]> git.saurik.com Git - apple/ld64.git/blob - src/rebase.cpp
projects / apple / ld64.git / blob
? search:


ad9b905eaed5e81d451dc48df1799083fbe239e2
[apple/ld64.git] / src / rebase.cpp
1 /* -*- mode: C++; c-basic-offset: 4; tab-width: 4 -*-
2 *
3 * Copyright (c) 2006 Apple Computer, Inc. All rights reserved.
4 *
5 * @APPLE_LICENSE_HEADER_START@
6 *
7 * This file contains Original Code and/or Modifications of Original Code
8 * as defined in and that are subject to the Apple Public Source License
9 * Version 2.0 (the 'License'). You may not use this file except in
10 * compliance with the License. Please obtain a copy of the License at
11 * http://www.opensource.apple.com/apsl/ and read it before using this
12 * file.
13 *
14 * The Original Code and all software distributed under the License are
15 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
16 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
17 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
19 * Please see the License for the specific language governing rights and
20 * limitations under the License.
21 *
22 * @APPLE_LICENSE_HEADER_END@
23 */
24
25 #include <sys/types.h>
26 #include <sys/stat.h>
27 #include <sys/mman.h>
28 #include <mach/mach.h>
29 #include <limits.h>
30 #include <stdarg.h>
31 #include <stdio.h>
32 #include <fcntl.h>
33 #include <errno.h>
34 #include <unistd.h>
35 #include <vector>
36 #include <set>
37
38
39 #include "MachOFileAbstraction.hpp"
40 #include "Architectures.hpp"
41
42 static bool verbose = false;
43
44 __attribute__((noreturn))
45 void throwf(const char* format, ...)
46 {
47 va_list list;
48 char* p;
49 va_start(list, format);
50 vasprintf(&p, format, list);
51 va_end(list);
52
53 const char* t = p;
54 throw t;
55 }
56
57
58 class AbstractRebaser
59 {
60 public:
61 virtual cpu_type_t getArchitecture() const = 0;
62 virtual uint64_t getBaseAddress() const = 0;
63 virtual uint64_t getVMSize() const = 0;
64 virtual void setBaseAddress(uint64_t) = 0;
65 };
66
67
68 template <typename A>
69 class Rebaser : public AbstractRebaser
70 {
71 public:
72 Rebaser(const void* machHeader);
73 virtual ~Rebaser() {}
74
75 virtual cpu_type_t getArchitecture() const;
76 virtual uint64_t getBaseAddress() const;
77 virtual uint64_t getVMSize() const;
78 virtual void setBaseAddress(uint64_t);
79
80 private:
81 typedef typename A::P P;
82 typedef typename A::P::E E;
83 typedef typename A::P::uint_t pint_t;
84
85 struct vmmap { pint_t vmaddr; pint_t vmsize; pint_t fileoff; };
86
87 void setRelocBase();
88 void buildSectionTable();
89 void adjustLoadCommands();
90 void adjustSymbolTable();
91 void adjustDATA();
92 void doLocalRelocation(const macho_relocation_info<P>* reloc);
93 pint_t* mappedAddressForVMAddress(uint32_t vmaddress);
94
95 const macho_header<P>* fHeader;
96 pint_t fOrignalVMRelocBaseAddress;
97 pint_t fSlide;
98 pint_t fRelocBase;
99 std::vector<vmmap> fVMMApping;
100 };
101
102
103
104 class MultiArchRebaser
105 {
106 public:
107 MultiArchRebaser(const char* path, bool writable=false);
108 ~MultiArchRebaser();
109
110 const std::vector<AbstractRebaser*>& getArchs() const { return fRebasers; }
111 void commit();
112
113 private:
114 std::vector<AbstractRebaser*> fRebasers;
115 void* fMappingAddress;
116 uint64_t fFileSize;
117 };
118
119
120
121 MultiArchRebaser::MultiArchRebaser(const char* path, bool writable)
122 : fMappingAddress(0), fFileSize(0)
123 {
124 // map in whole file
125 int fd = ::open(path, (writable ? O_RDWR : O_RDONLY), 0);
126 if ( fd == -1 )
127 throwf("can't open file, errno=%d", errno);
128 struct stat stat_buf;
129 if ( fstat(fd, &stat_buf) == -1)
130 throwf("can't stat open file %s, errno=%d", path, errno);
131 if ( stat_buf.st_size < 20 )
132 throwf("file too small %s", path);
133 const int prot = writable ? (PROT_READ | PROT_WRITE) : PROT_READ;
134 const int flags = writable ? (MAP_FILE | MAP_SHARED) : (MAP_FILE | MAP_PRIVATE);
135 uint8_t* p = (uint8_t*)::mmap(NULL, stat_buf.st_size, prot, flags, fd, 0);
136 if ( p == (uint8_t*)(-1) )
137 throwf("can't map file %s, errno=%d", path, errno);
138 ::close(fd);
139
140 // if fat file, process each architecture
141 const fat_header* fh = (fat_header*)p;
142 const mach_header* mh = (mach_header*)p;
143 if ( fh->magic == OSSwapBigToHostInt32(FAT_MAGIC) ) {
144 // Fat header is always big-endian
145 const struct fat_arch* archs = (struct fat_arch*)(p + sizeof(struct fat_header));
146 for (unsigned long i=0; i < OSSwapBigToHostInt32(fh->nfat_arch); ++i) {
147 uint32_t fileOffset = OSSwapBigToHostInt32(archs[i].offset);
148 try {
149 switch ( OSSwapBigToHostInt32(archs[i].cputype) ) {
150 case CPU_TYPE_POWERPC:
151 fRebasers.push_back(new Rebaser<ppc>(&p[fileOffset]));
152 break;
153 case CPU_TYPE_POWERPC64:
154 fRebasers.push_back(new Rebaser<ppc64>(&p[fileOffset]));
155 break;
156 case CPU_TYPE_I386:
157 fRebasers.push_back(new Rebaser<x86>(&p[fileOffset]));
158 break;
159 case CPU_TYPE_X86_64:
160 fRebasers.push_back(new Rebaser<x86_64>(&p[fileOffset]));
161 break;
162 case CPU_TYPE_ARM:
163 fRebasers.push_back(new Rebaser<arm>(&p[fileOffset]));
164 break;
165 default:
166 throw "unknown file format";
167 }
168 }
169 catch (const char* msg) {
170 fprintf(stderr, "rebase warning: %s for %s\n", msg, path);
171 }
172 }
173 }
174 else {
175 try {
176 if ( (OSSwapBigToHostInt32(mh->magic) == MH_MAGIC) && (OSSwapBigToHostInt32(mh->cputype) == CPU_TYPE_POWERPC)) {
177 fRebasers.push_back(new Rebaser<ppc>(mh));
178 }
179 else if ( (OSSwapBigToHostInt32(mh->magic) == MH_MAGIC_64) && (OSSwapBigToHostInt32(mh->cputype) == CPU_TYPE_POWERPC64)) {
180 fRebasers.push_back(new Rebaser<ppc64>(mh));
181 }
182 else if ( (OSSwapLittleToHostInt32(mh->magic) == MH_MAGIC) && (OSSwapLittleToHostInt32(mh->cputype) == CPU_TYPE_I386)) {
183 fRebasers.push_back(new Rebaser<x86>(mh));
184 }
185 else if ( (OSSwapLittleToHostInt32(mh->magic) == MH_MAGIC_64) && (OSSwapLittleToHostInt32(mh->cputype) == CPU_TYPE_X86_64)) {
186 fRebasers.push_back(new Rebaser<x86_64>(mh));
187 }
188 else if ( (OSSwapLittleToHostInt32(mh->magic) == MH_MAGIC) && (OSSwapLittleToHostInt32(mh->cputype) == CPU_TYPE_ARM)) {
189 fRebasers.push_back(new Rebaser<arm>(mh));
190 }
191 else {
192 throw "unknown file format";
193 }
194 }
195 catch (const char* msg) {
196 fprintf(stderr, "rebase warning: %s for %s\n", msg, path);
197 }
198 }
199
200 fMappingAddress = p;
201 fFileSize = stat_buf.st_size;
202 }
203
204
205 MultiArchRebaser::~MultiArchRebaser()
206 {
207 ::munmap(fMappingAddress, fFileSize);
208 }
209
210 void MultiArchRebaser::commit()
211 {
212 ::msync(fMappingAddress, fFileSize, MS_ASYNC);
213 }
214
215
216
217 template <typename A>
218 Rebaser<A>::Rebaser(const void* machHeader)
219 : fHeader((const macho_header<P>*)machHeader)
220 {
221 switch ( fHeader->filetype() ) {
222 case MH_DYLIB:
223 if ( (fHeader->flags() & MH_SPLIT_SEGS) != 0 )
224 throw "split-seg dylibs cannot be rebased";
225 break;
226 case MH_BUNDLE:
227 break;
228 default:
229 throw "file is not a dylib or bundle";
230 }
231
232 }
233
234 template <> cpu_type_t Rebaser<ppc>::getArchitecture() const { return CPU_TYPE_POWERPC; }
235 template <> cpu_type_t Rebaser<ppc64>::getArchitecture() const { return CPU_TYPE_POWERPC64; }
236 template <> cpu_type_t Rebaser<x86>::getArchitecture() const { return CPU_TYPE_I386; }
237 template <> cpu_type_t Rebaser<x86_64>::getArchitecture() const { return CPU_TYPE_X86_64; }
238 template <> cpu_type_t Rebaser<arm>::getArchitecture() const { return CPU_TYPE_ARM; }
239
240 template <typename A>
241 uint64_t Rebaser<A>::getBaseAddress() const
242 {
243 uint64_t lowestSegmentAddress = LLONG_MAX;
244 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((uint8_t*)fHeader + sizeof(macho_header<P>));
245 const uint32_t cmd_count = fHeader->ncmds();
246 const macho_load_command<P>* cmd = cmds;
247 for (uint32_t i = 0; i < cmd_count; ++i) {
248 if ( cmd->cmd() == macho_segment_command<P>::CMD ) {
249 const macho_segment_command<P>* segCmd = (const macho_segment_command<P>*)cmd;
250 if ( segCmd->vmaddr() < lowestSegmentAddress ) {
251 lowestSegmentAddress = segCmd->vmaddr();
252 }
253 }
254 cmd = (const macho_load_command<P>*)(((uint8_t*)cmd)+cmd->cmdsize());
255 }
256 return lowestSegmentAddress;
257 }
258
259 template <typename A>
260 uint64_t Rebaser<A>::getVMSize() const
261 {
262 const macho_segment_command<P>* highestSegmentCmd = NULL;
263 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((uint8_t*)fHeader + sizeof(macho_header<P>));
264 const uint32_t cmd_count = fHeader->ncmds();
265 const macho_load_command<P>* cmd = cmds;
266 for (uint32_t i = 0; i < cmd_count; ++i) {
267 if ( cmd->cmd() == macho_segment_command<P>::CMD ) {
268 const macho_segment_command<P>* segCmd = (const macho_segment_command<P>*)cmd;
269 if ( (highestSegmentCmd == NULL) || (segCmd->vmaddr() > highestSegmentCmd->vmaddr()) ) {
270 highestSegmentCmd = segCmd;
271 }
272 }
273 cmd = (const macho_load_command<P>*)(((uint8_t*)cmd)+cmd->cmdsize());
274 }
275
276 return ((highestSegmentCmd->vmaddr() + highestSegmentCmd->vmsize() - this->getBaseAddress() + 4095) & (-4096));
277 }
278
279
280 template <typename A>
281 void Rebaser<A>::setBaseAddress(uint64_t addr)
282 {
283 // calculate slide
284 fSlide = addr - this->getBaseAddress();
285
286 // compute base address for relocations
287 this->setRelocBase();
288
289 // build cache of section index to section
290 this->buildSectionTable();
291
292 // update load commands
293 this->adjustLoadCommands();
294
295 // update symbol table
296 this->adjustSymbolTable();
297
298 // update writable segments that have internal pointers
299 this->adjustDATA();
300 }
301
302 template <typename A>
303 void Rebaser<A>::adjustLoadCommands()
304 {
305 const macho_segment_command<P>* highestSegmentCmd = NULL;
306 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((uint8_t*)fHeader + sizeof(macho_header<P>));
307 const uint32_t cmd_count = fHeader->ncmds();
308 const macho_load_command<P>* cmd = cmds;
309 for (uint32_t i = 0; i < cmd_count; ++i) {
310 switch ( cmd->cmd() ) {
311 case LC_ID_DYLIB:
312 if ( (fHeader->flags() & MH_PREBOUND) != 0 ) {
313 // clear timestamp so that any prebound clients are invalidated
314 macho_dylib_command<P>* dylib = (macho_dylib_command<P>*)cmd;
315 dylib->set_timestamp(1);
316 }
317 break;
318 case LC_LOAD_DYLIB:
319 case LC_LOAD_WEAK_DYLIB:
320 if ( (fHeader->flags() & MH_PREBOUND) != 0 ) {
321 // clear expected timestamps so that this image will load with invalid prebinding
322 macho_dylib_command<P>* dylib = (macho_dylib_command<P>*)cmd;
323 dylib->set_timestamp(2);
324 }
325 break;
326 case macho_routines_command<P>::CMD:
327 // update -init command
328 {
329 struct macho_routines_command<P>* routines = (struct macho_routines_command<P>*)cmd;
330 routines->set_init_address(routines->init_address() + fSlide);
331 }
332 break;
333 case macho_segment_command<P>::CMD:
334 // update segment commands
335 {
336 macho_segment_command<P>* seg = (macho_segment_command<P>*)cmd;
337 seg->set_vmaddr(seg->vmaddr() + fSlide);
338 macho_section<P>* const sectionsStart = (macho_section<P>*)((char*)seg + sizeof(macho_segment_command<P>));
339 macho_section<P>* const sectionsEnd = &sectionsStart[seg->nsects()];
340 for(macho_section<P>* sect = sectionsStart; sect < sectionsEnd; ++sect) {
341 sect->set_addr(sect->addr() + fSlide);
342 }
343 }
344 break;
345 }
346 cmd = (const macho_load_command<P>*)(((uint8_t*)cmd)+cmd->cmdsize());
347 }
348 }
349
350
351 template <typename A>
352 void Rebaser<A>::buildSectionTable()
353 {
354 // build vector of sections
355 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((uint8_t*)fHeader + sizeof(macho_header<P>));
356 const uint32_t cmd_count = fHeader->ncmds();
357 const macho_load_command<P>* cmd = cmds;
358 for (uint32_t i = 0; i < cmd_count; ++i) {
359 if ( cmd->cmd() == macho_segment_command<P>::CMD ) {
360 const macho_segment_command<P>* seg = (macho_segment_command<P>*)cmd;
361 vmmap mapping;
362 mapping.vmaddr = seg->vmaddr();
363 mapping.vmsize = seg->vmsize();
364 mapping.fileoff = seg->fileoff();
365 fVMMApping.push_back(mapping);
366 }
367 cmd = (const macho_load_command<P>*)(((uint8_t*)cmd)+cmd->cmdsize());
368 }
369 }
370
371
372 template <typename A>
373 void Rebaser<A>::adjustSymbolTable()
374 {
375 const macho_dysymtab_command<P>* dysymtab = NULL;
376 macho_nlist<P>* symbolTable = NULL;
377
378 // get symbol table info
379 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((uint8_t*)fHeader + sizeof(macho_header<P>));
380 const uint32_t cmd_count = fHeader->ncmds();
381 const macho_load_command<P>* cmd = cmds;
382 for (uint32_t i = 0; i < cmd_count; ++i) {
383 switch (cmd->cmd()) {
384 case LC_SYMTAB:
385 {
386 const macho_symtab_command<P>* symtab = (macho_symtab_command<P>*)cmd;
387 symbolTable = (macho_nlist<P>*)(((uint8_t*)fHeader) + symtab->symoff());
388 }
389 break;
390 case LC_DYSYMTAB:
391 dysymtab = (macho_dysymtab_command<P>*)cmd;
392 break;
393 }
394 cmd = (const macho_load_command<P>*)(((uint8_t*)cmd)+cmd->cmdsize());
395 }
396
397 // walk all exports and slide their n_value
398 macho_nlist<P>* lastExport = &symbolTable[dysymtab->iextdefsym()+dysymtab->nextdefsym()];
399 for (macho_nlist<P>* entry = &symbolTable[dysymtab->iextdefsym()]; entry < lastExport; ++entry) {
400 if ( (entry->n_type() & N_TYPE) == N_SECT )
401 entry->set_n_value(entry->n_value() + fSlide);
402 }
403
404 // walk all local symbols and slide their n_value
405 macho_nlist<P>* lastLocal = &symbolTable[dysymtab->ilocalsym()+dysymtab->nlocalsym()];
406 for (macho_nlist<P>* entry = &symbolTable[dysymtab->ilocalsym()]; entry < lastLocal; ++entry) {
407 if ( entry->n_sect() != NO_SECT )
408 entry->set_n_value(entry->n_value() + fSlide);
409 }
410
411 // FIXME ¥¥¥ adjust dylib_module if it exists
412 }
413
414 template <typename A>
415 void Rebaser<A>::adjustDATA()
416 {
417 const macho_dysymtab_command<P>* dysymtab = NULL;
418
419 // get symbol table info
420 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((uint8_t*)fHeader + sizeof(macho_header<P>));
421 const uint32_t cmd_count = fHeader->ncmds();
422 const macho_load_command<P>* cmd = cmds;
423 for (uint32_t i = 0; i < cmd_count; ++i) {
424 switch (cmd->cmd()) {
425 case LC_DYSYMTAB:
426 dysymtab = (macho_dysymtab_command<P>*)cmd;
427 break;
428 }
429 cmd = (const macho_load_command<P>*)(((uint8_t*)cmd)+cmd->cmdsize());
430 }
431
432 // walk all local relocations and slide every pointer
433 const macho_relocation_info<P>* const relocsStart = (macho_relocation_info<P>*)(((uint8_t*)fHeader) + dysymtab->locreloff());
434 const macho_relocation_info<P>* const relocsEnd = &relocsStart[dysymtab->nlocrel()];
435 for (const macho_relocation_info<P>* reloc=relocsStart; reloc < relocsEnd; ++reloc) {
436 this->doLocalRelocation(reloc);
437 }
438
439 // walk non-lazy-pointers and slide the ones that are LOCAL
440 cmd = cmds;
441 for (uint32_t i = 0; i < cmd_count; ++i) {
442 if ( cmd->cmd() == macho_segment_command<P>::CMD ) {
443 const macho_segment_command<P>* seg = (macho_segment_command<P>*)cmd;
444 const macho_section<P>* const sectionsStart = (macho_section<P>*)((char*)seg + sizeof(macho_segment_command<P>));
445 const macho_section<P>* const sectionsEnd = &sectionsStart[seg->nsects()];
446 const uint32_t* const indirectTable = (uint32_t*)(((uint8_t*)fHeader) + dysymtab->indirectsymoff());
447 for(const macho_section<P>* sect = sectionsStart; sect < sectionsEnd; ++sect) {
448 if ( (sect->flags() & SECTION_TYPE) == S_NON_LAZY_SYMBOL_POINTERS ) {
449 const uint32_t indirectTableOffset = sect->reserved1();
450 uint32_t pointerCount = sect->size() / sizeof(pint_t);
451 pint_t* nonLazyPointer = (pint_t*)(((uint8_t*)fHeader) + sect->offset());
452 for (uint32_t i=0; i < pointerCount; ++i, ++nonLazyPointer) {
453 if ( E::get32(indirectTable[indirectTableOffset + i]) == INDIRECT_SYMBOL_LOCAL ) {
454 P::setP(*nonLazyPointer, A::P::getP(*nonLazyPointer) + fSlide);
455 }
456 }
457 }
458 }
459 }
460 cmd = (const macho_load_command<P>*)(((uint8_t*)cmd)+cmd->cmdsize());
461 }
462
463 }
464
465
466 template <typename A>
467 typename A::P::uint_t* Rebaser<A>::mappedAddressForVMAddress(uint32_t vmaddress)
468 {
469 for(typename std::vector<vmmap>::iterator it = fVMMApping.begin(); it != fVMMApping.end(); ++it) {
470 //fprintf(stderr, "vmaddr=0x%08lX, vmsize=0x%08lX\n", it->vmaddr, it->vmsize);
471 if ( (vmaddress >= it->vmaddr) && (vmaddress < (it->vmaddr+it->vmsize)) ) {
472 return (pint_t*)((vmaddress - it->vmaddr) + it->fileoff + (uint8_t*)fHeader);
473 }
474 }
475 throwf("reloc address 0x%08X not found", vmaddress);
476 }
477
478
479 template <>
480 void Rebaser<x86_64>::doLocalRelocation(const macho_relocation_info<x86_64::P>* reloc)
481 {
482 if ( reloc->r_type() == X86_64_RELOC_UNSIGNED ) {
483 pint_t* addr = mappedAddressForVMAddress(reloc->r_address() + fOrignalVMRelocBaseAddress);
484 P::setP(*addr, P::getP(*addr) + fSlide);
485 }
486 else {
487 throw "invalid relocation type";
488 }
489 }
490
491 template <>
492 void Rebaser<ppc>::doLocalRelocation(const macho_relocation_info<P>* reloc)
493 {
494 if ( (reloc->r_address() & R_SCATTERED) == 0 ) {
495 if ( reloc->r_type() == GENERIC_RELOC_VANILLA ) {
496 pint_t* addr = mappedAddressForVMAddress(reloc->r_address() + fOrignalVMRelocBaseAddress);
497 P::setP(*addr, P::getP(*addr) + fSlide);
498 }
499 }
500 else {
501 macho_scattered_relocation_info<P>* sreloc = (macho_scattered_relocation_info<P>*)reloc;
502 if ( sreloc->r_type() == PPC_RELOC_PB_LA_PTR ) {
503 sreloc->set_r_value( sreloc->r_value() + fSlide );
504 }
505 else {
506 throw "cannot rebase final linked image with scattered relocations";
507 }
508 }
509 }
510
511 template <>
512 void Rebaser<x86>::doLocalRelocation(const macho_relocation_info<P>* reloc)
513 {
514 if ( (reloc->r_address() & R_SCATTERED) == 0 ) {
515 if ( reloc->r_type() == GENERIC_RELOC_VANILLA ) {
516 pint_t* addr = mappedAddressForVMAddress(reloc->r_address() + fOrignalVMRelocBaseAddress);
517 P::setP(*addr, P::getP(*addr) + fSlide);
518 }
519 }
520 else {
521 macho_scattered_relocation_info<P>* sreloc = (macho_scattered_relocation_info<P>*)reloc;
522 if ( sreloc->r_type() == GENERIC_RELOC_PB_LA_PTR ) {
523 sreloc->set_r_value( sreloc->r_value() + fSlide );
524 }
525 else {
526 throw "cannot rebase final linked image with scattered relocations";
527 }
528 }
529 }
530
531 template <>
532 void Rebaser<arm>::doLocalRelocation(const macho_relocation_info<P>* reloc)
533 {
534 if ( (reloc->r_address() & R_SCATTERED) == 0 ) {
535 if ( reloc->r_type() == ARM_RELOC_VANILLA ) {
536 pint_t* addr = mappedAddressForVMAddress(reloc->r_address() + fOrignalVMRelocBaseAddress);
537 P::setP(*addr, P::getP(*addr) + fSlide);
538 }
539 }
540 else {
541 macho_scattered_relocation_info<P>* sreloc = (macho_scattered_relocation_info<P>*)reloc;
542 if ( sreloc->r_type() == ARM_RELOC_PB_LA_PTR ) {
543 sreloc->set_r_value( sreloc->r_value() + fSlide );
544 }
545 else {
546 throw "cannot rebase final linked image with scattered relocations";
547 }
548 }
549 }
550
551 template <typename A>
552 void Rebaser<A>::doLocalRelocation(const macho_relocation_info<P>* reloc)
553 {
554 if ( (reloc->r_address() & R_SCATTERED) == 0 ) {
555 if ( reloc->r_type() == GENERIC_RELOC_VANILLA ) {
556 pint_t* addr = mappedAddressForVMAddress(reloc->r_address() + fOrignalVMRelocBaseAddress);
557 P::setP(*addr, P::getP(*addr) + fSlide);
558 }
559 }
560 else {
561 throw "cannot rebase final linked image with scattered relocations";
562 }
563 }
564
565
566 template <typename A>
567 void Rebaser<A>::setRelocBase()
568 {
569 // reloc addresses are from the start of the mapped file (base address)
570 fRelocBase = (pint_t)fHeader;
571 fOrignalVMRelocBaseAddress = this->getBaseAddress();
572 //fprintf(stderr, "fOrignalVMRelocBaseAddress=0x%08X\n", fOrignalVMRelocBaseAddress);
573 }
574
575 template <>
576 void Rebaser<ppc64>::setRelocBase()
577 {
578 // reloc addresses either:
579 // 1) from the base address if no writable segment is > 4GB from base address
580 // 2) from start of first writable segment
581 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((uint8_t*)fHeader + sizeof(macho_header<P>));
582 const uint32_t cmd_count = fHeader->ncmds();
583 const macho_load_command<P>* cmd = cmds;
584 for (uint32_t i = 0; i < cmd_count; ++i) {
585 if ( cmd->cmd() == macho_segment_command<P>::CMD ) {
586 const macho_segment_command<P>* segCmd = (const macho_segment_command<P>*)cmd;
587 if ( segCmd->initprot() & VM_PROT_WRITE ) {
588 if ( (segCmd->vmaddr() + segCmd->vmsize() - this->getBaseAddress()) > 0x100000000ULL ) {
589 // found writable segment with address > 4GB past base address
590 fRelocBase = segCmd->fileoff() + (pint_t)fHeader;
591 fOrignalVMRelocBaseAddress = segCmd->vmaddr();
592 return;
593 }
594 }
595 }
596 cmd = (const macho_load_command<P>*)(((uint8_t*)cmd)+cmd->cmdsize());
597 }
598 // just use base address
599 fRelocBase = (pint_t)fHeader;
600 fOrignalVMRelocBaseAddress = this->getBaseAddress();
601 }
602
603 template <>
604 void Rebaser<x86_64>::setRelocBase()
605 {
606 // reloc addresses are always based from the start of the first writable segment
607 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((uint8_t*)fHeader + sizeof(macho_header<P>));
608 const uint32_t cmd_count = fHeader->ncmds();
609 const macho_load_command<P>* cmd = cmds;
610 for (uint32_t i = 0; i < cmd_count; ++i) {
611 if ( cmd->cmd() == macho_segment_command<P>::CMD ) {
612 const macho_segment_command<P>* segCmd = (const macho_segment_command<P>*)cmd;
613 if ( segCmd->initprot() & VM_PROT_WRITE ) {
614 fRelocBase = segCmd->fileoff() + (pint_t)fHeader;
615 fOrignalVMRelocBaseAddress = segCmd->vmaddr();
616 return;
617 }
618 }
619 cmd = (const macho_load_command<P>*)(((uint8_t*)cmd)+cmd->cmdsize());
620 }
621 throw "no writable segment";
622 }
623
624
625 static void copyFile(const char* srcFile, const char* dstFile)
626 {
627 // open files
628 int src = open(srcFile, O_RDONLY);
629 if ( src == -1 )
630 throwf("can't open file %s, errno=%d", srcFile, errno);
631 struct stat stat_buf;
632 if ( fstat(src, &stat_buf) == -1)
633 throwf("can't stat open file %s, errno=%d", srcFile, errno);
634
635 // create new file with all same permissions to hold copy of dylib
636 ::unlink(dstFile);
637 int dst = open(dstFile, O_CREAT | O_RDWR | O_TRUNC, stat_buf.st_mode);
638 if ( dst == -1 )
639 throwf("can't create temp file %s, errnor=%d", dstFile, errno);
640
641 // mark source as "don't cache"
642 (void)fcntl(src, F_NOCACHE, 1);
643 // we want to cache the dst because we are about to map it in and modify it
644
645 // copy permission bits
646 if ( chmod(dstFile, stat_buf.st_mode & 07777) == -1 )
647 throwf("can't chmod temp file %s, errno=%d", dstFile, errno);
648 if ( chown(dstFile, stat_buf.st_uid, stat_buf.st_gid) == -1)
649 throwf("can't chown temp file %s, errno=%d", dstFile, errno);
650
651 // copy contents
652 ssize_t len;
653 const uint32_t kBufferSize = 128*1024;
654 static uint8_t* buffer = NULL;
655 if ( buffer == NULL ) {
656 vm_address_t addr = 0;
657 if ( vm_allocate(mach_task_self(), &addr, kBufferSize, true /*find range*/) == KERN_SUCCESS )
658 buffer = (uint8_t*)addr;
659 else
660 throw "can't allcoate copy buffer";
661 }
662 while ( (len = read(src, buffer, kBufferSize)) > 0 ) {
663 if ( write(dst, buffer, len) == -1 )
664 throwf("write failure copying feil %s, errno=%d", dstFile, errno);
665 }
666
667 // close files
668 int result1 = close(dst);
669 int result2 = close(src);
670 if ( (result1 != 0) || (result2 != 0) )
671 throw "can't close file";
672 }
673
674
675 // scan dylibs and collect size info
676 // calculate new base address for each dylib
677 // rebase each file
678 // copy to temp and mmap
679 // update content
680 // unmap/flush
681 // rename
682
683 struct archInfo {
684 cpu_type_t arch;
685 uint64_t vmSize;
686 uint64_t orgBase;
687 uint64_t newBase;
688 };
689
690 struct fileInfo
691 {
692 fileInfo(const char* p) : path(p) {}
693
694 const char* path;
695 std::vector<archInfo> archs;
696 };
697
698 //
699 // add archInfos to fileInfo for every slice of a fat file
700 // for ppc, there may be duplicate architectures (with different sub-types)
701 //
702 static void setSizes(fileInfo& info, const std::set<cpu_type_t>& onlyArchs)
703 {
704 const MultiArchRebaser mar(info.path);
705 const std::vector<AbstractRebaser*>& rebasers = mar.getArchs();
706 for(std::set<cpu_type_t>::iterator ait=onlyArchs.begin(); ait != onlyArchs.end(); ++ait) {
707 for(std::vector<AbstractRebaser*>::const_iterator rit=rebasers.begin(); rit != rebasers.end(); ++rit) {
708 AbstractRebaser* rebaser = *rit;
709 if ( rebaser->getArchitecture() == *ait ) {
710 archInfo ai;
711 ai.arch = *ait;
712 ai.vmSize = rebaser->getVMSize();
713 ai.orgBase = rebaser->getBaseAddress();
714 ai.newBase = 0;
715 //fprintf(stderr, "base=0x%llX, size=0x%llX\n", ai.orgBase, ai.vmSize);
716 info.archs.push_back(ai);
717 }
718 }
719 }
720 }
721
722 static const char* nameForArch(cpu_type_t arch)
723 {
724 switch( arch ) {
725 case CPU_TYPE_POWERPC:
726 return "ppc";
727 case CPU_TYPE_POWERPC64:
728 return "ppca64";
729 case CPU_TYPE_I386:
730 return "i386";
731 case CPU_TYPE_X86_64:
732 return "x86_64";
733 case CPU_TYPE_ARM:
734 return "arm";
735 }
736 return "unknown";
737 }
738
739 static void rebase(const fileInfo& info)
740 {
741 // generate temp file name
742 char realFilePath[PATH_MAX];
743 if ( realpath(info.path, realFilePath) == NULL ) {
744 throwf("realpath() failed on %s, errno=%d", info.path, errno);
745 }
746 const char* tempPath;
747 asprintf((char**)&tempPath, "%s_rebase", realFilePath);
748
749 // copy whole file to temp file
750 copyFile(info.path, tempPath);
751
752 try {
753 // rebase temp file
754 MultiArchRebaser mar(tempPath, true);
755 const std::vector<AbstractRebaser*>& rebasers = mar.getArchs();
756 for(std::vector<archInfo>::const_iterator fait=info.archs.begin(); fait != info.archs.end(); ++fait) {
757 for(std::vector<AbstractRebaser*>::const_iterator rit=rebasers.begin(); rit != rebasers.end(); ++rit) {
758 if ( (*rit)->getArchitecture() == fait->arch ) {
759 (*rit)->setBaseAddress(fait->newBase);
760 if ( verbose )
761 printf("%8s 0x%0llX -> 0x%0llX %s\n", nameForArch(fait->arch), fait->orgBase, fait->newBase, info.path);
762 }
763 }
764 }
765
766 // flush temp file out to disk
767 mar.commit();
768
769 // rename
770 int result = rename(tempPath, info.path);
771 if ( result != 0 ) {
772 throwf("can't swap temporary rebased file: rename(%s,%s) returned errno=%d", tempPath, info.path, errno);
773 }
774
775 // make sure every really gets out to disk
776 ::sync();
777 }
778 catch (const char* msg) {
779 // delete temp file
780 ::unlink(tempPath);
781
782 // throw exception with file name added
783 const char* newMsg;
784 asprintf((char**)&newMsg, "%s for file %s", msg, info.path);
785 throw newMsg;
786 }
787 }
788
789 static uint64_t totalVMSize(cpu_type_t arch, std::vector<fileInfo>& files)
790 {
791 uint64_t totalSize = 0;
792 for(std::vector<fileInfo>::iterator fit=files.begin(); fit != files.end(); ++fit) {
793 fileInfo& fi = *fit;
794 for(std::vector<archInfo>::iterator fait=fi.archs.begin(); fait != fi.archs.end(); ++fait) {
795 if ( fait->arch == arch )
796 totalSize += fait->vmSize;
797 }
798 }
799 return totalSize;
800 }
801
802 static uint64_t startAddress(cpu_type_t arch, std::vector<fileInfo>& files, uint64_t lowAddress, uint64_t highAddress)
803 {
804 if ( lowAddress != 0 )
805 return lowAddress;
806 else if ( highAddress != 0 ) {
807 uint64_t totalSize = totalVMSize(arch, files);
808 if ( highAddress < totalSize )
809 throwf("cannot use -high_address 0x%X because total size of images is greater: 0x%X", highAddress, totalSize);
810 return highAddress - totalSize;
811 }
812 else {
813 if ( (arch == CPU_TYPE_I386) || (arch == CPU_TYPE_POWERPC) ) {
814 // place dylibs below dyld
815 uint64_t topAddr = 0x8FE00000;
816 uint64_t totalSize = totalVMSize(arch, files);
817 if ( totalSize > topAddr )
818 throwf("total size of images (0x%X) does not fit below 0x8FE00000", totalSize);
819 return topAddr - totalSize;
820 }
821 else if ( arch == CPU_TYPE_POWERPC64 ) {
822 return 0x200000000ULL;
823 }
824 else if ( arch == CPU_TYPE_X86_64 ) {
825 return 0x200000000ULL;
826 }
827 else if ( arch == CPU_TYPE_ARM ) {
828 // place dylibs below dyld
829 uint64_t topAddr = 0x2FE00000;
830 uint64_t totalSize = totalVMSize(arch, files);
831 if ( totalSize > topAddr )
832 throwf("total size of images (0x%X) does not fit below 0x2FE00000", totalSize);
833 return topAddr - totalSize;
834 }
835 else
836 throw "unknown architecture";
837 }
838 }
839
840 static void usage()
841 {
842 fprintf(stderr, "rebase [-low_address] [-high_address] [-v] [-arch <arch>] files...\n");
843 }
844
845
846 int main(int argc, const char* argv[])
847 {
848 std::vector<fileInfo> files;
849 std::set<cpu_type_t> onlyArchs;
850 uint64_t lowAddress = 0;
851 uint64_t highAddress = 0;
852
853 try {
854 // parse command line options
855 char* endptr;
856 for(int i=1; i < argc; ++i) {
857 const char* arg = argv[i];
858 if ( arg[0] == '-' ) {
859 if ( strcmp(arg, "-v") == 0 ) {
860 verbose = true;
861 }
862 else if ( strcmp(arg, "-low_address") == 0 ) {
863 lowAddress = strtoull(argv[++i], &endptr, 16);
864 }
865 else if ( strcmp(arg, "-high_address") == 0 ) {
866 highAddress = strtoull(argv[++i], &endptr, 16);
867 }
868 else if ( strcmp(arg, "-arch") == 0 ) {
869 const char* arch = argv[++i];
870 if ( strcmp(arch, "ppc") == 0 )
871 onlyArchs.insert(CPU_TYPE_POWERPC);
872 else if ( strcmp(arch, "ppc64") == 0 )
873 onlyArchs.insert(CPU_TYPE_POWERPC64);
874 else if ( strcmp(arch, "i386") == 0 )
875 onlyArchs.insert(CPU_TYPE_I386);
876 else if ( strcmp(arch, "x86_64") == 0 )
877 onlyArchs.insert(CPU_TYPE_X86_64);
878 else if ( strcmp(arch, "arm") == 0 )
879 onlyArchs.insert(CPU_TYPE_ARM);
880 else if ( strcmp(arch, "armv6") == 0 )
881 onlyArchs.insert(CPU_TYPE_ARM);
882 else
883 throwf("unknown architecture %s", arch);
884 }
885 else {
886 usage();
887 throwf("unknown option: %s\n", arg);
888 }
889 }
890 else {
891 files.push_back(fileInfo(arg));
892 }
893 }
894
895 if ( files.size() == 0 )
896 throw "no files specified";
897
898 // use all architectures if no restrictions specified
899 if ( onlyArchs.size() == 0 ) {
900 onlyArchs.insert(CPU_TYPE_POWERPC);
901 onlyArchs.insert(CPU_TYPE_POWERPC64);
902 onlyArchs.insert(CPU_TYPE_I386);
903 onlyArchs.insert(CPU_TYPE_X86_64);
904 onlyArchs.insert(CPU_TYPE_ARM);
905 }
906
907 // scan files and collect sizes
908 for(std::vector<fileInfo>::iterator it=files.begin(); it != files.end(); ++it) {
909 setSizes(*it, onlyArchs);
910 }
911
912 // assign new base address for each arch
913 for(std::set<cpu_type_t>::iterator ait=onlyArchs.begin(); ait != onlyArchs.end(); ++ait) {
914 cpu_type_t arch = *ait;
915 uint64_t baseAddress = startAddress(arch, files, lowAddress, highAddress);
916 for(std::vector<fileInfo>::iterator fit=files.begin(); fit != files.end(); ++fit) {
917 fileInfo& fi = *fit;
918 for(std::vector<archInfo>::iterator fait=fi.archs.begin(); fait != fi.archs.end(); ++fait) {
919 if ( fait->arch == arch ) {
920 fait->newBase = baseAddress;
921 baseAddress += fait->vmSize;
922 baseAddress = (baseAddress + 4095) & (-4096); // page align
923 }
924 }
925 }
926 }
927
928 // rebase each file if it contains something rebaseable
929 for(std::vector<fileInfo>::iterator it=files.begin(); it != files.end(); ++it) {
930 fileInfo& fi = *it;
931 if ( fi.archs.size() > 0 )
932 rebase(fi);
933 }
934
935 }
936 catch (const char* msg) {
937 fprintf(stderr, "rebase failed: %s\n", msg);
938 return 1;
939 }
940
941 return 0;
942 }
943
944
945
mirror of ld64