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, " %jd%s",
263 (intmax_t)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 * Note that we do not set cg_initediblk at all.
402 * In this extension of a previous filesystem
403 * we have no inodes initialized for the cylinder
404 * group at all. The first access to that cylinder
405 * group will do the correct initialization.
408 acg.cg_magic = CG_MAGIC;
410 acg.cg_niblk = sblock.fs_ipg;
411 acg.cg_ndblk = dmax - cbase;
412 if (sblock.fs_contigsumsize > 0)
413 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
414 start = &acg.cg_space[0] - (u_char *)(&acg.cg_firstfield);
415 if (sblock.fs_magic == FS_UFS2_MAGIC) {
416 acg.cg_iusedoff = start;
418 acg.cg_old_ncyl = sblock.fs_old_cpg;
419 acg.cg_old_time = acg.cg_time;
421 acg.cg_old_niblk = acg.cg_niblk;
423 acg.cg_old_btotoff = start;
424 acg.cg_old_boff = acg.cg_old_btotoff +
425 sblock.fs_old_cpg * sizeof(int32_t);
426 acg.cg_iusedoff = acg.cg_old_boff +
427 sblock.fs_old_cpg * sizeof(u_int16_t);
429 acg.cg_freeoff = acg.cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
430 acg.cg_nextfreeoff = acg.cg_freeoff + howmany(sblock.fs_fpg, CHAR_BIT);
431 if (sblock.fs_contigsumsize > 0) {
432 acg.cg_clustersumoff =
433 roundup(acg.cg_nextfreeoff, sizeof(u_int32_t));
434 acg.cg_clustersumoff -= sizeof(u_int32_t);
435 acg.cg_clusteroff = acg.cg_clustersumoff +
436 (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
437 acg.cg_nextfreeoff = acg.cg_clusteroff +
438 howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
440 if (acg.cg_nextfreeoff > sblock.fs_cgsize) {
442 * This should never happen as we would have had that panic
443 * already on file system creation
445 errx(37, "panic: cylinder group too big");
447 acg.cg_cs.cs_nifree += sblock.fs_ipg;
449 for (i = 0; i < ROOTINO; i++) {
450 setbit(cg_inosused(&acg), i);
451 acg.cg_cs.cs_nifree--;
454 * XXX Newfs writes out two blocks of initialized inodes
455 * unconditionally. Should we check here to make sure that they
456 * were actually written?
458 if (sblock.fs_magic == FS_UFS1_MAGIC) {
459 bzero(iobuf, sblock.fs_bsize);
460 for (i = 2 * sblock.fs_frag; i < sblock.fs_ipg / INOPF(&sblock);
461 i += sblock.fs_frag) {
462 dp1 = (struct ufs1_dinode *)iobuf;
463 dp2 = (struct ufs2_dinode *)iobuf;
465 for (j = 0; j < INOPB(&sblock); j++)
466 if (sblock.fs_magic == FS_UFS1_MAGIC) {
467 dp1->di_gen = random();
470 dp2->di_gen = random();
474 wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno) + i),
475 sblock.fs_bsize, iobuf, fso, Nflag);
480 * In cylno 0, beginning space is reserved
481 * for boot and super blocks.
483 for (d = 0; d < dlower; d += sblock.fs_frag) {
484 blkno = d / sblock.fs_frag;
485 setblock(&sblock, cg_blksfree(&acg), blkno);
486 if (sblock.fs_contigsumsize > 0)
487 setbit(cg_clustersfree(&acg), blkno);
488 acg.cg_cs.cs_nbfree++;
490 sblock.fs_dsize += dlower;
492 sblock.fs_dsize += acg.cg_ndblk - dupper;
493 if ((i = dupper % sblock.fs_frag)) {
494 acg.cg_frsum[sblock.fs_frag - i]++;
495 for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
496 setbit(cg_blksfree(&acg), dupper);
497 acg.cg_cs.cs_nffree++;
500 for (d = dupper; d + sblock.fs_frag <= acg.cg_ndblk;
501 d += sblock.fs_frag) {
502 blkno = d / sblock.fs_frag;
503 setblock(&sblock, cg_blksfree(&acg), blkno);
504 if (sblock.fs_contigsumsize > 0)
505 setbit(cg_clustersfree(&acg), blkno);
506 acg.cg_cs.cs_nbfree++;
508 if (d < acg.cg_ndblk) {
509 acg.cg_frsum[acg.cg_ndblk - d]++;
510 for (; d < acg.cg_ndblk; d++) {
511 setbit(cg_blksfree(&acg), d);
512 acg.cg_cs.cs_nffree++;
515 if (sblock.fs_contigsumsize > 0) {
516 int32_t *sump = cg_clustersum(&acg);
517 u_char *mapp = cg_clustersfree(&acg);
522 for (i = 0; i < acg.cg_nclusterblks; i++) {
523 if ((map & bit) != 0)
526 if (run > sblock.fs_contigsumsize)
527 run = sblock.fs_contigsumsize;
531 if ((i & (CHAR_BIT - 1)) != CHAR_BIT - 1)
539 if (run > sblock.fs_contigsumsize)
540 run = sblock.fs_contigsumsize;
544 sblock.fs_cstotal.cs_ndir += acg.cg_cs.cs_ndir;
545 sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
546 sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
547 sblock.fs_cstotal.cs_nifree += acg.cg_cs.cs_nifree;
549 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
550 sblock.fs_bsize, (char *)&acg, fso, Nflag);
559 /* ******************************************************* frag_adjust ***** */
561 * Here we add or subtract (sign +1/-1) the available fragments in a given
562 * block to or from the fragment statistics. By subtracting before and adding
563 * after an operation on the free frag map we can easy update the fragment
564 * statistic, which seems to be otherwise a rather complex operation.
567 frag_adjust(ufs2_daddr_t frag, int sign)
569 DBG_FUNC("frag_adjust")
577 * Here frag only needs to point to any fragment in the block we want
580 for(f=rounddown(frag, sblock.fs_frag);
581 f<roundup(frag+1, sblock.fs_frag);
584 * Count contiguous free fragments.
586 if(isset(cg_blksfree(&acg), f)) {
589 if(fragsize && fragsize<sblock.fs_frag) {
591 * We found something in between.
593 acg.cg_frsum[fragsize]+=sign;
594 DBG_PRINT2("frag_adjust [%d]+=%d\n",
601 if(fragsize && fragsize<sblock.fs_frag) {
603 * We found something.
605 acg.cg_frsum[fragsize]+=sign;
606 DBG_PRINT2("frag_adjust [%d]+=%d\n",
610 DBG_PRINT2("frag_adjust [[%d]]+=%d\n",
618 /* ******************************************************* cond_bl_upd ***** */
620 * Here we conditionally update a pointer to a fragment. We check for all
621 * relocated blocks if any of its fragments is referenced by the current
622 * field, and update the pointer to the respective fragment in our new
623 * block. If we find a reference we write back the block immediately,
624 * as there is no easy way for our general block reading engine to figure
625 * out if a write back operation is needed.
628 cond_bl_upd(ufs2_daddr_t *block, struct gfs_bpp *field, int fsi, int fso,
631 DBG_FUNC("cond_bl_upd")
633 ufs2_daddr_t src, dst;
639 for (f = field; f->old != 0; f++) {
641 if (fragstoblks(&sblock, src) != f->old)
644 * The fragment is part of the block, so update.
646 dst = blkstofrags(&sblock, f->new);
647 fragnum = fragnum(&sblock, src);
648 *block = dst + fragnum;
650 DBG_PRINT3("scg (%jd->%jd)[%d] reference updated\n",
656 * Copy the block back immediately.
658 * XXX If src is is from an indirect block we have
659 * to implement copy on write here in case of
662 ibuf = malloc(sblock.fs_bsize);
664 errx(1, "malloc failed");
666 rdfs(fsbtodb(&sblock, src), (size_t)sblock.fs_bsize, ibuf, fsi);
667 wtfs(dst, (size_t)sblock.fs_bsize, ibuf, fso, Nflag);
670 * The same block can't be found again in this loop.
679 /* ************************************************************ updjcg ***** */
681 * Here we do all needed work for the former last cylinder group. It has to be
682 * changed in any case, even if the file system ended exactly on the end of
683 * this group, as there is some slightly inconsistent handling of the number
684 * of cylinders in the cylinder group. We start again by reading the cylinder
685 * group from disk. If the last block was not fully available, we first handle
686 * the missing fragments, then we handle all new full blocks in that file
687 * system and finally we handle the new last fragmented block in the file
688 * system. We again have to handle the fragment statistics rotational layout
689 * tables and cluster summary during all those operations.
692 updjcg(int cylno, time_t utime, int fsi, int fso, unsigned int Nflag)
695 ufs2_daddr_t cbase, dmax, dupper;
703 * Read the former last (joining) cylinder group from disk, and make
706 rdfs(fsbtodb(&osblock, cgtod(&osblock, cylno)),
707 (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
708 DBG_PRINT0("jcg read\n");
713 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
716 * If the cylinder group had already its new final size almost
717 * nothing is to be done ... except:
718 * For some reason the value of cg_ncyl in the last cylinder group has
719 * to be zero instead of fs_cpg. As this is now no longer the last
720 * cylinder group we have to change that value now to fs_cpg.
723 if(cgbase(&osblock, cylno+1) == osblock.fs_size) {
724 if (sblock.fs_magic == FS_UFS1_MAGIC)
725 acg.cg_old_ncyl=sblock.fs_old_cpg;
727 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
728 (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
729 DBG_PRINT0("jcg written\n");
739 * Set up some variables needed later.
741 cbase = cgbase(&sblock, cylno);
742 dmax = cbase + sblock.fs_fpg;
743 if (dmax > sblock.fs_size)
744 dmax = sblock.fs_size;
745 dupper = cgdmin(&sblock, cylno) - cbase;
746 if (cylno == 0) { /* XXX fscs may be relocated */
747 dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
751 * Set pointer to the cylinder summary for our cylinder group.
756 * Touch the cylinder group, update all fields in the cylinder group as
757 * needed, update the free space in the superblock.
760 if (cylno == sblock.fs_ncg - 1) {
762 * This is still the last cylinder group.
764 if (sblock.fs_magic == FS_UFS1_MAGIC)
766 sblock.fs_old_ncyl % sblock.fs_old_cpg;
768 acg.cg_old_ncyl = sblock.fs_old_cpg;
770 DBG_PRINT2("jcg dbg: %d %u",
774 if (sblock.fs_magic == FS_UFS1_MAGIC)
780 acg.cg_ndblk = dmax - cbase;
781 sblock.fs_dsize += acg.cg_ndblk-aocg.cg_ndblk;
782 if (sblock.fs_contigsumsize > 0) {
783 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
787 * Now we have to update the free fragment bitmap for our new free
788 * space. There again we have to handle the fragmentation and also
789 * the rotational layout tables and the cluster summary. This is
790 * also done per fragment for the first new block if the old file
791 * system end was not on a block boundary, per fragment for the new
792 * last block if the new file system end is not on a block boundary,
793 * and per block for all space in between.
795 * Handle the first new block here if it was partially available
798 if(osblock.fs_size % sblock.fs_frag) {
799 if(roundup(osblock.fs_size, sblock.fs_frag)<=sblock.fs_size) {
801 * The new space is enough to fill at least this
805 for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag)-1;
806 i>=osblock.fs_size-cbase;
808 setbit(cg_blksfree(&acg), i);
809 acg.cg_cs.cs_nffree++;
814 * Check if the fragment just created could join an
815 * already existing fragment at the former end of the
818 if(isblock(&sblock, cg_blksfree(&acg),
819 ((osblock.fs_size - cgbase(&sblock, cylno))/
822 * The block is now completely available.
824 DBG_PRINT0("block was\n");
825 acg.cg_frsum[osblock.fs_size%sblock.fs_frag]--;
826 acg.cg_cs.cs_nbfree++;
827 acg.cg_cs.cs_nffree-=sblock.fs_frag;
828 k=rounddown(osblock.fs_size-cbase,
830 updclst((osblock.fs_size-cbase)/sblock.fs_frag);
833 * Lets rejoin a possible partially growed
837 while(isset(cg_blksfree(&acg), i) &&
838 (i>=rounddown(osblock.fs_size-cbase,
850 * We only grow by some fragments within this last
853 for(i=sblock.fs_size-cbase-1;
854 i>=osblock.fs_size-cbase;
856 setbit(cg_blksfree(&acg), i);
857 acg.cg_cs.cs_nffree++;
861 * Lets rejoin a possible partially growed fragment.
864 while(isset(cg_blksfree(&acg), i) &&
865 (i>=rounddown(osblock.fs_size-cbase,
878 * Handle all new complete blocks here.
880 for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag);
881 i+sblock.fs_frag<=dmax-cbase; /* XXX <= or only < ? */
883 j = i / sblock.fs_frag;
884 setblock(&sblock, cg_blksfree(&acg), j);
886 acg.cg_cs.cs_nbfree++;
890 * Handle the last new block if there are stll some new fragments left.
891 * Here we don't have to bother about the cluster summary or the even
892 * the rotational layout table.
894 if (i < (dmax - cbase)) {
895 acg.cg_frsum[dmax - cbase - i]++;
896 for (; i < dmax - cbase; i++) {
897 setbit(cg_blksfree(&acg), i);
898 acg.cg_cs.cs_nffree++;
902 sblock.fs_cstotal.cs_nffree +=
903 (acg.cg_cs.cs_nffree - aocg.cg_cs.cs_nffree);
904 sblock.fs_cstotal.cs_nbfree +=
905 (acg.cg_cs.cs_nbfree - aocg.cg_cs.cs_nbfree);
907 * The following statistics are not changed here:
908 * sblock.fs_cstotal.cs_ndir
909 * sblock.fs_cstotal.cs_nifree
910 * As the statistics for this cylinder group are ready, copy it to
911 * the summary information array.
916 * Write the updated "joining" cylinder group back to disk.
918 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize,
919 (void *)&acg, fso, Nflag);
920 DBG_PRINT0("jcg written\n");
929 /* ********************************************************** updcsloc ***** */
931 * Here we update the location of the cylinder summary. We have two possible
932 * ways of growing the cylinder summary.
933 * (1) We can try to grow the summary in the current location, and relocate
934 * possibly used blocks within the current cylinder group.
935 * (2) Alternatively we can relocate the whole cylinder summary to the first
936 * new completely empty cylinder group. Once the cylinder summary is no
937 * longer in the beginning of the first cylinder group you should never
938 * use a version of fsck which is not aware of the possibility to have
939 * this structure in a non standard place.
940 * Option (1) is considered to be less intrusive to the structure of the file-
941 * system. So we try to stick to that whenever possible. If there is not enough
942 * space in the cylinder group containing the cylinder summary we have to use
943 * method (2). In case of active snapshots in the file system we probably can
944 * completely avoid implementing copy on write if we stick to method (2) only.
947 updcsloc(time_t utime, int fsi, int fso, unsigned int Nflag)
953 ufs2_daddr_t cbase, dupper, odupper, d, f, g;
963 if(howmany(sblock.fs_cssize, sblock.fs_fsize) ==
964 howmany(osblock.fs_cssize, osblock.fs_fsize)) {
966 * No new fragment needed.
971 ocscg=dtog(&osblock, osblock.fs_csaddr);
973 blocks = 1+howmany(sblock.fs_cssize, sblock.fs_bsize)-
974 howmany(osblock.fs_cssize, osblock.fs_bsize);
977 * Read original cylinder group from disk, and make a copy.
978 * XXX If Nflag is set in some very rare cases we now miss
979 * some changes done in updjcg by reading the unmodified
982 rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)),
983 (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
984 DBG_PRINT0("oscg read\n");
989 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
992 * Touch the cylinder group, set up local variables needed later
993 * and update the superblock.
998 * XXX In the case of having active snapshots we may need much more
999 * blocks for the copy on write. We need each block twice, and
1000 * also up to 8*3 blocks for indirect blocks for all possible
1003 if(/*((int)sblock.fs_time&0x3)>0||*/ cs->cs_nbfree < blocks) {
1005 * There is not enough space in the old cylinder group to
1006 * relocate all blocks as needed, so we relocate the whole
1007 * cylinder group summary to a new group. We try to use the
1008 * first complete new cylinder group just created. Within the
1009 * cylinder group we align the area immediately after the
1010 * cylinder group information location in order to be as
1011 * close as possible to the original implementation of ffs.
1013 * First we have to make sure we'll find enough space in the
1014 * new cylinder group. If not, then we currently give up.
1015 * We start with freeing everything which was used by the
1016 * fragments of the old cylinder summary in the current group.
1017 * Now we write back the group meta data, read in the needed
1018 * meta data from the new cylinder group, and start allocating
1019 * within that group. Here we can assume, the group to be
1020 * completely empty. Which makes the handling of fragments and
1021 * clusters a lot easier.
1024 if(sblock.fs_ncg-osblock.fs_ncg < 2) {
1025 errx(2, "panic: not enough space");
1029 * Point "d" to the first fragment not used by the cylinder
1032 d=osblock.fs_csaddr+(osblock.fs_cssize/osblock.fs_fsize);
1035 * Set up last cluster size ("lcs") already here. Calculate
1036 * the size for the trailing cluster just behind where "d"
1039 if(sblock.fs_contigsumsize > 0) {
1040 for(block=howmany(d%sblock.fs_fpg, sblock.fs_frag),
1041 lcs=0; lcs<sblock.fs_contigsumsize;
1043 if(isclr(cg_clustersfree(&acg), block)){
1050 * Point "d" to the last frag used by the cylinder summary.
1054 DBG_PRINT1("d=%jd\n",
1056 if((d+1)%sblock.fs_frag) {
1058 * The end of the cylinder summary is not a complete
1062 frag_adjust(d%sblock.fs_fpg, -1);
1063 for(; (d+1)%sblock.fs_frag; d--) {
1064 DBG_PRINT1("d=%jd\n",
1066 setbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1067 acg.cg_cs.cs_nffree++;
1068 sblock.fs_cstotal.cs_nffree++;
1071 * Point "d" to the last fragment of the last
1072 * (incomplete) block of the cylinder summary.
1075 frag_adjust(d%sblock.fs_fpg, 1);
1077 if(isblock(&sblock, cg_blksfree(&acg),
1078 (d%sblock.fs_fpg)/sblock.fs_frag)) {
1079 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1080 acg.cg_cs.cs_nffree-=sblock.fs_frag;
1081 acg.cg_cs.cs_nbfree++;
1082 sblock.fs_cstotal.cs_nffree-=sblock.fs_frag;
1083 sblock.fs_cstotal.cs_nbfree++;
1084 if(sblock.fs_contigsumsize > 0) {
1085 setbit(cg_clustersfree(&acg),
1086 (d%sblock.fs_fpg)/sblock.fs_frag);
1087 if(lcs < sblock.fs_contigsumsize) {
1093 cg_clustersum(&acg)[lcs]++;
1098 * Point "d" to the first fragment of the block before
1099 * the last incomplete block.
1104 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1105 for(d=rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr;
1106 d-=sblock.fs_frag) {
1108 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1109 setblock(&sblock, cg_blksfree(&acg),
1110 (d%sblock.fs_fpg)/sblock.fs_frag);
1111 acg.cg_cs.cs_nbfree++;
1112 sblock.fs_cstotal.cs_nbfree++;
1113 if(sblock.fs_contigsumsize > 0) {
1114 setbit(cg_clustersfree(&acg),
1115 (d%sblock.fs_fpg)/sblock.fs_frag);
1117 * The last cluster size is already set up.
1119 if(lcs < sblock.fs_contigsumsize) {
1121 cg_clustersum(&acg)[lcs]--;
1124 cg_clustersum(&acg)[lcs]++;
1131 * Now write the former cylinder group containing the cylinder
1132 * summary back to disk.
1134 wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)),
1135 (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1136 DBG_PRINT0("oscg written\n");
1137 DBG_DUMP_CG(&sblock,
1142 * Find the beginning of the new cylinder group containing the
1145 sblock.fs_csaddr=cgdmin(&sblock, osblock.fs_ncg);
1146 ncscg=dtog(&sblock, sblock.fs_csaddr);
1151 * If Nflag is specified, we would now read random data instead
1152 * of an empty cg structure from disk. So we can't simulate that
1156 DBG_PRINT0("nscg update skipped\n");
1162 * Read the future cylinder group containing the cylinder
1163 * summary from disk, and make a copy.
1165 rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1166 (size_t)sblock.fs_cgsize, (void *)&aocg, fsi);
1167 DBG_PRINT0("nscg read\n");
1168 DBG_DUMP_CG(&sblock,
1172 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
1175 * Allocate all complete blocks used by the new cylinder
1178 for(d=sblock.fs_csaddr; d+sblock.fs_frag <=
1179 sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize);
1180 d+=sblock.fs_frag) {
1181 clrblock(&sblock, cg_blksfree(&acg),
1182 (d%sblock.fs_fpg)/sblock.fs_frag);
1183 acg.cg_cs.cs_nbfree--;
1184 sblock.fs_cstotal.cs_nbfree--;
1185 if(sblock.fs_contigsumsize > 0) {
1186 clrbit(cg_clustersfree(&acg),
1187 (d%sblock.fs_fpg)/sblock.fs_frag);
1192 * Allocate all fragments used by the cylinder summary in the
1195 if(d<sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize)) {
1196 for(; d-sblock.fs_csaddr<
1197 sblock.fs_cssize/sblock.fs_fsize;
1199 clrbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1200 acg.cg_cs.cs_nffree--;
1201 sblock.fs_cstotal.cs_nffree--;
1203 acg.cg_cs.cs_nbfree--;
1204 acg.cg_cs.cs_nffree+=sblock.fs_frag;
1205 sblock.fs_cstotal.cs_nbfree--;
1206 sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1207 if(sblock.fs_contigsumsize > 0) {
1208 clrbit(cg_clustersfree(&acg),
1209 (d%sblock.fs_fpg)/sblock.fs_frag);
1212 frag_adjust(d%sblock.fs_fpg, +1);
1215 * XXX Handle the cluster statistics here in the case this
1216 * cylinder group is now almost full, and the remaining
1217 * space is less then the maximum cluster size. This is
1218 * probably not needed, as you would hardly find a file
1219 * system which has only MAXCSBUFS+FS_MAXCONTIG of free
1220 * space right behind the cylinder group information in
1221 * any new cylinder group.
1225 * Update our statistics in the cylinder summary.
1230 * Write the new cylinder group containing the cylinder summary
1233 wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1234 (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1235 DBG_PRINT0("nscg written\n");
1236 DBG_DUMP_CG(&sblock,
1244 * We have got enough of space in the current cylinder group, so we
1245 * can relocate just a few blocks, and let the summary information
1246 * grow in place where it is right now.
1250 cbase = cgbase(&osblock, ocscg); /* old and new are equal */
1251 dupper = sblock.fs_csaddr - cbase +
1252 howmany(sblock.fs_cssize, sblock.fs_fsize);
1253 odupper = osblock.fs_csaddr - cbase +
1254 howmany(osblock.fs_cssize, osblock.fs_fsize);
1256 sblock.fs_dsize -= dupper-odupper;
1259 * Allocate the space for the array of blocks to be relocated.
1261 bp=(struct gfs_bpp *)malloc(((dupper-odupper)/sblock.fs_frag+2)*
1262 sizeof(struct gfs_bpp));
1264 errx(1, "malloc failed");
1266 memset((char *)bp, 0, ((dupper-odupper)/sblock.fs_frag+2)*
1267 sizeof(struct gfs_bpp));
1270 * Lock all new frags needed for the cylinder group summary. This is
1271 * done per fragment in the first and last block of the new required
1272 * area, and per block for all other blocks.
1274 * Handle the first new block here (but only if some fragments where
1275 * already used for the cylinder summary).
1278 frag_adjust(odupper, -1);
1279 for(d=odupper; ((d<dupper)&&(d%sblock.fs_frag)); d++) {
1280 DBG_PRINT1("scg first frag check loop d=%jd\n",
1282 if(isclr(cg_blksfree(&acg), d)) {
1284 bp[ind].old=d/sblock.fs_frag;
1285 bp[ind].flags|=GFS_FL_FIRST;
1286 if(roundup(d, sblock.fs_frag) >= dupper) {
1287 bp[ind].flags|=GFS_FL_LAST;
1292 clrbit(cg_blksfree(&acg), d);
1293 acg.cg_cs.cs_nffree--;
1294 sblock.fs_cstotal.cs_nffree--;
1297 * No cluster handling is needed here, as there was at least
1298 * one fragment in use by the cylinder summary in the old
1300 * No block-free counter handling here as this block was not
1304 frag_adjust(odupper, 1);
1307 * Handle all needed complete blocks here.
1309 for(; d+sblock.fs_frag<=dupper; d+=sblock.fs_frag) {
1310 DBG_PRINT1("scg block check loop d=%jd\n",
1312 if(!isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1313 for(f=d; f<d+sblock.fs_frag; f++) {
1314 if(isset(cg_blksfree(&aocg), f)) {
1315 acg.cg_cs.cs_nffree--;
1316 sblock.fs_cstotal.cs_nffree--;
1319 clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1320 bp[ind].old=d/sblock.fs_frag;
1323 clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1324 acg.cg_cs.cs_nbfree--;
1325 sblock.fs_cstotal.cs_nbfree--;
1326 if(sblock.fs_contigsumsize > 0) {
1327 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1328 for(lcs=0, l=(d/sblock.fs_frag)+1;
1329 lcs<sblock.fs_contigsumsize;
1331 if(isclr(cg_clustersfree(&acg),l)){
1335 if(lcs < sblock.fs_contigsumsize) {
1336 cg_clustersum(&acg)[lcs+1]--;
1338 cg_clustersum(&acg)[lcs]++;
1344 * No fragment counter handling is needed here, as this finally
1345 * doesn't change after the relocation.
1350 * Handle all fragments needed in the last new affected block.
1353 frag_adjust(dupper-1, -1);
1355 if(isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1356 acg.cg_cs.cs_nbfree--;
1357 sblock.fs_cstotal.cs_nbfree--;
1358 acg.cg_cs.cs_nffree+=sblock.fs_frag;
1359 sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1360 if(sblock.fs_contigsumsize > 0) {
1361 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1362 for(lcs=0, l=(d/sblock.fs_frag)+1;
1363 lcs<sblock.fs_contigsumsize;
1365 if(isclr(cg_clustersfree(&acg),l)){
1369 if(lcs < sblock.fs_contigsumsize) {
1370 cg_clustersum(&acg)[lcs+1]--;
1372 cg_clustersum(&acg)[lcs]++;
1378 for(; d<dupper; d++) {
1379 DBG_PRINT1("scg second frag check loop d=%jd\n",
1381 if(isclr(cg_blksfree(&acg), d)) {
1382 bp[ind].old=d/sblock.fs_frag;
1383 bp[ind].flags|=GFS_FL_LAST;
1385 clrbit(cg_blksfree(&acg), d);
1386 acg.cg_cs.cs_nffree--;
1387 sblock.fs_cstotal.cs_nffree--;
1390 if(bp[ind].flags & GFS_FL_LAST) { /* we have to advance here */
1393 frag_adjust(dupper-1, 1);
1397 * If we found a block to relocate just do so.
1400 for(i=0; i<ind; i++) {
1401 if(!bp[i].old) { /* no more blocks listed */
1403 * XXX A relative blocknumber should not be
1404 * zero, which is not explicitly
1405 * guaranteed by our code.
1410 * Allocate a complete block in the same (current)
1413 bp[i].new=alloc()/sblock.fs_frag;
1416 * There is no frag_adjust() needed for the new block
1417 * as it will have no fragments yet :-).
1419 for(f=bp[i].old*sblock.fs_frag,
1420 g=bp[i].new*sblock.fs_frag;
1421 f<(bp[i].old+1)*sblock.fs_frag;
1423 if(isset(cg_blksfree(&aocg), f)) {
1424 setbit(cg_blksfree(&acg), g);
1425 acg.cg_cs.cs_nffree++;
1426 sblock.fs_cstotal.cs_nffree++;
1431 * Special handling is required if this was the first
1432 * block. We have to consider the fragments which were
1433 * used by the cylinder summary in the original block
1434 * which re to be free in the copy of our block. We
1435 * have to be careful if this first block happens to
1436 * be also the last block to be relocated.
1438 if(bp[i].flags & GFS_FL_FIRST) {
1439 for(f=bp[i].old*sblock.fs_frag,
1440 g=bp[i].new*sblock.fs_frag;
1443 setbit(cg_blksfree(&acg), g);
1444 acg.cg_cs.cs_nffree++;
1445 sblock.fs_cstotal.cs_nffree++;
1447 if(!(bp[i].flags & GFS_FL_LAST)) {
1448 frag_adjust(bp[i].new*sblock.fs_frag,1);
1453 * Special handling is required if this is the last
1454 * block to be relocated.
1456 if(bp[i].flags & GFS_FL_LAST) {
1457 frag_adjust(bp[i].new*sblock.fs_frag, 1);
1458 frag_adjust(bp[i].old*sblock.fs_frag, -1);
1460 f<roundup(dupper, sblock.fs_frag);
1462 if(isclr(cg_blksfree(&acg), f)) {
1463 setbit(cg_blksfree(&acg), f);
1464 acg.cg_cs.cs_nffree++;
1465 sblock.fs_cstotal.cs_nffree++;
1468 frag_adjust(bp[i].old*sblock.fs_frag, 1);
1472 * !!! Attach the cylindergroup offset here.
1474 bp[i].old+=cbase/sblock.fs_frag;
1475 bp[i].new+=cbase/sblock.fs_frag;
1478 * Copy the content of the block.
1481 * XXX Here we will have to implement a copy on write
1482 * in the case we have any active snapshots.
1484 rdfs(fsbtodb(&sblock, bp[i].old*sblock.fs_frag),
1485 (size_t)sblock.fs_bsize, (void *)&ablk, fsi);
1486 wtfs(fsbtodb(&sblock, bp[i].new*sblock.fs_frag),
1487 (size_t)sblock.fs_bsize, (void *)&ablk, fso, Nflag);
1488 DBG_DUMP_HEX(&sblock,
1489 "copied full block",
1490 (unsigned char *)&ablk);
1492 DBG_PRINT2("scg (%jd->%jd) block relocated\n",
1493 (intmax_t)bp[i].old,
1494 (intmax_t)bp[i].new);
1498 * Now we have to update all references to any fragment which
1499 * belongs to any block relocated. We iterate now over all
1500 * cylinder groups, within those over all non zero length
1503 for(cylno=0; cylno<osblock.fs_ncg; cylno++) {
1504 DBG_PRINT1("scg doing cg (%d)\n",
1506 for(inc=osblock.fs_ipg-1 ; inc>0 ; inc--) {
1507 updrefs(cylno, (ino_t)inc, bp, fsi, fso, Nflag);
1512 * All inodes are checked, now make sure the number of
1513 * references found make sense.
1515 for(i=0; i<ind; i++) {
1516 if(!bp[i].found || (bp[i].found>sblock.fs_frag)) {
1517 warnx("error: %jd refs found for block %jd.",
1518 (intmax_t)bp[i].found, (intmax_t)bp[i].old);
1524 * The following statistics are not changed here:
1525 * sblock.fs_cstotal.cs_ndir
1526 * sblock.fs_cstotal.cs_nifree
1527 * The following statistics were already updated on the fly:
1528 * sblock.fs_cstotal.cs_nffree
1529 * sblock.fs_cstotal.cs_nbfree
1530 * As the statistics for this cylinder group are ready, copy it to
1531 * the summary information array.
1537 * Write summary cylinder group back to disk.
1539 wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)), (size_t)sblock.fs_cgsize,
1540 (void *)&acg, fso, Nflag);
1541 DBG_PRINT0("scg written\n");
1542 DBG_DUMP_CG(&sblock,
1550 /* ************************************************************** rdfs ***** */
1552 * Here we read some block(s) from disk.
1555 rdfs(ufs2_daddr_t bno, size_t size, void *bf, int fsi)
1563 err(32, "rdfs: attempting to read negative block number");
1565 if (lseek(fsi, (off_t)bno * DEV_BSIZE, 0) < 0) {
1566 err(33, "rdfs: seek error: %jd", (intmax_t)bno);
1568 n = read(fsi, bf, size);
1569 if (n != (ssize_t)size) {
1570 err(34, "rdfs: read error: %jd", (intmax_t)bno);
1577 /* ************************************************************** wtfs ***** */
1579 * Here we write some block(s) to disk.
1582 wtfs(ufs2_daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag)
1593 if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0) {
1594 err(35, "wtfs: seek error: %ld", (long)bno);
1596 n = write(fso, bf, size);
1597 if (n != (ssize_t)size) {
1598 err(36, "wtfs: write error: %ld", (long)bno);
1605 /* ************************************************************* alloc ***** */
1607 * Here we allocate a free block in the current cylinder group. It is assumed,
1608 * that acg contains the current cylinder group. As we may take a block from
1609 * somewhere in the file system we have to handle cluster summary here.
1615 ufs2_daddr_t d, blkno;
1619 int dlower, dupper, dmax;
1623 if (acg.cg_magic != CG_MAGIC) {
1624 warnx("acg: bad magic number");
1628 if (acg.cg_cs.cs_nbfree == 0) {
1629 warnx("error: cylinder group ran out of space");
1634 * We start seeking for free blocks only from the space available after
1635 * the end of the new grown cylinder summary. Otherwise we allocate a
1636 * block here which we have to relocate a couple of seconds later again
1637 * again, and we are not prepared to to this anyway.
1640 dlower=cgsblock(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1641 dupper=cgdmin(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1642 dmax=cgbase(&sblock, acg.cg_cgx)+sblock.fs_fpg;
1643 if (dmax > sblock.fs_size) {
1644 dmax = sblock.fs_size;
1646 dmax-=cgbase(&sblock, acg.cg_cgx); /* retransform into cg */
1647 csmin=sblock.fs_csaddr-cgbase(&sblock, acg.cg_cgx);
1648 csmax=csmin+howmany(sblock.fs_cssize, sblock.fs_fsize);
1649 DBG_PRINT3("seek range: dl=%d, du=%d, dm=%d\n",
1653 DBG_PRINT2("range cont: csmin=%d, csmax=%d\n",
1657 for(d=0; (d<dlower && blkno==-1); d+=sblock.fs_frag) {
1658 if(d>=csmin && d<=csmax) {
1661 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1663 blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1667 for(d=dupper; (d<dmax && blkno==-1); d+=sblock.fs_frag) {
1668 if(d>=csmin && d<=csmax) {
1671 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1673 blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1678 warnx("internal error: couldn't find promised block in cg");
1684 * This is needed if the block was found already in the first loop.
1686 d=blkstofrags(&sblock, blkno);
1688 clrblock(&sblock, cg_blksfree(&acg), blkno);
1689 if (sblock.fs_contigsumsize > 0) {
1691 * Handle the cluster allocation bitmap.
1693 clrbit(cg_clustersfree(&acg), blkno);
1695 * We possibly have split a cluster here, so we have to do
1696 * recalculate the sizes of the remaining cluster halves now,
1697 * and use them for updating the cluster summary information.
1699 * Lets start with the blocks before our allocated block ...
1701 for(lcs1=0, l=blkno-1; lcs1<sblock.fs_contigsumsize;
1703 if(isclr(cg_clustersfree(&acg),l)){
1708 * ... and continue with the blocks right after our allocated
1711 for(lcs2=0, l=blkno+1; lcs2<sblock.fs_contigsumsize;
1713 if(isclr(cg_clustersfree(&acg),l)){
1719 * Now update all counters.
1721 cg_clustersum(&acg)[MIN(lcs1+lcs2+1,sblock.fs_contigsumsize)]--;
1723 cg_clustersum(&acg)[lcs1]++;
1726 cg_clustersum(&acg)[lcs2]++;
1730 * Update all statistics based on blocks.
1732 acg.cg_cs.cs_nbfree--;
1733 sblock.fs_cstotal.cs_nbfree--;
1739 /* *********************************************************** isblock ***** */
1741 * Here we check if all frags of a block are free. For more details again
1742 * please see the source of newfs(8), as this function is taken over almost
1746 isblock(struct fs *fs, unsigned char *cp, int h)
1753 switch (fs->fs_frag) {
1756 return (cp[h] == 0xff);
1758 mask = 0x0f << ((h & 0x1) << 2);
1760 return ((cp[h >> 1] & mask) == mask);
1762 mask = 0x03 << ((h & 0x3) << 1);
1764 return ((cp[h >> 2] & mask) == mask);
1766 mask = 0x01 << (h & 0x7);
1768 return ((cp[h >> 3] & mask) == mask);
1770 fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
1776 /* ********************************************************** clrblock ***** */
1778 * Here we allocate a complete block in the block map. For more details again
1779 * please see the source of newfs(8), as this function is taken over almost
1783 clrblock(struct fs *fs, unsigned char *cp, int h)
1785 DBG_FUNC("clrblock")
1789 switch ((fs)->fs_frag) {
1794 cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1797 cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1800 cp[h >> 3] &= ~(0x01 << (h & 0x7));
1803 warnx("clrblock bad fs_frag %d", fs->fs_frag);
1811 /* ********************************************************** setblock ***** */
1813 * Here we free a complete block in the free block map. For more details again
1814 * please see the source of newfs(8), as this function is taken over almost
1818 setblock(struct fs *fs, unsigned char *cp, int h)
1820 DBG_FUNC("setblock")
1824 switch (fs->fs_frag) {
1829 cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1832 cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1835 cp[h >> 3] |= (0x01 << (h & 0x7));
1838 warnx("setblock bad fs_frag %d", fs->fs_frag);
1846 /* ************************************************************ ginode ***** */
1848 * This function provides access to an individual inode. We find out in which
1849 * block the requested inode is located, read it from disk if needed, and
1850 * return the pointer into that block. We maintain a cache of one block to
1851 * not read the same block again and again if we iterate linearly over all
1854 static union dinode *
1855 ginode(ino_t inumber, int fsi, int cg)
1858 static ino_t startinum = 0; /* first inode in cached block */
1863 * The inumber passed in is relative to the cg, so use it here to see
1864 * if the inode has been allocated yet.
1866 if (isclr(cg_inosused(&aocg), inumber)) {
1871 * Now make the inumber relative to the entire inode space so it can
1872 * be sanity checked.
1874 inumber += (cg * sblock.fs_ipg);
1875 if (inumber < ROOTINO) {
1879 if (inumber > maxino)
1880 errx(8, "bad inode number %d to ginode", inumber);
1881 if (startinum == 0 ||
1882 inumber < startinum || inumber >= startinum + INOPB(&sblock)) {
1883 inoblk = fsbtodb(&sblock, ino_to_fsba(&sblock, inumber));
1884 rdfs(inoblk, (size_t)sblock.fs_bsize, inobuf, fsi);
1885 startinum = (inumber / INOPB(&sblock)) * INOPB(&sblock);
1888 if (sblock.fs_magic == FS_UFS1_MAGIC)
1889 return (union dinode *)((uintptr_t)inobuf +
1890 (inumber % INOPB(&sblock)) * sizeof(struct ufs1_dinode));
1891 return (union dinode *)((uintptr_t)inobuf +
1892 (inumber % INOPB(&sblock)) * sizeof(struct ufs2_dinode));
1895 /* ****************************************************** charsperline ***** */
1897 * Figure out how many lines our current terminal has. For more details again
1898 * please see the source of newfs(8), as this function is taken over almost
1904 DBG_FUNC("charsperline")
1912 if (ioctl(0, TIOCGWINSZ, &ws) != -1) {
1913 columns = ws.ws_col;
1915 if (columns == 0 && (cp = getenv("COLUMNS"))) {
1919 columns = 80; /* last resort */
1926 /* ****************************************************** get_dev_size ***** */
1928 * Get the size of the partition if we can't figure it out from the disklabel,
1929 * e.g. from vinum volumes.
1932 get_dev_size(int fd, int *size)
1937 if (ioctl(fd, DIOCGSECTORSIZE, §orsize) == -1)
1938 err(1,"DIOCGSECTORSIZE");
1939 if (ioctl(fd, DIOCGMEDIASIZE, &mediasize) == -1)
1940 err(1,"DIOCGMEDIASIZE");
1942 if (sectorsize <= 0)
1943 errx(1, "bogus sectorsize: %d", sectorsize);
1945 *size = mediasize / sectorsize;
1948 /* ************************************************************** main ***** */
1950 * growfs(8) is a utility which allows to increase the size of an existing
1951 * ufs file system. Currently this can only be done on unmounted file system.
1952 * It recognizes some command line options to specify the new desired size,
1953 * and it does some basic checkings. The old file system size is determined
1954 * and after some more checks like we can really access the new last block
1955 * on the disk etc. we calculate the new parameters for the superblock. After
1956 * having done this we just call growfs() which will do the work. Before
1957 * we finish the only thing left is to update the disklabel.
1958 * We still have to provide support for snapshots. Therefore we first have to
1959 * understand what data structures are always replicated in the snapshot on
1960 * creation, for all other blocks we touch during our procedure, we have to
1961 * keep the old blocks unchanged somewhere available for the snapshots. If we
1962 * are lucky, then we only have to handle our blocks to be relocated in that
1964 * Also we have to consider in what order we actually update the critical
1965 * data structures of the file system to make sure, that in case of a disaster
1966 * fsck(8) is still able to restore any lost data.
1967 * The foreseen last step then will be to provide for growing even mounted
1968 * file systems. There we have to extend the mount() system call to provide
1969 * userland access to the file system locking facility.
1972 main(int argc, char **argv)
1975 char *device, *special, *cp;
1977 unsigned int size=0;
1979 unsigned int Nflag=0;
1982 struct disklabel *lp;
1983 struct partition *pp;
1989 #endif /* FSMAXSNAP */
1993 while((ch=getopt(argc, argv, "Ns:vy")) != -1) {
1999 size=(size_t)atol(optarg);
2004 case 'v': /* for compatibility to newfs */
2024 * Now try to guess the (raw)device name.
2026 if (0 == strrchr(device, '/')) {
2028 * No path prefix was given, so try in that order:
2034 * FreeBSD now doesn't distinguish between raw and block
2035 * devices any longer, but it should still work this way.
2037 len=strlen(device)+strlen(_PATH_DEV)+2+strlen("vinum/");
2038 special=(char *)malloc(len);
2039 if(special == NULL) {
2040 errx(1, "malloc failed");
2042 snprintf(special, len, "%sr%s", _PATH_DEV, device);
2043 if (stat(special, &st) == -1) {
2044 snprintf(special, len, "%s%s", _PATH_DEV, device);
2045 if (stat(special, &st) == -1) {
2046 snprintf(special, len, "%svinum/r%s",
2048 if (stat(special, &st) == -1) {
2049 /* For now this is the 'last resort' */
2050 snprintf(special, len, "%svinum/%s",
2059 * Try to access our devices for writing ...
2064 fso = open(device, O_WRONLY);
2066 err(1, "%s", device);
2073 fsi = open(device, O_RDONLY);
2075 err(1, "%s", device);
2079 * Try to read a label and guess the slice if not specified. This
2080 * code should guess the right thing and avoid to bother the user
2081 * with the task of specifying the option -v on vinum volumes.
2083 cp=device+strlen(device)-1;
2084 lp = get_disklabel(fsi);
2088 pp = &lp->d_partitions[2];
2089 } else if (*cp>='a' && *cp<='h') {
2090 pp = &lp->d_partitions[*cp - 'a'];
2092 errx(1, "unknown device");
2094 p_size = pp->p_size;
2096 get_dev_size(fsi, &p_size);
2100 * Check if that partition is suitable for growing a file system.
2103 errx(1, "partition is unavailable");
2107 * Read the current superblock, and take a backup.
2109 for (i = 0; sblock_try[i] != -1; i++) {
2110 sblockloc = sblock_try[i] / DEV_BSIZE;
2111 rdfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&(osblock), fsi);
2112 if ((osblock.fs_magic == FS_UFS1_MAGIC ||
2113 (osblock.fs_magic == FS_UFS2_MAGIC &&
2114 osblock.fs_sblockloc == sblock_try[i])) &&
2115 osblock.fs_bsize <= MAXBSIZE &&
2116 osblock.fs_bsize >= (int32_t) sizeof(struct fs))
2119 if (sblock_try[i] == -1) {
2120 errx(1, "superblock not recognized");
2122 memcpy((void *)&fsun1, (void *)&fsun2, sizeof(fsun2));
2123 maxino = sblock.fs_ncg * sblock.fs_ipg;
2125 DBG_OPEN("/tmp/growfs.debug"); /* already here we need a superblock */
2126 DBG_DUMP_FS(&sblock,
2130 * Determine size to grow to. Default to the full size specified in
2133 sblock.fs_size = dbtofsb(&osblock, p_size);
2136 errx(1, "there is not enough space (%d < %d)",
2139 sblock.fs_size = dbtofsb(&osblock, size);
2143 * Are we really growing ?
2145 if(osblock.fs_size >= sblock.fs_size) {
2146 errx(1, "we are not growing (%jd->%jd)",
2147 (intmax_t)osblock.fs_size, (intmax_t)sblock.fs_size);
2153 * Check if we find an active snapshot.
2155 if(ExpertFlag == 0) {
2156 for(j=0; j<FSMAXSNAP; j++) {
2157 if(sblock.fs_snapinum[j]) {
2158 errx(1, "active snapshot found in file system\n"
2159 " please remove all snapshots before "
2162 if(!sblock.fs_snapinum[j]) { /* list is dense */
2169 if (ExpertFlag == 0 && Nflag == 0) {
2170 printf("We strongly recommend you to make a backup "
2171 "before growing the Filesystem\n\n"
2172 " Did you backup your data (Yes/No) ? ");
2173 fgets(reply, (int)sizeof(reply), stdin);
2174 if (strcmp(reply, "Yes\n")){
2175 printf("\n Nothing done \n");
2180 printf("new file systemsize is: %jd frags\n", (intmax_t)sblock.fs_size);
2183 * Try to access our new last block in the file system. Even if we
2184 * later on realize we have to abort our operation, on that block
2185 * there should be no data, so we can't destroy something yet.
2187 wtfs((ufs2_daddr_t)p_size-1, (size_t)DEV_BSIZE, (void *)&sblock,
2191 * Now calculate new superblock values and check for reasonable
2192 * bound for new file system size:
2193 * fs_size: is derived from label or user input
2194 * fs_dsize: should get updated in the routines creating or
2195 * updating the cylinder groups on the fly
2196 * fs_cstotal: should get updated in the routines creating or
2197 * updating the cylinder groups
2201 * Update the number of cylinders and cylinder groups in the file system.
2203 if (sblock.fs_magic == FS_UFS1_MAGIC) {
2204 sblock.fs_old_ncyl =
2205 sblock.fs_size * sblock.fs_old_nspf / sblock.fs_old_spc;
2206 if (sblock.fs_size * sblock.fs_old_nspf >
2207 sblock.fs_old_ncyl * sblock.fs_old_spc)
2208 sblock.fs_old_ncyl++;
2210 sblock.fs_ncg = howmany(sblock.fs_size, sblock.fs_fpg);
2211 maxino = sblock.fs_ncg * sblock.fs_ipg;
2213 if (sblock.fs_size % sblock.fs_fpg != 0 &&
2214 sblock.fs_size % sblock.fs_fpg < cgdmin(&sblock, sblock.fs_ncg)) {
2216 * The space in the new last cylinder group is too small,
2220 if (sblock.fs_magic == FS_UFS1_MAGIC)
2221 sblock.fs_old_ncyl = sblock.fs_ncg * sblock.fs_old_cpg;
2222 printf("Warning: %jd sector(s) cannot be allocated.\n",
2223 (intmax_t)fsbtodb(&sblock, sblock.fs_size % sblock.fs_fpg));
2224 sblock.fs_size = sblock.fs_ncg * sblock.fs_fpg;
2225 maxino -= sblock.fs_ipg;
2229 * Update the space for the cylinder group summary information in the
2230 * respective cylinder group data area.
2233 fragroundup(&sblock, sblock.fs_ncg * sizeof(struct csum));
2235 if(osblock.fs_size >= sblock.fs_size) {
2236 errx(1, "not enough new space");
2239 DBG_PRINT0("sblock calculated\n");
2242 * Ok, everything prepared, so now let's do the tricks.
2244 growfs(fsi, fso, Nflag);
2247 * Update the disk label.
2250 pp->p_fsize = sblock.fs_fsize;
2251 pp->p_frag = sblock.fs_frag;
2252 pp->p_cpg = sblock.fs_fpg;
2254 return_disklabel(fso, lp, Nflag);
2255 DBG_PRINT0("label rewritten\n");
2259 if(fso>-1) close(fso);
2267 /* ************************************************** return_disklabel ***** */
2269 * Write the updated disklabel back to disk.
2272 return_disklabel(int fd, struct disklabel *lp, unsigned int Nflag)
2274 DBG_FUNC("return_disklabel")
2290 * recalculate checksum
2292 while(ptr < (u_short *)&lp->d_partitions[lp->d_npartitions]) {
2297 if (ioctl(fd, DIOCWDINFO, (char *)lp) < 0) {
2298 errx(1, "DIOCWDINFO failed");
2307 /* ***************************************************** get_disklabel ***** */
2309 * Read the disklabel from disk.
2311 static struct disklabel *
2312 get_disklabel(int fd)
2314 DBG_FUNC("get_disklabel")
2315 static struct disklabel *lab;
2319 lab=(struct disklabel *)malloc(sizeof(struct disklabel));
2321 errx(1, "malloc failed");
2323 if (!ioctl(fd, DIOCGDINFO, (char *)lab))
2333 /* ************************************************************* usage ***** */
2335 * Dump a line of usage.
2344 fprintf(stderr, "usage: growfs [-Ny] [-s size] special\n");
2350 /* *********************************************************** updclst ***** */
2352 * This updates most parameters and the bitmap related to cluster. We have to
2353 * assume that sblock, osblock, acg are set up.
2363 if(sblock.fs_contigsumsize < 1) { /* no clustering */
2367 * update cluster allocation map
2369 setbit(cg_clustersfree(&acg), block);
2372 * update cluster summary table
2376 * calculate size for the trailing cluster
2378 for(block--; lcs<sblock.fs_contigsumsize; block--, lcs++ ) {
2379 if(isclr(cg_clustersfree(&acg), block)){
2384 if(lcs < sblock.fs_contigsumsize) {
2386 cg_clustersum(&acg)[lcs]--;
2389 cg_clustersum(&acg)[lcs]++;
2396 /* *********************************************************** updrefs ***** */
2398 * This updates all references to relocated blocks for the given inode. The
2399 * inode is given as number within the cylinder group, and the number of the
2403 updrefs(int cg, ino_t in, struct gfs_bpp *bp, int fsi, int fso, unsigned int
2407 ufs_lbn_t len, lbn, numblks;
2408 ufs2_daddr_t iptr, blksperindir;
2410 int i, mode, inodeupdated;
2414 ino = ginode(in, fsi, cg);
2419 mode = DIP(ino, di_mode) & IFMT;
2420 if (mode != IFDIR && mode != IFREG && mode != IFLNK) {
2422 return; /* only check DIR, FILE, LINK */
2424 if (mode == IFLNK &&
2425 DIP(ino, di_size) < (u_int64_t) sblock.fs_maxsymlinklen) {
2427 return; /* skip short symlinks */
2429 numblks = howmany(DIP(ino, di_size), sblock.fs_bsize);
2432 return; /* skip empty file */
2434 if (DIP(ino, di_blocks) == 0) {
2436 return; /* skip empty swiss cheesy file or old fastlink */
2438 DBG_PRINT2("scg checking inode (%d in %d)\n",
2443 * Check all the blocks.
2446 len = numblks < NDADDR ? numblks : NDADDR;
2447 for (i = 0; i < len; i++) {
2448 iptr = DIP(ino, di_db[i]);
2451 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2452 DIP_SET(ino, di_db[i], iptr);
2456 DBG_PRINT0("~~scg direct blocks checked\n");
2459 len = numblks - NDADDR;
2461 for (i = 0; len > 0 && i < NIADDR; i++) {
2462 iptr = DIP(ino, di_ib[i]);
2465 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2466 DIP_SET(ino, di_ib[i], iptr);
2469 indirchk(blksperindir, lbn, iptr, numblks, bp, fsi, fso, Nflag);
2470 blksperindir *= NINDIR(&sblock);
2471 lbn += blksperindir;
2472 len -= blksperindir;
2473 DBG_PRINT1("scg indirect_%d blocks checked\n", i + 1);
2476 wtfs(inoblk, sblock.fs_bsize, inobuf, fso, Nflag);
2483 * Recursively check all the indirect blocks.
2486 indirchk(ufs_lbn_t blksperindir, ufs_lbn_t lbn, ufs2_daddr_t blkno,
2487 ufs_lbn_t lastlbn, struct gfs_bpp *bp, int fsi, int fso, unsigned int Nflag)
2489 DBG_FUNC("indirchk")
2496 /* read in the indirect block. */
2497 ibuf = malloc(sblock.fs_bsize);
2499 errx(1, "malloc failed");
2500 rdfs(fsbtodb(&sblock, blkno), (size_t)sblock.fs_bsize, ibuf, fsi);
2501 last = howmany(lastlbn - lbn, blksperindir) < NINDIR(&sblock) ?
2502 howmany(lastlbn - lbn, blksperindir) : NINDIR(&sblock);
2503 for (i = 0; i < last; i++) {
2504 if (sblock.fs_magic == FS_UFS1_MAGIC)
2505 iptr = ((ufs1_daddr_t *)ibuf)[i];
2507 iptr = ((ufs2_daddr_t *)ibuf)[i];
2510 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2511 if (sblock.fs_magic == FS_UFS1_MAGIC)
2512 ((ufs1_daddr_t *)ibuf)[i] = iptr;
2514 ((ufs2_daddr_t *)ibuf)[i] = iptr;
2516 if (blksperindir == 1)
2518 indirchk(blksperindir / NINDIR(&sblock), lbn + blksperindir * i,
2519 iptr, lastlbn, bp, fsi, fso, Nflag);