add proc pointers to appropriate vnode operations

[unix-history] / usr / src / sys / ufs / ufs / ufs_inode.c

diff --git a/usr/src/sys/ufs/ufs/ufs_inode.c b/usr/src/sys/ufs/ufs/ufs_inode.c
index 9eda051..e1954e4 100644 (file)
--- a/usr/src/sys/ufs/ufs/ufs_inode.c
+++ b/usr/src/sys/ufs/ufs/ufs_inode.c
@@ -1,473 +1,642 @@
-/*     ufs_inode.c     4.12    82/06/10        */
-
-#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"
-#include "../h/inline.h"
-
-#define        INOHSZ  63
-#define        INOHASH(dev,ino)        (((dev)+(ino))%INOHSZ)
-short  inohash[INOHSZ];
-short  ifreel;
+/*
+ * Copyright (c) 1982, 1986, 1989 Regents of the University of California.
+ * All rights reserved.
+ *
+ * %sccs.include.redist.c%
+ *
+ *     @(#)ufs_inode.c 7.39 (Berkeley) %G%
+ */
+
+#include "param.h"
+#include "systm.h"
+#include "mount.h"
+#include "proc.h"
+#include "file.h"
+#include "buf.h"
+#include "vnode.h"
+#include "kernel.h"
+#include "malloc.h"
+
+#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 */

 
 /*
 
 /*

- * Initialize hash links for inodes
- * and build inode free list.
+ * Initialize hash links for inodes.

  */
  */

-ihinit()
+ufs_init()

 {
        register int i;
 {
        register int i;

-       register struct inode *ip = inode;
-
-       ifreel = 0;
-       for (i = 0; i < ninode-1; i++, ip++)
-               ip->i_hlink = i+1;
-       ip->i_hlink = -1;
-       for (i = 0; i < INOHSZ; i++)
-               inohash[i] = -1;
+       register union ihead *ih = ihead;
+
+#ifndef lint
+       if (VN_MAXPRIVATE < sizeof(struct inode))
+               panic("ihinit: too small");
+#endif /* not lint */
+       for (i = INOHSZ; --i >= 0; ih++) {
+               ih->ih_head[0] = ih;
+               ih->ih_head[1] = ih;
+       }
+#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.

- *
- * panic: no imt -- if the mounted file
- *     system is not in the mount table.
- *     "cannot happen"

  */
  */

-struct inode *
-iget(dev, fs, ino)
-       dev_t dev;
-       register struct fs *fs;
+iget(xp, ino, ipp)
+       struct inode *xp;

        ino_t ino;
        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:

-       if (getfs(dev) != fs)
-               panic("iget: bad fs");
-       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;
-                                       fs = mp->m_bufp->b_un.b_fs;
-                                       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) {
-               tablefull("inode");
-               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_dev = dev;

-       ip->i_fs = fs;

        ip->i_number = ino;
        ip->i_number = ino;

-       ip->i_flag = ILOCK;
-       ip->i_count++;
-       ip->i_lastr = 0;
-       bp = bread(dev, fsbtodb(fs, itod(fs, ino)), fs->fs_bsize);
+       insque(ip, ih);
+       ILOCK(ip);

        /*
        /*

-        * Check I/O errors
+        * Read in the disk contents for the inode.

         */
         */

-       if ((bp->b_flags&B_ERROR) != 0) {
-               brelse(bp);
+       if (error = bread(VFSTOUFS(mntp)->um_devvp, fsbtodb(fs, itod(fs, ino)),
+           (int)fs->fs_bsize, NOCRED, &bp)) {
+               /*
+                * The inode 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 += 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);

-       return (ip);
+       /*
+        * Initialize the associated vnode
+        */
+       vp = ITOV(ip);
+       vp->v_type = IFTOVT(ip->i_mode);
+       if (vp->v_type == VFIFO) {
+#ifdef FIFO
+               extern struct vnodeops fifo_inodeops;
+               vp->v_op = &fifo_inodeops;
+#else
+               iput(ip);
+               *ipp = 0;
+               return (EOPNOTSUPP);
+#endif /* FIFO */
+       }
+       if (vp->v_type == VCHR || vp->v_type == VBLK) {
+               vp->v_op = &spec_inodeops;
+               if (nvp = checkalias(vp, ip->i_rdev, mntp)) {
+                       /*
+                        * Reinitialize aliased inode.
+                        */
+                       vp = nvp;
+                       iq = VTOI(vp);
+                       iq->i_vnode = vp;
+                       iq->i_flag = 0;
+                       ILOCK(iq);
+                       iq->i_din = ip->i_din;
+                       iq->i_dev = dev;
+                       iq->i_number = ino;
+                       insque(iq, ih);
+                       /*
+                        * Discard unneeded vnode
+                        */
+                       ip->i_mode = 0;
+                       iput(ip);
+                       ip = iq;
+               }
+       }
+       if (ino == ROOTINO)
+               vp->v_flag |= VROOT;
+       /*
+        * Finish inode initialization.
+        */
+       ip->i_fs = fs;
+       ip->i_devvp = VFSTOUFS(mntp)->um_devvp;
+       VREF(ip->i_devvp);
+       /*
+        * Set up a generation number for this inode if it does not
+        * already have one. This should only happen on old filesystems.
+        */
+       if (ip->i_gen == 0) {
+               if (++nextgennumber < (u_long)time.tv_sec)
+                       nextgennumber = time.tv_sec;
+               ip->i_gen = nextgennumber;
+               if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0)
+                       ip->i_flag |= IMOD;
+       }
+       *ipp = ip;
+       return (0);

 }
 
 /*
 }
 
 /*

- * Decrement reference count of
- * an inode structure.
- * On the last reference,
- * write the inode out and if necessary,
- * truncate and deallocate the file.
+ * Unlock and decrement the reference count of an inode structure.

  */
 iput(ip)
        register struct inode *ip;
 {
 
  */
 iput(ip)
        register struct inode *ip;
 {
 

-       if ((ip->i_flag & ILOCK) == 0)
+       if ((ip->i_flag & ILOCKED) == 0)

                panic("iput");
                panic("iput");

-       iunlock(ip);
-       irele(ip);
+       IUNLOCK(ip);
+       vrele(ITOV(ip));

 }
 
 }
 

-irele(ip)
-       register struct inode *ip;
+/*
+ * Last reference to an inode, write the inode out and if necessary,
+ * truncate and deallocate the file.
+ */
+ufs_inactive(vp, p)
+       struct vnode *vp;
+       struct proc *p;

 {
 {

-       register int i, x;
-       register struct inode *jp;
-       int mode;
-
-       if (ip->i_count == 1) {
-               ip->i_flag |= ILOCK;
-               if (ip->i_nlink <= 0) {
-                       itrunc(ip);
-                       mode = ip->i_mode;
-                       ip->i_mode = 0;
-                       ip->i_flag |= IUPD|ICHG;
-                       ifree(ip, ip->i_number, mode);
-               }
-               IUPDAT(ip, &time, &time, 0);
-               iunlock(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");
+       register struct inode *ip = VTOI(vp);
+       int mode, error = 0;
+
+       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;

        }
        }

-       ip->i_count--;
+#endif
+       ip->i_flag = 0;
+       return (0);

 }
 
 /*
 }
 
 /*

- * Check accessed and update flags on
- * an inode structure.
- * If any is on, update the inode
- * with the current time.
- * If waitfor is given, then must insure
- * i/o order so wait for write to complete.
+ * Check accessed and update flags on an inode structure.
+ * If any is on, update the inode with the current time.
+ * If waitfor is given, then must ensure I/O order,
+ * so wait for write to complete.

  */
 iupdat(ip, ta, tm, waitfor)
        register struct inode *ip;
  */
 iupdat(ip, ta, tm, waitfor)
        register struct inode *ip;

-       time_t *ta, *tm;
+       struct timeval *ta, *tm;

        int waitfor;
 {
        int waitfor;
 {

-       register struct buf *bp;
+       struct buf *bp;
+       struct vnode *vp = ITOV(ip);

        struct dinode *dp;
        struct dinode *dp;

-       register struct fs *fp;
-
-       fp = ip->i_fs;
-       if ((ip->i_flag & (IUPD|IACC|ICHG)) != 0) {
-               if (fp->fs_ronly)
-                       return;
-               bp = bread(ip->i_dev, fsbtodb(fp, itod(fp, ip->i_number)),
-                       fp->fs_bsize);
-               if (bp->b_flags & B_ERROR) {
-                       brelse(bp);
-                       return;
-               }
-               if (ip->i_flag&IACC)
-                       ip->i_atime = *ta;
-               if (ip->i_flag&IUPD)
-                       ip->i_mtime = *tm;
-               if (ip->i_flag&ICHG)
-                       ip->i_ctime = time;
-               ip->i_flag &= ~(IUPD|IACC|ICHG);
-               dp = bp->b_un.b_dino + itoo(fp, ip->i_number);
-               dp->di_ic = ip->i_ic;
-               if (waitfor)
-                       bwrite(bp);
-               else
-                       bdwrite(bp);
+       register struct fs *fs;
+
+       fs = ip->i_fs;
+       if ((ip->i_flag & (IUPD|IACC|ICHG|IMOD)) == 0)
+               return (0);
+       if (vp->v_mount->mnt_flag & MNT_RDONLY)
+               return (0);
+       error = bread(ip->i_devvp, fsbtodb(fs, itod(fs, ip->i_number)),
+               (int)fs->fs_bsize, NOCRED, &bp);
+       if (error) {
+               brelse(bp);
+               return (error);
+       }
+       if (ip->i_flag&IACC)
+               ip->i_atime = ta->tv_sec;
+       if (ip->i_flag&IUPD)
+               ip->i_mtime = tm->tv_sec;
+       if (ip->i_flag&ICHG)
+               ip->i_ctime = time.tv_sec;
+       ip->i_flag &= ~(IUPD|IACC|ICHG|IMOD);
+       dp = bp->b_un.b_dino + itoo(fs, ip->i_number);
+       *dp = ip->i_din;
+       if (waitfor) {
+               return (bwrite(bp));
+       } else {
+               bdwrite(bp);
+               return (0);

        }
 }
 
        }
 }
 

+#define        SINGLE  0       /* index of single indirect block */
+#define        DOUBLE  1       /* index of double indirect block */
+#define        TRIPLE  2       /* index of triple indirect block */

 /*
 /*

- * 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;
-       struct inode itmp;
+       register daddr_t lastblock;
+       daddr_t bn, lbn, lastiblock[NIADDR];

        register struct fs *fs;
        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;

 
 

-       i = ip->i_mode & IFMT;
-       if (i != IFREG && i != IFDIR && i != IFLNK)
-               return;
-       /*
-        * Clean inode on disk before freeing blocks
-        * to insure no duplicates if system crashes.
-        */
-       itmp = *ip;
-       itmp.i_size = 0;
-       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);
-
+       vnode_pager_setsize(ITOV(oip), length);
+       if (oip->i_size <= length) {
+               oip->i_flag |= ICHG|IUPD;
+               error = iupdat(oip, &time, &time, 1);
+               return (error);
+       }

        /*
        /*

-        * Now return blocks to free list... if machine
-        * crashes, they will be harmless MISSING blocks.
+        * Calculate index into inode's block list of
+        * last direct and indirect blocks (if any)
+        * which we want to keep.  Lastblock is -1 when
+        * the file is truncated to 0.

         */
         */

-       dev = ip->i_dev;
-       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;
-               tloop(ip, bn, 1);
+       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);

        }
        /*
        }
        /*

-        * release single indirect blocks second
+        * Update file and block pointers
+        * on disk before we start freeing blocks.
+        * If we crash before free'ing blocks below,
+        * the blocks will be returned to the free list.
+        * lastiblock values are also normalized to -1
+        * for calls to indirtrunc below.

         */
         */

-       for (i = NIADDR - 2; i >= 0; i--) {
-               bn = ip->i_ib[i];
-               if (bn != (daddr_t)0) {
-                       ip->i_ib[i] = (daddr_t)0;
-                       tloop(ip, bn, 0);
+       tip = *oip;
+       tip.i_size = osize;
+       for (level = TRIPLE; level >= SINGLE; level--)
+               if (lastiblock[level] < 0) {
+                       oip->i_ib[level] = 0;
+                       lastiblock[level] = -1;

                }
                }

-       }
+       for (i = NDADDR - 1; i > lastblock; i--)
+               oip->i_db[i] = 0;
+       oip->i_flag |= ICHG|IUPD;
+       vinvalbuf(ITOV(oip), (length > 0));
+       allerror = iupdat(oip, &time, &time, MNT_WAIT);
+

        /*
        /*

-        * finally release direct blocks
+        * Indirect blocks first.

         */
         */

-       for (i = NDADDR - 1; i>=0; i--) {
-               bn = ip->i_db[i];
-               if (bn == (daddr_t)0)
-                       continue;
-               ip->i_db[i] = (daddr_t)0;
-               fre(ip, bn, (off_t)blksize(fs, ip, i));
+       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;

        }
        }

-       ip->i_size = 0;
+

        /*
        /*

-        * Inode was written and flags updated above.
-        * No need to modify flags here.
+        * All whole direct blocks or frags.

         */
         */

-}
+       for (i = NDADDR - 1; i > lastblock; i--) {
+               register off_t bsize;

 
 

-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;
-
-       bp = NULL;
-       fs = ip->i_fs;
-       for (i = NINDIR(fs) - 1; i >= 0; i--) {
-               if (bp == NULL) {
-                       bp = bread(ip->i_dev, fsbtodb(fs, bn), fs->fs_bsize);
-                       if (bp->b_flags & B_ERROR) {
-                               brelse(bp);
-                               return;
-                       }
-                       bap = bp->b_un.b_daddr;
-               }
-               nb = bap[i];
-               if (nb == (daddr_t)0)
+               bn = ip->i_db[i];
+               if (bn == 0)

                        continue;
                        continue;

-               if (indflg)
-                       tloop(ip, nb, 0);
-               else
-                       fre(ip, nb, fs->fs_bsize);
-       }
-       if (bp != NULL)
-               brelse(bp);
-       fre(ip, bn, fs->fs_bsize);
-}
-
-/*
- * Make a new file.
- */
-struct inode *
-maknode(mode)
-       int mode;
-{
-       register struct inode *ip;
-       ino_t ipref;
-
-       if ((mode & IFMT) == IFDIR)
-               ipref = dirpref(u.u_pdir->i_fs);
-       else
-               ipref = u.u_pdir->i_number;
-       ip = ialloc(u.u_pdir, ipref, mode);
-       if (ip == NULL) {
-               iput(u.u_pdir);
-               return(NULL);
+               ip->i_db[i] = 0;
+               bsize = (off_t)blksize(fs, ip, i);
+               blkfree(ip, bn, bsize);
+               blocksreleased += btodb(bsize);

        }
        }

-       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_pdir->i_gid;
+       if (lastblock < 0)
+               goto done;

 
        /*
 
        /*

-        * Make sure inode goes to disk before directory entry.
+        * Finally, look for a change in size of the
+        * last direct block; release any frags.

         */
         */

-       iupdat(ip, &time, &time, 1);
-       wdir(ip);
-       if (u.u_error) {
+       bn = ip->i_db[lastblock];
+       if (bn != 0) {
+               off_t oldspace, newspace;
+

                /*
                /*

-                * write error occurred trying to update directory
-                * so must deallocate the inode
+                * Calculate amount of space we're giving
+                * back as old block size minus new block size.

                 */
                 */

-               ip->i_nlink = 0;
-               ip->i_flag |= ICHG;
-               iput(ip);
-               return(NULL);
+               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);
+               }

        }
        }

-       return(ip);
+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);

 }
 
 /*
 }
 
 /*

- * Write a directory entry with
- * parameters left as side effects
- * to a call to namei.
+ * 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.

  */
  */

-wdir(ip)
-       struct inode *ip;
+indirtrunc(ip, bn, lastbn, level, countp)
+       register struct inode *ip;
+       daddr_t bn, lastbn;
+       int level;
+       long *countp;

 {
 {

-       register struct direct *dp, *ndp;
-       struct fs *fs;
+       register int i;

        struct buf *bp;
        struct buf *bp;

-       int lbn, bn, base;
-       int loc, dsize, spccnt, newsize;
-       char *dirbuf;
+       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;

 
 

-       u.u_dent.d_ino = ip->i_number;
-       u.u_segflg = 1;
-       newsize = DIRSIZ(&u.u_dent);

        /*
        /*

-        * if u.u_count == 0, a new directory block must be allocated.
+        * Calculate index in current block of last
+        * block to be kept.  -1 indicates the entire
+        * block so we need not calculate the index.

         */
         */

-       if (u.u_count == 0) {
-               u.u_dent.d_reclen = DIRBLKSIZ;
-               u.u_count = newsize;
-               u.u_base = (caddr_t)&u.u_dent;
-               writei(u.u_pdir);
-               iput(u.u_pdir);
-               return;
-       }
+       factor = 1;
+       for (i = SINGLE; i < level; i++)
+               factor *= NINDIR(fs);
+       last = lastbn;
+       if (lastbn > 0)
+               last /= factor;
+       nblocks = btodb(fs->fs_bsize);

        /*
        /*

-        * must read in an existing directory block
-        * to prepare to place the new entry into it.
+        * Get buffer of block pointers, zero those 
+        * entries corresponding to blocks to be free'd,
+        * and update on disk copy first.

         */
         */

-       fs = u.u_pdir->i_fs;
-       lbn = lblkno(fs, u.u_offset);
-       base = blkoff(fs, u.u_offset);
-       bn = fsbtodb(fs, bmap(u.u_pdir, lbn, B_WRITE, base + u.u_count));
-       if (u.u_offset + u.u_count > u.u_pdir->i_size)
-               u.u_pdir->i_size = u.u_offset + u.u_count;
-       bp = bread(u.u_pdir->i_dev, bn, blksize(fs, u.u_pdir, lbn));
-       if (bp->b_flags & B_ERROR) {
+#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);
                brelse(bp);

-               return;
+               *countp = 0;
+               return (error);

        }
        }

-       dirbuf = bp->b_un.b_addr + base;
-       dp = (struct direct *)dirbuf;
-       dsize = DIRSIZ(dp);
-       spccnt = dp->d_reclen - dsize;
+       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;
+

        /*
        /*

-        * if there is insufficient room to make an entry at this point
-        * namei insures that compacting from u.u_offset for u.u_count
-        * bytes will provide the necessary space.
+        * Recursively free totally unused blocks.

         */
         */

-       for (loc = dp->d_reclen; loc < u.u_count; ) {
-               ndp = (struct direct *)(dirbuf + loc);
-               if (dp->d_ino == 0) {
-                       spccnt += dsize;
-               } else {
-                       dp->d_reclen = dsize;
-                       dp = (struct direct *)((char *)dp + dsize);
+       for (i = NINDIR(fs) - 1; i > last; i--) {
+               nb = bap[i];
+               if (nb == 0)
+                       continue;
+               if (level > SINGLE) {
+                       error = indirtrunc(ip, nb, (daddr_t)-1, level - 1,
+                               &blkcount);
+                       if (error)
+                               allerror = error;
+                       blocksreleased += blkcount;

                }
                }

-               dsize = DIRSIZ(ndp);
-               spccnt += ndp->d_reclen - dsize;
-               loc += ndp->d_reclen;
-               bcopy(ndp, dp, dsize);
+               blkfree(ip, nb, (off_t)fs->fs_bsize);
+               blocksreleased += nblocks;

        }
        }

+

        /*
        /*

-        * Update the pointer fields in the previous entry (if any),
-        * copy in the new entry, and write out the block.
+        * Recursively free last partial block.

         */
         */

-       if (dp->d_ino == 0) {
-               if (spccnt + dsize < newsize)
-                       panic("wdir: compact failed");
-               u.u_dent.d_reclen = spccnt + dsize;
-       } else {
-               if (spccnt < newsize)
-                       panic("wdir: compact failed");
-               u.u_dent.d_reclen = spccnt;
-               dp->d_reclen = dsize;
-               dp = (struct direct *)((char *)dp + dsize);
+       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;
+               }

        }
        }

-       bcopy(&u.u_dent, dp, newsize);
-       bwrite(bp);
-       u.u_pdir->i_flag |= IUPD|ICHG;
-       iput(u.u_pdir);
+       FREE(copy, M_TEMP);
+       *countp = blocksreleased;
+       return (allerror);

 }
 
 }
 

-#ifdef ilock
-#undef ilock
-#endif
-#ifdef iunlock
-#undef iunlock
-#endif

 /*
  * Lock an inode. If its already locked, set the WANT bit and sleep.
  */
 /*
  * Lock an inode. If its already locked, set the WANT bit and sleep.
  */


@@ -475,11 +644,16 @@ ilock(ip)
        register struct inode *ip;
 {
 
        register struct inode *ip;
 {
 

-       while (ip->i_flag&ILOCK) {
+       while (ip->i_flag & ILOCKED) {

                ip->i_flag |= IWANT;
                ip->i_flag |= IWANT;

-               sleep((caddr_t)ip, PINOD);
+               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 |= ILOCK;
+       ip->i_spare1 = 0;
+       ip->i_spare0 = curproc->p_pid;
+       ip->i_flag |= ILOCKED;

 }
 
 /*
 }
 
 /*


@@ -489,7 +663,10 @@ iunlock(ip)
        register struct inode *ip;
 {
 
        register struct inode *ip;
 {
 

-       ip->i_flag &= ~ILOCK;
+       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);
        if (ip->i_flag&IWANT) {
                ip->i_flag &= ~IWANT;
                wakeup((caddr_t)ip);