1 //===-- interception_linux.cc -----------------------------------*- C++ -*-===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file is a part of AddressSanitizer, an address sanity checker. 11 // 12 // Windows-specific interception methods. 13 // 14 // This file is implementing several hooking techniques to intercept calls 15 // to functions. The hooks are dynamically installed by modifying the assembly 16 // code. 17 // 18 // The hooking techniques are making assumptions on the way the code is 19 // generated and are safe under these assumptions. 20 // 21 // On 64-bit architecture, there is no direct 64-bit jump instruction. To allow 22 // arbitrary branching on the whole memory space, the notion of trampoline 23 // region is used. A trampoline region is a memory space withing 2G boundary 24 // where it is safe to add custom assembly code to build 64-bit jumps. 25 // 26 // Hooking techniques 27 // ================== 28 // 29 // 1) Detour 30 // 31 // The Detour hooking technique is assuming the presence of an header with 32 // padding and an overridable 2-bytes nop instruction (mov edi, edi). The 33 // nop instruction can safely be replaced by a 2-bytes jump without any need 34 // to save the instruction. A jump to the target is encoded in the function 35 // header and the nop instruction is replaced by a short jump to the header. 36 // 37 // head: 5 x nop head: jmp <hook> 38 // func: mov edi, edi --> func: jmp short <head> 39 // [...] real: [...] 40 // 41 // This technique is only implemented on 32-bit architecture. 42 // Most of the time, Windows API are hookable with the detour technique. 43 // 44 // 2) Redirect Jump 45 // 46 // The redirect jump is applicable when the first instruction is a direct 47 // jump. The instruction is replaced by jump to the hook. 48 // 49 // func: jmp <label> --> func: jmp <hook> 50 // 51 // On an 64-bit architecture, a trampoline is inserted. 52 // 53 // func: jmp <label> --> func: jmp <tramp> 54 // [...] 55 // 56 // [trampoline] 57 // tramp: jmp QWORD [addr] 58 // addr: .bytes <hook> 59 // 60 // Note: <real> is equilavent to <label>. 61 // 62 // 3) HotPatch 63 // 64 // The HotPatch hooking is assuming the presence of an header with padding 65 // and a first instruction with at least 2-bytes. 66 // 67 // The reason to enforce the 2-bytes limitation is to provide the minimal 68 // space to encode a short jump. HotPatch technique is only rewriting one 69 // instruction to avoid breaking a sequence of instructions containing a 70 // branching target. 71 // 72 // Assumptions are enforced by MSVC compiler by using the /HOTPATCH flag. 73 // see: https://msdn.microsoft.com/en-us/library/ms173507.aspx 74 // Default padding length is 5 bytes in 32-bits and 6 bytes in 64-bits. 75 // 76 // head: 5 x nop head: jmp <hook> 77 // func: <instr> --> func: jmp short <head> 78 // [...] body: [...] 79 // 80 // [trampoline] 81 // real: <instr> 82 // jmp <body> 83 // 84 // On an 64-bit architecture: 85 // 86 // head: 6 x nop head: jmp QWORD [addr1] 87 // func: <instr> --> func: jmp short <head> 88 // [...] body: [...] 89 // 90 // [trampoline] 91 // addr1: .bytes <hook> 92 // real: <instr> 93 // jmp QWORD [addr2] 94 // addr2: .bytes <body> 95 // 96 // 4) Trampoline 97 // 98 // The Trampoline hooking technique is the most aggressive one. It is 99 // assuming that there is a sequence of instructions that can be safely 100 // replaced by a jump (enough room and no incoming branches). 101 // 102 // Unfortunately, these assumptions can't be safely presumed and code may 103 // be broken after hooking. 104 // 105 // func: <instr> --> func: jmp <hook> 106 // <instr> 107 // [...] body: [...] 108 // 109 // [trampoline] 110 // real: <instr> 111 // <instr> 112 // jmp <body> 113 // 114 // On an 64-bit architecture: 115 // 116 // func: <instr> --> func: jmp QWORD [addr1] 117 // <instr> 118 // [...] body: [...] 119 // 120 // [trampoline] 121 // addr1: .bytes <hook> 122 // real: <instr> 123 // <instr> 124 // jmp QWORD [addr2] 125 // addr2: .bytes <body> 126 //===----------------------------------------------------------------------===// 127 128 #include "interception.h" 129 130 #if SANITIZER_WINDOWS 131 #include "sanitizer_common/sanitizer_platform.h" 132 #define WIN32_LEAN_AND_MEAN 133 #include <windows.h> 134 135 namespace __interception { 136 137 static const int kAddressLength = FIRST_32_SECOND_64(4, 8); 138 static const int kJumpInstructionLength = 5; 139 static const int kShortJumpInstructionLength = 2; 140 static const int kIndirectJumpInstructionLength = 6; 141 static const int kBranchLength = 142 FIRST_32_SECOND_64(kJumpInstructionLength, kIndirectJumpInstructionLength); 143 static const int kDirectBranchLength = kBranchLength + kAddressLength; 144 145 static void InterceptionFailed() { 146 // Do we have a good way to abort with an error message here? 147 __debugbreak(); 148 } 149 150 static bool DistanceIsWithin2Gig(uptr from, uptr target) { 151 #if SANITIZER_WINDOWS64 152 if (from < target) 153 return target - from <= (uptr)0x7FFFFFFFU; 154 else 155 return from - target <= (uptr)0x80000000U; 156 #else 157 // In a 32-bit address space, the address calculation will wrap, so this check 158 // is unnecessary. 159 return true; 160 #endif 161 } 162 163 static uptr GetMmapGranularity() { 164 SYSTEM_INFO si; 165 GetSystemInfo(&si); 166 return si.dwAllocationGranularity; 167 } 168 169 static uptr RoundUpTo(uptr size, uptr boundary) { 170 return (size + boundary - 1) & ~(boundary - 1); 171 } 172 173 // FIXME: internal_str* and internal_mem* functions should be moved from the 174 // ASan sources into interception/. 175 176 static size_t _strlen(const char *str) { 177 const char* p = str; 178 while (*p != '\0') ++p; 179 return p - str; 180 } 181 182 static char* _strchr(char* str, char c) { 183 while (*str) { 184 if (*str == c) 185 return str; 186 ++str; 187 } 188 return nullptr; 189 } 190 191 static void _memset(void *p, int value, size_t sz) { 192 for (size_t i = 0; i < sz; ++i) 193 ((char*)p)[i] = (char)value; 194 } 195 196 static void _memcpy(void *dst, void *src, size_t sz) { 197 char *dst_c = (char*)dst, 198 *src_c = (char*)src; 199 for (size_t i = 0; i < sz; ++i) 200 dst_c[i] = src_c[i]; 201 } 202 203 static bool ChangeMemoryProtection( 204 uptr address, uptr size, DWORD *old_protection) { 205 return ::VirtualProtect((void*)address, size, 206 PAGE_EXECUTE_READWRITE, 207 old_protection) != FALSE; 208 } 209 210 static bool RestoreMemoryProtection( 211 uptr address, uptr size, DWORD old_protection) { 212 DWORD unused; 213 return ::VirtualProtect((void*)address, size, 214 old_protection, 215 &unused) != FALSE; 216 } 217 218 static bool IsMemoryPadding(uptr address, uptr size) { 219 u8* function = (u8*)address; 220 for (size_t i = 0; i < size; ++i) 221 if (function[i] != 0x90 && function[i] != 0xCC) 222 return false; 223 return true; 224 } 225 226 static const u8 kHintNop8Bytes[] = { 227 0x0F, 0x1F, 0x84, 0x00, 0x00, 0x00, 0x00, 0x00 228 }; 229 230 template<class T> 231 static bool FunctionHasPrefix(uptr address, const T &pattern) { 232 u8* function = (u8*)address - sizeof(pattern); 233 for (size_t i = 0; i < sizeof(pattern); ++i) 234 if (function[i] != pattern[i]) 235 return false; 236 return true; 237 } 238 239 static bool FunctionHasPadding(uptr address, uptr size) { 240 if (IsMemoryPadding(address - size, size)) 241 return true; 242 if (size <= sizeof(kHintNop8Bytes) && 243 FunctionHasPrefix(address, kHintNop8Bytes)) 244 return true; 245 return false; 246 } 247 248 static void WritePadding(uptr from, uptr size) { 249 _memset((void*)from, 0xCC, (size_t)size); 250 } 251 252 static void WriteJumpInstruction(uptr from, uptr target) { 253 if (!DistanceIsWithin2Gig(from + kJumpInstructionLength, target)) 254 InterceptionFailed(); 255 ptrdiff_t offset = target - from - kJumpInstructionLength; 256 *(u8*)from = 0xE9; 257 *(u32*)(from + 1) = offset; 258 } 259 260 static void WriteShortJumpInstruction(uptr from, uptr target) { 261 sptr offset = target - from - kShortJumpInstructionLength; 262 if (offset < -128 || offset > 127) 263 InterceptionFailed(); 264 *(u8*)from = 0xEB; 265 *(u8*)(from + 1) = (u8)offset; 266 } 267 268 #if SANITIZER_WINDOWS64 269 static void WriteIndirectJumpInstruction(uptr from, uptr indirect_target) { 270 // jmp [rip + <offset>] = FF 25 <offset> where <offset> is a relative 271 // offset. 272 // The offset is the distance from then end of the jump instruction to the 273 // memory location containing the targeted address. The displacement is still 274 // 32-bit in x64, so indirect_target must be located within +/- 2GB range. 275 int offset = indirect_target - from - kIndirectJumpInstructionLength; 276 if (!DistanceIsWithin2Gig(from + kIndirectJumpInstructionLength, 277 indirect_target)) { 278 InterceptionFailed(); 279 } 280 *(u16*)from = 0x25FF; 281 *(u32*)(from + 2) = offset; 282 } 283 #endif 284 285 static void WriteBranch( 286 uptr from, uptr indirect_target, uptr target) { 287 #if SANITIZER_WINDOWS64 288 WriteIndirectJumpInstruction(from, indirect_target); 289 *(u64*)indirect_target = target; 290 #else 291 (void)indirect_target; 292 WriteJumpInstruction(from, target); 293 #endif 294 } 295 296 static void WriteDirectBranch(uptr from, uptr target) { 297 #if SANITIZER_WINDOWS64 298 // Emit an indirect jump through immediately following bytes: 299 // jmp [rip + kBranchLength] 300 // .quad <target> 301 WriteBranch(from, from + kBranchLength, target); 302 #else 303 WriteJumpInstruction(from, target); 304 #endif 305 } 306 307 struct TrampolineMemoryRegion { 308 uptr content; 309 uptr allocated_size; 310 uptr max_size; 311 }; 312 313 static const uptr kTrampolineScanLimitRange = 1 << 31; // 2 gig 314 static const int kMaxTrampolineRegion = 1024; 315 static TrampolineMemoryRegion TrampolineRegions[kMaxTrampolineRegion]; 316 317 static void *AllocateTrampolineRegion(uptr image_address, size_t granularity) { 318 #if SANITIZER_WINDOWS64 319 uptr address = image_address; 320 uptr scanned = 0; 321 while (scanned < kTrampolineScanLimitRange) { 322 MEMORY_BASIC_INFORMATION info; 323 if (!::VirtualQuery((void*)address, &info, sizeof(info))) 324 return nullptr; 325 326 // Check whether a region can be allocated at |address|. 327 if (info.State == MEM_FREE && info.RegionSize >= granularity) { 328 void *page = ::VirtualAlloc((void*)RoundUpTo(address, granularity), 329 granularity, 330 MEM_RESERVE | MEM_COMMIT, 331 PAGE_EXECUTE_READWRITE); 332 return page; 333 } 334 335 // Move to the next region. 336 address = (uptr)info.BaseAddress + info.RegionSize; 337 scanned += info.RegionSize; 338 } 339 return nullptr; 340 #else 341 return ::VirtualAlloc(nullptr, 342 granularity, 343 MEM_RESERVE | MEM_COMMIT, 344 PAGE_EXECUTE_READWRITE); 345 #endif 346 } 347 348 // Used by unittests to release mapped memory space. 349 void TestOnlyReleaseTrampolineRegions() { 350 for (size_t bucket = 0; bucket < kMaxTrampolineRegion; ++bucket) { 351 TrampolineMemoryRegion *current = &TrampolineRegions[bucket]; 352 if (current->content == 0) 353 return; 354 ::VirtualFree((void*)current->content, 0, MEM_RELEASE); 355 current->content = 0; 356 } 357 } 358 359 static uptr AllocateMemoryForTrampoline(uptr image_address, size_t size) { 360 // Find a region within 2G with enough space to allocate |size| bytes. 361 TrampolineMemoryRegion *region = nullptr; 362 for (size_t bucket = 0; bucket < kMaxTrampolineRegion; ++bucket) { 363 TrampolineMemoryRegion* current = &TrampolineRegions[bucket]; 364 if (current->content == 0) { 365 // No valid region found, allocate a new region. 366 size_t bucket_size = GetMmapGranularity(); 367 void *content = AllocateTrampolineRegion(image_address, bucket_size); 368 if (content == nullptr) 369 return 0U; 370 371 current->content = (uptr)content; 372 current->allocated_size = 0; 373 current->max_size = bucket_size; 374 region = current; 375 break; 376 } else if (current->max_size - current->allocated_size > size) { 377 #if SANITIZER_WINDOWS64 378 // In 64-bits, the memory space must be allocated within 2G boundary. 379 uptr next_address = current->content + current->allocated_size; 380 if (next_address < image_address || 381 next_address - image_address >= 0x7FFF0000) 382 continue; 383 #endif 384 // The space can be allocated in the current region. 385 region = current; 386 break; 387 } 388 } 389 390 // Failed to find a region. 391 if (region == nullptr) 392 return 0U; 393 394 // Allocate the space in the current region. 395 uptr allocated_space = region->content + region->allocated_size; 396 region->allocated_size += size; 397 WritePadding(allocated_space, size); 398 399 return allocated_space; 400 } 401 402 // Returns 0 on error. 403 static size_t GetInstructionSize(uptr address, size_t* rel_offset = nullptr) { 404 switch (*(u64*)address) { 405 case 0x90909090909006EB: // stub: jmp over 6 x nop. 406 return 8; 407 } 408 409 switch (*(u8*)address) { 410 case 0x90: // 90 : nop 411 return 1; 412 413 case 0x50: // push eax / rax 414 case 0x51: // push ecx / rcx 415 case 0x52: // push edx / rdx 416 case 0x53: // push ebx / rbx 417 case 0x54: // push esp / rsp 418 case 0x55: // push ebp / rbp 419 case 0x56: // push esi / rsi 420 case 0x57: // push edi / rdi 421 case 0x5D: // pop ebp / rbp 422 return 1; 423 424 case 0x6A: // 6A XX = push XX 425 return 2; 426 427 case 0xb8: // b8 XX XX XX XX : mov eax, XX XX XX XX 428 case 0xB9: // b9 XX XX XX XX : mov ecx, XX XX XX XX 429 return 5; 430 431 // Cannot overwrite control-instruction. Return 0 to indicate failure. 432 case 0xE9: // E9 XX XX XX XX : jmp <label> 433 case 0xE8: // E8 XX XX XX XX : call <func> 434 case 0xC3: // C3 : ret 435 case 0xEB: // EB XX : jmp XX (short jump) 436 case 0x70: // 7Y YY : jy XX (short conditional jump) 437 case 0x71: 438 case 0x72: 439 case 0x73: 440 case 0x74: 441 case 0x75: 442 case 0x76: 443 case 0x77: 444 case 0x78: 445 case 0x79: 446 case 0x7A: 447 case 0x7B: 448 case 0x7C: 449 case 0x7D: 450 case 0x7E: 451 case 0x7F: 452 return 0; 453 } 454 455 switch (*(u16*)(address)) { 456 case 0x018A: // 8A 01 : mov al, byte ptr [ecx] 457 case 0xFF8B: // 8B FF : mov edi, edi 458 case 0xEC8B: // 8B EC : mov ebp, esp 459 case 0xc889: // 89 C8 : mov eax, ecx 460 case 0xC18B: // 8B C1 : mov eax, ecx 461 case 0xC033: // 33 C0 : xor eax, eax 462 case 0xC933: // 33 C9 : xor ecx, ecx 463 case 0xD233: // 33 D2 : xor edx, edx 464 return 2; 465 466 // Cannot overwrite control-instruction. Return 0 to indicate failure. 467 case 0x25FF: // FF 25 XX XX XX XX : jmp [XXXXXXXX] 468 return 0; 469 } 470 471 switch (0x00FFFFFF & *(u32*)address) { 472 case 0x24A48D: // 8D A4 24 XX XX XX XX : lea esp, [esp + XX XX XX XX] 473 return 7; 474 } 475 476 #if SANITIZER_WINDOWS64 477 switch (*(u8*)address) { 478 case 0xA1: // A1 XX XX XX XX XX XX XX XX : 479 // movabs eax, dword ptr ds:[XXXXXXXX] 480 return 9; 481 } 482 483 switch (*(u16*)address) { 484 case 0x5040: // push rax 485 case 0x5140: // push rcx 486 case 0x5240: // push rdx 487 case 0x5340: // push rbx 488 case 0x5440: // push rsp 489 case 0x5540: // push rbp 490 case 0x5640: // push rsi 491 case 0x5740: // push rdi 492 case 0x5441: // push r12 493 case 0x5541: // push r13 494 case 0x5641: // push r14 495 case 0x5741: // push r15 496 case 0x9066: // Two-byte NOP 497 return 2; 498 499 case 0x058B: // 8B 05 XX XX XX XX : mov eax, dword ptr [XX XX XX XX] 500 if (rel_offset) 501 *rel_offset = 2; 502 return 6; 503 } 504 505 switch (0x00FFFFFF & *(u32*)address) { 506 case 0xe58948: // 48 8b c4 : mov rbp, rsp 507 case 0xc18b48: // 48 8b c1 : mov rax, rcx 508 case 0xc48b48: // 48 8b c4 : mov rax, rsp 509 case 0xd9f748: // 48 f7 d9 : neg rcx 510 case 0xd12b48: // 48 2b d1 : sub rdx, rcx 511 case 0x07c1f6: // f6 c1 07 : test cl, 0x7 512 case 0xc98548: // 48 85 C9 : test rcx, rcx 513 case 0xc0854d: // 4d 85 c0 : test r8, r8 514 case 0xc2b60f: // 0f b6 c2 : movzx eax, dl 515 case 0xc03345: // 45 33 c0 : xor r8d, r8d 516 case 0xdb3345: // 45 33 DB : xor r11d, r11d 517 case 0xd98b4c: // 4c 8b d9 : mov r11, rcx 518 case 0xd28b4c: // 4c 8b d2 : mov r10, rdx 519 case 0xc98b4c: // 4C 8B C9 : mov r9, rcx 520 case 0xd2b60f: // 0f b6 d2 : movzx edx, dl 521 case 0xca2b48: // 48 2b ca : sub rcx, rdx 522 case 0x10b70f: // 0f b7 10 : movzx edx, WORD PTR [rax] 523 case 0xc00b4d: // 3d 0b c0 : or r8, r8 524 case 0xd18b48: // 48 8b d1 : mov rdx, rcx 525 case 0xdc8b4c: // 4c 8b dc : mov r11, rsp 526 case 0xd18b4c: // 4c 8b d1 : mov r10, rcx 527 return 3; 528 529 case 0xec8348: // 48 83 ec XX : sub rsp, XX 530 case 0xf88349: // 49 83 f8 XX : cmp r8, XX 531 case 0x588948: // 48 89 58 XX : mov QWORD PTR[rax + XX], rbx 532 return 4; 533 534 case 0xec8148: // 48 81 EC XX XX XX XX : sub rsp, XXXXXXXX 535 return 7; 536 537 case 0x058b48: // 48 8b 05 XX XX XX XX : 538 // mov rax, QWORD PTR [rip + XXXXXXXX] 539 case 0x25ff48: // 48 ff 25 XX XX XX XX : 540 // rex.W jmp QWORD PTR [rip + XXXXXXXX] 541 542 // Instructions having offset relative to 'rip' need offset adjustment. 543 if (rel_offset) 544 *rel_offset = 3; 545 return 7; 546 547 case 0x2444c7: // C7 44 24 XX YY YY YY YY 548 // mov dword ptr [rsp + XX], YYYYYYYY 549 return 8; 550 } 551 552 switch (*(u32*)(address)) { 553 case 0x24448b48: // 48 8b 44 24 XX : mov rax, QWORD ptr [rsp + XX] 554 case 0x246c8948: // 48 89 6C 24 XX : mov QWORD ptr [rsp + XX], rbp 555 case 0x245c8948: // 48 89 5c 24 XX : mov QWORD PTR [rsp + XX], rbx 556 case 0x24748948: // 48 89 74 24 XX : mov QWORD PTR [rsp + XX], rsi 557 case 0x244C8948: // 48 89 4C 24 XX : mov QWORD PTR [rsp + XX], rcx 558 return 5; 559 case 0x24648348: // 48 83 64 24 XX : and QWORD PTR [rsp + XX], YY 560 return 6; 561 } 562 563 #else 564 565 switch (*(u8*)address) { 566 case 0xA1: // A1 XX XX XX XX : mov eax, dword ptr ds:[XXXXXXXX] 567 return 5; 568 } 569 switch (*(u16*)address) { 570 case 0x458B: // 8B 45 XX : mov eax, dword ptr [ebp + XX] 571 case 0x5D8B: // 8B 5D XX : mov ebx, dword ptr [ebp + XX] 572 case 0x7D8B: // 8B 7D XX : mov edi, dword ptr [ebp + XX] 573 case 0xEC83: // 83 EC XX : sub esp, XX 574 case 0x75FF: // FF 75 XX : push dword ptr [ebp + XX] 575 return 3; 576 case 0xC1F7: // F7 C1 XX YY ZZ WW : test ecx, WWZZYYXX 577 case 0x25FF: // FF 25 XX YY ZZ WW : jmp dword ptr ds:[WWZZYYXX] 578 return 6; 579 case 0x3D83: // 83 3D XX YY ZZ WW TT : cmp TT, WWZZYYXX 580 return 7; 581 case 0x7D83: // 83 7D XX YY : cmp dword ptr [ebp + XX], YY 582 return 4; 583 } 584 585 switch (0x00FFFFFF & *(u32*)address) { 586 case 0x24448A: // 8A 44 24 XX : mov eal, dword ptr [esp + XX] 587 case 0x24448B: // 8B 44 24 XX : mov eax, dword ptr [esp + XX] 588 case 0x244C8B: // 8B 4C 24 XX : mov ecx, dword ptr [esp + XX] 589 case 0x24548B: // 8B 54 24 XX : mov edx, dword ptr [esp + XX] 590 case 0x24748B: // 8B 74 24 XX : mov esi, dword ptr [esp + XX] 591 case 0x247C8B: // 8B 7C 24 XX : mov edi, dword ptr [esp + XX] 592 return 4; 593 } 594 595 switch (*(u32*)address) { 596 case 0x2444B60F: // 0F B6 44 24 XX : movzx eax, byte ptr [esp + XX] 597 return 5; 598 } 599 #endif 600 601 // Unknown instruction! 602 // FIXME: Unknown instruction failures might happen when we add a new 603 // interceptor or a new compiler version. In either case, they should result 604 // in visible and readable error messages. However, merely calling abort() 605 // leads to an infinite recursion in CheckFailed. 606 InterceptionFailed(); 607 return 0; 608 } 609 610 // Returns 0 on error. 611 static size_t RoundUpToInstrBoundary(size_t size, uptr address) { 612 size_t cursor = 0; 613 while (cursor < size) { 614 size_t instruction_size = GetInstructionSize(address + cursor); 615 if (!instruction_size) 616 return 0; 617 cursor += instruction_size; 618 } 619 return cursor; 620 } 621 622 static bool CopyInstructions(uptr to, uptr from, size_t size) { 623 size_t cursor = 0; 624 while (cursor != size) { 625 size_t rel_offset = 0; 626 size_t instruction_size = GetInstructionSize(from + cursor, &rel_offset); 627 _memcpy((void*)(to + cursor), (void*)(from + cursor), 628 (size_t)instruction_size); 629 if (rel_offset) { 630 uptr delta = to - from; 631 uptr relocated_offset = *(u32*)(to + cursor + rel_offset) - delta; 632 #if SANITIZER_WINDOWS64 633 if (relocated_offset + 0x80000000U >= 0xFFFFFFFFU) 634 return false; 635 #endif 636 *(u32*)(to + cursor + rel_offset) = relocated_offset; 637 } 638 cursor += instruction_size; 639 } 640 return true; 641 } 642 643 644 #if !SANITIZER_WINDOWS64 645 bool OverrideFunctionWithDetour( 646 uptr old_func, uptr new_func, uptr *orig_old_func) { 647 const int kDetourHeaderLen = 5; 648 const u16 kDetourInstruction = 0xFF8B; 649 650 uptr header = (uptr)old_func - kDetourHeaderLen; 651 uptr patch_length = kDetourHeaderLen + kShortJumpInstructionLength; 652 653 // Validate that the function is hookable. 654 if (*(u16*)old_func != kDetourInstruction || 655 !IsMemoryPadding(header, kDetourHeaderLen)) 656 return false; 657 658 // Change memory protection to writable. 659 DWORD protection = 0; 660 if (!ChangeMemoryProtection(header, patch_length, &protection)) 661 return false; 662 663 // Write a relative jump to the redirected function. 664 WriteJumpInstruction(header, new_func); 665 666 // Write the short jump to the function prefix. 667 WriteShortJumpInstruction(old_func, header); 668 669 // Restore previous memory protection. 670 if (!RestoreMemoryProtection(header, patch_length, protection)) 671 return false; 672 673 if (orig_old_func) 674 *orig_old_func = old_func + kShortJumpInstructionLength; 675 676 return true; 677 } 678 #endif 679 680 bool OverrideFunctionWithRedirectJump( 681 uptr old_func, uptr new_func, uptr *orig_old_func) { 682 // Check whether the first instruction is a relative jump. 683 if (*(u8*)old_func != 0xE9) 684 return false; 685 686 if (orig_old_func) { 687 uptr relative_offset = *(u32*)(old_func + 1); 688 uptr absolute_target = old_func + relative_offset + kJumpInstructionLength; 689 *orig_old_func = absolute_target; 690 } 691 692 #if SANITIZER_WINDOWS64 693 // If needed, get memory space for a trampoline jump. 694 uptr trampoline = AllocateMemoryForTrampoline(old_func, kDirectBranchLength); 695 if (!trampoline) 696 return false; 697 WriteDirectBranch(trampoline, new_func); 698 #endif 699 700 // Change memory protection to writable. 701 DWORD protection = 0; 702 if (!ChangeMemoryProtection(old_func, kJumpInstructionLength, &protection)) 703 return false; 704 705 // Write a relative jump to the redirected function. 706 WriteJumpInstruction(old_func, FIRST_32_SECOND_64(new_func, trampoline)); 707 708 // Restore previous memory protection. 709 if (!RestoreMemoryProtection(old_func, kJumpInstructionLength, protection)) 710 return false; 711 712 return true; 713 } 714 715 bool OverrideFunctionWithHotPatch( 716 uptr old_func, uptr new_func, uptr *orig_old_func) { 717 const int kHotPatchHeaderLen = kBranchLength; 718 719 uptr header = (uptr)old_func - kHotPatchHeaderLen; 720 uptr patch_length = kHotPatchHeaderLen + kShortJumpInstructionLength; 721 722 // Validate that the function is hot patchable. 723 size_t instruction_size = GetInstructionSize(old_func); 724 if (instruction_size < kShortJumpInstructionLength || 725 !FunctionHasPadding(old_func, kHotPatchHeaderLen)) 726 return false; 727 728 if (orig_old_func) { 729 // Put the needed instructions into the trampoline bytes. 730 uptr trampoline_length = instruction_size + kDirectBranchLength; 731 uptr trampoline = AllocateMemoryForTrampoline(old_func, trampoline_length); 732 if (!trampoline) 733 return false; 734 if (!CopyInstructions(trampoline, old_func, instruction_size)) 735 return false; 736 WriteDirectBranch(trampoline + instruction_size, 737 old_func + instruction_size); 738 *orig_old_func = trampoline; 739 } 740 741 // If needed, get memory space for indirect address. 742 uptr indirect_address = 0; 743 #if SANITIZER_WINDOWS64 744 indirect_address = AllocateMemoryForTrampoline(old_func, kAddressLength); 745 if (!indirect_address) 746 return false; 747 #endif 748 749 // Change memory protection to writable. 750 DWORD protection = 0; 751 if (!ChangeMemoryProtection(header, patch_length, &protection)) 752 return false; 753 754 // Write jumps to the redirected function. 755 WriteBranch(header, indirect_address, new_func); 756 WriteShortJumpInstruction(old_func, header); 757 758 // Restore previous memory protection. 759 if (!RestoreMemoryProtection(header, patch_length, protection)) 760 return false; 761 762 return true; 763 } 764 765 bool OverrideFunctionWithTrampoline( 766 uptr old_func, uptr new_func, uptr *orig_old_func) { 767 768 size_t instructions_length = kBranchLength; 769 size_t padding_length = 0; 770 uptr indirect_address = 0; 771 772 if (orig_old_func) { 773 // Find out the number of bytes of the instructions we need to copy 774 // to the trampoline. 775 instructions_length = RoundUpToInstrBoundary(kBranchLength, old_func); 776 if (!instructions_length) 777 return false; 778 779 // Put the needed instructions into the trampoline bytes. 780 uptr trampoline_length = instructions_length + kDirectBranchLength; 781 uptr trampoline = AllocateMemoryForTrampoline(old_func, trampoline_length); 782 if (!trampoline) 783 return false; 784 if (!CopyInstructions(trampoline, old_func, instructions_length)) 785 return false; 786 WriteDirectBranch(trampoline + instructions_length, 787 old_func + instructions_length); 788 *orig_old_func = trampoline; 789 } 790 791 #if SANITIZER_WINDOWS64 792 // Check if the targeted address can be encoded in the function padding. 793 // Otherwise, allocate it in the trampoline region. 794 if (IsMemoryPadding(old_func - kAddressLength, kAddressLength)) { 795 indirect_address = old_func - kAddressLength; 796 padding_length = kAddressLength; 797 } else { 798 indirect_address = AllocateMemoryForTrampoline(old_func, kAddressLength); 799 if (!indirect_address) 800 return false; 801 } 802 #endif 803 804 // Change memory protection to writable. 805 uptr patch_address = old_func - padding_length; 806 uptr patch_length = instructions_length + padding_length; 807 DWORD protection = 0; 808 if (!ChangeMemoryProtection(patch_address, patch_length, &protection)) 809 return false; 810 811 // Patch the original function. 812 WriteBranch(old_func, indirect_address, new_func); 813 814 // Restore previous memory protection. 815 if (!RestoreMemoryProtection(patch_address, patch_length, protection)) 816 return false; 817 818 return true; 819 } 820 821 bool OverrideFunction( 822 uptr old_func, uptr new_func, uptr *orig_old_func) { 823 #if !SANITIZER_WINDOWS64 824 if (OverrideFunctionWithDetour(old_func, new_func, orig_old_func)) 825 return true; 826 #endif 827 if (OverrideFunctionWithRedirectJump(old_func, new_func, orig_old_func)) 828 return true; 829 if (OverrideFunctionWithHotPatch(old_func, new_func, orig_old_func)) 830 return true; 831 if (OverrideFunctionWithTrampoline(old_func, new_func, orig_old_func)) 832 return true; 833 return false; 834 } 835 836 static void **InterestingDLLsAvailable() { 837 static const char *InterestingDLLs[] = { 838 "kernel32.dll", 839 "msvcr100.dll", // VS2010 840 "msvcr110.dll", // VS2012 841 "msvcr120.dll", // VS2013 842 "vcruntime140.dll", // VS2015 843 "ucrtbase.dll", // Universal CRT 844 // NTDLL should go last as it exports some functions that we should 845 // override in the CRT [presumably only used internally]. 846 "ntdll.dll", NULL}; 847 static void *result[ARRAY_SIZE(InterestingDLLs)] = { 0 }; 848 if (!result[0]) { 849 for (size_t i = 0, j = 0; InterestingDLLs[i]; ++i) { 850 if (HMODULE h = GetModuleHandleA(InterestingDLLs[i])) 851 result[j++] = (void *)h; 852 } 853 } 854 return &result[0]; 855 } 856 857 namespace { 858 // Utility for reading loaded PE images. 859 template <typename T> class RVAPtr { 860 public: 861 RVAPtr(void *module, uptr rva) 862 : ptr_(reinterpret_cast<T *>(reinterpret_cast<char *>(module) + rva)) {} 863 operator T *() { return ptr_; } 864 T *operator->() { return ptr_; } 865 T *operator++() { return ++ptr_; } 866 867 private: 868 T *ptr_; 869 }; 870 } // namespace 871 872 // Internal implementation of GetProcAddress. At least since Windows 8, 873 // GetProcAddress appears to initialize DLLs before returning function pointers 874 // into them. This is problematic for the sanitizers, because they typically 875 // want to intercept malloc *before* MSVCRT initializes. Our internal 876 // implementation walks the export list manually without doing initialization. 877 uptr InternalGetProcAddress(void *module, const char *func_name) { 878 // Check that the module header is full and present. 879 RVAPtr<IMAGE_DOS_HEADER> dos_stub(module, 0); 880 RVAPtr<IMAGE_NT_HEADERS> headers(module, dos_stub->e_lfanew); 881 if (!module || dos_stub->e_magic != IMAGE_DOS_SIGNATURE || // "MZ" 882 headers->Signature != IMAGE_NT_SIGNATURE || // "PE\0\0" 883 headers->FileHeader.SizeOfOptionalHeader < 884 sizeof(IMAGE_OPTIONAL_HEADER)) { 885 return 0; 886 } 887 888 IMAGE_DATA_DIRECTORY *export_directory = 889 &headers->OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_EXPORT]; 890 if (export_directory->Size == 0) 891 return 0; 892 RVAPtr<IMAGE_EXPORT_DIRECTORY> exports(module, 893 export_directory->VirtualAddress); 894 RVAPtr<DWORD> functions(module, exports->AddressOfFunctions); 895 RVAPtr<DWORD> names(module, exports->AddressOfNames); 896 RVAPtr<WORD> ordinals(module, exports->AddressOfNameOrdinals); 897 898 for (DWORD i = 0; i < exports->NumberOfNames; i++) { 899 RVAPtr<char> name(module, names[i]); 900 if (!strcmp(func_name, name)) { 901 DWORD index = ordinals[i]; 902 RVAPtr<char> func(module, functions[index]); 903 904 // Handle forwarded functions. 905 DWORD offset = functions[index]; 906 if (offset >= export_directory->VirtualAddress && 907 offset < export_directory->VirtualAddress + export_directory->Size) { 908 // An entry for a forwarded function is a string with the following 909 // format: "<module> . <function_name>" that is stored into the 910 // exported directory. 911 char function_name[256]; 912 size_t funtion_name_length = _strlen(func); 913 if (funtion_name_length >= sizeof(function_name) - 1) 914 InterceptionFailed(); 915 916 _memcpy(function_name, func, funtion_name_length); 917 function_name[funtion_name_length] = '\0'; 918 char* separator = _strchr(function_name, '.'); 919 if (!separator) 920 InterceptionFailed(); 921 *separator = '\0'; 922 923 void* redirected_module = GetModuleHandleA(function_name); 924 if (!redirected_module) 925 InterceptionFailed(); 926 return InternalGetProcAddress(redirected_module, separator + 1); 927 } 928 929 return (uptr)(char *)func; 930 } 931 } 932 933 return 0; 934 } 935 936 bool OverrideFunction( 937 const char *func_name, uptr new_func, uptr *orig_old_func) { 938 bool hooked = false; 939 void **DLLs = InterestingDLLsAvailable(); 940 for (size_t i = 0; DLLs[i]; ++i) { 941 uptr func_addr = InternalGetProcAddress(DLLs[i], func_name); 942 if (func_addr && 943 OverrideFunction(func_addr, new_func, orig_old_func)) { 944 hooked = true; 945 } 946 } 947 return hooked; 948 } 949 950 bool OverrideImportedFunction(const char *module_to_patch, 951 const char *imported_module, 952 const char *function_name, uptr new_function, 953 uptr *orig_old_func) { 954 HMODULE module = GetModuleHandleA(module_to_patch); 955 if (!module) 956 return false; 957 958 // Check that the module header is full and present. 959 RVAPtr<IMAGE_DOS_HEADER> dos_stub(module, 0); 960 RVAPtr<IMAGE_NT_HEADERS> headers(module, dos_stub->e_lfanew); 961 if (!module || dos_stub->e_magic != IMAGE_DOS_SIGNATURE || // "MZ" 962 headers->Signature != IMAGE_NT_SIGNATURE || // "PE\0\0" 963 headers->FileHeader.SizeOfOptionalHeader < 964 sizeof(IMAGE_OPTIONAL_HEADER)) { 965 return false; 966 } 967 968 IMAGE_DATA_DIRECTORY *import_directory = 969 &headers->OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_IMPORT]; 970 971 // Iterate the list of imported DLLs. FirstThunk will be null for the last 972 // entry. 973 RVAPtr<IMAGE_IMPORT_DESCRIPTOR> imports(module, 974 import_directory->VirtualAddress); 975 for (; imports->FirstThunk != 0; ++imports) { 976 RVAPtr<const char> modname(module, imports->Name); 977 if (_stricmp(&*modname, imported_module) == 0) 978 break; 979 } 980 if (imports->FirstThunk == 0) 981 return false; 982 983 // We have two parallel arrays: the import address table (IAT) and the table 984 // of names. They start out containing the same data, but the loader rewrites 985 // the IAT to hold imported addresses and leaves the name table in 986 // OriginalFirstThunk alone. 987 RVAPtr<IMAGE_THUNK_DATA> name_table(module, imports->OriginalFirstThunk); 988 RVAPtr<IMAGE_THUNK_DATA> iat(module, imports->FirstThunk); 989 for (; name_table->u1.Ordinal != 0; ++name_table, ++iat) { 990 if (!IMAGE_SNAP_BY_ORDINAL(name_table->u1.Ordinal)) { 991 RVAPtr<IMAGE_IMPORT_BY_NAME> import_by_name( 992 module, name_table->u1.ForwarderString); 993 const char *funcname = &import_by_name->Name[0]; 994 if (strcmp(funcname, function_name) == 0) 995 break; 996 } 997 } 998 if (name_table->u1.Ordinal == 0) 999 return false; 1000 1001 // Now we have the correct IAT entry. Do the swap. We have to make the page 1002 // read/write first. 1003 if (orig_old_func) 1004 *orig_old_func = iat->u1.AddressOfData; 1005 DWORD old_prot, unused_prot; 1006 if (!VirtualProtect(&iat->u1.AddressOfData, 4, PAGE_EXECUTE_READWRITE, 1007 &old_prot)) 1008 return false; 1009 iat->u1.AddressOfData = new_function; 1010 if (!VirtualProtect(&iat->u1.AddressOfData, 4, old_prot, &unused_prot)) 1011 return false; // Not clear if this failure bothers us. 1012 return true; 1013 } 1014 1015 } // namespace __interception 1016 1017 #endif // SANITIZER_MAC 1018