]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/kern/mach_loader.c
xnu-4903.241.1.tar.gz
[apple/xnu.git] / bsd / kern / mach_loader.c
index 89181a010230ef3e49cb43ad0928f24252928470..18d2239baf6899d5e347a50ab66076c498224ef1 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2000-2007 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2010 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
 #include <sys/fcntl.h>
 #include <sys/ubc_internal.h>
 #include <sys/imgact.h>
+#include <sys/codesign.h>
+#include <sys/proc_uuid_policy.h>
+#include <sys/reason.h>
+#include <sys/kdebug.h>
 
 #include <mach/mach_types.h>
 #include <mach/vm_map.h>       /* vm_allocate() */
@@ -59,7 +63,9 @@
 
 #include <machine/vmparam.h>
 #include <machine/exec.h>
+#include <machine/pal_routines.h>
 
+#include <kern/ast.h>
 #include <kern/kern_types.h>
 #include <kern/cpu_number.h>
 #include <kern/mach_loader.h>
 #include <vm/vm_kern.h>
 #include <vm/vm_pager.h>
 #include <vm/vnode_pager.h>
-#include <vm/vm_protos.h> 
+#include <vm/vm_protos.h>
+#include <IOKit/IOReturn.h>    /* for kIOReturnNotPrivileged */
 
+#include <os/overflow.h>
 
-/*
- * XXX vm/pmap.h should not treat these prototypes as MACH_KERNEL_PRIVATE
- * when KERNEL is defined.
- */
-extern pmap_t  pmap_create(vm_map_size_t size, boolean_t is_64bit);
-extern void    pmap_switch(pmap_t);
+#if __x86_64__
+extern int bootarg_no32exec;    /* bsd_init.c */
+#endif
 
 /*
- * XXX kern/thread.h should not treat these prototypes as MACH_KERNEL_PRIVATE
+ * XXX vm/pmap.h should not treat these prototypes as MACH_KERNEL_PRIVATE
  * when KERNEL is defined.
  */
-extern kern_return_t   thread_setstatus(thread_t thread, int flavor,
-                               thread_state_t tstate,
-                               mach_msg_type_number_t count);
-
-extern kern_return_t    thread_state_initialize(thread_t thread);
-
+extern pmap_t  pmap_create(ledger_t ledger, vm_map_size_t size,
+                               boolean_t is_64bit);
 
 /* XXX should have prototypes in a shared header file */
 extern int     get_map_nentries(vm_map_t);
-extern kern_return_t   thread_userstack(thread_t, int, thread_state_t,
-                               unsigned int, mach_vm_offset_t *, int *);
-extern kern_return_t   thread_entrypoint(thread_t, int, thread_state_t,
-                               unsigned int, mach_vm_offset_t *);
 
 extern kern_return_t   memory_object_signed(memory_object_control_t control,
                                             boolean_t is_signed);
 
 /* An empty load_result_t */
-static load_result_t load_result_null = {
+static const load_result_t load_result_null = {
        .mach_header = MACH_VM_MIN_ADDRESS,
        .entry_point = MACH_VM_MIN_ADDRESS,
        .user_stack = MACH_VM_MIN_ADDRESS,
+       .user_stack_size = 0,
+       .user_stack_alloc = MACH_VM_MIN_ADDRESS,
+       .user_stack_alloc_size = 0,
        .all_image_info_addr = MACH_VM_MIN_ADDRESS,
        .all_image_info_size = 0,
        .thread_count = 0,
        .unixproc = 0,
        .dynlinker = 0,
-       .customstack = 0,
+       .needs_dynlinker = 0,
+       .validentry = 0,
+       .using_lcmain = 0,
+       .is_64bit_addr = 0,
+       .is_64bit_data = 0,
        .csflags = 0,
-       .uuid = { 0 }
+       .has_pagezero = 0,
+       .uuid = { 0 },
+       .min_vm_addr = MACH_VM_MAX_ADDRESS,
+       .max_vm_addr = MACH_VM_MIN_ADDRESS,
+       .cs_end_offset = 0,
+       .threadstate = NULL,
+       .threadstate_sz = 0
 };
 
 /*
@@ -135,7 +145,11 @@ parse_machfile(
        off_t                   file_offset,
        off_t                   macho_size,
        int                     depth,
-       load_result_t           *result
+       int64_t                 slide,
+       int64_t                 dyld_slide,
+       load_result_t           *result,
+       load_result_t           *binresult,
+       struct image_params     *imgp
 );
 
 static load_return_t
@@ -147,16 +161,26 @@ load_segment(
        off_t                           macho_size,
        struct vnode                    *vp,
        vm_map_t                        map,
+       int64_t                         slide,
+       load_result_t                   *result
+);
+
+static load_return_t
+load_uuid(
+       struct uuid_command             *uulp,
+       char                            *command_end,
        load_result_t                   *result
 );
 
-int load_code_signature(
+static load_return_t
+load_code_signature(
        struct linkedit_data_command    *lcp,
        struct vnode                    *vp,
        off_t                           macho_offset,
        off_t                           macho_size,
        cpu_type_t                      cputype,
-       load_result_t                   *result);
+       load_result_t                   *result,
+       struct image_params             *imgp);
        
 #if CONFIG_CODE_DECRYPTION
 static load_return_t
@@ -164,20 +188,27 @@ set_code_unprotect(
        struct encryption_info_command  *lcp,
        caddr_t                         addr,
        vm_map_t                        map,
-       struct vnode                    *vp);
+       int64_t                         slide,
+       struct vnode                    *vp,
+       off_t                           macho_offset,
+       cpu_type_t                      cputype,
+       cpu_subtype_t                   cpusubtype);
 #endif
 
-static load_return_t
-load_unixthread(
-       struct thread_command   *tcp,
-       thread_t                        thread,
-       load_result_t                   *result
+static
+load_return_t
+load_main(
+       struct entry_point_command      *epc,
+       thread_t                thread,
+       int64_t                         slide,
+       load_result_t           *result
 );
 
 static load_return_t
-load_thread(
+load_unixthread(
        struct thread_command   *tcp,
        thread_t                        thread,
+       int64_t                         slide,
        load_result_t                   *result
 );
 
@@ -185,16 +216,18 @@ static load_return_t
 load_threadstate(
        thread_t                thread,
        uint32_t        *ts,
-       uint32_t        total_size
+       uint32_t        total_size,
+       load_result_t *
 );
 
 static load_return_t
 load_threadstack(
        thread_t                thread,
-       uint32_t        *ts,
-       uint32_t        total_size,
-       user_addr_t     *user_stack,
-       int                             *customstack
+       uint32_t                *ts,
+       uint32_t                total_size,
+       mach_vm_offset_t        *user_stack,
+       int                     *customstack,
+       load_result_t           *result
 );
 
 static load_return_t
@@ -209,20 +242,24 @@ static load_return_t
 load_dylinker(
        struct dylinker_command *lcp,
        integer_t               archbits,
-       vm_map_t                                map,
-       thread_t                        thread,
-       int                                             depth,
-       load_result_t                   *result,
-       boolean_t                       is_64bit
+       vm_map_t                map,
+       thread_t                thread,
+       int                     depth,
+       int64_t                 slide,
+       load_result_t           *result,
+       struct image_params     *imgp
 );
 
+struct macho_data;
+
 static load_return_t
 get_macho_vnode(
-       char                            *path,
+       const char                      *path,
        integer_t               archbits,
        struct mach_header      *mach_header,
        off_t                   *file_offset,
        off_t                   *macho_size,
+       struct macho_data       *macho_data,
        struct vnode            **vpp
 );
 
@@ -246,7 +283,7 @@ widen_segment_command(const struct segment_command *scp32,
 static void
 note_all_image_info_section(const struct segment_command_64 *scp,
     boolean_t is64, size_t section_size, const void *sections,
-    load_result_t *result)
+    int64_t slide, load_result_t *result)
 {
        const union {
                struct section s32;
@@ -254,6 +291,7 @@ note_all_image_info_section(const struct segment_command_64 *scp,
        } *sectionp;
        unsigned int i;
 
+     
        if (strncmp(scp->segname, "__DATA", sizeof(scp->segname)) != 0)
                return;
        for (i = 0; i < scp->nsects; ++i) {
@@ -263,6 +301,7 @@ note_all_image_info_section(const struct segment_command_64 *scp,
                    sizeof(sectionp->s64.sectname))) {
                        result->all_image_info_addr =
                            is64 ? sectionp->s64.addr : sectionp->s32.addr;
+                       result->all_image_info_addr += slide;
                        result->all_image_info_size =
                            is64 ? sectionp->s64.size : sectionp->s32.size;
                        return;
@@ -270,122 +309,251 @@ note_all_image_info_section(const struct segment_command_64 *scp,
        }
 }
 
+#if __arm64__
+/*
+ * Allow bypassing some security rules (hard pagezero, no write+execute)
+ * in exchange for better binary compatibility for legacy apps built
+ * before 16KB-alignment was enforced.
+ */
+const int fourk_binary_compatibility_unsafe = TRUE;
+const int fourk_binary_compatibility_allow_wx = FALSE;
+#endif /* __arm64__ */
 
 load_return_t
 load_machfile(
        struct image_params     *imgp,
        struct mach_header      *header,
        thread_t                thread,
-       vm_map_t                new_map,
+       vm_map_t                *mapp,
        load_result_t           *result
 )
 {
        struct vnode            *vp = imgp->ip_vp;
        off_t                   file_offset = imgp->ip_arch_offset;
        off_t                   macho_size = imgp->ip_arch_size;
-       
+       off_t                   file_size = imgp->ip_vattr->va_data_size;
        pmap_t                  pmap = 0;       /* protected by create_map */
        vm_map_t                map;
-       vm_map_t                old_map;
-       task_t                  old_task = TASK_NULL; /* protected by create_map */
        load_result_t           myresult;
        load_return_t           lret;
-       boolean_t create_map = FALSE;
-       int spawn = (imgp->ip_flags & IMGPF_SPAWN);
+       boolean_t enforce_hard_pagezero = TRUE;
+       int in_exec = (imgp->ip_flags & IMGPF_EXEC);
        task_t task = current_task();
+       proc_t p = current_proc();
+       int64_t                 aslr_page_offset = 0;
+       int64_t                 dyld_aslr_page_offset = 0;
+       int64_t                 aslr_section_size = 0;
+       int64_t                 aslr_section_offset = 0;
+       kern_return_t           kret;
+
+       if (macho_size > file_size) {
+               return(LOAD_BADMACHO);
+       }
+
+       result->is_64bit_addr = ((imgp->ip_flags & IMGPF_IS_64BIT_ADDR) == IMGPF_IS_64BIT_ADDR);
+       result->is_64bit_data = ((imgp->ip_flags & IMGPF_IS_64BIT_DATA) == IMGPF_IS_64BIT_DATA);
 
-       if (new_map == VM_MAP_NULL) {
-               create_map = TRUE;
-               old_task = current_task();
+       task_t ledger_task;
+       if (imgp->ip_new_thread) {
+               ledger_task = get_threadtask(imgp->ip_new_thread);
+       } else {
+               ledger_task = task;
        }
+       pmap = pmap_create(get_task_ledger(ledger_task),
+                          (vm_map_size_t) 0,
+                          result->is_64bit_addr);
+       map = vm_map_create(pmap,
+                       0,
+                       vm_compute_max_offset(result->is_64bit_addr),
+                       TRUE);
 
-       /*
-        * If we are spawning, we have created backing objects for the process
-        * already, which include non-lazily creating the task map.  So we
-        * are going to switch out the task map with one appropriate for the
-        * bitness of the image being loaded.
+#if defined(__arm64__)
+       if (result->is_64bit_addr) {
+               /* enforce 16KB alignment of VM map entries */
+               vm_map_set_page_shift(map, SIXTEENK_PAGE_SHIFT);
+       } else {
+               vm_map_set_page_shift(map, page_shift_user32);
+       }
+#elif (__ARM_ARCH_7K__ >= 2) && defined(PLATFORM_WatchOS)
+       /* enforce 16KB alignment for watch targets with new ABI */
+       vm_map_set_page_shift(map, SIXTEENK_PAGE_SHIFT);
+#endif /* __arm64__ */
+
+#ifndef        CONFIG_ENFORCE_SIGNED_CODE
+       /* This turns off faulting for executable pages, which allows
+        * to circumvent Code Signing Enforcement. The per process
+        * flag (CS_ENFORCEMENT) is not set yet, but we can use the
+        * global flag.
         */
-       if (spawn) {
-               create_map = TRUE;
-               old_task = get_threadtask(thread);
+       if ( !cs_process_global_enforcement() && (header->flags & MH_ALLOW_STACK_EXECUTION) ) {
+               vm_map_disable_NX(map);
+               // TODO: Message Trace or log that this is happening
        }
+#endif
 
-       if (create_map) {
-               pmap = pmap_create((vm_map_size_t) 0, (imgp->ip_flags & IMGPF_IS_64BIT));
-               map = vm_map_create(pmap,
-                               0,
-                               vm_compute_max_offset((imgp->ip_flags & IMGPF_IS_64BIT)),
-                               TRUE);
-       } else
-               map = new_map;
+       /* Forcibly disallow execution from data pages on even if the arch
+        * normally permits it. */
+       if ((header->flags & MH_NO_HEAP_EXECUTION) && !(imgp->ip_flags & IMGPF_ALLOW_DATA_EXEC))
+               vm_map_disallow_data_exec(map);
+
+       /*
+        * Compute a random offset for ASLR, and an independent random offset for dyld.
+        */
+       if (!(imgp->ip_flags & IMGPF_DISABLE_ASLR)) {
+               vm_map_get_max_aslr_slide_section(map, &aslr_section_offset, &aslr_section_size);
+               aslr_section_offset = (random() % aslr_section_offset) * aslr_section_size;
+
+               aslr_page_offset = random();
+               aslr_page_offset %= vm_map_get_max_aslr_slide_pages(map);
+               aslr_page_offset <<= vm_map_page_shift(map);
+
+               dyld_aslr_page_offset = random();
+               dyld_aslr_page_offset %= vm_map_get_max_loader_aslr_slide_pages(map);
+               dyld_aslr_page_offset <<= vm_map_page_shift(map);
+
+               aslr_page_offset += aslr_section_offset;
+       }
 
-       if ( (header->flags & MH_ALLOW_STACK_EXECUTION) )
-               vm_map_disable_NX(map);
-               
        if (!result)
                result = &myresult;
 
        *result = load_result_null;
 
+       /*
+        * re-set the bitness on the load result since we cleared the load result above.
+        */
+       result->is_64bit_addr = ((imgp->ip_flags & IMGPF_IS_64BIT_ADDR) == IMGPF_IS_64BIT_ADDR);
+       result->is_64bit_data = ((imgp->ip_flags & IMGPF_IS_64BIT_DATA) == IMGPF_IS_64BIT_DATA);
+
        lret = parse_machfile(vp, map, thread, header, file_offset, macho_size,
-                             0, result);
+                             0, aslr_page_offset, dyld_aslr_page_offset, result,
+                             NULL, imgp);
 
        if (lret != LOAD_SUCCESS) {
-               if (create_map) {
-                       vm_map_deallocate(map); /* will lose pmap reference too */
-               }
+               vm_map_deallocate(map); /* will lose pmap reference too */
                return(lret);
        }
 
+#if __x86_64__
+       /*
+        * On x86, for compatibility, don't enforce the hard page-zero restriction for 32-bit binaries.
+        */
+       if (!result->is_64bit_addr) {
+               enforce_hard_pagezero = FALSE;
+       }
+
+       /*
+        * For processes with IMGPF_HIGH_BITS_ASLR, add a few random high bits
+        * to the start address for "anywhere" memory allocations.
+        */
+#define VM_MAP_HIGH_START_BITS_COUNT 8
+#define VM_MAP_HIGH_START_BITS_SHIFT 27
+       if (result->is_64bit_addr &&
+           (imgp->ip_flags & IMGPF_HIGH_BITS_ASLR)) {
+               int random_bits;
+               vm_map_offset_t high_start;
+
+               random_bits = random();
+               random_bits &= (1 << VM_MAP_HIGH_START_BITS_COUNT)-1;
+               high_start = (((vm_map_offset_t)random_bits)
+                             << VM_MAP_HIGH_START_BITS_SHIFT);
+               vm_map_set_high_start(map, high_start);
+       }
+#endif /* __x86_64__ */
+
        /*
-        * For 64-bit users, check for presence of a 4GB page zero
-        * which will enable the kernel to share the user's address space
-        * and hence avoid TLB flushes on kernel entry/exit
+        * Check to see if the page zero is enforced by the map->min_offset.
         */ 
+       if (enforce_hard_pagezero &&
+           (vm_map_has_hard_pagezero(map, 0x1000) == FALSE)) {
+#if __arm64__
+               if (!result->is_64bit_addr && /* not 64-bit address space */
+                   !(header->flags & MH_PIE) &&          /* not PIE */
+                   (vm_map_page_shift(map) != FOURK_PAGE_SHIFT ||
+                    PAGE_SHIFT != FOURK_PAGE_SHIFT) && /* page size != 4KB */
+                   result->has_pagezero &&     /* has a "soft" page zero */
+                   fourk_binary_compatibility_unsafe) {
+                       /*
+                        * For backwards compatibility of "4K" apps on
+                        * a 16K system, do not enforce a hard page zero...
+                        */
+               } else
+#endif /* __arm64__ */
+               {
+                       vm_map_deallocate(map); /* will lose pmap reference too */
+                       return (LOAD_BADMACHO);
+               }
+       }
 
-       if ((imgp->ip_flags & IMGPF_IS_64BIT) &&
-            vm_map_has_4GB_pagezero(map))
-               vm_map_set_4GB_pagezero(map);
+       vm_commit_pagezero_status(map);
 
        /*
-        *      Commit to new map.
-        *
-        *      Swap the new map for the old, which  consumes our new map
-        *      reference but each leaves us responsible for the old_map reference.
-        *      That lets us get off the pmap associated with it, and
-        *      then we can release it.
+        * If this is an exec, then we are going to destroy the old
+        * task, and it's correct to halt it; if it's spawn, the
+        * task is not yet running, and it makes no sense.
         */
-
-        if (create_map) {
+       if (in_exec) {
                /*
-                * If this is an exec, then we are going to destory the old
-                * task, and it's correct to halt it; if it's spawn, the
-                * task is not yet running, and it makes no sense.
+                * Mark the task as halting and start the other
+                * threads towards terminating themselves.  Then
+                * make sure any threads waiting for a process
+                * transition get informed that we are committed to
+                * this transition, and then finally complete the
+                * task halting (wait for threads and then cleanup
+                * task resources).
+                *
+                * NOTE: task_start_halt() makes sure that no new
+                * threads are created in the task during the transition.
+                * We need to mark the workqueue as exiting before we
+                * wait for threads to terminate (at the end of which
+                * we no longer have a prohibition on thread creation).
+                *
+                * Finally, clean up any lingering workqueue data structures
+                * that may have been left behind by the workqueue threads
+                * as they exited (and then clean up the work queue itself).
                 */
-               if (!spawn) {
-                       /*
-                        * Mark the task as halting and start the other
-                        * threads towards terminating themselves.  Then
-                        * make sure any threads waiting for a process
-                        * transition get informed that we are committed to
-                        * this transition, and then finally complete the
-                        * task halting (wait for threads and then cleanup
-                        * task resources).
-                        */
-                       task_start_halt(task);
-                       proc_transcommit(current_proc(), 0);
-                       task_complete_halt(task);
+               kret = task_start_halt(task);
+               if (kret != KERN_SUCCESS) {
+                       vm_map_deallocate(map); /* will lose pmap reference too */
+                       return (LOAD_FAILURE);
                }
-               old_map = swap_task_map(old_task, thread, map);
-               vm_map_clear_4GB_pagezero(old_map);
-               /* XXX L4 : For spawn the current task isn't running... */
-               if (!spawn)
-                       pmap_switch(pmap);      /* Make sure we are using the new pmap */
-               vm_map_deallocate(old_map);
+               proc_transcommit(p, 0);
+               workq_mark_exiting(p);
+               task_complete_halt(task);
+               workq_exit(p);
+
+               /*
+                * Roll up accounting info to new task. The roll up is done after
+                * task_complete_halt to make sure the thread accounting info is
+                * rolled up to current_task.
+                */
+               task_rollup_accounting_info(get_threadtask(thread), task);
        }
+       *mapp = map;
+
+#ifdef CONFIG_32BIT_TELEMETRY
+       if (!result->is_64bit_data) {
+               /*
+                * This may not need to be an AST; we merely need to ensure that
+                * we gather telemetry at the point where all of the information
+                * that we want has been added to the process.
+                */
+               task_set_32bit_log_flag(get_threadtask(thread));
+               act_set_astbsd(thread);
+       }
+#endif /* CONFIG_32BIT_TELEMETRY */
+
        return(LOAD_SUCCESS);
 }
 
+int macho_printf = 0;
+#define MACHO_PRINTF(args)                             \
+       do {                                            \
+               if (macho_printf) {                     \
+                       printf args;                    \
+               }                                       \
+       } while (0)
+
 /*
  * The file size of a mach-o file is limited to 32 bits; this is because
  * this is the limit on the kalloc() of enough bytes for a mach_header and
@@ -393,7 +561,9 @@ load_machfile(
  * bits in the file format itself.  We read into the kernel buffer the
  * commands section, and then parse it in order to parse the mach-o file
  * format load_command segment(s).  We are only interested in a subset of
- * the total set of possible commands.
+ * the total set of possible commands. If "map"==VM_MAP_NULL or
+ * "thread"==THREAD_NULL, do not make permament VM modifications,
+ * just preflight the parse.
  */
 static
 load_return_t
@@ -405,29 +575,44 @@ parse_machfile(
        off_t                   file_offset,
        off_t                   macho_size,
        int                     depth,
-       load_result_t           *result
+       int64_t                 aslr_offset,
+       int64_t                 dyld_aslr_offset,
+       load_result_t           *result,
+       load_result_t           *binresult,
+       struct image_params     *imgp
 )
 {
        uint32_t                ncmds;
        struct load_command     *lcp;
        struct dylinker_command *dlp = 0;
-       struct uuid_command     *uulp = 0;
        integer_t               dlarchbits = 0;
        void *                  control;
        load_return_t           ret = LOAD_SUCCESS;
-       caddr_t                 addr;
-       void *                  kl_addr;
-       vm_size_t               size,kl_size;
+       void *                  addr;
+       vm_size_t               alloc_size, cmds_size;
        size_t                  offset;
        size_t                  oldoffset;      /* for overflow check */
        int                     pass;
        proc_t                  p = current_proc();             /* XXXX */
        int                     error;
-       int resid=0;
-       task_t task;
+       int                     resid = 0;
        size_t                  mach_header_sz = sizeof(struct mach_header);
        boolean_t               abi64;
        boolean_t               got_code_signatures = FALSE;
+       boolean_t               found_header_segment = FALSE;
+       boolean_t               found_xhdr = FALSE;
+       int64_t                 slide = 0;
+       boolean_t               dyld_no_load_addr = FALSE;
+       boolean_t               is_dyld = FALSE;
+       vm_map_offset_t         effective_page_mask = MAX(PAGE_MASK, vm_map_page_mask(map));
+#if __arm64__
+       uint32_t                pagezero_end = 0;
+       uint32_t                executable_end = 0;
+       uint32_t                writable_start = 0;
+       vm_map_size_t           effective_page_size;
+
+       effective_page_size = MAX(PAGE_SIZE, vm_map_page_size(map));
+#endif /* __arm64__ */
 
        if (header->magic == MH_MAGIC_64 ||
            header->magic == MH_CIGAM_64) {
@@ -437,45 +622,55 @@ parse_machfile(
        /*
         *      Break infinite recursion
         */
-       if (depth > 6) {
+       if (depth > 1) {
                return(LOAD_FAILURE);
        }
 
-       task = (task_t)get_threadtask(thread);
-
        depth++;
 
        /*
         *      Check to see if right machine type.
         */
-       if (((cpu_type_t)(header->cputype & ~CPU_ARCH_MASK) != cpu_type()) ||
-           !grade_binary(header->cputype, 
+       if (((cpu_type_t)(header->cputype & ~CPU_ARCH_MASK) != (cpu_type() & ~CPU_ARCH_MASK)) ||
+           !grade_binary(header->cputype,
                header->cpusubtype & ~CPU_SUBTYPE_MASK))
                return(LOAD_BADARCH);
-               
+
+#if __x86_64__
+       if (bootarg_no32exec && (header->cputype == CPU_TYPE_X86)) {
+               return(LOAD_BADARCH_X86);
+       }
+#endif
+
        abi64 = ((header->cputype & CPU_ARCH_ABI64) == CPU_ARCH_ABI64);
-               
+
        switch (header->filetype) {
-       
-       case MH_OBJECT:
+
        case MH_EXECUTE:
-       case MH_PRELOAD:
                if (depth != 1) {
                        return (LOAD_FAILURE);
                }
-               break;
-               
-       case MH_FVMLIB:
-       case MH_DYLIB:
-               if (depth == 1) {
+#if CONFIG_EMBEDDED
+               if (header->flags & MH_DYLDLINK) {
+                       /* Check properties of dynamic executables */
+                       if (!(header->flags & MH_PIE) && pie_required(header->cputype, header->cpusubtype & ~CPU_SUBTYPE_MASK)) {
+                               return (LOAD_FAILURE);
+                       }
+                       result->needs_dynlinker = TRUE;
+               } else {
+                       /* Check properties of static executables (disallowed except for development) */
+#if !(DEVELOPMENT || DEBUG)
                        return (LOAD_FAILURE);
+#endif
                }
-               break;
+#endif /* CONFIG_EMBEDDED */
 
+               break;
        case MH_DYLINKER:
                if (depth != 2) {
                        return (LOAD_FAILURE);
                }
+               is_dyld = TRUE;
                break;
                
        default:
@@ -487,42 +682,147 @@ parse_machfile(
         */
        control = ubc_getobject(vp, UBC_FLAGS_NONE);
 
-       /*
-        *      Map portion that must be accessible directly into
-        *      kernel's map.
-        */
-       if ((off_t)(mach_header_sz + header->sizeofcmds) > macho_size)
-               return(LOAD_BADMACHO);
-
-       /*
-        *      Round size of Mach-O commands up to page boundry.
-        */
-       size = round_page(mach_header_sz + header->sizeofcmds);
-       if (size <= 0)
-               return(LOAD_BADMACHO);
+       /* ensure header + sizeofcmds falls within the file */
+       if (os_add_overflow(mach_header_sz, header->sizeofcmds, &cmds_size) ||
+                       (off_t)cmds_size > macho_size ||
+                       round_page_overflow(cmds_size, &alloc_size)) {
+               return LOAD_BADMACHO;
+       }
 
        /*
         * Map the load commands into kernel memory.
         */
-       addr = 0;
-       kl_size = size;
-       kl_addr = kalloc(size);
-       addr = (caddr_t)kl_addr;
-       if (addr == NULL)
-               return(LOAD_NOSPACE);
+       addr = kalloc(alloc_size);
+       if (addr == NULL) {
+               return LOAD_NOSPACE;
+       }
 
-       error = vn_rdwr(UIO_READ, vp, addr, size, file_offset,
+       error = vn_rdwr(UIO_READ, vp, addr, alloc_size, file_offset,
            UIO_SYSSPACE, 0, kauth_cred_get(), &resid, p);
        if (error) {
-               if (kl_addr )
-                       kfree(kl_addr, kl_size);
-               return(LOAD_IOERROR);
+               kfree(addr, alloc_size);
+               return LOAD_IOERROR;
        }
-       
+
+       if (resid) {
+               /* We must be able to read in as much as the mach_header indicated */
+               kfree(addr, alloc_size);
+               return LOAD_BADMACHO;
+       }
+
+       /*
+        *      For PIE and dyld, slide everything by the ASLR offset.
+        */
+       if ((header->flags & MH_PIE) || is_dyld) {
+               slide = aslr_offset;
+       }
+
        /*
-        *      Scan through the commands, processing each one as necessary.
+        *  Scan through the commands, processing each one as necessary.
+        *  We parse in three passes through the headers:
+        *  0: determine if TEXT and DATA boundary can be page-aligned
+        *  1: thread state, uuid, code signature
+        *  2: segments
+        *  3: dyld, encryption, check entry point
         */
-       for (pass = 1; pass <= 2; pass++) {
+
+       boolean_t slide_realign = FALSE;
+#if __arm64__
+       if (!abi64) {
+               slide_realign = TRUE;
+       }
+#endif
+
+       for (pass = 0; pass <= 3; pass++) {
+
+               if (pass == 0 && !slide_realign && !is_dyld) {
+                       /* if we dont need to realign the slide or determine dyld's load
+                        * address, pass 0 can be skipped */
+                       continue;
+               } else if (pass == 1) {
+#if __arm64__
+                       boolean_t       is_pie;
+                       int64_t         adjust;
+
+                       is_pie = ((header->flags & MH_PIE) != 0);
+                       if (pagezero_end != 0 &&
+                           pagezero_end < effective_page_size) {
+                               /* need at least 1 page for PAGEZERO */
+                               adjust = effective_page_size;
+                               MACHO_PRINTF(("pagezero boundary at "
+                                             "0x%llx; adjust slide from "
+                                             "0x%llx to 0x%llx%s\n",
+                                             (uint64_t) pagezero_end,
+                                             slide,
+                                             slide + adjust,
+                                             (is_pie
+                                              ? ""
+                                              : " BUT NO PIE ****** :-(")));
+                               if (is_pie) {
+                                       slide += adjust;
+                                       pagezero_end += adjust;
+                                       executable_end += adjust;
+                                       writable_start += adjust;
+                               }
+                       }
+                       if (pagezero_end != 0) {
+                               result->has_pagezero = TRUE;
+                       }
+                       if (executable_end == writable_start && 
+                           (executable_end & effective_page_mask) != 0 &&
+                           (executable_end & FOURK_PAGE_MASK) == 0) {
+
+                               /*
+                                * The TEXT/DATA boundary is 4K-aligned but
+                                * not page-aligned.  Adjust the slide to make
+                                * it page-aligned and avoid having a page
+                                * with both write and execute permissions.
+                                */
+                               adjust =
+                                       (effective_page_size -
+                                        (executable_end & effective_page_mask));
+                               MACHO_PRINTF(("page-unaligned X-W boundary at "
+                                             "0x%llx; adjust slide from "
+                                             "0x%llx to 0x%llx%s\n",
+                                             (uint64_t) executable_end,
+                                             slide,
+                                             slide + adjust,
+                                             (is_pie
+                                              ? ""
+                                              : " BUT NO PIE ****** :-(")));
+                               if (is_pie)
+                                       slide += adjust;
+                       }
+#endif /* __arm64__ */
+
+                       if (dyld_no_load_addr && binresult) {
+                               /*
+                                * The dyld Mach-O does not specify a load address. Try to locate
+                                * it right after the main binary. If binresult == NULL, load
+                                * directly to the given slide.
+                                */
+                               slide = vm_map_round_page(slide + binresult->max_vm_addr, effective_page_mask);
+                       }
+               }
+
+               /*
+                * Check that the entry point is contained in an executable segments
+                */ 
+               if ((pass == 3) && (!result->using_lcmain && result->validentry == 0)) {
+                       thread_state_initialize(thread);
+                       ret = LOAD_FAILURE;
+                       break;
+               }
+
+               /*
+                * Check that some segment maps the start of the mach-o file, which is
+                * needed by the dynamic loader to read the mach headers, etc.
+                */
+               if ((pass == 3) && (found_header_segment == FALSE)) {
+                       ret = LOAD_BADMACHO;
+                       break;
+               }
+
                /*
                 * Loop through each of the load_commands indicated by the
                 * Mach-O header; if an absurd value is provided, we just
@@ -531,13 +831,20 @@ parse_machfile(
                 */
                offset = mach_header_sz;
                ncmds = header->ncmds;
+
                while (ncmds--) {
+
+                       /* ensure enough space for a minimal load command */
+                       if (offset + sizeof(struct load_command) > cmds_size) {
+                               ret = LOAD_BADMACHO;
+                               break;
+                       }
+
                        /*
                         *      Get a pointer to the command.
                         */
                        lcp = (struct load_command *)(addr + offset);
                        oldoffset = offset;
-                       offset += lcp->cmdsize;
 
                        /*
                         * Perform prevalidation of the struct load_command
@@ -547,9 +854,9 @@ parse_machfile(
                         * straddle or exist past the reserved section at the
                         * start of the image.
                         */
-                       if (oldoffset > offset ||
-                           lcp->cmdsize < sizeof(struct load_command) ||
-                           offset > header->sizeofcmds + mach_header_sz) {
+                       if (os_add_overflow(offset, lcp->cmdsize, &offset) ||
+                                       lcp->cmdsize < sizeof(struct load_command) ||
+                                       offset > cmds_size) {
                                ret = LOAD_BADMACHO;
                                break;
                        }
@@ -559,36 +866,157 @@ parse_machfile(
                         * intervention is required.
                         */
                        switch(lcp->cmd) {
-                       case LC_SEGMENT:
-                       case LC_SEGMENT_64:
-                               if (pass != 1)
+                       case LC_SEGMENT: {
+                               struct segment_command *scp = (struct segment_command *) lcp;
+                               if (pass == 0) {
+                                       if (is_dyld && scp->vmaddr == 0 && scp->fileoff == 0) {
+                                               dyld_no_load_addr = TRUE;
+                                               if (!slide_realign) {
+                                                       /* got what we need, bail early on pass 0 */
+                                                       continue;
+                                               }
+                                       }
+
+#if __arm64__
+                                       assert(!abi64);
+
+                                       if (scp->initprot == 0 && scp->maxprot == 0 && scp->vmaddr == 0) {
+                                               /* PAGEZERO */
+                                               if (os_add3_overflow(scp->vmaddr, scp->vmsize, slide, &pagezero_end)) {
+                                                       ret = LOAD_BADMACHO;
+                                                       break;
+                                               }
+                                       }
+                                       if (scp->initprot & VM_PROT_EXECUTE) {
+                                               /* TEXT */
+                                               if (os_add3_overflow(scp->vmaddr, scp->vmsize, slide, &executable_end)) {
+                                                       ret = LOAD_BADMACHO;
+                                                       break;
+                                               }
+                                       }
+                                       if (scp->initprot & VM_PROT_WRITE) {
+                                               /* DATA */
+                                               if (os_add_overflow(scp->vmaddr, slide, &writable_start)) {
+                                                       ret = LOAD_BADMACHO;
+                                                       break;
+                                               }
+                                       }
+#endif /* __arm64__ */
+                                       break;
+                               }
+
+                               if (pass == 1 && !strncmp(scp->segname, "__XHDR", sizeof(scp->segname))) {
+                                       found_xhdr = TRUE;
+                               }
+
+                               if (pass != 2)
+                                       break;
+
+                               if (abi64) {
+                                       /*
+                                        * Having an LC_SEGMENT command for the
+                                        * wrong ABI is invalid <rdar://problem/11021230>
+                                        */
+                                       ret = LOAD_BADMACHO;
                                        break;
+                               }
+
                                ret = load_segment(lcp,
-                                                  header->filetype,
-                                                  control,
-                                                  file_offset,
-                                                  macho_size,
-                                                  vp,
-                                                  map,
-                                                  result);
+                                                  header->filetype,
+                                                  control,
+                                                  file_offset,
+                                                  macho_size,
+                                                  vp,
+                                                  map,
+                                                  slide,
+                                                  result);
+                               if (ret == LOAD_SUCCESS && scp->fileoff == 0 && scp->filesize > 0) {
+                                       /* Enforce a single segment mapping offset zero, with R+X
+                                        * protection. */
+                                       if (found_header_segment ||
+                                                       ((scp->initprot & (VM_PROT_READ|VM_PROT_EXECUTE)) != (VM_PROT_READ|VM_PROT_EXECUTE))) {
+                                               ret = LOAD_BADMACHO;
+                                               break;
+                                       }
+                                       found_header_segment = TRUE;
+                               }
+
                                break;
-                       case LC_THREAD:
+                       }
+                       case LC_SEGMENT_64: {
+                               struct segment_command_64 *scp64 = (struct segment_command_64 *) lcp;
+
+                               if (pass == 0) {
+                                       if (is_dyld && scp64->vmaddr == 0 && scp64->fileoff == 0) {
+                                               dyld_no_load_addr = TRUE;
+                                               if (!slide_realign) {
+                                                       /* got what we need, bail early on pass 0 */
+                                                       continue;
+                                               }
+                                       }
+                               }
+
+                               if (pass == 1 && !strncmp(scp64->segname, "__XHDR", sizeof(scp64->segname))) {
+                                       found_xhdr = TRUE;
+                               }
+
                                if (pass != 2)
                                        break;
-                               ret = load_thread((struct thread_command *)lcp,
-                                                  thread,
-                                                 result);
+
+                               if (!abi64) {
+                                       /*
+                                        * Having an LC_SEGMENT_64 command for the
+                                        * wrong ABI is invalid <rdar://problem/11021230>
+                                        */
+                                       ret = LOAD_BADMACHO;
+                                       break;
+                               }
+
+                               ret = load_segment(lcp,
+                                                  header->filetype,
+                                                  control,
+                                                  file_offset,
+                                                  macho_size,
+                                                  vp,
+                                                  map,
+                                                  slide,
+                                                  result);
+
+                               if (ret == LOAD_SUCCESS && scp64->fileoff == 0 && scp64->filesize > 0) {
+                                       /* Enforce a single segment mapping offset zero, with R+X
+                                        * protection. */
+                                       if (found_header_segment ||
+                                                       ((scp64->initprot & (VM_PROT_READ|VM_PROT_EXECUTE)) != (VM_PROT_READ|VM_PROT_EXECUTE))) {
+                                               ret = LOAD_BADMACHO;
+                                               break;
+                                       }
+                                       found_header_segment = TRUE;
+                               }
+
                                break;
+                       }
                        case LC_UNIXTHREAD:
-                               if (pass != 2)
+                               if (pass != 1)
                                        break;
                                ret = load_unixthread(
                                                 (struct thread_command *) lcp,
-                                                  thread,
+                                                thread,
+                                                slide,
+                                                result);
+                               break;
+                       case LC_MAIN:
+                               if (pass != 1)
+                                       break;
+                               if (depth != 1)
+                                       break;
+                               ret = load_main(
+                                                (struct entry_point_command *) lcp,
+                                                thread,
+                                                slide,
                                                 result);
                                break;
                        case LC_LOAD_DYLINKER:
-                               if (pass != 2)
+                               if (pass != 3)
                                        break;
                                if ((depth == 1) && (dlp == 0)) {
                                        dlp = (struct dylinker_command *)lcp;
@@ -598,14 +1026,15 @@ parse_machfile(
                                }
                                break;
                        case LC_UUID:
-                               if (pass == 2 && depth == 1) {
-                                       uulp = (struct uuid_command *)lcp;
-                                       memcpy(&result->uuid[0], &uulp->uuid[0], sizeof(result->uuid));
+                               if (pass == 1 && depth == 1) {
+                                       ret = load_uuid((struct uuid_command *) lcp,
+                                                       (char *)addr + cmds_size,
+                                                       result);
                                }
                                break;
                        case LC_CODE_SIGNATURE:
                                /* CODE SIGNING */
-                               if (pass != 2)
+                               if (pass != 1)
                                        break;
                                /* pager -> uip ->
                                   load signatures & store in uip
@@ -617,34 +1046,112 @@ parse_machfile(
                                        file_offset,
                                        macho_size,
                                        header->cputype,
-                                       (depth == 1) ? result : NULL);
+                                       result,
+                                       imgp);
                                if (ret != LOAD_SUCCESS) {
                                        printf("proc %d: load code signature error %d "
                                               "for file \"%s\"\n",
                                               p->p_pid, ret, vp->v_name);
-                                       ret = LOAD_SUCCESS; /* ignore error */
+                                       /*
+                                        * Allow injections to be ignored on devices w/o enforcement enabled
+                                        */
+                                       if (!cs_process_global_enforcement())
+                                           ret = LOAD_SUCCESS; /* ignore error */
+
                                } else {
                                        got_code_signatures = TRUE;
                                }
+
+                               if (got_code_signatures) {
+                                       unsigned tainted = CS_VALIDATE_TAINTED;
+                                       boolean_t valid = FALSE;
+                                       vm_size_t off = 0;
+
+
+                                       if (cs_debug > 10)
+                                               printf("validating initial pages of %s\n", vp->v_name);
+                                       
+                                       while (off < alloc_size && ret == LOAD_SUCCESS) {
+                                            tainted = CS_VALIDATE_TAINTED;
+
+                                            valid = cs_validate_range(vp,
+                                                                      NULL,
+                                                                      file_offset + off,
+                                                                      addr + off,
+                                                                      PAGE_SIZE,
+                                                                      &tainted);
+                                            if (!valid || (tainted & CS_VALIDATE_TAINTED)) {
+                                                    if (cs_debug)
+                                                            printf("CODE SIGNING: %s[%d]: invalid initial page at offset %lld validated:%d tainted:%d csflags:0x%x\n", 
+                                                                   vp->v_name, p->p_pid, (long long)(file_offset + off), valid, tainted, result->csflags);
+                                                    if (cs_process_global_enforcement() ||
+                                                        (result->csflags & (CS_HARD|CS_KILL|CS_ENFORCEMENT))) {
+                                                            ret = LOAD_FAILURE;
+                                                    }
+                                                    result->csflags &= ~CS_VALID;
+                                            }
+                                            off += PAGE_SIZE;
+                                       }
+                               }
+
                                break;
 #if CONFIG_CODE_DECRYPTION
                        case LC_ENCRYPTION_INFO:
-                               if (pass != 2)
+                       case LC_ENCRYPTION_INFO_64:
+                               if (pass != 3)
                                        break;
                                ret = set_code_unprotect(
                                        (struct encryption_info_command *) lcp,
-                                       addr, map, vp);
+                                       addr, map, slide, vp, file_offset,
+                                       header->cputype, header->cpusubtype);
                                if (ret != LOAD_SUCCESS) {
+                                       os_reason_t load_failure_reason = OS_REASON_NULL;
                                        printf("proc %d: set_code_unprotect() error %d "
                                               "for file \"%s\"\n",
                                               p->p_pid, ret, vp->v_name);
-                                       /* Don't let the app run if it's 
+                                       /* 
+                                        * Don't let the app run if it's 
                                         * encrypted but we failed to set up the
-                                        * decrypter */
-                                        psignal(p, SIGKILL);
+                                        * decrypter. If the keys are missing it will
+                                        * return LOAD_DECRYPTFAIL.
+                                        */
+                                        if (ret == LOAD_DECRYPTFAIL) {
+                                               /* failed to load due to missing FP keys */
+                                               proc_lock(p);
+                                               p->p_lflag |= P_LTERM_DECRYPTFAIL;
+                                               proc_unlock(p);
+
+                                               KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
+                                                               p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_FAIRPLAY_DECRYPT, 0, 0);
+                                               load_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_FAIRPLAY_DECRYPT);
+                                        } else {
+
+                                               KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
+                                                               p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_DECRYPT, 0, 0);
+                                               load_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_DECRYPT);
+                                        }
+
+                                        assert(load_failure_reason != OS_REASON_NULL);
+                                        psignal_with_reason(p, SIGKILL, load_failure_reason);
                                }
                                break;
 #endif
+#if __arm64__
+                       case LC_VERSION_MIN_IPHONEOS: {
+                               struct version_min_command *vmc;
+
+                               if (pass != 1) {
+                                       break;
+                               }
+                               vmc = (struct version_min_command *) lcp;
+                               if (vmc->sdk < (12 << 16)) {
+                                       /* app built with a pre-iOS12 SDK: apply legacy footprint mitigation */
+                                       result->legacy_footprint = TRUE;
+                               }
+//                             printf("FBDP %s:%d vp %p (%s) sdk %d.%d.%d -> legacy_footprint=%d\n", __FUNCTION__, __LINE__, vp, vp->v_name, (vmc->sdk >> 16), ((vmc->sdk & 0xFF00) >> 8), (vmc->sdk & 0xFF), result->legacy_footprint);
+                               break;
+                       }
+#endif /* __arm64__ */
                        default:
                                /* Other commands are ignored by the kernel */
                                ret = LOAD_SUCCESS;
@@ -656,51 +1163,53 @@ parse_machfile(
                if (ret != LOAD_SUCCESS)
                        break;
        }
-       if (ret == LOAD_SUCCESS) { 
-           if (! got_code_signatures) {
-                   struct cs_blob *blob;
-                   /* no embedded signatures: look for detached ones */
-                   blob = ubc_cs_blob_get(vp, -1, file_offset);
-                   if (blob != NULL) {
-                           /* get flags to be applied to the process */
-                           result->csflags |= blob->csb_flags;
-                   }
-           }
 
-           if (dlp != 0)
-                       ret = load_dylinker(dlp, dlarchbits, map, thread, depth, result, abi64);
+       if (ret == LOAD_SUCCESS) {
+               if(!got_code_signatures && cs_process_global_enforcement()) {
+                       ret = LOAD_FAILURE;
+               }
 
-           if(depth == 1) {
-               if (result->thread_count == 0) {
+               /* Make sure if we need dyld, we got it */
+               if (result->needs_dynlinker && !dlp) {
                        ret = LOAD_FAILURE;
-               } else if ( abi64 ) {
-#ifdef __ppc__
-                       /* Map in 64-bit commpage */
+               }
+
+               if ((ret == LOAD_SUCCESS) && (dlp != 0)) {
                        /*
-                        * PPC51: ppc64 is limited to 51-bit addresses.
-                        * Memory above that limit is handled specially
-                        * at the pmap level.
-                        *
-                        * <rdar://6640492> -- wrong task for vfork()/spawn()
+                        * load the dylinker, and slide it by the independent DYLD ASLR
+                        * offset regardless of the PIE-ness of the main binary.
                         */
-                       pmap_map_sharedpage(current_task(), get_map_pmap(map));
-#endif /* __ppc__ */
+                       ret = load_dylinker(dlp, dlarchbits, map, thread, depth,
+                                           dyld_aslr_offset, result, imgp);
                }
+
+               if ((ret == LOAD_SUCCESS) && (depth == 1)) {
+                       if (result->thread_count == 0) {
+                               ret = LOAD_FAILURE;
+                       }
+#if CONFIG_ENFORCE_SIGNED_CODE
+                       if (result->needs_dynlinker && !(result->csflags & CS_DYLD_PLATFORM)) {
+                               ret = LOAD_FAILURE;
+                       }
+#endif
            }
        }
 
-       if (kl_addr )
-               kfree(kl_addr, kl_size);
+       if (ret == LOAD_BADMACHO && found_xhdr) {
+               ret = LOAD_BADMACHO_UPX;
+       }
+
+       kfree(addr, alloc_size);
 
-       return(ret);
+       return ret;
 }
 
 #if CONFIG_CODE_DECRYPTION
 
-#define        APPLE_UNPROTECTED_HEADER_SIZE   (3 * PAGE_SIZE_64)
+#define        APPLE_UNPROTECTED_HEADER_SIZE   (3 * 4096)
 
 static load_return_t
-unprotect_segment(
+unprotect_dsmos_segment(
        uint64_t        file_off,
        uint64_t        file_size,
        struct vnode    *vp,
@@ -740,9 +1249,32 @@ unprotect_segment(
                crypt_info.crypt_end = NULL;
 #pragma unused(vp, macho_offset)
                crypt_info.crypt_ops = (void *)0x2e69cf40;
+               vm_map_offset_t crypto_backing_offset;
+               crypto_backing_offset = -1; /* i.e. use map entry's offset */
+#if VM_MAP_DEBUG_APPLE_PROTECT
+               if (vm_map_debug_apple_protect) {
+                       struct proc *p;
+                       p = current_proc();
+                       printf("APPLE_PROTECT: %d[%s] map %p "
+                              "[0x%llx:0x%llx] %s(%s)\n",
+                              p->p_pid, p->p_comm, map,
+                              (uint64_t) map_addr,
+                              (uint64_t) (map_addr + map_size),
+                              __FUNCTION__, vp->v_name);
+               }
+#endif /* VM_MAP_DEBUG_APPLE_PROTECT */
+
+               /* The DSMOS pager can only be used by apple signed code */
+               struct cs_blob * blob = csvnode_get_blob(vp, file_off);
+               if( blob == NULL || !blob->csb_platform_binary || blob->csb_platform_path)
+               {
+                       return LOAD_FAILURE;
+               }
+
                kr = vm_map_apple_protected(map,
                                            map_addr,
                                            map_addr + map_size,
+                                           crypto_backing_offset,
                                            &crypt_info);
        }
 
@@ -753,7 +1285,7 @@ unprotect_segment(
 }
 #else  /* CONFIG_CODE_DECRYPTION */
 static load_return_t
-unprotect_segment(
+unprotect_dsmos_segment(
        __unused        uint64_t        file_off,
        __unused        uint64_t        file_size,
        __unused        struct vnode    *vp,
@@ -766,76 +1298,398 @@ unprotect_segment(
 }
 #endif /* CONFIG_CODE_DECRYPTION */
 
+
+/*
+ * map_segment:
+ *     Maps a Mach-O segment, taking care of mis-alignment (wrt the system
+ *     page size) issues.
+ * 
+ *     The mapping might result in 1, 2 or 3 map entries:
+ *     1. for the first page, which could be overlap with the previous
+ *        mapping,
+ *     2. for the center (if applicable),
+ *     3. for the last page, which could overlap with the next mapping.
+ *
+ *     For each of those map entries, we might have to interpose a
+ *     "fourk_pager" to deal with mis-alignment wrt the system page size,
+ *     either in the mapping address and/or size or the file offset and/or
+ *     size.
+ *     The "fourk_pager" itself would be mapped with proper alignment
+ *     wrt the system page size and would then be populated with the
+ *     information about the intended mapping, with a "4KB" granularity.
+ */
+static kern_return_t
+map_segment(
+       vm_map_t                map,
+       vm_map_offset_t         vm_start,
+       vm_map_offset_t         vm_end,
+       memory_object_control_t control,
+       vm_map_offset_t         file_start,
+       vm_map_offset_t         file_end,
+       vm_prot_t               initprot,
+       vm_prot_t               maxprot,
+       load_result_t           *result)
+{
+       vm_map_offset_t cur_offset, cur_start, cur_end;
+       kern_return_t   ret;
+       vm_map_offset_t effective_page_mask;
+       vm_map_kernel_flags_t vmk_flags, cur_vmk_flags;
+       
+       if (vm_end < vm_start ||
+           file_end < file_start) {
+               return LOAD_BADMACHO;
+       }
+       if (vm_end == vm_start ||
+           file_end == file_start) {
+               /* nothing to map... */
+               return LOAD_SUCCESS;
+       }
+
+       effective_page_mask = MAX(PAGE_MASK, vm_map_page_mask(map));
+
+       vmk_flags = VM_MAP_KERNEL_FLAGS_NONE;
+       if (vm_map_page_aligned(vm_start, effective_page_mask) &&
+           vm_map_page_aligned(vm_end, effective_page_mask) &&
+           vm_map_page_aligned(file_start, effective_page_mask) &&
+           vm_map_page_aligned(file_end, effective_page_mask)) {
+               /* all page-aligned and map-aligned: proceed */
+       } else {
+#if __arm64__
+               /* use an intermediate "4K" pager */
+               vmk_flags.vmkf_fourk = TRUE;
+#else /* __arm64__ */
+               panic("map_segment: unexpected mis-alignment "
+                     "vm[0x%llx:0x%llx] file[0x%llx:0x%llx]\n",
+                     (uint64_t) vm_start,
+                     (uint64_t) vm_end,
+                     (uint64_t) file_start,
+                     (uint64_t) file_end);
+#endif /* __arm64__ */
+       }
+
+       cur_offset = 0;
+       cur_start = vm_start;
+       cur_end = vm_start;
+#if __arm64__
+       if (!vm_map_page_aligned(vm_start, effective_page_mask)) {
+               /* one 4K pager for the 1st page */
+               cur_end = vm_map_round_page(cur_start, effective_page_mask);
+               if (cur_end > vm_end) {
+                       cur_end = vm_start + (file_end - file_start);
+               }
+               if (control != MEMORY_OBJECT_CONTROL_NULL) {
+                       ret = vm_map_enter_mem_object_control(
+                               map,
+                               &cur_start,
+                               cur_end - cur_start,
+                               (mach_vm_offset_t)0,
+                               VM_FLAGS_FIXED,
+                               vmk_flags,
+                               VM_KERN_MEMORY_NONE,
+                               control,
+                               file_start + cur_offset,
+                               TRUE, /* copy */
+                               initprot, maxprot,
+                               VM_INHERIT_DEFAULT);
+               } else {
+                       ret = vm_map_enter_mem_object(
+                               map,
+                               &cur_start,
+                               cur_end - cur_start,
+                               (mach_vm_offset_t)0,
+                               VM_FLAGS_FIXED,
+                               vmk_flags,
+                               VM_KERN_MEMORY_NONE,
+                               IPC_PORT_NULL,
+                               0, /* offset */
+                               TRUE, /* copy */
+                               initprot, maxprot,
+                               VM_INHERIT_DEFAULT);
+               }
+               if (ret != KERN_SUCCESS) {
+                       return (LOAD_NOSPACE);
+               }
+               cur_offset += cur_end - cur_start;
+       }
+#endif /* __arm64__ */
+       if (cur_end >= vm_start + (file_end - file_start)) {
+               /* all mapped: done */
+               goto done;
+       }
+       if (vm_map_round_page(cur_end, effective_page_mask) >=
+           vm_map_trunc_page(vm_start + (file_end - file_start),
+                             effective_page_mask)) {
+               /* no middle */
+       } else {
+               cur_start = cur_end;
+               if ((vm_start & effective_page_mask) !=
+                   (file_start & effective_page_mask)) {
+                       /* one 4K pager for the middle */
+                       cur_vmk_flags = vmk_flags;
+               } else {
+                       /* regular mapping for the middle */
+                       cur_vmk_flags = VM_MAP_KERNEL_FLAGS_NONE;
+               }
+
+#if CONFIG_EMBEDDED
+               (void) result;
+#else /* CONFIG_EMBEDDED */
+               /*
+                * This process doesn't have its new csflags (from
+                * the image being loaded) yet, so tell VM to override the
+                * current process's CS_ENFORCEMENT for this mapping.
+                */
+               if (result->csflags & CS_ENFORCEMENT) {
+                       cur_vmk_flags.vmkf_cs_enforcement = TRUE;
+               } else {
+                       cur_vmk_flags.vmkf_cs_enforcement = FALSE;
+               }
+               cur_vmk_flags.vmkf_cs_enforcement_override = TRUE;
+#endif /* CONFIG_EMBEDDED */
+
+               cur_end = vm_map_trunc_page(vm_start + (file_end -
+                                                       file_start),
+                                           effective_page_mask);
+               if (control != MEMORY_OBJECT_CONTROL_NULL) {
+                       ret = vm_map_enter_mem_object_control(
+                               map,
+                               &cur_start,
+                               cur_end - cur_start,
+                               (mach_vm_offset_t)0,
+                               VM_FLAGS_FIXED,
+                               cur_vmk_flags,
+                               VM_KERN_MEMORY_NONE,
+                               control,
+                               file_start + cur_offset,
+                               TRUE, /* copy */
+                               initprot, maxprot,
+                               VM_INHERIT_DEFAULT);
+               } else {
+                       ret = vm_map_enter_mem_object(
+                               map,
+                               &cur_start,
+                               cur_end - cur_start,
+                               (mach_vm_offset_t)0,
+                               VM_FLAGS_FIXED,
+                               cur_vmk_flags,
+                               VM_KERN_MEMORY_NONE,
+                               IPC_PORT_NULL,
+                               0, /* offset */
+                               TRUE, /* copy */
+                               initprot, maxprot,
+                               VM_INHERIT_DEFAULT);
+               }
+               if (ret != KERN_SUCCESS) {
+                       return (LOAD_NOSPACE);
+               }
+               cur_offset += cur_end - cur_start;
+       }
+       if (cur_end >= vm_start + (file_end - file_start)) {
+               /* all mapped: done */
+               goto done;
+       }
+       cur_start = cur_end;
+#if __arm64__
+       if (!vm_map_page_aligned(vm_start + (file_end - file_start),
+                                effective_page_mask)) {
+               /* one 4K pager for the last page */
+               cur_end = vm_start + (file_end - file_start);
+               if (control != MEMORY_OBJECT_CONTROL_NULL) {
+                       ret = vm_map_enter_mem_object_control(
+                               map,
+                               &cur_start,
+                               cur_end - cur_start,
+                               (mach_vm_offset_t)0,
+                               VM_FLAGS_FIXED,
+                               vmk_flags,
+                               VM_KERN_MEMORY_NONE,
+                               control,
+                               file_start + cur_offset,
+                               TRUE, /* copy */
+                               initprot, maxprot,
+                               VM_INHERIT_DEFAULT);
+               } else {
+                       ret = vm_map_enter_mem_object(
+                               map,
+                               &cur_start,
+                               cur_end - cur_start,
+                               (mach_vm_offset_t)0,
+                               VM_FLAGS_FIXED,
+                               vmk_flags,
+                               VM_KERN_MEMORY_NONE,
+                               IPC_PORT_NULL,
+                               0, /* offset */
+                               TRUE, /* copy */
+                               initprot, maxprot,
+                               VM_INHERIT_DEFAULT);
+               }
+               if (ret != KERN_SUCCESS) {
+                       return (LOAD_NOSPACE);
+               }
+               cur_offset += cur_end - cur_start;
+       }
+#endif /* __arm64__ */
+done:
+       assert(cur_end >= vm_start + (file_end - file_start));
+       return LOAD_SUCCESS;
+}
+
 static
 load_return_t
 load_segment(
-       struct load_command             *lcp,
-       uint32_t                        filetype,
-       void *                          control,
-       off_t                           pager_offset,
-       off_t                           macho_size,
-       struct vnode                    *vp,
-       vm_map_t                        map,
-       load_result_t           *result
-)
+       struct load_command     *lcp,
+       uint32_t                filetype,
+       void *                  control,
+       off_t                   pager_offset,
+       off_t                   macho_size,
+       struct vnode            *vp,
+       vm_map_t                map,
+       int64_t                 slide,
+       load_result_t           *result)
 {
        struct segment_command_64 segment_command, *scp;
        kern_return_t           ret;
-       mach_vm_offset_t        map_addr, map_offset;
-       mach_vm_size_t          map_size, seg_size, delta_size;
+       vm_map_size_t           delta_size;
        vm_prot_t               initprot;
        vm_prot_t               maxprot;
        size_t                  segment_command_size, total_section_size,
                                single_section_size;
-       
+       vm_map_offset_t         file_offset, file_size;
+       vm_map_offset_t         vm_offset, vm_size;
+       vm_map_offset_t         vm_start, vm_end, vm_end_aligned;
+       vm_map_offset_t         file_start, file_end;
+       kern_return_t           kr;
+       boolean_t               verbose;
+       vm_map_size_t           effective_page_size;
+       vm_map_offset_t         effective_page_mask;
+#if __arm64__
+       vm_map_kernel_flags_t   vmk_flags;
+       boolean_t               fourk_align;
+#endif /* __arm64__ */
+
+       effective_page_size = MAX(PAGE_SIZE, vm_map_page_size(map));
+       effective_page_mask = MAX(PAGE_MASK, vm_map_page_mask(map));
+
+       verbose = FALSE;
        if (LC_SEGMENT_64 == lcp->cmd) {
                segment_command_size = sizeof(struct segment_command_64);
                single_section_size  = sizeof(struct section_64);
-               scp = (struct segment_command_64 *)lcp;
+#if __arm64__
+               /* 64-bit binary: should already be 16K-aligned */
+               fourk_align = FALSE;
+#endif /* __arm64__ */
        } else {
                segment_command_size = sizeof(struct segment_command);
                single_section_size  = sizeof(struct section);
-               scp = &segment_command;
-               widen_segment_command((struct segment_command *)lcp, scp);
+#if __arm64__
+               /* 32-bit binary: might need 4K-alignment */
+               if (effective_page_size != FOURK_PAGE_SIZE) {
+                       /* not using 4K page size: need fourk_pager */
+                       fourk_align = TRUE;
+                       verbose = TRUE;
+               } else {
+                       /* using 4K page size: no need for re-alignment */
+                       fourk_align = FALSE;
+               }
+#endif /* __arm64__ */
        }
        if (lcp->cmdsize < segment_command_size)
                return (LOAD_BADMACHO);
        total_section_size = lcp->cmdsize - segment_command_size;
 
+       if (LC_SEGMENT_64 == lcp->cmd) {
+               scp = (struct segment_command_64 *)lcp;
+       } else {
+               scp = &segment_command;
+               widen_segment_command((struct segment_command *)lcp, scp);
+       }
+
+       if (verbose) {
+               MACHO_PRINTF(("+++ load_segment %s "
+                             "vm[0x%llx:0x%llx] file[0x%llx:0x%llx] "
+                             "prot %d/%d flags 0x%x\n",
+                             scp->segname,
+                             (uint64_t)(slide + scp->vmaddr),
+                             (uint64_t)(slide + scp->vmaddr + scp->vmsize),
+                             pager_offset + scp->fileoff,
+                             pager_offset + scp->fileoff + scp->filesize,
+                             scp->initprot,
+                             scp->maxprot,
+                             scp->flags));
+       }
+
        /*
         * Make sure what we get from the file is really ours (as specified
         * by macho_size).
         */
        if (scp->fileoff + scp->filesize < scp->fileoff ||
-           scp->fileoff + scp->filesize > (uint64_t)macho_size)
+           scp->fileoff + scp->filesize > (uint64_t)macho_size) {
                return (LOAD_BADMACHO);
+       }
        /*
         * Ensure that the number of sections specified would fit
         * within the load command size.
         */
-       if (total_section_size / single_section_size < scp->nsects)
+       if (total_section_size / single_section_size < scp->nsects) {
                return (LOAD_BADMACHO);
+       }
        /*
         * Make sure the segment is page-aligned in the file.
         */
-       if ((scp->fileoff & PAGE_MASK_64) != 0)
+       file_offset = pager_offset + scp->fileoff;      /* limited to 32 bits */
+       file_size = scp->filesize;
+#if __arm64__
+       if (fourk_align) {
+               if ((file_offset & FOURK_PAGE_MASK) != 0) {
+                       /*
+                        * we can't mmap() it if it's not at least 4KB-aligned
+                        * in the file
+                        */
+                       return LOAD_BADMACHO;
+               }
+       } else
+#endif /* __arm64__ */
+       if ((file_offset & PAGE_MASK_64) != 0 ||
+               /* we can't mmap() it if it's not page-aligned in the file */
+           (file_offset & vm_map_page_mask(map)) != 0) {
+               /*
+                * The 1st test would have failed if the system's page size
+                * was what this process believe is the page size, so let's
+                * fail here too for the sake of consistency.
+                */
                return (LOAD_BADMACHO);
+       }
 
        /*
-        *      Round sizes to page size.
+        * If we have a code signature attached for this slice
+        * require that the segments are within the signed part
+        * of the file.
         */
-       seg_size = round_page_64(scp->vmsize);
-       map_size = round_page_64(scp->filesize);
-       map_addr = trunc_page_64(scp->vmaddr); /* JVXXX note that in XNU TOT this is round instead of trunc for 64 bits */
-       if (seg_size == 0)
-               return (KERN_SUCCESS);
-       /* XXX (4596982) this interferes with Rosetta, so limit to 64-bit tasks */
-       if (map_addr == 0 &&
-           map_size == 0 &&
-           seg_size != 0 &&
-           scp->cmd == LC_SEGMENT_64 &&
+       if (result->cs_end_offset &&
+           result->cs_end_offset < (off_t)scp->fileoff &&
+           result->cs_end_offset - scp->fileoff < scp->filesize)
+        {
+               if (cs_debug)
+                       printf("section outside code signature\n");
+               return LOAD_BADMACHO;
+       }
+
+       vm_offset = scp->vmaddr + slide;
+       vm_size = scp->vmsize;
+
+       if (vm_size == 0)
+               return (LOAD_SUCCESS);
+       if (scp->vmaddr == 0 &&
+           file_size == 0 &&
+           vm_size != 0 &&
            (scp->initprot & VM_PROT_ALL) == VM_PROT_NONE &&
            (scp->maxprot & VM_PROT_ALL) == VM_PROT_NONE) {
+               /*
+                * For PIE, extend page zero rather than moving it.  Extending
+                * page zero keeps early allocations from falling predictably
+                * between the end of page zero and the beginning of the first
+                * slid segment.
+                */
                /*
                 * This is a "page zero" segment:  it starts at address 0,
                 * is not mapped from the binary file and is not accessible.
@@ -843,41 +1697,150 @@ load_segment(
                 * make it completely off limits by raising the VM map's
                 * minimum offset.
                 */
-               ret = vm_map_raise_min_offset(map, seg_size);
+               vm_end = vm_offset + vm_size;
+               if (vm_end < vm_offset) {
+                       return (LOAD_BADMACHO);
+               }
+               if (verbose) {
+                       MACHO_PRINTF(("++++++ load_segment: "
+                                     "page_zero up to 0x%llx\n",
+                                     (uint64_t) vm_end));
+               }
+#if __arm64__
+               if (fourk_align) {
+                       /* raise min_offset as much as page-alignment allows */
+                       vm_end_aligned = vm_map_trunc_page(vm_end,
+                                                          effective_page_mask);
+               } else
+#endif /* __arm64__ */
+               {
+                       vm_end = vm_map_round_page(vm_end,
+                                                  PAGE_MASK_64);
+                       vm_end_aligned = vm_end;
+               }
+               ret = vm_map_raise_min_offset(map,
+                                             vm_end_aligned);
+#if __arm64__
+               if (ret == 0 &&
+                   vm_end > vm_end_aligned) {
+                       /* use fourk_pager to map the rest of pagezero */
+                       assert(fourk_align);
+                       vmk_flags = VM_MAP_KERNEL_FLAGS_NONE;
+                       vmk_flags.vmkf_fourk = TRUE;
+                       ret = vm_map_enter_mem_object(
+                               map,
+                               &vm_end_aligned,
+                               vm_end - vm_end_aligned,
+                               (mach_vm_offset_t) 0,   /* mask */
+                               VM_FLAGS_FIXED,
+                               vmk_flags,
+                               VM_KERN_MEMORY_NONE,
+                               IPC_PORT_NULL,
+                               0,
+                               FALSE,  /* copy */
+                               (scp->initprot & VM_PROT_ALL),
+                               (scp->maxprot & VM_PROT_ALL),
+                               VM_INHERIT_DEFAULT);
+               }
+#endif /* __arm64__ */
+                       
                if (ret != KERN_SUCCESS) {
                        return (LOAD_FAILURE);
                }
                return (LOAD_SUCCESS);
+       } else {
+#if CONFIG_EMBEDDED
+               /* not PAGEZERO: should not be mapped at address 0 */
+               if (filetype != MH_DYLINKER && scp->vmaddr == 0) {
+                       return LOAD_BADMACHO;
+               }
+#endif /* CONFIG_EMBEDDED */
+       }
+
+#if __arm64__
+       if (fourk_align) {
+               /* 4K-align */
+               file_start = vm_map_trunc_page(file_offset,
+                                              FOURK_PAGE_MASK);
+               file_end = vm_map_round_page(file_offset + file_size,
+                                            FOURK_PAGE_MASK);
+               vm_start = vm_map_trunc_page(vm_offset,
+                                            FOURK_PAGE_MASK);
+               vm_end = vm_map_round_page(vm_offset + vm_size,
+                                          FOURK_PAGE_MASK);
+               if (!strncmp(scp->segname, "__LINKEDIT", 11) &&
+                   page_aligned(file_start) &&
+                   vm_map_page_aligned(file_start, vm_map_page_mask(map)) &&
+                   page_aligned(vm_start) &&
+                   vm_map_page_aligned(vm_start, vm_map_page_mask(map))) {
+                       /* XXX last segment: ignore mis-aligned tail */
+                       file_end = vm_map_round_page(file_end,
+                                                    effective_page_mask);
+                       vm_end = vm_map_round_page(vm_end,
+                                                  effective_page_mask);
+               }
+       } else
+#endif /* __arm64__ */
+       {
+               file_start = vm_map_trunc_page(file_offset,
+                                              effective_page_mask);
+               file_end = vm_map_round_page(file_offset + file_size,
+                                            effective_page_mask);
+               vm_start = vm_map_trunc_page(vm_offset,
+                                            effective_page_mask);
+               vm_end = vm_map_round_page(vm_offset + vm_size,
+                                          effective_page_mask);
        }
 
-       map_offset = pager_offset + scp->fileoff;       /* limited to 32 bits */
+       if (vm_start < result->min_vm_addr)
+               result->min_vm_addr = vm_start;
+       if (vm_end > result->max_vm_addr)
+               result->max_vm_addr = vm_end;
+
+       if (map == VM_MAP_NULL)
+               return (LOAD_SUCCESS);
 
-       if (map_size > 0) {
+       if (vm_size > 0) {
                initprot = (scp->initprot) & VM_PROT_ALL;
                maxprot = (scp->maxprot) & VM_PROT_ALL;
                /*
                 *      Map a copy of the file into the address space.
                 */
-               ret = vm_map_enter_mem_object_control(map,
-                               &map_addr, map_size, (mach_vm_offset_t)0,
-                               VM_FLAGS_FIXED, control, map_offset, TRUE,
-                               initprot, maxprot,
-                               VM_INHERIT_DEFAULT);
-               if (ret != KERN_SUCCESS)
-                       return (LOAD_NOSPACE);
-       
+               if (verbose) {
+                       MACHO_PRINTF(("++++++ load_segment: "
+                                     "mapping at vm [0x%llx:0x%llx] of "
+                                     "file [0x%llx:0x%llx]\n",
+                                     (uint64_t) vm_start,
+                                     (uint64_t) vm_end,
+                                     (uint64_t) file_start,
+                                     (uint64_t) file_end));
+               }
+               ret = map_segment(map,
+                                 vm_start,
+                                 vm_end,
+                                 control,
+                                 file_start,
+                                 file_end,
+                                 initprot,
+                                 maxprot,
+                                 result);
+               if (ret) {
+                       return LOAD_NOSPACE;
+               }
+
+#if FIXME
                /*
                 *      If the file didn't end on a page boundary,
                 *      we need to zero the leftover.
                 */
                delta_size = map_size - scp->filesize;
-#if FIXME
                if (delta_size > 0) {
                        mach_vm_offset_t        tmp;
        
-                       ret = mach_vm_allocate(kernel_map, &tmp, delta_size, VM_FLAGS_ANYWHERE);
-                       if (ret != KERN_SUCCESS)
+                       ret = mach_vm_allocate_kernel(kernel_map, &tmp, delta_size, VM_FLAGS_ANYWHERE, VM_KERN_MEMORY_BSD);
+                       if (ret != KERN_SUCCESS) {
                                return(LOAD_RESOURCE);
+                       }
        
                        if (copyout(tmp, map_addr + scp->filesize,
                                                                delta_size)) {
@@ -896,152 +1859,234 @@ load_segment(
         *      than the size from the file, we need to allocate
         *      zero fill memory for the rest.
         */
-       delta_size = seg_size - map_size;
+       if ((vm_end - vm_start) > (file_end - file_start)) {
+               delta_size = (vm_end - vm_start) - (file_end - file_start);
+       } else {
+               delta_size = 0;
+       }
        if (delta_size > 0) {
-               mach_vm_offset_t tmp = map_addr + map_size;
-
-               ret = mach_vm_map(map, &tmp, delta_size, 0, VM_FLAGS_FIXED,
-                                 NULL, 0, FALSE,
-                                 scp->initprot, scp->maxprot,
-                                 VM_INHERIT_DEFAULT);
-               if (ret != KERN_SUCCESS)
+               mach_vm_offset_t tmp;
+
+               tmp = vm_start + (file_end - file_start);
+               if (verbose) {
+                       MACHO_PRINTF(("++++++ load_segment: "
+                                     "delta mapping vm [0x%llx:0x%llx]\n",
+                                     (uint64_t) tmp,
+                                     (uint64_t) (tmp + delta_size)));
+               }
+               kr = map_segment(map,
+                                tmp,
+                                tmp + delta_size,
+                                MEMORY_OBJECT_CONTROL_NULL,
+                                0,
+                                delta_size,
+                                scp->initprot,
+                                scp->maxprot,
+                                result);
+               if (kr != KERN_SUCCESS) {
                        return(LOAD_NOSPACE);
+               }
        }
 
        if ( (scp->fileoff == 0) && (scp->filesize != 0) )
-               result->mach_header = map_addr;
+               result->mach_header = vm_offset;
 
        if (scp->flags & SG_PROTECTED_VERSION_1) {
-               ret = unprotect_segment(scp->fileoff,
-                                       scp->filesize,
-                                       vp,
-                                       pager_offset,
-                                       map,
-                                       map_addr,
-                                       map_size);
+               ret = unprotect_dsmos_segment(file_start,
+                                             file_end - file_start,
+                                             vp,
+                                             pager_offset,
+                                             map,
+                                             vm_start,
+                                             vm_end - vm_start);
+               if (ret != LOAD_SUCCESS) {
+                       return ret;
+               }
        } else {
                ret = LOAD_SUCCESS;
        }
-       if (LOAD_SUCCESS == ret && filetype == MH_DYLINKER &&
-           result->all_image_info_addr == MACH_VM_MIN_ADDRESS)
+
+       if (LOAD_SUCCESS == ret &&
+           filetype == MH_DYLINKER &&
+           result->all_image_info_addr == MACH_VM_MIN_ADDRESS) {
                note_all_image_info_section(scp,
-                   LC_SEGMENT_64 == lcp->cmd, single_section_size,
-                   (const char *)lcp + segment_command_size, result);
+                                           LC_SEGMENT_64 == lcp->cmd,
+                                           single_section_size,
+                                           ((const char *)lcp +
+                                            segment_command_size),
+                                           slide,
+                                           result);
+       }
+
+       if (result->entry_point != MACH_VM_MIN_ADDRESS) {
+               if ((result->entry_point >= vm_offset) && (result->entry_point < (vm_offset + vm_size))) {
+                       if ((scp->initprot & (VM_PROT_READ|VM_PROT_EXECUTE)) == (VM_PROT_READ|VM_PROT_EXECUTE)) {
+                               result->validentry = 1;
+                       } else {
+                               /* right range but wrong protections, unset if previously validated */
+                               result->validentry = 0;
+                       }
+               }
+       }
 
        return ret;
 }
 
 static
 load_return_t
-load_thread(
-       struct thread_command   *tcp,
-       thread_t                        thread,
+load_uuid(
+       struct uuid_command     *uulp,
+       char                    *command_end,
        load_result_t           *result
 )
 {
-       kern_return_t   kret;
-       load_return_t   lret;
-       task_t                  task;
-       int customstack=0;
+               /*
+                * We need to check the following for this command:
+                * - The command size should be atleast the size of struct uuid_command
+                * - The UUID part of the command should be completely within the mach-o header
+                */
 
-       if (tcp->cmdsize < sizeof(*tcp))
-               return (LOAD_BADMACHO);
-       task = get_threadtask(thread);
+               if ((uulp->cmdsize < sizeof(struct uuid_command)) ||
+                   (((char *)uulp + sizeof(struct uuid_command)) > command_end)) {
+                       return (LOAD_BADMACHO);
+               }
+               
+               memcpy(&result->uuid[0], &uulp->uuid[0], sizeof(result->uuid));
+               return (LOAD_SUCCESS);
+}
 
-       /* if count is 0; same as thread */
+static
+load_return_t
+load_main(
+       struct entry_point_command      *epc,
+       thread_t                thread,
+       int64_t                         slide,
+       load_result_t           *result
+)
+{
+       mach_vm_offset_t addr;
+       kern_return_t   ret;
+       
+       if (epc->cmdsize < sizeof(*epc))
+               return (LOAD_BADMACHO);
        if (result->thread_count != 0) {
-               kret = thread_create(task, &thread);
-               if (kret != KERN_SUCCESS)
-                       return(LOAD_RESOURCE);
-               thread_deallocate(thread);
-       }
-
-       lret = load_threadstate(thread,
-                      (uint32_t *)(((vm_offset_t)tcp) + 
-                               sizeof(struct thread_command)),
-                      tcp->cmdsize - sizeof(struct thread_command));
-       if (lret != LOAD_SUCCESS)
-               return (lret);
-
-       if (result->thread_count == 0) {
-               lret = load_threadstack(thread,
-                               (uint32_t *)(((vm_offset_t)tcp) + 
-                                       sizeof(struct thread_command)),
-                               tcp->cmdsize - sizeof(struct thread_command),
-                               &result->user_stack,
-                               &customstack);
-               if (customstack)
-                               result->customstack = 1;
-               else
-                               result->customstack = 0;
-                       
-               if (lret != LOAD_SUCCESS)
-                       return(lret);
-
-               lret = load_threadentry(thread,
-                               (uint32_t *)(((vm_offset_t)tcp) + 
-                                       sizeof(struct thread_command)),
-                               tcp->cmdsize - sizeof(struct thread_command),
-                               &result->entry_point);
-               if (lret != LOAD_SUCCESS)
-                       return(lret);
+               return (LOAD_FAILURE);
        }
+
+       if (thread == THREAD_NULL)
+               return (LOAD_SUCCESS);
+       
        /*
-        *      Resume thread now, note that this means that the thread
-        *      commands should appear after all the load commands to
-        *      be sure they don't reference anything not yet mapped.
+        * LC_MAIN specifies stack size but not location.
+        * Add guard page to allocation size (MAXSSIZ includes guard page).
         */
-       else
-               thread_resume(thread);
-               
+       if (epc->stacksize) {
+               if (os_add_overflow(epc->stacksize, 4*PAGE_SIZE, &result->user_stack_size)) {
+                       /*
+                        * We are going to immediately throw away this result, but we want
+                        * to make sure we aren't loading a dangerously close to
+                        * overflowing value, since this will have a guard page added to it
+                        * and be rounded to page boundaries
+                        */
+                       return LOAD_BADMACHO;
+               }
+               result->user_stack_size = epc->stacksize;
+               if (os_add_overflow(epc->stacksize, PAGE_SIZE, &result->user_stack_alloc_size)) {
+                       return LOAD_BADMACHO;
+               }
+       } else {
+               result->user_stack_alloc_size = MAXSSIZ;
+       }
+
+       /* use default location for stack */
+       ret = thread_userstackdefault(&addr, result->is_64bit_addr);
+       if (ret != KERN_SUCCESS)
+               return(LOAD_FAILURE);
+
+       /* The stack slides down from the default location */
+       result->user_stack = addr;
+       result->user_stack -= slide;
+
+       if (result->using_lcmain || result->entry_point != MACH_VM_MIN_ADDRESS) {
+               /* Already processed LC_MAIN or LC_UNIXTHREAD */
+               return (LOAD_FAILURE);
+       }
+
+       /* kernel does *not* use entryoff from LC_MAIN.  Dyld uses it. */
+       result->needs_dynlinker = TRUE;
+       result->using_lcmain = TRUE;
+
+       ret = thread_state_initialize( thread );
+       if (ret != KERN_SUCCESS) {
+               return(LOAD_FAILURE);
+       }
+
+       result->unixproc = TRUE;
        result->thread_count++;
 
        return(LOAD_SUCCESS);
 }
 
+
 static
 load_return_t
 load_unixthread(
        struct thread_command   *tcp,
        thread_t                thread,
+       int64_t                         slide,
        load_result_t           *result
 )
 {
        load_return_t   ret;
        int customstack =0;
-       
+       mach_vm_offset_t addr;
        if (tcp->cmdsize < sizeof(*tcp))
                return (LOAD_BADMACHO);
        if (result->thread_count != 0) {
-printf("load_unixthread: already have a thread!");
                return (LOAD_FAILURE);
        }
+
+       if (thread == THREAD_NULL)
+               return (LOAD_SUCCESS);
        
        ret = load_threadstack(thread,
-                      (uint32_t *)(((vm_offset_t)tcp) + 
-                               sizeof(struct thread_command)),
-                      tcp->cmdsize - sizeof(struct thread_command),
-                      &result->user_stack,
-                          &customstack);
+                               (uint32_t *)(((vm_offset_t)tcp) +
+                                       sizeof(struct thread_command)),
+                               tcp->cmdsize - sizeof(struct thread_command),
+                               &addr, &customstack, result);
        if (ret != LOAD_SUCCESS)
                return(ret);
 
-       if (customstack)
-                       result->customstack = 1;
-       else
-                       result->customstack = 0;
+       /* LC_UNIXTHREAD optionally specifies stack size and location */
+
+       if (!customstack) {
+               result->user_stack_alloc_size = MAXSSIZ;
+       }
+
+       /* The stack slides down from the default location */
+       result->user_stack = addr;
+       result->user_stack -= slide;
+
        ret = load_threadentry(thread,
-                      (uint32_t *)(((vm_offset_t)tcp) + 
-                               sizeof(struct thread_command)),
-                      tcp->cmdsize - sizeof(struct thread_command),
-                      &result->entry_point);
+                               (uint32_t *)(((vm_offset_t)tcp) +
+                                       sizeof(struct thread_command)),
+                               tcp->cmdsize - sizeof(struct thread_command),
+                               &addr);
        if (ret != LOAD_SUCCESS)
                return(ret);
 
+       if (result->using_lcmain || result->entry_point != MACH_VM_MIN_ADDRESS) {
+               /* Already processed LC_MAIN or LC_UNIXTHREAD */
+               return (LOAD_FAILURE);
+       }
+
+       result->entry_point = addr;
+       result->entry_point += slide;
+
        ret = load_threadstate(thread,
-                      (uint32_t *)(((vm_offset_t)tcp) + 
-                               sizeof(struct thread_command)),
-                      tcp->cmdsize - sizeof(struct thread_command));
+                               (uint32_t *)(((vm_offset_t)tcp) + sizeof(struct thread_command)),
+                               tcp->cmdsize - sizeof(struct thread_command),
+                               result);
        if (ret != LOAD_SUCCESS)
                return (ret);
 
@@ -1056,55 +2101,68 @@ load_return_t
 load_threadstate(
        thread_t        thread,
        uint32_t        *ts,
-       uint32_t        total_size
+       uint32_t        total_size,
+       load_result_t   *result
 )
 {
-       kern_return_t   ret;
        uint32_t        size;
        int             flavor;
        uint32_t        thread_size;
+       uint32_t        *local_ts = NULL;
+       uint32_t        local_ts_size = 0;
+       int             ret;
+
+       (void)thread;
+
+       if (total_size > 0) {
+               local_ts_size = total_size;
+               local_ts = kalloc(local_ts_size);
+               if (local_ts == NULL) {
+                       return LOAD_FAILURE;
+               }
+               memcpy(local_ts, ts, local_ts_size);
+               ts = local_ts;
+       }
 
-    ret = thread_state_initialize( thread );
-    if (ret != KERN_SUCCESS) {
-        return(LOAD_FAILURE);
-    }
-    
        /*
-        *      Set the new thread state; iterate through the state flavors in
-     *  the mach-o file.
+        * Validate the new thread state; iterate through the state flavors in
+        * the Mach-O file.
+        * XXX: we should validate the machine state here, to avoid failing at
+        * activation time where we can't bail out cleanly.
         */
        while (total_size > 0) {
                flavor = *ts++;
                size = *ts++;
-               if (UINT32_MAX-2 < size ||
-                   UINT32_MAX/sizeof(uint32_t) < size+2)
-                       return (LOAD_BADMACHO);
-               thread_size = (size+2)*sizeof(uint32_t);
-               if (thread_size > total_size)
-                       return(LOAD_BADMACHO);
-               total_size -= thread_size;
-               /*
-                * Third argument is a kernel space pointer; it gets cast
-                * to the appropriate type in machine_thread_set_state()
-                * based on the value of flavor.
-                */
-               ret = thread_setstatus(thread, flavor, (thread_state_t)ts, size);
-               if (ret != KERN_SUCCESS) {
-                       return(LOAD_FAILURE);
+
+               if (os_add_and_mul_overflow(size, 2, sizeof(uint32_t), &thread_size) ||
+                   os_sub_overflow(total_size, thread_size, &total_size)) {
+                       ret = LOAD_BADMACHO;
+                       goto bad;
                }
+
                ts += size;     /* ts is a (uint32_t *) */
        }
-       return(LOAD_SUCCESS);
+
+       result->threadstate = local_ts;
+       result->threadstate_sz = local_ts_size;
+       return LOAD_SUCCESS;
+
+bad:
+       if (local_ts) {
+               kfree(local_ts, local_ts_size);
+       }
+       return ret;
 }
 
 static
 load_return_t
 load_threadstack(
-       thread_t        thread,
-       uint32_t        *ts,
-       uint32_t        total_size,
-       user_addr_t     *user_stack,
-       int *customstack
+       thread_t                thread,
+       uint32_t                *ts,
+       uint32_t                total_size,
+       mach_vm_offset_t        *user_stack,
+       int                     *customstack,
+       load_result_t           *result
 )
 {
        kern_return_t   ret;
@@ -1128,7 +2186,7 @@ load_threadstack(
                 * to the appropriate type in thread_userstack() based on
                 * the value of flavor.
                 */
-               ret = thread_userstack(thread, flavor, (thread_state_t)ts, size, user_stack, customstack);
+               ret = thread_userstack(thread, flavor, (thread_state_t)ts, size, user_stack, customstack, result->is_64bit_data);
                if (ret != KERN_SUCCESS) {
                        return(LOAD_FAILURE);
                }
@@ -1179,164 +2237,162 @@ load_threadentry(
        return(LOAD_SUCCESS);
 }
 
+struct macho_data {
+       struct nameidata        __nid;
+       union macho_vnode_header {
+               struct mach_header      mach_header;
+               struct fat_header       fat_header;
+               char    __pad[512];
+       } __header;
+};
 
-static
-load_return_t
+#define DEFAULT_DYLD_PATH "/usr/lib/dyld"
+
+#if (DEVELOPMENT || DEBUG)
+extern char dyld_alt_path[];
+extern int use_alt_dyld;
+#endif
+
+static uint64_t get_va_fsid(struct vnode_attr *vap)
+{
+       if (VATTR_IS_SUPPORTED(vap, va_fsid64)) {
+               return *(uint64_t *)&vap->va_fsid64;
+       } else {
+               return vap->va_fsid;
+       }
+}
+
+static load_return_t
 load_dylinker(
        struct dylinker_command *lcp,
        integer_t               archbits,
        vm_map_t                map,
        thread_t        thread,
        int                     depth,
+       int64_t                 slide,
        load_result_t           *result,
-       boolean_t               is_64bit
+       struct image_params     *imgp
 )
 {
-       char                    *name;
-       char                    *p;
+       const char              *name;
        struct vnode            *vp = NULLVP;   /* set by get_macho_vnode() */
-       struct mach_header      header;
+       struct mach_header      *header;
        off_t                   file_offset = 0; /* set by get_macho_vnode() */
        off_t                   macho_size = 0; /* set by get_macho_vnode() */
-       vm_map_t                copy_map;
-       load_result_t           myresult;
+       load_result_t           *myresult;
        kern_return_t           ret;
-       vm_map_copy_t   tmp;
-       mach_vm_offset_t        dyl_start, map_addr;
-       mach_vm_size_t          dyl_length;
+       struct macho_data       *macho_data;
+       struct {
+               struct mach_header      __header;
+               load_result_t           __myresult;
+               struct macho_data       __macho_data;
+       } *dyld_data;
+
+       if (lcp->cmdsize < sizeof(*lcp) || lcp->name.offset >= lcp->cmdsize)
+               return LOAD_BADMACHO;
 
-       if (lcp->cmdsize < sizeof(*lcp))
-               return (LOAD_BADMACHO);
+       name = (const char *)lcp + lcp->name.offset;
 
-       name = (char *)lcp + lcp->name.offset;
-       /*
-        *      Check for a proper null terminated string.
-        */
-       p = name;
-       do {
-               if (p >= (char *)lcp + lcp->cmdsize)
-                       return(LOAD_BADMACHO);
-       } while (*p++);
+       /* Check for a proper null terminated string. */
+       size_t maxsz = lcp->cmdsize - lcp->name.offset;
+       size_t namelen = strnlen(name, maxsz);
+       if (namelen >= maxsz) {
+               return LOAD_BADMACHO;
+       }
 
-       ret = get_macho_vnode(name, archbits, &header, &file_offset, &macho_size, &vp);
-       if (ret)
-               return (ret);
-                       
-       myresult = load_result_null;
+#if (DEVELOPMENT || DEBUG)
+
+    /*
+     * rdar://23680808
+     * If an alternate dyld has been specified via boot args, check
+     * to see if PROC_UUID_ALT_DYLD_POLICY has been set on this
+     * executable and redirect the kernel to load that linker.
+     */
+
+    if (use_alt_dyld) {
+        int policy_error;
+        uint32_t policy_flags = 0;
+        int32_t policy_gencount = 0;
+        
+        policy_error = proc_uuid_policy_lookup(result->uuid, &policy_flags, &policy_gencount);
+        if (policy_error == 0) {
+            if (policy_flags & PROC_UUID_ALT_DYLD_POLICY) {
+                name = dyld_alt_path;
+            }
+        }
+    }
+#endif
 
-       /*
-        *      First try to map dyld in directly.  This should work most of
-        *      the time since there shouldn't normally be something already
-        *      mapped to its address.
-        */
+#if !(DEVELOPMENT || DEBUG)
+       if (0 != strcmp(name, DEFAULT_DYLD_PATH)) {
+               return (LOAD_BADMACHO);
+       }
+#endif
 
-       ret = parse_machfile(vp, map, thread, &header, file_offset, macho_size,
-                               depth, &myresult);
+       /* Allocate wad-of-data from heap to reduce excessively deep stacks */
 
-       /*
-        *      If it turned out something was in the way, then we'll take
-        *      take this longer path to map dyld into a temporary map and
-        *      copy it into destination map at a different address.
-        */
+       MALLOC(dyld_data, void *, sizeof (*dyld_data), M_TEMP, M_WAITOK);
+       header = &dyld_data->__header;
+       myresult = &dyld_data->__myresult;
+       macho_data = &dyld_data->__macho_data;
 
-       if (ret == LOAD_NOSPACE) {
+       ret = get_macho_vnode(name, archbits, header,
+           &file_offset, &macho_size, macho_data, &vp);
+       if (ret)
+               goto novp_out;
 
-               /*
-                *      Load the Mach-O.
-                *      Use a temporary map to do the work.
-                */
-               copy_map = vm_map_create(pmap_create(vm_map_round_page(macho_size),
-                                                    is_64bit),
-                                        get_map_min(map), get_map_max(map), TRUE);
-               if (VM_MAP_NULL == copy_map) {
-                       ret = LOAD_RESOURCE;
-                       goto out;
-               }
-       
-               myresult = load_result_null;
+       *myresult = load_result_null;
+       myresult->is_64bit_addr = result->is_64bit_addr;
+       myresult->is_64bit_data = result->is_64bit_data;
 
-               ret = parse_machfile(vp, copy_map, thread, &header,
-                                       file_offset, macho_size,
-                                       depth, &myresult);
-       
-               if (ret) {
-                       vm_map_deallocate(copy_map);
-                       goto out;
-               }
-       
-               if (get_map_nentries(copy_map) > 0) {
-       
-                       dyl_start = mach_get_vm_start(copy_map);
-                       dyl_length = mach_get_vm_end(copy_map) - dyl_start;
-       
-                       map_addr = dyl_start;
-                       ret = mach_vm_allocate(map, &map_addr, dyl_length, VM_FLAGS_ANYWHERE);
-       
-                       if (ret != KERN_SUCCESS) {
-                               vm_map_deallocate(copy_map);
-                               ret = LOAD_NOSPACE;
-                               goto out;
-                       
-                       }
+       ret = parse_machfile(vp, map, thread, header, file_offset,
+                            macho_size, depth, slide, 0, myresult, result, imgp);
 
-                       ret = vm_map_copyin(copy_map,
-                                           (vm_map_address_t)dyl_start,
-                                           (vm_map_size_t)dyl_length,
-                                           TRUE, &tmp);
-                       if (ret != KERN_SUCCESS) {
-                               (void) vm_map_remove(map,
-                                            vm_map_trunc_page(map_addr),
-                                            vm_map_round_page(map_addr + dyl_length),
-                                            VM_MAP_NO_FLAGS);
-                               vm_map_deallocate(copy_map);
-                               goto out;
-                       }
-       
-                       ret = vm_map_copy_overwrite(map,
-                                            (vm_map_address_t)map_addr,
-                                            tmp, FALSE);
-                       if (ret != KERN_SUCCESS) {
-                               vm_map_copy_discard(tmp);
-                               (void) vm_map_remove(map,
-                                            vm_map_trunc_page(map_addr),
-                                            vm_map_round_page(map_addr + dyl_length),
-                                            VM_MAP_NO_FLAGS);
-                               vm_map_deallocate(copy_map);
-                               goto out;
-                       }
-       
-                       if (map_addr != dyl_start) {
-                               myresult.entry_point += (map_addr - dyl_start);
-                               myresult.all_image_info_addr +=
-                                   (map_addr - dyl_start);
-                       }
-               } else {
-                       ret = LOAD_FAILURE;
+       if (ret == LOAD_SUCCESS) {
+               if (result->threadstate) {
+                       /* don't use the app's threadstate if we have a dyld */
+                       kfree(result->threadstate, result->threadstate_sz);
                }
+               result->threadstate = myresult->threadstate;
+               result->threadstate_sz = myresult->threadstate_sz;
 
-               vm_map_deallocate(copy_map);
-       }
-       
-       if (ret == LOAD_SUCCESS) {              
                result->dynlinker = TRUE;
-               result->entry_point = myresult.entry_point;
-               result->all_image_info_addr = myresult.all_image_info_addr;
-               result->all_image_info_size = myresult.all_image_info_size;
+               result->entry_point = myresult->entry_point;
+               result->validentry = myresult->validentry;
+               result->all_image_info_addr = myresult->all_image_info_addr;
+               result->all_image_info_size = myresult->all_image_info_size;
+               if (myresult->platform_binary) {
+                       result->csflags |= CS_DYLD_PLATFORM;
+               }
        }
-out:
+
+       struct vnode_attr va;
+       VATTR_INIT(&va);
+       VATTR_WANTED(&va, va_fsid64);
+       VATTR_WANTED(&va, va_fsid);
+       VATTR_WANTED(&va, va_fileid);
+       int error = vnode_getattr(vp, &va, imgp->ip_vfs_context);
+       if (error == 0) {
+               imgp->ip_dyld_fsid = get_va_fsid(&va);
+               imgp->ip_dyld_fsobjid = va.va_fileid;
+       }
+
        vnode_put(vp);
+novp_out:
+       FREE(dyld_data, M_TEMP);
        return (ret);
 
 }
 
-int
+static load_return_t
 load_code_signature(
        struct linkedit_data_command    *lcp,
        struct vnode                    *vp,
        off_t                           macho_offset,
        off_t                           macho_size,
        cpu_type_t                      cputype,
-       load_result_t                   *result)
+       load_result_t                   *result,
+       struct image_params             *imgp)
 {
        int             ret;
        kern_return_t   kr;
@@ -1355,19 +2411,45 @@ load_code_signature(
                goto out;
        }
 
-       blob = ubc_cs_blob_get(vp, cputype, -1);
+       blob = ubc_cs_blob_get(vp, cputype, macho_offset);
+
        if (blob != NULL) {
                /* we already have a blob for this vnode and cputype */
-               if (blob->csb_cpu_type == cputype &&
-                   blob->csb_base_offset == macho_offset &&
-                   blob->csb_mem_size == lcp->datasize) {
-                       /* it matches the blob we want here: we're done */
-                       ret = LOAD_SUCCESS;
-               } else {
+               if (blob->csb_cpu_type != cputype ||
+                   blob->csb_base_offset != macho_offset) {
                        /* the blob has changed for this vnode: fail ! */
                        ret = LOAD_BADMACHO;
+                       goto out;
                }
-               goto out;
+
+               /* It matches the blob we want here, let's verify the version */
+               if (ubc_cs_generation_check(vp) == 0) {
+                       /* No need to revalidate, we're good! */
+                       ret = LOAD_SUCCESS;
+                       goto out;
+               }
+
+               /* That blob may be stale, let's revalidate. */
+               error = ubc_cs_blob_revalidate(vp, blob, imgp, 0);
+               if (error == 0) {
+                       /* Revalidation succeeded, we're good! */
+                       ret = LOAD_SUCCESS;
+                       goto out;
+               }
+
+               if (error != EAGAIN) {
+                       printf("load_code_signature: revalidation failed: %d\n", error);
+                       ret = LOAD_FAILURE;
+                       goto out;
+               }
+
+               assert(error == EAGAIN);
+
+               /*
+                * Revalidation was not possible for this blob. We just continue as if there was no blob,
+                * rereading the signature, and ubc_cs_blob_add will do the right thing.
+                */
+               blob = NULL;
        }
 
        blob_size = lcp->datasize;
@@ -1396,21 +2478,34 @@ load_code_signature(
        if (ubc_cs_blob_add(vp,
                            cputype,
                            macho_offset,
-                           addr,
-                           lcp->datasize)) {
+                           &addr,
+                           lcp->datasize,
+                           imgp,
+                           0,
+                           &blob)) {
+               if (addr) {
+                       ubc_cs_blob_deallocate(addr, blob_size);
+               }
                ret = LOAD_FAILURE;
                goto out;
        } else {
                /* ubc_cs_blob_add() has consumed "addr" */
                addr = 0;
        }
-               
-       blob = ubc_cs_blob_get(vp, cputype, -1);
 
+#if CHECK_CS_VALIDATION_BITMAP
+       ubc_cs_validation_bitmap_allocate( vp );
+#endif
+               
        ret = LOAD_SUCCESS;
 out:
-       if (result && ret == LOAD_SUCCESS) {
+       if (ret == LOAD_SUCCESS) {
+               if (blob == NULL)
+                       panic("success, but no blob!");
+
                result->csflags |= blob->csb_flags;
+               result->platform_binary = blob->csb_platform_binary;
+               result->cs_end_offset = blob->csb_end_offset;
        }
        if (addr != 0) {
                ubc_cs_blob_deallocate(addr, blob_size);
@@ -1425,24 +2520,28 @@ out:
 
 static load_return_t
 set_code_unprotect(
-                  struct encryption_info_command *eip,
-                  caddr_t addr,        
-                  vm_map_t map,
-                  struct vnode *vp)
+       struct encryption_info_command *eip,
+       caddr_t addr,   
+       vm_map_t map,
+       int64_t slide,
+       struct vnode *vp,
+       off_t macho_offset,
+       cpu_type_t cputype,
+       cpu_subtype_t cpusubtype)
 {
-       int result, len;
-       char vpath[MAXPATHLEN];
+       int error, len;
        pager_crypt_info_t crypt_info;
        const char * cryptname = 0;
+       char *vpath;
        
        size_t offset;
        struct segment_command_64 *seg64;
        struct segment_command *seg32;
        vm_map_offset_t map_offset, map_size;
+       vm_object_offset_t crypto_backing_offset;
        kern_return_t kr;
 
-       if (eip->cmdsize < sizeof(*eip))
-               return LOAD_BADMACHO;
+       if (eip->cmdsize < sizeof(*eip)) return LOAD_BADMACHO;
        
        switch(eip->cryptid) {
                case 0:
@@ -1460,18 +2559,43 @@ set_code_unprotect(
                        return LOAD_BADMACHO;
        }
        
+       if (map == VM_MAP_NULL) return (LOAD_SUCCESS);
+       if (NULL == text_crypter_create) return LOAD_FAILURE;
+
+       MALLOC_ZONE(vpath, char *, MAXPATHLEN, M_NAMEI, M_WAITOK);
+       if(vpath == NULL) return LOAD_FAILURE;
+       
        len = MAXPATHLEN;
-       result = vn_getpath(vp, vpath, &len);
-       if(result) return result;
+       error = vn_getpath(vp, vpath, &len);
+       if (error) {
+               FREE_ZONE(vpath, MAXPATHLEN, M_NAMEI);
+               return LOAD_FAILURE;
+       }
        
        /* set up decrypter first */
-       if(NULL==text_crypter_create) return LOAD_FAILURE;
-       kr=text_crypter_create(&crypt_info, cryptname, (void*)vpath);
+       crypt_file_data_t crypt_data = {
+               .filename = vpath,
+               .cputype = cputype,
+               .cpusubtype = cpusubtype};
+       kr=text_crypter_create(&crypt_info, cryptname, (void*)&crypt_data);
+#if VM_MAP_DEBUG_APPLE_PROTECT
+       if (vm_map_debug_apple_protect) {
+               struct proc *p;
+               p  = current_proc();
+               printf("APPLE_PROTECT: %d[%s] map %p %s(%s) -> 0x%x\n",
+                      p->p_pid, p->p_comm, map, __FUNCTION__, vpath, kr);
+       }
+#endif /* VM_MAP_DEBUG_APPLE_PROTECT */
+       FREE_ZONE(vpath, MAXPATHLEN, M_NAMEI);
        
        if(kr) {
                printf("set_code_unprotect: unable to create decrypter %s, kr=%d\n",
                       cryptname, kr);
-               return LOAD_RESOURCE;
+               if (kr == kIOReturnNotPrivileged) {
+                       /* text encryption returned decryption failure */
+                       return(LOAD_DECRYPTFAIL);
+                }else
+                       return LOAD_RESOURCE;
        }
        
        /* this is terrible, but we have to rescan the load commands to find the
@@ -1498,8 +2622,9 @@ set_code_unprotect(
                                if ((seg64->fileoff <= eip->cryptoff) &&
                                    (seg64->fileoff+seg64->filesize >= 
                                     eip->cryptoff+eip->cryptsize)) {
-                                       map_offset = seg64->vmaddr + eip->cryptoff - seg64->fileoff;
+                                       map_offset = seg64->vmaddr + eip->cryptoff - seg64->fileoff + slide;
                                        map_size = eip->cryptsize;
+                                       crypto_backing_offset = macho_offset + eip->cryptoff;
                                        goto remap_now;
                                }
                        case LC_SEGMENT:
@@ -1507,8 +2632,9 @@ set_code_unprotect(
                                if ((seg32->fileoff <= eip->cryptoff) &&
                                    (seg32->fileoff+seg32->filesize >= 
                                     eip->cryptoff+eip->cryptsize)) {
-                                       map_offset = seg32->vmaddr + eip->cryptoff - seg32->fileoff;
+                                       map_offset = seg32->vmaddr + eip->cryptoff - seg32->fileoff + slide;
                                        map_size = eip->cryptsize;
+                                       crypto_backing_offset = macho_offset + eip->cryptoff;
                                        goto remap_now;
                                }
                }
@@ -1519,10 +2645,16 @@ set_code_unprotect(
        
 remap_now:
        /* now remap using the decrypter */
-       kr = vm_map_apple_protected(map, map_offset, map_offset+map_size, &crypt_info);
-       if(kr) {
+       MACHO_PRINTF(("+++ set_code_unprotect: vm[0x%llx:0x%llx]\n",
+                     (uint64_t) map_offset,
+                     (uint64_t) (map_offset+map_size)));
+       kr = vm_map_apple_protected(map,
+                                   map_offset,
+                                   map_offset+map_size,
+                                   crypto_backing_offset,
+                                   &crypt_info);
+       if (kr) {
                printf("set_code_unprotect(): mapping failed with %x\n", kr);
-               crypt_info.crypt_end(crypt_info.crypt_ops);
                return LOAD_PROTECT;
        }
        
@@ -1540,11 +2672,12 @@ remap_now:
 static
 load_return_t
 get_macho_vnode(
-       char                    *path,
+       const char              *path,
        integer_t               archbits,
        struct mach_header      *mach_header,
        off_t                   *file_offset,
        off_t                   *macho_size,
+       struct macho_data       *data,
        struct vnode            **vpp
 )
 {
@@ -1552,19 +2685,14 @@ get_macho_vnode(
        vfs_context_t           ctx = vfs_context_current();
        proc_t                  p = vfs_context_proc(ctx);
        kauth_cred_t            kerncred;
-       struct nameidata nid, *ndp;
+       struct nameidata        *ndp = &data->__nid;
        boolean_t               is_fat;
        struct fat_arch         fat_arch;
-       int                     error = LOAD_SUCCESS;
+       int                     error;
        int resid;
-       union {
-               struct mach_header      mach_header;
-               struct fat_header       fat_header;
-               char    pad[512];
-       } header;
+       union macho_vnode_header *header = &data->__header;
        off_t fsize = (off_t)0;
-       int err2;
-       
+
        /*
         * Capture the kernel credential for use in the actual read of the
         * file, since the user doing the execution may have execute rights
@@ -1575,10 +2703,8 @@ get_macho_vnode(
         */
        kerncred = vfs_context_ucred(vfs_context_kernel());
 
-       ndp = &nid;
-       
        /* init the namei data to point the file user's program name */
-       NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, CAST_USER_ADDR_T(path), ctx);
+       NDINIT(ndp, LOOKUP, OP_OPEN, FOLLOW | LOCKLEAF, UIO_SYSSPACE, CAST_USER_ADDR_T(path), ctx);
 
        if ((error = namei(ndp)) != 0) {
                if (error == ENOENT) {
@@ -1590,7 +2716,7 @@ get_macho_vnode(
        }
        nameidone(ndp);
        vp = ndp->ni_vp;
-       
+
        /* check for regular file */
        if (vp->v_type != VREG) {
                error = LOAD_PROTECT;
@@ -1610,7 +2736,7 @@ get_macho_vnode(
        }
 
        /* check access */
-       if ((error = vnode_authorize(vp, NULL, KAUTH_VNODE_EXECUTE, ctx)) != 0) {
+       if ((error = vnode_authorize(vp, NULL, KAUTH_VNODE_EXECUTE | KAUTH_VNODE_READ_DATA, ctx)) != 0) {
                error = LOAD_PROTECT;
                goto bad1;
        }
@@ -1621,41 +2747,58 @@ get_macho_vnode(
                goto bad1;
        }
 
-       if ((error = vn_rdwr(UIO_READ, vp, (caddr_t)&header, sizeof(header), 0,
+       if ((error = vn_rdwr(UIO_READ, vp, (caddr_t)header, sizeof (*header), 0,
            UIO_SYSSPACE, IO_NODELOCKED, kerncred, &resid, p)) != 0) {
                error = LOAD_IOERROR;
                goto bad2;
        }
-       
-       if (header.mach_header.magic == MH_MAGIC ||
-           header.mach_header.magic == MH_MAGIC_64)
-           is_fat = FALSE;
-       else if (header.fat_header.magic == FAT_MAGIC ||
-                header.fat_header.magic == FAT_CIGAM)
+
+       if (resid) {
+               error = LOAD_BADMACHO;
+               goto bad2;
+       }
+
+       if (header->mach_header.magic == MH_MAGIC ||
+           header->mach_header.magic == MH_MAGIC_64) {
+               is_fat = FALSE;
+       } else if (OSSwapBigToHostInt32(header->fat_header.magic) == FAT_MAGIC) {
            is_fat = TRUE;
-       else {
-           error = LOAD_BADMACHO;
-           goto bad2;
+       else {
+               error = LOAD_BADMACHO;
+               goto bad2;
        }
 
        if (is_fat) {
+
+               error = fatfile_validate_fatarches((vm_offset_t)(&header->fat_header),
+                                               sizeof(*header));
+               if (error != LOAD_SUCCESS) {
+                       goto bad2;
+               }
+
                /* Look up our architecture in the fat file. */
-               error = fatfile_getarch_with_bits(vp, archbits, (vm_offset_t)(&header.fat_header), &fat_arch);
+               error = fatfile_getarch_with_bits(archbits,
+                                               (vm_offset_t)(&header->fat_header), sizeof(*header), &fat_arch);
                if (error != LOAD_SUCCESS)
                        goto bad2;
 
                /* Read the Mach-O header out of it */
-               error = vn_rdwr(UIO_READ, vp, (caddr_t)&header.mach_header,
-                               sizeof(header.mach_header), fat_arch.offset,
-                               UIO_SYSSPACE, IO_NODELOCKED, kerncred, &resid, p);
+               error = vn_rdwr(UIO_READ, vp, (caddr_t)&header->mach_header,
+                   sizeof (header->mach_header), fat_arch.offset,
+                   UIO_SYSSPACE, IO_NODELOCKED, kerncred, &resid, p);
                if (error) {
                        error = LOAD_IOERROR;
                        goto bad2;
                }
 
+               if (resid) {
+                       error = LOAD_BADMACHO;
+                       goto bad2;
+               }
+
                /* Is this really a Mach-O? */
-               if (header.mach_header.magic != MH_MAGIC &&
-                   header.mach_header.magic != MH_MAGIC_64) {
+               if (header->mach_header.magic != MH_MAGIC &&
+                   header->mach_header.magic != MH_MAGIC_64) {
                        error = LOAD_BADMACHO;
                        goto bad2;
                }
@@ -1673,25 +2816,23 @@ get_macho_vnode(
                 * required, since the dynamic linker might work, but we will
                 * refuse to load it because of this check.
                 */
-               if ((cpu_type_t)(header.mach_header.cputype & CPU_ARCH_MASK) != archbits)
-                       return(LOAD_BADARCH);
+               if ((cpu_type_t)(header->mach_header.cputype & CPU_ARCH_MASK) != archbits) {
+                       error = LOAD_BADARCH;
+                       goto bad2;
+               }
 
                *file_offset = 0;
                *macho_size = fsize;
        }
 
-       *mach_header = header.mach_header;
+       *mach_header = header->mach_header;
        *vpp = vp;
 
        ubc_setsize(vp, fsize);
-       
        return (error);
 
 bad2:
-       err2 = VNOP_CLOSE(vp, FREAD, ctx);
-       vnode_put(vp);
-       return (error);
-
+       (void) VNOP_CLOSE(vp, FREAD, ctx);
 bad1:
        vnode_put(vp);
        return(error);