First off, in that case it seems better to read explicit sizes of data instead of read_some which reads whatever is available.
In this pattern, it becomes easier to track "remaining bytes receivable" than m_fileSize.
Here's some minor re-shufflings that made your code into a self-contained example. It expects a server to send a line of text giving the payload size and output filename, followed by the contents of that file. An example server can be run with netcat e.g.:
(stat -c '%soutput.dat' main.cpp; cat main.cpp) | netcat -l -p 6969
Live On Coliru
#include <boost/asio.hpp>
#include <fstream>
#include <iostream>
using boost::system::error_code;
using boost::asio::ip::tcp;
struct Session : std::enable_shared_from_this<Session> {
Session(boost::asio::io_context& io, uint16_t port)
: m_socket(io)
{
m_socket.connect({{}, port});
}
void Start();
void DoReadFileContent(size_t transferred = 0);
private:
std::array<char, 1024> m_buffer;
std::streamsize m_remainingSize = 0;
std::string m_fileName = "noname.dat";
std::ofstream m_outputFile;
tcp::socket m_socket;
};
void Session::Start() {
// Reading a size (in text for simplicity) and subsequently receive as many bytes
//
// I'm keeping this sync for simplicity, because you probably already have
// this coded somehwere
boost::asio::streambuf buf;
error_code ec;
auto n = read_until(m_socket, buf, "\n", ec);
std::istream is(&buf);
if (is >> m_remainingSize && getline(is, m_fileName)) {
std::cerr << "Protocol trace: n:" << n << ", fileName:" << m_fileName << " payload_size:" << m_remainingSize << "\n";
m_outputFile.exceptions(std::ios::failbit | std::ios::badbit);
m_outputFile.open(m_fileName, std::ios::binary);
// write excess buffer contents as part of payload
if (buf.size()) {
std::cerr << "Writing " << buf.size() << " bytes\n";
m_remainingSize -= buf.size();
m_outputFile << &buf;
}
DoReadFileContent();
} else {
std::cerr << "Protocol error, payload_size expected\n";
}
}
void Session::DoReadFileContent(size_t transferred) {
if (transferred > 0) {
std::cerr << "Writing " << transferred << " bytes\n";
m_remainingSize -= transferred;
m_outputFile.write(m_buffer.data(), transferred);
}
if (m_remainingSize <= 0) {
std::cout << "Completed file: " << m_fileName << std::endl;
return;
}
auto self = shared_from_this();
auto expect = std::min(size_t(m_remainingSize), m_buffer.size());
std::cout << "Trying to receive next " << expect << " bytes" << std::endl;
async_read(m_socket,
boost::asio::buffer(m_buffer.data(), expect),
[this, self](error_code ec, size_t arg_bytes) {
std::cerr << "async_read: " << ec.message() << " - " << arg_bytes << " bytes\n";
if (!ec) {
DoReadFileContent(arg_bytes);
}
});
}
int main() {
boost::asio::io_context io;
std::make_shared<Session>(io, 6868) // download from port 6868
->Start();
io.run(); // complete
}
Testing with
(stat -c '%soutput.dat' main.cpp; cat main.cpp) | netcat -l -p 6868&
./a.out
md5sum main.cpp output.dat
Prints, e.g.:
Protocol trace: n:15, fileName:output.dat payload_size:2654
Trying to receive next 1024 bytes
async_read: Success - 1024 bytes
Writing 1024 bytes
Trying to receive next 1024 bytes
async_read: Success - 1024 bytes
Writing 1024 bytes
Trying to receive next 606 bytes
async_read: Success - 606 bytes
Writing 606 bytes
Completed file: output.dat
The last two lines
b4eec7203f6a1dcbfbf3d298c7ec0832 main.cpp
b4eec7203f6a1dcbfbf3d298c7ec0832 output.dat
indicate that the received file is identical to the original.
Notes:
packets are delivered in unspecified sizes, on my system e.g. the same file is received as:
Protocol trace: n:15, fileName:output.dat payload_size:2654
Writing 497 bytes
Trying to receive next 1024 bytes
async_read: Success - 1024 bytes
Writing 1024 bytes
Trying to receive next 1024 bytes
async_read: Success - 1024 bytes
Writing 1024 bytes
Trying to receive next 109 bytes
async_read: Success - 109 bytes
Writing 109 bytes
Completed file: output.dat
b4eec7203f6a1dcbfbf3d298c7ec0832 main.cpp
b4eec7203f6a1dcbfbf3d298c7ec0832 output.dat
Note that it starts out with 497 bytes already in the input buffer from the read_until.
- The protocol is not secure:
- the file names should be validated. Just imagine what happens if the file would be '/home/sehe/myimportant_file.txt' or worse, say /dev/sde1 and we have permissions to do raw block device access...
- you might want to specify a amximum size for streambuf, so that if you get a fuzzer that doesn't ever send a
'\n' you wouldn't just gobble up all RAM
- the error handling on file IO is very rough. I used io exceptions, but you probably want to check for
m_outputFile.good() instead at various places