assign a new generation number when allocating an inode
[unix-history] / usr / src / sys / ufs / lfs / lfs_inode.c
index 21e64cf..257339b 100644 (file)
@@ -1,21 +1,64 @@
-/*     lfs_inode.c     4.6     81/10/11        */
-
-#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;
+/*
+ * Copyright (c) 1982, 1986, 1989 Regents of the University of California.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms are permitted
+ * provided that the above copyright notice and this paragraph are
+ * duplicated in all such forms and that any documentation,
+ * advertising materials, and other materials related to such
+ * distribution and use acknowledge that the software was developed
+ * by the University of California, Berkeley.  The name of the
+ * University may not be used to endorse or promote products derived
+ * from this software without specific prior written permission.
+ * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
+ * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ *     @(#)lfs_inode.c 7.7 (Berkeley) %G%
+ */
+
+#include "param.h"
+#include "systm.h"
+#include "mount.h"
+#include "user.h"
+#include "file.h"
+#include "buf.h"
+#include "cmap.h"
+#include "vnode.h"
+#include "../ufs/inode.h"
+#include "../ufs/fs.h"
+#include "../ufs/ufsmount.h"
+#ifdef QUOTA
+#include "../ufs/quota.h"
+#endif
+#include "kernel.h"
+#include "malloc.h"
+
+#define        INOHSZ  512
+#if    ((INOHSZ&(INOHSZ-1)) == 0)
+#define        INOHASH(dev,ino)        (((dev)+(ino))&(INOHSZ-1))
+#else
+#define        INOHASH(dev,ino)        (((unsigned)((dev)+(ino)))%INOHSZ)
+#endif
+
+#define INSFREE(ip) {\
+       if (ifreeh) { \
+               *ifreet = (ip); \
+               (ip)->i_freeb = ifreet; \
+       } else { \
+               ifreeh = (ip); \
+               (ip)->i_freeb = &ifreeh; \
+       } \
+       (ip)->i_freef = NULL; \
+       ifreet = &(ip)->i_freef; \
+}
+
+union ihead {                          /* inode LRU cache, Chris Maltby */
+       union  ihead *ih_head[2];
+       struct inode *ih_chain[2];
+} ihead[INOHSZ];
+
+struct inode *ifreeh, **ifreet, *bdevlisth;
 
 /*
  * Initialize hash links for inodes
 
 /*
  * Initialize hash links for inodes
@@ -25,135 +68,360 @@ ihinit()
 {
        register int i;
        register struct inode *ip = inode;
 {
        register int i;
        register struct inode *ip = inode;
+       register union  ihead *ih = ihead;
 
 
-       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;
-}
-
-/*
- * Find an inode if it is incore.
- * This is the equivalent, for inodes,
- * of ``incore'' in bio.c or ``pfind'' in subr.c.
- */
-struct inode *
-ifind(dev, ino)
-dev_t dev;
-ino_t ino;
-{
-       register struct inode *ip;
-
-       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);
+       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;
+       ITOV(ip)->v_data = (qaddr_t)ip;
+       for (i = ninode; --i > 0; ) {
+               ++ip;
+               ip->i_forw = ip;
+               ip->i_back = ip;
+               ITOV(ip)->v_data = (qaddr_t)ip;
+               *ifreet = ip;
+               ip->i_freeb = ifreet;
+               ifreet = &ip->i_freef;
+       }
+       ip->i_freef = NULL;
 }
 
 /*
 }
 
 /*
- * 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, 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;
+       register struct inode *ip, *iq;
+       register struct vnode *vp;
+       struct inode *nip;
+       struct buf *bp;
+       struct dinode tdip, *dp;
+       union  ihead *ih;
+       int error;
 
 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) {
+       ih = &ihead[INOHASH(dev, ino)];
+       for (ip = ih->ih_chain[0]; ip != (struct inode *)ih; ip = ip->i_forw)
+               if (ino == ip->i_number && dev == ip->i_dev) {
+                       /*
+                        * Following is essentially an inline expanded
+                        * copy of igrab(), expanded inline for speed,
+                        * and so that the test for a mounted on inode
+                        * can be deferred until after we are sure that
+                        * the inode isn't busy.
+                        */
+                       if ((ip->i_flag&ILOCKED) != 0) {
                                ip->i_flag |= IWANT;
                                sleep((caddr_t)ip, PINOD);
                                goto loop;
                        }
                                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");
+                       vp = ITOV(ip);
+                       if (vp->v_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 |= ILOCK;
-                       return(ip);
+                       ILOCK(ip);
+                       vp->v_count++;
+                       *ipp = ip;
+                       return(0);
                }
                }
-               ip = &inode[ip->i_hlink];
+       if (error = getnewino(dev, ino, &nip)) {
+               *ipp = 0;
+               return (error);
        }
        }
-       if(ifreel < 0) {
+       ip = nip;
+       /*
+        * Read in the disk contents for the inode.
+        */
+       if (error = bread(VFSTOUFS(mntp)->um_devvp, fsbtodb(fs, itod(fs, ino)),
+           (int)fs->fs_bsize, &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. We also
+                * lose its inumber, just in case.
+                */
+               remque(ip);
+               ip->i_forw = ip;
+               ip->i_back = ip;
+               ip->i_number = 0;
+               INSFREE(ip);
+               iunlock(ip);
+               ip->i_flag = 0;
+               brelse(bp);
+               *ipp = 0;
+               return(error);
+       }
+       /*
+        * Check to see if the new inode represents a block device
+        * for which we already have an inode (either because of
+        * bdevvp() or because of a different inode representing
+        * the same block device). If such an alias exists, put the
+        * just allocated inode back on the free list, and replace
+        * the contents of the existing inode with the contents of
+        * the new inode.
+        */
+       dp = bp->b_un.b_dino;
+       dp += itoo(fs, ino);
+       if ((dp->di_mode & IFMT) != IFBLK) {
+               ip->i_ic = dp->di_ic;
+               brelse(bp);
+       } else {
+again:
+               for (iq = bdevlisth; iq; iq = iq->i_devlst) {
+                       if (dp->di_rdev != ITOV(iq)->v_rdev)
+                               continue;
+                       igrab(iq);
+                       if (dp->di_rdev != ITOV(iq)->v_rdev) {
+                               iput(iq);
+                               goto again;
+                       }
+                       /*
+                        * Discard unneeded inode.
+                        */
+                       remque(ip);
+                       ip->i_forw = ip;
+                       ip->i_back = ip;
+                       ip->i_number = 0;
+                       INSFREE(ip);
+                       iunlock(ip);
+                       ip->i_flag = 0;
+                       /*
+                        * Reinitialize aliased inode.
+                        * We must release the buffer that we just read
+                        * before doing the iupdat() to avoid a possible
+                        * deadlock with updating an inode in the same
+                        * disk block.
+                        */
+                       ip = iq;
+                       vp = ITOV(iq);
+                       tdip.di_ic = dp->di_ic;
+                       brelse(bp);
+                       error = iupdat(ip, &time, &time, 1);
+                       ip->i_ic = tdip.di_ic;
+                       remque(ip);
+                       insque(ip, ih);
+                       ip->i_dev = dev;
+                       ip->i_number = ino;
+                       if (ip->i_devvp) {
+                               vrele(ip->i_devvp);
+                               ip->i_devvp = 0;
+                       }
+                       cache_purge(vp);
+                       break;
+               }
+               if (iq == 0) {
+                       ip->i_ic = dp->di_ic;
+                       brelse(bp);
+                       ip->i_devlst = bdevlisth;
+                       bdevlisth = ip;
+               }
+       }
+       /*
+        * Finish inode initialization.
+        */
+       ip->i_fs = fs;
+       ip->i_devvp = VFSTOUFS(mntp)->um_devvp;
+       ip->i_devvp->v_count++;
+       /*
+        * Initialize the associated vnode
+        */
+       vp = ITOV(ip);
+       vinit(vp, mntp, IFTOVT(ip->i_mode), &ufs_vnodeops);
+       if (vp->v_type == VCHR || vp->v_type == VBLK) {
+               vp->v_rdev = ip->i_rdev;
+               vp->v_op = &blk_vnodeops;
+       }
+       if (ino == ROOTINO)
+               vp->v_flag |= VROOT;
+#ifdef QUOTA
+       if (ip->i_mode != 0)
+               ip->i_dquot = inoquota(ip);
+#endif
+       *ipp = ip;
+       return (0);
+}
+
+/*
+ * Allocate a new inode.
+ *
+ * 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.
+ */
+getnewino(dev, ino, ipp)
+       dev_t dev;
+       ino_t ino;
+       struct inode **ipp;
+{
+       union ihead *ih;
+       register struct inode *ip, *iq;
+       register struct vnode *vp;
+
+       /*
+        * Remove the next inode from the free list.
+        */
+       if ((ip = ifreeh) == NULL) {
                tablefull("inode");
                tablefull("inode");
-               u.u_error = ENFILE;
-               return(NULL);
+               *ipp = 0;
+               return(ENFILE);
        }
        }
-       ip = &inode[ifreel];
-       ifreel = ip->i_hlink;
-       ip->i_hlink = inohash[slot];
-       inohash[slot] = ip - inode;
+       vp = ITOV(ip);
+       if (vp->v_count)
+               panic("free inode isn't");
+       if (iq = ip->i_freef)
+               iq->i_freeb = &ifreeh;
+       ifreeh = iq;
+       ip->i_freef = NULL;
+       ip->i_freeb = NULL;
+       /*
+        * Now to take inode off the hash chain it was on
+        * (initially, or after an iflush, it is on a "hash chain"
+        * consisting entirely of itself, and pointed to by no-one)
+        * and put it on the chain for its new (ino, dev) pair.
+        */
+       remque(ip);
        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));
+       if (dev != NODEV) {
+               ih = &ihead[INOHASH(dev, ino)];
+               insque(ip, ih);
+       }
+       ip->i_flag = 0;
+       ILOCK(ip);
+       ip->i_lastr = 0;
+#endif SECSIZE
        /*
        /*
-        * Check I/O errors
+        * Purge old data structures associated with the inode.
         */
         */
-       if((bp->b_flags&B_ERROR) != 0) {
-               brelse(bp);
-               iput(ip);
-               return(NULL);
+       cache_purge(vp);
+       if (ip->i_devvp) {
+               vrele(ip->i_devvp);
+               ip->i_devvp = 0;
        }
        }
-       dp = bp->b_un.b_dino;
-       dp += itoo(ino);
-       iexpand(ip, dp);
-       brelse(bp);
-       return(ip);
+#ifdef QUOTA
+       dqrele(ip->i_dquot);
+       ip->i_dquot = NODQUOT;
+#endif
+       if (vp->v_type == VBLK) {
+               if (bdevlisth == ip) {
+                       bdevlisth = ip->i_devlst;
+               } else {
+                       for (iq = bdevlisth; iq; iq = iq->i_devlst) {
+                               if (iq->i_devlst != ip)
+                                       continue;
+                               iq->i_devlst = ip->i_devlst;
+                               break;
+                       }
+                       if (iq == NULL)
+                               panic("missing bdev");
+               }
+       }
+       *ipp = ip;
+       return (0);
 }
 
 }
 
-iexpand(ip, dp)
-register struct inode *ip;
-register struct dinode *dp;
+/*
+ * Convert a pointer to an inode into a reference to an inode.
+ *
+ * This is basically the internal piece of iget (after the
+ * inode pointer is located) but without the test for mounted
+ * filesystems.  It is caller's responsibility to check that
+ * the inode pointer is valid.
+ */
+igrab(ip)
+       register struct inode *ip;
 {
 {
-       register char *p1, *p2;
-       register int i;
+       register struct vnode *vp = ITOV(ip);
+
+       while ((ip->i_flag&ILOCKED) != 0) {
+               ip->i_flag |= IWANT;
+               sleep((caddr_t)ip, PINOD);
+       }
+       if (vp->v_count == 0) {         /* ino on free list */
+               register struct inode *iq;
 
 
-       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 (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;
        }
        }
+       vp->v_count++;
+       ILOCK(ip);
+}
+
+/*
+ * Create a vnode for a block device.
+ * Used for root filesystem, argdev, and swap areas.
+ */
+bdevvp(dev, vpp)
+       dev_t dev;
+       struct vnode **vpp;
+{
+       register struct inode *ip;
+       register struct vnode *vp;
+       struct inode *nip;
+       int error;
+
+       /*
+        * Check for the existence of an existing vnode.
+        */
+again:
+       for (ip = bdevlisth; ip; ip = ip->i_devlst) {
+               vp = ITOV(ip);
+               if (dev != vp->v_rdev)
+                       continue;
+               igrab(ip);
+               if (dev != vp->v_rdev) {
+                       iput(ip);
+                       goto again;
+               }
+               IUNLOCK(ip);
+               *vpp = vp;
+               return (0);
+       }
+       if (error = getnewino(NODEV, (ino_t)0, &nip)) {
+               *vpp = 0;
+               return (error);
+       }
+       ip = nip;
+       ip->i_fs = 0;
+       ip->i_devlst = bdevlisth;
+       bdevlisth = ip;
+       vp = ITOV(ip);
+       vinit(vp, 0, VBLK, &blk_vnodeops);
+       vp->v_rdev = dev;
+       IUNLOCK(ip);
+       *vpp = vp;
+       return (0);
 }
 
 /*
 }
 
 /*
@@ -164,42 +432,49 @@ register struct dinode *dp;
  * truncate and deallocate the file.
  */
 iput(ip)
  * truncate and deallocate the file.
  */
 iput(ip)
-register struct inode *ip;
+       register struct inode *ip;
 {
 {
-       register int i, x;
-       register struct inode *jp;
-
-       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, 0);
-               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");
-               }
-done:
-               ip->i_hlink = ifreel;
-               ifreel = x;
-               ip->i_flag = 0;
-               ip->i_number = 0;
-       } else
-               prele(ip);
-       ip->i_count--;
+
+       if ((ip->i_flag & ILOCKED) == 0)
+               panic("iput");
+       IUNLOCK(ip);
+       vrele(ITOV(ip));
+}
+
+
+ufs_inactive(vp)
+       struct vnode *vp;
+{
+       register struct inode *ip = VTOI(vp);
+       int mode, error;
+
+       if (ITOV(ip)->v_count != 0)
+               panic("ufs_inactive: not inactive");
+       ILOCK(ip);
+       if (ip->i_nlink <= 0 && (ITOV(ip)->v_mount->m_flag&M_RDONLY) == 0) {
+               error = itrunc(ip, (u_long)0);
+               mode = ip->i_mode;
+               ip->i_mode = 0;
+               ip->i_rdev = 0;
+               ip->i_flag |= IUPD|ICHG;
+               ifree(ip, ip->i_number, mode);
+#ifdef QUOTA
+               (void) chkiq(ip->i_dev, ip, ip->i_uid, 0);
+               dqrele(ip->i_dquot);
+               ip->i_dquot = NODQUOT;
+#endif
+       }
+       IUPDAT(ip, &time, &time, 0);
+       IUNLOCK(ip);
+       ip->i_flag = 0;
+       /*
+        * Put the inode on the end of the free list.
+        * Possibly in some cases it would be better to
+        * put the inode at the head of the free list,
+        * (eg: where i_mode == 0 || i_number == 0).
+        */
+       INSFREE(ip);
+       return (error);
 }
 
 /*
 }
 
 /*
@@ -211,239 +486,442 @@ done:
  * i/o order so wait for write to complete.
  */
 iupdat(ip, ta, tm, waitfor)
  * i/o order so wait for write to complete.
  */
 iupdat(ip, ta, tm, waitfor)
-register struct inode *ip;
-time_t *ta, *tm;
-int 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;
-
-       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)
-                               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);
-               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->m_flag & M_RDONLY)
+               return (0);
+       error = bread(ip->i_devvp, fsbtodb(fs, itod(fs, ip->i_number)),
+               (int)fs->fs_bsize, &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->di_ic = ip->i_ic;
+       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)
+       register struct inode *oip;
+       u_long length;
 {
 {
-       register i;
-       dev_t dev;
-       daddr_t bn;
-       struct inode itmp;
-
-       if (ip->i_vfdcnt)
-               panic("itrunc");
-       i = ip->i_mode & IFMT;
-       if (i!=IFREG && i!=IFDIR)
-               return;
+       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 error, allerror = 0;
+       struct inode tip;
 
 
+       if (oip->i_size <= length) {
+               oip->i_flag |= ICHG|IUPD;
+               error = iupdat(oip, &time, &time, 1);
+               return (error);
+       }
        /*
        /*
-        * Clean inode on disk before freeing blocks
-        * to insure no duplicates if system crashes.
+        * 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.
         */
         */
-       itmp = *ip;
-       itmp.i_size = 0;
-       for (i = 0; i < NADDR; i++)
-               itmp.i_un.i_addr[i] = 0;
-       itmp.i_flag |= ICHG|IUPD;
-       iupdat(&itmp, &time, &time, 1);
-       ip->i_flag &= ~(IUPD|IACC|ICHG);
-
+       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);
        /*
        /*
-        * Now return blocks to free list... if machine
-        * crashes, they will be harmless MISSING blocks.
+        * 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.
         */
         */
-       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) {
-
-               default:
-                       free(dev, bn);
-                       break;
-
-               case NADDR-3:
-                       tloop(dev, bn, 0, 0);
-                       break;
-
-               case NADDR-2:
-                       tloop(dev, bn, 1, 0);
-                       break;
-
-               case NADDR-1:
-                       tloop(dev, bn, 1, 1);
+       osize = oip->i_size;
+       offset = blkoff(fs, length);
+       if (offset == 0) {
+               oip->i_size = length;
+       } else {
+               lbn = lblkno(fs, length);
+               error = balloc(oip, lbn, offset, &bn, B_CLRBUF);
+               if (error)
+                       return (error);
+               if ((long)bn < 0)
+                       panic("itrunc: hole");
+               oip->i_size = length;
+               size = blksize(fs, oip, lbn);
+               count = howmany(size, CLBYTES);
+                       munhash(oip->i_devvp, bn + i * CLBYTES / DEV_BSIZE);
+               error = bread(oip->i_devvp, bn, size, &bp);
+               if (error) {
+                       oip->i_size = osize;
+                       brelse(bp);
+                       return (error);
                }
                }
+               bzero(bp->b_un.b_addr + offset, (unsigned)(size - offset));
+               bdwrite(bp);
        }
        }
-       ip->i_size = 0;
        /*
        /*
-        * Inode was written and flags updated above.
-        * No need to modify flags here.
+        * 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;
+       allerror = syncip(oip);
 
 
-tloop(dev, bn, f1, f2)
-dev_t dev;
-daddr_t bn;
-{
-       register i;
-       register struct buf *bp;
-       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;
+       /*
+        * 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;
                        }
                        }
-                       bap = bp->b_un.b_daddr;
                }
                }
-               nb = bap[i];
-               if(nb == (daddr_t)0)
+               if (lastiblock[level] >= 0)
+                       goto done;
+       }
+
+       /*
+        * All whole direct blocks or frags.
+        */
+       for (i = NDADDR - 1; i > lastblock; i--) {
+               register off_t bsize;
+
+               bn = ip->i_db[i];
+               if (bn == 0)
                        continue;
                        continue;
-               if(f1) {
-                       brelse(bp);
-                       bp = NULL;
-                       tloop(dev, nb, f2, 0);
-               } else
-                       free(dev, nb);
+               ip->i_db[i] = 0;
+               bsize = (off_t)blksize(fs, ip, i);
+               blkfree(ip, bn, bsize);
+               blocksreleased += btodb(bsize);
        }
        }
-       if(bp != NULL)
-               brelse(bp);
-       free(dev, bn);
+       if (lastblock < 0)
+               goto done;
+
+       /*
+        * Finally, look for a change in size of the
+        * last direct block; release any frags.
+        */
+       bn = ip->i_db[lastblock];
+       if (bn != 0) {
+               off_t oldspace, newspace;
+
+               /*
+                * Calculate amount of space we're giving
+                * back as old block size minus new block size.
+                */
+               oldspace = blksize(fs, ip, lastblock);
+               ip->i_size = length;
+               newspace = blksize(fs, ip, lastblock);
+               if (newspace == 0)
+                       panic("itrunc: newspace");
+               if (oldspace - newspace > 0) {
+                       /*
+                        * Block number of space to be free'd is
+                        * the old block # plus the number of frags
+                        * required for the storage we're keeping.
+                        */
+                       bn += numfrags(fs, newspace);
+                       blkfree(ip, bn, oldspace - newspace);
+                       blocksreleased += btodb(oldspace - newspace);
+               }
+       }
+done:
+/* BEGIN PARANOIA */
+       for (level = SINGLE; level <= TRIPLE; level++)
+               if (ip->i_ib[level] != oip->i_ib[level])
+                       panic("itrunc1");
+       for (i = 0; i < NDADDR; i++)
+               if (ip->i_db[i] != oip->i_db[i])
+                       panic("itrunc2");
+/* END PARANOIA */
+       oip->i_blocks -= blocksreleased;
+       if (oip->i_blocks < 0)                  /* sanity */
+               oip->i_blocks = 0;
+       oip->i_flag |= ICHG;
+#ifdef QUOTA
+       (void) chkdq(oip, -blocksreleased, 0);
+#endif
+       return (allerror);
 }
 
 /*
 }
 
 /*
- * Make a new file.
+ * 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.
  */
  */
-struct inode *
-maknode(mode)
-{
+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;
 
 
-       ip = ialloc(u.u_pdir->i_dev);
-       if(ip == NULL) {
-               iput(u.u_pdir);
-               return(NULL);
+       /*
+        * 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, &bp);
+       if (error) {
+               brelse(bp);
+               *countp = 0;
+               return (error);
        }
        }
-       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;
+       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));
+       error = bwrite(bp);
+       if (error)
+               allerror = error;
+       bap = copy;
 
        /*
 
        /*
-        * Make sure inode goes to disk before directory entry.
+        * Recursively free totally unused blocks.
         */
         */
-       iupdat(ip, &time, &time, 1);
+       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;
+               }
+               blkfree(ip, nb, (off_t)fs->fs_bsize);
+               blocksreleased += nblocks;
+       }
 
 
-       wdir(ip);
-       return(ip);
+       /*
+        * 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);
 }
 
 /*
 }
 
 /*
- * Write a directory entry with
- * parameters left as side effects
- * to a call to namei.
+ * 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).
  */
  */
-wdir(ip)
-struct inode *ip;
+#ifdef QUOTA
+iflush(dev, iq)
+       dev_t dev;
+       struct inode *iq;
+#else
+iflush(dev)
+       dev_t dev;
+#endif
 {
 {
+       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);
-}
-
-#ifdef plock
-#undef plock
+       for (ip = inode; ip < inodeNINODE; ip++) {
+#ifdef QUOTA
+               if (ip != iq && ip->i_dev == dev)
+#else
+               if (ip->i_dev == dev)
 #endif
 #endif
-#ifdef prele
-#undef prele
+                       if (ITOV(ip)->v_count)
+                               return (EBUSY);
+                       else {
+                               remque(ip);
+                               ip->i_forw = ip;
+                               ip->i_back = ip;
+                               /*
+                                * as v_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
 #endif
+                               if (ip->i_devvp) {
+                                       vrele(ip->i_devvp);
+                                       ip->i_devvp = 0;
+                               }
+                       }
+       }
+       return (0);
+}
+
 /*
 /*
- * Lock an inode (should be called ilock).
- * If its already locked,
- * set the WANT bit and sleep.
+ * Lock an inode. If its already locked, set the WANT bit and sleep.
  */
  */
-plock(ip)
-register struct inode *ip;
+ilock(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);
+               (void) sleep((caddr_t)ip, PINOD);
        }
        }
-       ip->i_flag |= ILOCK;
+       ip->i_flag |= ILOCKED;
 }
 
 /*
 }
 
 /*
- * Unlock an inode.
- * If WANT bit is on,
- * wakeup.
+ * Unlock an inode.  If WANT bit is on, wakeup.
  */
  */
-prele(ip)
-register struct inode *ip;
+iunlock(ip)
+       register struct inode *ip;
 {
 
 {
 
-       ip->i_flag &= ~ILOCK;
-       if(ip->i_flag&IWANT) {
+       if ((ip->i_flag & ILOCKED) == 0)
+               printf("unlocking unlocked inode %d on dev 0x%x\n",
+                       ip->i_number, ip->i_dev);
+       ip->i_flag &= ~ILOCKED;
+       if (ip->i_flag&IWANT) {
                ip->i_flag &= ~IWANT;
                wakeup((caddr_t)ip);
        }
 }
                ip->i_flag &= ~IWANT;
                wakeup((caddr_t)ip);
        }
 }
+
+/*
+ * Check mode permission on inode pointer. Mode is READ, WRITE or EXEC.
+ * The mode is shifted to select the owner/group/other fields. The
+ * super user is granted all permissions.
+ *
+ * NB: Called from vnode op table. It seems this could all be done
+ * using vattr's but...
+ */
+iaccess(ip, mode, cred)
+       register struct inode *ip;
+       register int mode;
+       struct ucred *cred;
+{
+       register gid_t *gp;
+       register struct vnode *vp = ITOV(ip);
+       int i;
+
+       /*
+        * If you're the super-user,
+        * you always get access.
+        */
+       if (cred->cr_uid == 0)
+               return (0);
+       /*
+        * Access check is based on only one of owner, group, public.
+        * If not owner, then check group. If not a member of the
+        * group, then check public access.
+        */
+       if (cred->cr_uid != ip->i_uid) {
+               mode >>= 3;
+               gp = cred->cr_groups;
+               for (i = 0; i < cred->cr_ngroups; i++, gp++)
+                       if (ip->i_gid == *gp)
+                               goto found;
+               mode >>= 3;
+found:
+               ;
+       }
+       if ((ip->i_mode & mode) != 0)
+               return (0);
+       return (EACCES);
+}