ufs/ffs/ffs_inode.c

/*
 * Copyright (c) 1982, 1986, 1989 Regents of the University of California.
 * All rights reserved.
 *
 * %sccs.include.redist.c%
 *
 *	@(#)ffs_inode.c	7.58 (Berkeley) 06/27/92
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/mount.h>
#include <sys/proc.h>
#include <sys/file.h>
#include <sys/buf.h>
#include <sys/vnode.h>
#include <sys/kernel.h>
#include <sys/malloc.h>

#include <vm/vm.h>

#include <ufs/ufs/quota.h>
#include <ufs/ufs/inode.h>
#include <ufs/ufs/ufsmount.h>
#include <ufs/ufs/ufs_extern.h>

#include <ufs/ffs/fs.h>
#include <ufs/ffs/ffs_extern.h>

static int ffs_indirtrunc __P((struct inode *, daddr_t, daddr_t, int, long *));

extern u_long nextgennumber;

int
ffs_init()
{
	return (ufs_init());
}

/*
 * Look up a UFS dinode number to find its incore vnode.
 * If it is not in core, read it in from the specified device.
 * If it is in core, wait for the lock bit to clear, then
 * return the inode locked. Detection and handling of mount
 * points must be done by the calling routine.
 */
ffs_vget (ap)
	struct vop_vget_args *ap;
{
	register struct fs *fs;
	register struct inode *ip;
	struct ufsmount *ump;
	struct buf *bp;
	struct dinode *dp;
	struct vnode *vp;
	union ihead *ih;
	dev_t dev;
	int i, type, error;

	ump = VFSTOUFS(ap->a_mp);
	dev = ump->um_dev;
	if ((*ap->a_vpp = ufs_ihashget(dev, ap->a_ino)) != NULL)
		return (0);

	/* Allocate a new vnode/inode. */
	if (error = getnewvnode(VT_UFS, ap->a_mp, ffs_vnodeop_p, &vp)) {
		*ap->a_vpp = NULL;
		return (error);
	}
	type = ump->um_devvp->v_tag == VT_MFS ? M_MFSNODE : M_FFSNODE; /* XXX */
	MALLOC(ip, struct inode *, sizeof(struct inode), type, M_WAITOK);
	vp->v_data = ip;
	ip->i_vnode = vp;
	ip->i_flag = 0;
	ip->i_devvp = 0;
	ip->i_mode = 0;
	ip->i_diroff = 0;
	ip->i_lockf = 0;
	ip->i_fs = fs = ump->um_fs;
	ip->i_dev = dev;
	ip->i_number = ap->a_ino;
#ifdef QUOTA
	for (i = 0; i < MAXQUOTAS; i++)
		ip->i_dquot[i] = NODQUOT;
#endif
	/*
	 * Put it onto its hash chain and lock it so that other requests for
	 * this inode will block if they arrive while we are sleeping waiting
	 * for old data structures to be purged or for the contents of the
	 * disk portion of this inode to be read.
	 */
	ufs_ihashins(ip);

	/* Read in the disk contents for the inode, copy into the inode. */
	if (error = bread(ump->um_devvp, fsbtodb(fs, itod(fs, ap->a_ino)),
	    (int)fs->fs_bsize, NOCRED, &bp)) {
		/*
		 * The inode does not contain anything useful, so it would
		 * be misleading to leave it on its hash chain. It will be
		 * returned to the free list by ufs_iput().
		 */
		remque(ip);
		ip->i_forw = ip;
		ip->i_back = ip;

		/* Unlock and discard unneeded inode. */
		ufs_iput(ip);
		brelse(bp);
		*ap->a_vpp = NULL;
		return (error);
	}
	dp = bp->b_un.b_dino;
	dp += itoo(fs, ap->a_ino);
	ip->i_din = *dp;
	brelse(bp);

	/*
	 * Initialize the vnode from the inode, check for aliases.
	 * Note that the underlying vnode may have changed.
	 */
	if (error = ufs_vinit(ap->a_mp, ffs_specop_p, FFS_FIFOOPS, &vp)) {
		ufs_iput(ip);
		*ap->a_vpp = NULL;
		return (error);
	}
	/*
	 * Finish inode initialization now that aliasing has been resolved.
	 */
	ip->i_devvp = ump->um_devvp;
	VREF(ip->i_devvp);
	/*
	 * Set up a generation number for this inode if it does not
	 * already have one. This should only happen on old filesystems.
	 */
	if (ip->i_gen == 0) {
		if (++nextgennumber < (u_long)time.tv_sec)
			nextgennumber = time.tv_sec;
		ip->i_gen = nextgennumber;
		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0)
			ip->i_flag |= IMOD;
	}
	/*
	 * Ensure that uid and gid are correct. This is a temporary
	 * fix until fsck has been changed to do the update.
	 */
	if (fs->fs_inodefmt < FS_44INODEFMT) {		/* XXX */
		ip->i_uid = ip->i_din.di_ouid;		/* XXX */
		ip->i_gid = ip->i_din.di_ogid;		/* XXX */
	}						/* XXX */

	*ap->a_vpp = vp;
	return (0);
}

/*
 * Update the access, modified, and inode change times as specified
 * by the IACC, IUPD, and ICHG flags respectively. The IMOD flag
 * is used to specify that the inode needs to be updated but that
 * the times have already been set. The access and modified times
 * are taken from the second and third parameters; the inode change
 * time is always taken from the current time. If waitfor is set,
 * then wait for the disk write of the inode to complete.
 */
int
ffs_update (ap)
	struct vop_update_args *ap;
{
	struct buf *bp;
	struct inode *ip;
	struct dinode *dp;
	register struct fs *fs;
	int error;

	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
		return (0);
	ip = VTOI(ap->a_vp);
	if ((ip->i_flag & (IUPD|IACC|ICHG|IMOD)) == 0)
		return (0);
	if (ip->i_flag&IACC)
		ip->i_atime.ts_sec = ap->a_ta->tv_sec;
	if (ip->i_flag&IUPD) {
		ip->i_mtime.ts_sec = ap->a_tm->tv_sec;
		ip->i_modrev++;
	}
	if (ip->i_flag&ICHG)
		ip->i_ctime.ts_sec = time.tv_sec;
	ip->i_flag &= ~(IUPD|IACC|ICHG|IMOD);
	fs = ip->i_fs;
	/*
	 * Ensure that uid and gid are correct. This is a temporary
	 * fix until fsck has been changed to do the update.
	 */
	if (fs->fs_inodefmt < FS_44INODEFMT) {		/* XXX */
		ip->i_din.di_ouid = ip->i_uid;		/* XXX */
		ip->i_din.di_ogid = ip->i_gid;		/* XXX */
	}						/* XXX */
	if (error = bread(ip->i_devvp, fsbtodb(fs, itod(fs, ip->i_number)),
		(int)fs->fs_bsize, NOCRED, &bp)) {
		brelse(bp);
		return (error);
	}
	dp = bp->b_un.b_dino + itoo(fs, ip->i_number);
	*dp = ip->i_din;
	if (ap->a_waitfor)
		return (bwrite(bp));
	else {
		bdwrite(bp);
		return (0);
	}
}

#define	SINGLE	0	/* index of single indirect block */
#define	DOUBLE	1	/* index of double indirect block */
#define	TRIPLE	2	/* index of triple indirect block */
/*
 * Truncate the inode ip to at most length size.  Free affected disk
 * blocks -- the blocks of the file are removed in reverse order.
 *
 * NB: triple indirect blocks are untested.
 */
ffs_truncate (ap)
	struct vop_truncate_args *ap;
{
	USES_VOP_UPDATE;
	register struct vnode *ovp = ap->a_vp;
	register daddr_t lastblock;
	register struct inode *oip;
	daddr_t bn, lbn, lastiblock[NIADDR];
	register struct fs *fs;
	register struct inode *ip;
	struct buf *bp;
	int offset, size, level;
	long count, nblocks, blocksreleased = 0;
	register int i;
	int aflags, error, allerror;
	struct inode tip;
	off_t osize;

	oip = VTOI(ovp);
	if (ovp->v_type == VLNK && ovp->v_mount->mnt_maxsymlinklen > 0) {
#ifdef DIAGNOSTIC
		if (ap->a_length != 0)
			panic("ffs_truncate: partial truncate of symlink");
#endif
		bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
		oip->i_size = 0;
		oip->i_flag |= ICHG|IUPD;
		return (VOP_UPDATE(ovp, &time, &time, 1));
	}
	if (oip->i_size <= ap->a_length) {
		oip->i_flag |= ICHG|IUPD;
		return (VOP_UPDATE(ovp, &time, &time, 1));
	}
	vnode_pager_setsize(ovp, (u_long)ap->a_length);
	/*
	 * Calculate index into inode's block list of
	 * last direct and indirect blocks (if any)
	 * which we want to keep.  Lastblock is -1 when
	 * the file is truncated to 0.
	 */
	fs = oip->i_fs;
	lastblock = lblkno(fs, ap->a_length + fs->fs_bsize - 1) - 1;
	lastiblock[SINGLE] = lastblock - NDADDR;
	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
	nblocks = btodb(fs->fs_bsize);
	/*
	 * Update the size of the file. If the file is not being
	 * truncated to a block boundry, the contents of the
	 * partial block following the end of the file must be
	 * zero'ed in case it ever become accessable again because
	 * of subsequent file growth.
	 */
	osize = oip->i_size;
	offset = blkoff(fs, ap->a_length);
	if (offset == 0) {
		oip->i_size = ap->a_length;
	} else {
		lbn = lblkno(fs, ap->a_length);
		aflags = B_CLRBUF;
		if (ap->a_flags & IO_SYNC)
			aflags |= B_SYNC;
#ifdef QUOTA
		if (error = getinoquota(oip))
			return (error);
#endif
		if (error = ffs_balloc(oip, lbn, offset, ap->a_cred, &bp, aflags))
			return (error);
		oip->i_size = ap->a_length;
		size = blksize(fs, oip, lbn);
		(void) vnode_pager_uncache(ovp);
		bzero(bp->b_un.b_addr + offset, (unsigned)(size - offset));
		allocbuf(bp, size);
		if (ap->a_flags & IO_SYNC)
			bwrite(bp);
		else
			bdwrite(bp);
	}
	/*
	 * Update file and block pointers on disk before we start freeing
	 * blocks.  If we crash before free'ing blocks below, the blocks
	 * will be returned to the free list.  lastiblock values are also
	 * normalized to -1 for calls to ffs_indirtrunc below.
	 */
	tip = *oip;
	tip.i_size = osize;
	for (level = TRIPLE; level >= SINGLE; level--)
		if (lastiblock[level] < 0) {
			oip->i_ib[level] = 0;
			lastiblock[level] = -1;
		}
	for (i = NDADDR - 1; i > lastblock; i--)
		oip->i_db[i] = 0;
	oip->i_flag |= ICHG|IUPD;
	allerror = vinvalbuf(ovp, ap->a_length > 0, ap->a_cred, ap->a_p);
	if (error = VOP_UPDATE(ovp, &time, &time, MNT_WAIT))
		allerror = error;

	/*
	 * Indirect blocks first.
	 */
	ip = &tip;
	for (level = TRIPLE; level >= SINGLE; level--) {
		bn = ip->i_ib[level];
		if (bn != 0) {
			error = ffs_indirtrunc(ip,
			    bn, lastiblock[level], level, &count);
			if (error)
				allerror = error;
			blocksreleased += count;
			if (lastiblock[level] < 0) {
				ip->i_ib[level] = 0;
				ffs_blkfree(ip, bn, fs->fs_bsize);
				blocksreleased += nblocks;
			}
		}
		if (lastiblock[level] >= 0)
			goto done;
	}

	/*
	 * All whole direct blocks or frags.
	 */
	for (i = NDADDR - 1; i > lastblock; i--) {
		register long bsize;

		bn = ip->i_db[i];
		if (bn == 0)
			continue;
		ip->i_db[i] = 0;
		bsize = blksize(fs, ip, i);
		ffs_blkfree(ip, bn, bsize);
		blocksreleased += btodb(bsize);
	}
	if (lastblock < 0)
		goto done;

	/*
	 * Finally, look for a change in size of the
	 * last direct block; release any frags.
	 */
	bn = ip->i_db[lastblock];
	if (bn != 0) {
		long oldspace, newspace;

		/*
		 * Calculate amount of space we're giving
		 * back as old block size minus new block size.
		 */
		oldspace = blksize(fs, ip, lastblock);
		ip->i_size = ap->a_length;
		newspace = blksize(fs, ip, lastblock);
		if (newspace == 0)
			panic("itrunc: newspace");
		if (oldspace - newspace > 0) {
			/*
			 * Block number of space to be free'd is
			 * the old block # plus the number of frags
			 * required for the storage we're keeping.
			 */
			bn += numfrags(fs, newspace);
			ffs_blkfree(ip, bn, oldspace - newspace);
			blocksreleased += btodb(oldspace - newspace);
		}
	}
done:
/* BEGIN PARANOIA */
	for (level = SINGLE; level <= TRIPLE; level++)
		if (ip->i_ib[level] != oip->i_ib[level])
			panic("itrunc1");
	for (i = 0; i < NDADDR; i++)
		if (ip->i_db[i] != oip->i_db[i])
			panic("itrunc2");
/* END PARANOIA */
	oip->i_blocks -= blocksreleased;
	if (oip->i_blocks < 0)			/* sanity */
		oip->i_blocks = 0;
	oip->i_flag |= ICHG;
#ifdef QUOTA
	if (!getinoquota(oip))
		(void) chkdq(oip, -blocksreleased, NOCRED, 0);
#endif
	return (allerror);
}

/*
 * Release blocks associated with the inode ip and stored in the indirect
 * block bn.  Blocks are free'd in LIFO order up to (but not including)
 * lastbn.  If level is greater than SINGLE, the block is an indirect block
 * and recursive calls to indirtrunc must be used to cleanse other indirect
 * blocks.
 *
 * NB: triple indirect blocks are untested.
 */
static int
ffs_indirtrunc(ip, bn, lastbn, level, countp)
	register struct inode *ip;
	daddr_t bn, lastbn;
	int level;
	long *countp;
{
	register int i;
	struct buf *bp;
	register struct fs *fs = ip->i_fs;
	register daddr_t *bap;
	daddr_t *copy, nb, last;
	long blkcount, factor;
	int nblocks, blocksreleased = 0;
	int error, allerror = 0;

	/*
	 * Calculate index in current block of last
	 * block to be kept.  -1 indicates the entire
	 * block so we need not calculate the index.
	 */
	factor = 1;
	for (i = SINGLE; i < level; i++)
		factor *= NINDIR(fs);
	last = lastbn;
	if (lastbn > 0)
		last /= factor;
	nblocks = btodb(fs->fs_bsize);
	/*
	 * Get buffer of block pointers, zero those
	 * entries corresponding to blocks to be free'd,
	 * and update on disk copy first.
	 */
	error = bread(ip->i_devvp, fsbtodb(fs, bn), (int)fs->fs_bsize,
		NOCRED, &bp);
	if (error) {
		brelse(bp);
		*countp = 0;
		return (error);
	}
	bap = bp->b_un.b_daddr;
	MALLOC(copy, daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
	bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize);
	bzero((caddr_t)&bap[last + 1],
	  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t));
	if (last == -1)
		bp->b_flags |= B_INVAL;
	error = bwrite(bp);
	if (error)
		allerror = error;
	bap = copy;

	/*
	 * Recursively free totally unused blocks.
	 */
	for (i = NINDIR(fs) - 1; i > last; i--) {
		nb = bap[i];
		if (nb == 0)
			continue;
		if (level > SINGLE) {
			if (error = ffs_indirtrunc(ip,
			    nb, (daddr_t)-1, level - 1, &blkcount))
				allerror = error;
			blocksreleased += blkcount;
		}
		ffs_blkfree(ip, nb, fs->fs_bsize);
		blocksreleased += nblocks;
	}

	/*
	 * Recursively free last partial block.
	 */
	if (level > SINGLE && lastbn >= 0) {
		last = lastbn % factor;
		nb = bap[i];
		if (nb != 0) {
			if (error =
			    ffs_indirtrunc(ip, nb, last, level - 1, &blkcount))
				allerror = error;
			blocksreleased += blkcount;
		}
	}
	FREE(copy, M_TEMP);
	*countp = blocksreleased;
	return (allerror);
}