]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sbin/growfs/growfs.c
Uupdate code to vendor rev. 4183 (release/2.8)
[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 static 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         time_t  modtime;
178         uint    cylno;
179         int     i, j, width;
180         char    tmpbuf[100];
181 #ifdef FSIRAND
182         static int      randinit=0;
183
184         DBG_ENTER;
185
186         if (!randinit) {
187                 randinit = 1;
188                 srandomdev();
189         }
190 #else /* not FSIRAND */
191
192         DBG_ENTER;
193
194 #endif /* FSIRAND */
195         time(&modtime);
196
197         /*
198          * Get the cylinder summary into the memory.
199          */
200         fscs = (struct csum *)calloc((size_t)1, (size_t)sblock.fs_cssize);
201         if(fscs == NULL) {
202                 errx(1, "calloc failed");
203         }
204         for (i = 0; i < osblock.fs_cssize; i += osblock.fs_bsize) {
205                 rdfs(fsbtodb(&osblock, osblock.fs_csaddr +
206                     numfrags(&osblock, i)), (size_t)MIN(osblock.fs_cssize - i,
207                     osblock.fs_bsize), (void *)(((char *)fscs)+i), fsi);
208         }
209
210 #ifdef FS_DEBUG
211 {
212         struct csum     *dbg_csp;
213         int     dbg_csc;
214         char    dbg_line[80];
215
216         dbg_csp=fscs;
217         for(dbg_csc=0; dbg_csc<osblock.fs_ncg; dbg_csc++) {
218                 snprintf(dbg_line, sizeof(dbg_line),
219                     "%d. old csum in old location", dbg_csc);
220                 DBG_DUMP_CSUM(&osblock,
221                     dbg_line,
222                     dbg_csp++);
223         }
224 }
225 #endif /* FS_DEBUG */
226         DBG_PRINT0("fscs read\n");
227
228         /*
229          * Do all needed changes in the former last cylinder group.
230          */
231         updjcg(osblock.fs_ncg-1, modtime, fsi, fso, Nflag);
232
233         /*
234          * Dump out summary information about file system.
235          */
236 #       define B2MBFACTOR (1 / (1024.0 * 1024.0))
237         printf("growfs: %.1fMB (%jd sectors) block size %d, fragment size %d\n",
238             (float)sblock.fs_size * sblock.fs_fsize * B2MBFACTOR,
239             (intmax_t)fsbtodb(&sblock, sblock.fs_size), sblock.fs_bsize,
240             sblock.fs_fsize);
241         printf("\tusing %d cylinder groups of %.2fMB, %d blks, %d inodes.\n",
242             sblock.fs_ncg, (float)sblock.fs_fpg * sblock.fs_fsize * B2MBFACTOR,
243             sblock.fs_fpg / sblock.fs_frag, sblock.fs_ipg);
244         if (sblock.fs_flags & FS_DOSOFTDEP)
245                 printf("\twith soft updates\n");
246 #       undef B2MBFACTOR
247
248         /*
249          * Now build the cylinders group blocks and
250          * then print out indices of cylinder groups.
251          */
252         printf("super-block backups (for fsck -b #) at:\n");
253         i = 0;
254         width = charsperline();
255
256         /*
257          * Iterate for only the new cylinder groups.
258          */
259         for (cylno = osblock.fs_ncg; cylno < sblock.fs_ncg; cylno++) {
260                 initcg(cylno, modtime, fso, Nflag);
261                 j = sprintf(tmpbuf, " %jd%s",
262                     (intmax_t)fsbtodb(&sblock, cgsblock(&sblock, cylno)),
263                     cylno < (sblock.fs_ncg-1) ? "," : "" );
264                 if (i + j >= width) {
265                         printf("\n");
266                         i = 0;
267                 }
268                 i += j;
269                 printf("%s", tmpbuf);
270                 fflush(stdout);
271         }
272         printf("\n");
273
274         /*
275          * Do all needed changes in the first cylinder group.
276          * allocate blocks in new location
277          */
278         updcsloc(modtime, fsi, fso, Nflag);
279
280         /*
281          * Now write the cylinder summary back to disk.
282          */
283         for (i = 0; i < sblock.fs_cssize; i += sblock.fs_bsize) {
284                 wtfs(fsbtodb(&sblock, sblock.fs_csaddr + numfrags(&sblock, i)),
285                     (size_t)MIN(sblock.fs_cssize - i, sblock.fs_bsize),
286                     (void *)(((char *)fscs) + i), fso, Nflag);
287         }
288         DBG_PRINT0("fscs written\n");
289
290 #ifdef FS_DEBUG
291 {
292         struct csum     *dbg_csp;
293         int     dbg_csc;
294         char    dbg_line[80];
295
296         dbg_csp=fscs;
297         for(dbg_csc=0; dbg_csc<sblock.fs_ncg; dbg_csc++) {
298                 snprintf(dbg_line, sizeof(dbg_line),
299                     "%d. new csum in new location", dbg_csc);
300                 DBG_DUMP_CSUM(&sblock,
301                     dbg_line,
302                     dbg_csp++);
303         }
304 }
305 #endif /* FS_DEBUG */
306
307         /*
308          * Now write the new superblock back to disk.
309          */
310         sblock.fs_time = modtime;
311         wtfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
312         DBG_PRINT0("sblock written\n");
313         DBG_DUMP_FS(&sblock,
314             "new initial sblock");
315
316         /*
317          * Clean up the dynamic fields in our superblock copies.
318          */
319         sblock.fs_fmod = 0;
320         sblock.fs_clean = 1;
321         sblock.fs_ronly = 0;
322         sblock.fs_cgrotor = 0;
323         sblock.fs_state = 0;
324         memset((void *)&sblock.fs_fsmnt, 0, sizeof(sblock.fs_fsmnt));
325         sblock.fs_flags &= FS_DOSOFTDEP;
326
327         /*
328          * XXX
329          * The following fields are currently distributed from the superblock
330          * to the copies:
331          *     fs_minfree
332          *     fs_rotdelay
333          *     fs_maxcontig
334          *     fs_maxbpg
335          *     fs_minfree,
336          *     fs_optim
337          *     fs_flags regarding SOFTPDATES
338          *
339          * We probably should rather change the summary for the cylinder group
340          * statistics here to the value of what would be in there, if the file
341          * system were created initially with the new size. Therefor we still
342          * need to find an easy way of calculating that.
343          * Possibly we can try to read the first superblock copy and apply the
344          * "diffed" stats between the old and new superblock by still copying
345          * certain parameters onto that.
346          */
347
348         /*
349          * Write out the duplicate super blocks.
350          */
351         for (cylno = 0; cylno < sblock.fs_ncg; cylno++) {
352                 wtfs(fsbtodb(&sblock, cgsblock(&sblock, cylno)),
353                     (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
354         }
355         DBG_PRINT0("sblock copies written\n");
356         DBG_DUMP_FS(&sblock,
357             "new other sblocks");
358
359         DBG_LEAVE;
360         return;
361 }
362
363 /* ************************************************************ initcg ***** */
364 /*
365  * This creates a new cylinder group structure, for more details please see
366  * the source of newfs(8), as this function is taken over almost unchanged.
367  * As this is never called for the first cylinder group, the special
368  * provisions for that case are removed here.
369  */
370 static void
371 initcg(int cylno, time_t modtime, int fso, unsigned int Nflag)
372 {
373         DBG_FUNC("initcg")
374         static caddr_t iobuf;
375         long blkno, start;
376         ufs2_daddr_t i, cbase, dmax;
377 #ifdef FSIRAND
378         struct ufs1_dinode *dp1;
379 #endif
380         struct csum *cs;
381         uint d, dupper, dlower;
382
383         if (iobuf == NULL && (iobuf = malloc(sblock.fs_bsize * 3)) == NULL)
384                 errx(37, "panic: cannot allocate I/O buffer");
385
386         /*
387          * Determine block bounds for cylinder group.
388          * Allow space for super block summary information in first
389          * cylinder group.
390          */
391         cbase = cgbase(&sblock, cylno);
392         dmax = cbase + sblock.fs_fpg;
393         if (dmax > sblock.fs_size)
394                 dmax = sblock.fs_size;
395         dlower = cgsblock(&sblock, cylno) - cbase;
396         dupper = cgdmin(&sblock, cylno) - cbase;
397         if (cylno == 0) /* XXX fscs may be relocated */
398                 dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
399         cs = &fscs[cylno];
400         memset(&acg, 0, sblock.fs_cgsize);
401         acg.cg_time = modtime;
402         acg.cg_magic = CG_MAGIC;
403         acg.cg_cgx = cylno;
404         acg.cg_niblk = sblock.fs_ipg;
405         acg.cg_initediblk = sblock.fs_ipg < 2 * INOPB(&sblock) ?
406             sblock.fs_ipg : 2 * INOPB(&sblock);
407         acg.cg_ndblk = dmax - cbase;
408         if (sblock.fs_contigsumsize > 0)
409                 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
410         start = &acg.cg_space[0] - (u_char *)(&acg.cg_firstfield);
411         if (sblock.fs_magic == FS_UFS2_MAGIC) {
412                 acg.cg_iusedoff = start;
413         } else {
414                 acg.cg_old_ncyl = sblock.fs_old_cpg;
415                 acg.cg_old_time = acg.cg_time;
416                 acg.cg_time = 0;
417                 acg.cg_old_niblk = acg.cg_niblk;
418                 acg.cg_niblk = 0;
419                 acg.cg_initediblk = 0;
420                 acg.cg_old_btotoff = start;
421                 acg.cg_old_boff = acg.cg_old_btotoff +
422                     sblock.fs_old_cpg * sizeof(int32_t);
423                 acg.cg_iusedoff = acg.cg_old_boff +
424                     sblock.fs_old_cpg * sizeof(u_int16_t);
425         }
426         acg.cg_freeoff = acg.cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
427         acg.cg_nextfreeoff = acg.cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT);
428         if (sblock.fs_contigsumsize > 0) {
429                 acg.cg_clustersumoff =
430                     roundup(acg.cg_nextfreeoff, sizeof(u_int32_t));
431                 acg.cg_clustersumoff -= sizeof(u_int32_t);
432                 acg.cg_clusteroff = acg.cg_clustersumoff +
433                     (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
434                 acg.cg_nextfreeoff = acg.cg_clusteroff +
435                     howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
436         }
437         if (acg.cg_nextfreeoff > (unsigned)sblock.fs_cgsize) {
438                 /*
439                  * This should never happen as we would have had that panic
440                  * already on file system creation
441                  */
442                 errx(37, "panic: cylinder group too big");
443         }
444         acg.cg_cs.cs_nifree += sblock.fs_ipg;
445         if (cylno == 0)
446                 for (i = 0; i < ROOTINO; i++) {
447                         setbit(cg_inosused(&acg), i);
448                         acg.cg_cs.cs_nifree--;
449                 }
450         /*
451          * For the old file system, we have to initialize all the inodes.
452          */
453         if (sblock.fs_magic == FS_UFS1_MAGIC) {
454                 bzero(iobuf, sblock.fs_bsize);
455                 for (i = 0; i < sblock.fs_ipg / INOPF(&sblock);
456                      i += sblock.fs_frag) {
457 #ifdef FSIRAND
458                         dp1 = (struct ufs1_dinode *)(void *)iobuf;
459                         for (j = 0; j < INOPB(&sblock); j++) {
460                                 dp1->di_gen = random();
461                                 dp1++;
462                         }
463 #endif
464                         wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno) + i),
465                             sblock.fs_bsize, iobuf, fso, Nflag);
466                 }
467         }
468         if (cylno > 0) {
469                 /*
470                  * In cylno 0, beginning space is reserved
471                  * for boot and super blocks.
472                  */
473                 for (d = 0; d < dlower; d += sblock.fs_frag) {
474                         blkno = d / sblock.fs_frag;
475                         setblock(&sblock, cg_blksfree(&acg), blkno);
476                         if (sblock.fs_contigsumsize > 0)
477                                 setbit(cg_clustersfree(&acg), blkno);
478                         acg.cg_cs.cs_nbfree++;
479                 }
480                 sblock.fs_dsize += dlower;
481         }
482         sblock.fs_dsize += acg.cg_ndblk - dupper;
483         if ((i = dupper % sblock.fs_frag)) {
484                 acg.cg_frsum[sblock.fs_frag - i]++;
485                 for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
486                         setbit(cg_blksfree(&acg), dupper);
487                         acg.cg_cs.cs_nffree++;
488                 }
489         }
490         for (d = dupper; d + sblock.fs_frag <= acg.cg_ndblk;
491              d += sblock.fs_frag) {
492                 blkno = d / sblock.fs_frag;
493                 setblock(&sblock, cg_blksfree(&acg), blkno);
494                 if (sblock.fs_contigsumsize > 0)
495                         setbit(cg_clustersfree(&acg), blkno);
496                 acg.cg_cs.cs_nbfree++;
497         }
498         if (d < acg.cg_ndblk) {
499                 acg.cg_frsum[acg.cg_ndblk - d]++;
500                 for (; d < acg.cg_ndblk; d++) {
501                         setbit(cg_blksfree(&acg), d);
502                         acg.cg_cs.cs_nffree++;
503                 }
504         }
505         if (sblock.fs_contigsumsize > 0) {
506                 int32_t *sump = cg_clustersum(&acg);
507                 u_char *mapp = cg_clustersfree(&acg);
508                 int map = *mapp++;
509                 int bit = 1;
510                 int run = 0;
511
512                 for (i = 0; i < acg.cg_nclusterblks; i++) {
513                         if ((map & bit) != 0)
514                                 run++;
515                         else if (run != 0) {
516                                 if (run > sblock.fs_contigsumsize)
517                                         run = sblock.fs_contigsumsize;
518                                 sump[run]++;
519                                 run = 0;
520                         }
521                         if ((i & (CHAR_BIT - 1)) != CHAR_BIT - 1)
522                                 bit <<= 1;
523                         else {
524                                 map = *mapp++;
525                                 bit = 1;
526                         }
527                 }
528                 if (run != 0) {
529                         if (run > sblock.fs_contigsumsize)
530                                 run = sblock.fs_contigsumsize;
531                         sump[run]++;
532                 }
533         }
534         sblock.fs_cstotal.cs_ndir += acg.cg_cs.cs_ndir;
535         sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
536         sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
537         sblock.fs_cstotal.cs_nifree += acg.cg_cs.cs_nifree;
538         *cs = acg.cg_cs;
539
540         memcpy(iobuf, &acg, sblock.fs_cgsize);
541         memset(iobuf + sblock.fs_cgsize, '\0',
542             sblock.fs_bsize * 3 - sblock.fs_cgsize);
543
544         wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
545             sblock.fs_bsize * 3, iobuf, fso, Nflag);
546         DBG_DUMP_CG(&sblock, "new cg", &acg);
547
548         DBG_LEAVE;
549         return;
550 }
551
552 /* ******************************************************* frag_adjust ***** */
553 /*
554  * Here we add or subtract (sign +1/-1) the available fragments in a given
555  * block to or from the fragment statistics. By subtracting before and adding
556  * after an operation on the free frag map we can easy update the fragment
557  * statistic, which seems to be otherwise a rather complex operation.
558  */
559 static void
560 frag_adjust(ufs2_daddr_t frag, int sign)
561 {
562         DBG_FUNC("frag_adjust")
563         int fragsize;
564         int f;
565
566         DBG_ENTER;
567
568         fragsize=0;
569         /*
570          * Here frag only needs to point to any fragment in the block we want
571          * to examine.
572          */
573         for(f=rounddown(frag, sblock.fs_frag);
574             f<roundup(frag+1, sblock.fs_frag);
575             f++) {
576                 /*
577                  * Count contiguous free fragments.
578                  */
579                 if(isset(cg_blksfree(&acg), f)) {
580                         fragsize++;
581                 } else {
582                         if(fragsize && fragsize<sblock.fs_frag) {
583                                 /*
584                                  * We found something in between.
585                                  */
586                                 acg.cg_frsum[fragsize]+=sign;
587                                 DBG_PRINT2("frag_adjust [%d]+=%d\n",
588                                     fragsize,
589                                     sign);
590                         }
591                         fragsize=0;
592                 }
593         }
594         if(fragsize && fragsize<sblock.fs_frag) {
595                 /*
596                  * We found something.
597                  */
598                 acg.cg_frsum[fragsize]+=sign;
599                 DBG_PRINT2("frag_adjust [%d]+=%d\n",
600                     fragsize,
601                     sign);
602         }
603         DBG_PRINT2("frag_adjust [[%d]]+=%d\n",
604             fragsize,
605             sign);
606
607         DBG_LEAVE;
608         return;
609 }
610
611 /* ******************************************************* cond_bl_upd ***** */
612 /*
613  * Here we conditionally update a pointer to a fragment. We check for all
614  * relocated blocks if any of its fragments is referenced by the current
615  * field, and update the pointer to the respective fragment in our new
616  * block.  If we find a reference we write back the block immediately,
617  * as there is no easy way for our general block reading engine to figure
618  * out if a write back operation is needed.
619  */
620 static int
621 cond_bl_upd(ufs2_daddr_t *block, struct gfs_bpp *field, int fsi, int fso,
622     unsigned int Nflag)
623 {
624         DBG_FUNC("cond_bl_upd")
625         struct gfs_bpp *f;
626         ufs2_daddr_t src, dst;
627         int fragnum;
628         void *ibuf;
629
630         DBG_ENTER;
631
632         for (f = field; f->old != 0; f++) {
633                 src = *block;
634                 if (fragstoblks(&sblock, src) != f->old)
635                         continue;
636                 /*
637                  * The fragment is part of the block, so update.
638                  */
639                 dst = blkstofrags(&sblock, f->new);
640                 fragnum = fragnum(&sblock, src);
641                 *block = dst + fragnum;
642                 f->found++;
643                 DBG_PRINT3("scg (%jd->%jd)[%d] reference updated\n",
644                     (intmax_t)f->old,
645                     (intmax_t)f->new,
646                     fragnum);
647
648                 /*
649                  * Copy the block back immediately.
650                  *
651                  * XXX  If src is from an indirect block we have
652                  *      to implement copy on write here in case of
653                  *      active snapshots.
654                  */
655                 ibuf = malloc(sblock.fs_bsize);
656                 if (!ibuf)
657                         errx(1, "malloc failed");
658                 src -= fragnum;
659                 rdfs(fsbtodb(&sblock, src), (size_t)sblock.fs_bsize, ibuf, fsi);
660                 wtfs(dst, (size_t)sblock.fs_bsize, ibuf, fso, Nflag);
661                 free(ibuf);
662                 /*
663                  * The same block can't be found again in this loop.
664                  */
665                 return (1);
666         }
667
668         DBG_LEAVE;
669         return (0);
670 }
671
672 /* ************************************************************ updjcg ***** */
673 /*
674  * Here we do all needed work for the former last cylinder group. It has to be
675  * changed in any case, even if the file system ended exactly on the end of
676  * this group, as there is some slightly inconsistent handling of the number
677  * of cylinders in the cylinder group. We start again by reading the cylinder
678  * group from disk. If the last block was not fully available, we first handle
679  * the missing fragments, then we handle all new full blocks in that file
680  * system and finally we handle the new last fragmented block in the file
681  * system.  We again have to handle the fragment statistics rotational layout
682  * tables and cluster summary during all those operations.
683  */
684 static void
685 updjcg(int cylno, time_t modtime, int fsi, int fso, unsigned int Nflag)
686 {
687         DBG_FUNC("updjcg")
688         ufs2_daddr_t    cbase, dmax, dupper;
689         struct csum     *cs;
690         int     i,k;
691         int     j=0;
692
693         DBG_ENTER;
694
695         /*
696          * Read the former last (joining) cylinder group from disk, and make
697          * a copy.
698          */
699         rdfs(fsbtodb(&osblock, cgtod(&osblock, cylno)),
700             (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
701         DBG_PRINT0("jcg read\n");
702         DBG_DUMP_CG(&sblock,
703             "old joining cg",
704             &aocg);
705
706         memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
707
708         /*
709          * If the cylinder group had already its new final size almost
710          * nothing is to be done ... except:
711          * For some reason the value of cg_ncyl in the last cylinder group has
712          * to be zero instead of fs_cpg. As this is now no longer the last
713          * cylinder group we have to change that value now to fs_cpg.
714          */
715
716         if(cgbase(&osblock, cylno+1) == osblock.fs_size) {
717                 if (sblock.fs_magic == FS_UFS1_MAGIC)
718                         acg.cg_old_ncyl=sblock.fs_old_cpg;
719
720                 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
721                     (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
722                 DBG_PRINT0("jcg written\n");
723                 DBG_DUMP_CG(&sblock,
724                     "new joining cg",
725                     &acg);
726
727                 DBG_LEAVE;
728                 return;
729         }
730
731         /*
732          * Set up some variables needed later.
733          */
734         cbase = cgbase(&sblock, cylno);
735         dmax = cbase + sblock.fs_fpg;
736         if (dmax > sblock.fs_size)
737                 dmax = sblock.fs_size;
738         dupper = cgdmin(&sblock, cylno) - cbase;
739         if (cylno == 0) { /* XXX fscs may be relocated */
740                 dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
741         }
742
743         /*
744          * Set pointer to the cylinder summary for our cylinder group.
745          */
746         cs = fscs + cylno;
747
748         /*
749          * Touch the cylinder group, update all fields in the cylinder group as
750          * needed, update the free space in the superblock.
751          */
752         acg.cg_time = modtime;
753         if ((unsigned)cylno == sblock.fs_ncg - 1) {
754                 /*
755                  * This is still the last cylinder group.
756                  */
757                 if (sblock.fs_magic == FS_UFS1_MAGIC)
758                         acg.cg_old_ncyl =
759                             sblock.fs_old_ncyl % sblock.fs_old_cpg;
760         } else {
761                 acg.cg_old_ncyl = sblock.fs_old_cpg;
762         }
763         DBG_PRINT2("jcg dbg: %d %u",
764             cylno,
765             sblock.fs_ncg);
766 #ifdef FS_DEBUG
767         if (sblock.fs_magic == FS_UFS1_MAGIC)
768                 DBG_PRINT2("%d %u",
769                     acg.cg_old_ncyl,
770                     sblock.fs_old_cpg);
771 #endif
772         DBG_PRINT0("\n");
773         acg.cg_ndblk = dmax - cbase;
774         sblock.fs_dsize += acg.cg_ndblk-aocg.cg_ndblk;
775         if (sblock.fs_contigsumsize > 0) {
776                 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
777         }
778
779         /*
780          * Now we have to update the free fragment bitmap for our new free
781          * space.  There again we have to handle the fragmentation and also
782          * the rotational layout tables and the cluster summary.  This is
783          * also done per fragment for the first new block if the old file
784          * system end was not on a block boundary, per fragment for the new
785          * last block if the new file system end is not on a block boundary,
786          * and per block for all space in between.
787          *
788          * Handle the first new block here if it was partially available
789          * before.
790          */
791         if(osblock.fs_size % sblock.fs_frag) {
792                 if(roundup(osblock.fs_size, sblock.fs_frag)<=sblock.fs_size) {
793                         /*
794                          * The new space is enough to fill at least this
795                          * block
796                          */
797                         j=0;
798                         for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag)-1;
799                             i>=osblock.fs_size-cbase;
800                             i--) {
801                                 setbit(cg_blksfree(&acg), i);
802                                 acg.cg_cs.cs_nffree++;
803                                 j++;
804                         }
805
806                         /*
807                          * Check if the fragment just created could join an
808                          * already existing fragment at the former end of the
809                          * file system.
810                          */
811                         if(isblock(&sblock, cg_blksfree(&acg),
812                             ((osblock.fs_size - cgbase(&sblock, cylno))/
813                             sblock.fs_frag))) {
814                                 /*
815                                  * The block is now completely available.
816                                  */
817                                 DBG_PRINT0("block was\n");
818                                 acg.cg_frsum[osblock.fs_size%sblock.fs_frag]--;
819                                 acg.cg_cs.cs_nbfree++;
820                                 acg.cg_cs.cs_nffree-=sblock.fs_frag;
821                                 k=rounddown(osblock.fs_size-cbase,
822                                     sblock.fs_frag);
823                                 updclst((osblock.fs_size-cbase)/sblock.fs_frag);
824                         } else {
825                                 /*
826                                  * Lets rejoin a possible partially growed
827                                  * fragment.
828                                  */
829                                 k=0;
830                                 while(isset(cg_blksfree(&acg), i) &&
831                                     (i>=rounddown(osblock.fs_size-cbase,
832                                     sblock.fs_frag))) {
833                                         i--;
834                                         k++;
835                                 }
836                                 if(k) {
837                                         acg.cg_frsum[k]--;
838                                 }
839                                 acg.cg_frsum[k+j]++;
840                         }
841                 } else {
842                         /*
843                          * We only grow by some fragments within this last
844                          * block.
845                          */
846                         for(i=sblock.fs_size-cbase-1;
847                                 i>=osblock.fs_size-cbase;
848                                 i--) {
849                                 setbit(cg_blksfree(&acg), i);
850                                 acg.cg_cs.cs_nffree++;
851                                 j++;
852                         }
853                         /*
854                          * Lets rejoin a possible partially growed fragment.
855                          */
856                         k=0;
857                         while(isset(cg_blksfree(&acg), i) &&
858                             (i>=rounddown(osblock.fs_size-cbase,
859                             sblock.fs_frag))) {
860                                 i--;
861                                 k++;
862                         }
863                         if(k) {
864                                 acg.cg_frsum[k]--;
865                         }
866                         acg.cg_frsum[k+j]++;
867                 }
868         }
869
870         /*
871          * Handle all new complete blocks here.
872          */
873         for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag);
874             i+sblock.fs_frag<=dmax-cbase;       /* XXX <= or only < ? */
875             i+=sblock.fs_frag) {
876                 j = i / sblock.fs_frag;
877                 setblock(&sblock, cg_blksfree(&acg), j);
878                 updclst(j);
879                 acg.cg_cs.cs_nbfree++;
880         }
881
882         /*
883          * Handle the last new block if there are stll some new fragments left.
884          * Here we don't have to bother about the cluster summary or the even
885          * the rotational layout table.
886          */
887         if (i < (dmax - cbase)) {
888                 acg.cg_frsum[dmax - cbase - i]++;
889                 for (; i < dmax - cbase; i++) {
890                         setbit(cg_blksfree(&acg), i);
891                         acg.cg_cs.cs_nffree++;
892                 }
893         }
894
895         sblock.fs_cstotal.cs_nffree +=
896             (acg.cg_cs.cs_nffree - aocg.cg_cs.cs_nffree);
897         sblock.fs_cstotal.cs_nbfree +=
898             (acg.cg_cs.cs_nbfree - aocg.cg_cs.cs_nbfree);
899         /*
900          * The following statistics are not changed here:
901          *     sblock.fs_cstotal.cs_ndir
902          *     sblock.fs_cstotal.cs_nifree
903          * As the statistics for this cylinder group are ready, copy it to
904          * the summary information array.
905          */
906         *cs = acg.cg_cs;
907
908         /*
909          * Write the updated "joining" cylinder group back to disk.
910          */
911         wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize,
912             (void *)&acg, fso, Nflag);
913         DBG_PRINT0("jcg written\n");
914         DBG_DUMP_CG(&sblock,
915             "new joining cg",
916             &acg);
917
918         DBG_LEAVE;
919         return;
920 }
921
922 /* ********************************************************** updcsloc ***** */
923 /*
924  * Here we update the location of the cylinder summary. We have two possible
925  * ways of growing the cylinder summary.
926  * (1)  We can try to grow the summary in the current location, and relocate
927  *      possibly used blocks within the current cylinder group.
928  * (2)  Alternatively we can relocate the whole cylinder summary to the first
929  *      new completely empty cylinder group. Once the cylinder summary is no
930  *      longer in the beginning of the first cylinder group you should never
931  *      use a version of fsck which is not aware of the possibility to have
932  *      this structure in a non standard place.
933  * Option (1) is considered to be less intrusive to the structure of the file-
934  * system. So we try to stick to that whenever possible. If there is not enough
935  * space in the cylinder group containing the cylinder summary we have to use
936  * method (2). In case of active snapshots in the file system we probably can
937  * completely avoid implementing copy on write if we stick to method (2) only.
938  */
939 static void
940 updcsloc(time_t modtime, int fsi, int fso, unsigned int Nflag)
941 {
942         DBG_FUNC("updcsloc")
943         struct csum     *cs;
944         int     ocscg, ncscg;
945         int     blocks;
946         ufs2_daddr_t    cbase, dupper, odupper, d, f, g;
947         int     ind, inc;
948         uint    cylno;
949         struct gfs_bpp  *bp;
950         int     i, l;
951         int     lcs=0;
952         int     block;
953
954         DBG_ENTER;
955
956         if(howmany(sblock.fs_cssize, sblock.fs_fsize) ==
957             howmany(osblock.fs_cssize, osblock.fs_fsize)) {
958                 /*
959                  * No new fragment needed.
960                  */
961                 DBG_LEAVE;
962                 return;
963         }
964         ocscg=dtog(&osblock, osblock.fs_csaddr);
965         cs=fscs+ocscg;
966         blocks = 1+howmany(sblock.fs_cssize, sblock.fs_bsize)-
967             howmany(osblock.fs_cssize, osblock.fs_bsize);
968
969         /*
970          * Read original cylinder group from disk, and make a copy.
971          * XXX  If Nflag is set in some very rare cases we now miss
972          *      some changes done in updjcg by reading the unmodified
973          *      block from disk.
974          */
975         rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)),
976             (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
977         DBG_PRINT0("oscg read\n");
978         DBG_DUMP_CG(&sblock,
979             "old summary cg",
980             &aocg);
981
982         memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
983
984         /*
985          * Touch the cylinder group, set up local variables needed later
986          * and update the superblock.
987          */
988         acg.cg_time = modtime;
989
990         /*
991          * XXX  In the case of having active snapshots we may need much more
992          *      blocks for the copy on write. We need each block twice, and
993          *      also up to 8*3 blocks for indirect blocks for all possible
994          *      references.
995          */
996         if(/*((int)sblock.fs_time&0x3)>0||*/ cs->cs_nbfree < blocks) {
997                 /*
998                  * There is not enough space in the old cylinder group to
999                  * relocate all blocks as needed, so we relocate the whole
1000                  * cylinder group summary to a new group. We try to use the
1001                  * first complete new cylinder group just created. Within the
1002                  * cylinder group we align the area immediately after the
1003                  * cylinder group information location in order to be as
1004                  * close as possible to the original implementation of ffs.
1005                  *
1006                  * First we have to make sure we'll find enough space in the
1007                  * new cylinder group. If not, then we currently give up.
1008                  * We start with freeing everything which was used by the
1009                  * fragments of the old cylinder summary in the current group.
1010                  * Now we write back the group meta data, read in the needed
1011                  * meta data from the new cylinder group, and start allocating
1012                  * within that group. Here we can assume, the group to be
1013                  * completely empty. Which makes the handling of fragments and
1014                  * clusters a lot easier.
1015                  */
1016                 DBG_TRC;
1017                 if(sblock.fs_ncg-osblock.fs_ncg < 2) {
1018                         errx(2, "panic: not enough space");
1019                 }
1020
1021                 /*
1022                  * Point "d" to the first fragment not used by the cylinder
1023                  * summary.
1024                  */
1025                 d=osblock.fs_csaddr+(osblock.fs_cssize/osblock.fs_fsize);
1026
1027                 /*
1028                  * Set up last cluster size ("lcs") already here. Calculate
1029                  * the size for the trailing cluster just behind where "d"
1030                  * points to.
1031                  */
1032                 if(sblock.fs_contigsumsize > 0) {
1033                         for(block=howmany(d%sblock.fs_fpg, sblock.fs_frag),
1034                             lcs=0; lcs<sblock.fs_contigsumsize;
1035                             block++, lcs++) {
1036                                 if(isclr(cg_clustersfree(&acg), block)){
1037                                         break;
1038                                 }
1039                         }
1040                 }
1041
1042                 /*
1043                  * Point "d" to the last frag used by the cylinder summary.
1044                  */
1045                 d--;
1046
1047                 DBG_PRINT1("d=%jd\n",
1048                     (intmax_t)d);
1049                 if((d+1)%sblock.fs_frag) {
1050                         /*
1051                          * The end of the cylinder summary is not a complete
1052                          * block.
1053                          */
1054                         DBG_TRC;
1055                         frag_adjust(d%sblock.fs_fpg, -1);
1056                         for(; (d+1)%sblock.fs_frag; d--) {
1057                                 DBG_PRINT1("d=%jd\n",
1058                                     (intmax_t)d);
1059                                 setbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1060                                 acg.cg_cs.cs_nffree++;
1061                                 sblock.fs_cstotal.cs_nffree++;
1062                         }
1063                         /*
1064                          * Point "d" to the last fragment of the last
1065                          * (incomplete) block of the cylinder summary.
1066                          */
1067                         d++;
1068                         frag_adjust(d%sblock.fs_fpg, 1);
1069
1070                         if(isblock(&sblock, cg_blksfree(&acg),
1071                             (d%sblock.fs_fpg)/sblock.fs_frag)) {
1072                                 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1073                                 acg.cg_cs.cs_nffree-=sblock.fs_frag;
1074                                 acg.cg_cs.cs_nbfree++;
1075                                 sblock.fs_cstotal.cs_nffree-=sblock.fs_frag;
1076                                 sblock.fs_cstotal.cs_nbfree++;
1077                                 if(sblock.fs_contigsumsize > 0) {
1078                                         setbit(cg_clustersfree(&acg),
1079                                             (d%sblock.fs_fpg)/sblock.fs_frag);
1080                                         if(lcs < sblock.fs_contigsumsize) {
1081                                                 if(lcs) {
1082                                                         cg_clustersum(&acg)
1083                                                             [lcs]--;
1084                                                 }
1085                                                 lcs++;
1086                                                 cg_clustersum(&acg)[lcs]++;
1087                                         }
1088                                 }
1089                         }
1090                         /*
1091                          * Point "d" to the first fragment of the block before
1092                          * the last incomplete block.
1093                          */
1094                         d--;
1095                 }
1096
1097                 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1098                 for(d=rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr;
1099                     d-=sblock.fs_frag) {
1100                         DBG_TRC;
1101                         DBG_PRINT1("d=%jd\n", (intmax_t)d);
1102                         setblock(&sblock, cg_blksfree(&acg),
1103                             (d%sblock.fs_fpg)/sblock.fs_frag);
1104                         acg.cg_cs.cs_nbfree++;
1105                         sblock.fs_cstotal.cs_nbfree++;
1106                         if(sblock.fs_contigsumsize > 0) {
1107                                 setbit(cg_clustersfree(&acg),
1108                                     (d%sblock.fs_fpg)/sblock.fs_frag);
1109                                 /*
1110                                  * The last cluster size is already set up.
1111                                  */
1112                                 if(lcs < sblock.fs_contigsumsize) {
1113                                         if(lcs) {
1114                                                 cg_clustersum(&acg)[lcs]--;
1115                                         }
1116                                         lcs++;
1117                                         cg_clustersum(&acg)[lcs]++;
1118                                 }
1119                         }
1120                 }
1121                 *cs = acg.cg_cs;
1122
1123                 /*
1124                  * Now write the former cylinder group containing the cylinder
1125                  * summary back to disk.
1126                  */
1127                 wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)),
1128                     (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1129                 DBG_PRINT0("oscg written\n");
1130                 DBG_DUMP_CG(&sblock,
1131                     "old summary cg",
1132                     &acg);
1133
1134                 /*
1135                  * Find the beginning of the new cylinder group containing the
1136                  * cylinder summary.
1137                  */
1138                 sblock.fs_csaddr=cgdmin(&sblock, osblock.fs_ncg);
1139                 ncscg=dtog(&sblock, sblock.fs_csaddr);
1140                 cs=fscs+ncscg;
1141
1142
1143                 /*
1144                  * If Nflag is specified, we would now read random data instead
1145                  * of an empty cg structure from disk. So we can't simulate that
1146                  * part for now.
1147                  */
1148                 if(Nflag) {
1149                         DBG_PRINT0("nscg update skipped\n");
1150                         DBG_LEAVE;
1151                         return;
1152                 }
1153
1154                 /*
1155                  * Read the future cylinder group containing the cylinder
1156                  * summary from disk, and make a copy.
1157                  */
1158                 rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1159                     (size_t)sblock.fs_cgsize, (void *)&aocg, fsi);
1160                 DBG_PRINT0("nscg read\n");
1161                 DBG_DUMP_CG(&sblock,
1162                     "new summary cg",
1163                     &aocg);
1164
1165                 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
1166
1167                 /*
1168                  * Allocate all complete blocks used by the new cylinder
1169                  * summary.
1170                  */
1171                 for(d=sblock.fs_csaddr; d+sblock.fs_frag <=
1172                     sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize);
1173                     d+=sblock.fs_frag) {
1174                         clrblock(&sblock, cg_blksfree(&acg),
1175                             (d%sblock.fs_fpg)/sblock.fs_frag);
1176                         acg.cg_cs.cs_nbfree--;
1177                         sblock.fs_cstotal.cs_nbfree--;
1178                         if(sblock.fs_contigsumsize > 0) {
1179                                 clrbit(cg_clustersfree(&acg),
1180                                     (d%sblock.fs_fpg)/sblock.fs_frag);
1181                         }
1182                 }
1183
1184                 /*
1185                  * Allocate all fragments used by the cylinder summary in the
1186                  * last block.
1187                  */
1188                 if(d<sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize)) {
1189                         for(; d-sblock.fs_csaddr<
1190                             sblock.fs_cssize/sblock.fs_fsize;
1191                             d++) {
1192                                 clrbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1193                                 acg.cg_cs.cs_nffree--;
1194                                 sblock.fs_cstotal.cs_nffree--;
1195                         }
1196                         acg.cg_cs.cs_nbfree--;
1197                         acg.cg_cs.cs_nffree+=sblock.fs_frag;
1198                         sblock.fs_cstotal.cs_nbfree--;
1199                         sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1200                         if(sblock.fs_contigsumsize > 0) {
1201                                 clrbit(cg_clustersfree(&acg),
1202                                     (d%sblock.fs_fpg)/sblock.fs_frag);
1203                         }
1204
1205                         frag_adjust(d%sblock.fs_fpg, +1);
1206                 }
1207                 /*
1208                  * XXX  Handle the cluster statistics here in the case this
1209                  *      cylinder group is now almost full, and the remaining
1210                  *      space is less then the maximum cluster size. This is
1211                  *      probably not needed, as you would hardly find a file
1212                  *      system which has only MAXCSBUFS+FS_MAXCONTIG of free
1213                  *      space right behind the cylinder group information in
1214                  *      any new cylinder group.
1215                  */
1216
1217                 /*
1218                  * Update our statistics in the cylinder summary.
1219                  */
1220                 *cs = acg.cg_cs;
1221
1222                 /*
1223                  * Write the new cylinder group containing the cylinder summary
1224                  * back to disk.
1225                  */
1226                 wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1227                     (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1228                 DBG_PRINT0("nscg written\n");
1229                 DBG_DUMP_CG(&sblock,
1230                     "new summary cg",
1231                     &acg);
1232
1233                 DBG_LEAVE;
1234                 return;
1235         }
1236         /*
1237          * We have got enough of space in the current cylinder group, so we
1238          * can relocate just a few blocks, and let the summary information
1239          * grow in place where it is right now.
1240          */
1241         DBG_TRC;
1242
1243         cbase = cgbase(&osblock, ocscg);        /* old and new are equal */
1244         dupper = sblock.fs_csaddr - cbase +
1245             howmany(sblock.fs_cssize, sblock.fs_fsize);
1246         odupper = osblock.fs_csaddr - cbase +
1247             howmany(osblock.fs_cssize, osblock.fs_fsize);
1248
1249         sblock.fs_dsize -= dupper-odupper;
1250
1251         /*
1252          * Allocate the space for the array of blocks to be relocated.
1253          */
1254         bp=(struct gfs_bpp *)malloc(((dupper-odupper)/sblock.fs_frag+2)*
1255             sizeof(struct gfs_bpp));
1256         if(bp == NULL) {
1257                 errx(1, "malloc failed");
1258         }
1259         memset((char *)bp, 0, ((dupper-odupper)/sblock.fs_frag+2)*
1260             sizeof(struct gfs_bpp));
1261
1262         /*
1263          * Lock all new frags needed for the cylinder group summary. This is
1264          * done per fragment in the first and last block of the new required
1265          * area, and per block for all other blocks.
1266          *
1267          * Handle the first new block here (but only if some fragments where
1268          * already used for the cylinder summary).
1269          */
1270         ind=0;
1271         frag_adjust(odupper, -1);
1272         for(d=odupper; ((d<dupper)&&(d%sblock.fs_frag)); d++) {
1273                 DBG_PRINT1("scg first frag check loop d=%jd\n",
1274                     (intmax_t)d);
1275                 if(isclr(cg_blksfree(&acg), d)) {
1276                         if (!ind) {
1277                                 bp[ind].old=d/sblock.fs_frag;
1278                                 bp[ind].flags|=GFS_FL_FIRST;
1279                                 if(roundup(d, sblock.fs_frag) >= dupper) {
1280                                         bp[ind].flags|=GFS_FL_LAST;
1281                                 }
1282                                 ind++;
1283                         }
1284                 } else {
1285                         clrbit(cg_blksfree(&acg), d);
1286                         acg.cg_cs.cs_nffree--;
1287                         sblock.fs_cstotal.cs_nffree--;
1288                 }
1289                 /*
1290                  * No cluster handling is needed here, as there was at least
1291                  * one fragment in use by the cylinder summary in the old
1292                  * file system.
1293                  * No block-free counter handling here as this block was not
1294                  * a free block.
1295                  */
1296         }
1297         frag_adjust(odupper, 1);
1298
1299         /*
1300          * Handle all needed complete blocks here.
1301          */
1302         for(; d+sblock.fs_frag<=dupper; d+=sblock.fs_frag) {
1303                 DBG_PRINT1("scg block check loop d=%jd\n",
1304                     (intmax_t)d);
1305                 if(!isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1306                         for(f=d; f<d+sblock.fs_frag; f++) {
1307                                 if(isset(cg_blksfree(&aocg), f)) {
1308                                         acg.cg_cs.cs_nffree--;
1309                                         sblock.fs_cstotal.cs_nffree--;
1310                                 }
1311                         }
1312                         clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1313                         bp[ind].old=d/sblock.fs_frag;
1314                         ind++;
1315                 } else {
1316                         clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1317                         acg.cg_cs.cs_nbfree--;
1318                         sblock.fs_cstotal.cs_nbfree--;
1319                         if(sblock.fs_contigsumsize > 0) {
1320                                 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1321                                 for(lcs=0, l=(d/sblock.fs_frag)+1;
1322                                     lcs<sblock.fs_contigsumsize;
1323                                     l++, lcs++ ) {
1324                                         if(isclr(cg_clustersfree(&acg),l)){
1325                                                 break;
1326                                         }
1327                                 }
1328                                 if(lcs < sblock.fs_contigsumsize) {
1329                                         cg_clustersum(&acg)[lcs+1]--;
1330                                         if(lcs) {
1331                                                 cg_clustersum(&acg)[lcs]++;
1332                                         }
1333                                 }
1334                         }
1335                 }
1336                 /*
1337                  * No fragment counter handling is needed here, as this finally
1338                  * doesn't change after the relocation.
1339                  */
1340         }
1341
1342         /*
1343          * Handle all fragments needed in the last new affected block.
1344          */
1345         if(d<dupper) {
1346                 frag_adjust(dupper-1, -1);
1347
1348                 if(isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1349                         acg.cg_cs.cs_nbfree--;
1350                         sblock.fs_cstotal.cs_nbfree--;
1351                         acg.cg_cs.cs_nffree+=sblock.fs_frag;
1352                         sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1353                         if(sblock.fs_contigsumsize > 0) {
1354                                 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1355                                 for(lcs=0, l=(d/sblock.fs_frag)+1;
1356                                     lcs<sblock.fs_contigsumsize;
1357                                     l++, lcs++ ) {
1358                                         if(isclr(cg_clustersfree(&acg),l)){
1359                                                 break;
1360                                         }
1361                                 }
1362                                 if(lcs < sblock.fs_contigsumsize) {
1363                                         cg_clustersum(&acg)[lcs+1]--;
1364                                         if(lcs) {
1365                                                 cg_clustersum(&acg)[lcs]++;
1366                                         }
1367                                 }
1368                         }
1369                 }
1370
1371                 for(; d<dupper; d++) {
1372                         DBG_PRINT1("scg second frag check loop d=%jd\n",
1373                             (intmax_t)d);
1374                         if(isclr(cg_blksfree(&acg), d)) {
1375                                 bp[ind].old=d/sblock.fs_frag;
1376                                 bp[ind].flags|=GFS_FL_LAST;
1377                         } else {
1378                                 clrbit(cg_blksfree(&acg), d);
1379                                 acg.cg_cs.cs_nffree--;
1380                                 sblock.fs_cstotal.cs_nffree--;
1381                         }
1382                 }
1383                 if(bp[ind].flags & GFS_FL_LAST) { /* we have to advance here */
1384                         ind++;
1385                 }
1386                 frag_adjust(dupper-1, 1);
1387         }
1388
1389         /*
1390          * If we found a block to relocate just do so.
1391          */
1392         if(ind) {
1393                 for(i=0; i<ind; i++) {
1394                         if(!bp[i].old) { /* no more blocks listed */
1395                                 /*
1396                                  * XXX  A relative blocknumber should not be
1397                                  *      zero, which is not explicitly
1398                                  *      guaranteed by our code.
1399                                  */
1400                                 break;
1401                         }
1402                         /*
1403                          * Allocate a complete block in the same (current)
1404                          * cylinder group.
1405                          */
1406                         bp[i].new=alloc()/sblock.fs_frag;
1407
1408                         /*
1409                          * There is no frag_adjust() needed for the new block
1410                          * as it will have no fragments yet :-).
1411                          */
1412                         for(f=bp[i].old*sblock.fs_frag,
1413                             g=bp[i].new*sblock.fs_frag;
1414                             f<(bp[i].old+1)*sblock.fs_frag;
1415                             f++, g++) {
1416                                 if(isset(cg_blksfree(&aocg), f)) {
1417                                         setbit(cg_blksfree(&acg), g);
1418                                         acg.cg_cs.cs_nffree++;
1419                                         sblock.fs_cstotal.cs_nffree++;
1420                                 }
1421                         }
1422
1423                         /*
1424                          * Special handling is required if this was the first
1425                          * block. We have to consider the fragments which were
1426                          * used by the cylinder summary in the original block
1427                          * which re to be free in the copy of our block.  We
1428                          * have to be careful if this first block happens to
1429                          * be also the last block to be relocated.
1430                          */
1431                         if(bp[i].flags & GFS_FL_FIRST) {
1432                                 for(f=bp[i].old*sblock.fs_frag,
1433                                     g=bp[i].new*sblock.fs_frag;
1434                                     f<odupper;
1435                                     f++, g++) {
1436                                         setbit(cg_blksfree(&acg), g);
1437                                         acg.cg_cs.cs_nffree++;
1438                                         sblock.fs_cstotal.cs_nffree++;
1439                                 }
1440                                 if(!(bp[i].flags & GFS_FL_LAST)) {
1441                                         frag_adjust(bp[i].new*sblock.fs_frag,1);
1442                                 }
1443                         }
1444
1445                         /*
1446                          * Special handling is required if this is the last
1447                          * block to be relocated.
1448                          */
1449                         if(bp[i].flags & GFS_FL_LAST) {
1450                                 frag_adjust(bp[i].new*sblock.fs_frag, 1);
1451                                 frag_adjust(bp[i].old*sblock.fs_frag, -1);
1452                                 for(f=dupper;
1453                                     f<roundup(dupper, sblock.fs_frag);
1454                                     f++) {
1455                                         if(isclr(cg_blksfree(&acg), f)) {
1456                                                 setbit(cg_blksfree(&acg), f);
1457                                                 acg.cg_cs.cs_nffree++;
1458                                                 sblock.fs_cstotal.cs_nffree++;
1459                                         }
1460                                 }
1461                                 frag_adjust(bp[i].old*sblock.fs_frag, 1);
1462                         }
1463
1464                         /*
1465                          * !!! Attach the cylindergroup offset here.
1466                          */
1467                         bp[i].old+=cbase/sblock.fs_frag;
1468                         bp[i].new+=cbase/sblock.fs_frag;
1469
1470                         /*
1471                          * Copy the content of the block.
1472                          */
1473                         /*
1474                          * XXX  Here we will have to implement a copy on write
1475                          *      in the case we have any active snapshots.
1476                          */
1477                         rdfs(fsbtodb(&sblock, bp[i].old*sblock.fs_frag),
1478                             (size_t)sblock.fs_bsize, (void *)&ablk, fsi);
1479                         wtfs(fsbtodb(&sblock, bp[i].new*sblock.fs_frag),
1480                             (size_t)sblock.fs_bsize, (void *)&ablk, fso, Nflag);
1481                         DBG_DUMP_HEX(&sblock,
1482                             "copied full block",
1483                             (unsigned char *)&ablk);
1484
1485                         DBG_PRINT2("scg (%jd->%jd) block relocated\n",
1486                             (intmax_t)bp[i].old,
1487                             (intmax_t)bp[i].new);
1488                 }
1489
1490                 /*
1491                  * Now we have to update all references to any fragment which
1492                  * belongs to any block relocated. We iterate now over all
1493                  * cylinder groups, within those over all non zero length
1494                  * inodes.
1495                  */
1496                 for(cylno=0; cylno<osblock.fs_ncg; cylno++) {
1497                         DBG_PRINT1("scg doing cg (%d)\n",
1498                             cylno);
1499                         for(inc=osblock.fs_ipg-1 ; inc>0 ; inc--) {
1500                                 updrefs(cylno, (ino_t)inc, bp, fsi, fso, Nflag);
1501                         }
1502                 }
1503
1504                 /*
1505                  * All inodes are checked, now make sure the number of
1506                  * references found make sense.
1507                  */
1508                 for(i=0; i<ind; i++) {
1509                         if(!bp[i].found || (bp[i].found>sblock.fs_frag)) {
1510                                 warnx("error: %jd refs found for block %jd.",
1511                                     (intmax_t)bp[i].found, (intmax_t)bp[i].old);
1512                         }
1513
1514                 }
1515         }
1516         /*
1517          * The following statistics are not changed here:
1518          *     sblock.fs_cstotal.cs_ndir
1519          *     sblock.fs_cstotal.cs_nifree
1520          * The following statistics were already updated on the fly:
1521          *     sblock.fs_cstotal.cs_nffree
1522          *     sblock.fs_cstotal.cs_nbfree
1523          * As the statistics for this cylinder group are ready, copy it to
1524          * the summary information array.
1525          */
1526
1527         *cs = acg.cg_cs;
1528
1529         /*
1530          * Write summary cylinder group back to disk.
1531          */
1532         wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)), (size_t)sblock.fs_cgsize,
1533             (void *)&acg, fso, Nflag);
1534         DBG_PRINT0("scg written\n");
1535         DBG_DUMP_CG(&sblock,
1536             "new summary cg",
1537             &acg);
1538
1539         DBG_LEAVE;
1540         return;
1541 }
1542
1543 /* ************************************************************** rdfs ***** */
1544 /*
1545  * Here we read some block(s) from disk.
1546  */
1547 static void
1548 rdfs(ufs2_daddr_t bno, size_t size, void *bf, int fsi)
1549 {
1550         DBG_FUNC("rdfs")
1551         ssize_t n;
1552
1553         DBG_ENTER;
1554
1555         if (bno < 0) {
1556                 err(32, "rdfs: attempting to read negative block number");
1557         }
1558         if (lseek(fsi, (off_t)bno * DEV_BSIZE, 0) < 0) {
1559                 err(33, "rdfs: seek error: %jd", (intmax_t)bno);
1560         }
1561         n = read(fsi, bf, size);
1562         if (n != (ssize_t)size) {
1563                 err(34, "rdfs: read error: %jd", (intmax_t)bno);
1564         }
1565
1566         DBG_LEAVE;
1567         return;
1568 }
1569
1570 /* ************************************************************** wtfs ***** */
1571 /*
1572  * Here we write some block(s) to disk.
1573  */
1574 static void
1575 wtfs(ufs2_daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag)
1576 {
1577         DBG_FUNC("wtfs")
1578         ssize_t n;
1579
1580         DBG_ENTER;
1581
1582         if (Nflag) {
1583                 DBG_LEAVE;
1584                 return;
1585         }
1586         if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0) {
1587                 err(35, "wtfs: seek error: %ld", (long)bno);
1588         }
1589         n = write(fso, bf, size);
1590         if (n != (ssize_t)size) {
1591                 err(36, "wtfs: write error: %ld", (long)bno);
1592         }
1593
1594         DBG_LEAVE;
1595         return;
1596 }
1597
1598 /* ************************************************************* alloc ***** */
1599 /*
1600  * Here we allocate a free block in the current cylinder group. It is assumed,
1601  * that acg contains the current cylinder group. As we may take a block from
1602  * somewhere in the file system we have to handle cluster summary here.
1603  */
1604 static ufs2_daddr_t
1605 alloc(void)
1606 {
1607         DBG_FUNC("alloc")
1608         ufs2_daddr_t    d, blkno;
1609         int     lcs1, lcs2;
1610         int     l;
1611         int     csmin, csmax;
1612         int     dlower, dupper, dmax;
1613
1614         DBG_ENTER;
1615
1616         if (acg.cg_magic != CG_MAGIC) {
1617                 warnx("acg: bad magic number");
1618                 DBG_LEAVE;
1619                 return (0);
1620         }
1621         if (acg.cg_cs.cs_nbfree == 0) {
1622                 warnx("error: cylinder group ran out of space");
1623                 DBG_LEAVE;
1624                 return (0);
1625         }
1626         /*
1627          * We start seeking for free blocks only from the space available after
1628          * the end of the new grown cylinder summary. Otherwise we allocate a
1629          * block here which we have to relocate a couple of seconds later again
1630          * again, and we are not prepared to to this anyway.
1631          */
1632         blkno=-1;
1633         dlower=cgsblock(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1634         dupper=cgdmin(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1635         dmax=cgbase(&sblock, acg.cg_cgx)+sblock.fs_fpg;
1636         if (dmax > sblock.fs_size) {
1637                 dmax = sblock.fs_size;
1638         }
1639         dmax-=cgbase(&sblock, acg.cg_cgx); /* retransform into cg */
1640         csmin=sblock.fs_csaddr-cgbase(&sblock, acg.cg_cgx);
1641         csmax=csmin+howmany(sblock.fs_cssize, sblock.fs_fsize);
1642         DBG_PRINT3("seek range: dl=%d, du=%d, dm=%d\n",
1643             dlower,
1644             dupper,
1645             dmax);
1646         DBG_PRINT2("range cont: csmin=%d, csmax=%d\n",
1647             csmin,
1648             csmax);
1649
1650         for(d=0; (d<dlower && blkno==-1); d+=sblock.fs_frag) {
1651                 if(d>=csmin && d<=csmax) {
1652                         continue;
1653                 }
1654                 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1655                     d))) {
1656                         blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1657                         break;
1658                 }
1659         }
1660         for(d=dupper; (d<dmax && blkno==-1); d+=sblock.fs_frag) {
1661                 if(d>=csmin && d<=csmax) {
1662                         continue;
1663                 }
1664                 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1665                     d))) {
1666                         blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1667                         break;
1668                 }
1669         }
1670         if(blkno==-1) {
1671                 warnx("internal error: couldn't find promised block in cg");
1672                 DBG_LEAVE;
1673                 return (0);
1674         }
1675
1676         /*
1677          * This is needed if the block was found already in the first loop.
1678          */
1679         d=blkstofrags(&sblock, blkno);
1680
1681         clrblock(&sblock, cg_blksfree(&acg), blkno);
1682         if (sblock.fs_contigsumsize > 0) {
1683                 /*
1684                  * Handle the cluster allocation bitmap.
1685                  */
1686                 clrbit(cg_clustersfree(&acg), blkno);
1687                 /*
1688                  * We possibly have split a cluster here, so we have to do
1689                  * recalculate the sizes of the remaining cluster halves now,
1690                  * and use them for updating the cluster summary information.
1691                  *
1692                  * Lets start with the blocks before our allocated block ...
1693                  */
1694                 for(lcs1=0, l=blkno-1; lcs1<sblock.fs_contigsumsize;
1695                     l--, lcs1++ ) {
1696                         if(isclr(cg_clustersfree(&acg),l)){
1697                                 break;
1698                         }
1699                 }
1700                 /*
1701                  * ... and continue with the blocks right after our allocated
1702                  * block.
1703                  */
1704                 for(lcs2=0, l=blkno+1; lcs2<sblock.fs_contigsumsize;
1705                     l++, lcs2++ ) {
1706                         if(isclr(cg_clustersfree(&acg),l)){
1707                                 break;
1708                         }
1709                 }
1710
1711                 /*
1712                  * Now update all counters.
1713                  */
1714                 cg_clustersum(&acg)[MIN(lcs1+lcs2+1,sblock.fs_contigsumsize)]--;
1715                 if(lcs1) {
1716                         cg_clustersum(&acg)[lcs1]++;
1717                 }
1718                 if(lcs2) {
1719                         cg_clustersum(&acg)[lcs2]++;
1720                 }
1721         }
1722         /*
1723          * Update all statistics based on blocks.
1724          */
1725         acg.cg_cs.cs_nbfree--;
1726         sblock.fs_cstotal.cs_nbfree--;
1727
1728         DBG_LEAVE;
1729         return (d);
1730 }
1731
1732 /* *********************************************************** isblock ***** */
1733 /*
1734  * Here we check if all frags of a block are free. For more details again
1735  * please see the source of newfs(8), as this function is taken over almost
1736  * unchanged.
1737  */
1738 static int
1739 isblock(struct fs *fs, unsigned char *cp, int h)
1740 {
1741         DBG_FUNC("isblock")
1742         unsigned char   mask;
1743
1744         DBG_ENTER;
1745
1746         switch (fs->fs_frag) {
1747         case 8:
1748                 DBG_LEAVE;
1749                 return (cp[h] == 0xff);
1750         case 4:
1751                 mask = 0x0f << ((h & 0x1) << 2);
1752                 DBG_LEAVE;
1753                 return ((cp[h >> 1] & mask) == mask);
1754         case 2:
1755                 mask = 0x03 << ((h & 0x3) << 1);
1756                 DBG_LEAVE;
1757                 return ((cp[h >> 2] & mask) == mask);
1758         case 1:
1759                 mask = 0x01 << (h & 0x7);
1760                 DBG_LEAVE;
1761                 return ((cp[h >> 3] & mask) == mask);
1762         default:
1763                 fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
1764                 DBG_LEAVE;
1765                 return (0);
1766         }
1767 }
1768
1769 /* ********************************************************** clrblock ***** */
1770 /*
1771  * Here we allocate a complete block in the block map. For more details again
1772  * please see the source of newfs(8), as this function is taken over almost
1773  * unchanged.
1774  */
1775 static void
1776 clrblock(struct fs *fs, unsigned char *cp, int h)
1777 {
1778         DBG_FUNC("clrblock")
1779
1780         DBG_ENTER;
1781
1782         switch ((fs)->fs_frag) {
1783         case 8:
1784                 cp[h] = 0;
1785                 break;
1786         case 4:
1787                 cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1788                 break;
1789         case 2:
1790                 cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1791                 break;
1792         case 1:
1793                 cp[h >> 3] &= ~(0x01 << (h & 0x7));
1794                 break;
1795         default:
1796                 warnx("clrblock bad fs_frag %d", fs->fs_frag);
1797                 break;
1798         }
1799
1800         DBG_LEAVE;
1801         return;
1802 }
1803
1804 /* ********************************************************** setblock ***** */
1805 /*
1806  * Here we free a complete block in the free block map. For more details again
1807  * please see the source of newfs(8), as this function is taken over almost
1808  * unchanged.
1809  */
1810 static void
1811 setblock(struct fs *fs, unsigned char *cp, int h)
1812 {
1813         DBG_FUNC("setblock")
1814
1815         DBG_ENTER;
1816
1817         switch (fs->fs_frag) {
1818         case 8:
1819                 cp[h] = 0xff;
1820                 break;
1821         case 4:
1822                 cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1823                 break;
1824         case 2:
1825                 cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1826                 break;
1827         case 1:
1828                 cp[h >> 3] |= (0x01 << (h & 0x7));
1829                 break;
1830         default:
1831                 warnx("setblock bad fs_frag %d", fs->fs_frag);
1832                 break;
1833         }
1834
1835         DBG_LEAVE;
1836         return;
1837 }
1838
1839 /* ************************************************************ ginode ***** */
1840 /*
1841  * This function provides access to an individual inode. We find out in which
1842  * block the requested inode is located, read it from disk if needed, and
1843  * return the pointer into that block. We maintain a cache of one block to
1844  * not read the same block again and again if we iterate linearly over all
1845  * inodes.
1846  */
1847 static union dinode *
1848 ginode(ino_t inumber, int fsi, int cg)
1849 {
1850         DBG_FUNC("ginode")
1851         static ino_t    startinum = 0;  /* first inode in cached block */
1852
1853         DBG_ENTER;
1854
1855         /*
1856          * The inumber passed in is relative to the cg, so use it here to see
1857          * if the inode has been allocated yet.
1858          */
1859         if (isclr(cg_inosused(&aocg), inumber)) {
1860                 DBG_LEAVE;
1861                 return NULL;
1862         }
1863         /*
1864          * Now make the inumber relative to the entire inode space so it can
1865          * be sanity checked.
1866          */
1867         inumber += (cg * sblock.fs_ipg);
1868         if (inumber < ROOTINO) {
1869                 DBG_LEAVE;
1870                 return NULL;
1871         }
1872         if (inumber > maxino)
1873                 errx(8, "bad inode number %d to ginode", inumber);
1874         if (startinum == 0 ||
1875             inumber < startinum || inumber >= startinum + INOPB(&sblock)) {
1876                 inoblk = fsbtodb(&sblock, ino_to_fsba(&sblock, inumber));
1877                 rdfs(inoblk, (size_t)sblock.fs_bsize, inobuf, fsi);
1878                 startinum = (inumber / INOPB(&sblock)) * INOPB(&sblock);
1879         }
1880         DBG_LEAVE;
1881         if (sblock.fs_magic == FS_UFS1_MAGIC)
1882                 return (union dinode *)((uintptr_t)inobuf +
1883                     (inumber % INOPB(&sblock)) * sizeof(struct ufs1_dinode));
1884         return (union dinode *)((uintptr_t)inobuf +
1885             (inumber % INOPB(&sblock)) * sizeof(struct ufs2_dinode));
1886 }
1887
1888 /* ****************************************************** charsperline ***** */
1889 /*
1890  * Figure out how many lines our current terminal has. For more details again
1891  * please see the source of newfs(8), as this function is taken over almost
1892  * unchanged.
1893  */
1894 static int
1895 charsperline(void)
1896 {
1897         DBG_FUNC("charsperline")
1898         int     columns;
1899         char    *cp;
1900         struct winsize  ws;
1901
1902         DBG_ENTER;
1903
1904         columns = 0;
1905         if (ioctl(0, TIOCGWINSZ, &ws) != -1) {
1906                 columns = ws.ws_col;
1907         }
1908         if (columns == 0 && (cp = getenv("COLUMNS"))) {
1909                 columns = atoi(cp);
1910         }
1911         if (columns == 0) {
1912                 columns = 80;   /* last resort */
1913         }
1914
1915         DBG_LEAVE;
1916         return columns;
1917 }
1918
1919 /* ****************************************************** get_dev_size ***** */
1920 /*
1921  * Get the size of the partition if we can't figure it out from the disklabel,
1922  * e.g. from vinum volumes.
1923  */
1924 static void
1925 get_dev_size(int fd, int *size)
1926 {
1927    int sectorsize;
1928    off_t mediasize;
1929
1930    if (ioctl(fd, DIOCGSECTORSIZE, &sectorsize) == -1)
1931         err(1,"DIOCGSECTORSIZE");
1932    if (ioctl(fd, DIOCGMEDIASIZE, &mediasize) == -1)
1933         err(1,"DIOCGMEDIASIZE");
1934
1935    if (sectorsize <= 0)
1936        errx(1, "bogus sectorsize: %d", sectorsize);
1937
1938    *size = mediasize / sectorsize;
1939 }
1940
1941 /* ************************************************************** main ***** */
1942 /*
1943  * growfs(8)  is a utility which allows to increase the size of an existing
1944  * ufs file system. Currently this can only be done on unmounted file system.
1945  * It recognizes some command line options to specify the new desired size,
1946  * and it does some basic checkings. The old file system size is determined
1947  * and after some more checks like we can really access the new last block
1948  * on the disk etc. we calculate the new parameters for the superblock. After
1949  * having done this we just call growfs() which will do the work.  Before
1950  * we finish the only thing left is to update the disklabel.
1951  * We still have to provide support for snapshots. Therefore we first have to
1952  * understand what data structures are always replicated in the snapshot on
1953  * creation, for all other blocks we touch during our procedure, we have to
1954  * keep the old blocks unchanged somewhere available for the snapshots. If we
1955  * are lucky, then we only have to handle our blocks to be relocated in that
1956  * way.
1957  * Also we have to consider in what order we actually update the critical
1958  * data structures of the file system to make sure, that in case of a disaster
1959  * fsck(8) is still able to restore any lost data.
1960  * The foreseen last step then will be to provide for growing even mounted
1961  * file systems. There we have to extend the mount() system call to provide
1962  * userland access to the file system locking facility.
1963  */
1964 int
1965 main(int argc, char **argv)
1966 {
1967         DBG_FUNC("main")
1968         char    *device, *special, *cp;
1969         int     ch;
1970         unsigned int    size=0;
1971         size_t  len;
1972         unsigned int    Nflag=0;
1973         int     ExpertFlag=0;
1974         struct stat     st;
1975         struct disklabel        *lp;
1976         struct partition        *pp;
1977         int     i,fsi,fso;
1978     u_int32_t p_size;
1979         char    reply[5];
1980 #ifdef FSMAXSNAP
1981         int     j;
1982 #endif /* FSMAXSNAP */
1983
1984         DBG_ENTER;
1985
1986         while((ch=getopt(argc, argv, "Ns:vy")) != -1) {
1987                 switch(ch) {
1988                 case 'N':
1989                         Nflag=1;
1990                         break;
1991                 case 's':
1992                         size=(size_t)atol(optarg);
1993                         if(size<1) {
1994                                 usage();
1995                         }
1996                         break;
1997                 case 'v': /* for compatibility to newfs */
1998                         break;
1999                 case 'y':
2000                         ExpertFlag=1;
2001                         break;
2002                 case '?':
2003                         /* FALLTHROUGH */
2004                 default:
2005                         usage();
2006                 }
2007         }
2008         argc -= optind;
2009         argv += optind;
2010
2011         if(argc != 1) {
2012                 usage();
2013         }
2014         device=*argv;
2015
2016         /*
2017          * Now try to guess the (raw)device name.
2018          */
2019         if (0 == strrchr(device, '/')) {
2020                 /*
2021                  * No path prefix was given, so try in that order:
2022                  *     /dev/r%s
2023                  *     /dev/%s
2024                  *     /dev/vinum/r%s
2025                  *     /dev/vinum/%s.
2026                  *
2027                  * FreeBSD now doesn't distinguish between raw and block
2028                  * devices any longer, but it should still work this way.
2029                  */
2030                 len=strlen(device)+strlen(_PATH_DEV)+2+strlen("vinum/");
2031                 special=(char *)malloc(len);
2032                 if(special == NULL) {
2033                         errx(1, "malloc failed");
2034                 }
2035                 snprintf(special, len, "%sr%s", _PATH_DEV, device);
2036                 if (stat(special, &st) == -1) {
2037                         snprintf(special, len, "%s%s", _PATH_DEV, device);
2038                         if (stat(special, &st) == -1) {
2039                                 snprintf(special, len, "%svinum/r%s",
2040                                     _PATH_DEV, device);
2041                                 if (stat(special, &st) == -1) {
2042                                         /* For now this is the 'last resort' */
2043                                         snprintf(special, len, "%svinum/%s",
2044                                             _PATH_DEV, device);
2045                                 }
2046                         }
2047                 }
2048                 device = special;
2049         }
2050
2051         /*
2052          * Try to access our devices for writing ...
2053          */
2054         if (Nflag) {
2055                 fso = -1;
2056         } else {
2057                 fso = open(device, O_WRONLY);
2058                 if (fso < 0) {
2059                         err(1, "%s", device);
2060                 }
2061         }
2062
2063         /*
2064          * ... and reading.
2065          */
2066         fsi = open(device, O_RDONLY);
2067         if (fsi < 0) {
2068                 err(1, "%s", device);
2069         }
2070
2071         /*
2072          * Try to read a label and guess the slice if not specified. This
2073          * code should guess the right thing and avoid to bother the user
2074          * with the task of specifying the option -v on vinum volumes.
2075          */
2076         cp=device+strlen(device)-1;
2077         lp = get_disklabel(fsi);
2078         pp = NULL;
2079     if (lp != NULL) {
2080         if (isdigit(*cp)) {
2081             pp = &lp->d_partitions[2];
2082         } else if (*cp>='a' && *cp<='h') {
2083             pp = &lp->d_partitions[*cp - 'a'];
2084         } else {
2085             errx(1, "unknown device");
2086         }
2087         p_size = pp->p_size;
2088     } else {
2089         get_dev_size(fsi, &p_size);
2090     }
2091
2092         /*
2093          * Check if that partition is suitable for growing a file system.
2094          */
2095         if (p_size < 1) {
2096                 errx(1, "partition is unavailable");
2097         }
2098
2099         /*
2100          * Read the current superblock, and take a backup.
2101          */
2102         for (i = 0; sblock_try[i] != -1; i++) {
2103                 sblockloc = sblock_try[i] / DEV_BSIZE;
2104                 rdfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&(osblock), fsi);
2105                 if ((osblock.fs_magic == FS_UFS1_MAGIC ||
2106                      (osblock.fs_magic == FS_UFS2_MAGIC &&
2107                       osblock.fs_sblockloc == sblock_try[i])) &&
2108                     osblock.fs_bsize <= MAXBSIZE &&
2109                     osblock.fs_bsize >= (int32_t) sizeof(struct fs))
2110                         break;
2111         }
2112         if (sblock_try[i] == -1) {
2113                 errx(1, "superblock not recognized");
2114         }
2115         memcpy((void *)&fsun1, (void *)&fsun2, sizeof(fsun2));
2116         maxino = sblock.fs_ncg * sblock.fs_ipg;
2117
2118         DBG_OPEN("/tmp/growfs.debug"); /* already here we need a superblock */
2119         DBG_DUMP_FS(&sblock,
2120             "old sblock");
2121
2122         /*
2123          * Determine size to grow to. Default to the full size specified in
2124          * the disk label.
2125          */
2126         sblock.fs_size = dbtofsb(&osblock, p_size);
2127         if (size != 0) {
2128                 if (size > p_size){
2129                         errx(1, "there is not enough space (%d < %d)",
2130                             p_size, size);
2131                 }
2132                 sblock.fs_size = dbtofsb(&osblock, size);
2133         }
2134
2135         /*
2136          * Are we really growing ?
2137          */
2138         if(osblock.fs_size >= sblock.fs_size) {
2139                 errx(1, "we are not growing (%jd->%jd)",
2140                     (intmax_t)osblock.fs_size, (intmax_t)sblock.fs_size);
2141         }
2142
2143
2144 #ifdef FSMAXSNAP
2145         /*
2146          * Check if we find an active snapshot.
2147          */
2148         if(ExpertFlag == 0) {
2149                 for(j=0; j<FSMAXSNAP; j++) {
2150                         if(sblock.fs_snapinum[j]) {
2151                                 errx(1, "active snapshot found in file system; "
2152                                     "please remove all snapshots before "
2153                                     "using growfs");
2154                         }
2155                         if(!sblock.fs_snapinum[j]) { /* list is dense */
2156                                 break;
2157                         }
2158                 }
2159         }
2160 #endif
2161
2162         if (ExpertFlag == 0 && Nflag == 0) {
2163                 printf("We strongly recommend you to make a backup "
2164                     "before growing the file system.\n"
2165                     "Did you backup your data (Yes/No)? ");
2166                 fgets(reply, (int)sizeof(reply), stdin);
2167                 if (strcmp(reply, "Yes\n")){
2168                         printf("\nNothing done\n");
2169                         exit (0);
2170                 }
2171         }
2172
2173         printf("New file system size is %jd frags\n", (intmax_t)sblock.fs_size);
2174
2175         /*
2176          * Try to access our new last block in the file system. Even if we
2177          * later on realize we have to abort our operation, on that block
2178          * there should be no data, so we can't destroy something yet.
2179          */
2180         wtfs((ufs2_daddr_t)p_size-1, (size_t)DEV_BSIZE, (void *)&sblock,
2181             fso, Nflag);
2182
2183         /*
2184          * Now calculate new superblock values and check for reasonable
2185          * bound for new file system size:
2186          *     fs_size:    is derived from label or user input
2187          *     fs_dsize:   should get updated in the routines creating or
2188          *                 updating the cylinder groups on the fly
2189          *     fs_cstotal: should get updated in the routines creating or
2190          *                 updating the cylinder groups
2191          */
2192
2193         /*
2194          * Update the number of cylinders and cylinder groups in the file system.
2195          */
2196         if (sblock.fs_magic == FS_UFS1_MAGIC) {
2197                 sblock.fs_old_ncyl =
2198                     sblock.fs_size * sblock.fs_old_nspf / sblock.fs_old_spc;
2199                 if (sblock.fs_size * sblock.fs_old_nspf >
2200                     sblock.fs_old_ncyl * sblock.fs_old_spc)
2201                         sblock.fs_old_ncyl++;
2202         }
2203         sblock.fs_ncg = howmany(sblock.fs_size, sblock.fs_fpg);
2204         maxino = sblock.fs_ncg * sblock.fs_ipg;
2205
2206         if (sblock.fs_size % sblock.fs_fpg != 0 &&
2207             sblock.fs_size % sblock.fs_fpg < cgdmin(&sblock, sblock.fs_ncg)) {
2208                 /*
2209                  * The space in the new last cylinder group is too small,
2210                  * so revert back.
2211                  */
2212                 sblock.fs_ncg--;
2213                 if (sblock.fs_magic == FS_UFS1_MAGIC)
2214                         sblock.fs_old_ncyl = sblock.fs_ncg * sblock.fs_old_cpg;
2215                 printf("Warning: %jd sector(s) cannot be allocated.\n",
2216                     (intmax_t)fsbtodb(&sblock, sblock.fs_size % sblock.fs_fpg));
2217                 sblock.fs_size = sblock.fs_ncg * sblock.fs_fpg;
2218                 maxino -= sblock.fs_ipg;
2219         }
2220
2221         /*
2222          * Update the space for the cylinder group summary information in the
2223          * respective cylinder group data area.
2224          */
2225         sblock.fs_cssize =
2226             fragroundup(&sblock, sblock.fs_ncg * sizeof(struct csum));
2227
2228         if(osblock.fs_size >= sblock.fs_size) {
2229                 errx(1, "not enough new space");
2230         }
2231
2232         DBG_PRINT0("sblock calculated\n");
2233
2234         /*
2235          * Ok, everything prepared, so now let's do the tricks.
2236          */
2237         growfs(fsi, fso, Nflag);
2238
2239         /*
2240          * Update the disk label.
2241          */
2242     if (!unlabeled) {
2243         pp->p_fsize = sblock.fs_fsize;
2244         pp->p_frag = sblock.fs_frag;
2245         pp->p_cpg = sblock.fs_fpg;
2246
2247         return_disklabel(fso, lp, Nflag);
2248         DBG_PRINT0("label rewritten\n");
2249     }
2250
2251         close(fsi);
2252         if(fso>-1) close(fso);
2253
2254         DBG_CLOSE;
2255
2256         DBG_LEAVE;
2257         return 0;
2258 }
2259
2260 /* ************************************************** return_disklabel ***** */
2261 /*
2262  * Write the updated disklabel back to disk.
2263  */
2264 static void
2265 return_disklabel(int fd, struct disklabel *lp, unsigned int Nflag)
2266 {
2267         DBG_FUNC("return_disklabel")
2268         u_short sum;
2269         u_short *ptr;
2270
2271         DBG_ENTER;
2272
2273         if(!lp) {
2274                 DBG_LEAVE;
2275                 return;
2276         }
2277         if(!Nflag) {
2278                 lp->d_checksum=0;
2279                 sum = 0;
2280                 ptr=(u_short *)lp;
2281
2282                 /*
2283                  * recalculate checksum
2284                  */
2285                 while(ptr < (u_short *)&lp->d_partitions[lp->d_npartitions]) {
2286                         sum ^= *ptr++;
2287                 }
2288                 lp->d_checksum=sum;
2289
2290                 if (ioctl(fd, DIOCWDINFO, (char *)lp) < 0) {
2291                         errx(1, "DIOCWDINFO failed");
2292                 }
2293         }
2294         free(lp);
2295
2296         DBG_LEAVE;
2297         return ;
2298 }
2299
2300 /* ***************************************************** get_disklabel ***** */
2301 /*
2302  * Read the disklabel from disk.
2303  */
2304 static struct disklabel *
2305 get_disklabel(int fd)
2306 {
2307         DBG_FUNC("get_disklabel")
2308         static struct   disklabel *lab;
2309
2310         DBG_ENTER;
2311
2312         lab=(struct disklabel *)malloc(sizeof(struct disklabel));
2313         if (!lab)
2314                 errx(1, "malloc failed");
2315
2316     if (!ioctl(fd, DIOCGDINFO, (char *)lab))
2317         return (lab);
2318
2319     unlabeled++;
2320
2321         DBG_LEAVE;
2322         return (NULL);
2323 }
2324
2325
2326 /* ************************************************************* usage ***** */
2327 /*
2328  * Dump a line of usage.
2329  */
2330 static void
2331 usage(void)
2332 {
2333         DBG_FUNC("usage")
2334
2335         DBG_ENTER;
2336
2337         fprintf(stderr, "usage: growfs [-Ny] [-s size] special\n");
2338
2339         DBG_LEAVE;
2340         exit(1);
2341 }
2342
2343 /* *********************************************************** updclst ***** */
2344 /*
2345  * This updates most parameters and the bitmap related to cluster. We have to
2346  * assume that sblock, osblock, acg are set up.
2347  */
2348 static void
2349 updclst(int block)
2350 {
2351         DBG_FUNC("updclst")
2352         static int      lcs=0;
2353
2354         DBG_ENTER;
2355
2356         if(sblock.fs_contigsumsize < 1) { /* no clustering */
2357                 return;
2358         }
2359         /*
2360          * update cluster allocation map
2361          */
2362         setbit(cg_clustersfree(&acg), block);
2363
2364         /*
2365          * update cluster summary table
2366          */
2367         if(!lcs) {
2368                 /*
2369                  * calculate size for the trailing cluster
2370                  */
2371                 for(block--; lcs<sblock.fs_contigsumsize; block--, lcs++ ) {
2372                         if(isclr(cg_clustersfree(&acg), block)){
2373                                 break;
2374                         }
2375                 }
2376         }
2377         if(lcs < sblock.fs_contigsumsize) {
2378                 if(lcs) {
2379                         cg_clustersum(&acg)[lcs]--;
2380                 }
2381                 lcs++;
2382                 cg_clustersum(&acg)[lcs]++;
2383         }
2384
2385         DBG_LEAVE;
2386         return;
2387 }
2388
2389 /* *********************************************************** updrefs ***** */
2390 /*
2391  * This updates all references to relocated blocks for the given inode.  The
2392  * inode is given as number within the cylinder group, and the number of the
2393  * cylinder group.
2394  */
2395 static void
2396 updrefs(int cg, ino_t in, struct gfs_bpp *bp, int fsi, int fso, unsigned int
2397     Nflag)
2398 {
2399         DBG_FUNC("updrefs")
2400         ufs_lbn_t       len, lbn, numblks;
2401         ufs2_daddr_t    iptr, blksperindir;
2402         union dinode    *ino;
2403         int             i, mode, inodeupdated;
2404
2405         DBG_ENTER;
2406
2407         ino = ginode(in, fsi, cg);
2408         if (ino == NULL) {
2409                 DBG_LEAVE;
2410                 return;
2411         }
2412         mode = DIP(ino, di_mode) & IFMT;
2413         if (mode != IFDIR && mode != IFREG && mode != IFLNK) {
2414                 DBG_LEAVE;
2415                 return; /* only check DIR, FILE, LINK */
2416         }
2417         if (mode == IFLNK && 
2418             DIP(ino, di_size) < (u_int64_t) sblock.fs_maxsymlinklen) {
2419                 DBG_LEAVE;
2420                 return; /* skip short symlinks */
2421         }
2422         numblks = howmany(DIP(ino, di_size), sblock.fs_bsize);
2423         if (numblks == 0) {
2424                 DBG_LEAVE;
2425                 return; /* skip empty file */
2426         }
2427         if (DIP(ino, di_blocks) == 0) {
2428                 DBG_LEAVE;
2429                 return; /* skip empty swiss cheesy file or old fastlink */
2430         }
2431         DBG_PRINT2("scg checking inode (%d in %d)\n",
2432             in,
2433             cg);
2434
2435         /*
2436          * Check all the blocks.
2437          */
2438         inodeupdated = 0;
2439         len = numblks < NDADDR ? numblks : NDADDR;
2440         for (i = 0; i < len; i++) {
2441                 iptr = DIP(ino, di_db[i]);
2442                 if (iptr == 0)
2443                         continue;
2444                 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2445                         DIP_SET(ino, di_db[i], iptr);
2446                         inodeupdated++;
2447                 }
2448         }
2449         DBG_PRINT0("~~scg direct blocks checked\n");
2450
2451         blksperindir = 1;
2452         len = numblks - NDADDR;
2453         lbn = NDADDR;
2454         for (i = 0; len > 0 && i < NIADDR; i++) {
2455                 iptr = DIP(ino, di_ib[i]);
2456                 if (iptr == 0)
2457                         continue;
2458                 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2459                         DIP_SET(ino, di_ib[i], iptr);
2460                         inodeupdated++;
2461                 }
2462                 indirchk(blksperindir, lbn, iptr, numblks, bp, fsi, fso, Nflag);
2463                 blksperindir *= NINDIR(&sblock);
2464                 lbn += blksperindir;
2465                 len -= blksperindir;
2466                 DBG_PRINT1("scg indirect_%d blocks checked\n", i + 1);
2467         }
2468         if (inodeupdated)
2469                 wtfs(inoblk, sblock.fs_bsize, inobuf, fso, Nflag);
2470
2471         DBG_LEAVE;
2472         return;
2473 }
2474
2475 /*
2476  * Recursively check all the indirect blocks.
2477  */
2478 static void
2479 indirchk(ufs_lbn_t blksperindir, ufs_lbn_t lbn, ufs2_daddr_t blkno,
2480     ufs_lbn_t lastlbn, struct gfs_bpp *bp, int fsi, int fso, unsigned int Nflag)
2481 {
2482         DBG_FUNC("indirchk")
2483         void *ibuf;
2484         int i, last;
2485         ufs2_daddr_t iptr;
2486
2487         DBG_ENTER;
2488
2489         /* read in the indirect block. */
2490         ibuf = malloc(sblock.fs_bsize);
2491         if (!ibuf)
2492                 errx(1, "malloc failed");
2493         rdfs(fsbtodb(&sblock, blkno), (size_t)sblock.fs_bsize, ibuf, fsi);
2494         last = howmany(lastlbn - lbn, blksperindir) < NINDIR(&sblock) ?
2495             howmany(lastlbn - lbn, blksperindir) : NINDIR(&sblock);
2496         for (i = 0; i < last; i++) {
2497                 if (sblock.fs_magic == FS_UFS1_MAGIC)
2498                         iptr = ((ufs1_daddr_t *)ibuf)[i];
2499                 else
2500                         iptr = ((ufs2_daddr_t *)ibuf)[i];
2501                 if (iptr == 0)
2502                         continue;
2503                 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2504                         if (sblock.fs_magic == FS_UFS1_MAGIC)
2505                                 ((ufs1_daddr_t *)ibuf)[i] = iptr;
2506                         else
2507                                 ((ufs2_daddr_t *)ibuf)[i] = iptr;
2508                 }
2509                 if (blksperindir == 1)
2510                         continue;
2511                 indirchk(blksperindir / NINDIR(&sblock), lbn + blksperindir * i,
2512                     iptr, lastlbn, bp, fsi, fso, Nflag);
2513         }
2514         free(ibuf);
2515
2516         DBG_LEAVE;
2517         return;
2518 }