// Write one member out to the file.
bool
Archive::writeMember(
const ArchiveMember& member,
raw_fd_ostream& ARFile,
std::string* ErrMsg
) {
uint64_t filepos = ARFile.tell();
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) {
ErrorOr<std::unique_ptr<MemoryBuffer> > FileOrErr =
MemoryBuffer::getFile(member.getPath());
if (!FileOrErr) {
if (ErrMsg)
*ErrMsg = FileOrErr.getError().message();
return true;
}
mFile = FileOrErr.get().release();
data = mFile->getBufferStart();
fSize = mFile->getBufferSize();
}
int hdrSize = fSize;
// Compute the fields of the header
ArchiveMemberHeader Hdr;
bool writeLongName = fillHeader(member,Hdr,hdrSize);
// Write header to archive file
ARFile.write((char*)&Hdr, sizeof(Hdr));
// Write the long filename if its long
if (writeLongName) {
StringRef Name = sys::path::filename(member.getPath());
ARFile.write(Name.data(), Name.size());
}
// Write the (possibly compressed) member's content to the file.
ARFile.write(data,fSize);
// Make sure the member is an even length
if ((ARFile.tell() & 1) == 1)
ARFile << ARFILE_PAD;
// Close the mapped file if it was opened
delete mFile;
return false;
}
// 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;
}
// 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) 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);
std::string mbrPath = sys::path::filename(mbr.getPath());
// 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 (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;
}
// 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();
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()) {
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;
}
bool Archive::getMembersBySymbol(std::string name,
std::vector<Symtab *> &matches) {
if (!symbolTableParsed && !parseSymbolTable())
return false;
std::pair<std::multimap<string, ArchiveMember *>::iterator,
std::multimap<string, ArchiveMember *>::iterator> range_it;
range_it = membersBySymbol.equal_range(name);
auto begin = range_it.first;
auto end = range_it.second;
for (; begin != end; ++begin) {
ArchiveMember *member = begin->second;
Symtab *img = member->getSymtab();
if (!img && !parseMember(img, member)) return false;
matches.push_back(img);
}
return true;
}
bool Archive::getMemberByGlobalSymbol(Symtab *&img, string& symbol_name)
{
if( !symbolTableParsed ) {
if( !parseSymbolTable() ) {
return false;
}
}
std::pair<std::multimap<string, ArchiveMember *>::iterator,
std::multimap<string, ArchiveMember *>::iterator> range_it;
range_it = membersBySymbol.equal_range(symbol_name);
// Symbol not found in symbol table
if( range_it.first == range_it.second ) {
serr = No_Such_Member;
errMsg = MEMBER_DNE;
return false;
}
ArchiveMember *foundMember = range_it.first->second;
// Duplicate symbol found in symbol table
++(range_it.first);
if( range_it.first != range_it.second ) {
serr = Duplicate_Symbol;
errMsg = symbol_name;
return false;
}
img = foundMember->getSymtab();
if( img == NULL ) {
if( !parseMember(img, foundMember) ) {
return false;
}
}
return true;
}
// Load just the symbol table from the archive file
bool
Archive::loadSymbolTable(std::string* ErrorMsg) {
// Set up parsing
members.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;
}
// 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 (mbr.isLLVMSymbolTable()) {
memcpy(hdr.name,ARFILE_LLVM_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;
}
// Write one member out to the file.
bool
Archive::writeMember(
const ArchiveMember& member,
std::ofstream& ARFile,
bool CreateSymbolTable,
bool TruncateNames,
bool ShouldCompress,
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) {
mFile = MemoryBuffer::getFile(member.getPath().c_str(), ErrMsg);
if (mFile == 0)
return true;
data = mFile->getBufferStart();
fSize = mFile->getBufferSize();
}
// Now that we have the data in memory, update the
// symbol table if its a bitcode file.
if (CreateSymbolTable && member.isBitcode()) {
std::vector<std::string> symbols;
std::string FullMemberName = archPath.str() + "(" + member.getPath().str()
+ ")";
Module* M =
GetBitcodeSymbols((const unsigned char*)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;
}
// 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;
}
// 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 seenSymbolTable = false;
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 if (mbr->isLLVMSymbolTable()) {
// This is the LLVM symbol table for the archive. If we've seen it
// already, its an error. Otherwise, parse the symbol table and move on.
if (seenSymbolTable) {
if (error)
*error = "invalid archive: multiple symbol tables";
return false;
}
if (!parseSymbolTable(mbr->getData(), mbr->getSize(), error))
return false;
seenSymbolTable = true;
At += mbr->getSize();
if ((intptr_t(At) & 1) == 1)
At++;
delete mbr; // We don't need this member in the list of members.
} 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;
}
// Look up multiple symbols in the symbol table and return a set of
// ModuleProviders that define those symbols.
bool
Archive::findModulesDefiningSymbols(std::set<std::string>& symbols,
std::set<ModuleProvider*>& 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.toString() + "(" +
mbr->getPath().toString() + ")";
ModuleProvider* MP =
GetBitcodeSymbols((const unsigned char*)At, mbr->getSize(),
FullMemberName, symbols, error);
if (MP) {
// 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 ModuleProvider and the ArchiveMember into the table of
// modules.
modules.insert(std::make_pair(offset, std::make_pair(MP, mbr)));
} else {
if (error)
*error = "Can't parse bitcode member: " +
mbr->getPath().toString() + ": " + *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.
for (std::set<std::string>::iterator I=symbols.begin(),
E=symbols.end(); I != E;) {
// See if this symbol exists
ModuleProvider* mp = findModuleDefiningSymbol(*I,error);
if (mp) {
// The symbol exists, insert the ModuleProvider into our result,
// duplicates wil be ignored
result.insert(mp);
// Remove the symbol now that its been resolved, being careful to
// post-increment the iterator.
symbols.erase(I++);
} else {
++I;
}
}
return true;
}