NTW/hw1/server.cc

// vim: set ts=2 sw=2 et tw=80:
/*
 * Insidious Proxy - Claudio Maggioni
 *
 * Sources and references used to write this program:
 * - https://git.maggioni.xyz/maggicl/UniversalPizzoccheri (own work from
 *   high-school), used for code snippets on sockets
 * - Cppreference.com and man 3, for documentation (as usual)
 * - http://acme.com/software/micro_proxy/, for general inspiration for the high
 *   level implementation and code snippets for DNS queries.
 *
 * Dependemcies:
 * - http://www.graphicsmagick.org/project.html for imagemagick bindings in
 *   order to rotate the image. On macos, install with 'brew install
 *   graphicsmagick'.
 *
 * Building:
 * On MacOS, run 'make macos'. Otherwise use simply 'make'.
 */

#include <cctype>
#include <arpa/inet.h>
#include <Magick++.h>
#include <sys/types.h>
#include <cstdint>
#include <sys/socket.h>
#include <cstring>
#include <cstdio>
#include <cstdlib>
#include <unistd.h>
#include <pthread.h>
#include <iostream>
#include <map>
#include <vector>
#include <netdb.h>

using namespace std;

#define MAX_CLIENTS 128
#define BUFFER 8192

void* request_thread(void*);

// Do not kill process when socket connection is broken. Error is already
// handled
static void broken_pipe(int signo) {}

int sock_conn_d;

// Close socket when shutting down
static void sigint(int signo) {
  close(sock_conn_d);
  exit(0);
}

enum orientation {
  LEFT = 0, FLIP = 1, RIGHT = 2
};

enum orientation alteration;

bool parse_alteration(const char*, enum orientation*);

int main(int argc, char** argv) {
  srand(time(NULL));

  signal(SIGPIPE, broken_pipe);
  Magick::InitializeMagick(*argv);

  if(argc < 3) {
    cerr << "Give port as first arg and image_rotation as second arg" << endl;
    exit(255);
  }

  if (!parse_alteration(argv[2], &alteration)) {
    cerr << "image_rotation can only be 'flip', 'clockwise', 'counterclockwise'"
      "or 'random'" << endl;
    exit(255);
  }

  int sock_data_d;
  struct sockaddr_in server;

  if((sock_conn_d = socket(AF_INET, SOCK_STREAM, 0)) == -1) {
    perror("server socket not created");
    exit(1);
  }

  memset(&server, 0, sizeof(server)); // zero the memory

  server.sin_family = AF_INET;
  server.sin_addr.s_addr = inet_addr("0.0.0.0");
  server.sin_port = htons(atoi(argv[1]));

  int reusePort = 1;
  setsockopt(sock_conn_d, SOL_SOCKET, SO_REUSEPORT, &reusePort,
      sizeof(reusePort));

  if(::bind(sock_conn_d, (struct sockaddr*)&server, sizeof(server)) == -1) {
    perror("server socket not bound");
    exit(2);
  }

  while(true) {
    if(listen(sock_conn_d, MAX_CLIENTS) != 0) {
      perror("cannot listen on socket");
      close(sock_conn_d);
      exit(3);
    }

    printf("listening...\n");

    struct sockaddr_in client;
    socklen_t client_len = sizeof(client);
    sock_data_d = accept(sock_conn_d, (struct sockaddr*)&client, &client_len);

    if(sock_data_d == -1) {
      perror("cannot accept client");
      close(sock_conn_d);
      exit(4);
    }

    int* fd = (int*) malloc(sizeof(int));
    if (!fd) {
      perror("thread creation failed");
      exit(1);
    }

    *fd = sock_data_d;

    pthread_t req;
    pthread_create(&req, NULL, request_thread, (void*) fd);
  }
}

/*
 * Parses argv[2] in order to set alteration using the constraints given in the
 * assignment. 'random' sets a random alteration fixed for the entire execution
 * of the program. Returns true when argv[2] is valid, false if not
 */
bool parse_alteration(const char* arg, enum orientation* p) {
  if (!strcmp(arg, "counterclockwise")) {
    *p = LEFT;
  } else if (!strcmp(arg, "flip")) {
    *p = FLIP;
  } else if (!strcmp(arg, "clockwise")) {
    *p = RIGHT;
  } else if (!strcmp(arg, "random")) {
    *p = (enum orientation) (rand() % 3);
  } else {
    return false;
  }

  return true;
}

/**
 * Returns true if from the given file the first two chars that are read from
 * the current seeking point are '\r' and '\n'. Removes those two chars from the
 * stream if present, otherwise no character will apprar read
 */
bool empty_line(FILE* in) {
  char a = fgetc(in), b = fgetc(in);
  if (a == '\r' && b == '\n') {
    return true;
  } else {
    ungetc(b, in);
    ungetc(a, in);
    return false;
  }
}

/**
 * Reads from the given file HTTP header formats, and writes them to the given
 * std::map. Halts when the header section is terminated (first occurrence of
 * '\r\n\r\n'). Returns false at the first malformed headers found, true if no
 * are found
 */
bool parse_headers(FILE* in, map<string, string>& headers) {
  while(true) {
    // probe first two chars to see if body will start
    if (empty_line(in)) {
      break;
    }

    string header_name;
    char c;

    while((c = fgetc(in)) != ':') {
      if (c == EOF || c == '\r' || c == '\n' || !isascii(c)) {
        return false;
      } else {
        header_name += c;
      }
    }

    do {
      c = fgetc(in);
    } while(isspace(c) && c != '\n' && c != '\r');

    ungetc(c, in);

    string value = "";
    do {
      c = fgetc(in);
      value += c;
    } while(c != '\r');
    value.pop_back();
    fgetc(in); // \n

    headers[header_name] = value;
  }

  return true;
}

/**
 * Returns true if it is customary for the given HTTP method to have a body
 */
bool has_body(const char* method) {
  return !strcmp(method, "POST") || !strcmp(method, "PUT");
}

/**
 * Returns true if the given string contains a valid HTTP 1.1 chunk start
 * delimiter, false otherwise.
 */
bool is_chunk_start(const char* buf) {
  const char* i;
  for (i = buf; isdigit(*i); i++);
  if (*i != '\r') return false;
  i++;
  if (*i != '\n') return false;
  i++;
  return *i == '\0';
}

/**
 * Returns true if the given buffer contains a valid HTTP 1.1 chunk end
 * delimiter at its very end, false otherwise.
 */
bool is_chunk_end(vector<uint8_t>& body) {
  const char b = body.back();
  body.pop_back();

  const char a = body.back();
  body.push_back(b);

  return a == '\r' && b == '\n';
}

/**
 * Fetch and parse HTTP body from given file, assuming the header has already
 * been parsed. Parses bodies either delimited by Content-Length or endoded
 * using HTTP 1.1 chunds. If the body appears to not meet this criteria (by
 * inspecting the HTTP headers), false is returned, otherwise returns true.
 */
bool fetch_body(FILE* in, vector<uint8_t>& body,
    const map<string, string> headers, bool strip_chunked) {
  bool chunked;
  size_t length;

  {
    auto p = headers.find("Content-Length");
    chunked = p == headers.end();

    if (!chunked) {
      length = atol(p->second.c_str());
    } else {
      auto q = headers.find("Transfer-Encoding");
      if (q == headers.end() || q->second != "chunked") {
        return false;
      }
    }
  }

  if (!chunked) {
    // read Content-Length bytes
    uint8_t buf[BUFFER];
    size_t r;

    for(size_t w = 0; w < length; w += r) {
      r = fread(buf, 1, (length - w) > BUFFER ? BUFFER : (length - w), in);
      if (r == -1) return false;
      body.insert(end(body), buf, buf + r);
    }
  } else {
    // This was implemented before Prof. Carzaniga said chunked encoding is not
    // required. I am leaving this just because it is already done.
    // Read chunks and search for final chunk 0\r\n\r\n
    char buf[BUFFER + 1];
    buf[BUFFER] = '\0';
    bool will_chunk_start = true;

    while (true) {
      size_t len;
      for (len = 0; len < BUFFER; len++) {
        if ((buf[len] = fgetc(in)) == '\n') {
          buf[len + 1] = '\0';
          break;
        }
      }

      if (!strcmp(buf, "0\r\n") && will_chunk_start) {
        break;
      }

      if (will_chunk_start && is_chunk_start(buf) && strip_chunked) {
        continue;
      }

      body.insert(end(body), begin(buf), end(buf));
      body.pop_back();

      will_chunk_start = is_chunk_end(body);
    }

    char a = fgetc(in), b = fgetc(in);

    if (!strip_chunked) {
      body.insert(end(body), begin(buf), end(buf));
      body.push_back(a); // \r
      body.push_back(b); // \n
    }
  }

  return true;
}

/**
 * Sends an HTTP response with status code and message given. Then proceeds to
 * close the given input and output files and terminates the current thread.
 */
void send_error(FILE* in, FILE* out, const char* protocol, const int status,
    const string message) {
  const char* msg = message.c_str();

  fprintf(out, "%s %d %s\r\n", protocol, status, msg);
  fprintf(out, "Content-Type: text/html; charset=utf-8\r\n");
  fprintf(out, "Connection: close\r\n");

  cerr << status << ' ' << msg << endl;

  char message_buf[4096];
  message_buf[4095] = '\0';
  snprintf(message_buf, 4095, "<!DOCTYPE html>\n"
      "<html>\n"
      "  <head>\n"
      "    <meta charset=\"utf-8\"/>\n"
      "    <title>%s</title>\n"
      "  </head>\n"
      "  <body>\n"
      "    <h1>%s</h1>\n"
      "  </body>\n"
      "</html>\n", msg, msg);

  fprintf(out, "Content-Length: %lu\r\n\r\n", strlen(message_buf));
  fprintf(out, "%s", message_buf);

  fclose(in);
  fflush(out);
  fclose(out);
  pthread_exit(NULL);
}

/**
 * Converts url in a zero-terminated string for the host name, copied in a
 * pre-allocated char array given as copy_in. Port is returned
 * as integer, -1 on error
 */
int find_host_port(const char* url, char* copy_in) {
  const char* i = url;
  size_t c = 0;

  while (*i != ':' && *i != '\0' && *i != '/') {
    if (c >= 256 || (!isalnum(*i) && *i != '-' && *i != '.')) return -1;
    *copy_in = *i;
    copy_in++;
    i++;
    c++;
  }

  if (c == 0) return -1;

  *copy_in = '\0';
  i++;
  if (*i == '\0' || *i == '/') return 80;

  unsigned p;
  if (sscanf(i, "%5u", &p) == 1) {
    return p < 65536 ? p : 80;
  } else {
    return 80;
  }
}

struct forward {
  int in;
  int out;
};

bool total_write(uint8_t* data, size_t n, int out) {
  for(size_t w = 0; w < n;) {
    ssize_t written = write(out, data + w, n - w);
    if (written == -1) {
      return false;
    }
    w += written;
  }
  return true;
}

void* forwarder_thread(void* data) {
  struct forward* f = (struct forward*) data;
  const size_t BUF_SIZE = 65536;

  uint8_t buffer[BUF_SIZE];
  while (true) {
    size_t r = read(f->in, buffer, BUF_SIZE);

    if (r == 0) {
      break;
    }

    if (!total_write(buffer, r, f->out)) {
#if DEBUG
      cerr << "Closing CONNECT forwarder" << endl;
#endif
      return NULL;
    }
  }

  return NULL;
}

/**
 * Returns a socket file descriptof of a newly opened socket to the given host.
 */
int open_client_socket(FILE* in, FILE* out, const char* protocol,
    const char* host) {
  int port;
  char hostname[257];

  if ((port = find_host_port(host, hostname)) == -1) {
    send_error(in, out, protocol, 500, "Hostname parse error");
  }

  int socketfd = socket(AF_INET, SOCK_STREAM, 0);
  if (socketfd == -1) {
    send_error(in, out, protocol, 500, "TCP socket connection error");
  }

  struct hostent *he = gethostbyname(hostname);
  if (!he) {
    send_error(in, out, protocol, 404, "Unknown host");
  }

  struct sockaddr_in locale;
  memset (&locale, 0, sizeof(struct sockaddr_in));

  locale.sin_family =	AF_INET;
  locale.sin_addr.s_addr = htonl(INADDR_ANY);
  locale.sin_port =	htons(0);

  if (::bind(socketfd, (struct sockaddr*) &locale, sizeof(locale)) == -1) {
    send_error(in, out, protocol, 500, "TCP socket connection error");
  }

  struct sockaddr_in remote;
  memset (&remote, 0, sizeof(struct sockaddr_in));

  remote.sin_family = he->h_addrtype;
  memmove(&remote.sin_addr, he->h_addr, he->h_length);
  remote.sin_port = htons((uint16_t) port);

  if (connect(socketfd, (struct sockaddr*) &remote, sizeof(remote)) == -1) {
    send_error(in, out, protocol, 500, "TCP socket connection error");
  }

  return socketfd;
}

/**
 * Handle HTTP connect method. Opens a client socket to the host in *url (by
 * doing a dns query first). Then forks in another threads and starts relaying
 * data both from the HTTP client socket to the newly opened socket and
 * videversa.
 */
void handle_connect(FILE* in, FILE* out, const char* protocol, char* url,
    const int fd) {

  int socketfd = open_client_socket(in, out, protocol, url);

  struct forward from = { .in = fd, .out = socketfd },
                 to = { .in = socketfd, .out = fd };

  fprintf(out, "%s %d %s\r\n\r\n", protocol, 200, "Connection established");
  fflush(out);

#if DEBUG
  cerr << "CONNECT: Connection established" << endl;
#endif

  pthread_t ffrom;
  pthread_create(&ffrom, NULL, forwarder_thread, &from);

  forwarder_thread(&to);

  pthread_join(ffrom, NULL);
}

void* request_thread(void* data) {
  int fd = *((int*) data);
  FILE* in = fdopen(dup(*((int*) data)), "r");
  FILE* out = fdopen(dup(*((int*) data)), "w");
  free(data);

  while (true) {

    char method[10];
    char url[8000];
    char protocol[10];
    map<string, string> headers;
    vector<uint8_t> body;

    int i;
    if ((i = fscanf(in, "%10s %8000s %10s\r\n", method, url, protocol)) < 3) {
      if (i < 0) break;
      send_error(in, out, protocol, 400, "Bad request line");
    }

#if DEBUG
    cerr << "METHOD: " << method << " URL: " << url << " PROTOCOL: " << protocol
      << endl;
#endif

    if (!parse_headers(in, headers)) {
      send_error(in, out, protocol, 400, "Malformed header");
    }

    if (strncmp(url, "http", 4) && strcmp(method, "CONNECT")) {
      send_error(in, out, protocol, 400, "Request line must contain absolute"
          " URL (this is a proxy)");
    }

    bool has_body = fetch_body(in, body, headers, false);

#if DEBUG
    cerr << "has_body: "  << has_body << endl;
    cerr << "Request body parsed" << endl;
#endif

    if (!strcmp(method, "CONNECT")) {
      handle_connect(in, out, protocol, url, fd);
      break;
    } else {
      // Delete Proxy-Connection header
      {
        auto i = headers.find("Proxy-Connection");
        if (i != headers.end()) headers.erase(i);
      }

      if (strncmp(url, "http://", 7)) {
        send_error(in, out, protocol, 400, "Protocol in URL not supported");
      }

      char* host = url + 7;

      int serverfd = open_client_socket(in, out, protocol, host);
      FILE* s_in = fdopen(dup(serverfd), "r");
      FILE* s_out = fdopen(dup(serverfd), "w");

      fprintf(s_out, "%s %s %s\r\n", method, url, protocol);
      for (auto i = headers.begin(); i != headers.end(); i++) {
        fprintf(s_out, "%s: %s\r\n", i->first.c_str(), i->second.c_str());
      }
      fprintf(s_out, "\r\n");

      if (has_body) {
        for (auto i = body.begin(); i < body.end(); i++) {
          fputc(*i, s_out);
        }
      }

      fflush(s_out);

      char protocol_r[10];
      char message[8000];
      unsigned code;

      const char* image = NULL;
      const char* const PNG = "PNG";
      const char* const JPEG = "JPEG";

      fscanf(s_in, "%10s %3u ", protocol_r, &code);
      fgets(message, 8000, s_in);
      message[strlen(message) - 2] = '\0';

#if DEBUG
      cerr << "Response: STATUS: " << code << " MESSAGE: " << message << endl;
#endif

      headers.clear();
      parse_headers(s_in, headers);

      {
        auto a = headers.find("Content-Type");
        if (a != headers.end()) {
          if (a->second == "image/jpeg") image = JPEG;
          else if (a->second == "image/png") image = PNG;
        }
      }

      body.clear();
      has_body = fetch_body(s_in, body, headers, true);

      fclose(s_in);
      fclose(s_out);
      close(serverfd);

      Magick::Blob output;
      if (has_body && image) {
        try {
          Magick::Blob my_blob(&body[0], body.size());

          Magick::Image to_rotate;

          try {
            to_rotate = Magick::Image(my_blob);
          } catch(Magick::Warning& ignored) {}

          try {
            to_rotate.magick(image);
          } catch(Magick::Warning& ignored) {}

          try {
            switch (alteration) {
              case FLIP: to_rotate.flip(); break;
              case LEFT: to_rotate.rotate(-90); break;
              case RIGHT: to_rotate.rotate(90); break;
            }
          } catch(Magick::Warning& ignored) {}

          try {
            to_rotate.write(&output);
          } catch(Magick::Warning& ignored) {}

          fprintf(out, "%s %u %s\n", protocol_r, code, message);

          headers["Content-Length"] = to_string(output.length());
        } catch (Magick::Exception &error) {
          cout << "Magick++ image conversion failed: " << error.what()
            << endl;
          send_error(in, out, protocol, 500, "Image conversion failed");
        }
      } else if (has_body) {
        headers["Content-Length"] = to_string(body.size());
      } else {
        headers["Content-Length"] = "0";
      }

      fprintf(out, "%s %u %s\n", protocol_r, code, message);

      headers["Connection"] = "keep-alive";
      for (auto i = headers.begin(); i != headers.end(); i++) {
        string header = i->first + ": " + i->second + "\r\n";
        fprintf(out, "%s", header.c_str());
      }
      fprintf(out, "\r\n");
      fflush(out);

      if (has_body && image) {
        total_write((uint8_t*) output.data(), output.length(), fd);
      } else if (has_body) {
        total_write(&body[0], body.size(), fd);
      }

      fflush(out);
    }
  }

  cout << "closing data socket...\n";

  fclose(in);
  fflush(out);
  fclose(out);
  close(fd);

  pthread_exit(NULL);
}