add proc pointers to appropriate vnode operations
[unix-history] / usr / src / sys / ufs / ufs / ufs_inode.c
index f8d38b7..e1954e4 100644 (file)
-/*     ufs_inode.c     3.2     %H%     */
-
-#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/ino.h"
-#include "../h/filsys.h"
-#include "../h/conf.h"
-#include "../h/buf.h"
-#include "../h/inline.h"
-
-#define        INOHSZ  63
-#define        INOHASH(dev,ino)        (((dev)+(ino))%INOHSZ)
-short  inohash[INOHSZ];
-short  ifreel;
-
 /*
 /*
- * Initialize hash links for inodes
- * and build inode free list.
+ * 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%
  */
  */
-ihinit()
-{
-       register int i;
 
 
-       ifreel = 0;
-       for (i = 0; i < NINODE - 1; i++)
-               inode[i].i_hlink = i+1;
-       inode[NINODE - 1].i_hlink = -1;
-       for (i = 0; i < INOHSZ; i++)
-               inohash[i] = -1;
-}
+#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"
+
+#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)        (((unsigned)((dev)+(ino)))%INOHSZ)
+#endif
+
+union ihead {
+       union  ihead *ih_head[2];
+       struct inode *ih_chain[2];
+} ihead[INOHSZ];
+
+int prtactive; /* 1 => print out reclaim of active vnodes */
 
 /*
 
 /*
- * Find an inode if it is incore.
- * This is the equivalent, for inodes,
- * of ``incore'' in bio.c or ``pfind'' in subr.c.
+ * Initialize hash links for inodes.
  */
  */
-struct inode *
-ifind(dev, ino)
-dev_t dev;
-ino_t ino;
+ufs_init()
 {
 {
-       register struct inode *ip;
+       register int i;
+       register union ihead *ih = ihead;
 
 
-       for (ip = &inode[inohash[INOHASH(dev,ino)]]; ip != &inode[-1];
-           ip = &inode[ip->i_hlink])
-               if (ino==ip->i_number && dev==ip->i_dev)
-                       return (ip);
-       return ((struct inode *)0);
+#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;
+       }
+#ifdef QUOTA
+       dqinit();
+#endif /* QUOTA */
 }
 
 /*
 }
 
 /*
- * 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 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.
  * In all cases, a pointer to a locked
  * inode structure is returned.
- *
- * printf warning: no inodes -- if the inode
- *     structure is full
- * panic: no imt -- if the mounted file
- *     system is not in the mount table.
- *     "cannot happen"
  */
  */
-struct inode *
-iget(dev, ino)
-dev_t dev;
-ino_t ino;
+iget(xp, ino, ipp)
+       struct inode *xp;
+       ino_t ino;
+       struct inode **ipp;
 {
 {
-       register struct inode *ip;
-       register struct mount *mp;
-       register struct buf *bp;
-       register struct dinode *dp;
-       register int slot;
+       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;
 
 
+       ih = &ihead[INOHASH(dev, ino)];
 loop:
 loop:
-       slot = INOHASH(dev, ino);
-       ip = &inode[inohash[slot]];
-       while (ip != &inode[-1]) {
-               if(ino == ip->i_number && dev == ip->i_dev) {
-                       if((ip->i_flag&ILOCK) != 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;
-                                       ino = ROOTINO;
-                                       goto loop;
-                               }
-                               panic("no imt");
-                       }
-                       ip->i_count++;
-                       ip->i_flag |= ILOCK;
-                       return(ip);
+       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;
                }
                }
-               ip = &inode[ip->i_hlink];
+               if (vget(ITOV(ip)))
+                       goto loop;
+               *ipp = ip;
+               return(0);
        }
        }
-       if(ifreel < 0) {
-               printf("Inode table overflow\n");
-               u.u_error = ENFILE;
-               return(NULL);
+       /*
+        * Allocate a new inode.
+        */
+       if (error = getnewvnode(VT_UFS, mntp, &ufs_vnodeops, &nvp)) {
+               *ipp = 0;
+               return (error);
        }
        }
-       ip = &inode[ifreel];
-       ifreel = ip->i_hlink;
-       ip->i_hlink = inohash[slot];
-       inohash[slot] = ip - inode;
+       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
+       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_number = ino;
        ip->i_dev = dev;
        ip->i_number = ino;
-       ip->i_flag = ILOCK;
-       ip->i_count++;
-       ip->i_un.i_lastr = 0;
-       bp = bread(dev, itod(ino));
+       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 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.
+                */
+               remque(ip);
+               ip->i_forw = ip;
+               ip->i_back = ip;
+               /*
+                * Unlock and discard unneeded inode.
+                */
                iput(ip);
                iput(ip);
-               return(NULL);
+               brelse(bp);
+               *ipp = 0;
+               return (error);
        }
        dp = bp->b_un.b_dino;
        }
        dp = bp->b_un.b_dino;
-       dp += itoo(ino);
-       iexpand(ip, dp);
+       dp += itoo(fs, ino);
+       ip->i_din = *dp;
        brelse(bp);
        brelse(bp);
-       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);
 }
 
 }
 
-iexpand(ip, dp)
-register struct inode *ip;
-register struct dinode *dp;
+/*
+ * Unlock and decrement the reference count of an inode structure.
+ */
+iput(ip)
+       register struct inode *ip;
 {
 {
-       register char *p1, *p2;
-       register int i;
 
 
-       ip->i_mode = dp->di_mode;
-       ip->i_nlink = dp->di_nlink;
-       ip->i_uid = dp->di_uid;
-       ip->i_gid = dp->di_gid;
-       ip->i_size = dp->di_size;
-       p1 = (char *)ip->i_un.i_addr;
-       p2 = (char *)dp->di_addr;
-       for(i=0; i<NADDR; i++) {
-               *p1++ = *p2++;
-               *p1++ = *p2++;
-               *p1++ = *p2++;
-               *p1++ = 0;
-       }
+       if ((ip->i_flag & ILOCKED) == 0)
+               panic("iput");
+       IUNLOCK(ip);
+       vrele(ITOV(ip));
 }
 
 /*
 }
 
 /*
- * Decrement reference count of
- * an inode structure.
- * On the last reference,
- * write the inode out and if necessary,
+ * Last reference to an inode, write the inode out and if necessary,
  * truncate and deallocate the file.
  */
  * truncate and deallocate the file.
  */
-iput(ip)
-register struct inode *ip;
+ufs_inactive(vp, p)
+       struct vnode *vp;
+       struct proc *p;
 {
 {
-       register int i, x;
-       register struct inode *jp;
+       register struct inode *ip = VTOI(vp);
+       int mode, error = 0;
 
 
-       if(ip->i_count == 1) {
-               ip->i_flag |= ILOCK;
-               if(ip->i_nlink <= 0) {
-                       itrunc(ip);
-                       ip->i_mode = 0;
-                       ip->i_flag |= IUPD|ICHG;
-                       ifree(ip->i_dev, ip->i_number);
-               }
-               IUPDAT(ip, &time, &time);
-               prele(ip);
-               i = INOHASH(ip->i_dev, ip->i_number);
-               x = ip - inode;
-               if (inohash[i] == x) {
-                       inohash[i] = ip->i_hlink;
-               } else {
-                       for (jp = &inode[inohash[i]]; jp != &inode[-1];
-                           jp = &inode[jp->i_hlink])
-                               if (jp->i_hlink == x) {
-                                       jp->i_hlink = ip->i_hlink;
-                                       goto done;
-                               }
-                       panic("iput");
+       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
+               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;
                }
                }
-done:
-               ip->i_hlink = ifreel;
-               ifreel = x;
-               ip->i_flag = 0;
-               ip->i_number = 0;
-       } else
-               prele(ip);
-       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.
+ * 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)
-register struct inode *ip;
-time_t *ta, *tm;
+iupdat(ip, ta, tm, waitfor)
+       register struct inode *ip;
+       struct timeval *ta, *tm;
+       int waitfor;
 {
 {
-       register struct buf *bp;
+       struct buf *bp;
+       struct vnode *vp = ITOV(ip);
        struct dinode *dp;
        struct dinode *dp;
-       register char *p1, *p2;
-       register int i;
+       register struct fs *fs;
 
 
-       if((ip->i_flag&(IUPD|IACC|ICHG)) != 0) {
-               if(getfs(ip->i_dev)->s_ronly)
-                       return;
-               bp = bread(ip->i_dev, itod(ip->i_number));
-               if (bp->b_flags & B_ERROR) {
-                       brelse(bp);
-                       return;
-               }
-               dp = bp->b_un.b_dino;
-               dp += itoo(ip->i_number);
-               dp->di_mode = ip->i_mode;
-               dp->di_nlink = ip->i_nlink;
-               dp->di_uid = ip->i_uid;
-               dp->di_gid = ip->i_gid;
-               dp->di_size = ip->i_size;
-               p1 = (char *)dp->di_addr;
-               p2 = (char *)ip->i_un.i_addr;
-               for(i=0; i<NADDR; i++) {
-                       *p1++ = *p2++;
-                       *p1++ = *p2++;
-                       *p1++ = *p2++;
-                       if(*p2++ != 0 && (ip->i_mode&IFMT)!=IFMPC
-                          && (ip->i_mode&IFMT)!=IFMPB)
-                               printf("iaddress > 2^24\n");
-               }
-               if(ip->i_flag&IACC)
-                       dp->di_atime = *ta;
-               if(ip->i_flag&IUPD)
-                       dp->di_mtime = *tm;
-               if(ip->i_flag&ICHG)
-                       dp->di_ctime = time;
-               ip->i_flag &= ~(IUPD|IACC|ICHG);
+       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);
                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 */
 /*
 /*
- * Free all the disk blocks associated
- * with the specified inode structure.
- * The blocks of the file are removed
- * in reverse order. This FILO
- * algorithm will tend to maintain
- * a contiguous free list much longer
- * than FIFO.
+ * 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)
-register struct inode *ip;
+itrunc(oip, length, flags)
+       register struct inode *oip;
+       u_long length;
+       int flags;
 {
 {
-       register i;
-       dev_t dev;
-       daddr_t bn;
-
-       if (ip->i_vfdcnt)
-               panic("itrunc");
-       i = ip->i_mode & IFMT;
-       if (i!=IFREG && i!=IFDIR)
-               return;
-       dev = ip->i_dev;
-       for(i=NADDR-1; i>=0; i--) {
-               bn = ip->i_un.i_addr[i];
-               if(bn == (daddr_t)0)
-                       continue;
-               ip->i_un.i_addr[i] = (daddr_t)0;
-               switch(i) {
+       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, 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);
+       }
+       /*
+        * Calculate index into inode's block list of
+        * last direct and indirect blocks (if any)
+        * which we want to keep.  Lastblock is -1 when
+        * the file is truncated to 0.
+        */
+       fs = oip->i_fs;
+       lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
+       lastiblock[SINGLE] = lastblock - NDADDR;
+       lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
+       lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
+       nblocks = btodb(fs->fs_bsize);
+       /*
+        * Update the size of the file. If the file is not being
+        * truncated to a block boundry, the contents of the
+        * partial block following the end of the file must be
+        * zero'ed in case it ever become accessable again because
+        * of subsequent file growth.
+        */
+       osize = oip->i_size;
+       offset = blkoff(fs, 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
+               if (error = getinoquota(oip))
+                       return (error);
+#endif
+               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);
+       }
+       /*
+        * 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.
+        */
+       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);
 
 
-               default:
-                       free(dev, bn);
-                       break;
+       /*
+        * 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;
+       }
 
 
-               case NADDR-3:
-                       tloop(dev, bn, 0, 0);
-                       break;
+       /*
+        * All whole direct blocks or frags.
+        */
+       for (i = NDADDR - 1; i > lastblock; i--) {
+               register off_t bsize;
 
 
-               case NADDR-2:
-                       tloop(dev, bn, 1, 0);
-                       break;
+               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 (lastblock < 0)
+               goto done;
 
 
-               case NADDR-1:
-                       tloop(dev, bn, 1, 1);
+       /*
+        * 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);
                }
        }
                }
        }
-       ip->i_size = 0;
-       ip->i_flag |= ICHG|IUPD;
+done:
+/* BEGIN PARANOIA */
+       for (level = SINGLE; level <= TRIPLE; level++)
+               if (ip->i_ib[level] != oip->i_ib[level])
+                       panic("itrunc1");
+       for (i = 0; i < NDADDR; i++)
+               if (ip->i_db[i] != oip->i_db[i])
+                       panic("itrunc2");
+/* END PARANOIA */
+       oip->i_blocks -= blocksreleased;
+       if (oip->i_blocks < 0)                  /* sanity */
+               oip->i_blocks = 0;
+       oip->i_flag |= ICHG;
+#ifdef QUOTA
+       if (!getinoquota(oip))
+               (void) chkdq(oip, -blocksreleased, NOCRED, 0);
+#endif
+       return (allerror);
 }
 
 }
 
-tloop(dev, bn, f1, f2)
-dev_t dev;
-daddr_t bn;
+/*
+ * 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, 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 daddr_t *bap;
-       daddr_t nb;
-
-       bp = NULL;
-       for(i=NINDIR-1; i>=0; i--) {
-               if(bp == NULL) {
-                       bp = bread(dev, bn);
-                       if (bp->b_flags & B_ERROR) {
-                               brelse(bp);
-                               return;
-                       }
-                       bap = bp->b_un.b_daddr;
-               }
+       daddr_t *copy, nb, last;
+       long blkcount, factor;
+       int nblocks, blocksreleased = 0;
+       int error, allerror = 0;
+
+       /*
+        * Calculate index in current block of last
+        * block to be kept.  -1 indicates the entire
+        * block so we need not calculate the index.
+        */
+       factor = 1;
+       for (i = SINGLE; i < level; i++)
+               factor *= NINDIR(fs);
+       last = lastbn;
+       if (lastbn > 0)
+               last /= factor;
+       nblocks = btodb(fs->fs_bsize);
+       /*
+        * Get buffer of block pointers, zero those 
+        * entries corresponding to blocks to be free'd,
+        * and update on disk copy first.
+        */
+#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];
                nb = bap[i];
-               if(nb == (daddr_t)0)
+               if (nb == 0)
                        continue;
                        continue;
-               if(f1) {
-                       brelse(bp);
-                       bp = NULL;
-                       tloop(dev, nb, f2, 0);
-               } else
-                       free(dev, nb);
+               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);
-       free(dev, bn);
+
+       /*
+        * 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;
+               }
+       }
+       FREE(copy, M_TEMP);
+       *countp = blocksreleased;
+       return (allerror);
 }
 
 /*
 }
 
 /*
- * Make a new file.
+ * Lock an inode. If its already locked, set the WANT bit and sleep.
  */
  */
-struct inode *
-maknode(mode)
-{
+ilock(ip)
        register struct inode *ip;
        register struct inode *ip;
+{
 
 
-       ip = ialloc(u.u_pdir->i_dev);
-       if(ip == NULL) {
-               iput(u.u_pdir);
-               return(NULL);
+       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_flag |= IACC|IUPD|ICHG;
-       if((mode&IFMT) == 0)
-               mode |= IFREG;
-       ip->i_mode = mode & ~u.u_cmask;
-       ip->i_nlink = 1;
-       ip->i_uid = u.u_uid;
-       ip->i_gid = u.u_gid;
-       wdir(ip);
-       return(ip);
+       ip->i_spare1 = 0;
+       ip->i_spare0 = curproc->p_pid;
+       ip->i_flag |= ILOCKED;
 }
 
 /*
 }
 
 /*
- * Write a directory entry with
- * parameters left as side effects
- * to a call to namei.
+ * Unlock an inode.  If WANT bit is on, wakeup.
  */
  */
-wdir(ip)
-struct inode *ip;
+iunlock(ip)
+       register struct inode *ip;
 {
 
 {
 
-       u.u_dent.d_ino = ip->i_number;
-       bcopy((caddr_t)u.u_dbuf, (caddr_t)u.u_dent.d_name, DIRSIZ);
-       u.u_count = sizeof(struct direct);
-       u.u_segflg = 1;
-       u.u_base = (caddr_t)&u.u_dent;
-       writei(u.u_pdir);
-       iput(u.u_pdir);
+       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);
+       }
 }
 }