[217] | 1 | /*******************************************************************************
|
---|
| 2 | mach_inject.c
|
---|
| 3 | Copyright (c) 2003-2005 Jonathan 'Wolf' Rentzsch: <http://rentzsch.com>
|
---|
| 4 | Some rights reserved: <http://creativecommons.org/licenses/by/2.0/>
|
---|
[153] | 5 |
|
---|
[217] | 6 | ***************************************************************************/
|
---|
[153] | 7 |
|
---|
| 8 | #include "mach_inject.h"
|
---|
| 9 |
|
---|
[217] | 10 | #include <mach-o/dyld.h>
|
---|
| 11 | #include <mach-o/getsect.h>
|
---|
| 12 | #include <mach/mach.h>
|
---|
| 13 | #include <sys/stat.h>
|
---|
| 14 | #include <sys/errno.h>
|
---|
| 15 | #include <assert.h>
|
---|
| 16 | #include <stdlib.h> // for malloc()
|
---|
| 17 | #include <stdio.h> // for printf()
|
---|
| 18 | #include <mach-o/fat.h> // for fat structure decoding
|
---|
| 19 | #include <mach-o/arch.h> // to know which is local arch
|
---|
| 20 | #include <fcntl.h> // for open/close
|
---|
| 21 | // for mmap()
|
---|
| 22 | #include <sys/types.h>
|
---|
| 23 | #include <sys/mman.h>
|
---|
[153] | 24 |
|
---|
[217] | 25 | #ifndef COMPILE_TIME_ASSERT( exp )
|
---|
[153] | 26 | #define COMPILE_TIME_ASSERT( exp ) { switch (0) { case 0: case (exp):; } }
|
---|
| 27 | #endif
|
---|
[217] | 28 | #define ASSERT_CAST( CAST_TO, CAST_FROM ) \
|
---|
| 29 | COMPILE_TIME_ASSERT( sizeof(CAST_TO)==sizeof(CAST_FROM) )
|
---|
[153] | 30 |
|
---|
[217] | 31 | #if defined(__i386__)
|
---|
| 32 | void* fixedUpImageFromImage (
|
---|
| 33 | const void *image,
|
---|
| 34 | unsigned long imageSize,
|
---|
| 35 | unsigned int jumpTableOffset,
|
---|
| 36 | unsigned int jumpTableSize,
|
---|
| 37 | ptrdiff_t fixUpOffset);
|
---|
| 38 | #endif /* __i386__ */
|
---|
| 39 |
|
---|
| 40 | /*******************************************************************************
|
---|
[153] | 41 | *
|
---|
| 42 | * Interface
|
---|
| 43 | *
|
---|
[217] | 44 | *******************************************************************************/
|
---|
[153] | 45 | #pragma mark -
|
---|
| 46 | #pragma mark (Interface)
|
---|
| 47 |
|
---|
| 48 | mach_error_t
|
---|
| 49 | mach_inject(
|
---|
| 50 | const mach_inject_entry threadEntry,
|
---|
| 51 | const void *paramBlock,
|
---|
| 52 | size_t paramSize,
|
---|
| 53 | pid_t targetProcess,
|
---|
[217] | 54 | vm_size_t stackSize )
|
---|
| 55 | {
|
---|
| 56 | assert( threadEntry );
|
---|
| 57 | assert( targetProcess > 0 );
|
---|
| 58 | assert( stackSize == 0 || stackSize > 1024 );
|
---|
[153] | 59 |
|
---|
| 60 | // Find the image.
|
---|
| 61 | const void *image;
|
---|
| 62 | unsigned long imageSize;
|
---|
[217] | 63 | unsigned int jumpTableOffset;
|
---|
| 64 | unsigned int jumpTableSize;
|
---|
| 65 | mach_error_t err = machImageForPointer( threadEntry, &image, &imageSize, &jumpTableOffset, &jumpTableSize );
|
---|
| 66 |
|
---|
[153] | 67 | // Initialize stackSize to default if requested.
|
---|
| 68 | if( stackSize == 0 )
|
---|
[217] | 69 | /** @bug
|
---|
| 70 | We only want an 8K default, fix the plop-in-the-middle code below.
|
---|
| 71 | */
|
---|
[153] | 72 | stackSize = 16 * 1024;
|
---|
| 73 |
|
---|
| 74 | // Convert PID to Mach Task ref.
|
---|
| 75 | mach_port_t remoteTask = 0;
|
---|
[217] | 76 | if( !err ) {
|
---|
[153] | 77 | err = task_for_pid( mach_task_self(), targetProcess, &remoteTask );
|
---|
[217] | 78 | #if defined(__i386__)
|
---|
| 79 | if (err == 5) fprintf(stderr, "Could not access task for pid %d. You probably need to add user to procmod group\n", targetProcess);
|
---|
| 80 | #endif
|
---|
| 81 | }
|
---|
[153] | 82 |
|
---|
[217] | 83 | /** @todo
|
---|
| 84 | Would be nice to just allocate one block for both the remote stack
|
---|
| 85 | *and* the remoteCode (including the parameter data block once that's
|
---|
| 86 | written.
|
---|
[153] | 87 | */
|
---|
| 88 |
|
---|
| 89 | // Allocate the remoteStack.
|
---|
[217] | 90 | vm_address_t remoteStack = (vm_address_t)NULL;
|
---|
[153] | 91 | if( !err )
|
---|
| 92 | err = vm_allocate( remoteTask, &remoteStack, stackSize, 1 );
|
---|
| 93 |
|
---|
| 94 | // Allocate the code.
|
---|
[217] | 95 | vm_address_t remoteCode = (vm_address_t)NULL;
|
---|
[153] | 96 | if( !err )
|
---|
| 97 | err = vm_allocate( remoteTask, &remoteCode, imageSize, 1 );
|
---|
| 98 | if( !err ) {
|
---|
| 99 | ASSERT_CAST( pointer_t, image );
|
---|
[217] | 100 | #if defined (__ppc__) || defined (__ppc64__)
|
---|
[153] | 101 | err = vm_write( remoteTask, remoteCode, (pointer_t) image, imageSize );
|
---|
[217] | 102 | #elif defined (__i386__)
|
---|
| 103 | // on intel, jump table use relative jump instructions (jmp), which means
|
---|
| 104 | // the offset needs to be corrected. We thus copy the image and fix the offset by hand.
|
---|
| 105 | ptrdiff_t fixUpOffset = (ptrdiff_t) (image - remoteCode);
|
---|
| 106 | void * fixedUpImage = fixedUpImageFromImage(image, imageSize, jumpTableOffset, jumpTableSize, fixUpOffset);
|
---|
| 107 | err = vm_write( remoteTask, remoteCode, (pointer_t) fixedUpImage, imageSize );
|
---|
| 108 | free(fixedUpImage);
|
---|
| 109 | #endif
|
---|
[153] | 110 | }
|
---|
| 111 |
|
---|
| 112 | // Allocate the paramBlock if specified.
|
---|
[217] | 113 | vm_address_t remoteParamBlock = (vm_address_t)NULL;
|
---|
[153] | 114 | if( !err && paramBlock != NULL && paramSize ) {
|
---|
| 115 | err = vm_allocate( remoteTask, &remoteParamBlock, paramSize, 1 );
|
---|
| 116 | if( !err ) {
|
---|
| 117 | ASSERT_CAST( pointer_t, paramBlock );
|
---|
[217] | 118 | err = vm_write( remoteTask, remoteParamBlock,
|
---|
| 119 | (pointer_t) paramBlock, paramSize );
|
---|
[153] | 120 | }
|
---|
| 121 | }
|
---|
| 122 |
|
---|
| 123 | // Calculate offsets.
|
---|
| 124 | ptrdiff_t threadEntryOffset, imageOffset;
|
---|
| 125 | if( !err ) {
|
---|
[217] | 126 | //assert( (void*)threadEntry >= image && (void*)threadEntry <= (image+imageSize) );
|
---|
[153] | 127 | ASSERT_CAST( void*, threadEntry );
|
---|
[217] | 128 | threadEntryOffset = ((void*) threadEntry) - image;
|
---|
[153] | 129 |
|
---|
| 130 | ASSERT_CAST( void*, remoteCode );
|
---|
[217] | 131 | imageOffset = ((void*) remoteCode) - image;
|
---|
[153] | 132 | }
|
---|
| 133 |
|
---|
| 134 | // Allocate the thread.
|
---|
| 135 | thread_act_t remoteThread;
|
---|
[217] | 136 | #if defined (__ppc__) || defined (__ppc64__)
|
---|
[153] | 137 | if( !err ) {
|
---|
| 138 | ppc_thread_state_t remoteThreadState;
|
---|
| 139 |
|
---|
[217] | 140 | /** @bug
|
---|
| 141 | Stack math should be more sophisticated than this (ala redzone).
|
---|
| 142 | */
|
---|
[153] | 143 | remoteStack += stackSize / 2;
|
---|
| 144 |
|
---|
| 145 | bzero( &remoteThreadState, sizeof(remoteThreadState) );
|
---|
| 146 |
|
---|
| 147 | ASSERT_CAST( unsigned int, remoteCode );
|
---|
| 148 | remoteThreadState.srr0 = (unsigned int) remoteCode;
|
---|
| 149 | remoteThreadState.srr0 += threadEntryOffset;
|
---|
| 150 | assert( remoteThreadState.srr0 < (remoteCode + imageSize) );
|
---|
| 151 |
|
---|
| 152 | ASSERT_CAST( unsigned int, remoteStack );
|
---|
| 153 | remoteThreadState.r1 = (unsigned int) remoteStack;
|
---|
| 154 |
|
---|
| 155 | ASSERT_CAST( unsigned int, imageOffset );
|
---|
| 156 | remoteThreadState.r3 = (unsigned int) imageOffset;
|
---|
| 157 |
|
---|
| 158 | ASSERT_CAST( unsigned int, remoteParamBlock );
|
---|
| 159 | remoteThreadState.r4 = (unsigned int) remoteParamBlock;
|
---|
| 160 |
|
---|
| 161 | ASSERT_CAST( unsigned int, paramSize );
|
---|
| 162 | remoteThreadState.r5 = (unsigned int) paramSize;
|
---|
| 163 |
|
---|
| 164 | ASSERT_CAST( unsigned int, 0xDEADBEEF );
|
---|
| 165 | remoteThreadState.lr = (unsigned int) 0xDEADBEEF;
|
---|
| 166 |
|
---|
[217] | 167 | #if 0
|
---|
[153] | 168 | printf( "remoteCode start: %p\n", (void*) remoteCode );
|
---|
| 169 | printf( "remoteCode size: %ld\n", imageSize );
|
---|
| 170 | printf( "remoteCode pc: %p\n", (void*) remoteThreadState.srr0 );
|
---|
[217] | 171 | printf( "remoteCode end: %p\n",
|
---|
| 172 | (void*) (((char*)remoteCode)+imageSize) );
|
---|
[153] | 173 | fflush(0);
|
---|
[217] | 174 | #endif
|
---|
[153] | 175 |
|
---|
| 176 | err = thread_create_running( remoteTask, PPC_THREAD_STATE,
|
---|
| 177 | (thread_state_t) &remoteThreadState, PPC_THREAD_STATE_COUNT,
|
---|
| 178 | &remoteThread );
|
---|
| 179 | }
|
---|
[217] | 180 | #elif defined (__i386__)
|
---|
| 181 | if( !err ) {
|
---|
| 182 |
|
---|
| 183 | i386_thread_state_t remoteThreadState;
|
---|
| 184 | bzero( &remoteThreadState, sizeof(remoteThreadState) );
|
---|
| 185 |
|
---|
| 186 | vm_address_t dummy_thread_struct = remoteStack;
|
---|
| 187 | remoteStack += (stackSize / 2); // this is the real stack
|
---|
| 188 | // (*) increase the stack, since we're simulating a CALL instruction, which normally pushes return address on the stack
|
---|
| 189 | remoteStack -= 4;
|
---|
| 190 |
|
---|
| 191 | #define PARAM_COUNT 4
|
---|
| 192 | #define STACK_CONTENTS_SIZE ((1+PARAM_COUNT) * sizeof(unsigned int))
|
---|
| 193 | unsigned int stackContents[1 + PARAM_COUNT]; // 1 for the return address and 1 for each param
|
---|
| 194 | // first entry is return address (see above *)
|
---|
| 195 | stackContents[0] = 0xDEADBEEF; // invalid return address.
|
---|
| 196 | // then we push function parameters one by one.
|
---|
| 197 | stackContents[1] = imageOffset;
|
---|
| 198 | stackContents[2] = remoteParamBlock;
|
---|
| 199 | stackContents[3] = paramSize;
|
---|
| 200 | // We use the remote stack we allocated as the fake thread struct. We should probably use a dedicated memory zone.
|
---|
| 201 | // We don't fill it with 0, vm_allocate did it for us
|
---|
| 202 | stackContents[4] = dummy_thread_struct;
|
---|
| 203 |
|
---|
| 204 | // push stackContents
|
---|
| 205 | err = vm_write( remoteTask, remoteStack,
|
---|
| 206 | (pointer_t) stackContents, STACK_CONTENTS_SIZE);
|
---|
| 207 |
|
---|
| 208 | // set remote Program Counter
|
---|
| 209 | remoteThreadState.eip = (unsigned int) (remoteCode);
|
---|
| 210 | remoteThreadState.eip += threadEntryOffset;
|
---|
| 211 |
|
---|
| 212 | // set remote Stack Pointer
|
---|
| 213 | ASSERT_CAST( unsigned int, remoteStack );
|
---|
| 214 | remoteThreadState.esp = (unsigned int) remoteStack;
|
---|
| 215 |
|
---|
| 216 | #if 0
|
---|
| 217 | printf( "remoteCode start: %p\n", (void*) remoteCode );
|
---|
| 218 | printf( "remoteCode size: %ld\n", imageSize );
|
---|
| 219 | printf( "remoteCode pc: %p\n", (void*) remoteThreadState.eip );
|
---|
| 220 | printf( "remoteCode end: %p\n",
|
---|
| 221 | (void*) (((char*)remoteCode)+imageSize) );
|
---|
| 222 | fflush(0);
|
---|
| 223 | #endif
|
---|
| 224 | // create thread and launch it
|
---|
| 225 | err = thread_create_running( remoteTask, i386_THREAD_STATE,
|
---|
| 226 | (thread_state_t) &remoteThreadState, i386_THREAD_STATE_COUNT,
|
---|
| 227 | &remoteThread );
|
---|
| 228 | }
|
---|
| 229 | #else
|
---|
| 230 | #error architecture not supported
|
---|
| 231 | #endif
|
---|
[153] | 232 |
|
---|
| 233 | if( err ) {
|
---|
| 234 | if( remoteParamBlock )
|
---|
| 235 | vm_deallocate( remoteTask, remoteParamBlock, paramSize );
|
---|
| 236 | if( remoteCode )
|
---|
| 237 | vm_deallocate( remoteTask, remoteCode, imageSize );
|
---|
| 238 | if( remoteStack )
|
---|
| 239 | vm_deallocate( remoteTask, remoteStack, stackSize );
|
---|
| 240 | }
|
---|
| 241 |
|
---|
| 242 | return err;
|
---|
| 243 | }
|
---|
| 244 |
|
---|
| 245 | mach_error_t
|
---|
| 246 | machImageForPointer(
|
---|
| 247 | const void *pointer,
|
---|
| 248 | const void **image,
|
---|
[217] | 249 | unsigned long *size,
|
---|
| 250 | unsigned int *jumpTableOffset,
|
---|
| 251 | unsigned int *jumpTableSize )
|
---|
| 252 | {
|
---|
| 253 | assert( pointer );
|
---|
| 254 | assert( image );
|
---|
| 255 | assert( size );
|
---|
[153] | 256 |
|
---|
| 257 | unsigned long p = (unsigned long) pointer;
|
---|
| 258 |
|
---|
[217] | 259 | if (jumpTableOffset && jumpTableSize) {
|
---|
| 260 | *jumpTableOffset = 0;
|
---|
| 261 | *jumpTableSize = 0;
|
---|
| 262 | }
|
---|
| 263 |
|
---|
[153] | 264 | unsigned long imageIndex, imageCount = _dyld_image_count();
|
---|
| 265 | for( imageIndex = 0; imageIndex < imageCount; imageIndex++ ) {
|
---|
[217] | 266 | const struct mach_header *header = _dyld_get_image_header( imageIndex );
|
---|
| 267 | const struct section *section = getsectbynamefromheader( header,
|
---|
| 268 | SEG_TEXT,
|
---|
| 269 | SECT_TEXT );
|
---|
[153] | 270 | long start = section->addr + _dyld_get_image_vmaddr_slide( imageIndex );
|
---|
| 271 | long stop = start + section->size;
|
---|
| 272 | if( p >= start && p <= stop ) {
|
---|
[217] | 273 | // It is truely insane we have to stat() the file system in order
|
---|
| 274 | // to discover the size of an in-memory data structure.
|
---|
| 275 | const char *imageName = _dyld_get_image_name( imageIndex );
|
---|
| 276 | assert( imageName );
|
---|
[153] | 277 | struct stat sb;
|
---|
| 278 | if( stat( imageName, &sb ) )
|
---|
| 279 | return unix_err( errno );
|
---|
| 280 | if( image ) {
|
---|
| 281 | ASSERT_CAST( void*, header );
|
---|
| 282 | *image = (void*) header;
|
---|
| 283 | }
|
---|
| 284 | if( size ) {
|
---|
| 285 | ;//assertUInt32( st_size );
|
---|
| 286 | *size = sb.st_size;
|
---|
[217] | 287 |
|
---|
| 288 | // needed for Universal binaries. Check if file is fat and get image size from there.
|
---|
| 289 | int fd = open (imageName, O_RDONLY);
|
---|
| 290 | size_t mapSize = *size;
|
---|
| 291 | char * fileImage = mmap (NULL, mapSize, PROT_READ, MAP_FILE, fd, 0);
|
---|
| 292 |
|
---|
| 293 | struct fat_header* fatHeader = (struct fat_header *)fileImage;
|
---|
| 294 | if (fatHeader->magic == OSSwapBigToHostInt32(FAT_MAGIC)) {
|
---|
| 295 | //printf("This is a fat binary\n");
|
---|
| 296 | uint32_t archCount = OSSwapBigToHostInt32(fatHeader->nfat_arch);
|
---|
| 297 |
|
---|
| 298 | NXArchInfo const *localArchInfo = NXGetLocalArchInfo();
|
---|
| 299 |
|
---|
| 300 | struct fat_arch* arch = (struct fat_arch *)(fileImage + sizeof(struct fat_header));
|
---|
| 301 | struct fat_arch* matchingArch = NULL;
|
---|
| 302 |
|
---|
| 303 | int archIndex = 0;
|
---|
| 304 | for (archIndex = 0; archIndex < archCount; archIndex++) {
|
---|
| 305 | cpu_type_t cpuType = OSSwapBigToHostInt32(arch[archIndex].cputype);
|
---|
| 306 | cpu_subtype_t cpuSubtype = OSSwapBigToHostInt32(arch[archIndex].cpusubtype);
|
---|
| 307 |
|
---|
| 308 | if (localArchInfo->cputype == cpuType) {
|
---|
| 309 | matchingArch = arch + archIndex;
|
---|
| 310 | if (localArchInfo->cpusubtype == cpuSubtype) break;
|
---|
| 311 | }
|
---|
| 312 | }
|
---|
| 313 |
|
---|
| 314 | if (matchingArch) {
|
---|
| 315 | *size = OSSwapBigToHostInt32(matchingArch->size);
|
---|
| 316 | //printf ("found arch-specific size : %p\n", *size);
|
---|
| 317 | }
|
---|
| 318 | }
|
---|
| 319 |
|
---|
| 320 | munmap (fileImage, mapSize);
|
---|
| 321 | close (fd);
|
---|
[153] | 322 | }
|
---|
[217] | 323 | if (jumpTableOffset && jumpTableSize) {
|
---|
| 324 | const struct section * jumpTableSection = getsectbynamefromheader( header,
|
---|
| 325 | "__IMPORT",
|
---|
| 326 | "__jump_table" );
|
---|
| 327 | if (jumpTableSection) {
|
---|
| 328 | *jumpTableOffset = jumpTableSection->offset;
|
---|
| 329 | *jumpTableSize = jumpTableSection->size;
|
---|
| 330 | }
|
---|
| 331 | }
|
---|
[153] | 332 | return err_none;
|
---|
| 333 | }
|
---|
| 334 | }
|
---|
| 335 |
|
---|
| 336 | return err_threadEntry_image_not_found;
|
---|
[217] | 337 | }
|
---|
| 338 |
|
---|
| 339 | #if defined(__i386__)
|
---|
| 340 | void* fixedUpImageFromImage (
|
---|
| 341 | const void *image,
|
---|
| 342 | unsigned long imageSize,
|
---|
| 343 | unsigned int jumpTableOffset,
|
---|
| 344 | unsigned int jumpTableSize,
|
---|
| 345 | ptrdiff_t fixUpOffset)
|
---|
| 346 | {
|
---|
| 347 | // first copy the full image
|
---|
| 348 | void *fixedUpImage = (void *) malloc ((size_t)imageSize);
|
---|
| 349 | bcopy(image, fixedUpImage, imageSize);
|
---|
| 350 |
|
---|
| 351 | // address of jump table in copied image
|
---|
| 352 | void *jumpTable = fixedUpImage + jumpTableOffset;
|
---|
| 353 | // each JMP instruction is 5 bytes (E9 xx xx xx xx) where E9 is the opcode for JMP
|
---|
| 354 | int jumpTableCount = jumpTableSize / 5;
|
---|
| 355 |
|
---|
| 356 | // skip first "E9"
|
---|
| 357 | jumpTable++;
|
---|
| 358 |
|
---|
| 359 | int entry=0;
|
---|
| 360 | for (entry = 0; entry < jumpTableCount; entry++) {
|
---|
| 361 | unsigned int jmpValue = *((unsigned int *)jumpTable);
|
---|
| 362 | jmpValue += fixUpOffset;
|
---|
| 363 | *((unsigned int *)jumpTable) = jmpValue;
|
---|
| 364 | jumpTable+=5;
|
---|
| 365 | }
|
---|
| 366 |
|
---|
| 367 | return fixedUpImage;
|
---|
| 368 | }
|
---|
| 369 | #endif /* __i386__ */ |
---|