1 //===-- ArchiveWriter.cpp - Write LLVM archive files ----------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by Reid Spencer and is distributed under the
6 // University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // Builds up an LLVM archive file (.a) containing LLVM bytecode.
12 //===----------------------------------------------------------------------===//
14 #include "ArchiveInternals.h"
15 #include "llvm/Bytecode/Reader.h"
16 #include "llvm/Support/FileUtilities.h"
17 #include "llvm/Support/Compressor.h"
18 #include "llvm/System/Signals.h"
25 // Write an integer using variable bit rate encoding. This saves a few bytes
26 // per entry in the symbol table.
27 inline void writeInteger(unsigned num, std::ofstream& ARFile) {
29 if (num < 0x80) { // done?
30 ARFile << (unsigned char)num;
34 // Nope, we are bigger than a character, output the next 7 bits and set the
35 // high bit to say that there is more coming...
36 ARFile << (unsigned char)(0x80 | ((unsigned char)num & 0x7F));
37 num >>= 7; // Shift out 7 bits now...
41 // Compute how many bytes are taken by a given VBR encoded value. This is needed
42 // to pre-compute the size of the symbol table.
43 inline unsigned numVbrBytes(unsigned num) {
45 // Note that the following nested ifs are somewhat equivalent to a binary
46 // search. We split it in half by comparing against 2^14 first. This allows
47 // most reasonable values to be done in 2 comparisons instead of 1 for
48 // small ones and four for large ones. We expect this to access file offsets
49 // in the 2^10 to 2^24 range and symbol lengths in the 2^0 to 2^8 range,
50 // so this approach is reasonable.
61 return 5; // anything >= 2^28 takes 5 bytes
64 // Create an empty archive.
66 Archive::CreateEmpty(const sys::Path& FilePath ) {
67 Archive* result = new Archive(FilePath,false);
71 // Fill the ArchiveMemberHeader with the information from a member. If
72 // TruncateNames is true, names are flattened to 15 chars or less. The sz field
73 // is provided here instead of coming from the mbr because the member might be
74 // stored compressed and the compressed size is not the ArchiveMember's size.
75 // Furthermore compressed files have negative size fields to identify them as
78 Archive::fillHeader(const ArchiveMember &mbr, ArchiveMemberHeader& hdr,
79 int sz, bool TruncateNames) const {
81 // Set the permissions mode, uid and gid
84 sprintf(buffer, "%-8o", mbr.getMode());
85 memcpy(hdr.mode,buffer,8);
86 sprintf(buffer, "%-6u", mbr.getUser());
87 memcpy(hdr.uid,buffer,6);
88 sprintf(buffer, "%-6u", mbr.getGroup());
89 memcpy(hdr.gid,buffer,6);
94 sprintf(&buffer[1],"%-9u",(unsigned)-sz);
96 sprintf(buffer, "%-10u", (unsigned)sz);
98 memcpy(hdr.size,buffer,10);
100 // Set the last modification date
101 uint64_t secondsSinceEpoch = mbr.getModTime().toEpochTime();
102 sprintf(buffer,"%-12u", unsigned(secondsSinceEpoch));
103 memcpy(hdr.date,buffer,12);
105 // Set the name field in one of its various flavors.
106 bool writeLongName = false;
107 const std::string& mbrPath = mbr.getPath().get();
108 if (mbr.isStringTable()) {
109 memcpy(hdr.name,ARFILE_STRTAB_NAME,16);
110 } else if (mbr.isForeignSymbolTable()) {
111 memcpy(hdr.name,ARFILE_SYMTAB_NAME,16);
112 } else if (mbr.isLLVMSymbolTable()) {
113 memcpy(hdr.name,ARFILE_LLVM_SYMTAB_NAME,16);
114 } else if (TruncateNames) {
115 const char* nm = mbrPath.c_str();
116 unsigned len = mbrPath.length();
117 size_t slashpos = mbrPath.rfind('/');
118 if (slashpos != std::string::npos) {
124 memcpy(hdr.name,nm,len);
126 } else if (mbrPath.length() < 16 && mbrPath.find('/') == std::string::npos) {
127 mbrPath.copy(hdr.name,mbrPath.length());
128 hdr.name[mbrPath.length()] = '/';
130 std::string nm = "#1/";
131 nm += utostr(mbrPath.length());
132 nm.copy(hdr.name,nm.length());
133 writeLongName = true;
135 return writeLongName;
138 // Insert a file into the archive before some other member. This also takes care
139 // of extracting the necessary flags and information from the file.
141 Archive::addFileBefore(const sys::Path& filePath, iterator where) {
142 assert(filePath.exists() && "Can't add a non-existent file");
144 ArchiveMember* mbr = new ArchiveMember(this);
147 mbr->path = filePath;
148 mbr->path.getStatusInfo(mbr->info);
151 bool hasSlash = filePath.get().find('/') != std::string::npos;
153 flags |= ArchiveMember::HasPathFlag;
154 if (hasSlash || filePath.get().length() > 15)
155 flags |= ArchiveMember::HasLongFilenameFlag;
157 mbr->path.getMagicNumber(magic,4);
158 switch (sys::IdentifyFileType(magic.c_str(),4)) {
159 case sys::BytecodeFileType:
160 flags |= ArchiveMember::BytecodeFlag;
162 case sys::CompressedBytecodeFileType:
163 flags |= ArchiveMember::CompressedBytecodeFlag;
169 members.insert(where,mbr);
172 // Write one member out to the file.
174 Archive::writeMember(
175 const ArchiveMember& member,
176 std::ofstream& ARFile,
177 bool CreateSymbolTable,
182 unsigned filepos = ARFile.tellp();
185 // Get the data and its size either from the
186 // member's in-memory data or directly from the file.
187 size_t fSize = member.getSize();
188 const char* data = (const char*)member.getData();
189 sys::MappedFile* mFile = 0;
191 mFile = new sys::MappedFile(member.getPath());
192 data = (const char*) mFile->map();
193 fSize = mFile->size();
196 // Now that we have the data in memory, update the
197 // symbol table if its a bytecode file.
198 if (CreateSymbolTable &&
199 (member.isBytecode() || member.isCompressedBytecode())) {
200 std::vector<std::string> symbols;
201 ModuleProvider* MP = GetBytecodeSymbols(
202 (const unsigned char*)data,fSize,member.getPath().get(), symbols);
204 // If the bytecode parsed successfully
206 for (std::vector<std::string>::iterator SI = symbols.begin(),
207 SE = symbols.end(); SI != SE; ++SI) {
209 std::pair<SymTabType::iterator,bool> Res =
210 symTab.insert(std::make_pair(*SI,filepos));
213 symTabSize += SI->length() +
214 numVbrBytes(SI->length()) +
215 numVbrBytes(filepos);
218 // We don't need this module any more.
221 throw std::string("Can't parse bytecode member: ") +
222 member.getPath().get();
226 // Determine if we actually should compress this member
229 !member.isForeignSymbolTable() &&
230 !member.isLLVMSymbolTable() &&
231 !member.isCompressed() &&
232 !member.isCompressedBytecode());
234 // Perform the compression. Note that if the file is uncompressed bytecode
235 // then we turn the file into compressed bytecode rather than treating it as
236 // compressed data. This is necessary since it allows us to determine that the
237 // file contains bytecode instead of looking like a regular compressed data
238 // member. A compressed bytecode file has its content compressed but has a
239 // magic number of "llvc". This acounts for the +/-4 arithmetic in the code
244 if (member.isBytecode()) {
248 fSize = Compressor::compressToNewBuffer(
249 data,fSize,output,Compressor::COMP_TYPE_ZLIB);
251 if (member.isBytecode())
259 // Compute the fields of the header
260 ArchiveMemberHeader Hdr;
261 bool writeLongName = fillHeader(member,Hdr,hdrSize,TruncateNames);
263 // Write header to archive file
264 ARFile.write((char*)&Hdr, sizeof(Hdr));
266 // Write the long filename if its long
268 ARFile << member.getPath().c_str();
272 // Make sure we write the compressed bytecode magic number if we should.
273 if (willCompress && member.isBytecode())
274 ARFile.write("llvc",4);
276 // Write the (possibly compressed) member's content to the file.
277 ARFile.write(data,fSize);
279 // Make sure the member is an even length
280 if (ARFile.tellp() % 2 != 0)
281 ARFile << ARFILE_PAD;
283 // Free the compressed data, if necessary
288 // Close the mapped file if it was opened
295 // Write out the LLVM symbol table as an archive member to the file.
297 Archive::writeSymbolTable(std::ofstream& ARFile) {
299 // Construct the symbol table's header
300 ArchiveMemberHeader Hdr;
302 memcpy(Hdr.name,ARFILE_LLVM_SYMTAB_NAME,16);
303 uint64_t secondsSinceEpoch = sys::TimeValue::now().toEpochTime();
305 sprintf(buffer,"%-12u", unsigned(secondsSinceEpoch));
306 memcpy(Hdr.date,buffer,12);
307 sprintf(buffer,"%-10u",symTabSize);
308 memcpy(Hdr.size,buffer,10);
311 ARFile.write((char*)&Hdr, sizeof(Hdr));
313 // Save the starting position of the symbol tables data content.
314 unsigned startpos = ARFile.tellp();
316 // Write out the symbols sequentially
317 for ( Archive::SymTabType::iterator I = symTab.begin(), E = symTab.end();
320 // Write out the file index
321 writeInteger(I->second, ARFile);
322 // Write out the length of the symbol
323 writeInteger(I->first.length(), ARFile);
324 // Write out the symbol
325 ARFile.write(I->first.data(), I->first.length());
328 // Now that we're done with the symbol table, get the ending file position
329 unsigned endpos = ARFile.tellp();
331 // Make sure that the amount we wrote is what we pre-computed. This is
332 // critical for file integrity purposes.
333 assert(endpos - startpos == symTabSize && "Invalid symTabSize computation");
335 // Make sure the symbol table is even sized
336 if (symTabSize % 2 != 0 )
337 ARFile << ARFILE_PAD;
340 // Write the entire archive to the file specified when the archive was created.
341 // This writes to a temporary file first. Options are for creating a symbol
342 // table, flattening the file names (no directories, 15 chars max) and
343 // compressing each archive member.
345 Archive::writeToDisk(bool CreateSymbolTable, bool TruncateNames, bool Compress){
347 // Make sure they haven't opened up the file, not loaded it,
348 // but are now trying to write it which would wipe out the file.
349 assert(!(members.empty() && mapfile->size() > 8) &&
350 "Can't write an archive not opened for writing");
352 // Create a temporary file to store the archive in
353 sys::Path TmpArchive = archPath;
354 TmpArchive.createTemporaryFile();
356 // Make sure the temporary gets removed if we crash
357 sys::RemoveFileOnSignal(TmpArchive);
359 // Ensure we can remove the temporary even in the face of an exception
361 // Create archive file for output.
362 std::ofstream ArchiveFile(TmpArchive.c_str());
364 // Check for errors opening or creating archive file.
365 if ( !ArchiveFile.is_open() || ArchiveFile.bad() ) {
366 throw std::string("Error opening archive file: ") + archPath.get();
369 // If we're creating a symbol table, reset it now
370 if (CreateSymbolTable) {
375 // Write magic string to archive.
376 ArchiveFile << ARFILE_MAGIC;
378 // Loop over all member files, and write them out. Note that this also
379 // builds the symbol table, symTab.
380 for ( MembersList::iterator I = begin(), E = end(); I != E; ++I) {
381 writeMember(*I,ArchiveFile,CreateSymbolTable,TruncateNames,Compress);
384 // Close archive file.
387 // Write the symbol table
388 if (CreateSymbolTable) {
389 // At this point we have written a file that is a legal archive but it
390 // doesn't have a symbol table in it. To aid in faster reading and to
391 // ensure compatibility with other archivers we need to put the symbol
392 // table first in the file. Unfortunately, this means mapping the file
393 // we just wrote back in and copying it to the destination file.
395 // Map in the archive we just wrote.
396 sys::MappedFile arch(TmpArchive);
397 const char* base = (const char*) arch.map();
399 // Open the final file to write and check it.
400 std::ofstream FinalFile(archPath.c_str());
401 if ( !FinalFile.is_open() || FinalFile.bad() ) {
402 throw std::string("Error opening archive file: ") + archPath.get();
405 // Write the file magic number
406 FinalFile << ARFILE_MAGIC;
408 // If there is a foreign symbol table, put it into the file now.
410 writeMember(*foreignST, FinalFile, false, false, false);
413 // Put out the LLVM symbol table now.
414 writeSymbolTable(FinalFile);
416 // Copy the temporary file contents being sure to skip the file's magic
418 FinalFile.write(base + sizeof(ARFILE_MAGIC)-1,
419 arch.size()-sizeof(ARFILE_MAGIC)+1);
424 TmpArchive.destroyFile();
427 // We don't have to insert the symbol table, so just renaming the temp
428 // file to the correct name will suffice.
429 TmpArchive.renameFile(archPath);
432 // Make sure we clean up.
433 if (TmpArchive.exists())
434 TmpArchive.destroyFile();