| 1 | /* Dump Emacs in Mach-O format for use on Mac OS X. |
| 2 | Copyright (C) 2001, 2002, 2003, 2004, 2005, |
| 3 | 2006, 2007, 2008 Free Software Foundation, Inc. |
| 4 | |
| 5 | This file is part of GNU Emacs. |
| 6 | |
| 7 | GNU Emacs is free software: you can redistribute it and/or modify |
| 8 | it under the terms of the GNU General Public License as published by |
| 9 | the Free Software Foundation, either version 3 of the License, or |
| 10 | (at your option) any later version. |
| 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 |
| 18 | along with GNU Emacs. If not, see <http://www.gnu.org/licenses/>. */ |
| 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 |
| 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. |
| 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> |
| 97 | #include <mach-o/reloc.h> |
| 98 | #if defined (__ppc__) |
| 99 | #include <mach-o/ppc/reloc.h> |
| 100 | #endif |
| 101 | #include <config.h> |
| 102 | #undef malloc |
| 103 | #undef realloc |
| 104 | #undef free |
| 105 | #ifdef HAVE_MALLOC_MALLOC_H |
| 106 | #include <malloc/malloc.h> |
| 107 | #else |
| 108 | #include <objc/malloc.h> |
| 109 | #endif |
| 110 | |
| 111 | #include <assert.h> |
| 112 | |
| 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 |
| 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 | |
| 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. */ |
| 149 | static struct region_t *region_list_head = 0; |
| 150 | static struct region_t *region_list_tail = 0; |
| 151 | |
| 152 | /* Pointer to array of load commands. */ |
| 153 | static struct load_command **lca; |
| 154 | |
| 155 | /* Number of load commands. */ |
| 156 | static int nlc; |
| 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. */ |
| 161 | static vm_address_t infile_lc_highest_addr = 0; |
| 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. */ |
| 168 | static unsigned long text_seg_lowest_offset = 0x10000000; |
| 169 | |
| 170 | /* Mach header. */ |
| 171 | static struct mach_header mh; |
| 172 | |
| 173 | /* Offset at which the next load command should be written. */ |
| 174 | static unsigned long curr_header_offset = sizeof (struct mach_header); |
| 175 | |
| 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)) |
| 181 | |
| 182 | static int infd, outfd; |
| 183 | |
| 184 | static int in_dumped_exec = 0; |
| 185 | |
| 186 | static malloc_zone_t *emacs_zone; |
| 187 | |
| 188 | /* file offset of input file's data segment */ |
| 189 | static off_t data_segment_old_fileoff = 0; |
| 190 | |
| 191 | static struct segment_command *data_segment_scp; |
| 192 | |
| 193 | /* Read N bytes from infd into memory starting at address DEST. |
| 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 | |
| 201 | /* Write COUNT bytes from memory starting at address SRC to outfd |
| 202 | starting at offset DEST. Return true if successful, false |
| 203 | otherwise. */ |
| 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 | |
| 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 | |
| 221 | bzero (buf, UNEXEC_COPY_BUFSZ); |
| 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. */ |
| 239 | static int |
| 240 | unexec_copy (off_t dest, off_t src, ssize_t count) |
| 241 | { |
| 242 | ssize_t bytes_read; |
| 243 | ssize_t bytes_to_read; |
| 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 | { |
| 255 | bytes_to_read = count > UNEXEC_COPY_BUFSZ ? UNEXEC_COPY_BUFSZ : count; |
| 256 | bytes_read = read (infd, buf, bytes_to_read); |
| 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 |
| 270 | unexec_error (char *format, ...) |
| 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 | { |
| 300 | printf ("%#10lx %#8lx ", (long) address, (long) size); |
| 301 | print_prot (prot); |
| 302 | putchar (' '); |
| 303 | print_prot (max_prot); |
| 304 | putchar ('\n'); |
| 305 | } |
| 306 | |
| 307 | static void |
| 308 | print_region_list () |
| 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 | |
| 318 | static void |
| 319 | print_regions () |
| 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); |
| 338 | |
| 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 |
| 349 | build_region_list () |
| 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)); |
| 391 | |
| 392 | if (!r) |
| 393 | unexec_error ("cannot allocate region structure"); |
| 394 | |
| 395 | r->address = address; |
| 396 | r->size = size; |
| 397 | r->protection = info.protection; |
| 398 | r->max_protection = info.max_protection; |
| 399 | |
| 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 | } |
| 411 | |
| 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 | } |
| 417 | |
| 418 | address += size; |
| 419 | } |
| 420 | |
| 421 | printf ("--- List of Regions to be Dumped ---\n"); |
| 422 | print_region_list (); |
| 423 | } |
| 424 | |
| 425 | |
| 426 | #define MAX_UNEXEC_REGIONS 400 |
| 427 | |
| 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]; |
| 434 | |
| 435 | static void |
| 436 | unexec_regions_recorder (task_t task, void *rr, unsigned type, |
| 437 | vm_range_t *ranges, unsigned num) |
| 438 | { |
| 439 | vm_address_t p; |
| 440 | vm_size_t filesize; |
| 441 | |
| 442 | while (num && num_unexec_regions < MAX_UNEXEC_REGIONS) |
| 443 | { |
| 444 | /* Subtract the size of trailing null bytes from filesize. It |
| 445 | can be smaller than vmsize in segment commands. In such a |
| 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; |
| 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)); |
| 456 | ranges++; num--; |
| 457 | } |
| 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 | |
| 467 | static void |
| 468 | find_emacs_zone_regions () |
| 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); |
| 478 | |
| 479 | if (num_unexec_regions == MAX_UNEXEC_REGIONS) |
| 480 | unexec_error ("find_emacs_zone_regions: too many regions"); |
| 481 | } |
| 482 | |
| 483 | static int |
| 484 | unexec_regions_sort_compare (const void *a, const void *b) |
| 485 | { |
| 486 | vm_address_t aa = ((unexec_region_info *) a)->range.address; |
| 487 | vm_address_t bb = ((unexec_region_info *) b)->range.address; |
| 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 |
| 498 | unexec_regions_merge () |
| 499 | { |
| 500 | int i, n; |
| 501 | unexec_region_info r; |
| 502 | vm_size_t padsize; |
| 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]; |
| 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 | } |
| 515 | for (i = 1; i < num_unexec_regions; i++) |
| 516 | { |
| 517 | if (r.range.address + r.range.size == unexec_regions[i].range.address |
| 518 | && r.range.size - r.filesize < 2 * pagesize) |
| 519 | { |
| 520 | r.filesize = r.range.size + unexec_regions[i].filesize; |
| 521 | r.range.size += unexec_regions[i].range.size; |
| 522 | } |
| 523 | else |
| 524 | { |
| 525 | unexec_regions[n++] = r; |
| 526 | r = unexec_regions[i]; |
| 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 | } |
| 538 | } |
| 539 | } |
| 540 | unexec_regions[n++] = r; |
| 541 | num_unexec_regions = n; |
| 542 | } |
| 543 | |
| 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: |
| 553 | #ifndef _LP64 |
| 554 | printf ("LC_SEGMENT "); |
| 555 | #else |
| 556 | printf ("LC_SEGMENT_64 "); |
| 557 | #endif |
| 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; |
| 580 | #ifdef LC_UUID |
| 581 | case LC_UUID: |
| 582 | printf ("LC_UUID "); |
| 583 | break; |
| 584 | #endif |
| 585 | default: |
| 586 | printf ("unknown "); |
| 587 | } |
| 588 | } |
| 589 | |
| 590 | static void |
| 591 | print_load_command (struct load_command *lc) |
| 592 | { |
| 593 | print_load_command_name (lc->cmd); |
| 594 | printf ("%8d", lc->cmdsize); |
| 595 | |
| 596 | if (lc->cmd == LC_SEGMENT) |
| 597 | { |
| 598 | struct segment_command *scp; |
| 599 | struct section *sectp; |
| 600 | int j; |
| 601 | |
| 602 | scp = (struct segment_command *) lc; |
| 603 | printf (" %-16.16s %#10lx %#8lx\n", |
| 604 | scp->segname, (long) (scp->vmaddr), (long) (scp->vmsize)); |
| 605 | |
| 606 | sectp = (struct section *) (scp + 1); |
| 607 | for (j = 0; j < scp->nsects; j++) |
| 608 | { |
| 609 | printf (" %-16.16s %#10lx %#8lx\n", |
| 610 | sectp->sectname, (long) (sectp->addr), (long) (sectp->size)); |
| 611 | sectp++; |
| 612 | } |
| 613 | } |
| 614 | else |
| 615 | printf ("\n"); |
| 616 | } |
| 617 | |
| 618 | /* Read header and load commands from input file. Store the latter in |
| 619 | the global array lca. Store the total number of load commands in |
| 620 | global variable nlc. */ |
| 621 | static void |
| 622 | read_load_commands () |
| 623 | { |
| 624 | int i; |
| 625 | |
| 626 | if (!unexec_read (&mh, sizeof (struct mach_header))) |
| 627 | unexec_error ("cannot read mach-o header"); |
| 628 | |
| 629 | if (mh.magic != MH_MAGIC) |
| 630 | unexec_error ("input file not in Mach-O format"); |
| 631 | |
| 632 | if (mh.filetype != MH_EXECUTE) |
| 633 | unexec_error ("input Mach-O file is not an executable object file"); |
| 634 | |
| 635 | #if VERBOSE |
| 636 | printf ("--- Header Information ---\n"); |
| 637 | printf ("Magic = 0x%08x\n", mh.magic); |
| 638 | printf ("CPUType = %d\n", mh.cputype); |
| 639 | printf ("CPUSubType = %d\n", mh.cpusubtype); |
| 640 | printf ("FileType = 0x%x\n", mh.filetype); |
| 641 | printf ("NCmds = %d\n", mh.ncmds); |
| 642 | printf ("SizeOfCmds = %d\n", mh.sizeofcmds); |
| 643 | printf ("Flags = 0x%08x\n", mh.flags); |
| 644 | #endif |
| 645 | |
| 646 | nlc = mh.ncmds; |
| 647 | lca = (struct load_command **) malloc (nlc * sizeof (struct load_command *)); |
| 648 | |
| 649 | for (i = 0; i < nlc; i++) |
| 650 | { |
| 651 | struct load_command lc; |
| 652 | /* Load commands are variable-size: so read the command type and |
| 653 | size first and then read the rest. */ |
| 654 | if (!unexec_read (&lc, sizeof (struct load_command))) |
| 655 | unexec_error ("cannot read load command"); |
| 656 | lca[i] = (struct load_command *) malloc (lc.cmdsize); |
| 657 | memcpy (lca[i], &lc, sizeof (struct load_command)); |
| 658 | if (!unexec_read (lca[i] + 1, lc.cmdsize - sizeof (struct load_command))) |
| 659 | unexec_error ("cannot read content of load command"); |
| 660 | if (lc.cmd == LC_SEGMENT) |
| 661 | { |
| 662 | struct segment_command *scp = (struct segment_command *) lca[i]; |
| 663 | |
| 664 | if (scp->vmaddr + scp->vmsize > infile_lc_highest_addr) |
| 665 | infile_lc_highest_addr = scp->vmaddr + scp->vmsize; |
| 666 | |
| 667 | if (strncmp (scp->segname, SEG_TEXT, 16) == 0) |
| 668 | { |
| 669 | struct section *sectp = (struct section *) (scp + 1); |
| 670 | int j; |
| 671 | |
| 672 | for (j = 0; j < scp->nsects; j++) |
| 673 | if (sectp->offset < text_seg_lowest_offset) |
| 674 | text_seg_lowest_offset = sectp->offset; |
| 675 | } |
| 676 | } |
| 677 | } |
| 678 | |
| 679 | printf ("Highest address of load commands in input file: %#8x\n", |
| 680 | infile_lc_highest_addr); |
| 681 | |
| 682 | printf ("Lowest offset of all sections in __TEXT segment: %#8lx\n", |
| 683 | text_seg_lowest_offset); |
| 684 | |
| 685 | printf ("--- List of Load Commands in Input File ---\n"); |
| 686 | printf ("# cmd cmdsize name address size\n"); |
| 687 | |
| 688 | for (i = 0; i < nlc; i++) |
| 689 | { |
| 690 | printf ("%1d ", i); |
| 691 | print_load_command (lca[i]); |
| 692 | } |
| 693 | } |
| 694 | |
| 695 | /* Copy a LC_SEGMENT load command other than the __DATA segment from |
| 696 | the input file to the output file, adjusting the file offset of the |
| 697 | segment and the file offsets of sections contained in it. */ |
| 698 | static void |
| 699 | copy_segment (struct load_command *lc) |
| 700 | { |
| 701 | struct segment_command *scp = (struct segment_command *) lc; |
| 702 | unsigned long old_fileoff = scp->fileoff; |
| 703 | struct section *sectp; |
| 704 | int j; |
| 705 | |
| 706 | scp->fileoff = curr_file_offset; |
| 707 | |
| 708 | sectp = (struct section *) (scp + 1); |
| 709 | for (j = 0; j < scp->nsects; j++) |
| 710 | { |
| 711 | sectp->offset += curr_file_offset - old_fileoff; |
| 712 | sectp++; |
| 713 | } |
| 714 | |
| 715 | printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n", |
| 716 | scp->segname, (long) (scp->fileoff), (long) (scp->filesize), |
| 717 | (long) (scp->vmsize), (long) (scp->vmaddr)); |
| 718 | |
| 719 | if (!unexec_copy (scp->fileoff, old_fileoff, scp->filesize)) |
| 720 | unexec_error ("cannot copy segment from input to output file"); |
| 721 | curr_file_offset += ROUNDUP_TO_PAGE_BOUNDARY (scp->filesize); |
| 722 | |
| 723 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) |
| 724 | unexec_error ("cannot write load command to header"); |
| 725 | |
| 726 | curr_header_offset += lc->cmdsize; |
| 727 | } |
| 728 | |
| 729 | /* Copy a LC_SEGMENT load command for the __DATA segment in the input |
| 730 | file to the output file. We assume that only one such segment load |
| 731 | command exists in the input file and it contains the sections |
| 732 | __data, __bss, __common, __la_symbol_ptr, __nl_symbol_ptr, and |
| 733 | __dyld. The first three of these should be dumped from memory and |
| 734 | the rest should be copied from the input file. Note that the |
| 735 | sections __bss and __common contain no data in the input file |
| 736 | because their flag fields have the value S_ZEROFILL. Dumping these |
| 737 | from memory makes it necessary to adjust file offset fields in |
| 738 | subsequently dumped load commands. Then, create new __DATA segment |
| 739 | load commands for regions on the region list other than the one |
| 740 | corresponding to the __DATA segment in the input file. */ |
| 741 | static void |
| 742 | copy_data_segment (struct load_command *lc) |
| 743 | { |
| 744 | struct segment_command *scp = (struct segment_command *) lc; |
| 745 | struct section *sectp; |
| 746 | int j; |
| 747 | unsigned long header_offset, old_file_offset; |
| 748 | |
| 749 | /* The new filesize of the segment is set to its vmsize because data |
| 750 | blocks for segments must start at region boundaries. Note that |
| 751 | this may leave unused locations at the end of the segment data |
| 752 | block because the total of the sizes of all sections in the |
| 753 | segment is generally smaller than vmsize. */ |
| 754 | scp->filesize = scp->vmsize; |
| 755 | |
| 756 | printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n", |
| 757 | scp->segname, curr_file_offset, (long)(scp->filesize), |
| 758 | (long)(scp->vmsize), (long) (scp->vmaddr)); |
| 759 | |
| 760 | /* Offsets in the output file for writing the next section structure |
| 761 | and segment data block, respectively. */ |
| 762 | header_offset = curr_header_offset + sizeof (struct segment_command); |
| 763 | |
| 764 | sectp = (struct section *) (scp + 1); |
| 765 | for (j = 0; j < scp->nsects; j++) |
| 766 | { |
| 767 | old_file_offset = sectp->offset; |
| 768 | sectp->offset = sectp->addr - scp->vmaddr + curr_file_offset; |
| 769 | /* The __data section is dumped from memory. The __bss and |
| 770 | __common sections are also dumped from memory but their flag |
| 771 | fields require changing (from S_ZEROFILL to S_REGULAR). The |
| 772 | other three kinds of sections are just copied from the input |
| 773 | file. */ |
| 774 | if (strncmp (sectp->sectname, SECT_DATA, 16) == 0) |
| 775 | { |
| 776 | if (!unexec_write (sectp->offset, (void *) sectp->addr, sectp->size)) |
| 777 | unexec_error ("cannot write section %s", SECT_DATA); |
| 778 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) |
| 779 | unexec_error ("cannot write section %s's header", SECT_DATA); |
| 780 | } |
| 781 | else if (strncmp (sectp->sectname, SECT_COMMON, 16) == 0) |
| 782 | { |
| 783 | sectp->flags = S_REGULAR; |
| 784 | if (!unexec_write (sectp->offset, (void *) sectp->addr, sectp->size)) |
| 785 | unexec_error ("cannot write section %s", sectp->sectname); |
| 786 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) |
| 787 | unexec_error ("cannot write section %s's header", sectp->sectname); |
| 788 | } |
| 789 | else if (strncmp (sectp->sectname, SECT_BSS, 16) == 0) |
| 790 | { |
| 791 | extern char *my_endbss_static; |
| 792 | unsigned long my_size; |
| 793 | |
| 794 | sectp->flags = S_REGULAR; |
| 795 | |
| 796 | /* Clear uninitialized local variables in statically linked |
| 797 | libraries. In particular, function pointers stored by |
| 798 | libSystemStub.a, which is introduced in Mac OS X 10.4 for |
| 799 | binary compatibility with respect to long double, are |
| 800 | cleared so that they will be reinitialized when the |
| 801 | dumped binary is executed on other versions of OS. */ |
| 802 | my_size = (unsigned long)my_endbss_static - sectp->addr; |
| 803 | if (!(sectp->addr <= (unsigned long)my_endbss_static |
| 804 | && my_size <= sectp->size)) |
| 805 | unexec_error ("my_endbss_static is not in section %s", |
| 806 | sectp->sectname); |
| 807 | if (!unexec_write (sectp->offset, (void *) sectp->addr, my_size)) |
| 808 | unexec_error ("cannot write section %s", sectp->sectname); |
| 809 | if (!unexec_write_zero (sectp->offset + my_size, |
| 810 | sectp->size - my_size)) |
| 811 | unexec_error ("cannot write section %s", sectp->sectname); |
| 812 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) |
| 813 | unexec_error ("cannot write section %s's header", sectp->sectname); |
| 814 | } |
| 815 | else if (strncmp (sectp->sectname, "__la_symbol_ptr", 16) == 0 |
| 816 | || strncmp (sectp->sectname, "__nl_symbol_ptr", 16) == 0 |
| 817 | || strncmp (sectp->sectname, "__la_sym_ptr2", 16) == 0 |
| 818 | || strncmp (sectp->sectname, "__dyld", 16) == 0 |
| 819 | || strncmp (sectp->sectname, "__const", 16) == 0 |
| 820 | || strncmp (sectp->sectname, "__cfstring", 16) == 0 |
| 821 | || strncmp (sectp->sectname, "__gcc_except_tab", 16) == 0 |
| 822 | || strncmp (sectp->sectname, "__objc_", 7) == 0) |
| 823 | { |
| 824 | if (!unexec_copy (sectp->offset, old_file_offset, sectp->size)) |
| 825 | unexec_error ("cannot copy section %s", sectp->sectname); |
| 826 | if (!unexec_write (header_offset, sectp, sizeof (struct section))) |
| 827 | unexec_error ("cannot write section %s's header", sectp->sectname); |
| 828 | } |
| 829 | else |
| 830 | unexec_error ("unrecognized section name in __DATA segment"); |
| 831 | |
| 832 | printf (" section %-16.16s at %#8lx - %#8lx (sz: %#8lx)\n", |
| 833 | sectp->sectname, (long) (sectp->offset), |
| 834 | (long) (sectp->offset + sectp->size), (long) (sectp->size)); |
| 835 | |
| 836 | header_offset += sizeof (struct section); |
| 837 | sectp++; |
| 838 | } |
| 839 | |
| 840 | curr_file_offset += ROUNDUP_TO_PAGE_BOUNDARY (scp->filesize); |
| 841 | |
| 842 | if (!unexec_write (curr_header_offset, scp, sizeof (struct segment_command))) |
| 843 | unexec_error ("cannot write header of __DATA segment"); |
| 844 | curr_header_offset += lc->cmdsize; |
| 845 | |
| 846 | /* Create new __DATA segment load commands for regions on the region |
| 847 | list that do not corresponding to any segment load commands in |
| 848 | the input file. |
| 849 | */ |
| 850 | for (j = 0; j < num_unexec_regions; j++) |
| 851 | { |
| 852 | struct segment_command sc; |
| 853 | |
| 854 | sc.cmd = LC_SEGMENT; |
| 855 | sc.cmdsize = sizeof (struct segment_command); |
| 856 | strncpy (sc.segname, SEG_DATA, 16); |
| 857 | sc.vmaddr = unexec_regions[j].range.address; |
| 858 | sc.vmsize = unexec_regions[j].range.size; |
| 859 | sc.fileoff = curr_file_offset; |
| 860 | sc.filesize = unexec_regions[j].filesize; |
| 861 | sc.maxprot = VM_PROT_READ | VM_PROT_WRITE; |
| 862 | sc.initprot = VM_PROT_READ | VM_PROT_WRITE; |
| 863 | sc.nsects = 0; |
| 864 | sc.flags = 0; |
| 865 | |
| 866 | printf ("Writing segment %-16.16s @ %#8lx (%#8lx/%#8lx @ %#10lx)\n", |
| 867 | sc.segname, (long) (sc.fileoff), (long) (sc.filesize), |
| 868 | (long) (sc.vmsize), (long) (sc.vmaddr)); |
| 869 | |
| 870 | if (!unexec_write (sc.fileoff, (void *) sc.vmaddr, sc.filesize)) |
| 871 | unexec_error ("cannot write new __DATA segment"); |
| 872 | curr_file_offset += ROUNDUP_TO_PAGE_BOUNDARY (sc.filesize); |
| 873 | |
| 874 | if (!unexec_write (curr_header_offset, &sc, sc.cmdsize)) |
| 875 | unexec_error ("cannot write new __DATA segment's header"); |
| 876 | curr_header_offset += sc.cmdsize; |
| 877 | mh.ncmds++; |
| 878 | } |
| 879 | } |
| 880 | |
| 881 | /* Copy a LC_SYMTAB load command from the input file to the output |
| 882 | file, adjusting the file offset fields. */ |
| 883 | static void |
| 884 | copy_symtab (struct load_command *lc, long delta) |
| 885 | { |
| 886 | struct symtab_command *stp = (struct symtab_command *) lc; |
| 887 | |
| 888 | stp->symoff += delta; |
| 889 | stp->stroff += delta; |
| 890 | |
| 891 | printf ("Writing LC_SYMTAB command\n"); |
| 892 | |
| 893 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) |
| 894 | unexec_error ("cannot write symtab command to header"); |
| 895 | |
| 896 | curr_header_offset += lc->cmdsize; |
| 897 | } |
| 898 | |
| 899 | /* Fix up relocation entries. */ |
| 900 | static void |
| 901 | unrelocate (const char *name, off_t reloff, int nrel, vm_address_t base) |
| 902 | { |
| 903 | int i, unreloc_count; |
| 904 | struct relocation_info reloc_info; |
| 905 | struct scattered_relocation_info *sc_reloc_info |
| 906 | = (struct scattered_relocation_info *) &reloc_info; |
| 907 | vm_address_t location; |
| 908 | |
| 909 | for (unreloc_count = 0, i = 0; i < nrel; i++) |
| 910 | { |
| 911 | if (lseek (infd, reloff, L_SET) != reloff) |
| 912 | unexec_error ("unrelocate: %s:%d cannot seek to reloc_info", name, i); |
| 913 | if (!unexec_read (&reloc_info, sizeof (reloc_info))) |
| 914 | unexec_error ("unrelocate: %s:%d cannot read reloc_info", name, i); |
| 915 | reloff += sizeof (reloc_info); |
| 916 | |
| 917 | if (sc_reloc_info->r_scattered == 0) |
| 918 | switch (reloc_info.r_type) |
| 919 | { |
| 920 | case GENERIC_RELOC_VANILLA: |
| 921 | location = base + reloc_info.r_address; |
| 922 | if (location >= data_segment_scp->vmaddr |
| 923 | && location < (data_segment_scp->vmaddr |
| 924 | + data_segment_scp->vmsize)) |
| 925 | { |
| 926 | off_t src_off = data_segment_old_fileoff |
| 927 | + (location - data_segment_scp->vmaddr); |
| 928 | off_t dst_off = data_segment_scp->fileoff |
| 929 | + (location - data_segment_scp->vmaddr); |
| 930 | |
| 931 | if (!unexec_copy (dst_off, src_off, 1 << reloc_info.r_length)) |
| 932 | unexec_error ("unrelocate: %s:%d cannot copy original value", |
| 933 | name, i); |
| 934 | unreloc_count++; |
| 935 | } |
| 936 | break; |
| 937 | default: |
| 938 | unexec_error ("unrelocate: %s:%d cannot handle type = %d", |
| 939 | name, i, reloc_info.r_type); |
| 940 | } |
| 941 | else |
| 942 | switch (sc_reloc_info->r_type) |
| 943 | { |
| 944 | #if defined (__ppc__) |
| 945 | case PPC_RELOC_PB_LA_PTR: |
| 946 | /* nothing to do for prebound lazy pointer */ |
| 947 | break; |
| 948 | #endif |
| 949 | default: |
| 950 | unexec_error ("unrelocate: %s:%d cannot handle scattered type = %d", |
| 951 | name, i, sc_reloc_info->r_type); |
| 952 | } |
| 953 | } |
| 954 | |
| 955 | if (nrel > 0) |
| 956 | printf ("Fixed up %d/%d %s relocation entries in data segment.\n", |
| 957 | unreloc_count, nrel, name); |
| 958 | } |
| 959 | |
| 960 | #if __ppc64__ |
| 961 | /* Rebase r_address in the relocation table. */ |
| 962 | static void |
| 963 | rebase_reloc_address (off_t reloff, int nrel, long linkedit_delta, long diff) |
| 964 | { |
| 965 | int i; |
| 966 | struct relocation_info reloc_info; |
| 967 | struct scattered_relocation_info *sc_reloc_info |
| 968 | = (struct scattered_relocation_info *) &reloc_info; |
| 969 | |
| 970 | for (i = 0; i < nrel; i++, reloff += sizeof (reloc_info)) |
| 971 | { |
| 972 | if (lseek (infd, reloff - linkedit_delta, L_SET) |
| 973 | != reloff - linkedit_delta) |
| 974 | unexec_error ("rebase_reloc_table: cannot seek to reloc_info"); |
| 975 | if (!unexec_read (&reloc_info, sizeof (reloc_info))) |
| 976 | unexec_error ("rebase_reloc_table: cannot read reloc_info"); |
| 977 | |
| 978 | if (sc_reloc_info->r_scattered == 0 |
| 979 | && reloc_info.r_type == GENERIC_RELOC_VANILLA) |
| 980 | { |
| 981 | reloc_info.r_address -= diff; |
| 982 | if (!unexec_write (reloff, &reloc_info, sizeof (reloc_info))) |
| 983 | unexec_error ("rebase_reloc_table: cannot write reloc_info"); |
| 984 | } |
| 985 | } |
| 986 | } |
| 987 | #endif |
| 988 | |
| 989 | /* Copy a LC_DYSYMTAB load command from the input file to the output |
| 990 | file, adjusting the file offset fields. */ |
| 991 | static void |
| 992 | copy_dysymtab (struct load_command *lc, long delta) |
| 993 | { |
| 994 | struct dysymtab_command *dstp = (struct dysymtab_command *) lc; |
| 995 | vm_address_t base; |
| 996 | |
| 997 | #ifdef _LP64 |
| 998 | #if __ppc64__ |
| 999 | { |
| 1000 | int i; |
| 1001 | |
| 1002 | base = 0; |
| 1003 | for (i = 0; i < nlc; i++) |
| 1004 | if (lca[i]->cmd == LC_SEGMENT) |
| 1005 | { |
| 1006 | struct segment_command *scp = (struct segment_command *) lca[i]; |
| 1007 | |
| 1008 | if (scp->vmaddr + scp->vmsize > 0x100000000 |
| 1009 | && (scp->initprot & VM_PROT_WRITE) != 0) |
| 1010 | { |
| 1011 | base = data_segment_scp->vmaddr; |
| 1012 | break; |
| 1013 | } |
| 1014 | } |
| 1015 | } |
| 1016 | #else |
| 1017 | /* First writable segment address. */ |
| 1018 | base = data_segment_scp->vmaddr; |
| 1019 | #endif |
| 1020 | #else |
| 1021 | /* First segment address in the file (unless MH_SPLIT_SEGS set). */ |
| 1022 | base = 0; |
| 1023 | #endif |
| 1024 | |
| 1025 | unrelocate ("local", dstp->locreloff, dstp->nlocrel, base); |
| 1026 | unrelocate ("external", dstp->extreloff, dstp->nextrel, base); |
| 1027 | |
| 1028 | if (dstp->nextrel > 0) { |
| 1029 | dstp->extreloff += delta; |
| 1030 | } |
| 1031 | |
| 1032 | if (dstp->nlocrel > 0) { |
| 1033 | dstp->locreloff += delta; |
| 1034 | } |
| 1035 | |
| 1036 | if (dstp->nindirectsyms > 0) |
| 1037 | dstp->indirectsymoff += delta; |
| 1038 | |
| 1039 | printf ("Writing LC_DYSYMTAB command\n"); |
| 1040 | |
| 1041 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) |
| 1042 | unexec_error ("cannot write symtab command to header"); |
| 1043 | |
| 1044 | curr_header_offset += lc->cmdsize; |
| 1045 | |
| 1046 | #if __ppc64__ |
| 1047 | /* Check if the relocation base needs to be changed. */ |
| 1048 | if (base == 0) |
| 1049 | { |
| 1050 | vm_address_t newbase = 0; |
| 1051 | int i; |
| 1052 | |
| 1053 | for (i = 0; i < num_unexec_regions; i++) |
| 1054 | if (unexec_regions[i].range.address + unexec_regions[i].range.size |
| 1055 | > 0x100000000) |
| 1056 | { |
| 1057 | newbase = data_segment_scp->vmaddr; |
| 1058 | break; |
| 1059 | } |
| 1060 | |
| 1061 | if (newbase) |
| 1062 | { |
| 1063 | rebase_reloc_address (dstp->locreloff, dstp->nlocrel, delta, newbase); |
| 1064 | rebase_reloc_address (dstp->extreloff, dstp->nextrel, delta, newbase); |
| 1065 | } |
| 1066 | } |
| 1067 | #endif |
| 1068 | } |
| 1069 | |
| 1070 | /* Copy a LC_TWOLEVEL_HINTS load command from the input file to the output |
| 1071 | file, adjusting the file offset fields. */ |
| 1072 | static void |
| 1073 | copy_twolevelhints (struct load_command *lc, long delta) |
| 1074 | { |
| 1075 | struct twolevel_hints_command *tlhp = (struct twolevel_hints_command *) lc; |
| 1076 | |
| 1077 | if (tlhp->nhints > 0) { |
| 1078 | tlhp->offset += delta; |
| 1079 | } |
| 1080 | |
| 1081 | printf ("Writing LC_TWOLEVEL_HINTS command\n"); |
| 1082 | |
| 1083 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) |
| 1084 | unexec_error ("cannot write two level hint command to header"); |
| 1085 | |
| 1086 | curr_header_offset += lc->cmdsize; |
| 1087 | } |
| 1088 | |
| 1089 | /* Copy other kinds of load commands from the input file to the output |
| 1090 | file, ones that do not require adjustments of file offsets. */ |
| 1091 | static void |
| 1092 | copy_other (struct load_command *lc) |
| 1093 | { |
| 1094 | printf ("Writing "); |
| 1095 | print_load_command_name (lc->cmd); |
| 1096 | printf (" command\n"); |
| 1097 | |
| 1098 | if (!unexec_write (curr_header_offset, lc, lc->cmdsize)) |
| 1099 | unexec_error ("cannot write symtab command to header"); |
| 1100 | |
| 1101 | curr_header_offset += lc->cmdsize; |
| 1102 | } |
| 1103 | |
| 1104 | /* Loop through all load commands and dump them. Then write the Mach |
| 1105 | header. */ |
| 1106 | static void |
| 1107 | dump_it () |
| 1108 | { |
| 1109 | int i; |
| 1110 | long linkedit_delta = 0; |
| 1111 | |
| 1112 | printf ("--- Load Commands written to Output File ---\n"); |
| 1113 | |
| 1114 | for (i = 0; i < nlc; i++) |
| 1115 | switch (lca[i]->cmd) |
| 1116 | { |
| 1117 | case LC_SEGMENT: |
| 1118 | { |
| 1119 | struct segment_command *scp = (struct segment_command *) lca[i]; |
| 1120 | if (strncmp (scp->segname, SEG_DATA, 16) == 0) |
| 1121 | { |
| 1122 | /* save data segment file offset and segment_command for |
| 1123 | unrelocate */ |
| 1124 | if (data_segment_old_fileoff) |
| 1125 | unexec_error ("cannot handle multiple DATA segments" |
| 1126 | " in input file"); |
| 1127 | data_segment_old_fileoff = scp->fileoff; |
| 1128 | data_segment_scp = scp; |
| 1129 | |
| 1130 | copy_data_segment (lca[i]); |
| 1131 | } |
| 1132 | else |
| 1133 | { |
| 1134 | if (strncmp (scp->segname, SEG_LINKEDIT, 16) == 0) |
| 1135 | { |
| 1136 | if (linkedit_delta) |
| 1137 | unexec_error ("cannot handle multiple LINKEDIT segments" |
| 1138 | " in input file"); |
| 1139 | linkedit_delta = curr_file_offset - scp->fileoff; |
| 1140 | } |
| 1141 | |
| 1142 | copy_segment (lca[i]); |
| 1143 | } |
| 1144 | } |
| 1145 | break; |
| 1146 | case LC_SYMTAB: |
| 1147 | copy_symtab (lca[i], linkedit_delta); |
| 1148 | break; |
| 1149 | case LC_DYSYMTAB: |
| 1150 | copy_dysymtab (lca[i], linkedit_delta); |
| 1151 | break; |
| 1152 | case LC_TWOLEVEL_HINTS: |
| 1153 | copy_twolevelhints (lca[i], linkedit_delta); |
| 1154 | break; |
| 1155 | default: |
| 1156 | copy_other (lca[i]); |
| 1157 | break; |
| 1158 | } |
| 1159 | |
| 1160 | if (curr_header_offset > text_seg_lowest_offset) |
| 1161 | unexec_error ("not enough room for load commands for new __DATA segments"); |
| 1162 | |
| 1163 | printf ("%ld unused bytes follow Mach-O header\n", |
| 1164 | text_seg_lowest_offset - curr_header_offset); |
| 1165 | |
| 1166 | mh.sizeofcmds = curr_header_offset - sizeof (struct mach_header); |
| 1167 | if (!unexec_write (0, &mh, sizeof (struct mach_header))) |
| 1168 | unexec_error ("cannot write final header contents"); |
| 1169 | } |
| 1170 | |
| 1171 | /* Take a snapshot of Emacs and make a Mach-O format executable file |
| 1172 | from it. The file names of the output and input files are outfile |
| 1173 | and infile, respectively. The three other parameters are |
| 1174 | ignored. */ |
| 1175 | void |
| 1176 | unexec (char *outfile, char *infile, void *start_data, void *start_bss, |
| 1177 | void *entry_address) |
| 1178 | { |
| 1179 | if (in_dumped_exec) |
| 1180 | unexec_error ("Unexec from a dumped executable is not supported."); |
| 1181 | |
| 1182 | pagesize = getpagesize (); |
| 1183 | infd = open (infile, O_RDONLY, 0); |
| 1184 | if (infd < 0) |
| 1185 | { |
| 1186 | unexec_error ("cannot open input file `%s'", infile); |
| 1187 | } |
| 1188 | |
| 1189 | outfd = open (outfile, O_WRONLY | O_TRUNC | O_CREAT, 0755); |
| 1190 | if (outfd < 0) |
| 1191 | { |
| 1192 | close (infd); |
| 1193 | unexec_error ("cannot open output file `%s'", outfile); |
| 1194 | } |
| 1195 | |
| 1196 | build_region_list (); |
| 1197 | read_load_commands (); |
| 1198 | |
| 1199 | find_emacs_zone_regions (); |
| 1200 | unexec_regions_merge (); |
| 1201 | |
| 1202 | in_dumped_exec = 1; |
| 1203 | |
| 1204 | dump_it (); |
| 1205 | |
| 1206 | close (outfd); |
| 1207 | } |
| 1208 | |
| 1209 | |
| 1210 | void |
| 1211 | unexec_init_emacs_zone () |
| 1212 | { |
| 1213 | emacs_zone = malloc_create_zone (0, 0); |
| 1214 | malloc_set_zone_name (emacs_zone, "EmacsZone"); |
| 1215 | } |
| 1216 | |
| 1217 | #ifndef MACOSX_MALLOC_MULT16 |
| 1218 | #define MACOSX_MALLOC_MULT16 1 |
| 1219 | #endif |
| 1220 | |
| 1221 | typedef struct unexec_malloc_header { |
| 1222 | union { |
| 1223 | char c[8]; |
| 1224 | size_t size; |
| 1225 | } u; |
| 1226 | } unexec_malloc_header_t; |
| 1227 | |
| 1228 | #if MACOSX_MALLOC_MULT16 |
| 1229 | |
| 1230 | #define ptr_in_unexec_regions(p) ((((vm_address_t) (p)) & 8) != 0) |
| 1231 | |
| 1232 | #else |
| 1233 | |
| 1234 | int |
| 1235 | ptr_in_unexec_regions (void *ptr) |
| 1236 | { |
| 1237 | int i; |
| 1238 | |
| 1239 | for (i = 0; i < num_unexec_regions; i++) |
| 1240 | if ((vm_address_t) ptr - unexec_regions[i].range.address |
| 1241 | < unexec_regions[i].range.size) |
| 1242 | return 1; |
| 1243 | |
| 1244 | return 0; |
| 1245 | } |
| 1246 | |
| 1247 | #endif |
| 1248 | |
| 1249 | void * |
| 1250 | unexec_malloc (size_t size) |
| 1251 | { |
| 1252 | if (in_dumped_exec) |
| 1253 | { |
| 1254 | void *p; |
| 1255 | |
| 1256 | p = malloc (size); |
| 1257 | #if MACOSX_MALLOC_MULT16 |
| 1258 | assert (((vm_address_t) p % 16) == 0); |
| 1259 | #endif |
| 1260 | return p; |
| 1261 | } |
| 1262 | else |
| 1263 | { |
| 1264 | unexec_malloc_header_t *ptr; |
| 1265 | |
| 1266 | ptr = (unexec_malloc_header_t *) |
| 1267 | malloc_zone_malloc (emacs_zone, size + sizeof (unexec_malloc_header_t)); |
| 1268 | ptr->u.size = size; |
| 1269 | ptr++; |
| 1270 | #if MACOSX_MALLOC_MULT16 |
| 1271 | assert (((vm_address_t) ptr % 16) == 8); |
| 1272 | #endif |
| 1273 | return (void *) ptr; |
| 1274 | } |
| 1275 | } |
| 1276 | |
| 1277 | void * |
| 1278 | unexec_realloc (void *old_ptr, size_t new_size) |
| 1279 | { |
| 1280 | if (in_dumped_exec) |
| 1281 | { |
| 1282 | void *p; |
| 1283 | |
| 1284 | if (ptr_in_unexec_regions (old_ptr)) |
| 1285 | { |
| 1286 | size_t old_size = ((unexec_malloc_header_t *) old_ptr)[-1].u.size; |
| 1287 | size_t size = new_size > old_size ? old_size : new_size; |
| 1288 | |
| 1289 | p = (size_t *) malloc (new_size); |
| 1290 | if (size) |
| 1291 | memcpy (p, old_ptr, size); |
| 1292 | } |
| 1293 | else |
| 1294 | { |
| 1295 | p = realloc (old_ptr, new_size); |
| 1296 | } |
| 1297 | #if MACOSX_MALLOC_MULT16 |
| 1298 | assert (((vm_address_t) p % 16) == 0); |
| 1299 | #endif |
| 1300 | return p; |
| 1301 | } |
| 1302 | else |
| 1303 | { |
| 1304 | unexec_malloc_header_t *ptr; |
| 1305 | |
| 1306 | ptr = (unexec_malloc_header_t *) |
| 1307 | malloc_zone_realloc (emacs_zone, (unexec_malloc_header_t *) old_ptr - 1, |
| 1308 | new_size + sizeof (unexec_malloc_header_t)); |
| 1309 | ptr->u.size = new_size; |
| 1310 | ptr++; |
| 1311 | #if MACOSX_MALLOC_MULT16 |
| 1312 | assert (((vm_address_t) ptr % 16) == 8); |
| 1313 | #endif |
| 1314 | return (void *) ptr; |
| 1315 | } |
| 1316 | } |
| 1317 | |
| 1318 | void |
| 1319 | unexec_free (void *ptr) |
| 1320 | { |
| 1321 | if (ptr == NULL) |
| 1322 | return; |
| 1323 | if (in_dumped_exec) |
| 1324 | { |
| 1325 | if (!ptr_in_unexec_regions (ptr)) |
| 1326 | free (ptr); |
| 1327 | } |
| 1328 | else |
| 1329 | malloc_zone_free (emacs_zone, (unexec_malloc_header_t *) ptr - 1); |
| 1330 | } |
| 1331 | |
| 1332 | /* arch-tag: 1a784f7b-a184-4c4f-9544-da8619593d72 |
| 1333 | (do not change this comment) */ |