1 //===- ArchiveReader.cpp - Code to read LLVM bytecode from .a files -------===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by the LLVM research group and is distributed under
6 // the University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the ReadArchiveFile interface, which allows a linker to
11 // read all of the LLVM bytecode files contained in a .a file. This file
12 // understands the standard system .a file format. This can only handle the .a
13 // variant prevalent on Linux systems so far, but may be extended. See
14 // information in this source file for more information:
15 // http://sources.redhat.com/cgi-bin/cvsweb.cgi/src/bfd/archive.c?cvsroot=src
17 //===----------------------------------------------------------------------===//
19 #include "llvm/Bytecode/Reader.h"
20 #include "llvm/Module.h"
21 #include "Config/sys/stat.h"
22 #include "Config/sys/mman.h"
23 #include "Config/fcntl.h"
36 char fmag[2]; // Always equal to '`\n'
40 UserObject, // A user .o/.bc file
41 Unknown, // Unknown file, just ignore it
42 SVR4LongFilename, // a "//" section used for long file names
43 ArchiveSymbolTable, // Symbol table produced by ranlib.
47 /// getObjectType - Determine the type of object that this header represents.
48 /// This is capable of parsing the variety of special sections used for various
51 static enum ObjectType getObjectType(ar_hdr *H, unsigned char *MemberData,
53 // Check for sections with special names...
54 if (!memcmp(H->name, "__.SYMDEF ", 16))
55 return ArchiveSymbolTable;
56 if (!memcmp(H->name, "__.SYMDEF SORTED", 16))
57 return ArchiveSymbolTable;
58 if (!memcmp(H->name, "// ", 16))
59 return SVR4LongFilename;
61 // Check to see if it looks like an llvm object file...
62 if (Size >= 4 && !memcmp(MemberData, "llvm", 4))
68 static inline bool Error(std::string *ErrorStr, const char *Message) {
69 if (ErrorStr) *ErrorStr = Message;
73 static bool ParseSymbolTableSection(unsigned char *Buffer, unsigned Size,
75 // Currently not supported (succeeds without doing anything)
79 static bool ReadArchiveBuffer(const std::string &ArchiveName,
80 unsigned char *Buffer, unsigned Length,
81 std::vector<Module*> &Objects,
82 std::string *ErrorStr) {
83 if (Length < 8 || memcmp(Buffer, "!<arch>\n", 8))
84 return Error(ErrorStr, "signature incorrect for an archive file!");
85 Buffer += 8; Length -= 8; // Skip the magic string.
87 std::vector<char> LongFilenames;
89 while (Length >= sizeof(ar_hdr)) {
90 ar_hdr *Hdr = (ar_hdr*)Buffer;
91 unsigned SizeFromHeader = atoi(Hdr->size);
92 if (SizeFromHeader + sizeof(ar_hdr) > Length)
93 return Error(ErrorStr, "invalid record length in archive file!");
95 unsigned char *MemberData = Buffer + sizeof(ar_hdr);
96 unsigned MemberSize = SizeFromHeader;
97 // Get name of archive member.
98 char *startp = Hdr->name;
99 char *endp = (char *) memchr (startp, '/', sizeof(ar_hdr));
100 if (memcmp (Hdr->name, "#1/", 3) == 0) {
101 // 4.4BSD/MacOSX long filenames are abbreviated as "#1/L", where L is an
102 // ASCII-coded decimal number representing the length of the name buffer,
103 // which is prepended to the archive member's contents.
104 unsigned NameLength = atoi (&Hdr->name[3]);
105 startp = (char *) MemberData;
106 endp = startp + NameLength;
107 MemberData += NameLength;
108 MemberSize -= NameLength;
109 } else if (startp == endp && isdigit (Hdr->name[1])) {
110 // SVR4 long filenames are abbreviated as "/I", where I is
111 // an ASCII-coded decimal index into the LongFilenames vector.
112 unsigned NameIndex = atoi (&Hdr->name[1]);
113 assert (LongFilenames.size () > NameIndex
114 && "SVR4-style long filename for archive member not found");
115 startp = &LongFilenames[NameIndex];
116 endp = strchr (startp, '/');
119 // 4.4BSD/MacOSX *short* filenames are not guaranteed to have a
120 // terminator. Start at the end of the field and backtrack over spaces.
121 endp = startp + sizeof(Hdr->name);
122 while (endp[-1] == ' ')
127 // We now have the beginning and the end of the object name.
128 // Convert this into a dynamically allocated std::string to pass
129 // to the routines that create the Module object. We do this
130 // (I think) because the created Module object will outlive this function,
131 // but statically declared std::string's won't.
133 std::string MemberName (startp, endp);
134 std::string * FullMemberName;
135 FullMemberName = new std::string (ArchiveName + "(" + MemberName + ")");
137 switch (getObjectType(Hdr, MemberData, MemberSize)) {
138 case SVR4LongFilename:
139 // If this is a long filename section, read all of the file names into the
140 // LongFilenames vector.
141 LongFilenames.assign (MemberData, MemberData + MemberSize);
144 Module *M = ParseBytecodeBuffer(MemberData, MemberSize,
145 *(FullMemberName), ErrorStr);
147 Objects.push_back(M);
150 case ArchiveSymbolTable:
151 if (ParseSymbolTableSection(MemberData, MemberSize, ErrorStr))
155 std::cerr << "ReadArchiveBuffer: WARNING: Skipping unknown file: "
156 << *(FullMemberName) << "\n";
157 break; // Just ignore unknown files.
160 // Round SizeFromHeader up to an even number...
161 SizeFromHeader = (SizeFromHeader+1)/2*2;
162 Buffer += sizeof(ar_hdr)+SizeFromHeader; // Move to the next entry
163 Length -= sizeof(ar_hdr)+SizeFromHeader;
170 // ReadArchiveFile - Read bytecode files from the specified .a file, returning
171 // true on error, or false on success. This does not support reading files from
174 bool ReadArchiveFile(const std::string &Filename, std::vector<Module*> &Objects,
175 std::string *ErrorStr) {
176 int FD = open(Filename.c_str(), O_RDONLY);
178 return Error(ErrorStr, "Error opening file!");
180 // Stat the file to get its length...
182 if (fstat(FD, &StatBuf) == -1 || StatBuf.st_size == 0)
183 return Error(ErrorStr, "Error stat'ing file!");
185 // mmap in the file all at once...
186 int Length = StatBuf.st_size;
187 unsigned char *Buffer = (unsigned char*)mmap(0, Length, PROT_READ,
189 if (Buffer == (unsigned char*)MAP_FAILED)
190 return Error(ErrorStr, "Error mmapping file!");
192 // Parse the archive files we mmap'ped in
193 bool Result = ReadArchiveBuffer(Filename, Buffer, Length, Objects, ErrorStr);
195 // Unmmap the archive...
196 munmap((char*)Buffer, Length);
198 if (Result) // Free any loaded objects
199 while (!Objects.empty()) {
200 delete Objects.back();
207 } // End llvm namespace