1 //===-- llvm/Bitcode/Archive.h - LLVM Bitcode Archive -----------*- C++ -*-===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This header file declares the Archive and ArchiveMember classes that provide 11 // manipulation of LLVM Archive files. The implementation is provided by the 12 // lib/Bitcode/Archive library. This library is used to read and write 13 // archive (*.a) files that contain LLVM bitcode files (or others). 14 // 15 //===----------------------------------------------------------------------===// 16 17 #ifndef LLVM_BITCODE_ARCHIVE_H 18 #define LLVM_BITCODE_ARCHIVE_H 19 20 #include "llvm/ADT/ilist.h" 21 #include "llvm/ADT/ilist_node.h" 22 #include "llvm/Support/Path.h" 23 #include <map> 24 #include <set> 25 26 namespace llvm { 27 class MemoryBuffer; 28 29 // Forward declare classes 30 class Module; // From VMCore 31 class Archive; // Declared below 32 class ArchiveMemberHeader; // Internal implementation class 33 class LLVMContext; // Global data 34 35 /// This class is the main class manipulated by users of the Archive class. It 36 /// holds information about one member of the Archive. It is also the element 37 /// stored by the Archive's ilist, the Archive's main abstraction. Because of 38 /// the special requirements of archive files, users are not permitted to 39 /// construct ArchiveMember instances. You should obtain them from the methods 40 /// of the Archive class instead. 41 /// @brief This class represents a single archive member. 42 class ArchiveMember : public ilist_node<ArchiveMember> { 43 /// @name Types 44 /// @{ 45 public: 46 /// These flags are used internally by the archive member to specify various 47 /// characteristics of the member. The various "is" methods below provide 48 /// access to the flags. The flags are not user settable. 49 enum Flags { 50 SVR4SymbolTableFlag = 1, ///< Member is a SVR4 symbol table 51 BSD4SymbolTableFlag = 2, ///< Member is a BSD4 symbol table 52 LLVMSymbolTableFlag = 4, ///< Member is an LLVM symbol table 53 BitcodeFlag = 8, ///< Member is bitcode 54 HasPathFlag = 16, ///< Member has a full or partial path 55 HasLongFilenameFlag = 32, ///< Member uses the long filename syntax 56 StringTableFlag = 64 ///< Member is an ar(1) format string table 57 }; 58 59 /// @} 60 /// @name Accessors 61 /// @{ 62 public: 63 /// @returns the parent Archive instance 64 /// @brief Get the archive associated with this member getArchive()65 Archive* getArchive() const { return parent; } 66 67 /// @returns the path to the Archive's file 68 /// @brief Get the path to the archive member getPath()69 const sys::Path& getPath() const { return path; } 70 71 /// The "user" is the owner of the file per Unix security. This may not 72 /// have any applicability on non-Unix systems but is a required component 73 /// of the "ar" file format. 74 /// @brief Get the user associated with this archive member. getUser()75 unsigned getUser() const { return info.getUser(); } 76 77 /// The "group" is the owning group of the file per Unix security. This 78 /// may not have any applicability on non-Unix systems but is a required 79 /// component of the "ar" file format. 80 /// @brief Get the group associated with this archive member. getGroup()81 unsigned getGroup() const { return info.getGroup(); } 82 83 /// The "mode" specifies the access permissions for the file per Unix 84 /// security. This may not have any applicability on non-Unix systems but is 85 /// a required component of the "ar" file format. 86 /// @brief Get the permission mode associated with this archive member. getMode()87 unsigned getMode() const { return info.getMode(); } 88 89 /// This method returns the time at which the archive member was last 90 /// modified when it was not in the archive. 91 /// @brief Get the time of last modification of the archive member. getModTime()92 sys::TimeValue getModTime() const { return info.getTimestamp(); } 93 94 /// @returns the size of the archive member in bytes. 95 /// @brief Get the size of the archive member. getSize()96 uint64_t getSize() const { return info.getSize(); } 97 98 /// This method returns the total size of the archive member as it 99 /// appears on disk. This includes the file content, the header, the 100 /// long file name if any, and the padding. 101 /// @brief Get total on-disk member size. 102 unsigned getMemberSize() const; 103 104 /// This method will return a pointer to the in-memory content of the 105 /// archive member, if it is available. If the data has not been loaded 106 /// into memory, the return value will be null. 107 /// @returns a pointer to the member's data. 108 /// @brief Get the data content of the archive member getData()109 const char* getData() const { return data; } 110 111 /// @returns true iff the member is a SVR4 (non-LLVM) symbol table 112 /// @brief Determine if this member is a SVR4 symbol table. isSVR4SymbolTable()113 bool isSVR4SymbolTable() const { return flags&SVR4SymbolTableFlag; } 114 115 /// @returns true iff the member is a BSD4.4 (non-LLVM) symbol table 116 /// @brief Determine if this member is a BSD4.4 symbol table. isBSD4SymbolTable()117 bool isBSD4SymbolTable() const { return flags&BSD4SymbolTableFlag; } 118 119 /// @returns true iff the archive member is the LLVM symbol table 120 /// @brief Determine if this member is the LLVM symbol table. isLLVMSymbolTable()121 bool isLLVMSymbolTable() const { return flags&LLVMSymbolTableFlag; } 122 123 /// @returns true iff the archive member is the ar(1) string table 124 /// @brief Determine if this member is the ar(1) string table. isStringTable()125 bool isStringTable() const { return flags&StringTableFlag; } 126 127 /// @returns true iff the archive member is a bitcode file. 128 /// @brief Determine if this member is a bitcode file. isBitcode()129 bool isBitcode() const { return flags&BitcodeFlag; } 130 131 /// @returns true iff the file name contains a path (directory) component. 132 /// @brief Determine if the member has a path hasPath()133 bool hasPath() const { return flags&HasPathFlag; } 134 135 /// Long filenames are an artifact of the ar(1) file format which allows 136 /// up to sixteen characters in its header and doesn't allow a path 137 /// separator character (/). To avoid this, a "long format" member name is 138 /// allowed that doesn't have this restriction. This method determines if 139 /// that "long format" is used for this member. 140 /// @returns true iff the file name uses the long form 141 /// @brief Determine if the member has a long file name hasLongFilename()142 bool hasLongFilename() const { return flags&HasLongFilenameFlag; } 143 144 /// This method returns the status info (like Unix stat(2)) for the archive 145 /// member. The status info provides the file's size, permissions, and 146 /// modification time. The contents of the Path::StatusInfo structure, other 147 /// than the size and modification time, may not have utility on non-Unix 148 /// systems. 149 /// @returns the status info for the archive member 150 /// @brief Obtain the status info for the archive member getFileStatus()151 const sys::FileStatus &getFileStatus() const { return info; } 152 153 /// This method causes the archive member to be replaced with the contents 154 /// of the file specified by \p File. The contents of \p this will be 155 /// updated to reflect the new data from \p File. The \p File must exist and 156 /// be readable on entry to this method. 157 /// @returns true if an error occurred, false otherwise 158 /// @brief Replace contents of archive member with a new file. 159 bool replaceWith(const sys::Path &aFile, std::string* ErrMsg); 160 161 /// @} 162 /// @name Data 163 /// @{ 164 private: 165 Archive* parent; ///< Pointer to parent archive 166 sys::PathWithStatus path; ///< Path of file containing the member 167 sys::FileStatus info; ///< Status info (size,mode,date) 168 unsigned flags; ///< Flags about the archive member 169 const char* data; ///< Data for the member 170 171 /// @} 172 /// @name Constructors 173 /// @{ 174 public: 175 /// The default constructor is only used by the Archive's iplist when it 176 /// constructs the list's sentry node. 177 ArchiveMember(); 178 179 private: 180 /// Used internally by the Archive class to construct an ArchiveMember. 181 /// The contents of the ArchiveMember are filled out by the Archive class. 182 explicit ArchiveMember(Archive *PAR); 183 184 // So Archive can construct an ArchiveMember 185 friend class llvm::Archive; 186 /// @} 187 }; 188 189 /// This class defines the interface to LLVM Archive files. The Archive class 190 /// presents the archive file as an ilist of ArchiveMember objects. The members 191 /// can be rearranged in any fashion either by directly editing the ilist or by 192 /// using editing methods on the Archive class (recommended). The Archive 193 /// class also provides several ways of accessing the archive file for various 194 /// purposes such as editing and linking. Full symbol table support is provided 195 /// for loading only those files that resolve symbols. Note that read 196 /// performance of this library is _crucial_ for performance of JIT type 197 /// applications and the linkers. Consequently, the implementation of the class 198 /// is optimized for reading. 199 class Archive { 200 201 /// @name Types 202 /// @{ 203 public: 204 /// This is the ilist type over which users may iterate to examine 205 /// the contents of the archive 206 /// @brief The ilist type of ArchiveMembers that Archive contains. 207 typedef iplist<ArchiveMember> MembersList; 208 209 /// @brief Forward mutable iterator over ArchiveMember 210 typedef MembersList::iterator iterator; 211 212 /// @brief Forward immutable iterator over ArchiveMember 213 typedef MembersList::const_iterator const_iterator; 214 215 /// @brief Reverse mutable iterator over ArchiveMember 216 typedef std::reverse_iterator<iterator> reverse_iterator; 217 218 /// @brief Reverse immutable iterator over ArchiveMember 219 typedef std::reverse_iterator<const_iterator> const_reverse_iterator; 220 221 /// @brief The in-memory version of the symbol table 222 typedef std::map<std::string,unsigned> SymTabType; 223 224 /// @} 225 /// @name ilist accessor methods 226 /// @{ 227 public: begin()228 inline iterator begin() { return members.begin(); } begin()229 inline const_iterator begin() const { return members.begin(); } end()230 inline iterator end () { return members.end(); } end()231 inline const_iterator end () const { return members.end(); } 232 rbegin()233 inline reverse_iterator rbegin() { return members.rbegin(); } rbegin()234 inline const_reverse_iterator rbegin() const { return members.rbegin(); } rend()235 inline reverse_iterator rend () { return members.rend(); } rend()236 inline const_reverse_iterator rend () const { return members.rend(); } 237 size()238 inline size_t size() const { return members.size(); } empty()239 inline bool empty() const { return members.empty(); } front()240 inline const ArchiveMember& front() const { return members.front(); } front()241 inline ArchiveMember& front() { return members.front(); } back()242 inline const ArchiveMember& back() const { return members.back(); } back()243 inline ArchiveMember& back() { return members.back(); } 244 245 /// @} 246 /// @name ilist mutator methods 247 /// @{ 248 public: 249 /// This method splices a \p src member from an archive (possibly \p this), 250 /// to a position just before the member given by \p dest in \p this. When 251 /// the archive is written, \p src will be written in its new location. 252 /// @brief Move a member to a new location splice(iterator dest,Archive & arch,iterator src)253 inline void splice(iterator dest, Archive& arch, iterator src) 254 { return members.splice(dest,arch.members,src); } 255 256 /// This method erases a \p target member from the archive. When the 257 /// archive is written, it will no longer contain \p target. The associated 258 /// ArchiveMember is deleted. 259 /// @brief Erase a member. erase(iterator target)260 inline iterator erase(iterator target) { return members.erase(target); } 261 262 /// @} 263 /// @name Constructors 264 /// @{ 265 public: 266 /// Create an empty archive file and associate it with the \p Filename. This 267 /// method does not actually create the archive disk file. It creates an 268 /// empty Archive object. If the writeToDisk method is called, the archive 269 /// file \p Filename will be created at that point, with whatever content 270 /// the returned Archive object has at that time. 271 /// @returns An Archive* that represents the new archive file. 272 /// @brief Create an empty Archive. 273 static Archive* CreateEmpty( 274 const sys::Path& Filename,///< Name of the archive to (eventually) create. 275 LLVMContext& C ///< Context to use for global information 276 ); 277 278 /// Open an existing archive and load its contents in preparation for 279 /// editing. After this call, the member ilist is completely populated based 280 /// on the contents of the archive file. You should use this form of open if 281 /// you intend to modify the archive or traverse its contents (e.g. for 282 /// printing). 283 /// @brief Open and load an archive file 284 static Archive* OpenAndLoad( 285 const sys::Path& filePath, ///< The file path to open and load 286 LLVMContext& C, ///< The context to use for global information 287 std::string* ErrorMessage ///< An optional error string 288 ); 289 290 /// This method opens an existing archive file from \p Filename and reads in 291 /// its symbol table without reading in any of the archive's members. This 292 /// reduces both I/O and cpu time in opening the archive if it is to be used 293 /// solely for symbol lookup (e.g. during linking). The \p Filename must 294 /// exist and be an archive file or an error will be returned. This form 295 /// of opening the archive is intended for read-only operations that need to 296 /// locate members via the symbol table for link editing. Since the archve 297 /// members are not read by this method, the archive will appear empty upon 298 /// return. If editing operations are performed on the archive, they will 299 /// completely replace the contents of the archive! It is recommended that 300 /// if this form of opening the archive is used that only the symbol table 301 /// lookup methods (getSymbolTable, findModuleDefiningSymbol, and 302 /// findModulesDefiningSymbols) be used. 303 /// @returns an Archive* that represents the archive file, or null on error. 304 /// @brief Open an existing archive and load its symbols. 305 static Archive* OpenAndLoadSymbols( 306 const sys::Path& Filename, ///< Name of the archive file to open 307 LLVMContext& C, ///< The context to use for global info 308 std::string* ErrorMessage=0 ///< An optional error string 309 ); 310 311 /// This destructor cleans up the Archive object, releases all memory, and 312 /// closes files. It does nothing with the archive file on disk. If you 313 /// haven't used the writeToDisk method by the time the destructor is 314 /// called, all changes to the archive will be lost. 315 /// @brief Destruct in-memory archive 316 ~Archive(); 317 318 /// @} 319 /// @name Accessors 320 /// @{ 321 public: 322 /// @returns the path to the archive file. 323 /// @brief Get the archive path. getPath()324 const sys::Path& getPath() { return archPath; } 325 326 /// This method is provided so that editing methods can be invoked directly 327 /// on the Archive's iplist of ArchiveMember. However, it is recommended 328 /// that the usual STL style iterator interface be used instead. 329 /// @returns the iplist of ArchiveMember 330 /// @brief Get the iplist of the members getMembers()331 MembersList& getMembers() { return members; } 332 333 /// This method allows direct query of the Archive's symbol table. The 334 /// symbol table is a std::map of std::string (the symbol) to unsigned (the 335 /// file offset). Note that for efficiency reasons, the offset stored in 336 /// the symbol table is not the actual offset. It is the offset from the 337 /// beginning of the first "real" file member (after the symbol table). Use 338 /// the getFirstFileOffset() to obtain that offset and add this value to the 339 /// offset in the symbol table to obtain the real file offset. Note that 340 /// there is purposefully no interface provided by Archive to look up 341 /// members by their offset. Use the findModulesDefiningSymbols and 342 /// findModuleDefiningSymbol methods instead. 343 /// @returns the Archive's symbol table. 344 /// @brief Get the archive's symbol table getSymbolTable()345 const SymTabType& getSymbolTable() { return symTab; } 346 347 /// This method returns the offset in the archive file to the first "real" 348 /// file member. Archive files, on disk, have a signature and might have a 349 /// symbol table that precedes the first actual file member. This method 350 /// allows you to determine what the size of those fields are. 351 /// @returns the offset to the first "real" file member in the archive. 352 /// @brief Get the offset to the first "real" file member in the archive. getFirstFileOffset()353 unsigned getFirstFileOffset() { return firstFileOffset; } 354 355 /// This method will scan the archive for bitcode modules, interpret them 356 /// and return a vector of the instantiated modules in \p Modules. If an 357 /// error occurs, this method will return true. If \p ErrMessage is not null 358 /// and an error occurs, \p *ErrMessage will be set to a string explaining 359 /// the error that occurred. 360 /// @returns true if an error occurred 361 /// @brief Instantiate all the bitcode modules located in the archive 362 bool getAllModules(std::vector<Module*>& Modules, std::string* ErrMessage); 363 364 /// This accessor looks up the \p symbol in the archive's symbol table and 365 /// returns the associated module that defines that symbol. This method can 366 /// be called as many times as necessary. This is handy for linking the 367 /// archive into another module based on unresolved symbols. Note that the 368 /// Module returned by this accessor should not be deleted by the caller. It 369 /// is managed internally by the Archive class. It is possible that multiple 370 /// calls to this accessor will return the same Module instance because the 371 /// associated module defines multiple symbols. 372 /// @returns The Module* found or null if the archive does not contain a 373 /// module that defines the \p symbol. 374 /// @brief Look up a module by symbol name. 375 Module* findModuleDefiningSymbol( 376 const std::string& symbol, ///< Symbol to be sought 377 std::string* ErrMessage ///< Error message storage, if non-zero 378 ); 379 380 /// This method is similar to findModuleDefiningSymbol but allows lookup of 381 /// more than one symbol at a time. If \p symbols contains a list of 382 /// undefined symbols in some module, then calling this method is like 383 /// making one complete pass through the archive to resolve symbols but is 384 /// more efficient than looking at the individual members. Note that on 385 /// exit, the symbols resolved by this method will be removed from \p 386 /// symbols to ensure they are not re-searched on a subsequent call. If 387 /// you need to retain the list of symbols, make a copy. 388 /// @brief Look up multiple symbols in the archive. 389 bool findModulesDefiningSymbols( 390 std::set<std::string>& symbols, ///< Symbols to be sought 391 SmallVectorImpl<Module*>& modules, ///< The modules matching \p symbols 392 std::string* ErrMessage ///< Error msg storage, if non-zero 393 ); 394 395 /// This method determines whether the archive is a properly formed llvm 396 /// bitcode archive. It first makes sure the symbol table has been loaded 397 /// and has a non-zero size. If it does, then it is an archive. If not, 398 /// then it tries to load all the bitcode modules of the archive. Finally, 399 /// it returns whether it was successful. 400 /// @returns true if the archive is a proper llvm bitcode archive 401 /// @brief Determine whether the archive is a proper llvm bitcode archive. 402 bool isBitcodeArchive(); 403 404 /// @} 405 /// @name Mutators 406 /// @{ 407 public: 408 /// This method is the only way to get the archive written to disk. It 409 /// creates or overwrites the file specified when \p this was created 410 /// or opened. The arguments provide options for writing the archive. If 411 /// \p CreateSymbolTable is true, the archive is scanned for bitcode files 412 /// and a symbol table of the externally visible function and global 413 /// variable names is created. If \p TruncateNames is true, the names of the 414 /// archive members will have their path component stripped and the file 415 /// name will be truncated at 15 characters. If \p Compress is specified, 416 /// all archive members will be compressed before being written. If 417 /// \p PrintSymTab is true, the symbol table will be printed to std::cout. 418 /// @returns true if an error occurred, \p error set to error message; 419 /// returns false if the writing succeeded. 420 /// @brief Write (possibly modified) archive contents to disk 421 bool writeToDisk( 422 bool CreateSymbolTable=false, ///< Create Symbol table 423 bool TruncateNames=false, ///< Truncate the filename to 15 chars 424 std::string* ErrMessage=0 ///< If non-null, where error msg is set 425 ); 426 427 /// This method adds a new file to the archive. The \p filename is examined 428 /// to determine just enough information to create an ArchiveMember object 429 /// which is then inserted into the Archive object's ilist at the location 430 /// given by \p where. 431 /// @returns true if an error occurred, false otherwise 432 /// @brief Add a file to the archive. 433 bool addFileBefore( 434 const sys::Path& filename, ///< The file to be added 435 iterator where, ///< Insertion point 436 std::string* ErrMsg ///< Optional error message location 437 ); 438 439 /// @} 440 /// @name Implementation 441 /// @{ 442 protected: 443 /// @brief Construct an Archive for \p filename and optionally map it 444 /// into memory. 445 explicit Archive(const sys::Path& filename, LLVMContext& C); 446 447 /// @param data The symbol table data to be parsed 448 /// @param len The length of the symbol table data 449 /// @param error Set to address of a std::string to get error messages 450 /// @returns false on error 451 /// @brief Parse the symbol table at \p data. 452 bool parseSymbolTable(const void* data,unsigned len,std::string* error); 453 454 /// @returns A fully populated ArchiveMember or 0 if an error occurred. 455 /// @brief Parse the header of a member starting at \p At 456 ArchiveMember* parseMemberHeader( 457 const char*&At, ///< The pointer to the location we're parsing 458 const char*End, ///< The pointer to the end of the archive 459 std::string* error ///< Optional error message catcher 460 ); 461 462 /// @param ErrMessage Set to address of a std::string to get error messages 463 /// @returns false on error 464 /// @brief Check that the archive signature is correct 465 bool checkSignature(std::string* ErrMessage); 466 467 /// @param ErrMessage Set to address of a std::string to get error messages 468 /// @returns false on error 469 /// @brief Load the entire archive. 470 bool loadArchive(std::string* ErrMessage); 471 472 /// @param ErrMessage Set to address of a std::string to get error messages 473 /// @returns false on error 474 /// @brief Load just the symbol table. 475 bool loadSymbolTable(std::string* ErrMessage); 476 477 /// @brief Write the symbol table to an ofstream. 478 void writeSymbolTable(std::ofstream& ARFile); 479 480 /// Writes one ArchiveMember to an ofstream. If an error occurs, returns 481 /// false, otherwise true. If an error occurs and error is non-null then 482 /// it will be set to an error message. 483 /// @returns false if writing member succeeded, 484 /// returns true if writing member failed, \p error set to error message. 485 bool writeMember( 486 const ArchiveMember& member, ///< The member to be written 487 std::ofstream& ARFile, ///< The file to write member onto 488 bool CreateSymbolTable, ///< Should symbol table be created? 489 bool TruncateNames, ///< Should names be truncated to 11 chars? 490 std::string* ErrMessage ///< If non-null, place were error msg is set 491 ); 492 493 /// @brief Fill in an ArchiveMemberHeader from ArchiveMember. 494 bool fillHeader(const ArchiveMember&mbr, 495 ArchiveMemberHeader& hdr,int sz, bool TruncateNames) const; 496 497 /// @brief Maps archive into memory 498 bool mapToMemory(std::string* ErrMsg); 499 500 /// @brief Frees all the members and unmaps the archive file. 501 void cleanUpMemory(); 502 503 /// This type is used to keep track of bitcode modules loaded from the 504 /// symbol table. It maps the file offset to a pair that consists of the 505 /// associated ArchiveMember and the Module. 506 /// @brief Module mapping type 507 typedef std::map<unsigned,std::pair<Module*,ArchiveMember*> > 508 ModuleMap; 509 510 511 /// @} 512 /// @name Data 513 /// @{ 514 protected: 515 sys::Path archPath; ///< Path to the archive file we read/write 516 MembersList members; ///< The ilist of ArchiveMember 517 MemoryBuffer *mapfile; ///< Raw Archive contents mapped into memory 518 const char* base; ///< Base of the memory mapped file data 519 SymTabType symTab; ///< The symbol table 520 std::string strtab; ///< The string table for long file names 521 unsigned symTabSize; ///< Size in bytes of symbol table 522 unsigned firstFileOffset; ///< Offset to first normal file. 523 ModuleMap modules; ///< The modules loaded via symbol lookup. 524 ArchiveMember* foreignST; ///< This holds the foreign symbol table. 525 LLVMContext& Context; ///< This holds global data. 526 /// @} 527 /// @name Hidden 528 /// @{ 529 private: 530 Archive(); ///< Do not implement 531 Archive(const Archive&); ///< Do not implement 532 Archive& operator=(const Archive&); ///< Do not implement 533 /// @} 534 }; 535 536 } // End llvm namespace 537 538 #endif 539