• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/wasm/module-decoder.h"
6 
7 #include "src/base/functional.h"
8 #include "src/base/platform/platform.h"
9 #include "src/base/platform/wrappers.h"
10 #include "src/flags/flags.h"
11 #include "src/init/v8.h"
12 #include "src/logging/counters.h"
13 #include "src/logging/metrics.h"
14 #include "src/objects/objects-inl.h"
15 #include "src/utils/ostreams.h"
16 #include "src/wasm/canonical-types.h"
17 #include "src/wasm/decoder.h"
18 #include "src/wasm/function-body-decoder-impl.h"
19 #include "src/wasm/init-expr-interface.h"
20 #include "src/wasm/struct-types.h"
21 #include "src/wasm/wasm-constants.h"
22 #include "src/wasm/wasm-engine.h"
23 #include "src/wasm/wasm-limits.h"
24 #include "src/wasm/wasm-opcodes-inl.h"
25 
26 namespace v8 {
27 namespace internal {
28 namespace wasm {
29 
30 #define TRACE(...)                                    \
31   do {                                                \
32     if (FLAG_trace_wasm_decoder) PrintF(__VA_ARGS__); \
33   } while (false)
34 
35 namespace {
36 
37 constexpr char kNameString[] = "name";
38 constexpr char kSourceMappingURLString[] = "sourceMappingURL";
39 constexpr char kCompilationHintsString[] = "compilationHints";
40 constexpr char kBranchHintsString[] = "metadata.code.branch_hint";
41 constexpr char kDebugInfoString[] = ".debug_info";
42 constexpr char kExternalDebugInfoString[] = "external_debug_info";
43 
ExternalKindName(ImportExportKindCode kind)44 const char* ExternalKindName(ImportExportKindCode kind) {
45   switch (kind) {
46     case kExternalFunction:
47       return "function";
48     case kExternalTable:
49       return "table";
50     case kExternalMemory:
51       return "memory";
52     case kExternalGlobal:
53       return "global";
54     case kExternalTag:
55       return "tag";
56   }
57   return "unknown";
58 }
59 
60 }  // namespace
61 
SectionName(SectionCode code)62 const char* SectionName(SectionCode code) {
63   switch (code) {
64     case kUnknownSectionCode:
65       return "Unknown";
66     case kTypeSectionCode:
67       return "Type";
68     case kImportSectionCode:
69       return "Import";
70     case kFunctionSectionCode:
71       return "Function";
72     case kTableSectionCode:
73       return "Table";
74     case kMemorySectionCode:
75       return "Memory";
76     case kGlobalSectionCode:
77       return "Global";
78     case kExportSectionCode:
79       return "Export";
80     case kStartSectionCode:
81       return "Start";
82     case kCodeSectionCode:
83       return "Code";
84     case kElementSectionCode:
85       return "Element";
86     case kDataSectionCode:
87       return "Data";
88     case kTagSectionCode:
89       return "Tag";
90     case kDataCountSectionCode:
91       return "DataCount";
92     case kNameSectionCode:
93       return kNameString;
94     case kSourceMappingURLSectionCode:
95       return kSourceMappingURLString;
96     case kDebugInfoSectionCode:
97       return kDebugInfoString;
98     case kExternalDebugInfoSectionCode:
99       return kExternalDebugInfoString;
100     case kCompilationHintsSectionCode:
101       return kCompilationHintsString;
102     case kBranchHintsSectionCode:
103       return kBranchHintsString;
104     default:
105       return "<unknown>";
106   }
107 }
108 
109 namespace {
110 
validate_utf8(Decoder * decoder,WireBytesRef string)111 bool validate_utf8(Decoder* decoder, WireBytesRef string) {
112   return unibrow::Utf8::ValidateEncoding(
113       decoder->start() + decoder->GetBufferRelativeOffset(string.offset()),
114       string.length());
115 }
116 
117 // Reads a length-prefixed string, checking that it is within bounds. Returns
118 // the offset of the string, and the length as an out parameter.
consume_string(Decoder * decoder,bool validate_utf8,const char * name)119 WireBytesRef consume_string(Decoder* decoder, bool validate_utf8,
120                             const char* name) {
121   uint32_t length = decoder->consume_u32v("string length");
122   uint32_t offset = decoder->pc_offset();
123   const byte* string_start = decoder->pc();
124   // Consume bytes before validation to guarantee that the string is not oob.
125   if (length > 0) {
126     decoder->consume_bytes(length, name);
127     if (decoder->ok() && validate_utf8 &&
128         !unibrow::Utf8::ValidateEncoding(string_start, length)) {
129       decoder->errorf(string_start, "%s: no valid UTF-8 string", name);
130     }
131   }
132   return {offset, decoder->failed() ? 0 : length};
133 }
134 
135 namespace {
IdentifyUnknownSectionInternal(Decoder * decoder)136 SectionCode IdentifyUnknownSectionInternal(Decoder* decoder) {
137   WireBytesRef string = consume_string(decoder, true, "section name");
138   if (decoder->failed()) {
139     return kUnknownSectionCode;
140   }
141   const byte* section_name_start =
142       decoder->start() + decoder->GetBufferRelativeOffset(string.offset());
143 
144   TRACE("  +%d  section name        : \"%.*s\"\n",
145         static_cast<int>(section_name_start - decoder->start()),
146         string.length() < 20 ? string.length() : 20, section_name_start);
147 
148   using SpecialSectionPair = std::pair<base::Vector<const char>, SectionCode>;
149   static constexpr SpecialSectionPair kSpecialSections[]{
150       {base::StaticCharVector(kNameString), kNameSectionCode},
151       {base::StaticCharVector(kSourceMappingURLString),
152        kSourceMappingURLSectionCode},
153       {base::StaticCharVector(kCompilationHintsString),
154        kCompilationHintsSectionCode},
155       {base::StaticCharVector(kBranchHintsString), kBranchHintsSectionCode},
156       {base::StaticCharVector(kDebugInfoString), kDebugInfoSectionCode},
157       {base::StaticCharVector(kExternalDebugInfoString),
158        kExternalDebugInfoSectionCode}};
159 
160   auto name_vec = base::Vector<const char>::cast(
161       base::VectorOf(section_name_start, string.length()));
162   for (auto& special_section : kSpecialSections) {
163     if (name_vec == special_section.first) return special_section.second;
164   }
165 
166   return kUnknownSectionCode;
167 }
168 }  // namespace
169 
170 // An iterator over the sections in a wasm binary module.
171 // Automatically skips all unknown sections.
172 class WasmSectionIterator {
173  public:
WasmSectionIterator(Decoder * decoder)174   explicit WasmSectionIterator(Decoder* decoder)
175       : decoder_(decoder),
176         section_code_(kUnknownSectionCode),
177         section_start_(decoder->pc()),
178         section_end_(decoder->pc()) {
179     next();
180   }
181 
more() const182   bool more() const { return decoder_->ok() && decoder_->more(); }
183 
section_code() const184   SectionCode section_code() const { return section_code_; }
185 
section_start() const186   const byte* section_start() const { return section_start_; }
187 
section_length() const188   uint32_t section_length() const {
189     return static_cast<uint32_t>(section_end_ - section_start_);
190   }
191 
payload() const192   base::Vector<const uint8_t> payload() const {
193     return {payload_start_, payload_length()};
194   }
195 
payload_start() const196   const byte* payload_start() const { return payload_start_; }
197 
payload_length() const198   uint32_t payload_length() const {
199     return static_cast<uint32_t>(section_end_ - payload_start_);
200   }
201 
section_end() const202   const byte* section_end() const { return section_end_; }
203 
204   // Advances to the next section, checking that decoding the current section
205   // stopped at {section_end_}.
advance(bool move_to_section_end=false)206   void advance(bool move_to_section_end = false) {
207     if (move_to_section_end && decoder_->pc() < section_end_) {
208       decoder_->consume_bytes(
209           static_cast<uint32_t>(section_end_ - decoder_->pc()));
210     }
211     if (decoder_->pc() != section_end_) {
212       const char* msg = decoder_->pc() < section_end_ ? "shorter" : "longer";
213       decoder_->errorf(decoder_->pc(),
214                        "section was %s than expected size "
215                        "(%u bytes expected, %zu decoded)",
216                        msg, section_length(),
217                        static_cast<size_t>(decoder_->pc() - section_start_));
218     }
219     next();
220   }
221 
222  private:
223   Decoder* decoder_;
224   SectionCode section_code_;
225   const byte* section_start_;
226   const byte* payload_start_;
227   const byte* section_end_;
228 
229   // Reads the section code/name at the current position and sets up
230   // the embedder fields.
next()231   void next() {
232     if (!decoder_->more()) {
233       section_code_ = kUnknownSectionCode;
234       return;
235     }
236     section_start_ = decoder_->pc();
237     uint8_t section_code = decoder_->consume_u8("section code");
238     // Read and check the section size.
239     uint32_t section_length = decoder_->consume_u32v("section length");
240 
241     payload_start_ = decoder_->pc();
242     if (decoder_->checkAvailable(section_length)) {
243       // Get the limit of the section within the module.
244       section_end_ = payload_start_ + section_length;
245     } else {
246       // The section would extend beyond the end of the module.
247       section_end_ = payload_start_;
248     }
249 
250     if (section_code == kUnknownSectionCode) {
251       // Check for the known "name", "sourceMappingURL", or "compilationHints"
252       // section.
253       // To identify the unknown section we set the end of the decoder bytes to
254       // the end of the custom section, so that we do not read the section name
255       // beyond the end of the section.
256       const byte* module_end = decoder_->end();
257       decoder_->set_end(section_end_);
258       section_code = IdentifyUnknownSectionInternal(decoder_);
259       if (decoder_->ok()) decoder_->set_end(module_end);
260       // As a side effect, the above function will forward the decoder to after
261       // the identifier string.
262       payload_start_ = decoder_->pc();
263     } else if (!IsValidSectionCode(section_code)) {
264       decoder_->errorf(decoder_->pc(), "unknown section code #0x%02x",
265                        section_code);
266       section_code = kUnknownSectionCode;
267     }
268     section_code_ = decoder_->failed() ? kUnknownSectionCode
269                                        : static_cast<SectionCode>(section_code);
270 
271     if (section_code_ == kUnknownSectionCode && section_end_ > decoder_->pc()) {
272       // skip to the end of the unknown section.
273       uint32_t remaining = static_cast<uint32_t>(section_end_ - decoder_->pc());
274       decoder_->consume_bytes(remaining, "section payload");
275     }
276   }
277 };
278 
279 }  // namespace
280 
281 // The main logic for decoding the bytes of a module.
282 class ModuleDecoderImpl : public Decoder {
283  public:
ModuleDecoderImpl(const WasmFeatures & enabled,ModuleOrigin origin)284   explicit ModuleDecoderImpl(const WasmFeatures& enabled, ModuleOrigin origin)
285       : Decoder(nullptr, nullptr),
286         enabled_features_(enabled),
287         origin_(origin) {}
288 
ModuleDecoderImpl(const WasmFeatures & enabled,const byte * module_start,const byte * module_end,ModuleOrigin origin)289   ModuleDecoderImpl(const WasmFeatures& enabled, const byte* module_start,
290                     const byte* module_end, ModuleOrigin origin)
291       : Decoder(module_start, module_end),
292         enabled_features_(enabled),
293         module_start_(module_start),
294         module_end_(module_end),
295         origin_(origin) {
296     if (end_ < start_) {
297       error(start_, "end is less than start");
298       end_ = start_;
299     }
300   }
301 
onFirstError()302   void onFirstError() override {
303     pc_ = end_;  // On error, terminate section decoding loop.
304   }
305 
DumpModule(const base::Vector<const byte> module_bytes)306   void DumpModule(const base::Vector<const byte> module_bytes) {
307     std::string path;
308     if (FLAG_dump_wasm_module_path) {
309       path = FLAG_dump_wasm_module_path;
310       if (path.size() &&
311           !base::OS::isDirectorySeparator(path[path.size() - 1])) {
312         path += base::OS::DirectorySeparator();
313       }
314     }
315     // File are named `HASH.{ok,failed}.wasm`.
316     size_t hash = base::hash_range(module_bytes.begin(), module_bytes.end());
317     base::EmbeddedVector<char, 32> buf;
318     SNPrintF(buf, "%016zx.%s.wasm", hash, ok() ? "ok" : "failed");
319     path += buf.begin();
320     size_t rv = 0;
321     if (FILE* file = base::OS::FOpen(path.c_str(), "wb")) {
322       rv = fwrite(module_bytes.begin(), module_bytes.length(), 1, file);
323       base::Fclose(file);
324     }
325     if (rv != 1) {
326       OFStream os(stderr);
327       os << "Error while dumping wasm file to " << path << std::endl;
328     }
329   }
330 
StartDecoding(Counters * counters,AccountingAllocator * allocator)331   void StartDecoding(Counters* counters, AccountingAllocator* allocator) {
332     CHECK_NULL(module_);
333     SetCounters(counters);
334     module_.reset(
335         new WasmModule(std::make_unique<Zone>(allocator, "signatures")));
336     module_->initial_pages = 0;
337     module_->maximum_pages = 0;
338     module_->mem_export = false;
339     module_->origin = origin_;
340   }
341 
DecodeModuleHeader(base::Vector<const uint8_t> bytes,uint8_t offset)342   void DecodeModuleHeader(base::Vector<const uint8_t> bytes, uint8_t offset) {
343     if (failed()) return;
344     Reset(bytes, offset);
345 
346     const byte* pos = pc_;
347     uint32_t magic_word = consume_u32("wasm magic");
348 #define BYTES(x) (x & 0xFF), (x >> 8) & 0xFF, (x >> 16) & 0xFF, (x >> 24) & 0xFF
349     if (magic_word != kWasmMagic) {
350       errorf(pos,
351              "expected magic word %02x %02x %02x %02x, "
352              "found %02x %02x %02x %02x",
353              BYTES(kWasmMagic), BYTES(magic_word));
354     }
355 
356     pos = pc_;
357     {
358       uint32_t magic_version = consume_u32("wasm version");
359       if (magic_version != kWasmVersion) {
360         errorf(pos,
361                "expected version %02x %02x %02x %02x, "
362                "found %02x %02x %02x %02x",
363                BYTES(kWasmVersion), BYTES(magic_version));
364       }
365     }
366 #undef BYTES
367   }
368 
CheckSectionOrder(SectionCode section_code,SectionCode prev_section_code,SectionCode next_section_code)369   bool CheckSectionOrder(SectionCode section_code,
370                          SectionCode prev_section_code,
371                          SectionCode next_section_code) {
372     if (next_ordered_section_ > next_section_code) {
373       errorf(pc(), "The %s section must appear before the %s section",
374              SectionName(section_code), SectionName(next_section_code));
375       return false;
376     }
377     if (next_ordered_section_ <= prev_section_code) {
378       next_ordered_section_ = prev_section_code + 1;
379     }
380     return true;
381   }
382 
CheckUnorderedSection(SectionCode section_code)383   bool CheckUnorderedSection(SectionCode section_code) {
384     if (has_seen_unordered_section(section_code)) {
385       errorf(pc(), "Multiple %s sections not allowed",
386              SectionName(section_code));
387       return false;
388     }
389     set_seen_unordered_section(section_code);
390     return true;
391   }
392 
DecodeSection(SectionCode section_code,base::Vector<const uint8_t> bytes,uint32_t offset,bool verify_functions=true)393   void DecodeSection(SectionCode section_code,
394                      base::Vector<const uint8_t> bytes, uint32_t offset,
395                      bool verify_functions = true) {
396     if (failed()) return;
397     Reset(bytes, offset);
398     TRACE("Section: %s\n", SectionName(section_code));
399     TRACE("Decode Section %p - %p\n", bytes.begin(), bytes.end());
400 
401     // Check if the section is out-of-order.
402     if (section_code < next_ordered_section_ &&
403         section_code < kFirstUnorderedSection) {
404       errorf(pc(), "unexpected section <%s>", SectionName(section_code));
405       return;
406     }
407 
408     switch (section_code) {
409       case kUnknownSectionCode:
410         break;
411       case kDataCountSectionCode:
412         if (!CheckUnorderedSection(section_code)) return;
413         // If wasm-gc is enabled, we allow the data cound section anywhere in
414         // the module.
415         if (!enabled_features_.has_gc() &&
416             !CheckSectionOrder(section_code, kElementSectionCode,
417                                kCodeSectionCode)) {
418           return;
419         }
420         break;
421       case kTagSectionCode:
422         if (!CheckUnorderedSection(section_code)) return;
423         if (!CheckSectionOrder(section_code, kMemorySectionCode,
424                                kGlobalSectionCode)) {
425           return;
426         }
427         break;
428       case kNameSectionCode:
429         // TODO(titzer): report out of place name section as a warning.
430         // Be lenient with placement of name section. All except first
431         // occurrence are ignored.
432       case kSourceMappingURLSectionCode:
433         // sourceMappingURL is a custom section and currently can occur anywhere
434         // in the module. In case of multiple sourceMappingURL sections, all
435         // except the first occurrence are ignored.
436       case kDebugInfoSectionCode:
437         // .debug_info is a custom section containing core DWARF information
438         // if produced by compiler. Its presence likely means that Wasm was
439         // built in a debug mode.
440       case kExternalDebugInfoSectionCode:
441         // external_debug_info is a custom section containing a reference to an
442         // external symbol file.
443       case kCompilationHintsSectionCode:
444         // TODO(frgossen): report out of place compilation hints section as a
445         // warning.
446         // Be lenient with placement of compilation hints section. All except
447         // first occurrence after function section and before code section are
448         // ignored.
449         break;
450       case kBranchHintsSectionCode:
451         // TODO(yuri): report out of place branch hints section as a
452         // warning.
453         // Be lenient with placement of compilation hints section. All except
454         // first occurrence after function section and before code section are
455         // ignored.
456         break;
457       default:
458         next_ordered_section_ = section_code + 1;
459         break;
460     }
461 
462     switch (section_code) {
463       case kUnknownSectionCode:
464         break;
465       case kTypeSectionCode:
466         DecodeTypeSection();
467         break;
468       case kImportSectionCode:
469         DecodeImportSection();
470         break;
471       case kFunctionSectionCode:
472         DecodeFunctionSection();
473         break;
474       case kTableSectionCode:
475         DecodeTableSection();
476         break;
477       case kMemorySectionCode:
478         DecodeMemorySection();
479         break;
480       case kGlobalSectionCode:
481         DecodeGlobalSection();
482         break;
483       case kExportSectionCode:
484         DecodeExportSection();
485         break;
486       case kStartSectionCode:
487         DecodeStartSection();
488         break;
489       case kCodeSectionCode:
490         DecodeCodeSection(verify_functions);
491         break;
492       case kElementSectionCode:
493         DecodeElementSection();
494         break;
495       case kDataSectionCode:
496         DecodeDataSection();
497         break;
498       case kNameSectionCode:
499         DecodeNameSection();
500         break;
501       case kSourceMappingURLSectionCode:
502         DecodeSourceMappingURLSection();
503         break;
504       case kDebugInfoSectionCode:
505         // If there is an explicit source map, prefer it over DWARF info.
506         if (module_->debug_symbols.type == WasmDebugSymbols::Type::None) {
507           module_->debug_symbols = {WasmDebugSymbols::Type::EmbeddedDWARF, {}};
508         }
509         consume_bytes(static_cast<uint32_t>(end_ - start_), ".debug_info");
510         break;
511       case kExternalDebugInfoSectionCode:
512         DecodeExternalDebugInfoSection();
513         break;
514       case kCompilationHintsSectionCode:
515         if (enabled_features_.has_compilation_hints()) {
516           DecodeCompilationHintsSection();
517         } else {
518           // Ignore this section when feature was disabled. It is an optional
519           // custom section anyways.
520           consume_bytes(static_cast<uint32_t>(end_ - start_), nullptr);
521         }
522         break;
523       case kBranchHintsSectionCode:
524         if (enabled_features_.has_branch_hinting()) {
525           DecodeBranchHintsSection();
526         } else {
527           // Ignore this section when feature was disabled. It is an optional
528           // custom section anyways.
529           consume_bytes(static_cast<uint32_t>(end_ - start_), nullptr);
530         }
531         break;
532       case kDataCountSectionCode:
533         DecodeDataCountSection();
534         break;
535       case kTagSectionCode:
536         if (enabled_features_.has_eh()) {
537           DecodeTagSection();
538         } else {
539           errorf(pc(),
540                  "unexpected section <%s> (enable with --experimental-wasm-eh)",
541                  SectionName(section_code));
542         }
543         break;
544       default:
545         errorf(pc(), "unexpected section <%s>", SectionName(section_code));
546         return;
547     }
548 
549     if (pc() != bytes.end()) {
550       const char* msg = pc() < bytes.end() ? "shorter" : "longer";
551       errorf(pc(),
552              "section was %s than expected size "
553              "(%zu bytes expected, %zu decoded)",
554              msg, bytes.size(), static_cast<size_t>(pc() - bytes.begin()));
555     }
556   }
557 
consume_base_type_definition()558   TypeDefinition consume_base_type_definition() {
559     DCHECK(enabled_features_.has_gc());
560     uint8_t kind = consume_u8("type kind");
561     switch (kind) {
562       case kWasmFunctionTypeCode: {
563         const FunctionSig* sig = consume_sig(module_->signature_zone.get());
564         return {sig, kNoSuperType};
565       }
566       case kWasmStructTypeCode: {
567         const StructType* type = consume_struct(module_->signature_zone.get());
568         return {type, kNoSuperType};
569       }
570       case kWasmArrayTypeCode: {
571         const ArrayType* type = consume_array(module_->signature_zone.get());
572         return {type, kNoSuperType};
573       }
574       case kWasmFunctionNominalCode:
575       case kWasmArrayNominalCode:
576       case kWasmStructNominalCode:
577         errorf(pc() - 1,
578                "mixing nominal and isorecursive types is not allowed");
579         return {};
580       default:
581         errorf(pc() - 1, "unknown type form: %d", kind);
582         return {};
583     }
584   }
585 
check_supertype(uint32_t supertype)586   bool check_supertype(uint32_t supertype) {
587     if (V8_UNLIKELY(supertype >= module_->types.size())) {
588       errorf(pc(), "type %zu: forward-declared supertype %d",
589              module_->types.size(), supertype);
590       return false;
591     }
592     return true;
593   }
594 
consume_nominal_type_definition()595   TypeDefinition consume_nominal_type_definition() {
596     DCHECK(enabled_features_.has_gc());
597     size_t num_types = module_->types.size();
598     uint8_t kind = consume_u8("type kind");
599     switch (kind) {
600       case kWasmFunctionNominalCode: {
601         const FunctionSig* sig = consume_sig(module_->signature_zone.get());
602         uint32_t super_index = kNoSuperType;
603         HeapType super_type = consume_super_type();
604         if (super_type.is_index()) {
605           super_index = super_type.representation();
606         } else if (V8_UNLIKELY(super_type != HeapType::kFunc)) {
607           errorf(pc() - 1, "type %zu: invalid supertype %d", num_types,
608                  super_type.code());
609           return {};
610         }
611         return {sig, super_index};
612       }
613       case kWasmStructNominalCode: {
614         const StructType* type = consume_struct(module_->signature_zone.get());
615         uint32_t super_index = kNoSuperType;
616         HeapType super_type = consume_super_type();
617         if (super_type.is_index()) {
618           super_index = super_type.representation();
619         } else if (V8_UNLIKELY(super_type != HeapType::kData)) {
620           errorf(pc() - 1, "type %zu: invalid supertype %d", num_types,
621                  super_type.code());
622           return {};
623         }
624         return {type, super_index};
625       }
626       case kWasmArrayNominalCode: {
627         const ArrayType* type = consume_array(module_->signature_zone.get());
628         uint32_t super_index = kNoSuperType;
629         HeapType super_type = consume_super_type();
630         if (super_type.is_index()) {
631           super_index = super_type.representation();
632         } else if (V8_UNLIKELY(super_type != HeapType::kData)) {
633           errorf(pc() - 1, "type %zu: invalid supertype %d", num_types,
634                  super_type.code());
635           return {};
636         }
637         return {type, super_index};
638       }
639       case kWasmFunctionTypeCode:
640       case kWasmArrayTypeCode:
641       case kWasmStructTypeCode:
642       case kWasmSubtypeCode:
643       case kWasmRecursiveTypeGroupCode:
644         errorf(pc() - 1,
645                "mixing nominal and isorecursive types is not allowed");
646         return {};
647       default:
648         errorf(pc() - 1, "unknown type form: %d", kind);
649         return {};
650     }
651   }
652 
consume_subtype_definition()653   TypeDefinition consume_subtype_definition() {
654     DCHECK(enabled_features_.has_gc());
655     uint8_t kind = read_u8<Decoder::kFullValidation>(pc(), "type kind");
656     if (kind == kWasmSubtypeCode) {
657       consume_bytes(1, "subtype definition");
658       constexpr uint32_t kMaximumSupertypes = 1;
659       uint32_t supertype_count =
660           consume_count("supertype count", kMaximumSupertypes);
661       uint32_t supertype =
662           supertype_count == 1 ? consume_u32v("supertype") : kNoSuperType;
663       if (!check_supertype(supertype)) return {};
664       TypeDefinition type = consume_base_type_definition();
665       type.supertype = supertype;
666       return type;
667     } else {
668       return consume_base_type_definition();
669     }
670   }
671 
DecodeTypeSection()672   void DecodeTypeSection() {
673     TypeCanonicalizer* type_canon = GetTypeCanonicalizer();
674     uint32_t types_count = consume_count("types count", kV8MaxWasmTypes);
675 
676     // Non wasm-gc type section decoding.
677     if (!enabled_features_.has_gc()) {
678       module_->types.reserve(types_count);
679       for (uint32_t i = 0; i < types_count; ++i) {
680         TRACE("DecodeSignature[%d] module+%d\n", i,
681               static_cast<int>(pc_ - start_));
682         expect_u8("signature definition", kWasmFunctionTypeCode);
683         const FunctionSig* sig = consume_sig(module_->signature_zone.get());
684         if (!ok()) break;
685         module_->add_signature(sig, kNoSuperType);
686         if (FLAG_wasm_type_canonicalization) {
687           type_canon->AddRecursiveGroup(module_.get(), 1);
688         }
689       }
690       return;
691     }
692 
693     if (types_count > 0) {
694       uint8_t first_type_opcode = this->read_u8<Decoder::kFullValidation>(pc());
695       if (first_type_opcode == kWasmFunctionNominalCode ||
696           first_type_opcode == kWasmStructNominalCode ||
697           first_type_opcode == kWasmArrayNominalCode) {
698         // wasm-gc nominal type section decoding.
699         // In a nominal module, all types belong in the same recursive group. We
700         // use the type vector's capacity to mark the end of the current
701         // recursive group.
702         module_->types.reserve(types_count);
703         for (uint32_t i = 0; ok() && i < types_count; ++i) {
704           TRACE("DecodeType[%d] module+%d\n", i,
705                 static_cast<int>(pc_ - start_));
706           TypeDefinition type = consume_nominal_type_definition();
707           if (ok()) module_->add_type(type);
708         }
709         if (ok() && FLAG_wasm_type_canonicalization) {
710           type_canon->AddRecursiveGroup(module_.get(), types_count);
711         }
712       } else {
713         // wasm-gc isorecursive type section decoding.
714         for (uint32_t i = 0; ok() && i < types_count; ++i) {
715           TRACE("DecodeType[%d] module+%d\n", i,
716                 static_cast<int>(pc_ - start_));
717           uint8_t kind = read_u8<Decoder::kFullValidation>(pc(), "type kind");
718           if (kind == kWasmRecursiveTypeGroupCode) {
719             consume_bytes(1, "rec. group definition");
720             uint32_t group_size =
721                 consume_count("recursive group size", kV8MaxWasmTypes);
722             if (module_->types.size() + group_size > kV8MaxWasmTypes) {
723               errorf(pc(), "Type definition count exeeds maximum %zu",
724                      kV8MaxWasmTypes);
725               return;
726             }
727             // Reserve space for the current recursive group, so we are
728             // allowed to reference its elements.
729             module_->types.reserve(module_->types.size() + group_size);
730             for (uint32_t i = 0; i < group_size; i++) {
731               TypeDefinition type = consume_subtype_definition();
732               if (ok()) module_->add_type(type);
733             }
734             if (ok() && FLAG_wasm_type_canonicalization) {
735               type_canon->AddRecursiveGroup(module_.get(), group_size);
736             }
737           } else {
738             TypeDefinition type = consume_subtype_definition();
739             if (ok()) {
740               module_->add_type(type);
741               if (FLAG_wasm_type_canonicalization) {
742                 type_canon->AddRecursiveGroup(module_.get(), 1);
743               }
744             }
745           }
746         }
747       }
748     }
749 
750     // Check validity of explicitly defined supertypes.
751     const WasmModule* module = module_.get();
752     for (uint32_t i = 0; ok() && i < types_count; ++i) {
753       uint32_t explicit_super = module_->supertype(i);
754       if (explicit_super == kNoSuperType) continue;
755       DCHECK_LT(explicit_super, types_count);  // {consume_super_type} checks.
756       int depth = GetSubtypingDepth(module, i);
757       if (depth > static_cast<int>(kV8MaxRttSubtypingDepth)) {
758         errorf("type %d: subtyping depth is greater than allowed", i);
759         continue;
760       }
761       // TODO(7748): Replace this with a DCHECK once we reject inheritance
762       // cycles for nominal modules.
763       if (depth == -1) {
764         errorf("type %d: cyclic inheritance", i);
765         continue;
766       }
767       if (!ValidSubtypeDefinition(i, explicit_super, module, module)) {
768         errorf("type %d has invalid explicit supertype %d", i, explicit_super);
769         continue;
770       }
771     }
772     module_->signature_map.Freeze();
773   }
774 
DecodeImportSection()775   void DecodeImportSection() {
776     uint32_t import_table_count =
777         consume_count("imports count", kV8MaxWasmImports);
778     module_->import_table.reserve(import_table_count);
779     for (uint32_t i = 0; ok() && i < import_table_count; ++i) {
780       TRACE("DecodeImportTable[%d] module+%d\n", i,
781             static_cast<int>(pc_ - start_));
782 
783       module_->import_table.push_back({
784           {0, 0},             // module_name
785           {0, 0},             // field_name
786           kExternalFunction,  // kind
787           0                   // index
788       });
789       WasmImport* import = &module_->import_table.back();
790       const byte* pos = pc_;
791       import->module_name = consume_string(this, true, "module name");
792       import->field_name = consume_string(this, true, "field name");
793       import->kind =
794           static_cast<ImportExportKindCode>(consume_u8("import kind"));
795       switch (import->kind) {
796         case kExternalFunction: {
797           // ===== Imported function ===========================================
798           import->index = static_cast<uint32_t>(module_->functions.size());
799           module_->num_imported_functions++;
800           module_->functions.push_back({nullptr,        // sig
801                                         import->index,  // func_index
802                                         0,              // sig_index
803                                         {0, 0},         // code
804                                         0,              // feedback slots
805                                         true,           // imported
806                                         false,          // exported
807                                         false});        // declared
808           WasmFunction* function = &module_->functions.back();
809           function->sig_index =
810               consume_sig_index(module_.get(), &function->sig);
811           break;
812         }
813         case kExternalTable: {
814           // ===== Imported table ==============================================
815           import->index = static_cast<uint32_t>(module_->tables.size());
816           module_->num_imported_tables++;
817           module_->tables.emplace_back();
818           WasmTable* table = &module_->tables.back();
819           table->imported = true;
820           const byte* type_position = pc();
821           ValueType type = consume_reference_type();
822           if (!WasmTable::IsValidTableType(type, module_.get())) {
823             errorf(type_position, "Invalid table type %s", type.name().c_str());
824             break;
825           }
826           table->type = type;
827           uint8_t flags = validate_table_flags("element count");
828           consume_resizable_limits(
829               "element count", "elements", std::numeric_limits<uint32_t>::max(),
830               &table->initial_size, &table->has_maximum_size,
831               std::numeric_limits<uint32_t>::max(), &table->maximum_size,
832               flags);
833           break;
834         }
835         case kExternalMemory: {
836           // ===== Imported memory =============================================
837           if (!AddMemory(module_.get())) break;
838           uint8_t flags = validate_memory_flags(&module_->has_shared_memory,
839                                                 &module_->is_memory64);
840           consume_resizable_limits(
841               "memory", "pages", kSpecMaxMemoryPages, &module_->initial_pages,
842               &module_->has_maximum_pages, kSpecMaxMemoryPages,
843               &module_->maximum_pages, flags);
844           break;
845         }
846         case kExternalGlobal: {
847           // ===== Imported global =============================================
848           import->index = static_cast<uint32_t>(module_->globals.size());
849           module_->globals.push_back({kWasmVoid, false, {}, {0}, true, false});
850           WasmGlobal* global = &module_->globals.back();
851           global->type = consume_value_type();
852           global->mutability = consume_mutability();
853           if (global->mutability) {
854             module_->num_imported_mutable_globals++;
855           }
856           break;
857         }
858         case kExternalTag: {
859           // ===== Imported tag ================================================
860           if (!enabled_features_.has_eh()) {
861             errorf(pos, "unknown import kind 0x%02x", import->kind);
862             break;
863           }
864           import->index = static_cast<uint32_t>(module_->tags.size());
865           const WasmTagSig* tag_sig = nullptr;
866           consume_exception_attribute();  // Attribute ignored for now.
867           consume_tag_sig_index(module_.get(), &tag_sig);
868           module_->tags.emplace_back(tag_sig);
869           break;
870         }
871         default:
872           errorf(pos, "unknown import kind 0x%02x", import->kind);
873           break;
874       }
875     }
876   }
877 
DecodeFunctionSection()878   void DecodeFunctionSection() {
879     uint32_t functions_count =
880         consume_count("functions count", kV8MaxWasmFunctions);
881     auto counter =
882         SELECT_WASM_COUNTER(GetCounters(), origin_, wasm_functions_per, module);
883     counter->AddSample(static_cast<int>(functions_count));
884     DCHECK_EQ(module_->functions.size(), module_->num_imported_functions);
885     uint32_t total_function_count =
886         module_->num_imported_functions + functions_count;
887     module_->functions.reserve(total_function_count);
888     module_->num_declared_functions = functions_count;
889     for (uint32_t i = 0; i < functions_count; ++i) {
890       uint32_t func_index = static_cast<uint32_t>(module_->functions.size());
891       module_->functions.push_back({nullptr,     // sig
892                                     func_index,  // func_index
893                                     0,           // sig_index
894                                     {0, 0},      // code
895                                     0,           // feedback slots
896                                     false,       // imported
897                                     false,       // exported
898                                     false});     // declared
899       WasmFunction* function = &module_->functions.back();
900       function->sig_index = consume_sig_index(module_.get(), &function->sig);
901       if (!ok()) return;
902     }
903     DCHECK_EQ(module_->functions.size(), total_function_count);
904   }
905 
DecodeTableSection()906   void DecodeTableSection() {
907     uint32_t table_count = consume_count("table count", kV8MaxWasmTables);
908 
909     for (uint32_t i = 0; ok() && i < table_count; i++) {
910       module_->tables.emplace_back();
911       WasmTable* table = &module_->tables.back();
912       const byte* type_position = pc();
913       ValueType table_type = consume_reference_type();
914       if (!WasmTable::IsValidTableType(table_type, module_.get())) {
915         error(type_position,
916               "Currently, only externref and function references are allowed "
917               "as table types");
918         continue;
919       }
920       table->type = table_type;
921       uint8_t flags = validate_table_flags("table elements");
922       consume_resizable_limits(
923           "table elements", "elements", std::numeric_limits<uint32_t>::max(),
924           &table->initial_size, &table->has_maximum_size,
925           std::numeric_limits<uint32_t>::max(), &table->maximum_size, flags);
926       if (!table_type.is_defaultable()) {
927         table->initial_value = consume_init_expr(module_.get(), table_type);
928       }
929     }
930   }
931 
DecodeMemorySection()932   void DecodeMemorySection() {
933     uint32_t memory_count = consume_count("memory count", kV8MaxWasmMemories);
934 
935     for (uint32_t i = 0; ok() && i < memory_count; i++) {
936       if (!AddMemory(module_.get())) break;
937       uint8_t flags = validate_memory_flags(&module_->has_shared_memory,
938                                             &module_->is_memory64);
939       consume_resizable_limits("memory", "pages", kSpecMaxMemoryPages,
940                                &module_->initial_pages,
941                                &module_->has_maximum_pages, kSpecMaxMemoryPages,
942                                &module_->maximum_pages, flags);
943     }
944   }
945 
DecodeGlobalSection()946   void DecodeGlobalSection() {
947     uint32_t globals_count = consume_count("globals count", kV8MaxWasmGlobals);
948     uint32_t imported_globals = static_cast<uint32_t>(module_->globals.size());
949     // It is important to not resize the globals vector from the beginning,
950     // because we use its current size when decoding the initializer.
951     module_->globals.reserve(imported_globals + globals_count);
952     for (uint32_t i = 0; ok() && i < globals_count; ++i) {
953       TRACE("DecodeGlobal[%d] module+%d\n", i, static_cast<int>(pc_ - start_));
954       ValueType type = consume_value_type();
955       bool mutability = consume_mutability();
956       if (failed()) break;
957       ConstantExpression init = consume_init_expr(module_.get(), type);
958       module_->globals.push_back({type, mutability, init, {0}, false, false});
959     }
960     if (ok()) CalculateGlobalOffsets(module_.get());
961   }
962 
DecodeExportSection()963   void DecodeExportSection() {
964     uint32_t export_table_count =
965         consume_count("exports count", kV8MaxWasmExports);
966     module_->export_table.reserve(export_table_count);
967     for (uint32_t i = 0; ok() && i < export_table_count; ++i) {
968       TRACE("DecodeExportTable[%d] module+%d\n", i,
969             static_cast<int>(pc_ - start_));
970 
971       module_->export_table.push_back({
972           {0, 0},             // name
973           kExternalFunction,  // kind
974           0                   // index
975       });
976       WasmExport* exp = &module_->export_table.back();
977 
978       exp->name = consume_string(this, true, "field name");
979 
980       const byte* pos = pc();
981       exp->kind = static_cast<ImportExportKindCode>(consume_u8("export kind"));
982       switch (exp->kind) {
983         case kExternalFunction: {
984           WasmFunction* func = nullptr;
985           exp->index =
986               consume_func_index(module_.get(), &func, "export function index");
987 
988           if (failed()) break;
989           DCHECK_NOT_NULL(func);
990 
991           module_->num_exported_functions++;
992           func->exported = true;
993           // Exported functions are considered "declared".
994           func->declared = true;
995           break;
996         }
997         case kExternalTable: {
998           WasmTable* table = nullptr;
999           exp->index = consume_table_index(module_.get(), &table);
1000           if (table) table->exported = true;
1001           break;
1002         }
1003         case kExternalMemory: {
1004           uint32_t index = consume_u32v("memory index");
1005           // TODO(titzer): This should become more regular
1006           // once we support multiple memories.
1007           if (!module_->has_memory || index != 0) {
1008             error("invalid memory index != 0");
1009           }
1010           module_->mem_export = true;
1011           break;
1012         }
1013         case kExternalGlobal: {
1014           WasmGlobal* global = nullptr;
1015           exp->index = consume_global_index(module_.get(), &global);
1016           if (global) {
1017             global->exported = true;
1018           }
1019           break;
1020         }
1021         case kExternalTag: {
1022           if (!enabled_features_.has_eh()) {
1023             errorf(pos, "invalid export kind 0x%02x", exp->kind);
1024             break;
1025           }
1026           WasmTag* tag = nullptr;
1027           exp->index = consume_tag_index(module_.get(), &tag);
1028           break;
1029         }
1030         default:
1031           errorf(pos, "invalid export kind 0x%02x", exp->kind);
1032           break;
1033       }
1034     }
1035     // Check for duplicate exports (except for asm.js).
1036     if (ok() && origin_ == kWasmOrigin && module_->export_table.size() > 1) {
1037       std::vector<WasmExport> sorted_exports(module_->export_table);
1038 
1039       auto cmp_less = [this](const WasmExport& a, const WasmExport& b) {
1040         // Return true if a < b.
1041         if (a.name.length() != b.name.length()) {
1042           return a.name.length() < b.name.length();
1043         }
1044         const byte* left = start() + GetBufferRelativeOffset(a.name.offset());
1045         const byte* right = start() + GetBufferRelativeOffset(b.name.offset());
1046         return memcmp(left, right, a.name.length()) < 0;
1047       };
1048       std::stable_sort(sorted_exports.begin(), sorted_exports.end(), cmp_less);
1049 
1050       auto it = sorted_exports.begin();
1051       WasmExport* last = &*it++;
1052       for (auto end = sorted_exports.end(); it != end; last = &*it++) {
1053         DCHECK(!cmp_less(*it, *last));  // Vector must be sorted.
1054         if (!cmp_less(*last, *it)) {
1055           const byte* pc = start() + GetBufferRelativeOffset(it->name.offset());
1056           TruncatedUserString<> name(pc, it->name.length());
1057           errorf(pc, "Duplicate export name '%.*s' for %s %d and %s %d",
1058                  name.length(), name.start(), ExternalKindName(last->kind),
1059                  last->index, ExternalKindName(it->kind), it->index);
1060           break;
1061         }
1062       }
1063     }
1064   }
1065 
DecodeStartSection()1066   void DecodeStartSection() {
1067     WasmFunction* func;
1068     const byte* pos = pc_;
1069     module_->start_function_index =
1070         consume_func_index(module_.get(), &func, "start function index");
1071     if (func &&
1072         (func->sig->parameter_count() > 0 || func->sig->return_count() > 0)) {
1073       error(pos, "invalid start function: non-zero parameter or return count");
1074     }
1075   }
1076 
DecodeElementSection()1077   void DecodeElementSection() {
1078     uint32_t element_count =
1079         consume_count("element count", FLAG_wasm_max_table_size);
1080 
1081     for (uint32_t i = 0; i < element_count; ++i) {
1082       WasmElemSegment segment = consume_element_segment_header();
1083       if (failed()) return;
1084       DCHECK_NE(segment.type, kWasmBottom);
1085 
1086       uint32_t num_elem =
1087           consume_count("number of elements", max_table_init_entries());
1088 
1089       for (uint32_t j = 0; j < num_elem; j++) {
1090         ConstantExpression entry =
1091             segment.element_type == WasmElemSegment::kExpressionElements
1092                 ? consume_init_expr(module_.get(), segment.type)
1093                 : ConstantExpression::RefFunc(
1094                       consume_element_func_index(segment.type));
1095         if (failed()) return;
1096         segment.entries.push_back(entry);
1097       }
1098       module_->elem_segments.push_back(std::move(segment));
1099     }
1100   }
1101 
DecodeCodeSection(bool verify_functions)1102   void DecodeCodeSection(bool verify_functions) {
1103     StartCodeSection();
1104     uint32_t code_section_start = pc_offset();
1105     uint32_t functions_count = consume_u32v("functions count");
1106     CheckFunctionsCount(functions_count, code_section_start);
1107     for (uint32_t i = 0; ok() && i < functions_count; ++i) {
1108       const byte* pos = pc();
1109       uint32_t size = consume_u32v("body size");
1110       if (size > kV8MaxWasmFunctionSize) {
1111         errorf(pos, "size %u > maximum function size %zu", size,
1112                kV8MaxWasmFunctionSize);
1113         return;
1114       }
1115       uint32_t offset = pc_offset();
1116       consume_bytes(size, "function body");
1117       if (failed()) break;
1118       DecodeFunctionBody(i, size, offset, verify_functions);
1119     }
1120     DCHECK_GE(pc_offset(), code_section_start);
1121     set_code_section(code_section_start, pc_offset() - code_section_start);
1122   }
1123 
StartCodeSection()1124   void StartCodeSection() {
1125     if (ok()) {
1126       // Make sure global offset were calculated before they get accessed during
1127       // function compilation.
1128       CalculateGlobalOffsets(module_.get());
1129     }
1130   }
1131 
CheckFunctionsCount(uint32_t functions_count,uint32_t error_offset)1132   bool CheckFunctionsCount(uint32_t functions_count, uint32_t error_offset) {
1133     if (functions_count != module_->num_declared_functions) {
1134       errorf(error_offset, "function body count %u mismatch (%u expected)",
1135              functions_count, module_->num_declared_functions);
1136       return false;
1137     }
1138     return true;
1139   }
1140 
DecodeFunctionBody(uint32_t index,uint32_t length,uint32_t offset,bool verify_functions)1141   void DecodeFunctionBody(uint32_t index, uint32_t length, uint32_t offset,
1142                           bool verify_functions) {
1143     WasmFunction* function =
1144         &module_->functions[index + module_->num_imported_functions];
1145     function->code = {offset, length};
1146     if (verify_functions) {
1147       ModuleWireBytes bytes(module_start_, module_end_);
1148       VerifyFunctionBody(module_->signature_zone->allocator(),
1149                          index + module_->num_imported_functions, bytes,
1150                          module_.get(), function);
1151     }
1152   }
1153 
CheckDataSegmentsCount(uint32_t data_segments_count)1154   bool CheckDataSegmentsCount(uint32_t data_segments_count) {
1155     if (has_seen_unordered_section(kDataCountSectionCode) &&
1156         data_segments_count != module_->num_declared_data_segments) {
1157       errorf(pc(), "data segments count %u mismatch (%u expected)",
1158              data_segments_count, module_->num_declared_data_segments);
1159       return false;
1160     }
1161     return true;
1162   }
1163 
DecodeDataSection()1164   void DecodeDataSection() {
1165     uint32_t data_segments_count =
1166         consume_count("data segments count", kV8MaxWasmDataSegments);
1167     if (!CheckDataSegmentsCount(data_segments_count)) return;
1168 
1169     module_->data_segments.reserve(data_segments_count);
1170     for (uint32_t i = 0; ok() && i < data_segments_count; ++i) {
1171       const byte* pos = pc();
1172       TRACE("DecodeDataSegment[%d] module+%d\n", i,
1173             static_cast<int>(pc_ - start_));
1174 
1175       bool is_active;
1176       uint32_t memory_index;
1177       ConstantExpression dest_addr;
1178       consume_data_segment_header(&is_active, &memory_index, &dest_addr);
1179       if (failed()) break;
1180 
1181       if (is_active) {
1182         if (!module_->has_memory) {
1183           error("cannot load data without memory");
1184           break;
1185         }
1186         if (memory_index != 0) {
1187           errorf(pos, "illegal memory index %u != 0", memory_index);
1188           break;
1189         }
1190       }
1191 
1192       uint32_t source_length = consume_u32v("source size");
1193       uint32_t source_offset = pc_offset();
1194 
1195       if (is_active) {
1196         module_->data_segments.emplace_back(std::move(dest_addr));
1197       } else {
1198         module_->data_segments.emplace_back();
1199       }
1200 
1201       WasmDataSegment* segment = &module_->data_segments.back();
1202 
1203       consume_bytes(source_length, "segment data");
1204       if (failed()) break;
1205 
1206       segment->source = {source_offset, source_length};
1207     }
1208   }
1209 
DecodeNameSection()1210   void DecodeNameSection() {
1211     // TODO(titzer): find a way to report name errors as warnings.
1212     // Ignore all but the first occurrence of name section.
1213     if (!has_seen_unordered_section(kNameSectionCode)) {
1214       set_seen_unordered_section(kNameSectionCode);
1215       // Use an inner decoder so that errors don't fail the outer decoder.
1216       Decoder inner(start_, pc_, end_, buffer_offset_);
1217       // Decode all name subsections.
1218       // Be lenient with their order.
1219       while (inner.ok() && inner.more()) {
1220         uint8_t name_type = inner.consume_u8("name type");
1221         if (name_type & 0x80) inner.error("name type if not varuint7");
1222 
1223         uint32_t name_payload_len = inner.consume_u32v("name payload length");
1224         if (!inner.checkAvailable(name_payload_len)) break;
1225 
1226         // Decode module name, ignore the rest.
1227         // Function and local names will be decoded when needed.
1228         if (name_type == NameSectionKindCode::kModuleCode) {
1229           WireBytesRef name = consume_string(&inner, false, "module name");
1230           if (inner.ok() && validate_utf8(&inner, name)) {
1231             module_->name = name;
1232           }
1233         } else {
1234           inner.consume_bytes(name_payload_len, "name subsection payload");
1235         }
1236       }
1237     }
1238     // Skip the whole names section in the outer decoder.
1239     consume_bytes(static_cast<uint32_t>(end_ - start_), nullptr);
1240   }
1241 
DecodeSourceMappingURLSection()1242   void DecodeSourceMappingURLSection() {
1243     Decoder inner(start_, pc_, end_, buffer_offset_);
1244     WireBytesRef url = wasm::consume_string(&inner, true, "module name");
1245     if (inner.ok() &&
1246         module_->debug_symbols.type != WasmDebugSymbols::Type::SourceMap) {
1247       module_->debug_symbols = {WasmDebugSymbols::Type::SourceMap, url};
1248     }
1249     set_seen_unordered_section(kSourceMappingURLSectionCode);
1250     consume_bytes(static_cast<uint32_t>(end_ - start_), nullptr);
1251   }
1252 
DecodeExternalDebugInfoSection()1253   void DecodeExternalDebugInfoSection() {
1254     Decoder inner(start_, pc_, end_, buffer_offset_);
1255     WireBytesRef url =
1256         wasm::consume_string(&inner, true, "external symbol file");
1257     // If there is an explicit source map, prefer it over DWARF info.
1258     if (inner.ok() &&
1259         module_->debug_symbols.type != WasmDebugSymbols::Type::SourceMap) {
1260       module_->debug_symbols = {WasmDebugSymbols::Type::ExternalDWARF, url};
1261       set_seen_unordered_section(kExternalDebugInfoSectionCode);
1262     }
1263     consume_bytes(static_cast<uint32_t>(end_ - start_), nullptr);
1264   }
1265 
DecodeCompilationHintsSection()1266   void DecodeCompilationHintsSection() {
1267     TRACE("DecodeCompilationHints module+%d\n", static_cast<int>(pc_ - start_));
1268 
1269     // TODO(frgossen): Find a way to report compilation hint errors as warnings.
1270     // All except first occurrence after function section and before code
1271     // section are ignored.
1272     const bool before_function_section =
1273         next_ordered_section_ <= kFunctionSectionCode;
1274     const bool after_code_section = next_ordered_section_ > kCodeSectionCode;
1275     if (before_function_section || after_code_section ||
1276         has_seen_unordered_section(kCompilationHintsSectionCode)) {
1277       return;
1278     }
1279     set_seen_unordered_section(kCompilationHintsSectionCode);
1280 
1281     // TODO(frgossen) Propagate errors to outer decoder in experimental phase.
1282     // We should use an inner decoder later and propagate its errors as
1283     // warnings.
1284     Decoder& decoder = *this;
1285     // Decoder decoder(start_, pc_, end_, buffer_offset_);
1286 
1287     // Ensure exactly one compilation hint per function.
1288     uint32_t hint_count = decoder.consume_u32v("compilation hint count");
1289     if (hint_count != module_->num_declared_functions) {
1290       decoder.errorf(decoder.pc(), "Expected %u compilation hints (%u found)",
1291                      module_->num_declared_functions, hint_count);
1292     }
1293 
1294     // Decode sequence of compilation hints.
1295     if (decoder.ok()) {
1296       module_->compilation_hints.reserve(hint_count);
1297     }
1298     for (uint32_t i = 0; decoder.ok() && i < hint_count; i++) {
1299       TRACE("DecodeCompilationHints[%d] module+%d\n", i,
1300             static_cast<int>(pc_ - start_));
1301 
1302       // Compilation hints are encoded in one byte each.
1303       // +-------+----------+---------------+----------+
1304       // | 2 bit | 2 bit    | 2 bit         | 2 bit    |
1305       // | ...   | Top tier | Baseline tier | Strategy |
1306       // +-------+----------+---------------+----------+
1307       uint8_t hint_byte = decoder.consume_u8("compilation hint");
1308       if (!decoder.ok()) break;
1309 
1310       // Validate the hint_byte.
1311       // For the compilation strategy, all 2-bit values are valid. For the tier,
1312       // only 0x0, 0x1, and 0x2 are allowed.
1313       static_assert(
1314           static_cast<int>(WasmCompilationHintTier::kDefault) == 0 &&
1315               static_cast<int>(WasmCompilationHintTier::kBaseline) == 1 &&
1316               static_cast<int>(WasmCompilationHintTier::kOptimized) == 2,
1317           "The check below assumes that 0x03 is the only invalid 2-bit number "
1318           "for a compilation tier");
1319       if (((hint_byte >> 2) & 0x03) == 0x03 ||
1320           ((hint_byte >> 4) & 0x03) == 0x03) {
1321         decoder.errorf(decoder.pc(),
1322                        "Invalid compilation hint %#04x (invalid tier 0x03)",
1323                        hint_byte);
1324         break;
1325       }
1326 
1327       // Decode compilation hint.
1328       WasmCompilationHint hint;
1329       hint.strategy =
1330           static_cast<WasmCompilationHintStrategy>(hint_byte & 0x03);
1331       hint.baseline_tier =
1332           static_cast<WasmCompilationHintTier>((hint_byte >> 2) & 0x03);
1333       hint.top_tier =
1334           static_cast<WasmCompilationHintTier>((hint_byte >> 4) & 0x03);
1335 
1336       // Ensure that the top tier never downgrades a compilation result. If
1337       // baseline and top tier are the same compilation will be invoked only
1338       // once.
1339       if (hint.top_tier < hint.baseline_tier &&
1340           hint.top_tier != WasmCompilationHintTier::kDefault) {
1341         decoder.errorf(decoder.pc(),
1342                        "Invalid compilation hint %#04x (forbidden downgrade)",
1343                        hint_byte);
1344       }
1345 
1346       // Happily accept compilation hint.
1347       if (decoder.ok()) {
1348         module_->compilation_hints.push_back(std::move(hint));
1349       }
1350     }
1351 
1352     // If section was invalid reset compilation hints.
1353     if (decoder.failed()) {
1354       module_->compilation_hints.clear();
1355     }
1356 
1357     // @TODO(frgossen) Skip the whole compilation hints section in the outer
1358     // decoder if inner decoder was used.
1359     // consume_bytes(static_cast<uint32_t>(end_ - start_), nullptr);
1360   }
1361 
DecodeBranchHintsSection()1362   void DecodeBranchHintsSection() {
1363     TRACE("DecodeBranchHints module+%d\n", static_cast<int>(pc_ - start_));
1364     if (!has_seen_unordered_section(kBranchHintsSectionCode)) {
1365       set_seen_unordered_section(kBranchHintsSectionCode);
1366       // Use an inner decoder so that errors don't fail the outer decoder.
1367       Decoder inner(start_, pc_, end_, buffer_offset_);
1368       BranchHintInfo branch_hints;
1369 
1370       uint32_t func_count = inner.consume_u32v("number of functions");
1371       // Keep track of the previous function index to validate the ordering
1372       int64_t last_func_idx = -1;
1373       for (uint32_t i = 0; i < func_count; i++) {
1374         uint32_t func_idx = inner.consume_u32v("function index");
1375         if (int64_t(func_idx) <= last_func_idx) {
1376           inner.errorf("Invalid function index: %d", func_idx);
1377           break;
1378         }
1379         last_func_idx = func_idx;
1380         uint32_t num_hints = inner.consume_u32v("number of hints");
1381         BranchHintMap func_branch_hints;
1382         TRACE("DecodeBranchHints[%d] module+%d\n", func_idx,
1383               static_cast<int>(inner.pc() - inner.start()));
1384         // Keep track of the previous branch offset to validate the ordering
1385         int64_t last_br_off = -1;
1386         for (uint32_t j = 0; j < num_hints; ++j) {
1387           uint32_t br_off = inner.consume_u32v("branch instruction offset");
1388           if (int64_t(br_off) <= last_br_off) {
1389             inner.errorf("Invalid branch offset: %d", br_off);
1390             break;
1391           }
1392           last_br_off = br_off;
1393           uint32_t data_size = inner.consume_u32v("data size");
1394           if (data_size != 1) {
1395             inner.errorf("Invalid data size: %#x. Expected 1.", data_size);
1396             break;
1397           }
1398           uint32_t br_dir = inner.consume_u8("branch direction");
1399           TRACE("DecodeBranchHints[%d][%d] module+%d\n", func_idx, br_off,
1400                 static_cast<int>(inner.pc() - inner.start()));
1401           WasmBranchHint hint;
1402           switch (br_dir) {
1403             case 0:
1404               hint = WasmBranchHint::kUnlikely;
1405               break;
1406             case 1:
1407               hint = WasmBranchHint::kLikely;
1408               break;
1409             default:
1410               hint = WasmBranchHint::kNoHint;
1411               inner.errorf(inner.pc(), "Invalid branch hint %#x", br_dir);
1412               break;
1413           }
1414           if (!inner.ok()) {
1415             break;
1416           }
1417           func_branch_hints.insert(br_off, hint);
1418         }
1419         if (!inner.ok()) {
1420           break;
1421         }
1422         branch_hints.emplace(func_idx, std::move(func_branch_hints));
1423       }
1424       // Extra unexpected bytes are an error.
1425       if (inner.more()) {
1426         inner.errorf("Unexpected extra bytes: %d\n",
1427                      static_cast<int>(inner.pc() - inner.start()));
1428       }
1429       // If everything went well, accept the hints for the module.
1430       if (inner.ok()) {
1431         module_->branch_hints = std::move(branch_hints);
1432       }
1433     }
1434     // Skip the whole branch hints section in the outer decoder.
1435     consume_bytes(static_cast<uint32_t>(end_ - start_), nullptr);
1436   }
1437 
DecodeDataCountSection()1438   void DecodeDataCountSection() {
1439     module_->num_declared_data_segments =
1440         consume_count("data segments count", kV8MaxWasmDataSegments);
1441   }
1442 
DecodeTagSection()1443   void DecodeTagSection() {
1444     uint32_t tag_count = consume_count("tag count", kV8MaxWasmTags);
1445     for (uint32_t i = 0; ok() && i < tag_count; ++i) {
1446       TRACE("DecodeTag[%d] module+%d\n", i, static_cast<int>(pc_ - start_));
1447       const WasmTagSig* tag_sig = nullptr;
1448       consume_exception_attribute();  // Attribute ignored for now.
1449       consume_tag_sig_index(module_.get(), &tag_sig);
1450       module_->tags.emplace_back(tag_sig);
1451     }
1452   }
1453 
CheckMismatchedCounts()1454   bool CheckMismatchedCounts() {
1455     // The declared vs. defined function count is normally checked when
1456     // decoding the code section, but we have to check it here too in case the
1457     // code section is absent.
1458     if (module_->num_declared_functions != 0) {
1459       DCHECK_LT(module_->num_imported_functions, module_->functions.size());
1460       // We know that the code section has been decoded if the first
1461       // non-imported function has its code set.
1462       if (!module_->functions[module_->num_imported_functions].code.is_set()) {
1463         errorf(pc(), "function count is %u, but code section is absent",
1464                module_->num_declared_functions);
1465         return false;
1466       }
1467     }
1468     // Perform a similar check for the DataCount and Data sections, where data
1469     // segments are declared but the Data section is absent.
1470     if (!CheckDataSegmentsCount(
1471             static_cast<uint32_t>(module_->data_segments.size()))) {
1472       return false;
1473     }
1474     return true;
1475   }
1476 
FinishDecoding(bool verify_functions=true)1477   ModuleResult FinishDecoding(bool verify_functions = true) {
1478     if (ok() && CheckMismatchedCounts()) {
1479       // We calculate the global offsets here, because there may not be a
1480       // global section and code section that would have triggered the
1481       // calculation before. Even without the globals section the calculation
1482       // is needed because globals can also be defined in the import section.
1483       CalculateGlobalOffsets(module_.get());
1484     }
1485 
1486     ModuleResult result = toResult(std::move(module_));
1487     if (verify_functions && result.ok() && intermediate_error_.has_error()) {
1488       // Copy error message and location.
1489       return ModuleResult{std::move(intermediate_error_)};
1490     }
1491     return result;
1492   }
1493 
set_code_section(uint32_t offset,uint32_t size)1494   void set_code_section(uint32_t offset, uint32_t size) {
1495     module_->code = {offset, size};
1496   }
1497 
1498   // Decodes an entire module.
DecodeModule(Counters * counters,AccountingAllocator * allocator,bool verify_functions=true)1499   ModuleResult DecodeModule(Counters* counters, AccountingAllocator* allocator,
1500                             bool verify_functions = true) {
1501     StartDecoding(counters, allocator);
1502     uint32_t offset = 0;
1503     base::Vector<const byte> orig_bytes(start(), end() - start());
1504     DecodeModuleHeader(base::VectorOf(start(), end() - start()), offset);
1505     if (failed()) {
1506       return FinishDecoding(verify_functions);
1507     }
1508     // Size of the module header.
1509     offset += 8;
1510     Decoder decoder(start_ + offset, end_, offset);
1511 
1512     WasmSectionIterator section_iter(&decoder);
1513 
1514     while (ok()) {
1515       // Shift the offset by the section header length
1516       offset += section_iter.payload_start() - section_iter.section_start();
1517       if (section_iter.section_code() != SectionCode::kUnknownSectionCode) {
1518         DecodeSection(section_iter.section_code(), section_iter.payload(),
1519                       offset, verify_functions);
1520       }
1521       // Shift the offset by the remaining section payload
1522       offset += section_iter.payload_length();
1523       if (!section_iter.more()) break;
1524       section_iter.advance(true);
1525     }
1526 
1527     if (FLAG_dump_wasm_module) DumpModule(orig_bytes);
1528 
1529     if (decoder.failed()) {
1530       return decoder.toResult<std::unique_ptr<WasmModule>>(nullptr);
1531     }
1532 
1533     return FinishDecoding(verify_functions);
1534   }
1535 
1536   // Decodes a single anonymous function starting at {start_}.
DecodeSingleFunction(Zone * zone,const ModuleWireBytes & wire_bytes,const WasmModule * module,std::unique_ptr<WasmFunction> function)1537   FunctionResult DecodeSingleFunction(Zone* zone,
1538                                       const ModuleWireBytes& wire_bytes,
1539                                       const WasmModule* module,
1540                                       std::unique_ptr<WasmFunction> function) {
1541     pc_ = start_;
1542     expect_u8("type form", kWasmFunctionTypeCode);
1543     if (!ok()) return FunctionResult{std::move(intermediate_error_)};
1544     function->sig = consume_sig(zone);
1545     function->code = {off(pc_), static_cast<uint32_t>(end_ - pc_)};
1546 
1547     if (ok())
1548       VerifyFunctionBody(zone->allocator(), 0, wire_bytes, module,
1549                          function.get());
1550 
1551     if (intermediate_error_.has_error()) {
1552       return FunctionResult{std::move(intermediate_error_)};
1553     }
1554 
1555     return FunctionResult(std::move(function));
1556   }
1557 
1558   // Decodes a single function signature at {start}.
DecodeFunctionSignature(Zone * zone,const byte * start)1559   const FunctionSig* DecodeFunctionSignature(Zone* zone, const byte* start) {
1560     pc_ = start;
1561     if (!expect_u8("type form", kWasmFunctionTypeCode)) return nullptr;
1562     const FunctionSig* result = consume_sig(zone);
1563     return ok() ? result : nullptr;
1564   }
1565 
DecodeInitExprForTesting(ValueType expected)1566   ConstantExpression DecodeInitExprForTesting(ValueType expected) {
1567     return consume_init_expr(module_.get(), expected);
1568   }
1569 
shared_module() const1570   const std::shared_ptr<WasmModule>& shared_module() const { return module_; }
1571 
GetCounters() const1572   Counters* GetCounters() const {
1573     DCHECK_NOT_NULL(counters_);
1574     return counters_;
1575   }
1576 
SetCounters(Counters * counters)1577   void SetCounters(Counters* counters) {
1578     DCHECK_NULL(counters_);
1579     counters_ = counters;
1580   }
1581 
1582  private:
1583   const WasmFeatures enabled_features_;
1584   std::shared_ptr<WasmModule> module_;
1585   const byte* module_start_ = nullptr;
1586   const byte* module_end_ = nullptr;
1587   Counters* counters_ = nullptr;
1588   // The type section is the first section in a module.
1589   uint8_t next_ordered_section_ = kFirstSectionInModule;
1590   // We store next_ordered_section_ as uint8_t instead of SectionCode so that
1591   // we can increment it. This static_assert should make sure that SectionCode
1592   // does not get bigger than uint8_t accidentially.
1593   static_assert(sizeof(ModuleDecoderImpl::next_ordered_section_) ==
1594                     sizeof(SectionCode),
1595                 "type mismatch");
1596   uint32_t seen_unordered_sections_ = 0;
1597   static_assert(kBitsPerByte *
1598                         sizeof(ModuleDecoderImpl::seen_unordered_sections_) >
1599                     kLastKnownModuleSection,
1600                 "not enough bits");
1601   WasmError intermediate_error_;
1602   ModuleOrigin origin_;
1603   AccountingAllocator allocator_;
1604   Zone init_expr_zone_{&allocator_, "initializer expression zone"};
1605 
has_seen_unordered_section(SectionCode section_code)1606   bool has_seen_unordered_section(SectionCode section_code) {
1607     return seen_unordered_sections_ & (1 << section_code);
1608   }
1609 
set_seen_unordered_section(SectionCode section_code)1610   void set_seen_unordered_section(SectionCode section_code) {
1611     seen_unordered_sections_ |= 1 << section_code;
1612   }
1613 
off(const byte * ptr)1614   uint32_t off(const byte* ptr) {
1615     return static_cast<uint32_t>(ptr - start_) + buffer_offset_;
1616   }
1617 
AddMemory(WasmModule * module)1618   bool AddMemory(WasmModule* module) {
1619     if (module->has_memory) {
1620       error("At most one memory is supported");
1621       return false;
1622     } else {
1623       module->has_memory = true;
1624       return true;
1625     }
1626   }
1627 
1628   // Calculate individual global offsets and total size of globals table. This
1629   // function should be called after all globals have been defined, which is
1630   // after the import section and the global section, but before the global
1631   // offsets are accessed, e.g. by the function compilers. The moment when this
1632   // function should be called is not well-defined, as the global section may
1633   // not exist. Therefore this function is called multiple times.
CalculateGlobalOffsets(WasmModule * module)1634   void CalculateGlobalOffsets(WasmModule* module) {
1635     if (module->globals.empty() || module->untagged_globals_buffer_size != 0 ||
1636         module->tagged_globals_buffer_size != 0) {
1637       // This function has already been executed before, so we don't have to
1638       // execute it again.
1639       return;
1640     }
1641     uint32_t untagged_offset = 0;
1642     uint32_t tagged_offset = 0;
1643     uint32_t num_imported_mutable_globals = 0;
1644     for (WasmGlobal& global : module->globals) {
1645       if (global.mutability && global.imported) {
1646         global.index = num_imported_mutable_globals++;
1647       } else if (global.type.is_reference()) {
1648         global.offset = tagged_offset;
1649         // All entries in the tagged_globals_buffer have size 1.
1650         tagged_offset++;
1651       } else {
1652         int size = global.type.value_kind_size();
1653         untagged_offset = (untagged_offset + size - 1) & ~(size - 1);  // align
1654         global.offset = untagged_offset;
1655         untagged_offset += size;
1656       }
1657     }
1658     module->untagged_globals_buffer_size = untagged_offset;
1659     module->tagged_globals_buffer_size = tagged_offset;
1660   }
1661 
1662   // Verifies the body (code) of a given function.
VerifyFunctionBody(AccountingAllocator * allocator,uint32_t func_num,const ModuleWireBytes & wire_bytes,const WasmModule * module,WasmFunction * function)1663   void VerifyFunctionBody(AccountingAllocator* allocator, uint32_t func_num,
1664                           const ModuleWireBytes& wire_bytes,
1665                           const WasmModule* module, WasmFunction* function) {
1666     WasmFunctionName func_name(function,
1667                                wire_bytes.GetNameOrNull(function, module));
1668     if (FLAG_trace_wasm_decoder) {
1669       StdoutStream{} << "Verifying wasm function " << func_name << std::endl;
1670     }
1671     FunctionBody body = {
1672         function->sig, function->code.offset(),
1673         start_ + GetBufferRelativeOffset(function->code.offset()),
1674         start_ + GetBufferRelativeOffset(function->code.end_offset())};
1675 
1676     WasmFeatures unused_detected_features = WasmFeatures::None();
1677     DecodeResult result = VerifyWasmCode(allocator, enabled_features_, module,
1678                                          &unused_detected_features, body);
1679 
1680     // If the decode failed and this is the first error, set error code and
1681     // location.
1682     if (result.failed() && intermediate_error_.empty()) {
1683       // Wrap the error message from the function decoder.
1684       std::ostringstream error_msg;
1685       error_msg << "in function " << func_name << ": "
1686                 << result.error().message();
1687       intermediate_error_ = WasmError{result.error().offset(), error_msg.str()};
1688     }
1689   }
1690 
consume_sig_index(WasmModule * module,const FunctionSig ** sig)1691   uint32_t consume_sig_index(WasmModule* module, const FunctionSig** sig) {
1692     const byte* pos = pc_;
1693     uint32_t sig_index = consume_u32v("signature index");
1694     if (!module->has_signature(sig_index)) {
1695       errorf(pos, "signature index %u out of bounds (%d signatures)", sig_index,
1696              static_cast<int>(module->types.size()));
1697       *sig = nullptr;
1698       return 0;
1699     }
1700     *sig = module->signature(sig_index);
1701     return sig_index;
1702   }
1703 
consume_tag_sig_index(WasmModule * module,const FunctionSig ** sig)1704   uint32_t consume_tag_sig_index(WasmModule* module, const FunctionSig** sig) {
1705     const byte* pos = pc_;
1706     uint32_t sig_index = consume_sig_index(module, sig);
1707     if (*sig && (*sig)->return_count() != 0) {
1708       errorf(pos, "tag signature %u has non-void return", sig_index);
1709       *sig = nullptr;
1710       return 0;
1711     }
1712     return sig_index;
1713   }
1714 
consume_count(const char * name,size_t maximum)1715   uint32_t consume_count(const char* name, size_t maximum) {
1716     const byte* p = pc_;
1717     uint32_t count = consume_u32v(name);
1718     if (count > maximum) {
1719       errorf(p, "%s of %u exceeds internal limit of %zu", name, count, maximum);
1720       return static_cast<uint32_t>(maximum);
1721     }
1722     return count;
1723   }
1724 
consume_func_index(WasmModule * module,WasmFunction ** func,const char * name)1725   uint32_t consume_func_index(WasmModule* module, WasmFunction** func,
1726                               const char* name) {
1727     return consume_index(name, &module->functions, func);
1728   }
1729 
consume_global_index(WasmModule * module,WasmGlobal ** global)1730   uint32_t consume_global_index(WasmModule* module, WasmGlobal** global) {
1731     return consume_index("global index", &module->globals, global);
1732   }
1733 
consume_table_index(WasmModule * module,WasmTable ** table)1734   uint32_t consume_table_index(WasmModule* module, WasmTable** table) {
1735     return consume_index("table index", &module->tables, table);
1736   }
1737 
consume_tag_index(WasmModule * module,WasmTag ** tag)1738   uint32_t consume_tag_index(WasmModule* module, WasmTag** tag) {
1739     return consume_index("tag index", &module->tags, tag);
1740   }
1741 
1742   template <typename T>
consume_index(const char * name,std::vector<T> * vector,T ** ptr)1743   uint32_t consume_index(const char* name, std::vector<T>* vector, T** ptr) {
1744     const byte* pos = pc_;
1745     uint32_t index = consume_u32v(name);
1746     if (index >= vector->size()) {
1747       errorf(pos, "%s %u out of bounds (%d entr%s)", name, index,
1748              static_cast<int>(vector->size()),
1749              vector->size() == 1 ? "y" : "ies");
1750       *ptr = nullptr;
1751       return 0;
1752     }
1753     *ptr = &(*vector)[index];
1754     return index;
1755   }
1756 
validate_table_flags(const char * name)1757   uint8_t validate_table_flags(const char* name) {
1758     uint8_t flags = consume_u8("table limits flags");
1759     STATIC_ASSERT(kNoMaximum < kWithMaximum);
1760     if (V8_UNLIKELY(flags > kWithMaximum)) {
1761       errorf(pc() - 1, "invalid %s limits flags", name);
1762     }
1763     return flags;
1764   }
1765 
validate_memory_flags(bool * has_shared_memory,bool * is_memory64)1766   uint8_t validate_memory_flags(bool* has_shared_memory, bool* is_memory64) {
1767     uint8_t flags = consume_u8("memory limits flags");
1768     *has_shared_memory = false;
1769     switch (flags) {
1770       case kNoMaximum:
1771       case kWithMaximum:
1772         break;
1773       case kSharedNoMaximum:
1774       case kSharedWithMaximum:
1775         if (!enabled_features_.has_threads()) {
1776           errorf(pc() - 1,
1777                  "invalid memory limits flags 0x%x (enable via "
1778                  "--experimental-wasm-threads)",
1779                  flags);
1780         }
1781         *has_shared_memory = true;
1782         // V8 does not support shared memory without a maximum.
1783         if (flags == kSharedNoMaximum) {
1784           errorf(pc() - 1,
1785                  "memory limits flags must have maximum defined if shared is "
1786                  "true");
1787         }
1788         break;
1789       case kMemory64NoMaximum:
1790       case kMemory64WithMaximum:
1791         if (!enabled_features_.has_memory64()) {
1792           errorf(pc() - 1,
1793                  "invalid memory limits flags 0x%x (enable via "
1794                  "--experimental-wasm-memory64)",
1795                  flags);
1796         }
1797         *is_memory64 = true;
1798         break;
1799       default:
1800         errorf(pc() - 1, "invalid memory limits flags 0x%x", flags);
1801         break;
1802     }
1803     return flags;
1804   }
1805 
consume_resizable_limits(const char * name,const char * units,uint32_t max_initial,uint32_t * initial,bool * has_max,uint32_t max_maximum,uint32_t * maximum,uint8_t flags)1806   void consume_resizable_limits(const char* name, const char* units,
1807                                 uint32_t max_initial, uint32_t* initial,
1808                                 bool* has_max, uint32_t max_maximum,
1809                                 uint32_t* maximum, uint8_t flags) {
1810     const byte* pos = pc();
1811     // For memory64 we need to read the numbers as LEB-encoded 64-bit unsigned
1812     // integer. All V8 limits are still within uint32_t range though.
1813     const bool is_memory64 =
1814         flags == kMemory64NoMaximum || flags == kMemory64WithMaximum;
1815     uint64_t initial_64 = is_memory64 ? consume_u64v("initial size")
1816                                       : consume_u32v("initial size");
1817     if (initial_64 > max_initial) {
1818       errorf(pos,
1819              "initial %s size (%" PRIu64
1820              " %s) is larger than implementation limit (%u)",
1821              name, initial_64, units, max_initial);
1822     }
1823     *initial = static_cast<uint32_t>(initial_64);
1824     if (flags & 1) {
1825       *has_max = true;
1826       pos = pc();
1827       uint64_t maximum_64 = is_memory64 ? consume_u64v("maximum size")
1828                                         : consume_u32v("maximum size");
1829       if (maximum_64 > max_maximum) {
1830         errorf(pos,
1831                "maximum %s size (%" PRIu64
1832                " %s) is larger than implementation limit (%u)",
1833                name, maximum_64, units, max_maximum);
1834       }
1835       if (maximum_64 < *initial) {
1836         errorf(pos,
1837                "maximum %s size (%" PRIu64 " %s) is less than initial (%u %s)",
1838                name, maximum_64, units, *initial, units);
1839       }
1840       *maximum = static_cast<uint32_t>(maximum_64);
1841     } else {
1842       *has_max = false;
1843       *maximum = max_initial;
1844     }
1845   }
1846 
1847   // Consumes a byte, and emits an error if it does not equal {expected}.
expect_u8(const char * name,uint8_t expected)1848   bool expect_u8(const char* name, uint8_t expected) {
1849     const byte* pos = pc();
1850     uint8_t value = consume_u8(name);
1851     if (value != expected) {
1852       errorf(pos, "expected %s 0x%02x, got 0x%02x", name, expected, value);
1853       return false;
1854     }
1855     return true;
1856   }
1857 
consume_init_expr(WasmModule * module,ValueType expected)1858   ConstantExpression consume_init_expr(WasmModule* module, ValueType expected) {
1859     uint32_t length;
1860 
1861     // The error message mimics the one generated by the {WasmFullDecoder}.
1862 #define TYPE_CHECK(found)                                             \
1863   if (V8_UNLIKELY(!IsSubtypeOf(found, expected, module_.get()))) {    \
1864     errorf(pc() + 1,                                                  \
1865            "type error in init. expression[0] (expected %s, got %s)", \
1866            expected.name().c_str(), found.name().c_str());            \
1867     return {};                                                        \
1868   }
1869 
1870     // To avoid initializing a {WasmFullDecoder} for the most common
1871     // expressions, we replicate their decoding and validation here. The
1872     // manually handled cases correspond to {ConstantExpression}'s kinds.
1873     // We need to make sure to check that the expression ends in {kExprEnd};
1874     // otherwise, it is just the first operand of a composite expression, and we
1875     // fall back to the default case.
1876     if (!more()) {
1877       error("Beyond end of code");
1878       return {};
1879     }
1880     switch (static_cast<WasmOpcode>(*pc())) {
1881       case kExprI32Const: {
1882         int32_t value =
1883             read_i32v<kFullValidation>(pc() + 1, &length, "i32.const");
1884         if (V8_UNLIKELY(failed())) return {};
1885         if (V8_LIKELY(lookahead(1 + length, kExprEnd))) {
1886           TYPE_CHECK(kWasmI32)
1887           consume_bytes(length + 2);
1888           return ConstantExpression::I32Const(value);
1889         }
1890         break;
1891       }
1892       case kExprRefFunc: {
1893         uint32_t index =
1894             read_u32v<kFullValidation>(pc() + 1, &length, "ref.func");
1895         if (V8_UNLIKELY(failed())) return {};
1896         if (V8_LIKELY(lookahead(1 + length, kExprEnd))) {
1897           if (V8_UNLIKELY(index >= module_->functions.size())) {
1898             errorf(pc() + 1, "function index %u out of bounds", index);
1899             return {};
1900           }
1901           ValueType type =
1902               enabled_features_.has_typed_funcref()
1903                   ? ValueType::Ref(module_->functions[index].sig_index,
1904                                    kNonNullable)
1905                   : kWasmFuncRef;
1906           TYPE_CHECK(type)
1907           module_->functions[index].declared = true;
1908           consume_bytes(length + 2);
1909           return ConstantExpression::RefFunc(index);
1910         }
1911         break;
1912       }
1913       case kExprRefNull: {
1914         HeapType type = value_type_reader::read_heap_type<kFullValidation>(
1915             this, pc() + 1, &length, module_.get(), enabled_features_);
1916         if (V8_UNLIKELY(failed())) return {};
1917         if (V8_LIKELY(lookahead(1 + length, kExprEnd))) {
1918           TYPE_CHECK(ValueType::Ref(type, kNullable))
1919           consume_bytes(length + 2);
1920           return ConstantExpression::RefNull(type.representation());
1921         }
1922         break;
1923       }
1924       default:
1925         break;
1926     }
1927 #undef TYPE_CHECK
1928 
1929     auto sig = FixedSizeSignature<ValueType>::Returns(expected);
1930     FunctionBody body(&sig, buffer_offset_, pc_, end_);
1931     WasmFeatures detected;
1932     WasmFullDecoder<Decoder::kFullValidation, InitExprInterface,
1933                     kInitExpression>
1934         decoder(&init_expr_zone_, module, enabled_features_, &detected, body,
1935                 module);
1936 
1937     uint32_t offset = this->pc_offset();
1938 
1939     decoder.DecodeFunctionBody();
1940 
1941     this->pc_ = decoder.end();
1942 
1943     if (decoder.failed()) {
1944       error(decoder.error().offset(), decoder.error().message().c_str());
1945       return {};
1946     }
1947 
1948     if (!decoder.interface().end_found()) {
1949       error("Initializer expression is missing 'end'");
1950       return {};
1951     }
1952 
1953     return ConstantExpression::WireBytes(
1954         offset, static_cast<uint32_t>(decoder.end() - decoder.start()));
1955   }
1956 
1957   // Read a mutability flag
consume_mutability()1958   bool consume_mutability() {
1959     byte val = consume_u8("mutability");
1960     if (val > 1) error(pc_ - 1, "invalid mutability");
1961     return val != 0;
1962   }
1963 
consume_value_type()1964   ValueType consume_value_type() {
1965     uint32_t type_length;
1966     ValueType result = value_type_reader::read_value_type<kFullValidation>(
1967         this, this->pc(), &type_length, module_.get(),
1968         origin_ == kWasmOrigin ? enabled_features_ : WasmFeatures::None());
1969     consume_bytes(type_length, "value type");
1970     return result;
1971   }
1972 
consume_super_type()1973   HeapType consume_super_type() {
1974     return value_type_reader::consume_heap_type(this, module_.get(),
1975                                                 enabled_features_);
1976   }
1977 
consume_storage_type()1978   ValueType consume_storage_type() {
1979     uint8_t opcode = read_u8<kFullValidation>(this->pc());
1980     switch (opcode) {
1981       case kI8Code:
1982         consume_bytes(1, "i8");
1983         return kWasmI8;
1984       case kI16Code:
1985         consume_bytes(1, "i16");
1986         return kWasmI16;
1987       default:
1988         // It is not a packed type, so it has to be a value type.
1989         return consume_value_type();
1990     }
1991   }
1992 
1993   // Reads a reference type for tables and element segment headers.
consume_reference_type()1994   ValueType consume_reference_type() {
1995     const byte* position = pc();
1996     ValueType result = consume_value_type();
1997     if (!result.is_reference()) {
1998       error(position, "expected reference type");
1999     }
2000     return result;
2001   }
2002 
consume_sig(Zone * zone)2003   const FunctionSig* consume_sig(Zone* zone) {
2004     // Parse parameter types.
2005     uint32_t param_count =
2006         consume_count("param count", kV8MaxWasmFunctionParams);
2007     if (failed()) return nullptr;
2008     std::vector<ValueType> params;
2009     for (uint32_t i = 0; ok() && i < param_count; ++i) {
2010       params.push_back(consume_value_type());
2011     }
2012     std::vector<ValueType> returns;
2013 
2014     // Parse return types.
2015     uint32_t return_count =
2016         consume_count("return count", kV8MaxWasmFunctionReturns);
2017     if (failed()) return nullptr;
2018     for (uint32_t i = 0; ok() && i < return_count; ++i) {
2019       returns.push_back(consume_value_type());
2020     }
2021     if (failed()) return nullptr;
2022 
2023     // FunctionSig stores the return types first.
2024     ValueType* buffer = zone->NewArray<ValueType>(param_count + return_count);
2025     uint32_t b = 0;
2026     for (uint32_t i = 0; i < return_count; ++i) buffer[b++] = returns[i];
2027     for (uint32_t i = 0; i < param_count; ++i) buffer[b++] = params[i];
2028 
2029     return zone->New<FunctionSig>(return_count, param_count, buffer);
2030   }
2031 
consume_struct(Zone * zone)2032   const StructType* consume_struct(Zone* zone) {
2033     uint32_t field_count = consume_count("field count", kV8MaxWasmStructFields);
2034     if (failed()) return nullptr;
2035     ValueType* fields = zone->NewArray<ValueType>(field_count);
2036     bool* mutabilities = zone->NewArray<bool>(field_count);
2037     for (uint32_t i = 0; ok() && i < field_count; ++i) {
2038       fields[i] = consume_storage_type();
2039       mutabilities[i] = consume_mutability();
2040     }
2041     if (failed()) return nullptr;
2042     uint32_t* offsets = zone->NewArray<uint32_t>(field_count);
2043     return zone->New<StructType>(field_count, offsets, fields, mutabilities);
2044   }
2045 
consume_array(Zone * zone)2046   const ArrayType* consume_array(Zone* zone) {
2047     ValueType element_type = consume_storage_type();
2048     bool mutability = consume_mutability();
2049     if (failed()) return nullptr;
2050     return zone->New<ArrayType>(element_type, mutability);
2051   }
2052 
2053   // Consume the attribute field of an exception.
consume_exception_attribute()2054   uint32_t consume_exception_attribute() {
2055     const byte* pos = pc_;
2056     uint32_t attribute = consume_u32v("exception attribute");
2057     if (attribute != kExceptionAttribute) {
2058       errorf(pos, "exception attribute %u not supported", attribute);
2059       return 0;
2060     }
2061     return attribute;
2062   }
2063 
consume_element_segment_header()2064   WasmElemSegment consume_element_segment_header() {
2065     const byte* pos = pc();
2066 
2067     // The mask for the bit in the flag which indicates if the segment is
2068     // active or not (0 is active).
2069     constexpr uint8_t kNonActiveMask = 1 << 0;
2070     // The mask for the bit in the flag which indicates:
2071     // - for active tables, if the segment has an explicit table index field.
2072     // - for non-active tables, whether the table is declarative (vs. passive).
2073     constexpr uint8_t kHasTableIndexOrIsDeclarativeMask = 1 << 1;
2074     // The mask for the bit in the flag which indicates if the functions of this
2075     // segment are defined as function indices (0) or init. expressions (1).
2076     constexpr uint8_t kExpressionsAsElementsMask = 1 << 2;
2077     constexpr uint8_t kFullMask = kNonActiveMask |
2078                                   kHasTableIndexOrIsDeclarativeMask |
2079                                   kExpressionsAsElementsMask;
2080 
2081     uint32_t flag = consume_u32v("flag");
2082     if ((flag & kFullMask) != flag) {
2083       errorf(pos, "illegal flag value %u. Must be between 0 and 7", flag);
2084       return {};
2085     }
2086 
2087     const WasmElemSegment::Status status =
2088         (flag & kNonActiveMask) ? (flag & kHasTableIndexOrIsDeclarativeMask)
2089                                       ? WasmElemSegment::kStatusDeclarative
2090                                       : WasmElemSegment::kStatusPassive
2091                                 : WasmElemSegment::kStatusActive;
2092     const bool is_active = status == WasmElemSegment::kStatusActive;
2093 
2094     WasmElemSegment::ElementType element_type =
2095         flag & kExpressionsAsElementsMask
2096             ? WasmElemSegment::kExpressionElements
2097             : WasmElemSegment::kFunctionIndexElements;
2098 
2099     const bool has_table_index =
2100         is_active && (flag & kHasTableIndexOrIsDeclarativeMask);
2101     uint32_t table_index = has_table_index ? consume_u32v("table index") : 0;
2102     if (is_active && table_index >= module_->tables.size()) {
2103       errorf(pos, "out of bounds%s table index %u",
2104              has_table_index ? " implicit" : "", table_index);
2105       return {};
2106     }
2107     ValueType table_type =
2108         is_active ? module_->tables[table_index].type : kWasmBottom;
2109 
2110     ConstantExpression offset;
2111     if (is_active) {
2112       offset = consume_init_expr(module_.get(), kWasmI32);
2113       // Failed to parse offset initializer, return early.
2114       if (failed()) return {};
2115     }
2116 
2117     // Denotes an active segment without table index, type, or element kind.
2118     const bool backwards_compatible_mode =
2119         is_active && !(flag & kHasTableIndexOrIsDeclarativeMask);
2120     ValueType type;
2121     if (element_type == WasmElemSegment::kExpressionElements) {
2122       type =
2123           backwards_compatible_mode ? kWasmFuncRef : consume_reference_type();
2124       if (is_active && !IsSubtypeOf(type, table_type, this->module_.get())) {
2125         errorf(pos,
2126                "Element segment of type %s is not a subtype of referenced "
2127                "table %u (of type %s)",
2128                type.name().c_str(), table_index, table_type.name().c_str());
2129         return {};
2130       }
2131     } else {
2132       if (!backwards_compatible_mode) {
2133         // We have to check that there is an element kind of type Function. All
2134         // other element kinds are not valid yet.
2135         uint8_t val = consume_u8("element kind");
2136         if (static_cast<ImportExportKindCode>(val) != kExternalFunction) {
2137           errorf(pos, "illegal element kind 0x%x. Must be 0x%x", val,
2138                  kExternalFunction);
2139           return {};
2140         }
2141       }
2142       if (!is_active) {
2143         // Declarative and passive segments without explicit type are funcref.
2144         type = kWasmFuncRef;
2145       } else {
2146         type = table_type;
2147         // Active segments with function indices must reference a function
2148         // table. TODO(7748): Add support for anyref tables when we have them.
2149         if (!IsSubtypeOf(table_type, kWasmFuncRef, this->module_.get())) {
2150           errorf(pos,
2151                  "An active element segment with function indices as elements "
2152                  "must reference a table of %s. Instead, table %u of type %s "
2153                  "is referenced.",
2154                  enabled_features_.has_typed_funcref()
2155                      ? "a subtype of type funcref"
2156                      : "type funcref",
2157                  table_index, table_type.name().c_str());
2158           return {};
2159         }
2160       }
2161     }
2162 
2163     if (is_active) {
2164       return {type, table_index, std::move(offset), element_type};
2165     } else {
2166       return {type, status, element_type};
2167     }
2168   }
2169 
consume_data_segment_header(bool * is_active,uint32_t * index,ConstantExpression * offset)2170   void consume_data_segment_header(bool* is_active, uint32_t* index,
2171                                    ConstantExpression* offset) {
2172     const byte* pos = pc();
2173     uint32_t flag = consume_u32v("flag");
2174 
2175     // Some flag values are only valid for specific proposals.
2176     if (flag != SegmentFlags::kActiveNoIndex &&
2177         flag != SegmentFlags::kPassive &&
2178         flag != SegmentFlags::kActiveWithIndex) {
2179       errorf(pos, "illegal flag value %u. Must be 0, 1, or 2", flag);
2180       return;
2181     }
2182 
2183     // We know now that the flag is valid. Time to read the rest.
2184     ValueType expected_type = module_->is_memory64 ? kWasmI64 : kWasmI32;
2185     if (flag == SegmentFlags::kActiveNoIndex) {
2186       *is_active = true;
2187       *index = 0;
2188       *offset = consume_init_expr(module_.get(), expected_type);
2189       return;
2190     }
2191     if (flag == SegmentFlags::kPassive) {
2192       *is_active = false;
2193       return;
2194     }
2195     if (flag == SegmentFlags::kActiveWithIndex) {
2196       *is_active = true;
2197       *index = consume_u32v("memory index");
2198       *offset = consume_init_expr(module_.get(), expected_type);
2199     }
2200   }
2201 
consume_element_func_index(ValueType expected)2202   uint32_t consume_element_func_index(ValueType expected) {
2203     WasmFunction* func = nullptr;
2204     const byte* initial_pc = pc();
2205     uint32_t index =
2206         consume_func_index(module_.get(), &func, "element function index");
2207     if (failed()) return index;
2208     DCHECK_NOT_NULL(func);
2209     DCHECK_EQ(index, func->func_index);
2210     ValueType entry_type = ValueType::Ref(func->sig_index, kNonNullable);
2211     if (V8_UNLIKELY(!IsSubtypeOf(entry_type, expected, module_.get()))) {
2212       errorf(initial_pc,
2213              "Invalid type in element entry: expected %s, got %s instead.",
2214              expected.name().c_str(), entry_type.name().c_str());
2215       return index;
2216     }
2217     func->declared = true;
2218     return index;
2219   }
2220 };
2221 
DecodeWasmModule(const WasmFeatures & enabled,const byte * module_start,const byte * module_end,bool verify_functions,ModuleOrigin origin,Counters * counters,std::shared_ptr<metrics::Recorder> metrics_recorder,v8::metrics::Recorder::ContextId context_id,DecodingMethod decoding_method,AccountingAllocator * allocator)2222 ModuleResult DecodeWasmModule(
2223     const WasmFeatures& enabled, const byte* module_start,
2224     const byte* module_end, bool verify_functions, ModuleOrigin origin,
2225     Counters* counters, std::shared_ptr<metrics::Recorder> metrics_recorder,
2226     v8::metrics::Recorder::ContextId context_id, DecodingMethod decoding_method,
2227     AccountingAllocator* allocator) {
2228   size_t size = module_end - module_start;
2229   CHECK_LE(module_start, module_end);
2230   size_t max_size = max_module_size();
2231   if (size > max_size) {
2232     return ModuleResult{
2233         WasmError{0, "size > maximum module size (%zu): %zu", max_size, size}};
2234   }
2235   // TODO(bradnelson): Improve histogram handling of size_t.
2236   auto size_counter =
2237       SELECT_WASM_COUNTER(counters, origin, wasm, module_size_bytes);
2238   size_counter->AddSample(static_cast<int>(size));
2239   // Signatures are stored in zone memory, which have the same lifetime
2240   // as the {module}.
2241   ModuleDecoderImpl decoder(enabled, module_start, module_end, origin);
2242   v8::metrics::WasmModuleDecoded metrics_event;
2243   base::ElapsedTimer timer;
2244   timer.Start();
2245   base::ThreadTicks thread_ticks = base::ThreadTicks::IsSupported()
2246                                        ? base::ThreadTicks::Now()
2247                                        : base::ThreadTicks();
2248   ModuleResult result =
2249       decoder.DecodeModule(counters, allocator, verify_functions);
2250 
2251   // Record event metrics.
2252   metrics_event.wall_clock_duration_in_us = timer.Elapsed().InMicroseconds();
2253   timer.Stop();
2254   if (!thread_ticks.IsNull()) {
2255     metrics_event.cpu_duration_in_us =
2256         (base::ThreadTicks::Now() - thread_ticks).InMicroseconds();
2257   }
2258   metrics_event.success = decoder.ok() && result.ok();
2259   metrics_event.async = decoding_method == DecodingMethod::kAsync ||
2260                         decoding_method == DecodingMethod::kAsyncStream;
2261   metrics_event.streamed = decoding_method == DecodingMethod::kSyncStream ||
2262                            decoding_method == DecodingMethod::kAsyncStream;
2263   if (result.ok()) {
2264     metrics_event.function_count = result.value()->num_declared_functions;
2265   } else if (auto&& module = decoder.shared_module()) {
2266     metrics_event.function_count = module->num_declared_functions;
2267   }
2268   metrics_event.module_size_in_bytes = size;
2269   metrics_recorder->DelayMainThreadEvent(metrics_event, context_id);
2270 
2271   return result;
2272 }
2273 
ModuleDecoder(const WasmFeatures & enabled)2274 ModuleDecoder::ModuleDecoder(const WasmFeatures& enabled)
2275     : enabled_features_(enabled) {}
2276 
2277 ModuleDecoder::~ModuleDecoder() = default;
2278 
shared_module() const2279 const std::shared_ptr<WasmModule>& ModuleDecoder::shared_module() const {
2280   return impl_->shared_module();
2281 }
2282 
StartDecoding(Counters * counters,std::shared_ptr<metrics::Recorder> metrics_recorder,v8::metrics::Recorder::ContextId context_id,AccountingAllocator * allocator,ModuleOrigin origin)2283 void ModuleDecoder::StartDecoding(
2284     Counters* counters, std::shared_ptr<metrics::Recorder> metrics_recorder,
2285     v8::metrics::Recorder::ContextId context_id, AccountingAllocator* allocator,
2286     ModuleOrigin origin) {
2287   DCHECK_NULL(impl_);
2288   impl_.reset(new ModuleDecoderImpl(enabled_features_, origin));
2289   impl_->StartDecoding(counters, allocator);
2290 }
2291 
DecodeModuleHeader(base::Vector<const uint8_t> bytes,uint32_t offset)2292 void ModuleDecoder::DecodeModuleHeader(base::Vector<const uint8_t> bytes,
2293                                        uint32_t offset) {
2294   impl_->DecodeModuleHeader(bytes, offset);
2295 }
2296 
DecodeSection(SectionCode section_code,base::Vector<const uint8_t> bytes,uint32_t offset,bool verify_functions)2297 void ModuleDecoder::DecodeSection(SectionCode section_code,
2298                                   base::Vector<const uint8_t> bytes,
2299                                   uint32_t offset, bool verify_functions) {
2300   impl_->DecodeSection(section_code, bytes, offset, verify_functions);
2301 }
2302 
DecodeFunctionBody(uint32_t index,uint32_t length,uint32_t offset,bool verify_functions)2303 void ModuleDecoder::DecodeFunctionBody(uint32_t index, uint32_t length,
2304                                        uint32_t offset, bool verify_functions) {
2305   impl_->DecodeFunctionBody(index, length, offset, verify_functions);
2306 }
2307 
StartCodeSection()2308 void ModuleDecoder::StartCodeSection() { impl_->StartCodeSection(); }
2309 
CheckFunctionsCount(uint32_t functions_count,uint32_t error_offset)2310 bool ModuleDecoder::CheckFunctionsCount(uint32_t functions_count,
2311                                         uint32_t error_offset) {
2312   return impl_->CheckFunctionsCount(functions_count, error_offset);
2313 }
2314 
FinishDecoding(bool verify_functions)2315 ModuleResult ModuleDecoder::FinishDecoding(bool verify_functions) {
2316   return impl_->FinishDecoding(verify_functions);
2317 }
2318 
set_code_section(uint32_t offset,uint32_t size)2319 void ModuleDecoder::set_code_section(uint32_t offset, uint32_t size) {
2320   return impl_->set_code_section(offset, size);
2321 }
2322 
IdentifyUnknownSection(ModuleDecoder * decoder,base::Vector<const uint8_t> bytes,uint32_t offset,SectionCode * result)2323 size_t ModuleDecoder::IdentifyUnknownSection(ModuleDecoder* decoder,
2324                                              base::Vector<const uint8_t> bytes,
2325                                              uint32_t offset,
2326                                              SectionCode* result) {
2327   if (!decoder->ok()) return 0;
2328   decoder->impl_->Reset(bytes, offset);
2329   *result = IdentifyUnknownSectionInternal(decoder->impl_.get());
2330   return decoder->impl_->pc() - bytes.begin();
2331 }
2332 
ok()2333 bool ModuleDecoder::ok() { return impl_->ok(); }
2334 
DecodeWasmSignatureForTesting(const WasmFeatures & enabled,Zone * zone,const byte * start,const byte * end)2335 const FunctionSig* DecodeWasmSignatureForTesting(const WasmFeatures& enabled,
2336                                                  Zone* zone, const byte* start,
2337                                                  const byte* end) {
2338   ModuleDecoderImpl decoder(enabled, start, end, kWasmOrigin);
2339   return decoder.DecodeFunctionSignature(zone, start);
2340 }
2341 
DecodeWasmInitExprForTesting(const WasmFeatures & enabled,const byte * start,const byte * end,ValueType expected)2342 ConstantExpression DecodeWasmInitExprForTesting(const WasmFeatures& enabled,
2343                                                 const byte* start,
2344                                                 const byte* end,
2345                                                 ValueType expected) {
2346   ModuleDecoderImpl decoder(enabled, start, end, kWasmOrigin);
2347   AccountingAllocator allocator;
2348   decoder.StartDecoding(nullptr, &allocator);
2349   return decoder.DecodeInitExprForTesting(expected);
2350 }
2351 
DecodeWasmFunctionForTesting(const WasmFeatures & enabled,Zone * zone,const ModuleWireBytes & wire_bytes,const WasmModule * module,const byte * function_start,const byte * function_end,Counters * counters)2352 FunctionResult DecodeWasmFunctionForTesting(
2353     const WasmFeatures& enabled, Zone* zone, const ModuleWireBytes& wire_bytes,
2354     const WasmModule* module, const byte* function_start,
2355     const byte* function_end, Counters* counters) {
2356   size_t size = function_end - function_start;
2357   CHECK_LE(function_start, function_end);
2358   if (size > kV8MaxWasmFunctionSize) {
2359     return FunctionResult{WasmError{0,
2360                                     "size > maximum function size (%zu): %zu",
2361                                     kV8MaxWasmFunctionSize, size}};
2362   }
2363   ModuleDecoderImpl decoder(enabled, function_start, function_end, kWasmOrigin);
2364   decoder.SetCounters(counters);
2365   return decoder.DecodeSingleFunction(zone, wire_bytes, module,
2366                                       std::make_unique<WasmFunction>());
2367 }
2368 
DecodeAsmJsOffsets(base::Vector<const uint8_t> encoded_offsets)2369 AsmJsOffsetsResult DecodeAsmJsOffsets(
2370     base::Vector<const uint8_t> encoded_offsets) {
2371   std::vector<AsmJsOffsetFunctionEntries> functions;
2372 
2373   Decoder decoder(encoded_offsets);
2374   uint32_t functions_count = decoder.consume_u32v("functions count");
2375   // Consistency check.
2376   DCHECK_GE(encoded_offsets.size(), functions_count);
2377   functions.reserve(functions_count);
2378 
2379   for (uint32_t i = 0; i < functions_count; ++i) {
2380     uint32_t size = decoder.consume_u32v("table size");
2381     if (size == 0) {
2382       functions.emplace_back();
2383       continue;
2384     }
2385     DCHECK(decoder.checkAvailable(size));
2386     const byte* table_end = decoder.pc() + size;
2387     uint32_t locals_size = decoder.consume_u32v("locals size");
2388     int function_start_position = decoder.consume_u32v("function start pos");
2389     int function_end_position = function_start_position;
2390     int last_byte_offset = locals_size;
2391     int last_asm_position = function_start_position;
2392     std::vector<AsmJsOffsetEntry> func_asm_offsets;
2393     func_asm_offsets.reserve(size / 4);  // conservative estimation
2394     // Add an entry for the stack check, associated with position 0.
2395     func_asm_offsets.push_back(
2396         {0, function_start_position, function_start_position});
2397     while (decoder.pc() < table_end) {
2398       DCHECK(decoder.ok());
2399       last_byte_offset += decoder.consume_u32v("byte offset delta");
2400       int call_position =
2401           last_asm_position + decoder.consume_i32v("call position delta");
2402       int to_number_position =
2403           call_position + decoder.consume_i32v("to_number position delta");
2404       last_asm_position = to_number_position;
2405       if (decoder.pc() == table_end) {
2406         // The last entry is the function end marker.
2407         DCHECK_EQ(call_position, to_number_position);
2408         function_end_position = call_position;
2409       } else {
2410         func_asm_offsets.push_back(
2411             {last_byte_offset, call_position, to_number_position});
2412       }
2413     }
2414     DCHECK_EQ(decoder.pc(), table_end);
2415     functions.emplace_back(AsmJsOffsetFunctionEntries{
2416         function_start_position, function_end_position,
2417         std::move(func_asm_offsets)});
2418   }
2419   DCHECK(decoder.ok());
2420   DCHECK(!decoder.more());
2421 
2422   return decoder.toResult(AsmJsOffsets{std::move(functions)});
2423 }
2424 
DecodeCustomSections(const byte * start,const byte * end)2425 std::vector<CustomSectionOffset> DecodeCustomSections(const byte* start,
2426                                                       const byte* end) {
2427   Decoder decoder(start, end);
2428   decoder.consume_bytes(4, "wasm magic");
2429   decoder.consume_bytes(4, "wasm version");
2430 
2431   std::vector<CustomSectionOffset> result;
2432 
2433   while (decoder.more()) {
2434     byte section_code = decoder.consume_u8("section code");
2435     uint32_t section_length = decoder.consume_u32v("section length");
2436     uint32_t section_start = decoder.pc_offset();
2437     if (section_code != 0) {
2438       // Skip known sections.
2439       decoder.consume_bytes(section_length, "section bytes");
2440       continue;
2441     }
2442     uint32_t name_length = decoder.consume_u32v("name length");
2443     uint32_t name_offset = decoder.pc_offset();
2444     decoder.consume_bytes(name_length, "section name");
2445     uint32_t payload_offset = decoder.pc_offset();
2446     if (section_length < (payload_offset - section_start)) {
2447       decoder.error("invalid section length");
2448       break;
2449     }
2450     uint32_t payload_length = section_length - (payload_offset - section_start);
2451     decoder.consume_bytes(payload_length);
2452     if (decoder.failed()) break;
2453     result.push_back({{section_start, section_length},
2454                       {name_offset, name_length},
2455                       {payload_offset, payload_length}});
2456   }
2457 
2458   return result;
2459 }
2460 
2461 namespace {
2462 
FindNameSection(Decoder * decoder)2463 bool FindNameSection(Decoder* decoder) {
2464   static constexpr int kModuleHeaderSize = 8;
2465   decoder->consume_bytes(kModuleHeaderSize, "module header");
2466 
2467   WasmSectionIterator section_iter(decoder);
2468 
2469   while (decoder->ok() && section_iter.more() &&
2470          section_iter.section_code() != kNameSectionCode) {
2471     section_iter.advance(true);
2472   }
2473   if (!section_iter.more()) return false;
2474 
2475   // Reset the decoder to not read beyond the name section end.
2476   decoder->Reset(section_iter.payload(), decoder->pc_offset());
2477   return true;
2478 }
2479 
2480 }  // namespace
2481 
DecodeFunctionNames(const byte * module_start,const byte * module_end,std::unordered_map<uint32_t,WireBytesRef> * names)2482 void DecodeFunctionNames(const byte* module_start, const byte* module_end,
2483                          std::unordered_map<uint32_t, WireBytesRef>* names) {
2484   DCHECK_NOT_NULL(names);
2485   DCHECK(names->empty());
2486 
2487   Decoder decoder(module_start, module_end);
2488   if (FindNameSection(&decoder)) {
2489     while (decoder.ok() && decoder.more()) {
2490       uint8_t name_type = decoder.consume_u8("name type");
2491       if (name_type & 0x80) break;  // no varuint7
2492 
2493       uint32_t name_payload_len = decoder.consume_u32v("name payload length");
2494       if (!decoder.checkAvailable(name_payload_len)) break;
2495 
2496       if (name_type != NameSectionKindCode::kFunctionCode) {
2497         decoder.consume_bytes(name_payload_len, "name subsection payload");
2498         continue;
2499       }
2500       uint32_t functions_count = decoder.consume_u32v("functions count");
2501 
2502       for (; decoder.ok() && functions_count > 0; --functions_count) {
2503         uint32_t function_index = decoder.consume_u32v("function index");
2504         WireBytesRef name = consume_string(&decoder, false, "function name");
2505 
2506         // Be lenient with errors in the name section: Ignore non-UTF8 names.
2507         // You can even assign to the same function multiple times (last valid
2508         // one wins).
2509         if (decoder.ok() && validate_utf8(&decoder, name)) {
2510           names->insert(std::make_pair(function_index, name));
2511         }
2512       }
2513     }
2514   }
2515 }
2516 
DecodeNameMap(base::Vector<const uint8_t> module_bytes,uint8_t name_section_kind)2517 NameMap DecodeNameMap(base::Vector<const uint8_t> module_bytes,
2518                       uint8_t name_section_kind) {
2519   Decoder decoder(module_bytes);
2520   if (!FindNameSection(&decoder)) return NameMap{{}};
2521 
2522   std::vector<NameAssoc> names;
2523   while (decoder.ok() && decoder.more()) {
2524     uint8_t name_type = decoder.consume_u8("name type");
2525     if (name_type & 0x80) break;  // no varuint7
2526 
2527     uint32_t name_payload_len = decoder.consume_u32v("name payload length");
2528     if (!decoder.checkAvailable(name_payload_len)) break;
2529 
2530     if (name_type != name_section_kind) {
2531       decoder.consume_bytes(name_payload_len, "name subsection payload");
2532       continue;
2533     }
2534 
2535     uint32_t count = decoder.consume_u32v("names count");
2536     for (uint32_t i = 0; i < count; i++) {
2537       uint32_t index = decoder.consume_u32v("index");
2538       WireBytesRef name = consume_string(&decoder, false, "name");
2539       if (!decoder.ok()) break;
2540       if (index > kMaxInt) continue;
2541       if (!validate_utf8(&decoder, name)) continue;
2542       names.emplace_back(static_cast<int>(index), name);
2543     }
2544   }
2545   std::stable_sort(names.begin(), names.end(), NameAssoc::IndexLess{});
2546   return NameMap{std::move(names)};
2547 }
2548 
DecodeIndirectNameMap(base::Vector<const uint8_t> module_bytes,uint8_t name_section_kind)2549 IndirectNameMap DecodeIndirectNameMap(base::Vector<const uint8_t> module_bytes,
2550                                       uint8_t name_section_kind) {
2551   Decoder decoder(module_bytes);
2552   if (!FindNameSection(&decoder)) return IndirectNameMap{{}};
2553 
2554   std::vector<IndirectNameMapEntry> entries;
2555   while (decoder.ok() && decoder.more()) {
2556     uint8_t name_type = decoder.consume_u8("name type");
2557     if (name_type & 0x80) break;  // no varuint7
2558 
2559     uint32_t name_payload_len = decoder.consume_u32v("name payload length");
2560     if (!decoder.checkAvailable(name_payload_len)) break;
2561 
2562     if (name_type != name_section_kind) {
2563       decoder.consume_bytes(name_payload_len, "name subsection payload");
2564       continue;
2565     }
2566 
2567     uint32_t outer_count = decoder.consume_u32v("outer count");
2568     for (uint32_t i = 0; i < outer_count; ++i) {
2569       uint32_t outer_index = decoder.consume_u32v("outer index");
2570       if (outer_index > kMaxInt) continue;
2571       std::vector<NameAssoc> names;
2572       uint32_t inner_count = decoder.consume_u32v("inner count");
2573       for (uint32_t k = 0; k < inner_count; ++k) {
2574         uint32_t inner_index = decoder.consume_u32v("inner index");
2575         WireBytesRef name = consume_string(&decoder, false, "name");
2576         if (!decoder.ok()) break;
2577         if (inner_index > kMaxInt) continue;
2578         // Ignore non-utf8 names.
2579         if (!validate_utf8(&decoder, name)) continue;
2580         names.emplace_back(static_cast<int>(inner_index), name);
2581       }
2582       // Use stable sort to get deterministic names (the first one declared)
2583       // even in the presence of duplicates.
2584       std::stable_sort(names.begin(), names.end(), NameAssoc::IndexLess{});
2585       entries.emplace_back(static_cast<int>(outer_index), std::move(names));
2586     }
2587   }
2588   std::stable_sort(entries.begin(), entries.end(),
2589                    IndirectNameMapEntry::IndexLess{});
2590   return IndirectNameMap{std::move(entries)};
2591 }
2592 
2593 #undef TRACE
2594 
2595 }  // namespace wasm
2596 }  // namespace internal
2597 }  // namespace v8
2598