Home | History | Annotate | Line # | Download | only in resize_ffs
resize_ffs.c revision 1.11
      1 /*	$NetBSD: resize_ffs.c,v 1.11 2007/12/15 16:32:06 perry Exp $	*/
      2 /* From sources sent on February 17, 2003 */
      3 /*-
      4  * As its sole author, I explicitly place this code in the public
      5  *  domain.  Anyone may use it for any purpose (though I would
      6  *  appreciate credit where it is due).
      7  *
      8  *					der Mouse
      9  *
     10  *			       mouse (at) rodents.montreal.qc.ca
     11  *		     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B
     12  */
     13 /*
     14  * resize_ffs:
     15  *
     16  * Resize a filesystem.  Is capable of both growing and shrinking.
     17  *
     18  * Usage: resize_ffs filesystem newsize
     19  *
     20  * Example: resize_ffs /dev/rsd1e 29574
     21  *
     22  * newsize is in DEV_BSIZE units (ie, disk sectors, usually 512 bytes
     23  *  each).
     24  *
     25  * Note: this currently requires gcc to build, since it is written
     26  *  depending on gcc-specific features, notably nested function
     27  *  definitions (which in at least a few cases depend on the lexical
     28  *  scoping gcc provides, so they can't be trivially moved outside).
     29  *
     30  * It will not do anything useful with filesystems in other than
     31  *  host-native byte order.  This really should be fixed (it's largely
     32  *  a historical accident; the original version of this program is
     33  *  older than bi-endian support in FFS).
     34  *
     35  * Many thanks go to John Kohl <jtk (at) NetBSD.org> for finding bugs: the
     36  *  one responsible for the "realloccgblk: can't find blk in cyl"
     37  *  problem and a more minor one which left fs_dsize wrong when
     38  *  shrinking.  (These actually indicate bugs in fsck too - it should
     39  *  have caught and fixed them.)
     40  *
     41  */
     42 
     43 #include <sys/cdefs.h>
     44 #include <stdio.h>
     45 #include <errno.h>
     46 #include <fcntl.h>
     47 #include <stdlib.h>
     48 #include <unistd.h>
     49 #include <strings.h>
     50 #include <err.h>
     51 #include <sys/stat.h>
     52 #include <sys/mman.h>
     53 #include <sys/param.h>		/* MAXFRAG */
     54 #include <ufs/ffs/fs.h>
     55 #include <ufs/ufs/dir.h>
     56 #include <ufs/ufs/dinode.h>
     57 #include <ufs/ufs/ufs_bswap.h>	/* ufs_rw32 */
     58 
     59 /* Suppress warnings about unused arguments */
     60 #if	defined(__GNUC__) &&				\
     61 	( (__GNUC__ > 2) ||				\
     62 	  ( (__GNUC__ == 2) &&				\
     63 	    defined(__GNUC_MINOR__) &&			\
     64 	    (__GNUC_MINOR__ >= 7) ) )
     65 #define UNUSED_ARG(x) x __attribute__((__unused__))
     66 #define INLINE inline
     67 #else
     68 #define UNUSED_ARG(x) x
     69 #define INLINE			/**/
     70 #endif
     71 
     72 /* new size of filesystem, in sectors */
     73 static int newsize;
     74 
     75 /* fd open onto disk device */
     76 static int fd;
     77 
     78 /* must we break up big I/O operations - see checksmallio() */
     79 static int smallio;
     80 
     81 /* size of a cg, in bytes, rounded up to a frag boundary */
     82 static int cgblksz;
     83 
     84 /* possible superblock localtions */
     85 static int search[] = SBLOCKSEARCH;
     86 /* location of the superblock */
     87 static off_t where;
     88 
     89 /* Superblocks. */
     90 static struct fs *oldsb;	/* before we started */
     91 static struct fs *newsb;	/* copy to work with */
     92 /* Buffer to hold the above.  Make sure it's aligned correctly. */
     93 static char sbbuf[2 * SBLOCKSIZE] __attribute__((__aligned__(__alignof__(struct fs))));
     94 
     95 /* a cg's worth of brand new squeaky-clean inodes */
     96 static struct ufs1_dinode *zinodes;
     97 
     98 /* pointers to the in-core cgs, read off disk and possibly modified */
     99 static struct cg **cgs;
    100 
    101 /* pointer to csum array - the stuff pointed to on-disk by fs_csaddr */
    102 static struct csum *csums;
    103 
    104 /* per-cg flags, indexed by cg number */
    105 static unsigned char *cgflags;
    106 #define CGF_DIRTY   0x01	/* needs to be written to disk */
    107 #define CGF_BLKMAPS 0x02	/* block bitmaps need rebuilding */
    108 #define CGF_INOMAPS 0x04	/* inode bitmaps need rebuilding */
    109 
    110 /* when shrinking, these two arrays record how we want blocks to move.	 */
    111 /*  if blkmove[i] is j, the frag that started out as frag #i should end	 */
    112 /*  up as frag #j.  inomove[i]=j means, similarly, that the inode that	 */
    113 /*  started out as inode i should end up as inode j.			 */
    114 static unsigned int *blkmove;
    115 static unsigned int *inomove;
    116 
    117 /* in-core copies of all inodes in the fs, indexed by inumber */
    118 static struct ufs1_dinode *inodes;
    119 
    120 /* per-inode flags, indexed by inumber */
    121 static unsigned char *iflags;
    122 #define IF_DIRTY  0x01		/* needs to be written to disk */
    123 #define IF_BDIRTY 0x02		/* like DIRTY, but is set on first inode in a
    124 				 * block of inodes, and applies to the whole
    125 				 * block. */
    126 
    127 /* Old FFS1 macros */
    128 #define cg_blktot(cgp, ns) \
    129     (cg_chkmagic(cgp, ns) ? \
    130     ((int32_t *)((u_int8_t *)(cgp) + ufs_rw32((cgp)->cg_old_btotoff, (ns)))) \
    131     : (((struct ocg *)(cgp))->cg_btot))
    132 #define cg_blks(fs, cgp, cylno, ns) \
    133     (cg_chkmagic(cgp, ns) ? \
    134     ((int16_t *)((u_int8_t *)(cgp) + ufs_rw32((cgp)->cg_old_boff, (ns))) + \
    135 	(cylno) * (fs)->fs_old_nrpos) \
    136     : (((struct ocg *)(cgp))->cg_b[cylno]))
    137 #define cbtocylno(fs, bno) \
    138    (fsbtodb(fs, bno) / (fs)->fs_old_spc)
    139 #define cbtorpos(fs, bno) \
    140     ((fs)->fs_old_nrpos <= 1 ? 0 : \
    141      (fsbtodb(fs, bno) % (fs)->fs_old_spc / \
    142       (fs)->fs_old_nsect * (fs)->fs_old_trackskew + \
    143       fsbtodb(fs, bno) % (fs)->fs_old_spc % \
    144       (fs)->fs_old_nsect * (fs)->fs_old_interleave) %\
    145     (fs)->fs_old_nsect * (fs)->fs_old_nrpos / (fs)->fs_old_npsect)
    146 #define dblksize(fs, dip, lbn) \
    147     (((lbn) >= NDADDR || (dip)->di_size >= lblktosize(fs, (lbn) + 1)) \
    148     ? (fs)->fs_bsize \
    149     : (fragroundup(fs, blkoff(fs, (dip)->di_size))))
    150 
    151 
    152 /*
    153  * Number of disk sectors per block/fragment; assumes DEV_BSIZE byte
    154  * sector size.
    155  */
    156 #define NSPB(fs)	((fs)->fs_old_nspf << (fs)->fs_fragshift)
    157 #define NSPF(fs)	((fs)->fs_old_nspf)
    158 
    159 /*
    160  * See if we need to break up large I/O operations.  This should never
    161  *  be needed, but under at least one <version,platform> combination,
    162  *  large enough disk transfers to the raw device hang.  So if we're
    163  *  talking to a character special device, play it safe; in this case,
    164  *  readat() and writeat() break everything up into pieces no larger
    165  *  than 8K, doing multiple syscalls for larger operations.
    166  */
    167 static void
    168 checksmallio(void)
    169 {
    170 	struct stat stb;
    171 
    172 	fstat(fd, &stb);
    173 	smallio = ((stb.st_mode & S_IFMT) == S_IFCHR);
    174 }
    175 /*
    176  * Read size bytes starting at blkno into buf.  blkno is in DEV_BSIZE
    177  *  units, ie, after fsbtodb(); size is in bytes.
    178  */
    179 static void
    180 readat(off_t blkno, void *buf, int size)
    181 {
    182 	/* Seek to the correct place. */
    183 	if (lseek(fd, blkno * DEV_BSIZE, L_SET) < 0)
    184 		err(1, "lseek failed");
    185 
    186 	/* See if we have to break up the transfer... */
    187 	if (smallio) {
    188 		char *bp;	/* pointer into buf */
    189 		int left;	/* bytes left to go */
    190 		int n;		/* number to do this time around */
    191 		int rv;		/* syscall return value */
    192 		bp = buf;
    193 		left = size;
    194 		while (left > 0) {
    195 			n = (left > 8192) ? 8192 : left;
    196 			rv = read(fd, bp, n);
    197 			if (rv < 0)
    198 				err(1, "read failed");
    199 			if (rv != n)
    200 				errx(1, "read: wanted %d, got %d", n, rv);
    201 			bp += n;
    202 			left -= n;
    203 		}
    204 	} else {
    205 		int rv;
    206 		rv = read(fd, buf, size);
    207 		if (rv < 0)
    208 			err(1, "read failed");
    209 		if (rv != size)
    210 			errx(1, "read: wanted %d, got %d", size, rv);
    211 	}
    212 }
    213 /*
    214  * Write size bytes from buf starting at blkno.  blkno is in DEV_BSIZE
    215  *  units, ie, after fsbtodb(); size is in bytes.
    216  */
    217 static void
    218 writeat(off_t blkno, const void *buf, int size)
    219 {
    220 	/* Seek to the correct place. */
    221 	if (lseek(fd, blkno * DEV_BSIZE, L_SET) < 0)
    222 		err(1, "lseek failed");
    223 	/* See if we have to break up the transfer... */
    224 	if (smallio) {
    225 		const char *bp;	/* pointer into buf */
    226 		int left;	/* bytes left to go */
    227 		int n;		/* number to do this time around */
    228 		int rv;		/* syscall return value */
    229 		bp = buf;
    230 		left = size;
    231 		while (left > 0) {
    232 			n = (left > 8192) ? 8192 : left;
    233 			rv = write(fd, bp, n);
    234 			if (rv < 0)
    235 				err(1, "write failed");
    236 			if (rv != n)
    237 				errx(1, "write: wanted %d, got %d", n, rv);
    238 			bp += n;
    239 			left -= n;
    240 		}
    241 	} else {
    242 		int rv;
    243 		rv = write(fd, buf, size);
    244 		if (rv < 0)
    245 			err(1, "write failed");
    246 		if (rv != size)
    247 			errx(1, "write: wanted %d, got %d", size, rv);
    248 	}
    249 }
    250 /*
    251  * Never-fail versions of malloc() and realloc(), and an allocation
    252  *  routine (which also never fails) for allocating memory that will
    253  *  never be freed until exit.
    254  */
    255 
    256 /*
    257  * Never-fail malloc.
    258  */
    259 static void *
    260 nfmalloc(size_t nb, const char *tag)
    261 {
    262 	void *rv;
    263 
    264 	rv = malloc(nb);
    265 	if (rv)
    266 		return (rv);
    267 	err(1, "Can't allocate %lu bytes for %s",
    268 	    (unsigned long int) nb, tag);
    269 }
    270 /*
    271  * Never-fail realloc.
    272  */
    273 static void *
    274 nfrealloc(void *blk, size_t nb, const char *tag)
    275 {
    276 	void *rv;
    277 
    278 	rv = realloc(blk, nb);
    279 	if (rv)
    280 		return (rv);
    281 	err(1, "Can't re-allocate %lu bytes for %s",
    282 	    (unsigned long int) nb, tag);
    283 }
    284 /*
    285  * Allocate memory that will never be freed or reallocated.  Arguably
    286  *  this routine should handle small allocations by chopping up pages,
    287  *  but that's not worth the bother; it's not called more than a
    288  *  handful of times per run, and if the allocations are that small the
    289  *  waste in giving each one its own page is ignorable.
    290  */
    291 static void *
    292 alloconce(size_t nb, const char *tag)
    293 {
    294 	void *rv;
    295 
    296 	rv = mmap(0, nb, PROT_READ | PROT_WRITE, MAP_ANON | MAP_PRIVATE, -1, 0);
    297 	if (rv != MAP_FAILED)
    298 		return (rv);
    299 	err(1, "Can't map %lu bytes for %s",
    300 	    (unsigned long int) nb, tag);
    301 }
    302 /*
    303  * Load the cgs and csums off disk.  Also allocates the space to load
    304  *  them into and initializes the per-cg flags.
    305  */
    306 static void
    307 loadcgs(void)
    308 {
    309 	int cg;
    310 	char *cgp;
    311 
    312 	cgblksz = roundup(oldsb->fs_cgsize, oldsb->fs_fsize);
    313 	cgs = nfmalloc(oldsb->fs_ncg * sizeof(struct cg *), "cg pointers");
    314 	cgp = alloconce(oldsb->fs_ncg * cgblksz, "cgs");
    315 	cgflags = nfmalloc(oldsb->fs_ncg, "cg flags");
    316 	csums = nfmalloc(oldsb->fs_cssize, "cg summary");
    317 	for (cg = 0; cg < oldsb->fs_ncg; cg++) {
    318 		cgs[cg] = (struct cg *) cgp;
    319 		readat(fsbtodb(oldsb, cgtod(oldsb, cg)), cgp, cgblksz);
    320 		cgflags[cg] = 0;
    321 		cgp += cgblksz;
    322 	}
    323 	readat(fsbtodb(oldsb, oldsb->fs_csaddr), csums, oldsb->fs_cssize);
    324 }
    325 /*
    326  * Set n bits, starting with bit #base, in the bitmap pointed to by
    327  *  bitvec (which is assumed to be large enough to include bits base
    328  *  through base+n-1).
    329  */
    330 static void
    331 set_bits(unsigned char *bitvec, unsigned int base, unsigned int n)
    332 {
    333 	if (n < 1)
    334 		return;		/* nothing to do */
    335 	if (base & 7) {		/* partial byte at beginning */
    336 		if (n <= 8 - (base & 7)) {	/* entirely within one byte */
    337 			bitvec[base >> 3] |= (~((~0U) << n)) << (base & 7);
    338 			return;
    339 		}
    340 		bitvec[base >> 3] |= (~0U) << (base & 7);
    341 		n -= 8 - (base & 7);
    342 		base = (base & ~7) + 8;
    343 	}
    344 	if (n >= 8) {		/* do full bytes */
    345 		memset(bitvec + (base >> 3), 0xff, n >> 3);
    346 		base += n & ~7;
    347 		n &= 7;
    348 	}
    349 	if (n) {		/* partial byte at end */
    350 		bitvec[base >> 3] |= ~((~0U) << n);
    351 	}
    352 }
    353 /*
    354  * Clear n bits, starting with bit #base, in the bitmap pointed to by
    355  *  bitvec (which is assumed to be large enough to include bits base
    356  *  through base+n-1).  Code parallels set_bits().
    357  */
    358 static void
    359 clr_bits(unsigned char *bitvec, int base, int n)
    360 {
    361 	if (n < 1)
    362 		return;
    363 	if (base & 7) {
    364 		if (n <= 8 - (base & 7)) {
    365 			bitvec[base >> 3] &= ~((~((~0U) << n)) << (base & 7));
    366 			return;
    367 		}
    368 		bitvec[base >> 3] &= ~((~0U) << (base & 7));
    369 		n -= 8 - (base & 7);
    370 		base = (base & ~7) + 8;
    371 	}
    372 	if (n >= 8) {
    373 		bzero(bitvec + (base >> 3), n >> 3);
    374 		base += n & ~7;
    375 		n &= 7;
    376 	}
    377 	if (n) {
    378 		bitvec[base >> 3] &= (~0U) << n;
    379 	}
    380 }
    381 /*
    382  * Test whether bit #bit is set in the bitmap pointed to by bitvec.
    383  */
    384 INLINE static int
    385 bit_is_set(unsigned char *bitvec, int bit)
    386 {
    387 	return (bitvec[bit >> 3] & (1 << (bit & 7)));
    388 }
    389 /*
    390  * Test whether bit #bit is clear in the bitmap pointed to by bitvec.
    391  */
    392 INLINE static int
    393 bit_is_clr(unsigned char *bitvec, int bit)
    394 {
    395 	return (!bit_is_set(bitvec, bit));
    396 }
    397 /*
    398  * Test whether a whole block of bits is set in a bitmap.  This is
    399  *  designed for testing (aligned) disk blocks in a bit-per-frag
    400  *  bitmap; it has assumptions wired into it based on that, essentially
    401  *  that the entire block fits into a single byte.  This returns true
    402  *  iff _all_ the bits are set; it is not just the complement of
    403  *  blk_is_clr on the same arguments (unless blkfrags==1).
    404  */
    405 INLINE static int
    406 blk_is_set(unsigned char *bitvec, int blkbase, int blkfrags)
    407 {
    408 	unsigned int mask;
    409 
    410 	mask = (~((~0U) << blkfrags)) << (blkbase & 7);
    411 	return ((bitvec[blkbase >> 3] & mask) == mask);
    412 }
    413 /*
    414  * Test whether a whole block of bits is clear in a bitmap.  See
    415  *  blk_is_set (above) for assumptions.  This returns true iff _all_
    416  *  the bits are clear; it is not just the complement of blk_is_set on
    417  *  the same arguments (unless blkfrags==1).
    418  */
    419 INLINE static int
    420 blk_is_clr(unsigned char *bitvec, int blkbase, int blkfrags)
    421 {
    422 	unsigned int mask;
    423 
    424 	mask = (~((~0U) << blkfrags)) << (blkbase & 7);
    425 	return ((bitvec[blkbase >> 3] & mask) == 0);
    426 }
    427 /*
    428  * Initialize a new cg.  Called when growing.  Assumes memory has been
    429  *  allocated but not otherwise set up.  This code sets the fields of
    430  *  the cg, initializes the bitmaps (and cluster summaries, if
    431  *  applicable), updates both per-cylinder summary info and the global
    432  *  summary info in newsb; it also writes out new inodes for the cg.
    433  *
    434  * This code knows it can never be called for cg 0, which makes it a
    435  *  bit simpler than it would otherwise be.
    436  */
    437 static void
    438 initcg(int cgn)
    439 {
    440 	struct cg *cg;		/* The in-core cg, of course */
    441 	int base;		/* Disk address of cg base */
    442 	int dlow;		/* Size of pre-cg data area */
    443 	int dhigh;		/* Offset of post-inode data area, from base */
    444 	int dmax;		/* Offset of end of post-inode data area */
    445 	int i;			/* Generic loop index */
    446 	int n;			/* Generic count */
    447 
    448 	cg = cgs[cgn];
    449 	/* Place the data areas */
    450 	base = cgbase(newsb, cgn);
    451 	dlow = cgsblock(newsb, cgn) - base;
    452 	dhigh = cgdmin(newsb, cgn) - base;
    453 	dmax = newsb->fs_size - base;
    454 	if (dmax > newsb->fs_fpg)
    455 		dmax = newsb->fs_fpg;
    456 	/*
    457          * Clear out the cg - assumes all-0-bytes is the correct way
    458          * to initialize fields we don't otherwise touch, which is
    459          * perhaps not the right thing to do, but it's what fsck and
    460          * mkfs do.
    461          */
    462 	bzero(cg, newsb->fs_cgsize);
    463 	cg->cg_time = newsb->fs_time;
    464 	cg->cg_magic = CG_MAGIC;
    465 	cg->cg_cgx = cgn;
    466 	cg->cg_old_ncyl = newsb->fs_old_cpg;
    467 	/* fsck whines if the cg->cg_old_ncyl value in the last cg is fs_old_cpg
    468 	 * instead of zero, when fs_old_cpg is the correct value. */
    469 	/* XXX fix once fsck is fixed */
    470 	if ((cgn == newsb->fs_ncg - 1) /* && (newsb->fs_old_ncyl % newsb->fs_old_cpg) */ ) {
    471 		cg->cg_old_ncyl = newsb->fs_old_ncyl % newsb->fs_old_cpg;
    472 	}
    473 	cg->cg_niblk = newsb->fs_ipg;
    474 	cg->cg_ndblk = dmax;
    475 	/* Set up the bitmap pointers.  We have to be careful to lay out the
    476 	 * cg _exactly_ the way mkfs and fsck do it, since fsck compares the
    477 	 * _entire_ cg against a recomputed cg, and whines if there is any
    478 	 * mismatch, including the bitmap offsets. */
    479 	/* XXX update this comment when fsck is fixed */
    480 	cg->cg_old_btotoff = &cg->cg_space[0] - (unsigned char *) cg;
    481 	cg->cg_old_boff = cg->cg_old_btotoff
    482 	    + (newsb->fs_old_cpg * sizeof(int32_t));
    483 	cg->cg_iusedoff = cg->cg_old_boff +
    484 	    (newsb->fs_old_cpg * newsb->fs_old_nrpos * sizeof(int16_t));
    485 	cg->cg_freeoff = cg->cg_iusedoff + howmany(newsb->fs_ipg, NBBY);
    486 	if (newsb->fs_contigsumsize > 0) {
    487 		cg->cg_nclusterblks = cg->cg_ndblk / newsb->fs_frag;
    488 		cg->cg_clustersumoff = cg->cg_freeoff +
    489 		    howmany(newsb->fs_old_cpg * newsb->fs_old_spc / NSPF(newsb),
    490 		    NBBY) - sizeof(int32_t);
    491 		cg->cg_clustersumoff =
    492 		    roundup(cg->cg_clustersumoff, sizeof(int32_t));
    493 		cg->cg_clusteroff = cg->cg_clustersumoff +
    494 		    ((newsb->fs_contigsumsize + 1) * sizeof(int32_t));
    495 		cg->cg_nextfreeoff = cg->cg_clusteroff +
    496 		    howmany(newsb->fs_old_cpg * newsb->fs_old_spc / NSPB(newsb),
    497 		    NBBY);
    498 		n = dlow / newsb->fs_frag;
    499 		if (n > 0) {
    500 			set_bits(cg_clustersfree(cg, 0), 0, n);
    501 			cg_clustersum(cg, 0)[(n > newsb->fs_contigsumsize) ?
    502 			    newsb->fs_contigsumsize : n]++;
    503 		}
    504 	} else {
    505 		cg->cg_nextfreeoff = cg->cg_freeoff +
    506 		    howmany(newsb->fs_old_cpg * newsb->fs_old_spc / NSPF(newsb),
    507 		    NBBY);
    508 	}
    509 	/* Mark the data areas as free; everything else is marked busy by the
    510 	 * bzero up at the top. */
    511 	set_bits(cg_blksfree(cg, 0), 0, dlow);
    512 	set_bits(cg_blksfree(cg, 0), dhigh, dmax - dhigh);
    513 	/* Initialize summary info */
    514 	cg->cg_cs.cs_ndir = 0;
    515 	cg->cg_cs.cs_nifree = newsb->fs_ipg;
    516 	cg->cg_cs.cs_nbfree = dlow / newsb->fs_frag;
    517 	cg->cg_cs.cs_nffree = 0;
    518 
    519 	/* This is the simplest way of doing this; we perhaps could compute
    520 	 * the correct cg_blktot()[] and cg_blks()[] values other ways, but it
    521 	 * would be complicated and hardly seems worth the effort.  (The
    522 	 * reason there isn't frag-at-beginning and frag-at-end code here,
    523 	 * like the code below for the post-inode data area, is that the
    524 	 * pre-sb data area always starts at 0, and thus is block-aligned, and
    525 	 * always ends at the sb, which is block-aligned.) */
    526 	for (i = 0; i < dlow; i += newsb->fs_frag) {
    527 		cg_blktot(cg, 0)[cbtocylno(newsb, i)]++;
    528 		cg_blks(newsb, cg, cbtocylno(newsb, i), 0)[cbtorpos(newsb, i)]++;
    529 	}
    530 	/* Deal with a partial block at the beginning of the post-inode area.
    531 	 * I'm not convinced this can happen - I think the inodes are always
    532 	 * block-aligned and always an integral number of blocks - but it's
    533 	 * cheap to do the right thing just in case. */
    534 	if (dhigh % newsb->fs_frag) {
    535 		n = newsb->fs_frag - (dhigh % newsb->fs_frag);
    536 		cg->cg_frsum[n]++;
    537 		cg->cg_cs.cs_nffree += n;
    538 		dhigh += n;
    539 	}
    540 	n = (dmax - dhigh) / newsb->fs_frag;
    541 	/* We have n full-size blocks in the post-inode data area. */
    542 	if (n > 0) {
    543 		cg->cg_cs.cs_nbfree += n;
    544 		if (newsb->fs_contigsumsize > 0) {
    545 			i = dhigh / newsb->fs_frag;
    546 			set_bits(cg_clustersfree(cg, 0), i, n);
    547 			cg_clustersum(cg, 0)[(n > newsb->fs_contigsumsize) ?
    548 			    newsb->fs_contigsumsize : n]++;
    549 		}
    550 		for (i = n; i > 0; i--) {
    551 			cg_blktot(cg, 0)[cbtocylno(newsb, dhigh)]++;
    552 			cg_blks(newsb, cg,
    553 			    cbtocylno(newsb, dhigh), 0)[cbtorpos(newsb,
    554 				dhigh)]++;
    555 			dhigh += newsb->fs_frag;
    556 		}
    557 	}
    558 	/* Deal with any leftover frag at the end of the cg. */
    559 	i = dmax - dhigh;
    560 	if (i) {
    561 		cg->cg_frsum[i]++;
    562 		cg->cg_cs.cs_nffree += i;
    563 	}
    564 	/* Update the csum info. */
    565 	csums[cgn] = cg->cg_cs;
    566 	newsb->fs_cstotal.cs_nffree += cg->cg_cs.cs_nffree;
    567 	newsb->fs_cstotal.cs_nbfree += cg->cg_cs.cs_nbfree;
    568 	newsb->fs_cstotal.cs_nifree += cg->cg_cs.cs_nifree;
    569 	/* Write out the cleared inodes. */
    570 	writeat(fsbtodb(newsb, cgimin(newsb, cgn)), zinodes,
    571 	    newsb->fs_ipg * sizeof(struct ufs1_dinode));
    572 	/* Dirty the cg. */
    573 	cgflags[cgn] |= CGF_DIRTY;
    574 }
    575 /*
    576  * Find free space, at least nfrags consecutive frags of it.  Pays no
    577  *  attention to block boundaries, but refuses to straddle cg
    578  *  boundaries, even if the disk blocks involved are in fact
    579  *  consecutive.  Return value is the frag number of the first frag of
    580  *  the block, or -1 if no space was found.  Uses newsb for sb values,
    581  *  and assumes the cgs[] structures correctly describe the area to be
    582  *  searched.
    583  *
    584  * XXX is there a bug lurking in the ignoring of block boundaries by
    585  *  the routine used by fragmove() in evict_data()?  Can an end-of-file
    586  *  frag legally straddle a block boundary?  If not, this should be
    587  *  cloned and fixed to stop at block boundaries for that use.  The
    588  *  current one may still be needed for csum info motion, in case that
    589  *  takes up more than a whole block (is the csum info allowed to begin
    590  *  partway through a block and continue into the following block?).
    591  *
    592  * If we wrap off the end of the filesystem back to the beginning, we
    593  *  can end up searching the end of the filesystem twice.  I ignore
    594  *  this inefficiency, since if that happens we're going to croak with
    595  *  a no-space error anyway, so it happens at most once.
    596  */
    597 static int
    598 find_freespace(unsigned int nfrags)
    599 {
    600 	static int hand = 0;	/* hand rotates through all frags in the fs */
    601 	int cgsize;		/* size of the cg hand currently points into */
    602 	int cgn;		/* number of cg hand currently points into */
    603 	int fwc;		/* frag-within-cg number of frag hand points
    604 				 * to */
    605 	int run;		/* length of run of free frags seen so far */
    606 	int secondpass;		/* have we wrapped from end of fs to
    607 				 * beginning? */
    608 	unsigned char *bits;	/* cg_blksfree()[] for cg hand points into */
    609 
    610 	cgn = dtog(newsb, hand);
    611 	fwc = dtogd(newsb, hand);
    612 	secondpass = (hand == 0);
    613 	run = 0;
    614 	bits = cg_blksfree(cgs[cgn], 0);
    615 	cgsize = cgs[cgn]->cg_ndblk;
    616 	while (1) {
    617 		if (bit_is_set(bits, fwc)) {
    618 			run++;
    619 			if (run >= nfrags)
    620 				return (hand + 1 - run);
    621 		} else {
    622 			run = 0;
    623 		}
    624 		hand++;
    625 		fwc++;
    626 		if (fwc >= cgsize) {
    627 			fwc = 0;
    628 			cgn++;
    629 			if (cgn >= newsb->fs_ncg) {
    630 				hand = 0;
    631 				if (secondpass)
    632 					return (-1);
    633 				secondpass = 1;
    634 				cgn = 0;
    635 			}
    636 			bits = cg_blksfree(cgs[cgn], 0);
    637 			cgsize = cgs[cgn]->cg_ndblk;
    638 			run = 0;
    639 		}
    640 	}
    641 }
    642 /*
    643  * Find a free block of disk space.  Finds an entire block of frags,
    644  *  all of which are free.  Return value is the frag number of the
    645  *  first frag of the block, or -1 if no space was found.  Uses newsb
    646  *  for sb values, and assumes the cgs[] structures correctly describe
    647  *  the area to be searched.
    648  *
    649  * See find_freespace(), above, for remarks about hand wrapping around.
    650  */
    651 static int
    652 find_freeblock(void)
    653 {
    654 	static int hand = 0;	/* hand rotates through all frags in fs */
    655 	int cgn;		/* cg number of cg hand points into */
    656 	int fwc;		/* frag-within-cg number of frag hand points
    657 				 * to */
    658 	int cgsize;		/* size of cg hand points into */
    659 	int secondpass;		/* have we wrapped from end to beginning? */
    660 	unsigned char *bits;	/* cg_blksfree()[] for cg hand points into */
    661 
    662 	cgn = dtog(newsb, hand);
    663 	fwc = dtogd(newsb, hand);
    664 	secondpass = (hand == 0);
    665 	bits = cg_blksfree(cgs[cgn], 0);
    666 	cgsize = blknum(newsb, cgs[cgn]->cg_ndblk);
    667 	while (1) {
    668 		if (blk_is_set(bits, fwc, newsb->fs_frag))
    669 			return (hand);
    670 		fwc += newsb->fs_frag;
    671 		hand += newsb->fs_frag;
    672 		if (fwc >= cgsize) {
    673 			fwc = 0;
    674 			cgn++;
    675 			if (cgn >= newsb->fs_ncg) {
    676 				hand = 0;
    677 				if (secondpass)
    678 					return (-1);
    679 				secondpass = 1;
    680 				cgn = 0;
    681 			}
    682 			bits = cg_blksfree(cgs[cgn], 0);
    683 			cgsize = blknum(newsb, cgs[cgn]->cg_ndblk);
    684 		}
    685 	}
    686 }
    687 /*
    688  * Find a free inode, returning its inumber or -1 if none was found.
    689  *  Uses newsb for sb values, and assumes the cgs[] structures
    690  *  correctly describe the area to be searched.
    691  *
    692  * See find_freespace(), above, for remarks about hand wrapping around.
    693  */
    694 static int
    695 find_freeinode(void)
    696 {
    697 	static int hand = 0;	/* hand rotates through all inodes in fs */
    698 	int cgn;		/* cg number of cg hand points into */
    699 	int iwc;		/* inode-within-cg number of inode hand points
    700 				 * to */
    701 	int secondpass;		/* have we wrapped from end to beginning? */
    702 	unsigned char *bits;	/* cg_inosused()[] for cg hand points into */
    703 
    704 	cgn = hand / newsb->fs_ipg;
    705 	iwc = hand % newsb->fs_ipg;
    706 	secondpass = (hand == 0);
    707 	bits = cg_inosused(cgs[cgn], 0);
    708 	while (1) {
    709 		if (bit_is_clr(bits, iwc))
    710 			return (hand);
    711 		hand++;
    712 		iwc++;
    713 		if (iwc >= newsb->fs_ipg) {
    714 			iwc = 0;
    715 			cgn++;
    716 			if (cgn >= newsb->fs_ncg) {
    717 				hand = 0;
    718 				if (secondpass)
    719 					return (-1);
    720 				secondpass = 1;
    721 				cgn = 0;
    722 			}
    723 			bits = cg_inosused(cgs[cgn], 0);
    724 		}
    725 	}
    726 }
    727 /*
    728  * Mark a frag as free.  Sets the frag's bit in the cg_blksfree bitmap
    729  *  for the appropriate cg, and marks the cg as dirty.
    730  */
    731 static void
    732 free_frag(int fno)
    733 {
    734 	int cgn;
    735 
    736 	cgn = dtog(newsb, fno);
    737 	set_bits(cg_blksfree(cgs[cgn], 0), dtogd(newsb, fno), 1);
    738 	cgflags[cgn] |= CGF_DIRTY | CGF_BLKMAPS;
    739 }
    740 /*
    741  * Allocate a frag.  Clears the frag's bit in the cg_blksfree bitmap
    742  *  for the appropriate cg, and marks the cg as dirty.
    743  */
    744 static void
    745 alloc_frag(int fno)
    746 {
    747 	int cgn;
    748 
    749 	cgn = dtog(newsb, fno);
    750 	clr_bits(cg_blksfree(cgs[cgn], 0), dtogd(newsb, fno), 1);
    751 	cgflags[cgn] |= CGF_DIRTY | CGF_BLKMAPS;
    752 }
    753 /*
    754  * Fix up the csum array.  If shrinking, this involves freeing zero or
    755  *  more frags; if growing, it involves allocating them, or if the
    756  *  frags being grown into aren't free, finding space elsewhere for the
    757  *  csum info.  (If the number of occupied frags doesn't change,
    758  *  nothing happens here.)
    759  */
    760 static void
    761 csum_fixup(void)
    762 {
    763 	int nold;		/* # frags in old csum info */
    764 	int ntot;		/* # frags in new csum info */
    765 	int nnew;		/* ntot-nold */
    766 	int newloc;		/* new location for csum info, if necessary */
    767 	int i;			/* generic loop index */
    768 	int j;			/* generic loop index */
    769 	int f;			/* "from" frag number, if moving */
    770 	int t;			/* "to" frag number, if moving */
    771 	int cgn;		/* cg number, used when shrinking */
    772 
    773 	ntot = howmany(newsb->fs_cssize, newsb->fs_fsize);
    774 	nold = howmany(oldsb->fs_cssize, newsb->fs_fsize);
    775 	nnew = ntot - nold;
    776 	/* First, if there's no change in frag counts, it's easy. */
    777 	if (nnew == 0)
    778 		return;
    779 	/* Next, if we're shrinking, it's almost as easy.  Just free up any
    780 	 * frags in the old area we no longer need. */
    781 	if (nnew < 0) {
    782 		for ((i = newsb->fs_csaddr + ntot - 1), (j = nnew);
    783 		    j < 0;
    784 		    i--, j++) {
    785 			free_frag(i);
    786 		}
    787 		return;
    788 	}
    789 	/* We must be growing.  Check to see that the new csum area fits
    790 	 * within the filesystem.  I think this can never happen, since for
    791 	 * the csum area to grow, we must be adding at least one cg, so the
    792 	 * old csum area can't be this close to the end of the new filesystem.
    793 	 * But it's a cheap check. */
    794 	/* XXX what if csum info is at end of cg and grows into next cg, what
    795 	 * if it spills over onto the next cg's backup superblock?  Can this
    796 	 * happen? */
    797 	if (newsb->fs_csaddr + ntot <= newsb->fs_size) {
    798 		/* Okay, it fits - now,  see if the space we want is free. */
    799 		for ((i = newsb->fs_csaddr + nold), (j = nnew);
    800 		    j > 0;
    801 		    i++, j--) {
    802 			cgn = dtog(newsb, i);
    803 			if (bit_is_clr(cg_blksfree(cgs[cgn], 0),
    804 				dtogd(newsb, i)))
    805 				break;
    806 		}
    807 		if (j <= 0) {
    808 			/* Win win - all the frags we want are free. Allocate
    809 			 * 'em and we're all done.  */
    810 			for ((i = newsb->fs_csaddr + ntot - nnew), (j = nnew); j > 0; i++, j--) {
    811 				alloc_frag(i);
    812 			}
    813 			return;
    814 		}
    815 	}
    816 	/* We have to move the csum info, sigh.  Look for new space, free old
    817 	 * space, and allocate new.  Update fs_csaddr.  We don't copy anything
    818 	 * on disk at this point; the csum info will be written to the
    819 	 * then-current fs_csaddr as part of the final flush. */
    820 	newloc = find_freespace(ntot);
    821 	if (newloc < 0) {
    822 		printf("Sorry, no space available for new csums\n");
    823 		exit(1);
    824 	}
    825 	for (i = 0, f = newsb->fs_csaddr, t = newloc; i < ntot; i++, f++, t++) {
    826 		if (i < nold) {
    827 			free_frag(f);
    828 		}
    829 		alloc_frag(t);
    830 	}
    831 	newsb->fs_csaddr = newloc;
    832 }
    833 /*
    834  * Recompute newsb->fs_dsize.  Just scans all cgs, adding the number of
    835  *  data blocks in that cg to the total.
    836  */
    837 static void
    838 recompute_fs_dsize(void)
    839 {
    840 	int i;
    841 
    842 	newsb->fs_dsize = 0;
    843 	for (i = 0; i < newsb->fs_ncg; i++) {
    844 		int dlow;	/* size of before-sb data area */
    845 		int dhigh;	/* offset of post-inode data area */
    846 		int dmax;	/* total size of cg */
    847 		int base;	/* base of cg, since cgsblock() etc add it in */
    848 		base = cgbase(newsb, i);
    849 		dlow = cgsblock(newsb, i) - base;
    850 		dhigh = cgdmin(newsb, i) - base;
    851 		dmax = newsb->fs_size - base;
    852 		if (dmax > newsb->fs_fpg)
    853 			dmax = newsb->fs_fpg;
    854 		newsb->fs_dsize += dlow + dmax - dhigh;
    855 	}
    856 	/* Space in cg 0 before cgsblock is boot area, not free space! */
    857 	newsb->fs_dsize -= cgsblock(newsb, 0) - cgbase(newsb, 0);
    858 	/* And of course the csum info takes up space. */
    859 	newsb->fs_dsize -= howmany(newsb->fs_cssize, newsb->fs_fsize);
    860 }
    861 /*
    862  * Return the current time.  We call this and assign, rather than
    863  *  calling time() directly, as insulation against OSes where fs_time
    864  *  is not a time_t.
    865  */
    866 static time_t
    867 timestamp(void)
    868 {
    869 	time_t t;
    870 
    871 	time(&t);
    872 	return (t);
    873 }
    874 /*
    875  * Grow the filesystem.
    876  */
    877 static void
    878 grow(void)
    879 {
    880 	int i;
    881 
    882 	/* Update the timestamp. */
    883 	newsb->fs_time = timestamp();
    884 	/* Allocate and clear the new-inode area, in case we add any cgs. */
    885 	zinodes = alloconce(newsb->fs_ipg * sizeof(struct ufs1_dinode),
    886                             "zeroed inodes");
    887 	bzero(zinodes, newsb->fs_ipg * sizeof(struct ufs1_dinode));
    888 	/* Update the size. */
    889 	newsb->fs_size = dbtofsb(newsb, newsize);
    890 	/* Did we actually not grow?  (This can happen if newsize is less than
    891 	 * a frag larger than the old size - unlikely, but no excuse to
    892 	 * misbehave if it happens.) */
    893 	if (newsb->fs_size == oldsb->fs_size)
    894 		return;
    895 	/* Check that the new last sector (frag, actually) is writable.  Since
    896 	 * it's at least one frag larger than it used to be, we know we aren't
    897 	 * overwriting anything important by this.  (The choice of sbbuf as
    898 	 * what to write is irrelevant; it's just something handy that's known
    899 	 * to be at least one frag in size.) */
    900 	writeat(newsb->fs_size - 1, &sbbuf, newsb->fs_fsize);
    901 	/* Update fs_old_ncyl and fs_ncg. */
    902 	newsb->fs_old_ncyl = (newsb->fs_size * NSPF(newsb)) / newsb->fs_old_spc;
    903 	newsb->fs_ncg = howmany(newsb->fs_old_ncyl, newsb->fs_old_cpg);
    904 	/* Does the last cg end before the end of its inode area? There is no
    905 	 * reason why this couldn't be handled, but it would complicate a lot
    906 	 * of code (in all filesystem code - fsck, kernel, etc) because of the
    907 	 * potential partial inode area, and the gain in space would be
    908 	 * minimal, at most the pre-sb data area. */
    909 	if (cgdmin(newsb, newsb->fs_ncg - 1) > newsb->fs_size) {
    910 		newsb->fs_ncg--;
    911 		newsb->fs_old_ncyl = newsb->fs_ncg * newsb->fs_old_cpg;
    912 		newsb->fs_size = (newsb->fs_old_ncyl * newsb->fs_old_spc) / NSPF(newsb);
    913 		printf("Warning: last cylinder group is too small;\n");
    914 		printf("    dropping it.  New size = %lu.\n",
    915 		    (unsigned long int) fsbtodb(newsb, newsb->fs_size));
    916 	}
    917 	/* Find out how big the csum area is, and realloc csums if bigger. */
    918 	newsb->fs_cssize = fragroundup(newsb,
    919 	    newsb->fs_ncg * sizeof(struct csum));
    920 	if (newsb->fs_cssize > oldsb->fs_cssize)
    921 		csums = nfrealloc(csums, newsb->fs_cssize, "new cg summary");
    922 	/* If we're adding any cgs, realloc structures and set up the new cgs. */
    923 	if (newsb->fs_ncg > oldsb->fs_ncg) {
    924 		char *cgp;
    925 		cgs = nfrealloc(cgs, newsb->fs_ncg * sizeof(struct cg *),
    926                                 "cg pointers");
    927 		cgflags = nfrealloc(cgflags, newsb->fs_ncg, "cg flags");
    928 		bzero(cgflags + oldsb->fs_ncg, newsb->fs_ncg - oldsb->fs_ncg);
    929 		cgp = alloconce((newsb->fs_ncg - oldsb->fs_ncg) * cgblksz,
    930                                 "cgs");
    931 		for (i = oldsb->fs_ncg; i < newsb->fs_ncg; i++) {
    932 			cgs[i] = (struct cg *) cgp;
    933 			initcg(i);
    934 			cgp += cgblksz;
    935 		}
    936 		cgs[oldsb->fs_ncg - 1]->cg_old_ncyl = oldsb->fs_old_cpg;
    937 		cgflags[oldsb->fs_ncg - 1] |= CGF_DIRTY;
    938 	}
    939 	/* If the old fs ended partway through a cg, we have to update the old
    940 	 * last cg (though possibly not to a full cg!). */
    941 	if (oldsb->fs_size % oldsb->fs_fpg) {
    942 		struct cg *cg;
    943 		int newcgsize;
    944 		int prevcgtop;
    945 		int oldcgsize;
    946 		cg = cgs[oldsb->fs_ncg - 1];
    947 		cgflags[oldsb->fs_ncg - 1] |= CGF_DIRTY | CGF_BLKMAPS;
    948 		prevcgtop = oldsb->fs_fpg * (oldsb->fs_ncg - 1);
    949 		newcgsize = newsb->fs_size - prevcgtop;
    950 		if (newcgsize > newsb->fs_fpg)
    951 			newcgsize = newsb->fs_fpg;
    952 		oldcgsize = oldsb->fs_size % oldsb->fs_fpg;
    953 		set_bits(cg_blksfree(cg, 0), oldcgsize, newcgsize - oldcgsize);
    954 		cg->cg_old_ncyl = howmany(newcgsize * NSPF(newsb), newsb->fs_old_spc);
    955 		cg->cg_ndblk = newcgsize;
    956 	}
    957 	/* Fix up the csum info, if necessary. */
    958 	csum_fixup();
    959 	/* Make fs_dsize match the new reality. */
    960 	recompute_fs_dsize();
    961 }
    962 /*
    963  * Call (*fn)() for each inode, passing the inode and its inumber.  The
    964  *  number of cylinder groups is pased in, so this can be used to map
    965  *  over either the old or the new filesystem's set of inodes.
    966  */
    967 static void
    968      map_inodes(void (*fn) (struct ufs1_dinode * di, unsigned int, void *arg), int ncg, void *cbarg) {
    969 	int i;
    970 	int ni;
    971 
    972 	ni = oldsb->fs_ipg * ncg;
    973 	for (i = 0; i < ni; i++)
    974 		(*fn) (inodes + i, i, cbarg);
    975 }
    976 /* Values for the third argument to the map function for
    977  * map_inode_data_blocks.  MDB_DATA indicates the block is contains
    978  * file data; MDB_INDIR_PRE and MDB_INDIR_POST indicate that it's an
    979  * indirect block.  The MDB_INDIR_PRE call is made before the indirect
    980  * block pointers are followed and the pointed-to blocks scanned,
    981  * MDB_INDIR_POST after.
    982  */
    983 #define MDB_DATA       1
    984 #define MDB_INDIR_PRE  2
    985 #define MDB_INDIR_POST 3
    986 
    987 typedef void (*mark_callback_t) (unsigned int blocknum, unsigned int nfrags, unsigned int blksize, int opcode);
    988 
    989 /* Helper function - handles a data block.  Calls the callback
    990  * function and returns number of bytes occupied in file (actually,
    991  * rounded up to a frag boundary).  The name is historical.  */
    992 static int
    993 markblk(mark_callback_t fn, struct ufs1_dinode * di, int bn, off_t o)
    994 {
    995 	int sz;
    996 	int nb;
    997 	if (o >= di->di_size)
    998 		return (0);
    999 	sz = dblksize(newsb, di, lblkno(newsb, o));
   1000 	nb = (sz > di->di_size - o) ? di->di_size - o : sz;
   1001 	if (bn)
   1002 		(*fn) (bn, numfrags(newsb, sz), nb, MDB_DATA);
   1003 	return (sz);
   1004 }
   1005 /* Helper function - handles an indirect block.  Makes the
   1006  * MDB_INDIR_PRE callback for the indirect block, loops over the
   1007  * pointers and recurses, and makes the MDB_INDIR_POST callback.
   1008  * Returns the number of bytes occupied in file, as does markblk().
   1009  * For the sake of update_for_data_move(), we read the indirect block
   1010  * _after_ making the _PRE callback.  The name is historical.  */
   1011 static int
   1012 markiblk(mark_callback_t fn, struct ufs1_dinode * di, int bn, off_t o, int lev)
   1013 {
   1014 	int i;
   1015 	int j;
   1016 	int tot;
   1017 	static int32_t indirblk1[howmany(MAXBSIZE, sizeof(int32_t))];
   1018 	static int32_t indirblk2[howmany(MAXBSIZE, sizeof(int32_t))];
   1019 	static int32_t indirblk3[howmany(MAXBSIZE, sizeof(int32_t))];
   1020 	static int32_t *indirblks[3] = {
   1021 		&indirblk1[0], &indirblk2[0], &indirblk3[0]
   1022 	};
   1023 	if (lev < 0)
   1024 		return (markblk(fn, di, bn, o));
   1025 	if (bn == 0) {
   1026 		for (i = newsb->fs_bsize;
   1027 		    lev >= 0;
   1028 		    i *= NINDIR(newsb), lev--);
   1029 		return (i);
   1030 	}
   1031 	(*fn) (bn, newsb->fs_frag, newsb->fs_bsize, MDB_INDIR_PRE);
   1032 	readat(fsbtodb(newsb, bn), indirblks[lev], newsb->fs_bsize);
   1033 	tot = 0;
   1034 	for (i = 0; i < NINDIR(newsb); i++) {
   1035 		j = markiblk(fn, di, indirblks[lev][i], o, lev - 1);
   1036 		if (j == 0)
   1037 			break;
   1038 		o += j;
   1039 		tot += j;
   1040 	}
   1041 	(*fn) (bn, newsb->fs_frag, newsb->fs_bsize, MDB_INDIR_POST);
   1042 	return (tot);
   1043 }
   1044 
   1045 
   1046 /*
   1047  * Call (*fn)() for each data block for an inode.  This routine assumes
   1048  *  the inode is known to be of a type that has data blocks (file,
   1049  *  directory, or non-fast symlink).  The called function is:
   1050  *
   1051  * (*fn)(unsigned int blkno, unsigned int nf, unsigned int nb, int op)
   1052  *
   1053  *  where blkno is the frag number, nf is the number of frags starting
   1054  *  at blkno (always <= fs_frag), nb is the number of bytes that belong
   1055  *  to the file (usually nf*fs_frag, often less for the last block/frag
   1056  *  of a file).
   1057  */
   1058 static void
   1059 map_inode_data_blocks(struct ufs1_dinode * di, mark_callback_t fn)
   1060 {
   1061 	off_t o;		/* offset within  inode */
   1062 	int inc;		/* increment for o - maybe should be off_t? */
   1063 	int b;			/* index within di_db[] and di_ib[] arrays */
   1064 
   1065 	/* Scan the direct blocks... */
   1066 	o = 0;
   1067 	for (b = 0; b < NDADDR; b++) {
   1068 		inc = markblk(fn, di, di->di_db[b], o);
   1069 		if (inc == 0)
   1070 			break;
   1071 		o += inc;
   1072 	}
   1073 	/* ...and the indirect blocks. */
   1074 	if (inc) {
   1075 		for (b = 0; b < NIADDR; b++) {
   1076 			inc = markiblk(fn, di, di->di_ib[b], o, b);
   1077 			if (inc == 0)
   1078 				return;
   1079 			o += inc;
   1080 		}
   1081 	}
   1082 }
   1083 
   1084 static void
   1085 dblk_callback(struct ufs1_dinode * di, unsigned int inum, void *arg)
   1086 {
   1087 	mark_callback_t fn;
   1088 	fn = (mark_callback_t) arg;
   1089 	switch (di->di_mode & IFMT) {
   1090 	case IFLNK:
   1091 		if (di->di_size > newsb->fs_maxsymlinklen) {
   1092 	case IFDIR:
   1093 	case IFREG:
   1094 			map_inode_data_blocks(di, fn);
   1095 		}
   1096 		break;
   1097 	}
   1098 }
   1099 /*
   1100  * Make a callback call, a la map_inode_data_blocks, for all data
   1101  *  blocks in the entire fs.  This is used only once, in
   1102  *  update_for_data_move, but it's out at top level because the complex
   1103  *  downward-funarg nesting that would otherwise result seems to give
   1104  *  gcc gastric distress.
   1105  */
   1106 static void
   1107 map_data_blocks(mark_callback_t fn, int ncg)
   1108 {
   1109 	map_inodes(&dblk_callback, ncg, (void *) fn);
   1110 }
   1111 /*
   1112  * Initialize the blkmove array.
   1113  */
   1114 static void
   1115 blkmove_init(void)
   1116 {
   1117 	int i;
   1118 
   1119 	blkmove = alloconce(oldsb->fs_size * sizeof(*blkmove), "blkmove");
   1120 	for (i = 0; i < oldsb->fs_size; i++)
   1121 		blkmove[i] = i;
   1122 }
   1123 /*
   1124  * Load the inodes off disk.  Allocates the structures and initializes
   1125  *  them - the inodes from disk, the flags to zero.
   1126  */
   1127 static void
   1128 loadinodes(void)
   1129 {
   1130 	int cg;
   1131 	struct ufs1_dinode *iptr;
   1132 
   1133 	inodes = alloconce(oldsb->fs_ncg * oldsb->fs_ipg * sizeof(struct ufs1_dinode), "inodes");
   1134 	iflags = alloconce(oldsb->fs_ncg * oldsb->fs_ipg, "inode flags");
   1135 	bzero(iflags, oldsb->fs_ncg * oldsb->fs_ipg);
   1136 	iptr = inodes;
   1137 	for (cg = 0; cg < oldsb->fs_ncg; cg++) {
   1138 		readat(fsbtodb(oldsb, cgimin(oldsb, cg)), iptr,
   1139 		    oldsb->fs_ipg * sizeof(struct ufs1_dinode));
   1140 		iptr += oldsb->fs_ipg;
   1141 	}
   1142 }
   1143 /*
   1144  * Report a filesystem-too-full problem.
   1145  */
   1146 static void
   1147 toofull(void)
   1148 {
   1149 	printf("Sorry, would run out of data blocks\n");
   1150 	exit(1);
   1151 }
   1152 /*
   1153  * Record a desire to move "n" frags from "from" to "to".
   1154  */
   1155 static void
   1156 mark_move(unsigned int from, unsigned int to, unsigned int n)
   1157 {
   1158 	for (; n > 0; n--)
   1159 		blkmove[from++] = to++;
   1160 }
   1161 /* Helper function - evict n frags, starting with start (cg-relative).
   1162  * The free bitmap is scanned, unallocated frags are ignored, and
   1163  * each block of consecutive allocated frags is moved as a unit.
   1164  */
   1165 static void
   1166 fragmove(struct cg * cg, int base, unsigned int start, unsigned int n)
   1167 {
   1168 	int i;
   1169 	int run;
   1170 	run = 0;
   1171 	for (i = 0; i <= n; i++) {
   1172 		if ((i < n) && bit_is_clr(cg_blksfree(cg, 0), start + i)) {
   1173 			run++;
   1174 		} else {
   1175 			if (run > 0) {
   1176 				int off;
   1177 				off = find_freespace(run);
   1178 				if (off < 0)
   1179 					toofull();
   1180 				mark_move(base + start + i - run, off, run);
   1181 				set_bits(cg_blksfree(cg, 0), start + i - run,
   1182 				    run);
   1183 				clr_bits(cg_blksfree(cgs[dtog(oldsb, off)], 0),
   1184 				    dtogd(oldsb, off), run);
   1185 			}
   1186 			run = 0;
   1187 		}
   1188 	}
   1189 }
   1190 /*
   1191  * Evict all data blocks from the given cg, starting at minfrag (based
   1192  *  at the beginning of the cg), for length nfrag.  The eviction is
   1193  *  assumed to be entirely data-area; this should not be called with a
   1194  *  range overlapping the metadata structures in the cg.  It also
   1195  *  assumes minfrag points into the given cg; it will misbehave if this
   1196  *  is not true.
   1197  *
   1198  * See the comment header on find_freespace() for one possible bug
   1199  *  lurking here.
   1200  */
   1201 static void
   1202 evict_data(struct cg * cg, unsigned int minfrag, unsigned int nfrag)
   1203 {
   1204 	int base;		/* base of cg (in frags from beginning of fs) */
   1205 
   1206 
   1207 	base = cgbase(oldsb, cg->cg_cgx);
   1208 	/* Does the boundary fall in the middle of a block?  To avoid breaking
   1209 	 * between frags allocated as consecutive, we always evict the whole
   1210 	 * block in this case, though one could argue we should check to see
   1211 	 * if the frag before or after the break is unallocated. */
   1212 	if (minfrag % oldsb->fs_frag) {
   1213 		int n;
   1214 		n = minfrag % oldsb->fs_frag;
   1215 		minfrag -= n;
   1216 		nfrag += n;
   1217 	}
   1218 	/* Do whole blocks.  If a block is wholly free, skip it; if wholly
   1219 	 * allocated, move it in toto.  If neither, call fragmove() to move
   1220 	 * the frags to new locations. */
   1221 	while (nfrag >= oldsb->fs_frag) {
   1222 		if (!blk_is_set(cg_blksfree(cg, 0), minfrag, oldsb->fs_frag)) {
   1223 			if (blk_is_clr(cg_blksfree(cg, 0), minfrag,
   1224 				oldsb->fs_frag)) {
   1225 				int off;
   1226 				off = find_freeblock();
   1227 				if (off < 0)
   1228 					toofull();
   1229 				mark_move(base + minfrag, off, oldsb->fs_frag);
   1230 				set_bits(cg_blksfree(cg, 0), minfrag,
   1231 				    oldsb->fs_frag);
   1232 				clr_bits(cg_blksfree(cgs[dtog(oldsb, off)], 0),
   1233 				    dtogd(oldsb, off), oldsb->fs_frag);
   1234 			} else {
   1235 				fragmove(cg, base, minfrag, oldsb->fs_frag);
   1236 			}
   1237 		}
   1238 		minfrag += oldsb->fs_frag;
   1239 		nfrag -= oldsb->fs_frag;
   1240 	}
   1241 	/* Clean up any sub-block amount left over. */
   1242 	if (nfrag) {
   1243 		fragmove(cg, base, minfrag, nfrag);
   1244 	}
   1245 }
   1246 /*
   1247  * Move all data blocks according to blkmove.  We have to be careful,
   1248  *  because we may be updating indirect blocks that will themselves be
   1249  *  getting moved, or inode int32_t arrays that point to indirect
   1250  *  blocks that will be moved.  We call this before
   1251  *  update_for_data_move, and update_for_data_move does inodes first,
   1252  *  then indirect blocks in preorder, so as to make sure that the
   1253  *  filesystem is self-consistent at all points, for better crash
   1254  *  tolerance.  (We can get away with this only because all the writes
   1255  *  done by perform_data_move() are writing into space that's not used
   1256  *  by the old filesystem.)  If we crash, some things may point to the
   1257  *  old data and some to the new, but both copies are the same.  The
   1258  *  only wrong things should be csum info and free bitmaps, which fsck
   1259  *  is entirely capable of cleaning up.
   1260  *
   1261  * Since blkmove_init() initializes all blocks to move to their current
   1262  *  locations, we can have two blocks marked as wanting to move to the
   1263  *  same location, but only two and only when one of them is the one
   1264  *  that was already there.  So if blkmove[i]==i, we ignore that entry
   1265  *  entirely - for unallocated blocks, we don't want it (and may be
   1266  *  putting something else there), and for allocated blocks, we don't
   1267  *  want to copy it anywhere.
   1268  */
   1269 static void
   1270 perform_data_move(void)
   1271 {
   1272 	int i;
   1273 	int run;
   1274 	int maxrun;
   1275 	char buf[65536];
   1276 
   1277 	maxrun = sizeof(buf) / newsb->fs_fsize;
   1278 	run = 0;
   1279 	for (i = 0; i < oldsb->fs_size; i++) {
   1280 		if ((blkmove[i] == i) ||
   1281 		    (run >= maxrun) ||
   1282 		    ((run > 0) &&
   1283 			(blkmove[i] != blkmove[i - 1] + 1))) {
   1284 			if (run > 0) {
   1285 				readat(fsbtodb(oldsb, i - run), &buf[0],
   1286 				    run << oldsb->fs_fshift);
   1287 				writeat(fsbtodb(oldsb, blkmove[i - run]),
   1288 				    &buf[0], run << oldsb->fs_fshift);
   1289 			}
   1290 			run = 0;
   1291 		}
   1292 		if (blkmove[i] != i)
   1293 			run++;
   1294 	}
   1295 	if (run > 0) {
   1296 		readat(fsbtodb(oldsb, i - run), &buf[0],
   1297 		    run << oldsb->fs_fshift);
   1298 		writeat(fsbtodb(oldsb, blkmove[i - run]), &buf[0],
   1299 		    run << oldsb->fs_fshift);
   1300 	}
   1301 }
   1302 /*
   1303  * This modifies an array of int32_t, according to blkmove.  This is
   1304  *  used to update inode block arrays and indirect blocks to point to
   1305  *  the new locations of data blocks.
   1306  *
   1307  * Return value is the number of int32_ts that needed updating; in
   1308  *  particular, the return value is zero iff nothing was modified.
   1309  */
   1310 static int
   1311 movemap_blocks(int32_t * vec, int n)
   1312 {
   1313 	int rv;
   1314 
   1315 	rv = 0;
   1316 	for (; n > 0; n--, vec++) {
   1317 		if (blkmove[*vec] != *vec) {
   1318 			*vec = blkmove[*vec];
   1319 			rv++;
   1320 		}
   1321 	}
   1322 	return (rv);
   1323 }
   1324 static void
   1325 moveblocks_callback(struct ufs1_dinode * di, unsigned int inum, void *arg)
   1326 {
   1327 	switch (di->di_mode & IFMT) {
   1328 	case IFLNK:
   1329 		if (di->di_size > oldsb->fs_maxsymlinklen) {
   1330 	case IFDIR:
   1331 	case IFREG:
   1332 			/* don't || these two calls; we need their
   1333 			 * side-effects */
   1334 			if (movemap_blocks(&di->di_db[0], NDADDR)) {
   1335 				iflags[inum] |= IF_DIRTY;
   1336 			}
   1337 			if (movemap_blocks(&di->di_ib[0], NIADDR)) {
   1338 				iflags[inum] |= IF_DIRTY;
   1339 			}
   1340 		}
   1341 		break;
   1342 	}
   1343 }
   1344 
   1345 static void
   1346 moveindir_callback(unsigned int off, unsigned int nfrag, unsigned int nbytes, int kind)
   1347 {
   1348 	if (kind == MDB_INDIR_PRE) {
   1349 		int32_t blk[howmany(MAXBSIZE, sizeof(int32_t))];
   1350 		readat(fsbtodb(oldsb, off), &blk[0], oldsb->fs_bsize);
   1351 		if (movemap_blocks(&blk[0], NINDIR(oldsb))) {
   1352 			writeat(fsbtodb(oldsb, off), &blk[0], oldsb->fs_bsize);
   1353 		}
   1354 	}
   1355 }
   1356 /*
   1357  * Update all inode data arrays and indirect blocks to point to the new
   1358  *  locations of data blocks.  See the comment header on
   1359  *  perform_data_move for some ordering considerations.
   1360  */
   1361 static void
   1362 update_for_data_move(void)
   1363 {
   1364 	map_inodes(&moveblocks_callback, oldsb->fs_ncg, NULL);
   1365 	map_data_blocks(&moveindir_callback, oldsb->fs_ncg);
   1366 }
   1367 /*
   1368  * Initialize the inomove array.
   1369  */
   1370 static void
   1371 inomove_init(void)
   1372 {
   1373 	int i;
   1374 
   1375 	inomove = alloconce(oldsb->fs_ipg * oldsb->fs_ncg * sizeof(*inomove),
   1376                             "inomove");
   1377 	for (i = (oldsb->fs_ipg * oldsb->fs_ncg) - 1; i >= 0; i--)
   1378 		inomove[i] = i;
   1379 }
   1380 /*
   1381  * Flush all dirtied inodes to disk.  Scans the inode flags array; for
   1382  *  each dirty inode, it sets the BDIRTY bit on the first inode in the
   1383  *  block containing the dirty inode.  Then it scans by blocks, and for
   1384  *  each marked block, writes it.
   1385  */
   1386 static void
   1387 flush_inodes(void)
   1388 {
   1389 	int i;
   1390 	int ni;
   1391 	int m;
   1392 
   1393 	ni = newsb->fs_ipg * newsb->fs_ncg;
   1394 	m = INOPB(newsb) - 1;
   1395 	for (i = 0; i < ni; i++) {
   1396 		if (iflags[i] & IF_DIRTY) {
   1397 			iflags[i & ~m] |= IF_BDIRTY;
   1398 		}
   1399 	}
   1400 	m++;
   1401 	for (i = 0; i < ni; i += m) {
   1402 		if (iflags[i] & IF_BDIRTY) {
   1403 			writeat(fsbtodb(newsb, ino_to_fsba(newsb, i)),
   1404 			    inodes + i, newsb->fs_bsize);
   1405 		}
   1406 	}
   1407 }
   1408 /*
   1409  * Evict all inodes from the specified cg.  shrink() already checked
   1410  *  that there were enough free inodes, so the no-free-inodes check is
   1411  *  a can't-happen.  If it does trip, the filesystem should be in good
   1412  *  enough shape for fsck to fix; see the comment on perform_data_move
   1413  *  for the considerations in question.
   1414  */
   1415 static void
   1416 evict_inodes(struct cg * cg)
   1417 {
   1418 	int inum;
   1419 	int i;
   1420 	int fi;
   1421 
   1422 	inum = newsb->fs_ipg * cg->cg_cgx;
   1423 	for (i = 0; i < newsb->fs_ipg; i++, inum++) {
   1424 		if (inodes[inum].di_mode != 0) {
   1425 			fi = find_freeinode();
   1426 			if (fi < 0) {
   1427 				printf("Sorry, inodes evaporated - "
   1428 				    "filesystem probably needs fsck\n");
   1429 				exit(1);
   1430 			}
   1431 			inomove[inum] = fi;
   1432 			clr_bits(cg_inosused(cg, 0), i, 1);
   1433 			set_bits(cg_inosused(cgs[ino_to_cg(newsb, fi)], 0),
   1434 			    fi % newsb->fs_ipg, 1);
   1435 		}
   1436 	}
   1437 }
   1438 /*
   1439  * Move inodes from old locations to new.  Does not actually write
   1440  *  anything to disk; just copies in-core and sets dirty bits.
   1441  *
   1442  * We have to be careful here for reasons similar to those mentioned in
   1443  *  the comment header on perform_data_move, above: for the sake of
   1444  *  crash tolerance, we want to make sure everything is present at both
   1445  *  old and new locations before we update pointers.  So we call this
   1446  *  first, then flush_inodes() to get them out on disk, then update
   1447  *  directories to match.
   1448  */
   1449 static void
   1450 perform_inode_move(void)
   1451 {
   1452 	int i;
   1453 	int ni;
   1454 
   1455 	ni = oldsb->fs_ipg * oldsb->fs_ncg;
   1456 	for (i = 0; i < ni; i++) {
   1457 		if (inomove[i] != i) {
   1458 			inodes[inomove[i]] = inodes[i];
   1459 			iflags[inomove[i]] = iflags[i] | IF_DIRTY;
   1460 		}
   1461 	}
   1462 }
   1463 /*
   1464  * Update the directory contained in the nb bytes at buf, to point to
   1465  *  inodes' new locations.
   1466  */
   1467 static int
   1468 update_dirents(char *buf, int nb)
   1469 {
   1470 	int rv;
   1471 #define d ((struct direct *)buf)
   1472 
   1473 	rv = 0;
   1474 	while (nb > 0) {
   1475 		if (inomove[d->d_ino] != d->d_ino) {
   1476 			rv++;
   1477 			d->d_ino = inomove[d->d_ino];
   1478 		}
   1479 		nb -= d->d_reclen;
   1480 		buf += d->d_reclen;
   1481 	}
   1482 	return (rv);
   1483 #undef d
   1484 }
   1485 /*
   1486  * Callback function for map_inode_data_blocks, for updating a
   1487  *  directory to point to new inode locations.
   1488  */
   1489 static void
   1490 update_dir_data(unsigned int bn, unsigned int size, unsigned int nb, int kind)
   1491 {
   1492 	if (kind == MDB_DATA) {
   1493 		union {
   1494 			struct direct d;
   1495 			char ch[MAXBSIZE];
   1496 		}     buf;
   1497 		readat(fsbtodb(oldsb, bn), &buf, size << oldsb->fs_fshift);
   1498 		if (update_dirents((char *) &buf, nb)) {
   1499 			writeat(fsbtodb(oldsb, bn), &buf,
   1500 			    size << oldsb->fs_fshift);
   1501 		}
   1502 	}
   1503 }
   1504 static void
   1505 dirmove_callback(struct ufs1_dinode * di, unsigned int inum, void *arg)
   1506 {
   1507 	switch (di->di_mode & IFMT) {
   1508 	case IFDIR:
   1509 		map_inode_data_blocks(di, &update_dir_data);
   1510 		break;
   1511 	}
   1512 }
   1513 /*
   1514  * Update directory entries to point to new inode locations.
   1515  */
   1516 static void
   1517 update_for_inode_move(void)
   1518 {
   1519 	map_inodes(&dirmove_callback, newsb->fs_ncg, NULL);
   1520 }
   1521 /*
   1522  * Shrink the filesystem.
   1523  */
   1524 static void
   1525 shrink(void)
   1526 {
   1527 	int i;
   1528 
   1529 	/* Load the inodes off disk - we'll need 'em. */
   1530 	loadinodes();
   1531 	/* Update the timestamp. */
   1532 	newsb->fs_time = timestamp();
   1533 	/* Update the size figures. */
   1534 	newsb->fs_size = dbtofsb(newsb, newsize);
   1535 	newsb->fs_old_ncyl = (newsb->fs_size * NSPF(newsb)) / newsb->fs_old_spc;
   1536 	newsb->fs_ncg = howmany(newsb->fs_old_ncyl, newsb->fs_old_cpg);
   1537 	/* Does the (new) last cg end before the end of its inode area?  See
   1538 	 * the similar code in grow() for more on this. */
   1539 	if (cgdmin(newsb, newsb->fs_ncg - 1) > newsb->fs_size) {
   1540 		newsb->fs_ncg--;
   1541 		newsb->fs_old_ncyl = newsb->fs_ncg * newsb->fs_old_cpg;
   1542 		newsb->fs_size = (newsb->fs_old_ncyl * newsb->fs_old_spc) / NSPF(newsb);
   1543 		printf("Warning: last cylinder group is too small;\n");
   1544 		printf("    dropping it.  New size = %lu.\n",
   1545 		    (unsigned long int) fsbtodb(newsb, newsb->fs_size));
   1546 	}
   1547 	/* Let's make sure we're not being shrunk into oblivion. */
   1548 	if (newsb->fs_ncg < 1) {
   1549 		printf("Size too small - filesystem would have no cylinders\n");
   1550 		exit(1);
   1551 	}
   1552 	/* Initialize for block motion. */
   1553 	blkmove_init();
   1554 	/* Update csum size, then fix up for the new size */
   1555 	newsb->fs_cssize = fragroundup(newsb,
   1556 	    newsb->fs_ncg * sizeof(struct csum));
   1557 	csum_fixup();
   1558 	/* Evict data from any cgs being wholly eliminated */
   1559 	for (i = newsb->fs_ncg; i < oldsb->fs_ncg; i++) {
   1560 		int base;
   1561 		int dlow;
   1562 		int dhigh;
   1563 		int dmax;
   1564 		base = cgbase(oldsb, i);
   1565 		dlow = cgsblock(oldsb, i) - base;
   1566 		dhigh = cgdmin(oldsb, i) - base;
   1567 		dmax = oldsb->fs_size - base;
   1568 		if (dmax > cgs[i]->cg_ndblk)
   1569 			dmax = cgs[i]->cg_ndblk;
   1570 		evict_data(cgs[i], 0, dlow);
   1571 		evict_data(cgs[i], dhigh, dmax - dhigh);
   1572 		newsb->fs_cstotal.cs_ndir -= cgs[i]->cg_cs.cs_ndir;
   1573 		newsb->fs_cstotal.cs_nifree -= cgs[i]->cg_cs.cs_nifree;
   1574 		newsb->fs_cstotal.cs_nffree -= cgs[i]->cg_cs.cs_nffree;
   1575 		newsb->fs_cstotal.cs_nbfree -= cgs[i]->cg_cs.cs_nbfree;
   1576 	}
   1577 	/* Update the new last cg. */
   1578 	cgs[newsb->fs_ncg - 1]->cg_ndblk = newsb->fs_size -
   1579 	    ((newsb->fs_ncg - 1) * newsb->fs_fpg);
   1580 	/* Is the new last cg partial?  If so, evict any data from the part
   1581 	 * being shrunken away. */
   1582 	if (newsb->fs_size % newsb->fs_fpg) {
   1583 		struct cg *cg;
   1584 		int oldcgsize;
   1585 		int newcgsize;
   1586 		cg = cgs[newsb->fs_ncg - 1];
   1587 		newcgsize = newsb->fs_size % newsb->fs_fpg;
   1588 		oldcgsize = oldsb->fs_size - ((newsb->fs_ncg - 1) & oldsb->fs_fpg);
   1589 		if (oldcgsize > oldsb->fs_fpg)
   1590 			oldcgsize = oldsb->fs_fpg;
   1591 		evict_data(cg, newcgsize, oldcgsize - newcgsize);
   1592 		clr_bits(cg_blksfree(cg, 0), newcgsize, oldcgsize - newcgsize);
   1593 	}
   1594 	/* Find out whether we would run out of inodes.  (Note we haven't
   1595 	 * actually done anything to the filesystem yet; all those evict_data
   1596 	 * calls just update blkmove.) */
   1597 	{
   1598 		int slop;
   1599 		slop = 0;
   1600 		for (i = 0; i < newsb->fs_ncg; i++)
   1601 			slop += cgs[i]->cg_cs.cs_nifree;
   1602 		for (; i < oldsb->fs_ncg; i++)
   1603 			slop -= oldsb->fs_ipg - cgs[i]->cg_cs.cs_nifree;
   1604 		if (slop < 0) {
   1605 			printf("Sorry, would run out of inodes\n");
   1606 			exit(1);
   1607 		}
   1608 	}
   1609 	/* Copy data, then update pointers to data.  See the comment header on
   1610 	 * perform_data_move for ordering considerations. */
   1611 	perform_data_move();
   1612 	update_for_data_move();
   1613 	/* Now do inodes.  Initialize, evict, move, update - see the comment
   1614 	 * header on perform_inode_move. */
   1615 	inomove_init();
   1616 	for (i = newsb->fs_ncg; i < oldsb->fs_ncg; i++)
   1617 		evict_inodes(cgs[i]);
   1618 	perform_inode_move();
   1619 	flush_inodes();
   1620 	update_for_inode_move();
   1621 	/* Recompute all the bitmaps; most of them probably need it anyway,
   1622 	 * the rest are just paranoia and not wanting to have to bother
   1623 	 * keeping track of exactly which ones require it. */
   1624 	for (i = 0; i < newsb->fs_ncg; i++)
   1625 		cgflags[i] |= CGF_DIRTY | CGF_BLKMAPS | CGF_INOMAPS;
   1626 	/* Update the cg_old_ncyl value for the last cylinder.  The condition is
   1627 	 * commented out because fsck whines if not - see the similar
   1628 	 * condition in grow() for more. */
   1629 	/* XXX fix once fsck is fixed */
   1630 	/* if (newsb->fs_old_ncyl % newsb->fs_old_cpg) XXX */
   1631 /*XXXJTK*/
   1632 	cgs[newsb->fs_ncg - 1]->cg_old_ncyl =
   1633 	    newsb->fs_old_ncyl % newsb->fs_old_cpg;
   1634 	/* Make fs_dsize match the new reality. */
   1635 	recompute_fs_dsize();
   1636 }
   1637 /*
   1638  * Recompute the block totals, block cluster summaries, and rotational
   1639  *  position summaries, for a given cg (specified by number), based on
   1640  *  its free-frag bitmap (cg_blksfree()[]).
   1641  */
   1642 static void
   1643 rescan_blkmaps(int cgn)
   1644 {
   1645 	struct cg *cg;
   1646 	int f;
   1647 	int b;
   1648 	int blkfree;
   1649 	int blkrun;
   1650 	int fragrun;
   1651 	int fwb;
   1652 
   1653 	cg = cgs[cgn];
   1654 	/* Subtract off the current totals from the sb's summary info */
   1655 	newsb->fs_cstotal.cs_nffree -= cg->cg_cs.cs_nffree;
   1656 	newsb->fs_cstotal.cs_nbfree -= cg->cg_cs.cs_nbfree;
   1657 	/* Clear counters and bitmaps. */
   1658 	cg->cg_cs.cs_nffree = 0;
   1659 	cg->cg_cs.cs_nbfree = 0;
   1660 	bzero(&cg->cg_frsum[0], MAXFRAG * sizeof(cg->cg_frsum[0]));
   1661 	bzero(&cg_blktot(cg, 0)[0],
   1662 	    newsb->fs_old_cpg * sizeof(cg_blktot(cg, 0)[0]));
   1663 	bzero(&cg_blks(newsb, cg, 0, 0)[0],
   1664 	    newsb->fs_old_cpg * newsb->fs_old_nrpos *
   1665 	    sizeof(cg_blks(newsb, cg, 0, 0)[0]));
   1666 	if (newsb->fs_contigsumsize > 0) {
   1667 		cg->cg_nclusterblks = cg->cg_ndblk / newsb->fs_frag;
   1668 		bzero(&cg_clustersum(cg, 0)[1],
   1669 		    newsb->fs_contigsumsize *
   1670 		    sizeof(cg_clustersum(cg, 0)[1]));
   1671 		bzero(&cg_clustersfree(cg, 0)[0],
   1672 		    howmany((newsb->fs_old_cpg * newsb->fs_old_spc) / NSPB(newsb),
   1673 			NBBY));
   1674 	}
   1675 	/* Scan the free-frag bitmap.  Runs of free frags are kept track of
   1676 	 * with fragrun, and recorded into cg_frsum[] and cg_cs.cs_nffree; on
   1677 	 * each block boundary, entire free blocks are recorded as well. */
   1678 	blkfree = 1;
   1679 	blkrun = 0;
   1680 	fragrun = 0;
   1681 	f = 0;
   1682 	b = 0;
   1683 	fwb = 0;
   1684 	while (f < cg->cg_ndblk) {
   1685 		if (bit_is_set(cg_blksfree(cg, 0), f)) {
   1686 			fragrun++;
   1687 		} else {
   1688 			blkfree = 0;
   1689 			if (fragrun > 0) {
   1690 				cg->cg_frsum[fragrun]++;
   1691 				cg->cg_cs.cs_nffree += fragrun;
   1692 			}
   1693 			fragrun = 0;
   1694 		}
   1695 		f++;
   1696 		fwb++;
   1697 		if (fwb >= newsb->fs_frag) {
   1698 			if (blkfree) {
   1699 				cg->cg_cs.cs_nbfree++;
   1700 				if (newsb->fs_contigsumsize > 0)
   1701 					set_bits(cg_clustersfree(cg, 0), b, 1);
   1702 				cg_blktot(cg, 0)[cbtocylno(newsb, f - newsb->fs_frag)]++;
   1703 				cg_blks(newsb, cg,
   1704 				    cbtocylno(newsb, f - newsb->fs_frag),
   1705 				    0)[cbtorpos(newsb, f - newsb->fs_frag)]++;
   1706 				blkrun++;
   1707 			} else {
   1708 				if (fragrun > 0) {
   1709 					cg->cg_frsum[fragrun]++;
   1710 					cg->cg_cs.cs_nffree += fragrun;
   1711 				}
   1712 				if (newsb->fs_contigsumsize > 0) {
   1713 					if (blkrun > 0) {
   1714 						cg_clustersum(cg, 0)[(blkrun > newsb->fs_contigsumsize) ? newsb->fs_contigsumsize : blkrun]++;
   1715 					}
   1716 				}
   1717 				blkrun = 0;
   1718 			}
   1719 			fwb = 0;
   1720 			b++;
   1721 			blkfree = 1;
   1722 			fragrun = 0;
   1723 		}
   1724 	}
   1725 	if (fragrun > 0) {
   1726 		cg->cg_frsum[fragrun]++;
   1727 		cg->cg_cs.cs_nffree += fragrun;
   1728 	}
   1729 	if ((blkrun > 0) && (newsb->fs_contigsumsize > 0)) {
   1730 		cg_clustersum(cg, 0)[(blkrun > newsb->fs_contigsumsize) ?
   1731 		    newsb->fs_contigsumsize : blkrun]++;
   1732 	}
   1733 	/*
   1734          * Put the updated summary info back into csums, and add it
   1735          * back into the sb's summary info.  Then mark the cg dirty.
   1736          */
   1737 	csums[cgn] = cg->cg_cs;
   1738 	newsb->fs_cstotal.cs_nffree += cg->cg_cs.cs_nffree;
   1739 	newsb->fs_cstotal.cs_nbfree += cg->cg_cs.cs_nbfree;
   1740 	cgflags[cgn] |= CGF_DIRTY;
   1741 }
   1742 /*
   1743  * Recompute the cg_inosused()[] bitmap, and the cs_nifree and cs_ndir
   1744  *  values, for a cg, based on the in-core inodes for that cg.
   1745  */
   1746 static void
   1747 rescan_inomaps(int cgn)
   1748 {
   1749 	struct cg *cg;
   1750 	int inum;
   1751 	int iwc;
   1752 
   1753 	cg = cgs[cgn];
   1754 	newsb->fs_cstotal.cs_ndir -= cg->cg_cs.cs_ndir;
   1755 	newsb->fs_cstotal.cs_nifree -= cg->cg_cs.cs_nifree;
   1756 	cg->cg_cs.cs_ndir = 0;
   1757 	cg->cg_cs.cs_nifree = 0;
   1758 	bzero(&cg_inosused(cg, 0)[0], howmany(newsb->fs_ipg, NBBY));
   1759 	inum = cgn * newsb->fs_ipg;
   1760 	if (cgn == 0) {
   1761 		set_bits(cg_inosused(cg, 0), 0, 2);
   1762 		iwc = 2;
   1763 		inum += 2;
   1764 	} else {
   1765 		iwc = 0;
   1766 	}
   1767 	for (; iwc < newsb->fs_ipg; iwc++, inum++) {
   1768 		switch (inodes[inum].di_mode & IFMT) {
   1769 		case 0:
   1770 			cg->cg_cs.cs_nifree++;
   1771 			break;
   1772 		case IFDIR:
   1773 			cg->cg_cs.cs_ndir++;
   1774 			/* fall through */
   1775 		default:
   1776 			set_bits(cg_inosused(cg, 0), iwc, 1);
   1777 			break;
   1778 		}
   1779 	}
   1780 	csums[cgn] = cg->cg_cs;
   1781 	newsb->fs_cstotal.cs_ndir += cg->cg_cs.cs_ndir;
   1782 	newsb->fs_cstotal.cs_nifree += cg->cg_cs.cs_nifree;
   1783 	cgflags[cgn] |= CGF_DIRTY;
   1784 }
   1785 /*
   1786  * Flush cgs to disk, recomputing anything they're marked as needing.
   1787  */
   1788 static void
   1789 flush_cgs(void)
   1790 {
   1791 	int i;
   1792 
   1793 	for (i = 0; i < newsb->fs_ncg; i++) {
   1794 		if (cgflags[i] & CGF_BLKMAPS) {
   1795 			rescan_blkmaps(i);
   1796 		}
   1797 		if (cgflags[i] & CGF_INOMAPS) {
   1798 			rescan_inomaps(i);
   1799 		}
   1800 		if (cgflags[i] & CGF_DIRTY) {
   1801 			cgs[i]->cg_rotor = 0;
   1802 			cgs[i]->cg_frotor = 0;
   1803 			cgs[i]->cg_irotor = 0;
   1804 			writeat(fsbtodb(newsb, cgtod(newsb, i)), cgs[i],
   1805 			    cgblksz);
   1806 		}
   1807 	}
   1808 	writeat(fsbtodb(newsb, newsb->fs_csaddr), csums, newsb->fs_cssize);
   1809 }
   1810 /*
   1811  * Write the superblock, both to the main superblock and to each cg's
   1812  *  alternative superblock.
   1813  */
   1814 static void
   1815 write_sbs(void)
   1816 {
   1817 	int i;
   1818 
   1819 	writeat(where /  DEV_BSIZE, newsb, SBLOCKSIZE);
   1820 	for (i = 0; i < newsb->fs_ncg; i++) {
   1821 		writeat(fsbtodb(newsb, cgsblock(newsb, i)), newsb, SBLOCKSIZE);
   1822 	}
   1823 }
   1824 /*
   1825  * main().
   1826  */
   1827 int main(int, char **);
   1828 int
   1829 main(int ac, char **av)
   1830 {
   1831 	size_t i;
   1832 	if (ac != 3) {
   1833 		fprintf(stderr, "usage: %s filesystem new-size\n",
   1834 		    getprogname());
   1835 		exit(1);
   1836 	}
   1837 	fd = open(av[1], O_RDWR, 0);
   1838 	if (fd < 0)
   1839 		err(1, "Cannot open `%s'", av[1]);
   1840 	checksmallio();
   1841 	newsize = atoi(av[2]);
   1842 	oldsb = (struct fs *) & sbbuf;
   1843 	newsb = (struct fs *) (SBLOCKSIZE + (char *) &sbbuf);
   1844 	for (where = search[i = 0]; search[i] != -1; where = search[++i]) {
   1845 		readat(where / DEV_BSIZE, oldsb, SBLOCKSIZE);
   1846 		if (oldsb->fs_magic == FS_UFS1_MAGIC)
   1847 			break;
   1848 		if (where == SBLOCK_UFS2)
   1849 			continue;
   1850 		if (oldsb->fs_old_flags & FS_FLAGS_UPDATED)
   1851 			err(1, "Cannot resize ffsv2 format suberblock!");
   1852 	}
   1853 	if (where == (off_t)-1)
   1854 		errx(1, "Bad magic number");
   1855 	oldsb->fs_qbmask = ~(int64_t) oldsb->fs_bmask;
   1856 	oldsb->fs_qfmask = ~(int64_t) oldsb->fs_fmask;
   1857 	if (oldsb->fs_ipg % INOPB(oldsb)) {
   1858 		printf("ipg[%d] %% INOPB[%d] != 0\n", (int) oldsb->fs_ipg,
   1859 		    (int) INOPB(oldsb));
   1860 		exit(1);
   1861 	}
   1862 	/* The superblock is bigger than struct fs (there are trailing tables,
   1863 	 * of non-fixed size); make sure we copy the whole thing.  SBLOCKSIZE may
   1864 	 * be an over-estimate, but we do this just once, so being generous is
   1865 	 * cheap. */
   1866 	bcopy(oldsb, newsb, SBLOCKSIZE);
   1867 	loadcgs();
   1868 	if (newsize > fsbtodb(oldsb, oldsb->fs_size)) {
   1869 		grow();
   1870 	} else if (newsize < fsbtodb(oldsb, oldsb->fs_size)) {
   1871 		shrink();
   1872 	}
   1873 	flush_cgs();
   1874 	write_sbs();
   1875 	exit(0);
   1876 }
   1877