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