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 TOOLS_LLVM_AR_ARCHIVE_H
18 #define TOOLS_LLVM_AR_ARCHIVE_H
20 #include "llvm/ADT/ilist.h"
21 #include "llvm/ADT/ilist_node.h"
22 #include "llvm/Support/Path.h"
23 #include "llvm/Support/TimeValue.h"
30 // Forward declare classes
31 class Module; // From VMCore
32 class Archive; // Declared below
33 class ArchiveMemberHeader; // Internal implementation class
34 class LLVMContext; // Global data
36 /// This class is the main class manipulated by users of the Archive class. It
37 /// holds information about one member of the Archive. It is also the element
38 /// stored by the Archive's ilist, the Archive's main abstraction. Because of
39 /// the special requirements of archive files, users are not permitted to
40 /// construct ArchiveMember instances. You should obtain them from the methods
41 /// of the Archive class instead.
42 /// @brief This class represents a single archive member.
43 class ArchiveMember : public ilist_node<ArchiveMember> {
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 SVR4SymbolTableFlag = 1, ///< Member is a SVR4 symbol table
52 BSD4SymbolTableFlag = 2, ///< Member is a BSD4 symbol table
53 BitcodeFlag = 4, ///< Member is bitcode
54 HasPathFlag = 8, ///< Member has a full or partial path
55 HasLongFilenameFlag = 16, ///< Member uses the long filename syntax
56 StringTableFlag = 32 ///< Member is an ar(1) format string table
63 /// @returns the parent Archive instance
64 /// @brief Get the archive associated with this member
65 Archive* getArchive() const { return parent; }
67 /// @returns the path to the Archive's file
68 /// @brief Get the path to the archive member
69 StringRef getPath() const { return path; }
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.
75 unsigned getUser() const { return User; }
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.
81 unsigned getGroup() const { return Group; }
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.
87 unsigned getMode() const { return Mode; }
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.
92 sys::TimeValue getModTime() const { return ModTime; }
94 /// @returns the size of the archive member in bytes.
95 /// @brief Get the size of the archive member.
96 uint64_t getSize() const { return Size; }
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;
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
109 const char* getData() const { return data; }
111 /// @returns true iff the member is a SVR4 (non-LLVM) symbol table
112 /// @brief Determine if this member is a SVR4 symbol table.
113 bool isSVR4SymbolTable() const { return flags&SVR4SymbolTableFlag; }
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.
117 bool isBSD4SymbolTable() const { return flags&BSD4SymbolTableFlag; }
119 /// @returns true iff the archive member is the ar(1) string table
120 /// @brief Determine if this member is the ar(1) string table.
121 bool isStringTable() const { return flags&StringTableFlag; }
123 /// @returns true iff the archive member is a bitcode file.
124 /// @brief Determine if this member is a bitcode file.
125 bool isBitcode() const { return flags&BitcodeFlag; }
127 /// @returns true iff the file name contains a path (directory) component.
128 /// @brief Determine if the member has a path
129 bool hasPath() const { return flags&HasPathFlag; }
131 /// Long filenames are an artifact of the ar(1) file format which allows
132 /// up to sixteen characters in its header and doesn't allow a path
133 /// separator character (/). To avoid this, a "long format" member name is
134 /// allowed that doesn't have this restriction. This method determines if
135 /// that "long format" is used for this member.
136 /// @returns true iff the file name uses the long form
137 /// @brief Determine if the member has a long file name
138 bool hasLongFilename() const { return flags&HasLongFilenameFlag; }
140 /// This method causes the archive member to be replaced with the contents
141 /// of the file specified by \p File. The contents of \p this will be
142 /// updated to reflect the new data from \p File. The \p File must exist and
143 /// be readable on entry to this method.
144 /// @returns true if an error occurred, false otherwise
145 /// @brief Replace contents of archive member with a new file.
146 bool replaceWith(StringRef aFile, std::string* ErrMsg);
152 Archive *parent; ///< Pointer to parent archive
153 std::string path; ///< Path of file containing the member
157 sys::TimeValue ModTime;
159 unsigned flags; ///< Flags about the archive member
160 const char *data; ///< Data for the member
163 /// @name Constructors
166 /// The default constructor is only used by the Archive's iplist when it
167 /// constructs the list's sentry node.
171 /// Used internally by the Archive class to construct an ArchiveMember.
172 /// The contents of the ArchiveMember are filled out by the Archive class.
173 explicit ArchiveMember(Archive *PAR);
175 // So Archive can construct an ArchiveMember
176 friend class llvm::Archive;
180 /// This class defines the interface to LLVM Archive files. The Archive class
181 /// presents the archive file as an ilist of ArchiveMember objects. The members
182 /// can be rearranged in any fashion either by directly editing the ilist or by
183 /// using editing methods on the Archive class (recommended). The Archive
184 /// class also provides several ways of accessing the archive file for various
185 /// purposes such as editing and linking. Full symbol table support is provided
186 /// for loading only those files that resolve symbols. Note that read
187 /// performance of this library is _crucial_ for performance of JIT type
188 /// applications and the linkers. Consequently, the implementation of the class
189 /// is optimized for reading.
195 /// This is the ilist type over which users may iterate to examine
196 /// the contents of the archive
197 /// @brief The ilist type of ArchiveMembers that Archive contains.
198 typedef iplist<ArchiveMember> MembersList;
200 /// @brief Forward mutable iterator over ArchiveMember
201 typedef MembersList::iterator iterator;
203 /// @brief Forward immutable iterator over ArchiveMember
204 typedef MembersList::const_iterator const_iterator;
206 /// @brief Reverse mutable iterator over ArchiveMember
207 typedef std::reverse_iterator<iterator> reverse_iterator;
209 /// @brief Reverse immutable iterator over ArchiveMember
210 typedef std::reverse_iterator<const_iterator> const_reverse_iterator;
212 /// @brief The in-memory version of the symbol table
213 typedef std::map<std::string,unsigned> SymTabType;
216 /// @name ilist accessor methods
219 inline iterator begin() { return members.begin(); }
220 inline const_iterator begin() const { return members.begin(); }
221 inline iterator end () { return members.end(); }
222 inline const_iterator end () const { return members.end(); }
224 inline reverse_iterator rbegin() { return members.rbegin(); }
225 inline const_reverse_iterator rbegin() const { return members.rbegin(); }
226 inline reverse_iterator rend () { return members.rend(); }
227 inline const_reverse_iterator rend () const { return members.rend(); }
229 inline size_t size() const { return members.size(); }
230 inline bool empty() const { return members.empty(); }
231 inline const ArchiveMember& front() const { return members.front(); }
232 inline ArchiveMember& front() { return members.front(); }
233 inline const ArchiveMember& back() const { return members.back(); }
234 inline ArchiveMember& back() { return members.back(); }
237 /// @name ilist mutator methods
240 /// This method splices a \p src member from an archive (possibly \p this),
241 /// to a position just before the member given by \p dest in \p this. When
242 /// the archive is written, \p src will be written in its new location.
243 /// @brief Move a member to a new location
244 inline void splice(iterator dest, Archive& arch, iterator src)
245 { return members.splice(dest,arch.members,src); }
247 /// This method erases a \p target member from the archive. When the
248 /// archive is written, it will no longer contain \p target. The associated
249 /// ArchiveMember is deleted.
250 /// @brief Erase a member.
251 inline iterator erase(iterator target) { return members.erase(target); }
254 /// @name Constructors
257 /// Create an empty archive file and associate it with the \p Filename. This
258 /// method does not actually create the archive disk file. It creates an
259 /// empty Archive object. If the writeToDisk method is called, the archive
260 /// file \p Filename will be created at that point, with whatever content
261 /// the returned Archive object has at that time.
262 /// @returns An Archive* that represents the new archive file.
263 /// @brief Create an empty Archive.
264 static Archive *CreateEmpty(
265 StringRef Filename, ///< Name of the archive to (eventually) create.
266 LLVMContext &C ///< Context to use for global information
269 /// Open an existing archive and load its contents in preparation for
270 /// editing. After this call, the member ilist is completely populated based
271 /// on the contents of the archive file. You should use this form of open if
272 /// you intend to modify the archive or traverse its contents (e.g. for
274 /// @brief Open and load an archive file
275 static Archive *OpenAndLoad(
276 StringRef filePath, ///< The file path to open and load
277 LLVMContext &C, ///< The context to use for global information
278 std::string *ErrorMessage ///< An optional error string
281 /// This destructor cleans up the Archive object, releases all memory, and
282 /// closes files. It does nothing with the archive file on disk. If you
283 /// haven't used the writeToDisk method by the time the destructor is
284 /// called, all changes to the archive will be lost.
285 /// @brief Destruct in-memory archive
292 /// @returns the path to the archive file.
293 /// @brief Get the archive path.
294 StringRef getPath() { return archPath; }
296 /// This method is provided so that editing methods can be invoked directly
297 /// on the Archive's iplist of ArchiveMember. However, it is recommended
298 /// that the usual STL style iterator interface be used instead.
299 /// @returns the iplist of ArchiveMember
300 /// @brief Get the iplist of the members
301 MembersList& getMembers() { return members; }
303 /// This method allows direct query of the Archive's symbol table. The
304 /// symbol table is a std::map of std::string (the symbol) to unsigned (the
305 /// file offset). Note that for efficiency reasons, the offset stored in
306 /// the symbol table is not the actual offset. It is the offset from the
307 /// beginning of the first "real" file member (after the symbol table). Use
308 /// the getFirstFileOffset() to obtain that offset and add this value to the
309 /// offset in the symbol table to obtain the real file offset. Note that
310 /// there is purposefully no interface provided by Archive to look up
311 /// members by their offset. Use the findModulesDefiningSymbols and
312 /// findModuleDefiningSymbol methods instead.
313 /// @returns the Archive's symbol table.
314 /// @brief Get the archive's symbol table
315 const SymTabType& getSymbolTable() { return symTab; }
317 /// This method returns the offset in the archive file to the first "real"
318 /// file member. Archive files, on disk, have a signature and might have a
319 /// symbol table that precedes the first actual file member. This method
320 /// allows you to determine what the size of those fields are.
321 /// @returns the offset to the first "real" file member in the archive.
322 /// @brief Get the offset to the first "real" file member in the archive.
323 unsigned getFirstFileOffset() { return firstFileOffset; }
325 /// This method will scan the archive for bitcode modules, interpret them
326 /// and return a vector of the instantiated modules in \p Modules. If an
327 /// error occurs, this method will return true. If \p ErrMessage is not null
328 /// and an error occurs, \p *ErrMessage will be set to a string explaining
329 /// the error that occurred.
330 /// @returns true if an error occurred
331 /// @brief Instantiate all the bitcode modules located in the archive
332 bool getAllModules(std::vector<Module*>& Modules, std::string* ErrMessage);
334 /// This accessor looks up the \p symbol in the archive's symbol table and
335 /// returns the associated module that defines that symbol. This method can
336 /// be called as many times as necessary. This is handy for linking the
337 /// archive into another module based on unresolved symbols. Note that the
338 /// Module returned by this accessor should not be deleted by the caller. It
339 /// is managed internally by the Archive class. It is possible that multiple
340 /// calls to this accessor will return the same Module instance because the
341 /// associated module defines multiple symbols.
342 /// @returns The Module* found or null if the archive does not contain a
343 /// module that defines the \p symbol.
344 /// @brief Look up a module by symbol name.
345 Module* findModuleDefiningSymbol(
346 const std::string& symbol, ///< Symbol to be sought
347 std::string* ErrMessage ///< Error message storage, if non-zero
350 /// This method is similar to findModuleDefiningSymbol but allows lookup of
351 /// more than one symbol at a time. If \p symbols contains a list of
352 /// undefined symbols in some module, then calling this method is like
353 /// making one complete pass through the archive to resolve symbols but is
354 /// more efficient than looking at the individual members. Note that on
355 /// exit, the symbols resolved by this method will be removed from \p
356 /// symbols to ensure they are not re-searched on a subsequent call. If
357 /// you need to retain the list of symbols, make a copy.
358 /// @brief Look up multiple symbols in the archive.
359 bool findModulesDefiningSymbols(
360 std::set<std::string>& symbols, ///< Symbols to be sought
361 SmallVectorImpl<Module*>& modules, ///< The modules matching \p symbols
362 std::string* ErrMessage ///< Error msg storage, if non-zero
365 /// This method determines whether the archive is a properly formed llvm
366 /// bitcode archive. It first makes sure the symbol table has been loaded
367 /// and has a non-zero size. If it does, then it is an archive. If not,
368 /// then it tries to load all the bitcode modules of the archive. Finally,
369 /// it returns whether it was successful.
370 /// @returns true if the archive is a proper llvm bitcode archive
371 /// @brief Determine whether the archive is a proper llvm bitcode archive.
372 bool isBitcodeArchive();
378 /// This method is the only way to get the archive written to disk. It
379 /// creates or overwrites the file specified when \p this was created
380 /// or opened. The arguments provide options for writing the archive. If
381 /// \p CreateSymbolTable is true, the archive is scanned for bitcode files
382 /// and a symbol table of the externally visible function and global
383 /// variable names is created. If \p TruncateNames is true, the names of the
384 /// archive members will have their path component stripped and the file
385 /// name will be truncated at 15 characters. If \p Compress is specified,
386 /// all archive members will be compressed before being written. If
387 /// \p PrintSymTab is true, the symbol table will be printed to std::cout.
388 /// @returns true if an error occurred, \p error set to error message;
389 /// returns false if the writing succeeded.
390 /// @brief Write (possibly modified) archive contents to disk
392 bool CreateSymbolTable=false, ///< Create Symbol table
393 bool TruncateNames=false, ///< Truncate the filename to 15 chars
394 std::string* ErrMessage=0 ///< If non-null, where error msg is set
397 /// This method adds a new file to the archive. The \p filename is examined
398 /// to determine just enough information to create an ArchiveMember object
399 /// which is then inserted into the Archive object's ilist at the location
400 /// given by \p where.
401 /// @returns true if an error occurred, false otherwise
402 /// @brief Add a file to the archive.
403 bool addFileBefore(StringRef filename, ///< The file to be added
404 iterator where, ///< Insertion point
405 std::string *ErrMsg ///< Optional error message location
409 /// @name Implementation
412 /// @brief Construct an Archive for \p filename and optionally map it
414 explicit Archive(StringRef filename, LLVMContext& C);
416 /// @returns A fully populated ArchiveMember or 0 if an error occurred.
417 /// @brief Parse the header of a member starting at \p At
418 ArchiveMember* parseMemberHeader(
419 const char*&At, ///< The pointer to the location we're parsing
420 const char*End, ///< The pointer to the end of the archive
421 std::string* error ///< Optional error message catcher
424 /// @param ErrMessage Set to address of a std::string to get error messages
425 /// @returns false on error
426 /// @brief Check that the archive signature is correct
427 bool checkSignature(std::string* ErrMessage);
429 /// @param ErrMessage Set to address of a std::string to get error messages
430 /// @returns false on error
431 /// @brief Load the entire archive.
432 bool loadArchive(std::string* ErrMessage);
434 /// @param ErrMessage Set to address of a std::string to get error messages
435 /// @returns false on error
436 /// @brief Load just the symbol table.
437 bool loadSymbolTable(std::string* ErrMessage);
439 /// Writes one ArchiveMember to an ofstream. If an error occurs, returns
440 /// false, otherwise true. If an error occurs and error is non-null then
441 /// it will be set to an error message.
442 /// @returns false if writing member succeeded,
443 /// returns true if writing member failed, \p error set to error message.
445 const ArchiveMember& member, ///< The member to be written
446 std::ofstream& ARFile, ///< The file to write member onto
447 bool CreateSymbolTable, ///< Should symbol table be created?
448 bool TruncateNames, ///< Should names be truncated to 11 chars?
449 std::string* ErrMessage ///< If non-null, place were error msg is set
452 /// @brief Fill in an ArchiveMemberHeader from ArchiveMember.
453 bool fillHeader(const ArchiveMember&mbr,
454 ArchiveMemberHeader& hdr,int sz, bool TruncateNames) const;
456 /// @brief Maps archive into memory
457 bool mapToMemory(std::string* ErrMsg);
459 /// @brief Frees all the members and unmaps the archive file.
460 void cleanUpMemory();
462 /// This type is used to keep track of bitcode modules loaded from the
463 /// symbol table. It maps the file offset to a pair that consists of the
464 /// associated ArchiveMember and the Module.
465 /// @brief Module mapping type
466 typedef std::map<unsigned,std::pair<Module*,ArchiveMember*> >
474 std::string archPath; ///< Path to the archive file we read/write
475 MembersList members; ///< The ilist of ArchiveMember
476 MemoryBuffer *mapfile; ///< Raw Archive contents mapped into memory
477 const char* base; ///< Base of the memory mapped file data
478 SymTabType symTab; ///< The symbol table
479 std::string strtab; ///< The string table for long file names
480 unsigned symTabSize; ///< Size in bytes of symbol table
481 unsigned firstFileOffset; ///< Offset to first normal file.
482 ModuleMap modules; ///< The modules loaded via symbol lookup.
483 ArchiveMember* foreignST; ///< This holds the foreign symbol table.
484 LLVMContext& Context; ///< This holds global data.
489 Archive() LLVM_DELETED_FUNCTION;
490 Archive(const Archive&) LLVM_DELETED_FUNCTION;
491 Archive& operator=(const Archive&) LLVM_DELETED_FUNCTION;
495 } // End llvm namespace