]> CyberLeo.Net >> Repos - FreeBSD/releng/7.2.git/blob - sbin/growfs/growfs.c
Create releng/7.2 from stable/7 in preparation for 7.2-RELEASE.
[FreeBSD/releng/7.2.git] / sbin / growfs / growfs.c
1 /*
2  * Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz
3  * Copyright (c) 1980, 1989, 1993 The Regents of the University of California.
4  * All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * Christoph Herrmann and Thomas-Henning von Kamptz, Munich and Frankfurt.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgment:
19  *      This product includes software developed by the University of
20  *      California, Berkeley and its contributors, as well as Christoph
21  *      Herrmann and Thomas-Henning von Kamptz.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  * $TSHeader: src/sbin/growfs/growfs.c,v 1.5 2000/12/12 19:31:00 tomsoft Exp $
39  *
40  */
41
42 #ifndef lint
43 static const char copyright[] =
44 "@(#) Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz\n\
45 Copyright (c) 1980, 1989, 1993 The Regents of the University of California.\n\
46 All rights reserved.\n";
47 #endif /* not lint */
48
49 #include <sys/cdefs.h>
50 __FBSDID("$FreeBSD$");
51
52 /* ********************************************************** INCLUDES ***** */
53 #include <sys/param.h>
54 #include <sys/disklabel.h>
55 #include <sys/ioctl.h>
56 #include <sys/stat.h>
57 #include <sys/disk.h>
58
59 #include <stdio.h>
60 #include <paths.h>
61 #include <ctype.h>
62 #include <err.h>
63 #include <fcntl.h>
64 #include <limits.h>
65 #include <stdlib.h>
66 #include <stdint.h>
67 #include <string.h>
68 #include <time.h>
69 #include <unistd.h>
70 #include <ufs/ufs/dinode.h>
71 #include <ufs/ffs/fs.h>
72
73 #include "debug.h"
74
75 /* *************************************************** GLOBALS & TYPES ***** */
76 #ifdef FS_DEBUG
77 int     _dbg_lvl_ = (DL_INFO);  /* DL_TRC */
78 #endif /* FS_DEBUG */
79
80 static union {
81         struct fs       fs;
82         char    pad[SBLOCKSIZE];
83 } fsun1, fsun2;
84 #define sblock  fsun1.fs        /* the new superblock */
85 #define osblock fsun2.fs        /* the old superblock */
86
87 /*
88  * Possible superblock locations ordered from most to least likely.
89  */
90 static int sblock_try[] = SBLOCKSEARCH;
91 static ufs2_daddr_t sblockloc;
92
93 static union {
94         struct cg       cg;
95         char    pad[MAXBSIZE];
96 } cgun1, cgun2;
97 #define acg     cgun1.cg        /* a cylinder cgroup (new) */
98 #define aocg    cgun2.cg        /* an old cylinder group */
99
100 static char     ablk[MAXBSIZE]; /* a block */
101
102 static struct csum      *fscs;  /* cylinder summary */
103
104 union dinode {
105         struct ufs1_dinode dp1;
106         struct ufs2_dinode dp2;
107 };
108 #define DIP(dp, field) \
109         ((sblock.fs_magic == FS_UFS1_MAGIC) ? \
110         (uint32_t)(dp)->dp1.field : (dp)->dp2.field)
111 #define DIP_SET(dp, field, val) do { \
112         if (sblock.fs_magic == FS_UFS1_MAGIC) \
113                 (dp)->dp1.field = (val); \
114         else \
115                 (dp)->dp2.field = (val); \
116         } while (0)
117 static ufs2_daddr_t     inoblk;                 /* inode block address */
118 static char             inobuf[MAXBSIZE];       /* inode block */
119 ino_t                   maxino;                 /* last valid inode */
120 static int              unlabeled;     /* unlabeled partition, e.g. vinum volume etc. */
121
122 /*
123  * An array of elements of type struct gfs_bpp describes all blocks to
124  * be relocated in order to free the space needed for the cylinder group
125  * summary for all cylinder groups located in the first cylinder group.
126  */
127 struct gfs_bpp {
128         ufs2_daddr_t    old;            /* old block number */
129         ufs2_daddr_t    new;            /* new block number */
130 #define GFS_FL_FIRST    1
131 #define GFS_FL_LAST     2
132         unsigned int    flags;  /* special handling required */
133         int     found;          /* how many references were updated */
134 };
135
136 /* ******************************************************** PROTOTYPES ***** */
137 static void     growfs(int, int, unsigned int);
138 static void     rdfs(ufs2_daddr_t, size_t, void *, int);
139 static void     wtfs(ufs2_daddr_t, size_t, void *, int, unsigned int);
140 static ufs2_daddr_t alloc(void);
141 static int      charsperline(void);
142 static void     usage(void);
143 static int      isblock(struct fs *, unsigned char *, int);
144 static void     clrblock(struct fs *, unsigned char *, int);
145 static void     setblock(struct fs *, unsigned char *, int);
146 static void     initcg(int, time_t, int, unsigned int);
147 static void     updjcg(int, time_t, int, int, unsigned int);
148 static void     updcsloc(time_t, int, int, unsigned int);
149 static struct disklabel *get_disklabel(int);
150 static void     return_disklabel(int, struct disklabel *, unsigned int);
151 static union dinode *ginode(ino_t, int, int);
152 static void     frag_adjust(ufs2_daddr_t, int);
153 static int      cond_bl_upd(ufs2_daddr_t *, struct gfs_bpp *, int, int,
154                     unsigned int);
155 static void     updclst(int);
156 static void     updrefs(int, ino_t, struct gfs_bpp *, int, int, unsigned int);
157 static void     indirchk(ufs_lbn_t, ufs_lbn_t, ufs2_daddr_t, ufs_lbn_t,
158                     struct gfs_bpp *, int, int, unsigned int);
159 static void     get_dev_size(int, int *);
160
161 /* ************************************************************ growfs ***** */
162 /*
163  * Here we actually start growing the file system. We basically read the
164  * cylinder summary from the first cylinder group as we want to update
165  * this on the fly during our various operations. First we handle the
166  * changes in the former last cylinder group. Afterwards we create all new
167  * cylinder groups.  Now we handle the cylinder group containing the
168  * cylinder summary which might result in a relocation of the whole
169  * structure.  In the end we write back the updated cylinder summary, the
170  * new superblock, and slightly patched versions of the super block
171  * copies.
172  */
173 static void
174 growfs(int fsi, int fso, unsigned int Nflag)
175 {
176         DBG_FUNC("growfs")
177         int     i;
178         int     cylno, j;
179         time_t  utime;
180         int     width;
181         char    tmpbuf[100];
182 #ifdef FSIRAND
183         static int      randinit=0;
184
185         DBG_ENTER;
186
187         if (!randinit) {
188                 randinit = 1;
189                 srandomdev();
190         }
191 #else /* not FSIRAND */
192
193         DBG_ENTER;
194
195 #endif /* FSIRAND */
196         time(&utime);
197
198         /*
199          * Get the cylinder summary into the memory.
200          */
201         fscs = (struct csum *)calloc((size_t)1, (size_t)sblock.fs_cssize);
202         if(fscs == NULL) {
203                 errx(1, "calloc failed");
204         }
205         for (i = 0; i < osblock.fs_cssize; i += osblock.fs_bsize) {
206                 rdfs(fsbtodb(&osblock, osblock.fs_csaddr +
207                     numfrags(&osblock, i)), (size_t)MIN(osblock.fs_cssize - i,
208                     osblock.fs_bsize), (void *)(((char *)fscs)+i), fsi);
209         }
210
211 #ifdef FS_DEBUG
212 {
213         struct csum     *dbg_csp;
214         int     dbg_csc;
215         char    dbg_line[80];
216
217         dbg_csp=fscs;
218         for(dbg_csc=0; dbg_csc<osblock.fs_ncg; dbg_csc++) {
219                 snprintf(dbg_line, sizeof(dbg_line),
220                     "%d. old csum in old location", dbg_csc);
221                 DBG_DUMP_CSUM(&osblock,
222                     dbg_line,
223                     dbg_csp++);
224         }
225 }
226 #endif /* FS_DEBUG */
227         DBG_PRINT0("fscs read\n");
228
229         /*
230          * Do all needed changes in the former last cylinder group.
231          */
232         updjcg(osblock.fs_ncg-1, utime, fsi, fso, Nflag);
233
234         /*
235          * Dump out summary information about file system.
236          */
237 #       define B2MBFACTOR (1 / (1024.0 * 1024.0))
238         printf("growfs: %.1fMB (%jd sectors) block size %d, fragment size %d\n",
239             (float)sblock.fs_size * sblock.fs_fsize * B2MBFACTOR,
240             (intmax_t)fsbtodb(&sblock, sblock.fs_size), sblock.fs_bsize,
241             sblock.fs_fsize);
242         printf("\tusing %d cylinder groups of %.2fMB, %d blks, %d inodes.\n",
243             sblock.fs_ncg, (float)sblock.fs_fpg * sblock.fs_fsize * B2MBFACTOR,
244             sblock.fs_fpg / sblock.fs_frag, sblock.fs_ipg);
245         if (sblock.fs_flags & FS_DOSOFTDEP)
246                 printf("\twith soft updates\n");
247 #       undef B2MBFACTOR
248
249         /*
250          * Now build the cylinders group blocks and
251          * then print out indices of cylinder groups.
252          */
253         printf("super-block backups (for fsck -b #) at:\n");
254         i = 0;
255         width = charsperline();
256
257         /*
258          * Iterate for only the new cylinder groups.
259          */
260         for (cylno = osblock.fs_ncg; cylno < sblock.fs_ncg; cylno++) {
261                 initcg(cylno, utime, fso, Nflag);
262                 j = sprintf(tmpbuf, " %jd%s",
263                     (intmax_t)fsbtodb(&sblock, cgsblock(&sblock, cylno)),
264                     cylno < (sblock.fs_ncg-1) ? "," : "" );
265                 if (i + j >= width) {
266                         printf("\n");
267                         i = 0;
268                 }
269                 i += j;
270                 printf("%s", tmpbuf);
271                 fflush(stdout);
272         }
273         printf("\n");
274
275         /*
276          * Do all needed changes in the first cylinder group.
277          * allocate blocks in new location
278          */
279         updcsloc(utime, fsi, fso, Nflag);
280
281         /*
282          * Now write the cylinder summary back to disk.
283          */
284         for (i = 0; i < sblock.fs_cssize; i += sblock.fs_bsize) {
285                 wtfs(fsbtodb(&sblock, sblock.fs_csaddr + numfrags(&sblock, i)),
286                     (size_t)MIN(sblock.fs_cssize - i, sblock.fs_bsize),
287                     (void *)(((char *)fscs) + i), fso, Nflag);
288         }
289         DBG_PRINT0("fscs written\n");
290
291 #ifdef FS_DEBUG
292 {
293         struct csum     *dbg_csp;
294         int     dbg_csc;
295         char    dbg_line[80];
296
297         dbg_csp=fscs;
298         for(dbg_csc=0; dbg_csc<sblock.fs_ncg; dbg_csc++) {
299                 snprintf(dbg_line, sizeof(dbg_line),
300                     "%d. new csum in new location", dbg_csc);
301                 DBG_DUMP_CSUM(&sblock,
302                     dbg_line,
303                     dbg_csp++);
304         }
305 }
306 #endif /* FS_DEBUG */
307
308         /*
309          * Now write the new superblock back to disk.
310          */
311         sblock.fs_time = utime;
312         wtfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
313         DBG_PRINT0("sblock written\n");
314         DBG_DUMP_FS(&sblock,
315             "new initial sblock");
316
317         /*
318          * Clean up the dynamic fields in our superblock copies.
319          */
320         sblock.fs_fmod = 0;
321         sblock.fs_clean = 1;
322         sblock.fs_ronly = 0;
323         sblock.fs_cgrotor = 0;
324         sblock.fs_state = 0;
325         memset((void *)&sblock.fs_fsmnt, 0, sizeof(sblock.fs_fsmnt));
326         sblock.fs_flags &= FS_DOSOFTDEP;
327
328         /*
329          * XXX
330          * The following fields are currently distributed from the superblock
331          * to the copies:
332          *     fs_minfree
333          *     fs_rotdelay
334          *     fs_maxcontig
335          *     fs_maxbpg
336          *     fs_minfree,
337          *     fs_optim
338          *     fs_flags regarding SOFTPDATES
339          *
340          * We probably should rather change the summary for the cylinder group
341          * statistics here to the value of what would be in there, if the file
342          * system were created initially with the new size. Therefor we still
343          * need to find an easy way of calculating that.
344          * Possibly we can try to read the first superblock copy and apply the
345          * "diffed" stats between the old and new superblock by still copying
346          * certain parameters onto that.
347          */
348
349         /*
350          * Write out the duplicate super blocks.
351          */
352         for (cylno = 0; cylno < sblock.fs_ncg; cylno++) {
353                 wtfs(fsbtodb(&sblock, cgsblock(&sblock, cylno)),
354                     (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
355         }
356         DBG_PRINT0("sblock copies written\n");
357         DBG_DUMP_FS(&sblock,
358             "new other sblocks");
359
360         DBG_LEAVE;
361         return;
362 }
363
364 /* ************************************************************ initcg ***** */
365 /*
366  * This creates a new cylinder group structure, for more details please see
367  * the source of newfs(8), as this function is taken over almost unchanged.
368  * As this is never called for the first cylinder group, the special
369  * provisions for that case are removed here.
370  */
371 static void
372 initcg(int cylno, time_t utime, int fso, unsigned int Nflag)
373 {
374         DBG_FUNC("initcg")
375         static void *iobuf;
376         long d, dlower, dupper, blkno, start;
377         ufs2_daddr_t i, cbase, dmax;
378         struct ufs1_dinode *dp1;
379         struct ufs2_dinode *dp2;
380         struct csum *cs;
381
382         if (iobuf == NULL && (iobuf = malloc(sblock.fs_bsize)) == NULL) {
383                 errx(37, "panic: cannot allocate I/O buffer");
384         }
385         /*
386          * Determine block bounds for cylinder group.
387          * Allow space for super block summary information in first
388          * cylinder group.
389          */
390         cbase = cgbase(&sblock, cylno);
391         dmax = cbase + sblock.fs_fpg;
392         if (dmax > sblock.fs_size)
393                 dmax = sblock.fs_size;
394         dlower = cgsblock(&sblock, cylno) - cbase;
395         dupper = cgdmin(&sblock, cylno) - cbase;
396         if (cylno == 0) /* XXX fscs may be relocated */
397                 dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
398         cs = &fscs[cylno];
399         memset(&acg, 0, sblock.fs_cgsize);
400         acg.cg_time = utime;
401         acg.cg_magic = CG_MAGIC;
402         acg.cg_cgx = cylno;
403         acg.cg_niblk = sblock.fs_ipg;
404         acg.cg_initediblk = sblock.fs_ipg;
405         acg.cg_ndblk = dmax - cbase;
406         if (sblock.fs_contigsumsize > 0)
407                 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
408         start = &acg.cg_space[0] - (u_char *)(&acg.cg_firstfield);
409         if (sblock.fs_magic == FS_UFS2_MAGIC) {
410                 acg.cg_iusedoff = start;
411         } else {
412                 acg.cg_old_ncyl = sblock.fs_old_cpg;
413                 acg.cg_old_time = acg.cg_time;
414                 acg.cg_time = 0;
415                 acg.cg_old_niblk = acg.cg_niblk;
416                 acg.cg_niblk = 0;
417                 acg.cg_initediblk = 0;
418                 acg.cg_old_btotoff = start;
419                 acg.cg_old_boff = acg.cg_old_btotoff +
420                     sblock.fs_old_cpg * sizeof(int32_t);
421                 acg.cg_iusedoff = acg.cg_old_boff +
422                     sblock.fs_old_cpg * sizeof(u_int16_t);
423         }
424         acg.cg_freeoff = acg.cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
425         acg.cg_nextfreeoff = acg.cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT);
426         if (sblock.fs_contigsumsize > 0) {
427                 acg.cg_clustersumoff =
428                     roundup(acg.cg_nextfreeoff, sizeof(u_int32_t));
429                 acg.cg_clustersumoff -= sizeof(u_int32_t);
430                 acg.cg_clusteroff = acg.cg_clustersumoff +
431                     (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
432                 acg.cg_nextfreeoff = acg.cg_clusteroff +
433                     howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
434         }
435         if (acg.cg_nextfreeoff > sblock.fs_cgsize) {
436                 /*
437                  * This should never happen as we would have had that panic
438                  * already on file system creation
439                  */
440                 errx(37, "panic: cylinder group too big");
441         }
442         acg.cg_cs.cs_nifree += sblock.fs_ipg;
443         if (cylno == 0)
444                 for (i = 0; i < ROOTINO; i++) {
445                         setbit(cg_inosused(&acg), i);
446                         acg.cg_cs.cs_nifree--;
447                 }
448         /*
449          * XXX Newfs writes out two blocks of initialized inodes
450          *     unconditionally.  Should we check here to make sure that they
451          *     were actually written?
452          */
453         if (sblock.fs_magic == FS_UFS1_MAGIC) {
454                 bzero(iobuf, sblock.fs_bsize);
455                 for (i = 2 * sblock.fs_frag; i < sblock.fs_ipg / INOPF(&sblock);
456                      i += sblock.fs_frag) {
457                         dp1 = (struct ufs1_dinode *)iobuf;
458                         dp2 = (struct ufs2_dinode *)iobuf;
459 #ifdef FSIRAND
460                         for (j = 0; j < INOPB(&sblock); j++)
461                                 if (sblock.fs_magic == FS_UFS1_MAGIC) {
462                                         dp1->di_gen = random();
463                                         dp1++;
464                                 } else {
465                                         dp2->di_gen = random();
466                                         dp2++;
467                                 }
468 #endif
469                         wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno) + i),
470                             sblock.fs_bsize, iobuf, fso, Nflag);
471                 }
472         }
473         if (cylno > 0) {
474                 /*
475                  * In cylno 0, beginning space is reserved
476                  * for boot and super blocks.
477                  */
478                 for (d = 0; d < dlower; d += sblock.fs_frag) {
479                         blkno = d / sblock.fs_frag;
480                         setblock(&sblock, cg_blksfree(&acg), blkno);
481                         if (sblock.fs_contigsumsize > 0)
482                                 setbit(cg_clustersfree(&acg), blkno);
483                         acg.cg_cs.cs_nbfree++;
484                 }
485                 sblock.fs_dsize += dlower;
486         }
487         sblock.fs_dsize += acg.cg_ndblk - dupper;
488         if ((i = dupper % sblock.fs_frag)) {
489                 acg.cg_frsum[sblock.fs_frag - i]++;
490                 for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
491                         setbit(cg_blksfree(&acg), dupper);
492                         acg.cg_cs.cs_nffree++;
493                 }
494         }
495         for (d = dupper; d + sblock.fs_frag <= acg.cg_ndblk;
496              d += sblock.fs_frag) {
497                 blkno = d / sblock.fs_frag;
498                 setblock(&sblock, cg_blksfree(&acg), blkno);
499                 if (sblock.fs_contigsumsize > 0)
500                         setbit(cg_clustersfree(&acg), blkno);
501                 acg.cg_cs.cs_nbfree++;
502         }
503         if (d < acg.cg_ndblk) {
504                 acg.cg_frsum[acg.cg_ndblk - d]++;
505                 for (; d < acg.cg_ndblk; d++) {
506                         setbit(cg_blksfree(&acg), d);
507                         acg.cg_cs.cs_nffree++;
508                 }
509         }
510         if (sblock.fs_contigsumsize > 0) {
511                 int32_t *sump = cg_clustersum(&acg);
512                 u_char *mapp = cg_clustersfree(&acg);
513                 int map = *mapp++;
514                 int bit = 1;
515                 int run = 0;
516
517                 for (i = 0; i < acg.cg_nclusterblks; i++) {
518                         if ((map & bit) != 0)
519                                 run++;
520                         else if (run != 0) {
521                                 if (run > sblock.fs_contigsumsize)
522                                         run = sblock.fs_contigsumsize;
523                                 sump[run]++;
524                                 run = 0;
525                         }
526                         if ((i & (CHAR_BIT - 1)) != CHAR_BIT - 1)
527                                 bit <<= 1;
528                         else {
529                                 map = *mapp++;
530                                 bit = 1;
531                         }
532                 }
533                 if (run != 0) {
534                         if (run > sblock.fs_contigsumsize)
535                                 run = sblock.fs_contigsumsize;
536                         sump[run]++;
537                 }
538         }
539         sblock.fs_cstotal.cs_ndir += acg.cg_cs.cs_ndir;
540         sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
541         sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
542         sblock.fs_cstotal.cs_nifree += acg.cg_cs.cs_nifree;
543         *cs = acg.cg_cs;
544         wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
545                 sblock.fs_bsize, (char *)&acg, fso, Nflag);
546         DBG_DUMP_CG(&sblock,
547             "new cg",
548             &acg);
549
550         DBG_LEAVE;
551         return;
552 }
553
554 /* ******************************************************* frag_adjust ***** */
555 /*
556  * Here we add or subtract (sign +1/-1) the available fragments in a given
557  * block to or from the fragment statistics. By subtracting before and adding
558  * after an operation on the free frag map we can easy update the fragment
559  * statistic, which seems to be otherwise a rather complex operation.
560  */
561 static void
562 frag_adjust(ufs2_daddr_t frag, int sign)
563 {
564         DBG_FUNC("frag_adjust")
565         int fragsize;
566         int f;
567
568         DBG_ENTER;
569
570         fragsize=0;
571         /*
572          * Here frag only needs to point to any fragment in the block we want
573          * to examine.
574          */
575         for(f=rounddown(frag, sblock.fs_frag);
576             f<roundup(frag+1, sblock.fs_frag);
577             f++) {
578                 /*
579                  * Count contiguous free fragments.
580                  */
581                 if(isset(cg_blksfree(&acg), f)) {
582                         fragsize++;
583                 } else {
584                         if(fragsize && fragsize<sblock.fs_frag) {
585                                 /*
586                                  * We found something in between.
587                                  */
588                                 acg.cg_frsum[fragsize]+=sign;
589                                 DBG_PRINT2("frag_adjust [%d]+=%d\n",
590                                     fragsize,
591                                     sign);
592                         }
593                         fragsize=0;
594                 }
595         }
596         if(fragsize && fragsize<sblock.fs_frag) {
597                 /*
598                  * We found something.
599                  */
600                 acg.cg_frsum[fragsize]+=sign;
601                 DBG_PRINT2("frag_adjust [%d]+=%d\n",
602                     fragsize,
603                     sign);
604         }
605         DBG_PRINT2("frag_adjust [[%d]]+=%d\n",
606             fragsize,
607             sign);
608
609         DBG_LEAVE;
610         return;
611 }
612
613 /* ******************************************************* cond_bl_upd ***** */
614 /*
615  * Here we conditionally update a pointer to a fragment. We check for all
616  * relocated blocks if any of its fragments is referenced by the current
617  * field, and update the pointer to the respective fragment in our new
618  * block.  If we find a reference we write back the block immediately,
619  * as there is no easy way for our general block reading engine to figure
620  * out if a write back operation is needed.
621  */
622 static int
623 cond_bl_upd(ufs2_daddr_t *block, struct gfs_bpp *field, int fsi, int fso,
624     unsigned int Nflag)
625 {
626         DBG_FUNC("cond_bl_upd")
627         struct gfs_bpp *f;
628         ufs2_daddr_t src, dst;
629         int fragnum;
630         void *ibuf;
631
632         DBG_ENTER;
633
634         for (f = field; f->old != 0; f++) {
635                 src = *block;
636                 if (fragstoblks(&sblock, src) != f->old)
637                         continue;
638                 /*
639                  * The fragment is part of the block, so update.
640                  */
641                 dst = blkstofrags(&sblock, f->new);
642                 fragnum = fragnum(&sblock, src);
643                 *block = dst + fragnum;
644                 f->found++;
645                 DBG_PRINT3("scg (%jd->%jd)[%d] reference updated\n",
646                     (intmax_t)f->old,
647                     (intmax_t)f->new,
648                     fragnum);
649
650                 /*
651                  * Copy the block back immediately.
652                  *
653                  * XXX  If src is is from an indirect block we have
654                  *      to implement copy on write here in case of
655                  *      active snapshots.
656                  */
657                 ibuf = malloc(sblock.fs_bsize);
658                 if (!ibuf)
659                         errx(1, "malloc failed");
660                 src -= fragnum;
661                 rdfs(fsbtodb(&sblock, src), (size_t)sblock.fs_bsize, ibuf, fsi);
662                 wtfs(dst, (size_t)sblock.fs_bsize, ibuf, fso, Nflag);
663                 free(ibuf);
664                 /*
665                  * The same block can't be found again in this loop.
666                  */
667                 return (1);
668         }
669
670         DBG_LEAVE;
671         return (0);
672 }
673
674 /* ************************************************************ updjcg ***** */
675 /*
676  * Here we do all needed work for the former last cylinder group. It has to be
677  * changed in any case, even if the file system ended exactly on the end of
678  * this group, as there is some slightly inconsistent handling of the number
679  * of cylinders in the cylinder group. We start again by reading the cylinder
680  * group from disk. If the last block was not fully available, we first handle
681  * the missing fragments, then we handle all new full blocks in that file
682  * system and finally we handle the new last fragmented block in the file
683  * system.  We again have to handle the fragment statistics rotational layout
684  * tables and cluster summary during all those operations.
685  */
686 static void
687 updjcg(int cylno, time_t utime, int fsi, int fso, unsigned int Nflag)
688 {
689         DBG_FUNC("updjcg")
690         ufs2_daddr_t    cbase, dmax, dupper;
691         struct csum     *cs;
692         int     i,k;
693         int     j=0;
694
695         DBG_ENTER;
696
697         /*
698          * Read the former last (joining) cylinder group from disk, and make
699          * a copy.
700          */
701         rdfs(fsbtodb(&osblock, cgtod(&osblock, cylno)),
702             (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
703         DBG_PRINT0("jcg read\n");
704         DBG_DUMP_CG(&sblock,
705             "old joining cg",
706             &aocg);
707
708         memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
709
710         /*
711          * If the cylinder group had already its new final size almost
712          * nothing is to be done ... except:
713          * For some reason the value of cg_ncyl in the last cylinder group has
714          * to be zero instead of fs_cpg. As this is now no longer the last
715          * cylinder group we have to change that value now to fs_cpg.
716          */
717
718         if(cgbase(&osblock, cylno+1) == osblock.fs_size) {
719                 if (sblock.fs_magic == FS_UFS1_MAGIC)
720                         acg.cg_old_ncyl=sblock.fs_old_cpg;
721
722                 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
723                     (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
724                 DBG_PRINT0("jcg written\n");
725                 DBG_DUMP_CG(&sblock,
726                     "new joining cg",
727                     &acg);
728
729                 DBG_LEAVE;
730                 return;
731         }
732
733         /*
734          * Set up some variables needed later.
735          */
736         cbase = cgbase(&sblock, cylno);
737         dmax = cbase + sblock.fs_fpg;
738         if (dmax > sblock.fs_size)
739                 dmax = sblock.fs_size;
740         dupper = cgdmin(&sblock, cylno) - cbase;
741         if (cylno == 0) { /* XXX fscs may be relocated */
742                 dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
743         }
744
745         /*
746          * Set pointer to the cylinder summary for our cylinder group.
747          */
748         cs = fscs + cylno;
749
750         /*
751          * Touch the cylinder group, update all fields in the cylinder group as
752          * needed, update the free space in the superblock.
753          */
754         acg.cg_time = utime;
755         if (cylno == sblock.fs_ncg - 1) {
756                 /*
757                  * This is still the last cylinder group.
758                  */
759                 if (sblock.fs_magic == FS_UFS1_MAGIC)
760                         acg.cg_old_ncyl =
761                             sblock.fs_old_ncyl % sblock.fs_old_cpg;
762         } else {
763                 acg.cg_old_ncyl = sblock.fs_old_cpg;
764         }
765         DBG_PRINT2("jcg dbg: %d %u",
766             cylno,
767             sblock.fs_ncg);
768 #ifdef FS_DEBUG
769         if (sblock.fs_magic == FS_UFS1_MAGIC)
770                 DBG_PRINT2("%d %u",
771                     acg.cg_old_ncyl,
772                     sblock.fs_old_cpg);
773 #endif
774         DBG_PRINT0("\n");
775         acg.cg_ndblk = dmax - cbase;
776         sblock.fs_dsize += acg.cg_ndblk-aocg.cg_ndblk;
777         if (sblock.fs_contigsumsize > 0) {
778                 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
779         }
780
781         /*
782          * Now we have to update the free fragment bitmap for our new free
783          * space.  There again we have to handle the fragmentation and also
784          * the rotational layout tables and the cluster summary.  This is
785          * also done per fragment for the first new block if the old file
786          * system end was not on a block boundary, per fragment for the new
787          * last block if the new file system end is not on a block boundary,
788          * and per block for all space in between.
789          *
790          * Handle the first new block here if it was partially available
791          * before.
792          */
793         if(osblock.fs_size % sblock.fs_frag) {
794                 if(roundup(osblock.fs_size, sblock.fs_frag)<=sblock.fs_size) {
795                         /*
796                          * The new space is enough to fill at least this
797                          * block
798                          */
799                         j=0;
800                         for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag)-1;
801                             i>=osblock.fs_size-cbase;
802                             i--) {
803                                 setbit(cg_blksfree(&acg), i);
804                                 acg.cg_cs.cs_nffree++;
805                                 j++;
806                         }
807
808                         /*
809                          * Check if the fragment just created could join an
810                          * already existing fragment at the former end of the
811                          * file system.
812                          */
813                         if(isblock(&sblock, cg_blksfree(&acg),
814                             ((osblock.fs_size - cgbase(&sblock, cylno))/
815                             sblock.fs_frag))) {
816                                 /*
817                                  * The block is now completely available.
818                                  */
819                                 DBG_PRINT0("block was\n");
820                                 acg.cg_frsum[osblock.fs_size%sblock.fs_frag]--;
821                                 acg.cg_cs.cs_nbfree++;
822                                 acg.cg_cs.cs_nffree-=sblock.fs_frag;
823                                 k=rounddown(osblock.fs_size-cbase,
824                                     sblock.fs_frag);
825                                 updclst((osblock.fs_size-cbase)/sblock.fs_frag);
826                         } else {
827                                 /*
828                                  * Lets rejoin a possible partially growed
829                                  * fragment.
830                                  */
831                                 k=0;
832                                 while(isset(cg_blksfree(&acg), i) &&
833                                     (i>=rounddown(osblock.fs_size-cbase,
834                                     sblock.fs_frag))) {
835                                         i--;
836                                         k++;
837                                 }
838                                 if(k) {
839                                         acg.cg_frsum[k]--;
840                                 }
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;
850                                 i--) {
851                                 setbit(cg_blksfree(&acg), i);
852                                 acg.cg_cs.cs_nffree++;
853                                 j++;
854                         }
855                         /*
856                          * Lets rejoin a possible partially growed fragment.
857                          */
858                         k=0;
859                         while(isset(cg_blksfree(&acg), i) &&
860                             (i>=rounddown(osblock.fs_size-cbase,
861                             sblock.fs_frag))) {
862                                 i--;
863                                 k++;
864                         }
865                         if(k) {
866                                 acg.cg_frsum[k]--;
867                         }
868                         acg.cg_frsum[k+j]++;
869                 }
870         }
871
872         /*
873          * Handle all new complete blocks here.
874          */
875         for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag);
876             i+sblock.fs_frag<=dmax-cbase;       /* XXX <= or only < ? */
877             i+=sblock.fs_frag) {
878                 j = i / sblock.fs_frag;
879                 setblock(&sblock, cg_blksfree(&acg), j);
880                 updclst(j);
881                 acg.cg_cs.cs_nbfree++;
882         }
883
884         /*
885          * Handle the last new block if there are stll some new fragments left.
886          * Here we don't have to bother about the cluster summary or the even
887          * the rotational layout table.
888          */
889         if (i < (dmax - cbase)) {
890                 acg.cg_frsum[dmax - cbase - i]++;
891                 for (; i < dmax - cbase; i++) {
892                         setbit(cg_blksfree(&acg), i);
893                         acg.cg_cs.cs_nffree++;
894                 }
895         }
896
897         sblock.fs_cstotal.cs_nffree +=
898             (acg.cg_cs.cs_nffree - aocg.cg_cs.cs_nffree);
899         sblock.fs_cstotal.cs_nbfree +=
900             (acg.cg_cs.cs_nbfree - aocg.cg_cs.cs_nbfree);
901         /*
902          * The following statistics are not changed here:
903          *     sblock.fs_cstotal.cs_ndir
904          *     sblock.fs_cstotal.cs_nifree
905          * As the statistics for this cylinder group are ready, copy it to
906          * the summary information array.
907          */
908         *cs = acg.cg_cs;
909
910         /*
911          * Write the updated "joining" cylinder group back to disk.
912          */
913         wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize,
914             (void *)&acg, fso, Nflag);
915         DBG_PRINT0("jcg written\n");
916         DBG_DUMP_CG(&sblock,
917             "new joining cg",
918             &acg);
919
920         DBG_LEAVE;
921         return;
922 }
923
924 /* ********************************************************** updcsloc ***** */
925 /*
926  * Here we update the location of the cylinder summary. We have two possible
927  * ways of growing the cylinder summary.
928  * (1)  We can try to grow the summary in the current location, and relocate
929  *      possibly used blocks within the current cylinder group.
930  * (2)  Alternatively we can relocate the whole cylinder summary to the first
931  *      new completely empty cylinder group. Once the cylinder summary is no
932  *      longer in the beginning of the first cylinder group you should never
933  *      use a version of fsck which is not aware of the possibility to have
934  *      this structure in a non standard place.
935  * Option (1) is considered to be less intrusive to the structure of the file-
936  * system. So we try to stick to that whenever possible. If there is not enough
937  * space in the cylinder group containing the cylinder summary we have to use
938  * method (2). In case of active snapshots in the file system we probably can
939  * completely avoid implementing copy on write if we stick to method (2) only.
940  */
941 static void
942 updcsloc(time_t utime, int fsi, int fso, unsigned int Nflag)
943 {
944         DBG_FUNC("updcsloc")
945         struct csum     *cs;
946         int     ocscg, ncscg;
947         int     blocks;
948         ufs2_daddr_t    cbase, dupper, odupper, d, f, g;
949         int     ind;
950         int     cylno, inc;
951         struct gfs_bpp  *bp;
952         int     i, l;
953         int     lcs=0;
954         int     block;
955
956         DBG_ENTER;
957
958         if(howmany(sblock.fs_cssize, sblock.fs_fsize) ==
959             howmany(osblock.fs_cssize, osblock.fs_fsize)) {
960                 /*
961                  * No new fragment needed.
962                  */
963                 DBG_LEAVE;
964                 return;
965         }
966         ocscg=dtog(&osblock, osblock.fs_csaddr);
967         cs=fscs+ocscg;
968         blocks = 1+howmany(sblock.fs_cssize, sblock.fs_bsize)-
969             howmany(osblock.fs_cssize, osblock.fs_bsize);
970
971         /*
972          * Read original cylinder group from disk, and make a copy.
973          * XXX  If Nflag is set in some very rare cases we now miss
974          *      some changes done in updjcg by reading the unmodified
975          *      block from disk.
976          */
977         rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)),
978             (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
979         DBG_PRINT0("oscg read\n");
980         DBG_DUMP_CG(&sblock,
981             "old summary cg",
982             &aocg);
983
984         memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
985
986         /*
987          * Touch the cylinder group, set up local variables needed later
988          * and update the superblock.
989          */
990         acg.cg_time = utime;
991
992         /*
993          * XXX  In the case of having active snapshots we may need much more
994          *      blocks for the copy on write. We need each block twice, and
995          *      also up to 8*3 blocks for indirect blocks for all possible
996          *      references.
997          */
998         if(/*((int)sblock.fs_time&0x3)>0||*/ cs->cs_nbfree < blocks) {
999                 /*
1000                  * There is not enough space in the old cylinder group to
1001                  * relocate all blocks as needed, so we relocate the whole
1002                  * cylinder group summary to a new group. We try to use the
1003                  * first complete new cylinder group just created. Within the
1004                  * cylinder group we align the area immediately after the
1005                  * cylinder group information location in order to be as
1006                  * close as possible to the original implementation of ffs.
1007                  *
1008                  * First we have to make sure we'll find enough space in the
1009                  * new cylinder group. If not, then we currently give up.
1010                  * We start with freeing everything which was used by the
1011                  * fragments of the old cylinder summary in the current group.
1012                  * Now we write back the group meta data, read in the needed
1013                  * meta data from the new cylinder group, and start allocating
1014                  * within that group. Here we can assume, the group to be
1015                  * completely empty. Which makes the handling of fragments and
1016                  * clusters a lot easier.
1017                  */
1018                 DBG_TRC;
1019                 if(sblock.fs_ncg-osblock.fs_ncg < 2) {
1020                         errx(2, "panic: not enough space");
1021                 }
1022
1023                 /*
1024                  * Point "d" to the first fragment not used by the cylinder
1025                  * summary.
1026                  */
1027                 d=osblock.fs_csaddr+(osblock.fs_cssize/osblock.fs_fsize);
1028
1029                 /*
1030                  * Set up last cluster size ("lcs") already here. Calculate
1031                  * the size for the trailing cluster just behind where "d"
1032                  * points to.
1033                  */
1034                 if(sblock.fs_contigsumsize > 0) {
1035                         for(block=howmany(d%sblock.fs_fpg, sblock.fs_frag),
1036                             lcs=0; lcs<sblock.fs_contigsumsize;
1037                             block++, lcs++) {
1038                                 if(isclr(cg_clustersfree(&acg), block)){
1039                                         break;
1040                                 }
1041                         }
1042                 }
1043
1044                 /*
1045                  * Point "d" to the last frag used by the cylinder summary.
1046                  */
1047                 d--;
1048
1049                 DBG_PRINT1("d=%jd\n",
1050                     (intmax_t)d);
1051                 if((d+1)%sblock.fs_frag) {
1052                         /*
1053                          * The end of the cylinder summary is not a complete
1054                          * block.
1055                          */
1056                         DBG_TRC;
1057                         frag_adjust(d%sblock.fs_fpg, -1);
1058                         for(; (d+1)%sblock.fs_frag; d--) {
1059                                 DBG_PRINT1("d=%jd\n",
1060                                     (intmax_t)d);
1061                                 setbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1062                                 acg.cg_cs.cs_nffree++;
1063                                 sblock.fs_cstotal.cs_nffree++;
1064                         }
1065                         /*
1066                          * Point "d" to the last fragment of the last
1067                          * (incomplete) block of the cylinder summary.
1068                          */
1069                         d++;
1070                         frag_adjust(d%sblock.fs_fpg, 1);
1071
1072                         if(isblock(&sblock, cg_blksfree(&acg),
1073                             (d%sblock.fs_fpg)/sblock.fs_frag)) {
1074                                 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1075                                 acg.cg_cs.cs_nffree-=sblock.fs_frag;
1076                                 acg.cg_cs.cs_nbfree++;
1077                                 sblock.fs_cstotal.cs_nffree-=sblock.fs_frag;
1078                                 sblock.fs_cstotal.cs_nbfree++;
1079                                 if(sblock.fs_contigsumsize > 0) {
1080                                         setbit(cg_clustersfree(&acg),
1081                                             (d%sblock.fs_fpg)/sblock.fs_frag);
1082                                         if(lcs < sblock.fs_contigsumsize) {
1083                                                 if(lcs) {
1084                                                         cg_clustersum(&acg)
1085                                                             [lcs]--;
1086                                                 }
1087                                                 lcs++;
1088                                                 cg_clustersum(&acg)[lcs]++;
1089                                         }
1090                                 }
1091                         }
1092                         /*
1093                          * Point "d" to the first fragment of the block before
1094                          * the last incomplete block.
1095                          */
1096                         d--;
1097                 }
1098
1099                 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1100                 for(d=rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr;
1101                     d-=sblock.fs_frag) {
1102                         DBG_TRC;
1103                         DBG_PRINT1("d=%jd\n", (intmax_t)d);
1104                         setblock(&sblock, cg_blksfree(&acg),
1105                             (d%sblock.fs_fpg)/sblock.fs_frag);
1106                         acg.cg_cs.cs_nbfree++;
1107                         sblock.fs_cstotal.cs_nbfree++;
1108                         if(sblock.fs_contigsumsize > 0) {
1109                                 setbit(cg_clustersfree(&acg),
1110                                     (d%sblock.fs_fpg)/sblock.fs_frag);
1111                                 /*
1112                                  * The last cluster size is already set up.
1113                                  */
1114                                 if(lcs < sblock.fs_contigsumsize) {
1115                                         if(lcs) {
1116                                                 cg_clustersum(&acg)[lcs]--;
1117                                         }
1118                                         lcs++;
1119                                         cg_clustersum(&acg)[lcs]++;
1120                                 }
1121                         }
1122                 }
1123                 *cs = acg.cg_cs;
1124
1125                 /*
1126                  * Now write the former cylinder group containing the cylinder
1127                  * summary back to disk.
1128                  */
1129                 wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)),
1130                     (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1131                 DBG_PRINT0("oscg written\n");
1132                 DBG_DUMP_CG(&sblock,
1133                     "old summary cg",
1134                     &acg);
1135
1136                 /*
1137                  * Find the beginning of the new cylinder group containing the
1138                  * cylinder summary.
1139                  */
1140                 sblock.fs_csaddr=cgdmin(&sblock, osblock.fs_ncg);
1141                 ncscg=dtog(&sblock, sblock.fs_csaddr);
1142                 cs=fscs+ncscg;
1143
1144
1145                 /*
1146                  * If Nflag is specified, we would now read random data instead
1147                  * of an empty cg structure from disk. So we can't simulate that
1148                  * part for now.
1149                  */
1150                 if(Nflag) {
1151                         DBG_PRINT0("nscg update skipped\n");
1152                         DBG_LEAVE;
1153                         return;
1154                 }
1155
1156                 /*
1157                  * Read the future cylinder group containing the cylinder
1158                  * summary from disk, and make a copy.
1159                  */
1160                 rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1161                     (size_t)sblock.fs_cgsize, (void *)&aocg, fsi);
1162                 DBG_PRINT0("nscg read\n");
1163                 DBG_DUMP_CG(&sblock,
1164                     "new summary cg",
1165                     &aocg);
1166
1167                 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
1168
1169                 /*
1170                  * Allocate all complete blocks used by the new cylinder
1171                  * summary.
1172                  */
1173                 for(d=sblock.fs_csaddr; d+sblock.fs_frag <=
1174                     sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize);
1175                     d+=sblock.fs_frag) {
1176                         clrblock(&sblock, cg_blksfree(&acg),
1177                             (d%sblock.fs_fpg)/sblock.fs_frag);
1178                         acg.cg_cs.cs_nbfree--;
1179                         sblock.fs_cstotal.cs_nbfree--;
1180                         if(sblock.fs_contigsumsize > 0) {
1181                                 clrbit(cg_clustersfree(&acg),
1182                                     (d%sblock.fs_fpg)/sblock.fs_frag);
1183                         }
1184                 }
1185
1186                 /*
1187                  * Allocate all fragments used by the cylinder summary in the
1188                  * last block.
1189                  */
1190                 if(d<sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize)) {
1191                         for(; d-sblock.fs_csaddr<
1192                             sblock.fs_cssize/sblock.fs_fsize;
1193                             d++) {
1194                                 clrbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1195                                 acg.cg_cs.cs_nffree--;
1196                                 sblock.fs_cstotal.cs_nffree--;
1197                         }
1198                         acg.cg_cs.cs_nbfree--;
1199                         acg.cg_cs.cs_nffree+=sblock.fs_frag;
1200                         sblock.fs_cstotal.cs_nbfree--;
1201                         sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1202                         if(sblock.fs_contigsumsize > 0) {
1203                                 clrbit(cg_clustersfree(&acg),
1204                                     (d%sblock.fs_fpg)/sblock.fs_frag);
1205                         }
1206
1207                         frag_adjust(d%sblock.fs_fpg, +1);
1208                 }
1209                 /*
1210                  * XXX  Handle the cluster statistics here in the case this
1211                  *      cylinder group is now almost full, and the remaining
1212                  *      space is less then the maximum cluster size. This is
1213                  *      probably not needed, as you would hardly find a file
1214                  *      system which has only MAXCSBUFS+FS_MAXCONTIG of free
1215                  *      space right behind the cylinder group information in
1216                  *      any new cylinder group.
1217                  */
1218
1219                 /*
1220                  * Update our statistics in the cylinder summary.
1221                  */
1222                 *cs = acg.cg_cs;
1223
1224                 /*
1225                  * Write the new cylinder group containing the cylinder summary
1226                  * back to disk.
1227                  */
1228                 wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1229                     (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1230                 DBG_PRINT0("nscg written\n");
1231                 DBG_DUMP_CG(&sblock,
1232                     "new summary cg",
1233                     &acg);
1234
1235                 DBG_LEAVE;
1236                 return;
1237         }
1238         /*
1239          * We have got enough of space in the current cylinder group, so we
1240          * can relocate just a few blocks, and let the summary information
1241          * grow in place where it is right now.
1242          */
1243         DBG_TRC;
1244
1245         cbase = cgbase(&osblock, ocscg);        /* old and new are equal */
1246         dupper = sblock.fs_csaddr - cbase +
1247             howmany(sblock.fs_cssize, sblock.fs_fsize);
1248         odupper = osblock.fs_csaddr - cbase +
1249             howmany(osblock.fs_cssize, osblock.fs_fsize);
1250
1251         sblock.fs_dsize -= dupper-odupper;
1252
1253         /*
1254          * Allocate the space for the array of blocks to be relocated.
1255          */
1256         bp=(struct gfs_bpp *)malloc(((dupper-odupper)/sblock.fs_frag+2)*
1257             sizeof(struct gfs_bpp));
1258         if(bp == NULL) {
1259                 errx(1, "malloc failed");
1260         }
1261         memset((char *)bp, 0, ((dupper-odupper)/sblock.fs_frag+2)*
1262             sizeof(struct gfs_bpp));
1263
1264         /*
1265          * Lock all new frags needed for the cylinder group summary. This is
1266          * done per fragment in the first and last block of the new required
1267          * area, and per block for all other blocks.
1268          *
1269          * Handle the first new block here (but only if some fragments where
1270          * already used for the cylinder summary).
1271          */
1272         ind=0;
1273         frag_adjust(odupper, -1);
1274         for(d=odupper; ((d<dupper)&&(d%sblock.fs_frag)); d++) {
1275                 DBG_PRINT1("scg first frag check loop d=%jd\n",
1276                     (intmax_t)d);
1277                 if(isclr(cg_blksfree(&acg), d)) {
1278                         if (!ind) {
1279                                 bp[ind].old=d/sblock.fs_frag;
1280                                 bp[ind].flags|=GFS_FL_FIRST;
1281                                 if(roundup(d, sblock.fs_frag) >= dupper) {
1282                                         bp[ind].flags|=GFS_FL_LAST;
1283                                 }
1284                                 ind++;
1285                         }
1286                 } else {
1287                         clrbit(cg_blksfree(&acg), d);
1288                         acg.cg_cs.cs_nffree--;
1289                         sblock.fs_cstotal.cs_nffree--;
1290                 }
1291                 /*
1292                  * No cluster handling is needed here, as there was at least
1293                  * one fragment in use by the cylinder summary in the old
1294                  * file system.
1295                  * No block-free counter handling here as this block was not
1296                  * a free block.
1297                  */
1298         }
1299         frag_adjust(odupper, 1);
1300
1301         /*
1302          * Handle all needed complete blocks here.
1303          */
1304         for(; d+sblock.fs_frag<=dupper; d+=sblock.fs_frag) {
1305                 DBG_PRINT1("scg block check loop d=%jd\n",
1306                     (intmax_t)d);
1307                 if(!isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1308                         for(f=d; f<d+sblock.fs_frag; f++) {
1309                                 if(isset(cg_blksfree(&aocg), f)) {
1310                                         acg.cg_cs.cs_nffree--;
1311                                         sblock.fs_cstotal.cs_nffree--;
1312                                 }
1313                         }
1314                         clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1315                         bp[ind].old=d/sblock.fs_frag;
1316                         ind++;
1317                 } else {
1318                         clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1319                         acg.cg_cs.cs_nbfree--;
1320                         sblock.fs_cstotal.cs_nbfree--;
1321                         if(sblock.fs_contigsumsize > 0) {
1322                                 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1323                                 for(lcs=0, l=(d/sblock.fs_frag)+1;
1324                                     lcs<sblock.fs_contigsumsize;
1325                                     l++, lcs++ ) {
1326                                         if(isclr(cg_clustersfree(&acg),l)){
1327                                                 break;
1328                                         }
1329                                 }
1330                                 if(lcs < sblock.fs_contigsumsize) {
1331                                         cg_clustersum(&acg)[lcs+1]--;
1332                                         if(lcs) {
1333                                                 cg_clustersum(&acg)[lcs]++;
1334                                         }
1335                                 }
1336                         }
1337                 }
1338                 /*
1339                  * No fragment counter handling is needed here, as this finally
1340                  * doesn't change after the relocation.
1341                  */
1342         }
1343
1344         /*
1345          * Handle all fragments needed in the last new affected block.
1346          */
1347         if(d<dupper) {
1348                 frag_adjust(dupper-1, -1);
1349
1350                 if(isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1351                         acg.cg_cs.cs_nbfree--;
1352                         sblock.fs_cstotal.cs_nbfree--;
1353                         acg.cg_cs.cs_nffree+=sblock.fs_frag;
1354                         sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1355                         if(sblock.fs_contigsumsize > 0) {
1356                                 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1357                                 for(lcs=0, l=(d/sblock.fs_frag)+1;
1358                                     lcs<sblock.fs_contigsumsize;
1359                                     l++, lcs++ ) {
1360                                         if(isclr(cg_clustersfree(&acg),l)){
1361                                                 break;
1362                                         }
1363                                 }
1364                                 if(lcs < sblock.fs_contigsumsize) {
1365                                         cg_clustersum(&acg)[lcs+1]--;
1366                                         if(lcs) {
1367                                                 cg_clustersum(&acg)[lcs]++;
1368                                         }
1369                                 }
1370                         }
1371                 }
1372
1373                 for(; d<dupper; d++) {
1374                         DBG_PRINT1("scg second frag check loop d=%jd\n",
1375                             (intmax_t)d);
1376                         if(isclr(cg_blksfree(&acg), d)) {
1377                                 bp[ind].old=d/sblock.fs_frag;
1378                                 bp[ind].flags|=GFS_FL_LAST;
1379                         } else {
1380                                 clrbit(cg_blksfree(&acg), d);
1381                                 acg.cg_cs.cs_nffree--;
1382                                 sblock.fs_cstotal.cs_nffree--;
1383                         }
1384                 }
1385                 if(bp[ind].flags & GFS_FL_LAST) { /* we have to advance here */
1386                         ind++;
1387                 }
1388                 frag_adjust(dupper-1, 1);
1389         }
1390
1391         /*
1392          * If we found a block to relocate just do so.
1393          */
1394         if(ind) {
1395                 for(i=0; i<ind; i++) {
1396                         if(!bp[i].old) { /* no more blocks listed */
1397                                 /*
1398                                  * XXX  A relative blocknumber should not be
1399                                  *      zero, which is not explicitly
1400                                  *      guaranteed by our code.
1401                                  */
1402                                 break;
1403                         }
1404                         /*
1405                          * Allocate a complete block in the same (current)
1406                          * cylinder group.
1407                          */
1408                         bp[i].new=alloc()/sblock.fs_frag;
1409
1410                         /*
1411                          * There is no frag_adjust() needed for the new block
1412                          * as it will have no fragments yet :-).
1413                          */
1414                         for(f=bp[i].old*sblock.fs_frag,
1415                             g=bp[i].new*sblock.fs_frag;
1416                             f<(bp[i].old+1)*sblock.fs_frag;
1417                             f++, g++) {
1418                                 if(isset(cg_blksfree(&aocg), f)) {
1419                                         setbit(cg_blksfree(&acg), g);
1420                                         acg.cg_cs.cs_nffree++;
1421                                         sblock.fs_cstotal.cs_nffree++;
1422                                 }
1423                         }
1424
1425                         /*
1426                          * Special handling is required if this was the first
1427                          * block. We have to consider the fragments which were
1428                          * used by the cylinder summary in the original block
1429                          * which re to be free in the copy of our block.  We
1430                          * have to be careful if this first block happens to
1431                          * be also the last block to be relocated.
1432                          */
1433                         if(bp[i].flags & GFS_FL_FIRST) {
1434                                 for(f=bp[i].old*sblock.fs_frag,
1435                                     g=bp[i].new*sblock.fs_frag;
1436                                     f<odupper;
1437                                     f++, g++) {
1438                                         setbit(cg_blksfree(&acg), g);
1439                                         acg.cg_cs.cs_nffree++;
1440                                         sblock.fs_cstotal.cs_nffree++;
1441                                 }
1442                                 if(!(bp[i].flags & GFS_FL_LAST)) {
1443                                         frag_adjust(bp[i].new*sblock.fs_frag,1);
1444                                 }
1445                         }
1446
1447                         /*
1448                          * Special handling is required if this is the last
1449                          * block to be relocated.
1450                          */
1451                         if(bp[i].flags & GFS_FL_LAST) {
1452                                 frag_adjust(bp[i].new*sblock.fs_frag, 1);
1453                                 frag_adjust(bp[i].old*sblock.fs_frag, -1);
1454                                 for(f=dupper;
1455                                     f<roundup(dupper, sblock.fs_frag);
1456                                     f++) {
1457                                         if(isclr(cg_blksfree(&acg), f)) {
1458                                                 setbit(cg_blksfree(&acg), f);
1459                                                 acg.cg_cs.cs_nffree++;
1460                                                 sblock.fs_cstotal.cs_nffree++;
1461                                         }
1462                                 }
1463                                 frag_adjust(bp[i].old*sblock.fs_frag, 1);
1464                         }
1465
1466                         /*
1467                          * !!! Attach the cylindergroup offset here.
1468                          */
1469                         bp[i].old+=cbase/sblock.fs_frag;
1470                         bp[i].new+=cbase/sblock.fs_frag;
1471
1472                         /*
1473                          * Copy the content of the block.
1474                          */
1475                         /*
1476                          * XXX  Here we will have to implement a copy on write
1477                          *      in the case we have any active snapshots.
1478                          */
1479                         rdfs(fsbtodb(&sblock, bp[i].old*sblock.fs_frag),
1480                             (size_t)sblock.fs_bsize, (void *)&ablk, fsi);
1481                         wtfs(fsbtodb(&sblock, bp[i].new*sblock.fs_frag),
1482                             (size_t)sblock.fs_bsize, (void *)&ablk, fso, Nflag);
1483                         DBG_DUMP_HEX(&sblock,
1484                             "copied full block",
1485                             (unsigned char *)&ablk);
1486
1487                         DBG_PRINT2("scg (%jd->%jd) block relocated\n",
1488                             (intmax_t)bp[i].old,
1489                             (intmax_t)bp[i].new);
1490                 }
1491
1492                 /*
1493                  * Now we have to update all references to any fragment which
1494                  * belongs to any block relocated. We iterate now over all
1495                  * cylinder groups, within those over all non zero length
1496                  * inodes.
1497                  */
1498                 for(cylno=0; cylno<osblock.fs_ncg; cylno++) {
1499                         DBG_PRINT1("scg doing cg (%d)\n",
1500                             cylno);
1501                         for(inc=osblock.fs_ipg-1 ; inc>0 ; inc--) {
1502                                 updrefs(cylno, (ino_t)inc, bp, fsi, fso, Nflag);
1503                         }
1504                 }
1505
1506                 /*
1507                  * All inodes are checked, now make sure the number of
1508                  * references found make sense.
1509                  */
1510                 for(i=0; i<ind; i++) {
1511                         if(!bp[i].found || (bp[i].found>sblock.fs_frag)) {
1512                                 warnx("error: %jd refs found for block %jd.",
1513                                     (intmax_t)bp[i].found, (intmax_t)bp[i].old);
1514                         }
1515
1516                 }
1517         }
1518         /*
1519          * The following statistics are not changed here:
1520          *     sblock.fs_cstotal.cs_ndir
1521          *     sblock.fs_cstotal.cs_nifree
1522          * The following statistics were already updated on the fly:
1523          *     sblock.fs_cstotal.cs_nffree
1524          *     sblock.fs_cstotal.cs_nbfree
1525          * As the statistics for this cylinder group are ready, copy it to
1526          * the summary information array.
1527          */
1528
1529         *cs = acg.cg_cs;
1530
1531         /*
1532          * Write summary cylinder group back to disk.
1533          */
1534         wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)), (size_t)sblock.fs_cgsize,
1535             (void *)&acg, fso, Nflag);
1536         DBG_PRINT0("scg written\n");
1537         DBG_DUMP_CG(&sblock,
1538             "new summary cg",
1539             &acg);
1540
1541         DBG_LEAVE;
1542         return;
1543 }
1544
1545 /* ************************************************************** rdfs ***** */
1546 /*
1547  * Here we read some block(s) from disk.
1548  */
1549 static void
1550 rdfs(ufs2_daddr_t bno, size_t size, void *bf, int fsi)
1551 {
1552         DBG_FUNC("rdfs")
1553         ssize_t n;
1554
1555         DBG_ENTER;
1556
1557         if (bno < 0) {
1558                 err(32, "rdfs: attempting to read negative block number");
1559         }
1560         if (lseek(fsi, (off_t)bno * DEV_BSIZE, 0) < 0) {
1561                 err(33, "rdfs: seek error: %jd", (intmax_t)bno);
1562         }
1563         n = read(fsi, bf, size);
1564         if (n != (ssize_t)size) {
1565                 err(34, "rdfs: read error: %jd", (intmax_t)bno);
1566         }
1567
1568         DBG_LEAVE;
1569         return;
1570 }
1571
1572 /* ************************************************************** wtfs ***** */
1573 /*
1574  * Here we write some block(s) to disk.
1575  */
1576 static void
1577 wtfs(ufs2_daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag)
1578 {
1579         DBG_FUNC("wtfs")
1580         ssize_t n;
1581
1582         DBG_ENTER;
1583
1584         if (Nflag) {
1585                 DBG_LEAVE;
1586                 return;
1587         }
1588         if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0) {
1589                 err(35, "wtfs: seek error: %ld", (long)bno);
1590         }
1591         n = write(fso, bf, size);
1592         if (n != (ssize_t)size) {
1593                 err(36, "wtfs: write error: %ld", (long)bno);
1594         }
1595
1596         DBG_LEAVE;
1597         return;
1598 }
1599
1600 /* ************************************************************* alloc ***** */
1601 /*
1602  * Here we allocate a free block in the current cylinder group. It is assumed,
1603  * that acg contains the current cylinder group. As we may take a block from
1604  * somewhere in the file system we have to handle cluster summary here.
1605  */
1606 static ufs2_daddr_t
1607 alloc(void)
1608 {
1609         DBG_FUNC("alloc")
1610         ufs2_daddr_t    d, blkno;
1611         int     lcs1, lcs2;
1612         int     l;
1613         int     csmin, csmax;
1614         int     dlower, dupper, dmax;
1615
1616         DBG_ENTER;
1617
1618         if (acg.cg_magic != CG_MAGIC) {
1619                 warnx("acg: bad magic number");
1620                 DBG_LEAVE;
1621                 return (0);
1622         }
1623         if (acg.cg_cs.cs_nbfree == 0) {
1624                 warnx("error: cylinder group ran out of space");
1625                 DBG_LEAVE;
1626                 return (0);
1627         }
1628         /*
1629          * We start seeking for free blocks only from the space available after
1630          * the end of the new grown cylinder summary. Otherwise we allocate a
1631          * block here which we have to relocate a couple of seconds later again
1632          * again, and we are not prepared to to this anyway.
1633          */
1634         blkno=-1;
1635         dlower=cgsblock(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1636         dupper=cgdmin(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1637         dmax=cgbase(&sblock, acg.cg_cgx)+sblock.fs_fpg;
1638         if (dmax > sblock.fs_size) {
1639                 dmax = sblock.fs_size;
1640         }
1641         dmax-=cgbase(&sblock, acg.cg_cgx); /* retransform into cg */
1642         csmin=sblock.fs_csaddr-cgbase(&sblock, acg.cg_cgx);
1643         csmax=csmin+howmany(sblock.fs_cssize, sblock.fs_fsize);
1644         DBG_PRINT3("seek range: dl=%d, du=%d, dm=%d\n",
1645             dlower,
1646             dupper,
1647             dmax);
1648         DBG_PRINT2("range cont: csmin=%d, csmax=%d\n",
1649             csmin,
1650             csmax);
1651
1652         for(d=0; (d<dlower && blkno==-1); d+=sblock.fs_frag) {
1653                 if(d>=csmin && d<=csmax) {
1654                         continue;
1655                 }
1656                 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1657                     d))) {
1658                         blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1659                         break;
1660                 }
1661         }
1662         for(d=dupper; (d<dmax && blkno==-1); d+=sblock.fs_frag) {
1663                 if(d>=csmin && d<=csmax) {
1664                         continue;
1665                 }
1666                 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1667                     d))) {
1668                         blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1669                         break;
1670                 }
1671         }
1672         if(blkno==-1) {
1673                 warnx("internal error: couldn't find promised block in cg");
1674                 DBG_LEAVE;
1675                 return (0);
1676         }
1677
1678         /*
1679          * This is needed if the block was found already in the first loop.
1680          */
1681         d=blkstofrags(&sblock, blkno);
1682
1683         clrblock(&sblock, cg_blksfree(&acg), blkno);
1684         if (sblock.fs_contigsumsize > 0) {
1685                 /*
1686                  * Handle the cluster allocation bitmap.
1687                  */
1688                 clrbit(cg_clustersfree(&acg), blkno);
1689                 /*
1690                  * We possibly have split a cluster here, so we have to do
1691                  * recalculate the sizes of the remaining cluster halves now,
1692                  * and use them for updating the cluster summary information.
1693                  *
1694                  * Lets start with the blocks before our allocated block ...
1695                  */
1696                 for(lcs1=0, l=blkno-1; lcs1<sblock.fs_contigsumsize;
1697                     l--, lcs1++ ) {
1698                         if(isclr(cg_clustersfree(&acg),l)){
1699                                 break;
1700                         }
1701                 }
1702                 /*
1703                  * ... and continue with the blocks right after our allocated
1704                  * block.
1705                  */
1706                 for(lcs2=0, l=blkno+1; lcs2<sblock.fs_contigsumsize;
1707                     l++, lcs2++ ) {
1708                         if(isclr(cg_clustersfree(&acg),l)){
1709                                 break;
1710                         }
1711                 }
1712
1713                 /*
1714                  * Now update all counters.
1715                  */
1716                 cg_clustersum(&acg)[MIN(lcs1+lcs2+1,sblock.fs_contigsumsize)]--;
1717                 if(lcs1) {
1718                         cg_clustersum(&acg)[lcs1]++;
1719                 }
1720                 if(lcs2) {
1721                         cg_clustersum(&acg)[lcs2]++;
1722                 }
1723         }
1724         /*
1725          * Update all statistics based on blocks.
1726          */
1727         acg.cg_cs.cs_nbfree--;
1728         sblock.fs_cstotal.cs_nbfree--;
1729
1730         DBG_LEAVE;
1731         return (d);
1732 }
1733
1734 /* *********************************************************** isblock ***** */
1735 /*
1736  * Here we check if all frags of a block are free. For more details again
1737  * please see the source of newfs(8), as this function is taken over almost
1738  * unchanged.
1739  */
1740 static int
1741 isblock(struct fs *fs, unsigned char *cp, int h)
1742 {
1743         DBG_FUNC("isblock")
1744         unsigned char   mask;
1745
1746         DBG_ENTER;
1747
1748         switch (fs->fs_frag) {
1749         case 8:
1750                 DBG_LEAVE;
1751                 return (cp[h] == 0xff);
1752         case 4:
1753                 mask = 0x0f << ((h & 0x1) << 2);
1754                 DBG_LEAVE;
1755                 return ((cp[h >> 1] & mask) == mask);
1756         case 2:
1757                 mask = 0x03 << ((h & 0x3) << 1);
1758                 DBG_LEAVE;
1759                 return ((cp[h >> 2] & mask) == mask);
1760         case 1:
1761                 mask = 0x01 << (h & 0x7);
1762                 DBG_LEAVE;
1763                 return ((cp[h >> 3] & mask) == mask);
1764         default:
1765                 fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
1766                 DBG_LEAVE;
1767                 return (0);
1768         }
1769 }
1770
1771 /* ********************************************************** clrblock ***** */
1772 /*
1773  * Here we allocate a complete block in the block map. For more details again
1774  * please see the source of newfs(8), as this function is taken over almost
1775  * unchanged.
1776  */
1777 static void
1778 clrblock(struct fs *fs, unsigned char *cp, int h)
1779 {
1780         DBG_FUNC("clrblock")
1781
1782         DBG_ENTER;
1783
1784         switch ((fs)->fs_frag) {
1785         case 8:
1786                 cp[h] = 0;
1787                 break;
1788         case 4:
1789                 cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1790                 break;
1791         case 2:
1792                 cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1793                 break;
1794         case 1:
1795                 cp[h >> 3] &= ~(0x01 << (h & 0x7));
1796                 break;
1797         default:
1798                 warnx("clrblock bad fs_frag %d", fs->fs_frag);
1799                 break;
1800         }
1801
1802         DBG_LEAVE;
1803         return;
1804 }
1805
1806 /* ********************************************************** setblock ***** */
1807 /*
1808  * Here we free a complete block in the free block map. For more details again
1809  * please see the source of newfs(8), as this function is taken over almost
1810  * unchanged.
1811  */
1812 static void
1813 setblock(struct fs *fs, unsigned char *cp, int h)
1814 {
1815         DBG_FUNC("setblock")
1816
1817         DBG_ENTER;
1818
1819         switch (fs->fs_frag) {
1820         case 8:
1821                 cp[h] = 0xff;
1822                 break;
1823         case 4:
1824                 cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1825                 break;
1826         case 2:
1827                 cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1828                 break;
1829         case 1:
1830                 cp[h >> 3] |= (0x01 << (h & 0x7));
1831                 break;
1832         default:
1833                 warnx("setblock bad fs_frag %d", fs->fs_frag);
1834                 break;
1835         }
1836
1837         DBG_LEAVE;
1838         return;
1839 }
1840
1841 /* ************************************************************ ginode ***** */
1842 /*
1843  * This function provides access to an individual inode. We find out in which
1844  * block the requested inode is located, read it from disk if needed, and
1845  * return the pointer into that block. We maintain a cache of one block to
1846  * not read the same block again and again if we iterate linearly over all
1847  * inodes.
1848  */
1849 static union dinode *
1850 ginode(ino_t inumber, int fsi, int cg)
1851 {
1852         DBG_FUNC("ginode")
1853         static ino_t    startinum = 0;  /* first inode in cached block */
1854
1855         DBG_ENTER;
1856
1857         /*
1858          * The inumber passed in is relative to the cg, so use it here to see
1859          * if the inode has been allocated yet.
1860          */
1861         if (isclr(cg_inosused(&aocg), inumber)) {
1862                 DBG_LEAVE;
1863                 return NULL;
1864         }
1865         /*
1866          * Now make the inumber relative to the entire inode space so it can
1867          * be sanity checked.
1868          */
1869         inumber += (cg * sblock.fs_ipg);
1870         if (inumber < ROOTINO) {
1871                 DBG_LEAVE;
1872                 return NULL;
1873         }
1874         if (inumber > maxino)
1875                 errx(8, "bad inode number %d to ginode", inumber);
1876         if (startinum == 0 ||
1877             inumber < startinum || inumber >= startinum + INOPB(&sblock)) {
1878                 inoblk = fsbtodb(&sblock, ino_to_fsba(&sblock, inumber));
1879                 rdfs(inoblk, (size_t)sblock.fs_bsize, inobuf, fsi);
1880                 startinum = (inumber / INOPB(&sblock)) * INOPB(&sblock);
1881         }
1882         DBG_LEAVE;
1883         if (sblock.fs_magic == FS_UFS1_MAGIC)
1884                 return (union dinode *)((uintptr_t)inobuf +
1885                     (inumber % INOPB(&sblock)) * sizeof(struct ufs1_dinode));
1886         return (union dinode *)((uintptr_t)inobuf +
1887             (inumber % INOPB(&sblock)) * sizeof(struct ufs2_dinode));
1888 }
1889
1890 /* ****************************************************** charsperline ***** */
1891 /*
1892  * Figure out how many lines our current terminal has. For more details again
1893  * please see the source of newfs(8), as this function is taken over almost
1894  * unchanged.
1895  */
1896 static int
1897 charsperline(void)
1898 {
1899         DBG_FUNC("charsperline")
1900         int     columns;
1901         char    *cp;
1902         struct winsize  ws;
1903
1904         DBG_ENTER;
1905
1906         columns = 0;
1907         if (ioctl(0, TIOCGWINSZ, &ws) != -1) {
1908                 columns = ws.ws_col;
1909         }
1910         if (columns == 0 && (cp = getenv("COLUMNS"))) {
1911                 columns = atoi(cp);
1912         }
1913         if (columns == 0) {
1914                 columns = 80;   /* last resort */
1915         }
1916
1917         DBG_LEAVE;
1918         return columns;
1919 }
1920
1921 /* ****************************************************** get_dev_size ***** */
1922 /*
1923  * Get the size of the partition if we can't figure it out from the disklabel,
1924  * e.g. from vinum volumes.
1925  */
1926 static void
1927 get_dev_size(int fd, int *size)
1928 {
1929    int sectorsize;
1930    off_t mediasize;
1931
1932    if (ioctl(fd, DIOCGSECTORSIZE, &sectorsize) == -1)
1933         err(1,"DIOCGSECTORSIZE");
1934    if (ioctl(fd, DIOCGMEDIASIZE, &mediasize) == -1)
1935         err(1,"DIOCGMEDIASIZE");
1936
1937    if (sectorsize <= 0)
1938        errx(1, "bogus sectorsize: %d", sectorsize);
1939
1940    *size = mediasize / sectorsize;
1941 }
1942
1943 /* ************************************************************** main ***** */
1944 /*
1945  * growfs(8)  is a utility which allows to increase the size of an existing
1946  * ufs file system. Currently this can only be done on unmounted file system.
1947  * It recognizes some command line options to specify the new desired size,
1948  * and it does some basic checkings. The old file system size is determined
1949  * and after some more checks like we can really access the new last block
1950  * on the disk etc. we calculate the new parameters for the superblock. After
1951  * having done this we just call growfs() which will do the work.  Before
1952  * we finish the only thing left is to update the disklabel.
1953  * We still have to provide support for snapshots. Therefore we first have to
1954  * understand what data structures are always replicated in the snapshot on
1955  * creation, for all other blocks we touch during our procedure, we have to
1956  * keep the old blocks unchanged somewhere available for the snapshots. If we
1957  * are lucky, then we only have to handle our blocks to be relocated in that
1958  * way.
1959  * Also we have to consider in what order we actually update the critical
1960  * data structures of the file system to make sure, that in case of a disaster
1961  * fsck(8) is still able to restore any lost data.
1962  * The foreseen last step then will be to provide for growing even mounted
1963  * file systems. There we have to extend the mount() system call to provide
1964  * userland access to the file system locking facility.
1965  */
1966 int
1967 main(int argc, char **argv)
1968 {
1969         DBG_FUNC("main")
1970         char    *device, *special, *cp;
1971         int     ch;
1972         unsigned int    size=0;
1973         size_t  len;
1974         unsigned int    Nflag=0;
1975         int     ExpertFlag=0;
1976         struct stat     st;
1977         struct disklabel        *lp;
1978         struct partition        *pp;
1979         int     i,fsi,fso;
1980     u_int32_t p_size;
1981         char    reply[5];
1982 #ifdef FSMAXSNAP
1983         int     j;
1984 #endif /* FSMAXSNAP */
1985
1986         DBG_ENTER;
1987
1988         while((ch=getopt(argc, argv, "Ns:vy")) != -1) {
1989                 switch(ch) {
1990                 case 'N':
1991                         Nflag=1;
1992                         break;
1993                 case 's':
1994                         size=(size_t)atol(optarg);
1995                         if(size<1) {
1996                                 usage();
1997                         }
1998                         break;
1999                 case 'v': /* for compatibility to newfs */
2000                         break;
2001                 case 'y':
2002                         ExpertFlag=1;
2003                         break;
2004                 case '?':
2005                         /* FALLTHROUGH */
2006                 default:
2007                         usage();
2008                 }
2009         }
2010         argc -= optind;
2011         argv += optind;
2012
2013         if(argc != 1) {
2014                 usage();
2015         }
2016         device=*argv;
2017
2018         /*
2019          * Now try to guess the (raw)device name.
2020          */
2021         if (0 == strrchr(device, '/')) {
2022                 /*
2023                  * No path prefix was given, so try in that order:
2024                  *     /dev/r%s
2025                  *     /dev/%s
2026                  *     /dev/vinum/r%s
2027                  *     /dev/vinum/%s.
2028                  *
2029                  * FreeBSD now doesn't distinguish between raw and block
2030                  * devices any longer, but it should still work this way.
2031                  */
2032                 len=strlen(device)+strlen(_PATH_DEV)+2+strlen("vinum/");
2033                 special=(char *)malloc(len);
2034                 if(special == NULL) {
2035                         errx(1, "malloc failed");
2036                 }
2037                 snprintf(special, len, "%sr%s", _PATH_DEV, device);
2038                 if (stat(special, &st) == -1) {
2039                         snprintf(special, len, "%s%s", _PATH_DEV, device);
2040                         if (stat(special, &st) == -1) {
2041                                 snprintf(special, len, "%svinum/r%s",
2042                                     _PATH_DEV, device);
2043                                 if (stat(special, &st) == -1) {
2044                                         /* For now this is the 'last resort' */
2045                                         snprintf(special, len, "%svinum/%s",
2046                                             _PATH_DEV, device);
2047                                 }
2048                         }
2049                 }
2050                 device = special;
2051         }
2052
2053         /*
2054          * Try to access our devices for writing ...
2055          */
2056         if (Nflag) {
2057                 fso = -1;
2058         } else {
2059                 fso = open(device, O_WRONLY);
2060                 if (fso < 0) {
2061                         err(1, "%s", device);
2062                 }
2063         }
2064
2065         /*
2066          * ... and reading.
2067          */
2068         fsi = open(device, O_RDONLY);
2069         if (fsi < 0) {
2070                 err(1, "%s", device);
2071         }
2072
2073         /*
2074          * Try to read a label and guess the slice if not specified. This
2075          * code should guess the right thing and avoid to bother the user
2076          * with the task of specifying the option -v on vinum volumes.
2077          */
2078         cp=device+strlen(device)-1;
2079         lp = get_disklabel(fsi);
2080         pp = NULL;
2081     if (lp != NULL) {
2082         if (isdigit(*cp)) {
2083             pp = &lp->d_partitions[2];
2084         } else if (*cp>='a' && *cp<='h') {
2085             pp = &lp->d_partitions[*cp - 'a'];
2086         } else {
2087             errx(1, "unknown device");
2088         }
2089         p_size = pp->p_size;
2090     } else {
2091         get_dev_size(fsi, &p_size);
2092     }
2093
2094         /*
2095          * Check if that partition is suitable for growing a file system.
2096          */
2097         if (p_size < 1) {
2098                 errx(1, "partition is unavailable");
2099         }
2100
2101         /*
2102          * Read the current superblock, and take a backup.
2103          */
2104         for (i = 0; sblock_try[i] != -1; i++) {
2105                 sblockloc = sblock_try[i] / DEV_BSIZE;
2106                 rdfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&(osblock), fsi);
2107                 if ((osblock.fs_magic == FS_UFS1_MAGIC ||
2108                      (osblock.fs_magic == FS_UFS2_MAGIC &&
2109                       osblock.fs_sblockloc == sblock_try[i])) &&
2110                     osblock.fs_bsize <= MAXBSIZE &&
2111                     osblock.fs_bsize >= (int32_t) sizeof(struct fs))
2112                         break;
2113         }
2114         if (sblock_try[i] == -1) {
2115                 errx(1, "superblock not recognized");
2116         }
2117         memcpy((void *)&fsun1, (void *)&fsun2, sizeof(fsun2));
2118         maxino = sblock.fs_ncg * sblock.fs_ipg;
2119
2120         DBG_OPEN("/tmp/growfs.debug"); /* already here we need a superblock */
2121         DBG_DUMP_FS(&sblock,
2122             "old sblock");
2123
2124         /*
2125          * Determine size to grow to. Default to the full size specified in
2126          * the disk label.
2127          */
2128         sblock.fs_size = dbtofsb(&osblock, p_size);
2129         if (size != 0) {
2130                 if (size > p_size){
2131                         errx(1, "there is not enough space (%d < %d)",
2132                             p_size, size);
2133                 }
2134                 sblock.fs_size = dbtofsb(&osblock, size);
2135         }
2136
2137         /*
2138          * Are we really growing ?
2139          */
2140         if(osblock.fs_size >= sblock.fs_size) {
2141                 errx(1, "we are not growing (%jd->%jd)",
2142                     (intmax_t)osblock.fs_size, (intmax_t)sblock.fs_size);
2143         }
2144
2145
2146 #ifdef FSMAXSNAP
2147         /*
2148          * Check if we find an active snapshot.
2149          */
2150         if(ExpertFlag == 0) {
2151                 for(j=0; j<FSMAXSNAP; j++) {
2152                         if(sblock.fs_snapinum[j]) {
2153                                 errx(1, "active snapshot found in file system\n"
2154                                     "   please remove all snapshots before "
2155                                     "using growfs");
2156                         }
2157                         if(!sblock.fs_snapinum[j]) { /* list is dense */
2158                                 break;
2159                         }
2160                 }
2161         }
2162 #endif
2163
2164         if (ExpertFlag == 0 && Nflag == 0) {
2165                 printf("We strongly recommend you to make a backup "
2166                     "before growing the Filesystem\n\n"
2167                     " Did you backup your data (Yes/No) ? ");
2168                 fgets(reply, (int)sizeof(reply), stdin);
2169                 if (strcmp(reply, "Yes\n")){
2170                         printf("\n Nothing done \n");
2171                         exit (0);
2172                 }
2173         }
2174
2175         printf("new file systemsize is: %jd frags\n", (intmax_t)sblock.fs_size);
2176
2177         /*
2178          * Try to access our new last block in the file system. Even if we
2179          * later on realize we have to abort our operation, on that block
2180          * there should be no data, so we can't destroy something yet.
2181          */
2182         wtfs((ufs2_daddr_t)p_size-1, (size_t)DEV_BSIZE, (void *)&sblock,
2183             fso, Nflag);
2184
2185         /*
2186          * Now calculate new superblock values and check for reasonable
2187          * bound for new file system size:
2188          *     fs_size:    is derived from label or user input
2189          *     fs_dsize:   should get updated in the routines creating or
2190          *                 updating the cylinder groups on the fly
2191          *     fs_cstotal: should get updated in the routines creating or
2192          *                 updating the cylinder groups
2193          */
2194
2195         /*
2196          * Update the number of cylinders and cylinder groups in the file system.
2197          */
2198         if (sblock.fs_magic == FS_UFS1_MAGIC) {
2199                 sblock.fs_old_ncyl =
2200                     sblock.fs_size * sblock.fs_old_nspf / sblock.fs_old_spc;
2201                 if (sblock.fs_size * sblock.fs_old_nspf >
2202                     sblock.fs_old_ncyl * sblock.fs_old_spc)
2203                         sblock.fs_old_ncyl++;
2204         }
2205         sblock.fs_ncg = howmany(sblock.fs_size, sblock.fs_fpg);
2206         maxino = sblock.fs_ncg * sblock.fs_ipg;
2207
2208         if (sblock.fs_size % sblock.fs_fpg != 0 &&
2209             sblock.fs_size % sblock.fs_fpg < cgdmin(&sblock, sblock.fs_ncg)) {
2210                 /*
2211                  * The space in the new last cylinder group is too small,
2212                  * so revert back.
2213                  */
2214                 sblock.fs_ncg--;
2215                 if (sblock.fs_magic == FS_UFS1_MAGIC)
2216                         sblock.fs_old_ncyl = sblock.fs_ncg * sblock.fs_old_cpg;
2217                 printf("Warning: %jd sector(s) cannot be allocated.\n",
2218                     (intmax_t)fsbtodb(&sblock, sblock.fs_size % sblock.fs_fpg));
2219                 sblock.fs_size = sblock.fs_ncg * sblock.fs_fpg;
2220         }
2221
2222         /*
2223          * Update the space for the cylinder group summary information in the
2224          * respective cylinder group data area.
2225          */
2226         sblock.fs_cssize =
2227             fragroundup(&sblock, sblock.fs_ncg * sizeof(struct csum));
2228
2229         if(osblock.fs_size >= sblock.fs_size) {
2230                 errx(1, "not enough new space");
2231         }
2232
2233         DBG_PRINT0("sblock calculated\n");
2234
2235         /*
2236          * Ok, everything prepared, so now let's do the tricks.
2237          */
2238         growfs(fsi, fso, Nflag);
2239
2240         /*
2241          * Update the disk label.
2242          */
2243     if (!unlabeled) {
2244         pp->p_fsize = sblock.fs_fsize;
2245         pp->p_frag = sblock.fs_frag;
2246         pp->p_cpg = sblock.fs_fpg;
2247
2248         return_disklabel(fso, lp, Nflag);
2249         DBG_PRINT0("label rewritten\n");
2250     }
2251
2252         close(fsi);
2253         if(fso>-1) close(fso);
2254
2255         DBG_CLOSE;
2256
2257         DBG_LEAVE;
2258         return 0;
2259 }
2260
2261 /* ************************************************** return_disklabel ***** */
2262 /*
2263  * Write the updated disklabel back to disk.
2264  */
2265 static void
2266 return_disklabel(int fd, struct disklabel *lp, unsigned int Nflag)
2267 {
2268         DBG_FUNC("return_disklabel")
2269         u_short sum;
2270         u_short *ptr;
2271
2272         DBG_ENTER;
2273
2274         if(!lp) {
2275                 DBG_LEAVE;
2276                 return;
2277         }
2278         if(!Nflag) {
2279                 lp->d_checksum=0;
2280                 sum = 0;
2281                 ptr=(u_short *)lp;
2282
2283                 /*
2284                  * recalculate checksum
2285                  */
2286                 while(ptr < (u_short *)&lp->d_partitions[lp->d_npartitions]) {
2287                         sum ^= *ptr++;
2288                 }
2289                 lp->d_checksum=sum;
2290
2291                 if (ioctl(fd, DIOCWDINFO, (char *)lp) < 0) {
2292                         errx(1, "DIOCWDINFO failed");
2293                 }
2294         }
2295         free(lp);
2296
2297         DBG_LEAVE;
2298         return ;
2299 }
2300
2301 /* ***************************************************** get_disklabel ***** */
2302 /*
2303  * Read the disklabel from disk.
2304  */
2305 static struct disklabel *
2306 get_disklabel(int fd)
2307 {
2308         DBG_FUNC("get_disklabel")
2309         static struct   disklabel *lab;
2310
2311         DBG_ENTER;
2312
2313         lab=(struct disklabel *)malloc(sizeof(struct disklabel));
2314         if (!lab)
2315                 errx(1, "malloc failed");
2316
2317     if (!ioctl(fd, DIOCGDINFO, (char *)lab))
2318         return (lab);
2319
2320     unlabeled++;
2321
2322         DBG_LEAVE;
2323         return (NULL);
2324 }
2325
2326
2327 /* ************************************************************* usage ***** */
2328 /*
2329  * Dump a line of usage.
2330  */
2331 static void
2332 usage(void)
2333 {
2334         DBG_FUNC("usage")
2335
2336         DBG_ENTER;
2337
2338         fprintf(stderr, "usage: growfs [-Ny] [-s size] special\n");
2339
2340         DBG_LEAVE;
2341         exit(1);
2342 }
2343
2344 /* *********************************************************** updclst ***** */
2345 /*
2346  * This updates most parameters and the bitmap related to cluster. We have to
2347  * assume that sblock, osblock, acg are set up.
2348  */
2349 static void
2350 updclst(int block)
2351 {
2352         DBG_FUNC("updclst")
2353         static int      lcs=0;
2354
2355         DBG_ENTER;
2356
2357         if(sblock.fs_contigsumsize < 1) { /* no clustering */
2358                 return;
2359         }
2360         /*
2361          * update cluster allocation map
2362          */
2363         setbit(cg_clustersfree(&acg), block);
2364
2365         /*
2366          * update cluster summary table
2367          */
2368         if(!lcs) {
2369                 /*
2370                  * calculate size for the trailing cluster
2371                  */
2372                 for(block--; lcs<sblock.fs_contigsumsize; block--, lcs++ ) {
2373                         if(isclr(cg_clustersfree(&acg), block)){
2374                                 break;
2375                         }
2376                 }
2377         }
2378         if(lcs < sblock.fs_contigsumsize) {
2379                 if(lcs) {
2380                         cg_clustersum(&acg)[lcs]--;
2381                 }
2382                 lcs++;
2383                 cg_clustersum(&acg)[lcs]++;
2384         }
2385
2386         DBG_LEAVE;
2387         return;
2388 }
2389
2390 /* *********************************************************** updrefs ***** */
2391 /*
2392  * This updates all references to relocated blocks for the given inode.  The
2393  * inode is given as number within the cylinder group, and the number of the
2394  * cylinder group.
2395  */
2396 static void
2397 updrefs(int cg, ino_t in, struct gfs_bpp *bp, int fsi, int fso, unsigned int
2398     Nflag)
2399 {
2400         DBG_FUNC("updrefs")
2401         ufs_lbn_t       len, lbn, numblks;
2402         ufs2_daddr_t    iptr, blksperindir;
2403         union dinode    *ino;
2404         int             i, mode, inodeupdated;
2405
2406         DBG_ENTER;
2407
2408         ino = ginode(in, fsi, cg);
2409         if (ino == NULL) {
2410                 DBG_LEAVE;
2411                 return;
2412         }
2413         mode = DIP(ino, di_mode) & IFMT;
2414         if (mode != IFDIR && mode != IFREG && mode != IFLNK) {
2415                 DBG_LEAVE;
2416                 return; /* only check DIR, FILE, LINK */
2417         }
2418         if (mode == IFLNK && 
2419             DIP(ino, di_size) < (u_int64_t) sblock.fs_maxsymlinklen) {
2420                 DBG_LEAVE;
2421                 return; /* skip short symlinks */
2422         }
2423         numblks = howmany(DIP(ino, di_size), sblock.fs_bsize);
2424         if (numblks == 0) {
2425                 DBG_LEAVE;
2426                 return; /* skip empty file */
2427         }
2428         if (DIP(ino, di_blocks) == 0) {
2429                 DBG_LEAVE;
2430                 return; /* skip empty swiss cheesy file or old fastlink */
2431         }
2432         DBG_PRINT2("scg checking inode (%d in %d)\n",
2433             in,
2434             cg);
2435
2436         /*
2437          * Check all the blocks.
2438          */
2439         inodeupdated = 0;
2440         len = numblks < NDADDR ? numblks : NDADDR;
2441         for (i = 0; i < len; i++) {
2442                 iptr = DIP(ino, di_db[i]);
2443                 if (iptr == 0)
2444                         continue;
2445                 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2446                         DIP_SET(ino, di_db[i], iptr);
2447                         inodeupdated++;
2448                 }
2449         }
2450         DBG_PRINT0("~~scg direct blocks checked\n");
2451
2452         blksperindir = 1;
2453         len = numblks - NDADDR;
2454         lbn = NDADDR;
2455         for (i = 0; len > 0 && i < NIADDR; i++) {
2456                 iptr = DIP(ino, di_ib[i]);
2457                 if (iptr == 0)
2458                         continue;
2459                 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2460                         DIP_SET(ino, di_ib[i], iptr);
2461                         inodeupdated++;
2462                 }
2463                 indirchk(blksperindir, lbn, iptr, numblks, bp, fsi, fso, Nflag);
2464                 blksperindir *= NINDIR(&sblock);
2465                 lbn += blksperindir;
2466                 len -= blksperindir;
2467                 DBG_PRINT1("scg indirect_%d blocks checked\n", i + 1);
2468         }
2469         if (inodeupdated)
2470                 wtfs(inoblk, sblock.fs_bsize, inobuf, fso, Nflag);
2471
2472         DBG_LEAVE;
2473         return;
2474 }
2475
2476 /*
2477  * Recursively check all the indirect blocks.
2478  */
2479 static void
2480 indirchk(ufs_lbn_t blksperindir, ufs_lbn_t lbn, ufs2_daddr_t blkno,
2481     ufs_lbn_t lastlbn, struct gfs_bpp *bp, int fsi, int fso, unsigned int Nflag)
2482 {
2483         DBG_FUNC("indirchk")
2484         void *ibuf;
2485         int i, last;
2486         ufs2_daddr_t iptr;
2487
2488         DBG_ENTER;
2489
2490         /* read in the indirect block. */
2491         ibuf = malloc(sblock.fs_bsize);
2492         if (!ibuf)
2493                 errx(1, "malloc failed");
2494         rdfs(fsbtodb(&sblock, blkno), (size_t)sblock.fs_bsize, ibuf, fsi);
2495         last = howmany(lastlbn - lbn, blksperindir) < NINDIR(&sblock) ?
2496             howmany(lastlbn - lbn, blksperindir) : NINDIR(&sblock);
2497         for (i = 0; i < last; i++) {
2498                 if (sblock.fs_magic == FS_UFS1_MAGIC)
2499                         iptr = ((ufs1_daddr_t *)ibuf)[i];
2500                 else
2501                         iptr = ((ufs2_daddr_t *)ibuf)[i];
2502                 if (iptr == 0)
2503                         continue;
2504                 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2505                         if (sblock.fs_magic == FS_UFS1_MAGIC)
2506                                 ((ufs1_daddr_t *)ibuf)[i] = iptr;
2507                         else
2508                                 ((ufs2_daddr_t *)ibuf)[i] = iptr;
2509                 }
2510                 if (blksperindir == 1)
2511                         continue;
2512                 indirchk(blksperindir / NINDIR(&sblock), lbn + blksperindir * i,
2513                     iptr, lastlbn, bp, fsi, fso, Nflag);
2514         }
2515         free(ibuf);
2516
2517         DBG_LEAVE;
2518         return;
2519 }