Home | History | Annotate | Line # | Download | only in resize_ffs
resize_ffs.c revision 1.40
      1  1.40  riastrad /*	$NetBSD: resize_ffs.c,v 1.40 2015/03/28 17:25:33 riastradh Exp $	*/
      2   1.1       jtk /* From sources sent on February 17, 2003 */
      3   1.1       jtk /*-
      4   1.1       jtk  * As its sole author, I explicitly place this code in the public
      5   1.1       jtk  *  domain.  Anyone may use it for any purpose (though I would
      6   1.1       jtk  *  appreciate credit where it is due).
      7   1.1       jtk  *
      8   1.1       jtk  *					der Mouse
      9   1.1       jtk  *
     10   1.1       jtk  *			       mouse (at) rodents.montreal.qc.ca
     11   1.1       jtk  *		     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B
     12   1.1       jtk  */
     13   1.1       jtk /*
     14   1.3       wiz  * resize_ffs:
     15   1.1       jtk  *
     16  1.24       wiz  * Resize a file system.  Is capable of both growing and shrinking.
     17   1.1       jtk  *
     18  1.24       wiz  * Usage: resize_ffs [-s newsize] [-y] file_system
     19   1.1       jtk  *
     20  1.15       riz  * Example: resize_ffs -s 29574 /dev/rsd1e
     21   1.1       jtk  *
     22   1.1       jtk  * newsize is in DEV_BSIZE units (ie, disk sectors, usually 512 bytes
     23   1.1       jtk  *  each).
     24   1.1       jtk  *
     25   1.1       jtk  * Note: this currently requires gcc to build, since it is written
     26   1.1       jtk  *  depending on gcc-specific features, notably nested function
     27   1.1       jtk  *  definitions (which in at least a few cases depend on the lexical
     28   1.1       jtk  *  scoping gcc provides, so they can't be trivially moved outside).
     29   1.1       jtk  *
     30   1.5      salo  * Many thanks go to John Kohl <jtk (at) NetBSD.org> for finding bugs: the
     31   1.1       jtk  *  one responsible for the "realloccgblk: can't find blk in cyl"
     32   1.1       jtk  *  problem and a more minor one which left fs_dsize wrong when
     33   1.1       jtk  *  shrinking.  (These actually indicate bugs in fsck too - it should
     34   1.1       jtk  *  have caught and fixed them.)
     35   1.1       jtk  *
     36   1.1       jtk  */
     37   1.1       jtk 
     38  1.11     perry #include <sys/cdefs.h>
     39  1.40  riastrad __RCSID("$NetBSD: resize_ffs.c,v 1.40 2015/03/28 17:25:33 riastradh Exp $");
     40  1.29  dholland 
     41  1.13      haad #include <sys/disk.h>
     42  1.13      haad #include <sys/disklabel.h>
     43  1.13      haad #include <sys/dkio.h>
     44  1.13      haad #include <sys/ioctl.h>
     45   1.1       jtk #include <sys/stat.h>
     46   1.1       jtk #include <sys/mman.h>
     47   1.1       jtk #include <sys/param.h>		/* MAXFRAG */
     48   1.1       jtk #include <ufs/ffs/fs.h>
     49  1.25       riz #include <ufs/ffs/ffs_extern.h>
     50   1.1       jtk #include <ufs/ufs/dir.h>
     51   1.1       jtk #include <ufs/ufs/dinode.h>
     52   1.1       jtk #include <ufs/ufs/ufs_bswap.h>	/* ufs_rw32 */
     53   1.1       jtk 
     54  1.15       riz #include <err.h>
     55  1.15       riz #include <errno.h>
     56  1.15       riz #include <fcntl.h>
     57  1.15       riz #include <stdio.h>
     58  1.15       riz #include <stdlib.h>
     59  1.15       riz #include <strings.h>
     60  1.15       riz #include <unistd.h>
     61  1.15       riz 
     62  1.24       wiz /* new size of file system, in sectors */
     63  1.30  dholland static int64_t newsize;
     64   1.1       jtk 
     65  1.23       riz /* fd open onto disk device or file */
     66   1.1       jtk static int fd;
     67   1.1       jtk 
     68   1.1       jtk /* must we break up big I/O operations - see checksmallio() */
     69   1.1       jtk static int smallio;
     70   1.1       jtk 
     71   1.1       jtk /* size of a cg, in bytes, rounded up to a frag boundary */
     72   1.1       jtk static int cgblksz;
     73   1.1       jtk 
     74   1.4  christos /* possible superblock localtions */
     75   1.4  christos static int search[] = SBLOCKSEARCH;
     76   1.4  christos /* location of the superblock */
     77   1.4  christos static off_t where;
     78   1.4  christos 
     79   1.1       jtk /* Superblocks. */
     80   1.1       jtk static struct fs *oldsb;	/* before we started */
     81   1.1       jtk static struct fs *newsb;	/* copy to work with */
     82   1.1       jtk /* Buffer to hold the above.  Make sure it's aligned correctly. */
     83  1.15       riz static char sbbuf[2 * SBLOCKSIZE]
     84  1.15       riz 	__attribute__((__aligned__(__alignof__(struct fs))));
     85   1.1       jtk 
     86  1.25       riz union dinode {
     87  1.25       riz 	struct ufs1_dinode dp1;
     88  1.25       riz 	struct ufs2_dinode dp2;
     89  1.25       riz };
     90  1.25       riz #define DIP(dp, field)							      \
     91  1.25       riz 	((is_ufs2) ?							      \
     92  1.25       riz 	    (dp)->dp2.field : (dp)->dp1.field)
     93  1.25       riz 
     94  1.25       riz #define DIP_ASSIGN(dp, field, value)					      \
     95  1.25       riz 	do {								      \
     96  1.25       riz 		if (is_ufs2)						      \
     97  1.25       riz 			(dp)->dp2.field = (value);			      \
     98  1.25       riz 		else							      \
     99  1.25       riz 			(dp)->dp1.field = (value);			      \
    100  1.25       riz 	} while (0)
    101  1.25       riz 
    102   1.1       jtk /* a cg's worth of brand new squeaky-clean inodes */
    103   1.4  christos static struct ufs1_dinode *zinodes;
    104   1.1       jtk 
    105   1.1       jtk /* pointers to the in-core cgs, read off disk and possibly modified */
    106   1.1       jtk static struct cg **cgs;
    107   1.1       jtk 
    108   1.1       jtk /* pointer to csum array - the stuff pointed to on-disk by fs_csaddr */
    109   1.1       jtk static struct csum *csums;
    110   1.1       jtk 
    111   1.1       jtk /* per-cg flags, indexed by cg number */
    112   1.1       jtk static unsigned char *cgflags;
    113   1.1       jtk #define CGF_DIRTY   0x01	/* needs to be written to disk */
    114   1.1       jtk #define CGF_BLKMAPS 0x02	/* block bitmaps need rebuilding */
    115   1.1       jtk #define CGF_INOMAPS 0x04	/* inode bitmaps need rebuilding */
    116   1.1       jtk 
    117   1.1       jtk /* when shrinking, these two arrays record how we want blocks to move.	 */
    118   1.1       jtk /*  if blkmove[i] is j, the frag that started out as frag #i should end	 */
    119   1.1       jtk /*  up as frag #j.  inomove[i]=j means, similarly, that the inode that	 */
    120   1.1       jtk /*  started out as inode i should end up as inode j.			 */
    121   1.1       jtk static unsigned int *blkmove;
    122   1.1       jtk static unsigned int *inomove;
    123   1.1       jtk 
    124   1.1       jtk /* in-core copies of all inodes in the fs, indexed by inumber */
    125  1.25       riz union dinode *inodes;
    126  1.25       riz 
    127  1.25       riz void *ibuf;	/* ptr to fs block-sized buffer for reading/writing inodes */
    128  1.25       riz 
    129  1.25       riz /* byteswapped inodes */
    130  1.25       riz union dinode *sinodes;
    131   1.1       jtk 
    132   1.1       jtk /* per-inode flags, indexed by inumber */
    133   1.1       jtk static unsigned char *iflags;
    134   1.1       jtk #define IF_DIRTY  0x01		/* needs to be written to disk */
    135   1.1       jtk #define IF_BDIRTY 0x02		/* like DIRTY, but is set on first inode in a
    136   1.1       jtk 				 * block of inodes, and applies to the whole
    137   1.1       jtk 				 * block. */
    138   1.1       jtk 
    139  1.15       riz /* resize_ffs works directly on dinodes, adapt blksize() */
    140  1.30  dholland #define dblksize(fs, dip, lbn, filesize) \
    141  1.37  dholland 	(((lbn) >= UFS_NDADDR || (uint64_t)(filesize) >= ffs_lblktosize(fs, (lbn) + 1)) \
    142  1.25       riz 	    ? (fs)->fs_bsize						       \
    143  1.37  dholland 	    : (ffs_fragroundup(fs, ffs_blkoff(fs, (filesize)))))
    144   1.4  christos 
    145   1.4  christos 
    146   1.4  christos /*
    147  1.25       riz  * Number of disk sectors per block/fragment
    148  1.28  dholland  */
    149  1.36  dholland #define NSPB(fs)	(FFS_FSBTODB((fs),1) << (fs)->fs_fragshift)
    150  1.36  dholland #define NSPF(fs)	(FFS_FSBTODB((fs),1))
    151   1.4  christos 
    152  1.23       riz /* global flags */
    153  1.23       riz int is_ufs2 = 0;
    154  1.23       riz int needswap = 0;
    155  1.23       riz 
    156  1.13      haad static void usage(void) __dead;
    157  1.13      haad 
    158   1.1       jtk /*
    159   1.1       jtk  * See if we need to break up large I/O operations.  This should never
    160   1.1       jtk  *  be needed, but under at least one <version,platform> combination,
    161   1.1       jtk  *  large enough disk transfers to the raw device hang.  So if we're
    162   1.1       jtk  *  talking to a character special device, play it safe; in this case,
    163   1.1       jtk  *  readat() and writeat() break everything up into pieces no larger
    164   1.1       jtk  *  than 8K, doing multiple syscalls for larger operations.
    165   1.1       jtk  */
    166   1.1       jtk static void
    167   1.1       jtk checksmallio(void)
    168   1.1       jtk {
    169   1.1       jtk 	struct stat stb;
    170   1.1       jtk 
    171   1.1       jtk 	fstat(fd, &stb);
    172   1.1       jtk 	smallio = ((stb.st_mode & S_IFMT) == S_IFCHR);
    173   1.1       jtk }
    174  1.19       riz 
    175  1.19       riz static int
    176  1.19       riz isplainfile(void)
    177  1.19       riz {
    178  1.19       riz 	struct stat stb;
    179  1.19       riz 
    180  1.19       riz 	fstat(fd, &stb);
    181  1.19       riz 	return S_ISREG(stb.st_mode);
    182  1.19       riz }
    183   1.1       jtk /*
    184   1.1       jtk  * Read size bytes starting at blkno into buf.  blkno is in DEV_BSIZE
    185  1.36  dholland  *  units, ie, after FFS_FSBTODB(); size is in bytes.
    186   1.1       jtk  */
    187   1.1       jtk static void
    188   1.1       jtk readat(off_t blkno, void *buf, int size)
    189   1.1       jtk {
    190   1.1       jtk 	/* Seek to the correct place. */
    191   1.4  christos 	if (lseek(fd, blkno * DEV_BSIZE, L_SET) < 0)
    192  1.15       riz 		err(EXIT_FAILURE, "lseek failed");
    193   1.4  christos 
    194   1.1       jtk 	/* See if we have to break up the transfer... */
    195   1.1       jtk 	if (smallio) {
    196   1.1       jtk 		char *bp;	/* pointer into buf */
    197   1.1       jtk 		int left;	/* bytes left to go */
    198   1.1       jtk 		int n;		/* number to do this time around */
    199   1.1       jtk 		int rv;		/* syscall return value */
    200   1.1       jtk 		bp = buf;
    201   1.1       jtk 		left = size;
    202   1.1       jtk 		while (left > 0) {
    203   1.1       jtk 			n = (left > 8192) ? 8192 : left;
    204   1.1       jtk 			rv = read(fd, bp, n);
    205   1.4  christos 			if (rv < 0)
    206  1.13      haad 				err(EXIT_FAILURE, "read failed");
    207   1.4  christos 			if (rv != n)
    208  1.15       riz 				errx(EXIT_FAILURE,
    209  1.15       riz 				    "read: wanted %d, got %d", n, rv);
    210   1.1       jtk 			bp += n;
    211   1.1       jtk 			left -= n;
    212   1.1       jtk 		}
    213   1.1       jtk 	} else {
    214   1.1       jtk 		int rv;
    215   1.1       jtk 		rv = read(fd, buf, size);
    216   1.4  christos 		if (rv < 0)
    217  1.13      haad 			err(EXIT_FAILURE, "read failed");
    218   1.4  christos 		if (rv != size)
    219  1.25       riz 			errx(EXIT_FAILURE, "read: wanted %d, got %d",
    220  1.25       riz 			    size, rv);
    221   1.1       jtk 	}
    222   1.1       jtk }
    223   1.1       jtk /*
    224   1.1       jtk  * Write size bytes from buf starting at blkno.  blkno is in DEV_BSIZE
    225  1.36  dholland  *  units, ie, after FFS_FSBTODB(); size is in bytes.
    226   1.1       jtk  */
    227   1.1       jtk static void
    228   1.1       jtk writeat(off_t blkno, const void *buf, int size)
    229   1.1       jtk {
    230   1.1       jtk 	/* Seek to the correct place. */
    231   1.4  christos 	if (lseek(fd, blkno * DEV_BSIZE, L_SET) < 0)
    232  1.13      haad 		err(EXIT_FAILURE, "lseek failed");
    233   1.1       jtk 	/* See if we have to break up the transfer... */
    234   1.1       jtk 	if (smallio) {
    235   1.1       jtk 		const char *bp;	/* pointer into buf */
    236   1.1       jtk 		int left;	/* bytes left to go */
    237   1.1       jtk 		int n;		/* number to do this time around */
    238   1.1       jtk 		int rv;		/* syscall return value */
    239   1.1       jtk 		bp = buf;
    240   1.1       jtk 		left = size;
    241   1.1       jtk 		while (left > 0) {
    242   1.1       jtk 			n = (left > 8192) ? 8192 : left;
    243   1.1       jtk 			rv = write(fd, bp, n);
    244   1.4  christos 			if (rv < 0)
    245  1.13      haad 				err(EXIT_FAILURE, "write failed");
    246   1.4  christos 			if (rv != n)
    247  1.15       riz 				errx(EXIT_FAILURE,
    248  1.15       riz 				    "write: wanted %d, got %d", n, rv);
    249   1.1       jtk 			bp += n;
    250   1.1       jtk 			left -= n;
    251   1.1       jtk 		}
    252   1.1       jtk 	} else {
    253   1.1       jtk 		int rv;
    254   1.1       jtk 		rv = write(fd, buf, size);
    255   1.4  christos 		if (rv < 0)
    256  1.13      haad 			err(EXIT_FAILURE, "write failed");
    257   1.4  christos 		if (rv != size)
    258  1.15       riz 			errx(EXIT_FAILURE,
    259  1.15       riz 			    "write: wanted %d, got %d", size, rv);
    260   1.1       jtk 	}
    261   1.1       jtk }
    262   1.1       jtk /*
    263   1.1       jtk  * Never-fail versions of malloc() and realloc(), and an allocation
    264   1.1       jtk  *  routine (which also never fails) for allocating memory that will
    265   1.1       jtk  *  never be freed until exit.
    266   1.1       jtk  */
    267   1.1       jtk 
    268   1.1       jtk /*
    269   1.1       jtk  * Never-fail malloc.
    270   1.1       jtk  */
    271   1.1       jtk static void *
    272   1.1       jtk nfmalloc(size_t nb, const char *tag)
    273   1.1       jtk {
    274   1.1       jtk 	void *rv;
    275   1.1       jtk 
    276   1.1       jtk 	rv = malloc(nb);
    277   1.1       jtk 	if (rv)
    278   1.1       jtk 		return (rv);
    279  1.13      haad 	err(EXIT_FAILURE, "Can't allocate %lu bytes for %s",
    280   1.4  christos 	    (unsigned long int) nb, tag);
    281   1.1       jtk }
    282   1.1       jtk /*
    283   1.1       jtk  * Never-fail realloc.
    284   1.1       jtk  */
    285   1.1       jtk static void *
    286   1.1       jtk nfrealloc(void *blk, size_t nb, const char *tag)
    287   1.1       jtk {
    288   1.1       jtk 	void *rv;
    289   1.1       jtk 
    290   1.1       jtk 	rv = realloc(blk, nb);
    291   1.1       jtk 	if (rv)
    292   1.1       jtk 		return (rv);
    293  1.13      haad 	err(EXIT_FAILURE, "Can't re-allocate %lu bytes for %s",
    294   1.4  christos 	    (unsigned long int) nb, tag);
    295   1.1       jtk }
    296   1.1       jtk /*
    297   1.1       jtk  * Allocate memory that will never be freed or reallocated.  Arguably
    298   1.1       jtk  *  this routine should handle small allocations by chopping up pages,
    299   1.1       jtk  *  but that's not worth the bother; it's not called more than a
    300   1.1       jtk  *  handful of times per run, and if the allocations are that small the
    301   1.1       jtk  *  waste in giving each one its own page is ignorable.
    302   1.1       jtk  */
    303   1.1       jtk static void *
    304   1.1       jtk alloconce(size_t nb, const char *tag)
    305   1.1       jtk {
    306   1.1       jtk 	void *rv;
    307   1.1       jtk 
    308   1.1       jtk 	rv = mmap(0, nb, PROT_READ | PROT_WRITE, MAP_ANON | MAP_PRIVATE, -1, 0);
    309   1.1       jtk 	if (rv != MAP_FAILED)
    310   1.1       jtk 		return (rv);
    311  1.13      haad 	err(EXIT_FAILURE, "Can't map %lu bytes for %s",
    312   1.4  christos 	    (unsigned long int) nb, tag);
    313   1.1       jtk }
    314   1.1       jtk /*
    315   1.1       jtk  * Load the cgs and csums off disk.  Also allocates the space to load
    316   1.1       jtk  *  them into and initializes the per-cg flags.
    317   1.1       jtk  */
    318   1.1       jtk static void
    319   1.1       jtk loadcgs(void)
    320   1.1       jtk {
    321   1.1       jtk 	int cg;
    322   1.1       jtk 	char *cgp;
    323   1.1       jtk 
    324   1.1       jtk 	cgblksz = roundup(oldsb->fs_cgsize, oldsb->fs_fsize);
    325  1.32  christos 	cgs = nfmalloc(oldsb->fs_ncg * sizeof(*cgs), "cg pointers");
    326   1.1       jtk 	cgp = alloconce(oldsb->fs_ncg * cgblksz, "cgs");
    327   1.1       jtk 	cgflags = nfmalloc(oldsb->fs_ncg, "cg flags");
    328   1.1       jtk 	csums = nfmalloc(oldsb->fs_cssize, "cg summary");
    329   1.1       jtk 	for (cg = 0; cg < oldsb->fs_ncg; cg++) {
    330   1.1       jtk 		cgs[cg] = (struct cg *) cgp;
    331  1.36  dholland 		readat(FFS_FSBTODB(oldsb, cgtod(oldsb, cg)), cgp, cgblksz);
    332  1.25       riz 		if (needswap)
    333  1.25       riz 			ffs_cg_swap(cgs[cg],cgs[cg],oldsb);
    334   1.1       jtk 		cgflags[cg] = 0;
    335   1.1       jtk 		cgp += cgblksz;
    336   1.1       jtk 	}
    337  1.36  dholland 	readat(FFS_FSBTODB(oldsb, oldsb->fs_csaddr), csums, oldsb->fs_cssize);
    338  1.25       riz 	if (needswap)
    339  1.25       riz 		ffs_csum_swap(csums,csums,oldsb->fs_cssize);
    340   1.1       jtk }
    341   1.1       jtk /*
    342   1.1       jtk  * Set n bits, starting with bit #base, in the bitmap pointed to by
    343   1.1       jtk  *  bitvec (which is assumed to be large enough to include bits base
    344   1.1       jtk  *  through base+n-1).
    345   1.1       jtk  */
    346   1.1       jtk static void
    347   1.1       jtk set_bits(unsigned char *bitvec, unsigned int base, unsigned int n)
    348   1.1       jtk {
    349   1.1       jtk 	if (n < 1)
    350   1.1       jtk 		return;		/* nothing to do */
    351   1.1       jtk 	if (base & 7) {		/* partial byte at beginning */
    352   1.1       jtk 		if (n <= 8 - (base & 7)) {	/* entirely within one byte */
    353   1.1       jtk 			bitvec[base >> 3] |= (~((~0U) << n)) << (base & 7);
    354   1.1       jtk 			return;
    355   1.1       jtk 		}
    356   1.1       jtk 		bitvec[base >> 3] |= (~0U) << (base & 7);
    357   1.1       jtk 		n -= 8 - (base & 7);
    358   1.1       jtk 		base = (base & ~7) + 8;
    359   1.1       jtk 	}
    360   1.1       jtk 	if (n >= 8) {		/* do full bytes */
    361   1.1       jtk 		memset(bitvec + (base >> 3), 0xff, n >> 3);
    362   1.1       jtk 		base += n & ~7;
    363   1.1       jtk 		n &= 7;
    364   1.1       jtk 	}
    365   1.1       jtk 	if (n) {		/* partial byte at end */
    366   1.1       jtk 		bitvec[base >> 3] |= ~((~0U) << n);
    367   1.1       jtk 	}
    368   1.1       jtk }
    369   1.1       jtk /*
    370   1.1       jtk  * Clear n bits, starting with bit #base, in the bitmap pointed to by
    371   1.1       jtk  *  bitvec (which is assumed to be large enough to include bits base
    372   1.1       jtk  *  through base+n-1).  Code parallels set_bits().
    373   1.1       jtk  */
    374   1.1       jtk static void
    375   1.1       jtk clr_bits(unsigned char *bitvec, int base, int n)
    376   1.1       jtk {
    377   1.1       jtk 	if (n < 1)
    378   1.1       jtk 		return;
    379   1.1       jtk 	if (base & 7) {
    380   1.1       jtk 		if (n <= 8 - (base & 7)) {
    381   1.1       jtk 			bitvec[base >> 3] &= ~((~((~0U) << n)) << (base & 7));
    382   1.1       jtk 			return;
    383   1.1       jtk 		}
    384   1.1       jtk 		bitvec[base >> 3] &= ~((~0U) << (base & 7));
    385   1.1       jtk 		n -= 8 - (base & 7);
    386   1.1       jtk 		base = (base & ~7) + 8;
    387   1.1       jtk 	}
    388   1.1       jtk 	if (n >= 8) {
    389  1.25       riz 		memset(bitvec + (base >> 3), 0, n >> 3);
    390   1.1       jtk 		base += n & ~7;
    391   1.1       jtk 		n &= 7;
    392   1.1       jtk 	}
    393   1.1       jtk 	if (n) {
    394   1.1       jtk 		bitvec[base >> 3] &= (~0U) << n;
    395   1.1       jtk 	}
    396   1.1       jtk }
    397   1.1       jtk /*
    398   1.1       jtk  * Test whether bit #bit is set in the bitmap pointed to by bitvec.
    399   1.1       jtk  */
    400  1.13      haad static int
    401   1.1       jtk bit_is_set(unsigned char *bitvec, int bit)
    402   1.1       jtk {
    403   1.1       jtk 	return (bitvec[bit >> 3] & (1 << (bit & 7)));
    404   1.1       jtk }
    405   1.1       jtk /*
    406   1.1       jtk  * Test whether bit #bit is clear in the bitmap pointed to by bitvec.
    407   1.1       jtk  */
    408  1.13      haad static int
    409   1.1       jtk bit_is_clr(unsigned char *bitvec, int bit)
    410   1.1       jtk {
    411   1.1       jtk 	return (!bit_is_set(bitvec, bit));
    412   1.1       jtk }
    413   1.1       jtk /*
    414   1.1       jtk  * Test whether a whole block of bits is set in a bitmap.  This is
    415   1.1       jtk  *  designed for testing (aligned) disk blocks in a bit-per-frag
    416   1.1       jtk  *  bitmap; it has assumptions wired into it based on that, essentially
    417   1.1       jtk  *  that the entire block fits into a single byte.  This returns true
    418   1.1       jtk  *  iff _all_ the bits are set; it is not just the complement of
    419   1.1       jtk  *  blk_is_clr on the same arguments (unless blkfrags==1).
    420   1.1       jtk  */
    421  1.13      haad static int
    422   1.1       jtk blk_is_set(unsigned char *bitvec, int blkbase, int blkfrags)
    423   1.1       jtk {
    424   1.1       jtk 	unsigned int mask;
    425   1.1       jtk 
    426   1.1       jtk 	mask = (~((~0U) << blkfrags)) << (blkbase & 7);
    427   1.1       jtk 	return ((bitvec[blkbase >> 3] & mask) == mask);
    428   1.1       jtk }
    429   1.1       jtk /*
    430   1.1       jtk  * Test whether a whole block of bits is clear in a bitmap.  See
    431   1.1       jtk  *  blk_is_set (above) for assumptions.  This returns true iff _all_
    432   1.1       jtk  *  the bits are clear; it is not just the complement of blk_is_set on
    433   1.1       jtk  *  the same arguments (unless blkfrags==1).
    434   1.1       jtk  */
    435  1.13      haad static int
    436   1.1       jtk blk_is_clr(unsigned char *bitvec, int blkbase, int blkfrags)
    437   1.1       jtk {
    438   1.1       jtk 	unsigned int mask;
    439   1.1       jtk 
    440   1.1       jtk 	mask = (~((~0U) << blkfrags)) << (blkbase & 7);
    441   1.1       jtk 	return ((bitvec[blkbase >> 3] & mask) == 0);
    442   1.1       jtk }
    443   1.1       jtk /*
    444   1.1       jtk  * Initialize a new cg.  Called when growing.  Assumes memory has been
    445   1.1       jtk  *  allocated but not otherwise set up.  This code sets the fields of
    446   1.1       jtk  *  the cg, initializes the bitmaps (and cluster summaries, if
    447   1.1       jtk  *  applicable), updates both per-cylinder summary info and the global
    448   1.1       jtk  *  summary info in newsb; it also writes out new inodes for the cg.
    449   1.1       jtk  *
    450   1.1       jtk  * This code knows it can never be called for cg 0, which makes it a
    451   1.1       jtk  *  bit simpler than it would otherwise be.
    452   1.1       jtk  */
    453   1.1       jtk static void
    454   1.1       jtk initcg(int cgn)
    455   1.1       jtk {
    456   1.1       jtk 	struct cg *cg;		/* The in-core cg, of course */
    457   1.1       jtk 	int base;		/* Disk address of cg base */
    458   1.1       jtk 	int dlow;		/* Size of pre-cg data area */
    459   1.1       jtk 	int dhigh;		/* Offset of post-inode data area, from base */
    460   1.1       jtk 	int dmax;		/* Offset of end of post-inode data area */
    461   1.1       jtk 	int i;			/* Generic loop index */
    462   1.1       jtk 	int n;			/* Generic count */
    463  1.25       riz 	int start;		/* start of cg maps */
    464   1.1       jtk 
    465   1.1       jtk 	cg = cgs[cgn];
    466   1.1       jtk 	/* Place the data areas */
    467   1.1       jtk 	base = cgbase(newsb, cgn);
    468   1.1       jtk 	dlow = cgsblock(newsb, cgn) - base;
    469   1.1       jtk 	dhigh = cgdmin(newsb, cgn) - base;
    470   1.1       jtk 	dmax = newsb->fs_size - base;
    471   1.1       jtk 	if (dmax > newsb->fs_fpg)
    472   1.1       jtk 		dmax = newsb->fs_fpg;
    473  1.25       riz 	start = &cg->cg_space[0] - (unsigned char *) cg;
    474   1.1       jtk 	/*
    475   1.1       jtk          * Clear out the cg - assumes all-0-bytes is the correct way
    476   1.1       jtk          * to initialize fields we don't otherwise touch, which is
    477   1.1       jtk          * perhaps not the right thing to do, but it's what fsck and
    478   1.1       jtk          * mkfs do.
    479   1.1       jtk          */
    480  1.25       riz 	memset(cg, 0, newsb->fs_cgsize);
    481  1.20    mhitch 	if (newsb->fs_old_flags & FS_FLAGS_UPDATED)
    482  1.20    mhitch 		cg->cg_time = newsb->fs_time;
    483   1.1       jtk 	cg->cg_magic = CG_MAGIC;
    484   1.1       jtk 	cg->cg_cgx = cgn;
    485  1.25       riz 	cg->cg_niblk = newsb->fs_ipg;
    486  1.25       riz 	cg->cg_ndblk = dmax;
    487  1.25       riz 
    488  1.25       riz 	if (is_ufs2) {
    489  1.25       riz 		cg->cg_time = newsb->fs_time;
    490  1.35  dholland 		cg->cg_initediblk = newsb->fs_ipg < 2 * FFS_INOPB(newsb) ?
    491  1.35  dholland 		    newsb->fs_ipg : 2 * FFS_INOPB(newsb);
    492  1.25       riz 		cg->cg_iusedoff = start;
    493  1.25       riz 	} else {
    494  1.25       riz 		cg->cg_old_time = newsb->fs_time;
    495  1.25       riz 		cg->cg_old_niblk = cg->cg_niblk;
    496  1.25       riz 		cg->cg_niblk = 0;
    497  1.25       riz 		cg->cg_initediblk = 0;
    498  1.28  dholland 
    499  1.28  dholland 
    500  1.25       riz 		cg->cg_old_ncyl = newsb->fs_old_cpg;
    501  1.25       riz 		/* Update the cg_old_ncyl value for the last cylinder. */
    502  1.25       riz 		if (cgn == newsb->fs_ncg - 1) {
    503  1.25       riz 			if ((newsb->fs_old_flags & FS_FLAGS_UPDATED) == 0)
    504  1.25       riz 				cg->cg_old_ncyl = newsb->fs_old_ncyl %
    505  1.25       riz 				    newsb->fs_old_cpg;
    506  1.25       riz 		}
    507  1.25       riz 
    508  1.25       riz 		/* Set up the bitmap pointers.  We have to be careful
    509  1.25       riz 		 * to lay out the cg _exactly_ the way mkfs and fsck
    510  1.25       riz 		 * do it, since fsck compares the _entire_ cg against
    511  1.25       riz 		 * a recomputed cg, and whines if there is any
    512  1.25       riz 		 * mismatch, including the bitmap offsets. */
    513  1.25       riz 		/* XXX update this comment when fsck is fixed */
    514  1.25       riz 		cg->cg_old_btotoff = start;
    515  1.25       riz 		cg->cg_old_boff = cg->cg_old_btotoff
    516  1.25       riz 		    + (newsb->fs_old_cpg * sizeof(int32_t));
    517  1.25       riz 		cg->cg_iusedoff = cg->cg_old_boff +
    518  1.25       riz 		    (newsb->fs_old_cpg * newsb->fs_old_nrpos * sizeof(int16_t));
    519   1.1       jtk 	}
    520   1.1       jtk 	cg->cg_freeoff = cg->cg_iusedoff + howmany(newsb->fs_ipg, NBBY);
    521   1.1       jtk 	if (newsb->fs_contigsumsize > 0) {
    522   1.1       jtk 		cg->cg_nclusterblks = cg->cg_ndblk / newsb->fs_frag;
    523   1.1       jtk 		cg->cg_clustersumoff = cg->cg_freeoff +
    524  1.25       riz 		    howmany(newsb->fs_fpg, NBBY) - sizeof(int32_t);
    525   1.1       jtk 		cg->cg_clustersumoff =
    526   1.1       jtk 		    roundup(cg->cg_clustersumoff, sizeof(int32_t));
    527   1.1       jtk 		cg->cg_clusteroff = cg->cg_clustersumoff +
    528   1.1       jtk 		    ((newsb->fs_contigsumsize + 1) * sizeof(int32_t));
    529   1.1       jtk 		cg->cg_nextfreeoff = cg->cg_clusteroff +
    530  1.38  dholland 		    howmany(ffs_fragstoblks(newsb,newsb->fs_fpg), NBBY);
    531   1.1       jtk 		n = dlow / newsb->fs_frag;
    532   1.1       jtk 		if (n > 0) {
    533   1.1       jtk 			set_bits(cg_clustersfree(cg, 0), 0, n);
    534   1.1       jtk 			cg_clustersum(cg, 0)[(n > newsb->fs_contigsumsize) ?
    535   1.1       jtk 			    newsb->fs_contigsumsize : n]++;
    536   1.1       jtk 		}
    537   1.1       jtk 	} else {
    538   1.1       jtk 		cg->cg_nextfreeoff = cg->cg_freeoff +
    539  1.25       riz 		    howmany(newsb->fs_fpg, NBBY);
    540   1.1       jtk 	}
    541   1.1       jtk 	/* Mark the data areas as free; everything else is marked busy by the
    542  1.25       riz 	 * memset() up at the top. */
    543   1.1       jtk 	set_bits(cg_blksfree(cg, 0), 0, dlow);
    544   1.1       jtk 	set_bits(cg_blksfree(cg, 0), dhigh, dmax - dhigh);
    545   1.1       jtk 	/* Initialize summary info */
    546   1.1       jtk 	cg->cg_cs.cs_ndir = 0;
    547   1.1       jtk 	cg->cg_cs.cs_nifree = newsb->fs_ipg;
    548   1.1       jtk 	cg->cg_cs.cs_nbfree = dlow / newsb->fs_frag;
    549   1.1       jtk 	cg->cg_cs.cs_nffree = 0;
    550  1.28  dholland 
    551  1.25       riz 	/* This is the simplest way of doing this; we perhaps could
    552  1.25       riz 	 * compute the correct cg_blktot()[] and cg_blks()[] values
    553  1.25       riz 	 * other ways, but it would be complicated and hardly seems
    554  1.25       riz 	 * worth the effort.  (The reason there isn't
    555  1.25       riz 	 * frag-at-beginning and frag-at-end code here, like the code
    556  1.25       riz 	 * below for the post-inode data area, is that the pre-sb data
    557  1.25       riz 	 * area always starts at 0, and thus is block-aligned, and
    558  1.25       riz 	 * always ends at the sb, which is block-aligned.) */
    559  1.28  dholland 	if ((newsb->fs_old_flags & FS_FLAGS_UPDATED) == 0)
    560  1.25       riz 		for (i = 0; i < dlow; i += newsb->fs_frag) {
    561  1.25       riz 			old_cg_blktot(cg, 0)[old_cbtocylno(newsb, i)]++;
    562  1.25       riz 			old_cg_blks(newsb, cg,
    563  1.25       riz 			    old_cbtocylno(newsb, i),
    564  1.25       riz 			    0)[old_cbtorpos(newsb, i)]++;
    565  1.25       riz 		}
    566   1.1       jtk 
    567   1.1       jtk 	/* Deal with a partial block at the beginning of the post-inode area.
    568   1.1       jtk 	 * I'm not convinced this can happen - I think the inodes are always
    569   1.1       jtk 	 * block-aligned and always an integral number of blocks - but it's
    570   1.1       jtk 	 * cheap to do the right thing just in case. */
    571   1.1       jtk 	if (dhigh % newsb->fs_frag) {
    572   1.1       jtk 		n = newsb->fs_frag - (dhigh % newsb->fs_frag);
    573   1.1       jtk 		cg->cg_frsum[n]++;
    574   1.1       jtk 		cg->cg_cs.cs_nffree += n;
    575   1.1       jtk 		dhigh += n;
    576   1.1       jtk 	}
    577   1.1       jtk 	n = (dmax - dhigh) / newsb->fs_frag;
    578   1.1       jtk 	/* We have n full-size blocks in the post-inode data area. */
    579   1.1       jtk 	if (n > 0) {
    580   1.1       jtk 		cg->cg_cs.cs_nbfree += n;
    581   1.1       jtk 		if (newsb->fs_contigsumsize > 0) {
    582   1.1       jtk 			i = dhigh / newsb->fs_frag;
    583   1.1       jtk 			set_bits(cg_clustersfree(cg, 0), i, n);
    584   1.1       jtk 			cg_clustersum(cg, 0)[(n > newsb->fs_contigsumsize) ?
    585   1.1       jtk 			    newsb->fs_contigsumsize : n]++;
    586   1.1       jtk 		}
    587  1.25       riz 		if (is_ufs2 == 0)
    588  1.25       riz 			for (i = n; i > 0; i--) {
    589  1.25       riz 				old_cg_blktot(cg, 0)[old_cbtocylno(newsb,
    590  1.25       riz 					    dhigh)]++;
    591  1.25       riz 				old_cg_blks(newsb, cg,
    592  1.25       riz 				    old_cbtocylno(newsb, dhigh),
    593  1.25       riz 				    0)[old_cbtorpos(newsb,
    594  1.25       riz 					    dhigh)]++;
    595  1.25       riz 				dhigh += newsb->fs_frag;
    596  1.25       riz 			}
    597  1.25       riz 	}
    598  1.25       riz 	if (is_ufs2 == 0) {
    599  1.25       riz 		/* Deal with any leftover frag at the end of the cg. */
    600  1.25       riz 		i = dmax - dhigh;
    601  1.25       riz 		if (i) {
    602  1.25       riz 			cg->cg_frsum[i]++;
    603  1.25       riz 			cg->cg_cs.cs_nffree += i;
    604   1.1       jtk 		}
    605   1.1       jtk 	}
    606   1.1       jtk 	/* Update the csum info. */
    607   1.1       jtk 	csums[cgn] = cg->cg_cs;
    608   1.1       jtk 	newsb->fs_cstotal.cs_nffree += cg->cg_cs.cs_nffree;
    609   1.1       jtk 	newsb->fs_cstotal.cs_nbfree += cg->cg_cs.cs_nbfree;
    610   1.1       jtk 	newsb->fs_cstotal.cs_nifree += cg->cg_cs.cs_nifree;
    611  1.25       riz 	if (is_ufs2 == 0)
    612  1.25       riz 		/* Write out the cleared inodes. */
    613  1.36  dholland 		writeat(FFS_FSBTODB(newsb, cgimin(newsb, cgn)), zinodes,
    614  1.32  christos 		    newsb->fs_ipg * sizeof(*zinodes));
    615   1.1       jtk 	/* Dirty the cg. */
    616   1.1       jtk 	cgflags[cgn] |= CGF_DIRTY;
    617   1.1       jtk }
    618   1.1       jtk /*
    619   1.1       jtk  * Find free space, at least nfrags consecutive frags of it.  Pays no
    620   1.1       jtk  *  attention to block boundaries, but refuses to straddle cg
    621   1.1       jtk  *  boundaries, even if the disk blocks involved are in fact
    622   1.1       jtk  *  consecutive.  Return value is the frag number of the first frag of
    623   1.1       jtk  *  the block, or -1 if no space was found.  Uses newsb for sb values,
    624   1.1       jtk  *  and assumes the cgs[] structures correctly describe the area to be
    625   1.1       jtk  *  searched.
    626   1.1       jtk  *
    627   1.1       jtk  * XXX is there a bug lurking in the ignoring of block boundaries by
    628   1.1       jtk  *  the routine used by fragmove() in evict_data()?  Can an end-of-file
    629   1.1       jtk  *  frag legally straddle a block boundary?  If not, this should be
    630   1.1       jtk  *  cloned and fixed to stop at block boundaries for that use.  The
    631   1.1       jtk  *  current one may still be needed for csum info motion, in case that
    632   1.1       jtk  *  takes up more than a whole block (is the csum info allowed to begin
    633   1.1       jtk  *  partway through a block and continue into the following block?).
    634   1.1       jtk  *
    635  1.24       wiz  * If we wrap off the end of the file system back to the beginning, we
    636  1.24       wiz  *  can end up searching the end of the file system twice.  I ignore
    637   1.1       jtk  *  this inefficiency, since if that happens we're going to croak with
    638   1.1       jtk  *  a no-space error anyway, so it happens at most once.
    639   1.1       jtk  */
    640   1.1       jtk static int
    641   1.1       jtk find_freespace(unsigned int nfrags)
    642   1.1       jtk {
    643   1.1       jtk 	static int hand = 0;	/* hand rotates through all frags in the fs */
    644   1.1       jtk 	int cgsize;		/* size of the cg hand currently points into */
    645   1.1       jtk 	int cgn;		/* number of cg hand currently points into */
    646   1.1       jtk 	int fwc;		/* frag-within-cg number of frag hand points
    647   1.1       jtk 				 * to */
    648  1.30  dholland 	unsigned int run;	/* length of run of free frags seen so far */
    649   1.1       jtk 	int secondpass;		/* have we wrapped from end of fs to
    650   1.1       jtk 				 * beginning? */
    651   1.1       jtk 	unsigned char *bits;	/* cg_blksfree()[] for cg hand points into */
    652   1.1       jtk 
    653   1.1       jtk 	cgn = dtog(newsb, hand);
    654   1.1       jtk 	fwc = dtogd(newsb, hand);
    655   1.1       jtk 	secondpass = (hand == 0);
    656   1.1       jtk 	run = 0;
    657   1.1       jtk 	bits = cg_blksfree(cgs[cgn], 0);
    658   1.1       jtk 	cgsize = cgs[cgn]->cg_ndblk;
    659   1.1       jtk 	while (1) {
    660   1.1       jtk 		if (bit_is_set(bits, fwc)) {
    661   1.1       jtk 			run++;
    662   1.1       jtk 			if (run >= nfrags)
    663   1.1       jtk 				return (hand + 1 - run);
    664   1.1       jtk 		} else {
    665   1.1       jtk 			run = 0;
    666   1.1       jtk 		}
    667   1.1       jtk 		hand++;
    668   1.1       jtk 		fwc++;
    669   1.1       jtk 		if (fwc >= cgsize) {
    670   1.1       jtk 			fwc = 0;
    671   1.1       jtk 			cgn++;
    672   1.1       jtk 			if (cgn >= newsb->fs_ncg) {
    673   1.1       jtk 				hand = 0;
    674   1.1       jtk 				if (secondpass)
    675   1.1       jtk 					return (-1);
    676   1.1       jtk 				secondpass = 1;
    677   1.1       jtk 				cgn = 0;
    678   1.1       jtk 			}
    679   1.1       jtk 			bits = cg_blksfree(cgs[cgn], 0);
    680   1.1       jtk 			cgsize = cgs[cgn]->cg_ndblk;
    681   1.1       jtk 			run = 0;
    682   1.1       jtk 		}
    683   1.1       jtk 	}
    684   1.1       jtk }
    685   1.1       jtk /*
    686   1.1       jtk  * Find a free block of disk space.  Finds an entire block of frags,
    687   1.1       jtk  *  all of which are free.  Return value is the frag number of the
    688   1.1       jtk  *  first frag of the block, or -1 if no space was found.  Uses newsb
    689   1.1       jtk  *  for sb values, and assumes the cgs[] structures correctly describe
    690   1.1       jtk  *  the area to be searched.
    691   1.1       jtk  *
    692   1.1       jtk  * See find_freespace(), above, for remarks about hand wrapping around.
    693   1.1       jtk  */
    694   1.1       jtk static int
    695   1.1       jtk find_freeblock(void)
    696   1.1       jtk {
    697   1.1       jtk 	static int hand = 0;	/* hand rotates through all frags in fs */
    698   1.1       jtk 	int cgn;		/* cg number of cg hand points into */
    699   1.1       jtk 	int fwc;		/* frag-within-cg number of frag hand points
    700   1.1       jtk 				 * to */
    701   1.1       jtk 	int cgsize;		/* size of cg hand points into */
    702   1.1       jtk 	int secondpass;		/* have we wrapped from end to beginning? */
    703   1.1       jtk 	unsigned char *bits;	/* cg_blksfree()[] for cg hand points into */
    704   1.1       jtk 
    705   1.1       jtk 	cgn = dtog(newsb, hand);
    706   1.1       jtk 	fwc = dtogd(newsb, hand);
    707   1.1       jtk 	secondpass = (hand == 0);
    708   1.1       jtk 	bits = cg_blksfree(cgs[cgn], 0);
    709  1.38  dholland 	cgsize = ffs_blknum(newsb, cgs[cgn]->cg_ndblk);
    710   1.1       jtk 	while (1) {
    711   1.1       jtk 		if (blk_is_set(bits, fwc, newsb->fs_frag))
    712   1.1       jtk 			return (hand);
    713   1.1       jtk 		fwc += newsb->fs_frag;
    714   1.1       jtk 		hand += newsb->fs_frag;
    715   1.1       jtk 		if (fwc >= cgsize) {
    716   1.1       jtk 			fwc = 0;
    717   1.1       jtk 			cgn++;
    718   1.1       jtk 			if (cgn >= newsb->fs_ncg) {
    719   1.1       jtk 				hand = 0;
    720   1.1       jtk 				if (secondpass)
    721   1.1       jtk 					return (-1);
    722   1.1       jtk 				secondpass = 1;
    723   1.1       jtk 				cgn = 0;
    724   1.1       jtk 			}
    725   1.1       jtk 			bits = cg_blksfree(cgs[cgn], 0);
    726  1.38  dholland 			cgsize = ffs_blknum(newsb, cgs[cgn]->cg_ndblk);
    727   1.1       jtk 		}
    728   1.1       jtk 	}
    729   1.1       jtk }
    730   1.1       jtk /*
    731   1.1       jtk  * Find a free inode, returning its inumber or -1 if none was found.
    732   1.1       jtk  *  Uses newsb for sb values, and assumes the cgs[] structures
    733   1.1       jtk  *  correctly describe the area to be searched.
    734   1.1       jtk  *
    735   1.1       jtk  * See find_freespace(), above, for remarks about hand wrapping around.
    736   1.1       jtk  */
    737   1.1       jtk static int
    738   1.1       jtk find_freeinode(void)
    739   1.1       jtk {
    740   1.1       jtk 	static int hand = 0;	/* hand rotates through all inodes in fs */
    741   1.1       jtk 	int cgn;		/* cg number of cg hand points into */
    742   1.1       jtk 	int iwc;		/* inode-within-cg number of inode hand points
    743   1.1       jtk 				 * to */
    744   1.1       jtk 	int secondpass;		/* have we wrapped from end to beginning? */
    745   1.1       jtk 	unsigned char *bits;	/* cg_inosused()[] for cg hand points into */
    746   1.1       jtk 
    747   1.1       jtk 	cgn = hand / newsb->fs_ipg;
    748   1.1       jtk 	iwc = hand % newsb->fs_ipg;
    749   1.1       jtk 	secondpass = (hand == 0);
    750   1.1       jtk 	bits = cg_inosused(cgs[cgn], 0);
    751   1.1       jtk 	while (1) {
    752   1.1       jtk 		if (bit_is_clr(bits, iwc))
    753   1.1       jtk 			return (hand);
    754   1.1       jtk 		hand++;
    755   1.1       jtk 		iwc++;
    756   1.1       jtk 		if (iwc >= newsb->fs_ipg) {
    757   1.1       jtk 			iwc = 0;
    758   1.1       jtk 			cgn++;
    759   1.1       jtk 			if (cgn >= newsb->fs_ncg) {
    760   1.1       jtk 				hand = 0;
    761   1.1       jtk 				if (secondpass)
    762   1.1       jtk 					return (-1);
    763   1.1       jtk 				secondpass = 1;
    764   1.1       jtk 				cgn = 0;
    765   1.1       jtk 			}
    766   1.1       jtk 			bits = cg_inosused(cgs[cgn], 0);
    767   1.1       jtk 		}
    768   1.1       jtk 	}
    769   1.1       jtk }
    770   1.1       jtk /*
    771   1.1       jtk  * Mark a frag as free.  Sets the frag's bit in the cg_blksfree bitmap
    772   1.1       jtk  *  for the appropriate cg, and marks the cg as dirty.
    773   1.1       jtk  */
    774   1.1       jtk static void
    775   1.1       jtk free_frag(int fno)
    776   1.1       jtk {
    777   1.1       jtk 	int cgn;
    778   1.1       jtk 
    779   1.1       jtk 	cgn = dtog(newsb, fno);
    780   1.1       jtk 	set_bits(cg_blksfree(cgs[cgn], 0), dtogd(newsb, fno), 1);
    781   1.1       jtk 	cgflags[cgn] |= CGF_DIRTY | CGF_BLKMAPS;
    782   1.1       jtk }
    783   1.1       jtk /*
    784   1.1       jtk  * Allocate a frag.  Clears the frag's bit in the cg_blksfree bitmap
    785   1.1       jtk  *  for the appropriate cg, and marks the cg as dirty.
    786   1.1       jtk  */
    787   1.1       jtk static void
    788   1.1       jtk alloc_frag(int fno)
    789   1.1       jtk {
    790   1.1       jtk 	int cgn;
    791   1.1       jtk 
    792   1.1       jtk 	cgn = dtog(newsb, fno);
    793   1.1       jtk 	clr_bits(cg_blksfree(cgs[cgn], 0), dtogd(newsb, fno), 1);
    794   1.1       jtk 	cgflags[cgn] |= CGF_DIRTY | CGF_BLKMAPS;
    795   1.1       jtk }
    796   1.1       jtk /*
    797   1.1       jtk  * Fix up the csum array.  If shrinking, this involves freeing zero or
    798   1.1       jtk  *  more frags; if growing, it involves allocating them, or if the
    799   1.1       jtk  *  frags being grown into aren't free, finding space elsewhere for the
    800   1.1       jtk  *  csum info.  (If the number of occupied frags doesn't change,
    801   1.1       jtk  *  nothing happens here.)
    802   1.1       jtk  */
    803   1.1       jtk static void
    804   1.1       jtk csum_fixup(void)
    805   1.1       jtk {
    806   1.1       jtk 	int nold;		/* # frags in old csum info */
    807   1.1       jtk 	int ntot;		/* # frags in new csum info */
    808   1.1       jtk 	int nnew;		/* ntot-nold */
    809   1.1       jtk 	int newloc;		/* new location for csum info, if necessary */
    810   1.1       jtk 	int i;			/* generic loop index */
    811   1.1       jtk 	int j;			/* generic loop index */
    812   1.1       jtk 	int f;			/* "from" frag number, if moving */
    813   1.1       jtk 	int t;			/* "to" frag number, if moving */
    814   1.1       jtk 	int cgn;		/* cg number, used when shrinking */
    815   1.1       jtk 
    816   1.1       jtk 	ntot = howmany(newsb->fs_cssize, newsb->fs_fsize);
    817   1.1       jtk 	nold = howmany(oldsb->fs_cssize, newsb->fs_fsize);
    818   1.1       jtk 	nnew = ntot - nold;
    819   1.1       jtk 	/* First, if there's no change in frag counts, it's easy. */
    820   1.1       jtk 	if (nnew == 0)
    821   1.1       jtk 		return;
    822   1.1       jtk 	/* Next, if we're shrinking, it's almost as easy.  Just free up any
    823   1.1       jtk 	 * frags in the old area we no longer need. */
    824   1.1       jtk 	if (nnew < 0) {
    825   1.1       jtk 		for ((i = newsb->fs_csaddr + ntot - 1), (j = nnew);
    826   1.1       jtk 		    j < 0;
    827   1.1       jtk 		    i--, j++) {
    828   1.1       jtk 			free_frag(i);
    829   1.1       jtk 		}
    830   1.1       jtk 		return;
    831   1.1       jtk 	}
    832   1.1       jtk 	/* We must be growing.  Check to see that the new csum area fits
    833  1.24       wiz 	 * within the file system.  I think this can never happen, since for
    834   1.1       jtk 	 * the csum area to grow, we must be adding at least one cg, so the
    835  1.24       wiz 	 * old csum area can't be this close to the end of the new file system.
    836   1.1       jtk 	 * But it's a cheap check. */
    837   1.1       jtk 	/* XXX what if csum info is at end of cg and grows into next cg, what
    838   1.1       jtk 	 * if it spills over onto the next cg's backup superblock?  Can this
    839   1.1       jtk 	 * happen? */
    840   1.1       jtk 	if (newsb->fs_csaddr + ntot <= newsb->fs_size) {
    841   1.1       jtk 		/* Okay, it fits - now,  see if the space we want is free. */
    842   1.1       jtk 		for ((i = newsb->fs_csaddr + nold), (j = nnew);
    843   1.1       jtk 		    j > 0;
    844   1.1       jtk 		    i++, j--) {
    845   1.1       jtk 			cgn = dtog(newsb, i);
    846   1.1       jtk 			if (bit_is_clr(cg_blksfree(cgs[cgn], 0),
    847   1.1       jtk 				dtogd(newsb, i)))
    848   1.1       jtk 				break;
    849   1.1       jtk 		}
    850   1.1       jtk 		if (j <= 0) {
    851   1.1       jtk 			/* Win win - all the frags we want are free. Allocate
    852   1.1       jtk 			 * 'em and we're all done.  */
    853  1.25       riz 			for ((i = newsb->fs_csaddr + ntot - nnew),
    854  1.25       riz 				 (j = nnew); j > 0; i++, j--) {
    855   1.1       jtk 				alloc_frag(i);
    856   1.1       jtk 			}
    857   1.1       jtk 			return;
    858   1.1       jtk 		}
    859   1.1       jtk 	}
    860   1.1       jtk 	/* We have to move the csum info, sigh.  Look for new space, free old
    861   1.1       jtk 	 * space, and allocate new.  Update fs_csaddr.  We don't copy anything
    862   1.1       jtk 	 * on disk at this point; the csum info will be written to the
    863   1.1       jtk 	 * then-current fs_csaddr as part of the final flush. */
    864   1.1       jtk 	newloc = find_freespace(ntot);
    865  1.32  christos 	if (newloc < 0)
    866  1.32  christos 		errx(EXIT_FAILURE, "Sorry, no space available for new csums");
    867   1.1       jtk 	for (i = 0, f = newsb->fs_csaddr, t = newloc; i < ntot; i++, f++, t++) {
    868   1.1       jtk 		if (i < nold) {
    869   1.1       jtk 			free_frag(f);
    870   1.1       jtk 		}
    871   1.1       jtk 		alloc_frag(t);
    872   1.1       jtk 	}
    873   1.1       jtk 	newsb->fs_csaddr = newloc;
    874   1.1       jtk }
    875   1.1       jtk /*
    876   1.1       jtk  * Recompute newsb->fs_dsize.  Just scans all cgs, adding the number of
    877   1.1       jtk  *  data blocks in that cg to the total.
    878   1.1       jtk  */
    879   1.1       jtk static void
    880   1.1       jtk recompute_fs_dsize(void)
    881   1.1       jtk {
    882   1.1       jtk 	int i;
    883   1.1       jtk 
    884   1.1       jtk 	newsb->fs_dsize = 0;
    885   1.1       jtk 	for (i = 0; i < newsb->fs_ncg; i++) {
    886   1.1       jtk 		int dlow;	/* size of before-sb data area */
    887   1.1       jtk 		int dhigh;	/* offset of post-inode data area */
    888   1.1       jtk 		int dmax;	/* total size of cg */
    889   1.1       jtk 		int base;	/* base of cg, since cgsblock() etc add it in */
    890   1.1       jtk 		base = cgbase(newsb, i);
    891   1.1       jtk 		dlow = cgsblock(newsb, i) - base;
    892   1.1       jtk 		dhigh = cgdmin(newsb, i) - base;
    893   1.1       jtk 		dmax = newsb->fs_size - base;
    894   1.1       jtk 		if (dmax > newsb->fs_fpg)
    895   1.1       jtk 			dmax = newsb->fs_fpg;
    896   1.1       jtk 		newsb->fs_dsize += dlow + dmax - dhigh;
    897   1.1       jtk 	}
    898   1.1       jtk 	/* Space in cg 0 before cgsblock is boot area, not free space! */
    899   1.1       jtk 	newsb->fs_dsize -= cgsblock(newsb, 0) - cgbase(newsb, 0);
    900   1.1       jtk 	/* And of course the csum info takes up space. */
    901   1.1       jtk 	newsb->fs_dsize -= howmany(newsb->fs_cssize, newsb->fs_fsize);
    902   1.1       jtk }
    903   1.1       jtk /*
    904   1.1       jtk  * Return the current time.  We call this and assign, rather than
    905   1.1       jtk  *  calling time() directly, as insulation against OSes where fs_time
    906   1.1       jtk  *  is not a time_t.
    907   1.1       jtk  */
    908   1.1       jtk static time_t
    909   1.1       jtk timestamp(void)
    910   1.1       jtk {
    911   1.1       jtk 	time_t t;
    912   1.1       jtk 
    913   1.1       jtk 	time(&t);
    914   1.1       jtk 	return (t);
    915   1.1       jtk }
    916   1.1       jtk /*
    917  1.24       wiz  * Grow the file system.
    918   1.1       jtk  */
    919   1.1       jtk static void
    920   1.1       jtk grow(void)
    921   1.1       jtk {
    922   1.1       jtk 	int i;
    923   1.1       jtk 
    924   1.1       jtk 	/* Update the timestamp. */
    925   1.1       jtk 	newsb->fs_time = timestamp();
    926   1.1       jtk 	/* Allocate and clear the new-inode area, in case we add any cgs. */
    927  1.32  christos 	zinodes = alloconce(newsb->fs_ipg * sizeof(*zinodes), "zeroed inodes");
    928  1.32  christos 	memset(zinodes, 0, newsb->fs_ipg * sizeof(*zinodes));
    929   1.1       jtk 	/* Update the size. */
    930  1.36  dholland 	newsb->fs_size = FFS_DBTOFSB(newsb, newsize);
    931   1.1       jtk 	/* Did we actually not grow?  (This can happen if newsize is less than
    932   1.1       jtk 	 * a frag larger than the old size - unlikely, but no excuse to
    933   1.1       jtk 	 * misbehave if it happens.) */
    934  1.13      haad 	if (newsb->fs_size == oldsb->fs_size) {
    935  1.32  christos 		printf("New fs size %"PRIu64" = old fs size %"PRIu64
    936  1.15       riz 		    ", not growing.\n", newsb->fs_size, oldsb->fs_size);
    937   1.1       jtk 		return;
    938  1.13      haad 	}
    939   1.1       jtk 	/* Check that the new last sector (frag, actually) is writable.  Since
    940   1.1       jtk 	 * it's at least one frag larger than it used to be, we know we aren't
    941   1.1       jtk 	 * overwriting anything important by this.  (The choice of sbbuf as
    942   1.1       jtk 	 * what to write is irrelevant; it's just something handy that's known
    943   1.1       jtk 	 * to be at least one frag in size.) */
    944  1.36  dholland 	writeat(FFS_FSBTODB(newsb,newsb->fs_size - 1), &sbbuf, newsb->fs_fsize);
    945  1.25       riz 	if (is_ufs2)
    946  1.25       riz 		newsb->fs_ncg = howmany(newsb->fs_size, newsb->fs_fpg);
    947  1.25       riz 	else {
    948  1.25       riz 		/* Update fs_old_ncyl and fs_ncg. */
    949  1.25       riz 		newsb->fs_old_ncyl = howmany(newsb->fs_size * NSPF(newsb),
    950  1.25       riz 		    newsb->fs_old_spc);
    951  1.25       riz 		newsb->fs_ncg = howmany(newsb->fs_old_ncyl, newsb->fs_old_cpg);
    952  1.25       riz 	}
    953  1.28  dholland 
    954   1.1       jtk 	/* Does the last cg end before the end of its inode area? There is no
    955   1.1       jtk 	 * reason why this couldn't be handled, but it would complicate a lot
    956  1.24       wiz 	 * of code (in all file system code - fsck, kernel, etc) because of the
    957   1.1       jtk 	 * potential partial inode area, and the gain in space would be
    958   1.1       jtk 	 * minimal, at most the pre-sb data area. */
    959   1.1       jtk 	if (cgdmin(newsb, newsb->fs_ncg - 1) > newsb->fs_size) {
    960   1.1       jtk 		newsb->fs_ncg--;
    961   1.4  christos 		newsb->fs_old_ncyl = newsb->fs_ncg * newsb->fs_old_cpg;
    962  1.15       riz 		newsb->fs_size = (newsb->fs_old_ncyl * newsb->fs_old_spc)
    963  1.15       riz 		    / NSPF(newsb);
    964   1.1       jtk 		printf("Warning: last cylinder group is too small;\n");
    965   1.1       jtk 		printf("    dropping it.  New size = %lu.\n",
    966  1.36  dholland 		    (unsigned long int) FFS_FSBTODB(newsb, newsb->fs_size));
    967   1.1       jtk 	}
    968   1.1       jtk 	/* Find out how big the csum area is, and realloc csums if bigger. */
    969  1.37  dholland 	newsb->fs_cssize = ffs_fragroundup(newsb,
    970   1.1       jtk 	    newsb->fs_ncg * sizeof(struct csum));
    971   1.1       jtk 	if (newsb->fs_cssize > oldsb->fs_cssize)
    972   1.1       jtk 		csums = nfrealloc(csums, newsb->fs_cssize, "new cg summary");
    973  1.25       riz 	/* If we're adding any cgs, realloc structures and set up the new
    974  1.25       riz 	   cgs. */
    975   1.1       jtk 	if (newsb->fs_ncg > oldsb->fs_ncg) {
    976   1.1       jtk 		char *cgp;
    977  1.32  christos 		cgs = nfrealloc(cgs, newsb->fs_ncg * sizeof(*cgs),
    978   1.1       jtk                                 "cg pointers");
    979   1.1       jtk 		cgflags = nfrealloc(cgflags, newsb->fs_ncg, "cg flags");
    980  1.25       riz 		memset(cgflags + oldsb->fs_ncg, 0,
    981  1.25       riz 		    newsb->fs_ncg - oldsb->fs_ncg);
    982   1.1       jtk 		cgp = alloconce((newsb->fs_ncg - oldsb->fs_ncg) * cgblksz,
    983   1.1       jtk                                 "cgs");
    984   1.1       jtk 		for (i = oldsb->fs_ncg; i < newsb->fs_ncg; i++) {
    985   1.1       jtk 			cgs[i] = (struct cg *) cgp;
    986   1.1       jtk 			initcg(i);
    987   1.1       jtk 			cgp += cgblksz;
    988   1.1       jtk 		}
    989   1.4  christos 		cgs[oldsb->fs_ncg - 1]->cg_old_ncyl = oldsb->fs_old_cpg;
    990   1.1       jtk 		cgflags[oldsb->fs_ncg - 1] |= CGF_DIRTY;
    991   1.1       jtk 	}
    992   1.1       jtk 	/* If the old fs ended partway through a cg, we have to update the old
    993   1.1       jtk 	 * last cg (though possibly not to a full cg!). */
    994   1.1       jtk 	if (oldsb->fs_size % oldsb->fs_fpg) {
    995   1.1       jtk 		struct cg *cg;
    996   1.1       jtk 		int newcgsize;
    997   1.1       jtk 		int prevcgtop;
    998   1.1       jtk 		int oldcgsize;
    999   1.1       jtk 		cg = cgs[oldsb->fs_ncg - 1];
   1000   1.1       jtk 		cgflags[oldsb->fs_ncg - 1] |= CGF_DIRTY | CGF_BLKMAPS;
   1001   1.1       jtk 		prevcgtop = oldsb->fs_fpg * (oldsb->fs_ncg - 1);
   1002   1.1       jtk 		newcgsize = newsb->fs_size - prevcgtop;
   1003   1.1       jtk 		if (newcgsize > newsb->fs_fpg)
   1004   1.1       jtk 			newcgsize = newsb->fs_fpg;
   1005   1.1       jtk 		oldcgsize = oldsb->fs_size % oldsb->fs_fpg;
   1006   1.1       jtk 		set_bits(cg_blksfree(cg, 0), oldcgsize, newcgsize - oldcgsize);
   1007  1.20    mhitch 		cg->cg_old_ncyl = oldsb->fs_old_cpg;
   1008   1.1       jtk 		cg->cg_ndblk = newcgsize;
   1009   1.1       jtk 	}
   1010   1.1       jtk 	/* Fix up the csum info, if necessary. */
   1011   1.1       jtk 	csum_fixup();
   1012   1.1       jtk 	/* Make fs_dsize match the new reality. */
   1013   1.1       jtk 	recompute_fs_dsize();
   1014   1.1       jtk }
   1015   1.1       jtk /*
   1016   1.1       jtk  * Call (*fn)() for each inode, passing the inode and its inumber.  The
   1017   1.1       jtk  *  number of cylinder groups is pased in, so this can be used to map
   1018  1.24       wiz  *  over either the old or the new file system's set of inodes.
   1019   1.1       jtk  */
   1020   1.1       jtk static void
   1021  1.25       riz map_inodes(void (*fn) (union dinode * di, unsigned int, void *arg),
   1022  1.15       riz 	   int ncg, void *cbarg) {
   1023   1.1       jtk 	int i;
   1024   1.1       jtk 	int ni;
   1025   1.1       jtk 
   1026   1.1       jtk 	ni = oldsb->fs_ipg * ncg;
   1027   1.1       jtk 	for (i = 0; i < ni; i++)
   1028   1.1       jtk 		(*fn) (inodes + i, i, cbarg);
   1029   1.1       jtk }
   1030   1.1       jtk /* Values for the third argument to the map function for
   1031   1.1       jtk  * map_inode_data_blocks.  MDB_DATA indicates the block is contains
   1032   1.1       jtk  * file data; MDB_INDIR_PRE and MDB_INDIR_POST indicate that it's an
   1033   1.1       jtk  * indirect block.  The MDB_INDIR_PRE call is made before the indirect
   1034   1.1       jtk  * block pointers are followed and the pointed-to blocks scanned,
   1035   1.1       jtk  * MDB_INDIR_POST after.
   1036   1.1       jtk  */
   1037   1.1       jtk #define MDB_DATA       1
   1038   1.1       jtk #define MDB_INDIR_PRE  2
   1039   1.1       jtk #define MDB_INDIR_POST 3
   1040   1.1       jtk 
   1041  1.30  dholland typedef void (*mark_callback_t) (off_t blocknum, unsigned int nfrags,
   1042  1.15       riz 				 unsigned int blksize, int opcode);
   1043   1.1       jtk 
   1044   1.1       jtk /* Helper function - handles a data block.  Calls the callback
   1045   1.1       jtk  * function and returns number of bytes occupied in file (actually,
   1046   1.1       jtk  * rounded up to a frag boundary).  The name is historical.  */
   1047   1.1       jtk static int
   1048  1.30  dholland markblk(mark_callback_t fn, union dinode * di, off_t bn, off_t o)
   1049   1.1       jtk {
   1050   1.1       jtk 	int sz;
   1051   1.1       jtk 	int nb;
   1052  1.30  dholland 	off_t filesize;
   1053  1.26  dholland 
   1054  1.30  dholland 	filesize = DIP(di,di_size);
   1055  1.30  dholland 	if (o >= filesize)
   1056   1.1       jtk 		return (0);
   1057  1.37  dholland 	sz = dblksize(newsb, di, ffs_lblkno(newsb, o), filesize);
   1058  1.30  dholland 	nb = (sz > filesize - o) ? filesize - o : sz;
   1059   1.1       jtk 	if (bn)
   1060  1.37  dholland 		(*fn) (bn, ffs_numfrags(newsb, sz), nb, MDB_DATA);
   1061   1.1       jtk 	return (sz);
   1062   1.1       jtk }
   1063   1.1       jtk /* Helper function - handles an indirect block.  Makes the
   1064   1.1       jtk  * MDB_INDIR_PRE callback for the indirect block, loops over the
   1065   1.1       jtk  * pointers and recurses, and makes the MDB_INDIR_POST callback.
   1066   1.1       jtk  * Returns the number of bytes occupied in file, as does markblk().
   1067   1.1       jtk  * For the sake of update_for_data_move(), we read the indirect block
   1068   1.1       jtk  * _after_ making the _PRE callback.  The name is historical.  */
   1069   1.1       jtk static int
   1070  1.30  dholland markiblk(mark_callback_t fn, union dinode * di, off_t bn, off_t o, int lev)
   1071   1.1       jtk {
   1072   1.1       jtk 	int i;
   1073   1.1       jtk 	int j;
   1074  1.30  dholland 	unsigned k;
   1075   1.1       jtk 	int tot;
   1076   1.2    martin 	static int32_t indirblk1[howmany(MAXBSIZE, sizeof(int32_t))];
   1077   1.2    martin 	static int32_t indirblk2[howmany(MAXBSIZE, sizeof(int32_t))];
   1078   1.2    martin 	static int32_t indirblk3[howmany(MAXBSIZE, sizeof(int32_t))];
   1079   1.2    martin 	static int32_t *indirblks[3] = {
   1080   1.1       jtk 		&indirblk1[0], &indirblk2[0], &indirblk3[0]
   1081   1.1       jtk 	};
   1082  1.26  dholland 
   1083   1.1       jtk 	if (lev < 0)
   1084   1.1       jtk 		return (markblk(fn, di, bn, o));
   1085   1.1       jtk 	if (bn == 0) {
   1086   1.1       jtk 		for (i = newsb->fs_bsize;
   1087   1.1       jtk 		    lev >= 0;
   1088  1.35  dholland 		    i *= FFS_NINDIR(newsb), lev--);
   1089   1.1       jtk 		return (i);
   1090   1.1       jtk 	}
   1091   1.1       jtk 	(*fn) (bn, newsb->fs_frag, newsb->fs_bsize, MDB_INDIR_PRE);
   1092  1.36  dholland 	readat(FFS_FSBTODB(newsb, bn), indirblks[lev], newsb->fs_bsize);
   1093  1.25       riz 	if (needswap)
   1094  1.30  dholland 		for (k = 0; k < howmany(MAXBSIZE, sizeof(int32_t)); k++)
   1095  1.30  dholland 			indirblks[lev][k] = bswap32(indirblks[lev][k]);
   1096   1.1       jtk 	tot = 0;
   1097  1.35  dholland 	for (i = 0; i < FFS_NINDIR(newsb); i++) {
   1098   1.1       jtk 		j = markiblk(fn, di, indirblks[lev][i], o, lev - 1);
   1099   1.1       jtk 		if (j == 0)
   1100   1.1       jtk 			break;
   1101   1.1       jtk 		o += j;
   1102   1.1       jtk 		tot += j;
   1103   1.1       jtk 	}
   1104   1.1       jtk 	(*fn) (bn, newsb->fs_frag, newsb->fs_bsize, MDB_INDIR_POST);
   1105   1.1       jtk 	return (tot);
   1106   1.1       jtk }
   1107   1.1       jtk 
   1108   1.1       jtk 
   1109   1.1       jtk /*
   1110   1.1       jtk  * Call (*fn)() for each data block for an inode.  This routine assumes
   1111   1.1       jtk  *  the inode is known to be of a type that has data blocks (file,
   1112   1.1       jtk  *  directory, or non-fast symlink).  The called function is:
   1113   1.1       jtk  *
   1114   1.1       jtk  * (*fn)(unsigned int blkno, unsigned int nf, unsigned int nb, int op)
   1115   1.1       jtk  *
   1116   1.1       jtk  *  where blkno is the frag number, nf is the number of frags starting
   1117   1.1       jtk  *  at blkno (always <= fs_frag), nb is the number of bytes that belong
   1118   1.1       jtk  *  to the file (usually nf*fs_frag, often less for the last block/frag
   1119   1.1       jtk  *  of a file).
   1120   1.1       jtk  */
   1121   1.1       jtk static void
   1122  1.25       riz map_inode_data_blocks(union dinode * di, mark_callback_t fn)
   1123   1.1       jtk {
   1124   1.1       jtk 	off_t o;		/* offset within  inode */
   1125   1.1       jtk 	int inc;		/* increment for o - maybe should be off_t? */
   1126   1.1       jtk 	int b;			/* index within di_db[] and di_ib[] arrays */
   1127   1.1       jtk 
   1128   1.1       jtk 	/* Scan the direct blocks... */
   1129   1.1       jtk 	o = 0;
   1130  1.34  dholland 	for (b = 0; b < UFS_NDADDR; b++) {
   1131  1.25       riz 		inc = markblk(fn, di, DIP(di,di_db[b]), o);
   1132   1.1       jtk 		if (inc == 0)
   1133   1.1       jtk 			break;
   1134   1.1       jtk 		o += inc;
   1135   1.1       jtk 	}
   1136   1.1       jtk 	/* ...and the indirect blocks. */
   1137   1.1       jtk 	if (inc) {
   1138  1.34  dholland 		for (b = 0; b < UFS_NIADDR; b++) {
   1139  1.25       riz 			inc = markiblk(fn, di, DIP(di,di_ib[b]), o, b);
   1140   1.1       jtk 			if (inc == 0)
   1141   1.1       jtk 				return;
   1142   1.1       jtk 			o += inc;
   1143   1.1       jtk 		}
   1144   1.1       jtk 	}
   1145   1.1       jtk }
   1146   1.1       jtk 
   1147   1.1       jtk static void
   1148  1.25       riz dblk_callback(union dinode * di, unsigned int inum, void *arg)
   1149   1.1       jtk {
   1150   1.1       jtk 	mark_callback_t fn;
   1151  1.30  dholland 	off_t filesize;
   1152  1.26  dholland 
   1153  1.30  dholland 	filesize = DIP(di,di_size);
   1154   1.1       jtk 	fn = (mark_callback_t) arg;
   1155  1.25       riz 	switch (DIP(di,di_mode) & IFMT) {
   1156   1.1       jtk 	case IFLNK:
   1157  1.31  dholland 		if (filesize <= newsb->fs_maxsymlinklen) {
   1158  1.31  dholland 			break;
   1159  1.31  dholland 		}
   1160  1.31  dholland 		/* FALLTHROUGH */
   1161   1.1       jtk 	case IFDIR:
   1162   1.1       jtk 	case IFREG:
   1163  1.31  dholland 		map_inode_data_blocks(di, fn);
   1164   1.1       jtk 		break;
   1165   1.1       jtk 	}
   1166   1.1       jtk }
   1167   1.1       jtk /*
   1168   1.1       jtk  * Make a callback call, a la map_inode_data_blocks, for all data
   1169   1.1       jtk  *  blocks in the entire fs.  This is used only once, in
   1170   1.1       jtk  *  update_for_data_move, but it's out at top level because the complex
   1171   1.1       jtk  *  downward-funarg nesting that would otherwise result seems to give
   1172   1.1       jtk  *  gcc gastric distress.
   1173   1.1       jtk  */
   1174   1.1       jtk static void
   1175   1.1       jtk map_data_blocks(mark_callback_t fn, int ncg)
   1176   1.1       jtk {
   1177   1.1       jtk 	map_inodes(&dblk_callback, ncg, (void *) fn);
   1178   1.1       jtk }
   1179   1.1       jtk /*
   1180   1.1       jtk  * Initialize the blkmove array.
   1181   1.1       jtk  */
   1182   1.1       jtk static void
   1183   1.1       jtk blkmove_init(void)
   1184   1.1       jtk {
   1185   1.1       jtk 	int i;
   1186   1.1       jtk 
   1187   1.1       jtk 	blkmove = alloconce(oldsb->fs_size * sizeof(*blkmove), "blkmove");
   1188   1.1       jtk 	for (i = 0; i < oldsb->fs_size; i++)
   1189   1.1       jtk 		blkmove[i] = i;
   1190   1.1       jtk }
   1191   1.1       jtk /*
   1192   1.1       jtk  * Load the inodes off disk.  Allocates the structures and initializes
   1193   1.1       jtk  *  them - the inodes from disk, the flags to zero.
   1194   1.1       jtk  */
   1195   1.1       jtk static void
   1196   1.1       jtk loadinodes(void)
   1197   1.1       jtk {
   1198  1.25       riz 	int imax, ino, i, j;
   1199  1.25       riz 	struct ufs1_dinode *dp1 = NULL;
   1200  1.25       riz 	struct ufs2_dinode *dp2 = NULL;
   1201  1.28  dholland 
   1202  1.25       riz 	/* read inodes one fs block at a time and copy them */
   1203   1.1       jtk 
   1204  1.15       riz 	inodes = alloconce(oldsb->fs_ncg * oldsb->fs_ipg *
   1205  1.25       riz 	    sizeof(union dinode), "inodes");
   1206   1.1       jtk 	iflags = alloconce(oldsb->fs_ncg * oldsb->fs_ipg, "inode flags");
   1207  1.25       riz 	memset(iflags, 0, oldsb->fs_ncg * oldsb->fs_ipg);
   1208  1.28  dholland 
   1209  1.25       riz 	ibuf = nfmalloc(oldsb->fs_bsize,"inode block buf");
   1210  1.25       riz 	if (is_ufs2)
   1211  1.25       riz 		dp2 = (struct ufs2_dinode *)ibuf;
   1212  1.25       riz 	else
   1213  1.25       riz 		dp1 = (struct ufs1_dinode *)ibuf;
   1214  1.28  dholland 
   1215  1.25       riz 	for (ino = 0,imax = oldsb->fs_ipg * oldsb->fs_ncg; ino < imax; ) {
   1216  1.36  dholland 		readat(FFS_FSBTODB(oldsb, ino_to_fsba(oldsb, ino)), ibuf,
   1217  1.25       riz 		    oldsb->fs_bsize);
   1218  1.25       riz 
   1219  1.25       riz 		for (i = 0; i < oldsb->fs_inopb; i++) {
   1220  1.25       riz 			if (is_ufs2) {
   1221  1.25       riz 				if (needswap) {
   1222  1.25       riz 					ffs_dinode2_swap(&(dp2[i]), &(dp2[i]));
   1223  1.34  dholland 					for (j = 0; j < UFS_NDADDR + UFS_NIADDR; j++)
   1224  1.25       riz 						dp2[i].di_db[j] =
   1225  1.25       riz 						    bswap32(dp2[i].di_db[j]);
   1226  1.25       riz 				}
   1227  1.25       riz 				memcpy(&inodes[ino].dp2, &dp2[i],
   1228  1.32  christos 				    sizeof(inodes[ino].dp2));
   1229  1.25       riz 			} else {
   1230  1.25       riz 				if (needswap) {
   1231  1.25       riz 					ffs_dinode1_swap(&(dp1[i]), &(dp1[i]));
   1232  1.34  dholland 					for (j = 0; j < UFS_NDADDR + UFS_NIADDR; j++)
   1233  1.25       riz 						dp1[i].di_db[j] =
   1234  1.25       riz 						    bswap32(dp1[i].di_db[j]);
   1235  1.25       riz 				}
   1236  1.25       riz 				memcpy(&inodes[ino].dp1, &dp1[i],
   1237  1.32  christos 				    sizeof(inodes[ino].dp1));
   1238  1.25       riz 			}
   1239  1.25       riz 			    if (++ino > imax)
   1240  1.25       riz 				    errx(EXIT_FAILURE,
   1241  1.25       riz 					"Exceeded number of inodes");
   1242  1.25       riz 		}
   1243  1.25       riz 
   1244   1.1       jtk 	}
   1245   1.1       jtk }
   1246   1.1       jtk /*
   1247  1.24       wiz  * Report a file-system-too-full problem.
   1248   1.1       jtk  */
   1249  1.32  christos __dead static void
   1250   1.1       jtk toofull(void)
   1251   1.1       jtk {
   1252  1.32  christos 	errx(EXIT_FAILURE, "Sorry, would run out of data blocks");
   1253   1.1       jtk }
   1254   1.1       jtk /*
   1255   1.1       jtk  * Record a desire to move "n" frags from "from" to "to".
   1256   1.1       jtk  */
   1257   1.1       jtk static void
   1258   1.1       jtk mark_move(unsigned int from, unsigned int to, unsigned int n)
   1259   1.1       jtk {
   1260   1.1       jtk 	for (; n > 0; n--)
   1261   1.1       jtk 		blkmove[from++] = to++;
   1262   1.1       jtk }
   1263   1.1       jtk /* Helper function - evict n frags, starting with start (cg-relative).
   1264   1.1       jtk  * The free bitmap is scanned, unallocated frags are ignored, and
   1265   1.1       jtk  * each block of consecutive allocated frags is moved as a unit.
   1266   1.1       jtk  */
   1267   1.1       jtk static void
   1268   1.1       jtk fragmove(struct cg * cg, int base, unsigned int start, unsigned int n)
   1269   1.1       jtk {
   1270  1.30  dholland 	unsigned int i;
   1271   1.1       jtk 	int run;
   1272  1.26  dholland 
   1273   1.1       jtk 	run = 0;
   1274   1.1       jtk 	for (i = 0; i <= n; i++) {
   1275   1.1       jtk 		if ((i < n) && bit_is_clr(cg_blksfree(cg, 0), start + i)) {
   1276   1.1       jtk 			run++;
   1277   1.1       jtk 		} else {
   1278   1.1       jtk 			if (run > 0) {
   1279   1.1       jtk 				int off;
   1280   1.1       jtk 				off = find_freespace(run);
   1281   1.1       jtk 				if (off < 0)
   1282   1.1       jtk 					toofull();
   1283   1.1       jtk 				mark_move(base + start + i - run, off, run);
   1284   1.1       jtk 				set_bits(cg_blksfree(cg, 0), start + i - run,
   1285   1.1       jtk 				    run);
   1286   1.1       jtk 				clr_bits(cg_blksfree(cgs[dtog(oldsb, off)], 0),
   1287   1.1       jtk 				    dtogd(oldsb, off), run);
   1288   1.1       jtk 			}
   1289   1.1       jtk 			run = 0;
   1290   1.1       jtk 		}
   1291   1.1       jtk 	}
   1292   1.1       jtk }
   1293   1.1       jtk /*
   1294   1.1       jtk  * Evict all data blocks from the given cg, starting at minfrag (based
   1295   1.1       jtk  *  at the beginning of the cg), for length nfrag.  The eviction is
   1296   1.1       jtk  *  assumed to be entirely data-area; this should not be called with a
   1297   1.1       jtk  *  range overlapping the metadata structures in the cg.  It also
   1298   1.1       jtk  *  assumes minfrag points into the given cg; it will misbehave if this
   1299   1.1       jtk  *  is not true.
   1300   1.1       jtk  *
   1301   1.1       jtk  * See the comment header on find_freespace() for one possible bug
   1302   1.1       jtk  *  lurking here.
   1303   1.1       jtk  */
   1304   1.1       jtk static void
   1305  1.30  dholland evict_data(struct cg * cg, unsigned int minfrag, int nfrag)
   1306   1.1       jtk {
   1307  1.25       riz 	int base;	/* base of cg (in frags from beginning of fs) */
   1308   1.1       jtk 
   1309   1.1       jtk 	base = cgbase(oldsb, cg->cg_cgx);
   1310  1.25       riz 	/* Does the boundary fall in the middle of a block?  To avoid
   1311  1.25       riz 	 * breaking between frags allocated as consecutive, we always
   1312  1.25       riz 	 * evict the whole block in this case, though one could argue
   1313  1.25       riz 	 * we should check to see if the frag before or after the
   1314  1.25       riz 	 * break is unallocated. */
   1315   1.1       jtk 	if (minfrag % oldsb->fs_frag) {
   1316   1.1       jtk 		int n;
   1317   1.1       jtk 		n = minfrag % oldsb->fs_frag;
   1318   1.1       jtk 		minfrag -= n;
   1319   1.1       jtk 		nfrag += n;
   1320   1.1       jtk 	}
   1321  1.25       riz 	/* Do whole blocks.  If a block is wholly free, skip it; if
   1322  1.25       riz 	 * wholly allocated, move it in toto.  If neither, call
   1323  1.25       riz 	 * fragmove() to move the frags to new locations. */
   1324   1.1       jtk 	while (nfrag >= oldsb->fs_frag) {
   1325   1.1       jtk 		if (!blk_is_set(cg_blksfree(cg, 0), minfrag, oldsb->fs_frag)) {
   1326   1.1       jtk 			if (blk_is_clr(cg_blksfree(cg, 0), minfrag,
   1327   1.1       jtk 				oldsb->fs_frag)) {
   1328   1.1       jtk 				int off;
   1329   1.1       jtk 				off = find_freeblock();
   1330   1.1       jtk 				if (off < 0)
   1331   1.1       jtk 					toofull();
   1332   1.1       jtk 				mark_move(base + minfrag, off, oldsb->fs_frag);
   1333   1.1       jtk 				set_bits(cg_blksfree(cg, 0), minfrag,
   1334   1.1       jtk 				    oldsb->fs_frag);
   1335   1.1       jtk 				clr_bits(cg_blksfree(cgs[dtog(oldsb, off)], 0),
   1336   1.1       jtk 				    dtogd(oldsb, off), oldsb->fs_frag);
   1337   1.1       jtk 			} else {
   1338   1.1       jtk 				fragmove(cg, base, minfrag, oldsb->fs_frag);
   1339   1.1       jtk 			}
   1340   1.1       jtk 		}
   1341   1.1       jtk 		minfrag += oldsb->fs_frag;
   1342   1.1       jtk 		nfrag -= oldsb->fs_frag;
   1343   1.1       jtk 	}
   1344   1.1       jtk 	/* Clean up any sub-block amount left over. */
   1345   1.1       jtk 	if (nfrag) {
   1346   1.1       jtk 		fragmove(cg, base, minfrag, nfrag);
   1347   1.1       jtk 	}
   1348   1.1       jtk }
   1349   1.1       jtk /*
   1350   1.1       jtk  * Move all data blocks according to blkmove.  We have to be careful,
   1351   1.1       jtk  *  because we may be updating indirect blocks that will themselves be
   1352   1.2    martin  *  getting moved, or inode int32_t arrays that point to indirect
   1353   1.1       jtk  *  blocks that will be moved.  We call this before
   1354   1.1       jtk  *  update_for_data_move, and update_for_data_move does inodes first,
   1355   1.1       jtk  *  then indirect blocks in preorder, so as to make sure that the
   1356  1.24       wiz  *  file system is self-consistent at all points, for better crash
   1357   1.1       jtk  *  tolerance.  (We can get away with this only because all the writes
   1358   1.1       jtk  *  done by perform_data_move() are writing into space that's not used
   1359  1.24       wiz  *  by the old file system.)  If we crash, some things may point to the
   1360   1.1       jtk  *  old data and some to the new, but both copies are the same.  The
   1361   1.1       jtk  *  only wrong things should be csum info and free bitmaps, which fsck
   1362   1.1       jtk  *  is entirely capable of cleaning up.
   1363   1.1       jtk  *
   1364   1.1       jtk  * Since blkmove_init() initializes all blocks to move to their current
   1365   1.1       jtk  *  locations, we can have two blocks marked as wanting to move to the
   1366   1.1       jtk  *  same location, but only two and only when one of them is the one
   1367   1.1       jtk  *  that was already there.  So if blkmove[i]==i, we ignore that entry
   1368   1.1       jtk  *  entirely - for unallocated blocks, we don't want it (and may be
   1369   1.1       jtk  *  putting something else there), and for allocated blocks, we don't
   1370   1.1       jtk  *  want to copy it anywhere.
   1371   1.1       jtk  */
   1372   1.1       jtk static void
   1373   1.1       jtk perform_data_move(void)
   1374   1.1       jtk {
   1375   1.1       jtk 	int i;
   1376   1.1       jtk 	int run;
   1377   1.1       jtk 	int maxrun;
   1378   1.1       jtk 	char buf[65536];
   1379   1.1       jtk 
   1380   1.1       jtk 	maxrun = sizeof(buf) / newsb->fs_fsize;
   1381   1.1       jtk 	run = 0;
   1382   1.1       jtk 	for (i = 0; i < oldsb->fs_size; i++) {
   1383  1.30  dholland 		if ((blkmove[i] == (unsigned)i /*XXX cast*/) ||
   1384   1.1       jtk 		    (run >= maxrun) ||
   1385   1.1       jtk 		    ((run > 0) &&
   1386   1.1       jtk 			(blkmove[i] != blkmove[i - 1] + 1))) {
   1387   1.1       jtk 			if (run > 0) {
   1388  1.36  dholland 				readat(FFS_FSBTODB(oldsb, i - run), &buf[0],
   1389   1.1       jtk 				    run << oldsb->fs_fshift);
   1390  1.36  dholland 				writeat(FFS_FSBTODB(oldsb, blkmove[i - run]),
   1391   1.1       jtk 				    &buf[0], run << oldsb->fs_fshift);
   1392   1.1       jtk 			}
   1393   1.1       jtk 			run = 0;
   1394   1.1       jtk 		}
   1395  1.30  dholland 		if (blkmove[i] != (unsigned)i /*XXX cast*/)
   1396   1.1       jtk 			run++;
   1397   1.1       jtk 	}
   1398   1.1       jtk 	if (run > 0) {
   1399  1.36  dholland 		readat(FFS_FSBTODB(oldsb, i - run), &buf[0],
   1400   1.1       jtk 		    run << oldsb->fs_fshift);
   1401  1.36  dholland 		writeat(FFS_FSBTODB(oldsb, blkmove[i - run]), &buf[0],
   1402   1.1       jtk 		    run << oldsb->fs_fshift);
   1403   1.1       jtk 	}
   1404   1.1       jtk }
   1405   1.1       jtk /*
   1406   1.2    martin  * This modifies an array of int32_t, according to blkmove.  This is
   1407   1.1       jtk  *  used to update inode block arrays and indirect blocks to point to
   1408   1.1       jtk  *  the new locations of data blocks.
   1409   1.1       jtk  *
   1410   1.2    martin  * Return value is the number of int32_ts that needed updating; in
   1411   1.1       jtk  *  particular, the return value is zero iff nothing was modified.
   1412   1.1       jtk  */
   1413   1.1       jtk static int
   1414   1.2    martin movemap_blocks(int32_t * vec, int n)
   1415   1.1       jtk {
   1416   1.1       jtk 	int rv;
   1417  1.26  dholland 
   1418   1.1       jtk 	rv = 0;
   1419   1.1       jtk 	for (; n > 0; n--, vec++) {
   1420  1.30  dholland 		if (blkmove[*vec] != (unsigned)*vec /*XXX cast*/) {
   1421   1.1       jtk 			*vec = blkmove[*vec];
   1422   1.1       jtk 			rv++;
   1423   1.1       jtk 		}
   1424   1.1       jtk 	}
   1425   1.1       jtk 	return (rv);
   1426   1.1       jtk }
   1427   1.1       jtk static void
   1428  1.25       riz moveblocks_callback(union dinode * di, unsigned int inum, void *arg)
   1429   1.1       jtk {
   1430  1.30  dholland 	int32_t *dblkptr, *iblkptr;
   1431  1.26  dholland 
   1432  1.25       riz 	switch (DIP(di,di_mode) & IFMT) {
   1433   1.1       jtk 	case IFLNK:
   1434  1.30  dholland 		if ((off_t)DIP(di,di_size) <= oldsb->fs_maxsymlinklen) {
   1435  1.27  dholland 			break;
   1436  1.27  dholland 		}
   1437  1.27  dholland 		/* FALLTHROUGH */
   1438   1.1       jtk 	case IFDIR:
   1439   1.1       jtk 	case IFREG:
   1440  1.25       riz 		if (is_ufs2) {
   1441  1.30  dholland 			/* XXX these are not int32_t and this is WRONG! */
   1442  1.30  dholland 			dblkptr = (void *) &(di->dp2.di_db[0]);
   1443  1.30  dholland 			iblkptr = (void *) &(di->dp2.di_ib[0]);
   1444  1.25       riz 		} else {
   1445  1.25       riz 			dblkptr = &(di->dp1.di_db[0]);
   1446  1.25       riz 			iblkptr = &(di->dp1.di_ib[0]);
   1447  1.25       riz 		}
   1448  1.26  dholland 		/*
   1449  1.26  dholland 		 * Don't || these two calls; we need their
   1450  1.26  dholland 		 * side-effects.
   1451  1.26  dholland 		 */
   1452  1.34  dholland 		if (movemap_blocks(dblkptr, UFS_NDADDR)) {
   1453  1.28  dholland 			iflags[inum] |= IF_DIRTY;
   1454  1.28  dholland 		}
   1455  1.34  dholland 		if (movemap_blocks(iblkptr, UFS_NIADDR)) {
   1456  1.28  dholland 			iflags[inum] |= IF_DIRTY;
   1457  1.28  dholland 		}
   1458   1.1       jtk 		break;
   1459   1.1       jtk 	}
   1460   1.1       jtk }
   1461   1.1       jtk 
   1462   1.1       jtk static void
   1463  1.30  dholland moveindir_callback(off_t off, unsigned int nfrag, unsigned int nbytes,
   1464  1.15       riz 		   int kind)
   1465   1.1       jtk {
   1466  1.30  dholland 	unsigned int i;
   1467  1.26  dholland 
   1468   1.1       jtk 	if (kind == MDB_INDIR_PRE) {
   1469   1.2    martin 		int32_t blk[howmany(MAXBSIZE, sizeof(int32_t))];
   1470  1.36  dholland 		readat(FFS_FSBTODB(oldsb, off), &blk[0], oldsb->fs_bsize);
   1471  1.25       riz 		if (needswap)
   1472  1.25       riz 			for (i = 0; i < howmany(MAXBSIZE, sizeof(int32_t)); i++)
   1473  1.25       riz 				blk[i] = bswap32(blk[i]);
   1474  1.35  dholland 		if (movemap_blocks(&blk[0], FFS_NINDIR(oldsb))) {
   1475  1.25       riz 			if (needswap)
   1476  1.25       riz 				for (i = 0; i < howmany(MAXBSIZE,
   1477  1.25       riz 					sizeof(int32_t)); i++)
   1478  1.25       riz 					blk[i] = bswap32(blk[i]);
   1479  1.36  dholland 			writeat(FFS_FSBTODB(oldsb, off), &blk[0], oldsb->fs_bsize);
   1480   1.1       jtk 		}
   1481   1.1       jtk 	}
   1482   1.1       jtk }
   1483   1.1       jtk /*
   1484   1.1       jtk  * Update all inode data arrays and indirect blocks to point to the new
   1485   1.1       jtk  *  locations of data blocks.  See the comment header on
   1486   1.1       jtk  *  perform_data_move for some ordering considerations.
   1487   1.1       jtk  */
   1488   1.1       jtk static void
   1489   1.1       jtk update_for_data_move(void)
   1490   1.1       jtk {
   1491   1.1       jtk 	map_inodes(&moveblocks_callback, oldsb->fs_ncg, NULL);
   1492   1.1       jtk 	map_data_blocks(&moveindir_callback, oldsb->fs_ncg);
   1493   1.1       jtk }
   1494   1.1       jtk /*
   1495   1.1       jtk  * Initialize the inomove array.
   1496   1.1       jtk  */
   1497   1.1       jtk static void
   1498   1.1       jtk inomove_init(void)
   1499   1.1       jtk {
   1500   1.1       jtk 	int i;
   1501   1.1       jtk 
   1502   1.1       jtk 	inomove = alloconce(oldsb->fs_ipg * oldsb->fs_ncg * sizeof(*inomove),
   1503   1.1       jtk                             "inomove");
   1504   1.1       jtk 	for (i = (oldsb->fs_ipg * oldsb->fs_ncg) - 1; i >= 0; i--)
   1505   1.1       jtk 		inomove[i] = i;
   1506   1.1       jtk }
   1507   1.1       jtk /*
   1508   1.1       jtk  * Flush all dirtied inodes to disk.  Scans the inode flags array; for
   1509   1.1       jtk  *  each dirty inode, it sets the BDIRTY bit on the first inode in the
   1510   1.1       jtk  *  block containing the dirty inode.  Then it scans by blocks, and for
   1511   1.1       jtk  *  each marked block, writes it.
   1512   1.1       jtk  */
   1513   1.1       jtk static void
   1514   1.1       jtk flush_inodes(void)
   1515   1.1       jtk {
   1516  1.25       riz 	int i, j, k, na, ni, m;
   1517  1.25       riz 	struct ufs1_dinode *dp1 = NULL;
   1518  1.25       riz 	struct ufs2_dinode *dp2 = NULL;
   1519   1.1       jtk 
   1520  1.34  dholland 	na = UFS_NDADDR + UFS_NIADDR;
   1521   1.1       jtk 	ni = newsb->fs_ipg * newsb->fs_ncg;
   1522  1.35  dholland 	m = FFS_INOPB(newsb) - 1;
   1523   1.1       jtk 	for (i = 0; i < ni; i++) {
   1524   1.1       jtk 		if (iflags[i] & IF_DIRTY) {
   1525   1.1       jtk 			iflags[i & ~m] |= IF_BDIRTY;
   1526   1.1       jtk 		}
   1527   1.1       jtk 	}
   1528   1.1       jtk 	m++;
   1529  1.25       riz 
   1530  1.25       riz 	if (is_ufs2)
   1531  1.25       riz 		dp2 = (struct ufs2_dinode *)ibuf;
   1532  1.25       riz 	else
   1533  1.25       riz 		dp1 = (struct ufs1_dinode *)ibuf;
   1534  1.28  dholland 
   1535   1.1       jtk 	for (i = 0; i < ni; i += m) {
   1536   1.1       jtk 		if (iflags[i] & IF_BDIRTY) {
   1537  1.25       riz 			if (is_ufs2)
   1538  1.25       riz 				for (j = 0; j < m; j++) {
   1539  1.25       riz 					dp2[j] = inodes[i + j].dp2;
   1540  1.25       riz 					if (needswap) {
   1541  1.25       riz 						for (k = 0; k < na; k++)
   1542  1.25       riz 							dp2[j].di_db[k]=
   1543  1.25       riz 							    bswap32(dp2[j].di_db[k]);
   1544  1.25       riz 						ffs_dinode2_swap(&dp2[j],
   1545  1.25       riz 						    &dp2[j]);
   1546  1.25       riz 					}
   1547  1.25       riz 				}
   1548  1.25       riz 			else
   1549  1.25       riz 				for (j = 0; j < m; j++) {
   1550  1.25       riz 					dp1[j] = inodes[i + j].dp1;
   1551  1.25       riz 					if (needswap) {
   1552  1.25       riz 						for (k = 0; k < na; k++)
   1553  1.25       riz 							dp1[j].di_db[k]=
   1554  1.25       riz 							    bswap32(dp1[j].di_db[k]);
   1555  1.25       riz 						ffs_dinode1_swap(&dp1[j],
   1556  1.25       riz 						    &dp1[j]);
   1557  1.25       riz 					}
   1558  1.25       riz 				}
   1559  1.28  dholland 
   1560  1.36  dholland 			writeat(FFS_FSBTODB(newsb, ino_to_fsba(newsb, i)),
   1561  1.25       riz 			    ibuf, newsb->fs_bsize);
   1562   1.1       jtk 		}
   1563   1.1       jtk 	}
   1564   1.1       jtk }
   1565   1.1       jtk /*
   1566   1.1       jtk  * Evict all inodes from the specified cg.  shrink() already checked
   1567   1.1       jtk  *  that there were enough free inodes, so the no-free-inodes check is
   1568  1.24       wiz  *  a can't-happen.  If it does trip, the file system should be in good
   1569   1.1       jtk  *  enough shape for fsck to fix; see the comment on perform_data_move
   1570   1.1       jtk  *  for the considerations in question.
   1571   1.1       jtk  */
   1572   1.1       jtk static void
   1573   1.1       jtk evict_inodes(struct cg * cg)
   1574   1.1       jtk {
   1575   1.1       jtk 	int inum;
   1576   1.1       jtk 	int i;
   1577   1.1       jtk 	int fi;
   1578   1.1       jtk 
   1579   1.1       jtk 	inum = newsb->fs_ipg * cg->cg_cgx;
   1580   1.1       jtk 	for (i = 0; i < newsb->fs_ipg; i++, inum++) {
   1581  1.25       riz 		if (DIP(inodes + inum,di_mode) != 0) {
   1582   1.1       jtk 			fi = find_freeinode();
   1583  1.32  christos 			if (fi < 0)
   1584  1.32  christos 				errx(EXIT_FAILURE, "Sorry, inodes evaporated - "
   1585  1.32  christos 				    "file system probably needs fsck");
   1586   1.1       jtk 			inomove[inum] = fi;
   1587   1.1       jtk 			clr_bits(cg_inosused(cg, 0), i, 1);
   1588   1.1       jtk 			set_bits(cg_inosused(cgs[ino_to_cg(newsb, fi)], 0),
   1589   1.1       jtk 			    fi % newsb->fs_ipg, 1);
   1590   1.1       jtk 		}
   1591   1.1       jtk 	}
   1592   1.1       jtk }
   1593   1.1       jtk /*
   1594   1.1       jtk  * Move inodes from old locations to new.  Does not actually write
   1595   1.1       jtk  *  anything to disk; just copies in-core and sets dirty bits.
   1596   1.1       jtk  *
   1597   1.1       jtk  * We have to be careful here for reasons similar to those mentioned in
   1598   1.1       jtk  *  the comment header on perform_data_move, above: for the sake of
   1599   1.1       jtk  *  crash tolerance, we want to make sure everything is present at both
   1600   1.1       jtk  *  old and new locations before we update pointers.  So we call this
   1601   1.1       jtk  *  first, then flush_inodes() to get them out on disk, then update
   1602   1.1       jtk  *  directories to match.
   1603   1.1       jtk  */
   1604   1.1       jtk static void
   1605   1.1       jtk perform_inode_move(void)
   1606   1.1       jtk {
   1607  1.30  dholland 	unsigned int i;
   1608  1.30  dholland 	unsigned int ni;
   1609   1.1       jtk 
   1610   1.1       jtk 	ni = oldsb->fs_ipg * oldsb->fs_ncg;
   1611   1.1       jtk 	for (i = 0; i < ni; i++) {
   1612   1.1       jtk 		if (inomove[i] != i) {
   1613   1.1       jtk 			inodes[inomove[i]] = inodes[i];
   1614   1.1       jtk 			iflags[inomove[i]] = iflags[i] | IF_DIRTY;
   1615   1.1       jtk 		}
   1616   1.1       jtk 	}
   1617   1.1       jtk }
   1618   1.1       jtk /*
   1619   1.1       jtk  * Update the directory contained in the nb bytes at buf, to point to
   1620   1.1       jtk  *  inodes' new locations.
   1621   1.1       jtk  */
   1622   1.1       jtk static int
   1623   1.1       jtk update_dirents(char *buf, int nb)
   1624   1.1       jtk {
   1625   1.1       jtk 	int rv;
   1626   1.1       jtk #define d ((struct direct *)buf)
   1627  1.25       riz #define s32(x) (needswap?bswap32((x)):(x))
   1628  1.25       riz #define s16(x) (needswap?bswap16((x)):(x))
   1629  1.28  dholland 
   1630   1.1       jtk 	rv = 0;
   1631   1.1       jtk 	while (nb > 0) {
   1632  1.25       riz 		if (inomove[s32(d->d_ino)] != s32(d->d_ino)) {
   1633   1.1       jtk 			rv++;
   1634  1.25       riz 			d->d_ino = s32(inomove[s32(d->d_ino)]);
   1635   1.1       jtk 		}
   1636  1.25       riz 		nb -= s16(d->d_reclen);
   1637  1.25       riz 		buf += s16(d->d_reclen);
   1638   1.1       jtk 	}
   1639   1.1       jtk 	return (rv);
   1640   1.1       jtk #undef d
   1641  1.25       riz #undef s32
   1642  1.25       riz #undef s16
   1643   1.1       jtk }
   1644   1.1       jtk /*
   1645   1.1       jtk  * Callback function for map_inode_data_blocks, for updating a
   1646   1.1       jtk  *  directory to point to new inode locations.
   1647   1.1       jtk  */
   1648   1.1       jtk static void
   1649  1.30  dholland update_dir_data(off_t bn, unsigned int size, unsigned int nb, int kind)
   1650   1.1       jtk {
   1651   1.1       jtk 	if (kind == MDB_DATA) {
   1652   1.1       jtk 		union {
   1653   1.1       jtk 			struct direct d;
   1654   1.1       jtk 			char ch[MAXBSIZE];
   1655   1.1       jtk 		}     buf;
   1656  1.36  dholland 		readat(FFS_FSBTODB(oldsb, bn), &buf, size << oldsb->fs_fshift);
   1657   1.1       jtk 		if (update_dirents((char *) &buf, nb)) {
   1658  1.36  dholland 			writeat(FFS_FSBTODB(oldsb, bn), &buf,
   1659   1.1       jtk 			    size << oldsb->fs_fshift);
   1660   1.1       jtk 		}
   1661   1.1       jtk 	}
   1662   1.1       jtk }
   1663   1.1       jtk static void
   1664  1.25       riz dirmove_callback(union dinode * di, unsigned int inum, void *arg)
   1665   1.1       jtk {
   1666  1.25       riz 	switch (DIP(di,di_mode) & IFMT) {
   1667   1.1       jtk 	case IFDIR:
   1668   1.1       jtk 		map_inode_data_blocks(di, &update_dir_data);
   1669   1.1       jtk 		break;
   1670   1.1       jtk 	}
   1671   1.1       jtk }
   1672   1.1       jtk /*
   1673   1.1       jtk  * Update directory entries to point to new inode locations.
   1674   1.1       jtk  */
   1675   1.1       jtk static void
   1676   1.1       jtk update_for_inode_move(void)
   1677   1.1       jtk {
   1678   1.1       jtk 	map_inodes(&dirmove_callback, newsb->fs_ncg, NULL);
   1679   1.1       jtk }
   1680   1.1       jtk /*
   1681  1.24       wiz  * Shrink the file system.
   1682   1.1       jtk  */
   1683   1.1       jtk static void
   1684   1.1       jtk shrink(void)
   1685   1.1       jtk {
   1686   1.1       jtk 	int i;
   1687   1.1       jtk 
   1688   1.1       jtk 	/* Load the inodes off disk - we'll need 'em. */
   1689   1.1       jtk 	loadinodes();
   1690   1.1       jtk 	/* Update the timestamp. */
   1691   1.1       jtk 	newsb->fs_time = timestamp();
   1692   1.1       jtk 	/* Update the size figures. */
   1693  1.36  dholland 	newsb->fs_size = FFS_DBTOFSB(newsb, newsize);
   1694  1.25       riz 	if (is_ufs2)
   1695  1.25       riz 		newsb->fs_ncg = howmany(newsb->fs_size, newsb->fs_fpg);
   1696  1.25       riz 	else {
   1697  1.25       riz 		newsb->fs_old_ncyl = howmany(newsb->fs_size * NSPF(newsb),
   1698  1.25       riz 		    newsb->fs_old_spc);
   1699  1.25       riz 		newsb->fs_ncg = howmany(newsb->fs_old_ncyl, newsb->fs_old_cpg);
   1700  1.25       riz 	}
   1701   1.1       jtk 	/* Does the (new) last cg end before the end of its inode area?  See
   1702   1.1       jtk 	 * the similar code in grow() for more on this. */
   1703   1.1       jtk 	if (cgdmin(newsb, newsb->fs_ncg - 1) > newsb->fs_size) {
   1704   1.1       jtk 		newsb->fs_ncg--;
   1705  1.25       riz 		if (is_ufs2 == 0) {
   1706  1.25       riz 			newsb->fs_old_ncyl = newsb->fs_ncg * newsb->fs_old_cpg;
   1707  1.25       riz 			newsb->fs_size = (newsb->fs_old_ncyl *
   1708  1.25       riz 			    newsb->fs_old_spc) / NSPF(newsb);
   1709  1.25       riz 		} else
   1710  1.25       riz 			newsb->fs_size = newsb->fs_ncg * newsb->fs_fpg;
   1711  1.28  dholland 
   1712   1.1       jtk 		printf("Warning: last cylinder group is too small;\n");
   1713   1.1       jtk 		printf("    dropping it.  New size = %lu.\n",
   1714  1.36  dholland 		    (unsigned long int) FFS_FSBTODB(newsb, newsb->fs_size));
   1715   1.1       jtk 	}
   1716   1.1       jtk 	/* Let's make sure we're not being shrunk into oblivion. */
   1717  1.32  christos 	if (newsb->fs_ncg < 1)
   1718  1.32  christos 		errx(EXIT_FAILURE, "Size too small - file system would "
   1719  1.32  christos 		    "have no cylinders");
   1720   1.1       jtk 	/* Initialize for block motion. */
   1721   1.1       jtk 	blkmove_init();
   1722   1.1       jtk 	/* Update csum size, then fix up for the new size */
   1723  1.37  dholland 	newsb->fs_cssize = ffs_fragroundup(newsb,
   1724   1.1       jtk 	    newsb->fs_ncg * sizeof(struct csum));
   1725   1.1       jtk 	csum_fixup();
   1726   1.8       snj 	/* Evict data from any cgs being wholly eliminated */
   1727   1.1       jtk 	for (i = newsb->fs_ncg; i < oldsb->fs_ncg; i++) {
   1728   1.1       jtk 		int base;
   1729   1.1       jtk 		int dlow;
   1730   1.1       jtk 		int dhigh;
   1731   1.1       jtk 		int dmax;
   1732   1.1       jtk 		base = cgbase(oldsb, i);
   1733   1.1       jtk 		dlow = cgsblock(oldsb, i) - base;
   1734   1.1       jtk 		dhigh = cgdmin(oldsb, i) - base;
   1735   1.1       jtk 		dmax = oldsb->fs_size - base;
   1736   1.1       jtk 		if (dmax > cgs[i]->cg_ndblk)
   1737   1.1       jtk 			dmax = cgs[i]->cg_ndblk;
   1738   1.1       jtk 		evict_data(cgs[i], 0, dlow);
   1739   1.1       jtk 		evict_data(cgs[i], dhigh, dmax - dhigh);
   1740   1.1       jtk 		newsb->fs_cstotal.cs_ndir -= cgs[i]->cg_cs.cs_ndir;
   1741   1.1       jtk 		newsb->fs_cstotal.cs_nifree -= cgs[i]->cg_cs.cs_nifree;
   1742   1.1       jtk 		newsb->fs_cstotal.cs_nffree -= cgs[i]->cg_cs.cs_nffree;
   1743   1.1       jtk 		newsb->fs_cstotal.cs_nbfree -= cgs[i]->cg_cs.cs_nbfree;
   1744   1.1       jtk 	}
   1745   1.1       jtk 	/* Update the new last cg. */
   1746   1.1       jtk 	cgs[newsb->fs_ncg - 1]->cg_ndblk = newsb->fs_size -
   1747   1.1       jtk 	    ((newsb->fs_ncg - 1) * newsb->fs_fpg);
   1748   1.1       jtk 	/* Is the new last cg partial?  If so, evict any data from the part
   1749   1.1       jtk 	 * being shrunken away. */
   1750   1.1       jtk 	if (newsb->fs_size % newsb->fs_fpg) {
   1751   1.1       jtk 		struct cg *cg;
   1752   1.1       jtk 		int oldcgsize;
   1753   1.1       jtk 		int newcgsize;
   1754   1.1       jtk 		cg = cgs[newsb->fs_ncg - 1];
   1755   1.1       jtk 		newcgsize = newsb->fs_size % newsb->fs_fpg;
   1756  1.15       riz 		oldcgsize = oldsb->fs_size - ((newsb->fs_ncg - 1) &
   1757  1.15       riz 		    oldsb->fs_fpg);
   1758   1.1       jtk 		if (oldcgsize > oldsb->fs_fpg)
   1759   1.1       jtk 			oldcgsize = oldsb->fs_fpg;
   1760   1.1       jtk 		evict_data(cg, newcgsize, oldcgsize - newcgsize);
   1761   1.1       jtk 		clr_bits(cg_blksfree(cg, 0), newcgsize, oldcgsize - newcgsize);
   1762   1.1       jtk 	}
   1763  1.25       riz 	/* Find out whether we would run out of inodes.  (Note we
   1764  1.25       riz 	 * haven't actually done anything to the file system yet; all
   1765  1.25       riz 	 * those evict_data calls just update blkmove.) */
   1766   1.1       jtk 	{
   1767   1.1       jtk 		int slop;
   1768   1.1       jtk 		slop = 0;
   1769   1.1       jtk 		for (i = 0; i < newsb->fs_ncg; i++)
   1770   1.1       jtk 			slop += cgs[i]->cg_cs.cs_nifree;
   1771   1.1       jtk 		for (; i < oldsb->fs_ncg; i++)
   1772   1.1       jtk 			slop -= oldsb->fs_ipg - cgs[i]->cg_cs.cs_nifree;
   1773  1.32  christos 		if (slop < 0)
   1774  1.32  christos 			errx(EXIT_FAILURE, "Sorry, would run out of inodes");
   1775   1.1       jtk 	}
   1776  1.25       riz 	/* Copy data, then update pointers to data.  See the comment
   1777  1.25       riz 	 * header on perform_data_move for ordering considerations. */
   1778   1.1       jtk 	perform_data_move();
   1779   1.1       jtk 	update_for_data_move();
   1780  1.25       riz 	/* Now do inodes.  Initialize, evict, move, update - see the
   1781  1.25       riz 	 * comment header on perform_inode_move. */
   1782   1.1       jtk 	inomove_init();
   1783   1.1       jtk 	for (i = newsb->fs_ncg; i < oldsb->fs_ncg; i++)
   1784   1.1       jtk 		evict_inodes(cgs[i]);
   1785   1.1       jtk 	perform_inode_move();
   1786   1.1       jtk 	flush_inodes();
   1787   1.1       jtk 	update_for_inode_move();
   1788   1.1       jtk 	/* Recompute all the bitmaps; most of them probably need it anyway,
   1789   1.1       jtk 	 * the rest are just paranoia and not wanting to have to bother
   1790   1.1       jtk 	 * keeping track of exactly which ones require it. */
   1791   1.1       jtk 	for (i = 0; i < newsb->fs_ncg; i++)
   1792   1.1       jtk 		cgflags[i] |= CGF_DIRTY | CGF_BLKMAPS | CGF_INOMAPS;
   1793  1.14       riz 	/* Update the cg_old_ncyl value for the last cylinder. */
   1794  1.20    mhitch 	if ((newsb->fs_old_flags & FS_FLAGS_UPDATED) == 0)
   1795  1.14       riz 		cgs[newsb->fs_ncg - 1]->cg_old_ncyl =
   1796  1.20    mhitch 		    newsb->fs_old_ncyl % newsb->fs_old_cpg;
   1797   1.1       jtk 	/* Make fs_dsize match the new reality. */
   1798   1.1       jtk 	recompute_fs_dsize();
   1799   1.1       jtk }
   1800   1.1       jtk /*
   1801   1.1       jtk  * Recompute the block totals, block cluster summaries, and rotational
   1802   1.1       jtk  *  position summaries, for a given cg (specified by number), based on
   1803   1.1       jtk  *  its free-frag bitmap (cg_blksfree()[]).
   1804   1.1       jtk  */
   1805   1.1       jtk static void
   1806   1.1       jtk rescan_blkmaps(int cgn)
   1807   1.1       jtk {
   1808   1.1       jtk 	struct cg *cg;
   1809   1.1       jtk 	int f;
   1810   1.1       jtk 	int b;
   1811   1.1       jtk 	int blkfree;
   1812   1.1       jtk 	int blkrun;
   1813   1.1       jtk 	int fragrun;
   1814   1.1       jtk 	int fwb;
   1815   1.1       jtk 
   1816   1.1       jtk 	cg = cgs[cgn];
   1817   1.1       jtk 	/* Subtract off the current totals from the sb's summary info */
   1818   1.1       jtk 	newsb->fs_cstotal.cs_nffree -= cg->cg_cs.cs_nffree;
   1819   1.1       jtk 	newsb->fs_cstotal.cs_nbfree -= cg->cg_cs.cs_nbfree;
   1820   1.1       jtk 	/* Clear counters and bitmaps. */
   1821   1.1       jtk 	cg->cg_cs.cs_nffree = 0;
   1822   1.1       jtk 	cg->cg_cs.cs_nbfree = 0;
   1823  1.25       riz 	memset(&cg->cg_frsum[0], 0, MAXFRAG * sizeof(cg->cg_frsum[0]));
   1824  1.25       riz 	memset(&old_cg_blktot(cg, 0)[0], 0,
   1825  1.15       riz 	    newsb->fs_old_cpg * sizeof(old_cg_blktot(cg, 0)[0]));
   1826  1.25       riz 	memset(&old_cg_blks(newsb, cg, 0, 0)[0], 0,
   1827   1.4  christos 	    newsb->fs_old_cpg * newsb->fs_old_nrpos *
   1828  1.15       riz 	    sizeof(old_cg_blks(newsb, cg, 0, 0)[0]));
   1829   1.1       jtk 	if (newsb->fs_contigsumsize > 0) {
   1830   1.1       jtk 		cg->cg_nclusterblks = cg->cg_ndblk / newsb->fs_frag;
   1831  1.25       riz 		memset(&cg_clustersum(cg, 0)[1], 0,
   1832   1.1       jtk 		    newsb->fs_contigsumsize *
   1833   1.1       jtk 		    sizeof(cg_clustersum(cg, 0)[1]));
   1834  1.25       riz 		if (is_ufs2)
   1835  1.25       riz 			memset(&cg_clustersfree(cg, 0)[0], 0,
   1836  1.25       riz 			    howmany(newsb->fs_fpg / NSPB(newsb), NBBY));
   1837  1.25       riz 		else
   1838  1.25       riz 			memset(&cg_clustersfree(cg, 0)[0], 0,
   1839  1.25       riz 			    howmany((newsb->fs_old_cpg * newsb->fs_old_spc) /
   1840  1.25       riz 				NSPB(newsb), NBBY));
   1841  1.25       riz 	}
   1842  1.25       riz 	/* Scan the free-frag bitmap.  Runs of free frags are kept
   1843  1.25       riz 	 * track of with fragrun, and recorded into cg_frsum[] and
   1844  1.25       riz 	 * cg_cs.cs_nffree; on each block boundary, entire free blocks
   1845  1.25       riz 	 * are recorded as well. */
   1846   1.1       jtk 	blkfree = 1;
   1847   1.1       jtk 	blkrun = 0;
   1848   1.1       jtk 	fragrun = 0;
   1849   1.1       jtk 	f = 0;
   1850   1.1       jtk 	b = 0;
   1851   1.1       jtk 	fwb = 0;
   1852   1.1       jtk 	while (f < cg->cg_ndblk) {
   1853   1.1       jtk 		if (bit_is_set(cg_blksfree(cg, 0), f)) {
   1854   1.1       jtk 			fragrun++;
   1855   1.1       jtk 		} else {
   1856   1.1       jtk 			blkfree = 0;
   1857   1.1       jtk 			if (fragrun > 0) {
   1858   1.1       jtk 				cg->cg_frsum[fragrun]++;
   1859   1.1       jtk 				cg->cg_cs.cs_nffree += fragrun;
   1860   1.1       jtk 			}
   1861   1.1       jtk 			fragrun = 0;
   1862   1.1       jtk 		}
   1863   1.1       jtk 		f++;
   1864   1.1       jtk 		fwb++;
   1865   1.1       jtk 		if (fwb >= newsb->fs_frag) {
   1866   1.1       jtk 			if (blkfree) {
   1867   1.1       jtk 				cg->cg_cs.cs_nbfree++;
   1868   1.1       jtk 				if (newsb->fs_contigsumsize > 0)
   1869   1.1       jtk 					set_bits(cg_clustersfree(cg, 0), b, 1);
   1870  1.25       riz 				if (is_ufs2 == 0) {
   1871  1.25       riz 					old_cg_blktot(cg, 0)[
   1872  1.25       riz 						old_cbtocylno(newsb,
   1873  1.25       riz 						    f - newsb->fs_frag)]++;
   1874  1.25       riz 					old_cg_blks(newsb, cg,
   1875  1.25       riz 					    old_cbtocylno(newsb,
   1876  1.25       riz 						f - newsb->fs_frag),
   1877  1.25       riz 					    0)[old_cbtorpos(newsb,
   1878  1.25       riz 						    f - newsb->fs_frag)]++;
   1879  1.25       riz 				}
   1880   1.1       jtk 				blkrun++;
   1881   1.1       jtk 			} else {
   1882   1.1       jtk 				if (fragrun > 0) {
   1883   1.1       jtk 					cg->cg_frsum[fragrun]++;
   1884   1.1       jtk 					cg->cg_cs.cs_nffree += fragrun;
   1885   1.1       jtk 				}
   1886   1.1       jtk 				if (newsb->fs_contigsumsize > 0) {
   1887   1.1       jtk 					if (blkrun > 0) {
   1888  1.15       riz 						cg_clustersum(cg, 0)[(blkrun
   1889  1.15       riz 						    > newsb->fs_contigsumsize)
   1890  1.15       riz 						    ? newsb->fs_contigsumsize
   1891  1.15       riz 						    : blkrun]++;
   1892   1.1       jtk 					}
   1893   1.1       jtk 				}
   1894   1.1       jtk 				blkrun = 0;
   1895   1.1       jtk 			}
   1896   1.1       jtk 			fwb = 0;
   1897   1.1       jtk 			b++;
   1898   1.1       jtk 			blkfree = 1;
   1899   1.1       jtk 			fragrun = 0;
   1900   1.1       jtk 		}
   1901   1.1       jtk 	}
   1902   1.1       jtk 	if (fragrun > 0) {
   1903   1.1       jtk 		cg->cg_frsum[fragrun]++;
   1904   1.1       jtk 		cg->cg_cs.cs_nffree += fragrun;
   1905   1.1       jtk 	}
   1906   1.1       jtk 	if ((blkrun > 0) && (newsb->fs_contigsumsize > 0)) {
   1907   1.1       jtk 		cg_clustersum(cg, 0)[(blkrun > newsb->fs_contigsumsize) ?
   1908   1.1       jtk 		    newsb->fs_contigsumsize : blkrun]++;
   1909   1.1       jtk 	}
   1910   1.1       jtk 	/*
   1911   1.1       jtk          * Put the updated summary info back into csums, and add it
   1912   1.1       jtk          * back into the sb's summary info.  Then mark the cg dirty.
   1913   1.1       jtk          */
   1914   1.1       jtk 	csums[cgn] = cg->cg_cs;
   1915   1.1       jtk 	newsb->fs_cstotal.cs_nffree += cg->cg_cs.cs_nffree;
   1916   1.1       jtk 	newsb->fs_cstotal.cs_nbfree += cg->cg_cs.cs_nbfree;
   1917   1.1       jtk 	cgflags[cgn] |= CGF_DIRTY;
   1918   1.1       jtk }
   1919   1.1       jtk /*
   1920   1.1       jtk  * Recompute the cg_inosused()[] bitmap, and the cs_nifree and cs_ndir
   1921   1.1       jtk  *  values, for a cg, based on the in-core inodes for that cg.
   1922   1.1       jtk  */
   1923   1.1       jtk static void
   1924   1.1       jtk rescan_inomaps(int cgn)
   1925   1.1       jtk {
   1926   1.1       jtk 	struct cg *cg;
   1927   1.1       jtk 	int inum;
   1928   1.1       jtk 	int iwc;
   1929   1.1       jtk 
   1930   1.1       jtk 	cg = cgs[cgn];
   1931   1.1       jtk 	newsb->fs_cstotal.cs_ndir -= cg->cg_cs.cs_ndir;
   1932   1.1       jtk 	newsb->fs_cstotal.cs_nifree -= cg->cg_cs.cs_nifree;
   1933   1.1       jtk 	cg->cg_cs.cs_ndir = 0;
   1934   1.1       jtk 	cg->cg_cs.cs_nifree = 0;
   1935  1.25       riz 	memset(&cg_inosused(cg, 0)[0], 0, howmany(newsb->fs_ipg, NBBY));
   1936   1.1       jtk 	inum = cgn * newsb->fs_ipg;
   1937   1.1       jtk 	if (cgn == 0) {
   1938   1.1       jtk 		set_bits(cg_inosused(cg, 0), 0, 2);
   1939   1.1       jtk 		iwc = 2;
   1940   1.1       jtk 		inum += 2;
   1941   1.1       jtk 	} else {
   1942   1.1       jtk 		iwc = 0;
   1943   1.1       jtk 	}
   1944   1.1       jtk 	for (; iwc < newsb->fs_ipg; iwc++, inum++) {
   1945  1.25       riz 		switch (DIP(inodes + inum, di_mode) & IFMT) {
   1946   1.1       jtk 		case 0:
   1947   1.1       jtk 			cg->cg_cs.cs_nifree++;
   1948   1.1       jtk 			break;
   1949   1.1       jtk 		case IFDIR:
   1950   1.1       jtk 			cg->cg_cs.cs_ndir++;
   1951  1.31  dholland 			/* FALLTHROUGH */
   1952   1.1       jtk 		default:
   1953   1.1       jtk 			set_bits(cg_inosused(cg, 0), iwc, 1);
   1954   1.1       jtk 			break;
   1955   1.1       jtk 		}
   1956   1.1       jtk 	}
   1957   1.1       jtk 	csums[cgn] = cg->cg_cs;
   1958   1.1       jtk 	newsb->fs_cstotal.cs_ndir += cg->cg_cs.cs_ndir;
   1959   1.1       jtk 	newsb->fs_cstotal.cs_nifree += cg->cg_cs.cs_nifree;
   1960   1.1       jtk 	cgflags[cgn] |= CGF_DIRTY;
   1961   1.1       jtk }
   1962   1.1       jtk /*
   1963   1.1       jtk  * Flush cgs to disk, recomputing anything they're marked as needing.
   1964   1.1       jtk  */
   1965   1.1       jtk static void
   1966   1.1       jtk flush_cgs(void)
   1967   1.1       jtk {
   1968   1.1       jtk 	int i;
   1969   1.1       jtk 
   1970   1.1       jtk 	for (i = 0; i < newsb->fs_ncg; i++) {
   1971   1.1       jtk 		if (cgflags[i] & CGF_BLKMAPS) {
   1972   1.1       jtk 			rescan_blkmaps(i);
   1973   1.1       jtk 		}
   1974   1.1       jtk 		if (cgflags[i] & CGF_INOMAPS) {
   1975   1.1       jtk 			rescan_inomaps(i);
   1976   1.1       jtk 		}
   1977   1.1       jtk 		if (cgflags[i] & CGF_DIRTY) {
   1978   1.1       jtk 			cgs[i]->cg_rotor = 0;
   1979   1.1       jtk 			cgs[i]->cg_frotor = 0;
   1980   1.1       jtk 			cgs[i]->cg_irotor = 0;
   1981  1.25       riz 			if (needswap)
   1982  1.25       riz 				ffs_cg_swap(cgs[i],cgs[i],newsb);
   1983  1.36  dholland 			writeat(FFS_FSBTODB(newsb, cgtod(newsb, i)), cgs[i],
   1984   1.1       jtk 			    cgblksz);
   1985   1.1       jtk 		}
   1986   1.1       jtk 	}
   1987  1.25       riz 	if (needswap)
   1988  1.25       riz 		ffs_csum_swap(csums,csums,newsb->fs_cssize);
   1989  1.36  dholland 	writeat(FFS_FSBTODB(newsb, newsb->fs_csaddr), csums, newsb->fs_cssize);
   1990   1.1       jtk }
   1991   1.1       jtk /*
   1992   1.1       jtk  * Write the superblock, both to the main superblock and to each cg's
   1993   1.1       jtk  *  alternative superblock.
   1994   1.1       jtk  */
   1995   1.1       jtk static void
   1996   1.1       jtk write_sbs(void)
   1997   1.1       jtk {
   1998   1.1       jtk 	int i;
   1999   1.1       jtk 
   2000  1.20    mhitch 	if (newsb->fs_magic == FS_UFS1_MAGIC &&
   2001  1.20    mhitch 	    (newsb->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
   2002  1.20    mhitch 		newsb->fs_old_time = newsb->fs_time;
   2003  1.20    mhitch 	    	newsb->fs_old_size = newsb->fs_size;
   2004  1.20    mhitch 	    	/* we don't update fs_csaddr */
   2005  1.20    mhitch 	    	newsb->fs_old_dsize = newsb->fs_dsize;
   2006  1.20    mhitch 		newsb->fs_old_cstotal.cs_ndir = newsb->fs_cstotal.cs_ndir;
   2007  1.20    mhitch 		newsb->fs_old_cstotal.cs_nbfree = newsb->fs_cstotal.cs_nbfree;
   2008  1.20    mhitch 		newsb->fs_old_cstotal.cs_nifree = newsb->fs_cstotal.cs_nifree;
   2009  1.20    mhitch 		newsb->fs_old_cstotal.cs_nffree = newsb->fs_cstotal.cs_nffree;
   2010  1.20    mhitch 		/* fill fs_old_postbl_start with 256 bytes of 0xff? */
   2011  1.20    mhitch 	}
   2012  1.25       riz 	/* copy newsb back to oldsb, so we can use it for offsets if
   2013  1.25       riz 	   newsb has been swapped for writing to disk */
   2014  1.25       riz 	memcpy(oldsb, newsb, SBLOCKSIZE);
   2015  1.25       riz 	if (needswap)
   2016  1.25       riz 		ffs_sb_swap(newsb,newsb);
   2017  1.10    bouyer 	writeat(where /  DEV_BSIZE, newsb, SBLOCKSIZE);
   2018  1.25       riz 	for (i = 0; i < oldsb->fs_ncg; i++) {
   2019  1.36  dholland 		writeat(FFS_FSBTODB(oldsb, cgsblock(oldsb, i)), newsb, SBLOCKSIZE);
   2020   1.1       jtk 	}
   2021   1.1       jtk }
   2022  1.13      haad 
   2023  1.30  dholland static off_t
   2024  1.13      haad get_dev_size(char *dev_name)
   2025  1.13      haad {
   2026  1.13      haad 	struct dkwedge_info dkw;
   2027  1.13      haad 	struct partition *pp;
   2028  1.13      haad 	struct disklabel lp;
   2029  1.39  riastrad 	struct stat st;
   2030  1.13      haad 	size_t ptn;
   2031  1.28  dholland 
   2032  1.13      haad 	/* Get info about partition/wedge */
   2033  1.40  riastrad 	if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1)
   2034  1.39  riastrad 		return dkw.dkw_size;
   2035  1.40  riastrad 	if (ioctl(fd, DIOCGDINFO, &lp) != -1) {
   2036  1.13      haad 		ptn = strchr(dev_name, '\0')[-1] - 'a';
   2037  1.13      haad 		if (ptn >= lp.d_npartitions)
   2038  1.13      haad 			return 0;
   2039  1.13      haad 		pp = &lp.d_partitions[ptn];
   2040  1.13      haad 		return pp->p_size;
   2041  1.13      haad 	}
   2042  1.40  riastrad 	if (fstat(fd, &st) != -1 && S_ISREG(st.st_mode))
   2043  1.39  riastrad 		return st.st_size;
   2044  1.13      haad 
   2045  1.39  riastrad 	return 0;
   2046  1.13      haad }
   2047  1.13      haad 
   2048   1.1       jtk /*
   2049   1.1       jtk  * main().
   2050   1.1       jtk  */
   2051   1.1       jtk int
   2052  1.13      haad main(int argc, char **argv)
   2053   1.1       jtk {
   2054  1.13      haad 	int ch;
   2055  1.13      haad 	int ExpertFlag;
   2056  1.13      haad 	int SFlag;
   2057   1.4  christos 	size_t i;
   2058  1.13      haad 
   2059  1.23       riz 	char *special;
   2060  1.13      haad 	char reply[5];
   2061  1.28  dholland 
   2062  1.13      haad 	newsize = 0;
   2063  1.13      haad 	ExpertFlag = 0;
   2064  1.13      haad 	SFlag = 0;
   2065  1.28  dholland 
   2066  1.13      haad 	while ((ch = getopt(argc, argv, "s:y")) != -1) {
   2067  1.13      haad 		switch (ch) {
   2068  1.13      haad 		case 's':
   2069  1.13      haad 			SFlag = 1;
   2070  1.30  dholland 			newsize = strtoll(optarg, NULL, 10);
   2071  1.13      haad 			if(newsize < 1) {
   2072  1.13      haad 				usage();
   2073  1.13      haad 			}
   2074  1.13      haad 			break;
   2075  1.13      haad 		case 'y':
   2076  1.13      haad 			ExpertFlag = 1;
   2077  1.13      haad 			break;
   2078  1.13      haad 		case '?':
   2079  1.13      haad 			/* FALLTHROUGH */
   2080  1.13      haad 		default:
   2081  1.13      haad 			usage();
   2082  1.13      haad 		}
   2083  1.13      haad 	}
   2084  1.13      haad 	argc -= optind;
   2085  1.13      haad 	argv += optind;
   2086  1.13      haad 
   2087  1.13      haad 	if (argc != 1) {
   2088  1.13      haad 		usage();
   2089  1.13      haad 	}
   2090  1.13      haad 
   2091  1.23       riz 	special = *argv;
   2092  1.13      haad 
   2093  1.13      haad 	if (ExpertFlag == 0) {
   2094  1.24       wiz 		printf("It's required to manually run fsck on file system "
   2095  1.13      haad 		    "before you can resize it\n\n"
   2096  1.13      haad 		    " Did you run fsck on your disk (Yes/No) ? ");
   2097  1.13      haad 		fgets(reply, (int)sizeof(reply), stdin);
   2098  1.13      haad 		if (strcasecmp(reply, "Yes\n")) {
   2099  1.13      haad 			printf("\n Nothing done \n");
   2100  1.13      haad 			exit(EXIT_SUCCESS);
   2101  1.13      haad 		}
   2102   1.1       jtk 	}
   2103  1.28  dholland 
   2104  1.23       riz 	fd = open(special, O_RDWR, 0);
   2105   1.4  christos 	if (fd < 0)
   2106  1.23       riz 		err(EXIT_FAILURE, "Can't open `%s'", special);
   2107   1.1       jtk 	checksmallio();
   2108  1.13      haad 
   2109  1.13      haad 	if (SFlag == 0) {
   2110  1.23       riz 		newsize = get_dev_size(special);
   2111  1.13      haad 		if (newsize == 0)
   2112  1.15       riz 			err(EXIT_FAILURE,
   2113  1.24       wiz 			    "Can't resize file system, newsize not known.");
   2114  1.13      haad 	}
   2115  1.28  dholland 
   2116   1.1       jtk 	oldsb = (struct fs *) & sbbuf;
   2117   1.4  christos 	newsb = (struct fs *) (SBLOCKSIZE + (char *) &sbbuf);
   2118   1.4  christos 	for (where = search[i = 0]; search[i] != -1; where = search[++i]) {
   2119   1.9    bouyer 		readat(where / DEV_BSIZE, oldsb, SBLOCKSIZE);
   2120  1.23       riz 		switch (oldsb->fs_magic) {
   2121  1.23       riz 		case FS_UFS2_MAGIC:
   2122  1.31  dholland 			is_ufs2 = 1;
   2123  1.23       riz 			/* FALLTHROUGH */
   2124  1.23       riz 		case FS_UFS1_MAGIC:
   2125  1.23       riz 			needswap = 0;
   2126  1.23       riz 			break;
   2127  1.23       riz 		case FS_UFS2_MAGIC_SWAPPED:
   2128  1.23       riz  			is_ufs2 = 1;
   2129  1.23       riz 			/* FALLTHROUGH */
   2130  1.23       riz 		case FS_UFS1_MAGIC_SWAPPED:
   2131  1.23       riz 			needswap = 1;
   2132  1.23       riz 			break;
   2133  1.23       riz 		default:
   2134  1.23       riz 			continue;
   2135  1.23       riz 		}
   2136  1.23       riz 		if (!is_ufs2 && where == SBLOCK_UFS2)
   2137  1.16       riz 			continue;
   2138  1.23       riz 		break;
   2139   1.1       jtk 	}
   2140   1.4  christos 	if (where == (off_t)-1)
   2141  1.13      haad 		errx(EXIT_FAILURE, "Bad magic number");
   2142  1.25       riz 	if (needswap)
   2143  1.25       riz 		ffs_sb_swap(oldsb,oldsb);
   2144  1.20    mhitch 	if (oldsb->fs_magic == FS_UFS1_MAGIC &&
   2145  1.20    mhitch 	    (oldsb->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
   2146  1.20    mhitch 		oldsb->fs_csaddr = oldsb->fs_old_csaddr;
   2147  1.20    mhitch 		oldsb->fs_size = oldsb->fs_old_size;
   2148  1.20    mhitch 		oldsb->fs_dsize = oldsb->fs_old_dsize;
   2149  1.20    mhitch 		oldsb->fs_cstotal.cs_ndir = oldsb->fs_old_cstotal.cs_ndir;
   2150  1.20    mhitch 		oldsb->fs_cstotal.cs_nbfree = oldsb->fs_old_cstotal.cs_nbfree;
   2151  1.20    mhitch 		oldsb->fs_cstotal.cs_nifree = oldsb->fs_old_cstotal.cs_nifree;
   2152  1.20    mhitch 		oldsb->fs_cstotal.cs_nffree = oldsb->fs_old_cstotal.cs_nffree;
   2153  1.20    mhitch 		/* any others? */
   2154  1.20    mhitch 		printf("Resizing with ffsv1 superblock\n");
   2155  1.20    mhitch 	}
   2156  1.25       riz 
   2157   1.1       jtk 	oldsb->fs_qbmask = ~(int64_t) oldsb->fs_bmask;
   2158   1.1       jtk 	oldsb->fs_qfmask = ~(int64_t) oldsb->fs_fmask;
   2159  1.35  dholland 	if (oldsb->fs_ipg % FFS_INOPB(oldsb))
   2160  1.35  dholland 		errx(EXIT_FAILURE, "ipg[%d] %% FFS_INOPB[%d] != 0",
   2161  1.35  dholland 		    (int) oldsb->fs_ipg, (int) FFS_INOPB(oldsb));
   2162  1.25       riz 	/* The superblock is bigger than struct fs (there are trailing
   2163  1.25       riz 	 * tables, of non-fixed size); make sure we copy the whole
   2164  1.25       riz 	 * thing.  SBLOCKSIZE may be an over-estimate, but we do this
   2165  1.25       riz 	 * just once, so being generous is cheap. */
   2166  1.25       riz 	memcpy(newsb, oldsb, SBLOCKSIZE);
   2167   1.1       jtk 	loadcgs();
   2168  1.36  dholland 	if (newsize > FFS_FSBTODB(oldsb, oldsb->fs_size)) {
   2169   1.1       jtk 		grow();
   2170  1.36  dholland 	} else if (newsize < FFS_FSBTODB(oldsb, oldsb->fs_size)) {
   2171  1.25       riz 		if (is_ufs2)
   2172  1.25       riz 			errx(EXIT_FAILURE,"shrinking not supported for ufs2");
   2173   1.1       jtk 		shrink();
   2174   1.1       jtk 	}
   2175   1.1       jtk 	flush_cgs();
   2176   1.1       jtk 	write_sbs();
   2177  1.19       riz 	if (isplainfile())
   2178  1.19       riz 		ftruncate(fd,newsize * DEV_BSIZE);
   2179  1.13      haad 	return 0;
   2180  1.13      haad }
   2181  1.13      haad 
   2182  1.13      haad static void
   2183  1.13      haad usage(void)
   2184  1.13      haad {
   2185  1.13      haad 
   2186  1.25       riz 	(void)fprintf(stderr, "usage: %s [-y] [-s size] special\n",
   2187  1.25       riz 	    getprogname());
   2188  1.13      haad 	exit(EXIT_FAILURE);
   2189   1.1       jtk }
   2190