]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/kern/ubc_subr.c
xnu-1228.15.4.tar.gz
[apple/xnu.git] / bsd / kern / ubc_subr.c
index 5fb09e492a7ae470452963c338b056ed54a1004f..df8047ea8201c215bd9114c6248224b9ca2d04a0 100644 (file)
@@ -1,16 +1,19 @@
 /*
 /*
- * Copyright (c) 1999-2002 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 1999-2008 Apple Inc. All rights reserved.
  *
  *
- * @APPLE_LICENSE_HEADER_START@
- * 
- * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
  * 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
  * 
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
- * compliance with the License. Please obtain a copy of the License at
- * http://www.opensource.apple.com/apsl/ and read it before using this
- * file.
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ * 
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
  * 
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * 
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
@@ -20,7 +23,7 @@
  * Please see the License for the specific language governing rights and
  * limitations under the License.
  * 
  * Please see the License for the specific language governing rights and
  * limitations under the License.
  * 
- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /* 
  *     File:   ubc_subr.c
  */
 /* 
  *     File:   ubc_subr.c
  *
  */ 
 
  *
  */ 
 
-#undef DIAGNOSTIC
-#define DIAGNOSTIC 1
-
 #include <sys/types.h>
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/lock.h>
 #include <sys/types.h>
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/lock.h>
-#include <sys/ubc.h>
-#include <sys/mount.h>
-#include <sys/vnode.h>
-#include <sys/ubc.h>
+#include <sys/mman.h>
+#include <sys/mount_internal.h>
+#include <sys/vnode_internal.h>
+#include <sys/ubc_internal.h>
 #include <sys/ucred.h>
 #include <sys/ucred.h>
-#include <sys/proc.h>
+#include <sys/proc_internal.h>
+#include <sys/kauth.h>
 #include <sys/buf.h>
 #include <sys/buf.h>
+#include <sys/user.h>
+#include <sys/codesign.h>
 
 #include <mach/mach_types.h>
 #include <mach/memory_object_types.h>
 
 #include <mach/mach_types.h>
 #include <mach/memory_object_types.h>
+#include <mach/memory_object_control.h>
+#include <mach/vm_map.h>
+#include <mach/upl.h>
 
 
+#include <kern/kern_types.h>
+#include <kern/kalloc.h>
 #include <kern/zalloc.h>
 #include <kern/zalloc.h>
+#include <kern/thread.h>
+#include <vm/vm_kern.h>
+#include <vm/vm_protos.h> /* last */
+
+#include <libkern/crypto/sha1.h>
+
+#include <security/mac_framework.h>
+
+/* XXX These should be in a BSD accessible Mach header, but aren't. */
+extern kern_return_t memory_object_pages_resident(memory_object_control_t,
+                                                       boolean_t *);
+extern kern_return_t   memory_object_signed(memory_object_control_t control,
+                                            boolean_t is_signed);
+extern void Debugger(const char *message);
+
+
+/* XXX no one uses this interface! */
+kern_return_t ubc_page_op_with_control(
+       memory_object_control_t  control,
+       off_t                    f_offset,
+       int                      ops,
+       ppnum_t                  *phys_entryp,
+       int                      *flagsp);
+
 
 #if DIAGNOSTIC
 #if defined(assert)
 #undef assert()
 #endif
 #define assert(cond)    \
 
 #if DIAGNOSTIC
 #if defined(assert)
 #undef assert()
 #endif
 #define assert(cond)    \
-    ((void) ((cond) ? 0 : panic("%s:%d (%s)", __FILE__, __LINE__, # cond)))
+    ((void) ((cond) ? 0 : panic("Assert failed: %s", # cond)))
 #else
 #include <kern/assert.h>
 #endif /* DIAGNOSTIC */
 
 #else
 #include <kern/assert.h>
 #endif /* DIAGNOSTIC */
 
+static int ubc_info_init_internal(struct vnode *vp, int withfsize, off_t filesize);
+static int ubc_umcallback(vnode_t, void *);
+static int ubc_msync_internal(vnode_t, off_t, off_t, off_t *, int, int *);
+static void ubc_cs_free(struct ubc_info *uip);
+
 struct zone    *ubc_info_zone;
 
 struct zone    *ubc_info_zone;
 
-/* lock for changes to struct UBC */
-static __inline__ void
-ubc_lock(struct vnode *vp)
-{
-       /* For now, just use the v_interlock */
-       simple_lock(&vp->v_interlock);
-}
 
 
-/* unlock */
-static __inline__ void
-ubc_unlock(struct vnode *vp)
+/*
+ * CODESIGNING
+ * Routines to navigate code signing data structures in the kernel...
+ */
+static boolean_t
+cs_valid_range(
+       const void *start,
+       const void *end,
+       const void *lower_bound,
+       const void *upper_bound)
 {
 {
-       /* For now, just use the v_interlock */
-       simple_unlock(&vp->v_interlock);
+       if (upper_bound < lower_bound ||
+           end < start) {
+               return FALSE;
+       }
+
+       if (start < lower_bound ||
+           end > upper_bound) {
+               return FALSE;
+       }
+
+       return TRUE;
 }
 
 /*
 }
 
 /*
- * Serialize the requests to the VM
- * Returns:
- *             0       -       Failure
- *             1       -       Sucessful in acquiring the lock
- *             2       -       Sucessful in acquiring the lock recursively
- *                             do not call ubc_unbusy()
- *                             [This is strange, but saves 4 bytes in struct ubc_info]
+ * Magic numbers used by Code Signing
  */
  */
-static int
-ubc_busy(struct vnode *vp)
-{
-       register struct ubc_info        *uip;
+enum {
+       CSMAGIC_REQUIREMENT = 0xfade0c00,               /* single Requirement blob */
+       CSMAGIC_REQUIREMENTS = 0xfade0c01,              /* Requirements vector (internal requirements) */
+       CSMAGIC_CODEDIRECTORY = 0xfade0c02,             /* CodeDirectory blob */
+       CSMAGIC_EMBEDDED_SIGNATURE = 0xfade0cc0, /* embedded form of signature data */
+       CSMAGIC_EMBEDDED_SIGNATURE_OLD = 0xfade0b02,    /* XXX */
+       CSMAGIC_DETACHED_SIGNATURE = 0xfade0cc1, /* multi-arch collection of embedded signatures */
+       
+       CSSLOT_CODEDIRECTORY = 0,                               /* slot index for CodeDirectory */
+};
 
 
-       if (!UBCINFOEXISTS(vp))
-               return (0);
 
 
-       uip = vp->v_ubcinfo;
+/*
+ * Structure of an embedded-signature SuperBlob
+ */
+typedef struct __BlobIndex {
+       uint32_t type;                                  /* type of entry */
+       uint32_t offset;                                /* offset of entry */
+} CS_BlobIndex;
+
+typedef struct __SuperBlob {
+       uint32_t magic;                                 /* magic number */
+       uint32_t length;                                /* total length of SuperBlob */
+       uint32_t count;                                 /* number of index entries following */
+       CS_BlobIndex index[];                   /* (count) entries */
+       /* followed by Blobs in no particular order as indicated by offsets in index */
+} CS_SuperBlob;
 
 
-       while (ISSET(uip->ui_flags, UI_BUSY)) {
 
 
-               if (uip->ui_owner == (void *)current_act())
-                       return (2);
+/*
+ * C form of a CodeDirectory.
+ */
+typedef struct __CodeDirectory {
+       uint32_t magic;                                 /* magic number (CSMAGIC_CODEDIRECTORY) */
+       uint32_t length;                                /* total length of CodeDirectory blob */
+       uint32_t version;                               /* compatibility version */
+       uint32_t flags;                                 /* setup and mode flags */
+       uint32_t hashOffset;                    /* offset of hash slot element at index zero */
+       uint32_t identOffset;                   /* offset of identifier string */
+       uint32_t nSpecialSlots;                 /* number of special hash slots */
+       uint32_t nCodeSlots;                    /* number of ordinary (code) hash slots */
+       uint32_t codeLimit;                             /* limit to main image signature range */
+       uint8_t hashSize;                               /* size of each hash in bytes */
+       uint8_t hashType;                               /* type of hash (cdHashType* constants) */
+       uint8_t spare1;                                 /* unused (must be zero) */
+       uint8_t pageSize;                               /* log2(page size in bytes); 0 => infinite */
+       uint32_t spare2;                                /* unused (must be zero) */
+       /* followed by dynamic content as located by offset fields above */
+} CS_CodeDirectory;
 
 
-               SET(uip->ui_flags, UI_WANTED);
-               (void) tsleep((caddr_t)&vp->v_ubcinfo, PINOD, "ubcbusy", 0);
 
 
-               if (!UBCINFOEXISTS(vp))
-                       return (0);
-       }
-       uip->ui_owner = (void *)current_act();
+/*
+ * Locate the CodeDirectory from an embedded signature blob
+ */
+static const 
+CS_CodeDirectory *findCodeDirectory(
+       const CS_SuperBlob *embedded,
+       char *lower_bound,
+       char *upper_bound)
+{
+       const CS_CodeDirectory *cd = NULL;
+
+       if (embedded &&
+           cs_valid_range(embedded, embedded + 1, lower_bound, upper_bound) &&
+           ntohl(embedded->magic) == CSMAGIC_EMBEDDED_SIGNATURE) {
+               const CS_BlobIndex *limit;
+               const CS_BlobIndex *p;
+
+               limit = &embedded->index[ntohl(embedded->count)];
+               if (!cs_valid_range(&embedded->index[0], limit,
+                                   lower_bound, upper_bound)) {
+                       return NULL;
+               }
+               for (p = embedded->index; p < limit; ++p) {
+                       if (ntohl(p->type) == CSSLOT_CODEDIRECTORY) {
+                               const unsigned char *base;
 
 
-       SET(uip->ui_flags, UI_BUSY);
+                               base = (const unsigned char *)embedded;
+                               cd = (const CS_CodeDirectory *)(base + ntohl(p->offset));
+                               break;
+                       }
+               }
+       } else {
+               /*
+                * Detached signatures come as a bare CS_CodeDirectory,
+                * without a blob.
+                */
+               cd = (const CS_CodeDirectory *) embedded;
+       }
+       if (cd &&
+           cs_valid_range(cd, cd + 1, lower_bound, upper_bound) &&
+           cs_valid_range(cd, (const char *) cd + ntohl(cd->length),
+                          lower_bound, upper_bound) &&
+           cs_valid_range(cd, (const char *) cd + ntohl(cd->hashOffset),
+                          lower_bound, upper_bound) &&
+           cs_valid_range(cd, (const char *) cd +
+                          ntohl(cd->hashOffset) +
+                          (ntohl(cd->nCodeSlots) * SHA1_RESULTLEN),
+                          lower_bound, upper_bound) &&
+           
+           ntohl(cd->magic) == CSMAGIC_CODEDIRECTORY) {
+               return cd;
+       }
 
 
-       return (1);
+       // not found or not a valid code directory
+       return NULL;
 }
 
 }
 
-static void
-ubc_unbusy(struct vnode *vp)
-{
-       register struct ubc_info        *uip;
 
 
-       if (!UBCINFOEXISTS(vp)) {
-               wakeup((caddr_t)&vp->v_ubcinfo);
-               return;
+/*
+ * Locating a page hash
+ */
+static const unsigned char *
+hashes(
+       const CS_CodeDirectory *cd,
+       unsigned page,
+       char *lower_bound,
+       char *upper_bound)
+{
+       const unsigned char *base, *top, *hash;
+       uint32_t nCodeSlots;
+
+       assert(cs_valid_range(cd, cd + 1, lower_bound, upper_bound));
+
+       base = (const unsigned char *)cd + ntohl(cd->hashOffset);
+       nCodeSlots = ntohl(cd->nCodeSlots);
+       top = base + nCodeSlots * SHA1_RESULTLEN;
+       if (!cs_valid_range(base, top, 
+                           lower_bound, upper_bound) ||
+           page > nCodeSlots) {
+               return NULL;
        }
        }
-       uip = vp->v_ubcinfo;
-       CLR(uip->ui_flags, UI_BUSY);
-       uip->ui_owner = (void *)NULL;
+       assert(page < nCodeSlots);
 
 
-       if (ISSET(uip->ui_flags, UI_WANTED)) {
-               CLR(uip->ui_flags, UI_WANTED);
-               wakeup((caddr_t)&vp->v_ubcinfo);
+       hash = base + page * SHA1_RESULTLEN;
+       if (!cs_valid_range(hash, hash + SHA1_RESULTLEN,
+                           lower_bound, upper_bound)) {
+               hash = NULL;
        }
        }
+
+       return hash;
 }
 }
+/*
+ * CODESIGNING
+ * End of routines to navigate code signing data structures in the kernel.
+ */
+
 
 /*
 
 /*
- *     Initialization of the zone for Unified Buffer Cache.
+ * ubc_init
+ * 
+ * Initialization of the zone for Unified Buffer Cache.
+ *
+ * Parameters: (void)
+ *
+ * Returns:    (void)
+ *
+ * Implicit returns:
+ *             ubc_info_zone(global)   initialized for subsequent allocations
  */
 __private_extern__ void
  */
 __private_extern__ void
-ubc_init()
+ubc_init(void)
 {
        int     i;
 
        i = (vm_size_t) sizeof (struct ubc_info);
 {
        int     i;
 
        i = (vm_size_t) sizeof (struct ubc_info);
-       /* XXX  the number of elements should be tied in to maxvnodes */
+
        ubc_info_zone = zinit (i, 10000*i, 8192, "ubc_info zone");
        ubc_info_zone = zinit (i, 10000*i, 8192, "ubc_info zone");
-       return;
 }
 
 }
 
+
 /*
 /*
- *     Initialize a ubc_info structure for a vnode.
+ * ubc_info_init
+ *
+ * Allocate and attach an empty ubc_info structure to a vnode
+ *
+ * Parameters: vp                      Pointer to the vnode
+ *
+ * Returns:    0                       Success
+ *     vnode_size:ENOMEM               Not enough space
+ *     vnode_size:???                  Other error from vnode_getattr
+ *
  */
 int
 ubc_info_init(struct vnode *vp)
  */
 int
 ubc_info_init(struct vnode *vp)
+{
+       return(ubc_info_init_internal(vp, 0, 0));
+}
+
+
+/*
+ * ubc_info_init_withsize
+ *
+ * Allocate and attach a sized ubc_info structure to a vnode
+ *
+ * Parameters: vp                      Pointer to the vnode
+ *             filesize                The size of the file
+ *
+ * Returns:    0                       Success
+ *     vnode_size:ENOMEM               Not enough space
+ *     vnode_size:???                  Other error from vnode_getattr
+ */
+int
+ubc_info_init_withsize(struct vnode *vp, off_t filesize)
+{
+       return(ubc_info_init_internal(vp, 1, filesize));
+}
+
+
+/*
+ * ubc_info_init_internal
+ *
+ * Allocate and attach a ubc_info structure to a vnode
+ *
+ * Parameters: vp                      Pointer to the vnode
+ *             withfsize{0,1}          Zero if the size should be obtained
+ *                                     from the vnode; otherwise, use filesize
+ *             filesize                The size of the file, if withfsize == 1
+ *
+ * Returns:    0                       Success
+ *     vnode_size:ENOMEM               Not enough space
+ *     vnode_size:???                  Other error from vnode_getattr
+ *
+ * Notes:      We call a blocking zalloc(), and the zone was created as an
+ *             expandable and collectable zone, so if no memory is available,
+ *             it is possible for zalloc() to block indefinitely.  zalloc()
+ *             may also panic if the zone of zones is exhausted, since it's
+ *             NOT expandable.
+ *
+ *             We unconditionally call vnode_pager_setup(), even if this is
+ *             a reuse of a ubc_info; in that case, we should probably assert
+ *             that it does not already have a pager association, but do not.
+ *
+ *             Since memory_object_create_named() can only fail from receiving
+ *             an invalid pager argument, the explicit check and panic is
+ *             merely precautionary.
+ */
+static int
+ubc_info_init_internal(vnode_t vp, int withfsize, off_t filesize)
 {
        register struct ubc_info        *uip;
        void *  pager;
 {
        register struct ubc_info        *uip;
        void *  pager;
-       struct vattr    vattr;
-       struct proc *p = current_proc();
        int error = 0;
        kern_return_t kret;
        memory_object_control_t control;
 
        int error = 0;
        kern_return_t kret;
        memory_object_control_t control;
 
-       if (!UBCISVALID(vp))
-               return (EINVAL);
+       uip = vp->v_ubcinfo;
 
 
-       ubc_lock(vp);
-       if (ISSET(vp->v_flag,  VUINIT)) {
-               /*
-                * other thread is already doing this
-                * wait till done
-                */
-               while (ISSET(vp->v_flag,  VUINIT)) {
-                       SET(vp->v_flag, VUWANT); /* XXX overloaded! */
-                       ubc_unlock(vp);
-                       (void) tsleep((caddr_t)vp, PINOD, "ubcinfo", 0);
-                       ubc_lock(vp);
-               }
-               ubc_unlock(vp);
-               return (0);
-       } else {
-               SET(vp->v_flag, VUINIT);
-       }
+       /*
+        * If there is not already a ubc_info attached to the vnode, we
+        * attach one; otherwise, we will reuse the one that's there.
+        */
+       if (uip == UBC_INFO_NULL) {
 
 
-       uip = vp->v_ubcinfo;
-       if ((uip == UBC_INFO_NULL) || (uip == UBC_NOINFO)) {
-               ubc_unlock(vp);
                uip = (struct ubc_info *) zalloc(ubc_info_zone);
                uip = (struct ubc_info *) zalloc(ubc_info_zone);
-               uip->ui_pager = MEMORY_OBJECT_NULL;
-               uip->ui_control = MEMORY_OBJECT_CONTROL_NULL;
-               uip->ui_flags = UI_INITED;
+               bzero((char *)uip, sizeof(struct ubc_info));
+
                uip->ui_vnode = vp;
                uip->ui_vnode = vp;
+               uip->ui_flags = UI_INITED;
                uip->ui_ucred = NOCRED;
                uip->ui_ucred = NOCRED;
-               uip->ui_refcount = 1;
-               uip->ui_size = 0;
-               uip->ui_mapped = 0;
-               uip->ui_owner = (void *)NULL;
-               ubc_lock(vp);
        }
        }
-#if DIAGNOSTIC
-       else
-               Debugger("ubc_info_init: already");
-#endif /* DIAGNOSTIC */
-       
        assert(uip->ui_flags != UI_NONE);
        assert(uip->ui_vnode == vp);
 
        assert(uip->ui_flags != UI_NONE);
        assert(uip->ui_vnode == vp);
 
-#if 0
-       if(ISSET(uip->ui_flags, UI_HASPAGER))
-               goto done;
-#endif /* 0 */
-
        /* now set this ubc_info in the vnode */
        vp->v_ubcinfo = uip;
        /* now set this ubc_info in the vnode */
        vp->v_ubcinfo = uip;
-       SET(uip->ui_flags, UI_HASPAGER);
-       ubc_unlock(vp);
+
+       /*
+        * Allocate a pager object for this vnode
+        *
+        * XXX The value of the pager parameter is currently ignored.
+        * XXX Presumably, this API changed to avoid the race between
+        * XXX setting the pager and the UI_HASPAGER flag.
+        */
        pager = (void *)vnode_pager_setup(vp, uip->ui_pager);
        assert(pager);
        pager = (void *)vnode_pager_setup(vp, uip->ui_pager);
        assert(pager);
-       ubc_setpager(vp, pager);
 
        /*
 
        /*
-        * Note: We can not use VOP_GETATTR() to get accurate
-        * value of ui_size. Thanks to NFS.
-        * nfs_getattr() can call vinvalbuf() and in this case
-        * ubc_info is not set up to deal with that.
+        * Explicitly set the pager into the ubc_info, after setting the
+        * UI_HASPAGER flag.
+        */
+       SET(uip->ui_flags, UI_HASPAGER);
+       uip->ui_pager = pager;
+
+       /*
+        * Note: We can not use VNOP_GETATTR() to get accurate
+        * value of ui_size because this may be an NFS vnode, and
+        * nfs_getattr() can call vinvalbuf(); if this happens,
+        * ubc_info is not set up to deal with that event.
         * So use bogus size.
         */
 
         * So use bogus size.
         */
 
@@ -247,70 +439,91 @@ ubc_info_init(struct vnode *vp)
        assert(control);
        uip->ui_control = control;      /* cache the value of the mo control */
        SET(uip->ui_flags, UI_HASOBJREF);       /* with a named reference */
        assert(control);
        uip->ui_control = control;      /* cache the value of the mo control */
        SET(uip->ui_flags, UI_HASOBJREF);       /* with a named reference */
-       /* create a pager reference on the vnode */
-       error = vnode_pager_vget(vp);
-       if (error)
-               panic("ubc_info_init: vnode_pager_vget error = %d", error);
-
-       /* initialize the size */
-       error = VOP_GETATTR(vp, &vattr, p->p_ucred, p);
-
-       ubc_lock(vp);
-       uip->ui_size = (error ? 0: vattr.va_size);
-
-done:
-       CLR(vp->v_flag, VUINIT);
-       if (ISSET(vp->v_flag, VUWANT)) {
-               CLR(vp->v_flag, VUWANT);
-               ubc_unlock(vp);
-               wakeup((caddr_t)vp);
-       } else 
-               ubc_unlock(vp);
+
+       if (withfsize == 0) {
+               /* initialize the size */
+               error = vnode_size(vp, &uip->ui_size, vfs_context_current());
+               if (error)
+                       uip->ui_size = 0;
+       } else {
+               uip->ui_size = filesize;
+       }
+       vp->v_lflag |= VNAMED_UBC;      /* vnode has a named ubc reference */
 
        return (error);
 }
 
 
        return (error);
 }
 
-/* Free the ubc_info */
+
+/*
+ * ubc_info_free
+ *
+ * Free a ubc_info structure
+ *
+ * Parameters: uip                     A pointer to the ubc_info to free
+ *
+ * Returns:    (void)
+ *
+ * Notes:      If there is a credential that has subsequently been associated
+ *             with the ubc_info via a call to ubc_setcred(), the reference
+ *             to the credential is dropped.
+ *
+ *             It's actually impossible for a ubc_info.ui_control to take the
+ *             value MEMORY_OBJECT_CONTROL_NULL.
+ */
 static void
 ubc_info_free(struct ubc_info *uip)
 {
 static void
 ubc_info_free(struct ubc_info *uip)
 {
-       struct ucred *credp;
-       
-       credp = uip->ui_ucred;
-       if (credp != NOCRED) {
-               uip->ui_ucred = NOCRED;
-               crfree(credp);
+       if (IS_VALID_CRED(uip->ui_ucred)) {
+               kauth_cred_unref(&uip->ui_ucred);
        }
 
        if (uip->ui_control != MEMORY_OBJECT_CONTROL_NULL)
                memory_object_control_deallocate(uip->ui_control);
        }
 
        if (uip->ui_control != MEMORY_OBJECT_CONTROL_NULL)
                memory_object_control_deallocate(uip->ui_control);
+       
+       cluster_release(uip);
+       ubc_cs_free(uip);
 
 
-       zfree(ubc_info_zone, (vm_offset_t)uip);
+       zfree(ubc_info_zone, uip);
        return;
 }
 
        return;
 }
 
+
 void
 ubc_info_deallocate(struct ubc_info *uip)
 {
 void
 ubc_info_deallocate(struct ubc_info *uip)
 {
-
-       assert(uip->ui_refcount > 0);
-
-    if (uip->ui_refcount-- == 1) {
-               struct vnode *vp;
-
-               vp = uip->ui_vnode;
-               if (ISSET(uip->ui_flags, UI_WANTED)) {
-                       CLR(uip->ui_flags, UI_WANTED);
-                       wakeup((caddr_t)&vp->v_ubcinfo);
-               }
-
-               ubc_info_free(uip);
-       }
+        ubc_info_free(uip);
 }
 
 }
 
+
 /*
 /*
- * Communicate with VM the size change of the file
- * returns 1 on success, 0 on failure
+ * ubc_setsize
+ *
+ * Tell the  VM that the the size of the file represented by the vnode has
+ * changed
+ *
+ * Parameters: vp                      The vp whose backing file size is
+ *                                     being changed
+ *             nsize                   The new size of the backing file
+ *
+ * Returns:    1                       Success
+ *             0                       Failure
+ *
+ * Notes:      This function will indicate failure if the new size that's
+ *             being attempted to be set is negative.
+ *
+ *             This function will fail if there is no ubc_info currently
+ *             associated with the vnode.
+ *
+ *             This function will indicate success it the new size is the
+ *             same or larger than the old size (in this case, the remainder
+ *             of the file will require modification or use of an existing upl
+ *             to access successfully).
+ *
+ *             This function will fail if the new file size is smaller, and
+ *             the memory region being invalidated was unable to actually be
+ *             invalidated and/or the last page could not be flushed, if the
+ *             new size is not aligned to a page boundary.  This is usually
+ *             indicative of an I/O error.
  */
 int
 ubc_setsize(struct vnode *vp, off_t nsize)
  */
 int
 ubc_setsize(struct vnode *vp, off_t nsize)
@@ -319,20 +532,19 @@ ubc_setsize(struct vnode *vp, off_t nsize)
        off_t lastpg, olastpgend, lastoff;
        struct ubc_info *uip;
        memory_object_control_t control;
        off_t lastpg, olastpgend, lastoff;
        struct ubc_info *uip;
        memory_object_control_t control;
-       kern_return_t kret;
+       kern_return_t kret = KERN_SUCCESS;
 
        if (nsize < (off_t)0)
                return (0);
 
 
        if (nsize < (off_t)0)
                return (0);
 
-       if (UBCINVALID(vp))
-               return (0);
-
        if (!UBCINFOEXISTS(vp))
                return (0);
 
        uip = vp->v_ubcinfo;
        if (!UBCINFOEXISTS(vp))
                return (0);
 
        uip = vp->v_ubcinfo;
-       osize = uip->ui_size;   /* call ubc_getsize() ??? */
-       /* Update the size before flushing the VM */
+       osize = uip->ui_size;
+       /*
+        * Update the size before flushing the VM
+        */
        uip->ui_size = nsize;
 
        if (nsize >= osize)     /* Nothing more to do */
        uip->ui_size = nsize;
 
        if (nsize >= osize)     /* Nothing more to do */
@@ -341,853 +553,1044 @@ ubc_setsize(struct vnode *vp, off_t nsize)
        /*
         * When the file shrinks, invalidate the pages beyond the
         * new size. Also get rid of garbage beyond nsize on the
        /*
         * When the file shrinks, invalidate the pages beyond the
         * new size. Also get rid of garbage beyond nsize on the
-        * last page. The ui_size already has the nsize. This
-        * insures that the pageout would not write beyond the new
-        * end of the file.
+        * last page. The ui_size already has the nsize, so any
+        * subsequent page-in will zero-fill the tail properly
         */
         */
-
        lastpg = trunc_page_64(nsize);
        olastpgend = round_page_64(osize);
        control = uip->ui_control;
        assert(control);
        lastoff = (nsize & PAGE_MASK_64);
 
        lastpg = trunc_page_64(nsize);
        olastpgend = round_page_64(osize);
        control = uip->ui_control;
        assert(control);
        lastoff = (nsize & PAGE_MASK_64);
 
-       /*
-        * If length is multiple of page size, we should not flush
-        * invalidating is sufficient
-        */
-        if (!lastoff) {
-        /* invalidate last page and old contents beyond nsize */
-        kret = memory_object_lock_request(control,
-                    (memory_object_offset_t)lastpg,
-                    (memory_object_size_t)(olastpgend - lastpg),
-                    MEMORY_OBJECT_RETURN_NONE, MEMORY_OBJECT_DATA_FLUSH,
-                    VM_PROT_NO_CHANGE);
-        if (kret != KERN_SUCCESS)
-            printf("ubc_setsize: invalidate failed (error = %d)\n", kret);
-
-               return ((kret == KERN_SUCCESS) ? 1 : 0);
-        }
-
-       /* flush the last page */
-       kret = memory_object_lock_request(control,
-                               (memory_object_offset_t)lastpg,
-                               PAGE_SIZE_64,
-                               MEMORY_OBJECT_RETURN_DIRTY, FALSE,
-                               VM_PROT_NO_CHANGE);
-
-       if (kret == KERN_SUCCESS) {
-               /* invalidate last page and old contents beyond nsize */
-               kret = memory_object_lock_request(control,
-                                       (memory_object_offset_t)lastpg,
-                                       (memory_object_size_t)(olastpgend - lastpg),
-                                       MEMORY_OBJECT_RETURN_NONE, MEMORY_OBJECT_DATA_FLUSH,
-                                       VM_PROT_NO_CHANGE);
+       if (lastoff) {
+               upl_t           upl;
+               upl_page_info_t *pl;
+
+
+               /*
+                * new EOF ends up in the middle of a page
+                * zero the tail of this page if its currently
+                * present in the cache
+                */
+               kret = ubc_create_upl(vp, lastpg, PAGE_SIZE, &upl, &pl, UPL_SET_LITE);
+               
                if (kret != KERN_SUCCESS)
                if (kret != KERN_SUCCESS)
-                       printf("ubc_setsize: invalidate failed (error = %d)\n", kret);
-       } else
-               printf("ubc_setsize: flush failed (error = %d)\n", kret);
+                       panic("ubc_setsize: ubc_create_upl (error = %d)\n", kret);
 
 
+               if (upl_valid_page(pl, 0))
+                       cluster_zero(upl, (uint32_t)lastoff, PAGE_SIZE - (uint32_t)lastoff, NULL);
+
+               ubc_upl_abort_range(upl, 0, PAGE_SIZE, UPL_ABORT_FREE_ON_EMPTY);
+
+               lastpg += PAGE_SIZE_64;
+       }
+       if (olastpgend > lastpg) {
+               /*
+                * invalidate the pages beyond the new EOF page
+                *
+                */
+               kret = memory_object_lock_request(control,
+                                                 (memory_object_offset_t)lastpg,
+                                                 (memory_object_size_t)(olastpgend - lastpg), NULL, NULL,
+                                                 MEMORY_OBJECT_RETURN_NONE, MEMORY_OBJECT_DATA_FLUSH,
+                                                 VM_PROT_NO_CHANGE);
+               if (kret != KERN_SUCCESS)
+                       printf("ubc_setsize: invalidate failed (error = %d)\n", kret);
+       }
        return ((kret == KERN_SUCCESS) ? 1 : 0);
 }
 
        return ((kret == KERN_SUCCESS) ? 1 : 0);
 }
 
+
 /*
 /*
- * Get the size of the file
+ * ubc_getsize
+ *
+ * Get the size of the file assocated with the specified vnode
+ *
+ * Parameters: vp                      The vnode whose size is of interest
+ *
+ * Returns:    0                       There is no ubc_info associated with
+ *                                     this vnode, or the size is zero
+ *             !0                      The size of the file
+ *
+ * Notes:      Using this routine, it is not possible for a caller to
+ *             successfully distinguish between a vnode associate with a zero
+ *             length file, and a vnode with no associated ubc_info.  The
+ *             caller therefore needs to not care, or needs to ensure that
+ *             they have previously successfully called ubc_info_init() or
+ *             ubc_info_init_withsize().
  */
 off_t
 ubc_getsize(struct vnode *vp)
 {
  */
 off_t
 ubc_getsize(struct vnode *vp)
 {
+       /* people depend on the side effect of this working this way
+        * as they call this for directory 
+        */
+       if (!UBCINFOEXISTS(vp))
+               return ((off_t)0);
        return (vp->v_ubcinfo->ui_size);
 }
 
        return (vp->v_ubcinfo->ui_size);
 }
 
-/*
- * Caller indicate that the object corresponding to the vnode 
- * can not be cached in object cache. Make it so.
- * returns 1 on success, 0 on failure
- */
-int
-ubc_uncache(struct vnode *vp)
-{
-       kern_return_t kret;
-       struct ubc_info *uip;
-       int    recursed;
-       memory_object_control_t control;
-       memory_object_perf_info_data_t   perf;
-
-       if (!UBCINFOEXISTS(vp))
-               return (0);
-
-       if ((recursed = ubc_busy(vp)) == 0)
-               return (0);
-
-       uip = vp->v_ubcinfo;
-
-       assert(uip != UBC_INFO_NULL);
-
-       /*
-        * AGE it so that vfree() can make sure that it
-        * would get recycled soon after the last reference is gone
-        * This will insure that .nfs turds would not linger
-        */
-       vagevp(vp);
-
-       /* set the "do not cache" bit */
-       SET(uip->ui_flags, UI_DONTCACHE);
-
-       control = uip->ui_control;
-       assert(control);
-
-       perf.cluster_size = PAGE_SIZE; /* XXX use real cluster_size. */
-       perf.may_cache = FALSE;
-       kret = memory_object_change_attributes(control,
-                               MEMORY_OBJECT_PERFORMANCE_INFO,
-                               (memory_object_info_t) &perf,
-                               MEMORY_OBJECT_PERF_INFO_COUNT);
-
-       if (kret != KERN_SUCCESS) {
-               printf("ubc_uncache: memory_object_change_attributes_named "
-                       "kret = %d", kret);
-               if (recursed == 1)
-                       ubc_unbusy(vp);
-               return (0);
-       }
-
-       ubc_release_named(vp);
-
-       if (recursed == 1)
-               ubc_unbusy(vp);
-       return (1);
-}
 
 /*
 
 /*
- * call ubc_clean() and ubc_uncache() on all the vnodes
- * for this mount point.
- * returns 1 on success, 0 on failure
+ * ubc_umount
+ *
+ * Call ubc_sync_range(vp, 0, EOF, UBC_PUSHALL) on all the vnodes for this
+ * mount point
+ *
+ * Parameters: mp                      The mount point
+ *
+ * Returns:    0                       Success
+ *
+ * Notes:      There is no failure indication for this function.
+ *
+ *             This function is used in the unmount path; since it may block
+ *             I/O indefinitely, it should not be used in the forced unmount
+ *             path, since a device unavailability could also block that
+ *             indefinitely.
+ *
+ *             Because there is no device ejection interlock on USB, FireWire,
+ *             or similar devices, it's possible that an ejection that begins
+ *             subsequent to the vnode_iterate() completing, either on one of
+ *             those devices, or a network mount for which the server quits
+ *             responding, etc., may cause the caller to block indefinitely.
  */
 __private_extern__ int
 ubc_umount(struct mount *mp)
 {
  */
 __private_extern__ int
 ubc_umount(struct mount *mp)
 {
-       struct proc *p = current_proc();
-       struct vnode *vp, *nvp;
-       int ret = 1;
-
-loop:
-       simple_lock(&mntvnode_slock);
-       for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) {
-               if (vp->v_mount != mp) {
-                       simple_unlock(&mntvnode_slock);
-                       goto loop;
-               }
-               nvp = vp->v_mntvnodes.le_next;
-               simple_unlock(&mntvnode_slock);
-               if (UBCINFOEXISTS(vp)) {
-
-                       /*
-                        * Must get a valid reference on the vnode
-                        * before callig UBC functions
-                        */
-                       if (vget(vp, 0, p)) {
-                               ret = 0;
-                               simple_lock(&mntvnode_slock);
-                               continue; /* move on to the next vnode */
-                       }
-                       ret &= ubc_clean(vp, 0); /* do not invalidate */
-                       ret &= ubc_uncache(vp);
-                       vrele(vp);
-               }
-               simple_lock(&mntvnode_slock);
-       }
-       simple_unlock(&mntvnode_slock);
-       return (ret);
+       vnode_iterate(mp, 0, ubc_umcallback, 0);
+       return(0);
 }
 
 }
 
+
 /*
 /*
- * Call ubc_unmount() for all filesystems.
- * The list is traversed in reverse order
- * of mounting to avoid dependencies.
+ * ubc_umcallback
+ *
+ * Used by ubc_umount() as an internal implementation detail; see ubc_umount()
+ * and vnode_iterate() for details of implementation.
  */
  */
-__private_extern__ void
-ubc_unmountall()
+static int
+ubc_umcallback(vnode_t vp, __unused void * args)
 {
 {
-       struct mount *mp, *nmp;
 
 
-       /*
-        * Since this only runs when rebooting, it is not interlocked.
-        */
-       for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
-               nmp = mp->mnt_list.cqe_prev;
-               (void) ubc_umount(mp);
+       if (UBCINFOEXISTS(vp)) {
+
+               (void) ubc_msync(vp, (off_t)0, ubc_getsize(vp), NULL, UBC_PUSHALL);
        }
        }
+       return (VNODE_RETURNED);
 }
 
 }
 
-/* Get the credentials */
-struct ucred *
+
+/*
+ * ubc_getcred
+ *
+ * Get the credentials currently active for the ubc_info associated with the
+ * vnode.
+ *
+ * Parameters: vp                      The vnode whose ubc_info credentials
+ *                                     are to be retrieved
+ *
+ * Returns:    !NOCRED                 The credentials
+ *             NOCRED                  If there is no ubc_info for the vnode,
+ *                                     or if there is one, but it has not had
+ *                                     any credentials associated with it via
+ *                                     a call to ubc_setcred()
+ */
+kauth_cred_t
 ubc_getcred(struct vnode *vp)
 {
 ubc_getcred(struct vnode *vp)
 {
-       struct ubc_info *uip;
-
-       uip = vp->v_ubcinfo;
+        if (UBCINFOEXISTS(vp))
+               return (vp->v_ubcinfo->ui_ucred);
 
 
-       if (UBCINVALID(vp))
-               return (NOCRED);
-
-       return (uip->ui_ucred);
+       return (NOCRED);
 }
 
 }
 
+
 /*
 /*
- * Set the credentials
- * existing credentials are not changed
- * returns 1 on success and 0 on failure
+ * ubc_setthreadcred
+ *
+ * If they are not already set, set the credentials of the ubc_info structure
+ * associated with the vnode to those of the supplied thread; otherwise leave
+ * them alone.
+ *
+ * Parameters: vp                      The vnode whose ubc_info creds are to
+ *                                     be set
+ *             p                       The process whose credentials are to
+ *                                     be used, if not running on an assumed
+ *                                     credential
+ *             thread                  The thread whose credentials are to
+ *                                     be used
+ *
+ * Returns:    1                       This vnode has no associated ubc_info
+ *             0                       Success
+ *
+ * Notes:      This function takes a proc parameter to account for bootstrap
+ *             issues where a task or thread may call this routine, either
+ *             before credentials have been initialized by bsd_init(), or if
+ *             there is no BSD info asscoiate with a mach thread yet.  This
+ *             is known to happen in both the initial swap and memory mapping
+ *             calls.
+ *
+ *             This function is generally used only in the following cases:
+ *
+ *             o       a memory mapped file via the mmap() system call
+ *             o       a memory mapped file via the deprecated map_fd() call
+ *             o       a swap store backing file
+ *             o       subsequent to a successful write via vn_write()
+ *
+ *             The information is then used by the NFS client in order to
+ *             cons up a wire message in either the page-in or page-out path.
+ *
+ *             There are two potential problems with the use of this API:
+ *
+ *             o       Because the write path only set it on a successful
+ *                     write, there is a race window between setting the
+ *                     credential and its use to evict the pages to the
+ *                     remote file server
+ *
+ *             o       Because a page-in may occur prior to a write, the
+ *                     credential may not be set at this time, if the page-in
+ *                     is not the result of a mapping established via mmap()
+ *                     or map_fd().
+ *
+ *             In both these cases, this will be triggered from the paging
+ *             path, which will instead use the credential of the current
+ *             process, which in this case is either the dynamic_pager or
+ *             the kernel task, both of which utilize "root" credentials.
+ *
+ *             This may potentially permit operations to occur which should
+ *             be denied, or it may cause to be denied operations which
+ *             should be permitted, depending on the configuration of the NFS
+ *             server.
  */
 int
  */
 int
-ubc_setcred(struct vnode *vp, struct proc *p)
+ubc_setthreadcred(struct vnode *vp, proc_t p, thread_t thread)
 {
        struct ubc_info *uip;
 {
        struct ubc_info *uip;
-       struct ucred *credp;
+       kauth_cred_t credp;
+       struct uthread  *uthread = get_bsdthread_info(thread);
 
 
-       uip = vp->v_ubcinfo;
+       if (!UBCINFOEXISTS(vp))
+               return (1); 
 
 
-       if (UBCINVALID(vp))
-               return (0); 
+       vnode_lock(vp);
 
 
+       uip = vp->v_ubcinfo;
        credp = uip->ui_ucred;
        credp = uip->ui_ucred;
-       if (credp == NOCRED) {
-               crhold(p->p_ucred);
-               uip->ui_ucred = p->p_ucred;
+
+       if (!IS_VALID_CRED(credp)) {
+               /* use per-thread cred, if assumed identity, else proc cred */
+               if (uthread == NULL || (uthread->uu_flag & UT_SETUID) == 0) {
+                       uip->ui_ucred = kauth_cred_proc_ref(p);
+               } else {
+                       uip->ui_ucred = uthread->uu_ucred;
+                       kauth_cred_ref(uip->ui_ucred);
+               }
        } 
        } 
+       vnode_unlock(vp);
 
 
-       return (1);
+       return (0);
 }
 
 }
 
-/* Get the pager */
-__private_extern__ memory_object_t
-ubc_getpager(struct vnode *vp)
-{
-       struct ubc_info *uip;
-
-       uip = vp->v_ubcinfo;
-
-       if (UBCINVALID(vp))
-               return (0);
-
-       return (uip->ui_pager);
-}
 
 /*
 
 /*
- * Get the memory object associated with this vnode
- * If the vnode was reactivated, memory object would not exist.
- * Unless "do not rectivate" was specified, look it up using the pager.
- * If hold was requested create an object reference of one does not
- * exist already.
+ * ubc_setcred
+ *
+ * If they are not already set, set the credentials of the ubc_info structure
+ * associated with the vnode to those of the process; otherwise leave them
+ * alone.
+ *
+ * Parameters: vp                      The vnode whose ubc_info creds are to
+ *                                     be set
+ *             p                       The process whose credentials are to
+ *                                     be used
+ *
+ * Returns:    0                       This vnode has no associated ubc_info
+ *             1                       Success
+ *
+ * Notes:      The return values for this function are inverted from nearly
+ *             all other uses in the kernel.
+ *
+ *             See also ubc_setthreadcred(), above.
+ *
+ *             This function is considered deprecated, and generally should
+ *             not be used, as it is incompatible with per-thread credentials;
+ *             it exists for legacy KPI reasons.
+ *
+ * DEPRECATION:        ubc_setcred() is being deprecated. Please use 
+ *             ubc_setthreadcred() instead.
  */
  */
-
-memory_object_control_t
-ubc_getobject(struct vnode *vp, int flags)
-{
-       struct ubc_info *uip;
-       int    recursed;
-       memory_object_control_t control;
-
-       if (UBCINVALID(vp))
-               return (0);
-
-       if (flags & UBC_FOR_PAGEOUT)
-               return(vp->v_ubcinfo->ui_control);
-
-       if ((recursed = ubc_busy(vp)) == 0)
-               return (0);
-
-       uip = vp->v_ubcinfo;
-       control = uip->ui_control;
-
-       if ((flags & UBC_HOLDOBJECT) && (!ISSET(uip->ui_flags, UI_HASOBJREF))) {
-
-               /*
-                * Take a temporary reference on the ubc info so that it won't go
-                * away during our recovery attempt.
-                */
-               ubc_lock(vp);
-               uip->ui_refcount++;
-               ubc_unlock(vp);
-               if (memory_object_recover_named(control, TRUE) == KERN_SUCCESS) {
-                       SET(uip->ui_flags, UI_HASOBJREF);
-               } else {
-                       control = MEMORY_OBJECT_CONTROL_NULL;
-               }
-               if (recursed == 1)
-                       ubc_unbusy(vp);
-               ubc_info_deallocate(uip);
-
-       } else {
-               if (recursed == 1)
-                       ubc_unbusy(vp);
-       }
-
-       return (control);
-}
-
-/* Set the pager */
 int
 int
-ubc_setpager(struct vnode *vp, memory_object_t pager)
+ubc_setcred(struct vnode *vp, proc_t p)
 {
        struct ubc_info *uip;
 {
        struct ubc_info *uip;
+       kauth_cred_t credp;
 
 
-       uip = vp->v_ubcinfo;
+       /* If there is no ubc_info, deny the operation */
+       if ( !UBCINFOEXISTS(vp))
+               return (0); 
 
 
-       if (UBCINVALID(vp))
-               return (0);
+       /*
+        * Check to see if there is already a credential reference in the
+        * ubc_info; if there is not, take one on the supplied credential.
+        */
+       vnode_lock(vp);
+       uip = vp->v_ubcinfo;
+       credp = uip->ui_ucred;
+       if (!IS_VALID_CRED(credp)) {
+               uip->ui_ucred = kauth_cred_proc_ref(p);
+       } 
+       vnode_unlock(vp);
 
 
-       uip->ui_pager = pager;
        return (1);
 }
 
        return (1);
 }
 
-int 
-ubc_setflags(struct vnode * vp, int  flags)
-{
-       struct ubc_info *uip;
-
-       if (UBCINVALID(vp))
-               return (0);
-
-       uip = vp->v_ubcinfo;
-
-       SET(uip->ui_flags, flags);
-
-       return (1);     
-} 
 
 
-int 
-ubc_clearflags(struct vnode * vp, int  flags)
+/*
+ * ubc_getpager
+ *
+ * Get the pager associated with the ubc_info associated with the vnode.
+ *
+ * Parameters: vp                      The vnode to obtain the pager from
+ *
+ * Returns:    !VNODE_PAGER_NULL       The memory_object_t for the pager
+ *             VNODE_PAGER_NULL        There is no ubc_info for this vnode
+ *
+ * Notes:      For each vnode that has a ubc_info associated with it, that
+ *             ubc_info SHALL have a pager associated with it, so in the
+ *             normal case, it's impossible to return VNODE_PAGER_NULL for
+ *             a vnode with an associated ubc_info.
+ */
+__private_extern__ memory_object_t
+ubc_getpager(struct vnode *vp)
 {
 {
-       struct ubc_info *uip;
-
-       if (UBCINVALID(vp))
-               return (0);
-
-       uip = vp->v_ubcinfo;
+        if (UBCINFOEXISTS(vp))
+               return (vp->v_ubcinfo->ui_pager);
 
 
-       CLR(uip->ui_flags, flags);
-
-       return (1);     
-} 
+       return (0);
+}
 
 
 
 
-int 
-ubc_issetflags(struct vnode * vp, int  flags)
+/*
+ * ubc_getobject
+ *
+ * Get the memory object control associated with the ubc_info associated with
+ * the vnode
+ *
+ * Parameters: vp                      The vnode to obtain the memory object
+ *                                     from
+ *             flags                   DEPRECATED
+ *
+ * Returns:    !MEMORY_OBJECT_CONTROL_NULL
+ *             MEMORY_OBJECT_CONTROL_NULL
+ *
+ * Notes:      Historically, if the flags were not "do not reactivate", this
+ *             function would look up the memory object using the pager if
+ *             it did not exist (this could be the case if the vnode had
+ *             been previously reactivated).  The flags would also permit a
+ *             hold to be requested, which would have created an object
+ *             reference, if one had not already existed.  This usage is
+ *             deprecated, as it would permit a race between finding and
+ *             taking the reference vs. a single reference being dropped in
+ *             another thread.
+ */
+memory_object_control_t
+ubc_getobject(struct vnode *vp, __unused int flags)
 {
 {
-       struct ubc_info *uip;
+        if (UBCINFOEXISTS(vp))
+               return((vp->v_ubcinfo->ui_control));
 
 
-       if (UBCINVALID(vp))
-               return (0);
-
-       uip = vp->v_ubcinfo;
+       return (MEMORY_OBJECT_CONTROL_NULL);
+}
 
 
-       return (ISSET(uip->ui_flags, flags));
-} 
 
 
+/*
+ * ubc_blktooff
+ *
+ * Convert a given block number to a memory backing object (file) offset for a
+ * given vnode
+ *
+ * Parameters: vp                      The vnode in which the block is located
+ *             blkno                   The block number to convert
+ *
+ * Returns:    !-1                     The offset into the backing object
+ *             -1                      There is no ubc_info associated with
+ *                                     the vnode
+ *             -1                      An error occurred in the underlying VFS
+ *                                     while translating the block to an
+ *                                     offset; the most likely cause is that
+ *                                     the caller specified a block past the
+ *                                     end of the file, but this could also be
+ *                                     any other error from VNOP_BLKTOOFF().
+ *
+ * Note:       Representing the error in band loses some information, but does
+ *             not occlude a valid offset, since an off_t of -1 is normally
+ *             used to represent EOF.  If we had a more reliable constant in
+ *             our header files for it (i.e. explicitly cast to an off_t), we
+ *             would use it here instead.
+ */
 off_t
 off_t
-ubc_blktooff(struct vnode *vp, daddr_t blkno)
+ubc_blktooff(vnode_t vp, daddr64_t blkno)
 {
 {
-       off_t file_offset;
+       off_t file_offset = -1;
        int error;
 
        int error;
 
-    if (UBCINVALID(vp))
-        return ((off_t)-1);
-
-       error = VOP_BLKTOOFF(vp, blkno, &file_offset);
-       if (error)
-               file_offset = -1;
+       if (UBCINFOEXISTS(vp)) {
+               error = VNOP_BLKTOOFF(vp, blkno, &file_offset);
+               if (error)
+                       file_offset = -1;
+       }
 
        return (file_offset);
 }
 
 
        return (file_offset);
 }
 
-daddr_t
-ubc_offtoblk(struct vnode *vp, off_t offset)
+
+/*
+ * ubc_offtoblk
+ *
+ * Convert a given offset in a memory backing object into a block number for a
+ * given vnode
+ *
+ * Parameters: vp                      The vnode in which the offset is
+ *                                     located
+ *             offset                  The offset into the backing object
+ *
+ * Returns:    !-1                     The returned block number
+ *             -1                      There is no ubc_info associated with
+ *                                     the vnode
+ *             -1                      An error occurred in the underlying VFS
+ *                                     while translating the block to an
+ *                                     offset; the most likely cause is that
+ *                                     the caller specified a block past the
+ *                                     end of the file, but this could also be
+ *                                     any other error from VNOP_OFFTOBLK().
+ *
+ * Note:       Representing the error in band loses some information, but does
+ *             not occlude a valid block number, since block numbers exceed
+ *             the valid range for offsets, due to their relative sizes.  If
+ *             we had a more reliable constant than -1 in our header files
+ *             for it (i.e. explicitly cast to an daddr64_t), we would use it
+ *             here instead.
+ */
+daddr64_t
+ubc_offtoblk(vnode_t vp, off_t offset)
 {
 {
-       daddr_t blkno;
+       daddr64_t blkno = -1;
        int error = 0;
 
        int error = 0;
 
-    if (UBCINVALID(vp)) { 
-        return ((daddr_t)-1);
-    }   
-
-       error = VOP_OFFTOBLK(vp, offset, &blkno);
-       if (error)
-               blkno = -1;
+       if (UBCINFOEXISTS(vp)) {
+               error = VNOP_OFFTOBLK(vp, offset, &blkno);
+               if (error)
+                       blkno = -1;
+       }
 
        return (blkno);
 }
 
 
        return (blkno);
 }
 
+
 /*
 /*
- * Cause the file data in VM to be pushed out to the storage
- * it also causes all currently valid pages to be released
- * returns 1 on success, 0 on failure
+ * ubc_pages_resident
+ *
+ * Determine whether or not a given vnode has pages resident via the memory
+ * object control associated with the ubc_info associated with the vnode
+ *
+ * Parameters: vp                      The vnode we want to know about
+ *
+ * Returns:    1                       Yes
+ *             0                       No
  */
 int
  */
 int
-ubc_clean(struct vnode *vp, int invalidate)
+ubc_pages_resident(vnode_t vp)
 {
 {
-       off_t size;
-       struct ubc_info *uip;
-       memory_object_control_t control;
-       kern_return_t kret;
-       int flags = 0;
-
-       if (UBCINVALID(vp))
-               return (0);
-
+       kern_return_t           kret;
+       boolean_t                       has_pages_resident;
+       
        if (!UBCINFOEXISTS(vp))
                return (0);
        if (!UBCINFOEXISTS(vp))
                return (0);
-
+                       
        /*
        /*
-        * if invalidate was requested, write dirty data and then discard
-        * the resident pages
+        * The following call may fail if an invalid ui_control is specified,
+        * or if there is no VM object associated with the control object.  In
+        * either case, reacting to it as if there were no pages resident will
+        * result in correct behavior.
         */
         */
-       if (invalidate)
-               flags = (MEMORY_OBJECT_DATA_FLUSH | MEMORY_OBJECT_DATA_NO_CHANGE);
-
-       uip = vp->v_ubcinfo;
-       size = uip->ui_size;    /* call ubc_getsize() ??? */
-
-       control = uip->ui_control;
-       assert(control);
-
-       cluster_release(vp);
-       vp->v_clen = 0;
-
-       /* Write the dirty data in the file and discard cached pages */
-       kret = memory_object_lock_request(control,
-                               (memory_object_offset_t)0,
-                               (memory_object_size_t)round_page_64(size),
-                               MEMORY_OBJECT_RETURN_ALL, flags,
-                               VM_PROT_NO_CHANGE);
-
+       kret = memory_object_pages_resident(vp->v_ubcinfo->ui_control, &has_pages_resident);
+       
        if (kret != KERN_SUCCESS)
        if (kret != KERN_SUCCESS)
-               printf("ubc_clean: clean failed (error = %d)\n", kret);
-
-       return ((kret == KERN_SUCCESS) ? 1 : 0);
+               return (0);
+               
+       if (has_pages_resident == TRUE)
+               return (1);
+               
+       return (0);
 }
 
 }
 
+
 /*
 /*
- * Cause the file data in VM to be pushed out to the storage
- * currently valid pages are NOT invalidated
- * returns 1 on success, 0 on failure
+ * ubc_sync_range
+ *
+ * Clean and/or invalidate a range in the memory object that backs this vnode
+ *
+ * Parameters: vp                      The vnode whose associated ubc_info's
+ *                                     associated memory object is to have a
+ *                                     range invalidated within it
+ *             beg_off                 The start of the range, as an offset
+ *             end_off                 The end of the range, as an offset
+ *             flags                   See ubc_msync_internal()
+ *
+ * Returns:    1                       Success
+ *             0                       Failure
+ *
+ * Notes:      see ubc_msync_internal() for more detailed information.
+ *
+ * DEPRECATED: This interface is obsolete due to a failure to return error
+ *             information needed in order to correct failures.  The currently
+ *             recommended interface is ubc_msync().
  */
 int
  */
 int
-ubc_pushdirty(struct vnode *vp)
+ubc_sync_range(vnode_t vp, off_t beg_off, off_t end_off, int flags)
 {
 {
-       off_t size;
-       struct ubc_info *uip;
-       memory_object_control_t control;
-       kern_return_t kret;
-
-       if (UBCINVALID(vp))
-               return (0);
-
-       if (!UBCINFOEXISTS(vp))
-               return (0);
-
-       uip = vp->v_ubcinfo;
-       size = uip->ui_size;    /* call ubc_getsize() ??? */
-
-       control = uip->ui_control;
-       assert(control);
-
-       vp->v_flag &= ~VHASDIRTY;
-       vp->v_clen = 0;
-
-       /* Write the dirty data in the file and discard cached pages */
-       kret = memory_object_lock_request(control,
-                               (memory_object_offset_t)0,
-                               (memory_object_size_t)round_page_64(size),
-                               MEMORY_OBJECT_RETURN_DIRTY, FALSE,
-                               VM_PROT_NO_CHANGE);
-
-       if (kret != KERN_SUCCESS)
-               printf("ubc_pushdirty: flush failed (error = %d)\n", kret);
-
-       return ((kret == KERN_SUCCESS) ? 1 : 0);
+        return (ubc_msync_internal(vp, beg_off, end_off, NULL, flags, NULL));
 }
 
 }
 
+
 /*
 /*
- * Cause the file data in VM to be pushed out to the storage
- * currently valid pages are NOT invalidated
- * returns 1 on success, 0 on failure
+ * ubc_msync
+ *
+ * Clean and/or invalidate a range in the memory object that backs this vnode
+ *
+ * Parameters: vp                      The vnode whose associated ubc_info's
+ *                                     associated memory object is to have a
+ *                                     range invalidated within it
+ *             beg_off                 The start of the range, as an offset
+ *             end_off                 The end of the range, as an offset
+ *             resid_off               The address of an off_t supplied by the
+ *                                     caller; may be set to NULL to ignore
+ *             flags                   See ubc_msync_internal()
+ *
+ * Returns:    0                       Success
+ *             !0                      Failure; an errno is returned
+ *
+ * Implicit Returns:
+ *             *resid_off, modified    If non-NULL, the  contents are ALWAYS
+ *                                     modified; they are initialized to the
+ *                                     beg_off, and in case of an I/O error,
+ *                                     the difference between beg_off and the
+ *                                     current value will reflect what was
+ *                                     able to be written before the error
+ *                                     occurred.  If no error is returned, the
+ *                                     value of the resid_off is undefined; do
+ *                                     NOT use it in place of end_off if you
+ *                                     intend to increment from the end of the
+ *                                     last call and call iteratively.
+ *
+ * Notes:      see ubc_msync_internal() for more detailed information.
+ *
  */
  */
-int
-ubc_pushdirty_range(struct vnode *vp, off_t offset, off_t size)
+errno_t
+ubc_msync(vnode_t vp, off_t beg_off, off_t end_off, off_t *resid_off, int flags)
 {
 {
-       struct ubc_info *uip;
-       memory_object_control_t control;
-       kern_return_t kret;
-
-       if (UBCINVALID(vp))
-               return (0);
-
-       if (!UBCINFOEXISTS(vp))
-               return (0);
-
-       uip = vp->v_ubcinfo;
-
-       control = uip->ui_control;
-       assert(control);
-
-       /* Write any dirty pages in the requested range of the file: */
-       kret = memory_object_lock_request(control,
-                               (memory_object_offset_t)offset,
-                               (memory_object_size_t)round_page_64(size),
-                               MEMORY_OBJECT_RETURN_DIRTY, FALSE,
-                               VM_PROT_NO_CHANGE);
+        int retval;
+       int io_errno = 0;
+       
+       if (resid_off)
+               *resid_off = beg_off;
 
 
-       if (kret != KERN_SUCCESS)
-               printf("ubc_pushdirty_range: flush failed (error = %d)\n", kret);
+        retval = ubc_msync_internal(vp, beg_off, end_off, resid_off, flags, &io_errno);
 
 
-       return ((kret == KERN_SUCCESS) ? 1 : 0);
+       if (retval == 0 && io_errno == 0)
+               return (EINVAL);
+       return (io_errno);
 }
 
 }
 
+
 /*
 /*
- * Make sure the vm object does not vanish 
- * returns 1 if the hold count was incremented
- * returns 0 if the hold count was not incremented
- * This return value should be used to balance 
- * ubc_hold() and ubc_rele().
+ * Clean and/or invalidate a range in the memory object that backs this vnode
+ *
+ * Parameters: vp                      The vnode whose associated ubc_info's
+ *                                     associated memory object is to have a
+ *                                     range invalidated within it
+ *             beg_off                 The start of the range, as an offset
+ *             end_off                 The end of the range, as an offset
+ *             resid_off               The address of an off_t supplied by the
+ *                                     caller; may be set to NULL to ignore
+ *             flags                   MUST contain at least one of the flags
+ *                                     UBC_INVALIDATE, UBC_PUSHDIRTY, or
+ *                                     UBC_PUSHALL; if UBC_PUSHDIRTY is used,
+ *                                     UBC_SYNC may also be specified to cause
+ *                                     this function to block until the
+ *                                     operation is complete.  The behavior
+ *                                     of UBC_SYNC is otherwise undefined.
+ *             io_errno                The address of an int to contain the
+ *                                     errno from a failed I/O operation, if
+ *                                     one occurs; may be set to NULL to
+ *                                     ignore
+ *
+ * Returns:    1                       Success
+ *             0                       Failure
+ *
+ * Implicit Returns:
+ *             *resid_off, modified    The contents of this offset MAY be
+ *                                     modified; in case of an I/O error, the
+ *                                     difference between beg_off and the
+ *                                     current value will reflect what was
+ *                                     able to be written before the error
+ *                                     occurred.
+ *             *io_errno, modified     The contents of this offset are set to
+ *                                     an errno, if an error occurs; if the
+ *                                     caller supplies an io_errno parameter,
+ *                                     they should be careful to initialize it
+ *                                     to 0 before calling this function to
+ *                                     enable them to distinguish an error
+ *                                     with a valid *resid_off from an invalid
+ *                                     one, and to avoid potentially falsely
+ *                                     reporting an error, depending on use.
+ *
+ * Notes:      If there is no ubc_info associated with the vnode supplied,
+ *             this function immediately returns success.
+ *
+ *             If the value of end_off is less than or equal to beg_off, this
+ *             function immediately returns success; that is, end_off is NOT
+ *             inclusive.
+ *
+ *             IMPORTANT: one of the flags UBC_INVALIDATE, UBC_PUSHDIRTY, or
+ *             UBC_PUSHALL MUST be specified; that is, it is NOT possible to
+ *             attempt to block on in-progress I/O by calling this function
+ *             with UBC_PUSHDIRTY, and then later call it with just UBC_SYNC
+ *             in order to block pending on the I/O already in progress.
+ *
+ *             The start offset is truncated to the page boundary and the
+ *             size is adjusted to include the last page in the range; that
+ *             is, end_off on exactly a page boundary will not change if it
+ *             is rounded, and the range of bytes written will be from the
+ *             truncate beg_off to the rounded (end_off - 1).
  */
  */
-int
-ubc_hold(struct vnode *vp)
+static int
+ubc_msync_internal(vnode_t vp, off_t beg_off, off_t end_off, off_t *resid_off, int flags, int *io_errno)
 {
 {
-       struct ubc_info *uip;
-       int    recursed;
-       memory_object_control_t object;
-
-retry:
-
-       if (UBCINVALID(vp))
-               return (0);
-
-       ubc_lock(vp);
-       if (ISSET(vp->v_flag,  VUINIT)) {
-               /*
-                * other thread is not done initializing this
-                * yet, wait till it's done and try again
+       memory_object_size_t    tsize;
+       kern_return_t           kret;
+       int request_flags = 0;
+       int flush_flags   = MEMORY_OBJECT_RETURN_NONE;
+       
+       if ( !UBCINFOEXISTS(vp))
+               return (0);
+       if ((flags & (UBC_INVALIDATE | UBC_PUSHDIRTY | UBC_PUSHALL)) == 0)
+               return (0);
+       if (end_off <= beg_off)
+               return (1);
+
+       if (flags & UBC_INVALIDATE)
+               /*
+                * discard the resident pages
                 */
                 */
-               while (ISSET(vp->v_flag,  VUINIT)) {
-                       SET(vp->v_flag, VUWANT); /* XXX overloaded! */
-                       ubc_unlock(vp);
-                       (void) tsleep((caddr_t)vp, PINOD, "ubchold", 0);
-                       ubc_lock(vp);
-               }
-               ubc_unlock(vp);
-               goto retry;
-       }
-       ubc_unlock(vp);
-
-       if ((recursed = ubc_busy(vp)) == 0) {
-               /* must be invalid or dying vnode */
-               assert(UBCINVALID(vp) ||
-                       ((vp->v_flag & VXLOCK) || (vp->v_flag & VTERMINATE)));
-               return (0);
-       }
+               request_flags = (MEMORY_OBJECT_DATA_FLUSH | MEMORY_OBJECT_DATA_NO_CHANGE);
 
 
-       uip = vp->v_ubcinfo;
-       assert(uip->ui_control != MEMORY_OBJECT_CONTROL_NULL);
-
-       ubc_lock(vp);
-       uip->ui_refcount++;
-       ubc_unlock(vp);
-
-       if (!ISSET(uip->ui_flags, UI_HASOBJREF)) {
-               if (memory_object_recover_named(uip->ui_control, TRUE)
-                       != KERN_SUCCESS) {
-                       if (recursed == 1)
-                               ubc_unbusy(vp);
-                       ubc_info_deallocate(uip);
-                       return (0);
-               }
-               SET(uip->ui_flags, UI_HASOBJREF);
-       }
-       if (recursed == 1)
-               ubc_unbusy(vp);
+       if (flags & UBC_SYNC)
+               /*
+                * wait for all the I/O to complete before returning
+                */
+               request_flags |= MEMORY_OBJECT_IO_SYNC;
 
 
-       assert(uip->ui_refcount > 0);
+       if (flags & UBC_PUSHDIRTY)
+               /*
+                * we only return the dirty pages in the range
+                */
+               flush_flags = MEMORY_OBJECT_RETURN_DIRTY;
 
 
-       return (1);
+       if (flags & UBC_PUSHALL)
+               /*
+                * then return all the interesting pages in the range (both
+                * dirty and precious) to the pager
+                */
+               flush_flags = MEMORY_OBJECT_RETURN_ALL;
+
+       beg_off = trunc_page_64(beg_off);
+       end_off = round_page_64(end_off);
+       tsize   = (memory_object_size_t)end_off - beg_off;
+
+       /* flush and/or invalidate pages in the range requested */
+       kret = memory_object_lock_request(vp->v_ubcinfo->ui_control,
+                                         beg_off, tsize,
+                                         (memory_object_offset_t *)resid_off,
+                                         io_errno, flush_flags, request_flags,
+                                         VM_PROT_NO_CHANGE);
+       
+       return ((kret == KERN_SUCCESS) ? 1 : 0);
 }
 
 }
 
-/*
- * Drop the holdcount.
- * release the reference on the vm object if the this is "uncached"
- * ubc_info.
- */
-void
-ubc_rele(struct vnode *vp)
-{
-       struct ubc_info *uip;
-
-       if (UBCINVALID(vp))
-               return;
-
-       if (!UBCINFOEXISTS(vp)) {
-               /* nothing more to do for a dying vnode */
-               if ((vp->v_flag & VXLOCK) || (vp->v_flag & VTERMINATE))
-                       return;
-               panic("ubc_rele: can not");
-       }
-
-       uip = vp->v_ubcinfo;
-
-       if (uip->ui_refcount == 1)
-               panic("ubc_rele: ui_refcount");
-
-       --uip->ui_refcount;
-
-       if ((uip->ui_refcount == 1)
-               && ISSET(uip->ui_flags, UI_DONTCACHE))
-               (void) ubc_release_named(vp);
-
-       return;
-}
 
 /*
 
 /*
- * The vnode is mapped explicitly, mark it so.
+ * ubc_msync_internal
+ *
+ * Explicitly map a vnode that has an associate ubc_info, and add a reference
+ * to it for the ubc system, if there isn't one already, so it will not be
+ * recycled while it's in use, and set flags on the ubc_info to indicate that
+ * we have done this
+ *
+ * Parameters: vp                      The vnode to map
+ *             flags                   The mapping flags for the vnode; this
+ *                                     will be a combination of one or more of
+ *                                     PROT_READ, PROT_WRITE, and PROT_EXEC
+ *
+ * Returns:    0                       Success
+ *             EPERM                   Permission was denied
+ *
+ * Notes:      An I/O reference on the vnode must already be held on entry
+ *
+ *             If there is no ubc_info associated with the vnode, this function
+ *             will return success.
+ *
+ *             If a permission error occurs, this function will return
+ *             failure; all other failures will cause this function to return
+ *             success.
+ *
+ *             IMPORTANT: This is an internal use function, and its symbols
+ *             are not exported, hence its error checking is not very robust.
+ *             It is primarily used by:
+ *
+ *             o       mmap(), when mapping a file
+ *             o       The deprecated map_fd() interface, when mapping a file
+ *             o       When mapping a shared file (a shared library in the
+ *                     shared segment region)
+ *             o       When loading a program image during the exec process
+ *
+ *             ...all of these uses ignore the return code, and any fault that
+ *             results later because of a failure is handled in the fix-up path
+ *             of the fault handler.  The interface exists primarily as a
+ *             performance hint.
+ *
+ *             Given that third party implementation of the type of interfaces
+ *             that would use this function, such as alternative executable
+ *             formats, etc., are unsupported, this function is not exported
+ *             for general use.
+ *
+ *             The extra reference is held until the VM system unmaps the
+ *             vnode from its own context to maintain a vnode reference in
+ *             cases like open()/mmap()/close(), which leave the backing
+ *             object referenced by a mapped memory region in a process
+ *             address space.
  */
  */
-__private_extern__ void
-ubc_map(struct vnode *vp)
+__private_extern__ int
+ubc_map(vnode_t vp, int flags)
 {
        struct ubc_info *uip;
 {
        struct ubc_info *uip;
+       int error = 0;
+       int need_ref = 0;
+       int need_wakeup = 0;
 
 
-       if (UBCINVALID(vp))
-               return;
+       if (UBCINFOEXISTS(vp)) {
 
 
-       if (!UBCINFOEXISTS(vp))
-               return;
+               vnode_lock(vp);
+               uip = vp->v_ubcinfo;
 
 
-       ubc_lock(vp);
-       uip = vp->v_ubcinfo;
+               while (ISSET(uip->ui_flags, UI_MAPBUSY)) {
+                       SET(uip->ui_flags, UI_MAPWAITING);
+                       (void) msleep(&uip->ui_flags, &vp->v_lock,
+                                     PRIBIO, "ubc_map", NULL);
+               }
+               SET(uip->ui_flags, UI_MAPBUSY);
+               vnode_unlock(vp);
 
 
-       SET(uip->ui_flags, UI_WASMAPPED);
-       uip->ui_mapped = 1;
-       ubc_unlock(vp);
+               error = VNOP_MMAP(vp, flags, vfs_context_current());
 
 
-       return;
-}
+               if (error != EPERM)
+                       error = 0;
 
 
-/*
- * Release the memory object reference on the vnode
- * only if it is not in use
- * Return 1 if the reference was released, 0 otherwise.
- */
-int
-ubc_release_named(struct vnode *vp)
-{
-       struct ubc_info *uip;
-       int    recursed;
-       memory_object_control_t control;
-       kern_return_t kret = KERN_FAILURE;
+               vnode_lock_spin(vp);
 
 
-       if (UBCINVALID(vp))
-               return (0);
-
-       if ((recursed = ubc_busy(vp)) == 0)
-               return (0);
-       uip = vp->v_ubcinfo;
-
-       /* can not release held or mapped vnodes */
-       if (ISSET(uip->ui_flags, UI_HASOBJREF) && 
-               (uip->ui_refcount == 1) && !uip->ui_mapped) {
-               control = uip->ui_control;
-               assert(control);
+               if (error == 0) {
+                       if ( !ISSET(uip->ui_flags, UI_ISMAPPED))
+                               need_ref = 1;
+                       SET(uip->ui_flags, (UI_WASMAPPED | UI_ISMAPPED));
+               }
+               CLR(uip->ui_flags, UI_MAPBUSY);
 
 
-               // XXXdbg
-               if (vp->v_flag & VDELETED) {
-                   ubc_setsize(vp, (off_t)0);
+               if (ISSET(uip->ui_flags, UI_MAPWAITING)) {
+                       CLR(uip->ui_flags, UI_MAPWAITING);
+                       need_wakeup = 1;
                }
                }
+               vnode_unlock(vp);
 
 
-               CLR(uip->ui_flags, UI_HASOBJREF);
-               kret = memory_object_release_name(control,
-                               MEMORY_OBJECT_RESPECT_CACHE);
-       }
+               if (need_wakeup)
+                       wakeup(&uip->ui_flags);
 
 
-       if (recursed == 1)
-               ubc_unbusy(vp);
-       return ((kret != KERN_SUCCESS) ? 0 : 1);
+               if (need_ref)
+                       vnode_ref(vp);
+       }
+       return (error);
 }
 
 }
 
-/*
- * This function used to called by extensions directly.  Some may
- * still exist with this behavior.  In those cases, we will do the
- * release as part of reclaiming or cleaning the vnode.  We don't
- * need anything explicit - so just stub this out until those callers
- * get cleaned up.
- */
-int
-ubc_release(
-       struct vnode    *vp)
-{
-       return 0;
-}
 
 /*
 
 /*
- * destroy the named reference for a given vnode
+ * ubc_destroy_named
+ *
+ * Destroy the named memory object associated with the ubc_info control object
+ * associated with the designated vnode, if there is a ubc_info associated
+ * with the vnode, and a control object is associated with it
+ *
+ * Parameters: vp                      The designated vnode
+ *
+ * Returns:    (void)
+ *
+ * Notes:      This function is called on vnode termination for all vnodes,
+ *             and must therefore not assume that there is a ubc_info that is
+ *             associated with the vnode, nor that there is a control object
+ *             associated with the ubc_info.
+ *
+ *             If all the conditions necessary are present, this function
+ *             calls memory_object_destory(), which will in turn end up
+ *             calling ubc_unmap() to release any vnode references that were
+ *             established via ubc_map().
+ *
+ *             IMPORTANT: This is an internal use function that is used
+ *             exclusively by the internal use function vclean().
  */
  */
-__private_extern__ int
-ubc_destroy_named(
-       struct vnode    *vp)
+__private_extern__ void
+ubc_destroy_named(vnode_t vp)
 {
        memory_object_control_t control;
 {
        memory_object_control_t control;
-       struct proc *p;
        struct ubc_info *uip;
        kern_return_t kret;
 
        struct ubc_info *uip;
        kern_return_t kret;
 
-       /*
-        * We may already have had the object terminated
-        * and the ubcinfo released as a side effect of
-        * some earlier processing.  If so, pretend we did
-        * it, because it probably was a result of our
-        * efforts.
-        */
-       if (!UBCINFOEXISTS(vp))
-               return (1);
-
-       uip = vp->v_ubcinfo;
-
-       /* can not destroy held vnodes */
-       if (uip->ui_refcount > 1)
-               return (0);
+       if (UBCINFOEXISTS(vp)) {
+               uip = vp->v_ubcinfo;
 
 
-       /* 
-        * Terminate the memory object.
-        * memory_object_destroy() will result in
-        * vnode_pager_no_senders(). 
-        * That will release the pager reference
-        * and the vnode will move to the free list.
-        */
-       control = ubc_getobject(vp, UBC_HOLDOBJECT);
-       if (control != MEMORY_OBJECT_CONTROL_NULL) {
-
-               if (ISSET(vp->v_flag, VTERMINATE))
-                       panic("ubc_destroy_named: already teminating");
-               SET(vp->v_flag, VTERMINATE);
-
-               kret = memory_object_destroy(control, 0);
-               if (kret != KERN_SUCCESS)
-                       return (0);
-
-               /* 
-                * memory_object_destroy() is asynchronous
-                * with respect to vnode_pager_no_senders().
-                * wait for vnode_pager_no_senders() to clear
-                * VTERMINATE
-                */
-               while (ISSET(vp->v_flag, VTERMINATE)) {
-                       SET(vp->v_flag, VTERMWANT);
-                       (void)tsleep((caddr_t)&vp->v_ubcinfo,
-                                                PINOD, "ubc_destroy_named", 0);
+               /* Terminate the memory object  */
+               control = ubc_getobject(vp, UBC_HOLDOBJECT);
+               if (control != MEMORY_OBJECT_CONTROL_NULL) {
+                       kret = memory_object_destroy(control, 0);
+                       if (kret != KERN_SUCCESS)
+                               panic("ubc_destroy_named: memory_object_destroy failed");
                }
        }
                }
        }
-       return (1);
 }
 
 
 /*
 }
 
 
 /*
- * Invalidate a range in the memory object that backs this
- * vnode. The offset is truncated to the page boundary and the
- * size is adjusted to include the last page in the range.
+ * ubc_isinuse
+ *
+ * Determine whether or not a vnode is currently in use by ubc at a level in
+ * excess of the requested busycount
+ *
+ * Parameters: vp                      The vnode to check
+ *             busycount               The threshold busy count, used to bias
+ *                                     the count usually already held by the
+ *                                     caller to avoid races
+ *
+ * Returns:    1                       The vnode is in use over the threshold
+ *             0                       The vnode is not in use over the
+ *                                     threshold
+ *
+ * Notes:      Because the vnode is only held locked while actually asking
+ *             the use count, this function only represents a snapshot of the
+ *             current state of the vnode.  If more accurate information is
+ *             required, an additional busycount should be held by the caller
+ *             and a non-zero busycount used.
+ *
+ *             If there is no ubc_info associated with the vnode, this
+ *             function will report that the vnode is not in use by ubc.
  */
 int
  */
 int
-ubc_invalidate(struct vnode *vp, off_t offset, size_t size)
+ubc_isinuse(struct vnode *vp, int busycount)
 {
 {
-       struct ubc_info *uip;
-       memory_object_control_t control;
-       kern_return_t kret;
-       off_t toff;
-       size_t tsize;
-
-       if (UBCINVALID(vp))
+       if ( !UBCINFOEXISTS(vp))
                return (0);
                return (0);
-
-       if (!UBCINFOEXISTS(vp))
-               return (0);
-
-       toff = trunc_page_64(offset);
-       tsize = (size_t)(round_page_64(offset+size) - toff);
-       uip = vp->v_ubcinfo;
-       control = uip->ui_control;
-       assert(control);
-
-       /* invalidate pages in the range requested */
-       kret = memory_object_lock_request(control,
-                               (memory_object_offset_t)toff,
-                               (memory_object_size_t)tsize,
-                               MEMORY_OBJECT_RETURN_NONE,
-                               (MEMORY_OBJECT_DATA_NO_CHANGE| MEMORY_OBJECT_DATA_FLUSH),
-                               VM_PROT_NO_CHANGE);
-       if (kret != KERN_SUCCESS)
-               printf("ubc_invalidate: invalidate failed (error = %d)\n", kret);
-
-       return ((kret == KERN_SUCCESS) ? 1 : 0);
+       return(ubc_isinuse_locked(vp, busycount, 0));
 }
 
 }
 
+
 /*
 /*
- * Find out whether a vnode is in use by UBC
- * Returns 1 if file is in use by UBC, 0 if not
+ * ubc_isinuse_locked
+ *
+ * Determine whether or not a vnode is currently in use by ubc at a level in
+ * excess of the requested busycount
+ *
+ * Parameters: vp                      The vnode to check
+ *             busycount               The threshold busy count, used to bias
+ *                                     the count usually already held by the
+ *                                     caller to avoid races
+ *             locked                  True if the vnode is already locked by
+ *                                     the caller
+ *
+ * Returns:    1                       The vnode is in use over the threshold
+ *             0                       The vnode is not in use over the
+ *                                     threshold
+ *
+ * Notes:      If the vnode is not locked on entry, it is locked while
+ *             actually asking the use count.  If this is the case, this
+ *             function only represents a snapshot of the current state of
+ *             the vnode.  If more accurate information is required, the
+ *             vnode lock should be held by the caller, otherwise an
+ *             additional busycount should be held by the caller and a
+ *             non-zero busycount used.
+ *
+ *             If there is no ubc_info associated with the vnode, this
+ *             function will report that the vnode is not in use by ubc.
  */
 int
  */
 int
-ubc_isinuse(struct vnode *vp, int busycount)
+ubc_isinuse_locked(struct vnode *vp, int busycount, int locked)
 {
 {
-       if (!UBCINFOEXISTS(vp))
-               return (0);
+       int retval = 0;
 
 
-       if (busycount == 0) {
-               printf("ubc_isinuse: called without a valid reference"
-                   ": v_tag = %d\v", vp->v_tag);
-               vprint("ubc_isinuse", vp);
-               return (0);
-       }
 
 
-       if (vp->v_usecount > busycount+1)
-               return (1);
+       if (!locked)
+               vnode_lock(vp);
 
 
-       if ((vp->v_usecount == busycount+1)
-               && (vp->v_ubcinfo->ui_mapped == 1))
-               return (1);
-       else
-               return (0);
+       if ((vp->v_usecount - vp->v_kusecount) > busycount)
+               retval = 1;
+
+       if (!locked)
+               vnode_unlock(vp);
+       return (retval);
 }
 
 }
 
+
 /*
 /*
- * The backdoor routine to clear the ui_mapped.
- * MUST only be called by the VM
+ * ubc_unmap
  *
  *
- * Note that this routine is not called under funnel. There are numerous
- * things about the calling sequence that make this work on SMP.
- * Any code change in those paths can break this.
+ * Reverse the effects of a ubc_map() call for a given vnode
  *
  *
+ * Parameters: vp                      vnode to unmap from ubc
+ *
+ * Returns:    (void)
+ *
+ * Notes:      This is an internal use function used by vnode_pager_unmap().
+ *             It will attempt to obtain a reference on the supplied vnode,
+ *             and if it can do so, and there is an associated ubc_info, and
+ *             the flags indicate that it was mapped via ubc_map(), then the
+ *             flag is cleared, the mapping removed, and the reference taken
+ *             by ubc_map() is released.
+ *
+ *             IMPORTANT: This MUST only be called by the VM
+ *             to prevent race conditions.
  */
 __private_extern__ void
 ubc_unmap(struct vnode *vp)
 {
        struct ubc_info *uip;
  */
 __private_extern__ void
 ubc_unmap(struct vnode *vp)
 {
        struct ubc_info *uip;
-       boolean_t       funnel_state;
-       if (UBCINVALID(vp))
-               return;
+       int     need_rele = 0;
+       int     need_wakeup = 0;
+       
+       if (vnode_getwithref(vp))
+               return;
 
 
-       if (!UBCINFOEXISTS(vp))
-               return;
+       if (UBCINFOEXISTS(vp)) {
+               vnode_lock(vp);
+               uip = vp->v_ubcinfo;
 
 
-       ubc_lock(vp);
-       uip = vp->v_ubcinfo;
-       uip->ui_mapped = 0;
-       if ((uip->ui_refcount > 1) || !ISSET(uip->ui_flags, UI_DONTCACHE)) {
-               ubc_unlock(vp);
-               return;
-       }
-       ubc_unlock(vp);
+               while (ISSET(uip->ui_flags, UI_MAPBUSY)) {
+                       SET(uip->ui_flags, UI_MAPWAITING);
+                       (void) msleep(&uip->ui_flags, &vp->v_lock,
+                                     PRIBIO, "ubc_unmap", NULL);
+               }
+               SET(uip->ui_flags, UI_MAPBUSY);
+
+               if (ISSET(uip->ui_flags, UI_ISMAPPED)) {
+               CLR(uip->ui_flags, UI_ISMAPPED);
+                       need_rele = 1;
+               }
+               vnode_unlock(vp);
+       
+               if (need_rele) {
+                       (void) VNOP_MNOMAP(vp, vfs_context_current());
+                       vnode_rele(vp);
+               }
+
+               vnode_lock_spin(vp);
+
+               CLR(uip->ui_flags, UI_MAPBUSY);
+               if (ISSET(uip->ui_flags, UI_MAPWAITING)) {
+                       CLR(uip->ui_flags, UI_MAPWAITING);
+                       need_wakeup = 1;
+               }
+               vnode_unlock(vp);
+
+               if (need_wakeup)
+                   wakeup(&uip->ui_flags);
 
 
-       funnel_state = thread_funnel_set(kernel_flock, TRUE);
-       (void) ubc_release_named(vp);
-       (void) thread_funnel_set(kernel_flock, funnel_state);
+       }
+       /*
+        * the drop of the vnode ref will cleanup
+        */
+       vnode_put(vp);
 }
 
 }
 
+
+/*
+ * ubc_page_op
+ *
+ * Manipulate individual page state for a vnode with an associated ubc_info
+ * with an associated memory object control.
+ *
+ * Parameters: vp                      The vnode backing the page
+ *             f_offset                A file offset interior to the page
+ *             ops                     The operations to perform, as a bitmap
+ *                                     (see below for more information)
+ *             phys_entryp             The address of a ppnum_t; may be NULL
+ *                                     to ignore
+ *             flagsp                  A pointer to an int to contain flags;
+ *                                     may be NULL to ignore
+ *
+ * Returns:    KERN_SUCCESS            Success
+ *             KERN_INVALID_ARGUMENT   If the memory object control has no VM
+ *                                     object associated
+ *             KERN_INVALID_OBJECT     If UPL_POP_PHYSICAL and the object is
+ *                                     not physically contiguous
+ *             KERN_INVALID_OBJECT     If !UPL_POP_PHYSICAL and the object is
+ *                                     physically contiguous
+ *             KERN_FAILURE            If the page cannot be looked up
+ *
+ * Implicit Returns:
+ *             *phys_entryp (modified) If phys_entryp is non-NULL and
+ *                                     UPL_POP_PHYSICAL
+ *             *flagsp (modified)      If flagsp is non-NULL and there was
+ *                                     !UPL_POP_PHYSICAL and a KERN_SUCCESS
+ *
+ * Notes:      For object boundaries, it is considerably more efficient to
+ *             ensure that f_offset is in fact on a page boundary, as this
+ *             will avoid internal use of the hash table to identify the
+ *             page, and would therefore skip a number of early optimizations.
+ *             Since this is a page operation anyway, the caller should try
+ *             to pass only a page aligned offset because of this.
+ *
+ *             *flagsp may be modified even if this function fails.  If it is
+ *             modified, it will contain the condition of the page before the
+ *             requested operation was attempted; these will only include the
+ *             bitmap flags, and not the PL_POP_PHYSICAL, UPL_POP_DUMP,
+ *             UPL_POP_SET, or UPL_POP_CLR bits.
+ *
+ *             The flags field may contain a specific operation, such as
+ *             UPL_POP_PHYSICAL or UPL_POP_DUMP:
+ *
+ *             o       UPL_POP_PHYSICAL        Fail if not contiguous; if
+ *                                             *phys_entryp and successful, set
+ *                                             *phys_entryp
+ *             o       UPL_POP_DUMP            Dump the specified page
+ *
+ *             Otherwise, it is treated as a bitmap of one or more page
+ *             operations to perform on the final memory object; allowable
+ *             bit values are:
+ *
+ *             o       UPL_POP_DIRTY           The page is dirty
+ *             o       UPL_POP_PAGEOUT         The page is paged out
+ *             o       UPL_POP_PRECIOUS        The page is precious
+ *             o       UPL_POP_ABSENT          The page is absent
+ *             o       UPL_POP_BUSY            The page is busy
+ *
+ *             If the page status is only being queried and not modified, then
+ *             not other bits should be specified.  However, if it is being
+ *             modified, exactly ONE of the following bits should be set:
+ *
+ *             o       UPL_POP_SET             Set the current bitmap bits
+ *             o       UPL_POP_CLR             Clear the current bitmap bits
+ *
+ *             Thus to effect a combination of setting an clearing, it may be
+ *             necessary to call this function twice.  If this is done, the
+ *             set should be used before the clear, since clearing may trigger
+ *             a wakeup on the destination page, and if the page is backed by
+ *             an encrypted swap file, setting will trigger the decryption
+ *             needed before the wakeup occurs.
+ */
 kern_return_t
 ubc_page_op(
        struct vnode    *vp,
 kern_return_t
 ubc_page_op(
        struct vnode    *vp,
@@ -1208,22 +1611,66 @@ ubc_page_op(
                                      phys_entryp,
                                      flagsp));
 }
                                      phys_entryp,
                                      flagsp));
 }
-                                     
-__private_extern__ kern_return_t
-ubc_page_op_with_control(
-       memory_object_control_t  control,
-       off_t                    f_offset,
-       int                      ops,
-       ppnum_t                  *phys_entryp,
-       int                      *flagsp)
-{
-       return (memory_object_page_op(control,
-                                     (memory_object_offset_t)f_offset,
-                                     ops,
-                                     phys_entryp,
-                                     flagsp));
-}
-                                     
+
+
+/*
+ * ubc_range_op
+ *
+ * Manipulate page state for a range of memory for a vnode with an associated
+ * ubc_info with an associated memory object control, when page level state is
+ * not required to be returned from the call (i.e. there are no phys_entryp or
+ * flagsp parameters to this call, and it takes a range which may contain
+ * multiple pages, rather than an offset interior to a single page).
+ *
+ * Parameters: vp                      The vnode backing the page
+ *             f_offset_beg            A file offset interior to the start page
+ *             f_offset_end            A file offset interior to the end page
+ *             ops                     The operations to perform, as a bitmap
+ *                                     (see below for more information)
+ *             range                   The address of an int; may be NULL to
+ *                                     ignore
+ *
+ * Returns:    KERN_SUCCESS            Success
+ *             KERN_INVALID_ARGUMENT   If the memory object control has no VM
+ *                                     object associated
+ *             KERN_INVALID_OBJECT     If the object is physically contiguous
+ *
+ * Implicit Returns:
+ *             *range (modified)       If range is non-NULL, its contents will
+ *                                     be modified to contain the number of
+ *                                     bytes successfully operated upon.
+ *
+ * Notes:      IMPORTANT: This function cannot be used on a range that
+ *             consists of physically contiguous pages.
+ *
+ *             For object boundaries, it is considerably more efficient to
+ *             ensure that f_offset_beg and f_offset_end are in fact on page
+ *             boundaries, as this will avoid internal use of the hash table
+ *             to identify the page, and would therefore skip a number of
+ *             early optimizations.  Since this is an operation on a set of
+ *             pages anyway, the caller should try to pass only a page aligned
+ *             offsets because of this.
+ *
+ *             *range will be modified only if this function succeeds.
+ *
+ *             The flags field MUST contain a specific operation; allowable
+ *             values are:
+ *
+ *             o       UPL_ROP_ABSENT  Returns the extent of the range
+ *                                     presented which is absent, starting
+ *                                     with the start address presented
+ *
+ *             o       UPL_ROP_PRESENT Returns the extent of the range
+ *                                     presented which is present (resident),
+ *                                     starting with the start address
+ *                                     presented
+ *             o       UPL_ROP_DUMP    Dump the pages which are found in the
+ *                                     target object for the target range.
+ *
+ *             IMPORTANT: For UPL_ROP_ABSENT and UPL_ROP_PRESENT; if there are
+ *             multiple regions in the range, only the first matching region
+ *             is returned.
+ */
 kern_return_t
 ubc_range_op(
        struct vnode    *vp,
 kern_return_t
 ubc_range_op(
        struct vnode    *vp,
@@ -1244,20 +1691,58 @@ ubc_range_op(
                                      ops,
                                      range));
 }
                                      ops,
                                      range));
 }
-                                     
+
+
+/*
+ * ubc_create_upl
+ *
+ * Given a vnode, cause the population of a portion of the vm_object; based on
+ * the nature of the request, the pages returned may contain valid data, or
+ * they may be uninitialized.
+ *
+ * Parameters: vp                      The vnode from which to create the upl
+ *             f_offset                The start offset into the backing store
+ *                                     represented by the vnode
+ *             bufsize                 The size of the upl to create
+ *             uplp                    Pointer to the upl_t to receive the
+ *                                     created upl; MUST NOT be NULL
+ *             plp                     Pointer to receive the internal page
+ *                                     list for the created upl; MAY be NULL
+ *                                     to ignore
+ *
+ * Returns:    KERN_SUCCESS            The requested upl has been created
+ *             KERN_INVALID_ARGUMENT   The bufsize argument is not an even
+ *                                     multiple of the page size
+ *             KERN_INVALID_ARGUMENT   There is no ubc_info associated with
+ *                                     the vnode, or there is no memory object
+ *                                     control associated with the ubc_info 
+ *     memory_object_upl_request:KERN_INVALID_VALUE
+ *                                     The supplied upl_flags argument is
+ *                                     invalid
+ * Implicit Returns:
+ *             *uplp (modified)        
+ *             *plp (modified)         If non-NULL, the value of *plp will be
+ *                                     modified to point to the internal page
+ *                                     list; this modification may occur even
+ *                                     if this function is unsuccessful, in
+ *                                     which case the contents may be invalid
+ *
+ * Note:       If successful, the returned *uplp MUST subsequently be freed
+ *             via a call to ubc_upl_commit(), ubc_upl_commit_range(),
+ *             ubc_upl_abort(), or ubc_upl_abort_range().
+ */
 kern_return_t
 ubc_create_upl(
        struct vnode    *vp,
 kern_return_t
 ubc_create_upl(
        struct vnode    *vp,
-       off_t                   f_offset,
-       long                    bufsize,
-       upl_t                   *uplp,
+       off_t           f_offset,
+       long            bufsize,
+       upl_t           *uplp,
        upl_page_info_t **plp,
        upl_page_info_t **plp,
-       int                             uplflags)
+       int             uplflags)
 {
        memory_object_control_t         control;
 {
        memory_object_control_t         control;
-       int                             count;
+       mach_msg_type_number_t          count;
        int                             ubcflags;
        int                             ubcflags;
-       off_t                           file_offset;
        kern_return_t                   kr;
        
        if (bufsize & 0xfff)
        kern_return_t                   kr;
        
        if (bufsize & 0xfff)
@@ -1279,14 +1764,47 @@ ubc_create_upl(
        } else
                uplflags |= (UPL_NO_SYNC|UPL_CLEAN_IN_PLACE|UPL_SET_INTERNAL);
        count = 0;
        } else
                uplflags |= (UPL_NO_SYNC|UPL_CLEAN_IN_PLACE|UPL_SET_INTERNAL);
        count = 0;
-       kr = memory_object_upl_request(control, f_offset, bufsize,
-                                                                  uplp, NULL, &count, uplflags);
+
+       kr = memory_object_upl_request(control, f_offset, bufsize, uplp, NULL, &count, uplflags);
        if (plp != NULL)
                        *plp = UPL_GET_INTERNAL_PAGE_LIST(*uplp);
        return kr;
 }
        if (plp != NULL)
                        *plp = UPL_GET_INTERNAL_PAGE_LIST(*uplp);
        return kr;
 }
-                                     
+               
+                                                             
+/*
+ * ubc_upl_maxbufsize
+ *
+ * Return the maximum bufsize ubc_create_upl( ) will take.
+ *
+ * Parameters: none
+ *
+ * Returns:    maximum size buffer (in bytes) ubc_create_upl( ) will take.
+ */
+upl_size_t 
+ubc_upl_maxbufsize(
+       void)
+{
+       return(MAX_UPL_SIZE * PAGE_SIZE);
+}
 
 
+/*
+ * ubc_upl_map
+ *
+ * Map the page list assocated with the supplied upl into the kernel virtual
+ * address space at the virtual address indicated by the dst_addr argument;
+ * the entire upl is mapped
+ *
+ * Parameters: upl                     The upl to map
+ *             dst_addr                The address at which to map the upl
+ *
+ * Returns:    KERN_SUCCESS            The upl has been mapped
+ *             KERN_INVALID_ARGUMENT   The upl is UPL_NULL
+ *             KERN_FAILURE            The upl is already mapped
+ *     vm_map_enter:KERN_INVALID_ARGUMENT
+ *                                     A failure code from vm_map_enter() due
+ *                                     to an invalid argument
+ */
 kern_return_t
 ubc_upl_map(
        upl_t           upl,
 kern_return_t
 ubc_upl_map(
        upl_t           upl,
@@ -1296,6 +1814,18 @@ ubc_upl_map(
 }
 
 
 }
 
 
+/*
+ * ubc_upl_unmap
+ *
+ * Unmap the page list assocated with the supplied upl from the kernel virtual
+ * address space; the entire upl is unmapped.
+ *
+ * Parameters: upl                     The upl to unmap
+ *
+ * Returns:    KERN_SUCCESS            The upl has been unmapped
+ *             KERN_FAILURE            The upl is not currently mapped
+ *             KERN_INVALID_ARGUMENT   If the upl is UPL_NULL
+ */
 kern_return_t
 ubc_upl_unmap(
        upl_t   upl)
 kern_return_t
 ubc_upl_unmap(
        upl_t   upl)
@@ -1303,6 +1833,30 @@ ubc_upl_unmap(
        return(vm_upl_unmap(kernel_map, upl));
 }
 
        return(vm_upl_unmap(kernel_map, upl));
 }
 
+
+/*
+ * ubc_upl_commit
+ *
+ * Commit the contents of the upl to the backing store
+ *
+ * Parameters: upl                     The upl to commit
+ *
+ * Returns:    KERN_SUCCESS            The upl has been committed
+ *             KERN_INVALID_ARGUMENT   The supplied upl was UPL_NULL
+ *             KERN_FAILURE            The supplied upl does not represent
+ *                                     device memory, and the offset plus the
+ *                                     size would exceed the actual size of
+ *                                     the upl
+ *
+ * Notes:      In practice, the only return value for this function should be
+ *             KERN_SUCCESS, unless there has been data structure corruption;
+ *             since the upl is deallocated regardless of success or failure,
+ *             there's really nothing to do about this other than panic.
+ *
+ *             IMPORTANT: Use of this function should not be mixed with use of
+ *             ubc_upl_commit_range(), due to the unconditional deallocation
+ *             by this function.
+ */
 kern_return_t
 ubc_upl_commit(
        upl_t                   upl)
 kern_return_t
 ubc_upl_commit(
        upl_t                   upl)
@@ -1311,12 +1865,64 @@ ubc_upl_commit(
        kern_return_t   kr;
 
        pl = UPL_GET_INTERNAL_PAGE_LIST(upl);
        kern_return_t   kr;
 
        pl = UPL_GET_INTERNAL_PAGE_LIST(upl);
-       kr = upl_commit(upl, pl, MAX_UPL_TRANSFER);
+       kr = upl_commit(upl, pl, MAX_UPL_SIZE);
        upl_deallocate(upl);
        return kr;
 }
 
 
        upl_deallocate(upl);
        return kr;
 }
 
 
+/*
+ * ubc_upl_commit
+ *
+ * Commit the contents of the specified range of the upl to the backing store
+ *
+ * Parameters: upl                     The upl to commit
+ *             offset                  The offset into the upl
+ *             size                    The size of the region to be committed,
+ *                                     starting at the specified offset
+ *             flags                   commit type (see below)
+ *
+ * Returns:    KERN_SUCCESS            The range has been committed
+ *             KERN_INVALID_ARGUMENT   The supplied upl was UPL_NULL
+ *             KERN_FAILURE            The supplied upl does not represent
+ *                                     device memory, and the offset plus the
+ *                                     size would exceed the actual size of
+ *                                     the upl
+ *
+ * Notes:      IMPORTANT: If the commit is successful, and the object is now
+ *             empty, the upl will be deallocated.  Since the caller cannot
+ *             check that this is the case, the UPL_COMMIT_FREE_ON_EMPTY flag
+ *             should generally only be used when the offset is 0 and the size
+ *             is equal to the upl size.
+ *
+ *             The flags argument is a bitmap of flags on the rage of pages in
+ *             the upl to be committed; allowable flags are:
+ *
+ *             o       UPL_COMMIT_FREE_ON_EMPTY        Free the upl when it is
+ *                                                     both empty and has been
+ *                                                     successfully committed
+ *             o       UPL_COMMIT_CLEAR_DIRTY          Clear each pages dirty
+ *                                                     bit; will prevent a
+ *                                                     later pageout
+ *             o       UPL_COMMIT_SET_DIRTY            Set each pages dirty
+ *                                                     bit; will cause a later
+ *                                                     pageout
+ *             o       UPL_COMMIT_INACTIVATE           Clear each pages
+ *                                                     reference bit; the page
+ *                                                     will not be accessed
+ *             o       UPL_COMMIT_ALLOW_ACCESS         Unbusy each page; pages
+ *                                                     become busy when an
+ *                                                     IOMemoryDescriptor is
+ *                                                     mapped or redirected,
+ *                                                     and we have to wait for
+ *                                                     an IOKit driver
+ *
+ *             The flag UPL_COMMIT_NOTIFY_EMPTY is used internally, and should
+ *             not be specified by the caller.
+ *
+ *             The UPL_COMMIT_CLEAR_DIRTY and UPL_COMMIT_SET_DIRTY flags are
+ *             mutually exclusive, and should not be combined.
+ */
 kern_return_t
 ubc_upl_commit_range(
        upl_t                   upl,
 kern_return_t
 ubc_upl_commit_range(
        upl_t                   upl,
@@ -1331,17 +1937,67 @@ ubc_upl_commit_range(
        if (flags & UPL_COMMIT_FREE_ON_EMPTY)
                flags |= UPL_COMMIT_NOTIFY_EMPTY;
 
        if (flags & UPL_COMMIT_FREE_ON_EMPTY)
                flags |= UPL_COMMIT_NOTIFY_EMPTY;
 
+       if (flags & UPL_COMMIT_KERNEL_ONLY_FLAGS) {
+               return KERN_INVALID_ARGUMENT;
+       }
+
        pl = UPL_GET_INTERNAL_PAGE_LIST(upl);
 
        kr = upl_commit_range(upl, offset, size, flags,
        pl = UPL_GET_INTERNAL_PAGE_LIST(upl);
 
        kr = upl_commit_range(upl, offset, size, flags,
-                                                 pl, MAX_UPL_TRANSFER, &empty);
+                                                 pl, MAX_UPL_SIZE, &empty);
 
        if((flags & UPL_COMMIT_FREE_ON_EMPTY) && empty)
                upl_deallocate(upl);
 
        return kr;
 }
 
        if((flags & UPL_COMMIT_FREE_ON_EMPTY) && empty)
                upl_deallocate(upl);
 
        return kr;
 }
-       
+
+
+/*
+ * ubc_upl_abort_range
+ *
+ * Abort the contents of the specified range of the specified upl
+ *
+ * Parameters: upl                     The upl to abort
+ *             offset                  The offset into the upl
+ *             size                    The size of the region to be aborted,
+ *                                     starting at the specified offset
+ *             abort_flags             abort type (see below)
+ *
+ * Returns:    KERN_SUCCESS            The range has been aborted
+ *             KERN_INVALID_ARGUMENT   The supplied upl was UPL_NULL
+ *             KERN_FAILURE            The supplied upl does not represent
+ *                                     device memory, and the offset plus the
+ *                                     size would exceed the actual size of
+ *                                     the upl
+ *
+ * Notes:      IMPORTANT: If the abort is successful, and the object is now
+ *             empty, the upl will be deallocated.  Since the caller cannot
+ *             check that this is the case, the UPL_ABORT_FREE_ON_EMPTY flag
+ *             should generally only be used when the offset is 0 and the size
+ *             is equal to the upl size.
+ *
+ *             The abort_flags argument is a bitmap of flags on the range of
+ *             pages in the upl to be aborted; allowable flags are:
+ *
+ *             o       UPL_ABORT_FREE_ON_EMPTY Free the upl when it is both
+ *                                             empty and has been successfully
+ *                                             aborted
+ *             o       UPL_ABORT_RESTART       The operation must be restarted
+ *             o       UPL_ABORT_UNAVAILABLE   The pages are unavailable
+ *             o       UPL_ABORT_ERROR         An I/O error occurred
+ *             o       UPL_ABORT_DUMP_PAGES    Just free the pages
+ *             o       UPL_ABORT_NOTIFY_EMPTY  RESERVED
+ *             o       UPL_ABORT_ALLOW_ACCESS  RESERVED
+ *
+ *             The UPL_ABORT_NOTIFY_EMPTY is an internal use flag and should
+ *             not be specified by the caller.  It is intended to fulfill the
+ *             same role as UPL_COMMIT_NOTIFY_EMPTY does in the function
+ *             ubc_upl_commit_range(), but is never referenced internally.
+ *
+ *             The UPL_ABORT_ALLOW_ACCESS is defined, but neither set nor
+ *             referenced; do not use it.
+ */
 kern_return_t
 ubc_upl_abort_range(
        upl_t                   upl,
 kern_return_t
 ubc_upl_abort_range(
        upl_t                   upl,
@@ -1363,6 +2019,49 @@ ubc_upl_abort_range(
        return kr;
 }
 
        return kr;
 }
 
+
+/*
+ * ubc_upl_abort
+ *
+ * Abort the contents of the specified upl
+ *
+ * Parameters: upl                     The upl to abort
+ *             abort_type              abort type (see below)
+ *
+ * Returns:    KERN_SUCCESS            The range has been aborted
+ *             KERN_INVALID_ARGUMENT   The supplied upl was UPL_NULL
+ *             KERN_FAILURE            The supplied upl does not represent
+ *                                     device memory, and the offset plus the
+ *                                     size would exceed the actual size of
+ *                                     the upl
+ *
+ * Notes:      IMPORTANT: If the abort is successful, and the object is now
+ *             empty, the upl will be deallocated.  Since the caller cannot
+ *             check that this is the case, the UPL_ABORT_FREE_ON_EMPTY flag
+ *             should generally only be used when the offset is 0 and the size
+ *             is equal to the upl size.
+ *
+ *             The abort_type is a bitmap of flags on the range of
+ *             pages in the upl to be aborted; allowable flags are:
+ *
+ *             o       UPL_ABORT_FREE_ON_EMPTY Free the upl when it is both
+ *                                             empty and has been successfully
+ *                                             aborted
+ *             o       UPL_ABORT_RESTART       The operation must be restarted
+ *             o       UPL_ABORT_UNAVAILABLE   The pages are unavailable
+ *             o       UPL_ABORT_ERROR         An I/O error occurred
+ *             o       UPL_ABORT_DUMP_PAGES    Just free the pages
+ *             o       UPL_ABORT_NOTIFY_EMPTY  RESERVED
+ *             o       UPL_ABORT_ALLOW_ACCESS  RESERVED
+ *
+ *             The UPL_ABORT_NOTIFY_EMPTY is an internal use flag and should
+ *             not be specified by the caller.  It is intended to fulfill the
+ *             same role as UPL_COMMIT_NOTIFY_EMPTY does in the function
+ *             ubc_upl_commit_range(), but is never referenced internally.
+ *
+ *             The UPL_ABORT_ALLOW_ACCESS is defined, but neither set nor
+ *             referenced; do not use it.
+ */
 kern_return_t
 ubc_upl_abort(
        upl_t                   upl,
 kern_return_t
 ubc_upl_abort(
        upl_t                   upl,
@@ -1375,9 +2074,627 @@ ubc_upl_abort(
        return kr;
 }
 
        return kr;
 }
 
+
+/*
+ * ubc_upl_pageinfo
+ *
+ *  Retrieve the internal page list for the specified upl
+ *
+ * Parameters: upl                     The upl to obtain the page list from
+ *
+ * Returns:    !NULL                   The (upl_page_info_t *) for the page
+ *                                     list internal to the upl
+ *             NULL                    Error/no page list associated
+ *
+ * Notes:      IMPORTANT: The function is only valid on internal objects
+ *             where the list request was made with the UPL_INTERNAL flag.
+ *
+ *             This function is a utility helper function, since some callers
+ *             may not have direct access to the header defining the macro,
+ *             due to abstraction layering constraints.
+ */
 upl_page_info_t *
 ubc_upl_pageinfo(
        upl_t                   upl)
 {             
        return (UPL_GET_INTERNAL_PAGE_LIST(upl));
 }
 upl_page_info_t *
 ubc_upl_pageinfo(
        upl_t                   upl)
 {             
        return (UPL_GET_INTERNAL_PAGE_LIST(upl));
 }
+
+
+int 
+UBCINFOEXISTS(struct vnode * vp)
+{
+        return((vp) && ((vp)->v_type == VREG) && ((vp)->v_ubcinfo != UBC_INFO_NULL));
+}
+
+
+/*
+ * CODE SIGNING
+ */
+#define CS_BLOB_PAGEABLE 0
+static volatile SInt32 cs_blob_size = 0;
+static volatile SInt32 cs_blob_count = 0;
+static SInt32 cs_blob_size_peak = 0;
+static UInt32 cs_blob_size_max = 0;
+static SInt32 cs_blob_count_peak = 0;
+extern int cs_debug;
+
+int cs_validation = 1;
+
+SYSCTL_INT(_vm, OID_AUTO, cs_validation, CTLFLAG_RW, &cs_validation, 0, "Do validate code signatures");
+SYSCTL_INT(_vm, OID_AUTO, cs_blob_count, CTLFLAG_RD, &cs_blob_count, 0, "Current number of code signature blobs");
+SYSCTL_INT(_vm, OID_AUTO, cs_blob_size, CTLFLAG_RD, &cs_blob_size, 0, "Current size of all code signature blobs");
+SYSCTL_INT(_vm, OID_AUTO, cs_blob_count_peak, CTLFLAG_RD, &cs_blob_count_peak, 0, "Peak number of code signature blobs");
+SYSCTL_INT(_vm, OID_AUTO, cs_blob_size_peak, CTLFLAG_RD, &cs_blob_size_peak, 0, "Peak size of code signature blobs");
+SYSCTL_INT(_vm, OID_AUTO, cs_blob_size_max, CTLFLAG_RD, &cs_blob_size_max, 0, "Size of biggest code signature blob");
+
+kern_return_t
+ubc_cs_blob_allocate(
+       vm_offset_t     *blob_addr_p,
+       vm_size_t       *blob_size_p)
+{
+       kern_return_t   kr;
+
+#if CS_BLOB_PAGEABLE
+       *blob_size_p = round_page(*blob_size_p);
+       kr = kmem_alloc(kernel_map, blob_addr_p, *blob_size_p);
+#else  /* CS_BLOB_PAGEABLE */
+       *blob_addr_p = (vm_offset_t) kalloc(*blob_size_p);
+       if (*blob_addr_p == 0) {
+               kr = KERN_NO_SPACE;
+       } else {
+               kr = KERN_SUCCESS;
+       }
+#endif /* CS_BLOB_PAGEABLE */
+       return kr;
+}
+
+void
+ubc_cs_blob_deallocate(
+       vm_offset_t     blob_addr,
+       vm_size_t       blob_size)
+{
+#if CS_BLOB_PAGEABLE
+       kmem_free(kernel_map, blob_addr, blob_size);
+#else  /* CS_BLOB_PAGEABLE */
+       kfree((void *) blob_addr, blob_size);
+#endif /* CS_BLOB_PAGEABLE */
+}
+       
+int
+ubc_cs_blob_add(
+       struct vnode    *vp,
+       cpu_type_t      cputype,
+       off_t           base_offset,
+       vm_address_t    addr,
+       vm_size_t       size)
+{
+       kern_return_t           kr;
+       struct ubc_info         *uip;
+       struct cs_blob          *blob, *oblob;
+       int                     error;
+       ipc_port_t              blob_handle;
+       memory_object_size_t    blob_size;
+       const CS_CodeDirectory *cd;
+       off_t                   blob_start_offset, blob_end_offset;
+       SHA1_CTX                sha1ctxt;
+
+       blob_handle = IPC_PORT_NULL;
+
+       blob = (struct cs_blob *) kalloc(sizeof (struct cs_blob));
+       if (blob == NULL) {
+               return ENOMEM;
+       }
+
+#if CS_BLOB_PAGEABLE
+       /* get a memory entry on the blob */
+       blob_size = (memory_object_size_t) size;
+       kr = mach_make_memory_entry_64(kernel_map,
+                                      &blob_size,
+                                      addr,
+                                      VM_PROT_READ,
+                                      &blob_handle,
+                                      IPC_PORT_NULL);
+       if (kr != KERN_SUCCESS) {
+               error = ENOMEM;
+               goto out;
+       }
+       if (memory_object_round_page(blob_size) !=
+           (memory_object_size_t) round_page(size)) {
+               printf("ubc_cs_blob_add: size mismatch 0x%llx 0x%x !?\n",
+                      blob_size, size);
+               panic("XXX FBDP size mismatch 0x%llx 0x%x\n", blob_size, size);
+               error = EINVAL;
+               goto out;
+       }
+#else
+       blob_size = (memory_object_size_t) size;
+       blob_handle = IPC_PORT_NULL;
+#endif
+
+       /* fill in the new blob */
+       blob->csb_cpu_type = cputype;
+       blob->csb_base_offset = base_offset;
+       blob->csb_mem_size = size;
+       blob->csb_mem_offset = 0;
+       blob->csb_mem_handle = blob_handle;
+       blob->csb_mem_kaddr = addr;
+       
+       /*
+        * Validate the blob's contents
+        */
+       cd = findCodeDirectory(
+               (const CS_SuperBlob *) addr, 
+               (char *) addr, 
+               (char *) addr + blob->csb_mem_size);
+       if (cd == NULL) {
+               /* no code directory => useless blob ! */
+               blob->csb_flags = 0;
+               blob->csb_start_offset = 0;
+               blob->csb_end_offset = 0;
+       } else {
+               unsigned char *sha1_base;
+               int sha1_size;
+
+               blob->csb_flags = ntohl(cd->flags) | CS_VALID;
+               blob->csb_end_offset = round_page(ntohl(cd->codeLimit));
+               blob->csb_start_offset = (blob->csb_end_offset -
+                                         (ntohl(cd->nCodeSlots) * PAGE_SIZE));
+               /* compute the blob's SHA1 hash */
+               sha1_base = (const unsigned char *) cd;
+               sha1_size = ntohl(cd->length);
+               SHA1Init(&sha1ctxt);
+               SHA1Update(&sha1ctxt, sha1_base, sha1_size);
+               SHA1Final(blob->csb_sha1, &sha1ctxt);
+       }
+
+       /* 
+        * Let policy module check whether the blob's signature is accepted.
+        */
+#if CONFIG_MACF
+       error = mac_vnode_check_signature(vp, blob->csb_sha1, (void*)addr, size);
+       if (error) 
+               goto out;
+#endif 
+       
+       /*
+        * Validate the blob's coverage
+        */
+       blob_start_offset = blob->csb_base_offset + blob->csb_start_offset;
+       blob_end_offset = blob->csb_base_offset + blob->csb_end_offset;
+
+       if (blob_start_offset >= blob_end_offset ||
+           blob_start_offset < 0 ||
+           blob_end_offset <= 0) {
+               /* reject empty or backwards blob */
+               error = EINVAL;
+               goto out;
+       }
+
+       vnode_lock(vp);
+       if (! UBCINFOEXISTS(vp)) {
+               vnode_unlock(vp);
+               error = ENOENT;
+               goto out;
+       }
+       uip = vp->v_ubcinfo;
+
+       /* check if this new blob overlaps with an existing blob */
+       for (oblob = uip->cs_blobs;
+            oblob != NULL;
+            oblob = oblob->csb_next) {
+                off_t oblob_start_offset, oblob_end_offset;
+
+                oblob_start_offset = (oblob->csb_base_offset +
+                                      oblob->csb_start_offset);
+                oblob_end_offset = (oblob->csb_base_offset +
+                                    oblob->csb_end_offset);
+                if (blob_start_offset >= oblob_end_offset ||
+                    blob_end_offset <= oblob_start_offset) {
+                        /* no conflict with this existing blob */
+                } else {
+                        /* conflict ! */
+                        if (blob_start_offset == oblob_start_offset &&
+                            blob_end_offset == oblob_end_offset &&
+                            blob->csb_mem_size == oblob->csb_mem_size &&
+                            blob->csb_flags == oblob->csb_flags &&
+                            (blob->csb_cpu_type == CPU_TYPE_ANY ||
+                             oblob->csb_cpu_type == CPU_TYPE_ANY ||
+                             blob->csb_cpu_type == oblob->csb_cpu_type) &&
+                            !bcmp(blob->csb_sha1,
+                                  oblob->csb_sha1,
+                                  SHA1_RESULTLEN)) {
+                                /*
+                                 * We already have this blob:
+                                 * we'll return success but
+                                 * throw away the new blob.
+                                 */
+                                if (oblob->csb_cpu_type == CPU_TYPE_ANY) {
+                                        /*
+                                         * The old blob matches this one
+                                         * but doesn't have any CPU type.
+                                         * Update it with whatever the caller
+                                         * provided this time.
+                                         */
+                                        oblob->csb_cpu_type = cputype;
+                                }
+                                vnode_unlock(vp);
+                                error = EAGAIN;
+                                goto out;
+                        } else {
+                                /* different blob: reject the new one */
+                                vnode_unlock(vp);
+                                error = EALREADY;
+                                goto out;
+                        }
+                }
+
+       }
+
+
+       /* mark this vnode's VM object as having "signed pages" */
+       kr = memory_object_signed(uip->ui_control, TRUE);
+       if (kr != KERN_SUCCESS) {
+               vnode_unlock(vp);
+               error = ENOENT;
+               goto out;
+       }
+
+       /*
+        * Add this blob to the list of blobs for this vnode.
+        * We always add at the front of the list and we never remove a
+        * blob from the list, so ubc_cs_get_blobs() can return whatever
+        * the top of the list was and that list will remain valid
+        * while we validate a page, even after we release the vnode's lock.
+        */
+       blob->csb_next = uip->cs_blobs;
+       uip->cs_blobs = blob;
+
+       OSAddAtomic(+1, &cs_blob_count);
+       if (cs_blob_count > cs_blob_count_peak) {
+               cs_blob_count_peak = cs_blob_count; /* XXX atomic ? */
+       }
+       OSAddAtomic(+blob->csb_mem_size, &cs_blob_size);
+       if (cs_blob_size > cs_blob_size_peak) {
+               cs_blob_size_peak = cs_blob_size; /* XXX atomic ? */
+       }
+       if (blob->csb_mem_size > cs_blob_size_max) {
+               cs_blob_size_max = blob->csb_mem_size;
+       }
+
+       if (cs_debug) {
+               proc_t p;
+
+               p = current_proc();
+               printf("CODE SIGNING: proc %d(%s) "
+                      "loaded %s signatures for file (%s) "
+                      "range 0x%llx:0x%llx flags 0x%x\n",
+                      p->p_pid, p->p_comm,
+                      blob->csb_cpu_type == -1 ? "detached" : "embedded",
+                      vnode_name(vp),
+                      blob->csb_base_offset + blob->csb_start_offset,
+                      blob->csb_base_offset + blob->csb_end_offset,
+                      blob->csb_flags);
+       }
+
+       vnode_unlock(vp);
+
+       error = 0;      /* success ! */
+
+out:
+       if (error) {
+               /* we failed; release what we allocated */
+               if (blob) {
+                       kfree(blob, sizeof (*blob));
+                       blob = NULL;
+               }
+               if (blob_handle != IPC_PORT_NULL) {
+                       mach_memory_entry_port_release(blob_handle);
+                       blob_handle = IPC_PORT_NULL;
+               }
+       }
+
+       if (error == EAGAIN) {
+               /*
+                * See above:  error is EAGAIN if we were asked 
+                * to add an existing blob again.  We cleaned the new
+                * blob and we want to return success.
+                */
+               error = 0;
+               /*
+                * Since we're not failing, consume the data we received.
+                */
+               ubc_cs_blob_deallocate(addr, size);
+       }
+
+       return error;
+}
+
+
+struct cs_blob *
+ubc_cs_blob_get(
+       struct vnode    *vp,
+       cpu_type_t      cputype,
+       off_t           offset)
+{
+       struct ubc_info *uip;
+       struct cs_blob  *blob;
+       off_t offset_in_blob;
+
+       vnode_lock_spin(vp);
+
+       if (! UBCINFOEXISTS(vp)) {
+               blob = NULL;
+               goto out;
+       }
+
+       uip = vp->v_ubcinfo;
+       for (blob = uip->cs_blobs;
+            blob != NULL;
+            blob = blob->csb_next) {
+               if (cputype != -1 && blob->csb_cpu_type == cputype) {
+                       break;
+               }
+               if (offset != -1) {
+                       offset_in_blob = offset - blob->csb_base_offset;
+                       if (offset_in_blob >= blob->csb_start_offset &&
+                           offset_in_blob < blob->csb_end_offset) {
+                               /* our offset is covered by this blob */
+                               break;
+                       }
+               }
+       }
+
+out:
+       vnode_unlock(vp);
+
+       return blob;
+}
+
+static void
+ubc_cs_free(
+       struct ubc_info *uip)
+{
+       struct cs_blob  *blob, *next_blob;
+
+       for (blob = uip->cs_blobs;
+            blob != NULL;
+            blob = next_blob) {
+               next_blob = blob->csb_next;
+               if (blob->csb_mem_kaddr != 0) {
+                       ubc_cs_blob_deallocate(blob->csb_mem_kaddr,
+                                              blob->csb_mem_size);
+                       blob->csb_mem_kaddr = 0;
+               }
+               if (blob->csb_mem_handle != IPC_PORT_NULL) {
+                       mach_memory_entry_port_release(blob->csb_mem_handle);
+               }
+               blob->csb_mem_handle = IPC_PORT_NULL;
+               OSAddAtomic(-1, &cs_blob_count);
+               OSAddAtomic(-blob->csb_mem_size, &cs_blob_size);
+               kfree(blob, sizeof (*blob));
+       }
+       uip->cs_blobs = NULL;
+}
+
+struct cs_blob *
+ubc_get_cs_blobs(
+       struct vnode    *vp)
+{
+       struct ubc_info *uip;
+       struct cs_blob  *blobs;
+
+       vnode_lock_spin(vp);
+
+       if (! UBCINFOEXISTS(vp)) {
+               blobs = NULL;
+               goto out;
+       }
+
+       uip = vp->v_ubcinfo;
+       blobs = uip->cs_blobs;
+
+out:
+       vnode_unlock(vp);
+
+       return blobs;
+}
+
+unsigned long cs_validate_page_no_hash = 0;
+unsigned long cs_validate_page_bad_hash = 0;
+boolean_t
+cs_validate_page(
+       void                    *_blobs,
+       memory_object_offset_t  page_offset,
+       const void              *data,
+       boolean_t               *tainted)
+{
+       SHA1_CTX                sha1ctxt;
+       unsigned char           actual_hash[SHA1_RESULTLEN];
+       unsigned char           expected_hash[SHA1_RESULTLEN];
+       boolean_t               found_hash;
+       struct cs_blob          *blobs, *blob;
+       const CS_CodeDirectory  *cd;
+       const CS_SuperBlob      *embedded;
+       off_t                   start_offset, end_offset;
+       const unsigned char     *hash;
+       boolean_t               validated;
+       off_t                   offset; /* page offset in the file */
+       size_t                  size;
+       off_t                   codeLimit = 0;
+       char                    *lower_bound, *upper_bound;
+       vm_offset_t             kaddr, blob_addr;
+       vm_size_t               ksize;
+       kern_return_t           kr;
+
+       offset = page_offset;
+
+       /* retrieve the expected hash */
+       found_hash = FALSE;
+       blobs = (struct cs_blob *) _blobs;
+
+       for (blob = blobs;
+            blob != NULL;
+            blob = blob->csb_next) {
+               offset = page_offset - blob->csb_base_offset;
+               if (offset < blob->csb_start_offset ||
+                   offset >= blob->csb_end_offset) {
+                       /* our page is not covered by this blob */
+                       continue;
+               }
+
+               /* map the blob in the kernel address space */
+               kaddr = blob->csb_mem_kaddr;
+               if (kaddr == 0) {
+                       ksize = (vm_size_t) (blob->csb_mem_size +
+                                            blob->csb_mem_offset);
+                       kr = vm_map(kernel_map,
+                                   &kaddr,
+                                   ksize,
+                                   0,
+                                   VM_FLAGS_ANYWHERE,
+                                   blob->csb_mem_handle,
+                                   0,
+                                   TRUE,
+                                   VM_PROT_READ,
+                                   VM_PROT_READ,
+                                   VM_INHERIT_NONE);
+                       if (kr != KERN_SUCCESS) {
+                               /* XXX FBDP what to do !? */
+                               printf("cs_validate_page: failed to map blob, "
+                                      "size=0x%x kr=0x%x\n",
+                                      blob->csb_mem_size, kr);
+                               break;
+                       }
+               }
+               blob_addr = kaddr + blob->csb_mem_offset;
+               
+               lower_bound = CAST_DOWN(char *, blob_addr);
+               upper_bound = lower_bound + blob->csb_mem_size;
+
+               embedded = (const CS_SuperBlob *) blob_addr;
+               cd = findCodeDirectory(embedded, lower_bound, upper_bound);
+               if (cd != NULL) {
+                       if (cd->pageSize != PAGE_SHIFT ||
+                           cd->hashType != 0x1 ||
+                           cd->hashSize != SHA1_RESULTLEN) {
+                               /* bogus blob ? */
+                               continue;
+                       }
+                           
+                       end_offset = round_page(ntohl(cd->codeLimit));
+                       start_offset = end_offset - (ntohl(cd->nCodeSlots) * PAGE_SIZE);
+                       offset = page_offset - blob->csb_base_offset;
+                       if (offset < start_offset ||
+                           offset >= end_offset) {
+                               /* our page is not covered by this blob */
+                               continue;
+                       }
+
+                       codeLimit = ntohl(cd->codeLimit);
+                       hash = hashes(cd, atop(offset),
+                                     lower_bound, upper_bound);
+                       if (hash != NULL) {
+                               bcopy(hash, expected_hash,
+                                     sizeof (expected_hash));
+                               found_hash = TRUE;
+                       }
+
+                       break;
+               }
+       }
+
+       if (found_hash == FALSE) {
+               /*
+                * We can't verify this page because there is no signature
+                * for it (yet).  It's possible that this part of the object
+                * is not signed, or that signatures for that part have not
+                * been loaded yet.
+                * Report that the page has not been validated and let the
+                * caller decide if it wants to accept it or not.
+                */
+               cs_validate_page_no_hash++;
+               if (cs_debug > 1) {
+                       printf("CODE SIGNING: cs_validate_page: "
+                              "off 0x%llx: no hash to validate !?\n",
+                              page_offset);
+               }
+               validated = FALSE;
+               *tainted = FALSE;
+       } else {
+
+               size = PAGE_SIZE;
+               const uint32_t *asha1, *esha1;
+               if (offset + size > codeLimit) {
+                       /* partial page at end of segment */
+                       assert(offset < codeLimit);
+                       size = codeLimit & PAGE_MASK;
+               }
+               /* compute the actual page's SHA1 hash */
+               SHA1Init(&sha1ctxt);
+               SHA1UpdateUsePhysicalAddress(&sha1ctxt, data, size);
+               SHA1Final(actual_hash, &sha1ctxt);
+
+               asha1 = (const uint32_t *) actual_hash;
+               esha1 = (const uint32_t *) expected_hash;
+
+               if (bcmp(expected_hash, actual_hash, SHA1_RESULTLEN) != 0) {
+                       if (cs_debug) {
+                               printf("CODE SIGNING: cs_validate_page: "
+                                      "off 0x%llx size 0x%lx: "
+                                      "actual [0x%x 0x%x 0x%x 0x%x 0x%x] != "
+                                      "expected [0x%x 0x%x 0x%x 0x%x 0x%x]\n",
+                                      page_offset, size,
+                                      asha1[0], asha1[1], asha1[2],
+                                      asha1[3], asha1[4],
+                                      esha1[0], esha1[1], esha1[2],
+                                      esha1[3], esha1[4]);
+                       }
+                       cs_validate_page_bad_hash++;
+                       *tainted = TRUE;
+               } else {
+                       if (cs_debug > 1) {
+                               printf("CODE SIGNING: cs_validate_page: "
+                                      "off 0x%llx size 0x%lx: SHA1 OK\n",
+                                      page_offset, size);
+                       }
+                       *tainted = FALSE;
+               }
+               validated = TRUE;
+       }
+       
+       return validated;
+}
+
+int
+ubc_cs_getcdhash(
+       vnode_t         vp,
+       off_t           offset,
+       unsigned char   *cdhash)
+{
+       struct cs_blob *blobs, *blob;
+       off_t rel_offset;
+
+       blobs = ubc_get_cs_blobs(vp);
+       for (blob = blobs;
+            blob != NULL;
+            blob = blob->csb_next) {
+               /* compute offset relative to this blob */
+               rel_offset = offset - blob->csb_base_offset;
+               if (rel_offset >= blob->csb_start_offset &&
+                   rel_offset < blob->csb_end_offset) {
+                       /* this blob does cover our "offset" ! */
+                       break;
+               }
+       }
+
+       if (blob == NULL) {
+               /* we didn't find a blob covering "offset" */
+               return EBADEXEC; /* XXX any better error ? */
+       }
+
+       /* get the SHA1 hash of that blob */
+       bcopy(blob->csb_sha1, cdhash, sizeof (blob->csb_sha1));
+
+       return 0;
+}