Commit | Line | Data |
---|---|---|
e0f712ba AC |
1 | /* Dump Emacs in Mach-O format for use on Mac OS X. |
2 | Copyright (C) 2001, 2002 Free Software Foundation, Inc. | |
3 | ||
4 | This file is part of GNU Emacs. | |
5 | ||
6 | GNU Emacs is free software; you can redistribute it and/or modify | |
7 | it under the terms of the GNU General Public License as published by | |
8 | the Free Software Foundation; either version 2, or (at your option) | |
9 | any later version. | |
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 | |
17 | along with GNU Emacs; see the file COPYING. If not, write to | |
18 | the Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
19 | Boston, MA 02111-1307, USA. */ | |
20 | ||
21 | /* Contributed by Andrew Choi (akochoi@mac.com). */ | |
22 | ||
23 | /* Documentation note. | |
24 | ||
25 | Consult the following documents/files for a description of the | |
26 | Mach-O format: the file loader.h, man pages for Mach-O and ld, old | |
27 | NEXTSTEP documents of the Mach-O format. The tool otool dumps the | |
28 | mach header (-h option) and the load commands (-l option) in a | |
29 | Mach-O file. The tool nm on Mac OS X displays the symbol table in | |
30 | a Mach-O file. For examples of unexec for the Mach-O format, see | |
31 | the file unexnext.c in the GNU Emacs distribution, the file | |
32 | unexdyld.c in the Darwin port of GNU Emacs 20.7, and unexdyld.c in | |
33 | the Darwin port of XEmacs 21.1. Also the Darwin Libc source | |
34 | contains the source code for malloc_freezedry and malloc_jumpstart. | |
35 | Read that to see what they do. This file was written completely | |
36 | from scratch, making use of information from the above sources. */ | |
37 | ||
38 | /* The Mac OS X implementation of unexec makes use of Darwin's `zone' | |
39 | memory allocator. All calls to malloc, realloc, and free in Emacs | |
40 | are redirected to unexec_malloc, unexec_realloc, and unexec_free in | |
41 | this file. When temacs is run, all memory requests are handled in | |
42 | the zone EmacsZone. The Darwin memory allocator library calls | |
43 | maintain the data structures to manage this zone. Dumping writes | |
44 | its contents to data segments of the executable file. When emacs | |
45 | is run, the loader recreates the contents of the zone in memory. | |
46 | However since the initialization routine of the zone memory | |
47 | allocator is run again, this `zone' can no longer be used as a | |
48 | heap. That is why emacs uses the ordinary malloc system call to | |
49 | allocate memory. Also, when a block of memory needs to be | |
50 | reallocated and the new size is larger than the old one, a new | |
51 | block must be obtained by malloc and the old contents copied to | |
52 | it. */ | |
53 | ||
54 | /* Peculiarity of the Mach-O files generated by ld in Mac OS X | |
55 | (possible causes of future bugs if changed). | |
56 | ||
57 | The file offset of the start of the __TEXT segment is zero. Since | |
58 | the Mach header and load commands are located at the beginning of a | |
59 | Mach-O file, copying the contents of the __TEXT segment from the | |
60 | input file overwrites them in the output file. Despite this, | |
61 | unexec works fine as written below because the segment load command | |
62 | for __TEXT appears, and is therefore processed, before all other | |
63 | load commands except the segment load command for __PAGEZERO, which | |
64 | remains unchanged. | |
65 | ||
66 | Although the file offset of the start of the __TEXT segment is | |
67 | zero, none of the sections it contains actually start there. In | |
68 | fact, the earliest one starts a few hundred bytes beyond the end of | |
69 | the last load command. The linker option -headerpad controls the | |
70 | minimum size of this padding. Its setting can be changed in | |
71 | s/darwin.h. A value of 0x300, e.g., leaves room for about 15 | |
72 | additional load commands for the newly created __DATA segments (at | |
73 | 56 bytes each). Unexec fails if there is not enough room for these | |
74 | new segments. | |
75 | ||
76 | The __TEXT segment contains the sections __text, __cstring, | |
77 | __picsymbol_stub, and __const and the __DATA segment contains the | |
78 | sections __data, __la_symbol_ptr, __nl_symbol_ptr, __dyld, __bss, | |
79 | and __common. The other segments do not contain any sections. | |
80 | These sections are copied from the input file to the output file, | |
81 | except for __data, __bss, and __common, which are dumped from | |
82 | memory. The types of the sections __bss and __common are changed | |
83 | from S_ZEROFILL to S_REGULAR. Note that the number of sections and | |
84 | their relative order in the input and output files remain | |
85 | unchanged. Otherwise all n_sect fields in the nlist records in the | |
86 | symbol table (specified by the LC_SYMTAB load command) will have to | |
87 | be changed accordingly. | |
88 | */ | |
89 | ||
90 | #include <stdio.h> | |
91 | #include <stdlib.h> | |
92 | #include <fcntl.h> | |
93 | #include <stdarg.h> | |
94 | #include <sys/types.h> | |
95 | #include <unistd.h> | |
96 | #include <mach/mach.h> | |
97 | #include <mach-o/loader.h> | |
98 | #include <objc/malloc.h> | |
99 | ||
100 | #define VERBOSE 1 | |
101 | ||
102 | /* Size of buffer used to copy data from the input file to the output | |
103 | file in function unexec_copy. */ | |
104 | #define UNEXEC_COPY_BUFSZ 1024 | |
105 | ||
106 | /* Regions with memory addresses above this value are assumed to be | |
107 | mapped to dynamically loaded libraries and will not be dumped. */ | |
108 | #define VM_DATA_TOP (20 * 1024 * 1024) | |
109 | ||
110 | /* Used by malloc_freezedry and malloc_jumpstart. */ | |
111 | int malloc_cookie; | |
112 | ||
113 | /* Type of an element on the list of regions to be dumped. */ | |
114 | struct region_t { | |
115 | vm_address_t address; | |
116 | vm_size_t size; | |
117 | vm_prot_t protection; | |
118 | vm_prot_t max_protection; | |
119 | ||
120 | struct region_t *next; | |
121 | }; | |
122 | ||
123 | /* Head and tail of the list of regions to be dumped. */ | |
124 | struct region_t *region_list_head = 0; | |
125 | struct region_t *region_list_tail = 0; | |
126 | ||
127 | /* Pointer to array of load commands. */ | |
128 | struct load_command **lca; | |
129 | ||
130 | /* Number of load commands. */ | |
131 | int nlc; | |
132 | ||
133 | /* The highest VM address of segments loaded by the input file. | |
134 | Regions with addresses beyond this are assumed to be allocated | |
135 | dynamically and thus require dumping. */ | |
136 | vm_address_t infile_lc_highest_addr = 0; | |
137 | ||
138 | /* The lowest file offset used by the all sections in the __TEXT | |
139 | segments. This leaves room at the beginning of the file to store | |
140 | the Mach-O header. Check this value against header size to ensure | |
141 | the added load commands for the new __DATA segments did not | |
142 | overwrite any of the sections in the __TEXT segment. */ | |
143 | unsigned long text_seg_lowest_offset = 0x10000000; | |
144 | ||
145 | /* Mach header. */ | |
146 | struct mach_header mh; | |
147 | ||
148 | /* Offset at which the next load command should be written. */ | |
149 | unsigned long curr_header_offset = sizeof (struct mach_header); | |
150 | ||
151 | /* Current adjustment that needs to be made to offset values because | |
152 | of additional data segments. */ | |
153 | unsigned long delta = 0; | |
154 | ||
155 | int infd, outfd; | |
156 | ||
157 | int in_dumped_exec = 0; | |
158 | ||
159 | malloc_zone_t *emacs_zone; | |
160 | ||
161 | /* Read n bytes from infd into memory starting at address dest. | |
162 | Return true if successful, false otherwise. */ | |
163 | static int | |
164 | unexec_read (void *dest, size_t n) | |
165 | { | |
166 | return n == read (infd, dest, n); | |
167 | } | |
168 | ||
169 | /* Write n bytes from memory starting at address src to outfd starting | |
170 | at offset dest. Return true if successful, false otherwise. */ | |
171 | static int | |
172 | unexec_write (off_t dest, const void *src, size_t count) | |
173 | { | |
174 | if (lseek (outfd, dest, SEEK_SET) != dest) | |
175 | return 0; | |
176 | ||
177 | return write (outfd, src, count) == count; | |
178 | } | |
179 | ||
180 | /* Copy n bytes from starting offset src in infd to starting offset | |
181 | dest in outfd. Return true if successful, false otherwise. */ | |
182 | static int | |
183 | unexec_copy (off_t dest, off_t src, ssize_t count) | |
184 | { | |
185 | ssize_t bytes_read; | |
186 | ||
187 | char buf[UNEXEC_COPY_BUFSZ]; | |
188 | ||
189 | if (lseek (infd, src, SEEK_SET) != src) | |
190 | return 0; | |
191 | ||
192 | if (lseek (outfd, dest, SEEK_SET) != dest) | |
193 | return 0; | |
194 | ||
195 | while (count > 0) | |
196 | { | |
197 | bytes_read = read (infd, buf, UNEXEC_COPY_BUFSZ); | |
198 | if (bytes_read <= 0) | |
199 | return 0; | |
200 | if (write (outfd, buf, bytes_read) != bytes_read) | |
201 | return 0; | |
202 | count -= bytes_read; | |
203 | } | |
204 | ||
205 | return 1; | |
206 | } | |
207 | ||
208 | /* Debugging and informational messages routines. */ | |
209 | ||
210 | static void | |
211 | unexec_error (char *format, ...) | |
212 | { | |
213 | va_list ap; | |
214 | ||
215 | va_start (ap, format); | |
216 | fprintf (stderr, "unexec: "); | |
217 | vfprintf (stderr, format, ap); | |
218 | fprintf (stderr, "\n"); | |
219 | va_end (ap); | |
220 | exit (1); | |
221 | } | |
222 | ||
223 | static void | |
224 | print_prot (vm_prot_t prot) | |
225 | { | |
226 | if (prot == VM_PROT_NONE) | |
227 | printf ("none"); | |
228 | else | |
229 | { | |
230 | putchar (prot & VM_PROT_READ ? 'r' : ' '); | |
231 | putchar (prot & VM_PROT_WRITE ? 'w' : ' '); | |
232 | putchar (prot & VM_PROT_EXECUTE ? 'x' : ' '); | |
233 | putchar (' '); | |
234 | } | |
235 | } | |
236 | ||
237 | static void | |
238 | print_region (vm_address_t address, vm_size_t size, vm_prot_t prot, | |
239 | vm_prot_t max_prot) | |
240 | { | |
241 | printf ("%#10x %#8x ", address, size); | |
242 | print_prot (prot); | |
243 | putchar (' '); | |
244 | print_prot (max_prot); | |
245 | putchar ('\n'); | |
246 | } | |
247 | ||
248 | static void | |
249 | print_region_list () | |
250 | { | |
251 | struct region_t *r; | |
252 | ||
253 | printf (" address size prot maxp\n"); | |
254 | ||
255 | for (r = region_list_head; r; r = r->next) | |
256 | print_region (r->address, r->size, r->protection, r->max_protection); | |
257 | } | |
258 | ||
259 | void | |
260 | print_regions () | |
261 | { | |
262 | task_t target_task = mach_task_self (); | |
263 | vm_address_t address = (vm_address_t) 0; | |
264 | vm_size_t size; | |
265 | struct vm_region_basic_info info; | |
266 | mach_msg_type_number_t info_count = VM_REGION_BASIC_INFO_COUNT; | |
267 | mach_port_t object_name; | |
268 | ||
269 | printf (" address size prot maxp\n"); | |
270 | ||
271 | while (vm_region (target_task, &address, &size, VM_REGION_BASIC_INFO, | |
272 | (vm_region_info_t) &info, &info_count, &object_name) | |
273 | == KERN_SUCCESS && info_count == VM_REGION_BASIC_INFO_COUNT) | |
274 | { | |
275 | print_region (address, size, info.protection, info.max_protection); | |
276 | ||
277 | if (object_name != MACH_PORT_NULL) | |
278 | mach_port_deallocate (target_task, object_name); | |
279 | ||
280 | address += size; | |
281 | } | |
282 | } | |
283 | ||
284 | /* Build the list of regions that need to be dumped. Regions with | |
285 | addresses above VM_DATA_TOP are omitted. Adjacent regions with | |
286 | identical protection are merged. Note that non-writable regions | |
287 | cannot be omitted because they some regions created at run time are | |
288 | read-only. */ | |
289 | static void | |
290 | build_region_list () | |
291 | { | |
292 | task_t target_task = mach_task_self (); | |
293 | vm_address_t address = (vm_address_t) 0; | |
294 | vm_size_t size; | |
295 | struct vm_region_basic_info info; | |
296 | mach_msg_type_number_t info_count = VM_REGION_BASIC_INFO_COUNT; | |
297 | mach_port_t object_name; | |
298 | struct region_t *r; | |
299 | ||
300 | #if VERBOSE | |
301 | printf ("--- List of All Regions ---\n"); | |
302 | printf (" address size prot maxp\n"); | |
303 | #endif | |
304 | ||
305 | while (vm_region (target_task, &address, &size, VM_REGION_BASIC_INFO, | |
306 | (vm_region_info_t) &info, &info_count, &object_name) | |
307 | == KERN_SUCCESS && info_count == VM_REGION_BASIC_INFO_COUNT) | |
308 | { | |
309 | /* Done when we reach addresses of shared libraries, which are | |
310 | loaded in high memory. */ | |
311 | if (address >= VM_DATA_TOP) | |
312 | break; | |
313 | ||
314 | #if VERBOSE | |
315 | print_region (address, size, info.protection, info.max_protection); | |
316 | #endif | |
317 | ||
318 | /* If a region immediately follows the previous one (the one | |
319 | most recently added to the list) and has identical | |
320 | protection, merge it with the latter. Otherwise create a | |
321 | new list element for it. */ | |
322 | if (region_list_tail | |
323 | && info.protection == region_list_tail->protection | |
324 | && info.max_protection == region_list_tail->max_protection | |
325 | && region_list_tail->address + region_list_tail->size == address) | |
326 | { | |
327 | region_list_tail->size += size; | |
328 | } | |
329 | else | |
330 | { | |
331 | r = (struct region_t *) malloc (sizeof (struct region_t)); | |
332 | ||
333 | if (!r) | |
334 | unexec_error ("cannot allocate region structure"); | |
335 | ||
336 | r->address = address; | |
337 | r->size = size; | |
338 | r->protection = info.protection; | |
339 | r->max_protection = info.max_protection; | |
340 | ||
341 | r->next = 0; | |
342 | if (region_list_head == 0) | |
343 | { | |
344 | region_list_head = r; | |
345 | region_list_tail = r; | |
346 | } | |
347 | else | |
348 | { | |
349 | region_list_tail->next = r; | |
350 | region_list_tail = r; | |
351 | } | |
352 | ||
353 | /* Deallocate (unused) object name returned by | |
354 | vm_region. */ | |
355 | if (object_name != MACH_PORT_NULL) | |
356 | mach_port_deallocate (target_task, object_name); | |
357 | } | |
358 | ||
359 | address += size; | |
360 | } | |
361 | ||
362 | printf ("--- List of Regions to be Dumped ---\n"); | |
363 | print_region_list (); | |
364 | } | |
365 | ||
366 | ||
367 | #define MAX_UNEXEC_REGIONS 30 | |
368 | ||
369 | int num_unexec_regions; | |
370 | vm_range_t unexec_regions[MAX_UNEXEC_REGIONS]; | |
371 | ||
372 | static void | |
373 | unexec_regions_recorder (task_t task, void *rr, unsigned type, | |
374 | vm_range_t *ranges, unsigned num) | |
375 | { | |
376 | while (num && num_unexec_regions < MAX_UNEXEC_REGIONS) | |
377 | { | |
378 | unexec_regions[num_unexec_regions++] = *ranges; | |
379 | printf ("%#8x (sz: %#8x)\n", ranges->address, ranges->size); | |
380 | ranges++; num--; | |
381 | } | |
382 | if (num_unexec_regions == MAX_UNEXEC_REGIONS) | |
383 | fprintf (stderr, "malloc_freezedry_recorder: too many regions\n"); | |
384 | } | |
385 | ||
386 | static kern_return_t | |
387 | unexec_reader (task_t task, vm_address_t address, vm_size_t size, void **ptr) | |
388 | { | |
389 | *ptr = (void *) address; | |
390 | return KERN_SUCCESS; | |
391 | } | |
392 | ||
393 | void | |
394 | find_emacs_zone_regions () | |
395 | { | |
396 | num_unexec_regions = 0; | |
397 | ||
398 | emacs_zone->introspect->enumerator (mach_task_self(), 0, | |
399 | MALLOC_PTR_REGION_RANGE_TYPE | |
400 | | MALLOC_ADMIN_REGION_RANGE_TYPE, | |
401 | (vm_address_t) emacs_zone, | |
402 | unexec_reader, | |
403 | unexec_regions_recorder); | |
404 | } | |
405 | ||
406 | ||
407 | /* More informational messages routines. */ | |
408 | ||
409 | static void | |
410 | print_load_command_name (int lc) | |
411 | { | |
412 | switch (lc) | |
413 | { | |
414 | case LC_SEGMENT: | |
415 | printf ("LC_SEGMENT "); | |
416 | break; | |
417 | case LC_LOAD_DYLINKER: | |
418 | printf ("LC_LOAD_DYLINKER "); | |
419 | break; | |
420 | case LC_LOAD_DYLIB: | |
421 | printf ("LC_LOAD_DYLIB "); | |
422 | break; | |
423 | case LC_SYMTAB: | |
424 | printf ("LC_SYMTAB "); | |
425 | break; | |
426 | case LC_DYSYMTAB: | |
427 | printf ("LC_DYSYMTAB "); | |
428 | break; | |
429 | case LC_UNIXTHREAD: | |
430 | printf ("LC_UNIXTHREAD "); | |
431 | break; | |
432 | case LC_PREBOUND_DYLIB: | |
433 | printf ("LC_PREBOUND_DYLIB"); | |
434 | break; | |
435 | case LC_TWOLEVEL_HINTS: | |
436 | printf ("LC_TWOLEVEL_HINTS"); | |
437 | break; | |
438 | default: | |
439 | printf ("unknown "); | |
440 | } | |
441 | } | |
442 | ||
443 | static void | |
444 | print_load_command (struct load_command *lc) | |
445 | { | |
446 | print_load_command_name (lc->cmd); | |
447 | printf ("%8d", lc->cmdsize); | |
448 | ||
449 | if (lc->cmd == LC_SEGMENT) | |
450 | { | |
451 | struct segment_command *scp; | |
452 | struct section *sectp; | |
453 | int j; | |
454 | ||
455 | scp = (struct segment_command *) lc; | |
456 | printf (" %-16.16s %#10x %#8x\n", | |
457 | scp->segname, scp->vmaddr, scp->vmsize); | |
458 | ||
459 | sectp = (struct section *) (scp + 1); | |
460 | for (j = 0; j < scp->nsects; j++) | |
461 | { | |
462 | printf (" %-16.16s %#10x %#8x\n", | |
463 | sectp->sectname, sectp->addr, sectp->size); | |
464 | sectp++; | |
465 | } | |
466 | } | |
467 | else | |
468 | printf ("\n"); | |
469 | } | |
470 | ||
471 | /* Read header and load commands from input file. Store the latter in | |
472 | the global array lca. Store the total number of load commands in | |
473 | global variable nlc. */ | |
474 | static void | |
475 | read_load_commands () | |
476 | { | |
477 | int n, i, j; | |
478 | ||
479 | if (!unexec_read (&mh, sizeof (struct mach_header))) | |
480 | unexec_error ("cannot read mach-o header"); | |
481 | ||
482 | if (mh.magic != MH_MAGIC) | |
483 | unexec_error ("input file not in Mach-O format"); | |
484 | ||
485 | if (mh.filetype != MH_EXECUTE) | |
486 | unexec_error ("input Mach-O file is not an executable object file"); | |
487 | ||
488 | #if VERBOSE | |
489 | printf ("--- Header Information ---\n"); | |
490 | printf ("Magic = 0x%08x\n", mh.magic); | |
491 | printf ("CPUType = %d\n", mh.cputype); | |
492 | printf ("CPUSubType = %d\n", mh.cpusubtype); | |
493 | printf ("FileType = 0x%x\n", mh.filetype); | |
494 | printf ("NCmds = %d\n", mh.ncmds); | |
495 | printf ("SizeOfCmds = %d\n", mh.sizeofcmds); | |
496 | printf ("Flags = 0x%08x\n", mh.flags); | |
497 | #endif | |
498 | ||
499 | nlc = mh.ncmds; | |
500 | lca = (struct load_command **) malloc (nlc * sizeof (struct load_command *)); | |
501 | ||
502 | for (i = 0; i < nlc; i++) | |
503 | { | |
504 | struct load_command lc; | |
505 | /* Load commands are variable-size: so read the command type and | |
506 | size first and then read the rest. */ | |
507 | if (!unexec_read (&lc, sizeof (struct load_command))) | |
508 | unexec_error ("cannot read load command"); | |
509 | lca[i] = (struct load_command *) malloc (lc.cmdsize); | |
510 | memcpy (lca[i], &lc, sizeof (struct load_command)); | |
511 | if (!unexec_read (lca[i] + 1, lc.cmdsize - sizeof (struct load_command))) | |
512 | unexec_error ("cannot read content of load command"); | |
513 | if (lc.cmd == LC_SEGMENT) | |
514 | { | |
515 | struct segment_command *scp = (struct segment_command *) lca[i]; | |
516 | ||
517 | if (scp->vmaddr + scp->vmsize > infile_lc_highest_addr) | |
518 | infile_lc_highest_addr = scp->vmaddr + scp->vmsize; | |
519 | ||
520 | if (strncmp (scp->segname, SEG_TEXT, 16) == 0) | |
521 | { | |
522 | struct section *sectp = (struct section *) (scp + 1); | |
523 | int j; | |
524 | ||
525 | for (j = 0; j < scp->nsects; j++) | |
526 | if (sectp->offset < text_seg_lowest_offset) | |
527 | text_seg_lowest_offset = sectp->offset; | |
528 | } | |
529 | } | |
530 | } | |
531 | ||
532 | printf ("Highest address of load commands in input file: %#8x\n", | |
533 | infile_lc_highest_addr); | |
534 | ||
535 | printf ("Lowest offset of all sections in __TEXT segment: %#8x\n", | |
536 | text_seg_lowest_offset); | |
537 | ||
538 | printf ("--- List of Load Commands in Input File ---\n"); | |
539 | printf ("# cmd cmdsize name address size\n"); | |
540 | ||
541 | for (i = 0; i < nlc; i++) | |
542 | { | |
543 | printf ("%1d ", i); | |
544 | print_load_command (lca[i]); | |
545 | } | |
546 | } | |
547 | ||
548 | /* Copy a LC_SEGMENT load command other than the __DATA segment from | |
549 | the input file to the output file, adjusting the file offset of the | |
550 | segment and the file offsets of sections contained in it. */ | |
551 | static void | |
552 | copy_segment (struct load_command *lc) | |
553 | { | |
554 | struct segment_command *scp = (struct segment_command *) lc; | |
555 | unsigned long old_fileoff = scp->fileoff; | |
556 | struct section *sectp; | |
557 | int j; | |
558 | ||
559 | scp->fileoff += delta; | |
560 | ||
561 | sectp = (struct section *) (scp + 1); | |
562 | for (j = 0; j < scp->nsects; j++) | |
563 | { | |
564 | sectp->offset += delta; | |
565 | sectp++; | |
566 | } | |
567 | ||
568 | printf ("Writing segment %-16.16s at %#8x - %#8x (sz: %#8x)\n", | |
569 | scp->segname, scp->fileoff, scp->fileoff + scp->filesize, | |
570 | scp->filesize); | |
571 | ||
572 | if (!unexec_copy (scp->fileoff, old_fileoff, scp->filesize)) | |
573 | unexec_error ("cannot copy segment from input to output file"); | |
574 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
575 | unexec_error ("cannot write load command to header"); | |
576 | ||
577 | curr_header_offset += lc->cmdsize; | |
578 | } | |
579 | ||
580 | /* Copy a LC_SEGMENT load command for the __DATA segment in the input | |
581 | file to the output file. We assume that only one such segment load | |
582 | command exists in the input file and it contains the sections | |
583 | __data, __bss, __common, __la_symbol_ptr, __nl_symbol_ptr, and | |
584 | __dyld. The first three of these should be dumped from memory and | |
585 | the rest should be copied from the input file. Note that the | |
586 | sections __bss and __common contain no data in the input file | |
587 | because their flag fields have the value S_ZEROFILL. Dumping these | |
588 | from memory makes it necessary to adjust file offset fields in | |
589 | subsequently dumped load commands. Then, create new __DATA segment | |
590 | load commands for regions on the region list other than the one | |
591 | corresponding to the __DATA segment in the input file. */ | |
592 | static void | |
593 | copy_data_segment (struct load_command *lc) | |
594 | { | |
595 | struct segment_command *scp = (struct segment_command *) lc; | |
596 | struct section *sectp; | |
597 | int j; | |
598 | unsigned long header_offset, file_offset, old_file_offset; | |
599 | struct region_t *r; | |
600 | ||
601 | printf ("Writing segment %-16.16s at %#8x - %#8x (sz: %#8x)\n", | |
602 | scp->segname, scp->fileoff, scp->fileoff + scp->filesize, | |
603 | scp->filesize); | |
604 | ||
605 | if (delta != 0) | |
606 | unexec_error ("cannot handle multiple DATA segments in input file"); | |
607 | ||
608 | /* Offsets in the output file for writing the next section structure | |
609 | and segment data block, respectively. */ | |
610 | header_offset = curr_header_offset + sizeof (struct segment_command); | |
611 | ||
612 | sectp = (struct section *) (scp + 1); | |
613 | for (j = 0; j < scp->nsects; j++) | |
614 | { | |
615 | old_file_offset = sectp->offset; | |
616 | sectp->offset = sectp->addr - scp->vmaddr + scp->fileoff; | |
617 | /* The __data section is dumped from memory. The __bss and | |
618 | __common sections are also dumped from memory but their flag | |
619 | fields require changing (from S_ZEROFILL to S_REGULAR). The | |
620 | other three kinds of sections are just copied from the input | |
621 | file. */ | |
622 | if (strncmp (sectp->sectname, SECT_DATA, 16) == 0) | |
623 | { | |
624 | if (!unexec_write (sectp->offset, (void *) sectp->addr, sectp->size)) | |
625 | unexec_error ("cannot write section %s", SECT_DATA); | |
626 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) | |
627 | unexec_error ("cannot write section %s's header", SECT_DATA); | |
628 | } | |
629 | else if (strncmp (sectp->sectname, SECT_BSS, 16) == 0 | |
630 | || strncmp (sectp->sectname, SECT_COMMON, 16) == 0) | |
631 | { | |
632 | sectp->flags = S_REGULAR; | |
633 | if (!unexec_write (sectp->offset, (void *) sectp->addr, sectp->size)) | |
634 | unexec_error ("cannot write section %s", SECT_DATA); | |
635 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) | |
636 | unexec_error ("cannot write section %s's header", SECT_DATA); | |
637 | } | |
638 | else if (strncmp (sectp->sectname, "__la_symbol_ptr", 16) == 0 | |
639 | || strncmp (sectp->sectname, "__nl_symbol_ptr", 16) == 0 | |
640 | || strncmp (sectp->sectname, "__dyld", 16) == 0 | |
641 | || strncmp (sectp->sectname, "__const", 16) == 0) | |
642 | { | |
643 | if (!unexec_copy (sectp->offset, old_file_offset, sectp->size)) | |
644 | unexec_error ("cannot copy section %s", sectp->sectname); | |
645 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) | |
646 | unexec_error ("cannot write section %s's header", sectp->sectname); | |
647 | } | |
648 | else | |
649 | unexec_error ("unrecognized section name in __DATA segment"); | |
650 | ||
651 | printf (" section %-16.16s at %#8x - %#8x (sz: %#8x)\n", | |
652 | sectp->sectname, sectp->offset, sectp->offset + sectp->size, | |
653 | sectp->size); | |
654 | ||
655 | header_offset += sizeof (struct section); | |
656 | sectp++; | |
657 | } | |
658 | ||
659 | /* The new filesize of the segment is set to its vmsize because data | |
660 | blocks for segments must start at region boundaries. Note that | |
661 | this may leave unused locations at the end of the segment data | |
662 | block because the total of the sizes of all sections in the | |
663 | segment is generally smaller than vmsize. */ | |
664 | delta = scp->vmsize - scp->filesize; | |
665 | scp->filesize = scp->vmsize; | |
666 | if (!unexec_write (curr_header_offset, scp, sizeof (struct segment_command))) | |
667 | unexec_error ("cannot write header of __DATA segment"); | |
668 | curr_header_offset += lc->cmdsize; | |
669 | ||
670 | /* Create new __DATA segment load commands for regions on the region | |
671 | list that do not corresponding to any segment load commands in | |
672 | the input file. | |
673 | */ | |
674 | file_offset = scp->fileoff + scp->filesize; | |
675 | for (j = 0; j < num_unexec_regions; j++) | |
676 | { | |
677 | struct segment_command sc; | |
678 | ||
679 | sc.cmd = LC_SEGMENT; | |
680 | sc.cmdsize = sizeof (struct segment_command); | |
681 | strncpy (sc.segname, SEG_DATA, 16); | |
682 | sc.vmaddr = unexec_regions[j].address; | |
683 | sc.vmsize = unexec_regions[j].size; | |
684 | sc.fileoff = file_offset; | |
685 | sc.filesize = unexec_regions[j].size; | |
686 | sc.maxprot = VM_PROT_READ | VM_PROT_WRITE; | |
687 | sc.initprot = VM_PROT_READ | VM_PROT_WRITE; | |
688 | sc.nsects = 0; | |
689 | sc.flags = 0; | |
690 | ||
691 | printf ("Writing segment %-16.16s at %#8x - %#8x (sz: %#8x)\n", | |
692 | sc.segname, sc.fileoff, sc.fileoff + sc.filesize, | |
693 | sc.filesize); | |
694 | ||
695 | if (!unexec_write (sc.fileoff, (void *) sc.vmaddr, sc.vmsize)) | |
696 | unexec_error ("cannot write new __DATA segment"); | |
697 | delta += sc.filesize; | |
698 | file_offset += sc.filesize; | |
699 | ||
700 | if (!unexec_write (curr_header_offset, &sc, sc.cmdsize)) | |
701 | unexec_error ("cannot write new __DATA segment's header"); | |
702 | curr_header_offset += sc.cmdsize; | |
703 | mh.ncmds++; | |
704 | } | |
705 | } | |
706 | ||
707 | /* Copy a LC_SYMTAB load command from the input file to the output | |
708 | file, adjusting the file offset fields. */ | |
709 | static void | |
710 | copy_symtab (struct load_command *lc) | |
711 | { | |
712 | struct symtab_command *stp = (struct symtab_command *) lc; | |
713 | ||
714 | stp->symoff += delta; | |
715 | stp->stroff += delta; | |
716 | ||
717 | printf ("Writing LC_SYMTAB command\n"); | |
718 | ||
719 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
720 | unexec_error ("cannot write symtab command to header"); | |
721 | ||
722 | curr_header_offset += lc->cmdsize; | |
723 | } | |
724 | ||
725 | /* Copy a LC_DYSYMTAB load command from the input file to the output | |
726 | file, adjusting the file offset fields. */ | |
727 | static void | |
728 | copy_dysymtab (struct load_command *lc) | |
729 | { | |
730 | struct dysymtab_command *dstp = (struct dysymtab_command *) lc; | |
731 | ||
732 | /* If Mach-O executable is not prebound, relocation entries need | |
733 | fixing up. This is not supported currently. */ | |
734 | if (!(mh.flags & MH_PREBOUND) && (dstp->nextrel != 0 || dstp->nlocrel != 0)) | |
735 | unexec_error ("cannot handle LC_DYSYMTAB with relocation entries"); | |
736 | ||
737 | if (dstp->nextrel > 0) { | |
738 | dstp->extreloff += delta; | |
739 | } | |
740 | ||
741 | if (dstp->nlocrel > 0) { | |
742 | dstp->locreloff += delta; | |
743 | } | |
744 | ||
745 | if (dstp->nindirectsyms > 0) | |
746 | dstp->indirectsymoff += delta; | |
747 | ||
748 | printf ("Writing LC_DYSYMTAB command\n"); | |
749 | ||
750 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
751 | unexec_error ("cannot write symtab command to header"); | |
752 | ||
753 | curr_header_offset += lc->cmdsize; | |
754 | } | |
755 | ||
40e6ff95 ST |
756 | /* Copy a LC_TWOLEVEL_HINTS load command from the input file to the output |
757 | file, adjusting the file offset fields. */ | |
758 | static void | |
759 | copy_twolevelhints (struct load_command *lc) | |
760 | { | |
761 | struct twolevel_hints_command *tlhp = (struct twolevel_hints_command *) lc; | |
762 | ||
763 | if (tlhp->nhints > 0) { | |
764 | tlhp->offset += delta; | |
765 | } | |
766 | ||
767 | printf ("Writing LC_TWOLEVEL_HINTS command\n"); | |
768 | ||
769 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
770 | unexec_error ("cannot write two level hint command to header"); | |
771 | ||
772 | curr_header_offset += lc->cmdsize; | |
773 | } | |
774 | ||
e0f712ba AC |
775 | /* Copy other kinds of load commands from the input file to the output |
776 | file, ones that do not require adjustments of file offsets. */ | |
777 | static void | |
778 | copy_other (struct load_command *lc) | |
779 | { | |
780 | printf ("Writing "); | |
781 | print_load_command_name (lc->cmd); | |
782 | printf (" command\n"); | |
783 | ||
784 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) | |
785 | unexec_error ("cannot write symtab command to header"); | |
786 | ||
787 | curr_header_offset += lc->cmdsize; | |
788 | } | |
789 | ||
790 | /* Loop through all load commands and dump them. Then write the Mach | |
791 | header. */ | |
792 | static void | |
793 | dump_it () | |
794 | { | |
795 | int i; | |
796 | ||
797 | printf ("--- Load Commands written to Output File ---\n"); | |
798 | ||
799 | for (i = 0; i < nlc; i++) | |
800 | switch (lca[i]->cmd) | |
801 | { | |
802 | case LC_SEGMENT: | |
803 | { | |
804 | struct segment_command *scp = (struct segment_command *) lca[i]; | |
805 | if (strncmp (scp->segname, SEG_DATA, 16) == 0) | |
806 | { | |
807 | copy_data_segment (lca[i]); | |
808 | } | |
809 | else | |
810 | { | |
811 | copy_segment (lca[i]); | |
812 | } | |
813 | } | |
814 | break; | |
815 | case LC_SYMTAB: | |
816 | copy_symtab (lca[i]); | |
817 | break; | |
818 | case LC_DYSYMTAB: | |
819 | copy_dysymtab (lca[i]); | |
820 | break; | |
40e6ff95 ST |
821 | case LC_TWOLEVEL_HINTS: |
822 | copy_twolevelhints (lca[i]); | |
823 | break; | |
e0f712ba AC |
824 | default: |
825 | copy_other (lca[i]); | |
826 | break; | |
827 | } | |
828 | ||
829 | if (curr_header_offset > text_seg_lowest_offset) | |
830 | unexec_error ("not enough room for load commands for new __DATA segments"); | |
831 | ||
832 | printf ("%d unused bytes follow Mach-O header\n", | |
833 | text_seg_lowest_offset - curr_header_offset); | |
834 | ||
835 | mh.sizeofcmds = curr_header_offset - sizeof (struct mach_header); | |
836 | if (!unexec_write (0, &mh, sizeof (struct mach_header))) | |
837 | unexec_error ("cannot write final header contents"); | |
838 | } | |
839 | ||
840 | /* Take a snapshot of Emacs and make a Mach-O format executable file | |
841 | from it. The file names of the output and input files are outfile | |
842 | and infile, respectively. The three other parameters are | |
843 | ignored. */ | |
844 | void | |
845 | unexec (char *outfile, char *infile, void *start_data, void *start_bss, | |
846 | void *entry_address) | |
847 | { | |
848 | infd = open (infile, O_RDONLY, 0); | |
849 | if (infd < 0) | |
850 | { | |
851 | unexec_error ("cannot open input file `%s'", infile); | |
852 | } | |
853 | ||
854 | outfd = open (outfile, O_WRONLY | O_TRUNC | O_CREAT, 0755); | |
855 | if (outfd < 0) | |
856 | { | |
857 | close (infd); | |
858 | unexec_error ("cannot open output file `%s'", outfile); | |
859 | } | |
860 | ||
861 | build_region_list (); | |
862 | read_load_commands (); | |
863 | ||
864 | find_emacs_zone_regions (); | |
865 | ||
866 | in_dumped_exec = 1; | |
867 | ||
868 | dump_it (); | |
869 | ||
870 | close (outfd); | |
871 | } | |
872 | ||
873 | ||
874 | void | |
875 | unexec_init_emacs_zone () | |
876 | { | |
877 | emacs_zone = malloc_create_zone (0, 0); | |
878 | malloc_set_zone_name (emacs_zone, "EmacsZone"); | |
879 | } | |
880 | ||
881 | int | |
882 | ptr_in_unexec_regions (void *ptr) | |
883 | { | |
884 | int i; | |
885 | ||
886 | for (i = 0; i < num_unexec_regions; i++) | |
887 | if ((vm_address_t) ptr - unexec_regions[i].address | |
888 | < unexec_regions[i].size) | |
889 | return 1; | |
890 | ||
891 | return 0; | |
892 | } | |
893 | ||
894 | void * | |
895 | unexec_malloc (size_t size) | |
896 | { | |
897 | if (in_dumped_exec) | |
898 | return malloc (size); | |
899 | else | |
900 | return malloc_zone_malloc (emacs_zone, size); | |
901 | } | |
902 | ||
903 | void * | |
904 | unexec_realloc (void *old_ptr, size_t new_size) | |
905 | { | |
906 | if (in_dumped_exec) | |
907 | if (ptr_in_unexec_regions (old_ptr)) | |
908 | { | |
909 | char *p = malloc (new_size); | |
910 | /* 2002-04-15 T. Ikegami <ikegami@adam.uprr.pr>. The original | |
911 | code to get size failed to reallocate read_buffer | |
912 | (lread.c). */ | |
82122254 | 913 | int old_size = malloc_default_zone()->size (emacs_zone, old_ptr); |
e0f712ba AC |
914 | int size = new_size > old_size ? old_size : new_size; |
915 | ||
916 | if (size) | |
917 | memcpy (p, old_ptr, size); | |
918 | return p; | |
919 | } | |
920 | else | |
921 | return realloc (old_ptr, new_size); | |
922 | else | |
923 | return malloc_zone_realloc (emacs_zone, old_ptr, new_size); | |
924 | } | |
925 | ||
926 | void | |
927 | unexec_free (void *ptr) | |
928 | { | |
929 | if (in_dumped_exec) | |
930 | { | |
931 | if (!ptr_in_unexec_regions (ptr)) | |
932 | free (ptr); | |
933 | } | |
934 | else | |
935 | malloc_zone_free (emacs_zone, ptr); | |
936 | } |