1 //===-- interception_linux.cc -----------------------------------*- C++ -*-===// 2 // 3 // This file is distributed under the University of Illinois Open Source 4 // License. See LICENSE.TXT for details. 5 // 6 //===----------------------------------------------------------------------===// 7 // 8 // This file is a part of AddressSanitizer, an address sanity checker. 9 // 10 // Windows-specific interception methods. 11 // 12 // This file is implementing several hooking techniques to intercept calls 13 // to functions. The hooks are dynamically installed by modifying the assembly 14 // code. 15 // 16 // The hooking techniques are making assumptions on the way the code is 17 // generated and are safe under these assumptions. 18 // 19 // On 64-bit architecture, there is no direct 64-bit jump instruction. To allow 20 // arbitrary branching on the whole memory space, the notion of trampoline 21 // region is used. A trampoline region is a memory space withing 2G boundary 22 // where it is safe to add custom assembly code to build 64-bit jumps. 23 // 24 // Hooking techniques 25 // ================== 26 // 27 // 1) Detour 28 // 29 // The Detour hooking technique is assuming the presence of an header with 30 // padding and an overridable 2-bytes nop instruction (mov edi, edi). The 31 // nop instruction can safely be replaced by a 2-bytes jump without any need 32 // to save the instruction. A jump to the target is encoded in the function 33 // header and the nop instruction is replaced by a short jump to the header. 34 // 35 // head: 5 x nop head: jmp <hook> 36 // func: mov edi, edi --> func: jmp short <head> 37 // [...] real: [...] 38 // 39 // This technique is only implemented on 32-bit architecture. 40 // Most of the time, Windows API are hookable with the detour technique. 41 // 42 // 2) Redirect Jump 43 // 44 // The redirect jump is applicable when the first instruction is a direct 45 // jump. The instruction is replaced by jump to the hook. 46 // 47 // func: jmp <label> --> func: jmp <hook> 48 // 49 // On an 64-bit architecture, a trampoline is inserted. 50 // 51 // func: jmp <label> --> func: jmp <tramp> 52 // [...] 53 // 54 // [trampoline] 55 // tramp: jmp QWORD [addr] 56 // addr: .bytes <hook> 57 // 58 // Note: <real> is equilavent to <label>. 59 // 60 // 3) HotPatch 61 // 62 // The HotPatch hooking is assuming the presence of an header with padding 63 // and a first instruction with at least 2-bytes. 64 // 65 // The reason to enforce the 2-bytes limitation is to provide the minimal 66 // space to encode a short jump. HotPatch technique is only rewriting one 67 // instruction to avoid breaking a sequence of instructions containing a 68 // branching target. 69 // 70 // Assumptions are enforced by MSVC compiler by using the /HOTPATCH flag. 71 // see: https://msdn.microsoft.com/en-us/library/ms173507.aspx 72 // Default padding length is 5 bytes in 32-bits and 6 bytes in 64-bits. 73 // 74 // head: 5 x nop head: jmp <hook> 75 // func: <instr> --> func: jmp short <head> 76 // [...] body: [...] 77 // 78 // [trampoline] 79 // real: <instr> 80 // jmp <body> 81 // 82 // On an 64-bit architecture: 83 // 84 // head: 6 x nop head: jmp QWORD [addr1] 85 // func: <instr> --> func: jmp short <head> 86 // [...] body: [...] 87 // 88 // [trampoline] 89 // addr1: .bytes <hook> 90 // real: <instr> 91 // jmp QWORD [addr2] 92 // addr2: .bytes <body> 93 // 94 // 4) Trampoline 95 // 96 // The Trampoline hooking technique is the most aggressive one. It is 97 // assuming that there is a sequence of instructions that can be safely 98 // replaced by a jump (enough room and no incoming branches). 99 // 100 // Unfortunately, these assumptions can't be safely presumed and code may 101 // be broken after hooking. 102 // 103 // func: <instr> --> func: jmp <hook> 104 // <instr> 105 // [...] body: [...] 106 // 107 // [trampoline] 108 // real: <instr> 109 // <instr> 110 // jmp <body> 111 // 112 // On an 64-bit architecture: 113 // 114 // func: <instr> --> func: jmp QWORD [addr1] 115 // <instr> 116 // [...] body: [...] 117 // 118 // [trampoline] 119 // addr1: .bytes <hook> 120 // real: <instr> 121 // <instr> 122 // jmp QWORD [addr2] 123 // addr2: .bytes <body> 124 //===----------------------------------------------------------------------===// 125 126 #include "interception.h" 127 128 #if SANITIZER_WINDOWS 129 #include "sanitizer_common/sanitizer_platform.h" 130 #define WIN32_LEAN_AND_MEAN 131 #include <windows.h> 132 133 namespace __interception { 134 135 static const int kAddressLength = FIRST_32_SECOND_64(4, 8); 136 static const int kJumpInstructionLength = 5; 137 static const int kShortJumpInstructionLength = 2; 138 static const int kIndirectJumpInstructionLength = 6; 139 static const int kBranchLength = 140 FIRST_32_SECOND_64(kJumpInstructionLength, kIndirectJumpInstructionLength); 141 static const int kDirectBranchLength = kBranchLength + kAddressLength; 142 143 static void InterceptionFailed() { 144 // Do we have a good way to abort with an error message here? 145 __debugbreak(); 146 } 147 148 static bool DistanceIsWithin2Gig(uptr from, uptr target) { 149 #if SANITIZER_WINDOWS64 150 if (from < target) 151 return target - from <= (uptr)0x7FFFFFFFU; 152 else 153 return from - target <= (uptr)0x80000000U; 154 #else 155 // In a 32-bit address space, the address calculation will wrap, so this check 156 // is unnecessary. 157 return true; 158 #endif 159 } 160 161 static uptr GetMmapGranularity() { 162 SYSTEM_INFO si; 163 GetSystemInfo(&si); 164 return si.dwAllocationGranularity; 165 } 166 167 static uptr RoundUpTo(uptr size, uptr boundary) { 168 return (size + boundary - 1) & ~(boundary - 1); 169 } 170 171 // FIXME: internal_str* and internal_mem* functions should be moved from the 172 // ASan sources into interception/. 173 174 static size_t _strlen(const char *str) { 175 const char* p = str; 176 while (*p != '\0') ++p; 177 return p - str; 178 } 179 180 static char* _strchr(char* str, char c) { 181 while (*str) { 182 if (*str == c) 183 return str; 184 ++str; 185 } 186 return nullptr; 187 } 188 189 static void _memset(void *p, int value, size_t sz) { 190 for (size_t i = 0; i < sz; ++i) 191 ((char*)p)[i] = (char)value; 192 } 193 194 static void _memcpy(void *dst, void *src, size_t sz) { 195 char *dst_c = (char*)dst, 196 *src_c = (char*)src; 197 for (size_t i = 0; i < sz; ++i) 198 dst_c[i] = src_c[i]; 199 } 200 201 static bool ChangeMemoryProtection( 202 uptr address, uptr size, DWORD *old_protection) { 203 return ::VirtualProtect((void*)address, size, 204 PAGE_EXECUTE_READWRITE, 205 old_protection) != FALSE; 206 } 207 208 static bool RestoreMemoryProtection( 209 uptr address, uptr size, DWORD old_protection) { 210 DWORD unused; 211 return ::VirtualProtect((void*)address, size, 212 old_protection, 213 &unused) != FALSE; 214 } 215 216 static bool IsMemoryPadding(uptr address, uptr size) { 217 u8* function = (u8*)address; 218 for (size_t i = 0; i < size; ++i) 219 if (function[i] != 0x90 && function[i] != 0xCC) 220 return false; 221 return true; 222 } 223 224 static const u8 kHintNop8Bytes[] = { 225 0x0F, 0x1F, 0x84, 0x00, 0x00, 0x00, 0x00, 0x00 226 }; 227 228 template<class T> 229 static bool FunctionHasPrefix(uptr address, const T &pattern) { 230 u8* function = (u8*)address - sizeof(pattern); 231 for (size_t i = 0; i < sizeof(pattern); ++i) 232 if (function[i] != pattern[i]) 233 return false; 234 return true; 235 } 236 237 static bool FunctionHasPadding(uptr address, uptr size) { 238 if (IsMemoryPadding(address - size, size)) 239 return true; 240 if (size <= sizeof(kHintNop8Bytes) && 241 FunctionHasPrefix(address, kHintNop8Bytes)) 242 return true; 243 return false; 244 } 245 246 static void WritePadding(uptr from, uptr size) { 247 _memset((void*)from, 0xCC, (size_t)size); 248 } 249 250 static void WriteJumpInstruction(uptr from, uptr target) { 251 if (!DistanceIsWithin2Gig(from + kJumpInstructionLength, target)) 252 InterceptionFailed(); 253 ptrdiff_t offset = target - from - kJumpInstructionLength; 254 *(u8*)from = 0xE9; 255 *(u32*)(from + 1) = offset; 256 } 257 258 static void WriteShortJumpInstruction(uptr from, uptr target) { 259 sptr offset = target - from - kShortJumpInstructionLength; 260 if (offset < -128 || offset > 127) 261 InterceptionFailed(); 262 *(u8*)from = 0xEB; 263 *(u8*)(from + 1) = (u8)offset; 264 } 265 266 #if SANITIZER_WINDOWS64 267 static void WriteIndirectJumpInstruction(uptr from, uptr indirect_target) { 268 // jmp [rip + <offset>] = FF 25 <offset> where <offset> is a relative 269 // offset. 270 // The offset is the distance from then end of the jump instruction to the 271 // memory location containing the targeted address. The displacement is still 272 // 32-bit in x64, so indirect_target must be located within +/- 2GB range. 273 int offset = indirect_target - from - kIndirectJumpInstructionLength; 274 if (!DistanceIsWithin2Gig(from + kIndirectJumpInstructionLength, 275 indirect_target)) { 276 InterceptionFailed(); 277 } 278 *(u16*)from = 0x25FF; 279 *(u32*)(from + 2) = offset; 280 } 281 #endif 282 283 static void WriteBranch( 284 uptr from, uptr indirect_target, uptr target) { 285 #if SANITIZER_WINDOWS64 286 WriteIndirectJumpInstruction(from, indirect_target); 287 *(u64*)indirect_target = target; 288 #else 289 (void)indirect_target; 290 WriteJumpInstruction(from, target); 291 #endif 292 } 293 294 static void WriteDirectBranch(uptr from, uptr target) { 295 #if SANITIZER_WINDOWS64 296 // Emit an indirect jump through immediately following bytes: 297 // jmp [rip + kBranchLength] 298 // .quad <target> 299 WriteBranch(from, from + kBranchLength, target); 300 #else 301 WriteJumpInstruction(from, target); 302 #endif 303 } 304 305 struct TrampolineMemoryRegion { 306 uptr content; 307 uptr allocated_size; 308 uptr max_size; 309 }; 310 311 static const uptr kTrampolineScanLimitRange = 1 << 31; // 2 gig 312 static const int kMaxTrampolineRegion = 1024; 313 static TrampolineMemoryRegion TrampolineRegions[kMaxTrampolineRegion]; 314 315 static void *AllocateTrampolineRegion(uptr image_address, size_t granularity) { 316 #if SANITIZER_WINDOWS64 317 uptr address = image_address; 318 uptr scanned = 0; 319 while (scanned < kTrampolineScanLimitRange) { 320 MEMORY_BASIC_INFORMATION info; 321 if (!::VirtualQuery((void*)address, &info, sizeof(info))) 322 return nullptr; 323 324 // Check whether a region can be allocated at |address|. 325 if (info.State == MEM_FREE && info.RegionSize >= granularity) { 326 void *page = ::VirtualAlloc((void*)RoundUpTo(address, granularity), 327 granularity, 328 MEM_RESERVE | MEM_COMMIT, 329 PAGE_EXECUTE_READWRITE); 330 return page; 331 } 332 333 // Move to the next region. 334 address = (uptr)info.BaseAddress + info.RegionSize; 335 scanned += info.RegionSize; 336 } 337 return nullptr; 338 #else 339 return ::VirtualAlloc(nullptr, 340 granularity, 341 MEM_RESERVE | MEM_COMMIT, 342 PAGE_EXECUTE_READWRITE); 343 #endif 344 } 345 346 // Used by unittests to release mapped memory space. 347 void TestOnlyReleaseTrampolineRegions() { 348 for (size_t bucket = 0; bucket < kMaxTrampolineRegion; ++bucket) { 349 TrampolineMemoryRegion *current = &TrampolineRegions[bucket]; 350 if (current->content == 0) 351 return; 352 ::VirtualFree((void*)current->content, 0, MEM_RELEASE); 353 current->content = 0; 354 } 355 } 356 357 static uptr AllocateMemoryForTrampoline(uptr image_address, size_t size) { 358 // Find a region within 2G with enough space to allocate |size| bytes. 359 TrampolineMemoryRegion *region = nullptr; 360 for (size_t bucket = 0; bucket < kMaxTrampolineRegion; ++bucket) { 361 TrampolineMemoryRegion* current = &TrampolineRegions[bucket]; 362 if (current->content == 0) { 363 // No valid region found, allocate a new region. 364 size_t bucket_size = GetMmapGranularity(); 365 void *content = AllocateTrampolineRegion(image_address, bucket_size); 366 if (content == nullptr) 367 return 0U; 368 369 current->content = (uptr)content; 370 current->allocated_size = 0; 371 current->max_size = bucket_size; 372 region = current; 373 break; 374 } else if (current->max_size - current->allocated_size > size) { 375 #if SANITIZER_WINDOWS64 376 // In 64-bits, the memory space must be allocated within 2G boundary. 377 uptr next_address = current->content + current->allocated_size; 378 if (next_address < image_address || 379 next_address - image_address >= 0x7FFF0000) 380 continue; 381 #endif 382 // The space can be allocated in the current region. 383 region = current; 384 break; 385 } 386 } 387 388 // Failed to find a region. 389 if (region == nullptr) 390 return 0U; 391 392 // Allocate the space in the current region. 393 uptr allocated_space = region->content + region->allocated_size; 394 region->allocated_size += size; 395 WritePadding(allocated_space, size); 396 397 return allocated_space; 398 } 399 400 // Returns 0 on error. 401 static size_t GetInstructionSize(uptr address, size_t* rel_offset = nullptr) { 402 switch (*(u64*)address) { 403 case 0x90909090909006EB: // stub: jmp over 6 x nop. 404 return 8; 405 } 406 407 switch (*(u8*)address) { 408 case 0x90: // 90 : nop 409 return 1; 410 411 case 0x50: // push eax / rax 412 case 0x51: // push ecx / rcx 413 case 0x52: // push edx / rdx 414 case 0x53: // push ebx / rbx 415 case 0x54: // push esp / rsp 416 case 0x55: // push ebp / rbp 417 case 0x56: // push esi / rsi 418 case 0x57: // push edi / rdi 419 case 0x5D: // pop ebp / rbp 420 return 1; 421 422 case 0x6A: // 6A XX = push XX 423 return 2; 424 425 case 0xb8: // b8 XX XX XX XX : mov eax, XX XX XX XX 426 case 0xB9: // b9 XX XX XX XX : mov ecx, XX XX XX XX 427 return 5; 428 429 // Cannot overwrite control-instruction. Return 0 to indicate failure. 430 case 0xE9: // E9 XX XX XX XX : jmp <label> 431 case 0xE8: // E8 XX XX XX XX : call <func> 432 case 0xC3: // C3 : ret 433 case 0xEB: // EB XX : jmp XX (short jump) 434 case 0x70: // 7Y YY : jy XX (short conditional jump) 435 case 0x71: 436 case 0x72: 437 case 0x73: 438 case 0x74: 439 case 0x75: 440 case 0x76: 441 case 0x77: 442 case 0x78: 443 case 0x79: 444 case 0x7A: 445 case 0x7B: 446 case 0x7C: 447 case 0x7D: 448 case 0x7E: 449 case 0x7F: 450 return 0; 451 } 452 453 switch (*(u16*)(address)) { 454 case 0x018A: // 8A 01 : mov al, byte ptr [ecx] 455 case 0xFF8B: // 8B FF : mov edi, edi 456 case 0xEC8B: // 8B EC : mov ebp, esp 457 case 0xc889: // 89 C8 : mov eax, ecx 458 case 0xC18B: // 8B C1 : mov eax, ecx 459 case 0xC033: // 33 C0 : xor eax, eax 460 case 0xC933: // 33 C9 : xor ecx, ecx 461 case 0xD233: // 33 D2 : xor edx, edx 462 return 2; 463 464 // Cannot overwrite control-instruction. Return 0 to indicate failure. 465 case 0x25FF: // FF 25 XX XX XX XX : jmp [XXXXXXXX] 466 return 0; 467 } 468 469 switch (0x00FFFFFF & *(u32*)address) { 470 case 0x24A48D: // 8D A4 24 XX XX XX XX : lea esp, [esp + XX XX XX XX] 471 return 7; 472 } 473 474 #if SANITIZER_WINDOWS64 475 switch (*(u8*)address) { 476 case 0xA1: // A1 XX XX XX XX XX XX XX XX : 477 // movabs eax, dword ptr ds:[XXXXXXXX] 478 return 9; 479 } 480 481 switch (*(u16*)address) { 482 case 0x5040: // push rax 483 case 0x5140: // push rcx 484 case 0x5240: // push rdx 485 case 0x5340: // push rbx 486 case 0x5440: // push rsp 487 case 0x5540: // push rbp 488 case 0x5640: // push rsi 489 case 0x5740: // push rdi 490 case 0x5441: // push r12 491 case 0x5541: // push r13 492 case 0x5641: // push r14 493 case 0x5741: // push r15 494 case 0x9066: // Two-byte NOP 495 return 2; 496 497 case 0x058B: // 8B 05 XX XX XX XX : mov eax, dword ptr [XX XX XX XX] 498 if (rel_offset) 499 *rel_offset = 2; 500 return 6; 501 } 502 503 switch (0x00FFFFFF & *(u32*)address) { 504 case 0xe58948: // 48 8b c4 : mov rbp, rsp 505 case 0xc18b48: // 48 8b c1 : mov rax, rcx 506 case 0xc48b48: // 48 8b c4 : mov rax, rsp 507 case 0xd9f748: // 48 f7 d9 : neg rcx 508 case 0xd12b48: // 48 2b d1 : sub rdx, rcx 509 case 0x07c1f6: // f6 c1 07 : test cl, 0x7 510 case 0xc98548: // 48 85 C9 : test rcx, rcx 511 case 0xc0854d: // 4d 85 c0 : test r8, r8 512 case 0xc2b60f: // 0f b6 c2 : movzx eax, dl 513 case 0xc03345: // 45 33 c0 : xor r8d, r8d 514 case 0xdb3345: // 45 33 DB : xor r11d, r11d 515 case 0xd98b4c: // 4c 8b d9 : mov r11, rcx 516 case 0xd28b4c: // 4c 8b d2 : mov r10, rdx 517 case 0xc98b4c: // 4C 8B C9 : mov r9, rcx 518 case 0xd2b60f: // 0f b6 d2 : movzx edx, dl 519 case 0xca2b48: // 48 2b ca : sub rcx, rdx 520 case 0x10b70f: // 0f b7 10 : movzx edx, WORD PTR [rax] 521 case 0xc00b4d: // 3d 0b c0 : or r8, r8 522 case 0xd18b48: // 48 8b d1 : mov rdx, rcx 523 case 0xdc8b4c: // 4c 8b dc : mov r11, rsp 524 case 0xd18b4c: // 4c 8b d1 : mov r10, rcx 525 return 3; 526 527 case 0xec8348: // 48 83 ec XX : sub rsp, XX 528 case 0xf88349: // 49 83 f8 XX : cmp r8, XX 529 case 0x588948: // 48 89 58 XX : mov QWORD PTR[rax + XX], rbx 530 return 4; 531 532 case 0xec8148: // 48 81 EC XX XX XX XX : sub rsp, XXXXXXXX 533 return 7; 534 535 case 0x058b48: // 48 8b 05 XX XX XX XX : 536 // mov rax, QWORD PTR [rip + XXXXXXXX] 537 case 0x25ff48: // 48 ff 25 XX XX XX XX : 538 // rex.W jmp QWORD PTR [rip + XXXXXXXX] 539 540 // Instructions having offset relative to 'rip' need offset adjustment. 541 if (rel_offset) 542 *rel_offset = 3; 543 return 7; 544 545 case 0x2444c7: // C7 44 24 XX YY YY YY YY 546 // mov dword ptr [rsp + XX], YYYYYYYY 547 return 8; 548 } 549 550 switch (*(u32*)(address)) { 551 case 0x24448b48: // 48 8b 44 24 XX : mov rax, QWORD ptr [rsp + XX] 552 case 0x246c8948: // 48 89 6C 24 XX : mov QWORD ptr [rsp + XX], rbp 553 case 0x245c8948: // 48 89 5c 24 XX : mov QWORD PTR [rsp + XX], rbx 554 case 0x24748948: // 48 89 74 24 XX : mov QWORD PTR [rsp + XX], rsi 555 case 0x244C8948: // 48 89 4C 24 XX : mov QWORD PTR [rsp + XX], rcx 556 return 5; 557 case 0x24648348: // 48 83 64 24 XX : and QWORD PTR [rsp + XX], YY 558 return 6; 559 } 560 561 #else 562 563 switch (*(u8*)address) { 564 case 0xA1: // A1 XX XX XX XX : mov eax, dword ptr ds:[XXXXXXXX] 565 return 5; 566 } 567 switch (*(u16*)address) { 568 case 0x458B: // 8B 45 XX : mov eax, dword ptr [ebp + XX] 569 case 0x5D8B: // 8B 5D XX : mov ebx, dword ptr [ebp + XX] 570 case 0x7D8B: // 8B 7D XX : mov edi, dword ptr [ebp + XX] 571 case 0xEC83: // 83 EC XX : sub esp, XX 572 case 0x75FF: // FF 75 XX : push dword ptr [ebp + XX] 573 return 3; 574 case 0xC1F7: // F7 C1 XX YY ZZ WW : test ecx, WWZZYYXX 575 case 0x25FF: // FF 25 XX YY ZZ WW : jmp dword ptr ds:[WWZZYYXX] 576 return 6; 577 case 0x3D83: // 83 3D XX YY ZZ WW TT : cmp TT, WWZZYYXX 578 return 7; 579 case 0x7D83: // 83 7D XX YY : cmp dword ptr [ebp + XX], YY 580 return 4; 581 } 582 583 switch (0x00FFFFFF & *(u32*)address) { 584 case 0x24448A: // 8A 44 24 XX : mov eal, dword ptr [esp + XX] 585 case 0x24448B: // 8B 44 24 XX : mov eax, dword ptr [esp + XX] 586 case 0x244C8B: // 8B 4C 24 XX : mov ecx, dword ptr [esp + XX] 587 case 0x24548B: // 8B 54 24 XX : mov edx, dword ptr [esp + XX] 588 case 0x24748B: // 8B 74 24 XX : mov esi, dword ptr [esp + XX] 589 case 0x247C8B: // 8B 7C 24 XX : mov edi, dword ptr [esp + XX] 590 return 4; 591 } 592 593 switch (*(u32*)address) { 594 case 0x2444B60F: // 0F B6 44 24 XX : movzx eax, byte ptr [esp + XX] 595 return 5; 596 } 597 #endif 598 599 // Unknown instruction! 600 // FIXME: Unknown instruction failures might happen when we add a new 601 // interceptor or a new compiler version. In either case, they should result 602 // in visible and readable error messages. However, merely calling abort() 603 // leads to an infinite recursion in CheckFailed. 604 InterceptionFailed(); 605 return 0; 606 } 607 608 // Returns 0 on error. 609 static size_t RoundUpToInstrBoundary(size_t size, uptr address) { 610 size_t cursor = 0; 611 while (cursor < size) { 612 size_t instruction_size = GetInstructionSize(address + cursor); 613 if (!instruction_size) 614 return 0; 615 cursor += instruction_size; 616 } 617 return cursor; 618 } 619 620 static bool CopyInstructions(uptr to, uptr from, size_t size) { 621 size_t cursor = 0; 622 while (cursor != size) { 623 size_t rel_offset = 0; 624 size_t instruction_size = GetInstructionSize(from + cursor, &rel_offset); 625 _memcpy((void*)(to + cursor), (void*)(from + cursor), 626 (size_t)instruction_size); 627 if (rel_offset) { 628 uptr delta = to - from; 629 uptr relocated_offset = *(u32*)(to + cursor + rel_offset) - delta; 630 #if SANITIZER_WINDOWS64 631 if (relocated_offset + 0x80000000U >= 0xFFFFFFFFU) 632 return false; 633 #endif 634 *(u32*)(to + cursor + rel_offset) = relocated_offset; 635 } 636 cursor += instruction_size; 637 } 638 return true; 639 } 640 641 642 #if !SANITIZER_WINDOWS64 643 bool OverrideFunctionWithDetour( 644 uptr old_func, uptr new_func, uptr *orig_old_func) { 645 const int kDetourHeaderLen = 5; 646 const u16 kDetourInstruction = 0xFF8B; 647 648 uptr header = (uptr)old_func - kDetourHeaderLen; 649 uptr patch_length = kDetourHeaderLen + kShortJumpInstructionLength; 650 651 // Validate that the function is hookable. 652 if (*(u16*)old_func != kDetourInstruction || 653 !IsMemoryPadding(header, kDetourHeaderLen)) 654 return false; 655 656 // Change memory protection to writable. 657 DWORD protection = 0; 658 if (!ChangeMemoryProtection(header, patch_length, &protection)) 659 return false; 660 661 // Write a relative jump to the redirected function. 662 WriteJumpInstruction(header, new_func); 663 664 // Write the short jump to the function prefix. 665 WriteShortJumpInstruction(old_func, header); 666 667 // Restore previous memory protection. 668 if (!RestoreMemoryProtection(header, patch_length, protection)) 669 return false; 670 671 if (orig_old_func) 672 *orig_old_func = old_func + kShortJumpInstructionLength; 673 674 return true; 675 } 676 #endif 677 678 bool OverrideFunctionWithRedirectJump( 679 uptr old_func, uptr new_func, uptr *orig_old_func) { 680 // Check whether the first instruction is a relative jump. 681 if (*(u8*)old_func != 0xE9) 682 return false; 683 684 if (orig_old_func) { 685 uptr relative_offset = *(u32*)(old_func + 1); 686 uptr absolute_target = old_func + relative_offset + kJumpInstructionLength; 687 *orig_old_func = absolute_target; 688 } 689 690 #if SANITIZER_WINDOWS64 691 // If needed, get memory space for a trampoline jump. 692 uptr trampoline = AllocateMemoryForTrampoline(old_func, kDirectBranchLength); 693 if (!trampoline) 694 return false; 695 WriteDirectBranch(trampoline, new_func); 696 #endif 697 698 // Change memory protection to writable. 699 DWORD protection = 0; 700 if (!ChangeMemoryProtection(old_func, kJumpInstructionLength, &protection)) 701 return false; 702 703 // Write a relative jump to the redirected function. 704 WriteJumpInstruction(old_func, FIRST_32_SECOND_64(new_func, trampoline)); 705 706 // Restore previous memory protection. 707 if (!RestoreMemoryProtection(old_func, kJumpInstructionLength, protection)) 708 return false; 709 710 return true; 711 } 712 713 bool OverrideFunctionWithHotPatch( 714 uptr old_func, uptr new_func, uptr *orig_old_func) { 715 const int kHotPatchHeaderLen = kBranchLength; 716 717 uptr header = (uptr)old_func - kHotPatchHeaderLen; 718 uptr patch_length = kHotPatchHeaderLen + kShortJumpInstructionLength; 719 720 // Validate that the function is hot patchable. 721 size_t instruction_size = GetInstructionSize(old_func); 722 if (instruction_size < kShortJumpInstructionLength || 723 !FunctionHasPadding(old_func, kHotPatchHeaderLen)) 724 return false; 725 726 if (orig_old_func) { 727 // Put the needed instructions into the trampoline bytes. 728 uptr trampoline_length = instruction_size + kDirectBranchLength; 729 uptr trampoline = AllocateMemoryForTrampoline(old_func, trampoline_length); 730 if (!trampoline) 731 return false; 732 if (!CopyInstructions(trampoline, old_func, instruction_size)) 733 return false; 734 WriteDirectBranch(trampoline + instruction_size, 735 old_func + instruction_size); 736 *orig_old_func = trampoline; 737 } 738 739 // If needed, get memory space for indirect address. 740 uptr indirect_address = 0; 741 #if SANITIZER_WINDOWS64 742 indirect_address = AllocateMemoryForTrampoline(old_func, kAddressLength); 743 if (!indirect_address) 744 return false; 745 #endif 746 747 // Change memory protection to writable. 748 DWORD protection = 0; 749 if (!ChangeMemoryProtection(header, patch_length, &protection)) 750 return false; 751 752 // Write jumps to the redirected function. 753 WriteBranch(header, indirect_address, new_func); 754 WriteShortJumpInstruction(old_func, header); 755 756 // Restore previous memory protection. 757 if (!RestoreMemoryProtection(header, patch_length, protection)) 758 return false; 759 760 return true; 761 } 762 763 bool OverrideFunctionWithTrampoline( 764 uptr old_func, uptr new_func, uptr *orig_old_func) { 765 766 size_t instructions_length = kBranchLength; 767 size_t padding_length = 0; 768 uptr indirect_address = 0; 769 770 if (orig_old_func) { 771 // Find out the number of bytes of the instructions we need to copy 772 // to the trampoline. 773 instructions_length = RoundUpToInstrBoundary(kBranchLength, old_func); 774 if (!instructions_length) 775 return false; 776 777 // Put the needed instructions into the trampoline bytes. 778 uptr trampoline_length = instructions_length + kDirectBranchLength; 779 uptr trampoline = AllocateMemoryForTrampoline(old_func, trampoline_length); 780 if (!trampoline) 781 return false; 782 if (!CopyInstructions(trampoline, old_func, instructions_length)) 783 return false; 784 WriteDirectBranch(trampoline + instructions_length, 785 old_func + instructions_length); 786 *orig_old_func = trampoline; 787 } 788 789 #if SANITIZER_WINDOWS64 790 // Check if the targeted address can be encoded in the function padding. 791 // Otherwise, allocate it in the trampoline region. 792 if (IsMemoryPadding(old_func - kAddressLength, kAddressLength)) { 793 indirect_address = old_func - kAddressLength; 794 padding_length = kAddressLength; 795 } else { 796 indirect_address = AllocateMemoryForTrampoline(old_func, kAddressLength); 797 if (!indirect_address) 798 return false; 799 } 800 #endif 801 802 // Change memory protection to writable. 803 uptr patch_address = old_func - padding_length; 804 uptr patch_length = instructions_length + padding_length; 805 DWORD protection = 0; 806 if (!ChangeMemoryProtection(patch_address, patch_length, &protection)) 807 return false; 808 809 // Patch the original function. 810 WriteBranch(old_func, indirect_address, new_func); 811 812 // Restore previous memory protection. 813 if (!RestoreMemoryProtection(patch_address, patch_length, protection)) 814 return false; 815 816 return true; 817 } 818 819 bool OverrideFunction( 820 uptr old_func, uptr new_func, uptr *orig_old_func) { 821 #if !SANITIZER_WINDOWS64 822 if (OverrideFunctionWithDetour(old_func, new_func, orig_old_func)) 823 return true; 824 #endif 825 if (OverrideFunctionWithRedirectJump(old_func, new_func, orig_old_func)) 826 return true; 827 if (OverrideFunctionWithHotPatch(old_func, new_func, orig_old_func)) 828 return true; 829 if (OverrideFunctionWithTrampoline(old_func, new_func, orig_old_func)) 830 return true; 831 return false; 832 } 833 834 static void **InterestingDLLsAvailable() { 835 static const char *InterestingDLLs[] = { 836 "kernel32.dll", 837 "msvcr100.dll", // VS2010 838 "msvcr110.dll", // VS2012 839 "msvcr120.dll", // VS2013 840 "vcruntime140.dll", // VS2015 841 "ucrtbase.dll", // Universal CRT 842 // NTDLL should go last as it exports some functions that we should 843 // override in the CRT [presumably only used internally]. 844 "ntdll.dll", NULL}; 845 static void *result[ARRAY_SIZE(InterestingDLLs)] = { 0 }; 846 if (!result[0]) { 847 for (size_t i = 0, j = 0; InterestingDLLs[i]; ++i) { 848 if (HMODULE h = GetModuleHandleA(InterestingDLLs[i])) 849 result[j++] = (void *)h; 850 } 851 } 852 return &result[0]; 853 } 854 855 namespace { 856 // Utility for reading loaded PE images. 857 template <typename T> class RVAPtr { 858 public: 859 RVAPtr(void *module, uptr rva) 860 : ptr_(reinterpret_cast<T *>(reinterpret_cast<char *>(module) + rva)) {} 861 operator T *() { return ptr_; } 862 T *operator->() { return ptr_; } 863 T *operator++() { return ++ptr_; } 864 865 private: 866 T *ptr_; 867 }; 868 } // namespace 869 870 // Internal implementation of GetProcAddress. At least since Windows 8, 871 // GetProcAddress appears to initialize DLLs before returning function pointers 872 // into them. This is problematic for the sanitizers, because they typically 873 // want to intercept malloc *before* MSVCRT initializes. Our internal 874 // implementation walks the export list manually without doing initialization. 875 uptr InternalGetProcAddress(void *module, const char *func_name) { 876 // Check that the module header is full and present. 877 RVAPtr<IMAGE_DOS_HEADER> dos_stub(module, 0); 878 RVAPtr<IMAGE_NT_HEADERS> headers(module, dos_stub->e_lfanew); 879 if (!module || dos_stub->e_magic != IMAGE_DOS_SIGNATURE || // "MZ" 880 headers->Signature != IMAGE_NT_SIGNATURE || // "PE\0\0" 881 headers->FileHeader.SizeOfOptionalHeader < 882 sizeof(IMAGE_OPTIONAL_HEADER)) { 883 return 0; 884 } 885 886 IMAGE_DATA_DIRECTORY *export_directory = 887 &headers->OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_EXPORT]; 888 if (export_directory->Size == 0) 889 return 0; 890 RVAPtr<IMAGE_EXPORT_DIRECTORY> exports(module, 891 export_directory->VirtualAddress); 892 RVAPtr<DWORD> functions(module, exports->AddressOfFunctions); 893 RVAPtr<DWORD> names(module, exports->AddressOfNames); 894 RVAPtr<WORD> ordinals(module, exports->AddressOfNameOrdinals); 895 896 for (DWORD i = 0; i < exports->NumberOfNames; i++) { 897 RVAPtr<char> name(module, names[i]); 898 if (!strcmp(func_name, name)) { 899 DWORD index = ordinals[i]; 900 RVAPtr<char> func(module, functions[index]); 901 902 // Handle forwarded functions. 903 DWORD offset = functions[index]; 904 if (offset >= export_directory->VirtualAddress && 905 offset < export_directory->VirtualAddress + export_directory->Size) { 906 // An entry for a forwarded function is a string with the following 907 // format: "<module> . <function_name>" that is stored into the 908 // exported directory. 909 char function_name[256]; 910 size_t funtion_name_length = _strlen(func); 911 if (funtion_name_length >= sizeof(function_name) - 1) 912 InterceptionFailed(); 913 914 _memcpy(function_name, func, funtion_name_length); 915 function_name[funtion_name_length] = '\0'; 916 char* separator = _strchr(function_name, '.'); 917 if (!separator) 918 InterceptionFailed(); 919 *separator = '\0'; 920 921 void* redirected_module = GetModuleHandleA(function_name); 922 if (!redirected_module) 923 InterceptionFailed(); 924 return InternalGetProcAddress(redirected_module, separator + 1); 925 } 926 927 return (uptr)(char *)func; 928 } 929 } 930 931 return 0; 932 } 933 934 bool OverrideFunction( 935 const char *func_name, uptr new_func, uptr *orig_old_func) { 936 bool hooked = false; 937 void **DLLs = InterestingDLLsAvailable(); 938 for (size_t i = 0; DLLs[i]; ++i) { 939 uptr func_addr = InternalGetProcAddress(DLLs[i], func_name); 940 if (func_addr && 941 OverrideFunction(func_addr, new_func, orig_old_func)) { 942 hooked = true; 943 } 944 } 945 return hooked; 946 } 947 948 bool OverrideImportedFunction(const char *module_to_patch, 949 const char *imported_module, 950 const char *function_name, uptr new_function, 951 uptr *orig_old_func) { 952 HMODULE module = GetModuleHandleA(module_to_patch); 953 if (!module) 954 return false; 955 956 // Check that the module header is full and present. 957 RVAPtr<IMAGE_DOS_HEADER> dos_stub(module, 0); 958 RVAPtr<IMAGE_NT_HEADERS> headers(module, dos_stub->e_lfanew); 959 if (!module || dos_stub->e_magic != IMAGE_DOS_SIGNATURE || // "MZ" 960 headers->Signature != IMAGE_NT_SIGNATURE || // "PE\0\0" 961 headers->FileHeader.SizeOfOptionalHeader < 962 sizeof(IMAGE_OPTIONAL_HEADER)) { 963 return false; 964 } 965 966 IMAGE_DATA_DIRECTORY *import_directory = 967 &headers->OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_IMPORT]; 968 969 // Iterate the list of imported DLLs. FirstThunk will be null for the last 970 // entry. 971 RVAPtr<IMAGE_IMPORT_DESCRIPTOR> imports(module, 972 import_directory->VirtualAddress); 973 for (; imports->FirstThunk != 0; ++imports) { 974 RVAPtr<const char> modname(module, imports->Name); 975 if (_stricmp(&*modname, imported_module) == 0) 976 break; 977 } 978 if (imports->FirstThunk == 0) 979 return false; 980 981 // We have two parallel arrays: the import address table (IAT) and the table 982 // of names. They start out containing the same data, but the loader rewrites 983 // the IAT to hold imported addresses and leaves the name table in 984 // OriginalFirstThunk alone. 985 RVAPtr<IMAGE_THUNK_DATA> name_table(module, imports->OriginalFirstThunk); 986 RVAPtr<IMAGE_THUNK_DATA> iat(module, imports->FirstThunk); 987 for (; name_table->u1.Ordinal != 0; ++name_table, ++iat) { 988 if (!IMAGE_SNAP_BY_ORDINAL(name_table->u1.Ordinal)) { 989 RVAPtr<IMAGE_IMPORT_BY_NAME> import_by_name( 990 module, name_table->u1.ForwarderString); 991 const char *funcname = &import_by_name->Name[0]; 992 if (strcmp(funcname, function_name) == 0) 993 break; 994 } 995 } 996 if (name_table->u1.Ordinal == 0) 997 return false; 998 999 // Now we have the correct IAT entry. Do the swap. We have to make the page 1000 // read/write first. 1001 if (orig_old_func) 1002 *orig_old_func = iat->u1.AddressOfData; 1003 DWORD old_prot, unused_prot; 1004 if (!VirtualProtect(&iat->u1.AddressOfData, 4, PAGE_EXECUTE_READWRITE, 1005 &old_prot)) 1006 return false; 1007 iat->u1.AddressOfData = new_function; 1008 if (!VirtualProtect(&iat->u1.AddressOfData, 4, old_prot, &unused_prot)) 1009 return false; // Not clear if this failure bothers us. 1010 return true; 1011 } 1012 1013 } // namespace __interception 1014 1015 #endif // SANITIZER_MAC 1016