PhpFile *FileRepository::checkoutFile(const std::string &rname, const struct stat &s) {
PhpFile *ret = NULL;
Lock lock(s_lock);
string name;
if (rname[0] == '/') {
name = rname;
} else if (RuntimeOption::SourceRoot.empty()) {
name = Process::GetCurrentDirectory() + "/" + rname;
} else {
name = RuntimeOption::SourceRoot + "/" + rname;
}
hphp_hash_map<string, PhpFile*, string_hash>::iterator it =
m_files.find(name);
if (it == m_files.end()) {
ret = readFile(name, s);
if (ret) {
m_files[name] = ret;
}
} else {
if (it->second->isChanged(s)) {
ret = readFile(name, s);
if (ret) {
it->second->decRef();
it->second = ret;
}
} else {
ret = it->second;
}
}
if (ret) ret->incRef();
return ret;
}
PhpFile *FileRepository::readFile(const string &name,
const struct stat &s,
bool &created) {
created = false;
vector<StaticStatementPtr> sts;
Block::VariableIndices variableIndices;
if (s.st_size > StringData::LenMask) {
throw FatalErrorException(0, "file %s is too big", name.c_str());
}
int fileSize = s.st_size;
int fd = open(name.c_str(), O_RDONLY);
if (!fd) return NULL; // ignore file open exception
char *input = (char *)malloc(fileSize + 1);
if (!input) return NULL;
int nbytes = read(fd, input, fileSize);
close(fd);
input[fileSize] = 0;
String strHelper(input, fileSize, AttachString);
if (nbytes != fileSize) return NULL;
string md5;
string relPath;
string srcRoot;
if (RuntimeOption::SandboxCheckMd5) {
md5 = StringUtil::MD5(strHelper).c_str();
srcRoot = SourceRootInfo::GetCurrentSourceRoot();
if (srcRoot.empty()) srcRoot = RuntimeOption::SourceRoot;
int srcRootLen = srcRoot.size();
if (srcRootLen) {
if (!strncmp(name.c_str(), srcRoot.c_str(), srcRootLen)) {
relPath = string(name.c_str() + srcRootLen);
}
}
hphp_hash_map<string, PhpFile*, string_hash>::iterator it =
s_md5Files.find(md5);
if (it != s_md5Files.end()) {
PhpFile *f = it->second;
if (!relPath.empty() && relPath == f->getRelPath()) {
return it->second;
}
}
}
StatementPtr stmt =
Parser::ParseString(input, name.c_str(), sts, variableIndices);
if (stmt) {
created = true;
PhpFile *p = new PhpFile(stmt, sts, variableIndices,
name, srcRoot, relPath, md5);
return p;
}
return NULL;
}
String FileRepository::translateFileName(StringData *file) {
ParsedFilesMap::const_accessor acc;
if (!s_files.find(acc, file)) return file;
string srcRoot(SourceRootInfo::GetCurrentSourceRoot());
if (srcRoot.empty()) return file;
PhpFile *f = acc->second->getPhpFile();
const string &parsedSrcRoot = f->getSrcRoot();
if (srcRoot == parsedSrcRoot) return file;
int len = parsedSrcRoot.size();
if (len > 0 && file->size() > len &&
strncmp(file->data(), parsedSrcRoot.c_str(), len) == 0) {
return srcRoot + (file->data() + len);
}
return file;
}
String FileRepository::translateFileName(const string &file) {
if (!RuntimeOption::SandboxCheckMd5) return file;
hphp_hash_map<string, PhpFileWrapper*, string_hash>::const_iterator iter =
s_files.find(file);
if (iter == s_files.end()) return file;
string srcRoot(SourceRootInfo::GetCurrentSourceRoot());
if (srcRoot.empty()) srcRoot = RuntimeOption::SourceRoot;
if (srcRoot.empty()) return file;
PhpFile *f = iter->second->getPhpFile();
string parsedSrcRoot = f->getSrcRoot();
if (srcRoot == parsedSrcRoot) return file;
unsigned int len = parsedSrcRoot.size();
if (len > 0 && file.size() > len &&
strncmp(file.data(), parsedSrcRoot.c_str(), len) == 0) {
return srcRoot + file.substr(len);
}
return file;
}
void FileRepository::setFileInfo(const StringData *name,
const string& md5,
FileInfo &fileInfo,
bool fromRepo) {
int md5len;
char* md5str;
// Incorporate the path into the md5 that is used as the key for file
// repository lookups. This assures that even if two PHP files have
// identical content, separate units exist for them (so that
// Unit::filepath() and Unit::dirpath() work correctly).
string s = md5 + '\0' + name->data();
md5str = string_md5(s.c_str(), s.size(), false, md5len);
fileInfo.m_md5 = string(md5str, md5len);
free(md5str);
if (fromRepo) {
fileInfo.m_unitMd5 = md5;
} else {
fileInfo.m_unitMd5 = unitMd5(md5);
}
fileInfo.m_srcRoot = SourceRootInfo::GetCurrentSourceRoot();
int srcRootLen = fileInfo.m_srcRoot.size();
if (srcRootLen) {
if (!strncmp(name->data(), fileInfo.m_srcRoot.c_str(), srcRootLen)) {
fileInfo.m_relPath = string(name->data() + srcRootLen);
}
}
ReadLock lock(s_md5Lock);
Md5FileMap::iterator it = s_md5Files.find(fileInfo.m_md5);
if (it != s_md5Files.end()) {
PhpFile *f = it->second;
if (!fileInfo.m_relPath.empty() &&
fileInfo.m_relPath == f->getRelPath()) {
assert(fileInfo.m_md5 == f->getMd5());
fileInfo.m_phpFile = f;
}
}
}
PhpFile *FileRepository::checkoutFile(const string &rname,
const struct stat &s) {
PhpFile *ret = NULL;
Lock lock(s_lock);
string name;
if (rname[0] == '/') {
name = rname;
} else if (RuntimeOption::SourceRoot.empty()) {
name = Process::GetCurrentDirectory() + "/" + rname;
} else {
name = RuntimeOption::SourceRoot + "/" + rname;
}
hphp_hash_map<string, PhpFileWrapper*, string_hash>::iterator it =
s_files.find(name);
if (it == s_files.end()) {
bool created;
ret = readFile(name, s, created);
if (ret) {
s_files[name] = new PhpFileWrapper(s, ret);
ret->incRef();
if (created) {
if (RuntimeOption::SandboxCheckMd5) {
s_md5Files[ret->getMd5()] = ret;
}
} else {
if (RuntimeOption::SandboxCheckMd5) {
ASSERT(s_md5Files.find(ret->getMd5())->second == ret);
}
}
}
} else if (it->second->isChanged(s)) {
bool created;
ret = readFile(name, s, created);
if (ret) {
PhpFile *f = it->second->getPhpFile();
if (LIKELY(created)) {
assert(ret != f);
} else {
if (RuntimeOption::SandboxCheckMd5) {
ASSERT(s_md5Files.find(ret->getMd5())->second == ret);
}
}
if (LIKELY(f != ret)) {
string oldMd5 = f->getMd5();
if (f->decRef() == 0) {
onDelete(f);
}
if (RuntimeOption::SandboxCheckMd5) {
s_md5Files[ret->getMd5()] = ret;
}
delete(it->second);
it->second = new PhpFileWrapper(s, ret);
ret->incRef();
}
}
} else {
ret = it->second->getPhpFile();
}
if (ret) {
ret->incRef();
}
return ret;
}
PhpFile *FileRepository::checkoutFile(StringData *rname,
const struct stat &s) {
FileInfo fileInfo;
PhpFile *ret = nullptr;
String name(rname);
bool isPlainFile = File::IsPlainFilePath(name);
if (isPlainFile) {
if (rname->data()[0] != '/') {
name = String(SourceRootInfo::GetCurrentSourceRoot()) + name;
}
// Get the common fast path out of the way with as little locking
// as possible: it's in the map and has not changed on disk
ParsedFilesMap::const_accessor acc;
if (s_files.find(acc, name.get()) && !acc->second->isChanged(s)) {
TRACE(1, "FR fast path hit %s\n", rname->data());
ret = acc->second->getPhpFile();
return ret;
}
} else {
// Do the read before we get the repo lock, since it will call into the
// stream library and will needs its own locks.
if (isAuthoritativeRepo()) {
throw FatalErrorException(
"including urls doesn't work in RepoAuthoritative mode"
);
}
Stream::Wrapper* w = Stream::getWrapperFromURI(name);
File* f = w->open(name, "r", 0, null_variant);
if (!f) return nullptr;
StringBuffer sb;
sb.read(f);
fileInfo.m_inputString = sb.detach();
computeMd5(name.get(), fileInfo);
}
TRACE(1, "FR fast path miss: %s\n", rname->data());
const StringData *n = makeStaticString(name.get());
PhpFile* toKill = nullptr;
SCOPE_EXIT {
// run this after acc is destroyed (and its lock released)
if (toKill) toKill->decRefAndDelete();
};
ParsedFilesMap::accessor acc;
bool isNew = s_files.insert(acc, n);
PhpFileWrapper* old = acc->second;
SCOPE_EXIT {
// run this just before acc is released
if (old && old != acc->second) {
if (old->getPhpFile() != acc->second->getPhpFile()) {
toKill = old->getPhpFile();
}
delete old;
}
if (!acc->second) s_files.erase(acc);
};
assert(isNew || old); // We don't leave null entries around.
bool isChanged = !isNew && old->isChanged(s);
if (isNew || isChanged) {
if (isPlainFile && !readFile(n, s, fileInfo)) {
TRACE(1, "File disappeared between stat and FR::readNewFile: %s\n",
rname->data());
return nullptr;
}
ret = fileInfo.m_phpFile;
if (isChanged && ret == old->getPhpFile()) {
// The file changed but had the same contents.
if (debug && md5Enabled()) {
ReadLock lock(s_md5Lock);
assert(s_md5Files.find(ret->getMd5())->second == ret);
}
return ret;
}
} else if (!isNew) {
// Somebody might have loaded the file between when we dropped
// our read lock and got the write lock
ret = old->getPhpFile();
return ret;
}
// If we get here the file was not in s_files or has changed on disk
if (!ret) {
// Try to read Unit from .hhbc repo.
ret = readHhbc(n->data(), fileInfo);
}
if (!ret) {
if (isAuthoritativeRepo()) {
raise_error("Tried to parse %s in repo authoritative mode", n->data());
}
ret = parseFile(n->data(), fileInfo);
if (!ret) return nullptr;
}
assert(ret != nullptr);
if (isNew) {
acc->second = new PhpFileWrapper(s, ret);
ret->incRef();
ret->setId(Transl::TargetCache::allocBit());
} else {
PhpFile *f = old->getPhpFile();
if (f != ret) {
ret->setId(f->getId());
ret->incRef();
}
acc->second = new PhpFileWrapper(s, ret);
}
if (md5Enabled()) {
WriteLock lock(s_md5Lock);
s_md5Files[ret->getMd5()] = ret;
}
return ret;
}
PhpFile *FileRepository::checkoutFile(StringData *rname,
const struct stat &s) {
FileInfo fileInfo;
PhpFile *ret = nullptr;
String name(rname);
if (rname->data()[0] != '/') {
name = String(SourceRootInfo::GetCurrentSourceRoot()) + name;
}
{
// Get the common fast path out of the way with as little locking
// as possible: it's in the map and has not changed on disk
ParsedFilesMap::const_accessor acc;
if (s_files.find(acc, name.get()) && !acc->second->isChanged(s)) {
TRACE(1, "FR fast path hit %s\n", rname->data());
ret = acc->second->getPhpFile();
ret->incRef();
return ret;
}
}
TRACE(1, "FR fast path miss: %s\n", rname->data());
bool interceptsEnabled = s_interceptsEnabled;
const StringData *n = StringData::GetStaticString(name.get());
ParsedFilesMap::accessor acc;
bool isNew = s_files.insert(acc, n);
assert(isNew || acc->second); // We don't leave null entries around.
bool isChanged = !isNew && acc->second->isChanged(s);
if (isNew || isChanged) {
if (!readFile(n, s, fileInfo)) {
// Be sure to get rid of the new reference to it.
s_files.erase(acc);
TRACE(1, "File disappeared between stat and FR::readNewFile: %s\n",
rname->data());
return nullptr;
}
ret = fileInfo.m_phpFile;
if (isChanged && ret == acc->second->getPhpFile()) {
// The file changed but had the same contents.
if (debug && md5Enabled()) {
ReadLock lock(s_md5Lock);
assert(s_md5Files.find(ret->getMd5())->second == ret);
}
ret->incRef();
return ret;
}
} else if (!isNew) {
// Somebody might have loaded the file between when we dropped
// our read lock and got the write lock
ret = acc->second->getPhpFile();
ret->incRef();
return ret;
}
// If we get here the file was not in s_files or has changed on disk
if (!ret) {
// Try to read Unit from .hhbc repo.
ret = readHhbc(n->data(), fileInfo);
}
if (!ret) {
if (isAuthoritativeRepo()) {
raise_error("Tried to parse %s in repo authoritative mode", n->data());
}
ret = parseFile(n->data(), fileInfo);
if (!ret) {
s_files.erase(acc);
return nullptr;
}
}
assert(ret != nullptr);
if (isNew) {
acc->second = new PhpFileWrapper(s, ret);
ret->incRef();
ret->setId(VM::Transl::TargetCache::allocBit());
} else {
PhpFile *f = acc->second->getPhpFile();
if (f != ret) {
ret->setId(f->getId());
tx64->invalidateFile(f); // f has changed
}
f->decRefAndDelete();
delete acc->second;
acc->second = new PhpFileWrapper(s, ret);
ret->incRef();
}
if (md5Enabled()) {
WriteLock lock(s_md5Lock);
// make sure intercepts are enabled for the functions within the
// new units
// Since we have the write lock on s_md5lock, s_interceptsEnabled
// can't change, and we are serialized wrt enableIntercepts
// (i.e., this will execute either before or after
// enableIntercepts).
if (interceptsEnabled != s_interceptsEnabled) {
// intercepts were enabled since the time we created the unit
ret->unit()->enableIntercepts();
}
s_md5Files[ret->getMd5()] = ret;
}
ret->incRef();
return ret;
}
