diff options
Diffstat (limited to 'tools/llvm-ar')
-rw-r--r-- | tools/llvm-ar/Archive.cpp | 254 | ||||
-rw-r--r-- | tools/llvm-ar/Archive.h | 524 | ||||
-rw-r--r-- | tools/llvm-ar/ArchiveInternals.h | 87 | ||||
-rw-r--r-- | tools/llvm-ar/ArchiveReader.cpp | 557 | ||||
-rw-r--r-- | tools/llvm-ar/ArchiveWriter.cpp | 429 | ||||
-rw-r--r-- | tools/llvm-ar/CMakeLists.txt | 5 | ||||
-rw-r--r-- | tools/llvm-ar/LLVMBuild.txt | 1 | ||||
-rw-r--r-- | tools/llvm-ar/Makefile | 2 | ||||
-rw-r--r-- | tools/llvm-ar/llvm-ar.cpp | 2 |
9 files changed, 1857 insertions, 4 deletions
diff --git a/tools/llvm-ar/Archive.cpp b/tools/llvm-ar/Archive.cpp new file mode 100644 index 0000000000..3e9b9b2c25 --- /dev/null +++ b/tools/llvm-ar/Archive.cpp @@ -0,0 +1,254 @@ +//===-- Archive.cpp - Generic LLVM archive functions ------------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file contains the implementation of the Archive and ArchiveMember +// classes that is common to both reading and writing archives.. +// +//===----------------------------------------------------------------------===// + +#include "Archive.h" +#include "ArchiveInternals.h" +#include "llvm/Bitcode/ReaderWriter.h" +#include "llvm/IR/Module.h" +#include "llvm/Support/FileSystem.h" +#include "llvm/Support/MemoryBuffer.h" +#include "llvm/Support/Process.h" +#include "llvm/Support/system_error.h" +#include <cstring> +#include <memory> +using namespace llvm; + +// getMemberSize - compute the actual physical size of the file member as seen +// on disk. This isn't the size of member's payload. Use getSize() for that. +unsigned +ArchiveMember::getMemberSize() const { + // Basically its the file size plus the header size + unsigned result = info.fileSize + sizeof(ArchiveMemberHeader); + + // If it has a long filename, include the name length + if (hasLongFilename()) + result += path.str().length() + 1; + + // If its now odd lengthed, include the padding byte + if (result % 2 != 0 ) + result++; + + return result; +} + +// This default constructor is only use by the ilist when it creates its +// sentry node. We give it specific static values to make it stand out a bit. +ArchiveMember::ArchiveMember() + : parent(0), path("--invalid--"), flags(0), data(0) +{ + info.user = sys::Process::GetCurrentUserId(); + info.group = sys::Process::GetCurrentGroupId(); + info.mode = 0777; + info.fileSize = 0; + info.modTime = sys::TimeValue::now(); +} + +// This is the constructor that the Archive class uses when it is building or +// reading an archive. It just defaults a few things and ensures the parent is +// set for the iplist. The Archive class fills in the ArchiveMember's data. +// This is required because correctly setting the data may depend on other +// things in the Archive. +ArchiveMember::ArchiveMember(Archive* PAR) + : parent(PAR), path(), flags(0), data(0) +{ +} + +// This method allows an ArchiveMember to be replaced with the data for a +// different file, presumably as an update to the member. It also makes sure +// the flags are reset correctly. +bool ArchiveMember::replaceWith(const sys::Path& newFile, std::string* ErrMsg) { + bool Exists; + if (sys::fs::exists(newFile.str(), Exists) || !Exists) { + if (ErrMsg) + *ErrMsg = "Can not replace an archive member with a non-existent file"; + return true; + } + + data = 0; + path = newFile; + + // SVR4 symbol tables have an empty name + if (path.str() == ARFILE_SVR4_SYMTAB_NAME) + flags |= SVR4SymbolTableFlag; + else + flags &= ~SVR4SymbolTableFlag; + + // BSD4.4 symbol tables have a special name + if (path.str() == ARFILE_BSD4_SYMTAB_NAME) + flags |= BSD4SymbolTableFlag; + else + flags &= ~BSD4SymbolTableFlag; + + // String table name + if (path.str() == ARFILE_STRTAB_NAME) + flags |= StringTableFlag; + else + flags &= ~StringTableFlag; + + // If it has a slash then it has a path + bool hasSlash = path.str().find('/') != std::string::npos; + if (hasSlash) + flags |= HasPathFlag; + else + flags &= ~HasPathFlag; + + // If it has a slash or its over 15 chars then its a long filename format + if (hasSlash || path.str().length() > 15) + flags |= HasLongFilenameFlag; + else + flags &= ~HasLongFilenameFlag; + + // Get the signature and status info + const char* signature = (const char*) data; + SmallString<4> magic; + if (!signature) { + sys::fs::get_magic(path.str(), magic.capacity(), magic); + signature = magic.c_str(); + const sys::FileStatus *FSinfo = path.getFileStatus(false, ErrMsg); + if (FSinfo) + info = *FSinfo; + else + return true; + } + + // Determine what kind of file it is. + if (sys::fs::identify_magic(StringRef(signature, 4)) == + sys::fs::file_magic::bitcode) + flags |= BitcodeFlag; + else + flags &= ~BitcodeFlag; + + return false; +} + +// Archive constructor - this is the only constructor that gets used for the +// Archive class. Everything else (default,copy) is deprecated. This just +// initializes and maps the file into memory, if requested. +Archive::Archive(const sys::Path& filename, LLVMContext& C) + : archPath(filename), members(), mapfile(0), base(0), symTab(), strtab(), + symTabSize(0), firstFileOffset(0), modules(), foreignST(0), Context(C) { +} + +bool +Archive::mapToMemory(std::string* ErrMsg) { + OwningPtr<MemoryBuffer> File; + if (error_code ec = MemoryBuffer::getFile(archPath.c_str(), File)) { + if (ErrMsg) + *ErrMsg = ec.message(); + return true; + } + mapfile = File.take(); + base = mapfile->getBufferStart(); + return false; +} + +void Archive::cleanUpMemory() { + // Shutdown the file mapping + delete mapfile; + mapfile = 0; + base = 0; + + // Forget the entire symbol table + symTab.clear(); + symTabSize = 0; + + firstFileOffset = 0; + + // Free the foreign symbol table member + if (foreignST) { + delete foreignST; + foreignST = 0; + } + + // Delete any Modules and ArchiveMember's we've allocated as a result of + // symbol table searches. + for (ModuleMap::iterator I=modules.begin(), E=modules.end(); I != E; ++I ) { + delete I->second.first; + delete I->second.second; + } +} + +// Archive destructor - just clean up memory +Archive::~Archive() { + cleanUpMemory(); +} + + + +static void getSymbols(Module*M, std::vector<std::string>& symbols) { + // Loop over global variables + for (Module::global_iterator GI = M->global_begin(), GE=M->global_end(); GI != GE; ++GI) + if (!GI->isDeclaration() && !GI->hasLocalLinkage()) + if (!GI->getName().empty()) + symbols.push_back(GI->getName()); + + // Loop over functions + for (Module::iterator FI = M->begin(), FE = M->end(); FI != FE; ++FI) + if (!FI->isDeclaration() && !FI->hasLocalLinkage()) + if (!FI->getName().empty()) + symbols.push_back(FI->getName()); + + // Loop over aliases + for (Module::alias_iterator AI = M->alias_begin(), AE = M->alias_end(); + AI != AE; ++AI) { + if (AI->hasName()) + symbols.push_back(AI->getName()); + } +} + +// Get just the externally visible defined symbols from the bitcode +bool llvm::GetBitcodeSymbols(const sys::Path& fName, + LLVMContext& Context, + std::vector<std::string>& symbols, + std::string* ErrMsg) { + OwningPtr<MemoryBuffer> Buffer; + if (error_code ec = MemoryBuffer::getFileOrSTDIN(fName.c_str(), Buffer)) { + if (ErrMsg) *ErrMsg = "Could not open file '" + fName.str() + "'" + ": " + + ec.message(); + return true; + } + + Module *M = ParseBitcodeFile(Buffer.get(), Context, ErrMsg); + if (!M) + return true; + + // Get the symbols + getSymbols(M, symbols); + + // Done with the module. + delete M; + return true; +} + +Module* +llvm::GetBitcodeSymbols(const char *BufPtr, unsigned Length, + const std::string& ModuleID, + LLVMContext& Context, + std::vector<std::string>& symbols, + std::string* ErrMsg) { + // Get the module. + OwningPtr<MemoryBuffer> Buffer( + MemoryBuffer::getMemBufferCopy(StringRef(BufPtr, Length),ModuleID.c_str())); + + Module *M = ParseBitcodeFile(Buffer.get(), Context, ErrMsg); + if (!M) + return 0; + + // Get the symbols + getSymbols(M, symbols); + + // Done with the module. Note that it's the caller's responsibility to delete + // the Module. + return M; +} diff --git a/tools/llvm-ar/Archive.h b/tools/llvm-ar/Archive.h new file mode 100644 index 0000000000..bd76298c42 --- /dev/null +++ b/tools/llvm-ar/Archive.h @@ -0,0 +1,524 @@ +//===-- llvm/Bitcode/Archive.h - LLVM Bitcode Archive -----------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This header file declares the Archive and ArchiveMember classes that provide +// manipulation of LLVM Archive files. The implementation is provided by the +// lib/Bitcode/Archive library. This library is used to read and write +// archive (*.a) files that contain LLVM bitcode files (or others). +// +//===----------------------------------------------------------------------===// + +#ifndef TOOLS_LLVM_AR_ARCHIVE_H +#define TOOLS_LLVM_AR_ARCHIVE_H + +#include "llvm/ADT/ilist.h" +#include "llvm/ADT/ilist_node.h" +#include "llvm/Support/Path.h" +#include "llvm/Support/PathV1.h" +#include <map> +#include <set> + +namespace llvm { + class MemoryBuffer; + +// Forward declare classes +class Module; // From VMCore +class Archive; // Declared below +class ArchiveMemberHeader; // Internal implementation class +class LLVMContext; // Global data + +/// This class is the main class manipulated by users of the Archive class. It +/// holds information about one member of the Archive. It is also the element +/// stored by the Archive's ilist, the Archive's main abstraction. Because of +/// the special requirements of archive files, users are not permitted to +/// construct ArchiveMember instances. You should obtain them from the methods +/// of the Archive class instead. +/// @brief This class represents a single archive member. +class ArchiveMember : public ilist_node<ArchiveMember> { + /// @name Types + /// @{ + public: + /// These flags are used internally by the archive member to specify various + /// characteristics of the member. The various "is" methods below provide + /// access to the flags. The flags are not user settable. + enum Flags { + SVR4SymbolTableFlag = 1, ///< Member is a SVR4 symbol table + BSD4SymbolTableFlag = 2, ///< Member is a BSD4 symbol table + BitcodeFlag = 4, ///< Member is bitcode + HasPathFlag = 8, ///< Member has a full or partial path + HasLongFilenameFlag = 16, ///< Member uses the long filename syntax + StringTableFlag = 32 ///< Member is an ar(1) format string table + }; + + /// @} + /// @name Accessors + /// @{ + public: + /// @returns the parent Archive instance + /// @brief Get the archive associated with this member + Archive* getArchive() const { return parent; } + + /// @returns the path to the Archive's file + /// @brief Get the path to the archive member + const sys::Path& getPath() const { return path; } + + /// The "user" is the owner of the file per Unix security. This may not + /// have any applicability on non-Unix systems but is a required component + /// of the "ar" file format. + /// @brief Get the user associated with this archive member. + unsigned getUser() const { return info.getUser(); } + + /// The "group" is the owning group of the file per Unix security. This + /// may not have any applicability on non-Unix systems but is a required + /// component of the "ar" file format. + /// @brief Get the group associated with this archive member. + unsigned getGroup() const { return info.getGroup(); } + + /// The "mode" specifies the access permissions for the file per Unix + /// security. This may not have any applicability on non-Unix systems but is + /// a required component of the "ar" file format. + /// @brief Get the permission mode associated with this archive member. + unsigned getMode() const { return info.getMode(); } + + /// This method returns the time at which the archive member was last + /// modified when it was not in the archive. + /// @brief Get the time of last modification of the archive member. + sys::TimeValue getModTime() const { return info.getTimestamp(); } + + /// @returns the size of the archive member in bytes. + /// @brief Get the size of the archive member. + uint64_t getSize() const { return info.getSize(); } + + /// This method returns the total size of the archive member as it + /// appears on disk. This includes the file content, the header, the + /// long file name if any, and the padding. + /// @brief Get total on-disk member size. + unsigned getMemberSize() const; + + /// This method will return a pointer to the in-memory content of the + /// archive member, if it is available. If the data has not been loaded + /// into memory, the return value will be null. + /// @returns a pointer to the member's data. + /// @brief Get the data content of the archive member + const char* getData() const { return data; } + + /// @returns true iff the member is a SVR4 (non-LLVM) symbol table + /// @brief Determine if this member is a SVR4 symbol table. + bool isSVR4SymbolTable() const { return flags&SVR4SymbolTableFlag; } + + /// @returns true iff the member is a BSD4.4 (non-LLVM) symbol table + /// @brief Determine if this member is a BSD4.4 symbol table. + bool isBSD4SymbolTable() const { return flags&BSD4SymbolTableFlag; } + + /// @returns true iff the archive member is the ar(1) string table + /// @brief Determine if this member is the ar(1) string table. + bool isStringTable() const { return flags&StringTableFlag; } + + /// @returns true iff the archive member is a bitcode file. + /// @brief Determine if this member is a bitcode file. + bool isBitcode() const { return flags&BitcodeFlag; } + + /// @returns true iff the file name contains a path (directory) component. + /// @brief Determine if the member has a path + bool hasPath() const { return flags&HasPathFlag; } + + /// Long filenames are an artifact of the ar(1) file format which allows + /// up to sixteen characters in its header and doesn't allow a path + /// separator character (/). To avoid this, a "long format" member name is + /// allowed that doesn't have this restriction. This method determines if + /// that "long format" is used for this member. + /// @returns true iff the file name uses the long form + /// @brief Determine if the member has a long file name + bool hasLongFilename() const { return flags&HasLongFilenameFlag; } + + /// This method returns the status info (like Unix stat(2)) for the archive + /// member. The status info provides the file's size, permissions, and + /// modification time. The contents of the Path::StatusInfo structure, other + /// than the size and modification time, may not have utility on non-Unix + /// systems. + /// @returns the status info for the archive member + /// @brief Obtain the status info for the archive member + const sys::FileStatus &getFileStatus() const { return info; } + + /// This method causes the archive member to be replaced with the contents + /// of the file specified by \p File. The contents of \p this will be + /// updated to reflect the new data from \p File. The \p File must exist and + /// be readable on entry to this method. + /// @returns true if an error occurred, false otherwise + /// @brief Replace contents of archive member with a new file. + bool replaceWith(const sys::Path &aFile, std::string* ErrMsg); + + /// @} + /// @name Data + /// @{ + private: + Archive* parent; ///< Pointer to parent archive + sys::PathWithStatus path; ///< Path of file containing the member + sys::FileStatus info; ///< Status info (size,mode,date) + unsigned flags; ///< Flags about the archive member + const char* data; ///< Data for the member + + /// @} + /// @name Constructors + /// @{ + public: + /// The default constructor is only used by the Archive's iplist when it + /// constructs the list's sentry node. + ArchiveMember(); + + private: + /// Used internally by the Archive class to construct an ArchiveMember. + /// The contents of the ArchiveMember are filled out by the Archive class. + explicit ArchiveMember(Archive *PAR); + + // So Archive can construct an ArchiveMember + friend class llvm::Archive; + /// @} +}; + +/// This class defines the interface to LLVM Archive files. The Archive class +/// presents the archive file as an ilist of ArchiveMember objects. The members +/// can be rearranged in any fashion either by directly editing the ilist or by +/// using editing methods on the Archive class (recommended). The Archive +/// class also provides several ways of accessing the archive file for various +/// purposes such as editing and linking. Full symbol table support is provided +/// for loading only those files that resolve symbols. Note that read +/// performance of this library is _crucial_ for performance of JIT type +/// applications and the linkers. Consequently, the implementation of the class +/// is optimized for reading. +class Archive { + + /// @name Types + /// @{ + public: + /// This is the ilist type over which users may iterate to examine + /// the contents of the archive + /// @brief The ilist type of ArchiveMembers that Archive contains. + typedef iplist<ArchiveMember> MembersList; + + /// @brief Forward mutable iterator over ArchiveMember + typedef MembersList::iterator iterator; + + /// @brief Forward immutable iterator over ArchiveMember + typedef MembersList::const_iterator const_iterator; + + /// @brief Reverse mutable iterator over ArchiveMember + typedef std::reverse_iterator<iterator> reverse_iterator; + + /// @brief Reverse immutable iterator over ArchiveMember + typedef std::reverse_iterator<const_iterator> const_reverse_iterator; + + /// @brief The in-memory version of the symbol table + typedef std::map<std::string,unsigned> SymTabType; + + /// @} + /// @name ilist accessor methods + /// @{ + public: + inline iterator begin() { return members.begin(); } + inline const_iterator begin() const { return members.begin(); } + inline iterator end () { return members.end(); } + inline const_iterator end () const { return members.end(); } + + inline reverse_iterator rbegin() { return members.rbegin(); } + inline const_reverse_iterator rbegin() const { return members.rbegin(); } + inline reverse_iterator rend () { return members.rend(); } + inline const_reverse_iterator rend () const { return members.rend(); } + + inline size_t size() const { return members.size(); } + inline bool empty() const { return members.empty(); } + inline const ArchiveMember& front() const { return members.front(); } + inline ArchiveMember& front() { return members.front(); } + inline const ArchiveMember& back() const { return members.back(); } + inline ArchiveMember& back() { return members.back(); } + + /// @} + /// @name ilist mutator methods + /// @{ + public: + /// This method splices a \p src member from an archive (possibly \p this), + /// to a position just before the member given by \p dest in \p this. When + /// the archive is written, \p src will be written in its new location. + /// @brief Move a member to a new location + inline void splice(iterator dest, Archive& arch, iterator src) + { return members.splice(dest,arch.members,src); } + + /// This method erases a \p target member from the archive. When the + /// archive is written, it will no longer contain \p target. The associated + /// ArchiveMember is deleted. + /// @brief Erase a member. + inline iterator erase(iterator target) { return members.erase(target); } + + /// @} + /// @name Constructors + /// @{ + public: + /// Create an empty archive file and associate it with the \p Filename. This + /// method does not actually create the archive disk file. It creates an + /// empty Archive object. If the writeToDisk method is called, the archive + /// file \p Filename will be created at that point, with whatever content + /// the returned Archive object has at that time. + /// @returns An Archive* that represents the new archive file. + /// @brief Create an empty Archive. + static Archive* CreateEmpty( + const sys::Path& Filename,///< Name of the archive to (eventually) create. + LLVMContext& C ///< Context to use for global information + ); + + /// Open an existing archive and load its contents in preparation for + /// editing. After this call, the member ilist is completely populated based + /// on the contents of the archive file. You should use this form of open if + /// you intend to modify the archive or traverse its contents (e.g. for + /// printing). + /// @brief Open and load an archive file + static Archive* OpenAndLoad( + const sys::Path& filePath, ///< The file path to open and load + LLVMContext& C, ///< The context to use for global information + std::string* ErrorMessage ///< An optional error string + ); + + /// This method opens an existing archive file from \p Filename and reads in + /// its symbol table without reading in any of the archive's members. This + /// reduces both I/O and cpu time in opening the archive if it is to be used + /// solely for symbol lookup (e.g. during linking). The \p Filename must + /// exist and be an archive file or an error will be returned. This form + /// of opening the archive is intended for read-only operations that need to + /// locate members via the symbol table for link editing. Since the archve + /// members are not read by this method, the archive will appear empty upon + /// return. If editing operations are performed on the archive, they will + /// completely replace the contents of the archive! It is recommended that + /// if this form of opening the archive is used that only the symbol table + /// lookup methods (getSymbolTable, findModuleDefiningSymbol, and + /// findModulesDefiningSymbols) be used. + /// @returns an Archive* that represents the archive file, or null on error. + /// @brief Open an existing archive and load its symbols. + static Archive* OpenAndLoadSymbols( + const sys::Path& Filename, ///< Name of the archive file to open + LLVMContext& C, ///< The context to use for global info + std::string* ErrorMessage=0 ///< An optional error string + ); + + /// This destructor cleans up the Archive object, releases all memory, and + /// closes files. It does nothing with the archive file on disk. If you + /// haven't used the writeToDisk method by the time the destructor is + /// called, all changes to the archive will be lost. + /// @brief Destruct in-memory archive + ~Archive(); + + /// @} + /// @name Accessors + /// @{ + public: + /// @returns the path to the archive file. + /// @brief Get the archive path. + const sys::Path& getPath() { return archPath; } + + /// This method is provided so that editing methods can be invoked directly + /// on the Archive's iplist of ArchiveMember. However, it is recommended + /// that the usual STL style iterator interface be used instead. + /// @returns the iplist of ArchiveMember + /// @brief Get the iplist of the members + MembersList& getMembers() { return members; } + + /// This method allows direct query of the Archive's symbol table. The + /// symbol table is a std::map of std::string (the symbol) to unsigned (the + /// file offset). Note that for efficiency reasons, the offset stored in + /// the symbol table is not the actual offset. It is the offset from the + /// beginning of the first "real" file member (after the symbol table). Use + /// the getFirstFileOffset() to obtain that offset and add this value to the + /// offset in the symbol table to obtain the real file offset. Note that + /// there is purposefully no interface provided by Archive to look up + /// members by their offset. Use the findModulesDefiningSymbols and + /// findModuleDefiningSymbol methods instead. + /// @returns the Archive's symbol table. + /// @brief Get the archive's symbol table + const SymTabType& getSymbolTable() { return symTab; } + + /// This method returns the offset in the archive file to the first "real" + /// file member. Archive files, on disk, have a signature and might have a + /// symbol table that precedes the first actual file member. This method + /// allows you to determine what the size of those fields are. + /// @returns the offset to the first "real" file member in the archive. + /// @brief Get the offset to the first "real" file member in the archive. + unsigned getFirstFileOffset() { return firstFileOffset; } + + /// This method will scan the archive for bitcode modules, interpret them + /// and return a vector of the instantiated modules in \p Modules. If an + /// error occurs, this method will return true. If \p ErrMessage is not null + /// and an error occurs, \p *ErrMessage will be set to a string explaining + /// the error that occurred. + /// @returns true if an error occurred + /// @brief Instantiate all the bitcode modules located in the archive + bool getAllModules(std::vector<Module*>& Modules, std::string* ErrMessage); + + /// This accessor looks up the \p symbol in the archive's symbol table and + /// returns the associated module that defines that symbol. This method can + /// be called as many times as necessary. This is handy for linking the + /// archive into another module based on unresolved symbols. Note that the + /// Module returned by this accessor should not be deleted by the caller. It + /// is managed internally by the Archive class. It is possible that multiple + /// calls to this accessor will return the same Module instance because the + /// associated module defines multiple symbols. + /// @returns The Module* found or null if the archive does not contain a + /// module that defines the \p symbol. + /// @brief Look up a module by symbol name. + Module* findModuleDefiningSymbol( + const std::string& symbol, ///< Symbol to be sought + std::string* ErrMessage ///< Error message storage, if non-zero + ); + + /// This method is similar to findModuleDefiningSymbol but allows lookup of + /// more than one symbol at a time. If \p symbols contains a list of + /// undefined symbols in some module, then calling this method is like + /// making one complete pass through the archive to resolve symbols but is + /// more efficient than looking at the individual members. Note that on + /// exit, the symbols resolved by this method will be removed from \p + /// symbols to ensure they are not re-searched on a subsequent call. If + /// you need to retain the list of symbols, make a copy. + /// @brief Look up multiple symbols in the archive. + bool findModulesDefiningSymbols( + std::set<std::string>& symbols, ///< Symbols to be sought + SmallVectorImpl<Module*>& modules, ///< The modules matching \p symbols + std::string* ErrMessage ///< Error msg storage, if non-zero + ); + + /// This method determines whether the archive is a properly formed llvm + /// bitcode archive. It first makes sure the symbol table has been loaded + /// and has a non-zero size. If it does, then it is an archive. If not, + /// then it tries to load all the bitcode modules of the archive. Finally, + /// it returns whether it was successful. + /// @returns true if the archive is a proper llvm bitcode archive + /// @brief Determine whether the archive is a proper llvm bitcode archive. + bool isBitcodeArchive(); + + /// @} + /// @name Mutators + /// @{ + public: + /// This method is the only way to get the archive written to disk. It + /// creates or overwrites the file specified when \p this was created + /// or opened. The arguments provide options for writing the archive. If + /// \p CreateSymbolTable is true, the archive is scanned for bitcode files + /// and a symbol table of the externally visible function and global + /// variable names is created. If \p TruncateNames is true, the names of the + /// archive members will have their path component stripped and the file + /// name will be truncated at 15 characters. If \p Compress is specified, + /// all archive members will be compressed before being written. If + /// \p PrintSymTab is true, the symbol table will be printed to std::cout. + /// @returns true if an error occurred, \p error set to error message; + /// returns false if the writing succeeded. + /// @brief Write (possibly modified) archive contents to disk + bool writeToDisk( + bool CreateSymbolTable=false, ///< Create Symbol table + bool TruncateNames=false, ///< Truncate the filename to 15 chars + std::string* ErrMessage=0 ///< If non-null, where error msg is set + ); + + /// This method adds a new file to the archive. The \p filename is examined + /// to determine just enough information to create an ArchiveMember object + /// which is then inserted into the Archive object's ilist at the location + /// given by \p where. + /// @returns true if an error occurred, false otherwise + /// @brief Add a file to the archive. + bool addFileBefore( + const sys::Path& filename, ///< The file to be added + iterator where, ///< Insertion point + std::string* ErrMsg ///< Optional error message location + ); + + /// @} + /// @name Implementation + /// @{ + protected: + /// @brief Construct an Archive for \p filename and optionally map it + /// into memory. + explicit Archive(const sys::Path& filename, LLVMContext& C); + + /// @returns A fully populated ArchiveMember or 0 if an error occurred. + /// @brief Parse the header of a member starting at \p At + ArchiveMember* parseMemberHeader( + const char*&At, ///< The pointer to the location we're parsing + const char*End, ///< The pointer to the end of the archive + std::string* error ///< Optional error message catcher + ); + + /// @param ErrMessage Set to address of a std::string to get error messages + /// @returns false on error + /// @brief Check that the archive signature is correct + bool checkSignature(std::string* ErrMessage); + + /// @param ErrMessage Set to address of a std::string to get error messages + /// @returns false on error + /// @brief Load the entire archive. + bool loadArchive(std::string* ErrMessage); + + /// @param ErrMessage Set to address of a std::string to get error messages + /// @returns false on error + /// @brief Load just the symbol table. + bool loadSymbolTable(std::string* ErrMessage); + + /// Writes one ArchiveMember to an ofstream. If an error occurs, returns + /// false, otherwise true. If an error occurs and error is non-null then + /// it will be set to an error message. + /// @returns false if writing member succeeded, + /// returns true if writing member failed, \p error set to error message. + bool writeMember( + const ArchiveMember& member, ///< The member to be written + std::ofstream& ARFile, ///< The file to write member onto + bool CreateSymbolTable, ///< Should symbol table be created? + bool TruncateNames, ///< Should names be truncated to 11 chars? + std::string* ErrMessage ///< If non-null, place were error msg is set + ); + + /// @brief Fill in an ArchiveMemberHeader from ArchiveMember. + bool fillHeader(const ArchiveMember&mbr, + ArchiveMemberHeader& hdr,int sz, bool TruncateNames) const; + + /// @brief Maps archive into memory + bool mapToMemory(std::string* ErrMsg); + + /// @brief Frees all the members and unmaps the archive file. + void cleanUpMemory(); + + /// This type is used to keep track of bitcode modules loaded from the + /// symbol table. It maps the file offset to a pair that consists of the + /// associated ArchiveMember and the Module. + /// @brief Module mapping type + typedef std::map<unsigned,std::pair<Module*,ArchiveMember*> > + ModuleMap; + + + /// @} + /// @name Data + /// @{ + protected: + sys::Path archPath; ///< Path to the archive file we read/write + MembersList members; ///< The ilist of ArchiveMember + MemoryBuffer *mapfile; ///< Raw Archive contents mapped into memory + const char* base; ///< Base of the memory mapped file data + SymTabType symTab; ///< The symbol table + std::string strtab; ///< The string table for long file names + unsigned symTabSize; ///< Size in bytes of symbol table + unsigned firstFileOffset; ///< Offset to first normal file. + ModuleMap modules; ///< The modules loaded via symbol lookup. + ArchiveMember* foreignST; ///< This holds the foreign symbol table. + LLVMContext& Context; ///< This holds global data. + /// @} + /// @name Hidden + /// @{ + private: + Archive() LLVM_DELETED_FUNCTION; + Archive(const Archive&) LLVM_DELETED_FUNCTION; + Archive& operator=(const Archive&) LLVM_DELETED_FUNCTION; + /// @} +}; + +} // End llvm namespace + +#endif diff --git a/tools/llvm-ar/ArchiveInternals.h b/tools/llvm-ar/ArchiveInternals.h new file mode 100644 index 0000000000..e906d7a059 --- /dev/null +++ b/tools/llvm-ar/ArchiveInternals.h @@ -0,0 +1,87 @@ +//===-- lib/Archive/ArchiveInternals.h -------------------------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Internal implementation header for LLVM Archive files. +// +//===----------------------------------------------------------------------===// + +#ifndef TOOLS_LLVM_AR_ARCHIVEINTERNALS_H +#define TOOLS_LLVM_AR_ARCHIVEINTERNALS_H + +#include "Archive.h" +#include "llvm/ADT/StringExtras.h" +#include "llvm/Support/TimeValue.h" +#include <cstring> + +#define ARFILE_MAGIC "!<arch>\n" ///< magic string +#define ARFILE_MAGIC_LEN (sizeof(ARFILE_MAGIC)-1) ///< length of magic string +#define ARFILE_SVR4_SYMTAB_NAME "/ " ///< SVR4 symtab entry name +#define ARFILE_BSD4_SYMTAB_NAME "__.SYMDEF SORTED" ///< BSD4 symtab entry name +#define ARFILE_STRTAB_NAME "// " ///< Name of string table +#define ARFILE_PAD "\n" ///< inter-file align padding +#define ARFILE_MEMBER_MAGIC "`\n" ///< fmag field magic # + +namespace llvm { + + class LLVMContext; + + /// The ArchiveMemberHeader structure is used internally for bitcode + /// archives. + /// The header precedes each file member in the archive. This structure is + /// defined using character arrays for direct and correct interpretation + /// regardless of the endianess of the machine that produced it. + /// @brief Archive File Member Header + class ArchiveMemberHeader { + /// @name Data + /// @{ + public: + char name[16]; ///< Name of the file member. + char date[12]; ///< File date, decimal seconds since Epoch + char uid[6]; ///< user id in ASCII decimal + char gid[6]; ///< group id in ASCII decimal + char mode[8]; ///< file mode in ASCII octal + char size[10]; ///< file size in ASCII decimal + char fmag[2]; ///< Always contains ARFILE_MAGIC_TERMINATOR + + /// @} + /// @name Methods + /// @{ + public: + void init() { + memset(name,' ',16); + memset(date,' ',12); + memset(uid,' ',6); + memset(gid,' ',6); + memset(mode,' ',8); + memset(size,' ',10); + fmag[0] = '`'; + fmag[1] = '\n'; + } + + bool checkSignature() const { + return 0 == memcmp(fmag, ARFILE_MEMBER_MAGIC,2); + } + }; + + // Get just the externally visible defined symbols from the bitcode + bool GetBitcodeSymbols(const sys::Path& fName, + LLVMContext& Context, + std::vector<std::string>& symbols, + std::string* ErrMsg); + + Module* GetBitcodeSymbols(const char *Buffer, unsigned Length, + const std::string& ModuleID, + LLVMContext& Context, + std::vector<std::string>& symbols, + std::string* ErrMsg); +} + +#endif + +// vim: sw=2 ai diff --git a/tools/llvm-ar/ArchiveReader.cpp b/tools/llvm-ar/ArchiveReader.cpp new file mode 100644 index 0000000000..8ad01dca0b --- /dev/null +++ b/tools/llvm-ar/ArchiveReader.cpp @@ -0,0 +1,557 @@ +//===-- ArchiveReader.cpp - Read LLVM archive files -------------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Builds up standard unix archive files (.a) containing LLVM bitcode. +// +//===----------------------------------------------------------------------===// + +#include "Archive.h" +#include "ArchiveInternals.h" +#include "llvm/ADT/OwningPtr.h" +#include "llvm/ADT/SmallPtrSet.h" +#include "llvm/Bitcode/ReaderWriter.h" +#include "llvm/IR/Module.h" +#include "llvm/Support/FileSystem.h" +#include "llvm/Support/MemoryBuffer.h" +#include <cstdio> +#include <cstdlib> +using namespace llvm; + +/// Read a variable-bit-rate encoded unsigned integer +static inline unsigned readInteger(const char*&At, const char*End) { + unsigned Shift = 0; + unsigned Result = 0; + + do { + if (At == End) + return Result; + Result |= (unsigned)((*At++) & 0x7F) << Shift; + Shift += 7; + } while (At[-1] & 0x80); + return Result; +} + +// This member parses an ArchiveMemberHeader that is presumed to be pointed to +// by At. The At pointer is updated to the byte just after the header, which +// can be variable in size. +ArchiveMember* +Archive::parseMemberHeader(const char*& At, const char* End, std::string* error) +{ + if (At + sizeof(ArchiveMemberHeader) >= End) { + if (error) + *error = "Unexpected end of file"; + return 0; + } + + // Cast archive member header + const ArchiveMemberHeader* Hdr = (const ArchiveMemberHeader*)At; + At += sizeof(ArchiveMemberHeader); + + int flags = 0; + int MemberSize = atoi(Hdr->size); + assert(MemberSize >= 0); + + // Check the size of the member for sanity + if (At + MemberSize > End) { + if (error) + *error = "invalid member length in archive file"; + return 0; + } + + // Check the member signature + if (!Hdr->checkSignature()) { + if (error) + *error = "invalid file member signature"; + return 0; + } + + // Convert and check the member name + // The empty name ( '/' and 15 blanks) is for a foreign (non-LLVM) symbol + // table. The special name "//" and 14 blanks is for a string table, used + // for long file names. This library doesn't generate either of those but + // it will accept them. If the name starts with #1/ and the remainder is + // digits, then those digits specify the length of the name that is + // stored immediately following the header. Anything else is a regular, short + // filename that is terminated with a '/' and blanks. + + std::string pathname; + switch (Hdr->name[0]) { + case '#': + if (Hdr->name[1] == '1' && Hdr->name[2] == '/') { + if (isdigit(Hdr->name[3])) { + unsigned len = atoi(&Hdr->name[3]); + const char *nulp = (const char *)memchr(At, '\0', len); + pathname.assign(At, nulp != 0 ? (uintptr_t)(nulp - At) : len); + At += len; + MemberSize -= len; + flags |= ArchiveMember::HasLongFilenameFlag; + } else { + if (error) + *error = "invalid long filename"; + return 0; + } + } + break; + case '/': + if (Hdr->name[1]== '/') { + if (0 == memcmp(Hdr->name, ARFILE_STRTAB_NAME, 16)) { + pathname.assign(ARFILE_STRTAB_NAME); + flags |= ArchiveMember::StringTableFlag; + } else { + if (error) + *error = "invalid string table name"; + return 0; + } + } else if (Hdr->name[1] == ' ') { + if (0 == memcmp(Hdr->name, ARFILE_SVR4_SYMTAB_NAME, 16)) { + pathname.assign(ARFILE_SVR4_SYMTAB_NAME); + flags |= ArchiveMember::SVR4SymbolTableFlag; + } else { + if (error) + *error = "invalid SVR4 symbol table name"; + return 0; + } + } else if (isdigit(Hdr->name[1])) { + unsigned index = atoi(&Hdr->name[1]); + if (index < strtab.length()) { + const char* namep = strtab.c_str() + index; + const char* endp = strtab.c_str() + strtab.length(); + const char* p = namep; + const char* last_p = p; + while (p < endp) { + if (*p == '\n' && *last_p == '/') { + pathname.assign(namep, last_p - namep); + flags |= ArchiveMember::HasLongFilenameFlag; + break; + } + last_p = p; + p++; + } + if (p >= endp) { + if (error) + *error = "missing name terminator in string table"; + return 0; + } + } else { + if (error) + *error = "name index beyond string table"; + return 0; + } + } + break; + case '_': + if (Hdr->name[1] == '_' && + (0 == memcmp(Hdr->name, ARFILE_BSD4_SYMTAB_NAME, 16))) { + pathname.assign(ARFILE_BSD4_SYMTAB_NAME); + flags |= ArchiveMember::BSD4SymbolTableFlag; + break; + } + /* FALL THROUGH */ + + default: + const char* slash = (const char*) memchr(Hdr->name, '/', 16); + if (slash == 0) + slash = Hdr->name + 16; + pathname.assign(Hdr->name, slash - Hdr->name); + break; + } + + // Determine if this is a bitcode file + if (sys::fs::identify_magic(StringRef(At, 4)) == + sys::fs::file_magic::bitcode) + flags |= ArchiveMember::BitcodeFlag; + else + flags &= ~ArchiveMember::BitcodeFlag; + + // Instantiate the ArchiveMember to be filled + ArchiveMember* member = new ArchiveMember(this); + + // Fill in fields of the ArchiveMember + member->parent = this; + member->path.set(pathname); + member->info.fileSize = MemberSize; + member->info.modTime.fromEpochTime(atoi(Hdr->date)); + unsigned int mode; + sscanf(Hdr->mode, "%o", &mode); + member->info.mode = mode; + member->info.user = atoi(Hdr->uid); + member->info.group = atoi(Hdr->gid); + member->flags = flags; + member->data = At; + + return member; +} + +bool +Archive::checkSignature(std::string* error) { + // Check the magic string at file's header + if (mapfile->getBufferSize() < 8 || memcmp(base, ARFILE_MAGIC, 8)) { + if (error) + *error = "invalid signature for an archive file"; + return false; + } + return true; +} + +// This function loads the entire archive and fully populates its ilist with +// the members of the archive file. This is typically used in preparation for +// editing the contents of the archive. +bool +Archive::loadArchive(std::string* error) { + + // Set up parsing + members.clear(); + symTab.clear(); + const char *At = base; + const char *End = mapfile->getBufferEnd(); + + if (!checkSignature(error)) + return false; + + At += 8; // Skip the magic string. + + bool foundFirstFile = false; + while (At < End) { + // parse the member header + const char* Save = At; + ArchiveMember* mbr = parseMemberHeader(At, End, error); + if (!mbr) + return false; + + // check if this is the foreign symbol table + if (mbr->isSVR4SymbolTable() || mbr->isBSD4SymbolTable()) { + // We just save this but don't do anything special + // with it. It doesn't count as the "first file". + if (foreignST) { + // What? Multiple foreign symbol tables? Just chuck it + // and retain the last one found. + delete foreignST; + } + foreignST = mbr; + At += mbr->getSize(); + if ((intptr_t(At) & 1) == 1) + At++; + } else if (mbr->isStringTable()) { + // Simply suck the entire string table into a string + // variable. This will be used to get the names of the + // members that use the "/ddd" format for their names + // (SVR4 style long names). + strtab.assign(At, mbr->getSize()); + At += mbr->getSize(); + if ((intptr_t(At) & 1) == 1) + At++; + delete mbr; + } else { + // This is just a regular file. If its the first one, save its offset. + // Otherwise just push it on the list and move on to the next file. + if (!foundFirstFile) { + firstFileOffset = Save - base; + foundFirstFile = true; + } + members.push_back(mbr); + At += mbr->getSize(); + if ((intptr_t(At) & 1) == 1) + At++; + } + } + return true; +} + +// Open and completely load the archive file. +Archive* +Archive::OpenAndLoad(const sys::Path& File, LLVMContext& C, + std::string* ErrorMessage) { + OwningPtr<Archive> result ( new Archive(File, C)); + if (result->mapToMemory(ErrorMessage)) + return NULL; + if (!result->loadArchive(ErrorMessage)) + return NULL; + return result.take(); +} + +// Get all the bitcode modules from the archive +bool +Archive::getAllModules(std::vector<Module*>& Modules, + std::string* ErrMessage) { + + for (iterator I=begin(), E=end(); I != E; ++I) { + if (I->isBitcode()) { + std::string FullMemberName = archPath.str() + + "(" + I->getPath().str() + ")"; + MemoryBuffer *Buffer = + MemoryBuffer::getMemBufferCopy(StringRef(I->getData(), I->getSize()), + FullMemberName.c_str()); + + Module *M = ParseBitcodeFile(Buffer, Context, ErrMessage); + delete Buffer; + if (!M) + return true; + + Modules.push_back(M); + } + } + return false; +} + +// Load just the symbol table from the archive file +bool +Archive::loadSymbolTable(std::string* ErrorMsg) { + + // Set up parsing + members.clear(); + symTab.clear(); + const char *At = base; + const char *End = mapfile->getBufferEnd(); + + // Make sure we're dealing with an archive + if (!checkSignature(ErrorMsg)) + return false; + + At += 8; // Skip signature + + // Parse the first file member header + const char* FirstFile = At; + ArchiveMember* mbr = parseMemberHeader(At, End, ErrorMsg); + if (!mbr) + return false; + + if (mbr->isSVR4SymbolTable() || mbr->isBSD4SymbolTable()) { + // Skip the foreign symbol table, we don't do anything with it + At += mbr->getSize(); + if ((intptr_t(At) & 1) == 1) + At++; + delete mbr; + + // Read the next one + FirstFile = At; + mbr = parseMemberHeader(At, End, ErrorMsg); + if (!mbr) { + delete mbr; + return false; + } + } + + if (mbr->isStringTable()) { + // Process the string table entry + strtab.assign((const char*)mbr->getData(), mbr->getSize()); + At += mbr->getSize(); + if ((intptr_t(At) & 1) == 1) + At++; + delete mbr; + // Get the next one + FirstFile = At; + mbr = parseMemberHeader(At, End, ErrorMsg); + if (!mbr) { + delete mbr; + return false; + } + } + + // There's no symbol table in the file. We have to rebuild it from scratch + // because the intent of this method is to get the symbol table loaded so + // it can be searched efficiently. + // Add the member to the members list + members.push_back(mbr); + + firstFileOffset = FirstFile - base; + return true; +} + +// Open the archive and load just the symbol tables +Archive* Archive::OpenAndLoadSymbols(const sys::Path& File, + LLVMContext& C, + std::string* ErrorMessage) { + OwningPtr<Archive> result ( new Archive(File, C) ); + if (result->mapToMemory(ErrorMessage)) + return NULL; + if (!result->loadSymbolTable(ErrorMessage)) + return NULL; + return result.take(); +} + +// Look up one symbol in the symbol table and return the module that defines +// that symbol. +Module* +Archive::findModuleDefiningSymbol(const std::string& symbol, + std::string* ErrMsg) { + SymTabType::iterator SI = symTab.find(symbol); + if (SI == symTab.end()) + return 0; + + // The symbol table was previously constructed assuming that the members were + // written without the symbol table header. Because VBR encoding is used, the + // values could not be adjusted to account for the offset of the symbol table + // because that could affect the size of the symbol table due to VBR encoding. + // We now have to account for this by adjusting the offset by the size of the + // symbol table and its header. + unsigned fileOffset = + SI->second + // offset in symbol-table-less file + firstFileOffset; // add offset to first "real" file in archive + + // See if the module is already loaded + ModuleMap::iterator MI = modules.find(fileOffset); + if (MI != modules.end()) + return MI->second.first; + + // Module hasn't been loaded yet, we need to load it + const char* modptr = base + fileOffset; + ArchiveMember* mbr = parseMemberHeader(modptr, mapfile->getBufferEnd(), + ErrMsg); + if (!mbr) + return 0; + + // Now, load the bitcode module to get the Module. + std::string FullMemberName = archPath.str() + "(" + + mbr->getPath().str() + ")"; + MemoryBuffer *Buffer = + MemoryBuffer::getMemBufferCopy(StringRef(mbr->getData(), mbr->getSize()), + FullMemberName.c_str()); + + Module *m = getLazyBitcodeModule(Buffer, Context, ErrMsg); + if (!m) + return 0; + + modules.insert(std::make_pair(fileOffset, std::make_pair(m, mbr))); + + return m; +} + +// Look up multiple symbols in the symbol table and return a set of +// Modules that define those symbols. +bool +Archive::findModulesDefiningSymbols(std::set<std::string>& symbols, + SmallVectorImpl<Module*>& result, + std::string* error) { + if (!mapfile || !base) { + if (error) + *error = "Empty archive invalid for finding modules defining symbols"; + return false; + } + + if (symTab.empty()) { + // We don't have a symbol table, so we must build it now but lets also + // make sure that we populate the modules table as we do this to ensure + // that we don't load them twice when findModuleDefiningSymbol is called + // below. + + // Get a pointer to the first file + const char* At = base + firstFileOffset; + const char* End = mapfile->getBufferEnd(); + + while ( At < End) { + // Compute the offset to be put in the symbol table + unsigned offset = At - base - firstFileOffset; + + // Parse the file's header + ArchiveMember* mbr = parseMemberHeader(At, End, error); + if (!mbr) + return false; + + // If it contains symbols + if (mbr->isBitcode()) { + // Get the symbols + std::vector<std::string> symbols; + std::string FullMemberName = archPath.str() + "(" + + mbr->getPath().str() + ")"; + Module* M = + GetBitcodeSymbols(At, mbr->getSize(), FullMemberName, Context, + symbols, error); + + if (M) { + // Insert the module's symbols into the symbol table + for (std::vector<std::string>::iterator I = symbols.begin(), + E=symbols.end(); I != E; ++I ) { + symTab.insert(std::make_pair(*I, offset)); + } + // Insert the Module and the ArchiveMember into the table of + // modules. + modules.insert(std::make_pair(offset, std::make_pair(M, mbr))); + } else { + if (error) + *error = "Can't parse bitcode member: " + + mbr->getPath().str() + ": " + *error; + delete mbr; + return false; + } + } + + // Go to the next file location + At += mbr->getSize(); + if ((intptr_t(At) & 1) == 1) + At++; + } + } + + // At this point we have a valid symbol table (one way or another) so we + // just use it to quickly find the symbols requested. + + SmallPtrSet<Module*, 16> Added; + for (std::set<std::string>::iterator I=symbols.begin(), + Next = I, + E=symbols.end(); I != E; I = Next) { + // Increment Next before we invalidate it. + ++Next; + + // See if this symbol exists + Module* m = findModuleDefiningSymbol(*I,error); + if (!m) + continue; + bool NewMember = Added.insert(m); + if (!NewMember) + continue; + + // The symbol exists, insert the Module into our result. + result.push_back(m); + + // Remove the symbol now that its been resolved. + symbols.erase(I); + } + return true; +} + +bool Archive::isBitcodeArchive() { + // Make sure the symTab has been loaded. In most cases this should have been + // done when the archive was constructed, but still, this is just in case. + if (symTab.empty()) + if (!loadSymbolTable(0)) + return false; + + // Now that we know it's been loaded, return true + // if it has a size + if (symTab.size()) return true; + + // We still can't be sure it isn't a bitcode archive + if (!loadArchive(0)) + return false; + + std::vector<Module *> Modules; + std::string ErrorMessage; + + // Scan the archive, trying to load a bitcode member. We only load one to + // see if this works. + for (iterator I = begin(), E = end(); I != E; ++I) { + if (!I->isBitcode()) + continue; + + std::string FullMemberName = + archPath.str() + "(" + I->getPath().str() + ")"; + + MemoryBuffer *Buffer = + MemoryBuffer::getMemBufferCopy(StringRef(I->getData(), I->getSize()), + FullMemberName.c_str()); + Module *M = ParseBitcodeFile(Buffer, Context); + delete Buffer; + if (!M) + return false; // Couldn't parse bitcode, not a bitcode archive. + delete M; + return true; + } + + return false; +} diff --git a/tools/llvm-ar/ArchiveWriter.cpp b/tools/llvm-ar/ArchiveWriter.cpp new file mode 100644 index 0000000000..24f5dbc0dc --- /dev/null +++ b/tools/llvm-ar/ArchiveWriter.cpp @@ -0,0 +1,429 @@ +//===-- ArchiveWriter.cpp - Write LLVM archive files ----------------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Builds up an LLVM archive file (.a) containing LLVM bitcode. +// +//===----------------------------------------------------------------------===// + +#include "Archive.h" +#include "ArchiveInternals.h" +#include "llvm/ADT/OwningPtr.h" +#include "llvm/Bitcode/ReaderWriter.h" +#include "llvm/IR/Module.h" +#include "llvm/Support/FileSystem.h" +#include "llvm/Support/MemoryBuffer.h" +#include "llvm/Support/Process.h" +#include "llvm/Support/Signals.h" +#include "llvm/Support/system_error.h" +#include <fstream> +#include <iomanip> +#include <ostream> +using namespace llvm; + +// Write an integer using variable bit rate encoding. This saves a few bytes +// per entry in the symbol table. +static inline void writeInteger(unsigned num, std::ofstream& ARFile) { + while (1) { + if (num < 0x80) { // done? + ARFile << (unsigned char)num; + return; + } + + // Nope, we are bigger than a character, output the next 7 bits and set the + // high bit to say that there is more coming... + ARFile << (unsigned char)(0x80 | ((unsigned char)num & 0x7F)); + num >>= 7; // Shift out 7 bits now... + } +} + +// Compute how many bytes are taken by a given VBR encoded value. This is needed +// to pre-compute the size of the symbol table. +static inline unsigned numVbrBytes(unsigned num) { + + // Note that the following nested ifs are somewhat equivalent to a binary + // search. We split it in half by comparing against 2^14 first. This allows + // most reasonable values to be done in 2 comparisons instead of 1 for + // small ones and four for large ones. We expect this to access file offsets + // in the 2^10 to 2^24 range and symbol lengths in the 2^0 to 2^8 range, + // so this approach is reasonable. + if (num < 1<<14) { + if (num < 1<<7) + return 1; + else + return 2; + } + if (num < 1<<21) + return 3; + + if (num < 1<<28) + return 4; + return 5; // anything >= 2^28 takes 5 bytes +} + +// Create an empty archive. +Archive* Archive::CreateEmpty(const sys::Path& FilePath, LLVMContext& C) { + Archive* result = new Archive(FilePath, C); + return result; +} + +// Fill the ArchiveMemberHeader with the information from a member. If +// TruncateNames is true, names are flattened to 15 chars or less. The sz field +// is provided here instead of coming from the mbr because the member might be +// stored compressed and the compressed size is not the ArchiveMember's size. +// Furthermore compressed files have negative size fields to identify them as +// compressed. +bool +Archive::fillHeader(const ArchiveMember &mbr, ArchiveMemberHeader& hdr, + int sz, bool TruncateNames) const { + + // Set the permissions mode, uid and gid + hdr.init(); + char buffer[32]; + sprintf(buffer, "%-8o", mbr.getMode()); + memcpy(hdr.mode,buffer,8); + sprintf(buffer, "%-6u", mbr.getUser()); + memcpy(hdr.uid,buffer,6); + sprintf(buffer, "%-6u", mbr.getGroup()); + memcpy(hdr.gid,buffer,6); + + // Set the last modification date + uint64_t secondsSinceEpoch = mbr.getModTime().toEpochTime(); + sprintf(buffer,"%-12u", unsigned(secondsSinceEpoch)); + memcpy(hdr.date,buffer,12); + + // Get rid of trailing blanks in the name + std::string mbrPath = mbr.getPath().str(); + size_t mbrLen = mbrPath.length(); + while (mbrLen > 0 && mbrPath[mbrLen-1] == ' ') { + mbrPath.erase(mbrLen-1,1); + mbrLen--; + } + + // Set the name field in one of its various flavors. + bool writeLongName = false; + if (mbr.isStringTable()) { + memcpy(hdr.name,ARFILE_STRTAB_NAME,16); + } else if (mbr.isSVR4SymbolTable()) { + memcpy(hdr.name,ARFILE_SVR4_SYMTAB_NAME,16); + } else if (mbr.isBSD4SymbolTable()) { + memcpy(hdr.name,ARFILE_BSD4_SYMTAB_NAME,16); + } else if (TruncateNames) { + const char* nm = mbrPath.c_str(); + unsigned len = mbrPath.length(); + size_t slashpos = mbrPath.rfind('/'); + if (slashpos != std::string::npos) { + nm += slashpos + 1; + len -= slashpos +1; + } + if (len > 15) + len = 15; + memcpy(hdr.name,nm,len); + hdr.name[len] = '/'; + } else if (mbrPath.length() < 16 && mbrPath.find('/') == std::string::npos) { + memcpy(hdr.name,mbrPath.c_str(),mbrPath.length()); + hdr.name[mbrPath.length()] = '/'; + } else { + std::string nm = "#1/"; + nm += utostr(mbrPath.length()); + memcpy(hdr.name,nm.data(),nm.length()); + if (sz < 0) + sz -= mbrPath.length(); + else + sz += mbrPath.length(); + writeLongName = true; + } + + // Set the size field + if (sz < 0) { + buffer[0] = '-'; + sprintf(&buffer[1],"%-9u",(unsigned)-sz); + } else { + sprintf(buffer, "%-10u", (unsigned)sz); + } + memcpy(hdr.size,buffer,10); + + return writeLongName; +} + +// Insert a file into the archive before some other member. This also takes care +// of extracting the necessary flags and information from the file. +bool +Archive::addFileBefore(const sys::Path& filePath, iterator where, + std::string* ErrMsg) { + bool Exists; + if (sys::fs::exists(filePath.str(), Exists) || !Exists) { + if (ErrMsg) + *ErrMsg = "Can not add a non-existent file to archive"; + return true; + } + + ArchiveMember* mbr = new ArchiveMember(this); + + mbr->data = 0; + mbr->path = filePath; + const sys::FileStatus *FSInfo = mbr->path.getFileStatus(false, ErrMsg); + if (!FSInfo) { + delete mbr; + return true; + } + mbr->info = *FSInfo; + + unsigned flags = 0; + bool hasSlash = filePath.str().find('/') != std::string::npos; + if (hasSlash) + flags |= ArchiveMember::HasPathFlag; + if (hasSlash || filePath.str().length() > 15) + flags |= ArchiveMember::HasLongFilenameFlag; + + sys::fs::file_magic type; + if (sys::fs::identify_magic(mbr->path.str(), type)) + type = sys::fs::file_magic::unknown; + switch (type) { + case sys::fs::file_magic::bitcode: + flags |= ArchiveMember::BitcodeFlag; + break; + default: + break; + } + mbr->flags = flags; + members.insert(where,mbr); + return false; +} + +// Write one member out to the file. +bool +Archive::writeMember( + const ArchiveMember& member, + std::ofstream& ARFile, + bool CreateSymbolTable, + bool TruncateNames, + std::string* ErrMsg +) { + + unsigned filepos = ARFile.tellp(); + filepos -= 8; + + // Get the data and its size either from the + // member's in-memory data or directly from the file. + size_t fSize = member.getSize(); + const char *data = (const char*)member.getData(); + MemoryBuffer *mFile = 0; + if (!data) { + OwningPtr<MemoryBuffer> File; + if (error_code ec = MemoryBuffer::getFile(member.getPath().c_str(), File)) { + if (ErrMsg) + *ErrMsg = ec.message(); + return true; + } + mFile = File.take(); + data = mFile->getBufferStart(); + fSize = mFile->getBufferSize(); + } + + // Now that we have the data in memory, update the + // symbol table if it's a bitcode file. + if (CreateSymbolTable && member.isBitcode()) { + std::vector<std::string> symbols; + std::string FullMemberName = archPath.str() + "(" + member.getPath().str() + + ")"; + Module* M = + GetBitcodeSymbols(data, fSize, FullMemberName, Context, symbols, ErrMsg); + + // If the bitcode parsed successfully + if ( M ) { + for (std::vector<std::string>::iterator SI = symbols.begin(), + SE = symbols.end(); SI != SE; ++SI) { + + std::pair<SymTabType::iterator,bool> Res = + symTab.insert(std::make_pair(*SI,filepos)); + + if (Res.second) { + symTabSize += SI->length() + + numVbrBytes(SI->length()) + + numVbrBytes(filepos); + } + } + // We don't need this module any more. + delete M; + } else { + delete mFile; + if (ErrMsg) + *ErrMsg = "Can't parse bitcode member: " + member.getPath().str() + + ": " + *ErrMsg; + return true; + } + } + + int hdrSize = fSize; + + // Compute the fields of the header + ArchiveMemberHeader Hdr; + bool writeLongName = fillHeader(member,Hdr,hdrSize,TruncateNames); + + // Write header to archive file + ARFile.write((char*)&Hdr, sizeof(Hdr)); + + // Write the long filename if its long + if (writeLongName) { + ARFile.write(member.getPath().str().data(), + member.getPath().str().length()); + } + + // Write the (possibly compressed) member's content to the file. + ARFile.write(data,fSize); + + // Make sure the member is an even length + if ((ARFile.tellp() & 1) == 1) + ARFile << ARFILE_PAD; + + // Close the mapped file if it was opened + delete mFile; + return false; +} + +// Write the entire archive to the file specified when the archive was created. +// This writes to a temporary file first. Options are for creating a symbol +// table, flattening the file names (no directories, 15 chars max) and +// compressing each archive member. +bool +Archive::writeToDisk(bool CreateSymbolTable, bool TruncateNames, + std::string* ErrMsg) +{ + // Make sure they haven't opened up the file, not loaded it, + // but are now trying to write it which would wipe out the file. + if (members.empty() && mapfile && mapfile->getBufferSize() > 8) { + if (ErrMsg) + *ErrMsg = "Can't write an archive not opened for writing"; + return true; + } + + // Create a temporary file to store the archive in + sys::Path TmpArchive = archPath; + if (TmpArchive.createTemporaryFileOnDisk(ErrMsg)) + return true; + + // Make sure the temporary gets removed if we crash + sys::RemoveFileOnSignal(TmpArchive.str()); + + // Create archive file for output. + std::ios::openmode io_mode = std::ios::out | std::ios::trunc | + std::ios::binary; + std::ofstream ArchiveFile(TmpArchive.c_str(), io_mode); + + // Check for errors opening or creating archive file. + if (!ArchiveFile.is_open() || ArchiveFile.bad()) { + TmpArchive.eraseFromDisk(); + if (ErrMsg) + *ErrMsg = "Error opening archive file: " + archPath.str(); + return true; + } + + // If we're creating a symbol table, reset it now + if (CreateSymbolTable) { + symTabSize = 0; + symTab.clear(); + } + + // Write magic string to archive. + ArchiveFile << ARFILE_MAGIC; + + // Loop over all member files, and write them out. Note that this also + // builds the symbol table, symTab. + for (MembersList::iterator I = begin(), E = end(); I != E; ++I) { + if (writeMember(*I, ArchiveFile, CreateSymbolTable, + TruncateNames, ErrMsg)) { + TmpArchive.eraseFromDisk(); + ArchiveFile.close(); + return true; + } + } + + // Close archive file. + ArchiveFile.close(); + + // Write the symbol table + if (CreateSymbolTable) { + // At this point we have written a file that is a legal archive but it + // doesn't have a symbol table in it. To aid in faster reading and to + // ensure compatibility with other archivers we need to put the symbol + // table first in the file. Unfortunately, this means mapping the file + // we just wrote back in and copying it to the destination file. + sys::Path FinalFilePath = archPath; + + // Map in the archive we just wrote. + { + OwningPtr<MemoryBuffer> arch; + if (error_code ec = MemoryBuffer::getFile(TmpArchive.c_str(), arch)) { + if (ErrMsg) + *ErrMsg = ec.message(); + return true; + } + const char* base = arch->getBufferStart(); + + // Open another temporary file in order to avoid invalidating the + // mmapped data + if (FinalFilePath.createTemporaryFileOnDisk(ErrMsg)) + return true; + sys::RemoveFileOnSignal(FinalFilePath.str()); + + std::ofstream FinalFile(FinalFilePath.c_str(), io_mode); + if (!FinalFile.is_open() || FinalFile.bad()) { + TmpArchive.eraseFromDisk(); + if (ErrMsg) + *ErrMsg = "Error opening archive file: " + FinalFilePath.str(); + return true; + } + + // Write the file magic number + FinalFile << ARFILE_MAGIC; + + // If there is a foreign symbol table, put it into the file now. Most + // ar(1) implementations require the symbol table to be first but llvm-ar + // can deal with it being after a foreign symbol table. This ensures + // compatibility with other ar(1) implementations as well as allowing the + // archive to store both native .o and LLVM .bc files, both indexed. + if (foreignST) { + if (writeMember(*foreignST, FinalFile, false, false, ErrMsg)) { + FinalFile.close(); + TmpArchive.eraseFromDisk(); + return true; + } + } + + // Copy the temporary file contents being sure to skip the file's magic + // number. + FinalFile.write(base + sizeof(ARFILE_MAGIC)-1, + arch->getBufferSize()-sizeof(ARFILE_MAGIC)+1); + + // Close up shop + FinalFile.close(); + } // free arch. + + // Move the final file over top of TmpArchive + if (FinalFilePath.renamePathOnDisk(TmpArchive, ErrMsg)) + return true; + } + + // Before we replace the actual archive, we need to forget all the + // members, since they point to data in that old archive. We need to do + // this because we cannot replace an open file on Windows. + cleanUpMemory(); + + if (TmpArchive.renamePathOnDisk(archPath, ErrMsg)) + return true; + + // Set correct read and write permissions after temporary file is moved + // to final destination path. + if (archPath.makeReadableOnDisk(ErrMsg)) + return true; + if (archPath.makeWriteableOnDisk(ErrMsg)) + return true; + + return false; +} diff --git a/tools/llvm-ar/CMakeLists.txt b/tools/llvm-ar/CMakeLists.txt index 70eb7603fd..68095df714 100644 --- a/tools/llvm-ar/CMakeLists.txt +++ b/tools/llvm-ar/CMakeLists.txt @@ -1,7 +1,10 @@ -set(LLVM_LINK_COMPONENTS archive) +set(LLVM_LINK_COMPONENTS support bitreader) add_llvm_tool(llvm-ar llvm-ar.cpp + ArchiveWriter.cpp + ArchiveReader.cpp + Archive.cpp ) # TODO: Support check-local. diff --git a/tools/llvm-ar/LLVMBuild.txt b/tools/llvm-ar/LLVMBuild.txt index 1f61a32014..236b4659f7 100644 --- a/tools/llvm-ar/LLVMBuild.txt +++ b/tools/llvm-ar/LLVMBuild.txt @@ -19,4 +19,3 @@ type = Tool name = llvm-ar parent = Tools -required_libraries = Archive diff --git a/tools/llvm-ar/Makefile b/tools/llvm-ar/Makefile index fafb14bc12..9e782cfca5 100644 --- a/tools/llvm-ar/Makefile +++ b/tools/llvm-ar/Makefile @@ -9,7 +9,7 @@ LEVEL := ../.. TOOLNAME := llvm-ar -LINK_COMPONENTS := archive +LINK_COMPONENTS := bitreader support # This tool has no plugins, optimize startup time. TOOL_NO_EXPORTS := 1 diff --git a/tools/llvm-ar/llvm-ar.cpp b/tools/llvm-ar/llvm-ar.cpp index 243a845e3c..2a4a3deecd 100644 --- a/tools/llvm-ar/llvm-ar.cpp +++ b/tools/llvm-ar/llvm-ar.cpp @@ -12,8 +12,8 @@ // //===----------------------------------------------------------------------===// +#include "Archive.h" #include "llvm/IR/LLVMContext.h" -#include "llvm/Bitcode/Archive.h" #include "llvm/IR/Module.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/FileSystem.h" |