Commit | Line | Data |
---|---|---|
e0f712ba | 1 | /* Dump Emacs in Mach-O format for use on Mac OS X. |
acaf905b | 2 | Copyright (C) 2001-2012 Free Software Foundation, Inc. |
e0f712ba AC |
3 | |
4 | This file is part of GNU Emacs. | |
5 | ||
9ec0b715 | 6 | GNU Emacs is free software: you can redistribute it and/or modify |
e0f712ba | 7 | it under the terms of the GNU General Public License as published by |
9ec0b715 GM |
8 | the Free Software Foundation, either version 3 of the License, or |
9 | (at your option) any later version. | |
e0f712ba AC |
10 | |
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. | |
15 | ||
16 | You should have received a copy of the GNU General Public License | |
9ec0b715 | 17 | along with GNU Emacs. If not, see <http://www.gnu.org/licenses/>. */ |
e0f712ba AC |
18 | |
19 | /* Contributed by Andrew Choi (akochoi@mac.com). */ | |
20 | ||
21 | /* Documentation note. | |
22 | ||
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. */ | |
35 | ||
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 | |
50 | it. */ | |
51 | ||
52 | /* Peculiarity of the Mach-O files generated by ld in Mac OS X | |
53 | (possible causes of future bugs if changed). | |
54 | ||
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 | |
62 | remains unchanged. | |
63 | ||
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 | |
c57038f8 YM |
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 | |
72 | segments. | |
e0f712ba AC |
73 | |
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. | |
86 | */ | |
87 | ||
e885315d JD |
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. */ | |
92 | ||
e0f712ba | 93 | #include <stdlib.h> |
8beb828a JD |
94 | #include <config.h> |
95 | #undef malloc | |
96 | #undef realloc | |
97 | #undef free | |
ce701a33 PE |
98 | |
99 | #include "unexec.h" | |
100 | ||
e885315d JD |
101 | #include <stdio.h> |
102 | #include <fcntl.h> | |
103 | #include <stdarg.h> | |
104 | #include <sys/types.h> | |
e0f712ba AC |
105 | #include <unistd.h> |
106 | #include <mach/mach.h> | |
107 | #include <mach-o/loader.h> | |
043131c4 AC |
108 | #include <mach-o/reloc.h> |
109 | #if defined (__ppc__) | |
110 | #include <mach-o/ppc/reloc.h> | |
111 | #endif | |
7f900522 | 112 | #ifdef HAVE_MALLOC_MALLOC_H |
f7f3a65f ST |
113 | #include <malloc/malloc.h> |
114 | #else | |
e0f712ba | 115 | #include <objc/malloc.h> |
f7f3a65f ST |
116 | #endif |
117 | ||
40ef0695 YM |
118 | #include <assert.h> |
119 | ||
73da71f9 YM |
120 | #ifdef _LP64 |
121 | #define mach_header mach_header_64 | |
122 | #define segment_command segment_command_64 | |
123 | #undef VM_REGION_BASIC_INFO_COUNT | |
124 | #define VM_REGION_BASIC_INFO_COUNT VM_REGION_BASIC_INFO_COUNT_64 | |
125 | #undef VM_REGION_BASIC_INFO | |
126 | #define VM_REGION_BASIC_INFO VM_REGION_BASIC_INFO_64 | |
127 | #undef LC_SEGMENT | |
128 | #define LC_SEGMENT LC_SEGMENT_64 | |
129 | #define vm_region vm_region_64 | |
130 | #define section section_64 | |
131 | #undef MH_MAGIC | |
132 | #define MH_MAGIC MH_MAGIC_64 | |
133 | #endif | |
e0f712ba AC |
134 | |
135 | #define VERBOSE 1 | |
136 | ||
137 | /* Size of buffer used to copy data from the input file to the output | |
138 | file in function unexec_copy. */ | |
139 | #define UNEXEC_COPY_BUFSZ 1024 | |
140 | ||
141 | /* Regions with memory addresses above this value are assumed to be | |
142 | mapped to dynamically loaded libraries and will not be dumped. */ | |
143 | #define VM_DATA_TOP (20 * 1024 * 1024) | |
144 | ||
e0f712ba AC |
145 | /* Type of an element on the list of regions to be dumped. */ |
146 | struct region_t { | |
147 | vm_address_t address; | |
148 | vm_size_t size; | |
149 | vm_prot_t protection; | |
150 | vm_prot_t max_protection; | |
151 | ||
152 | struct region_t *next; | |
153 | }; | |
154 | ||
155 | /* Head and tail of the list of regions to be dumped. */ | |
c57038f8 YM |
156 | static struct region_t *region_list_head = 0; |
157 | static struct region_t *region_list_tail = 0; | |
e0f712ba AC |
158 | |
159 | /* Pointer to array of load commands. */ | |
c57038f8 | 160 | static struct load_command **lca; |
e0f712ba AC |
161 | |
162 | /* Number of load commands. */ | |
c57038f8 | 163 | static int nlc; |
e0f712ba AC |
164 | |
165 | /* The highest VM address of segments loaded by the input file. | |
166 | Regions with addresses beyond this are assumed to be allocated | |
167 | dynamically and thus require dumping. */ | |
c57038f8 | 168 | static vm_address_t infile_lc_highest_addr = 0; |
e0f712ba AC |
169 | |
170 | /* The lowest file offset used by the all sections in the __TEXT | |
171 | segments. This leaves room at the beginning of the file to store | |
172 | the Mach-O header. Check this value against header size to ensure | |
173 | the added load commands for the new __DATA segments did not | |
174 | overwrite any of the sections in the __TEXT segment. */ | |
c57038f8 | 175 | static unsigned long text_seg_lowest_offset = 0x10000000; |
e0f712ba AC |
176 | |
177 | /* Mach header. */ | |
c57038f8 | 178 | static struct mach_header mh; |
e0f712ba AC |
179 | |
180 | /* Offset at which the next load command should be written. */ | |
c57038f8 | 181 | static unsigned long curr_header_offset = sizeof (struct mach_header); |
e0f712ba | 182 | |
73da71f9 YM |
183 | /* Offset at which the next segment should be written. */ |
184 | static unsigned long curr_file_offset = 0; | |
185 | ||
186 | static unsigned long pagesize; | |
187 | #define ROUNDUP_TO_PAGE_BOUNDARY(x) (((x) + pagesize - 1) & ~(pagesize - 1)) | |
e0f712ba | 188 | |
c57038f8 | 189 | static int infd, outfd; |
e0f712ba | 190 | |
c57038f8 | 191 | static int in_dumped_exec = 0; |
e0f712ba | 192 | |
c57038f8 | 193 | static malloc_zone_t *emacs_zone; |
e0f712ba | 194 | |
043131c4 | 195 | /* file offset of input file's data segment */ |
c57038f8 | 196 | static off_t data_segment_old_fileoff = 0; |
043131c4 | 197 | |
c57038f8 | 198 | static struct segment_command *data_segment_scp; |
043131c4 | 199 | |
433456d7 | 200 | /* Read N bytes from infd into memory starting at address DEST. |
e0f712ba AC |
201 | Return true if successful, false otherwise. */ |
202 | static int | |
203 | unexec_read (void *dest, size_t n) | |
204 | { | |
205 | return n == read (infd, dest, n); | |
206 | } | |
207 | ||
433456d7 YM |
208 | /* Write COUNT bytes from memory starting at address SRC to outfd |
209 | starting at offset DEST. Return true if successful, false | |
210 | otherwise. */ | |
e0f712ba AC |
211 | static int |
212 | unexec_write (off_t dest, const void *src, size_t count) | |
213 | { | |
214 | if (lseek (outfd, dest, SEEK_SET) != dest) | |
215 | return 0; | |
216 | ||
217 | return write (outfd, src, count) == count; | |
218 | } | |
219 | ||
433456d7 YM |
220 | /* Write COUNT bytes of zeros to outfd starting at offset DEST. |
221 | Return true if successful, false otherwise. */ | |
222 | static int | |
223 | unexec_write_zero (off_t dest, size_t count) | |
224 | { | |
225 | char buf[UNEXEC_COPY_BUFSZ]; | |
226 | ssize_t bytes; | |
227 | ||
72af86bd | 228 | memset (buf, 0, UNEXEC_COPY_BUFSZ); |
433456d7 YM |
229 | if (lseek (outfd, dest, SEEK_SET) != dest) |
230 | return 0; | |
231 | ||
232 | while (count > 0) | |
233 | { | |
234 | bytes = count > UNEXEC_COPY_BUFSZ ? UNEXEC_COPY_BUFSZ : count; | |
235 | if (write (outfd, buf, bytes) != bytes) | |
236 | return 0; | |
237 | count -= bytes; | |
238 | } | |
239 | ||
240 | return 1; | |
241 | } | |
242 | ||
243 | /* Copy COUNT bytes from starting offset SRC in infd to starting | |
244 | offset DEST in outfd. Return true if successful, false | |
245 | otherwise. */ | |
e0f712ba AC |
246 | static int |
247 | unexec_copy (off_t dest, off_t src, ssize_t count) | |
248 | { | |
249 | ssize_t bytes_read; | |
911c78b4 | 250 | ssize_t bytes_to_read; |
e0f712ba AC |
251 | |
252 | char buf[UNEXEC_COPY_BUFSZ]; | |
253 | ||
254 | if (lseek (infd, src, SEEK_SET) != src) | |
255 | return 0; | |
256 | ||
257 | if (lseek (outfd, dest, SEEK_SET) != dest) | |
258 | return 0; | |
259 | ||
260 | while (count > 0) | |
261 | { | |
911c78b4 ST |
262 | bytes_to_read = count > UNEXEC_COPY_BUFSZ ? UNEXEC_COPY_BUFSZ : count; |
263 | bytes_read = read (infd, buf, bytes_to_read); | |
e0f712ba AC |
264 | if (bytes_read <= 0) |
265 | return 0; | |
266 | if (write (outfd, buf, bytes_read) != bytes_read) | |
267 | return 0; | |
268 | count -= bytes_read; | |
269 | } | |
270 | ||
271 | return 1; | |
272 | } | |
273 | ||
274 | /* Debugging and informational messages routines. */ | |
275 | ||
845ca893 | 276 | static _Noreturn void |
7aee76f4 | 277 | unexec_error (const char *format, ...) |
e0f712ba AC |
278 | { |
279 | va_list ap; | |
280 | ||
281 | va_start (ap, format); | |
282 | fprintf (stderr, "unexec: "); | |
283 | vfprintf (stderr, format, ap); | |
284 | fprintf (stderr, "\n"); | |
285 | va_end (ap); | |
286 | exit (1); | |
287 | } | |
288 | ||
289 | static void | |
290 | print_prot (vm_prot_t prot) | |
291 | { | |
292 | if (prot == VM_PROT_NONE) | |
293 | printf ("none"); | |
294 | else | |
295 | { | |
296 | putchar (prot & VM_PROT_READ ? 'r' : ' '); | |
297 | putchar (prot & VM_PROT_WRITE ? 'w' : ' '); | |
298 | putchar (prot & VM_PROT_EXECUTE ? 'x' : ' '); | |
299 | putchar (' '); | |
300 | } | |
301 | } | |
302 | ||
303 | static void | |
304 | print_region (vm_address_t address, vm_size_t size, vm_prot_t prot, | |
305 | vm_prot_t max_prot) | |
306 | { | |
73da71f9 | 307 | printf ("%#10lx %#8lx ", (long) address, (long) size); |
e0f712ba AC |
308 | print_prot (prot); |
309 | putchar (' '); | |
310 | print_prot (max_prot); | |
311 | putchar ('\n'); | |
312 | } | |
313 | ||
314 | static void | |
3d608a86 | 315 | print_region_list (void) |
e0f712ba AC |
316 | { |
317 | struct region_t *r; | |
318 | ||
319 | printf (" address size prot maxp\n"); | |
320 | ||
321 | for (r = region_list_head; r; r = r->next) | |
322 | print_region (r->address, r->size, r->protection, r->max_protection); | |
323 | } | |
324 | ||
c57038f8 | 325 | static void |
3d608a86 | 326 | print_regions (void) |
e0f712ba AC |
327 | { |
328 | task_t target_task = mach_task_self (); | |
329 | vm_address_t address = (vm_address_t) 0; | |
330 | vm_size_t size; | |
331 | struct vm_region_basic_info info; | |
332 | mach_msg_type_number_t info_count = VM_REGION_BASIC_INFO_COUNT; | |
333 | mach_port_t object_name; | |
334 | ||
335 | printf (" address size prot maxp\n"); | |
336 | ||
337 | while (vm_region (target_task, &address, &size, VM_REGION_BASIC_INFO, | |
338 | (vm_region_info_t) &info, &info_count, &object_name) | |
339 | == KERN_SUCCESS && info_count == VM_REGION_BASIC_INFO_COUNT) | |
340 | { | |
341 | print_region (address, size, info.protection, info.max_protection); | |
342 | ||
343 | if (object_name != MACH_PORT_NULL) | |
344 | mach_port_deallocate (target_task, object_name); | |
177c0ea7 | 345 | |
e0f712ba AC |
346 | address += size; |
347 | } | |
348 | } | |
349 | ||
350 | /* Build the list of regions that need to be dumped. Regions with | |
351 | addresses above VM_DATA_TOP are omitted. Adjacent regions with | |
352 | identical protection are merged. Note that non-writable regions | |
353 | cannot be omitted because they some regions created at run time are | |
354 | read-only. */ | |
355 | static void | |
3d608a86 | 356 | build_region_list (void) |
e0f712ba AC |
357 | { |
358 | task_t target_task = mach_task_self (); | |
359 | vm_address_t address = (vm_address_t) 0; | |
360 | vm_size_t size; | |
361 | struct vm_region_basic_info info; | |
362 | mach_msg_type_number_t info_count = VM_REGION_BASIC_INFO_COUNT; | |
363 | mach_port_t object_name; | |
364 | struct region_t *r; | |
365 | ||
366 | #if VERBOSE | |
367 | printf ("--- List of All Regions ---\n"); | |
368 | printf (" address size prot maxp\n"); | |
369 | #endif | |
370 | ||
371 | while (vm_region (target_task, &address, &size, VM_REGION_BASIC_INFO, | |
372 | (vm_region_info_t) &info, &info_count, &object_name) | |
373 | == KERN_SUCCESS && info_count == VM_REGION_BASIC_INFO_COUNT) | |
374 | { | |
375 | /* Done when we reach addresses of shared libraries, which are | |
376 | loaded in high memory. */ | |
377 | if (address >= VM_DATA_TOP) | |
378 | break; | |
379 | ||
380 | #if VERBOSE | |
381 | print_region (address, size, info.protection, info.max_protection); | |
382 | #endif | |
383 | ||
384 | /* If a region immediately follows the previous one (the one | |
385 | most recently added to the list) and has identical | |
386 | protection, merge it with the latter. Otherwise create a | |
387 | new list element for it. */ | |
388 | if (region_list_tail | |
389 | && info.protection == region_list_tail->protection | |
390 | && info.max_protection == region_list_tail->max_protection | |
391 | && region_list_tail->address + region_list_tail->size == address) | |
392 | { | |
393 | region_list_tail->size += size; | |
394 | } | |
395 | else | |
396 | { | |
38182d90 | 397 | r = malloc (sizeof *r); |
177c0ea7 | 398 | |
e0f712ba AC |
399 | if (!r) |
400 | unexec_error ("cannot allocate region structure"); | |
177c0ea7 | 401 | |
e0f712ba AC |
402 | r->address = address; |
403 | r->size = size; | |
404 | r->protection = info.protection; | |
405 | r->max_protection = info.max_protection; | |
177c0ea7 | 406 | |
e0f712ba AC |
407 | r->next = 0; |
408 | if (region_list_head == 0) | |
409 | { | |
410 | region_list_head = r; | |
411 | region_list_tail = r; | |
412 | } | |
413 | else | |
414 | { | |
415 | region_list_tail->next = r; | |
416 | region_list_tail = r; | |
417 | } | |
177c0ea7 | 418 | |
e0f712ba AC |
419 | /* Deallocate (unused) object name returned by |
420 | vm_region. */ | |
421 | if (object_name != MACH_PORT_NULL) | |
422 | mach_port_deallocate (target_task, object_name); | |
423 | } | |
177c0ea7 | 424 | |
e0f712ba AC |
425 | address += size; |
426 | } | |
427 | ||
428 | printf ("--- List of Regions to be Dumped ---\n"); | |
429 | print_region_list (); | |
430 | } | |
431 | ||
432 | ||
73da71f9 | 433 | #define MAX_UNEXEC_REGIONS 400 |
e0f712ba | 434 | |
c57038f8 YM |
435 | static int num_unexec_regions; |
436 | typedef struct { | |
437 | vm_range_t range; | |
438 | vm_size_t filesize; | |
439 | } unexec_region_info; | |
440 | static unexec_region_info unexec_regions[MAX_UNEXEC_REGIONS]; | |
e0f712ba AC |
441 | |
442 | static void | |
443 | unexec_regions_recorder (task_t task, void *rr, unsigned type, | |
444 | vm_range_t *ranges, unsigned num) | |
445 | { | |
c57038f8 YM |
446 | vm_address_t p; |
447 | vm_size_t filesize; | |
448 | ||
e0f712ba AC |
449 | while (num && num_unexec_regions < MAX_UNEXEC_REGIONS) |
450 | { | |
ae7c60a9 | 451 | /* Subtract the size of trailing null bytes from filesize. It |
c57038f8 | 452 | can be smaller than vmsize in segment commands. In such a |
ae7c60a9 YM |
453 | case, trailing bytes are initialized with zeros. */ |
454 | for (p = ranges->address + ranges->size; p > ranges->address; p--) | |
455 | if (*(((char *) p)-1)) | |
456 | break; | |
457 | filesize = p - ranges->address; | |
c57038f8 YM |
458 | |
459 | unexec_regions[num_unexec_regions].filesize = filesize; | |
460 | unexec_regions[num_unexec_regions++].range = *ranges; | |
461 | printf ("%#10lx (sz: %#8lx/%#8lx)\n", (long) (ranges->address), | |
462 | (long) filesize, (long) (ranges->size)); | |
e0f712ba AC |
463 | ranges++; num--; |
464 | } | |
e0f712ba AC |
465 | } |
466 | ||
467 | static kern_return_t | |
468 | unexec_reader (task_t task, vm_address_t address, vm_size_t size, void **ptr) | |
469 | { | |
470 | *ptr = (void *) address; | |
471 | return KERN_SUCCESS; | |
472 | } | |
473 | ||
c57038f8 | 474 | static void |
3d608a86 | 475 | find_emacs_zone_regions (void) |
e0f712ba AC |
476 | { |
477 | num_unexec_regions = 0; | |
478 | ||
5e617bc2 | 479 | emacs_zone->introspect->enumerator (mach_task_self (), 0, |
e0f712ba AC |
480 | MALLOC_PTR_REGION_RANGE_TYPE |
481 | | MALLOC_ADMIN_REGION_RANGE_TYPE, | |
482 | (vm_address_t) emacs_zone, | |
483 | unexec_reader, | |
484 | unexec_regions_recorder); | |
73da71f9 YM |
485 | |
486 | if (num_unexec_regions == MAX_UNEXEC_REGIONS) | |
487 | unexec_error ("find_emacs_zone_regions: too many regions"); | |
e0f712ba AC |
488 | } |
489 | ||
1dd7ccf2 AC |
490 | static int |
491 | unexec_regions_sort_compare (const void *a, const void *b) | |
492 | { | |
c57038f8 YM |
493 | vm_address_t aa = ((unexec_region_info *) a)->range.address; |
494 | vm_address_t bb = ((unexec_region_info *) b)->range.address; | |
1dd7ccf2 AC |
495 | |
496 | if (aa < bb) | |
497 | return -1; | |
498 | else if (aa > bb) | |
499 | return 1; | |
500 | else | |
501 | return 0; | |
502 | } | |
503 | ||
504 | static void | |
3d608a86 | 505 | unexec_regions_merge (void) |
1dd7ccf2 AC |
506 | { |
507 | int i, n; | |
c57038f8 | 508 | unexec_region_info r; |
ae7c60a9 | 509 | vm_size_t padsize; |
1dd7ccf2 AC |
510 | |
511 | qsort (unexec_regions, num_unexec_regions, sizeof (unexec_regions[0]), | |
512 | &unexec_regions_sort_compare); | |
513 | n = 0; | |
514 | r = unexec_regions[0]; | |
ae7c60a9 YM |
515 | padsize = r.range.address & (pagesize - 1); |
516 | if (padsize) | |
517 | { | |
518 | r.range.address -= padsize; | |
519 | r.range.size += padsize; | |
520 | r.filesize += padsize; | |
521 | } | |
1dd7ccf2 AC |
522 | for (i = 1; i < num_unexec_regions; i++) |
523 | { | |
c57038f8 YM |
524 | if (r.range.address + r.range.size == unexec_regions[i].range.address |
525 | && r.range.size - r.filesize < 2 * pagesize) | |
1dd7ccf2 | 526 | { |
c57038f8 YM |
527 | r.filesize = r.range.size + unexec_regions[i].filesize; |
528 | r.range.size += unexec_regions[i].range.size; | |
1dd7ccf2 AC |
529 | } |
530 | else | |
531 | { | |
532 | unexec_regions[n++] = r; | |
533 | r = unexec_regions[i]; | |
ae7c60a9 YM |
534 | padsize = r.range.address & (pagesize - 1); |
535 | if (padsize) | |
536 | { | |
537 | if ((unexec_regions[n-1].range.address | |
538 | + unexec_regions[n-1].range.size) == r.range.address) | |
539 | unexec_regions[n-1].range.size -= padsize; | |
540 | ||
541 | r.range.address -= padsize; | |
542 | r.range.size += padsize; | |
543 | r.filesize += padsize; | |
544 | } | |
1dd7ccf2 AC |
545 | } |
546 | } | |
547 | unexec_regions[n++] = r; | |
548 | num_unexec_regions = n; | |
549 | } | |
550 | ||
e0f712ba AC |
551 | |
552 | /* More informational messages routines. */ | |
553 | ||
554 | static void | |
555 | print_load_command_name (int lc) | |
556 | { | |
557 | switch (lc) | |
558 | { | |
559 | case LC_SEGMENT: | |
73da71f9 | 560 | #ifndef _LP64 |
e0f712ba | 561 | printf ("LC_SEGMENT "); |
73da71f9 YM |
562 | #else |
563 | printf ("LC_SEGMENT_64 "); | |
564 | #endif | |
e0f712ba AC |
565 | break; |
566 | case LC_LOAD_DYLINKER: | |
567 | printf ("LC_LOAD_DYLINKER "); | |
568 | break; | |
569 | case LC_LOAD_DYLIB: | |
570 | printf ("LC_LOAD_DYLIB "); | |
571 | break; | |
572 | case LC_SYMTAB: | |
573 | printf ("LC_SYMTAB "); | |
574 | break; | |
575 | case LC_DYSYMTAB: | |
576 | printf ("LC_DYSYMTAB "); | |
577 | break; | |
578 | case LC_UNIXTHREAD: | |
579 | printf ("LC_UNIXTHREAD "); | |
580 | break; | |
581 | case LC_PREBOUND_DYLIB: | |
582 | printf ("LC_PREBOUND_DYLIB"); | |
583 | break; | |
584 | case LC_TWOLEVEL_HINTS: | |
585 | printf ("LC_TWOLEVEL_HINTS"); | |
586 | break; | |
ae7c60a9 YM |
587 | #ifdef LC_UUID |
588 | case LC_UUID: | |
589 | printf ("LC_UUID "); | |
590 | break; | |
6bcd6333 YM |
591 | #endif |
592 | #ifdef LC_DYLD_INFO | |
593 | case LC_DYLD_INFO: | |
594 | printf ("LC_DYLD_INFO "); | |
595 | break; | |
596 | case LC_DYLD_INFO_ONLY: | |
597 | printf ("LC_DYLD_INFO_ONLY"); | |
598 | break; | |
7963fa06 YM |
599 | #endif |
600 | #ifdef LC_VERSION_MIN_MACOSX | |
601 | case LC_VERSION_MIN_MACOSX: | |
602 | printf ("LC_VERSION_MIN_MACOSX"); | |
603 | break; | |
604 | #endif | |
605 | #ifdef LC_FUNCTION_STARTS | |
606 | case LC_FUNCTION_STARTS: | |
607 | printf ("LC_FUNCTION_STARTS"); | |
608 | break; | |
ae7c60a9 | 609 | #endif |
e0f712ba AC |
610 | default: |
611 | printf ("unknown "); | |
612 | } | |
613 | } | |
614 | ||
615 | static void | |
616 | print_load_command (struct load_command *lc) | |
617 | { | |
618 | print_load_command_name (lc->cmd); | |
619 | printf ("%8d", lc->cmdsize); | |
620 | ||
621 | if (lc->cmd == LC_SEGMENT) | |
622 | { | |
623 | struct segment_command *scp; | |
624 | struct section *sectp; | |
625 | int j; | |
626 | ||
627 | scp = (struct segment_command *) lc; | |
73da71f9 YM |
628 | printf (" %-16.16s %#10lx %#8lx\n", |
629 | scp->segname, (long) (scp->vmaddr), (long) (scp->vmsize)); | |
e0f712ba AC |
630 | |
631 | sectp = (struct section *) (scp + 1); | |
632 | for (j = 0; j < scp->nsects; j++) | |
633 | { | |
73da71f9 YM |
634 | printf (" %-16.16s %#10lx %#8lx\n", |
635 | sectp->sectname, (long) (sectp->addr), (long) (sectp->size)); | |
e0f712ba AC |
636 | sectp++; |
637 | } | |
638 | } | |
639 | else | |
640 | printf ("\n"); | |
641 | } | |
642 | ||
643 | /* Read header and load commands from input file. Store the latter in | |
644 | the global array lca. Store the total number of load commands in | |
645 | global variable nlc. */ | |
646 | static void | |
3d608a86 | 647 | read_load_commands (void) |
e0f712ba | 648 | { |
7f900522 | 649 | int i; |
e0f712ba AC |
650 | |
651 | if (!unexec_read (&mh, sizeof (struct mach_header))) | |
652 | unexec_error ("cannot read mach-o header"); | |
653 | ||
654 | if (mh.magic != MH_MAGIC) | |
655 | unexec_error ("input file not in Mach-O format"); | |
656 | ||
657 | if (mh.filetype != MH_EXECUTE) | |
658 | unexec_error ("input Mach-O file is not an executable object file"); | |
659 | ||
660 | #if VERBOSE | |
661 | printf ("--- Header Information ---\n"); | |
662 | printf ("Magic = 0x%08x\n", mh.magic); | |
663 | printf ("CPUType = %d\n", mh.cputype); | |
664 | printf ("CPUSubType = %d\n", mh.cpusubtype); | |
665 | printf ("FileType = 0x%x\n", mh.filetype); | |
666 | printf ("NCmds = %d\n", mh.ncmds); | |
667 | printf ("SizeOfCmds = %d\n", mh.sizeofcmds); | |
668 | printf ("Flags = 0x%08x\n", mh.flags); | |
669 | #endif | |
670 | ||
671 | nlc = mh.ncmds; | |
38182d90 | 672 | lca = malloc (nlc * sizeof *lca); |
177c0ea7 | 673 | |
e0f712ba AC |
674 | for (i = 0; i < nlc; i++) |
675 | { | |
676 | struct load_command lc; | |
677 | /* Load commands are variable-size: so read the command type and | |
678 | size first and then read the rest. */ | |
679 | if (!unexec_read (&lc, sizeof (struct load_command))) | |
680 | unexec_error ("cannot read load command"); | |
38182d90 | 681 | lca[i] = malloc (lc.cmdsize); |
e0f712ba AC |
682 | memcpy (lca[i], &lc, sizeof (struct load_command)); |
683 | if (!unexec_read (lca[i] + 1, lc.cmdsize - sizeof (struct load_command))) | |
684 | unexec_error ("cannot read content of load command"); | |
685 | if (lc.cmd == LC_SEGMENT) | |
686 | { | |
687 | struct segment_command *scp = (struct segment_command *) lca[i]; | |
177c0ea7 | 688 | |
e0f712ba AC |
689 | if (scp->vmaddr + scp->vmsize > infile_lc_highest_addr) |
690 | infile_lc_highest_addr = scp->vmaddr + scp->vmsize; | |
691 | ||
692 | if (strncmp (scp->segname, SEG_TEXT, 16) == 0) | |
693 | { | |
694 | struct section *sectp = (struct section *) (scp + 1); | |
695 | int j; | |
696 | ||
697 | for (j = 0; j < scp->nsects; j++) | |
698 | if (sectp->offset < text_seg_lowest_offset) | |
699 | text_seg_lowest_offset = sectp->offset; | |
700 | } | |
701 | } | |
702 | } | |
703 | ||
3d608a86 J |
704 | printf ("Highest address of load commands in input file: %#8lx\n", |
705 | (unsigned long)infile_lc_highest_addr); | |
e0f712ba | 706 | |
c57038f8 | 707 | printf ("Lowest offset of all sections in __TEXT segment: %#8lx\n", |
e0f712ba AC |
708 | text_seg_lowest_offset); |
709 | ||
710 | printf ("--- List of Load Commands in Input File ---\n"); | |
711 | printf ("# cmd cmdsize name address size\n"); | |
712 | ||
713 | for (i = 0; i < nlc; i++) | |
714 | { | |
715 | printf ("%1d ", i); | |
716 | print_load_command (lca[i]); | |
717 | } | |
718 | } | |
719 | ||
720 | /* Copy a LC_SEGMENT load command other than the __DATA segment from | |
721 | the input file to the output file, adjusting the file offset of the | |
722 | segment and the file offsets of sections contained in it. */ | |
723 | static void | |
724 | copy_segment (struct load_command *lc) | |
725 | { | |
726 | struct segment_command *scp = (struct segment_command *) lc; | |
727 | unsigned long old_fileoff = scp->fileoff; | |
728 | struct section *sectp; | |
729 | int j; | |
730 | ||
73da71f9 | 731 | scp->fileoff = curr_file_offset; |
e0f712ba AC |
732 | |
733 | sectp = (struct section *) (scp + 1); | |
734 | for (j = 0; j < scp->nsects; j++) | |
735 | { | |
73da71f9 | 736 | sectp->offset += curr_file_offset - old_fileoff; |
e0f712ba AC |
737 | sectp++; |
738 | } | |
739 | ||
c57038f8 YM |
740 | printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n", |
741 | scp->segname, (long) (scp->fileoff), (long) (scp->filesize), | |
742 | (long) (scp->vmsize), (long) (scp->vmaddr)); | |
e0f712ba AC |
743 | |
744 | if (!unexec_copy (scp->fileoff, old_fileoff, scp->filesize)) | |
745 | unexec_error ("cannot copy segment from input to output file"); | |
73da71f9 YM |
746 | curr_file_offset += ROUNDUP_TO_PAGE_BOUNDARY (scp->filesize); |
747 | ||
e0f712ba AC |
748 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) |
749 | unexec_error ("cannot write load command to header"); | |
750 | ||
751 | curr_header_offset += lc->cmdsize; | |
752 | } | |
753 | ||
754 | /* Copy a LC_SEGMENT load command for the __DATA segment in the input | |
755 | file to the output file. We assume that only one such segment load | |
756 | command exists in the input file and it contains the sections | |
757 | __data, __bss, __common, __la_symbol_ptr, __nl_symbol_ptr, and | |
758 | __dyld. The first three of these should be dumped from memory and | |
759 | the rest should be copied from the input file. Note that the | |
760 | sections __bss and __common contain no data in the input file | |
761 | because their flag fields have the value S_ZEROFILL. Dumping these | |
762 | from memory makes it necessary to adjust file offset fields in | |
763 | subsequently dumped load commands. Then, create new __DATA segment | |
764 | load commands for regions on the region list other than the one | |
765 | corresponding to the __DATA segment in the input file. */ | |
766 | static void | |
767 | copy_data_segment (struct load_command *lc) | |
768 | { | |
769 | struct segment_command *scp = (struct segment_command *) lc; | |
770 | struct section *sectp; | |
771 | int j; | |
c57038f8 YM |
772 | unsigned long header_offset, old_file_offset; |
773 | ||
774 | /* The new filesize of the segment is set to its vmsize because data | |
775 | blocks for segments must start at region boundaries. Note that | |
776 | this may leave unused locations at the end of the segment data | |
777 | block because the total of the sizes of all sections in the | |
778 | segment is generally smaller than vmsize. */ | |
779 | scp->filesize = scp->vmsize; | |
e0f712ba | 780 | |
c57038f8 YM |
781 | printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n", |
782 | scp->segname, curr_file_offset, (long)(scp->filesize), | |
783 | (long)(scp->vmsize), (long) (scp->vmaddr)); | |
e0f712ba AC |
784 | |
785 | /* Offsets in the output file for writing the next section structure | |
786 | and segment data block, respectively. */ | |
787 | header_offset = curr_header_offset + sizeof (struct segment_command); | |
788 | ||
789 | sectp = (struct section *) (scp + 1); | |
790 | for (j = 0; j < scp->nsects; j++) | |
791 | { | |
792 | old_file_offset = sectp->offset; | |
73da71f9 | 793 | sectp->offset = sectp->addr - scp->vmaddr + curr_file_offset; |
e0f712ba AC |
794 | /* The __data section is dumped from memory. The __bss and |
795 | __common sections are also dumped from memory but their flag | |
796 | fields require changing (from S_ZEROFILL to S_REGULAR). The | |
797 | other three kinds of sections are just copied from the input | |
798 | file. */ | |
799 | if (strncmp (sectp->sectname, SECT_DATA, 16) == 0) | |
800 | { | |
801 | if (!unexec_write (sectp->offset, (void *) sectp->addr, sectp->size)) | |
802 | unexec_error ("cannot write section %s", SECT_DATA); | |
803 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) | |
804 | unexec_error ("cannot write section %s's header", SECT_DATA); | |
805 | } | |
433456d7 | 806 | else if (strncmp (sectp->sectname, SECT_COMMON, 16) == 0) |
e0f712ba AC |
807 | { |
808 | sectp->flags = S_REGULAR; | |
809 | if (!unexec_write (sectp->offset, (void *) sectp->addr, sectp->size)) | |
f3dd7312 | 810 | unexec_error ("cannot write section %.16s", sectp->sectname); |
e0f712ba | 811 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) |
f3dd7312 | 812 | unexec_error ("cannot write section %.16s's header", sectp->sectname); |
433456d7 YM |
813 | } |
814 | else if (strncmp (sectp->sectname, SECT_BSS, 16) == 0) | |
815 | { | |
816 | extern char *my_endbss_static; | |
817 | unsigned long my_size; | |
818 | ||
819 | sectp->flags = S_REGULAR; | |
820 | ||
821 | /* Clear uninitialized local variables in statically linked | |
822 | libraries. In particular, function pointers stored by | |
823 | libSystemStub.a, which is introduced in Mac OS X 10.4 for | |
824 | binary compatibility with respect to long double, are | |
825 | cleared so that they will be reinitialized when the | |
826 | dumped binary is executed on other versions of OS. */ | |
827 | my_size = (unsigned long)my_endbss_static - sectp->addr; | |
828 | if (!(sectp->addr <= (unsigned long)my_endbss_static | |
829 | && my_size <= sectp->size)) | |
f3dd7312 | 830 | unexec_error ("my_endbss_static is not in section %.16s", |
433456d7 YM |
831 | sectp->sectname); |
832 | if (!unexec_write (sectp->offset, (void *) sectp->addr, my_size)) | |
f3dd7312 | 833 | unexec_error ("cannot write section %.16s", sectp->sectname); |
433456d7 YM |
834 | if (!unexec_write_zero (sectp->offset + my_size, |
835 | sectp->size - my_size)) | |
f3dd7312 | 836 | unexec_error ("cannot write section %.16s", sectp->sectname); |
433456d7 | 837 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) |
f3dd7312 | 838 | unexec_error ("cannot write section %.16s's header", sectp->sectname); |
e0f712ba AC |
839 | } |
840 | else if (strncmp (sectp->sectname, "__la_symbol_ptr", 16) == 0 | |
841 | || strncmp (sectp->sectname, "__nl_symbol_ptr", 16) == 0 | |
c47cbdfd | 842 | || strncmp (sectp->sectname, "__got", 16) == 0 |
427c5b1b | 843 | || strncmp (sectp->sectname, "__la_sym_ptr2", 16) == 0 |
e0f712ba | 844 | || strncmp (sectp->sectname, "__dyld", 16) == 0 |
7290a344 | 845 | || strncmp (sectp->sectname, "__const", 16) == 0 |
b2411edf YM |
846 | || strncmp (sectp->sectname, "__cfstring", 16) == 0 |
847 | || strncmp (sectp->sectname, "__gcc_except_tab", 16) == 0 | |
6bcd6333 | 848 | || strncmp (sectp->sectname, "__program_vars", 16) == 0 |
157e99e4 JW |
849 | || strncmp (sectp->sectname, "__mod_init_func", 16) == 0 |
850 | || strncmp (sectp->sectname, "__mod_term_func", 16) == 0 | |
b2411edf | 851 | || strncmp (sectp->sectname, "__objc_", 7) == 0) |
e0f712ba AC |
852 | { |
853 | if (!unexec_copy (sectp->offset, old_file_offset, sectp->size)) | |
f3dd7312 | 854 | unexec_error ("cannot copy section %.16s", sectp->sectname); |
e0f712ba | 855 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) |
f3dd7312 | 856 | unexec_error ("cannot write section %.16s's header", sectp->sectname); |
e0f712ba AC |
857 | } |
858 | else | |
f3dd7312 AS |
859 | unexec_error ("unrecognized section %.16s in __DATA segment", |
860 | sectp->sectname); | |
177c0ea7 | 861 | |
73da71f9 YM |
862 | printf (" section %-16.16s at %#8lx - %#8lx (sz: %#8lx)\n", |
863 | sectp->sectname, (long) (sectp->offset), | |
864 | (long) (sectp->offset + sectp->size), (long) (sectp->size)); | |
e0f712ba AC |
865 | |
866 | header_offset += sizeof (struct section); | |
867 | sectp++; | |
868 | } | |
869 | ||
c57038f8 YM |
870 | curr_file_offset += ROUNDUP_TO_PAGE_BOUNDARY (scp->filesize); |
871 | ||
e0f712ba AC |
872 | if (!unexec_write (curr_header_offset, scp, sizeof (struct segment_command))) |
873 | unexec_error ("cannot write header of __DATA segment"); | |
874 | curr_header_offset += lc->cmdsize; | |
875 | ||
876 | /* Create new __DATA segment load commands for regions on the region | |
877 | list that do not corresponding to any segment load commands in | |
878 | the input file. | |
73da71f9 | 879 | */ |
e0f712ba AC |
880 | for (j = 0; j < num_unexec_regions; j++) |
881 | { | |
882 | struct segment_command sc; | |
177c0ea7 | 883 | |
e0f712ba AC |
884 | sc.cmd = LC_SEGMENT; |
885 | sc.cmdsize = sizeof (struct segment_command); | |
886 | strncpy (sc.segname, SEG_DATA, 16); | |
c57038f8 YM |
887 | sc.vmaddr = unexec_regions[j].range.address; |
888 | sc.vmsize = unexec_regions[j].range.size; | |
73da71f9 | 889 | sc.fileoff = curr_file_offset; |
c57038f8 | 890 | sc.filesize = unexec_regions[j].filesize; |
e0f712ba AC |
891 | sc.maxprot = VM_PROT_READ | VM_PROT_WRITE; |
892 | sc.initprot = VM_PROT_READ | VM_PROT_WRITE; | |
893 | sc.nsects = 0; | |
894 | sc.flags = 0; | |
177c0ea7 | 895 | |
c57038f8 YM |
896 | printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n", |
897 | sc.segname, (long) (sc.fileoff), (long) (sc.filesize), | |
898 | (long) (sc.vmsize), (long) (sc.vmaddr)); | |
e0f712ba | 899 | |
c57038f8 | 900 | if (!unexec_write (sc.fileoff, (void *) sc.vmaddr, sc.filesize)) |
e0f712ba | 901 | unexec_error ("cannot write new __DATA segment"); |
73da71f9 | 902 | curr_file_offset += ROUNDUP_TO_PAGE_BOUNDARY (sc.filesize); |
177c0ea7 | 903 | |
e0f712ba AC |
904 | if (!unexec_write (curr_header_offset, &sc, sc.cmdsize)) |
905 | unexec_error ("cannot write new __DATA segment's header"); | |
906 | curr_header_offset += sc.cmdsize; | |
907 | mh.ncmds++; | |
908 | } | |
909 | } | |
910 | ||
911 | /* Copy a LC_SYMTAB load command from the input file to the output | |
912 | file, adjusting the file offset fields. */ | |
913 | static void | |
73da71f9 | 914 | copy_symtab (struct load_command *lc, long delta) |
e0f712ba AC |
915 | { |
916 | struct symtab_command *stp = (struct symtab_command *) lc; | |
917 | ||
918 | stp->symoff += delta; | |
919 | stp->stroff += delta; | |
920 | ||
921 | printf ("Writing LC_SYMTAB command\n"); | |
922 | ||
923 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
924 | unexec_error ("cannot write symtab command to header"); | |
925 | ||
926 | curr_header_offset += lc->cmdsize; | |
927 | } | |
928 | ||
043131c4 AC |
929 | /* Fix up relocation entries. */ |
930 | static void | |
7aee2da7 | 931 | unrelocate (const char *name, off_t reloff, int nrel, vm_address_t base) |
043131c4 AC |
932 | { |
933 | int i, unreloc_count; | |
934 | struct relocation_info reloc_info; | |
935 | struct scattered_relocation_info *sc_reloc_info | |
936 | = (struct scattered_relocation_info *) &reloc_info; | |
7aee2da7 | 937 | vm_address_t location; |
043131c4 AC |
938 | |
939 | for (unreloc_count = 0, i = 0; i < nrel; i++) | |
940 | { | |
941 | if (lseek (infd, reloff, L_SET) != reloff) | |
942 | unexec_error ("unrelocate: %s:%d cannot seek to reloc_info", name, i); | |
943 | if (!unexec_read (&reloc_info, sizeof (reloc_info))) | |
944 | unexec_error ("unrelocate: %s:%d cannot read reloc_info", name, i); | |
945 | reloff += sizeof (reloc_info); | |
946 | ||
947 | if (sc_reloc_info->r_scattered == 0) | |
948 | switch (reloc_info.r_type) | |
949 | { | |
950 | case GENERIC_RELOC_VANILLA: | |
7aee2da7 | 951 | location = base + reloc_info.r_address; |
b2411edf YM |
952 | if (location >= data_segment_scp->vmaddr |
953 | && location < (data_segment_scp->vmaddr | |
954 | + data_segment_scp->vmsize)) | |
043131c4 AC |
955 | { |
956 | off_t src_off = data_segment_old_fileoff | |
b2411edf | 957 | + (location - data_segment_scp->vmaddr); |
043131c4 | 958 | off_t dst_off = data_segment_scp->fileoff |
b2411edf | 959 | + (location - data_segment_scp->vmaddr); |
043131c4 AC |
960 | |
961 | if (!unexec_copy (dst_off, src_off, 1 << reloc_info.r_length)) | |
962 | unexec_error ("unrelocate: %s:%d cannot copy original value", | |
963 | name, i); | |
964 | unreloc_count++; | |
965 | } | |
966 | break; | |
967 | default: | |
968 | unexec_error ("unrelocate: %s:%d cannot handle type = %d", | |
969 | name, i, reloc_info.r_type); | |
970 | } | |
971 | else | |
972 | switch (sc_reloc_info->r_type) | |
973 | { | |
974 | #if defined (__ppc__) | |
975 | case PPC_RELOC_PB_LA_PTR: | |
976 | /* nothing to do for prebound lazy pointer */ | |
977 | break; | |
978 | #endif | |
979 | default: | |
980 | unexec_error ("unrelocate: %s:%d cannot handle scattered type = %d", | |
981 | name, i, sc_reloc_info->r_type); | |
982 | } | |
983 | } | |
984 | ||
985 | if (nrel > 0) | |
986 | printf ("Fixed up %d/%d %s relocation entries in data segment.\n", | |
987 | unreloc_count, nrel, name); | |
988 | } | |
989 | ||
7aee2da7 YM |
990 | #if __ppc64__ |
991 | /* Rebase r_address in the relocation table. */ | |
992 | static void | |
993 | rebase_reloc_address (off_t reloff, int nrel, long linkedit_delta, long diff) | |
994 | { | |
995 | int i; | |
996 | struct relocation_info reloc_info; | |
997 | struct scattered_relocation_info *sc_reloc_info | |
998 | = (struct scattered_relocation_info *) &reloc_info; | |
999 | ||
1000 | for (i = 0; i < nrel; i++, reloff += sizeof (reloc_info)) | |
1001 | { | |
1002 | if (lseek (infd, reloff - linkedit_delta, L_SET) | |
1003 | != reloff - linkedit_delta) | |
1004 | unexec_error ("rebase_reloc_table: cannot seek to reloc_info"); | |
1005 | if (!unexec_read (&reloc_info, sizeof (reloc_info))) | |
1006 | unexec_error ("rebase_reloc_table: cannot read reloc_info"); | |
1007 | ||
1008 | if (sc_reloc_info->r_scattered == 0 | |
1009 | && reloc_info.r_type == GENERIC_RELOC_VANILLA) | |
1010 | { | |
1011 | reloc_info.r_address -= diff; | |
1012 | if (!unexec_write (reloff, &reloc_info, sizeof (reloc_info))) | |
1013 | unexec_error ("rebase_reloc_table: cannot write reloc_info"); | |
1014 | } | |
1015 | } | |
1016 | } | |
1017 | #endif | |
1018 | ||
e0f712ba AC |
1019 | /* Copy a LC_DYSYMTAB load command from the input file to the output |
1020 | file, adjusting the file offset fields. */ | |
1021 | static void | |
73da71f9 | 1022 | copy_dysymtab (struct load_command *lc, long delta) |
e0f712ba AC |
1023 | { |
1024 | struct dysymtab_command *dstp = (struct dysymtab_command *) lc; | |
7aee2da7 | 1025 | vm_address_t base; |
e0f712ba | 1026 | |
7aee2da7 YM |
1027 | #ifdef _LP64 |
1028 | #if __ppc64__ | |
1029 | { | |
1030 | int i; | |
1031 | ||
1032 | base = 0; | |
1033 | for (i = 0; i < nlc; i++) | |
1034 | if (lca[i]->cmd == LC_SEGMENT) | |
1035 | { | |
1036 | struct segment_command *scp = (struct segment_command *) lca[i]; | |
1037 | ||
1038 | if (scp->vmaddr + scp->vmsize > 0x100000000 | |
1039 | && (scp->initprot & VM_PROT_WRITE) != 0) | |
1040 | { | |
1041 | base = data_segment_scp->vmaddr; | |
1042 | break; | |
1043 | } | |
1044 | } | |
1045 | } | |
1046 | #else | |
1047 | /* First writable segment address. */ | |
1048 | base = data_segment_scp->vmaddr; | |
1049 | #endif | |
1050 | #else | |
1051 | /* First segment address in the file (unless MH_SPLIT_SEGS set). */ | |
1052 | base = 0; | |
1053 | #endif | |
1054 | ||
1055 | unrelocate ("local", dstp->locreloff, dstp->nlocrel, base); | |
1056 | unrelocate ("external", dstp->extreloff, dstp->nextrel, base); | |
e0f712ba AC |
1057 | |
1058 | if (dstp->nextrel > 0) { | |
1059 | dstp->extreloff += delta; | |
1060 | } | |
1061 | ||
1062 | if (dstp->nlocrel > 0) { | |
1063 | dstp->locreloff += delta; | |
1064 | } | |
1065 | ||
1066 | if (dstp->nindirectsyms > 0) | |
1067 | dstp->indirectsymoff += delta; | |
1068 | ||
1069 | printf ("Writing LC_DYSYMTAB command\n"); | |
1070 | ||
1071 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
1072 | unexec_error ("cannot write symtab command to header"); | |
1073 | ||
1074 | curr_header_offset += lc->cmdsize; | |
7aee2da7 YM |
1075 | |
1076 | #if __ppc64__ | |
1077 | /* Check if the relocation base needs to be changed. */ | |
1078 | if (base == 0) | |
1079 | { | |
1080 | vm_address_t newbase = 0; | |
1081 | int i; | |
1082 | ||
1083 | for (i = 0; i < num_unexec_regions; i++) | |
1084 | if (unexec_regions[i].range.address + unexec_regions[i].range.size | |
1085 | > 0x100000000) | |
1086 | { | |
1087 | newbase = data_segment_scp->vmaddr; | |
1088 | break; | |
1089 | } | |
1090 | ||
1091 | if (newbase) | |
1092 | { | |
1093 | rebase_reloc_address (dstp->locreloff, dstp->nlocrel, delta, newbase); | |
1094 | rebase_reloc_address (dstp->extreloff, dstp->nextrel, delta, newbase); | |
1095 | } | |
1096 | } | |
1097 | #endif | |
e0f712ba AC |
1098 | } |
1099 | ||
40e6ff95 ST |
1100 | /* Copy a LC_TWOLEVEL_HINTS load command from the input file to the output |
1101 | file, adjusting the file offset fields. */ | |
1102 | static void | |
73da71f9 | 1103 | copy_twolevelhints (struct load_command *lc, long delta) |
40e6ff95 ST |
1104 | { |
1105 | struct twolevel_hints_command *tlhp = (struct twolevel_hints_command *) lc; | |
1106 | ||
1107 | if (tlhp->nhints > 0) { | |
1108 | tlhp->offset += delta; | |
1109 | } | |
1110 | ||
1111 | printf ("Writing LC_TWOLEVEL_HINTS command\n"); | |
1112 | ||
1113 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
1114 | unexec_error ("cannot write two level hint command to header"); | |
1115 | ||
1116 | curr_header_offset += lc->cmdsize; | |
1117 | } | |
1118 | ||
6bcd6333 YM |
1119 | #ifdef LC_DYLD_INFO |
1120 | /* Copy a LC_DYLD_INFO(_ONLY) load command from the input file to the output | |
1121 | file, adjusting the file offset fields. */ | |
1122 | static void | |
1123 | copy_dyld_info (struct load_command *lc, long delta) | |
1124 | { | |
1125 | struct dyld_info_command *dip = (struct dyld_info_command *) lc; | |
1126 | ||
1127 | if (dip->rebase_off > 0) | |
1128 | dip->rebase_off += delta; | |
1129 | if (dip->bind_off > 0) | |
1130 | dip->bind_off += delta; | |
1131 | if (dip->weak_bind_off > 0) | |
1132 | dip->weak_bind_off += delta; | |
1133 | if (dip->lazy_bind_off > 0) | |
1134 | dip->lazy_bind_off += delta; | |
1135 | if (dip->export_off > 0) | |
1136 | dip->export_off += delta; | |
1137 | ||
1138 | printf ("Writing "); | |
1139 | print_load_command_name (lc->cmd); | |
1140 | printf (" command\n"); | |
1141 | ||
1142 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
1143 | unexec_error ("cannot write dyld info command to header"); | |
1144 | ||
1145 | curr_header_offset += lc->cmdsize; | |
1146 | } | |
1147 | #endif | |
1148 | ||
7963fa06 YM |
1149 | #ifdef LC_FUNCTION_STARTS |
1150 | /* Copy a LC_FUNCTION_STARTS load command from the input file to the | |
1151 | output file, adjusting the data offset field. */ | |
1152 | static void | |
1153 | copy_linkedit_data (struct load_command *lc, long delta) | |
1154 | { | |
1155 | struct linkedit_data_command *ldp = (struct linkedit_data_command *) lc; | |
1156 | ||
1157 | if (ldp->dataoff > 0) | |
1158 | ldp->dataoff += delta; | |
1159 | ||
1160 | printf ("Writing "); | |
1161 | print_load_command_name (lc->cmd); | |
1162 | printf (" command\n"); | |
1163 | ||
1164 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
1165 | unexec_error ("cannot write linkedit data command to header"); | |
1166 | ||
1167 | curr_header_offset += lc->cmdsize; | |
1168 | } | |
1169 | #endif | |
1170 | ||
e0f712ba AC |
1171 | /* Copy other kinds of load commands from the input file to the output |
1172 | file, ones that do not require adjustments of file offsets. */ | |
1173 | static void | |
1174 | copy_other (struct load_command *lc) | |
1175 | { | |
1176 | printf ("Writing "); | |
1177 | print_load_command_name (lc->cmd); | |
1178 | printf (" command\n"); | |
1179 | ||
1180 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
1181 | unexec_error ("cannot write symtab command to header"); | |
1182 | ||
1183 | curr_header_offset += lc->cmdsize; | |
1184 | } | |
1185 | ||
1186 | /* Loop through all load commands and dump them. Then write the Mach | |
1187 | header. */ | |
1188 | static void | |
3d608a86 | 1189 | dump_it (void) |
e0f712ba AC |
1190 | { |
1191 | int i; | |
73da71f9 | 1192 | long linkedit_delta = 0; |
e0f712ba AC |
1193 | |
1194 | printf ("--- Load Commands written to Output File ---\n"); | |
1195 | ||
1196 | for (i = 0; i < nlc; i++) | |
1197 | switch (lca[i]->cmd) | |
1198 | { | |
1199 | case LC_SEGMENT: | |
1200 | { | |
1201 | struct segment_command *scp = (struct segment_command *) lca[i]; | |
1202 | if (strncmp (scp->segname, SEG_DATA, 16) == 0) | |
1203 | { | |
043131c4 AC |
1204 | /* save data segment file offset and segment_command for |
1205 | unrelocate */ | |
73da71f9 YM |
1206 | if (data_segment_old_fileoff) |
1207 | unexec_error ("cannot handle multiple DATA segments" | |
1208 | " in input file"); | |
043131c4 AC |
1209 | data_segment_old_fileoff = scp->fileoff; |
1210 | data_segment_scp = scp; | |
1211 | ||
e0f712ba AC |
1212 | copy_data_segment (lca[i]); |
1213 | } | |
1214 | else | |
1215 | { | |
73da71f9 YM |
1216 | if (strncmp (scp->segname, SEG_LINKEDIT, 16) == 0) |
1217 | { | |
1218 | if (linkedit_delta) | |
1219 | unexec_error ("cannot handle multiple LINKEDIT segments" | |
1220 | " in input file"); | |
1221 | linkedit_delta = curr_file_offset - scp->fileoff; | |
1222 | } | |
1223 | ||
e0f712ba AC |
1224 | copy_segment (lca[i]); |
1225 | } | |
1226 | } | |
1227 | break; | |
1228 | case LC_SYMTAB: | |
73da71f9 | 1229 | copy_symtab (lca[i], linkedit_delta); |
e0f712ba AC |
1230 | break; |
1231 | case LC_DYSYMTAB: | |
73da71f9 | 1232 | copy_dysymtab (lca[i], linkedit_delta); |
e0f712ba | 1233 | break; |
40e6ff95 | 1234 | case LC_TWOLEVEL_HINTS: |
73da71f9 | 1235 | copy_twolevelhints (lca[i], linkedit_delta); |
40e6ff95 | 1236 | break; |
6bcd6333 YM |
1237 | #ifdef LC_DYLD_INFO |
1238 | case LC_DYLD_INFO: | |
1239 | case LC_DYLD_INFO_ONLY: | |
1240 | copy_dyld_info (lca[i], linkedit_delta); | |
1241 | break; | |
7963fa06 YM |
1242 | #endif |
1243 | #ifdef LC_FUNCTION_STARTS | |
1244 | case LC_FUNCTION_STARTS: | |
1245 | copy_linkedit_data (lca[i], linkedit_delta); | |
1246 | break; | |
6bcd6333 | 1247 | #endif |
e0f712ba AC |
1248 | default: |
1249 | copy_other (lca[i]); | |
1250 | break; | |
1251 | } | |
1252 | ||
1253 | if (curr_header_offset > text_seg_lowest_offset) | |
1254 | unexec_error ("not enough room for load commands for new __DATA segments"); | |
1255 | ||
c57038f8 | 1256 | printf ("%ld unused bytes follow Mach-O header\n", |
e0f712ba AC |
1257 | text_seg_lowest_offset - curr_header_offset); |
1258 | ||
1259 | mh.sizeofcmds = curr_header_offset - sizeof (struct mach_header); | |
1260 | if (!unexec_write (0, &mh, sizeof (struct mach_header))) | |
1261 | unexec_error ("cannot write final header contents"); | |
1262 | } | |
1263 | ||
1264 | /* Take a snapshot of Emacs and make a Mach-O format executable file | |
1265 | from it. The file names of the output and input files are outfile | |
1266 | and infile, respectively. The three other parameters are | |
1267 | ignored. */ | |
381259ef | 1268 | void |
dd5ecd6b | 1269 | unexec (const char *outfile, const char *infile) |
e0f712ba | 1270 | { |
6dc5c8a7 YM |
1271 | if (in_dumped_exec) |
1272 | unexec_error ("Unexec from a dumped executable is not supported."); | |
1273 | ||
73da71f9 | 1274 | pagesize = getpagesize (); |
e0f712ba AC |
1275 | infd = open (infile, O_RDONLY, 0); |
1276 | if (infd < 0) | |
1277 | { | |
1278 | unexec_error ("cannot open input file `%s'", infile); | |
1279 | } | |
177c0ea7 | 1280 | |
e0f712ba AC |
1281 | outfd = open (outfile, O_WRONLY | O_TRUNC | O_CREAT, 0755); |
1282 | if (outfd < 0) | |
1283 | { | |
1284 | close (infd); | |
1285 | unexec_error ("cannot open output file `%s'", outfile); | |
1286 | } | |
1287 | ||
1288 | build_region_list (); | |
1289 | read_load_commands (); | |
1290 | ||
1291 | find_emacs_zone_regions (); | |
1dd7ccf2 | 1292 | unexec_regions_merge (); |
e0f712ba AC |
1293 | |
1294 | in_dumped_exec = 1; | |
1295 | ||
1296 | dump_it (); | |
1297 | ||
1298 | close (outfd); | |
1299 | } | |
1300 | ||
1301 | ||
1302 | void | |
3d608a86 | 1303 | unexec_init_emacs_zone (void) |
e0f712ba AC |
1304 | { |
1305 | emacs_zone = malloc_create_zone (0, 0); | |
1306 | malloc_set_zone_name (emacs_zone, "EmacsZone"); | |
1307 | } | |
1308 | ||
40ef0695 YM |
1309 | #ifndef MACOSX_MALLOC_MULT16 |
1310 | #define MACOSX_MALLOC_MULT16 1 | |
1311 | #endif | |
1312 | ||
1313 | typedef struct unexec_malloc_header { | |
1314 | union { | |
1315 | char c[8]; | |
1316 | size_t size; | |
1317 | } u; | |
1318 | } unexec_malloc_header_t; | |
1319 | ||
1320 | #if MACOSX_MALLOC_MULT16 | |
1321 | ||
1322 | #define ptr_in_unexec_regions(p) ((((vm_address_t) (p)) & 8) != 0) | |
1323 | ||
1324 | #else | |
1325 | ||
e0f712ba AC |
1326 | int |
1327 | ptr_in_unexec_regions (void *ptr) | |
1328 | { | |
1329 | int i; | |
1330 | ||
1331 | for (i = 0; i < num_unexec_regions; i++) | |
c57038f8 YM |
1332 | if ((vm_address_t) ptr - unexec_regions[i].range.address |
1333 | < unexec_regions[i].range.size) | |
e0f712ba AC |
1334 | return 1; |
1335 | ||
1336 | return 0; | |
1337 | } | |
1338 | ||
40ef0695 YM |
1339 | #endif |
1340 | ||
e0f712ba AC |
1341 | void * |
1342 | unexec_malloc (size_t size) | |
1343 | { | |
1344 | if (in_dumped_exec) | |
40ef0695 YM |
1345 | { |
1346 | void *p; | |
1347 | ||
1348 | p = malloc (size); | |
1349 | #if MACOSX_MALLOC_MULT16 | |
1350 | assert (((vm_address_t) p % 16) == 0); | |
1351 | #endif | |
1352 | return p; | |
1353 | } | |
e0f712ba | 1354 | else |
40ef0695 YM |
1355 | { |
1356 | unexec_malloc_header_t *ptr; | |
1357 | ||
1358 | ptr = (unexec_malloc_header_t *) | |
1359 | malloc_zone_malloc (emacs_zone, size + sizeof (unexec_malloc_header_t)); | |
1360 | ptr->u.size = size; | |
1361 | ptr++; | |
1362 | #if MACOSX_MALLOC_MULT16 | |
1363 | assert (((vm_address_t) ptr % 16) == 8); | |
1364 | #endif | |
1365 | return (void *) ptr; | |
1366 | } | |
e0f712ba AC |
1367 | } |
1368 | ||
1369 | void * | |
1370 | unexec_realloc (void *old_ptr, size_t new_size) | |
1371 | { | |
1372 | if (in_dumped_exec) | |
40ef0695 YM |
1373 | { |
1374 | void *p; | |
1375 | ||
1376 | if (ptr_in_unexec_regions (old_ptr)) | |
1377 | { | |
40ef0695 YM |
1378 | size_t old_size = ((unexec_malloc_header_t *) old_ptr)[-1].u.size; |
1379 | size_t size = new_size > old_size ? old_size : new_size; | |
1380 | ||
38182d90 | 1381 | p = malloc (new_size); |
40ef0695 YM |
1382 | if (size) |
1383 | memcpy (p, old_ptr, size); | |
1384 | } | |
1385 | else | |
1386 | { | |
1387 | p = realloc (old_ptr, new_size); | |
1388 | } | |
1389 | #if MACOSX_MALLOC_MULT16 | |
1390 | assert (((vm_address_t) p % 16) == 0); | |
1391 | #endif | |
1392 | return p; | |
1393 | } | |
e0f712ba | 1394 | else |
40ef0695 YM |
1395 | { |
1396 | unexec_malloc_header_t *ptr; | |
1397 | ||
1398 | ptr = (unexec_malloc_header_t *) | |
1399 | malloc_zone_realloc (emacs_zone, (unexec_malloc_header_t *) old_ptr - 1, | |
1400 | new_size + sizeof (unexec_malloc_header_t)); | |
1401 | ptr->u.size = new_size; | |
1402 | ptr++; | |
1403 | #if MACOSX_MALLOC_MULT16 | |
1404 | assert (((vm_address_t) ptr % 16) == 8); | |
1405 | #endif | |
1406 | return (void *) ptr; | |
1407 | } | |
e0f712ba AC |
1408 | } |
1409 | ||
1410 | void | |
1411 | unexec_free (void *ptr) | |
1412 | { | |
9c5e177e JM |
1413 | if (ptr == NULL) |
1414 | return; | |
e0f712ba AC |
1415 | if (in_dumped_exec) |
1416 | { | |
1417 | if (!ptr_in_unexec_regions (ptr)) | |
1418 | free (ptr); | |
1419 | } | |
1420 | else | |
40ef0695 | 1421 | malloc_zone_free (emacs_zone, (unexec_malloc_header_t *) ptr - 1); |
e0f712ba | 1422 | } |