/* $OpenBSD: growfs.c,v 1.11 2004/03/15 08:52:01 deraadt Exp $ */ /* * Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz * Copyright (c) 1980, 1989, 1993 The Regents of the University of California. * All rights reserved. * * This code is derived from software contributed to Berkeley by * Christoph Herrmann and Thomas-Henning von Kamptz, Munich and Frankfurt. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgment: * This product includes software developed by the University of * California, Berkeley and its contributors, as well as Christoph * Herrmann and Thomas-Henning von Kamptz. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $TSHeader: src/sbin/growfs/growfs.c,v 1.5 2000/12/12 19:31:00 tomsoft Exp $ * */ #ifndef lint static const char copyright[] = "@(#) Copyright (c) 2000 Christoph Herrmann, Thomas-Henning von Kamptz\n\ Copyright (c) 1980, 1989, 1993 The Regents of the University of California.\n\ All rights reserved.\n"; static const char rcsid[] = "$OpenBSD: growfs.c,v 1.11 2004/03/15 08:52:01 deraadt Exp $"; #endif /* not lint */ /* ********************************************************** INCLUDES ***** */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "debug.h" /* *************************************************** GLOBALS & TYPES ***** */ #ifdef FS_DEBUG int _dbg_lvl_ = (DL_INFO); /* DL_TRC */ #endif /* FS_DEBUG */ static union { struct fs fs; char pad[SBSIZE]; } fsun1, fsun2; #define sblock fsun1.fs /* the new superblock */ #define osblock fsun2.fs /* the old superblock */ static union { struct cg cg; char pad[MAXBSIZE]; } cgun1, cgun2; #define acg cgun1.cg /* a cylinder cgroup (new) */ #define aocg cgun2.cg /* an old cylinder group */ static char ablk[MAXBSIZE]; /* a block */ static char i1blk[MAXBSIZE]; /* some indirect blocks */ static char i2blk[MAXBSIZE]; static char i3blk[MAXBSIZE]; /* where to write back updated blocks */ static daddr_t in_src, i1_src, i2_src, i3_src; /* what object contains the reference */ enum pointer_source { GFS_PS_INODE, GFS_PS_IND_BLK_LVL1, GFS_PS_IND_BLK_LVL2, GFS_PS_IND_BLK_LVL3 }; static struct csum *fscs; /* cylinder summary */ static struct ufs1_dinode zino[MAXBSIZE / sizeof(struct ufs1_dinode)]; /* some inodes */ /* * An array of elements of type struct gfs_bpp describes all blocks to * be relocated in order to free the space needed for the cylinder group * summary for all cylinder groups located in the first cylinder group. */ struct gfs_bpp { daddr_t old; /* old block number */ daddr_t new; /* new block number */ #define GFS_FL_FIRST 1 #define GFS_FL_LAST 2 unsigned int flags; /* special handling required */ int found; /* how many references were updated */ }; /* ******************************************************** PROTOTYPES ***** */ static void growfs(int, int, unsigned int); static void rdfs(daddr_t, size_t, void *, int); static void wtfs(daddr_t, size_t, void *, int, unsigned int); static daddr_t alloc(void); static int charsperline(void); static void usage(void); static int isblock(struct fs *, unsigned char *, int); static void clrblock(struct fs *, unsigned char *, int); static void setblock(struct fs *, unsigned char *, int); static void initcg(int, time_t, int, unsigned int); static void updjcg(int, time_t, int, int, unsigned int); static void updcsloc(time_t, int, int, unsigned int); static struct disklabel *get_disklabel(int); static void return_disklabel(int, struct disklabel *, unsigned int); static struct ufs1_dinode *ginode(ino_t, int, int); static void frag_adjust(daddr_t, int); static void cond_bl_upd(ufs_daddr_t *, struct gfs_bpp *, enum pointer_source, int, unsigned int); static void updclst(int); static void updrefs(int, ino_t, struct gfs_bpp *, int, int, unsigned int); /* ************************************************************ growfs ***** */ /* * Here we actually start growing the filesystem. We basically read the * cylinder summary from the first cylinder group as we want to update * this on the fly during our various operations. First we handle the * changes in the former last cylinder group. Afterwards we create all new * cylinder groups. Now we handle the cylinder group containing the * cylinder summary which might result in a relocation of the whole * structure. In the end we write back the updated cylinder summary, the * new superblock, and slightly patched versions of the super block * copies. */ static void growfs(int fsi, int fso, unsigned int Nflag) { DBG_FUNC("growfs") int i; int cylno, j; time_t utime; int width; char tmpbuf[100]; DBG_ENTER; time(&utime); /* * Get the cylinder summary into the memory. */ fscs = calloc((size_t)1, (size_t)sblock.fs_cssize); if (fscs == NULL) errx(1, "calloc failed"); for (i = 0; i < osblock.fs_cssize; i += osblock.fs_bsize) { rdfs(fsbtodb(&osblock, osblock.fs_csaddr + numfrags(&osblock, i)), (size_t)MIN(osblock.fs_cssize - i, osblock.fs_bsize), (((char *)fscs) + i), fsi); } #ifdef FS_DEBUG { struct csum *dbg_csp; int dbg_csc; char dbg_line[80]; dbg_csp = fscs; for (dbg_csc = 0; dbg_csc < osblock.fs_ncg; dbg_csc++) { snprintf(dbg_line, sizeof(dbg_line), "%d. old csum in old location", dbg_csc); DBG_DUMP_CSUM(&osblock, dbg_line, dbg_csp++); } } #endif /* FS_DEBUG */ DBG_PRINT0("fscs read\n"); /* * Do all needed changes in the former last cylinder group. */ updjcg(osblock.fs_ncg - 1, utime, fsi, fso, Nflag); /* * Dump out summary information about filesystem. */ printf("growfs:\t%d sectors in %d %s of %d tracks, %d sectors\n", sblock.fs_size * NSPF(&sblock), sblock.fs_ncyl, "cylinders", sblock.fs_ntrak, sblock.fs_nsect); #define B2MBFACTOR (1 / (1024.0 * 1024.0)) printf("\t%.1fMB in %d cyl groups (%d c/g, %.2fMB/g, %d i/g)\n", (float)sblock.fs_size * sblock.fs_fsize * B2MBFACTOR, sblock.fs_ncg, sblock.fs_cpg, (float)sblock.fs_fpg * sblock.fs_fsize * B2MBFACTOR, sblock.fs_ipg); #undef B2MBFACTOR /* * Now build the cylinders group blocks and * then print out indices of cylinder groups. */ printf("superblock backups (for fsck -b #) at:\n"); i = 0; width = charsperline(); /* * Iterate for only the new cylinder groups. */ for (cylno = osblock.fs_ncg; cylno < sblock.fs_ncg; cylno++) { initcg(cylno, utime, fso, Nflag); j = snprintf(tmpbuf, sizeof tmpbuf, " %d%s", (int)fsbtodb(&sblock, cgsblock(&sblock, cylno)), cylno < (sblock.fs_ncg - 1) ? "," : ""); if (i + j >= width) { printf("\n"); i = 0; } i += j; printf("%s", tmpbuf); fflush(stdout); } printf("\n"); /* * Do all needed changes in the first cylinder group. * allocate blocks in new location */ updcsloc(utime, fsi, fso, Nflag); /* * Now write the cylinder summary back to disk. */ for (i = 0; i < sblock.fs_cssize; i += sblock.fs_bsize) { wtfs(fsbtodb(&sblock, sblock.fs_csaddr + numfrags(&sblock, i)), (size_t)MIN(sblock.fs_cssize - i, sblock.fs_bsize), (((char *)fscs) + i), fso, Nflag); } DBG_PRINT0("fscs written\n"); #ifdef FS_DEBUG { struct csum *dbg_csp; int dbg_csc; char dbg_line[80]; dbg_csp = fscs; for (dbg_csc = 0; dbg_csc < sblock.fs_ncg; dbg_csc++) { snprintf(dbg_line, sizeof(dbg_line), "%d. new csum in new location", dbg_csc); DBG_DUMP_CSUM(&sblock, dbg_line, dbg_csp++); } } #endif /* FS_DEBUG */ /* * Now write the new superblock back to disk. */ sblock.fs_time = utime; wtfs((daddr_t)(SBOFF / DEV_BSIZE), (size_t)SBSIZE, &sblock, fso, Nflag); DBG_PRINT0("sblock written\n"); DBG_DUMP_FS(&sblock, "new initial sblock"); /* * Clean up the dynamic fields in our superblock copies. */ sblock.fs_fmod = 0; sblock.fs_clean = 1; sblock.fs_ronly = 0; sblock.fs_cgrotor = 0; sblock.fs_state = 0; memset(&sblock.fs_fsmnt, 0, sizeof(sblock.fs_fsmnt)); sblock.fs_flags &= FS_DOSOFTDEP; /* * XXX * The following fields are currently distributed from the superblock * to the copies: * fs_minfree * fs_rotdelay * fs_maxcontig * fs_maxbpg * fs_minfree, * fs_optim * fs_flags regarding SOFTUPDATES * * We probably should rather change the summary for the cylinder group * statistics here to the value of what would be in there, if the file * system were created initially with the new size. Therefor we still * need to find an easy way of calculating that. * Possibly we can try to read the first superblock copy and apply the * "diffed" stats between the old and new superblock by still copying * certain parameters onto that. */ /* * Write out the duplicate superblocks. */ for (cylno = 0; cylno < sblock.fs_ncg; cylno++) { wtfs(fsbtodb(&sblock, cgsblock(&sblock, cylno)), (size_t)SBSIZE, &sblock, fso, Nflag); } DBG_PRINT0("sblock copies written\n"); DBG_DUMP_FS(&sblock, "new other sblocks"); DBG_LEAVE; } /* ************************************************************ initcg ***** */ /* * This creates a new cylinder group structure, for more details please see * the source of newfs(8), as this function is taken over almost unchanged. * As this is never called for the first cylinder group, the special * provisions for that case are removed here. */ static void initcg(int cylno, time_t utime, int fso, unsigned int Nflag) { DBG_FUNC("initcg") daddr_t cbase, d, dlower, dupper, dmax, blkno; int i; struct csum *cs; int j; DBG_ENTER; /* * Determine block bounds for cylinder group. */ cbase = cgbase(&sblock, cylno); dmax = cbase + sblock.fs_fpg; if (dmax > sblock.fs_size) dmax = sblock.fs_size; dlower = cgsblock(&sblock, cylno) - cbase; dupper = cgdmin(&sblock, cylno) - cbase; if (cylno == 0) /* XXX fscs may be relocated */ dupper += howmany(sblock.fs_cssize, sblock.fs_fsize); cs = fscs + cylno; memset(&acg, 0, (size_t)sblock.fs_cgsize); acg.cg_time = utime; acg.cg_magic = CG_MAGIC; acg.cg_cgx = cylno; if (cylno == sblock.fs_ncg - 1) acg.cg_ncyl = sblock.fs_ncyl % sblock.fs_cpg; else acg.cg_ncyl = sblock.fs_cpg; acg.cg_niblk = sblock.fs_ipg; acg.cg_ndblk = dmax - cbase; if (sblock.fs_contigsumsize > 0) acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag; acg.cg_btotoff = &acg.cg_space[0] - (u_char *)(&acg.cg_firstfield); acg.cg_boff = acg.cg_btotoff + sblock.fs_cpg * sizeof(int32_t); acg.cg_iusedoff = acg.cg_boff + sblock.fs_cpg * sblock.fs_nrpos * sizeof(u_int16_t); acg.cg_freeoff = acg.cg_iusedoff + howmany(sblock.fs_ipg, NBBY); if (sblock.fs_contigsumsize <= 0) { acg.cg_nextfreeoff = acg.cg_freeoff + howmany(sblock.fs_cpg * sblock.fs_spc / NSPF(&sblock), NBBY); } else { acg.cg_clustersumoff = acg.cg_freeoff + howmany (sblock.fs_cpg * sblock.fs_spc / NSPF(&sblock), NBBY) - sizeof(u_int32_t); acg.cg_clustersumoff = roundup(acg.cg_clustersumoff, sizeof(u_int32_t)); acg.cg_clusteroff = acg.cg_clustersumoff + (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t); acg.cg_nextfreeoff = acg.cg_clusteroff + howmany (sblock.fs_cpg * sblock.fs_spc / NSPB(&sblock), NBBY); } if (acg.cg_nextfreeoff-(int)(&acg.cg_firstfield) > sblock.fs_cgsize) { /* * XXX This should never happen as we would have had that panic * already on filesystem creation */ errx(37, "panic: cylinder group too big"); } acg.cg_cs.cs_nifree += sblock.fs_ipg; if (cylno == 0) for (i = 0; (size_t)i < ROOTINO; i++) { setbit(cg_inosused(&acg), i); acg.cg_cs.cs_nifree--; } for (i = 0; i < sblock.fs_ipg / INOPF(&sblock); i += sblock.fs_frag) { for (j = 0; (unsigned)j < sblock.fs_bsize / sizeof(struct ufs1_dinode); j++) zino[j].di_gen = arc4random(); wtfs(fsbtodb(&sblock, cgimin(&sblock, cylno) + i), (size_t)sblock.fs_bsize, zino, fso, Nflag); } for (d = 0; d < dlower; d += sblock.fs_frag) { blkno = d / sblock.fs_frag; setblock(&sblock, cg_blksfree(&acg), blkno); if (sblock.fs_contigsumsize > 0) { setbit(cg_clustersfree(&acg), blkno); } acg.cg_cs.cs_nbfree++; cg_blktot(&acg)[cbtocylno(&sblock, d)]++; cg_blks(&sblock, &acg, cbtocylno(&sblock, d)) [cbtorpos(&sblock, d)]++; } sblock.fs_dsize += dlower; sblock.fs_dsize += acg.cg_ndblk - dupper; if ((i = dupper % sblock.fs_frag)) { acg.cg_frsum[sblock.fs_frag - i]++; for (d = dupper + sblock.fs_frag - i; dupper < d; dupper++) { setbit(cg_blksfree(&acg), dupper); acg.cg_cs.cs_nffree++; } } for (d = dupper; d + sblock.fs_frag <= dmax - cbase;) { blkno = d / sblock.fs_frag; setblock(&sblock, cg_blksfree(&acg), blkno); if (sblock.fs_contigsumsize > 0) { setbit(cg_clustersfree(&acg), blkno); } acg.cg_cs.cs_nbfree++; cg_blktot(&acg)[cbtocylno(&sblock, d)]++; cg_blks(&sblock, &acg, cbtocylno(&sblock, d)) [cbtorpos(&sblock, d)]++; d += sblock.fs_frag; } if (d < dmax - cbase) { acg.cg_frsum[dmax - cbase - d]++; for (; d < dmax - cbase; d++) { setbit(cg_blksfree(&acg), d); acg.cg_cs.cs_nffree++; } } if (sblock.fs_contigsumsize > 0) { int32_t *sump = cg_clustersum(&acg); u_char *mapp = cg_clustersfree(&acg); int map = *mapp++; int bit = 1; int run = 0; for (i = 0; i < acg.cg_nclusterblks; i++) { if ((map & bit) != 0) run++; else if (run != 0) { if (run > sblock.fs_contigsumsize) run = sblock.fs_contigsumsize; sump[run]++; run = 0; } if ((i & (NBBY - 1)) != (NBBY - 1)) bit <<= 1; else { map = *mapp++; bit = 1; } } if (run != 0) { if (run > sblock.fs_contigsumsize) { run = sblock.fs_contigsumsize; } sump[run]++; } } sblock.fs_cstotal.cs_ndir += acg.cg_cs.cs_ndir; sblock.fs_cstotal.cs_nffree += acg.cg_cs.cs_nffree; sblock.fs_cstotal.cs_nbfree += acg.cg_cs.cs_nbfree; sblock.fs_cstotal.cs_nifree += acg.cg_cs.cs_nifree; *cs = acg.cg_cs; wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_bsize, &acg, fso, Nflag); DBG_DUMP_CG(&sblock, "new cg", &acg); DBG_LEAVE; } /* ******************************************************* frag_adjust ***** */ /* * Here we add or subtract (sign +1/-1) the available fragments in a given * block to or from the fragment statistics. By subtracting before and adding * after an operation on the free frag map we can easy update the fragment * statistic, which seems to be otherwise an rather complex operation. */ static void frag_adjust(daddr_t frag, int sign) { DBG_FUNC("frag_adjust") int fragsize; int f; DBG_ENTER; fragsize = 0; /* * Here frag only needs to point to any fragment in the block we want * to examine. */ #define rounddown(x, y) (((x)/(y))*(y)) for (f = rounddown(frag, sblock.fs_frag); f < roundup(frag + 1, sblock.fs_frag); f++) { /* * Count contiguous free fragments. */ if (isset(cg_blksfree(&acg), f)) fragsize++; else { if (fragsize && fragsize < sblock.fs_frag) { /* * We found something in between. */ acg.cg_frsum[fragsize]+=sign; DBG_PRINT2("frag_adjust [%d]+=%d\n", fragsize, sign); } fragsize = 0; } } if (fragsize && fragsize < sblock.fs_frag) { /* * We found something. */ acg.cg_frsum[fragsize]+=sign; DBG_PRINT2("frag_adjust [%d]+=%d\n", fragsize, sign); } DBG_PRINT2("frag_adjust [[%d]]+=%d\n", fragsize, sign); DBG_LEAVE; } /* ******************************************************* cond_bl_upd ***** */ /* * Here we conditionally update a pointer to a fragment. We check for all * relocated blocks if any of it's fragments is referenced by the current * field, and update the pointer to the respective fragment in our new * block. If we find a reference we write back the block immediately, * as there is no easy way for our general block reading engine to figure * out if a write back operation is needed. */ static void cond_bl_upd(ufs_daddr_t *block, struct gfs_bpp *field, enum pointer_source source, int fso, unsigned int Nflag) { DBG_FUNC("cond_bl_upd") struct gfs_bpp *f; char *src; daddr_t dst = 0; DBG_ENTER; f = field; while (f->old) { /* for all old blocks */ if (*block/sblock.fs_frag == f->old) { /* * The fragment is part of the block, so update. */ *block = (f->new * sblock.fs_frag + (*block % sblock.fs_frag)); f->found++; DBG_PRINT3("scg (%d->%d)[%d] reference updated\n", f->old, f->new, *block % sblock.fs_frag); /* Write the block back to disk immediately */ switch (source) { case GFS_PS_INODE: src = ablk; dst = in_src; break; case GFS_PS_IND_BLK_LVL1: src = i1blk; dst = i1_src; break; case GFS_PS_IND_BLK_LVL2: src = i2blk; dst = i2_src; break; case GFS_PS_IND_BLK_LVL3: src = i3blk; dst = i3_src; break; default: /* error */ src = NULL; break; } if (src) { /* * XXX If src is not of type inode we have to * implement copy on write here in case * of active snapshots. */ wtfs(dst, (size_t)sblock.fs_bsize, src, fso, Nflag); } /* * The same block can't be found again in this loop. */ break; } f++; } DBG_LEAVE; } /* ************************************************************ updjcg ***** */ /* * Here we do all needed work for the former last cylinder group. It has to be * changed in any case, even if the filesystem ended exactly on the end of * this group, as there is some slightly inconsistent handling of the number * of cylinders in the cylinder group. We start again by reading the cylinder * group from disk. If the last block was not fully available, we first handle * the missing fragments, then we handle all new full blocks in that file * system and finally we handle the new last fragmented block in the file * system. We again have to handle the fragment statistics rotational layout * tables and cluster summary during all those operations. */ static void updjcg(int cylno, time_t utime, int fsi, int fso, unsigned int Nflag) { DBG_FUNC("updjcg") daddr_t cbase, dmax, dupper; struct csum *cs; int i, k; int j = 0; DBG_ENTER; /* * Read the former last (joining) cylinder group from disk, and make * a copy. */ rdfs(fsbtodb(&osblock, cgtod(&osblock, cylno)), (size_t)osblock.fs_cgsize, &aocg, fsi); DBG_PRINT0("jcg read\n"); DBG_DUMP_CG(&sblock, "old joining cg", &aocg); memcpy(&cgun1, &cgun2, sizeof(cgun2)); /* * If the cylinder group had already it's new final size almost * nothing is to be done ... except: * For some reason the value of cg_ncyl in the last cylinder group has * to be zero instead of fs_cpg. As this is now no longer the last * cylinder group we have to change that value now to fs_cpg. */ if (cgbase(&osblock, cylno + 1) == osblock.fs_size) { acg.cg_ncyl = sblock.fs_cpg; wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize, &acg, fso, Nflag); DBG_PRINT0("jcg written\n"); DBG_DUMP_CG(&sblock, "new joining cg", &acg); DBG_LEAVE; return; } /* * Set up some variables needed later. */ cbase = cgbase(&sblock, cylno); dmax = cbase + sblock.fs_fpg; if (dmax > sblock.fs_size) dmax = sblock.fs_size; dupper = cgdmin(&sblock, cylno) - cbase; if (cylno == 0) /* XXX fscs may be relocated */ dupper += howmany(sblock.fs_cssize, sblock.fs_fsize); /* * Set pointer to the cylinder summary for our cylinder group. */ cs = fscs + cylno; /* * Touch the cylinder group, update all fields in the cylinder group as * needed, update the free space in the superblock. */ acg.cg_time = utime; if (cylno == sblock.fs_ncg - 1) { /* * This is still the last cylinder group. */ acg.cg_ncyl = sblock.fs_ncyl % sblock.fs_cpg; } else { acg.cg_ncyl = sblock.fs_cpg; } DBG_PRINT4("jcg dbg: %d %u %d %u\n", cylno, sblock.fs_ncg, acg.cg_ncyl, sblock.fs_cpg); acg.cg_ndblk = dmax - cbase; sblock.fs_dsize += acg.cg_ndblk - aocg.cg_ndblk; if (sblock.fs_contigsumsize > 0) acg.cg_nclusterblks = acg.cg_ndblk / sblock.fs_frag; /* * Now we have to update the free fragment bitmap for our new free * space. There again we have to handle the fragmentation and also * the rotational layout tables and the cluster summary. This is * also done per fragment for the first new block if the old file * system end was not on a block boundary, per fragment for the new * last block if the new filesystem end is not on a block boundary, * and per block for all space in between. * * Handle the first new block here if it was partially available * before. */ if (osblock.fs_size % sblock.fs_frag) { if (roundup(osblock.fs_size, sblock.fs_frag)<=sblock.fs_size) { /* * The new space is enough to fill at least this * block */ j = 0; for (i = roundup(osblock.fs_size - cbase, sblock.fs_frag)-1; i >= osblock.fs_size - cbase; i--) { setbit(cg_blksfree(&acg), i); acg.cg_cs.cs_nffree++; j++; } /* * Check if the fragment just created could join an * already existing fragment at the former end of the * filesystem. */ if (isblock(&sblock, cg_blksfree(&acg), ((osblock.fs_size - cgbase(&sblock, cylno))/ sblock.fs_frag))) { /* * The block is now completely available */ DBG_PRINT0("block was\n"); acg.cg_frsum[osblock.fs_size % sblock.fs_frag]--; acg.cg_cs.cs_nbfree++; acg.cg_cs.cs_nffree-=sblock.fs_frag; k = rounddown(osblock.fs_size - cbase, sblock.fs_frag); cg_blktot(&acg)[cbtocylno(&sblock, k)]++; cg_blks(&sblock, &acg, cbtocylno(&sblock, k)) [cbtorpos(&sblock, k)]++; updclst((osblock.fs_size - cbase)/sblock.fs_frag); } else { /* * Lets rejoin a possible partially growed * fragment. */ k = 0; while (isset(cg_blksfree(&acg), i) && (i >= rounddown(osblock.fs_size - cbase, sblock.fs_frag))) { i--; k++; } if (k) acg.cg_frsum[k]--; acg.cg_frsum[k + j]++; } } else { /* * We only grow by some fragments within this last * block. */ for (i = sblock.fs_size - cbase - 1; i >= osblock.fs_size - cbase; i--) { setbit(cg_blksfree(&acg), i); acg.cg_cs.cs_nffree++; j++; } /* * Lets rejoin a possible partially growed fragment. */ k = 0; while (isset(cg_blksfree(&acg), i) && (i >= rounddown(osblock.fs_size - cbase, sblock.fs_frag))) { i--; k++; } if (k) acg.cg_frsum[k]--; acg.cg_frsum[k + j]++; } } /* * Handle all new complete blocks here. */ for (i = roundup(osblock.fs_size - cbase, sblock.fs_frag); i + sblock.fs_frag<=dmax-cbase; /* XXX <= or only < ? */ i += sblock.fs_frag) { j = i / sblock.fs_frag; setblock(&sblock, cg_blksfree(&acg), j); updclst(j); acg.cg_cs.cs_nbfree++; cg_blktot(&acg)[cbtocylno(&sblock, i)]++; cg_blks(&sblock, &acg, cbtocylno(&sblock, i)) [cbtorpos(&sblock, i)]++; } /* * Handle the last new block if there are stll some new fragments left. * Here we don't have to bother about the cluster summary or the even * the rotational layout table. */ if (i < (dmax - cbase)) { acg.cg_frsum[dmax - cbase - i]++; for (; i < dmax - cbase; i++) { setbit(cg_blksfree(&acg), i); acg.cg_cs.cs_nffree++; } } sblock.fs_cstotal.cs_nffree += (acg.cg_cs.cs_nffree - aocg.cg_cs.cs_nffree); sblock.fs_cstotal.cs_nbfree += (acg.cg_cs.cs_nbfree - aocg.cg_cs.cs_nbfree); /* * The following statistics are not changed here: * sblock.fs_cstotal.cs_ndir * sblock.fs_cstotal.cs_nifree * As the statistics for this cylinder group are ready, copy it to * the summary information array. */ *cs = acg.cg_cs; /* * Write the updated "joining" cylinder group back to disk. */ wtfs(fsbtodb(&sblock, cgtod(&sblock, cylno)), (size_t)sblock.fs_cgsize, &acg, fso, Nflag); DBG_PRINT0("jcg written\n"); DBG_DUMP_CG(&sblock, "new joining cg", &acg); DBG_LEAVE; } /* ********************************************************** updcsloc ***** */ /* * Here we update the location of the cylinder summary. We have two possible * ways of growing the cylinder summary. * (1) We can try to grow the summary in the current location, and relocate * possibly used blocks within the current cylinder group. * (2) Alternatively we can relocate the whole cylinder summary to the first * new completely empty cylinder group. Once the cylinder summary is no * longer in the beginning of the first cylinder group you should never * use a version of fsck which is not aware of the possibility to have * this structure in a non standard place. * Option (1) is considered to be less intrusive to the structure of the file- * system. So we try to stick to that whenever possible. If there is not enough * space in the cylinder group containing the cylinder summary we have to use * method (2). In case of active snapshots in the filesystem we probably can * completely avoid implementing copy on write if we stick to method (2) only. */ static void updcsloc(time_t utime, int fsi, int fso, unsigned int Nflag) { DBG_FUNC("updcsloc") struct csum *cs; int ocscg, ncscg; int blocks; daddr_t cbase, dupper, odupper, d, f, g; int ind; int cylno, inc; struct gfs_bpp *bp; int i, l; int lcs = 0; int block; DBG_ENTER; if (howmany(sblock.fs_cssize, sblock.fs_fsize) == howmany(osblock.fs_cssize, osblock.fs_fsize)) { /* * No new fragment needed. */ DBG_LEAVE; return; } ocscg = dtog(&osblock, osblock.fs_csaddr); cs = fscs + ocscg; blocks = 1+howmany(sblock.fs_cssize, sblock.fs_bsize)- howmany(osblock.fs_cssize, osblock.fs_bsize); /* * Read original cylinder group from disk, and make a copy. * XXX If Nflag is set in some very rare cases we now miss * some changes done in updjcg by reading the unmodified * block from disk. */ rdfs(fsbtodb(&osblock, cgtod(&osblock, ocscg)), (size_t)osblock.fs_cgsize, &aocg, fsi); DBG_PRINT0("oscg read\n"); DBG_DUMP_CG(&sblock, "old summary cg", &aocg); memcpy(&cgun1, &cgun2, sizeof(cgun2)); /* * Touch the cylinder group, set up local variables needed later * and update the superblock. */ acg.cg_time = utime; /* * XXX In the case of having active snapshots we may need much more * blocks for the copy on write. We need each block twice, and * also up to 8*3 blocks for indirect blocks for all possible * references. */ if (/* ((int)sblock.fs_time & 0x3) > 0 || */ cs->cs_nbfree < blocks) { /* * There is not enough space in the old cylinder group to * relocate all blocks as needed, so we relocate the whole * cylinder group summary to a new group. We try to use the * first complete new cylinder group just created. Within the * cylinder group we align the area immediately after the * cylinder group information location in order to be as * close as possible to the original implementation of ffs. * * First we have to make sure we'll find enough space in the * new cylinder group. If not, then we currently give up. * We start with freeing everything which was used by the * fragments of the old cylinder summary in the current group. * Now we write back the group meta data, read in the needed * meta data from the new cylinder group, and start allocating * within that group. Here we can assume, the group to be * completely empty. Which makes the handling of fragments and * clusters a lot easier. */ DBG_TRC; if (sblock.fs_ncg - osblock.fs_ncg < 2) errx(2, "panic: not enough space"); /* * Point "d" to the first fragment not used by the cylinder * summary. */ d = osblock.fs_csaddr+(osblock.fs_cssize/osblock.fs_fsize); /* * Set up last cluster size ("lcs") already here. Calculate * the size for the trailing cluster just behind where "d" * points to. */ if (sblock.fs_contigsumsize > 0) { for (block = howmany(d % sblock.fs_fpg, sblock.fs_frag), lcs = 0; lcs < sblock.fs_contigsumsize; block++, lcs++) { if (isclr(cg_clustersfree(&acg), block)) break; } } /* * Point "d" to the last frag used by the cylinder summary. */ d--; DBG_PRINT1("d=%d\n", d); if ((d + 1) % sblock.fs_frag) { /* * The end of the cylinder summary is not a complete * block. */ DBG_TRC; frag_adjust(d % sblock.fs_fpg, -1); for (; (d + 1) % sblock.fs_frag; d--) { DBG_PRINT1("d=%d\n", d); setbit(cg_blksfree(&acg), d % sblock.fs_fpg); acg.cg_cs.cs_nffree++; sblock.fs_cstotal.cs_nffree++; } /* * Point "d" to the last fragment of the last * (incomplete) block of the cylinder summary. */ d++; frag_adjust(d % sblock.fs_fpg, 1); if (isblock(&sblock, cg_blksfree(&acg), (d % sblock.fs_fpg)/sblock.fs_frag)) { DBG_PRINT1("d=%d\n", d); acg.cg_cs.cs_nffree-=sblock.fs_frag; acg.cg_cs.cs_nbfree++; sblock.fs_cstotal.cs_nffree-=sblock.fs_frag; sblock.fs_cstotal.cs_nbfree++; cg_blktot(&acg)[cbtocylno(&sblock, d % sblock.fs_fpg)]++; cg_blks(&sblock, &acg, cbtocylno(&sblock, d % sblock.fs_fpg))[cbtorpos(&sblock, d % sblock.fs_fpg)]++; if (sblock.fs_contigsumsize > 0) { setbit(cg_clustersfree(&acg), (d % sblock.fs_fpg)/sblock.fs_frag); if (lcs < sblock.fs_contigsumsize) { if (lcs) cg_clustersum(&acg)[lcs]--; lcs++; cg_clustersum(&acg)[lcs]++; } } } /* * Point "d" to the first fragment of the block before * the last incomplete block. */ d--; } DBG_PRINT1("d=%d\n", d); for (d = rounddown(d, sblock.fs_frag); d >= osblock.fs_csaddr; d -= sblock.fs_frag) { DBG_TRC; DBG_PRINT1("d=%d\n", d); setblock(&sblock, cg_blksfree(&acg), (d % sblock.fs_fpg)/sblock.fs_frag); acg.cg_cs.cs_nbfree++; sblock.fs_cstotal.cs_nbfree++; cg_blktot(&acg)[cbtocylno(&sblock, d % sblock.fs_fpg)]++; cg_blks(&sblock, &acg, cbtocylno(&sblock, d % sblock.fs_fpg))[cbtorpos(&sblock, d % sblock.fs_fpg)]++; if (sblock.fs_contigsumsize > 0) { setbit(cg_clustersfree(&acg), (d % sblock.fs_fpg)/sblock.fs_frag); /* * The last cluster size is already set up. */ if (lcs < sblock.fs_contigsumsize) { if (lcs) cg_clustersum(&acg)[lcs]--; lcs++; cg_clustersum(&acg)[lcs]++; } } } *cs = acg.cg_cs; /* * Now write the former cylinder group containing the cylinder * summary back to disk. */ wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)), (size_t)sblock.fs_cgsize, &acg, fso, Nflag); DBG_PRINT0("oscg written\n"); DBG_DUMP_CG(&sblock, "old summary cg", &acg); /* * Find the beginning of the new cylinder group containing the * cylinder summary. */ sblock.fs_csaddr = cgdmin(&sblock, osblock.fs_ncg); ncscg = dtog(&sblock, sblock.fs_csaddr); cs = fscs + ncscg; /* * If Nflag is specified, we would now read random data instead * of an empty cg structure from disk. So we can't simulate that * part for now. */ if (Nflag) { DBG_PRINT0("nscg update skipped\n"); DBG_LEAVE; return; } /* * Read the future cylinder group containing the cylinder * summary from disk, and make a copy. */ rdfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)), (size_t)sblock.fs_cgsize, &aocg, fsi); DBG_PRINT0("nscg read\n"); DBG_DUMP_CG(&sblock, "new summary cg", &aocg); memcpy(&cgun1, &cgun2, sizeof(cgun2)); /* * Allocate all complete blocks used by the new cylinder * summary. */ for (d = sblock.fs_csaddr; d + sblock.fs_frag <= sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize); d+=sblock.fs_frag) { clrblock(&sblock, cg_blksfree(&acg), (d%sblock.fs_fpg)/sblock.fs_frag); acg.cg_cs.cs_nbfree--; sblock.fs_cstotal.cs_nbfree--; cg_blktot(&acg)[cbtocylno(&sblock, d%sblock.fs_fpg)]--; cg_blks(&sblock, &acg, cbtocylno(&sblock, d%sblock.fs_fpg))[cbtorpos(&sblock, d%sblock.fs_fpg)]--; if (sblock.fs_contigsumsize > 0) { clrbit(cg_clustersfree(&acg), (d%sblock.fs_fpg)/sblock.fs_frag); } } /* * Allocate all fragments used by the cylinder summary in the * last block. */ if (d < sblock.fs_csaddr+(sblock.fs_cssize/sblock.fs_fsize)) { for (; d - sblock.fs_csaddr< sblock.fs_cssize/sblock.fs_fsize; d++) { clrbit(cg_blksfree(&acg), d%sblock.fs_fpg); acg.cg_cs.cs_nffree--; sblock.fs_cstotal.cs_nffree--; } acg.cg_cs.cs_nbfree--; acg.cg_cs.cs_nffree+=sblock.fs_frag; sblock.fs_cstotal.cs_nbfree--; sblock.fs_cstotal.cs_nffree+=sblock.fs_frag; cg_blktot(&acg)[cbtocylno(&sblock, d%sblock.fs_fpg)]--; cg_blks(&sblock, &acg, cbtocylno(&sblock, d%sblock.fs_fpg))[cbtorpos(&sblock, d%sblock.fs_fpg)]--; if (sblock.fs_contigsumsize > 0) { clrbit(cg_clustersfree(&acg), (d%sblock.fs_fpg)/sblock.fs_frag); } frag_adjust(d%sblock.fs_fpg, +1); } /* * XXX Handle the cluster statistics here in the case this * cylinder group is now almost full, and the remaining * space is less than the maximum cluster size. This is * probably not needed, as you would hardly find a file * system which has only MAXCSBUFS+FS_MAXCONTIG of free * space right behind the cylinder group information in * any new cylinder group. */ /* * Update our statistics in the cylinder summary. */ *cs = acg.cg_cs; /* * Write the new cylinder group containing the cylinder summary * back to disk. */ wtfs(fsbtodb(&sblock, cgtod(&sblock, ncscg)), (size_t)sblock.fs_cgsize, &acg, fso, Nflag); DBG_PRINT0("nscg written\n"); DBG_DUMP_CG(&sblock, "new summary cg", &acg); DBG_LEAVE; return; } /* * We have got enough of space in the current cylinder group, so we * can relocate just a few blocks, and let the summary information * grow in place where it is right now. */ DBG_TRC; cbase = cgbase(&osblock, ocscg); /* old and new are equal */ dupper = sblock.fs_csaddr - cbase + howmany(sblock.fs_cssize, sblock.fs_fsize); odupper = osblock.fs_csaddr - cbase + howmany(osblock.fs_cssize, osblock.fs_fsize); sblock.fs_dsize -= dupper - odupper; /* * Allocate the space for the array of blocks to be relocated. */ bp= malloc(((dupper - odupper)/sblock.fs_frag + 2) * sizeof(struct gfs_bpp)); if (bp == NULL) errx(1, "malloc failed"); memset(bp, 0, ((dupper - odupper) / sblock.fs_frag + 2) * sizeof(struct gfs_bpp)); /* * Lock all new frags needed for the cylinder group summary. This is * done per fragment in the first and last block of the new required * area, and per block for all other blocks. * * Handle the first new block here (but only if some fragments where * already used for the cylinder summary). */ ind = 0; frag_adjust(odupper, -1); for (d = odupper; ((d < dupper) && (d % sblock.fs_frag)); d++) { DBG_PRINT1("scg first frag check loop d=%d\n", d); if (isclr(cg_blksfree(&acg), d)) { if (!ind) { bp[ind].old = d / sblock.fs_frag; bp[ind].flags|=GFS_FL_FIRST; if (roundup(d, sblock.fs_frag) >= dupper) bp[ind].flags |= GFS_FL_LAST; ind++; } } else { clrbit(cg_blksfree(&acg), d); acg.cg_cs.cs_nffree--; sblock.fs_cstotal.cs_nffree--; } /* * No cluster handling is needed here, as there was at least * one fragment in use by the cylinder summary in the old * filesystem. * No block - free counter handling here as this block was not * a free block. */ } frag_adjust(odupper, 1); /* * Handle all needed complete blocks here. */ for (; d + sblock.fs_frag<=dupper; d+=sblock.fs_frag) { DBG_PRINT1("scg block check loop d=%d\n", d); if (!isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) { for (f = d; f < d + sblock.fs_frag; f++) { if (isset(cg_blksfree(&aocg), f)) { acg.cg_cs.cs_nffree--; sblock.fs_cstotal.cs_nffree--; } } clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag); bp[ind].old = d / sblock.fs_frag; ind++; } else { clrblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag); acg.cg_cs.cs_nbfree--; sblock.fs_cstotal.cs_nbfree--; cg_blktot(&acg)[cbtocylno(&sblock, d)]--; cg_blks(&sblock, &acg, cbtocylno(&sblock, d)) [cbtorpos(&sblock, d)]--; if (sblock.fs_contigsumsize > 0) { clrbit(cg_clustersfree(&acg), d / sblock.fs_frag); for (lcs = 0, l = (d / sblock.fs_frag) + 1; lcs < sblock.fs_contigsumsize; l++, lcs++) { if (isclr(cg_clustersfree(&acg), l)) break; } if (lcs < sblock.fs_contigsumsize) { cg_clustersum(&acg)[lcs + 1]--; if (lcs) cg_clustersum(&acg)[lcs]++; } } } /* * No fragment counter handling is needed here, as this finally * doesn't change after the relocation. */ } /* * Handle all fragments needed in the last new affected block. */ if (d < dupper) { frag_adjust(dupper - 1, -1); if (isblock(&sblock, cg_blksfree(&acg), d/sblock.fs_frag)) { acg.cg_cs.cs_nbfree--; sblock.fs_cstotal.cs_nbfree--; acg.cg_cs.cs_nffree+=sblock.fs_frag; sblock.fs_cstotal.cs_nffree+=sblock.fs_frag; cg_blktot(&acg)[cbtocylno(&sblock, d)]--; cg_blks(&sblock, &acg, cbtocylno(&sblock, d)) [cbtorpos(&sblock, d)]--; if (sblock.fs_contigsumsize > 0) { clrbit(cg_clustersfree(&acg), d / sblock.fs_frag); for (lcs = 0, l = (d / sblock.fs_frag) + 1; lcs < sblock.fs_contigsumsize; l++, lcs++) { if (isclr(cg_clustersfree(&acg), l)) break; } if (lcs < sblock.fs_contigsumsize) { cg_clustersum(&acg)[lcs + 1]--; if (lcs) cg_clustersum(&acg)[lcs]++; } } } for (; d < dupper; d++) { DBG_PRINT1("scg second frag check loop d=%d\n", d); if (isclr(cg_blksfree(&acg), d)) { bp[ind].old = d / sblock.fs_frag; bp[ind].flags |= GFS_FL_LAST; } else { clrbit(cg_blksfree(&acg), d); acg.cg_cs.cs_nffree--; sblock.fs_cstotal.cs_nffree--; } } if (bp[ind].flags & GFS_FL_LAST) /* we have to advance here */ ind++; frag_adjust(dupper - 1, 1); } /* * If we found a block to relocate just do so. */ if (ind) { for (i = 0; i < ind; i++) { if (!bp[i].old) { /* no more blocks listed */ /* * XXX A relative blocknumber should not be * zero, which is not explicitly * guaranteed by our code. */ break; } /* * Allocate a complete block in the same (current) * cylinder group. */ bp[i].new = alloc() / sblock.fs_frag; /* * There is no frag_adjust() needed for the new block * as it will have no fragments yet :-). */ for (f = bp[i].old * sblock.fs_frag, g = bp[i].new * sblock.fs_frag; f < (bp[i].old + 1) * sblock.fs_frag; f++, g++) { if (isset(cg_blksfree(&aocg), f)) { setbit(cg_blksfree(&acg), g); acg.cg_cs.cs_nffree++; sblock.fs_cstotal.cs_nffree++; } } /* * Special handling is required if this was the first * block. We have to consider the fragments which were * used by the cylinder summary in the original block * which re to be free in the copy of our block. We * have to be careful if this first block happens to * be also the last block to be relocated. */ if (bp[i].flags & GFS_FL_FIRST) { for (f = bp[i].old * sblock.fs_frag, g = bp[i].new * sblock.fs_frag; f < odupper; f++, g++) { setbit(cg_blksfree(&acg), g); acg.cg_cs.cs_nffree++; sblock.fs_cstotal.cs_nffree++; } if (!(bp[i].flags & GFS_FL_LAST)) frag_adjust(bp[i].new * sblock.fs_frag,1); } /* * Special handling is required if this is the last * block to be relocated. */ if (bp[i].flags & GFS_FL_LAST) { frag_adjust(bp[i].new * sblock.fs_frag, 1); frag_adjust(bp[i].old * sblock.fs_frag, -1); for (f = dupper; f < roundup(dupper, sblock.fs_frag); f++) { if (isclr(cg_blksfree(&acg), f)) { setbit(cg_blksfree(&acg), f); acg.cg_cs.cs_nffree++; sblock.fs_cstotal.cs_nffree++; } } frag_adjust(bp[i].old * sblock.fs_frag, 1); } /* * !!! Attach the cylindergroup offset here. */ bp[i].old += cbase / sblock.fs_frag; bp[i].new += cbase / sblock.fs_frag; /* * Copy the content of the block. */ /* * XXX Here we will have to implement a copy on write * in the case we have any active snapshots. */ rdfs(fsbtodb(&sblock, bp[i].old * sblock.fs_frag), (size_t)sblock.fs_bsize, &ablk, fsi); wtfs(fsbtodb(&sblock, bp[i].new * sblock.fs_frag), (size_t)sblock.fs_bsize, &ablk, fso, Nflag); DBG_DUMP_HEX(&sblock, "copied full block", (unsigned char *)&ablk); DBG_PRINT2("scg (%d->%d) block relocated\n", bp[i].old, bp[i].new); } /* * Now we have to update all references to any fragment which * belongs to any block relocated. We iterate now over all * cylinder groups, within those over all non zero length * inodes. */ for (cylno = 0; cylno < osblock.fs_ncg; cylno++) { DBG_PRINT1("scg doing cg (%d)\n", cylno); for (inc = osblock.fs_ipg - 1; inc >= 0; inc--) { updrefs(cylno, (ino_t)inc, bp, fsi, fso, Nflag); } } /* * All inodes are checked, now make sure the number of * references found make sense. */ for (i = 0; i < ind; i++) { if (!bp[i].found || (bp[i].found>sblock.fs_frag)) { warnx("error: %d refs found for block %d.", bp[i].found, bp[i].old); } } } /* * The following statistics are not changed here: * sblock.fs_cstotal.cs_ndir * sblock.fs_cstotal.cs_nifree * The following statistics were already updated on the fly: * sblock.fs_cstotal.cs_nffree * sblock.fs_cstotal.cs_nbfree * As the statistics for this cylinder group are ready, copy it to * the summary information array. */ *cs = acg.cg_cs; /* * Write summary cylinder group back to disk. */ wtfs(fsbtodb(&sblock, cgtod(&sblock, ocscg)), (size_t)sblock.fs_cgsize, &acg, fso, Nflag); DBG_PRINT0("scg written\n"); DBG_DUMP_CG(&sblock, "new summary cg", &acg); DBG_LEAVE; } /* ************************************************************** rdfs ***** */ /* * Here we read some block(s) from disk. */ static void rdfs(daddr_t bno, size_t size, void *bf, int fsi) { DBG_FUNC("rdfs") ssize_t n; DBG_ENTER; if (lseek(fsi, (off_t)bno * DEV_BSIZE, 0) < 0) { err(33, "rdfs: seek error: %ld", (long)bno); } n = read(fsi, bf, size); if (n != (ssize_t)size) { err(34, "rdfs: read error: %ld", (long)bno); } DBG_LEAVE; } /* ************************************************************** wtfs ***** */ /* * Here we write some block(s) to disk. */ static void wtfs(daddr_t bno, size_t size, void *bf, int fso, unsigned int Nflag) { DBG_FUNC("wtfs") ssize_t n; DBG_ENTER; if (Nflag) { DBG_LEAVE; return; } if (lseek(fso, (off_t)bno * DEV_BSIZE, SEEK_SET) < 0) err(35, "wtfs: seek error: %ld", (long)bno); n = write(fso, bf, size); if (n != (ssize_t)size) err(36, "wtfs: write error: %ld", (long)bno); DBG_LEAVE; } /* ************************************************************* alloc ***** */ /* * Here we allocate a free block in the current cylinder group. It is assumed, * that acg contains the current cylinder group. As we may take a block from * somewhere in the filesystem we have to handle cluster summary here. */ static daddr_t alloc(void) { DBG_FUNC("alloc") daddr_t d, blkno; int lcs1, lcs2; int l; int csmin, csmax; int dlower, dupper, dmax; DBG_ENTER; if (acg.cg_magic != CG_MAGIC) { warnx("acg: bad magic number"); DBG_LEAVE; return (0); } if (acg.cg_cs.cs_nbfree == 0) { warnx("error: cylinder group ran out of space"); DBG_LEAVE; return (0); } /* * We start seeking for free blocks only from the space available after * the end of the new grown cylinder summary. Otherwise we allocate a * block here which we have to relocate a couple of seconds later again * again, and we are not prepared to to this anyway. */ blkno = -1; dlower = cgsblock(&sblock, acg.cg_cgx) - cgbase(&sblock, acg.cg_cgx); dupper = cgdmin(&sblock, acg.cg_cgx) - cgbase(&sblock, acg.cg_cgx); dmax = cgbase(&sblock, acg.cg_cgx) + sblock.fs_fpg; if (dmax > sblock.fs_size) { dmax = sblock.fs_size; } dmax -= cgbase(&sblock, acg.cg_cgx); /* retransform into cg */ csmin = sblock.fs_csaddr - cgbase(&sblock, acg.cg_cgx); csmax = csmin + howmany(sblock.fs_cssize, sblock.fs_fsize); DBG_PRINT3("seek range: dl=%d, du=%d, dm=%d\n", dlower, dupper, dmax); DBG_PRINT2("range cont: csmin=%d, csmax=%d\n", csmin, csmax); for (d = 0; (d < dlower && blkno == -1); d += sblock.fs_frag) { if (d >= csmin && d<=csmax) { continue; } if (isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock, d))) { blkno = fragstoblks(&sblock, d);/* Yeah found a block */ break; } } for (d = dupper; (d < dmax && blkno == -1); d += sblock.fs_frag) { if (d >= csmin && d <= csmax) { continue; } if (isblock(&sblock, cg_blksfree(&acg), fragstoblks(&sblock, d))) { blkno = fragstoblks(&sblock, d);/* Yeah found a block */ break; } } if (blkno==-1) { warnx("internal error: couldn't find promised block in cg"); DBG_LEAVE; return (0); } /* * This is needed if the block was found already in the first loop. */ d = blkstofrags(&sblock, blkno); clrblock(&sblock, cg_blksfree(&acg), blkno); if (sblock.fs_contigsumsize > 0) { /* * Handle the cluster allocation bitmap. */ clrbit(cg_clustersfree(&acg), blkno); /* * We possibly have split a cluster here, so we have to do * recalculate the sizes of the remaining cluster halves now, * and use them for updating the cluster summary information. * * Lets start with the blocks before our allocated block ... */ for (lcs1 = 0, l = blkno - 1; lcs1fs_frag) { case 8: DBG_LEAVE; return (cp[h] == 0xff); case 4: mask = 0x0f << ((h & 0x1) << 2); DBG_LEAVE; return ((cp[h >> 1] & mask) == mask); case 2: mask = 0x03 << ((h & 0x3) << 1); DBG_LEAVE; return ((cp[h >> 2] & mask) == mask); case 1: mask = 0x01 << (h & 0x7); DBG_LEAVE; return ((cp[h >> 3] & mask) == mask); default: fprintf(stderr, "isblock bad fs_frag %d\n", fs->fs_frag); DBG_LEAVE; return (0); } } /* ********************************************************** clrblock ***** */ /* * Here we allocate a complete block in the block map. For more details again * please see the source of newfs(8), as this function is taken over almost * unchanged. */ static void clrblock(struct fs *fs, unsigned char *cp, int h) { DBG_FUNC("clrblock") DBG_ENTER; switch ((fs)->fs_frag) { case 8: cp[h] = 0; break; case 4: cp[h >> 1] &= ~(0x0f << ((h & 0x1) << 2)); break; case 2: cp[h >> 2] &= ~(0x03 << ((h & 0x3) << 1)); break; case 1: cp[h >> 3] &= ~(0x01 << (h & 0x7)); break; default: warnx("clrblock bad fs_frag %d", fs->fs_frag); break; } DBG_LEAVE; } /* ********************************************************** setblock ***** */ /* * Here we free a complete block in the free block map. For more details again * please see the source of newfs(8), as this function is taken over almost * unchanged. */ static void setblock(struct fs *fs, unsigned char *cp, int h) { DBG_FUNC("setblock") DBG_ENTER; switch (fs->fs_frag) { case 8: cp[h] = 0xff; break; case 4: cp[h >> 1] |= (0x0f << ((h & 0x1) << 2)); break; case 2: cp[h >> 2] |= (0x03 << ((h & 0x3) << 1)); break; case 1: cp[h >> 3] |= (0x01 << (h & 0x7)); break; default: warnx("setblock bad fs_frag %d", fs->fs_frag); break; } DBG_LEAVE; } /* ************************************************************ ginode ***** */ /* * This function provides access to an individual inode. We find out in which * block the requested inode is located, read it from disk if needed, and * return the pointer into that block. We maintain a cache of one block to * not read the same block again and again if we iterate linearly over all * inodes. */ static struct ufs1_dinode * ginode(ino_t inumber, int fsi, int cg) { DBG_FUNC("ginode") ufs_daddr_t iblk; static ino_t startinum = 0; /* first inode in cached block */ struct ufs1_dinode *pi; DBG_ENTER; pi = (struct ufs1_dinode *)ablk; inumber += (cg * sblock.fs_ipg); if (startinum == 0 || inumber < startinum || inumber >= startinum + INOPB(&sblock)) { /* * The block needed is not cached, so we have to read it from * disk now. */ iblk = ino_to_fsba(&sblock, inumber); in_src = fsbtodb(&sblock, iblk); rdfs(in_src, (size_t)sblock.fs_bsize, &ablk, fsi); startinum = (inumber / INOPB(&sblock)) * INOPB(&sblock); } DBG_LEAVE; return (&(pi[inumber % INOPB(&sblock)])); } /* ****************************************************** charsperline ***** */ /* * Figure out how many lines our current terminal has. For more details again * please see the source of newfs(8), as this function is taken over almost * unchanged. */ static int charsperline(void) { DBG_FUNC("charsperline") int columns; char *cp; struct winsize ws; DBG_ENTER; columns = 0; if (ioctl(0, TIOCGWINSZ, &ws) != -1) { columns = ws.ws_col; } if (columns == 0 && (cp = getenv("COLUMNS"))) { columns = atoi(cp); } if (columns == 0) { columns = 80; /* last resort */ } DBG_LEAVE; return columns; } /* ************************************************************** main ***** */ /* * growfs(8) is a utility which allows to increase the size of an existing * ufs filesystem. Currently this can only be done on unmounted file system. * It recognizes some command line options to specify the new desired size, * and it does some basic checkings. The old filesystem size is determined * and after some more checks like we can really access the new last block * on the disk etc. we calculate the new parameters for the superblock. After * having done this we just call growfs() which will do the work. Before * we finish the only thing left is to update the disklabel. * We still have to provide support for snapshots. Therefore we first have to * understand what data structures are always replicated in the snapshot on * creation, for all other blocks we touch during our procedure, we have to * keep the old blocks unchanged somewhere available for the snapshots. If we * are lucky, then we only have to handle our blocks to be relocated in that * way. * Also we have to consider in what order we actually update the critical * data structures of the filesystem to make sure, that in case of a disaster * fsck(8) is still able to restore any lost data. * The foreseen last step then will be to provide for growing even mounted * file systems. There we have to extend the mount() system call to provide * userland access to the filesystem locking facility. */ int main(int argc, char **argv) { DBG_FUNC("main") char *device, *rdev; int ch; unsigned int size = 0; unsigned int Nflag = 0; int ExpertFlag = 0; struct stat st; struct disklabel *lp; struct partition *pp; int fsi, fso; char reply[5]; #ifdef FSMAXSNAP int j; #endif /* FSMAXSNAP */ DBG_ENTER; while ((ch = getopt(argc, argv, "Ns:y")) != -1) { switch (ch) { case 'N': Nflag = 1; break; case 's': size = (size_t)atol(optarg); if (size < 1) { usage(); } break; case 'y': ExpertFlag = 1; break; case '?': /* FALLTHROUGH */ default: usage(); } } argc -= optind; argv += optind; if (argc != 1) { usage(); } device = *argv; /* * Rather than guessing, use opendev() to get the device * name, which we open for reading. */ if ((fsi = opendev(device, O_RDONLY, 0, &rdev)) < 0) err(1, "%s", rdev); /* * Try to access our device for writing ... */ if (Nflag) { fso = -1; } else { fso = open(rdev, O_WRONLY); if (fso < 0) err(1, "%s", rdev); } /* * Now we have a file descriptor for our device, fstat() it to * figure out the partition number. */ if (fstat(fsi, &st) != 0) err(1, "%s: fstat()", rdev); /* * Try to read a label from the disk. Then get the partition from the * device minor number, using DISKPART(). Probably don't need to * check against getmaxpartitions(). */ lp = get_disklabel(fsi); if (DISKPART(st.st_rdev) < getmaxpartitions()) pp = &lp->d_partitions[DISKPART(st.st_rdev)]; else errx(1, "%s: invalid partition number %u", rdev, DISKPART(st.st_rdev)); /* * Check if that partition looks suited for growing a filesystem. */ if (pp->p_size < 1) errx(1, "partition is unavailable"); if (pp->p_fstype != FS_BSDFFS) errx(1, "partition not 4.2BSD"); /* * Read the current superblock, and take a backup. */ rdfs((daddr_t)(SBOFF / DEV_BSIZE), (size_t)SBSIZE, &osblock, fsi); if (osblock.fs_magic != FS_MAGIC) errx(1, "superblock not recognized"); memcpy(&fsun1, &fsun2, sizeof(fsun2)); DBG_OPEN("/tmp/growfs.debug"); /* already here we need a superblock */ DBG_DUMP_FS(&sblock, "old sblock"); /* * Determine size to grow to. Default to the full size specified in * the disk label. */ sblock.fs_size = dbtofsb(&osblock, pp->p_size); if (size != 0) { if (size > pp->p_size) { errx(1, "There is not enough space (%d < %d)", pp->p_size, size); } sblock.fs_size = dbtofsb(&osblock, size); } /* * Are we really growing ? */ if (osblock.fs_size >= sblock.fs_size) { errx(1, "we are not growing (%d->%d)", osblock.fs_size, sblock.fs_size); } #ifdef FSMAXSNAP /* * Check if we find an active snapshot. */ if (ExpertFlag == 0) { for (j = 0; j < FSMAXSNAP; j++) { if (sblock.fs_snapinum[j]) { errx(1, "active snapshot found in filesystem\n" " please remove all snapshots before " "using growfs"); } if (!sblock.fs_snapinum[j]) /* list is dense */ break; } } #endif if (ExpertFlag == 0 && Nflag == 0) { printf("We strongly recommend you make a backup " "before growing the Filesystem\n\n" " Did you backup your data (Yes/No) ? "); fgets(reply, (int)sizeof(reply), stdin); if (strcmp(reply, "Yes\n")) { printf("\n Nothing done \n"); exit (0); } } printf("new file system size is: %d frags\n", sblock.fs_size); /* * Try to access our new last block in the filesystem. Even if we * later on realize we have to abort our operation, on that block * there should be no data, so we can't destroy something yet. */ wtfs((daddr_t)pp->p_size - 1, (size_t)DEV_BSIZE, &sblock, fso, Nflag); /* * Now calculate new superblock values and check for reasonable * bound for new filesystem size: * fs_size: is derived from label or user input * fs_dsize: should get updated in the routines creating or * updating the cylinder groups on the fly * fs_cstotal: should get updated in the routines creating or * updating the cylinder groups */ /* * Update the number of cylinders in the filesystem. */ sblock.fs_ncyl = sblock.fs_size * NSPF(&sblock) / sblock.fs_spc; if (sblock.fs_size * NSPF(&sblock) > sblock.fs_ncyl * sblock.fs_spc) sblock.fs_ncyl++; /* * Update the number of cylinder groups in the filesystem. */ sblock.fs_ncg = sblock.fs_ncyl / sblock.fs_cpg; if (sblock.fs_ncyl % sblock.fs_cpg) sblock.fs_ncg++; if ((sblock.fs_size - (sblock.fs_ncg - 1) * sblock.fs_fpg) < sblock.fs_fpg && cgdmin(&sblock, (sblock.fs_ncg - 1))- cgbase(&sblock, (sblock.fs_ncg - 1)) > (sblock.fs_size - (sblock.fs_ncg - 1) * sblock.fs_fpg)) { /* * The space in the new last cylinder group is too small, * so revert back. */ sblock.fs_ncg--; #if 1 /* this is a bit more safe */ sblock.fs_ncyl = sblock.fs_ncg * sblock.fs_cpg; #else sblock.fs_ncyl -= sblock.fs_ncyl % sblock.fs_cpg; #endif sblock.fs_ncyl -= sblock.fs_ncyl % sblock.fs_cpg; printf("Warning: %d sector(s) cannot be allocated.\n", (sblock.fs_size-(sblock.fs_ncg)*sblock.fs_fpg) * NSPF(&sblock)); sblock.fs_size = sblock.fs_ncyl * sblock.fs_spc / NSPF(&sblock); } /* * Update the space for the cylinder group summary information in the * respective cylinder group data area. */ sblock.fs_cssize = fragroundup(&sblock, sblock.fs_ncg * sizeof(struct csum)); if (osblock.fs_size >= sblock.fs_size) { errx(1, "not enough new space"); } DBG_PRINT0("sblock calculated\n"); /* * Ok, everything prepared, so now let's do the tricks. */ growfs(fsi, fso, Nflag); /* * Update the disk label. */ pp->p_fsize = sblock.fs_fsize; pp->p_frag = sblock.fs_frag; pp->p_cpg = sblock.fs_cpg; return_disklabel(fso, lp, Nflag); DBG_PRINT0("label rewritten\n"); close(fsi); if (fso > -1) close(fso); DBG_CLOSE; DBG_LEAVE; return (0); } /* ************************************************** return_disklabel ***** */ /* * Write the updated disklabel back to disk. */ static void return_disklabel(int fd, struct disklabel *lp, unsigned int Nflag) { DBG_FUNC("return_disklabel") u_short sum; u_short *ptr; DBG_ENTER; if (!lp) { DBG_LEAVE; return; } if (!Nflag) { lp->d_checksum = 0; sum = 0; ptr = (u_short *)lp; /* * recalculate checksum */ while (ptr < (u_short *)&lp->d_partitions[lp->d_npartitions]) sum ^= *ptr++; lp->d_checksum = sum; if (ioctl(fd, DIOCWDINFO, (char *)lp) < 0) errx(1, "DIOCWDINFO failed"); } free(lp); DBG_LEAVE; } /* ***************************************************** get_disklabel ***** */ /* * Read the disklabel from disk. */ static struct disklabel * get_disklabel(int fd) { DBG_FUNC("get_disklabel") static struct disklabel *lab; DBG_ENTER; lab = malloc(sizeof(struct disklabel)); if (!lab) errx(1, "malloc failed"); if (ioctl(fd, DIOCGDINFO, (char *)lab) < 0) errx(1, "DIOCGDINFO failed"); DBG_LEAVE; return (lab); } /* ************************************************************* usage ***** */ /* * Dump a line of usage. */ static void usage(void) { DBG_FUNC("usage") DBG_ENTER; fprintf(stderr, "usage: growfs [-Ny] [-s size] special\n"); DBG_LEAVE; exit(1); } /* *********************************************************** updclst ***** */ /* * This updates most parameters and the bitmap related to cluster. We have to * assume, that sblock, osblock, acg are set up. */ static void updclst(int block) { DBG_FUNC("updclst") static int lcs = 0; DBG_ENTER; if (sblock.fs_contigsumsize < 1) /* no clustering */ return; /* * update cluster allocation map */ setbit(cg_clustersfree(&acg), block); /* * update cluster summary table */ if (!lcs) { /* * calculate size for the trailing cluster */ for (block--; lcs < sblock.fs_contigsumsize; block--, lcs++) { if (isclr(cg_clustersfree(&acg), block)) break; } } if (lcs < sblock.fs_contigsumsize) { if (lcs) cg_clustersum(&acg)[lcs]--; lcs++; cg_clustersum(&acg)[lcs]++; } DBG_LEAVE; } /* *********************************************************** updrefs ***** */ /* * This updates all references to relocated blocks for the given inode. The * inode is given as number within the cylinder group, and the number of the * cylinder group. */ static void updrefs(int cg, ino_t in, struct gfs_bpp *bp, int fsi, int fso, unsigned int Nflag) { DBG_FUNC("updrefs") unsigned int ictr, ind2ctr, ind3ctr; ufs_daddr_t *iptr, *ind2ptr, *ind3ptr; struct ufs1_dinode *ino; int remaining_blocks; DBG_ENTER; /* * XXX We should skip unused inodes even from beeing read from disk * here by using the bitmap. */ ino = ginode(in, fsi, cg); if (!((ino->di_mode & IFMT) == IFDIR || (ino->di_mode & IFMT) == IFREG || (ino->di_mode & IFMT) == IFLNK)) { DBG_LEAVE; return; /* only check DIR, FILE, LINK */ } if (((ino->di_mode & IFMT) == IFLNK) && (ino->di_size < MAXSYMLINKLEN)) { DBG_LEAVE; return; /* skip short symlinks */ } if (!ino->di_size) { DBG_LEAVE; return; /* skip empty file */ } if (!ino->di_blocks) { DBG_LEAVE; return; /* skip empty swiss cheesy file or old fastlink */ } DBG_PRINT2("scg checking inode (%d in %d)\n", in, cg); /* * Start checking all direct blocks. */ remaining_blocks = howmany(ino->di_size, sblock.fs_bsize); for (ictr = 0; ictr < MIN(NDADDR, (unsigned int)remaining_blocks); ictr++) { iptr = &(ino->di_db[ictr]); if (*iptr) cond_bl_upd(iptr, bp, GFS_PS_INODE, fso, Nflag); } DBG_PRINT0("~~scg direct blocks checked\n"); remaining_blocks -= NDADDR; if (remaining_blocks < 0) { DBG_LEAVE; return; } if (ino->di_ib[0]) { /* * Start checking first indirect block */ cond_bl_upd(&(ino->di_ib[0]), bp, GFS_PS_INODE, fso, Nflag); i1_src = fsbtodb(&sblock, ino->di_ib[0]); rdfs(i1_src, (size_t)sblock.fs_bsize, &i1blk, fsi); for (ictr = 0; ictr < MIN(howmany(sblock.fs_bsize, sizeof(ufs_daddr_t)), (unsigned int)remaining_blocks); ictr++) { iptr = &((ufs_daddr_t *)&i1blk)[ictr]; if (*iptr) { cond_bl_upd(iptr, bp, GFS_PS_IND_BLK_LVL1, fso, Nflag); } } } DBG_PRINT0("scg indirect_1 blocks checked\n"); remaining_blocks -= howmany(sblock.fs_bsize, sizeof(ufs_daddr_t)); if (remaining_blocks<0) { DBG_LEAVE; return; } if (ino->di_ib[1]) { /* * Start checking second indirect block */ cond_bl_upd(&(ino->di_ib[1]), bp, GFS_PS_INODE, fso, Nflag); i2_src = fsbtodb(&sblock, ino->di_ib[1]); rdfs(i2_src, (size_t)sblock.fs_bsize, &i2blk, fsi); for (ind2ctr = 0; ind2ctr < howmany(sblock.fs_bsize, sizeof(ufs_daddr_t)); ind2ctr++) { ind2ptr = &((ufs_daddr_t *)&i2blk)[ind2ctr]; if (!*ind2ptr) continue; cond_bl_upd(ind2ptr, bp, GFS_PS_IND_BLK_LVL2, fso, Nflag); i1_src = fsbtodb(&sblock, *ind2ptr); rdfs(i1_src, (size_t)sblock.fs_bsize, &i1blk, fsi); for (ictr = 0; ictr < MIN(howmany((unsigned int) sblock.fs_bsize, sizeof(ufs_daddr_t)), (unsigned int)remaining_blocks); ictr++) { iptr = &((ufs_daddr_t *)&i1blk)[ictr]; if (*iptr) { cond_bl_upd(iptr, bp, GFS_PS_IND_BLK_LVL1, fso, Nflag); } } } } DBG_PRINT0("scg indirect_2 blocks checked\n"); #define SQUARE(a) ((a)*(a)) remaining_blocks -= SQUARE(howmany(sblock.fs_bsize, sizeof(ufs_daddr_t))); #undef SQUARE if (remaining_blocks < 0) { DBG_LEAVE; return; } if (ino->di_ib[2]) { /* * Start checking third indirect block */ cond_bl_upd(&(ino->di_ib[2]), bp, GFS_PS_INODE, fso, Nflag); i3_src = fsbtodb(&sblock, ino->di_ib[2]); rdfs(i3_src, (size_t)sblock.fs_bsize, &i3blk, fsi); for (ind3ctr = 0; ind3ctr < howmany(sblock.fs_bsize, sizeof(ufs_daddr_t)); ind3ctr++) { ind3ptr = &((ufs_daddr_t *)&i3blk)[ind3ctr]; if (!*ind3ptr) continue; cond_bl_upd(ind3ptr, bp, GFS_PS_IND_BLK_LVL3, fso, Nflag); i2_src = fsbtodb(&sblock, *ind3ptr); rdfs(i2_src, (size_t)sblock.fs_bsize, &i2blk, fsi); for (ind2ctr = 0; ind2ctr < howmany(sblock.fs_bsize, sizeof(ufs_daddr_t)); ind2ctr++) { ind2ptr = &((ufs_daddr_t *)&i2blk)[ind2ctr]; if (!*ind2ptr) continue; cond_bl_upd(ind2ptr, bp, GFS_PS_IND_BLK_LVL2, fso, Nflag); i1_src = fsbtodb(&sblock, *ind2ptr); rdfs(i1_src, (size_t)sblock.fs_bsize, &i1blk, fsi); for (ictr = 0; ictr < MIN(howmany(sblock.fs_bsize, sizeof(ufs_daddr_t)), (unsigned int)remaining_blocks); ictr++) { iptr = &((ufs_daddr_t *)&i1blk)[ictr]; if (*iptr) { cond_bl_upd(iptr, bp, GFS_PS_IND_BLK_LVL1, fso, Nflag); } } } } } DBG_PRINT0("scg indirect_3 blocks checked\n"); DBG_LEAVE; }