#include "socket.hpp"
#include <netinet/in.h>
#include <netinet/tcp.h>
#include <sys/socket.h>
#include <sys/time.h>
#include <sys/types.h>
#include <fcntl.h>
#include <unistd.h>
#include <cerrno>
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include "mmo.hpp"
#include "utils.hpp"
fd_set readfds;
int fd_max;
int currentuse;
const uint32_t RFIFO_SIZE = 65536;
const uint32_t WFIFO_SIZE = 65536;
struct socket_data *session[FD_SETSIZE];
/// Discard all input
static
void null_parse(int fd);
/// Default parser for new connections
static
void(*default_func_parse)(int) = null_parse;
void set_defaultparse(void(*defaultparse)(int))
{
default_func_parse = defaultparse;
}
/// Read from socket to the queue
static
void recv_to_fifo(int fd)
{
if (session[fd]->eof)
return;
ssize_t len = read(fd, session[fd]->rdata + session[fd]->rdata_size,
RFIFOSPACE(fd));
if (len > 0)
{
session[fd]->rdata_size += len;
session[fd]->connected = 1;
}
else
{
session[fd]->eof = 1;
}
}
static
void send_from_fifo(int fd)
{
if (session[fd]->eof)
return;
ssize_t len = write(fd, session[fd]->wdata, session[fd]->wdata_size);
if (len > 0)
{
session[fd]->wdata_size -= len;
if (len < (ssize_t)session[fd]->wdata_size)
{
memmove(session[fd]->wdata, session[fd]->wdata + len,
session[fd]->wdata_size);
}
session[fd]->connected = 1;
}
else
{
session[fd]->eof = 1;
}
}
static
void null_parse(int fd)
{
printf("null_parse : %d\n", fd);
RFIFOSKIP(fd, RFIFOREST(fd));
}
static
void connect_client(int listen_fd)
{
struct sockaddr_in client_address;
socklen_t len = sizeof(client_address);
int fd = accept(listen_fd, (struct sockaddr *) &client_address, &len);
if (fd == -1)
{
perror("accept");
return;
}
if (fd_max <= fd)
{
fd_max = fd + 1;
}
if (!free_fds())
{
fprintf(stderr, "softlimit reached, disconnecting : %d\n", fd);
delete_session(fd);
return;
}
const int yes = 1;
/// Allow to bind() again after the server restarts.
// Since the socket is still in the TIME_WAIT, there's a possibility
// that formerly lost packets might be delivered and confuse the server.
setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes);
/// Send packets as soon as possible
/// even if the kernel thinks there is too little for it to be worth it!
// I'm not convinced this is a good idea; although in minimizes the
// latency for an individual write, it increases traffic in general.
setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &yes, sizeof yes);
FD_SET(fd, &readfds);
fcntl(fd, F_SETFL, O_NONBLOCK);
CREATE(session[fd], struct socket_data, 1);
CREATE(session[fd]->rdata, uint8_t, RFIFO_SIZE);
CREATE(session[fd]->wdata, uint8_t, WFIFO_SIZE);
session[fd]->max_rdata = RFIFO_SIZE;
session[fd]->max_wdata = WFIFO_SIZE;
session[fd]->func_recv = recv_to_fifo;
session[fd]->func_send = send_from_fifo;
session[fd]->func_parse = default_func_parse;
session[fd]->client_addr = client_address;
session[fd]->created = time(NULL);
session[fd]->connected = 0;
currentuse++;
}
int make_listen_port(uint16_t port)
{
struct sockaddr_in server_address;
int fd = socket(AF_INET, SOCK_STREAM, 0);
if (fd == -1)
{
perror("socket");
return -1;
}
if (fd_max <= fd)
fd_max = fd + 1;
fcntl(fd, F_SETFL, O_NONBLOCK);
const int yes = 1;
/// Allow to bind() again after the server restarts.
// Since the socket is still in the TIME_WAIT, there's a possibility
// that formerly lost packets might be delivered and confuse the server.
setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes);
/// Send packets as soon as possible
/// even if the kernel thinks there is too little for it to be worth it!
// I'm not convinced this is a good idea; although in minimizes the
// latency for an individual write, it increases traffic in general.
setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &yes, sizeof yes);
server_address.sin_family = AF_INET;
server_address.sin_addr.s_addr = htonl(INADDR_ANY);
server_address.sin_port = htons(port);
if (bind(fd, (struct sockaddr *) &server_address,
sizeof(server_address)) == -1)
{
perror("bind");
exit(1);
}
if (listen(fd, 5) == -1)
{ /* error */
perror("listen");
exit(1);
}
FD_SET(fd, &readfds);
CREATE(session[fd], struct socket_data, 1);
session[fd]->func_recv = connect_client;
session[fd]->created = time(NULL);
session[fd]->connected = 1;
currentuse++;
return fd;
}
int make_connection(uint32_t ip, uint16_t port)
{
struct sockaddr_in server_address;
int fd = socket(AF_INET, SOCK_STREAM, 0);
if (fd == -1)
{
perror("socket");
return -1;
}
if (fd_max <= fd)
fd_max = fd + 1;
const int yes = 1;
/// Allow to bind() again after the server restarts.
// Since the socket is still in the TIME_WAIT, there's a possibility
// that formerly lost packets might be delivered and confuse the server.
setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes);
/// Send packets as soon as possible
/// even if the kernel thinks there is too little for it to be worth it!
// I'm not convinced this is a good idea; although in minimizes the
// latency for an individual write, it increases traffic in general.
setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &yes, sizeof yes);
server_address.sin_family = AF_INET;
server_address.sin_addr.s_addr = ip;
server_address.sin_port = htons(port);
fcntl(fd, F_SETFL, O_NONBLOCK);
/// Errors not caught - we must not block
/// Let the main select() loop detect when we know the state
connect(fd, (struct sockaddr *) &server_address,
sizeof(struct sockaddr_in));
FD_SET(fd, &readfds);
CREATE(session[fd], struct socket_data, 1);
CREATE(session[fd]->rdata, uint8_t, RFIFO_SIZE);
CREATE(session[fd]->wdata, uint8_t, WFIFO_SIZE);
session[fd]->max_rdata = RFIFO_SIZE;
session[fd]->max_wdata = WFIFO_SIZE;
session[fd]->func_recv = recv_to_fifo;
session[fd]->func_send = send_from_fifo;
session[fd]->func_parse = default_func_parse;
session[fd]->created = time(NULL);
session[fd]->connected = 1;
currentuse++;
return fd;
}
void delete_session(int fd)
{
if (fd < 0 || fd >= FD_SETSIZE)
return;
// If this was the highest fd, decrease it
// We could add a loop to decrement fd_max further for every null session,
// but this is cheap and good enough for the typical case
if (fd == fd_max - 1)
fd_max--;
FD_CLR(fd, &readfds);
if (session[fd])
{
free(session[fd]->rdata);
free(session[fd]->wdata);
free(session[fd]->session_data);
free(session[fd]);
}
session[fd] = NULL;
// just close() would try to keep sending buffers
shutdown(fd, SHUT_RDWR);
close(fd);
currentuse--;
if (currentuse < 0)
{
fprintf(stderr, "delete_session: current sessions negative!\n");
currentuse = 0;
}
return;
}
void realloc_fifo(int fd, size_t rfifo_size, size_t wfifo_size)
{
struct socket_data *s = session[fd];
if (s->max_rdata != rfifo_size && s->rdata_size < rfifo_size)
{
RECREATE(s->rdata, uint8_t, rfifo_size);
s->max_rdata = rfifo_size;
}
if (s->max_wdata != wfifo_size && s->wdata_size < wfifo_size)
{
RECREATE(s->wdata, uint8_t, wfifo_size);
s->max_wdata = wfifo_size;
}
}
void WFIFOSET(int fd, size_t len)
{
struct socket_data *s = session[fd];
if (s->wdata_size + len + 16384 > s->max_wdata)
{
realloc_fifo(fd, s->max_rdata, s->max_wdata << 1);
printf("socket: %d wdata expanded to %d bytes.\n", fd, s->max_wdata);
}
if (s->wdata_size + len + 2048 < s->max_wdata)
s->wdata_size += len;
else
fprintf(stderr, "socket: %d wdata lost !!\n", fd), abort();
}
void do_sendrecv(uint32_t next)
{
fd_set rfd = readfds, wfd;
FD_ZERO(&wfd);
for (int i = 0; i < fd_max; i++)
{
if (session[i] && session[i]->wdata_size)
FD_SET(i, &wfd);
}
struct timeval timeout;
timeout.tv_sec = next / 1000;
timeout.tv_usec = next % 1000 * 1000;
if (select(fd_max, &rfd, &wfd, NULL, &timeout) <= 0)
return;
for (int i = 0; i < fd_max; i++)
{
if (!session[i])
continue;
if (FD_ISSET(i, &wfd))
{
if (session[i]->func_send)
//send_from_fifo(i);
session[i]->func_send(i);
}
if (FD_ISSET(i, &rfd))
{
if (session[i]->func_recv)
//recv_to_fifo(i);
//or connect_client(i);
session[i]->func_recv(i);
}
}
}
void do_parsepacket(void)
{
for (int i = 0; i < fd_max; i++)
{
if (!session[i])
continue;
if (!session[i]->connected
&& time(NULL) - session[i]->created > CONNECT_TIMEOUT)
{
printf("Session #%d timed out\n", i);
session[i]->eof = 1;
}
if (!session[i]->rdata_size && !session[i]->eof)
continue;
if (session[i]->func_parse)
{
session[i]->func_parse(i);
/// some func_parse may call delete_session
if (!session[i])
continue;
}
/// Reclaim buffer space for what was read
RFIFOFLUSH(i);
}
}
void do_socket(void)
{
FD_ZERO(&readfds);
currentuse = 3;
}
void RFIFOSKIP(int fd, size_t len)
{
struct socket_data *s = session[fd];
s->rdata_pos += len;
if (s->rdata_size < s->rdata_pos)
{
fprintf(stderr, "too many skip\n");
abort();
}
}
void fclose_(FILE * fp)
{
if (fclose(fp))
perror("fclose"), abort();
currentuse--;
}
FILE *fopen_(const char *path, const char *mode)
{
FILE *f = fopen(path, mode);
if (f)
currentuse++;
return f;
}
bool free_fds(void)
{
return currentuse < SOFT_LIMIT;
}