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


3328e89e20ccac18842ae54ceece00968f14025e
[apple/dyld.git] / src / ImageLoaderMachO.cpp
1 /* -*- mode: C++; c-basic-offset: 4; tab-width: 4 -*-
2 *
3 * Copyright (c) 2004-2005 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 <string.h>
26 #include <fcntl.h>
27 #include <sys/types.h>
28 #include <sys/stat.h>
29 #include <sys/mman.h>
30 #include <mach/shared_memory_server.h>
31 #include <mach/mach.h>
32 #include <mach/thread_status.h>
33 #include <mach-o/loader.h>
34 #include <mach-o/reloc.h>
35 #include <mach-o/nlist.h>
36 #include <sys/sysctl.h>
37 #if __ppc__ || __ppc64__
38 #include <mach-o/ppc/reloc.h>
39 #endif
40
41 #include "ImageLoaderMachO.h"
42 #include "mach-o/dyld_gdb.h"
43
44 // no header for this yet, rdar://problem/3850825
45 extern "C" void sys_icache_invalidate(void *, size_t);
46
47 // optimize strcmp for ppc
48 #if __ppc__
49 #include <ppc_intrinsics.h>
50 #else
51 #define astrcmp(a,b) strcmp(a,b)
52 #endif
53
54 // relocation_info.r_length field has value 3 for 64-bit executables and value 2 for 32-bit executables
55 #if __LP64__
56 #define RELOC_SIZE 3
57 #define LC_SEGMENT_COMMAND LC_SEGMENT_64
58 #define LC_ROUTINES_COMMAND LC_ROUTINES_64
59 struct macho_header : public mach_header_64 {};
60 struct macho_segment_command : public segment_command_64 {};
61 struct macho_section : public section_64 {};
62 struct macho_nlist : public nlist_64 {};
63 struct macho_routines_command : public routines_command_64 {};
64 #else
65 #define RELOC_SIZE 2
66 #define LC_SEGMENT_COMMAND LC_SEGMENT
67 #define LC_ROUTINES_COMMAND LC_ROUTINES
68 struct macho_header : public mach_header {};
69 struct macho_segment_command : public segment_command {};
70 struct macho_section : public section {};
71 struct macho_nlist : public nlist {};
72 struct macho_routines_command : public routines_command {};
73 #endif
74
75
76 uint32_t ImageLoaderMachO::fgHintedBinaryTreeSearchs = 0;
77 uint32_t ImageLoaderMachO::fgUnhintedBinaryTreeSearchs = 0;
78
79
80 //#define LINKEDIT_USAGE_DEBUG 1
81
82 #if LINKEDIT_USAGE_DEBUG
83 #include <set>
84 static std::set<uintptr_t> sLinkEditPageBuckets;
85
86 namespace dyld {
87 extern ImageLoader* findImageContainingAddress(const void* addr);
88 };
89
90 static void noteAccessedLinkEditAddress(const void* addr)
91 {
92 uintptr_t page = ((uintptr_t)addr) & (-4096);
93 sLinkEditPageBuckets.insert(page);
94 fprintf(stderr, "dyld: accessing page 0x%08lX in __LINKEDIT of %s\n", page, dyld::findImageContainingAddress(addr)->getPath());
95 }
96 #endif
97
98 // only way to share initialization in C++
99 void ImageLoaderMachO::init()
100 {
101 fMachOData = NULL;
102 fLinkEditBase = NULL;
103 fSymbolTable = NULL;
104 fStrings = NULL;
105 fDynamicInfo = NULL;
106 fSlide = 0;
107 fIsSplitSeg = false;
108 fHasSubLibraries= false;
109 fHasSubUmbrella = false;
110 fDashInit = NULL;
111 fModInitSection = NULL;
112 fModTermSection = NULL;
113 fDATAdyld = NULL;
114 fImageNotifySection = NULL;
115 fTwoLevelHints = NULL;
116 fDylibID = NULL;
117 fReExportThruFramework = NULL;
118 fTextSegmentWithFixups = NULL;
119 }
120
121 // create image by copying an in-memory mach-o file
122 ImageLoaderMachO::ImageLoaderMachO(const char* moduleName, const struct mach_header* mh, uint64_t len, const LinkContext& context)
123 : ImageLoader(moduleName)
124 {
125 // clean slate
126 this->init();
127
128 // temporary use this buffer until TEXT is mapped in
129 fMachOData = (const uint8_t*)mh;
130
131 // create segments
132 this->instantiateSegments((const uint8_t*)mh);
133
134 // map segments
135 if ( mh->filetype != MH_EXECUTE )
136 ImageLoader::mapSegments((const void*)mh, len, context);
137
138 // get pointers to interesting things
139 this->parseLoadCmds();
140 }
141
142
143 // create image by mapping in a mach-o file
144 ImageLoaderMachO::ImageLoaderMachO(const char* path, int fd, const uint8_t firstPage[4096], uint64_t offsetInFat,
145 uint64_t lenInFat, const struct stat& info, const LinkContext& context)
146 : ImageLoader(path, offsetInFat, info)
147 {
148 // clean slate
149 this->init();
150
151 // read load commands
152 const unsigned int dataSize = sizeof(macho_header) + ((macho_header*)firstPage)->sizeofcmds;
153 uint8_t buffer[dataSize];
154 const uint8_t* fileData = firstPage;
155 if ( dataSize > 4096 ) {
156 // only read more if cmds take up more space than first page
157 fileData = buffer;
158 memcpy(buffer, firstPage, 4096);
159 pread(fd, &buffer[4096], dataSize-4096, offsetInFat+4096);
160 }
161
162 // temporary use this buffer until TEXT is mapped in
163 fMachOData = fileData;
164
165 // the meaning of many fields changes in split seg mach-o files
166 fIsSplitSeg = ((((macho_header*)fileData)->flags & MH_SPLIT_SEGS) != 0) && (((macho_header*)fileData)->filetype == MH_DYLIB);
167
168 // create segments
169 this->instantiateSegments(fileData);
170
171 // map segments, except for main executable which is already mapped in by kernel
172 if ( ((macho_header*)fileData)->filetype != MH_EXECUTE )
173 this->mapSegments(fd, offsetInFat, lenInFat, info.st_size, context);
174
175 // get pointers to interesting things
176 this->parseLoadCmds();
177 }
178
179
180
181
182 void ImageLoaderMachO::instantiateSegments(const uint8_t* fileData)
183 {
184 const uint32_t cmd_count = ((macho_header*)fileData)->ncmds;
185 const struct load_command* const cmds = (struct load_command*)&fileData[sizeof(macho_header)];
186
187 // construct Segment object for each LC_SEGMENT cmd and add to list
188 const struct load_command* cmd = cmds;
189 for (unsigned long i = 0; i < cmd_count; ++i) {
190 if ( cmd->cmd == LC_SEGMENT_COMMAND ) {
191 fSegments.push_back(new SegmentMachO((struct macho_segment_command*)cmd, this, fileData));
192 }
193 cmd = (const struct load_command*)(((char*)cmd)+cmd->cmdsize);
194 }
195 }
196
197
198
199 bool ImageLoaderMachO::segmentsMustSlideTogether() const
200 {
201 return true;
202 }
203
204 bool ImageLoaderMachO::segmentsCanSlide() const
205 {
206 const macho_header* mh = (macho_header*)fMachOData;
207 return ( (mh->filetype == MH_DYLIB) || (mh->filetype == MH_BUNDLE) );
208 }
209
210 bool ImageLoaderMachO::isBundle() const
211 {
212 const macho_header* mh = (macho_header*)fMachOData;
213 return ( mh->filetype == MH_BUNDLE );
214 }
215
216 bool ImageLoaderMachO::isDylib() const
217 {
218 const macho_header* mh = (macho_header*)fMachOData;
219 return ( mh->filetype == MH_DYLIB );
220 }
221
222 bool ImageLoaderMachO::forceFlat() const
223 {
224 const macho_header* mh = (macho_header*)fMachOData;
225 return ( (mh->flags & MH_FORCE_FLAT) != 0 );
226 }
227
228 bool ImageLoaderMachO::usesTwoLevelNameSpace() const
229 {
230 const macho_header* mh = (macho_header*)fMachOData;
231 return ( (mh->flags & MH_TWOLEVEL) != 0 );
232 }
233
234 bool ImageLoaderMachO::isPrebindable() const
235 {
236 const macho_header* mh = (macho_header*)fMachOData;
237 return ( (mh->flags & MH_PREBOUND) != 0 );
238 }
239
240 bool ImageLoaderMachO::hasCoalescedExports() const
241 {
242 const macho_header* mh = (macho_header*)fMachOData;
243 return ( (mh->flags & MH_WEAK_DEFINES) != 0 );
244 }
245
246 bool ImageLoaderMachO::needsCoalescing() const
247 {
248 const macho_header* mh = (macho_header*)fMachOData;
249 return ( (mh->flags & MH_BINDS_TO_WEAK) != 0 );
250 }
251
252 #if !__LP64__ // split segs not supported for 64-bits
253
254 #if 1 // hack until kernel headers and glue are in system
255 struct _shared_region_mapping_np {
256 mach_vm_address_t address;
257 mach_vm_size_t size;
258 mach_vm_offset_t file_offset;
259 vm_prot_t max_prot; /* read/write/execute/COW/ZF */
260 vm_prot_t init_prot; /* read/write/execute/COW/ZF */
261 };
262 struct _shared_region_range_np {
263 mach_vm_address_t address;
264 mach_vm_size_t size;
265 };
266
267 // Called by dyld.
268 // Requests the kernel to map a number of regions from the fd into the
269 // shared sections address range (0x90000000-0xAFFFFFFF).
270 // If shared_region_make_private_np() has not been called by this process,
271 // the file mapped in is seen in the address space of all processes that
272 // participate in using the shared region.
273 // If shared_region_make_private_np() _has_ been called by this process,
274 // the file mapped in is only seen by this process.
275 // If the slide parameter is not NULL and then regions cannot be mapped
276 // as requested, the kernel will try to map the file in at a different
277 // address in the shared region and return the distance slid.
278 // If the mapping requesting cannot be fulfilled, returns non-zero.
279 static int
280 _shared_region_map_file_np(
281 int fd, // file descriptor to map into shared region
282 unsigned int regionCount, // number of entres in array of regions
283 const _shared_region_mapping_np regions[], // the array of regions to map
284 uint64_t* slide) // the amount all regions were slid, NULL means don't attempt to slide
285 {
286 //fprintf(stderr, "%s(%i, %u, %8p, %8p)\n", __func__, fd, regionCount, regions, slide);
287 //for ( unsigned int i=0; i < regionCount; ++i) {
288 // fprintf(stderr, "\taddress=0x%08llX, size=0x%08llX\n", regions[i].address, regions[i].size);
289 //}
290 int r = syscall(299, fd, regionCount, regions, slide);
291 // if(0 != r)
292 // fprintf(stderr, "%s(%i, %u, %8p, %8p) errno=%i (%s)\n", __func__, fd, regionCount, regions, slide, errno, strerror(errno));
293 return r;
294 }
295 // Called by dyld if shared_region_map_file() fails.
296 // Requests the kernel to take this process out of using the shared region.
297 // The specified ranges are created as private copies from the shared region for this process.
298 static int
299 _shared_region_make_private_np(
300 unsigned int rangeCount, // number of entres in array of msrp_range
301 const _shared_region_range_np ranges[]) // the array of shared regions to make private
302 {
303 //fprintf(stderr, "%s(%u, %8p)\n", __func__, rangeCount, ranges);
304 int r = syscall(300, rangeCount, ranges);
305 // if(0 != r)
306 // fprintf(stderr, "%s(%u, %8p) errno=%i (%s)\n", __func__, rangeCount, ranges, errno, strerror(errno));
307 return r;
308 }
309 #define KERN_SHREG_PRIVATIZABLE 54
310 #endif // hack until kernel headers and glue are in system
311
312 static uintptr_t sNextAltLoadAddress
313 #if __ppc_
314 = 0xC0000000;
315 #else
316 = 0;
317 #endif
318
319 static int
320 _shared_region_map_file_with_mmap(
321 int fd, // file descriptor to map into shared region
322 unsigned int regionCount, // number of entres in array of regions
323 const _shared_region_mapping_np regions[]) // the array of regions to map
324 {
325 // map in each region
326 for(unsigned int i=0; i < regionCount; ++i) {
327 void* mmapAddress = (void*)(uintptr_t)(regions[i].address);
328 size_t size = regions[i].size;
329 if ( (regions[i].init_prot & VM_PROT_ZF) != 0 ) {
330 // do nothing already vm_allocate() which zero fills
331 }
332 else {
333 int protection = 0;
334 if ( regions[i].init_prot & VM_PROT_EXECUTE )
335 protection |= PROT_EXEC;
336 if ( regions[i].init_prot & VM_PROT_READ )
337 protection |= PROT_READ;
338 if ( regions[i].init_prot & VM_PROT_WRITE )
339 protection |= PROT_WRITE;
340 off_t offset = regions[i].file_offset;
341 //fprintf(stderr, "mmap(%p, 0x%08lX, block=0x%08X, %s\n", mmapAddress, size, biggestDiff, fPath);
342 mmapAddress = mmap(mmapAddress, size, protection, MAP_FILE | MAP_FIXED | MAP_PRIVATE, fd, offset);
343 if ( mmapAddress == ((void*)(-1)) )
344 throw "mmap error";
345 }
346 }
347
348 return 0;
349 }
350
351
352 static
353 bool
354 hasSharedRegionMapFile(void)
355 {
356 int mib[CTL_MAXNAME];
357 int value = 0;
358 size_t size;
359
360 mib[0] = CTL_KERN;
361 mib[1] = KERN_SHREG_PRIVATIZABLE;
362 size = sizeof (int);
363 if (sysctl(mib, 2, &value, &size, NULL, 0) != 0) {
364 value = 0;
365 }
366
367 return 0 != value;
368 }
369
370 int
371 ImageLoaderMachO::sharedRegionMapFilePrivateOutside(int fd,
372 uint64_t offsetInFat,
373 uint64_t lenInFat,
374 uint64_t fileLen,
375 const LinkContext& context)
376 {
377 const unsigned int segmentCount = fSegments.size();
378 const unsigned int extraZeroFillEntries = getExtraZeroFillEntriesCount();
379 const unsigned int regionCount = segmentCount+extraZeroFillEntries;
380 _shared_region_mapping_np regions[regionCount];
381 initMappingTable(offsetInFat, regions);
382 int r = -1;
383 // find space somewhere to allocate split seg
384 bool foundRoom = false;
385 vm_size_t biggestDiff = 0;
386 while ( ! foundRoom ) {
387 foundRoom = true;
388 for(unsigned int i=0; i < regionCount; ++i) {
389 vm_address_t addr = sNextAltLoadAddress + regions[i].address - regions[0].address;
390 vm_size_t size = regions[i].size ;
391 r = vm_allocate(mach_task_self(), &addr, size, false /*only this range*/);
392 if ( 0 != r ) {
393 // no room here, deallocate what has succeeded so far
394 for(unsigned int j=0; j < i; ++j) {
395 vm_address_t addr = sNextAltLoadAddress + regions[j].address - regions[0].address;
396 vm_size_t size = regions[j].size ;
397 (void)vm_deallocate(mach_task_self(), addr, size);
398 }
399 sNextAltLoadAddress += 0x00100000; // skip ahead 1MB and try again
400 if ( (sNextAltLoadAddress & 0xF0000000) == 0x90000000 )
401 throw "can't map split seg anywhere";
402 foundRoom = false;
403 break;
404 }
405 vm_size_t high = (regions[i].address + size - regions[0].address) & 0x0FFFFFFF;
406 if ( high > biggestDiff )
407 biggestDiff = high;
408 }
409 }
410
411 // map in each region
412 uintptr_t slide = sNextAltLoadAddress - regions[0].address;
413 this->setSlide(slide);
414 for(unsigned int i=0; i < regionCount; ++i) {
415 if ( (regions[i].init_prot & VM_PROT_ZF) != 0 ) {
416 // do nothing vm_allocate() zero-fills by default
417 }
418 else {
419 void* mmapAddress = (void*)(uintptr_t)(regions[i].address + slide);
420 size_t size = regions[i].size;
421 int protection = 0;
422 if ( regions[i].init_prot & VM_PROT_EXECUTE )
423 protection |= PROT_EXEC;
424 if ( regions[i].init_prot & VM_PROT_READ )
425 protection |= PROT_READ;
426 if ( regions[i].init_prot & VM_PROT_WRITE )
427 protection |= PROT_WRITE;
428 off_t offset = regions[i].file_offset;
429 //fprintf(stderr, "mmap(%p, 0x%08lX, block=0x%08X, %s\n", mmapAddress, size, biggestDiff, fPath);
430 mmapAddress = mmap(mmapAddress, size, protection, MAP_FILE | MAP_FIXED | MAP_PRIVATE, fd, offset);
431 if ( mmapAddress == ((void*)(-1)) )
432 throw "mmap error";
433 }
434 }
435 // set so next maps right after this one
436 sNextAltLoadAddress += biggestDiff;
437 sNextAltLoadAddress = (sNextAltLoadAddress + 4095) & (-4096);
438
439 // logging
440 if ( context.verboseMapping ) {
441 fprintf(stderr, "dyld: Mapping split-seg outside shared region, slid by 0x%08lX %s\n", this->fSlide, this->getPath());
442 for(unsigned int segIndex=0,entryIndex=0; segIndex < segmentCount; ++segIndex, ++entryIndex){
443 Segment* seg = fSegments[segIndex];
444 const _shared_region_mapping_np* entry = &regions[entryIndex];
445 if ( (entry->init_prot & VM_PROT_ZF) == 0 )
446 fprintf(stderr, "%18s at 0x%08lX->0x%08lX\n",
447 seg->getName(), seg->getActualLoadAddress(), seg->getActualLoadAddress()+seg->getFileSize()-1);
448 if ( entryIndex < (regionCount-1) ) {
449 const _shared_region_mapping_np* nextEntry = &regions[entryIndex+1];
450 if ( (nextEntry->init_prot & VM_PROT_ZF) != 0 ) {
451 uint64_t segOffset = nextEntry->address - entry->address;
452 fprintf(stderr, "%18s at 0x%08lX->0x%08lX (zerofill)\n",
453 seg->getName(), (uintptr_t)(seg->getActualLoadAddress() + segOffset), (uintptr_t)(seg->getActualLoadAddress() + segOffset + nextEntry->size - 1));
454 ++entryIndex;
455 }
456 }
457 }
458 }
459
460 return r;
461 }
462
463
464 void ImageLoaderMachO::mapSegments(int fd, uint64_t offsetInFat, uint64_t lenInFat, uint64_t fileLen, const LinkContext& context)
465 {
466 enum SharedRegionState
467 {
468 kSharedRegionStartState = 0,
469 kSharedRegionLoadFileState,
470 kSharedRegionMapFileState,
471 kSharedRegionMapFilePrivateState,
472 kSharedRegionMapFilePrivateMMapState,
473 kSharedRegionMapFilePrivateOutsideState,
474 };
475 static SharedRegionState sSharedRegionState = kSharedRegionStartState;
476
477 // non-split segment libraries handled by super class
478 if ( !fIsSplitSeg )
479 return ImageLoader::mapSegments(fd, offsetInFat, lenInFat, fileLen, context);
480
481 if ( kSharedRegionStartState == sSharedRegionState ) {
482 if ( hasSharedRegionMapFile() ) {
483 if ( context.slideAndPackDylibs ) {
484 sharedRegionMakePrivate(context);
485 // remove underlying submap and block out 0x90000000 to 0xAFFFFFFF
486 vm_address_t addr = (vm_address_t)0x90000000;
487 vm_deallocate(mach_task_self(), addr, 0x20000000);
488 vm_allocate(mach_task_self(), &addr, 0x20000000, false);
489 sSharedRegionState = kSharedRegionMapFilePrivateMMapState;
490 }
491 else if ( context.sharedRegionMode == kUsePrivateSharedRegion ) {
492 sharedRegionMakePrivate(context);
493 sSharedRegionState = kSharedRegionMapFilePrivateState;
494 }
495 else if ( context.sharedRegionMode == kDontUseSharedRegion ) {
496 sSharedRegionState = kSharedRegionMapFilePrivateOutsideState;
497 }
498 else {
499 sSharedRegionState = kSharedRegionMapFileState;
500 }
501 }
502 else {
503 sSharedRegionState = kSharedRegionLoadFileState;
504 }
505 }
506
507 if ( kSharedRegionLoadFileState == sSharedRegionState ) {
508 if ( 0 != sharedRegionLoadFile(fd, offsetInFat, lenInFat, fileLen, context) ) {
509 sSharedRegionState = kSharedRegionMapFilePrivateOutsideState;
510 }
511 }
512 else
513 if ( kSharedRegionMapFileState == sSharedRegionState ) {
514 if ( 0 != sharedRegionMapFile(fd, offsetInFat, lenInFat, fileLen, context) ) {
515 sharedRegionMakePrivate(context);
516 sSharedRegionState = kSharedRegionMapFilePrivateState;
517 }
518 }
519
520 if ( (kSharedRegionMapFilePrivateState == sSharedRegionState) || (kSharedRegionMapFilePrivateMMapState == sSharedRegionState) ) {
521 if ( 0 != sharedRegionMapFilePrivate(fd, offsetInFat, lenInFat, fileLen, context, (kSharedRegionMapFilePrivateMMapState == sSharedRegionState)) ) {
522 sSharedRegionState = kSharedRegionMapFilePrivateOutsideState;
523 }
524 }
525
526 if ( kSharedRegionMapFilePrivateOutsideState == sSharedRegionState ) {
527 if ( 0 != sharedRegionMapFilePrivateOutside(fd, offsetInFat, lenInFat, fileLen, context) ) {
528 throw "mapping error";
529 }
530 }
531 }
532
533 unsigned int
534 ImageLoaderMachO::getExtraZeroFillEntriesCount()
535 {
536 // calculate mapping entries
537 const unsigned int segmentCount = fSegments.size();
538 unsigned int extraZeroFillEntries = 0;
539 for(unsigned int i=0; i < segmentCount; ++i){
540 Segment* seg = fSegments[i];
541 if ( seg->hasTrailingZeroFill() )
542 ++extraZeroFillEntries;
543 }
544
545 return extraZeroFillEntries;
546 }
547
548 void
549 ImageLoaderMachO::initMappingTable(uint64_t offsetInFat,
550 _shared_region_mapping_np *mappingTable)
551 {
552 unsigned int segmentCount = fSegments.size();
553 for(unsigned int segIndex=0,entryIndex=0; segIndex < segmentCount; ++segIndex, ++entryIndex){
554 Segment* seg = fSegments[segIndex];
555 _shared_region_mapping_np* entry = &mappingTable[entryIndex];
556 entry->address = seg->getActualLoadAddress();
557 entry->size = seg->getFileSize();
558 entry->file_offset = seg->getFileOffset() + offsetInFat;
559 entry->init_prot = VM_PROT_NONE;
560 if ( !seg->unaccessible() ) {
561 if ( seg->executable() )
562 entry->init_prot |= VM_PROT_EXECUTE;
563 if ( seg->readable() )
564 entry->init_prot |= VM_PROT_READ;
565 if ( seg->writeable() )
566 entry->init_prot |= VM_PROT_WRITE | VM_PROT_COW;
567 }
568 entry->max_prot = entry->init_prot;
569 if ( seg->hasTrailingZeroFill() ) {
570 _shared_region_mapping_np* zfentry = &mappingTable[++entryIndex];
571 zfentry->address = entry->address + seg->getFileSize();
572 zfentry->size = seg->getSize() - seg->getFileSize();
573 zfentry->file_offset = 0;
574 zfentry->init_prot = entry->init_prot | VM_PROT_COW | VM_PROT_ZF;
575 zfentry->max_prot = zfentry->init_prot;
576 }
577 }
578 }
579
580 int
581 ImageLoaderMachO::sharedRegionMakePrivate(const LinkContext& context)
582 {
583 if ( context.verboseMapping )
584 fprintf(stderr, "dyld: making shared regions private\n");
585
586 // shared mapping failed, so make private copy of shared region and try mapping private
587 RegionsVector allRegions;
588 context.getAllMappedRegions(allRegions);
589 std::vector<_shared_region_range_np> splitSegRegions;
590 const unsigned int allRegiontCount = allRegions.size();
591 for(unsigned int i=0; i < allRegiontCount; ++i){
592 MappedRegion region = allRegions[i];
593 uint8_t highByte = region.address >> 28;
594 if ( (highByte == 9) || (highByte == 0xA) ) {
595 _shared_region_range_np splitRegion;
596 splitRegion.address = region.address;
597 splitRegion.size = region.size;
598 splitSegRegions.push_back(splitRegion);
599 }
600 }
601 int result = _shared_region_make_private_np(splitSegRegions.size(), &splitSegRegions[0]);
602 // notify gdb or other lurkers that this process is no longer using the shared region
603 dyld_all_image_infos.processDetachedFromSharedRegion = true;
604 return result;
605 }
606
607 int
608 ImageLoaderMachO::sharedRegionMapFile(int fd,
609 uint64_t offsetInFat,
610 uint64_t lenInFat,
611 uint64_t fileLen,
612 const LinkContext& context)
613 {
614 // build table of segments to map
615 const unsigned int segmentCount = fSegments.size();
616 const unsigned int extraZeroFillEntries = getExtraZeroFillEntriesCount();
617 const unsigned int mappingTableCount = segmentCount+extraZeroFillEntries;
618 _shared_region_mapping_np mappingTable[mappingTableCount];
619 initMappingTable(offsetInFat, mappingTable);
620 // uint64_t slide;
621 uint64_t *slidep = NULL;
622
623 // try to map it in shared
624 int r = _shared_region_map_file_np(fd, mappingTableCount, mappingTable, slidep);
625 if ( 0 == r ) {
626 if(NULL != slidep && 0 != *slidep) {
627 // update with actual load addresses
628 }
629 if ( context.verboseMapping ) {
630 fprintf(stderr, "dyld: Mapping split-seg shared %s\n", this->getPath());
631 for(unsigned int segIndex=0,entryIndex=0; segIndex < segmentCount; ++segIndex, ++entryIndex){
632 Segment* seg = fSegments[segIndex];
633 const _shared_region_mapping_np* entry = &mappingTable[entryIndex];
634 if ( (entry->init_prot & VM_PROT_ZF) == 0 )
635 fprintf(stderr, "%18s at 0x%08lX->0x%08lX\n",
636 seg->getName(), seg->getActualLoadAddress(), seg->getActualLoadAddress()+seg->getFileSize()-1);
637 if ( entryIndex < (mappingTableCount-1) ) {
638 const _shared_region_mapping_np* nextEntry = &mappingTable[entryIndex+1];
639 if ( (nextEntry->init_prot & VM_PROT_ZF) != 0 ) {
640 uint64_t segOffset = nextEntry->address - entry->address;
641 fprintf(stderr, "%18s at 0x%08lX->0x%08lX\n",
642 seg->getName(), (uintptr_t)(seg->getActualLoadAddress() + segOffset), (uintptr_t)(seg->getActualLoadAddress() + segOffset + nextEntry->size - 1));
643 ++entryIndex;
644 }
645 }
646 }
647 }
648 }
649 return r;
650 }
651
652
653 int
654 ImageLoaderMachO::sharedRegionMapFilePrivate(int fd,
655 uint64_t offsetInFat,
656 uint64_t lenInFat,
657 uint64_t fileLen,
658 const LinkContext& context,
659 bool usemmap)
660 {
661 const unsigned int segmentCount = fSegments.size();
662
663 // adjust base address of segments to pack next to last dylib
664 if ( context.slideAndPackDylibs ) {
665 uintptr_t lowestReadOnly = (uintptr_t)(-1);
666 uintptr_t lowestWritable = (uintptr_t)(-1);
667 for(unsigned int segIndex=0; segIndex < segmentCount; ++segIndex){
668 Segment* seg = fSegments[segIndex];
669 uintptr_t segEnd = seg->getActualLoadAddress();
670 if ( seg->writeable() ) {
671 if ( segEnd < lowestWritable )
672 lowestWritable = segEnd;
673 }
674 else {
675 if ( segEnd < lowestReadOnly )
676 lowestReadOnly = segEnd;
677 }
678 }
679 uintptr_t baseAddress;
680 if ( lowestWritable - 256*1024*1024 < lowestReadOnly )
681 baseAddress = lowestWritable - 256*1024*1024;
682 else
683 baseAddress = lowestReadOnly;
684 // record that we want dylb slid to fgNextSplitSegAddress
685 this->setSlide(fgNextSplitSegAddress - baseAddress);
686 }
687
688 // build table of segments to map
689 const unsigned int extraZeroFillEntries = getExtraZeroFillEntriesCount();
690 const unsigned int mappingTableCount = segmentCount+extraZeroFillEntries;
691 _shared_region_mapping_np mappingTable[mappingTableCount];
692 initMappingTable(offsetInFat, mappingTable);
693 uint64_t slide = 0;
694
695 // try map it in privately (don't allow sliding if we pre-calculated the load address to pack dylibs)
696 int r;
697 if ( usemmap )
698 r = _shared_region_map_file_with_mmap(fd, mappingTableCount, mappingTable);
699 else
700 r = _shared_region_map_file_np(fd, mappingTableCount, mappingTable, context.slideAndPackDylibs ? NULL : &slide);
701 if ( 0 == r ) {
702 if ( 0 != slide ) {
703 slide = (slide) & (-4096); // round down to page boundary
704 this->setSlide(slide);
705 }
706 if ( context.verboseMapping ) {
707 if ( slide == 0 )
708 fprintf(stderr, "dyld: Mapping split-seg un-shared %s\n", this->getPath());
709 else
710 fprintf(stderr, "dyld: Mapping split-seg un-shared slid by 0x%08llX %s\n", slide, this->getPath());
711 for(unsigned int segIndex=0,entryIndex=0; segIndex < segmentCount; ++segIndex, ++entryIndex){
712 Segment* seg = fSegments[segIndex];
713 const _shared_region_mapping_np* entry = &mappingTable[entryIndex];
714 if ( (entry->init_prot & VM_PROT_ZF) == 0 )
715 fprintf(stderr, "%18s at 0x%08lX->0x%08lX\n",
716 seg->getName(), seg->getActualLoadAddress(), seg->getActualLoadAddress()+seg->getFileSize()-1);
717 if ( entryIndex < (mappingTableCount-1) ) {
718 const _shared_region_mapping_np* nextEntry = &mappingTable[entryIndex+1];
719 if ( (nextEntry->init_prot & VM_PROT_ZF) != 0 ) {
720 uint64_t segOffset = nextEntry->address - entry->address;
721 fprintf(stderr, "%18s at 0x%08lX->0x%08lX (zerofill)\n",
722 seg->getName(), (uintptr_t)(seg->getActualLoadAddress() + segOffset), (uintptr_t)(seg->getActualLoadAddress() + segOffset + nextEntry->size - 1));
723 ++entryIndex;
724 }
725 }
726 }
727 }
728 if ( context.slideAndPackDylibs ) {
729 // calculate where next split-seg dylib can load
730 uintptr_t largestReadOnly = 0;
731 uintptr_t largestWritable = 0;
732 for (unsigned int segIndex=0; segIndex < segmentCount; ++segIndex) {
733 Segment* seg = fSegments[segIndex];
734 uintptr_t segEnd = seg->getActualLoadAddress()+seg->getSize();
735 segEnd = (segEnd+4095) & (-4096); // page align
736 if ( seg->writeable() ) {
737 if ( segEnd > largestWritable )
738 largestWritable = segEnd;
739 }
740 else {
741 if ( segEnd > largestReadOnly )
742 largestReadOnly = segEnd;
743 }
744 }
745 if ( largestWritable - 256*1024*1024 > largestReadOnly )
746 fgNextSplitSegAddress = largestWritable - 256*1024*1024;
747 else
748 fgNextSplitSegAddress = largestReadOnly;
749 }
750 }
751 if ( context.slideAndPackDylibs && (r != 0) )
752 throwf("can't rebase split-seg dylib %s because shared_region_map_file_np() returned %d", this->getPath(), r);
753
754 return r;
755 }
756
757
758 int
759 ImageLoaderMachO::sharedRegionLoadFile(int fd, uint64_t offsetInFat, uint64_t lenInFat, uint64_t fileLen, const LinkContext& context)
760 {
761
762 // map in split segment file at random address, then tell kernel to share it
763 void* loadAddress = 0;
764 loadAddress = mmap(NULL, fileLen, PROT_READ, MAP_FILE, fd, 0);
765 if ( loadAddress == ((void*)(-1)) )
766 throw "mmap error";
767
768 // calculate mapping entries
769 const unsigned int segmentCount = fSegments.size();
770 unsigned int extraZeroFillEntries = getExtraZeroFillEntriesCount();
771
772 // build table of segments to map
773 const unsigned int mappingTableCount = segmentCount+extraZeroFillEntries;
774 const uintptr_t baseAddress = fSegments[0]->getPreferredLoadAddress();
775 sf_mapping mappingTable[mappingTableCount];
776 initMappingTable(offsetInFat, mappingTable, baseAddress);
777
778
779 // use load_shared_file() to map all segments at once
780 int flags = 0; // might need to set NEW_LOCAL_SHARED_REGIONS on first use
781 static bool firstTime = true;
782 if ( firstTime ) {
783 // when NEW_LOCAL_SHARED_REGIONS bit is set, this process will get is own shared region
784 // this is used by Xcode to prevent development libraries from polluting the global shared segment
785 if ( context.sharedRegionMode == kUsePrivateSharedRegion )
786 flags |= NEW_LOCAL_SHARED_REGIONS;
787 firstTime = false;
788 }
789
790 caddr_t base_address = (caddr_t)baseAddress;
791 kern_return_t r;
792 r = load_shared_file( (char*)fPath, // path of file to map shared
793 (char*)loadAddress, // beginning of local copy of sharable pages in file
794 fileLen, // end of shareable pages in file
795 &base_address, // beginning of address range to map
796 mappingTableCount, // number of entres in array of sf_mapping
797 mappingTable, // the array of sf_mapping
798 &flags); // in/out flags
799 if ( 0 != r ) {
800 // try again but tell kernel it is ok to slide
801 flags |= ALTERNATE_LOAD_SITE;
802 r = load_shared_file((char*)fPath,(char*)loadAddress, fileLen, &base_address,
803 mappingTableCount, mappingTable, &flags);
804 }
805
806 // unmap file from random address now that they are (hopefully) mapped into the shared region
807 munmap(loadAddress, fileLen);
808
809 if ( 0 == r ) {
810 if ( base_address != (caddr_t)baseAddress )
811 this->setSlide((uintptr_t)base_address - baseAddress);
812 if ( context.verboseMapping ) {
813 if ( base_address != (caddr_t)baseAddress )
814 fprintf(stderr, "dyld: Mapping split-seg load_shared_alt_region %s\n", this->getPath());
815 else
816 fprintf(stderr, "dyld: Mapping split-seg load_shared %s\n", this->getPath());
817 for(unsigned int segIndex=0,entryIndex=0; segIndex < segmentCount; ++segIndex, ++entryIndex){
818 Segment* seg = fSegments[segIndex];
819 const sf_mapping* entry = &mappingTable[entryIndex];
820 if ( (entry->protection & VM_PROT_ZF) == 0 )
821 fprintf(stderr, "%18s at 0x%08lX->0x%08lX\n",
822 seg->getName(), seg->getActualLoadAddress(), seg->getActualLoadAddress()+seg->getFileSize()-1);
823 if ( entryIndex < (mappingTableCount-1) ) {
824 const sf_mapping* nextEntry = &mappingTable[entryIndex+1];
825 if ( (nextEntry->protection & VM_PROT_ZF) != 0 ) {
826 fprintf(stderr, "%18s at 0x%08lX->0x%08lX\n",
827 seg->getName(), (uintptr_t)(nextEntry->mapping_offset + base_address), (uintptr_t)(nextEntry->mapping_offset + base_address + nextEntry->size - 1));
828 ++entryIndex;
829 }
830 }
831 }
832 }
833 }
834 return r;
835 }
836 void
837 ImageLoaderMachO::initMappingTable(uint64_t offsetInFat,
838 sf_mapping *mappingTable,
839 uintptr_t baseAddress)
840 {
841 unsigned int segmentCount = fSegments.size();
842 for(unsigned int segIndex=0,entryIndex=0; segIndex < segmentCount; ++segIndex, ++entryIndex){
843 Segment* seg = fSegments[segIndex];
844 sf_mapping* entry = &mappingTable[entryIndex];
845 entry->mapping_offset = seg->getPreferredLoadAddress() - baseAddress;
846 entry->size = seg->getFileSize();
847 entry->file_offset = seg->getFileOffset() + offsetInFat;
848 entry->protection = VM_PROT_NONE;
849 if ( !seg->unaccessible() ) {
850 if ( seg->executable() )
851 entry->protection |= VM_PROT_EXECUTE;
852 if ( seg->readable() )
853 entry->protection |= VM_PROT_READ;
854 if ( seg->writeable() )
855 entry->protection |= VM_PROT_WRITE | VM_PROT_COW;
856 }
857
858 entry->cksum = 0;
859 if ( seg->hasTrailingZeroFill() ) {
860 sf_mapping* zfentry = &mappingTable[++entryIndex];
861 zfentry->mapping_offset = entry->mapping_offset + seg->getFileSize();
862 zfentry->size = seg->getSize() - seg->getFileSize();
863 zfentry->file_offset = 0;
864 zfentry->protection = entry->protection | VM_PROT_COW | VM_PROT_ZF;
865 zfentry->cksum = 0;
866 }
867 }
868 }
869
870 #endif // !__LP64__ split segs not supported for 64-bits
871
872
873 void ImageLoaderMachO::setSlide(intptr_t slide)
874 {
875 fSlide = slide;
876 }
877
878 void ImageLoaderMachO::parseLoadCmds()
879 {
880 // now that segments are mapped in, get real fMachOData, fLinkEditBase, and fSlide
881 const unsigned int segmentCount = fSegments.size();
882 for(unsigned int i=0; i < segmentCount; ++i){
883 Segment* seg = fSegments[i];
884 // set up pointer to __LINKEDIT segment
885 if ( strcmp(seg->getName(),"__LINKEDIT") == 0 )
886 fLinkEditBase = (uint8_t*)(seg->getActualLoadAddress() - seg->getFileOffset());
887 // __TEXT segment always starts at beginning of file and contains mach_header and load commands
888 if ( strcmp(seg->getName(),"__TEXT") == 0 ) {
889 if ( seg->hasFixUps() )
890 fTextSegmentWithFixups = (SegmentMachO*)seg;
891 }
892 // some segment always starts at beginning of file and contains mach_header and load commands
893 if ( (seg->getFileOffset() == 0) && (seg->getFileSize() != 0) ) {
894 fMachOData = (uint8_t*)(seg->getActualLoadAddress());
895 }
896 }
897
898 // walk load commands (mapped in at start of __TEXT segment)
899 const uint32_t cmd_count = ((macho_header*)fMachOData)->ncmds;
900 const struct load_command* const cmds = (struct load_command*)&fMachOData[sizeof(macho_header)];
901 const struct load_command* cmd = cmds;
902 for (uint32_t i = 0; i < cmd_count; ++i) {
903 switch (cmd->cmd) {
904 case LC_SYMTAB:
905 {
906 const struct symtab_command* symtab = (struct symtab_command*)cmd;
907 fStrings = (const char*)&fLinkEditBase[symtab->stroff];
908 fSymbolTable = (struct macho_nlist*)(&fLinkEditBase[symtab->symoff]);
909 }
910 break;
911 case LC_DYSYMTAB:
912 fDynamicInfo = (struct dysymtab_command*)cmd;
913 break;
914 case LC_SUB_UMBRELLA:
915 fHasSubUmbrella = true;
916 break;
917 case LC_SUB_FRAMEWORK:
918 {
919 const struct sub_framework_command* subf = (struct sub_framework_command*)cmd;
920 fReExportThruFramework = (char*)cmd + subf->umbrella.offset;
921 }
922 break;
923 case LC_SUB_LIBRARY:
924 fHasSubLibraries = true;
925 break;
926 case LC_ROUTINES_COMMAND:
927 fDashInit = (struct macho_routines_command*)cmd;
928 break;
929 case LC_SEGMENT_COMMAND:
930 {
931 const struct macho_segment_command* seg = (struct macho_segment_command*)cmd;
932 const bool isDataSeg = (strcmp(seg->segname, "__DATA") == 0);
933 const struct macho_section* const sectionsStart = (struct macho_section*)((char*)seg + sizeof(struct macho_segment_command));
934 const struct macho_section* const sectionsEnd = &sectionsStart[seg->nsects];
935 for (const struct macho_section* sect=sectionsStart; sect < sectionsEnd; ++sect) {
936 const uint8_t type = sect->flags & SECTION_TYPE;
937 if ( type == S_MOD_INIT_FUNC_POINTERS )
938 fModInitSection = sect;
939 else if ( type == S_MOD_TERM_FUNC_POINTERS )
940 fModTermSection = sect;
941 else if ( isDataSeg && (strcmp(sect->sectname, "__dyld") == 0) ) {
942 fDATAdyld = sect;
943 }
944 else if ( isDataSeg && (strcmp(sect->sectname, "__image_notify") == 0) )
945 fImageNotifySection = sect;
946 }
947 }
948 break;
949 case LC_TWOLEVEL_HINTS:
950 fTwoLevelHints = (struct twolevel_hints_command*)cmd;
951 break;
952 case LC_ID_DYLIB:
953 {
954 fDylibID = (struct dylib_command*)cmd;
955 }
956 break;
957 case LC_LOAD_WEAK_DYLIB:
958 // do nothing, just prevent LC_REQ_DYLD exception from occuring
959 break;
960 default:
961 if ( (cmd->cmd & LC_REQ_DYLD) != 0 )
962 throwf("unknown required load command 0x%08X", cmd->cmd);
963 }
964 cmd = (const struct load_command*)(((char*)cmd)+cmd->cmdsize);
965 }
966 }
967
968
969
970
971 const char* ImageLoaderMachO::getInstallPath() const
972 {
973 if ( fDylibID != NULL ) {
974 return (char*)fDylibID + fDylibID->dylib.name.offset;
975 }
976 return NULL;
977 }
978
979 // test if this image is re-exported through parent (the image that loaded this one)
980 bool ImageLoaderMachO::isSubframeworkOf(const LinkContext& context, const ImageLoader* parent) const
981 {
982 if ( fReExportThruFramework != NULL ) {
983 // need to match LC_SUB_FRAMEWORK string against the leaf name of the install location of parent...
984 const char* parentInstallPath = parent->getInstallPath();
985 if ( parentInstallPath != NULL ) {
986 const char* lastSlash = strrchr(parentInstallPath, '/');
987 if ( lastSlash != NULL ) {
988 if ( strcmp(&lastSlash[1], fReExportThruFramework) == 0 )
989 return true;
990 if ( context.imageSuffix != NULL ) {
991 // when DYLD_IMAGE_SUFFIX is used, lastSlash string needs imageSuffix removed from end
992 char reexportAndSuffix[strlen(context.imageSuffix)+strlen(fReExportThruFramework)+1];
993 strcpy(reexportAndSuffix, fReExportThruFramework);
994 strcat(reexportAndSuffix, context.imageSuffix);
995 if ( strcmp(&lastSlash[1], reexportAndSuffix) == 0 )
996 return true;
997 }
998 }
999 }
1000 }
1001 return false;
1002 }
1003
1004 // test if child is re-exported
1005 bool ImageLoaderMachO::hasSubLibrary(const LinkContext& context, const ImageLoader* child) const
1006 {
1007 if ( fHasSubLibraries ) {
1008 // need to match LC_SUB_LIBRARY string against the leaf name (without extension) of the install location of child...
1009 const char* childInstallPath = child->getInstallPath();
1010 if ( childInstallPath != NULL ) {
1011 const char* lastSlash = strrchr(childInstallPath, '/');
1012 if ( lastSlash != NULL ) {
1013 const char* firstDot = strchr(lastSlash, '.');
1014 int len;
1015 if ( firstDot == NULL )
1016 len = strlen(lastSlash);
1017 else
1018 len = firstDot-lastSlash-1;
1019 char childLeafName[len+1];
1020 strncpy(childLeafName, &lastSlash[1], len);
1021 childLeafName[len] = '\0';
1022 const uint32_t cmd_count = ((macho_header*)fMachOData)->ncmds;
1023 const struct load_command* const cmds = (struct load_command*)&fMachOData[sizeof(macho_header)];
1024 const struct load_command* cmd = cmds;
1025 for (uint32_t i = 0; i < cmd_count; ++i) {
1026 switch (cmd->cmd) {
1027 case LC_SUB_LIBRARY:
1028 {
1029 const struct sub_library_command* lib = (struct sub_library_command*)cmd;
1030 const char* aSubLibName = (char*)cmd + lib->sub_library.offset;
1031 if ( strcmp(aSubLibName, childLeafName) == 0 )
1032 return true;
1033 if ( context.imageSuffix != NULL ) {
1034 // when DYLD_IMAGE_SUFFIX is used, childLeafName string needs imageSuffix removed from end
1035 char aSubLibNameAndSuffix[strlen(context.imageSuffix)+strlen(aSubLibName)+1];
1036 strcpy(aSubLibNameAndSuffix, aSubLibName);
1037 strcat(aSubLibNameAndSuffix, context.imageSuffix);
1038 if ( strcmp(aSubLibNameAndSuffix, childLeafName) == 0 )
1039 return true;
1040 }
1041 }
1042 break;
1043 }
1044 cmd = (const struct load_command*)(((char*)cmd)+cmd->cmdsize);
1045 }
1046 }
1047 }
1048 }
1049 if ( fHasSubUmbrella ) {
1050 // need to match LC_SUB_UMBRELLA string against the leaf name of install location of child...
1051 const char* childInstallPath = child->getInstallPath();
1052 if ( childInstallPath != NULL ) {
1053 const char* lastSlash = strrchr(childInstallPath, '/');
1054 if ( lastSlash != NULL ) {
1055 const uint32_t cmd_count = ((macho_header*)fMachOData)->ncmds;
1056 const struct load_command* const cmds = (struct load_command*)&fMachOData[sizeof(macho_header)];
1057 const struct load_command* cmd = cmds;
1058 for (uint32_t i = 0; i < cmd_count; ++i) {
1059 switch (cmd->cmd) {
1060 case LC_SUB_UMBRELLA:
1061 {
1062 const struct sub_umbrella_command* um = (struct sub_umbrella_command*)cmd;
1063 const char* aSubUmbrellaName = (char*)cmd + um->sub_umbrella.offset;
1064 if ( strcmp(aSubUmbrellaName, &lastSlash[1]) == 0 )
1065 return true;
1066 if ( context.imageSuffix != NULL ) {
1067 // when DYLD_IMAGE_SUFFIX is used, lastSlash string needs imageSuffix removed from end
1068 char umbrellaAndSuffix[strlen(context.imageSuffix)+strlen(aSubUmbrellaName)+1];
1069 strcpy(umbrellaAndSuffix, aSubUmbrellaName);
1070 strcat(umbrellaAndSuffix, context.imageSuffix);
1071 if ( strcmp(umbrellaAndSuffix, &lastSlash[1]) == 0 )
1072 return true;
1073 }
1074 }
1075 break;
1076 }
1077 cmd = (const struct load_command*)(((char*)cmd)+cmd->cmdsize);
1078 }
1079 }
1080 }
1081 }
1082 return false;
1083 }
1084
1085
1086 void* ImageLoaderMachO::getMain() const
1087 {
1088 const uint32_t cmd_count = ((macho_header*)fMachOData)->ncmds;
1089 const struct load_command* const cmds = (struct load_command*)&fMachOData[sizeof(macho_header)];
1090 const struct load_command* cmd = cmds;
1091 for (unsigned long i = 0; i < cmd_count; ++i) {
1092 switch (cmd->cmd) {
1093 case LC_UNIXTHREAD:
1094 {
1095 #if __ppc__
1096 const ppc_thread_state_t* registers = (ppc_thread_state_t*)(((char*)cmd) + 16);
1097 return (void*)registers->srr0;
1098 #elif __ppc64__
1099 const ppc_thread_state64_t* registers = (ppc_thread_state64_t*)(((char*)cmd) + 16);
1100 return (void*)registers->srr0;
1101 #elif __i386__
1102 const i386_thread_state_t* registers = (i386_thread_state_t*)(((char*)cmd) + 16);
1103 return (void*)registers->eip;
1104 #else
1105 #warning need processor specific code
1106 #endif
1107 }
1108 break;
1109 }
1110 cmd = (const struct load_command*)(((char*)cmd)+cmd->cmdsize);
1111 }
1112 return NULL;
1113 }
1114
1115
1116 uint32_t ImageLoaderMachO::doGetDependentLibraryCount()
1117 {
1118 const uint32_t cmd_count = ((macho_header*)fMachOData)->ncmds;
1119 const struct load_command* const cmds = (struct load_command*)&fMachOData[sizeof(macho_header)];
1120 uint32_t count = 0;
1121 const struct load_command* cmd = cmds;
1122 for (unsigned long i = 0; i < cmd_count; ++i) {
1123 switch (cmd->cmd) {
1124 case LC_LOAD_DYLIB:
1125 case LC_LOAD_WEAK_DYLIB:
1126 ++count;
1127 break;
1128 }
1129 cmd = (const struct load_command*)(((char*)cmd)+cmd->cmdsize);
1130 }
1131 return count;
1132 }
1133
1134 void ImageLoaderMachO::doGetDependentLibraries(DependentLibrary libs[])
1135 {
1136 uint32_t index = 0;
1137 const uint32_t cmd_count = ((macho_header*)fMachOData)->ncmds;
1138 const struct load_command* const cmds = (struct load_command*)&fMachOData[sizeof(macho_header)];
1139 const struct load_command* cmd = cmds;
1140 for (unsigned long i = 0; i < cmd_count; ++i) {
1141 switch (cmd->cmd) {
1142 case LC_LOAD_DYLIB:
1143 case LC_LOAD_WEAK_DYLIB:
1144 {
1145 const struct dylib_command* dylib = (struct dylib_command*)cmd;
1146 DependentLibrary* lib = &libs[index++];
1147 lib->name = (char*)cmd + dylib->dylib.name.offset;
1148 //lib->name = strdup((char*)cmd + dylib->dylib.name.offset);
1149 lib->image = NULL;
1150 lib->info.checksum = dylib->dylib.timestamp;
1151 lib->info.minVersion = dylib->dylib.compatibility_version;
1152 lib->info.maxVersion = dylib->dylib.current_version;
1153 lib->required = (cmd->cmd == LC_LOAD_DYLIB);
1154 lib->checksumMatches = false;
1155 lib->isReExported = false;
1156 }
1157 break;
1158 }
1159 cmd = (const struct load_command*)(((char*)cmd)+cmd->cmdsize);
1160 }
1161 }
1162
1163 ImageLoader::LibraryInfo ImageLoaderMachO::doGetLibraryInfo()
1164 {
1165 LibraryInfo info;
1166 if ( fDylibID != NULL ) {
1167 info.minVersion = fDylibID->dylib.compatibility_version;
1168 info.maxVersion = fDylibID->dylib.current_version;
1169 info.checksum = fDylibID->dylib.timestamp;
1170 }
1171 else {
1172 info.minVersion = 0;
1173 info.maxVersion = 0;
1174 info.checksum = 0;
1175 }
1176 return info;
1177 }
1178
1179
1180 uintptr_t ImageLoaderMachO::getRelocBase()
1181 {
1182 if ( fIsSplitSeg ) {
1183 // in split segment libraries r_address is offset from first writable segment
1184 const unsigned int segmentCount = fSegments.size();
1185 for(unsigned int i=0; i < segmentCount; ++i){
1186 Segment* seg = fSegments[i];
1187 if ( seg->writeable() ) {
1188 return seg->getActualLoadAddress();
1189 }
1190 }
1191 }
1192
1193 // in non-split segment libraries r_address is offset from first segment
1194 return fSegments[0]->getActualLoadAddress();
1195 }
1196
1197 #if __ppc__
1198 static inline void otherRelocsPPC(uintptr_t* locationToFix, uint8_t relocationType, uint16_t otherHalf, uintptr_t slide)
1199 {
1200 // low 16 bits of 32-bit ppc instructions need fixing
1201 struct ppcInstruction { uint16_t opcode; int16_t immediateValue; };
1202 ppcInstruction* instruction = (ppcInstruction*)locationToFix;
1203 //uint32_t before = *((uint32_t*)locationToFix);
1204 switch ( relocationType )
1205 {
1206 case PPC_RELOC_LO16:
1207 instruction->immediateValue = ((otherHalf << 16) | instruction->immediateValue) + slide;
1208 break;
1209 case PPC_RELOC_HI16:
1210 instruction->immediateValue = ((((instruction->immediateValue << 16) | otherHalf) + slide) >> 16);
1211 break;
1212 case PPC_RELOC_HA16:
1213 int16_t signedOtherHalf = (int16_t)(otherHalf & 0xffff);
1214 uint32_t temp = (instruction->immediateValue << 16) + signedOtherHalf + slide;
1215 if ( (temp & 0x00008000) != 0 )
1216 temp += 0x00008000;
1217 instruction->immediateValue = temp >> 16;
1218 }
1219 //uint32_t after = *((uint32_t*)locationToFix);
1220 //fprintf(stderr, "dyld: ppc fixup %0p type %d from 0x%08X to 0x%08X\n", locationToFix, relocationType, before, after);
1221 }
1222 #endif
1223
1224 void ImageLoaderMachO::doRebase(const LinkContext& context)
1225 {
1226 // if prebound and loaded at prebound address, then no need to rebase
1227 // Note: you might think that the check for allDependentLibrariesAsWhenPreBound() is not needed
1228 // but it is. If a dependent library changed, this image's lazy pointers into that library
1229 // need to be updated (reset back to lazy binding handler). That work is done most easily
1230 // here because there is a PPC_RELOC_PB_LA_PTR reloc record for each lazy pointer.
1231 if ( this->usablePrebinding(context) && this->usesTwoLevelNameSpace() ) {
1232 // skip rebasing cause prebound and prebinding not disabled
1233 ++fgImagesWithUsedPrebinding; // bump totals for statistics
1234 return;
1235 }
1236
1237 // print why prebinding was not used
1238 if ( context.verbosePrebinding ) {
1239 if ( !this->isPrebindable() ) {
1240 fprintf(stderr, "dyld: image not prebound, so could not use prebinding in %s\n", this->getPath());
1241 }
1242 else if ( fSlide != 0 ) {
1243 fprintf(stderr, "dyld: image slid, so could not use prebinding in %s\n", this->getPath());
1244 }
1245 else if ( !this->allDependentLibrariesAsWhenPreBound() ) {
1246 fprintf(stderr, "dyld: dependent libraries changed, so could not use prebinding in %s\n", this->getPath());
1247 }
1248 else if ( !this->usesTwoLevelNameSpace() ){
1249 fprintf(stderr, "dyld: image uses flat-namespace so, parts of prebinding ignored %s\n", this->getPath());
1250 }
1251 else {
1252 fprintf(stderr, "dyld: environment variable disabled use of prebinding in %s\n", this->getPath());
1253 }
1254 }
1255
1256 // if there are __TEXT fixups, temporarily make __TEXT writable
1257 if ( fTextSegmentWithFixups != NULL )
1258 fTextSegmentWithFixups->tempWritable();
1259
1260 // cache this value that is used in the following loop
1261 register const uintptr_t slide = this->fSlide;
1262
1263 // loop through all local (internal) relocation records
1264 const uintptr_t relocBase = this->getRelocBase();
1265 const relocation_info* const relocsStart = (struct relocation_info*)(&fLinkEditBase[fDynamicInfo->locreloff]);
1266 const relocation_info* const relocsEnd = &relocsStart[fDynamicInfo->nlocrel];
1267 for (const relocation_info* reloc=relocsStart; reloc < relocsEnd; ++reloc) {
1268 if ( (reloc->r_address & R_SCATTERED) == 0 ) {
1269 if ( reloc->r_symbolnum == R_ABS ) {
1270 // ignore absolute relocations
1271 }
1272 else if (reloc->r_length == RELOC_SIZE) {
1273 switch(reloc->r_type) {
1274 case GENERIC_RELOC_VANILLA:
1275 *((uintptr_t*)(reloc->r_address + relocBase)) += slide;
1276 break;
1277 #if __ppc__
1278 case PPC_RELOC_HI16:
1279 case PPC_RELOC_LO16:
1280 case PPC_RELOC_HA16:
1281 // some tools leave object file relocations in linked images
1282 otherRelocsPPC((uintptr_t*)(reloc->r_address + relocBase), reloc->r_type, reloc[1].r_address, slide);
1283 ++reloc; // these relocations come in pairs, skip next
1284 break;
1285 #endif
1286 default:
1287 throw "unknown local relocation type";
1288 }
1289 }
1290 else {
1291 throw "bad local relocation length";
1292 }
1293 }
1294 else {
1295 const struct scattered_relocation_info* sreloc = (struct scattered_relocation_info*)reloc;
1296 if (sreloc->r_length == RELOC_SIZE) {
1297 uintptr_t* locationToFix = (uintptr_t*)(sreloc->r_address + relocBase);
1298 switch(sreloc->r_type) {
1299 case GENERIC_RELOC_VANILLA:
1300 *locationToFix += slide;
1301 break;
1302 #if __ppc__ || __ppc64__
1303 case PPC_RELOC_PB_LA_PTR:
1304 // should only see these in prebound images, and we got here so prebinding is being ignored
1305 *locationToFix = sreloc->r_value + slide;
1306 break;
1307 #endif
1308 #if __ppc__
1309 case PPC_RELOC_HI16:
1310 case PPC_RELOC_LO16:
1311 case PPC_RELOC_HA16:
1312 // Metrowerks compiler sometimes leaves object file relocations in linked images???
1313 ++reloc; // these relocations come in pairs, get next one
1314 otherRelocsPPC(locationToFix, sreloc->r_type, reloc->r_address, slide);
1315 break;
1316 #endif
1317 #if __i386__
1318 case GENERIC_RELOC_PB_LA_PTR:
1319 // should only see these in prebound images, and we got here so prebinding is being ignored
1320 *locationToFix = sreloc->r_value + slide;
1321 break;
1322 #endif
1323 default:
1324 throw "unknown local scattered relocation type";
1325 }
1326 }
1327 else {
1328 throw "bad local scattered relocation length";
1329 }
1330 }
1331 }
1332
1333 // if there were __TEXT fixups, restore write protection
1334 if ( fTextSegmentWithFixups != NULL ) {
1335 fTextSegmentWithFixups->setPermissions();
1336 sys_icache_invalidate((void*)fTextSegmentWithFixups->getActualLoadAddress(), fTextSegmentWithFixups->getSize());
1337 }
1338
1339 // update stats
1340 fgTotalRebaseFixups += fDynamicInfo->nlocrel;
1341 }
1342
1343
1344 const struct macho_nlist* ImageLoaderMachO::binarySearchWithToc(const char* key, const char stringPool[], const struct macho_nlist symbols[],
1345 const struct dylib_table_of_contents toc[], uint32_t symbolCount, uint32_t hintIndex)
1346 {
1347 int32_t high = symbolCount-1;
1348 int32_t mid = hintIndex;
1349
1350 // handle out of range hint
1351 if ( mid >= (int32_t)symbolCount ) {
1352 mid = symbolCount/2;
1353 ++ImageLoaderMachO::fgUnhintedBinaryTreeSearchs;
1354 }
1355 else {
1356 ++ImageLoaderMachO::fgHintedBinaryTreeSearchs;
1357 }
1358
1359 for (int32_t low = 0; low <= high; mid = (low+high)/2) {
1360 const uint32_t index = toc[mid].symbol_index;
1361 const struct macho_nlist* pivot = &symbols[index];
1362 const char* pivotStr = &stringPool[pivot->n_un.n_strx];
1363 #if LINKEDIT_USAGE_DEBUG
1364 noteAccessedLinkEditAddress(&toc[mid]);
1365 noteAccessedLinkEditAddress(pivot);
1366 noteAccessedLinkEditAddress(pivotStr);
1367 #endif
1368 int cmp = astrcmp(key, pivotStr);
1369 if ( cmp == 0 )
1370 return pivot;
1371 if ( cmp > 0 ) {
1372 // key > pivot
1373 low = mid + 1;
1374 }
1375 else {
1376 // key < pivot
1377 high = mid - 1;
1378 }
1379 }
1380 return NULL;
1381 }
1382
1383 const struct macho_nlist* ImageLoaderMachO::binarySearch(const char* key, const char stringPool[], const struct macho_nlist symbols[], uint32_t symbolCount)
1384 {
1385 ++ImageLoaderMachO::fgUnhintedBinaryTreeSearchs;
1386 const struct macho_nlist* base = symbols;
1387 for (uint32_t n = symbolCount; n > 0; n /= 2) {
1388 const struct macho_nlist* pivot = &base[n/2];
1389 const char* pivotStr = &stringPool[pivot->n_un.n_strx];
1390 #if LINKEDIT_USAGE_DEBUG
1391 noteAccessedLinkEditAddress(pivot);
1392 noteAccessedLinkEditAddress(pivotStr);
1393 #endif
1394 int cmp = astrcmp(key, pivotStr);
1395 if ( cmp == 0 )
1396 return pivot;
1397 if ( cmp > 0 ) {
1398 // key > pivot
1399 // move base to symbol after pivot
1400 base = &pivot[1];
1401 --n;
1402 }
1403 else {
1404 // key < pivot
1405 // keep same base
1406 }
1407 }
1408 return NULL;
1409 }
1410
1411 const ImageLoader::Symbol* ImageLoaderMachO::findExportedSymbol(const char* name, const void* hint, bool searchReExports, ImageLoader** foundIn) const
1412 {
1413 const struct macho_nlist* sym = NULL;
1414 const struct twolevel_hint* theHint = (struct twolevel_hint*)hint;
1415 if ( fDynamicInfo->tocoff == 0 )
1416 sym = binarySearch(name, fStrings, &fSymbolTable[fDynamicInfo->iextdefsym], fDynamicInfo->nextdefsym);
1417 else {
1418 uint32_t start = fDynamicInfo->nextdefsym;
1419 if ( theHint != NULL )
1420 start = theHint->itoc;
1421 if ( (theHint == NULL) || (theHint->isub_image == 0) ) {
1422 sym = binarySearchWithToc(name, fStrings, fSymbolTable, (dylib_table_of_contents*)&fLinkEditBase[fDynamicInfo->tocoff],
1423 fDynamicInfo->ntoc, start);
1424 }
1425 }
1426 if ( sym != NULL ) {
1427 if ( foundIn != NULL )
1428 *foundIn = (ImageLoader*)this;
1429
1430 return (const Symbol*)sym;
1431 }
1432
1433 if ( searchReExports ) {
1434 // hint might tell us to try a particular subimage
1435 if ( (theHint != NULL) && (theHint->isub_image > 0) && (theHint->isub_image <= fLibrariesCount) ) {
1436 // isub_image is an index into a list that is sorted non-rexported images first
1437 uint32_t index = 0;
1438 ImageLoader* target = NULL;
1439 // pass one, only look at sub-frameworks
1440 for (uint32_t i=0; i < fLibrariesCount; ++i) {
1441 DependentLibrary& libInfo = fLibraries[i];
1442 if ( libInfo.isSubFramework && (libInfo.image != NULL)) {
1443 if ( ++index == theHint->isub_image ) {
1444 target = libInfo.image;
1445 break;
1446 }
1447 }
1448 }
1449 if (target != NULL) {
1450 // pass two, only look at non-sub-framework-reexports
1451 for (uint32_t i=0; i < fLibrariesCount; ++i) {
1452 DependentLibrary& libInfo = fLibraries[i];
1453 if ( libInfo.isReExported && !libInfo.isSubFramework && (libInfo.image != NULL) ) {
1454 if ( ++index == theHint->isub_image ) {
1455 target = libInfo.image;
1456 break;
1457 }
1458 }
1459 }
1460 }
1461 if (target != NULL) {
1462 const Symbol* result = target->findExportedSymbol(name, NULL, searchReExports, foundIn);
1463 if ( result != NULL )
1464 return result;
1465 }
1466 }
1467
1468 // hint failed, try all sub images
1469 // pass one, only look at sub-frameworks
1470 for(unsigned int i=0; i < fLibrariesCount; ++i){
1471 DependentLibrary& libInfo = fLibraries[i];
1472 if ( (libInfo.image != NULL) && libInfo.isSubFramework ) {
1473 const Symbol* result = libInfo.image->findExportedSymbol(name, NULL, searchReExports, foundIn);
1474 if ( result != NULL )
1475 return result;
1476 }
1477 }
1478 // pass two, only look at non-sub-framework-reexports
1479 for(unsigned int i=0; i < fLibrariesCount; ++i){
1480 DependentLibrary& libInfo = fLibraries[i];
1481 if ( (libInfo.image != NULL) && libInfo.isReExported && !libInfo.isSubFramework ) {
1482 const Symbol* result = libInfo.image->findExportedSymbol(name, NULL, searchReExports, foundIn);
1483 if ( result != NULL )
1484 return result;
1485 }
1486 }
1487 }
1488
1489 // last change: the hint is wrong (non-zero but actually in this image)
1490 if ( (theHint != NULL) && (theHint->isub_image != 0) ) {
1491 sym = binarySearchWithToc(name, fStrings, fSymbolTable, (dylib_table_of_contents*)&fLinkEditBase[fDynamicInfo->tocoff],
1492 fDynamicInfo->ntoc, fDynamicInfo->nextdefsym);
1493 if ( sym != NULL ) {
1494 if ( foundIn != NULL )
1495 *foundIn = (ImageLoader*)this;
1496 return (const Symbol*)sym;
1497 }
1498 }
1499
1500 return NULL;
1501 }
1502
1503
1504 uintptr_t ImageLoaderMachO::getExportedSymbolAddress(const Symbol* sym) const
1505 {
1506 const struct macho_nlist* nlistSym = (const struct macho_nlist*)sym;
1507 return nlistSym->n_value + fSlide;
1508 }
1509
1510 ImageLoader::DefinitionFlags ImageLoaderMachO::getExportedSymbolInfo(const Symbol* sym) const
1511 {
1512 const struct macho_nlist* nlistSym = (const struct macho_nlist*)sym;
1513 if ( (nlistSym->n_desc & N_WEAK_DEF) != 0 )
1514 return kWeakDefinition;
1515 return kNoDefinitionOptions;
1516 }
1517
1518 const char* ImageLoaderMachO::getExportedSymbolName(const Symbol* sym) const
1519 {
1520 const struct macho_nlist* nlistSym = (const struct macho_nlist*)sym;
1521 return &fStrings[nlistSym->n_un.n_strx];
1522 }
1523
1524 uint32_t ImageLoaderMachO::getExportedSymbolCount() const
1525 {
1526 return fDynamicInfo->nextdefsym;
1527 }
1528
1529
1530 const ImageLoader::Symbol* ImageLoaderMachO::getIndexedExportedSymbol(uint32_t index) const
1531 {
1532 if ( index < fDynamicInfo->nextdefsym ) {
1533 const struct macho_nlist* sym = &fSymbolTable[fDynamicInfo->iextdefsym + index];
1534 return (const ImageLoader::Symbol*)sym;
1535 }
1536 return NULL;
1537 }
1538
1539
1540 uint32_t ImageLoaderMachO::getImportedSymbolCount() const
1541 {
1542 return fDynamicInfo->nundefsym;
1543 }
1544
1545
1546 const ImageLoader::Symbol* ImageLoaderMachO::getIndexedImportedSymbol(uint32_t index) const
1547 {
1548 if ( index < fDynamicInfo->nundefsym ) {
1549 const struct macho_nlist* sym = &fSymbolTable[fDynamicInfo->iundefsym + index];
1550 return (const ImageLoader::Symbol*)sym;
1551 }
1552 return NULL;
1553 }
1554
1555
1556 ImageLoader::ReferenceFlags ImageLoaderMachO::geImportedSymbolInfo(const ImageLoader::Symbol* sym) const
1557 {
1558 const struct macho_nlist* nlistSym = (const struct macho_nlist*)sym;
1559 ImageLoader::ReferenceFlags flags = kNoReferenceOptions;
1560 if ( ((nlistSym->n_type & N_TYPE) == N_UNDF) && (nlistSym->n_value != 0) )
1561 flags |= ImageLoader::kTentativeDefinition;
1562 if ( (nlistSym->n_desc & N_WEAK_REF) != 0 )
1563 flags |= ImageLoader::kWeakReference;
1564 return flags;
1565 }
1566
1567
1568 const char* ImageLoaderMachO::getImportedSymbolName(const ImageLoader::Symbol* sym) const
1569 {
1570 const struct macho_nlist* nlistSym = (const struct macho_nlist*)sym;
1571 return &fStrings[nlistSym->n_un.n_strx];
1572 }
1573
1574
1575 bool ImageLoaderMachO::getSectionContent(const char* segmentName, const char* sectionName, void** start, size_t* length)
1576 {
1577 const uint32_t cmd_count = ((macho_header*)fMachOData)->ncmds;
1578 const struct load_command* const cmds = (struct load_command*)&fMachOData[sizeof(macho_header)];
1579 const struct load_command* cmd = cmds;
1580 for (uint32_t i = 0; i < cmd_count; ++i) {
1581 switch (cmd->cmd) {
1582 case LC_SEGMENT_COMMAND:
1583 {
1584 const struct macho_segment_command* seg = (struct macho_segment_command*)cmd;
1585 const struct macho_section* const sectionsStart = (struct macho_section*)((char*)seg + sizeof(struct macho_segment_command));
1586 const struct macho_section* const sectionsEnd = &sectionsStart[seg->nsects];
1587 for (const struct macho_section* sect=sectionsStart; sect < sectionsEnd; ++sect) {
1588 if ( (strcmp(sect->segname, segmentName) == 0) && (strcmp(sect->sectname, sectionName) == 0) ) {
1589 *start = (uintptr_t*)(sect->addr + fSlide);
1590 *length = sect->size;
1591 return true;
1592 }
1593 }
1594 }
1595 break;
1596 }
1597 cmd = (const struct load_command*)(((char*)cmd)+cmd->cmdsize);
1598 }
1599 return false;
1600 }
1601
1602
1603 bool ImageLoaderMachO::findSection(const void* imageInterior, const char** segmentName, const char** sectionName, size_t* sectionOffset)
1604 {
1605 const uint32_t cmd_count = ((macho_header*)fMachOData)->ncmds;
1606 const struct load_command* const cmds = (struct load_command*)&fMachOData[sizeof(macho_header)];
1607 const struct load_command* cmd = cmds;
1608 const uintptr_t unslidInteriorAddress = (uintptr_t)imageInterior - this->getSlide();
1609 for (uint32_t i = 0; i < cmd_count; ++i) {
1610 switch (cmd->cmd) {
1611 case LC_SEGMENT_COMMAND:
1612 {
1613 const struct macho_segment_command* seg = (struct macho_segment_command*)cmd;
1614 if ( (unslidInteriorAddress >= seg->vmaddr) && (unslidInteriorAddress < (seg->vmaddr+seg->vmsize)) ) {
1615 const struct macho_section* const sectionsStart = (struct macho_section*)((char*)seg + sizeof(struct macho_segment_command));
1616 const struct macho_section* const sectionsEnd = &sectionsStart[seg->nsects];
1617 for (const struct macho_section* sect=sectionsStart; sect < sectionsEnd; ++sect) {
1618 if ((sect->addr <= unslidInteriorAddress) && (unslidInteriorAddress < (sect->addr+sect->size))) {
1619 if ( segmentName != NULL )
1620 *segmentName = sect->segname;
1621 if ( sectionName != NULL )
1622 *sectionName = sect->sectname;
1623 if ( sectionOffset != NULL )
1624 *sectionOffset = unslidInteriorAddress - sect->addr;
1625 return true;
1626 }
1627 }
1628 }
1629 }
1630 break;
1631 }
1632 cmd = (const struct load_command*)(((char*)cmd)+cmd->cmdsize);
1633 }
1634 return false;
1635 }
1636
1637
1638 bool ImageLoaderMachO::symbolRequiresCoalescing(const struct macho_nlist* symbol)
1639 {
1640 // if a define and weak ==> coalesced
1641 if ( ((symbol->n_type & N_TYPE) == N_SECT) && ((symbol->n_desc & N_WEAK_DEF) != 0) )
1642 return true;
1643 // if an undefine and not referencing a weak symbol ==> coalesced
1644 if ( ((symbol->n_type & N_TYPE) != N_SECT) && ((symbol->n_desc & N_REF_TO_WEAK) != 0) )
1645 return true;
1646
1647 // regular symbol
1648 return false;
1649 }
1650
1651
1652 static void __attribute__((noreturn)) throwSymbolNotFound(const char* symbol, const char* referencedFrom, const char* expectedIn)
1653 {
1654 const char* formatString = "Symbol not found: %s\n Referenced from: %s\n Expected in: %s\n";
1655 char buf[strlen(symbol)+strlen(referencedFrom)+strlen(expectedIn)+strlen(formatString)];
1656 sprintf(buf, formatString, symbol, referencedFrom, expectedIn);
1657 throw strdup(buf); // this is a leak if exception doesn't halt program
1658 }
1659
1660 uintptr_t ImageLoaderMachO::resolveUndefined(const LinkContext& context, const struct macho_nlist* undefinedSymbol, bool twoLevel, ImageLoader** foundIn)
1661 {
1662 const char* symbolName = &fStrings[undefinedSymbol->n_un.n_strx];
1663
1664 if ( context.bindFlat || !twoLevel ) {
1665 // flat lookup
1666 const Symbol* sym;
1667 if ( context.flatExportFinder(symbolName, &sym, foundIn) )
1668 return (*foundIn)->getExportedSymbolAddress(sym);
1669 // if a bundle is loaded privately the above will not find its exports
1670 if ( this->isBundle() && this->hasHiddenExports() ) {
1671 // look in self for needed symbol
1672 sym = this->findExportedSymbol(symbolName, NULL, false, foundIn);
1673 if ( sym != NULL )
1674 return (*foundIn)->getExportedSymbolAddress(sym);
1675 }
1676 if ( ((undefinedSymbol->n_type & N_PEXT) != 0) || ((undefinedSymbol->n_type & N_TYPE) == N_SECT) ) {
1677 // could be a multi-module private_extern internal reference
1678 // the static linker squirrels away the target address in n_value
1679 uintptr_t addr = undefinedSymbol->n_value + this->fSlide;
1680 *foundIn = this;
1681 return addr;
1682 }
1683 if ( (undefinedSymbol->n_desc & N_WEAK_REF) != 0 ) {
1684 // definition can't be found anywhere
1685 // if reference is weak_import, then it is ok, just return 0
1686 return 0;
1687 }
1688 throwSymbolNotFound(symbolName, this->getPath(), "flat namespace");
1689 }
1690 else {
1691 // symbol requires searching images with coalesced symbols
1692 if ( this->needsCoalescing() && symbolRequiresCoalescing(undefinedSymbol) ) {
1693 const Symbol* sym;
1694 if ( context.coalescedExportFinder(symbolName, &sym, foundIn) )
1695 return (*foundIn)->getExportedSymbolAddress(sym);
1696 //throwSymbolNotFound(symbolName, this->getPath(), "coalesced namespace");
1697 //fprintf(stderr, "dyld: coalesced symbol %s not found in any coalesced image, falling back to two-level lookup", symbolName);
1698 }
1699
1700 // two level lookup
1701 void* hint = NULL;
1702 ImageLoader* target = NULL;
1703 uint8_t ord = GET_LIBRARY_ORDINAL(undefinedSymbol->n_desc);
1704 if ( ord == EXECUTABLE_ORDINAL ) {
1705 target = context.mainExecutable;
1706 }
1707 else if ( ord == SELF_LIBRARY_ORDINAL ) {
1708 target = this;
1709 }
1710 else if ( ord == DYNAMIC_LOOKUP_ORDINAL ) {
1711 // rnielsen: HACKHACK
1712 // flat lookup
1713 const Symbol* sym;
1714 if ( context.flatExportFinder(symbolName, &sym, foundIn) )
1715 return (*foundIn)->getExportedSymbolAddress(sym);
1716 // no image has exports this symbol
1717 // either report error or hope ZeroLink can just-in-time load an image
1718 context.undefinedHandler(symbolName);
1719 // try looking again
1720 if ( context.flatExportFinder(symbolName, &sym, foundIn) )
1721 return (*foundIn)->getExportedSymbolAddress(sym);
1722
1723 throwSymbolNotFound(symbolName, this->getPath(), "dynamic lookup");
1724 }
1725 else if ( ord <= fLibrariesCount ) {
1726 DependentLibrary& libInfo = fLibraries[ord-1];
1727 target = libInfo.image;
1728 if ( (target == NULL) && (((undefinedSymbol->n_desc & N_WEAK_REF) != 0) || !libInfo.required) ) {
1729 // if target library not loaded and reference is weak or library is weak return 0
1730 return 0;
1731 }
1732 }
1733 else {
1734 throw "corrupt binary, library ordinal too big";
1735 }
1736
1737 if ( target == NULL ) {
1738 fprintf(stderr, "resolveUndefined(%s) in %s\n", symbolName, this->getPath());
1739 throw "symbol not found";
1740 }
1741
1742 // interpret hint
1743 if ( fTwoLevelHints != NULL ) {
1744 uint32_t symIndex = undefinedSymbol - fSymbolTable;
1745 int32_t undefinedIndex = symIndex - fDynamicInfo->iundefsym;
1746 if ( (undefinedIndex >= 0) && ((uint32_t)undefinedIndex < fDynamicInfo->nundefsym) ) {
1747 const struct twolevel_hint* hints = (struct twolevel_hint*)(&fLinkEditBase[fTwoLevelHints->offset]);
1748 const struct twolevel_hint* theHint = &hints[undefinedIndex];
1749 hint = (void*)theHint;
1750 }
1751 }
1752
1753 const Symbol* sym = target->findExportedSymbol(symbolName, hint, true, foundIn);
1754 if ( sym!= NULL ) {
1755 return (*foundIn)->getExportedSymbolAddress(sym);
1756 }
1757 else if ( (undefinedSymbol->n_type & N_PEXT) != 0 ) {
1758 // don't know why the static linker did not eliminate the internal reference to a private extern definition
1759 *foundIn = this;
1760 return undefinedSymbol->n_value + fSlide;
1761 }
1762 else if ( (undefinedSymbol->n_desc & N_WEAK_REF) != 0 ) {
1763 // if definition not found and reference is weak return 0
1764 return 0;
1765 }
1766
1767 // nowhere to be found
1768 throwSymbolNotFound(symbolName, this->getPath(), target->getPath());
1769 }
1770 }
1771
1772 // returns if 'addr' is within the address range of section 'sectionIndex'
1773 // fSlide is not used. 'addr' is assumed to be a prebound address in this image
1774 bool ImageLoaderMachO::isAddrInSection(uintptr_t addr, uint8_t sectionIndex)
1775 {
1776 uint8_t currentSectionIndex = 1;
1777 const uint32_t cmd_count = ((macho_header*)fMachOData)->ncmds;
1778 const struct load_command* const cmds = (struct load_command*)&fMachOData[sizeof(macho_header)];
1779 const struct load_command* cmd = cmds;
1780 for (unsigned long i = 0; i < cmd_count; ++i) {
1781 if ( cmd->cmd == LC_SEGMENT_COMMAND ) {
1782 const struct macho_segment_command* seg = (struct macho_segment_command*)cmd;
1783 if ( (currentSectionIndex <= sectionIndex) && (sectionIndex < currentSectionIndex+seg->nsects) ) {
1784 // 'sectionIndex' is in this segment, get section info
1785 const struct macho_section* const sectionsStart = (struct macho_section*)((char*)seg + sizeof(struct macho_segment_command));
1786 const struct macho_section* const section = &sectionsStart[sectionIndex-currentSectionIndex];
1787 return ( (section->addr <= addr) && (addr < section->addr+section->size) );
1788 }
1789 else {
1790 // 'sectionIndex' not in this segment, skip to next segment
1791 currentSectionIndex += seg->nsects;
1792 }
1793 }
1794 cmd = (const struct load_command*)(((char*)cmd)+cmd->cmdsize);
1795 }
1796
1797 return false;
1798 }
1799
1800 void ImageLoaderMachO::doBindExternalRelocations(const LinkContext& context, bool onlyCoalescedSymbols)
1801 {
1802 const uintptr_t relocBase = this->getRelocBase();
1803 const bool twoLevel = this->usesTwoLevelNameSpace();
1804 const bool prebound = this->isPrebindable();
1805
1806 // if there are __TEXT fixups, temporarily make __TEXT writable
1807 if ( fTextSegmentWithFixups != NULL )
1808 fTextSegmentWithFixups->tempWritable();
1809
1810 // cache last lookup
1811 const struct macho_nlist* lastUndefinedSymbol = 0;
1812 uintptr_t symbolAddr = 0;
1813 ImageLoader* image = NULL;
1814
1815 // loop through all external relocation records and bind each
1816 const relocation_info* const relocsStart = (struct relocation_info*)(&fLinkEditBase[fDynamicInfo->extreloff]);
1817 const relocation_info* const relocsEnd = &relocsStart[fDynamicInfo->nextrel];
1818 for (const relocation_info* reloc=relocsStart; reloc < relocsEnd; ++reloc) {
1819 if (reloc->r_length == RELOC_SIZE) {
1820 switch(reloc->r_type) {
1821 case GENERIC_RELOC_VANILLA:
1822 {
1823 const struct macho_nlist* undefinedSymbol = &fSymbolTable[reloc->r_symbolnum];
1824 // if only processing coalesced symbols and this one does not require coalesceing, skip to next
1825 if ( onlyCoalescedSymbols && !symbolRequiresCoalescing(undefinedSymbol) )
1826 continue;
1827 uintptr_t* location = ((uintptr_t*)(reloc->r_address + relocBase));
1828 uintptr_t value = *location;
1829 #if __i386__
1830 if ( reloc->r_pcrel ) {
1831 value += (uintptr_t)location + 4 - fSlide;
1832 }
1833 #endif
1834 if ( prebound ) {
1835 // we are doing relocations, so prebinding was not usable
1836 // in a prebound executable, the n_value field of an undefined symbol is set to the address where the symbol was found when prebound
1837 // so, subtracting that gives the initial displacement which we need to add to the newly found symbol address
1838 // if mach-o relocation structs had an "addend" field this complication would not be necessary.
1839 if ( ((undefinedSymbol->n_type & N_TYPE) == N_SECT) && ((undefinedSymbol->n_desc & N_WEAK_DEF) != 0) ) {
1840 // weak symbols need special casing, since *location may have been prebound to a definition in another image.
1841 // If *location is currently prebound to somewhere in the same section as the weak definition, we assume
1842 // that we can subtract off the weak symbol address to get the addend.
1843 // If prebound elsewhere, we've lost the addend and have to assume it is zero.
1844 // The prebinding to elsewhere only happens with 10.4+ update_prebinding which only operates on a small set of Apple dylibs
1845 if ( (value == undefinedSymbol->n_value) || this->isAddrInSection(value, undefinedSymbol->n_sect) )
1846 value -= undefinedSymbol->n_value;
1847 else
1848 value = 0;
1849 }
1850 else {
1851 // is undefined or non-weak symbol, so do subtraction to get addend
1852 value -= undefinedSymbol->n_value;
1853 }
1854 }
1855 // if undefinedSymbol is same as last time, then symbolAddr and image will resolve to the same too
1856 if ( undefinedSymbol != lastUndefinedSymbol ) {
1857 symbolAddr = this->resolveUndefined(context, undefinedSymbol, twoLevel, &image);
1858 lastUndefinedSymbol = undefinedSymbol;
1859 }
1860 if ( context.verboseBind ) {
1861 const char *path = NULL;
1862 if(NULL != image) {
1863 path = image->getShortName();
1864 }
1865 if(0 == value) {
1866 fprintf(stderr, "dyld: bind: %s:0x%08lx = %s:%s, *0x%08lx = 0x%08lx\n",
1867 this->getShortName(), (uintptr_t)location,
1868 path, &fStrings[undefinedSymbol->n_un.n_strx], (uintptr_t)location, symbolAddr);
1869 }
1870 else {
1871 fprintf(stderr, "dyld: bind: %s:0x%08lx = %s:%s, *0x%08lx = 0x%08lx + %ld\n",
1872 this->getShortName(), (uintptr_t)location,
1873 path, &fStrings[undefinedSymbol->n_un.n_strx], (uintptr_t)location, symbolAddr, value);
1874 }
1875 }
1876 value += symbolAddr;
1877 #if __i386__
1878 if ( reloc->r_pcrel ) {
1879 *location = value - ((uintptr_t)location + 4);
1880 }
1881 else {
1882 // don't dirty page if prebound value was correct
1883 if ( !prebound || (*location != value) )
1884 *location = value;
1885 }
1886 #else
1887 // don't dirty page if prebound value was correct
1888 if ( !prebound || (*location != value) )
1889 *location = value;
1890 #endif
1891 }
1892 break;
1893 default:
1894 throw "unknown external relocation type";
1895 }
1896 }
1897 else {
1898 throw "bad external relocation length";
1899 }
1900 }
1901
1902 // if there were __TEXT fixups, restore write protection
1903 if ( fTextSegmentWithFixups != NULL ) {
1904 fTextSegmentWithFixups->setPermissions();
1905 sys_icache_invalidate((void*)fTextSegmentWithFixups->getActualLoadAddress(), fTextSegmentWithFixups->getSize());
1906 }
1907
1908 // update stats
1909 fgTotalBindFixups += fDynamicInfo->nextrel;
1910 }
1911
1912 const mach_header* ImageLoaderMachO::machHeader() const
1913 {
1914 return (mach_header*)fMachOData;
1915 }
1916
1917 uintptr_t ImageLoaderMachO::getSlide() const
1918 {
1919 return fSlide;
1920 }
1921
1922 // hmm. maybe this should be up in ImageLoader??
1923 const void* ImageLoaderMachO::getBaseAddress() const
1924 {
1925 Segment* seg = fSegments[0];
1926 return (const void*)seg->getActualLoadAddress();
1927 }
1928
1929
1930 uintptr_t ImageLoaderMachO::doBindLazySymbol(uintptr_t* lazyPointer, const LinkContext& context)
1931 {
1932 // scan for all non-lazy-pointer sections
1933 const bool twoLevel = this->usesTwoLevelNameSpace();
1934 const uint32_t cmd_count = ((macho_header*)fMachOData)->ncmds;
1935 const struct load_command* const cmds = (struct load_command*)&fMachOData[sizeof(macho_header)];
1936 const struct load_command* cmd = cmds;
1937 const uint32_t* const indirectTable = (uint32_t*)&fLinkEditBase[fDynamicInfo->indirectsymoff];
1938 for (uint32_t i = 0; i < cmd_count; ++i) {
1939 switch (cmd->cmd) {
1940 case LC_SEGMENT_COMMAND:
1941 {
1942 const struct macho_segment_command* seg = (struct macho_segment_command*)cmd;
1943 const struct macho_section* const sectionsStart = (struct macho_section*)((char*)seg + sizeof(struct macho_segment_command));
1944 const struct macho_section* const sectionsEnd = &sectionsStart[seg->nsects];
1945 for (const struct macho_section* sect=sectionsStart; sect < sectionsEnd; ++sect) {
1946 const uint8_t type = sect->flags & SECTION_TYPE;
1947 if ( type == S_LAZY_SYMBOL_POINTERS ) {
1948 const uint32_t pointerCount = sect->size / sizeof(uintptr_t);
1949 uintptr_t* const symbolPointers = (uintptr_t*)(sect->addr + fSlide);
1950 if ( (lazyPointer >= symbolPointers) && (lazyPointer < &symbolPointers[pointerCount]) ) {
1951 const uint32_t indirectTableOffset = sect->reserved1;
1952 const uint32_t lazyIndex = lazyPointer - symbolPointers;
1953 uint32_t symbolIndex = indirectTable[indirectTableOffset + lazyIndex];
1954 if ( symbolIndex != INDIRECT_SYMBOL_ABS && symbolIndex != INDIRECT_SYMBOL_LOCAL ) {
1955 ImageLoader *image = NULL;
1956 const char *path = NULL;
1957 uintptr_t symbolAddr = this->resolveUndefined(context, &fSymbolTable[symbolIndex], twoLevel, &image);
1958 if ( context.verboseBind ) {
1959 if(NULL == path && NULL != image) {
1960 path = image->getShortName();
1961 }
1962 fprintf(stderr, "dyld: bind: %s:%s$%s = %s:%s, *0x%08lx = 0x%08lx\n",
1963 this->getShortName(), &fStrings[fSymbolTable[symbolIndex].n_un.n_strx], "lazy_ptr",
1964 path, &fStrings[fSymbolTable[symbolIndex].n_un.n_strx], (uintptr_t)&symbolPointers[lazyIndex], symbolAddr);
1965 }
1966 if ( NULL != context.bindingHandler ) {
1967 if(NULL == path && NULL != image) {
1968 path = image->getPath();
1969 }
1970 symbolAddr = (uintptr_t)context.bindingHandler(path, &fStrings[fSymbolTable[symbolIndex].n_un.n_strx], (void *)symbolAddr);
1971 }
1972 symbolPointers[lazyIndex] = symbolAddr;
1973 // update stats
1974 fgTotalLazyBindFixups++;
1975 return symbolPointers[lazyIndex];
1976 }
1977 }
1978 }
1979 }
1980 }
1981 break;
1982 }
1983 cmd = (const struct load_command*)(((char*)cmd)+cmd->cmdsize);
1984 }
1985 throw "lazy pointer not found";
1986 }
1987
1988
1989
1990 void ImageLoaderMachO::doBindIndirectSymbolPointers(const LinkContext& context, BindingLaziness bindness, bool onlyCoalescedSymbols)
1991 {
1992 // scan for all non-lazy-pointer sections
1993 const bool twoLevel = this->usesTwoLevelNameSpace();
1994 const uint32_t cmd_count = ((macho_header*)fMachOData)->ncmds;
1995 const struct load_command* const cmds = (struct load_command*)&fMachOData[sizeof(macho_header)];
1996 const struct load_command* cmd = cmds;
1997 const uint32_t* const indirectTable = (uint32_t*)&fLinkEditBase[fDynamicInfo->indirectsymoff];
1998 for (uint32_t i = 0; i < cmd_count; ++i) {
1999 switch (cmd->cmd) {
2000 case LC_SEGMENT_COMMAND:
2001 {
2002 const struct macho_segment_command* seg = (struct macho_segment_command*)cmd;
2003 const struct macho_section* const sectionsStart = (struct macho_section*)((char*)seg + sizeof(struct macho_segment_command));
2004 const struct macho_section* const sectionsEnd = &sectionsStart[seg->nsects];
2005 for (const struct macho_section* sect=sectionsStart; sect < sectionsEnd; ++sect) {
2006 const uint8_t type = sect->flags & SECTION_TYPE;
2007 const uint32_t pointerCount = sect->size / sizeof(uintptr_t);
2008 if ( type == S_NON_LAZY_SYMBOL_POINTERS ) {
2009 if ( (bindness == kLazyOnly) || (bindness == kLazyOnlyNoDependents) )
2010 continue;
2011 }
2012 else if ( type == S_LAZY_SYMBOL_POINTERS ) {
2013 // process each symbol pointer in this section
2014 fgTotalPossibleLazyBindFixups += pointerCount;
2015 if ( bindness == kNonLazyOnly )
2016 continue;
2017 }
2018 else {
2019 continue;
2020 }
2021 const uint32_t indirectTableOffset = sect->reserved1;
2022 uintptr_t* const symbolPointers = (uintptr_t*)(sect->addr + fSlide);
2023 for (uint32_t j=0; j < pointerCount; ++j) {
2024 uint32_t symbolIndex = indirectTable[indirectTableOffset + j];
2025 if ( symbolIndex == INDIRECT_SYMBOL_LOCAL) {
2026 symbolPointers[j] += this->fSlide;
2027 }
2028 else if ( symbolIndex == INDIRECT_SYMBOL_ABS) {
2029 // do nothing since already has absolute address
2030 }
2031 else {
2032 const struct macho_nlist* sym = &fSymbolTable[symbolIndex];
2033 if ( symbolIndex == 0 ) {
2034 // This could be rdar://problem/3534709
2035 if ( ((const macho_header*)fMachOData)->filetype == MH_EXECUTE ) {
2036 static bool alreadyWarned = false;
2037 if ( (sym->n_type & N_TYPE) != N_UNDF ) {
2038 // The indirect table parallels the (non)lazy pointer sections. For
2039 // instance, to find info about the fifth lazy pointer you look at the
2040 // fifth entry in the indirect table. (try otool -Iv on a file).
2041 // The entry in the indirect table contains an index into the symbol table.
2042
2043 // The bug in ld caused the entry in the indirect table to be zero
2044 // (instead of a magic value that means a local symbol). So, if the
2045 // symbolIndex == 0, we may be encountering the bug, or 0 may be a valid
2046 // symbol table index. The check I put in place is to see if the zero'th
2047 // symbol table entry is an import entry (usually it is a local symbol
2048 // definition).
2049 if ( context.verboseWarnings && !alreadyWarned ) {
2050 fprintf(stderr, "dyld: malformed executable '%s', skipping indirect symbol to %s\n",
2051 this->getPath(), &fStrings[sym->n_un.n_strx]);
2052 alreadyWarned = true;
2053 }
2054 continue;
2055 }
2056 }
2057 }
2058 ImageLoader *image = NULL;
2059 // if only processing coalesced symbols and this one does not require coalesceing, skip to next
2060 if ( onlyCoalescedSymbols && !symbolRequiresCoalescing(sym) )
2061 continue;
2062 uintptr_t symbolAddr;
2063 symbolAddr = resolveUndefined(context, sym, twoLevel, &image);
2064 if ( context.verboseBind ) {
2065 const char *path = NULL;
2066 if(NULL != image) {
2067 path = image->getShortName();
2068 }
2069 const char *typeName;
2070 if ( type == S_LAZY_SYMBOL_POINTERS ) {
2071 typeName = "lazy_ptr";
2072 }
2073 else {
2074 typeName = "non_lazy_ptr";
2075 }
2076 fprintf(stderr, "dyld: bind: %s:%s$%s = %s:%s, *0x%08lx = 0x%08lx\n",
2077 this->getShortName(), &fStrings[sym->n_un.n_strx], typeName,
2078 path, &fStrings[sym->n_un.n_strx], (uintptr_t)&symbolPointers[j], symbolAddr);
2079 }
2080 symbolPointers[j] = symbolAddr;
2081 }
2082 }
2083 // update stats
2084 fgTotalBindFixups += pointerCount;
2085 }
2086 }
2087 break;
2088 }
2089 cmd = (const struct load_command*)(((char*)cmd)+cmd->cmdsize);
2090 }
2091 }
2092
2093 /*
2094 * The address of these symbols are written in to the (__DATA,__dyld) section
2095 * at the following offsets:
2096 * at offset 0 stub_binding_helper_interface
2097 * at offset 4 _dyld_func_lookup
2098 * at offset 8 start_debug_thread
2099 * The 'C' types (if any) for these symbols are ignored here and all are
2100 * declared as longs so the assignment of their address in to the section will
2101 * not require a cast. stub_binding_helper_interface is really a label in the
2102 * assembly code interface for the stub binding. It does not have a meaningful
2103 * 'C' type. _dyld_func_lookup is the routine in dyld_libfuncs.c.
2104 * start_debug_thread is the routine in debug.c.
2105 *
2106 * For ppc the image's stub_binding_binding_helper is read from:
2107 * at offset 20 the image's stub_binding_binding_helper address
2108 * and saved into to the image structure.
2109 */
2110 struct DATAdyld {
2111 void* dyldLazyBinder; // filled in at launch by dyld to point into dyld to &stub_binding_helper_interface
2112 void* dyldFuncLookup; // filled in at launch by dyld to point into dyld to &_dyld_func_lookup
2113 void* startDebugThread; // debugger interface ???
2114 void* debugPort; // debugger interface ???
2115 void* debugThread; // debugger interface ???
2116 void* stubBindHelper; // filled in at static link time to point to stub helper in image
2117 void* coreDebug; // ???
2118 };
2119
2120 // These are defined in dyldStartup.s
2121 extern "C" void stub_binding_helper();
2122 extern "C" bool dyld_func_lookup(const char* name, uintptr_t* address);
2123
2124
2125 void ImageLoaderMachO::setupLazyPointerHandler()
2126 {
2127 if ( fDATAdyld != NULL ) {
2128 struct DATAdyld* dd = (struct DATAdyld*)(fDATAdyld->addr + fSlide);
2129 if ( fDATAdyld->size > offsetof(DATAdyld, dyldLazyBinder) ) {
2130 if ( dd->dyldLazyBinder != (void*)&stub_binding_helper )
2131 dd->dyldLazyBinder = (void*)&stub_binding_helper;
2132 }
2133 if ( fDATAdyld->size > offsetof(DATAdyld, dyldFuncLookup) ) {
2134 if ( dd->dyldFuncLookup != (void*)&dyld_func_lookup )
2135 dd->dyldFuncLookup = (void*)&dyld_func_lookup;
2136 }
2137 //if ( fDATAdyld->size > offsetof(DATAdyld, startDebugThread) )
2138 // dd->startDebugThread = &start_debug_thread;
2139 #ifdef __ppc__
2140 //if ( fDATAdyld->size > offsetof(DATAdyld, stubBindHelper) )
2141 // save = dd->stubBindHelper;
2142 #endif
2143 }
2144 }
2145
2146 bool ImageLoaderMachO::usablePrebinding(const LinkContext& context) const
2147 {
2148 // if prebound and loaded at prebound address, and all libraries are same as when this was prebound, then no need to bind
2149 if ( this->isPrebindable() && this->allDependentLibrariesAsWhenPreBound() && (this->getSlide() == 0) ) {
2150 // allow environment variables to disable prebinding
2151 if ( context.bindFlat )
2152 return false;
2153 switch ( context.prebindUsage ) {
2154 case kUseAllPrebinding:
2155 return true;
2156 case kUseSplitSegPrebinding:
2157 return this->fIsSplitSeg;
2158 case kUseAllButAppPredbinding:
2159 return (this != context.mainExecutable);
2160 case kUseNoPrebinding:
2161 return false;
2162 }
2163 }
2164 return false;
2165 }
2166
2167 void ImageLoaderMachO::doBind(const LinkContext& context, BindingLaziness bindness)
2168 {
2169 // set dyld entry points in image
2170 this->setupLazyPointerHandler();
2171
2172 // if prebound and loaded at prebound address, and all libraries are same as when this was prebound, then no need to bind
2173 // note: flat-namespace binaries need to be imports rebound (even if correctly prebound)
2174 if ( this->usablePrebinding(context) && this->usesTwoLevelNameSpace() ) {
2175 // if image has coalesced symbols, then these need to be rebound
2176 if ( this->needsCoalescing() ) {
2177 this->doBindExternalRelocations(context, true);
2178 this->doBindIndirectSymbolPointers(context, kLazyAndNonLazy, true);
2179 }
2180 // skip binding because prebound and prebinding not disabled
2181 return;
2182 }
2183
2184 // values bound by name are stored two different ways in mach-o
2185 switch (bindness) {
2186 case kNonLazyOnly:
2187 case kLazyAndNonLazy:
2188 // external relocations are used for data initialized to external symbols
2189 this->doBindExternalRelocations(context, false);
2190 break;
2191 case kLazyOnly:
2192 case kLazyOnlyNoDependents:
2193 break;
2194 }
2195 // "indirect symbols" are used for code references to external symbols
2196 this->doBindIndirectSymbolPointers(context, bindness, false);
2197 }
2198
2199
2200
2201 void ImageLoaderMachO::doImageInit(const LinkContext& context)
2202 {
2203 if ( fDashInit != NULL ) {
2204 Initializer func = (Initializer)(fDashInit->init_address + fSlide);
2205 if ( context.verboseInit )
2206 fprintf(stderr, "dyld: calling -init function 0x%p in %s\n", func, this->getPath());
2207 func(context.argc, context.argv, context.envp, context.apple);
2208 }
2209 }
2210
2211 void ImageLoaderMachO::doModInitFunctions(const LinkContext& context)
2212 {
2213 if ( fModInitSection != NULL ) {
2214 Initializer* inits = (Initializer*)(fModInitSection->addr + fSlide);
2215 const uint32_t count = fModInitSection->size / sizeof(uintptr_t);
2216 for (uint32_t i=0; i < count; ++i) {
2217 Initializer func = inits[i];
2218 if ( context.verboseInit )
2219 fprintf(stderr, "dyld: calling initializer function %p in %s\n", func, this->getPath());
2220 func(context.argc, context.argv, context.envp, context.apple);
2221 }
2222 }
2223 }
2224
2225
2226 void ImageLoaderMachO::doInitialization(const LinkContext& context)
2227 {
2228 // mach-o has -init and static initializers
2229 doImageInit(context);
2230 doModInitFunctions(context);
2231 }
2232
2233 bool ImageLoaderMachO::needsInitialization()
2234 {
2235 return ( (fDashInit != NULL) || (fModInitSection != NULL) );
2236 }
2237
2238
2239 bool ImageLoaderMachO::needsTermination()
2240 {
2241 return ( fModTermSection != NULL );
2242 }
2243
2244 bool ImageLoaderMachO::hasImageNotification()
2245 {
2246 return ( fImageNotifySection != NULL );
2247 }
2248
2249
2250 void ImageLoaderMachO::doTermination(const LinkContext& context)
2251 {
2252 if ( fModTermSection != NULL ) {
2253 Terminator* terms = (Terminator*)(fModTermSection->addr + fSlide);
2254 const uint32_t count = fModTermSection->size / sizeof(uintptr_t);
2255 for (uint32_t i=count; i > 0; --i) {
2256 Terminator func = terms[i-1];
2257 if ( context.verboseInit )
2258 fprintf(stderr, "dyld: calling terminaton function %p in %s\n", func, this->getPath());
2259 func();
2260 }
2261 }
2262 }
2263
2264 void ImageLoaderMachO::doNotification(enum dyld_image_mode mode, uint32_t infoCount, const struct dyld_image_info info[])
2265 {
2266 if ( fImageNotifySection != NULL ) {
2267 dyld_image_notifier* notes = (dyld_image_notifier*)(fImageNotifySection->addr + fSlide);
2268 const uint32_t count = fImageNotifySection->size / sizeof(uintptr_t);
2269 for (uint32_t i=count; i > 0; --i) {
2270 dyld_image_notifier func = notes[i-1];
2271 func(mode, infoCount, info);
2272 }
2273 }
2274 }
2275
2276 void ImageLoaderMachO::printStatistics(unsigned int imageCount)
2277 {
2278 ImageLoader::printStatistics(imageCount);
2279 fprintf(stderr, "total hinted binary tree searches: %d\n", fgHintedBinaryTreeSearchs);
2280 fprintf(stderr, "total unhinted binary tree searches: %d\n", fgUnhintedBinaryTreeSearchs);
2281
2282 #if LINKEDIT_USAGE_DEBUG
2283 fprintf(stderr, "linkedit pages accessed (%lu):\n", sLinkEditPageBuckets.size());
2284 #endif
2285 }
2286
2287 void ImageLoaderMachO::doPrebinding(const LinkContext& context, time_t timestamp, uint8_t* fileToPrebind)
2288 {
2289 // update __DATA segment
2290 this->applyPrebindingToDATA(fileToPrebind);
2291
2292 // update load commands
2293 this->applyPrebindingToLoadCommands(context, fileToPrebind, timestamp);
2294
2295 // update symbol table
2296 this->applyPrebindingToLinkEdit(context, fileToPrebind);
2297 }
2298
2299 void ImageLoaderMachO::applyPrebindingToDATA(uint8_t* fileToPrebind)
2300 {
2301 const unsigned int segmentCount = fSegments.size();
2302 for(unsigned int i=0; i < segmentCount; ++i) {
2303 SegmentMachO* seg = (SegmentMachO*)fSegments[i];
2304 if ( seg->writeable() ) {
2305 memcpy(&fileToPrebind[seg->fFileOffset], (void*)seg->getActualLoadAddress(), seg->fFileSize);
2306 }
2307 }
2308 }
2309
2310 void ImageLoaderMachO::applyPrebindingToLoadCommands(const LinkContext& context, uint8_t* fileToPrebind, time_t timestamp)
2311 {
2312 macho_header* mh = (macho_header*)fileToPrebind;
2313 const uint32_t cmd_count = mh->ncmds;
2314 const struct load_command* const cmds = (struct load_command*)&fileToPrebind[sizeof(macho_header)];
2315 const struct load_command* cmd = cmds;
2316 for (uint32_t i = 0; i < cmd_count; ++i) {
2317 switch (cmd->cmd) {
2318 case LC_LOAD_DYLIB:
2319 case LC_LOAD_WEAK_DYLIB:
2320 {
2321 // update each dylib load command with the timestamp of the target dylib
2322 struct dylib_command* dylib = (struct dylib_command*)cmd;
2323 const char* name = (char*)cmd + dylib->dylib.name.offset;
2324 for (const DependentLibrary* dl=fLibraries; dl < &fLibraries[fLibrariesCount]; dl++) {
2325 if (strcmp(dl->name, name) == 0 ) {
2326 // found matching DependentLibrary for this load command
2327 ImageLoaderMachO* targetImage = (ImageLoaderMachO*)(dl->image); // !!! assume only mach-o images are prebound
2328 if ( ! targetImage->isPrebindable() )
2329 throw "dependent dylib is not prebound";
2330 // if the target is currently being re-prebound then its timestamp will be the same as this one
2331 if ( ! targetImage->usablePrebinding(context) ) {
2332 dylib->dylib.timestamp = timestamp;
2333 }
2334 else {
2335 // otherwise dependent library is already correctly prebound, so use its checksum
2336 dylib->dylib.timestamp = targetImage->doGetLibraryInfo().checksum;
2337 }
2338 break;
2339 }
2340 }
2341 }
2342 break;
2343 case LC_ID_DYLIB:
2344 {
2345 // update the ID of this library with the new timestamp
2346 struct dylib_command* dylib = (struct dylib_command*)cmd;
2347 dylib->dylib.timestamp = timestamp;
2348 }
2349 break;
2350 case LC_SEGMENT_COMMAND:
2351 // if dylib was rebased, update segment commands
2352 if ( fSlide != 0 ) {
2353 struct macho_segment_command* seg = (struct macho_segment_command*)cmd;
2354 seg->vmaddr += fSlide;
2355 struct macho_section* const sectionsStart = (struct macho_section*)((char*)seg + sizeof(struct macho_segment_command));
2356 struct macho_section* const sectionsEnd = &sectionsStart[seg->nsects];
2357 for (struct macho_section* sect=sectionsStart; sect < sectionsEnd; ++sect) {
2358 sect->addr += fSlide;
2359 }
2360 }
2361 break;
2362 case LC_ROUTINES_COMMAND:
2363 // if dylib was rebased, update -init command
2364 if ( fSlide != 0 ) {
2365 struct macho_routines_command* routines = (struct macho_routines_command*)cmd;
2366 routines->init_address += fSlide;
2367 }
2368 break;
2369 }
2370 cmd = (const struct load_command*)(((char*)cmd)+cmd->cmdsize);
2371 }
2372 }
2373
2374 void ImageLoaderMachO::applyPrebindingToLinkEdit(const LinkContext& context, uint8_t* fileToPrebind)
2375 {
2376 // In prebound images, the n_value of the symbol table entry for is the prebound address
2377 // This is needed when prebinding can't be used, to back solve for any possible addend in non-lazy pointers
2378 const char* stringPool = NULL;
2379 struct macho_nlist* symbolTable = NULL;
2380 const struct dysymtab_command* dysymtab = NULL;
2381
2382 // get symbol table info
2383 macho_header* mh = (macho_header*)fileToPrebind;
2384 const uint32_t cmd_count = mh->ncmds;
2385 const struct load_command* const cmds = (struct load_command*)&fileToPrebind[sizeof(macho_header)];
2386 const struct load_command* cmd = cmds;
2387 for (uint32_t i = 0; i < cmd_count; ++i) {
2388 switch (cmd->cmd) {
2389 case LC_SYMTAB:
2390 {
2391 const struct symtab_command* symtab = (struct symtab_command*)cmd;
2392 stringPool = (const char*)&fileToPrebind[symtab->stroff];
2393 symbolTable = (struct macho_nlist*)(&fileToPrebind[symtab->symoff]);
2394 }
2395 break;
2396 case LC_DYSYMTAB:
2397 dysymtab = (struct dysymtab_command*)cmd;
2398 break;
2399 }
2400 cmd = (const struct load_command*)(((char*)cmd)+cmd->cmdsize);
2401 }
2402
2403 // walk all imports and re-resolve their n_value (needed incase prebinding is invalid)
2404 struct macho_nlist* lastImport = &symbolTable[dysymtab->iundefsym+dysymtab->nundefsym];
2405 for (struct macho_nlist* entry = &symbolTable[dysymtab->iundefsym]; entry < lastImport; ++entry) {
2406 ImageLoader* dummy;
2407 entry->n_value = this->resolveUndefined(context, entry, this->usesTwoLevelNameSpace(), &dummy);
2408 }
2409
2410 // walk all exports and slide their n_value
2411 struct macho_nlist* lastExport = &symbolTable[dysymtab->iextdefsym+dysymtab->nextdefsym];
2412 for (struct macho_nlist* entry = &symbolTable[dysymtab->iextdefsym]; entry < lastExport; ++entry) {
2413 if ( (entry->n_type & N_TYPE) == N_SECT )
2414 entry->n_value += fSlide;
2415 }
2416
2417 // walk all local symbols and slide their n_value
2418 struct macho_nlist* lastLocal = &symbolTable[dysymtab->ilocalsym+dysymtab->nlocalsym];
2419 for (struct macho_nlist* entry = &symbolTable[dysymtab->ilocalsym]; entry < lastLocal; ++entry) {
2420 if ( entry->n_sect != NO_SECT )
2421 entry->n_value += fSlide;
2422 }
2423
2424 // walk all local relocations and reset every PPC_RELOC_PB_LA_PTR r_value
2425 relocation_info* const relocsStart = (struct relocation_info*)(&fileToPrebind[dysymtab->locreloff]);
2426 relocation_info* const relocsEnd = &relocsStart[dysymtab->nlocrel];
2427 for (relocation_info* reloc=relocsStart; reloc < relocsEnd; ++reloc) {
2428 if ( (reloc->r_address & R_SCATTERED) != 0 ) {
2429 struct scattered_relocation_info* sreloc = (struct scattered_relocation_info*)reloc;
2430 if (sreloc->r_length == RELOC_SIZE) {
2431 switch(sreloc->r_type) {
2432 #if __ppc__ || __ppc64__
2433 case PPC_RELOC_PB_LA_PTR:
2434 #elif __i386__
2435 case GENERIC_RELOC_PB_LA_PTR:
2436 #else
2437 #error unknown architecture
2438 #endif
2439 sreloc->r_value += fSlide;
2440 break;
2441 }
2442 }
2443 }
2444 }
2445
2446 // if multi-module, fix up objc_addr (10.4 and later runtime does not use this, but we want to keep file checksum consistent)
2447 if ( dysymtab->nmodtab != 0 ) {
2448 dylib_module* const modulesStart = (struct dylib_module*)(&fileToPrebind[dysymtab->modtaboff]);
2449 dylib_module* const modulesEnd = &modulesStart[dysymtab->nmodtab];
2450 for (dylib_module* module=modulesStart; module < modulesEnd; ++module) {
2451 if ( module->objc_module_info_size != 0 ) {
2452 module->objc_module_info_addr += fSlide;
2453 }
2454 }
2455 }
2456 }
2457
2458 // file on disk has been reprebound, but we are still mapped to old file
2459 void ImageLoaderMachO::prebindUnmap(const LinkContext& context)
2460 {
2461 // this removes all mappings to the old file, so the kernel will unlink (delete) it.
2462 // We need to leave the load commands and __LINKEDIT in place
2463 for (std::vector<class Segment*>::iterator it=fSegments.begin(); it != fSegments.end(); ++it) {
2464 void* segmentAddress = (void*)((*it)->getActualLoadAddress());
2465 uintptr_t segmentSize = (*it)->getSize();
2466 //fprintf(stderr, "unmapping segment %s at %p for %s\n", (*it)->getName(), segmentAddress, this->getPath());
2467 // save load commands at beginning of __TEXT segment
2468 if ( segmentAddress == fMachOData ) {
2469 // typically load commands are one or two pages in size, so ok to alloc on stack
2470 uint32_t loadCmdSize = sizeof(macho_header) + ((macho_header*)fMachOData)->sizeofcmds;
2471 uint32_t loadCmdPages = (loadCmdSize+4095) & (-4096);
2472 uint8_t loadcommands[loadCmdPages];
2473 memcpy(loadcommands, fMachOData, loadCmdPages);
2474 // unmap whole __TEXT segment
2475 munmap((void*)(fMachOData), segmentSize);
2476 // allocate and copy back mach_header and load commands
2477 vm_address_t addr = (vm_address_t)fMachOData;
2478 int r2 = vm_allocate(mach_task_self(), &addr, loadCmdPages, false /*at this address*/);
2479 if ( r2 != 0 )
2480 fprintf(stderr, "prebindUnmap() vm_allocate for __TEXT %d failed\n", loadCmdPages);
2481 memcpy((void*)fMachOData, loadcommands, loadCmdPages);
2482 //fprintf(stderr, "copying back load commands to %p size=%u for %s\n", segmentAddress, loadCmdPages, this->getPath());
2483 }
2484 else if ( strcmp((*it)->getName(), "__LINKEDIT") == 0 ) {
2485 uint32_t linkEditSize = segmentSize;
2486 uint32_t linkEditPages = (linkEditSize+4095) & (-4096);
2487 void* linkEditTmp = malloc(linkEditPages);
2488 memcpy(linkEditTmp, segmentAddress, linkEditPages);
2489 // unmap whole __LINKEDIT segment
2490 munmap(segmentAddress, segmentSize);
2491 vm_address_t addr = (vm_address_t)segmentAddress;
2492 int r2 = vm_allocate(mach_task_self(), &addr, linkEditPages, false /*at this address*/);
2493 if ( r2 != 0 )
2494 fprintf(stderr, "prebindUnmap() vm_allocate for __LINKEDIT %d failed\n", linkEditPages);
2495 memcpy(segmentAddress, linkEditTmp, linkEditPages);
2496 //fprintf(stderr, "copying back __LINKEDIT to %p size=%u for %s\n", segmentAddress, linkEditPages, this->getPath());
2497 free(linkEditTmp);
2498 }
2499 else {
2500 // unmap any other segment
2501 munmap((void*)(segmentAddress), (*it)->getSize());
2502 }
2503 }
2504 }
2505
2506
2507
2508 SegmentMachO::SegmentMachO(const struct macho_segment_command* cmd, ImageLoaderMachO* image, const uint8_t* fileData)
2509 : fImage(image), fSize(cmd->vmsize), fFileSize(cmd->filesize), fFileOffset(cmd->fileoff), fPreferredLoadAddress(cmd->vmaddr),
2510 fVMProtection(cmd->initprot), fHasFixUps(false), fUnMapOnDestruction(false)
2511 {
2512 strncpy(fName, cmd->segname, 16);
2513 fName[16] = '\0';
2514 // scan sections for fix-up bit
2515 const struct macho_section* const sectionsStart = (struct macho_section*)((char*)cmd + sizeof(struct macho_segment_command));
2516 const struct macho_section* const sectionsEnd = &sectionsStart[cmd->nsects];
2517 for (const struct macho_section* sect=sectionsStart; sect < sectionsEnd; ++sect) {
2518 if ( (sect->flags & (S_ATTR_EXT_RELOC | S_ATTR_LOC_RELOC)) != 0 )
2519 fHasFixUps = true;
2520 }
2521 }
2522
2523 SegmentMachO::~SegmentMachO()
2524 {
2525 if ( fUnMapOnDestruction ) {
2526 //fprintf(stderr, "unmapping segment %s at 0x%08lX\n", getName(), getActualLoadAddress());
2527 munmap((void*)(this->getActualLoadAddress()), this->getSize());
2528 }
2529 }
2530
2531 const ImageLoader* SegmentMachO::getImage()
2532 {
2533 return fImage;
2534 }
2535
2536 const char* SegmentMachO::getName()
2537 {
2538 return fName;
2539 }
2540
2541 uintptr_t SegmentMachO::getSize()
2542 {
2543 return fSize;
2544 }
2545
2546 uintptr_t SegmentMachO::getFileSize()
2547 {
2548 return fFileSize;
2549 }
2550
2551 uintptr_t SegmentMachO::getFileOffset()
2552 {
2553 return fFileOffset;
2554 }
2555
2556 bool SegmentMachO::readable()
2557 {
2558 return ( (fVMProtection & VM_PROT_READ) != 0);
2559 }
2560
2561 bool SegmentMachO::writeable()
2562 {
2563 return ((fVMProtection & VM_PROT_WRITE) != 0);
2564 }
2565
2566 bool SegmentMachO::executable()
2567 {
2568 return ((fVMProtection & VM_PROT_EXECUTE) != 0);
2569 }
2570
2571 bool SegmentMachO::unaccessible()
2572 {
2573 return (fVMProtection == 0);
2574 }
2575
2576 bool SegmentMachO::hasFixUps()
2577 {
2578 return fHasFixUps;
2579 }
2580
2581 uintptr_t SegmentMachO::getActualLoadAddress()
2582 {
2583 return fPreferredLoadAddress + fImage->fSlide;
2584 }
2585
2586 uintptr_t SegmentMachO::getPreferredLoadAddress()
2587 {
2588 return fPreferredLoadAddress;
2589 }
2590
2591 bool SegmentMachO::hasPreferredLoadAddress()
2592 {
2593 return (fPreferredLoadAddress != 0);
2594 }
2595
2596 void SegmentMachO::setUnMapWhenDestructed(bool unmap)
2597 {
2598 fUnMapOnDestruction = unmap;
2599 }
2600
2601 static uint32_t *buildCRCTable(void)
2602 {
2603 uint32_t *table = new uint32_t[256];
2604 uint32_t p = 0xedb88320UL; // standard CRC-32 polynomial
2605
2606 for (unsigned int i = 0; i < 256; i++) {
2607 uint32_t c = i;
2608 for (unsigned int j = 0; j < 8; j++) {
2609 if ( c & 1 ) c = p ^ (c >> 1);
2610 else c = c >> 1;
2611 }
2612 table[i] = c;
2613 }
2614
2615 return table;
2616 }
2617
2618 uint32_t SegmentMachO::crc32()
2619 {
2620 if ( !readable() ) return 0;
2621
2622 static uint32_t *crcTable = NULL;
2623 if ( !crcTable ) crcTable = buildCRCTable();
2624
2625 uint32_t crc = ~(uint32_t)0;
2626 uint8_t *p = (uint8_t *)getActualLoadAddress();
2627 uint8_t *end = p + getSize();
2628 while ( p < end ) {
2629 crc = crcTable[(crc & 0xff) ^ (*p++)] ^ (crc >> 8);
2630 }
2631 return crc ^ ~(uint32_t)0;
2632 }
2633
2634
2635
2636
2637
mirror of dyld