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"
31 // Forward declare classes
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 SVR4SymbolTableFlag = 1, ///< Member is a SVR4 symbol table
53 BSD4SymbolTableFlag = 2, ///< Member is a BSD4 symbol table
54 BitcodeFlag = 4, ///< Member is bitcode
55 HasPathFlag = 8, ///< Member has a full or partial path
56 HasLongFilenameFlag = 16, ///< Member uses the long filename syntax
57 StringTableFlag = 32 ///< 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 StringRef 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 User; }
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 Group; }
84 /// The "mode" specifies the access permissions for the file per Unix
85 /// security. This may not have any applicability 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 Mode; }
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 ModTime; }
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 Size; }
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 char* getData() const { return data; }
112 /// @returns true iff the member is a SVR4 (non-LLVM) symbol table
113 /// @brief Determine if this member is a SVR4 symbol table.
114 bool isSVR4SymbolTable() const { return flags&SVR4SymbolTableFlag; }
116 /// @returns true iff the member is a BSD4.4 (non-LLVM) symbol table
117 /// @brief Determine if this member is a BSD4.4 symbol table.
118 bool isBSD4SymbolTable() const { return flags&BSD4SymbolTableFlag; }
120 /// @returns true iff the archive member is the ar(1) string table
121 /// @brief Determine if this member is the ar(1) string table.
122 bool isStringTable() const { return flags&StringTableFlag; }
124 /// @returns true iff the archive member is a bitcode file.
125 /// @brief Determine if this member is a bitcode file.
126 bool isBitcode() const { return flags&BitcodeFlag; }
128 /// @returns true iff the file name contains a path (directory) component.
129 /// @brief Determine if the member has a path
130 bool hasPath() const { return flags&HasPathFlag; }
132 /// Long filenames are an artifact of the ar(1) file format which allows
133 /// up to sixteen characters in its header and doesn't allow a path
134 /// separator character (/). To avoid this, a "long format" member name is
135 /// allowed that doesn't have this restriction. This method determines if
136 /// that "long format" is used for this member.
137 /// @returns true iff the file name uses the long form
138 /// @brief Determine if the member has a long file name
139 bool hasLongFilename() const { return flags&HasLongFilenameFlag; }
141 /// This method causes the archive member to be replaced with the contents
142 /// of the file specified by \p File. The contents of \p this will be
143 /// updated to reflect the new data from \p File. The \p File must exist and
144 /// be readable on entry to this method.
145 /// @returns true if an error occurred, false otherwise
146 /// @brief Replace contents of archive member with a new file.
147 bool replaceWith(StringRef aFile, std::string* ErrMsg);
153 Archive *parent; ///< Pointer to parent archive
154 std::string path; ///< Path of file containing the member
158 sys::TimeValue ModTime;
160 unsigned flags; ///< Flags about the archive member
161 const char *data; ///< Data for the member
164 /// @name Constructors
167 /// The default constructor is only used by the Archive's iplist when it
168 /// constructs the list's sentry node.
172 /// Used internally by the Archive class to construct an ArchiveMember.
173 /// The contents of the ArchiveMember are filled out by the Archive class.
174 explicit ArchiveMember(Archive *PAR);
176 // So Archive can construct an ArchiveMember
177 friend class llvm::Archive;
181 /// This class defines the interface to LLVM Archive files. The Archive class
182 /// presents the archive file as an ilist of ArchiveMember objects. The members
183 /// can be rearranged in any fashion either by directly editing the ilist or by
184 /// using editing methods on the Archive class (recommended). The Archive
185 /// class also provides several ways of accessing the archive file for various
186 /// purposes such as editing and linking. Full symbol table support is provided
187 /// for loading only those files that resolve symbols. Note that read
188 /// performance of this library is _crucial_ for performance of JIT type
189 /// applications and the linkers. Consequently, the implementation of the class
190 /// is optimized for reading.
196 /// This is the ilist type over which users may iterate to examine
197 /// the contents of the archive
198 /// @brief The ilist type of ArchiveMembers that Archive contains.
199 typedef iplist<ArchiveMember> MembersList;
201 /// @brief Forward mutable iterator over ArchiveMember
202 typedef MembersList::iterator iterator;
204 /// @brief Forward immutable iterator over ArchiveMember
205 typedef MembersList::const_iterator const_iterator;
207 /// @brief Reverse mutable iterator over ArchiveMember
208 typedef std::reverse_iterator<iterator> reverse_iterator;
210 /// @brief Reverse immutable iterator over ArchiveMember
211 typedef std::reverse_iterator<const_iterator> const_reverse_iterator;
213 /// @brief The in-memory version of the symbol table
214 typedef std::map<std::string,unsigned> SymTabType;
217 /// @name ilist accessor methods
220 inline iterator begin() { return members.begin(); }
221 inline const_iterator begin() const { return members.begin(); }
222 inline iterator end () { return members.end(); }
223 inline const_iterator end () const { return members.end(); }
225 inline reverse_iterator rbegin() { return members.rbegin(); }
226 inline const_reverse_iterator rbegin() const { return members.rbegin(); }
227 inline reverse_iterator rend () { return members.rend(); }
228 inline const_reverse_iterator rend () const { return members.rend(); }
230 inline size_t size() const { return members.size(); }
231 inline bool empty() const { return members.empty(); }
232 inline const ArchiveMember& front() const { return members.front(); }
233 inline ArchiveMember& front() { return members.front(); }
234 inline const ArchiveMember& back() const { return members.back(); }
235 inline ArchiveMember& back() { return members.back(); }
238 /// @name ilist mutator methods
241 /// This method splices a \p src member from an archive (possibly \p this),
242 /// to a position just before the member given by \p dest in \p this. When
243 /// the archive is written, \p src will be written in its new location.
244 /// @brief Move a member to a new location
245 inline void splice(iterator dest, Archive& arch, iterator src)
246 { return members.splice(dest,arch.members,src); }
248 /// This method erases a \p target member from the archive. When the
249 /// archive is written, it will no longer contain \p target. The associated
250 /// ArchiveMember is deleted.
251 /// @brief Erase a member.
252 inline iterator erase(iterator target) { return members.erase(target); }
255 /// @name Constructors
258 /// Create an empty archive file and associate it with the \p Filename. This
259 /// method does not actually create the archive disk file. It creates an
260 /// empty Archive object. If the writeToDisk method is called, the archive
261 /// file \p Filename will be created at that point, with whatever content
262 /// the returned Archive object has at that time.
263 /// @returns An Archive* that represents the new archive file.
264 /// @brief Create an empty Archive.
265 static Archive *CreateEmpty(
266 StringRef Filename, ///< Name of the archive to (eventually) create.
267 LLVMContext &C ///< Context to use for global information
270 /// Open an existing archive and load its contents in preparation for
271 /// editing. After this call, the member ilist is completely populated based
272 /// on the contents of the archive file. You should use this form of open if
273 /// you intend to modify the archive or traverse its contents (e.g. for
275 /// @brief Open and load an archive file
276 static Archive *OpenAndLoad(
277 StringRef filePath, ///< The file path to open and load
278 LLVMContext &C, ///< The context to use for global information
279 std::string *ErrorMessage ///< An optional error string
282 /// This destructor cleans up the Archive object, releases all memory, and
283 /// closes files. It does nothing with the archive file on disk. If you
284 /// haven't used the writeToDisk method by the time the destructor is
285 /// called, all changes to the archive will be lost.
286 /// @brief Destruct in-memory archive
293 /// @returns the path to the archive file.
294 /// @brief Get the archive path.
295 StringRef getPath() { return archPath; }
297 /// This method is provided so that editing methods can be invoked directly
298 /// on the Archive's iplist of ArchiveMember. However, it is recommended
299 /// that the usual STL style iterator interface be used instead.
300 /// @returns the iplist of ArchiveMember
301 /// @brief Get the iplist of the members
302 MembersList& getMembers() { return members; }
304 /// This method allows direct query of the Archive's symbol table. The
305 /// symbol table is a std::map of std::string (the symbol) to unsigned (the
306 /// file offset). Note that for efficiency reasons, the offset stored in
307 /// the symbol table is not the actual offset. It is the offset from the
308 /// beginning of the first "real" file member (after the symbol table). Use
309 /// the getFirstFileOffset() to obtain that offset and add this value to the
310 /// offset in the symbol table to obtain the real file offset. Note that
311 /// there is purposefully no interface provided by Archive to look up
312 /// members by their offset. Use the findModulesDefiningSymbols and
313 /// findModuleDefiningSymbol methods instead.
314 /// @returns the Archive's symbol table.
315 /// @brief Get the archive's symbol table
316 const SymTabType& getSymbolTable() { return symTab; }
318 /// This method returns the offset in the archive file to the first "real"
319 /// file member. Archive files, on disk, have a signature and might have a
320 /// symbol table that precedes the first actual file member. This method
321 /// allows you to determine what the size of those fields are.
322 /// @returns the offset to the first "real" file member in the archive.
323 /// @brief Get the offset to the first "real" file member in the archive.
324 unsigned getFirstFileOffset() { return firstFileOffset; }
330 /// This method is the only way to get the archive written to disk. It
331 /// creates or overwrites the file specified when \p this was created
332 /// or opened. The arguments provide options for writing the archive. If
333 /// \p CreateSymbolTable is true, the archive is scanned for bitcode files
334 /// and a symbol table of the externally visible function and global
335 /// variable names is created. If \p TruncateNames is true, the names of the
336 /// archive members will have their path component stripped and the file
337 /// name will be truncated at 15 characters. If \p Compress is specified,
338 /// all archive members will be compressed before being written. If
339 /// \p PrintSymTab is true, the symbol table will be printed to std::cout.
340 /// @returns true if an error occurred, \p error set to error message;
341 /// returns false if the writing succeeded.
342 /// @brief Write (possibly modified) archive contents to disk
344 bool CreateSymbolTable=false, ///< Create Symbol table
345 bool TruncateNames=false, ///< Truncate the filename to 15 chars
346 std::string* ErrMessage=0 ///< If non-null, where error msg is set
349 /// This method adds a new file to the archive. The \p filename is examined
350 /// to determine just enough information to create an ArchiveMember object
351 /// which is then inserted into the Archive object's ilist at the location
352 /// given by \p where.
353 /// @returns true if an error occurred, false otherwise
354 /// @brief Add a file to the archive.
355 bool addFileBefore(StringRef filename, ///< The file to be added
356 iterator where, ///< Insertion point
357 std::string *ErrMsg ///< Optional error message location
361 /// @name Implementation
364 /// @brief Construct an Archive for \p filename and optionally map it
366 explicit Archive(StringRef filename, LLVMContext& C);
368 /// @returns A fully populated ArchiveMember or 0 if an error occurred.
369 /// @brief Parse the header of a member starting at \p At
370 ArchiveMember* parseMemberHeader(
371 const char*&At, ///< The pointer to the location we're parsing
372 const char*End, ///< The pointer to the end of the archive
373 std::string* error ///< Optional error message catcher
376 /// @param ErrMessage Set to address of a std::string to get error messages
377 /// @returns false on error
378 /// @brief Check that the archive signature is correct
379 bool checkSignature(std::string* ErrMessage);
381 /// @param ErrMessage Set to address of a std::string to get error messages
382 /// @returns false on error
383 /// @brief Load the entire archive.
384 bool loadArchive(std::string* ErrMessage);
386 /// @param ErrMessage Set to address of a std::string to get error messages
387 /// @returns false on error
388 /// @brief Load just the symbol table.
389 bool loadSymbolTable(std::string* ErrMessage);
391 /// Writes one ArchiveMember to an ofstream. If an error occurs, returns
392 /// false, otherwise true. If an error occurs and error is non-null then
393 /// it will be set to an error message.
394 /// @returns false if writing member succeeded,
395 /// returns true if writing member failed, \p error set to error message.
397 const ArchiveMember& member, ///< The member to be written
398 std::ofstream& ARFile, ///< The file to write member onto
399 bool CreateSymbolTable, ///< Should symbol table be created?
400 bool TruncateNames, ///< Should names be truncated to 11 chars?
401 std::string* ErrMessage ///< If non-null, place were error msg is set
404 /// @brief Fill in an ArchiveMemberHeader from ArchiveMember.
405 bool fillHeader(const ArchiveMember&mbr,
406 ArchiveMemberHeader& hdr,int sz, bool TruncateNames) const;
408 /// @brief Maps archive into memory
409 bool mapToMemory(std::string* ErrMsg);
411 /// @brief Frees all the members and unmaps the archive file.
412 void cleanUpMemory();
414 /// This type is used to keep track of bitcode modules loaded from the
415 /// symbol table. It maps the file offset to a pair that consists of the
416 /// associated ArchiveMember and the Module.
417 /// @brief Module mapping type
418 typedef std::map<unsigned,std::pair<Module*,ArchiveMember*> >
426 std::string archPath; ///< Path to the archive file we read/write
427 MembersList members; ///< The ilist of ArchiveMember
428 MemoryBuffer *mapfile; ///< Raw Archive contents mapped into memory
429 const char* base; ///< Base of the memory mapped file data
430 SymTabType symTab; ///< The symbol table
431 std::string strtab; ///< The string table for long file names
432 unsigned symTabSize; ///< Size in bytes of symbol table
433 unsigned firstFileOffset; ///< Offset to first normal file.
434 ModuleMap modules; ///< The modules loaded via symbol lookup.
435 ArchiveMember* foreignST; ///< This holds the foreign symbol table.
436 LLVMContext& Context; ///< This holds global data.
441 Archive() LLVM_DELETED_FUNCTION;
442 Archive(const Archive&) LLVM_DELETED_FUNCTION;
443 Archive& operator=(const Archive&) LLVM_DELETED_FUNCTION;
447 } // End llvm namespace