]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/vfs/vfs_subr.c
xnu-344.49.tar.gz
[apple/xnu.git] / bsd / vfs / vfs_subr.c
index 216c1a0e4ff7127f9c5eb34e18960bdb15331fb9..742af69fd44626bc2368ebc33679018898f2ff49 100644 (file)
@@ -1,21 +1,24 @@
 /*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2001 Apple Computer, Inc. All rights reserved.
  *
  * @APPLE_LICENSE_HEADER_START@
  * 
- * The contents of this file constitute Original Code as defined in and
- * are subject to the Apple Public Source License Version 1.1 (the
- * "License").  You may not use this file except in compliance with the
- * License.  Please obtain a copy of the License at
- * http://www.apple.com/publicsource and read it before using this file.
+ * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
  * 
- * This Original Code and all software distributed under the License are
- * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this
+ * file.
+ * 
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
- * License for the specific language governing rights and limitations
- * under the License.
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
  * 
  * @APPLE_LICENSE_HEADER_END@
  */
@@ -64,6 +67,7 @@
  * External virtual filesystem routines
  */
 
+#undef DIAGNOSTIC
 #define DIAGNOSTIC 1
 
 #include <sys/param.h>
 
 #include <miscfs/specfs/specdev.h>
 
+#include <mach/mach_types.h>
+#include <mach/memory_object_types.h>
+
+
 enum vtype iftovt_tab[16] = {
        VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
        VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
@@ -100,19 +108,10 @@ int       vttoif_tab[9] = {
 
 static void vfree(struct vnode *vp);
 static void vinactive(struct vnode *vp);
-extern int vnreclaim(int count);
+static int vnreclaim(int count);
 extern kern_return_t 
        adjust_vm_object_cache(vm_size_t oval, vm_size_t nval);
 
-/*
- * Insq/Remq for the vnode usage lists.
- */
-#define        bufinsvn(bp, dp)        LIST_INSERT_HEAD(dp, bp, b_vnbufs)
-#define        bufremvn(bp) {                                                  \
-       LIST_REMOVE(bp, b_vnbufs);                                      \
-       (bp)->b_vnbufs.le_next = NOLIST;                                \
-}
-
 TAILQ_HEAD(freelst, vnode) vnode_free_list;    /* vnode free list */
 TAILQ_HEAD(inactivelst, vnode) vnode_inactive_list;    /* vnode inactive list */
 struct mntlist mountlist;                      /* mounted filesystem list */
@@ -207,7 +206,7 @@ unsigned long vnodetarget;          /* target for vnreclaim() */
  * cache. Having too few vnodes on the free list causes serious disk
  * thrashing as we cycle through them.
  */
-#define VNODE_FREE_MIN         100     /* freelist should have at least these many */
+#define VNODE_FREE_MIN         300     /* freelist should have at least these many */
 
 /*
  * We need to get vnodes back from the VM object cache when a certain #
@@ -226,7 +225,7 @@ unsigned long vnodetarget;          /* target for vnreclaim() */
 /*
  * Initialize the vnode management data structures.
  */
-void
+__private_extern__ void
 vntblinit()
 {
        extern struct lock__bsd__       exchangelock;
@@ -252,11 +251,16 @@ vntblinit()
 }
 
 /* Reset the VM Object Cache with the values passed in */
-kern_return_t
+__private_extern__ kern_return_t
 reset_vmobjectcache(unsigned int val1, unsigned int val2)
 {
        vm_size_t oval = val1 - VNODE_FREE_MIN;
-       vm_size_t nval = val2 - VNODE_FREE_MIN;
+       vm_size_t nval;
+       
+       if(val2 < VNODE_FREE_MIN)
+               nval = 0;
+       else
+               nval = val2 - VNODE_FREE_MIN;
 
        return(adjust_vm_object_cache(oval, nval));
 }
@@ -334,6 +338,11 @@ vfs_rootmountalloc(fstypename, devname, mpp)
                return (ENODEV);
        mp = _MALLOC_ZONE((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
        bzero((char *)mp, (u_long)sizeof(struct mount));
+
+    /* Initialize the default IO constraints */
+    mp->mnt_maxreadcnt = mp->mnt_maxwritecnt = MAXPHYS;
+    mp->mnt_segreadcnt = mp->mnt_segwritecnt = 32;
+
        lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
        (void)vfs_busy(mp, LK_NOWAIT, 0, p);
        LIST_INIT(&mp->mnt_vnodelist);
@@ -626,6 +635,20 @@ insmntque(vp, mp)
        simple_unlock(&mntvnode_slock);
 }
 
+__inline void
+vpwakeup(struct vnode *vp)
+{
+       if (vp) {
+               if (--vp->v_numoutput < 0)
+                       panic("vpwakeup: neg numoutput");
+               if ((vp->v_flag & VBWAIT || vp->v_flag & VTHROTTLED)
+                   && vp->v_numoutput <= 0) {
+                       vp->v_flag &= ~(VBWAIT|VTHROTTLED);
+                       wakeup((caddr_t)&vp->v_numoutput);
+               }
+       }
+}
+
 /*
  * Update outstanding I/O count and do wakeup if requested.
  */
@@ -633,19 +656,8 @@ void
 vwakeup(bp)
        register struct buf *bp;
 {
-       register struct vnode *vp;
-
        CLR(bp->b_flags, B_WRITEINPROG);
-       if (vp = bp->b_vp) {
-               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 2");
-                       vp->v_flag &= ~VBWAIT;
-                       wakeup((caddr_t)&vp->v_numoutput);
-               }
-       }
+       vpwakeup(bp->b_vp);
 }
 
 /*
@@ -668,12 +680,22 @@ vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
                if (error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) {
                        return (error);
                }
-               if (vp->v_dirtyblkhd.lh_first != NULL || (vp->v_flag & VHASDIRTY))
-                       panic("vinvalbuf: dirty bufs");
+
+               // XXXdbg - if there are dirty bufs, wait for 'em if they're busy
+               for (bp=vp->v_dirtyblkhd.lh_first; bp; bp=nbp) {
+                   nbp = bp->b_vnbufs.le_next;
+                   if (ISSET(bp->b_flags, B_BUSY)) {
+                       SET(bp->b_flags, B_WANTED);
+                       tsleep((caddr_t)bp, slpflag | (PRIBIO + 1), "vinvalbuf", 0);
+                       nbp = vp->v_dirtyblkhd.lh_first;
+                   } else {
+                       panic("vinvalbuf: dirty buf (vp 0x%x, bp 0x%x)", vp, bp);
+                   }
+               }
        }
 
        for (;;) {
-               if ((blist = vp->v_cleanblkhd.lh_first) && flags & V_SAVEMETA)
+               if ((blist = vp->v_cleanblkhd.lh_first) && (flags & V_SAVEMETA))
                        while (blist && blist->b_lblkno < 0)
                                blist = blist->b_vnbufs.le_next;
                if (!blist && (blist = vp->v_dirtyblkhd.lh_first) &&
@@ -685,7 +707,7 @@ vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
 
                for (bp = blist; bp; bp = nbp) {
                        nbp = bp->b_vnbufs.le_next;
-                       if (flags & V_SAVEMETA && bp->b_lblkno < 0)
+                       if ((flags & V_SAVEMETA) && bp->b_lblkno < 0)
                                continue;
                        s = splbio();
                        if (ISSET(bp->b_flags, B_BUSY)) {
@@ -711,7 +733,13 @@ vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
                                (void) VOP_BWRITE(bp);
                                break;
                        }
-                       SET(bp->b_flags, B_INVAL);
+
+                       if (bp->b_flags & B_LOCKED) {
+                               panic("vinvalbuf: bp @ 0x%x is locked!\n", bp);
+                               break;
+                       } else {
+                               SET(bp->b_flags, B_INVAL);
+                       }
                        brelse(bp);
                }
        }
@@ -721,82 +749,6 @@ vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
        return (0);
 }
 
-/*
- * Associate a buffer with a vnode.
- */
-void
-bgetvp(vp, bp)
-       register struct vnode *vp;
-       register struct buf *bp;
-{
-
-       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.
- */
-void
-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.le_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.
- */
-void
-reassignbuf(bp, newvp)
-       register struct buf *bp;
-       register struct vnode *newvp;
-{
-       register struct buflists *listheadp;
-
-       if (newvp == NULL) {
-               printf("reassignbuf: NULL");
-               return;
-       }
-       /*
-        * Delete from old vnode list, if on one.
-        */
-       if (bp->b_vnbufs.le_next != NOLIST)
-               bufremvn(bp);
-       /*
-        * If dirty, put on list of dirty buffers;
-        * otherwise insert onto list of clean buffers.
-        */
-       if (ISSET(bp->b_flags, B_DELWRI))
-               listheadp = &newvp->v_dirtyblkhd;
-       else
-               listheadp = &newvp->v_cleanblkhd;
-       bufinsvn(bp, listheadp);
-}
-
 /*
  * Create a vnode for a block device.
  * Used for root filesystem, argdev, and swap areas.
@@ -908,12 +860,14 @@ loop:
 }
 
 /*
- * Grab a particular vnode from the free list, increment its
- * reference count and lock it. The vnode lock bit is set the
- * vnode is being eliminated in vgone. The process is awakened
- * when the transition is completed, and an error returned to
- * indicate that the vnode is no longer usable (possibly having
- * been changed to a new file system type).
+ * Get a reference on a particular vnode and lock it if requested.
+ * If the vnode was on the inactive list, remove it from the list.
+ * If the vnode was on the free list, remove it from the list and
+ * move it to inactive list as needed.
+ * The vnode lock bit is set if the vnode is being eliminated in
+ * vgone. The process is awakened when the transition is completed,
+ * and an error returned to indicate that the vnode is no longer
+ * usable (possibly having been changed to a new file system type).
  */
 int
 vget(vp, flags, p)
@@ -923,6 +877,8 @@ vget(vp, flags, p)
 {
        int error = 0;
 
+retry:
+
        /*
         * If the vnode is in the process of being cleaned out for
         * another use, we wait for the cleaning to finish and then
@@ -934,7 +890,7 @@ vget(vp, flags, p)
        if ((vp->v_flag & VXLOCK) || (vp->v_flag & VORECLAIM)) {
                vp->v_flag |= VXWANT;
                simple_unlock(&vp->v_interlock);
-               tsleep((caddr_t)vp, PINOD, "vget", 0);
+               (void)tsleep((caddr_t)vp, PINOD, "vget", 0);
                return (ENOENT);
        }
 
@@ -945,13 +901,26 @@ vget(vp, flags, p)
        if (ISSET(vp->v_flag, VTERMINATE)) {
                SET(vp->v_flag, VTERMWANT);
                simple_unlock(&vp->v_interlock);
-               tsleep((caddr_t)&vp->v_ubcinfo, PINOD, "vclean", 0);
+               (void)tsleep((caddr_t)&vp->v_ubcinfo, PINOD, "vclean", 0);
                return (ENOENT);
        }
 
+       /*
+        * if the vnode is being initialized,
+        * wait for it to finish initialization
+        */
+       if (ISSET(vp->v_flag,  VUINIT)) {
+               if (ISSET(vp->v_flag,  VUINIT)) {
+                       SET(vp->v_flag, VUWANT);
+                       simple_unlock(&vp->v_interlock);
+                       (void) tsleep((caddr_t)vp, PINOD, "vget2", 0);
+                       goto retry;
+               }
+       }
+
        simple_lock(&vnode_free_list_slock);
-       /* If on the free list, remove it from there */
        if (vp->v_usecount == 0) {
+               /* If on the free list, remove it from there */
                if (VONLIST(vp))
                        VREMFREE("vget", vp);
        } else {
@@ -966,74 +935,99 @@ vget(vp, flags, p)
        VINACTIVECHECK("vget", vp, 0);
 
        simple_unlock(&vnode_free_list_slock);
+
        if (++vp->v_usecount <= 0)
                panic("vget: v_usecount");                     
 
-       if (flags & LK_TYPE_MASK) {
-               if (error = vn_lock(vp, flags | LK_INTERLOCK, p)) {
-                       /*
-                        * If the vnode was not active in the first place
-                        * must not call vrele() as VOP_INACTIVE() is not
-                        * required.
-                        * So inlined part of vrele() here.
-                        */
-                       simple_lock(&vp->v_interlock);
-                       if (--vp->v_usecount == 1) {
-                               if (UBCINFOEXISTS(vp)) {
-                                       vinactive(vp);
-                                       simple_unlock(&vp->v_interlock);
-                                       return (error);
-                               }
-                       }
-                       if (vp->v_usecount > 0) {
-                               simple_unlock(&vp->v_interlock);
-                               return (error);
-                       }
-                       if (vp->v_usecount < 0)
-                               panic("vget: negative usecount (%d)", vp->v_usecount);
-                       vfree(vp);
-                       simple_unlock(&vp->v_interlock);
+       /*
+        * Recover named reference as needed
+        */
+       if (UBCISVALID(vp) && !ubc_issetflags(vp, UI_HASOBJREF)) {
+               simple_unlock(&vp->v_interlock);
+               if (ubc_getobject(vp, UBC_HOLDOBJECT)) {
+                       error = ENOENT;
+                       goto errout;
                }
-               return (error);
+               simple_lock(&vp->v_interlock);
+       }
+
+       if (flags & LK_TYPE_MASK) {
+               if (error = vn_lock(vp, flags | LK_INTERLOCK, p))
+                       goto errout;
+               return (0);
        }
 
+       if ((flags & LK_INTERLOCK) == 0)
+               simple_unlock(&vp->v_interlock);
+       return (0);
+
+errout:
        /*
-        * If this is a valid UBC vnode, if usecount is 1 and if
-        * this vnode was mapped in the past, it is likely
-        * that ubc_info freed due to the memory object getting recycled.
-        * Just re-initialize the ubc_info.
+        * If the vnode was not active in the first place
+        * must not call vrele() as VOP_INACTIVE() is not
+        * required.
+        * So inlined part of vrele() here.
         */
-       if ((vp->v_usecount == 1) && UBCISVALID(vp)) {
-               if (UBCINFOMISSING(vp))
-                       panic("vget: lost ubc_info");
-
-               if (ISSET(vp->v_flag, VTERMINATE)) {
-                       /* 
-                        * vnode is being terminated.
-                        * wait for vnode_pager_no_senders() to clear
-                        * VTERMINATE
-                        */
-                       SET(vp->v_flag, VTERMWANT);
+       simple_lock(&vp->v_interlock);
+       if (--vp->v_usecount == 1) {
+               if (UBCINFOEXISTS(vp)) {
+                       vinactive(vp);
                        simple_unlock(&vp->v_interlock);
-                       tsleep((caddr_t)&vp->v_ubcinfo, PINOD, "vclean", 0);
-                       /* return error */
-                       return (ENOENT);
+                       return (error);
                }
-
-               if ((!UBCINFOEXISTS(vp)) && ISSET(vp->v_flag, VWASMAPPED)) {
-                       simple_unlock(&vp->v_interlock);
-                       ubc_info_init(vp);
-                       simple_lock(&vp->v_interlock);
-               } else
-                       panic("vget: stolen ubc_info");
-
-               if (!ubc_issetflags(vp, UI_HASOBJREF))
-                       if (ubc_getobject(vp, (UBC_NOREACTIVATE|UBC_HOLDOBJECT)))
-                               panic("vget: null object");
        }
-out:
-       if ((flags & LK_INTERLOCK) == 0)
+       if (vp->v_usecount > 0) {
                simple_unlock(&vp->v_interlock);
+               return (error);
+       }
+       if (vp->v_usecount < 0)
+               panic("vget: negative usecount (%d)", vp->v_usecount);
+       vfree(vp);
+       simple_unlock(&vp->v_interlock);
+       return (error);
+}
+
+/*
+ * Get a pager reference on the particular vnode.
+ *
+ * This is called from ubc_info_init() and it is asumed that
+ * the vnode is neither on the free list on on the inactive list.
+ * It is also assumed that the vnode is neither being recycled
+ * by vgonel nor being terminated by vnode_pager_vrele().
+ *
+ * The vnode interlock is NOT held by the caller.
+ */
+__private_extern__ int
+vnode_pager_vget(vp)
+       struct vnode *vp;
+{
+       simple_lock(&vp->v_interlock);
+       if (UBCINFOMISSING(vp))
+               panic("vnode_pager_vget: stolen ubc_info");
+
+       if (!UBCINFOEXISTS(vp))
+               panic("vnode_pager_vget: lost ubc_info");
+
+       if ((vp->v_flag & VXLOCK) || (vp->v_flag & VORECLAIM))
+               panic("vnode_pager_vget: already being reclaimd");
+
+       if (ISSET(vp->v_flag, VTERMINATE))
+               panic("vnode_pager_vget: already being terminated");
+
+       simple_lock(&vnode_free_list_slock);
+       /* The vnode should not be on ANY list */
+       if (VONLIST(vp))
+               panic("vnode_pager_vget: still on the list");
+
+       /* The vnode should not be on the inactive list here */
+       VINACTIVECHECK("vnode_pager_vget", vp, 0);
+       simple_unlock(&vnode_free_list_slock);
+
+       /* After all those checks, now do the real work :-) */
+       if (++vp->v_usecount <= 0)
+               panic("vnode_pager_vget: v_usecount");                     
+       simple_unlock(&vp->v_interlock);
+
        return (0);
 }
 
@@ -1242,10 +1236,6 @@ vput(vp)
 {
        struct proc *p = current_proc();        /* XXX */
 
-#if DIAGNOSTIC
-       if (vp == NULL)
-               panic("vput: null vp");
-#endif
        simple_lock(&vp->v_interlock);
        if (--vp->v_usecount == 1) {
                if (UBCINFOEXISTS(vp)) {
@@ -1298,10 +1288,6 @@ vrele(vp)
 {
        struct proc *p = current_proc();        /* XXX */
 
-#if DIAGNOSTIC
-       if (vp == NULL)
-               panic("vrele: null vp");
-#endif
        simple_lock(&vp->v_interlock);
        if (--vp->v_usecount == 1) {
                if (UBCINFOEXISTS(vp)) {
@@ -1361,7 +1347,6 @@ void
 vagevp(vp)
        struct vnode *vp;
 {
-       assert(vp);
        simple_lock(&vp->v_interlock);
        vp->v_flag |= VAGE;
        simple_unlock(&vp->v_interlock);
@@ -1417,7 +1402,7 @@ vflush(mp, skipvp, flags)
        struct vnode *skipvp;
        int flags;
 {
-       struct proc *p = current_proc();        /* XXX */
+       struct proc *p = current_proc();
        struct vnode *vp, *nvp;
        int busy = 0;
 
@@ -1435,9 +1420,9 @@ loop:
 
                simple_lock(&vp->v_interlock);
                /*
-                * Skip over a vnodes marked VSYSTEM.
+                * Skip over a vnodes marked VSYSTEM or VNOFLUSH.
                 */
-               if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
+               if ((flags & SKIPSYSTEM) && ((vp->v_flag & VSYSTEM) || (vp->v_flag & VNOFLUSH))) {
                        simple_unlock(&vp->v_interlock);
                        continue;
                }
@@ -1492,7 +1477,7 @@ loop:
                busy++;
        }
        simple_unlock(&mntvnode_slock);
-       if (busy)
+       if (busy && ((flags & FORCECLOSE)==0))
                return (EBUSY);
        return (0);
 }
@@ -1508,8 +1493,8 @@ vclean(vp, flags, p)
        struct proc *p;
 {
        int active;
-       void *obj;
        int removed = 0;
+       int didhold;
 
        /*
         * if the vnode is not obtained by calling getnewvnode() we
@@ -1559,42 +1544,40 @@ vclean(vp, flags, p)
        }
 
        /* Clean the pages in VM. */
-       if ((active) && UBCINFOEXISTS(vp)) {
+       if (active && (flags & DOCLOSE))
+               VOP_CLOSE(vp, IO_NDELAY, NOCRED, p);
+
+       /* Clean the pages in VM. */
+       didhold = ubc_hold(vp);
+       if ((active) && (didhold))
                (void)ubc_clean(vp, 0); /* do not invalidate */
-       }
 
        /*
         * Clean out any buffers associated with the vnode.
         */
        if (flags & DOCLOSE) {
                if (vp->v_tag == VT_NFS)
-                       nfs_vinvalbuf(vp, V_SAVE, NOCRED, p, 0);
-               else
-                       vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
-       }
-       /*
-        * If purging an active vnode, it must be closed and
-        * deactivated before being reclaimed. Note that the
-        * VOP_INACTIVE will unlock the vnode.
-        */
-       if (active) {
-               if (flags & DOCLOSE)
-                       VOP_CLOSE(vp, IO_NDELAY, NOCRED, p);
+            nfs_vinvalbuf(vp, V_SAVE, NOCRED, p, 0);
+        else
+            vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
+    }
+
+       if (active)
                VOP_INACTIVE(vp, p);
-       } else {
-               /*
-                * Any other processes trying to obtain this lock must first
-                * wait for VXLOCK to clear, then call the new lock operation.
-                */
+       else
                VOP_UNLOCK(vp, 0, p);
+
+       /* Destroy ubc named reference */
+    if (didhold) {
+        ubc_rele(vp);
+               ubc_destroy_named(vp);
        }
+
        /*
         * Reclaim the vnode.
         */
        if (VOP_RECLAIM(vp, p))
                panic("vclean: cannot reclaim");
-       if (active)
-               vrele(vp);
        cache_purge(vp);
        if (vp->v_vnlock) {
                if ((vp->v_vnlock->lk_flags & LK_DRAINED) == 0)
@@ -1607,41 +1590,6 @@ vclean(vp, flags, p)
        vp->v_op = dead_vnodeop_p;
        vp->v_tag = VT_NON;
 
-       /*
-        * v_data is reclaimed by VOP_RECLAIM, all the vnode
-        * operation generated by the code below would be directed
-        * to the deadfs
-        */
-       if (UBCINFOEXISTS(vp)) {
-               /* vnode is dying, destroy the object */
-               if (ubc_issetflags(vp, UI_HASOBJREF)) {
-                       obj = ubc_getobject(vp, UBC_NOREACTIVATE);
-                       if (obj == NULL)
-                               panic("vclean: null object");
-                       if (ISSET(vp->v_flag, VTERMINATE))
-                               panic("vclean: already teminating");
-                       SET(vp->v_flag, VTERMINATE);
-
-                       ubc_clearflags(vp, UI_HASOBJREF);
-                       memory_object_destroy(obj, 0);
-
-                       /* 
-                        * memory_object_destroy() is asynchronous with respect
-                        * to vnode_pager_no_senders().
-                        * wait for vnode_pager_no_senders() to clear
-                        * VTERMINATE
-                        */
-                       while (ISSET(vp->v_flag, VTERMINATE)) {
-                               SET(vp->v_flag, VTERMWANT);
-                               tsleep((caddr_t)&vp->v_ubcinfo, PINOD, "vclean", 0);
-                       }
-                       if (UBCINFOEXISTS(vp)) {
-                               ubc_info_free(vp);
-                               vp->v_ubcinfo = UBC_NOINFO;  /* catch bad accesses */
-                       }
-               }
-       }
-
        /*
         * Done with purge, notify sleepers of the grim news.
         */
@@ -1650,6 +1598,9 @@ vclean(vp, flags, p)
                vp->v_flag &= ~VXWANT;
                wakeup((caddr_t)vp);
        }
+
+       if (active)
+               vrele(vp);
 }
 
 /*
@@ -1664,7 +1615,7 @@ vop_revoke(ap)
        } */ *ap;
 {
        struct vnode *vp, *vq;
-       struct proc *p = current_proc();        /* XXX */
+       struct proc *p = current_proc();
 
 #if DIAGNOSTIC
        if ((ap->a_flags & REVOKEALL) == 0)
@@ -1683,7 +1634,7 @@ vop_revoke(ap)
                        while (vp->v_flag & VXLOCK) {
                                vp->v_flag |= VXWANT;
                                simple_unlock(&vp->v_interlock);
-                               tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0);
+                               (void)tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0);
                        }
                        return (0);
                }
@@ -1748,7 +1699,7 @@ void
 vgone(vp)
        struct vnode *vp;
 {
-       struct proc *p = current_proc();        /* XXX */
+       struct proc *p = current_proc();
 
        simple_lock(&vp->v_interlock);
        vgonel(vp, p);
@@ -1782,7 +1733,7 @@ vgonel(vp, p)
                while (vp->v_flag & VXLOCK) {
                        vp->v_flag |= VXWANT;
                        simple_unlock(&vp->v_interlock);
-                       tsleep((caddr_t)vp, PINOD, "vgone", 0);
+                       (void)tsleep((caddr_t)vp, PINOD, "vgone", 0);
                }
                return;
        }
@@ -1941,6 +1892,8 @@ vprint(label, vp)
                strcat(buf, "|VTEXT");
        if (vp->v_flag & VSYSTEM)
                strcat(buf, "|VSYSTEM");
+       if (vp->v_flag & VNOFLUSH)
+               strcat(buf, "|VNOFLUSH");
        if (vp->v_flag & VXLOCK)
                strcat(buf, "|VXLOCK");
        if (vp->v_flag & VXWANT)
@@ -1967,7 +1920,7 @@ vprint(label, vp)
 void
 printlockedvnodes()
 {
-       struct proc *p = current_proc();        /* XXX */
+       struct proc *p = current_proc();
        struct mount *mp, *nmp;
        struct vnode *vp;
 
@@ -2005,10 +1958,19 @@ vfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
        size_t newlen;
        struct proc *p;
 {
-       struct ctldebug *cdp;
        struct vfsconf *vfsp;
 
-       if (name[0] == VFS_NUMMNTOPS) {
+       /*
+        * The VFS_NUMMNTOPS shouldn't be at name[0] since
+        * is a VFS generic variable. So now we must check
+        * namelen so we don't end up covering any UFS
+        * variables (sinc UFS vfc_typenum is 1).
+        *
+        * It should have been:
+        *    name[0]:  VFS_GENERIC
+        *    name[1]:  VFS_NUMMNTOPS
+        */
+       if (namelen == 1 && name[0] == VFS_NUMMNTOPS) {
                extern unsigned int vfs_nummntops;
                return (sysctl_rdint(oldp, oldlenp, newp, vfs_nummntops));
        }
@@ -2149,11 +2111,11 @@ vfs_mountedon(vp)
  * Unmount all filesystems. The list is traversed in reverse order
  * of mounting to avoid dependencies.
  */
-void
+__private_extern__ void
 vfs_unmountall()
 {
        struct mount *mp, *nmp;
-       struct proc *p = current_proc();        /* XXX */
+       struct proc *p = current_proc();
 
        /*
         * Since this only runs when rebooting, it is not interlocked.
@@ -2166,7 +2128,7 @@ vfs_unmountall()
 
 /*
  * Build hash lists of net addresses and hang them off the mount point.
- * Called by ufs_mount() to set up the lists of export addresses.
+ * Called by vfs_export() to set up the lists of export addresses.
  */
 static int
 vfs_hang_addrlist(mp, nep, argp)
@@ -2343,7 +2305,7 @@ vfs_export_lookup(mp, nep, nam)
  * try to reclaim vnodes from the memory 
  * object cache
  */
-int
+static int
 vm_object_cache_reclaim(int count)
 {
        int cnt;
@@ -2360,11 +2322,10 @@ vm_object_cache_reclaim(int count)
  * and then try to reclaim some vnodes from the memory 
  * object cache
  */
-int
+static int
 vnreclaim(int count)
 {
-       int cnt, i, loopcnt;
-       void *obj;
+       int i, loopcnt;
        struct vnode *vp;
        int err;
        struct proc *p;
@@ -2390,163 +2351,123 @@ restart:
        for (vp = TAILQ_FIRST(&vnode_inactive_list);
                        (vp != NULLVP) && (i < count);
                        vp = TAILQ_NEXT(vp, v_freelist)) {
+               
+               if (!simple_lock_try(&vp->v_interlock))
+                       continue;
 
-               if (simple_lock_try(&vp->v_interlock)) {
-                       if (vp->v_usecount != 1)
-                               panic("vnreclaim: v_usecount");
-
-                       if(!UBCINFOEXISTS(vp)) {
-                               if (vp->v_type == VBAD) {
-                                       VREMINACTIVE("vnreclaim", vp);
-                                       simple_unlock(&vp->v_interlock);
-                                       continue;
-                               } else
-                                       panic("non UBC vnode on inactive list");
-                               /* Should not reach here */
-                       }
+               if (vp->v_usecount != 1)
+                       panic("vnreclaim: v_usecount");
 
-                       /* If vnode is already being reclaimed, wait */
-                       if ((vp->v_flag & VXLOCK) || (vp->v_flag & VORECLAIM)) {
-                               vp->v_flag |= VXWANT;
+               if(!UBCINFOEXISTS(vp)) {
+                       if (vp->v_type == VBAD) {
+                               VREMINACTIVE("vnreclaim", vp);
                                simple_unlock(&vp->v_interlock);
-                               simple_unlock(&vnode_free_list_slock);
-                               (void)tsleep((caddr_t)vp, PINOD, "vocr", 0);
-                               goto restart;
-                       }
+                               continue;
+                       } else
+                               panic("non UBC vnode on inactive list");
+                               /* Should not reach here */
+               }
 
-                       VREMINACTIVE("vnreclaim", vp);
+               /* If vnode is already being reclaimed, wait */
+               if ((vp->v_flag & VXLOCK) || (vp->v_flag & VORECLAIM)) {
+                       vp->v_flag |= VXWANT;
+                       simple_unlock(&vp->v_interlock);
                        simple_unlock(&vnode_free_list_slock);
+                       (void)tsleep((caddr_t)vp, PINOD, "vocr", 0);
+                       goto restart;
+               }
 
-                       /* held vnodes must not be reclaimed */
-                       if (vp->v_ubcinfo->ui_holdcnt)  { /* XXX */
-                               vinactive(vp);
-                               simple_unlock(&vp->v_interlock);
-                               goto restart;
-                       }
+               VREMINACTIVE("vnreclaim", vp);
+               simple_unlock(&vnode_free_list_slock);
 
-                       if (ubc_issetflags(vp, UI_WASMAPPED)) {
-                               /*
-                                * We should not reclaim as it is likely
-                                * to be in use. Let it die a natural death.
-                                * Release the UBC reference if one exists
-                                * and put it back at the tail.
-                                */
-                               if (ubc_issetflags(vp, UI_HASOBJREF)) {
-                                       obj = ubc_getobject(vp, UBC_NOREACTIVATE);
-                                       if (obj == NULL)
-                                               panic("vnreclaim: null object");
-                                       /* release the reference gained by ubc_info_init() */
-                                       ubc_clearflags(vp, UI_HASOBJREF);
-                                       simple_unlock(&vp->v_interlock);
-                                       vm_object_deallocate(obj);
-                                       /*
-                                        * The vnode interlock was release. 
-                                        * vm_object_deallocate() might have blocked.
-                                        * It is possible that the object was terminated.
-                                        * It is also possible that the vnode was
-                                        * reactivated. Evaluate the state again.
-                                        */ 
-                                       if (UBCINFOEXISTS(vp)) {
-                                               simple_lock(&vp->v_interlock);
-                                               if ((vp->v_usecount == 1) && !VONLIST(vp))
-                                                       vinactive(vp);
-                                               simple_unlock(&vp->v_interlock);
-                                       }
-                               } else {
-                                       vinactive(vp);
-                                       simple_unlock(&vp->v_interlock);
-                               }
-                       } else {
-                               VORECLAIM_ENABLE(vp);
-
-                               /*
-                                * scrub the dirty pages and invalidate the buffers
-                                */
-                               p = current_proc();
-                               err = vn_lock(vp, LK_EXCLUSIVE|LK_INTERLOCK, p); 
-                               if (err) {
-                                       /* cannot reclaim */
+               if (ubc_issetflags(vp, UI_WASMAPPED)) {
+                       /*
+                        * We should not reclaim as it is likely
+                        * to be in use. Let it die a natural death.
+                        * Release the UBC reference if one exists
+                        * and put it back at the tail.
+                        */
+                       simple_unlock(&vp->v_interlock);
+                       if (ubc_release_named(vp)) {
+                               if (UBCINFOEXISTS(vp)) {
                                        simple_lock(&vp->v_interlock);
-                                       vinactive(vp);
-                                       VORECLAIM_DISABLE(vp);
+                                       if (vp->v_usecount == 1 && !VONLIST(vp))
+                                               vinactive(vp);
                                        simple_unlock(&vp->v_interlock);
-                                       goto restart;
                                }
-                               simple_lock(&vp->v_interlock);
-                               if(vp->v_usecount != 1)
-                                       panic("VOCR: usecount race");
+                       } else {
+                           simple_lock(&vp->v_interlock);
+                               vinactive(vp);
                                simple_unlock(&vp->v_interlock);
+                       }
+               } else {
+                       int didhold;
 
-                               /*
-                                * If the UBC reference on the memory object
-                                * was already lost, regain it. This will
-                                * keep the memory object alive for rest of the
-                                * reclaim and finally this reference would
-                                * be lost by memory_object_destroy()
-                                */
-                               obj = ubc_getobject(vp, (UBC_NOREACTIVATE|UBC_HOLDOBJECT));
-                               if (obj == (void *)NULL)
-                                       panic("vnreclaim: null object");
-
-                               /* clean up the state in VM without invalidating */
-                               (void)ubc_clean(vp, 0);
+                       VORECLAIM_ENABLE(vp);
 
-                               /* flush and invalidate buffers associated with the vnode */
-                               if (vp->v_tag == VT_NFS)
-                                       nfs_vinvalbuf(vp, V_SAVE, NOCRED, p, 0);
-                               else
-                                       vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
-
-                               /*
-                                * It is not necessary to call ubc_uncache()
-                                * here because memory_object_destroy() marks
-                                * the memory object non cachable already
-                                *
-                                * Need to release the vnode lock before calling
-                                * vm_object_deallocate() to avoid deadlock
-                                * when the vnode goes through vop_inactive
-                                *
-                                * Note: for the v_usecount == 1 case, VOP_INACTIVE
-                                * has not yet been called.  Call it now while vp is
-                                * still locked, it will also release the lock.
-                                */
-                               if (vp->v_usecount == 1)
-                                       VOP_INACTIVE(vp, p);
-                               else
-                                       VOP_UNLOCK(vp, 0, p);
-
-                               /* 
-                                * This vnode is ready to be reclaimed.
-                                * Terminate the memory object.
-                                * memory_object_destroy() will result in
-                                * vnode_pager_no_senders(). 
-                                * That will release the pager reference
-                                * and the vnode will move to the free list.
-                                */
-                               if (ISSET(vp->v_flag, VTERMINATE))
-                                       panic("vnreclaim: already teminating");
-                               SET(vp->v_flag, VTERMINATE);
-
-                               memory_object_destroy(obj, 0);
-
-                               /* 
-                                * memory_object_destroy() is asynchronous with respect
-                                * to vnode_pager_no_senders().
-                                * wait for vnode_pager_no_senders() to clear
-                                * VTERMINATE
-                                */
-                               while (ISSET(vp->v_flag, VTERMINATE)) {
-                                       SET(vp->v_flag, VTERMWANT);
-                                       tsleep((caddr_t)&vp->v_ubcinfo, PINOD, "vnreclaim", 0);
-                               }
+                       /*
+                        * scrub the dirty pages and invalidate the buffers
+                        */
+                       p = current_proc();
+                       err = vn_lock(vp, LK_EXCLUSIVE|LK_INTERLOCK, p); 
+                       if (err) {
+                               /* cannot reclaim */
                                simple_lock(&vp->v_interlock);
+                               vinactive(vp);
                                VORECLAIM_DISABLE(vp);
                                i++;
                                simple_unlock(&vp->v_interlock);
+                               goto restart;
                        }
-                       /* inactive list lock was released, must restart */
-                       goto restart;
+
+                       /* keep the vnode alive so we can kill it */
+                       simple_lock(&vp->v_interlock);
+                       if(vp->v_usecount != 1)
+                               panic("VOCR: usecount race");
+                       vp->v_usecount++;
+                       simple_unlock(&vp->v_interlock);
+
+                       /* clean up the state in VM without invalidating */
+                       didhold = ubc_hold(vp);
+                       if (didhold)
+                               (void)ubc_clean(vp, 0);
+
+                       /* flush and invalidate buffers associated with the vnode */
+                       if (vp->v_tag == VT_NFS)
+                               nfs_vinvalbuf(vp, V_SAVE, NOCRED, p, 0);
+                       else
+                               vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
+
+                       /*
+                        * Note: for the v_usecount == 2 case, VOP_INACTIVE
+                        * has not yet been called.  Call it now while vp is
+                        * still locked, it will also release the lock.
+                        */
+                       if (vp->v_usecount == 2)
+                               VOP_INACTIVE(vp, p);
+                       else
+                               VOP_UNLOCK(vp, 0, p);
+
+                       if (didhold)
+                               ubc_rele(vp);
+
+                       /*
+                        * destroy the ubc named reference.
+                        * If we can't because it is held for I/Os
+                        * in progress, just put it back on the inactive
+                        * list and move on.  Otherwise, the paging reference
+                        * is toast (and so is this vnode?).
+                        */
+                       if (ubc_destroy_named(vp)) {
+                           i++;
+                       }
+                       simple_lock(&vp->v_interlock);
+                       VORECLAIM_DISABLE(vp);
+                       simple_unlock(&vp->v_interlock);
+                       vrele(vp);  /* release extra use we added here */
                }
+               /* inactive list lock was released, must restart */
+               goto restart;
        }
        simple_unlock(&vnode_free_list_slock);
 
@@ -2566,7 +2487,7 @@ out:
  * AGE the vnode so that it gets recycled quickly.
  * Check lock status to decide whether to call vput() or vrele().
  */
-void
+__private_extern__ void
 vnode_pager_vrele(struct vnode *vp)
 {
 
@@ -2613,22 +2534,22 @@ vnode_pager_vrele(struct vnode *vp)
        if (!ISSET(vp->v_flag, VTERMINATE))
                SET(vp->v_flag, VTERMINATE);
        if (UBCINFOEXISTS(vp)) {
+               struct ubc_info *uip = vp->v_ubcinfo;
+
                if (ubc_issetflags(vp, UI_WASMAPPED))
                        SET(vp->v_flag, VWASMAPPED);
 
-               if ((vp->v_ubcinfo->ui_holdcnt) /* XXX */
-                       && !(vp->v_flag & VXLOCK))
-                       panic("vnode_pager_vrele: freeing held ubc_info");
-
-               simple_unlock(&vp->v_interlock);
-               ubc_info_free(vp);
                vp->v_ubcinfo = UBC_NOINFO;  /* catch bad accesses */
+               simple_unlock(&vp->v_interlock);
+               ubc_info_deallocate(uip);
        } else {
                if ((vp->v_type == VBAD) && ((vp)->v_ubcinfo != UBC_INFO_NULL) 
                        && ((vp)->v_ubcinfo != UBC_NOINFO)) {
-                       simple_unlock(&vp->v_interlock);
-                       ubc_info_free(vp);
+                       struct ubc_info *uip = vp->v_ubcinfo;
+
                        vp->v_ubcinfo = UBC_NOINFO;  /* catch bad accesses */
+                       simple_unlock(&vp->v_interlock);
+                       ubc_info_deallocate(uip);
                } else {
                        simple_unlock(&vp->v_interlock);
                }
@@ -2659,7 +2580,6 @@ int walk_vnodes_debug=0;
 void
 walk_allvnodes()
 {
-       struct proc *p = current_proc();        /* XXX */
        struct mount *mp, *nmp;
        struct vnode *vp;
        int cnt = 0;
@@ -2697,3 +2617,100 @@ walk_allvnodes()
        printf("%d - inactive\n", cnt);
 }
 #endif /* DIAGNOSTIC */
+
+void
+vfs_io_attributes(vp, flags, iosize, vectors)
+       struct vnode    *vp;
+       int     flags;  /* B_READ or B_WRITE */
+       int     *iosize;
+       int     *vectors;
+{
+       struct mount *mp;
+
+       /* start with "reasonable" defaults */
+       *iosize = MAXPHYS;
+       *vectors = 32;
+
+       mp = vp->v_mount;
+       if (mp != NULL) {
+               switch (flags) {
+               case B_READ:
+                       *iosize = mp->mnt_maxreadcnt;
+                       *vectors = mp->mnt_segreadcnt;
+                       break;
+               case B_WRITE:
+                       *iosize = mp->mnt_maxwritecnt;
+                       *vectors = mp->mnt_segwritecnt;
+                       break;
+               default:
+                       break;
+               }
+       }
+
+       return;
+}
+
+#include <dev/disk.h>
+
+int
+vfs_init_io_attributes(devvp, mp)
+       struct vnode *devvp;
+       struct mount *mp;
+{
+       int error;
+       off_t readblockcnt;
+       off_t writeblockcnt;
+       off_t readsegcnt;
+       off_t writesegcnt;
+       u_long blksize;
+
+       u_int64_t temp;
+
+       struct proc *p = current_proc();
+       struct  ucred *cred = p->p_ucred;
+
+       if ((error = VOP_IOCTL(devvp, DKIOCGETMAXBLOCKCOUNTREAD,
+                               (caddr_t)&readblockcnt, 0, cred, p)))
+               return (error);
+
+       if ((error = VOP_IOCTL(devvp, DKIOCGETMAXBLOCKCOUNTWRITE,
+                               (caddr_t)&writeblockcnt, 0, cred, p)))
+               return (error);
+
+       if ((error = VOP_IOCTL(devvp, DKIOCGETMAXSEGMENTCOUNTREAD,
+                               (caddr_t)&readsegcnt, 0, cred, p)))
+               return (error);
+
+       if ((error = VOP_IOCTL(devvp, DKIOCGETMAXSEGMENTCOUNTWRITE,
+                               (caddr_t)&writesegcnt, 0, cred, p)))
+               return (error);
+
+       if ((error = VOP_IOCTL(devvp, DKIOCGETBLOCKSIZE,
+                               (caddr_t)&blksize, 0, cred, p)))
+               return (error);
+
+       temp = readblockcnt * blksize;
+       temp = (temp > UINT32_MAX) ? (UINT32_MAX / blksize) * blksize : temp;
+       mp->mnt_maxreadcnt = (u_int32_t)temp;
+
+       temp = writeblockcnt * blksize;
+       temp = (temp > UINT32_MAX) ? (UINT32_MAX / blksize) * blksize : temp;
+       mp->mnt_maxwritecnt = (u_int32_t)temp;
+
+       temp = (readsegcnt > UINT16_MAX) ? UINT16_MAX : readsegcnt;
+       mp->mnt_segreadcnt = (u_int16_t)temp;
+
+       temp = (writesegcnt > UINT16_MAX) ? UINT16_MAX : writesegcnt;
+       mp->mnt_segwritecnt = (u_int16_t)temp;
+
+#if 0
+       printf("--- IO attributes for mount point 0x%08x ---\n", mp);
+       printf("\tmnt_maxreadcnt = 0x%x", mp->mnt_maxreadcnt);
+       printf("\tmnt_maxwritecnt = 0x%x\n", mp->mnt_maxwritecnt);
+       printf("\tmnt_segreadcnt = 0x%x", mp->mnt_segreadcnt);
+       printf("\tmnt_segwritecnt = 0x%x\n", mp->mnt_segwritecnt);
+#endif /* 0 */
+
+       return (error);
+}
+