1 /* Dump Emacs in Mach-O format for use on Mac OS X.
2 Copyright (C) 2001-2012 Free Software Foundation, Inc.
4 This file is part of GNU Emacs.
6 GNU Emacs is free software: you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation, either version 3 of the License, or
9 (at your option) any later version.
11 GNU Emacs is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with GNU Emacs. If not, see <http://www.gnu.org/licenses/>. */
19 /* Contributed by Andrew Choi (akochoi@mac.com). */
21 /* Documentation note.
23 Consult the following documents/files for a description of the
24 Mach-O format: the file loader.h, man pages for Mach-O and ld, old
25 NEXTSTEP documents of the Mach-O format. The tool otool dumps the
26 mach header (-h option) and the load commands (-l option) in a
27 Mach-O file. The tool nm on Mac OS X displays the symbol table in
28 a Mach-O file. For examples of unexec for the Mach-O format, see
29 the file unexnext.c in the GNU Emacs distribution, the file
30 unexdyld.c in the Darwin port of GNU Emacs 20.7, and unexdyld.c in
31 the Darwin port of XEmacs 21.1. Also the Darwin Libc source
32 contains the source code for malloc_freezedry and malloc_jumpstart.
33 Read that to see what they do. This file was written completely
34 from scratch, making use of information from the above sources. */
36 /* The Mac OS X implementation of unexec makes use of Darwin's `zone'
37 memory allocator. All calls to malloc, realloc, and free in Emacs
38 are redirected to unexec_malloc, unexec_realloc, and unexec_free in
39 this file. When temacs is run, all memory requests are handled in
40 the zone EmacsZone. The Darwin memory allocator library calls
41 maintain the data structures to manage this zone. Dumping writes
42 its contents to data segments of the executable file. When emacs
43 is run, the loader recreates the contents of the zone in memory.
44 However since the initialization routine of the zone memory
45 allocator is run again, this `zone' can no longer be used as a
46 heap. That is why emacs uses the ordinary malloc system call to
47 allocate memory. Also, when a block of memory needs to be
48 reallocated and the new size is larger than the old one, a new
49 block must be obtained by malloc and the old contents copied to
52 /* Peculiarity of the Mach-O files generated by ld in Mac OS X
53 (possible causes of future bugs if changed).
55 The file offset of the start of the __TEXT segment is zero. Since
56 the Mach header and load commands are located at the beginning of a
57 Mach-O file, copying the contents of the __TEXT segment from the
58 input file overwrites them in the output file. Despite this,
59 unexec works fine as written below because the segment load command
60 for __TEXT appears, and is therefore processed, before all other
61 load commands except the segment load command for __PAGEZERO, which
64 Although the file offset of the start of the __TEXT segment is
65 zero, none of the sections it contains actually start there. In
66 fact, the earliest one starts a few hundred bytes beyond the end of
67 the last load command. The linker option -headerpad controls the
68 minimum size of this padding. Its setting can be changed in
69 s/darwin.h. A value of 0x690, e.g., leaves room for 30 additional
70 load commands for the newly created __DATA segments (at 56 bytes
71 each). Unexec fails if there is not enough room for these new
74 The __TEXT segment contains the sections __text, __cstring,
75 __picsymbol_stub, and __const and the __DATA segment contains the
76 sections __data, __la_symbol_ptr, __nl_symbol_ptr, __dyld, __bss,
77 and __common. The other segments do not contain any sections.
78 These sections are copied from the input file to the output file,
79 except for __data, __bss, and __common, which are dumped from
80 memory. The types of the sections __bss and __common are changed
81 from S_ZEROFILL to S_REGULAR. Note that the number of sections and
82 their relative order in the input and output files remain
83 unchanged. Otherwise all n_sect fields in the nlist records in the
84 symbol table (specified by the LC_SYMTAB load command) will have to
85 be changed accordingly.
88 /* config.h #define:s malloc/realloc/free and then includes stdlib.h.
89 We want the undefined versions, but if config.h includes stdlib.h
90 with the #define:s in place, the prototypes will be wrong and we get
91 warnings. To prevent that, include stdlib.h before config.h. */
104 #include <sys/types.h>
106 #include <mach/mach.h>
107 #include <mach-o/loader.h>
108 #include <mach-o/reloc.h>
109 #if defined (__ppc__)
110 #include <mach-o/ppc/reloc.h>
112 #ifdef HAVE_MALLOC_MALLOC_H
113 #include <malloc/malloc.h>
115 #include <objc/malloc.h>
121 #define mach_header mach_header_64
122 #define segment_command segment_command_64
123 #undef VM_REGION_BASIC_INFO_COUNT
124 #define VM_REGION_BASIC_INFO_COUNT VM_REGION_BASIC_INFO_COUNT_64
125 #undef VM_REGION_BASIC_INFO
126 #define VM_REGION_BASIC_INFO VM_REGION_BASIC_INFO_64
128 #define LC_SEGMENT LC_SEGMENT_64
129 #define vm_region vm_region_64
130 #define section section_64
132 #define MH_MAGIC MH_MAGIC_64
137 /* Size of buffer used to copy data from the input file to the output
138 file in function unexec_copy. */
139 #define UNEXEC_COPY_BUFSZ 1024
141 /* Regions with memory addresses above this value are assumed to be
142 mapped to dynamically loaded libraries and will not be dumped. */
143 #define VM_DATA_TOP (20 * 1024 * 1024)
145 /* Type of an element on the list of regions to be dumped. */
147 vm_address_t address
;
149 vm_prot_t protection
;
150 vm_prot_t max_protection
;
152 struct region_t
*next
;
155 /* Head and tail of the list of regions to be dumped. */
156 static struct region_t
*region_list_head
= 0;
157 static struct region_t
*region_list_tail
= 0;
159 /* Pointer to array of load commands. */
160 static struct load_command
**lca
;
162 /* Number of load commands. */
165 /* The highest VM address of segments loaded by the input file.
166 Regions with addresses beyond this are assumed to be allocated
167 dynamically and thus require dumping. */
168 static vm_address_t infile_lc_highest_addr
= 0;
170 /* The lowest file offset used by the all sections in the __TEXT
171 segments. This leaves room at the beginning of the file to store
172 the Mach-O header. Check this value against header size to ensure
173 the added load commands for the new __DATA segments did not
174 overwrite any of the sections in the __TEXT segment. */
175 static unsigned long text_seg_lowest_offset
= 0x10000000;
178 static struct mach_header mh
;
180 /* Offset at which the next load command should be written. */
181 static unsigned long curr_header_offset
= sizeof (struct mach_header
);
183 /* Offset at which the next segment should be written. */
184 static unsigned long curr_file_offset
= 0;
186 static unsigned long pagesize
;
187 #define ROUNDUP_TO_PAGE_BOUNDARY(x) (((x) + pagesize - 1) & ~(pagesize - 1))
189 static int infd
, outfd
;
191 static int in_dumped_exec
= 0;
193 static malloc_zone_t
*emacs_zone
;
195 /* file offset of input file's data segment */
196 static off_t data_segment_old_fileoff
= 0;
198 static struct segment_command
*data_segment_scp
;
200 /* Read N bytes from infd into memory starting at address DEST.
201 Return true if successful, false otherwise. */
203 unexec_read (void *dest
, size_t n
)
205 return n
== read (infd
, dest
, n
);
208 /* Write COUNT bytes from memory starting at address SRC to outfd
209 starting at offset DEST. Return true if successful, false
212 unexec_write (off_t dest
, const void *src
, size_t count
)
214 if (lseek (outfd
, dest
, SEEK_SET
) != dest
)
217 return write (outfd
, src
, count
) == count
;
220 /* Write COUNT bytes of zeros to outfd starting at offset DEST.
221 Return true if successful, false otherwise. */
223 unexec_write_zero (off_t dest
, size_t count
)
225 char buf
[UNEXEC_COPY_BUFSZ
];
228 memset (buf
, 0, UNEXEC_COPY_BUFSZ
);
229 if (lseek (outfd
, dest
, SEEK_SET
) != dest
)
234 bytes
= count
> UNEXEC_COPY_BUFSZ
? UNEXEC_COPY_BUFSZ
: count
;
235 if (write (outfd
, buf
, bytes
) != bytes
)
243 /* Copy COUNT bytes from starting offset SRC in infd to starting
244 offset DEST in outfd. Return true if successful, false
247 unexec_copy (off_t dest
, off_t src
, ssize_t count
)
250 ssize_t bytes_to_read
;
252 char buf
[UNEXEC_COPY_BUFSZ
];
254 if (lseek (infd
, src
, SEEK_SET
) != src
)
257 if (lseek (outfd
, dest
, SEEK_SET
) != dest
)
262 bytes_to_read
= count
> UNEXEC_COPY_BUFSZ
? UNEXEC_COPY_BUFSZ
: count
;
263 bytes_read
= read (infd
, buf
, bytes_to_read
);
266 if (write (outfd
, buf
, bytes_read
) != bytes_read
)
274 /* Debugging and informational messages routines. */
276 static _Noreturn
void
277 unexec_error (const char *format
, ...)
281 va_start (ap
, format
);
282 fprintf (stderr
, "unexec: ");
283 vfprintf (stderr
, format
, ap
);
284 fprintf (stderr
, "\n");
290 print_prot (vm_prot_t prot
)
292 if (prot
== VM_PROT_NONE
)
296 putchar (prot
& VM_PROT_READ
? 'r' : ' ');
297 putchar (prot
& VM_PROT_WRITE
? 'w' : ' ');
298 putchar (prot
& VM_PROT_EXECUTE
? 'x' : ' ');
304 print_region (vm_address_t address
, vm_size_t size
, vm_prot_t prot
,
307 printf ("%#10lx %#8lx ", (long) address
, (long) size
);
310 print_prot (max_prot
);
315 print_region_list (void)
319 printf (" address size prot maxp\n");
321 for (r
= region_list_head
; r
; r
= r
->next
)
322 print_region (r
->address
, r
->size
, r
->protection
, r
->max_protection
);
328 task_t target_task
= mach_task_self ();
329 vm_address_t address
= (vm_address_t
) 0;
331 struct vm_region_basic_info info
;
332 mach_msg_type_number_t info_count
= VM_REGION_BASIC_INFO_COUNT
;
333 mach_port_t object_name
;
335 printf (" address size prot maxp\n");
337 while (vm_region (target_task
, &address
, &size
, VM_REGION_BASIC_INFO
,
338 (vm_region_info_t
) &info
, &info_count
, &object_name
)
339 == KERN_SUCCESS
&& info_count
== VM_REGION_BASIC_INFO_COUNT
)
341 print_region (address
, size
, info
.protection
, info
.max_protection
);
343 if (object_name
!= MACH_PORT_NULL
)
344 mach_port_deallocate (target_task
, object_name
);
350 /* Build the list of regions that need to be dumped. Regions with
351 addresses above VM_DATA_TOP are omitted. Adjacent regions with
352 identical protection are merged. Note that non-writable regions
353 cannot be omitted because they some regions created at run time are
356 build_region_list (void)
358 task_t target_task
= mach_task_self ();
359 vm_address_t address
= (vm_address_t
) 0;
361 struct vm_region_basic_info info
;
362 mach_msg_type_number_t info_count
= VM_REGION_BASIC_INFO_COUNT
;
363 mach_port_t object_name
;
367 printf ("--- List of All Regions ---\n");
368 printf (" address size prot maxp\n");
371 while (vm_region (target_task
, &address
, &size
, VM_REGION_BASIC_INFO
,
372 (vm_region_info_t
) &info
, &info_count
, &object_name
)
373 == KERN_SUCCESS
&& info_count
== VM_REGION_BASIC_INFO_COUNT
)
375 /* Done when we reach addresses of shared libraries, which are
376 loaded in high memory. */
377 if (address
>= VM_DATA_TOP
)
381 print_region (address
, size
, info
.protection
, info
.max_protection
);
384 /* If a region immediately follows the previous one (the one
385 most recently added to the list) and has identical
386 protection, merge it with the latter. Otherwise create a
387 new list element for it. */
389 && info
.protection
== region_list_tail
->protection
390 && info
.max_protection
== region_list_tail
->max_protection
391 && region_list_tail
->address
+ region_list_tail
->size
== address
)
393 region_list_tail
->size
+= size
;
397 r
= malloc (sizeof *r
);
400 unexec_error ("cannot allocate region structure");
402 r
->address
= address
;
404 r
->protection
= info
.protection
;
405 r
->max_protection
= info
.max_protection
;
408 if (region_list_head
== 0)
410 region_list_head
= r
;
411 region_list_tail
= r
;
415 region_list_tail
->next
= r
;
416 region_list_tail
= r
;
419 /* Deallocate (unused) object name returned by
421 if (object_name
!= MACH_PORT_NULL
)
422 mach_port_deallocate (target_task
, object_name
);
428 printf ("--- List of Regions to be Dumped ---\n");
429 print_region_list ();
433 #define MAX_UNEXEC_REGIONS 400
435 static int num_unexec_regions
;
439 } unexec_region_info
;
440 static unexec_region_info unexec_regions
[MAX_UNEXEC_REGIONS
];
443 unexec_regions_recorder (task_t task
, void *rr
, unsigned type
,
444 vm_range_t
*ranges
, unsigned num
)
449 while (num
&& num_unexec_regions
< MAX_UNEXEC_REGIONS
)
451 /* Subtract the size of trailing null bytes from filesize. It
452 can be smaller than vmsize in segment commands. In such a
453 case, trailing bytes are initialized with zeros. */
454 for (p
= ranges
->address
+ ranges
->size
; p
> ranges
->address
; p
--)
455 if (*(((char *) p
)-1))
457 filesize
= p
- ranges
->address
;
459 unexec_regions
[num_unexec_regions
].filesize
= filesize
;
460 unexec_regions
[num_unexec_regions
++].range
= *ranges
;
461 printf ("%#10lx (sz: %#8lx/%#8lx)\n", (long) (ranges
->address
),
462 (long) filesize
, (long) (ranges
->size
));
468 unexec_reader (task_t task
, vm_address_t address
, vm_size_t size
, void **ptr
)
470 *ptr
= (void *) address
;
475 find_emacs_zone_regions (void)
477 num_unexec_regions
= 0;
479 emacs_zone
->introspect
->enumerator (mach_task_self (), 0,
480 MALLOC_PTR_REGION_RANGE_TYPE
481 | MALLOC_ADMIN_REGION_RANGE_TYPE
,
482 (vm_address_t
) emacs_zone
,
484 unexec_regions_recorder
);
486 if (num_unexec_regions
== MAX_UNEXEC_REGIONS
)
487 unexec_error ("find_emacs_zone_regions: too many regions");
491 unexec_regions_sort_compare (const void *a
, const void *b
)
493 vm_address_t aa
= ((unexec_region_info
*) a
)->range
.address
;
494 vm_address_t bb
= ((unexec_region_info
*) b
)->range
.address
;
505 unexec_regions_merge (void)
508 unexec_region_info r
;
511 qsort (unexec_regions
, num_unexec_regions
, sizeof (unexec_regions
[0]),
512 &unexec_regions_sort_compare
);
514 r
= unexec_regions
[0];
515 padsize
= r
.range
.address
& (pagesize
- 1);
518 r
.range
.address
-= padsize
;
519 r
.range
.size
+= padsize
;
520 r
.filesize
+= padsize
;
522 for (i
= 1; i
< num_unexec_regions
; i
++)
524 if (r
.range
.address
+ r
.range
.size
== unexec_regions
[i
].range
.address
525 && r
.range
.size
- r
.filesize
< 2 * pagesize
)
527 r
.filesize
= r
.range
.size
+ unexec_regions
[i
].filesize
;
528 r
.range
.size
+= unexec_regions
[i
].range
.size
;
532 unexec_regions
[n
++] = r
;
533 r
= unexec_regions
[i
];
534 padsize
= r
.range
.address
& (pagesize
- 1);
537 if ((unexec_regions
[n
-1].range
.address
538 + unexec_regions
[n
-1].range
.size
) == r
.range
.address
)
539 unexec_regions
[n
-1].range
.size
-= padsize
;
541 r
.range
.address
-= padsize
;
542 r
.range
.size
+= padsize
;
543 r
.filesize
+= padsize
;
547 unexec_regions
[n
++] = r
;
548 num_unexec_regions
= n
;
552 /* More informational messages routines. */
555 print_load_command_name (int lc
)
561 printf ("LC_SEGMENT ");
563 printf ("LC_SEGMENT_64 ");
566 case LC_LOAD_DYLINKER
:
567 printf ("LC_LOAD_DYLINKER ");
570 printf ("LC_LOAD_DYLIB ");
573 printf ("LC_SYMTAB ");
576 printf ("LC_DYSYMTAB ");
579 printf ("LC_UNIXTHREAD ");
581 case LC_PREBOUND_DYLIB
:
582 printf ("LC_PREBOUND_DYLIB");
584 case LC_TWOLEVEL_HINTS
:
585 printf ("LC_TWOLEVEL_HINTS");
594 printf ("LC_DYLD_INFO ");
596 case LC_DYLD_INFO_ONLY
:
597 printf ("LC_DYLD_INFO_ONLY");
600 #ifdef LC_VERSION_MIN_MACOSX
601 case LC_VERSION_MIN_MACOSX
:
602 printf ("LC_VERSION_MIN_MACOSX");
605 #ifdef LC_FUNCTION_STARTS
606 case LC_FUNCTION_STARTS
:
607 printf ("LC_FUNCTION_STARTS");
615 #ifdef LC_SOURCE_VERSION
616 case LC_SOURCE_VERSION
:
617 printf ("LC_SOURCE_VERSION");
620 #ifdef LC_DYLIB_CODE_SIGN_DRS
621 case LC_DYLIB_CODE_SIGN_DRS
:
622 printf ("LC_DYLIB_CODE_SIGN_DRS");
631 print_load_command (struct load_command
*lc
)
633 print_load_command_name (lc
->cmd
);
634 printf ("%8d", lc
->cmdsize
);
636 if (lc
->cmd
== LC_SEGMENT
)
638 struct segment_command
*scp
;
639 struct section
*sectp
;
642 scp
= (struct segment_command
*) lc
;
643 printf (" %-16.16s %#10lx %#8lx\n",
644 scp
->segname
, (long) (scp
->vmaddr
), (long) (scp
->vmsize
));
646 sectp
= (struct section
*) (scp
+ 1);
647 for (j
= 0; j
< scp
->nsects
; j
++)
649 printf (" %-16.16s %#10lx %#8lx\n",
650 sectp
->sectname
, (long) (sectp
->addr
), (long) (sectp
->size
));
658 /* Read header and load commands from input file. Store the latter in
659 the global array lca. Store the total number of load commands in
660 global variable nlc. */
662 read_load_commands (void)
666 if (!unexec_read (&mh
, sizeof (struct mach_header
)))
667 unexec_error ("cannot read mach-o header");
669 if (mh
.magic
!= MH_MAGIC
)
670 unexec_error ("input file not in Mach-O format");
672 if (mh
.filetype
!= MH_EXECUTE
)
673 unexec_error ("input Mach-O file is not an executable object file");
676 printf ("--- Header Information ---\n");
677 printf ("Magic = 0x%08x\n", mh
.magic
);
678 printf ("CPUType = %d\n", mh
.cputype
);
679 printf ("CPUSubType = %d\n", mh
.cpusubtype
);
680 printf ("FileType = 0x%x\n", mh
.filetype
);
681 printf ("NCmds = %d\n", mh
.ncmds
);
682 printf ("SizeOfCmds = %d\n", mh
.sizeofcmds
);
683 printf ("Flags = 0x%08x\n", mh
.flags
);
687 lca
= malloc (nlc
* sizeof *lca
);
689 for (i
= 0; i
< nlc
; i
++)
691 struct load_command lc
;
692 /* Load commands are variable-size: so read the command type and
693 size first and then read the rest. */
694 if (!unexec_read (&lc
, sizeof (struct load_command
)))
695 unexec_error ("cannot read load command");
696 lca
[i
] = malloc (lc
.cmdsize
);
697 memcpy (lca
[i
], &lc
, sizeof (struct load_command
));
698 if (!unexec_read (lca
[i
] + 1, lc
.cmdsize
- sizeof (struct load_command
)))
699 unexec_error ("cannot read content of load command");
700 if (lc
.cmd
== LC_SEGMENT
)
702 struct segment_command
*scp
= (struct segment_command
*) lca
[i
];
704 if (scp
->vmaddr
+ scp
->vmsize
> infile_lc_highest_addr
)
705 infile_lc_highest_addr
= scp
->vmaddr
+ scp
->vmsize
;
707 if (strncmp (scp
->segname
, SEG_TEXT
, 16) == 0)
709 struct section
*sectp
= (struct section
*) (scp
+ 1);
712 for (j
= 0; j
< scp
->nsects
; j
++)
713 if (sectp
->offset
< text_seg_lowest_offset
)
714 text_seg_lowest_offset
= sectp
->offset
;
719 printf ("Highest address of load commands in input file: %#8lx\n",
720 (unsigned long)infile_lc_highest_addr
);
722 printf ("Lowest offset of all sections in __TEXT segment: %#8lx\n",
723 text_seg_lowest_offset
);
725 printf ("--- List of Load Commands in Input File ---\n");
726 printf ("# cmd cmdsize name address size\n");
728 for (i
= 0; i
< nlc
; i
++)
731 print_load_command (lca
[i
]);
735 /* Copy a LC_SEGMENT load command other than the __DATA segment from
736 the input file to the output file, adjusting the file offset of the
737 segment and the file offsets of sections contained in it. */
739 copy_segment (struct load_command
*lc
)
741 struct segment_command
*scp
= (struct segment_command
*) lc
;
742 unsigned long old_fileoff
= scp
->fileoff
;
743 struct section
*sectp
;
746 scp
->fileoff
= curr_file_offset
;
748 sectp
= (struct section
*) (scp
+ 1);
749 for (j
= 0; j
< scp
->nsects
; j
++)
751 sectp
->offset
+= curr_file_offset
- old_fileoff
;
755 printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n",
756 scp
->segname
, (long) (scp
->fileoff
), (long) (scp
->filesize
),
757 (long) (scp
->vmsize
), (long) (scp
->vmaddr
));
759 if (!unexec_copy (scp
->fileoff
, old_fileoff
, scp
->filesize
))
760 unexec_error ("cannot copy segment from input to output file");
761 curr_file_offset
+= ROUNDUP_TO_PAGE_BOUNDARY (scp
->filesize
);
763 if (!unexec_write (curr_header_offset
, lc
, lc
->cmdsize
))
764 unexec_error ("cannot write load command to header");
766 curr_header_offset
+= lc
->cmdsize
;
769 /* Copy a LC_SEGMENT load command for the __DATA segment in the input
770 file to the output file. We assume that only one such segment load
771 command exists in the input file and it contains the sections
772 __data, __bss, __common, __la_symbol_ptr, __nl_symbol_ptr, and
773 __dyld. The first three of these should be dumped from memory and
774 the rest should be copied from the input file. Note that the
775 sections __bss and __common contain no data in the input file
776 because their flag fields have the value S_ZEROFILL. Dumping these
777 from memory makes it necessary to adjust file offset fields in
778 subsequently dumped load commands. Then, create new __DATA segment
779 load commands for regions on the region list other than the one
780 corresponding to the __DATA segment in the input file. */
782 copy_data_segment (struct load_command
*lc
)
784 struct segment_command
*scp
= (struct segment_command
*) lc
;
785 struct section
*sectp
;
787 unsigned long header_offset
, old_file_offset
;
789 /* The new filesize of the segment is set to its vmsize because data
790 blocks for segments must start at region boundaries. Note that
791 this may leave unused locations at the end of the segment data
792 block because the total of the sizes of all sections in the
793 segment is generally smaller than vmsize. */
794 scp
->filesize
= scp
->vmsize
;
796 printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n",
797 scp
->segname
, curr_file_offset
, (long)(scp
->filesize
),
798 (long)(scp
->vmsize
), (long) (scp
->vmaddr
));
800 /* Offsets in the output file for writing the next section structure
801 and segment data block, respectively. */
802 header_offset
= curr_header_offset
+ sizeof (struct segment_command
);
804 sectp
= (struct section
*) (scp
+ 1);
805 for (j
= 0; j
< scp
->nsects
; j
++)
807 old_file_offset
= sectp
->offset
;
808 sectp
->offset
= sectp
->addr
- scp
->vmaddr
+ curr_file_offset
;
809 /* The __data section is dumped from memory. The __bss and
810 __common sections are also dumped from memory but their flag
811 fields require changing (from S_ZEROFILL to S_REGULAR). The
812 other three kinds of sections are just copied from the input
814 if (strncmp (sectp
->sectname
, SECT_DATA
, 16) == 0)
816 extern char my_edata
[];
817 unsigned long my_size
;
819 /* The __data section is basically dumped from memory. But
820 initialized data in statically linked libraries are
821 copied from the input file. In particular,
822 add_image_hook.names and add_image_hook.pointers stored
823 by libarclite_macosx.a, are restored so that they will be
824 reinitialized when the dumped binary is executed. */
825 my_size
= (unsigned long)my_edata
- sectp
->addr
;
826 if (!(sectp
->addr
<= (unsigned long)my_edata
827 && my_size
<= sectp
->size
))
828 unexec_error ("my_edata is not in section %s", SECT_DATA
);
829 if (!unexec_write (sectp
->offset
, (void *) sectp
->addr
, my_size
))
830 unexec_error ("cannot write section %s", SECT_DATA
);
831 if (!unexec_copy (sectp
->offset
+ my_size
, old_file_offset
+ my_size
,
832 sectp
->size
- my_size
))
833 unexec_error ("cannot copy section %s", SECT_DATA
);
834 if (!unexec_write (header_offset
, sectp
, sizeof (struct section
)))
835 unexec_error ("cannot write section %s's header", SECT_DATA
);
837 else if (strncmp (sectp
->sectname
, SECT_COMMON
, 16) == 0)
839 sectp
->flags
= S_REGULAR
;
840 if (!unexec_write (sectp
->offset
, (void *) sectp
->addr
, sectp
->size
))
841 unexec_error ("cannot write section %.16s", sectp
->sectname
);
842 if (!unexec_write (header_offset
, sectp
, sizeof (struct section
)))
843 unexec_error ("cannot write section %.16s's header", sectp
->sectname
);
845 else if (strncmp (sectp
->sectname
, SECT_BSS
, 16) == 0)
847 extern char *my_endbss_static
;
848 unsigned long my_size
;
850 sectp
->flags
= S_REGULAR
;
852 /* Clear uninitialized local variables in statically linked
853 libraries. In particular, function pointers stored by
854 libSystemStub.a, which is introduced in Mac OS X 10.4 for
855 binary compatibility with respect to long double, are
856 cleared so that they will be reinitialized when the
857 dumped binary is executed on other versions of OS. */
858 my_size
= (unsigned long)my_endbss_static
- sectp
->addr
;
859 if (!(sectp
->addr
<= (unsigned long)my_endbss_static
860 && my_size
<= sectp
->size
))
861 unexec_error ("my_endbss_static is not in section %.16s",
863 if (!unexec_write (sectp
->offset
, (void *) sectp
->addr
, my_size
))
864 unexec_error ("cannot write section %.16s", sectp
->sectname
);
865 if (!unexec_write_zero (sectp
->offset
+ my_size
,
866 sectp
->size
- my_size
))
867 unexec_error ("cannot write section %.16s", sectp
->sectname
);
868 if (!unexec_write (header_offset
, sectp
, sizeof (struct section
)))
869 unexec_error ("cannot write section %.16s's header", sectp
->sectname
);
871 else if (strncmp (sectp
->sectname
, "__la_symbol_ptr", 16) == 0
872 || strncmp (sectp
->sectname
, "__nl_symbol_ptr", 16) == 0
873 || strncmp (sectp
->sectname
, "__got", 16) == 0
874 || strncmp (sectp
->sectname
, "__la_sym_ptr2", 16) == 0
875 || strncmp (sectp
->sectname
, "__dyld", 16) == 0
876 || strncmp (sectp
->sectname
, "__const", 16) == 0
877 || strncmp (sectp
->sectname
, "__cfstring", 16) == 0
878 || strncmp (sectp
->sectname
, "__gcc_except_tab", 16) == 0
879 || strncmp (sectp
->sectname
, "__program_vars", 16) == 0
880 || strncmp (sectp
->sectname
, "__mod_init_func", 16) == 0
881 || strncmp (sectp
->sectname
, "__mod_term_func", 16) == 0
882 || strncmp (sectp
->sectname
, "__objc_", 7) == 0)
884 if (!unexec_copy (sectp
->offset
, old_file_offset
, sectp
->size
))
885 unexec_error ("cannot copy section %.16s", sectp
->sectname
);
886 if (!unexec_write (header_offset
, sectp
, sizeof (struct section
)))
887 unexec_error ("cannot write section %.16s's header", sectp
->sectname
);
890 unexec_error ("unrecognized section %.16s in __DATA segment",
893 printf (" section %-16.16s at %#8lx - %#8lx (sz: %#8lx)\n",
894 sectp
->sectname
, (long) (sectp
->offset
),
895 (long) (sectp
->offset
+ sectp
->size
), (long) (sectp
->size
));
897 header_offset
+= sizeof (struct section
);
901 curr_file_offset
+= ROUNDUP_TO_PAGE_BOUNDARY (scp
->filesize
);
903 if (!unexec_write (curr_header_offset
, scp
, sizeof (struct segment_command
)))
904 unexec_error ("cannot write header of __DATA segment");
905 curr_header_offset
+= lc
->cmdsize
;
907 /* Create new __DATA segment load commands for regions on the region
908 list that do not corresponding to any segment load commands in
911 for (j
= 0; j
< num_unexec_regions
; j
++)
913 struct segment_command sc
;
916 sc
.cmdsize
= sizeof (struct segment_command
);
917 strncpy (sc
.segname
, SEG_DATA
, 16);
918 sc
.vmaddr
= unexec_regions
[j
].range
.address
;
919 sc
.vmsize
= unexec_regions
[j
].range
.size
;
920 sc
.fileoff
= curr_file_offset
;
921 sc
.filesize
= unexec_regions
[j
].filesize
;
922 sc
.maxprot
= VM_PROT_READ
| VM_PROT_WRITE
;
923 sc
.initprot
= VM_PROT_READ
| VM_PROT_WRITE
;
927 printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n",
928 sc
.segname
, (long) (sc
.fileoff
), (long) (sc
.filesize
),
929 (long) (sc
.vmsize
), (long) (sc
.vmaddr
));
931 if (!unexec_write (sc
.fileoff
, (void *) sc
.vmaddr
, sc
.filesize
))
932 unexec_error ("cannot write new __DATA segment");
933 curr_file_offset
+= ROUNDUP_TO_PAGE_BOUNDARY (sc
.filesize
);
935 if (!unexec_write (curr_header_offset
, &sc
, sc
.cmdsize
))
936 unexec_error ("cannot write new __DATA segment's header");
937 curr_header_offset
+= sc
.cmdsize
;
942 /* Copy a LC_SYMTAB load command from the input file to the output
943 file, adjusting the file offset fields. */
945 copy_symtab (struct load_command
*lc
, long delta
)
947 struct symtab_command
*stp
= (struct symtab_command
*) lc
;
949 stp
->symoff
+= delta
;
950 stp
->stroff
+= delta
;
952 printf ("Writing LC_SYMTAB command\n");
954 if (!unexec_write (curr_header_offset
, lc
, lc
->cmdsize
))
955 unexec_error ("cannot write symtab command to header");
957 curr_header_offset
+= lc
->cmdsize
;
960 /* Fix up relocation entries. */
962 unrelocate (const char *name
, off_t reloff
, int nrel
, vm_address_t base
)
964 int i
, unreloc_count
;
965 struct relocation_info reloc_info
;
966 struct scattered_relocation_info
*sc_reloc_info
967 = (struct scattered_relocation_info
*) &reloc_info
;
968 vm_address_t location
;
970 for (unreloc_count
= 0, i
= 0; i
< nrel
; i
++)
972 if (lseek (infd
, reloff
, L_SET
) != reloff
)
973 unexec_error ("unrelocate: %s:%d cannot seek to reloc_info", name
, i
);
974 if (!unexec_read (&reloc_info
, sizeof (reloc_info
)))
975 unexec_error ("unrelocate: %s:%d cannot read reloc_info", name
, i
);
976 reloff
+= sizeof (reloc_info
);
978 if (sc_reloc_info
->r_scattered
== 0)
979 switch (reloc_info
.r_type
)
981 case GENERIC_RELOC_VANILLA
:
982 location
= base
+ reloc_info
.r_address
;
983 if (location
>= data_segment_scp
->vmaddr
984 && location
< (data_segment_scp
->vmaddr
985 + data_segment_scp
->vmsize
))
987 off_t src_off
= data_segment_old_fileoff
988 + (location
- data_segment_scp
->vmaddr
);
989 off_t dst_off
= data_segment_scp
->fileoff
990 + (location
- data_segment_scp
->vmaddr
);
992 if (!unexec_copy (dst_off
, src_off
, 1 << reloc_info
.r_length
))
993 unexec_error ("unrelocate: %s:%d cannot copy original value",
999 unexec_error ("unrelocate: %s:%d cannot handle type = %d",
1000 name
, i
, reloc_info
.r_type
);
1003 switch (sc_reloc_info
->r_type
)
1005 #if defined (__ppc__)
1006 case PPC_RELOC_PB_LA_PTR
:
1007 /* nothing to do for prebound lazy pointer */
1011 unexec_error ("unrelocate: %s:%d cannot handle scattered type = %d",
1012 name
, i
, sc_reloc_info
->r_type
);
1017 printf ("Fixed up %d/%d %s relocation entries in data segment.\n",
1018 unreloc_count
, nrel
, name
);
1022 /* Rebase r_address in the relocation table. */
1024 rebase_reloc_address (off_t reloff
, int nrel
, long linkedit_delta
, long diff
)
1027 struct relocation_info reloc_info
;
1028 struct scattered_relocation_info
*sc_reloc_info
1029 = (struct scattered_relocation_info
*) &reloc_info
;
1031 for (i
= 0; i
< nrel
; i
++, reloff
+= sizeof (reloc_info
))
1033 if (lseek (infd
, reloff
- linkedit_delta
, L_SET
)
1034 != reloff
- linkedit_delta
)
1035 unexec_error ("rebase_reloc_table: cannot seek to reloc_info");
1036 if (!unexec_read (&reloc_info
, sizeof (reloc_info
)))
1037 unexec_error ("rebase_reloc_table: cannot read reloc_info");
1039 if (sc_reloc_info
->r_scattered
== 0
1040 && reloc_info
.r_type
== GENERIC_RELOC_VANILLA
)
1042 reloc_info
.r_address
-= diff
;
1043 if (!unexec_write (reloff
, &reloc_info
, sizeof (reloc_info
)))
1044 unexec_error ("rebase_reloc_table: cannot write reloc_info");
1050 /* Copy a LC_DYSYMTAB load command from the input file to the output
1051 file, adjusting the file offset fields. */
1053 copy_dysymtab (struct load_command
*lc
, long delta
)
1055 struct dysymtab_command
*dstp
= (struct dysymtab_command
*) lc
;
1064 for (i
= 0; i
< nlc
; i
++)
1065 if (lca
[i
]->cmd
== LC_SEGMENT
)
1067 struct segment_command
*scp
= (struct segment_command
*) lca
[i
];
1069 if (scp
->vmaddr
+ scp
->vmsize
> 0x100000000
1070 && (scp
->initprot
& VM_PROT_WRITE
) != 0)
1072 base
= data_segment_scp
->vmaddr
;
1078 /* First writable segment address. */
1079 base
= data_segment_scp
->vmaddr
;
1082 /* First segment address in the file (unless MH_SPLIT_SEGS set). */
1086 unrelocate ("local", dstp
->locreloff
, dstp
->nlocrel
, base
);
1087 unrelocate ("external", dstp
->extreloff
, dstp
->nextrel
, base
);
1089 if (dstp
->nextrel
> 0) {
1090 dstp
->extreloff
+= delta
;
1093 if (dstp
->nlocrel
> 0) {
1094 dstp
->locreloff
+= delta
;
1097 if (dstp
->nindirectsyms
> 0)
1098 dstp
->indirectsymoff
+= delta
;
1100 printf ("Writing LC_DYSYMTAB command\n");
1102 if (!unexec_write (curr_header_offset
, lc
, lc
->cmdsize
))
1103 unexec_error ("cannot write symtab command to header");
1105 curr_header_offset
+= lc
->cmdsize
;
1108 /* Check if the relocation base needs to be changed. */
1111 vm_address_t newbase
= 0;
1114 for (i
= 0; i
< num_unexec_regions
; i
++)
1115 if (unexec_regions
[i
].range
.address
+ unexec_regions
[i
].range
.size
1118 newbase
= data_segment_scp
->vmaddr
;
1124 rebase_reloc_address (dstp
->locreloff
, dstp
->nlocrel
, delta
, newbase
);
1125 rebase_reloc_address (dstp
->extreloff
, dstp
->nextrel
, delta
, newbase
);
1131 /* Copy a LC_TWOLEVEL_HINTS load command from the input file to the output
1132 file, adjusting the file offset fields. */
1134 copy_twolevelhints (struct load_command
*lc
, long delta
)
1136 struct twolevel_hints_command
*tlhp
= (struct twolevel_hints_command
*) lc
;
1138 if (tlhp
->nhints
> 0) {
1139 tlhp
->offset
+= delta
;
1142 printf ("Writing LC_TWOLEVEL_HINTS command\n");
1144 if (!unexec_write (curr_header_offset
, lc
, lc
->cmdsize
))
1145 unexec_error ("cannot write two level hint command to header");
1147 curr_header_offset
+= lc
->cmdsize
;
1151 /* Copy a LC_DYLD_INFO(_ONLY) load command from the input file to the output
1152 file, adjusting the file offset fields. */
1154 copy_dyld_info (struct load_command
*lc
, long delta
)
1156 struct dyld_info_command
*dip
= (struct dyld_info_command
*) lc
;
1158 if (dip
->rebase_off
> 0)
1159 dip
->rebase_off
+= delta
;
1160 if (dip
->bind_off
> 0)
1161 dip
->bind_off
+= delta
;
1162 if (dip
->weak_bind_off
> 0)
1163 dip
->weak_bind_off
+= delta
;
1164 if (dip
->lazy_bind_off
> 0)
1165 dip
->lazy_bind_off
+= delta
;
1166 if (dip
->export_off
> 0)
1167 dip
->export_off
+= delta
;
1169 printf ("Writing ");
1170 print_load_command_name (lc
->cmd
);
1171 printf (" command\n");
1173 if (!unexec_write (curr_header_offset
, lc
, lc
->cmdsize
))
1174 unexec_error ("cannot write dyld info command to header");
1176 curr_header_offset
+= lc
->cmdsize
;
1180 #ifdef LC_FUNCTION_STARTS
1181 /* Copy a LC_FUNCTION_STARTS/LC_DYLIB_CODE_SIGN_DRS load command from
1182 the input file to the output file, adjusting the data offset
1185 copy_linkedit_data (struct load_command
*lc
, long delta
)
1187 struct linkedit_data_command
*ldp
= (struct linkedit_data_command
*) lc
;
1189 if (ldp
->dataoff
> 0)
1190 ldp
->dataoff
+= delta
;
1192 printf ("Writing ");
1193 print_load_command_name (lc
->cmd
);
1194 printf (" command\n");
1196 if (!unexec_write (curr_header_offset
, lc
, lc
->cmdsize
))
1197 unexec_error ("cannot write linkedit data command to header");
1199 curr_header_offset
+= lc
->cmdsize
;
1203 /* Copy other kinds of load commands from the input file to the output
1204 file, ones that do not require adjustments of file offsets. */
1206 copy_other (struct load_command
*lc
)
1208 printf ("Writing ");
1209 print_load_command_name (lc
->cmd
);
1210 printf (" command\n");
1212 if (!unexec_write (curr_header_offset
, lc
, lc
->cmdsize
))
1213 unexec_error ("cannot write symtab command to header");
1215 curr_header_offset
+= lc
->cmdsize
;
1218 /* Loop through all load commands and dump them. Then write the Mach
1224 long linkedit_delta
= 0;
1226 printf ("--- Load Commands written to Output File ---\n");
1228 for (i
= 0; i
< nlc
; i
++)
1229 switch (lca
[i
]->cmd
)
1233 struct segment_command
*scp
= (struct segment_command
*) lca
[i
];
1234 if (strncmp (scp
->segname
, SEG_DATA
, 16) == 0)
1236 /* save data segment file offset and segment_command for
1238 if (data_segment_old_fileoff
)
1239 unexec_error ("cannot handle multiple DATA segments"
1241 data_segment_old_fileoff
= scp
->fileoff
;
1242 data_segment_scp
= scp
;
1244 copy_data_segment (lca
[i
]);
1248 if (strncmp (scp
->segname
, SEG_LINKEDIT
, 16) == 0)
1251 unexec_error ("cannot handle multiple LINKEDIT segments"
1253 linkedit_delta
= curr_file_offset
- scp
->fileoff
;
1256 copy_segment (lca
[i
]);
1261 copy_symtab (lca
[i
], linkedit_delta
);
1264 copy_dysymtab (lca
[i
], linkedit_delta
);
1266 case LC_TWOLEVEL_HINTS
:
1267 copy_twolevelhints (lca
[i
], linkedit_delta
);
1271 case LC_DYLD_INFO_ONLY
:
1272 copy_dyld_info (lca
[i
], linkedit_delta
);
1275 #ifdef LC_FUNCTION_STARTS
1276 case LC_FUNCTION_STARTS
:
1277 #ifdef LC_DYLIB_CODE_SIGN_DRS
1278 case LC_DYLIB_CODE_SIGN_DRS
:
1280 copy_linkedit_data (lca
[i
], linkedit_delta
);
1284 copy_other (lca
[i
]);
1288 if (curr_header_offset
> text_seg_lowest_offset
)
1289 unexec_error ("not enough room for load commands for new __DATA segments");
1291 printf ("%ld unused bytes follow Mach-O header\n",
1292 text_seg_lowest_offset
- curr_header_offset
);
1294 mh
.sizeofcmds
= curr_header_offset
- sizeof (struct mach_header
);
1295 if (!unexec_write (0, &mh
, sizeof (struct mach_header
)))
1296 unexec_error ("cannot write final header contents");
1299 /* Take a snapshot of Emacs and make a Mach-O format executable file
1300 from it. The file names of the output and input files are outfile
1301 and infile, respectively. The three other parameters are
1304 unexec (const char *outfile
, const char *infile
)
1307 unexec_error ("Unexec from a dumped executable is not supported.");
1309 pagesize
= getpagesize ();
1310 infd
= open (infile
, O_RDONLY
, 0);
1313 unexec_error ("cannot open input file `%s'", infile
);
1316 outfd
= open (outfile
, O_WRONLY
| O_TRUNC
| O_CREAT
, 0755);
1320 unexec_error ("cannot open output file `%s'", outfile
);
1323 build_region_list ();
1324 read_load_commands ();
1326 find_emacs_zone_regions ();
1327 unexec_regions_merge ();
1338 unexec_init_emacs_zone (void)
1340 emacs_zone
= malloc_create_zone (0, 0);
1341 malloc_set_zone_name (emacs_zone
, "EmacsZone");
1344 #ifndef MACOSX_MALLOC_MULT16
1345 #define MACOSX_MALLOC_MULT16 1
1348 typedef struct unexec_malloc_header
{
1353 } unexec_malloc_header_t
;
1355 #if MACOSX_MALLOC_MULT16
1357 #define ptr_in_unexec_regions(p) ((((vm_address_t) (p)) & 8) != 0)
1362 ptr_in_unexec_regions (void *ptr
)
1366 for (i
= 0; i
< num_unexec_regions
; i
++)
1367 if ((vm_address_t
) ptr
- unexec_regions
[i
].range
.address
1368 < unexec_regions
[i
].range
.size
)
1377 unexec_malloc (size_t size
)
1384 #if MACOSX_MALLOC_MULT16
1385 assert (((vm_address_t
) p
% 16) == 0);
1391 unexec_malloc_header_t
*ptr
;
1393 ptr
= (unexec_malloc_header_t
*)
1394 malloc_zone_malloc (emacs_zone
, size
+ sizeof (unexec_malloc_header_t
));
1397 #if MACOSX_MALLOC_MULT16
1398 assert (((vm_address_t
) ptr
% 16) == 8);
1400 return (void *) ptr
;
1405 unexec_realloc (void *old_ptr
, size_t new_size
)
1411 if (ptr_in_unexec_regions (old_ptr
))
1413 size_t old_size
= ((unexec_malloc_header_t
*) old_ptr
)[-1].u
.size
;
1414 size_t size
= new_size
> old_size
? old_size
: new_size
;
1416 p
= malloc (new_size
);
1418 memcpy (p
, old_ptr
, size
);
1422 p
= realloc (old_ptr
, new_size
);
1424 #if MACOSX_MALLOC_MULT16
1425 assert (((vm_address_t
) p
% 16) == 0);
1431 unexec_malloc_header_t
*ptr
;
1433 ptr
= (unexec_malloc_header_t
*)
1434 malloc_zone_realloc (emacs_zone
, (unexec_malloc_header_t
*) old_ptr
- 1,
1435 new_size
+ sizeof (unexec_malloc_header_t
));
1436 ptr
->u
.size
= new_size
;
1438 #if MACOSX_MALLOC_MULT16
1439 assert (((vm_address_t
) ptr
% 16) == 8);
1441 return (void *) ptr
;
1446 unexec_free (void *ptr
)
1452 if (!ptr_in_unexec_regions (ptr
))
1456 malloc_zone_free (emacs_zone
, (unexec_malloc_header_t
*) ptr
- 1);