2 * Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz
3 * Copyright (c) 1980, 1989, 1993 The Regents of the University of California.
6 * This code is derived from software contributed to Berkeley by
7 * Christoph Herrmann and Thomas-Henning von Kamptz, Munich and Frankfurt.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgment:
19 * This product includes software developed by the University of
20 * California, Berkeley and its contributors, as well as Christoph
21 * Herrmann and Thomas-Henning von Kamptz.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * $TSHeader: src/sbin/growfs/growfs.c,v 1.5 2000/12/12 19:31:00 tomsoft Exp $
43 static const char copyright[] =
44 "@(#) Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz\n\
45 Copyright (c) 1980, 1989, 1993 The Regents of the University of California.\n\
46 All rights reserved.\n";
49 #include <sys/cdefs.h>
50 __FBSDID("$FreeBSD$");
52 /* ********************************************************** INCLUDES ***** */
53 #include <sys/param.h>
54 #include <sys/disklabel.h>
55 #include <sys/ioctl.h>
70 #include <ufs/ufs/dinode.h>
71 #include <ufs/ffs/fs.h>
75 /* *************************************************** GLOBALS & TYPES ***** */
77 int _dbg_lvl_ = (DL_INFO); /* DL_TRC */
84 #define sblock fsun1.fs /* the new superblock */
85 #define osblock fsun2.fs /* the old superblock */
88 * Possible superblock locations ordered from most to least likely.
90 static int sblock_try[] = SBLOCKSEARCH;
91 static ufs2_daddr_t sblockloc;
97 #define acg cgun1.cg /* a cylinder cgroup (new) */
98 #define aocg cgun2.cg /* an old cylinder group */
100 static char ablk[MAXBSIZE]; /* a block */
102 static struct csum *fscs; /* cylinder summary */
105 struct ufs1_dinode dp1;
106 struct ufs2_dinode dp2;
108 #define DIP(dp, field) \
109 ((sblock.fs_magic == FS_UFS1_MAGIC) ? \
110 (uint32_t)(dp)->dp1.field : (dp)->dp2.field)
111 #define DIP_SET(dp, field, val) do { \
112 if (sblock.fs_magic == FS_UFS1_MAGIC) \
113 (dp)->dp1.field = (val); \
115 (dp)->dp2.field = (val); \
117 static ufs2_daddr_t inoblk; /* inode block address */
118 static char inobuf[MAXBSIZE]; /* inode block */
119 ino_t maxino; /* last valid inode */
120 static int unlabeled; /* unlabeled partition, e.g. vinum volume etc. */
123 * An array of elements of type struct gfs_bpp describes all blocks to
124 * be relocated in order to free the space needed for the cylinder group
125 * summary for all cylinder groups located in the first cylinder group.
128 ufs2_daddr_t old; /* old block number */
129 ufs2_daddr_t new; /* new block number */
130 #define GFS_FL_FIRST 1
131 #define GFS_FL_LAST 2
132 unsigned int flags; /* special handling required */
133 int found; /* how many references were updated */
136 /* ******************************************************** PROTOTYPES ***** */
137 static void growfs(int, int, unsigned int);
138 static void rdfs(ufs2_daddr_t, size_t, void *, int);
139 static void wtfs(ufs2_daddr_t, size_t, void *, int, unsigned int);
140 static ufs2_daddr_t alloc(void);
141 static int charsperline(void);
142 static void usage(void);
143 static int isblock(struct fs *, unsigned char *, int);
144 static void clrblock(struct fs *, unsigned char *, int);
145 static void setblock(struct fs *, unsigned char *, int);
146 static void initcg(int, time_t, int, unsigned int);
147 static void updjcg(int, time_t, int, int, unsigned int);
148 static void updcsloc(time_t, int, int, unsigned int);
149 static struct disklabel *get_disklabel(int);
150 static void return_disklabel(int, struct disklabel *, unsigned int);
151 static union dinode *ginode(ino_t, int, int);
152 static void frag_adjust(ufs2_daddr_t, int);
153 static int cond_bl_upd(ufs2_daddr_t *, struct gfs_bpp *, int, int,
155 static void updclst(int);
156 static void updrefs(int, ino_t, struct gfs_bpp *, int, int, unsigned int);
157 static void indirchk(ufs_lbn_t, ufs_lbn_t, ufs2_daddr_t, ufs_lbn_t,
158 struct gfs_bpp *, int, int, unsigned int);
159 static void get_dev_size(int, int *);
161 /* ************************************************************ growfs ***** */
163 * Here we actually start growing the file system. We basically read the
164 * cylinder summary from the first cylinder group as we want to update
165 * this on the fly during our various operations. First we handle the
166 * changes in the former last cylinder group. Afterwards we create all new
167 * cylinder groups. Now we handle the cylinder group containing the
168 * cylinder summary which might result in a relocation of the whole
169 * structure. In the end we write back the updated cylinder summary, the
170 * new superblock, and slightly patched versions of the super block
174 growfs(int fsi, int fso, unsigned int Nflag)
182 static int randinit=0;
190 #else /* not FSIRAND */
198 * Get the cylinder summary into the memory.
200 fscs = (struct csum *)calloc((size_t)1, (size_t)sblock.fs_cssize);
202 errx(1, "calloc failed");
204 for (i = 0; i < osblock.fs_cssize; i += osblock.fs_bsize) {
205 rdfs(fsbtodb(&osblock, osblock.fs_csaddr +
206 numfrags(&osblock, i)), (size_t)MIN(osblock.fs_cssize - i,
207 osblock.fs_bsize), (void *)(((char *)fscs)+i), fsi);
212 struct csum *dbg_csp;
217 for(dbg_csc=0; dbg_csc<osblock.fs_ncg; dbg_csc++) {
218 snprintf(dbg_line, sizeof(dbg_line),
219 "%d. old csum in old location", dbg_csc);
220 DBG_DUMP_CSUM(&osblock,
225 #endif /* FS_DEBUG */
226 DBG_PRINT0("fscs read\n");
229 * Do all needed changes in the former last cylinder group.
231 updjcg(osblock.fs_ncg-1, utime, fsi, fso, Nflag);
234 * Dump out summary information about file system.
236 # define B2MBFACTOR (1 / (1024.0 * 1024.0))
237 printf("growfs: %.1fMB (%jd sectors) block size %d, fragment size %d\n",
238 (float)sblock.fs_size * sblock.fs_fsize * B2MBFACTOR,
239 (intmax_t)fsbtodb(&sblock, sblock.fs_size), sblock.fs_bsize,
241 printf("\tusing %d cylinder groups of %.2fMB, %d blks, %d inodes.\n",
242 sblock.fs_ncg, (float)sblock.fs_fpg * sblock.fs_fsize * B2MBFACTOR,
243 sblock.fs_fpg / sblock.fs_frag, sblock.fs_ipg);
244 if (sblock.fs_flags & FS_DOSOFTDEP)
245 printf("\twith soft updates\n");
249 * Now build the cylinders group blocks and
250 * then print out indices of cylinder groups.
252 printf("super-block backups (for fsck -b #) at:\n");
254 width = charsperline();
257 * Iterate for only the new cylinder groups.
259 for (cylno = osblock.fs_ncg; cylno < sblock.fs_ncg; cylno++) {
260 initcg(cylno, utime, fso, Nflag);
261 j = sprintf(tmpbuf, " %jd%s",
262 (intmax_t)fsbtodb(&sblock, cgsblock(&sblock, cylno)),
263 cylno < (sblock.fs_ncg-1) ? "," : "" );
264 if (i + j >= width) {
269 printf("%s", tmpbuf);
275 * Do all needed changes in the first cylinder group.
276 * allocate blocks in new location
278 updcsloc(utime, fsi, fso, Nflag);
281 * Now write the cylinder summary back to disk.
283 for (i = 0; i < sblock.fs_cssize; i += sblock.fs_bsize) {
284 wtfs(fsbtodb(&sblock, sblock.fs_csaddr + numfrags(&sblock, i)),
285 (size_t)MIN(sblock.fs_cssize - i, sblock.fs_bsize),
286 (void *)(((char *)fscs) + i), fso, Nflag);
288 DBG_PRINT0("fscs written\n");
292 struct csum *dbg_csp;
297 for(dbg_csc=0; dbg_csc<sblock.fs_ncg; dbg_csc++) {
298 snprintf(dbg_line, sizeof(dbg_line),
299 "%d. new csum in new location", dbg_csc);
300 DBG_DUMP_CSUM(&sblock,
305 #endif /* FS_DEBUG */
308 * Now write the new superblock back to disk.
310 sblock.fs_time = utime;
311 wtfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
312 DBG_PRINT0("sblock written\n");
314 "new initial sblock");
317 * Clean up the dynamic fields in our superblock copies.
322 sblock.fs_cgrotor = 0;
324 memset((void *)&sblock.fs_fsmnt, 0, sizeof(sblock.fs_fsmnt));
325 sblock.fs_flags &= FS_DOSOFTDEP;
329 * The following fields are currently distributed from the superblock
337 * fs_flags regarding SOFTPDATES
339 * We probably should rather change the summary for the cylinder group
340 * statistics here to the value of what would be in there, if the file
341 * system were created initially with the new size. Therefor we still
342 * need to find an easy way of calculating that.
343 * Possibly we can try to read the first superblock copy and apply the
344 * "diffed" stats between the old and new superblock by still copying
345 * certain parameters onto that.
349 * Write out the duplicate super blocks.
351 for (cylno = 0; cylno < sblock.fs_ncg; cylno++) {
352 wtfs(fsbtodb(&sblock, cgsblock(&sblock, cylno)),
353 (size_t)SBLOCKSIZE, (void *)&sblock, fso, Nflag);
355 DBG_PRINT0("sblock copies written\n");
357 "new other sblocks");
363 /* ************************************************************ initcg ***** */
365 * This creates a new cylinder group structure, for more details please see
366 * the source of newfs(8), as this function is taken over almost unchanged.
367 * As this is never called for the first cylinder group, the special
368 * provisions for that case are removed here.
371 initcg(int cylno, time_t utime, int fso, unsigned int Nflag)
374 static caddr_t iobuf;
376 ufs2_daddr_t i, cbase, dmax;
377 struct ufs1_dinode *dp1;
379 uint d, dupper, dlower;
381 if (iobuf == NULL && (iobuf = malloc(sblock.fs_bsize * 3)) == NULL)
382 errx(37, "panic: cannot allocate I/O buffer");
385 * Determine block bounds for cylinder group.
386 * Allow space for super block summary information in first
389 cbase = cgbase(&sblock, cylno);
390 dmax = cbase + sblock.fs_fpg;
391 if (dmax > sblock.fs_size)
392 dmax = sblock.fs_size;
393 dlower = cgsblock(&sblock, cylno) - cbase;
394 dupper = cgdmin(&sblock, cylno) - cbase;
395 if (cylno == 0) /* XXX fscs may be relocated */
396 dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
398 memset(&acg, 0, sblock.fs_cgsize);
400 acg.cg_magic = CG_MAGIC;
402 acg.cg_niblk = sblock.fs_ipg;
403 acg.cg_initediblk = sblock.fs_ipg < 2 * INOPB(&sblock) ?
404 sblock.fs_ipg : 2 * INOPB(&sblock);
405 acg.cg_ndblk = dmax - cbase;
406 if (sblock.fs_contigsumsize > 0)
407 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
408 start = &acg.cg_space[0] - (u_char *)(&acg.cg_firstfield);
409 if (sblock.fs_magic == FS_UFS2_MAGIC) {
410 acg.cg_iusedoff = start;
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 > (unsigned)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 * For the old file system, we have to initialize all the inodes.
451 if (sblock.fs_magic == FS_UFS1_MAGIC) {
452 bzero(iobuf, sblock.fs_bsize);
453 for (i = 0; i < sblock.fs_ipg / INOPF(&sblock);
454 i += sblock.fs_frag) {
455 dp1 = (struct ufs1_dinode *)(void *)iobuf;
457 for (j = 0; j < INOPB(&sblock); j++) {
458 dp1->di_gen = random();
462 wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno) + i),
463 sblock.fs_bsize, iobuf, fso, Nflag);
468 * In cylno 0, beginning space is reserved
469 * for boot and super blocks.
471 for (d = 0; d < dlower; d += sblock.fs_frag) {
472 blkno = d / sblock.fs_frag;
473 setblock(&sblock, cg_blksfree(&acg), blkno);
474 if (sblock.fs_contigsumsize > 0)
475 setbit(cg_clustersfree(&acg), blkno);
476 acg.cg_cs.cs_nbfree++;
478 sblock.fs_dsize += dlower;
480 sblock.fs_dsize += acg.cg_ndblk - dupper;
481 if ((i = dupper % sblock.fs_frag)) {
482 acg.cg_frsum[sblock.fs_frag - i]++;
483 for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) {
484 setbit(cg_blksfree(&acg), dupper);
485 acg.cg_cs.cs_nffree++;
488 for (d = dupper; d + sblock.fs_frag <= acg.cg_ndblk;
489 d += sblock.fs_frag) {
490 blkno = d / sblock.fs_frag;
491 setblock(&sblock, cg_blksfree(&acg), blkno);
492 if (sblock.fs_contigsumsize > 0)
493 setbit(cg_clustersfree(&acg), blkno);
494 acg.cg_cs.cs_nbfree++;
496 if (d < acg.cg_ndblk) {
497 acg.cg_frsum[acg.cg_ndblk - d]++;
498 for (; d < acg.cg_ndblk; d++) {
499 setbit(cg_blksfree(&acg), d);
500 acg.cg_cs.cs_nffree++;
503 if (sblock.fs_contigsumsize > 0) {
504 int32_t *sump = cg_clustersum(&acg);
505 u_char *mapp = cg_clustersfree(&acg);
510 for (i = 0; i < acg.cg_nclusterblks; i++) {
511 if ((map & bit) != 0)
514 if (run > sblock.fs_contigsumsize)
515 run = sblock.fs_contigsumsize;
519 if ((i & (CHAR_BIT - 1)) != CHAR_BIT - 1)
527 if (run > sblock.fs_contigsumsize)
528 run = sblock.fs_contigsumsize;
532 sblock.fs_cstotal.cs_ndir += acg.cg_cs.cs_ndir;
533 sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree;
534 sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree;
535 sblock.fs_cstotal.cs_nifree += acg.cg_cs.cs_nifree;
538 memcpy(iobuf, &acg, sblock.fs_cgsize);
539 memset(iobuf + sblock.fs_cgsize, '\0',
540 sblock.fs_bsize * 3 - sblock.fs_cgsize);
542 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
543 sblock.fs_bsize * 3, iobuf, fso, Nflag);
544 DBG_DUMP_CG(&sblock, "new cg", &acg);
550 /* ******************************************************* frag_adjust ***** */
552 * Here we add or subtract (sign +1/-1) the available fragments in a given
553 * block to or from the fragment statistics. By subtracting before and adding
554 * after an operation on the free frag map we can easy update the fragment
555 * statistic, which seems to be otherwise a rather complex operation.
558 frag_adjust(ufs2_daddr_t frag, int sign)
560 DBG_FUNC("frag_adjust")
568 * Here frag only needs to point to any fragment in the block we want
571 for(f=rounddown(frag, sblock.fs_frag);
572 f<roundup(frag+1, sblock.fs_frag);
575 * Count contiguous free fragments.
577 if(isset(cg_blksfree(&acg), f)) {
580 if(fragsize && fragsize<sblock.fs_frag) {
582 * We found something in between.
584 acg.cg_frsum[fragsize]+=sign;
585 DBG_PRINT2("frag_adjust [%d]+=%d\n",
592 if(fragsize && fragsize<sblock.fs_frag) {
594 * We found something.
596 acg.cg_frsum[fragsize]+=sign;
597 DBG_PRINT2("frag_adjust [%d]+=%d\n",
601 DBG_PRINT2("frag_adjust [[%d]]+=%d\n",
609 /* ******************************************************* cond_bl_upd ***** */
611 * Here we conditionally update a pointer to a fragment. We check for all
612 * relocated blocks if any of its fragments is referenced by the current
613 * field, and update the pointer to the respective fragment in our new
614 * block. If we find a reference we write back the block immediately,
615 * as there is no easy way for our general block reading engine to figure
616 * out if a write back operation is needed.
619 cond_bl_upd(ufs2_daddr_t *block, struct gfs_bpp *field, int fsi, int fso,
622 DBG_FUNC("cond_bl_upd")
624 ufs2_daddr_t src, dst;
630 for (f = field; f->old != 0; f++) {
632 if (fragstoblks(&sblock, src) != f->old)
635 * The fragment is part of the block, so update.
637 dst = blkstofrags(&sblock, f->new);
638 fragnum = fragnum(&sblock, src);
639 *block = dst + fragnum;
641 DBG_PRINT3("scg (%jd->%jd)[%d] reference updated\n",
647 * Copy the block back immediately.
649 * XXX If src is from an indirect block we have
650 * to implement copy on write here in case of
653 ibuf = malloc(sblock.fs_bsize);
655 errx(1, "malloc failed");
657 rdfs(fsbtodb(&sblock, src), (size_t)sblock.fs_bsize, ibuf, fsi);
658 wtfs(dst, (size_t)sblock.fs_bsize, ibuf, fso, Nflag);
661 * The same block can't be found again in this loop.
670 /* ************************************************************ updjcg ***** */
672 * Here we do all needed work for the former last cylinder group. It has to be
673 * changed in any case, even if the file system ended exactly on the end of
674 * this group, as there is some slightly inconsistent handling of the number
675 * of cylinders in the cylinder group. We start again by reading the cylinder
676 * group from disk. If the last block was not fully available, we first handle
677 * the missing fragments, then we handle all new full blocks in that file
678 * system and finally we handle the new last fragmented block in the file
679 * system. We again have to handle the fragment statistics rotational layout
680 * tables and cluster summary during all those operations.
683 updjcg(int cylno, time_t utime, int fsi, int fso, unsigned int Nflag)
686 ufs2_daddr_t cbase, dmax, dupper;
694 * Read the former last (joining) cylinder group from disk, and make
697 rdfs(fsbtodb(&osblock, cgtod(&osblock, cylno)),
698 (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
699 DBG_PRINT0("jcg read\n");
704 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
707 * If the cylinder group had already its new final size almost
708 * nothing is to be done ... except:
709 * For some reason the value of cg_ncyl in the last cylinder group has
710 * to be zero instead of fs_cpg. As this is now no longer the last
711 * cylinder group we have to change that value now to fs_cpg.
714 if(cgbase(&osblock, cylno+1) == osblock.fs_size) {
715 if (sblock.fs_magic == FS_UFS1_MAGIC)
716 acg.cg_old_ncyl=sblock.fs_old_cpg;
718 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)),
719 (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
720 DBG_PRINT0("jcg written\n");
730 * Set up some variables needed later.
732 cbase = cgbase(&sblock, cylno);
733 dmax = cbase + sblock.fs_fpg;
734 if (dmax > sblock.fs_size)
735 dmax = sblock.fs_size;
736 dupper = cgdmin(&sblock, cylno) - cbase;
737 if (cylno == 0) { /* XXX fscs may be relocated */
738 dupper += howmany(sblock.fs_cssize, sblock.fs_fsize);
742 * Set pointer to the cylinder summary for our cylinder group.
747 * Touch the cylinder group, update all fields in the cylinder group as
748 * needed, update the free space in the superblock.
751 if ((unsigned)cylno == sblock.fs_ncg - 1) {
753 * This is still the last cylinder group.
755 if (sblock.fs_magic == FS_UFS1_MAGIC)
757 sblock.fs_old_ncyl % sblock.fs_old_cpg;
759 acg.cg_old_ncyl = sblock.fs_old_cpg;
761 DBG_PRINT2("jcg dbg: %d %u",
765 if (sblock.fs_magic == FS_UFS1_MAGIC)
771 acg.cg_ndblk = dmax - cbase;
772 sblock.fs_dsize += acg.cg_ndblk-aocg.cg_ndblk;
773 if (sblock.fs_contigsumsize > 0) {
774 acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag;
778 * Now we have to update the free fragment bitmap for our new free
779 * space. There again we have to handle the fragmentation and also
780 * the rotational layout tables and the cluster summary. This is
781 * also done per fragment for the first new block if the old file
782 * system end was not on a block boundary, per fragment for the new
783 * last block if the new file system end is not on a block boundary,
784 * and per block for all space in between.
786 * Handle the first new block here if it was partially available
789 if(osblock.fs_size % sblock.fs_frag) {
790 if(roundup(osblock.fs_size, sblock.fs_frag)<=sblock.fs_size) {
792 * The new space is enough to fill at least this
796 for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag)-1;
797 i>=osblock.fs_size-cbase;
799 setbit(cg_blksfree(&acg), i);
800 acg.cg_cs.cs_nffree++;
805 * Check if the fragment just created could join an
806 * already existing fragment at the former end of the
809 if(isblock(&sblock, cg_blksfree(&acg),
810 ((osblock.fs_size - cgbase(&sblock, cylno))/
813 * The block is now completely available.
815 DBG_PRINT0("block was\n");
816 acg.cg_frsum[osblock.fs_size%sblock.fs_frag]--;
817 acg.cg_cs.cs_nbfree++;
818 acg.cg_cs.cs_nffree-=sblock.fs_frag;
819 k=rounddown(osblock.fs_size-cbase,
821 updclst((osblock.fs_size-cbase)/sblock.fs_frag);
824 * Lets rejoin a possible partially growed
828 while(isset(cg_blksfree(&acg), i) &&
829 (i>=rounddown(osblock.fs_size-cbase,
841 * We only grow by some fragments within this last
844 for(i=sblock.fs_size-cbase-1;
845 i>=osblock.fs_size-cbase;
847 setbit(cg_blksfree(&acg), i);
848 acg.cg_cs.cs_nffree++;
852 * Lets rejoin a possible partially growed fragment.
855 while(isset(cg_blksfree(&acg), i) &&
856 (i>=rounddown(osblock.fs_size-cbase,
869 * Handle all new complete blocks here.
871 for(i=roundup(osblock.fs_size-cbase, sblock.fs_frag);
872 i+sblock.fs_frag<=dmax-cbase; /* XXX <= or only < ? */
874 j = i / sblock.fs_frag;
875 setblock(&sblock, cg_blksfree(&acg), j);
877 acg.cg_cs.cs_nbfree++;
881 * Handle the last new block if there are stll some new fragments left.
882 * Here we don't have to bother about the cluster summary or the even
883 * the rotational layout table.
885 if (i < (dmax - cbase)) {
886 acg.cg_frsum[dmax - cbase - i]++;
887 for (; i < dmax - cbase; i++) {
888 setbit(cg_blksfree(&acg), i);
889 acg.cg_cs.cs_nffree++;
893 sblock.fs_cstotal.cs_nffree +=
894 (acg.cg_cs.cs_nffree - aocg.cg_cs.cs_nffree);
895 sblock.fs_cstotal.cs_nbfree +=
896 (acg.cg_cs.cs_nbfree - aocg.cg_cs.cs_nbfree);
898 * The following statistics are not changed here:
899 * sblock.fs_cstotal.cs_ndir
900 * sblock.fs_cstotal.cs_nifree
901 * As the statistics for this cylinder group are ready, copy it to
902 * the summary information array.
907 * Write the updated "joining" cylinder group back to disk.
909 wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize,
910 (void *)&acg, fso, Nflag);
911 DBG_PRINT0("jcg written\n");
920 /* ********************************************************** updcsloc ***** */
922 * Here we update the location of the cylinder summary. We have two possible
923 * ways of growing the cylinder summary.
924 * (1) We can try to grow the summary in the current location, and relocate
925 * possibly used blocks within the current cylinder group.
926 * (2) Alternatively we can relocate the whole cylinder summary to the first
927 * new completely empty cylinder group. Once the cylinder summary is no
928 * longer in the beginning of the first cylinder group you should never
929 * use a version of fsck which is not aware of the possibility to have
930 * this structure in a non standard place.
931 * Option (1) is considered to be less intrusive to the structure of the file-
932 * system. So we try to stick to that whenever possible. If there is not enough
933 * space in the cylinder group containing the cylinder summary we have to use
934 * method (2). In case of active snapshots in the file system we probably can
935 * completely avoid implementing copy on write if we stick to method (2) only.
938 updcsloc(time_t utime, int fsi, int fso, unsigned int Nflag)
944 ufs2_daddr_t cbase, dupper, odupper, d, f, g;
954 if(howmany(sblock.fs_cssize, sblock.fs_fsize) ==
955 howmany(osblock.fs_cssize, osblock.fs_fsize)) {
957 * No new fragment needed.
962 ocscg=dtog(&osblock, osblock.fs_csaddr);
964 blocks = 1+howmany(sblock.fs_cssize, sblock.fs_bsize)-
965 howmany(osblock.fs_cssize, osblock.fs_bsize);
968 * Read original cylinder group from disk, and make a copy.
969 * XXX If Nflag is set in some very rare cases we now miss
970 * some changes done in updjcg by reading the unmodified
973 rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)),
974 (size_t)osblock.fs_cgsize, (void *)&aocg, fsi);
975 DBG_PRINT0("oscg read\n");
980 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
983 * Touch the cylinder group, set up local variables needed later
984 * and update the superblock.
989 * XXX In the case of having active snapshots we may need much more
990 * blocks for the copy on write. We need each block twice, and
991 * also up to 8*3 blocks for indirect blocks for all possible
994 if(/*((int)sblock.fs_time&0x3)>0||*/ cs->cs_nbfree < blocks) {
996 * There is not enough space in the old cylinder group to
997 * relocate all blocks as needed, so we relocate the whole
998 * cylinder group summary to a new group. We try to use the
999 * first complete new cylinder group just created. Within the
1000 * cylinder group we align the area immediately after the
1001 * cylinder group information location in order to be as
1002 * close as possible to the original implementation of ffs.
1004 * First we have to make sure we'll find enough space in the
1005 * new cylinder group. If not, then we currently give up.
1006 * We start with freeing everything which was used by the
1007 * fragments of the old cylinder summary in the current group.
1008 * Now we write back the group meta data, read in the needed
1009 * meta data from the new cylinder group, and start allocating
1010 * within that group. Here we can assume, the group to be
1011 * completely empty. Which makes the handling of fragments and
1012 * clusters a lot easier.
1015 if(sblock.fs_ncg-osblock.fs_ncg < 2) {
1016 errx(2, "panic: not enough space");
1020 * Point "d" to the first fragment not used by the cylinder
1023 d=osblock.fs_csaddr+(osblock.fs_cssize/osblock.fs_fsize);
1026 * Set up last cluster size ("lcs") already here. Calculate
1027 * the size for the trailing cluster just behind where "d"
1030 if(sblock.fs_contigsumsize > 0) {
1031 for(block=howmany(d%sblock.fs_fpg, sblock.fs_frag),
1032 lcs=0; lcs<sblock.fs_contigsumsize;
1034 if(isclr(cg_clustersfree(&acg), block)){
1041 * Point "d" to the last frag used by the cylinder summary.
1045 DBG_PRINT1("d=%jd\n",
1047 if((d+1)%sblock.fs_frag) {
1049 * The end of the cylinder summary is not a complete
1053 frag_adjust(d%sblock.fs_fpg, -1);
1054 for(; (d+1)%sblock.fs_frag; d--) {
1055 DBG_PRINT1("d=%jd\n",
1057 setbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1058 acg.cg_cs.cs_nffree++;
1059 sblock.fs_cstotal.cs_nffree++;
1062 * Point "d" to the last fragment of the last
1063 * (incomplete) block of the cylinder summary.
1066 frag_adjust(d%sblock.fs_fpg, 1);
1068 if(isblock(&sblock, cg_blksfree(&acg),
1069 (d%sblock.fs_fpg)/sblock.fs_frag)) {
1070 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1071 acg.cg_cs.cs_nffree-=sblock.fs_frag;
1072 acg.cg_cs.cs_nbfree++;
1073 sblock.fs_cstotal.cs_nffree-=sblock.fs_frag;
1074 sblock.fs_cstotal.cs_nbfree++;
1075 if(sblock.fs_contigsumsize > 0) {
1076 setbit(cg_clustersfree(&acg),
1077 (d%sblock.fs_fpg)/sblock.fs_frag);
1078 if(lcs < sblock.fs_contigsumsize) {
1084 cg_clustersum(&acg)[lcs]++;
1089 * Point "d" to the first fragment of the block before
1090 * the last incomplete block.
1095 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1096 for(d=rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr;
1097 d-=sblock.fs_frag) {
1099 DBG_PRINT1("d=%jd\n", (intmax_t)d);
1100 setblock(&sblock, cg_blksfree(&acg),
1101 (d%sblock.fs_fpg)/sblock.fs_frag);
1102 acg.cg_cs.cs_nbfree++;
1103 sblock.fs_cstotal.cs_nbfree++;
1104 if(sblock.fs_contigsumsize > 0) {
1105 setbit(cg_clustersfree(&acg),
1106 (d%sblock.fs_fpg)/sblock.fs_frag);
1108 * The last cluster size is already set up.
1110 if(lcs < sblock.fs_contigsumsize) {
1112 cg_clustersum(&acg)[lcs]--;
1115 cg_clustersum(&acg)[lcs]++;
1122 * Now write the former cylinder group containing the cylinder
1123 * summary back to disk.
1125 wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)),
1126 (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1127 DBG_PRINT0("oscg written\n");
1128 DBG_DUMP_CG(&sblock,
1133 * Find the beginning of the new cylinder group containing the
1136 sblock.fs_csaddr=cgdmin(&sblock, osblock.fs_ncg);
1137 ncscg=dtog(&sblock, sblock.fs_csaddr);
1142 * If Nflag is specified, we would now read random data instead
1143 * of an empty cg structure from disk. So we can't simulate that
1147 DBG_PRINT0("nscg update skipped\n");
1153 * Read the future cylinder group containing the cylinder
1154 * summary from disk, and make a copy.
1156 rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1157 (size_t)sblock.fs_cgsize, (void *)&aocg, fsi);
1158 DBG_PRINT0("nscg read\n");
1159 DBG_DUMP_CG(&sblock,
1163 memcpy((void *)&cgun1, (void *)&cgun2, sizeof(cgun2));
1166 * Allocate all complete blocks used by the new cylinder
1169 for(d=sblock.fs_csaddr; d+sblock.fs_frag <=
1170 sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize);
1171 d+=sblock.fs_frag) {
1172 clrblock(&sblock, cg_blksfree(&acg),
1173 (d%sblock.fs_fpg)/sblock.fs_frag);
1174 acg.cg_cs.cs_nbfree--;
1175 sblock.fs_cstotal.cs_nbfree--;
1176 if(sblock.fs_contigsumsize > 0) {
1177 clrbit(cg_clustersfree(&acg),
1178 (d%sblock.fs_fpg)/sblock.fs_frag);
1183 * Allocate all fragments used by the cylinder summary in the
1186 if(d<sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize)) {
1187 for(; d-sblock.fs_csaddr<
1188 sblock.fs_cssize/sblock.fs_fsize;
1190 clrbit(cg_blksfree(&acg), d%sblock.fs_fpg);
1191 acg.cg_cs.cs_nffree--;
1192 sblock.fs_cstotal.cs_nffree--;
1194 acg.cg_cs.cs_nbfree--;
1195 acg.cg_cs.cs_nffree+=sblock.fs_frag;
1196 sblock.fs_cstotal.cs_nbfree--;
1197 sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1198 if(sblock.fs_contigsumsize > 0) {
1199 clrbit(cg_clustersfree(&acg),
1200 (d%sblock.fs_fpg)/sblock.fs_frag);
1203 frag_adjust(d%sblock.fs_fpg, +1);
1206 * XXX Handle the cluster statistics here in the case this
1207 * cylinder group is now almost full, and the remaining
1208 * space is less then the maximum cluster size. This is
1209 * probably not needed, as you would hardly find a file
1210 * system which has only MAXCSBUFS+FS_MAXCONTIG of free
1211 * space right behind the cylinder group information in
1212 * any new cylinder group.
1216 * Update our statistics in the cylinder summary.
1221 * Write the new cylinder group containing the cylinder summary
1224 wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)),
1225 (size_t)sblock.fs_cgsize, (void *)&acg, fso, Nflag);
1226 DBG_PRINT0("nscg written\n");
1227 DBG_DUMP_CG(&sblock,
1235 * We have got enough of space in the current cylinder group, so we
1236 * can relocate just a few blocks, and let the summary information
1237 * grow in place where it is right now.
1241 cbase = cgbase(&osblock, ocscg); /* old and new are equal */
1242 dupper = sblock.fs_csaddr - cbase +
1243 howmany(sblock.fs_cssize, sblock.fs_fsize);
1244 odupper = osblock.fs_csaddr - cbase +
1245 howmany(osblock.fs_cssize, osblock.fs_fsize);
1247 sblock.fs_dsize -= dupper-odupper;
1250 * Allocate the space for the array of blocks to be relocated.
1252 bp=(struct gfs_bpp *)malloc(((dupper-odupper)/sblock.fs_frag+2)*
1253 sizeof(struct gfs_bpp));
1255 errx(1, "malloc failed");
1257 memset((char *)bp, 0, ((dupper-odupper)/sblock.fs_frag+2)*
1258 sizeof(struct gfs_bpp));
1261 * Lock all new frags needed for the cylinder group summary. This is
1262 * done per fragment in the first and last block of the new required
1263 * area, and per block for all other blocks.
1265 * Handle the first new block here (but only if some fragments where
1266 * already used for the cylinder summary).
1269 frag_adjust(odupper, -1);
1270 for(d=odupper; ((d<dupper)&&(d%sblock.fs_frag)); d++) {
1271 DBG_PRINT1("scg first frag check loop d=%jd\n",
1273 if(isclr(cg_blksfree(&acg), d)) {
1275 bp[ind].old=d/sblock.fs_frag;
1276 bp[ind].flags|=GFS_FL_FIRST;
1277 if(roundup(d, sblock.fs_frag) >= dupper) {
1278 bp[ind].flags|=GFS_FL_LAST;
1283 clrbit(cg_blksfree(&acg), d);
1284 acg.cg_cs.cs_nffree--;
1285 sblock.fs_cstotal.cs_nffree--;
1288 * No cluster handling is needed here, as there was at least
1289 * one fragment in use by the cylinder summary in the old
1291 * No block-free counter handling here as this block was not
1295 frag_adjust(odupper, 1);
1298 * Handle all needed complete blocks here.
1300 for(; d+sblock.fs_frag<=dupper; d+=sblock.fs_frag) {
1301 DBG_PRINT1("scg block check loop d=%jd\n",
1303 if(!isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1304 for(f=d; f<d+sblock.fs_frag; f++) {
1305 if(isset(cg_blksfree(&aocg), f)) {
1306 acg.cg_cs.cs_nffree--;
1307 sblock.fs_cstotal.cs_nffree--;
1310 clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1311 bp[ind].old=d/sblock.fs_frag;
1314 clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag);
1315 acg.cg_cs.cs_nbfree--;
1316 sblock.fs_cstotal.cs_nbfree--;
1317 if(sblock.fs_contigsumsize > 0) {
1318 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1319 for(lcs=0, l=(d/sblock.fs_frag)+1;
1320 lcs<sblock.fs_contigsumsize;
1322 if(isclr(cg_clustersfree(&acg),l)){
1326 if(lcs < sblock.fs_contigsumsize) {
1327 cg_clustersum(&acg)[lcs+1]--;
1329 cg_clustersum(&acg)[lcs]++;
1335 * No fragment counter handling is needed here, as this finally
1336 * doesn't change after the relocation.
1341 * Handle all fragments needed in the last new affected block.
1344 frag_adjust(dupper-1, -1);
1346 if(isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) {
1347 acg.cg_cs.cs_nbfree--;
1348 sblock.fs_cstotal.cs_nbfree--;
1349 acg.cg_cs.cs_nffree+=sblock.fs_frag;
1350 sblock.fs_cstotal.cs_nffree+=sblock.fs_frag;
1351 if(sblock.fs_contigsumsize > 0) {
1352 clrbit(cg_clustersfree(&acg), d/sblock.fs_frag);
1353 for(lcs=0, l=(d/sblock.fs_frag)+1;
1354 lcs<sblock.fs_contigsumsize;
1356 if(isclr(cg_clustersfree(&acg),l)){
1360 if(lcs < sblock.fs_contigsumsize) {
1361 cg_clustersum(&acg)[lcs+1]--;
1363 cg_clustersum(&acg)[lcs]++;
1369 for(; d<dupper; d++) {
1370 DBG_PRINT1("scg second frag check loop d=%jd\n",
1372 if(isclr(cg_blksfree(&acg), d)) {
1373 bp[ind].old=d/sblock.fs_frag;
1374 bp[ind].flags|=GFS_FL_LAST;
1376 clrbit(cg_blksfree(&acg), d);
1377 acg.cg_cs.cs_nffree--;
1378 sblock.fs_cstotal.cs_nffree--;
1381 if(bp[ind].flags & GFS_FL_LAST) { /* we have to advance here */
1384 frag_adjust(dupper-1, 1);
1388 * If we found a block to relocate just do so.
1391 for(i=0; i<ind; i++) {
1392 if(!bp[i].old) { /* no more blocks listed */
1394 * XXX A relative blocknumber should not be
1395 * zero, which is not explicitly
1396 * guaranteed by our code.
1401 * Allocate a complete block in the same (current)
1404 bp[i].new=alloc()/sblock.fs_frag;
1407 * There is no frag_adjust() needed for the new block
1408 * as it will have no fragments yet :-).
1410 for(f=bp[i].old*sblock.fs_frag,
1411 g=bp[i].new*sblock.fs_frag;
1412 f<(bp[i].old+1)*sblock.fs_frag;
1414 if(isset(cg_blksfree(&aocg), f)) {
1415 setbit(cg_blksfree(&acg), g);
1416 acg.cg_cs.cs_nffree++;
1417 sblock.fs_cstotal.cs_nffree++;
1422 * Special handling is required if this was the first
1423 * block. We have to consider the fragments which were
1424 * used by the cylinder summary in the original block
1425 * which re to be free in the copy of our block. We
1426 * have to be careful if this first block happens to
1427 * be also the last block to be relocated.
1429 if(bp[i].flags & GFS_FL_FIRST) {
1430 for(f=bp[i].old*sblock.fs_frag,
1431 g=bp[i].new*sblock.fs_frag;
1434 setbit(cg_blksfree(&acg), g);
1435 acg.cg_cs.cs_nffree++;
1436 sblock.fs_cstotal.cs_nffree++;
1438 if(!(bp[i].flags & GFS_FL_LAST)) {
1439 frag_adjust(bp[i].new*sblock.fs_frag,1);
1444 * Special handling is required if this is the last
1445 * block to be relocated.
1447 if(bp[i].flags & GFS_FL_LAST) {
1448 frag_adjust(bp[i].new*sblock.fs_frag, 1);
1449 frag_adjust(bp[i].old*sblock.fs_frag, -1);
1451 f<roundup(dupper, sblock.fs_frag);
1453 if(isclr(cg_blksfree(&acg), f)) {
1454 setbit(cg_blksfree(&acg), f);
1455 acg.cg_cs.cs_nffree++;
1456 sblock.fs_cstotal.cs_nffree++;
1459 frag_adjust(bp[i].old*sblock.fs_frag, 1);
1463 * !!! Attach the cylindergroup offset here.
1465 bp[i].old+=cbase/sblock.fs_frag;
1466 bp[i].new+=cbase/sblock.fs_frag;
1469 * Copy the content of the block.
1472 * XXX Here we will have to implement a copy on write
1473 * in the case we have any active snapshots.
1475 rdfs(fsbtodb(&sblock, bp[i].old*sblock.fs_frag),
1476 (size_t)sblock.fs_bsize, (void *)&ablk, fsi);
1477 wtfs(fsbtodb(&sblock, bp[i].new*sblock.fs_frag),
1478 (size_t)sblock.fs_bsize, (void *)&ablk, fso, Nflag);
1479 DBG_DUMP_HEX(&sblock,
1480 "copied full block",
1481 (unsigned char *)&ablk);
1483 DBG_PRINT2("scg (%jd->%jd) block relocated\n",
1484 (intmax_t)bp[i].old,
1485 (intmax_t)bp[i].new);
1489 * Now we have to update all references to any fragment which
1490 * belongs to any block relocated. We iterate now over all
1491 * cylinder groups, within those over all non zero length
1494 for(cylno=0; cylno<osblock.fs_ncg; cylno++) {
1495 DBG_PRINT1("scg doing cg (%d)\n",
1497 for(inc=osblock.fs_ipg-1 ; inc>0 ; inc--) {
1498 updrefs(cylno, (ino_t)inc, bp, fsi, fso, Nflag);
1503 * All inodes are checked, now make sure the number of
1504 * references found make sense.
1506 for(i=0; i<ind; i++) {
1507 if(!bp[i].found || (bp[i].found>sblock.fs_frag)) {
1508 warnx("error: %jd refs found for block %jd.",
1509 (intmax_t)bp[i].found, (intmax_t)bp[i].old);
1515 * The following statistics are not changed here:
1516 * sblock.fs_cstotal.cs_ndir
1517 * sblock.fs_cstotal.cs_nifree
1518 * The following statistics were already updated on the fly:
1519 * sblock.fs_cstotal.cs_nffree
1520 * sblock.fs_cstotal.cs_nbfree
1521 * As the statistics for this cylinder group are ready, copy it to
1522 * the summary information array.
1528 * Write summary cylinder group back to disk.
1530 wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)), (size_t)sblock.fs_cgsize,
1531 (void *)&acg, fso, Nflag);
1532 DBG_PRINT0("scg written\n");
1533 DBG_DUMP_CG(&sblock,
1541 /* ************************************************************** rdfs ***** */
1543 * Here we read some block(s) from disk.
1546 rdfs(ufs2_daddr_t bno, size_t size, void *bf, int fsi)
1554 err(32, "rdfs: attempting to read negative block number");
1556 if (lseek(fsi, (off_t)bno * DEV_BSIZE, 0) < 0) {
1557 err(33, "rdfs: seek error: %jd", (intmax_t)bno);
1559 n = read(fsi, bf, size);
1560 if (n != (ssize_t)size) {
1561 err(34, "rdfs: read error: %jd", (intmax_t)bno);
1568 /* ************************************************************** wtfs ***** */
1570 * Here we write some block(s) to disk.
1573 wtfs(ufs2_daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag)
1584 if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0) {
1585 err(35, "wtfs: seek error: %ld", (long)bno);
1587 n = write(fso, bf, size);
1588 if (n != (ssize_t)size) {
1589 err(36, "wtfs: write error: %ld", (long)bno);
1596 /* ************************************************************* alloc ***** */
1598 * Here we allocate a free block in the current cylinder group. It is assumed,
1599 * that acg contains the current cylinder group. As we may take a block from
1600 * somewhere in the file system we have to handle cluster summary here.
1606 ufs2_daddr_t d, blkno;
1610 int dlower, dupper, dmax;
1614 if (acg.cg_magic != CG_MAGIC) {
1615 warnx("acg: bad magic number");
1619 if (acg.cg_cs.cs_nbfree == 0) {
1620 warnx("error: cylinder group ran out of space");
1625 * We start seeking for free blocks only from the space available after
1626 * the end of the new grown cylinder summary. Otherwise we allocate a
1627 * block here which we have to relocate a couple of seconds later again
1628 * again, and we are not prepared to to this anyway.
1631 dlower=cgsblock(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1632 dupper=cgdmin(&sblock, acg.cg_cgx)-cgbase(&sblock, acg.cg_cgx);
1633 dmax=cgbase(&sblock, acg.cg_cgx)+sblock.fs_fpg;
1634 if (dmax > sblock.fs_size) {
1635 dmax = sblock.fs_size;
1637 dmax-=cgbase(&sblock, acg.cg_cgx); /* retransform into cg */
1638 csmin=sblock.fs_csaddr-cgbase(&sblock, acg.cg_cgx);
1639 csmax=csmin+howmany(sblock.fs_cssize, sblock.fs_fsize);
1640 DBG_PRINT3("seek range: dl=%d, du=%d, dm=%d\n",
1644 DBG_PRINT2("range cont: csmin=%d, csmax=%d\n",
1648 for(d=0; (d<dlower && blkno==-1); d+=sblock.fs_frag) {
1649 if(d>=csmin && d<=csmax) {
1652 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1654 blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1658 for(d=dupper; (d<dmax && blkno==-1); d+=sblock.fs_frag) {
1659 if(d>=csmin && d<=csmax) {
1662 if(isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock,
1664 blkno = fragstoblks(&sblock, d);/* Yeah found a block */
1669 warnx("internal error: couldn't find promised block in cg");
1675 * This is needed if the block was found already in the first loop.
1677 d=blkstofrags(&sblock, blkno);
1679 clrblock(&sblock, cg_blksfree(&acg), blkno);
1680 if (sblock.fs_contigsumsize > 0) {
1682 * Handle the cluster allocation bitmap.
1684 clrbit(cg_clustersfree(&acg), blkno);
1686 * We possibly have split a cluster here, so we have to do
1687 * recalculate the sizes of the remaining cluster halves now,
1688 * and use them for updating the cluster summary information.
1690 * Lets start with the blocks before our allocated block ...
1692 for(lcs1=0, l=blkno-1; lcs1<sblock.fs_contigsumsize;
1694 if(isclr(cg_clustersfree(&acg),l)){
1699 * ... and continue with the blocks right after our allocated
1702 for(lcs2=0, l=blkno+1; lcs2<sblock.fs_contigsumsize;
1704 if(isclr(cg_clustersfree(&acg),l)){
1710 * Now update all counters.
1712 cg_clustersum(&acg)[MIN(lcs1+lcs2+1,sblock.fs_contigsumsize)]--;
1714 cg_clustersum(&acg)[lcs1]++;
1717 cg_clustersum(&acg)[lcs2]++;
1721 * Update all statistics based on blocks.
1723 acg.cg_cs.cs_nbfree--;
1724 sblock.fs_cstotal.cs_nbfree--;
1730 /* *********************************************************** isblock ***** */
1732 * Here we check if all frags of a block are free. For more details again
1733 * please see the source of newfs(8), as this function is taken over almost
1737 isblock(struct fs *fs, unsigned char *cp, int h)
1744 switch (fs->fs_frag) {
1747 return (cp[h] == 0xff);
1749 mask = 0x0f << ((h & 0x1) << 2);
1751 return ((cp[h >> 1] & mask) == mask);
1753 mask = 0x03 << ((h & 0x3) << 1);
1755 return ((cp[h >> 2] & mask) == mask);
1757 mask = 0x01 << (h & 0x7);
1759 return ((cp[h >> 3] & mask) == mask);
1761 fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag);
1767 /* ********************************************************** clrblock ***** */
1769 * Here we allocate a complete block in the block map. For more details again
1770 * please see the source of newfs(8), as this function is taken over almost
1774 clrblock(struct fs *fs, unsigned char *cp, int h)
1776 DBG_FUNC("clrblock")
1780 switch ((fs)->fs_frag) {
1785 cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2));
1788 cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1));
1791 cp[h >> 3] &= ~(0x01 << (h & 0x7));
1794 warnx("clrblock bad fs_frag %d", fs->fs_frag);
1802 /* ********************************************************** setblock ***** */
1804 * Here we free a complete block in the free block map. For more details again
1805 * please see the source of newfs(8), as this function is taken over almost
1809 setblock(struct fs *fs, unsigned char *cp, int h)
1811 DBG_FUNC("setblock")
1815 switch (fs->fs_frag) {
1820 cp[h >> 1] |= (0x0f << ((h & 0x1) << 2));
1823 cp[h >> 2] |= (0x03 << ((h & 0x3) << 1));
1826 cp[h >> 3] |= (0x01 << (h & 0x7));
1829 warnx("setblock bad fs_frag %d", fs->fs_frag);
1837 /* ************************************************************ ginode ***** */
1839 * This function provides access to an individual inode. We find out in which
1840 * block the requested inode is located, read it from disk if needed, and
1841 * return the pointer into that block. We maintain a cache of one block to
1842 * not read the same block again and again if we iterate linearly over all
1845 static union dinode *
1846 ginode(ino_t inumber, int fsi, int cg)
1849 static ino_t startinum = 0; /* first inode in cached block */
1854 * The inumber passed in is relative to the cg, so use it here to see
1855 * if the inode has been allocated yet.
1857 if (isclr(cg_inosused(&aocg), inumber)) {
1862 * Now make the inumber relative to the entire inode space so it can
1863 * be sanity checked.
1865 inumber += (cg * sblock.fs_ipg);
1866 if (inumber < ROOTINO) {
1870 if (inumber > maxino)
1871 errx(8, "bad inode number %d to ginode", inumber);
1872 if (startinum == 0 ||
1873 inumber < startinum || inumber >= startinum + INOPB(&sblock)) {
1874 inoblk = fsbtodb(&sblock, ino_to_fsba(&sblock, inumber));
1875 rdfs(inoblk, (size_t)sblock.fs_bsize, inobuf, fsi);
1876 startinum = (inumber / INOPB(&sblock)) * INOPB(&sblock);
1879 if (sblock.fs_magic == FS_UFS1_MAGIC)
1880 return (union dinode *)((uintptr_t)inobuf +
1881 (inumber % INOPB(&sblock)) * sizeof(struct ufs1_dinode));
1882 return (union dinode *)((uintptr_t)inobuf +
1883 (inumber % INOPB(&sblock)) * sizeof(struct ufs2_dinode));
1886 /* ****************************************************** charsperline ***** */
1888 * Figure out how many lines our current terminal has. For more details again
1889 * please see the source of newfs(8), as this function is taken over almost
1895 DBG_FUNC("charsperline")
1903 if (ioctl(0, TIOCGWINSZ, &ws) != -1) {
1904 columns = ws.ws_col;
1906 if (columns == 0 && (cp = getenv("COLUMNS"))) {
1910 columns = 80; /* last resort */
1917 /* ****************************************************** get_dev_size ***** */
1919 * Get the size of the partition if we can't figure it out from the disklabel,
1920 * e.g. from vinum volumes.
1923 get_dev_size(int fd, int *size)
1928 if (ioctl(fd, DIOCGSECTORSIZE, §orsize) == -1)
1929 err(1,"DIOCGSECTORSIZE");
1930 if (ioctl(fd, DIOCGMEDIASIZE, &mediasize) == -1)
1931 err(1,"DIOCGMEDIASIZE");
1933 if (sectorsize <= 0)
1934 errx(1, "bogus sectorsize: %d", sectorsize);
1936 *size = mediasize / sectorsize;
1939 /* ************************************************************** main ***** */
1941 * growfs(8) is a utility which allows to increase the size of an existing
1942 * ufs file system. Currently this can only be done on unmounted file system.
1943 * It recognizes some command line options to specify the new desired size,
1944 * and it does some basic checkings. The old file system size is determined
1945 * and after some more checks like we can really access the new last block
1946 * on the disk etc. we calculate the new parameters for the superblock. After
1947 * having done this we just call growfs() which will do the work. Before
1948 * we finish the only thing left is to update the disklabel.
1949 * We still have to provide support for snapshots. Therefore we first have to
1950 * understand what data structures are always replicated in the snapshot on
1951 * creation, for all other blocks we touch during our procedure, we have to
1952 * keep the old blocks unchanged somewhere available for the snapshots. If we
1953 * are lucky, then we only have to handle our blocks to be relocated in that
1955 * Also we have to consider in what order we actually update the critical
1956 * data structures of the file system to make sure, that in case of a disaster
1957 * fsck(8) is still able to restore any lost data.
1958 * The foreseen last step then will be to provide for growing even mounted
1959 * file systems. There we have to extend the mount() system call to provide
1960 * userland access to the file system locking facility.
1963 main(int argc, char **argv)
1966 char *device, *special, *cp;
1968 unsigned int size=0;
1970 unsigned int Nflag=0;
1973 struct disklabel *lp;
1974 struct partition *pp;
1980 #endif /* FSMAXSNAP */
1984 while((ch=getopt(argc, argv, "Ns:vy")) != -1) {
1990 size=(size_t)atol(optarg);
1995 case 'v': /* for compatibility to newfs */
2015 * Now try to guess the (raw)device name.
2017 if (0 == strrchr(device, '/')) {
2019 * No path prefix was given, so try in that order:
2025 * FreeBSD now doesn't distinguish between raw and block
2026 * devices any longer, but it should still work this way.
2028 len=strlen(device)+strlen(_PATH_DEV)+2+strlen("vinum/");
2029 special=(char *)malloc(len);
2030 if(special == NULL) {
2031 errx(1, "malloc failed");
2033 snprintf(special, len, "%sr%s", _PATH_DEV, device);
2034 if (stat(special, &st) == -1) {
2035 snprintf(special, len, "%s%s", _PATH_DEV, device);
2036 if (stat(special, &st) == -1) {
2037 snprintf(special, len, "%svinum/r%s",
2039 if (stat(special, &st) == -1) {
2040 /* For now this is the 'last resort' */
2041 snprintf(special, len, "%svinum/%s",
2050 * Try to access our devices for writing ...
2055 fso = open(device, O_WRONLY);
2057 err(1, "%s", device);
2064 fsi = open(device, O_RDONLY);
2066 err(1, "%s", device);
2070 * Try to read a label and guess the slice if not specified. This
2071 * code should guess the right thing and avoid to bother the user
2072 * with the task of specifying the option -v on vinum volumes.
2074 cp=device+strlen(device)-1;
2075 lp = get_disklabel(fsi);
2079 pp = &lp->d_partitions[2];
2080 } else if (*cp>='a' && *cp<='h') {
2081 pp = &lp->d_partitions[*cp - 'a'];
2083 errx(1, "unknown device");
2085 p_size = pp->p_size;
2087 get_dev_size(fsi, &p_size);
2091 * Check if that partition is suitable for growing a file system.
2094 errx(1, "partition is unavailable");
2098 * Read the current superblock, and take a backup.
2100 for (i = 0; sblock_try[i] != -1; i++) {
2101 sblockloc = sblock_try[i] / DEV_BSIZE;
2102 rdfs(sblockloc, (size_t)SBLOCKSIZE, (void *)&(osblock), fsi);
2103 if ((osblock.fs_magic == FS_UFS1_MAGIC ||
2104 (osblock.fs_magic == FS_UFS2_MAGIC &&
2105 osblock.fs_sblockloc == sblock_try[i])) &&
2106 osblock.fs_bsize <= MAXBSIZE &&
2107 osblock.fs_bsize >= (int32_t) sizeof(struct fs))
2110 if (sblock_try[i] == -1) {
2111 errx(1, "superblock not recognized");
2113 memcpy((void *)&fsun1, (void *)&fsun2, sizeof(fsun2));
2114 maxino = sblock.fs_ncg * sblock.fs_ipg;
2116 DBG_OPEN("/tmp/growfs.debug"); /* already here we need a superblock */
2117 DBG_DUMP_FS(&sblock,
2121 * Determine size to grow to. Default to the full size specified in
2124 sblock.fs_size = dbtofsb(&osblock, p_size);
2127 errx(1, "there is not enough space (%d < %d)",
2130 sblock.fs_size = dbtofsb(&osblock, size);
2134 * Are we really growing ?
2136 if(osblock.fs_size >= sblock.fs_size) {
2137 errx(1, "we are not growing (%jd->%jd)",
2138 (intmax_t)osblock.fs_size, (intmax_t)sblock.fs_size);
2144 * Check if we find an active snapshot.
2146 if(ExpertFlag == 0) {
2147 for(j=0; j<FSMAXSNAP; j++) {
2148 if(sblock.fs_snapinum[j]) {
2149 errx(1, "active snapshot found in file system\n"
2150 " please remove all snapshots before "
2153 if(!sblock.fs_snapinum[j]) { /* list is dense */
2160 if (ExpertFlag == 0 && Nflag == 0) {
2161 printf("We strongly recommend you to make a backup "
2162 "before growing the Filesystem\n\n"
2163 " Did you backup your data (Yes/No) ? ");
2164 fgets(reply, (int)sizeof(reply), stdin);
2165 if (strcasecmp(reply, "Yes\n")){
2166 printf("\n Nothing done \n");
2171 printf("new file systemsize is: %jd frags\n", (intmax_t)sblock.fs_size);
2174 * Try to access our new last block in the file system. Even if we
2175 * later on realize we have to abort our operation, on that block
2176 * there should be no data, so we can't destroy something yet.
2178 wtfs((ufs2_daddr_t)p_size-1, (size_t)DEV_BSIZE, (void *)&sblock,
2182 * Now calculate new superblock values and check for reasonable
2183 * bound for new file system size:
2184 * fs_size: is derived from label or user input
2185 * fs_dsize: should get updated in the routines creating or
2186 * updating the cylinder groups on the fly
2187 * fs_cstotal: should get updated in the routines creating or
2188 * updating the cylinder groups
2192 * Update the number of cylinders and cylinder groups in the file system.
2194 if (sblock.fs_magic == FS_UFS1_MAGIC) {
2195 sblock.fs_old_ncyl =
2196 sblock.fs_size * sblock.fs_old_nspf / sblock.fs_old_spc;
2197 if (sblock.fs_size * sblock.fs_old_nspf >
2198 sblock.fs_old_ncyl * sblock.fs_old_spc)
2199 sblock.fs_old_ncyl++;
2201 sblock.fs_ncg = howmany(sblock.fs_size, sblock.fs_fpg);
2202 maxino = sblock.fs_ncg * sblock.fs_ipg;
2204 if (sblock.fs_size % sblock.fs_fpg != 0 &&
2205 sblock.fs_size % sblock.fs_fpg < cgdmin(&sblock, sblock.fs_ncg)) {
2207 * The space in the new last cylinder group is too small,
2211 if (sblock.fs_magic == FS_UFS1_MAGIC)
2212 sblock.fs_old_ncyl = sblock.fs_ncg * sblock.fs_old_cpg;
2213 printf("Warning: %jd sector(s) cannot be allocated.\n",
2214 (intmax_t)fsbtodb(&sblock, sblock.fs_size % sblock.fs_fpg));
2215 sblock.fs_size = sblock.fs_ncg * sblock.fs_fpg;
2216 maxino -= sblock.fs_ipg;
2220 * Update the space for the cylinder group summary information in the
2221 * respective cylinder group data area.
2224 fragroundup(&sblock, sblock.fs_ncg * sizeof(struct csum));
2226 if(osblock.fs_size >= sblock.fs_size) {
2227 errx(1, "not enough new space");
2230 DBG_PRINT0("sblock calculated\n");
2233 * Ok, everything prepared, so now let's do the tricks.
2235 growfs(fsi, fso, Nflag);
2238 * Update the disk label.
2241 pp->p_fsize = sblock.fs_fsize;
2242 pp->p_frag = sblock.fs_frag;
2243 pp->p_cpg = sblock.fs_fpg;
2245 return_disklabel(fso, lp, Nflag);
2246 DBG_PRINT0("label rewritten\n");
2250 if(fso>-1) close(fso);
2258 /* ************************************************** return_disklabel ***** */
2260 * Write the updated disklabel back to disk.
2263 return_disklabel(int fd, struct disklabel *lp, unsigned int Nflag)
2265 DBG_FUNC("return_disklabel")
2281 * recalculate checksum
2283 while(ptr < (u_short *)&lp->d_partitions[lp->d_npartitions]) {
2288 if (ioctl(fd, DIOCWDINFO, (char *)lp) < 0) {
2289 errx(1, "DIOCWDINFO failed");
2298 /* ***************************************************** get_disklabel ***** */
2300 * Read the disklabel from disk.
2302 static struct disklabel *
2303 get_disklabel(int fd)
2305 DBG_FUNC("get_disklabel")
2306 static struct disklabel *lab;
2310 lab=(struct disklabel *)malloc(sizeof(struct disklabel));
2312 errx(1, "malloc failed");
2314 if (!ioctl(fd, DIOCGDINFO, (char *)lab))
2324 /* ************************************************************* usage ***** */
2326 * Dump a line of usage.
2335 fprintf(stderr, "usage: growfs [-Ny] [-s size] special\n");
2341 /* *********************************************************** updclst ***** */
2343 * This updates most parameters and the bitmap related to cluster. We have to
2344 * assume that sblock, osblock, acg are set up.
2354 if(sblock.fs_contigsumsize < 1) { /* no clustering */
2358 * update cluster allocation map
2360 setbit(cg_clustersfree(&acg), block);
2363 * update cluster summary table
2367 * calculate size for the trailing cluster
2369 for(block--; lcs<sblock.fs_contigsumsize; block--, lcs++ ) {
2370 if(isclr(cg_clustersfree(&acg), block)){
2375 if(lcs < sblock.fs_contigsumsize) {
2377 cg_clustersum(&acg)[lcs]--;
2380 cg_clustersum(&acg)[lcs]++;
2387 /* *********************************************************** updrefs ***** */
2389 * This updates all references to relocated blocks for the given inode. The
2390 * inode is given as number within the cylinder group, and the number of the
2394 updrefs(int cg, ino_t in, struct gfs_bpp *bp, int fsi, int fso, unsigned int
2398 ufs_lbn_t len, lbn, numblks;
2399 ufs2_daddr_t iptr, blksperindir;
2401 int i, mode, inodeupdated;
2405 ino = ginode(in, fsi, cg);
2410 mode = DIP(ino, di_mode) & IFMT;
2411 if (mode != IFDIR && mode != IFREG && mode != IFLNK) {
2413 return; /* only check DIR, FILE, LINK */
2415 if (mode == IFLNK &&
2416 DIP(ino, di_size) < (u_int64_t) sblock.fs_maxsymlinklen) {
2418 return; /* skip short symlinks */
2420 numblks = howmany(DIP(ino, di_size), sblock.fs_bsize);
2423 return; /* skip empty file */
2425 if (DIP(ino, di_blocks) == 0) {
2427 return; /* skip empty swiss cheesy file or old fastlink */
2429 DBG_PRINT2("scg checking inode (%d in %d)\n",
2434 * Check all the blocks.
2437 len = numblks < NDADDR ? numblks : NDADDR;
2438 for (i = 0; i < len; i++) {
2439 iptr = DIP(ino, di_db[i]);
2442 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2443 DIP_SET(ino, di_db[i], iptr);
2447 DBG_PRINT0("~~scg direct blocks checked\n");
2450 len = numblks - NDADDR;
2452 for (i = 0; len > 0 && i < NIADDR; i++) {
2453 iptr = DIP(ino, di_ib[i]);
2456 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2457 DIP_SET(ino, di_ib[i], iptr);
2460 indirchk(blksperindir, lbn, iptr, numblks, bp, fsi, fso, Nflag);
2461 blksperindir *= NINDIR(&sblock);
2462 lbn += blksperindir;
2463 len -= blksperindir;
2464 DBG_PRINT1("scg indirect_%d blocks checked\n", i + 1);
2467 wtfs(inoblk, sblock.fs_bsize, inobuf, fso, Nflag);
2474 * Recursively check all the indirect blocks.
2477 indirchk(ufs_lbn_t blksperindir, ufs_lbn_t lbn, ufs2_daddr_t blkno,
2478 ufs_lbn_t lastlbn, struct gfs_bpp *bp, int fsi, int fso, unsigned int Nflag)
2480 DBG_FUNC("indirchk")
2487 /* read in the indirect block. */
2488 ibuf = malloc(sblock.fs_bsize);
2490 errx(1, "malloc failed");
2491 rdfs(fsbtodb(&sblock, blkno), (size_t)sblock.fs_bsize, ibuf, fsi);
2492 last = howmany(lastlbn - lbn, blksperindir) < NINDIR(&sblock) ?
2493 howmany(lastlbn - lbn, blksperindir) : NINDIR(&sblock);
2494 for (i = 0; i < last; i++) {
2495 if (sblock.fs_magic == FS_UFS1_MAGIC)
2496 iptr = ((ufs1_daddr_t *)ibuf)[i];
2498 iptr = ((ufs2_daddr_t *)ibuf)[i];
2501 if (cond_bl_upd(&iptr, bp, fsi, fso, Nflag)) {
2502 if (sblock.fs_magic == FS_UFS1_MAGIC)
2503 ((ufs1_daddr_t *)ibuf)[i] = iptr;
2505 ((ufs2_daddr_t *)ibuf)[i] = iptr;
2507 if (blksperindir == 1)
2509 indirchk(blksperindir / NINDIR(&sblock), lbn + blksperindir * i,
2510 iptr, lastlbn, bp, fsi, fso, Nflag);