2 * Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz
3 * Copyright (c) 1980, 1989, 1993 The Regents of the University of California.
6 * This code is derived from software contributed to Berkeley by
7 * Christoph Herrmann and Thomas-Henning von Kamptz, Munich and Frankfurt.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
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.
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
38 * $TSHeader: src/sbin/growfs/growfs.c,v 1.5 2000/12/12 19:31:00 tomsoft Exp $
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";
49 #include <sys/cdefs.h>
50 __FBSDID("$FreeBSD$");
52 /* ********************************************************** INCLUDES ***** */
53 #include <sys/param.h>
54 #include <sys/disklabel.h>
55 #include <sys/ioctl.h>
70 #include <ufs/ufs/dinode.h>
71 #include <ufs/ffs/fs.h>
75 /* *************************************************** GLOBALS & TYPES ***** */
77 int _dbg_lvl_ = (DL_INFO); /* DL_TRC */
84 #define sblock fsun1.fs /* the new superblock */
85 #define osblock fsun2.fs /* the old superblock */
88 * Possible superblock locations ordered from most to least likely.
90 static int sblock_try[] = SBLOCKSEARCH;
91 static ufs2_daddr_t sblockloc;
97 #define acg cgun1.cg /* a cylinder cgroup (new) */
98 #define aocg cgun2.cg /* an old cylinder group */
100 static char ablk[MAXBSIZE]; /* a block */
102 static struct csum *fscs; /* cylinder summary */
105 struct ufs1_dinode dp1;
106 struct ufs2_dinode dp2;
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); \
115 (dp)->dp2.field = (val); \
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. */
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.
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 */
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,
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 *);
161 /* ************************************************************ growfs ***** */
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
174 growfs(int fsi, int fso, unsigned int Nflag)
182 static int randinit=0;
190 #else /* not FSIRAND */
198 * Get the cylinder summary into the memory.
200 fscs = (struct csum *)calloc((size_t)1, (size_t)sblock.fs_cssize);
202 errx(1, "calloc failed");
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);
212 struct csum *dbg_csp;
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,
225 #endif /* FS_DEBUG */
226 DBG_PRINT0("fscs read\n");
229 * Do all needed changes in the former last cylinder group.
231 updjcg(osblock.fs_ncg-1, utime, fsi, fso, Nflag);
234 * Dump out summary information about file system.
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,
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");
249 * Now build the cylinders group blocks and
250 * then print out indices of cylinder groups.
252 printf("super-block backups (for fsck -b #) at:\n");
254 width = charsperline();
257 * Iterate for only the new cylinder groups.
259 for (cylno = osblock.fs_ncg; cylno < sblock.fs_ncg; cylno++) {
260 initcg(cylno, utime, 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) {
269 printf("%s", tmpbuf);
275 * Do all needed changes in the first cylinder group.
276 * allocate blocks in new location
278 updcsloc(utime, fsi, fso, Nflag);
281 * Now write the cylinder summary back to disk.
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);
288 DBG_PRINT0("fscs written\n");
292 struct csum *dbg_csp;
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,
305 #endif /* FS_DEBUG */
308 * Now write the new superblock back to disk.
310 sblock.fs_time = utime;
311 wtfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
312 DBG_PRINT0("sblock written\n");
314 "new initial sblock");
317 * Clean up the dynamic fields in our superblock copies.
322 sblock.fs_cgrotor = 0;
324 memset((void *)&sblock.fs_fsmnt, 0, sizeof(sblock.fs_fsmnt));
325 sblock.fs_flags &= FS_DOSOFTDEP;
329 * The following fields are currently distributed from the superblock
337 * fs_flags regarding SOFTPDATES
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.
349 * Write out the duplicate super blocks.
351 for (cylno = 0; cylno < sblock.fs_ncg; cylno++) {
352 wtfs(fsbtodb(&sblock, cgsblock(&sblock, cylno)),
353 (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
355 DBG_PRINT0("sblock copies written\n");
357 "new other sblocks");
363 /* ************************************************************ initcg ***** */
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.
371 initcg(int cylno, time_t utime, int fso, unsigned int Nflag)
376 ufs2_daddr_t i, cbase, dmax;
377 struct ufs1_dinode *dp1;
379 uint d, dupper, dlower;
381 if (iobuf == NULL && (iobuf = malloc(sblock.fs_bsize)) == NULL) {
382 errx(37, "panic: cannot allocate I/O buffer");
385 * Determine block bounds for cylinder group.
386 * Allow space for super block summary information in first
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);
398 memset(&acg, 0, sblock.fs_cgsize);
400 acg.cg_magic = CG_MAGIC;
402 acg.cg_niblk = sblock.fs_ipg;
403 acg.cg_initediblk = sblock.fs_ipg;
404 acg.cg_ndblk = dmax - cbase;
405 if (sblock.fs_contigsumsize > 0)
406 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
407 start = &acg.cg_space[0] - (u_char *)(&acg.cg_firstfield);
408 if (sblock.fs_magic == FS_UFS2_MAGIC) {
409 acg.cg_iusedoff = start;
411 acg.cg_old_ncyl = sblock.fs_old_cpg;
412 acg.cg_old_time = acg.cg_time;
414 acg.cg_old_niblk = acg.cg_niblk;
416 acg.cg_initediblk = 0;
417 acg.cg_old_btotoff = start;
418 acg.cg_old_boff = acg.cg_old_btotoff +
419 sblock.fs_old_cpg * sizeof(int32_t);
420 acg.cg_iusedoff = acg.cg_old_boff +
421 sblock.fs_old_cpg * sizeof(u_int16_t);
423 acg.cg_freeoff = acg.cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
424 acg.cg_nextfreeoff = acg.cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT);
425 if (sblock.fs_contigsumsize > 0) {
426 acg.cg_clustersumoff =
427 roundup(acg.cg_nextfreeoff, sizeof(u_int32_t));
428 acg.cg_clustersumoff -= sizeof(u_int32_t);
429 acg.cg_clusteroff = acg.cg_clustersumoff +
430 (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
431 acg.cg_nextfreeoff = acg.cg_clusteroff +
432 howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
434 if (acg.cg_nextfreeoff > (unsigned)sblock.fs_cgsize) {
436 * This should never happen as we would have had that panic
437 * already on file system creation
439 errx(37, "panic: cylinder group too big");
441 acg.cg_cs.cs_nifree += sblock.fs_ipg;
443 for (i = 0; i < ROOTINO; i++) {
444 setbit(cg_inosused(&acg), i);
445 acg.cg_cs.cs_nifree--;
448 * For the old file system, we have to initialize all the inodes.
450 if (sblock.fs_magic == FS_UFS1_MAGIC) {
451 bzero(iobuf, sblock.fs_bsize);
452 for (i = 0; i < sblock.fs_ipg / INOPF(&sblock);
453 i += sblock.fs_frag) {
454 dp1 = (struct ufs1_dinode *)iobuf;
456 for (j = 0; j < INOPB(&sblock); j++) {
457 dp1->di_gen = random();
461 wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno) + i),
462 sblock.fs_bsize, iobuf, fso, Nflag);
467 * In cylno 0, beginning space is reserved
468 * for boot and super blocks.
470 for (d = 0; d < dlower; d += sblock.fs_frag) {
471 blkno = d / sblock.fs_frag;
472 setblock(&sblock, cg_blksfree(&acg), blkno);
473 if (sblock.fs_contigsumsize > 0)
474 setbit(cg_clustersfree(&acg), blkno);
475 acg.cg_cs.cs_nbfree++;
477 sblock.fs_dsize += dlower;
479 sblock.fs_dsize += acg.cg_ndblk - dupper;
480 if ((i = dupper % sblock.fs_frag)) {
481 acg.cg_frsum[sblock.fs_frag - i]++;
482 for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
483 setbit(cg_blksfree(&acg), dupper);
484 acg.cg_cs.cs_nffree++;
487 for (d = dupper; d + sblock.fs_frag <= acg.cg_ndblk;
488 d += sblock.fs_frag) {
489 blkno = d / sblock.fs_frag;
490 setblock(&sblock, cg_blksfree(&acg), blkno);
491 if (sblock.fs_contigsumsize > 0)
492 setbit(cg_clustersfree(&acg), blkno);
493 acg.cg_cs.cs_nbfree++;
495 if (d < acg.cg_ndblk) {
496 acg.cg_frsum[acg.cg_ndblk - d]++;
497 for (; d < acg.cg_ndblk; d++) {
498 setbit(cg_blksfree(&acg), d);
499 acg.cg_cs.cs_nffree++;
502 if (sblock.fs_contigsumsize > 0) {
503 int32_t *sump = cg_clustersum(&acg);
504 u_char *mapp = cg_clustersfree(&acg);
509 for (i = 0; i < acg.cg_nclusterblks; i++) {
510 if ((map & bit) != 0)
513 if (run > sblock.fs_contigsumsize)
514 run = sblock.fs_contigsumsize;
518 if ((i & (CHAR_BIT - 1)) != CHAR_BIT - 1)
526 if (run > sblock.fs_contigsumsize)
527 run = sblock.fs_contigsumsize;
531 sblock.fs_cstotal.cs_ndir += acg.cg_cs.cs_ndir;
532 sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
533 sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
534 sblock.fs_cstotal.cs_nifree += acg.cg_cs.cs_nifree;
536 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
537 sblock.fs_bsize, (char *)&acg, fso, Nflag);
546 /* ******************************************************* frag_adjust ***** */
548 * Here we add or subtract (sign +1/-1) the available fragments in a given
549 * block to or from the fragment statistics. By subtracting before and adding
550 * after an operation on the free frag map we can easy update the fragment
551 * statistic, which seems to be otherwise a rather complex operation.
554 frag_adjust(ufs2_daddr_t frag, int sign)
556 DBG_FUNC("frag_adjust")
564 * Here frag only needs to point to any fragment in the block we want
567 for(f=rounddown(frag, sblock.fs_frag);
568 f<roundup(frag+1, sblock.fs_frag);
571 * Count contiguous free fragments.
573 if(isset(cg_blksfree(&acg), f)) {
576 if(fragsize && fragsize<sblock.fs_frag) {
578 * We found something in between.
580 acg.cg_frsum[fragsize]+=sign;
581 DBG_PRINT2("frag_adjust [%d]+=%d\n",
588 if(fragsize && fragsize<sblock.fs_frag) {
590 * We found something.
592 acg.cg_frsum[fragsize]+=sign;
593 DBG_PRINT2("frag_adjust [%d]+=%d\n",
597 DBG_PRINT2("frag_adjust [[%d]]+=%d\n",
605 /* ******************************************************* cond_bl_upd ***** */
607 * Here we conditionally update a pointer to a fragment. We check for all
608 * relocated blocks if any of its fragments is referenced by the current
609 * field, and update the pointer to the respective fragment in our new
610 * block. If we find a reference we write back the block immediately,
611 * as there is no easy way for our general block reading engine to figure
612 * out if a write back operation is needed.
615 cond_bl_upd(ufs2_daddr_t *block, struct gfs_bpp *field, int fsi, int fso,
618 DBG_FUNC("cond_bl_upd")
620 ufs2_daddr_t src, dst;
626 for (f = field; f->old != 0; f++) {
628 if (fragstoblks(&sblock, src) != f->old)
631 * The fragment is part of the block, so update.
633 dst = blkstofrags(&sblock, f->new);
634 fragnum = fragnum(&sblock, src);
635 *block = dst + fragnum;
637 DBG_PRINT3("scg (%jd->%jd)[%d] reference updated\n",
643 * Copy the block back immediately.
645 * XXX If src is is from an indirect block we have
646 * to implement copy on write here in case of
649 ibuf = malloc(sblock.fs_bsize);
651 errx(1, "malloc failed");
653 rdfs(fsbtodb(&sblock, src), (size_t)sblock.fs_bsize, ibuf, fsi);
654 wtfs(dst, (size_t)sblock.fs_bsize, ibuf, fso, Nflag);
657 * The same block can't be found again in this loop.
666 /* ************************************************************ updjcg ***** */
668 * Here we do all needed work for the former last cylinder group. It has to be
669 * changed in any case, even if the file system ended exactly on the end of
670 * this group, as there is some slightly inconsistent handling of the number
671 * of cylinders in the cylinder group. We start again by reading the cylinder
672 * group from disk. If the last block was not fully available, we first handle
673 * the missing fragments, then we handle all new full blocks in that file
674 * system and finally we handle the new last fragmented block in the file
675 * system. We again have to handle the fragment statistics rotational layout
676 * tables and cluster summary during all those operations.
679 updjcg(int cylno, time_t utime, int fsi, int fso, unsigned int Nflag)
682 ufs2_daddr_t cbase, dmax, dupper;
690 * Read the former last (joining) cylinder group from disk, and make
693 rdfs(fsbtodb(&osblock, cgtod(&osblock, cylno)),
694 (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
695 DBG_PRINT0("jcg read\n");
700 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
703 * If the cylinder group had already its new final size almost
704 * nothing is to be done ... except:
705 * For some reason the value of cg_ncyl in the last cylinder group has
706 * to be zero instead of fs_cpg. As this is now no longer the last
707 * cylinder group we have to change that value now to fs_cpg.
710 if(cgbase(&osblock, cylno+1) == osblock.fs_size) {
711 if (sblock.fs_magic == FS_UFS1_MAGIC)
712 acg.cg_old_ncyl=sblock.fs_old_cpg;
714 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
715 (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
716 DBG_PRINT0("jcg written\n");
726 * Set up some variables needed later.
728 cbase = cgbase(&sblock, cylno);
729 dmax = cbase + sblock.fs_fpg;
730 if (dmax > sblock.fs_size)
731 dmax = sblock.fs_size;
732 dupper = cgdmin(&sblock, cylno) - cbase;
733 if (cylno == 0) { /* XXX fscs may be relocated */
734 dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
738 * Set pointer to the cylinder summary for our cylinder group.
743 * Touch the cylinder group, update all fields in the cylinder group as
744 * needed, update the free space in the superblock.
747 if ((unsigned)cylno == sblock.fs_ncg - 1) {
749 * This is still the last cylinder group.
751 if (sblock.fs_magic == FS_UFS1_MAGIC)
753 sblock.fs_old_ncyl % sblock.fs_old_cpg;
755 acg.cg_old_ncyl = sblock.fs_old_cpg;
757 DBG_PRINT2("jcg dbg: %d %u",
761 if (sblock.fs_magic == FS_UFS1_MAGIC)
767 acg.cg_ndblk = dmax - cbase;
768 sblock.fs_dsize += acg.cg_ndblk-aocg.cg_ndblk;
769 if (sblock.fs_contigsumsize > 0) {
770 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
774 * Now we have to update the free fragment bitmap for our new free
775 * space. There again we have to handle the fragmentation and also
776 * the rotational layout tables and the cluster summary. This is
777 * also done per fragment for the first new block if the old file
778 * system end was not on a block boundary, per fragment for the new
779 * last block if the new file system end is not on a block boundary,
780 * and per block for all space in between.
782 * Handle the first new block here if it was partially available
785 if(osblock.fs_size % sblock.fs_frag) {
786 if(roundup(osblock.fs_size, sblock.fs_frag)<=sblock.fs_size) {
788 * The new space is enough to fill at least this
792 for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag)-1;
793 i>=osblock.fs_size-cbase;
795 setbit(cg_blksfree(&acg), i);
796 acg.cg_cs.cs_nffree++;
801 * Check if the fragment just created could join an
802 * already existing fragment at the former end of the
805 if(isblock(&sblock, cg_blksfree(&acg),
806 ((osblock.fs_size - cgbase(&sblock, cylno))/
809 * The block is now completely available.
811 DBG_PRINT0("block was\n");
812 acg.cg_frsum[osblock.fs_size%sblock.fs_frag]--;
813 acg.cg_cs.cs_nbfree++;
814 acg.cg_cs.cs_nffree-=sblock.fs_frag;
815 k=rounddown(osblock.fs_size-cbase,
817 updclst((osblock.fs_size-cbase)/sblock.fs_frag);
820 * Lets rejoin a possible partially growed
824 while(isset(cg_blksfree(&acg), i) &&
825 (i>=rounddown(osblock.fs_size-cbase,
837 * We only grow by some fragments within this last
840 for(i=sblock.fs_size-cbase-1;
841 i>=osblock.fs_size-cbase;
843 setbit(cg_blksfree(&acg), i);
844 acg.cg_cs.cs_nffree++;
848 * Lets rejoin a possible partially growed fragment.
851 while(isset(cg_blksfree(&acg), i) &&
852 (i>=rounddown(osblock.fs_size-cbase,
865 * Handle all new complete blocks here.
867 for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag);
868 i+sblock.fs_frag<=dmax-cbase; /* XXX <= or only < ? */
870 j = i / sblock.fs_frag;
871 setblock(&sblock, cg_blksfree(&acg), j);
873 acg.cg_cs.cs_nbfree++;
877 * Handle the last new block if there are stll some new fragments left.
878 * Here we don't have to bother about the cluster summary or the even
879 * the rotational layout table.
881 if (i < (dmax - cbase)) {
882 acg.cg_frsum[dmax - cbase - i]++;
883 for (; i < dmax - cbase; i++) {
884 setbit(cg_blksfree(&acg), i);
885 acg.cg_cs.cs_nffree++;
889 sblock.fs_cstotal.cs_nffree +=
890 (acg.cg_cs.cs_nffree - aocg.cg_cs.cs_nffree);
891 sblock.fs_cstotal.cs_nbfree +=
892 (acg.cg_cs.cs_nbfree - aocg.cg_cs.cs_nbfree);
894 * The following statistics are not changed here:
895 * sblock.fs_cstotal.cs_ndir
896 * sblock.fs_cstotal.cs_nifree
897 * As the statistics for this cylinder group are ready, copy it to
898 * the summary information array.
903 * Write the updated "joining" cylinder group back to disk.
905 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize,
906 (void *)&acg, fso, Nflag);
907 DBG_PRINT0("jcg written\n");
916 /* ********************************************************** updcsloc ***** */
918 * Here we update the location of the cylinder summary. We have two possible
919 * ways of growing the cylinder summary.
920 * (1) We can try to grow the summary in the current location, and relocate
921 * possibly used blocks within the current cylinder group.
922 * (2) Alternatively we can relocate the whole cylinder summary to the first
923 * new completely empty cylinder group. Once the cylinder summary is no
924 * longer in the beginning of the first cylinder group you should never
925 * use a version of fsck which is not aware of the possibility to have
926 * this structure in a non standard place.
927 * Option (1) is considered to be less intrusive to the structure of the file-
928 * system. So we try to stick to that whenever possible. If there is not enough
929 * space in the cylinder group containing the cylinder summary we have to use
930 * method (2). In case of active snapshots in the file system we probably can
931 * completely avoid implementing copy on write if we stick to method (2) only.
934 updcsloc(time_t utime, int fsi, int fso, unsigned int Nflag)
940 ufs2_daddr_t cbase, dupper, odupper, d, f, g;
950 if(howmany(sblock.fs_cssize, sblock.fs_fsize) ==
951 howmany(osblock.fs_cssize, osblock.fs_fsize)) {
953 * No new fragment needed.
958 ocscg=dtog(&osblock, osblock.fs_csaddr);
960 blocks = 1+howmany(sblock.fs_cssize, sblock.fs_bsize)-
961 howmany(osblock.fs_cssize, osblock.fs_bsize);
964 * Read original cylinder group from disk, and make a copy.
965 * XXX If Nflag is set in some very rare cases we now miss
966 * some changes done in updjcg by reading the unmodified
969 rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)),
970 (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
971 DBG_PRINT0("oscg read\n");
976 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
979 * Touch the cylinder group, set up local variables needed later
980 * and update the superblock.
985 * XXX In the case of having active snapshots we may need much more
986 * blocks for the copy on write. We need each block twice, and
987 * also up to 8*3 blocks for indirect blocks for all possible
990 if(/*((int)sblock.fs_time&0x3)>0||*/ cs->cs_nbfree < blocks) {
992 * There is not enough space in the old cylinder group to
993 * relocate all blocks as needed, so we relocate the whole
994 * cylinder group summary to a new group. We try to use the
995 * first complete new cylinder group just created. Within the
996 * cylinder group we align the area immediately after the
997 * cylinder group information location in order to be as
998 * close as possible to the original implementation of ffs.
1000 * First we have to make sure we'll find enough space in the
1001 * new cylinder group. If not, then we currently give up.
1002 * We start with freeing everything which was used by the
1003 * fragments of the old cylinder summary in the current group.
1004 * Now we write back the group meta data, read in the needed
1005 * meta data from the new cylinder group, and start allocating
1006 * within that group. Here we can assume, the group to be
1007 * completely empty. Which makes the handling of fragments and
1008 * clusters a lot easier.
1011 if(sblock.fs_ncg-osblock.fs_ncg < 2) {
1012 errx(2, "panic: not enough space");
1016 * Point "d" to the first fragment not used by the cylinder
1019 d=osblock.fs_csaddr+(osblock.fs_cssize/osblock.fs_fsize);
1022 * Set up last cluster size ("lcs") already here. Calculate
1023 * the size for the trailing cluster just behind where "d"
1026 if(sblock.fs_contigsumsize > 0) {
1027 for(block=howmany(d%sblock.fs_fpg, sblock.fs_frag),
1028 lcs=0; lcs<sblock.fs_contigsumsize;
1030 if(isclr(cg_clustersfree(&acg), block)){
1037 * Point "d" to the last frag used by the cylinder summary.
1041 DBG_PRINT1("d=%jd\n",
1043 if((d+1)%sblock.fs_frag) {
1045 * The end of the cylinder summary is not a complete
1049 frag_adjust(d%sblock.fs_fpg, -1);
1050 for(; (d+1)%sblock.fs_frag; d--) {
1051 DBG_PRINT1("d=%jd\n",
1053 setbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1054 acg.cg_cs.cs_nffree++;
1055 sblock.fs_cstotal.cs_nffree++;
1058 * Point "d" to the last fragment of the last
1059 * (incomplete) block of the cylinder summary.
1062 frag_adjust(d%sblock.fs_fpg, 1);
1064 if(isblock(&sblock, cg_blksfree(&acg),
1065 (d%sblock.fs_fpg)/sblock.fs_frag)) {
1066 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1067 acg.cg_cs.cs_nffree-=sblock.fs_frag;
1068 acg.cg_cs.cs_nbfree++;
1069 sblock.fs_cstotal.cs_nffree-=sblock.fs_frag;
1070 sblock.fs_cstotal.cs_nbfree++;
1071 if(sblock.fs_contigsumsize > 0) {
1072 setbit(cg_clustersfree(&acg),
1073 (d%sblock.fs_fpg)/sblock.fs_frag);
1074 if(lcs < sblock.fs_contigsumsize) {
1080 cg_clustersum(&acg)[lcs]++;
1085 * Point "d" to the first fragment of the block before
1086 * the last incomplete block.
1091 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1092 for(d=rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr;
1093 d-=sblock.fs_frag) {
1095 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1096 setblock(&sblock, cg_blksfree(&acg),
1097 (d%sblock.fs_fpg)/sblock.fs_frag);
1098 acg.cg_cs.cs_nbfree++;
1099 sblock.fs_cstotal.cs_nbfree++;
1100 if(sblock.fs_contigsumsize > 0) {
1101 setbit(cg_clustersfree(&acg),
1102 (d%sblock.fs_fpg)/sblock.fs_frag);
1104 * The last cluster size is already set up.
1106 if(lcs < sblock.fs_contigsumsize) {
1108 cg_clustersum(&acg)[lcs]--;
1111 cg_clustersum(&acg)[lcs]++;
1118 * Now write the former cylinder group containing the cylinder
1119 * summary back to disk.
1121 wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)),
1122 (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1123 DBG_PRINT0("oscg written\n");
1124 DBG_DUMP_CG(&sblock,
1129 * Find the beginning of the new cylinder group containing the
1132 sblock.fs_csaddr=cgdmin(&sblock, osblock.fs_ncg);
1133 ncscg=dtog(&sblock, sblock.fs_csaddr);
1138 * If Nflag is specified, we would now read random data instead
1139 * of an empty cg structure from disk. So we can't simulate that
1143 DBG_PRINT0("nscg update skipped\n");
1149 * Read the future cylinder group containing the cylinder
1150 * summary from disk, and make a copy.
1152 rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1153 (size_t)sblock.fs_cgsize, (void *)&aocg, fsi);
1154 DBG_PRINT0("nscg read\n");
1155 DBG_DUMP_CG(&sblock,
1159 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
1162 * Allocate all complete blocks used by the new cylinder
1165 for(d=sblock.fs_csaddr; d+sblock.fs_frag <=
1166 sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize);
1167 d+=sblock.fs_frag) {
1168 clrblock(&sblock, cg_blksfree(&acg),
1169 (d%sblock.fs_fpg)/sblock.fs_frag);
1170 acg.cg_cs.cs_nbfree--;
1171 sblock.fs_cstotal.cs_nbfree--;
1172 if(sblock.fs_contigsumsize > 0) {
1173 clrbit(cg_clustersfree(&acg),
1174 (d%sblock.fs_fpg)/sblock.fs_frag);
1179 * Allocate all fragments used by the cylinder summary in the
1182 if(d<sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize)) {
1183 for(; d-sblock.fs_csaddr<
1184 sblock.fs_cssize/sblock.fs_fsize;
1186 clrbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1187 acg.cg_cs.cs_nffree--;
1188 sblock.fs_cstotal.cs_nffree--;
1190 acg.cg_cs.cs_nbfree--;
1191 acg.cg_cs.cs_nffree+=sblock.fs_frag;
1192 sblock.fs_cstotal.cs_nbfree--;
1193 sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1194 if(sblock.fs_contigsumsize > 0) {
1195 clrbit(cg_clustersfree(&acg),
1196 (d%sblock.fs_fpg)/sblock.fs_frag);
1199 frag_adjust(d%sblock.fs_fpg, +1);
1202 * XXX Handle the cluster statistics here in the case this
1203 * cylinder group is now almost full, and the remaining
1204 * space is less then the maximum cluster size. This is
1205 * probably not needed, as you would hardly find a file
1206 * system which has only MAXCSBUFS+FS_MAXCONTIG of free
1207 * space right behind the cylinder group information in
1208 * any new cylinder group.
1212 * Update our statistics in the cylinder summary.
1217 * Write the new cylinder group containing the cylinder summary
1220 wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1221 (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1222 DBG_PRINT0("nscg written\n");
1223 DBG_DUMP_CG(&sblock,
1231 * We have got enough of space in the current cylinder group, so we
1232 * can relocate just a few blocks, and let the summary information
1233 * grow in place where it is right now.
1237 cbase = cgbase(&osblock, ocscg); /* old and new are equal */
1238 dupper = sblock.fs_csaddr - cbase +
1239 howmany(sblock.fs_cssize, sblock.fs_fsize);
1240 odupper = osblock.fs_csaddr - cbase +
1241 howmany(osblock.fs_cssize, osblock.fs_fsize);
1243 sblock.fs_dsize -= dupper-odupper;
1246 * Allocate the space for the array of blocks to be relocated.
1248 bp=(struct gfs_bpp *)malloc(((dupper-odupper)/sblock.fs_frag+2)*
1249 sizeof(struct gfs_bpp));
1251 errx(1, "malloc failed");
1253 memset((char *)bp, 0, ((dupper-odupper)/sblock.fs_frag+2)*
1254 sizeof(struct gfs_bpp));
1257 * Lock all new frags needed for the cylinder group summary. This is
1258 * done per fragment in the first and last block of the new required
1259 * area, and per block for all other blocks.
1261 * Handle the first new block here (but only if some fragments where
1262 * already used for the cylinder summary).
1265 frag_adjust(odupper, -1);
1266 for(d=odupper; ((d<dupper)&&(d%sblock.fs_frag)); d++) {
1267 DBG_PRINT1("scg first frag check loop d=%jd\n",
1269 if(isclr(cg_blksfree(&acg), d)) {
1271 bp[ind].old=d/sblock.fs_frag;
1272 bp[ind].flags|=GFS_FL_FIRST;
1273 if(roundup(d, sblock.fs_frag) >= dupper) {
1274 bp[ind].flags|=GFS_FL_LAST;
1279 clrbit(cg_blksfree(&acg), d);
1280 acg.cg_cs.cs_nffree--;
1281 sblock.fs_cstotal.cs_nffree--;
1284 * No cluster handling is needed here, as there was at least
1285 * one fragment in use by the cylinder summary in the old
1287 * No block-free counter handling here as this block was not
1291 frag_adjust(odupper, 1);
1294 * Handle all needed complete blocks here.
1296 for(; d+sblock.fs_frag<=dupper; d+=sblock.fs_frag) {
1297 DBG_PRINT1("scg block check loop d=%jd\n",
1299 if(!isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1300 for(f=d; f<d+sblock.fs_frag; f++) {
1301 if(isset(cg_blksfree(&aocg), f)) {
1302 acg.cg_cs.cs_nffree--;
1303 sblock.fs_cstotal.cs_nffree--;
1306 clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1307 bp[ind].old=d/sblock.fs_frag;
1310 clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1311 acg.cg_cs.cs_nbfree--;
1312 sblock.fs_cstotal.cs_nbfree--;
1313 if(sblock.fs_contigsumsize > 0) {
1314 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1315 for(lcs=0, l=(d/sblock.fs_frag)+1;
1316 lcs<sblock.fs_contigsumsize;
1318 if(isclr(cg_clustersfree(&acg),l)){
1322 if(lcs < sblock.fs_contigsumsize) {
1323 cg_clustersum(&acg)[lcs+1]--;
1325 cg_clustersum(&acg)[lcs]++;
1331 * No fragment counter handling is needed here, as this finally
1332 * doesn't change after the relocation.
1337 * Handle all fragments needed in the last new affected block.
1340 frag_adjust(dupper-1, -1);
1342 if(isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1343 acg.cg_cs.cs_nbfree--;
1344 sblock.fs_cstotal.cs_nbfree--;
1345 acg.cg_cs.cs_nffree+=sblock.fs_frag;
1346 sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1347 if(sblock.fs_contigsumsize > 0) {
1348 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1349 for(lcs=0, l=(d/sblock.fs_frag)+1;
1350 lcs<sblock.fs_contigsumsize;
1352 if(isclr(cg_clustersfree(&acg),l)){
1356 if(lcs < sblock.fs_contigsumsize) {
1357 cg_clustersum(&acg)[lcs+1]--;
1359 cg_clustersum(&acg)[lcs]++;
1365 for(; d<dupper; d++) {
1366 DBG_PRINT1("scg second frag check loop d=%jd\n",
1368 if(isclr(cg_blksfree(&acg), d)) {
1369 bp[ind].old=d/sblock.fs_frag;
1370 bp[ind].flags|=GFS_FL_LAST;
1372 clrbit(cg_blksfree(&acg), d);
1373 acg.cg_cs.cs_nffree--;
1374 sblock.fs_cstotal.cs_nffree--;
1377 if(bp[ind].flags & GFS_FL_LAST) { /* we have to advance here */
1380 frag_adjust(dupper-1, 1);
1384 * If we found a block to relocate just do so.
1387 for(i=0; i<ind; i++) {
1388 if(!bp[i].old) { /* no more blocks listed */
1390 * XXX A relative blocknumber should not be
1391 * zero, which is not explicitly
1392 * guaranteed by our code.
1397 * Allocate a complete block in the same (current)
1400 bp[i].new=alloc()/sblock.fs_frag;
1403 * There is no frag_adjust() needed for the new block
1404 * as it will have no fragments yet :-).
1406 for(f=bp[i].old*sblock.fs_frag,
1407 g=bp[i].new*sblock.fs_frag;
1408 f<(bp[i].old+1)*sblock.fs_frag;
1410 if(isset(cg_blksfree(&aocg), f)) {
1411 setbit(cg_blksfree(&acg), g);
1412 acg.cg_cs.cs_nffree++;
1413 sblock.fs_cstotal.cs_nffree++;
1418 * Special handling is required if this was the first
1419 * block. We have to consider the fragments which were
1420 * used by the cylinder summary in the original block
1421 * which re to be free in the copy of our block. We
1422 * have to be careful if this first block happens to
1423 * be also the last block to be relocated.
1425 if(bp[i].flags & GFS_FL_FIRST) {
1426 for(f=bp[i].old*sblock.fs_frag,
1427 g=bp[i].new*sblock.fs_frag;
1430 setbit(cg_blksfree(&acg), g);
1431 acg.cg_cs.cs_nffree++;
1432 sblock.fs_cstotal.cs_nffree++;
1434 if(!(bp[i].flags & GFS_FL_LAST)) {
1435 frag_adjust(bp[i].new*sblock.fs_frag,1);
1440 * Special handling is required if this is the last
1441 * block to be relocated.
1443 if(bp[i].flags & GFS_FL_LAST) {
1444 frag_adjust(bp[i].new*sblock.fs_frag, 1);
1445 frag_adjust(bp[i].old*sblock.fs_frag, -1);
1447 f<roundup(dupper, sblock.fs_frag);
1449 if(isclr(cg_blksfree(&acg), f)) {
1450 setbit(cg_blksfree(&acg), f);
1451 acg.cg_cs.cs_nffree++;
1452 sblock.fs_cstotal.cs_nffree++;
1455 frag_adjust(bp[i].old*sblock.fs_frag, 1);
1459 * !!! Attach the cylindergroup offset here.
1461 bp[i].old+=cbase/sblock.fs_frag;
1462 bp[i].new+=cbase/sblock.fs_frag;
1465 * Copy the content of the block.
1468 * XXX Here we will have to implement a copy on write
1469 * in the case we have any active snapshots.
1471 rdfs(fsbtodb(&sblock, bp[i].old*sblock.fs_frag),
1472 (size_t)sblock.fs_bsize, (void *)&ablk, fsi);
1473 wtfs(fsbtodb(&sblock, bp[i].new*sblock.fs_frag),
1474 (size_t)sblock.fs_bsize, (void *)&ablk, fso, Nflag);
1475 DBG_DUMP_HEX(&sblock,
1476 "copied full block",
1477 (unsigned char *)&ablk);
1479 DBG_PRINT2("scg (%jd->%jd) block relocated\n",
1480 (intmax_t)bp[i].old,
1481 (intmax_t)bp[i].new);
1485 * Now we have to update all references to any fragment which
1486 * belongs to any block relocated. We iterate now over all
1487 * cylinder groups, within those over all non zero length
1490 for(cylno=0; cylno<osblock.fs_ncg; cylno++) {
1491 DBG_PRINT1("scg doing cg (%d)\n",
1493 for(inc=osblock.fs_ipg-1 ; inc>0 ; inc--) {
1494 updrefs(cylno, (ino_t)inc, bp, fsi, fso, Nflag);
1499 * All inodes are checked, now make sure the number of
1500 * references found make sense.
1502 for(i=0; i<ind; i++) {
1503 if(!bp[i].found || (bp[i].found>sblock.fs_frag)) {
1504 warnx("error: %jd refs found for block %jd.",
1505 (intmax_t)bp[i].found, (intmax_t)bp[i].old);
1511 * The following statistics are not changed here:
1512 * sblock.fs_cstotal.cs_ndir
1513 * sblock.fs_cstotal.cs_nifree
1514 * The following statistics were already updated on the fly:
1515 * sblock.fs_cstotal.cs_nffree
1516 * sblock.fs_cstotal.cs_nbfree
1517 * As the statistics for this cylinder group are ready, copy it to
1518 * the summary information array.
1524 * Write summary cylinder group back to disk.
1526 wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)), (size_t)sblock.fs_cgsize,
1527 (void *)&acg, fso, Nflag);
1528 DBG_PRINT0("scg written\n");
1529 DBG_DUMP_CG(&sblock,
1537 /* ************************************************************** rdfs ***** */
1539 * Here we read some block(s) from disk.
1542 rdfs(ufs2_daddr_t bno, size_t size, void *bf, int fsi)
1550 err(32, "rdfs: attempting to read negative block number");
1552 if (lseek(fsi, (off_t)bno * DEV_BSIZE, 0) < 0) {
1553 err(33, "rdfs: seek error: %jd", (intmax_t)bno);
1555 n = read(fsi, bf, size);
1556 if (n != (ssize_t)size) {
1557 err(34, "rdfs: read error: %jd", (intmax_t)bno);
1564 /* ************************************************************** wtfs ***** */
1566 * Here we write some block(s) to disk.
1569 wtfs(ufs2_daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag)
1580 if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0) {
1581 err(35, "wtfs: seek error: %ld", (long)bno);
1583 n = write(fso, bf, size);
1584 if (n != (ssize_t)size) {
1585 err(36, "wtfs: write error: %ld", (long)bno);
1592 /* ************************************************************* alloc ***** */
1594 * Here we allocate a free block in the current cylinder group. It is assumed,
1595 * that acg contains the current cylinder group. As we may take a block from
1596 * somewhere in the file system we have to handle cluster summary here.
1602 ufs2_daddr_t d, blkno;
1606 int dlower, dupper, dmax;
1610 if (acg.cg_magic != CG_MAGIC) {
1611 warnx("acg: bad magic number");
1615 if (acg.cg_cs.cs_nbfree == 0) {
1616 warnx("error: cylinder group ran out of space");
1621 * We start seeking for free blocks only from the space available after
1622 * the end of the new grown cylinder summary. Otherwise we allocate a
1623 * block here which we have to relocate a couple of seconds later again
1624 * again, and we are not prepared to to this anyway.
1627 dlower=cgsblock(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1628 dupper=cgdmin(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1629 dmax=cgbase(&sblock, acg.cg_cgx)+sblock.fs_fpg;
1630 if (dmax > sblock.fs_size) {
1631 dmax = sblock.fs_size;
1633 dmax-=cgbase(&sblock, acg.cg_cgx); /* retransform into cg */
1634 csmin=sblock.fs_csaddr-cgbase(&sblock, acg.cg_cgx);
1635 csmax=csmin+howmany(sblock.fs_cssize, sblock.fs_fsize);
1636 DBG_PRINT3("seek range: dl=%d, du=%d, dm=%d\n",
1640 DBG_PRINT2("range cont: csmin=%d, csmax=%d\n",
1644 for(d=0; (d<dlower && blkno==-1); d+=sblock.fs_frag) {
1645 if(d>=csmin && d<=csmax) {
1648 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1650 blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1654 for(d=dupper; (d<dmax && blkno==-1); d+=sblock.fs_frag) {
1655 if(d>=csmin && d<=csmax) {
1658 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1660 blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1665 warnx("internal error: couldn't find promised block in cg");
1671 * This is needed if the block was found already in the first loop.
1673 d=blkstofrags(&sblock, blkno);
1675 clrblock(&sblock, cg_blksfree(&acg), blkno);
1676 if (sblock.fs_contigsumsize > 0) {
1678 * Handle the cluster allocation bitmap.
1680 clrbit(cg_clustersfree(&acg), blkno);
1682 * We possibly have split a cluster here, so we have to do
1683 * recalculate the sizes of the remaining cluster halves now,
1684 * and use them for updating the cluster summary information.
1686 * Lets start with the blocks before our allocated block ...
1688 for(lcs1=0, l=blkno-1; lcs1<sblock.fs_contigsumsize;
1690 if(isclr(cg_clustersfree(&acg),l)){
1695 * ... and continue with the blocks right after our allocated
1698 for(lcs2=0, l=blkno+1; lcs2<sblock.fs_contigsumsize;
1700 if(isclr(cg_clustersfree(&acg),l)){
1706 * Now update all counters.
1708 cg_clustersum(&acg)[MIN(lcs1+lcs2+1,sblock.fs_contigsumsize)]--;
1710 cg_clustersum(&acg)[lcs1]++;
1713 cg_clustersum(&acg)[lcs2]++;
1717 * Update all statistics based on blocks.
1719 acg.cg_cs.cs_nbfree--;
1720 sblock.fs_cstotal.cs_nbfree--;
1726 /* *********************************************************** isblock ***** */
1728 * Here we check if all frags of a block are free. For more details again
1729 * please see the source of newfs(8), as this function is taken over almost
1733 isblock(struct fs *fs, unsigned char *cp, int h)
1740 switch (fs->fs_frag) {
1743 return (cp[h] == 0xff);
1745 mask = 0x0f << ((h & 0x1) << 2);
1747 return ((cp[h >> 1] & mask) == mask);
1749 mask = 0x03 << ((h & 0x3) << 1);
1751 return ((cp[h >> 2] & mask) == mask);
1753 mask = 0x01 << (h & 0x7);
1755 return ((cp[h >> 3] & mask) == mask);
1757 fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
1763 /* ********************************************************** clrblock ***** */
1765 * Here we allocate a complete block in the block map. For more details again
1766 * please see the source of newfs(8), as this function is taken over almost
1770 clrblock(struct fs *fs, unsigned char *cp, int h)
1772 DBG_FUNC("clrblock")
1776 switch ((fs)->fs_frag) {
1781 cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1784 cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1787 cp[h >> 3] &= ~(0x01 << (h & 0x7));
1790 warnx("clrblock bad fs_frag %d", fs->fs_frag);
1798 /* ********************************************************** setblock ***** */
1800 * Here we free a complete block in the free block map. For more details again
1801 * please see the source of newfs(8), as this function is taken over almost
1805 setblock(struct fs *fs, unsigned char *cp, int h)
1807 DBG_FUNC("setblock")
1811 switch (fs->fs_frag) {
1816 cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1819 cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1822 cp[h >> 3] |= (0x01 << (h & 0x7));
1825 warnx("setblock bad fs_frag %d", fs->fs_frag);
1833 /* ************************************************************ ginode ***** */
1835 * This function provides access to an individual inode. We find out in which
1836 * block the requested inode is located, read it from disk if needed, and
1837 * return the pointer into that block. We maintain a cache of one block to
1838 * not read the same block again and again if we iterate linearly over all
1841 static union dinode *
1842 ginode(ino_t inumber, int fsi, int cg)
1845 static ino_t startinum = 0; /* first inode in cached block */
1850 * The inumber passed in is relative to the cg, so use it here to see
1851 * if the inode has been allocated yet.
1853 if (isclr(cg_inosused(&aocg), inumber)) {
1858 * Now make the inumber relative to the entire inode space so it can
1859 * be sanity checked.
1861 inumber += (cg * sblock.fs_ipg);
1862 if (inumber < ROOTINO) {
1866 if (inumber > maxino)
1867 errx(8, "bad inode number %d to ginode", inumber);
1868 if (startinum == 0 ||
1869 inumber < startinum || inumber >= startinum + INOPB(&sblock)) {
1870 inoblk = fsbtodb(&sblock, ino_to_fsba(&sblock, inumber));
1871 rdfs(inoblk, (size_t)sblock.fs_bsize, inobuf, fsi);
1872 startinum = (inumber / INOPB(&sblock)) * INOPB(&sblock);
1875 if (sblock.fs_magic == FS_UFS1_MAGIC)
1876 return (union dinode *)((uintptr_t)inobuf +
1877 (inumber % INOPB(&sblock)) * sizeof(struct ufs1_dinode));
1878 return (union dinode *)((uintptr_t)inobuf +
1879 (inumber % INOPB(&sblock)) * sizeof(struct ufs2_dinode));
1882 /* ****************************************************** charsperline ***** */
1884 * Figure out how many lines our current terminal has. For more details again
1885 * please see the source of newfs(8), as this function is taken over almost
1891 DBG_FUNC("charsperline")
1899 if (ioctl(0, TIOCGWINSZ, &ws) != -1) {
1900 columns = ws.ws_col;
1902 if (columns == 0 && (cp = getenv("COLUMNS"))) {
1906 columns = 80; /* last resort */
1913 /* ****************************************************** get_dev_size ***** */
1915 * Get the size of the partition if we can't figure it out from the disklabel,
1916 * e.g. from vinum volumes.
1919 get_dev_size(int fd, int *size)
1924 if (ioctl(fd, DIOCGSECTORSIZE, §orsize) == -1)
1925 err(1,"DIOCGSECTORSIZE");
1926 if (ioctl(fd, DIOCGMEDIASIZE, &mediasize) == -1)
1927 err(1,"DIOCGMEDIASIZE");
1929 if (sectorsize <= 0)
1930 errx(1, "bogus sectorsize: %d", sectorsize);
1932 *size = mediasize / sectorsize;
1935 /* ************************************************************** main ***** */
1937 * growfs(8) is a utility which allows to increase the size of an existing
1938 * ufs file system. Currently this can only be done on unmounted file system.
1939 * It recognizes some command line options to specify the new desired size,
1940 * and it does some basic checkings. The old file system size is determined
1941 * and after some more checks like we can really access the new last block
1942 * on the disk etc. we calculate the new parameters for the superblock. After
1943 * having done this we just call growfs() which will do the work. Before
1944 * we finish the only thing left is to update the disklabel.
1945 * We still have to provide support for snapshots. Therefore we first have to
1946 * understand what data structures are always replicated in the snapshot on
1947 * creation, for all other blocks we touch during our procedure, we have to
1948 * keep the old blocks unchanged somewhere available for the snapshots. If we
1949 * are lucky, then we only have to handle our blocks to be relocated in that
1951 * Also we have to consider in what order we actually update the critical
1952 * data structures of the file system to make sure, that in case of a disaster
1953 * fsck(8) is still able to restore any lost data.
1954 * The foreseen last step then will be to provide for growing even mounted
1955 * file systems. There we have to extend the mount() system call to provide
1956 * userland access to the file system locking facility.
1959 main(int argc, char **argv)
1962 char *device, *special, *cp;
1964 unsigned int size=0;
1966 unsigned int Nflag=0;
1969 struct disklabel *lp;
1970 struct partition *pp;
1976 #endif /* FSMAXSNAP */
1980 while((ch=getopt(argc, argv, "Ns:vy")) != -1) {
1986 size=(size_t)atol(optarg);
1991 case 'v': /* for compatibility to newfs */
2011 * Now try to guess the (raw)device name.
2013 if (0 == strrchr(device, '/')) {
2015 * No path prefix was given, so try in that order:
2021 * FreeBSD now doesn't distinguish between raw and block
2022 * devices any longer, but it should still work this way.
2024 len=strlen(device)+strlen(_PATH_DEV)+2+strlen("vinum/");
2025 special=(char *)malloc(len);
2026 if(special == NULL) {
2027 errx(1, "malloc failed");
2029 snprintf(special, len, "%sr%s", _PATH_DEV, device);
2030 if (stat(special, &st) == -1) {
2031 snprintf(special, len, "%s%s", _PATH_DEV, device);
2032 if (stat(special, &st) == -1) {
2033 snprintf(special, len, "%svinum/r%s",
2035 if (stat(special, &st) == -1) {
2036 /* For now this is the 'last resort' */
2037 snprintf(special, len, "%svinum/%s",
2046 * Try to access our devices for writing ...
2051 fso = open(device, O_WRONLY);
2053 err(1, "%s", device);
2060 fsi = open(device, O_RDONLY);
2062 err(1, "%s", device);
2066 * Try to read a label and guess the slice if not specified. This
2067 * code should guess the right thing and avoid to bother the user
2068 * with the task of specifying the option -v on vinum volumes.
2070 cp=device+strlen(device)-1;
2071 lp = get_disklabel(fsi);
2075 pp = &lp->d_partitions[2];
2076 } else if (*cp>='a' && *cp<='h') {
2077 pp = &lp->d_partitions[*cp - 'a'];
2079 errx(1, "unknown device");
2081 p_size = pp->p_size;
2083 get_dev_size(fsi, &p_size);
2087 * Check if that partition is suitable for growing a file system.
2090 errx(1, "partition is unavailable");
2094 * Read the current superblock, and take a backup.
2096 for (i = 0; sblock_try[i] != -1; i++) {
2097 sblockloc = sblock_try[i] / DEV_BSIZE;
2098 rdfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&(osblock), fsi);
2099 if ((osblock.fs_magic == FS_UFS1_MAGIC ||
2100 (osblock.fs_magic == FS_UFS2_MAGIC &&
2101 osblock.fs_sblockloc == sblock_try[i])) &&
2102 osblock.fs_bsize <= MAXBSIZE &&
2103 osblock.fs_bsize >= (int32_t) sizeof(struct fs))
2106 if (sblock_try[i] == -1) {
2107 errx(1, "superblock not recognized");
2109 memcpy((void *)&fsun1, (void *)&fsun2, sizeof(fsun2));
2110 maxino = sblock.fs_ncg * sblock.fs_ipg;
2112 DBG_OPEN("/tmp/growfs.debug"); /* already here we need a superblock */
2113 DBG_DUMP_FS(&sblock,
2117 * Determine size to grow to. Default to the full size specified in
2120 sblock.fs_size = dbtofsb(&osblock, p_size);
2123 errx(1, "there is not enough space (%d < %d)",
2126 sblock.fs_size = dbtofsb(&osblock, size);
2130 * Are we really growing ?
2132 if(osblock.fs_size >= sblock.fs_size) {
2133 errx(1, "we are not growing (%jd->%jd)",
2134 (intmax_t)osblock.fs_size, (intmax_t)sblock.fs_size);
2140 * Check if we find an active snapshot.
2142 if(ExpertFlag == 0) {
2143 for(j=0; j<FSMAXSNAP; j++) {
2144 if(sblock.fs_snapinum[j]) {
2145 errx(1, "active snapshot found in file system\n"
2146 " please remove all snapshots before "
2149 if(!sblock.fs_snapinum[j]) { /* list is dense */
2156 if (ExpertFlag == 0 && Nflag == 0) {
2157 printf("We strongly recommend you to make a backup "
2158 "before growing the Filesystem\n\n"
2159 " Did you backup your data (Yes/No) ? ");
2160 fgets(reply, (int)sizeof(reply), stdin);
2161 if (strcmp(reply, "Yes\n")){
2162 printf("\n Nothing done \n");
2167 printf("new file systemsize is: %jd frags\n", (intmax_t)sblock.fs_size);
2170 * Try to access our new last block in the file system. Even if we
2171 * later on realize we have to abort our operation, on that block
2172 * there should be no data, so we can't destroy something yet.
2174 wtfs((ufs2_daddr_t)p_size-1, (size_t)DEV_BSIZE, (void *)&sblock,
2178 * Now calculate new superblock values and check for reasonable
2179 * bound for new file system size:
2180 * fs_size: is derived from label or user input
2181 * fs_dsize: should get updated in the routines creating or
2182 * updating the cylinder groups on the fly
2183 * fs_cstotal: should get updated in the routines creating or
2184 * updating the cylinder groups
2188 * Update the number of cylinders and cylinder groups in the file system.
2190 if (sblock.fs_magic == FS_UFS1_MAGIC) {
2191 sblock.fs_old_ncyl =
2192 sblock.fs_size * sblock.fs_old_nspf / sblock.fs_old_spc;
2193 if (sblock.fs_size * sblock.fs_old_nspf >
2194 sblock.fs_old_ncyl * sblock.fs_old_spc)
2195 sblock.fs_old_ncyl++;
2197 sblock.fs_ncg = howmany(sblock.fs_size, sblock.fs_fpg);
2198 maxino = sblock.fs_ncg * sblock.fs_ipg;
2200 if (sblock.fs_size % sblock.fs_fpg != 0 &&
2201 sblock.fs_size % sblock.fs_fpg < cgdmin(&sblock, sblock.fs_ncg)) {
2203 * The space in the new last cylinder group is too small,
2207 if (sblock.fs_magic == FS_UFS1_MAGIC)
2208 sblock.fs_old_ncyl = sblock.fs_ncg * sblock.fs_old_cpg;
2209 printf("Warning: %jd sector(s) cannot be allocated.\n",
2210 (intmax_t)fsbtodb(&sblock, sblock.fs_size % sblock.fs_fpg));
2211 sblock.fs_size = sblock.fs_ncg * sblock.fs_fpg;
2215 * Update the space for the cylinder group summary information in the
2216 * respective cylinder group data area.
2219 fragroundup(&sblock, sblock.fs_ncg * sizeof(struct csum));
2221 if(osblock.fs_size >= sblock.fs_size) {
2222 errx(1, "not enough new space");
2225 DBG_PRINT0("sblock calculated\n");
2228 * Ok, everything prepared, so now let's do the tricks.
2230 growfs(fsi, fso, Nflag);
2233 * Update the disk label.
2236 pp->p_fsize = sblock.fs_fsize;
2237 pp->p_frag = sblock.fs_frag;
2238 pp->p_cpg = sblock.fs_fpg;
2240 return_disklabel(fso, lp, Nflag);
2241 DBG_PRINT0("label rewritten\n");
2245 if(fso>-1) close(fso);
2253 /* ************************************************** return_disklabel ***** */
2255 * Write the updated disklabel back to disk.
2258 return_disklabel(int fd, struct disklabel *lp, unsigned int Nflag)
2260 DBG_FUNC("return_disklabel")
2276 * recalculate checksum
2278 while(ptr < (u_short *)&lp->d_partitions[lp->d_npartitions]) {
2283 if (ioctl(fd, DIOCWDINFO, (char *)lp) < 0) {
2284 errx(1, "DIOCWDINFO failed");
2293 /* ***************************************************** get_disklabel ***** */
2295 * Read the disklabel from disk.
2297 static struct disklabel *
2298 get_disklabel(int fd)
2300 DBG_FUNC("get_disklabel")
2301 static struct disklabel *lab;
2305 lab=(struct disklabel *)malloc(sizeof(struct disklabel));
2307 errx(1, "malloc failed");
2309 if (!ioctl(fd, DIOCGDINFO, (char *)lab))
2319 /* ************************************************************* usage ***** */
2321 * Dump a line of usage.
2330 fprintf(stderr, "usage: growfs [-Ny] [-s size] special\n");
2336 /* *********************************************************** updclst ***** */
2338 * This updates most parameters and the bitmap related to cluster. We have to
2339 * assume that sblock, osblock, acg are set up.
2349 if(sblock.fs_contigsumsize < 1) { /* no clustering */
2353 * update cluster allocation map
2355 setbit(cg_clustersfree(&acg), block);
2358 * update cluster summary table
2362 * calculate size for the trailing cluster
2364 for(block--; lcs<sblock.fs_contigsumsize; block--, lcs++ ) {
2365 if(isclr(cg_clustersfree(&acg), block)){
2370 if(lcs < sblock.fs_contigsumsize) {
2372 cg_clustersum(&acg)[lcs]--;
2375 cg_clustersum(&acg)[lcs]++;
2382 /* *********************************************************** updrefs ***** */
2384 * This updates all references to relocated blocks for the given inode. The
2385 * inode is given as number within the cylinder group, and the number of the
2389 updrefs(int cg, ino_t in, struct gfs_bpp *bp, int fsi, int fso, unsigned int
2393 ufs_lbn_t len, lbn, numblks;
2394 ufs2_daddr_t iptr, blksperindir;
2396 int i, mode, inodeupdated;
2400 ino = ginode(in, fsi, cg);
2405 mode = DIP(ino, di_mode) & IFMT;
2406 if (mode != IFDIR && mode != IFREG && mode != IFLNK) {
2408 return; /* only check DIR, FILE, LINK */
2410 if (mode == IFLNK &&
2411 DIP(ino, di_size) < (u_int64_t) sblock.fs_maxsymlinklen) {
2413 return; /* skip short symlinks */
2415 numblks = howmany(DIP(ino, di_size), sblock.fs_bsize);
2418 return; /* skip empty file */
2420 if (DIP(ino, di_blocks) == 0) {
2422 return; /* skip empty swiss cheesy file or old fastlink */
2424 DBG_PRINT2("scg checking inode (%d in %d)\n",
2429 * Check all the blocks.
2432 len = numblks < NDADDR ? numblks : NDADDR;
2433 for (i = 0; i < len; i++) {
2434 iptr = DIP(ino, di_db[i]);
2437 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2438 DIP_SET(ino, di_db[i], iptr);
2442 DBG_PRINT0("~~scg direct blocks checked\n");
2445 len = numblks - NDADDR;
2447 for (i = 0; len > 0 && i < NIADDR; i++) {
2448 iptr = DIP(ino, di_ib[i]);
2451 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2452 DIP_SET(ino, di_ib[i], iptr);
2455 indirchk(blksperindir, lbn, iptr, numblks, bp, fsi, fso, Nflag);
2456 blksperindir *= NINDIR(&sblock);
2457 lbn += blksperindir;
2458 len -= blksperindir;
2459 DBG_PRINT1("scg indirect_%d blocks checked\n", i + 1);
2462 wtfs(inoblk, sblock.fs_bsize, inobuf, fso, Nflag);
2469 * Recursively check all the indirect blocks.
2472 indirchk(ufs_lbn_t blksperindir, ufs_lbn_t lbn, ufs2_daddr_t blkno,
2473 ufs_lbn_t lastlbn, struct gfs_bpp *bp, int fsi, int fso, unsigned int Nflag)
2475 DBG_FUNC("indirchk")
2482 /* read in the indirect block. */
2483 ibuf = malloc(sblock.fs_bsize);
2485 errx(1, "malloc failed");
2486 rdfs(fsbtodb(&sblock, blkno), (size_t)sblock.fs_bsize, ibuf, fsi);
2487 last = howmany(lastlbn - lbn, blksperindir) < NINDIR(&sblock) ?
2488 howmany(lastlbn - lbn, blksperindir) : NINDIR(&sblock);
2489 for (i = 0; i < last; i++) {
2490 if (sblock.fs_magic == FS_UFS1_MAGIC)
2491 iptr = ((ufs1_daddr_t *)ibuf)[i];
2493 iptr = ((ufs2_daddr_t *)ibuf)[i];
2496 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2497 if (sblock.fs_magic == FS_UFS1_MAGIC)
2498 ((ufs1_daddr_t *)ibuf)[i] = iptr;
2500 ((ufs2_daddr_t *)ibuf)[i] = iptr;
2502 if (blksperindir == 1)
2504 indirchk(blksperindir / NINDIR(&sblock), lbn + blksperindir * i,
2505 iptr, lastlbn, bp, fsi, fso, Nflag);