BSD 4_4_Lite1 release
[unix-history] / usr / src / sys / ufs / lfs / lfs_inode.c
index 5d81a96..1a06aa2 100644 (file)
 /*
 /*
- * Copyright (c) 1982, 1986 Regents of the University of California.
- * All rights reserved.  The Berkeley software License Agreement
- * specifies the terms and conditions for redistribution.
+ * Copyright (c) 1986, 1989, 1991, 1993
+ *     The Regents of the University of California.  All rights reserved.
  *
  *
- *     @(#)lfs_inode.c 7.5 (Berkeley) %G%
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *     This product includes software developed by the University of
+ *     California, Berkeley and its contributors.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *     @(#)lfs_inode.c 8.5 (Berkeley) 12/30/93
  */
 
  */
 
-#include "param.h"
-#include "systm.h"
-#include "mount.h"
-#include "dir.h"
-#include "user.h"
-#include "inode.h"
-#include "fs.h"
-#include "buf.h"
-#include "cmap.h"
-#ifdef QUOTA
-#include "quota.h"
-#endif
-#include "kernel.h"
-#include "malloc.h"
+#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>
 
 
-#define        INOHSZ  512
-#if    ((INOHSZ&(INOHSZ-1)) == 0)
-#define        INOHASH(dev,ino)        (((dev)+(ino))&(INOHSZ-1))
-#else
-#define        INOHASH(dev,ino)        (((unsigned)((dev)+(ino)))%INOHSZ)
-#endif
+#include <vm/vm.h>
 
 
-union ihead {                          /* inode LRU cache, Chris Maltby */
-       union  ihead *ih_head[2];
-       struct inode *ih_chain[2];
-} ihead[INOHSZ];
+#include <ufs/ufs/quota.h>
+#include <ufs/ufs/inode.h>
+#include <ufs/ufs/ufsmount.h>
+#include <ufs/ufs/ufs_extern.h>
 
 
-struct inode *ifreeh, **ifreet;
+#include <ufs/lfs/lfs.h>
+#include <ufs/lfs/lfs_extern.h>
 
 
-/*
- * Initialize hash links for inodes
- * and build inode free list.
- */
-ihinit()
+int
+lfs_init()
 {
 {
-       register int i;
-       register struct inode *ip = inode;
-       register union  ihead *ih = ihead;
-
-       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;
+       return (ufs_init());
 }
 
 }
 
-#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;
+/* Search a block for a specific dinode. */
+struct dinode *
+lfs_ifind(fs, ino, dip)
+       struct lfs *fs;
        ino_t ino;
        ino_t ino;
+       register struct dinode *dip;
 {
 {
-       register struct inode *ip;
-       register union  ihead *ih;
+       register int cnt;
+       register struct dinode *ldip;
 
 
-       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);
-}
-#endif notdef
+       for (cnt = INOPB(fs), ldip = dip + (cnt - 1); cnt--; --ldip)
+               if (ldip->di_inumber == ino)
+                       return (ldip);
 
 
-/*
- * Look up an 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.
- * 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;
-       ino_t ino;
-{
-       register struct inode *ip;
-       register union  ihead *ih;
-       register struct mount *mp;
-       register struct buf *bp;
-       register struct dinode *dp;
-       register struct inode *iq;
-
-loop:
-       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) {
-                       /*
-                        * Following is essentially an inline expanded
-                        * copy of igrab(), expanded inline for speed,
-                        * and so that the test for a mounted on inode
-                        * can be deferred until after we are sure that
-                        * the inode isn't busy.
-                        */
-                       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_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);
-               }
-
-       if ((ip = ifreeh) == NULL) {
-               tablefull("inode");
-               u.u_error = ENFILE;
-               return(NULL);
-       }
-       if (ip->i_count)
-               panic("free inode isn't");
-       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
-        */
-       remque(ip);
-       insque(ip, ih);
-       ip->i_dev = dev;
-       ip->i_fs = fs;
-       ip->i_number = ino;
-       cacheinval(ip);
-       ip->i_flag = ILOCKED;
-       ip->i_count++;
-       ip->i_lastr = 0;
-#ifdef QUOTA
-       dqrele(ip->i_dquot);
-#endif
-#ifdef SECSIZE
-       bp = bread(dev, fsbtodb(fs, itod(fs, ino)), (int)fs->fs_bsize,
-           fs->fs_dbsize);
-#else SECSIZE
-       bp = bread(dev, fsbtodb(fs, itod(fs, ino)), (int)fs->fs_bsize);
-#endif SECSIZE
-       /*
-        * Check I/O errors
-        */
-       if ((bp->b_flags&B_ERROR) != 0) {
-               brelse(bp);
-               /*
-                * the inode doesn't 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.
-                */
-               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)
-                */
-               ip->i_number = 0;
-#ifdef QUOTA
-               ip->i_dquot = NODQUOT;
-#endif
-               iput(ip);
-               return(NULL);
-       }
-       dp = bp->b_un.b_dino;
-       dp += itoo(fs, ino);
-       ip->i_ic = dp->di_ic;
-       brelse(bp);
-#ifdef QUOTA
-       if (ip->i_mode == 0)
-               ip->i_dquot = NODQUOT;
-       else
-               ip->i_dquot = inoquota(ip);
-#endif
-       return (ip);
+       panic("lfs_ifind: dinode %u not found", ino);
+       /* NOTREACHED */
 }
 
 }
 
-/*
- * Convert a pointer to an inode into a reference to an inode.
- *
- * This is basically the internal piece of iget (after the
- * inode pointer is located) but without the test for mounted
- * filesystems.  It is caller's responsibility to check that
- * the inode pointer is valid.
- */
-igrab(ip)
-       register struct inode *ip;
+int
+lfs_update(ap)
+       struct vop_update_args /* {
+               struct vnode *a_vp;
+               struct timeval *a_access;
+               struct timeval *a_modify;
+               int a_waitfor;
+       } */ *ap;
 {
 {
-       while ((ip->i_flag&ILOCKED) != 0) {
-               ip->i_flag |= IWANT;
-               sleep((caddr_t)ip, PINOD);
-       }
-       if (ip->i_count == 0) {         /* ino on free list */
-               register struct inode *iq;
+       struct vnode *vp = ap->a_vp;
+       struct inode *ip;
 
 
-               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;
+       if (vp->v_mount->mnt_flag & MNT_RDONLY)
+               return (0);
+       ip = VTOI(vp);
+       if ((ip->i_flag &
+           (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0)
+               return (0);
+       if (ip->i_flag & IN_ACCESS)
+               ip->i_atime.ts_sec = ap->a_access->tv_sec;
+       if (ip->i_flag & IN_UPDATE) {
+               ip->i_mtime.ts_sec = ap->a_modify->tv_sec;
+               (ip)->i_modrev++;
        }
        }
-       ip->i_count++;
-       ip->i_flag |= ILOCKED;
-}
+       if (ip->i_flag & IN_CHANGE)
+               ip->i_ctime.ts_sec = time.tv_sec;
+       ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
 
 
-/*
- * Decrement reference count of
- * an inode structure.
- * On the last reference,
- * write the inode out and if necessary,
- * truncate and deallocate the file.
- */
-iput(ip)
-       register struct inode *ip;
-{
+       if (!(ip->i_flag & IN_MODIFIED))
+               ++(VFSTOUFS(vp->v_mount)->um_lfs->lfs_uinodes);
+       ip->i_flag |= IN_MODIFIED;
 
 
-       if ((ip->i_flag & ILOCKED) == 0)
-               panic("iput");
-       IUNLOCK(ip);
-       irele(ip);
+       /* If sync, push back the vnode and any dirty blocks it may have. */
+       return (ap->a_waitfor & LFS_SYNC ? lfs_vflush(vp) : 0);
 }
 
 }
 
-irele(ip)
-       register struct inode *ip;
-{
-       int mode;
+/* Update segment usage information when removing a block. */
+#define UPDATE_SEGUSE \
+       if (lastseg != -1) { \
+               LFS_SEGENTRY(sup, fs, lastseg, sup_bp); \
+               if ((num << fs->lfs_bshift) > sup->su_nbytes) \
+                       panic("lfs_truncate: negative bytes in segment %d\n", \
+                           lastseg); \
+               sup->su_nbytes -= num << fs->lfs_bshift; \
+               e1 = VOP_BWRITE(sup_bp); \
+               blocksreleased += num; \
+       }
 
 
-       if (ip->i_count == 1) {
-               ip->i_flag |= ILOCKED;
-               if (ip->i_nlink <= 0 && ip->i_fs->fs_ronly == 0) {
-                       itrunc(ip, (u_long)0);
-                       mode = ip->i_mode;
-                       ip->i_mode = 0;
-                       ip->i_rdev = 0;
-                       ip->i_flag |= IUPD|ICHG;
-                       ifree(ip, ip->i_number, mode);
-#ifdef QUOTA
-                       (void) chkiq(ip->i_dev, ip, ip->i_uid, 0);
-                       dqrele(ip->i_dquot);
-                       ip->i_dquot = NODQUOT;
-#endif
-               }
-               IUPDAT(ip, &time, &time, 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;
-       } else if (!(ip->i_flag & ILOCKED))
-               ITIMES(ip, &time, &time);
-       ip->i_count--;
+#define SEGDEC { \
+       if (daddr != 0) { \
+               if (lastseg != (seg = datosn(fs, daddr))) { \
+                       UPDATE_SEGUSE; \
+                       num = 1; \
+                       lastseg = seg; \
+               } else \
+                       ++num; \
+       } \
 }
 
 /*
 }
 
 /*
- * 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.
+ * Truncate the inode ip to at most length size.  Update segment usage
+ * table information.
  */
  */
-iupdat(ip, ta, tm, waitfor)
-       register struct inode *ip;
-       struct timeval *ta, *tm;
-       int waitfor;
+/* ARGSUSED */
+int
+lfs_truncate(ap)
+       struct vop_truncate_args /* {
+               struct vnode *a_vp;
+               off_t a_length;
+               int a_flags;
+               struct ucred *a_cred;
+               struct proc *a_p;
+       } */ *ap;
 {
 {
-       register struct buf *bp;
-       struct dinode *dp;
-       register struct fs *fs;
-
-       fs = ip->i_fs;
-       if ((ip->i_flag & (IUPD|IACC|ICHG|IMOD)) != 0) {
-               if (fs->fs_ronly)
-                       return;
-#ifdef SECSIZE
-               bp = bread(ip->i_dev, fsbtodb(fs, itod(fs, ip->i_number)),
-                       (int)fs->fs_bsize, fs->fs_dbsize);
-#else SECSIZE
-               bp = bread(ip->i_dev, fsbtodb(fs, itod(fs, ip->i_number)),
-                       (int)fs->fs_bsize);
-#endif SECSIZE
-               if (bp->b_flags & B_ERROR) {
-                       brelse(bp);
-                       return;
-               }
-               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->di_ic = ip->i_ic;
-               if (waitfor)
-                       bwrite(bp);
-               else
-                       bdwrite(bp);
+       register struct indir *inp;
+       register int i;
+       register daddr_t *daddrp;
+       register struct vnode *vp = ap->a_vp;
+       off_t length = ap->a_length;
+       struct buf *bp, *sup_bp;
+       struct timeval tv;
+       struct ifile *ifp;
+       struct inode *ip;
+       struct lfs *fs;
+       struct indir a[NIADDR + 2], a_end[NIADDR + 2];
+       SEGUSE *sup;
+       daddr_t daddr, lastblock, lbn, olastblock;
+       long off, a_released, blocksreleased, i_released;
+       int e1, e2, depth, lastseg, num, offset, seg, size;
+
+       ip = VTOI(vp);
+       tv = time;
+       if (vp->v_type == VLNK && vp->v_mount->mnt_maxsymlinklen > 0) {
+#ifdef DIAGNOSTIC
+               if (length != 0)
+                       panic("lfs_truncate: partial truncate of symlink");
+#endif
+               bzero((char *)&ip->i_shortlink, (u_int)ip->i_size);
+               ip->i_size = 0;
+               ip->i_flag |= IN_CHANGE | IN_UPDATE;
+               return (VOP_UPDATE(vp, &tv, &tv, 0));
        }
        }
-}
+       vnode_pager_setsize(vp, (u_long)length);
 
 
-#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.
- */
-itrunc(oip, length)
-       register struct inode *oip;
-       u_long length;
-{
-       register daddr_t lastblock;
-       daddr_t bn, lbn, lastiblock[NIADDR];
-       register struct fs *fs;
-       register struct inode *ip;
-       struct buf *bp;
-       int offset, osize, size, count, level;
-       long nblocks, blocksreleased = 0;
-       register int i;
-       dev_t dev;
-       struct inode tip;
-       extern long indirtrunc();
+       fs = ip->i_lfs;
 
 
-       if (oip->i_size <= length) {
-               oip->i_flag |= ICHG|IUPD;
-               iupdat(oip, &time, &time, 1);
-               return;
+       /* If length is larger than the file, just update the times. */
+       if (ip->i_size <= length) {
+               ip->i_flag |= IN_CHANGE | IN_UPDATE;
+               return (VOP_UPDATE(vp, &tv, &tv, 0));
        }
        }
+
        /*
        /*
-        * 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.
+        * Calculate index into inode's block list of last direct and indirect
+        * blocks (if any) which we want to keep.  Lastblock is 0 when the
+        * file is truncated to 0.
         */
         */
-       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);
+       lastblock = lblkno(fs, length + fs->lfs_bsize - 1);
+       olastblock = lblkno(fs, ip->i_size + fs->lfs_bsize - 1) - 1;
+
        /*
        /*
-        * 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.
+        * 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, length);
        offset = blkoff(fs, length);
-       if (offset == 0) {
-               oip->i_size = length;
-       else {
+       if (offset == 0)
+               ip->i_size = length;
+       else {
                lbn = lblkno(fs, length);
                lbn = lblkno(fs, length);
-               bn = fsbtodb(fs, bmap(oip, lbn, B_WRITE, offset));
-               if (u.u_error || (long)bn < 0)
-                       return;
-               oip->i_size = length;
-               size = blksize(fs, oip, lbn);
-               count = howmany(size, CLBYTES);
-               dev = oip->i_dev;
-               for (i = 0; i < count; i++)
-#ifdef SECSIZE
-                       munhash(dev, bn + i * CLBYTES / fs->fs_dbsize);
-#else SECSIZE
-                       munhash(dev, bn + i * CLBYTES / DEV_BSIZE);
-#endif SECSIZE
-               bp = bread(dev, bn, size);
-               if (bp->b_flags & B_ERROR) {
-                       u.u_error = EIO;
-                       oip->i_size = osize;
-                       brelse(bp);
-                       return;
-               }
-               bzero(bp->b_un.b_addr + offset, (unsigned)(size - offset));
-               bdwrite(bp);
+#ifdef QUOTA
+               if (e1 = getinoquota(ip))
+                       return (e1);
+#endif 
+               if (e1 = bread(vp, lbn, fs->lfs_bsize, NOCRED, &bp))
+                       return (e1);
+               ip->i_size = length;
+               size = blksize(fs);
+               (void)vnode_pager_uncache(vp);
+               bzero((char *)bp->b_data + offset, (u_int)(size - offset));
+               allocbuf(bp, size);
+               if (e1 = VOP_BWRITE(bp))
+                       return (e1);
        }
        /*
        }
        /*
-        * 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.
+        * Modify sup->su_nbyte counters for each deleted block; keep track
+        * of number of blocks removed for ip->i_blocks.
         */
         */
-       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;
-       syncip(oip);
-
-       /*
-        * Indirect blocks first.
-        */
-       ip = &tip;
-       for (level = TRIPLE; level >= SINGLE; level--) {
-               bn = ip->i_ib[level];
-               if (bn != 0) {
-                       blocksreleased +=
-                           indirtrunc(ip, bn, lastiblock[level], level);
-                       if (lastiblock[level] < 0) {
-                               ip->i_ib[level] = 0;
-                               blkfree(ip, bn, (off_t)fs->fs_bsize);
-                               blocksreleased += nblocks;
+       blocksreleased = 0;
+       num = 0;
+       lastseg = -1;
+
+       for (lbn = olastblock; lbn >= lastblock;) {
+               /* XXX use run length from bmap array to make this faster */
+               ufs_bmaparray(vp, lbn, &daddr, a, &depth, NULL);
+               if (lbn == olastblock)
+                       for (i = NIADDR + 2; i--;)
+                               a_end[i] = a[i];
+               switch (depth) {
+               case 0:                         /* Direct block. */
+                       daddr = ip->i_db[lbn];
+                       SEGDEC;
+                       ip->i_db[lbn] = 0;
+                       --lbn;
+                       break;
+#ifdef DIAGNOSTIC
+               case 1:                         /* An indirect block. */
+                       panic("lfs_truncate: ufs_bmaparray returned depth 1");
+                       /* NOTREACHED */
+#endif
+               default:                        /* Chain of indirect blocks. */
+                       inp = a + --depth;
+                       if (inp->in_off > 0 && lbn != lastblock) {
+                               lbn -= inp->in_off < lbn - lastblock ?
+                                   inp->in_off : lbn - lastblock;
+                               break;
+                       }
+                       for (; depth && (inp->in_off == 0 || lbn == lastblock);
+                           --inp, --depth) {
+                               if (bread(vp,
+                                   inp->in_lbn, fs->lfs_bsize, NOCRED, &bp))
+                                       panic("lfs_truncate: bread bno %d",
+                                           inp->in_lbn);
+                               daddrp = (daddr_t *)bp->b_data + inp->in_off;
+                               for (i = inp->in_off;
+                                   i++ <= a_end[depth].in_off;) {
+                                       daddr = *daddrp++;
+                                       SEGDEC;
+                               }
+                               a_end[depth].in_off = NINDIR(fs) - 1;
+                               if (inp->in_off == 0)
+                                       brelse (bp);
+                               else {
+                                       bzero((daddr_t *)bp->b_data +
+                                           inp->in_off, fs->lfs_bsize - 
+                                           inp->in_off * sizeof(daddr_t));
+                                       if (e1 = VOP_BWRITE(bp)) 
+                                               return (e1);
+                               }
+                       }
+                       if (depth == 0 && a[1].in_off == 0) {
+                               off = a[0].in_off;
+                               daddr = ip->i_ib[off];
+                               SEGDEC;
+                               ip->i_ib[off] = 0;
+                       }
+                       if (lbn == lastblock || lbn <= NDADDR)
+                               --lbn;
+                       else {
+                               lbn -= NINDIR(fs);
+                               if (lbn < lastblock)
+                                       lbn = lastblock;
                        }
                }
                        }
                }
-               if (lastiblock[level] >= 0)
-                       goto done;
        }
        }
+       UPDATE_SEGUSE;
 
 
-       /*
-        * All whole direct blocks or frags.
-        */
-       for (i = NDADDR - 1; i > lastblock; i--) {
-               register off_t bsize;
-
-               bn = ip->i_db[i];
-               if (bn == 0)
-                       continue;
-               ip->i_db[i] = 0;
-               bsize = (off_t)blksize(fs, ip, i);
-               blkfree(ip, bn, bsize);
-               blocksreleased += btodb(bsize);
+       /* If truncating the file to 0, update the version number. */
+       if (length == 0) {
+               LFS_IENTRY(ifp, fs, ip->i_number, bp);
+               ++ifp->if_version;
+               (void) VOP_BWRITE(bp);
        }
        }
-       if (lastblock < 0)
-               goto done;
 
 
+#ifdef DIAGNOSTIC
+       if (ip->i_blocks < fsbtodb(fs, blocksreleased)) {
+               printf("lfs_truncate: block count < 0\n");
+               blocksreleased = ip->i_blocks;
+       }
+#endif
+       ip->i_blocks -= fsbtodb(fs, blocksreleased);
+       fs->lfs_bfree +=  fsbtodb(fs, blocksreleased);
+       ip->i_flag |= IN_CHANGE | IN_UPDATE;
        /*
        /*
-        * Finally, look for a change in size of the
-        * last direct block; release any frags.
+        * Traverse dirty block list counting number of dirty buffers
+        * that are being deleted out of the cache, so that the lfs_avail
+        * field can be updated.
         */
         */
-       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) {
+       a_released = 0;
+       i_released = 0;
+       for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = bp->b_vnbufs.le_next)
+               if (bp->b_flags & B_LOCKED) {
+                       ++a_released;
                        /*
                        /*
-                        * Block number of space to be free'd is
-                        * the old block # plus the number of frags
-                        * required for the storage we're keeping.
+                        * XXX
+                        * When buffers are created in the cache, their block
+                        * number is set equal to their logical block number.
+                        * If that is still true, we are assuming that the
+                        * blocks are new (not yet on disk) and weren't
+                        * counted above.  However, there is a slight chance
+                        * that a block's disk address is equal to its logical
+                        * block number in which case, we'll get an overcounting
+                        * here.
                         */
                         */
-                       bn += numfrags(fs, newspace);
-                       blkfree(ip, bn, oldspace - newspace);
-                       blocksreleased += btodb(oldspace - newspace);
+                       if (bp->b_blkno == bp->b_lblkno)
+                               ++i_released;
                }
                }
+       blocksreleased = fsbtodb(fs, i_released);
+#ifdef DIAGNOSTIC
+       if (blocksreleased > ip->i_blocks) {
+               printf("lfs_inode: Warning! %s\n",
+                   "more blocks released from inode than are in inode");
+               blocksreleased = ip->i_blocks;
        }
        }
-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(oip, -blocksreleased, 0);
-#endif
-}
-
-/*
- * 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.
- */
-long
-indirtrunc(ip, bn, lastbn, level)
-       register struct inode *ip;
-       daddr_t bn, lastbn;
-       int level;
-{
-       register int i;
-       struct buf *bp;
-       register struct fs *fs = ip->i_fs;
-       register daddr_t *bap;
-       daddr_t *copy, nb, last;
-       long factor;
-       int blocksreleased = 0, nblocks;
-
-       /*
-        * 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
-       bp = bread(ip->i_dev, fsbtodb(fs, bn), (int)fs->fs_bsize);
-#endif SECSIZE
-       if (bp->b_flags&B_ERROR) {
-               brelse(bp);
-               return (0);
-       }
-       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));
-       bwrite(bp);
-       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)
-                       blocksreleased +=
-                           indirtrunc(ip, nb, (daddr_t)-1, level - 1);
-               blkfree(ip, nb, (off_t)fs->fs_bsize);
-               blocksreleased += nblocks;
-       }
-
-       /*
-        * Recursively free last partial block.
-        */
-       if (level > SINGLE && lastbn >= 0) {
-               last = lastbn % factor;
-               nb = bap[i];
-               if (nb != 0)
-                       blocksreleased += indirtrunc(ip, nb, last, level - 1);
-       }
-       FREE(copy, M_TEMP);
-       return (blocksreleased);
-}
-
-/*
- * 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).
- */
-#ifdef QUOTA
-iflush(dev, iq)
-       dev_t dev;
-       struct inode *iq;
-#else
-iflush(dev)
-       dev_t dev;
-#endif
-{
-       register struct inode *ip;
-
-       for (ip = inode; ip < inodeNINODE; ip++) {
-#ifdef QUOTA
-               if (ip != iq && ip->i_dev == dev)
-#else
-               if (ip->i_dev == dev)
 #endif
 #endif
-                       if (ip->i_count)
-                               return (EBUSY);
-                       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;
+       fs->lfs_bfree += blocksreleased;
+       ip->i_blocks -= blocksreleased;
+#ifdef DIAGNOSTIC
+       if (length == 0 && ip->i_blocks != 0)
+               printf("lfs_inode: Warning! %s%d%s\n",
+                   "Truncation to zero, but ", ip->i_blocks,
+                   " blocks left on inode");
 #endif
 #endif
-                       }
-       }
-       return (0);
-}
-
-/*
- * Lock an inode. If its already locked, set the WANT bit and sleep.
- */
-ilock(ip)
-       register struct inode *ip;
-{
-
-       ILOCK(ip);
-}
-
-/*
- * Unlock an inode.  If WANT bit is on, wakeup.
- */
-iunlock(ip)
-       register struct inode *ip;
-{
-
-       IUNLOCK(ip);
+       fs->lfs_avail += fsbtodb(fs, a_released);
+       e1 = vinvalbuf(vp, (length > 0) ? V_SAVE : 0, ap->a_cred, ap->a_p,
+           0, 0); 
+       e2 = VOP_UPDATE(vp, &tv, &tv, 0);
+       return (e1 ? e1 : e2 ? e2 : 0);
 }
 }