libsanitizer/interception/interception_win.cpp

   1 //===-- interception_win.cpp ------------------------------------*- C++ -*-===//
   2 //
   3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
   4 // See https://llvm.org/LICENSE.txt for license information.
   5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
   6 //
   7 //===----------------------------------------------------------------------===//
   8 //
   9 // This file is a part of AddressSanitizer, an address sanity checker.
  10 //
  11 // Windows-specific interception methods.
  12 //
  13 // This file is implementing several hooking techniques to intercept calls
  14 // to functions. The hooks are dynamically installed by modifying the assembly
  15 // code.
  16 //
  17 // The hooking techniques are making assumptions on the way the code is
  18 // generated and are safe under these assumptions.
  19 //
  20 // On 64-bit architecture, there is no direct 64-bit jump instruction. To allow
  21 // arbitrary branching on the whole memory space, the notion of trampoline
  22 // region is used. A trampoline region is a memory space withing 2G boundary
  23 // where it is safe to add custom assembly code to build 64-bit jumps.
  24 //
  25 // Hooking techniques
  26 // ==================
  27 //
  28 // 1) Detour
  29 //
  30 //    The Detour hooking technique is assuming the presence of an header with
  31 //    padding and an overridable 2-bytes nop instruction (mov edi, edi). The
  32 //    nop instruction can safely be replaced by a 2-bytes jump without any need
  33 //    to save the instruction. A jump to the target is encoded in the function
  34 //    header and the nop instruction is replaced by a short jump to the header.
  35 //
  36 //        head:  5 x nop                 head:  jmp <hook>
  37 //        func:  mov edi, edi    -->     func:  jmp short <head>
  38 //               [...]                   real:  [...]
  39 //
  40 //    This technique is only implemented on 32-bit architecture.
  41 //    Most of the time, Windows API are hookable with the detour technique.
  42 //
  43 // 2) Redirect Jump
  44 //
  45 //    The redirect jump is applicable when the first instruction is a direct
  46 //    jump. The instruction is replaced by jump to the hook.
  47 //
  48 //        func:  jmp <label>     -->     func:  jmp <hook>
  49 //
  50 //    On an 64-bit architecture, a trampoline is inserted.
  51 //
  52 //        func:  jmp <label>     -->     func:  jmp <tramp>
  53 //                                              [...]
  54 //
  55 //                                   [trampoline]
  56 //                                      tramp:  jmp QWORD [addr]
  57 //                                       addr:  .bytes <hook>
  58 //
  59 //    Note: <real> is equivalent to <label>.
  60 //
  61 // 3) HotPatch
  62 //
  63 //    The HotPatch hooking is assuming the presence of an header with padding
  64 //    and a first instruction with at least 2-bytes.
  65 //
  66 //    The reason to enforce the 2-bytes limitation is to provide the minimal
  67 //    space to encode a short jump. HotPatch technique is only rewriting one
  68 //    instruction to avoid breaking a sequence of instructions containing a
  69 //    branching target.
  70 //
  71 //    Assumptions are enforced by MSVC compiler by using the /HOTPATCH flag.
  72 //      see: https://msdn.microsoft.com/en-us/library/ms173507.aspx
  73 //    Default padding length is 5 bytes in 32-bits and 6 bytes in 64-bits.
  74 //
  75 //        head:   5 x nop                head:  jmp <hook>
  76 //        func:   <instr>        -->     func:  jmp short <head>
  77 //                [...]                  body:  [...]
  78 //
  79 //                                   [trampoline]
  80 //                                       real:  <instr>
  81 //                                              jmp <body>
  82 //
  83 //    On an 64-bit architecture:
  84 //
  85 //        head:   6 x nop                head:  jmp QWORD [addr1]
  86 //        func:   <instr>        -->     func:  jmp short <head>
  87 //                [...]                  body:  [...]
  88 //
  89 //                                   [trampoline]
  90 //                                      addr1:  .bytes <hook>
  91 //                                       real:  <instr>
  92 //                                              jmp QWORD [addr2]
  93 //                                      addr2:  .bytes <body>
  94 //
  95 // 4) Trampoline
  96 //
  97 //    The Trampoline hooking technique is the most aggressive one. It is
  98 //    assuming that there is a sequence of instructions that can be safely
  99 //    replaced by a jump (enough room and no incoming branches).
 100 //
 101 //    Unfortunately, these assumptions can't be safely presumed and code may
 102 //    be broken after hooking.
 103 //
 104 //        func:   <instr>        -->     func:  jmp <hook>
 105 //                <instr>
 106 //                [...]                  body:  [...]
 107 //
 108 //                                   [trampoline]
 109 //                                       real:  <instr>
 110 //                                              <instr>
 111 //                                              jmp <body>
 112 //
 113 //    On an 64-bit architecture:
 114 //
 115 //        func:   <instr>        -->     func:  jmp QWORD [addr1]
 116 //                <instr>
 117 //                [...]                  body:  [...]
 118 //
 119 //                                   [trampoline]
 120 //                                      addr1:  .bytes <hook>
 121 //                                       real:  <instr>
 122 //                                              <instr>
 123 //                                              jmp QWORD [addr2]
 124 //                                      addr2:  .bytes <body>
 125 //===----------------------------------------------------------------------===//
 126
 127 #include "interception.h"
 128
 129 #if SANITIZER_WINDOWS
 130 #include "sanitizer_common/sanitizer_platform.h"
 131 #define WIN32_LEAN_AND_MEAN
 132 #include <windows.h>
 133
 134 namespace __interception {
 135
 136 static const int kAddressLength = FIRST_32_SECOND_64(4, 8);
 137 static const int kJumpInstructionLength = 5;
 138 static const int kShortJumpInstructionLength = 2;
 139 UNUSED static const int kIndirectJumpInstructionLength = 6;
 140 static const int kBranchLength =
 141     FIRST_32_SECOND_64(kJumpInstructionLength, kIndirectJumpInstructionLength);
 142 static const int kDirectBranchLength = kBranchLength + kAddressLength;
 143
 144 #  if defined(_MSC_VER)
 145 #    define INTERCEPTION_FORMAT(f, a)
 146 #  else
 147 #    define INTERCEPTION_FORMAT(f, a) __attribute__((format(printf, f, a)))
 148 #  endif
 149
 150 static void (*ErrorReportCallback)(const char *format, ...)
 151     INTERCEPTION_FORMAT(1, 2);
 152
 153 void SetErrorReportCallback(void (*callback)(const char *format, ...)) {
 154   ErrorReportCallback = callback;
 155 }
 156
 157 #  define ReportError(...)                \
 158     do {                                  \
 159       if (ErrorReportCallback)            \
 160         ErrorReportCallback(__VA_ARGS__); \
 161     } while (0)
 162
 163 static void InterceptionFailed() {
 164   ReportError("interception_win: failed due to an unrecoverable error.\n");
 165   // This acts like an abort when no debugger is attached. According to an old
 166   // comment, calling abort() leads to an infinite recursion in CheckFailed.
 167   __debugbreak();
 168 }
 169
 170 static bool DistanceIsWithin2Gig(uptr from, uptr target) {
 171 #if SANITIZER_WINDOWS64
 172   if (from < target)
 173     return target - from <= (uptr)0x7FFFFFFFU;
 174   else
 175     return from - target <= (uptr)0x80000000U;
 176 #else
 177   // In a 32-bit address space, the address calculation will wrap, so this check
 178   // is unnecessary.
 179   return true;
 180 #endif
 181 }
 182
 183 static uptr GetMmapGranularity() {
 184   SYSTEM_INFO si;
 185   GetSystemInfo(&si);
 186   return si.dwAllocationGranularity;
 187 }
 188
 189 UNUSED static uptr RoundUpTo(uptr size, uptr boundary) {
 190   return (size + boundary - 1) & ~(boundary - 1);
 191 }
 192
 193 // FIXME: internal_str* and internal_mem* functions should be moved from the
 194 // ASan sources into interception/.
 195
 196 static size_t _strlen(const char *str) {
 197   const char* p = str;
 198   while (*p != '\0') ++p;
 199   return p - str;
 200 }
 201
 202 static char* _strchr(char* str, char c) {
 203   while (*str) {
 204     if (*str == c)
 205       return str;
 206     ++str;
 207   }
 208   return nullptr;
 209 }
 210
 211 static void _memset(void *p, int value, size_t sz) {
 212   for (size_t i = 0; i < sz; ++i)
 213     ((char*)p)[i] = (char)value;
 214 }
 215
 216 static void _memcpy(void *dst, void *src, size_t sz) {
 217   char *dst_c = (char*)dst,
 218        *src_c = (char*)src;
 219   for (size_t i = 0; i < sz; ++i)
 220     dst_c[i] = src_c[i];
 221 }
 222
 223 static bool ChangeMemoryProtection(
 224     uptr address, uptr size, DWORD *old_protection) {
 225   return ::VirtualProtect((void*)address, size,
 226                           PAGE_EXECUTE_READWRITE,
 227                           old_protection) != FALSE;
 228 }
 229
 230 static bool RestoreMemoryProtection(
 231     uptr address, uptr size, DWORD old_protection) {
 232   DWORD unused;
 233   return ::VirtualProtect((void*)address, size,
 234                           old_protection,
 235                           &unused) != FALSE;
 236 }
 237
 238 static bool IsMemoryPadding(uptr address, uptr size) {
 239   u8* function = (u8*)address;
 240   for (size_t i = 0; i < size; ++i)
 241     if (function[i] != 0x90 && function[i] != 0xCC)
 242       return false;
 243   return true;
 244 }
 245
 246 static const u8 kHintNop8Bytes[] = {
 247   0x0F, 0x1F, 0x84, 0x00, 0x00, 0x00, 0x00, 0x00
 248 };
 249
 250 template<class T>
 251 static bool FunctionHasPrefix(uptr address, const T &pattern) {
 252   u8* function = (u8*)address - sizeof(pattern);
 253   for (size_t i = 0; i < sizeof(pattern); ++i)
 254     if (function[i] != pattern[i])
 255       return false;
 256   return true;
 257 }
 258
 259 static bool FunctionHasPadding(uptr address, uptr size) {
 260   if (IsMemoryPadding(address - size, size))
 261     return true;
 262   if (size <= sizeof(kHintNop8Bytes) &&
 263       FunctionHasPrefix(address, kHintNop8Bytes))
 264     return true;
 265   return false;
 266 }
 267
 268 static void WritePadding(uptr from, uptr size) {
 269   _memset((void*)from, 0xCC, (size_t)size);
 270 }
 271
 272 static void WriteJumpInstruction(uptr from, uptr target) {
 273   if (!DistanceIsWithin2Gig(from + kJumpInstructionLength, target)) {
 274     ReportError(
 275         "interception_win: cannot write jmp further than 2GB away, from %p to "
 276         "%p.\n",
 277         (void *)from, (void *)target);
 278     InterceptionFailed();
 279   }
 280   ptrdiff_t offset = target - from - kJumpInstructionLength;
 281   *(u8*)from = 0xE9;
 282   *(u32*)(from + 1) = offset;
 283 }
 284
 285 static void WriteShortJumpInstruction(uptr from, uptr target) {
 286   sptr offset = target - from - kShortJumpInstructionLength;
 287   if (offset < -128 || offset > 127)
 288     InterceptionFailed();
 289   *(u8*)from = 0xEB;
 290   *(u8*)(from + 1) = (u8)offset;
 291 }
 292
 293 #if SANITIZER_WINDOWS64
 294 static void WriteIndirectJumpInstruction(uptr from, uptr indirect_target) {
 295   // jmp [rip + <offset>] = FF 25 <offset> where <offset> is a relative
 296   // offset.
 297   // The offset is the distance from then end of the jump instruction to the
 298   // memory location containing the targeted address. The displacement is still
 299   // 32-bit in x64, so indirect_target must be located within +/- 2GB range.
 300   int offset = indirect_target - from - kIndirectJumpInstructionLength;
 301   if (!DistanceIsWithin2Gig(from + kIndirectJumpInstructionLength,
 302                             indirect_target)) {
 303     ReportError(
 304         "interception_win: cannot write indirect jmp with target further than "
 305         "2GB away, from %p to %p.\n",
 306         (void *)from, (void *)indirect_target);
 307     InterceptionFailed();
 308   }
 309   *(u16*)from = 0x25FF;
 310   *(u32*)(from + 2) = offset;
 311 }
 312 #endif
 313
 314 static void WriteBranch(
 315     uptr from, uptr indirect_target, uptr target) {
 316 #if SANITIZER_WINDOWS64
 317   WriteIndirectJumpInstruction(from, indirect_target);
 318   *(u64*)indirect_target = target;
 319 #else
 320   (void)indirect_target;
 321   WriteJumpInstruction(from, target);
 322 #endif
 323 }
 324
 325 static void WriteDirectBranch(uptr from, uptr target) {
 326 #if SANITIZER_WINDOWS64
 327   // Emit an indirect jump through immediately following bytes:
 328   //   jmp [rip + kBranchLength]
 329   //   .quad <target>
 330   WriteBranch(from, from + kBranchLength, target);
 331 #else
 332   WriteJumpInstruction(from, target);
 333 #endif
 334 }
 335
 336 struct TrampolineMemoryRegion {
 337   uptr content;
 338   uptr allocated_size;
 339   uptr max_size;
 340 };
 341
 342 UNUSED static const uptr kTrampolineScanLimitRange = 1 << 31;  // 2 gig
 343 static const int kMaxTrampolineRegion = 1024;
 344 static TrampolineMemoryRegion TrampolineRegions[kMaxTrampolineRegion];
 345
 346 static void *AllocateTrampolineRegion(uptr image_address, size_t granularity) {
 347 #if SANITIZER_WINDOWS64
 348   uptr address = image_address;
 349   uptr scanned = 0;
 350   while (scanned < kTrampolineScanLimitRange) {
 351     MEMORY_BASIC_INFORMATION info;
 352     if (!::VirtualQuery((void*)address, &info, sizeof(info)))
 353       return nullptr;
 354
 355     // Check whether a region can be allocated at |address|.
 356     if (info.State == MEM_FREE && info.RegionSize >= granularity) {
 357       void *page = ::VirtualAlloc((void*)RoundUpTo(address, granularity),
 358                                   granularity,
 359                                   MEM_RESERVE | MEM_COMMIT,
 360                                   PAGE_EXECUTE_READWRITE);
 361       return page;
 362     }
 363
 364     // Move to the next region.
 365     address = (uptr)info.BaseAddress + info.RegionSize;
 366     scanned += info.RegionSize;
 367   }
 368   return nullptr;
 369 #else
 370   return ::VirtualAlloc(nullptr,
 371                         granularity,
 372                         MEM_RESERVE | MEM_COMMIT,
 373                         PAGE_EXECUTE_READWRITE);
 374 #endif
 375 }
 376
 377 // Used by unittests to release mapped memory space.
 378 void TestOnlyReleaseTrampolineRegions() {
 379   for (size_t bucket = 0; bucket < kMaxTrampolineRegion; ++bucket) {
 380     TrampolineMemoryRegion *current = &TrampolineRegions[bucket];
 381     if (current->content == 0)
 382       return;
 383     ::VirtualFree((void*)current->content, 0, MEM_RELEASE);
 384     current->content = 0;
 385   }
 386 }
 387
 388 static uptr AllocateMemoryForTrampoline(uptr image_address, size_t size) {
 389   // Find a region within 2G with enough space to allocate |size| bytes.
 390   TrampolineMemoryRegion *region = nullptr;
 391   for (size_t bucket = 0; bucket < kMaxTrampolineRegion; ++bucket) {
 392     TrampolineMemoryRegion* current = &TrampolineRegions[bucket];
 393     if (current->content == 0) {
 394       // No valid region found, allocate a new region.
 395       size_t bucket_size = GetMmapGranularity();
 396       void *content = AllocateTrampolineRegion(image_address, bucket_size);
 397       if (content == nullptr)
 398         return 0U;
 399
 400       current->content = (uptr)content;
 401       current->allocated_size = 0;
 402       current->max_size = bucket_size;
 403       region = current;
 404       break;
 405     } else if (current->max_size - current->allocated_size > size) {
 406 #if SANITIZER_WINDOWS64
 407         // In 64-bits, the memory space must be allocated within 2G boundary.
 408         uptr next_address = current->content + current->allocated_size;
 409         if (next_address < image_address ||
 410             next_address - image_address >= 0x7FFF0000)
 411           continue;
 412 #endif
 413       // The space can be allocated in the current region.
 414       region = current;
 415       break;
 416     }
 417   }
 418
 419   // Failed to find a region.
 420   if (region == nullptr)
 421     return 0U;
 422
 423   // Allocate the space in the current region.
 424   uptr allocated_space = region->content + region->allocated_size;
 425   region->allocated_size += size;
 426   WritePadding(allocated_space, size);
 427
 428   return allocated_space;
 429 }
 430
 431 // The following prologues cannot be patched because of the short jump
 432 // jumping to the patching region.
 433
 434 #if SANITIZER_WINDOWS64
 435 // ntdll!wcslen in Win11
 436 //   488bc1          mov     rax,rcx
 437 //   0fb710          movzx   edx,word ptr [rax]
 438 //   4883c002        add     rax,2
 439 //   6685d2          test    dx,dx
 440 //   75f4            jne     -12
 441 static const u8 kPrologueWithShortJump1[] = {
 442     0x48, 0x8b, 0xc1, 0x0f, 0xb7, 0x10, 0x48, 0x83,
 443     0xc0, 0x02, 0x66, 0x85, 0xd2, 0x75, 0xf4,
 444 };
 445
 446 // ntdll!strrchr in Win11
 447 //   4c8bc1          mov     r8,rcx
 448 //   8a01            mov     al,byte ptr [rcx]
 449 //   48ffc1          inc     rcx
 450 //   84c0            test    al,al
 451 //   75f7            jne     -9
 452 static const u8 kPrologueWithShortJump2[] = {
 453     0x4c, 0x8b, 0xc1, 0x8a, 0x01, 0x48, 0xff, 0xc1,
 454     0x84, 0xc0, 0x75, 0xf7,
 455 };
 456 #endif
 457
 458 // Returns 0 on error.
 459 static size_t GetInstructionSize(uptr address, size_t* rel_offset = nullptr) {
 460 #if SANITIZER_ARM64
 461   // An ARM64 instruction is 4 bytes long.
 462   return 4;
 463 #endif
 464
 465 #if SANITIZER_WINDOWS64
 466   if (memcmp((u8*)address, kPrologueWithShortJump1,
 467              sizeof(kPrologueWithShortJump1)) == 0 ||
 468       memcmp((u8*)address, kPrologueWithShortJump2,
 469              sizeof(kPrologueWithShortJump2)) == 0) {
 470     return 0;
 471   }
 472 #endif
 473
 474   switch (*(u64*)address) {
 475     case 0x90909090909006EB:  // stub: jmp over 6 x nop.
 476       return 8;
 477   }
 478
 479   switch (*(u8*)address) {
 480     case 0x90:  // 90 : nop
 481       return 1;
 482
 483     case 0x50:  // push eax / rax
 484     case 0x51:  // push ecx / rcx
 485     case 0x52:  // push edx / rdx
 486     case 0x53:  // push ebx / rbx
 487     case 0x54:  // push esp / rsp
 488     case 0x55:  // push ebp / rbp
 489     case 0x56:  // push esi / rsi
 490     case 0x57:  // push edi / rdi
 491     case 0x5D:  // pop ebp / rbp
 492       return 1;
 493
 494     case 0x6A:  // 6A XX = push XX
 495       return 2;
 496
 497     case 0xb8:  // b8 XX XX XX XX : mov eax, XX XX XX XX
 498     case 0xB9:  // b9 XX XX XX XX : mov ecx, XX XX XX XX
 499       return 5;
 500
 501     // Cannot overwrite control-instruction. Return 0 to indicate failure.
 502     case 0xE9:  // E9 XX XX XX XX : jmp <label>
 503     case 0xE8:  // E8 XX XX XX XX : call <func>
 504     case 0xC3:  // C3 : ret
 505     case 0xEB:  // EB XX : jmp XX (short jump)
 506     case 0x70:  // 7Y YY : jy XX (short conditional jump)
 507     case 0x71:
 508     case 0x72:
 509     case 0x73:
 510     case 0x74:
 511     case 0x75:
 512     case 0x76:
 513     case 0x77:
 514     case 0x78:
 515     case 0x79:
 516     case 0x7A:
 517     case 0x7B:
 518     case 0x7C:
 519     case 0x7D:
 520     case 0x7E:
 521     case 0x7F:
 522       return 0;
 523   }
 524
 525   switch (*(u16*)(address)) {
 526     case 0x018A:  // 8A 01 : mov al, byte ptr [ecx]
 527     case 0xFF8B:  // 8B FF : mov edi, edi
 528     case 0xEC8B:  // 8B EC : mov ebp, esp
 529     case 0xc889:  // 89 C8 : mov eax, ecx
 530     case 0xE589:  // 89 E5 : mov ebp, esp
 531     case 0xC18B:  // 8B C1 : mov eax, ecx
 532     case 0xC033:  // 33 C0 : xor eax, eax
 533     case 0xC933:  // 33 C9 : xor ecx, ecx
 534     case 0xD233:  // 33 D2 : xor edx, edx
 535       return 2;
 536
 537     // Cannot overwrite control-instruction. Return 0 to indicate failure.
 538     case 0x25FF:  // FF 25 XX XX XX XX : jmp [XXXXXXXX]
 539       return 0;
 540   }
 541
 542   switch (0x00FFFFFF & *(u32*)address) {
 543     case 0x24A48D:  // 8D A4 24 XX XX XX XX : lea esp, [esp + XX XX XX XX]
 544       return 7;
 545   }
 546
 547 #if SANITIZER_WINDOWS64
 548   switch (*(u8*)address) {
 549     case 0xA1:  // A1 XX XX XX XX XX XX XX XX :
 550                 //   movabs eax, dword ptr ds:[XXXXXXXX]
 551       return 9;
 552
 553     case 0x83:
 554       const u8 next_byte = *(u8*)(address + 1);
 555       const u8 mod = next_byte >> 6;
 556       const u8 rm = next_byte & 7;
 557       if (mod == 1 && rm == 4)
 558         return 5;  // 83 ModR/M SIB Disp8 Imm8
 559                    //   add|or|adc|sbb|and|sub|xor|cmp [r+disp8], imm8
 560   }
 561
 562   switch (*(u16*)address) {
 563     case 0x5040:  // push rax
 564     case 0x5140:  // push rcx
 565     case 0x5240:  // push rdx
 566     case 0x5340:  // push rbx
 567     case 0x5440:  // push rsp
 568     case 0x5540:  // push rbp
 569     case 0x5640:  // push rsi
 570     case 0x5740:  // push rdi
 571     case 0x5441:  // push r12
 572     case 0x5541:  // push r13
 573     case 0x5641:  // push r14
 574     case 0x5741:  // push r15
 575     case 0x9066:  // Two-byte NOP
 576     case 0xc084:  // test al, al
 577     case 0x018a:  // mov al, byte ptr [rcx]
 578       return 2;
 579
 580     case 0x058B:  // 8B 05 XX XX XX XX : mov eax, dword ptr [XX XX XX XX]
 581       if (rel_offset)
 582         *rel_offset = 2;
 583       return 6;
 584   }
 585
 586   switch (0x00FFFFFF & *(u32*)address) {
 587     case 0xe58948:    // 48 8b c4 : mov rbp, rsp
 588     case 0xc18b48:    // 48 8b c1 : mov rax, rcx
 589     case 0xc48b48:    // 48 8b c4 : mov rax, rsp
 590     case 0xd9f748:    // 48 f7 d9 : neg rcx
 591     case 0xd12b48:    // 48 2b d1 : sub rdx, rcx
 592     case 0x07c1f6:    // f6 c1 07 : test cl, 0x7
 593     case 0xc98548:    // 48 85 C9 : test rcx, rcx
 594     case 0xd28548:    // 48 85 d2 : test rdx, rdx
 595     case 0xc0854d:    // 4d 85 c0 : test r8, r8
 596     case 0xc2b60f:    // 0f b6 c2 : movzx eax, dl
 597     case 0xc03345:    // 45 33 c0 : xor r8d, r8d
 598     case 0xc93345:    // 45 33 c9 : xor r9d, r9d
 599     case 0xdb3345:    // 45 33 DB : xor r11d, r11d
 600     case 0xd98b4c:    // 4c 8b d9 : mov r11, rcx
 601     case 0xd28b4c:    // 4c 8b d2 : mov r10, rdx
 602     case 0xc98b4c:    // 4C 8B C9 : mov r9, rcx
 603     case 0xc18b4c:    // 4C 8B C1 : mov r8, rcx
 604     case 0xd2b60f:    // 0f b6 d2 : movzx edx, dl
 605     case 0xca2b48:    // 48 2b ca : sub rcx, rdx
 606     case 0x10b70f:    // 0f b7 10 : movzx edx, WORD PTR [rax]
 607     case 0xc00b4d:    // 3d 0b c0 : or r8, r8
 608     case 0xc08b41:    // 41 8b c0 : mov eax, r8d
 609     case 0xd18b48:    // 48 8b d1 : mov rdx, rcx
 610     case 0xdc8b4c:    // 4c 8b dc : mov r11, rsp
 611     case 0xd18b4c:    // 4c 8b d1 : mov r10, rcx
 612     case 0xE0E483:    // 83 E4 E0 : and esp, 0xFFFFFFE0
 613       return 3;
 614
 615     case 0xec8348:    // 48 83 ec XX : sub rsp, XX
 616     case 0xf88349:    // 49 83 f8 XX : cmp r8, XX
 617     case 0x588948:    // 48 89 58 XX : mov QWORD PTR[rax + XX], rbx
 618       return 4;
 619
 620     case 0xec8148:    // 48 81 EC XX XX XX XX : sub rsp, XXXXXXXX
 621       return 7;
 622
 623     case 0x058b48:    // 48 8b 05 XX XX XX XX :
 624                       //   mov rax, QWORD PTR [rip + XXXXXXXX]
 625     case 0x25ff48:    // 48 ff 25 XX XX XX XX :
 626                       //   rex.W jmp QWORD PTR [rip + XXXXXXXX]
 627     case 0x158D4C:    // 4c 8d 15 XX XX XX XX : lea r10, [rip + XX]
 628       // Instructions having offset relative to 'rip' need offset adjustment.
 629       if (rel_offset)
 630         *rel_offset = 3;
 631       return 7;
 632
 633     case 0x2444c7:    // C7 44 24 XX YY YY YY YY
 634                       //   mov dword ptr [rsp + XX], YYYYYYYY
 635       return 8;
 636   }
 637
 638   switch (*(u32*)(address)) {
 639     case 0x24448b48:  // 48 8b 44 24 XX : mov rax, QWORD ptr [rsp + XX]
 640     case 0x246c8948:  // 48 89 6C 24 XX : mov QWORD ptr [rsp + XX], rbp
 641     case 0x245c8948:  // 48 89 5c 24 XX : mov QWORD PTR [rsp + XX], rbx
 642     case 0x24748948:  // 48 89 74 24 XX : mov QWORD PTR [rsp + XX], rsi
 643     case 0x247c8948:  // 48 89 7c 24 XX : mov QWORD PTR [rsp + XX], rdi
 644     case 0x244C8948:  // 48 89 4C 24 XX : mov QWORD PTR [rsp + XX], rcx
 645     case 0x24548948:  // 48 89 54 24 XX : mov QWORD PTR [rsp + XX], rdx
 646     case 0x244c894c:  // 4c 89 4c 24 XX : mov QWORD PTR [rsp + XX], r9
 647     case 0x2444894c:  // 4c 89 44 24 XX : mov QWORD PTR [rsp + XX], r8
 648       return 5;
 649     case 0x24648348:  // 48 83 64 24 XX : and QWORD PTR [rsp + XX], YY
 650       return 6;
 651   }
 652
 653 #else
 654
 655   switch (*(u8*)address) {
 656     case 0xA1:  // A1 XX XX XX XX :  mov eax, dword ptr ds:[XXXXXXXX]
 657       return 5;
 658   }
 659   switch (*(u16*)address) {
 660     case 0x458B:  // 8B 45 XX : mov eax, dword ptr [ebp + XX]
 661     case 0x5D8B:  // 8B 5D XX : mov ebx, dword ptr [ebp + XX]
 662     case 0x7D8B:  // 8B 7D XX : mov edi, dword ptr [ebp + XX]
 663     case 0xEC83:  // 83 EC XX : sub esp, XX
 664     case 0x75FF:  // FF 75 XX : push dword ptr [ebp + XX]
 665       return 3;
 666     case 0xC1F7:  // F7 C1 XX YY ZZ WW : test ecx, WWZZYYXX
 667     case 0x25FF:  // FF 25 XX YY ZZ WW : jmp dword ptr ds:[WWZZYYXX]
 668       return 6;
 669     case 0x3D83:  // 83 3D XX YY ZZ WW TT : cmp TT, WWZZYYXX
 670       return 7;
 671     case 0x7D83:  // 83 7D XX YY : cmp dword ptr [ebp + XX], YY
 672       return 4;
 673   }
 674
 675   switch (0x00FFFFFF & *(u32*)address) {
 676     case 0x24448A:  // 8A 44 24 XX : mov eal, dword ptr [esp + XX]
 677     case 0x24448B:  // 8B 44 24 XX : mov eax, dword ptr [esp + XX]
 678     case 0x244C8B:  // 8B 4C 24 XX : mov ecx, dword ptr [esp + XX]
 679     case 0x24548B:  // 8B 54 24 XX : mov edx, dword ptr [esp + XX]
 680     case 0x245C8B:  // 8B 5C 24 XX : mov ebx, dword ptr [esp + XX]
 681     case 0x246C8B:  // 8B 6C 24 XX : mov ebp, dword ptr [esp + XX]
 682     case 0x24748B:  // 8B 74 24 XX : mov esi, dword ptr [esp + XX]
 683     case 0x247C8B:  // 8B 7C 24 XX : mov edi, dword ptr [esp + XX]
 684       return 4;
 685   }
 686
 687   switch (*(u32*)address) {
 688     case 0x2444B60F:  // 0F B6 44 24 XX : movzx eax, byte ptr [esp + XX]
 689       return 5;
 690   }
 691 #endif
 692
 693   // Unknown instruction! This might happen when we add a new interceptor, use
 694   // a new compiler version, or if Windows changed how some functions are
 695   // compiled. In either case, we print the address and 8 bytes of instructions
 696   // to notify the user about the error and to help identify the unknown
 697   // instruction. Don't treat this as a fatal error, though we can break the
 698   // debugger if one has been attached.
 699   u8 *bytes = (u8 *)address;
 700   ReportError(
 701       "interception_win: unhandled instruction at %p: %02x %02x %02x %02x %02x "
 702       "%02x %02x %02x\n",
 703       (void *)address, bytes[0], bytes[1], bytes[2], bytes[3], bytes[4],
 704       bytes[5], bytes[6], bytes[7]);
 705   if (::IsDebuggerPresent())
 706     __debugbreak();
 707   return 0;
 708 }
 709
 710 // Returns 0 on error.
 711 static size_t RoundUpToInstrBoundary(size_t size, uptr address) {
 712   size_t cursor = 0;
 713   while (cursor < size) {
 714     size_t instruction_size = GetInstructionSize(address + cursor);
 715     if (!instruction_size)
 716       return 0;
 717     cursor += instruction_size;
 718   }
 719   return cursor;
 720 }
 721
 722 static bool CopyInstructions(uptr to, uptr from, size_t size) {
 723   size_t cursor = 0;
 724   while (cursor != size) {
 725     size_t rel_offset = 0;
 726     size_t instruction_size = GetInstructionSize(from + cursor, &rel_offset);
 727     if (!instruction_size)
 728       return false;
 729     _memcpy((void *)(to + cursor), (void *)(from + cursor),
 730             (size_t)instruction_size);
 731     if (rel_offset) {
 732 #  if SANITIZER_WINDOWS64
 733       // we want to make sure that the new relative offset still fits in 32-bits
 734       // this will be untrue if relocated_offset \notin [-2**31, 2**31)
 735       s64 delta = to - from;
 736       s64 relocated_offset = *(s32 *)(to + cursor + rel_offset) - delta;
 737       if (-0x8000'0000ll > relocated_offset || relocated_offset > 0x7FFF'FFFFll)
 738         return false;
 739 #  else
 740       // on 32-bit, the relative offset will always be correct
 741       s32 delta = to - from;
 742       s32 relocated_offset = *(s32 *)(to + cursor + rel_offset) - delta;
 743 #  endif
 744       *(s32 *)(to + cursor + rel_offset) = relocated_offset;
 745     }
 746     cursor += instruction_size;
 747   }
 748   return true;
 749 }
 750
 751
 752 #if !SANITIZER_WINDOWS64
 753 bool OverrideFunctionWithDetour(
 754     uptr old_func, uptr new_func, uptr *orig_old_func) {
 755   const int kDetourHeaderLen = 5;
 756   const u16 kDetourInstruction = 0xFF8B;
 757
 758   uptr header = (uptr)old_func - kDetourHeaderLen;
 759   uptr patch_length = kDetourHeaderLen + kShortJumpInstructionLength;
 760
 761   // Validate that the function is hookable.
 762   if (*(u16*)old_func != kDetourInstruction ||
 763       !IsMemoryPadding(header, kDetourHeaderLen))
 764     return false;
 765
 766   // Change memory protection to writable.
 767   DWORD protection = 0;
 768   if (!ChangeMemoryProtection(header, patch_length, &protection))
 769     return false;
 770
 771   // Write a relative jump to the redirected function.
 772   WriteJumpInstruction(header, new_func);
 773
 774   // Write the short jump to the function prefix.
 775   WriteShortJumpInstruction(old_func, header);
 776
 777   // Restore previous memory protection.
 778   if (!RestoreMemoryProtection(header, patch_length, protection))
 779     return false;
 780
 781   if (orig_old_func)
 782     *orig_old_func = old_func + kShortJumpInstructionLength;
 783
 784   return true;
 785 }
 786 #endif
 787
 788 bool OverrideFunctionWithRedirectJump(
 789     uptr old_func, uptr new_func, uptr *orig_old_func) {
 790   // Check whether the first instruction is a relative jump.
 791   if (*(u8*)old_func != 0xE9)
 792     return false;
 793
 794   if (orig_old_func) {
 795     sptr relative_offset = *(s32 *)(old_func + 1);
 796     uptr absolute_target = old_func + relative_offset + kJumpInstructionLength;
 797     *orig_old_func = absolute_target;
 798   }
 799
 800 #if SANITIZER_WINDOWS64
 801   // If needed, get memory space for a trampoline jump.
 802   uptr trampoline = AllocateMemoryForTrampoline(old_func, kDirectBranchLength);
 803   if (!trampoline)
 804     return false;
 805   WriteDirectBranch(trampoline, new_func);
 806 #endif
 807
 808   // Change memory protection to writable.
 809   DWORD protection = 0;
 810   if (!ChangeMemoryProtection(old_func, kJumpInstructionLength, &protection))
 811     return false;
 812
 813   // Write a relative jump to the redirected function.
 814   WriteJumpInstruction(old_func, FIRST_32_SECOND_64(new_func, trampoline));
 815
 816   // Restore previous memory protection.
 817   if (!RestoreMemoryProtection(old_func, kJumpInstructionLength, protection))
 818     return false;
 819
 820   return true;
 821 }
 822
 823 bool OverrideFunctionWithHotPatch(
 824     uptr old_func, uptr new_func, uptr *orig_old_func) {
 825   const int kHotPatchHeaderLen = kBranchLength;
 826
 827   uptr header = (uptr)old_func - kHotPatchHeaderLen;
 828   uptr patch_length = kHotPatchHeaderLen + kShortJumpInstructionLength;
 829
 830   // Validate that the function is hot patchable.
 831   size_t instruction_size = GetInstructionSize(old_func);
 832   if (instruction_size < kShortJumpInstructionLength ||
 833       !FunctionHasPadding(old_func, kHotPatchHeaderLen))
 834     return false;
 835
 836   if (orig_old_func) {
 837     // Put the needed instructions into the trampoline bytes.
 838     uptr trampoline_length = instruction_size + kDirectBranchLength;
 839     uptr trampoline = AllocateMemoryForTrampoline(old_func, trampoline_length);
 840     if (!trampoline)
 841       return false;
 842     if (!CopyInstructions(trampoline, old_func, instruction_size))
 843       return false;
 844     WriteDirectBranch(trampoline + instruction_size,
 845                       old_func + instruction_size);
 846     *orig_old_func = trampoline;
 847   }
 848
 849   // If needed, get memory space for indirect address.
 850   uptr indirect_address = 0;
 851 #if SANITIZER_WINDOWS64
 852   indirect_address = AllocateMemoryForTrampoline(old_func, kAddressLength);
 853   if (!indirect_address)
 854     return false;
 855 #endif
 856
 857   // Change memory protection to writable.
 858   DWORD protection = 0;
 859   if (!ChangeMemoryProtection(header, patch_length, &protection))
 860     return false;
 861
 862   // Write jumps to the redirected function.
 863   WriteBranch(header, indirect_address, new_func);
 864   WriteShortJumpInstruction(old_func, header);
 865
 866   // Restore previous memory protection.
 867   if (!RestoreMemoryProtection(header, patch_length, protection))
 868     return false;
 869
 870   return true;
 871 }
 872
 873 bool OverrideFunctionWithTrampoline(
 874     uptr old_func, uptr new_func, uptr *orig_old_func) {
 875
 876   size_t instructions_length = kBranchLength;
 877   size_t padding_length = 0;
 878   uptr indirect_address = 0;
 879
 880   if (orig_old_func) {
 881     // Find out the number of bytes of the instructions we need to copy
 882     // to the trampoline.
 883     instructions_length = RoundUpToInstrBoundary(kBranchLength, old_func);
 884     if (!instructions_length)
 885       return false;
 886
 887     // Put the needed instructions into the trampoline bytes.
 888     uptr trampoline_length = instructions_length + kDirectBranchLength;
 889     uptr trampoline = AllocateMemoryForTrampoline(old_func, trampoline_length);
 890     if (!trampoline)
 891       return false;
 892     if (!CopyInstructions(trampoline, old_func, instructions_length))
 893       return false;
 894     WriteDirectBranch(trampoline + instructions_length,
 895                       old_func + instructions_length);
 896     *orig_old_func = trampoline;
 897   }
 898
 899 #if SANITIZER_WINDOWS64
 900   // Check if the targeted address can be encoded in the function padding.
 901   // Otherwise, allocate it in the trampoline region.
 902   if (IsMemoryPadding(old_func - kAddressLength, kAddressLength)) {
 903     indirect_address = old_func - kAddressLength;
 904     padding_length = kAddressLength;
 905   } else {
 906     indirect_address = AllocateMemoryForTrampoline(old_func, kAddressLength);
 907     if (!indirect_address)
 908       return false;
 909   }
 910 #endif
 911
 912   // Change memory protection to writable.
 913   uptr patch_address = old_func - padding_length;
 914   uptr patch_length = instructions_length + padding_length;
 915   DWORD protection = 0;
 916   if (!ChangeMemoryProtection(patch_address, patch_length, &protection))
 917     return false;
 918
 919   // Patch the original function.
 920   WriteBranch(old_func, indirect_address, new_func);
 921
 922   // Restore previous memory protection.
 923   if (!RestoreMemoryProtection(patch_address, patch_length, protection))
 924     return false;
 925
 926   return true;
 927 }
 928
 929 bool OverrideFunction(
 930     uptr old_func, uptr new_func, uptr *orig_old_func) {
 931 #if !SANITIZER_WINDOWS64
 932   if (OverrideFunctionWithDetour(old_func, new_func, orig_old_func))
 933     return true;
 934 #endif
 935   if (OverrideFunctionWithRedirectJump(old_func, new_func, orig_old_func))
 936     return true;
 937   if (OverrideFunctionWithHotPatch(old_func, new_func, orig_old_func))
 938     return true;
 939   if (OverrideFunctionWithTrampoline(old_func, new_func, orig_old_func))
 940     return true;
 941   return false;
 942 }
 943
 944 static void **InterestingDLLsAvailable() {
 945   static const char *InterestingDLLs[] = {
 946       "kernel32.dll",
 947       "msvcr100.dll",      // VS2010
 948       "msvcr110.dll",      // VS2012
 949       "msvcr120.dll",      // VS2013
 950       "vcruntime140.dll",  // VS2015
 951       "ucrtbase.dll",      // Universal CRT
 952 #if (defined(__MINGW32__) && defined(__i386__))
 953       "libc++.dll",        // libc++
 954       "libunwind.dll",     // libunwind
 955 #endif
 956       // NTDLL should go last as it exports some functions that we should
 957       // override in the CRT [presumably only used internally].
 958       "ntdll.dll", NULL};
 959   static void *result[ARRAY_SIZE(InterestingDLLs)] = { 0 };
 960   if (!result[0]) {
 961     for (size_t i = 0, j = 0; InterestingDLLs[i]; ++i) {
 962       if (HMODULE h = GetModuleHandleA(InterestingDLLs[i]))
 963         result[j++] = (void *)h;
 964     }
 965   }
 966   return &result[0];
 967 }
 968
 969 namespace {
 970 // Utility for reading loaded PE images.
 971 template <typename T> class RVAPtr {
 972  public:
 973   RVAPtr(void *module, uptr rva)
 974       : ptr_(reinterpret_cast<T *>(reinterpret_cast<char *>(module) + rva)) {}
 975   operator T *() { return ptr_; }
 976   T *operator->() { return ptr_; }
 977   T *operator++() { return ++ptr_; }
 978
 979  private:
 980   T *ptr_;
 981 };
 982 } // namespace
 983
 984 // Internal implementation of GetProcAddress. At least since Windows 8,
 985 // GetProcAddress appears to initialize DLLs before returning function pointers
 986 // into them. This is problematic for the sanitizers, because they typically
 987 // want to intercept malloc *before* MSVCRT initializes. Our internal
 988 // implementation walks the export list manually without doing initialization.
 989 uptr InternalGetProcAddress(void *module, const char *func_name) {
 990   // Check that the module header is full and present.
 991   RVAPtr<IMAGE_DOS_HEADER> dos_stub(module, 0);
 992   RVAPtr<IMAGE_NT_HEADERS> headers(module, dos_stub->e_lfanew);
 993   if (!module || dos_stub->e_magic != IMAGE_DOS_SIGNATURE ||  // "MZ"
 994       headers->Signature != IMAGE_NT_SIGNATURE ||             // "PE\0\0"
 995       headers->FileHeader.SizeOfOptionalHeader <
 996           sizeof(IMAGE_OPTIONAL_HEADER)) {
 997     return 0;
 998   }
 999
1000   IMAGE_DATA_DIRECTORY *export_directory =
1001       &headers->OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_EXPORT];
1002   if (export_directory->Size == 0)
1003     return 0;
1004   RVAPtr<IMAGE_EXPORT_DIRECTORY> exports(module,
1005                                          export_directory->VirtualAddress);
1006   RVAPtr<DWORD> functions(module, exports->AddressOfFunctions);
1007   RVAPtr<DWORD> names(module, exports->AddressOfNames);
1008   RVAPtr<WORD> ordinals(module, exports->AddressOfNameOrdinals);
1009
1010   for (DWORD i = 0; i < exports->NumberOfNames; i++) {
1011     RVAPtr<char> name(module, names[i]);
1012     if (!strcmp(func_name, name)) {
1013       DWORD index = ordinals[i];
1014       RVAPtr<char> func(module, functions[index]);
1015
1016       // Handle forwarded functions.
1017       DWORD offset = functions[index];
1018       if (offset >= export_directory->VirtualAddress &&
1019           offset < export_directory->VirtualAddress + export_directory->Size) {
1020         // An entry for a forwarded function is a string with the following
1021         // format: "<module> . <function_name>" that is stored into the
1022         // exported directory.
1023         char function_name[256];
1024         size_t funtion_name_length = _strlen(func);
1025         if (funtion_name_length >= sizeof(function_name) - 1)
1026           InterceptionFailed();
1027
1028         _memcpy(function_name, func, funtion_name_length);
1029         function_name[funtion_name_length] = '\0';
1030         char* separator = _strchr(function_name, '.');
1031         if (!separator)
1032           InterceptionFailed();
1033         *separator = '\0';
1034
1035         void* redirected_module = GetModuleHandleA(function_name);
1036         if (!redirected_module)
1037           InterceptionFailed();
1038         return InternalGetProcAddress(redirected_module, separator + 1);
1039       }
1040
1041       return (uptr)(char *)func;
1042     }
1043   }
1044
1045   return 0;
1046 }
1047
1048 bool OverrideFunction(
1049     const char *func_name, uptr new_func, uptr *orig_old_func) {
1050   bool hooked = false;
1051   void **DLLs = InterestingDLLsAvailable();
1052   for (size_t i = 0; DLLs[i]; ++i) {
1053     uptr func_addr = InternalGetProcAddress(DLLs[i], func_name);
1054     if (func_addr &&
1055         OverrideFunction(func_addr, new_func, orig_old_func)) {
1056       hooked = true;
1057     }
1058   }
1059   return hooked;
1060 }
1061
1062 bool OverrideImportedFunction(const char *module_to_patch,
1063                               const char *imported_module,
1064                               const char *function_name, uptr new_function,
1065                               uptr *orig_old_func) {
1066   HMODULE module = GetModuleHandleA(module_to_patch);
1067   if (!module)
1068     return false;
1069
1070   // Check that the module header is full and present.
1071   RVAPtr<IMAGE_DOS_HEADER> dos_stub(module, 0);
1072   RVAPtr<IMAGE_NT_HEADERS> headers(module, dos_stub->e_lfanew);
1073   if (!module || dos_stub->e_magic != IMAGE_DOS_SIGNATURE ||  // "MZ"
1074       headers->Signature != IMAGE_NT_SIGNATURE ||             // "PE\0\0"
1075       headers->FileHeader.SizeOfOptionalHeader <
1076           sizeof(IMAGE_OPTIONAL_HEADER)) {
1077     return false;
1078   }
1079
1080   IMAGE_DATA_DIRECTORY *import_directory =
1081       &headers->OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_IMPORT];
1082
1083   // Iterate the list of imported DLLs. FirstThunk will be null for the last
1084   // entry.
1085   RVAPtr<IMAGE_IMPORT_DESCRIPTOR> imports(module,
1086                                           import_directory->VirtualAddress);
1087   for (; imports->FirstThunk != 0; ++imports) {
1088     RVAPtr<const char> modname(module, imports->Name);
1089     if (_stricmp(&*modname, imported_module) == 0)
1090       break;
1091   }
1092   if (imports->FirstThunk == 0)
1093     return false;
1094
1095   // We have two parallel arrays: the import address table (IAT) and the table
1096   // of names. They start out containing the same data, but the loader rewrites
1097   // the IAT to hold imported addresses and leaves the name table in
1098   // OriginalFirstThunk alone.
1099   RVAPtr<IMAGE_THUNK_DATA> name_table(module, imports->OriginalFirstThunk);
1100   RVAPtr<IMAGE_THUNK_DATA> iat(module, imports->FirstThunk);
1101   for (; name_table->u1.Ordinal != 0; ++name_table, ++iat) {
1102     if (!IMAGE_SNAP_BY_ORDINAL(name_table->u1.Ordinal)) {
1103       RVAPtr<IMAGE_IMPORT_BY_NAME> import_by_name(
1104           module, name_table->u1.ForwarderString);
1105       const char *funcname = &import_by_name->Name[0];
1106       if (strcmp(funcname, function_name) == 0)
1107         break;
1108     }
1109   }
1110   if (name_table->u1.Ordinal == 0)
1111     return false;
1112
1113   // Now we have the correct IAT entry. Do the swap. We have to make the page
1114   // read/write first.
1115   if (orig_old_func)
1116     *orig_old_func = iat->u1.AddressOfData;
1117   DWORD old_prot, unused_prot;
1118   if (!VirtualProtect(&iat->u1.AddressOfData, 4, PAGE_EXECUTE_READWRITE,
1119                       &old_prot))
1120     return false;
1121   iat->u1.AddressOfData = new_function;
1122   if (!VirtualProtect(&iat->u1.AddressOfData, 4, old_prot, &unused_prot))
1123     return false;  // Not clear if this failure bothers us.
1124   return true;
1125 }
1126
1127 }  // namespace __interception
1128
1129 #endif  // SANITIZER_APPLE