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)
183 static int randinit=0;
191 #else /* not FSIRAND */
199 * Get the cylinder summary into the memory.
201 fscs = (struct csum *)calloc((size_t)1, (size_t)sblock.fs_cssize);
203 errx(1, "calloc failed");
205 for (i = 0; i < osblock.fs_cssize; i += osblock.fs_bsize) {
206 rdfs(fsbtodb(&osblock, osblock.fs_csaddr +
207 numfrags(&osblock, i)), (size_t)MIN(osblock.fs_cssize - i,
208 osblock.fs_bsize), (void *)(((char *)fscs)+i), fsi);
213 struct csum *dbg_csp;
218 for(dbg_csc=0; dbg_csc<osblock.fs_ncg; dbg_csc++) {
219 snprintf(dbg_line, sizeof(dbg_line),
220 "%d. old csum in old location", dbg_csc);
221 DBG_DUMP_CSUM(&osblock,
226 #endif /* FS_DEBUG */
227 DBG_PRINT0("fscs read\n");
230 * Do all needed changes in the former last cylinder group.
232 updjcg(osblock.fs_ncg-1, utime, fsi, fso, Nflag);
235 * Dump out summary information about file system.
237 # define B2MBFACTOR (1 / (1024.0 * 1024.0))
238 printf("growfs: %.1fMB (%jd sectors) block size %d, fragment size %d\n",
239 (float)sblock.fs_size * sblock.fs_fsize * B2MBFACTOR,
240 (intmax_t)fsbtodb(&sblock, sblock.fs_size), sblock.fs_bsize,
242 printf("\tusing %d cylinder groups of %.2fMB, %d blks, %d inodes.\n",
243 sblock.fs_ncg, (float)sblock.fs_fpg * sblock.fs_fsize * B2MBFACTOR,
244 sblock.fs_fpg / sblock.fs_frag, sblock.fs_ipg);
245 if (sblock.fs_flags & FS_DOSOFTDEP)
246 printf("\twith soft updates\n");
250 * Now build the cylinders group blocks and
251 * then print out indices of cylinder groups.
253 printf("super-block backups (for fsck -b #) at:\n");
255 width = charsperline();
258 * Iterate for only the new cylinder groups.
260 for (cylno = osblock.fs_ncg; cylno < sblock.fs_ncg; cylno++) {
261 initcg(cylno, utime, fso, Nflag);
262 j = sprintf(tmpbuf, " %d%s",
263 (int)fsbtodb(&sblock, cgsblock(&sblock, cylno)),
264 cylno < (sblock.fs_ncg-1) ? "," : "" );
265 if (i + j >= width) {
270 printf("%s", tmpbuf);
276 * Do all needed changes in the first cylinder group.
277 * allocate blocks in new location
279 updcsloc(utime, fsi, fso, Nflag);
282 * Now write the cylinder summary back to disk.
284 for (i = 0; i < sblock.fs_cssize; i += sblock.fs_bsize) {
285 wtfs(fsbtodb(&sblock, sblock.fs_csaddr + numfrags(&sblock, i)),
286 (size_t)MIN(sblock.fs_cssize - i, sblock.fs_bsize),
287 (void *)(((char *)fscs) + i), fso, Nflag);
289 DBG_PRINT0("fscs written\n");
293 struct csum *dbg_csp;
298 for(dbg_csc=0; dbg_csc<sblock.fs_ncg; dbg_csc++) {
299 snprintf(dbg_line, sizeof(dbg_line),
300 "%d. new csum in new location", dbg_csc);
301 DBG_DUMP_CSUM(&sblock,
306 #endif /* FS_DEBUG */
309 * Now write the new superblock back to disk.
311 sblock.fs_time = utime;
312 wtfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
313 DBG_PRINT0("sblock written\n");
315 "new initial sblock");
318 * Clean up the dynamic fields in our superblock copies.
323 sblock.fs_cgrotor = 0;
325 memset((void *)&sblock.fs_fsmnt, 0, sizeof(sblock.fs_fsmnt));
326 sblock.fs_flags &= FS_DOSOFTDEP;
330 * The following fields are currently distributed from the superblock
338 * fs_flags regarding SOFTPDATES
340 * We probably should rather change the summary for the cylinder group
341 * statistics here to the value of what would be in there, if the file
342 * system were created initially with the new size. Therefor we still
343 * need to find an easy way of calculating that.
344 * Possibly we can try to read the first superblock copy and apply the
345 * "diffed" stats between the old and new superblock by still copying
346 * certain parameters onto that.
350 * Write out the duplicate super blocks.
352 for (cylno = 0; cylno < sblock.fs_ncg; cylno++) {
353 wtfs(fsbtodb(&sblock, cgsblock(&sblock, cylno)),
354 (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
356 DBG_PRINT0("sblock copies written\n");
358 "new other sblocks");
364 /* ************************************************************ initcg ***** */
366 * This creates a new cylinder group structure, for more details please see
367 * the source of newfs(8), as this function is taken over almost unchanged.
368 * As this is never called for the first cylinder group, the special
369 * provisions for that case are removed here.
372 initcg(int cylno, time_t utime, int fso, unsigned int Nflag)
376 long d, dlower, dupper, blkno, start;
377 ufs2_daddr_t i, cbase, dmax;
378 struct ufs1_dinode *dp1;
379 struct ufs2_dinode *dp2;
382 if (iobuf == NULL && (iobuf = malloc(sblock.fs_bsize)) == NULL) {
383 errx(37, "panic: cannot allocate I/O buffer");
386 * Determine block bounds for cylinder group.
387 * Allow space for super block summary information in first
390 cbase = cgbase(&sblock, cylno);
391 dmax = cbase + sblock.fs_fpg;
392 if (dmax > sblock.fs_size)
393 dmax = sblock.fs_size;
394 dlower = cgsblock(&sblock, cylno) - cbase;
395 dupper = cgdmin(&sblock, cylno) - cbase;
396 if (cylno == 0) /* XXX fscs may be relocated */
397 dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
399 memset(&acg, 0, sblock.fs_cgsize);
401 acg.cg_magic = CG_MAGIC;
403 acg.cg_niblk = sblock.fs_ipg;
404 acg.cg_initediblk = sblock.fs_ipg;
405 acg.cg_ndblk = dmax - cbase;
406 if (sblock.fs_contigsumsize > 0)
407 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
408 start = &acg.cg_space[0] - (u_char *)(&acg.cg_firstfield);
409 if (sblock.fs_magic == FS_UFS2_MAGIC) {
410 acg.cg_iusedoff = start;
412 acg.cg_old_ncyl = sblock.fs_old_cpg;
413 acg.cg_old_time = acg.cg_time;
415 acg.cg_old_niblk = acg.cg_niblk;
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);
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);
435 if (acg.cg_nextfreeoff > sblock.fs_cgsize) {
437 * This should never happen as we would have had that panic
438 * already on file system creation
440 errx(37, "panic: cylinder group too big");
442 acg.cg_cs.cs_nifree += sblock.fs_ipg;
444 for (i = 0; i < ROOTINO; i++) {
445 setbit(cg_inosused(&acg), i);
446 acg.cg_cs.cs_nifree--;
449 * XXX Newfs writes out two blocks of initialized inodes
450 * unconditionally. Should we check here to make sure that they
451 * were actually written?
453 if (sblock.fs_magic == FS_UFS1_MAGIC) {
454 bzero(iobuf, sblock.fs_bsize);
455 for (i = 2 * sblock.fs_frag; i < sblock.fs_ipg / INOPF(&sblock);
456 i += sblock.fs_frag) {
457 dp1 = (struct ufs1_dinode *)iobuf;
458 dp2 = (struct ufs2_dinode *)iobuf;
460 for (j = 0; j < INOPB(&sblock); j++)
461 if (sblock.fs_magic == FS_UFS1_MAGIC) {
462 dp1->di_gen = random();
465 dp2->di_gen = random();
469 wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno) + i),
470 sblock.fs_bsize, iobuf, fso, Nflag);
475 * In cylno 0, beginning space is reserved
476 * for boot and super blocks.
478 for (d = 0; d < dlower; d += sblock.fs_frag) {
479 blkno = d / sblock.fs_frag;
480 setblock(&sblock, cg_blksfree(&acg), blkno);
481 if (sblock.fs_contigsumsize > 0)
482 setbit(cg_clustersfree(&acg), blkno);
483 acg.cg_cs.cs_nbfree++;
485 sblock.fs_dsize += dlower;
487 sblock.fs_dsize += acg.cg_ndblk - dupper;
488 if ((i = dupper % sblock.fs_frag)) {
489 acg.cg_frsum[sblock.fs_frag - i]++;
490 for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
491 setbit(cg_blksfree(&acg), dupper);
492 acg.cg_cs.cs_nffree++;
495 for (d = dupper; d + sblock.fs_frag <= acg.cg_ndblk;
496 d += sblock.fs_frag) {
497 blkno = d / sblock.fs_frag;
498 setblock(&sblock, cg_blksfree(&acg), blkno);
499 if (sblock.fs_contigsumsize > 0)
500 setbit(cg_clustersfree(&acg), blkno);
501 acg.cg_cs.cs_nbfree++;
503 if (d < acg.cg_ndblk) {
504 acg.cg_frsum[acg.cg_ndblk - d]++;
505 for (; d < acg.cg_ndblk; d++) {
506 setbit(cg_blksfree(&acg), d);
507 acg.cg_cs.cs_nffree++;
510 if (sblock.fs_contigsumsize > 0) {
511 int32_t *sump = cg_clustersum(&acg);
512 u_char *mapp = cg_clustersfree(&acg);
517 for (i = 0; i < acg.cg_nclusterblks; i++) {
518 if ((map & bit) != 0)
521 if (run > sblock.fs_contigsumsize)
522 run = sblock.fs_contigsumsize;
526 if ((i & (CHAR_BIT - 1)) != CHAR_BIT - 1)
534 if (run > sblock.fs_contigsumsize)
535 run = sblock.fs_contigsumsize;
539 sblock.fs_cstotal.cs_ndir += acg.cg_cs.cs_ndir;
540 sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
541 sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
542 sblock.fs_cstotal.cs_nifree += acg.cg_cs.cs_nifree;
544 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
545 sblock.fs_bsize, (char *)&acg, fso, Nflag);
554 /* ******************************************************* frag_adjust ***** */
556 * Here we add or subtract (sign +1/-1) the available fragments in a given
557 * block to or from the fragment statistics. By subtracting before and adding
558 * after an operation on the free frag map we can easy update the fragment
559 * statistic, which seems to be otherwise a rather complex operation.
562 frag_adjust(ufs2_daddr_t frag, int sign)
564 DBG_FUNC("frag_adjust")
572 * Here frag only needs to point to any fragment in the block we want
575 for(f=rounddown(frag, sblock.fs_frag);
576 f<roundup(frag+1, sblock.fs_frag);
579 * Count contiguous free fragments.
581 if(isset(cg_blksfree(&acg), f)) {
584 if(fragsize && fragsize<sblock.fs_frag) {
586 * We found something in between.
588 acg.cg_frsum[fragsize]+=sign;
589 DBG_PRINT2("frag_adjust [%d]+=%d\n",
596 if(fragsize && fragsize<sblock.fs_frag) {
598 * We found something.
600 acg.cg_frsum[fragsize]+=sign;
601 DBG_PRINT2("frag_adjust [%d]+=%d\n",
605 DBG_PRINT2("frag_adjust [[%d]]+=%d\n",
613 /* ******************************************************* cond_bl_upd ***** */
615 * Here we conditionally update a pointer to a fragment. We check for all
616 * relocated blocks if any of its fragments is referenced by the current
617 * field, and update the pointer to the respective fragment in our new
618 * block. If we find a reference we write back the block immediately,
619 * as there is no easy way for our general block reading engine to figure
620 * out if a write back operation is needed.
623 cond_bl_upd(ufs2_daddr_t *block, struct gfs_bpp *field, int fsi, int fso,
626 DBG_FUNC("cond_bl_upd")
628 ufs2_daddr_t src, dst;
635 for (f = field; f->old != 0; f++) {
637 if (fragstoblks(&sblock, src) != f->old)
640 * The fragment is part of the block, so update.
642 dst = blkstofrags(&sblock, f->new);
643 fragnum = fragnum(&sblock, src);
644 *block = dst + fragnum;
646 DBG_PRINT3("scg (%jd->%jd)[%d] reference updated\n",
652 * Copy the block back immediately.
654 * XXX If src is is from an indirect block we have
655 * to implement copy on write here in case of
658 ibuf = malloc(sblock.fs_bsize);
660 errx(1, "malloc failed");
662 rdfs(fsbtodb(&sblock, src), (size_t)sblock.fs_bsize, ibuf, fsi);
663 wtfs(dst, (size_t)sblock.fs_bsize, ibuf, fso, Nflag);
666 * The same block can't be found again in this loop.
675 /* ************************************************************ updjcg ***** */
677 * Here we do all needed work for the former last cylinder group. It has to be
678 * changed in any case, even if the file system ended exactly on the end of
679 * this group, as there is some slightly inconsistent handling of the number
680 * of cylinders in the cylinder group. We start again by reading the cylinder
681 * group from disk. If the last block was not fully available, we first handle
682 * the missing fragments, then we handle all new full blocks in that file
683 * system and finally we handle the new last fragmented block in the file
684 * system. We again have to handle the fragment statistics rotational layout
685 * tables and cluster summary during all those operations.
688 updjcg(int cylno, time_t utime, int fsi, int fso, unsigned int Nflag)
691 ufs2_daddr_t cbase, dmax, dupper;
699 * Read the former last (joining) cylinder group from disk, and make
702 rdfs(fsbtodb(&osblock, cgtod(&osblock, cylno)),
703 (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
704 DBG_PRINT0("jcg read\n");
709 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
712 * If the cylinder group had already its new final size almost
713 * nothing is to be done ... except:
714 * For some reason the value of cg_ncyl in the last cylinder group has
715 * to be zero instead of fs_cpg. As this is now no longer the last
716 * cylinder group we have to change that value now to fs_cpg.
719 if(cgbase(&osblock, cylno+1) == osblock.fs_size) {
720 if (sblock.fs_magic == FS_UFS1_MAGIC)
721 acg.cg_old_ncyl=sblock.fs_old_cpg;
723 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
724 (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
725 DBG_PRINT0("jcg written\n");
735 * Set up some variables needed later.
737 cbase = cgbase(&sblock, cylno);
738 dmax = cbase + sblock.fs_fpg;
739 if (dmax > sblock.fs_size)
740 dmax = sblock.fs_size;
741 dupper = cgdmin(&sblock, cylno) - cbase;
742 if (cylno == 0) { /* XXX fscs may be relocated */
743 dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
747 * Set pointer to the cylinder summary for our cylinder group.
752 * Touch the cylinder group, update all fields in the cylinder group as
753 * needed, update the free space in the superblock.
756 if (cylno == sblock.fs_ncg - 1) {
758 * This is still the last cylinder group.
760 if (sblock.fs_magic == FS_UFS1_MAGIC)
762 sblock.fs_old_ncyl % sblock.fs_old_cpg;
764 acg.cg_old_ncyl = sblock.fs_old_cpg;
766 DBG_PRINT2("jcg dbg: %d %u",
770 if (sblock.fs_magic == FS_UFS1_MAGIC)
776 acg.cg_ndblk = dmax - cbase;
777 sblock.fs_dsize += acg.cg_ndblk-aocg.cg_ndblk;
778 if (sblock.fs_contigsumsize > 0) {
779 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
783 * Now we have to update the free fragment bitmap for our new free
784 * space. There again we have to handle the fragmentation and also
785 * the rotational layout tables and the cluster summary. This is
786 * also done per fragment for the first new block if the old file
787 * system end was not on a block boundary, per fragment for the new
788 * last block if the new file system end is not on a block boundary,
789 * and per block for all space in between.
791 * Handle the first new block here if it was partially available
794 if(osblock.fs_size % sblock.fs_frag) {
795 if(roundup(osblock.fs_size, sblock.fs_frag)<=sblock.fs_size) {
797 * The new space is enough to fill at least this
801 for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag)-1;
802 i>=osblock.fs_size-cbase;
804 setbit(cg_blksfree(&acg), i);
805 acg.cg_cs.cs_nffree++;
810 * Check if the fragment just created could join an
811 * already existing fragment at the former end of the
814 if(isblock(&sblock, cg_blksfree(&acg),
815 ((osblock.fs_size - cgbase(&sblock, cylno))/
818 * The block is now completely available.
820 DBG_PRINT0("block was\n");
821 acg.cg_frsum[osblock.fs_size%sblock.fs_frag]--;
822 acg.cg_cs.cs_nbfree++;
823 acg.cg_cs.cs_nffree-=sblock.fs_frag;
824 k=rounddown(osblock.fs_size-cbase,
826 updclst((osblock.fs_size-cbase)/sblock.fs_frag);
829 * Lets rejoin a possible partially growed
833 while(isset(cg_blksfree(&acg), i) &&
834 (i>=rounddown(osblock.fs_size-cbase,
846 * We only grow by some fragments within this last
849 for(i=sblock.fs_size-cbase-1;
850 i>=osblock.fs_size-cbase;
852 setbit(cg_blksfree(&acg), i);
853 acg.cg_cs.cs_nffree++;
857 * Lets rejoin a possible partially growed fragment.
860 while(isset(cg_blksfree(&acg), i) &&
861 (i>=rounddown(osblock.fs_size-cbase,
874 * Handle all new complete blocks here.
876 for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag);
877 i+sblock.fs_frag<=dmax-cbase; /* XXX <= or only < ? */
879 j = i / sblock.fs_frag;
880 setblock(&sblock, cg_blksfree(&acg), j);
882 acg.cg_cs.cs_nbfree++;
886 * Handle the last new block if there are stll some new fragments left.
887 * Here we don't have to bother about the cluster summary or the even
888 * the rotational layout table.
890 if (i < (dmax - cbase)) {
891 acg.cg_frsum[dmax - cbase - i]++;
892 for (; i < dmax - cbase; i++) {
893 setbit(cg_blksfree(&acg), i);
894 acg.cg_cs.cs_nffree++;
898 sblock.fs_cstotal.cs_nffree +=
899 (acg.cg_cs.cs_nffree - aocg.cg_cs.cs_nffree);
900 sblock.fs_cstotal.cs_nbfree +=
901 (acg.cg_cs.cs_nbfree - aocg.cg_cs.cs_nbfree);
903 * The following statistics are not changed here:
904 * sblock.fs_cstotal.cs_ndir
905 * sblock.fs_cstotal.cs_nifree
906 * As the statistics for this cylinder group are ready, copy it to
907 * the summary information array.
912 * Write the updated "joining" cylinder group back to disk.
914 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize,
915 (void *)&acg, fso, Nflag);
916 DBG_PRINT0("jcg written\n");
925 /* ********************************************************** updcsloc ***** */
927 * Here we update the location of the cylinder summary. We have two possible
928 * ways of growing the cylinder summary.
929 * (1) We can try to grow the summary in the current location, and relocate
930 * possibly used blocks within the current cylinder group.
931 * (2) Alternatively we can relocate the whole cylinder summary to the first
932 * new completely empty cylinder group. Once the cylinder summary is no
933 * longer in the beginning of the first cylinder group you should never
934 * use a version of fsck which is not aware of the possibility to have
935 * this structure in a non standard place.
936 * Option (1) is considered to be less intrusive to the structure of the file-
937 * system. So we try to stick to that whenever possible. If there is not enough
938 * space in the cylinder group containing the cylinder summary we have to use
939 * method (2). In case of active snapshots in the file system we probably can
940 * completely avoid implementing copy on write if we stick to method (2) only.
943 updcsloc(time_t utime, int fsi, int fso, unsigned int Nflag)
949 ufs2_daddr_t cbase, dupper, odupper, d, f, g;
959 if(howmany(sblock.fs_cssize, sblock.fs_fsize) ==
960 howmany(osblock.fs_cssize, osblock.fs_fsize)) {
962 * No new fragment needed.
967 ocscg=dtog(&osblock, osblock.fs_csaddr);
969 blocks = 1+howmany(sblock.fs_cssize, sblock.fs_bsize)-
970 howmany(osblock.fs_cssize, osblock.fs_bsize);
973 * Read original cylinder group from disk, and make a copy.
974 * XXX If Nflag is set in some very rare cases we now miss
975 * some changes done in updjcg by reading the unmodified
978 rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)),
979 (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
980 DBG_PRINT0("oscg read\n");
985 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
988 * Touch the cylinder group, set up local variables needed later
989 * and update the superblock.
994 * XXX In the case of having active snapshots we may need much more
995 * blocks for the copy on write. We need each block twice, and
996 * also up to 8*3 blocks for indirect blocks for all possible
999 if(/*((int)sblock.fs_time&0x3)>0||*/ cs->cs_nbfree < blocks) {
1001 * There is not enough space in the old cylinder group to
1002 * relocate all blocks as needed, so we relocate the whole
1003 * cylinder group summary to a new group. We try to use the
1004 * first complete new cylinder group just created. Within the
1005 * cylinder group we align the area immediately after the
1006 * cylinder group information location in order to be as
1007 * close as possible to the original implementation of ffs.
1009 * First we have to make sure we'll find enough space in the
1010 * new cylinder group. If not, then we currently give up.
1011 * We start with freeing everything which was used by the
1012 * fragments of the old cylinder summary in the current group.
1013 * Now we write back the group meta data, read in the needed
1014 * meta data from the new cylinder group, and start allocating
1015 * within that group. Here we can assume, the group to be
1016 * completely empty. Which makes the handling of fragments and
1017 * clusters a lot easier.
1020 if(sblock.fs_ncg-osblock.fs_ncg < 2) {
1021 errx(2, "panic: not enough space");
1025 * Point "d" to the first fragment not used by the cylinder
1028 d=osblock.fs_csaddr+(osblock.fs_cssize/osblock.fs_fsize);
1031 * Set up last cluster size ("lcs") already here. Calculate
1032 * the size for the trailing cluster just behind where "d"
1035 if(sblock.fs_contigsumsize > 0) {
1036 for(block=howmany(d%sblock.fs_fpg, sblock.fs_frag),
1037 lcs=0; lcs<sblock.fs_contigsumsize;
1039 if(isclr(cg_clustersfree(&acg), block)){
1046 * Point "d" to the last frag used by the cylinder summary.
1050 DBG_PRINT1("d=%jd\n",
1052 if((d+1)%sblock.fs_frag) {
1054 * The end of the cylinder summary is not a complete
1058 frag_adjust(d%sblock.fs_fpg, -1);
1059 for(; (d+1)%sblock.fs_frag; d--) {
1060 DBG_PRINT1("d=%jd\n",
1062 setbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1063 acg.cg_cs.cs_nffree++;
1064 sblock.fs_cstotal.cs_nffree++;
1067 * Point "d" to the last fragment of the last
1068 * (incomplete) block of the cylinder summary.
1071 frag_adjust(d%sblock.fs_fpg, 1);
1073 if(isblock(&sblock, cg_blksfree(&acg),
1074 (d%sblock.fs_fpg)/sblock.fs_frag)) {
1075 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1076 acg.cg_cs.cs_nffree-=sblock.fs_frag;
1077 acg.cg_cs.cs_nbfree++;
1078 sblock.fs_cstotal.cs_nffree-=sblock.fs_frag;
1079 sblock.fs_cstotal.cs_nbfree++;
1080 if(sblock.fs_contigsumsize > 0) {
1081 setbit(cg_clustersfree(&acg),
1082 (d%sblock.fs_fpg)/sblock.fs_frag);
1083 if(lcs < sblock.fs_contigsumsize) {
1089 cg_clustersum(&acg)[lcs]++;
1094 * Point "d" to the first fragment of the block before
1095 * the last incomplete block.
1100 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1101 for(d=rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr;
1102 d-=sblock.fs_frag) {
1104 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1105 setblock(&sblock, cg_blksfree(&acg),
1106 (d%sblock.fs_fpg)/sblock.fs_frag);
1107 acg.cg_cs.cs_nbfree++;
1108 sblock.fs_cstotal.cs_nbfree++;
1109 if(sblock.fs_contigsumsize > 0) {
1110 setbit(cg_clustersfree(&acg),
1111 (d%sblock.fs_fpg)/sblock.fs_frag);
1113 * The last cluster size is already set up.
1115 if(lcs < sblock.fs_contigsumsize) {
1117 cg_clustersum(&acg)[lcs]--;
1120 cg_clustersum(&acg)[lcs]++;
1127 * Now write the former cylinder group containing the cylinder
1128 * summary back to disk.
1130 wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)),
1131 (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1132 DBG_PRINT0("oscg written\n");
1133 DBG_DUMP_CG(&sblock,
1138 * Find the beginning of the new cylinder group containing the
1141 sblock.fs_csaddr=cgdmin(&sblock, osblock.fs_ncg);
1142 ncscg=dtog(&sblock, sblock.fs_csaddr);
1147 * If Nflag is specified, we would now read random data instead
1148 * of an empty cg structure from disk. So we can't simulate that
1152 DBG_PRINT0("nscg update skipped\n");
1158 * Read the future cylinder group containing the cylinder
1159 * summary from disk, and make a copy.
1161 rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1162 (size_t)sblock.fs_cgsize, (void *)&aocg, fsi);
1163 DBG_PRINT0("nscg read\n");
1164 DBG_DUMP_CG(&sblock,
1168 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
1171 * Allocate all complete blocks used by the new cylinder
1174 for(d=sblock.fs_csaddr; d+sblock.fs_frag <=
1175 sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize);
1176 d+=sblock.fs_frag) {
1177 clrblock(&sblock, cg_blksfree(&acg),
1178 (d%sblock.fs_fpg)/sblock.fs_frag);
1179 acg.cg_cs.cs_nbfree--;
1180 sblock.fs_cstotal.cs_nbfree--;
1181 if(sblock.fs_contigsumsize > 0) {
1182 clrbit(cg_clustersfree(&acg),
1183 (d%sblock.fs_fpg)/sblock.fs_frag);
1188 * Allocate all fragments used by the cylinder summary in the
1191 if(d<sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize)) {
1192 for(; d-sblock.fs_csaddr<
1193 sblock.fs_cssize/sblock.fs_fsize;
1195 clrbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1196 acg.cg_cs.cs_nffree--;
1197 sblock.fs_cstotal.cs_nffree--;
1199 acg.cg_cs.cs_nbfree--;
1200 acg.cg_cs.cs_nffree+=sblock.fs_frag;
1201 sblock.fs_cstotal.cs_nbfree--;
1202 sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1203 if(sblock.fs_contigsumsize > 0) {
1204 clrbit(cg_clustersfree(&acg),
1205 (d%sblock.fs_fpg)/sblock.fs_frag);
1208 frag_adjust(d%sblock.fs_fpg, +1);
1211 * XXX Handle the cluster statistics here in the case this
1212 * cylinder group is now almost full, and the remaining
1213 * space is less then the maximum cluster size. This is
1214 * probably not needed, as you would hardly find a file
1215 * system which has only MAXCSBUFS+FS_MAXCONTIG of free
1216 * space right behind the cylinder group information in
1217 * any new cylinder group.
1221 * Update our statistics in the cylinder summary.
1226 * Write the new cylinder group containing the cylinder summary
1229 wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1230 (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1231 DBG_PRINT0("nscg written\n");
1232 DBG_DUMP_CG(&sblock,
1240 * We have got enough of space in the current cylinder group, so we
1241 * can relocate just a few blocks, and let the summary information
1242 * grow in place where it is right now.
1246 cbase = cgbase(&osblock, ocscg); /* old and new are equal */
1247 dupper = sblock.fs_csaddr - cbase +
1248 howmany(sblock.fs_cssize, sblock.fs_fsize);
1249 odupper = osblock.fs_csaddr - cbase +
1250 howmany(osblock.fs_cssize, osblock.fs_fsize);
1252 sblock.fs_dsize -= dupper-odupper;
1255 * Allocate the space for the array of blocks to be relocated.
1257 bp=(struct gfs_bpp *)malloc(((dupper-odupper)/sblock.fs_frag+2)*
1258 sizeof(struct gfs_bpp));
1260 errx(1, "malloc failed");
1262 memset((char *)bp, 0, ((dupper-odupper)/sblock.fs_frag+2)*
1263 sizeof(struct gfs_bpp));
1266 * Lock all new frags needed for the cylinder group summary. This is
1267 * done per fragment in the first and last block of the new required
1268 * area, and per block for all other blocks.
1270 * Handle the first new block here (but only if some fragments where
1271 * already used for the cylinder summary).
1274 frag_adjust(odupper, -1);
1275 for(d=odupper; ((d<dupper)&&(d%sblock.fs_frag)); d++) {
1276 DBG_PRINT1("scg first frag check loop d=%jd\n",
1278 if(isclr(cg_blksfree(&acg), d)) {
1280 bp[ind].old=d/sblock.fs_frag;
1281 bp[ind].flags|=GFS_FL_FIRST;
1282 if(roundup(d, sblock.fs_frag) >= dupper) {
1283 bp[ind].flags|=GFS_FL_LAST;
1288 clrbit(cg_blksfree(&acg), d);
1289 acg.cg_cs.cs_nffree--;
1290 sblock.fs_cstotal.cs_nffree--;
1293 * No cluster handling is needed here, as there was at least
1294 * one fragment in use by the cylinder summary in the old
1296 * No block-free counter handling here as this block was not
1300 frag_adjust(odupper, 1);
1303 * Handle all needed complete blocks here.
1305 for(; d+sblock.fs_frag<=dupper; d+=sblock.fs_frag) {
1306 DBG_PRINT1("scg block check loop d=%jd\n",
1308 if(!isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1309 for(f=d; f<d+sblock.fs_frag; f++) {
1310 if(isset(cg_blksfree(&aocg), f)) {
1311 acg.cg_cs.cs_nffree--;
1312 sblock.fs_cstotal.cs_nffree--;
1315 clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1316 bp[ind].old=d/sblock.fs_frag;
1319 clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1320 acg.cg_cs.cs_nbfree--;
1321 sblock.fs_cstotal.cs_nbfree--;
1322 if(sblock.fs_contigsumsize > 0) {
1323 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1324 for(lcs=0, l=(d/sblock.fs_frag)+1;
1325 lcs<sblock.fs_contigsumsize;
1327 if(isclr(cg_clustersfree(&acg),l)){
1331 if(lcs < sblock.fs_contigsumsize) {
1332 cg_clustersum(&acg)[lcs+1]--;
1334 cg_clustersum(&acg)[lcs]++;
1340 * No fragment counter handling is needed here, as this finally
1341 * doesn't change after the relocation.
1346 * Handle all fragments needed in the last new affected block.
1349 frag_adjust(dupper-1, -1);
1351 if(isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1352 acg.cg_cs.cs_nbfree--;
1353 sblock.fs_cstotal.cs_nbfree--;
1354 acg.cg_cs.cs_nffree+=sblock.fs_frag;
1355 sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1356 if(sblock.fs_contigsumsize > 0) {
1357 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1358 for(lcs=0, l=(d/sblock.fs_frag)+1;
1359 lcs<sblock.fs_contigsumsize;
1361 if(isclr(cg_clustersfree(&acg),l)){
1365 if(lcs < sblock.fs_contigsumsize) {
1366 cg_clustersum(&acg)[lcs+1]--;
1368 cg_clustersum(&acg)[lcs]++;
1374 for(; d<dupper; d++) {
1375 DBG_PRINT1("scg second frag check loop d=%jd\n",
1377 if(isclr(cg_blksfree(&acg), d)) {
1378 bp[ind].old=d/sblock.fs_frag;
1379 bp[ind].flags|=GFS_FL_LAST;
1381 clrbit(cg_blksfree(&acg), d);
1382 acg.cg_cs.cs_nffree--;
1383 sblock.fs_cstotal.cs_nffree--;
1386 if(bp[ind].flags & GFS_FL_LAST) { /* we have to advance here */
1389 frag_adjust(dupper-1, 1);
1393 * If we found a block to relocate just do so.
1396 for(i=0; i<ind; i++) {
1397 if(!bp[i].old) { /* no more blocks listed */
1399 * XXX A relative blocknumber should not be
1400 * zero, which is not explicitly
1401 * guaranteed by our code.
1406 * Allocate a complete block in the same (current)
1409 bp[i].new=alloc()/sblock.fs_frag;
1412 * There is no frag_adjust() needed for the new block
1413 * as it will have no fragments yet :-).
1415 for(f=bp[i].old*sblock.fs_frag,
1416 g=bp[i].new*sblock.fs_frag;
1417 f<(bp[i].old+1)*sblock.fs_frag;
1419 if(isset(cg_blksfree(&aocg), f)) {
1420 setbit(cg_blksfree(&acg), g);
1421 acg.cg_cs.cs_nffree++;
1422 sblock.fs_cstotal.cs_nffree++;
1427 * Special handling is required if this was the first
1428 * block. We have to consider the fragments which were
1429 * used by the cylinder summary in the original block
1430 * which re to be free in the copy of our block. We
1431 * have to be careful if this first block happens to
1432 * be also the last block to be relocated.
1434 if(bp[i].flags & GFS_FL_FIRST) {
1435 for(f=bp[i].old*sblock.fs_frag,
1436 g=bp[i].new*sblock.fs_frag;
1439 setbit(cg_blksfree(&acg), g);
1440 acg.cg_cs.cs_nffree++;
1441 sblock.fs_cstotal.cs_nffree++;
1443 if(!(bp[i].flags & GFS_FL_LAST)) {
1444 frag_adjust(bp[i].new*sblock.fs_frag,1);
1449 * Special handling is required if this is the last
1450 * block to be relocated.
1452 if(bp[i].flags & GFS_FL_LAST) {
1453 frag_adjust(bp[i].new*sblock.fs_frag, 1);
1454 frag_adjust(bp[i].old*sblock.fs_frag, -1);
1456 f<roundup(dupper, sblock.fs_frag);
1458 if(isclr(cg_blksfree(&acg), f)) {
1459 setbit(cg_blksfree(&acg), f);
1460 acg.cg_cs.cs_nffree++;
1461 sblock.fs_cstotal.cs_nffree++;
1464 frag_adjust(bp[i].old*sblock.fs_frag, 1);
1468 * !!! Attach the cylindergroup offset here.
1470 bp[i].old+=cbase/sblock.fs_frag;
1471 bp[i].new+=cbase/sblock.fs_frag;
1474 * Copy the content of the block.
1477 * XXX Here we will have to implement a copy on write
1478 * in the case we have any active snapshots.
1480 rdfs(fsbtodb(&sblock, bp[i].old*sblock.fs_frag),
1481 (size_t)sblock.fs_bsize, (void *)&ablk, fsi);
1482 wtfs(fsbtodb(&sblock, bp[i].new*sblock.fs_frag),
1483 (size_t)sblock.fs_bsize, (void *)&ablk, fso, Nflag);
1484 DBG_DUMP_HEX(&sblock,
1485 "copied full block",
1486 (unsigned char *)&ablk);
1488 DBG_PRINT2("scg (%jd->%jd) block relocated\n",
1489 (intmax_t)bp[i].old,
1490 (intmax_t)bp[i].new);
1494 * Now we have to update all references to any fragment which
1495 * belongs to any block relocated. We iterate now over all
1496 * cylinder groups, within those over all non zero length
1499 for(cylno=0; cylno<osblock.fs_ncg; cylno++) {
1500 DBG_PRINT1("scg doing cg (%d)\n",
1502 for(inc=osblock.fs_ipg-1 ; inc>0 ; inc--) {
1503 updrefs(cylno, (ino_t)inc, bp, fsi, fso, Nflag);
1508 * All inodes are checked, now make sure the number of
1509 * references found make sense.
1511 for(i=0; i<ind; i++) {
1512 if(!bp[i].found || (bp[i].found>sblock.fs_frag)) {
1513 warnx("error: %jd refs found for block %jd.",
1514 (intmax_t)bp[i].found, (intmax_t)bp[i].old);
1520 * The following statistics are not changed here:
1521 * sblock.fs_cstotal.cs_ndir
1522 * sblock.fs_cstotal.cs_nifree
1523 * The following statistics were already updated on the fly:
1524 * sblock.fs_cstotal.cs_nffree
1525 * sblock.fs_cstotal.cs_nbfree
1526 * As the statistics for this cylinder group are ready, copy it to
1527 * the summary information array.
1533 * Write summary cylinder group back to disk.
1535 wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)), (size_t)sblock.fs_cgsize,
1536 (void *)&acg, fso, Nflag);
1537 DBG_PRINT0("scg written\n");
1538 DBG_DUMP_CG(&sblock,
1546 /* ************************************************************** rdfs ***** */
1548 * Here we read some block(s) from disk.
1551 rdfs(ufs2_daddr_t bno, size_t size, void *bf, int fsi)
1559 err(32, "rdfs: attempting to read negative block number");
1561 if (lseek(fsi, (off_t)bno * DEV_BSIZE, 0) < 0) {
1562 err(33, "rdfs: seek error: %jd", (intmax_t)bno);
1564 n = read(fsi, bf, size);
1565 if (n != (ssize_t)size) {
1566 err(34, "rdfs: read error: %jd", (intmax_t)bno);
1573 /* ************************************************************** wtfs ***** */
1575 * Here we write some block(s) to disk.
1578 wtfs(ufs2_daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag)
1589 if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0) {
1590 err(35, "wtfs: seek error: %ld", (long)bno);
1592 n = write(fso, bf, size);
1593 if (n != (ssize_t)size) {
1594 err(36, "wtfs: write error: %ld", (long)bno);
1601 /* ************************************************************* alloc ***** */
1603 * Here we allocate a free block in the current cylinder group. It is assumed,
1604 * that acg contains the current cylinder group. As we may take a block from
1605 * somewhere in the file system we have to handle cluster summary here.
1611 ufs2_daddr_t d, blkno;
1615 int dlower, dupper, dmax;
1619 if (acg.cg_magic != CG_MAGIC) {
1620 warnx("acg: bad magic number");
1624 if (acg.cg_cs.cs_nbfree == 0) {
1625 warnx("error: cylinder group ran out of space");
1630 * We start seeking for free blocks only from the space available after
1631 * the end of the new grown cylinder summary. Otherwise we allocate a
1632 * block here which we have to relocate a couple of seconds later again
1633 * again, and we are not prepared to to this anyway.
1636 dlower=cgsblock(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1637 dupper=cgdmin(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1638 dmax=cgbase(&sblock, acg.cg_cgx)+sblock.fs_fpg;
1639 if (dmax > sblock.fs_size) {
1640 dmax = sblock.fs_size;
1642 dmax-=cgbase(&sblock, acg.cg_cgx); /* retransform into cg */
1643 csmin=sblock.fs_csaddr-cgbase(&sblock, acg.cg_cgx);
1644 csmax=csmin+howmany(sblock.fs_cssize, sblock.fs_fsize);
1645 DBG_PRINT3("seek range: dl=%d, du=%d, dm=%d\n",
1649 DBG_PRINT2("range cont: csmin=%d, csmax=%d\n",
1653 for(d=0; (d<dlower && blkno==-1); d+=sblock.fs_frag) {
1654 if(d>=csmin && d<=csmax) {
1657 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1659 blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1663 for(d=dupper; (d<dmax && blkno==-1); d+=sblock.fs_frag) {
1664 if(d>=csmin && d<=csmax) {
1667 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1669 blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1674 warnx("internal error: couldn't find promised block in cg");
1680 * This is needed if the block was found already in the first loop.
1682 d=blkstofrags(&sblock, blkno);
1684 clrblock(&sblock, cg_blksfree(&acg), blkno);
1685 if (sblock.fs_contigsumsize > 0) {
1687 * Handle the cluster allocation bitmap.
1689 clrbit(cg_clustersfree(&acg), blkno);
1691 * We possibly have split a cluster here, so we have to do
1692 * recalculate the sizes of the remaining cluster halves now,
1693 * and use them for updating the cluster summary information.
1695 * Lets start with the blocks before our allocated block ...
1697 for(lcs1=0, l=blkno-1; lcs1<sblock.fs_contigsumsize;
1699 if(isclr(cg_clustersfree(&acg),l)){
1704 * ... and continue with the blocks right after our allocated
1707 for(lcs2=0, l=blkno+1; lcs2<sblock.fs_contigsumsize;
1709 if(isclr(cg_clustersfree(&acg),l)){
1715 * Now update all counters.
1717 cg_clustersum(&acg)[MIN(lcs1+lcs2+1,sblock.fs_contigsumsize)]--;
1719 cg_clustersum(&acg)[lcs1]++;
1722 cg_clustersum(&acg)[lcs2]++;
1726 * Update all statistics based on blocks.
1728 acg.cg_cs.cs_nbfree--;
1729 sblock.fs_cstotal.cs_nbfree--;
1735 /* *********************************************************** isblock ***** */
1737 * Here we check if all frags of a block are free. For more details again
1738 * please see the source of newfs(8), as this function is taken over almost
1742 isblock(struct fs *fs, unsigned char *cp, int h)
1749 switch (fs->fs_frag) {
1752 return (cp[h] == 0xff);
1754 mask = 0x0f << ((h & 0x1) << 2);
1756 return ((cp[h >> 1] & mask) == mask);
1758 mask = 0x03 << ((h & 0x3) << 1);
1760 return ((cp[h >> 2] & mask) == mask);
1762 mask = 0x01 << (h & 0x7);
1764 return ((cp[h >> 3] & mask) == mask);
1766 fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
1772 /* ********************************************************** clrblock ***** */
1774 * Here we allocate a complete block in the block map. For more details again
1775 * please see the source of newfs(8), as this function is taken over almost
1779 clrblock(struct fs *fs, unsigned char *cp, int h)
1781 DBG_FUNC("clrblock")
1785 switch ((fs)->fs_frag) {
1790 cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1793 cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1796 cp[h >> 3] &= ~(0x01 << (h & 0x7));
1799 warnx("clrblock bad fs_frag %d", fs->fs_frag);
1807 /* ********************************************************** setblock ***** */
1809 * Here we free a complete block in the free block map. For more details again
1810 * please see the source of newfs(8), as this function is taken over almost
1814 setblock(struct fs *fs, unsigned char *cp, int h)
1816 DBG_FUNC("setblock")
1820 switch (fs->fs_frag) {
1825 cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1828 cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1831 cp[h >> 3] |= (0x01 << (h & 0x7));
1834 warnx("setblock bad fs_frag %d", fs->fs_frag);
1842 /* ************************************************************ ginode ***** */
1844 * This function provides access to an individual inode. We find out in which
1845 * block the requested inode is located, read it from disk if needed, and
1846 * return the pointer into that block. We maintain a cache of one block to
1847 * not read the same block again and again if we iterate linearly over all
1850 static union dinode *
1851 ginode(ino_t inumber, int fsi, int cg)
1854 static ino_t startinum = 0; /* first inode in cached block */
1859 * The inumber passed in is relative to the cg, so use it here to see
1860 * if the inode has been allocated yet.
1862 if (isclr(cg_inosused(&aocg), inumber)) {
1867 * Now make the inumber relative to the entire inode space so it can
1868 * be sanity checked.
1870 inumber += (cg * sblock.fs_ipg);
1871 if (inumber < ROOTINO) {
1875 if (inumber > maxino)
1876 errx(8, "bad inode number %d to ginode", inumber);
1877 if (startinum == 0 ||
1878 inumber < startinum || inumber >= startinum + INOPB(&sblock)) {
1879 inoblk = fsbtodb(&sblock, ino_to_fsba(&sblock, inumber));
1880 rdfs(inoblk, (size_t)sblock.fs_bsize, inobuf, fsi);
1881 startinum = (inumber / INOPB(&sblock)) * INOPB(&sblock);
1884 if (sblock.fs_magic == FS_UFS1_MAGIC)
1885 return (union dinode *)((uintptr_t)inobuf +
1886 (inumber % INOPB(&sblock)) * sizeof(struct ufs1_dinode));
1887 return (union dinode *)((uintptr_t)inobuf +
1888 (inumber % INOPB(&sblock)) * sizeof(struct ufs2_dinode));
1891 /* ****************************************************** charsperline ***** */
1893 * Figure out how many lines our current terminal has. For more details again
1894 * please see the source of newfs(8), as this function is taken over almost
1900 DBG_FUNC("charsperline")
1908 if (ioctl(0, TIOCGWINSZ, &ws) != -1) {
1909 columns = ws.ws_col;
1911 if (columns == 0 && (cp = getenv("COLUMNS"))) {
1915 columns = 80; /* last resort */
1922 /* ****************************************************** get_dev_size ***** */
1924 * Get the size of the partition if we can't figure it out from the disklabel,
1925 * e.g. from vinum volumes.
1928 get_dev_size(int fd, int *size)
1933 if (ioctl(fd, DIOCGSECTORSIZE, §orsize) == -1)
1934 err(1,"DIOCGSECTORSIZE");
1935 if (ioctl(fd, DIOCGMEDIASIZE, &mediasize) == -1)
1936 err(1,"DIOCGMEDIASIZE");
1938 if (sectorsize <= 0)
1939 errx(1, "bogus sectorsize: %d", sectorsize);
1941 *size = mediasize / sectorsize;
1944 /* ************************************************************** main ***** */
1946 * growfs(8) is a utility which allows to increase the size of an existing
1947 * ufs file system. Currently this can only be done on unmounted file system.
1948 * It recognizes some command line options to specify the new desired size,
1949 * and it does some basic checkings. The old file system size is determined
1950 * and after some more checks like we can really access the new last block
1951 * on the disk etc. we calculate the new parameters for the superblock. After
1952 * having done this we just call growfs() which will do the work. Before
1953 * we finish the only thing left is to update the disklabel.
1954 * We still have to provide support for snapshots. Therefore we first have to
1955 * understand what data structures are always replicated in the snapshot on
1956 * creation, for all other blocks we touch during our procedure, we have to
1957 * keep the old blocks unchanged somewhere available for the snapshots. If we
1958 * are lucky, then we only have to handle our blocks to be relocated in that
1960 * Also we have to consider in what order we actually update the critical
1961 * data structures of the file system to make sure, that in case of a disaster
1962 * fsck(8) is still able to restore any lost data.
1963 * The foreseen last step then will be to provide for growing even mounted
1964 * file systems. There we have to extend the mount() system call to provide
1965 * userland access to the file system locking facility.
1968 main(int argc, char **argv)
1971 char *device, *special, *cp;
1973 unsigned int size=0;
1975 unsigned int Nflag=0;
1978 struct disklabel *lp;
1979 struct partition *pp;
1985 #endif /* FSMAXSNAP */
1989 while((ch=getopt(argc, argv, "Ns:vy")) != -1) {
1995 size=(size_t)atol(optarg);
2000 case 'v': /* for compatibility to newfs */
2020 * Now try to guess the (raw)device name.
2022 if (0 == strrchr(device, '/')) {
2024 * No path prefix was given, so try in that order:
2030 * FreeBSD now doesn't distinguish between raw and block
2031 * devices any longer, but it should still work this way.
2033 len=strlen(device)+strlen(_PATH_DEV)+2+strlen("vinum/");
2034 special=(char *)malloc(len);
2035 if(special == NULL) {
2036 errx(1, "malloc failed");
2038 snprintf(special, len, "%sr%s", _PATH_DEV, device);
2039 if (stat(special, &st) == -1) {
2040 snprintf(special, len, "%s%s", _PATH_DEV, device);
2041 if (stat(special, &st) == -1) {
2042 snprintf(special, len, "%svinum/r%s",
2044 if (stat(special, &st) == -1) {
2045 /* For now this is the 'last resort' */
2046 snprintf(special, len, "%svinum/%s",
2055 * Try to access our devices for writing ...
2060 fso = open(device, O_WRONLY);
2062 err(1, "%s", device);
2069 fsi = open(device, O_RDONLY);
2071 err(1, "%s", device);
2075 * Try to read a label and guess the slice if not specified. This
2076 * code should guess the right thing and avoid to bother the user
2077 * with the task of specifying the option -v on vinum volumes.
2079 cp=device+strlen(device)-1;
2080 lp = get_disklabel(fsi);
2084 pp = &lp->d_partitions[2];
2085 } else if (*cp>='a' && *cp<='h') {
2086 pp = &lp->d_partitions[*cp - 'a'];
2088 errx(1, "unknown device");
2090 p_size = pp->p_size;
2092 get_dev_size(fsi, &p_size);
2096 * Check if that partition is suitable for growing a file system.
2099 errx(1, "partition is unavailable");
2103 * Read the current superblock, and take a backup.
2105 for (i = 0; sblock_try[i] != -1; i++) {
2106 sblockloc = sblock_try[i] / DEV_BSIZE;
2107 rdfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&(osblock), fsi);
2108 if ((osblock.fs_magic == FS_UFS1_MAGIC ||
2109 (osblock.fs_magic == FS_UFS2_MAGIC &&
2110 osblock.fs_sblockloc == sblock_try[i])) &&
2111 osblock.fs_bsize <= MAXBSIZE &&
2112 osblock.fs_bsize >= (int32_t) sizeof(struct fs))
2115 if (sblock_try[i] == -1) {
2116 errx(1, "superblock not recognized");
2118 memcpy((void *)&fsun1, (void *)&fsun2, sizeof(fsun2));
2119 maxino = sblock.fs_ncg * sblock.fs_ipg;
2121 DBG_OPEN("/tmp/growfs.debug"); /* already here we need a superblock */
2122 DBG_DUMP_FS(&sblock,
2126 * Determine size to grow to. Default to the full size specified in
2129 sblock.fs_size = dbtofsb(&osblock, p_size);
2132 errx(1, "there is not enough space (%d < %d)",
2135 sblock.fs_size = dbtofsb(&osblock, size);
2139 * Are we really growing ?
2141 if(osblock.fs_size >= sblock.fs_size) {
2142 errx(1, "we are not growing (%jd->%jd)",
2143 (intmax_t)osblock.fs_size, (intmax_t)sblock.fs_size);
2149 * Check if we find an active snapshot.
2151 if(ExpertFlag == 0) {
2152 for(j=0; j<FSMAXSNAP; j++) {
2153 if(sblock.fs_snapinum[j]) {
2154 errx(1, "active snapshot found in file system\n"
2155 " please remove all snapshots before "
2158 if(!sblock.fs_snapinum[j]) { /* list is dense */
2165 if (ExpertFlag == 0 && Nflag == 0) {
2166 printf("We strongly recommend you to make a backup "
2167 "before growing the Filesystem\n\n"
2168 " Did you backup your data (Yes/No) ? ");
2169 fgets(reply, (int)sizeof(reply), stdin);
2170 if (strcmp(reply, "Yes\n")){
2171 printf("\n Nothing done \n");
2176 printf("new file systemsize is: %jd frags\n", (intmax_t)sblock.fs_size);
2179 * Try to access our new last block in the file system. Even if we
2180 * later on realize we have to abort our operation, on that block
2181 * there should be no data, so we can't destroy something yet.
2183 wtfs((ufs2_daddr_t)p_size-1, (size_t)DEV_BSIZE, (void *)&sblock,
2187 * Now calculate new superblock values and check for reasonable
2188 * bound for new file system size:
2189 * fs_size: is derived from label or user input
2190 * fs_dsize: should get updated in the routines creating or
2191 * updating the cylinder groups on the fly
2192 * fs_cstotal: should get updated in the routines creating or
2193 * updating the cylinder groups
2197 * Update the number of cylinders and cylinder groups in the file system.
2199 if (sblock.fs_magic == FS_UFS1_MAGIC) {
2200 sblock.fs_old_ncyl =
2201 sblock.fs_size * sblock.fs_old_nspf / sblock.fs_old_spc;
2202 if (sblock.fs_size * sblock.fs_old_nspf >
2203 sblock.fs_old_ncyl * sblock.fs_old_spc)
2204 sblock.fs_old_ncyl++;
2206 sblock.fs_ncg = howmany(sblock.fs_size, sblock.fs_fpg);
2207 maxino = sblock.fs_ncg * sblock.fs_ipg;
2209 if (sblock.fs_size % sblock.fs_fpg != 0 &&
2210 sblock.fs_size % sblock.fs_fpg < cgdmin(&sblock, sblock.fs_ncg)) {
2212 * The space in the new last cylinder group is too small,
2216 if (sblock.fs_magic == FS_UFS1_MAGIC)
2217 sblock.fs_old_ncyl = sblock.fs_ncg * sblock.fs_old_cpg;
2218 printf("Warning: %jd sector(s) cannot be allocated.\n",
2219 (intmax_t)fsbtodb(&sblock, sblock.fs_size % sblock.fs_fpg));
2220 sblock.fs_size = sblock.fs_ncg * sblock.fs_fpg;
2224 * Update the space for the cylinder group summary information in the
2225 * respective cylinder group data area.
2228 fragroundup(&sblock, sblock.fs_ncg * sizeof(struct csum));
2230 if(osblock.fs_size >= sblock.fs_size) {
2231 errx(1, "not enough new space");
2234 DBG_PRINT0("sblock calculated\n");
2237 * Ok, everything prepared, so now let's do the tricks.
2239 growfs(fsi, fso, Nflag);
2242 * Update the disk label.
2245 pp->p_fsize = sblock.fs_fsize;
2246 pp->p_frag = sblock.fs_frag;
2247 pp->p_cpg = sblock.fs_fpg;
2249 return_disklabel(fso, lp, Nflag);
2250 DBG_PRINT0("label rewritten\n");
2254 if(fso>-1) close(fso);
2262 /* ************************************************** return_disklabel ***** */
2264 * Write the updated disklabel back to disk.
2267 return_disklabel(int fd, struct disklabel *lp, unsigned int Nflag)
2269 DBG_FUNC("return_disklabel")
2285 * recalculate checksum
2287 while(ptr < (u_short *)&lp->d_partitions[lp->d_npartitions]) {
2292 if (ioctl(fd, DIOCWDINFO, (char *)lp) < 0) {
2293 errx(1, "DIOCWDINFO failed");
2302 /* ***************************************************** get_disklabel ***** */
2304 * Read the disklabel from disk.
2306 static struct disklabel *
2307 get_disklabel(int fd)
2309 DBG_FUNC("get_disklabel")
2310 static struct disklabel *lab;
2314 lab=(struct disklabel *)malloc(sizeof(struct disklabel));
2316 errx(1, "malloc failed");
2318 if (!ioctl(fd, DIOCGDINFO, (char *)lab))
2328 /* ************************************************************* usage ***** */
2330 * Dump a line of usage.
2339 fprintf(stderr, "usage: growfs [-Ny] [-s size] special\n");
2345 /* *********************************************************** updclst ***** */
2347 * This updates most parameters and the bitmap related to cluster. We have to
2348 * assume that sblock, osblock, acg are set up.
2358 if(sblock.fs_contigsumsize < 1) { /* no clustering */
2362 * update cluster allocation map
2364 setbit(cg_clustersfree(&acg), block);
2367 * update cluster summary table
2371 * calculate size for the trailing cluster
2373 for(block--; lcs<sblock.fs_contigsumsize; block--, lcs++ ) {
2374 if(isclr(cg_clustersfree(&acg), block)){
2379 if(lcs < sblock.fs_contigsumsize) {
2381 cg_clustersum(&acg)[lcs]--;
2384 cg_clustersum(&acg)[lcs]++;
2391 /* *********************************************************** updrefs ***** */
2393 * This updates all references to relocated blocks for the given inode. The
2394 * inode is given as number within the cylinder group, and the number of the
2398 updrefs(int cg, ino_t in, struct gfs_bpp *bp, int fsi, int fso, unsigned int
2402 ufs_lbn_t len, lbn, numblks;
2403 ufs2_daddr_t iptr, blksperindir;
2405 int i, mode, inodeupdated;
2409 ino = ginode(in, fsi, cg);
2414 mode = DIP(ino, di_mode) & IFMT;
2415 if (mode != IFDIR && mode != IFREG && mode != IFLNK) {
2417 return; /* only check DIR, FILE, LINK */
2419 if (mode == IFLNK &&
2420 DIP(ino, di_size) < (u_int64_t) sblock.fs_maxsymlinklen) {
2422 return; /* skip short symlinks */
2424 numblks = howmany(DIP(ino, di_size), sblock.fs_bsize);
2427 return; /* skip empty file */
2429 if (DIP(ino, di_blocks) == 0) {
2431 return; /* skip empty swiss cheesy file or old fastlink */
2433 DBG_PRINT2("scg checking inode (%d in %d)\n",
2438 * Check all the blocks.
2441 len = numblks < NDADDR ? numblks : NDADDR;
2442 for (i = 0; i < len; i++) {
2443 iptr = DIP(ino, di_db[i]);
2446 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2447 DIP_SET(ino, di_db[i], iptr);
2451 DBG_PRINT0("~~scg direct blocks checked\n");
2454 len = numblks - NDADDR;
2456 for (i = 0; len > 0 && i < NIADDR; i++) {
2457 iptr = DIP(ino, di_ib[i]);
2460 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2461 DIP_SET(ino, di_ib[i], iptr);
2464 indirchk(blksperindir, lbn, iptr, numblks, bp, fsi, fso, Nflag);
2465 blksperindir *= NINDIR(&sblock);
2466 lbn += blksperindir;
2467 len -= blksperindir;
2468 DBG_PRINT1("scg indirect_%d blocks checked\n", i + 1);
2471 wtfs(inoblk, sblock.fs_bsize, inobuf, fso, Nflag);
2478 * Recursively check all the indirect blocks.
2481 indirchk(ufs_lbn_t blksperindir, ufs_lbn_t lbn, ufs2_daddr_t blkno,
2482 ufs_lbn_t lastlbn, struct gfs_bpp *bp, int fsi, int fso, unsigned int Nflag)
2484 DBG_FUNC("indirchk")
2491 /* read in the indirect block. */
2492 ibuf = malloc(sblock.fs_bsize);
2494 errx(1, "malloc failed");
2495 rdfs(fsbtodb(&sblock, blkno), (size_t)sblock.fs_bsize, ibuf, fsi);
2496 last = howmany(lastlbn - lbn, blksperindir) < NINDIR(&sblock) ?
2497 howmany(lastlbn - lbn, blksperindir) : NINDIR(&sblock);
2498 for (i = 0; i < last; i++) {
2499 if (sblock.fs_magic == FS_UFS1_MAGIC)
2500 iptr = ((ufs1_daddr_t *)ibuf)[i];
2502 iptr = ((ufs2_daddr_t *)ibuf)[i];
2505 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2506 if (sblock.fs_magic == FS_UFS1_MAGIC)
2507 ((ufs1_daddr_t *)ibuf)[i] = iptr;
2509 ((ufs2_daddr_t *)ibuf)[i] = iptr;
2511 if (blksperindir == 1)
2513 indirchk(blksperindir / NINDIR(&sblock), lbn + blksperindir * i,
2514 iptr, lastlbn, bp, fsi, fso, Nflag);