cleanups and bug fixes from John
[unix-history] / usr / src / sys / ufs / ffs / ffs_inode.c
index c7adeef..df2c16e 100644 (file)
-/*     ffs_inode.c     4.29    82/10/19        */
+/*
+ * Copyright (c) 1982, 1986, 1989 Regents of the University of California.
+ * All rights reserved.
+ *
+ * %sccs.include.redist.c%
+ *
+ *     @(#)ffs_inode.c 7.45 (Berkeley) %G%
+ */
 
 
-#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"
+#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  63
-#if    ((INOHSZ&(INOHSZ-1)) == 0)
-#define        INOHASH(dev,ino)        (((dev)+(ino))&(INOHSZ-1))
-#else
-#define        INOHASH(dev,ino)        (((dev)+(ino))%INOHSZ)
-#endif
+#include <ufs/ufs/quota.h>
+#include <ufs/ufs/inode.h>
+#include <ufs/ufs/ufsmount.h>
+#include <ufs/ufs/ufs_extern.h>
 
 
-union ihead {                          /* inode LRU cache, Chris Maltby */
-       union  ihead *ih_head[2];
-       struct inode *ih_chain[2];
-} ihead[INOHSZ];
+#include <ufs/ffs/fs.h>
+#include <ufs/ffs/ffs_extern.h>
 
 
-struct inode *ifreeh, **ifreet;
+static int ffs_indirtrunc __P((struct inode *, daddr_t, daddr_t, int, long *));
 
 
-/*
- * Initialize hash links for inodes
- * and build inode free list.
- */
-ihinit()
-{
-       register int i;
-       register struct inode *ip = inode;
-       register union  ihead *ih = ihead;
+extern u_long nextgennumber;
 
 
-       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;
+int
+ffs_init()
 {
 {
-       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);
+       return (ufs_init());
 }
 }
-#endif notdef
 
 /*
 
 /*
- * 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"
+ * 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.
  */
  */
-struct inode *
-iget(dev, fs, ino)
-       dev_t dev;
-       register struct fs *fs;
+ffs_vget(mntp, ino, vpp)
+       struct mount *mntp;
        ino_t ino;
        ino_t ino;
+       struct vnode **vpp;
 {
 {
+       register struct fs *fs;
        register struct inode *ip;
        register struct inode *ip;
-       register union  ihead *ih;
-       register struct mount *mp;
-       register struct buf *bp;
-       register struct dinode *dp;
-       register struct inode *iq;
+       struct ufsmount *ump;
+       struct buf *bp;
+       struct dinode *dp;
+       struct vnode *vp;
+       union ihead *ih;
+       dev_t dev;
+       int i, type, 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);
-               }
+       ump = VFSTOUFS(mntp);
+       dev = ump->um_dev;
+       if ((*vpp = ufs_ihashget(dev, ino)) != NULL)
+               return (0);
 
 
-       if ((ip = ifreeh) == NULL) {
-               tablefull("inode");
-               u.u_error = ENFILE;
-               return(NULL);
+       /* Allocate a new vnode/inode. */
+       if (error = getnewvnode(VT_UFS, mntp, &ffs_vnodeops, &vp)) {
+               *vpp = NULL;
+               return (error);
        }
        }
-       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);
-#ifdef QUOTA
-       dqrele(ip->i_dquot);
-#endif
+       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_dev = dev;
-       ip->i_fs = fs;
        ip->i_number = ino;
        ip->i_number = ino;
-       ip->i_flag = ILOCKED;
-       ip->i_count++;
-       ip->i_lastr = 0;
-       bp = bread(dev, fsbtodb(fs, itod(fs, ino)), (int)fs->fs_bsize);
+#ifdef QUOTA
+       for (i = 0; i < MAXQUOTAS; i++)
+               ip->i_dquot[i] = NODQUOT;
+#endif
        /*
        /*
-        * Check I/O errors
+        * 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.
         */
         */
-       if ((bp->b_flags&B_ERROR) != 0) {
-               brelse(bp);
+       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, ino)),
+           (int)fs->fs_bsize, NOCRED, &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.
+                * 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;
                 */
                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);
+
+               /* Unlock and discard unneeded inode. */
+               ufs_iput(ip);
+               brelse(bp);
+               *vpp = NULL;
+               return (error);
        }
        dp = bp->b_un.b_dino;
        dp += itoo(fs, ino);
        }
        dp = bp->b_un.b_dino;
        dp += itoo(fs, ino);
-       ip->i_ic = dp->di_ic;
+       ip->i_din = *dp;
        brelse(bp);
        brelse(bp);
-#ifdef QUOTA
-       if (ip->i_mode == 0)
-               ip->i_dquot = NODQUOT;
-       else
-               ip->i_dquot = inoquota(ip);
-#endif
-       return (ip);
-}
-
-/*
- * 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 & ILOCKED) == 0)
-               panic("iput");
-       iunlock(ip);
-       irele(ip);
-}
 
 
-irele(ip)
-       register struct inode *ip;
-{
-       int mode;
-
-       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);
-#ifdef QUOTA
-                       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;
+       /*
+        * Initialize the vnode from the inode, check for aliases.
+        * Note that the underlying vnode may have changed.
+        */
+       if (error = ufs_vinit(mntp, &ffs_specops, FFS_FIFOOPS, &vp)) {
+               ufs_iput(ip);
+               *vpp = NULL;
+               return (error);
        }
        }
-       ip->i_count--;
+       /*
+        * 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;
+       }
+       *vpp = vp;
+       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.
+ * 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.
  */
  */
-iupdat(ip, ta, tm, waitfor)
-       register struct inode *ip;
+int
+ffs_update(vp, ta, tm, waitfor)
+       register struct vnode *vp;
        struct timeval *ta, *tm;
        int waitfor;
 {
        struct timeval *ta, *tm;
        int waitfor;
 {
-       register struct buf *bp;
+       struct buf *bp;
+       struct inode *ip;
        struct dinode *dp;
        struct dinode *dp;
-       register struct fs *fp;
+       register struct fs *fs;
 
 
-       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)),
-                       (int)fp->fs_bsize);
-               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);
-               dp = bp->b_un.b_dino + itoo(fp, ip->i_number);
-               dp->di_ic = ip->i_ic;
-               if (waitfor)
-                       bwrite(bp);
-               else
-                       bdwrite(bp);
+       if (vp->v_mount->mnt_flag & MNT_RDONLY)
+               return (0);
+       ip = VTOI(vp);
+       if ((ip->i_flag & (IUPD|IACC|ICHG|IMOD)) == 0)
+               return (0);
+       if (ip->i_flag&IACC)
+               ip->i_atime = ta->tv_sec;
+       if (ip->i_flag&IUPD) {
+               ip->i_mtime = tm->tv_sec;
+               INCRQUAD(ip->i_modrev);
+       }
+       if (ip->i_flag&ICHG)
+               ip->i_ctime = time.tv_sec;
+       ip->i_flag &= ~(IUPD|IACC|ICHG|IMOD);
+
+       fs = ip->i_fs;
+       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 (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;
+ffs_truncate(ovp, length, flags)
+       register struct vnode *ovp;
+       u_long length;
+       int flags;
 {
 {
-       register i;
-       daddr_t bn;
-       struct inode itmp;
+       register daddr_t lastblock;
+       register struct inode *oip;
+       daddr_t bn, lbn, lastiblock[NIADDR];
        register struct fs *fs;
        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, &time, 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(ovp, length);
+       oip = VTOI(ovp);
+       if (oip->i_size <= length) {
+               oip->i_flag |= ICHG|IUPD;
+               error = ffs_update(ovp, &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
 #ifdef QUOTA
-               cnt +=
+               if (error = getinoquota(oip))
+                       return (error);
 #endif
 #endif
-                       tloop(ip, bn, 1);
+               if (error = ffs_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 ffs_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 = ffs_update(ovp, &time, &time, MNT_WAIT);
+
+       /*
+        * 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, (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--) {
+       for (i = NDADDR - 1; i > lastblock; i--) {
+               register off_t bsize;
+
                bn = ip->i_db[i];
                bn = ip->i_db[i];
-               if (bn == (daddr_t)0)
+               if (bn == 0)
                        continue;
                        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);
+               ffs_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.
         */
         */
-#ifdef QUOTA
-       (void) chkdq(ip, -cnt, 0);
-#endif
-}
-
-#ifdef QUOTA
-long
-#endif
-tloop(ip, bn, indflg)
-       register struct inode *ip;
-       daddr_t bn;
-       int indflg;
-{
-       register i;
-       register struct buf *bp;
-       register daddr_t *bap;
-       register struct fs *fs;
-       daddr_t nb;
-#ifdef QUOTA
-       register long cnt = 0;
-#endif
+       bn = ip->i_db[lastblock];
+       if (bn != 0) {
+               off_t oldspace, newspace;
 
 
-       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),
-                               (int)fs->fs_bsize);
-                       if (bp->b_flags & B_ERROR) {
-                               brelse(bp);
-                               return;
-                       }
-                       bap = bp->b_un.b_daddr;
-               }
-               nb = bap[i];
-               if (nb == (daddr_t)0)
-                       continue;
-               if (indflg) {
-#ifdef QUOTA
-                       cnt +=
-#endif
-                               tloop(ip, nb, 0);
-               } else {
-                       fre(ip, nb, (int)fs->fs_bsize);
-#ifdef QUOTA
-                       cnt += fs->fs_bsize / DEV_BSIZE;
-#endif
+               /*
+                * 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);
+                       ffs_blkfree(ip, bn, oldspace - newspace);
+                       blocksreleased += btodb(oldspace - newspace);
                }
        }
                }
        }
-       if (bp != NULL)
-               brelse(bp);
-       fre(ip, bn, (int)fs->fs_bsize);
+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
 #ifdef QUOTA
-       cnt += fs->fs_bsize / DEV_BSIZE;
-       return(cnt);
+       if (!getinoquota(oip))
+               (void) chkdq(oip, -blocksreleased, NOCRED, 0);
 #endif
 #endif
+       return (allerror);
 }
 
 /*
 }
 
 /*
- * 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)
+ * 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.
  *
  *
- * 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
+ * NB: triple indirect blocks are untested.
  */
  */
-#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++;
-       }
-       return (open);
-}
-
-/*
- * Lock an inode. If its already locked, set the WANT bit and sleep.
- */
-ilock(ip)
+static int
+ffs_indirtrunc(ip, bn, lastbn, level, countp)
        register struct inode *ip;
        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;
 
 
-       ILOCK(ip);
-}
+       /*
+        * 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;
 
 
-/*
- * Unlock an inode.  If WANT bit is on, wakeup.
- */
-iunlock(ip)
-       register struct inode *ip;
-{
+       /*
+        * 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, (off_t)fs->fs_bsize);
+               blocksreleased += nblocks;
+       }
 
 
-       IUNLOCK(ip);
+       /*
+        * 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);
 }
 }