X-Git-Url: https://git.subgeniuskitty.com/unix-history/.git/blobdiff_plain/a8d3bf7f08061516a44629fbaf4c7bb727122742..28708b7290eb3aa4837b4ce01932e3a4c63f078f:/usr/src/sys/ufs/ufs/ufs_inode.c

diff --git a/usr/src/sys/ufs/ufs/ufs_inode.c b/usr/src/sys/ufs/ufs/ufs_inode.c
index 4c2db27e54..e1954e4548 100644
--- a/usr/src/sys/ufs/ufs/ufs_inode.c
+++ b/usr/src/sys/ufs/ufs/ufs_inode.c
@@ -1,222 +1,220 @@
-/*	ufs_inode.c	4.26	82/10/17	*/
-
-#include "../h/param.h"
-#include "../h/systm.h"
-#include "../h/mount.h"
-#include "../h/dir.h"
-#include "../h/user.h"
-#include "../h/inode.h"
-#include "../h/fs.h"
-#include "../h/conf.h"
-#include "../h/buf.h"
-#ifdef QUOTA
-#include "../h/quota.h"
-#endif
-#include "../h/kernel.h"
+/*
+ * Copyright (c) 1982, 1986, 1989 Regents of the University of California.
+ * All rights reserved.
+ *
+ * %sccs.include.redist.c%
+ *
+ *	@(#)ufs_inode.c	7.39 (Berkeley) %G%
+ */
+
+#include "param.h"
+#include "systm.h"
+#include "mount.h"
+#include "proc.h"
+#include "file.h"
+#include "buf.h"
+#include "vnode.h"
+#include "kernel.h"
+#include "malloc.h"
 
-#define	INOHSZ	63
+#include "quota.h"
+#include "inode.h"
+#include "fs.h"
+#include "ufsmount.h"
+
+#define	INOHSZ	512
 #if	((INOHSZ&(INOHSZ-1)) == 0)
 #define	INOHASH(dev,ino)	(((dev)+(ino))&(INOHSZ-1))
 #else
-#define	INOHASH(dev,ino)	(((dev)+(ino))%INOHSZ)
+#define	INOHASH(dev,ino)	(((unsigned)((dev)+(ino)))%INOHSZ)
 #endif
 
-union ihead {				/* inode LRU cache, Chris Maltby */
+union ihead {
 	union  ihead *ih_head[2];
 	struct inode *ih_chain[2];
 } ihead[INOHSZ];
 
-struct inode *ifreeh, **ifreet;
+int prtactive;	/* 1 => print out reclaim of active vnodes */
 
 /*
- * Initialize hash links for inodes
- * and build inode free list.
+ * Initialize hash links for inodes.
  */
-ihinit()
+ufs_init()
 {
 	register int i;
-	register struct inode *ip = inode;
-	register union  ihead *ih = ihead;
+	register union ihead *ih = ihead;
 
+#ifndef lint
+	if (VN_MAXPRIVATE < sizeof(struct inode))
+		panic("ihinit: too small");
+#endif /* not lint */
 	for (i = INOHSZ; --i >= 0; ih++) {
 		ih->ih_head[0] = ih;
 		ih->ih_head[1] = ih;
 	}
-	ifreeh = ip;
-	ifreet = &ip->i_freef;
-	ip->i_freeb = &ifreeh;
-	ip->i_forw = ip;
-	ip->i_back = ip;
-	for (i = ninode; --i > 0; ) {
-		++ip;
-		ip->i_forw = ip;
-		ip->i_back = ip;
-		*ifreet = ip;
-		ip->i_freeb = ifreet;
-		ifreet = &ip->i_freef;
-	}
-	ip->i_freef = NULL;
-}
-
-#ifdef notdef
-/*
- * Find an inode if it is incore.
- * This is the equivalent, for inodes,
- * of ``incore'' in bio.c or ``pfind'' in subr.c.
- */
-struct inode *
-ifind(dev, ino)
-	dev_t dev;
-	ino_t ino;
-{
-	register struct inode *ip;
-	register union  ihead *ih;
-
-	ih = &ihead[INOHASH(dev, ino)];
-	for (ip = ih->ih_chain[0]; ip != (struct inode *)ih; ip = ip->i_forw)
-		if (ino==ip->i_number && dev==ip->i_dev)
-			return (ip);
-	return ((struct inode *)0);
+#ifdef QUOTA
+	dqinit();
+#endif /* QUOTA */
 }
-#endif notdef
 
 /*
- * Look up an inode by device,inumber.
+ * Look up an vnode/inode by device,inumber.
  * If it is in core (in the inode structure),
  * honor the locking protocol.
  * If it is not in core, read it in from the
  * specified device.
- * If the inode is mounted on, perform
- * the indicated indirection.
+ * Callers must check for mount points!!
  * In all cases, a pointer to a locked
  * inode structure is returned.
- *
- * panic: no imt -- if the mounted file
- *	system is not in the mount table.
- *	"cannot happen"
  */
-struct inode *
-iget(dev, fs, ino)
-	dev_t dev;
-	register struct fs *fs;
+iget(xp, ino, ipp)
+	struct inode *xp;
 	ino_t ino;
+	struct inode **ipp;
 {
-	register struct inode *ip;
-	register union  ihead *ih;
-	register struct mount *mp;
-	register struct buf *bp;
-	register struct dinode *dp;
-	register struct inode *iq;
+	dev_t dev = xp->i_dev;
+	struct mount *mntp = ITOV(xp)->v_mount;
+	register struct fs *fs = VFSTOUFS(mntp)->um_fs;
+	extern struct vnodeops ufs_vnodeops, spec_inodeops;
+	register struct inode *ip, *iq;
+	register struct vnode *vp;
+	struct vnode *nvp;
+	struct buf *bp;
+	struct dinode *dp;
+	union ihead *ih;
+	int i, error;
 
-loop:
-	if (getfs(dev) != fs)
-		panic("iget: bad fs");
 	ih = &ihead[INOHASH(dev, ino)];
-	for (ip = ih->ih_chain[0]; ip != (struct inode *)ih; ip = ip->i_forw)
-		if (ino == ip->i_number && dev == ip->i_dev) {
-			if ((ip->i_flag&ILOCKED) != 0) {
-				ip->i_flag |= IWANT;
-				sleep((caddr_t)ip, PINOD);
-				goto loop;
-			}
-			if ((ip->i_flag&IMOUNT) != 0) {
-				for (mp = &mount[0]; mp < &mount[NMOUNT]; mp++)
-					if(mp->m_inodp == ip) {
-						dev = mp->m_dev;
-						fs = mp->m_bufp->b_un.b_fs;
-						ino = ROOTINO;
-						goto loop;
-					}
-				panic("no imt");
-			}
-			if (ip->i_count == 0) {		/* ino on free list */
-				if (iq = ip->i_freef)
-					iq->i_freeb = ip->i_freeb;
-				else
-					ifreet = ip->i_freeb;
-				*ip->i_freeb = iq;
-				ip->i_freef = NULL;
-				ip->i_freeb = NULL;
-			}
-			ip->i_count++;
-			ip->i_flag |= ILOCKED;
-			return(ip);
+loop:
+	for (ip = ih->ih_chain[0]; ip != (struct inode *)ih; ip = ip->i_forw) {
+		if (ino != ip->i_number || dev != ip->i_dev)
+			continue;
+		if ((ip->i_flag&ILOCKED) != 0) {
+			ip->i_flag |= IWANT;
+			sleep((caddr_t)ip, PINOD);
+			goto loop;
 		}
-
-	if ((ip = ifreeh) == NULL) {
-		tablefull("inode");
-		u.u_error = ENFILE;
-		return(NULL);
+		if (vget(ITOV(ip)))
+			goto loop;
+		*ipp = ip;
+		return(0);
 	}
-	if (iq = ip->i_freef)
-		iq->i_freeb = &ifreeh;
-	ifreeh = iq;
-	ip->i_freef = NULL;
-	ip->i_freeb = NULL;
 	/*
-	 * Now to take inode off the hash chain it was on
-	 * (initially, or after an iflush, it is on a "hash chain"
-	 * consisting entirely of itself, and pointed to by no-one,
-	 * but that doesn't matter), and put it on the chain for
-	 * its new (ino, dev) pair
+	 * Allocate a new inode.
 	 */
-	remque(ip);
-	insque(ip, ih);
+	if (error = getnewvnode(VT_UFS, mntp, &ufs_vnodeops, &nvp)) {
+		*ipp = 0;
+		return (error);
+	}
+	ip = VTOI(nvp);
+	ip->i_vnode = nvp;
+	ip->i_flag = 0;
+	ip->i_devvp = 0;
+	ip->i_mode = 0;
+	ip->i_diroff = 0;
+	ip->i_lockf = 0;
 #ifdef QUOTA
-	dqrele(ip->i_dquot);
+	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.
+	 */
 	ip->i_dev = dev;
-	ip->i_fs = fs;
 	ip->i_number = ino;
-	ip->i_flag = ILOCKED;
-	ip->i_count++;
-	ip->i_lastr = 0;
-	bp = bread(dev, fsbtodb(fs, itod(fs, ino)), fs->fs_bsize);
+	insque(ip, ih);
+	ILOCK(ip);
 	/*
-	 * Check I/O errors
+	 * Read in the disk contents for the inode.
 	 */
-	if ((bp->b_flags&B_ERROR) != 0) {
-		brelse(bp);
+	if (error = bread(VFSTOUFS(mntp)->um_devvp, fsbtodb(fs, itod(fs, ino)),
+	    (int)fs->fs_bsize, NOCRED, &bp)) {
 		/*
-		 * the inode doesn't contain anything useful, so it would
+		 * The inode does not contain anything useful, so it would
 		 * be misleading to leave it on its hash chain.
-		 * 'iput' will take care of putting it back on the free list.
+		 * Iput() will take care of putting it back on the free list.
 		 */
 		remque(ip);
 		ip->i_forw = ip;
 		ip->i_back = ip;
 		/*
-		 * we also loose its inumber, just in case (as iput
-		 * doesn't do that any more) - but as it isn't on its
-		 * hash chain, I doubt if this is really necessary .. kre
-		 * (probably the two methods are interchangable)
+		 * Unlock and discard unneeded inode.
 		 */
-		ip->i_number = 0;
-#ifdef QUOTA
-		ip->i_dquot = NODQUOT;
-#endif
 		iput(ip);
-		return(NULL);
+		brelse(bp);
+		*ipp = 0;
+		return (error);
 	}
 	dp = bp->b_un.b_dino;
 	dp += itoo(fs, ino);
-	ip->i_ic = dp->di_ic;
+	ip->i_din = *dp;
 	brelse(bp);
-#ifdef QUOTA
-	if (ip->i_mode == 0)
-		ip->i_dquot = NODQUOT;
-	else
-		ip->i_dquot = inoquota(ip);
-#endif
-	return (ip);
+	/*
+	 * Initialize the associated vnode
+	 */
+	vp = ITOV(ip);
+	vp->v_type = IFTOVT(ip->i_mode);
+	if (vp->v_type == VFIFO) {
+#ifdef FIFO
+		extern struct vnodeops fifo_inodeops;
+		vp->v_op = &fifo_inodeops;
+#else
+		iput(ip);
+		*ipp = 0;
+		return (EOPNOTSUPP);
+#endif /* FIFO */
+	}
+	if (vp->v_type == VCHR || vp->v_type == VBLK) {
+		vp->v_op = &spec_inodeops;
+		if (nvp = checkalias(vp, ip->i_rdev, mntp)) {
+			/*
+			 * Reinitialize aliased inode.
+			 */
+			vp = nvp;
+			iq = VTOI(vp);
+			iq->i_vnode = vp;
+			iq->i_flag = 0;
+			ILOCK(iq);
+			iq->i_din = ip->i_din;
+			iq->i_dev = dev;
+			iq->i_number = ino;
+			insque(iq, ih);
+			/*
+			 * Discard unneeded vnode
+			 */
+			ip->i_mode = 0;
+			iput(ip);
+			ip = iq;
+		}
+	}
+	if (ino == ROOTINO)
+		vp->v_flag |= VROOT;
+	/*
+	 * Finish inode initialization.
+	 */
+	ip->i_fs = fs;
+	ip->i_devvp = VFSTOUFS(mntp)->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;
+	}
+	*ipp = ip;
+	return (0);
 }
 
 /*
- * Decrement reference count of
- * an inode structure.
- * On the last reference,
- * write the inode out and if necessary,
- * truncate and deallocate the file.
+ * Unlock and decrement the reference count of an inode structure.
  */
 iput(ip)
 	register struct inode *ip;
@@ -224,306 +222,419 @@ iput(ip)
 
 	if ((ip->i_flag & ILOCKED) == 0)
 		panic("iput");
-	iunlock(ip);
-	irele(ip);
+	IUNLOCK(ip);
+	vrele(ITOV(ip));
 }
 
-irele(ip)
-	register struct inode *ip;
+/*
+ * Last reference to an inode, write the inode out and if necessary,
+ * truncate and deallocate the file.
+ */
+ufs_inactive(vp, p)
+	struct vnode *vp;
+	struct proc *p;
 {
-	int mode;
+	register struct inode *ip = VTOI(vp);
+	int mode, error = 0;
 
-	if (ip->i_count == 1) {
-		ip->i_flag |= ILOCKED;
-		if (ip->i_nlink <= 0) {
-			itrunc(ip, 0);
-			mode = ip->i_mode;
-			ip->i_mode = 0;
-			ip->i_rdev = 0;
-			ip->i_flag |= IUPD|ICHG;
-			ifree(ip, ip->i_number, mode);
+	if (prtactive && vp->v_usecount != 0)
+		vprint("ufs_inactive: pushing active", vp);
+	/*
+	 * Get rid of inodes related to stale file handles.
+	 */
+	if (ip->i_mode == 0) {
+		if ((vp->v_flag & VXLOCK) == 0)
+			vgone(vp);
+		return (0);
+	}
+	ILOCK(ip);
+	if (ip->i_nlink <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 #ifdef QUOTA
-			chkiq(ip->i_dev, ip, ip->i_uid, 0);
-			dqrele(ip->i_dquot);
-			ip->i_dquot = NODQUOT;
+		if (!getinoquota(ip))
+			(void) chkiq(ip, -1, NOCRED, 0);
 #endif
+		error = itrunc(ip, (u_long)0, 0);
+		mode = ip->i_mode;
+		ip->i_mode = 0;
+		ip->i_rdev = 0;
+		ip->i_flag |= IUPD|ICHG;
+		ifree(ip, ip->i_number, mode);
+	}
+	IUPDAT(ip, &time, &time, 0);
+	IUNLOCK(ip);
+	ip->i_flag = 0;
+	/*
+	 * If we are done with the inode, reclaim it
+	 * so that it can be reused immediately.
+	 */
+	if (vp->v_usecount == 0 && ip->i_mode == 0)
+		vgone(vp);
+	return (error);
+}
+
+/*
+ * Reclaim an inode so that it can be used for other purposes.
+ */
+ufs_reclaim(vp)
+	register struct vnode *vp;
+{
+	register struct inode *ip = VTOI(vp);
+	int i;
+
+	if (prtactive && vp->v_usecount != 0)
+		vprint("ufs_reclaim: pushing active", vp);
+	/*
+	 * Remove the inode from its hash chain.
+	 */
+	remque(ip);
+	ip->i_forw = ip;
+	ip->i_back = ip;
+	/*
+	 * Purge old data structures associated with the inode.
+	 */
+	cache_purge(vp);
+	if (ip->i_devvp) {
+		vrele(ip->i_devvp);
+		ip->i_devvp = 0;
+	}
+#ifdef QUOTA
+	for (i = 0; i < MAXQUOTAS; i++) {
+		if (ip->i_dquot[i] != NODQUOT) {
+			dqrele(vp, ip->i_dquot[i]);
+			ip->i_dquot[i] = NODQUOT;
 		}
-		IUPDAT(ip, &time.tv_sec, &time.tv_sec, 0);
-		iunlock(ip);
-		ip->i_flag = 0;
-		/*
-		 * Put the inode on the end of the free list.
-		 * Possibly in some cases it would be better to
-		 * put the inode at the head of the free list,
-		 * (eg: where i_mode == 0 || i_number == 0)
-		 * but I will think about that later .. kre
-		 * (i_number is rarely 0 - only after an i/o error in iget,
-		 * where i_mode == 0, the inode will probably be wanted
-		 * again soon for an ialloc, so possibly we should keep it)
-		 */
-		if (ifreeh) {
-			*ifreet = ip;
-			ip->i_freeb = ifreet;
-		} else {
-			ifreeh = ip;
-			ip->i_freeb = &ifreeh;
-		}
-		ip->i_freef = NULL;
-		ifreet = &ip->i_freef;
 	}
-	ip->i_count--;
+#endif
+	ip->i_flag = 0;
+	return (0);
 }
 
 /*
- * Check accessed and update flags on
- * an inode structure.
- * If any is on, update the inode
- * with the current time.
- * If waitfor is given, then must insure
- * i/o order so wait for write to complete.
+ * Check accessed and update flags on an inode structure.
+ * If any is on, update the inode with the current time.
+ * If waitfor is given, then must ensure I/O order,
+ * so wait for write to complete.
  */
 iupdat(ip, ta, tm, waitfor)
 	register struct inode *ip;
-	time_t *ta, *tm;
+	struct timeval *ta, *tm;
 	int waitfor;
 {
-	register struct buf *bp;
+	struct buf *bp;
+	struct vnode *vp = ITOV(ip);
 	struct dinode *dp;
-	register struct fs *fp;
-
-	fp = ip->i_fs;
-	if ((ip->i_flag & (IUPD|IACC|ICHG)) != 0) {
-		if (fp->fs_ronly)
-			return;
-		bp = bread(ip->i_dev, fsbtodb(fp, itod(fp, ip->i_number)),
-			fp->fs_bsize);
-		if (bp->b_flags & B_ERROR) {
-			brelse(bp);
-			return;
-		}
-		if (ip->i_flag&IACC)
-			ip->i_atime = *ta;
-		if (ip->i_flag&IUPD)
-			ip->i_mtime = *tm;
-		if (ip->i_flag&ICHG)
-			ip->i_ctime = time.tv_sec;
-		ip->i_flag &= ~(IUPD|IACC|ICHG);
-		dp = bp->b_un.b_dino + itoo(fp, ip->i_number);
-		dp->di_ic = ip->i_ic;
-		if (waitfor)
-			bwrite(bp);
-		else
-			bdwrite(bp);
+	register struct fs *fs;
+
+	fs = ip->i_fs;
+	if ((ip->i_flag & (IUPD|IACC|ICHG|IMOD)) == 0)
+		return (0);
+	if (vp->v_mount->mnt_flag & MNT_RDONLY)
+		return (0);
+	error = bread(ip->i_devvp, fsbtodb(fs, itod(fs, ip->i_number)),
+		(int)fs->fs_bsize, NOCRED, &bp);
+	if (error) {
+		brelse(bp);
+		return (error);
+	}
+	if (ip->i_flag&IACC)
+		ip->i_atime = ta->tv_sec;
+	if (ip->i_flag&IUPD)
+		ip->i_mtime = tm->tv_sec;
+	if (ip->i_flag&ICHG)
+		ip->i_ctime = time.tv_sec;
+	ip->i_flag &= ~(IUPD|IACC|ICHG|IMOD);
+	dp = bp->b_un.b_dino + itoo(fs, ip->i_number);
+	*dp = ip->i_din;
+	if (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.
+ * 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.
  */
-itrunc(ip, length)
-	register struct inode *ip;
-	register int length;
+itrunc(oip, length, flags)
+	register struct inode *oip;
+	u_long length;
+	int flags;
 {
-	register i;
-	dev_t dev;
-	daddr_t bn;
-	struct inode itmp;
+	register daddr_t lastblock;
+	daddr_t bn, lbn, lastiblock[NIADDR];
 	register struct fs *fs;
-#ifdef QUOTA
-	register long cnt = 0;
-	long tloop();
-#endif
-	/*
-	 * Only plain files, directories and symbolic
-	 * links contain blocks.
-	 */
-	i = ip->i_mode & IFMT;
-	if (i != IFREG && i != IFDIR && i != IFLNK)
-		return;
-	if (ip->i_size <= length)
-		return;
-
-	/*
-	 * Clean inode on disk before freeing blocks
-	 * to insure no duplicates if system crashes.
-	 */
-	itmp = *ip;
-	itmp.i_size = length;
-	for (i = 0; i < NDADDR; i++)
-		itmp.i_db[i] = 0;
-	for (i = 0; i < NIADDR; i++)
-		itmp.i_ib[i] = 0;
-	itmp.i_flag |= ICHG|IUPD;
-	iupdat(&itmp, &time.tv_sec, &time.tv_sec, 1);
-	ip->i_flag &= ~(IUPD|IACC|ICHG);
+	register struct inode *ip;
+	struct buf *bp;
+	int offset, osize, size, level;
+	long count, nblocks, blocksreleased = 0;
+	register int i;
+	int aflags, error, allerror;
+	struct inode tip;
 
+	vnode_pager_setsize(ITOV(oip), length);
+	if (oip->i_size <= length) {
+		oip->i_flag |= ICHG|IUPD;
+		error = iupdat(oip, &time, &time, 1);
+		return (error);
+	}
 	/*
-	 * Now return blocks to free list... if machine
-	 * crashes, they will be harmless MISSING blocks.
+	 * 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 = ip->i_fs;
+	fs = oip->i_fs;
+	lastblock = lblkno(fs, 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);
 	/*
-	 * release double indirect block first
+	 * 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.
 	 */
-	bn = ip->i_ib[NIADDR-1];
-	if (bn != (daddr_t)0) {
-		ip->i_ib[NIADDR - 1] = (daddr_t)0;
+	osize = oip->i_size;
+	offset = blkoff(fs, length);
+	if (offset == 0) {
+		oip->i_size = length;
+	} else {
+		lbn = lblkno(fs, length);
+		aflags = B_CLRBUF;
+		if (flags & IO_SYNC)
+			aflags |= B_SYNC;
 #ifdef QUOTA
-		cnt +=
+		if (error = getinoquota(oip))
+			return (error);
 #endif
-			tloop(ip, bn, 1);
+		if (error = balloc(oip, lbn, offset, &bp, aflags))
+			return (error);
+		oip->i_size = length;
+		size = blksize(fs, oip, lbn);
+		(void) vnode_pager_uncache(ITOV(oip));
+		bzero(bp->b_un.b_addr + offset, (unsigned)(size - offset));
+		allocbuf(bp, size);
+		if (flags & IO_SYNC)
+			bwrite(bp);
+		else
+			bdwrite(bp);
 	}
 	/*
-	 * release single indirect blocks second
+	 * 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 indirtrunc below.
 	 */
-	for (i = NIADDR - 2; i >= 0; i--) {
-		bn = ip->i_ib[i];
-		if (bn != (daddr_t)0) {
-			ip->i_ib[i] = (daddr_t)0;
-#ifdef QUOTA
-			cnt +=
-#endif
-				tloop(ip, bn, 0);
+	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;
+	vinvalbuf(ITOV(oip), (length > 0));
+	allerror = iupdat(oip, &time, &time, MNT_WAIT);
+
+	/*
+	 * Indirect blocks first.
+	 */
+	ip = &tip;
+	for (level = TRIPLE; level >= SINGLE; level--) {
+		bn = ip->i_ib[level];
+		if (bn != 0) {
+			error = indirtrunc(ip, bn, lastiblock[level], level,
+				&count);
+			if (error)
+				allerror = error;
+			blocksreleased += count;
+			if (lastiblock[level] < 0) {
+				ip->i_ib[level] = 0;
+				blkfree(ip, bn, (off_t)fs->fs_bsize);
+				blocksreleased += nblocks;
+			}
 		}
+		if (lastiblock[level] >= 0)
+			goto done;
 	}
+
 	/*
-	 * finally release direct blocks
+	 * All whole direct blocks or frags.
 	 */
-	for (i = NDADDR - 1; i>=0; i--) {
-		register size;
+	for (i = NDADDR - 1; i > lastblock; i--) {
+		register off_t bsize;
 
 		bn = ip->i_db[i];
-		if (bn == (daddr_t)0)
+		if (bn == 0)
 			continue;
-		ip->i_db[i] = (daddr_t)0;
-#ifndef QUOTA
-		fre(ip, bn, (off_t)blksize(fs, ip, i));
-#else
-		fre(ip, bn, size = (off_t)blksize(fs, ip, i));
-		cnt += size / DEV_BSIZE;
-#endif
+		ip->i_db[i] = 0;
+		bsize = (off_t)blksize(fs, ip, i);
+		blkfree(ip, bn, bsize);
+		blocksreleased += btodb(bsize);
 	}
-	ip->i_size = 0;
+	if (lastblock < 0)
+		goto done;
+
 	/*
-	 * Inode was written and flags updated above.
-	 * No need to modify flags here.
+	 * Finally, look for a change in size of the
+	 * last direct block; release any frags.
 	 */
+	bn = ip->i_db[lastblock];
+	if (bn != 0) {
+		off_t 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 = 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);
+			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
-	(void) chkdq(ip, -cnt, 0);
+	if (!getinoquota(oip))
+		(void) chkdq(oip, -blocksreleased, NOCRED, 0);
 #endif
+	return (allerror);
 }
 
-#ifdef QUOTA
-long
-#endif
-tloop(ip, bn, indflg)
+/*
+ * 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.
+ */
+indirtrunc(ip, bn, lastbn, level, countp)
 	register struct inode *ip;
-	daddr_t bn;
-	int indflg;
+	daddr_t bn, lastbn;
+	int level;
+	long *countp;
 {
-	register i;
-	register struct buf *bp;
+	register int i;
+	struct buf *bp;
+	register struct fs *fs = ip->i_fs;
 	register daddr_t *bap;
-	register struct fs *fs;
-	daddr_t nb;
-#ifdef QUOTA
-	register long cnt = 0;
-#endif
+	daddr_t *copy, nb, last;
+	long blkcount, factor;
+	int nblocks, blocksreleased = 0;
+	int error, allerror = 0;
 
-	bp = NULL;
-	fs = ip->i_fs;
-	for (i = NINDIR(fs) - 1; i >= 0; i--) {
-		if (bp == NULL) {
-			bp = bread(ip->i_dev, fsbtodb(fs, bn), fs->fs_bsize);
-			if (bp->b_flags & B_ERROR) {
-				brelse(bp);
-				return;
-			}
-			bap = bp->b_un.b_daddr;
-		}
+	/*
+	 * 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.
+	 */
+#ifdef SECSIZE
+	bp = bread(ip->i_dev, fsbtodb(fs, bn), (int)fs->fs_bsize,
+	    fs->fs_dbsize);
+#else SECSIZE
+	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 == (daddr_t)0)
+		if (nb == 0)
 			continue;
-		if (indflg) {
-#ifdef QUOTA
-			cnt +=
-#endif
-				tloop(ip, nb, 0);
-		} else {
-			fre(ip, nb, fs->fs_bsize);
-#ifdef QUOTA
-			cnt += fs->fs_bsize / DEV_BSIZE;
-#endif
+		if (level > SINGLE) {
+			error = indirtrunc(ip, nb, (daddr_t)-1, level - 1,
+				&blkcount);
+			if (error)
+				allerror = error;
+			blocksreleased += blkcount;
 		}
+		blkfree(ip, nb, (off_t)fs->fs_bsize);
+		blocksreleased += nblocks;
 	}
-	if (bp != NULL)
-		brelse(bp);
-	fre(ip, bn, fs->fs_bsize);
-#ifdef QUOTA
-	cnt += fs->fs_bsize / DEV_BSIZE;
-	return(cnt);
-#endif
-}
-
-/*
- * remove any inodes in the inode cache belonging to dev
- *
- * There should not be any active ones, return error if any are found
- * (nb: this is a user error, not a system err)
- *
- * Also, count the references to dev by block devices - this really
- * has nothing to do with the object of the procedure, but as we have
- * to scan the inode table here anyway, we might as well get the
- * extra benefit.
- *
- * this is called from sumount()/sys3.c when dev is being unmounted
- */
-#ifdef QUOTA
-iflush(dev, iq)
-	dev_t dev;
-	struct inode *iq;
-#else
-iflush(dev)
-	dev_t dev;
-#endif
-{
-	register struct inode *ip;
-	register open = 0;
 
-	for (ip = inode; ip < inodeNINODE; ip++) {
-#ifdef QUOTA
-		if (ip != iq && ip->i_dev == dev)
-#else
-		if (ip->i_dev == dev)
-#endif
-			if (ip->i_count)
-				return(-1);
-			else {
-				remque(ip);
-				ip->i_forw = ip;
-				ip->i_back = ip;
-				/*
-				 * as i_count == 0, the inode was on the free
-				 * list already, just leave it there, it will
-				 * fall off the bottom eventually. We could
-				 * perhaps move it to the head of the free
-				 * list, but as umounts are done so
-				 * infrequently, we would gain very little,
-				 * while making the code bigger.
-				 */
-#ifdef QUOTA
-				dqrele(ip->i_dquot);
-				ip->i_dquot = NODQUOT;
-#endif
-			}
-		else if (ip->i_count && (ip->i_mode&IFMT)==IFBLK &&
-		    ip->i_rdev == dev)
-			open++;
+	/*
+	 * Recursively free last partial block.
+	 */
+	if (level > SINGLE && lastbn >= 0) {
+		last = lastbn % factor;
+		nb = bap[i];
+		if (nb != 0) {
+			error = indirtrunc(ip, nb, last, level - 1, &blkcount);
+			if (error)
+				allerror = error;
+			blocksreleased += blkcount;
+		}
 	}
-	return (open);
+	FREE(copy, M_TEMP);
+	*countp = blocksreleased;
+	return (allerror);
 }
 
 /*
@@ -533,7 +644,16 @@ ilock(ip)
 	register struct inode *ip;
 {
 
-	ILOCK(ip);
+	while (ip->i_flag & ILOCKED) {
+		ip->i_flag |= IWANT;
+		if (ip->i_spare0 == curproc->p_pid)
+			panic("locking against myself");
+		ip->i_spare1 = curproc->p_pid;
+		(void) sleep((caddr_t)ip, PINOD);
+	}
+	ip->i_spare1 = 0;
+	ip->i_spare0 = curproc->p_pid;
+	ip->i_flag |= ILOCKED;
 }
 
 /*
@@ -543,5 +663,12 @@ iunlock(ip)
 	register struct inode *ip;
 {
 
-	IUNLOCK(ip);
+	if ((ip->i_flag & ILOCKED) == 0)
+		vprint("iunlock: unlocked inode", ITOV(ip));
+	ip->i_spare0 = 0;
+	ip->i_flag &= ~ILOCKED;
+	if (ip->i_flag&IWANT) {
+		ip->i_flag &= ~IWANT;
+		wakeup((caddr_t)ip);
+	}
 }