]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/vfs/vfs_cache.c
xnu-792.12.6.tar.gz
[apple/xnu.git] / bsd / vfs / vfs_cache.c
index e4b72f55480481d909138b1c6941f6947c0c8e36..174b0ecb784f490a79ea9100f814c78d5b5b1bc2 100644 (file)
@@ -1,26 +1,31 @@
 /*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2003 Apple Computer, Inc. All rights reserved.
  *
- * @APPLE_LICENSE_HEADER_START@
+ * @APPLE_LICENSE_OSREFERENCE_HEADER_START@
  * 
- * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
- * 
- * 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
+ * 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.  The rights granted to you under the 
+ * License may not be used to create, or enable the creation or 
+ * redistribution of, unlawful or unlicensed copies of an Apple operating 
+ * system, or to circumvent, violate, or enable the circumvention or 
+ * violation of, any terms of an Apple operating system software license 
+ * agreement.
+ *
+ * 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, QUIET ENJOYMENT OR NON-INFRINGEMENT.
- * Please see the License for the specific language governing rights and
+ *
+ * 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, QUIET ENJOYMENT OR NON-INFRINGEMENT. 
+ * Please see the License for the specific language governing rights and 
  * limitations under the License.
- * 
- * @APPLE_LICENSE_HEADER_END@
+ *
+ * @APPLE_LICENSE_OSREFERENCE_HEADER_END@
  */
 /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
 /*
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/time.h>
-#include <sys/mount.h>
-#include <sys/vnode.h>
+#include <sys/mount_internal.h>
+#include <sys/vnode_internal.h>
 #include <sys/namei.h>
 #include <sys/errno.h>
 #include <sys/malloc.h>
+#include <sys/kauth.h>
+#include <sys/user.h>
 
 /*
  * Name caching works as follows:
  * If it is a "negative" entry, (i.e. for a name that is known NOT to
  * exist) the vnode pointer will be NULL.
  *
- * For simplicity (and economy of storage), names longer than
- * a maximum length of NCHNAMLEN are not cached; they occur
- * infrequently in any case, and are almost never of interest.
- *
  * Upon reaching the last segment of a path, if the reference
  * is for DELETE, or NOCACHE is set (rewrite), and the
  * name is located in the cache, it will be dropped.
 /*
  * Structures associated with name cacheing.
  */
-#define NCHHASH(dvp, cnp) \
-       (&nchashtbl[((dvp)->v_id + (cnp)->cn_hash) & nchash])
+
 LIST_HEAD(nchashhead, namecache) *nchashtbl;   /* Hash Table */
+u_long nchashmask;
 u_long nchash;                         /* size of hash table - 1 */
 long   numcache;                       /* number of cache entries allocated */
-TAILQ_HEAD(, namecache) nclruhead;     /* LRU chain */
+int    desiredNodes;
+int    desiredNegNodes;
+TAILQ_HEAD(, namecache) nchead;                /* chain of all name cache entries */
+TAILQ_HEAD(, namecache) neghead;       /* chain of only negative cache entries */
 struct nchstats nchstats;              /* cache effectiveness statistics */
-u_long nextvnodeid = 0;
-int doingcache = 1;                    /* 1 => enable the cache */
+
+/* vars for name cache list lock */
+lck_grp_t * namecache_lck_grp;
+lck_grp_attr_t * namecache_lck_grp_attr;
+lck_attr_t * namecache_lck_attr;
+lck_mtx_t * namecache_mtx_lock;
+
+static vnode_t cache_lookup_locked(vnode_t dvp, struct componentname *cnp);
+static int  remove_name_locked(const char *);
+static char *add_name_locked(const char *, size_t, u_int, u_int);
+static void init_string_table(void);
+static void cache_delete(struct namecache *, int);
+static void dump_string_table(void);
+
+static void init_crc32(void);
+static unsigned int crc32tab[256];
+
+
+#define NCHHASH(dvp, hash_val) \
+       (&nchashtbl[(dvp->v_id ^ (hash_val)) & nchashmask])
+
+
+
+//
+// This function builds the path to a filename in "buff".  The
+// length of the buffer *INCLUDING* the trailing zero byte is
+// returned in outlen.  NOTE: the length includes the trailing
+// zero byte and thus the length is one greater than what strlen
+// would return.  This is important and lots of code elsewhere
+// in the kernel assumes this behavior.
+//
+int
+build_path(vnode_t first_vp, char *buff, int buflen, int *outlen)
+{
+        vnode_t vp = first_vp;
+       char *end, *str;
+       int   len, ret=0, counter=0;
+
+       end = &buff[buflen-1];
+       *end = '\0';
+
+       /*
+        * if this is the root dir of a file system...
+        */
+       if (vp && (vp->v_flag & VROOT) && vp->v_mount) {
+               /*
+                * then if it's the root fs, just put in a '/' and get out of here
+                */
+               if (vp->v_mount->mnt_flag & MNT_ROOTFS) {
+                       *--end = '/';
+                       goto out;
+               } else {
+                       /*
+                        * else just use the covered vnode to get the mount path
+                        */
+                       vp = vp->v_mount->mnt_vnodecovered;
+               }
+       }
+       name_cache_lock();
+
+       while (vp && vp->v_parent != vp) {
+               /*
+                * the maximum depth of a file system hierarchy is MAXPATHLEN/2
+                * (with single-char names separated by slashes).  we panic if
+                * we've ever looped more than that.
+                */
+               if (counter++ > MAXPATHLEN/2) {
+                       panic("build_path: vnode parent chain is too long! vp 0x%x\n", vp);
+               }
+               str = vp->v_name;
+
+               if (str == NULL) {
+                       if (vp->v_parent != NULL) {
+                               ret = EINVAL;
+                       }
+                       break;
+               }
+               len = strlen(str);
+
+               /*
+                * check that there's enough space (make sure to include space for the '/')
+                */
+               if ((end - buff) < (len + 1)) {
+                       ret = ENOSPC;
+                       break;
+               }
+               /*
+                * copy it backwards
+                */
+               str += len;
+
+               for (; len > 0; len--) {
+                      *--end = *--str;
+               }
+               /*
+                * put in the path separator
+                */
+               *--end = '/';
+
+               /*
+                * walk up the chain (as long as we're not the root)  
+                */
+               if (vp == first_vp && (vp->v_flag & VROOT)) {
+                       if (vp->v_mount && vp->v_mount->mnt_vnodecovered) {
+                               vp = vp->v_mount->mnt_vnodecovered->v_parent;
+                       } else {
+                               vp = NULLVP;
+                       }
+               } else {
+                       vp = vp->v_parent;
+               }
+               /*
+                * check if we're crossing a mount point and
+                * switch the vp if we are.
+                */
+               if (vp && (vp->v_flag & VROOT) && vp->v_mount) {
+                       vp = vp->v_mount->mnt_vnodecovered;
+               }
+       }
+       name_cache_unlock();
+out:
+       /*
+        * slide it down to the beginning of the buffer
+        */
+       memmove(buff, end, &buff[buflen] - end);
+    
+       *outlen = &buff[buflen] - end;  // length includes the trailing zero byte
+       return ret;
+}
+
 
 /*
- * Delete an entry from its hash list and move it to the front
- * of the LRU list for immediate reuse.
+ * return NULLVP if vp's parent doesn't
+ * exist, or we can't get a valid iocount
+ * else return the parent of vp
  */
-#if DIAGNOSTIC
-#define PURGE(ncp)  {                                          \
-       if (ncp->nc_hash.le_prev == 0)                          \
-               panic("namecache purge le_prev");               \
-       if (ncp->nc_hash.le_next == ncp)                        \
-               panic("namecache purge le_next");               \
-       LIST_REMOVE(ncp, nc_hash);                              \
-       ncp->nc_hash.le_prev = 0;                               \
-       TAILQ_REMOVE(&nclruhead, ncp, nc_lru);                  \
-       TAILQ_INSERT_HEAD(&nclruhead, ncp, nc_lru);             \
+vnode_t
+vnode_getparent(vnode_t vp)
+{
+        vnode_t pvp = NULLVP;
+       int     pvid;
+
+        name_cache_lock();
+       /*
+        * v_parent is stable behind the name_cache lock
+        * however, the only thing we can really guarantee
+        * is that we've grabbed a valid iocount on the
+        * parent of 'vp' at the time we took the name_cache lock...
+        * once we drop the lock, vp could get re-parented
+        */
+       if ( (pvp = vp->v_parent) != NULLVP ) {
+               pvid = pvp->v_id;
+
+               name_cache_unlock();
+
+               if (vnode_getwithvid(pvp, pvid) != 0)
+                       pvp = NULL;
+       } else
+               name_cache_unlock();
+
+       return (pvp);
+}
+
+char *
+vnode_getname(vnode_t vp)
+{
+        char *name = NULL;
+
+        name_cache_lock();
+       
+       if (vp->v_name)
+               name = add_name_locked(vp->v_name, strlen(vp->v_name), 0, 0);
+       name_cache_unlock();
+
+       return (name);
 }
-#else
-#define PURGE(ncp)  {                                          \
-       LIST_REMOVE(ncp, nc_hash);                              \
-       ncp->nc_hash.le_prev = 0;                               \
-       TAILQ_REMOVE(&nclruhead, ncp, nc_lru);                  \
-       TAILQ_INSERT_HEAD(&nclruhead, ncp, nc_lru);             \
+
+void
+vnode_putname(char *name)
+{
+       name_cache_lock();
+
+       remove_name_locked(name);
+
+       name_cache_unlock();
+}
+
+
+/*
+ * if VNODE_UPDATE_PARENT, and we can take
+ * a reference on dvp, then update vp with
+ * it's new parent... if vp already has a parent,
+ * then drop the reference vp held on it
+ *
+ * if VNODE_UPDATE_NAME,
+ * then drop string ref on v_name if it exists, and if name is non-NULL
+ * then pick up a string reference on name and record it in v_name...
+ * optionally pass in the length and hashval of name if known
+ *
+ * if VNODE_UPDATE_CACHE, flush the name cache entries associated with vp
+ */
+void
+vnode_update_identity(vnode_t vp, vnode_t dvp, char *name, int name_len, int name_hashval, int flags)
+{
+       struct  namecache *ncp;
+        vnode_t        old_parentvp = NULLVP;
+
+
+       if (flags & VNODE_UPDATE_PARENT) {
+               if (dvp && vnode_ref(dvp) != 0)
+                       dvp = NULLVP;
+       } else
+               dvp = NULLVP;
+       name_cache_lock();
+
+       if ( (flags & VNODE_UPDATE_NAME) && (name != vp->v_name) ) {
+               if (vp->v_name != NULL) {
+                       remove_name_locked(vp->v_name);
+                       vp->v_name = NULL;
+               }
+               if (name && *name) {
+                       if (name_len == 0)
+                               name_len = strlen(name);
+                       vp->v_name = add_name_locked(name, name_len, name_hashval, 0);
+               }
+       }
+       if (flags & VNODE_UPDATE_PARENT) {
+               if (dvp != vp && dvp != vp->v_parent) {
+                       old_parentvp = vp->v_parent;
+                       vp->v_parent = dvp;
+                       dvp = NULLVP;
+
+                       if (old_parentvp)
+                               flags |= VNODE_UPDATE_CACHE;
+               }
+       }
+       if (flags & VNODE_UPDATE_CACHE) {
+               while ( (ncp = LIST_FIRST(&vp->v_nclinks)) )
+                       cache_delete(ncp, 1);
+       }
+       name_cache_unlock();
+       
+       if (dvp != NULLVP)
+               vnode_rele(dvp);
+       
+       if (old_parentvp) {
+               struct  uthread *ut;
+
+               ut = get_bsdthread_info(current_thread());
+
+               /*
+                * indicated to vnode_rele that it shouldn't do a
+                * vnode_reclaim at this time... instead it will
+                * chain the vnode to the uu_vreclaims list...
+                * we'll be responsible for calling vnode_reclaim
+                * on each of the vnodes in this list...
+                */
+               ut->uu_defer_reclaims = 1;
+               ut->uu_vreclaims = NULLVP;
+
+               while ( (vp = old_parentvp) != NULLVP ) {
+         
+                       vnode_lock(vp);
+
+                       vnode_rele_internal(vp, 0, 0, 1);
+
+                       /*
+                        * check to see if the vnode is now in the state
+                        * that would have triggered a vnode_reclaim in vnode_rele
+                        * if it is, we save it's parent pointer and then NULL
+                        * out the v_parent field... we'll drop the reference
+                        * that was held on the next iteration of this loop...
+                        * this short circuits a potential deep recursion if we
+                        * have a long chain of parents in this state... 
+                        * we'll sit in this loop until we run into
+                        * a parent in this chain that is not in this state
+                        *
+                        * make our check and the node_rele atomic
+                        * with respect to the current vnode we're working on
+                        * by holding the vnode lock
+                        * if vnode_rele deferred the vnode_reclaim and has put
+                        * this vnode on the list to be reaped by us, than
+                        * it has left this vnode with an iocount == 1
+                        */
+                       if ( (vp->v_iocount == 1) && (vp->v_usecount == 0) &&
+                            ((vp->v_lflag & (VL_MARKTERM | VL_TERMINATE | VL_DEAD)) == VL_MARKTERM)) {
+                               /*
+                                * vnode_rele wanted to do a vnode_reclaim on this vnode
+                                * it should be sitting on the head of the uu_vreclaims chain
+                                * pull the parent pointer now so that when we do the
+                                * vnode_reclaim for each of the vnodes in the uu_vreclaims
+                                * list, we won't recurse back through here
+                                */
+                               name_cache_lock();
+                               old_parentvp = vp->v_parent;
+                               vp->v_parent = NULLVP;
+                               name_cache_unlock();
+                       } else {
+                               /*
+                                * we're done... we ran into a vnode that isn't
+                                * being terminated
+                                */
+                               old_parentvp = NULLVP;
+                       }
+                       vnode_unlock(vp);
+               }
+               ut->uu_defer_reclaims = 0;
+
+               while ( (vp = ut->uu_vreclaims) != NULLVP) {
+                       ut->uu_vreclaims = vp->v_defer_reclaimlist;
+                       
+                       /*
+                        * vnode_put will drive the vnode_reclaim if
+                        * we are still the only reference on this vnode
+                        */
+                       vnode_put(vp);
+               }
+       }
 }
-#endif /* DIAGNOSTIC */
+
 
 /*
- * Move an entry that has been used to the tail of the LRU list
- * so that it will be preserved for future use.
+ * Mark a vnode as having multiple hard links.  HFS makes use of this
+ * because it keeps track of each link separately, and wants to know
+ * which link was actually used.
+ *
+ * This will cause the name cache to force a VNOP_LOOKUP on the vnode
+ * so that HFS can post-process the lookup.  Also, volfs will call
+ * VNOP_GETATTR2 to determine the parent, instead of using v_parent.
  */
-#define TOUCH(ncp)  {                                          \
-       if (ncp->nc_lru.tqe_next != 0) {                        \
-               TAILQ_REMOVE(&nclruhead, ncp, nc_lru);          \
-               TAILQ_INSERT_TAIL(&nclruhead, ncp, nc_lru);     \
-       }                                                       \
+void vnode_set_hard_link(vnode_t vp)
+{
+       vnode_lock(vp);
+
+       /*
+        * In theory, we're changing the vnode's identity as far as the
+        * name cache is concerned, so we ought to grab the name cache lock
+        * here.  However, there is already a race, and grabbing the name
+        * cache lock only makes the race window slightly smaller.
+        *
+        * The race happens because the vnode already exists in the name
+        * cache, and could be found by one thread before another thread
+        * can set the hard link flag.
+        */
+
+       vp->v_flag |= VISHARDLINK;
+
+       vnode_unlock(vp);
+}
+
+
+void vnode_uncache_credentials(vnode_t vp)
+{
+        kauth_cred_t ucred = NULL;
+
+        if (vp->v_cred) {
+               vnode_lock(vp);
+
+               ucred = vp->v_cred;
+               vp->v_cred = NULL;
+
+               vnode_unlock(vp);
+
+               if (ucred)
+                       kauth_cred_rele(ucred);
+       }
+}
+
+
+void vnode_cache_credentials(vnode_t vp, vfs_context_t context)
+{
+       kauth_cred_t ucred;
+       kauth_cred_t tcred = NOCRED;
+       struct timeval tv;
+
+       ucred = vfs_context_ucred(context);
+
+       if (vp->v_cred != ucred || (vp->v_mount->mnt_kern_flag & MNTK_AUTH_OPAQUE)) {
+               vnode_lock(vp);
+
+               microuptime(&tv);
+               vp->v_cred_timestamp = tv.tv_sec;
+
+               if (vp->v_cred != ucred) {
+                       kauth_cred_ref(ucred);
+       
+                       tcred = vp->v_cred;
+                       vp->v_cred = ucred;
+               }
+               vnode_unlock(vp);
+       
+               if (tcred)
+                       kauth_cred_rele(tcred);
+       }
+}
+
+/*     reverse_lookup - lookup by walking back up the parent chain while leveraging
+ *     use of the name cache lock in order to protect our starting vnode.
+ *     NOTE - assumes you already have search access to starting point.
+ *  returns 0 when we have reached the root, current working dir, or chroot root 
+ *
+ */
+int
+reverse_lookup(vnode_t start_vp, vnode_t *lookup_vpp, struct filedesc *fdp, vfs_context_t context, int *dp_authorized)
+{
+       int                             vid, done = 0;
+       int                             auth_opaque = 0;
+       vnode_t                 dp = start_vp;
+       vnode_t                 vp = NULLVP;
+       kauth_cred_t    ucred;
+       struct timeval  tv;
+
+       ucred = vfs_context_ucred(context);
+       *lookup_vpp = start_vp;
+
+       name_cache_lock();
+
+       if ( dp->v_mount && (dp->v_mount->mnt_kern_flag & MNTK_AUTH_OPAQUE) ) {
+               auth_opaque = 1;
+               microuptime(&tv);
+       }
+       for (;;) {
+               *dp_authorized = 0;
+
+               if (auth_opaque && ((tv.tv_sec - dp->v_cred_timestamp) > VCRED_EXPIRED))
+                       break;
+               if (dp->v_cred != ucred)
+                       break;
+               /*
+                * indicate that we're allowed to traverse this directory...
+                * even if we bail for some reason, this information is valid and is used
+                * to avoid doing a vnode_authorize
+                */
+               *dp_authorized = 1;
+
+               if ((dp->v_flag & VROOT) != 0   ||              /* Hit "/" */
+                   (dp == fdp->fd_cdir)        ||              /* Hit process's working directory */
+                   (dp == fdp->fd_rdir)) {                     /* Hit process chroot()-ed root */
+                       done = 1;
+                       break;
+               }
+
+               if ( (vp = dp->v_parent) == NULLVP)
+                       break;
+
+               dp = vp;
+               *lookup_vpp = dp;
+       } /* for (;;) */
+
+       vid = dp->v_id;
+       
+       name_cache_unlock();
+       
+       if (done == 0 && dp != start_vp) {
+               if (vnode_getwithvid(dp, vid) != 0) {
+                       *lookup_vpp = start_vp;
+               }
+       }
+
+       return((done == 1) ? 0 : -1);
+}
+
+int 
+cache_lookup_path(struct nameidata *ndp, struct componentname *cnp, vnode_t dp, vfs_context_t context, int *trailing_slash, int *dp_authorized)
+{
+       char            *cp;            /* pointer into pathname argument */
+       int             vid, vvid;
+       int             auth_opaque = 0;
+       vnode_t         vp = NULLVP;
+       vnode_t         tdp = NULLVP;
+       kauth_cred_t    ucred;
+       struct timeval tv;
+       unsigned int    hash;
+
+       ucred = vfs_context_ucred(context);
+       *trailing_slash = 0;
+
+       name_cache_lock();
+
+
+       if ( dp->v_mount && (dp->v_mount->mnt_kern_flag & MNTK_AUTH_OPAQUE) ) {
+               auth_opaque = 1;
+               microuptime(&tv);
+       }
+       for (;;) {
+               /*
+                * Search a directory.
+                *
+                * The cn_hash value is for use by cache_lookup
+                * The last component of the filename is left accessible via
+                * cnp->cn_nameptr for callers that need the name.
+                */
+               hash = 0;
+               cp = cnp->cn_nameptr;
+
+               while (*cp && (*cp != '/')) {
+                       hash ^= crc32tab[((hash >> 24) ^ (unsigned char)*cp++)];
+               }
+               /*
+                * the crc generator can legitimately generate
+                * a 0... however, 0 for us means that we
+                * haven't computed a hash, so use 1 instead
+                */
+               if (hash == 0)
+                       hash = 1;
+               cnp->cn_hash = hash;
+               cnp->cn_namelen = cp - cnp->cn_nameptr;
+
+               ndp->ni_pathlen -= cnp->cn_namelen;
+               ndp->ni_next = cp;
+
+               /*
+                * Replace multiple slashes by a single slash and trailing slashes
+                * by a null.  This must be done before VNOP_LOOKUP() because some
+                * fs's don't know about trailing slashes.  Remember if there were
+                * trailing slashes to handle symlinks, existing non-directories
+                * and non-existing files that won't be directories specially later.
+                */
+               while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
+                       cp++;
+                       ndp->ni_pathlen--;
+
+                       if (*cp == '\0') {
+                               *trailing_slash = 1;
+                               *ndp->ni_next = '\0';
+                       }
+               }
+               ndp->ni_next = cp;
+
+               cnp->cn_flags &= ~(MAKEENTRY | ISLASTCN | ISDOTDOT);
+
+               if (*cp == '\0')
+                       cnp->cn_flags |= ISLASTCN;
+
+               if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
+                       cnp->cn_flags |= ISDOTDOT;
+
+               *dp_authorized = 0;
+
+               if (auth_opaque && ((tv.tv_sec - dp->v_cred_timestamp) > VCRED_EXPIRED))
+                       break;
+
+               if (dp->v_cred != ucred)
+                       break;
+               /*
+                * indicate that we're allowed to traverse this directory...
+                * even if we fail the cache lookup or decide to bail for
+                * some other reason, this information is valid and is used
+                * to avoid doing a vnode_authorize before the call to VNOP_LOOKUP
+                */
+               *dp_authorized = 1;
+
+               if ( (cnp->cn_flags & (ISLASTCN | ISDOTDOT)) ) {
+                       if (cnp->cn_nameiop != LOOKUP)
+                               break;
+                       if (cnp->cn_flags & (LOCKPARENT | NOCACHE))
+                               break;
+                       if (cnp->cn_flags & ISDOTDOT) {
+                               /*
+                                * Quit here only if we can't use
+                                * the parent directory pointer or
+                                * don't have one.  Otherwise, we'll
+                                * use it below.
+                                */
+                               if ((dp->v_flag & VROOT) ||
+                                   dp->v_parent == NULLVP)
+                                       break;
+                       }
+               }
+
+               /*
+                * "." and ".." aren't supposed to be cached, so check
+                * for them before checking the cache.
+                */
+               if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.')
+                       vp = dp;
+               else if (cnp->cn_flags & ISDOTDOT)
+                       vp = dp->v_parent;
+               else {
+                       if ( (vp = cache_lookup_locked(dp, cnp)) == NULLVP)
+                               break;
+               }
+
+               if ( (cnp->cn_flags & ISLASTCN) )
+                       break;
+
+               if (vp->v_type != VDIR) {
+                       if (vp->v_type != VLNK)
+                               vp = NULL;
+                       break;
+               }
+               if (vp->v_mountedhere && ((cnp->cn_flags & NOCROSSMOUNT) == 0))
+                       break;
+
+               dp = vp;
+               vp = NULLVP;
+
+               cnp->cn_nameptr = ndp->ni_next + 1;
+               ndp->ni_pathlen--;
+               while (*cnp->cn_nameptr == '/') {
+                       cnp->cn_nameptr++;
+                       ndp->ni_pathlen--;
+               }
+       }
+       if (vp != NULLVP)
+               vvid = vp->v_id;
+       vid = dp->v_id;
+       
+       name_cache_unlock();
+
+
+       if ((vp != NULLVP) && (vp->v_type != VLNK) &&
+           ((cnp->cn_flags & (ISLASTCN | LOCKPARENT | WANTPARENT | SAVESTART)) == ISLASTCN)) {
+               /*
+                * if we've got a child and it's the last component, and 
+                * the lookup doesn't need to return the parent then we
+                * can skip grabbing an iocount on the parent, since all
+                * we're going to do with it is a vnode_put just before
+                * we return from 'lookup'.  If it's a symbolic link,
+                * we need the parent in case the link happens to be
+                * a relative pathname.
+                */
+               tdp = dp;
+               dp = NULLVP;
+       } else {
+need_dp:
+               /*
+                * return the last directory we looked at
+                * with an io reference held
+                */
+               if (dp == ndp->ni_usedvp) {
+                       /*
+                        * if this vnode matches the one passed in via USEDVP
+                        * than this context already holds an io_count... just
+                        * use vnode_get to get an extra ref for lookup to play
+                        * with... can't use the getwithvid variant here because
+                        * it will block behind a vnode_drain which would result
+                        * in a deadlock (since we already own an io_count that the
+                        * vnode_drain is waiting on)... vnode_get grabs the io_count
+                        * immediately w/o waiting... it always succeeds
+                        */
+                       vnode_get(dp);
+               } else if ( (vnode_getwithvid(dp, vid)) ) {
+                       /*
+                        * failure indicates the vnode
+                        * changed identity or is being
+                        * TERMINATED... in either case
+                        * punt this lookup
+                        */
+                       return (ENOENT);
+               }
+       }
+       if (vp != NULLVP) {
+               if ( (vnode_getwithvid(vp, vvid)) ) {
+                       vp = NULLVP;
+
+                       /*
+                        * can't get reference on the vp we'd like
+                        * to return... if we didn't grab a reference
+                        * on the directory (due to fast path bypass),
+                        * then we need to do it now... we can't return
+                        * with both ni_dvp and ni_vp NULL, and no 
+                        * error condition
+                        */
+                       if (dp == NULLVP) {
+                               dp = tdp;
+                               goto need_dp;
+                       }
+               }
+       }
+       ndp->ni_dvp = dp;
+       ndp->ni_vp  = vp;
+
+       return (0);
+}
+
+
+static vnode_t
+cache_lookup_locked(vnode_t dvp, struct componentname *cnp)
+{
+       register struct namecache *ncp;
+       register struct nchashhead *ncpp;
+       register long namelen = cnp->cn_namelen;
+       char *nameptr = cnp->cn_nameptr;
+       unsigned int hashval = (cnp->cn_hash & NCHASHMASK);
+       vnode_t vp;
+       
+       ncpp = NCHHASH(dvp, cnp->cn_hash);
+       LIST_FOREACH(ncp, ncpp, nc_hash) {
+               if ((ncp->nc_dvp == dvp) && (ncp->nc_hashval == hashval)) {
+                       if (memcmp(ncp->nc_name, nameptr, namelen) == 0 && ncp->nc_name[namelen] == 0)
+                               break;
+               }
+       }
+       if (ncp == 0)
+               /*
+                * We failed to find an entry
+                */
+               return (NULL);
+
+       vp = ncp->nc_vp;
+       if (vp && (vp->v_flag & VISHARDLINK)) {
+                       /*
+                        * The file system wants a VNOP_LOOKUP on this vnode
+                        */
+                       vp = NULL;
+       }
+       
+       return (vp);
+}
+
+
+//
+// Have to take a len argument because we may only need to
+// hash part of a componentname.
+//
+static unsigned int
+hash_string(const char *cp, int len)
+{
+    unsigned hash = 0;
+
+    if (len) {
+            while (len--) {
+                   hash ^= crc32tab[((hash >> 24) ^ (unsigned char)*cp++)];
+           }
+    } else {
+            while (*cp != '\0') {
+                   hash ^= crc32tab[((hash >> 24) ^ (unsigned char)*cp++)];
+           }
+    }
+    /*
+     * the crc generator can legitimately generate
+     * a 0... however, 0 for us means that we
+     * haven't computed a hash, so use 1 instead
+     */
+    if (hash == 0)
+            hash = 1;
+    return hash;
 }
 
+
 /*
  * Lookup an entry in the cache 
  *
@@ -160,71 +870,73 @@ cache_lookup(dvp, vpp, cnp)
        struct vnode **vpp;
        struct componentname *cnp;
 {
-       register struct namecache *ncp, *nnp;
+       register struct namecache *ncp;
        register struct nchashhead *ncpp;
+       register long namelen = cnp->cn_namelen;
+       char *nameptr = cnp->cn_nameptr;
+       unsigned int hashval = (cnp->cn_hash & NCHASHMASK);
+       uint32_t vid;
+       vnode_t  vp;
 
-       if (!doingcache) {
-               cnp->cn_flags &= ~MAKEENTRY;
-               return (0);
-       }
-       if (cnp->cn_namelen > NCHNAMLEN) {
-               nchstats.ncs_long++;
-               cnp->cn_flags &= ~MAKEENTRY;
-               return (0);
-       }
+       name_cache_lock();
 
-       ncpp = NCHHASH(dvp, cnp);
-       for (ncp = ncpp->lh_first; ncp != 0; ncp = nnp) {
-               nnp = ncp->nc_hash.le_next;
-               /* If one of the vp's went stale, don't bother anymore. */
-               if ((ncp->nc_dvpid != ncp->nc_dvp->v_id) ||
-                   (ncp->nc_vp && ncp->nc_vpid != ncp->nc_vp->v_id)) {
-                       nchstats.ncs_falsehits++;
-                       PURGE(ncp);
-                       continue;
-               }
-               /* Now that we know the vp's to be valid, is it ours ? */
-               if (ncp->nc_dvp == dvp &&
-                   ncp->nc_nlen == cnp->cn_namelen &&
-                   !bcmp(ncp->nc_name, cnp->cn_nameptr, (u_int)ncp->nc_nlen))
-                       break;
+       ncpp = NCHHASH(dvp, cnp->cn_hash);
+       LIST_FOREACH(ncp, ncpp, nc_hash) {
+               if ((ncp->nc_dvp == dvp) && (ncp->nc_hashval == hashval)) {
+                       if (memcmp(ncp->nc_name, nameptr, namelen) == 0 && ncp->nc_name[namelen] == 0)
+                               break;
+               }
        }
-
        /* We failed to find an entry */
        if (ncp == 0) {
                nchstats.ncs_miss++;
+               name_cache_unlock();
                return (0);
        }
 
        /* We don't want to have an entry, so dump it */
        if ((cnp->cn_flags & MAKEENTRY) == 0) {
                nchstats.ncs_badhits++;
-               PURGE(ncp);
+               cache_delete(ncp, 1);
+               name_cache_unlock();
                return (0);
        } 
+       vp = ncp->nc_vp;
 
        /* We found a "positive" match, return the vnode */
-        if (ncp->nc_vp) {
+        if (vp) {
                nchstats.ncs_goodhits++;
-               TOUCH(ncp);
-               *vpp = ncp->nc_vp;
+
+               vid = vp->v_id;
+               name_cache_unlock();
+
+               if (vnode_getwithvid(vp, vid)) {
+                       name_cache_lock();
+                       nchstats.ncs_badvid++;
+                       name_cache_unlock();
+                       return (0);
+               }
+               *vpp = vp;
                return (-1);
        }
 
        /* We found a negative match, and want to create it, so purge */
-       if (cnp->cn_nameiop == CREATE) {
+       if (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) {
                nchstats.ncs_badhits++;
-               PURGE(ncp);
+               cache_delete(ncp, 1);
+               name_cache_unlock();
                return (0);
        }
 
        /*
         * We found a "negative" match, ENOENT notifies client of this match.
-        * The nc_vpid field records whether this is a whiteout.
+        * The nc_whiteout field records whether this is a whiteout.
         */
        nchstats.ncs_neghits++;
-       TOUCH(ncp);
-       cnp->cn_flags |= ncp->nc_vpid;
+
+       if (ncp->nc_whiteout)
+               cnp->cn_flags |= ISWHITEOUT;
+       name_cache_unlock();
        return (ENOENT);
 }
 
@@ -237,66 +949,70 @@ cache_enter(dvp, vp, cnp)
        struct vnode *vp;
        struct componentname *cnp;
 {
-       register struct namecache *ncp;
+        register struct namecache *ncp, *negp;
        register struct nchashhead *ncpp;
 
-       if (!doingcache)
-               return;
+        if (cnp->cn_hash == 0)
+               cnp->cn_hash = hash_string(cnp->cn_nameptr, cnp->cn_namelen);
 
-       /*
-        * If an entry that is too long, is entered, bad things happen.
-        * cache_lookup acts as the sentinel to make sure longer names
-        * are not stored. This here will prevent outsiders from doing
-        * something that is unexpected.
-        */
-       if (cnp->cn_namelen > NCHNAMLEN)
-               panic("cache_enter: name too long");
+       name_cache_lock();
 
+       /* if the entry is for -ve caching vp is null */
+       if ((vp != NULLVP) && (LIST_FIRST(&vp->v_nclinks))) {
+               /*
+                * someone beat us to the punch..
+                * this vnode is already in the cache
+                */
+               name_cache_unlock();
+                       return;
+       }
        /*
         * We allocate a new entry if we are less than the maximum
-        * allowed and the one at the front of the LRU list is in use.
-        * Otherwise we use the one at the front of the LRU list.
+        * allowed and the one at the front of the list is in use.
+        * Otherwise we use the one at the front of the list.
         */
-       if (numcache < desiredvnodes &&
-           ((ncp = nclruhead.tqh_first) == NULL ||
-           ncp->nc_hash.le_prev != 0)) {
-               /* Add one more entry */
-               ncp = (struct namecache *)
-                       _MALLOC_ZONE((u_long)sizeof *ncp, M_CACHE, M_WAITOK);
+       if (numcache < desiredNodes &&
+           ((ncp = nchead.tqh_first) == NULL ||
+             ncp->nc_hash.le_prev != 0)) {
+               /*
+                * Allocate one more entry
+                */
+               ncp = (struct namecache *)_MALLOC_ZONE((u_long)sizeof *ncp, M_CACHE, M_WAITOK);
                numcache++;
-       } else if (ncp = nclruhead.tqh_first) {
-               /* reuse an old entry */
-               TAILQ_REMOVE(&nclruhead, ncp, nc_lru);
+       } else {
+               /*
+                * reuse an old entry
+                */
+               ncp = TAILQ_FIRST(&nchead);
+               TAILQ_REMOVE(&nchead, ncp, nc_entry);
+
                if (ncp->nc_hash.le_prev != 0) {
-#if DIAGNOSTIC
-                       if (ncp->nc_hash.le_next == ncp)
-                               panic("cache_enter: le_next");
-#endif
-                       LIST_REMOVE(ncp, nc_hash);
-                       ncp->nc_hash.le_prev = 0;
+                      /*
+                       * still in use... we need to
+                       * delete it before re-using it
+                       */
+                       nchstats.ncs_stolen++;
+                       cache_delete(ncp, 0);
                }
-       } else {
-               /* give up */
-               return;
        }
+       nchstats.ncs_enters++;
 
        /*
         * Fill in cache info, if vp is NULL this is a "negative" cache entry.
-        * For negative entries, we have to record whether it is a whiteout.
-        * the whiteout flag is stored in the nc_vpid field which is
-        * otherwise unused.
         */
        ncp->nc_vp = vp;
-       if (vp)
-               ncp->nc_vpid = vp->v_id;
-       else
-               ncp->nc_vpid = cnp->cn_flags & ISWHITEOUT;
        ncp->nc_dvp = dvp;
-       ncp->nc_dvpid = dvp->v_id;
-       ncp->nc_nlen = cnp->cn_namelen;
-       bcopy(cnp->cn_nameptr, ncp->nc_name, (unsigned)ncp->nc_nlen);
-       TAILQ_INSERT_TAIL(&nclruhead, ncp, nc_lru);
-       ncpp = NCHHASH(dvp, cnp);
+       ncp->nc_hashval = cnp->cn_hash;
+       ncp->nc_whiteout = FALSE;
+       ncp->nc_name = add_name_locked(cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_hash, 0);
+
+       /*
+        * make us the newest entry in the cache
+        * i.e. we'll be the last to be stolen
+        */
+       TAILQ_INSERT_TAIL(&nchead, ncp, nc_entry);
+
+       ncpp = NCHHASH(dvp, cnp->cn_hash);
 #if DIAGNOSTIC
        {
                register struct namecache *p;
@@ -306,67 +1022,505 @@ cache_enter(dvp, vp, cnp)
                                panic("cache_enter: duplicate");
        }
 #endif
+       /*
+        * make us available to be found via lookup
+        */
        LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
+
+       if (vp) {
+              /*
+               * add to the list of name cache entries
+               * that point at vp
+               */
+               LIST_INSERT_HEAD(&vp->v_nclinks, ncp, nc_un.nc_link);
+       } else {
+               /*
+                * this is a negative cache entry (vp == NULL)
+                * stick it on the negative cache list
+                * and record the whiteout state
+                */
+               TAILQ_INSERT_TAIL(&neghead, ncp, nc_un.nc_negentry);
+         
+               if (cnp->cn_flags & ISWHITEOUT)
+                       ncp->nc_whiteout = TRUE;
+               nchstats.ncs_negtotal++;
+
+               if (nchstats.ncs_negtotal > desiredNegNodes) {
+                      /*
+                       * if we've reached our desired limit
+                       * of negative cache entries, delete
+                       * the oldest
+                       */
+                       negp = TAILQ_FIRST(&neghead);
+                       TAILQ_REMOVE(&neghead, negp, nc_un.nc_negentry);
+
+                       cache_delete(negp, 1);
+               }
+       }
+       /*
+        * add us to the list of name cache entries that
+        * are children of dvp
+        */
+       LIST_INSERT_HEAD(&dvp->v_ncchildren, ncp, nc_child);
+
+       name_cache_unlock();
 }
 
+
+/*
+ * Initialize CRC-32 remainder table.
+ */
+static void init_crc32(void)
+{
+        /*
+        * the CRC-32 generator polynomial is:
+        *   x^32 + x^26 + x^23 + x^22 + x^16 + x^12 + x^10
+        *        + x^8  + x^7  + x^5  + x^4  + x^2  + x + 1
+        */
+        unsigned int crc32_polynomial = 0x04c11db7;
+       unsigned int i,j;
+
+       /*
+        * pre-calculate the CRC-32 remainder for each possible octet encoding
+        */
+       for (i = 0;  i < 256;  i++) {
+               unsigned int crc_rem = i << 24;
+
+               for (j = 0;  j < 8;  j++) {
+                       if (crc_rem & 0x80000000)
+                               crc_rem = (crc_rem << 1) ^ crc32_polynomial;
+                       else
+                               crc_rem = (crc_rem << 1);
+               }
+               crc32tab[i] = crc_rem;
+       }
+}
+
+
 /*
  * Name cache initialization, from vfs_init() when we are booting
  */
 void
-nchinit()
+nchinit(void)
+{
+       desiredNegNodes = (desiredvnodes / 10);
+       desiredNodes = desiredvnodes + desiredNegNodes;
+
+       TAILQ_INIT(&nchead);
+       TAILQ_INIT(&neghead);
+
+       init_crc32();
+
+       nchashtbl = hashinit(MAX(4096, (2 *desiredNodes)), M_CACHE, &nchash);
+       nchashmask = nchash;
+       nchash++;
+
+       init_string_table();
+       
+       /* Allocate mount list lock group attribute and group */
+       namecache_lck_grp_attr= lck_grp_attr_alloc_init();
+       lck_grp_attr_setstat(namecache_lck_grp_attr);
+
+       namecache_lck_grp = lck_grp_alloc_init("Name Cache",  namecache_lck_grp_attr);
+       
+       /* Allocate mount list lock attribute */
+       namecache_lck_attr = lck_attr_alloc_init();
+       //lck_attr_setdebug(namecache_lck_attr);
+
+       /* Allocate mount list lock */
+       namecache_mtx_lock = lck_mtx_alloc_init(namecache_lck_grp, namecache_lck_attr);
+
+
+}
+
+void
+name_cache_lock(void)
+{
+       lck_mtx_lock(namecache_mtx_lock);
+}
+
+void
+name_cache_unlock(void)
 {
+       lck_mtx_unlock(namecache_mtx_lock);
+
+}
+
+
+int
+resize_namecache(u_int newsize)
+{
+    struct nchashhead  *new_table;
+    struct nchashhead  *old_table;
+    struct nchashhead  *old_head, *head;
+    struct namecache   *entry, *next;
+    uint32_t           i, hashval;
+    int                        dNodes, dNegNodes;
+    u_long             new_size, old_size;
+
+    dNegNodes = (newsize / 10);
+    dNodes = newsize + dNegNodes;
+
+    // we don't support shrinking yet
+    if (dNodes < desiredNodes) {
+       return 0;
+    }
+    new_table = hashinit(2 * dNodes, M_CACHE, &nchashmask);
+    new_size  = nchashmask + 1;
+
+    if (new_table == NULL) {
+       return ENOMEM;
+    }
+
+    name_cache_lock();
+    // do the switch!
+    old_table = nchashtbl;
+    nchashtbl = new_table;
+    old_size  = nchash;
+    nchash    = new_size;
+
+    // walk the old table and insert all the entries into
+    // the new table
+    //
+    for(i=0; i < old_size; i++) {
+       old_head = &old_table[i];
+       for (entry=old_head->lh_first; entry != NULL; entry=next) {
+           //
+           // XXXdbg - Beware: this assumes that hash_string() does
+           //                  the same thing as what happens in
+           //                  lookup() over in vfs_lookup.c
+           hashval = hash_string(entry->nc_name, 0);
+           entry->nc_hashval = hashval;
+           head = NCHHASH(entry->nc_dvp, hashval);
+           
+           next = entry->nc_hash.le_next;
+           LIST_INSERT_HEAD(head, entry, nc_hash);
+       }
+    }
+    desiredNodes = dNodes;
+    desiredNegNodes = dNegNodes;
+    
+    name_cache_unlock();
+    FREE(old_table, M_CACHE);
 
-       TAILQ_INIT(&nclruhead);
-       nchashtbl = hashinit(desiredvnodes, M_CACHE, &nchash);
+    return 0;
 }
 
+static void
+cache_delete(struct namecache *ncp, int age_entry)
+{
+        nchstats.ncs_deletes++;
+
+        if (ncp->nc_vp) {
+               LIST_REMOVE(ncp, nc_un.nc_link);
+       } else {
+               TAILQ_REMOVE(&neghead, ncp, nc_un.nc_negentry);
+               nchstats.ncs_negtotal--;
+       }
+        LIST_REMOVE(ncp, nc_child);
+
+       LIST_REMOVE(ncp, nc_hash);
+       /*
+        * this field is used to indicate
+        * that the entry is in use and
+        * must be deleted before it can 
+        * be reused...
+        */
+       ncp->nc_hash.le_prev = NULL;
+
+       if (age_entry) {
+               /*
+                * make it the next one available
+                * for cache_enter's use
+                */
+               TAILQ_REMOVE(&nchead, ncp, nc_entry);
+               TAILQ_INSERT_HEAD(&nchead, ncp, nc_entry);
+       }
+       remove_name_locked(ncp->nc_name);
+       ncp->nc_name = NULL;
+}
+
+
 /*
- * Invalidate a all entries to particular vnode.
- * 
- * We actually just increment the v_id, that will do it. The entries will
- * be purged by lookup as they get found. If the v_id wraps around, we
- * need to ditch the entire cache, to avoid confusion. No valid vnode will
- * ever have (v_id == 0).
+ * purge the entry associated with the 
+ * specified vnode from the name cache
  */
 void
-cache_purge(vp)
-       struct vnode *vp;
+cache_purge(vnode_t vp)
+{
+        struct namecache *ncp;
+
+       if ((LIST_FIRST(&vp->v_nclinks) == NULL) && (LIST_FIRST(&vp->v_ncchildren) == NULL))
+               return;
+
+       name_cache_lock();
+
+       while ( (ncp = LIST_FIRST(&vp->v_nclinks)) )
+               cache_delete(ncp, 1);
+
+       while ( (ncp = LIST_FIRST(&vp->v_ncchildren)) )
+               cache_delete(ncp, 1);
+
+       name_cache_unlock();
+}
+
+/*
+ * Purge all negative cache entries that are children of the
+ * given vnode.  A case-insensitive file system (or any file
+ * system that has multiple equivalent names for the same
+ * directory entry) can use this when creating or renaming
+ * to remove negative entries that may no longer apply.
+ */
+void
+cache_purge_negatives(vnode_t vp)
 {
        struct namecache *ncp;
-       struct nchashhead *ncpp;
 
-       vp->v_id = ++nextvnodeid;
-       if (nextvnodeid != 0)
-               return;
-       for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
-               while (ncp = ncpp->lh_first)
-                       PURGE(ncp);
-       }
-       vp->v_id = ++nextvnodeid;
+       name_cache_lock();
+
+       LIST_FOREACH(ncp, &vp->v_ncchildren, nc_child)
+               if (ncp->nc_vp == NULL)
+                       cache_delete(ncp , 1);
+
+       name_cache_unlock();
 }
 
 /*
  * Flush all entries referencing a particular filesystem.
  *
  * Since we need to check it anyway, we will flush all the invalid
- * entriess at the same time.
+ * entries at the same time.
  */
 void
 cache_purgevfs(mp)
        struct mount *mp;
 {
        struct nchashhead *ncpp;
-       struct namecache *ncp, *nnp;
+       struct namecache *ncp;
 
+       name_cache_lock();
        /* Scan hash tables for applicable entries */
-       for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
-               for (ncp = ncpp->lh_first; ncp != 0; ncp = nnp) {
-                       nnp = ncp->nc_hash.le_next;
-                       if (ncp->nc_dvpid != ncp->nc_dvp->v_id ||
-                           (ncp->nc_vp && ncp->nc_vpid != ncp->nc_vp->v_id) ||
-                           ncp->nc_dvp->v_mount == mp) {
-                               PURGE(ncp);
+       for (ncpp = &nchashtbl[nchash - 1]; ncpp >= nchashtbl; ncpp--) {
+restart:         
+               for (ncp = ncpp->lh_first; ncp != 0; ncp = ncp->nc_hash.le_next) {
+                       if (ncp->nc_dvp->v_mount == mp) {
+                               cache_delete(ncp, 0);
+                               goto restart;
                        }
                }
        }
+       name_cache_unlock();
+}
+
+
+
+//
+// String ref routines
+//
+static LIST_HEAD(stringhead, string_t) *string_ref_table;
+static u_long   string_table_mask;
+static uint32_t max_chain_len=0;
+static struct stringhead *long_chain_head=NULL;
+static uint32_t filled_buckets=0;
+static uint32_t num_dups=0;
+static uint32_t nstrings=0;
+
+typedef struct string_t {
+    LIST_ENTRY(string_t)  hash_chain;
+    unsigned char        *str;
+    uint32_t              refcount;
+} string_t;
+
+
+
+static int
+resize_string_ref_table(void)
+{
+    struct stringhead *new_table;
+    struct stringhead *old_table;
+    struct stringhead *old_head, *head;
+    string_t          *entry, *next;
+    uint32_t           i, hashval;
+    u_long             new_mask, old_mask;
+
+    new_table = hashinit((string_table_mask + 1) * 2, M_CACHE, &new_mask);
+    if (new_table == NULL) {
+       return ENOMEM;
+    }
+
+    // do the switch!
+    old_table         = string_ref_table;
+    string_ref_table  = new_table;
+    old_mask          = string_table_mask;
+    string_table_mask = new_mask;
+
+    printf("resize: max chain len %d, new table size %d\n",
+          max_chain_len, new_mask + 1);
+    max_chain_len   = 0;
+    long_chain_head = NULL;
+    filled_buckets  = 0;
+
+    // walk the old table and insert all the entries into
+    // the new table
+    //
+    for(i=0; i <= old_mask; i++) {
+       old_head = &old_table[i];
+       for (entry=old_head->lh_first; entry != NULL; entry=next) {
+           hashval = hash_string(entry->str, 0);
+           head = &string_ref_table[hashval & string_table_mask];
+           if (head->lh_first == NULL) {
+               filled_buckets++;
+           }
+
+           next = entry->hash_chain.le_next;
+           LIST_INSERT_HEAD(head, entry, hash_chain);
+       }
+    }
+    
+    FREE(old_table, M_CACHE);
+
+    return 0;
+}
+
+
+static void
+init_string_table(void)
+{
+    string_ref_table = hashinit(4096, M_CACHE, &string_table_mask);
+}
+
+
+char *
+vfs_addname(const char *name, size_t len, u_int hashval, u_int flags)
+{
+        char * ptr;
+
+       name_cache_lock();
+       ptr = add_name_locked(name, len, hashval, flags);
+       name_cache_unlock();
+
+       return(ptr);
+}
+
+static char *
+add_name_locked(const char *name, size_t len, u_int hashval, __unused u_int flags)
+{
+    struct stringhead *head;
+    string_t          *entry;
+    uint32_t          chain_len = 0;
+    
+    //
+    // If the table gets more than 3/4 full, resize it
+    //
+    if (4*filled_buckets >= ((string_table_mask + 1) * 3)) {
+               if (resize_string_ref_table() != 0) {
+                       printf("failed to resize the hash table.\n");
+               }
+    }
+    if (hashval == 0) {
+       hashval = hash_string(name, 0);
+    }
+
+    head = &string_ref_table[hashval & string_table_mask];
+    for (entry=head->lh_first; entry != NULL; chain_len++, entry=entry->hash_chain.le_next) {
+       if (memcmp(entry->str, name, len) == 0 && entry->str[len] == '\0') {
+           entry->refcount++;
+           num_dups++;
+           break;
+       }
+    }
+
+    if (entry == NULL) {
+       // it wasn't already there so add it.
+       MALLOC(entry, string_t *, sizeof(string_t) + len + 1, M_TEMP, M_WAITOK);
+
+       // have to get "head" again because we could have blocked
+       // in malloc and thus head could have changed.
+       //
+       head = &string_ref_table[hashval & string_table_mask];
+       if (head->lh_first == NULL) {
+           filled_buckets++;
+       }
+
+       entry->str = (char *)((char *)entry + sizeof(string_t));
+       strncpy(entry->str, name, len);
+       entry->str[len] = '\0';
+       entry->refcount = 1;
+       LIST_INSERT_HEAD(head, entry, hash_chain);
+
+       if (chain_len > max_chain_len) {
+           max_chain_len   = chain_len;
+           long_chain_head = head;
+       }
+
+       nstrings++;
+    }
+    
+    return entry->str;
+}
+
+int
+vfs_removename(const char *nameref)
+{
+       int i;
+
+       name_cache_lock();
+       i = remove_name_locked(nameref);
+       name_cache_unlock();
+
+       return(i);
+       
+}
+
+
+static int
+remove_name_locked(const char *nameref)
+{
+    struct stringhead *head;
+    string_t          *entry;
+    uint32_t           hashval;
+    char * ptr;
+
+    hashval = hash_string(nameref, 0);
+    head = &string_ref_table[hashval & string_table_mask];
+    for (entry=head->lh_first; entry != NULL; entry=entry->hash_chain.le_next) {
+       if (entry->str == (unsigned char *)nameref) {
+           entry->refcount--;
+           if (entry->refcount == 0) {
+               LIST_REMOVE(entry, hash_chain);
+               if (head->lh_first == NULL) {
+                   filled_buckets--;
+               }
+               ptr = entry->str;
+               entry->str = NULL;
+               nstrings--;
+
+               FREE(entry, M_TEMP);
+           } else {
+               num_dups--;
+           }
+
+           return 0;
+       }
+    }
+
+    return ENOENT;
+}
+
+
+void
+dump_string_table(void)
+{
+    struct stringhead *head;
+    string_t          *entry;
+    u_long            i;
+    
+    name_cache_lock();
+    for (i = 0; i <= string_table_mask; i++) {
+       head = &string_ref_table[i];
+       for (entry=head->lh_first; entry != NULL; entry=entry->hash_chain.le_next) {
+           printf("%6d - %s\n", entry->refcount, entry->str);
+       }
+    }
+    name_cache_unlock();
 }