Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame^] | 1 | // Copyright 2015 the V8 project authors. All rights reserved. |
| 2 | // Use of this source code is governed by a BSD-style license that can be |
| 3 | // found in the LICENSE file. |
| 4 | |
| 5 | #include "src/macro-assembler.h" |
| 6 | #include "src/objects.h" |
| 7 | #include "src/v8.h" |
| 8 | |
| 9 | #include "src/wasm/decoder.h" |
| 10 | #include "src/wasm/module-decoder.h" |
| 11 | |
| 12 | namespace v8 { |
| 13 | namespace internal { |
| 14 | namespace wasm { |
| 15 | |
| 16 | #if DEBUG |
| 17 | #define TRACE(...) \ |
| 18 | do { \ |
| 19 | if (FLAG_trace_wasm_decoder) PrintF(__VA_ARGS__); \ |
| 20 | } while (false) |
| 21 | #else |
| 22 | #define TRACE(...) |
| 23 | #endif |
| 24 | |
| 25 | |
| 26 | // The main logic for decoding the bytes of a module. |
| 27 | class ModuleDecoder : public Decoder { |
| 28 | public: |
| 29 | ModuleDecoder(Zone* zone, const byte* module_start, const byte* module_end, |
| 30 | bool asm_js) |
| 31 | : Decoder(module_start, module_end), module_zone(zone), asm_js_(asm_js) { |
| 32 | result_.start = start_; |
| 33 | if (limit_ < start_) { |
| 34 | error(start_, "end is less than start"); |
| 35 | limit_ = start_; |
| 36 | } |
| 37 | } |
| 38 | |
| 39 | virtual void onFirstError() { |
| 40 | pc_ = limit_; // On error, terminate section decoding loop. |
| 41 | } |
| 42 | |
| 43 | // Decodes an entire module. |
| 44 | ModuleResult DecodeModule(WasmModule* module, bool verify_functions = true) { |
| 45 | pc_ = start_; |
| 46 | module->module_start = start_; |
| 47 | module->module_end = limit_; |
| 48 | module->min_mem_size_log2 = 0; |
| 49 | module->max_mem_size_log2 = 0; |
| 50 | module->mem_export = false; |
| 51 | module->mem_external = false; |
| 52 | module->globals = new std::vector<WasmGlobal>(); |
| 53 | module->signatures = new std::vector<FunctionSig*>(); |
| 54 | module->functions = new std::vector<WasmFunction>(); |
| 55 | module->data_segments = new std::vector<WasmDataSegment>(); |
| 56 | module->function_table = new std::vector<uint16_t>(); |
| 57 | |
| 58 | bool sections[kMaxModuleSectionCode]; |
| 59 | memset(sections, 0, sizeof(sections)); |
| 60 | |
| 61 | // Decode the module sections. |
| 62 | while (pc_ < limit_) { |
| 63 | TRACE("DecodeSection\n"); |
| 64 | WasmSectionDeclCode section = |
| 65 | static_cast<WasmSectionDeclCode>(u8("section")); |
| 66 | // Each section should appear at most once. |
| 67 | if (section < kMaxModuleSectionCode) { |
| 68 | CheckForPreviousSection(sections, section, false); |
| 69 | sections[section] = true; |
| 70 | } |
| 71 | |
| 72 | switch (section) { |
| 73 | case kDeclEnd: |
| 74 | // Terminate section decoding. |
| 75 | limit_ = pc_; |
| 76 | break; |
| 77 | case kDeclMemory: |
| 78 | module->min_mem_size_log2 = u8("min memory"); |
| 79 | module->max_mem_size_log2 = u8("max memory"); |
| 80 | module->mem_export = u8("export memory") != 0; |
| 81 | break; |
| 82 | case kDeclSignatures: { |
| 83 | int length; |
| 84 | uint32_t signatures_count = u32v(&length, "signatures count"); |
| 85 | module->signatures->reserve(SafeReserve(signatures_count)); |
| 86 | // Decode signatures. |
| 87 | for (uint32_t i = 0; i < signatures_count; i++) { |
| 88 | if (failed()) break; |
| 89 | TRACE("DecodeSignature[%d] module+%d\n", i, |
| 90 | static_cast<int>(pc_ - start_)); |
| 91 | FunctionSig* s = sig(); // read function sig. |
| 92 | module->signatures->push_back(s); |
| 93 | } |
| 94 | break; |
| 95 | } |
| 96 | case kDeclFunctions: { |
| 97 | // Functions require a signature table first. |
| 98 | CheckForPreviousSection(sections, kDeclSignatures, true); |
| 99 | int length; |
| 100 | uint32_t functions_count = u32v(&length, "functions count"); |
| 101 | module->functions->reserve(SafeReserve(functions_count)); |
| 102 | // Set up module environment for verification. |
| 103 | ModuleEnv menv; |
| 104 | menv.module = module; |
| 105 | menv.globals_area = 0; |
| 106 | menv.mem_start = 0; |
| 107 | menv.mem_end = 0; |
| 108 | menv.function_code = nullptr; |
| 109 | menv.asm_js = asm_js_; |
| 110 | // Decode functions. |
| 111 | for (uint32_t i = 0; i < functions_count; i++) { |
| 112 | if (failed()) break; |
| 113 | TRACE("DecodeFunction[%d] module+%d\n", i, |
| 114 | static_cast<int>(pc_ - start_)); |
| 115 | |
| 116 | module->functions->push_back( |
| 117 | {nullptr, 0, 0, 0, 0, 0, 0, false, false}); |
| 118 | WasmFunction* function = &module->functions->back(); |
| 119 | DecodeFunctionInModule(module, function, false); |
| 120 | } |
| 121 | if (ok() && verify_functions) { |
| 122 | for (uint32_t i = 0; i < functions_count; i++) { |
| 123 | if (failed()) break; |
| 124 | WasmFunction* function = &module->functions->at(i); |
| 125 | if (!function->external) { |
| 126 | VerifyFunctionBody(i, &menv, function); |
| 127 | if (result_.failed()) |
| 128 | error(result_.error_pc, result_.error_msg.get()); |
| 129 | } |
| 130 | } |
| 131 | } |
| 132 | break; |
| 133 | } |
| 134 | case kDeclGlobals: { |
| 135 | int length; |
| 136 | uint32_t globals_count = u32v(&length, "globals count"); |
| 137 | module->globals->reserve(SafeReserve(globals_count)); |
| 138 | // Decode globals. |
| 139 | for (uint32_t i = 0; i < globals_count; i++) { |
| 140 | if (failed()) break; |
| 141 | TRACE("DecodeGlobal[%d] module+%d\n", i, |
| 142 | static_cast<int>(pc_ - start_)); |
| 143 | module->globals->push_back({0, MachineType::Int32(), 0, false}); |
| 144 | WasmGlobal* global = &module->globals->back(); |
| 145 | DecodeGlobalInModule(global); |
| 146 | } |
| 147 | break; |
| 148 | } |
| 149 | case kDeclDataSegments: { |
| 150 | int length; |
| 151 | uint32_t data_segments_count = u32v(&length, "data segments count"); |
| 152 | module->data_segments->reserve(SafeReserve(data_segments_count)); |
| 153 | // Decode data segments. |
| 154 | for (uint32_t i = 0; i < data_segments_count; i++) { |
| 155 | if (failed()) break; |
| 156 | TRACE("DecodeDataSegment[%d] module+%d\n", i, |
| 157 | static_cast<int>(pc_ - start_)); |
| 158 | module->data_segments->push_back({0, 0, 0}); |
| 159 | WasmDataSegment* segment = &module->data_segments->back(); |
| 160 | DecodeDataSegmentInModule(segment); |
| 161 | } |
| 162 | break; |
| 163 | } |
| 164 | case kDeclFunctionTable: { |
| 165 | // An indirect function table requires functions first. |
| 166 | CheckForPreviousSection(sections, kDeclFunctions, true); |
| 167 | int length; |
| 168 | uint32_t function_table_count = u32v(&length, "function table count"); |
| 169 | module->function_table->reserve(SafeReserve(function_table_count)); |
| 170 | // Decode function table. |
| 171 | for (uint32_t i = 0; i < function_table_count; i++) { |
| 172 | if (failed()) break; |
| 173 | TRACE("DecodeFunctionTable[%d] module+%d\n", i, |
| 174 | static_cast<int>(pc_ - start_)); |
| 175 | uint16_t index = u16(); |
| 176 | if (index >= module->functions->size()) { |
| 177 | error(pc_ - 2, "invalid function index"); |
| 178 | break; |
| 179 | } |
| 180 | module->function_table->push_back(index); |
| 181 | } |
| 182 | break; |
| 183 | } |
| 184 | case kDeclWLL: { |
| 185 | // Reserved for experimentation by the Web Low-level Language project |
| 186 | // which is augmenting the binary encoding with source code meta |
| 187 | // information. This section does not affect the semantics of the code |
| 188 | // and can be ignored by the runtime. https://github.com/JSStats/wll |
| 189 | int length = 0; |
| 190 | uint32_t section_size = u32v(&length, "section size"); |
| 191 | if (pc_ + section_size > limit_ || pc_ + section_size < pc_) { |
| 192 | error(pc_ - length, "invalid section size"); |
| 193 | break; |
| 194 | } |
| 195 | pc_ += section_size; |
| 196 | break; |
| 197 | } |
| 198 | default: |
| 199 | error(pc_ - 1, nullptr, "unrecognized section 0x%02x", section); |
| 200 | break; |
| 201 | } |
| 202 | } |
| 203 | |
| 204 | return toResult(module); |
| 205 | } |
| 206 | |
| 207 | uint32_t SafeReserve(uint32_t count) { |
| 208 | // Avoid OOM by only reserving up to a certain size. |
| 209 | const uint32_t kMaxReserve = 20000; |
| 210 | return count < kMaxReserve ? count : kMaxReserve; |
| 211 | } |
| 212 | |
| 213 | void CheckForPreviousSection(bool* sections, WasmSectionDeclCode section, |
| 214 | bool present) { |
| 215 | if (section >= kMaxModuleSectionCode) return; |
| 216 | if (sections[section] == present) return; |
| 217 | const char* name = ""; |
| 218 | switch (section) { |
| 219 | case kDeclMemory: |
| 220 | name = "memory"; |
| 221 | break; |
| 222 | case kDeclSignatures: |
| 223 | name = "signatures"; |
| 224 | break; |
| 225 | case kDeclFunctions: |
| 226 | name = "function declaration"; |
| 227 | break; |
| 228 | case kDeclGlobals: |
| 229 | name = "global variable"; |
| 230 | break; |
| 231 | case kDeclDataSegments: |
| 232 | name = "data segment"; |
| 233 | break; |
| 234 | case kDeclFunctionTable: |
| 235 | name = "function table"; |
| 236 | break; |
| 237 | default: |
| 238 | name = ""; |
| 239 | break; |
| 240 | } |
| 241 | if (present) { |
| 242 | error(pc_ - 1, nullptr, "required %s section missing", name); |
| 243 | } else { |
| 244 | error(pc_ - 1, nullptr, "%s section already present", name); |
| 245 | } |
| 246 | } |
| 247 | |
| 248 | // Decodes a single anonymous function starting at {start_}. |
| 249 | FunctionResult DecodeSingleFunction(ModuleEnv* module_env, |
| 250 | WasmFunction* function) { |
| 251 | pc_ = start_; |
| 252 | function->sig = sig(); // read signature |
| 253 | function->name_offset = 0; // ---- name |
| 254 | function->code_start_offset = off(pc_ + 8); // ---- code start |
| 255 | function->code_end_offset = off(limit_); // ---- code end |
| 256 | function->local_int32_count = u16(); // read u16 |
| 257 | function->local_int64_count = u16(); // read u16 |
| 258 | function->local_float32_count = u16(); // read u16 |
| 259 | function->local_float64_count = u16(); // read u16 |
| 260 | function->exported = false; // ---- exported |
| 261 | function->external = false; // ---- external |
| 262 | |
| 263 | if (ok()) VerifyFunctionBody(0, module_env, function); |
| 264 | |
| 265 | FunctionResult result; |
| 266 | result.CopyFrom(result_); // Copy error code and location. |
| 267 | result.val = function; |
| 268 | return result; |
| 269 | } |
| 270 | |
| 271 | // Decodes a single function signature at {start}. |
| 272 | FunctionSig* DecodeFunctionSignature(const byte* start) { |
| 273 | pc_ = start; |
| 274 | FunctionSig* result = sig(); |
| 275 | return ok() ? result : nullptr; |
| 276 | } |
| 277 | |
| 278 | private: |
| 279 | Zone* module_zone; |
| 280 | ModuleResult result_; |
| 281 | bool asm_js_; |
| 282 | |
| 283 | uint32_t off(const byte* ptr) { return static_cast<uint32_t>(ptr - start_); } |
| 284 | |
| 285 | // Decodes a single global entry inside a module starting at {pc_}. |
| 286 | void DecodeGlobalInModule(WasmGlobal* global) { |
| 287 | global->name_offset = string("global name"); |
| 288 | global->type = mem_type(); |
| 289 | global->offset = 0; |
| 290 | global->exported = u8("exported") != 0; |
| 291 | } |
| 292 | |
| 293 | // Decodes a single function entry inside a module starting at {pc_}. |
| 294 | void DecodeFunctionInModule(WasmModule* module, WasmFunction* function, |
| 295 | bool verify_body = true) { |
| 296 | byte decl_bits = u8("function decl"); |
| 297 | |
| 298 | const byte* sigpos = pc_; |
| 299 | function->sig_index = u16("signature index"); |
| 300 | |
| 301 | if (function->sig_index >= module->signatures->size()) { |
| 302 | return error(sigpos, "invalid signature index"); |
| 303 | } else { |
| 304 | function->sig = module->signatures->at(function->sig_index); |
| 305 | } |
| 306 | |
| 307 | TRACE(" +%d <function attributes:%s%s%s%s%s>\n", |
| 308 | static_cast<int>(pc_ - start_), |
| 309 | decl_bits & kDeclFunctionName ? " name" : "", |
| 310 | decl_bits & kDeclFunctionImport ? " imported" : "", |
| 311 | decl_bits & kDeclFunctionLocals ? " locals" : "", |
| 312 | decl_bits & kDeclFunctionExport ? " exported" : "", |
| 313 | (decl_bits & kDeclFunctionImport) == 0 ? " body" : ""); |
| 314 | |
| 315 | if (decl_bits & kDeclFunctionName) { |
| 316 | function->name_offset = string("function name"); |
| 317 | } |
| 318 | |
| 319 | function->exported = decl_bits & kDeclFunctionExport; |
| 320 | |
| 321 | // Imported functions have no locals or body. |
| 322 | if (decl_bits & kDeclFunctionImport) { |
| 323 | function->external = true; |
| 324 | return; |
| 325 | } |
| 326 | |
| 327 | if (decl_bits & kDeclFunctionLocals) { |
| 328 | function->local_int32_count = u16("int32 count"); |
| 329 | function->local_int64_count = u16("int64 count"); |
| 330 | function->local_float32_count = u16("float32 count"); |
| 331 | function->local_float64_count = u16("float64 count"); |
| 332 | } |
| 333 | |
| 334 | uint16_t size = u16("body size"); |
| 335 | if (ok()) { |
| 336 | if ((pc_ + size) > limit_) { |
| 337 | return error(pc_, limit_, |
| 338 | "expected %d bytes for function body, fell off end", size); |
| 339 | } |
| 340 | function->code_start_offset = static_cast<uint32_t>(pc_ - start_); |
| 341 | function->code_end_offset = function->code_start_offset + size; |
| 342 | TRACE(" +%d %-20s: (%d bytes)\n", static_cast<int>(pc_ - start_), |
| 343 | "function body", size); |
| 344 | pc_ += size; |
| 345 | } |
| 346 | } |
| 347 | |
| 348 | // Decodes a single data segment entry inside a module starting at {pc_}. |
| 349 | void DecodeDataSegmentInModule(WasmDataSegment* segment) { |
| 350 | segment->dest_addr = |
| 351 | u32("destination"); // TODO(titzer): check it's within the memory size. |
| 352 | segment->source_offset = offset("source offset"); |
| 353 | segment->source_size = |
| 354 | u32("source size"); // TODO(titzer): check the size is reasonable. |
| 355 | segment->init = u8("init"); |
| 356 | } |
| 357 | |
| 358 | // Verifies the body (code) of a given function. |
| 359 | void VerifyFunctionBody(uint32_t func_num, ModuleEnv* menv, |
| 360 | WasmFunction* function) { |
| 361 | if (FLAG_trace_wasm_decode_time) { |
| 362 | // TODO(titzer): clean me up a bit. |
| 363 | OFStream os(stdout); |
| 364 | os << "Verifying WASM function:"; |
| 365 | if (function->name_offset > 0) { |
| 366 | os << menv->module->GetName(function->name_offset); |
| 367 | } |
| 368 | os << std::endl; |
| 369 | } |
| 370 | FunctionEnv fenv; |
| 371 | fenv.module = menv; |
| 372 | fenv.sig = function->sig; |
| 373 | fenv.local_int32_count = function->local_int32_count; |
| 374 | fenv.local_int64_count = function->local_int64_count; |
| 375 | fenv.local_float32_count = function->local_float32_count; |
| 376 | fenv.local_float64_count = function->local_float64_count; |
| 377 | fenv.SumLocals(); |
| 378 | |
| 379 | TreeResult result = |
| 380 | VerifyWasmCode(&fenv, start_, start_ + function->code_start_offset, |
| 381 | start_ + function->code_end_offset); |
| 382 | if (result.failed()) { |
| 383 | // Wrap the error message from the function decoder. |
| 384 | std::ostringstream str; |
| 385 | str << "in function #" << func_num << ": "; |
| 386 | // TODO(titzer): add function name for the user? |
| 387 | str << result; |
| 388 | std::string strval = str.str(); |
| 389 | const char* raw = strval.c_str(); |
| 390 | size_t len = strlen(raw); |
| 391 | char* buffer = new char[len]; |
| 392 | strncpy(buffer, raw, len); |
| 393 | buffer[len - 1] = 0; |
| 394 | |
| 395 | // Copy error code and location. |
| 396 | result_.CopyFrom(result); |
| 397 | result_.error_msg.Reset(buffer); |
| 398 | } |
| 399 | } |
| 400 | |
| 401 | // Reads a single 32-bit unsigned integer interpreted as an offset, checking |
| 402 | // the offset is within bounds and advances. |
| 403 | uint32_t offset(const char* name = nullptr) { |
| 404 | uint32_t offset = u32(name ? name : "offset"); |
| 405 | if (offset > static_cast<uint32_t>(limit_ - start_)) { |
| 406 | error(pc_ - sizeof(uint32_t), "offset out of bounds of module"); |
| 407 | } |
| 408 | return offset; |
| 409 | } |
| 410 | |
| 411 | // Reads a single 32-bit unsigned integer interpreted as an offset into the |
| 412 | // data and validating the string there and advances. |
| 413 | uint32_t string(const char* name = nullptr) { |
| 414 | return offset(name ? name : "string"); // TODO(titzer): validate string |
| 415 | } |
| 416 | |
| 417 | // Reads a single 8-bit integer, interpreting it as a local type. |
| 418 | LocalType local_type() { |
| 419 | byte val = u8("local type"); |
| 420 | LocalTypeCode t = static_cast<LocalTypeCode>(val); |
| 421 | switch (t) { |
| 422 | case kLocalVoid: |
| 423 | return kAstStmt; |
| 424 | case kLocalI32: |
| 425 | return kAstI32; |
| 426 | case kLocalI64: |
| 427 | return kAstI64; |
| 428 | case kLocalF32: |
| 429 | return kAstF32; |
| 430 | case kLocalF64: |
| 431 | return kAstF64; |
| 432 | default: |
| 433 | error(pc_ - 1, "invalid local type"); |
| 434 | return kAstStmt; |
| 435 | } |
| 436 | } |
| 437 | |
| 438 | // Reads a single 8-bit integer, interpreting it as a memory type. |
| 439 | MachineType mem_type() { |
| 440 | byte val = u8("memory type"); |
| 441 | MemTypeCode t = static_cast<MemTypeCode>(val); |
| 442 | switch (t) { |
| 443 | case kMemI8: |
| 444 | return MachineType::Int8(); |
| 445 | case kMemU8: |
| 446 | return MachineType::Uint8(); |
| 447 | case kMemI16: |
| 448 | return MachineType::Int16(); |
| 449 | case kMemU16: |
| 450 | return MachineType::Uint16(); |
| 451 | case kMemI32: |
| 452 | return MachineType::Int32(); |
| 453 | case kMemU32: |
| 454 | return MachineType::Uint32(); |
| 455 | case kMemI64: |
| 456 | return MachineType::Int64(); |
| 457 | case kMemU64: |
| 458 | return MachineType::Uint64(); |
| 459 | case kMemF32: |
| 460 | return MachineType::Float32(); |
| 461 | case kMemF64: |
| 462 | return MachineType::Float64(); |
| 463 | default: |
| 464 | error(pc_ - 1, "invalid memory type"); |
| 465 | return MachineType::None(); |
| 466 | } |
| 467 | } |
| 468 | |
| 469 | // Parses an inline function signature. |
| 470 | FunctionSig* sig() { |
| 471 | byte count = u8("param count"); |
| 472 | LocalType ret = local_type(); |
| 473 | FunctionSig::Builder builder(module_zone, ret == kAstStmt ? 0 : 1, count); |
| 474 | if (ret != kAstStmt) builder.AddReturn(ret); |
| 475 | |
| 476 | for (int i = 0; i < count; i++) { |
| 477 | LocalType param = local_type(); |
| 478 | if (param == kAstStmt) error(pc_ - 1, "invalid void parameter type"); |
| 479 | builder.AddParam(param); |
| 480 | } |
| 481 | return builder.Build(); |
| 482 | } |
| 483 | }; |
| 484 | |
| 485 | |
| 486 | // Helpers for nice error messages. |
| 487 | class ModuleError : public ModuleResult { |
| 488 | public: |
| 489 | explicit ModuleError(const char* msg) { |
| 490 | error_code = kError; |
| 491 | size_t len = strlen(msg) + 1; |
| 492 | char* result = new char[len]; |
| 493 | strncpy(result, msg, len); |
| 494 | result[len - 1] = 0; |
| 495 | error_msg.Reset(result); |
| 496 | } |
| 497 | }; |
| 498 | |
| 499 | |
| 500 | // Helpers for nice error messages. |
| 501 | class FunctionError : public FunctionResult { |
| 502 | public: |
| 503 | explicit FunctionError(const char* msg) { |
| 504 | error_code = kError; |
| 505 | size_t len = strlen(msg) + 1; |
| 506 | char* result = new char[len]; |
| 507 | strncpy(result, msg, len); |
| 508 | result[len - 1] = 0; |
| 509 | error_msg.Reset(result); |
| 510 | } |
| 511 | }; |
| 512 | |
| 513 | |
| 514 | ModuleResult DecodeWasmModule(Isolate* isolate, Zone* zone, |
| 515 | const byte* module_start, const byte* module_end, |
| 516 | bool verify_functions, bool asm_js) { |
| 517 | size_t size = module_end - module_start; |
| 518 | if (module_start > module_end) return ModuleError("start > end"); |
| 519 | if (size >= kMaxModuleSize) return ModuleError("size > maximum module size"); |
| 520 | WasmModule* module = new WasmModule(); |
| 521 | ModuleDecoder decoder(zone, module_start, module_end, asm_js); |
| 522 | return decoder.DecodeModule(module, verify_functions); |
| 523 | } |
| 524 | |
| 525 | |
| 526 | FunctionSig* DecodeWasmSignatureForTesting(Zone* zone, const byte* start, |
| 527 | const byte* end) { |
| 528 | ModuleDecoder decoder(zone, start, end, false); |
| 529 | return decoder.DecodeFunctionSignature(start); |
| 530 | } |
| 531 | |
| 532 | |
| 533 | FunctionResult DecodeWasmFunction(Isolate* isolate, Zone* zone, |
| 534 | ModuleEnv* module_env, |
| 535 | const byte* function_start, |
| 536 | const byte* function_end) { |
| 537 | size_t size = function_end - function_start; |
| 538 | if (function_start > function_end) return FunctionError("start > end"); |
| 539 | if (size > kMaxFunctionSize) |
| 540 | return FunctionError("size > maximum function size"); |
| 541 | WasmFunction* function = new WasmFunction(); |
| 542 | ModuleDecoder decoder(zone, function_start, function_end, false); |
| 543 | return decoder.DecodeSingleFunction(module_env, function); |
| 544 | } |
| 545 | } // namespace wasm |
| 546 | } // namespace internal |
| 547 | } // namespace v8 |