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/PathV1.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 info.getUser(); }
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 info.getGroup(); }
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 info.getMode(); }
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 info.getTimestamp(); }
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 info.getSize(); }
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 returns the status info (like Unix stat(2)) for the archive
141 /// member. The status info provides the file's size, permissions, and
142 /// modification time. The contents of the Path::StatusInfo structure, other
143 /// than the size and modification time, may not have utility on non-Unix
145 /// @returns the status info for the archive member
146 /// @brief Obtain the status info for the archive member
147 const sys::FileStatus &getFileStatus() const { return info; }
149 /// This method causes the archive member to be replaced with the contents
150 /// of the file specified by \p File. The contents of \p this will be
151 /// updated to reflect the new data from \p File. The \p File must exist and
152 /// be readable on entry to this method.
153 /// @returns true if an error occurred, false otherwise
154 /// @brief Replace contents of archive member with a new file.
155 bool replaceWith(const sys::Path &aFile, std::string* ErrMsg);
161 Archive* parent; ///< Pointer to parent archive
162 std::string path; ///< Path of file containing the member
163 sys::FileStatus info; ///< Status info (size,mode,date)
164 unsigned flags; ///< Flags about the archive member
165 const char* data; ///< Data for the member
168 /// @name Constructors
171 /// The default constructor is only used by the Archive's iplist when it
172 /// constructs the list's sentry node.
176 /// Used internally by the Archive class to construct an ArchiveMember.
177 /// The contents of the ArchiveMember are filled out by the Archive class.
178 explicit ArchiveMember(Archive *PAR);
180 // So Archive can construct an ArchiveMember
181 friend class llvm::Archive;
185 /// This class defines the interface to LLVM Archive files. The Archive class
186 /// presents the archive file as an ilist of ArchiveMember objects. The members
187 /// can be rearranged in any fashion either by directly editing the ilist or by
188 /// using editing methods on the Archive class (recommended). The Archive
189 /// class also provides several ways of accessing the archive file for various
190 /// purposes such as editing and linking. Full symbol table support is provided
191 /// for loading only those files that resolve symbols. Note that read
192 /// performance of this library is _crucial_ for performance of JIT type
193 /// applications and the linkers. Consequently, the implementation of the class
194 /// is optimized for reading.
200 /// This is the ilist type over which users may iterate to examine
201 /// the contents of the archive
202 /// @brief The ilist type of ArchiveMembers that Archive contains.
203 typedef iplist<ArchiveMember> MembersList;
205 /// @brief Forward mutable iterator over ArchiveMember
206 typedef MembersList::iterator iterator;
208 /// @brief Forward immutable iterator over ArchiveMember
209 typedef MembersList::const_iterator const_iterator;
211 /// @brief Reverse mutable iterator over ArchiveMember
212 typedef std::reverse_iterator<iterator> reverse_iterator;
214 /// @brief Reverse immutable iterator over ArchiveMember
215 typedef std::reverse_iterator<const_iterator> const_reverse_iterator;
217 /// @brief The in-memory version of the symbol table
218 typedef std::map<std::string,unsigned> SymTabType;
221 /// @name ilist accessor methods
224 inline iterator begin() { return members.begin(); }
225 inline const_iterator begin() const { return members.begin(); }
226 inline iterator end () { return members.end(); }
227 inline const_iterator end () const { return members.end(); }
229 inline reverse_iterator rbegin() { return members.rbegin(); }
230 inline const_reverse_iterator rbegin() const { return members.rbegin(); }
231 inline reverse_iterator rend () { return members.rend(); }
232 inline const_reverse_iterator rend () const { return members.rend(); }
234 inline size_t size() const { return members.size(); }
235 inline bool empty() const { return members.empty(); }
236 inline const ArchiveMember& front() const { return members.front(); }
237 inline ArchiveMember& front() { return members.front(); }
238 inline const ArchiveMember& back() const { return members.back(); }
239 inline ArchiveMember& back() { return members.back(); }
242 /// @name ilist mutator methods
245 /// This method splices a \p src member from an archive (possibly \p this),
246 /// to a position just before the member given by \p dest in \p this. When
247 /// the archive is written, \p src will be written in its new location.
248 /// @brief Move a member to a new location
249 inline void splice(iterator dest, Archive& arch, iterator src)
250 { return members.splice(dest,arch.members,src); }
252 /// This method erases a \p target member from the archive. When the
253 /// archive is written, it will no longer contain \p target. The associated
254 /// ArchiveMember is deleted.
255 /// @brief Erase a member.
256 inline iterator erase(iterator target) { return members.erase(target); }
259 /// @name Constructors
262 /// Create an empty archive file and associate it with the \p Filename. This
263 /// method does not actually create the archive disk file. It creates an
264 /// empty Archive object. If the writeToDisk method is called, the archive
265 /// file \p Filename will be created at that point, with whatever content
266 /// the returned Archive object has at that time.
267 /// @returns An Archive* that represents the new archive file.
268 /// @brief Create an empty Archive.
269 static Archive* CreateEmpty(
270 const sys::Path& Filename,///< Name of the archive to (eventually) create.
271 LLVMContext& C ///< Context to use for global information
274 /// Open an existing archive and load its contents in preparation for
275 /// editing. After this call, the member ilist is completely populated based
276 /// on the contents of the archive file. You should use this form of open if
277 /// you intend to modify the archive or traverse its contents (e.g. for
279 /// @brief Open and load an archive file
280 static Archive* OpenAndLoad(
281 const sys::Path& filePath, ///< The file path to open and load
282 LLVMContext& C, ///< The context to use for global information
283 std::string* ErrorMessage ///< An optional error string
286 /// This destructor cleans up the Archive object, releases all memory, and
287 /// closes files. It does nothing with the archive file on disk. If you
288 /// haven't used the writeToDisk method by the time the destructor is
289 /// called, all changes to the archive will be lost.
290 /// @brief Destruct in-memory archive
297 /// @returns the path to the archive file.
298 /// @brief Get the archive path.
299 const sys::Path& getPath() { return archPath; }
301 /// This method is provided so that editing methods can be invoked directly
302 /// on the Archive's iplist of ArchiveMember. However, it is recommended
303 /// that the usual STL style iterator interface be used instead.
304 /// @returns the iplist of ArchiveMember
305 /// @brief Get the iplist of the members
306 MembersList& getMembers() { return members; }
308 /// This method allows direct query of the Archive's symbol table. The
309 /// symbol table is a std::map of std::string (the symbol) to unsigned (the
310 /// file offset). Note that for efficiency reasons, the offset stored in
311 /// the symbol table is not the actual offset. It is the offset from the
312 /// beginning of the first "real" file member (after the symbol table). Use
313 /// the getFirstFileOffset() to obtain that offset and add this value to the
314 /// offset in the symbol table to obtain the real file offset. Note that
315 /// there is purposefully no interface provided by Archive to look up
316 /// members by their offset. Use the findModulesDefiningSymbols and
317 /// findModuleDefiningSymbol methods instead.
318 /// @returns the Archive's symbol table.
319 /// @brief Get the archive's symbol table
320 const SymTabType& getSymbolTable() { return symTab; }
322 /// This method returns the offset in the archive file to the first "real"
323 /// file member. Archive files, on disk, have a signature and might have a
324 /// symbol table that precedes the first actual file member. This method
325 /// allows you to determine what the size of those fields are.
326 /// @returns the offset to the first "real" file member in the archive.
327 /// @brief Get the offset to the first "real" file member in the archive.
328 unsigned getFirstFileOffset() { return firstFileOffset; }
330 /// This method will scan the archive for bitcode modules, interpret them
331 /// and return a vector of the instantiated modules in \p Modules. If an
332 /// error occurs, this method will return true. If \p ErrMessage is not null
333 /// and an error occurs, \p *ErrMessage will be set to a string explaining
334 /// the error that occurred.
335 /// @returns true if an error occurred
336 /// @brief Instantiate all the bitcode modules located in the archive
337 bool getAllModules(std::vector<Module*>& Modules, std::string* ErrMessage);
339 /// This accessor looks up the \p symbol in the archive's symbol table and
340 /// returns the associated module that defines that symbol. This method can
341 /// be called as many times as necessary. This is handy for linking the
342 /// archive into another module based on unresolved symbols. Note that the
343 /// Module returned by this accessor should not be deleted by the caller. It
344 /// is managed internally by the Archive class. It is possible that multiple
345 /// calls to this accessor will return the same Module instance because the
346 /// associated module defines multiple symbols.
347 /// @returns The Module* found or null if the archive does not contain a
348 /// module that defines the \p symbol.
349 /// @brief Look up a module by symbol name.
350 Module* findModuleDefiningSymbol(
351 const std::string& symbol, ///< Symbol to be sought
352 std::string* ErrMessage ///< Error message storage, if non-zero
355 /// This method is similar to findModuleDefiningSymbol but allows lookup of
356 /// more than one symbol at a time. If \p symbols contains a list of
357 /// undefined symbols in some module, then calling this method is like
358 /// making one complete pass through the archive to resolve symbols but is
359 /// more efficient than looking at the individual members. Note that on
360 /// exit, the symbols resolved by this method will be removed from \p
361 /// symbols to ensure they are not re-searched on a subsequent call. If
362 /// you need to retain the list of symbols, make a copy.
363 /// @brief Look up multiple symbols in the archive.
364 bool findModulesDefiningSymbols(
365 std::set<std::string>& symbols, ///< Symbols to be sought
366 SmallVectorImpl<Module*>& modules, ///< The modules matching \p symbols
367 std::string* ErrMessage ///< Error msg storage, if non-zero
370 /// This method determines whether the archive is a properly formed llvm
371 /// bitcode archive. It first makes sure the symbol table has been loaded
372 /// and has a non-zero size. If it does, then it is an archive. If not,
373 /// then it tries to load all the bitcode modules of the archive. Finally,
374 /// it returns whether it was successful.
375 /// @returns true if the archive is a proper llvm bitcode archive
376 /// @brief Determine whether the archive is a proper llvm bitcode archive.
377 bool isBitcodeArchive();
383 /// This method is the only way to get the archive written to disk. It
384 /// creates or overwrites the file specified when \p this was created
385 /// or opened. The arguments provide options for writing the archive. If
386 /// \p CreateSymbolTable is true, the archive is scanned for bitcode files
387 /// and a symbol table of the externally visible function and global
388 /// variable names is created. If \p TruncateNames is true, the names of the
389 /// archive members will have their path component stripped and the file
390 /// name will be truncated at 15 characters. If \p Compress is specified,
391 /// all archive members will be compressed before being written. If
392 /// \p PrintSymTab is true, the symbol table will be printed to std::cout.
393 /// @returns true if an error occurred, \p error set to error message;
394 /// returns false if the writing succeeded.
395 /// @brief Write (possibly modified) archive contents to disk
397 bool CreateSymbolTable=false, ///< Create Symbol table
398 bool TruncateNames=false, ///< Truncate the filename to 15 chars
399 std::string* ErrMessage=0 ///< If non-null, where error msg is set
402 /// This method adds a new file to the archive. The \p filename is examined
403 /// to determine just enough information to create an ArchiveMember object
404 /// which is then inserted into the Archive object's ilist at the location
405 /// given by \p where.
406 /// @returns true if an error occurred, false otherwise
407 /// @brief Add a file to the archive.
409 const sys::Path& filename, ///< The file to be added
410 iterator where, ///< Insertion point
411 std::string* ErrMsg ///< Optional error message location
415 /// @name Implementation
418 /// @brief Construct an Archive for \p filename and optionally map it
420 explicit Archive(const sys::Path& filename, LLVMContext& C);
422 /// @returns A fully populated ArchiveMember or 0 if an error occurred.
423 /// @brief Parse the header of a member starting at \p At
424 ArchiveMember* parseMemberHeader(
425 const char*&At, ///< The pointer to the location we're parsing
426 const char*End, ///< The pointer to the end of the archive
427 std::string* error ///< Optional error message catcher
430 /// @param ErrMessage Set to address of a std::string to get error messages
431 /// @returns false on error
432 /// @brief Check that the archive signature is correct
433 bool checkSignature(std::string* ErrMessage);
435 /// @param ErrMessage Set to address of a std::string to get error messages
436 /// @returns false on error
437 /// @brief Load the entire archive.
438 bool loadArchive(std::string* ErrMessage);
440 /// @param ErrMessage Set to address of a std::string to get error messages
441 /// @returns false on error
442 /// @brief Load just the symbol table.
443 bool loadSymbolTable(std::string* ErrMessage);
445 /// Writes one ArchiveMember to an ofstream. If an error occurs, returns
446 /// false, otherwise true. If an error occurs and error is non-null then
447 /// it will be set to an error message.
448 /// @returns false if writing member succeeded,
449 /// returns true if writing member failed, \p error set to error message.
451 const ArchiveMember& member, ///< The member to be written
452 std::ofstream& ARFile, ///< The file to write member onto
453 bool CreateSymbolTable, ///< Should symbol table be created?
454 bool TruncateNames, ///< Should names be truncated to 11 chars?
455 std::string* ErrMessage ///< If non-null, place were error msg is set
458 /// @brief Fill in an ArchiveMemberHeader from ArchiveMember.
459 bool fillHeader(const ArchiveMember&mbr,
460 ArchiveMemberHeader& hdr,int sz, bool TruncateNames) const;
462 /// @brief Maps archive into memory
463 bool mapToMemory(std::string* ErrMsg);
465 /// @brief Frees all the members and unmaps the archive file.
466 void cleanUpMemory();
468 /// This type is used to keep track of bitcode modules loaded from the
469 /// symbol table. It maps the file offset to a pair that consists of the
470 /// associated ArchiveMember and the Module.
471 /// @brief Module mapping type
472 typedef std::map<unsigned,std::pair<Module*,ArchiveMember*> >
480 sys::Path archPath; ///< Path to the archive file we read/write
481 MembersList members; ///< The ilist of ArchiveMember
482 MemoryBuffer *mapfile; ///< Raw Archive contents mapped into memory
483 const char* base; ///< Base of the memory mapped file data
484 SymTabType symTab; ///< The symbol table
485 std::string strtab; ///< The string table for long file names
486 unsigned symTabSize; ///< Size in bytes of symbol table
487 unsigned firstFileOffset; ///< Offset to first normal file.
488 ModuleMap modules; ///< The modules loaded via symbol lookup.
489 ArchiveMember* foreignST; ///< This holds the foreign symbol table.
490 LLVMContext& Context; ///< This holds global data.
495 Archive() LLVM_DELETED_FUNCTION;
496 Archive(const Archive&) LLVM_DELETED_FUNCTION;
497 Archive& operator=(const Archive&) LLVM_DELETED_FUNCTION;
501 } // End llvm namespace