1 /* Dump Emacs in Mach-O format for use on Mac OS X.
2 Copyright (C) 2001-2014 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. */
105 #include <sys/types.h>
107 #include <mach/mach.h>
108 #include <mach-o/loader.h>
109 #include <mach-o/reloc.h>
110 #if defined (__ppc__)
111 #include <mach-o/ppc/reloc.h>
113 #ifdef HAVE_MALLOC_MALLOC_H
114 #include <malloc/malloc.h>
116 #include <objc/malloc.h>
121 /* LC_DATA_IN_CODE is not defined in mach-o/loader.h on OS X 10.7.
122 But it is used if we build with "Command Line Tools for Xcode 4.5
123 (OS X Lion) - September 2012". */
124 #ifndef LC_DATA_IN_CODE
125 #define LC_DATA_IN_CODE 0x29 /* table of non-instructions in __text */
129 #define mach_header mach_header_64
130 #define segment_command segment_command_64
131 #undef VM_REGION_BASIC_INFO_COUNT
132 #define VM_REGION_BASIC_INFO_COUNT VM_REGION_BASIC_INFO_COUNT_64
133 #undef VM_REGION_BASIC_INFO
134 #define VM_REGION_BASIC_INFO VM_REGION_BASIC_INFO_64
136 #define LC_SEGMENT LC_SEGMENT_64
137 #define vm_region vm_region_64
138 #define section section_64
140 #define MH_MAGIC MH_MAGIC_64
145 /* Size of buffer used to copy data from the input file to the output
146 file in function unexec_copy. */
147 #define UNEXEC_COPY_BUFSZ 1024
149 /* Regions with memory addresses above this value are assumed to be
150 mapped to dynamically loaded libraries and will not be dumped. */
151 #define VM_DATA_TOP (20 * 1024 * 1024)
153 /* Type of an element on the list of regions to be dumped. */
155 vm_address_t address
;
157 vm_prot_t protection
;
158 vm_prot_t max_protection
;
160 struct region_t
*next
;
163 /* Head and tail of the list of regions to be dumped. */
164 static struct region_t
*region_list_head
= 0;
165 static struct region_t
*region_list_tail
= 0;
167 /* Pointer to array of load commands. */
168 static struct load_command
**lca
;
170 /* Number of load commands. */
173 /* The highest VM address of segments loaded by the input file.
174 Regions with addresses beyond this are assumed to be allocated
175 dynamically and thus require dumping. */
176 static vm_address_t infile_lc_highest_addr
= 0;
178 /* The lowest file offset used by the all sections in the __TEXT
179 segments. This leaves room at the beginning of the file to store
180 the Mach-O header. Check this value against header size to ensure
181 the added load commands for the new __DATA segments did not
182 overwrite any of the sections in the __TEXT segment. */
183 static unsigned long text_seg_lowest_offset
= 0x10000000;
186 static struct mach_header mh
;
188 /* Offset at which the next load command should be written. */
189 static unsigned long curr_header_offset
= sizeof (struct mach_header
);
191 /* Offset at which the next segment should be written. */
192 static unsigned long curr_file_offset
= 0;
194 static unsigned long pagesize
;
195 #define ROUNDUP_TO_PAGE_BOUNDARY(x) (((x) + pagesize - 1) & ~(pagesize - 1))
197 static int infd
, outfd
;
199 static int in_dumped_exec
= 0;
201 static malloc_zone_t
*emacs_zone
;
203 /* file offset of input file's data segment */
204 static off_t data_segment_old_fileoff
= 0;
206 static struct segment_command
*data_segment_scp
;
208 /* Read N bytes from infd into memory starting at address DEST.
209 Return true if successful, false otherwise. */
211 unexec_read (void *dest
, size_t n
)
213 return n
== read (infd
, dest
, n
);
216 /* Write COUNT bytes from memory starting at address SRC to outfd
217 starting at offset DEST. Return true if successful, false
220 unexec_write (off_t dest
, const void *src
, size_t count
)
222 if (lseek (outfd
, dest
, SEEK_SET
) != dest
)
225 return write (outfd
, src
, count
) == count
;
228 /* Write COUNT bytes of zeros to outfd starting at offset DEST.
229 Return true if successful, false otherwise. */
231 unexec_write_zero (off_t dest
, size_t count
)
233 char buf
[UNEXEC_COPY_BUFSZ
];
236 memset (buf
, 0, UNEXEC_COPY_BUFSZ
);
237 if (lseek (outfd
, dest
, SEEK_SET
) != dest
)
242 bytes
= count
> UNEXEC_COPY_BUFSZ
? UNEXEC_COPY_BUFSZ
: count
;
243 if (write (outfd
, buf
, bytes
) != bytes
)
251 /* Copy COUNT bytes from starting offset SRC in infd to starting
252 offset DEST in outfd. Return true if successful, false
255 unexec_copy (off_t dest
, off_t src
, ssize_t count
)
258 ssize_t bytes_to_read
;
260 char buf
[UNEXEC_COPY_BUFSZ
];
262 if (lseek (infd
, src
, SEEK_SET
) != src
)
265 if (lseek (outfd
, dest
, SEEK_SET
) != dest
)
270 bytes_to_read
= count
> UNEXEC_COPY_BUFSZ
? UNEXEC_COPY_BUFSZ
: count
;
271 bytes_read
= read (infd
, buf
, bytes_to_read
);
274 if (write (outfd
, buf
, bytes_read
) != bytes_read
)
282 /* Debugging and informational messages routines. */
284 static _Noreturn
void
285 unexec_error (const char *format
, ...)
289 va_start (ap
, format
);
290 fprintf (stderr
, "unexec: ");
291 vfprintf (stderr
, format
, ap
);
292 fprintf (stderr
, "\n");
298 print_prot (vm_prot_t prot
)
300 if (prot
== VM_PROT_NONE
)
304 putchar (prot
& VM_PROT_READ
? 'r' : ' ');
305 putchar (prot
& VM_PROT_WRITE
? 'w' : ' ');
306 putchar (prot
& VM_PROT_EXECUTE
? 'x' : ' ');
312 print_region (vm_address_t address
, vm_size_t size
, vm_prot_t prot
,
315 printf ("%#10lx %#8lx ", (long) address
, (long) size
);
318 print_prot (max_prot
);
323 print_region_list (void)
327 printf (" address size prot maxp\n");
329 for (r
= region_list_head
; r
; r
= r
->next
)
330 print_region (r
->address
, r
->size
, r
->protection
, r
->max_protection
);
336 task_t target_task
= mach_task_self ();
337 vm_address_t address
= (vm_address_t
) 0;
339 struct vm_region_basic_info info
;
340 mach_msg_type_number_t info_count
= VM_REGION_BASIC_INFO_COUNT
;
341 mach_port_t object_name
;
343 printf (" address size prot maxp\n");
345 while (vm_region (target_task
, &address
, &size
, VM_REGION_BASIC_INFO
,
346 (vm_region_info_t
) &info
, &info_count
, &object_name
)
347 == KERN_SUCCESS
&& info_count
== VM_REGION_BASIC_INFO_COUNT
)
349 print_region (address
, size
, info
.protection
, info
.max_protection
);
351 if (object_name
!= MACH_PORT_NULL
)
352 mach_port_deallocate (target_task
, object_name
);
358 /* Build the list of regions that need to be dumped. Regions with
359 addresses above VM_DATA_TOP are omitted. Adjacent regions with
360 identical protection are merged. Note that non-writable regions
361 cannot be omitted because they some regions created at run time are
364 build_region_list (void)
366 task_t target_task
= mach_task_self ();
367 vm_address_t address
= (vm_address_t
) 0;
369 struct vm_region_basic_info info
;
370 mach_msg_type_number_t info_count
= VM_REGION_BASIC_INFO_COUNT
;
371 mach_port_t object_name
;
375 printf ("--- List of All Regions ---\n");
376 printf (" address size prot maxp\n");
379 while (vm_region (target_task
, &address
, &size
, VM_REGION_BASIC_INFO
,
380 (vm_region_info_t
) &info
, &info_count
, &object_name
)
381 == KERN_SUCCESS
&& info_count
== VM_REGION_BASIC_INFO_COUNT
)
383 /* Done when we reach addresses of shared libraries, which are
384 loaded in high memory. */
385 if (address
>= VM_DATA_TOP
)
389 print_region (address
, size
, info
.protection
, info
.max_protection
);
392 /* If a region immediately follows the previous one (the one
393 most recently added to the list) and has identical
394 protection, merge it with the latter. Otherwise create a
395 new list element for it. */
397 && info
.protection
== region_list_tail
->protection
398 && info
.max_protection
== region_list_tail
->max_protection
399 && region_list_tail
->address
+ region_list_tail
->size
== address
)
401 region_list_tail
->size
+= size
;
405 r
= malloc (sizeof *r
);
408 unexec_error ("cannot allocate region structure");
410 r
->address
= address
;
412 r
->protection
= info
.protection
;
413 r
->max_protection
= info
.max_protection
;
416 if (region_list_head
== 0)
418 region_list_head
= r
;
419 region_list_tail
= r
;
423 region_list_tail
->next
= r
;
424 region_list_tail
= r
;
427 /* Deallocate (unused) object name returned by
429 if (object_name
!= MACH_PORT_NULL
)
430 mach_port_deallocate (target_task
, object_name
);
436 printf ("--- List of Regions to be Dumped ---\n");
437 print_region_list ();
441 #define MAX_UNEXEC_REGIONS 400
443 static int num_unexec_regions
;
447 } unexec_region_info
;
448 static unexec_region_info unexec_regions
[MAX_UNEXEC_REGIONS
];
451 unexec_regions_recorder (task_t task
, void *rr
, unsigned type
,
452 vm_range_t
*ranges
, unsigned num
)
457 while (num
&& num_unexec_regions
< MAX_UNEXEC_REGIONS
)
459 /* Subtract the size of trailing null bytes from filesize. It
460 can be smaller than vmsize in segment commands. In such a
461 case, trailing bytes are initialized with zeros. */
462 for (p
= ranges
->address
+ ranges
->size
; p
> ranges
->address
; p
--)
463 if (*(((char *) p
)-1))
465 filesize
= p
- ranges
->address
;
467 unexec_regions
[num_unexec_regions
].filesize
= filesize
;
468 unexec_regions
[num_unexec_regions
++].range
= *ranges
;
469 printf ("%#10lx (sz: %#8lx/%#8lx)\n", (long) (ranges
->address
),
470 (long) filesize
, (long) (ranges
->size
));
476 unexec_reader (task_t task
, vm_address_t address
, vm_size_t size
, void **ptr
)
478 *ptr
= (void *) address
;
483 find_emacs_zone_regions (void)
485 num_unexec_regions
= 0;
487 emacs_zone
->introspect
->enumerator (mach_task_self (), 0,
488 MALLOC_PTR_REGION_RANGE_TYPE
489 | MALLOC_ADMIN_REGION_RANGE_TYPE
,
490 (vm_address_t
) emacs_zone
,
492 unexec_regions_recorder
);
494 if (num_unexec_regions
== MAX_UNEXEC_REGIONS
)
495 unexec_error ("find_emacs_zone_regions: too many regions");
499 unexec_regions_sort_compare (const void *a
, const void *b
)
501 vm_address_t aa
= ((unexec_region_info
*) a
)->range
.address
;
502 vm_address_t bb
= ((unexec_region_info
*) b
)->range
.address
;
513 unexec_regions_merge (void)
516 unexec_region_info r
;
519 qsort (unexec_regions
, num_unexec_regions
, sizeof (unexec_regions
[0]),
520 &unexec_regions_sort_compare
);
522 r
= unexec_regions
[0];
523 padsize
= r
.range
.address
& (pagesize
- 1);
526 r
.range
.address
-= padsize
;
527 r
.range
.size
+= padsize
;
528 r
.filesize
+= padsize
;
530 for (i
= 1; i
< num_unexec_regions
; i
++)
532 if (r
.range
.address
+ r
.range
.size
== unexec_regions
[i
].range
.address
533 && r
.range
.size
- r
.filesize
< 2 * pagesize
)
535 r
.filesize
= r
.range
.size
+ unexec_regions
[i
].filesize
;
536 r
.range
.size
+= unexec_regions
[i
].range
.size
;
540 unexec_regions
[n
++] = r
;
541 r
= unexec_regions
[i
];
542 padsize
= r
.range
.address
& (pagesize
- 1);
545 if ((unexec_regions
[n
-1].range
.address
546 + unexec_regions
[n
-1].range
.size
) == r
.range
.address
)
547 unexec_regions
[n
-1].range
.size
-= padsize
;
549 r
.range
.address
-= padsize
;
550 r
.range
.size
+= padsize
;
551 r
.filesize
+= padsize
;
555 unexec_regions
[n
++] = r
;
556 num_unexec_regions
= n
;
560 /* More informational messages routines. */
563 print_load_command_name (int lc
)
569 printf ("LC_SEGMENT ");
571 printf ("LC_SEGMENT_64 ");
574 case LC_LOAD_DYLINKER
:
575 printf ("LC_LOAD_DYLINKER ");
578 printf ("LC_LOAD_DYLIB ");
581 printf ("LC_SYMTAB ");
584 printf ("LC_DYSYMTAB ");
587 printf ("LC_UNIXTHREAD ");
589 case LC_PREBOUND_DYLIB
:
590 printf ("LC_PREBOUND_DYLIB");
592 case LC_TWOLEVEL_HINTS
:
593 printf ("LC_TWOLEVEL_HINTS");
602 printf ("LC_DYLD_INFO ");
604 case LC_DYLD_INFO_ONLY
:
605 printf ("LC_DYLD_INFO_ONLY");
608 #ifdef LC_VERSION_MIN_MACOSX
609 case LC_VERSION_MIN_MACOSX
:
610 printf ("LC_VERSION_MIN_MACOSX");
613 #ifdef LC_FUNCTION_STARTS
614 case LC_FUNCTION_STARTS
:
615 printf ("LC_FUNCTION_STARTS");
623 #ifdef LC_DATA_IN_CODE
624 case LC_DATA_IN_CODE
:
625 printf ("LC_DATA_IN_CODE ");
628 #ifdef LC_SOURCE_VERSION
629 case LC_SOURCE_VERSION
:
630 printf ("LC_SOURCE_VERSION");
633 #ifdef LC_DYLIB_CODE_SIGN_DRS
634 case LC_DYLIB_CODE_SIGN_DRS
:
635 printf ("LC_DYLIB_CODE_SIGN_DRS");
644 print_load_command (struct load_command
*lc
)
646 print_load_command_name (lc
->cmd
);
647 printf ("%8d", lc
->cmdsize
);
649 if (lc
->cmd
== LC_SEGMENT
)
651 struct segment_command
*scp
;
652 struct section
*sectp
;
655 scp
= (struct segment_command
*) lc
;
656 printf (" %-16.16s %#10lx %#8lx\n",
657 scp
->segname
, (long) (scp
->vmaddr
), (long) (scp
->vmsize
));
659 sectp
= (struct section
*) (scp
+ 1);
660 for (j
= 0; j
< scp
->nsects
; j
++)
662 printf (" %-16.16s %#10lx %#8lx\n",
663 sectp
->sectname
, (long) (sectp
->addr
), (long) (sectp
->size
));
671 /* Read header and load commands from input file. Store the latter in
672 the global array lca. Store the total number of load commands in
673 global variable nlc. */
675 read_load_commands (void)
679 if (!unexec_read (&mh
, sizeof (struct mach_header
)))
680 unexec_error ("cannot read mach-o header");
682 if (mh
.magic
!= MH_MAGIC
)
683 unexec_error ("input file not in Mach-O format");
685 if (mh
.filetype
!= MH_EXECUTE
)
686 unexec_error ("input Mach-O file is not an executable object file");
689 printf ("--- Header Information ---\n");
690 printf ("Magic = 0x%08x\n", mh
.magic
);
691 printf ("CPUType = %d\n", mh
.cputype
);
692 printf ("CPUSubType = %d\n", mh
.cpusubtype
);
693 printf ("FileType = 0x%x\n", mh
.filetype
);
694 printf ("NCmds = %d\n", mh
.ncmds
);
695 printf ("SizeOfCmds = %d\n", mh
.sizeofcmds
);
696 printf ("Flags = 0x%08x\n", mh
.flags
);
700 lca
= malloc (nlc
* sizeof *lca
);
702 for (i
= 0; i
< nlc
; i
++)
704 struct load_command lc
;
705 /* Load commands are variable-size: so read the command type and
706 size first and then read the rest. */
707 if (!unexec_read (&lc
, sizeof (struct load_command
)))
708 unexec_error ("cannot read load command");
709 lca
[i
] = malloc (lc
.cmdsize
);
710 memcpy (lca
[i
], &lc
, sizeof (struct load_command
));
711 if (!unexec_read (lca
[i
] + 1, lc
.cmdsize
- sizeof (struct load_command
)))
712 unexec_error ("cannot read content of load command");
713 if (lc
.cmd
== LC_SEGMENT
)
715 struct segment_command
*scp
= (struct segment_command
*) lca
[i
];
717 if (scp
->vmaddr
+ scp
->vmsize
> infile_lc_highest_addr
)
718 infile_lc_highest_addr
= scp
->vmaddr
+ scp
->vmsize
;
720 if (strncmp (scp
->segname
, SEG_TEXT
, 16) == 0)
722 struct section
*sectp
= (struct section
*) (scp
+ 1);
725 for (j
= 0; j
< scp
->nsects
; j
++)
726 if (sectp
->offset
< text_seg_lowest_offset
)
727 text_seg_lowest_offset
= sectp
->offset
;
732 printf ("Highest address of load commands in input file: %#8lx\n",
733 (unsigned long)infile_lc_highest_addr
);
735 printf ("Lowest offset of all sections in __TEXT segment: %#8lx\n",
736 text_seg_lowest_offset
);
738 printf ("--- List of Load Commands in Input File ---\n");
739 printf ("# cmd cmdsize name address size\n");
741 for (i
= 0; i
< nlc
; i
++)
744 print_load_command (lca
[i
]);
748 /* Copy a LC_SEGMENT load command other than the __DATA segment from
749 the input file to the output file, adjusting the file offset of the
750 segment and the file offsets of sections contained in it. */
752 copy_segment (struct load_command
*lc
)
754 struct segment_command
*scp
= (struct segment_command
*) lc
;
755 unsigned long old_fileoff
= scp
->fileoff
;
756 struct section
*sectp
;
759 scp
->fileoff
= curr_file_offset
;
761 sectp
= (struct section
*) (scp
+ 1);
762 for (j
= 0; j
< scp
->nsects
; j
++)
764 sectp
->offset
+= curr_file_offset
- old_fileoff
;
768 printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n",
769 scp
->segname
, (long) (scp
->fileoff
), (long) (scp
->filesize
),
770 (long) (scp
->vmsize
), (long) (scp
->vmaddr
));
772 if (!unexec_copy (scp
->fileoff
, old_fileoff
, scp
->filesize
))
773 unexec_error ("cannot copy segment from input to output file");
774 curr_file_offset
+= ROUNDUP_TO_PAGE_BOUNDARY (scp
->filesize
);
776 if (!unexec_write (curr_header_offset
, lc
, lc
->cmdsize
))
777 unexec_error ("cannot write load command to header");
779 curr_header_offset
+= lc
->cmdsize
;
782 /* Copy a LC_SEGMENT load command for the __DATA segment in the input
783 file to the output file. We assume that only one such segment load
784 command exists in the input file and it contains the sections
785 __data, __bss, __common, __la_symbol_ptr, __nl_symbol_ptr, and
786 __dyld. The first three of these should be dumped from memory and
787 the rest should be copied from the input file. Note that the
788 sections __bss and __common contain no data in the input file
789 because their flag fields have the value S_ZEROFILL. Dumping these
790 from memory makes it necessary to adjust file offset fields in
791 subsequently dumped load commands. Then, create new __DATA segment
792 load commands for regions on the region list other than the one
793 corresponding to the __DATA segment in the input file. */
795 copy_data_segment (struct load_command
*lc
)
797 struct segment_command
*scp
= (struct segment_command
*) lc
;
798 struct section
*sectp
;
800 unsigned long header_offset
, old_file_offset
;
802 /* The new filesize of the segment is set to its vmsize because data
803 blocks for segments must start at region boundaries. Note that
804 this may leave unused locations at the end of the segment data
805 block because the total of the sizes of all sections in the
806 segment is generally smaller than vmsize. */
807 scp
->filesize
= scp
->vmsize
;
809 printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n",
810 scp
->segname
, curr_file_offset
, (long)(scp
->filesize
),
811 (long)(scp
->vmsize
), (long) (scp
->vmaddr
));
813 /* Offsets in the output file for writing the next section structure
814 and segment data block, respectively. */
815 header_offset
= curr_header_offset
+ sizeof (struct segment_command
);
817 sectp
= (struct section
*) (scp
+ 1);
818 for (j
= 0; j
< scp
->nsects
; j
++)
820 old_file_offset
= sectp
->offset
;
821 sectp
->offset
= sectp
->addr
- scp
->vmaddr
+ curr_file_offset
;
822 /* The __data section is dumped from memory. The __bss and
823 __common sections are also dumped from memory but their flag
824 fields require changing (from S_ZEROFILL to S_REGULAR). The
825 other three kinds of sections are just copied from the input
827 if (strncmp (sectp
->sectname
, SECT_DATA
, 16) == 0)
829 unsigned long my_size
;
831 /* The __data section is basically dumped from memory. But
832 initialized data in statically linked libraries are
833 copied from the input file. In particular,
834 add_image_hook.names and add_image_hook.pointers stored
835 by libarclite_macosx.a, are restored so that they will be
836 reinitialized when the dumped binary is executed. */
837 my_size
= (unsigned long)my_edata
- sectp
->addr
;
838 if (!(sectp
->addr
<= (unsigned long)my_edata
839 && my_size
<= sectp
->size
))
840 unexec_error ("my_edata is not in section %s", SECT_DATA
);
841 if (!unexec_write (sectp
->offset
, (void *) sectp
->addr
, my_size
))
842 unexec_error ("cannot write section %s", SECT_DATA
);
843 if (!unexec_copy (sectp
->offset
+ my_size
, old_file_offset
+ my_size
,
844 sectp
->size
- my_size
))
845 unexec_error ("cannot copy section %s", SECT_DATA
);
846 if (!unexec_write (header_offset
, sectp
, sizeof (struct section
)))
847 unexec_error ("cannot write section %s's header", SECT_DATA
);
849 else if (strncmp (sectp
->sectname
, SECT_COMMON
, 16) == 0)
851 sectp
->flags
= S_REGULAR
;
852 if (!unexec_write (sectp
->offset
, (void *) sectp
->addr
, sectp
->size
))
853 unexec_error ("cannot write section %.16s", sectp
->sectname
);
854 if (!unexec_write (header_offset
, sectp
, sizeof (struct section
)))
855 unexec_error ("cannot write section %.16s's header", sectp
->sectname
);
857 else if (strncmp (sectp
->sectname
, SECT_BSS
, 16) == 0)
859 unsigned long my_size
;
861 sectp
->flags
= S_REGULAR
;
863 /* Clear uninitialized local variables in statically linked
864 libraries. In particular, function pointers stored by
865 libSystemStub.a, which is introduced in Mac OS X 10.4 for
866 binary compatibility with respect to long double, are
867 cleared so that they will be reinitialized when the
868 dumped binary is executed on other versions of OS. */
869 my_size
= (unsigned long)my_endbss_static
- sectp
->addr
;
870 if (!(sectp
->addr
<= (unsigned long)my_endbss_static
871 && my_size
<= sectp
->size
))
872 unexec_error ("my_endbss_static is not in section %.16s",
874 if (!unexec_write (sectp
->offset
, (void *) sectp
->addr
, my_size
))
875 unexec_error ("cannot write section %.16s", sectp
->sectname
);
876 if (!unexec_write_zero (sectp
->offset
+ my_size
,
877 sectp
->size
- my_size
))
878 unexec_error ("cannot write section %.16s", sectp
->sectname
);
879 if (!unexec_write (header_offset
, sectp
, sizeof (struct section
)))
880 unexec_error ("cannot write section %.16s's header", sectp
->sectname
);
882 else if (strncmp (sectp
->sectname
, "__bss", 5) == 0
883 || strncmp (sectp
->sectname
, "__pu_bss", 8) == 0)
885 sectp
->flags
= S_REGULAR
;
887 /* These sections are produced by GCC 4.6+.
889 FIXME: We possibly ought to clear uninitialized local
890 variables in statically linked libraries like for
891 SECT_BSS (__bss) above, but setting up the markers we
892 need in lastfile.c would be rather messy. See
893 darwin_output_aligned_bss () in gcc/config/darwin.c for
894 the root of the problem, keeping in mind that the
895 sections are numbered by their alignment in GCC 4.6, but
896 by log2(alignment) in GCC 4.7. */
898 if (!unexec_write (sectp
->offset
, (void *) sectp
->addr
, sectp
->size
))
899 unexec_error ("cannot copy section %.16s", sectp
->sectname
);
900 if (!unexec_write (header_offset
, sectp
, sizeof (struct section
)))
901 unexec_error ("cannot write section %.16s's header", sectp
->sectname
);
903 else if (strncmp (sectp
->sectname
, "__la_symbol_ptr", 16) == 0
904 || strncmp (sectp
->sectname
, "__nl_symbol_ptr", 16) == 0
905 || strncmp (sectp
->sectname
, "__got", 16) == 0
906 || strncmp (sectp
->sectname
, "__la_sym_ptr2", 16) == 0
907 || strncmp (sectp
->sectname
, "__dyld", 16) == 0
908 || strncmp (sectp
->sectname
, "__const", 16) == 0
909 || strncmp (sectp
->sectname
, "__cfstring", 16) == 0
910 || strncmp (sectp
->sectname
, "__gcc_except_tab", 16) == 0
911 || strncmp (sectp
->sectname
, "__program_vars", 16) == 0
912 || strncmp (sectp
->sectname
, "__mod_init_func", 16) == 0
913 || strncmp (sectp
->sectname
, "__mod_term_func", 16) == 0
914 || strncmp (sectp
->sectname
, "__static_data", 16) == 0
915 || strncmp (sectp
->sectname
, "__objc_", 7) == 0)
917 if (!unexec_copy (sectp
->offset
, old_file_offset
, sectp
->size
))
918 unexec_error ("cannot copy section %.16s", sectp
->sectname
);
919 if (!unexec_write (header_offset
, sectp
, sizeof (struct section
)))
920 unexec_error ("cannot write section %.16s's header", sectp
->sectname
);
923 unexec_error ("unrecognized section %.16s in __DATA segment",
926 printf (" section %-16.16s at %#8lx - %#8lx (sz: %#8lx)\n",
927 sectp
->sectname
, (long) (sectp
->offset
),
928 (long) (sectp
->offset
+ sectp
->size
), (long) (sectp
->size
));
930 header_offset
+= sizeof (struct section
);
934 curr_file_offset
+= ROUNDUP_TO_PAGE_BOUNDARY (scp
->filesize
);
936 if (!unexec_write (curr_header_offset
, scp
, sizeof (struct segment_command
)))
937 unexec_error ("cannot write header of __DATA segment");
938 curr_header_offset
+= lc
->cmdsize
;
940 /* Create new __DATA segment load commands for regions on the region
941 list that do not corresponding to any segment load commands in
944 for (j
= 0; j
< num_unexec_regions
; j
++)
946 struct segment_command sc
;
949 sc
.cmdsize
= sizeof (struct segment_command
);
950 strncpy (sc
.segname
, SEG_DATA
, 16);
951 sc
.vmaddr
= unexec_regions
[j
].range
.address
;
952 sc
.vmsize
= unexec_regions
[j
].range
.size
;
953 sc
.fileoff
= curr_file_offset
;
954 sc
.filesize
= unexec_regions
[j
].filesize
;
955 sc
.maxprot
= VM_PROT_READ
| VM_PROT_WRITE
;
956 sc
.initprot
= VM_PROT_READ
| VM_PROT_WRITE
;
960 printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n",
961 sc
.segname
, (long) (sc
.fileoff
), (long) (sc
.filesize
),
962 (long) (sc
.vmsize
), (long) (sc
.vmaddr
));
964 if (!unexec_write (sc
.fileoff
, (void *) sc
.vmaddr
, sc
.filesize
))
965 unexec_error ("cannot write new __DATA segment");
966 curr_file_offset
+= ROUNDUP_TO_PAGE_BOUNDARY (sc
.filesize
);
968 if (!unexec_write (curr_header_offset
, &sc
, sc
.cmdsize
))
969 unexec_error ("cannot write new __DATA segment's header");
970 curr_header_offset
+= sc
.cmdsize
;
975 /* Copy a LC_SYMTAB load command from the input file to the output
976 file, adjusting the file offset fields. */
978 copy_symtab (struct load_command
*lc
, long delta
)
980 struct symtab_command
*stp
= (struct symtab_command
*) lc
;
982 stp
->symoff
+= delta
;
983 stp
->stroff
+= delta
;
985 printf ("Writing LC_SYMTAB command\n");
987 if (!unexec_write (curr_header_offset
, lc
, lc
->cmdsize
))
988 unexec_error ("cannot write symtab command to header");
990 curr_header_offset
+= lc
->cmdsize
;
993 /* Fix up relocation entries. */
995 unrelocate (const char *name
, off_t reloff
, int nrel
, vm_address_t base
)
997 int i
, unreloc_count
;
998 struct relocation_info reloc_info
;
999 struct scattered_relocation_info
*sc_reloc_info
1000 = (struct scattered_relocation_info
*) &reloc_info
;
1001 vm_address_t location
;
1003 for (unreloc_count
= 0, i
= 0; i
< nrel
; i
++)
1005 if (lseek (infd
, reloff
, L_SET
) != reloff
)
1006 unexec_error ("unrelocate: %s:%d cannot seek to reloc_info", name
, i
);
1007 if (!unexec_read (&reloc_info
, sizeof (reloc_info
)))
1008 unexec_error ("unrelocate: %s:%d cannot read reloc_info", name
, i
);
1009 reloff
+= sizeof (reloc_info
);
1011 if (sc_reloc_info
->r_scattered
== 0)
1012 switch (reloc_info
.r_type
)
1014 case GENERIC_RELOC_VANILLA
:
1015 location
= base
+ reloc_info
.r_address
;
1016 if (location
>= data_segment_scp
->vmaddr
1017 && location
< (data_segment_scp
->vmaddr
1018 + data_segment_scp
->vmsize
))
1020 off_t src_off
= data_segment_old_fileoff
1021 + (location
- data_segment_scp
->vmaddr
);
1022 off_t dst_off
= data_segment_scp
->fileoff
1023 + (location
- data_segment_scp
->vmaddr
);
1025 if (!unexec_copy (dst_off
, src_off
, 1 << reloc_info
.r_length
))
1026 unexec_error ("unrelocate: %s:%d cannot copy original value",
1032 unexec_error ("unrelocate: %s:%d cannot handle type = %d",
1033 name
, i
, reloc_info
.r_type
);
1036 switch (sc_reloc_info
->r_type
)
1038 #if defined (__ppc__)
1039 case PPC_RELOC_PB_LA_PTR
:
1040 /* nothing to do for prebound lazy pointer */
1044 unexec_error ("unrelocate: %s:%d cannot handle scattered type = %d",
1045 name
, i
, sc_reloc_info
->r_type
);
1050 printf ("Fixed up %d/%d %s relocation entries in data segment.\n",
1051 unreloc_count
, nrel
, name
);
1055 /* Rebase r_address in the relocation table. */
1057 rebase_reloc_address (off_t reloff
, int nrel
, long linkedit_delta
, long diff
)
1060 struct relocation_info reloc_info
;
1061 struct scattered_relocation_info
*sc_reloc_info
1062 = (struct scattered_relocation_info
*) &reloc_info
;
1064 for (i
= 0; i
< nrel
; i
++, reloff
+= sizeof (reloc_info
))
1066 if (lseek (infd
, reloff
- linkedit_delta
, L_SET
)
1067 != reloff
- linkedit_delta
)
1068 unexec_error ("rebase_reloc_table: cannot seek to reloc_info");
1069 if (!unexec_read (&reloc_info
, sizeof (reloc_info
)))
1070 unexec_error ("rebase_reloc_table: cannot read reloc_info");
1072 if (sc_reloc_info
->r_scattered
== 0
1073 && reloc_info
.r_type
== GENERIC_RELOC_VANILLA
)
1075 reloc_info
.r_address
-= diff
;
1076 if (!unexec_write (reloff
, &reloc_info
, sizeof (reloc_info
)))
1077 unexec_error ("rebase_reloc_table: cannot write reloc_info");
1083 /* Copy a LC_DYSYMTAB load command from the input file to the output
1084 file, adjusting the file offset fields. */
1086 copy_dysymtab (struct load_command
*lc
, long delta
)
1088 struct dysymtab_command
*dstp
= (struct dysymtab_command
*) lc
;
1097 for (i
= 0; i
< nlc
; i
++)
1098 if (lca
[i
]->cmd
== LC_SEGMENT
)
1100 struct segment_command
*scp
= (struct segment_command
*) lca
[i
];
1102 if (scp
->vmaddr
+ scp
->vmsize
> 0x100000000
1103 && (scp
->initprot
& VM_PROT_WRITE
) != 0)
1105 base
= data_segment_scp
->vmaddr
;
1111 /* First writable segment address. */
1112 base
= data_segment_scp
->vmaddr
;
1115 /* First segment address in the file (unless MH_SPLIT_SEGS set). */
1119 unrelocate ("local", dstp
->locreloff
, dstp
->nlocrel
, base
);
1120 unrelocate ("external", dstp
->extreloff
, dstp
->nextrel
, base
);
1122 if (dstp
->nextrel
> 0) {
1123 dstp
->extreloff
+= delta
;
1126 if (dstp
->nlocrel
> 0) {
1127 dstp
->locreloff
+= delta
;
1130 if (dstp
->nindirectsyms
> 0)
1131 dstp
->indirectsymoff
+= delta
;
1133 printf ("Writing LC_DYSYMTAB command\n");
1135 if (!unexec_write (curr_header_offset
, lc
, lc
->cmdsize
))
1136 unexec_error ("cannot write symtab command to header");
1138 curr_header_offset
+= lc
->cmdsize
;
1141 /* Check if the relocation base needs to be changed. */
1144 vm_address_t newbase
= 0;
1147 for (i
= 0; i
< num_unexec_regions
; i
++)
1148 if (unexec_regions
[i
].range
.address
+ unexec_regions
[i
].range
.size
1151 newbase
= data_segment_scp
->vmaddr
;
1157 rebase_reloc_address (dstp
->locreloff
, dstp
->nlocrel
, delta
, newbase
);
1158 rebase_reloc_address (dstp
->extreloff
, dstp
->nextrel
, delta
, newbase
);
1164 /* Copy a LC_TWOLEVEL_HINTS load command from the input file to the output
1165 file, adjusting the file offset fields. */
1167 copy_twolevelhints (struct load_command
*lc
, long delta
)
1169 struct twolevel_hints_command
*tlhp
= (struct twolevel_hints_command
*) lc
;
1171 if (tlhp
->nhints
> 0) {
1172 tlhp
->offset
+= delta
;
1175 printf ("Writing LC_TWOLEVEL_HINTS command\n");
1177 if (!unexec_write (curr_header_offset
, lc
, lc
->cmdsize
))
1178 unexec_error ("cannot write two level hint command to header");
1180 curr_header_offset
+= lc
->cmdsize
;
1184 /* Copy a LC_DYLD_INFO(_ONLY) load command from the input file to the output
1185 file, adjusting the file offset fields. */
1187 copy_dyld_info (struct load_command
*lc
, long delta
)
1189 struct dyld_info_command
*dip
= (struct dyld_info_command
*) lc
;
1191 if (dip
->rebase_off
> 0)
1192 dip
->rebase_off
+= delta
;
1193 if (dip
->bind_off
> 0)
1194 dip
->bind_off
+= delta
;
1195 if (dip
->weak_bind_off
> 0)
1196 dip
->weak_bind_off
+= delta
;
1197 if (dip
->lazy_bind_off
> 0)
1198 dip
->lazy_bind_off
+= delta
;
1199 if (dip
->export_off
> 0)
1200 dip
->export_off
+= delta
;
1202 printf ("Writing ");
1203 print_load_command_name (lc
->cmd
);
1204 printf (" command\n");
1206 if (!unexec_write (curr_header_offset
, lc
, lc
->cmdsize
))
1207 unexec_error ("cannot write dyld info command to header");
1209 curr_header_offset
+= lc
->cmdsize
;
1213 #ifdef LC_FUNCTION_STARTS
1214 /* Copy a LC_FUNCTION_STARTS/LC_DATA_IN_CODE/LC_DYLIB_CODE_SIGN_DRS
1215 load command from the input file to the output file, adjusting the
1216 data offset field. */
1218 copy_linkedit_data (struct load_command
*lc
, long delta
)
1220 struct linkedit_data_command
*ldp
= (struct linkedit_data_command
*) lc
;
1222 if (ldp
->dataoff
> 0)
1223 ldp
->dataoff
+= delta
;
1225 printf ("Writing ");
1226 print_load_command_name (lc
->cmd
);
1227 printf (" command\n");
1229 if (!unexec_write (curr_header_offset
, lc
, lc
->cmdsize
))
1230 unexec_error ("cannot write linkedit data command to header");
1232 curr_header_offset
+= lc
->cmdsize
;
1236 /* Copy other kinds of load commands from the input file to the output
1237 file, ones that do not require adjustments of file offsets. */
1239 copy_other (struct load_command
*lc
)
1241 printf ("Writing ");
1242 print_load_command_name (lc
->cmd
);
1243 printf (" command\n");
1245 if (!unexec_write (curr_header_offset
, lc
, lc
->cmdsize
))
1246 unexec_error ("cannot write symtab command to header");
1248 curr_header_offset
+= lc
->cmdsize
;
1251 /* Loop through all load commands and dump them. Then write the Mach
1257 long linkedit_delta
= 0;
1259 printf ("--- Load Commands written to Output File ---\n");
1261 for (i
= 0; i
< nlc
; i
++)
1262 switch (lca
[i
]->cmd
)
1266 struct segment_command
*scp
= (struct segment_command
*) lca
[i
];
1267 if (strncmp (scp
->segname
, SEG_DATA
, 16) == 0)
1269 /* save data segment file offset and segment_command for
1271 if (data_segment_old_fileoff
)
1272 unexec_error ("cannot handle multiple DATA segments"
1274 data_segment_old_fileoff
= scp
->fileoff
;
1275 data_segment_scp
= scp
;
1277 copy_data_segment (lca
[i
]);
1281 if (strncmp (scp
->segname
, SEG_LINKEDIT
, 16) == 0)
1284 unexec_error ("cannot handle multiple LINKEDIT segments"
1286 linkedit_delta
= curr_file_offset
- scp
->fileoff
;
1289 copy_segment (lca
[i
]);
1294 copy_symtab (lca
[i
], linkedit_delta
);
1297 copy_dysymtab (lca
[i
], linkedit_delta
);
1299 case LC_TWOLEVEL_HINTS
:
1300 copy_twolevelhints (lca
[i
], linkedit_delta
);
1304 case LC_DYLD_INFO_ONLY
:
1305 copy_dyld_info (lca
[i
], linkedit_delta
);
1308 #ifdef LC_FUNCTION_STARTS
1309 case LC_FUNCTION_STARTS
:
1310 #ifdef LC_DATA_IN_CODE
1311 case LC_DATA_IN_CODE
:
1313 #ifdef LC_DYLIB_CODE_SIGN_DRS
1314 case LC_DYLIB_CODE_SIGN_DRS
:
1316 copy_linkedit_data (lca
[i
], linkedit_delta
);
1320 copy_other (lca
[i
]);
1324 if (curr_header_offset
> text_seg_lowest_offset
)
1325 unexec_error ("not enough room for load commands for new __DATA segments"
1326 " (increase headerpad_extra in configure.in to at least %lX)",
1327 num_unexec_regions
* sizeof (struct segment_command
));
1329 printf ("%ld unused bytes follow Mach-O header\n",
1330 text_seg_lowest_offset
- curr_header_offset
);
1332 mh
.sizeofcmds
= curr_header_offset
- sizeof (struct mach_header
);
1333 if (!unexec_write (0, &mh
, sizeof (struct mach_header
)))
1334 unexec_error ("cannot write final header contents");
1337 /* Take a snapshot of Emacs and make a Mach-O format executable file
1338 from it. The file names of the output and input files are outfile
1339 and infile, respectively. The three other parameters are
1342 unexec (const char *outfile
, const char *infile
)
1345 unexec_error ("Unexec from a dumped executable is not supported.");
1347 pagesize
= getpagesize ();
1348 infd
= emacs_open (infile
, O_RDONLY
, 0);
1351 unexec_error ("cannot open input file `%s'", infile
);
1354 outfd
= emacs_open (outfile
, O_WRONLY
| O_TRUNC
| O_CREAT
, 0755);
1358 unexec_error ("cannot open output file `%s'", outfile
);
1361 build_region_list ();
1362 read_load_commands ();
1364 find_emacs_zone_regions ();
1365 unexec_regions_merge ();
1371 emacs_close (outfd
);
1376 unexec_init_emacs_zone (void)
1378 emacs_zone
= malloc_create_zone (0, 0);
1379 malloc_set_zone_name (emacs_zone
, "EmacsZone");
1382 #ifndef MACOSX_MALLOC_MULT16
1383 #define MACOSX_MALLOC_MULT16 1
1386 typedef struct unexec_malloc_header
{
1391 } unexec_malloc_header_t
;
1393 #if MACOSX_MALLOC_MULT16
1395 #define ptr_in_unexec_regions(p) ((((vm_address_t) (p)) & 8) != 0)
1400 ptr_in_unexec_regions (void *ptr
)
1404 for (i
= 0; i
< num_unexec_regions
; i
++)
1405 if ((vm_address_t
) ptr
- unexec_regions
[i
].range
.address
1406 < unexec_regions
[i
].range
.size
)
1415 unexec_malloc (size_t size
)
1422 #if MACOSX_MALLOC_MULT16
1423 assert (((vm_address_t
) p
% 16) == 0);
1429 unexec_malloc_header_t
*ptr
;
1431 ptr
= (unexec_malloc_header_t
*)
1432 malloc_zone_malloc (emacs_zone
, size
+ sizeof (unexec_malloc_header_t
));
1435 #if MACOSX_MALLOC_MULT16
1436 assert (((vm_address_t
) ptr
% 16) == 8);
1438 return (void *) ptr
;
1443 unexec_realloc (void *old_ptr
, size_t new_size
)
1449 if (ptr_in_unexec_regions (old_ptr
))
1451 size_t old_size
= ((unexec_malloc_header_t
*) old_ptr
)[-1].u
.size
;
1452 size_t size
= new_size
> old_size
? old_size
: new_size
;
1454 p
= malloc (new_size
);
1456 memcpy (p
, old_ptr
, size
);
1460 p
= realloc (old_ptr
, new_size
);
1462 #if MACOSX_MALLOC_MULT16
1463 assert (((vm_address_t
) p
% 16) == 0);
1469 unexec_malloc_header_t
*ptr
;
1471 ptr
= (unexec_malloc_header_t
*)
1472 malloc_zone_realloc (emacs_zone
, (unexec_malloc_header_t
*) old_ptr
- 1,
1473 new_size
+ sizeof (unexec_malloc_header_t
));
1474 ptr
->u
.size
= new_size
;
1476 #if MACOSX_MALLOC_MULT16
1477 assert (((vm_address_t
) ptr
% 16) == 8);
1479 return (void *) ptr
;
1484 unexec_free (void *ptr
)
1490 if (!ptr_in_unexec_regions (ptr
))
1494 malloc_zone_free (emacs_zone
, (unexec_malloc_header_t
*) ptr
- 1);