OpMsg OpMsg::parse(const Message& message) try {
// It is the caller's responsibility to call the correct parser for a given message type.
invariant(!message.empty());
invariant(message.operation() == dbMsg);
const uint32_t flags = OpMsg::flags(message);
uassert(ErrorCodes::IllegalOpMsgFlag,
str::stream() << "Message contains illegal flags value: Ob"
<< std::bitset<32>(flags).to_string(),
!containsUnknownRequiredFlags(flags));
constexpr int kCrc32Size = 4;
const bool haveChecksum = flags & kChecksumPresent;
const int checksumSize = haveChecksum ? kCrc32Size : 0;
// The sections begin after the flags and before the checksum (if present).
BufReader sectionsBuf(message.singleData().data() + sizeof(flags),
message.dataSize() - sizeof(flags) - checksumSize);
// TODO some validation may make more sense in the IDL parser. I've tagged them with comments.
bool haveBody = false;
OpMsg msg;
while (!sectionsBuf.atEof()) {
const auto sectionKind = sectionsBuf.read<Section>();
switch (sectionKind) {
case Section::kBody: {
uassert(40430, "Multiple body sections in message", !haveBody);
haveBody = true;
msg.body = sectionsBuf.read<Validated<BSONObj>>();
break;
}
case Section::kDocSequence: {
// We use an O(N^2) algorithm here and an O(N*M) algorithm below. These are fastest
// for the current small values of N, but would be problematic if it is large.
// If we need more document sequences, raise the limit and use a better algorithm.
uassert(ErrorCodes::TooManyDocumentSequences,
"Too many document sequences in OP_MSG",
msg.sequences.size() < 2); // Limit is <=2 since we are about to add one.
// The first 4 bytes are the total size, including themselves.
const auto remainingSize =
sectionsBuf.read<LittleEndian<int32_t>>() - sizeof(int32_t);
BufReader seqBuf(sectionsBuf.skip(remainingSize), remainingSize);
const auto name = seqBuf.readCStr();
uassert(40431,
str::stream() << "Duplicate document sequence: " << name,
!msg.getSequence(name)); // TODO IDL
msg.sequences.push_back({name.toString()});
while (!seqBuf.atEof()) {
msg.sequences.back().objs.push_back(seqBuf.read<Validated<BSONObj>>());
}
break;
}
default:
// Using uint32_t so we append as a decimal number rather than as a char.
uasserted(40432, str::stream() << "Unknown section kind " << uint32_t(sectionKind));
}
}
uassert(40587, "OP_MSG messages must have a body", haveBody);
// Detect duplicates between doc sequences and body. TODO IDL
// Technically this is O(N*M) but N is at most 2.
for (const auto& docSeq : msg.sequences) {
const char* name = docSeq.name.c_str(); // Pointer is redirected by next call.
auto inBody =
!dotted_path_support::extractElementAtPathOrArrayAlongPath(msg.body, name).eoo();
uassert(40433,
str::stream() << "Duplicate field between body and document sequence "
<< docSeq.name,
!inBody);
}
return msg;
} catch (const DBException& ex) {
LOG(1) << "invalid message: " << ex.code() << " " << redact(ex) << " -- "
<< redact(hexdump(message.singleData().view2ptr(), message.size()));
throw;
}
OpMsg OpMsg::parse(const Message& message) try {
// TODO some validation may make more sense in the IDL parser. I've tagged them with comments.
OpMsg msg;
// Use a separate BufReader for the flags since the flags can change how much room we have
// for sections.
BufReader(message.singleData().data(), message.dataSize()).read(msg.flags);
uassert(40429,
str::stream() << "Message contains illegal flags value: " << msg.flags,
!containsUnknownRequiredFlags(msg.flags));
invariant(!msg.isFlagSet(kChecksumPresent)); // TODO SERVER-28679 check checksum here.
constexpr int kCrc32Size = 4;
const int checksumSize = msg.isFlagSet(kChecksumPresent) ? kCrc32Size : 0;
BufReader sectionsBuf(message.singleData().data() + sizeof(msg.flags),
message.dataSize() - sizeof(msg.flags) - checksumSize);
bool haveBody = false;
while (!sectionsBuf.atEof()) {
const auto sectionKind = sectionsBuf.read<Section>();
switch (sectionKind) {
case Section::kBody: {
uassert(40430, "Multiple body sections in message", !haveBody);
haveBody = true;
msg.body = sectionsBuf.read<Validated<BSONObj>>();
break;
}
case Section::kDocSequence: {
// The first 4 bytes are the total size, including themselves.
const auto remainingSize = sectionsBuf.read<int32_t>() - sizeof(int32_t);
BufReader seqBuf(sectionsBuf.skip(remainingSize), remainingSize);
const auto name = seqBuf.readCStr();
uassert(40431,
str::stream() << "Duplicate document sequence: " << name,
!msg.getSequence(name)); // TODO IDL
msg.sequences.push_back({name.toString()});
while (!seqBuf.atEof()) {
msg.sequences.back().objs.push_back(seqBuf.read<Validated<BSONObj>>());
}
break;
}
default:
// Using uint32_t so we append as a decimal number rather than as a char.
uasserted(40432, str::stream() << "Unknown section kind " << uint32_t(sectionKind));
}
}
// Detect duplicates between doc sequences and body. TODO IDL
// Technically this is O(N*M) but N is at most 2.
for (const auto& docSeq : msg.sequences) {
const char* name = docSeq.name.c_str(); // Pointer is redirected by next call.
auto inBody =
!dotted_path_support::extractElementAtPathOrArrayAlongPath(msg.body, name).eoo();
uassert(40433,
str::stream() << "Duplicate field between body and document sequence "
<< docSeq.name,
!inBody);
}
return msg;
} catch (const DBException& ex) {
// TODO change to LOG(1).
log() << "invalid message: " << redact(ex) << ' '
<< hexdump(message.singleData().view2ptr(), message.size());
throw;
}
