| 1 | //===-- xray_interface.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 XRay, a dynamic runtime instrumentation system. |
| 10 | // |
| 11 | // Implementation of the API functions. |
| 12 | // |
| 13 | //===----------------------------------------------------------------------===// |
| 14 | |
| 15 | #include "xray_interface_internal.h" |
| 16 | |
| 17 | #include <cinttypes> |
| 18 | #include <cstdio> |
| 19 | #include <errno.h> |
| 20 | #include <limits> |
| 21 | #include <string.h> |
| 22 | #include <sys/mman.h> |
| 23 | |
| 24 | #if SANITIZER_FUCHSIA |
| 25 | #include <zircon/process.h> |
| 26 | #include <zircon/sanitizer.h> |
| 27 | #include <zircon/status.h> |
| 28 | #include <zircon/syscalls.h> |
| 29 | #endif |
| 30 | |
| 31 | #include "sanitizer_common/sanitizer_addrhashmap.h" |
| 32 | #include "sanitizer_common/sanitizer_common.h" |
| 33 | |
| 34 | #include "xray_defs.h" |
| 35 | #include "xray_flags.h" |
| 36 | |
| 37 | extern __sanitizer::SpinMutex XRayInstrMapMutex; |
| 38 | extern __sanitizer::atomic_uint8_t XRayInitialized; |
| 39 | extern __xray::XRaySledMap *XRayInstrMaps; |
| 40 | extern __sanitizer::atomic_uint32_t XRayNumObjects; |
| 41 | |
| 42 | namespace __xray { |
| 43 | |
| 44 | #if defined(__x86_64__) |
| 45 | static const int16_t cSledLength = 12; |
| 46 | #elif defined(__aarch64__) |
| 47 | static const int16_t cSledLength = 32; |
| 48 | #elif defined(__arm__) |
| 49 | static const int16_t cSledLength = 28; |
| 50 | #elif SANITIZER_LOONGARCH64 |
| 51 | static const int16_t cSledLength = 48; |
| 52 | #elif SANITIZER_MIPS32 |
| 53 | static const int16_t cSledLength = 48; |
| 54 | #elif SANITIZER_MIPS64 |
| 55 | static const int16_t cSledLength = 64; |
| 56 | #elif defined(__powerpc64__) |
| 57 | static const int16_t cSledLength = 8; |
| 58 | #elif defined(__hexagon__) |
| 59 | static const int16_t cSledLength = 20; |
| 60 | #elif defined(__riscv) && (__riscv_xlen == 64) |
| 61 | static const int16_t cSledLength = 68; |
| 62 | #elif defined(__riscv) && (__riscv_xlen == 32) |
| 63 | static const int16_t cSledLength = 52; |
| 64 | #elif defined(__s390x__) |
| 65 | static const int16_t cSledLength = 18; |
| 66 | #else |
| 67 | #error "Unsupported CPU Architecture" |
| 68 | #endif /* CPU architecture */ |
| 69 | |
| 70 | // This is the function to call when we encounter the entry or exit sleds. |
| 71 | atomic_uintptr_t XRayPatchedFunction SANITIZER_INTERFACE_ATTRIBUTE{.val_dont_use: 0}; |
| 72 | |
| 73 | // This is the function to call from the arg1-enabled sleds/trampolines. |
| 74 | atomic_uintptr_t XRayArgLogger SANITIZER_INTERFACE_ATTRIBUTE{.val_dont_use: 0}; |
| 75 | |
| 76 | // This is the function to call when we encounter a custom event log call. |
| 77 | atomic_uintptr_t XRayPatchedCustomEvent SANITIZER_INTERFACE_ATTRIBUTE{.val_dont_use: 0}; |
| 78 | |
| 79 | // This is the function to call when we encounter a typed event log call. |
| 80 | atomic_uintptr_t XRayPatchedTypedEvent SANITIZER_INTERFACE_ATTRIBUTE{.val_dont_use: 0}; |
| 81 | |
| 82 | // This is the global status to determine whether we are currently |
| 83 | // patching/unpatching. |
| 84 | atomic_uint8_t XRayPatching{.val_dont_use: 0}; |
| 85 | |
| 86 | struct TypeDescription { |
| 87 | uint32_t type_id; |
| 88 | std::size_t description_string_length; |
| 89 | }; |
| 90 | |
| 91 | using TypeDescriptorMapType = AddrHashMap<TypeDescription, 11>; |
| 92 | // An address map from immutable descriptors to type ids. |
| 93 | TypeDescriptorMapType TypeDescriptorAddressMap{}; |
| 94 | |
| 95 | atomic_uint32_t TypeEventDescriptorCounter{.val_dont_use: 0}; |
| 96 | |
| 97 | // MProtectHelper is an RAII wrapper for calls to mprotect(...) that will |
| 98 | // undo any successful mprotect(...) changes. This is used to make a page |
| 99 | // writeable and executable, and upon destruction if it was successful in |
| 100 | // doing so returns the page into a read-only and executable page. |
| 101 | // |
| 102 | // This is only used specifically for runtime-patching of the XRay |
| 103 | // instrumentation points. This assumes that the executable pages are |
| 104 | // originally read-and-execute only. |
| 105 | class MProtectHelper { |
| 106 | void *PageAlignedAddr; |
| 107 | std::size_t MProtectLen; |
| 108 | bool MustCleanup; |
| 109 | |
| 110 | public: |
| 111 | explicit MProtectHelper(void *PageAlignedAddr, |
| 112 | std::size_t MProtectLen, |
| 113 | std::size_t PageSize) XRAY_NEVER_INSTRUMENT |
| 114 | : PageAlignedAddr(PageAlignedAddr), |
| 115 | MProtectLen(MProtectLen), |
| 116 | MustCleanup(false) { |
| 117 | #if SANITIZER_FUCHSIA |
| 118 | MProtectLen = RoundUpTo(MProtectLen, PageSize); |
| 119 | #endif |
| 120 | } |
| 121 | |
| 122 | int MakeWriteable() XRAY_NEVER_INSTRUMENT { |
| 123 | #if SANITIZER_FUCHSIA |
| 124 | auto R = __sanitizer_change_code_protection( |
| 125 | reinterpret_cast<uintptr_t>(PageAlignedAddr), MProtectLen, true); |
| 126 | if (R != ZX_OK) { |
| 127 | Report("XRay: cannot change code protection: %s\n" , |
| 128 | _zx_status_get_string(R)); |
| 129 | return -1; |
| 130 | } |
| 131 | MustCleanup = true; |
| 132 | return 0; |
| 133 | #else |
| 134 | auto R = mprotect(addr: PageAlignedAddr, len: MProtectLen, |
| 135 | PROT_READ | PROT_WRITE | PROT_EXEC); |
| 136 | if (R != -1) |
| 137 | MustCleanup = true; |
| 138 | return R; |
| 139 | #endif |
| 140 | } |
| 141 | |
| 142 | ~MProtectHelper() XRAY_NEVER_INSTRUMENT { |
| 143 | if (MustCleanup) { |
| 144 | #if SANITIZER_FUCHSIA |
| 145 | auto R = __sanitizer_change_code_protection( |
| 146 | reinterpret_cast<uintptr_t>(PageAlignedAddr), MProtectLen, false); |
| 147 | if (R != ZX_OK) { |
| 148 | Report("XRay: cannot change code protection: %s\n" , |
| 149 | _zx_status_get_string(R)); |
| 150 | } |
| 151 | #else |
| 152 | mprotect(addr: PageAlignedAddr, len: MProtectLen, PROT_READ | PROT_EXEC); |
| 153 | #endif |
| 154 | } |
| 155 | } |
| 156 | }; |
| 157 | |
| 158 | namespace { |
| 159 | |
| 160 | bool isObjectLoaded(int32_t ObjId) { |
| 161 | SpinMutexLock Guard(&XRayInstrMapMutex); |
| 162 | if (ObjId < 0 || static_cast<uint32_t>(ObjId) >= |
| 163 | atomic_load(a: &XRayNumObjects, mo: memory_order_acquire)) { |
| 164 | return false; |
| 165 | } |
| 166 | return XRayInstrMaps[ObjId].Loaded; |
| 167 | } |
| 168 | |
| 169 | bool patchSled(const XRaySledEntry &Sled, bool Enable, int32_t FuncId, |
| 170 | const XRayTrampolines &Trampolines) XRAY_NEVER_INSTRUMENT { |
| 171 | bool Success = false; |
| 172 | switch (Sled.Kind) { |
| 173 | case XRayEntryType::ENTRY: |
| 174 | Success = patchFunctionEntry(Enable, FuncId, Sled, Trampolines, |
| 175 | /*LogArgs=*/false); |
| 176 | break; |
| 177 | case XRayEntryType::EXIT: |
| 178 | Success = patchFunctionExit(Enable, FuncId, Sled, Trampolines); |
| 179 | break; |
| 180 | case XRayEntryType::TAIL: |
| 181 | Success = patchFunctionTailExit(Enable, FuncId, Sled, Trampolines); |
| 182 | break; |
| 183 | case XRayEntryType::LOG_ARGS_ENTRY: |
| 184 | Success = patchFunctionEntry(Enable, FuncId, Sled, Trampolines, |
| 185 | /*LogArgs=*/true); |
| 186 | break; |
| 187 | case XRayEntryType::CUSTOM_EVENT: |
| 188 | Success = patchCustomEvent(Enable, FuncId, Sled); |
| 189 | break; |
| 190 | case XRayEntryType::TYPED_EVENT: |
| 191 | Success = patchTypedEvent(Enable, FuncId, Sled); |
| 192 | break; |
| 193 | default: |
| 194 | Report(format: "Unsupported sled kind '%" PRIu64 "' @%04x\n" , Sled.Address, |
| 195 | int(Sled.Kind)); |
| 196 | return false; |
| 197 | } |
| 198 | return Success; |
| 199 | } |
| 200 | |
| 201 | const XRayFunctionSledIndex |
| 202 | findFunctionSleds(int32_t FuncId, |
| 203 | const XRaySledMap &InstrMap) XRAY_NEVER_INSTRUMENT { |
| 204 | int32_t CurFn = 0; |
| 205 | uint64_t LastFnAddr = 0; |
| 206 | XRayFunctionSledIndex Index = {.Begin: nullptr, .Size: 0}; |
| 207 | |
| 208 | for (std::size_t I = 0; I < InstrMap.Entries && CurFn <= FuncId; I++) { |
| 209 | const auto &Sled = InstrMap.Sleds[I]; |
| 210 | const auto Function = Sled.function(); |
| 211 | if (Function != LastFnAddr) { |
| 212 | CurFn++; |
| 213 | LastFnAddr = Function; |
| 214 | } |
| 215 | |
| 216 | if (CurFn == FuncId) { |
| 217 | if (Index.Begin == nullptr) |
| 218 | Index.Begin = &Sled; |
| 219 | Index.Size = &Sled - Index.Begin + 1; |
| 220 | } |
| 221 | } |
| 222 | |
| 223 | return Index; |
| 224 | } |
| 225 | |
| 226 | XRayPatchingStatus patchFunction(int32_t FuncId, int32_t ObjId, |
| 227 | bool Enable) XRAY_NEVER_INSTRUMENT { |
| 228 | if (!atomic_load(a: &XRayInitialized, mo: memory_order_acquire)) |
| 229 | return XRayPatchingStatus::NOT_INITIALIZED; // Not initialized. |
| 230 | |
| 231 | uint8_t NotPatching = false; |
| 232 | if (!atomic_compare_exchange_strong( |
| 233 | a: &XRayPatching, cmp: &NotPatching, xchg: true, mo: memory_order_acq_rel)) |
| 234 | return XRayPatchingStatus::ONGOING; // Already patching. |
| 235 | |
| 236 | // Next, we look for the function index. |
| 237 | XRaySledMap InstrMap; |
| 238 | { |
| 239 | SpinMutexLock Guard(&XRayInstrMapMutex); |
| 240 | if (ObjId < 0 || static_cast<uint32_t>(ObjId) >= |
| 241 | atomic_load(a: &XRayNumObjects, mo: memory_order_acquire)) { |
| 242 | Report(format: "Unable to patch function: invalid sled map index: %d" , ObjId); |
| 243 | return XRayPatchingStatus::FAILED; |
| 244 | } |
| 245 | InstrMap = XRayInstrMaps[ObjId]; |
| 246 | } |
| 247 | |
| 248 | // If we don't have an index, we can't patch individual functions. |
| 249 | if (InstrMap.Functions == 0) |
| 250 | return XRayPatchingStatus::NOT_INITIALIZED; |
| 251 | |
| 252 | // Check if the corresponding DSO has been unloaded. |
| 253 | if (!InstrMap.Loaded) { |
| 254 | Report(format: "Invalid function id provided: %d\n" , FuncId); |
| 255 | return XRayPatchingStatus::NOT_INITIALIZED; |
| 256 | } |
| 257 | |
| 258 | // FuncId must be a positive number, less than the number of functions |
| 259 | // instrumented. |
| 260 | if (FuncId <= 0 || static_cast<size_t>(FuncId) > InstrMap.Functions) { |
| 261 | Report(format: "Invalid function id provided: %d\n" , FuncId); |
| 262 | return XRayPatchingStatus::FAILED; |
| 263 | } |
| 264 | |
| 265 | auto PackedId = __xray::MakePackedId(FnId: FuncId, ObjId); |
| 266 | |
| 267 | // Now we patch ths sleds for this specific function. |
| 268 | XRayFunctionSledIndex SledRange; |
| 269 | if (InstrMap.SledsIndex) { |
| 270 | SledRange = {.Begin: InstrMap.SledsIndex[FuncId - 1].fromPCRelative(), |
| 271 | .Size: InstrMap.SledsIndex[FuncId - 1].Size}; |
| 272 | } else { |
| 273 | SledRange = findFunctionSleds(FuncId, InstrMap); |
| 274 | } |
| 275 | |
| 276 | auto *f = SledRange.Begin; |
| 277 | bool SucceedOnce = false; |
| 278 | for (size_t i = 0; i != SledRange.Size; ++i) |
| 279 | SucceedOnce |= patchSled(Sled: f[i], Enable, FuncId: PackedId, Trampolines: InstrMap.Trampolines); |
| 280 | |
| 281 | atomic_store(a: &XRayPatching, v: false, mo: memory_order_release); |
| 282 | |
| 283 | if (!SucceedOnce) { |
| 284 | Report(format: "Failed patching any sled for function '%d'." , FuncId); |
| 285 | return XRayPatchingStatus::FAILED; |
| 286 | } |
| 287 | |
| 288 | return XRayPatchingStatus::SUCCESS; |
| 289 | } |
| 290 | |
| 291 | // controlPatching implements the common internals of the patching/unpatching |
| 292 | // implementation. |Enable| defines whether we're enabling or disabling the |
| 293 | // runtime XRay instrumentation. |
| 294 | // This function should only be called after ensuring that XRay is initialized |
| 295 | // and no other thread is currently patching. |
| 296 | XRayPatchingStatus controlPatchingObjectUnchecked(bool Enable, int32_t ObjId) { |
| 297 | XRaySledMap InstrMap; |
| 298 | { |
| 299 | SpinMutexLock Guard(&XRayInstrMapMutex); |
| 300 | if (ObjId < 0 || static_cast<uint32_t>(ObjId) >= |
| 301 | atomic_load(a: &XRayNumObjects, mo: memory_order_acquire)) { |
| 302 | Report(format: "Unable to patch functions: invalid sled map index: %d\n" , ObjId); |
| 303 | return XRayPatchingStatus::FAILED; |
| 304 | } |
| 305 | InstrMap = XRayInstrMaps[ObjId]; |
| 306 | } |
| 307 | if (InstrMap.Entries == 0) |
| 308 | return XRayPatchingStatus::NOT_INITIALIZED; |
| 309 | |
| 310 | if (Verbosity()) |
| 311 | Report(format: "Patching object %d with %d functions.\n" , ObjId, InstrMap.Entries); |
| 312 | |
| 313 | // Check if the corresponding DSO has been unloaded. |
| 314 | if (!InstrMap.Loaded) { |
| 315 | Report(format: "Object is not loaded at index: %d\n" , ObjId); |
| 316 | return XRayPatchingStatus::FAILED; |
| 317 | } |
| 318 | |
| 319 | uint32_t FuncId = 1; |
| 320 | uint64_t CurFun = 0; |
| 321 | |
| 322 | // First we want to find the bounds for which we have instrumentation points, |
| 323 | // and try to get as few calls to mprotect(...) as possible. We're assuming |
| 324 | // that all the sleds for the instrumentation map are contiguous as a single |
| 325 | // set of pages. When we do support dynamic shared object instrumentation, |
| 326 | // we'll need to do this for each set of page load offsets per DSO loaded. For |
| 327 | // now we're assuming we can mprotect the whole section of text between the |
| 328 | // minimum sled address and the maximum sled address (+ the largest sled |
| 329 | // size). |
| 330 | auto *MinSled = &InstrMap.Sleds[0]; |
| 331 | auto *MaxSled = &InstrMap.Sleds[InstrMap.Entries - 1]; |
| 332 | for (std::size_t I = 0; I < InstrMap.Entries; I++) { |
| 333 | const auto &Sled = InstrMap.Sleds[I]; |
| 334 | if (Sled.address() < MinSled->address()) |
| 335 | MinSled = &Sled; |
| 336 | if (Sled.address() > MaxSled->address()) |
| 337 | MaxSled = &Sled; |
| 338 | } |
| 339 | |
| 340 | const size_t PageSize = flags()->xray_page_size_override > 0 |
| 341 | ? flags()->xray_page_size_override |
| 342 | : GetPageSizeCached(); |
| 343 | if ((PageSize == 0) || ((PageSize & (PageSize - 1)) != 0)) { |
| 344 | Report(format: "System page size is not a power of two: %zu\n" , PageSize); |
| 345 | return XRayPatchingStatus::FAILED; |
| 346 | } |
| 347 | |
| 348 | void *PageAlignedAddr = |
| 349 | reinterpret_cast<void *>(MinSled->address() & ~(PageSize - 1)); |
| 350 | size_t MProtectLen = |
| 351 | (MaxSled->address() - reinterpret_cast<uptr>(PageAlignedAddr)) + |
| 352 | cSledLength; |
| 353 | MProtectHelper Protector(PageAlignedAddr, MProtectLen, PageSize); |
| 354 | if (Protector.MakeWriteable() == -1) { |
| 355 | Report(format: "Failed mprotect: %d\n" , errno); |
| 356 | return XRayPatchingStatus::FAILED; |
| 357 | } |
| 358 | |
| 359 | for (std::size_t I = 0; I < InstrMap.Entries; ++I) { |
| 360 | auto &Sled = InstrMap.Sleds[I]; |
| 361 | auto F = Sled.function(); |
| 362 | if (CurFun == 0) |
| 363 | CurFun = F; |
| 364 | if (F != CurFun) { |
| 365 | ++FuncId; |
| 366 | CurFun = F; |
| 367 | } |
| 368 | auto PackedId = __xray::MakePackedId(FnId: FuncId, ObjId); |
| 369 | patchSled(Sled, Enable, FuncId: PackedId, Trampolines: InstrMap.Trampolines); |
| 370 | } |
| 371 | atomic_store(a: &XRayPatching, v: false, mo: memory_order_release); |
| 372 | return XRayPatchingStatus::SUCCESS; |
| 373 | } |
| 374 | |
| 375 | // Controls patching for all registered objects. |
| 376 | // Returns: SUCCESS, if patching succeeds for all objects. |
| 377 | // NOT_INITIALIZED, if one or more objects returned NOT_INITIALIZED |
| 378 | // but none failed. |
| 379 | // FAILED, if patching of one or more objects failed. |
| 380 | XRayPatchingStatus controlPatching(bool Enable) XRAY_NEVER_INSTRUMENT { |
| 381 | if (!atomic_load(a: &XRayInitialized, mo: memory_order_acquire)) |
| 382 | return XRayPatchingStatus::NOT_INITIALIZED; // Not initialized. |
| 383 | |
| 384 | uint8_t NotPatching = false; |
| 385 | if (!atomic_compare_exchange_strong(a: &XRayPatching, cmp: &NotPatching, xchg: true, |
| 386 | mo: memory_order_acq_rel)) |
| 387 | return XRayPatchingStatus::ONGOING; // Already patching. |
| 388 | |
| 389 | auto XRayPatchingStatusResetter = at_scope_exit( |
| 390 | fn: [] { atomic_store(a: &XRayPatching, v: false, mo: memory_order_release); }); |
| 391 | |
| 392 | unsigned NumObjects = __xray_num_objects(); |
| 393 | |
| 394 | XRayPatchingStatus CombinedStatus{NOT_INITIALIZED}; |
| 395 | for (unsigned I = 0; I < NumObjects; ++I) { |
| 396 | if (!isObjectLoaded(ObjId: I)) |
| 397 | continue; |
| 398 | auto LastStatus = controlPatchingObjectUnchecked(Enable, ObjId: I); |
| 399 | switch (LastStatus) { |
| 400 | case SUCCESS: |
| 401 | if (CombinedStatus == NOT_INITIALIZED) |
| 402 | CombinedStatus = SUCCESS; |
| 403 | break; |
| 404 | case FAILED: |
| 405 | // Report failure, but try to patch the remaining objects |
| 406 | CombinedStatus = FAILED; |
| 407 | break; |
| 408 | case NOT_INITIALIZED: |
| 409 | // XRay has been initialized but there are no sleds available for this |
| 410 | // object. Try to patch remaining objects. |
| 411 | if (CombinedStatus != FAILED) |
| 412 | CombinedStatus = NOT_INITIALIZED; |
| 413 | break; |
| 414 | case ONGOING: |
| 415 | UNREACHABLE("Status ONGOING should not appear at this point" ); |
| 416 | } |
| 417 | } |
| 418 | return CombinedStatus; |
| 419 | } |
| 420 | |
| 421 | // Controls patching for one object. |
| 422 | XRayPatchingStatus controlPatching(bool Enable, |
| 423 | int32_t ObjId) XRAY_NEVER_INSTRUMENT { |
| 424 | |
| 425 | if (!atomic_load(a: &XRayInitialized, mo: memory_order_acquire)) |
| 426 | return XRayPatchingStatus::NOT_INITIALIZED; // Not initialized. |
| 427 | |
| 428 | uint8_t NotPatching = false; |
| 429 | if (!atomic_compare_exchange_strong(a: &XRayPatching, cmp: &NotPatching, xchg: true, |
| 430 | mo: memory_order_acq_rel)) |
| 431 | return XRayPatchingStatus::ONGOING; // Already patching. |
| 432 | |
| 433 | auto XRayPatchingStatusResetter = at_scope_exit( |
| 434 | fn: [] { atomic_store(a: &XRayPatching, v: false, mo: memory_order_release); }); |
| 435 | |
| 436 | return controlPatchingObjectUnchecked(Enable, ObjId); |
| 437 | } |
| 438 | |
| 439 | XRayPatchingStatus mprotectAndPatchFunction(int32_t FuncId, int32_t ObjId, |
| 440 | bool Enable) XRAY_NEVER_INSTRUMENT { |
| 441 | XRaySledMap InstrMap; |
| 442 | { |
| 443 | SpinMutexLock Guard(&XRayInstrMapMutex); |
| 444 | if (ObjId < 0 || static_cast<uint32_t>(ObjId) >= |
| 445 | atomic_load(a: &XRayNumObjects, mo: memory_order_acquire)) { |
| 446 | Report(format: "Unable to patch function: invalid sled map index: %d\n" , ObjId); |
| 447 | return XRayPatchingStatus::FAILED; |
| 448 | } |
| 449 | InstrMap = XRayInstrMaps[ObjId]; |
| 450 | } |
| 451 | |
| 452 | // Check if the corresponding DSO has been unloaded. |
| 453 | if (!InstrMap.Loaded) { |
| 454 | Report(format: "Object is not loaded at index: %d\n" , ObjId); |
| 455 | return XRayPatchingStatus::FAILED; |
| 456 | } |
| 457 | |
| 458 | // FuncId must be a positive number, less than the number of functions |
| 459 | // instrumented. |
| 460 | if (FuncId <= 0 || static_cast<size_t>(FuncId) > InstrMap.Functions) { |
| 461 | Report(format: "Invalid function id provided: %d\n" , FuncId); |
| 462 | return XRayPatchingStatus::FAILED; |
| 463 | } |
| 464 | |
| 465 | const size_t PageSize = flags()->xray_page_size_override > 0 |
| 466 | ? flags()->xray_page_size_override |
| 467 | : GetPageSizeCached(); |
| 468 | if ((PageSize == 0) || ((PageSize & (PageSize - 1)) != 0)) { |
| 469 | Report(format: "Provided page size is not a power of two: %zu\n" , PageSize); |
| 470 | return XRayPatchingStatus::FAILED; |
| 471 | } |
| 472 | |
| 473 | // Here we compute the minimum sled and maximum sled associated with a |
| 474 | // particular function ID. |
| 475 | XRayFunctionSledIndex SledRange; |
| 476 | if (InstrMap.SledsIndex) { |
| 477 | SledRange = {.Begin: InstrMap.SledsIndex[FuncId - 1].fromPCRelative(), |
| 478 | .Size: InstrMap.SledsIndex[FuncId - 1].Size}; |
| 479 | } else { |
| 480 | SledRange = findFunctionSleds(FuncId, InstrMap); |
| 481 | } |
| 482 | auto *f = SledRange.Begin; |
| 483 | auto *e = SledRange.Begin + SledRange.Size; |
| 484 | auto *MinSled = f; |
| 485 | auto *MaxSled = e - 1; |
| 486 | while (f != e) { |
| 487 | if (f->address() < MinSled->address()) |
| 488 | MinSled = f; |
| 489 | if (f->address() > MaxSled->address()) |
| 490 | MaxSled = f; |
| 491 | ++f; |
| 492 | } |
| 493 | |
| 494 | void *PageAlignedAddr = |
| 495 | reinterpret_cast<void *>(MinSled->address() & ~(PageSize - 1)); |
| 496 | size_t MProtectLen = |
| 497 | (MaxSled->address() - reinterpret_cast<uptr>(PageAlignedAddr)) + |
| 498 | cSledLength; |
| 499 | MProtectHelper Protector(PageAlignedAddr, MProtectLen, PageSize); |
| 500 | if (Protector.MakeWriteable() == -1) { |
| 501 | Report(format: "Failed mprotect: %d\n" , errno); |
| 502 | return XRayPatchingStatus::FAILED; |
| 503 | } |
| 504 | return patchFunction(FuncId, ObjId, Enable); |
| 505 | } |
| 506 | |
| 507 | } // namespace |
| 508 | |
| 509 | } // namespace __xray |
| 510 | |
| 511 | using namespace __xray; |
| 512 | |
| 513 | // The following functions are declared `extern "C" {...}` in the header, hence |
| 514 | // they're defined in the global namespace. |
| 515 | |
| 516 | int __xray_set_handler(void (*entry)(int32_t, |
| 517 | XRayEntryType)) XRAY_NEVER_INSTRUMENT { |
| 518 | if (atomic_load(a: &XRayInitialized, mo: memory_order_acquire)) { |
| 519 | |
| 520 | atomic_store(a: &__xray::XRayPatchedFunction, |
| 521 | v: reinterpret_cast<uintptr_t>(entry), mo: memory_order_release); |
| 522 | return 1; |
| 523 | } |
| 524 | return 0; |
| 525 | } |
| 526 | |
| 527 | int __xray_set_customevent_handler(void (*entry)(void *, size_t)) |
| 528 | XRAY_NEVER_INSTRUMENT { |
| 529 | if (atomic_load(a: &XRayInitialized, mo: memory_order_acquire)) { |
| 530 | atomic_store(a: &__xray::XRayPatchedCustomEvent, |
| 531 | v: reinterpret_cast<uintptr_t>(entry), mo: memory_order_release); |
| 532 | return 1; |
| 533 | } |
| 534 | return 0; |
| 535 | } |
| 536 | |
| 537 | int __xray_set_typedevent_handler(void (*entry)(size_t, const void *, |
| 538 | size_t)) XRAY_NEVER_INSTRUMENT { |
| 539 | if (atomic_load(a: &XRayInitialized, mo: memory_order_acquire)) { |
| 540 | atomic_store(a: &__xray::XRayPatchedTypedEvent, |
| 541 | v: reinterpret_cast<uintptr_t>(entry), mo: memory_order_release); |
| 542 | return 1; |
| 543 | } |
| 544 | return 0; |
| 545 | } |
| 546 | |
| 547 | int __xray_remove_handler() XRAY_NEVER_INSTRUMENT { |
| 548 | return __xray_set_handler(entry: nullptr); |
| 549 | } |
| 550 | |
| 551 | int __xray_remove_customevent_handler() XRAY_NEVER_INSTRUMENT { |
| 552 | return __xray_set_customevent_handler(entry: nullptr); |
| 553 | } |
| 554 | |
| 555 | int __xray_remove_typedevent_handler() XRAY_NEVER_INSTRUMENT { |
| 556 | return __xray_set_typedevent_handler(entry: nullptr); |
| 557 | } |
| 558 | |
| 559 | uint16_t __xray_register_event_type( |
| 560 | const char *const event_type) XRAY_NEVER_INSTRUMENT { |
| 561 | TypeDescriptorMapType::Handle h(&TypeDescriptorAddressMap, (uptr)event_type); |
| 562 | if (h.created()) { |
| 563 | h->type_id = atomic_fetch_add( |
| 564 | a: &TypeEventDescriptorCounter, v: 1, mo: memory_order_acq_rel); |
| 565 | h->description_string_length = strnlen(string: event_type, maxlen: 1024); |
| 566 | } |
| 567 | return h->type_id; |
| 568 | } |
| 569 | |
| 570 | XRayPatchingStatus __xray_patch() XRAY_NEVER_INSTRUMENT { |
| 571 | return controlPatching(Enable: true); |
| 572 | } |
| 573 | |
| 574 | XRayPatchingStatus __xray_patch_object(int32_t ObjId) XRAY_NEVER_INSTRUMENT { |
| 575 | return controlPatching(Enable: true, ObjId); |
| 576 | } |
| 577 | |
| 578 | XRayPatchingStatus __xray_unpatch() XRAY_NEVER_INSTRUMENT { |
| 579 | return controlPatching(Enable: false); |
| 580 | } |
| 581 | |
| 582 | XRayPatchingStatus __xray_unpatch_object(int32_t ObjId) XRAY_NEVER_INSTRUMENT { |
| 583 | return controlPatching(Enable: false, ObjId); |
| 584 | } |
| 585 | |
| 586 | XRayPatchingStatus __xray_patch_function(int32_t FuncId) XRAY_NEVER_INSTRUMENT { |
| 587 | auto Ids = __xray::UnpackId(PackedId: FuncId); |
| 588 | auto ObjId = Ids.first; |
| 589 | auto FnId = Ids.second; |
| 590 | return mprotectAndPatchFunction(FuncId: FnId, ObjId, Enable: true); |
| 591 | } |
| 592 | |
| 593 | XRayPatchingStatus |
| 594 | __xray_patch_function_in_object(int32_t FuncId, |
| 595 | int32_t ObjId) XRAY_NEVER_INSTRUMENT { |
| 596 | return mprotectAndPatchFunction(FuncId, ObjId, Enable: true); |
| 597 | } |
| 598 | |
| 599 | XRayPatchingStatus |
| 600 | __xray_unpatch_function(int32_t FuncId) XRAY_NEVER_INSTRUMENT { |
| 601 | auto Ids = __xray::UnpackId(PackedId: FuncId); |
| 602 | auto ObjId = Ids.first; |
| 603 | auto FnId = Ids.second; |
| 604 | return mprotectAndPatchFunction(FuncId: FnId, ObjId, Enable: false); |
| 605 | } |
| 606 | |
| 607 | XRayPatchingStatus |
| 608 | __xray_unpatch_function_in_object(int32_t FuncId, |
| 609 | int32_t ObjId) XRAY_NEVER_INSTRUMENT { |
| 610 | return mprotectAndPatchFunction(FuncId, ObjId, Enable: false); |
| 611 | } |
| 612 | |
| 613 | int __xray_set_handler_arg1(void (*entry)(int32_t, XRayEntryType, uint64_t)) { |
| 614 | if (!atomic_load(a: &XRayInitialized, mo: memory_order_acquire)) |
| 615 | return 0; |
| 616 | |
| 617 | // A relaxed write might not be visible even if the current thread gets |
| 618 | // scheduled on a different CPU/NUMA node. We need to wait for everyone to |
| 619 | // have this handler installed for consistency of collected data across CPUs. |
| 620 | atomic_store(a: &XRayArgLogger, v: reinterpret_cast<uint64_t>(entry), |
| 621 | mo: memory_order_release); |
| 622 | return 1; |
| 623 | } |
| 624 | |
| 625 | int __xray_remove_handler_arg1() { return __xray_set_handler_arg1(entry: nullptr); } |
| 626 | |
| 627 | uintptr_t |
| 628 | __xray_function_address(int32_t CombinedFuncId) XRAY_NEVER_INSTRUMENT { |
| 629 | auto Ids = __xray::UnpackId(PackedId: CombinedFuncId); |
| 630 | return __xray_function_address_in_object(FuncId: Ids.second, ObjId: Ids.first); |
| 631 | } |
| 632 | |
| 633 | uintptr_t __xray_function_address_in_object(int32_t FuncId, int32_t ObjId) |
| 634 | XRAY_NEVER_INSTRUMENT { |
| 635 | XRaySledMap InstrMap; |
| 636 | { |
| 637 | SpinMutexLock Guard(&XRayInstrMapMutex); |
| 638 | auto count = atomic_load(a: &XRayNumObjects, mo: memory_order_acquire); |
| 639 | if (ObjId < 0 || static_cast<uint32_t>(ObjId) >= count) { |
| 640 | Report(format: "Unable to determine function address: invalid sled map index %d " |
| 641 | "(size is %d)\n" , |
| 642 | ObjId, (int)count); |
| 643 | return 0; |
| 644 | } |
| 645 | InstrMap = XRayInstrMaps[ObjId]; |
| 646 | } |
| 647 | |
| 648 | if (FuncId <= 0 || static_cast<size_t>(FuncId) > InstrMap.Functions) |
| 649 | return 0; |
| 650 | const XRaySledEntry *Sled = |
| 651 | InstrMap.SledsIndex ? InstrMap.SledsIndex[FuncId - 1].fromPCRelative() |
| 652 | : findFunctionSleds(FuncId, InstrMap).Begin; |
| 653 | return Sled->function() |
| 654 | // On PPC, function entries are always aligned to 16 bytes. The beginning of a |
| 655 | // sled might be a local entry, which is always +8 based on the global entry. |
| 656 | // Always return the global entry. |
| 657 | #ifdef __PPC__ |
| 658 | & ~0xf |
| 659 | #endif |
| 660 | ; |
| 661 | } |
| 662 | |
| 663 | size_t __xray_max_function_id() XRAY_NEVER_INSTRUMENT { |
| 664 | return __xray_max_function_id_in_object(ObjId: 0); |
| 665 | } |
| 666 | |
| 667 | size_t __xray_max_function_id_in_object(int32_t ObjId) XRAY_NEVER_INSTRUMENT { |
| 668 | SpinMutexLock Guard(&XRayInstrMapMutex); |
| 669 | if (ObjId < 0 || static_cast<uint32_t>(ObjId) >= |
| 670 | atomic_load(a: &XRayNumObjects, mo: memory_order_acquire)) |
| 671 | return 0; |
| 672 | return XRayInstrMaps[ObjId].Functions; |
| 673 | } |
| 674 | |
| 675 | size_t __xray_num_objects() XRAY_NEVER_INSTRUMENT { |
| 676 | SpinMutexLock Guard(&XRayInstrMapMutex); |
| 677 | return atomic_load(a: &XRayNumObjects, mo: memory_order_acquire); |
| 678 | } |
| 679 | |
| 680 | int32_t __xray_unpack_function_id(int32_t PackedId) { |
| 681 | return __xray::UnpackId(PackedId).second; |
| 682 | } |
| 683 | |
| 684 | int32_t __xray_unpack_object_id(int32_t PackedId) { |
| 685 | return __xray::UnpackId(PackedId).first; |
| 686 | } |
| 687 | |
| 688 | int32_t __xray_pack_id(int32_t FuncId, int32_t ObjId) { |
| 689 | return __xray::MakePackedId(FnId: FuncId, ObjId); |
| 690 | } |
| 691 | |