SSLServer + set_mount_point() timing out and not sending all data

[jmscreation]

I'm going to try to keep this as concise as possible.

What's happening is I have an SSLServer that's running, but when I setup the mount point to get files automatically, it just won't send the entire html file over. It sends almost all of it, then just stops and eventually times out. It's almost like it's trying to send the entire file in one TCP packet. That just can't happen, yet it's trying to or something. I can't determine what's causing this to happen.

I didn't want to open this as an issue, but after re-programming a tiny class to reproduce my issue, I determined it wasn't in my code.

Below is my minimal code that reproduces the error. I can connect on my PC which works fine on https://localhost, however it barely works when I connect over the WiFi on my Mobile device, or even other computers in the network. Sometimes (once every 15 - 20 tried) it might load the whole html page. The index.html is 578KB, and includes 39.6KB of javascript files. The problem is definitely with the large file size not being able to be sent by the mount point system somehow. I really need this to work, as this is an important project I'm working on. TIA

test.h

#pragma once

#include <iostream>

#define CPPHTTPLIB_OPENSSL_SUPPORT
#include "httplib.h"

class TestServer {
public:
    httplib::SSLServer server;
    TestServer();
};

test.cpp

#include "test.h"

/*
    Test Server Class Implementation
*/


TestServer::TestServer(): server("certs\\server.pem", "certs\\server.key") {
    server.set_pre_routing_handler([&](const httplib::Request &req, httplib::Response &res) -> httplib::SSLServer::HandlerResponse {
        std::cout << "from -> " << req.remote_addr << ":" << req.remote_port << "\n";

        return httplib::SSLServer::HandlerResponse::Unhandled;
    });

    server.set_mount_point("/", ".\\html");
    
    server.listen("0.0.0.0", 443);
}

EDIT:

I've been doing a lot of testing and have also uncovered the following:

The return value of SSL_write(...) is -1 when using the debugger

This results in an SSL_ERROR_SYSCALL error from SSL_get_error(rval);

[jmscreation]

Okay so I've made changes to your library to get it to work most of the time.

Here's what I did:

I had to enable partial write mode, and disable auto retry mode. This might actually break the SSL client reading, I don't know. Please look over this.

Line

// SSL socket stream implementation
SSLSocketStream::SSLSocketStream(socket_t sock, SSL *ssl,
                                        time_t read_timeout_sec,
                                        time_t read_timeout_usec,
                                        time_t write_timeout_sec,
                                        time_t write_timeout_usec)
    : sock_(sock), ssl_(ssl), read_timeout_sec_(read_timeout_sec),
      read_timeout_usec_(read_timeout_usec),
      write_timeout_sec_(write_timeout_sec),
      write_timeout_usec_(write_timeout_usec) {
  //SSL_clear_mode(ssl, SSL_MODE_AUTO_RETRY); // removed
  SSL_set_mode(ssl, SSL_MODE_ENABLE_PARTIAL_WRITE); // added partial write capability
}

I completely rewrote the SSLSocketStream::write function to automatically keep writing data to the buffer until it reaches the end of the buffer, or an error occurs.

Line

ssize_t SSLSocketStream::write(const char *ptr, size_t size) {
  ssize_t bytes = -1;
  const char* end = ptr + size; // fixed end point
  if (is_writable()) {
    do {
      bytes = SSL_write(ssl_, ptr, static_cast<int>(size));
      if(bytes <= 0){ return -1; } // SSL error occurred
      ptr += bytes; // delta - move buffer
      size = size > bytes ? size - bytes : 0; // delta - resize buffer
    } while(ptr < end);
  }
  return bytes;
}

A minor change here because I thought it could be problematic. Doing !strm.write() will only cause an error response if the value is 0. This doesn't actually return an error response if the value is negative, which in fact was -1 because SSL_write() was constantly returning -1 (Which is an SSL_ERROR_SYSCALL error as mentioned above)

Line

// Body
  auto ret = true;
  if (req.method != "HEAD") {
    if (!res.body.empty()) {
      if (strm.write(res.body) <= 0) { ret = false; } // using <= 0 because !strm.write() does not catch negative values
    } else if (res.content_provider_) {
      if (write_content_with_provider(strm, req, res, boundary, content_type)) {
        res.content_provider_success_ = true;
      } else {
        res.content_provider_success_ = false;
        ret = false;
      }
    }
  }

This is what I've done so far. It is mostly working right now. It still returns -1 seemingly at random. I'm still really unsure why this is happening.

Edit:
After more testing, it's getting worse the more I test it. Seems like it's slowly turning back to square one. It will load sometimes, but most of the time just transfers some of the data, then SSL_write returns -1 and stops.

I will continue to update this thread if I run into any other issues or problems. Thanks for contributing this library to the opensource community, and keep up the great work.

[yhirose]

@jmscreation, could you try httplib.h at f76a798 to see if it fixes the problem on your machine?

[jmscreation]

I think there's an issue with this line:

#ifdef _WIN32
      while (--n >= 0 &&
             (err == SSL_ERROR_WANT_WRITE
              err == SSL_ERROR_SYSCALL && WSAGetLastError() == WSAETIMEDOUT)) {
#else

I've changed it to just so I could get it to compile:

while (--n >= 0 &&
             (err == SSL_ERROR_WANT_WRITE || err == SSL_ERROR_SYSCALL) && WSAGetLastError() == WSAETIMEDOUT) {

The same issue is happening. It works sometimes, but randomly it will crash and the page is blank. It doesn't receive 100% of the data.

[yhirose]

That's my mistake... It should have been as below. (Your change on the above isn't correct either...)
The code uses the same retry logic as SSLSocketStream::read uses.

#ifdef _WIN32
      while (--n >= 0 &&
             (err == SSL_ERROR_WANT_WRITE ||
              err == SSL_ERROR_SYSCALL && WSAGetLastError() == WSAETIMEDOUT)) {
#else

[jmscreation]

I've tried the new code as well, but same thing. It's all stemming from SSL_write() returning -1 for seemingly no reason what-so-ever.

Are you able to re-produce the same issue on your machine? Is it possible I have a broken version of OpenSSL?

I'm using Windows 10 Pro 64bit, with a 64bit static library of openssl
Compiler is MinGW_64 GCC 11.2.0

[yhirose]

I use the latest OpenSSL available on vcpkg with Visual C++, and it always works well on my machine. At this point, I don't have any clue, and I can't verify if it's an OpenSSL issue or an issue on your network environment. Sorry about that...

[jmscreation]

For further information, it's not my network environment. I've tested the basic HTTP server, and it works without fail.

I'm going to call it a night. I will continue researching in trying to identify this problem.
Thanks for your help.

[jmscreation]

Good news and bad news (bad news for me mostly)

I did a little more testing, and found out that it has something to do with my computer I'm running it on. It's really strange. There's some kind of hardware or driver issue.

I'll reopen this issue if I find out it starts happening with other systems. But for now, it's quite a frustrating problem that's related to my own hardware.

Thank you for taking the time to help me with this. Also, I do like the updated version you've added into that code as it does, in fact make it more stable! It also reflects with how the stream read function is designed. I would recommend making that an actual update to the next release.

Thanks again!

[yhirose]

Sure. I have just merged it. Hope your project goes well.

[jmscreation]

I've been working with my project a little bit and noticed the same issue is happening on lots of devices. So it seems the problem isn't just my hardware.

I'll continue to investigate the issue, but I think the problem lies in my current build of OpenSSL, and not this library. It's just a bit of a difficult issue I'm facing.

Edit:

I've gotten the latest version of OpenSSL and I'm still having the same problem.
WSAECONNABORTED seems to intermittently occur (8 out of 10 times). This is not an issue with my network or network hardware. Something strange is happening and I really want to get to the bottom of this.