffs/fs.h>
67
68#define MINIMUM(a, b)	(((a) < (b)) ? (a) : (b))
69#define MAXIMUM(a, b)	(((a) > (b)) ? (a) : (b))
70
71#include "debug.h"
72
73#define	rounddown(x, y)	(((x)/(y))*(y))
74#define	roundup(x, y)	((((x)+((y)-1))/(y))*(y))
75
76/* *************************************************** GLOBALS & TYPES ***** */
77#ifdef FS_DEBUG
78int	_dbg_lvl_ = (DL_INFO);	/* DL_TRC */
79#endif /* FS_DEBUG */
80
81static int quiet;		/* quiet flag */
82
83static union {
84	struct	fs fs;
85	char	pad[SBLOCKSIZE];
86} fsun1, fsun2;
87#define	sblock	fsun1.fs	/* the new superblock */
88#define	osblock	fsun2.fs	/* the old superblock */
89
90/*
91 * Possible superblock locations ordered from most to least likely.
92 */
93static int sblock_try[] = SBLOCKSEARCH;
94static daddr_t sblockloc;
95
96static union {
97	struct	cg cg;
98	char	pad[MAXBSIZE];
99} cgun1, cgun2;
100#define	acg	cgun1.cg	/* a cylinder cgroup (new) */
101#define	aocg	cgun2.cg	/* an old cylinder group */
102
103static char	ablk[MAXBSIZE];		/* a block */
104
105static struct csum	*fscs;	/* cylinder summary */
106
107union dinode {
108	struct ufs1_dinode dp1;
109	struct ufs2_dinode dp2;
110};
111#define	DIP(dp, field) \
112	((sblock.fs_magic == FS_UFS1_MAGIC) ? \
113	(uint32_t)(dp)->dp1.field : (dp)->dp2.field)
114#define	DIP_SET(dp, field, val) do { \
115	if (sblock.fs_magic == FS_UFS1_MAGIC) \
116		(dp)->dp1.field = (val); \
117	else \
118		(dp)->dp2.field = (val); \
119	} while (0)
120static daddr_t 	inoblk;			/* inode block address */
121static char		inobuf[MAXBSIZE];	/* inode block */
122ino_t			maxino;			/* last valid inode */
123
124/*
125 * An  array of elements of type struct gfs_bpp describes all blocks  to
126 * be relocated in order to free the space needed for the cylinder group
127 * summary for all cylinder groups located in the first cylinder group.
128 */
129struct gfs_bpp {
130	daddr_t		old;		/* old block number */
131	daddr_t		new;		/* new block number */
132#define GFS_FL_FIRST	1
133#define GFS_FL_LAST	2
134	unsigned int	flags;		/* special handling required */
135	int		found;		/* how many references were updated */
136};
137
138/* ******************************************************** PROTOTYPES ***** */
139static void	growfs(int, int, unsigned int);
140static void	rdfs(daddr_t, size_t, void *, int);
141static void	wtfs(daddr_t, size_t, void *, int, unsigned int);
142static daddr_t alloc(void);
143static int	charsperline(void);
144static void	usage(void);
145static int	isblock(struct fs *, unsigned char *, int);
146static void	clrblock(struct fs *, unsigned char *, int);
147static void	setblock(struct fs *, unsigned char *, int);
148static void	initcg(int, time_t, int, unsigned int);
149static void	updjcg(int, time_t, int, int, unsigned int);
150static void	updcsloc(time_t, int, int, unsigned int);
151static struct disklabel	*get_disklabel(int);
152static void	return_disklabel(int, struct disklabel *, unsigned int);
153static union dinode *ginode(ino_t, int, int);
154static void	frag_adjust(daddr_t, int);
155static int	cond_bl_upd(daddr_t *, struct gfs_bpp *, int, int,
156		    unsigned int);
157static void	updclst(int);
158static void	updrefs(int, ino_t, struct gfs_bpp *, int, int, unsigned int);
159static void	indirchk(daddr_t, daddr_t, daddr_t, daddr_t,
160		    struct gfs_bpp *, int, int, unsigned int);
161static void	ffs1_sb_update(struct fs *, daddr_t);
162
163/* ************************************************************ growfs ***** */
164/*
165 * Here  we actually start growing the filesystem. We basically  read  the
166 * cylinder  summary  from the first cylinder group as we want  to  update
167 * this  on  the fly during our various operations. First  we  handle  the
168 * changes in the former last cylinder group. Afterwards we create all new
169 * cylinder  groups.  Now  we handle the  cylinder  group  containing  the
170 * cylinder  summary  which  might result in a  relocation  of  the  whole
171 * structure.  In the end we write back the updated cylinder summary,  the
172 * new superblock, and slightly patched versions of the super block
173 * copies.
174 */
175static void
176growfs(int fsi, int fso, unsigned int Nflag)
177{
178	DBG_FUNC("growfs")
179	int	i;
180	int	cylno, j;
181	time_t	utime;
182	int	width;
183	char	tmpbuf[100];
184
185	DBG_ENTER;
186
187	time(&utime);
188
189	/*
190	 * Get the cylinder summary into the memory.
191	 */
192	fscs = calloc((size_t)1, (size_t)sblock.fs_cssize);
193	if (fscs == NULL)
194		errx(1, "calloc failed");
195	for (i = 0; i < osblock.fs_cssize; i += osblock.fs_bsize) {
196		rdfs(fsbtodb(&osblock, osblock.fs_csaddr +
197		    numfrags(&osblock, i)), (size_t)MINIMUM(osblock.fs_cssize - i,
198		    osblock.fs_bsize), (void *)(((char *)fscs)+i), fsi);
199	}
200
201#ifdef FS_DEBUG
202{
203	struct csum	*dbg_csp;
204	int	dbg_csc;
205	char	dbg_line[80];
206
207	dbg_csp = fscs;
208	for (dbg_csc = 0; dbg_csc < osblock.fs_ncg; dbg_csc++) {
209		snprintf(dbg_line, sizeof(dbg_line),
210		    "%d. old csum in old location", dbg_csc);
211		DBG_DUMP_CSUM(&osblock, dbg_line, dbg_csp++);
212	}
213}
214#endif /* FS_DEBUG */
215	DBG_PRINT0("fscs read\n");
216
217	/*
218	 * Do all needed changes in the former last cylinder group.
219	 */
220	updjcg(osblock.fs_ncg - 1, utime, fsi, fso, Nflag);
221
222	/*
223	 * Dump out summary information about filesystem.
224	 */
225#define B2MBFACTOR (1 / (1024.0 * 1024.0))
226	printf("growfs: %.1fMB (%jd sectors) block size %d, fragment size %d\n",
227	    (float)sblock.fs_size * sblock.fs_fsize * B2MBFACTOR,
228	    (intmax_t)fsbtodb(&sblock, sblock.fs_size), sblock.fs_bsize,
229	    sblock.fs_fsize);
230	printf("\tusing %d cylinder groups of %.2fMB, %d blks, %d inodes.\n",
231	    sblock.fs_ncg, (float)sblock.fs_fpg * sblock.fs_fsize * B2MBFACTOR,
232	    sblock.fs_fpg / sblock.fs_frag, sblock.fs_ipg);
233	if (sblock.fs_flags & FS_DOSOFTDEP)
234		printf("\twith soft updates\n");
235#undef B2MBFACTOR
236
237	/*
238	 * Now build the cylinders group blocks and
239	 * then print out indices of cylinder groups.
240	 */
241	if (!quiet)
242		printf("super-block backups (for fsck -b #) at:\n");
243	i = 0;
244	width = charsperline();
245
246	/*
247	 * Iterate for only the new cylinder groups.
248	 */
249	for (cylno = osblock.fs_ncg; cylno < sblock.fs_ncg; cylno++) {
250		initcg(cylno, utime, fso, Nflag);
251		if (quiet)
252			continue;
253		j = snprintf(tmpbuf, sizeof(tmpbuf), " %lld%s",
254		    fsbtodb(&sblock, cgsblock(&sblock, cylno)),
255		    cylno < (sblock.fs_ncg - 1) ? "," : "");
256		if (j >= sizeof(tmpbuf))
257			j = sizeof(tmpbuf) - 1;
258		if (j == -1 || i + j >= width) {
259			printf("\n");
260			i = 0;
261		}
262		i += j;
263		printf("%s", tmpbuf);
264		fflush(stdout);
265	}
266	if (!quiet)
267		printf("\n");
268
269	/*
270	 * Do all needed changes in the first cylinder group.
271	 * allocate blocks in new location
272	 */
273	updcsloc(utime, fsi, fso, Nflag);
274
275	/*
276	 * Now write the cylinder summary back to disk.
277	 */
278	for (i = 0; i < sblock.fs_cssize; i += sblock.fs_bsize) {
279		wtfs(fsbtodb(&sblock, sblock.fs_csaddr + numfrags(&sblock, i)),
280		    (size_t)MINIMUM(sblock.fs_cssize - i, sblock.fs_bsize),
281		    (void *)(((char *)fscs) + i), fso, Nflag);
282	}
283	DBG_PRINT0("fscs written\n");
284
285#ifdef FS_DEBUG
286{
287	struct csum	*dbg_csp;
288	int	dbg_csc;
289	char	dbg_line[80];
290
291	dbg_csp = fscs;
292	for (dbg_csc = 0; dbg_csc < sblock.fs_ncg; dbg_csc++) {
293		snprintf(dbg_line, sizeof(dbg_line),
294		    "%d. new csum in new location", dbg_csc);
295		DBG_DUMP_CSUM(&sblock, dbg_line, dbg_csp++);
296	}
297}
298#endif /* FS_DEBUG */
299
300	/*
301	 * Now write the new superblock back to disk.
302	 */
303	sblock.fs_time = utime;
304	sblock.fs_clean = 0;
305	if (sblock.fs_magic == FS_UFS1_MAGIC) {
306		sblock.fs_ffs1_time = (int32_t)sblock.fs_time;
307		sblock.fs_ffs1_size = (int32_t)sblock.fs_size;
308		sblock.fs_ffs1_dsize = (int32_t)sblock.fs_dsize;
309		sblock.fs_ffs1_csaddr = (int32_t)sblock.fs_csaddr;
310		sblock.fs_ffs1_cstotal.cs_ndir =
311		    (int32_t)sblock.fs_cstotal.cs_ndir;
312		sblock.fs_ffs1_cstotal.cs_nbfree =
313		    (int32_t)sblock.fs_cstotal.cs_nbfree;
314		sblock.fs_ffs1_cstotal.cs_nifree =
315		    (int32_t)sblock.fs_cstotal.cs_nifree;
316		sblock.fs_ffs1_cstotal.cs_nffree =
317		    (int32_t)sblock.fs_cstotal.cs_nffree;
318	}
319	wtfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
320	DBG_PRINT0("sblock written\n");
321	DBG_DUMP_FS(&sblock, "new initial sblock");
322
323	/*
324	 * Clean up the dynamic fields in our superblock copies.
325	 */
326	sblock.fs_fmod = 0;
327	sblock.fs_clean = 1;
328	sblock.fs_ronly = 0;
329	sblock.fs_cgrotor = 0;
330	sblock.fs_state = 0;
331	memset((void *)&sblock.fs_fsmnt, 0, sizeof(sblock.fs_fsmnt));
332	sblock.fs_flags &= FS_DOSOFTDEP;
333	if (sblock.fs_magic == FS_UFS1_MAGIC)
334		sblock.fs_ffs1_flags &= FS_DOSOFTDEP;
335
336	/*
337	 * XXX
338	 * The following fields are currently distributed from the  superblock
339	 * to the copies:
340	 *     fs_minfree
341	 *     fs_rotdelay
342	 *     fs_maxcontig
343	 *     fs_maxbpg
344	 *     fs_minfree,
345	 *     fs_optim
346	 *     fs_flags regarding SOFTPDATES
347	 *
348	 * We probably should rather change the summary for the cylinder group
349	 * statistics here to the value of what would be in there, if the file
350	 * system were created initially with the new size. Therefor we  still
351	 * need to find an easy way of calculating that.
352	 * Possibly we can try to read the first superblock copy and apply the
353	 * "diffed" stats between the old and new superblock by still  copying
354	 * certain parameters onto that.
355	 */
356
357	/*
358	 * Write out the duplicate superblocks.
359	 */
360	for (cylno = 0; cylno < sblock.fs_ncg; cylno++) {
361		wtfs(fsbtodb(&sblock, cgsblock(&sblock, cylno)),
362		    (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
363	}
364	DBG_PRINT0("sblock copies written\n");
365	DBG_DUMP_FS(&sblock, "new other sblocks");
366
367	DBG_LEAVE;
368}
369
370/* ************************************************************ initcg ***** */
371/*
372 * This creates a new cylinder group structure, for more details please  see
373 * the  source of newfs(8), as this function is taken over almost unchanged.
374 * As  this  is  never called for the  first  cylinder  group,  the  special
375 * provisions for that case are removed here.
376 */
377static void
378initcg(int cylno, time_t utime, int fso, unsigned int Nflag)
379{
380	DBG_FUNC("initcg")
381	static char *iobuf;
382	daddr_t d, dlower, dupper, blkno, start;
383	daddr_t i, cbase, dmax;
384	struct ufs1_dinode *dp1;
385	struct ufs2_dinode *dp2;
386	struct csum *cs;
387	ino_t j;
388	size_t iobufsize;
389
390	if (sblock.fs_bsize < SBLOCKSIZE)
391		iobufsize = SBLOCKSIZE + 3 * sblock.fs_bsize;
392	else
393		iobufsize = 4 * sblock.fs_bsize;
394
395	if (iobuf == NULL && (iobuf = malloc(iobufsize)) == NULL)
396		errx(37, "panic: cannot allocate I/O buffer");
397	bzero(iobuf, iobufsize);
398
399	/*
400	 * Determine block bounds for cylinder group.
401	 * Allow space for super block summary information in first
402	 * cylinder group.
403	 */
404	cbase = cgbase(&sblock, cylno);
405	dmax = cbase + sblock.fs_fpg;
406	if (dmax > sblock.fs_size)
407		dmax = sblock.fs_size;
408	dlower = cgsblock(&sblock, cylno) - cbase;
409	dupper = cgdmin(&sblock, cylno) - cbase;
410	if (cylno == 0) /* XXX fscs may be relocated */
411		dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
412	cs = &fscs[cylno];
413	memset(&acg, 0, sblock.fs_cgsize);
414	acg.cg_ffs2_time = utime;
415	acg.cg_magic = CG_MAGIC;
416	acg.cg_cgx = cylno;
417	acg.cg_ffs2_niblk = sblock.fs_ipg;
418	acg.cg_initediblk = MINIMUM(sblock.fs_ipg, 2 * INOPB(&sblock));
419	acg.cg_ndblk = dmax - cbase;
420	if (sblock.fs_contigsumsize > 0)
421		acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
422	start = sizeof(struct cg);
423	if (sblock.fs_magic == FS_UFS2_MAGIC) {
424		acg.cg_iusedoff = start;
425	} else {
426		if (cylno == sblock.fs_ncg - 1)
427			acg.cg_ncyl = sblock.fs_ncyl % sblock.fs_cpg;
428		else
429			acg.cg_ncyl = sblock.fs_cpg;
430		acg.cg_time = (int32_t)acg.cg_ffs2_time;
431		acg.cg_ffs2_time = 0;
432		acg.cg_niblk = (int16_t)acg.cg_ffs2_niblk;
433		acg.cg_ffs2_niblk = 0;
434		acg.cg_initediblk = 0;
435		acg.cg_btotoff = start;
436		acg.cg_boff = acg.cg_btotoff +
437		    sblock.fs_cpg * sizeof(int32_t);
438		acg.cg_iusedoff = acg.cg_boff +
439		    sblock.fs_cpg * sizeof(u_int16_t);
440	}
441	acg.cg_freeoff = acg.cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
442	acg.cg_nextfreeoff = acg.cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT);
443	if (sblock.fs_contigsumsize > 0) {
444		acg.cg_clustersumoff =
445		    roundup(acg.cg_nextfreeoff, sizeof(u_int32_t));
446		acg.cg_clustersumoff -= sizeof(u_int32_t);
447		acg.cg_clusteroff = acg.cg_clustersumoff +
448		    (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
449		acg.cg_nextfreeoff = acg.cg_clusteroff +
450		    howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
451	}
452	if (acg.cg_nextfreeoff > sblock.fs_cgsize) {
453		/*
454		 * This should never happen as we would have had that panic
455		 *     already on filesystem creation
456		 */
457		errx(37, "panic: cylinder group too big");
458	}
459	acg.cg_cs.cs_nifree += sblock.fs_ipg;
460	if (cylno == 0) {
461		for (i = 0; i < ROOTINO; i++) {
462			setbit(cg_inosused(&acg), i);
463			acg.cg_cs.cs_nifree--;
464		}
465	}
466	if (cylno > 0) {
467		/*
468		 * In cylno 0, beginning space is reserved
469		 * for boot and super blocks.
470		 */
471		for (d = 0; d < dlower; d += sblock.fs_frag) {
472			blkno = d / sblock.fs_frag;
473			setblock(&sblock, cg_blksfree(&acg), blkno);
474			if (sblock.fs_contigsumsize > 0)
475				setbit(cg_clustersfree(&acg), blkno);
476			acg.cg_cs.cs_nbfree++;
477		}
478		sblock.fs_dsize += dlower;
479	}
480	sblock.fs_dsize += acg.cg_ndblk - dupper;
481	if ((i = dupper % sblock.fs_frag)) {
482		acg.cg_frsum[sblock.fs_frag - i]++;
483		for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
484			setbit(cg_blksfree(&acg), dupper);
485			acg.cg_cs.cs_nffree++;
486		}
487	}
488	for (d = dupper; d + sblock.fs_frag <= acg.cg_ndblk;
489	     d += sblock.fs_frag) {
490		blkno = d / sblock.fs_frag;
491		setblock(&sblock, cg_blksfree(&acg), blkno);
492		if (sblock.fs_contigsumsize > 0)
493			setbit(cg_clustersfree(&acg), blkno);
494		acg.cg_cs.cs_nbfree++;
495	}
496	if (d < acg.cg_ndblk) {
497		acg.cg_frsum[acg.cg_ndblk - d]++;
498		for (; d < acg.cg_ndblk; d++) {
499			setbit(cg_blksfree(&acg), d);
500			acg.cg_cs.cs_nffree++;
501		}
502	}
503	if (sblock.fs_contigsumsize > 0) {
504		int32_t	*sump = cg_clustersum(&acg);
505		u_char	*mapp = cg_clustersfree(&acg);
506		int	map = *mapp++;
507		int	bit = 1;
508		int	run = 0;
509
510		for (i = 0; i < acg.cg_nclusterblks; i++) {
511			if ((map & bit) != 0)
512				run++;
513			else if (run != 0) {
514				if (run > sblock.fs_contigsumsize)
515					run = sblock.fs_contigsumsize;
516				sump[run]++;
517				run = 0;
518			}
519			if ((i & (CHAR_BIT - 1)) != CHAR_BIT - 1)
520				bit <<= 1;
521			else {
522				map = *mapp++;
523				bit = 1;
524			}
525		}
526		if (run != 0) {
527			if (run > sblock.fs_contigsumsize)
528				run = sblock.fs_contigsumsize;
529			sump[run]++;
530		}
531	}
532	sblock.fs_cstotal.cs_ndir += acg.cg_cs.cs_ndir;
533	sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
534	sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
535	sblock.fs_cstotal.cs_nifree += acg.cg_cs.cs_nifree;
536	*cs = acg.cg_cs;
537
538	/*
539	 * Write out the duplicate superblock, the cylinder group map
540	 * and two blocks worth of inodes in a single write.
541	 */
542	bcopy(&sblock, iobuf, SBLOCKSIZE);
543	start = sblock.fs_bsize > SBLOCKSIZE ? sblock.fs_bsize : SBLOCKSIZE;
544	bcopy(&acg, &iobuf[start], sblock.fs_cgsize);
545	start += sblock.fs_bsize;
546	dp1 = (struct ufs1_dinode *)&iobuf[start];
547	dp2 = (struct ufs2_dinode *)&iobuf[start];
548	for (i = MINIMUM(sblock.fs_ipg, 2 * INOPB(&sblock)); i != 0; i--) {
549		if (sblock.fs_magic == FS_UFS1_MAGIC) {
550			dp1->di_gen = arc4random();
551			dp1++;
552		} else {
553			dp2->di_gen = arc4random();
554			dp2++;
555		}
556	}
557	wtfs(fsbtodb(&sblock, cgsblock(&sblock, cylno)), iobufsize,
558	    iobuf, fso, Nflag);
559
560	/* Initialize inodes for FFS1. */
561	if (sblock.fs_magic == FS_UFS1_MAGIC) {
562		for (i = 2 * sblock.fs_frag; i < sblock.fs_ipg / INOPF(&sblock);
563		     i += sblock.fs_frag) {
564			dp1 = (struct ufs1_dinode *)&iobuf[start];
565			for (j = 0; j < INOPB(&sblock); j++) {
566				dp1->di_gen = arc4random();
567				dp1++;
568			}
569			wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno) + i),
570			    (size_t)sblock.fs_bsize, &iobuf[start], fso, Nflag);
571		}
572	}
573
574	DBG_DUMP_CG(&sblock, "new cg", &acg);
575
576	DBG_LEAVE;
577}
578
579/* ******************************************************* frag_adjust ***** */
580/*
581 * Here  we add or subtract (sign +1/-1) the available fragments in  a  given
582 * block to or from the fragment statistics. By subtracting before and adding
583 * after  an operation on the free frag map we can easy update  the  fragment
584 * statistic, which seems to be otherwise a rather complex operation.
585 */
586static void
587frag_adjust(daddr_t frag, int sign)
588{
589	DBG_FUNC("frag_adjust")
590	int fragsize;
591	int f;
592
593	DBG_ENTER;
594
595	fragsize = 0;
596	/*
597	 * Here frag only needs to point to any fragment in the block we want
598	 * to examine.
599	 */
600	for (f = rounddown(frag, sblock.fs_frag);
601	    f < roundup(frag + 1, sblock.fs_frag);
602	    f++) {
603		/*
604		 * Count contiguous free fragments.
605		 */
606		if (isset(cg_blksfree(&acg), f)) {
607			fragsize++;
608		} else {
609			if (fragsize && fragsize < sblock.fs_frag) {
610				/*
611				 * We found something in between.
612				 */
613				acg.cg_frsum[fragsize] += sign;
614				DBG_PRINT2("frag_adjust [%d]+=%d\n", fragsize,
615				    sign);
616			}
617			fragsize = 0;
618		}
619	}
620	if (fragsize && fragsize < sblock.fs_frag) {
621		/*
622		 * We found something.
623		 */
624		acg.cg_frsum[fragsize] += sign;
625		DBG_PRINT2("frag_adjust [%d]+=%d\n", fragsize, sign);
626	}
627	DBG_PRINT2("frag_adjust [[%d]]+=%d\n", fragsize, sign);
628
629	DBG_LEAVE;
630}
631
632/* ******************************************************* cond_bl_upd ***** */
633/*
634 * Here we conditionally update a pointer to a fragment. We check for all
635 * relocated blocks if any of its fragments is referenced by the current
636 * field,  and update the pointer to the respective fragment in  our  new
637 * block.  If  we find a reference we write back the  block  immediately,
638 * as there is no easy way for our general block reading engine to figure
639 * out if a write back operation is needed.
640 */
641static int
642cond_bl_upd(daddr_t *block, struct gfs_bpp *field, int fsi, int fso,
643    unsigned int Nflag)
644{
645	DBG_FUNC("cond_bl_upd")
646	struct gfs_bpp	*f;
647	daddr_t src, dst;
648	int fragnum;
649	void *ibuf;
650
651	DBG_ENTER;
652
653	for (f = field; f->old != 0; f++) {
654		src = *block;
655		if (fragstoblks(&sblock, src) != f->old)
656			continue;
657		/*
658		 * The fragment is part of the block, so update.
659		 */
660		dst = blkstofrags(&sblock, f->new);
661		fragnum = fragnum(&sblock, src);
662		*block = dst + fragnum;
663		f->found++;
664		DBG_PRINT3("scg (%jd->%jd)[%d] reference updated\n",
665		    (intmax_t)f->old, (intmax_t)f->new, fragnum);
666
667		/*
668		 * Copy the block back immediately.
669		 *
670		 * XXX	If src is from an indirect block we have
671		 *	to implement copy on write here in case of
672		 *	active snapshots.
673		 */
674		ibuf = malloc(sblock.fs_bsize);
675		if (!ibuf)
676			errx(1, "malloc failed");
677		src -= fragnum;
678		rdfs(fsbtodb(&sblock, src), (size_t)sblock.fs_bsize, ibuf, fsi);
679		wtfs(dst, (size_t)sblock.fs_bsize, ibuf, fso, Nflag);
680		free(ibuf);
681		/*
682		 * The same block can't be found again in this loop.
683		 */
684		return (1);
685	}
686
687	DBG_LEAVE;
688	return (0);
689}
690
691/* ************************************************************ updjcg ***** */
692/*
693 * Here we do all needed work for the former last cylinder group. It has to be
694 * changed  in  any case, even if the filesystem ended exactly on the  end  of
695 * this  group, as there is some slightly inconsistent handling of the  number
696 * of cylinders in the cylinder group. We start again by reading the  cylinder
697 * group from disk. If the last block was not fully available, we first handle
698 * the  missing  fragments, then we handle all new full blocks  in  that  file
699 * system  and  finally we handle the new last fragmented block  in  the  file
700 * system.  We again have to handle the fragment statistics rotational  layout
701 * tables and cluster summary during all those operations.
702 */
703static void
704updjcg(int cylno, time_t utime, int fsi, int fso, unsigned int Nflag)
705{
706	DBG_FUNC("updjcg")
707	daddr_t	cbase, dmax, dupper;
708	struct csum	*cs;
709	int	i, k;
710	int	j = 0;
711
712	DBG_ENTER;
713
714	/*
715	 * Read the former last (joining) cylinder group from disk, and make
716	 * a copy.
717	 */
718	rdfs(fsbtodb(&osblock, cgtod(&osblock, cylno)),
719	    (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
720	DBG_PRINT0("jcg read\n");
721	DBG_DUMP_CG(&sblock, "old joining cg", &aocg);
722
723	memcpy(&cgun1, &cgun2, sizeof(cgun2));
724
725	/*
726	 * If the cylinder group had already its new final size almost
727	 * nothing is to be done ... except:
728	 * For some reason the value of cg_ncyl in the last cylinder group has
729	 * to  be  zero instead of fs_cpg. As this is now no longer  the  last
730	 * cylinder group we have to change that value now to fs_cpg.
731	 */
732	if (cgbase(&osblock, cylno+1) == osblock.fs_size) {
733		if (sblock.fs_magic == FS_UFS1_MAGIC)
734			acg.cg_ncyl = sblock.fs_cpg;
735
736		wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
737		    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
738		DBG_PRINT0("jcg written\n");
739		DBG_DUMP_CG(&sblock, "new joining cg", &acg);
740
741		DBG_LEAVE;
742		return;
743	}
744
745	/*
746	 * Set up some variables needed later.
747	 */
748	cbase = cgbase(&sblock, cylno);
749	dmax = cbase + sblock.fs_fpg;
750	if (dmax > sblock.fs_size)
751		dmax = sblock.fs_size;
752	dupper = cgdmin(&sblock, cylno) - cbase;
753	if (cylno == 0)	/* XXX fscs may be relocated */
754		dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
755
756	/*
757	 * Set pointer to the cylinder summary for our cylinder group.
758	 */
759	cs = fscs + cylno;
760
761	/*
762	 * Touch the cylinder group, update all fields in the cylinder group as
763	 * needed, update the free space in the superblock.
764	 */
765	acg.cg_time = utime;
766	if (sblock.fs_magic == FS_UFS1_MAGIC) {
767		if (cylno == sblock.fs_ncg - 1) {
768			/*
769			 * This is still the last cylinder group.
770			 */
771			acg.cg_ncyl = sblock.fs_ncyl % sblock.fs_cpg;
772		} else {
773			acg.cg_ncyl = sblock.fs_cpg;
774		}
775	}
776	DBG_PRINT2("jcg dbg: %d %u", cylno, sblock.fs_ncg);
777#ifdef FS_DEBUG
778	if (sblock.fs_magic == FS_UFS1_MAGIC)
779		DBG_PRINT2("%d %u", acg.cg_ncyl, sblock.fs_cpg);
780#endif
781	DBG_PRINT0("\n");
782	acg.cg_ndblk = dmax - cbase;
783	sblock.fs_dsize += acg.cg_ndblk-aocg.cg_ndblk;
784	if (sblock.fs_contigsumsize > 0)
785		acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
786
787	/*
788	 * Now  we have to update the free fragment bitmap for our new  free
789	 * space.  There again we have to handle the fragmentation and  also
790	 * the  rotational  layout tables and the cluster summary.  This  is
791	 * also  done per fragment for the first new block if the  old  file
792	 * system end was not on a block boundary, per fragment for the  new
793	 * last block if the new filesystem end is not on a block boundary,
794	 * and per block for all space in between.
795	 *
796	 * Handle the first new block here if it was partially available
797	 * before.
798	 */
799	if (osblock.fs_size % sblock.fs_frag) {
800		if (roundup(osblock.fs_size, sblock.fs_frag) <= sblock.fs_size) {
801			/*
802			 * The new space is enough to fill at least this
803			 * block
804			 */
805			j = 0;
806			for (i = roundup(osblock.fs_size-cbase, sblock.fs_frag) - 1;
807			    i >= osblock.fs_size-cbase; i--) {
808				setbit(cg_blksfree(&acg), i);
809				acg.cg_cs.cs_nffree++;
810				j++;
811			}
812
813			/*
814			 * Check  if the fragment just created could join  an
815			 * already existing fragment at the former end of the
816			 * filesystem.
817			 */
818			if (isblock(&sblock, cg_blksfree(&acg),
819			    ((osblock.fs_size - cgbase(&sblock, cylno))/
820			    sblock.fs_frag))) {
821				/*
822				 * The block is now completely available.
823				 */
824				DBG_PRINT0("block was\n");
825				acg.cg_frsum[osblock.fs_size%sblock.fs_frag]--;
826				acg.cg_cs.cs_nbfree++;
827				acg.cg_cs.cs_nffree-=sblock.fs_frag;
828				k = rounddown(osblock.fs_size-cbase,
829				    sblock.fs_frag);
830				updclst((osblock.fs_size-cbase)/sblock.fs_frag);
831			} else {
832				/*
833				 * Lets rejoin a possible partially growed
834				 * fragment.
835				 */
836				k = 0;
837				while (isset(cg_blksfree(&acg), i) &&
838				    (i >= rounddown(osblock.fs_size - cbase,
839				    sblock.fs_frag))) {
840					i--;
841					k++;
842				}
843				if (k)
844					acg.cg_frsum[k]--;
845				acg.cg_frsum[k + j]++;
846			}
847		} else {
848			/*
849			 * We only grow by some fragments within this last
850			 * block.
851			 */
852			for (i = sblock.fs_size-cbase-1;
853			    i >= osblock.fs_size-cbase; i--) {
854				setbit(cg_blksfree(&acg), i);
855				acg.cg_cs.cs_nffree++;
856				j++;
857			}
858			/*
859			 * Lets rejoin a possible partially growed fragment.
860			 */
861			k = 0;
862			while (isset(cg_blksfree(&acg), i) &&
863			    (i >= rounddown(osblock.fs_size - cbase,
864			    sblock.fs_frag))) {
865				i--;
866				k++;
867			}
868			if (k)
869				acg.cg_frsum[k]--;
870			acg.cg_frsum[k + j]++;
871		}
872	}
873
874	/*
875	 * Handle all new complete blocks here.
876	 */
877	for (i = roundup(osblock.fs_size - cbase, sblock.fs_frag);
878	    i + sblock.fs_frag <= dmax-cbase;	/* XXX <= or only < ? */
879	    i += sblock.fs_frag) {
880		j = i / sblock.fs_frag;
881		setblock(&sblock, cg_blksfree(&acg), j);
882		updclst(j);
883		acg.cg_cs.cs_nbfree++;
884	}
885
886	/*
887	 * Handle the last new block if there are stll some new fragments left.
888	 * Here  we don't have to bother about the cluster summary or the  even
889	 * the rotational layout table.
890	 */
891	if (i < (dmax - cbase)) {
892		acg.cg_frsum[dmax - cbase - i]++;
893		for (; i < dmax - cbase; i++) {
894			setbit(cg_blksfree(&acg), i);
895			acg.cg_cs.cs_nffree++;
896		}
897	}
898
899	sblock.fs_cstotal.cs_nffree +=
900	    (acg.cg_cs.cs_nffree - aocg.cg_cs.cs_nffree);
901	sblock.fs_cstotal.cs_nbfree +=
902	    (acg.cg_cs.cs_nbfree - aocg.cg_cs.cs_nbfree);
903	/*
904	 * The following statistics are not changed here:
905	 *     sblock.fs_cstotal.cs_ndir
906	 *     sblock.fs_cstotal.cs_nifree
907	 * As the statistics for this cylinder group are ready, copy it to
908	 * the summary information array.
909	 */
910	*cs = acg.cg_cs;
911
912	/*
913	 * Write the updated "joining" cylinder group back to disk.
914	 */
915	wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize,
916	    (void *)&acg, fso, Nflag);
917	DBG_PRINT0("jcg written\n");
918	DBG_DUMP_CG(&sblock, "new joining cg", &acg);
919
920	DBG_LEAVE;
921}
922
923/* ********************************************************** updcsloc ***** */
924/*
925 * Here  we update the location of the cylinder summary. We have  two  possible
926 * ways of growing the cylinder summary.
927 * (1)	We can try to grow the summary in the current location, and  relocate
928 *	possibly used blocks within the current cylinder group.
929 * (2)	Alternatively we can relocate the whole cylinder summary to the first
930 *	new completely empty cylinder group. Once the cylinder summary is  no
931 *	longer in the beginning of the first cylinder group you should  never
932 *	use  a version of fsck which is not aware of the possibility to  have
933 *	this structure in a non standard place.
934 * Option (1) is considered to be less intrusive to the structure of the  file-
935 * system. So we try to stick to that whenever possible. If there is not enough
936 * space  in the cylinder group containing the cylinder summary we have to  use
937 * method  (2). In case of active snapshots in the filesystem we  probably  can
938 * completely avoid implementing copy on write if we stick to method (2) only.
939 */
940static void
941updcsloc(time_t utime, int fsi, int fso, unsigned int Nflag)
942{
943	DBG_FUNC("updcsloc")
944	struct csum	*cs;
945	int	ocscg, ncscg;
946	int	blocks;
947	daddr_t	cbase, dupper, odupper, d, f, g;
948	int	ind;
949	int	cylno, inc;
950	struct gfs_bpp	*bp;
951	int	i, l;
952	int	lcs = 0;
953	int	block;
954
955	DBG_ENTER;
956
957	if (howmany(sblock.fs_cssize, sblock.fs_fsize) ==
958	    howmany(osblock.fs_cssize, osblock.fs_fsize)) {
959		/*
960		 * No new fragment needed.
961		 */
962		DBG_LEAVE;
963		return;
964	}
965	ocscg = dtog(&osblock, osblock.fs_csaddr);
966	cs = fscs + ocscg;
967	blocks = 1+howmany(sblock.fs_cssize, sblock.fs_bsize)-
968	    howmany(osblock.fs_cssize, osblock.fs_bsize);
969
970	/*
971	 * Read original cylinder group from disk, and make a copy.
972	 * XXX	If Nflag is set in some very rare cases we now miss
973	 *	some changes done in updjcg by reading the unmodified
974	 *	block from disk.
975	 */
976	rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)),
977	    (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
978	DBG_PRINT0("oscg read\n");
979	DBG_DUMP_CG(&sblock, "old summary cg", &aocg);
980
981	memcpy(&cgun1, &cgun2, sizeof(cgun2));
982
983	/*
984	 * Touch the cylinder group, set up local variables needed later
985	 * and update the superblock.
986	 */
987	acg.cg_time = utime;
988
989	/*
990	 * XXX	In the case of having active snapshots we may need much more
991	 *	blocks for the copy on write. We need each block twice,  and
992	 *	also  up to 8*3 blocks for indirect blocks for all  possible
993	 *	references.
994	 */
995	if (/*((int)sblock.fs_time & 0x3) > 0 || */ cs->cs_nbfree < blocks) {
996		/*
997		 * There  is  not enough space in the old cylinder  group  to
998		 * relocate  all blocks as needed, so we relocate  the  whole
999		 * cylinder  group summary to a new group. We try to use  the
1000		 * first complete new cylinder group just created. Within the
1001		 * cylinder  group we align the area immediately  after  the
1002		 * cylinder  group  information location in order  to  be  as
1003		 * close as possible to the original implementation of ffs.
1004		 *
1005		 * First  we have to make sure we'll find enough space in  the
1006		 * new  cylinder  group. If not, then we  currently  give  up.
1007		 * We  start  with freeing everything which was  used  by  the
1008		 * fragments of the old cylinder summary in the current group.
1009		 * Now  we write back the group meta data, read in the  needed
1010		 * meta data from the new cylinder group, and start allocating
1011		 * within  that  group. Here we can assume, the  group  to  be
1012		 * completely empty. Which makes the handling of fragments and
1013		 * clusters a lot easier.
1014		 */
1015		DBG_TRC;
1016		if (sblock.fs_ncg-osblock.fs_ncg < 2)
1017			errx(2, "panic: not enough space");
1018
1019		/*
1020		 * Point "d" to the first fragment not used by the cylinder
1021		 * summary.
1022		 */
1023		d = osblock.fs_csaddr + (osblock.fs_cssize / osblock.fs_fsize);
1024
1025		/*
1026		 * Set up last cluster size ("lcs") already here. Calculate
1027		 * the size for the trailing cluster just behind where  "d"
1028		 * points to.
1029		 */
1030		if (sblock.fs_contigsumsize > 0) {
1031			for (block = howmany(d % sblock.fs_fpg, sblock.fs_frag),
1032			    lcs = 0; lcs < sblock.fs_contigsumsize;
1033			    block++, lcs++) {
1034				if (isclr(cg_clustersfree(&acg), block))
1035					break;
1036			}
1037		}
1038
1039		/*
1040		 * Point "d" to the last frag used by the cylinder summary.
1041		 */
1042		d--;
1043
1044		DBG_PRINT1("d=%jd\n", (intmax_t)d);
1045		if ((d + 1) % sblock.fs_frag) {
1046			/*
1047			 * The end of the cylinder summary is not a complete
1048			 * block.
1049			 */
1050			DBG_TRC;
1051			frag_adjust(d % sblock.fs_fpg, -1);
1052			for (; (d + 1) % sblock.fs_frag; d--) {
1053				DBG_PRINT1("d=%jd\n", (intmax_t)d);
1054				setbit(cg_blksfree(&acg), d % sblock.fs_fpg);
1055				acg.cg_cs.cs_nffree++;
1056				sblock.fs_cstotal.cs_nffree++;
1057			}
1058			/*
1059			 * Point  "d" to the last fragment of the  last
1060			 * (incomplete) block of the cylinder summary.
1061			 */
1062			d++;
1063			frag_adjust(d % sblock.fs_fpg, 1);
1064
1065			if (isblock(&sblock, cg_blksfree(&acg),
1066			    (d % sblock.fs_fpg) / sblock.fs_frag)) {
1067				DBG_PRINT1("d=%jd\n", (intmax_t)d);
1068				acg.cg_cs.cs_nffree -= sblock.fs_frag;
1069				acg.cg_cs.cs_nbfree++;
1070				sblock.fs_cstotal.cs_nffree -= sblock.fs_frag;
1071				sblock.fs_cstotal.cs_nbfree++;
1072				if (sblock.fs_contigsumsize > 0) {
1073					setbit(cg_clustersfree(&acg),
1074					    (d % sblock.fs_fpg) / sblock.fs_frag);
1075					if (lcs < sblock.fs_contigsumsize) {
1076						if (lcs) {
1077							cg_clustersum(&acg)
1078							    [lcs]--;
1079						}
1080						lcs++;
1081						cg_clustersum(&acg)[lcs]++;
1082					}
1083				}
1084			}
1085			/*
1086			 * Point "d" to the first fragment of the block before
1087			 * the last incomplete block.
1088			 */
1089			d--;
1090		}
1091
1092		DBG_PRINT1("d=%jd\n", (intmax_t)d);
1093		for (d = rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr;
1094		    d -= sblock.fs_frag) {
1095			DBG_TRC;
1096			DBG_PRINT1("d=%d\n", d);
1097			setblock(&sblock, cg_blksfree(&acg),
1098			    (d % sblock.fs_fpg) / sblock.fs_frag);
1099			acg.cg_cs.cs_nbfree++;
1100			sblock.fs_cstotal.cs_nbfree++;
1101			 if (sblock.fs_contigsumsize > 0) {
1102				setbit(cg_clustersfree(&acg),
1103				    (d % sblock.fs_fpg) / sblock.fs_frag);
1104				/*
1105				 * The last cluster size is already set up.
1106				 */
1107				if (lcs < sblock.fs_contigsumsize) {
1108					if (lcs) {
1109						cg_clustersum(&acg)[lcs]--;
1110					}
1111					lcs++;
1112					cg_clustersum(&acg)[lcs]++;
1113				}
1114			}
1115		}
1116		*cs = acg.cg_cs;
1117
1118		/*
1119		 * Now write the former cylinder group containing the cylinder
1120		 * summary back to disk.
1121		 */
1122		wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)),
1123		    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1124		DBG_PRINT0("oscg written\n");
1125		DBG_DUMP_CG(&sblock, "old summary cg", &acg);
1126
1127		/*
1128		 * Find the beginning of the new cylinder group containing the
1129		 * cylinder summary.
1130		 */
1131		sblock.fs_csaddr = cgdmin(&sblock, osblock.fs_ncg);
1132		ncscg = dtog(&sblock, sblock.fs_csaddr);
1133		cs = fscs + ncscg;
1134
1135
1136		/*
1137		 * If Nflag is specified, we would now read random data instead
1138		 * of an empty cg structure from disk. So we can't simulate that
1139		 * part for now.
1140		 */
1141		if (Nflag) {
1142			DBG_PRINT0("nscg update skipped\n");
1143			DBG_LEAVE;
1144			return;
1145		}
1146
1147		/*
1148		 * Read the future cylinder group containing the cylinder
1149		 * summary from disk, and make a copy.
1150		 */
1151		rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1152		    (size_t)sblock.fs_cgsize, &aocg, fsi);
1153		DBG_PRINT0("nscg read\n");
1154		DBG_DUMP_CG(&sblock, "new summary cg", &aocg);
1155
1156		memcpy(&cgun1, &cgun2, sizeof(cgun2));
1157
1158		/*
1159		 * Allocate all complete blocks used by the new cylinder
1160		 * summary.
1161		 */
1162		for (d = sblock.fs_csaddr; d + sblock.fs_frag <=
1163		    sblock.fs_csaddr + (sblock.fs_cssize / sblock.fs_fsize);
1164		    d += sblock.fs_frag) {
1165			clrblock(&sblock, cg_blksfree(&acg),
1166			    (d%sblock.fs_fpg)/sblock.fs_frag);
1167			acg.cg_cs.cs_nbfree--;
1168			sblock.fs_cstotal.cs_nbfree--;
1169			if (sblock.fs_contigsumsize > 0) {
1170				clrbit(cg_clustersfree(&acg),
1171				    (d % sblock.fs_fpg) / sblock.fs_frag);
1172			}
1173		}
1174
1175		/*
1176		 * Allocate all fragments used by the cylinder summary in the
1177		 * last block.
1178		 */
1179		if (d < sblock.fs_csaddr + (sblock.fs_cssize / sblock.fs_fsize)) {
1180			for (; d - sblock.fs_csaddr <
1181			    sblock.fs_cssize/sblock.fs_fsize;
1182			    d++) {
1183				clrbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1184				acg.cg_cs.cs_nffree--;
1185				sblock.fs_cstotal.cs_nffree--;
1186			}
1187			acg.cg_cs.cs_nbfree--;
1188			acg.cg_cs.cs_nffree += sblock.fs_frag;
1189			sblock.fs_cstotal.cs_nbfree--;
1190			sblock.fs_cstotal.cs_nffree += sblock.fs_frag;
1191			if (sblock.fs_contigsumsize > 0) {
1192				clrbit(cg_clustersfree(&acg),
1193				    (d%sblock.fs_fpg) / sblock.fs_frag);
1194			}
1195
1196			frag_adjust(d % sblock.fs_fpg, 1);
1197		}
1198		/*
1199		 * XXX	Handle the cluster statistics here in the case  this
1200		 *	cylinder group is now almost full, and the remaining
1201		 *	space is less then the maximum cluster size. This is
1202		 *	probably not needed, as you would hardly find a file
1203		 *	system which has only MAXCSBUFS+FS_MAXCONTIG of free
1204		 *	space right behind the cylinder group information in
1205		 *	any new cylinder group.
1206		 */
1207
1208		/*
1209		 * Update our statistics in the cylinder summary.
1210		 */
1211		*cs = acg.cg_cs;
1212
1213		/*
1214		 * Write the new cylinder group containing the cylinder summary
1215		 * back to disk.
1216		 */
1217		wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1218		    (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1219		DBG_PRINT0("nscg written\n");
1220		DBG_DUMP_CG(&sblock, "new summary cg", &acg);
1221
1222		DBG_LEAVE;
1223		return;
1224	}
1225	/*
1226	 * We have got enough of space in the current cylinder group, so we
1227	 * can relocate just a few blocks, and let the summary  information
1228	 * grow in place where it is right now.
1229	 */
1230	DBG_TRC;
1231
1232	cbase = cgbase(&osblock, ocscg);	/* old and new are equal */
1233	dupper = sblock.fs_csaddr - cbase +
1234	    howmany(sblock.fs_cssize, sblock.fs_fsize);
1235	odupper = osblock.fs_csaddr - cbase +
1236	    howmany(osblock.fs_cssize, osblock.fs_fsize);
1237
1238	sblock.fs_dsize -= dupper-odupper;
1239
1240	/*
1241	 * Allocate the space for the array of blocks to be relocated.
1242	 */
1243 	bp = (struct gfs_bpp *)calloc(((dupper-odupper) / sblock.fs_frag + 2),
1244	    sizeof(struct gfs_bpp));
1245	if (bp == NULL)
1246		errx(1, "calloc failed");
1247
1248	/*
1249	 * Lock all new frags needed for the cylinder group summary. This  is
1250	 * done per fragment in the first and last block of the new  required
1251	 * area, and per block for all other blocks.
1252	 *
1253	 * Handle the first new  block here (but only if some fragments where
1254	 * already used for the cylinder summary).
1255	 */
1256	ind = 0;
1257	frag_adjust(odupper, -1);
1258	for (d = odupper; ((d < dupper) && (d % sblock.fs_frag)); d++) {
1259		DBG_PRINT1("scg first frag check loop d=%jd\n", (intmax_t)d);
1260		if (isclr(cg_blksfree(&acg), d)) {
1261			if (!ind) {
1262				bp[ind].old = d / sblock.fs_frag;
1263				bp[ind].flags|=GFS_FL_FIRST;
1264				if (roundup(d, sblock.fs_frag) >= dupper)
1265					bp[ind].flags |= GFS_FL_LAST;
1266				ind++;
1267			}
1268		} else {
1269			clrbit(cg_blksfree(&acg), d);
1270			acg.cg_cs.cs_nffree--;
1271			sblock.fs_cstotal.cs_nffree--;
1272		}
1273		/*
1274		 * No cluster handling is needed here, as there was at least
1275		 * one  fragment in use by the cylinder summary in  the  old
1276		 * filesystem.
1277		 * No block - free counter handling here as this block was not
1278		 * a free block.
1279		 */
1280	}
1281	frag_adjust(odupper, 1);
1282
1283	/*
1284	 * Handle all needed complete blocks here.
1285	 */
1286	for (; d + sblock.fs_frag <= dupper; d += sblock.fs_frag) {
1287		DBG_PRINT1("scg block check loop d=%jd\n", (intmax_t)d);
1288		if (!isblock(&sblock, cg_blksfree(&acg), d / sblock.fs_frag)) {
1289			for (f = d; f < d + sblock.fs_frag; f++) {
1290				if (isset(cg_blksfree(&aocg), f)) {
1291					acg.cg_cs.cs_nffree--;
1292					sblock.fs_cstotal.cs_nffree--;
1293				}
1294			}
1295			clrblock(&sblock, cg_blksfree(&acg), d / sblock.fs_frag);
1296			bp[ind].old = d / sblock.fs_frag;
1297			ind++;
1298		} else {
1299			clrblock(&sblock, cg_blksfree(&acg), d / sblock.fs_frag);
1300			acg.cg_cs.cs_nbfree--;
1301			sblock.fs_cstotal.cs_nbfree--;
1302			if (sblock.fs_contigsumsize > 0) {
1303				clrbit(cg_clustersfree(&acg), d / sblock.fs_frag);
1304				for (lcs = 0, l = (d / sblock.fs_frag) + 1;
1305				    lcs < sblock.fs_contigsumsize;
1306				    l++, lcs++) {
1307					if (isclr(cg_clustersfree(&acg), l))
1308						break;
1309				}
1310				if (lcs < sblock.fs_contigsumsize) {
1311					cg_clustersum(&acg)[lcs + 1]--;
1312					if (lcs)
1313						cg_clustersum(&acg)[lcs]++;
1314				}
1315			}
1316		}
1317		/*
1318		 * No fragment counter handling is needed here, as this finally
1319		 * doesn't change after the relocation.
1320		 */
1321	}
1322
1323	/*
1324	 * Handle all fragments needed in the last new affected block.
1325	 */
1326	if (d < dupper) {
1327		frag_adjust(dupper - 1, -1);
1328
1329		if (isblock(&sblock, cg_blksfree(&acg), d / sblock.fs_frag)) {
1330			acg.cg_cs.cs_nbfree--;
1331			sblock.fs_cstotal.cs_nbfree--;
1332			acg.cg_cs.cs_nffree+=sblock.fs_frag;
1333			sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1334			if (sblock.fs_contigsumsize > 0) {
1335				clrbit(cg_clustersfree(&acg), d / sblock.fs_frag);
1336				for (lcs = 0, l = (d / sblock.fs_frag) + 1;
1337				    lcs < sblock.fs_contigsumsize;
1338				    l++, lcs++) {
1339					if (isclr(cg_clustersfree(&acg), l))
1340						break;
1341				}
1342				if (lcs < sblock.fs_contigsumsize) {
1343					cg_clustersum(&acg)[lcs + 1]--;
1344					if (lcs)
1345						cg_clustersum(&acg)[lcs]++;
1346				}
1347			}
1348		}
1349
1350		for (; d < dupper; d++) {
1351			DBG_PRINT1("scg second frag check loop d=%jd\n",
1352			    (intmax_t)d);
1353			if (isclr(cg_blksfree(&acg), d)) {
1354				bp[ind].old = d / sblock.fs_frag;
1355				bp[ind].flags |= GFS_FL_LAST;
1356			} else {
1357				clrbit(cg_blksfree(&acg), d);
1358				acg.cg_cs.cs_nffree--;
1359				sblock.fs_cstotal.cs_nffree--;
1360			}
1361		}
1362		if (bp[ind].flags & GFS_FL_LAST) /* we have to advance here */
1363			ind++;
1364		frag_adjust(dupper - 1, 1);
1365	}
1366
1367	/*
1368	 * If we found a block to relocate just do so.
1369	 */
1370	if (ind) {
1371		for (i = 0; i < ind; i++) {
1372			if (!bp[i].old) { /* no more blocks listed */
1373				/*
1374				 * XXX	A relative blocknumber should not be
1375				 *	zero,   which  is   not   explicitly
1376				 *	guaranteed by our code.
1377				 */
1378				break;
1379			}
1380			/*
1381			 * Allocate a complete block in the same (current)
1382			 * cylinder group.
1383			 */
1384			bp[i].new = alloc() / sblock.fs_frag;
1385
1386			/*
1387			 * There is no frag_adjust() needed for the new block
1388			 * as it will have no fragments yet :-).
1389			 */
1390			for (f = bp[i].old * sblock.fs_frag,
1391			    g = bp[i].new * sblock.fs_frag;
1392			    f < (bp[i].old + 1) * sblock.fs_frag;
1393			    f++, g++) {
1394				if (isset(cg_blksfree(&aocg), f)) {
1395					setbit(cg_blksfree(&acg), g);
1396					acg.cg_cs.cs_nffree++;
1397					sblock.fs_cstotal.cs_nffree++;
1398				}
1399			}
1400
1401			/*
1402			 * Special handling is required if this was the  first
1403			 * block. We have to consider the fragments which were
1404			 * used by the cylinder summary in the original  block
1405			 * which  re to be free in the copy of our  block.  We
1406			 * have  to be careful if this first block happens  to
1407			 * be also the last block to be relocated.
1408			 */
1409			if (bp[i].flags & GFS_FL_FIRST) {
1410				for (f = bp[i].old * sblock.fs_frag,
1411				    g = bp[i].new * sblock.fs_frag;
1412				    f < odupper;
1413				    f++, g++) {
1414					setbit(cg_blksfree(&acg), g);
1415					acg.cg_cs.cs_nffree++;
1416					sblock.fs_cstotal.cs_nffree++;
1417				}
1418				if (!(bp[i].flags & GFS_FL_LAST))
1419					frag_adjust(bp[i].new * sblock.fs_frag, 1);
1420			}
1421
1422			/*
1423			 * Special handling is required if this is the last
1424			 * block to be relocated.
1425			 */
1426			if (bp[i].flags & GFS_FL_LAST) {
1427				frag_adjust(bp[i].new * sblock.fs_frag, 1);
1428				frag_adjust(bp[i].old * sblock.fs_frag, -1);
1429				for (f = dupper;
1430				    f < roundup(dupper, sblock.fs_frag);
1431				    f++) {
1432					if (isclr(cg_blksfree(&acg), f)) {
1433						setbit(cg_blksfree(&acg), f);
1434						acg.cg_cs.cs_nffree++;
1435						sblock.fs_cstotal.cs_nffree++;
1436					}
1437				}
1438				frag_adjust(bp[i].old * sblock.fs_frag, 1);
1439			}
1440
1441			/*
1442			 * !!! Attach the cylindergroup offset here.
1443			 */
1444			bp[i].old += cbase / sblock.fs_frag;
1445			bp[i].new += cbase / sblock.fs_frag;
1446
1447			/*
1448			 * Copy the content of the block.
1449			 */
1450			/*
1451			 * XXX	Here we will have to implement a copy on write
1452			 *	in the case we have any active snapshots.
1453			 */
1454			rdfs(fsbtodb(&sblock, bp[i].old * sblock.fs_frag),
1455			    (size_t)sblock.fs_bsize, (void *)&ablk, fsi);
1456			wtfs(fsbtodb(&sblock, bp[i].new * sblock.fs_frag),
1457			    (size_t)sblock.fs_bsize, (void *)&ablk, fso, Nflag);
1458			DBG_DUMP_HEX(&sblock, "copied full block",
1459			    (unsigned char *)&ablk);
1460
1461			DBG_PRINT2("scg (%jd->%jd) block relocated\n",
1462			    (intmax_t)bp[i].old, (intmax_t)bp[i].new);
1463		}
1464
1465		/*
1466		 * Now we have to update all references to any fragment which
1467		 * belongs  to any block relocated. We iterate now  over  all
1468		 * cylinder  groups,  within those over all non  zero  length
1469		 * inodes.
1470		 */
1471		for (cylno = 0; cylno < osblock.fs_ncg; cylno++) {
1472			DBG_PRINT1("scg doing cg (%d)\n", cylno);
1473			for (inc = osblock.fs_ipg - 1; inc > 0; inc--) {
1474				updrefs(cylno, (ino_t)inc, bp, fsi, fso, Nflag);
1475			}
1476		}
1477
1478		/*
1479		 * All inodes are checked, now make sure the number of
1480		 * references found make sense.
1481		 */
1482		for (i = 0; i < ind; i++) {
1483			if (!bp[i].found || (bp[i].found > sblock.fs_frag)) {
1484				warnx("error: %jd refs found for block %jd.",
1485				    (intmax_t)bp[i].found, (intmax_t)bp[i].old);
1486			}
1487
1488		}
1489	}
1490	/*
1491	 * The following statistics are not changed here:
1492	 *     sblock.fs_cstotal.cs_ndir
1493	 *     sblock.fs_cstotal.cs_nifree
1494	 * The following statistics were already updated on the fly:
1495	 *     sblock.fs_cstotal.cs_nffree
1496	 *     sblock.fs_cstotal.cs_nbfree
1497	 * As the statistics for this cylinder group are ready, copy it to
1498	 * the summary information array.
1499	 */
1500
1501	*cs = acg.cg_cs;
1502
1503	/*
1504	 * Write summary cylinder group back to disk.
1505	 */
1506	wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)), (size_t)sblock.fs_cgsize,
1507	    (void *)&acg, fso, Nflag);
1508	DBG_PRINT0("scg written\n");
1509	DBG_DUMP_CG(&sblock, "new summary cg", &acg);
1510
1511	DBG_LEAVE;
1512}
1513
1514/* ************************************************************** rdfs ***** */
1515/*
1516 * Here we read some block(s) from disk.
1517 */
1518static void
1519rdfs(daddr_t bno, size_t size, void *bf, int fsi)
1520{
1521	DBG_FUNC("rdfs")
1522	ssize_t	n;
1523
1524	DBG_ENTER;
1525
1526	if (bno < 0) {
1527		err(32, "rdfs: attempting to read negative block number");
1528	}
1529	if (lseek(fsi, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0) {
1530		err(33, "rdfs: seek error: %jd", (intmax_t)bno);
1531	}
1532	n = read(fsi, bf, size);
1533	if (n != (ssize_t)size) {
1534		err(34, "rdfs: read error: %jd", (intmax_t)bno);
1535	}
1536
1537	DBG_LEAVE;
1538}
1539
1540/* ************************************************************** wtfs ***** */
1541/*
1542 * Here we write some block(s) to disk.
1543 */
1544static void
1545wtfs(daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag)
1546{
1547	DBG_FUNC("wtfs")
1548	ssize_t	n;
1549
1550	DBG_ENTER;
1551
1552	if (Nflag) {
1553		DBG_LEAVE;
1554		return;
1555	}
1556	if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0) {
1557		err(35, "wtfs: seek error: %ld", (long)bno);
1558	}
1559	n = write(fso, bf, size);
1560	if (n != (ssize_t)size) {
1561		err(36, "wtfs: write error: %ld", (long)bno);
1562	}
1563
1564	DBG_LEAVE;
1565}
1566
1567/* ************************************************************* alloc ***** */
1568/*
1569 * Here we allocate a free block in the current cylinder group. It is assumed,
1570 * that  acg contains the current cylinder group. As we may take a block  from
1571 * somewhere in the filesystem we have to handle cluster summary here.
1572 */
1573static daddr_t
1574alloc(void)
1575{
1576	DBG_FUNC("alloc")
1577	daddr_t	d, blkno;
1578	int	lcs1, lcs2;
1579	int	l;
1580	int	csmin, csmax;
1581	int	dlower, dupper, dmax;
1582
1583	DBG_ENTER;
1584
1585	if (acg.cg_magic != CG_MAGIC) {
1586		warnx("acg: bad magic number");
1587		DBG_LEAVE;
1588		return (0);
1589	}
1590	if (acg.cg_cs.cs_nbfree == 0) {
1591		warnx("error: cylinder group ran out of space");
1592		DBG_LEAVE;
1593		return (0);
1594	}
1595	/*
1596	 * We start seeking for free blocks only from the space available after
1597	 * the  end of the new grown cylinder summary. Otherwise we allocate  a
1598	 * block here which we have to relocate a couple of seconds later again
1599	 * again, and we are not prepared to to this anyway.
1600	 */
1601	blkno = -1;
1602	dlower = cgsblock(&sblock, acg.cg_cgx) - cgbase(&sblock, acg.cg_cgx);
1603	dupper = cgdmin(&sblock, acg.cg_cgx) - cgbase(&sblock, acg.cg_cgx);
1604	dmax = cgbase(&sblock, acg.cg_cgx) + sblock.fs_fpg;
1605	if (dmax > sblock.fs_size) {
1606		dmax = sblock.fs_size;
1607	}
1608	dmax -= cgbase(&sblock, acg.cg_cgx); /* retransform into cg */
1609	csmin=sblock.fs_csaddr-cgbase(&sblock, acg.cg_cgx);
1610	csmax = csmin + howmany(sblock.fs_cssize, sblock.fs_fsize);
1611	DBG_PRINT3("seek range: dl=%d, du=%d, dm=%d\n", dlower, dupper, dmax);
1612	DBG_PRINT2("range cont: csmin=%d, csmax=%d\n", csmin, csmax);
1613
1614	for (d = 0; (d < dlower && blkno == -1); d += sblock.fs_frag) {
1615		if (d >= csmin && d <= csmax) {
1616			continue;
1617		}
1618		if (isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1619		    d))) {
1620			blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1621			break;
1622		}
1623	}
1624	for (d = dupper; (d < dmax && blkno == -1); d += sblock.fs_frag) {
1625		if (d >= csmin && d <= csmax) {
1626			continue;
1627		}
1628		if (isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1629		    d))) {
1630			blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1631			break;
1632		}
1633	}
1634	if (blkno == -1) {
1635		warnx("internal error: couldn't find promised block in cg");
1636		DBG_LEAVE;
1637		return (0);
1638	}
1639
1640	/*
1641	 * This is needed if the block was found already in the first loop.
1642	 */
1643	d = blkstofrags(&sblock, blkno);
1644
1645	clrblock(&sblock, cg_blksfree(&acg), blkno);
1646	if (sblock.fs_contigsumsize > 0) {
1647		/*
1648		 * Handle the cluster allocation bitmap.
1649		 */
1650		clrbit(cg_clustersfree(&acg), blkno);
1651		/*
1652		 * We  possibly have split a cluster here, so we have  to  do
1653		 * recalculate the sizes of the remaining cluster halves now,
1654		 * and use them for updating the cluster summary information.
1655		 *
1656		 * Lets start with the blocks before our allocated block ...
1657		 */
1658		for (lcs1 = 0, l = blkno - 1; lcs1 < sblock.fs_contigsumsize;
1659		    l--, lcs1++) {
1660			if (isclr(cg_clustersfree(&acg), l))
1661				break;
1662		}
1663		/*
1664		 * ... and continue with the blocks right after our allocated
1665		 * block.
1666		 */
1667		for (lcs2 = 0, l = blkno + 1; lcs2 < sblock.fs_contigsumsize;
1668		    l++, lcs2++) {
1669			if (isclr(cg_clustersfree(&acg), l))
1670				break;
1671		}
1672
1673		/*
1674		 * Now update all counters.
1675		 */
1676		cg_clustersum(&acg)[MINIMUM(lcs1 + lcs2 + 1, sblock.fs_contigsumsize)]--;
1677		if (lcs1)
1678			cg_clustersum(&acg)[lcs1]++;
1679		if (lcs2)
1680			cg_clustersum(&acg)[lcs2]++;
1681	}
1682	/*
1683	 * Update all statistics based on blocks.
1684	 */
1685	acg.cg_cs.cs_nbfree--;
1686	sblock.fs_cstotal.cs_nbfree--;
1687
1688	DBG_LEAVE;
1689	return (d);
1690}
1691
1692/* *********************************************************** isblock ***** */
1693/*
1694 * Here  we check if all frags of a block are free. For more details  again
1695 * please see the source of newfs(8), as this function is taken over almost
1696 * unchanged.
1697 */
1698static int
1699isblock(struct fs *fs, unsigned char *cp, int h)
1700{
1701	DBG_FUNC("isblock")
1702	unsigned char	mask;
1703
1704	DBG_ENTER;
1705
1706	switch (fs->fs_frag) {
1707	case 8:
1708		DBG_LEAVE;
1709		return (cp[h] == 0xff);
1710	case 4:
1711		mask = 0x0f << ((h & 0x1) << 2);
1712		DBG_LEAVE;
1713		return ((cp[h >> 1] & mask) == mask);
1714	case 2:
1715		mask = 0x03 << ((h & 0x3) << 1);
1716		DBG_LEAVE;
1717		return ((cp[h >> 2] & mask) == mask);
1718	case 1:
1719		mask = 0x01 << (h & 0x7);
1720		DBG_LEAVE;
1721		return ((cp[h >> 3] & mask) == mask);
1722	default:
1723		fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
1724		DBG_LEAVE;
1725		return (0);
1726	}
1727}
1728
1729/* ********************************************************** clrblock ***** */
1730/*
1731 * Here we allocate a complete block in the block map. For more details again
1732 * please  see the source of newfs(8), as this function is taken over  almost
1733 * unchanged.
1734 */
1735static void
1736clrblock(struct fs *fs, unsigned char *cp, int h)
1737{
1738	DBG_FUNC("clrblock")
1739
1740	DBG_ENTER;
1741
1742	switch ((fs)->fs_frag) {
1743	case 8:
1744		cp[h] = 0;
1745		break;
1746	case 4:
1747		cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1748		break;
1749	case 2:
1750		cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1751		break;
1752	case 1:
1753		cp[h >> 3] &= ~(0x01 << (h & 0x7));
1754		break;
1755	default:
1756		warnx("clrblock bad fs_frag %d", fs->fs_frag);
1757		break;
1758	}
1759
1760	DBG_LEAVE;
1761}
1762
1763/* ********************************************************** setblock ***** */
1764/*
1765 * Here we free a complete block in the free block map. For more details again
1766 * please  see the source of newfs(8), as this function is taken  over  almost
1767 * unchanged.
1768 */
1769static void
1770setblock(struct fs *fs, unsigned char *cp, int h)
1771{
1772	DBG_FUNC("setblock")
1773
1774	DBG_ENTER;
1775
1776	switch (fs->fs_frag) {
1777	case 8:
1778		cp[h] = 0xff;
1779		break;
1780	case 4:
1781		cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1782		break;
1783	case 2:
1784		cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1785		break;
1786	case 1:
1787		cp[h >> 3] |= (0x01 << (h & 0x7));
1788		break;
1789	default:
1790		warnx("setblock bad fs_frag %d", fs->fs_frag);
1791		break;
1792	}
1793
1794	DBG_LEAVE;
1795}
1796
1797/* ************************************************************ ginode ***** */
1798/*
1799 * This function provides access to an individual inode. We find out in which
1800 * block  the  requested inode is located, read it from disk if  needed,  and
1801 * return  the pointer into that block. We maintain a cache of one  block  to
1802 * not  read the same block again and again if we iterate linearly  over  all
1803 * inodes.
1804 */
1805static union dinode *
1806ginode(ino_t inumber, int fsi, int cg)
1807{
1808	DBG_FUNC("ginode")
1809	static ino_t	startinum = 0;	/* first inode in cached block */
1810
1811	DBG_ENTER;
1812
1813	/*
1814	 * The inumber passed in is relative to the cg, so use it here to see
1815	 * if the inode has been allocated yet.
1816	 */
1817	if (isclr(cg_inosused(&aocg), inumber)) {
1818		DBG_LEAVE;
1819		return NULL;
1820	}
1821	/*
1822	 * Now make the inumber relative to the entire inode space so it can
1823	 * be sanity checked.
1824	 */
1825	inumber += (cg * sblock.fs_ipg);
1826	if (inumber < ROOTINO) {
1827		DBG_LEAVE;
1828		return NULL;
1829	}
1830	if (inumber > maxino)
1831		errx(8, "bad inode number %llu to ginode",
1832		    (unsigned long long)inumber);
1833	if (startinum == 0 ||
1834	    inumber < startinum || inumber >= startinum + INOPB(&sblock)) {
1835		inoblk = fsbtodb(&sblock, ino_to_fsba(&sblock, inumber));
1836		rdfs(inoblk, (size_t)sblock.fs_bsize, inobuf, fsi);
1837		startinum = (inumber / INOPB(&sblock)) * INOPB(&sblock);
1838	}
1839	DBG_LEAVE;
1840	if (sblock.fs_magic == FS_UFS1_MAGIC)
1841		return (union dinode *)((uintptr_t)inobuf +
1842		    (inumber % INOPB(&sblock)) * sizeof(struct ufs1_dinode));
1843	return (union dinode *)((uintptr_t)inobuf +
1844	    (inumber % INOPB(&sblock)) * sizeof(struct ufs2_dinode));
1845}
1846
1847/* ****************************************************** charsperline ***** */
1848/*
1849 * Figure out how many lines our current terminal has. For more details again
1850 * please  see the source of newfs(8), as this function is taken over  almost
1851 * unchanged.
1852 */
1853static int
1854charsperline(void)
1855{
1856	DBG_FUNC("charsperline")
1857	int	columns;
1858	char	*cp;
1859	struct winsize	ws;
1860
1861	DBG_ENTER;
1862
1863	columns = 0;
1864	if (ioctl(0, TIOCGWINSZ, &ws) != -1) {
1865		columns = ws.ws_col;
1866	}
1867	if (columns == 0 && (cp = getenv("COLUMNS"))) {
1868		columns = atoi(cp);
1869	}
1870	if (columns == 0) {
1871		columns = 80;	/* last resort */
1872	}
1873
1874	DBG_LEAVE;
1875	return columns;
1876}
1877
1878/* ************************************************************** main ***** */
1879/*
1880 * growfs(8)  is a utility which allows to increase the size of  an  existing
1881 * ufs filesystem. Currently this can only be done on unmounted file  system.
1882 * It  recognizes some command line options to specify the new desired  size,
1883 * and  it does some basic checkings. The old filesystem size is  determined
1884 * and  after some more checks like we can really access the new  last  block
1885 * on the disk etc. we calculate the new parameters for the superblock. After
1886 * having  done  this we just call growfs() which will do  the  work.  Before
1887 * we finish the only thing left is to update the disklabel.
1888 * We still have to provide support for snapshots. Therefore we first have to
1889 * understand  what data structures are always replicated in the snapshot  on
1890 * creation,  for all other blocks we touch during our procedure, we have  to
1891 * keep the old blocks unchanged somewhere available for the snapshots. If we
1892 * are lucky, then we only have to handle our blocks to be relocated in  that
1893 * way.
1894 * Also  we  have to consider in what order we actually update  the  critical
1895 * data structures of the filesystem to make sure, that in case of a disaster
1896 * fsck(8) is still able to restore any lost data.
1897 * The  foreseen last step then will be to provide for growing  even  mounted
1898 * file  systems. There we have to extend the mount() system call to  provide
1899 * userland access to the filesystem locking facility.
1900 */
1901int
1902main(int argc, char **argv)
1903{
1904	DBG_FUNC("main")
1905	char	*device, *lastsector;
1906	int	ch;
1907	long long	size = 0;
1908	unsigned int	Nflag = 0;
1909	int	ExpertFlag = 0;
1910	struct stat	st;
1911	struct disklabel	*lp;
1912	struct partition	*pp;
1913	int	i,fsi,fso;
1914	char	reply[5];
1915	const char *errstr;
1916#ifdef FSMAXSNAP
1917	int	j;
1918#endif /* FSMAXSNAP */
1919
1920	DBG_ENTER;
1921
1922	while ((ch = getopt(argc, argv, "Nqs:vy")) != -1) {
1923		switch (ch) {
1924		case 'N':
1925			Nflag = 1;
1926			break;
1927		case 'q':
1928			quiet = 1;
1929			break;
1930		case 's':
1931			size = strtonum(optarg, 1, LLONG_MAX, &errstr);
1932			if (errstr)
1933				usage();
1934			break;
1935		case 'v': /* for compatibility to newfs */
1936			break;
1937		case 'y':
1938			ExpertFlag = 1;
1939			break;
1940		case '?':
1941			/* FALLTHROUGH */
1942		default:
1943			usage();
1944		}
1945	}
1946	argc -= optind;
1947	argv += optind;
1948
1949	if (argc != 1)
1950		usage();
1951
1952	/*
1953	 * Rather than guessing, use opendev() to get the device
1954	 * name, which we open for reading.
1955	 */
1956	if ((fsi = opendev(*argv, O_RDONLY, 0, &device)) < 0)
1957		err(1, "%s", *argv);
1958
1959	/*
1960	 * Try to access our devices for writing ...
1961	 */
1962	if (Nflag) {
1963		fso = -1;
1964	} else {
1965		fso = open(device, O_WRONLY);
1966		if (fso < 0)
1967			err(1, "%s", device);
1968	}
1969
1970	/*
1971	 * Now we have a file descriptor for our device, fstat() it to
1972	 * figure out the partition number.
1973	 */
1974	if (fstat(fsi, &st) != 0)
1975		err(1, "%s: fstat()", device);
1976
1977	/*
1978	 * Try to read a label from the disk.  Then get the partition from the
1979	 * device minor number, using DISKPART().  Probably don't need to
1980	 * check against getmaxpartitions().
1981	 */
1982	lp = get_disklabel(fsi);
1983	if (DISKPART(st.st_rdev) < getmaxpartitions())
1984		pp = &lp->d_partitions[DISKPART(st.st_rdev)];
1985	else
1986		errx(1, "%s: invalid partition number %u",
1987		    device, DISKPART(st.st_rdev));
1988
1989	/*
1990	 * Check if that partition is suitable for growing a file system.
1991	 */
1992	if (DL_GETPSIZE(pp) < 1)
1993		errx(1, "partition is unavailable");
1994	if (pp->p_fstype != FS_BSDFFS)
1995		errx(1, "can only grow ffs partitions");
1996
1997	/*
1998	 * Read the current superblock, and take a backup.
1999	 */
2000	for (i = 0; sblock_try[i] != -1; i++) {
2001		sblockloc = sblock_try[i] / DEV_BSIZE;
2002		rdfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&(osblock), fsi);
2003		if ((osblock.fs_magic == FS_UFS1_MAGIC ||
2004		     (osblock.fs_magic == FS_UFS2_MAGIC &&
2005		      osblock.fs_sblockloc == sblock_try[i])) &&
2006		    osblock.fs_bsize <= MAXBSIZE &&
2007		    osblock.fs_bsize >= (int32_t) sizeof(struct fs))
2008			break;
2009	}
2010	if (sblock_try[i] == -1)
2011		errx(1, "superblock not recognized");
2012	if (osblock.fs_clean == 0)
2013		errx(1, "filesystem not clean - run fsck");
2014	if (sblock.fs_magic == FS_UFS1_MAGIC &&
2015	    (sblock.fs_ffs1_flags & FS_FLAGS_UPDATED) == 0)
2016		ffs1_sb_update(&sblock, sblock_try[i]);
2017	memcpy(&fsun1, &fsun2, sizeof(fsun2));
2018	maxino = sblock.fs_ncg * sblock.fs_ipg;
2019
2020	DBG_OPEN("/tmp/growfs.debug"); /* already here we need a superblock */
2021	DBG_DUMP_FS(&sblock, "old sblock");
2022
2023	/*
2024	 * Determine size to grow to. Default to the full size specified in
2025	 * the disk label.
2026	 */
2027	sblock.fs_size = dbtofsb(&osblock, DL_SECTOBLK(lp, DL_GETPSIZE(pp)));
2028	if (size != 0) {
2029		if (size > DL_GETPSIZE(pp)) {
2030			errx(1, "there is not enough space (%llu < %lld)",
2031			    DL_GETPSIZE(pp), size);
2032		}
2033		sblock.fs_size = dbtofsb(&osblock, DL_SECTOBLK(lp, size));
2034	}
2035
2036	/*
2037	 * Are we really growing ?
2038	 */
2039	if (osblock.fs_size >= sblock.fs_size) {
2040		errx(1, "we are not growing (%jd->%jd)",
2041		    (intmax_t)osblock.fs_size, (intmax_t)sblock.fs_size);
2042	}
2043
2044
2045#ifdef FSMAXSNAP
2046	/*
2047	 * Check if we find an active snapshot.
2048	 */
2049	if (ExpertFlag == 0) {
2050		for (j = 0; j < FSMAXSNAP; j++) {
2051			if (sblock.fs_snapinum[j]) {
2052				errx(1, "active snapshot found in filesystem\n"
2053				    "	please remove all snapshots before "
2054				    "using growfs");
2055			}
2056			if (!sblock.fs_snapinum[j])	/* list is dense */
2057				break;
2058		}
2059	}
2060#endif
2061
2062	if (ExpertFlag == 0 && Nflag == 0) {
2063		printf("We strongly recommend you to make a backup "
2064		    "before growing the Filesystem\n\n"
2065		    " Did you backup your data (Yes/No) ? ");
2066		if (fgets(reply, (int)sizeof(reply), stdin) == NULL ||
2067		    strncasecmp(reply, "Yes", 3)) {
2068			printf("\n Nothing done \n");
2069			exit (0);
2070		}
2071	}
2072
2073	if (!quiet)
2074		printf("new filesystem size is: %jd frags\n",
2075		    (intmax_t)sblock.fs_size);
2076
2077	/*
2078	 * Try to access our new last sector in the filesystem. Even if we
2079	 * later on realize we have to abort our operation, on that sector
2080	 * there should be no data, so we can't destroy something yet.
2081	 */
2082	lastsector = calloc(1, lp->d_secsize);
2083	if (!lastsector)
2084		err(1, "No memory for last sector test write");
2085	wtfs(DL_SECTOBLK(lp, DL_GETPSIZE(pp) - 1), lp->d_secsize,
2086	    lastsector, fso, Nflag);
2087	free(lastsector);
2088
2089	/*
2090	 * Now calculate new superblock values and check for reasonable
2091	 * bound for new filesystem size:
2092	 *     fs_size:    is derived from label or user input
2093	 *     fs_dsize:   should get updated in the routines creating or
2094	 *                 updating the cylinder groups on the fly
2095	 *     fs_cstotal: should get updated in the routines creating or
2096	 *                 updating the cylinder groups
2097	 */
2098
2099	/*
2100	 * Update the number of cylinders and cylinder groups in the file system.
2101	 */
2102	if (sblock.fs_magic == FS_UFS1_MAGIC) {
2103		sblock.fs_ncyl = sblock.fs_size * NSPF(&sblock) / sblock.fs_spc;
2104		if (sblock.fs_size * NSPF(&sblock) >
2105		    sblock.fs_ncyl * sblock.fs_spc)
2106		sblock.fs_ncyl++;
2107	}
2108	sblock.fs_ncg = howmany(sblock.fs_size, sblock.fs_fpg);
2109	maxino = sblock.fs_ncg * sblock.fs_ipg;
2110
2111	if (sblock.fs_size % sblock.fs_fpg != 0 &&
2112	    sblock.fs_size % sblock.fs_fpg < cgdmin(&sblock, sblock.fs_ncg)) {
2113		/*
2114		 * The space in the new last cylinder group is too small,
2115		 * so revert back.
2116		 */
2117		sblock.fs_ncg--;
2118		if (sblock.fs_magic == FS_UFS1_MAGIC)
2119			sblock.fs_ncyl = sblock.fs_ncg * sblock.fs_cpg;
2120		if (!quiet)
2121			printf("Warning: %jd sector(s) cannot be allocated.\n",
2122			    (intmax_t)fsbtodb(&sblock,
2123			    sblock.fs_size % sblock.fs_fpg));
2124		sblock.fs_size = sblock.fs_ncg * sblock.fs_fpg;
2125	}
2126
2127	/*
2128	 * Update the space for the cylinder group summary information in the
2129	 * respective cylinder group data area.
2130	 */
2131	sblock.fs_cssize =
2132	    fragroundup(&sblock, sblock.fs_ncg * sizeof(struct csum));
2133
2134	if (osblock.fs_size >= sblock.fs_size)
2135		errx(1, "not enough new space");
2136
2137	DBG_PRINT0("sblock calculated\n");
2138
2139	/*
2140	 * Ok, everything prepared, so now let's do the tricks.
2141	 */
2142	growfs(fsi, fso, Nflag);
2143
2144	/*
2145	 * Update the disk label.
2146	 */
2147	pp->p_fragblock =
2148	    DISKLABELV1_FFS_FRAGBLOCK(sblock.fs_fsize, sblock.fs_frag);
2149        pp->p_cpg = sblock.fs_fpg;
2150
2151	return_disklabel(fso, lp, Nflag);
2152	DBG_PRINT0("label rewritten\n");
2153
2154	close(fsi);
2155	if (fso > -1)
2156		close(fso);
2157
2158	DBG_CLOSE;
2159
2160	DBG_LEAVE;
2161	return 0;
2162}
2163
2164/* ************************************************** return_disklabel ***** */
2165/*
2166 * Write the updated disklabel back to disk.
2167 */
2168static void
2169return_disklabel(int fd, struct disklabel *lp, unsigned int Nflag)
2170{
2171	DBG_FUNC("return_disklabel")
2172	u_short	sum;
2173	u_short	*ptr;
2174
2175	DBG_ENTER;
2176
2177	if (!lp) {
2178		DBG_LEAVE;
2179		return;
2180	}
2181	if (!Nflag) {
2182		lp->d_checksum = 0;
2183		sum = 0;
2184		ptr = (u_short *)lp;
2185
2186		/*
2187		 * recalculate checksum
2188		 */
2189		while (ptr < (u_short *)&lp->d_partitions[lp->d_npartitions])
2190			sum ^= *ptr++;
2191		lp->d_checksum = sum;
2192
2193		if (ioctl(fd, DIOCWDINFO, (char *)lp) < 0)
2194			errx(1, "DIOCWDINFO failed");
2195	}
2196	free(lp);
2197
2198	DBG_LEAVE;
2199	return ;
2200}
2201
2202/* ***************************************************** get_disklabel ***** */
2203/*
2204 * Read the disklabel from disk.
2205 */
2206static struct disklabel *
2207get_disklabel(int fd)
2208{
2209	DBG_FUNC("get_disklabel")
2210	static struct	disklabel *lab;
2211
2212	DBG_ENTER;
2213
2214	lab = (struct disklabel *)malloc(sizeof(struct disklabel));
2215	if (!lab)
2216		errx(1, "malloc failed");
2217	if (ioctl(fd, DIOCGDINFO, (char *)lab) != 0)
2218		err(1, "DIOCGDINFO");
2219
2220	DBG_LEAVE;
2221	return (lab);
2222}
2223
2224
2225/* ************************************************************* usage ***** */
2226/*
2227 * Dump a line of usage.
2228 */
2229static void
2230usage(void)
2231{
2232	DBG_FUNC("usage")
2233
2234	DBG_ENTER;
2235
2236	fprintf(stderr, "usage: growfs [-Nqy] [-s size] special\n");
2237
2238	DBG_LEAVE;
2239	exit(1);
2240}
2241
2242/* *********************************************************** updclst ***** */
2243/*
2244 * This updates most parameters and the bitmap related to cluster. We have to
2245 * assume that sblock, osblock, acg are set up.
2246 */
2247static void
2248updclst(int block)
2249{
2250	DBG_FUNC("updclst")
2251	static int	lcs = 0;
2252
2253	DBG_ENTER;
2254
2255	if (sblock.fs_contigsumsize < 1)	/* no clustering */
2256		return;
2257
2258	/*
2259	 * update cluster allocation map
2260	 */
2261	setbit(cg_clustersfree(&acg), block);
2262
2263	/*
2264	 * update cluster summary table
2265	 */
2266	if (!lcs) {
2267		/*
2268		 * calculate size for the trailing cluster
2269		 */
2270		for (block--; lcs < sblock.fs_contigsumsize; block--, lcs++) {
2271			if (isclr(cg_clustersfree(&acg), block))
2272				break;
2273		}
2274	}
2275	if (lcs < sblock.fs_contigsumsize) {
2276		if (lcs)
2277			cg_clustersum(&acg)[lcs]--;
2278		lcs++;
2279		cg_clustersum(&acg)[lcs]++;
2280	}
2281
2282	DBG_LEAVE;
2283}
2284
2285/* *********************************************************** updrefs ***** */
2286/*
2287 * This updates all references to relocated blocks for the given inode.  The
2288 * inode is given as number within the cylinder group, and the number of the
2289 * cylinder group.
2290 */
2291static void
2292updrefs(int cg, ino_t in, struct gfs_bpp *bp, int fsi, int fso, unsigned int
2293    Nflag)
2294{
2295	DBG_FUNC("updrefs")
2296	daddr_t	len, lbn, numblks;
2297	daddr_t	iptr, blksperindir;
2298	union dinode	*ino;
2299	int		i, mode, inodeupdated;
2300
2301	DBG_ENTER;
2302
2303	ino = ginode(in, fsi, cg);
2304	if (ino == NULL) {
2305		DBG_LEAVE;
2306		return;
2307	}
2308	mode = DIP(ino, di_mode) & IFMT;
2309	if (mode != IFDIR && mode != IFREG && mode != IFLNK) {
2310		DBG_LEAVE;
2311		return; /* only check DIR, FILE, LINK */
2312	}
2313	if (mode == IFLNK &&
2314	    DIP(ino, di_size) < (u_int64_t) sblock.fs_maxsymlinklen) {
2315		DBG_LEAVE;
2316		return;	/* skip short symlinks */
2317	}
2318	numblks = howmany(DIP(ino, di_size), sblock.fs_bsize);
2319	if (numblks == 0) {
2320		DBG_LEAVE;
2321		return;	/* skip empty file */
2322	}
2323	if (DIP(ino, di_blocks) == 0) {
2324		DBG_LEAVE;
2325		return;	/* skip empty swiss cheesy file or old fastlink */
2326	}
2327	DBG_PRINT2("scg checking inode (%llu in %d)\n",
2328	    (unsigned long long)in, cg);
2329
2330	/*
2331	 * Check all the blocks.
2332	 */
2333	inodeupdated = 0;
2334	len = numblks < NDADDR ? numblks : NDADDR;
2335	for (i = 0; i < len; i++) {
2336		iptr = DIP(ino, di_db[i]);
2337		if (iptr == 0)
2338			continue;
2339		if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2340			DIP_SET(ino, di_db[i], iptr);
2341			inodeupdated++;
2342		}
2343	}
2344	DBG_PRINT0("~~scg direct blocks checked\n");
2345
2346	blksperindir = 1;
2347	len = numblks - NDADDR;
2348	lbn = NDADDR;
2349	for (i = 0; len > 0 && i < NIADDR; i++) {
2350		iptr = DIP(ino, di_ib[i]);
2351		if (iptr == 0)
2352			continue;
2353		if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2354			DIP_SET(ino, di_ib[i], iptr);
2355			inodeupdated++;
2356		}
2357		indirchk(blksperindir, lbn, iptr, numblks, bp, fsi, fso, Nflag);
2358		blksperindir *= NINDIR(&sblock);
2359		lbn += blksperindir;
2360		len -= blksperindir;
2361		DBG_PRINT1("scg indirect_%d blocks checked\n", i + 1);
2362	}
2363	if (inodeupdated)
2364		wtfs(inoblk, sblock.fs_bsize, inobuf, fso, Nflag);
2365
2366	DBG_LEAVE;
2367}
2368
2369/*
2370 * Recursively check all the indirect blocks.
2371 */
2372static void
2373indirchk(daddr_t blksperindir, daddr_t lbn, daddr_t blkno,
2374    daddr_t lastlbn, struct gfs_bpp *bp, int fsi, int fso, unsigned int Nflag)
2375{
2376	DBG_FUNC("indirchk")
2377	void *ibuf;
2378	int i, last;
2379	daddr_t iptr;
2380
2381	DBG_ENTER;
2382
2383	/* read in the indirect block. */
2384	ibuf = malloc(sblock.fs_bsize);
2385	if (!ibuf)
2386		errx(1, "malloc failed");
2387	rdfs(fsbtodb(&sblock, blkno), (size_t)sblock.fs_bsize, ibuf, fsi);
2388	last = howmany(lastlbn - lbn, blksperindir) < NINDIR(&sblock) ?
2389	    howmany(lastlbn - lbn, blksperindir) : NINDIR(&sblock);
2390	for (i = 0; i < last; i++) {
2391		if (sblock.fs_magic == FS_UFS1_MAGIC)
2392			iptr = ((int32_t *)ibuf)[i];
2393		else
2394			iptr = ((daddr_t *)ibuf)[i];
2395		if (iptr == 0)
2396			continue;
2397		if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2398			if (sblock.fs_magic == FS_UFS1_MAGIC)
2399				((int32_t *)ibuf)[i] = iptr;
2400			else
2401				((daddr_t *)ibuf)[i] = iptr;
2402		}
2403		if (blksperindir == 1)
2404			continue;
2405		indirchk(blksperindir / NINDIR(&sblock), lbn + blksperindir * i,
2406		    iptr, lastlbn, bp, fsi, fso, Nflag);
2407	}
2408	free(ibuf);
2409
2410	DBG_LEAVE;
2411}
2412
2413static void
2414ffs1_sb_update(struct fs *fs, daddr_t sbloc)
2415{
2416	fs->fs_flags = fs->fs_ffs1_flags;
2417	fs->fs_sblockloc = sbloc;
2418	fs->fs_maxbsize = fs->fs_bsize;
2419	fs->fs_time = fs->fs_ffs1_time;
2420	fs->fs_size = fs->fs_ffs1_size;
2421	fs->fs_dsize = fs->fs_ffs1_dsize;
2422	fs->fs_csaddr = fs->fs_ffs1_csaddr;
2423	fs->fs_cstotal.cs_ndir = fs->fs_ffs1_cstotal.cs_ndir;
2424	fs->fs_cstotal.cs_nbfree = fs->fs_ffs1_cstotal.cs_nbfree;
2425	fs->fs_cstotal.cs_nifree = fs->fs_ffs1_cstotal.cs_nifree;
2426	fs->fs_cstotal.cs_nffree = fs->fs_ffs1_cstotal.cs_nffree;
2427	fs->fs_ffs1_flags |= FS_FLAGS_UPDATED;
2428}
2429