]> CyberLeo.Net >> Repos - FreeBSD/releng/9.0.git/blob - sbin/growfs/growfs.c
Copy stable/9 to releng/9.0 as part of the FreeBSD 9.0-RELEASE release
[FreeBSD/releng/9.0.git] / sbin / growfs / growfs.c
1 /*
2  * Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz
3  * Copyright (c) 1980, 1989, 1993 The Regents of the University of California.
4  * All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * Christoph Herrmann and Thomas-Henning von Kamptz, Munich and Frankfurt.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgment:
19  *      This product includes software developed by the University of
20  *      California, Berkeley and its contributors, as well as Christoph
21  *      Herrmann and Thomas-Henning von Kamptz.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  * $TSHeader: src/sbin/growfs/growfs.c,v 1.5 2000/12/12 19:31:00 tomsoft Exp $
39  *
40  */
41
42 #ifndef lint
43 static const char copyright[] =
44 "@(#) Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz\n\
45 Copyright (c) 1980, 1989, 1993 The Regents of the University of California.\n\
46 All rights reserved.\n";
47 #endif /* not lint */
48
49 #include <sys/cdefs.h>
50 __FBSDID("$FreeBSD$");
51
52 /* ********************************************************** INCLUDES ***** */
53 #include <sys/param.h>
54 #include <sys/disklabel.h>
55 #include <sys/ioctl.h>
56 #include <sys/stat.h>
57 #include <sys/disk.h>
58
59 #include <stdio.h>
60 #include <paths.h>
61 #include <ctype.h>
62 #include <err.h>
63 #include <fcntl.h>
64 #include <limits.h>
65 #include <stdlib.h>
66 #include <stdint.h>
67 #include <string.h>
68 #include <time.h>
69 #include <unistd.h>
70 #include <ufs/ufs/dinode.h>
71 #include <ufs/ffs/fs.h>
72
73 #include "debug.h"
74
75 /* *************************************************** GLOBALS & TYPES ***** */
76 #ifdef FS_DEBUG
77 int     _dbg_lvl_ = (DL_INFO);  /* DL_TRC */
78 #endif /* FS_DEBUG */
79
80 static union {
81         struct fs       fs;
82         char    pad[SBLOCKSIZE];
83 } fsun1, fsun2;
84 #define sblock  fsun1.fs        /* the new superblock */
85 #define osblock fsun2.fs        /* the old superblock */
86
87 /*
88  * Possible superblock locations ordered from most to least likely.
89  */
90 static int sblock_try[] = SBLOCKSEARCH;
91 static ufs2_daddr_t sblockloc;
92
93 static union {
94         struct cg       cg;
95         char    pad[MAXBSIZE];
96 } cgun1, cgun2;
97 #define acg     cgun1.cg        /* a cylinder cgroup (new) */
98 #define aocg    cgun2.cg        /* an old cylinder group */
99
100 static char     ablk[MAXBSIZE]; /* a block */
101
102 static struct csum      *fscs;  /* cylinder summary */
103
104 union dinode {
105         struct ufs1_dinode dp1;
106         struct ufs2_dinode dp2;
107 };
108 #define DIP(dp, field) \
109         ((sblock.fs_magic == FS_UFS1_MAGIC) ? \
110         (uint32_t)(dp)->dp1.field : (dp)->dp2.field)
111 #define DIP_SET(dp, field, val) do { \
112         if (sblock.fs_magic == FS_UFS1_MAGIC) \
113                 (dp)->dp1.field = (val); \
114         else \
115                 (dp)->dp2.field = (val); \
116         } while (0)
117 static ufs2_daddr_t     inoblk;                 /* inode block address */
118 static char             inobuf[MAXBSIZE];       /* inode block */
119 ino_t                   maxino;                 /* last valid inode */
120 static int              unlabeled;     /* unlabeled partition, e.g. vinum volume etc. */
121
122 /*
123  * An array of elements of type struct gfs_bpp describes all blocks to
124  * be relocated in order to free the space needed for the cylinder group
125  * summary for all cylinder groups located in the first cylinder group.
126  */
127 struct gfs_bpp {
128         ufs2_daddr_t    old;            /* old block number */
129         ufs2_daddr_t    new;            /* new block number */
130 #define GFS_FL_FIRST    1
131 #define GFS_FL_LAST     2
132         unsigned int    flags;  /* special handling required */
133         int     found;          /* how many references were updated */
134 };
135
136 /* ******************************************************** PROTOTYPES ***** */
137 static void     growfs(int, int, unsigned int);
138 static void     rdfs(ufs2_daddr_t, size_t, void *, int);
139 static void     wtfs(ufs2_daddr_t, size_t, void *, int, unsigned int);
140 static ufs2_daddr_t alloc(void);
141 static int      charsperline(void);
142 static void     usage(void);
143 static int      isblock(struct fs *, unsigned char *, int);
144 static void     clrblock(struct fs *, unsigned char *, int);
145 static void     setblock(struct fs *, unsigned char *, int);
146 static void     initcg(int, time_t, int, unsigned int);
147 static void     updjcg(int, time_t, int, int, unsigned int);
148 static void     updcsloc(time_t, int, int, unsigned int);
149 static struct disklabel *get_disklabel(int);
150 static void     return_disklabel(int, struct disklabel *, unsigned int);
151 static union dinode *ginode(ino_t, int, int);
152 static void     frag_adjust(ufs2_daddr_t, int);
153 static int      cond_bl_upd(ufs2_daddr_t *, struct gfs_bpp *, int, int,
154                     unsigned int);
155 static void     updclst(int);
156 static void     updrefs(int, ino_t, struct gfs_bpp *, int, int, unsigned int);
157 static void     indirchk(ufs_lbn_t, ufs_lbn_t, ufs2_daddr_t, ufs_lbn_t,
158                     struct gfs_bpp *, int, int, unsigned int);
159 static void     get_dev_size(int, int *);
160
161 /* ************************************************************ growfs ***** */
162 /*
163  * Here we actually start growing the file system. We basically read the
164  * cylinder summary from the first cylinder group as we want to update
165  * this on the fly during our various operations. First we handle the
166  * changes in the former last cylinder group. Afterwards we create all new
167  * cylinder groups.  Now we handle the cylinder group containing the
168  * cylinder summary which might result in a relocation of the whole
169  * structure.  In the end we write back the updated cylinder summary, the
170  * new superblock, and slightly patched versions of the super block
171  * copies.
172  */
173 static void
174 growfs(int fsi, int fso, unsigned int Nflag)
175 {
176         DBG_FUNC("growfs")
177         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         struct ufs1_dinode *dp1;
378         struct csum *cs;
379         uint d, dupper, dlower;
380
381         if (iobuf == NULL && (iobuf = malloc(sblock.fs_bsize * 3)) == NULL)
382                 errx(37, "panic: cannot allocate I/O buffer");
383
384         /*
385          * Determine block bounds for cylinder group.
386          * Allow space for super block summary information in first
387          * cylinder group.
388          */
389         cbase = cgbase(&sblock, cylno);
390         dmax = cbase + sblock.fs_fpg;
391         if (dmax > sblock.fs_size)
392                 dmax = sblock.fs_size;
393         dlower = cgsblock(&sblock, cylno) - cbase;
394         dupper = cgdmin(&sblock, cylno) - cbase;
395         if (cylno == 0) /* XXX fscs may be relocated */
396                 dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
397         cs = &fscs[cylno];
398         memset(&acg, 0, sblock.fs_cgsize);
399         acg.cg_time = modtime;
400         acg.cg_magic = CG_MAGIC;
401         acg.cg_cgx = cylno;
402         acg.cg_niblk = sblock.fs_ipg;
403         acg.cg_initediblk = sblock.fs_ipg < 2 * INOPB(&sblock) ?
404             sblock.fs_ipg : 2 * INOPB(&sblock);
405         acg.cg_ndblk = dmax - cbase;
406         if (sblock.fs_contigsumsize > 0)
407                 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
408         start = &acg.cg_space[0] - (u_char *)(&acg.cg_firstfield);
409         if (sblock.fs_magic == FS_UFS2_MAGIC) {
410                 acg.cg_iusedoff = start;
411         } else {
412                 acg.cg_old_ncyl = sblock.fs_old_cpg;
413                 acg.cg_old_time = acg.cg_time;
414                 acg.cg_time = 0;
415                 acg.cg_old_niblk = acg.cg_niblk;
416                 acg.cg_niblk = 0;
417                 acg.cg_initediblk = 0;
418                 acg.cg_old_btotoff = start;
419                 acg.cg_old_boff = acg.cg_old_btotoff +
420                     sblock.fs_old_cpg * sizeof(int32_t);
421                 acg.cg_iusedoff = acg.cg_old_boff +
422                     sblock.fs_old_cpg * sizeof(u_int16_t);
423         }
424         acg.cg_freeoff = acg.cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
425         acg.cg_nextfreeoff = acg.cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT);
426         if (sblock.fs_contigsumsize > 0) {
427                 acg.cg_clustersumoff =
428                     roundup(acg.cg_nextfreeoff, sizeof(u_int32_t));
429                 acg.cg_clustersumoff -= sizeof(u_int32_t);
430                 acg.cg_clusteroff = acg.cg_clustersumoff +
431                     (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
432                 acg.cg_nextfreeoff = acg.cg_clusteroff +
433                     howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
434         }
435         if (acg.cg_nextfreeoff > (unsigned)sblock.fs_cgsize) {
436                 /*
437                  * This should never happen as we would have had that panic
438                  * already on file system creation
439                  */
440                 errx(37, "panic: cylinder group too big");
441         }
442         acg.cg_cs.cs_nifree += sblock.fs_ipg;
443         if (cylno == 0)
444                 for (i = 0; i < ROOTINO; i++) {
445                         setbit(cg_inosused(&acg), i);
446                         acg.cg_cs.cs_nifree--;
447                 }
448         /*
449          * For the old file system, we have to initialize all the inodes.
450          */
451         if (sblock.fs_magic == FS_UFS1_MAGIC) {
452                 bzero(iobuf, sblock.fs_bsize);
453                 for (i = 0; i < sblock.fs_ipg / INOPF(&sblock);
454                      i += sblock.fs_frag) {
455                         dp1 = (struct ufs1_dinode *)(void *)iobuf;
456 #ifdef FSIRAND
457                         for (j = 0; j < INOPB(&sblock); j++) {
458                                 dp1->di_gen = random();
459                                 dp1++;
460                         }
461 #endif
462                         wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno) + i),
463                             sblock.fs_bsize, iobuf, fso, Nflag);
464                 }
465         }
466         if (cylno > 0) {
467                 /*
468                  * In cylno 0, beginning space is reserved
469                  * for boot and super blocks.
470                  */
471                 for (d = 0; d < dlower; d += sblock.fs_frag) {
472                         blkno = d / sblock.fs_frag;
473                         setblock(&sblock, cg_blksfree(&acg), blkno);
474                         if (sblock.fs_contigsumsize > 0)
475                                 setbit(cg_clustersfree(&acg), blkno);
476                         acg.cg_cs.cs_nbfree++;
477                 }
478                 sblock.fs_dsize += dlower;
479         }
480         sblock.fs_dsize += acg.cg_ndblk - dupper;
481         if ((i = dupper % sblock.fs_frag)) {
482                 acg.cg_frsum[sblock.fs_frag - i]++;
483                 for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
484                         setbit(cg_blksfree(&acg), dupper);
485                         acg.cg_cs.cs_nffree++;
486                 }
487         }
488         for (d = dupper; d + sblock.fs_frag <= acg.cg_ndblk;
489              d += sblock.fs_frag) {
490                 blkno = d / sblock.fs_frag;
491                 setblock(&sblock, cg_blksfree(&acg), blkno);
492                 if (sblock.fs_contigsumsize > 0)
493                         setbit(cg_clustersfree(&acg), blkno);
494                 acg.cg_cs.cs_nbfree++;
495         }
496         if (d < acg.cg_ndblk) {
497                 acg.cg_frsum[acg.cg_ndblk - d]++;
498                 for (; d < acg.cg_ndblk; d++) {
499                         setbit(cg_blksfree(&acg), d);
500                         acg.cg_cs.cs_nffree++;
501                 }
502         }
503         if (sblock.fs_contigsumsize > 0) {
504                 int32_t *sump = cg_clustersum(&acg);
505                 u_char *mapp = cg_clustersfree(&acg);
506                 int map = *mapp++;
507                 int bit = 1;
508                 int run = 0;
509
510                 for (i = 0; i < acg.cg_nclusterblks; i++) {
511                         if ((map & bit) != 0)
512                                 run++;
513                         else if (run != 0) {
514                                 if (run > sblock.fs_contigsumsize)
515                                         run = sblock.fs_contigsumsize;
516                                 sump[run]++;
517                                 run = 0;
518                         }
519                         if ((i & (CHAR_BIT - 1)) != CHAR_BIT - 1)
520                                 bit <<= 1;
521                         else {
522                                 map = *mapp++;
523                                 bit = 1;
524                         }
525                 }
526                 if (run != 0) {
527                         if (run > sblock.fs_contigsumsize)
528                                 run = sblock.fs_contigsumsize;
529                         sump[run]++;
530                 }
531         }
532         sblock.fs_cstotal.cs_ndir += acg.cg_cs.cs_ndir;
533         sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
534         sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
535         sblock.fs_cstotal.cs_nifree += acg.cg_cs.cs_nifree;
536         *cs = acg.cg_cs;
537
538         memcpy(iobuf, &acg, sblock.fs_cgsize);
539         memset(iobuf + sblock.fs_cgsize, '\0',
540             sblock.fs_bsize * 3 - sblock.fs_cgsize);
541
542         wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
543             sblock.fs_bsize * 3, iobuf, fso, Nflag);
544         DBG_DUMP_CG(&sblock, "new cg", &acg);
545
546         DBG_LEAVE;
547         return;
548 }
549
550 /* ******************************************************* frag_adjust ***** */
551 /*
552  * Here we add or subtract (sign +1/-1) the available fragments in a given
553  * block to or from the fragment statistics. By subtracting before and adding
554  * after an operation on the free frag map we can easy update the fragment
555  * statistic, which seems to be otherwise a rather complex operation.
556  */
557 static void
558 frag_adjust(ufs2_daddr_t frag, int sign)
559 {
560         DBG_FUNC("frag_adjust")
561         int fragsize;
562         int f;
563
564         DBG_ENTER;
565
566         fragsize=0;
567         /*
568          * Here frag only needs to point to any fragment in the block we want
569          * to examine.
570          */
571         for(f=rounddown(frag, sblock.fs_frag);
572             f<roundup(frag+1, sblock.fs_frag);
573             f++) {
574                 /*
575                  * Count contiguous free fragments.
576                  */
577                 if(isset(cg_blksfree(&acg), f)) {
578                         fragsize++;
579                 } else {
580                         if(fragsize && fragsize<sblock.fs_frag) {
581                                 /*
582                                  * We found something in between.
583                                  */
584                                 acg.cg_frsum[fragsize]+=sign;
585                                 DBG_PRINT2("frag_adjust [%d]+=%d\n",
586                                     fragsize,
587                                     sign);
588                         }
589                         fragsize=0;
590                 }
591         }
592         if(fragsize && fragsize<sblock.fs_frag) {
593                 /*
594                  * We found something.
595                  */
596                 acg.cg_frsum[fragsize]+=sign;
597                 DBG_PRINT2("frag_adjust [%d]+=%d\n",
598                     fragsize,
599                     sign);
600         }
601         DBG_PRINT2("frag_adjust [[%d]]+=%d\n",
602             fragsize,
603             sign);
604
605         DBG_LEAVE;
606         return;
607 }
608
609 /* ******************************************************* cond_bl_upd ***** */
610 /*
611  * Here we conditionally update a pointer to a fragment. We check for all
612  * relocated blocks if any of its fragments is referenced by the current
613  * field, and update the pointer to the respective fragment in our new
614  * block.  If we find a reference we write back the block immediately,
615  * as there is no easy way for our general block reading engine to figure
616  * out if a write back operation is needed.
617  */
618 static int
619 cond_bl_upd(ufs2_daddr_t *block, struct gfs_bpp *field, int fsi, int fso,
620     unsigned int Nflag)
621 {
622         DBG_FUNC("cond_bl_upd")
623         struct gfs_bpp *f;
624         ufs2_daddr_t src, dst;
625         int fragnum;
626         void *ibuf;
627
628         DBG_ENTER;
629
630         for (f = field; f->old != 0; f++) {
631                 src = *block;
632                 if (fragstoblks(&sblock, src) != f->old)
633                         continue;
634                 /*
635                  * The fragment is part of the block, so update.
636                  */
637                 dst = blkstofrags(&sblock, f->new);
638                 fragnum = fragnum(&sblock, src);
639                 *block = dst + fragnum;
640                 f->found++;
641                 DBG_PRINT3("scg (%jd->%jd)[%d] reference updated\n",
642                     (intmax_t)f->old,
643                     (intmax_t)f->new,
644                     fragnum);
645
646                 /*
647                  * Copy the block back immediately.
648                  *
649                  * XXX  If src is from an indirect block we have
650                  *      to implement copy on write here in case of
651                  *      active snapshots.
652                  */
653                 ibuf = malloc(sblock.fs_bsize);
654                 if (!ibuf)
655                         errx(1, "malloc failed");
656                 src -= fragnum;
657                 rdfs(fsbtodb(&sblock, src), (size_t)sblock.fs_bsize, ibuf, fsi);
658                 wtfs(dst, (size_t)sblock.fs_bsize, ibuf, fso, Nflag);
659                 free(ibuf);
660                 /*
661                  * The same block can't be found again in this loop.
662                  */
663                 return (1);
664         }
665
666         DBG_LEAVE;
667         return (0);
668 }
669
670 /* ************************************************************ updjcg ***** */
671 /*
672  * Here we do all needed work for the former last cylinder group. It has to be
673  * changed in any case, even if the file system ended exactly on the end of
674  * this group, as there is some slightly inconsistent handling of the number
675  * of cylinders in the cylinder group. We start again by reading the cylinder
676  * group from disk. If the last block was not fully available, we first handle
677  * the missing fragments, then we handle all new full blocks in that file
678  * system and finally we handle the new last fragmented block in the file
679  * system.  We again have to handle the fragment statistics rotational layout
680  * tables and cluster summary during all those operations.
681  */
682 static void
683 updjcg(int cylno, time_t modtime, int fsi, int fso, unsigned int Nflag)
684 {
685         DBG_FUNC("updjcg")
686         ufs2_daddr_t    cbase, dmax, dupper;
687         struct csum     *cs;
688         int     i,k;
689         int     j=0;
690
691         DBG_ENTER;
692
693         /*
694          * Read the former last (joining) cylinder group from disk, and make
695          * a copy.
696          */
697         rdfs(fsbtodb(&osblock, cgtod(&osblock, cylno)),
698             (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
699         DBG_PRINT0("jcg read\n");
700         DBG_DUMP_CG(&sblock,
701             "old joining cg",
702             &aocg);
703
704         memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
705
706         /*
707          * If the cylinder group had already its new final size almost
708          * nothing is to be done ... except:
709          * For some reason the value of cg_ncyl in the last cylinder group has
710          * to be zero instead of fs_cpg. As this is now no longer the last
711          * cylinder group we have to change that value now to fs_cpg.
712          */
713
714         if(cgbase(&osblock, cylno+1) == osblock.fs_size) {
715                 if (sblock.fs_magic == FS_UFS1_MAGIC)
716                         acg.cg_old_ncyl=sblock.fs_old_cpg;
717
718                 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
719                     (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
720                 DBG_PRINT0("jcg written\n");
721                 DBG_DUMP_CG(&sblock,
722                     "new joining cg",
723                     &acg);
724
725                 DBG_LEAVE;
726                 return;
727         }
728
729         /*
730          * Set up some variables needed later.
731          */
732         cbase = cgbase(&sblock, cylno);
733         dmax = cbase + sblock.fs_fpg;
734         if (dmax > sblock.fs_size)
735                 dmax = sblock.fs_size;
736         dupper = cgdmin(&sblock, cylno) - cbase;
737         if (cylno == 0) { /* XXX fscs may be relocated */
738                 dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
739         }
740
741         /*
742          * Set pointer to the cylinder summary for our cylinder group.
743          */
744         cs = fscs + cylno;
745
746         /*
747          * Touch the cylinder group, update all fields in the cylinder group as
748          * needed, update the free space in the superblock.
749          */
750         acg.cg_time = modtime;
751         if ((unsigned)cylno == sblock.fs_ncg - 1) {
752                 /*
753                  * This is still the last cylinder group.
754                  */
755                 if (sblock.fs_magic == FS_UFS1_MAGIC)
756                         acg.cg_old_ncyl =
757                             sblock.fs_old_ncyl % sblock.fs_old_cpg;
758         } else {
759                 acg.cg_old_ncyl = sblock.fs_old_cpg;
760         }
761         DBG_PRINT2("jcg dbg: %d %u",
762             cylno,
763             sblock.fs_ncg);
764 #ifdef FS_DEBUG
765         if (sblock.fs_magic == FS_UFS1_MAGIC)
766                 DBG_PRINT2("%d %u",
767                     acg.cg_old_ncyl,
768                     sblock.fs_old_cpg);
769 #endif
770         DBG_PRINT0("\n");
771         acg.cg_ndblk = dmax - cbase;
772         sblock.fs_dsize += acg.cg_ndblk-aocg.cg_ndblk;
773         if (sblock.fs_contigsumsize > 0) {
774                 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
775         }
776
777         /*
778          * Now we have to update the free fragment bitmap for our new free
779          * space.  There again we have to handle the fragmentation and also
780          * the rotational layout tables and the cluster summary.  This is
781          * also done per fragment for the first new block if the old file
782          * system end was not on a block boundary, per fragment for the new
783          * last block if the new file system end is not on a block boundary,
784          * and per block for all space in between.
785          *
786          * Handle the first new block here if it was partially available
787          * before.
788          */
789         if(osblock.fs_size % sblock.fs_frag) {
790                 if(roundup(osblock.fs_size, sblock.fs_frag)<=sblock.fs_size) {
791                         /*
792                          * The new space is enough to fill at least this
793                          * block
794                          */
795                         j=0;
796                         for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag)-1;
797                             i>=osblock.fs_size-cbase;
798                             i--) {
799                                 setbit(cg_blksfree(&acg), i);
800                                 acg.cg_cs.cs_nffree++;
801                                 j++;
802                         }
803
804                         /*
805                          * Check if the fragment just created could join an
806                          * already existing fragment at the former end of the
807                          * file system.
808                          */
809                         if(isblock(&sblock, cg_blksfree(&acg),
810                             ((osblock.fs_size - cgbase(&sblock, cylno))/
811                             sblock.fs_frag))) {
812                                 /*
813                                  * The block is now completely available.
814                                  */
815                                 DBG_PRINT0("block was\n");
816                                 acg.cg_frsum[osblock.fs_size%sblock.fs_frag]--;
817                                 acg.cg_cs.cs_nbfree++;
818                                 acg.cg_cs.cs_nffree-=sblock.fs_frag;
819                                 k=rounddown(osblock.fs_size-cbase,
820                                     sblock.fs_frag);
821                                 updclst((osblock.fs_size-cbase)/sblock.fs_frag);
822                         } else {
823                                 /*
824                                  * Lets rejoin a possible partially growed
825                                  * fragment.
826                                  */
827                                 k=0;
828                                 while(isset(cg_blksfree(&acg), i) &&
829                                     (i>=rounddown(osblock.fs_size-cbase,
830                                     sblock.fs_frag))) {
831                                         i--;
832                                         k++;
833                                 }
834                                 if(k) {
835                                         acg.cg_frsum[k]--;
836                                 }
837                                 acg.cg_frsum[k+j]++;
838                         }
839                 } else {
840                         /*
841                          * We only grow by some fragments within this last
842                          * block.
843                          */
844                         for(i=sblock.fs_size-cbase-1;
845                                 i>=osblock.fs_size-cbase;
846                                 i--) {
847                                 setbit(cg_blksfree(&acg), i);
848                                 acg.cg_cs.cs_nffree++;
849                                 j++;
850                         }
851                         /*
852                          * Lets rejoin a possible partially growed fragment.
853                          */
854                         k=0;
855                         while(isset(cg_blksfree(&acg), i) &&
856                             (i>=rounddown(osblock.fs_size-cbase,
857                             sblock.fs_frag))) {
858                                 i--;
859                                 k++;
860                         }
861                         if(k) {
862                                 acg.cg_frsum[k]--;
863                         }
864                         acg.cg_frsum[k+j]++;
865                 }
866         }
867
868         /*
869          * Handle all new complete blocks here.
870          */
871         for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag);
872             i+sblock.fs_frag<=dmax-cbase;       /* XXX <= or only < ? */
873             i+=sblock.fs_frag) {
874                 j = i / sblock.fs_frag;
875                 setblock(&sblock, cg_blksfree(&acg), j);
876                 updclst(j);
877                 acg.cg_cs.cs_nbfree++;
878         }
879
880         /*
881          * Handle the last new block if there are stll some new fragments left.
882          * Here we don't have to bother about the cluster summary or the even
883          * the rotational layout table.
884          */
885         if (i < (dmax - cbase)) {
886                 acg.cg_frsum[dmax - cbase - i]++;
887                 for (; i < dmax - cbase; i++) {
888                         setbit(cg_blksfree(&acg), i);
889                         acg.cg_cs.cs_nffree++;
890                 }
891         }
892
893         sblock.fs_cstotal.cs_nffree +=
894             (acg.cg_cs.cs_nffree - aocg.cg_cs.cs_nffree);
895         sblock.fs_cstotal.cs_nbfree +=
896             (acg.cg_cs.cs_nbfree - aocg.cg_cs.cs_nbfree);
897         /*
898          * The following statistics are not changed here:
899          *     sblock.fs_cstotal.cs_ndir
900          *     sblock.fs_cstotal.cs_nifree
901          * As the statistics for this cylinder group are ready, copy it to
902          * the summary information array.
903          */
904         *cs = acg.cg_cs;
905
906         /*
907          * Write the updated "joining" cylinder group back to disk.
908          */
909         wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize,
910             (void *)&acg, fso, Nflag);
911         DBG_PRINT0("jcg written\n");
912         DBG_DUMP_CG(&sblock,
913             "new joining cg",
914             &acg);
915
916         DBG_LEAVE;
917         return;
918 }
919
920 /* ********************************************************** updcsloc ***** */
921 /*
922  * Here we update the location of the cylinder summary. We have two possible
923  * ways of growing the cylinder summary.
924  * (1)  We can try to grow the summary in the current location, and relocate
925  *      possibly used blocks within the current cylinder group.
926  * (2)  Alternatively we can relocate the whole cylinder summary to the first
927  *      new completely empty cylinder group. Once the cylinder summary is no
928  *      longer in the beginning of the first cylinder group you should never
929  *      use a version of fsck which is not aware of the possibility to have
930  *      this structure in a non standard place.
931  * Option (1) is considered to be less intrusive to the structure of the file-
932  * system. So we try to stick to that whenever possible. If there is not enough
933  * space in the cylinder group containing the cylinder summary we have to use
934  * method (2). In case of active snapshots in the file system we probably can
935  * completely avoid implementing copy on write if we stick to method (2) only.
936  */
937 static void
938 updcsloc(time_t modtime, int fsi, int fso, unsigned int Nflag)
939 {
940         DBG_FUNC("updcsloc")
941         struct csum     *cs;
942         int     ocscg, ncscg;
943         int     blocks;
944         ufs2_daddr_t    cbase, dupper, odupper, d, f, g;
945         int     ind, inc;
946         uint    cylno;
947         struct gfs_bpp  *bp;
948         int     i, l;
949         int     lcs=0;
950         int     block;
951
952         DBG_ENTER;
953
954         if(howmany(sblock.fs_cssize, sblock.fs_fsize) ==
955             howmany(osblock.fs_cssize, osblock.fs_fsize)) {
956                 /*
957                  * No new fragment needed.
958                  */
959                 DBG_LEAVE;
960                 return;
961         }
962         ocscg=dtog(&osblock, osblock.fs_csaddr);
963         cs=fscs+ocscg;
964         blocks = 1+howmany(sblock.fs_cssize, sblock.fs_bsize)-
965             howmany(osblock.fs_cssize, osblock.fs_bsize);
966
967         /*
968          * Read original cylinder group from disk, and make a copy.
969          * XXX  If Nflag is set in some very rare cases we now miss
970          *      some changes done in updjcg by reading the unmodified
971          *      block from disk.
972          */
973         rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)),
974             (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
975         DBG_PRINT0("oscg read\n");
976         DBG_DUMP_CG(&sblock,
977             "old summary cg",
978             &aocg);
979
980         memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
981
982         /*
983          * Touch the cylinder group, set up local variables needed later
984          * and update the superblock.
985          */
986         acg.cg_time = modtime;
987
988         /*
989          * XXX  In the case of having active snapshots we may need much more
990          *      blocks for the copy on write. We need each block twice, and
991          *      also up to 8*3 blocks for indirect blocks for all possible
992          *      references.
993          */
994         if(/*((int)sblock.fs_time&0x3)>0||*/ cs->cs_nbfree < blocks) {
995                 /*
996                  * There is not enough space in the old cylinder group to
997                  * relocate all blocks as needed, so we relocate the whole
998                  * cylinder group summary to a new group. We try to use the
999                  * first complete new cylinder group just created. Within the
1000                  * cylinder group we align the area immediately after the
1001                  * cylinder group information location in order to be as
1002                  * close as possible to the original implementation of ffs.
1003                  *
1004                  * First we have to make sure we'll find enough space in the
1005                  * new cylinder group. If not, then we currently give up.
1006                  * We start with freeing everything which was used by the
1007                  * fragments of the old cylinder summary in the current group.
1008                  * Now we write back the group meta data, read in the needed
1009                  * meta data from the new cylinder group, and start allocating
1010                  * within that group. Here we can assume, the group to be
1011                  * completely empty. Which makes the handling of fragments and
1012                  * clusters a lot easier.
1013                  */
1014                 DBG_TRC;
1015                 if(sblock.fs_ncg-osblock.fs_ncg < 2) {
1016                         errx(2, "panic: not enough space");
1017                 }
1018
1019                 /*
1020                  * Point "d" to the first fragment not used by the cylinder
1021                  * summary.
1022                  */
1023                 d=osblock.fs_csaddr+(osblock.fs_cssize/osblock.fs_fsize);
1024
1025                 /*
1026                  * Set up last cluster size ("lcs") already here. Calculate
1027                  * the size for the trailing cluster just behind where "d"
1028                  * points to.
1029                  */
1030                 if(sblock.fs_contigsumsize > 0) {
1031                         for(block=howmany(d%sblock.fs_fpg, sblock.fs_frag),
1032                             lcs=0; lcs<sblock.fs_contigsumsize;
1033                             block++, lcs++) {
1034                                 if(isclr(cg_clustersfree(&acg), block)){
1035                                         break;
1036                                 }
1037                         }
1038                 }
1039
1040                 /*
1041                  * Point "d" to the last frag used by the cylinder summary.
1042                  */
1043                 d--;
1044
1045                 DBG_PRINT1("d=%jd\n",
1046                     (intmax_t)d);
1047                 if((d+1)%sblock.fs_frag) {
1048                         /*
1049                          * The end of the cylinder summary is not a complete
1050                          * block.
1051                          */
1052                         DBG_TRC;
1053                         frag_adjust(d%sblock.fs_fpg, -1);
1054                         for(; (d+1)%sblock.fs_frag; d--) {
1055                                 DBG_PRINT1("d=%jd\n",
1056                                     (intmax_t)d);
1057                                 setbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1058                                 acg.cg_cs.cs_nffree++;
1059                                 sblock.fs_cstotal.cs_nffree++;
1060                         }
1061                         /*
1062                          * Point "d" to the last fragment of the last
1063                          * (incomplete) block of the cylinder summary.
1064                          */
1065                         d++;
1066                         frag_adjust(d%sblock.fs_fpg, 1);
1067
1068                         if(isblock(&sblock, cg_blksfree(&acg),
1069                             (d%sblock.fs_fpg)/sblock.fs_frag)) {
1070                                 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1071                                 acg.cg_cs.cs_nffree-=sblock.fs_frag;
1072                                 acg.cg_cs.cs_nbfree++;
1073                                 sblock.fs_cstotal.cs_nffree-=sblock.fs_frag;
1074                                 sblock.fs_cstotal.cs_nbfree++;
1075                                 if(sblock.fs_contigsumsize > 0) {
1076                                         setbit(cg_clustersfree(&acg),
1077                                             (d%sblock.fs_fpg)/sblock.fs_frag);
1078                                         if(lcs < sblock.fs_contigsumsize) {
1079                                                 if(lcs) {
1080                                                         cg_clustersum(&acg)
1081                                                             [lcs]--;
1082                                                 }
1083                                                 lcs++;
1084                                                 cg_clustersum(&acg)[lcs]++;
1085                                         }
1086                                 }
1087                         }
1088                         /*
1089                          * Point "d" to the first fragment of the block before
1090                          * the last incomplete block.
1091                          */
1092                         d--;
1093                 }
1094
1095                 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1096                 for(d=rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr;
1097                     d-=sblock.fs_frag) {
1098                         DBG_TRC;
1099                         DBG_PRINT1("d=%jd\n", (intmax_t)d);
1100                         setblock(&sblock, cg_blksfree(&acg),
1101                             (d%sblock.fs_fpg)/sblock.fs_frag);
1102                         acg.cg_cs.cs_nbfree++;
1103                         sblock.fs_cstotal.cs_nbfree++;
1104                         if(sblock.fs_contigsumsize > 0) {
1105                                 setbit(cg_clustersfree(&acg),
1106                                     (d%sblock.fs_fpg)/sblock.fs_frag);
1107                                 /*
1108                                  * The last cluster size is already set up.
1109                                  */
1110                                 if(lcs < sblock.fs_contigsumsize) {
1111                                         if(lcs) {
1112                                                 cg_clustersum(&acg)[lcs]--;
1113                                         }
1114                                         lcs++;
1115                                         cg_clustersum(&acg)[lcs]++;
1116                                 }
1117                         }
1118                 }
1119                 *cs = acg.cg_cs;
1120
1121                 /*
1122                  * Now write the former cylinder group containing the cylinder
1123                  * summary back to disk.
1124                  */
1125                 wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)),
1126                     (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1127                 DBG_PRINT0("oscg written\n");
1128                 DBG_DUMP_CG(&sblock,
1129                     "old summary cg",
1130                     &acg);
1131
1132                 /*
1133                  * Find the beginning of the new cylinder group containing the
1134                  * cylinder summary.
1135                  */
1136                 sblock.fs_csaddr=cgdmin(&sblock, osblock.fs_ncg);
1137                 ncscg=dtog(&sblock, sblock.fs_csaddr);
1138                 cs=fscs+ncscg;
1139
1140
1141                 /*
1142                  * If Nflag is specified, we would now read random data instead
1143                  * of an empty cg structure from disk. So we can't simulate that
1144                  * part for now.
1145                  */
1146                 if(Nflag) {
1147                         DBG_PRINT0("nscg update skipped\n");
1148                         DBG_LEAVE;
1149                         return;
1150                 }
1151
1152                 /*
1153                  * Read the future cylinder group containing the cylinder
1154                  * summary from disk, and make a copy.
1155                  */
1156                 rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1157                     (size_t)sblock.fs_cgsize, (void *)&aocg, fsi);
1158                 DBG_PRINT0("nscg read\n");
1159                 DBG_DUMP_CG(&sblock,
1160                     "new summary cg",
1161                     &aocg);
1162
1163                 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
1164
1165                 /*
1166                  * Allocate all complete blocks used by the new cylinder
1167                  * summary.
1168                  */
1169                 for(d=sblock.fs_csaddr; d+sblock.fs_frag <=
1170                     sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize);
1171                     d+=sblock.fs_frag) {
1172                         clrblock(&sblock, cg_blksfree(&acg),
1173                             (d%sblock.fs_fpg)/sblock.fs_frag);
1174                         acg.cg_cs.cs_nbfree--;
1175                         sblock.fs_cstotal.cs_nbfree--;
1176                         if(sblock.fs_contigsumsize > 0) {
1177                                 clrbit(cg_clustersfree(&acg),
1178                                     (d%sblock.fs_fpg)/sblock.fs_frag);
1179                         }
1180                 }
1181
1182                 /*
1183                  * Allocate all fragments used by the cylinder summary in the
1184                  * last block.
1185                  */
1186                 if(d<sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize)) {
1187                         for(; d-sblock.fs_csaddr<
1188                             sblock.fs_cssize/sblock.fs_fsize;
1189                             d++) {
1190                                 clrbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1191                                 acg.cg_cs.cs_nffree--;
1192                                 sblock.fs_cstotal.cs_nffree--;
1193                         }
1194                         acg.cg_cs.cs_nbfree--;
1195                         acg.cg_cs.cs_nffree+=sblock.fs_frag;
1196                         sblock.fs_cstotal.cs_nbfree--;
1197                         sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1198                         if(sblock.fs_contigsumsize > 0) {
1199                                 clrbit(cg_clustersfree(&acg),
1200                                     (d%sblock.fs_fpg)/sblock.fs_frag);
1201                         }
1202
1203                         frag_adjust(d%sblock.fs_fpg, +1);
1204                 }
1205                 /*
1206                  * XXX  Handle the cluster statistics here in the case this
1207                  *      cylinder group is now almost full, and the remaining
1208                  *      space is less then the maximum cluster size. This is
1209                  *      probably not needed, as you would hardly find a file
1210                  *      system which has only MAXCSBUFS+FS_MAXCONTIG of free
1211                  *      space right behind the cylinder group information in
1212                  *      any new cylinder group.
1213                  */
1214
1215                 /*
1216                  * Update our statistics in the cylinder summary.
1217                  */
1218                 *cs = acg.cg_cs;
1219
1220                 /*
1221                  * Write the new cylinder group containing the cylinder summary
1222                  * back to disk.
1223                  */
1224                 wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1225                     (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1226                 DBG_PRINT0("nscg written\n");
1227                 DBG_DUMP_CG(&sblock,
1228                     "new summary cg",
1229                     &acg);
1230
1231                 DBG_LEAVE;
1232                 return;
1233         }
1234         /*
1235          * We have got enough of space in the current cylinder group, so we
1236          * can relocate just a few blocks, and let the summary information
1237          * grow in place where it is right now.
1238          */
1239         DBG_TRC;
1240
1241         cbase = cgbase(&osblock, ocscg);        /* old and new are equal */
1242         dupper = sblock.fs_csaddr - cbase +
1243             howmany(sblock.fs_cssize, sblock.fs_fsize);
1244         odupper = osblock.fs_csaddr - cbase +
1245             howmany(osblock.fs_cssize, osblock.fs_fsize);
1246
1247         sblock.fs_dsize -= dupper-odupper;
1248
1249         /*
1250          * Allocate the space for the array of blocks to be relocated.
1251          */
1252         bp=(struct gfs_bpp *)malloc(((dupper-odupper)/sblock.fs_frag+2)*
1253             sizeof(struct gfs_bpp));
1254         if(bp == NULL) {
1255                 errx(1, "malloc failed");
1256         }
1257         memset((char *)bp, 0, ((dupper-odupper)/sblock.fs_frag+2)*
1258             sizeof(struct gfs_bpp));
1259
1260         /*
1261          * Lock all new frags needed for the cylinder group summary. This is
1262          * done per fragment in the first and last block of the new required
1263          * area, and per block for all other blocks.
1264          *
1265          * Handle the first new block here (but only if some fragments where
1266          * already used for the cylinder summary).
1267          */
1268         ind=0;
1269         frag_adjust(odupper, -1);
1270         for(d=odupper; ((d<dupper)&&(d%sblock.fs_frag)); d++) {
1271                 DBG_PRINT1("scg first frag check loop d=%jd\n",
1272                     (intmax_t)d);
1273                 if(isclr(cg_blksfree(&acg), d)) {
1274                         if (!ind) {
1275                                 bp[ind].old=d/sblock.fs_frag;
1276                                 bp[ind].flags|=GFS_FL_FIRST;
1277                                 if(roundup(d, sblock.fs_frag) >= dupper) {
1278                                         bp[ind].flags|=GFS_FL_LAST;
1279                                 }
1280                                 ind++;
1281                         }
1282                 } else {
1283                         clrbit(cg_blksfree(&acg), d);
1284                         acg.cg_cs.cs_nffree--;
1285                         sblock.fs_cstotal.cs_nffree--;
1286                 }
1287                 /*
1288                  * No cluster handling is needed here, as there was at least
1289                  * one fragment in use by the cylinder summary in the old
1290                  * file system.
1291                  * No block-free counter handling here as this block was not
1292                  * a free block.
1293                  */
1294         }
1295         frag_adjust(odupper, 1);
1296
1297         /*
1298          * Handle all needed complete blocks here.
1299          */
1300         for(; d+sblock.fs_frag<=dupper; d+=sblock.fs_frag) {
1301                 DBG_PRINT1("scg block check loop d=%jd\n",
1302                     (intmax_t)d);
1303                 if(!isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1304                         for(f=d; f<d+sblock.fs_frag; f++) {
1305                                 if(isset(cg_blksfree(&aocg), f)) {
1306                                         acg.cg_cs.cs_nffree--;
1307                                         sblock.fs_cstotal.cs_nffree--;
1308                                 }
1309                         }
1310                         clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1311                         bp[ind].old=d/sblock.fs_frag;
1312                         ind++;
1313                 } else {
1314                         clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1315                         acg.cg_cs.cs_nbfree--;
1316                         sblock.fs_cstotal.cs_nbfree--;
1317                         if(sblock.fs_contigsumsize > 0) {
1318                                 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1319                                 for(lcs=0, l=(d/sblock.fs_frag)+1;
1320                                     lcs<sblock.fs_contigsumsize;
1321                                     l++, lcs++ ) {
1322                                         if(isclr(cg_clustersfree(&acg),l)){
1323                                                 break;
1324                                         }
1325                                 }
1326                                 if(lcs < sblock.fs_contigsumsize) {
1327                                         cg_clustersum(&acg)[lcs+1]--;
1328                                         if(lcs) {
1329                                                 cg_clustersum(&acg)[lcs]++;
1330                                         }
1331                                 }
1332                         }
1333                 }
1334                 /*
1335                  * No fragment counter handling is needed here, as this finally
1336                  * doesn't change after the relocation.
1337                  */
1338         }
1339
1340         /*
1341          * Handle all fragments needed in the last new affected block.
1342          */
1343         if(d<dupper) {
1344                 frag_adjust(dupper-1, -1);
1345
1346                 if(isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1347                         acg.cg_cs.cs_nbfree--;
1348                         sblock.fs_cstotal.cs_nbfree--;
1349                         acg.cg_cs.cs_nffree+=sblock.fs_frag;
1350                         sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1351                         if(sblock.fs_contigsumsize > 0) {
1352                                 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1353                                 for(lcs=0, l=(d/sblock.fs_frag)+1;
1354                                     lcs<sblock.fs_contigsumsize;
1355                                     l++, lcs++ ) {
1356                                         if(isclr(cg_clustersfree(&acg),l)){
1357                                                 break;
1358                                         }
1359                                 }
1360                                 if(lcs < sblock.fs_contigsumsize) {
1361                                         cg_clustersum(&acg)[lcs+1]--;
1362                                         if(lcs) {
1363                                                 cg_clustersum(&acg)[lcs]++;
1364                                         }
1365                                 }
1366                         }
1367                 }
1368
1369                 for(; d<dupper; d++) {
1370                         DBG_PRINT1("scg second frag check loop d=%jd\n",
1371                             (intmax_t)d);
1372                         if(isclr(cg_blksfree(&acg), d)) {
1373                                 bp[ind].old=d/sblock.fs_frag;
1374                                 bp[ind].flags|=GFS_FL_LAST;
1375                         } else {
1376                                 clrbit(cg_blksfree(&acg), d);
1377                                 acg.cg_cs.cs_nffree--;
1378                                 sblock.fs_cstotal.cs_nffree--;
1379                         }
1380                 }
1381                 if(bp[ind].flags & GFS_FL_LAST) { /* we have to advance here */
1382                         ind++;
1383                 }
1384                 frag_adjust(dupper-1, 1);
1385         }
1386
1387         /*
1388          * If we found a block to relocate just do so.
1389          */
1390         if(ind) {
1391                 for(i=0; i<ind; i++) {
1392                         if(!bp[i].old) { /* no more blocks listed */
1393                                 /*
1394                                  * XXX  A relative blocknumber should not be
1395                                  *      zero, which is not explicitly
1396                                  *      guaranteed by our code.
1397                                  */
1398                                 break;
1399                         }
1400                         /*
1401                          * Allocate a complete block in the same (current)
1402                          * cylinder group.
1403                          */
1404                         bp[i].new=alloc()/sblock.fs_frag;
1405
1406                         /*
1407                          * There is no frag_adjust() needed for the new block
1408                          * as it will have no fragments yet :-).
1409                          */
1410                         for(f=bp[i].old*sblock.fs_frag,
1411                             g=bp[i].new*sblock.fs_frag;
1412                             f<(bp[i].old+1)*sblock.fs_frag;
1413                             f++, g++) {
1414                                 if(isset(cg_blksfree(&aocg), f)) {
1415                                         setbit(cg_blksfree(&acg), g);
1416                                         acg.cg_cs.cs_nffree++;
1417                                         sblock.fs_cstotal.cs_nffree++;
1418                                 }
1419                         }
1420
1421                         /*
1422                          * Special handling is required if this was the first
1423                          * block. We have to consider the fragments which were
1424                          * used by the cylinder summary in the original block
1425                          * which re to be free in the copy of our block.  We
1426                          * have to be careful if this first block happens to
1427                          * be also the last block to be relocated.
1428                          */
1429                         if(bp[i].flags & GFS_FL_FIRST) {
1430                                 for(f=bp[i].old*sblock.fs_frag,
1431                                     g=bp[i].new*sblock.fs_frag;
1432                                     f<odupper;
1433                                     f++, g++) {
1434                                         setbit(cg_blksfree(&acg), g);
1435                                         acg.cg_cs.cs_nffree++;
1436                                         sblock.fs_cstotal.cs_nffree++;
1437                                 }
1438                                 if(!(bp[i].flags & GFS_FL_LAST)) {
1439                                         frag_adjust(bp[i].new*sblock.fs_frag,1);
1440                                 }
1441                         }
1442
1443                         /*
1444                          * Special handling is required if this is the last
1445                          * block to be relocated.
1446                          */
1447                         if(bp[i].flags & GFS_FL_LAST) {
1448                                 frag_adjust(bp[i].new*sblock.fs_frag, 1);
1449                                 frag_adjust(bp[i].old*sblock.fs_frag, -1);
1450                                 for(f=dupper;
1451                                     f<roundup(dupper, sblock.fs_frag);
1452                                     f++) {
1453                                         if(isclr(cg_blksfree(&acg), f)) {
1454                                                 setbit(cg_blksfree(&acg), f);
1455                                                 acg.cg_cs.cs_nffree++;
1456                                                 sblock.fs_cstotal.cs_nffree++;
1457                                         }
1458                                 }
1459                                 frag_adjust(bp[i].old*sblock.fs_frag, 1);
1460                         }
1461
1462                         /*
1463                          * !!! Attach the cylindergroup offset here.
1464                          */
1465                         bp[i].old+=cbase/sblock.fs_frag;
1466                         bp[i].new+=cbase/sblock.fs_frag;
1467
1468                         /*
1469                          * Copy the content of the block.
1470                          */
1471                         /*
1472                          * XXX  Here we will have to implement a copy on write
1473                          *      in the case we have any active snapshots.
1474                          */
1475                         rdfs(fsbtodb(&sblock, bp[i].old*sblock.fs_frag),
1476                             (size_t)sblock.fs_bsize, (void *)&ablk, fsi);
1477                         wtfs(fsbtodb(&sblock, bp[i].new*sblock.fs_frag),
1478                             (size_t)sblock.fs_bsize, (void *)&ablk, fso, Nflag);
1479                         DBG_DUMP_HEX(&sblock,
1480                             "copied full block",
1481                             (unsigned char *)&ablk);
1482
1483                         DBG_PRINT2("scg (%jd->%jd) block relocated\n",
1484                             (intmax_t)bp[i].old,
1485                             (intmax_t)bp[i].new);
1486                 }
1487
1488                 /*
1489                  * Now we have to update all references to any fragment which
1490                  * belongs to any block relocated. We iterate now over all
1491                  * cylinder groups, within those over all non zero length
1492                  * inodes.
1493                  */
1494                 for(cylno=0; cylno<osblock.fs_ncg; cylno++) {
1495                         DBG_PRINT1("scg doing cg (%d)\n",
1496                             cylno);
1497                         for(inc=osblock.fs_ipg-1 ; inc>0 ; inc--) {
1498                                 updrefs(cylno, (ino_t)inc, bp, fsi, fso, Nflag);
1499                         }
1500                 }
1501
1502                 /*
1503                  * All inodes are checked, now make sure the number of
1504                  * references found make sense.
1505                  */
1506                 for(i=0; i<ind; i++) {
1507                         if(!bp[i].found || (bp[i].found>sblock.fs_frag)) {
1508                                 warnx("error: %jd refs found for block %jd.",
1509                                     (intmax_t)bp[i].found, (intmax_t)bp[i].old);
1510                         }
1511
1512                 }
1513         }
1514         /*
1515          * The following statistics are not changed here:
1516          *     sblock.fs_cstotal.cs_ndir
1517          *     sblock.fs_cstotal.cs_nifree
1518          * The following statistics were already updated on the fly:
1519          *     sblock.fs_cstotal.cs_nffree
1520          *     sblock.fs_cstotal.cs_nbfree
1521          * As the statistics for this cylinder group are ready, copy it to
1522          * the summary information array.
1523          */
1524
1525         *cs = acg.cg_cs;
1526
1527         /*
1528          * Write summary cylinder group back to disk.
1529          */
1530         wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)), (size_t)sblock.fs_cgsize,
1531             (void *)&acg, fso, Nflag);
1532         DBG_PRINT0("scg written\n");
1533         DBG_DUMP_CG(&sblock,
1534             "new summary cg",
1535             &acg);
1536
1537         DBG_LEAVE;
1538         return;
1539 }
1540
1541 /* ************************************************************** rdfs ***** */
1542 /*
1543  * Here we read some block(s) from disk.
1544  */
1545 static void
1546 rdfs(ufs2_daddr_t bno, size_t size, void *bf, int fsi)
1547 {
1548         DBG_FUNC("rdfs")
1549         ssize_t n;
1550
1551         DBG_ENTER;
1552
1553         if (bno < 0) {
1554                 err(32, "rdfs: attempting to read negative block number");
1555         }
1556         if (lseek(fsi, (off_t)bno * DEV_BSIZE, 0) < 0) {
1557                 err(33, "rdfs: seek error: %jd", (intmax_t)bno);
1558         }
1559         n = read(fsi, bf, size);
1560         if (n != (ssize_t)size) {
1561                 err(34, "rdfs: read error: %jd", (intmax_t)bno);
1562         }
1563
1564         DBG_LEAVE;
1565         return;
1566 }
1567
1568 /* ************************************************************** wtfs ***** */
1569 /*
1570  * Here we write some block(s) to disk.
1571  */
1572 static void
1573 wtfs(ufs2_daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag)
1574 {
1575         DBG_FUNC("wtfs")
1576         ssize_t n;
1577
1578         DBG_ENTER;
1579
1580         if (Nflag) {
1581                 DBG_LEAVE;
1582                 return;
1583         }
1584         if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0) {
1585                 err(35, "wtfs: seek error: %ld", (long)bno);
1586         }
1587         n = write(fso, bf, size);
1588         if (n != (ssize_t)size) {
1589                 err(36, "wtfs: write error: %ld", (long)bno);
1590         }
1591
1592         DBG_LEAVE;
1593         return;
1594 }
1595
1596 /* ************************************************************* alloc ***** */
1597 /*
1598  * Here we allocate a free block in the current cylinder group. It is assumed,
1599  * that acg contains the current cylinder group. As we may take a block from
1600  * somewhere in the file system we have to handle cluster summary here.
1601  */
1602 static ufs2_daddr_t
1603 alloc(void)
1604 {
1605         DBG_FUNC("alloc")
1606         ufs2_daddr_t    d, blkno;
1607         int     lcs1, lcs2;
1608         int     l;
1609         int     csmin, csmax;
1610         int     dlower, dupper, dmax;
1611
1612         DBG_ENTER;
1613
1614         if (acg.cg_magic != CG_MAGIC) {
1615                 warnx("acg: bad magic number");
1616                 DBG_LEAVE;
1617                 return (0);
1618         }
1619         if (acg.cg_cs.cs_nbfree == 0) {
1620                 warnx("error: cylinder group ran out of space");
1621                 DBG_LEAVE;
1622                 return (0);
1623         }
1624         /*
1625          * We start seeking for free blocks only from the space available after
1626          * the end of the new grown cylinder summary. Otherwise we allocate a
1627          * block here which we have to relocate a couple of seconds later again
1628          * again, and we are not prepared to to this anyway.
1629          */
1630         blkno=-1;
1631         dlower=cgsblock(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1632         dupper=cgdmin(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1633         dmax=cgbase(&sblock, acg.cg_cgx)+sblock.fs_fpg;
1634         if (dmax > sblock.fs_size) {
1635                 dmax = sblock.fs_size;
1636         }
1637         dmax-=cgbase(&sblock, acg.cg_cgx); /* retransform into cg */
1638         csmin=sblock.fs_csaddr-cgbase(&sblock, acg.cg_cgx);
1639         csmax=csmin+howmany(sblock.fs_cssize, sblock.fs_fsize);
1640         DBG_PRINT3("seek range: dl=%d, du=%d, dm=%d\n",
1641             dlower,
1642             dupper,
1643             dmax);
1644         DBG_PRINT2("range cont: csmin=%d, csmax=%d\n",
1645             csmin,
1646             csmax);
1647
1648         for(d=0; (d<dlower && blkno==-1); d+=sblock.fs_frag) {
1649                 if(d>=csmin && d<=csmax) {
1650                         continue;
1651                 }
1652                 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1653                     d))) {
1654                         blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1655                         break;
1656                 }
1657         }
1658         for(d=dupper; (d<dmax && blkno==-1); d+=sblock.fs_frag) {
1659                 if(d>=csmin && d<=csmax) {
1660                         continue;
1661                 }
1662                 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1663                     d))) {
1664                         blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1665                         break;
1666                 }
1667         }
1668         if(blkno==-1) {
1669                 warnx("internal error: couldn't find promised block in cg");
1670                 DBG_LEAVE;
1671                 return (0);
1672         }
1673
1674         /*
1675          * This is needed if the block was found already in the first loop.
1676          */
1677         d=blkstofrags(&sblock, blkno);
1678
1679         clrblock(&sblock, cg_blksfree(&acg), blkno);
1680         if (sblock.fs_contigsumsize > 0) {
1681                 /*
1682                  * Handle the cluster allocation bitmap.
1683                  */
1684                 clrbit(cg_clustersfree(&acg), blkno);
1685                 /*
1686                  * We possibly have split a cluster here, so we have to do
1687                  * recalculate the sizes of the remaining cluster halves now,
1688                  * and use them for updating the cluster summary information.
1689                  *
1690                  * Lets start with the blocks before our allocated block ...
1691                  */
1692                 for(lcs1=0, l=blkno-1; lcs1<sblock.fs_contigsumsize;
1693                     l--, lcs1++ ) {
1694                         if(isclr(cg_clustersfree(&acg),l)){
1695                                 break;
1696                         }
1697                 }
1698                 /*
1699                  * ... and continue with the blocks right after our allocated
1700                  * block.
1701                  */
1702                 for(lcs2=0, l=blkno+1; lcs2<sblock.fs_contigsumsize;
1703                     l++, lcs2++ ) {
1704                         if(isclr(cg_clustersfree(&acg),l)){
1705                                 break;
1706                         }
1707                 }
1708
1709                 /*
1710                  * Now update all counters.
1711                  */
1712                 cg_clustersum(&acg)[MIN(lcs1+lcs2+1,sblock.fs_contigsumsize)]--;
1713                 if(lcs1) {
1714                         cg_clustersum(&acg)[lcs1]++;
1715                 }
1716                 if(lcs2) {
1717                         cg_clustersum(&acg)[lcs2]++;
1718                 }
1719         }
1720         /*
1721          * Update all statistics based on blocks.
1722          */
1723         acg.cg_cs.cs_nbfree--;
1724         sblock.fs_cstotal.cs_nbfree--;
1725
1726         DBG_LEAVE;
1727         return (d);
1728 }
1729
1730 /* *********************************************************** isblock ***** */
1731 /*
1732  * Here we check if all frags of a block are free. For more details again
1733  * please see the source of newfs(8), as this function is taken over almost
1734  * unchanged.
1735  */
1736 static int
1737 isblock(struct fs *fs, unsigned char *cp, int h)
1738 {
1739         DBG_FUNC("isblock")
1740         unsigned char   mask;
1741
1742         DBG_ENTER;
1743
1744         switch (fs->fs_frag) {
1745         case 8:
1746                 DBG_LEAVE;
1747                 return (cp[h] == 0xff);
1748         case 4:
1749                 mask = 0x0f << ((h & 0x1) << 2);
1750                 DBG_LEAVE;
1751                 return ((cp[h >> 1] & mask) == mask);
1752         case 2:
1753                 mask = 0x03 << ((h & 0x3) << 1);
1754                 DBG_LEAVE;
1755                 return ((cp[h >> 2] & mask) == mask);
1756         case 1:
1757                 mask = 0x01 << (h & 0x7);
1758                 DBG_LEAVE;
1759                 return ((cp[h >> 3] & mask) == mask);
1760         default:
1761                 fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
1762                 DBG_LEAVE;
1763                 return (0);
1764         }
1765 }
1766
1767 /* ********************************************************** clrblock ***** */
1768 /*
1769  * Here we allocate a complete block in the block map. For more details again
1770  * please see the source of newfs(8), as this function is taken over almost
1771  * unchanged.
1772  */
1773 static void
1774 clrblock(struct fs *fs, unsigned char *cp, int h)
1775 {
1776         DBG_FUNC("clrblock")
1777
1778         DBG_ENTER;
1779
1780         switch ((fs)->fs_frag) {
1781         case 8:
1782                 cp[h] = 0;
1783                 break;
1784         case 4:
1785                 cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1786                 break;
1787         case 2:
1788                 cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1789                 break;
1790         case 1:
1791                 cp[h >> 3] &= ~(0x01 << (h & 0x7));
1792                 break;
1793         default:
1794                 warnx("clrblock bad fs_frag %d", fs->fs_frag);
1795                 break;
1796         }
1797
1798         DBG_LEAVE;
1799         return;
1800 }
1801
1802 /* ********************************************************** setblock ***** */
1803 /*
1804  * Here we free a complete block in the free block map. For more details again
1805  * please see the source of newfs(8), as this function is taken over almost
1806  * unchanged.
1807  */
1808 static void
1809 setblock(struct fs *fs, unsigned char *cp, int h)
1810 {
1811         DBG_FUNC("setblock")
1812
1813         DBG_ENTER;
1814
1815         switch (fs->fs_frag) {
1816         case 8:
1817                 cp[h] = 0xff;
1818                 break;
1819         case 4:
1820                 cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1821                 break;
1822         case 2:
1823                 cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1824                 break;
1825         case 1:
1826                 cp[h >> 3] |= (0x01 << (h & 0x7));
1827                 break;
1828         default:
1829                 warnx("setblock bad fs_frag %d", fs->fs_frag);
1830                 break;
1831         }
1832
1833         DBG_LEAVE;
1834         return;
1835 }
1836
1837 /* ************************************************************ ginode ***** */
1838 /*
1839  * This function provides access to an individual inode. We find out in which
1840  * block the requested inode is located, read it from disk if needed, and
1841  * return the pointer into that block. We maintain a cache of one block to
1842  * not read the same block again and again if we iterate linearly over all
1843  * inodes.
1844  */
1845 static union dinode *
1846 ginode(ino_t inumber, int fsi, int cg)
1847 {
1848         DBG_FUNC("ginode")
1849         static ino_t    startinum = 0;  /* first inode in cached block */
1850
1851         DBG_ENTER;
1852
1853         /*
1854          * The inumber passed in is relative to the cg, so use it here to see
1855          * if the inode has been allocated yet.
1856          */
1857         if (isclr(cg_inosused(&aocg), inumber)) {
1858                 DBG_LEAVE;
1859                 return NULL;
1860         }
1861         /*
1862          * Now make the inumber relative to the entire inode space so it can
1863          * be sanity checked.
1864          */
1865         inumber += (cg * sblock.fs_ipg);
1866         if (inumber < ROOTINO) {
1867                 DBG_LEAVE;
1868                 return NULL;
1869         }
1870         if (inumber > maxino)
1871                 errx(8, "bad inode number %d to ginode", inumber);
1872         if (startinum == 0 ||
1873             inumber < startinum || inumber >= startinum + INOPB(&sblock)) {
1874                 inoblk = fsbtodb(&sblock, ino_to_fsba(&sblock, inumber));
1875                 rdfs(inoblk, (size_t)sblock.fs_bsize, inobuf, fsi);
1876                 startinum = (inumber / INOPB(&sblock)) * INOPB(&sblock);
1877         }
1878         DBG_LEAVE;
1879         if (sblock.fs_magic == FS_UFS1_MAGIC)
1880                 return (union dinode *)((uintptr_t)inobuf +
1881                     (inumber % INOPB(&sblock)) * sizeof(struct ufs1_dinode));
1882         return (union dinode *)((uintptr_t)inobuf +
1883             (inumber % INOPB(&sblock)) * sizeof(struct ufs2_dinode));
1884 }
1885
1886 /* ****************************************************** charsperline ***** */
1887 /*
1888  * Figure out how many lines our current terminal has. For more details again
1889  * please see the source of newfs(8), as this function is taken over almost
1890  * unchanged.
1891  */
1892 static int
1893 charsperline(void)
1894 {
1895         DBG_FUNC("charsperline")
1896         int     columns;
1897         char    *cp;
1898         struct winsize  ws;
1899
1900         DBG_ENTER;
1901
1902         columns = 0;
1903         if (ioctl(0, TIOCGWINSZ, &ws) != -1) {
1904                 columns = ws.ws_col;
1905         }
1906         if (columns == 0 && (cp = getenv("COLUMNS"))) {
1907                 columns = atoi(cp);
1908         }
1909         if (columns == 0) {
1910                 columns = 80;   /* last resort */
1911         }
1912
1913         DBG_LEAVE;
1914         return columns;
1915 }
1916
1917 /* ****************************************************** get_dev_size ***** */
1918 /*
1919  * Get the size of the partition if we can't figure it out from the disklabel,
1920  * e.g. from vinum volumes.
1921  */
1922 static void
1923 get_dev_size(int fd, int *size)
1924 {
1925    int sectorsize;
1926    off_t mediasize;
1927
1928    if (ioctl(fd, DIOCGSECTORSIZE, &sectorsize) == -1)
1929         err(1,"DIOCGSECTORSIZE");
1930    if (ioctl(fd, DIOCGMEDIASIZE, &mediasize) == -1)
1931         err(1,"DIOCGMEDIASIZE");
1932
1933    if (sectorsize <= 0)
1934        errx(1, "bogus sectorsize: %d", sectorsize);
1935
1936    *size = mediasize / sectorsize;
1937 }
1938
1939 /* ************************************************************** main ***** */
1940 /*
1941  * growfs(8)  is a utility which allows to increase the size of an existing
1942  * ufs file system. Currently this can only be done on unmounted file system.
1943  * It recognizes some command line options to specify the new desired size,
1944  * and it does some basic checkings. The old file system size is determined
1945  * and after some more checks like we can really access the new last block
1946  * on the disk etc. we calculate the new parameters for the superblock. After
1947  * having done this we just call growfs() which will do the work.  Before
1948  * we finish the only thing left is to update the disklabel.
1949  * We still have to provide support for snapshots. Therefore we first have to
1950  * understand what data structures are always replicated in the snapshot on
1951  * creation, for all other blocks we touch during our procedure, we have to
1952  * keep the old blocks unchanged somewhere available for the snapshots. If we
1953  * are lucky, then we only have to handle our blocks to be relocated in that
1954  * way.
1955  * Also we have to consider in what order we actually update the critical
1956  * data structures of the file system to make sure, that in case of a disaster
1957  * fsck(8) is still able to restore any lost data.
1958  * The foreseen last step then will be to provide for growing even mounted
1959  * file systems. There we have to extend the mount() system call to provide
1960  * userland access to the file system locking facility.
1961  */
1962 int
1963 main(int argc, char **argv)
1964 {
1965         DBG_FUNC("main")
1966         char    *device, *special, *cp;
1967         int     ch;
1968         unsigned int    size=0;
1969         size_t  len;
1970         unsigned int    Nflag=0;
1971         int     ExpertFlag=0;
1972         struct stat     st;
1973         struct disklabel        *lp;
1974         struct partition        *pp;
1975         int     i,fsi,fso;
1976     u_int32_t p_size;
1977         char    reply[5];
1978 #ifdef FSMAXSNAP
1979         int     j;
1980 #endif /* FSMAXSNAP */
1981
1982         DBG_ENTER;
1983
1984         while((ch=getopt(argc, argv, "Ns:vy")) != -1) {
1985                 switch(ch) {
1986                 case 'N':
1987                         Nflag=1;
1988                         break;
1989                 case 's':
1990                         size=(size_t)atol(optarg);
1991                         if(size<1) {
1992                                 usage();
1993                         }
1994                         break;
1995                 case 'v': /* for compatibility to newfs */
1996                         break;
1997                 case 'y':
1998                         ExpertFlag=1;
1999                         break;
2000                 case '?':
2001                         /* FALLTHROUGH */
2002                 default:
2003                         usage();
2004                 }
2005         }
2006         argc -= optind;
2007         argv += optind;
2008
2009         if(argc != 1) {
2010                 usage();
2011         }
2012         device=*argv;
2013
2014         /*
2015          * Now try to guess the (raw)device name.
2016          */
2017         if (0 == strrchr(device, '/')) {
2018                 /*
2019                  * No path prefix was given, so try in that order:
2020                  *     /dev/r%s
2021                  *     /dev/%s
2022                  *     /dev/vinum/r%s
2023                  *     /dev/vinum/%s.
2024                  *
2025                  * FreeBSD now doesn't distinguish between raw and block
2026                  * devices any longer, but it should still work this way.
2027                  */
2028                 len=strlen(device)+strlen(_PATH_DEV)+2+strlen("vinum/");
2029                 special=(char *)malloc(len);
2030                 if(special == NULL) {
2031                         errx(1, "malloc failed");
2032                 }
2033                 snprintf(special, len, "%sr%s", _PATH_DEV, device);
2034                 if (stat(special, &st) == -1) {
2035                         snprintf(special, len, "%s%s", _PATH_DEV, device);
2036                         if (stat(special, &st) == -1) {
2037                                 snprintf(special, len, "%svinum/r%s",
2038                                     _PATH_DEV, device);
2039                                 if (stat(special, &st) == -1) {
2040                                         /* For now this is the 'last resort' */
2041                                         snprintf(special, len, "%svinum/%s",
2042                                             _PATH_DEV, device);
2043                                 }
2044                         }
2045                 }
2046                 device = special;
2047         }
2048
2049         /*
2050          * Try to access our devices for writing ...
2051          */
2052         if (Nflag) {
2053                 fso = -1;
2054         } else {
2055                 fso = open(device, O_WRONLY);
2056                 if (fso < 0) {
2057                         err(1, "%s", device);
2058                 }
2059         }
2060
2061         /*
2062          * ... and reading.
2063          */
2064         fsi = open(device, O_RDONLY);
2065         if (fsi < 0) {
2066                 err(1, "%s", device);
2067         }
2068
2069         /*
2070          * Try to read a label and guess the slice if not specified. This
2071          * code should guess the right thing and avoid to bother the user
2072          * with the task of specifying the option -v on vinum volumes.
2073          */
2074         cp=device+strlen(device)-1;
2075         lp = get_disklabel(fsi);
2076         pp = NULL;
2077     if (lp != NULL) {
2078         if (isdigit(*cp)) {
2079             pp = &lp->d_partitions[2];
2080         } else if (*cp>='a' && *cp<='h') {
2081             pp = &lp->d_partitions[*cp - 'a'];
2082         } else {
2083             errx(1, "unknown device");
2084         }
2085         p_size = pp->p_size;
2086     } else {
2087         get_dev_size(fsi, &p_size);
2088     }
2089
2090         /*
2091          * Check if that partition is suitable for growing a file system.
2092          */
2093         if (p_size < 1) {
2094                 errx(1, "partition is unavailable");
2095         }
2096
2097         /*
2098          * Read the current superblock, and take a backup.
2099          */
2100         for (i = 0; sblock_try[i] != -1; i++) {
2101                 sblockloc = sblock_try[i] / DEV_BSIZE;
2102                 rdfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&(osblock), fsi);
2103                 if ((osblock.fs_magic == FS_UFS1_MAGIC ||
2104                      (osblock.fs_magic == FS_UFS2_MAGIC &&
2105                       osblock.fs_sblockloc == sblock_try[i])) &&
2106                     osblock.fs_bsize <= MAXBSIZE &&
2107                     osblock.fs_bsize >= (int32_t) sizeof(struct fs))
2108                         break;
2109         }
2110         if (sblock_try[i] == -1) {
2111                 errx(1, "superblock not recognized");
2112         }
2113         memcpy((void *)&fsun1, (void *)&fsun2, sizeof(fsun2));
2114         maxino = sblock.fs_ncg * sblock.fs_ipg;
2115
2116         DBG_OPEN("/tmp/growfs.debug"); /* already here we need a superblock */
2117         DBG_DUMP_FS(&sblock,
2118             "old sblock");
2119
2120         /*
2121          * Determine size to grow to. Default to the full size specified in
2122          * the disk label.
2123          */
2124         sblock.fs_size = dbtofsb(&osblock, p_size);
2125         if (size != 0) {
2126                 if (size > p_size){
2127                         errx(1, "there is not enough space (%d < %d)",
2128                             p_size, size);
2129                 }
2130                 sblock.fs_size = dbtofsb(&osblock, size);
2131         }
2132
2133         /*
2134          * Are we really growing ?
2135          */
2136         if(osblock.fs_size >= sblock.fs_size) {
2137                 errx(1, "we are not growing (%jd->%jd)",
2138                     (intmax_t)osblock.fs_size, (intmax_t)sblock.fs_size);
2139         }
2140
2141
2142 #ifdef FSMAXSNAP
2143         /*
2144          * Check if we find an active snapshot.
2145          */
2146         if(ExpertFlag == 0) {
2147                 for(j=0; j<FSMAXSNAP; j++) {
2148                         if(sblock.fs_snapinum[j]) {
2149                                 errx(1, "active snapshot found in file system; "
2150                                     "please remove all snapshots before "
2151                                     "using growfs");
2152                         }
2153                         if(!sblock.fs_snapinum[j]) { /* list is dense */
2154                                 break;
2155                         }
2156                 }
2157         }
2158 #endif
2159
2160         if (ExpertFlag == 0 && Nflag == 0) {
2161                 printf("We strongly recommend you to make a backup "
2162                     "before growing the file system.\n"
2163                     "Did you backup your data (Yes/No)? ");
2164                 fgets(reply, (int)sizeof(reply), stdin);
2165                 if (strcmp(reply, "Yes\n")){
2166                         printf("\nNothing done\n");
2167                         exit (0);
2168                 }
2169         }
2170
2171         printf("New file system size is %jd frags\n", (intmax_t)sblock.fs_size);
2172
2173         /*
2174          * Try to access our new last block in the file system. Even if we
2175          * later on realize we have to abort our operation, on that block
2176          * there should be no data, so we can't destroy something yet.
2177          */
2178         wtfs((ufs2_daddr_t)p_size-1, (size_t)DEV_BSIZE, (void *)&sblock,
2179             fso, Nflag);
2180
2181         /*
2182          * Now calculate new superblock values and check for reasonable
2183          * bound for new file system size:
2184          *     fs_size:    is derived from label or user input
2185          *     fs_dsize:   should get updated in the routines creating or
2186          *                 updating the cylinder groups on the fly
2187          *     fs_cstotal: should get updated in the routines creating or
2188          *                 updating the cylinder groups
2189          */
2190
2191         /*
2192          * Update the number of cylinders and cylinder groups in the file system.
2193          */
2194         if (sblock.fs_magic == FS_UFS1_MAGIC) {
2195                 sblock.fs_old_ncyl =
2196                     sblock.fs_size * sblock.fs_old_nspf / sblock.fs_old_spc;
2197                 if (sblock.fs_size * sblock.fs_old_nspf >
2198                     sblock.fs_old_ncyl * sblock.fs_old_spc)
2199                         sblock.fs_old_ncyl++;
2200         }
2201         sblock.fs_ncg = howmany(sblock.fs_size, sblock.fs_fpg);
2202         maxino = sblock.fs_ncg * sblock.fs_ipg;
2203
2204         if (sblock.fs_size % sblock.fs_fpg != 0 &&
2205             sblock.fs_size % sblock.fs_fpg < cgdmin(&sblock, sblock.fs_ncg)) {
2206                 /*
2207                  * The space in the new last cylinder group is too small,
2208                  * so revert back.
2209                  */
2210                 sblock.fs_ncg--;
2211                 if (sblock.fs_magic == FS_UFS1_MAGIC)
2212                         sblock.fs_old_ncyl = sblock.fs_ncg * sblock.fs_old_cpg;
2213                 printf("Warning: %jd sector(s) cannot be allocated.\n",
2214                     (intmax_t)fsbtodb(&sblock, sblock.fs_size % sblock.fs_fpg));
2215                 sblock.fs_size = sblock.fs_ncg * sblock.fs_fpg;
2216                 maxino -= sblock.fs_ipg;
2217         }
2218
2219         /*
2220          * Update the space for the cylinder group summary information in the
2221          * respective cylinder group data area.
2222          */
2223         sblock.fs_cssize =
2224             fragroundup(&sblock, sblock.fs_ncg * sizeof(struct csum));
2225
2226         if(osblock.fs_size >= sblock.fs_size) {
2227                 errx(1, "not enough new space");
2228         }
2229
2230         DBG_PRINT0("sblock calculated\n");
2231
2232         /*
2233          * Ok, everything prepared, so now let's do the tricks.
2234          */
2235         growfs(fsi, fso, Nflag);
2236
2237         /*
2238          * Update the disk label.
2239          */
2240     if (!unlabeled) {
2241         pp->p_fsize = sblock.fs_fsize;
2242         pp->p_frag = sblock.fs_frag;
2243         pp->p_cpg = sblock.fs_fpg;
2244
2245         return_disklabel(fso, lp, Nflag);
2246         DBG_PRINT0("label rewritten\n");
2247     }
2248
2249         close(fsi);
2250         if(fso>-1) close(fso);
2251
2252         DBG_CLOSE;
2253
2254         DBG_LEAVE;
2255         return 0;
2256 }
2257
2258 /* ************************************************** return_disklabel ***** */
2259 /*
2260  * Write the updated disklabel back to disk.
2261  */
2262 static void
2263 return_disklabel(int fd, struct disklabel *lp, unsigned int Nflag)
2264 {
2265         DBG_FUNC("return_disklabel")
2266         u_short sum;
2267         u_short *ptr;
2268
2269         DBG_ENTER;
2270
2271         if(!lp) {
2272                 DBG_LEAVE;
2273                 return;
2274         }
2275         if(!Nflag) {
2276                 lp->d_checksum=0;
2277                 sum = 0;
2278                 ptr=(u_short *)lp;
2279
2280                 /*
2281                  * recalculate checksum
2282                  */
2283                 while(ptr < (u_short *)&lp->d_partitions[lp->d_npartitions]) {
2284                         sum ^= *ptr++;
2285                 }
2286                 lp->d_checksum=sum;
2287
2288                 if (ioctl(fd, DIOCWDINFO, (char *)lp) < 0) {
2289                         errx(1, "DIOCWDINFO failed");
2290                 }
2291         }
2292         free(lp);
2293
2294         DBG_LEAVE;
2295         return ;
2296 }
2297
2298 /* ***************************************************** get_disklabel ***** */
2299 /*
2300  * Read the disklabel from disk.
2301  */
2302 static struct disklabel *
2303 get_disklabel(int fd)
2304 {
2305         DBG_FUNC("get_disklabel")
2306         static struct   disklabel *lab;
2307
2308         DBG_ENTER;
2309
2310         lab=(struct disklabel *)malloc(sizeof(struct disklabel));
2311         if (!lab)
2312                 errx(1, "malloc failed");
2313
2314     if (!ioctl(fd, DIOCGDINFO, (char *)lab))
2315         return (lab);
2316
2317     unlabeled++;
2318
2319         DBG_LEAVE;
2320         return (NULL);
2321 }
2322
2323
2324 /* ************************************************************* usage ***** */
2325 /*
2326  * Dump a line of usage.
2327  */
2328 static void
2329 usage(void)
2330 {
2331         DBG_FUNC("usage")
2332
2333         DBG_ENTER;
2334
2335         fprintf(stderr, "usage: growfs [-Ny] [-s size] special\n");
2336
2337         DBG_LEAVE;
2338         exit(1);
2339 }
2340
2341 /* *********************************************************** updclst ***** */
2342 /*
2343  * This updates most parameters and the bitmap related to cluster. We have to
2344  * assume that sblock, osblock, acg are set up.
2345  */
2346 static void
2347 updclst(int block)
2348 {
2349         DBG_FUNC("updclst")
2350         static int      lcs=0;
2351
2352         DBG_ENTER;
2353
2354         if(sblock.fs_contigsumsize < 1) { /* no clustering */
2355                 return;
2356         }
2357         /*
2358          * update cluster allocation map
2359          */
2360         setbit(cg_clustersfree(&acg), block);
2361
2362         /*
2363          * update cluster summary table
2364          */
2365         if(!lcs) {
2366                 /*
2367                  * calculate size for the trailing cluster
2368                  */
2369                 for(block--; lcs<sblock.fs_contigsumsize; block--, lcs++ ) {
2370                         if(isclr(cg_clustersfree(&acg), block)){
2371                                 break;
2372                         }
2373                 }
2374         }
2375         if(lcs < sblock.fs_contigsumsize) {
2376                 if(lcs) {
2377                         cg_clustersum(&acg)[lcs]--;
2378                 }
2379                 lcs++;
2380                 cg_clustersum(&acg)[lcs]++;
2381         }
2382
2383         DBG_LEAVE;
2384         return;
2385 }
2386
2387 /* *********************************************************** updrefs ***** */
2388 /*
2389  * This updates all references to relocated blocks for the given inode.  The
2390  * inode is given as number within the cylinder group, and the number of the
2391  * cylinder group.
2392  */
2393 static void
2394 updrefs(int cg, ino_t in, struct gfs_bpp *bp, int fsi, int fso, unsigned int
2395     Nflag)
2396 {
2397         DBG_FUNC("updrefs")
2398         ufs_lbn_t       len, lbn, numblks;
2399         ufs2_daddr_t    iptr, blksperindir;
2400         union dinode    *ino;
2401         int             i, mode, inodeupdated;
2402
2403         DBG_ENTER;
2404
2405         ino = ginode(in, fsi, cg);
2406         if (ino == NULL) {
2407                 DBG_LEAVE;
2408                 return;
2409         }
2410         mode = DIP(ino, di_mode) & IFMT;
2411         if (mode != IFDIR && mode != IFREG && mode != IFLNK) {
2412                 DBG_LEAVE;
2413                 return; /* only check DIR, FILE, LINK */
2414         }
2415         if (mode == IFLNK && 
2416             DIP(ino, di_size) < (u_int64_t) sblock.fs_maxsymlinklen) {
2417                 DBG_LEAVE;
2418                 return; /* skip short symlinks */
2419         }
2420         numblks = howmany(DIP(ino, di_size), sblock.fs_bsize);
2421         if (numblks == 0) {
2422                 DBG_LEAVE;
2423                 return; /* skip empty file */
2424         }
2425         if (DIP(ino, di_blocks) == 0) {
2426                 DBG_LEAVE;
2427                 return; /* skip empty swiss cheesy file or old fastlink */
2428         }
2429         DBG_PRINT2("scg checking inode (%d in %d)\n",
2430             in,
2431             cg);
2432
2433         /*
2434          * Check all the blocks.
2435          */
2436         inodeupdated = 0;
2437         len = numblks < NDADDR ? numblks : NDADDR;
2438         for (i = 0; i < len; i++) {
2439                 iptr = DIP(ino, di_db[i]);
2440                 if (iptr == 0)
2441                         continue;
2442                 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2443                         DIP_SET(ino, di_db[i], iptr);
2444                         inodeupdated++;
2445                 }
2446         }
2447         DBG_PRINT0("~~scg direct blocks checked\n");
2448
2449         blksperindir = 1;
2450         len = numblks - NDADDR;
2451         lbn = NDADDR;
2452         for (i = 0; len > 0 && i < NIADDR; i++) {
2453                 iptr = DIP(ino, di_ib[i]);
2454                 if (iptr == 0)
2455                         continue;
2456                 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2457                         DIP_SET(ino, di_ib[i], iptr);
2458                         inodeupdated++;
2459                 }
2460                 indirchk(blksperindir, lbn, iptr, numblks, bp, fsi, fso, Nflag);
2461                 blksperindir *= NINDIR(&sblock);
2462                 lbn += blksperindir;
2463                 len -= blksperindir;
2464                 DBG_PRINT1("scg indirect_%d blocks checked\n", i + 1);
2465         }
2466         if (inodeupdated)
2467                 wtfs(inoblk, sblock.fs_bsize, inobuf, fso, Nflag);
2468
2469         DBG_LEAVE;
2470         return;
2471 }
2472
2473 /*
2474  * Recursively check all the indirect blocks.
2475  */
2476 static void
2477 indirchk(ufs_lbn_t blksperindir, ufs_lbn_t lbn, ufs2_daddr_t blkno,
2478     ufs_lbn_t lastlbn, struct gfs_bpp *bp, int fsi, int fso, unsigned int Nflag)
2479 {
2480         DBG_FUNC("indirchk")
2481         void *ibuf;
2482         int i, last;
2483         ufs2_daddr_t iptr;
2484
2485         DBG_ENTER;
2486
2487         /* read in the indirect block. */
2488         ibuf = malloc(sblock.fs_bsize);
2489         if (!ibuf)
2490                 errx(1, "malloc failed");
2491         rdfs(fsbtodb(&sblock, blkno), (size_t)sblock.fs_bsize, ibuf, fsi);
2492         last = howmany(lastlbn - lbn, blksperindir) < NINDIR(&sblock) ?
2493             howmany(lastlbn - lbn, blksperindir) : NINDIR(&sblock);
2494         for (i = 0; i < last; i++) {
2495                 if (sblock.fs_magic == FS_UFS1_MAGIC)
2496                         iptr = ((ufs1_daddr_t *)ibuf)[i];
2497                 else
2498                         iptr = ((ufs2_daddr_t *)ibuf)[i];
2499                 if (iptr == 0)
2500                         continue;
2501                 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2502                         if (sblock.fs_magic == FS_UFS1_MAGIC)
2503                                 ((ufs1_daddr_t *)ibuf)[i] = iptr;
2504                         else
2505                                 ((ufs2_daddr_t *)ibuf)[i] = iptr;
2506                 }
2507                 if (blksperindir == 1)
2508                         continue;
2509                 indirchk(blksperindir / NINDIR(&sblock), lbn + blksperindir * i,
2510                     iptr, lastlbn, bp, fsi, fso, Nflag);
2511         }
2512         free(ibuf);
2513
2514         DBG_LEAVE;
2515         return;
2516 }