BSD 4_4 release
[unix-history] / usr / src / sys / kern / vfs_subr.c
index ce47f3b..d11e8d3 100644 (file)
@@ -1,34 +1,74 @@
 /*
 /*
- * Copyright (c) 1989 The Regents of the University of California.
- * All rights reserved.
+ * Copyright (c) 1989, 1993
+ *     The 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 MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *     This product includes software developed by the University of
+ *     California, Berkeley and its contributors.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
  *
  *
- *     @(#)vfs_subr.c  7.26 (Berkeley) %G%
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *     @(#)vfs_subr.c  8.1 (Berkeley) 6/10/93
  */
 
 /*
  * External virtual filesystem routines
  */
 
  */
 
 /*
  * External virtual filesystem routines
  */
 
-#include "param.h"
-#include "mount.h"
-#include "time.h"
-#include "vnode.h"
-#include "namei.h"
-#include "ucred.h"
-#include "errno.h"
-#include "malloc.h"
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/proc.h>
+#include <sys/mount.h>
+#include <sys/time.h>
+#include <sys/vnode.h>
+#include <sys/stat.h>
+#include <sys/namei.h>
+#include <sys/ucred.h>
+#include <sys/buf.h>
+#include <sys/errno.h>
+#include <sys/malloc.h>
+
+#include <vm/vm.h>
+#include <sys/sysctl.h>
+
+#include <miscfs/specfs/specdev.h>
+
+enum vtype iftovt_tab[16] = {
+       VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
+       VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
+};
+int    vttoif_tab[9] = {
+       0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
+       S_IFSOCK, S_IFIFO, S_IFMT,
+};
+
+/*
+ * Insq/Remq for the vnode usage lists.
+ */
+#define        bufinsvn(bp, dp)        list_enter_head(dp, bp, struct buf *, b_vnbufs)
+#define        bufremvn(bp)            list_remove(bp, struct buf *, b_vnbufs)
 
 /*
  * Remove a mount point from the list of mounted filesystems.
 
 /*
  * Remove a mount point from the list of mounted filesystems.
@@ -41,9 +81,9 @@ vfs_remove(mp)
 
        if (mp == rootfs)
                panic("vfs_remove: unmounting root");
 
        if (mp == rootfs)
                panic("vfs_remove: unmounting root");
-       mp->m_prev->m_next = mp->m_next;
-       mp->m_next->m_prev = mp->m_prev;
-       mp->m_vnodecovered->v_mountedhere = (struct mount *)0;
+       mp->mnt_prev->mnt_next = mp->mnt_next;
+       mp->mnt_next->mnt_prev = mp->mnt_prev;
+       mp->mnt_vnodecovered->v_mountedhere = (struct mount *)0;
        vfs_unlock(mp);
 }
 
        vfs_unlock(mp);
 }
 
@@ -55,11 +95,11 @@ vfs_lock(mp)
        register struct mount *mp;
 {
 
        register struct mount *mp;
 {
 
-       while(mp->m_flag & M_MLOCK) {
-               mp->m_flag |= M_MWAIT;
+       while(mp->mnt_flag & MNT_MLOCK) {
+               mp->mnt_flag |= MNT_MWAIT;
                sleep((caddr_t)mp, PVFS);
        }
                sleep((caddr_t)mp, PVFS);
        }
-       mp->m_flag |= M_MLOCK;
+       mp->mnt_flag |= MNT_MLOCK;
        return (0);
 }
 
        return (0);
 }
 
@@ -72,151 +112,125 @@ vfs_unlock(mp)
        register struct mount *mp;
 {
 
        register struct mount *mp;
 {
 
-       if ((mp->m_flag & M_MLOCK) == 0)
-               panic("vfs_unlock: locked fs");
-       mp->m_flag &= ~M_MLOCK;
-       if (mp->m_flag & M_MWAIT) {
-               mp->m_flag &= ~M_MWAIT;
+       if ((mp->mnt_flag & MNT_MLOCK) == 0)
+               panic("vfs_unlock: not locked");
+       mp->mnt_flag &= ~MNT_MLOCK;
+       if (mp->mnt_flag & MNT_MWAIT) {
+               mp->mnt_flag &= ~MNT_MWAIT;
                wakeup((caddr_t)mp);
        }
 }
 
 /*
                wakeup((caddr_t)mp);
        }
 }
 
 /*
- * Lookup a mount point by filesystem identifier.
+ * Mark a mount point as busy.
+ * Used to synchronize access and to delay unmounting.
  */
  */
-struct mount *
-getvfs(fsid)
-       fsid_t *fsid;
-{
+vfs_busy(mp)
        register struct mount *mp;
        register struct mount *mp;
+{
 
 
-       mp = rootfs;
-       do {
-               if (mp->m_fsid.val[0] == fsid->val[0] &&
-                   mp->m_fsid.val[1] == fsid->val[1]) {
-                       return (mp);
-               }
-               mp = mp->m_next;
-       } while (mp != rootfs);
-       return ((struct mount *)0);
+       while(mp->mnt_flag & MNT_MPBUSY) {
+               mp->mnt_flag |= MNT_MPWANT;
+               sleep((caddr_t)&mp->mnt_flag, PVFS);
+       }
+       if (mp->mnt_flag & MNT_UNMOUNT)
+               return (1);
+       mp->mnt_flag |= MNT_MPBUSY;
+       return (0);
 }
 
 /*
 }
 
 /*
- * Set vnode attributes to VNOVAL
+ * Free a busy filesystem.
+ * Panic if filesystem is not busy.
  */
  */
-void vattr_null(vap)
-       register struct vattr *vap;
+vfs_unbusy(mp)
+       register struct mount *mp;
 {
 
 {
 
-       vap->va_type = VNON;
-       vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid =
-               vap->va_fsid = vap->va_fileid = vap->va_size =
-               vap->va_size1 = vap->va_blocksize = vap->va_rdev =
-               vap->va_bytes = vap->va_bytes1 =
-               vap->va_atime.tv_sec = vap->va_atime.tv_usec =
-               vap->va_mtime.tv_sec = vap->va_mtime.tv_usec =
-               vap->va_ctime.tv_sec = vap->va_ctime.tv_usec =
-               vap->va_flags = vap->va_gen = VNOVAL;
+       if ((mp->mnt_flag & MNT_MPBUSY) == 0)
+               panic("vfs_unbusy: not busy");
+       mp->mnt_flag &= ~MNT_MPBUSY;
+       if (mp->mnt_flag & MNT_MPWANT) {
+               mp->mnt_flag &= ~MNT_MPWANT;
+               wakeup((caddr_t)&mp->mnt_flag);
+       }
 }
 
 /*
 }
 
 /*
- * Initialize a nameidata structure
+ * Lookup a mount point by filesystem identifier.
  */
  */
-ndinit(ndp)
-       register struct nameidata *ndp;
+struct mount *
+getvfs(fsid)
+       fsid_t *fsid;
 {
 {
+       register struct mount *mp;
 
 
-       bzero((caddr_t)ndp, sizeof(struct nameidata));
-       ndp->ni_iov = &ndp->ni_nd.nd_iovec;
-       ndp->ni_iovcnt = 1;
-       ndp->ni_base = (caddr_t)&ndp->ni_dent;
-       ndp->ni_rw = UIO_WRITE;
-       ndp->ni_uioseg = UIO_SYSSPACE;
+       mp = rootfs;
+       do {
+               if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
+                   mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
+                       return (mp);
+               }
+               mp = mp->mnt_next;
+       } while (mp != rootfs);
+       return ((struct mount *)0);
 }
 
 /*
 }
 
 /*
- * Duplicate a nameidata structure
+ * Get a new unique fsid
  */
  */
-nddup(ndp, newndp)
-       register struct nameidata *ndp, *newndp;
+void
+getnewfsid(mp, mtype)
+       struct mount *mp;
+       int mtype;
 {
 {
+static u_short xxxfs_mntid;
+
+       fsid_t tfsid;
 
 
-       ndinit(newndp);
-       newndp->ni_cdir = ndp->ni_cdir;
-       VREF(newndp->ni_cdir);
-       newndp->ni_rdir = ndp->ni_rdir;
-       if (newndp->ni_rdir)
-               VREF(newndp->ni_rdir);
-       newndp->ni_cred = ndp->ni_cred;
-       crhold(newndp->ni_cred);
+       mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + 11, 0);  /* XXX */
+       mp->mnt_stat.f_fsid.val[1] = mtype;
+       if (xxxfs_mntid == 0)
+               ++xxxfs_mntid;
+       tfsid.val[0] = makedev(nblkdev, xxxfs_mntid);
+       tfsid.val[1] = mtype;
+       if (rootfs) {
+               while (getvfs(&tfsid)) {
+                       tfsid.val[0]++;
+                       xxxfs_mntid++;
+               }
+       }
+       mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
 }
 
 /*
 }
 
 /*
- * Release a nameidata structure
+ * Set vnode attributes to VNOVAL
  */
  */
-ndrele(ndp)
-       register struct nameidata *ndp;
+void vattr_null(vap)
+       register struct vattr *vap;
 {
 
 {
 
-       vrele(ndp->ni_cdir);
-       if (ndp->ni_rdir)
-               vrele(ndp->ni_rdir);
-       crfree(ndp->ni_cred);
+       vap->va_type = VNON;
+       vap->va_size = vap->va_bytes = VNOVAL;
+       vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid =
+               vap->va_fsid = vap->va_fileid =
+               vap->va_blocksize = vap->va_rdev =
+               vap->va_atime.ts_sec = vap->va_atime.ts_nsec =
+               vap->va_mtime.ts_sec = vap->va_mtime.ts_nsec =
+               vap->va_ctime.ts_sec = vap->va_ctime.ts_nsec =
+               vap->va_flags = vap->va_gen = VNOVAL;
+       vap->va_vaflags = 0;
 }
 
 /*
  * Routines having to do with the management of the vnode table.
  */
 }
 
 /*
  * Routines having to do with the management of the vnode table.
  */
-struct vnode *vfreeh, **vfreet;
-extern struct vnodeops dead_vnodeops, spec_vnodeops;
+struct vnode *vfreeh, **vfreet = &vfreeh;
+extern int (**dead_vnodeop_p)();
 extern void vclean();
 extern void vclean();
+long numvnodes;
+extern struct vattr va_null;
 
 
-#define        SPECHSZ 64
-#if    ((SPECHSZ&(SPECHSZ-1)) == 0)
-#define        SPECHASH(rdev)  (((rdev>>5)+(rdev))&(SPECHSZ-1))
-#else
-#define        SPECHASH(rdev)  (((unsigned)((rdev>>5)+(rdev)))%SPECHSZ)
-#endif
-struct vnode *speclisth[SPECHSZ];
-
-/*
- * Initialize the vnode structures and initialize each file system type.
- */
-vfsinit()
-{
-       register struct vnode *vp = vnode;
-       struct vfsops **vfsp;
-
-       /*
-        * Build vnode free list.
-        */
-       vfreeh = vp;
-       vfreet = &vp->v_freef;
-       vp->v_freeb = &vfreeh;
-       vp->v_op = &dead_vnodeops;
-       vp->v_type = VBAD;
-       for (vp++; vp < vnodeNVNODE; vp++) {
-               *vfreet = vp;
-               vp->v_freeb = vfreet;
-               vfreet = &vp->v_freef;
-               vp->v_op = &dead_vnodeops;
-               vp->v_type = VBAD;
-       }
-       vp--;
-       vp->v_freef = NULL;
-       /*
-        * Initialize the vnode name cache
-        */
-       nchinit();
-       /*
-        * Initialize each file system type.
-        */
-       for (vfsp = &vfssw[0]; vfsp <= &vfssw[MOUNT_MAXTYPE]; vfsp++) {
-               if (*vfsp == NULL)
-                       continue;
-               (*(*vfsp)->vfs_init)();
-       }
-}
 
 /*
  * Return the next vnode from the free list.
 
 /*
  * Return the next vnode from the free list.
@@ -224,40 +238,62 @@ vfsinit()
 getnewvnode(tag, mp, vops, vpp)
        enum vtagtype tag;
        struct mount *mp;
 getnewvnode(tag, mp, vops, vpp)
        enum vtagtype tag;
        struct mount *mp;
-       struct vnodeops *vops;
+       int (**vops)();
        struct vnode **vpp;
 {
        register struct vnode *vp, *vq;
        struct vnode **vpp;
 {
        register struct vnode *vp, *vq;
+       int s;
 
 
-       if ((vp = vfreeh) == NULL) {
-               tablefull("vnode");
-               *vpp = 0;
-               return (ENFILE);
+       if ((vfreeh == NULL && numvnodes < 2 * desiredvnodes) ||
+           numvnodes < desiredvnodes) {
+               vp = (struct vnode *)malloc((u_long)sizeof *vp,
+                   M_VNODE, M_WAITOK);
+               bzero((char *)vp, sizeof *vp);
+               numvnodes++;
+       } else {
+               if ((vp = vfreeh) == NULL) {
+                       tablefull("vnode");
+                       *vpp = 0;
+                       return (ENFILE);
+               }
+               if (vp->v_usecount)
+                       panic("free vnode isn't");
+               if (vq = vp->v_freef)
+                       vq->v_freeb = &vfreeh;
+               else
+                       vfreet = &vfreeh;
+               vfreeh = vq;
+               vp->v_freef = NULL;
+               vp->v_freeb = NULL;
+               vp->v_lease = NULL;
+               if (vp->v_type != VBAD)
+                       vgone(vp);
+#ifdef DIAGNOSTIC
+               if (vp->v_data)
+                       panic("cleaned vnode isn't");
+               s = splbio();
+               if (vp->v_numoutput)
+                       panic("Clean vnode has pending I/O's");
+               splx(s);
+#endif
+               vp->v_flag = 0;
+               vp->v_lastr = 0;
+               vp->v_lastw = 0;
+               vp->v_lasta = 0;
+               vp->v_cstart = 0;
+               vp->v_clen = 0;
+               vp->v_socket = 0;
        }
        }
-       if (vp->v_usecount)
-               panic("free vnode isn't");
-       if (vq = vp->v_freef)
-               vq->v_freeb = &vfreeh;
-       vfreeh = vq;
-       vp->v_freef = NULL;
-       vp->v_freeb = NULL;
-       if (vp->v_type != VBAD)
-               vgone(vp);
+       vp->v_ralen = 1;
        vp->v_type = VNON;
        vp->v_type = VNON;
-       vp->v_flag = 0;
-       vp->v_shlockc = 0;
-       vp->v_exlockc = 0;
-       vp->v_lastr = 0;
-       vp->v_socket = 0;
        cache_purge(vp);
        vp->v_tag = tag;
        vp->v_op = vops;
        insmntque(vp, mp);
        cache_purge(vp);
        vp->v_tag = tag;
        vp->v_op = vops;
        insmntque(vp, mp);
-       VREF(vp);
+       vp->v_usecount++;
        *vpp = vp;
        return (0);
 }
        *vpp = vp;
        return (0);
 }
-
 /*
  * Move a vnode from one mount queue to another.
  */
 /*
  * Move a vnode from one mount queue to another.
  */
@@ -265,7 +301,7 @@ insmntque(vp, mp)
        register struct vnode *vp;
        register struct mount *mp;
 {
        register struct vnode *vp;
        register struct mount *mp;
 {
-       struct vnode *vq;
+       register struct vnode *vq;
 
        /*
         * Delete from old mount point vnode list, if on one.
 
        /*
         * Delete from old mount point vnode list, if on one.
@@ -284,16 +320,177 @@ insmntque(vp, mp)
                vp->v_mountb = NULL;
                return;
        }
                vp->v_mountb = NULL;
                return;
        }
-       if (mp->m_mounth) {
-               vp->v_mountf = mp->m_mounth;
-               vp->v_mountb = &mp->m_mounth;
-               mp->m_mounth->v_mountb = &vp->v_mountf;
-               mp->m_mounth = vp;
-       } else {
-               mp->m_mounth = vp;
-               vp->v_mountb = &mp->m_mounth;
-               vp->v_mountf = NULL;
+       if (vq = mp->mnt_mounth)
+               vq->v_mountb = &vp->v_mountf;
+       vp->v_mountf = vq;
+       vp->v_mountb = &mp->mnt_mounth;
+       mp->mnt_mounth = vp;
+}
+
+/*
+ * Update outstanding I/O count and do wakeup if requested.
+ */
+vwakeup(bp)
+       register struct buf *bp;
+{
+       register struct vnode *vp;
+
+       bp->b_flags &= ~B_WRITEINPROG;
+       if (vp = bp->b_vp) {
+               vp->v_numoutput--;
+               if (vp->v_numoutput < 0)
+                       panic("vwakeup: neg numoutput");
+               if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) {
+                       if (vp->v_numoutput < 0)
+                               panic("vwakeup: neg numoutput");
+                       vp->v_flag &= ~VBWAIT;
+                       wakeup((caddr_t)&vp->v_numoutput);
+               }
+       }
+}
+
+/*
+ * Flush out and invalidate all buffers associated with a vnode.
+ * Called with the underlying object locked.
+ */
+int
+vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
+       register struct vnode *vp;
+       int flags;
+       struct ucred *cred;
+       struct proc *p;
+       int slpflag, slptimeo;
+{
+       register struct buf *bp;
+       struct buf *nbp, *blist;
+       int s, error;
+
+       if (flags & V_SAVE) {
+               if (error = VOP_FSYNC(vp, cred, MNT_WAIT, p))
+                       return (error);
+               if (vp->v_dirtyblkhd.le_next != NULL)
+                       panic("vinvalbuf: dirty bufs");
+       }
+       for (;;) {
+               if ((blist = vp->v_cleanblkhd.le_next) && flags & V_SAVEMETA)
+                       while (blist && blist->b_lblkno < 0)
+                               blist = blist->b_vnbufs.qe_next;
+               if (!blist && (blist = vp->v_dirtyblkhd.le_next) && 
+                   (flags & V_SAVEMETA))
+                       while (blist && blist->b_lblkno < 0)
+                               blist = blist->b_vnbufs.qe_next;
+               if (!blist)
+                       break;
+
+               for (bp = blist; bp; bp = nbp) {
+                       nbp = bp->b_vnbufs.qe_next;
+                       if (flags & V_SAVEMETA && bp->b_lblkno < 0)
+                               continue;
+                       s = splbio();
+                       if (bp->b_flags & B_BUSY) {
+                               bp->b_flags |= B_WANTED;
+                               error = tsleep((caddr_t)bp,
+                                       slpflag | (PRIBIO + 1), "vinvalbuf",
+                                       slptimeo);
+                               splx(s);
+                               if (error)
+                                       return (error);
+                               break;
+                       }
+                       bremfree(bp);
+                       bp->b_flags |= B_BUSY;
+                       splx(s);
+                       /*
+                        * XXX Since there are no node locks for NFS, I believe
+                        * there is a slight chance that a delayed write will
+                        * occur while sleeping just above, so check for it.
+                        */
+                       if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
+                               (void) VOP_BWRITE(bp);
+                               break;
+                       }
+                       bp->b_flags |= B_INVAL;
+                       brelse(bp);
+               }
+       }
+       if (!(flags & V_SAVEMETA) &&
+           (vp->v_dirtyblkhd.le_next || vp->v_cleanblkhd.le_next))
+               panic("vinvalbuf: flush failed");
+       return (0);
+}
+
+/*
+ * Associate a buffer with a vnode.
+ */
+bgetvp(vp, bp)
+       register struct vnode *vp;
+       register struct buf *bp;
+{
+       register struct vnode *vq;
+
+       if (bp->b_vp)
+               panic("bgetvp: not free");
+       VHOLD(vp);
+       bp->b_vp = vp;
+       if (vp->v_type == VBLK || vp->v_type == VCHR)
+               bp->b_dev = vp->v_rdev;
+       else
+               bp->b_dev = NODEV;
+       /*
+        * Insert onto list for new vnode.
+        */
+       bufinsvn(bp, &vp->v_cleanblkhd);
+}
+
+/*
+ * Disassociate a buffer from a vnode.
+ */
+brelvp(bp)
+       register struct buf *bp;
+{
+       struct vnode *vp;
+
+       if (bp->b_vp == (struct vnode *) 0)
+               panic("brelvp: NULL");
+       /*
+        * Delete from old vnode list, if on one.
+        */
+       if (bp->b_vnbufs.qe_next != NOLIST)
+               bufremvn(bp);
+       vp = bp->b_vp;
+       bp->b_vp = (struct vnode *) 0;
+       HOLDRELE(vp);
+}
+
+/*
+ * Reassign a buffer from one vnode to another.
+ * Used to assign file specific control information
+ * (indirect blocks) to the vnode to which they belong.
+ */
+reassignbuf(bp, newvp)
+       register struct buf *bp;
+       register struct vnode *newvp;
+{
+       register struct list_entry *listheadp;
+
+       if (newvp == NULL) {
+               printf("reassignbuf: NULL");
+               return;
        }
        }
+       /*
+        * Delete from old vnode list, if on one.
+        */
+       if (bp->b_vnbufs.qe_next != NOLIST)
+               bufremvn(bp);
+       /*
+        * If dirty, put on list of dirty buffers;
+        * otherwise insert onto list of clean buffers.
+        */
+       if (bp->b_flags & B_DELWRI)
+               listheadp = &newvp->v_dirtyblkhd;
+       else
+               listheadp = &newvp->v_cleanblkhd;
+       bufinsvn(bp, listheadp);
 }
 
 /*
 }
 
 /*
@@ -309,7 +506,9 @@ bdevvp(dev, vpp)
        struct vnode *nvp;
        int error;
 
        struct vnode *nvp;
        int error;
 
-       error = getnewvnode(VT_NON, (struct mount *)0, &spec_vnodeops, &nvp);
+       if (dev == NODEV)
+               return (0);
+       error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
        if (error) {
                *vpp = 0;
                return (error);
        if (error) {
                *vpp = 0;
                return (error);
@@ -342,7 +541,7 @@ checkalias(nvp, nvp_rdev, mp)
        struct vnode **vpp;
 
        if (nvp->v_type != VBLK && nvp->v_type != VCHR)
        struct vnode **vpp;
 
        if (nvp->v_type != VBLK && nvp->v_type != VCHR)
-               return ((struct vnode *)0);
+               return (NULLVP);
 
        vpp = &speclisth[SPECHASH(nvp_rdev)];
 loop:
 
        vpp = &speclisth[SPECHASH(nvp_rdev)];
 loop:
@@ -361,18 +560,19 @@ loop:
                break;
        }
        if (vp == NULL || vp->v_tag != VT_NON) {
                break;
        }
        if (vp == NULL || vp->v_tag != VT_NON) {
-               if (vp != NULL) {
-                       nvp->v_flag |= VALIASED;
-                       vp->v_flag |= VALIASED;
-                       vput(vp);
-               }
                MALLOC(nvp->v_specinfo, struct specinfo *,
                        sizeof(struct specinfo), M_VNODE, M_WAITOK);
                nvp->v_rdev = nvp_rdev;
                nvp->v_hashchain = vpp;
                nvp->v_specnext = *vpp;
                MALLOC(nvp->v_specinfo, struct specinfo *,
                        sizeof(struct specinfo), M_VNODE, M_WAITOK);
                nvp->v_rdev = nvp_rdev;
                nvp->v_hashchain = vpp;
                nvp->v_specnext = *vpp;
+               nvp->v_specflags = 0;
                *vpp = nvp;
                *vpp = nvp;
-               return ((struct vnode *)0);
+               if (vp != NULL) {
+                       nvp->v_flag |= VALIASED;
+                       vp->v_flag |= VALIASED;
+                       vput(vp);
+               }
+               return (NULLVP);
        }
        VOP_UNLOCK(vp);
        vclean(vp, 0);
        }
        VOP_UNLOCK(vp);
        vclean(vp, 0);
@@ -410,7 +610,7 @@ vget(vp)
                vp->v_freef = NULL;
                vp->v_freeb = NULL;
        }
                vp->v_freef = NULL;
                vp->v_freeb = NULL;
        }
-       VREF(vp);
+       vp->v_usecount++;
        VOP_LOCK(vp);
        return (0);
 }
        VOP_LOCK(vp);
        return (0);
 }
@@ -422,6 +622,8 @@ void vref(vp)
        struct vnode *vp;
 {
 
        struct vnode *vp;
 {
 
+       if (vp->v_usecount <= 0)
+               panic("vref used where vget required");
        vp->v_usecount++;
 }
 
        vp->v_usecount++;
 }
 
@@ -431,6 +633,7 @@ void vref(vp)
 void vput(vp)
        register struct vnode *vp;
 {
 void vput(vp)
        register struct vnode *vp;
 {
+
        VOP_UNLOCK(vp);
        vrele(vp);
 }
        VOP_UNLOCK(vp);
        vrele(vp);
 }
@@ -443,26 +646,24 @@ void vrele(vp)
        register struct vnode *vp;
 {
 
        register struct vnode *vp;
 {
 
+#ifdef DIAGNOSTIC
        if (vp == NULL)
                panic("vrele: null vp");
        if (vp == NULL)
                panic("vrele: null vp");
+#endif
        vp->v_usecount--;
        vp->v_usecount--;
-       if (vp->v_usecount < 0)
-               vprint("vrele: bad ref count", vp);
        if (vp->v_usecount > 0)
                return;
        if (vp->v_usecount > 0)
                return;
-       if (vfreeh == (struct vnode *)0) {
-               /*
-                * insert into empty list
-                */
-               vfreeh = vp;
-               vp->v_freeb = &vfreeh;
-       } else {
-               /*
-                * insert at tail of list
-                */
-               *vfreet = vp;
-               vp->v_freeb = vfreet;
+#ifdef DIAGNOSTIC
+       if (vp->v_usecount != 0 || vp->v_writecount != 0) {
+               vprint("vrele: bad ref count", vp);
+               panic("vrele: ref cnt");
        }
        }
+#endif
+       /*
+        * insert at tail of LRU list
+        */
+       *vfreet = vp;
+       vp->v_freeb = vfreet;
        vp->v_freef = NULL;
        vfreet = &vp->v_freef;
        VOP_INACTIVE(vp);
        vp->v_freef = NULL;
        vfreet = &vp->v_freef;
        VOP_INACTIVE(vp);
@@ -471,7 +672,7 @@ void vrele(vp)
 /*
  * Page or buffer structure gets a reference.
  */
 /*
  * Page or buffer structure gets a reference.
  */
-vhold(vp)
+void vhold(vp)
        register struct vnode *vp;
 {
 
        register struct vnode *vp;
 {
 
@@ -481,7 +682,7 @@ vhold(vp)
 /*
  * Page or buffer structure frees a reference.
  */
 /*
  * Page or buffer structure frees a reference.
  */
-holdrele(vp)
+void holdrele(vp)
        register struct vnode *vp;
 {
 
        register struct vnode *vp;
 {
 
@@ -498,7 +699,8 @@ holdrele(vp)
  * system error). If MNT_FORCE is specified, detach any active vnodes
  * that are found.
  */
  * system error). If MNT_FORCE is specified, detach any active vnodes
  * that are found.
  */
-int busyprt = 0;       /* patch to print out busy vnodes */
+int busyprt = 0;       /* print out busy vnodes */
+struct ctldebug debug1 = { "busyprt", &busyprt };
 
 vflush(mp, skipvp, flags)
        struct mount *mp;
 
 vflush(mp, skipvp, flags)
        struct mount *mp;
@@ -508,14 +710,30 @@ vflush(mp, skipvp, flags)
        register struct vnode *vp, *nvp;
        int busy = 0;
 
        register struct vnode *vp, *nvp;
        int busy = 0;
 
-       for (vp = mp->m_mounth; vp; vp = nvp) {
+       if ((mp->mnt_flag & MNT_MPBUSY) == 0)
+               panic("vflush: not busy");
+loop:
+       for (vp = mp->mnt_mounth; vp; vp = nvp) {
+               if (vp->v_mount != mp)
+                       goto loop;
                nvp = vp->v_mountf;
                /*
                 * Skip over a selected vnode.
                nvp = vp->v_mountf;
                /*
                 * Skip over a selected vnode.
-                * Used by ufs to skip over the quota structure inode.
                 */
                if (vp == skipvp)
                        continue;
                 */
                if (vp == skipvp)
                        continue;
+               /*
+                * Skip over a vnodes marked VSYSTEM.
+                */
+               if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM))
+                       continue;
+               /*
+                * If WRITECLOSE is set, only flush out regular file
+                * vnodes open for writing.
+                */
+               if ((flags & WRITECLOSE) &&
+                   (vp->v_writecount == 0 || vp->v_type != VREG))
+                       continue;
                /*
                 * With v_usecount == 0, all we need to do is clear
                 * out the vnode data structures and we are done.
                /*
                 * With v_usecount == 0, all we need to do is clear
                 * out the vnode data structures and we are done.
@@ -525,15 +743,16 @@ vflush(mp, skipvp, flags)
                        continue;
                }
                /*
                        continue;
                }
                /*
+                * If FORCECLOSE is set, forcibly close the vnode.
                 * For block or character devices, revert to an
                 * anonymous device. For all other files, just kill them.
                 */
                 * For block or character devices, revert to an
                 * anonymous device. For all other files, just kill them.
                 */
-               if (flags & MNT_FORCE) {
+               if (flags & FORCECLOSE) {
                        if (vp->v_type != VBLK && vp->v_type != VCHR) {
                                vgone(vp);
                        } else {
                                vclean(vp, 0);
                        if (vp->v_type != VBLK && vp->v_type != VCHR) {
                                vgone(vp);
                        } else {
                                vclean(vp, 0);
-                               vp->v_op = &spec_vnodeops;
+                               vp->v_op = spec_vnodeop_p;
                                insmntque(vp, (struct mount *)0);
                        }
                        continue;
                                insmntque(vp, (struct mount *)0);
                        }
                        continue;
@@ -550,11 +769,11 @@ vflush(mp, skipvp, flags)
 /*
  * Disassociate the underlying file system from a vnode.
  */
 /*
  * Disassociate the underlying file system from a vnode.
  */
-void vclean(vp, doclose)
+void
+vclean(vp, flags)
        register struct vnode *vp;
        register struct vnode *vp;
-       long doclose;
+       int flags;
 {
 {
-       struct vnodeops *origops;
        int active;
 
        /*
        int active;
 
        /*
@@ -565,6 +784,14 @@ void vclean(vp, doclose)
         */
        if (active = vp->v_usecount)
                VREF(vp);
         */
        if (active = vp->v_usecount)
                VREF(vp);
+       /*
+        * Even if the count is zero, the VOP_INACTIVE routine may still
+        * have the object locked while it cleans it out. The VOP_LOCK
+        * ensures that the VOP_INACTIVE routine is done with its work.
+        * For active vnodes, it ensures that no other activity can
+        * occur while the underlying object is being cleaned out.
+        */
+       VOP_LOCK(vp);
        /*
         * Prevent the vnode from being recycled or
         * brought into use while we clean it out.
        /*
         * Prevent the vnode from being recycled or
         * brought into use while we clean it out.
@@ -573,42 +800,37 @@ void vclean(vp, doclose)
                panic("vclean: deadlock");
        vp->v_flag |= VXLOCK;
        /*
                panic("vclean: deadlock");
        vp->v_flag |= VXLOCK;
        /*
-        * Even if the count is zero, the VOP_INACTIVE routine may still
-        * have the object locked while it cleans it out. The VOP_LOCK
-        * ensures that the VOP_INACTIVE routine is done with its work.
-        * For active vnodes, it ensures that no other activity can
-        * occur while the buffer list is being cleaned out.
+        * Clean out any buffers associated with the vnode.
         */
         */
-       VOP_LOCK(vp);
-       if (doclose)
-               vinvalbuf(vp, 1);
+       if (flags & DOCLOSE)
+               vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0);
        /*
        /*
-        * Prevent any further operations on the vnode from
-        * being passed through to the old file system.
+        * Any other processes trying to obtain this lock must first
+        * wait for VXLOCK to clear, then call the new lock operation.
         */
         */
-       origops = vp->v_op;
-       vp->v_op = &dead_vnodeops;
-       vp->v_tag = VT_NON;
+       VOP_UNLOCK(vp);
        /*
        /*
-        * If purging an active vnode, it must be unlocked, closed,
-        * and deactivated before being reclaimed.
+        * If purging an active vnode, it must be closed and
+        * deactivated before being reclaimed.
         */
         */
-       (*(origops->vn_unlock))(vp);
        if (active) {
        if (active) {
-               if (doclose)
-                       (*(origops->vn_close))(vp, 0, NOCRED);
-               (*(origops->vn_inactive))(vp);
+               if (flags & DOCLOSE)
+                       VOP_CLOSE(vp, IO_NDELAY, NOCRED, NULL);
+               VOP_INACTIVE(vp);
        }
        /*
         * Reclaim the vnode.
         */
        }
        /*
         * Reclaim the vnode.
         */
-       if ((*(origops->vn_reclaim))(vp))
+       if (VOP_RECLAIM(vp))
                panic("vclean: cannot reclaim");
        if (active)
                vrele(vp);
                panic("vclean: cannot reclaim");
        if (active)
                vrele(vp);
+
        /*
        /*
-        * Done with purge, notify sleepers in vget of the grim news.
+        * Done with purge, notify sleepers of the grim news.
         */
         */
+       vp->v_op = dead_vnodeop_p;
+       vp->v_tag = VT_NON;
        vp->v_flag &= ~VXLOCK;
        if (vp->v_flag & VXWANT) {
                vp->v_flag &= ~VXWANT;
        vp->v_flag &= ~VXLOCK;
        if (vp->v_flag & VXWANT) {
                vp->v_flag &= ~VXWANT;
@@ -625,14 +847,36 @@ void vgoneall(vp)
 {
        register struct vnode *vq;
 
 {
        register struct vnode *vq;
 
-       while (vp->v_flag & VALIASED) {
-               for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
-                       if (vq->v_rdev != vp->v_rdev ||
-                           vq->v_type != vp->v_type || vp == vq)
-                               continue;
-                       vgone(vq);
-                       break;
+       if (vp->v_flag & VALIASED) {
+               /*
+                * If a vgone (or vclean) is already in progress,
+                * wait until it is done and return.
+                */
+               if (vp->v_flag & VXLOCK) {
+                       vp->v_flag |= VXWANT;
+                       sleep((caddr_t)vp, PINOD);
+                       return;
+               }
+               /*
+                * Ensure that vp will not be vgone'd while we
+                * are eliminating its aliases.
+                */
+               vp->v_flag |= VXLOCK;
+               while (vp->v_flag & VALIASED) {
+                       for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
+                               if (vq->v_rdev != vp->v_rdev ||
+                                   vq->v_type != vp->v_type || vp == vq)
+                                       continue;
+                               vgone(vq);
+                               break;
+                       }
                }
                }
+               /*
+                * Remove the lock so that vgone below will
+                * really eliminate the vnode after which time
+                * vgone will awaken any sleepers.
+                */
+               vp->v_flag &= ~VXLOCK;
        }
        vgone(vp);
 }
        }
        vgone(vp);
 }
@@ -646,12 +890,20 @@ void vgone(vp)
 {
        register struct vnode *vq;
        struct vnode *vx;
 {
        register struct vnode *vq;
        struct vnode *vx;
-       long count;
 
 
+       /*
+        * If a vgone (or vclean) is already in progress,
+        * wait until it is done and return.
+        */
+       if (vp->v_flag & VXLOCK) {
+               vp->v_flag |= VXWANT;
+               sleep((caddr_t)vp, PINOD);
+               return;
+       }
        /*
         * Clean out the filesystem specific data.
         */
        /*
         * Clean out the filesystem specific data.
         */
-       vclean(vp, 1);
+       vclean(vp, DOCLOSE);
        /*
         * Delete from old mount point vnode list, if on one.
         */
        /*
         * Delete from old mount point vnode list, if on one.
         */
@@ -661,6 +913,7 @@ void vgone(vp)
                *vp->v_mountb = vq;
                vp->v_mountf = NULL;
                vp->v_mountb = NULL;
                *vp->v_mountb = vq;
                vp->v_mountf = NULL;
                vp->v_mountb = NULL;
+               vp->v_mount = NULL;
        }
        /*
         * If special device, remove it from special device alias list.
        }
        /*
         * If special device, remove it from special device alias list.
@@ -679,16 +932,18 @@ void vgone(vp)
                                panic("missing bdev");
                }
                if (vp->v_flag & VALIASED) {
                                panic("missing bdev");
                }
                if (vp->v_flag & VALIASED) {
-                       count = 0;
+                       vx = NULL;
                        for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
                        for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
-                               if (vq->v_rdev != vp->v_rdev)
+                               if (vq->v_rdev != vp->v_rdev ||
+                                   vq->v_type != vp->v_type)
                                        continue;
                                        continue;
-                               count++;
+                               if (vx)
+                                       break;
                                vx = vq;
                        }
                                vx = vq;
                        }
-                       if (count == 0)
+                       if (vx == NULL)
                                panic("missing alias");
                                panic("missing alias");
-                       if (count == 1)
+                       if (vq == NULL)
                                vx->v_flag &= ~VALIASED;
                        vp->v_flag &= ~VALIASED;
                }
                                vx->v_flag &= ~VALIASED;
                        vp->v_flag &= ~VALIASED;
                }
@@ -696,9 +951,10 @@ void vgone(vp)
                vp->v_specinfo = NULL;
        }
        /*
                vp->v_specinfo = NULL;
        }
        /*
-        * If it is on the freelist, move it to the head of the list.
+        * If it is on the freelist and not already at the head,
+        * move it to the head of the list.
         */
         */
-       if (vp->v_freeb) {
+       if (vp->v_freeb && vfreeh != vp) {
                if (vq = vp->v_freef)
                        vq->v_freeb = vp->v_freeb;
                else
                if (vq = vp->v_freef)
                        vq->v_freeb = vp->v_freeb;
                else
@@ -726,9 +982,9 @@ vfinddev(dev, type, vpp)
                if (dev != vp->v_rdev || type != vp->v_type)
                        continue;
                *vpp = vp;
                if (dev != vp->v_rdev || type != vp->v_type)
                        continue;
                *vpp = vp;
-               return (0);
+               return (1);
        }
        }
-       return (1);
+       return (0);
 }
 
 /*
 }
 
 /*
@@ -744,7 +1000,7 @@ vcount(vp)
                return (vp->v_usecount);
 loop:
        for (count = 0, vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
                return (vp->v_usecount);
 loop:
        for (count = 0, vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
-               if (vq->v_rdev != vp->v_rdev)
+               if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
                        continue;
                /*
                 * Alias, but not in use, so flush it out.
                        continue;
                /*
                 * Alias, but not in use, so flush it out.
@@ -762,7 +1018,7 @@ loop:
  * Print out a description of a vnode.
  */
 static char *typename[] =
  * Print out a description of a vnode.
  */
 static char *typename[] =
-       { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VBAD" };
+   { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" };
 
 vprint(label, vp)
        char *label;
 
 vprint(label, vp)
        char *label;
@@ -772,39 +1028,111 @@ vprint(label, vp)
 
        if (label != NULL)
                printf("%s: ", label);
 
        if (label != NULL)
                printf("%s: ", label);
-       printf("type %s, usecount %d, refcount %d,", typename[vp->v_type],
-               vp->v_usecount, vp->v_holdcnt);
+       printf("type %s, usecount %d, writecount %d, refcount %d,",
+               typename[vp->v_type], vp->v_usecount, vp->v_writecount,
+               vp->v_holdcnt);
        buf[0] = '\0';
        if (vp->v_flag & VROOT)
                strcat(buf, "|VROOT");
        if (vp->v_flag & VTEXT)
                strcat(buf, "|VTEXT");
        buf[0] = '\0';
        if (vp->v_flag & VROOT)
                strcat(buf, "|VROOT");
        if (vp->v_flag & VTEXT)
                strcat(buf, "|VTEXT");
+       if (vp->v_flag & VSYSTEM)
+               strcat(buf, "|VSYSTEM");
        if (vp->v_flag & VXLOCK)
                strcat(buf, "|VXLOCK");
        if (vp->v_flag & VXWANT)
                strcat(buf, "|VXWANT");
        if (vp->v_flag & VXLOCK)
                strcat(buf, "|VXLOCK");
        if (vp->v_flag & VXWANT)
                strcat(buf, "|VXWANT");
-       if (vp->v_flag & VEXLOCK)
-               strcat(buf, "|VEXLOCK");
-       if (vp->v_flag & VSHLOCK)
-               strcat(buf, "|VSHLOCK");
-       if (vp->v_flag & VLWAIT)
-               strcat(buf, "|VLWAIT");
-       if (vp->v_flag & VALIASED)
-               strcat(buf, "|VALIASED");
        if (vp->v_flag & VBWAIT)
                strcat(buf, "|VBWAIT");
        if (vp->v_flag & VBWAIT)
                strcat(buf, "|VBWAIT");
+       if (vp->v_flag & VALIASED)
+               strcat(buf, "|VALIASED");
        if (buf[0] != '\0')
                printf(" flags (%s)", &buf[1]);
        printf("\n\t");
        VOP_PRINT(vp);
 }
 
        if (buf[0] != '\0')
                printf(" flags (%s)", &buf[1]);
        printf("\n\t");
        VOP_PRINT(vp);
 }
 
-strcat(src, append)
-       register char *src, *append;
+#ifdef DEBUG
+/*
+ * List all of the locked vnodes in the system.
+ * Called when debugging the kernel.
+ */
+printlockedvnodes()
 {
 {
+       register struct mount *mp;
+       register struct vnode *vp;
 
 
-       for (; *src; ++src)
-               /* void */;
-       while (*src++ = *append++)
-               /* void */;
+       printf("Locked vnodes\n");
+       mp = rootfs;
+       do {
+               for (vp = mp->mnt_mounth; vp; vp = vp->v_mountf)
+                       if (VOP_ISLOCKED(vp))
+                               vprint((char *)0, vp);
+               mp = mp->mnt_next;
+       } while (mp != rootfs);
+}
+#endif
+
+int kinfo_vdebug = 1;
+int kinfo_vgetfailed;
+#define KINFO_VNODESLOP        10
+/*
+ * Dump vnode list (via sysctl).
+ * Copyout address of vnode followed by vnode.
+ */
+/* ARGSUSED */
+sysctl_vnode(where, sizep)
+       char *where;
+       size_t *sizep;
+{
+       register struct mount *mp = rootfs;
+       struct mount *omp;
+       struct vnode *vp;
+       register char *bp = where, *savebp;
+       char *ewhere;
+       int error;
+
+#define VPTRSZ sizeof (struct vnode *)
+#define VNODESZ        sizeof (struct vnode)
+       if (where == NULL) {
+               *sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
+               return (0);
+       }
+       ewhere = where + *sizep;
+               
+       do {
+               if (vfs_busy(mp)) {
+                       mp = mp->mnt_next;
+                       continue;
+               }
+               savebp = bp;
+again:
+               for (vp = mp->mnt_mounth; vp; vp = vp->v_mountf) {
+                       /*
+                        * Check that the vp is still associated with
+                        * this filesystem.  RACE: could have been
+                        * recycled onto the same filesystem.
+                        */
+                       if (vp->v_mount != mp) {
+                               if (kinfo_vdebug)
+                                       printf("kinfo: vp changed\n");
+                               bp = savebp;
+                               goto again;
+                       }
+                       if (bp + VPTRSZ + VNODESZ > ewhere) {
+                               *sizep = bp - where;
+                               return (ENOMEM);
+                       }
+                       if ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) ||
+                          (error = copyout((caddr_t)vp, bp + VPTRSZ, VNODESZ)))
+                               return (error);
+                       bp += VPTRSZ + VNODESZ;
+               }
+               omp = mp;
+               mp = mp->mnt_next;
+               vfs_unbusy(omp);
+       } while (mp != rootfs);
+
+       *sizep = bp - where;
+       return (0);
 }
 }