1 //===-- llvm/Bitcode/Archive.h - LLVM Bitcode 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/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"
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.
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.
50 CompressedFlag = 1, ///< Member is a normal compressed file
51 SVR4SymbolTableFlag = 2, ///< Member is a SVR4 symbol table
52 BSD4SymbolTableFlag = 4, ///< Member is a BSD4 symbol table
53 LLVMSymbolTableFlag = 8, ///< Member is an LLVM symbol table
54 BitcodeFlag = 16, ///< Member is bitcode
55 HasPathFlag = 64, ///< Member has a full or partial path
56 HasLongFilenameFlag = 128, ///< Member uses the long filename syntax
57 StringTableFlag = 256 ///< Member is an ar(1) format string table
64 /// @returns the parent Archive instance
65 /// @brief Get the archive associated with this member
66 Archive* getArchive() const { return parent; }
68 /// @returns the path to the Archive's file
69 /// @brief Get the path to the archive member
70 const sys::Path& getPath() const { return path; }
72 /// The "user" is the owner of the file per Unix security. This may not
73 /// have any applicability on non-Unix systems but is a required component
74 /// of the "ar" file format.
75 /// @brief Get the user associated with this archive member.
76 unsigned getUser() const { return info.getUser(); }
78 /// The "group" is the owning group of the file per Unix security. This
79 /// may not have any applicability on non-Unix systems but is a required
80 /// component of the "ar" file format.
81 /// @brief Get the group associated with this archive member.
82 unsigned getGroup() const { return info.getGroup(); }
84 /// The "mode" specifies the access permissions for the file per Unix
85 /// security. This may not have any applicabiity on non-Unix systems but is
86 /// a required component of the "ar" file format.
87 /// @brief Get the permission mode associated with this archive member.
88 unsigned getMode() const { return info.getMode(); }
90 /// This method returns the time at which the archive member was last
91 /// modified when it was not in the archive.
92 /// @brief Get the time of last modification of the archive member.
93 sys::TimeValue getModTime() const { return info.getTimestamp(); }
95 /// @returns the size of the archive member in bytes.
96 /// @brief Get the size of the archive member.
97 uint64_t getSize() const { return info.getSize(); }
99 /// This method returns the total size of the archive member as it
100 /// appears on disk. This includes the file content, the header, the
101 /// long file name if any, and the padding.
102 /// @brief Get total on-disk member size.
103 unsigned getMemberSize() const;
105 /// This method will return a pointer to the in-memory content of the
106 /// archive member, if it is available. If the data has not been loaded
107 /// into memory, the return value will be null.
108 /// @returns a pointer to the member's data.
109 /// @brief Get the data content of the archive member
110 const void* getData() const { return data; }
112 /// This method determines if the member is a regular compressed file.
113 /// @returns true iff the archive member is a compressed regular file.
114 /// @brief Determine if the member is a compressed regular file.
115 bool isCompressed() const { return flags&CompressedFlag; }
117 /// @returns true iff the member is a SVR4 (non-LLVM) symbol table
118 /// @brief Determine if this member is a SVR4 symbol table.
119 bool isSVR4SymbolTable() const { return flags&SVR4SymbolTableFlag; }
121 /// @returns true iff the member is a BSD4.4 (non-LLVM) symbol table
122 /// @brief Determine if this member is a BSD4.4 symbol table.
123 bool isBSD4SymbolTable() const { return flags&BSD4SymbolTableFlag; }
125 /// @returns true iff the archive member is the LLVM symbol table
126 /// @brief Determine if this member is the LLVM symbol table.
127 bool isLLVMSymbolTable() const { return flags&LLVMSymbolTableFlag; }
129 /// @returns true iff the archive member is the ar(1) string table
130 /// @brief Determine if this member is the ar(1) string table.
131 bool isStringTable() const { return flags&StringTableFlag; }
133 /// @returns true iff the archive member is a bitcode file.
134 /// @brief Determine if this member is a bitcode file.
135 bool isBitcode() const { return flags&BitcodeFlag; }
137 /// @returns true iff the file name contains a path (directory) component.
138 /// @brief Determine if the member has a path
139 bool hasPath() const { return flags&HasPathFlag; }
141 /// Long filenames are an artifact of the ar(1) file format which allows
142 /// up to sixteen characters in its header and doesn't allow a path
143 /// separator character (/). To avoid this, a "long format" member name is
144 /// allowed that doesn't have this restriction. This method determines if
145 /// that "long format" is used for this member.
146 /// @returns true iff the file name uses the long form
147 /// @brief Determin if the member has a long file name
148 bool hasLongFilename() const { return flags&HasLongFilenameFlag; }
150 /// This method returns the status info (like Unix stat(2)) for the archive
151 /// member. The status info provides the file's size, permissions, and
152 /// modification time. The contents of the Path::StatusInfo structure, other
153 /// than the size and modification time, may not have utility on non-Unix
155 /// @returns the status info for the archive member
156 /// @brief Obtain the status info for the archive member
157 const sys::FileStatus &getFileStatus() const { return info; }
159 /// This method causes the archive member to be replaced with the contents
160 /// of the file specified by \p File. The contents of \p this will be
161 /// updated to reflect the new data from \p File. The \p File must exist and
162 /// be readable on entry to this method.
163 /// @returns true if an error occurred, false otherwise
164 /// @brief Replace contents of archive member with a new file.
165 bool replaceWith(const sys::Path &aFile, std::string* ErrMsg);
168 /// @name ilist methods - do not use
171 const ArchiveMember *getNext() const { return next; }
172 const ArchiveMember *getPrev() const { return prev; }
173 ArchiveMember *getNext() { return next; }
174 ArchiveMember *getPrev() { return prev; }
175 void setPrev(ArchiveMember* p) { prev = p; }
176 void setNext(ArchiveMember* n) { next = n; }
182 ArchiveMember* next; ///< Pointer to next archive member
183 ArchiveMember* prev; ///< Pointer to previous archive member
184 Archive* parent; ///< Pointer to parent archive
185 sys::PathWithStatus path; ///< Path of file containing the member
186 sys::FileStatus info; ///< Status info (size,mode,date)
187 unsigned flags; ///< Flags about the archive member
188 const void* data; ///< Data for the member
191 /// @name Constructors
194 /// The default constructor is only used by the Archive's iplist when it
195 /// constructs the list's sentry node.
199 /// Used internally by the Archive class to construct an ArchiveMember.
200 /// The contents of the ArchiveMember are filled out by the Archive class.
201 ArchiveMember(Archive *PAR);
203 // So Archive can construct an ArchiveMember
204 friend class llvm::Archive;
208 /// This class defines the interface to LLVM Archive files. The Archive class
209 /// presents the archive file as an ilist of ArchiveMember objects. The members
210 /// can be rearranged in any fashion either by directly editing the ilist or by
211 /// using editing methods on the Archive class (recommended). The Archive
212 /// class also provides several ways of accessing the archive file for various
213 /// purposes such as editing and linking. Full symbol table support is provided
214 /// for loading only those files that resolve symbols. Note that read
215 /// performance of this library is _crucial_ for performance of JIT type
216 /// applications and the linkers. Consequently, the implementation of the class
217 /// is optimized for reading.
223 /// This is the ilist type over which users may iterate to examine
224 /// the contents of the archive
225 /// @brief The ilist type of ArchiveMembers that Archive contains.
226 typedef iplist<ArchiveMember> MembersList;
228 /// @brief Forward mutable iterator over ArchiveMember
229 typedef MembersList::iterator iterator;
231 /// @brief Forward immutable iterator over ArchiveMember
232 typedef MembersList::const_iterator const_iterator;
234 /// @brief Reverse mutable iterator over ArchiveMember
235 typedef std::reverse_iterator<iterator> reverse_iterator;
237 /// @brief Reverse immutable iterator over ArchiveMember
238 typedef std::reverse_iterator<const_iterator> const_reverse_iterator;
240 /// @brief The in-memory version of the symbol table
241 typedef std::map<std::string,unsigned> SymTabType;
244 /// @name ilist accessor methods
247 inline iterator begin() { return members.begin(); }
248 inline const_iterator begin() const { return members.begin(); }
249 inline iterator end () { return members.end(); }
250 inline const_iterator end () const { return members.end(); }
252 inline reverse_iterator rbegin() { return members.rbegin(); }
253 inline const_reverse_iterator rbegin() const { return members.rbegin(); }
254 inline reverse_iterator rend () { return members.rend(); }
255 inline const_reverse_iterator rend () const { return members.rend(); }
257 inline unsigned size() const { return members.size(); }
258 inline bool empty() const { return members.empty(); }
259 inline const ArchiveMember& front() const { return members.front(); }
260 inline ArchiveMember& front() { return members.front(); }
261 inline const ArchiveMember& back() const { return members.back(); }
262 inline ArchiveMember& back() { return members.back(); }
265 /// @name ilist mutator methods
268 /// This method splices a \p src member from an archive (possibly \p this),
269 /// to a position just before the member given by \p dest in \p this. When
270 /// the archive is written, \p src will be written in its new location.
271 /// @brief Move a member to a new location
272 inline void splice(iterator dest, Archive& arch, iterator src)
273 { return members.splice(dest,arch.members,src); }
275 /// This method erases a \p target member from the archive. When the
276 /// archive is written, it will no longer contain \p target. The associated
277 /// ArchiveMember is deleted.
278 /// @brief Erase a member.
279 inline iterator erase(iterator target) { return members.erase(target); }
282 /// @name Constructors
285 /// Create an empty archive file and associate it with the \p Filename. This
286 /// method does not actually create the archive disk file. It creates an
287 /// empty Archive object. If the writeToDisk method is called, the archive
288 /// file \p Filename will be created at that point, with whatever content
289 /// the returned Archive object has at that time.
290 /// @returns An Archive* that represents the new archive file.
291 /// @brief Create an empty Archive.
292 static Archive* CreateEmpty(
293 const sys::Path& Filename ///< Name of the archive to (eventually) create.
296 /// Open an existing archive and load its contents in preparation for
297 /// editing. After this call, the member ilist is completely populated based
298 /// on the contents of the archive file. You should use this form of open if
299 /// you intend to modify the archive or traverse its contents (e.g. for
301 /// @brief Open and load an archive file
302 static Archive* OpenAndLoad(
303 const sys::Path& filePath, ///< The file path to open and load
304 std::string* ErrorMessage ///< An optional error string
307 /// This method opens an existing archive file from \p Filename and reads in
308 /// its symbol table without reading in any of the archive's members. This
309 /// reduces both I/O and cpu time in opening the archive if it is to be used
310 /// solely for symbol lookup (e.g. during linking). The \p Filename must
311 /// exist and be an archive file or an exception will be thrown. This form
312 /// of opening the archive is intended for read-only operations that need to
313 /// locate members via the symbol table for link editing. Since the archve
314 /// members are not read by this method, the archive will appear empty upon
315 /// return. If editing operations are performed on the archive, they will
316 /// completely replace the contents of the archive! It is recommended that
317 /// if this form of opening the archive is used that only the symbol table
318 /// lookup methods (getSymbolTable, findModuleDefiningSymbol, and
319 /// findModulesDefiningSymbols) be used.
320 /// @throws std::string if an error occurs opening the file
321 /// @returns an Archive* that represents the archive file.
322 /// @brief Open an existing archive and load its symbols.
323 static Archive* OpenAndLoadSymbols(
324 const sys::Path& Filename, ///< Name of the archive file to open
325 std::string* ErrorMessage=0 ///< An optional error string
328 /// This destructor cleans up the Archive object, releases all memory, and
329 /// closes files. It does nothing with the archive file on disk. If you
330 /// haven't used the writeToDisk method by the time the destructor is
331 /// called, all changes to the archive will be lost.
332 /// @throws std::string if an error occurs
333 /// @brief Destruct in-memory archive
340 /// @returns the path to the archive file.
341 /// @brief Get the archive path.
342 const sys::Path& getPath() { return archPath; }
344 /// This method is provided so that editing methods can be invoked directly
345 /// on the Archive's iplist of ArchiveMember. However, it is recommended
346 /// that the usual STL style iterator interface be used instead.
347 /// @returns the iplist of ArchiveMember
348 /// @brief Get the iplist of the members
349 MembersList& getMembers() { return members; }
351 /// This method allows direct query of the Archive's symbol table. The
352 /// symbol table is a std::map of std::string (the symbol) to unsigned (the
353 /// file offset). Note that for efficiency reasons, the offset stored in
354 /// the symbol table is not the actual offset. It is the offset from the
355 /// beginning of the first "real" file member (after the symbol table). Use
356 /// the getFirstFileOffset() to obtain that offset and add this value to the
357 /// offset in the symbol table to obtain the real file offset. Note that
358 /// there is purposefully no interface provided by Archive to look up
359 /// members by their offset. Use the findModulesDefiningSymbols and
360 /// findModuleDefiningSymbol methods instead.
361 /// @returns the Archive's symbol table.
362 /// @brief Get the archive's symbol table
363 const SymTabType& getSymbolTable() { return symTab; }
365 /// This method returns the offset in the archive file to the first "real"
366 /// file member. Archive files, on disk, have a signature and might have a
367 /// symbol table that precedes the first actual file member. This method
368 /// allows you to determine what the size of those fields are.
369 /// @returns the offset to the first "real" file member in the archive.
370 /// @brief Get the offset to the first "real" file member in the archive.
371 unsigned getFirstFileOffset() { return firstFileOffset; }
373 /// This method will scan the archive for bitcode modules, interpret them
374 /// and return a vector of the instantiated modules in \p Modules. If an
375 /// error occurs, this method will return true. If \p ErrMessage is not null
376 /// and an error occurs, \p *ErrMessage will be set to a string explaining
377 /// the error that occurred.
378 /// @returns true if an error occurred
379 /// @brief Instantiate all the bitcode modules located in the archive
380 bool getAllModules(std::vector<Module*>& Modules, std::string* ErrMessage);
382 /// This accessor looks up the \p symbol in the archive's symbol table and
383 /// returns the associated module that defines that symbol. This method can
384 /// be called as many times as necessary. This is handy for linking the
385 /// archive into another module based on unresolved symbols. Note that the
386 /// ModuleProvider returned by this accessor should not be deleted by the
387 /// caller. It is managed internally by the Archive class. It is possible
388 /// that multiple calls to this accessor will return the same ModuleProvider
389 /// instance because the associated module defines multiple symbols.
390 /// @returns The ModuleProvider* found or null if the archive does not
391 /// contain a module that defines the \p symbol.
392 /// @brief Look up a module by symbol name.
393 ModuleProvider* findModuleDefiningSymbol(
394 const std::string& symbol, ///< Symbol to be sought
395 std::string* ErrMessage ///< Error message storage, if non-zero
398 /// This method is similar to findModuleDefiningSymbol but allows lookup of
399 /// more than one symbol at a time. If \p symbols contains a list of
400 /// undefined symbols in some module, then calling this method is like
401 /// making one complete pass through the archive to resolve symbols but is
402 /// more efficient than looking at the individual members. Note that on
403 /// exit, the symbols resolved by this method will be removed from \p
404 /// symbols to ensure they are not re-searched on a subsequent call. If
405 /// you need to retain the list of symbols, make a copy.
406 /// @brief Look up multiple symbols in the archive.
407 bool findModulesDefiningSymbols(
408 std::set<std::string>& symbols, ///< Symbols to be sought
409 std::set<ModuleProvider*>& modules, ///< The modules matching \p symbols
410 std::string* ErrMessage ///< Error msg storage, if non-zero
413 /// This method determines whether the archive is a properly formed llvm
414 /// bitcode archive. It first makes sure the symbol table has been loaded
415 /// and has a non-zero size. If it does, then it is an archive. If not,
416 /// then it tries to load all the bitcode modules of the archive. Finally,
417 /// it returns whether it was successfull.
418 /// @returns true if the archive is a proper llvm bitcode archive
419 /// @brief Determine whether the archive is a proper llvm bitcode archive.
420 bool isBitcodeArchive();
426 /// This method is the only way to get the archive written to disk. It
427 /// creates or overwrites the file specified when \p this was created
428 /// or opened. The arguments provide options for writing the archive. If
429 /// \p CreateSymbolTable is true, the archive is scanned for bitcode files
430 /// and a symbol table of the externally visible function and global
431 /// variable names is created. If \p TruncateNames is true, the names of the
432 /// archive members will have their path component stripped and the file
433 /// name will be truncated at 15 characters. If \p Compress is specified,
434 /// all archive members will be compressed before being written. If
435 /// \p PrintSymTab is true, the symbol table will be printed to std::cout.
436 /// @returns true if an error occurred, \p error set to error message
437 /// @returns false if the writing succeeded.
438 /// @brief Write (possibly modified) archive contents to disk
440 bool CreateSymbolTable=false, ///< Create Symbol table
441 bool TruncateNames=false, ///< Truncate the filename to 15 chars
442 bool Compress=false, ///< Compress files
443 std::string* ErrMessage=0 ///< If non-null, where error msg is set
446 /// This method adds a new file to the archive. The \p filename is examined
447 /// to determine just enough information to create an ArchiveMember object
448 /// which is then inserted into the Archive object's ilist at the location
449 /// given by \p where.
450 /// @returns true if an error occured, false otherwise
451 /// @brief Add a file to the archive.
453 const sys::Path& filename, ///< The file to be added
454 iterator where, ///< Insertion point
455 std::string* ErrMsg ///< Optional error message location
459 /// @name Implementation
462 /// @brief Construct an Archive for \p filename and optionally map it
464 Archive(const sys::Path& filename);
466 /// @param data The symbol table data to be parsed
467 /// @param len The length of the symbol table data
468 /// @param error Set to address of a std::string to get error messages
469 /// @returns false on error
470 /// @brief Parse the symbol table at \p data.
471 bool parseSymbolTable(const void* data,unsigned len,std::string* error);
473 /// @returns A fully populated ArchiveMember or 0 if an error occurred.
474 /// @brief Parse the header of a member starting at \p At
475 ArchiveMember* parseMemberHeader(
476 const char*&At, ///< The pointer to the location we're parsing
477 const char*End, ///< The pointer to the end of the archive
478 std::string* error ///< Optional error message catcher
481 /// @param ErrMessage Set to address of a std::string to get error messages
482 /// @returns false on error
483 /// @brief Check that the archive signature is correct
484 bool checkSignature(std::string* ErrMessage);
486 /// @param ErrMessage Set to address of a std::string to get error messages
487 /// @returns false on error
488 /// @brief Load the entire archive.
489 bool loadArchive(std::string* ErrMessage);
491 /// @param ErrMessage Set to address of a std::string to get error messages
492 /// @returns false on error
493 /// @brief Load just the symbol table.
494 bool loadSymbolTable(std::string* ErrMessage);
496 /// @brief Write the symbol table to an ofstream.
497 void writeSymbolTable(std::ofstream& ARFile);
499 /// Writes one ArchiveMember to an ofstream. If an error occurs, returns
500 /// false, otherwise true. If an error occurs and error is non-null then
501 /// it will be set to an error message.
502 /// @returns false Writing member succeeded
503 /// @returns true Writing member failed, \p error set to error message
505 const ArchiveMember& member, ///< The member to be written
506 std::ofstream& ARFile, ///< The file to write member onto
507 bool CreateSymbolTable, ///< Should symbol table be created?
508 bool TruncateNames, ///< Should names be truncated to 11 chars?
509 bool ShouldCompress, ///< Should the member be compressed?
510 std::string* ErrMessage ///< If non-null, place were error msg is set
513 /// @brief Fill in an ArchiveMemberHeader from ArchiveMember.
514 bool fillHeader(const ArchiveMember&mbr,
515 ArchiveMemberHeader& hdr,int sz, bool TruncateNames) const;
517 /// @brief Maps archive into memory
518 bool mapToMemory(std::string* ErrMsg);
520 /// @brief Frees all the members and unmaps the archive file.
521 void cleanUpMemory();
523 /// This type is used to keep track of bitcode modules loaded from the
524 /// symbol table. It maps the file offset to a pair that consists of the
525 /// associated ArchiveMember and the ModuleProvider.
526 /// @brief Module mapping type
527 typedef std::map<unsigned,std::pair<ModuleProvider*,ArchiveMember*> >
535 sys::Path archPath; ///< Path to the archive file we read/write
536 MembersList members; ///< The ilist of ArchiveMember
537 sys::MappedFile* mapfile; ///< Raw Archive contents mapped into memory
538 const char* base; ///< Base of the memory mapped file data
539 SymTabType symTab; ///< The symbol table
540 std::string strtab; ///< The string table for long file names
541 unsigned symTabSize; ///< Size in bytes of symbol table
542 unsigned firstFileOffset; ///< Offset to first normal file.
543 ModuleMap modules; ///< The modules loaded via symbol lookup.
544 ArchiveMember* foreignST; ///< This holds the foreign symbol table.
549 Archive(); ///< Do not implement
550 Archive(const Archive&); ///< Do not implement
551 Archive& operator=(const Archive&); ///< Do not implement
555 } // End llvm namespace