path: root/src/unexmacosx.c

/* Dump Emacs in Mach-O format for use on macOS.
   Copyright (C) 2001-2023 Free Software Foundation, Inc.

This file is part of GNU Emacs.

GNU Emacs is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or (at
your option) any later version.

GNU Emacs is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with GNU Emacs.  If not, see <https://www.gnu.org/licenses/>.  */

/* Contributed by Andrew Choi (akochoi@mac.com).  */

/* Documentation note.

   Consult the following documents/files for a description of the
   Mach-O format: the file loader.h, man pages for Mach-O and ld, old
   NEXTSTEP documents of the Mach-O format.  The tool otool dumps the
   mach header (-h option) and the load commands (-l option) in a
   Mach-O file.  The tool nm on macOS displays the symbol table in
   a Mach-O file.  For examples of unexec for the Mach-O format, see
   the file unexnext.c in the GNU Emacs distribution, the file
   unexdyld.c in the Darwin port of GNU Emacs 20.7, and unexdyld.c in
   the Darwin port of XEmacs 21.1.  Also the Darwin Libc source
   contains the source code for malloc_freezedry and malloc_jumpstart.
   Read that to see what they do.  This file was written completely
   from scratch, making use of information from the above sources.  */

/* The macOS implementation of unexec makes use of Darwin's `zone'
   memory allocator.  All calls to malloc, realloc, and free in Emacs
   are redirected to unexec_malloc, unexec_realloc, and unexec_free in
   this file.  When temacs is run, all memory requests are handled in
   the zone EmacsZone.  The Darwin memory allocator library calls
   maintain the data structures to manage this zone.  Dumping writes
   its contents to data segments of the executable file.  When emacs
   is run, the loader recreates the contents of the zone in memory.
   However since the initialization routine of the zone memory
   allocator is run again, this `zone' can no longer be used as a
   heap.  That is why emacs uses the ordinary malloc system call to
   allocate memory.  Also, when a block of memory needs to be
   reallocated and the new size is larger than the old one, a new
   block must be obtained by malloc and the old contents copied to
   it.  */

/* Peculiarity of the Mach-O files generated by ld in macOS
   (possible causes of future bugs if changed).

   The file offset of the start of the __TEXT segment is zero.  Since
   the Mach header and load commands are located at the beginning of a
   Mach-O file, copying the contents of the __TEXT segment from the
   input file overwrites them in the output file.  Despite this,
   unexec works fine as written below because the segment load command
   for __TEXT appears, and is therefore processed, before all other
   load commands except the segment load command for __PAGEZERO, which
   remains unchanged.

   Although the file offset of the start of the __TEXT segment is
   zero, none of the sections it contains actually start there.  In
   fact, the earliest one starts a few hundred bytes beyond the end of
   the last load command.  The linker option -headerpad controls the
   minimum size of this padding.  Its setting can be changed in
   s/darwin.h.  A value of 0x690, e.g., leaves room for 30 additional
   load commands for the newly created __DATA segments (at 56 bytes
   each).  Unexec fails if there is not enough room for these new
   segments.

   The __TEXT segment contains the sections __text, __cstring,
   __picsymbol_stub, and __const and the __DATA segment contains the
   sections __data, __la_symbol_ptr, __nl_symbol_ptr, __dyld, __bss,
   and __common.  The other segments do not contain any sections.
   These sections are copied from the input file to the output file,
   except for __data, __bss, and __common, which are dumped from
   memory.  The types of the sections __bss and __common are changed
   from S_ZEROFILL to S_REGULAR.  Note that the number of sections and
   their relative order in the input and output files remain
   unchanged.  Otherwise all n_sect fields in the nlist records in the
   symbol table (specified by the LC_SYMTAB load command) will have to
   be changed accordingly.
*/

#include <config.h>

/* Although <config.h> redefines malloc to unexec_malloc, etc., this
   file wants stdlib.h to declare the originals.  */
#undef malloc
#undef realloc
#undef free

#include <stdlib.h>

#include "unexec.h"
#include "lisp.h"
#include "sysstdio.h"

#include <errno.h>
#include <fcntl.h>
#include <stdarg.h>
#include <stdint.h>
#include <sys/types.h>
#include <unistd.h>
#include <mach/mach.h>
#include <mach/vm_map.h>
#include <mach-o/loader.h>
#include <mach-o/reloc.h>
#ifdef HAVE_MALLOC_MALLOC_H
#include <malloc/malloc.h>
#else
#include <objc/malloc.h>
#endif

#include <assert.h>

/* LC_DATA_IN_CODE is not defined in mach-o/loader.h on Mac OS X 10.7.
   But it is used if we build with "Command Line Tools for Xcode 4.5
   (Mac OS X Lion) - September 2012".  */
#ifndef LC_DATA_IN_CODE
#define LC_DATA_IN_CODE 0x29 /* table of non-instructions in __text */
#endif

#ifdef _LP64
#define mach_header			mach_header_64
#define segment_command			segment_command_64
#undef  VM_REGION_BASIC_INFO_COUNT
#define VM_REGION_BASIC_INFO_COUNT	VM_REGION_BASIC_INFO_COUNT_64
#undef  VM_REGION_BASIC_INFO
#define VM_REGION_BASIC_INFO		VM_REGION_BASIC_INFO_64
#undef  LC_SEGMENT
#define LC_SEGMENT			LC_SEGMENT_64
#define vm_region			vm_region_64
#define section				section_64
#undef MH_MAGIC
#define MH_MAGIC			MH_MAGIC_64
#endif

#define VERBOSE 1

/* Size of buffer used to copy data from the input file to the output
   file in function unexec_copy.  */
#define UNEXEC_COPY_BUFSZ 1024

/* Regions with memory addresses above this value are assumed to be
   mapped to dynamically loaded libraries and will not be dumped.  */
#define VM_DATA_TOP (20 * 1024 * 1024)

/* Type of an element on the list of regions to be dumped.  */
struct region_t {
  vm_address_t address;
  vm_size_t size;
  vm_prot_t protection;
  vm_prot_t max_protection;

  struct region_t *next;
};

/* Head and tail of the list of regions to be dumped.  */
static struct region_t *region_list_head = 0;
static struct region_t *region_list_tail = 0;

/* Pointer to array of load commands.  */
static struct load_command **lca;

/* Number of load commands.  */
static int nlc;

/* The highest VM address of segments loaded by the input file.
   Regions with addresses beyond this are assumed to be allocated
   dynamically and thus require dumping.  */
static vm_address_t infile_lc_highest_addr = 0;

/* The lowest file offset used by the all sections in the __TEXT
   segments.  This leaves room at the beginning of the file to store
   the Mach-O header.  Check this value against header size to ensure
   the added load commands for the new __DATA segments did not
   overwrite any of the sections in the __TEXT segment.  */
static unsigned long text_seg_lowest_offset = 0x10000000;

/* Mach header.  */
static struct mach_header mh;

/* Offset at which the next load command should be written.  */
static unsigned long curr_header_offset = sizeof (struct mach_header);

/* Offset at which the next segment should be written.  */
static unsigned long curr_file_offset = 0;

static unsigned long pagesize;
#define ROUNDUP_TO_PAGE_BOUNDARY(x)	(((x) + pagesize - 1) & ~(pagesize - 1))

static int infd, outfd;

static int in_dumped_exec = 0;

static malloc_zone_t *emacs_zone;

/* file offset of input file's data segment */
static off_t data_segment_old_fileoff = 0;

static struct segment_command *data_segment_scp;

/* Read N bytes from infd into memory starting at address DEST.
   Return true if successful, false otherwise.  */
static int
unexec_read (void *dest, size_t n)
{
  return n == read (infd, dest, n);
}

/* Write COUNT bytes from memory starting at address SRC to outfd
   starting at offset DEST.  Return true if successful, false
   otherwise.  */
static int
unexec_write (off_t dest, const void *src, size_t count)
{
  task_t task = mach_task_self();
  if (task == MACH_PORT_NULL || task == MACH_PORT_DEAD)
    return false;

  if (lseek (outfd, dest, SEEK_SET) != dest)
    return 0;

  /* We use the Mach virtual memory API to read our process memory
     because using src directly would be undefined behavior and fails
     under Address Sanitizer.  */
  bool success = false;
  vm_offset_t data;
  mach_msg_type_number_t data_count;
  if (vm_read (task, (uintptr_t) src, count, &data, &data_count)
      == KERN_SUCCESS)
    {
      success =
        write (outfd, (const void *) (uintptr_t) data, data_count) == count;
      vm_deallocate (task, data, data_count);
    }
  return success;
}

/* Write COUNT bytes of zeros to outfd starting at offset DEST.
   Return true if successful, false otherwise.  */
static int
unexec_write_zero (off_t dest, size_t count)
{
  char buf[UNEXEC_COPY_BUFSZ];
  ssize_t bytes;

  memset (buf, 0, UNEXEC_COPY_BUFSZ);
  if (lseek (outfd, dest, SEEK_SET) != dest)
    return 0;

  while (count > 0)
    {
      bytes = count > UNEXEC_COPY_BUFSZ ? UNEXEC_COPY_BUFSZ : count;
      if (write (outfd, buf, bytes) != bytes)
	return 0;
      count -= bytes;
    }

  return 1;
}

/* Copy COUNT bytes from starting offset SRC in infd to starting
   offset DEST in outfd.  Return true if successful, false
   otherwise.  */
static int
unexec_copy (off_t dest, off_t src, ssize_t count)
{
  ssize_t bytes_read;
  ssize_t bytes_to_read;

  char buf[UNEXEC_COPY_BUFSZ];

  if (lseek (infd, src, SEEK_SET) != src)
    return 0;

  if (lseek (outfd, dest, SEEK_SET) != dest)
    return 0;

  while (count > 0)
    {
      bytes_to_read = count > UNEXEC_COPY_BUFSZ ? UNEXEC_COPY_BUFSZ : count;
      bytes_read = read (infd, buf, bytes_to_read);
      if (bytes_read <= 0)
	return 0;
      if (write (outfd, buf, bytes_read) != bytes_read)
	return 0;
      count -= bytes_read;
    }

  return 1;
}

/* Debugging and informational messages routines.  */

static _Noreturn void
unexec_error (const char *format, ...)
{
  va_list ap;

  va_start (ap, format);
  fputs ("unexec: ", stderr);
  vfprintf (stderr, format, ap);
  putc ('\n', stderr);
  va_end (ap);
  exit (1);
}

static void
print_prot (vm_prot_t prot)
{
  if (prot == VM_PROT_NONE)
    printf ("none");
  else
    {
      putchar (prot & VM_PROT_READ ? 'r' : ' ');
      putchar (prot & VM_PROT_WRITE ? 'w' : ' ');
      putchar (prot & VM_PROT_EXECUTE ? 'x' : ' ');
      putchar (' ');
    }
}

static void
print_region (vm_address_t address, vm_size_t size, vm_prot_t prot,
	      vm_prot_t max_prot)
{
  printf ("%#10lx %#8lx ", (long) address, (long) size);
  print_prot (prot);
  putchar (' ');
  print_prot (max_prot);
  putchar ('\n');
}

static void
print_region_list (void)
{
  struct region_t *r;

  printf ("   address     size prot maxp\n");

  for (r = region_list_head; r; r = r->next)
    print_region (r->address, r->size, r->protection, r->max_protection);
}

/* Build the list of regions that need to be dumped.  Regions with
   addresses above VM_DATA_TOP are omitted.  Adjacent regions with
   identical protection are merged.  Note that non-writable regions
   cannot be omitted because they some regions created at run time are
   read-only.  */
static void
build_region_list (void)
{
  task_t target_task = mach_task_self ();
  vm_address_t address = (vm_address_t) 0;
  vm_size_t size;
  struct vm_region_basic_info info;
  mach_msg_type_number_t info_count = VM_REGION_BASIC_INFO_COUNT;
  mach_port_t object_name;
  struct region_t *r;

#if VERBOSE
  printf ("--- List of All Regions ---\n");
  printf ("   address     size prot maxp\n");
#endif

  while (vm_region (target_task, &address, &size, VM_REGION_BASIC_INFO,
		    (vm_region_info_t) &info, &info_count, &object_name)
	 == KERN_SUCCESS && info_count == VM_REGION_BASIC_INFO_COUNT)
    {
      /* Done when we reach addresses of shared libraries, which are
	 loaded in high memory.  */
      if (address >= VM_DATA_TOP)
	break;

#if VERBOSE
      print_region (address, size, info.protection, info.max_protection);
#endif

      /* If a region immediately follows the previous one (the one
	 most recently added to the list) and has identical
	 protection, merge it with the latter.  Otherwise create a
	 new list element for it.  */
      if (region_list_tail
	  && info.protection == region_list_tail->protection
	  && info.max_protection == region_list_tail->max_protection
	  && region_list_tail->address + region_list_tail->size == address)
	{
	  region_list_tail->size += size;
	}
      else
	{
	  r = malloc (sizeof *r);

	  if (!r)
	    unexec_error ("cannot allocate region structure");

	  r->address = address;
	  r->size = size;
	  r->protection = info.protection;
	  r->max_protection = info.max_protection;

	  r->next = 0;
	  if (region_list_head == 0)
	    {
	      region_list_head = r;
	      region_list_tail = r;
	    }
	  else
	    {
	      region_list_tail->next = r;
	      region_list_tail = r;
	    }

	  /* Deallocate (unused) object name returned by
	     vm_region.  */
	  if (object_name != MACH_PORT_NULL)
	    mach_port_deallocate (target_task, object_name);
	}

      address += size;
    }

  printf ("--- List of Regions to be Dumped ---\n");
  print_region_list ();
}


#define MAX_UNEXEC_REGIONS 400

static int num_unexec_regions;
typedef struct {
  vm_range_t range;
  vm_size_t filesize;
} unexec_region_info;
static unexec_region_info unexec_regions[MAX_UNEXEC_REGIONS];

static void
unexec_regions_recorder (task_t task, void *rr, unsigned type,
			 vm_range_t *ranges, unsigned num)
{
  vm_address_t p;
  vm_size_t filesize;

  while (num && num_unexec_regions < MAX_UNEXEC_REGIONS)
    {
      /* Subtract the size of trailing null bytes from filesize.  It
	 can be smaller than vmsize in segment commands.  In such a
	 case, trailing bytes are initialized with zeros.  */
      for (p = ranges->address + ranges->size; p > ranges->address; p--)
      	if (*(((char *) p)-1))
      	  break;
      filesize = p - ranges->address;

      unexec_regions[num_unexec_regions].filesize = filesize;
      unexec_regions[num_unexec_regions++].range = *ranges;
      printf ("%#10lx (sz: %#8lx/%#8lx)\n", (long) (ranges->address),
	      (long) filesize, (long) (ranges->size));
      ranges++; num--;
    }
}

static kern_return_t
unexec_reader (task_t task, vm_address_t address, vm_size_t size, void **ptr)
{
  *ptr = (void *) address;
  return KERN_SUCCESS;
}

static void
find_emacs_zone_regions (void)
{
  num_unexec_regions = 0;

  emacs_zone->introspect->enumerator (mach_task_self (), 0,
				      MALLOC_PTR_REGION_RANGE_TYPE
				      | MALLOC_ADMIN_REGION_RANGE_TYPE,
				      (vm_address_t) emacs_zone,
				      unexec_reader,
				      unexec_regions_recorder);

  if (num_unexec_regions == MAX_UNEXEC_REGIONS)
    unexec_error ("find_emacs_zone_regions: too many regions");
}

static int
unexec_regions_sort_compare (const void *a, const void *b)
{
  vm_address_t aa = ((unexec_region_info *) a)->range.address;
  vm_address_t bb = ((unexec_region_info *) b)->range.address;

  if (aa < bb)
    return -1;
  else if (aa > bb)
    return 1;
  else
    return 0;
}

static void
unexec_regions_merge (void)
{
  qsort (unexec_regions, num_unexec_regions, sizeof (unexec_regions[0]),
	 &unexec_regions_sort_compare);

  /* Align each region start address to a page boundary.  */
  for (unexec_region_info *cur = unexec_regions;
       cur < unexec_regions + num_unexec_regions; cur++)
    {
      vm_size_t padsize = cur->range.address & (pagesize - 1);
      if (padsize)
	{
	  cur->range.address -= padsize;
	  cur->range.size += padsize;
	  cur->filesize += padsize;

	  unexec_region_info *prev = cur == unexec_regions ? NULL : cur - 1;
	  if (prev
	      && prev->range.address + prev->range.size > cur->range.address)
	    {
	      prev->range.size = cur->range.address - prev->range.address;
	      if (prev->filesize > prev->range.size)
		prev->filesize = prev->range.size;
	    }
	}
    }

  int n = 0;
  unexec_region_info r = unexec_regions[0];
  for (int i = 1; i < num_unexec_regions; i++)
    {
      if (r.range.address + r.range.size == unexec_regions[i].range.address
	  && r.range.size - r.filesize < 2 * pagesize)
	{
	  r.filesize = r.range.size + unexec_regions[i].filesize;
	  r.range.size += unexec_regions[i].range.size;
	}
      else
	{
	  unexec_regions[n++] = r;
	  r = unexec_regions[i];
	}
    }
  unexec_regions[n++] = r;
  num_unexec_regions = n;
}


/* More informational messages routines.  */

static void
print_load_command_name (int lc)
{
  switch (lc)
    {
    case LC_SEGMENT:
#ifndef _LP64
      printf ("LC_SEGMENT       ");
#else
      printf ("LC_SEGMENT_64    ");
#endif
      break;
    case LC_LOAD_DYLINKER:
      printf ("LC_LOAD_DYLINKER ");
      break;
    case LC_LOAD_DYLIB:
      printf ("LC_LOAD_DYLIB    ");
      break;
    case LC_SYMTAB:
      printf ("LC_SYMTAB        ");
      break;
    case LC_DYSYMTAB:
      printf ("LC_DYSYMTAB      ");
      break;
    case LC_UNIXTHREAD:
      printf ("LC_UNIXTHREAD    ");
      break;
    case LC_PREBOUND_DYLIB:
      printf ("LC_PREBOUND_DYLIB");
      break;
    case LC_TWOLEVEL_HINTS:
      printf ("LC_TWOLEVEL_HINTS");
      break;
#ifdef LC_UUID
    case LC_UUID:
      printf ("LC_UUID          ");
      break;
#endif
#ifdef LC_DYLD_INFO
    case LC_DYLD_INFO:
      printf ("LC_DYLD_INFO     ");
      break;
    case LC_DYLD_INFO_ONLY:
      printf ("LC_DYLD_INFO_ONLY");
      break;
#endif
#ifdef LC_VERSION_MIN_MACOSX
    case LC_VERSION_MIN_MACOSX:
      printf ("LC_VERSION_MIN_MACOSX");
      break;
#endif
#ifdef LC_FUNCTION_STARTS
    case LC_FUNCTION_STARTS:
      printf ("LC_FUNCTION_STARTS");
      break;
#endif
#ifdef LC_MAIN
    case LC_MAIN:
      printf ("LC_MAIN          ");
      break;
#endif
#ifdef LC_DATA_IN_CODE
    case LC_DATA_IN_CODE:
      printf ("LC_DATA_IN_CODE  ");
      break;
#endif
#ifdef LC_SOURCE_VERSION
    case LC_SOURCE_VERSION:
      printf ("LC_SOURCE_VERSION");
      break;
#endif
#ifdef LC_DYLIB_CODE_SIGN_DRS
    case LC_DYLIB_CODE_SIGN_DRS:
      printf ("LC_DYLIB_CODE_SIGN_DRS");
      break;
#endif
    default:
      printf ("unknown          ");
    }
}

static void
print_load_command (struct load_command *lc)
{
  print_load_command_name (lc->cmd);
  printf ("%8d", lc->cmdsize);

  if (lc->cmd == LC_SEGMENT)
    {
      struct segment_command *scp;
      struct section *sectp;
      int j;

      scp = (struct segment_command *) lc;
      printf (" %-16.16s %#10lx %#8lx\n",
	      scp->segname, (long) (scp->vmaddr), (long) (scp->vmsize));

      sectp = (struct section *) (scp + 1);
      for (j = 0; j < scp->nsects; j++)
	{
	  printf ("                           %-16.16s %#10lx %#8lx\n",
		  sectp->sectname, (long) (sectp->addr), (long) (sectp->size));
	  sectp++;
	}
    }
  else
    printf ("\n");
}

/* Read header and load commands from input file.  Store the latter in
   the global array lca.  Store the total number of load commands in
   global variable nlc.  */
static void
read_load_commands (void)
{
  int i;

  if (!unexec_read (&mh, sizeof (struct mach_header)))
    unexec_error ("cannot read mach-o header");

  if (mh.magic != MH_MAGIC)
    unexec_error ("input file not in Mach-O format");

  if (mh.filetype != MH_EXECUTE)
    unexec_error ("input Mach-O file is not an executable object file");

#if VERBOSE
  printf ("--- Header Information ---\n");
  printf ("Magic = 0x%08x\n", mh.magic);
  printf ("CPUType = %d\n", mh.cputype);
  printf ("CPUSubType = %d\n", mh.cpusubtype);
  printf ("FileType = 0x%x\n", mh.filetype);
  printf ("NCmds = %d\n", mh.ncmds);
  printf ("SizeOfCmds = %d\n", mh.sizeofcmds);
  printf ("Flags = 0x%08x\n", mh.flags);
#endif

  nlc = mh.ncmds;
  lca = malloc (nlc * sizeof *lca);

  for (i = 0; i < nlc; i++)
    {
      struct load_command lc;
      /* Load commands are variable-size: so read the command type and
	 size first and then read the rest.  */
      if (!unexec_read (&lc, sizeof (struct load_command)))
        unexec_error ("cannot read load command");
      lca[i] = malloc (lc.cmdsize);
      memcpy (lca[i], &lc, sizeof (struct load_command));
      if (!unexec_read (lca[i] + 1, lc.cmdsize - sizeof (struct load_command)))
        unexec_error ("cannot read content of load command");
      if (lc.cmd == LC_SEGMENT)
	{
	  struct segment_command *scp = (struct segment_command *) lca[i];

	  if (scp->vmaddr + scp->vmsize > infile_lc_highest_addr)
	    infile_lc_highest_addr = scp->vmaddr + scp->vmsize;

	  if (strncmp (scp->segname, SEG_TEXT, 16) == 0)
	    {
	      struct section *sectp = (struct section *) (scp + 1);
	      int j;

	      for (j = 0; j < scp->nsects; j++)
		if (sectp->offset < text_seg_lowest_offset)
		  text_seg_lowest_offset = sectp->offset;
	    }
	}
    }

  printf ("Highest address of load commands in input file: %#8lx\n",
	  (unsigned long)infile_lc_highest_addr);

  printf ("Lowest offset of all sections in __TEXT segment: %#8lx\n",
	  text_seg_lowest_offset);

  printf ("--- List of Load Commands in Input File ---\n");
  printf ("# cmd              cmdsize name                address     size\n");

  for (i = 0; i < nlc; i++)
    {
      printf ("%1d ", i);
      print_load_command (lca[i]);
    }
}

/* Copy a LC_SEGMENT load command other than the __DATA segment from
   the input file to the output file, adjusting the file offset of the
   segment and the file offsets of sections contained in it.  */
static void
copy_segment (struct load_command *lc)
{
  struct segment_command *scp = (struct segment_command *) lc;
  unsigned long old_fileoff = scp->fileoff;
  struct section *sectp;
  int j;

  scp->fileoff = curr_file_offset;

  sectp = (struct section *) (scp + 1);
  for (j = 0; j < scp->nsects; j++)
    {
      sectp->offset += curr_file_offset - old_fileoff;
      sectp++;
    }

  printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n",
	  scp->segname, (long) (scp->fileoff), (long) (scp->filesize),
	  (long) (scp->vmsize), (long) (scp->vmaddr));

  if (!unexec_copy (scp->fileoff, old_fileoff, scp->filesize))
    unexec_error ("cannot copy segment from input to output file");
  curr_file_offset += ROUNDUP_TO_PAGE_BOUNDARY (scp->filesize);

  if (!unexec_write (curr_header_offset, lc, lc->cmdsize))
    unexec_error ("cannot write load command to header");

  curr_header_offset += lc->cmdsize;
}

/* Copy a LC_SEGMENT load command for the __DATA segment in the input
   file to the output file.  We assume that only one such segment load
   command exists in the input file and it contains the sections
   __data, __bss, __common, __la_symbol_ptr, __nl_symbol_ptr, and
   __dyld.  The first three of these should be dumped from memory and
   the rest should be copied from the input file.  Note that the
   sections __bss and __common contain no data in the input file
   because their flag fields have the value S_ZEROFILL.  Dumping these
   from memory makes it necessary to adjust file offset fields in
   subsequently dumped load commands.  Then, create new __DATA segment
   load commands for regions on the region list other than the one
   corresponding to the __DATA segment in the input file.  */
static void
copy_data_segment (struct load_command *lc)
{
  struct segment_command *scp = (struct segment_command *) lc;
  struct section *sectp;
  int j;
  unsigned long header_offset, old_file_offset;

  /* The new filesize of the segment is set to its vmsize because data
     blocks for segments must start at region boundaries.  Note that
     this may leave unused locations at the end of the segment data
     block because the total of the sizes of all sections in the
     segment is generally smaller than vmsize.  */
  scp->filesize = scp->vmsize;

  printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n",
	  scp->segname, curr_file_offset, (long)(scp->filesize),
	  (long)(scp->vmsize), (long) (scp->vmaddr));

  /* Offsets in the output file for writing the next section structure
     and segment data block, respectively.  */
  header_offset = curr_header_offset + sizeof (struct segment_command);

  sectp = (struct section *) (scp + 1);
  for (j = 0; j < scp->nsects; j++)
    {
      old_file_offset = sectp->offset;
      sectp->offset = sectp->addr - scp->vmaddr + curr_file_offset;
      /* The __data section is dumped from memory.  The __bss and
	 __common sections are also dumped from memory but their flag
	 fields require changing (from S_ZEROFILL to S_REGULAR).  The
	 other three kinds of sections are just copied from the input
	 file.  */
      if (strncmp (sectp->sectname, SECT_DATA, 16) == 0)
	{
	  unsigned long my_size;

	  /* The __data section is basically dumped from memory.  But
	     initialized data in statically linked libraries are
	     copied from the input file.  In particular,
	     add_image_hook.names and add_image_hook.pointers stored
	     by libarclite_macosx.a, are restored so that they will be
	     reinitialized when the dumped binary is executed.  */
	  my_size = (unsigned long)my_edata - sectp->addr;
	  if (!(sectp->addr <= (unsigned long)my_edata
		&& my_size <= sectp->size))
	    unexec_error ("my_edata is not in section %s", SECT_DATA);
	  if (!unexec_write (sectp->offset, (void *) sectp->addr, my_size))
	    unexec_error ("cannot write section %s", SECT_DATA);
	  if (!unexec_copy (sectp->offset + my_size, old_file_offset + my_size,
			    sectp->size - my_size))
	    unexec_error ("cannot copy section %s", SECT_DATA);
	  if (!unexec_write (header_offset, sectp, sizeof (struct section)))
	    unexec_error ("cannot write section %s's header", SECT_DATA);
	}
      else if (strncmp (sectp->sectname, SECT_COMMON, 16) == 0)
	{
	  sectp->flags = S_REGULAR;
	  if (!unexec_write (sectp->offset, (void *) sectp->addr, sectp->size))
	    unexec_error ("cannot write section %.16s", sectp->sectname);
	  if (!unexec_write (header_offset, sectp, sizeof (struct section)))
	    unexec_error ("cannot write section %.16s's header", sectp->sectname);
	}
      else if (strncmp (sectp->sectname, SECT_BSS, 16) == 0)
	{
	  unsigned long my_size;

	  sectp->flags = S_REGULAR;

	  /* Clear uninitialized local variables in statically linked
	     libraries.  In particular, function pointers stored by
	     libSystemStub.a, which is introduced in Mac OS X 10.4 for
	     binary compatibility with respect to long double, are
	     cleared so that they will be reinitialized when the
	     dumped binary is executed on other versions of OS.  */
	  my_size = (unsigned long)my_endbss_static - sectp->addr;
	  if (!(sectp->addr <= (unsigned long)my_endbss_static
		&& my_size <= sectp->size))
	    unexec_error ("my_endbss_static is not in section %.16s",
			  sectp->sectname);
	  if (!unexec_write (sectp->offset, (void *) sectp->addr, my_size))
	    unexec_error ("cannot write section %.16s", sectp->sectname);
	  if (!unexec_write_zero (sectp->offset + my_size,
				  sectp->size - my_size))
	    unexec_error ("cannot write section %.16s", sectp->sectname);
	  if (!unexec_write (header_offset, sectp, sizeof (struct section)))
	    unexec_error ("cannot write section %.16s's header", sectp->sectname);
	}
      else if (strncmp (sectp->sectname, "__bss", 5) == 0
	       || strncmp (sectp->sectname, "__pu_bss", 8) == 0)
	{
	  sectp->flags = S_REGULAR;

	  /* These sections are produced by GCC 4.6+.

	     FIXME: We possibly ought to clear uninitialized local
	     variables in statically linked libraries like for
	     SECT_BSS (__bss) above, but setting up the markers we
	     need in lastfile.c would be rather messy. See
	     darwin_output_aligned_bss () in gcc/config/darwin.c for
	     the root of the problem, keeping in mind that the
	     sections are numbered by their alignment in GCC 4.6, but
	     by log2(alignment) in GCC 4.7. */

	  if (!unexec_write (sectp->offset, (void *) sectp->addr, sectp->size))
	    unexec_error ("cannot copy section %.16s", sectp->sectname);
	  if (!unexec_write (header_offset, sectp, sizeof (struct section)))
	    unexec_error ("cannot write section %.16s's header", sectp->sectname);
	}
      else if (strncmp (sectp->sectname, "__la_symbol_ptr", 16) == 0
	       || strncmp (sectp->sectname, "__nl_symbol_ptr", 16) == 0
	       || strncmp (sectp->sectname, "__got", 16) == 0
	       || strncmp (sectp->sectname, "__la_sym_ptr2", 16) == 0
	       || strncmp (sectp->sectname, "__dyld", 16) == 0
	       || strncmp (sectp->sectname, "__const", 16) == 0
	       || strncmp (sectp->sectname, "__cfstring", 16) == 0
	       || strncmp (sectp->sectname, "__gcc_except_tab", 16) == 0
	       || strncmp (sectp->sectname, "__program_vars", 16) == 0
	       || strncmp (sectp->sectname, "__mod_init_func", 16) == 0
	       || strncmp (sectp->sectname, "__mod_term_func", 16) == 0
	       || strncmp (sectp->sectname, "__static_data", 16) == 0
	       || strncmp (sectp->sectname, "__objc_", 7) == 0)
	{
	  if (!unexec_copy (sectp->offset, old_file_offset, sectp->size))
	    unexec_error ("cannot copy section %.16s", sectp->sectname);
	  if (!unexec_write (header_offset, sectp, sizeof (struct section)))
	    unexec_error ("cannot write section %.16s's header", sectp->sectname);
	}
      else
	unexec_error ("unrecognized section %.16s in __DATA segment",
		      sectp->sectname);

      printf ("        section %-16.16s at %#8lx - %#8lx (sz: %#8lx)\n",
	      sectp->sectname, (long) (sectp->offset),
	      (long) (sectp->offset + sectp->size), (long) (sectp->size));

      header_offset += sizeof (struct section);
      sectp++;
    }

  curr_file_offset += ROUNDUP_TO_PAGE_BOUNDARY (scp->filesize);

  if (!unexec_write (curr_header_offset, scp, sizeof (struct segment_command)))
    unexec_error ("cannot write header of __DATA segment");
  curr_header_offset += lc->cmdsize;

  /* Create new __DATA segment load commands for regions on the region
     list that do not corresponding to any segment load commands in
     the input file.
  */
  for (j = 0; j < num_unexec_regions; j++)
    {
      struct segment_command sc;

      sc.cmd = LC_SEGMENT;
      sc.cmdsize = sizeof (struct segment_command);
      strncpy (sc.segname, SEG_DATA, 16);
      sc.vmaddr = unexec_regions[j].range.address;
      sc.vmsize = unexec_regions[j].range.size;
      sc.fileoff = curr_file_offset;
      sc.filesize = unexec_regions[j].filesize;
      sc.maxprot = VM_PROT_READ | VM_PROT_WRITE;
      sc.initprot = VM_PROT_READ | VM_PROT_WRITE;
      sc.nsects = 0;
      sc.flags = 0;

      printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n",
	      sc.segname, (long) (sc.fileoff), (long) (sc.filesize),
	      (long) (sc.vmsize), (long) (sc.vmaddr));

      if (!unexec_write (sc.fileoff, (void *) sc.vmaddr, sc.filesize))
	unexec_error ("cannot write new __DATA segment");
      curr_file_offset += ROUNDUP_TO_PAGE_BOUNDARY (sc.filesize);

      if (!unexec_write (curr_header_offset, &sc, sc.cmdsize))
	unexec_error ("cannot write new __DATA segment's header");
      curr_header_offset += sc.cmdsize;
      mh.ncmds++;
    }
}

/* Copy a LC_SYMTAB load command from the input file to the output
   file, adjusting the file offset fields.  */
static void
copy_symtab (struct load_command *lc, long delta)
{
  struct symtab_command *stp = (struct symtab_command *) lc;

  stp->symoff += delta;
  stp->stroff += delta;

  printf ("Writing LC_SYMTAB command\n");

  if (!unexec_write (curr_header_offset, lc, lc->cmdsize))
    unexec_error ("cannot write symtab command to header");

  curr_header_offset += lc->cmdsize;
}

/* Fix up relocation entries. */
static void
unrelocate (const char *name, off_t reloff, int nrel, vm_address_t base)
{
  int i, unreloc_count;
  struct relocation_info reloc_info;
  struct scattered_relocation_info *sc_reloc_info
    = (struct scattered_relocation_info *) &reloc_info;
  vm_address_t location;

  for (unreloc_count = 0, i = 0; i < nrel; i++)
    {
      if (lseek (infd, reloff, L_SET) != reloff)
	unexec_error ("unrelocate: %s:%d cannot seek to reloc_info", name, i);
      if (!unexec_read (&reloc_info, sizeof (reloc_info)))
	unexec_error ("unrelocate: %s:%d cannot read reloc_info", name, i);
      reloff += sizeof (reloc_info);

      if (sc_reloc_info->r_scattered == 0)
	switch (reloc_info.r_type)
	  {
	  case GENERIC_RELOC_VANILLA:
	    location = base + reloc_info.r_address;
	    if (location >= data_segment_scp->vmaddr
		&& location < (data_segment_scp->vmaddr
			       + data_segment_scp->vmsize))
	      {
		off_t src_off = data_segment_old_fileoff
		  + (location - data_segment_scp->vmaddr);
		off_t dst_off = data_segment_scp->fileoff
		  + (location - data_segment_scp->vmaddr);

		if (!unexec_copy (dst_off, src_off, 1 << reloc_info.r_length))
		  unexec_error ("unrelocate: %s:%d cannot copy original value",
				name, i);
		unreloc_count++;
	      }
	    break;
	  default:
	    unexec_error ("unrelocate: %s:%d cannot handle type = %d",
			  name, i, reloc_info.r_type);
	  }
      else
        unexec_error ("unrelocate: %s:%d cannot handle scattered type = %d",
                      name, i, sc_reloc_info->r_type);
    }

  if (nrel > 0)
    printf ("Fixed up %d/%d %s relocation entries in data segment.\n",
	    unreloc_count, nrel, name);
}

/* Copy a LC_DYSYMTAB load command from the input file to the output
   file, adjusting the file offset fields.  */
static void
copy_dysymtab (struct load_command *lc, long delta)
{
  struct dysymtab_command *dstp = (struct dysymtab_command *) lc;
  vm_address_t base;

#ifdef _LP64
  /* First writable segment address.  */
  base = data_segment_scp->vmaddr;
#else
  /* First segment address in the file (unless MH_SPLIT_SEGS set). */
  base = 0;
#endif

  unrelocate ("local", dstp->locreloff, dstp->nlocrel, base);
  unrelocate ("external", dstp->extreloff, dstp->nextrel, base);

  if (dstp->nextrel > 0) {
    dstp->extreloff += delta;
  }

  if (dstp->nlocrel > 0) {
    dstp->locreloff += delta;
  }

  if (dstp->nindirectsyms > 0)
    dstp->indirectsymoff += delta;

  printf ("Writing LC_DYSYMTAB command\n");

  if (!unexec_write (curr_header_offset, lc, lc->cmdsize))
    unexec_error ("cannot write symtab command to header");

  curr_header_offset += lc->cmdsize;
}

/* Copy a LC_TWOLEVEL_HINTS load command from the input file to the output
   file, adjusting the file offset fields.  */
static void
copy_twolevelhints (struct load_command *lc, long delta)
{
  struct twolevel_hints_command *tlhp = (struct twolevel_hints_command *) lc;

  if (tlhp->nhints > 0) {
    tlhp->offset += delta;
  }

  printf ("Writing LC_TWOLEVEL_HINTS command\n");

  if (!unexec_write (curr_header_offset, lc, lc->cmdsize))
    unexec_error ("cannot write two level hint command to header");

  curr_header_offset += lc->cmdsize;
}

#ifdef LC_DYLD_INFO
/* Copy a LC_DYLD_INFO(_ONLY) load command from the input file to the output
   file, adjusting the file offset fields.  */
static void
copy_dyld_info (struct load_command *lc, long delta)
{
  struct dyld_info_command *dip = (struct dyld_info_command *) lc;

  if (dip->rebase_off > 0)
    dip->rebase_off += delta;
  if (dip->bind_off > 0)
    dip->bind_off += delta;
  if (dip->weak_bind_off > 0)
    dip->weak_bind_off += delta;
  if (dip->lazy_bind_off > 0)
    dip->lazy_bind_off += delta;
  if (dip->export_off > 0)
    dip->export_off += delta;

  printf ("Writing ");
  print_load_command_name (lc->cmd);
  printf (" command\n");

  if (!unexec_write (curr_header_offset, lc, lc->cmdsize))
    unexec_error ("cannot write dyld info command to header");

  curr_header_offset += lc->cmdsize;
}
#endif

#ifdef LC_FUNCTION_STARTS
/* Copy a LC_FUNCTION_STARTS/LC_DATA_IN_CODE/LC_DYLIB_CODE_SIGN_DRS
   load command from the input file to the output file, adjusting the
   data offset field.  */
static void
copy_linkedit_data (struct load_command *lc, long delta)
{
  struct linkedit_data_command *ldp = (struct linkedit_data_command *) lc;

  if (ldp->dataoff > 0)
    ldp->dataoff += delta;

  printf ("Writing ");
  print_load_command_name (lc->cmd);
  printf (" command\n");

  if (!unexec_write (curr_header_offset, lc, lc->cmdsize))
    unexec_error ("cannot write linkedit data command to header");

  curr_header_offset += lc->cmdsize;
}
#endif

/* Copy other kinds of load commands from the input file to the output
   file, ones that do not require adjustments of file offsets.  */
static void
copy_other (struct load_command *lc)
{
  printf ("Writing ");
  print_load_command_name (lc->cmd);
  printf (" command\n");

  if (!unexec_write (curr_header_offset, lc, lc->cmdsize))
    unexec_error ("cannot write symtab command to header");

  curr_header_offset += lc->cmdsize;
}

/* Loop through all load commands and dump them.  Then write the Mach
   header.  */
static void
dump_it (void)
{
  int i;
  long linkedit_delta = 0;

  printf ("--- Load Commands written to Output File ---\n");

  for (i = 0; i < nlc; i++)
    switch (lca[i]->cmd)
      {
      case LC_SEGMENT:
	{
	  struct segment_command *scp = (struct segment_command *) lca[i];
	  if (strncmp (scp->segname, SEG_DATA, 16) == 0)
	    {
	      /* save data segment file offset and segment_command for
		 unrelocate */
	      if (data_segment_old_fileoff)
		unexec_error ("cannot handle multiple DATA segments"
			      " in input file");
	      data_segment_old_fileoff = scp->fileoff;
	      data_segment_scp = scp;

	      copy_data_segment (lca[i]);
	    }
	  else
	    {
	      if (strncmp (scp->segname, SEG_LINKEDIT, 16) == 0)
		{
		  if (linkedit_delta)
		    unexec_error ("cannot handle multiple LINKEDIT segments"
				  " in input file");
		  linkedit_delta = curr_file_offset - scp->fileoff;
		}

	      copy_segment (lca[i]);
	    }
	}
	break;
      case LC_SYMTAB:
	copy_symtab (lca[i], linkedit_delta);
	break;
      case LC_DYSYMTAB:
	copy_dysymtab (lca[i], linkedit_delta);
	break;
      case LC_TWOLEVEL_HINTS:
	copy_twolevelhints (lca[i], linkedit_delta);
	break;
#ifdef LC_DYLD_INFO
      case LC_DYLD_INFO:
      case LC_DYLD_INFO_ONLY:
	copy_dyld_info (lca[i], linkedit_delta);
	break;
#endif
#ifdef LC_FUNCTION_STARTS
      case LC_FUNCTION_STARTS:
#ifdef LC_DATA_IN_CODE
      case LC_DATA_IN_CODE:
#endif
#ifdef LC_DYLIB_CODE_SIGN_DRS
      case LC_DYLIB_CODE_SIGN_DRS:
#endif
	copy_linkedit_data (lca[i], linkedit_delta);
	break;
#endif
      default:
	copy_other (lca[i]);
	break;
      }

  if (curr_header_offset > text_seg_lowest_offset)
    unexec_error ("not enough room for load commands for new __DATA segments"
		  " (increase headerpad_extra in configure.in to at least %lX)",
		  num_unexec_regions * sizeof (struct segment_command));

  printf ("%ld unused bytes follow Mach-O header\n",
	  text_seg_lowest_offset - curr_header_offset);

  mh.sizeofcmds = curr_header_offset - sizeof (struct mach_header);
  if (!unexec_write (0, &mh, sizeof (struct mach_header)))
    unexec_error ("cannot write final header contents");
}

/* Take a snapshot of Emacs and make a Mach-O format executable file
   from it.  The file names of the output and input files are outfile
   and infile, respectively.  The three other parameters are
   ignored.  */
void
unexec (const char *outfile, const char *infile)
{
  if (in_dumped_exec)
    unexec_error ("Unexec from a dumped executable is not supported.");

  pagesize = getpagesize ();
  infd = emacs_open (infile, O_RDONLY, 0);
  if (infd < 0)
    {
      unexec_error ("%s: %s", infile, strerror (errno));
    }

  outfd = emacs_open (outfile, O_WRONLY | O_TRUNC | O_CREAT, 0777);
  if (outfd < 0)
    {
      emacs_close (infd);
      unexec_error ("%s: %s", outfile, strerror (errno));
    }

  build_region_list ();
  read_load_commands ();

  find_emacs_zone_regions ();
  unexec_regions_merge ();

  in_dumped_exec = 1;

  dump_it ();

  emacs_close (outfd);
}


void
unexec_init_emacs_zone (void)
{
  emacs_zone = malloc_create_zone (0, 0);
  malloc_set_zone_name (emacs_zone, "EmacsZone");
}

#ifndef MACOSX_MALLOC_MULT16
#define MACOSX_MALLOC_MULT16 1
#endif

typedef struct unexec_malloc_header {
  union {
    char c[8];
    size_t size;
  } u;
} unexec_malloc_header_t;

#if MACOSX_MALLOC_MULT16

#define ptr_in_unexec_regions(p) ((((vm_address_t) (p)) & 8) != 0)

#else

int
ptr_in_unexec_regions (void *ptr)
{
  int i;

  for (i = 0; i < num_unexec_regions; i++)
    if ((vm_address_t) ptr - unexec_regions[i].range.address
	< unexec_regions[i].range.size)
      return 1;

  return 0;
}

#endif

void *
unexec_malloc (size_t size)
{
  if (in_dumped_exec)
    {
      void *p;

      p = malloc (size);
#if MACOSX_MALLOC_MULT16
      assert (((vm_address_t) p % 16) == 0);
#endif
      return p;
    }
  else
    {
      unexec_malloc_header_t *ptr;

      ptr = (unexec_malloc_header_t *)
	malloc_zone_malloc (emacs_zone, size + sizeof (unexec_malloc_header_t));
      ptr->u.size = size;
      ptr++;
#if MACOSX_MALLOC_MULT16
      assert (((vm_address_t) ptr % 16) == 8);
#endif
      return (void *) ptr;
    }
}

void *
unexec_realloc (void *old_ptr, size_t new_size)
{
  if (in_dumped_exec)
    {
      void *p;

      if (ptr_in_unexec_regions (old_ptr))
	{
	  size_t old_size = ((unexec_malloc_header_t *) old_ptr)[-1].u.size;
	  size_t size = new_size > old_size ? old_size : new_size;

	  p = malloc (new_size);
	  if (size)
	    memcpy (p, old_ptr, size);
	}
      else
	{
	  p = realloc (old_ptr, new_size);
	}
#if MACOSX_MALLOC_MULT16
      assert (((vm_address_t) p % 16) == 0);
#endif
      return p;
    }
  else
    {
      unexec_malloc_header_t *ptr;

      ptr = (unexec_malloc_header_t *)
	malloc_zone_realloc (emacs_zone, (unexec_malloc_header_t *) old_ptr - 1,
			     new_size + sizeof (unexec_malloc_header_t));
      ptr->u.size = new_size;
      ptr++;
#if MACOSX_MALLOC_MULT16
      assert (((vm_address_t) ptr % 16) == 8);
#endif
      return (void *) ptr;
    }
}

void
unexec_free (void *ptr)
{
  if (ptr == NULL)
    return;
  if (in_dumped_exec)
    {
      if (!ptr_in_unexec_regions (ptr))
	free (ptr);
    }
  else
    malloc_zone_free (emacs_zone, (unexec_malloc_header_t *) ptr - 1);
}