Commit | Line | Data |
---|---|---|
e0f712ba | 1 | /* Dump Emacs in Mach-O format for use on Mac OS X. |
aaef169d | 2 | Copyright (C) 2001, 2002, 2003, 2004, 2005, |
114f9c96 | 3 | 2006, 2007, 2008, 2009, 2010 Free Software Foundation, Inc. |
e0f712ba AC |
4 | |
5 | This file is part of GNU Emacs. | |
6 | ||
9ec0b715 | 7 | GNU Emacs is free software: you can redistribute it and/or modify |
e0f712ba | 8 | it under the terms of the GNU General Public License as published by |
9ec0b715 GM |
9 | the Free Software Foundation, either version 3 of the License, or |
10 | (at your option) any later version. | |
e0f712ba AC |
11 | |
12 | GNU Emacs is distributed in the hope that it will be useful, | |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | GNU General Public License for more details. | |
16 | ||
17 | You should have received a copy of the GNU General Public License | |
9ec0b715 | 18 | along with GNU Emacs. If not, see <http://www.gnu.org/licenses/>. */ |
e0f712ba AC |
19 | |
20 | /* Contributed by Andrew Choi (akochoi@mac.com). */ | |
21 | ||
22 | /* Documentation note. | |
23 | ||
24 | Consult the following documents/files for a description of the | |
25 | Mach-O format: the file loader.h, man pages for Mach-O and ld, old | |
26 | NEXTSTEP documents of the Mach-O format. The tool otool dumps the | |
27 | mach header (-h option) and the load commands (-l option) in a | |
28 | Mach-O file. The tool nm on Mac OS X displays the symbol table in | |
29 | a Mach-O file. For examples of unexec for the Mach-O format, see | |
30 | the file unexnext.c in the GNU Emacs distribution, the file | |
31 | unexdyld.c in the Darwin port of GNU Emacs 20.7, and unexdyld.c in | |
32 | the Darwin port of XEmacs 21.1. Also the Darwin Libc source | |
33 | contains the source code for malloc_freezedry and malloc_jumpstart. | |
34 | Read that to see what they do. This file was written completely | |
35 | from scratch, making use of information from the above sources. */ | |
36 | ||
37 | /* The Mac OS X implementation of unexec makes use of Darwin's `zone' | |
38 | memory allocator. All calls to malloc, realloc, and free in Emacs | |
39 | are redirected to unexec_malloc, unexec_realloc, and unexec_free in | |
40 | this file. When temacs is run, all memory requests are handled in | |
41 | the zone EmacsZone. The Darwin memory allocator library calls | |
42 | maintain the data structures to manage this zone. Dumping writes | |
43 | its contents to data segments of the executable file. When emacs | |
44 | is run, the loader recreates the contents of the zone in memory. | |
45 | However since the initialization routine of the zone memory | |
46 | allocator is run again, this `zone' can no longer be used as a | |
47 | heap. That is why emacs uses the ordinary malloc system call to | |
48 | allocate memory. Also, when a block of memory needs to be | |
49 | reallocated and the new size is larger than the old one, a new | |
50 | block must be obtained by malloc and the old contents copied to | |
51 | it. */ | |
52 | ||
53 | /* Peculiarity of the Mach-O files generated by ld in Mac OS X | |
54 | (possible causes of future bugs if changed). | |
55 | ||
56 | The file offset of the start of the __TEXT segment is zero. Since | |
57 | the Mach header and load commands are located at the beginning of a | |
58 | Mach-O file, copying the contents of the __TEXT segment from the | |
59 | input file overwrites them in the output file. Despite this, | |
60 | unexec works fine as written below because the segment load command | |
61 | for __TEXT appears, and is therefore processed, before all other | |
62 | load commands except the segment load command for __PAGEZERO, which | |
63 | remains unchanged. | |
64 | ||
65 | Although the file offset of the start of the __TEXT segment is | |
66 | zero, none of the sections it contains actually start there. In | |
67 | fact, the earliest one starts a few hundred bytes beyond the end of | |
68 | the last load command. The linker option -headerpad controls the | |
69 | minimum size of this padding. Its setting can be changed in | |
c57038f8 YM |
70 | s/darwin.h. A value of 0x690, e.g., leaves room for 30 additional |
71 | load commands for the newly created __DATA segments (at 56 bytes | |
72 | each). Unexec fails if there is not enough room for these new | |
73 | segments. | |
e0f712ba AC |
74 | |
75 | The __TEXT segment contains the sections __text, __cstring, | |
76 | __picsymbol_stub, and __const and the __DATA segment contains the | |
77 | sections __data, __la_symbol_ptr, __nl_symbol_ptr, __dyld, __bss, | |
78 | and __common. The other segments do not contain any sections. | |
79 | These sections are copied from the input file to the output file, | |
80 | except for __data, __bss, and __common, which are dumped from | |
81 | memory. The types of the sections __bss and __common are changed | |
82 | from S_ZEROFILL to S_REGULAR. Note that the number of sections and | |
83 | their relative order in the input and output files remain | |
84 | unchanged. Otherwise all n_sect fields in the nlist records in the | |
85 | symbol table (specified by the LC_SYMTAB load command) will have to | |
86 | be changed accordingly. | |
87 | */ | |
88 | ||
89 | #include <stdio.h> | |
90 | #include <stdlib.h> | |
91 | #include <fcntl.h> | |
92 | #include <stdarg.h> | |
93 | #include <sys/types.h> | |
94 | #include <unistd.h> | |
95 | #include <mach/mach.h> | |
96 | #include <mach-o/loader.h> | |
043131c4 AC |
97 | #include <mach-o/reloc.h> |
98 | #if defined (__ppc__) | |
99 | #include <mach-o/ppc/reloc.h> | |
100 | #endif | |
7f900522 YM |
101 | #include <config.h> |
102 | #undef malloc | |
103 | #undef realloc | |
104 | #undef free | |
105 | #ifdef HAVE_MALLOC_MALLOC_H | |
f7f3a65f ST |
106 | #include <malloc/malloc.h> |
107 | #else | |
e0f712ba | 108 | #include <objc/malloc.h> |
f7f3a65f ST |
109 | #endif |
110 | ||
40ef0695 YM |
111 | #include <assert.h> |
112 | ||
73da71f9 YM |
113 | #ifdef _LP64 |
114 | #define mach_header mach_header_64 | |
115 | #define segment_command segment_command_64 | |
116 | #undef VM_REGION_BASIC_INFO_COUNT | |
117 | #define VM_REGION_BASIC_INFO_COUNT VM_REGION_BASIC_INFO_COUNT_64 | |
118 | #undef VM_REGION_BASIC_INFO | |
119 | #define VM_REGION_BASIC_INFO VM_REGION_BASIC_INFO_64 | |
120 | #undef LC_SEGMENT | |
121 | #define LC_SEGMENT LC_SEGMENT_64 | |
122 | #define vm_region vm_region_64 | |
123 | #define section section_64 | |
124 | #undef MH_MAGIC | |
125 | #define MH_MAGIC MH_MAGIC_64 | |
126 | #endif | |
e0f712ba AC |
127 | |
128 | #define VERBOSE 1 | |
129 | ||
130 | /* Size of buffer used to copy data from the input file to the output | |
131 | file in function unexec_copy. */ | |
132 | #define UNEXEC_COPY_BUFSZ 1024 | |
133 | ||
134 | /* Regions with memory addresses above this value are assumed to be | |
135 | mapped to dynamically loaded libraries and will not be dumped. */ | |
136 | #define VM_DATA_TOP (20 * 1024 * 1024) | |
137 | ||
e0f712ba AC |
138 | /* Type of an element on the list of regions to be dumped. */ |
139 | struct region_t { | |
140 | vm_address_t address; | |
141 | vm_size_t size; | |
142 | vm_prot_t protection; | |
143 | vm_prot_t max_protection; | |
144 | ||
145 | struct region_t *next; | |
146 | }; | |
147 | ||
148 | /* Head and tail of the list of regions to be dumped. */ | |
c57038f8 YM |
149 | static struct region_t *region_list_head = 0; |
150 | static struct region_t *region_list_tail = 0; | |
e0f712ba AC |
151 | |
152 | /* Pointer to array of load commands. */ | |
c57038f8 | 153 | static struct load_command **lca; |
e0f712ba AC |
154 | |
155 | /* Number of load commands. */ | |
c57038f8 | 156 | static int nlc; |
e0f712ba AC |
157 | |
158 | /* The highest VM address of segments loaded by the input file. | |
159 | Regions with addresses beyond this are assumed to be allocated | |
160 | dynamically and thus require dumping. */ | |
c57038f8 | 161 | static vm_address_t infile_lc_highest_addr = 0; |
e0f712ba AC |
162 | |
163 | /* The lowest file offset used by the all sections in the __TEXT | |
164 | segments. This leaves room at the beginning of the file to store | |
165 | the Mach-O header. Check this value against header size to ensure | |
166 | the added load commands for the new __DATA segments did not | |
167 | overwrite any of the sections in the __TEXT segment. */ | |
c57038f8 | 168 | static unsigned long text_seg_lowest_offset = 0x10000000; |
e0f712ba AC |
169 | |
170 | /* Mach header. */ | |
c57038f8 | 171 | static struct mach_header mh; |
e0f712ba AC |
172 | |
173 | /* Offset at which the next load command should be written. */ | |
c57038f8 | 174 | static unsigned long curr_header_offset = sizeof (struct mach_header); |
e0f712ba | 175 | |
73da71f9 YM |
176 | /* Offset at which the next segment should be written. */ |
177 | static unsigned long curr_file_offset = 0; | |
178 | ||
179 | static unsigned long pagesize; | |
180 | #define ROUNDUP_TO_PAGE_BOUNDARY(x) (((x) + pagesize - 1) & ~(pagesize - 1)) | |
e0f712ba | 181 | |
c57038f8 | 182 | static int infd, outfd; |
e0f712ba | 183 | |
c57038f8 | 184 | static int in_dumped_exec = 0; |
e0f712ba | 185 | |
c57038f8 | 186 | static malloc_zone_t *emacs_zone; |
e0f712ba | 187 | |
043131c4 | 188 | /* file offset of input file's data segment */ |
c57038f8 | 189 | static off_t data_segment_old_fileoff = 0; |
043131c4 | 190 | |
c57038f8 | 191 | static struct segment_command *data_segment_scp; |
043131c4 | 192 | |
433456d7 | 193 | /* Read N bytes from infd into memory starting at address DEST. |
e0f712ba AC |
194 | Return true if successful, false otherwise. */ |
195 | static int | |
196 | unexec_read (void *dest, size_t n) | |
197 | { | |
198 | return n == read (infd, dest, n); | |
199 | } | |
200 | ||
433456d7 YM |
201 | /* Write COUNT bytes from memory starting at address SRC to outfd |
202 | starting at offset DEST. Return true if successful, false | |
203 | otherwise. */ | |
e0f712ba AC |
204 | static int |
205 | unexec_write (off_t dest, const void *src, size_t count) | |
206 | { | |
207 | if (lseek (outfd, dest, SEEK_SET) != dest) | |
208 | return 0; | |
209 | ||
210 | return write (outfd, src, count) == count; | |
211 | } | |
212 | ||
433456d7 YM |
213 | /* Write COUNT bytes of zeros to outfd starting at offset DEST. |
214 | Return true if successful, false otherwise. */ | |
215 | static int | |
216 | unexec_write_zero (off_t dest, size_t count) | |
217 | { | |
218 | char buf[UNEXEC_COPY_BUFSZ]; | |
219 | ssize_t bytes; | |
220 | ||
72af86bd | 221 | memset (buf, 0, UNEXEC_COPY_BUFSZ); |
433456d7 YM |
222 | if (lseek (outfd, dest, SEEK_SET) != dest) |
223 | return 0; | |
224 | ||
225 | while (count > 0) | |
226 | { | |
227 | bytes = count > UNEXEC_COPY_BUFSZ ? UNEXEC_COPY_BUFSZ : count; | |
228 | if (write (outfd, buf, bytes) != bytes) | |
229 | return 0; | |
230 | count -= bytes; | |
231 | } | |
232 | ||
233 | return 1; | |
234 | } | |
235 | ||
236 | /* Copy COUNT bytes from starting offset SRC in infd to starting | |
237 | offset DEST in outfd. Return true if successful, false | |
238 | otherwise. */ | |
e0f712ba AC |
239 | static int |
240 | unexec_copy (off_t dest, off_t src, ssize_t count) | |
241 | { | |
242 | ssize_t bytes_read; | |
911c78b4 | 243 | ssize_t bytes_to_read; |
e0f712ba AC |
244 | |
245 | char buf[UNEXEC_COPY_BUFSZ]; | |
246 | ||
247 | if (lseek (infd, src, SEEK_SET) != src) | |
248 | return 0; | |
249 | ||
250 | if (lseek (outfd, dest, SEEK_SET) != dest) | |
251 | return 0; | |
252 | ||
253 | while (count > 0) | |
254 | { | |
911c78b4 ST |
255 | bytes_to_read = count > UNEXEC_COPY_BUFSZ ? UNEXEC_COPY_BUFSZ : count; |
256 | bytes_read = read (infd, buf, bytes_to_read); | |
e0f712ba AC |
257 | if (bytes_read <= 0) |
258 | return 0; | |
259 | if (write (outfd, buf, bytes_read) != bytes_read) | |
260 | return 0; | |
261 | count -= bytes_read; | |
262 | } | |
263 | ||
264 | return 1; | |
265 | } | |
266 | ||
267 | /* Debugging and informational messages routines. */ | |
268 | ||
269 | static void | |
7aee76f4 | 270 | unexec_error (const char *format, ...) |
e0f712ba AC |
271 | { |
272 | va_list ap; | |
273 | ||
274 | va_start (ap, format); | |
275 | fprintf (stderr, "unexec: "); | |
276 | vfprintf (stderr, format, ap); | |
277 | fprintf (stderr, "\n"); | |
278 | va_end (ap); | |
279 | exit (1); | |
280 | } | |
281 | ||
282 | static void | |
283 | print_prot (vm_prot_t prot) | |
284 | { | |
285 | if (prot == VM_PROT_NONE) | |
286 | printf ("none"); | |
287 | else | |
288 | { | |
289 | putchar (prot & VM_PROT_READ ? 'r' : ' '); | |
290 | putchar (prot & VM_PROT_WRITE ? 'w' : ' '); | |
291 | putchar (prot & VM_PROT_EXECUTE ? 'x' : ' '); | |
292 | putchar (' '); | |
293 | } | |
294 | } | |
295 | ||
296 | static void | |
297 | print_region (vm_address_t address, vm_size_t size, vm_prot_t prot, | |
298 | vm_prot_t max_prot) | |
299 | { | |
73da71f9 | 300 | printf ("%#10lx %#8lx ", (long) address, (long) size); |
e0f712ba AC |
301 | print_prot (prot); |
302 | putchar (' '); | |
303 | print_prot (max_prot); | |
304 | putchar ('\n'); | |
305 | } | |
306 | ||
307 | static void | |
3d608a86 | 308 | print_region_list (void) |
e0f712ba AC |
309 | { |
310 | struct region_t *r; | |
311 | ||
312 | printf (" address size prot maxp\n"); | |
313 | ||
314 | for (r = region_list_head; r; r = r->next) | |
315 | print_region (r->address, r->size, r->protection, r->max_protection); | |
316 | } | |
317 | ||
c57038f8 | 318 | static void |
3d608a86 | 319 | print_regions (void) |
e0f712ba AC |
320 | { |
321 | task_t target_task = mach_task_self (); | |
322 | vm_address_t address = (vm_address_t) 0; | |
323 | vm_size_t size; | |
324 | struct vm_region_basic_info info; | |
325 | mach_msg_type_number_t info_count = VM_REGION_BASIC_INFO_COUNT; | |
326 | mach_port_t object_name; | |
327 | ||
328 | printf (" address size prot maxp\n"); | |
329 | ||
330 | while (vm_region (target_task, &address, &size, VM_REGION_BASIC_INFO, | |
331 | (vm_region_info_t) &info, &info_count, &object_name) | |
332 | == KERN_SUCCESS && info_count == VM_REGION_BASIC_INFO_COUNT) | |
333 | { | |
334 | print_region (address, size, info.protection, info.max_protection); | |
335 | ||
336 | if (object_name != MACH_PORT_NULL) | |
337 | mach_port_deallocate (target_task, object_name); | |
177c0ea7 | 338 | |
e0f712ba AC |
339 | address += size; |
340 | } | |
341 | } | |
342 | ||
343 | /* Build the list of regions that need to be dumped. Regions with | |
344 | addresses above VM_DATA_TOP are omitted. Adjacent regions with | |
345 | identical protection are merged. Note that non-writable regions | |
346 | cannot be omitted because they some regions created at run time are | |
347 | read-only. */ | |
348 | static void | |
3d608a86 | 349 | build_region_list (void) |
e0f712ba AC |
350 | { |
351 | task_t target_task = mach_task_self (); | |
352 | vm_address_t address = (vm_address_t) 0; | |
353 | vm_size_t size; | |
354 | struct vm_region_basic_info info; | |
355 | mach_msg_type_number_t info_count = VM_REGION_BASIC_INFO_COUNT; | |
356 | mach_port_t object_name; | |
357 | struct region_t *r; | |
358 | ||
359 | #if VERBOSE | |
360 | printf ("--- List of All Regions ---\n"); | |
361 | printf (" address size prot maxp\n"); | |
362 | #endif | |
363 | ||
364 | while (vm_region (target_task, &address, &size, VM_REGION_BASIC_INFO, | |
365 | (vm_region_info_t) &info, &info_count, &object_name) | |
366 | == KERN_SUCCESS && info_count == VM_REGION_BASIC_INFO_COUNT) | |
367 | { | |
368 | /* Done when we reach addresses of shared libraries, which are | |
369 | loaded in high memory. */ | |
370 | if (address >= VM_DATA_TOP) | |
371 | break; | |
372 | ||
373 | #if VERBOSE | |
374 | print_region (address, size, info.protection, info.max_protection); | |
375 | #endif | |
376 | ||
377 | /* If a region immediately follows the previous one (the one | |
378 | most recently added to the list) and has identical | |
379 | protection, merge it with the latter. Otherwise create a | |
380 | new list element for it. */ | |
381 | if (region_list_tail | |
382 | && info.protection == region_list_tail->protection | |
383 | && info.max_protection == region_list_tail->max_protection | |
384 | && region_list_tail->address + region_list_tail->size == address) | |
385 | { | |
386 | region_list_tail->size += size; | |
387 | } | |
388 | else | |
389 | { | |
390 | r = (struct region_t *) malloc (sizeof (struct region_t)); | |
177c0ea7 | 391 | |
e0f712ba AC |
392 | if (!r) |
393 | unexec_error ("cannot allocate region structure"); | |
177c0ea7 | 394 | |
e0f712ba AC |
395 | r->address = address; |
396 | r->size = size; | |
397 | r->protection = info.protection; | |
398 | r->max_protection = info.max_protection; | |
177c0ea7 | 399 | |
e0f712ba AC |
400 | r->next = 0; |
401 | if (region_list_head == 0) | |
402 | { | |
403 | region_list_head = r; | |
404 | region_list_tail = r; | |
405 | } | |
406 | else | |
407 | { | |
408 | region_list_tail->next = r; | |
409 | region_list_tail = r; | |
410 | } | |
177c0ea7 | 411 | |
e0f712ba AC |
412 | /* Deallocate (unused) object name returned by |
413 | vm_region. */ | |
414 | if (object_name != MACH_PORT_NULL) | |
415 | mach_port_deallocate (target_task, object_name); | |
416 | } | |
177c0ea7 | 417 | |
e0f712ba AC |
418 | address += size; |
419 | } | |
420 | ||
421 | printf ("--- List of Regions to be Dumped ---\n"); | |
422 | print_region_list (); | |
423 | } | |
424 | ||
425 | ||
73da71f9 | 426 | #define MAX_UNEXEC_REGIONS 400 |
e0f712ba | 427 | |
c57038f8 YM |
428 | static int num_unexec_regions; |
429 | typedef struct { | |
430 | vm_range_t range; | |
431 | vm_size_t filesize; | |
432 | } unexec_region_info; | |
433 | static unexec_region_info unexec_regions[MAX_UNEXEC_REGIONS]; | |
e0f712ba AC |
434 | |
435 | static void | |
436 | unexec_regions_recorder (task_t task, void *rr, unsigned type, | |
437 | vm_range_t *ranges, unsigned num) | |
438 | { | |
c57038f8 YM |
439 | vm_address_t p; |
440 | vm_size_t filesize; | |
441 | ||
e0f712ba AC |
442 | while (num && num_unexec_regions < MAX_UNEXEC_REGIONS) |
443 | { | |
ae7c60a9 | 444 | /* Subtract the size of trailing null bytes from filesize. It |
c57038f8 | 445 | can be smaller than vmsize in segment commands. In such a |
ae7c60a9 YM |
446 | case, trailing bytes are initialized with zeros. */ |
447 | for (p = ranges->address + ranges->size; p > ranges->address; p--) | |
448 | if (*(((char *) p)-1)) | |
449 | break; | |
450 | filesize = p - ranges->address; | |
c57038f8 YM |
451 | |
452 | unexec_regions[num_unexec_regions].filesize = filesize; | |
453 | unexec_regions[num_unexec_regions++].range = *ranges; | |
454 | printf ("%#10lx (sz: %#8lx/%#8lx)\n", (long) (ranges->address), | |
455 | (long) filesize, (long) (ranges->size)); | |
e0f712ba AC |
456 | ranges++; num--; |
457 | } | |
e0f712ba AC |
458 | } |
459 | ||
460 | static kern_return_t | |
461 | unexec_reader (task_t task, vm_address_t address, vm_size_t size, void **ptr) | |
462 | { | |
463 | *ptr = (void *) address; | |
464 | return KERN_SUCCESS; | |
465 | } | |
466 | ||
c57038f8 | 467 | static void |
3d608a86 | 468 | find_emacs_zone_regions (void) |
e0f712ba AC |
469 | { |
470 | num_unexec_regions = 0; | |
471 | ||
472 | emacs_zone->introspect->enumerator (mach_task_self(), 0, | |
473 | MALLOC_PTR_REGION_RANGE_TYPE | |
474 | | MALLOC_ADMIN_REGION_RANGE_TYPE, | |
475 | (vm_address_t) emacs_zone, | |
476 | unexec_reader, | |
477 | unexec_regions_recorder); | |
73da71f9 YM |
478 | |
479 | if (num_unexec_regions == MAX_UNEXEC_REGIONS) | |
480 | unexec_error ("find_emacs_zone_regions: too many regions"); | |
e0f712ba AC |
481 | } |
482 | ||
1dd7ccf2 AC |
483 | static int |
484 | unexec_regions_sort_compare (const void *a, const void *b) | |
485 | { | |
c57038f8 YM |
486 | vm_address_t aa = ((unexec_region_info *) a)->range.address; |
487 | vm_address_t bb = ((unexec_region_info *) b)->range.address; | |
1dd7ccf2 AC |
488 | |
489 | if (aa < bb) | |
490 | return -1; | |
491 | else if (aa > bb) | |
492 | return 1; | |
493 | else | |
494 | return 0; | |
495 | } | |
496 | ||
497 | static void | |
3d608a86 | 498 | unexec_regions_merge (void) |
1dd7ccf2 AC |
499 | { |
500 | int i, n; | |
c57038f8 | 501 | unexec_region_info r; |
ae7c60a9 | 502 | vm_size_t padsize; |
1dd7ccf2 AC |
503 | |
504 | qsort (unexec_regions, num_unexec_regions, sizeof (unexec_regions[0]), | |
505 | &unexec_regions_sort_compare); | |
506 | n = 0; | |
507 | r = unexec_regions[0]; | |
ae7c60a9 YM |
508 | padsize = r.range.address & (pagesize - 1); |
509 | if (padsize) | |
510 | { | |
511 | r.range.address -= padsize; | |
512 | r.range.size += padsize; | |
513 | r.filesize += padsize; | |
514 | } | |
1dd7ccf2 AC |
515 | for (i = 1; i < num_unexec_regions; i++) |
516 | { | |
c57038f8 YM |
517 | if (r.range.address + r.range.size == unexec_regions[i].range.address |
518 | && r.range.size - r.filesize < 2 * pagesize) | |
1dd7ccf2 | 519 | { |
c57038f8 YM |
520 | r.filesize = r.range.size + unexec_regions[i].filesize; |
521 | r.range.size += unexec_regions[i].range.size; | |
1dd7ccf2 AC |
522 | } |
523 | else | |
524 | { | |
525 | unexec_regions[n++] = r; | |
526 | r = unexec_regions[i]; | |
ae7c60a9 YM |
527 | padsize = r.range.address & (pagesize - 1); |
528 | if (padsize) | |
529 | { | |
530 | if ((unexec_regions[n-1].range.address | |
531 | + unexec_regions[n-1].range.size) == r.range.address) | |
532 | unexec_regions[n-1].range.size -= padsize; | |
533 | ||
534 | r.range.address -= padsize; | |
535 | r.range.size += padsize; | |
536 | r.filesize += padsize; | |
537 | } | |
1dd7ccf2 AC |
538 | } |
539 | } | |
540 | unexec_regions[n++] = r; | |
541 | num_unexec_regions = n; | |
542 | } | |
543 | ||
e0f712ba AC |
544 | |
545 | /* More informational messages routines. */ | |
546 | ||
547 | static void | |
548 | print_load_command_name (int lc) | |
549 | { | |
550 | switch (lc) | |
551 | { | |
552 | case LC_SEGMENT: | |
73da71f9 | 553 | #ifndef _LP64 |
e0f712ba | 554 | printf ("LC_SEGMENT "); |
73da71f9 YM |
555 | #else |
556 | printf ("LC_SEGMENT_64 "); | |
557 | #endif | |
e0f712ba AC |
558 | break; |
559 | case LC_LOAD_DYLINKER: | |
560 | printf ("LC_LOAD_DYLINKER "); | |
561 | break; | |
562 | case LC_LOAD_DYLIB: | |
563 | printf ("LC_LOAD_DYLIB "); | |
564 | break; | |
565 | case LC_SYMTAB: | |
566 | printf ("LC_SYMTAB "); | |
567 | break; | |
568 | case LC_DYSYMTAB: | |
569 | printf ("LC_DYSYMTAB "); | |
570 | break; | |
571 | case LC_UNIXTHREAD: | |
572 | printf ("LC_UNIXTHREAD "); | |
573 | break; | |
574 | case LC_PREBOUND_DYLIB: | |
575 | printf ("LC_PREBOUND_DYLIB"); | |
576 | break; | |
577 | case LC_TWOLEVEL_HINTS: | |
578 | printf ("LC_TWOLEVEL_HINTS"); | |
579 | break; | |
ae7c60a9 YM |
580 | #ifdef LC_UUID |
581 | case LC_UUID: | |
582 | printf ("LC_UUID "); | |
583 | break; | |
6bcd6333 YM |
584 | #endif |
585 | #ifdef LC_DYLD_INFO | |
586 | case LC_DYLD_INFO: | |
587 | printf ("LC_DYLD_INFO "); | |
588 | break; | |
589 | case LC_DYLD_INFO_ONLY: | |
590 | printf ("LC_DYLD_INFO_ONLY"); | |
591 | break; | |
ae7c60a9 | 592 | #endif |
e0f712ba AC |
593 | default: |
594 | printf ("unknown "); | |
595 | } | |
596 | } | |
597 | ||
598 | static void | |
599 | print_load_command (struct load_command *lc) | |
600 | { | |
601 | print_load_command_name (lc->cmd); | |
602 | printf ("%8d", lc->cmdsize); | |
603 | ||
604 | if (lc->cmd == LC_SEGMENT) | |
605 | { | |
606 | struct segment_command *scp; | |
607 | struct section *sectp; | |
608 | int j; | |
609 | ||
610 | scp = (struct segment_command *) lc; | |
73da71f9 YM |
611 | printf (" %-16.16s %#10lx %#8lx\n", |
612 | scp->segname, (long) (scp->vmaddr), (long) (scp->vmsize)); | |
e0f712ba AC |
613 | |
614 | sectp = (struct section *) (scp + 1); | |
615 | for (j = 0; j < scp->nsects; j++) | |
616 | { | |
73da71f9 YM |
617 | printf (" %-16.16s %#10lx %#8lx\n", |
618 | sectp->sectname, (long) (sectp->addr), (long) (sectp->size)); | |
e0f712ba AC |
619 | sectp++; |
620 | } | |
621 | } | |
622 | else | |
623 | printf ("\n"); | |
624 | } | |
625 | ||
626 | /* Read header and load commands from input file. Store the latter in | |
627 | the global array lca. Store the total number of load commands in | |
628 | global variable nlc. */ | |
629 | static void | |
3d608a86 | 630 | read_load_commands (void) |
e0f712ba | 631 | { |
7f900522 | 632 | int i; |
e0f712ba AC |
633 | |
634 | if (!unexec_read (&mh, sizeof (struct mach_header))) | |
635 | unexec_error ("cannot read mach-o header"); | |
636 | ||
637 | if (mh.magic != MH_MAGIC) | |
638 | unexec_error ("input file not in Mach-O format"); | |
639 | ||
640 | if (mh.filetype != MH_EXECUTE) | |
641 | unexec_error ("input Mach-O file is not an executable object file"); | |
642 | ||
643 | #if VERBOSE | |
644 | printf ("--- Header Information ---\n"); | |
645 | printf ("Magic = 0x%08x\n", mh.magic); | |
646 | printf ("CPUType = %d\n", mh.cputype); | |
647 | printf ("CPUSubType = %d\n", mh.cpusubtype); | |
648 | printf ("FileType = 0x%x\n", mh.filetype); | |
649 | printf ("NCmds = %d\n", mh.ncmds); | |
650 | printf ("SizeOfCmds = %d\n", mh.sizeofcmds); | |
651 | printf ("Flags = 0x%08x\n", mh.flags); | |
652 | #endif | |
653 | ||
654 | nlc = mh.ncmds; | |
655 | lca = (struct load_command **) malloc (nlc * sizeof (struct load_command *)); | |
177c0ea7 | 656 | |
e0f712ba AC |
657 | for (i = 0; i < nlc; i++) |
658 | { | |
659 | struct load_command lc; | |
660 | /* Load commands are variable-size: so read the command type and | |
661 | size first and then read the rest. */ | |
662 | if (!unexec_read (&lc, sizeof (struct load_command))) | |
663 | unexec_error ("cannot read load command"); | |
664 | lca[i] = (struct load_command *) malloc (lc.cmdsize); | |
665 | memcpy (lca[i], &lc, sizeof (struct load_command)); | |
666 | if (!unexec_read (lca[i] + 1, lc.cmdsize - sizeof (struct load_command))) | |
667 | unexec_error ("cannot read content of load command"); | |
668 | if (lc.cmd == LC_SEGMENT) | |
669 | { | |
670 | struct segment_command *scp = (struct segment_command *) lca[i]; | |
177c0ea7 | 671 | |
e0f712ba AC |
672 | if (scp->vmaddr + scp->vmsize > infile_lc_highest_addr) |
673 | infile_lc_highest_addr = scp->vmaddr + scp->vmsize; | |
674 | ||
675 | if (strncmp (scp->segname, SEG_TEXT, 16) == 0) | |
676 | { | |
677 | struct section *sectp = (struct section *) (scp + 1); | |
678 | int j; | |
679 | ||
680 | for (j = 0; j < scp->nsects; j++) | |
681 | if (sectp->offset < text_seg_lowest_offset) | |
682 | text_seg_lowest_offset = sectp->offset; | |
683 | } | |
684 | } | |
685 | } | |
686 | ||
3d608a86 J |
687 | printf ("Highest address of load commands in input file: %#8lx\n", |
688 | (unsigned long)infile_lc_highest_addr); | |
e0f712ba | 689 | |
c57038f8 | 690 | printf ("Lowest offset of all sections in __TEXT segment: %#8lx\n", |
e0f712ba AC |
691 | text_seg_lowest_offset); |
692 | ||
693 | printf ("--- List of Load Commands in Input File ---\n"); | |
694 | printf ("# cmd cmdsize name address size\n"); | |
695 | ||
696 | for (i = 0; i < nlc; i++) | |
697 | { | |
698 | printf ("%1d ", i); | |
699 | print_load_command (lca[i]); | |
700 | } | |
701 | } | |
702 | ||
703 | /* Copy a LC_SEGMENT load command other than the __DATA segment from | |
704 | the input file to the output file, adjusting the file offset of the | |
705 | segment and the file offsets of sections contained in it. */ | |
706 | static void | |
707 | copy_segment (struct load_command *lc) | |
708 | { | |
709 | struct segment_command *scp = (struct segment_command *) lc; | |
710 | unsigned long old_fileoff = scp->fileoff; | |
711 | struct section *sectp; | |
712 | int j; | |
713 | ||
73da71f9 | 714 | scp->fileoff = curr_file_offset; |
e0f712ba AC |
715 | |
716 | sectp = (struct section *) (scp + 1); | |
717 | for (j = 0; j < scp->nsects; j++) | |
718 | { | |
73da71f9 | 719 | sectp->offset += curr_file_offset - old_fileoff; |
e0f712ba AC |
720 | sectp++; |
721 | } | |
722 | ||
c57038f8 YM |
723 | printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n", |
724 | scp->segname, (long) (scp->fileoff), (long) (scp->filesize), | |
725 | (long) (scp->vmsize), (long) (scp->vmaddr)); | |
e0f712ba AC |
726 | |
727 | if (!unexec_copy (scp->fileoff, old_fileoff, scp->filesize)) | |
728 | unexec_error ("cannot copy segment from input to output file"); | |
73da71f9 YM |
729 | curr_file_offset += ROUNDUP_TO_PAGE_BOUNDARY (scp->filesize); |
730 | ||
e0f712ba AC |
731 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) |
732 | unexec_error ("cannot write load command to header"); | |
733 | ||
734 | curr_header_offset += lc->cmdsize; | |
735 | } | |
736 | ||
737 | /* Copy a LC_SEGMENT load command for the __DATA segment in the input | |
738 | file to the output file. We assume that only one such segment load | |
739 | command exists in the input file and it contains the sections | |
740 | __data, __bss, __common, __la_symbol_ptr, __nl_symbol_ptr, and | |
741 | __dyld. The first three of these should be dumped from memory and | |
742 | the rest should be copied from the input file. Note that the | |
743 | sections __bss and __common contain no data in the input file | |
744 | because their flag fields have the value S_ZEROFILL. Dumping these | |
745 | from memory makes it necessary to adjust file offset fields in | |
746 | subsequently dumped load commands. Then, create new __DATA segment | |
747 | load commands for regions on the region list other than the one | |
748 | corresponding to the __DATA segment in the input file. */ | |
749 | static void | |
750 | copy_data_segment (struct load_command *lc) | |
751 | { | |
752 | struct segment_command *scp = (struct segment_command *) lc; | |
753 | struct section *sectp; | |
754 | int j; | |
c57038f8 YM |
755 | unsigned long header_offset, old_file_offset; |
756 | ||
757 | /* The new filesize of the segment is set to its vmsize because data | |
758 | blocks for segments must start at region boundaries. Note that | |
759 | this may leave unused locations at the end of the segment data | |
760 | block because the total of the sizes of all sections in the | |
761 | segment is generally smaller than vmsize. */ | |
762 | scp->filesize = scp->vmsize; | |
e0f712ba | 763 | |
c57038f8 YM |
764 | printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n", |
765 | scp->segname, curr_file_offset, (long)(scp->filesize), | |
766 | (long)(scp->vmsize), (long) (scp->vmaddr)); | |
e0f712ba AC |
767 | |
768 | /* Offsets in the output file for writing the next section structure | |
769 | and segment data block, respectively. */ | |
770 | header_offset = curr_header_offset + sizeof (struct segment_command); | |
771 | ||
772 | sectp = (struct section *) (scp + 1); | |
773 | for (j = 0; j < scp->nsects; j++) | |
774 | { | |
775 | old_file_offset = sectp->offset; | |
73da71f9 | 776 | sectp->offset = sectp->addr - scp->vmaddr + curr_file_offset; |
e0f712ba AC |
777 | /* The __data section is dumped from memory. The __bss and |
778 | __common sections are also dumped from memory but their flag | |
779 | fields require changing (from S_ZEROFILL to S_REGULAR). The | |
780 | other three kinds of sections are just copied from the input | |
781 | file. */ | |
782 | if (strncmp (sectp->sectname, SECT_DATA, 16) == 0) | |
783 | { | |
784 | if (!unexec_write (sectp->offset, (void *) sectp->addr, sectp->size)) | |
785 | unexec_error ("cannot write section %s", SECT_DATA); | |
786 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) | |
787 | unexec_error ("cannot write section %s's header", SECT_DATA); | |
788 | } | |
433456d7 | 789 | else if (strncmp (sectp->sectname, SECT_COMMON, 16) == 0) |
e0f712ba AC |
790 | { |
791 | sectp->flags = S_REGULAR; | |
792 | if (!unexec_write (sectp->offset, (void *) sectp->addr, sectp->size)) | |
433456d7 | 793 | unexec_error ("cannot write section %s", sectp->sectname); |
e0f712ba | 794 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) |
433456d7 YM |
795 | unexec_error ("cannot write section %s's header", sectp->sectname); |
796 | } | |
797 | else if (strncmp (sectp->sectname, SECT_BSS, 16) == 0) | |
798 | { | |
799 | extern char *my_endbss_static; | |
800 | unsigned long my_size; | |
801 | ||
802 | sectp->flags = S_REGULAR; | |
803 | ||
804 | /* Clear uninitialized local variables in statically linked | |
805 | libraries. In particular, function pointers stored by | |
806 | libSystemStub.a, which is introduced in Mac OS X 10.4 for | |
807 | binary compatibility with respect to long double, are | |
808 | cleared so that they will be reinitialized when the | |
809 | dumped binary is executed on other versions of OS. */ | |
810 | my_size = (unsigned long)my_endbss_static - sectp->addr; | |
811 | if (!(sectp->addr <= (unsigned long)my_endbss_static | |
812 | && my_size <= sectp->size)) | |
813 | unexec_error ("my_endbss_static is not in section %s", | |
814 | sectp->sectname); | |
815 | if (!unexec_write (sectp->offset, (void *) sectp->addr, my_size)) | |
816 | unexec_error ("cannot write section %s", sectp->sectname); | |
817 | if (!unexec_write_zero (sectp->offset + my_size, | |
818 | sectp->size - my_size)) | |
819 | unexec_error ("cannot write section %s", sectp->sectname); | |
820 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) | |
821 | unexec_error ("cannot write section %s's header", sectp->sectname); | |
e0f712ba AC |
822 | } |
823 | else if (strncmp (sectp->sectname, "__la_symbol_ptr", 16) == 0 | |
824 | || strncmp (sectp->sectname, "__nl_symbol_ptr", 16) == 0 | |
427c5b1b | 825 | || strncmp (sectp->sectname, "__la_sym_ptr2", 16) == 0 |
e0f712ba | 826 | || strncmp (sectp->sectname, "__dyld", 16) == 0 |
7290a344 | 827 | || strncmp (sectp->sectname, "__const", 16) == 0 |
b2411edf YM |
828 | || strncmp (sectp->sectname, "__cfstring", 16) == 0 |
829 | || strncmp (sectp->sectname, "__gcc_except_tab", 16) == 0 | |
6bcd6333 | 830 | || strncmp (sectp->sectname, "__program_vars", 16) == 0 |
b2411edf | 831 | || strncmp (sectp->sectname, "__objc_", 7) == 0) |
e0f712ba AC |
832 | { |
833 | if (!unexec_copy (sectp->offset, old_file_offset, sectp->size)) | |
834 | unexec_error ("cannot copy section %s", sectp->sectname); | |
835 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) | |
836 | unexec_error ("cannot write section %s's header", sectp->sectname); | |
837 | } | |
838 | else | |
839 | unexec_error ("unrecognized section name in __DATA segment"); | |
177c0ea7 | 840 | |
73da71f9 YM |
841 | printf (" section %-16.16s at %#8lx - %#8lx (sz: %#8lx)\n", |
842 | sectp->sectname, (long) (sectp->offset), | |
843 | (long) (sectp->offset + sectp->size), (long) (sectp->size)); | |
e0f712ba AC |
844 | |
845 | header_offset += sizeof (struct section); | |
846 | sectp++; | |
847 | } | |
848 | ||
c57038f8 YM |
849 | curr_file_offset += ROUNDUP_TO_PAGE_BOUNDARY (scp->filesize); |
850 | ||
e0f712ba AC |
851 | if (!unexec_write (curr_header_offset, scp, sizeof (struct segment_command))) |
852 | unexec_error ("cannot write header of __DATA segment"); | |
853 | curr_header_offset += lc->cmdsize; | |
854 | ||
855 | /* Create new __DATA segment load commands for regions on the region | |
856 | list that do not corresponding to any segment load commands in | |
857 | the input file. | |
73da71f9 | 858 | */ |
e0f712ba AC |
859 | for (j = 0; j < num_unexec_regions; j++) |
860 | { | |
861 | struct segment_command sc; | |
177c0ea7 | 862 | |
e0f712ba AC |
863 | sc.cmd = LC_SEGMENT; |
864 | sc.cmdsize = sizeof (struct segment_command); | |
865 | strncpy (sc.segname, SEG_DATA, 16); | |
c57038f8 YM |
866 | sc.vmaddr = unexec_regions[j].range.address; |
867 | sc.vmsize = unexec_regions[j].range.size; | |
73da71f9 | 868 | sc.fileoff = curr_file_offset; |
c57038f8 | 869 | sc.filesize = unexec_regions[j].filesize; |
e0f712ba AC |
870 | sc.maxprot = VM_PROT_READ | VM_PROT_WRITE; |
871 | sc.initprot = VM_PROT_READ | VM_PROT_WRITE; | |
872 | sc.nsects = 0; | |
873 | sc.flags = 0; | |
177c0ea7 | 874 | |
c57038f8 YM |
875 | printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n", |
876 | sc.segname, (long) (sc.fileoff), (long) (sc.filesize), | |
877 | (long) (sc.vmsize), (long) (sc.vmaddr)); | |
e0f712ba | 878 | |
c57038f8 | 879 | if (!unexec_write (sc.fileoff, (void *) sc.vmaddr, sc.filesize)) |
e0f712ba | 880 | unexec_error ("cannot write new __DATA segment"); |
73da71f9 | 881 | curr_file_offset += ROUNDUP_TO_PAGE_BOUNDARY (sc.filesize); |
177c0ea7 | 882 | |
e0f712ba AC |
883 | if (!unexec_write (curr_header_offset, &sc, sc.cmdsize)) |
884 | unexec_error ("cannot write new __DATA segment's header"); | |
885 | curr_header_offset += sc.cmdsize; | |
886 | mh.ncmds++; | |
887 | } | |
888 | } | |
889 | ||
890 | /* Copy a LC_SYMTAB load command from the input file to the output | |
891 | file, adjusting the file offset fields. */ | |
892 | static void | |
73da71f9 | 893 | copy_symtab (struct load_command *lc, long delta) |
e0f712ba AC |
894 | { |
895 | struct symtab_command *stp = (struct symtab_command *) lc; | |
896 | ||
897 | stp->symoff += delta; | |
898 | stp->stroff += delta; | |
899 | ||
900 | printf ("Writing LC_SYMTAB command\n"); | |
901 | ||
902 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
903 | unexec_error ("cannot write symtab command to header"); | |
904 | ||
905 | curr_header_offset += lc->cmdsize; | |
906 | } | |
907 | ||
043131c4 AC |
908 | /* Fix up relocation entries. */ |
909 | static void | |
7aee2da7 | 910 | unrelocate (const char *name, off_t reloff, int nrel, vm_address_t base) |
043131c4 AC |
911 | { |
912 | int i, unreloc_count; | |
913 | struct relocation_info reloc_info; | |
914 | struct scattered_relocation_info *sc_reloc_info | |
915 | = (struct scattered_relocation_info *) &reloc_info; | |
7aee2da7 | 916 | vm_address_t location; |
043131c4 AC |
917 | |
918 | for (unreloc_count = 0, i = 0; i < nrel; i++) | |
919 | { | |
920 | if (lseek (infd, reloff, L_SET) != reloff) | |
921 | unexec_error ("unrelocate: %s:%d cannot seek to reloc_info", name, i); | |
922 | if (!unexec_read (&reloc_info, sizeof (reloc_info))) | |
923 | unexec_error ("unrelocate: %s:%d cannot read reloc_info", name, i); | |
924 | reloff += sizeof (reloc_info); | |
925 | ||
926 | if (sc_reloc_info->r_scattered == 0) | |
927 | switch (reloc_info.r_type) | |
928 | { | |
929 | case GENERIC_RELOC_VANILLA: | |
7aee2da7 | 930 | location = base + reloc_info.r_address; |
b2411edf YM |
931 | if (location >= data_segment_scp->vmaddr |
932 | && location < (data_segment_scp->vmaddr | |
933 | + data_segment_scp->vmsize)) | |
043131c4 AC |
934 | { |
935 | off_t src_off = data_segment_old_fileoff | |
b2411edf | 936 | + (location - data_segment_scp->vmaddr); |
043131c4 | 937 | off_t dst_off = data_segment_scp->fileoff |
b2411edf | 938 | + (location - data_segment_scp->vmaddr); |
043131c4 AC |
939 | |
940 | if (!unexec_copy (dst_off, src_off, 1 << reloc_info.r_length)) | |
941 | unexec_error ("unrelocate: %s:%d cannot copy original value", | |
942 | name, i); | |
943 | unreloc_count++; | |
944 | } | |
945 | break; | |
946 | default: | |
947 | unexec_error ("unrelocate: %s:%d cannot handle type = %d", | |
948 | name, i, reloc_info.r_type); | |
949 | } | |
950 | else | |
951 | switch (sc_reloc_info->r_type) | |
952 | { | |
953 | #if defined (__ppc__) | |
954 | case PPC_RELOC_PB_LA_PTR: | |
955 | /* nothing to do for prebound lazy pointer */ | |
956 | break; | |
957 | #endif | |
958 | default: | |
959 | unexec_error ("unrelocate: %s:%d cannot handle scattered type = %d", | |
960 | name, i, sc_reloc_info->r_type); | |
961 | } | |
962 | } | |
963 | ||
964 | if (nrel > 0) | |
965 | printf ("Fixed up %d/%d %s relocation entries in data segment.\n", | |
966 | unreloc_count, nrel, name); | |
967 | } | |
968 | ||
7aee2da7 YM |
969 | #if __ppc64__ |
970 | /* Rebase r_address in the relocation table. */ | |
971 | static void | |
972 | rebase_reloc_address (off_t reloff, int nrel, long linkedit_delta, long diff) | |
973 | { | |
974 | int i; | |
975 | struct relocation_info reloc_info; | |
976 | struct scattered_relocation_info *sc_reloc_info | |
977 | = (struct scattered_relocation_info *) &reloc_info; | |
978 | ||
979 | for (i = 0; i < nrel; i++, reloff += sizeof (reloc_info)) | |
980 | { | |
981 | if (lseek (infd, reloff - linkedit_delta, L_SET) | |
982 | != reloff - linkedit_delta) | |
983 | unexec_error ("rebase_reloc_table: cannot seek to reloc_info"); | |
984 | if (!unexec_read (&reloc_info, sizeof (reloc_info))) | |
985 | unexec_error ("rebase_reloc_table: cannot read reloc_info"); | |
986 | ||
987 | if (sc_reloc_info->r_scattered == 0 | |
988 | && reloc_info.r_type == GENERIC_RELOC_VANILLA) | |
989 | { | |
990 | reloc_info.r_address -= diff; | |
991 | if (!unexec_write (reloff, &reloc_info, sizeof (reloc_info))) | |
992 | unexec_error ("rebase_reloc_table: cannot write reloc_info"); | |
993 | } | |
994 | } | |
995 | } | |
996 | #endif | |
997 | ||
e0f712ba AC |
998 | /* Copy a LC_DYSYMTAB load command from the input file to the output |
999 | file, adjusting the file offset fields. */ | |
1000 | static void | |
73da71f9 | 1001 | copy_dysymtab (struct load_command *lc, long delta) |
e0f712ba AC |
1002 | { |
1003 | struct dysymtab_command *dstp = (struct dysymtab_command *) lc; | |
7aee2da7 | 1004 | vm_address_t base; |
e0f712ba | 1005 | |
7aee2da7 YM |
1006 | #ifdef _LP64 |
1007 | #if __ppc64__ | |
1008 | { | |
1009 | int i; | |
1010 | ||
1011 | base = 0; | |
1012 | for (i = 0; i < nlc; i++) | |
1013 | if (lca[i]->cmd == LC_SEGMENT) | |
1014 | { | |
1015 | struct segment_command *scp = (struct segment_command *) lca[i]; | |
1016 | ||
1017 | if (scp->vmaddr + scp->vmsize > 0x100000000 | |
1018 | && (scp->initprot & VM_PROT_WRITE) != 0) | |
1019 | { | |
1020 | base = data_segment_scp->vmaddr; | |
1021 | break; | |
1022 | } | |
1023 | } | |
1024 | } | |
1025 | #else | |
1026 | /* First writable segment address. */ | |
1027 | base = data_segment_scp->vmaddr; | |
1028 | #endif | |
1029 | #else | |
1030 | /* First segment address in the file (unless MH_SPLIT_SEGS set). */ | |
1031 | base = 0; | |
1032 | #endif | |
1033 | ||
1034 | unrelocate ("local", dstp->locreloff, dstp->nlocrel, base); | |
1035 | unrelocate ("external", dstp->extreloff, dstp->nextrel, base); | |
e0f712ba AC |
1036 | |
1037 | if (dstp->nextrel > 0) { | |
1038 | dstp->extreloff += delta; | |
1039 | } | |
1040 | ||
1041 | if (dstp->nlocrel > 0) { | |
1042 | dstp->locreloff += delta; | |
1043 | } | |
1044 | ||
1045 | if (dstp->nindirectsyms > 0) | |
1046 | dstp->indirectsymoff += delta; | |
1047 | ||
1048 | printf ("Writing LC_DYSYMTAB command\n"); | |
1049 | ||
1050 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
1051 | unexec_error ("cannot write symtab command to header"); | |
1052 | ||
1053 | curr_header_offset += lc->cmdsize; | |
7aee2da7 YM |
1054 | |
1055 | #if __ppc64__ | |
1056 | /* Check if the relocation base needs to be changed. */ | |
1057 | if (base == 0) | |
1058 | { | |
1059 | vm_address_t newbase = 0; | |
1060 | int i; | |
1061 | ||
1062 | for (i = 0; i < num_unexec_regions; i++) | |
1063 | if (unexec_regions[i].range.address + unexec_regions[i].range.size | |
1064 | > 0x100000000) | |
1065 | { | |
1066 | newbase = data_segment_scp->vmaddr; | |
1067 | break; | |
1068 | } | |
1069 | ||
1070 | if (newbase) | |
1071 | { | |
1072 | rebase_reloc_address (dstp->locreloff, dstp->nlocrel, delta, newbase); | |
1073 | rebase_reloc_address (dstp->extreloff, dstp->nextrel, delta, newbase); | |
1074 | } | |
1075 | } | |
1076 | #endif | |
e0f712ba AC |
1077 | } |
1078 | ||
40e6ff95 ST |
1079 | /* Copy a LC_TWOLEVEL_HINTS load command from the input file to the output |
1080 | file, adjusting the file offset fields. */ | |
1081 | static void | |
73da71f9 | 1082 | copy_twolevelhints (struct load_command *lc, long delta) |
40e6ff95 ST |
1083 | { |
1084 | struct twolevel_hints_command *tlhp = (struct twolevel_hints_command *) lc; | |
1085 | ||
1086 | if (tlhp->nhints > 0) { | |
1087 | tlhp->offset += delta; | |
1088 | } | |
1089 | ||
1090 | printf ("Writing LC_TWOLEVEL_HINTS command\n"); | |
1091 | ||
1092 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
1093 | unexec_error ("cannot write two level hint command to header"); | |
1094 | ||
1095 | curr_header_offset += lc->cmdsize; | |
1096 | } | |
1097 | ||
6bcd6333 YM |
1098 | #ifdef LC_DYLD_INFO |
1099 | /* Copy a LC_DYLD_INFO(_ONLY) load command from the input file to the output | |
1100 | file, adjusting the file offset fields. */ | |
1101 | static void | |
1102 | copy_dyld_info (struct load_command *lc, long delta) | |
1103 | { | |
1104 | struct dyld_info_command *dip = (struct dyld_info_command *) lc; | |
1105 | ||
1106 | if (dip->rebase_off > 0) | |
1107 | dip->rebase_off += delta; | |
1108 | if (dip->bind_off > 0) | |
1109 | dip->bind_off += delta; | |
1110 | if (dip->weak_bind_off > 0) | |
1111 | dip->weak_bind_off += delta; | |
1112 | if (dip->lazy_bind_off > 0) | |
1113 | dip->lazy_bind_off += delta; | |
1114 | if (dip->export_off > 0) | |
1115 | dip->export_off += delta; | |
1116 | ||
1117 | printf ("Writing "); | |
1118 | print_load_command_name (lc->cmd); | |
1119 | printf (" command\n"); | |
1120 | ||
1121 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
1122 | unexec_error ("cannot write dyld info command to header"); | |
1123 | ||
1124 | curr_header_offset += lc->cmdsize; | |
1125 | } | |
1126 | #endif | |
1127 | ||
e0f712ba AC |
1128 | /* Copy other kinds of load commands from the input file to the output |
1129 | file, ones that do not require adjustments of file offsets. */ | |
1130 | static void | |
1131 | copy_other (struct load_command *lc) | |
1132 | { | |
1133 | printf ("Writing "); | |
1134 | print_load_command_name (lc->cmd); | |
1135 | printf (" command\n"); | |
1136 | ||
1137 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
1138 | unexec_error ("cannot write symtab command to header"); | |
1139 | ||
1140 | curr_header_offset += lc->cmdsize; | |
1141 | } | |
1142 | ||
1143 | /* Loop through all load commands and dump them. Then write the Mach | |
1144 | header. */ | |
1145 | static void | |
3d608a86 | 1146 | dump_it (void) |
e0f712ba AC |
1147 | { |
1148 | int i; | |
73da71f9 | 1149 | long linkedit_delta = 0; |
e0f712ba AC |
1150 | |
1151 | printf ("--- Load Commands written to Output File ---\n"); | |
1152 | ||
1153 | for (i = 0; i < nlc; i++) | |
1154 | switch (lca[i]->cmd) | |
1155 | { | |
1156 | case LC_SEGMENT: | |
1157 | { | |
1158 | struct segment_command *scp = (struct segment_command *) lca[i]; | |
1159 | if (strncmp (scp->segname, SEG_DATA, 16) == 0) | |
1160 | { | |
043131c4 AC |
1161 | /* save data segment file offset and segment_command for |
1162 | unrelocate */ | |
73da71f9 YM |
1163 | if (data_segment_old_fileoff) |
1164 | unexec_error ("cannot handle multiple DATA segments" | |
1165 | " in input file"); | |
043131c4 AC |
1166 | data_segment_old_fileoff = scp->fileoff; |
1167 | data_segment_scp = scp; | |
1168 | ||
e0f712ba AC |
1169 | copy_data_segment (lca[i]); |
1170 | } | |
1171 | else | |
1172 | { | |
73da71f9 YM |
1173 | if (strncmp (scp->segname, SEG_LINKEDIT, 16) == 0) |
1174 | { | |
1175 | if (linkedit_delta) | |
1176 | unexec_error ("cannot handle multiple LINKEDIT segments" | |
1177 | " in input file"); | |
1178 | linkedit_delta = curr_file_offset - scp->fileoff; | |
1179 | } | |
1180 | ||
e0f712ba AC |
1181 | copy_segment (lca[i]); |
1182 | } | |
1183 | } | |
1184 | break; | |
1185 | case LC_SYMTAB: | |
73da71f9 | 1186 | copy_symtab (lca[i], linkedit_delta); |
e0f712ba AC |
1187 | break; |
1188 | case LC_DYSYMTAB: | |
73da71f9 | 1189 | copy_dysymtab (lca[i], linkedit_delta); |
e0f712ba | 1190 | break; |
40e6ff95 | 1191 | case LC_TWOLEVEL_HINTS: |
73da71f9 | 1192 | copy_twolevelhints (lca[i], linkedit_delta); |
40e6ff95 | 1193 | break; |
6bcd6333 YM |
1194 | #ifdef LC_DYLD_INFO |
1195 | case LC_DYLD_INFO: | |
1196 | case LC_DYLD_INFO_ONLY: | |
1197 | copy_dyld_info (lca[i], linkedit_delta); | |
1198 | break; | |
1199 | #endif | |
e0f712ba AC |
1200 | default: |
1201 | copy_other (lca[i]); | |
1202 | break; | |
1203 | } | |
1204 | ||
1205 | if (curr_header_offset > text_seg_lowest_offset) | |
1206 | unexec_error ("not enough room for load commands for new __DATA segments"); | |
1207 | ||
c57038f8 | 1208 | printf ("%ld unused bytes follow Mach-O header\n", |
e0f712ba AC |
1209 | text_seg_lowest_offset - curr_header_offset); |
1210 | ||
1211 | mh.sizeofcmds = curr_header_offset - sizeof (struct mach_header); | |
1212 | if (!unexec_write (0, &mh, sizeof (struct mach_header))) | |
1213 | unexec_error ("cannot write final header contents"); | |
1214 | } | |
1215 | ||
1216 | /* Take a snapshot of Emacs and make a Mach-O format executable file | |
1217 | from it. The file names of the output and input files are outfile | |
1218 | and infile, respectively. The three other parameters are | |
1219 | ignored. */ | |
1220 | void | |
1221 | unexec (char *outfile, char *infile, void *start_data, void *start_bss, | |
1222 | void *entry_address) | |
1223 | { | |
6dc5c8a7 YM |
1224 | if (in_dumped_exec) |
1225 | unexec_error ("Unexec from a dumped executable is not supported."); | |
1226 | ||
73da71f9 | 1227 | pagesize = getpagesize (); |
e0f712ba AC |
1228 | infd = open (infile, O_RDONLY, 0); |
1229 | if (infd < 0) | |
1230 | { | |
1231 | unexec_error ("cannot open input file `%s'", infile); | |
1232 | } | |
177c0ea7 | 1233 | |
e0f712ba AC |
1234 | outfd = open (outfile, O_WRONLY | O_TRUNC | O_CREAT, 0755); |
1235 | if (outfd < 0) | |
1236 | { | |
1237 | close (infd); | |
1238 | unexec_error ("cannot open output file `%s'", outfile); | |
1239 | } | |
1240 | ||
1241 | build_region_list (); | |
1242 | read_load_commands (); | |
1243 | ||
1244 | find_emacs_zone_regions (); | |
1dd7ccf2 | 1245 | unexec_regions_merge (); |
e0f712ba AC |
1246 | |
1247 | in_dumped_exec = 1; | |
1248 | ||
1249 | dump_it (); | |
1250 | ||
1251 | close (outfd); | |
1252 | } | |
1253 | ||
1254 | ||
1255 | void | |
3d608a86 | 1256 | unexec_init_emacs_zone (void) |
e0f712ba AC |
1257 | { |
1258 | emacs_zone = malloc_create_zone (0, 0); | |
1259 | malloc_set_zone_name (emacs_zone, "EmacsZone"); | |
1260 | } | |
1261 | ||
40ef0695 YM |
1262 | #ifndef MACOSX_MALLOC_MULT16 |
1263 | #define MACOSX_MALLOC_MULT16 1 | |
1264 | #endif | |
1265 | ||
1266 | typedef struct unexec_malloc_header { | |
1267 | union { | |
1268 | char c[8]; | |
1269 | size_t size; | |
1270 | } u; | |
1271 | } unexec_malloc_header_t; | |
1272 | ||
1273 | #if MACOSX_MALLOC_MULT16 | |
1274 | ||
1275 | #define ptr_in_unexec_regions(p) ((((vm_address_t) (p)) & 8) != 0) | |
1276 | ||
1277 | #else | |
1278 | ||
e0f712ba AC |
1279 | int |
1280 | ptr_in_unexec_regions (void *ptr) | |
1281 | { | |
1282 | int i; | |
1283 | ||
1284 | for (i = 0; i < num_unexec_regions; i++) | |
c57038f8 YM |
1285 | if ((vm_address_t) ptr - unexec_regions[i].range.address |
1286 | < unexec_regions[i].range.size) | |
e0f712ba AC |
1287 | return 1; |
1288 | ||
1289 | return 0; | |
1290 | } | |
1291 | ||
40ef0695 YM |
1292 | #endif |
1293 | ||
e0f712ba AC |
1294 | void * |
1295 | unexec_malloc (size_t size) | |
1296 | { | |
1297 | if (in_dumped_exec) | |
40ef0695 YM |
1298 | { |
1299 | void *p; | |
1300 | ||
1301 | p = malloc (size); | |
1302 | #if MACOSX_MALLOC_MULT16 | |
1303 | assert (((vm_address_t) p % 16) == 0); | |
1304 | #endif | |
1305 | return p; | |
1306 | } | |
e0f712ba | 1307 | else |
40ef0695 YM |
1308 | { |
1309 | unexec_malloc_header_t *ptr; | |
1310 | ||
1311 | ptr = (unexec_malloc_header_t *) | |
1312 | malloc_zone_malloc (emacs_zone, size + sizeof (unexec_malloc_header_t)); | |
1313 | ptr->u.size = size; | |
1314 | ptr++; | |
1315 | #if MACOSX_MALLOC_MULT16 | |
1316 | assert (((vm_address_t) ptr % 16) == 8); | |
1317 | #endif | |
1318 | return (void *) ptr; | |
1319 | } | |
e0f712ba AC |
1320 | } |
1321 | ||
1322 | void * | |
1323 | unexec_realloc (void *old_ptr, size_t new_size) | |
1324 | { | |
1325 | if (in_dumped_exec) | |
40ef0695 YM |
1326 | { |
1327 | void *p; | |
1328 | ||
1329 | if (ptr_in_unexec_regions (old_ptr)) | |
1330 | { | |
40ef0695 YM |
1331 | size_t old_size = ((unexec_malloc_header_t *) old_ptr)[-1].u.size; |
1332 | size_t size = new_size > old_size ? old_size : new_size; | |
1333 | ||
0da46b6e | 1334 | p = (size_t *) malloc (new_size); |
40ef0695 YM |
1335 | if (size) |
1336 | memcpy (p, old_ptr, size); | |
1337 | } | |
1338 | else | |
1339 | { | |
1340 | p = realloc (old_ptr, new_size); | |
1341 | } | |
1342 | #if MACOSX_MALLOC_MULT16 | |
1343 | assert (((vm_address_t) p % 16) == 0); | |
1344 | #endif | |
1345 | return p; | |
1346 | } | |
e0f712ba | 1347 | else |
40ef0695 YM |
1348 | { |
1349 | unexec_malloc_header_t *ptr; | |
1350 | ||
1351 | ptr = (unexec_malloc_header_t *) | |
1352 | malloc_zone_realloc (emacs_zone, (unexec_malloc_header_t *) old_ptr - 1, | |
1353 | new_size + sizeof (unexec_malloc_header_t)); | |
1354 | ptr->u.size = new_size; | |
1355 | ptr++; | |
1356 | #if MACOSX_MALLOC_MULT16 | |
1357 | assert (((vm_address_t) ptr % 16) == 8); | |
1358 | #endif | |
1359 | return (void *) ptr; | |
1360 | } | |
e0f712ba AC |
1361 | } |
1362 | ||
1363 | void | |
1364 | unexec_free (void *ptr) | |
1365 | { | |
9c5e177e JM |
1366 | if (ptr == NULL) |
1367 | return; | |
e0f712ba AC |
1368 | if (in_dumped_exec) |
1369 | { | |
1370 | if (!ptr_in_unexec_regions (ptr)) | |
1371 | free (ptr); | |
1372 | } | |
1373 | else | |
40ef0695 | 1374 | malloc_zone_free (emacs_zone, (unexec_malloc_header_t *) ptr - 1); |
e0f712ba | 1375 | } |
ab5796a9 MB |
1376 | |
1377 | /* arch-tag: 1a784f7b-a184-4c4f-9544-da8619593d72 | |
1378 | (do not change this comment) */ |