]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/hfs/hfs_vnops.c
xnu-3247.1.106.tar.gz
[apple/xnu.git] / bsd / hfs / hfs_vnops.c
index 0f4848c249351ee8cbf3aa092fb5849e1f8080b4..dac4b088f814fb1ae4b13f377c2df5bfcd647994 100644 (file)
@@ -1,39 +1,57 @@
 /*
- * Copyright (c) 2000-2003 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2015 Apple Inc. All rights reserved.
  *
- * @APPLE_LICENSE_HEADER_START@
+ * @APPLE_OSREFERENCE_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.
+ * 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.
  * 
- * This Original Code and all software distributed under the License are
- * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * 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@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 
+#include <stdbool.h>
 #include <sys/systm.h>
+#include <sys/param.h>
 #include <sys/kernel.h>
-#include <sys/file.h>
+#include <sys/file_internal.h>
 #include <sys/dirent.h>
 #include <sys/stat.h>
 #include <sys/buf.h>
+#include <sys/buf_internal.h>
 #include <sys/mount.h>
-#include <sys/vnode.h>
+#include <sys/vnode_if.h>
+#include <sys/vnode_internal.h>
 #include <sys/malloc.h>
-#include <sys/namei.h>
 #include <sys/ubc.h>
+#include <sys/ubc_internal.h>
+#include <sys/paths.h>
 #include <sys/quota.h>
 #include <sys/time.h>
 #include <sys/disk.h>
+#include <sys/kauth.h>
+#include <sys/uio_internal.h>
+#include <sys/fsctl.h>
+#include <sys/xattr.h>
+#include <string.h>
+#include <sys/fsevents.h>
+#include <kern/kalloc.h>
 
 #include <miscfs/specfs/specdev.h>
 #include <miscfs/fifofs/fifo.h>
 #include <machine/spl.h>
 
 #include <sys/kdebug.h>
+#include <sys/sysctl.h>
+#include <stdbool.h>
 
 #include "hfs.h"
 #include "hfs_catalog.h"
 #include "hfs_cnode.h"
-#include "hfs_lockf.h"
 #include "hfs_dbg.h"
 #include "hfs_mount.h"
 #include "hfs_quota.h"
 #include "hfs_endian.h"
+#include "hfs_kdebug.h"
+#include "hfs_cprotect.h"
+
 
 #include "hfscommon/headers/BTreesInternal.h"
 #include "hfscommon/headers/FileMgrInternal.h"
 
-#define MAKE_DELETED_NAME(NAME,FID) \
-           (void) sprintf((NAME), "%s%d", HFS_DELETE_PREFIX, (FID))
-
 #define KNDETACH_VNLOCKED 0x00000001
 
-#define CARBON_TEMP_DIR_NAME   "Cleanup At Startup"
+/* Global vfs data structures for hfs */
+
+/* Always F_FULLFSYNC? 1=yes,0=no (default due to "various" reasons is 'no') */
+int always_do_fullfsync = 0;
+SYSCTL_DECL(_vfs_generic);
+SYSCTL_INT (_vfs_generic, OID_AUTO, always_do_fullfsync, CTLFLAG_RW | CTLFLAG_LOCKED, &always_do_fullfsync, 0, "always F_FULLFSYNC when fsync is called");
 
+int hfs_makenode(struct vnode *dvp, struct vnode **vpp,
+                        struct componentname *cnp, struct vnode_attr *vap,
+                        vfs_context_t ctx);
+int hfs_metasync(struct hfsmount *hfsmp, daddr64_t node, __unused struct proc *p);
+int hfs_metasync_all(struct hfsmount *hfsmp);
 
-/* Global vfs data structures for hfs */
+int hfs_removedir(struct vnode *, struct vnode *, struct componentname *,
+                         int, int);
+int hfs_removefile(struct vnode *, struct vnode *, struct componentname *,
+                          int, int, int, struct vnode *, int);
 
+/* Used here and in cnode teardown -- for symlinks */
+int hfs_removefile_callback(struct buf *bp, void *hfsmp);
+
+enum {
+       HFS_MOVE_DATA_INCLUDE_RSRC              = 1,
+};
+typedef uint32_t hfs_move_data_options_t;
 
-extern unsigned long strtoul(const char *, char **, int);
+static int hfs_move_data(cnode_t *from_cp, cnode_t *to_cp, 
+                                                hfs_move_data_options_t options);
+static int hfs_move_fork(filefork_t *srcfork, cnode_t *src, 
+                                                filefork_t *dstfork, cnode_t *dst);
 
-extern int groupmember(gid_t gid, struct ucred *cred);
+#if HFS_COMPRESSION
+static int hfs_move_compressed(cnode_t *from_vp, cnode_t *to_vp);
+#endif
 
-static int hfs_makenode(int mode, struct vnode *dvp, struct vnode **vpp,
-                        struct componentname *cnp);
-                        
-static int hfs_vgetrsrc(struct hfsmount *hfsmp, struct vnode *vp,
-                       struct vnode **rvpp, struct proc *p);
+decmpfs_cnode* hfs_lazy_init_decmpfs_cnode (struct cnode *cp);
 
-static int hfs_metasync(struct hfsmount *hfsmp, daddr_t node, struct proc *p);
+#if FIFO
+static int hfsfifo_read(struct vnop_read_args *);
+static int hfsfifo_write(struct vnop_write_args *);
+static int hfsfifo_close(struct vnop_close_args *);
 
-static int hfs_removedir(struct vnode *, struct vnode *, struct componentname *,
-                         int);
+extern int (**fifo_vnodeop_p)(void *);
+#endif /* FIFO */
 
-static int hfs_removefile(struct vnode *, struct vnode *, struct componentname *,
-                          int);
+int hfs_vnop_close(struct vnop_close_args*);
+int hfs_vnop_create(struct vnop_create_args*);
+int hfs_vnop_exchange(struct vnop_exchange_args*);
+int hfs_vnop_fsync(struct vnop_fsync_args*);
+int hfs_vnop_mkdir(struct vnop_mkdir_args*);
+int hfs_vnop_mknod(struct vnop_mknod_args*);
+int hfs_vnop_getattr(struct vnop_getattr_args*);
+int hfs_vnop_open(struct vnop_open_args*);
+int hfs_vnop_readdir(struct vnop_readdir_args*);
+int hfs_vnop_remove(struct vnop_remove_args*);
+int hfs_vnop_rename(struct vnop_rename_args*);
+int hfs_vnop_rmdir(struct vnop_rmdir_args*);
+int hfs_vnop_symlink(struct vnop_symlink_args*);
+int hfs_vnop_setattr(struct vnop_setattr_args*);
+int hfs_vnop_readlink(struct vnop_readlink_args *);
+int hfs_vnop_pathconf(struct vnop_pathconf_args *);
+int hfs_vnop_mmap(struct vnop_mmap_args *ap);
+int hfsspec_read(struct vnop_read_args *);
+int hfsspec_write(struct vnop_write_args *);
+int hfsspec_close(struct vnop_close_args *);
 
 /* Options for hfs_removedir and hfs_removefile */
-#define HFSRM_PARENT_LOCKED    0x01
-#define HFSRM_SKIP_RESERVE     0x02
-#define HFSRM_SAVE_NAME        0x04
-#define HFSRM_RENAMEOPTS       0x07
-
+#define HFSRM_SKIP_RESERVE  0x01
 
-int hfs_write_access(struct vnode *vp, struct ucred *cred, struct proc *p, Boolean considerFlags);
 
-int hfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
-                       struct proc *p);
-int hfs_chmod(struct vnode *vp, int mode, struct ucred *cred,
-                       struct proc *p);
-int hfs_chown(struct vnode *vp, uid_t uid, gid_t gid,
-                       struct ucred *cred, struct proc *p);
 
 /*****************************************************************************
 *
@@ -106,618 +156,1610 @@ int hfs_chown(struct vnode *vp, uid_t uid, gid_t gid,
 *****************************************************************************/
 
 /*
- * Create a regular file
-#% create      dvp     L U U
-#% create      vpp     - L -
-#
- vop_create {
-     IN WILLRELE struct vnode *dvp;
-     OUT struct vnode **vpp;
-     IN struct componentname *cnp;
-     IN struct vattr *vap;
-       
-     We are responsible for freeing the namei buffer,
-        it is done in hfs_makenode()
-*/
-
-static int
-hfs_create(ap)
-       struct vop_create_args /* {
-               struct vnode *a_dvp;
-               struct vnode **a_vpp;
-               struct componentname *a_cnp;
-               struct vattr *a_vap;
-       } */ *ap;
+ * Is the given cnode either the .journal or .journal_info_block file on
+ * a volume with an active journal?  Many VNOPs use this to deny access
+ * to those files.
+ *
+ * Note: the .journal file on a volume with an external journal still
+ * returns true here, even though it does not actually hold the contents
+ * of the volume's journal.
+ */
+static _Bool
+hfs_is_journal_file(struct hfsmount *hfsmp, struct cnode *cp)
 {
-       struct vattr *vap = ap->a_vap;
-
-       return (hfs_makenode(MAKEIMODE(vap->va_type, vap->va_mode),
-                               ap->a_dvp, ap->a_vpp, ap->a_cnp));
+       if (hfsmp->jnl != NULL &&
+           (cp->c_fileid == hfsmp->hfs_jnlinfoblkid ||
+            cp->c_fileid == hfsmp->hfs_jnlfileid)) {
+               return true;
+       } else {
+               return false;
+       }
 }
 
-
 /*
- * Mknod vnode call
-
-#% mknod       dvp     L U U
-#% mknod       vpp     - X -
-#
- vop_mknod {
-     IN WILLRELE struct vnode *dvp;
-     OUT WILLRELE struct vnode **vpp;
-     IN struct componentname *cnp;
-     IN struct vattr *vap;
-     */
-/* ARGSUSED */
+ * Create a regular file.
+ */
+int
+hfs_vnop_create(struct vnop_create_args *ap)
+{
+       /*
+        * We leave handling of certain race conditions here to the caller
+        * which will have a better understanding of the semantics it
+        * requires.  For example, if it turns out that the file exists,
+        * it would be wrong of us to return a reference to the existing
+        * file because the caller might not want that and it would be
+        * misleading to suggest the file had been created when it hadn't
+        * been.  Note that our NFS server code does not set the
+        * VA_EXCLUSIVE flag so you cannot assume that callers don't want
+        * EEXIST errors if it's not set.  The common case, where users
+        * are calling open with the O_CREAT mode, is handled in VFS; when
+        * we return EEXIST, it will loop and do the look-up again.
+        */
+       return hfs_makenode(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap, ap->a_context);
+}
 
-static int
-hfs_mknod(ap)
-       struct vop_mknod_args /* {
-               struct vnode *a_dvp;
-               struct vnode **a_vpp;
-               struct componentname *a_cnp;
-               struct vattr *a_vap;
-       } */ *ap;
+/*
+ * Make device special file.
+ */
+int
+hfs_vnop_mknod(struct vnop_mknod_args *ap)
 {
-       struct vattr *vap = ap->a_vap;
+       struct vnode_attr *vap = ap->a_vap;
+       struct vnode *dvp = ap->a_dvp;
        struct vnode **vpp = ap->a_vpp;
        struct cnode *cp;
        int error;
 
-       if (VTOVCB(ap->a_dvp)->vcbSigWord != kHFSPlusSigWord) {
-               VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
-               vput(ap->a_dvp);
-               return (EOPNOTSUPP);
+       if (VTOVCB(dvp)->vcbSigWord != kHFSPlusSigWord) {
+               return (ENOTSUP);
        }
 
        /* Create the vnode */
-       error = hfs_makenode(MAKEIMODE(vap->va_type, vap->va_mode),
-                            ap->a_dvp, vpp, ap->a_cnp);
+       error = hfs_makenode(dvp, vpp, ap->a_cnp, vap, ap->a_context);
        if (error)
                return (error);
+
        cp = VTOC(*vpp);
-       cp->c_flag |= C_ACCESS | C_CHANGE | C_UPDATE;
+       cp->c_touch_acctime = TRUE;
+       cp->c_touch_chgtime = TRUE;
+       cp->c_touch_modtime = TRUE;
+
        if ((vap->va_rdev != VNOVAL) &&
            (vap->va_type == VBLK || vap->va_type == VCHR))
                cp->c_rdev = vap->va_rdev;
-       /*
-        * Remove cnode so that it will be reloaded by lookup and
-        * checked to see if it is an alias of an existing vnode.
-        * Note: unlike UFS, we don't bash v_type here.
-        */
-       vput(*vpp);
-       vgone(*vpp);
-       *vpp = 0;
+
        return (0);
 }
 
-
-/*
- * Open called.
-#% open                vp      L L L
-#
- vop_open {
-     IN struct vnode *vp;
-     IN int mode;
-     IN struct ucred *cred;
-     IN struct proc *p;
-     */
-
-
+#if HFS_COMPRESSION
+/* 
+ *     hfs_ref_data_vp(): returns the data fork vnode for a given cnode. 
+ *     In the (hopefully rare) case where the data fork vnode is not 
+ *     present, it will use hfs_vget() to create a new vnode for the
+ *     data fork. 
+ *     
+ *     NOTE: If successful and a vnode is returned, the caller is responsible
+ *     for releasing the returned vnode with vnode_rele().
+ */
 static int
-hfs_open(ap)
-       struct vop_open_args /* {
-               struct vnode *a_vp;
-               int  a_mode;
-               struct ucred *a_cred;
-               struct proc *a_p;
-       } */ *ap;
+hfs_ref_data_vp(struct cnode *cp, struct vnode **data_vp, int skiplock)
 {
-       struct vnode *vp = ap->a_vp;
-       struct filefork *fp = VTOF(vp);
-       struct timeval tv;
-
-       /*
-        * Files marked append-only must be opened for appending.
-        */
-       if ((vp->v_type != VDIR) && (VTOC(vp)->c_flags & APPEND) &&
-           (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
-               return (EPERM);
+       int vref = 0;
 
-       if (ap->a_mode & O_EVTONLY) {
-               if (vp->v_type == VREG) {
-                       ++VTOF(vp)->ff_evtonly_refs;
-               } else {
-                       ++VTOC(vp)->c_evtonly_refs;
-               };
-       };
+       if (!data_vp || !cp) /* sanity check incoming parameters */
+               return EINVAL;
+       
+       /* maybe we should take the hfs cnode lock here, and if so, use the skiplock parameter to tell us not to */
+
+       if (!skiplock) hfs_lock(cp, HFS_SHARED_LOCK, HFS_LOCK_DEFAULT);
+       struct vnode *c_vp = cp->c_vp;
+       if (c_vp) {
+               /* we already have a data vnode */
+               *data_vp = c_vp;
+               vref = vnode_ref(*data_vp);
+               if (!skiplock) hfs_unlock(cp);
+               if (vref == 0) {
+                       return 0;
+               }
+               return EINVAL;
+       }
+       /* no data fork vnode in the cnode, so ask hfs for one. */
 
-       /*
-        * On the first (non-busy) open of a fragmented
-        * file attempt to de-frag it (if its less than 20MB).
-        */
-       if ((VTOHFS(vp)->hfs_flags & HFS_READ_ONLY) ||
-           !UBCISVALID(vp) || ubc_isinuse(vp, 1)) {
-               return (0);
+       if (!cp->c_rsrc_vp) {
+               /* if we don't have either a c_vp or c_rsrc_vp, we can't really do anything useful */
+               *data_vp = NULL;
+               if (!skiplock) hfs_unlock(cp);
+               return EINVAL;
        }
-       fp = VTOF(vp);
-       if (fp->ff_blocks &&
-           fp->ff_extents[7].blockCount != 0 &&
-           fp->ff_size <= (20 * 1024 * 1024)) {
-               /* 
-                * Wait until system bootup is done (3 min).
-                */
-               microuptime(&tv);
-               if (tv.tv_sec < (60 * 3)) {
-                       return (0);
+       
+       if (0 == hfs_vget(VTOHFS(cp->c_rsrc_vp), cp->c_cnid, data_vp, 1, 0) &&
+               0 != data_vp) {
+               vref = vnode_ref(*data_vp);
+               vnode_put(*data_vp);
+               if (!skiplock) hfs_unlock(cp);
+               if (vref == 0) {
+                       return 0;
                }
-               (void) hfs_relocate(vp, VTOVCB(vp)->nextAllocation + 4096, ap->a_cred, ap->a_p);
+               return EINVAL;
        }
-
-       return (0);
+       /* there was an error getting the vnode */
+       *data_vp = NULL;
+       if (!skiplock) hfs_unlock(cp);
+       return EINVAL;
 }
 
 /*
- * Close called.
- *
- * Update the times on the cnode.
-#% close       vp      U U U
-#
- vop_close {
-     IN struct vnode *vp;
-     IN int fflag;
-     IN struct ucred *cred;
-     IN struct proc *p;
-     */
-
-
-static int
-hfs_close(ap)
-       struct vop_close_args /* {
-               struct vnode *a_vp;
-               int a_fflag;
-               struct ucred *a_cred;
-               struct proc *a_p;
-       } */ *ap;
+ *     hfs_lazy_init_decmpfs_cnode(): returns the decmpfs_cnode for a cnode,
+ *     allocating it if necessary; returns NULL if there was an allocation error.
+ *  function is non-static so that it can be used from the FCNTL handler.
+ */
+decmpfs_cnode *
+hfs_lazy_init_decmpfs_cnode(struct cnode *cp)
 {
-       register struct vnode *vp = ap->a_vp;
-       register struct cnode *cp = VTOC(vp);
-       register struct filefork *fp = VTOF(vp);
-       struct proc *p = ap->a_p;
-       struct timeval tv;
-       off_t leof;
-       u_long blks, blocksize;
-       int devBlockSize;
-       int error;
-
-       simple_lock(&vp->v_interlock);
-       if ((!UBCISVALID(vp) && vp->v_usecount > 1)
-           || (UBCISVALID(vp) && ubc_isinuse(vp, 1))) {
-               tv = time;
-               CTIMES(cp, &tv, &tv);
+       if (!cp->c_decmp) {
+               decmpfs_cnode *dp = NULL;
+               MALLOC_ZONE(dp, decmpfs_cnode *, sizeof(decmpfs_cnode), M_DECMPFS_CNODE, M_WAITOK);
+               if (!dp) {
+                       /* error allocating a decmpfs cnode */
+                       return NULL;
+               }
+               decmpfs_cnode_init(dp);
+               if (!OSCompareAndSwapPtr(NULL, dp, (void * volatile *)&cp->c_decmp)) {
+                       /* another thread got here first, so free the decmpfs_cnode we allocated */
+                       decmpfs_cnode_destroy(dp);
+                       FREE_ZONE(dp, sizeof(*dp), M_DECMPFS_CNODE);
+               }
        }
-       simple_unlock(&vp->v_interlock);
-
-       if (ap->a_fflag & O_EVTONLY) {
-               if (vp->v_type == VREG) {
-                       --VTOF(vp)->ff_evtonly_refs;
-               } else {
-                       --VTOC(vp)->c_evtonly_refs;
-               };
-       };
+       
+       return cp->c_decmp;
+}
 
-       /*
-        * VOP_CLOSE can be called with vp locked (from vclean).
-        * We check for this case using VOP_ISLOCKED and bail.
-        * 
-        * XXX During a force unmount we won't do the cleanup below!
-        */
-       if (vp->v_type == VDIR || VOP_ISLOCKED(vp))
-               return (0);
+/*
+ *     hfs_file_is_compressed(): returns 1 if the file is compressed, and 0 (zero) if not.
+ *     if the file's compressed flag is set, makes sure that the decmpfs_cnode field
+ *     is allocated by calling hfs_lazy_init_decmpfs_cnode(), then makes sure it is populated,
+ *     or else fills it in via the decmpfs_file_is_compressed() function.
+ */
+int
+hfs_file_is_compressed(struct cnode *cp, int skiplock)
+{
+       int ret = 0;
+       
+       /* fast check to see if file is compressed. If flag is clear, just answer no */
+       if (!(cp->c_bsdflags & UF_COMPRESSED)) {
+               return 0;
+       }
 
-       leof = fp->ff_size;
+       decmpfs_cnode *dp = hfs_lazy_init_decmpfs_cnode(cp);
+       if (!dp) {
+               /* error allocating a decmpfs cnode, treat the file as uncompressed */
+               return 0;
+       }
        
-       if ((fp->ff_blocks > 0) &&
-           !ISSET(cp->c_flag, C_DELETED) &&
-           ((VTOHFS(vp)->hfs_flags & HFS_READ_ONLY) == 0)) {
-               enum vtype our_type = vp->v_type;
-               u_long our_id = vp->v_id;
-               int was_nocache = ISSET(vp->v_flag, VNOCACHE_DATA);
-
-               error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
-               if (error)
-                       return (0);
-               /*
-                * Since we can context switch in vn_lock our vnode
-                * could get recycled (eg umount -f).  Double check
-                * that its still ours.
-                */
-               if (vp->v_type != our_type || vp->v_id != our_id
-                   || cp != VTOC(vp) || !UBCINFOEXISTS(vp)) {
-                       VOP_UNLOCK(vp, 0, p);
-                       return (0);
+       /* flag was set, see if the decmpfs_cnode state is valid (zero == invalid) */
+       uint32_t decmpfs_state = decmpfs_cnode_get_vnode_state(dp);
+       switch(decmpfs_state) {
+               case FILE_IS_COMPRESSED:
+               case FILE_IS_CONVERTING: /* treat decompressing files as if they are compressed */
+                       return 1;
+               case FILE_IS_NOT_COMPRESSED:
+                       return 0;
+               /* otherwise the state is not cached yet */
+       }
+       
+       /* decmpfs hasn't seen this file yet, so call decmpfs_file_is_compressed() to init the decmpfs_cnode struct */
+       struct vnode *data_vp = NULL;
+       if (0 == hfs_ref_data_vp(cp, &data_vp, skiplock)) {
+               if (data_vp) {
+                       ret = decmpfs_file_is_compressed(data_vp, VTOCMP(data_vp)); // fill in decmpfs_cnode
+                       vnode_rele(data_vp);
                }
+       }
+       return ret;
+}
 
-               /*
-                * Last chance to explicitly zero out the areas
-                * that are currently marked invalid:
-                */
-               VOP_DEVBLOCKSIZE(cp->c_devvp, &devBlockSize);
-               (void) cluster_push(vp);
-               SET(vp->v_flag, VNOCACHE_DATA); /* Don't cache zeros */
-               while (!CIRCLEQ_EMPTY(&fp->ff_invalidranges)) {
-                       struct rl_entry *invalid_range = CIRCLEQ_FIRST(&fp->ff_invalidranges);
-                       off_t start = invalid_range->rl_start;
-                       off_t end = invalid_range->rl_end;
-               
-                       /* The range about to be written must be validated
-                        * first, so that VOP_CMAP() will return the
-                        * appropriate mapping for the cluster code:
-                        */
-                       rl_remove(start, end, &fp->ff_invalidranges);
-
-                       (void) cluster_write(vp, (struct uio *) 0, leof,
-                                       invalid_range->rl_end + 1, invalid_range->rl_start,
-                                       (off_t)0, devBlockSize, IO_HEADZEROFILL | IO_NOZERODIRTY);
+/*     hfs_uncompressed_size_of_compressed_file() - get the uncompressed size of the file.
+ *     if the caller has passed a valid vnode (has a ref count > 0), then hfsmp and fid are not required.
+ *     if the caller doesn't have a vnode, pass NULL in vp, and pass valid hfsmp and fid.
+ *     files size is returned in size (required)
+ *     if the indicated file is a directory (or something that doesn't have a data fork), then this call
+ *     will return an error and the caller should fall back to treating the item as an uncompressed file
+ */
+int
+hfs_uncompressed_size_of_compressed_file(struct hfsmount *hfsmp, struct vnode *vp, cnid_t fid, off_t *size, int skiplock)
+{
+       int ret = 0;
+       int putaway = 0;                                                                        /* flag to remember if we used hfs_vget() */
 
-                       if (ISSET(vp->v_flag, VHASDIRTY))
-                               (void) cluster_push(vp);
+       if (!size) {
+               return EINVAL;                                                                  /* no place to put the file size */
+       }
 
-                       cp->c_flag |= C_MODIFIED;
+       if (NULL == vp) {
+               if (!hfsmp || !fid) {                                                   /* make sure we have the required parameters */
+                       return EINVAL;
                }
-               cp->c_flag &= ~C_ZFWANTSYNC;
-               cp->c_zftimeout = 0;
-               blocksize = VTOVCB(vp)->blockSize;
-               blks = leof / blocksize;
-               if (((off_t)blks * (off_t)blocksize) != leof)
-                       blks++;
-               /*
-                * Shrink the peof to the smallest size neccessary to contain the leof.
-                */
-               if (blks < fp->ff_blocks)
-                       (void) VOP_TRUNCATE(vp, leof, IO_NDELAY, ap->a_cred, p);
-               (void) cluster_push(vp);
-
-               if (!was_nocache)
-                       CLR(vp->v_flag, VNOCACHE_DATA);
-               
-               /*
-                * If the VOP_TRUNCATE didn't happen to flush the vnode's
-                * information out to disk, force it to be updated now that
-                * all invalid ranges have been zero-filled and validated:
-                */
-               if (cp->c_flag & C_MODIFIED) {
-                       tv = time;
-                       VOP_UPDATE(vp, &tv, &tv, 0);
+               if (0 != hfs_vget(hfsmp, fid, &vp, skiplock, 0)) {              /* vnode is null, use hfs_vget() to get it */
+                       vp = NULL;
+               } else {
+                       putaway = 1;                                                            /* note that hfs_vget() was used to aquire the vnode */
                }
-               VOP_UNLOCK(vp, 0, p);
        }
-       if ((vp->v_flag & VSYSTEM) && (vp->v_usecount == 1))
-               vgone(vp);
-       return (0);
+       /* this double check for compression (hfs_file_is_compressed)
+        * ensures the cached size is present in case decmpfs hasn't 
+        * encountered this node yet.
+        */
+       if (vp) {
+               if (hfs_file_is_compressed(VTOC(vp), skiplock) ) {
+                       *size = decmpfs_cnode_get_vnode_cached_size(VTOCMP(vp));        /* file info will be cached now, so get size */
+               } else {
+                       if (VTOCMP(vp) && VTOCMP(vp)->cmp_type >= CMP_MAX) {
+                               if (VTOCMP(vp)->cmp_type != DATALESS_CMPFS_TYPE) {
+                                       // if we don't recognize this type, just use the real data fork size
+                                       if (VTOC(vp)->c_datafork) {
+                                               *size = VTOC(vp)->c_datafork->ff_size;
+                                               ret = 0;
+                                       } else {
+                                               ret = EINVAL;
+                                       }
+                               } else {
+                                       *size = decmpfs_cnode_get_vnode_cached_size(VTOCMP(vp));        /* file info will be cached now, so get size */
+                                       ret = 0;
+                               }
+                       } else {
+                               ret = EINVAL;
+                       }
+               }
+       }
+       
+       if (putaway) {          /* did we use hfs_vget() to get this vnode? */
+               vnode_put(vp);  /* if so, release it and set it to null */
+               vp = NULL;
+       }
+       return ret;
 }
 
-/*
-#% access      vp      L L L
-#
- vop_access {
-     IN struct vnode *vp;
-     IN int mode;
-     IN struct ucred *cred;
-     IN struct proc *p;
+int
+hfs_hides_rsrc(vfs_context_t ctx, struct cnode *cp, int skiplock)
+{
+       if (ctx == decmpfs_ctx)
+               return 0;
+       if (!hfs_file_is_compressed(cp, skiplock))
+               return 0;
+       return decmpfs_hides_rsrc(ctx, cp->c_decmp);
+}
 
-     */
+int
+hfs_hides_xattr(vfs_context_t ctx, struct cnode *cp, const char *name, int skiplock)
+{
+       if (ctx == decmpfs_ctx)
+               return 0;
+       if (!hfs_file_is_compressed(cp, skiplock))
+               return 0;
+       return decmpfs_hides_xattr(ctx, cp->c_decmp, name);
+}
+#endif /* HFS_COMPRESSION */
+               
 
-static int
-hfs_access(ap)
-       struct vop_access_args /* {
-               struct vnode *a_vp;
-               int a_mode;
-               struct ucred *a_cred;
-               struct proc *a_p;
-       } */ *ap;
+//
+// This function gets the doc_tombstone structure for the
+// current thread.  If the thread doesn't have one, the
+// structure is allocated.
+//
+static struct doc_tombstone *
+get_uthread_doc_tombstone(void)
 {
-       struct vnode *vp = ap->a_vp;
-       struct cnode *cp = VTOC(vp);
-       struct ucred *cred = ap->a_cred;
-       register gid_t *gp;
-       mode_t mode = ap->a_mode;
-       mode_t mask = 0;
-       int i;
-       int error;
+       struct  uthread *ut;
+       ut = get_bsdthread_info(current_thread());
 
-       /*
-        * Disallow write attempts on read-only file systems;
-        * unless the file is a socket, fifo, or a block or
-        * character device resident on the file system.
-        */
-       if (mode & VWRITE) {
-               switch (vp->v_type) {
-               case VDIR:
-               case VLNK:
-               case VREG:
-                       if (VTOHFS(vp)->hfs_flags & HFS_READ_ONLY)
-                               return (EROFS);
-#if QUOTA
-                       if ((error = hfs_getinoquota(cp)))
-                               return (error);
-#endif /* QUOTA */
-                       break;
+       if (ut->t_tombstone == NULL) {
+               ut->t_tombstone = kalloc(sizeof(struct doc_tombstone));
+               if (ut->t_tombstone) {
+                       memset(ut->t_tombstone, 0, sizeof(struct doc_tombstone));
                }
-               /* If immutable bit set, nobody gets to write it. */
-               if (cp->c_flags & IMMUTABLE)
-                       return (EPERM);
        }
+       
+       return ut->t_tombstone;
+}
 
+//
+// This routine clears out the current tombstone for the
+// current thread and if necessary passes the doc-id of
+// the tombstone on to the dst_cnode.
+//
+// If the doc-id transfers to dst_cnode, we also generate
+// a doc-id changed fsevent.  Unlike all the other fsevents,
+// doc-id changed events can only be generated here in HFS
+// where we have the necessary info.
+// 
+static void
+clear_tombstone_docid(struct  doc_tombstone *ut, __unused struct hfsmount *hfsmp, struct cnode *dst_cnode)
+{
+       uint32_t old_id = ut->t_lastop_document_id;
 
-       /* Otherwise, user id 0 always gets access. */
-       if (cred->cr_uid == 0)
-               return (0);
-
-       mask = 0;
-
-       /* Otherwise, check the owner. */
-       if ( (cp->c_uid == cred->cr_uid) || (cp->c_uid == UNKNOWNUID) ) {
-               if (mode & VEXEC)
-                       mask |= S_IXUSR;
-               if (mode & VREAD)
-                       mask |= S_IRUSR;
-               if (mode & VWRITE)
-                       mask |= S_IWUSR;
-               return ((cp->c_mode & mask) == mask ? 0 : EACCES);
-       }
+       ut->t_lastop_document_id = 0;
+       ut->t_lastop_parent = NULL;
+       ut->t_lastop_parent_vid = 0;
+       ut->t_lastop_filename[0] = '\0';
 
-       /* Otherwise, check the groups. */
-       if (! (VTOVFS(vp)->mnt_flag & MNT_UNKNOWNPERMISSIONS)) {
-               for (i = 0, gp = cred->cr_groups; i < cred->cr_ngroups; i++, gp++)
-                       if (cp->c_gid == *gp) {
-                               if (mode & VEXEC)
-                                       mask |= S_IXGRP;
-                               if (mode & VREAD)
-                                       mask |= S_IRGRP;
-                               if (mode & VWRITE)
-                                       mask |= S_IWGRP;
-                               return ((cp->c_mode & mask) == mask ? 0 : EACCES);
-                       }
+       //
+       // If the lastop item is still the same and needs to be cleared,
+       // clear it.
+       //
+       if (dst_cnode && old_id && ut->t_lastop_item && vnode_vid(ut->t_lastop_item) == ut->t_lastop_item_vid) {
+               //
+               // clear the document_id from the file that used to have it.
+               // XXXdbg - we need to lock the other vnode and make sure to
+               // update it on disk.
+               //
+               struct cnode *ocp = VTOC(ut->t_lastop_item);
+               struct FndrExtendedFileInfo *ofip = (struct FndrExtendedFileInfo *)((char *)&ocp->c_attr.ca_finderinfo + 16);
+
+               // printf("clearing doc-id from ino %d\n", ocp->c_desc.cd_cnid);
+               ofip->document_id = 0;
+               ocp->c_bsdflags &= ~UF_TRACKED;
+               ocp->c_flag |= C_MODIFIED;
+               /* cat_update(hfsmp, &ocp->c_desc, &ocp->c_attr, NULL, NULL); */
+
+       }
+
+#if CONFIG_FSE
+       if (dst_cnode && old_id) {
+               struct FndrExtendedFileInfo *fip = (struct FndrExtendedFileInfo *)((char *)&dst_cnode->c_attr.ca_finderinfo + 16);
+
+               add_fsevent(FSE_DOCID_CHANGED, vfs_context_current(),
+                           FSE_ARG_DEV, hfsmp->hfs_raw_dev,
+                           FSE_ARG_INO, (ino64_t)ut->t_lastop_fileid,    // src inode #
+                           FSE_ARG_INO, (ino64_t)dst_cnode->c_fileid,    // dst inode #
+                           FSE_ARG_INT32, (uint32_t)fip->document_id,
+                           FSE_ARG_DONE);
        }
-
-       /* Otherwise, check everyone else. */
-       if (mode & VEXEC)
-               mask |= S_IXOTH;
-       if (mode & VREAD)
-               mask |= S_IROTH;
-       if (mode & VWRITE)
-               mask |= S_IWOTH;
-       return ((cp->c_mode & mask) == mask ? 0 : EACCES);
+#endif
+       // last, clear these now that we're all done
+       ut->t_lastop_item     = NULL;
+       ut->t_lastop_fileid   = 0;
+       ut->t_lastop_item_vid = 0;
 }
 
 
-
-/*
-#% getattr     vp      = = =
-#
- vop_getattr {
-     IN struct vnode *vp;
-     IN struct vattr *vap;
-     IN struct ucred *cred;
-     IN struct proc *p;
-
-     */
-
-
-/* ARGSUSED */
+//
+// This function is used to filter out operations on temp
+// filenames.  We have to filter out operations on certain
+// temp filenames to work-around questionable application
+// behavior from apps like Autocad that perform unusual
+// sequences of file system operations for a "safe save".
 static int
-hfs_getattr(ap)
-       struct vop_getattr_args /* {
-               struct vnode *a_vp;
-               struct vattr *a_vap;
-               struct ucred *a_cred;
-               struct proc *a_p;
-       } */ *ap;
+is_ignorable_temp_name(const char *nameptr, int len)
 {
-       struct vnode *vp = ap->a_vp;
-       struct cnode *cp = VTOC(vp);
-       struct vattr *vap = ap->a_vap;
-       struct timeval tv;
-
-       tv = time;
-       CTIMES(cp, &tv, &tv);
-
-       vap->va_type = vp->v_type;
-       vap->va_mode = cp->c_mode;
-       vap->va_nlink = cp->c_nlink;
-       /*
-        * [2856576]  Since we are dynamically changing the owner, also
-        * effectively turn off the set-user-id and set-group-id bits,
-        * just like chmod(2) would when changing ownership.  This prevents
-        * a security hole where set-user-id programs run as whoever is
-        * logged on (or root if nobody is logged in yet!)
-        */
-       if (cp->c_uid == UNKNOWNUID) {
-               vap->va_mode &= ~(S_ISUID | S_ISGID);
-               vap->va_uid = ap->a_cred->cr_uid;
-       } else {
-               vap->va_uid = cp->c_uid;
+       if (len == 0) {
+               len = strlen(nameptr);
        }
-       vap->va_gid = cp->c_gid;
-       vap->va_fsid = cp->c_dev;
-       /*
-        * Exporting file IDs from HFS Plus:
-        *
-        * For "normal" files the c_fileid is the same value as the
-        * c_cnid.  But for hard link files, they are different - the
-        * c_cnid belongs to the active directory entry (ie the link)
-        * and the c_fileid is for the actual inode (ie the data file).
-        *
-        * The stat call (getattr) will always return the c_fileid
-        * and Carbon APIs, which are hardlink-ignorant, will always
-        * receive the c_cnid (from getattrlist).
-        */
-       vap->va_fileid = cp->c_fileid;
-       vap->va_atime.tv_sec = cp->c_atime;
-       vap->va_atime.tv_nsec = 0;
-       vap->va_mtime.tv_sec = cp->c_mtime;
-       vap->va_mtime.tv_nsec = cp->c_mtime_nsec;
-       vap->va_ctime.tv_sec = cp->c_ctime;
-       vap->va_ctime.tv_nsec = 0;
-       vap->va_gen = 0;
-       vap->va_flags = cp->c_flags;
-       vap->va_rdev = 0;
-       vap->va_blocksize = VTOVFS(vp)->mnt_stat.f_iosize;
-       vap->va_filerev = 0;
-       if (vp->v_type == VDIR) {
-               vap->va_size = cp->c_nlink * AVERAGE_HFSDIRENTRY_SIZE;
-               vap->va_bytes = 0;
-       } else {
-               vap->va_size = VTOF(vp)->ff_size;
-               vap->va_bytes = (u_quad_t)cp->c_blocks *
-                                   (u_quad_t)VTOVCB(vp)->blockSize;
-               if (vp->v_type == VBLK || vp->v_type == VCHR)
-                       vap->va_rdev = cp->c_rdev;
+       
+       if (   strncmp(nameptr, "atmp", 4) == 0
+          || (len > 4 && strncmp(nameptr+len-4, ".bak", 4) == 0)
+          || (len > 4 && strncmp(nameptr+len-4, ".tmp", 4) == 0)) {
+               return 1;
        }
-       return (0);
-}
-
-/*
- * Set attribute vnode op. called from several syscalls
-#% setattr     vp      L L L
-#
- vop_setattr {
-     IN struct vnode *vp;
-     IN struct vattr *vap;
-     IN struct ucred *cred;
-     IN struct proc *p;
 
-     */
+       return 0;
+}
 
+//
+// Decide if we need to save a tombstone or not.  Normally we always
+// save a tombstone - but if there already is one and the name we're
+// given is an ignorable name, then we will not save a tombstone.
+// 
 static int
-hfs_setattr(ap)
-       struct vop_setattr_args /* {
-               struct vnode *a_vp;
-               struct vattr *a_vap;
-               struct ucred *a_cred;
-               struct proc *a_p;
-       } */ *ap;
+should_save_docid_tombstone(struct doc_tombstone *ut, struct vnode *vp, struct componentname *cnp)
 {
-       struct vattr *vap = ap->a_vap;
-       struct vnode *vp = ap->a_vp;
-       struct cnode *cp = VTOC(vp);
-       struct ucred *cred = ap->a_cred;
-       struct proc *p = ap->a_p;
-       struct timeval atimeval, mtimeval;
-       int error;
-
-       /*
-        * Check for unsettable attributes.
-        */
-       if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
-           (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
-           (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
-           ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
-               return (EINVAL);
+       if (cnp->cn_nameptr == NULL) {
+               return 0;
        }
 
-       // XXXdbg
-       // don't allow people to set the attributes of symlinks
-       // (nfs has a bad habit of doing ths and it can cause
-       // problems for journaling).
-       //
-       if (vp->v_type == VLNK) {
+       if (ut->t_lastop_document_id && ut->t_lastop_item == vp && is_ignorable_temp_name(cnp->cn_nameptr, cnp->cn_namelen)) {
                return 0;
        }
 
+       return 1;
+}
 
 
-       if (vap->va_flags != VNOVAL) {
-               if (VTOHFS(vp)->hfs_flags & HFS_READ_ONLY)
-                       return (EROFS);
-               if ((error = hfs_chflags(vp, vap->va_flags, cred, p)))
-                       return (error);
-               if (vap->va_flags & (IMMUTABLE | APPEND))
-                       return (0);
+//
+// This function saves a tombstone for the given vnode and name.  The
+// tombstone represents the parent directory and name where the document
+// used to live and the document-id of that file.  This info is recorded
+// in the doc_tombstone structure hanging off the uthread (which assumes
+// that all safe-save operations happen on the same thread).
+//
+// If later on the same parent/name combo comes back into existence then
+// we'll preserve the doc-id from this vnode onto the new vnode.
+//
+static void
+save_tombstone(struct hfsmount *hfsmp, struct vnode *dvp, struct vnode *vp, struct componentname *cnp, int for_unlink)
+{
+       struct cnode *cp = VTOC(vp);
+       struct  doc_tombstone *ut;
+       ut = get_uthread_doc_tombstone();
+                               
+       if (for_unlink && vp->v_type == VREG && cp->c_linkcount > 1) {
+               //
+               // a regular file that is being unlinked and that is also
+               // hardlinked should not clear the UF_TRACKED state or
+               // mess with the tombstone because somewhere else in the
+               // file system the file is still alive.
+               // 
+               return;
+       }
+
+       ut->t_lastop_parent     = dvp;
+       ut->t_lastop_parent_vid = vnode_vid(dvp);
+       ut->t_lastop_fileid     = cp->c_fileid;
+       if (for_unlink) {
+               ut->t_lastop_item      = NULL;
+               ut->t_lastop_item_vid  = 0;
+       } else {
+               ut->t_lastop_item      = vp;
+               ut->t_lastop_item_vid  = vnode_vid(vp);
+       }
+               
+       strlcpy((char *)&ut->t_lastop_filename[0], cnp->cn_nameptr, sizeof(ut->t_lastop_filename));
+               
+       struct FndrExtendedFileInfo *fip = (struct FndrExtendedFileInfo *)((char *)&cp->c_attr.ca_finderinfo + 16);
+       ut->t_lastop_document_id = fip->document_id;
+
+       if (for_unlink) {
+               // clear this so it's never returned again
+               fip->document_id = 0;
+               cp->c_bsdflags &= ~UF_TRACKED;
+
+               if (ut->t_lastop_document_id) {
+                       (void) cat_update(hfsmp, &cp->c_desc, &cp->c_attr, NULL, NULL);
+
+#if CONFIG_FSE
+                       // this event is more of a "pending-delete" 
+                       add_fsevent(FSE_DOCID_CHANGED, vfs_context_current(),
+                                   FSE_ARG_DEV, hfsmp->hfs_raw_dev,
+                                   FSE_ARG_INO, (ino64_t)cp->c_fileid,       // src inode #
+                                   FSE_ARG_INO, (ino64_t)0,                  // dst inode #
+                                   FSE_ARG_INT32, ut->t_lastop_document_id,  // document id
+                                   FSE_ARG_DONE);
+#endif
+               }
        }
+}
 
-       if (cp->c_flags & (IMMUTABLE | APPEND))
+
+/*
+ * Open a file/directory.
+ */
+int
+hfs_vnop_open(struct vnop_open_args *ap)
+{
+       struct vnode *vp = ap->a_vp;
+       struct filefork *fp;
+       struct timeval tv;
+       int error;
+       static int past_bootup = 0;
+       struct cnode *cp = VTOC(vp);
+       struct hfsmount *hfsmp = VTOHFS(vp);
+       
+#if HFS_COMPRESSION
+       if (ap->a_mode & FWRITE) {
+               /* open for write */
+               if ( hfs_file_is_compressed(cp, 1) ) { /* 1 == don't take the cnode lock */
+                       /* opening a compressed file for write, so convert it to decompressed */
+                       struct vnode *data_vp = NULL;
+                       error = hfs_ref_data_vp(cp, &data_vp, 1); /* 1 == don't take the cnode lock */
+                       if (0 == error) {
+                               if (data_vp) {
+                                       error = decmpfs_decompress_file(data_vp, VTOCMP(data_vp), -1, 1, 0);
+                                       vnode_rele(data_vp);
+                               } else {
+                                       error = EINVAL;
+                               }
+                       }
+                       if (error != 0)
+                               return error;
+               }
+       } else {
+               /* open for read */
+               if (hfs_file_is_compressed(cp, 1) ) { /* 1 == don't take the cnode lock */
+                       if (VNODE_IS_RSRC(vp)) {
+                               /* opening the resource fork of a compressed file, so nothing to do */
+                       } else {
+                               /* opening a compressed file for read, make sure it validates */
+                               error = decmpfs_validate_compressed_file(vp, VTOCMP(vp));
+                               if (error != 0)
+                                       return error;
+                       }
+               }
+       }
+#endif
+
+       /*
+        * Files marked append-only must be opened for appending.
+        */
+       if ((cp->c_bsdflags & APPEND) && !vnode_isdir(vp) &&
+           (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
                return (EPERM);
 
-       // XXXdbg - don't allow modification of the journal or journal_info_block
-       if (VTOHFS(vp)->jnl && cp->c_datafork) {
-               struct HFSPlusExtentDescriptor *extd;
+       if (vnode_isreg(vp) && !UBCINFOEXISTS(vp))
+               return (EBUSY);  /* file is in use by the kernel */
+
+       /* Don't allow journal to be opened externally. */
+       if (hfs_is_journal_file(hfsmp, cp))
+               return (EPERM);
+
+       bool have_lock = false;
+
+#if CONFIG_PROTECT
+       if (ISSET(ap->a_mode, FENCRYPTED) && cp->c_cpentry && vnode_isreg(vp)) {
+               bool have_trunc_lock = false;
+
 
-               extd = &cp->c_datafork->ff_extents[0];
-               if (extd->startBlock == VTOVCB(vp)->vcbJinfoBlock || extd->startBlock == VTOHFS(vp)->jnl_start) {
-                       return EPERM;
+               if ((error = hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT))) {
+                       if (have_trunc_lock)
+                               hfs_unlock_truncate(cp, 0);
+                       return error;
+               }
+
+               have_lock = true;
+
+               if (cp->c_cpentry->cp_raw_open_count + 1
+                       < cp->c_cpentry->cp_raw_open_count) {
+                       // Overflow; too many raw opens on this file
+                       hfs_unlock(cp);
+                       if (have_trunc_lock)
+                               hfs_unlock_truncate(cp, 0);
+                       return ENFILE;
                }
+
+
+               if (have_trunc_lock)
+                       hfs_unlock_truncate(cp, 0);
+
+               ++cp->c_cpentry->cp_raw_open_count;
+       }
+#endif
+
+       if ((hfsmp->hfs_flags & HFS_READ_ONLY) ||
+           (hfsmp->jnl == NULL) ||
+#if NAMEDSTREAMS
+           !vnode_isreg(vp) || vnode_isinuse(vp, 0) || vnode_isnamedstream(vp)) {
+#else
+           !vnode_isreg(vp) || vnode_isinuse(vp, 0)) {
+#endif
+
+#if CONFIG_PROTECT
+               if (have_lock)
+                       hfs_unlock(cp);
+#endif
+
+               return (0);
        }
 
+       if (!have_lock && (error = hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT)))
+               return (error);
+
+#if QUOTA
+       /* If we're going to write to the file, initialize quotas. */
+       if ((ap->a_mode & FWRITE) && (hfsmp->hfs_flags & HFS_QUOTAS))
+               (void)hfs_getinoquota(cp);
+#endif /* QUOTA */
+
        /*
-        * Go through the fields and update iff not VNOVAL.
+        * On the first (non-busy) open of a fragmented
+        * file attempt to de-frag it (if its less than 20MB).
         */
-       if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
-               if (VTOHFS(vp)->hfs_flags & HFS_READ_ONLY)
-                       return (EROFS);
-               if ((error = hfs_chown(vp, vap->va_uid, vap->va_gid, cred, p)))
+       fp = VTOF(vp);
+       if (fp->ff_blocks &&
+           fp->ff_extents[7].blockCount != 0 &&
+           fp->ff_size <= (20 * 1024 * 1024)) {
+               int no_mods = 0;
+               struct timeval now;
+               /* 
+                * Wait until system bootup is done (3 min).
+                * And don't relocate a file that's been modified
+                * within the past minute -- this can lead to
+                * system thrashing.
+                */
+
+               if (!past_bootup) {
+                       microuptime(&tv);
+                       if (tv.tv_sec > (60*3)) {
+                               past_bootup = 1;
+                       }
+               }
+               
+               microtime(&now);
+               if ((now.tv_sec - cp->c_mtime) > 60) {  
+                       no_mods = 1;
+               } 
+               
+               if (past_bootup && no_mods) {
+                       (void) hfs_relocate(vp, hfsmp->nextAllocation + 4096,
+                                       vfs_context_ucred(ap->a_context),
+                                       vfs_context_proc(ap->a_context));
+               }
+       }
+
+       hfs_unlock(cp);
+
+       return (0);
+}
+
+
+/*
+ * Close a file/directory.
+ */
+int
+hfs_vnop_close(ap)
+       struct vnop_close_args /* {
+               struct vnode *a_vp;
+               int a_fflag;
+               vfs_context_t a_context;
+       } */ *ap;
+{
+       register struct vnode *vp = ap->a_vp;
+       register struct cnode *cp;
+       struct proc *p = vfs_context_proc(ap->a_context);
+       struct hfsmount *hfsmp;
+       int busy;
+       int tooktrunclock = 0;
+       int knownrefs = 0;
+
+       if ( hfs_lock(VTOC(vp), HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT) != 0)
+               return (0);
+       cp = VTOC(vp);
+       hfsmp = VTOHFS(vp);
+
+#if CONFIG_PROTECT
+       if (cp->c_cpentry && ISSET(ap->a_fflag, FENCRYPTED) && vnode_isreg(vp)) {
+               assert(cp->c_cpentry->cp_raw_open_count > 0);
+               --cp->c_cpentry->cp_raw_open_count;
+       }
+#endif
+
+       /* 
+        * If the rsrc fork is a named stream, it can cause the data fork to
+        * stay around, preventing de-allocation of these blocks. 
+        * Do checks for truncation on close. Purge extra extents if they exist.
+        * Make sure the vp is not a directory, and that it has a resource fork,
+        * and that resource fork is also a named stream.
+        */
+
+       if ((vp->v_type == VREG) && (cp->c_rsrc_vp)
+                       && (vnode_isnamedstream(cp->c_rsrc_vp))) {
+               uint32_t blks;
+
+               blks = howmany(VTOF(vp)->ff_size, VTOVCB(vp)->blockSize);
+               /*
+                * If there are extra blocks and there are only 2 refs on
+                * this vp (ourselves + rsrc fork holding ref on us), go ahead
+                * and try to truncate.
+                */
+               if ((blks < VTOF(vp)->ff_blocks) && (!vnode_isinuse(vp, 2))) {
+                       // release cnode lock; must acquire truncate lock BEFORE cnode lock
+                       hfs_unlock(cp);
+
+                       hfs_lock_truncate(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
+                       tooktrunclock = 1;
+
+                       if (hfs_lock(VTOC(vp), HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT) != 0) { 
+                               hfs_unlock_truncate(cp, HFS_LOCK_DEFAULT);
+                               // bail out if we can't re-acquire cnode lock
+                               return 0;
+                       }
+                       // now re-test to make sure it's still valid
+                       if (cp->c_rsrc_vp) {
+                               knownrefs = 1 + vnode_isnamedstream(cp->c_rsrc_vp);
+                               if (!vnode_isinuse(vp, knownrefs)){
+                                       // now we can truncate the file, if necessary
+                                       blks = howmany(VTOF(vp)->ff_size, VTOVCB(vp)->blockSize);
+                                       if (blks < VTOF(vp)->ff_blocks){
+                                               (void) hfs_truncate(vp, VTOF(vp)->ff_size, IO_NDELAY,
+                                                                                       0, ap->a_context);
+                                       }
+                               }
+                       }
+               }
+       }
+
+
+       // if we froze the fs and we're exiting, then "thaw" the fs 
+       if (hfsmp->hfs_freeze_state == HFS_FROZEN
+           && hfsmp->hfs_freezing_proc == p && proc_exiting(p)) {
+               hfs_thaw(hfsmp, p);
+       }
+
+       busy = vnode_isinuse(vp, 1);
+
+       if (busy) {
+               hfs_touchtimes(VTOHFS(vp), cp); 
+       }
+       if (vnode_isdir(vp)) {
+               hfs_reldirhints(cp, busy);
+       } else if (vnode_issystem(vp) && !busy) {
+               vnode_recycle(vp);
+       }
+
+       if (tooktrunclock){
+               hfs_unlock_truncate(cp, HFS_LOCK_DEFAULT);
+       }
+       hfs_unlock(cp);
+
+       if (ap->a_fflag & FWASWRITTEN) {
+               hfs_sync_ejectable(hfsmp);
+       }
+
+       return (0);
+}
+
+static bool hfs_should_generate_document_id(hfsmount_t *hfsmp, cnode_t *cp)
+{
+       return (!ISSET(hfsmp->hfs_flags, HFS_READ_ONLY)
+                       && ISSET(cp->c_bsdflags, UF_TRACKED)
+                       && cp->c_desc.cd_cnid != kHFSRootFolderID
+                       && (S_ISDIR(cp->c_mode) || S_ISREG(cp->c_mode) || S_ISLNK(cp->c_mode)));
+}
+
+/*
+ * Get basic attributes.
+ */
+int
+hfs_vnop_getattr(struct vnop_getattr_args *ap)
+{
+#define VNODE_ATTR_TIMES  \
+       (VNODE_ATTR_va_access_time|VNODE_ATTR_va_change_time|VNODE_ATTR_va_modify_time)
+#define VNODE_ATTR_AUTH  \
+       (VNODE_ATTR_va_mode | VNODE_ATTR_va_uid | VNODE_ATTR_va_gid | \
+         VNODE_ATTR_va_flags | VNODE_ATTR_va_acl)
+
+       struct vnode *vp = ap->a_vp;
+       struct vnode_attr *vap = ap->a_vap;
+       struct vnode *rvp = NULLVP;
+       struct hfsmount *hfsmp;
+       struct cnode *cp;
+       uint64_t data_size;
+       enum vtype v_type;
+       int error = 0;
+       cp = VTOC(vp);
+
+#if HFS_COMPRESSION
+       /* we need to inspect the decmpfs state of the file before we take the hfs cnode lock */
+       int compressed = 0;
+       int hide_size = 0;
+       off_t uncompressed_size = -1;
+       if (VATTR_IS_ACTIVE(vap, va_data_size) || VATTR_IS_ACTIVE(vap, va_total_alloc) || VATTR_IS_ACTIVE(vap, va_data_alloc) || VATTR_IS_ACTIVE(vap, va_total_size)) {
+               /* we only care about whether the file is compressed if asked for the uncompressed size */
+               if (VNODE_IS_RSRC(vp)) {
+                       /* if it's a resource fork, decmpfs may want us to hide the size */
+                       hide_size = hfs_hides_rsrc(ap->a_context, cp, 0);
+               } else {
+                       /* if it's a data fork, we need to know if it was compressed so we can report the uncompressed size */
+                       compressed = hfs_file_is_compressed(cp, 0);
+               }
+               if ((VATTR_IS_ACTIVE(vap, va_data_size) || VATTR_IS_ACTIVE(vap, va_total_size))) {
+                       // if it's compressed 
+                       if (compressed || (!VNODE_IS_RSRC(vp) && cp->c_decmp && cp->c_decmp->cmp_type >= CMP_MAX)) {
+                               if (0 != hfs_uncompressed_size_of_compressed_file(NULL, vp, 0, &uncompressed_size, 0)) {
+                                       /* failed to get the uncompressed size, we'll check for this later */
+                                       uncompressed_size = -1;
+                               } else {
+                                       // fake that it's compressed
+                                       compressed = 1;
+                               }
+                       }
+               }
+       }
+#endif
+
+       /*
+        * Shortcut for vnode_authorize path.  Each of the attributes
+        * in this set is updated atomically so we don't need to take
+        * the cnode lock to access them.
+        */
+       if ((vap->va_active & ~VNODE_ATTR_AUTH) == 0) {
+               /* Make sure file still exists. */
+               if (cp->c_flag & C_NOEXISTS)
+                       return (ENOENT);
+
+               vap->va_uid = cp->c_uid;
+               vap->va_gid = cp->c_gid;
+               vap->va_mode = cp->c_mode;
+               vap->va_flags = cp->c_bsdflags;
+               vap->va_supported |= VNODE_ATTR_AUTH & ~VNODE_ATTR_va_acl;
+
+               if ((cp->c_attr.ca_recflags & kHFSHasSecurityMask) == 0) {
+                       vap->va_acl = (kauth_acl_t) KAUTH_FILESEC_NONE;
+                       VATTR_SET_SUPPORTED(vap, va_acl);
+               }
+       
+               return (0);
+       }
+
+       hfsmp = VTOHFS(vp);
+       v_type = vnode_vtype(vp);
+
+       if (VATTR_IS_ACTIVE(vap, va_document_id)) {
+               uint32_t document_id;
+
+               if (cp->c_desc.cd_cnid == kHFSRootFolderID)
+                       document_id = kHFSRootFolderID;
+               else {
+                       /*
+                        * This is safe without a lock because we're just reading
+                        * a 32 bit aligned integer which should be atomic on all
+                        * platforms we support.
+                        */
+                       document_id = hfs_get_document_id(cp);
+
+                       if (!document_id && hfs_should_generate_document_id(hfsmp, cp)) {
+                               uint32_t new_document_id;
+
+                               error = hfs_generate_document_id(hfsmp, &new_document_id);
+                               if (error)
+                                       return error;
+
+                               error = hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
+                               if (error)
+                                       return error;
+
+                               bool want_docid_fsevent = false;
+
+                               // Need to check again now that we have the lock
+                               document_id = hfs_get_document_id(cp);
+                               if (!document_id && hfs_should_generate_document_id(hfsmp, cp)) {
+                                       cp->c_attr.ca_finderextendeddirinfo.document_id = document_id = new_document_id;
+                                       want_docid_fsevent = true;
+                                       SET(cp->c_flag, C_MODIFIED);
+                               }
+
+                               hfs_unlock(cp);
+
+                               if (want_docid_fsevent) {
+#if CONFIG_FSE
+                                       add_fsevent(FSE_DOCID_CHANGED, ap->a_context,
+                                                               FSE_ARG_DEV,   hfsmp->hfs_raw_dev,
+                                                               FSE_ARG_INO,   (ino64_t)0,             // src inode #
+                                                               FSE_ARG_INO,   (ino64_t)cp->c_fileid,  // dst inode #
+                                                               FSE_ARG_INT32, document_id,
+                                                               FSE_ARG_DONE);
+
+                                       if (need_fsevent(FSE_STAT_CHANGED, vp)) {
+                                               add_fsevent(FSE_STAT_CHANGED, ap->a_context, 
+                                                                       FSE_ARG_VNODE, vp, FSE_ARG_DONE);
+                                       }
+#endif
+                               }
+                       }
+               }
+
+               vap->va_document_id = document_id;
+               VATTR_SET_SUPPORTED(vap, va_document_id);
+       }
+
+       /*
+        * If time attributes are requested and we have cnode times
+        * that require updating, then acquire an exclusive lock on
+        * the cnode before updating the times.  Otherwise we can
+        * just acquire a shared lock.
+        */
+       if ((vap->va_active & VNODE_ATTR_TIMES) &&
+           (cp->c_touch_acctime || cp->c_touch_chgtime || cp->c_touch_modtime)) {
+               if ((error = hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT)))
                        return (error);
+               hfs_touchtimes(hfsmp, cp);
+
+               // downgrade to a shared lock since that's all we need from here on out
+               cp->c_lockowner = HFS_SHARED_OWNER;
+               lck_rw_lock_exclusive_to_shared(&cp->c_rwlock);
+
+       } else if ((error = hfs_lock(cp, HFS_SHARED_LOCK, HFS_LOCK_DEFAULT))) {
+               return (error);
        }
-       if (vap->va_size != VNOVAL) {
+
+       if (v_type == VDIR) {
+               data_size = (cp->c_entries + 2) * AVERAGE_HFSDIRENTRY_SIZE;
+
+               if (VATTR_IS_ACTIVE(vap, va_nlink)) {
+                       int nlink;
+       
+                       /*
+                        * For directories, the va_nlink is esentially a count
+                        * of the ".." references to a directory plus the "."
+                        * reference and the directory itself. So for HFS+ this
+                        * becomes the sub-directory count plus two.
+                        *
+                        * In the absence of a sub-directory count we use the
+                        * directory's item count.  This will be too high in
+                        * most cases since it also includes files.
+                        */
+                       if ((hfsmp->hfs_flags & HFS_FOLDERCOUNT) && 
+                           (cp->c_attr.ca_recflags & kHFSHasFolderCountMask))
+                               nlink = cp->c_attr.ca_dircount;  /* implied ".." entries */
+                       else
+                               nlink = cp->c_entries;
+
+                       /* Account for ourself and our "." entry */
+                       nlink += 2;  
+                        /* Hide our private directories. */
+                       if (cp->c_cnid == kHFSRootFolderID) {
+                               if (hfsmp->hfs_private_desc[FILE_HARDLINKS].cd_cnid != 0) {
+                                       --nlink;    
+                               }
+                               if (hfsmp->hfs_private_desc[DIR_HARDLINKS].cd_cnid != 0) {
+                                       --nlink;
+                               }
+                       }
+                       VATTR_RETURN(vap, va_nlink, (u_int64_t)nlink);
+               }               
+               if (VATTR_IS_ACTIVE(vap, va_nchildren)) {
+                       int entries;
+       
+                       entries = cp->c_entries;
+                       /* Hide our private files and directories. */
+                       if (cp->c_cnid == kHFSRootFolderID) {
+                               if (hfsmp->hfs_private_desc[FILE_HARDLINKS].cd_cnid != 0)
+                                       --entries;
+                               if (hfsmp->hfs_private_desc[DIR_HARDLINKS].cd_cnid != 0)
+                                       --entries;
+                               if (hfsmp->jnl || ((hfsmp->vcbAtrb & kHFSVolumeJournaledMask) && (hfsmp->hfs_flags & HFS_READ_ONLY)))
+                                       entries -= 2;   /* hide the journal files */
+                       }
+                       VATTR_RETURN(vap, va_nchildren, entries);
+               }
                /*
-                * Disallow write attempts on read-only file systems;
-                * unless the file is a socket, fifo, or a block or
-                * character device resident on the file system.
+                * The va_dirlinkcount is the count of real directory hard links.
+                * (i.e. its not the sum of the implied "." and ".." references)
+                */
+               if (VATTR_IS_ACTIVE(vap, va_dirlinkcount)) {
+                       VATTR_RETURN(vap, va_dirlinkcount, (uint32_t)cp->c_linkcount);
+               }
+       } else /* !VDIR */ {
+               data_size = VCTOF(vp, cp)->ff_size;
+
+               VATTR_RETURN(vap, va_nlink, (u_int64_t)cp->c_linkcount);
+               if (VATTR_IS_ACTIVE(vap, va_data_alloc)) {
+                       u_int64_t blocks;
+       
+#if HFS_COMPRESSION
+                       if (hide_size) {
+                               VATTR_RETURN(vap, va_data_alloc, 0);
+                       } else if (compressed) {
+                               /* for compressed files, we report all allocated blocks as belonging to the data fork */
+                               blocks = cp->c_blocks;
+                               VATTR_RETURN(vap, va_data_alloc, blocks * (u_int64_t)hfsmp->blockSize);
+                       }
+                       else
+#endif
+                       {
+                               blocks = VCTOF(vp, cp)->ff_blocks;
+                               VATTR_RETURN(vap, va_data_alloc, blocks * (u_int64_t)hfsmp->blockSize);
+                       }
+               }
+       }
+
+       /* conditional because 64-bit arithmetic can be expensive */
+       if (VATTR_IS_ACTIVE(vap, va_total_size)) {
+               if (v_type == VDIR) {
+                       VATTR_RETURN(vap, va_total_size, (cp->c_entries + 2) * AVERAGE_HFSDIRENTRY_SIZE);
+               } else {
+                       u_int64_t total_size = ~0ULL;
+                       struct cnode *rcp;
+#if HFS_COMPRESSION
+                       if (hide_size) {
+                               /* we're hiding the size of this file, so just return 0 */
+                               total_size = 0;
+                       } else if (compressed) {
+                               if (uncompressed_size == -1) {
+                                       /*
+                                        * We failed to get the uncompressed size above,
+                                        * so we'll fall back to the standard path below
+                                        * since total_size is still -1
+                                        */
+                               } else {
+                                       /* use the uncompressed size we fetched above */
+                                       total_size = uncompressed_size;
+                               }
+                       }
+#endif
+                       if (total_size == ~0ULL) {
+                               if (cp->c_datafork) {
+                                       total_size = cp->c_datafork->ff_size;
+                               }
+                               
+                               if (cp->c_blocks - VTOF(vp)->ff_blocks) {
+                                       /* We deal with rsrc fork vnode iocount at the end of the function */
+                                       error = hfs_vgetrsrc(hfsmp, vp, &rvp);
+                                       if (error) {
+                                               /*
+                                                * Note that we call hfs_vgetrsrc with error_on_unlinked
+                                                * set to FALSE.  This is because we may be invoked via
+                                                * fstat() on an open-unlinked file descriptor and we must 
+                                                * continue to support access to the rsrc fork until it disappears.
+                                                * The code at the end of this function will be
+                                                * responsible for releasing the iocount generated by 
+                                                * hfs_vgetrsrc.  This is because we can't drop the iocount
+                                                * without unlocking the cnode first.
+                                                */
+                                               goto out;
+                                       }
+                                       
+                                       rcp = VTOC(rvp);
+                                       if (rcp && rcp->c_rsrcfork) {
+                                               total_size += rcp->c_rsrcfork->ff_size;
+                                       }
+                               }
+                       }
+                       
+                       VATTR_RETURN(vap, va_total_size, total_size);
+               }
+       }
+       if (VATTR_IS_ACTIVE(vap, va_total_alloc)) {
+               if (v_type == VDIR) {
+                       VATTR_RETURN(vap, va_total_alloc, 0);
+               } else {
+                       VATTR_RETURN(vap, va_total_alloc, (u_int64_t)cp->c_blocks * (u_int64_t)hfsmp->blockSize);
+               }
+       }
+
+       /*
+        * If the VFS wants extended security data, and we know that we
+        * don't have any (because it never told us it was setting any)
+        * then we can return the supported bit and no data.  If we do
+        * have extended security, we can just leave the bit alone and
+        * the VFS will use the fallback path to fetch it.
+        */
+       if (VATTR_IS_ACTIVE(vap, va_acl)) {
+               if ((cp->c_attr.ca_recflags & kHFSHasSecurityMask) == 0) {
+                       vap->va_acl = (kauth_acl_t) KAUTH_FILESEC_NONE;
+                       VATTR_SET_SUPPORTED(vap, va_acl);
+               }
+       }
+
+       vap->va_access_time.tv_sec = cp->c_atime;
+       vap->va_access_time.tv_nsec = 0;
+       vap->va_create_time.tv_sec = cp->c_itime;
+       vap->va_create_time.tv_nsec = 0;
+       vap->va_modify_time.tv_sec = cp->c_mtime;
+       vap->va_modify_time.tv_nsec = 0;
+       vap->va_change_time.tv_sec = cp->c_ctime;
+       vap->va_change_time.tv_nsec = 0;
+       vap->va_backup_time.tv_sec = cp->c_btime;
+       vap->va_backup_time.tv_nsec = 0;        
+
+       /* See if we need to emit the date added field to the user */
+       if (VATTR_IS_ACTIVE(vap, va_addedtime)) {
+               u_int32_t dateadded = hfs_get_dateadded (cp);
+               if (dateadded) {
+                       vap->va_addedtime.tv_sec = dateadded;
+                       vap->va_addedtime.tv_nsec = 0;
+                       VATTR_SET_SUPPORTED (vap, va_addedtime);
+               }
+       }
+
+       /* XXX is this really a good 'optimal I/O size'? */
+       vap->va_iosize = hfsmp->hfs_logBlockSize;
+       vap->va_uid = cp->c_uid;
+       vap->va_gid = cp->c_gid;
+       vap->va_mode = cp->c_mode;
+       vap->va_flags = cp->c_bsdflags;
+
+       /*
+        * Exporting file IDs from HFS Plus:
+        *
+        * For "normal" files the c_fileid is the same value as the
+        * c_cnid.  But for hard link files, they are different - the
+        * c_cnid belongs to the active directory entry (ie the link)
+        * and the c_fileid is for the actual inode (ie the data file).
+        *
+        * The stat call (getattr) uses va_fileid and the Carbon APIs,
+        * which are hardlink-ignorant, will ask for va_linkid.
+        */
+       vap->va_fileid = (u_int64_t)cp->c_fileid;
+       /* 
+        * We need to use the origin cache for both hardlinked files 
+        * and directories. Hardlinked directories have multiple cnids 
+        * and parents (one per link). Hardlinked files also have their 
+        * own parents and link IDs separate from the indirect inode number. 
+        * If we don't use the cache, we could end up vending the wrong ID 
+        * because the cnode will only reflect the link that was looked up most recently.
+        */
+       if (cp->c_flag & C_HARDLINK) {
+               vap->va_linkid = (u_int64_t)hfs_currentcnid(cp);
+               vap->va_parentid = (u_int64_t)hfs_currentparent(cp, /* have_lock: */ true);
+       } else {
+               vap->va_linkid = (u_int64_t)cp->c_cnid;
+               vap->va_parentid = (u_int64_t)cp->c_parentcnid;
+       }
+       vap->va_fsid = hfsmp->hfs_raw_dev;
+       vap->va_filerev = 0;
+       vap->va_encoding = cp->c_encoding;
+       vap->va_rdev = (v_type == VBLK || v_type == VCHR) ? cp->c_rdev : 0;
+#if HFS_COMPRESSION
+       if (VATTR_IS_ACTIVE(vap, va_data_size)) {
+               if (hide_size)
+                       vap->va_data_size = 0;
+               else if (compressed) {
+                       if (uncompressed_size == -1) {
+                               /* failed to get the uncompressed size above, so just return data_size */
+                               vap->va_data_size = data_size;
+                       } else {
+                               /* use the uncompressed size we fetched above */
+                               vap->va_data_size = uncompressed_size;
+                       }
+               } else
+                       vap->va_data_size = data_size;
+               VATTR_SET_SUPPORTED(vap, va_data_size);
+       }
+#else
+       vap->va_data_size = data_size;
+       vap->va_supported |= VNODE_ATTR_va_data_size;
+#endif
+
+#if CONFIG_PROTECT
+       if (VATTR_IS_ACTIVE(vap, va_dataprotect_class)) {
+               vap->va_dataprotect_class = cp->c_cpentry ? CP_CLASS(cp->c_cpentry->cp_pclass) : 0;
+               VATTR_SET_SUPPORTED(vap, va_dataprotect_class);
+       }       
+#endif
+       if (VATTR_IS_ACTIVE(vap, va_write_gencount)) {
+               if (ubc_is_mapped_writable(vp)) {
+                       /*
+                        * Return 0 to the caller to indicate the file may be
+                        * changing.  There is no need for us to increment the
+                        * generation counter here because it gets done as part of
+                        * page-out and also when the file is unmapped (to account
+                        * for changes we might not have seen).
+                        */
+                       vap->va_write_gencount = 0;
+               } else {
+                       vap->va_write_gencount = hfs_get_gencount(cp);
+               }
+
+               VATTR_SET_SUPPORTED(vap, va_write_gencount);
+       }
+
+       /* Mark them all at once instead of individual VATTR_SET_SUPPORTED calls. */
+       vap->va_supported |= VNODE_ATTR_va_access_time |
+                                                VNODE_ATTR_va_create_time | VNODE_ATTR_va_modify_time |
+                            VNODE_ATTR_va_change_time| VNODE_ATTR_va_backup_time |
+                            VNODE_ATTR_va_iosize | VNODE_ATTR_va_uid |
+                            VNODE_ATTR_va_gid | VNODE_ATTR_va_mode |
+                            VNODE_ATTR_va_flags |VNODE_ATTR_va_fileid |
+                            VNODE_ATTR_va_linkid | VNODE_ATTR_va_parentid |
+                            VNODE_ATTR_va_fsid | VNODE_ATTR_va_filerev |
+                            VNODE_ATTR_va_encoding | VNODE_ATTR_va_rdev;
+
+       /* If this is the root, let VFS to find out the mount name, which 
+        * may be different from the real name.  Otherwise, we need to take care
+        * for hardlinked files, which need to be looked up, if necessary 
+        */
+       if (VATTR_IS_ACTIVE(vap, va_name) && (cp->c_cnid != kHFSRootFolderID)) {
+               struct cat_desc linkdesc;
+               int lockflags;
+               int uselinkdesc = 0;
+               cnid_t nextlinkid = 0;
+               cnid_t prevlinkid = 0;
+
+               /* Get the name for ATTR_CMN_NAME.  We need to take special care for hardlinks      
+                * here because the info. for the link ID requested by getattrlist may be
+                * different than what's currently in the cnode.  This is because the cnode     
+                * will be filled in with the information for the most recent link ID that went
+                * through namei/lookup().  If there are competing lookups for hardlinks that point 
+                * to the same inode, one (or more) getattrlists could be vended incorrect name information.
+                * Also, we need to beware of open-unlinked files which could have a namelen of 0.     
                 */
-               switch (vp->v_type) {
-               case VDIR:
-                       return (EISDIR);
-               case VLNK:
-               case VREG:
-                       if (VTOHFS(vp)->hfs_flags & HFS_READ_ONLY)
-                               return (EROFS);
-                       break;
-               default:
-                       break;
-               }
-               if ((error = VOP_TRUNCATE(vp, vap->va_size, 0, cred, p)))
+
+               if ((cp->c_flag & C_HARDLINK) && 
+                               ((cp->c_desc.cd_namelen == 0) || (vap->va_linkid != cp->c_cnid))) {
+                       /* 
+                        * If we have no name and our link ID is the raw inode number, then we may
+                        * have an open-unlinked file.  Go to the next link in this case.
+                        */
+                       if ((cp->c_desc.cd_namelen == 0) && (vap->va_linkid == cp->c_fileid)) {
+                               if ((error = hfs_lookup_siblinglinks(hfsmp, vap->va_linkid, &prevlinkid, &nextlinkid))){
+                                       goto out;
+                               }
+                       }       
+                       else {
+                               /* just use link obtained from vap above */
+                               nextlinkid = vap->va_linkid;
+                       }
+
+                       /* We need to probe the catalog for the descriptor corresponding to the link ID
+                        * stored in nextlinkid.  Note that we don't know if we have the exclusive lock
+                        * for the cnode here, so we can't just update the descriptor.  Instead,
+                        * we should just store the descriptor's value locally and then use it to pass
+                        * out the name value as needed below. 
+                        */ 
+                       if (nextlinkid){
+                               lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG, HFS_SHARED_LOCK);
+                               error = cat_findname(hfsmp, nextlinkid, &linkdesc);
+                               hfs_systemfile_unlock(hfsmp, lockflags);        
+                               if (error == 0) {
+                                       uselinkdesc = 1;
+                               }
+                       }
+               }
+
+               /* By this point, we've either patched up the name above and the c_desc
+                * points to the correct data, or it already did, in which case we just proceed
+                * by copying the name into the vap.  Note that we will never set va_name to
+                * supported if nextlinkid is never initialized.  This could happen in the degenerate
+                * case above involving the raw inode number, where it has no nextlinkid.  In this case
+                * we will simply not mark the name bit as supported.
+                */
+               if (uselinkdesc) {
+                       strlcpy(vap->va_name, (const char*) linkdesc.cd_nameptr, MAXPATHLEN);
+                       VATTR_SET_SUPPORTED(vap, va_name);
+                       cat_releasedesc(&linkdesc);
+               }       
+               else if (cp->c_desc.cd_namelen) {
+                       strlcpy(vap->va_name, (const char*) cp->c_desc.cd_nameptr, MAXPATHLEN);
+                       VATTR_SET_SUPPORTED(vap, va_name);
+               }
+       }
+
+out:
+       hfs_unlock(cp);
+       /*
+        * We need to vnode_put the rsrc fork vnode only *after* we've released
+        * the cnode lock, since vnode_put can trigger an inactive call, which 
+        * will go back into HFS and try to acquire a cnode lock.
+        */
+       if (rvp) {
+               vnode_put (rvp);
+       }
+
+       return (error);
+}
+
+int
+hfs_vnop_setattr(ap)
+       struct vnop_setattr_args /* {
+               struct vnode *a_vp;
+               struct vnode_attr *a_vap;
+               vfs_context_t a_context;
+       } */ *ap;
+{
+       struct vnode_attr *vap = ap->a_vap;
+       struct vnode *vp = ap->a_vp;
+       struct cnode *cp = NULL;
+       struct hfsmount *hfsmp;
+       kauth_cred_t cred = vfs_context_ucred(ap->a_context);
+       struct proc *p = vfs_context_proc(ap->a_context);
+       int error = 0;
+       uid_t nuid;
+       gid_t ngid;
+       time_t orig_ctime;
+
+       orig_ctime = VTOC(vp)->c_ctime;
+       
+#if HFS_COMPRESSION
+       int decmpfs_reset_state = 0;
+       /*
+        we call decmpfs_update_attributes even if the file is not compressed
+        because we want to update the incoming flags if the xattrs are invalid
+        */
+       error = decmpfs_update_attributes(vp, vap);
+       if (error)
+               return error;
+#endif
+       //
+       // if this is not a size-changing setattr and it is not just
+       // an atime update, then check for a snapshot.
+       //
+       if (!VATTR_IS_ACTIVE(vap, va_data_size) && !(vap->va_active == VNODE_ATTR_va_access_time)) {
+               check_for_tracked_file(vp, orig_ctime, NAMESPACE_HANDLER_METADATA_MOD, NSPACE_REARM_NO_ARG);
+       }
+
+#if CONFIG_PROTECT
+       /*
+        * All metadata changes should be allowed except a size-changing setattr, which
+        * has effects on file content and requires calling into cp_handle_vnop
+        * to have content protection check.
+        */
+       if (VATTR_IS_ACTIVE(vap, va_data_size)) {
+               if ((error = cp_handle_vnop(vp, CP_WRITE_ACCESS, 0)) != 0) {
                        return (error);
+               }
        }
-       cp = VTOC(vp);
-       if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
-               if (VTOHFS(vp)->hfs_flags & HFS_READ_ONLY)
-                       return (EROFS);
-               if (((error = hfs_owner_rights(VTOHFS(vp), cp->c_uid, cred, p, true)) != 0) &&
-                   ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
-                   (error = VOP_ACCESS(vp, VWRITE, cred, p)))) {
+#endif /* CONFIG_PROTECT */
+
+       hfsmp = VTOHFS(vp);
+
+       /* Don't allow modification of the journal. */
+       if (hfs_is_journal_file(hfsmp, VTOC(vp))) {
+               return (EPERM);
+       }
+
+       //
+       // Check if we'll need a document_id and if so, get it before we lock the
+       // the cnode to avoid any possible deadlock with the root vnode which has
+       // to get locked to get the document id
+       //
+       u_int32_t document_id=0;
+       if (VATTR_IS_ACTIVE(vap, va_flags) && (vap->va_flags & UF_TRACKED) && !(VTOC(vp)->c_bsdflags & UF_TRACKED)) {
+               struct FndrExtendedDirInfo *fip = (struct FndrExtendedDirInfo *)((char *)&(VTOC(vp)->c_attr.ca_finderinfo) + 16);
+               //
+               // If the document_id is not set, get a new one.  It will be set
+               // on the file down below once we hold the cnode lock.
+               //
+               if (fip->document_id == 0) {
+                       if (hfs_generate_document_id(hfsmp, &document_id) != 0) {
+                               document_id = 0;
+                       }
+               }
+       }
+
+
+       /*
+        * File size change request.
+        * We are guaranteed that this is not a directory, and that
+        * the filesystem object is writeable.
+        *
+        * NOTE: HFS COMPRESSION depends on the data_size being set *before* the bsd flags are updated
+        */
+       VATTR_SET_SUPPORTED(vap, va_data_size);
+       if (VATTR_IS_ACTIVE(vap, va_data_size) && !vnode_islnk(vp)) {
+#if HFS_COMPRESSION
+               /* keep the compressed state locked until we're done truncating the file */
+               decmpfs_cnode *dp = VTOCMP(vp);
+               if (!dp) {
+                       /*
+                        * call hfs_lazy_init_decmpfs_cnode() to make sure that the decmpfs_cnode
+                        * is filled in; we need a decmpfs_cnode to lock out decmpfs state changes
+                        * on this file while it's truncating
+                        */
+                       dp = hfs_lazy_init_decmpfs_cnode(VTOC(vp));
+                       if (!dp) {
+                               /* failed to allocate a decmpfs_cnode */
+                               return ENOMEM; /* what should this be? */
+                       }
+               }
+               
+               check_for_tracked_file(vp, orig_ctime, vap->va_data_size == 0 ? NAMESPACE_HANDLER_TRUNCATE_OP|NAMESPACE_HANDLER_DELETE_OP : NAMESPACE_HANDLER_TRUNCATE_OP, NULL);
+
+               decmpfs_lock_compressed_data(dp, 1);
+               if (hfs_file_is_compressed(VTOC(vp), 1)) {
+                       error = decmpfs_decompress_file(vp, dp, -1/*vap->va_data_size*/, 0, 1);
+                       if (error != 0) {
+                               decmpfs_unlock_compressed_data(dp, 1);
+                               return error;
+                       }
+               }
+#endif
+
+               // Take truncate lock
+               hfs_lock_truncate(VTOC(vp), HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
+
+               // hfs_truncate will deal with the cnode lock
+               error = hfs_truncate(vp, vap->va_data_size, vap->va_vaflags & 0xffff, 
+                                                        0, ap->a_context);
+
+               hfs_unlock_truncate(VTOC(vp), HFS_LOCK_DEFAULT);
+#if HFS_COMPRESSION
+               decmpfs_unlock_compressed_data(dp, 1);
+#endif
+               if (error)
+                       return error;
+       }
+       if (cp == NULL) {
+               if ((error = hfs_lock(VTOC(vp), HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT)))
                        return (error);
+               cp = VTOC(vp);
+       }
+
+       /*
+        * If it is just an access time update request by itself
+        * we know the request is from kernel level code, and we
+        * can delay it without being as worried about consistency.
+        * This change speeds up mmaps, in the rare case that they
+        * get caught behind a sync.
+        */
+
+       if (vap->va_active == VNODE_ATTR_va_access_time) {
+               cp->c_touch_acctime=TRUE;
+               goto out;
+       }
+
+
+
+       /*
+        * Owner/group change request.
+        * We are guaranteed that the new owner/group is valid and legal.
+        */
+       VATTR_SET_SUPPORTED(vap, va_uid);
+       VATTR_SET_SUPPORTED(vap, va_gid);
+       nuid = VATTR_IS_ACTIVE(vap, va_uid) ? vap->va_uid : (uid_t)VNOVAL;
+       ngid = VATTR_IS_ACTIVE(vap, va_gid) ? vap->va_gid : (gid_t)VNOVAL;
+       if (((nuid != (uid_t)VNOVAL) || (ngid != (gid_t)VNOVAL)) &&
+           ((error = hfs_chown(vp, nuid, ngid, cred, p)) != 0))
+               goto out;
+
+       /*
+        * Mode change request.
+        * We are guaranteed that the mode value is valid and that in
+        * conjunction with the owner and group, this change is legal.
+       */
+       VATTR_SET_SUPPORTED(vap, va_mode);
+       if (VATTR_IS_ACTIVE(vap, va_mode) &&
+           ((error = hfs_chmod(vp, (int)vap->va_mode, cred, p)) != 0))
+           goto out;
+
+       /*
+        * File flags change.
+        * We are guaranteed that only flags allowed to change given the
+        * current securelevel are being changed.
+        */
+       VATTR_SET_SUPPORTED(vap, va_flags);
+       if (VATTR_IS_ACTIVE(vap, va_flags)) {
+               u_int16_t *fdFlags;
+
+#if HFS_COMPRESSION
+               if ((cp->c_bsdflags ^ vap->va_flags) & UF_COMPRESSED) {
+                       /*
+                        * the UF_COMPRESSED was toggled, so reset our cached compressed state
+                        * but we don't want to actually do the update until we've released the cnode lock down below
+                        * NOTE: turning the flag off doesn't actually decompress the file, so that we can
+                        * turn off the flag and look at the "raw" file for debugging purposes
+                        */
+                       decmpfs_reset_state = 1;
+               }
+#endif
+               if ((vap->va_flags & UF_TRACKED) && !(cp->c_bsdflags & UF_TRACKED)) {
+                       struct FndrExtendedDirInfo *fip = (struct FndrExtendedDirInfo *)((char *)&cp->c_attr.ca_finderinfo + 16);
+
+                       //
+                       // we're marking this item UF_TRACKED.  if the document_id is
+                       // not set, get a new one and put it on the file.
+                       //
+                       if (fip->document_id == 0) {
+                               if (document_id != 0) {
+                                       // printf("SETATTR: assigning doc-id %d to %s (ino %d)\n", document_id, vp->v_name, cp->c_desc.cd_cnid);
+                                       fip->document_id = (uint32_t)document_id;
+#if CONFIG_FSE
+                                       add_fsevent(FSE_DOCID_CHANGED, ap->a_context,
+                                                   FSE_ARG_DEV, hfsmp->hfs_raw_dev,
+                                                   FSE_ARG_INO, (ino64_t)0,                // src inode #
+                                                   FSE_ARG_INO, (ino64_t)cp->c_fileid,     // dst inode #
+                                                   FSE_ARG_INT32, document_id,
+                                                   FSE_ARG_DONE);
+#endif
+                               } else {
+                                       // printf("hfs: could not acquire a new document_id for %s (ino %d)\n", vp->v_name, cp->c_desc.cd_cnid);
+                               }
+                       }
+
+               } else if (!(vap->va_flags & UF_TRACKED) && (cp->c_bsdflags & UF_TRACKED)) {
+                       //
+                       // UF_TRACKED is being cleared so clear the document_id
+                       //
+                       struct FndrExtendedDirInfo *fip = (struct FndrExtendedDirInfo *)((char *)&cp->c_attr.ca_finderinfo + 16);
+                       if (fip->document_id) {
+                               // printf("SETATTR: clearing doc-id %d from %s (ino %d)\n", fip->document_id, vp->v_name, cp->c_desc.cd_cnid);
+#if CONFIG_FSE
+                               add_fsevent(FSE_DOCID_CHANGED, ap->a_context,
+                                           FSE_ARG_DEV, hfsmp->hfs_raw_dev,
+                                           FSE_ARG_INO, (ino64_t)cp->c_fileid,          // src inode #
+                                           FSE_ARG_INO, (ino64_t)0,                     // dst inode #
+                                           FSE_ARG_INT32, fip->document_id,             // document id
+                                           FSE_ARG_DONE);
+#endif
+                               fip->document_id = 0;
+                               cp->c_bsdflags &= ~UF_TRACKED;
+                       }
+               }
+
+               cp->c_bsdflags = vap->va_flags;
+               cp->c_flag |= C_MODIFIED;
+               cp->c_touch_chgtime = TRUE;
+
+               
+               /*
+                * Mirror the UF_HIDDEN flag to the invisible bit of the Finder Info.
+                *
+                * The fdFlags for files and frFlags for folders are both 8 bytes
+                * into the userInfo (the first 16 bytes of the Finder Info).  They
+                * are both 16-bit fields.
+                */
+               fdFlags = (u_int16_t *) &cp->c_finderinfo[8];
+               if (vap->va_flags & UF_HIDDEN)
+                       *fdFlags |= OSSwapHostToBigConstInt16(kFinderInvisibleMask);
+               else
+                       *fdFlags &= ~OSSwapHostToBigConstInt16(kFinderInvisibleMask);
+       }
+
+       /*
+        * Timestamp updates.
+        */
+       VATTR_SET_SUPPORTED(vap, va_create_time);
+       VATTR_SET_SUPPORTED(vap, va_access_time);
+       VATTR_SET_SUPPORTED(vap, va_modify_time);
+       VATTR_SET_SUPPORTED(vap, va_backup_time);
+       VATTR_SET_SUPPORTED(vap, va_change_time);
+       if (VATTR_IS_ACTIVE(vap, va_create_time) ||
+           VATTR_IS_ACTIVE(vap, va_access_time) ||
+           VATTR_IS_ACTIVE(vap, va_modify_time) ||
+           VATTR_IS_ACTIVE(vap, va_backup_time)) {
+               if (VATTR_IS_ACTIVE(vap, va_create_time))
+                       cp->c_itime = vap->va_create_time.tv_sec;
+               if (VATTR_IS_ACTIVE(vap, va_access_time)) {
+                       cp->c_atime = vap->va_access_time.tv_sec;
+                       cp->c_touch_acctime = FALSE;
                }
-               if (vap->va_atime.tv_sec != VNOVAL)
-                       cp->c_flag |= C_ACCESS;
-               if (vap->va_mtime.tv_sec != VNOVAL) {
-                       cp->c_flag |= C_CHANGE | C_UPDATE;
+               if (VATTR_IS_ACTIVE(vap, va_modify_time)) {
+                       cp->c_mtime = vap->va_modify_time.tv_sec;
+                       cp->c_touch_modtime = FALSE;
+                       cp->c_touch_chgtime = TRUE;
+
+                       hfs_clear_might_be_dirty_flag(cp);
+
                        /*
                         * The utimes system call can reset the modification
                         * time but it doesn't know about HFS create times.
-                        * So we need to insure that the creation time is
+                        * So we need to ensure that the creation time is
                         * always at least as old as the modification time.
                         */
                        if ((VTOVCB(vp)->vcbSigWord == kHFSPlusSigWord) &&
-                           (cp->c_cnid != kRootDirID) &&
-                           (vap->va_mtime.tv_sec < cp->c_itime)) {
-                               cp->c_itime = vap->va_mtime.tv_sec;
+                           (cp->c_cnid != kHFSRootFolderID) &&
+                               !VATTR_IS_ACTIVE(vap, va_create_time) &&
+                           (cp->c_mtime < cp->c_itime)) {
+                               cp->c_itime = cp->c_mtime;
                        }
                }
-               atimeval.tv_sec = vap->va_atime.tv_sec;
-               atimeval.tv_usec = 0;
-               mtimeval.tv_sec = vap->va_mtime.tv_sec;
-               mtimeval.tv_usec = 0;
-               if ((error = VOP_UPDATE(vp, &atimeval, &mtimeval, 1)))
-                       return (error);
+               if (VATTR_IS_ACTIVE(vap, va_backup_time))
+                       cp->c_btime = vap->va_backup_time.tv_sec;
+               cp->c_flag |= C_MINOR_MOD;
        }
-       error = 0;
-       if (vap->va_mode != (mode_t)VNOVAL) {
-               if (VTOHFS(vp)->hfs_flags & HFS_READ_ONLY)
-                       return (EROFS);
-               error = hfs_chmod(vp, (int)vap->va_mode, cred, p);
+       
+       /*
+        * Set name encoding.
+        */
+       VATTR_SET_SUPPORTED(vap, va_encoding);
+       if (VATTR_IS_ACTIVE(vap, va_encoding)) {
+               cp->c_encoding = vap->va_encoding;
+               cp->c_flag |= C_MODIFIED;
+               hfs_setencodingbits(hfsmp, cp->c_encoding);
+       }
+
+       if ((error = hfs_update(vp, 0)) != 0)
+               goto out;
+out:
+       if (cp) {
+               /* Purge origin cache for cnode, since caller now has correct link ID for it 
+                * We purge it here since it was acquired for us during lookup, and we no longer need it.
+                */
+               if ((cp->c_flag & C_HARDLINK) && (vp->v_type != VDIR)){
+                       hfs_relorigin(cp, 0);
+               }
+
+               hfs_unlock(cp);
+#if HFS_COMPRESSION
+               if (decmpfs_reset_state) {
+                       /*
+                        * we've changed the UF_COMPRESSED flag, so reset the decmpfs state for this cnode
+                        * but don't do it while holding the hfs cnode lock
+                        */
+                       decmpfs_cnode *dp = VTOCMP(vp);
+                       if (!dp) {
+                               /*
+                                * call hfs_lazy_init_decmpfs_cnode() to make sure that the decmpfs_cnode
+                                * is filled in; we need a decmpfs_cnode to prevent decmpfs state changes
+                                * on this file if it's locked
+                                */
+                               dp = hfs_lazy_init_decmpfs_cnode(VTOC(vp));
+                               if (!dp) {
+                                       /* failed to allocate a decmpfs_cnode */
+                                       return ENOMEM; /* what should this be? */
+                               }
+                       }
+                       decmpfs_cnode_set_vnode_state(dp, FILE_TYPE_UNKNOWN, 0);
+               }
+#endif
        }
-       HFS_KNOTE(vp, NOTE_ATTRIB);
        return (error);
 }
 
@@ -726,65 +1768,48 @@ hfs_setattr(ap)
  * Change the mode on a file.
  * cnode must be locked before calling.
  */
-__private_extern__
 int
-hfs_chmod(vp, mode, cred, p)
-       register struct vnode *vp;
-       register int mode;
-       register struct ucred *cred;
-       struct proc *p;
+hfs_chmod(struct vnode *vp, int mode, __unused kauth_cred_t cred, __unused struct proc *p)
 {
        register struct cnode *cp = VTOC(vp);
-       int error;
 
        if (VTOVCB(vp)->vcbSigWord != kHFSPlusSigWord)
                return (0);
 
-       // XXXdbg - don't allow modification of the journal or journal_info_block
-       if (VTOHFS(vp)->jnl && cp && cp->c_datafork) {
-               struct HFSPlusExtentDescriptor *extd;
-
-               extd = &cp->c_datafork->ff_extents[0];
-               if (extd->startBlock == VTOVCB(vp)->vcbJinfoBlock || extd->startBlock == VTOHFS(vp)->jnl_start) {
-                       return EPERM;
-               }
+       // Don't allow modification of the journal or journal_info_block
+       if (hfs_is_journal_file(VTOHFS(vp), cp)) {
+               return EPERM;
        }
 
 #if OVERRIDE_UNKNOWN_PERMISSIONS
-       if (VTOVFS(vp)->mnt_flag & MNT_UNKNOWNPERMISSIONS) {
+       if (((unsigned int)vfs_flags(VTOVFS(vp))) & MNT_UNKNOWNPERMISSIONS) {
                return (0);
        };
 #endif
-       if ((error = hfs_owner_rights(VTOHFS(vp), cp->c_uid, cred, p, true)) != 0)
-               return (error);
-       if (cred->cr_uid) {
-               if (vp->v_type != VDIR && (mode & S_ISTXT))
-                       return (EFTYPE);
-               if (!groupmember(cp->c_gid, cred) && (mode & S_ISGID))
-                       return (EPERM);
-       }
-       cp->c_mode &= ~ALLPERMS;
-       cp->c_mode |= (mode & ALLPERMS);
-       cp->c_flag |= C_CHANGE;
+
+       mode_t new_mode = (cp->c_mode & ~ALLPERMS) | (mode & ALLPERMS);
+       if (new_mode != cp->c_mode) {
+               cp->c_mode = new_mode;
+               cp->c_flag |= C_MINOR_MOD;
+       }
+       cp->c_touch_chgtime = TRUE;
        return (0);
 }
 
 
-__private_extern__
 int
-hfs_write_access(struct vnode *vp, struct ucred *cred, struct proc *p, Boolean considerFlags)
+hfs_write_access(struct vnode *vp, kauth_cred_t cred, struct proc *p, Boolean considerFlags)
 {
        struct cnode *cp = VTOC(vp);
-       gid_t *gp;
        int retval = 0;
-       int i;
+       int is_member;
 
        /*
         * Disallow write attempts on read-only file systems;
         * unless the file is a socket, fifo, or a block or
         * character device resident on the file system.
         */
-       switch (vp->v_type) {
+       switch (vnode_vtype(vp)) {
        case VDIR:
        case VLNK:
        case VREG:
@@ -796,11 +1821,11 @@ hfs_write_access(struct vnode *vp, struct ucred *cred, struct proc *p, Boolean c
        }
  
        /* If immutable bit set, nobody gets to write it. */
-       if (considerFlags && (cp->c_flags & IMMUTABLE))
+       if (considerFlags && (cp->c_bsdflags & IMMUTABLE))
                return (EPERM);
 
        /* Otherwise, user id 0 always gets access. */
-       if (cred->cr_uid == 0)
+       if (!suser(cred, NULL))
                return (0);
 
        /* Otherwise, check the owner. */
@@ -808,9 +1833,8 @@ hfs_write_access(struct vnode *vp, struct ucred *cred, struct proc *p, Boolean c
                return ((cp->c_mode & S_IWUSR) == S_IWUSR ? 0 : EACCES);
  
        /* Otherwise, check the groups. */
-       for (i = 0, gp = cred->cr_groups; i < cred->cr_ngroups; i++, gp++) {
-               if (cp->c_gid == *gp)
-                       return ((cp->c_mode & S_IWGRP) == S_IWGRP ? 0 : EACCES);
+       if (kauth_cred_ismember_gid(cred, cp->c_gid, &is_member) == 0 && is_member) {
+               return ((cp->c_mode & S_IWGRP) == S_IWGRP ? 0 : EACCES);
        }
  
        /* Otherwise, check everyone else. */
@@ -818,118 +1842,85 @@ hfs_write_access(struct vnode *vp, struct ucred *cred, struct proc *p, Boolean c
 }
 
 
-
-/*
- * Change the flags on a file or directory.
- * cnode must be locked before calling.
- */
-__private_extern__
-int
-hfs_chflags(vp, flags, cred, p)
-       register struct vnode *vp;
-       register u_long flags;
-       register struct ucred *cred;
-       struct proc *p;
-{
-       register struct cnode *cp = VTOC(vp);
-       int retval;
-
-       if (VTOVCB(vp)->vcbSigWord == kHFSSigWord) {
-               if ((retval = hfs_write_access(vp, cred, p, false)) != 0) {
-                       return retval;
-               };
-       } else if ((retval = hfs_owner_rights(VTOHFS(vp), cp->c_uid, cred, p, true)) != 0) {
-               return retval;
-       };
-
-       if (cred->cr_uid == 0) {
-               if ((cp->c_flags & (SF_IMMUTABLE | SF_APPEND)) &&
-                       securelevel > 0) {
-                       return EPERM;
-               };
-               cp->c_flags = flags;
-       } else {
-               if (cp->c_flags & (SF_IMMUTABLE | SF_APPEND) ||
-                       (flags & UF_SETTABLE) != flags) {
-                       return EPERM;
-               };
-               cp->c_flags &= SF_SETTABLE;
-               cp->c_flags |= (flags & UF_SETTABLE);
-       }
-       cp->c_flag |= C_CHANGE;
-
-       return (0);
-}
-
-
 /*
  * Perform chown operation on cnode cp;
  * code must be locked prior to call.
  */
-__private_extern__
 int
-hfs_chown(vp, uid, gid, cred, p)
-       register struct vnode *vp;
-       uid_t uid;
-       gid_t gid;
-       struct ucred *cred;
-       struct proc *p;
+#if !QUOTA
+hfs_chown(struct vnode *vp, uid_t uid, gid_t gid, __unused kauth_cred_t cred,
+       __unused struct proc *p)
+#else 
+hfs_chown(struct vnode *vp, uid_t uid, gid_t gid, kauth_cred_t cred,
+       __unused struct proc *p)
+#endif
 {
        register struct cnode *cp = VTOC(vp);
        uid_t ouid;
        gid_t ogid;
-       int error = 0;
 #if QUOTA
+       int error = 0;
        register int i;
        int64_t change;
 #endif /* QUOTA */
 
        if (VTOVCB(vp)->vcbSigWord != kHFSPlusSigWord)
-               return (EOPNOTSUPP);
+               return (ENOTSUP);
 
-       if (VTOVFS(vp)->mnt_flag & MNT_UNKNOWNPERMISSIONS)
+       if (((unsigned int)vfs_flags(VTOVFS(vp))) & MNT_UNKNOWNPERMISSIONS)
                return (0);
        
        if (uid == (uid_t)VNOVAL)
                uid = cp->c_uid;
        if (gid == (gid_t)VNOVAL)
                gid = cp->c_gid;
+
+#if 0  /* we are guaranteed that this is already the case */
        /*
         * If we don't own the file, are trying to change the owner
         * of the file, or are not a member of the target group,
         * the caller must be superuser or the call fails.
         */
-       if ((cred->cr_uid != cp->c_uid || uid != cp->c_uid ||
-           (gid != cp->c_gid && !groupmember((gid_t)gid, cred))) &&
-           (error = suser(cred, &p->p_acflag)))
+       if ((kauth_cred_getuid(cred) != cp->c_uid || uid != cp->c_uid ||
+           (gid != cp->c_gid &&
+            (kauth_cred_ismember_gid(cred, gid, &is_member) || !is_member))) &&
+           (error = suser(cred, 0)))
                return (error);
+#endif
 
        ogid = cp->c_gid;
        ouid = cp->c_uid;
+
+       if (ouid == uid && ogid == gid) {
+               // No change, just set change time
+               cp->c_touch_chgtime = TRUE;
+               return 0;
+       }
+
 #if QUOTA
        if ((error = hfs_getinoquota(cp)))
                return (error);
        if (ouid == uid) {
-               dqrele(vp, cp->c_dquot[USRQUOTA]);
+               dqrele(cp->c_dquot[USRQUOTA]);
                cp->c_dquot[USRQUOTA] = NODQUOT;
        }
        if (ogid == gid) {
-               dqrele(vp, cp->c_dquot[GRPQUOTA]);
+               dqrele(cp->c_dquot[GRPQUOTA]);
                cp->c_dquot[GRPQUOTA] = NODQUOT;
        }
 
        /*
         * Eventually need to account for (fake) a block per directory
-        *if (vp->v_type == VDIR)
-        *change = VTOVCB(vp)->blockSize;
-        *else
+        * if (vnode_isdir(vp))
+        *     change = VTOHFS(vp)->blockSize;
+        * else
         */
 
        change = (int64_t)(cp->c_blocks) * (int64_t)VTOVCB(vp)->blockSize;
        (void) hfs_chkdq(cp, -change, cred, CHOWN);
        (void) hfs_chkiq(cp, -1, cred, CHOWN);
        for (i = 0; i < MAXQUOTAS; i++) {
-               dqrele(vp, cp->c_dquot[i]);
+               dqrele(cp->c_dquot[i]);
                cp->c_dquot[i] = NODQUOT;
        }
 #endif /* QUOTA */
@@ -938,11 +1929,11 @@ hfs_chown(vp, uid, gid, cred, p)
 #if QUOTA
        if ((error = hfs_getinoquota(cp)) == 0) {
                if (ouid == uid) {
-                       dqrele(vp, cp->c_dquot[USRQUOTA]);
+                       dqrele(cp->c_dquot[USRQUOTA]);
                        cp->c_dquot[USRQUOTA] = NODQUOT;
                }
                if (ogid == gid) {
-                       dqrele(vp, cp->c_dquot[GRPQUOTA]);
+                       dqrele(cp->c_dquot[GRPQUOTA]);
                        cp->c_dquot[GRPQUOTA] = NODQUOT;
                }
                if ((error = hfs_chkdq(cp, change, cred, CHOWN)) == 0) {
@@ -952,7 +1943,7 @@ hfs_chown(vp, uid, gid, cred, p)
                                (void) hfs_chkdq(cp, -change, cred, CHOWN|FORCE);
                }
                for (i = 0; i < MAXQUOTAS; i++) {
-                       dqrele(vp, cp->c_dquot[i]);
+                       dqrele(cp->c_dquot[i]);
                        cp->c_dquot[i] = NODQUOT;
                }
        }
@@ -960,11 +1951,11 @@ hfs_chown(vp, uid, gid, cred, p)
        cp->c_uid = ouid;
        if (hfs_getinoquota(cp) == 0) {
                if (ouid == uid) {
-                       dqrele(vp, cp->c_dquot[USRQUOTA]);
+                       dqrele(cp->c_dquot[USRQUOTA]);
                        cp->c_dquot[USRQUOTA] = NODQUOT;
                }
                if (ogid == gid) {
-                       dqrele(vp, cp->c_dquot[GRPQUOTA]);
+                       dqrele(cp->c_dquot[GRPQUOTA]);
                        cp->c_dquot[GRPQUOTA] = NODQUOT;
                }
                (void) hfs_chkdq(cp, change, cred, FORCE|CHOWN);
@@ -977,120 +1968,341 @@ good:
                panic("hfs_chown: lost quota");
 #endif /* QUOTA */
 
-       if (ouid != uid || ogid != gid)
-               cp->c_flag |= C_CHANGE;
-       if (ouid != uid && cred->cr_uid != 0)
-               cp->c_mode &= ~S_ISUID;
-       if (ogid != gid && cred->cr_uid != 0)
-               cp->c_mode &= ~S_ISGID;
+       /*
+        * Without quotas, we could probably make this a minor
+        * modification.
+        */
+       cp->c_flag |= C_MODIFIED;
+
+       /*
+         According to the SUSv3 Standard, chown() shall mark
+         for update the st_ctime field of the file.
+         (No exceptions mentioned)
+       */
+               cp->c_touch_chgtime = TRUE;
        return (0);
 }
 
+#if HFS_COMPRESSION
+/*
+ * Flush the resource fork if it exists.  vp is the data fork and has
+ * an iocount.
+ */
+static int hfs_flush_rsrc(vnode_t vp, vfs_context_t ctx)
+{
+       cnode_t *cp = VTOC(vp);
+
+       hfs_lock(cp, HFS_SHARED_LOCK, 0);
+
+       vnode_t rvp = cp->c_rsrc_vp;
+
+       if (!rvp) {
+               hfs_unlock(cp);
+               return 0;
+       }
+
+       int vid = vnode_vid(rvp);
+
+       hfs_unlock(cp);
+
+       int error = vnode_getwithvid(rvp, vid);
+
+       if (error)
+               return error == ENOENT ? 0 : error;
+
+       hfs_lock_truncate(cp, HFS_EXCLUSIVE_LOCK, 0);
+       hfs_lock_always(cp, HFS_EXCLUSIVE_LOCK);
+       hfs_filedone(rvp, ctx, HFS_FILE_DONE_NO_SYNC);
+       hfs_unlock(cp);
+       hfs_unlock_truncate(cp, 0);
+
+       error = ubc_msync(rvp, 0, ubc_getsize(rvp), NULL,
+                                         UBC_PUSHALL | UBC_SYNC);
+
+       vnode_put(rvp);
+
+       return error;
+}
+#endif // HFS_COMPRESSION
 
 /*
-#
-#% exchange fvp                L L L
-#% exchange tvp                L L L
-#
+ * hfs_vnop_exchange:
+ * 
+ * Inputs: 
+ *             'from' vnode/cnode
+ *             'to' vnode/cnode
+ *             options flag bits
+ *             vfs_context
+ * 
+ * Discussion:
+ * hfs_vnop_exchange is used to service the exchangedata(2) system call.
+ * Per the requirements of that system call, this function "swaps" some
+ * of the information that lives in one catalog record for some that
+ * lives in another.  Note that not everything is swapped; in particular, 
+ * the extent information stored in each cnode is kept local to that 
+ * cnode.  This allows existing file descriptor references to continue
+ * to operate on the same content, regardless of the location in the
+ * namespace that the file may have moved to.  See inline comments
+ * in the function for more information.
  */
- /*
-  * The hfs_exchange routine swaps the fork data in two files by
-  * exchanging some of the information in the cnode.  It is used
-  * to preserve the file ID when updating an existing file, in
-  * case the file is being tracked through its file ID. Typically
-  * its used after creating a new file during a safe-save.
-  */
-  
-static int
-hfs_exchange(ap)
-       struct vop_exchange_args /* {
+int
+hfs_vnop_exchange(ap)
+       struct vnop_exchange_args /* {
                struct vnode *a_fvp;
                struct vnode *a_tvp;
-               struct ucred *a_cred;
-               struct proc *a_p;
+               int a_options;
+               vfs_context_t a_context;
        } */ *ap;
 {
        struct vnode *from_vp = ap->a_fvp;
        struct vnode *to_vp = ap->a_tvp;
-       struct cnode *from_cp = VTOC(from_vp);
-       struct cnode *to_cp = VTOC(to_vp);
-       struct hfsmount *hfsmp = VTOHFS(from_vp);
+       struct cnode *from_cp;
+       struct cnode *to_cp;
+       struct hfsmount *hfsmp;
        struct cat_desc tempdesc;
        struct cat_attr tempattr;
-       int error = 0, started_tr = 0, grabbed_lock = 0;
-       cat_cookie_t cookie = {0};
+       const unsigned char *from_nameptr;
+       const unsigned char *to_nameptr;
+       char from_iname[32];
+       char to_iname[32];
+       uint32_t to_flag_special;
+       uint32_t from_flag_special;
+       cnid_t  from_parid;
+       cnid_t  to_parid;
+       int lockflags;
+       int error = 0, started_tr = 0, got_cookie = 0;
+       cat_cookie_t cookie;
+       time_t orig_from_ctime, orig_to_ctime;
+       bool have_cnode_locks = false, have_from_trunc_lock = false, have_to_trunc_lock = false;
+
+       /*
+        * VFS does the following checks:
+        * 1. Validate that both are files.
+        * 2. Validate that both are on the same mount.
+        * 3. Validate that they're not the same vnode.
+        */
+
+       from_cp = VTOC(from_vp);
+       to_cp = VTOC(to_vp);
+       hfsmp = VTOHFS(from_vp);
+
+       orig_from_ctime = from_cp->c_ctime;
+       orig_to_ctime = to_cp->c_ctime;
+
+#if CONFIG_PROTECT
+       /* 
+        * Do not allow exchangedata/F_MOVEDATAEXTENTS on data-protected filesystems 
+        * because the EAs will not be swapped.  As a result, the persistent keys would not
+        * match and the files will be garbage.
+        */
+       if (cp_fs_protected (vnode_mount(from_vp))) {
+               return EINVAL;
+       }
+#endif
+
+#if HFS_COMPRESSION
+       if (!ISSET(ap->a_options, FSOPT_EXCHANGE_DATA_ONLY)) {
+               if ( hfs_file_is_compressed(from_cp, 0) ) {
+                       if ( 0 != ( error = decmpfs_decompress_file(from_vp, VTOCMP(from_vp), -1, 0, 1) ) ) {
+                               return error;
+                       }
+               }
+
+               if ( hfs_file_is_compressed(to_cp, 0) ) {
+                       if ( 0 != ( error = decmpfs_decompress_file(to_vp, VTOCMP(to_vp), -1, 0, 1) ) ) {
+                               return error;
+                       }
+               }
+       }
+#endif // HFS_COMPRESSION
+
+       // Resource forks cannot be exchanged.
+       if (VNODE_IS_RSRC(from_vp) || VNODE_IS_RSRC(to_vp))
+               return EINVAL;
+
+       /* 
+        * Normally, we want to notify the user handlers about the event,
+        * except if it's a handler driving the event.
+        */
+       if ((ap->a_options & FSOPT_EXCHANGE_DATA_ONLY) == 0) {
+               check_for_tracked_file(from_vp, orig_from_ctime, NAMESPACE_HANDLER_WRITE_OP, NULL);
+               check_for_tracked_file(to_vp, orig_to_ctime, NAMESPACE_HANDLER_WRITE_OP, NULL);
+       } else {
+               /*
+                * This is currently used by mtmd so we should tidy up the
+                * file now because the data won't be used again in the
+                * destination file.
+                */
+               hfs_lock_truncate(from_cp, HFS_EXCLUSIVE_LOCK, 0);
+               hfs_lock_always(from_cp, HFS_EXCLUSIVE_LOCK);
+               hfs_filedone(from_vp, ap->a_context, HFS_FILE_DONE_NO_SYNC);
+               hfs_unlock(from_cp);
+               hfs_unlock_truncate(from_cp, 0);
+
+               // Flush all the data from the source file
+               error = ubc_msync(from_vp, 0, ubc_getsize(from_vp), NULL,
+                                                 UBC_PUSHALL | UBC_SYNC);
+               if (error)
+                       goto exit;
+
+#if HFS_COMPRESSION
+               /*
+                * If this is a compressed file, we need to do the same for
+                * the resource fork.
+                */
+               if (ISSET(from_cp->c_bsdflags, UF_COMPRESSED)) {
+                       error = hfs_flush_rsrc(from_vp, ap->a_context);
+                       if (error)
+                               goto exit;
+               }
+#endif
 
-       /* The files must be on the same volume. */
-       if (from_vp->v_mount != to_vp->v_mount)
-               return (EXDEV);
+               /*
+                * We're doing a data-swap so we need to take the truncate
+                * lock exclusively.  We need an exclusive lock because we
+                * will be completely truncating the source file and we must
+                * make sure nobody else sneaks in and trys to issue I/O
+                * whilst we don't have the cnode lock.
+                *
+                * After taking the truncate lock we do a quick check to
+                * verify there are no other references (including mmap
+                * references), but we must remember that this does not stop
+                * anybody coming in later and taking a reference.  We will
+                * have the truncate lock exclusively so that will prevent
+                * them from issuing any I/O.
+                */
 
-       /* Only normal files can be exchanged. */
-       if ((from_vp->v_type != VREG) || (to_vp->v_type != VREG) ||
-           (from_cp->c_flag & C_HARDLINK) || (to_cp->c_flag & C_HARDLINK) ||
-           VNODE_IS_RSRC(from_vp) || VNODE_IS_RSRC(to_vp))
-               return (EINVAL);
+               if (to_cp < from_cp) {
+                       hfs_lock_truncate(to_cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
+                       have_to_trunc_lock = true;
+               }
 
-       // XXXdbg - don't allow modification of the journal or journal_info_block
-       if (hfsmp->jnl) {
-               struct HFSPlusExtentDescriptor *extd;
+               hfs_lock_truncate(from_cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
+               have_from_trunc_lock = true;
 
-               if (from_cp->c_datafork) {
-                       extd = &from_cp->c_datafork->ff_extents[0];
-                       if (extd->startBlock == VTOVCB(from_vp)->vcbJinfoBlock || extd->startBlock == hfsmp->jnl_start) {
-                               return EPERM;
-                       }
+               /*
+                * Do an early check to verify the source is not in use by
+                * anyone.  We should be called from an FD opened as F_EVTONLY
+                * so that doesn't count as a reference.
+                */
+               if (vnode_isinuse(from_vp, 0)) {
+                       error = EBUSY;
+                       goto exit;
                }
 
-               if (to_cp->c_datafork) {
-                       extd = &to_cp->c_datafork->ff_extents[0];
-                       if (extd->startBlock == VTOVCB(to_vp)->vcbJinfoBlock || extd->startBlock == hfsmp->jnl_start) {
-                               return EPERM;
-                       }
+               if (to_cp >= from_cp) {
+                       hfs_lock_truncate(to_cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
+                       have_to_trunc_lock = true;
                }
        }
 
-       // XXXdbg
-       hfs_global_shared_lock_acquire(hfsmp);
-       grabbed_lock = 1;
-       if (hfsmp->jnl) {
-           if ((error = journal_start_transaction(hfsmp->jnl)) != 0) {
-                       goto Err_Exit;
-           }
-               started_tr = 1;
+       if ((error = hfs_lockpair(from_cp, to_cp, HFS_EXCLUSIVE_LOCK)))
+               goto exit;
+       have_cnode_locks = true;
+
+       // Don't allow modification of the journal or journal_info_block
+       if (hfs_is_journal_file(hfsmp, from_cp) ||
+           hfs_is_journal_file(hfsmp, to_cp)) {
+               error = EPERM;
+               goto exit;
+       }
+
+       /* 
+        * Ok, now that all of the pre-flighting is done, call the underlying
+        * function if needed.
+        */
+       if (ISSET(ap->a_options, FSOPT_EXCHANGE_DATA_ONLY)) {
+#if HFS_COMPRESSION
+               if (ISSET(from_cp->c_bsdflags, UF_COMPRESSED)) {
+                       error = hfs_move_compressed(from_cp, to_cp);
+                       goto exit;
+               }
+#endif
+
+               error = hfs_move_data(from_cp, to_cp, 0);
+               goto exit;
+       }
+
+       if ((error = hfs_start_transaction(hfsmp)) != 0) {
+           goto exit;
        }
+       started_tr = 1;
        
        /*
         * Reserve some space in the Catalog file.
         */
-       if ((error = cat_preflight(hfsmp, CAT_EXCHANGE, &cookie, ap->a_p))) {
-               goto Err_Exit;
+       if ((error = cat_preflight(hfsmp, CAT_EXCHANGE, &cookie, vfs_context_proc(ap->a_context)))) {
+               goto exit;
        }
-
-       /* Lock catalog b-tree */
-       error = hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_EXCLUSIVE, ap->a_p);
-       if (error) goto Err_Exit;
+       got_cookie = 1;
 
        /* The backend code always tries to delete the virtual
-        * extent id for exchanging files so we neeed to lock
+        * extent id for exchanging files so we need to lock
         * the extents b-tree.
         */
-       error = hfs_metafilelocking(hfsmp, kHFSExtentsFileID, LK_EXCLUSIVE, ap->a_p);
-       if (error) {
-               (void) hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_RELEASE, ap->a_p);
-               goto Err_Exit;
+       lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG | SFL_EXTENTS | SFL_ATTRIBUTE, HFS_EXCLUSIVE_LOCK);
+
+       /* Account for the location of the catalog objects. */
+       if (from_cp->c_flag & C_HARDLINK) {
+               MAKE_INODE_NAME(from_iname, sizeof(from_iname),
+                               from_cp->c_attr.ca_linkref);
+               from_nameptr = (unsigned char *)from_iname;
+               from_parid = hfsmp->hfs_private_desc[FILE_HARDLINKS].cd_cnid;
+               from_cp->c_hint = 0;
+       } else {
+               from_nameptr = from_cp->c_desc.cd_nameptr;
+               from_parid = from_cp->c_parentcnid;
+       }
+       if (to_cp->c_flag & C_HARDLINK) {
+               MAKE_INODE_NAME(to_iname, sizeof(to_iname),
+                               to_cp->c_attr.ca_linkref);
+               to_nameptr = (unsigned char *)to_iname;
+               to_parid = hfsmp->hfs_private_desc[FILE_HARDLINKS].cd_cnid;
+               to_cp->c_hint = 0;
+       } else {
+               to_nameptr = to_cp->c_desc.cd_nameptr;
+               to_parid = to_cp->c_parentcnid;
        }
 
-       /* Do the exchange */
-       error = MacToVFSError(ExchangeFileIDs(HFSTOVCB(hfsmp),
-                               from_cp->c_desc.cd_nameptr, to_cp->c_desc.cd_nameptr,
-                               from_cp->c_parentcnid, to_cp->c_parentcnid,
-                               from_cp->c_hint, to_cp->c_hint));
+       /*
+        * ExchangeFileIDs swaps the on-disk, or in-BTree extent information 
+        * attached to two different file IDs.  It also swaps the extent 
+        * information that may live in the extents-overflow B-Tree.
+        *
+        * We do this in a transaction as this may require a lot of B-Tree nodes
+        * to do completely, particularly if one of the files in question
+        * has a lot of extents. 
+        *
+        * For example, assume "file1" has fileID 50, and "file2" has fileID 52.
+        * For the on-disk records, which are assumed to be synced, we will
+        * first swap the resident inline-8 extents as part of the catalog records.
+        * Then we will swap any extents overflow records for each file.
+        * 
+        * When ExchangeFileIDs returns successfully, "file1" will have fileID 52, 
+        * and "file2" will have fileID 50.  However, note that this is only 
+        * approximately half of the work that exchangedata(2) will need to 
+        * accomplish.  In other words, we swap "too much" of the information 
+        * because if we only called ExchangeFileIDs, both the fileID and extent 
+        * information would be the invariants of this operation.  We don't 
+        * actually want that; we want to conclude with "file1" having 
+        * file ID 50, and "file2" having fileID 52.
+        * 
+        * The remainder of hfs_vnop_exchange will swap the file ID and other cnode
+        * data back to the proper ownership, while still allowing the cnode to remain
+        * pointing at the same set of extents that it did originally.
+        */
+       error = ExchangeFileIDs(hfsmp, from_nameptr, to_nameptr, from_parid,
+                               to_parid, from_cp->c_hint, to_cp->c_hint);
+       hfs_systemfile_unlock(hfsmp, lockflags);
 
-       (void) hfs_metafilelocking(hfsmp, kHFSExtentsFileID, LK_RELEASE, ap->a_p);
-       (void) hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_RELEASE, ap->a_p);
+       /*
+        * Note that we don't need to exchange any extended attributes
+        * since the attributes are keyed by file ID.
+        */
 
        if (error != E_NONE) {
-               goto Err_Exit;
+               error = MacToVFSError(error);
+               goto exit;
        }
 
        /* Purge the vnodes from the name cache */
@@ -1099,133 +2311,690 @@ hfs_exchange(ap)
        if (to_vp)
                cache_purge(to_vp);
 
-       /* Save a copy of from attributes before swapping. */
+       /* Bump both source and destination write counts before any swaps. */
+       {
+               hfs_incr_gencount (from_cp);
+               hfs_incr_gencount (to_cp);
+       }
+
+       /* Save a copy of "from" attributes before swapping. */
        bcopy(&from_cp->c_desc, &tempdesc, sizeof(struct cat_desc));
        bcopy(&from_cp->c_attr, &tempattr, sizeof(struct cat_attr));
+       
+       /* Save whether or not each cnode is a hardlink or has EAs */
+       from_flag_special = from_cp->c_flag & (C_HARDLINK | C_HASXATTRS);
+       to_flag_special = to_cp->c_flag & (C_HARDLINK | C_HASXATTRS);
+
+       /* Drop the special bits from each cnode */
+       from_cp->c_flag &= ~(C_HARDLINK | C_HASXATTRS);
+       to_cp->c_flag &= ~(C_HARDLINK | C_HASXATTRS);
 
        /*
-        * Swap the descriptors and all non-fork related attributes.
-        * (except the modify date)
+        * Now complete the in-memory portion of the copy.
+        *
+        * ExchangeFileIDs swaps the on-disk records involved.  We complete the
+        * operation by swapping the in-memory contents of the two files here.
+        * We swap the cnode descriptors, which contain name, BSD attributes,
+        * timestamps, etc, about the file.  
+        * 
+        * NOTE: We do *NOT* swap the fileforks of the two cnodes.  We have
+        * already swapped the on-disk extent information.  As long as we swap the 
+        * IDs, the in-line resident 8 extents that live in the filefork data
+        * structure will point to the right data for the new file ID if we leave 
+        * them alone.
+        *
+        * As a result, any file descriptor that points to a particular
+        * vnode (even though it should change names), will continue
+        * to point to the same content. 
         */
+
+       /* Copy the "to" -> "from" cnode */
        bcopy(&to_cp->c_desc, &from_cp->c_desc, sizeof(struct cat_desc));
 
        from_cp->c_hint = 0;
-       from_cp->c_fileid = from_cp->c_cnid;
+       /*
+        * If 'to' was a hardlink, then we copied over its link ID/CNID/(namespace ID) 
+        * when we bcopy'd the descriptor above.  However, the cnode attributes
+        * are not bcopied.  As a result, make sure to swap the file IDs of each item.
+        *
+        * Further, other hardlink attributes must be moved along in this swap:
+        * the linkcount, the linkref, and the firstlink all need to move
+        * along with the file IDs.  See note below regarding the flags and
+        * what moves vs. what does not.
+        *
+        * For Reference:
+        * linkcount == total # of hardlinks.
+        * linkref == the indirect inode pointer.
+        * firstlink == the first hardlink in the chain (written to the raw inode).
+        * These three are tied to the fileID and must move along with the rest of the data.
+        */ 
+       from_cp->c_fileid = to_cp->c_attr.ca_fileid;
+
        from_cp->c_itime = to_cp->c_itime;
        from_cp->c_btime = to_cp->c_btime;
        from_cp->c_atime = to_cp->c_atime;
        from_cp->c_ctime = to_cp->c_ctime;
        from_cp->c_gid = to_cp->c_gid;
        from_cp->c_uid = to_cp->c_uid;
-       from_cp->c_flags = to_cp->c_flags;
+       from_cp->c_bsdflags = to_cp->c_bsdflags;
        from_cp->c_mode = to_cp->c_mode;
+       from_cp->c_linkcount = to_cp->c_linkcount;
+       from_cp->c_attr.ca_linkref = to_cp->c_attr.ca_linkref;
+       from_cp->c_attr.ca_firstlink = to_cp->c_attr.ca_firstlink;
+
+       /* 
+        * The cnode flags need to stay with the cnode and not get transferred
+        * over along with everything else because they describe the content; they are
+        * not attributes that reflect changes specific to the file ID.  In general, 
+        * fields that are tied to the file ID are the ones that will move.
+        * 
+        * This reflects the fact that the file may have borrowed blocks, dirty metadata, 
+        * or other extents, which may not yet have been written to the catalog.  If 
+        * they were, they would have been transferred above in the ExchangeFileIDs call above...
+        *
+        * The flags that are special are:
+        * C_HARDLINK, C_HASXATTRS
+        * 
+        * These flags move with the item and file ID in the namespace since their
+        * state is tied to that of the file ID.
+        * 
+        * So to transfer the flags, we have to take the following steps
+        * 1) Store in a localvar whether or not the special bits are set.
+        * 2) Drop the special bits from the current flags
+        * 3) swap the special flag bits to their destination
+        */      
+       from_cp->c_flag |= to_flag_special | C_MODIFIED;
+       from_cp->c_attr.ca_recflags = to_cp->c_attr.ca_recflags;
        bcopy(to_cp->c_finderinfo, from_cp->c_finderinfo, 32);
 
+
+       /* Copy the "from" -> "to" cnode */
        bcopy(&tempdesc, &to_cp->c_desc, sizeof(struct cat_desc));
        to_cp->c_hint = 0;
-       to_cp->c_fileid = to_cp->c_cnid;
+       /* 
+        * Pull the file ID from the tempattr we copied above. We can't assume 
+        * it is the same as the CNID.
+        */
+       to_cp->c_fileid = tempattr.ca_fileid;
        to_cp->c_itime = tempattr.ca_itime;
        to_cp->c_btime = tempattr.ca_btime;
        to_cp->c_atime = tempattr.ca_atime;
        to_cp->c_ctime = tempattr.ca_ctime;
        to_cp->c_gid = tempattr.ca_gid;
        to_cp->c_uid = tempattr.ca_uid;
-       to_cp->c_flags = tempattr.ca_flags;
+       to_cp->c_bsdflags = tempattr.ca_flags;
        to_cp->c_mode = tempattr.ca_mode;
+       to_cp->c_linkcount = tempattr.ca_linkcount;
+       to_cp->c_attr.ca_linkref = tempattr.ca_linkref;
+       to_cp->c_attr.ca_firstlink = tempattr.ca_firstlink;
+
+       /* 
+        * Only OR in the "from" flags into our cnode flags below. 
+        * Leave the rest of the flags alone.
+        */
+       to_cp->c_flag |= from_flag_special | C_MODIFIED;
+
+       to_cp->c_attr.ca_recflags = tempattr.ca_recflags;
        bcopy(tempattr.ca_finderinfo, to_cp->c_finderinfo, 32);
 
-       /* Reinsert into the cnode hash under new file IDs*/
-       hfs_chashremove(from_cp);
-       hfs_chashremove(to_cp);
 
-       hfs_chashinsert(from_cp);
-       hfs_chashinsert(to_cp);
+       /* Rehash the cnodes using their new file IDs */
+       hfs_chash_rehash(hfsmp, from_cp, to_cp);
 
        /*
         * When a file moves out of "Cleanup At Startup"
         * we can drop its NODUMP status.
         */
-       if ((from_cp->c_flags & UF_NODUMP) &&
+       if ((from_cp->c_bsdflags & UF_NODUMP) &&
            (from_cp->c_parentcnid != to_cp->c_parentcnid)) {
-               from_cp->c_flags &= ~UF_NODUMP;
-               from_cp->c_flag |= C_CHANGE;
+               from_cp->c_bsdflags &= ~UF_NODUMP;
+               from_cp->c_touch_chgtime = TRUE;
        }
-       if ((to_cp->c_flags & UF_NODUMP) &&
+       if ((to_cp->c_bsdflags & UF_NODUMP) &&
            (to_cp->c_parentcnid != from_cp->c_parentcnid)) {
-               to_cp->c_flags &= ~UF_NODUMP;
-               to_cp->c_flag |= C_CHANGE;
+               to_cp->c_bsdflags &= ~UF_NODUMP;
+               to_cp->c_touch_chgtime = TRUE;
+       }
+
+exit:
+       if (got_cookie) {
+               cat_postflight(hfsmp, &cookie, vfs_context_proc(ap->a_context));
+       }
+       if (started_tr) {
+           hfs_end_transaction(hfsmp);
        }
 
-       HFS_KNOTE(from_vp, NOTE_ATTRIB);
-       HFS_KNOTE(to_vp, NOTE_ATTRIB);
+       if (have_cnode_locks)
+               hfs_unlockpair(from_cp, to_cp);
 
-Err_Exit:
-       cat_postflight(hfsmp, &cookie, ap->a_p);
+       if (have_from_trunc_lock)
+               hfs_unlock_truncate(from_cp, 0);
 
-       // XXXdbg
-       if (started_tr) {
-           journal_end_transaction(hfsmp->jnl);
+       if (have_to_trunc_lock)
+               hfs_unlock_truncate(to_cp, 0);
+
+       return (error);
+}
+
+#if HFS_COMPRESSION
+/*
+ * This function is used specifically for the case when a namespace
+ * handler is trying to steal data before it's deleted.  Note that we
+ * don't bother deleting the xattr from the source because it will get
+ * deleted a short time later anyway.
+ *
+ * cnodes must be locked
+ */
+static int hfs_move_compressed(cnode_t *from_cp, cnode_t *to_cp)
+{
+       int     ret;
+       void   *data    = NULL;
+
+       CLR(from_cp->c_bsdflags, UF_COMPRESSED);
+       SET(from_cp->c_flag, C_MODIFIED);
+
+       ret = hfs_move_data(from_cp, to_cp, HFS_MOVE_DATA_INCLUDE_RSRC);
+       if (ret)
+               goto exit;
+
+       /*
+        * Transfer the xattr that decmpfs uses.  Ideally, this code
+        * should be with the other decmpfs code but it's file system
+        * agnostic and this path is currently, and likely to remain, HFS+
+        * specific.  It's easier and more performant if we implement it
+        * here.
+        */
+
+       size_t size = MAX_DECMPFS_XATTR_SIZE;
+       MALLOC(data, void *, size, M_TEMP, M_WAITOK);
+
+       ret = hfs_xattr_read(from_cp->c_vp, DECMPFS_XATTR_NAME, data, &size);
+       if (ret)
+               goto exit;
+
+       ret = hfs_xattr_write(to_cp->c_vp, DECMPFS_XATTR_NAME, data, size);
+       if (ret)
+               goto exit;
+
+       SET(to_cp->c_bsdflags, UF_COMPRESSED);
+       SET(to_cp->c_flag, C_MODIFIED);
+
+exit:
+       if (data)
+               FREE(data, M_TEMP);
+
+       return ret;
+}
+#endif // HFS_COMPRESSION
+
+int
+hfs_vnop_mmap(struct vnop_mmap_args *ap)
+{
+       struct vnode *vp = ap->a_vp;
+       cnode_t *cp = VTOC(vp);
+       int error;
+       
+       if (VNODE_IS_RSRC(vp)) {
+               /* allow pageins of the resource fork */
+       } else {
+               int compressed = hfs_file_is_compressed(cp, 1); /* 1 == don't take the cnode lock */
+               time_t orig_ctime = cp->c_ctime;
+               
+               if (!compressed && (cp->c_bsdflags & UF_COMPRESSED)) {
+                       error = check_for_dataless_file(vp, NAMESPACE_HANDLER_READ_OP);
+                       if (error != 0) {
+                               return error;
+                       }
+               }
+
+               if (ap->a_fflags & PROT_WRITE) {
+                       check_for_tracked_file(vp, orig_ctime, NAMESPACE_HANDLER_WRITE_OP, NULL);
+               }
        }
-       if (grabbed_lock) {
-               hfs_global_shared_lock_release(hfsmp);
+
+       //
+       // NOTE: we return ENOTSUP because we want the cluster layer
+       //       to actually do all the real work.
+       //
+       return (ENOTSUP);
+}
+
+static errno_t hfs_vnop_mnomap(struct vnop_mnomap_args *ap)
+{
+       vnode_t vp = ap->a_vp;
+
+       /*
+        * Whilst the file was mapped, there may not have been any
+        * page-outs so we need to increment the generation counter now.
+        * Unfortunately this may lead to a change in the generation
+        * counter when no actual change has been made, but there is
+        * little we can do about that with our current architecture.
+        */
+       if (ubc_is_mapped_writable(vp)) {
+               cnode_t *cp = VTOC(vp);
+               hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_ALLOW_NOEXISTS);
+               hfs_incr_gencount(cp);
+
+               /*
+                * We don't want to set the modification time here since a
+                * change to that is not acceptable if no changes were made.
+                * Instead we set a flag so that if we get any page-outs we
+                * know to update the modification time.  It's possible that
+                * they weren't actually because of changes made whilst the
+                * file was mapped but that's not easy to fix now.
+                */
+               SET(cp->c_flag, C_MIGHT_BE_DIRTY_FROM_MAPPING);
+
+               hfs_unlock(cp);
        }
 
-       return (error);
+       return 0;
 }
 
+/*
+ * Mark the resource fork as needing a ubc_setsize when we drop the
+ * cnode lock later.
+ */
+static void hfs_rsrc_setsize(cnode_t *cp)
+{
+       /*
+        * We need to take an iocount if we don't have one.  vnode_get
+        * will return ENOENT if the vnode is terminating which is what we
+        * want as it's not safe to call ubc_setsize in that case.
+        */
+       if (cp->c_rsrc_vp && !vnode_get(cp->c_rsrc_vp)) {
+               // Shouldn't happen, but better safe...
+               if (ISSET(cp->c_flag, C_NEED_RVNODE_PUT))
+                       vnode_put(cp->c_rsrc_vp);
+               SET(cp->c_flag, C_NEED_RVNODE_PUT | C_NEED_RSRC_SETSIZE);
+       }
+}
 
 /*
+ * hfs_move_data
+ *
+ * This is a non-symmetric variant of exchangedata.  In this function,
+ * the contents of the data fork (and optionally the resource fork)
+ * are moved from from_cp to to_cp.
+ * 
+ * The cnodes must be locked. 
+ *
+ * The cnode pointed to by 'to_cp' *must* be empty prior to invoking
+ * this function.  We impose this restriction because we may not be
+ * able to fully delete the entire file's contents in a single
+ * transaction, particularly if it has a lot of extents.  In the
+ * normal file deletion codepath, the file is screened for two
+ * conditions: 1) bigger than 400MB, and 2) more than 8 extents.  If
+ * so, the file is relocated to the hidden directory and the deletion
+ * is broken up into multiple truncates.  We can't do that here
+ * because both files need to exist in the namespace. The main reason
+ * this is imposed is that we may have to touch a whole lot of bitmap
+ * blocks if there are many extents.
+ * 
+ * Any data written to 'from_cp' after this call completes is not
+ * guaranteed to be moved.
+ * 
+ * Arguments:
+ * cnode_t *from_cp : source file
+ * cnode_t *to_cp   : destination file; must be empty
+ * 
+ * Returns:
+ *  
+ *  EBUSY - File has been deleted or is in use
+ *     EFBIG - Destination file was not empty
+ *  EIO   - An I/O error
+ *     0         - success
+ *  other - Other errors that can be returned from called functions
+ */
+int hfs_move_data(cnode_t *from_cp, cnode_t *to_cp, 
+                                 hfs_move_data_options_t options)
+{      
+       hfsmount_t *hfsmp                       = VTOHFS(from_cp->c_vp);
+       int error                                       = 0;
+       int lockflags                           = 0;
+       bool return_EIO_on_error        = false;
+       const bool include_rsrc         = ISSET(options, HFS_MOVE_DATA_INCLUDE_RSRC);
 
-#% fsync       vp      L L L
-#
- vop_fsync {
-     IN struct vnode *vp;
-     IN struct ucred *cred;
-     IN int waitfor;
-     IN struct proc *p;
+       /* Verify that neither source/dest file is open-unlinked */
+       if (ISSET(from_cp->c_flag, C_DELETED | C_NOEXISTS)
+               || ISSET(to_cp->c_flag, C_DELETED | C_NOEXISTS)) {
+               return EBUSY;
+       }
 
-     */
-static int
-hfs_fsync(ap)
-       struct vop_fsync_args /* {
-               struct vnode *a_vp;
-               struct ucred *a_cred;
-               int a_waitfor;
-               struct proc *a_p;
-       } */ *ap;
+       /* 
+        * Verify the source file is not in use by anyone besides us.
+        *
+        * This function is typically invoked by a namespace handler 
+        * process responding to a temporarily stalled system call.  
+        * The FD that it is working off of is opened O_EVTONLY, so
+        * it really has no active usecounts (the kusecount from O_EVTONLY
+        * is subtracted from the total usecounts).
+        * 
+        * As a result, we shouldn't have any active usecounts against
+        * this vnode when we go to check it below.
+        */
+       if (vnode_isinuse(from_cp->c_vp, 0))
+               return EBUSY;
+
+       if (include_rsrc && from_cp->c_rsrc_vp) {
+               if (vnode_isinuse(from_cp->c_rsrc_vp, 0))
+                       return EBUSY;
+
+               /* 
+                * In the code below, if the destination file doesn't have a
+                * c_rsrcfork then we don't create it which means we we cannot
+                * transfer the ff_invalidranges and cf_vblocks fields.  These
+                * shouldn't be set because we flush the resource fork before
+                * calling this function but there is a tiny window when we
+                * did not have any locks...
+                */
+               if (!to_cp->c_rsrcfork
+                       && (!TAILQ_EMPTY(&from_cp->c_rsrcfork->ff_invalidranges)
+                               || from_cp->c_rsrcfork->ff_unallocblocks)) {
+                       /*
+                        * The file isn't really busy now but something did slip
+                        * in and tinker with the file while we didn't have any
+                        * locks, so this is the most meaningful return code for
+                        * the caller.
+                        */
+                       return EBUSY;
+               }
+       }
+
+       // Check the destination file is empty
+       if (to_cp->c_datafork->ff_blocks 
+               || to_cp->c_datafork->ff_size
+               || (include_rsrc
+                       && (to_cp->c_blocks
+                               || (to_cp->c_rsrcfork && to_cp->c_rsrcfork->ff_size)))) {
+               return EFBIG;
+       }
+
+       if ((error = hfs_start_transaction (hfsmp)))
+               return error;
+
+       lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG | SFL_EXTENTS | SFL_ATTRIBUTE, 
+                                                                       HFS_EXCLUSIVE_LOCK);
+
+       // filefork_t is 128 bytes which should be OK
+       filefork_t rfork_buf, *from_rfork = NULL;
+
+       if (include_rsrc) {
+               from_rfork = from_cp->c_rsrcfork;
+
+               /*
+                * Creating resource fork vnodes is expensive, so just get get
+                * the fork data if we need it.
+                */
+               if (!from_rfork && hfs_has_rsrc(from_cp)) {
+                       from_rfork = &rfork_buf;
+
+                       from_rfork->ff_cp = from_cp;
+                       TAILQ_INIT(&from_rfork->ff_invalidranges);
+
+                       error = cat_idlookup(hfsmp, from_cp->c_fileid, 0, 1, NULL, NULL,
+                                                                &from_rfork->ff_data);
+
+                       if (error)
+                               goto exit;
+               }
+       }
+
+       /*
+        * From here on, any failures mean that we might be leaving things
+        * in a weird or inconsistent state.  Ideally, we should back out
+        * all the changes, but to do that properly we need to fix
+        * MoveData.  We'll save fixing that for another time.  For now,
+        * just return EIO in all cases to the caller so that they know.
+        */
+       return_EIO_on_error = true;
+
+       bool data_overflow_extents = overflow_extents(from_cp->c_datafork);
+
+       // Move the data fork
+       if ((error = hfs_move_fork (from_cp->c_datafork, from_cp,
+                                                               to_cp->c_datafork, to_cp))) {
+               goto exit;
+       }
+
+       SET(from_cp->c_flag, C_NEED_DATA_SETSIZE);
+       SET(to_cp->c_flag, C_NEED_DATA_SETSIZE);
+
+       // We move the resource fork later
+
+       /* 
+        * Note that because all we're doing is moving the extents around,
+        * we can probably do this in a single transaction: Each extent
+        * record (group of 8) is 64 bytes.  A extent overflow B-Tree node
+        * is typically 4k.  This means each node can hold roughly ~60
+        * extent records == (480 extents).
+        *
+        * If a file was massively fragmented and had 20k extents, this
+        * means we'd roughly touch 20k/480 == 41 to 42 nodes, plus the
+        * index nodes, for half of the operation.  (inserting or
+        * deleting). So if we're manipulating 80-100 nodes, this is
+        * basically 320k of data to write to the journal in a bad case.
+        */
+       if (data_overflow_extents) {
+               if ((error = MoveData(hfsmp, from_cp->c_cnid, to_cp->c_cnid, 0)))
+                       goto exit;
+       }
+
+       if (from_rfork && overflow_extents(from_rfork)) {
+               if ((error = MoveData(hfsmp, from_cp->c_cnid, to_cp->c_cnid, 1)))
+                       goto exit;
+       }
+
+       // Touch times
+       from_cp->c_touch_acctime = TRUE;
+       from_cp->c_touch_chgtime = TRUE;
+       from_cp->c_touch_modtime = TRUE;
+       hfs_touchtimes(hfsmp, from_cp);
+
+       to_cp->c_touch_acctime = TRUE;
+       to_cp->c_touch_chgtime = TRUE;
+       to_cp->c_touch_modtime = TRUE;
+       hfs_touchtimes(hfsmp, to_cp);
+
+       struct cat_fork dfork_buf;
+       const struct cat_fork *dfork, *rfork;
+
+       dfork = hfs_prepare_fork_for_update(to_cp->c_datafork, NULL,
+                                                                               &dfork_buf, hfsmp->blockSize);
+       rfork = hfs_prepare_fork_for_update(from_rfork, NULL,
+                                                                               &rfork_buf.ff_data, hfsmp->blockSize);
+
+       // Update the catalog nodes, to_cp first
+       if ((error = cat_update(hfsmp, &to_cp->c_desc, &to_cp->c_attr, 
+                                                       dfork, rfork))) {
+               goto exit;
+       }
+
+       CLR(to_cp->c_flag, C_MODIFIED | C_MINOR_MOD);
+
+       // Update in-memory resource fork data here
+       if (from_rfork) {
+               // Update c_blocks
+               uint32_t moving = from_rfork->ff_blocks + from_rfork->ff_unallocblocks;
+
+               from_cp->c_blocks -= moving;
+               to_cp->c_blocks   += moving;
+
+               // Update to_cp's resource data if it has it
+               filefork_t *to_rfork = to_cp->c_rsrcfork;
+               if (to_rfork) {
+                       to_rfork->ff_invalidranges = from_rfork->ff_invalidranges;
+                       to_rfork->ff_data                  = from_rfork->ff_data;
+
+                       // Deal with ubc_setsize
+                       hfs_rsrc_setsize(to_cp);
+               }
+
+               // Wipe out the resource fork in from_cp
+               rl_init(&from_rfork->ff_invalidranges);
+               bzero(&from_rfork->ff_data, sizeof(from_rfork->ff_data));
+
+               // Deal with ubc_setsize
+               hfs_rsrc_setsize(from_cp);
+       }
+
+       // Currently unnecessary, but might be useful in future...
+       dfork = hfs_prepare_fork_for_update(from_cp->c_datafork, NULL, &dfork_buf,
+                                                                               hfsmp->blockSize);
+       rfork = hfs_prepare_fork_for_update(from_rfork, NULL, &rfork_buf.ff_data,
+                                                                               hfsmp->blockSize);
+
+       // Update from_cp
+       if ((error = cat_update(hfsmp, &from_cp->c_desc, &from_cp->c_attr, 
+                                                       dfork, rfork))) {
+               goto exit;
+       }
+
+       CLR(from_cp->c_flag, C_MODIFIED | C_MINOR_MOD);
+
+exit:
+       if (lockflags) {
+               hfs_systemfile_unlock(hfsmp, lockflags);
+               hfs_end_transaction(hfsmp);
+       }
+
+       if (error && error != EIO && return_EIO_on_error) {
+               printf("hfs_move_data: encountered error %d\n", error);
+               error = EIO;
+       }
+
+       return error;
+}
+
+/* 
+ * Move all of the catalog and runtime data in srcfork to dstfork.
+ * 
+ * This allows us to maintain the invalid ranges across the move data
+ * operation so we don't need to force all of the pending IO right
+ * now. In addition, we move all non overflow-extent extents into the
+ * destination here.
+ *
+ * The destination fork must be empty and should have been checked
+ * prior to calling this.
+ */
+static int hfs_move_fork(filefork_t *srcfork, cnode_t *src_cp,
+                                                filefork_t *dstfork, cnode_t *dst_cp) 
+{
+       // Move the invalid ranges
+       TAILQ_SWAP(&dstfork->ff_invalidranges, &srcfork->ff_invalidranges,
+                          rl_entry, rl_link);
+       rl_remove_all(&srcfork->ff_invalidranges);
+
+       // Move the fork data (copy whole structure)
+       dstfork->ff_data = srcfork->ff_data;
+       bzero(&srcfork->ff_data, sizeof(srcfork->ff_data));
+
+       // Update c_blocks
+       src_cp->c_blocks -= dstfork->ff_blocks + dstfork->ff_unallocblocks;
+       dst_cp->c_blocks += dstfork->ff_blocks + dstfork->ff_unallocblocks;
+
+       return 0;
+}
+
+
+#include <i386/panic_hooks.h>
+
+struct hfs_fsync_panic_hook {
+       panic_hook_t hook;
+       struct cnode *cp;
+};
+
+static void hfs_fsync_panic_hook(panic_hook_t *hook_)
+{
+       struct hfs_fsync_panic_hook *hook = (struct hfs_fsync_panic_hook *)hook_;
+       extern int kdb_log(const char *fmt, ...);
+
+       // Get the physical region just before cp
+       panic_phys_range_t range;
+       uint64_t phys;
+
+       if (panic_phys_range_before(hook->cp, &phys, &range)) {
+               kdb_log("cp = %p, phys = %p, prev (%p: %p-%p)\n", 
+                               hook->cp, phys, range.type, range.phys_start,
+                               range.phys_start + range.len);
+       } else
+               kdb_log("cp = %p, phys = %p, prev (!)\n", hook->cp, phys);
+
+       panic_dump_mem((void *)(((vm_offset_t)hook->cp - 4096) & ~4095), 12288);
+
+       kdb_log("\n");
+}
+
+
+/*
+ *  cnode must be locked
+ */
+int
+hfs_fsync(struct vnode *vp, int waitfor, hfs_fsync_mode_t fsyncmode, struct proc *p)
 {
-       struct vnode *vp = ap->a_vp;
        struct cnode *cp = VTOC(vp);
        struct filefork *fp = NULL;
        int retval = 0;
-       register struct buf *bp;
+       struct hfsmount *hfsmp = VTOHFS(vp);
        struct timeval tv;
-       struct buf *nbp;
-       struct hfsmount *hfsmp = VTOHFS(ap->a_vp);
-       int s;
-       int wait;
-       int retry = 0;
+       int waitdata;           /* attributes necessary for data retrieval */
+       int wait;               /* all other attributes (e.g. atime, etc.) */
+       int lockflag;
+       int took_trunc_lock = 0;
+       int locked_buffers = 0;
+       int fsync_default = 1;
 
-       wait = (ap->a_waitfor == MNT_WAIT);
+       /*
+        * Applications which only care about data integrity rather than full
+        * file integrity may opt out of (delay) expensive metadata update
+        * operations as a performance optimization.
+        */
+       wait = (waitfor == MNT_WAIT);
+       waitdata = (waitfor == MNT_DWAIT) | wait;
 
+       if (always_do_fullfsync)
+               fsyncmode = HFS_FSYNC_FULL;
+       if (fsyncmode != HFS_FSYNC)
+               fsync_default = 0;
+       
        /* HFS directories don't have any data blocks. */
-       if (vp->v_type == VDIR)
+       if (vnode_isdir(vp))
                goto metasync;
+       fp = VTOF(vp);
 
        /*
         * For system files flush the B-tree header and
         * for regular files write out any clusters
         */
-       if (vp->v_flag & VSYSTEM) {
+       if (vnode_issystem(vp)) {
            if (VTOF(vp)->fcbBTCBPtr != NULL) {
                        // XXXdbg
                        if (hfsmp->jnl == NULL) {
                                BTFlushPath(VTOF(vp));
                        }
            }
-       } else if (UBCINFOEXISTS(vp))
-               (void) cluster_push(vp);
+       } else if (UBCINFOEXISTS(vp)) {
+               hfs_unlock(cp);
+               hfs_lock_truncate(cp, HFS_SHARED_LOCK, HFS_LOCK_DEFAULT);
+               took_trunc_lock = 1;
+
+               struct hfs_fsync_panic_hook hook;
+               hook.cp = cp;
+               panic_hook(&hook.hook, hfs_fsync_panic_hook);
 
+               if (fp->ff_unallocblocks != 0) {
+                       hfs_unlock_truncate(cp, HFS_LOCK_DEFAULT);
+
+                       hfs_lock_truncate(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
+               }
+
+               panic_unhook(&hook.hook);
+
+               /* Don't hold cnode lock when calling into cluster layer. */
+               (void) cluster_push(vp, waitdata ? IO_SYNC : 0);
+
+               hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_ALLOW_NOEXISTS);
+       }
        /*
         * When MNT_WAIT is requested and the zero fill timeout
         * has expired then we must explicitly zero out any areas
@@ -1233,162 +3002,139 @@ hfs_fsync(ap)
         *
         * Files with NODUMP can bypass zero filling here.
         */
-       if ((wait || (cp->c_flag & C_ZFWANTSYNC)) &&
-           ((cp->c_flags & UF_NODUMP) == 0) &&
-           UBCINFOEXISTS(vp) && (fp = VTOF(vp)) &&
-           cp->c_zftimeout != 0) {
-               int devblksize;
-               int was_nocache;
-
-               if (time.tv_sec < cp->c_zftimeout) {
+       if (fp && (((cp->c_flag & C_ALWAYS_ZEROFILL) && !TAILQ_EMPTY(&fp->ff_invalidranges)) ||
+           ((wait || (cp->c_flag & C_ZFWANTSYNC)) &&
+               ((cp->c_bsdflags & UF_NODUMP) == 0) &&
+               UBCINFOEXISTS(vp) && (vnode_issystem(vp) ==0) &&
+               cp->c_zftimeout != 0))) {
+
+               microuptime(&tv);
+               if ((cp->c_flag & C_ALWAYS_ZEROFILL) == 0 && fsync_default && tv.tv_sec < (long)cp->c_zftimeout) {
                        /* Remember that a force sync was requested. */
                        cp->c_flag |= C_ZFWANTSYNC;
-                       goto loop;
-               }       
-               VOP_DEVBLOCKSIZE(cp->c_devvp, &devblksize);
-               was_nocache = ISSET(vp->v_flag, VNOCACHE_DATA);
-               SET(vp->v_flag, VNOCACHE_DATA); /* Don't cache zeros */
-
-               while (!CIRCLEQ_EMPTY(&fp->ff_invalidranges)) {
-                       struct rl_entry *invalid_range = CIRCLEQ_FIRST(&fp->ff_invalidranges);
-                       off_t start = invalid_range->rl_start;
-                       off_t end = invalid_range->rl_end;
-               
-                       /* The range about to be written must be validated
-                        * first, so that VOP_CMAP() will return the
-                        * appropriate mapping for the cluster code:
-                        */
-                       rl_remove(start, end, &fp->ff_invalidranges);
-
-                       (void) cluster_write(vp, (struct uio *) 0,
-                                       fp->ff_size,
-                                       invalid_range->rl_end + 1,
-                                       invalid_range->rl_start,
-                                       (off_t)0, devblksize,
-                                       IO_HEADZEROFILL | IO_NOZERODIRTY);
-                       cp->c_flag |= C_MODIFIED;
+                       goto datasync;
+               }
+               if (!TAILQ_EMPTY(&fp->ff_invalidranges)) {
+                       if (!took_trunc_lock || (cp->c_truncatelockowner == HFS_SHARED_OWNER)) {
+                               hfs_unlock(cp);
+                               if (took_trunc_lock) {
+                                       hfs_unlock_truncate(cp, HFS_LOCK_DEFAULT);
+                               }
+                               hfs_lock_truncate(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
+                               hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_ALLOW_NOEXISTS);
+                               took_trunc_lock = 1;
+                       }
+                       hfs_flush_invalid_ranges(vp);
+                       hfs_unlock(cp);
+                       (void) cluster_push(vp, waitdata ? IO_SYNC : 0);
+                       hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_ALLOW_NOEXISTS);
                }
-               (void) cluster_push(vp);
-               if (!was_nocache)
-                       CLR(vp->v_flag, VNOCACHE_DATA);
-               cp->c_flag &= ~C_ZFWANTSYNC;
-               cp->c_zftimeout = 0;
        }
+datasync:
+       if (took_trunc_lock) {
+               hfs_unlock_truncate(cp, HFS_LOCK_DEFAULT);
+               took_trunc_lock = 0;
+       }
+       /*
+        * if we have a journal and if journal_active() returns != 0 then the
+        * we shouldn't do anything to a locked block (because it is part 
+        * of a transaction).  otherwise we'll just go through the normal 
+        * code path and flush the buffer.  note journal_active() can return
+        * -1 if the journal is invalid -- however we still need to skip any 
+        * locked blocks as they get cleaned up when we finish the transaction
+        * or close the journal.
+        */
+       // if (hfsmp->jnl && journal_active(hfsmp->jnl) >= 0)
+       if (hfsmp->jnl)
+               lockflag = BUF_SKIP_LOCKED;
+       else
+               lockflag = 0;
 
        /*
         * Flush all dirty buffers associated with a vnode.
+        * Record how many of them were dirty AND locked (if necessary).
         */
-loop:
-       s = splbio();
-       for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = nbp) {
-               nbp = bp->b_vnbufs.le_next;
-               if ((bp->b_flags & B_BUSY))
-                       continue;
-               if ((bp->b_flags & B_DELWRI) == 0)
-                       panic("hfs_fsync: bp 0x% not dirty (hfsmp 0x%x)", bp, hfsmp);
-               // XXXdbg
-               if (hfsmp->jnl && (bp->b_flags & B_LOCKED)) {
-                       if ((bp->b_flags & B_META) == 0) {
-                               panic("hfs: bp @ 0x%x is locked but not meta! jnl 0x%x\n",
-                                         bp, hfsmp->jnl);
-                       }
-                       // if journal_active() returns >= 0 then the journal is ok and we 
-                       // shouldn't do anything to this locked block (because it is part 
-                       // of a transaction).  otherwise we'll just go through the normal 
-                       // code path and flush the buffer.
-                       if (journal_active(hfsmp->jnl) >= 0) {
-                               continue;
-                       }
+       locked_buffers = buf_flushdirtyblks_skipinfo(vp, waitdata, lockflag, "hfs_fsync");
+       if ((lockflag & BUF_SKIP_LOCKED) && (locked_buffers) && (vnode_vtype(vp) == VLNK)) {
+               /* 
+                * If there are dirty symlink buffers, then we may need to take action
+                * to prevent issues later on if we are journaled. If we're fsyncing a 
+                * symlink vnode then we are in one of three cases:
+                * 
+                * 1) automatic sync has fired.  In this case, we don't want the behavior to change.
+                * 
+                * 2) Someone has opened the FD for the symlink (not what it points to)
+                * and has issued an fsync against it.  This should be rare, and we don't
+                * want the behavior to change.
+                * 
+                * 3) We are being called by a vclean which is trying to reclaim this
+                * symlink vnode.  If this is the case, then allowing this fsync to 
+                * proceed WITHOUT flushing the journal could result in the vclean 
+                * invalidating the buffer's blocks before the journal transaction is
+                * written to disk.  To prevent this, we force a journal flush 
+                * if the vnode is in the middle of a recycle (VL_TERMINATE or VL_DEAD is set).
+                */
+               if (vnode_isrecycled(vp)) {
+                       fsync_default = 0;
                }
+       }
 
-               bremfree(bp);
-               bp->b_flags |= B_BUSY;
-               /* Clear B_LOCKED, should only be set on meta files */
-               bp->b_flags &= ~B_LOCKED;
+metasync:
+       if (vnode_isreg(vp) && vnode_issystem(vp)) {
+               if (VTOF(vp)->fcbBTCBPtr != NULL) {
+                       microuptime(&tv);
+                       BTSetLastSync(VTOF(vp), tv.tv_sec);
+               }
+               cp->c_touch_acctime = FALSE;
+               cp->c_touch_chgtime = FALSE;
+               cp->c_touch_modtime = FALSE;
+       } else if ( !(vp->v_flag & VSWAP) ) /* User file */ {
+               retval = hfs_update(vp, HFS_UPDATE_FORCE);
 
-               splx(s);
                /*
-                * Wait for I/O associated with indirect blocks to complete,
-                * since there is no way to quickly wait for them below.
+                * When MNT_WAIT is requested push out the catalog record for
+                * this file.  If they asked for a full fsync, we can skip this
+                * because the journal_flush or hfs_metasync_all will push out
+                * all of the metadata changes.
                 */
-               if (bp->b_vp == vp || ap->a_waitfor == MNT_NOWAIT)
-                       (void) bawrite(bp);
-               else
-                       (void) VOP_BWRITE(bp);
-               goto loop;
-       }
-
-       if (wait) {
-               while (vp->v_numoutput) {
-                       vp->v_flag |= VBWAIT;
-                       tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "hfs_fsync", 0);
+               if ((retval == 0) && wait && fsync_default && cp->c_hint &&
+                   !ISSET(cp->c_flag, C_DELETED | C_NOEXISTS)) {
+                       hfs_metasync(VTOHFS(vp), (daddr64_t)cp->c_hint, p);
                }
 
-               // XXXdbg -- is checking for hfsmp->jnl == NULL the right
-               //           thing to do?
-               if (hfsmp->jnl == NULL && vp->v_dirtyblkhd.lh_first) {
-                       /* still have some dirty buffers */
-                       if (retry++ > 10) {
-                               vprint("hfs_fsync: dirty", vp);
-                               splx(s);
-                               /*
-                                * Looks like the requests are not
-                                * getting queued to the driver.
-                                * Retrying here causes a cpu bound loop.
-                                * Yield to the other threads and hope
-                                * for the best.
-                                */
-                               (void)tsleep((caddr_t)&vp->v_numoutput,
-                                       PRIBIO + 1, "hfs_fsync", hz/10);
-                               retry = 0;
+               /*
+                * If this was a full fsync, make sure all metadata
+                * changes get to stable storage.
+                */
+               if (!fsync_default) {
+                       if (hfsmp->jnl) {
+                               if (fsyncmode == HFS_FSYNC_FULL)
+                                       hfs_flush(hfsmp, HFS_FLUSH_FULL);
+                               else
+                                       hfs_flush(hfsmp,
+                                           HFS_FLUSH_JOURNAL_BARRIER);
                        } else {
-                               splx(s);
+                               retval = hfs_metasync_all(hfsmp);
+                               /* XXX need to pass context! */
+                               hfs_flush(hfsmp, HFS_FLUSH_CACHE);
                        }
-                       /* try again */
-                       goto loop;
                }
        }
-       splx(s);
 
-metasync:
-       tv = time;
-       if (vp->v_flag & VSYSTEM) {
-               if (VTOF(vp)->fcbBTCBPtr != NULL)
-                       BTSetLastSync(VTOF(vp), tv.tv_sec);
-               cp->c_flag &= ~(C_ACCESS | C_CHANGE | C_MODIFIED | C_UPDATE);
-       } else /* User file */ {
-               retval = VOP_UPDATE(ap->a_vp, &tv, &tv, wait);
-
-               /* When MNT_WAIT is requested push out any delayed meta data */
-               if ((retval == 0) && wait && cp->c_hint &&
-                   !ISSET(cp->c_flag, C_DELETED | C_NOEXISTS)) {
-                       hfs_metasync(VTOHFS(vp), cp->c_hint, ap->a_p);
-               }
-
-               // make sure that we've really been called from the user
-               // fsync() and if so push out any pending transactions 
-               // that this file might is a part of (and get them on
-               // stable storage).
-               if (vp->v_flag & VFULLFSYNC) {
-                   if (hfsmp->jnl) {
-                       journal_flush(hfsmp->jnl);
-                   } else {
-                       VOP_IOCTL(hfsmp->hfs_devvp, DKIOCSYNCHRONIZECACHE, NULL, FWRITE, NOCRED, ap->a_p);
-                   }
-               }
-       }
+       if (!hfs_is_dirty(cp) && !ISSET(cp->c_flag, C_DELETED))
+               vnode_cleardirty(vp);
 
        return (retval);
 }
 
+
 /* Sync an hfs catalog b-tree node */
-static int
-hfs_metasync(struct hfsmount *hfsmp, daddr_t node, struct proc *p)
+int
+hfs_metasync(struct hfsmount *hfsmp, daddr64_t node, __unused struct proc *p)
 {
-       struct vnode *vp;
-       struct buf *bp;
-       struct buf *nbp;
-       int s;
+       vnode_t vp;
+       buf_t   bp;
+       int lockflags;
 
        vp = HFSTOVCB(hfsmp)->catalogRefNum;
 
@@ -1397,169 +3143,289 @@ hfs_metasync(struct hfsmount *hfsmp, daddr_t node, struct proc *p)
                return 0;
        }
 
-       if (hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_EXCLUSIVE, p) != 0)
-               return (0);
-
+       lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG, HFS_EXCLUSIVE_LOCK);
        /*
         * Look for a matching node that has been delayed
         * but is not part of a set (B_LOCKED).
+        *
+        * BLK_ONLYVALID causes buf_getblk to return a
+        * buf_t for the daddr64_t specified only if it's
+        * currently resident in the cache... the size
+        * parameter to buf_getblk is ignored when this flag
+        * is set
         */
-       s = splbio();
-       for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = nbp) {
-               nbp = bp->b_vnbufs.le_next;
-               if (bp->b_flags & B_BUSY)
-                       continue;
-               if (bp->b_lblkno == node) {
-                       if (bp->b_flags & B_LOCKED)
-                               break;
-
-                       bremfree(bp);
-                       bp->b_flags |= B_BUSY;
-                       splx(s);
-                       (void) VOP_BWRITE(bp);
-                       goto exit;
-               }
+       bp = buf_getblk(vp, node, 0, 0, 0, BLK_META | BLK_ONLYVALID);
+
+       if (bp) {
+               if ((buf_flags(bp) & (B_LOCKED | B_DELWRI)) == B_DELWRI)
+                       (void) VNOP_BWRITE(bp);
+               else
+                       buf_brelse(bp);
        }
-       splx(s);
-exit:
-       (void) hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_RELEASE, p);
+
+       hfs_systemfile_unlock(hfsmp, lockflags);
 
        return (0);
 }
 
-__private_extern__
+
+/*
+ * Sync all hfs B-trees.  Use this instead of journal_flush for a volume
+ * without a journal.  Note that the volume bitmap does not get written;
+ * we rely on fsck_hfs to fix that up (which it can do without any loss
+ * of data).
+ */
+int
+hfs_metasync_all(struct hfsmount *hfsmp)
+{
+       int lockflags;
+
+       /* Lock all of the B-trees so we get a mutually consistent state */
+       lockflags = hfs_systemfile_lock(hfsmp,
+               SFL_CATALOG|SFL_EXTENTS|SFL_ATTRIBUTE, HFS_EXCLUSIVE_LOCK);
+
+       /* Sync each of the B-trees */
+       if (hfsmp->hfs_catalog_vp)
+               hfs_btsync(hfsmp->hfs_catalog_vp, 0);
+       if (hfsmp->hfs_extents_vp)
+               hfs_btsync(hfsmp->hfs_extents_vp, 0);
+       if (hfsmp->hfs_attribute_vp)
+               hfs_btsync(hfsmp->hfs_attribute_vp, 0);
+       
+       /* Wait for all of the writes to complete */
+       if (hfsmp->hfs_catalog_vp)
+               vnode_waitforwrites(hfsmp->hfs_catalog_vp, 0, 0, 0, "hfs_metasync_all");
+       if (hfsmp->hfs_extents_vp)
+               vnode_waitforwrites(hfsmp->hfs_extents_vp, 0, 0, 0, "hfs_metasync_all");
+       if (hfsmp->hfs_attribute_vp)
+               vnode_waitforwrites(hfsmp->hfs_attribute_vp, 0, 0, 0, "hfs_metasync_all");
+
+       hfs_systemfile_unlock(hfsmp, lockflags);
+       
+       return 0;
+}
+
+
+/*ARGSUSED 1*/
+static int
+hfs_btsync_callback(struct buf *bp, __unused void *dummy)
+{
+       buf_clearflags(bp, B_LOCKED);
+       (void) buf_bawrite(bp);
+
+       return(BUF_CLAIMED);
+}
+
+
 int
 hfs_btsync(struct vnode *vp, int sync_transaction)
 {
        struct cnode *cp = VTOC(vp);
-       register struct buf *bp;
        struct timeval tv;
-       struct buf *nbp;
-       struct hfsmount *hfsmp = VTOHFS(vp);
-       int s;
+       int    flags = 0;
 
+       if (sync_transaction)
+               flags |= BUF_SKIP_NONLOCKED;
        /*
         * Flush all dirty buffers associated with b-tree.
         */
-loop:
-       s = splbio();
-
-       for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = nbp) {
-               nbp = bp->b_vnbufs.le_next;
-               if ((bp->b_flags & B_BUSY))
-                       continue;
-               if ((bp->b_flags & B_DELWRI) == 0)
-                       panic("hfs_btsync: not dirty (bp 0x%x hfsmp 0x%x)", bp, hfsmp);
-
-               // XXXdbg
-               if (hfsmp->jnl && (bp->b_flags & B_LOCKED)) {
-                       if ((bp->b_flags & B_META) == 0) {
-                               panic("hfs: bp @ 0x%x is locked but not meta! jnl 0x%x\n",
-                                         bp, hfsmp->jnl);
-                       }
-                       // if journal_active() returns >= 0 then the journal is ok and we 
-                       // shouldn't do anything to this locked block (because it is part 
-                       // of a transaction).  otherwise we'll just go through the normal 
-                       // code path and flush the buffer.
-                       if (journal_active(hfsmp->jnl) >= 0) {
-                           continue;
-                       }
-               }
+       buf_iterate(vp, hfs_btsync_callback, flags, 0);
 
-               if (sync_transaction && !(bp->b_flags & B_LOCKED))
-                       continue;
-
-               bremfree(bp);
-               bp->b_flags |= B_BUSY;
-               bp->b_flags &= ~B_LOCKED;
-
-               splx(s);
-
-               (void) bawrite(bp);
-
-               goto loop;
-       }
-       splx(s);
-
-       tv = time;
-       if ((vp->v_flag & VSYSTEM) && (VTOF(vp)->fcbBTCBPtr != NULL))
+       microuptime(&tv);
+       if (vnode_issystem(vp) && (VTOF(vp)->fcbBTCBPtr != NULL))
                (void) BTSetLastSync(VTOF(vp), tv.tv_sec);
-       cp->c_flag &= ~(C_ACCESS | C_CHANGE | C_MODIFIED | C_UPDATE);
+       cp->c_touch_acctime = FALSE;
+       cp->c_touch_chgtime = FALSE;
+       cp->c_touch_modtime = FALSE;
 
        return 0;
 }
 
 /*
- * Rmdir system call.
-#% rmdir       dvp     L U U
-#% rmdir       vp      L U U
-#
- vop_rmdir {
-     IN WILLRELE struct vnode *dvp;
-     IN WILLRELE struct vnode *vp;
-     IN struct componentname *cnp;
-
+ * Remove a directory.
  */
-static int
-hfs_rmdir(ap)
-       struct vop_rmdir_args /* {
+int
+hfs_vnop_rmdir(ap)
+       struct vnop_rmdir_args /* {
                struct vnode *a_dvp;
                struct vnode *a_vp;
                struct componentname *a_cnp;
+               vfs_context_t a_context;
        } */ *ap;
 {
-       return (hfs_removedir(ap->a_dvp, ap->a_vp, ap->a_cnp, 0));
+       struct vnode *dvp = ap->a_dvp;
+       struct vnode *vp = ap->a_vp;
+       struct cnode *dcp = VTOC(dvp);
+       struct cnode *cp = VTOC(vp);
+       int error;
+       time_t orig_ctime;
+
+       orig_ctime = VTOC(vp)->c_ctime;
+
+       if (!S_ISDIR(cp->c_mode)) {
+               return (ENOTDIR);
+       }
+       if (dvp == vp) {
+               return (EINVAL);
+       }
+
+       check_for_tracked_file(vp, orig_ctime, NAMESPACE_HANDLER_DELETE_OP, NULL);
+       cp = VTOC(vp);
+
+       if ((error = hfs_lockpair(dcp, cp, HFS_EXCLUSIVE_LOCK))) {
+               return (error);
+       }
+
+       /* Check for a race with rmdir on the parent directory */
+       if (dcp->c_flag & (C_DELETED | C_NOEXISTS)) {
+               hfs_unlockpair (dcp, cp);
+               return ENOENT;
+       }
+
+       //
+       // if the item is tracked but doesn't have a document_id, assign one and generate an fsevent for it
+       //
+       if ((cp->c_bsdflags & UF_TRACKED) && ((struct FndrExtendedDirInfo *)((char *)&cp->c_attr.ca_finderinfo + 16))->document_id == 0) {
+               uint32_t newid;
+
+               hfs_unlockpair(dcp, cp);
+
+               if (hfs_generate_document_id(VTOHFS(vp), &newid) == 0) {
+                       hfs_lockpair(dcp, cp, HFS_EXCLUSIVE_LOCK);
+                       ((struct FndrExtendedDirInfo *)((char *)&cp->c_attr.ca_finderinfo + 16))->document_id = newid;
+#if CONFIG_FSE
+                       add_fsevent(FSE_DOCID_CHANGED, vfs_context_current(),
+                                   FSE_ARG_DEV,   VTOHFS(vp)->hfs_raw_dev,
+                                   FSE_ARG_INO,   (ino64_t)0,             // src inode #
+                                   FSE_ARG_INO,   (ino64_t)cp->c_fileid,  // dst inode #
+                                   FSE_ARG_INT32, newid,
+                                   FSE_ARG_DONE);
+#endif
+               } else {
+                       // XXXdbg - couldn't get a new docid... what to do?  can't really fail the rm...
+                       hfs_lockpair(dcp, cp, HFS_EXCLUSIVE_LOCK);
+               }
+       }
+
+       error = hfs_removedir(dvp, vp, ap->a_cnp, 0, 0);
+
+       hfs_unlockpair(dcp, cp);
+
+       return (error);
 }
 
 /*
- * hfs_removedir
+ * Remove a directory
+ *
+ * Both dvp and vp cnodes are locked
  */
-static int
-hfs_removedir(dvp, vp, cnp, options)
-       struct vnode *dvp;
-       struct vnode *vp;
-       struct componentname *cnp;
-       int options;
+int
+hfs_removedir(struct vnode *dvp, struct vnode *vp, struct componentname *cnp,
+              int skip_reserve, int only_unlink)
 {
-       struct proc *p = cnp->cn_proc;
        struct cnode *cp;
        struct cnode *dcp;
        struct hfsmount * hfsmp;
-       struct timeval tv;
-       cat_cookie_t cookie = {0};
-       int error = 0, started_tr = 0, grabbed_lock = 0;
+       struct cat_desc desc;
+       int lockflags;
+       int error = 0, started_tr = 0;
 
        cp = VTOC(vp);
        dcp = VTOC(dvp);
        hfsmp = VTOHFS(vp);
 
        if (dcp == cp) {
-               vrele(dvp);
-               vput(vp);
                return (EINVAL);        /* cannot remove "." */
        }
-
-#if QUOTA
-       (void)hfs_getinoquota(cp);
-#endif
-       // XXXdbg
-       hfs_global_shared_lock_acquire(hfsmp);
-       grabbed_lock = 1;
-       if (hfsmp->jnl) {
-           if ((error = journal_start_transaction(hfsmp->jnl)) != 0) {
-                       goto out;
-           }
-               started_tr = 1;
+       if (cp->c_flag & (C_NOEXISTS | C_DELETED)) {
+               return (0);
+       }
+       if (cp->c_entries != 0) {
+               return (ENOTEMPTY);
+       }
+       
+       /*
+        * If the directory is open or in use (e.g. opendir() or current working
+        * directory for some process); wait for inactive/reclaim to actually
+        * remove cnode from the catalog.  Both inactive and reclaim codepaths are capable
+        * of removing open-unlinked directories from the catalog, as well as getting rid
+        * of EAs still on the element.  So change only_unlink to true, so that it will get 
+        * cleaned up below.
+        *
+        * Otherwise, we can get into a weird old mess where the directory has C_DELETED,
+        * but it really means C_NOEXISTS because the item was actually removed from the 
+        * catalog.  Then when we try to remove the entry from the catalog later on, it won't
+        * really be there anymore.  
+        */
+       if (vnode_isinuse(vp, 0))  {
+               only_unlink = 1;
        }
 
-       if (!(options & HFSRM_SKIP_RESERVE)) {
-               /*
-                * Reserve some space in the Catalog file.
+       /* Deal with directory hardlinks */
+       if (cp->c_flag & C_HARDLINK) {
+               /* 
+                * Note that if we have a directory which was a hardlink at any point,
+                * its actual directory data is stored in the directory inode in the hidden
+                * directory rather than the leaf element(s) present in the namespace.
+                * 
+                * If there are still other hardlinks to this directory, 
+                * then we'll just eliminate this particular link and the vnode will still exist.
+                * If this is the last link to an empty directory, then we'll open-unlink the 
+                * directory and it will be only tagged with C_DELETED (as opposed to C_NOEXISTS).
+                * 
+                * We could also return EBUSY here. 
                 */
-               if ((error = cat_preflight(hfsmp, CAT_DELETE, &cookie, p))) {
-                       goto out;
-               }
+               
+               return hfs_unlink(hfsmp, dvp, vp, cnp, skip_reserve);
+       }
+       
+       /*
+        * In a few cases, we may want to allow the directory to persist in an
+        * open-unlinked state.  If the directory is being open-unlinked (still has usecount
+        * references), or if it has EAs, or if it was being deleted as part of a rename, 
+        * then we go ahead and move it to the hidden directory. 
+        *
+        * If the directory is being open-unlinked, then we want to keep the catalog entry 
+        * alive so that future EA calls and fchmod/fstat etc. do not cause issues later.
+        * 
+        * If the directory had EAs, then we want to use the open-unlink trick so that the 
+        * EA removal is not done in one giant transaction.  Otherwise, it could cause a panic
+        * due to overflowing the journal.
+        * 
+        * Finally, if it was deleted as part of a rename, we move it to the hidden directory
+        * in order to maintain rename atomicity.  
+        * 
+        * Note that the allow_dirs argument to hfs_removefile specifies that it is
+        * supposed to handle directories for this case.
+     */
+               
+       if (((hfsmp->hfs_attribute_vp != NULL) &&
+           ((cp->c_attr.ca_recflags & kHFSHasAttributesMask) != 0)) ||
+               (only_unlink != 0)) {
+               
+               int ret = hfs_removefile(dvp, vp, cnp, 0, 0, 1, NULL, only_unlink);
+               /* 
+                * Even though hfs_vnop_rename calls vnode_recycle for us on tvp we call 
+                * it here just in case we were invoked by rmdir() on a directory that had 
+                * EAs.  To ensure that we start reclaiming the space as soon as possible,
+                * we call vnode_recycle on the directory.
+                */
+               vnode_recycle(vp);
+               
+               return ret;
+               
+       }
+
+       dcp->c_flag |= C_DIR_MODIFICATION;
+
+#if QUOTA
+       if (hfsmp->hfs_flags & HFS_QUOTAS)
+               (void)hfs_getinoquota(cp);
+#endif
+       if ((error = hfs_start_transaction(hfsmp)) != 0) {
+           goto out;
        }
+       started_tr = 1;
 
        /*
         * Verify the directory is empty (and valid).
@@ -1568,11 +3434,7 @@ hfs_removedir(dvp, vp, cnp, options)
         *  the current directory and thus be
         *  non-empty.)
         */
-       if (cp->c_entries != 0) {
-               error = ENOTEMPTY;
-               goto out;
-       }
-       if ((dcp->c_flags & APPEND) || (cp->c_flags & (IMMUTABLE | APPEND))) {
+       if ((dcp->c_bsdflags & APPEND) || (cp->c_bsdflags & (IMMUTABLE | APPEND))) {
                error = EPERM;
                goto out;
        }
@@ -1580,501 +3442,849 @@ hfs_removedir(dvp, vp, cnp, options)
        /* Remove the entry from the namei cache: */
        cache_purge(vp);
 
-       /* Lock catalog b-tree */
-       error = hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_EXCLUSIVE, p);
-       if (error) goto out;
+       /* 
+        * Protect against a race with rename by using the component
+        * name passed in and parent id from dvp (instead of using 
+        * the cp->c_desc which may have changed).
+        */
+       desc.cd_nameptr = (const u_int8_t *)cnp->cn_nameptr;
+       desc.cd_namelen = cnp->cn_namelen;
+       desc.cd_parentcnid = dcp->c_fileid;
+       desc.cd_cnid = cp->c_cnid;
+       desc.cd_flags = CD_ISDIR;
+       desc.cd_encoding = cp->c_encoding;
+       desc.cd_hint = 0;
+
+       if (!hfs_valid_cnode(hfsmp, dvp, cnp, cp->c_fileid, NULL, &error)) {
+           error = 0;
+           goto out;
+       }
 
-       if (cp->c_entries > 0)
-               panic("hfs_rmdir: attempting to delete a non-empty directory!");
        /* Remove entry from catalog */
-       error = cat_delete(hfsmp, &cp->c_desc, &cp->c_attr);
+       lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG | SFL_ATTRIBUTE | SFL_BITMAP, HFS_EXCLUSIVE_LOCK);
+
+       if (!skip_reserve) {
+               /*
+                * Reserve some space in the Catalog file.
+                */
+               if ((error = cat_preflight(hfsmp, CAT_DELETE, NULL, 0))) {
+                       hfs_systemfile_unlock(hfsmp, lockflags);
+                       goto out;
+               }
+       }
+
+       error = cat_delete(hfsmp, &desc, &cp->c_attr);
+
+       if (!error) {
+               //
+               // if skip_reserve == 1 then we're being called from hfs_vnop_rename() and thus
+               // we don't need to touch the document_id as it's handled by the rename code.
+               // otherwise it's a normal remove and we need to save the document id in the
+               // per thread struct and clear it from the cnode.
+               //
+               struct  doc_tombstone *ut;
+               ut = get_uthread_doc_tombstone();
+               if (!skip_reserve && (cp->c_bsdflags & UF_TRACKED) && should_save_docid_tombstone(ut, vp, cnp)) {
+               
+                       if (ut->t_lastop_document_id) {
+                               clear_tombstone_docid(ut, hfsmp, NULL);
+                       }
+                       save_tombstone(hfsmp, dvp, vp, cnp, 1);
+
+               }
+
+               /* The parent lost a child */
+               if (dcp->c_entries > 0)
+                       dcp->c_entries--;
+               DEC_FOLDERCOUNT(hfsmp, dcp->c_attr);
+               dcp->c_dirchangecnt++;
+               hfs_incr_gencount(dcp);
+
+               dcp->c_touch_chgtime = TRUE;
+               dcp->c_touch_modtime = TRUE;
+               dcp->c_flag |= C_MODIFIED;
 
-       /* Unlock catalog b-tree */
-       (void) hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_RELEASE, p);
-       if (error) goto out;
+               hfs_update(dcp->c_vp, 0);
+       }
+
+       hfs_systemfile_unlock(hfsmp, lockflags);
+
+       if (error)
+               goto out;
 
 #if QUOTA
-       (void)hfs_chkiq(cp, -1, NOCRED, 0);
+       if (hfsmp->hfs_flags & HFS_QUOTAS)
+               (void)hfs_chkiq(cp, -1, NOCRED, 0);
 #endif /* QUOTA */
 
-       /* The parent lost a child */
-       if (dcp->c_entries > 0)
-               dcp->c_entries--;
-       if (dcp->c_nlink > 0)
-               dcp->c_nlink--;
-       dcp->c_flag |= C_CHANGE | C_UPDATE;
-       tv = time;
-       (void) VOP_UPDATE(dvp, &tv, &tv, 0);
-       HFS_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
-
        hfs_volupdate(hfsmp, VOL_RMDIR, (dcp->c_cnid == kHFSRootFolderID));
 
-       cp->c_mode = 0;  /* Makes the vnode go away...see inactive */
+       /* Mark C_NOEXISTS since the catalog entry is now gone */
        cp->c_flag |= C_NOEXISTS;
+
 out:
-       if (!(options & HFSRM_PARENT_LOCKED)) {
-               vput(dvp);
+       dcp->c_flag &= ~C_DIR_MODIFICATION;
+       wakeup((caddr_t)&dcp->c_flag);
+
+       if (started_tr) { 
+           hfs_end_transaction(hfsmp);
+       }
+
+       return (error);
+}
+
+
+/*
+ * Remove a file or link.
+ */
+int
+hfs_vnop_remove(ap)
+       struct vnop_remove_args /* {
+               struct vnode *a_dvp;
+               struct vnode *a_vp;
+               struct componentname *a_cnp;
+               int a_flags;
+               vfs_context_t a_context;
+       } */ *ap;
+{
+       struct vnode *dvp = ap->a_dvp;
+       struct vnode *vp = ap->a_vp;
+       struct cnode *dcp = VTOC(dvp);
+       struct cnode *cp;
+       struct vnode *rvp = NULL;
+       int error=0, recycle_rsrc=0;
+       int recycle_vnode = 0;
+       uint32_t rsrc_vid = 0;
+       time_t orig_ctime;
+
+       if (dvp == vp) {
+               return (EINVAL);
+       }
+
+       orig_ctime = VTOC(vp)->c_ctime;
+       if (!vnode_isnamedstream(vp) && ((ap->a_flags & VNODE_REMOVE_SKIP_NAMESPACE_EVENT) == 0)) {
+               error = check_for_tracked_file(vp, orig_ctime, NAMESPACE_HANDLER_DELETE_OP, NULL);
+               if (error) {
+                       // XXXdbg - decide on a policy for handling namespace handler failures!
+                       // for now we just let them proceed.
+               }               
+       }
+       error = 0;
+
+       cp = VTOC(vp);
+
+relock:
+
+       hfs_lock_truncate(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
+
+       if ((error = hfs_lockpair(dcp, cp, HFS_EXCLUSIVE_LOCK))) {
+               hfs_unlock_truncate(cp, HFS_LOCK_DEFAULT);
+               if (rvp) {
+                       vnode_put (rvp);
+               }       
+               return (error);
+       }
+       //
+       // if the item is tracked but doesn't have a document_id, assign one and generate an fsevent for it
+       //
+       if ((cp->c_bsdflags & UF_TRACKED) && ((struct FndrExtendedDirInfo *)((char *)&cp->c_attr.ca_finderinfo + 16))->document_id == 0) {
+               uint32_t newid;
+
+               hfs_unlockpair(dcp, cp);
+
+               if (hfs_generate_document_id(VTOHFS(vp), &newid) == 0) {
+                       hfs_lockpair(dcp, cp, HFS_EXCLUSIVE_LOCK);
+                       ((struct FndrExtendedDirInfo *)((char *)&cp->c_attr.ca_finderinfo + 16))->document_id = newid;
+#if CONFIG_FSE
+                       add_fsevent(FSE_DOCID_CHANGED, vfs_context_current(),
+                                   FSE_ARG_DEV,   VTOHFS(vp)->hfs_raw_dev,
+                                   FSE_ARG_INO,   (ino64_t)0,             // src inode #
+                                   FSE_ARG_INO,   (ino64_t)cp->c_fileid,  // dst inode #
+                                   FSE_ARG_INT32, newid,
+                                   FSE_ARG_DONE);
+#endif
+               } else {
+                       // XXXdbg - couldn't get a new docid... what to do?  can't really fail the rm...
+                       hfs_lockpair(dcp, cp, HFS_EXCLUSIVE_LOCK);
+               }
        }
-       HFS_KNOTE(vp, NOTE_DELETE);
-       vput(vp);
+       
+       /*
+        * Lazily respond to determining if there is a valid resource fork
+        * vnode attached to 'cp' if it is a regular file or symlink.  
+        * If the vnode does not exist, then we may proceed without having to
+        * create it.
+        *
+        * If, however, it does exist, then we need to acquire an iocount on the
+        * vnode after acquiring its vid.  This ensures that if we have to do I/O
+        * against it, it can't get recycled from underneath us in the middle
+        * of this call.
+        *
+        * Note: this function may be invoked for directory hardlinks, so just skip these
+        * steps if 'vp' is a directory.
+        */
+
+       if ((vp->v_type == VLNK) || (vp->v_type == VREG)) {
+               if ((cp->c_rsrc_vp) && (rvp == NULL)) {
+                       /* We need to acquire the rsrc vnode */
+                       rvp = cp->c_rsrc_vp;
+                       rsrc_vid = vnode_vid (rvp);
+
+                       /* Unlock everything to acquire iocount on the rsrc vnode */    
+                       hfs_unlock_truncate (cp, HFS_LOCK_DEFAULT);
+                       hfs_unlockpair (dcp, cp);
+                       /* Use the vid to maintain identity on rvp */
+                       if (vnode_getwithvid(rvp, rsrc_vid)) {
+                               /*
+                                * If this fails, then it was recycled or 
+                                * reclaimed in the interim.  Reset fields and
+                                * start over.
+                                */
+                               rvp = NULL;
+                               rsrc_vid = 0;
+                       }
+                       goto relock;
+               }
+       }
+
+       /* 
+        * Check to see if we raced rmdir for the parent directory 
+        * hfs_removefile already checks for a race on vp/cp
+        */
+       if (dcp->c_flag & (C_DELETED | C_NOEXISTS)) {
+               error = ENOENT;
+               goto rm_done;   
+       }
+
+       error = hfs_removefile(dvp, vp, ap->a_cnp, ap->a_flags, 0, 0, NULL, 0);
+       
+       /*
+        * If the remove succeeded in deleting the file, then we may need to mark
+        * the resource fork for recycle so that it is reclaimed as quickly
+        * as possible.  If it were not recycled quickly, then this resource fork
+        * vnode could keep a v_parent reference on the data fork, which prevents it
+        * from going through reclaim (by giving it extra usecounts), except in the force-
+        * unmount case.  
+        * 
+        * However, a caveat:  we need to continue to supply resource fork
+        * access to open-unlinked files even if the resource fork is not open.  This is
+        * a requirement for the compressed files work.  Luckily, hfs_vgetrsrc will handle
+        * this already if the data fork has been re-parented to the hidden directory.
+        * 
+        * As a result, all we really need to do here is mark the resource fork vnode
+        * for recycle.  If it goes out of core, it can be brought in again if needed.  
+        * If the cnode was instead marked C_NOEXISTS, then there wouldn't be any 
+        * more work.
+        */
+       if (error == 0) {
+               hfs_hotfile_deleted(vp);
+               
+               if (rvp) {
+               recycle_rsrc = 1;
+               }
+               /* 
+                * If the target was actually removed from the catalog schedule it for
+                * full reclamation/inactivation.  We hold an iocount on it so it should just
+                * get marked with MARKTERM
+                */
+               if (cp->c_flag & C_NOEXISTS) {
+                       recycle_vnode = 1;
+               }
+       }
+
+
+       /*
+        * Drop the truncate lock before unlocking the cnode
+        * (which can potentially perform a vnode_put and
+        * recycle the vnode which in turn might require the
+        * truncate lock)
+        */
+rm_done:
+       hfs_unlockpair(dcp, cp);
+       hfs_unlock_truncate(cp, HFS_LOCK_DEFAULT);
 
-       if (!(options & HFSRM_SKIP_RESERVE)) {
-               cat_postflight(hfsmp, &cookie, p);
-       }
-       // XXXdbg
-       if (started_tr) { 
-           journal_end_transaction(hfsmp->jnl);
+       if (recycle_rsrc) {
+               /* inactive or reclaim on rvp will clean up the blocks from the rsrc fork */
+               vnode_recycle(rvp);
+       } 
+       if (recycle_vnode) {
+               vnode_recycle (vp);
        }
-       if (grabbed_lock) {
-               hfs_global_shared_lock_release(hfsmp);
+
+       if (rvp) {
+               /* drop iocount on rsrc fork, was obtained at beginning of fxn */
+               vnode_put(rvp);
        }
 
        return (error);
 }
 
-/*
 
-#% remove      dvp     L U U
-#% remove      vp      L U U
-#
- vop_remove {
-     IN WILLRELE struct vnode *dvp;
-     IN WILLRELE struct vnode *vp;
-     IN struct componentname *cnp;
+int
+hfs_removefile_callback(struct buf *bp, void *hfsmp) {
 
-     */
+        if ( !(buf_flags(bp) & B_META))
+               panic("hfs: symlink bp @ %p is not marked meta-data!\n", bp);
+       /*
+        * it's part of the current transaction, kill it.
+        */
+       journal_kill_block(((struct hfsmount *)hfsmp)->jnl, bp);
 
-static int
-hfs_remove(ap)
-       struct vop_remove_args /* {
-               struct vnode *a_dvp;
-               struct vnode *a_vp;
-               struct componentname *a_cnp;
-       } */ *ap;
-{
-       return (hfs_removefile(ap->a_dvp, ap->a_vp, ap->a_cnp, 0));
+       return (BUF_CLAIMED);
 }
 
-
-
 /*
  * hfs_removefile
  *
- * Similar to hfs_remove except there are additional options.
+ * Similar to hfs_vnop_remove except there are additional options.
+ * This function may be used to remove directories if they have
+ * lots of EA's -- note the 'allow_dirs' argument.
+ *
+ * This function is able to delete blocks & fork data for the resource
+ * fork even if it does not exist in core (and have a backing vnode).  
+ * It should infer the correct behavior based on the number of blocks
+ * in the cnode and whether or not the resource fork pointer exists or 
+ * not.  As a result, one only need pass in the 'vp' corresponding to the
+ * data fork of this file (or main vnode in the case of a directory).  
+ * Passing in a resource fork will result in an error.
+ *
+ * Because we do not create any vnodes in this function, we are not at 
+ * risk of deadlocking against ourselves by double-locking.
+ *
+ * Requires cnode and truncate locks to be held.
  */
-static int
-hfs_removefile(dvp, vp, cnp, options)
-       struct vnode *dvp;
-       struct vnode *vp;
-       struct componentname *cnp;
-       int options;
+int
+hfs_removefile(struct vnode *dvp, struct vnode *vp, struct componentname *cnp,
+               int flags, int skip_reserve, int allow_dirs, 
+                          __unused struct vnode *rvp, int only_unlink)
 {
-       struct vnode *rvp = NULL;
        struct cnode *cp;
        struct cnode *dcp;
+       struct vnode *rsrc_vp = NULL;
        struct hfsmount *hfsmp;
-       struct proc *p = cnp->cn_proc;
+       struct cat_desc desc;
+       struct timeval tv;
        int dataforkbusy = 0;
        int rsrcforkbusy = 0;
-       int truncated = 0;
-       struct timeval tv;
-       cat_cookie_t cookie = {0};
+       int lockflags;
        int error = 0;
-       int started_tr = 0, grabbed_lock = 0;
-       int refcount, isbigfile = 0;
-
-       /* Directories should call hfs_rmdir! */
-       if (vp->v_type == VDIR) {
-               error = EISDIR;
-               goto out;
-       }
+       int started_tr = 0;
+       int isbigfile = 0, defer_remove=0, isdir=0;
+       int update_vh = 0;
 
        cp = VTOC(vp);
        dcp = VTOC(dvp);
        hfsmp = VTOHFS(vp);
-       
-       if (cp->c_parentcnid != dcp->c_cnid) {
-               error = EINVAL;
-               goto out;
+
+       /* Check if we lost a race post lookup. */
+       if (cp->c_flag & (C_NOEXISTS | C_DELETED)) {
+               return (0);
+       }
+
+       if (!hfs_valid_cnode(hfsmp, dvp, cnp, cp->c_fileid, NULL, &error)) {
+           return 0;
        }
 
        /* Make sure a remove is permitted */
-       if ((cp->c_flags & (IMMUTABLE | APPEND)) ||
-           (VTOC(dvp)->c_flags & APPEND) ||
-           VNODE_IS_RSRC(vp)) {
-               error = EPERM;
-               goto out;
+       if (VNODE_IS_RSRC(vp)) {
+               return (EPERM);
+       }
+       else {
+               /* 
+                * We know it's a data fork.
+                * Probe the cnode to see if we have a valid resource fork
+                * in hand or not.
+                */
+               rsrc_vp = cp->c_rsrc_vp;
+       }
+
+       /* Don't allow deleting the journal or journal_info_block. */
+       if (hfs_is_journal_file(hfsmp, cp)) {
+               return (EPERM);
        }
 
        /*
-        * Aquire a vnode for a non-empty resource fork.
-        * (needed for VOP_TRUNCATE)
+        * Hard links require special handling.
         */
-       if (cp->c_blocks - VTOF(vp)->ff_blocks) {
-               error = hfs_vgetrsrc(hfsmp, vp, &rvp, p);
-               if (error)
-                       goto out;
+       if (cp->c_flag & C_HARDLINK) {
+               if ((flags & VNODE_REMOVE_NODELETEBUSY) && vnode_isinuse(vp, 0)) {
+                       return (EBUSY);
+               } else {
+                       /* A directory hard link with a link count of one is 
+                        * treated as a regular directory.  Therefore it should 
+                        * only be removed using rmdir().
+                        */
+                       if ((vnode_isdir(vp) == 1) && (cp->c_linkcount == 1) && 
+                           (allow_dirs == 0)) {
+                               return (EPERM);
+                       }
+                       return hfs_unlink(hfsmp, dvp, vp, cnp, skip_reserve);
+               }
        }
 
-       // XXXdbg - don't allow deleting the journal or journal_info_block
-       if (hfsmp->jnl && cp->c_datafork) {
-               struct HFSPlusExtentDescriptor *extd;
-
-               extd = &cp->c_datafork->ff_extents[0];
-               if (extd->startBlock == HFSTOVCB(hfsmp)->vcbJinfoBlock || extd->startBlock == hfsmp->jnl_start) {
-                       error = EPERM;
-                       goto out;
-               }
+       /* Directories should call hfs_rmdir! (unless they have a lot of attributes) */
+       if (vnode_isdir(vp)) {
+               if (allow_dirs == 0)
+                       return (EPERM);  /* POSIX */
+               isdir = 1;
+       }
+       /* Sanity check the parent ids. */
+       if ((cp->c_parentcnid != hfsmp->hfs_private_desc[FILE_HARDLINKS].cd_cnid) &&
+           (cp->c_parentcnid != dcp->c_fileid)) {
+               return (EINVAL);
        }
 
+       dcp->c_flag |= C_DIR_MODIFICATION;
+
+       // this guy is going away so mark him as such
+       cp->c_flag |= C_DELETED;
+
+
+       /* Remove our entry from the namei cache. */
+       cache_purge(vp);
+       
        /*
-        * Check if this file is being used.
-        *
-        * The namei done for the remove took a reference on the
-        * vnode (vp).  And we took a ref on the resource vnode (rvp).
-        * Hence set 1 in the tookref parameter of ubc_isinuse().
+        * If the caller was operating on a file (as opposed to a 
+        * directory with EAs), then we need to figure out
+        * whether or not it has a valid resource fork vnode.
+        * 
+        * If there was a valid resource fork vnode, then we need
+        * to use hfs_truncate to eliminate its data.  If there is
+        * no vnode, then we hold the cnode lock which would
+        * prevent it from being created.  As a result, 
+        * we can use the data deletion functions which do not
+        * require that a cnode/vnode pair exist.
         */
-       if (VTOC(vp)->c_flag & C_VPREFHELD) {
-           refcount = 2;
-       } else {
-           refcount = 1;
+
+       /* Check if this file is being used. */
+       if (isdir == 0) {
+               dataforkbusy = vnode_isinuse(vp, 0);
+               /*  
+                * At this point, we know that 'vp' points to the 
+                * a data fork because we checked it up front. And if 
+                * there is no rsrc fork, rsrc_vp will be NULL.
+                */
+               if (rsrc_vp && (cp->c_blocks - VTOF(vp)->ff_blocks)) {
+                       rsrcforkbusy = vnode_isinuse(rsrc_vp, 0);
+               }
        }
-       if (UBCISVALID(vp) && ubc_isinuse(vp, refcount))
-               dataforkbusy = 1;
-       if (rvp && UBCISVALID(rvp) && ubc_isinuse(rvp, 1))
-               rsrcforkbusy = 1;
+       
+       /* Check if we have to break the deletion into multiple pieces. */
+       if (isdir == 0)
+               isbigfile = cp->c_datafork->ff_size >= HFS_BIGFILE_SIZE;
 
-       // need this to check if we have to break the deletion 
-       // into multiple pieces
-       isbigfile = (VTOC(vp)->c_datafork->ff_size >= HFS_BIGFILE_SIZE);
+       /* Check if the file has xattrs.  If it does we'll have to delete them in
+          individual transactions in case there are too many */
+       if ((hfsmp->hfs_attribute_vp != NULL) &&
+           (cp->c_attr.ca_recflags & kHFSHasAttributesMask) != 0) {
+           defer_remove = 1;
+       }
+       
+       /* If we are explicitly told to only unlink item and move to hidden dir, then do it */
+       if (only_unlink) {
+               defer_remove = 1;
+       }
 
        /*
         * Carbon semantics prohibit deleting busy files.
-        * (enforced when NODELETEBUSY is requested)
+        * (enforced when VNODE_REMOVE_NODELETEBUSY is requested)
         */
-       if ((dataforkbusy || rsrcforkbusy) &&
-           ((cnp->cn_flags & NODELETEBUSY) ||
-            (hfsmp->hfs_privdir_desc.cd_cnid == 0))) {
-               error = EBUSY;
-               goto out;
+       if (dataforkbusy || rsrcforkbusy) {
+               if ((flags & VNODE_REMOVE_NODELETEBUSY) ||
+                   (hfsmp->hfs_private_desc[FILE_HARDLINKS].cd_cnid == 0)) {
+                       error = EBUSY;
+                       goto out;
+               }
        }
 
 #if QUOTA
-       (void)hfs_getinoquota(cp);
+       if (hfsmp->hfs_flags & HFS_QUOTAS)
+               (void)hfs_getinoquota(cp);
 #endif /* QUOTA */
-
-       // XXXdbg
-       hfs_global_shared_lock_acquire(hfsmp);
-       grabbed_lock = 1;
-       if (hfsmp->jnl) {
-           if ((error = journal_start_transaction(hfsmp->jnl)) != 0) {
-                       goto out;
-           }
-           started_tr = 1;
-       }
-
-       if (!(options & HFSRM_SKIP_RESERVE)) {
+       
+       /* 
+        * Do a ubc_setsize to indicate we need to wipe contents if:
+        *  1) item is a regular file.
+        *  2) Neither fork is busy AND we are not told to unlink this. 
+        *
+        * We need to check for the defer_remove since it can be set without 
+        * having a busy data or rsrc fork   
+        */
+       if (isdir == 0 && (!dataforkbusy || !rsrcforkbusy) && (defer_remove == 0)) {
                /*
-                * Reserve some space in the Catalog file.
+                * A ubc_setsize can cause a pagein so defer it
+                * until after the cnode lock is dropped.  The
+                * cnode lock cannot be dropped/reacquired here
+                * since we might already hold the journal lock.
                 */
-               if ((error = cat_preflight(hfsmp, CAT_DELETE, &cookie, p))) {
-                       goto out;
+               if (!dataforkbusy && cp->c_datafork->ff_blocks && !isbigfile) {
+                       cp->c_flag |= C_NEED_DATA_SETSIZE;
+               }
+               if (!rsrcforkbusy && rsrc_vp) {
+                       cp->c_flag |= C_NEED_RSRC_SETSIZE;
                }
        }
 
-       /* Remove our entry from the namei cache. */
-       cache_purge(vp);
+       if ((error = hfs_start_transaction(hfsmp)) != 0) {
+           goto out;
+       }
+       started_tr = 1;
 
        // XXXdbg - if we're journaled, kill any dirty symlink buffers 
-       if (hfsmp->jnl && vp->v_type == VLNK && vp->v_dirtyblkhd.lh_first) {
-           struct buf *bp, *nbp;
-
-         recheck:
-           for (bp=vp->v_dirtyblkhd.lh_first; bp; bp=nbp) {
-                       nbp = bp->b_vnbufs.le_next;
-                       
-                       if ((bp->b_flags & B_BUSY)) {
-                               // if it was busy, someone else must be dealing
-                               // with it so just move on.
-                               continue;
-                       }
-
-                       if (!(bp->b_flags & B_META)) {
-                               panic("hfs: symlink bp @ 0x%x is not marked meta-data!\n", bp);
-                       }
-
-                       // if it's part of the current transaction, kill it.
-                       if (bp->b_flags & B_LOCKED) {
-                               bremfree(bp);
-                               bp->b_flags |= B_BUSY;
-                               journal_kill_block(hfsmp->jnl, bp);
-                               goto recheck;
-                       }
-           }
+       if (hfsmp->jnl && vnode_islnk(vp) && (defer_remove == 0)) {
+               buf_iterate(vp, hfs_removefile_callback, BUF_SKIP_NONLOCKED, (void *)hfsmp);
        }
-       // XXXdbg
 
        /*
-        * Truncate any non-busy forks.  Busy forks will
-        * get trucated when their vnode goes inactive.
+        * Prepare to truncate any non-busy forks.  Busy forks will
+        * get truncated when their vnode goes inactive.
+        * Note that we will only enter this region if we
+        * can avoid creating an open-unlinked file.  If 
+        * either region is busy, we will have to create an open
+        * unlinked file.
         *
-        * (Note: hard links are truncated in VOP_INACTIVE)
+        * Since we are deleting the file, we need to stagger the runtime
+        * modifications to do things in such a way that a crash won't 
+        * result in us getting overlapped extents or any other 
+        * bad inconsistencies.  As such, we call prepare_release_storage
+        * which updates the UBC, updates quota information, and releases
+        * any loaned blocks that belong to this file.  No actual 
+        * truncation or bitmap manipulation is done until *AFTER*
+        * the catalog record is removed. 
         */
-       if ((cp->c_flag & C_HARDLINK) == 0) {
-               int mode = cp->c_mode;
-
+       if (isdir == 0 && (!dataforkbusy && !rsrcforkbusy) && (only_unlink == 0)) {
+               
                if (!dataforkbusy && !isbigfile && cp->c_datafork->ff_blocks != 0) {
-                       cp->c_mode = 0;  /* Suppress VOP_UPDATES */
-                       error = VOP_TRUNCATE(vp, (off_t)0, IO_NDELAY, NOCRED, p);
-                       cp->c_mode = mode;
-                       if (error)
+                       
+                       error = hfs_prepare_release_storage (hfsmp, vp);
+                       if (error) {
                                goto out;
-                       truncated = 1;
+                       }
+                       update_vh = 1;
                }
-               if (!rsrcforkbusy && rvp) {
-                       cp->c_mode = 0;            /* Suppress VOP_UPDATES */
-                       error = VOP_TRUNCATE(rvp, (off_t)0, IO_NDELAY, NOCRED, p);
-                       cp->c_mode = mode;
-                       if (error)
+               
+               /*
+                * If the resource fork vnode does not exist, we can skip this step.
+                */
+               if (!rsrcforkbusy && rsrc_vp) {
+                       error = hfs_prepare_release_storage (hfsmp, rsrc_vp);
+                       if (error) {
                                goto out;
-                       truncated = 1;
+                       }
+                       update_vh = 1;
                }
        }
-       /*
-        * There are 3 remove cases to consider:
-        *   1. File is a hardlink    ==> remove the link
-        *   2. File is busy (in use) ==> move/rename the file
-        *   3. File is not in use    ==> remove the file
-        */
-
-       if (cp->c_flag & C_HARDLINK) {
-               struct cat_desc desc;
-
-               if ((cnp->cn_flags & HASBUF) == 0 ||
-                   cnp->cn_nameptr[0] == '\0') {
-                       error = ENOENT; /* name missing! */
-                       goto out;
-               }
-
-               /* Setup a descriptor for the link */
-               bzero(&desc, sizeof(desc));
-               desc.cd_nameptr = cnp->cn_nameptr;
-               desc.cd_namelen = cnp->cn_namelen;
-               desc.cd_parentcnid = dcp->c_cnid;
-               /* XXX - if cnid is out of sync then the wrong thread rec will get deleted. */
-               desc.cd_cnid = cp->c_cnid;
-
-               /* Lock catalog b-tree */
-               error = hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_EXCLUSIVE, p);
-               if (error)
-                       goto out;
-
-               /* Delete the link record */
-               error = cat_delete(hfsmp, &desc, &cp->c_attr);
-
-               if ((error == 0) && (--cp->c_nlink < 1)) {
-                       char inodename[32];
-                       char delname[32];
-                       struct cat_desc to_desc;
-                       struct cat_desc from_desc;
-
-                       /*
-                        * This is now esentially an open deleted file.
-                        * Rename it to reflect this state which makes
-                        * orphan file cleanup easier (see hfs_remove_orphans).
-                        * Note: a rename failure here is not fatal.
-                        */     
-                       MAKE_INODE_NAME(inodename, cp->c_rdev);
-                       bzero(&from_desc, sizeof(from_desc));
-                       from_desc.cd_nameptr = inodename;
-                       from_desc.cd_namelen = strlen(inodename);
-                       from_desc.cd_parentcnid = hfsmp->hfs_privdir_desc.cd_cnid;
-                       from_desc.cd_flags = 0;
-                       from_desc.cd_cnid = cp->c_fileid;
-
-                       MAKE_DELETED_NAME(delname, cp->c_fileid);               
-                       bzero(&to_desc, sizeof(to_desc));
-                       to_desc.cd_nameptr = delname;
-                       to_desc.cd_namelen = strlen(delname);
-                       to_desc.cd_parentcnid = hfsmp->hfs_privdir_desc.cd_cnid;
-                       to_desc.cd_flags = 0;
-                       to_desc.cd_cnid = cp->c_fileid;
        
-                       (void) cat_rename(hfsmp, &from_desc, &hfsmp->hfs_privdir_desc,
-                                         &to_desc, (struct cat_desc *)NULL);
-                       cp->c_flag |= C_DELETED;
-               }
-
-               /* Unlock the Catalog */
-               (void) hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_RELEASE, p);
-
-               if (error != 0)
-                       goto out;
-
-               cp->c_flag |= C_CHANGE;
-               tv = time;
-               (void) VOP_UPDATE(vp, &tv, &tv, 0);
-
-               hfs_volupdate(hfsmp, VOL_RMFILE, (dcp->c_cnid == kHFSRootFolderID));
+       /* 
+        * Protect against a race with rename by using the component
+        * name passed in and parent id from dvp (instead of using 
+        * the cp->c_desc which may have changed).   Also, be aware that
+        * because we allow directories to be passed in, we need to special case
+        * this temporary descriptor in case we were handed a directory.
+        */
+       if (isdir) {
+               desc.cd_flags = CD_ISDIR;
+       }
+       else {
+               desc.cd_flags = 0;
+       }
+       desc.cd_encoding = cp->c_desc.cd_encoding;
+       desc.cd_nameptr = (const u_int8_t *)cnp->cn_nameptr;
+       desc.cd_namelen = cnp->cn_namelen;
+       desc.cd_parentcnid = dcp->c_fileid;
+       desc.cd_hint = cp->c_desc.cd_hint;
+       desc.cd_cnid = cp->c_cnid;
+       microtime(&tv);
 
-       } else if (dataforkbusy || rsrcforkbusy || isbigfile) {
+       /*
+        * There are two cases to consider:
+        *  1. File/Dir is busy/big/defer_remove ==> move/rename the file/dir
+        *  2. File is not in use ==> remove the file
+        * 
+        * We can get a directory in case 1 because it may have had lots of attributes,
+        * which need to get removed here.
+        */
+       if (dataforkbusy || rsrcforkbusy || isbigfile || defer_remove) {
                char delname[32];
                struct cat_desc to_desc;
                struct cat_desc todir_desc;
 
                /*
-                * Orphan this file (move to hidden directory).
+                * Orphan this file or directory (move to hidden directory).
+                * Again, we need to take care that we treat directories as directories,
+                * and files as files.  Because directories with attributes can be passed in
+                * check to make sure that we have a directory or a file before filling in the 
+                * temporary descriptor's flags.  We keep orphaned directories AND files in
+                * the FILE_HARDLINKS private directory since we're generalizing over all
+                * orphaned filesystem objects.
                 */
                bzero(&todir_desc, sizeof(todir_desc));
                todir_desc.cd_parentcnid = 2;
 
-               MAKE_DELETED_NAME(delname, cp->c_fileid);               
+               MAKE_DELETED_NAME(delname, sizeof(delname), cp->c_fileid);
                bzero(&to_desc, sizeof(to_desc));
-               to_desc.cd_nameptr = delname;
+               to_desc.cd_nameptr = (const u_int8_t *)delname;
                to_desc.cd_namelen = strlen(delname);
-               to_desc.cd_parentcnid = hfsmp->hfs_privdir_desc.cd_cnid;
-               to_desc.cd_flags = 0;
+               to_desc.cd_parentcnid = hfsmp->hfs_private_desc[FILE_HARDLINKS].cd_cnid;
+               if (isdir) {
+                       to_desc.cd_flags = CD_ISDIR;
+               }
+               else {
+                       to_desc.cd_flags = 0;
+               }
                to_desc.cd_cnid = cp->c_cnid;
 
-               /* Lock catalog b-tree */
-               error = hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_EXCLUSIVE, p);
-               if (error)
-                       goto out;
+               lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG, HFS_EXCLUSIVE_LOCK);
+               if (!skip_reserve) {
+                       if ((error = cat_preflight(hfsmp, CAT_RENAME, NULL, 0))) {
+                               hfs_systemfile_unlock(hfsmp, lockflags);
+                               goto out;
+                       }
+               }
 
-               error = cat_rename(hfsmp, &cp->c_desc, &todir_desc,
+               error = cat_rename(hfsmp, &desc, &todir_desc,
                                &to_desc, (struct cat_desc *)NULL);
 
-               // XXXdbg - only bump this count if we were successful
                if (error == 0) {
-                       hfsmp->hfs_privdir_attr.ca_entries++;
-               }
-               (void)cat_update(hfsmp, &hfsmp->hfs_privdir_desc,
-                               &hfsmp->hfs_privdir_attr, NULL, NULL);
-
-               /* Unlock the Catalog */
-               (void) hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_RELEASE, p);
-               if (error) goto out;
-
-               cp->c_flag |= C_CHANGE | C_DELETED | C_NOEXISTS;
-               --cp->c_nlink;
-               tv = time;
-               (void) VOP_UPDATE(vp, &tv, &tv, 0);
+                       hfsmp->hfs_private_attr[FILE_HARDLINKS].ca_entries++;
+                       if (isdir == 1) {
+                               INC_FOLDERCOUNT(hfsmp, hfsmp->hfs_private_attr[FILE_HARDLINKS]);
+                       }
+                       (void) cat_update(hfsmp, &hfsmp->hfs_private_desc[FILE_HARDLINKS],
+                                         &hfsmp->hfs_private_attr[FILE_HARDLINKS], NULL, NULL);
+
+                       /* Update the parent directory */
+                       if (dcp->c_entries > 0)
+                               dcp->c_entries--;
+                       if (isdir == 1) {
+                               DEC_FOLDERCOUNT(hfsmp, dcp->c_attr);
+                       }
+                       dcp->c_dirchangecnt++;
+                       hfs_incr_gencount(dcp);
 
-       } else /* Not busy */ {
+                       dcp->c_ctime = tv.tv_sec;
+                       dcp->c_mtime = tv.tv_sec;
+                       (void) cat_update(hfsmp, &dcp->c_desc, &dcp->c_attr, NULL, NULL);
 
-               if (cp->c_blocks > 0) {
-#if 0
-                       panic("hfs_remove: attempting to delete a non-empty file!");
-#else
-                       printf("hfs_remove: attempting to delete a non-empty file %s\n",
-                               cp->c_desc.cd_nameptr);
-                       error = EBUSY;
-                       goto out;
-#endif
+                       /* Update the file or directory's state */
+                       cp->c_flag |= C_DELETED;
+                       cp->c_ctime = tv.tv_sec;
+                       --cp->c_linkcount;
+                       (void) cat_update(hfsmp, &to_desc, &cp->c_attr, NULL, NULL);
                }
-
-               /* Lock catalog b-tree */
-               error = hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_EXCLUSIVE, p);
+               hfs_systemfile_unlock(hfsmp, lockflags);
                if (error)
                        goto out;
 
-               error = cat_delete(hfsmp, &cp->c_desc, &cp->c_attr);
+       } 
+       else {
+               /*
+                * Nobody is using this item; we can safely remove everything.
+                */
+               struct filefork *temp_rsrc_fork = NULL;
+#if QUOTA
+               off_t savedbytes;
+               int blksize = hfsmp->blockSize;
+#endif
+               u_int32_t fileid = cp->c_fileid;
+       
+               /* 
+                * Figure out if we need to read the resource fork data into 
+                * core before wiping out the catalog record.  
+                *
+                * 1) Must not be a directory
+                * 2) cnode's c_rsrcfork ptr must be NULL.
+                * 3) rsrc fork must have actual blocks 
+                */
+               if ((isdir == 0) && (cp->c_rsrcfork == NULL) && 
+                               (cp->c_blocks - VTOF(vp)->ff_blocks)) {
+                       /*
+                        * The resource fork vnode & filefork did not exist.
+                        * Create a temporary one for use in this function only. 
+                        */
+                       MALLOC_ZONE (temp_rsrc_fork, struct filefork *, sizeof (struct filefork), M_HFSFORK, M_WAITOK);
+                       bzero(temp_rsrc_fork, sizeof(struct filefork));
+                       temp_rsrc_fork->ff_cp = cp;
+                       rl_init(&temp_rsrc_fork->ff_invalidranges);
+               }       
+
+               lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG | SFL_ATTRIBUTE | SFL_BITMAP, HFS_EXCLUSIVE_LOCK);
 
-               if (error && error != ENXIO && error != ENOENT && truncated) {
-                       if ((cp->c_datafork && cp->c_datafork->ff_size != 0) ||
-                               (cp->c_rsrcfork && cp->c_rsrcfork->ff_size != 0)) {
-                               panic("hfs: remove: couldn't delete a truncated file! (%d, data sz %lld; rsrc sz %lld)",
-                                         error, cp->c_datafork->ff_size, cp->c_rsrcfork->ff_size);
-                       } else {
-                               printf("hfs: remove: strangely enough, deleting truncated file %s (%d) got err %d\n",
-                                          cp->c_desc.cd_nameptr, cp->c_attr.ca_fileid, error);
+               /* Look up the resource fork first, if necessary */
+               if (temp_rsrc_fork) {
+                       error = cat_lookup (hfsmp, &desc, 1, 0, (struct cat_desc*) NULL, 
+                                       (struct cat_attr*) NULL, &temp_rsrc_fork->ff_data, NULL);
+                       if (error) {
+                               FREE_ZONE (temp_rsrc_fork, sizeof(struct filefork), M_HFSFORK);
+                               hfs_systemfile_unlock (hfsmp, lockflags);
+                               goto out;
                        }
                }
 
-               /* Unlock the Catalog */
-               (void) hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_RELEASE, p);
-               if (error) goto out;
+               if (!skip_reserve) {
+                       if ((error = cat_preflight(hfsmp, CAT_DELETE, NULL, 0))) {
+                               if (temp_rsrc_fork) {
+                                       FREE_ZONE (temp_rsrc_fork, sizeof(struct filefork), M_HFSFORK);
+                               }
+                               hfs_systemfile_unlock(hfsmp, lockflags);
+                               goto out;
+                       }
+               }
+               
+               error = cat_delete(hfsmp, &desc, &cp->c_attr);
+               
+               if (error && error != ENXIO && error != ENOENT) {
+                       printf("hfs_removefile: deleting file %s (id=%d) vol=%s err=%d\n",
+                                  cp->c_desc.cd_nameptr, cp->c_attr.ca_fileid, hfsmp->vcbVN, error);
+               }
+               
+               if (error == 0) {
+                       /* Update the parent directory */
+                       if (dcp->c_entries > 0)
+                               dcp->c_entries--;
+                       dcp->c_dirchangecnt++;
+                       hfs_incr_gencount(dcp);
+
+                       dcp->c_ctime = tv.tv_sec;
+                       dcp->c_mtime = tv.tv_sec;
+                       (void) cat_update(hfsmp, &dcp->c_desc, &dcp->c_attr, NULL, NULL);
+               }
+               hfs_systemfile_unlock(hfsmp, lockflags);
 
+               if (error) {
+                       if (temp_rsrc_fork) {
+                               FREE_ZONE (temp_rsrc_fork, sizeof(struct filefork), M_HFSFORK);
+                       }
+                       goto out;
+               }
+               
+               /* 
+                * Now that we've wiped out the catalog record, the file effectively doesn't
+                * exist anymore. So update the quota records to reflect the loss of the 
+                * data fork and the resource fork. 
+                */
 #if QUOTA
-               (void)hfs_chkiq(cp, -1, NOCRED, 0);
-#endif /* QUOTA */
+               if (cp->c_datafork->ff_blocks > 0) {
+                       savedbytes = ((off_t)cp->c_datafork->ff_blocks * (off_t)blksize);
+                       (void) hfs_chkdq(cp, (int64_t)-(savedbytes), NOCRED, 0);
+               }
+               
+               /*
+                * We may have just deleted the catalog record for a resource fork even 
+                * though it did not exist in core as a vnode. However, just because there 
+                * was a resource fork pointer in the cnode does not mean that it had any blocks.
+                */
+               if (temp_rsrc_fork || cp->c_rsrcfork) {
+                       if (cp->c_rsrcfork) {
+                               if (cp->c_rsrcfork->ff_blocks > 0) {
+                                       savedbytes = ((off_t)cp->c_rsrcfork->ff_blocks * (off_t)blksize);
+                                       (void) hfs_chkdq(cp, (int64_t)-(savedbytes), NOCRED, 0);
+                               }
+                       }       
+                       else {
+                               /* we must have used a temporary fork */
+                               savedbytes = ((off_t)temp_rsrc_fork->ff_blocks * (off_t)blksize);       
+                               (void) hfs_chkdq(cp, (int64_t)-(savedbytes), NOCRED, 0);
+                       }
+               }
+               
+               if (hfsmp->hfs_flags & HFS_QUOTAS) {
+                       (void)hfs_chkiq(cp, -1, NOCRED, 0);
+               }
+#endif
+               
+               /* 
+                * If we didn't get any errors deleting the catalog entry, then go ahead
+                * and release the backing store now.  The filefork pointers are still valid.
+                */
+               if (temp_rsrc_fork) {   
+                       error = hfs_release_storage (hfsmp, cp->c_datafork, temp_rsrc_fork, fileid);
+               }
+               else {
+                       /* if cp->c_rsrcfork == NULL, hfs_release_storage will skip over it. */
+                       error = hfs_release_storage (hfsmp, cp->c_datafork, cp->c_rsrcfork, fileid);
+               }
+               if (error) {
+                       /* 
+                        * If we encountered an error updating the extents and bitmap,
+                        * mark the volume inconsistent.  At this point, the catalog record has
+                        * already been deleted, so we can't recover it at this point. We need
+                        * to proceed and update the volume header and mark the cnode C_NOEXISTS.
+                        * The subsequent fsck should be able to recover the free space for us.
+                        */
+                       hfs_mark_inconsistent(hfsmp, HFS_OP_INCOMPLETE);
+               }
+               else {
+                       /* reset update_vh to 0, since hfs_release_storage should have done it for us */
+                       update_vh = 0;
+               }
+
+               /* Get rid of the temporary rsrc fork */
+               if (temp_rsrc_fork) {
+                       FREE_ZONE (temp_rsrc_fork, sizeof(struct filefork), M_HFSFORK);
+               }
 
-               cp->c_mode = 0;
-               truncated  = 0;    // because the catalog entry is gone
-               cp->c_flag |= C_CHANGE | C_NOEXISTS;
-               --cp->c_nlink;
+               cp->c_flag |= C_NOEXISTS;
+               cp->c_flag &= ~C_DELETED;
+               
+               cp->c_touch_chgtime = TRUE;
+               --cp->c_linkcount;
+               
+               /* 
+                * We must never get a directory if we're in this else block.  We could 
+                * accidentally drop the number of files in the volume header if we did.
+                */
                hfs_volupdate(hfsmp, VOL_RMFILE, (dcp->c_cnid == kHFSRootFolderID));
+               
+       }
+
+       //
+       // if skip_reserve == 1 then we're being called from hfs_vnop_rename() and thus
+       // we don't need to touch the document_id as it's handled by the rename code.
+       // otherwise it's a normal remove and we need to save the document id in the
+       // per thread struct and clear it from the cnode.
+       //
+       struct  doc_tombstone *ut;
+       ut = get_uthread_doc_tombstone();
+       if (!error && !skip_reserve && (cp->c_bsdflags & UF_TRACKED) && should_save_docid_tombstone(ut, vp, cnp)) {
+
+               if (ut->t_lastop_document_id) {
+                       clear_tombstone_docid(ut, hfsmp, NULL);
+               }
+               save_tombstone(hfsmp, dvp, vp, cnp, 1);
+
        }
 
+
        /*
         * All done with this cnode's descriptor...
         *
-        * Note: all future catalog calls for this cnode must be
-        * by fileid only.  This is OK for HFS (which doesn't have
-        * file thread records) since HFS doesn't support hard
-        * links or the removal of busy files.
+        * Note: all future catalog calls for this cnode must be by
+        * fileid only.  This is OK for HFS (which doesn't have file
+        * thread records) since HFS doesn't support the removal of
+        * busy files.
         */
        cat_releasedesc(&cp->c_desc);
 
-       /* In all three cases the parent lost a child */
-       if (dcp->c_entries > 0)
-               dcp->c_entries--;
-       if (dcp->c_nlink > 0)
-               dcp->c_nlink--;
-       dcp->c_flag |= C_CHANGE | C_UPDATE;
-       tv = time;
-       (void) VOP_UPDATE(dvp, &tv, &tv, 0);
-       HFS_KNOTE(dvp, NOTE_WRITE);
-
 out:
-       /* All done with component name... */
-       if ((options & HFSRM_SAVE_NAME) == 0 &&
-           (cnp != 0) &&
-           (cnp->cn_flags & (HASBUF | SAVENAME)) == (HASBUF | SAVENAME)) {
-               char *tmp = cnp->cn_pnbuf;
-               cnp->cn_pnbuf = NULL;
-               cnp->cn_flags &= ~HASBUF;
-               FREE_ZONE(tmp, cnp->cn_pnlen, M_NAMEI);
-       }
-
-       if (!(options & HFSRM_SKIP_RESERVE)) {
-               cat_postflight(hfsmp, &cookie, p);
-       }
-
-       /* Commit the truncation to the catalog record */
-       if (truncated) {
-           cp->c_flag |= C_CHANGE | C_UPDATE | C_FORCEUPDATE;
-           tv = time;
-           (void) VOP_UPDATE(vp, &tv, &tv, 0);
+       if (error) {
+           cp->c_flag &= ~C_DELETED;
        }
+       
+       if (update_vh) {
+               /* 
+                * If we bailed out earlier, we may need to update the volume header
+                * to deal with the borrowed blocks accounting. 
+                */
+               hfs_volupdate (hfsmp, VOL_UPDATE, 0);
+       }       
 
-       // XXXdbg
        if (started_tr) {
-           journal_end_transaction(hfsmp->jnl);
-       }
-       if (grabbed_lock) {
-               hfs_global_shared_lock_release(hfsmp);
+           hfs_end_transaction(hfsmp);
        }
 
-       HFS_KNOTE(vp, NOTE_DELETE);
-       if (rvp) {
-               HFS_KNOTE(rvp, NOTE_DELETE);
-               vrele(rvp);
-       };
-
-       if (error) {
-               vput(vp);
-       } else {
-               VOP_UNLOCK(vp, 0, p);
-               // XXXdbg - try to prevent the lost ubc_info panic
-               if ((cp->c_flag & C_HARDLINK) == 0 || cp->c_nlink == 0) {
-                       (void) ubc_uncache(vp);
-               }
-               vrele(vp);
-       }
-       if (!(options & HFSRM_PARENT_LOCKED)) {
-               vput(dvp);
-       }
+       dcp->c_flag &= ~C_DIR_MODIFICATION;
+       wakeup((caddr_t)&dcp->c_flag);
 
        return (error);
 }
@@ -2083,14 +4293,19 @@ out:
 __private_extern__ void
 replace_desc(struct cnode *cp, struct cat_desc *cdp)
 {
+       // fixes 4348457 and 4463138
+       if (&cp->c_desc == cdp) {
+           return;
+       }
+
        /* First release allocated name buffer */
        if (cp->c_desc.cd_flags & CD_HASBUF && cp->c_desc.cd_nameptr != 0) {
-               char *name = cp->c_desc.cd_nameptr;
+               const u_int8_t *name = cp->c_desc.cd_nameptr;
 
                cp->c_desc.cd_nameptr = 0;
                cp->c_desc.cd_namelen = 0;
                cp->c_desc.cd_flags &= ~CD_HASBUF;
-               remove_name(name);
+               vfs_removename((const char *)name);
        }
        bcopy(cdp, &cp->c_desc, sizeof(cp->c_desc));
 
@@ -2101,108 +4316,364 @@ replace_desc(struct cnode *cp, struct cat_desc *cdp)
 }
 
 
-/*
-#
-#% rename      fdvp   U U U
-#% rename      fvp    U U U
-#% rename      tdvp   L U U
-#% rename      tvp    X U U
-#
-*/
 /*
  * Rename a cnode.
  *
- * The VFS layer guarantees that source and destination will
- * either both be directories, or both not be directories.
+ * The VFS layer guarantees that:
+ *   - source and destination will either both be directories, or
+ *     both not be directories.
+ *   - all the vnodes are from the same file system
  *
- * When the target is a directory, hfs_rename must ensure
- * that it is empty.
+ * When the target is a directory, HFS must ensure that its empty.
  *
- * The rename system call is responsible for freeing
- * the pathname buffers (ie no need to call VOP_ABORTOP).
+ * Note that this function requires up to 6 vnodes in order to work properly
+ * if it is operating on files (and not on directories).  This is because only
+ * files can have resource forks, and we now require iocounts to be held on the
+ * vnodes corresponding to the resource forks (if applicable) as well as
+ * the files or directories undergoing rename.  The problem with not holding 
+ * iocounts on the resource fork vnodes is that it can lead to a deadlock 
+ * situation: The rsrc fork of the source file may be recycled and reclaimed 
+ * in order to provide a vnode for the destination file's rsrc fork.  Since
+ * data and rsrc forks share the same cnode, we'd eventually try to lock the
+ * source file's cnode in order to sync its rsrc fork to disk, but it's already 
+ * been locked.  By taking the rsrc fork vnodes up front we ensure that they 
+ * cannot be recycled, and that the situation mentioned above cannot happen.
  */
-
-static int
-hfs_rename(ap)
-       struct vop_rename_args  /* {
+int
+hfs_vnop_rename(ap)
+       struct vnop_rename_args  /* {
                struct vnode *a_fdvp;
                struct vnode *a_fvp;
                struct componentname *a_fcnp;
                struct vnode *a_tdvp;
                struct vnode *a_tvp;
                struct componentname *a_tcnp;
+               vfs_context_t a_context;
        } */ *ap;
 {
        struct vnode *tvp = ap->a_tvp;
        struct vnode *tdvp = ap->a_tdvp;
        struct vnode *fvp = ap->a_fvp;
        struct vnode *fdvp = ap->a_fdvp;
+       /*
+        * Note that we only need locals for the target/destination's
+        * resource fork vnode (and only if necessary).  We don't care if the
+        * source has a resource fork vnode or not.
+        */
+       struct vnode *tvp_rsrc = NULLVP;        
+       uint32_t tvp_rsrc_vid = 0;
        struct componentname *tcnp = ap->a_tcnp;
        struct componentname *fcnp = ap->a_fcnp;
-       struct proc *p = fcnp->cn_proc;
-       struct cnode *fcp = NULL;
-       struct cnode *fdcp = NULL;
-       struct cnode *tdcp = VTOC(tdvp);
+       struct proc *p = vfs_context_proc(ap->a_context);
+       struct cnode *fcp;
+       struct cnode *fdcp;
+       struct cnode *tdcp;
+       struct cnode *tcp;
+       struct cnode *error_cnode;
        struct cat_desc from_desc;
        struct cat_desc to_desc;
        struct cat_desc out_desc;
-       struct hfsmount *hfsmp = NULL;
-       struct timeval tv;
-       cat_cookie_t cookie = {0};
-       int fdvp_locked, fvp_locked, tdvp_locked, tvp_locked;
-       int tvp_deleted;
-       int started_tr = 0, grabbed_lock = 0;
-       int error = 0;
+       struct hfsmount *hfsmp;
+       cat_cookie_t cookie;
+       int tvp_deleted = 0;
+       int started_tr = 0, got_cookie = 0;
+       int took_trunc_lock = 0;
+       int lockflags;
+       int error;
+       time_t orig_from_ctime, orig_to_ctime;
+       int emit_rename = 1;
+       int emit_delete = 1;
+       int is_tracked = 0;
+       int unlocked;
+
+       orig_from_ctime = VTOC(fvp)->c_ctime;
+       if (tvp && VTOC(tvp)) {
+               orig_to_ctime = VTOC(tvp)->c_ctime;
+       } else {
+               orig_to_ctime = ~0;
+       }
 
+       hfsmp = VTOHFS(tdvp);
+       /* 
+        * Do special case checks here.  If fvp == tvp then we need to check the
+        * cnode with locks held.
+        */
+       if (fvp == tvp) {
+               int is_hardlink = 0;
+               /* 
+                * In this case, we do *NOT* ever emit a DELETE event.  
+                * We may not necessarily emit a RENAME event 
+                */     
+               emit_delete = 0;
+               if ((error = hfs_lock(VTOC(fvp), HFS_SHARED_LOCK, HFS_LOCK_DEFAULT))) {
+                       return error;
+               }
+               /* Check to see if the item is a hardlink or not */
+               is_hardlink = (VTOC(fvp)->c_flag & C_HARDLINK);
+               hfs_unlock (VTOC(fvp));
+               
+               /* 
+                * If the item is not a hardlink, then case sensitivity must be off, otherwise
+                * two names should not resolve to the same cnode unless they were case variants.
+                */
+               if (is_hardlink) {
+                       emit_rename = 0;
+                       /*
+                        * Hardlinks are a little trickier.  We only want to emit a rename event
+                        * if the item is a hardlink, the parent directories are the same, case sensitivity
+                        * is off, and the case folded names are the same.  See the fvp == tvp case below for more
+                        * info.
+                        */
 
-       /* Establish our vnode lock state. */
-       tdvp_locked = 1;
-       tvp_locked = (tvp != 0);
-       fdvp_locked = 0;
-       fvp_locked = 0;
-       tvp_deleted = 0;
+                       if ((fdvp == tdvp) && ((hfsmp->hfs_flags & HFS_CASE_SENSITIVE) == 0)) {
+                               if (hfs_namecmp((const u_int8_t *)fcnp->cn_nameptr, fcnp->cn_namelen,
+                                                       (const u_int8_t *)tcnp->cn_nameptr, tcnp->cn_namelen) == 0) {
+                                       /* Then in this case only it is ok to emit a rename */
+                                       emit_rename = 1;
+                               }
+                       }
+               }
+       }
+       if (emit_rename) {
+               /* c_bsdflags should only be assessed while holding the cnode lock.  
+                * This is not done consistently throughout the code and can result 
+                * in race.  This will be fixed via rdar://12181064
+                */
+               if (VTOC(fvp)->c_bsdflags & UF_TRACKED) {
+                       is_tracked = 1;
+               }
+               check_for_tracked_file(fvp, orig_from_ctime, NAMESPACE_HANDLER_RENAME_OP, NULL);
+       }
 
-       /*
-        * Check for cross-device rename.
+       if (tvp && VTOC(tvp)) {
+               if (emit_delete) {
+                       check_for_tracked_file(tvp, orig_to_ctime, NAMESPACE_HANDLER_DELETE_OP, NULL);
+               }
+       }
+
+retry:
+       /* When tvp exists, take the truncate lock for hfs_removefile(). */
+       if (tvp && (vnode_isreg(tvp) || vnode_islnk(tvp))) {
+               hfs_lock_truncate(VTOC(tvp), HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
+               took_trunc_lock = 1;
+       }
+
+relock:
+       error = hfs_lockfour(VTOC(fdvp), VTOC(fvp), VTOC(tdvp), tvp ? VTOC(tvp) : NULL,
+                            HFS_EXCLUSIVE_LOCK, &error_cnode);
+       if (error) {
+               if (took_trunc_lock) {
+                       hfs_unlock_truncate(VTOC(tvp), HFS_LOCK_DEFAULT);
+                       took_trunc_lock = 0;
+               }
+
+               /* 
+                * We hit an error path.  If we were trying to re-acquire the locks
+                * after coming through here once, we might have already obtained
+                * an iocount on tvp's resource fork vnode.  Drop that before dealing
+                * with the failure.  Note this is safe -- since we are in an
+                * error handling path, we can't be holding the cnode locks.
+                */
+               if (tvp_rsrc) {
+                       vnode_put (tvp_rsrc);
+                       tvp_rsrc_vid = 0;
+                       tvp_rsrc = NULL;
+               }
+
+               /* 
+                * tvp might no longer exist.  If the cause of the lock failure 
+                * was tvp, then we can try again with tvp/tcp set to NULL.  
+                * This is ok because the vfs syscall will vnode_put the vnodes 
+                * after we return from hfs_vnop_rename.
+                */
+               if ((error == ENOENT) && (tvp != NULL) && (error_cnode == VTOC(tvp))) { 
+                       tcp = NULL;
+                       tvp = NULL;
+                       goto retry;
+               }
+
+               /* If we want to reintroduce notifications for failed renames, this
+                  is the place to do it. */
+
+               return (error);
+       }
+
+       fdcp = VTOC(fdvp);
+       fcp = VTOC(fvp);
+       tdcp = VTOC(tdvp);
+       tcp = tvp ? VTOC(tvp) : NULL;
+
+       //
+       // if the item is tracked but doesn't have a document_id, assign one and generate an fsevent for it
+       //
+       unlocked = 0;
+       if ((fcp->c_bsdflags & UF_TRACKED) && ((struct FndrExtendedDirInfo *)((char *)&fcp->c_attr.ca_finderinfo + 16))->document_id == 0) {
+               uint32_t newid;
+
+               hfs_unlockfour(VTOC(fdvp), VTOC(fvp), VTOC(tdvp), tvp ? VTOC(tvp) : NULL);
+               unlocked = 1;
+
+               if (hfs_generate_document_id(hfsmp, &newid) == 0) {
+                       hfs_lock(fcp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
+                       ((struct FndrExtendedDirInfo *)((char *)&fcp->c_attr.ca_finderinfo + 16))->document_id = newid;
+#if CONFIG_FSE
+                       add_fsevent(FSE_DOCID_CHANGED, vfs_context_current(),
+                                   FSE_ARG_DEV,   hfsmp->hfs_raw_dev,
+                                   FSE_ARG_INO,   (ino64_t)0,             // src inode #
+                                   FSE_ARG_INO,   (ino64_t)fcp->c_fileid,  // dst inode #
+                                   FSE_ARG_INT32, newid,
+                                   FSE_ARG_DONE);
+#endif
+                       hfs_unlock(fcp);
+               } else {
+                       // XXXdbg - couldn't get a new docid... what to do?  can't really fail the rename...
+               }
+
+               //
+               // check if we're going to need to fix tcp as well.  if we aren't, go back relock
+               // everything.  otherwise continue on and fix up tcp as well before relocking.
+               //
+               if (tcp == NULL || !(tcp->c_bsdflags & UF_TRACKED) || ((struct FndrExtendedDirInfo *)((char *)&tcp->c_attr.ca_finderinfo + 16))->document_id != 0) {
+                       goto relock;
+               }
+       }
+
+       //
+       // same thing for tcp if it's set
+       //
+       if (tcp && (tcp->c_bsdflags & UF_TRACKED) && ((struct FndrExtendedDirInfo *)((char *)&tcp->c_attr.ca_finderinfo + 16))->document_id == 0) {
+               uint32_t newid;
+
+               if (!unlocked) {
+                       hfs_unlockfour(VTOC(fdvp), VTOC(fvp), VTOC(tdvp), tvp ? VTOC(tvp) : NULL);
+                       unlocked = 1;
+               }
+
+               if (hfs_generate_document_id(hfsmp, &newid) == 0) {
+                       hfs_lock(tcp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
+                       ((struct FndrExtendedDirInfo *)((char *)&tcp->c_attr.ca_finderinfo + 16))->document_id = newid;
+#if CONFIG_FSE
+                       add_fsevent(FSE_DOCID_CHANGED, vfs_context_current(),
+                                   FSE_ARG_DEV,   hfsmp->hfs_raw_dev,
+                                   FSE_ARG_INO,   (ino64_t)0,             // src inode #
+                                   FSE_ARG_INO,   (ino64_t)tcp->c_fileid,  // dst inode #
+                                   FSE_ARG_INT32, newid,
+                                   FSE_ARG_DONE);
+#endif
+                       hfs_unlock(tcp);
+               } else {
+                       // XXXdbg - couldn't get a new docid... what to do?  can't really fail the rename...
+               }
+
+               // go back up and relock everything.  next time through the if statement won't be true
+               // and we'll skip over this block of code.
+               goto relock;
+       }
+
+
+
+       /* 
+        * Acquire iocounts on the destination's resource fork vnode 
+        * if necessary. If dst/src are files and the dst has a resource 
+        * fork vnode, then we need to try and acquire an iocount on the rsrc vnode. 
+        * If it does not exist, then we don't care and can skip it.
+        */
+       if ((vnode_isreg(fvp)) || (vnode_islnk(fvp))) {
+               if ((tvp) && (tcp->c_rsrc_vp) && (tvp_rsrc == NULL)) {
+                       tvp_rsrc = tcp->c_rsrc_vp;
+                       /*
+                        * We can look at the vid here because we're holding the 
+                        * cnode lock on the underlying cnode for this rsrc vnode. 
+                        */
+                       tvp_rsrc_vid = vnode_vid (tvp_rsrc);
+
+                       /* Unlock everything to acquire iocount on this rsrc vnode */
+                       if (took_trunc_lock) {
+                               hfs_unlock_truncate (VTOC(tvp), HFS_LOCK_DEFAULT);
+                               took_trunc_lock = 0;    
+                       }       
+                       hfs_unlockfour(fdcp, fcp, tdcp, tcp);
+
+                       if (vnode_getwithvid (tvp_rsrc, tvp_rsrc_vid)) {
+                               /* iocount acquisition failed.  Reset fields and start over.. */
+                               tvp_rsrc_vid = 0;
+                               tvp_rsrc = NULL;
+                       }
+                       goto retry;
+               }
+       }
+
+
+
+       /* Ensure we didn't race src or dst parent directories with rmdir. */
+       if (fdcp->c_flag & (C_NOEXISTS | C_DELETED)) {
+               error = ENOENT;
+               goto out;
+       }
+
+       if (tdcp->c_flag & (C_NOEXISTS | C_DELETED)) {
+               error = ENOENT;
+               goto out;       
+       }
+
+
+       /* Check for a race against unlink.  The hfs_valid_cnode checks validate
+        * the parent/child relationship with fdcp and tdcp, as well as the
+        * component name of the target cnodes.  
         */
-       if ((fvp->v_mount != tdvp->v_mount) ||
-           (tvp && (fvp->v_mount != tvp->v_mount))) {
-               error = EXDEV;
+       if ((fcp->c_flag & (C_NOEXISTS | C_DELETED)) || !hfs_valid_cnode(hfsmp, fdvp, fcnp, fcp->c_fileid, NULL, &error)) {
+               error = ENOENT;
                goto out;
        }
 
+       if (tcp && ((tcp->c_flag & (C_NOEXISTS | C_DELETED)) || !hfs_valid_cnode(hfsmp, tdvp, tcnp, tcp->c_fileid, NULL, &error))) {
+           //
+           // hmm, the destination vnode isn't valid any more.
+           // in this case we can just drop him and pretend he
+           // never existed in the first place.
+           //
+           if (took_trunc_lock) {
+                       hfs_unlock_truncate(VTOC(tvp), HFS_LOCK_DEFAULT);
+                       took_trunc_lock = 0;
+           }
+               error = 0;
+
+           hfs_unlockfour(fdcp, fcp, tdcp, tcp);
+
+           tcp = NULL;
+           tvp = NULL;
+           
+           // retry the locking with tvp null'ed out
+           goto retry;
+       }
+
+       fdcp->c_flag |= C_DIR_MODIFICATION;
+       if (fdvp != tdvp) {
+           tdcp->c_flag |= C_DIR_MODIFICATION;
+       }
+
        /*
-        * When fvp matches tvp they must be case variants
-        * or hard links.
-        *
-        * In some cases tvp will be locked in other cases
-        * it be unlocked with no reference.  Normalize the
-        * state here (unlocked with a reference) so that
-        * we can exit in a known state.
+        * Disallow renaming of a directory hard link if the source and 
+        * destination parent directories are different, or a directory whose 
+        * descendant is a directory hard link and the one of the ancestors
+        * of the destination directory is a directory hard link.
         */
-       if (fvp == tvp) {
-               if (VOP_ISLOCKED(tvp) &&
-                   (VTOC(tvp)->c_lock.lk_lockholder == p->p_pid) &&
-                   (VTOC(tvp)->c_lock.lk_lockthread == current_thread())) {
-                       vput(tvp);
+       if (vnode_isdir(fvp) && (fdvp != tdvp)) {
+               if (fcp->c_flag & C_HARDLINK) {
+                       error = EPERM;
+                       goto out;
                }
-               tvp = NULL;
-               tvp_locked = 0;
-
-               /*
-                * If this a hard link with different parents
-                * and its not a case variant then keep tvp
-                * around for removal.
-                */
-               if ((VTOC(fvp)->c_flag & C_HARDLINK) &&
-                   ((fdvp != tdvp) ||
-                    (hfs_namecmp(fcnp->cn_nameptr, fcnp->cn_namelen,
-                                 tcnp->cn_nameptr, tcnp->cn_namelen) != 0))) {
-                       tvp = fvp;
-                       vref(tvp);
+               if (fcp->c_attr.ca_recflags & kHFSHasChildLinkMask) {
+                   lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG, HFS_SHARED_LOCK);
+                   if (cat_check_link_ancestry(hfsmp, tdcp->c_fileid, 0)) {
+                               error = EPERM;
+                               hfs_systemfile_unlock(hfsmp, lockflags);
+                               goto out;
+                       }
+                       hfs_systemfile_unlock(hfsmp, lockflags);
                }
        }
-        
+
        /*
         * The following edge case is caught here:
         * (to cannot be a descendent of from)
@@ -2218,7 +4689,7 @@ hfs_rename(ap)
         *     /
         *    o tvp
         */
-       if (tdcp->c_parentcnid == VTOC(fvp)->c_cnid) {
+       if (tdcp->c_parentcnid == fcp->c_fileid) {
                error = EINVAL;
                goto out;
        }
@@ -2238,7 +4709,7 @@ hfs_rename(ap)
         *     /
         *    o fvp
         */
-       if (tvp && (tvp->v_type == VDIR) && (VTOC(tvp)->c_entries != 0)) {
+       if (tvp && vnode_isdir(tvp) && (tcp->c_entries != 0) && fvp != tvp) {
                error = ENOTEMPTY;
                goto out;
        }
@@ -2260,14 +4731,11 @@ hfs_rename(ap)
        /*
         * Make sure "from" vnode and its parent are changeable.
         */
-       if ((VTOC(fvp)->c_flags & (IMMUTABLE | APPEND)) ||
-           (VTOC(fdvp)->c_flags & APPEND)) {
+       if ((fcp->c_bsdflags & (IMMUTABLE | APPEND)) || (fdcp->c_bsdflags & APPEND)) {
                error = EPERM;
                goto out;
        }
 
-       hfsmp = VTOHFS(tdvp);
-
        /*
         * If the destination parent directory is "sticky", then the
         * user must own the parent directory, or the destination of
@@ -2275,158 +4743,141 @@ hfs_rename(ap)
         * (except by root). This implements append-only directories.
         *
         * Note that checks for immutable and write access are done
-        * by the call to VOP_REMOVE.
+        * by the call to hfs_removefile.
         */
        if (tvp && (tdcp->c_mode & S_ISTXT) &&
-           (tcnp->cn_cred->cr_uid != 0) &&
-           (tcnp->cn_cred->cr_uid != tdcp->c_uid) &&
-           (hfs_owner_rights(hfsmp, VTOC(tvp)->c_uid, tcnp->cn_cred, p, false)) ) {
+           (suser(vfs_context_ucred(tcnp->cn_context), NULL)) &&
+           (kauth_cred_getuid(vfs_context_ucred(tcnp->cn_context)) != tdcp->c_uid) &&
+           (hfs_owner_rights(hfsmp, tcp->c_uid, vfs_context_ucred(tcnp->cn_context), p, false)) ) {
+               error = EPERM;
+               goto out;
+       }
+
+       /* Don't allow modification of the journal or journal_info_block */
+       if (hfs_is_journal_file(hfsmp, fcp) ||
+           (tcp && hfs_is_journal_file(hfsmp, tcp))) {
                error = EPERM;
                goto out;
        }
 
 #if QUOTA
        if (tvp)
-               (void)hfs_getinoquota(VTOC(tvp));
+               (void)hfs_getinoquota(tcp);
 #endif
+       /* Preflighting done, take fvp out of the name space. */
+       cache_purge(fvp);
 
+#if CONFIG_SECLUDED_RENAME
        /*
-        * Lock all the vnodes before starting a journal transaction.
-        */
-
-       /*
-        * Simple case (same parent) - just lock child (fvp).
-        */
-       if (fdvp == tdvp) {
-               if (error = vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY, p))
-                       goto out;
-               fvp_locked = 1;
-               goto vnlocked;
-       }
-
-       /*
-        * If fdvp is the parent of tdvp then we'll need to
-        * drop tdvp's lock before acquiring a lock on fdvp.
-        *
-        *          fdvp
-        *           o 
-        *          / \
-        *         /   \
-        *   tdvp o     o fvp
-        *         \
-        *          \
-        *           o tvp
-        *
-        *
-        * If the parent directories are unrelated then we'll
-        * need to aquire their vnode locks in vnode address
-        * order.  Otherwise we can race with another rename
-        * call that involves the same vnodes except that to
-        * and from are switched and potentially deadlock.
-        * [ie rename("a/b", "c/d") vs rename("c/d", "a/b")]
+        * Check for "secure" rename that imposes additional restrictions on the
+        * source vnode.  We wait until here to check in order to prevent a race
+        * with other threads that manage to look up fvp, but their open or link
+        * is blocked by our locks.  At this point, with fvp out of the name cache,
+        * and holding the lock on fdvp, no other thread can find fvp.
         *
-        * If its not either of the two above cases then we
-        * can safely lock fdvp and fvp.
+        * TODO: Do we need to limit these checks to regular files only?
         */
-       if ((VTOC(fdvp)->c_cnid == VTOC(tdvp)->c_parentcnid) ||
-           ((VTOC(tdvp)->c_cnid != VTOC(fdvp)->c_parentcnid) &&
-            (fdvp < tdvp))) {
-
-               /* Drop locks on tvp and tdvp */
-               if (tvp_locked) {
-                       VOP_UNLOCK(tvp, 0, p);
-                       tvp_locked = 0;
-               }
-               VOP_UNLOCK(tdvp, 0, p);
-               tdvp_locked = 0;
-
-               /* Aquire locks in correct order */
-               if ((error = vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY, p)))
-                       goto out;
-               fdvp_locked = 1;
-               if ((error = vn_lock(tdvp, LK_EXCLUSIVE | LK_RETRY, p)))
-                       goto out;
-               tdvp_locked = 1;
-
+       if (fcnp->cn_flags & CN_SECLUDE_RENAME) {
+               if (vnode_isdir(fvp)) {
+                       error = EISDIR;
+                       goto out;
+               }
+           
                /*
-                * Now that the parents are locked only one thread
-                * can continue. So the lock order of the children
-                * doesn't really matter
+                * Neither fork of source may be open or memory mapped.
+                * We also don't want it in use by any other system call.
+                * The file must not have hard links.
+                *
+                * We can't simply use vnode_isinuse() because that does not
+                * count opens with O_EVTONLY.  We don't want a malicious
+                * process using O_EVTONLY to subvert a secluded rename.
                 */
-               if (tvp == fvp) {
-                       if ((error = vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, p)))
-                               goto out;
-                       tvp_locked = 1;
-               } else {
-                       if (tvp) {
-                               if ((error = vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, p)))
-                                       goto out;
-                               tvp_locked = 1;
-                       }
-                       if ((error = vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY, p)))
-                               goto out;
-                       fvp_locked = 1;
+               if (fcp->c_linkcount != 1) {
+                       error = EMLINK;
+                       goto out;
                }
 
-       } else /* OK to lock fdvp and fvp */ {
-               if ((error = vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY, p)))
+               if (fcp->c_rsrc_vp && (fcp->c_rsrc_vp->v_usecount > 0 ||
+                                      fcp->c_rsrc_vp->v_iocount > 0)) {
+                       /* Resource fork is in use (including O_EVTONLY) */
+                       error = EBUSY;
                        goto out;
-               fdvp_locked = 1;
-               if (error = vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY, p))
+               }
+               if (fcp->c_vp && (fcp->c_vp->v_usecount > (fcp->c_rsrc_vp ? 1 : 0) ||
+                                 fcp->c_vp->v_iocount > 1)) {
+                       /*
+                        * Data fork is in use, including O_EVTONLY, but not
+                        * including a reference from the resource fork.
+                        */
+                       error = EBUSY;
                        goto out;
-               if (tvp == fvp)
-                       tvp_locked = 1;
-               else
-                       fvp_locked = 1;
-       }
-
-vnlocked:
-       fdcp = VTOC(fdvp);
-       fcp = VTOC(fvp);
-
-       /* 
-        * While fvp is still locked, purge it from the name cache and
-        * grab it's c_cnid value.  Note that the removal of tvp (below)
-        * can drop fvp's lock when fvp == tvp.
-        */
-       cache_purge(fvp);
-
-       /*
-        * When a file moves out of "Cleanup At Startup"
-        * we can drop its NODUMP status.
-        */
-       if ((fcp->c_flags & UF_NODUMP) &&
-           (fvp->v_type == VREG) &&
-           (fdvp != tdvp) &&
-           (fdcp->c_desc.cd_nameptr != NULL) &&
-           (strcmp(fdcp->c_desc.cd_nameptr, CARBON_TEMP_DIR_NAME) == 0)) {
-               fcp->c_flags &= ~UF_NODUMP;
-               fcp->c_flag |= C_CHANGE;
-               tv = time;
-               (void) VOP_UPDATE(fvp, &tv, &tv, 0);
+               }
        }
-
+#endif
+    
        bzero(&from_desc, sizeof(from_desc));
-       from_desc.cd_nameptr = fcnp->cn_nameptr;
+       from_desc.cd_nameptr = (const u_int8_t *)fcnp->cn_nameptr;
        from_desc.cd_namelen = fcnp->cn_namelen;
-       from_desc.cd_parentcnid = fdcp->c_cnid;
+       from_desc.cd_parentcnid = fdcp->c_fileid;
        from_desc.cd_flags = fcp->c_desc.cd_flags & ~(CD_HASBUF | CD_DECOMPOSED);
        from_desc.cd_cnid = fcp->c_cnid;
 
        bzero(&to_desc, sizeof(to_desc));
-       to_desc.cd_nameptr = tcnp->cn_nameptr;
+       to_desc.cd_nameptr = (const u_int8_t *)tcnp->cn_nameptr;
        to_desc.cd_namelen = tcnp->cn_namelen;
-       to_desc.cd_parentcnid = tdcp->c_cnid;
+       to_desc.cd_parentcnid = tdcp->c_fileid;
        to_desc.cd_flags = fcp->c_desc.cd_flags & ~(CD_HASBUF | CD_DECOMPOSED);
        to_desc.cd_cnid = fcp->c_cnid;
 
-       hfs_global_shared_lock_acquire(hfsmp);
-       grabbed_lock = 1;
-       if (hfsmp->jnl) {
-           if ((error = journal_start_transaction(hfsmp->jnl)) != 0) {
-                       goto out;
+       if ((error = hfs_start_transaction(hfsmp)) != 0) {
+           goto out;
+       }
+       started_tr = 1;
+
+       /* hfs_vnop_link() and hfs_vnop_rename() set kHFSHasChildLinkMask 
+        * inside a journal transaction and without holding a cnode lock.  
+        * As setting of this bit depends on being in journal transaction for 
+        * concurrency, check this bit again after we start journal transaction for rename
+        * to ensure that this directory does not have any descendant that
+        * is a directory hard link. 
+        */
+       if (vnode_isdir(fvp) && (fdvp != tdvp)) {
+               if (fcp->c_attr.ca_recflags & kHFSHasChildLinkMask) {
+                   lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG, HFS_SHARED_LOCK);
+                   if (cat_check_link_ancestry(hfsmp, tdcp->c_fileid, 0)) {
+                               error = EPERM;
+                               hfs_systemfile_unlock(hfsmp, lockflags);
+                               goto out;
+                       }
+                       hfs_systemfile_unlock(hfsmp, lockflags);
+               }
+       }
+
+       // if it's a hardlink then re-lookup the name so
+       // that we get the correct cnid in from_desc (see
+       // the comment in hfs_removefile for more details)
+       //
+       if (fcp->c_flag & C_HARDLINK) {
+           struct cat_desc tmpdesc;
+           cnid_t real_cnid;
+
+           tmpdesc.cd_nameptr = (const u_int8_t *)fcnp->cn_nameptr;
+           tmpdesc.cd_namelen = fcnp->cn_namelen;
+           tmpdesc.cd_parentcnid = fdcp->c_fileid;
+           tmpdesc.cd_hint = fdcp->c_childhint;
+           tmpdesc.cd_flags = fcp->c_desc.cd_flags & CD_ISDIR;
+           tmpdesc.cd_encoding = 0;
+           
+           lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG, HFS_SHARED_LOCK);
+
+           if (cat_lookup(hfsmp, &tmpdesc, 0, 0, NULL, NULL, NULL, &real_cnid) != 0) {
+               hfs_systemfile_unlock(hfsmp, lockflags);
+               goto out;
            }
-               started_tr = 1;
+
+           // use the real cnid instead of whatever happened to be there
+           from_desc.cd_cnid = real_cnid;
+           hfs_systemfile_unlock(hfsmp, lockflags);
        }
 
        /*
@@ -2435,358 +4886,720 @@ vnlocked:
        if ((error = cat_preflight(hfsmp, CAT_RENAME + CAT_DELETE, &cookie, p))) {
                goto out;
        }
+       got_cookie = 1;
 
        /*
-        * If the destination exists then it needs to be removed.
+        * If the destination exists then it may need to be removed.  
+        * 
+        * Due to HFS's locking system, we should always move the 
+        * existing 'tvp' element to the hidden directory in hfs_vnop_rename.
+        * Because the VNOP_LOOKUP call enters and exits the filesystem independently
+        * of the actual vnop that it was trying to do (stat, link, readlink),
+        * we must release the cnode lock of that element during the interim to 
+        * do MAC checking, vnode authorization, and other calls.  In that time, 
+        * the item can be deleted (or renamed over). However, only in the rename 
+        * case is it inappropriate to return ENOENT from any of those calls.  Either 
+        * the call should return information about the old element (stale), or get 
+        * information about the newer element that we are about to write in its place.  
+        * 
+        * HFS lookup has been modified to detect a rename and re-drive its 
+        * lookup internally. For other calls that have already succeeded in 
+        * their lookup call and are waiting to acquire the cnode lock in order 
+        * to proceed, that cnode lock will not fail due to the cnode being marked 
+        * C_NOEXISTS, because it won't have been marked as such.  It will only 
+        * have C_DELETED.  Thus, they will simply act on the stale open-unlinked
+        * element.  All future callers will get the new element.
+        *
+        * To implement this behavior, we pass the "only_unlink" argument to 
+        * hfs_removefile and hfs_removedir.  This will result in the vnode acting 
+        * as though it is open-unlinked.  Additionally, when we are done moving the 
+        * element to the hidden directory, we vnode_recycle the target so that it is 
+        * reclaimed as soon as possible.  Reclaim and inactive are both 
+        * capable of clearing out unused blocks for an open-unlinked file or dir.
         */
-
        if (tvp) {
-               if (tvp != fvp)
-                       cache_purge(tvp);
+               //
+               // if the destination has a document id, we need to preserve it
+               //
+               if (fvp != tvp) {
+                       uint32_t document_id;
+                       struct FndrExtendedDirInfo *ffip = (struct FndrExtendedDirInfo *)((char *)&fcp->c_attr.ca_finderinfo + 16);
+                       struct FndrExtendedDirInfo *tfip = (struct FndrExtendedDirInfo *)((char *)&tcp->c_attr.ca_finderinfo + 16);
+                       
+                       if (ffip->document_id && tfip->document_id) {
+                               // both documents are tracked.  only save a tombstone from tcp and do nothing else.
+                               save_tombstone(hfsmp, tdvp, tvp, tcnp, 0);
+                       } else {
+                               struct  doc_tombstone *ut;
+                               ut = get_uthread_doc_tombstone();
+                               
+                               document_id = tfip->document_id;
+                               tfip->document_id = 0;
+                       
+                               if (document_id != 0) {
+                                       // clear UF_TRACKED as well since tcp is now no longer tracked
+                                       tcp->c_bsdflags &= ~UF_TRACKED;
+                                       (void) cat_update(hfsmp, &tcp->c_desc, &tcp->c_attr, NULL, NULL);
+                               }
+
+                               if (ffip->document_id == 0 && document_id != 0) {
+                                       // printf("RENAME: preserving doc-id %d onto %s (from ino %d, to ino %d)\n", document_id, tcp->c_desc.cd_nameptr, tcp->c_desc.cd_cnid, fcp->c_desc.cd_cnid);
+                                       fcp->c_bsdflags |= UF_TRACKED;
+                                       ffip->document_id = document_id;
+                                       
+                                       (void) cat_update(hfsmp, &fcp->c_desc, &fcp->c_attr, NULL, NULL);
+#if CONFIG_FSE
+                                       add_fsevent(FSE_DOCID_CHANGED, vfs_context_current(),
+                                                   FSE_ARG_DEV, hfsmp->hfs_raw_dev,
+                                                   FSE_ARG_INO, (ino64_t)tcp->c_fileid,           // src inode #
+                                                   FSE_ARG_INO, (ino64_t)fcp->c_fileid,           // dst inode #
+                                                   FSE_ARG_INT32, (uint32_t)ffip->document_id,
+                                                   FSE_ARG_DONE);
+#endif
+                               } else if ((fcp->c_bsdflags & UF_TRACKED) && should_save_docid_tombstone(ut, fvp, fcnp)) {
+
+                                       if (ut->t_lastop_document_id) {
+                                               clear_tombstone_docid(ut, hfsmp, NULL);
+                                       }
+                                       save_tombstone(hfsmp, fdvp, fvp, fcnp, 0);
+
+                                       //printf("RENAME: (dest-exists): saving tombstone doc-id %lld @ %s (ino %d)\n",
+                                       //       ut->t_lastop_document_id, ut->t_lastop_filename, fcp->c_desc.cd_cnid);
+                               }
+                       }
+               }
+
                /*
-                * Note that hfs_removedir and hfs_removefile
-                * will keep tdvp locked with a reference.
-                * But tvp will lose its lock and reference.
+                * When fvp matches tvp they could be case variants
+                * or matching hard links.
                 */
-               if (tvp->v_type == VDIR)
-                       error = hfs_removedir(tdvp, tvp, tcnp, HFSRM_RENAMEOPTS);
-               else
-                       error = hfs_removefile(tdvp, tvp, tcnp, HFSRM_RENAMEOPTS);
+               if (fvp == tvp) {
+                       if (!(fcp->c_flag & C_HARDLINK)) {
+                               /* 
+                                * If they're not hardlinks, then fvp == tvp must mean we 
+                                * are using case-insensitive HFS because case-sensitive would
+                                * not use the same vnode for both.  In this case we just update
+                                * the catalog for: a -> A
+                                */
+                               goto skip_rm;  /* simple case variant */
 
-               if (tvp == fvp)
-                       fvp_locked = 0;
-               tvp = NULL;
-               tvp_locked = 0;
-               tvp_deleted = 1;
-               if (error)
+                       }
+                       /* For all cases below, we must be using hardlinks */   
+                       else if ((fdvp != tdvp) ||
+                                  (hfsmp->hfs_flags & HFS_CASE_SENSITIVE)) {
+                               /*
+                                * If the parent directories are not the same, AND the two items
+                                * are hardlinks, posix says to do nothing:
+                                * dir1/fred <-> dir2/bob   and the op was mv dir1/fred -> dir2/bob
+                                * We just return 0 in this case.
+                                *
+                                * If case sensitivity is on, and we are using hardlinks 
+                                * then renaming is supposed to do nothing.
+                                * dir1/fred <-> dir2/FRED, and op == mv dir1/fred -> dir2/FRED
+                                */
+                               goto out;  /* matching hardlinks, nothing to do */
+
+                       } else if (hfs_namecmp((const u_int8_t *)fcnp->cn_nameptr, fcnp->cn_namelen,
+                                              (const u_int8_t *)tcnp->cn_nameptr, tcnp->cn_namelen) == 0) {
+                               /*
+                                * If we get here, then the following must be true:
+                                * a) We are running case-insensitive HFS+.
+                                * b) Both paths 'fvp' and 'tvp' are in the same parent directory.
+                                * c) the two names are case-variants of each other.
+                                *
+                                * In this case, we are really only dealing with a single catalog record
+                                * whose name is being updated.
+                                * 
+                                * op is dir1/fred -> dir1/FRED
+                                * 
+                                * We need to special case the name matching, because if
+                                * dir1/fred <-> dir1/bob were the two links, and the 
+                                * op was dir1/fred -> dir1/bob
+                                * That would fail/do nothing.
+                                */
+                               goto skip_rm;  /* case-variant hardlink in the same dir */
+                       } else {
+                               goto out;  /* matching hardlink, nothing to do */
+                       }
+               }
+
+               
+               if (vnode_isdir(tvp)) {
+                       /*
+                        * hfs_removedir will eventually call hfs_removefile on the directory
+                        * we're working on, because only hfs_removefile does the renaming of the
+                        * item to the hidden directory.  The directory will stay around in the
+                        * hidden directory with C_DELETED until it gets an inactive or a reclaim.
+                        * That way, we can destroy all of the EAs as needed and allow new ones to be
+                        * written.
+                        */
+                       error = hfs_removedir(tdvp, tvp, tcnp, HFSRM_SKIP_RESERVE, 1);
+               }
+               else {
+                       error = hfs_removefile(tdvp, tvp, tcnp, 0, HFSRM_SKIP_RESERVE, 0, NULL, 1);
+                       
+                       /*
+                        * If the destination file had a resource fork vnode, then we need to get rid of
+                        * its blocks when there are no more references to it.  Because the call to
+                        * hfs_removefile above always open-unlinks things, we need to force an inactive/reclaim
+                        * on the resource fork vnode, in order to prevent block leaks.  Otherwise,
+                        * the resource fork vnode could prevent the data fork vnode from going out of scope
+                        * because it holds a v_parent reference on it.  So we mark it for termination
+                        * with a call to vnode_recycle. hfs_vnop_reclaim has been modified so that it 
+                        * can clean up the blocks of open-unlinked files and resource forks. 
+                        *
+                        * We can safely call vnode_recycle on the resource fork because we took an iocount
+                        * reference on it at the beginning of the function. 
+                        */ 
+                       
+                       if ((error == 0) && (tcp->c_flag & C_DELETED) && (tvp_rsrc)) {
+                               vnode_recycle(tvp_rsrc);
+                       }
+               }
+
+               if (error) {
                        goto out;
-       }
+               }
+               
+               tvp_deleted = 1;
+               
+               /* Mark 'tcp' as being deleted due to a rename */
+               tcp->c_flag |= C_RENAMED;
+               
+               /*
+                * Aggressively mark tvp/tcp for termination to ensure that we recover all blocks
+                * as quickly as possible.
+                */
+               vnode_recycle(tvp);
+       } else {
+               struct  doc_tombstone *ut;
+               ut = get_uthread_doc_tombstone();
+               
+               //
+               // There is nothing at the destination.  If the file being renamed is
+               // tracked, save a "tombstone" of the document_id.  If the file is
+               // not a tracked file, then see if it needs to inherit a tombstone.
+               //
+               // NOTE: we do not save a tombstone if the file being renamed begins
+               //       with "atmp" which is done to work-around AutoCad's bizarre
+               //       5-step un-safe save behavior
+               //
+               if (fcp->c_bsdflags & UF_TRACKED) {
+                       if (should_save_docid_tombstone(ut, fvp, fcnp)) {
+                               save_tombstone(hfsmp, fdvp, fvp, fcnp, 0);
+                               
+                               //printf("RENAME: (no dest): saving tombstone doc-id %lld @ %s (ino %d)\n",
+                               //       ut->t_lastop_document_id, ut->t_lastop_filename, fcp->c_desc.cd_cnid);
+                       } else {
+                               // intentionally do nothing
+                       }
+               } else if (   ut->t_lastop_document_id != 0
+                          && tdvp == ut->t_lastop_parent
+                          && vnode_vid(tdvp) == ut->t_lastop_parent_vid
+                          && strcmp((char *)ut->t_lastop_filename, (char *)tcnp->cn_nameptr) == 0) {
+
+                       //printf("RENAME: %s (ino %d) inheriting doc-id %lld\n", tcnp->cn_nameptr, fcp->c_desc.cd_cnid, ut->t_lastop_document_id);
+                       struct FndrExtendedFileInfo *fip = (struct FndrExtendedFileInfo *)((char *)&fcp->c_attr.ca_finderinfo + 16);
+                       fcp->c_bsdflags |= UF_TRACKED;
+                       fip->document_id = ut->t_lastop_document_id;
+                       cat_update(hfsmp, &fcp->c_desc, &fcp->c_attr, NULL, NULL);
+                       
+                       clear_tombstone_docid(ut, hfsmp, fcp);    // will send the docid-changed fsevent
 
+               } else if (ut->t_lastop_document_id && should_save_docid_tombstone(ut, fvp, fcnp) && should_save_docid_tombstone(ut, tvp, tcnp)) {
+                       // no match, clear the tombstone
+                       //printf("RENAME: clearing the tombstone %lld @ %s\n", ut->t_lastop_document_id, ut->t_lastop_filename);
+                       clear_tombstone_docid(ut, hfsmp, NULL);
+               }
+                          
+       }
+skip_rm:
        /*
-        * All done with tvp and fvp
+        * All done with tvp and fvp. 
+        * 
+        * We also jump to this point if there was no destination observed during lookup and namei.
+        * However, because only iocounts are held at the VFS layer, there is nothing preventing a 
+        * competing thread from racing us and creating a file or dir at the destination of this rename 
+        * operation.  If this occurs, it may cause us to get a spurious EEXIST out of the cat_rename 
+        * call below.  To preserve rename's atomicity, we need to signal VFS to re-drive the 
+        * namei/lookup and restart the rename operation.  EEXIST is an allowable errno to be bubbled 
+        * out of the rename syscall, but not for this reason, since it is a synonym errno for ENOTEMPTY.
+        * To signal VFS, we return ERECYCLE (which is also used for lookup restarts). This errno
+        * will be swallowed and it will restart the operation.
         */
-
-       /* Lock catalog b-tree */
-       error = hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_EXCLUSIVE, p);
-       if (error)
-                goto out;
-
+       
+       lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG, HFS_EXCLUSIVE_LOCK);
        error = cat_rename(hfsmp, &from_desc, &tdcp->c_desc, &to_desc, &out_desc);
-
-       /* Unlock catalog b-tree */
-       (void) hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_RELEASE, p);
+       hfs_systemfile_unlock(hfsmp, lockflags);
 
        if (error) {
+               if (error == EEXIST) {
+                       error = ERECYCLE;
+               }
                goto out;
        }
 
-       /* Update cnode's catalog descriptor */
-       if (fvp_locked) {
-               replace_desc(fcp, &out_desc);
-               fcp->c_parentcnid = tdcp->c_cnid;
-               fcp->c_hint = 0;
+       /* Invalidate negative cache entries in the destination directory */
+       if (tdcp->c_flag & C_NEG_ENTRIES) {
+               cache_purge_negatives(tdvp);
+               tdcp->c_flag &= ~C_NEG_ENTRIES;
        }
 
-       hfs_volupdate(hfsmp, fvp->v_type == VDIR ? VOL_RMDIR : VOL_RMFILE,
+       /* Update cnode's catalog descriptor */
+       replace_desc(fcp, &out_desc);
+       fcp->c_parentcnid = tdcp->c_fileid;
+       fcp->c_hint = 0;
+
+       /* Now indicate this cnode needs to have date-added written to the finderinfo */
+       fcp->c_flag |= C_NEEDS_DATEADDED;
+       (void) hfs_update (fvp, 0);
+
+
+       hfs_volupdate(hfsmp, vnode_isdir(fvp) ? VOL_RMDIR : VOL_RMFILE,
                      (fdcp->c_cnid == kHFSRootFolderID));
-       hfs_volupdate(hfsmp, fvp->v_type == VDIR ? VOL_MKDIR : VOL_MKFILE,
+       hfs_volupdate(hfsmp, vnode_isdir(fvp) ? VOL_MKDIR : VOL_MKFILE,
                      (tdcp->c_cnid == kHFSRootFolderID));
 
        /* Update both parent directories. */
-       tv = time;
        if (fdvp != tdvp) {
-               tdcp->c_nlink++;
+               if (vnode_isdir(fvp)) {
+                       /* If the source directory has directory hard link 
+                        * descendants, set the kHFSHasChildLinkBit in the 
+                        * destination parent hierarchy 
+                        */
+                       if ((fcp->c_attr.ca_recflags & kHFSHasChildLinkMask) && 
+                           !(tdcp->c_attr.ca_recflags & kHFSHasChildLinkMask)) {
+
+                               tdcp->c_attr.ca_recflags |= kHFSHasChildLinkMask;
+
+                               error = cat_set_childlinkbit(hfsmp, tdcp->c_parentcnid);
+                               if (error) {
+                                       printf ("hfs_vnop_rename: error updating parent chain for %u\n", tdcp->c_cnid);
+                                       error = 0;
+                               }
+                       }
+                       INC_FOLDERCOUNT(hfsmp, tdcp->c_attr);
+                       DEC_FOLDERCOUNT(hfsmp, fdcp->c_attr);
+               }
                tdcp->c_entries++;
-               if (fdcp->c_nlink > 0)
-                       fdcp->c_nlink--;
+               tdcp->c_dirchangecnt++;
+               tdcp->c_flag |= C_MODIFIED;
+               hfs_incr_gencount(tdcp);
+
                if (fdcp->c_entries > 0)
                        fdcp->c_entries--;
-               fdcp->c_flag |= C_CHANGE | C_UPDATE;
-               (void) VOP_UPDATE(fdvp, &tv, &tv, 0);
+               fdcp->c_dirchangecnt++;
+               fdcp->c_flag |= C_MODIFIED;
+               fdcp->c_touch_chgtime = TRUE;
+               fdcp->c_touch_modtime = TRUE;
+
+               if (ISSET(fcp->c_flag, C_HARDLINK)) {
+                       hfs_relorigin(fcp, fdcp->c_fileid);
+                       if (fdcp->c_fileid != fdcp->c_cnid)
+                               hfs_relorigin(fcp, fdcp->c_cnid);
+               }
+
+               (void) hfs_update(fdvp, 0);
        }
+       hfs_incr_gencount(fdcp);
+               
        tdcp->c_childhint = out_desc.cd_hint;   /* Cache directory's location */
-       tdcp->c_flag |= C_CHANGE | C_UPDATE;
-       (void) VOP_UPDATE(tdvp, &tv, &tv, 0);
+       tdcp->c_touch_chgtime = TRUE;
+       tdcp->c_touch_modtime = TRUE;
+
+       (void) hfs_update(tdvp, 0);
+
+       /* Update the vnode's name now that the rename has completed. */
+       vnode_update_identity(fvp, tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, 
+                       tcnp->cn_hash, (VNODE_UPDATE_PARENT | VNODE_UPDATE_NAME));
+       
+       /* 
+        * At this point, we may have a resource fork vnode attached to the 
+        * 'from' vnode.  If it exists, we will want to update its name, because
+        * it contains the old name + _PATH_RSRCFORKSPEC. ("/..namedfork/rsrc").
+        *
+        * Note that the only thing we need to update here is the name attached to
+        * the vnode, since a resource fork vnode does not have a separate resource
+        * cnode -- it's still 'fcp'.
+        */
+       if (fcp->c_rsrc_vp) {
+               char* rsrc_path = NULL;
+               int len;
+
+               /* Create a new temporary buffer that's going to hold the new name */
+               MALLOC_ZONE (rsrc_path, caddr_t, MAXPATHLEN, M_NAMEI, M_WAITOK);
+               len = snprintf (rsrc_path, MAXPATHLEN, "%s%s", tcnp->cn_nameptr, _PATH_RSRCFORKSPEC);
+               len = MIN(len, MAXPATHLEN);
 
+               /* 
+                * vnode_update_identity will do the following for us:
+                * 1) release reference on the existing rsrc vnode's name.
+                * 2) copy/insert new name into the name cache
+                * 3) attach the new name to the resource vnode
+                * 4) update the vnode's vid
+                */
+               vnode_update_identity (fcp->c_rsrc_vp, fvp, rsrc_path, len, 0, (VNODE_UPDATE_NAME | VNODE_UPDATE_CACHE));
+               
+               /* Free the memory associated with the resource fork's name */
+               FREE_ZONE (rsrc_path, MAXPATHLEN, M_NAMEI);     
+       }
 out:
-       if (hfsmp) {
+       if (got_cookie) {
                cat_postflight(hfsmp, &cookie, p);
        }
        if (started_tr) {
-           journal_end_transaction(hfsmp->jnl);
-       }
-       if (grabbed_lock) {
-               hfs_global_shared_lock_release(hfsmp);
+           hfs_end_transaction(hfsmp);
        }
 
-       /* Note that if hfs_removedir or hfs_removefile was invoked above they will already have
-          generated a NOTE_WRITE for tdvp and a NOTE_DELETE for tvp.
-        */
-       if (error == 0) {
-               HFS_KNOTE(fvp, NOTE_RENAME);
-               HFS_KNOTE(fdvp, NOTE_WRITE);
-               if (tdvp != fdvp) HFS_KNOTE(tdvp, NOTE_WRITE);
-       };
-       if (fvp_locked) {
-               VOP_UNLOCK(fvp, 0, p);
-       }
-       if (fdvp_locked) {
-               VOP_UNLOCK(fdvp, 0, p);
-       }
-       if (tdvp_locked) {
-               VOP_UNLOCK(tdvp, 0, p);
+       fdcp->c_flag &= ~C_DIR_MODIFICATION;
+       wakeup((caddr_t)&fdcp->c_flag);
+       if (fdvp != tdvp) {
+           tdcp->c_flag &= ~C_DIR_MODIFICATION;
+           wakeup((caddr_t)&tdcp->c_flag);
        }
-       if (tvp_locked) {
-               VOP_UNLOCK(tvp, 0, p);
+
+       hfs_unlockfour(fdcp, fcp, tdcp, tcp);
+
+       if (took_trunc_lock) {
+               hfs_unlock_truncate(VTOC(tvp), HFS_LOCK_DEFAULT);       
        }
 
-       vrele(fvp);
-       vrele(fdvp);
-       if (tvp)
-               vrele(tvp);
-       vrele(tdvp);
+       /* Now vnode_put the resource forks vnodes if necessary */
+       if (tvp_rsrc) {
+               vnode_put(tvp_rsrc);
+               tvp_rsrc = NULL;        
+       }
 
        /* After tvp is removed the only acceptable error is EIO */
        if (error && tvp_deleted)
                error = EIO;
 
+       /* If we want to reintroduce notifications for renames, this is the
+          place to do it. */
+
        return (error);
 }
 
 
-
 /*
- * Mkdir system call
-#% mkdir       dvp     L U U
-#% mkdir       vpp     - L -
-#
- vop_mkdir {
-     IN WILLRELE struct vnode *dvp;
-     OUT struct vnode **vpp;
-     IN struct componentname *cnp;
-     IN struct vattr *vap;
-
-     We are responsible for freeing the namei buffer,
-        it is done in hfs_makenode()
-*/
-
-static int
-hfs_mkdir(ap)
-       struct vop_mkdir_args /* {
-               struct vnode *a_dvp;
-               struct vnode **a_vpp;
-               struct componentname *a_cnp;
-               struct vattr *a_vap;
-       } */ *ap;
+ * Make a directory.
+ */
+int
+hfs_vnop_mkdir(struct vnop_mkdir_args *ap)
 {
-       struct vattr *vap = ap->a_vap;
-
-       return (hfs_makenode(MAKEIMODE(vap->va_type, vap->va_mode),
-                            ap->a_dvp, ap->a_vpp, ap->a_cnp));
+       /***** HACK ALERT ********/
+       ap->a_cnp->cn_flags |= MAKEENTRY;
+       return hfs_makenode(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap, ap->a_context);
 }
 
 
 /*
- * symlink -- make a symbolic link
-#% symlink     dvp     L U U
-#% symlink     vpp     - U -
-#
-# XXX - note that the return vnode has already been VRELE'ed
-#      by the filesystem layer.  To use it you must use vget,
-#      possibly with a further namei.
-#
- vop_symlink {
-     IN WILLRELE struct vnode *dvp;
-     OUT WILLRELE struct vnode **vpp;
-     IN struct componentname *cnp;
-     IN struct vattr *vap;
-     IN char *target;
-
-     We are responsible for freeing the namei buffer, 
-        it is done in hfs_makenode().
-
-*/
-
-static int
-hfs_symlink(ap)
-       struct vop_symlink_args /* {
-               struct vnode *a_dvp;
-               struct vnode **a_vpp;
-               struct componentname *a_cnp;
-               struct vattr *a_vap;
-               char *a_target;
-       } */ *ap;
+ * Create a symbolic link.
+ */
+int
+hfs_vnop_symlink(struct vnop_symlink_args *ap)
 {
-       register struct vnode *vp, **vpp = ap->a_vpp;
+       struct vnode **vpp = ap->a_vpp;
+       struct vnode *dvp = ap->a_dvp;
+       struct vnode *vp = NULL;
+       struct cnode *cp = NULL;
        struct hfsmount *hfsmp;
        struct filefork *fp;
-       int len, error;
        struct buf *bp = NULL;
+       char *datap;
+       int started_tr = 0;
+       u_int32_t len;
+       int error;
 
        /* HFS standard disks don't support symbolic links */
-       if (VTOVCB(ap->a_dvp)->vcbSigWord != kHFSPlusSigWord) {
-               VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
-               vput(ap->a_dvp);
-               return (EOPNOTSUPP);
-       }
+       if (VTOVCB(dvp)->vcbSigWord != kHFSPlusSigWord)
+               return (ENOTSUP);
 
        /* Check for empty target name */
-       if (ap->a_target[0] == 0) {
-               VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
-               vput(ap->a_dvp);
+       if (ap->a_target[0] == 0)
                return (EINVAL);
-       }
 
+       hfsmp = VTOHFS(dvp);
+       len = strlen(ap->a_target);
 
-       hfsmp = VTOHFS(ap->a_dvp);
+       /* Check for free space */
+       if (((u_int64_t)hfs_freeblks(hfsmp, 0) * (u_int64_t)hfsmp->blockSize) < len) {
+               return (ENOSPC);
+       }
 
        /* Create the vnode */
-       if ((error = hfs_makenode(S_IFLNK | ap->a_vap->va_mode,
-                                                         ap->a_dvp, vpp, ap->a_cnp))) {
-               return (error);
+       ap->a_vap->va_mode |= S_IFLNK;
+       if ((error = hfs_makenode(dvp, vpp, ap->a_cnp, ap->a_vap, ap->a_context))) {
+               goto out;
        }
-
        vp = *vpp;
-       len = strlen(ap->a_target);
+       if ((error = hfs_lock(VTOC(vp), HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT))) {
+               goto out;
+       }
+       cp = VTOC(vp);
        fp = VTOF(vp);
 
+       if (cp->c_flag & (C_NOEXISTS | C_DELETED)) {
+           goto out;
+       }
+
 #if QUOTA
-       (void)hfs_getinoquota(VTOC(vp));
+       (void)hfs_getinoquota(cp);
 #endif /* QUOTA */
 
-       // XXXdbg
-       hfs_global_shared_lock_acquire(hfsmp);
-       if (hfsmp->jnl) {
-           if ((error = journal_start_transaction(hfsmp->jnl)) != 0) {
-                       hfs_global_shared_lock_release(hfsmp);
-                       vput(vp);
-                       return error;
-           }
+       if ((error = hfs_start_transaction(hfsmp)) != 0) {
+           goto out;
        }
+       started_tr = 1;
 
-       /* Allocate space for the link */
-       error = VOP_TRUNCATE(vp, len, IO_NOZEROFILL,
-                             ap->a_cnp->cn_cred, ap->a_cnp->cn_proc);
-       if (error)
-               goto out;       /* XXX need to remove link */
+       /*
+        * Allocate space for the link.
+        *
+        * Since we're already inside a transaction,
+        *
+        * Don't need truncate lock since a symlink is treated as a system file.
+        */
+       error = hfs_truncate(vp, len, IO_NOZEROFILL, 0, ap->a_context);
+
+       /* On errors, remove the symlink file */
+       if (error) {
+               /*
+                * End the transaction so we don't re-take the cnode lock
+                * below while inside a transaction (lock order violation).
+                */
+               hfs_end_transaction(hfsmp);
+
+               /* hfs_removefile() requires holding the truncate lock */
+               hfs_unlock(cp);
+               hfs_lock_truncate(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
+               hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_ALLOW_NOEXISTS);
+
+               if (hfs_start_transaction(hfsmp) != 0) {
+                       started_tr = 0;
+                       hfs_unlock_truncate(cp, HFS_LOCK_DEFAULT);
+                       goto out;
+               }
+               
+               (void) hfs_removefile(dvp, vp, ap->a_cnp, 0, 0, 0, NULL, 0);
+               hfs_unlock_truncate(cp, HFS_LOCK_DEFAULT);
+               goto out;       
+       }
 
        /* Write the link to disk */
-       bp = getblk(vp, 0, roundup((int)fp->ff_size, VTOHFS(vp)->hfs_phys_block_size),
+       bp = buf_getblk(vp, (daddr64_t)0, roundup((int)fp->ff_size, hfsmp->hfs_physical_block_size),
                        0, 0, BLK_META);
        if (hfsmp->jnl) {
                journal_modify_block_start(hfsmp->jnl, bp);
        }
-       bzero(bp->b_data, bp->b_bufsize);
-       bcopy(ap->a_target, bp->b_data, len);
+       datap = (char *)buf_dataptr(bp);
+       bzero(datap, buf_size(bp));
+       bcopy(ap->a_target, datap, len);
+
        if (hfsmp->jnl) {
-               journal_modify_block_end(hfsmp->jnl, bp);
+               journal_modify_block_end(hfsmp->jnl, bp, NULL, NULL);
        } else {
-               bawrite(bp);
+               buf_bawrite(bp);
        }
 out:
-       if (hfsmp->jnl) {
-               journal_end_transaction(hfsmp->jnl);
+       if (started_tr)
+           hfs_end_transaction(hfsmp);
+       if ((cp != NULL) && (vp != NULL)) {
+               hfs_unlock(cp);
+       }
+       if (error) {
+               if (vp) {
+                       vnode_put(vp);
+               }
+               *vpp = NULL;
        }
-       hfs_global_shared_lock_release(hfsmp);
-       vput(vp);
        return (error);
 }
 
 
-/*
- * Dummy dirents to simulate the "." and ".." entries of the directory
- * in a hfs filesystem.  HFS doesn't provide these on disk.  Note that
- * the size of these entries is the smallest needed to represent them
- * (only 12 byte each).
- */
-static hfsdotentry  rootdots[2] = {
-       {
-               1,                              /* d_fileno */
-               sizeof(struct hfsdotentry),     /* d_reclen */
-               DT_DIR,                         /* d_type */
-               1,                              /* d_namlen */
-               "."                             /* d_name */
-    },
-    {
-               1,                              /* d_fileno */
-               sizeof(struct hfsdotentry),     /* d_reclen */
-               DT_DIR,                         /* d_type */
-               2,                              /* d_namlen */
-               ".."                            /* d_name */
-       }
+/* structures to hold a "." or ".." directory entry */
+struct hfs_stddotentry {
+       u_int32_t       d_fileno;   /* unique file number */
+       u_int16_t       d_reclen;   /* length of this structure */
+       u_int8_t        d_type;     /* dirent file type */
+       u_int8_t        d_namlen;   /* len of filename */
+       char            d_name[4];  /* "." or ".." */
 };
 
-/*     4.3 Note:
-*      There is some confusion as to what the semantics of uio_offset are.
-*      In ufs, it represents the actual byte offset within the directory
-*      "file."  HFS, however, just uses it as an entry counter - essentially
-*      assuming that it has no meaning except to the hfs_readdir function.
-*      This approach would be more efficient here, but some callers may
-*      assume the uio_offset acts like a byte offset.  NFS in fact
-*      monkeys around with the offset field a lot between readdir calls.
-*
-*      The use of the resid uiop->uio_resid and uiop->uio_iov->iov_len
-*      fields is a mess as well.  The libc function readdir() returns
-*      NULL (indicating the end of a directory) when either
-*      the getdirentries() syscall (which calls this and returns
-*      the size of the buffer passed in less the value of uiop->uio_resid)
-*      returns 0, or a direct record with a d_reclen of zero.
-*      nfs_server.c:rfs_readdir(), on the other hand, checks for the end
-*      of the directory by testing uiop->uio_resid == 0.  The solution
-*      is to pad the size of the last struct direct in a given
-*      block to fill the block if we are not at the end of the directory.
-*/
+struct hfs_extdotentry {
+       u_int64_t  d_fileno;   /* unique file number */
+       u_int64_t  d_seekoff;  /* seek offset (optional, used by servers) */
+       u_int16_t  d_reclen;   /* length of this structure */
+       u_int16_t  d_namlen;   /* len of filename */
+       u_int8_t   d_type;     /* dirent file type */
+       u_char     d_name[3];  /* "." or ".." */
+};
 
+typedef union {
+       struct hfs_stddotentry  std;
+       struct hfs_extdotentry  ext;
+} hfs_dotentry_t;
 
 /*
- * NOTE: We require a minimal buffer size of DIRBLKSIZ for two reasons. One, it is the same value
- * returned be stat() call as the block size. This is mentioned in the man page for getdirentries():
- * "Nbytes must be greater than or equal to the block size associated with the file,
- * see stat(2)". Might as well settle on the same size of ufs. Second, this makes sure there is enough
- * room for the . and .. entries that have to added manually.
+ *  hfs_vnop_readdir reads directory entries into the buffer pointed
+ *  to by uio, in a filesystem independent format.  Up to uio_resid
+ *  bytes of data can be transferred.  The data in the buffer is a
+ *  series of packed dirent structures where each one contains the
+ *  following entries:
+ *
+ *     u_int32_t   d_fileno;              // file number of entry
+ *     u_int16_t   d_reclen;              // length of this record
+ *     u_int8_t    d_type;                // file type
+ *     u_int8_t    d_namlen;              // length of string in d_name
+ *     char        d_name[MAXNAMELEN+1];  // null terminated file name
+ *
+ *  The current position (uio_offset) refers to the next block of
+ *  entries.  The offset can only be set to a value previously
+ *  returned by hfs_vnop_readdir or zero.  This offset does not have
+ *  to match the number of bytes returned (in uio_resid).
+ *
+ *  In fact, the offset used by HFS is essentially an index (26 bits)
+ *  with a tag (6 bits).  The tag is for associating the next request
+  *  with the current request.  This enables us to have multiple threads
+ *  reading the directory while the directory is also being modified.
+ *
+ *  Each tag/index pair is tied to a unique directory hint.  The hint
+ *  contains information (filename) needed to build the catalog b-tree
+ *  key for finding the next set of entries.
+ *
+ * If the directory is marked as deleted-but-in-use (cp->c_flag & C_DELETED),
+ * do NOT synthesize entries for "." and "..".
  */
-
-/*                     
-#% readdir     vp      L L L
-#
-vop_readdir {
-    IN struct vnode *vp;
-    INOUT struct uio *uio;
-    IN struct ucred *cred;
-    INOUT int *eofflag;
-    OUT int *ncookies;
-    INOUT u_long **cookies;
-    */
-static int
-hfs_readdir(ap)
-       struct vop_readdir_args /* {
-               struct vnode *vp;
-               struct uio *uio;
-               struct ucred *cred;
-               int *eofflag;
-               int *ncookies;
-               u_long **cookies;
+int
+hfs_vnop_readdir(ap)
+       struct vnop_readdir_args /* {
+               vnode_t a_vp;
+               uio_t a_uio;
+               int a_flags;
+               int *a_eofflag;
+               int *a_numdirent;
+               vfs_context_t a_context;
        } */ *ap;
 {
-       register struct uio *uio = ap->a_uio;
-       struct cnode *cp = VTOC(ap->a_vp);
-       struct hfsmount *hfsmp = VTOHFS(ap->a_vp);
-       struct proc *p = current_proc();
-       off_t off = uio->uio_offset;
-       int retval = 0;
+       struct vnode *vp = ap->a_vp;
+       uio_t uio = ap->a_uio;
+       struct cnode *cp;
+       struct hfsmount *hfsmp;
+       directoryhint_t *dirhint = NULL;
+       directoryhint_t localhint;
+       off_t offset;
+       off_t startoffset;
+       int error = 0;
        int eofflag = 0;
-       void *user_start = NULL;
-       int   user_len;
+       user_addr_t user_start = 0;
+       user_size_t user_len = 0;
+       int index;
+       unsigned int tag;
+       int items;
+       int lockflags;
+       int extended;
+       int nfs_cookies;
+       cnid_t cnid_hint = 0;
+       int bump_valence = 0;
+
+       items = 0;
+       startoffset = offset = uio_offset(uio);
+       extended = (ap->a_flags & VNODE_READDIR_EXTENDED);
+       nfs_cookies = extended && (ap->a_flags & VNODE_READDIR_REQSEEKOFF);
+
+       /* Sanity check the uio data. */
+       if (uio_iovcnt(uio) > 1)
+               return (EINVAL);
 
-       int ncookies=0;
-       u_long *cookies=NULL;
-       u_long *cookiep=NULL;
-       /* We assume it's all one big buffer... */
-       if (uio->uio_iovcnt > 1 || uio->uio_resid < AVERAGE_HFSDIRENTRY_SIZE)
-               return EINVAL;
+       if (VTOC(vp)->c_bsdflags & UF_COMPRESSED) {
+               int compressed = hfs_file_is_compressed(VTOC(vp), 0);  /* 0 == take the cnode lock */
+               if (VTOCMP(vp) != NULL && !compressed) {
+                       error = check_for_dataless_file(vp, NAMESPACE_HANDLER_READ_OP);
+                       if (error) {
+                               return error;
+                       }
+               }
+       }
 
-       // XXXdbg
+       cp = VTOC(vp);
+       hfsmp = VTOHFS(vp);
+
+       /* Note that the dirhint calls require an exclusive lock. */
+       if ((error = hfs_lock(VTOC(vp), HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT)))
+               return (error);
+
+       /* Pick up cnid hint (if any). */
+       if (nfs_cookies) {
+               cnid_hint = (cnid_t)(uio_offset(uio) >> 32);
+               uio_setoffset(uio, uio_offset(uio) & 0x00000000ffffffffLL);
+               if (cnid_hint == INT_MAX) { /* searching pass the last item */
+                       eofflag = 1;
+                       goto out;
+               }
+       }
+       /*
+        * Synthesize entries for "." and "..", unless the directory has
+        * been deleted, but not closed yet (lazy delete in progress).
+        */
+       if (offset == 0 && !(cp->c_flag & C_DELETED)) {
+               hfs_dotentry_t  dotentry[2];
+               size_t  uiosize;
+
+               if (extended) {
+                       struct hfs_extdotentry *entry = &dotentry[0].ext;
+
+                       entry->d_fileno = cp->c_cnid;
+                       entry->d_reclen = sizeof(struct hfs_extdotentry);
+                       entry->d_type = DT_DIR;
+                       entry->d_namlen = 1;
+                       entry->d_name[0] = '.';
+                       entry->d_name[1] = '\0';
+                       entry->d_name[2] = '\0';
+                       entry->d_seekoff = 1;
+
+                       ++entry;
+                       entry->d_fileno = cp->c_parentcnid;
+                       entry->d_reclen = sizeof(struct hfs_extdotentry);
+                       entry->d_type = DT_DIR;
+                       entry->d_namlen = 2;
+                       entry->d_name[0] = '.';
+                       entry->d_name[1] = '.';
+                       entry->d_name[2] = '\0';
+                       entry->d_seekoff = 2;
+                       uiosize = 2 * sizeof(struct hfs_extdotentry);
+               } else {
+                       struct hfs_stddotentry *entry = &dotentry[0].std;
+
+                       entry->d_fileno = cp->c_cnid;
+                       entry->d_reclen = sizeof(struct hfs_stddotentry);
+                       entry->d_type = DT_DIR;
+                       entry->d_namlen = 1;
+                       *(int *)&entry->d_name[0] = 0;
+                       entry->d_name[0] = '.';
+
+                       ++entry;
+                       entry->d_fileno = cp->c_parentcnid;
+                       entry->d_reclen = sizeof(struct hfs_stddotentry);
+                       entry->d_type = DT_DIR;
+                       entry->d_namlen = 2;
+                       *(int *)&entry->d_name[0] = 0;
+                       entry->d_name[0] = '.';
+                       entry->d_name[1] = '.';
+                       uiosize = 2 * sizeof(struct hfs_stddotentry);
+               }
+               if ((error = uiomove((caddr_t)&dotentry, uiosize, uio))) {
+                       goto out;
+               }
+               offset += 2;
+       }
+
+       /* 
+        * Intentionally avoid checking the valence here.  If we
+        * have FS corruption that reports the valence is 0, even though it
+        * has contents, we might artificially skip over iterating 
+        * this directory.
+        */
+       
+       //
        // We have to lock the user's buffer here so that we won't
        // fault on it after we've acquired a shared lock on the
        // catalog file.  The issue is that you can get a 3-way
@@ -2795,7 +5608,7 @@ hfs_readdir(ap)
        // here and we can't service our page fault because VM is
        // blocked trying to start a transaction as a result of
        // trying to free up pages for our page fault.  It's messy
-       // but it does happen on dual-procesors that are paging
+       // but it does happen on dual-processors that are paging
        // heavily (see radar 3082639 for more info).  By locking
        // the buffer up-front we prevent ourselves from faulting
        // while holding the shared catalog file lock.
@@ -2804,168 +5617,187 @@ hfs_readdir(ap)
        // currently (10/30/02) that can fault on user data with a
        // shared lock on the catalog file.
        //
-       if (hfsmp->jnl && uio->uio_segflg == UIO_USERSPACE) {
-               user_start = uio->uio_iov->iov_base;
-               user_len   = uio->uio_iov->iov_len;
+       if (hfsmp->jnl && uio_isuserspace(uio)) {
+               user_start = uio_curriovbase(uio);
+               user_len = uio_curriovlen(uio);
 
-               if ((retval = vslock(user_start, user_len)) != 0) {
-                       return retval;
+               if ((error = vslock(user_start, user_len)) != 0) {
+                       user_start = 0;
+                       goto out;
                }
        }
-
-       /* Create the entries for . and .. */
-       if (uio->uio_offset < sizeof(rootdots)) {
-               caddr_t dep;
-               size_t dotsize;
-               
-               rootdots[0].d_fileno = cp->c_cnid;
-               rootdots[1].d_fileno = cp->c_parentcnid;
-
-               if (uio->uio_offset == 0) {
-                       dep = (caddr_t) &rootdots[0];
-                       dotsize = 2* sizeof(struct hfsdotentry);
-               } else if (uio->uio_offset == sizeof(struct hfsdotentry)) {
-                       dep = (caddr_t) &rootdots[1];
-                       dotsize = sizeof(struct hfsdotentry);
-               } else {
-                       retval = EINVAL;
-                       goto Exit;
+       /* Convert offset into a catalog directory index. */
+       index = (offset & HFS_INDEX_MASK) - 2;
+       tag = offset & ~HFS_INDEX_MASK;
+
+       /* Lock catalog during cat_findname and cat_getdirentries. */
+       lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG, HFS_SHARED_LOCK);
+
+       /* When called from NFS, try and resolve a cnid hint. */
+       if (nfs_cookies && cnid_hint != 0) {
+               if (cat_findname(hfsmp, cnid_hint, &localhint.dh_desc) == 0) {
+                       if ( localhint.dh_desc.cd_parentcnid == cp->c_fileid) {
+                               localhint.dh_index = index - 1;
+                               localhint.dh_time = 0;
+                               bzero(&localhint.dh_link, sizeof(localhint.dh_link));
+                               dirhint = &localhint;  /* don't forget to release the descriptor */
+                       } else {
+                               cat_releasedesc(&localhint.dh_desc);
+                       }
                }
-
-               retval = uiomove(dep, dotsize, uio);
-               if (retval != 0)
-                       goto Exit;
        }
 
-       if (ap->a_ncookies != NULL) {
-               /*
-                * These cookies are handles that allow NFS to restart
-                * scanning through a directory.  If a directory is large
-                * enough, NFS will issue a successive readdir() with a
-                * uio->uio_offset that is equal to one of these cookies.
-                *
-                * The cookies that we generate are synthesized byte-offsets.
-                * The offset is where the dirent the dirent would be if the
-                * directory were an array of packed dirent structs.  It is
-                * synthetic because that's not how directories are stored in
-                * HFS but other code expects that the cookie is a byte offset.
-                *
-                * We have to pre-allocate the cookies because cat_getdirentries()
-                * is the only one that can properly synthesize the offsets (since
-                * it may have to skip over entries and only it knows the true
-                * virtual offset of any particular directory entry). So we allocate
-                * a cookie table here and pass it in to cat_getdirentries().
-                *
-                * Note that the handling of "." and ".." is mostly done here but
-                * cat_getdirentries() is aware of.
-                *
-                * Only the NFS server uses cookies so fortunately this code is
-                * not executed unless the NFS server is issuing the readdir
-                * request.
-                *
-                * Also note that the NFS server is the one responsible for
-                * free'ing the cookies even though we allocated them.  Ick.
-                *
-                * We allocate a reasonable number of entries for the size of
-                * the buffer that we're going to fill in.  cat_getdirentries()
-                * is smart enough to not overflow if there's more room in the
-                * buffer but not enough room in the cookie table.
-                */
-               if (uio->uio_segflg != UIO_SYSSPACE)
-                       panic("hfs_readdir: unexpected uio from NFS server");
-
-               ncookies = uio->uio_iov->iov_len / (AVERAGE_HFSDIRENTRY_SIZE/2);
-               MALLOC(cookies, u_long *, ncookies * sizeof(u_long), M_TEMP, M_WAITOK);
+       /* Get a directory hint (cnode must be locked exclusive) */
+       if (dirhint == NULL) {
+               dirhint = hfs_getdirhint(cp, ((index - 1) & HFS_INDEX_MASK) | tag, 0);
 
-               *ap->a_ncookies = ncookies;
-               *ap->a_cookies = cookies;
-               
-               /* handle cookies for "." and ".." */
-               if (off == 0) {
-                       cookies[0] = 0;
-                       cookies[1] = sizeof(struct hfsdotentry);
-               } else if (off == sizeof(struct hfsdotentry)) {
-                       cookies[0] = sizeof(struct hfsdotentry);
+               /* Hide tag from catalog layer. */
+               dirhint->dh_index &= HFS_INDEX_MASK;
+               if (dirhint->dh_index == HFS_INDEX_MASK) {
+                       dirhint->dh_index = -1;
                }
        }
+       
+       if (index == 0) {
+               dirhint->dh_threadhint = cp->c_dirthreadhint;
+       } 
+       else {
+               /*
+                * If we have a non-zero index, there is a possibility that during the last
+                * call to hfs_vnop_readdir we hit EOF for this directory.  If that is the case
+                * then we don't want to return any new entries for the caller.  Just return 0
+                * items, mark the eofflag, and bail out.  Because we won't have done any work, the 
+                * code at the end of the function will release the dirhint for us.  
+                *
+                * Don't forget to unlock the catalog lock on the way out, too.
+                */
+               if (dirhint->dh_desc.cd_flags & CD_EOF) {
+                       error = 0;
+                       eofflag = 1;
+                       uio_setoffset(uio, startoffset);
+                       hfs_systemfile_unlock (hfsmp, lockflags);
 
-       /* If there are no children then we're done */  
-       if (cp->c_entries == 0) {
-               eofflag = 1;
-               retval = 0;
-               if (cookies) {
-                   cookies[0] = 0;
-                   cookies[1] = sizeof(struct hfsdotentry);
+                       goto seekoffcalc;
                }
-               goto Exit;
        }
 
-       /* Lock catalog b-tree */
-       retval = hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_SHARED, p);
-       if (retval) goto Exit;
+       /* Pack the buffer with dirent entries. */
+       error = cat_getdirentries(hfsmp, cp->c_entries, dirhint, uio, ap->a_flags, &items, &eofflag);
 
-       retval = cat_getdirentries(hfsmp, &cp->c_desc, cp->c_entries, uio, &eofflag, cookies, ncookies);
+       if (index == 0 && error == 0) {
+               cp->c_dirthreadhint = dirhint->dh_threadhint;
+       }
 
-       /* Unlock catalog b-tree */
-       (void) hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_RELEASE, p);
+       hfs_systemfile_unlock(hfsmp, lockflags);
 
-       if (retval != E_NONE) {
-               goto Exit;
+       if (error != 0) {
+               goto out;
        }
        
-       /* were we already past eof ? */
-       if (uio->uio_offset == off) {
-               retval = E_NONE;
-               goto Exit;
+       /* Get index to the next item */
+       index += items;
+       
+       if (items >= (int)cp->c_entries) {
+               eofflag = 1;
        }
        
-       cp->c_flag |= C_ACCESS;
+       /* 
+        * Detect valence FS corruption.
+        *
+        * We are holding the cnode lock exclusive, so there should not be
+        * anybody modifying the valence field of this cnode.  If we enter
+        * this block, that means we observed filesystem corruption, because
+        * this directory reported a valence of 0, yet we found at least one
+        * item.  In this case, we need to minimally self-heal this
+        * directory to prevent userland from tripping over a directory
+        * that appears empty (getattr of valence reports 0), but actually 
+        * has contents. 
+        *
+        * We'll force the cnode update at the end of the function after 
+        * completing all of the normal getdirentries steps.
+        */ 
+       if ((cp->c_entries == 0) && (items > 0)) {
+               /* disk corruption */
+               cp->c_entries++;
+               /* Mark the cnode as dirty. */
+               cp->c_flag |= C_MODIFIED;
+               printf("hfs_vnop_readdir: repairing valence to non-zero! \n");
+               bump_valence++;
+       }
 
-Exit:;
-       if (hfsmp->jnl && user_start) {
-               vsunlock(user_start, user_len, TRUE);
+
+       /* Convert catalog directory index back into an offset. */
+       while (tag == 0)
+               tag = (++cp->c_dirhinttag) << HFS_INDEX_BITS;   
+       uio_setoffset(uio, (index + 2) | tag);
+       dirhint->dh_index |= tag;
+
+seekoffcalc:
+       cp->c_touch_acctime = TRUE;
+
+       if (ap->a_numdirent) {
+               if (startoffset == 0)
+                       items += 2;
+               *ap->a_numdirent = items;
        }
 
-       if (ap->a_eofflag)
+out:
+       if (user_start) {
+               vsunlock(user_start, user_len, TRUE);
+       }
+       /* If we didn't do anything then go ahead and dump the hint. */
+       if ((dirhint != NULL) &&
+           (dirhint != &localhint) &&
+           (uio_offset(uio) == startoffset)) {
+               hfs_reldirhint(cp, dirhint);
+               eofflag = 1;
+       }
+       if (ap->a_eofflag) {
                *ap->a_eofflag = eofflag;
+       }
+       if (dirhint == &localhint) {
+               cat_releasedesc(&localhint.dh_desc);
+       }
 
-    return (retval);
+       if (bump_valence) {
+               /* force the update before dropping the cnode lock*/
+               hfs_update(vp, 0);
+       }
+
+       hfs_unlock(cp);
+       
+       return (error);
 }
 
 
 /*
- * Return target name of a symbolic link
-#% readlink    vp      L L L
-#
- vop_readlink {
-     IN struct vnode *vp;
-     INOUT struct uio *uio;
-     IN struct ucred *cred;
-     */
-
-static int
-hfs_readlink(ap)
-       struct vop_readlink_args /* {
+ * Read contents of a symbolic link.
+ */
+int
+hfs_vnop_readlink(ap)
+       struct vnop_readlink_args /* {
                struct vnode *a_vp;
                struct uio *a_uio;
-               struct ucred *a_cred;
+               vfs_context_t a_context;
        } */ *ap;
 {
-       int retval;
        struct vnode *vp = ap->a_vp;
        struct cnode *cp;
        struct filefork *fp;
+       int error;
 
-       if (vp->v_type != VLNK)
+       if (!vnode_islnk(vp))
                return (EINVAL);
  
+       if ((error = hfs_lock(VTOC(vp), HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT)))
+               return (error);
        cp = VTOC(vp);
        fp = VTOF(vp);
    
        /* Zero length sym links are not allowed */
        if (fp->ff_size == 0 || fp->ff_size > MAXPATHLEN) {
-               VTOVCB(vp)->vcbFlags |= kHFS_DamagedVolume;
-               return (EINVAL);
+               error = EINVAL;
+               goto exit;
        }
     
        /* Cache the path so we don't waste buffer cache resources */
@@ -2973,33 +5805,35 @@ hfs_readlink(ap)
                struct buf *bp = NULL;
 
                MALLOC(fp->ff_symlinkptr, char *, fp->ff_size, M_TEMP, M_WAITOK);
-               retval = meta_bread(vp, 0,
-                               roundup((int)fp->ff_size,
-                                       VTOHFS(vp)->hfs_phys_block_size),
-                                               ap->a_cred, &bp);
-               if (retval) {
+               if (fp->ff_symlinkptr == NULL) {
+                       error = ENOMEM;
+                       goto exit;
+               }
+               error = (int)buf_meta_bread(vp, (daddr64_t)0,
+                                           roundup((int)fp->ff_size, VTOHFS(vp)->hfs_physical_block_size),
+                                           vfs_context_ucred(ap->a_context), &bp);
+               if (error) {
                        if (bp)
-                               brelse(bp);
+                               buf_brelse(bp);
                        if (fp->ff_symlinkptr) {
                                FREE(fp->ff_symlinkptr, M_TEMP);
                                fp->ff_symlinkptr = NULL;
                        }
-                       return (retval);
+                       goto exit;
                }
-               bcopy(bp->b_data, fp->ff_symlinkptr, (size_t)fp->ff_size);
-               if (bp) {
-                       if (VTOHFS(vp)->jnl && (bp->b_flags & B_LOCKED) == 0) {
-                               bp->b_flags |= B_INVAL;         /* data no longer needed */
-                       }
-                       brelse(bp);
+               bcopy((char *)buf_dataptr(bp), fp->ff_symlinkptr, (size_t)fp->ff_size);
+
+               if (VTOHFS(vp)->jnl && (buf_flags(bp) & B_LOCKED) == 0) {
+                       buf_markinvalid(bp);            /* data no longer needed */
                }
+               buf_brelse(bp);
        }
-       retval = uiomove((caddr_t)fp->ff_symlinkptr, (int)fp->ff_size, ap->a_uio);
-#if 1
+       error = uiomove((caddr_t)fp->ff_symlinkptr, (int)fp->ff_size, ap->a_uio);
+
        /*
         * Keep track blocks read
         */
-       if ((VTOHFS(vp)->hfc_stage == HFC_RECORDING) && (retval == 0)) {
+       if ((VTOHFS(vp)->hfc_stage == HFC_RECORDING) && (error == 0)) {
                
                /*
                 * If this file hasn't been seen since the start of
@@ -3011,156 +5845,71 @@ hfs_readlink(ap)
                        VTOF(vp)->ff_bytesread += fp->ff_size;
                
        //      if (VTOF(vp)->ff_bytesread > fp->ff_size)
-       //              cp->c_flag |= C_ACCESS;
-       }
-#endif
-       return (retval);
-}
-
-/*
- * Lock an cnode. If its already locked, set the WANT bit and sleep.
-#% lock                vp      U L U
-#
- vop_lock {
-     IN struct vnode *vp;
-     IN int flags;
-     IN struct proc *p;
-     */
-
-static int
-hfs_lock(ap)
-       struct vop_lock_args /* {
-               struct vnode *a_vp;
-               int a_flags;
-               struct proc *a_p;
-       } */ *ap;
-{
-       struct vnode *vp = ap->a_vp;
-       struct cnode *cp = VTOC(vp);
-
-       return (lockmgr(&cp->c_lock, ap->a_flags, &vp->v_interlock, ap->a_p));
-}
-
-/*
- * Unlock an cnode.
-#% unlock      vp      L U L
-#
- vop_unlock {
-     IN struct vnode *vp;
-     IN int flags;
-     IN struct proc *p;
-
-     */
-static int
-hfs_unlock(ap)
-       struct vop_unlock_args /* {
-               struct vnode *a_vp;
-               int a_flags;
-               struct proc *a_p;
-       } */ *ap;
-{
-       struct vnode *vp = ap->a_vp;
-       struct cnode *cp = VTOC(vp);
-#if 0
-       if (!lockstatus(&cp->c_lock)) {
-               printf("hfs_unlock: vnode %s wasn't locked!\n",
-                       cp->c_desc.cd_nameptr ? cp->c_desc.cd_nameptr : "");
+       //              cp->c_touch_acctime = TRUE;
        }
-#endif
-       return (lockmgr(&cp->c_lock, ap->a_flags | LK_RELEASE,
-               &vp->v_interlock, ap->a_p));
-}
-
-
-/*
- * Print out the contents of a cnode.
-#% print       vp      = = =
-#
- vop_print {
-     IN struct vnode *vp;
-     */
-static int
-hfs_print(ap)
-       struct vop_print_args /* {
-               struct vnode *a_vp;
-       } */ *ap;
-{
-       struct vnode * vp = ap->a_vp;
-       struct cnode *cp = VTOC(vp);
 
-       printf("tag VT_HFS, cnid %d, on dev %d, %d", cp->c_cnid,
-               major(cp->c_dev), minor(cp->c_dev));
-#if FIFO
-       if (vp->v_type == VFIFO)
-               fifo_printinfo(vp);
-#endif /* FIFO */
-       lockmgr_printinfo(&cp->c_lock);
-       printf("\n");
-       return (0);
+exit:
+       hfs_unlock(cp);
+       return (error);
 }
 
 
 /*
- * Check for a locked cnode.
-#% islocked    vp      = = =
-#
- vop_islocked {
-     IN struct vnode *vp;
-
-     */
-static int
-hfs_islocked(ap)
-       struct vop_islocked_args /* {
-               struct vnode *a_vp;
-       } */ *ap;
-{
-       return (lockstatus(&VTOC(ap->a_vp)->c_lock));
-}
-
-/*
-
-#% pathconf    vp      L L L
-#
- vop_pathconf {
-     IN struct vnode *vp;
-     IN int name;
-     OUT register_t *retval;
-
-     */
-static int
-hfs_pathconf(ap)
-       struct vop_pathconf_args /* {
+ * Get configurable pathname variables.
+ */
+int
+hfs_vnop_pathconf(ap)
+       struct vnop_pathconf_args /* {
                struct vnode *a_vp;
                int a_name;
                int *a_retval;
+               vfs_context_t a_context;
        } */ *ap;
 {
-       int retval = 0;
 
+       int std_hfs = (VTOHFS(ap->a_vp)->hfs_flags & HFS_STANDARD);
        switch (ap->a_name) {
        case _PC_LINK_MAX:
-               if (VTOVCB(ap->a_vp)->vcbSigWord == kHFSPlusSigWord)
+               if (std_hfs == 0){
                        *ap->a_retval = HFS_LINK_MAX;
-               else
+               }
+#if CONFIG_HFS_STD
+               else {
                        *ap->a_retval = 1;
+               }
+#endif
                break;
        case _PC_NAME_MAX:
-               *ap->a_retval = kHFSPlusMaxFileNameBytes;       /* max # of characters x max utf8 representation */
+               if (std_hfs == 0) {
+                       *ap->a_retval = kHFSPlusMaxFileNameChars;  /* 255 */
+               }
+#if CONFIG_HFS_STD
+               else {
+                       *ap->a_retval = kHFSMaxFileNameChars;  /* 31 */
+               }
+#endif
                break;
        case _PC_PATH_MAX:
-               *ap->a_retval = PATH_MAX; /* 1024 */
+               *ap->a_retval = PATH_MAX;  /* 1024 */
                break;
        case _PC_PIPE_BUF:
                *ap->a_retval = PIPE_BUF;
                break;
        case _PC_CHOWN_RESTRICTED:
-               *ap->a_retval = 1;
+               *ap->a_retval = 200112;         /* _POSIX_CHOWN_RESTRICTED */
                break;
        case _PC_NO_TRUNC:
-               *ap->a_retval = 0;
+               *ap->a_retval = 200112;         /* _POSIX_NO_TRUNC */
                break;
        case _PC_NAME_CHARS_MAX:
-               *ap->a_retval = kHFSPlusMaxFileNameChars;
+               if (std_hfs == 0) {
+                       *ap->a_retval = kHFSPlusMaxFileNameChars; /* 255 */
+               }
+#if CONFIG_HFS_STD
+               else {
+                       *ap->a_retval = kHFSMaxFileNameChars; /* 31 */
+               }
+#endif
                break;
        case _PC_CASE_SENSITIVE:
                if (VTOHFS(ap->a_vp)->hfs_flags & HFS_CASE_SENSITIVE)
@@ -3171,385 +5920,400 @@ hfs_pathconf(ap)
        case _PC_CASE_PRESERVING:
                *ap->a_retval = 1;
                break;
+       case _PC_FILESIZEBITS:
+               /* number of bits to store max file size */
+               if (std_hfs == 0) {
+                       *ap->a_retval = 64;     
+               }
+#if CONFIG_HFS_STD
+               else {
+                       *ap->a_retval = 32;
+               }
+#endif
+               break;
+       case _PC_XATTR_SIZE_BITS:
+               /* Number of bits to store maximum extended attribute size */
+               *ap->a_retval = HFS_XATTR_SIZE_BITS;
+               break;
        default:
-               retval = EINVAL;
+               return (EINVAL);
        }
 
-       return (retval);
+       return (0);
 }
 
-
 /*
- * Advisory record locking support
-#% advlock     vp      U U U
-#
- vop_advlock {
-     IN struct vnode *vp;
-     IN caddr_t id;
-     IN int op;
-     IN struct flock *fl;
-     IN int flags;
-
-     */
-static int
-hfs_advlock(ap)
-       struct vop_advlock_args /* {
-               struct vnode *a_vp;
-               caddr_t  a_id;
-               int a_op;
-               struct flock *a_fl;
-               int a_flags;
-       } */ *ap;
+ * Prepares a fork for cat_update by making sure ff_size and ff_blocks
+ * are no bigger than the valid data on disk thus reducing the chance
+ * of exposing uninitialised data in the event of a non clean unmount.
+ * fork_buf is where to put the temporary copy if required.  (It can
+ * be inside pfork.)
+ */
+const struct cat_fork *
+hfs_prepare_fork_for_update(filefork_t *ff,
+                                                       const struct cat_fork *cf,
+                                                       struct cat_fork *cf_buf,
+                                                       uint32_t block_size)
 {
-       struct vnode *vp = ap->a_vp;
-       struct flock *fl = ap->a_fl;
-       struct hfslockf *lock;
-       struct filefork *fork;
-       off_t start, end;
-       int retval;
+       if (!ff)
+               return NULL;
 
-       /* Only regular files can have locks */
-       if (vp->v_type != VREG)
-               return (EISDIR);
+       if (!cf)
+               cf = &ff->ff_data;
+       if (!cf_buf)
+               cf_buf = &ff->ff_data;
 
-       fork = VTOF(ap->a_vp);
-       /*
-        * Avoid the common case of unlocking when cnode has no locks.
-        */
-       if (fork->ff_lockf == (struct hfslockf *)0) {
-               if (ap->a_op != F_SETLK) {
-                       fl->l_type = F_UNLCK;
-                       return (0);
-               }
-       }
-       /*
-        * Convert the flock structure into a start and end.
-        */
-       start = 0;
-       switch (fl->l_whence) {
-       case SEEK_SET:
-       case SEEK_CUR:
-               /*
-                * Caller is responsible for adding any necessary offset
-                * when SEEK_CUR is used.
-                */
-               start = fl->l_start;
-               break;
-       case SEEK_END:
-               start = fork->ff_size + fl->l_start;
-               break;
-       default:
-               return (EINVAL);
-       }
+       off_t max_size = ff->ff_size;
+
+       // Check first invalid range
+       if (!TAILQ_EMPTY(&ff->ff_invalidranges))
+               max_size = TAILQ_FIRST(&ff->ff_invalidranges)->rl_start;
 
-       if (fl->l_len == 0)
-               end = -1;
-       else if (fl->l_len > 0)
-               end = start + fl->l_len - 1;
-       else { /* l_len is negative */
-               end = start - 1;
-               start += fl->l_len;
+       if (!ff->ff_unallocblocks && ff->ff_size <= max_size)
+               return cf; // Nothing to do
+
+       if (ff->ff_blocks < ff->ff_unallocblocks) {
+               panic("hfs: ff_blocks %d is less than unalloc blocks %d\n",
+                         ff->ff_blocks, ff->ff_unallocblocks);
        }
-       if (start < 0)
-               return (EINVAL);
+
+       struct cat_fork *out = cf_buf;
+
+       if (out != cf)
+               bcopy(cf, out, sizeof(*cf));
+
+       // Adjust cf_blocks for cf_vblocks
+       out->cf_blocks -= out->cf_vblocks;
 
        /*
-        * Create the hfslockf structure
-        */
-       MALLOC(lock, struct hfslockf *, sizeof *lock, M_LOCKF, M_WAITOK);
-       lock->lf_start = start;
-       lock->lf_end = end;
-       lock->lf_id = ap->a_id;
-       lock->lf_fork = fork;
-       lock->lf_type = fl->l_type;
-       lock->lf_next = (struct hfslockf *)0;
-       TAILQ_INIT(&lock->lf_blkhd);
-       lock->lf_flags = ap->a_flags;
-       /*
-        * Do the requested operation.
+        * Here we trim the size with the updated cf_blocks.  This is
+        * probably unnecessary now because the invalid ranges should
+        * catch this (but that wasn't always the case).
         */
-       switch(ap->a_op) {
-       case F_SETLK:
-               retval = hfs_setlock(lock);
-               break;
-       case F_UNLCK:
-               retval = hfs_clearlock(lock);
-               FREE(lock, M_LOCKF);
-               break;
-       case F_GETLK:
-               retval = hfs_getlock(lock, fl);
-               FREE(lock, M_LOCKF);
-               break;
-       default:
-               retval = EINVAL;
-               _FREE(lock, M_LOCKF);
-            break;
-       }
-
-       return (retval);
-}
+       off_t alloc_bytes = hfs_blk_to_bytes(out->cf_blocks, block_size);
+       if (out->cf_size > alloc_bytes)
+               out->cf_size = alloc_bytes;
 
+       // Trim cf_size to first invalid range
+       if (out->cf_size > max_size)
+               out->cf_size = max_size;
 
+       return out;
+}
 
 /*
- * Update the access, modified, and node change times as specified
- * by the C_ACCESS, C_UPDATE, and C_CHANGE flags respectively. The
- * C_MODIFIED flag is used to specify that the node needs to be
- * updated but that the times have already been set. The access and
- * modified times are input parameters but the node change time is
- * always taken from the current time. If waitfor is set, then wait
- * for the disk write of the node to complete.
+ * Update a cnode's on-disk metadata.
+ *
+ * The cnode must be locked exclusive.  See declaration for possible
+ * options.
  */
-/*
-#% update      vp      L L L
-       IN struct vnode *vp;
-       IN struct timeval *access;
-       IN struct timeval *modify;
-       IN int waitfor;
-*/
-static int
-hfs_update(ap)
-       struct vop_update_args /* {
-               struct vnode *a_vp;
-               struct timeval *a_access;
-               struct timeval *a_modify;
-               int a_waitfor;
-       } */ *ap;
+int
+hfs_update(struct vnode *vp, int options)
 {
-       struct vnode *vp = ap->a_vp;
-       struct cnode *cp = VTOC(ap->a_vp);
+       struct cnode *cp = VTOC(vp);
        struct proc *p;
-       struct cat_fork *dataforkp = NULL;
-       struct cat_fork *rsrcforkp = NULL;
+       const struct cat_fork *dataforkp = NULL;
+       const struct cat_fork *rsrcforkp = NULL;
        struct cat_fork datafork;
-       int updateflag;
+       struct cat_fork rsrcfork;
        struct hfsmount *hfsmp;
+       int lockflags;
        int error;
+       uint32_t tstate = 0;
+
+       if (ISSET(cp->c_flag, C_NOEXISTS))
+               return 0;
 
+       p = current_proc();
        hfsmp = VTOHFS(vp);
 
-       /* XXX do we really want to clear the sytem cnode flags here???? */
-       if (((vp->v_flag & VSYSTEM) && (cp->c_cnid < kHFSFirstUserCatalogNodeID))||
-           (VTOHFS(vp)->hfs_flags & HFS_READ_ONLY) ||
-           (cp->c_mode == 0)) {
-               cp->c_flag &= ~(C_ACCESS | C_CHANGE | C_MODIFIED | C_UPDATE);
+       if (((vnode_issystem(vp) && (cp->c_cnid < kHFSFirstUserCatalogNodeID))) || 
+               hfsmp->hfs_catalog_vp == NULL){
                return (0);
        }
-
-       updateflag = cp->c_flag & (C_ACCESS | C_CHANGE | C_MODIFIED | C_UPDATE | C_FORCEUPDATE);
-
-       /* Nothing to update. */
-       if (updateflag == 0) {
+       if ((hfsmp->hfs_flags & HFS_READ_ONLY) || (cp->c_mode == 0)) {
+               CLR(cp->c_flag, C_MODIFIED | C_MINOR_MOD | C_NEEDS_DATEADDED);
+               cp->c_touch_acctime = 0;
+               cp->c_touch_chgtime = 0;
+               cp->c_touch_modtime = 0;
                return (0);
        }
-       /* HFS standard doesn't have access times. */
-       if ((updateflag == C_ACCESS) && (VTOVCB(vp)->vcbSigWord == kHFSSigWord)) {
-               return (0);
+       if (kdebug_enable) {
+               if (cp->c_touch_acctime || cp->c_atime != cp->c_attr.ca_atimeondisk)
+                       tstate |= DBG_HFS_UPDATE_ACCTIME;
+               if (cp->c_touch_modtime)
+                       tstate |= DBG_HFS_UPDATE_MODTIME;
+               if (cp->c_touch_chgtime)
+                       tstate |= DBG_HFS_UPDATE_CHGTIME;
+
+               if (cp->c_flag & C_MODIFIED)
+                       tstate |= DBG_HFS_UPDATE_MODIFIED;
+               if (ISSET(options, HFS_UPDATE_FORCE))
+                       tstate |= DBG_HFS_UPDATE_FORCE;
+               if (cp->c_flag & C_NEEDS_DATEADDED)
+                       tstate |= DBG_HFS_UPDATE_DATEADDED;
+               if (cp->c_flag & C_MINOR_MOD)
+                       tstate |= DBG_HFS_UPDATE_MINOR;
+       }
+       hfs_touchtimes(hfsmp, cp);
+
+       if (!ISSET(cp->c_flag, C_MODIFIED | C_MINOR_MOD)
+               && !hfs_should_save_atime(cp)) {
+               // Nothing to update
+               return 0;
        }
-       if (updateflag & C_ACCESS) {
+
+       KDBG(HFSDBG_UPDATE | DBG_FUNC_START, VM_KERNEL_ADDRPERM(vp), tstate);
+
+       bool check_txn = false;
+
+       if (!ISSET(options, HFS_UPDATE_FORCE) && !ISSET(cp->c_flag, C_MODIFIED)) {
                /*
-                * When the access time is the only thing changing
-                * then make sure its sufficiently newer before
-                * committing it to disk.
+                * This must be a minor modification.  If the current
+                * transaction already has an update for this node, then we
+                * bundle in the modification.
                 */
-               if ((updateflag == C_ACCESS) &&
-                   (ap->a_access->tv_sec < (cp->c_atime + ATIME_ONDISK_ACCURACY))) {
-                       return (0);
+               if (hfsmp->jnl
+                       && journal_current_txn(hfsmp->jnl) == cp->c_update_txn) {
+                       check_txn = true;
+               } else {
+                       tstate |= DBG_HFS_UPDATE_SKIPPED;
+                       error = 0;
+                       goto exit;
                }
-               cp->c_atime = ap->a_access->tv_sec;
-       }
-       if (updateflag & C_UPDATE) {
-               cp->c_mtime = ap->a_modify->tv_sec;
-               cp->c_mtime_nsec = ap->a_modify->tv_usec * 1000;
        }
-       if (updateflag & C_CHANGE) {
-               cp->c_ctime = time.tv_sec;
-               /*
-                * HFS dates that WE set must be adjusted for DST
-                */
-               if ((VTOVCB(vp)->vcbSigWord == kHFSSigWord) && gTimeZone.tz_dsttime) {
-                       cp->c_ctime += 3600;
-                       cp->c_mtime = cp->c_ctime;
-               }
+
+       if ((error = hfs_start_transaction(hfsmp)) != 0)
+               goto exit;
+
+       if (check_txn
+               && journal_current_txn(hfsmp->jnl) != cp->c_update_txn) {
+               hfs_end_transaction(hfsmp);
+               tstate |= DBG_HFS_UPDATE_SKIPPED;
+               error = 0;
+               goto exit;
        }
-       
+
        if (cp->c_datafork)
                dataforkp = &cp->c_datafork->ff_data;
        if (cp->c_rsrcfork)
                rsrcforkp = &cp->c_rsrcfork->ff_data;
 
-       p = current_proc();
-
-       /*
-        * For delayed allocations updates are
-        * postponed until an fsync or the file
-        * gets written to disk.
-        *
-        * Deleted files can defer meta data updates until inactive.
-        *
-        * If we're ever called with the C_FORCEUPDATE flag though
-        * we have to do the update.
-        */
-       if (ISSET(cp->c_flag, C_FORCEUPDATE) == 0 &&
-           (ISSET(cp->c_flag, C_DELETED) || 
-           (dataforkp && cp->c_datafork->ff_unallocblocks) ||
-           (rsrcforkp && cp->c_rsrcfork->ff_unallocblocks))) {
-               if (updateflag & (C_CHANGE | C_UPDATE))
-                       hfs_volupdate(hfsmp, VOL_UPDATE, 0);    
-               cp->c_flag &= ~(C_ACCESS | C_CHANGE | C_UPDATE);
-               cp->c_flag |= C_MODIFIED;
-
-               HFS_KNOTE(vp, NOTE_ATTRIB);
+    /*
+     * Modify the values passed to cat_update based on whether or not
+     * the file has invalid ranges or borrowed blocks.
+     */
+    dataforkp = hfs_prepare_fork_for_update(cp->c_datafork, NULL, &datafork, hfsmp->blockSize);
+       rsrcforkp = hfs_prepare_fork_for_update(cp->c_rsrcfork, NULL, &rsrcfork, hfsmp->blockSize);
 
-               return (0);
-       }
+       if (__improbable(kdebug_enable & KDEBUG_TRACE)) {
+               long dbg_parms[NUMPARMS];
+               int  dbg_namelen;
 
+               dbg_namelen = NUMPARMS * sizeof(long);
+               vn_getpath(vp, (char *)dbg_parms, &dbg_namelen);
 
-       // XXXdbg
-       hfs_global_shared_lock_acquire(hfsmp);
-       if (hfsmp->jnl) {
-               if ((error = journal_start_transaction(hfsmp->jnl)) != 0) {
-                       hfs_global_shared_lock_release(hfsmp);
-                       return error;
-           }
-       }
-                       
+               if (dbg_namelen < (int)sizeof(dbg_parms))
+                       memset((char *)dbg_parms + dbg_namelen, 0, sizeof(dbg_parms) - dbg_namelen);
 
-       /*
-        * For files with invalid ranges (holes) the on-disk
-        * field representing the size of the file (cf_size)
-        * must be no larger than the start of the first hole.
-        */
-       if (dataforkp && !CIRCLEQ_EMPTY(&cp->c_datafork->ff_invalidranges)) {
-               bcopy(dataforkp, &datafork, sizeof(datafork));
-               datafork.cf_size = CIRCLEQ_FIRST(&cp->c_datafork->ff_invalidranges)->rl_start;
-               dataforkp = &datafork;
-       } else if (dataforkp && (cp->c_datafork->ff_unallocblocks != 0)) {
-               // always make sure the block count and the size 
-               // of the file match the number of blocks actually
-               // allocated to the file on disk
-               bcopy(dataforkp, &datafork, sizeof(datafork));
-               // make sure that we don't assign a negative block count
-               if (cp->c_datafork->ff_blocks < cp->c_datafork->ff_unallocblocks) {
-                   panic("hfs: ff_blocks %d is less than unalloc blocks %d\n",
-                         cp->c_datafork->ff_blocks, cp->c_datafork->ff_unallocblocks);
-               }
-               datafork.cf_blocks = (cp->c_datafork->ff_blocks - cp->c_datafork->ff_unallocblocks);
-               datafork.cf_size   = datafork.cf_blocks * HFSTOVCB(hfsmp)->blockSize;
-               dataforkp = &datafork;
+               kdebug_lookup_gen_events(dbg_parms, dbg_namelen, (void *)vp, TRUE);
        }
 
        /*
         * Lock the Catalog b-tree file.
-        * A shared lock is sufficient since an update doesn't change
-        * the tree and the lock on vp protects the cnode.
         */
-       error = hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_SHARED, p);
-       if (error) {
-               if (hfsmp->jnl) {
-                       journal_end_transaction(hfsmp->jnl);
-               }
-               hfs_global_shared_lock_release(hfsmp);
-               return (error);
-       }
+       lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG, HFS_EXCLUSIVE_LOCK);
 
-       /* XXX - waitfor is not enforced */
        error = cat_update(hfsmp, &cp->c_desc, &cp->c_attr, dataforkp, rsrcforkp);
 
-        /* Unlock the Catalog b-tree file. */
-       (void) hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_RELEASE, p);
+       if (hfsmp->jnl)
+               cp->c_update_txn = journal_current_txn(hfsmp->jnl);
 
-       if (updateflag & (C_CHANGE | C_UPDATE | C_FORCEUPDATE))
-               hfs_volupdate(hfsmp, VOL_UPDATE, 0);    
+       hfs_systemfile_unlock(hfsmp, lockflags);
 
-       /* After the updates are finished, clear the flags */
-       cp->c_flag &= ~(C_ACCESS | C_CHANGE | C_MODIFIED | C_UPDATE | C_FORCEUPDATE);
+       CLR(cp->c_flag, C_MODIFIED | C_MINOR_MOD);
 
-       // XXXdbg
-       if (hfsmp->jnl) {
-           journal_end_transaction(hfsmp->jnl);
-       }
-       hfs_global_shared_lock_release(hfsmp);
+       hfs_end_transaction(hfsmp);
 
-       HFS_KNOTE(vp, NOTE_ATTRIB);
-       
-       return (error);
+exit:
+
+       KDBG(HFSDBG_UPDATE | DBG_FUNC_END, VM_KERNEL_ADDRPERM(vp), tstate, error);
+
+       return error;
 }
 
 /*
  * Allocate a new node
- *
- * Upon leaving, namei buffer must be freed.
- *
  */
-static int
-hfs_makenode(mode, dvp, vpp, cnp)
-       int mode;
-       struct vnode *dvp;
-       struct vnode **vpp;
-       struct componentname *cnp;
+int
+hfs_makenode(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
+             struct vnode_attr *vap, vfs_context_t ctx)
 {
-       struct cnode *cp;
-       struct cnode *dcp;
+       struct cnode *cp = NULL;
+       struct cnode *dcp = NULL;
        struct vnode *tvp;
        struct hfsmount *hfsmp;
-       struct timeval tv;
-       struct proc *p;
        struct cat_desc in_desc, out_desc;
        struct cat_attr attr;
-       cat_cookie_t cookie = {0};
-       int error, started_tr = 0, grabbed_lock = 0;
+       struct timeval tv;
+       int lockflags;
+       int error, started_tr = 0;
        enum vtype vnodetype;
+       int mode;
+       int newvnode_flags = 0;
+       u_int32_t gnv_flags = 0;
+       int protectable_target = 0;
+       int nocache = 0;
+
+#if CONFIG_PROTECT
+       struct cprotect *entry = NULL;
+       int32_t cp_class = -1;
+
+       /* 
+        * By default, it's OK for AKS to overrride our target class preferences.
+        */
+       uint32_t keywrap_flags = CP_KEYWRAP_DIFFCLASS;
+
+       if (VATTR_IS_ACTIVE(vap, va_dataprotect_class)) {
+               cp_class = (int32_t)vap->va_dataprotect_class;
+               /* 
+                * Since the user specifically requested this target class be used,
+                * we want to fail this creation operation if we cannot wrap to their 
+                * target class. The CP_KEYWRAP_DIFFCLASS bit says that it is OK to
+                * use a different class than the one specified, so we turn that off
+                * now.
+                */
+               keywrap_flags &= ~CP_KEYWRAP_DIFFCLASS;
+       }
+       int protected_mount = 0;        
+#endif
 
-       p = cnp->cn_proc;
+
+       if ((error = hfs_lock(VTOC(dvp), HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT)))
+               return (error);
+
+       /* set the cnode pointer only after successfully acquiring lock */
        dcp = VTOC(dvp);
+
+       /* Don't allow creation of new entries in open-unlinked directories */
+       if ((error = hfs_checkdeleted(dcp))) {
+               hfs_unlock(dcp);
+               return error;
+       }
+
+       dcp->c_flag |= C_DIR_MODIFICATION;
+
        hfsmp = VTOHFS(dvp);
+
        *vpp = NULL;
        tvp = NULL;
-       bzero(&out_desc, sizeof(out_desc));
+       out_desc.cd_flags = 0;
+       out_desc.cd_nameptr = NULL;
 
-       if ((mode & S_IFMT) == 0)
-               mode |= S_IFREG;
-       vnodetype = IFTOVT(mode);
+       vnodetype = vap->va_type;
+       if (vnodetype == VNON)
+               vnodetype = VREG;
+       mode = MAKEIMODE(vnodetype, vap->va_mode);
 
-       /* Check if unmount in progress */
-       if (VTOVFS(dvp)->mnt_kern_flag & MNTK_UNMOUNT) {
-               error = EPERM;
-               goto exit;
+       if (S_ISDIR (mode) || S_ISREG (mode)) {
+               protectable_target = 1;
        }
+       
+
        /* Check if were out of usable disk space. */
-       if ((suser(cnp->cn_cred, NULL) != 0) && (hfs_freeblks(hfsmp, 1) <= 0)) {
+       if ((hfs_freeblks(hfsmp, 1) == 0) && (vfs_context_suser(ctx) != 0)) {
                error = ENOSPC;
                goto exit;
        }
 
+       microtime(&tv);
+
        /* Setup the default attributes */
        bzero(&attr, sizeof(attr));
        attr.ca_mode = mode;
-       attr.ca_nlink = vnodetype == VDIR ? 2 : 1;
-       attr.ca_mtime = time.tv_sec;
-       attr.ca_mtime_nsec = time.tv_usec * 1000;
-       if ((VTOVCB(dvp)->vcbSigWord == kHFSSigWord) && gTimeZone.tz_dsttime) {
-               attr.ca_mtime += 3600;  /* Same as what hfs_update does */
-       }
-       attr.ca_atime = attr.ca_ctime = attr.ca_itime = attr.ca_mtime;
-       if (VTOVFS(dvp)->mnt_flag & MNT_UNKNOWNPERMISSIONS) {
-               attr.ca_uid = hfsmp->hfs_uid;
-               attr.ca_gid = hfsmp->hfs_gid;
+       attr.ca_linkcount = 1;
+       if (VATTR_IS_ACTIVE(vap, va_rdev)) {
+               attr.ca_rdev = vap->va_rdev;
+       }
+       if (VATTR_IS_ACTIVE(vap, va_create_time)) {
+               VATTR_SET_SUPPORTED(vap, va_create_time);
+               attr.ca_itime = vap->va_create_time.tv_sec;
        } else {
-               if (vnodetype == VLNK)
-                       attr.ca_uid = dcp->c_uid;
-               else
-                       attr.ca_uid = cnp->cn_cred->cr_uid;
-               attr.ca_gid = dcp->c_gid;
+               attr.ca_itime = tv.tv_sec;
+       }
+#if CONFIG_HFS_STD
+       if ((hfsmp->hfs_flags & HFS_STANDARD) && gTimeZone.tz_dsttime) {
+               attr.ca_itime += 3600;  /* Same as what hfs_update does */
+       }
+#endif
+       attr.ca_atime = attr.ca_ctime = attr.ca_mtime = attr.ca_itime;
+       attr.ca_atimeondisk = attr.ca_atime;
+       if (VATTR_IS_ACTIVE(vap, va_flags)) {
+               VATTR_SET_SUPPORTED(vap, va_flags);
+               attr.ca_flags = vap->va_flags;
+       }
+       
+       /* 
+        * HFS+ only: all files get ThreadExists
+        * HFSX only: dirs get HasFolderCount
+        */
+       if (!(hfsmp->hfs_flags & HFS_STANDARD)) {
+               if (vnodetype == VDIR) {
+                       if (hfsmp->hfs_flags & HFS_FOLDERCOUNT)
+                               attr.ca_recflags = kHFSHasFolderCountMask;
+               } else {
+                       attr.ca_recflags = kHFSThreadExistsMask;
+               }
+       }
+
+#if CONFIG_PROTECT     
+       if (cp_fs_protected(hfsmp->hfs_mp)) {
+               protected_mount = 1;
        }
        /*
-        * Don't tag as a special file (BLK or CHR) until *after*
-        * hfs_getnewvnode is called.  This insures that any
-        * alias checking is defered until hfs_mknod completes.
+        * On a content-protected HFS+/HFSX filesystem, files and directories
+        * cannot be created without atomically setting/creating the EA that 
+        * contains the protection class metadata and keys at the same time, in
+        * the same transaction.  As a result, pre-set the "EAs exist" flag
+        * on the cat_attr for protectable catalog record creations.  This will
+        * cause the cnode creation routine in hfs_getnewvnode to mark the cnode
+        * as having EAs.
+        */
+       if ((protected_mount) && (protectable_target)) {
+               attr.ca_recflags |= kHFSHasAttributesMask;
+               /* delay entering in the namecache */
+               nocache = 1;
+       }
+#endif
+
+
+       /* 
+        * Add the date added to the item. See above, as
+        * all of the dates are set to the itime.
         */
-       if (vnodetype == VBLK || vnodetype == VCHR)
-               attr.ca_mode = (attr.ca_mode & ~S_IFMT) | S_IFREG;
+       hfs_write_dateadded (&attr, attr.ca_atime);
+
+       /* Initialize the gen counter to 1 */
+       hfs_write_gencount(&attr, (uint32_t)1);
+
+       attr.ca_uid = vap->va_uid;
+       attr.ca_gid = vap->va_gid;
+       VATTR_SET_SUPPORTED(vap, va_mode);
+       VATTR_SET_SUPPORTED(vap, va_uid);
+       VATTR_SET_SUPPORTED(vap, va_gid);
+
+#if QUOTA
+       /* check to see if this node's creation would cause us to go over
+        * quota.  If so, abort this operation.
+        */
+       if (hfsmp->hfs_flags & HFS_QUOTAS) {
+               if ((error = hfs_quotacheck(hfsmp, 1, attr.ca_uid, attr.ca_gid,
+                                                                       vfs_context_ucred(ctx)))) {
+                       goto exit;
+               }
+       }       
+#endif
+
 
        /* Tag symlinks with a type and creator. */
        if (vnodetype == VLNK) {
@@ -3559,431 +6323,649 @@ hfs_makenode(mode, dvp, vpp, cnp)
                fip->fdType    = SWAP_BE32(kSymLinkFileType);
                fip->fdCreator = SWAP_BE32(kSymLinkCreator);
        }
-       if ((attr.ca_mode & S_ISGID) &&
-           !groupmember(dcp->c_gid, cnp->cn_cred) &&
-           suser(cnp->cn_cred, NULL)) {
-               attr.ca_mode &= ~S_ISGID;
-       }
-       if (cnp->cn_flags & ISWHITEOUT)
-               attr.ca_flags |= UF_OPAQUE;
 
        /* Setup the descriptor */
-       bzero(&in_desc, sizeof(in_desc));
-       in_desc.cd_nameptr = cnp->cn_nameptr;
+       in_desc.cd_nameptr = (const u_int8_t *)cnp->cn_nameptr;
        in_desc.cd_namelen = cnp->cn_namelen;
-       in_desc.cd_parentcnid = dcp->c_cnid;
+       in_desc.cd_parentcnid = dcp->c_fileid;
        in_desc.cd_flags = S_ISDIR(mode) ? CD_ISDIR : 0;
+       in_desc.cd_hint = dcp->c_childhint;
+       in_desc.cd_encoding = 0;
 
-       // XXXdbg
-       hfs_global_shared_lock_acquire(hfsmp);
-       grabbed_lock = 1;
-       if (hfsmp->jnl) {
-           if ((error = journal_start_transaction(hfsmp->jnl)) != 0) {
+#if CONFIG_PROTECT
+       /*
+        * To preserve file creation atomicity with regards to the content protection EA,
+        * we must create the file in the catalog and then write out its EA in the same
+        * transaction.  
+        * 
+        * We only denote the target class in this EA; key generation is not completed
+        * until the file has been inserted into the catalog and will be done
+        * in a separate transaction.
+        */
+       if ((protected_mount) && (protectable_target)) {
+               error = cp_setup_newentry(hfsmp, dcp, cp_class, attr.ca_mode, &entry);
+               if (error) {
                        goto exit;
-           }
-               started_tr = 1;
+               }
+       }
+#endif
+
+       if ((error = hfs_start_transaction(hfsmp)) != 0) {
+           goto exit;
+       }
+       started_tr = 1;
+
+       // have to also lock the attribute file because cat_create() needs
+       // to check that any fileID it wants to use does not have orphaned
+       // attributes in it.
+       lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG | SFL_ATTRIBUTE, HFS_EXCLUSIVE_LOCK);
+       cnid_t new_id;
+
+       /* Reserve some space in the Catalog file. */
+       if ((error = cat_preflight(hfsmp, CAT_CREATE, NULL, 0))) {
+               hfs_systemfile_unlock(hfsmp, lockflags);
+               goto exit;
+       }
+
+       if ((error = cat_acquire_cnid(hfsmp, &new_id))) {
+               hfs_systemfile_unlock (hfsmp, lockflags);
+               goto exit;
+       }
+
+       error = cat_create(hfsmp, new_id, &in_desc, &attr, &out_desc);
+       if (error == 0) {
+               /* Update the parent directory */
+               dcp->c_childhint = out_desc.cd_hint;    /* Cache directory's location */
+               dcp->c_entries++;
+
+               if (vnodetype == VDIR) {
+                       INC_FOLDERCOUNT(hfsmp, dcp->c_attr);
+               }
+               dcp->c_dirchangecnt++;
+               hfs_incr_gencount(dcp);
+
+               dcp->c_touch_chgtime = dcp->c_touch_modtime = true;
+               dcp->c_flag |= C_MODIFIED;
+
+               hfs_update(dcp->c_vp, 0);
+
+#if CONFIG_PROTECT
+               /*
+                * If we are creating a content protected file, now is when
+                * we create the EA. We must create it in the same transaction
+                * that creates the file.  We can also guarantee that the file 
+                * MUST exist because we are still holding the catalog lock
+                * at this point.
+                */
+               if ((attr.ca_fileid != 0) && (protected_mount) && (protectable_target)) {
+                       error = cp_setxattr (NULL, entry, hfsmp, attr.ca_fileid, XATTR_CREATE);
+                       
+                       if (error) {
+                               int delete_err;
+                               /* 
+                                * If we fail the EA creation, then we need to delete the file. 
+                                * Luckily, we are still holding all of the right locks.
+                                */
+                               delete_err = cat_delete (hfsmp, &out_desc, &attr);
+                               if (delete_err == 0) {
+                                       /* Update the parent directory */
+                                       if (dcp->c_entries > 0)
+                                               dcp->c_entries--;
+                                       dcp->c_dirchangecnt++;
+                                       dcp->c_ctime = tv.tv_sec;
+                                       dcp->c_mtime = tv.tv_sec;
+                                       (void) cat_update(hfsmp, &dcp->c_desc, &dcp->c_attr, NULL, NULL);
+                               }
+
+                               /* Emit EINVAL if we fail to create EA*/
+                               error = EINVAL;
+                       }
+               }               
+#endif
+       }
+       hfs_systemfile_unlock(hfsmp, lockflags);
+       if (error)
+               goto exit;
+
+       uint32_t txn = hfsmp->jnl ? journal_current_txn(hfsmp->jnl) : 0;
+
+       /* Invalidate negative cache entries in the directory */
+       if (dcp->c_flag & C_NEG_ENTRIES) {
+               cache_purge_negatives(dvp);
+               dcp->c_flag &= ~C_NEG_ENTRIES;
+       }
+
+       hfs_volupdate(hfsmp, vnodetype == VDIR ? VOL_MKDIR : VOL_MKFILE,
+               (dcp->c_cnid == kHFSRootFolderID));
+
+       // XXXdbg
+       // have to end the transaction here before we call hfs_getnewvnode()
+       // because that can cause us to try and reclaim a vnode on a different
+       // file system which could cause us to start a transaction which can
+       // deadlock with someone on that other file system (since we could be
+       // holding two transaction locks as well as various vnodes and we did
+       // not obtain the locks on them in the proper order).
+       //
+       // NOTE: this means that if the quota check fails or we have to update
+       //       the change time on a block-special device that those changes
+       //       will happen as part of independent transactions.
+       //
+       if (started_tr) {
+           hfs_end_transaction(hfsmp);
+           started_tr = 0;
+       }
+
+#if CONFIG_PROTECT
+       /* 
+        * At this point, we must have encountered success with writing the EA.
+        * Destroy our temporary cprotect (which had no keys). 
+        */
+
+       if ((attr.ca_fileid != 0) && (protected_mount) && (protectable_target))  {
+               cp_entry_destroy (hfsmp, entry);
+               entry = NULL;
+       }
+#endif
+       gnv_flags |= GNV_CREATE;
+       if (nocache) {
+               gnv_flags |= GNV_NOCACHE;
        }
 
        /*
-        * Reserve some space in the Catalog file.
-        *
-        * (we also add CAT_DELETE since our getnewvnode
-        *  request can cause an hfs_inactive call to
-        *  delete an unlinked file)
+        * Create a vnode for the object just created.
+        * 
+        * NOTE: Maintaining the cnode lock on the parent directory is important,
+        * as it prevents race conditions where other threads want to look up entries 
+        * in the directory and/or add things as we are in the process of creating
+        * the vnode below.  However, this has the potential for causing a 
+        * double lock panic when dealing with shadow files on a HFS boot partition. 
+        * The panic could occur if we are not cleaning up after ourselves properly 
+        * when done with a shadow file or in the error cases.  The error would occur if we 
+        * try to create a new vnode, and then end up reclaiming another shadow vnode to 
+        * create the new one.  However, if everything is working properly, this should
+        * be a non-issue as we would never enter that reclaim codepath.
+        * 
+        * The cnode is locked on successful return.
         */
-       if ((error = cat_preflight(hfsmp, CAT_CREATE | CAT_DELETE, &cookie, p))) {
+       error = hfs_getnewvnode(hfsmp, dvp, cnp, &out_desc, gnv_flags, &attr, 
+                                                       NULL, &tvp, &newvnode_flags);
+       if (error)
                goto exit;
+
+       cp = VTOC(tvp);
+
+       cp->c_update_txn = txn;
+
+       struct  doc_tombstone *ut;
+       ut = get_uthread_doc_tombstone();
+       if (   ut->t_lastop_document_id != 0 
+           && ut->t_lastop_parent == dvp
+           && ut->t_lastop_parent_vid == vnode_vid(dvp)
+           && strcmp((char *)ut->t_lastop_filename, (const char *)cp->c_desc.cd_nameptr) == 0) {
+               struct FndrExtendedDirInfo *fip = (struct FndrExtendedDirInfo *)((char *)&cp->c_attr.ca_finderinfo + 16);
+
+               //printf("CREATE: preserving doc-id %lld on %s\n", ut->t_lastop_document_id, ut->t_lastop_filename);
+               fip->document_id = (uint32_t)(ut->t_lastop_document_id & 0xffffffff);
+
+               cp->c_bsdflags |= UF_TRACKED;
+               cp->c_flag |= C_MODIFIED;
+
+               if ((error = hfs_start_transaction(hfsmp)) == 0) {
+                       lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG, HFS_EXCLUSIVE_LOCK);
+
+                       (void) cat_update(hfsmp, &cp->c_desc, &cp->c_attr, NULL, NULL);
+
+                       hfs_systemfile_unlock (hfsmp, lockflags);
+                       (void) hfs_end_transaction(hfsmp);
+               }
+
+               clear_tombstone_docid(ut, hfsmp, cp);       // will send the docid-changed fsevent
+       } else if (ut->t_lastop_document_id != 0) {
+               int len = cnp->cn_namelen;
+               if (len == 0) {
+                       len = strlen(cnp->cn_nameptr);
+               }
+
+               if (is_ignorable_temp_name(cnp->cn_nameptr, cnp->cn_namelen)) {
+                       // printf("CREATE: not clearing tombstone because %s is a temp name.\n", cnp->cn_nameptr);
+               } else {
+                       // Clear the tombstone because the thread is not recreating the same path
+                       // printf("CREATE: clearing tombstone because %s is NOT a temp name.\n", cnp->cn_nameptr);
+                       clear_tombstone_docid(ut, hfsmp, NULL);
+               }
        }
 
-       /* Lock catalog b-tree */
-       error = hfs_metafilelocking(VTOHFS(dvp), kHFSCatalogFileID, LK_EXCLUSIVE, p);
-       if (error)
-               goto exit;
+       if ((hfsmp->hfs_flags & HFS_CS_HOTFILE_PIN) && (vnode_isfastdevicecandidate(dvp) && !vnode_isautocandidate(dvp))) {
+
+               //printf("hfs: flagging %s (fileid: %d) as VFASTDEVCANDIDATE (dvp name: %s)\n",
+               //       cnp->cn_nameptr ? cnp->cn_nameptr : "<NONAME>",
+               //       cp->c_fileid,
+               //       dvp->v_name ? dvp->v_name : "no-dir-name");
+
+               //
+               // On new files we set the FastDevCandidate flag so that
+               // any new blocks allocated to it will be pinned.
+               //
+               cp->c_attr.ca_recflags |= kHFSFastDevCandidateMask;
+               vnode_setfastdevicecandidate(tvp);
+
+               //
+               // properly inherit auto-cached flags
+               //
+               if (vnode_isautocandidate(dvp)) {
+                       cp->c_attr.ca_recflags |= kHFSAutoCandidateMask;
+                       vnode_setautocandidate(tvp);
+               }
 
-       error = cat_create(hfsmp, &in_desc, &attr, &out_desc);
 
-       /* Unlock catalog b-tree */
-       (void) hfs_metafilelocking(VTOHFS(dvp), kHFSCatalogFileID, LK_RELEASE, p);              
-       if (error)
-               goto exit;
+               //
+               // We also want to add it to the hotfile adoption list so
+               // that it will eventually land in the hotfile btree
+               //
+               (void) hfs_addhotfile(tvp);
+       }
        
-       /* Update the parent directory */
-       dcp->c_childhint = out_desc.cd_hint;    /* Cache directory's location */
-       dcp->c_nlink++;
-       dcp->c_entries++;
-       dcp->c_flag |= C_CHANGE | C_UPDATE;
-       tv = time;
-       (void) VOP_UPDATE(dvp, &tv, &tv, 0);
-       if (vnodetype == VDIR) {
-               HFS_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
-       } else {
-               HFS_KNOTE(dvp, NOTE_WRITE);
-       };
+       *vpp = tvp;
 
-       hfs_volupdate(hfsmp, vnodetype == VDIR ? VOL_MKDIR : VOL_MKFILE,
-               (dcp->c_cnid == kHFSRootFolderID));
+#if CONFIG_PROTECT
+       /* 
+        * Now that we have a vnode-in-hand, generate keys for this namespace item.
+        * If we fail to create the keys, then attempt to delete the item from the 
+        * namespace.  If we can't delete the item, that's not desirable but also not fatal..
+        *  All of the places which deal with restoring/unwrapping keys must also be 
+        * prepared to encounter an entry that does not have keys. 
+        */
+       if ((protectable_target) && (protected_mount)) {
+               struct cprotect *keyed_entry = NULL;
 
-       // XXXdbg
-       // have to end the transaction here before we call hfs_getnewvnode()
-       // because that can cause us to try and reclaim a vnode on a different
-       // file system which could cause us to start a transaction which can
-       // deadlock with someone on that other file system (since we could be
-       // holding two transaction locks as well as various vnodes and we did
-       // not obtain the locks on them in the proper order).
-    //
-       // NOTE: this means that if the quota check fails or we have to update
-       //       the change time on a block-special device that those changes
-       //       will happen as part of independent transactions.
-       //
-       if (started_tr) {
-               journal_end_transaction(hfsmp->jnl);
-               started_tr = 0;
-       }
-       if (grabbed_lock) {
-               hfs_global_shared_lock_release(hfsmp);
-               grabbed_lock = 0;
-       }
+               if (cp->c_cpentry == NULL) {
+                       panic ("hfs_makenode: no cpentry for cnode (%p)", cp);
+               }
 
-       /* Create a vnode for the object just created: */
-       error = hfs_getnewvnode(hfsmp, NULL, &out_desc, 0, &attr, NULL, &tvp);
-       if (error)
-               goto exit;
+               error = cp_generate_keys (hfsmp, cp, CP_CLASS(cp->c_cpentry->cp_pclass), keywrap_flags, &keyed_entry);
+               if (error == 0) {
+                       /* 
+                        * Upon success, the keys were generated and written out. 
+                        * Update the cp pointer in the cnode.
+                        */
+                       cp_replace_entry (hfsmp, cp, keyed_entry);
+                       if (nocache) {
+                               cache_enter (dvp, tvp, cnp);
+                       }
+               }
+               else {
+                       /* If key creation OR the setxattr failed, emit EPERM to userland */
+                       error = EPERM;
 
-       // XXXdbg
-       cache_enter(dvp, tvp, cnp);
+                       /* 
+                        * Beware! This slightly violates the lock ordering for the
+                        * cnode/vnode 'tvp'.  Ordinarily, you must acquire the truncate lock
+                        * which guards file size changes before acquiring the normal cnode lock
+                        * and calling hfs_removefile on an item.  
+                        * 
+                        * However, in this case, we are still holding the directory lock so 
+                        * 'tvp' is not lookup-able and it was a newly created vnode so it 
+                        * cannot have any content yet. The only reason we are initiating 
+                        * the removefile is because we could not generate content protection keys
+                        * for this namespace item. Note also that we pass a '1' in the allow_dirs
+                        * argument for hfs_removefile because we may be creating a directory here.
+                        * 
+                        * All this to say that while it is technically a violation it is 
+                        * impossible to race with another thread for this cnode so it is safe.
+                        */
+                       int err = hfs_removefile (dvp, tvp, cnp, 0, 0, 1, NULL, 0); 
+                       if (err) {
+                               printf("hfs_makenode: removefile failed (%d) for CP entry %p\n", err, tvp);
+                       }
 
-#if QUOTA
-       cp = VTOC(tvp);
-       /* 
-        * We call hfs_chkiq with FORCE flag so that if we
-        * fall through to the rmdir we actually have 
-        * accounted for the inode
-       */
-       if ((error = hfs_getinoquota(cp)) ||
-           (error = hfs_chkiq(cp, 1, cnp->cn_cred, FORCE))) {
-               if (tvp->v_type == VDIR)
-                       VOP_RMDIR(dvp,tvp, cnp);
-               else
-                       VOP_REMOVE(dvp,tvp, cnp);
+                       /* Release the cnode lock and mark the vnode for termination */ 
+                       hfs_unlock (cp);
+                       err = vnode_recycle (tvp);
+                       if (err) {
+                               printf("hfs_makenode: vnode_recycle failed (%d) for CP entry %p\n", err, tvp);
+                       }
 
-               // because VOP_RMDIR and VOP_REMOVE already
-               // have done the vput()
-               dvp = NULL;
-               goto exit;
+                       /* Drop the iocount on the new vnode to force reclamation/recycling */
+                       vnode_put (tvp);
+                       cp = NULL;
+                       *vpp = NULL;
+               }       
        }
-#endif /* QUOTA */
+#endif
 
-       /*
-        * restore vtype and mode for VBLK and VCHR
+#if QUOTA
+       /* 
+        * Once we create this vnode, we need to initialize its quota data 
+        * structures, if necessary.  We know that it is OK to just go ahead and 
+        * initialize because we've already validated earlier (through the hfs_quotacheck 
+        * function) to see if creating this cnode/vnode would cause us to go over quota. 
         */
-       if (vnodetype == VBLK || vnodetype == VCHR) {
-               struct cnode *cp;
-
-               cp = VTOC(tvp);
-               cp->c_mode = mode;
-               tvp->v_type = IFTOVT(mode);
-               cp->c_flag |= C_CHANGE;
-               tv = time;
-               if ((error = VOP_UPDATE(tvp, &tv, &tv, 1))) {
-                       vput(tvp);
-                       goto exit;
+       if (hfsmp->hfs_flags & HFS_QUOTAS) {
+               if (cp) {
+                       /* cp could have been zeroed earlier */ 
+                       (void) hfs_getinoquota(cp); 
                }
        }
+#endif
 
-       *vpp = tvp;
 exit:
        cat_releasedesc(&out_desc);
+       
+#if CONFIG_PROTECT
+       /*  
+        * We may have jumped here in error-handling various situations above.
+        * If we haven't already dumped the temporary CP used to initialize
+        * the file atomically, then free it now. cp_entry_destroy should null
+        * out the pointer if it was called already.
+        */
+       if (entry) {
+               cp_entry_destroy (hfsmp, entry);
+               entry = NULL;
+       }       
+#endif
 
-       cat_postflight(hfsmp, &cookie, p);
-
-       if ((cnp->cn_flags & (HASBUF | SAVESTART)) == HASBUF) {
-               char *tmp = cnp->cn_pnbuf;
-               cnp->cn_pnbuf = NULL;
-               cnp->cn_flags &= ~HASBUF;
-               FREE_ZONE(tmp, cnp->cn_pnlen, M_NAMEI);
-       }
        /*
-        * Check if a file is located in the "Cleanup At Startup"
-        * directory.  If it is then tag it as NODUMP so that we
-        * can be lazy about zero filling data holes.
+        * Make sure we release cnode lock on dcp.
         */
-       if ((error == 0) && dvp && (vnodetype == VREG) &&
-           (dcp->c_desc.cd_nameptr != NULL) &&
-           (strcmp(dcp->c_desc.cd_nameptr, CARBON_TEMP_DIR_NAME) == 0)) {
-               struct vnode *ddvp;
-               cnid_t parid;
-
-               parid = dcp->c_parentcnid;
-               vput(dvp);
-               dvp = NULL;
-
-               /*
-                * The parent of "Cleanup At Startup" should
-                * have the ASCII name of the userid.
-                */
-               if (VFS_VGET(HFSTOVFS(hfsmp), &parid, &ddvp) == 0) {
-                       if (VTOC(ddvp)->c_desc.cd_nameptr) {
-                               uid_t uid;
-
-                               uid = strtoul(VTOC(ddvp)->c_desc.cd_nameptr, 0, 0);
-                               if (uid == cp->c_uid || uid == cnp->cn_cred->cr_uid) {
-                               cp->c_flags |= UF_NODUMP;
-                               cp->c_flag |= C_CHANGE;
-                               }
-                       }
-                       vput(ddvp);
-               }
+       if (dcp) {
+               dcp->c_flag &= ~C_DIR_MODIFICATION;
+               wakeup((caddr_t)&dcp->c_flag);
+               
+               hfs_unlock(dcp);
        }
-       if (dvp)
-               vput(dvp);
-
-       if (started_tr) {
-           journal_end_transaction(hfsmp->jnl);
-               started_tr = 0;
+       if (error == 0 && cp != NULL) {
+               hfs_unlock(cp);
        }
-       if (grabbed_lock) {
-               hfs_global_shared_lock_release(hfsmp);
-               grabbed_lock = 0;
+       if (started_tr) {
+           hfs_end_transaction(hfsmp);
+           started_tr = 0;
        }
 
        return (error);
 }
 
 
-static int
-hfs_vgetrsrc(struct hfsmount *hfsmp, struct vnode *vp, struct vnode **rvpp, struct proc *p)
+/*
+ * hfs_vgetrsrc acquires a resource fork vnode corresponding to the
+ * cnode that is found in 'vp'.  The cnode should be locked upon entry
+ * and will be returned locked, but it may be dropped temporarily.
+ *
+ * If the resource fork vnode does not exist, HFS will attempt to acquire an
+ * empty (uninitialized) vnode from VFS so as to avoid deadlocks with
+ * jetsam. If we let the normal getnewvnode code produce the vnode for us
+ * we would be doing so while holding the cnode lock of our cnode.
+ * 
+ * On success, *rvpp wlll hold the resource fork vnode with an
+ * iocount.  *Don't* forget the vnode_put.
+ */
+int
+hfs_vgetrsrc(struct hfsmount *hfsmp, struct vnode *vp, struct vnode **rvpp)
 {
-       struct vnode *rvp;
+       struct vnode *rvp = NULLVP;
+       struct vnode *empty_rvp = NULLVP;
+       struct vnode *dvp = NULLVP;
        struct cnode *cp = VTOC(vp);
        int error;
+       int vid;
 
+       if (vnode_vtype(vp) == VDIR) {
+               return EINVAL;
+       }
+       
+restart:
+       /* Attempt to use existing vnode */
        if ((rvp = cp->c_rsrc_vp)) {
-               /* Use exising vnode */
-               error = vget(rvp, 0, p);
+               vid = vnode_vid(rvp);
+
+               // vnode_getwithvid can block so we need to drop the cnode lock
+               hfs_unlock(cp);
+
+               error = vnode_getwithvid(rvp, vid);
+
+               hfs_lock_always(cp, HFS_EXCLUSIVE_LOCK);
+
+               /*
+                * When our lock was relinquished, the resource fork
+                * could have been recycled.  Check for this and try
+                * again.
+                */
+               if (error == ENOENT)
+                       goto restart;
+
                if (error) {
-                       char * name = VTOC(vp)->c_desc.cd_nameptr;
+                       const char * name = (const char *)VTOC(vp)->c_desc.cd_nameptr;
 
                        if (name)
-                               printf("hfs_vgetrsrc: couldn't get"
-                                       " resource fork for %s\n", name);
+                               printf("hfs_vgetrsrc: couldn't get resource"
+                                      " fork for %s, vol=%s, err=%d\n", name, hfsmp->vcbVN, error);
                        return (error);
                }
        } else {
                struct cat_fork rsrcfork;
+               struct componentname cn;
+               struct cat_desc *descptr = NULL;
+               struct cat_desc to_desc;
+               char delname[32];
+               int lockflags;
+               int newvnode_flags = 0;
 
-               /* Lock catalog b-tree */
-               error = hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_SHARED, p);
-               if (error)
-                       return (error);
-
-               /* Get resource fork data */
-               error = cat_lookup(hfsmp, &cp->c_desc, 1, (struct cat_desc *)0,
-                               (struct cat_attr *)0, &rsrcfork);
+               /* 
+                * In this case, we don't currently see a resource fork vnode attached
+                * to this cnode.  In most cases, we were called from a read-only VNOP
+                * like getattr, so it should be safe to drop the cnode lock and then 
+                * re-acquire it.  
+                *
+                * Here, we drop the lock so that we can acquire an empty/husk
+                * vnode so that we don't deadlock against jetsam.  
+                *
+                * It does not currently appear possible to hold the truncate lock via
+                * FS re-entrancy when we get to this point. (8/2014)
+                */
+               hfs_unlock (cp);
 
-               /* Unlock the Catalog */
-               (void) hfs_metafilelocking(hfsmp, kHFSCatalogFileID, LK_RELEASE, p);
-               if (error)
-                       return (error);
-               
-               error = hfs_getnewvnode(hfsmp, cp, &cp->c_desc, 1, &cp->c_attr,
-                                       &rsrcfork, &rvp);
-               if (error)
-                       return (error);
-       }
+               error = vnode_create_empty (&empty_rvp); 
 
-       *rvpp = rvp;
-       return (0);
-}
+               hfs_lock_always (cp, HFS_EXCLUSIVE_LOCK);
 
+               if (error) {
+                       /* If acquiring the 'empty' vnode failed, then nothing to clean up */
+                       return error;
+               }
 
-static void
-filt_hfsdetach(struct knote *kn)
-{
-       struct vnode *vp;
-       int result;
-       struct proc *p = current_proc();
+               /* 
+                * We could have raced with another thread here while we dropped our cnode
+                * lock.  See if the cnode now has a resource fork vnode and restart if appropriate.
+                *
+                * Note: We just released the cnode lock, so there is a possibility that the
+                * cnode that we just acquired has been deleted or even removed from disk
+                * completely, though this is unlikely. If the file is open-unlinked, the 
+                * check below will resolve it for us.  If it has been completely 
+                * removed (even from the catalog!), then when we examine the catalog 
+                * directly, below, while holding the catalog lock, we will not find the
+                * item and we can fail out properly.
+                */
+               if (cp->c_rsrc_vp) {
+                       /* Drop the empty vnode before restarting */
+                       vnode_put (empty_rvp);
+                       empty_rvp = NULL;
+                       rvp = NULL;
+                       goto restart;
+               }
        
-       vp = (struct vnode *)kn->kn_hook;
-       if (1) {        /* ! KNDETACH_VNLOCKED */
-               result = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
-               if (result) return;
-       };
+               /*
+                * hfs_vgetsrc may be invoked for a cnode that has already been marked
+                * C_DELETED.  This is because we need to continue to provide rsrc
+                * fork access to open-unlinked files.  In this case, build a fake descriptor
+                * like in hfs_removefile.  If we don't do this, buildkey will fail in
+                * cat_lookup because this cnode has no name in its descriptor.
+                */
+               if ((cp->c_flag & C_DELETED ) && (cp->c_desc.cd_namelen == 0)) {
+                       bzero (&to_desc, sizeof(to_desc));
+                       bzero (delname, 32);
+                       MAKE_DELETED_NAME(delname, sizeof(delname), cp->c_fileid);
+                       to_desc.cd_nameptr = (const u_int8_t*) delname;
+                       to_desc.cd_namelen = strlen(delname);
+                       to_desc.cd_parentcnid = hfsmp->hfs_private_desc[FILE_HARDLINKS].cd_cnid;
+                       to_desc.cd_flags = 0;
+                       to_desc.cd_cnid = cp->c_cnid;
 
-       result = KNOTE_DETACH(&VTOC(vp)->c_knotes, kn);
-       
-       if (1) {        /* ! KNDETACH_VNLOCKED */
-               VOP_UNLOCK(vp, 0, p);
-       };
-}
+                       descptr = &to_desc;
+               }
+               else {
+                       descptr = &cp->c_desc;
+               }
 
-/*ARGSUSED*/
-static int
-filt_hfsread(struct knote *kn, long hint)
-{
-       struct vnode *vp = (struct vnode *)kn->kn_fp->f_data;
 
-       if (hint == NOTE_REVOKE) {
+               lockflags = hfs_systemfile_lock(hfsmp, SFL_CATALOG, HFS_SHARED_LOCK);
+
                /*
-                * filesystem is gone, so set the EOF flag and schedule 
-                * the knote for deletion.
+                * We call cat_idlookup (instead of cat_lookup) below because we can't
+                * trust the descriptor in the provided cnode for lookups at this point.  
+                * Between the time of the original lookup of this vnode and now, the 
+                * descriptor could have gotten swapped or replaced.  If this occurred, 
+                * the parent/name combo originally desired may not necessarily be provided
+                * if we use the descriptor.  Even worse, if the vnode represents
+                * a hardlink, we could have removed one of the links from the namespace
+                * but left the descriptor alone, since hfs_unlink does not invalidate
+                * the descriptor in the cnode if other links still point to the inode.
+                * 
+                * Consider the following (slightly contrived) scenario:
+                * /tmp/a <--> /tmp/b (hardlinks).
+                * 1. Thread A: open rsrc fork on /tmp/b.
+                * 1a. Thread A: does lookup, goes out to lunch right before calling getnamedstream.
+                * 2. Thread B does 'mv /foo/b /tmp/b'
+                * 2. Thread B succeeds.
+                * 3. Thread A comes back and wants rsrc fork info for /tmp/b.  
+                * 
+                * Even though the hardlink backing /tmp/b is now eliminated, the descriptor
+                * is not removed/updated during the unlink process.  So, if you were to
+                * do a lookup on /tmp/b, you'd acquire an entirely different record's resource
+                * fork.
+                * 
+                * As a result, we use the fileid, which should be invariant for the lifetime
+                * of the cnode (possibly barring calls to exchangedata).
+                *
+                * Addendum: We can't do the above for HFS standard since we aren't guaranteed to
+                * have thread records for files.  They were only required for directories.  So
+                * we need to do the lookup with the catalog name. This is OK since hardlinks were
+                * never allowed on HFS standard.
                 */
-               kn->kn_flags |= (EV_EOF | EV_ONESHOT);
-               return (1);
-       }
 
-       kn->kn_data = VTOF(vp)->ff_size - kn->kn_fp->f_offset;
-       return (kn->kn_data != 0);
-}
+               /* Get resource fork data */
+               if ((hfsmp->hfs_flags & HFS_STANDARD) == 0) {
+                       error = cat_idlookup (hfsmp, cp->c_fileid, 0, 1, NULL, NULL, &rsrcfork);
+               }
+#if CONFIG_HFS_STD
+               else {
+                       /* 
+                        * HFS standard only:
+                        * 
+                        * Get the resource fork for this item with a cat_lookup call, but do not 
+                        * force a case lookup since HFS standard is case-insensitive only. We 
+                        * don't want the descriptor; just the fork data here. If we tried to
+                        * do a ID lookup (via thread record -> catalog record), then we might fail
+                        * prematurely since, as noted above, thread records were not strictly required
+                        * on files in HFS.
+                        */
+                       error = cat_lookup (hfsmp, descptr, 1, 0, (struct cat_desc*)NULL, 
+                                       (struct cat_attr*)NULL, &rsrcfork, NULL);
+               }
+#endif
 
-/*ARGSUSED*/
-static int
-filt_hfswrite(struct knote *kn, long hint)
-{
-       if (hint == NOTE_REVOKE) {
+               hfs_systemfile_unlock(hfsmp, lockflags);
+               if (error) {
+                       /* Drop our 'empty' vnode ! */
+                       vnode_put (empty_rvp);
+                       return (error);
+               }
                /*
-                * filesystem is gone, so set the EOF flag and schedule 
-                * the knote for deletion.
+                * Supply hfs_getnewvnode with a component name. 
                 */
-               kn->kn_flags |= (EV_EOF | EV_ONESHOT);
-       }
-       
-       kn->kn_data = 0;
-       return (1);
-}
-
-static int
-filt_hfsvnode(struct knote *kn, long hint)
-{
-
-       if (kn->kn_sfflags & hint)
-               kn->kn_fflags |= hint;
-       if (hint == NOTE_REVOKE) {
-               kn->kn_flags |= EV_EOF;
-               return (1);
-       }
-       return (kn->kn_fflags != 0);
-}
-
-static struct filterops hfsread_filtops = 
-       { 1, NULL, filt_hfsdetach, filt_hfsread };
-static struct filterops hfswrite_filtops = 
-       { 1, NULL, filt_hfsdetach, filt_hfswrite };
-static struct filterops hfsvnode_filtops = 
-       { 1, NULL, filt_hfsdetach, filt_hfsvnode };
-
-/*
- #
- #% kqfilt_add vp      L L L
- #
- vop_kqfilt_add
-       IN struct vnode *vp;
-       IN struct knote *kn;
-       IN struct proc *p;
- */
-static int
-hfs_kqfilt_add(ap)
-       struct vop_kqfilt_add_args /* {
-               struct vnode *a_vp;
-               struct knote *a_kn;
-               struct proc *p;
-       } */ *ap;
-{
-       struct vnode *vp = ap->a_vp;
-       struct knote *kn = ap->a_kn;
-
-       switch (kn->kn_filter) {
-       case EVFILT_READ:
-               if (vp->v_type == VREG) {
-                       kn->kn_fop = &hfsread_filtops;
-               } else {
-                       return EINVAL;
-               };
-               break;
-       case EVFILT_WRITE:
-               if (vp->v_type == VREG) {
-                       kn->kn_fop = &hfswrite_filtops;
-               } else {
-                       return EINVAL;
-               };
-               break;
-       case EVFILT_VNODE:
-               kn->kn_fop = &hfsvnode_filtops;
-               break;
-       default:
-               return (1);
-       }
-
-       kn->kn_hook = (caddr_t)vp;
-
-       /* simple_lock(&vp->v_pollinfo.vpi_lock); */
-       KNOTE_ATTACH(&VTOC(vp)->c_knotes, kn);
-       /* simple_unlock(&vp->v_pollinfo.vpi_lock); */
+               cn.cn_pnbuf = NULL;
+               if (descptr->cd_nameptr) {
+                       MALLOC_ZONE(cn.cn_pnbuf, caddr_t, MAXPATHLEN, M_NAMEI, M_WAITOK);
+                       cn.cn_nameiop = LOOKUP;
+                       cn.cn_flags = ISLASTCN | HASBUF;
+                       cn.cn_context = NULL;
+                       cn.cn_pnlen = MAXPATHLEN;
+                       cn.cn_nameptr = cn.cn_pnbuf;
+                       cn.cn_hash = 0;
+                       cn.cn_consume = 0;
+                       cn.cn_namelen = snprintf(cn.cn_nameptr, MAXPATHLEN,
+                                                "%s%s", descptr->cd_nameptr,
+                                                _PATH_RSRCFORKSPEC);
+                       // Should never happen because cn.cn_nameptr won't ever be long...
+                       if (cn.cn_namelen >= MAXPATHLEN) {
+                               FREE_ZONE(cn.cn_pnbuf, cn.cn_pnlen, M_NAMEI);
+                               /* Drop our 'empty' vnode ! */
+                               vnode_put (empty_rvp);
+                               return ENAMETOOLONG;
+                               
+                       }
+               }
+               dvp = vnode_getparent(vp);
+               
+               /* 
+                * We are about to call hfs_getnewvnode and pass in the vnode that we acquired
+                * earlier when we were not holding any locks. The semantics of GNV_USE_VP require that
+                * either hfs_getnewvnode consume the vnode and vend it back to us, properly initialized,
+                * or it will consume/dispose of it properly if it errors out.
+                */ 
+               rvp = empty_rvp;
+               
+               error = hfs_getnewvnode(hfsmp, dvp, cn.cn_pnbuf ? &cn : NULL,
+                                       descptr, (GNV_WANTRSRC | GNV_SKIPLOCK | GNV_USE_VP), 
+                                                               &cp->c_attr, &rsrcfork, &rvp, &newvnode_flags);
+                       
+               if (dvp)
+                       vnode_put(dvp);
+               if (cn.cn_pnbuf)
+                       FREE_ZONE(cn.cn_pnbuf, cn.cn_pnlen, M_NAMEI);
+               if (error)
+                       return (error);
+       }  /* End 'else' for rsrc fork not existing */
 
+       *rvpp = rvp;
        return (0);
 }
 
-/*
- #
- #% kqfilt_remove      vp      L L L
- #
- vop_kqfilt_remove
-       IN struct vnode *vp;
-       IN uintptr_t ident;
-       IN struct proc *p;
- */
-static int
-hfs_kqfilt_remove(ap)
-       struct vop_kqfilt_remove_args /* {
-               struct vnode *a_vp;
-               uintptr_t ident;
-               struct proc *p;
-       } */ *ap;
-{
-       struct vnode *vp = ap->a_vp;
-       uintptr_t ident = ap->a_ident;
-       int result;
-
-       result = ENOTSUP; /* XXX */
-       
-       return (result);
-}
-
 /*
  * Wrapper for special device reads
  */
-static int
+int
 hfsspec_read(ap)
-       struct vop_read_args /* {
+       struct vnop_read_args /* {
                struct vnode *a_vp;
                struct uio *a_uio;
                int  a_ioflag;
-               struct ucred *a_cred;
+               vfs_context_t a_context;
        } */ *ap;
 {
        /*
         * Set access flag.
         */
-       VTOC(ap->a_vp)->c_flag |= C_ACCESS;
-       return (VOCALL (spec_vnodeop_p, VOFFSET(vop_read), ap));
+       VTOC(ap->a_vp)->c_touch_acctime = TRUE;
+       return (VOCALL (spec_vnodeop_p, VOFFSET(vnop_read), ap));
 }
 
 /*
  * Wrapper for special device writes
  */
-static int
+int
 hfsspec_write(ap)
-       struct vop_write_args /* {
+       struct vnop_write_args /* {
                struct vnode *a_vp;
                struct uio *a_uio;
                int  a_ioflag;
-               struct ucred *a_cred;
+               vfs_context_t a_context;
        } */ *ap;
 {
        /*
         * Set update and change flags.
         */
-       VTOC(ap->a_vp)->c_flag |= C_CHANGE | C_UPDATE;
-       return (VOCALL (spec_vnodeop_p, VOFFSET(vop_write), ap));
+       VTOC(ap->a_vp)->c_touch_chgtime = TRUE;
+       VTOC(ap->a_vp)->c_touch_modtime = TRUE;
+       return (VOCALL (spec_vnodeop_p, VOFFSET(vnop_write), ap));
 }
 
 /*
@@ -3991,23 +6973,25 @@ hfsspec_write(ap)
  *
  * Update the times on the cnode then do device close.
  */
-static int
+int
 hfsspec_close(ap)
-       struct vop_close_args /* {
+       struct vnop_close_args /* {
                struct vnode *a_vp;
                int  a_fflag;
-               struct ucred *a_cred;
-               struct proc *a_p;
+               vfs_context_t a_context;
        } */ *ap;
 {
        struct vnode *vp = ap->a_vp;
-       struct cnode *cp = VTOC(vp);
+       struct cnode *cp;
 
-       simple_lock(&vp->v_interlock);
-       if (ap->a_vp->v_usecount > 1)
-               CTIMES(cp, &time, &time);
-       simple_unlock(&vp->v_interlock);
-       return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap));
+       if (vnode_isinuse(ap->a_vp, 0)) {
+               if (hfs_lock(VTOC(vp), HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT) == 0) {
+                       cp = VTOC(vp);
+                       hfs_touchtimes(VTOHFS(vp), cp);
+                       hfs_unlock(cp);
+               }
+       }
+       return (VOCALL (spec_vnodeop_p, VOFFSET(vnop_close), ap));
 }
 
 #if FIFO
@@ -4016,20 +7000,18 @@ hfsspec_close(ap)
  */
 static int
 hfsfifo_read(ap)
-       struct vop_read_args /* {
+       struct vnop_read_args /* {
                struct vnode *a_vp;
                struct uio *a_uio;
                int  a_ioflag;
-               struct ucred *a_cred;
+               vfs_context_t a_context;
        } */ *ap;
 {
-       extern int (**fifo_vnodeop_p)(void *);
-
        /*
         * Set access flag.
         */
-       VTOC(ap->a_vp)->c_flag |= C_ACCESS;
-       return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_read), ap));
+       VTOC(ap->a_vp)->c_touch_acctime = TRUE;
+       return (VOCALL (fifo_vnodeop_p, VOFFSET(vnop_read), ap));
 }
 
 /*
@@ -4037,20 +7019,19 @@ hfsfifo_read(ap)
  */
 static int
 hfsfifo_write(ap)
-       struct vop_write_args /* {
+       struct vnop_write_args /* {
                struct vnode *a_vp;
                struct uio *a_uio;
                int  a_ioflag;
-               struct ucred *a_cred;
+               vfs_context_t a_context;
        } */ *ap;
 {
-       extern int (**fifo_vnodeop_p)(void *);
-
        /*
         * Set update and change flags.
         */
-       VTOC(ap->a_vp)->c_flag |= C_CHANGE | C_UPDATE;
-       return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_write), ap));
+       VTOC(ap->a_vp)->c_touch_chgtime = TRUE;
+       VTOC(ap->a_vp)->c_touch_modtime = TRUE;
+       return (VOCALL (fifo_vnodeop_p, VOFFSET(vnop_write), ap));
 }
 
 /*
@@ -4060,272 +7041,337 @@ hfsfifo_write(ap)
  */
 static int
 hfsfifo_close(ap)
-       struct vop_close_args /* {
+       struct vnop_close_args /* {
                struct vnode *a_vp;
                int  a_fflag;
-               struct ucred *a_cred;
-               struct proc *a_p;
+               vfs_context_t a_context;
        } */ *ap;
 {
-       extern int (**fifo_vnodeop_p)(void *);
        struct vnode *vp = ap->a_vp;
-       struct cnode *cp = VTOC(vp);
+       struct cnode *cp;
 
-       simple_lock(&vp->v_interlock);
-       if (ap->a_vp->v_usecount > 1)
-               CTIMES(cp, &time, &time);
-       simple_unlock(&vp->v_interlock);
-       return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap));
+       if (vnode_isinuse(ap->a_vp, 1)) {
+               if (hfs_lock(VTOC(vp), HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT) == 0) {
+                       cp = VTOC(vp);
+                       hfs_touchtimes(VTOHFS(vp), cp);
+                       hfs_unlock(cp);
+               }
+       }
+       return (VOCALL (fifo_vnodeop_p, VOFFSET(vnop_close), ap));
 }
 
-/*
- * kqfilt_add wrapper for fifos.
- *
- * Fall through to hfs kqfilt_add routines if needed 
+
+#endif /* FIFO */
+
+/* 
+ * Getter for the document_id 
+ * the document_id is stored in FndrExtendedFileInfo/FndrExtendedDirInfo
  */
-int
-hfsfifo_kqfilt_add(ap)
-       struct vop_kqfilt_add_args *ap;
+static u_int32_t 
+hfs_get_document_id_internal(const uint8_t *finderinfo, mode_t mode)
 {
-       extern int (**fifo_vnodeop_p)(void *);
-       int error;
+       const uint8_t *finfo = NULL;
+       u_int32_t doc_id = 0;
+       
+       /* overlay the FinderInfo to the correct pointer, and advance */
+       finfo = finderinfo + 16;
+
+       if (S_ISDIR(mode) || S_ISREG(mode)) {
+               const struct FndrExtendedFileInfo *extinfo = (const struct FndrExtendedFileInfo *)finfo;
+               doc_id = extinfo->document_id;
+       } else if (S_ISDIR(mode)) {
+               const struct FndrExtendedDirInfo *extinfo = (const struct FndrExtendedDirInfo *)finfo;
+               doc_id = extinfo->document_id;
+       }       
+
+       return doc_id;
+}
 
-       error = VOCALL(fifo_vnodeop_p, VOFFSET(vop_kqfilt_add), ap);
-       if (error)
-               error = hfs_kqfilt_add(ap);
-       return (error);
+
+/* getter(s) for document id */
+u_int32_t
+hfs_get_document_id(struct cnode *cp)
+{
+       return (hfs_get_document_id_internal((u_int8_t*)cp->c_finderinfo,
+           cp->c_attr.ca_mode));
+}
+
+/* If you have finderinfo and mode, you can use this */
+u_int32_t
+hfs_get_document_id_from_blob(const uint8_t *finderinfo, mode_t mode)
+{
+       return (hfs_get_document_id_internal(finderinfo, mode));
 }
 
 /*
- * kqfilt_remove wrapper for fifos.
- *
- * Fall through to hfs kqfilt_remove routines if needed 
+ * Synchronize a file's in-core state with that on disk.
  */
 int
-hfsfifo_kqfilt_remove(ap)
-       struct vop_kqfilt_remove_args *ap;
+hfs_vnop_fsync(ap)
+       struct vnop_fsync_args /* {
+               struct vnode *a_vp;
+               int a_waitfor;
+               vfs_context_t a_context;
+       } */ *ap;
 {
-       extern int (**fifo_vnodeop_p)(void *);
+       struct vnode* vp = ap->a_vp;
        int error;
 
-       error = VOCALL(fifo_vnodeop_p, VOFFSET(vop_kqfilt_remove), ap);
+       /* Note: We check hfs flags instead of vfs mount flag because during 
+        * read-write update, hfs marks itself read-write much earlier than
+        * the vfs, and hence won't result in skipping of certain writes like 
+        * zero'ing out of unused nodes, creation of hotfiles btree, etc. 
+        */
+       if (VTOHFS(vp)->hfs_flags & HFS_READ_ONLY) {
+               return 0;               
+       }
+
+       /*
+        * No need to call cp_handle_vnop to resolve fsync().  Any dirty data
+        * should have caused the keys to be unwrapped at the time the data was
+        * put into the UBC, either at mmap/pagein/read-write.  If we did manage
+        * to let this by, then strategy will auto-resolve for us.
+        * 
+        * We also need to allow ENOENT lock errors since unlink
+        * system call can call VNOP_FSYNC during vclean.
+        */
+       error = hfs_lock(VTOC(vp), HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
        if (error)
-               error = hfs_kqfilt_remove(ap);
+               return (0);
+
+       error = hfs_fsync(vp, ap->a_waitfor, 0, vfs_context_proc(ap->a_context));
+
+       hfs_unlock(VTOC(vp));
        return (error);
 }
 
-#endif /* FIFO */
+int (**hfs_vnodeop_p)(void *);
 
+#define VOPFUNC int (*)(void *)
 
-/*****************************************************************************
-*
-*      VOP Tables
-*
-*****************************************************************************/
-int hfs_cache_lookup();        /* in hfs_lookup.c */
-int hfs_lookup();      /* in hfs_lookup.c */
-int hfs_read();                /* in hfs_readwrite.c */
-int hfs_write();       /* in hfs_readwrite.c */
-int hfs_ioctl();       /* in hfs_readwrite.c */
-int hfs_select();      /* in hfs_readwrite.c */
-int hfs_bmap();                /* in hfs_readwrite.c */
-int hfs_strategy();    /* in hfs_readwrite.c */
-int hfs_truncate();    /* in hfs_readwrite.c */
-int hfs_allocate();    /* in hfs_readwrite.c */
-int hfs_pagein();      /* in hfs_readwrite.c */
-int hfs_pageout();     /* in hfs_readwrite.c */
-int hfs_search();      /* in hfs_search.c */
-int hfs_bwrite();      /* in hfs_readwrite.c */
-int hfs_link();                /* in hfs_link.c */
-int hfs_blktooff();    /* in hfs_readwrite.c */
-int hfs_offtoblk();    /* in hfs_readwrite.c */
-int hfs_cmap();                /* in hfs_readwrite.c */
-int hfs_getattrlist(); /* in hfs_attrlist.c */
-int hfs_setattrlist(); /* in hfs_attrlist.c */
-int hfs_readdirattr(); /* in hfs_attrlist.c */
-int hfs_inactive();    /* in hfs_cnode.c */
-int hfs_reclaim();     /* in hfs_cnode.c */
 
-int (**hfs_vnodeop_p)(void *);
+#if CONFIG_HFS_STD
+int (**hfs_std_vnodeop_p) (void *);
+static int hfs_readonly_op (__unused void* ap) { return (EROFS); }
 
-#define VOPFUNC int (*)(void *)
+/* 
+ * In 10.6 and forward, HFS Standard is read-only and deprecated.  The vnop table below
+ * is for use with HFS standard to block out operations that would modify the file system
+ */
+
+struct vnodeopv_entry_desc hfs_standard_vnodeop_entries[] = {
+    { &vnop_default_desc, (VOPFUNC)vn_default_error },
+    { &vnop_lookup_desc, (VOPFUNC)hfs_vnop_lookup },           /* lookup */
+    { &vnop_create_desc, (VOPFUNC)hfs_readonly_op },           /* create (READONLY) */
+    { &vnop_mknod_desc, (VOPFUNC)hfs_readonly_op },             /* mknod (READONLY) */
+    { &vnop_open_desc, (VOPFUNC)hfs_vnop_open },                       /* open */
+    { &vnop_close_desc, (VOPFUNC)hfs_vnop_close },             /* close */
+    { &vnop_getattr_desc, (VOPFUNC)hfs_vnop_getattr },         /* getattr */
+    { &vnop_setattr_desc, (VOPFUNC)hfs_readonly_op },          /* setattr */
+    { &vnop_read_desc, (VOPFUNC)hfs_vnop_read },                       /* read */
+    { &vnop_write_desc, (VOPFUNC)hfs_readonly_op },            /* write (READONLY) */
+    { &vnop_ioctl_desc, (VOPFUNC)hfs_vnop_ioctl },             /* ioctl */
+    { &vnop_select_desc, (VOPFUNC)hfs_vnop_select },           /* select */
+    { &vnop_revoke_desc, (VOPFUNC)nop_revoke },                        /* revoke */
+    { &vnop_exchange_desc, (VOPFUNC)hfs_readonly_op },         /* exchange  (READONLY)*/
+    { &vnop_mmap_desc, (VOPFUNC)err_mmap },                    /* mmap */
+    { &vnop_fsync_desc, (VOPFUNC)hfs_readonly_op},             /* fsync (READONLY) */
+    { &vnop_remove_desc, (VOPFUNC)hfs_readonly_op },           /* remove (READONLY) */
+    { &vnop_link_desc, (VOPFUNC)hfs_readonly_op },                     /* link ( READONLLY) */
+    { &vnop_rename_desc, (VOPFUNC)hfs_readonly_op },           /* rename (READONLY)*/
+    { &vnop_mkdir_desc, (VOPFUNC)hfs_readonly_op },             /* mkdir (READONLY) */
+    { &vnop_rmdir_desc, (VOPFUNC)hfs_readonly_op },            /* rmdir (READONLY) */
+    { &vnop_symlink_desc, (VOPFUNC)hfs_readonly_op },         /* symlink (READONLY) */
+    { &vnop_readdir_desc, (VOPFUNC)hfs_vnop_readdir },         /* readdir */
+    { &vnop_readdirattr_desc, (VOPFUNC)hfs_vnop_readdirattr }, /* readdirattr */
+    { &vnop_readlink_desc, (VOPFUNC)hfs_vnop_readlink },               /* readlink */
+    { &vnop_inactive_desc, (VOPFUNC)hfs_vnop_inactive },               /* inactive */
+    { &vnop_reclaim_desc, (VOPFUNC)hfs_vnop_reclaim },         /* reclaim */
+    { &vnop_strategy_desc, (VOPFUNC)hfs_vnop_strategy },               /* strategy */
+    { &vnop_pathconf_desc, (VOPFUNC)hfs_vnop_pathconf },               /* pathconf */
+    { &vnop_advlock_desc, (VOPFUNC)err_advlock },              /* advlock */
+    { &vnop_allocate_desc, (VOPFUNC)hfs_readonly_op },         /* allocate (READONLY) */
+#if CONFIG_SEARCHFS
+    { &vnop_searchfs_desc, (VOPFUNC)hfs_vnop_search },         /* search fs */
+#else
+    { &vnop_searchfs_desc, (VOPFUNC)err_searchfs },            /* search fs */
+#endif
+    { &vnop_bwrite_desc, (VOPFUNC)hfs_readonly_op },           /* bwrite (READONLY) */
+    { &vnop_pagein_desc, (VOPFUNC)hfs_vnop_pagein },           /* pagein */
+    { &vnop_pageout_desc,(VOPFUNC) hfs_readonly_op },          /* pageout (READONLY)  */
+    { &vnop_copyfile_desc, (VOPFUNC)hfs_readonly_op },         /* copyfile (READONLY)*/
+    { &vnop_blktooff_desc, (VOPFUNC)hfs_vnop_blktooff },               /* blktooff */
+    { &vnop_offtoblk_desc, (VOPFUNC)hfs_vnop_offtoblk },               /* offtoblk */
+    { &vnop_blockmap_desc, (VOPFUNC)hfs_vnop_blockmap },                       /* blockmap */
+    { &vnop_getxattr_desc, (VOPFUNC)hfs_vnop_getxattr},
+    { &vnop_setxattr_desc, (VOPFUNC)hfs_readonly_op},         /* set xattr (READONLY) */
+    { &vnop_removexattr_desc, (VOPFUNC)hfs_readonly_op},      /* remove xattr (READONLY) */
+    { &vnop_listxattr_desc, (VOPFUNC)hfs_vnop_listxattr},
+#if NAMEDSTREAMS
+    { &vnop_getnamedstream_desc, (VOPFUNC)hfs_vnop_getnamedstream },
+    { &vnop_makenamedstream_desc, (VOPFUNC)hfs_readonly_op }, 
+    { &vnop_removenamedstream_desc, (VOPFUNC)hfs_readonly_op },
+#endif
+    { &vnop_getattrlistbulk_desc, (VOPFUNC)hfs_vnop_getattrlistbulk }, /* getattrlistbulk */
+    { NULL, (VOPFUNC)NULL }
+};
+
+struct vnodeopv_desc hfs_std_vnodeop_opv_desc =
+{ &hfs_std_vnodeop_p, hfs_standard_vnodeop_entries };
+#endif
 
+/* VNOP table for HFS+ */
 struct vnodeopv_entry_desc hfs_vnodeop_entries[] = {
-    { &vop_default_desc, (VOPFUNC)vn_default_error },
-    { &vop_lookup_desc, (VOPFUNC)hfs_cache_lookup },           /* lookup */
-    { &vop_create_desc, (VOPFUNC)hfs_create },                 /* create */
-    { &vop_mknod_desc, (VOPFUNC)hfs_mknod },                   /* mknod */
-    { &vop_open_desc, (VOPFUNC)hfs_open },                     /* open */
-    { &vop_close_desc, (VOPFUNC)hfs_close },                   /* close */
-    { &vop_access_desc, (VOPFUNC)hfs_access },                 /* access */
-    { &vop_getattr_desc, (VOPFUNC)hfs_getattr },               /* getattr */
-    { &vop_setattr_desc, (VOPFUNC)hfs_setattr },               /* setattr */
-    { &vop_read_desc, (VOPFUNC)hfs_read },                     /* read */
-    { &vop_write_desc, (VOPFUNC)hfs_write },                   /* write */
-    { &vop_ioctl_desc, (VOPFUNC)hfs_ioctl },                   /* ioctl */
-    { &vop_select_desc, (VOPFUNC)hfs_select },                 /* select */
-    { &vop_revoke_desc, (VOPFUNC)nop_revoke },                 /* revoke */
-    { &vop_exchange_desc, (VOPFUNC)hfs_exchange },             /* exchange */
-    { &vop_mmap_desc, (VOPFUNC)err_mmap },                     /* mmap */
-    { &vop_fsync_desc, (VOPFUNC)hfs_fsync },                   /* fsync */
-    { &vop_seek_desc, (VOPFUNC)nop_seek },                     /* seek */
-    { &vop_remove_desc, (VOPFUNC)hfs_remove },                 /* remove */
-    { &vop_link_desc, (VOPFUNC)hfs_link },                     /* link */
-    { &vop_rename_desc, (VOPFUNC)hfs_rename },                 /* rename */
-    { &vop_mkdir_desc, (VOPFUNC)hfs_mkdir },                   /* mkdir */
-    { &vop_rmdir_desc, (VOPFUNC)hfs_rmdir },                   /* rmdir */
-    { &vop_mkcomplex_desc, (VOPFUNC)err_mkcomplex },           /* mkcomplex */
-    { &vop_getattrlist_desc, (VOPFUNC)hfs_getattrlist },  /* getattrlist */
-    { &vop_setattrlist_desc, (VOPFUNC)hfs_setattrlist },  /* setattrlist */
-    { &vop_symlink_desc, (VOPFUNC)hfs_symlink },               /* symlink */
-    { &vop_readdir_desc, (VOPFUNC)hfs_readdir },               /* readdir */
-    { &vop_readdirattr_desc, (VOPFUNC)hfs_readdirattr },  /* readdirattr */
-    { &vop_readlink_desc, (VOPFUNC)hfs_readlink },             /* readlink */
-    { &vop_abortop_desc, (VOPFUNC)nop_abortop },               /* abortop */
-    { &vop_inactive_desc, (VOPFUNC)hfs_inactive },             /* inactive */
-    { &vop_reclaim_desc, (VOPFUNC)hfs_reclaim },               /* reclaim */
-    { &vop_lock_desc, (VOPFUNC)hfs_lock },                     /* lock */
-    { &vop_unlock_desc, (VOPFUNC)hfs_unlock },                 /* unlock */
-    { &vop_bmap_desc, (VOPFUNC)hfs_bmap },                     /* bmap */
-    { &vop_strategy_desc, (VOPFUNC)hfs_strategy },             /* strategy */
-    { &vop_print_desc, (VOPFUNC)hfs_print },                   /* print */
-    { &vop_islocked_desc, (VOPFUNC)hfs_islocked },             /* islocked */
-    { &vop_pathconf_desc, (VOPFUNC)hfs_pathconf },             /* pathconf */
-    { &vop_advlock_desc, (VOPFUNC)hfs_advlock },               /* advlock */
-    { &vop_reallocblks_desc, (VOPFUNC)err_reallocblks },  /* reallocblks */
-    { &vop_truncate_desc, (VOPFUNC)hfs_truncate },             /* truncate */
-    { &vop_allocate_desc, (VOPFUNC)hfs_allocate },             /* allocate */
-    { &vop_update_desc, (VOPFUNC)hfs_update },                 /* update */
-    { &vop_searchfs_desc, (VOPFUNC)hfs_search },               /* search fs */
-    { &vop_bwrite_desc, (VOPFUNC)hfs_bwrite },                 /* bwrite */
-    { &vop_pagein_desc, (VOPFUNC)hfs_pagein },                 /* pagein */
-    { &vop_pageout_desc,(VOPFUNC) hfs_pageout },               /* pageout */
-    { &vop_copyfile_desc, (VOPFUNC)err_copyfile },             /* copyfile */
-    { &vop_blktooff_desc, (VOPFUNC)hfs_blktooff },             /* blktooff */
-    { &vop_offtoblk_desc, (VOPFUNC)hfs_offtoblk },             /* offtoblk */
-    { &vop_cmap_desc, (VOPFUNC)hfs_cmap },                     /* cmap */
-       { &vop_kqfilt_add_desc, (VOPFUNC)hfs_kqfilt_add },              /* kqfilt_add */
-       { &vop_kqfilt_remove_desc, (VOPFUNC)hfs_kqfilt_remove },                /* kqfilt_remove */
+    { &vnop_default_desc, (VOPFUNC)vn_default_error },
+    { &vnop_lookup_desc, (VOPFUNC)hfs_vnop_lookup },           /* lookup */
+    { &vnop_create_desc, (VOPFUNC)hfs_vnop_create },           /* create */
+    { &vnop_mknod_desc, (VOPFUNC)hfs_vnop_mknod },             /* mknod */
+    { &vnop_open_desc, (VOPFUNC)hfs_vnop_open },                       /* open */
+    { &vnop_close_desc, (VOPFUNC)hfs_vnop_close },             /* close */
+    { &vnop_getattr_desc, (VOPFUNC)hfs_vnop_getattr },         /* getattr */
+    { &vnop_setattr_desc, (VOPFUNC)hfs_vnop_setattr },         /* setattr */
+    { &vnop_read_desc, (VOPFUNC)hfs_vnop_read },                       /* read */
+    { &vnop_write_desc, (VOPFUNC)hfs_vnop_write },             /* write */
+    { &vnop_ioctl_desc, (VOPFUNC)hfs_vnop_ioctl },             /* ioctl */
+    { &vnop_select_desc, (VOPFUNC)hfs_vnop_select },           /* select */
+    { &vnop_revoke_desc, (VOPFUNC)nop_revoke },                        /* revoke */
+    { &vnop_exchange_desc, (VOPFUNC)hfs_vnop_exchange },               /* exchange */
+    { &vnop_mmap_desc, (VOPFUNC)hfs_vnop_mmap },                       /* mmap */
+    { &vnop_fsync_desc, (VOPFUNC)hfs_vnop_fsync },             /* fsync */
+    { &vnop_remove_desc, (VOPFUNC)hfs_vnop_remove },           /* remove */
+    { &vnop_link_desc, (VOPFUNC)hfs_vnop_link },                       /* link */
+    { &vnop_rename_desc, (VOPFUNC)hfs_vnop_rename },           /* rename */
+    { &vnop_mkdir_desc, (VOPFUNC)hfs_vnop_mkdir },             /* mkdir */
+    { &vnop_rmdir_desc, (VOPFUNC)hfs_vnop_rmdir },             /* rmdir */
+    { &vnop_symlink_desc, (VOPFUNC)hfs_vnop_symlink },         /* symlink */
+    { &vnop_readdir_desc, (VOPFUNC)hfs_vnop_readdir },         /* readdir */
+    { &vnop_readdirattr_desc, (VOPFUNC)hfs_vnop_readdirattr }, /* readdirattr */
+    { &vnop_readlink_desc, (VOPFUNC)hfs_vnop_readlink },               /* readlink */
+    { &vnop_inactive_desc, (VOPFUNC)hfs_vnop_inactive },               /* inactive */
+    { &vnop_reclaim_desc, (VOPFUNC)hfs_vnop_reclaim },         /* reclaim */
+    { &vnop_strategy_desc, (VOPFUNC)hfs_vnop_strategy },               /* strategy */
+    { &vnop_pathconf_desc, (VOPFUNC)hfs_vnop_pathconf },               /* pathconf */
+    { &vnop_advlock_desc, (VOPFUNC)err_advlock },              /* advlock */
+    { &vnop_allocate_desc, (VOPFUNC)hfs_vnop_allocate },               /* allocate */
+#if CONFIG_SEARCHFS
+    { &vnop_searchfs_desc, (VOPFUNC)hfs_vnop_search },         /* search fs */
+#else
+    { &vnop_searchfs_desc, (VOPFUNC)err_searchfs },            /* search fs */
+#endif
+    { &vnop_bwrite_desc, (VOPFUNC)hfs_vnop_bwrite },           /* bwrite */
+    { &vnop_pagein_desc, (VOPFUNC)hfs_vnop_pagein },           /* pagein */
+    { &vnop_pageout_desc,(VOPFUNC) hfs_vnop_pageout },         /* pageout */
+    { &vnop_copyfile_desc, (VOPFUNC)err_copyfile },            /* copyfile */
+    { &vnop_blktooff_desc, (VOPFUNC)hfs_vnop_blktooff },               /* blktooff */
+    { &vnop_offtoblk_desc, (VOPFUNC)hfs_vnop_offtoblk },               /* offtoblk */
+    { &vnop_blockmap_desc, (VOPFUNC)hfs_vnop_blockmap },                       /* blockmap */
+    { &vnop_getxattr_desc, (VOPFUNC)hfs_vnop_getxattr},
+    { &vnop_setxattr_desc, (VOPFUNC)hfs_vnop_setxattr},
+    { &vnop_removexattr_desc, (VOPFUNC)hfs_vnop_removexattr},
+    { &vnop_listxattr_desc, (VOPFUNC)hfs_vnop_listxattr},
+#if NAMEDSTREAMS
+    { &vnop_getnamedstream_desc, (VOPFUNC)hfs_vnop_getnamedstream },
+    { &vnop_makenamedstream_desc, (VOPFUNC)hfs_vnop_makenamedstream },
+    { &vnop_removenamedstream_desc, (VOPFUNC)hfs_vnop_removenamedstream },
+#endif
+    { &vnop_getattrlistbulk_desc, (VOPFUNC)hfs_vnop_getattrlistbulk }, /* getattrlistbulk */
+       { &vnop_mnomap_desc, (VOPFUNC)hfs_vnop_mnomap },
     { NULL, (VOPFUNC)NULL }
 };
 
 struct vnodeopv_desc hfs_vnodeop_opv_desc =
 { &hfs_vnodeop_p, hfs_vnodeop_entries };
 
+
+/* Spec Op vnop table for HFS+ */
 int (**hfs_specop_p)(void *);
 struct vnodeopv_entry_desc hfs_specop_entries[] = {
-       { &vop_default_desc, (VOPFUNC)vn_default_error },
-       { &vop_lookup_desc, (VOPFUNC)spec_lookup },             /* lookup */
-       { &vop_create_desc, (VOPFUNC)spec_create },             /* create */
-       { &vop_mknod_desc, (VOPFUNC)spec_mknod },               /* mknod */
-       { &vop_open_desc, (VOPFUNC)spec_open },                 /* open */
-       { &vop_close_desc, (VOPFUNC)hfsspec_close },            /* close */
-       { &vop_access_desc, (VOPFUNC)hfs_access },              /* access */
-       { &vop_getattr_desc, (VOPFUNC)hfs_getattr },            /* getattr */
-       { &vop_setattr_desc, (VOPFUNC)hfs_setattr },            /* setattr */
-       { &vop_read_desc, (VOPFUNC)hfsspec_read },              /* read */
-       { &vop_write_desc, (VOPFUNC)hfsspec_write },            /* write */
-       { &vop_lease_desc, (VOPFUNC)spec_lease_check },         /* lease */
-       { &vop_ioctl_desc, (VOPFUNC)spec_ioctl },               /* ioctl */
-       { &vop_select_desc, (VOPFUNC)spec_select },             /* select */
-       { &vop_revoke_desc, (VOPFUNC)spec_revoke },             /* revoke */
-       { &vop_mmap_desc, (VOPFUNC)spec_mmap },                 /* mmap */
-       { &vop_fsync_desc, (VOPFUNC)hfs_fsync },                /* fsync */
-       { &vop_seek_desc, (VOPFUNC)spec_seek },                 /* seek */
-       { &vop_remove_desc, (VOPFUNC)spec_remove },             /* remove */
-       { &vop_link_desc, (VOPFUNC)spec_link },                 /* link */
-       { &vop_rename_desc, (VOPFUNC)spec_rename },             /* rename */
-       { &vop_mkdir_desc, (VOPFUNC)spec_mkdir },               /* mkdir */
-       { &vop_rmdir_desc, (VOPFUNC)spec_rmdir },               /* rmdir */
-       { &vop_getattrlist_desc, (VOPFUNC)hfs_getattrlist },
-       { &vop_symlink_desc, (VOPFUNC)spec_symlink },           /* symlink */
-       { &vop_readdir_desc, (VOPFUNC)spec_readdir },           /* readdir */
-       { &vop_readlink_desc, (VOPFUNC)spec_readlink },         /* readlink */
-       { &vop_abortop_desc, (VOPFUNC)spec_abortop },           /* abortop */
-       { &vop_inactive_desc, (VOPFUNC)hfs_inactive },          /* inactive */
-       { &vop_reclaim_desc, (VOPFUNC)hfs_reclaim },            /* reclaim */
-       { &vop_lock_desc, (VOPFUNC)hfs_lock },                  /* lock */
-       { &vop_unlock_desc, (VOPFUNC)hfs_unlock },              /* unlock */
-       { &vop_bmap_desc, (VOPFUNC)spec_bmap },                 /* bmap */
-       { &vop_strategy_desc, (VOPFUNC)spec_strategy },         /* strategy */
-       { &vop_print_desc, (VOPFUNC)hfs_print },                /* print */
-       { &vop_islocked_desc, (VOPFUNC)hfs_islocked },          /* islocked */
-       { &vop_pathconf_desc, (VOPFUNC)spec_pathconf },         /* pathconf */
-       { &vop_advlock_desc, (VOPFUNC)spec_advlock },           /* advlock */
-       { &vop_blkatoff_desc, (VOPFUNC)spec_blkatoff },         /* blkatoff */
-       { &vop_valloc_desc, (VOPFUNC)spec_valloc },             /* valloc */
-       { &vop_reallocblks_desc, (VOPFUNC)spec_reallocblks },   /* reallocblks */
-       { &vop_vfree_desc, (VOPFUNC)err_vfree },                /* vfree */
-       { &vop_truncate_desc, (VOPFUNC)spec_truncate },         /* truncate */
-       { &vop_update_desc, (VOPFUNC)hfs_update },              /* update */
-       { &vop_bwrite_desc, (VOPFUNC)hfs_bwrite },
-       { &vop_devblocksize_desc, (VOPFUNC)spec_devblocksize }, /* devblocksize */
-       { &vop_pagein_desc, (VOPFUNC)hfs_pagein },              /* Pagein */
-       { &vop_pageout_desc, (VOPFUNC)hfs_pageout },            /* Pageout */
-        { &vop_copyfile_desc, (VOPFUNC)err_copyfile },         /* copyfile */
-       { &vop_blktooff_desc, (VOPFUNC)hfs_blktooff },          /* blktooff */
-       { &vop_offtoblk_desc, (VOPFUNC)hfs_offtoblk },          /* offtoblk */
+       { &vnop_default_desc, (VOPFUNC)vn_default_error },
+       { &vnop_lookup_desc, (VOPFUNC)spec_lookup },            /* lookup */
+       { &vnop_create_desc, (VOPFUNC)spec_create },            /* create */
+       { &vnop_mknod_desc, (VOPFUNC)spec_mknod },              /* mknod */
+       { &vnop_open_desc, (VOPFUNC)spec_open },                        /* open */
+       { &vnop_close_desc, (VOPFUNC)hfsspec_close },           /* close */
+       { &vnop_getattr_desc, (VOPFUNC)hfs_vnop_getattr },      /* getattr */
+       { &vnop_setattr_desc, (VOPFUNC)hfs_vnop_setattr },      /* setattr */
+       { &vnop_read_desc, (VOPFUNC)hfsspec_read },             /* read */
+       { &vnop_write_desc, (VOPFUNC)hfsspec_write },           /* write */
+       { &vnop_ioctl_desc, (VOPFUNC)spec_ioctl },              /* ioctl */
+       { &vnop_select_desc, (VOPFUNC)spec_select },            /* select */
+       { &vnop_revoke_desc, (VOPFUNC)spec_revoke },            /* revoke */
+       { &vnop_mmap_desc, (VOPFUNC)spec_mmap },                        /* mmap */
+       { &vnop_fsync_desc, (VOPFUNC)hfs_vnop_fsync },          /* fsync */
+       { &vnop_remove_desc, (VOPFUNC)spec_remove },            /* remove */
+       { &vnop_link_desc, (VOPFUNC)spec_link },                        /* link */
+       { &vnop_rename_desc, (VOPFUNC)spec_rename },            /* rename */
+       { &vnop_mkdir_desc, (VOPFUNC)spec_mkdir },              /* mkdir */
+       { &vnop_rmdir_desc, (VOPFUNC)spec_rmdir },              /* rmdir */
+       { &vnop_symlink_desc, (VOPFUNC)spec_symlink },          /* symlink */
+       { &vnop_readdir_desc, (VOPFUNC)spec_readdir },          /* readdir */
+       { &vnop_readlink_desc, (VOPFUNC)spec_readlink },                /* readlink */
+       { &vnop_inactive_desc, (VOPFUNC)hfs_vnop_inactive },    /* inactive */
+       { &vnop_reclaim_desc, (VOPFUNC)hfs_vnop_reclaim },      /* reclaim */
+       { &vnop_strategy_desc, (VOPFUNC)spec_strategy },                /* strategy */
+       { &vnop_pathconf_desc, (VOPFUNC)spec_pathconf },                /* pathconf */
+       { &vnop_advlock_desc, (VOPFUNC)err_advlock },           /* advlock */
+       { &vnop_bwrite_desc, (VOPFUNC)hfs_vnop_bwrite },
+       { &vnop_pagein_desc, (VOPFUNC)hfs_vnop_pagein },                /* Pagein */
+       { &vnop_pageout_desc, (VOPFUNC)hfs_vnop_pageout },      /* Pageout */
+    { &vnop_copyfile_desc, (VOPFUNC)err_copyfile },            /* copyfile */
+       { &vnop_blktooff_desc, (VOPFUNC)hfs_vnop_blktooff },    /* blktooff */
+       { &vnop_offtoblk_desc, (VOPFUNC)hfs_vnop_offtoblk },    /* offtoblk */
+       { &vnop_getxattr_desc, (VOPFUNC)hfs_vnop_getxattr},
+       { &vnop_setxattr_desc, (VOPFUNC)hfs_vnop_setxattr},
+       { &vnop_removexattr_desc, (VOPFUNC)hfs_vnop_removexattr},
+       { &vnop_listxattr_desc, (VOPFUNC)hfs_vnop_listxattr},
        { (struct vnodeop_desc*)NULL, (VOPFUNC)NULL }
 };
 struct vnodeopv_desc hfs_specop_opv_desc =
        { &hfs_specop_p, hfs_specop_entries };
 
 #if FIFO
+/* HFS+ FIFO VNOP table  */
 int (**hfs_fifoop_p)(void *);
 struct vnodeopv_entry_desc hfs_fifoop_entries[] = {
-       { &vop_default_desc, (VOPFUNC)vn_default_error },
-       { &vop_lookup_desc, (VOPFUNC)fifo_lookup },             /* lookup */
-       { &vop_create_desc, (VOPFUNC)fifo_create },             /* create */
-       { &vop_mknod_desc, (VOPFUNC)fifo_mknod },               /* mknod */
-       { &vop_open_desc, (VOPFUNC)fifo_open },                 /* open */
-       { &vop_close_desc, (VOPFUNC)hfsfifo_close },            /* close */
-       { &vop_access_desc, (VOPFUNC)hfs_access },              /* access */
-       { &vop_getattr_desc, (VOPFUNC)hfs_getattr },            /* getattr */
-       { &vop_setattr_desc, (VOPFUNC)hfs_setattr },            /* setattr */
-       { &vop_read_desc, (VOPFUNC)hfsfifo_read },              /* read */
-       { &vop_write_desc, (VOPFUNC)hfsfifo_write },            /* write */
-       { &vop_lease_desc, (VOPFUNC)fifo_lease_check },         /* lease */
-       { &vop_ioctl_desc, (VOPFUNC)fifo_ioctl },               /* ioctl */
-       { &vop_select_desc, (VOPFUNC)fifo_select },             /* select */
-       { &vop_revoke_desc, (VOPFUNC)fifo_revoke },             /* revoke */
-       { &vop_mmap_desc, (VOPFUNC)fifo_mmap },                 /* mmap */
-       { &vop_fsync_desc, (VOPFUNC)hfs_fsync },                /* fsync */
-       { &vop_seek_desc, (VOPFUNC)fifo_seek },                 /* seek */
-       { &vop_remove_desc, (VOPFUNC)fifo_remove },             /* remove */
-       { &vop_link_desc, (VOPFUNC)fifo_link },                 /* link */
-       { &vop_rename_desc, (VOPFUNC)fifo_rename },             /* rename */
-       { &vop_mkdir_desc, (VOPFUNC)fifo_mkdir },               /* mkdir */
-       { &vop_rmdir_desc, (VOPFUNC)fifo_rmdir },               /* rmdir */
-       { &vop_getattrlist_desc, (VOPFUNC)hfs_getattrlist },
-       { &vop_symlink_desc, (VOPFUNC)fifo_symlink },           /* symlink */
-       { &vop_readdir_desc, (VOPFUNC)fifo_readdir },           /* readdir */
-       { &vop_readlink_desc, (VOPFUNC)fifo_readlink },         /* readlink */
-       { &vop_abortop_desc, (VOPFUNC)fifo_abortop },           /* abortop */
-       { &vop_inactive_desc, (VOPFUNC)hfs_inactive },          /* inactive */
-       { &vop_reclaim_desc, (VOPFUNC)hfs_reclaim },            /* reclaim */
-       { &vop_lock_desc, (VOPFUNC)hfs_lock },                  /* lock */
-       { &vop_unlock_desc, (VOPFUNC)hfs_unlock },              /* unlock */
-       { &vop_bmap_desc, (VOPFUNC)fifo_bmap },                 /* bmap */
-       { &vop_strategy_desc, (VOPFUNC)fifo_strategy },         /* strategy */
-       { &vop_print_desc, (VOPFUNC)hfs_print },                /* print */
-       { &vop_islocked_desc, (VOPFUNC)hfs_islocked },          /* islocked */
-       { &vop_pathconf_desc, (VOPFUNC)fifo_pathconf },         /* pathconf */
-       { &vop_advlock_desc, (VOPFUNC)fifo_advlock },           /* advlock */
-       { &vop_blkatoff_desc, (VOPFUNC)fifo_blkatoff },         /* blkatoff */
-       { &vop_valloc_desc, (VOPFUNC)fifo_valloc },             /* valloc */
-       { &vop_reallocblks_desc, (VOPFUNC)fifo_reallocblks },   /* reallocblks */
-       { &vop_vfree_desc, (VOPFUNC)err_vfree },                /* vfree */
-       { &vop_truncate_desc, (VOPFUNC)fifo_truncate },         /* truncate */
-       { &vop_update_desc, (VOPFUNC)hfs_update },              /* update */
-       { &vop_bwrite_desc, (VOPFUNC)hfs_bwrite },
-       { &vop_pagein_desc, (VOPFUNC)hfs_pagein },              /* Pagein */
-       { &vop_pageout_desc, (VOPFUNC)hfs_pageout },            /* Pageout */
-    { &vop_copyfile_desc, (VOPFUNC)err_copyfile },             /* copyfile */
-       { &vop_blktooff_desc, (VOPFUNC)hfs_blktooff },          /* blktooff */
-       { &vop_offtoblk_desc, (VOPFUNC)hfs_offtoblk },          /* offtoblk */
-       { &vop_cmap_desc, (VOPFUNC)hfs_cmap },                  /* cmap */
-       { &vop_kqfilt_add_desc, (VOPFUNC)hfsfifo_kqfilt_add },  /* kqfilt_add */
-       { &vop_kqfilt_remove_desc, (VOPFUNC)hfsfifo_kqfilt_remove },  /* kqfilt_remove */
+       { &vnop_default_desc, (VOPFUNC)vn_default_error },
+       { &vnop_lookup_desc, (VOPFUNC)fifo_lookup },            /* lookup */
+       { &vnop_create_desc, (VOPFUNC)fifo_create },            /* create */
+       { &vnop_mknod_desc, (VOPFUNC)fifo_mknod },              /* mknod */
+       { &vnop_open_desc, (VOPFUNC)fifo_open },                        /* open */
+       { &vnop_close_desc, (VOPFUNC)hfsfifo_close },           /* close */
+       { &vnop_getattr_desc, (VOPFUNC)hfs_vnop_getattr },      /* getattr */
+       { &vnop_setattr_desc, (VOPFUNC)hfs_vnop_setattr },      /* setattr */
+       { &vnop_read_desc, (VOPFUNC)hfsfifo_read },             /* read */
+       { &vnop_write_desc, (VOPFUNC)hfsfifo_write },           /* write */
+       { &vnop_ioctl_desc, (VOPFUNC)fifo_ioctl },              /* ioctl */
+       { &vnop_select_desc, (VOPFUNC)fifo_select },            /* select */
+       { &vnop_revoke_desc, (VOPFUNC)fifo_revoke },            /* revoke */
+       { &vnop_mmap_desc, (VOPFUNC)fifo_mmap },                        /* mmap */
+       { &vnop_fsync_desc, (VOPFUNC)hfs_vnop_fsync },          /* fsync */
+       { &vnop_remove_desc, (VOPFUNC)fifo_remove },            /* remove */
+       { &vnop_link_desc, (VOPFUNC)fifo_link },                        /* link */
+       { &vnop_rename_desc, (VOPFUNC)fifo_rename },            /* rename */
+       { &vnop_mkdir_desc, (VOPFUNC)fifo_mkdir },              /* mkdir */
+       { &vnop_rmdir_desc, (VOPFUNC)fifo_rmdir },              /* rmdir */
+       { &vnop_symlink_desc, (VOPFUNC)fifo_symlink },          /* symlink */
+       { &vnop_readdir_desc, (VOPFUNC)fifo_readdir },          /* readdir */
+       { &vnop_readlink_desc, (VOPFUNC)fifo_readlink },                /* readlink */
+       { &vnop_inactive_desc, (VOPFUNC)hfs_vnop_inactive },    /* inactive */
+       { &vnop_reclaim_desc, (VOPFUNC)hfs_vnop_reclaim },      /* reclaim */
+       { &vnop_strategy_desc, (VOPFUNC)fifo_strategy },                /* strategy */
+       { &vnop_pathconf_desc, (VOPFUNC)fifo_pathconf },                /* pathconf */
+       { &vnop_advlock_desc, (VOPFUNC)err_advlock },           /* advlock */
+       { &vnop_bwrite_desc, (VOPFUNC)hfs_vnop_bwrite },
+       { &vnop_pagein_desc, (VOPFUNC)hfs_vnop_pagein },                /* Pagein */
+       { &vnop_pageout_desc, (VOPFUNC)hfs_vnop_pageout },      /* Pageout */
+       { &vnop_copyfile_desc, (VOPFUNC)err_copyfile },                 /* copyfile */
+       { &vnop_blktooff_desc, (VOPFUNC)hfs_vnop_blktooff },    /* blktooff */
+       { &vnop_offtoblk_desc, (VOPFUNC)hfs_vnop_offtoblk },    /* offtoblk */
+       { &vnop_blockmap_desc, (VOPFUNC)hfs_vnop_blockmap },            /* blockmap */
+       { &vnop_getxattr_desc, (VOPFUNC)hfs_vnop_getxattr},
+       { &vnop_setxattr_desc, (VOPFUNC)hfs_vnop_setxattr},
+       { &vnop_removexattr_desc, (VOPFUNC)hfs_vnop_removexattr},
+       { &vnop_listxattr_desc, (VOPFUNC)hfs_vnop_listxattr},
        { (struct vnodeop_desc*)NULL, (VOPFUNC)NULL }
 };
 struct vnodeopv_desc hfs_fifoop_opv_desc =