1 //===- llvm/Bitcode/BitcodeReader.h - Bitcode reader ------------*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This header defines interfaces to read LLVM bitcode files/streams.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #ifndef LLVM_BITCODE_BITCODEREADER_H
14 #define LLVM_BITCODE_BITCODEREADER_H
15 
16 #include "llvm/ADT/ArrayRef.h"
17 #include "llvm/ADT/StringRef.h"
18 #include "llvm/Bitstream/BitCodeEnums.h"
19 #include "llvm/IR/GlobalValue.h"
20 #include "llvm/Support/Endian.h"
21 #include "llvm/Support/Error.h"
22 #include "llvm/Support/ErrorOr.h"
23 #include "llvm/Support/MemoryBufferRef.h"
24 #include <cstdint>
25 #include <memory>
26 #include <optional>
27 #include <string>
28 #include <system_error>
29 #include <vector>
30 namespace llvm {
31 
32 class LLVMContext;
33 class Module;
34 class MemoryBuffer;
35 class Metadata;
36 class ModuleSummaryIndex;
37 class Type;
38 class Value;
39 
40 // Callback to override the data layout string of an imported bitcode module.
41 // The first argument is the target triple, the second argument the data layout
42 // string from the input, or a default string. It will be used if the callback
43 // returns std::nullopt.
44 typedef std::function<std::optional<std::string>(StringRef, StringRef)>
45     DataLayoutCallbackFuncTy;
46 
47 typedef std::function<Type *(unsigned)> GetTypeByIDTy;
48 
49 typedef std::function<unsigned(unsigned, unsigned)> GetContainedTypeIDTy;
50 
51 typedef std::function<void(Value *, unsigned, GetTypeByIDTy,
52                            GetContainedTypeIDTy)>
53     ValueTypeCallbackTy;
54 
55 typedef std::function<void(Metadata **, unsigned, GetTypeByIDTy,
56                            GetContainedTypeIDTy)>
57     MDTypeCallbackTy;
58 
59 // These functions are for converting Expected/Error values to
60 // ErrorOr/std::error_code for compatibility with legacy clients. FIXME:
61 // Remove these functions once no longer needed by the C and libLTO APIs.
62 
63 std::error_code errorToErrorCodeAndEmitErrors(LLVMContext &Ctx, Error Err);
64 
65 template <typename T>
66 ErrorOr<T> expectedToErrorOrAndEmitErrors(LLVMContext &Ctx, Expected<T> Val) {
67   if (!Val)
68     return errorToErrorCodeAndEmitErrors(Ctx, Val.takeError());
69   return std::move(*Val);
70 }
71 
72 struct ParserCallbacks {
73   std::optional<DataLayoutCallbackFuncTy> DataLayout;
74   /// The ValueType callback is called for every function definition or
75   /// declaration and allows accessing the type information, also behind
76   /// pointers. This can be useful, when the opaque pointer upgrade cleans all
77   /// type information behind pointers.
78   /// The second argument to ValueTypeCallback is the type ID of the
79   /// function, the two passed functions can be used to extract type
80   /// information.
81   std::optional<ValueTypeCallbackTy> ValueType;
82   /// The MDType callback is called for every value in metadata.
83   std::optional<MDTypeCallbackTy> MDType;
84 
85   ParserCallbacks() = default;
86   explicit ParserCallbacks(DataLayoutCallbackFuncTy DataLayout)
87       : DataLayout(DataLayout) {}
88 };
89 
90   struct BitcodeFileContents;
91 
92   /// Basic information extracted from a bitcode module to be used for LTO.
93   struct BitcodeLTOInfo {
94     bool IsThinLTO;
95     bool HasSummary;
96     bool EnableSplitLTOUnit;
97     bool UnifiedLTO;
98   };
99 
100   /// Represents a module in a bitcode file.
101   class BitcodeModule {
102     // This covers the identification (if present) and module blocks.
103     ArrayRef<uint8_t> Buffer;
104     StringRef ModuleIdentifier;
105 
106     // The string table used to interpret this module.
107     StringRef Strtab;
108 
109     // The bitstream location of the IDENTIFICATION_BLOCK.
110     uint64_t IdentificationBit;
111 
112     // The bitstream location of this module's MODULE_BLOCK.
113     uint64_t ModuleBit;
114 
115     BitcodeModule(ArrayRef<uint8_t> Buffer, StringRef ModuleIdentifier,
116                   uint64_t IdentificationBit, uint64_t ModuleBit)
117         : Buffer(Buffer), ModuleIdentifier(ModuleIdentifier),
118           IdentificationBit(IdentificationBit), ModuleBit(ModuleBit) {}
119 
120     // Calls the ctor.
121     friend Expected<BitcodeFileContents>
122     getBitcodeFileContents(MemoryBufferRef Buffer);
123 
124     Expected<std::unique_ptr<Module>>
125     getModuleImpl(LLVMContext &Context, bool MaterializeAll,
126                   bool ShouldLazyLoadMetadata, bool IsImporting,
127                   ParserCallbacks Callbacks = {});
128 
129   public:
130     StringRef getBuffer() const {
131       return StringRef((const char *)Buffer.begin(), Buffer.size());
132     }
133 
134     StringRef getStrtab() const { return Strtab; }
135 
136     StringRef getModuleIdentifier() const { return ModuleIdentifier; }
137 
138     /// Read the bitcode module and prepare for lazy deserialization of function
139     /// bodies. If ShouldLazyLoadMetadata is true, lazily load metadata as well.
140     /// If IsImporting is true, this module is being parsed for ThinLTO
141     /// importing into another module.
142     Expected<std::unique_ptr<Module>>
143     getLazyModule(LLVMContext &Context, bool ShouldLazyLoadMetadata,
144                   bool IsImporting, ParserCallbacks Callbacks = {});
145 
146     /// Read the entire bitcode module and return it.
147     Expected<std::unique_ptr<Module>>
148     parseModule(LLVMContext &Context, ParserCallbacks Callbacks = {});
149 
150     /// Returns information about the module to be used for LTO: whether to
151     /// compile with ThinLTO, and whether it has a summary.
152     Expected<BitcodeLTOInfo> getLTOInfo();
153 
154     /// Parse the specified bitcode buffer, returning the module summary index.
155     Expected<std::unique_ptr<ModuleSummaryIndex>> getSummary();
156 
157     /// Parse the specified bitcode buffer and merge its module summary index
158     /// into CombinedIndex.
159     Error
160     readSummary(ModuleSummaryIndex &CombinedIndex, StringRef ModulePath,
161                 std::function<bool(GlobalValue::GUID)> IsPrevailing = nullptr);
162   };
163 
164   struct BitcodeFileContents {
165     std::vector<BitcodeModule> Mods;
166     StringRef Symtab, StrtabForSymtab;
167   };
168 
169   /// Returns the contents of a bitcode file. This includes the raw contents of
170   /// the symbol table embedded in the bitcode file. Clients which require a
171   /// symbol table should prefer to use irsymtab::read instead of this function
172   /// because it creates a reader for the irsymtab and handles upgrading bitcode
173   /// files without a symbol table or with an old symbol table.
174   Expected<BitcodeFileContents> getBitcodeFileContents(MemoryBufferRef Buffer);
175 
176   /// Returns a list of modules in the specified bitcode buffer.
177   Expected<std::vector<BitcodeModule>>
178   getBitcodeModuleList(MemoryBufferRef Buffer);
179 
180   /// Read the header of the specified bitcode buffer and prepare for lazy
181   /// deserialization of function bodies. If ShouldLazyLoadMetadata is true,
182   /// lazily load metadata as well. If IsImporting is true, this module is
183   /// being parsed for ThinLTO importing into another module.
184   Expected<std::unique_ptr<Module>>
185   getLazyBitcodeModule(MemoryBufferRef Buffer, LLVMContext &Context,
186                        bool ShouldLazyLoadMetadata = false,
187                        bool IsImporting = false,
188                        ParserCallbacks Callbacks = {});
189 
190   /// Like getLazyBitcodeModule, except that the module takes ownership of
191   /// the memory buffer if successful. If successful, this moves Buffer. On
192   /// error, this *does not* move Buffer. If IsImporting is true, this module is
193   /// being parsed for ThinLTO importing into another module.
194   Expected<std::unique_ptr<Module>> getOwningLazyBitcodeModule(
195       std::unique_ptr<MemoryBuffer> &&Buffer, LLVMContext &Context,
196       bool ShouldLazyLoadMetadata = false, bool IsImporting = false,
197       ParserCallbacks Callbacks = {});
198 
199   /// Read the header of the specified bitcode buffer and extract just the
200   /// triple information. If successful, this returns a string. On error, this
201   /// returns "".
202   Expected<std::string> getBitcodeTargetTriple(MemoryBufferRef Buffer);
203 
204   /// Return true if \p Buffer contains a bitcode file with ObjC code (category
205   /// or class) in it.
206   Expected<bool> isBitcodeContainingObjCCategory(MemoryBufferRef Buffer);
207 
208   /// Read the header of the specified bitcode buffer and extract just the
209   /// producer string information. If successful, this returns a string. On
210   /// error, this returns "".
211   Expected<std::string> getBitcodeProducerString(MemoryBufferRef Buffer);
212 
213   /// Read the specified bitcode file, returning the module.
214   Expected<std::unique_ptr<Module>>
215   parseBitcodeFile(MemoryBufferRef Buffer, LLVMContext &Context,
216                    ParserCallbacks Callbacks = {});
217 
218   /// Returns LTO information for the specified bitcode file.
219   Expected<BitcodeLTOInfo> getBitcodeLTOInfo(MemoryBufferRef Buffer);
220 
221   /// Parse the specified bitcode buffer, returning the module summary index.
222   Expected<std::unique_ptr<ModuleSummaryIndex>>
223   getModuleSummaryIndex(MemoryBufferRef Buffer);
224 
225   /// Parse the specified bitcode buffer and merge the index into CombinedIndex.
226   Error readModuleSummaryIndex(MemoryBufferRef Buffer,
227                                ModuleSummaryIndex &CombinedIndex);
228 
229   /// Parse the module summary index out of an IR file and return the module
230   /// summary index object if found, or an empty summary if not. If Path refers
231   /// to an empty file and IgnoreEmptyThinLTOIndexFile is true, then
232   /// this function will return nullptr.
233   Expected<std::unique_ptr<ModuleSummaryIndex>>
234   getModuleSummaryIndexForFile(StringRef Path,
235                                bool IgnoreEmptyThinLTOIndexFile = false);
236 
237   /// isBitcodeWrapper - Return true if the given bytes are the magic bytes
238   /// for an LLVM IR bitcode wrapper.
239   inline bool isBitcodeWrapper(const unsigned char *BufPtr,
240                                const unsigned char *BufEnd) {
241     // See if you can find the hidden message in the magic bytes :-).
242     // (Hint: it's a little-endian encoding.)
243     return BufPtr != BufEnd &&
244            BufPtr[0] == 0xDE &&
245            BufPtr[1] == 0xC0 &&
246            BufPtr[2] == 0x17 &&
247            BufPtr[3] == 0x0B;
248   }
249 
250   /// isRawBitcode - Return true if the given bytes are the magic bytes for
251   /// raw LLVM IR bitcode (without a wrapper).
252   inline bool isRawBitcode(const unsigned char *BufPtr,
253                            const unsigned char *BufEnd) {
254     // These bytes sort of have a hidden message, but it's not in
255     // little-endian this time, and it's a little redundant.
256     return BufPtr != BufEnd &&
257            BufPtr[0] == 'B' &&
258            BufPtr[1] == 'C' &&
259            BufPtr[2] == 0xc0 &&
260            BufPtr[3] == 0xde;
261   }
262 
263   /// isBitcode - Return true if the given bytes are the magic bytes for
264   /// LLVM IR bitcode, either with or without a wrapper.
265   inline bool isBitcode(const unsigned char *BufPtr,
266                         const unsigned char *BufEnd) {
267     return isBitcodeWrapper(BufPtr, BufEnd) ||
268            isRawBitcode(BufPtr, BufEnd);
269   }
270 
271   /// SkipBitcodeWrapperHeader - Some systems wrap bc files with a special
272   /// header for padding or other reasons.  The format of this header is:
273   ///
274   /// struct bc_header {
275   ///   uint32_t Magic;         // 0x0B17C0DE
276   ///   uint32_t Version;       // Version, currently always 0.
277   ///   uint32_t BitcodeOffset; // Offset to traditional bitcode file.
278   ///   uint32_t BitcodeSize;   // Size of traditional bitcode file.
279   ///   ... potentially other gunk ...
280   /// };
281   ///
282   /// This function is called when we find a file with a matching magic number.
283   /// In this case, skip down to the subsection of the file that is actually a
284   /// BC file.
285   /// If 'VerifyBufferSize' is true, check that the buffer is large enough to
286   /// contain the whole bitcode file.
287   inline bool SkipBitcodeWrapperHeader(const unsigned char *&BufPtr,
288                                        const unsigned char *&BufEnd,
289                                        bool VerifyBufferSize) {
290     // Must contain the offset and size field!
291     if (unsigned(BufEnd - BufPtr) < BWH_SizeField + 4)
292       return true;
293 
294     unsigned Offset = support::endian::read32le(&BufPtr[BWH_OffsetField]);
295     unsigned Size = support::endian::read32le(&BufPtr[BWH_SizeField]);
296     uint64_t BitcodeOffsetEnd = (uint64_t)Offset + (uint64_t)Size;
297 
298     // Verify that Offset+Size fits in the file.
299     if (VerifyBufferSize && BitcodeOffsetEnd > uint64_t(BufEnd-BufPtr))
300       return true;
301     BufPtr += Offset;
302     BufEnd = BufPtr+Size;
303     return false;
304   }
305 
306   APInt readWideAPInt(ArrayRef<uint64_t> Vals, unsigned TypeBits);
307 
308   const std::error_category &BitcodeErrorCategory();
309   enum class BitcodeError { CorruptedBitcode = 1 };
310   inline std::error_code make_error_code(BitcodeError E) {
311     return std::error_code(static_cast<int>(E), BitcodeErrorCategory());
312   }
313 
314 } // end namespace llvm
315 
316 namespace std {
317 
318 template <> struct is_error_code_enum<llvm::BitcodeError> : std::true_type {};
319 
320 } // end namespace std
321 
322 #endif // LLVM_BITCODE_BITCODEREADER_H
323