1 //===-- ArchiveReader.cpp - Read LLVM archive files -------------*- C++ -*-===//
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 standard unix archive files (.a) containing LLVM bytecode.
12 //===----------------------------------------------------------------------===//
14 #include "ArchiveInternals.h"
15 #include "llvm/Bytecode/Reader.h"
19 /// Read a variable-bit-rate encoded unsigned integer
20 inline unsigned readInteger(const char*&At, const char*End) {
26 throw std::string("Ran out of data reading vbr_uint!");
27 Result |= (unsigned)((*At++) & 0x7F) << Shift;
29 } while (At[-1] & 0x80);
33 // Completely parse the Archive's symbol table and populate symTab member var.
35 Archive::parseSymbolTable(const void* data, unsigned size) {
36 const char* At = (const char*) data;
37 const char* End = At + size;
39 unsigned offset = readInteger(At, End);
40 unsigned length = readInteger(At, End);
41 if (At + length > End)
42 throw std::string("malformed symbol table");
43 // we don't care if it can't be inserted (duplicate entry)
44 symTab.insert(std::make_pair(std::string(At, length), offset));
50 // This member parses an ArchiveMemberHeader that is presumed to be pointed to
51 // by At. The At pointer is updated to the byte just after the header, which
52 // can be variable in size.
54 Archive::parseMemberHeader(const char*& At, const char* End) {
55 assert(At + sizeof(ArchiveMemberHeader) < End && "Not enough data");
57 // Cast archive member header
58 ArchiveMemberHeader* Hdr = (ArchiveMemberHeader*)At;
59 At += sizeof(ArchiveMemberHeader);
61 // Instantiate the ArchiveMember to be filled
62 ArchiveMember* member = new ArchiveMember(this);
64 // Extract the size and determine if the file is
65 // compressed or not (negative length).
67 int MemberSize = atoi(Hdr->size);
69 flags |= ArchiveMember::CompressedFlag;
70 MemberSize = -MemberSize;
73 // Check the size of the member for sanity
74 if (At + MemberSize > End)
75 throw std::string("invalid member length in archive file");
77 // Check the member signature
78 if (!Hdr->checkSignature())
79 throw std::string("invalid file member signature");
81 // Convert and check the member name
82 // The empty name ( '/' and 15 blanks) is for a foreign (non-LLVM) symbol
83 // table. The special name "//" and 14 blanks is for a string table, used
84 // for long file names. This library doesn't generate either of those but
85 // it will accept them. If the name starts with #1/ and the remainder is
86 // digits, then those digits specify the length of the name that is
87 // stored immediately following the header. The special name
88 // __LLVM_SYM_TAB__ identifies the symbol table for LLVM bytecode.
89 // Anything else is a regular, short filename that is terminated with
94 switch (Hdr->name[0]) {
96 if (Hdr->name[1] == '1' && Hdr->name[2] == '/') {
97 if (isdigit(Hdr->name[3])) {
98 unsigned len = atoi(&Hdr->name[3]);
99 pathname.assign(At, len);
102 flags |= ArchiveMember::HasLongFilenameFlag;
104 throw std::string("invalid long filename");
105 } else if (Hdr->name[1] == '_' &&
106 (0 == memcmp(Hdr->name, ARFILE_LLVM_SYMTAB_NAME, 16))) {
107 // The member is using a long file name (>15 chars) format.
108 // This format is standard for 4.4BSD and Mac OSX operating
109 // systems. LLVM uses it similarly. In this format, the
110 // remainder of the name field (after #1/) specifies the
111 // length of the file name which occupy the first bytes of
112 // the member's data. The pathname already has the #1/ stripped.
113 pathname.assign(ARFILE_LLVM_SYMTAB_NAME);
114 flags |= ArchiveMember::LLVMSymbolTableFlag;
118 if (Hdr->name[1]== '/') {
119 if (0 == memcmp(Hdr->name, ARFILE_STRTAB_NAME, 16)) {
120 pathname.assign(ARFILE_STRTAB_NAME);
121 flags |= ArchiveMember::StringTableFlag;
123 throw std::string("invalid string table name");
125 } else if (Hdr->name[1] == ' ') {
126 if (0 == memcmp(Hdr->name, ARFILE_SVR4_SYMTAB_NAME, 16)) {
127 pathname.assign(ARFILE_SVR4_SYMTAB_NAME);
128 flags |= ArchiveMember::SVR4SymbolTableFlag;
130 throw std::string("invalid SVR4 symbol table name");
132 } else if (isdigit(Hdr->name[1])) {
133 unsigned index = atoi(&Hdr->name[1]);
134 if (index < strtab.length()) {
135 const char* namep = strtab.c_str() + index;
136 const char* endp = strtab.c_str() + strtab.length();
137 const char* p = namep;
138 const char* last_p = p;
140 if (*p == '\n' && *last_p == '/') {
141 pathname.assign(namep, last_p - namep);
142 flags |= ArchiveMember::HasLongFilenameFlag;
149 throw std::string("missing name termiantor in string table");
151 throw std::string("name index beyond string table");
156 if (Hdr->name[1] == '_' &&
157 (0 == memcmp(Hdr->name, ARFILE_BSD4_SYMTAB_NAME, 16))) {
158 pathname.assign(ARFILE_BSD4_SYMTAB_NAME);
159 flags |= ArchiveMember::BSD4SymbolTableFlag;
165 char* slash = (char*) memchr(Hdr->name, '/', 16);
167 slash = Hdr->name + 16;
168 pathname.assign(Hdr->name, slash - Hdr->name);
172 // Determine if this is a bytecode file
173 switch (sys::IdentifyFileType(At, 4)) {
174 case sys::BytecodeFileType:
175 flags |= ArchiveMember::BytecodeFlag;
177 case sys::CompressedBytecodeFileType:
178 flags |= ArchiveMember::CompressedBytecodeFlag;
179 flags &= ~ArchiveMember::CompressedFlag;
182 flags &= ~(ArchiveMember::BytecodeFlag|
183 ArchiveMember::CompressedBytecodeFlag);
187 // Fill in fields of the ArchiveMember
190 member->parent = this;
191 member->path.setFile(pathname);
192 member->info.fileSize = MemberSize;
193 member->info.modTime.fromEpochTime(atoi(Hdr->date));
194 sscanf(Hdr->mode, "%o", &(member->info.mode));
195 member->info.user = atoi(Hdr->uid);
196 member->info.group = atoi(Hdr->gid);
197 member->flags = flags;
204 Archive::checkSignature() {
205 // Check the magic string at file's header
206 if (mapfile->size() < 8 || memcmp(base, ARFILE_MAGIC, 8))
207 throw std::string("invalid signature for an archive file");
210 // This function loads the entire archive and fully populates its ilist with
211 // the members of the archive file. This is typically used in preparation for
212 // editing the contents of the archive.
214 Archive::loadArchive() {
219 const char *At = base;
220 const char *End = base + mapfile->size();
223 At += 8; // Skip the magic string.
225 bool seenSymbolTable = false;
226 bool foundFirstFile = false;
228 // parse the member header
229 const char* Save = At;
230 ArchiveMember* mbr = parseMemberHeader(At, End);
232 // check if this is the foreign symbol table
233 if (mbr->isSVR4SymbolTable() || mbr->isBSD4SymbolTable()) {
234 // We just save this but don't do anything special
235 // with it. It doesn't count as the "first file".
237 // What? Multiple foreign symbol tables? Just chuck it
238 // and retain the last one found.
242 At += mbr->getSize();
243 if ((intptr_t(At) & 1) == 1)
245 } else if (mbr->isStringTable()) {
246 // Simply suck the entire string table into a string
247 // variable. This will be used to get the names of the
248 // members that use the "/ddd" format for their names
249 // (SVR4 style long names).
250 strtab.assign(At, mbr->getSize());
251 At += mbr->getSize();
252 if ((intptr_t(At) & 1) == 1)
255 } else if (mbr->isLLVMSymbolTable()) {
256 // This is the LLVM symbol table for the archive. If we've seen it
257 // already, its an error. Otherwise, parse the symbol table and move on.
259 throw std::string("invalid archive: multiple symbol tables");
260 parseSymbolTable(mbr->getData(), mbr->getSize());
261 seenSymbolTable = true;
262 At += mbr->getSize();
263 if ((intptr_t(At) & 1) == 1)
265 delete mbr; // We don't need this member in the list of members.
267 // This is just a regular file. If its the first one, save its offset.
268 // Otherwise just push it on the list and move on to the next file.
269 if (!foundFirstFile) {
270 firstFileOffset = Save - base;
271 foundFirstFile = true;
273 members.push_back(mbr);
274 At += mbr->getSize();
275 if ((intptr_t(At) & 1) == 1)
281 // Open and completely load the archive file.
283 Archive::OpenAndLoad(const sys::Path& file) {
285 Archive* result = new Archive(file, true);
287 result->loadArchive();
292 // Get all the bytecode modules from the archive
294 Archive::getAllModules(std::vector<Module*>& Modules, std::string* ErrMessage) {
296 for (iterator I=begin(), E=end(); I != E; ++I) {
297 if (I->isBytecode() || I->isCompressedBytecode()) {
298 std::string FullMemberName = archPath.get() +
299 "(" + I->getPath().get() + ")";
300 Module* M = ParseBytecodeBuffer((const unsigned char*)I->getData(),
301 I->getSize(), FullMemberName, ErrMessage);
305 Modules.push_back(M);
311 // Load just the symbol table from the archive file
313 Archive::loadSymbolTable() {
318 const char *At = base;
319 const char *End = base + mapfile->size();
321 // Make sure we're dealing with an archive
324 At += 8; // Skip signature
326 // Parse the first file member header
327 const char* FirstFile = At;
328 ArchiveMember* mbr = parseMemberHeader(At, End);
330 if (mbr->isSVR4SymbolTable() || mbr->isBSD4SymbolTable()) {
331 // Skip the foreign symbol table, we don't do anything with it
332 At += mbr->getSize();
333 if ((intptr_t(At) & 1) == 1)
339 mbr = parseMemberHeader(At, End);
342 if (mbr->isStringTable()) {
343 // Process the string table entry
344 strtab.assign((const char*)mbr->getData(), mbr->getSize());
345 At += mbr->getSize();
346 if ((intptr_t(At) & 1) == 1)
351 mbr = parseMemberHeader(At, End);
354 // See if its the symbol table
355 if (mbr->isLLVMSymbolTable()) {
356 parseSymbolTable(mbr->getData(), mbr->getSize());
357 FirstFile = At + mbr->getSize();
358 if ((intptr_t(At) & 1) == 1)
361 // There's no symbol table in the file. We have to rebuild it from scratch
362 // because the intent of this method is to get the symbol table loaded so
363 // it can be searched efficiently.
364 // Add the member to the members list
365 members.push_back(mbr);
368 firstFileOffset = FirstFile - base;
371 // Open the archive and load just the symbol tables
373 Archive::OpenAndLoadSymbols(const sys::Path& file) {
374 Archive* result = new Archive(file, true);
376 result->loadSymbolTable();
381 // Look up one symbol in the symbol table and return a ModuleProvider for the
382 // module that defines that symbol.
384 Archive::findModuleDefiningSymbol(const std::string& symbol) {
385 SymTabType::iterator SI = symTab.find(symbol);
386 if (SI == symTab.end())
389 // The symbol table was previously constructed assuming that the members were
390 // written without the symbol table header. Because VBR encoding is used, the
391 // values could not be adjusted to account for the offset of the symbol table
392 // because that could affect the size of the symbol table due to VBR encoding.
393 // We now have to account for this by adjusting the offset by the size of the
394 // symbol table and its header.
395 unsigned fileOffset =
396 SI->second + // offset in symbol-table-less file
397 firstFileOffset; // add offset to first "real" file in archive
399 // See if the module is already loaded
400 ModuleMap::iterator MI = modules.find(fileOffset);
401 if (MI != modules.end())
402 return MI->second.first;
404 // Module hasn't been loaded yet, we need to load it
405 const char* modptr = base + fileOffset;
406 ArchiveMember* mbr = parseMemberHeader(modptr, base + mapfile->size());
408 // Now, load the bytecode module to get the ModuleProvider
409 std::string FullMemberName = archPath.get() + "(" +
410 mbr->getPath().get() + ")";
411 ModuleProvider* mp = getBytecodeBufferModuleProvider(
412 (const unsigned char*) mbr->getData(), mbr->getSize(),
415 modules.insert(std::make_pair(fileOffset, std::make_pair(mp, mbr)));
420 // Look up multiple symbols in the symbol table and return a set of
421 // ModuleProviders that define those symbols.
423 Archive::findModulesDefiningSymbols(std::set<std::string>& symbols,
424 std::set<ModuleProvider*>& result)
426 assert(mapfile && base && "Can't findModulesDefiningSymbols on new archive");
427 if (symTab.empty()) {
428 // We don't have a symbol table, so we must build it now but lets also
429 // make sure that we populate the modules table as we do this to ensure
430 // that we don't load them twice when findModuleDefiningSymbol is called
433 // Get a pointer to the first file
434 const char* At = ((const char*)base) + firstFileOffset;
435 const char* End = ((const char*)base) + mapfile->size();
438 // Compute the offset to be put in the symbol table
439 unsigned offset = At - base - firstFileOffset;
441 // Parse the file's header
442 ArchiveMember* mbr = parseMemberHeader(At, End);
444 // If it contains symbols
445 if (mbr->isBytecode() || mbr->isCompressedBytecode()) {
447 std::vector<std::string> symbols;
448 std::string FullMemberName = archPath.get() + "(" +
449 mbr->getPath().get() + ")";
450 ModuleProvider* MP = GetBytecodeSymbols((const unsigned char*)At,
451 mbr->getSize(), FullMemberName, symbols);
454 // Insert the module's symbols into the symbol table
455 for (std::vector<std::string>::iterator I = symbols.begin(),
456 E=symbols.end(); I != E; ++I ) {
457 symTab.insert(std::make_pair(*I, offset));
459 // Insert the ModuleProvider and the ArchiveMember into the table of
461 modules.insert(std::make_pair(offset, std::make_pair(MP, mbr)));
463 throw std::string("Can't parse bytecode member: ") +
464 mbr->getPath().get();
468 // Go to the next file location
469 At += mbr->getSize();
470 if ((intptr_t(At) & 1) == 1)
475 // At this point we have a valid symbol table (one way or another) so we
476 // just use it to quickly find the symbols requested.
478 for (std::set<std::string>::iterator I=symbols.begin(),
479 E=symbols.end(); I != E;) {
480 // See if this symbol exists
481 ModuleProvider* mp = findModuleDefiningSymbol(*I);
483 // The symbol exists, insert the ModuleProvider into our result,
484 // duplicates wil be ignored
487 // Remove the symbol now that its been resolved, being careful to
488 // post-increment the iterator.