]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sbin/growfs/growfs.c
This commit was generated by cvs2svn to compensate for changes in r157016,
[FreeBSD/FreeBSD.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, " %d%s",
263                     (int)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         f = field;
635         for (f = field; f->old != 0; f++) {
636                 src = *block;
637                 if (fragstoblks(&sblock, src) != f->old)
638                         continue;
639                 /*
640                  * The fragment is part of the block, so update.
641                  */
642                 dst = blkstofrags(&sblock, f->new);
643                 fragnum = fragnum(&sblock, src);
644                 *block = dst + fragnum;
645                 f->found++;
646                 DBG_PRINT3("scg (%jd->%jd)[%d] reference updated\n",
647                     (intmax_t)f->old,
648                     (intmax_t)f->new,
649                     fragnum);
650
651                 /*
652                  * Copy the block back immediately.
653                  *
654                  * XXX  If src is is from an indirect block we have
655                  *      to implement copy on write here in case of
656                  *      active snapshots.
657                  */
658                 ibuf = malloc(sblock.fs_bsize);
659                 if (!ibuf)
660                         errx(1, "malloc failed");
661                 src -= fragnum;
662                 rdfs(fsbtodb(&sblock, src), (size_t)sblock.fs_bsize, ibuf, fsi);
663                 wtfs(dst, (size_t)sblock.fs_bsize, ibuf, fso, Nflag);
664                 free(ibuf);
665                 /*
666                  * The same block can't be found again in this loop.
667                  */
668                 return (1);
669         }
670
671         DBG_LEAVE;
672         return (0);
673 }
674
675 /* ************************************************************ updjcg ***** */
676 /*
677  * Here we do all needed work for the former last cylinder group. It has to be
678  * changed in any case, even if the file system ended exactly on the end of
679  * this group, as there is some slightly inconsistent handling of the number
680  * of cylinders in the cylinder group. We start again by reading the cylinder
681  * group from disk. If the last block was not fully available, we first handle
682  * the missing fragments, then we handle all new full blocks in that file
683  * system and finally we handle the new last fragmented block in the file
684  * system.  We again have to handle the fragment statistics rotational layout
685  * tables and cluster summary during all those operations.
686  */
687 static void
688 updjcg(int cylno, time_t utime, int fsi, int fso, unsigned int Nflag)
689 {
690         DBG_FUNC("updjcg")
691         ufs2_daddr_t    cbase, dmax, dupper;
692         struct csum     *cs;
693         int     i,k;
694         int     j=0;
695
696         DBG_ENTER;
697
698         /*
699          * Read the former last (joining) cylinder group from disk, and make
700          * a copy.
701          */
702         rdfs(fsbtodb(&osblock, cgtod(&osblock, cylno)),
703             (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
704         DBG_PRINT0("jcg read\n");
705         DBG_DUMP_CG(&sblock,
706             "old joining cg",
707             &aocg);
708
709         memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
710
711         /*
712          * If the cylinder group had already its new final size almost
713          * nothing is to be done ... except:
714          * For some reason the value of cg_ncyl in the last cylinder group has
715          * to be zero instead of fs_cpg. As this is now no longer the last
716          * cylinder group we have to change that value now to fs_cpg.
717          */
718
719         if(cgbase(&osblock, cylno+1) == osblock.fs_size) {
720                 if (sblock.fs_magic == FS_UFS1_MAGIC)
721                         acg.cg_old_ncyl=sblock.fs_old_cpg;
722
723                 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
724                     (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
725                 DBG_PRINT0("jcg written\n");
726                 DBG_DUMP_CG(&sblock,
727                     "new joining cg",
728                     &acg);
729
730                 DBG_LEAVE;
731                 return;
732         }
733
734         /*
735          * Set up some variables needed later.
736          */
737         cbase = cgbase(&sblock, cylno);
738         dmax = cbase + sblock.fs_fpg;
739         if (dmax > sblock.fs_size)
740                 dmax = sblock.fs_size;
741         dupper = cgdmin(&sblock, cylno) - cbase;
742         if (cylno == 0) { /* XXX fscs may be relocated */
743                 dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
744         }
745
746         /*
747          * Set pointer to the cylinder summary for our cylinder group.
748          */
749         cs = fscs + cylno;
750
751         /*
752          * Touch the cylinder group, update all fields in the cylinder group as
753          * needed, update the free space in the superblock.
754          */
755         acg.cg_time = utime;
756         if (cylno == sblock.fs_ncg - 1) {
757                 /*
758                  * This is still the last cylinder group.
759                  */
760                 if (sblock.fs_magic == FS_UFS1_MAGIC)
761                         acg.cg_old_ncyl =
762                             sblock.fs_old_ncyl % sblock.fs_old_cpg;
763         } else {
764                 acg.cg_old_ncyl = sblock.fs_old_cpg;
765         }
766         DBG_PRINT2("jcg dbg: %d %u",
767             cylno,
768             sblock.fs_ncg);
769 #ifdef FS_DEBUG
770         if (sblock.fs_magic == FS_UFS1_MAGIC)
771                 DBG_PRINT2("%d %u",
772                     acg.cg_old_ncyl,
773                     sblock.fs_old_cpg);
774 #endif
775         DBG_PRINT0("\n");
776         acg.cg_ndblk = dmax - cbase;
777         sblock.fs_dsize += acg.cg_ndblk-aocg.cg_ndblk;
778         if (sblock.fs_contigsumsize > 0) {
779                 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
780         }
781
782         /*
783          * Now we have to update the free fragment bitmap for our new free
784          * space.  There again we have to handle the fragmentation and also
785          * the rotational layout tables and the cluster summary.  This is
786          * also done per fragment for the first new block if the old file
787          * system end was not on a block boundary, per fragment for the new
788          * last block if the new file system end is not on a block boundary,
789          * and per block for all space in between.
790          *
791          * Handle the first new block here if it was partially available
792          * before.
793          */
794         if(osblock.fs_size % sblock.fs_frag) {
795                 if(roundup(osblock.fs_size, sblock.fs_frag)<=sblock.fs_size) {
796                         /*
797                          * The new space is enough to fill at least this
798                          * block
799                          */
800                         j=0;
801                         for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag)-1;
802                             i>=osblock.fs_size-cbase;
803                             i--) {
804                                 setbit(cg_blksfree(&acg), i);
805                                 acg.cg_cs.cs_nffree++;
806                                 j++;
807                         }
808
809                         /*
810                          * Check if the fragment just created could join an
811                          * already existing fragment at the former end of the
812                          * file system.
813                          */
814                         if(isblock(&sblock, cg_blksfree(&acg),
815                             ((osblock.fs_size - cgbase(&sblock, cylno))/
816                             sblock.fs_frag))) {
817                                 /*
818                                  * The block is now completely available.
819                                  */
820                                 DBG_PRINT0("block was\n");
821                                 acg.cg_frsum[osblock.fs_size%sblock.fs_frag]--;
822                                 acg.cg_cs.cs_nbfree++;
823                                 acg.cg_cs.cs_nffree-=sblock.fs_frag;
824                                 k=rounddown(osblock.fs_size-cbase,
825                                     sblock.fs_frag);
826                                 updclst((osblock.fs_size-cbase)/sblock.fs_frag);
827                         } else {
828                                 /*
829                                  * Lets rejoin a possible partially growed
830                                  * fragment.
831                                  */
832                                 k=0;
833                                 while(isset(cg_blksfree(&acg), i) &&
834                                     (i>=rounddown(osblock.fs_size-cbase,
835                                     sblock.fs_frag))) {
836                                         i--;
837                                         k++;
838                                 }
839                                 if(k) {
840                                         acg.cg_frsum[k]--;
841                                 }
842                                 acg.cg_frsum[k+j]++;
843                         }
844                 } else {
845                         /*
846                          * We only grow by some fragments within this last
847                          * block.
848                          */
849                         for(i=sblock.fs_size-cbase-1;
850                                 i>=osblock.fs_size-cbase;
851                                 i--) {
852                                 setbit(cg_blksfree(&acg), i);
853                                 acg.cg_cs.cs_nffree++;
854                                 j++;
855                         }
856                         /*
857                          * Lets rejoin a possible partially growed fragment.
858                          */
859                         k=0;
860                         while(isset(cg_blksfree(&acg), i) &&
861                             (i>=rounddown(osblock.fs_size-cbase,
862                             sblock.fs_frag))) {
863                                 i--;
864                                 k++;
865                         }
866                         if(k) {
867                                 acg.cg_frsum[k]--;
868                         }
869                         acg.cg_frsum[k+j]++;
870                 }
871         }
872
873         /*
874          * Handle all new complete blocks here.
875          */
876         for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag);
877             i+sblock.fs_frag<=dmax-cbase;       /* XXX <= or only < ? */
878             i+=sblock.fs_frag) {
879                 j = i / sblock.fs_frag;
880                 setblock(&sblock, cg_blksfree(&acg), j);
881                 updclst(j);
882                 acg.cg_cs.cs_nbfree++;
883         }
884
885         /*
886          * Handle the last new block if there are stll some new fragments left.
887          * Here we don't have to bother about the cluster summary or the even
888          * the rotational layout table.
889          */
890         if (i < (dmax - cbase)) {
891                 acg.cg_frsum[dmax - cbase - i]++;
892                 for (; i < dmax - cbase; i++) {
893                         setbit(cg_blksfree(&acg), i);
894                         acg.cg_cs.cs_nffree++;
895                 }
896         }
897
898         sblock.fs_cstotal.cs_nffree +=
899             (acg.cg_cs.cs_nffree - aocg.cg_cs.cs_nffree);
900         sblock.fs_cstotal.cs_nbfree +=
901             (acg.cg_cs.cs_nbfree - aocg.cg_cs.cs_nbfree);
902         /*
903          * The following statistics are not changed here:
904          *     sblock.fs_cstotal.cs_ndir
905          *     sblock.fs_cstotal.cs_nifree
906          * As the statistics for this cylinder group are ready, copy it to
907          * the summary information array.
908          */
909         *cs = acg.cg_cs;
910
911         /*
912          * Write the updated "joining" cylinder group back to disk.
913          */
914         wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize,
915             (void *)&acg, fso, Nflag);
916         DBG_PRINT0("jcg written\n");
917         DBG_DUMP_CG(&sblock,
918             "new joining cg",
919             &acg);
920
921         DBG_LEAVE;
922         return;
923 }
924
925 /* ********************************************************** updcsloc ***** */
926 /*
927  * Here we update the location of the cylinder summary. We have two possible
928  * ways of growing the cylinder summary.
929  * (1)  We can try to grow the summary in the current location, and relocate
930  *      possibly used blocks within the current cylinder group.
931  * (2)  Alternatively we can relocate the whole cylinder summary to the first
932  *      new completely empty cylinder group. Once the cylinder summary is no
933  *      longer in the beginning of the first cylinder group you should never
934  *      use a version of fsck which is not aware of the possibility to have
935  *      this structure in a non standard place.
936  * Option (1) is considered to be less intrusive to the structure of the file-
937  * system. So we try to stick to that whenever possible. If there is not enough
938  * space in the cylinder group containing the cylinder summary we have to use
939  * method (2). In case of active snapshots in the file system we probably can
940  * completely avoid implementing copy on write if we stick to method (2) only.
941  */
942 static void
943 updcsloc(time_t utime, int fsi, int fso, unsigned int Nflag)
944 {
945         DBG_FUNC("updcsloc")
946         struct csum     *cs;
947         int     ocscg, ncscg;
948         int     blocks;
949         ufs2_daddr_t    cbase, dupper, odupper, d, f, g;
950         int     ind;
951         int     cylno, inc;
952         struct gfs_bpp  *bp;
953         int     i, l;
954         int     lcs=0;
955         int     block;
956
957         DBG_ENTER;
958
959         if(howmany(sblock.fs_cssize, sblock.fs_fsize) ==
960             howmany(osblock.fs_cssize, osblock.fs_fsize)) {
961                 /*
962                  * No new fragment needed.
963                  */
964                 DBG_LEAVE;
965                 return;
966         }
967         ocscg=dtog(&osblock, osblock.fs_csaddr);
968         cs=fscs+ocscg;
969         blocks = 1+howmany(sblock.fs_cssize, sblock.fs_bsize)-
970             howmany(osblock.fs_cssize, osblock.fs_bsize);
971
972         /*
973          * Read original cylinder group from disk, and make a copy.
974          * XXX  If Nflag is set in some very rare cases we now miss
975          *      some changes done in updjcg by reading the unmodified
976          *      block from disk.
977          */
978         rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)),
979             (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
980         DBG_PRINT0("oscg read\n");
981         DBG_DUMP_CG(&sblock,
982             "old summary cg",
983             &aocg);
984
985         memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
986
987         /*
988          * Touch the cylinder group, set up local variables needed later
989          * and update the superblock.
990          */
991         acg.cg_time = utime;
992
993         /*
994          * XXX  In the case of having active snapshots we may need much more
995          *      blocks for the copy on write. We need each block twice, and
996          *      also up to 8*3 blocks for indirect blocks for all possible
997          *      references.
998          */
999         if(/*((int)sblock.fs_time&0x3)>0||*/ cs->cs_nbfree < blocks) {
1000                 /*
1001                  * There is not enough space in the old cylinder group to
1002                  * relocate all blocks as needed, so we relocate the whole
1003                  * cylinder group summary to a new group. We try to use the
1004                  * first complete new cylinder group just created. Within the
1005                  * cylinder group we align the area immediately after the
1006                  * cylinder group information location in order to be as
1007                  * close as possible to the original implementation of ffs.
1008                  *
1009                  * First we have to make sure we'll find enough space in the
1010                  * new cylinder group. If not, then we currently give up.
1011                  * We start with freeing everything which was used by the
1012                  * fragments of the old cylinder summary in the current group.
1013                  * Now we write back the group meta data, read in the needed
1014                  * meta data from the new cylinder group, and start allocating
1015                  * within that group. Here we can assume, the group to be
1016                  * completely empty. Which makes the handling of fragments and
1017                  * clusters a lot easier.
1018                  */
1019                 DBG_TRC;
1020                 if(sblock.fs_ncg-osblock.fs_ncg < 2) {
1021                         errx(2, "panic: not enough space");
1022                 }
1023
1024                 /*
1025                  * Point "d" to the first fragment not used by the cylinder
1026                  * summary.
1027                  */
1028                 d=osblock.fs_csaddr+(osblock.fs_cssize/osblock.fs_fsize);
1029
1030                 /*
1031                  * Set up last cluster size ("lcs") already here. Calculate
1032                  * the size for the trailing cluster just behind where "d"
1033                  * points to.
1034                  */
1035                 if(sblock.fs_contigsumsize > 0) {
1036                         for(block=howmany(d%sblock.fs_fpg, sblock.fs_frag),
1037                             lcs=0; lcs<sblock.fs_contigsumsize;
1038                             block++, lcs++) {
1039                                 if(isclr(cg_clustersfree(&acg), block)){
1040                                         break;
1041                                 }
1042                         }
1043                 }
1044
1045                 /*
1046                  * Point "d" to the last frag used by the cylinder summary.
1047                  */
1048                 d--;
1049
1050                 DBG_PRINT1("d=%jd\n",
1051                     (intmax_t)d);
1052                 if((d+1)%sblock.fs_frag) {
1053                         /*
1054                          * The end of the cylinder summary is not a complete
1055                          * block.
1056                          */
1057                         DBG_TRC;
1058                         frag_adjust(d%sblock.fs_fpg, -1);
1059                         for(; (d+1)%sblock.fs_frag; d--) {
1060                                 DBG_PRINT1("d=%jd\n",
1061                                     (intmax_t)d);
1062                                 setbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1063                                 acg.cg_cs.cs_nffree++;
1064                                 sblock.fs_cstotal.cs_nffree++;
1065                         }
1066                         /*
1067                          * Point "d" to the last fragment of the last
1068                          * (incomplete) block of the cylinder summary.
1069                          */
1070                         d++;
1071                         frag_adjust(d%sblock.fs_fpg, 1);
1072
1073                         if(isblock(&sblock, cg_blksfree(&acg),
1074                             (d%sblock.fs_fpg)/sblock.fs_frag)) {
1075                                 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1076                                 acg.cg_cs.cs_nffree-=sblock.fs_frag;
1077                                 acg.cg_cs.cs_nbfree++;
1078                                 sblock.fs_cstotal.cs_nffree-=sblock.fs_frag;
1079                                 sblock.fs_cstotal.cs_nbfree++;
1080                                 if(sblock.fs_contigsumsize > 0) {
1081                                         setbit(cg_clustersfree(&acg),
1082                                             (d%sblock.fs_fpg)/sblock.fs_frag);
1083                                         if(lcs < sblock.fs_contigsumsize) {
1084                                                 if(lcs) {
1085                                                         cg_clustersum(&acg)
1086                                                             [lcs]--;
1087                                                 }
1088                                                 lcs++;
1089                                                 cg_clustersum(&acg)[lcs]++;
1090                                         }
1091                                 }
1092                         }
1093                         /*
1094                          * Point "d" to the first fragment of the block before
1095                          * the last incomplete block.
1096                          */
1097                         d--;
1098                 }
1099
1100                 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1101                 for(d=rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr;
1102                     d-=sblock.fs_frag) {
1103                         DBG_TRC;
1104                         DBG_PRINT1("d=%jd\n", (intmax_t)d);
1105                         setblock(&sblock, cg_blksfree(&acg),
1106                             (d%sblock.fs_fpg)/sblock.fs_frag);
1107                         acg.cg_cs.cs_nbfree++;
1108                         sblock.fs_cstotal.cs_nbfree++;
1109                         if(sblock.fs_contigsumsize > 0) {
1110                                 setbit(cg_clustersfree(&acg),
1111                                     (d%sblock.fs_fpg)/sblock.fs_frag);
1112                                 /*
1113                                  * The last cluster size is already set up.
1114                                  */
1115                                 if(lcs < sblock.fs_contigsumsize) {
1116                                         if(lcs) {
1117                                                 cg_clustersum(&acg)[lcs]--;
1118                                         }
1119                                         lcs++;
1120                                         cg_clustersum(&acg)[lcs]++;
1121                                 }
1122                         }
1123                 }
1124                 *cs = acg.cg_cs;
1125
1126                 /*
1127                  * Now write the former cylinder group containing the cylinder
1128                  * summary back to disk.
1129                  */
1130                 wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)),
1131                     (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1132                 DBG_PRINT0("oscg written\n");
1133                 DBG_DUMP_CG(&sblock,
1134                     "old summary cg",
1135                     &acg);
1136
1137                 /*
1138                  * Find the beginning of the new cylinder group containing the
1139                  * cylinder summary.
1140                  */
1141                 sblock.fs_csaddr=cgdmin(&sblock, osblock.fs_ncg);
1142                 ncscg=dtog(&sblock, sblock.fs_csaddr);
1143                 cs=fscs+ncscg;
1144
1145
1146                 /*
1147                  * If Nflag is specified, we would now read random data instead
1148                  * of an empty cg structure from disk. So we can't simulate that
1149                  * part for now.
1150                  */
1151                 if(Nflag) {
1152                         DBG_PRINT0("nscg update skipped\n");
1153                         DBG_LEAVE;
1154                         return;
1155                 }
1156
1157                 /*
1158                  * Read the future cylinder group containing the cylinder
1159                  * summary from disk, and make a copy.
1160                  */
1161                 rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1162                     (size_t)sblock.fs_cgsize, (void *)&aocg, fsi);
1163                 DBG_PRINT0("nscg read\n");
1164                 DBG_DUMP_CG(&sblock,
1165                     "new summary cg",
1166                     &aocg);
1167
1168                 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
1169
1170                 /*
1171                  * Allocate all complete blocks used by the new cylinder
1172                  * summary.
1173                  */
1174                 for(d=sblock.fs_csaddr; d+sblock.fs_frag <=
1175                     sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize);
1176                     d+=sblock.fs_frag) {
1177                         clrblock(&sblock, cg_blksfree(&acg),
1178                             (d%sblock.fs_fpg)/sblock.fs_frag);
1179                         acg.cg_cs.cs_nbfree--;
1180                         sblock.fs_cstotal.cs_nbfree--;
1181                         if(sblock.fs_contigsumsize > 0) {
1182                                 clrbit(cg_clustersfree(&acg),
1183                                     (d%sblock.fs_fpg)/sblock.fs_frag);
1184                         }
1185                 }
1186
1187                 /*
1188                  * Allocate all fragments used by the cylinder summary in the
1189                  * last block.
1190                  */
1191                 if(d<sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize)) {
1192                         for(; d-sblock.fs_csaddr<
1193                             sblock.fs_cssize/sblock.fs_fsize;
1194                             d++) {
1195                                 clrbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1196                                 acg.cg_cs.cs_nffree--;
1197                                 sblock.fs_cstotal.cs_nffree--;
1198                         }
1199                         acg.cg_cs.cs_nbfree--;
1200                         acg.cg_cs.cs_nffree+=sblock.fs_frag;
1201                         sblock.fs_cstotal.cs_nbfree--;
1202                         sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1203                         if(sblock.fs_contigsumsize > 0) {
1204                                 clrbit(cg_clustersfree(&acg),
1205                                     (d%sblock.fs_fpg)/sblock.fs_frag);
1206                         }
1207
1208                         frag_adjust(d%sblock.fs_fpg, +1);
1209                 }
1210                 /*
1211                  * XXX  Handle the cluster statistics here in the case this
1212                  *      cylinder group is now almost full, and the remaining
1213                  *      space is less then the maximum cluster size. This is
1214                  *      probably not needed, as you would hardly find a file
1215                  *      system which has only MAXCSBUFS+FS_MAXCONTIG of free
1216                  *      space right behind the cylinder group information in
1217                  *      any new cylinder group.
1218                  */
1219
1220                 /*
1221                  * Update our statistics in the cylinder summary.
1222                  */
1223                 *cs = acg.cg_cs;
1224
1225                 /*
1226                  * Write the new cylinder group containing the cylinder summary
1227                  * back to disk.
1228                  */
1229                 wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1230                     (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1231                 DBG_PRINT0("nscg written\n");
1232                 DBG_DUMP_CG(&sblock,
1233                     "new summary cg",
1234                     &acg);
1235
1236                 DBG_LEAVE;
1237                 return;
1238         }
1239         /*
1240          * We have got enough of space in the current cylinder group, so we
1241          * can relocate just a few blocks, and let the summary information
1242          * grow in place where it is right now.
1243          */
1244         DBG_TRC;
1245
1246         cbase = cgbase(&osblock, ocscg);        /* old and new are equal */
1247         dupper = sblock.fs_csaddr - cbase +
1248             howmany(sblock.fs_cssize, sblock.fs_fsize);
1249         odupper = osblock.fs_csaddr - cbase +
1250             howmany(osblock.fs_cssize, osblock.fs_fsize);
1251
1252         sblock.fs_dsize -= dupper-odupper;
1253
1254         /*
1255          * Allocate the space for the array of blocks to be relocated.
1256          */
1257         bp=(struct gfs_bpp *)malloc(((dupper-odupper)/sblock.fs_frag+2)*
1258             sizeof(struct gfs_bpp));
1259         if(bp == NULL) {
1260                 errx(1, "malloc failed");
1261         }
1262         memset((char *)bp, 0, ((dupper-odupper)/sblock.fs_frag+2)*
1263             sizeof(struct gfs_bpp));
1264
1265         /*
1266          * Lock all new frags needed for the cylinder group summary. This is
1267          * done per fragment in the first and last block of the new required
1268          * area, and per block for all other blocks.
1269          *
1270          * Handle the first new block here (but only if some fragments where
1271          * already used for the cylinder summary).
1272          */
1273         ind=0;
1274         frag_adjust(odupper, -1);
1275         for(d=odupper; ((d<dupper)&&(d%sblock.fs_frag)); d++) {
1276                 DBG_PRINT1("scg first frag check loop d=%jd\n",
1277                     (intmax_t)d);
1278                 if(isclr(cg_blksfree(&acg), d)) {
1279                         if (!ind) {
1280                                 bp[ind].old=d/sblock.fs_frag;
1281                                 bp[ind].flags|=GFS_FL_FIRST;
1282                                 if(roundup(d, sblock.fs_frag) >= dupper) {
1283                                         bp[ind].flags|=GFS_FL_LAST;
1284                                 }
1285                                 ind++;
1286                         }
1287                 } else {
1288                         clrbit(cg_blksfree(&acg), d);
1289                         acg.cg_cs.cs_nffree--;
1290                         sblock.fs_cstotal.cs_nffree--;
1291                 }
1292                 /*
1293                  * No cluster handling is needed here, as there was at least
1294                  * one fragment in use by the cylinder summary in the old
1295                  * file system.
1296                  * No block-free counter handling here as this block was not
1297                  * a free block.
1298                  */
1299         }
1300         frag_adjust(odupper, 1);
1301
1302         /*
1303          * Handle all needed complete blocks here.
1304          */
1305         for(; d+sblock.fs_frag<=dupper; d+=sblock.fs_frag) {
1306                 DBG_PRINT1("scg block check loop d=%jd\n",
1307                     (intmax_t)d);
1308                 if(!isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1309                         for(f=d; f<d+sblock.fs_frag; f++) {
1310                                 if(isset(cg_blksfree(&aocg), f)) {
1311                                         acg.cg_cs.cs_nffree--;
1312                                         sblock.fs_cstotal.cs_nffree--;
1313                                 }
1314                         }
1315                         clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1316                         bp[ind].old=d/sblock.fs_frag;
1317                         ind++;
1318                 } else {
1319                         clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1320                         acg.cg_cs.cs_nbfree--;
1321                         sblock.fs_cstotal.cs_nbfree--;
1322                         if(sblock.fs_contigsumsize > 0) {
1323                                 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1324                                 for(lcs=0, l=(d/sblock.fs_frag)+1;
1325                                     lcs<sblock.fs_contigsumsize;
1326                                     l++, lcs++ ) {
1327                                         if(isclr(cg_clustersfree(&acg),l)){
1328                                                 break;
1329                                         }
1330                                 }
1331                                 if(lcs < sblock.fs_contigsumsize) {
1332                                         cg_clustersum(&acg)[lcs+1]--;
1333                                         if(lcs) {
1334                                                 cg_clustersum(&acg)[lcs]++;
1335                                         }
1336                                 }
1337                         }
1338                 }
1339                 /*
1340                  * No fragment counter handling is needed here, as this finally
1341                  * doesn't change after the relocation.
1342                  */
1343         }
1344
1345         /*
1346          * Handle all fragments needed in the last new affected block.
1347          */
1348         if(d<dupper) {
1349                 frag_adjust(dupper-1, -1);
1350
1351                 if(isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1352                         acg.cg_cs.cs_nbfree--;
1353                         sblock.fs_cstotal.cs_nbfree--;
1354                         acg.cg_cs.cs_nffree+=sblock.fs_frag;
1355                         sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1356                         if(sblock.fs_contigsumsize > 0) {
1357                                 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1358                                 for(lcs=0, l=(d/sblock.fs_frag)+1;
1359                                     lcs<sblock.fs_contigsumsize;
1360                                     l++, lcs++ ) {
1361                                         if(isclr(cg_clustersfree(&acg),l)){
1362                                                 break;
1363                                         }
1364                                 }
1365                                 if(lcs < sblock.fs_contigsumsize) {
1366                                         cg_clustersum(&acg)[lcs+1]--;
1367                                         if(lcs) {
1368                                                 cg_clustersum(&acg)[lcs]++;
1369                                         }
1370                                 }
1371                         }
1372                 }
1373
1374                 for(; d<dupper; d++) {
1375                         DBG_PRINT1("scg second frag check loop d=%jd\n",
1376                             (intmax_t)d);
1377                         if(isclr(cg_blksfree(&acg), d)) {
1378                                 bp[ind].old=d/sblock.fs_frag;
1379                                 bp[ind].flags|=GFS_FL_LAST;
1380                         } else {
1381                                 clrbit(cg_blksfree(&acg), d);
1382                                 acg.cg_cs.cs_nffree--;
1383                                 sblock.fs_cstotal.cs_nffree--;
1384                         }
1385                 }
1386                 if(bp[ind].flags & GFS_FL_LAST) { /* we have to advance here */
1387                         ind++;
1388                 }
1389                 frag_adjust(dupper-1, 1);
1390         }
1391
1392         /*
1393          * If we found a block to relocate just do so.
1394          */
1395         if(ind) {
1396                 for(i=0; i<ind; i++) {
1397                         if(!bp[i].old) { /* no more blocks listed */
1398                                 /*
1399                                  * XXX  A relative blocknumber should not be
1400                                  *      zero, which is not explicitly
1401                                  *      guaranteed by our code.
1402                                  */
1403                                 break;
1404                         }
1405                         /*
1406                          * Allocate a complete block in the same (current)
1407                          * cylinder group.
1408                          */
1409                         bp[i].new=alloc()/sblock.fs_frag;
1410
1411                         /*
1412                          * There is no frag_adjust() needed for the new block
1413                          * as it will have no fragments yet :-).
1414                          */
1415                         for(f=bp[i].old*sblock.fs_frag,
1416                             g=bp[i].new*sblock.fs_frag;
1417                             f<(bp[i].old+1)*sblock.fs_frag;
1418                             f++, g++) {
1419                                 if(isset(cg_blksfree(&aocg), f)) {
1420                                         setbit(cg_blksfree(&acg), g);
1421                                         acg.cg_cs.cs_nffree++;
1422                                         sblock.fs_cstotal.cs_nffree++;
1423                                 }
1424                         }
1425
1426                         /*
1427                          * Special handling is required if this was the first
1428                          * block. We have to consider the fragments which were
1429                          * used by the cylinder summary in the original block
1430                          * which re to be free in the copy of our block.  We
1431                          * have to be careful if this first block happens to
1432                          * be also the last block to be relocated.
1433                          */
1434                         if(bp[i].flags & GFS_FL_FIRST) {
1435                                 for(f=bp[i].old*sblock.fs_frag,
1436                                     g=bp[i].new*sblock.fs_frag;
1437                                     f<odupper;
1438                                     f++, g++) {
1439                                         setbit(cg_blksfree(&acg), g);
1440                                         acg.cg_cs.cs_nffree++;
1441                                         sblock.fs_cstotal.cs_nffree++;
1442                                 }
1443                                 if(!(bp[i].flags & GFS_FL_LAST)) {
1444                                         frag_adjust(bp[i].new*sblock.fs_frag,1);
1445                                 }
1446                         }
1447
1448                         /*
1449                          * Special handling is required if this is the last
1450                          * block to be relocated.
1451                          */
1452                         if(bp[i].flags & GFS_FL_LAST) {
1453                                 frag_adjust(bp[i].new*sblock.fs_frag, 1);
1454                                 frag_adjust(bp[i].old*sblock.fs_frag, -1);
1455                                 for(f=dupper;
1456                                     f<roundup(dupper, sblock.fs_frag);
1457                                     f++) {
1458                                         if(isclr(cg_blksfree(&acg), f)) {
1459                                                 setbit(cg_blksfree(&acg), f);
1460                                                 acg.cg_cs.cs_nffree++;
1461                                                 sblock.fs_cstotal.cs_nffree++;
1462                                         }
1463                                 }
1464                                 frag_adjust(bp[i].old*sblock.fs_frag, 1);
1465                         }
1466
1467                         /*
1468                          * !!! Attach the cylindergroup offset here.
1469                          */
1470                         bp[i].old+=cbase/sblock.fs_frag;
1471                         bp[i].new+=cbase/sblock.fs_frag;
1472
1473                         /*
1474                          * Copy the content of the block.
1475                          */
1476                         /*
1477                          * XXX  Here we will have to implement a copy on write
1478                          *      in the case we have any active snapshots.
1479                          */
1480                         rdfs(fsbtodb(&sblock, bp[i].old*sblock.fs_frag),
1481                             (size_t)sblock.fs_bsize, (void *)&ablk, fsi);
1482                         wtfs(fsbtodb(&sblock, bp[i].new*sblock.fs_frag),
1483                             (size_t)sblock.fs_bsize, (void *)&ablk, fso, Nflag);
1484                         DBG_DUMP_HEX(&sblock,
1485                             "copied full block",
1486                             (unsigned char *)&ablk);
1487
1488                         DBG_PRINT2("scg (%jd->%jd) block relocated\n",
1489                             (intmax_t)bp[i].old,
1490                             (intmax_t)bp[i].new);
1491                 }
1492
1493                 /*
1494                  * Now we have to update all references to any fragment which
1495                  * belongs to any block relocated. We iterate now over all
1496                  * cylinder groups, within those over all non zero length
1497                  * inodes.
1498                  */
1499                 for(cylno=0; cylno<osblock.fs_ncg; cylno++) {
1500                         DBG_PRINT1("scg doing cg (%d)\n",
1501                             cylno);
1502                         for(inc=osblock.fs_ipg-1 ; inc>0 ; inc--) {
1503                                 updrefs(cylno, (ino_t)inc, bp, fsi, fso, Nflag);
1504                         }
1505                 }
1506
1507                 /*
1508                  * All inodes are checked, now make sure the number of
1509                  * references found make sense.
1510                  */
1511                 for(i=0; i<ind; i++) {
1512                         if(!bp[i].found || (bp[i].found>sblock.fs_frag)) {
1513                                 warnx("error: %jd refs found for block %jd.",
1514                                     (intmax_t)bp[i].found, (intmax_t)bp[i].old);
1515                         }
1516
1517                 }
1518         }
1519         /*
1520          * The following statistics are not changed here:
1521          *     sblock.fs_cstotal.cs_ndir
1522          *     sblock.fs_cstotal.cs_nifree
1523          * The following statistics were already updated on the fly:
1524          *     sblock.fs_cstotal.cs_nffree
1525          *     sblock.fs_cstotal.cs_nbfree
1526          * As the statistics for this cylinder group are ready, copy it to
1527          * the summary information array.
1528          */
1529
1530         *cs = acg.cg_cs;
1531
1532         /*
1533          * Write summary cylinder group back to disk.
1534          */
1535         wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)), (size_t)sblock.fs_cgsize,
1536             (void *)&acg, fso, Nflag);
1537         DBG_PRINT0("scg written\n");
1538         DBG_DUMP_CG(&sblock,
1539             "new summary cg",
1540             &acg);
1541
1542         DBG_LEAVE;
1543         return;
1544 }
1545
1546 /* ************************************************************** rdfs ***** */
1547 /*
1548  * Here we read some block(s) from disk.
1549  */
1550 static void
1551 rdfs(ufs2_daddr_t bno, size_t size, void *bf, int fsi)
1552 {
1553         DBG_FUNC("rdfs")
1554         ssize_t n;
1555
1556         DBG_ENTER;
1557
1558         if (bno < 0) {
1559                 err(32, "rdfs: attempting to read negative block number");
1560         }
1561         if (lseek(fsi, (off_t)bno * DEV_BSIZE, 0) < 0) {
1562                 err(33, "rdfs: seek error: %jd", (intmax_t)bno);
1563         }
1564         n = read(fsi, bf, size);
1565         if (n != (ssize_t)size) {
1566                 err(34, "rdfs: read error: %jd", (intmax_t)bno);
1567         }
1568
1569         DBG_LEAVE;
1570         return;
1571 }
1572
1573 /* ************************************************************** wtfs ***** */
1574 /*
1575  * Here we write some block(s) to disk.
1576  */
1577 static void
1578 wtfs(ufs2_daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag)
1579 {
1580         DBG_FUNC("wtfs")
1581         ssize_t n;
1582
1583         DBG_ENTER;
1584
1585         if (Nflag) {
1586                 DBG_LEAVE;
1587                 return;
1588         }
1589         if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0) {
1590                 err(35, "wtfs: seek error: %ld", (long)bno);
1591         }
1592         n = write(fso, bf, size);
1593         if (n != (ssize_t)size) {
1594                 err(36, "wtfs: write error: %ld", (long)bno);
1595         }
1596
1597         DBG_LEAVE;
1598         return;
1599 }
1600
1601 /* ************************************************************* alloc ***** */
1602 /*
1603  * Here we allocate a free block in the current cylinder group. It is assumed,
1604  * that acg contains the current cylinder group. As we may take a block from
1605  * somewhere in the file system we have to handle cluster summary here.
1606  */
1607 static ufs2_daddr_t
1608 alloc(void)
1609 {
1610         DBG_FUNC("alloc")
1611         ufs2_daddr_t    d, blkno;
1612         int     lcs1, lcs2;
1613         int     l;
1614         int     csmin, csmax;
1615         int     dlower, dupper, dmax;
1616
1617         DBG_ENTER;
1618
1619         if (acg.cg_magic != CG_MAGIC) {
1620                 warnx("acg: bad magic number");
1621                 DBG_LEAVE;
1622                 return (0);
1623         }
1624         if (acg.cg_cs.cs_nbfree == 0) {
1625                 warnx("error: cylinder group ran out of space");
1626                 DBG_LEAVE;
1627                 return (0);
1628         }
1629         /*
1630          * We start seeking for free blocks only from the space available after
1631          * the end of the new grown cylinder summary. Otherwise we allocate a
1632          * block here which we have to relocate a couple of seconds later again
1633          * again, and we are not prepared to to this anyway.
1634          */
1635         blkno=-1;
1636         dlower=cgsblock(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1637         dupper=cgdmin(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1638         dmax=cgbase(&sblock, acg.cg_cgx)+sblock.fs_fpg;
1639         if (dmax > sblock.fs_size) {
1640                 dmax = sblock.fs_size;
1641         }
1642         dmax-=cgbase(&sblock, acg.cg_cgx); /* retransform into cg */
1643         csmin=sblock.fs_csaddr-cgbase(&sblock, acg.cg_cgx);
1644         csmax=csmin+howmany(sblock.fs_cssize, sblock.fs_fsize);
1645         DBG_PRINT3("seek range: dl=%d, du=%d, dm=%d\n",
1646             dlower,
1647             dupper,
1648             dmax);
1649         DBG_PRINT2("range cont: csmin=%d, csmax=%d\n",
1650             csmin,
1651             csmax);
1652
1653         for(d=0; (d<dlower && blkno==-1); d+=sblock.fs_frag) {
1654                 if(d>=csmin && d<=csmax) {
1655                         continue;
1656                 }
1657                 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1658                     d))) {
1659                         blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1660                         break;
1661                 }
1662         }
1663         for(d=dupper; (d<dmax && blkno==-1); d+=sblock.fs_frag) {
1664                 if(d>=csmin && d<=csmax) {
1665                         continue;
1666                 }
1667                 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1668                     d))) {
1669                         blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1670                         break;
1671                 }
1672         }
1673         if(blkno==-1) {
1674                 warnx("internal error: couldn't find promised block in cg");
1675                 DBG_LEAVE;
1676                 return (0);
1677         }
1678
1679         /*
1680          * This is needed if the block was found already in the first loop.
1681          */
1682         d=blkstofrags(&sblock, blkno);
1683
1684         clrblock(&sblock, cg_blksfree(&acg), blkno);
1685         if (sblock.fs_contigsumsize > 0) {
1686                 /*
1687                  * Handle the cluster allocation bitmap.
1688                  */
1689                 clrbit(cg_clustersfree(&acg), blkno);
1690                 /*
1691                  * We possibly have split a cluster here, so we have to do
1692                  * recalculate the sizes of the remaining cluster halves now,
1693                  * and use them for updating the cluster summary information.
1694                  *
1695                  * Lets start with the blocks before our allocated block ...
1696                  */
1697                 for(lcs1=0, l=blkno-1; lcs1<sblock.fs_contigsumsize;
1698                     l--, lcs1++ ) {
1699                         if(isclr(cg_clustersfree(&acg),l)){
1700                                 break;
1701                         }
1702                 }
1703                 /*
1704                  * ... and continue with the blocks right after our allocated
1705                  * block.
1706                  */
1707                 for(lcs2=0, l=blkno+1; lcs2<sblock.fs_contigsumsize;
1708                     l++, lcs2++ ) {
1709                         if(isclr(cg_clustersfree(&acg),l)){
1710                                 break;
1711                         }
1712                 }
1713
1714                 /*
1715                  * Now update all counters.
1716                  */
1717                 cg_clustersum(&acg)[MIN(lcs1+lcs2+1,sblock.fs_contigsumsize)]--;
1718                 if(lcs1) {
1719                         cg_clustersum(&acg)[lcs1]++;
1720                 }
1721                 if(lcs2) {
1722                         cg_clustersum(&acg)[lcs2]++;
1723                 }
1724         }
1725         /*
1726          * Update all statistics based on blocks.
1727          */
1728         acg.cg_cs.cs_nbfree--;
1729         sblock.fs_cstotal.cs_nbfree--;
1730
1731         DBG_LEAVE;
1732         return (d);
1733 }
1734
1735 /* *********************************************************** isblock ***** */
1736 /*
1737  * Here we check if all frags of a block are free. For more details again
1738  * please see the source of newfs(8), as this function is taken over almost
1739  * unchanged.
1740  */
1741 static int
1742 isblock(struct fs *fs, unsigned char *cp, int h)
1743 {
1744         DBG_FUNC("isblock")
1745         unsigned char   mask;
1746
1747         DBG_ENTER;
1748
1749         switch (fs->fs_frag) {
1750         case 8:
1751                 DBG_LEAVE;
1752                 return (cp[h] == 0xff);
1753         case 4:
1754                 mask = 0x0f << ((h & 0x1) << 2);
1755                 DBG_LEAVE;
1756                 return ((cp[h >> 1] & mask) == mask);
1757         case 2:
1758                 mask = 0x03 << ((h & 0x3) << 1);
1759                 DBG_LEAVE;
1760                 return ((cp[h >> 2] & mask) == mask);
1761         case 1:
1762                 mask = 0x01 << (h & 0x7);
1763                 DBG_LEAVE;
1764                 return ((cp[h >> 3] & mask) == mask);
1765         default:
1766                 fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
1767                 DBG_LEAVE;
1768                 return (0);
1769         }
1770 }
1771
1772 /* ********************************************************** clrblock ***** */
1773 /*
1774  * Here we allocate a complete block in the block map. For more details again
1775  * please see the source of newfs(8), as this function is taken over almost
1776  * unchanged.
1777  */
1778 static void
1779 clrblock(struct fs *fs, unsigned char *cp, int h)
1780 {
1781         DBG_FUNC("clrblock")
1782
1783         DBG_ENTER;
1784
1785         switch ((fs)->fs_frag) {
1786         case 8:
1787                 cp[h] = 0;
1788                 break;
1789         case 4:
1790                 cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1791                 break;
1792         case 2:
1793                 cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1794                 break;
1795         case 1:
1796                 cp[h >> 3] &= ~(0x01 << (h & 0x7));
1797                 break;
1798         default:
1799                 warnx("clrblock bad fs_frag %d", fs->fs_frag);
1800                 break;
1801         }
1802
1803         DBG_LEAVE;
1804         return;
1805 }
1806
1807 /* ********************************************************** setblock ***** */
1808 /*
1809  * Here we free a complete block in the free block map. For more details again
1810  * please see the source of newfs(8), as this function is taken over almost
1811  * unchanged.
1812  */
1813 static void
1814 setblock(struct fs *fs, unsigned char *cp, int h)
1815 {
1816         DBG_FUNC("setblock")
1817
1818         DBG_ENTER;
1819
1820         switch (fs->fs_frag) {
1821         case 8:
1822                 cp[h] = 0xff;
1823                 break;
1824         case 4:
1825                 cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1826                 break;
1827         case 2:
1828                 cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1829                 break;
1830         case 1:
1831                 cp[h >> 3] |= (0x01 << (h & 0x7));
1832                 break;
1833         default:
1834                 warnx("setblock bad fs_frag %d", fs->fs_frag);
1835                 break;
1836         }
1837
1838         DBG_LEAVE;
1839         return;
1840 }
1841
1842 /* ************************************************************ ginode ***** */
1843 /*
1844  * This function provides access to an individual inode. We find out in which
1845  * block the requested inode is located, read it from disk if needed, and
1846  * return the pointer into that block. We maintain a cache of one block to
1847  * not read the same block again and again if we iterate linearly over all
1848  * inodes.
1849  */
1850 static union dinode *
1851 ginode(ino_t inumber, int fsi, int cg)
1852 {
1853         DBG_FUNC("ginode")
1854         static ino_t    startinum = 0;  /* first inode in cached block */
1855
1856         DBG_ENTER;
1857
1858         /*
1859          * The inumber passed in is relative to the cg, so use it here to see
1860          * if the inode has been allocated yet.
1861          */
1862         if (isclr(cg_inosused(&aocg), inumber)) {
1863                 DBG_LEAVE;
1864                 return NULL;
1865         }
1866         /*
1867          * Now make the inumber relative to the entire inode space so it can
1868          * be sanity checked.
1869          */
1870         inumber += (cg * sblock.fs_ipg);
1871         if (inumber < ROOTINO) {
1872                 DBG_LEAVE;
1873                 return NULL;
1874         }
1875         if (inumber > maxino)
1876                 errx(8, "bad inode number %d to ginode", inumber);
1877         if (startinum == 0 ||
1878             inumber < startinum || inumber >= startinum + INOPB(&sblock)) {
1879                 inoblk = fsbtodb(&sblock, ino_to_fsba(&sblock, inumber));
1880                 rdfs(inoblk, (size_t)sblock.fs_bsize, inobuf, fsi);
1881                 startinum = (inumber / INOPB(&sblock)) * INOPB(&sblock);
1882         }
1883         DBG_LEAVE;
1884         if (sblock.fs_magic == FS_UFS1_MAGIC)
1885                 return (union dinode *)((uintptr_t)inobuf +
1886                     (inumber % INOPB(&sblock)) * sizeof(struct ufs1_dinode));
1887         return (union dinode *)((uintptr_t)inobuf +
1888             (inumber % INOPB(&sblock)) * sizeof(struct ufs2_dinode));
1889 }
1890
1891 /* ****************************************************** charsperline ***** */
1892 /*
1893  * Figure out how many lines our current terminal has. For more details again
1894  * please see the source of newfs(8), as this function is taken over almost
1895  * unchanged.
1896  */
1897 static int
1898 charsperline(void)
1899 {
1900         DBG_FUNC("charsperline")
1901         int     columns;
1902         char    *cp;
1903         struct winsize  ws;
1904
1905         DBG_ENTER;
1906
1907         columns = 0;
1908         if (ioctl(0, TIOCGWINSZ, &ws) != -1) {
1909                 columns = ws.ws_col;
1910         }
1911         if (columns == 0 && (cp = getenv("COLUMNS"))) {
1912                 columns = atoi(cp);
1913         }
1914         if (columns == 0) {
1915                 columns = 80;   /* last resort */
1916         }
1917
1918         DBG_LEAVE;
1919         return columns;
1920 }
1921
1922 /* ****************************************************** get_dev_size ***** */
1923 /*
1924  * Get the size of the partition if we can't figure it out from the disklabel,
1925  * e.g. from vinum volumes.
1926  */
1927 static void
1928 get_dev_size(int fd, int *size)
1929 {
1930    int sectorsize;
1931    off_t mediasize;
1932
1933    if (ioctl(fd, DIOCGSECTORSIZE, &sectorsize) == -1)
1934         err(1,"DIOCGSECTORSIZE");
1935    if (ioctl(fd, DIOCGMEDIASIZE, &mediasize) == -1)
1936         err(1,"DIOCGMEDIASIZE");
1937
1938    if (sectorsize <= 0)
1939        errx(1, "bogus sectorsize: %d", sectorsize);
1940
1941    *size = mediasize / sectorsize;
1942 }
1943
1944 /* ************************************************************** main ***** */
1945 /*
1946  * growfs(8)  is a utility which allows to increase the size of an existing
1947  * ufs file system. Currently this can only be done on unmounted file system.
1948  * It recognizes some command line options to specify the new desired size,
1949  * and it does some basic checkings. The old file system size is determined
1950  * and after some more checks like we can really access the new last block
1951  * on the disk etc. we calculate the new parameters for the superblock. After
1952  * having done this we just call growfs() which will do the work.  Before
1953  * we finish the only thing left is to update the disklabel.
1954  * We still have to provide support for snapshots. Therefore we first have to
1955  * understand what data structures are always replicated in the snapshot on
1956  * creation, for all other blocks we touch during our procedure, we have to
1957  * keep the old blocks unchanged somewhere available for the snapshots. If we
1958  * are lucky, then we only have to handle our blocks to be relocated in that
1959  * way.
1960  * Also we have to consider in what order we actually update the critical
1961  * data structures of the file system to make sure, that in case of a disaster
1962  * fsck(8) is still able to restore any lost data.
1963  * The foreseen last step then will be to provide for growing even mounted
1964  * file systems. There we have to extend the mount() system call to provide
1965  * userland access to the file system locking facility.
1966  */
1967 int
1968 main(int argc, char **argv)
1969 {
1970         DBG_FUNC("main")
1971         char    *device, *special, *cp;
1972         int     ch;
1973         unsigned int    size=0;
1974         size_t  len;
1975         unsigned int    Nflag=0;
1976         int     ExpertFlag=0;
1977         struct stat     st;
1978         struct disklabel        *lp;
1979         struct partition        *pp;
1980         int     i,fsi,fso;
1981     u_int32_t p_size;
1982         char    reply[5];
1983 #ifdef FSMAXSNAP
1984         int     j;
1985 #endif /* FSMAXSNAP */
1986
1987         DBG_ENTER;
1988
1989         while((ch=getopt(argc, argv, "Ns:vy")) != -1) {
1990                 switch(ch) {
1991                 case 'N':
1992                         Nflag=1;
1993                         break;
1994                 case 's':
1995                         size=(size_t)atol(optarg);
1996                         if(size<1) {
1997                                 usage();
1998                         }
1999                         break;
2000                 case 'v': /* for compatibility to newfs */
2001                         break;
2002                 case 'y':
2003                         ExpertFlag=1;
2004                         break;
2005                 case '?':
2006                         /* FALLTHROUGH */
2007                 default:
2008                         usage();
2009                 }
2010         }
2011         argc -= optind;
2012         argv += optind;
2013
2014         if(argc != 1) {
2015                 usage();
2016         }
2017         device=*argv;
2018
2019         /*
2020          * Now try to guess the (raw)device name.
2021          */
2022         if (0 == strrchr(device, '/')) {
2023                 /*
2024                  * No path prefix was given, so try in that order:
2025                  *     /dev/r%s
2026                  *     /dev/%s
2027                  *     /dev/vinum/r%s
2028                  *     /dev/vinum/%s.
2029                  *
2030                  * FreeBSD now doesn't distinguish between raw and block
2031                  * devices any longer, but it should still work this way.
2032                  */
2033                 len=strlen(device)+strlen(_PATH_DEV)+2+strlen("vinum/");
2034                 special=(char *)malloc(len);
2035                 if(special == NULL) {
2036                         errx(1, "malloc failed");
2037                 }
2038                 snprintf(special, len, "%sr%s", _PATH_DEV, device);
2039                 if (stat(special, &st) == -1) {
2040                         snprintf(special, len, "%s%s", _PATH_DEV, device);
2041                         if (stat(special, &st) == -1) {
2042                                 snprintf(special, len, "%svinum/r%s",
2043                                     _PATH_DEV, device);
2044                                 if (stat(special, &st) == -1) {
2045                                         /* For now this is the 'last resort' */
2046                                         snprintf(special, len, "%svinum/%s",
2047                                             _PATH_DEV, device);
2048                                 }
2049                         }
2050                 }
2051                 device = special;
2052         }
2053
2054         /*
2055          * Try to access our devices for writing ...
2056          */
2057         if (Nflag) {
2058                 fso = -1;
2059         } else {
2060                 fso = open(device, O_WRONLY);
2061                 if (fso < 0) {
2062                         err(1, "%s", device);
2063                 }
2064         }
2065
2066         /*
2067          * ... and reading.
2068          */
2069         fsi = open(device, O_RDONLY);
2070         if (fsi < 0) {
2071                 err(1, "%s", device);
2072         }
2073
2074         /*
2075          * Try to read a label and guess the slice if not specified. This
2076          * code should guess the right thing and avoid to bother the user
2077          * with the task of specifying the option -v on vinum volumes.
2078          */
2079         cp=device+strlen(device)-1;
2080         lp = get_disklabel(fsi);
2081         pp = NULL;
2082     if (lp != NULL) {
2083         if (isdigit(*cp)) {
2084             pp = &lp->d_partitions[2];
2085         } else if (*cp>='a' && *cp<='h') {
2086             pp = &lp->d_partitions[*cp - 'a'];
2087         } else {
2088             errx(1, "unknown device");
2089         }
2090         p_size = pp->p_size;
2091     } else {
2092         get_dev_size(fsi, &p_size);
2093     }
2094
2095         /*
2096          * Check if that partition is suitable for growing a file system.
2097          */
2098         if (p_size < 1) {
2099                 errx(1, "partition is unavailable");
2100         }
2101
2102         /*
2103          * Read the current superblock, and take a backup.
2104          */
2105         for (i = 0; sblock_try[i] != -1; i++) {
2106                 sblockloc = sblock_try[i] / DEV_BSIZE;
2107                 rdfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&(osblock), fsi);
2108                 if ((osblock.fs_magic == FS_UFS1_MAGIC ||
2109                      (osblock.fs_magic == FS_UFS2_MAGIC &&
2110                       osblock.fs_sblockloc == sblock_try[i])) &&
2111                     osblock.fs_bsize <= MAXBSIZE &&
2112                     osblock.fs_bsize >= (int32_t) sizeof(struct fs))
2113                         break;
2114         }
2115         if (sblock_try[i] == -1) {
2116                 errx(1, "superblock not recognized");
2117         }
2118         memcpy((void *)&fsun1, (void *)&fsun2, sizeof(fsun2));
2119         maxino = sblock.fs_ncg * sblock.fs_ipg;
2120
2121         DBG_OPEN("/tmp/growfs.debug"); /* already here we need a superblock */
2122         DBG_DUMP_FS(&sblock,
2123             "old sblock");
2124
2125         /*
2126          * Determine size to grow to. Default to the full size specified in
2127          * the disk label.
2128          */
2129         sblock.fs_size = dbtofsb(&osblock, p_size);
2130         if (size != 0) {
2131                 if (size > p_size){
2132                         errx(1, "there is not enough space (%d < %d)",
2133                             p_size, size);
2134                 }
2135                 sblock.fs_size = dbtofsb(&osblock, size);
2136         }
2137
2138         /*
2139          * Are we really growing ?
2140          */
2141         if(osblock.fs_size >= sblock.fs_size) {
2142                 errx(1, "we are not growing (%jd->%jd)",
2143                     (intmax_t)osblock.fs_size, (intmax_t)sblock.fs_size);
2144         }
2145
2146
2147 #ifdef FSMAXSNAP
2148         /*
2149          * Check if we find an active snapshot.
2150          */
2151         if(ExpertFlag == 0) {
2152                 for(j=0; j<FSMAXSNAP; j++) {
2153                         if(sblock.fs_snapinum[j]) {
2154                                 errx(1, "active snapshot found in file system\n"
2155                                     "   please remove all snapshots before "
2156                                     "using growfs");
2157                         }
2158                         if(!sblock.fs_snapinum[j]) { /* list is dense */
2159                                 break;
2160                         }
2161                 }
2162         }
2163 #endif
2164
2165         if (ExpertFlag == 0 && Nflag == 0) {
2166                 printf("We strongly recommend you to make a backup "
2167                     "before growing the Filesystem\n\n"
2168                     " Did you backup your data (Yes/No) ? ");
2169                 fgets(reply, (int)sizeof(reply), stdin);
2170                 if (strcmp(reply, "Yes\n")){
2171                         printf("\n Nothing done \n");
2172                         exit (0);
2173                 }
2174         }
2175
2176         printf("new file systemsize is: %jd frags\n", (intmax_t)sblock.fs_size);
2177
2178         /*
2179          * Try to access our new last block in the file system. Even if we
2180          * later on realize we have to abort our operation, on that block
2181          * there should be no data, so we can't destroy something yet.
2182          */
2183         wtfs((ufs2_daddr_t)p_size-1, (size_t)DEV_BSIZE, (void *)&sblock,
2184             fso, Nflag);
2185
2186         /*
2187          * Now calculate new superblock values and check for reasonable
2188          * bound for new file system size:
2189          *     fs_size:    is derived from label or user input
2190          *     fs_dsize:   should get updated in the routines creating or
2191          *                 updating the cylinder groups on the fly
2192          *     fs_cstotal: should get updated in the routines creating or
2193          *                 updating the cylinder groups
2194          */
2195
2196         /*
2197          * Update the number of cylinders and cylinder groups in the file system.
2198          */
2199         if (sblock.fs_magic == FS_UFS1_MAGIC) {
2200                 sblock.fs_old_ncyl =
2201                     sblock.fs_size * sblock.fs_old_nspf / sblock.fs_old_spc;
2202                 if (sblock.fs_size * sblock.fs_old_nspf >
2203                     sblock.fs_old_ncyl * sblock.fs_old_spc)
2204                         sblock.fs_old_ncyl++;
2205         }
2206         sblock.fs_ncg = howmany(sblock.fs_size, sblock.fs_fpg);
2207         maxino = sblock.fs_ncg * sblock.fs_ipg;
2208
2209         if (sblock.fs_size % sblock.fs_fpg != 0 &&
2210             sblock.fs_size % sblock.fs_fpg < cgdmin(&sblock, sblock.fs_ncg)) {
2211                 /*
2212                  * The space in the new last cylinder group is too small,
2213                  * so revert back.
2214                  */
2215                 sblock.fs_ncg--;
2216                 if (sblock.fs_magic == FS_UFS1_MAGIC)
2217                         sblock.fs_old_ncyl = sblock.fs_ncg * sblock.fs_old_cpg;
2218                 printf("Warning: %jd sector(s) cannot be allocated.\n",
2219                     (intmax_t)fsbtodb(&sblock, sblock.fs_size % sblock.fs_fpg));
2220                 sblock.fs_size = sblock.fs_ncg * sblock.fs_fpg;
2221         }
2222
2223         /*
2224          * Update the space for the cylinder group summary information in the
2225          * respective cylinder group data area.
2226          */
2227         sblock.fs_cssize =
2228             fragroundup(&sblock, sblock.fs_ncg * sizeof(struct csum));
2229
2230         if(osblock.fs_size >= sblock.fs_size) {
2231                 errx(1, "not enough new space");
2232         }
2233
2234         DBG_PRINT0("sblock calculated\n");
2235
2236         /*
2237          * Ok, everything prepared, so now let's do the tricks.
2238          */
2239         growfs(fsi, fso, Nflag);
2240
2241         /*
2242          * Update the disk label.
2243          */
2244     if (!unlabeled) {
2245         pp->p_fsize = sblock.fs_fsize;
2246         pp->p_frag = sblock.fs_frag;
2247         pp->p_cpg = sblock.fs_fpg;
2248
2249         return_disklabel(fso, lp, Nflag);
2250         DBG_PRINT0("label rewritten\n");
2251     }
2252
2253         close(fsi);
2254         if(fso>-1) close(fso);
2255
2256         DBG_CLOSE;
2257
2258         DBG_LEAVE;
2259         return 0;
2260 }
2261
2262 /* ************************************************** return_disklabel ***** */
2263 /*
2264  * Write the updated disklabel back to disk.
2265  */
2266 static void
2267 return_disklabel(int fd, struct disklabel *lp, unsigned int Nflag)
2268 {
2269         DBG_FUNC("return_disklabel")
2270         u_short sum;
2271         u_short *ptr;
2272
2273         DBG_ENTER;
2274
2275         if(!lp) {
2276                 DBG_LEAVE;
2277                 return;
2278         }
2279         if(!Nflag) {
2280                 lp->d_checksum=0;
2281                 sum = 0;
2282                 ptr=(u_short *)lp;
2283
2284                 /*
2285                  * recalculate checksum
2286                  */
2287                 while(ptr < (u_short *)&lp->d_partitions[lp->d_npartitions]) {
2288                         sum ^= *ptr++;
2289                 }
2290                 lp->d_checksum=sum;
2291
2292                 if (ioctl(fd, DIOCWDINFO, (char *)lp) < 0) {
2293                         errx(1, "DIOCWDINFO failed");
2294                 }
2295         }
2296         free(lp);
2297
2298         DBG_LEAVE;
2299         return ;
2300 }
2301
2302 /* ***************************************************** get_disklabel ***** */
2303 /*
2304  * Read the disklabel from disk.
2305  */
2306 static struct disklabel *
2307 get_disklabel(int fd)
2308 {
2309         DBG_FUNC("get_disklabel")
2310         static struct   disklabel *lab;
2311
2312         DBG_ENTER;
2313
2314         lab=(struct disklabel *)malloc(sizeof(struct disklabel));
2315         if (!lab)
2316                 errx(1, "malloc failed");
2317
2318     if (!ioctl(fd, DIOCGDINFO, (char *)lab))
2319         return (lab);
2320
2321     unlabeled++;
2322
2323         DBG_LEAVE;
2324         return (NULL);
2325 }
2326
2327
2328 /* ************************************************************* usage ***** */
2329 /*
2330  * Dump a line of usage.
2331  */
2332 static void
2333 usage(void)
2334 {
2335         DBG_FUNC("usage")
2336
2337         DBG_ENTER;
2338
2339         fprintf(stderr, "usage: growfs [-Ny] [-s size] special\n");
2340
2341         DBG_LEAVE;
2342         exit(1);
2343 }
2344
2345 /* *********************************************************** updclst ***** */
2346 /*
2347  * This updates most parameters and the bitmap related to cluster. We have to
2348  * assume that sblock, osblock, acg are set up.
2349  */
2350 static void
2351 updclst(int block)
2352 {
2353         DBG_FUNC("updclst")
2354         static int      lcs=0;
2355
2356         DBG_ENTER;
2357
2358         if(sblock.fs_contigsumsize < 1) { /* no clustering */
2359                 return;
2360         }
2361         /*
2362          * update cluster allocation map
2363          */
2364         setbit(cg_clustersfree(&acg), block);
2365
2366         /*
2367          * update cluster summary table
2368          */
2369         if(!lcs) {
2370                 /*
2371                  * calculate size for the trailing cluster
2372                  */
2373                 for(block--; lcs<sblock.fs_contigsumsize; block--, lcs++ ) {
2374                         if(isclr(cg_clustersfree(&acg), block)){
2375                                 break;
2376                         }
2377                 }
2378         }
2379         if(lcs < sblock.fs_contigsumsize) {
2380                 if(lcs) {
2381                         cg_clustersum(&acg)[lcs]--;
2382                 }
2383                 lcs++;
2384                 cg_clustersum(&acg)[lcs]++;
2385         }
2386
2387         DBG_LEAVE;
2388         return;
2389 }
2390
2391 /* *********************************************************** updrefs ***** */
2392 /*
2393  * This updates all references to relocated blocks for the given inode.  The
2394  * inode is given as number within the cylinder group, and the number of the
2395  * cylinder group.
2396  */
2397 static void
2398 updrefs(int cg, ino_t in, struct gfs_bpp *bp, int fsi, int fso, unsigned int
2399     Nflag)
2400 {
2401         DBG_FUNC("updrefs")
2402         ufs_lbn_t       len, lbn, numblks;
2403         ufs2_daddr_t    iptr, blksperindir;
2404         union dinode    *ino;
2405         int             i, mode, inodeupdated;
2406
2407         DBG_ENTER;
2408
2409         ino = ginode(in, fsi, cg);
2410         if (ino == NULL) {
2411                 DBG_LEAVE;
2412                 return;
2413         }
2414         mode = DIP(ino, di_mode) & IFMT;
2415         if (mode != IFDIR && mode != IFREG && mode != IFLNK) {
2416                 DBG_LEAVE;
2417                 return; /* only check DIR, FILE, LINK */
2418         }
2419         if (mode == IFLNK && 
2420             DIP(ino, di_size) < (u_int64_t) sblock.fs_maxsymlinklen) {
2421                 DBG_LEAVE;
2422                 return; /* skip short symlinks */
2423         }
2424         numblks = howmany(DIP(ino, di_size), sblock.fs_bsize);
2425         if (numblks == 0) {
2426                 DBG_LEAVE;
2427                 return; /* skip empty file */
2428         }
2429         if (DIP(ino, di_blocks) == 0) {
2430                 DBG_LEAVE;
2431                 return; /* skip empty swiss cheesy file or old fastlink */
2432         }
2433         DBG_PRINT2("scg checking inode (%d in %d)\n",
2434             in,
2435             cg);
2436
2437         /*
2438          * Check all the blocks.
2439          */
2440         inodeupdated = 0;
2441         len = numblks < NDADDR ? numblks : NDADDR;
2442         for (i = 0; i < len; i++) {
2443                 iptr = DIP(ino, di_db[i]);
2444                 if (iptr == 0)
2445                         continue;
2446                 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2447                         DIP_SET(ino, di_db[i], iptr);
2448                         inodeupdated++;
2449                 }
2450         }
2451         DBG_PRINT0("~~scg direct blocks checked\n");
2452
2453         blksperindir = 1;
2454         len = numblks - NDADDR;
2455         lbn = NDADDR;
2456         for (i = 0; len > 0 && i < NIADDR; i++) {
2457                 iptr = DIP(ino, di_ib[i]);
2458                 if (iptr == 0)
2459                         continue;
2460                 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2461                         DIP_SET(ino, di_ib[i], iptr);
2462                         inodeupdated++;
2463                 }
2464                 indirchk(blksperindir, lbn, iptr, numblks, bp, fsi, fso, Nflag);
2465                 blksperindir *= NINDIR(&sblock);
2466                 lbn += blksperindir;
2467                 len -= blksperindir;
2468                 DBG_PRINT1("scg indirect_%d blocks checked\n", i + 1);
2469         }
2470         if (inodeupdated)
2471                 wtfs(inoblk, sblock.fs_bsize, inobuf, fso, Nflag);
2472
2473         DBG_LEAVE;
2474         return;
2475 }
2476
2477 /*
2478  * Recursively check all the indirect blocks.
2479  */
2480 static void
2481 indirchk(ufs_lbn_t blksperindir, ufs_lbn_t lbn, ufs2_daddr_t blkno,
2482     ufs_lbn_t lastlbn, struct gfs_bpp *bp, int fsi, int fso, unsigned int Nflag)
2483 {
2484         DBG_FUNC("indirchk")
2485         void *ibuf;
2486         int i, last;
2487         ufs2_daddr_t iptr;
2488
2489         DBG_ENTER;
2490
2491         /* read in the indirect block. */
2492         ibuf = malloc(sblock.fs_bsize);
2493         if (!ibuf)
2494                 errx(1, "malloc failed");
2495         rdfs(fsbtodb(&sblock, blkno), (size_t)sblock.fs_bsize, ibuf, fsi);
2496         last = howmany(lastlbn - lbn, blksperindir) < NINDIR(&sblock) ?
2497             howmany(lastlbn - lbn, blksperindir) : NINDIR(&sblock);
2498         for (i = 0; i < last; i++) {
2499                 if (sblock.fs_magic == FS_UFS1_MAGIC)
2500                         iptr = ((ufs1_daddr_t *)ibuf)[i];
2501                 else
2502                         iptr = ((ufs2_daddr_t *)ibuf)[i];
2503                 if (iptr == 0)
2504                         continue;
2505                 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2506                         if (sblock.fs_magic == FS_UFS1_MAGIC)
2507                                 ((ufs1_daddr_t *)ibuf)[i] = iptr;
2508                         else
2509                                 ((ufs2_daddr_t *)ibuf)[i] = iptr;
2510                 }
2511                 if (blksperindir == 1)
2512                         continue;
2513                 indirchk(blksperindir / NINDIR(&sblock), lbn + blksperindir * i,
2514                     iptr, lastlbn, bp, fsi, fso, Nflag);
2515         }
2516         free(ibuf);
2517
2518         DBG_LEAVE;
2519         return;
2520 }