1 //===-- llvm/Bytecode/Archive.h - LLVM Bytecode Archive ---------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by Reid Spencer and is distributed under the
6 // University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
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/Bytecode/Archive library. This library is used to read and write
13 // archive (*.a) files that contain LLVM bytecode files (or others).
15 //===----------------------------------------------------------------------===//
17 #ifndef LLVM_BYTECODE_ARCHIVE_H
18 #define LLVM_BYTECODE_ARCHIVE_H
20 #include "llvm/ADT/ilist"
21 #include "llvm/System/Path.h"
22 #include "llvm/System/MappedFile.h"
29 // Forward declare classes
30 class ModuleProvider; // From VMCore
31 class Module; // From VMCore
32 class Archive; // Declared below
33 class ArchiveMemberHeader; // Internal implementation class
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.
47 /// These flags are used internally by the archive member to specify various
48 /// characteristics of the member. The various "is" methods below provide
49 /// access to the flags. The flags are not user settable.
51 CompressedFlag = 1, ///< Member is a normal compressed file
52 SVR4SymbolTableFlag = 2, ///< Member is a SVR4 symbol table
53 BSD4SymbolTableFlag = 4, ///< Member is a BSD4 symbol table
54 LLVMSymbolTableFlag = 8, ///< Member is an LLVM symbol table
55 BytecodeFlag = 16, ///< Member is uncompressed bytecode
56 CompressedBytecodeFlag = 32, ///< Member is compressed bytecode
57 HasPathFlag = 64, ///< Member has a full or partial path
58 HasLongFilenameFlag = 128, ///< Member uses the long filename syntax
59 StringTableFlag = 256 ///< Member is an ar(1) format string table
66 /// @returns the parent Archive instance
67 /// @brief Get the archive associated with this member
68 Archive* getArchive() const { return parent; }
70 /// @returns the path to the Archive's file
71 /// @brief Get the path to the archive member
72 const sys::Path& getPath() const { return path; }
74 /// The "user" is the owner of the file per Unix security. This may not
75 /// have any applicability on non-Unix systems but is a required component
76 /// of the "ar" file format.
77 /// @brief Get the user associated with this archive member.
78 unsigned getUser() const { return info.user; }
80 /// The "group" is the owning group of the file per Unix security. This
81 /// may not have any applicability on non-Unix systems but is a required
82 /// component of the "ar" file format.
83 /// @brief Get the group associated with this archive member.
84 unsigned getGroup() const { return info.group; }
86 /// The "mode" specifies the access permissions for the file per Unix
87 /// security. This may not have any applicabiity on non-Unix systems but is
88 /// a required component of the "ar" file format.
89 /// @brief Get the permission mode associated with this archive member.
90 unsigned getMode() const { return info.mode; }
92 /// This method returns the time at which the archive member was last
93 /// modified when it was not in the archive.
94 /// @brief Get the time of last modification of the archive member.
95 sys::TimeValue getModTime() const { return info.modTime; }
97 /// @returns the size of the archive member in bytes.
98 /// @brief Get the size of the archive member.
99 unsigned getSize() const { return info.fileSize; }
101 /// This method returns the total size of the archive member as it
102 /// appears on disk. This includes the file content, the header, the
103 /// long file name if any, and the padding.
104 /// @brief Get total on-disk member size.
105 unsigned getMemberSize() const;
107 /// This method will return a pointer to the in-memory content of the
108 /// archive member, if it is available. If the data has not been loaded
109 /// into memory, the return value will be null.
110 /// @returns a pointer to the member's data.
111 /// @brief Get the data content of the archive member
112 const void* getData() const { return data; }
114 /// This method determines if the member is a regular compressed file. Note
115 /// that compressed bytecode files will yield "false" for this method.
116 /// @see isCompressedBytecode()
117 /// @returns true iff the archive member is a compressed regular file.
118 /// @brief Determine if the member is a compressed regular file.
119 bool isCompressed() const { return flags&CompressedFlag; }
121 /// @returns true iff the member is a SVR4 (non-LLVM) symbol table
122 /// @brief Determine if this member is a SVR4 symbol table.
123 bool isSVR4SymbolTable() const { return flags&SVR4SymbolTableFlag; }
125 /// @returns true iff the member is a BSD4.4 (non-LLVM) symbol table
126 /// @brief Determine if this member is a BSD4.4 symbol table.
127 bool isBSD4SymbolTable() const { return flags&BSD4SymbolTableFlag; }
129 /// @returns true iff the archive member is the LLVM symbol table
130 /// @brief Determine if this member is the LLVM symbol table.
131 bool isLLVMSymbolTable() const { return flags&LLVMSymbolTableFlag; }
133 /// @returns true iff the archive member is the ar(1) string table
134 /// @brief Determine if this member is the ar(1) string table.
135 bool isStringTable() const { return flags&StringTableFlag; }
137 /// @returns true iff the archive member is an uncompressed bytecode file.
138 /// @brief Determine if this member is a bytecode file.
139 bool isBytecode() const { return flags&BytecodeFlag; }
141 /// @returns true iff the archive member is a compressed bytecode file.
142 /// @brief Determine if the member is a compressed bytecode file.
143 bool isCompressedBytecode() const { return flags&CompressedBytecodeFlag;}
145 /// @returns true iff the file name contains a path (directory) component.
146 /// @brief Determine if the member has a path
147 bool hasPath() const { return flags&HasPathFlag; }
149 /// Long filenames are an artifact of the ar(1) file format which allows
150 /// up to sixteen characters in its header and doesn't allow a path
151 /// separator character (/). To avoid this, a "long format" member name is
152 /// allowed that doesn't have this restriction. This method determines if
153 /// that "long format" is used for this member.
154 /// @returns true iff the file name uses the long form
155 /// @brief Determin if the member has a long file name
156 bool hasLongFilename() const { return flags&HasLongFilenameFlag; }
158 /// This method returns the status info (like Unix stat(2)) for the archive
159 /// member. The status info provides the file's size, permissions, and
160 /// modification time. The contents of the Path::StatusInfo structure, other
161 /// than the size and modification time, may not have utility on non-Unix
163 /// @returns the status info for the archive member
164 /// @brief Obtain the status info for the archive member
165 const sys::Path::StatusInfo& getStatusInfo() const { return info; }
167 /// This method causes the archive member to be replaced with the contents
168 /// of the file specified by \p File. The contents of \p this will be
169 /// updated to reflect the new data from \p File. The \p File must exist and
170 /// be readable on entry to this method.
171 /// @brief Replace contents of archive member with a new file.
172 void replaceWith(const sys::Path& aFile);
175 /// @name ilist methods - do not use
178 const ArchiveMember *getNext() const { return next; }
179 const ArchiveMember *getPrev() const { return prev; }
180 ArchiveMember *getNext() { return next; }
181 ArchiveMember *getPrev() { return prev; }
182 void setPrev(ArchiveMember* p) { prev = p; }
183 void setNext(ArchiveMember* n) { next = n; }
189 ArchiveMember* next; ///< Pointer to next archive member
190 ArchiveMember* prev; ///< Pointer to previous archive member
191 Archive* parent; ///< Pointer to parent archive
192 sys::Path path; ///< Path of file containing the member
193 sys::Path::StatusInfo info; ///< Status info (size,mode,date)
194 unsigned flags; ///< Flags about the archive member
195 const void* data; ///< Data for the member
198 /// @name Constructors
201 /// The default constructor is only used by the Archive's iplist when it
202 /// constructs the list's sentry node.
206 /// Used internally by the Archive class to construct an ArchiveMember.
207 /// The contents of the ArchiveMember are filled out by the Archive class.
208 ArchiveMember( Archive* PAR );
210 // So Archive can construct an ArchiveMember
211 friend class llvm::Archive;
215 /// This class defines the interface to LLVM Archive files. The Archive class
216 /// presents the archive file as an ilist of ArchiveMember objects. The members
217 /// can be rearranged in any fashion either by directly editing the ilist or by
218 /// using editing methods on the Archive class (recommended). The Archive
219 /// class also provides several ways of accessing the archive file for various
220 /// purposes such as editing and linking. Full symbol table support is provided
221 /// for loading only those files that resolve symbols. Note that read
222 /// performance of this library is _crucial_ for performance of JIT type
223 /// applications and the linkers. Consequently, the implementation of the class
224 /// is optimized for reading.
230 /// This is the ilist type over which users may iterate to examine
231 /// the contents of the archive
232 /// @brief The ilist type of ArchiveMembers that Archive contains.
233 typedef iplist<ArchiveMember> MembersList;
235 /// @brief Forward mutable iterator over ArchiveMember
236 typedef MembersList::iterator iterator;
238 /// @brief Forward immutable iterator over ArchiveMember
239 typedef MembersList::const_iterator const_iterator;
241 /// @brief Reverse mutable iterator over ArchiveMember
242 typedef std::reverse_iterator<iterator> reverse_iterator;
244 /// @brief Reverse immutable iterator over ArchiveMember
245 typedef std::reverse_iterator<const_iterator> const_reverse_iterator;
247 /// @brief The in-memory version of the symbol table
248 typedef std::map<std::string,unsigned> SymTabType;
251 /// @name ilist accessor methods
254 inline iterator begin() { return members.begin(); }
255 inline const_iterator begin() const { return members.begin(); }
256 inline iterator end () { return members.end(); }
257 inline const_iterator end () const { return members.end(); }
259 inline reverse_iterator rbegin() { return members.rbegin(); }
260 inline const_reverse_iterator rbegin() const { return members.rbegin(); }
261 inline reverse_iterator rend () { return members.rend(); }
262 inline const_reverse_iterator rend () const { return members.rend(); }
264 inline unsigned size() const { return members.size(); }
265 inline bool empty() const { return members.empty(); }
266 inline const ArchiveMember& front() const { return members.front(); }
267 inline ArchiveMember& front() { return members.front(); }
268 inline const ArchiveMember& back() const { return members.back(); }
269 inline ArchiveMember& back() { return members.back(); }
272 /// @name ilist mutator methods
275 /// This method splices a \p src member from an archive (possibly \p this),
276 /// to a position just before the member given by \p dest in \p this. When
277 /// the archive is written, \p src will be written in its new location.
278 /// @brief Move a member to a new location
279 inline void splice(iterator dest, Archive& arch, iterator src)
280 { return members.splice(dest,arch.members,src); }
282 /// This method erases a \p target member from the archive. When the
283 /// archive is written, it will no longer contain \p target. The associated
284 /// ArchiveMember is deleted.
285 /// @brief Erase a member.
286 inline iterator erase(iterator target) { return members.erase(target); }
289 /// @name Constructors
292 /// Create an empty archive file and associate it with the \p Filename. This
293 /// method does not actually create the archive disk file. It creates an
294 /// empty Archive object. If the writeToDisk method is called, the archive
295 /// file \p Filename will be created at that point, with whatever content
296 /// the returned Archive object has at that time.
297 /// @returns An Archive* that represents the new archive file.
298 /// @brief Create an empty Archive.
299 static Archive* CreateEmpty(
300 const sys::Path& Filename ///< Name of the archive to (eventually) create.
303 /// Open an existing archive and load its contents in preparation for
304 /// editing. After this call, the member ilist is completely populated based
305 /// on the contents of the archive file. You should use this form of open if
306 /// you intend to modify the archive or traverse its contents (e.g. for
308 /// @brief Open and load an archive file
309 static Archive* OpenAndLoad(
310 const sys::Path& filePath, ///< The file path to open and load
311 std::string* ErrorMessage = 0 ///< An optional error string
314 /// This method opens an existing archive file from \p Filename and reads in
315 /// its symbol table without reading in any of the archive's members. This
316 /// reduces both I/O and cpu time in opening the archive if it is to be used
317 /// solely for symbol lookup (e.g. during linking). The \p Filename must
318 /// exist and be an archive file or an exception will be thrown. This form
319 /// of opening the archive is intended for read-only operations that need to
320 /// locate members via the symbol table for link editing. Since the archve
321 /// members are not read by this method, the archive will appear empty upon
322 /// return. If editing operations are performed on the archive, they will
323 /// completely replace the contents of the archive! It is recommended that
324 /// if this form of opening the archive is used that only the symbol table
325 /// lookup methods (getSymbolTable, findModuleDefiningSymbol, and
326 /// findModulesDefiningSymbols) be used.
327 /// @throws std::string if an error occurs opening the file
328 /// @returns an Archive* that represents the archive file.
329 /// @brief Open an existing archive and load its symbols.
330 static Archive* OpenAndLoadSymbols(
331 const sys::Path& Filename, ///< Name of the archive file to open
332 std::string* ErrorMessage=0 ///< An optional error string
335 /// This destructor cleans up the Archive object, releases all memory, and
336 /// closes files. It does nothing with the archive file on disk. If you
337 /// haven't used the writeToDisk method by the time the destructor is
338 /// called, all changes to the archive will be lost.
339 /// @throws std::string if an error occurs
340 /// @brief Destruct in-memory archive
347 /// @returns the path to the archive file.
348 /// @brief Get the archive path.
349 const sys::Path& getPath() { return archPath; }
351 /// This method is provided so that editing methods can be invoked directly
352 /// on the Archive's iplist of ArchiveMember. However, it is recommended
353 /// that the usual STL style iterator interface be used instead.
354 /// @returns the iplist of ArchiveMember
355 /// @brief Get the iplist of the members
356 MembersList& getMembers() { return members; }
358 /// This method allows direct query of the Archive's symbol table. The
359 /// symbol table is a std::map of std::string (the symbol) to unsigned (the
360 /// file offset). Note that for efficiency reasons, the offset stored in
361 /// the symbol table is not the actual offset. It is the offset from the
362 /// beginning of the first "real" file member (after the symbol table). Use
363 /// the getFirstFileOffset() to obtain that offset and add this value to the
364 /// offset in the symbol table to obtain the real file offset. Note that
365 /// there is purposefully no interface provided by Archive to look up
366 /// members by their offset. Use the findModulesDefiningSymbols and
367 /// findModuleDefiningSymbol methods instead.
368 /// @returns the Archive's symbol table.
369 /// @brief Get the archive's symbol table
370 const SymTabType& getSymbolTable() { return symTab; }
372 /// This method returns the offset in the archive file to the first "real"
373 /// file member. Archive files, on disk, have a signature and might have a
374 /// symbol table that precedes the first actual file member. This method
375 /// allows you to determine what the size of those fields are.
376 /// @returns the offset to the first "real" file member in the archive.
377 /// @brief Get the offset to the first "real" file member in the archive.
378 unsigned getFirstFileOffset() { return firstFileOffset; }
380 /// This method will scan the archive for bytecode modules, interpret them
381 /// and return a vector of the instantiated modules in \p Modules. If an
382 /// error occurs, this method will return true. If \p ErrMessage is not null
383 /// and an error occurs, \p *ErrMessage will be set to a string explaining
384 /// the error that occurred.
385 /// @returns true if an error occurred
386 /// @brief Instantiate all the bytecode modules located in the archive
387 bool getAllModules(std::vector<Module*>& Modules, std::string* ErrMessage);
389 /// This accessor looks up the \p symbol in the archive's symbol table and
390 /// returns the associated module that defines that symbol. This method can
391 /// be called as many times as necessary. This is handy for linking the
392 /// archive into another module based on unresolved symbols. Note that the
393 /// ModuleProvider returned by this accessor should not be deleted by the
394 /// caller. It is managed internally by the Archive class. It is possible
395 /// that multiple calls to this accessor will return the same ModuleProvider
396 /// instance because the associated module defines multiple symbols.
397 /// @returns The ModuleProvider* found or null if the archive does not
398 /// contain a module that defines the \p symbol.
399 /// @brief Look up a module by symbol name.
400 ModuleProvider* findModuleDefiningSymbol(
401 const std::string& symbol ///< Symbol to be sought
404 /// This method is similar to findModuleDefiningSymbol but allows lookup of
405 /// more than one symbol at a time. If \p symbols contains a list of
406 /// undefined symbols in some module, then calling this method is like
407 /// making one complete pass through the archive to resolve symbols but is
408 /// more efficient than looking at the individual members. Note that on
409 /// exit, the symbols resolved by this method will be removed from \p
410 /// symbols to ensure they are not re-searched on a subsequent call. If
411 /// you need to retain the list of symbols, make a copy.
412 /// @brief Look up multiple symbols in the archive.
413 void findModulesDefiningSymbols(
414 std::set<std::string>& symbols, ///< Symbols to be sought
415 std::set<ModuleProvider*>& modules ///< The modules matching \p symbols
418 /// This method determines whether the archive is a properly formed llvm
419 /// bytecode archive. It first makes sure the symbol table has been loaded
420 /// and has a non-zero size. If it does, then it is an archive. If not,
421 /// then it tries to load all the bytecode modules of the archive. Finally,
422 /// it returns whether it was successfull.
423 /// @returns true if the archive is a proper llvm bytecode archive
424 /// @brief Determine whether the archive is a proper llvm bytecode archive.
425 bool isBytecodeArchive();
431 /// This method is the only way to get the archive written to disk. It
432 /// creates or overwrites the file specified when \p this was created
433 /// or opened. The arguments provide options for writing the archive. If
434 /// \p CreateSymbolTable is true, the archive is scanned for bytecode files
435 /// and a symbol table of the externally visible function and global
436 /// variable names is created. If \p TruncateNames is true, the names of the
437 /// archive members will have their path component stripped and the file
438 /// name will be truncated at 15 characters. If \p Compress is specified,
439 /// all archive members will be compressed before being written. If
440 /// \p PrintSymTab is true, the symbol table will be printed to std::cout.
441 /// @throws std::string if an error occurs
442 /// @brief Write (possibly modified) archive contents to disk
444 bool CreateSymbolTable=false, ///< Create Symbol table
445 bool TruncateNames=false, ///< Truncate the filename to 15 chars
446 bool Compress=false ///< Compress files
449 /// This method adds a new file to the archive. The \p filename is examined
450 /// to determine just enough information to create an ArchiveMember object
451 /// which is then inserted into the Archive object's ilist at the location
452 /// given by \p where.
453 /// @throws std::string if an error occurs reading the \p filename.
455 /// @brief Add a file to the archive.
456 void addFileBefore(const sys::Path& filename, iterator where);
459 /// @name Implementation
462 /// @brief Construct an Archive for \p filename and optionally map it
464 Archive(const sys::Path& filename, bool map = false );
466 /// @brief Parse the symbol table at \p data.
467 void parseSymbolTable(const void* data,unsigned len);
469 /// @brief Parse the header of a member starting at \p At
470 ArchiveMember* parseMemberHeader(const char*&At,const char*End);
472 /// @brief Check that the archive signature is correct
473 void checkSignature();
475 /// @brief Load the entire archive.
478 /// @brief Load just the symbol table.
479 void loadSymbolTable();
481 /// @brief Write the symbol table to an ofstream.
482 void writeSymbolTable(std::ofstream& ARFile);
484 /// @brief Write one ArchiveMember to an ofstream.
485 void writeMember(const ArchiveMember& member, std::ofstream& ARFile,
486 bool CreateSymbolTable, bool TruncateNames, bool ShouldCompress);
488 /// @brief Fill in an ArchiveMemberHeader from ArchiveMember.
489 bool fillHeader(const ArchiveMember&mbr,
490 ArchiveMemberHeader& hdr,int sz, bool TruncateNames) const;
492 /// @brief Frees all the members and unmaps the archive file.
493 void cleanUpMemory();
495 /// This type is used to keep track of bytecode modules loaded from the
496 /// symbol table. It maps the file offset to a pair that consists of the
497 /// associated ArchiveMember and the ModuleProvider.
498 /// @brief Module mapping type
499 typedef std::map<unsigned,std::pair<ModuleProvider*,ArchiveMember*> >
506 sys::Path archPath; ///< Path to the archive file we read/write
507 MembersList members; ///< The ilist of ArchiveMember
508 sys::MappedFile* mapfile; ///< Raw Archive contents mapped into memory
509 const char* base; ///< Base of the memory mapped file data
510 SymTabType symTab; ///< The symbol table
511 std::string strtab; ///< The string table for long file names
512 unsigned symTabSize; ///< Size in bytes of symbol table
513 unsigned firstFileOffset; ///< Offset to first normal file.
514 ModuleMap modules; ///< The modules loaded via symbol lookup.
515 ArchiveMember* foreignST; ///< This holds the foreign symbol table.
521 Archive(); ///< Do not implement
522 Archive(const Archive&); ///< Do not implement
523 Archive& operator=(const Archive&); ///< Do not implement
527 } // End llvm namespace