incentive/psopeeps-newserv
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
6ae08e9b059b249509591862c3066028b36a4619
psopeeps-newserv/src/IPStackSimulator.cc
T

1539 lines
55 KiB
C++
Raw Blame History

#include "IPStackSimulator.hh"
#include <arpa/inet.h>
#include <event2/buffer.h>
#include <event2/bufferevent.h>
#include <event2/listener.h>
#include <netinet/in.h>
#include <stdint.h>
#include <string.h>
#include <phosg/Network.hh>
#include <phosg/Random.hh>
#include <phosg/Time.hh>
#include <string>
#include "DNSServer.hh"
#include "IPFrameInfo.hh"
#include "Loggers.hh"
using namespace std;
static const size_t DEFAULT_RESEND_PUSH_USECS = 200000; // 200ms
static string unescape_hdlc_frame(const void* data, size_t size) {
StringReader r(data, size);
if (r.get_u8(data) != 0x7E) {
throw runtime_error("HDLC frame does not begin with 7E");
}
string ret("\x7E", 1);
while (r.get_u8(false) != 0x7E) {
uint8_t ch = r.get_u8();
if (ch == 0x7D) {
ch = r.get_u8();
if (ch == 0x7E) {
throw runtime_error("abort sequence received");
}
ret.push_back(ch ^ 0x20);
} else {
ret.push_back(ch);
}
}
ret.push_back(0x7E);
return ret;
}
static string unescape_hdlc_frame(const string& data) {
return unescape_hdlc_frame(data.data(), data.size());
}
static string escape_hdlc_frame(const void* data, size_t size, uint32_t escape_control_character_flags = 0xFFFFFFFF) {
if (size < 2) {
throw runtime_error("HDLC frame too small for start and end sentinels");
}
StringReader r(data, size);
if (r.pget_u8(size - 1) != 0x7E) {
throw runtime_error("HDLC frame does not end with 7E");
}
r.truncate(size - 1);
if (r.get_u8() != 0x7E) {
throw runtime_error("HDLC frame does not begin with 7E");
}
string ret("\x7E", 1);
while (!r.eof()) {
uint8_t ch = r.get_u8();
if ((ch == 0x7D) || (ch == 0x7E) || ((ch < 0x20) && ((escape_control_character_flags >> ch) & 1))) {
ret.push_back(0x7D);
ret.push_back(ch ^ 0x20);
} else {
ret.push_back(ch);
}
}
ret.push_back(0x7E);
return ret;
}
static string escape_hdlc_frame(const string& data, uint32_t escape_control_character_flags = 0xFFFFFFFF) {
return escape_hdlc_frame(data.data(), data.size(), escape_control_character_flags);
}
// Note: these functions exist because seq nums are allowed to wrap around the
// 32-bit integer space by design. We have to do the subtraction before the
// comparison to allow integer overflow to occur if needed.
static inline bool seq_num_less(uint32_t a, uint32_t b) {
return (a - b) & 0x80000000;
}
static inline bool seq_num_less_or_equal(uint32_t a, uint32_t b) {
return (a == b) || seq_num_less(a, b);
}
static inline bool seq_num_greater(uint32_t a, uint32_t b) {
return (b - a) & 0x80000000;
}
static __attribute__((unused)) inline bool seq_num_greater_or_equal(uint32_t a, uint32_t b) {
return (a == b) || seq_num_greater(a, b);
}
string IPStackSimulator::str_for_ipv4_netloc(uint32_t addr, uint16_t port) {
be_uint32_t be_addr = addr;
char addr_str[INET_ADDRSTRLEN];
if (!inet_ntop(AF_INET, &be_addr, addr_str, INET_ADDRSTRLEN)) {
return string_printf("<UNKNOWN>:%hu", port);
} else {
return string_printf("%s:%hu", addr_str, port);
}
}
string IPStackSimulator::str_for_tcp_connection(shared_ptr<const IPClient> c, const IPClient::TCPConnection& conn) {
uint64_t key = IPStackSimulator::tcp_conn_key_for_connection(conn);
string server_netloc_str = str_for_ipv4_netloc(conn.server_addr, conn.server_port);
string client_netloc_str = str_for_ipv4_netloc(c->ipv4_addr, conn.client_port);
int fd = bufferevent_getfd(c->bev.get());
return string_printf("%d+%016" PRIX64 " (%s -> %s)",
fd, key, client_netloc_str.c_str(), server_netloc_str.c_str());
}
IPStackSimulator::IPStackSimulator(
shared_ptr<struct event_base> base,
shared_ptr<ServerState> state)
: base(base),
state(state),
pcap_text_log_file(state->ip_stack_debug ? fopen("IPStackSimulator-Log.txt", "wt") : nullptr) {
this->host_mac_address_bytes.clear(0x90);
this->broadcast_mac_address_bytes.clear(0xFF);
}
IPStackSimulator::~IPStackSimulator() {
if (this->pcap_text_log_file) {
fclose(this->pcap_text_log_file);
}
}
void IPStackSimulator::listen(const string& name, const string& socket_path, Protocol proto) {
int fd = ::listen(socket_path, 0, SOMAXCONN);
ip_stack_simulator_log.info("Listening on Unix socket %s on fd %d as %s", socket_path.c_str(), fd, name.c_str());
this->add_socket(name, fd, proto);
}
void IPStackSimulator::listen(const string& name, const string& addr, int port, Protocol proto) {
if (port == 0) {
this->listen(name, addr, proto);
} else {
int fd = ::listen(addr, port, SOMAXCONN);
string netloc_str = render_netloc(addr, port);
ip_stack_simulator_log.info("Listening on TCP interface %s on fd %d as %s", netloc_str.c_str(), fd, name.c_str());
this->add_socket(name, fd, proto);
}
}
void IPStackSimulator::listen(const string& name, int port, Protocol proto) {
this->listen(name, "", port, proto);
}
void IPStackSimulator::add_socket(const string& name, int fd, Protocol proto) {
unique_listener l(
evconnlistener_new(
this->base.get(),
IPStackSimulator::dispatch_on_listen_accept,
this,
LEV_OPT_REUSEABLE,
0,
fd),
evconnlistener_free);
this->listening_sockets.emplace(piecewise_construct, forward_as_tuple(fd), forward_as_tuple(name, proto, std::move(l)));
}
uint32_t IPStackSimulator::connect_address_for_remote_address(uint32_t remote_addr) {
// Use an address not on the same subnet as the client, so that PSO Plus and
// Episode III will think they're talking to a remote network and won't reject
// the connection.
if ((remote_addr & 0xFF000000) != 0x23000000) {
return 0x23232323;
} else {
return 0x24242424;
}
}
IPStackSimulator::IPClient::IPClient(shared_ptr<IPStackSimulator> sim, Protocol protocol, struct bufferevent* bev)
: sim(sim),
bev(bev, bufferevent_free),
protocol(protocol),
mac_addr(0),
ipv4_addr(0),
idle_timeout_event(event_new(sim->base.get(), -1, EV_TIMEOUT, &IPStackSimulator::IPClient::dispatch_on_idle_timeout, this), event_free) {
uint64_t idle_timeout_usecs = sim->state->client_idle_timeout_usecs;
struct timeval tv = usecs_to_timeval(idle_timeout_usecs);
event_add(this->idle_timeout_event.get(), &tv);
}
void IPStackSimulator::IPClient::dispatch_on_idle_timeout(evutil_socket_t, short, void* ctx) {
reinterpret_cast<IPStackSimulator::IPClient*>(ctx)->on_idle_timeout();
}
void IPStackSimulator::IPClient::on_idle_timeout() {
auto sim = this->sim.lock();
if (sim) {
ip_stack_simulator_log.info("Idle timeout expired on virtual network %d", bufferevent_getfd(this->bev.get()));
sim->disconnect_client(this->bev.get());
} else {
ip_stack_simulator_log.info("Idle timeout expired on virtual network %d, but simulator is missing", bufferevent_getfd(this->bev.get()));
}
}
static void flush_and_free_bufferevent(struct bufferevent* bev) {
bufferevent_flush(bev, EV_READ | EV_WRITE, BEV_FINISHED);
bufferevent_free(bev);
}
IPStackSimulator::IPClient::TCPConnection::TCPConnection()
: server_bev(nullptr, flush_and_free_bufferevent),
pending_data(evbuffer_new(), evbuffer_free),
resend_push_event(nullptr, event_free),
awaiting_first_ack(true),
server_addr(0),
server_port(0),
client_port(0),
next_client_seq(0),
acked_server_seq(0),
resend_push_usecs(DEFAULT_RESEND_PUSH_USECS),
next_push_max_frame_size(1024),
max_frame_size(1024),
bytes_received(0),
bytes_sent(0) {}
void IPStackSimulator::disconnect_client(struct bufferevent* bev) {
ip_stack_simulator_log.info("Virtual network %d disconnected", bufferevent_getfd(bev));
this->bev_to_client.erase(bev);
}
void IPStackSimulator::dispatch_on_listen_accept(
struct evconnlistener* listener, evutil_socket_t fd,
struct sockaddr* address, int socklen, void* ctx) {
reinterpret_cast<IPStackSimulator*>(ctx)->on_listen_accept(
listener, fd, address, socklen);
}
void IPStackSimulator::on_listen_accept(struct evconnlistener* listener,
evutil_socket_t fd, struct sockaddr*, int) {
int listen_fd = evconnlistener_get_fd(listener);
const ListeningSocket* listening_socket;
try {
listening_socket = &this->listening_sockets.at(listen_fd);
} catch (const out_of_range&) {
ip_stack_simulator_log.info("Virtual network %d connected via unknown listener %d; disconnecting", fd, listen_fd);
close(fd);
return;
}
ip_stack_simulator_log.info("Virtual network %d connected via %s", fd, listening_socket->name.c_str());
struct bufferevent* bev = bufferevent_socket_new(this->base.get(), fd,
BEV_OPT_CLOSE_ON_FREE | BEV_OPT_DEFER_CALLBACKS);
auto c = make_shared<IPClient>(this->shared_from_this(), listening_socket->protocol, bev);
this->bev_to_client.emplace(make_pair(bev, c));
bufferevent_setcb(bev, &IPStackSimulator::dispatch_on_client_input, nullptr,
&IPStackSimulator::dispatch_on_client_error, this);
bufferevent_enable(bev, EV_READ | EV_WRITE);
}
void IPStackSimulator::dispatch_on_listen_error(
struct evconnlistener* listener, void* ctx) {
reinterpret_cast<IPStackSimulator*>(ctx)->on_listen_error(listener);
}
void IPStackSimulator::on_listen_error(struct evconnlistener* listener) {
int err = EVUTIL_SOCKET_ERROR();
ip_stack_simulator_log.error("Failure on listening socket %d: %d (%s)",
evconnlistener_get_fd(listener), err, evutil_socket_error_to_string(err));
event_base_loopexit(this->base.get(), nullptr);
}
void IPStackSimulator::dispatch_on_client_input(
struct bufferevent* bev, void* ctx) {
reinterpret_cast<IPStackSimulator*>(ctx)->on_client_input(bev);
}
void IPStackSimulator::on_client_input(struct bufferevent* bev) {
struct evbuffer* buf = bufferevent_get_input(bev);
shared_ptr<IPClient> c;
try {
c = this->bev_to_client.at(bev);
} catch (const out_of_range&) {
size_t bytes = evbuffer_get_length(buf);
ip_stack_simulator_log.warning("Ignoring data received from unregistered virtual network (0x%zX bytes)",
bytes);
evbuffer_drain(buf, bytes);
return;
}
auto sim = c->sim.lock();
uint64_t idle_timeout_usecs = sim ? sim->state->client_idle_timeout_usecs : 60000000;
struct timeval tv = usecs_to_timeval(idle_timeout_usecs);
event_add(c->idle_timeout_event.get(), &tv);
switch (c->protocol) {
case Protocol::ETHERNET_TAPSERVER:
case Protocol::HDLC_TAPSERVER:
while (evbuffer_get_length(buf) >= 2) {
uint16_t frame_size;
evbuffer_copyout(buf, &frame_size, 2);
if (evbuffer_get_length(buf) < static_cast<size_t>(frame_size + 2)) {
break; // No complete frame available; done for now
}
evbuffer_drain(buf, 2);
string frame(frame_size, '\0');
evbuffer_remove(buf, frame.data(), frame.size());
try {
this->on_client_frame(c, frame);
} catch (const exception& e) {
if (ip_stack_simulator_log.warning("Failed to process frame: %s", e.what())) {
print_data(stderr, frame);
}
}
}
break;
case Protocol::HDLC_RAW:
while (evbuffer_get_length(buf) >= 2) {
struct evbuffer_ptr res = evbuffer_search(buf, "\x7E", 1, nullptr);
if (res.pos < 0) {
break;
}
size_t start_offset = res.pos;
if (evbuffer_ptr_set(buf, &res, 1, EVBUFFER_PTR_ADD)) {
ip_stack_simulator_log.warning("Cannot advance search for end of frame");
break;
}
struct evbuffer_ptr end_res = evbuffer_search(buf, "\x7E", 1, &res);
if (end_res.pos < 0) {
break;
}
size_t frame_size = end_res.pos + 1 - start_offset;
if (start_offset) {
evbuffer_drain(buf, start_offset);
}
string frame(frame_size, '\0');
evbuffer_remove(buf, frame.data(), frame.size());
try {
this->on_client_frame(c, frame);
} catch (const exception& e) {
if (ip_stack_simulator_log.warning("Failed to process frame: %s", e.what())) {
print_data(stderr, frame);
}
}
}
break;
}
}
void IPStackSimulator::dispatch_on_client_error(
struct bufferevent* bev, short events, void* ctx) {
reinterpret_cast<IPStackSimulator*>(ctx)->on_client_error(bev, events);
}
void IPStackSimulator::on_client_error(struct bufferevent* bev, short events) {
if (events & BEV_EVENT_ERROR) {
int err = EVUTIL_SOCKET_ERROR();
ip_stack_simulator_log.warning("Virtual network caused error %d (%s)", err,
evutil_socket_error_to_string(err));
}
if (events & (BEV_EVENT_EOF | BEV_EVENT_ERROR)) {
this->disconnect_client(bev);
}
}
void IPStackSimulator::send_layer3_frame(shared_ptr<IPClient> c, FrameInfo::Protocol proto, const string& data) const {
this->send_layer3_frame(c, proto, data.data(), data.size());
}
void IPStackSimulator::send_layer3_frame(shared_ptr<IPClient> c, FrameInfo::Protocol proto, const void* data, size_t size) const {
struct evbuffer* out_buf = bufferevent_get_output(c->bev.get());
switch (c->protocol) {
case Protocol::ETHERNET_TAPSERVER: {
EthernetHeader ether;
ether.dest_mac = c->mac_addr;
ether.src_mac = this->host_mac_address_bytes;
switch (proto) {
case FrameInfo::Protocol::NONE:
throw logic_error("layer 3 protocol not specified");
case FrameInfo::Protocol::LCP:
throw logic_error("cannot send LCP frame over Ethernet");
case FrameInfo::Protocol::IPV4:
ether.protocol = 0x0800;
break;
case FrameInfo::Protocol::ARP:
ether.protocol = 0x0806;
break;
default:
throw logic_error("unknown layer 3 protocol");
}
le_uint16_t frame_size = size + sizeof(EthernetHeader);
evbuffer_add(out_buf, &frame_size, 2);
evbuffer_add(out_buf, &ether, sizeof(ether));
evbuffer_add(out_buf, data, size);
if (this->pcap_text_log_file) {
StringWriter w;
w.write(&ether, sizeof(ether));
w.write(data, size);
this->log_frame(w.str());
}
break;
}
case Protocol::HDLC_TAPSERVER:
case Protocol::HDLC_RAW: {
HDLCHeader hdlc;
hdlc.start_sentinel1 = 0x7E;
hdlc.address = 0xFF;
hdlc.control = 0x03;
switch (proto) {
case FrameInfo::Protocol::NONE:
throw logic_error("layer 3 protocol not specified");
case FrameInfo::Protocol::LCP:
hdlc.protocol = 0xC021;
break;
case FrameInfo::Protocol::PAP:
hdlc.protocol = 0xC023;
break;
case FrameInfo::Protocol::IPCP:
hdlc.protocol = 0x8021;
break;
case FrameInfo::Protocol::IPV4:
hdlc.protocol = 0x0021;
break;
case FrameInfo::Protocol::ARP:
throw runtime_error("cannot send ARP packets over HDLC");
default:
throw logic_error("unknown layer 3 protocol");
}
StringWriter w;
w.put(hdlc);
w.write(data, size);
w.put_u16l(FrameInfo::computed_hdlc_checksum(w.str().data() + 1, w.size() - 1));
w.put_u8(0x7E);
string escaped = escape_hdlc_frame(w.str(), c->hdlc_escape_control_character_flags);
if (ip_stack_simulator_log.debug("Sending HDLC frame to virtual network (escaped to %zX bytes)", escaped.size())) {
print_data(stderr, w.str());
}
if (c->protocol == Protocol::HDLC_TAPSERVER) {
le_uint16_t frame_size = escaped.size();
evbuffer_add(out_buf, &frame_size, 2);
}
evbuffer_add(out_buf, escaped.data(), escaped.size());
if (this->pcap_text_log_file) {
this->log_frame(escaped);
}
break;
}
default:
throw logic_error("unknown link type");
}
}
void IPStackSimulator::on_client_frame(shared_ptr<IPClient> c, const string& frame) {
FrameInfo::LinkType link_type = (c->protocol == Protocol::ETHERNET_TAPSERVER)
? FrameInfo::LinkType::ETHERNET
: FrameInfo::LinkType::HDLC;
const string* effective_data = &frame;
string hdlc_unescaped_data;
if (link_type == FrameInfo::LinkType::HDLC) {
hdlc_unescaped_data = unescape_hdlc_frame(frame);
effective_data = &hdlc_unescaped_data;
}
if (ip_stack_simulator_log.debug("Virtual network sent frame")) {
print_data(stderr, *effective_data);
}
this->log_frame(*effective_data);
FrameInfo fi(link_type, *effective_data);
if (ip_stack_simulator_log.should_log(LogLevel::DEBUG)) {
string fi_header = fi.header_str();
ip_stack_simulator_log.debug("Frame header: %s", fi_header.c_str());
}
if (fi.ether) {
if (c->mac_addr.is_filled_with(0)) {
c->mac_addr = fi.ether->src_mac;
} else if ((fi.ether->src_mac != c->mac_addr) && (fi.ether->src_mac != this->broadcast_mac_address_bytes)) {
throw runtime_error("client sent IPv4 packet from different MAC address");
}
} else if (fi.hdlc) {
uint16_t expected_checksum = fi.computed_hdlc_checksum();
uint16_t stored_checksum = fi.stored_hdlc_checksum();
if (expected_checksum != stored_checksum) {
throw runtime_error(string_printf(
"HDLC checksum is incorrect (%04hX expected, %04hX received)",
expected_checksum, stored_checksum));
}
} else {
throw runtime_error("frame is not Ethernet or HDLC");
}
if (fi.lcp) {
this->on_client_lcp_frame(c, fi);
} else if (fi.pap) {
this->on_client_pap_frame(c, fi);
} else if (fi.ipcp) {
this->on_client_ipcp_frame(c, fi);
} else if (fi.arp) {
this->on_client_arp_frame(c, fi);
} else if (fi.ipv4) {
uint16_t expected_ipv4_checksum = fi.computed_ipv4_header_checksum();
if (fi.ipv4->checksum != expected_ipv4_checksum) {
throw runtime_error(string_printf(
"IPv4 header checksum is incorrect (%04hX expected, %04hX received)",
expected_ipv4_checksum, fi.ipv4->checksum.load()));
}
if ((fi.ipv4->src_addr != c->ipv4_addr) && (fi.ipv4->src_addr != 0)) {
throw runtime_error("client sent IPv4 packet from different IPv4 address");
}
if (fi.udp) {
uint16_t expected_udp_checksum = fi.computed_udp4_checksum();
if (fi.udp->checksum != expected_udp_checksum) {
throw runtime_error(string_printf(
"UDP checksum is incorrect (%04hX expected, %04hX received)",
expected_udp_checksum, fi.udp->checksum.load()));
}
this->on_client_udp_frame(c, fi);
} else if (fi.tcp) {
uint16_t expected_tcp_checksum = fi.computed_tcp4_checksum();
if (fi.tcp->checksum != expected_tcp_checksum) {
throw runtime_error(string_printf(
"TCP checksum is incorrect (%04hX expected, %04hX received)",
expected_tcp_checksum, fi.tcp->checksum.load()));
}
this->on_client_tcp_frame(c, fi);
} else {
throw runtime_error("frame uses unsupported IPv4 protocol");
}
} else {
throw runtime_error("frame is not IPv4");
}
}
void IPStackSimulator::on_client_lcp_frame(shared_ptr<IPClient> c, const FrameInfo& fi) {
switch (fi.lcp->command) {
case 0x01: { // Configure-Request
auto opts_r = fi.read_payload();
while (!opts_r.eof()) {
uint8_t opt = opts_r.get_u8();
string opt_data = opts_r.read(opts_r.get_u8() - 2);
StringReader opt_data_r(opt_data);
switch (opt) {
case 0x01: // Maximum receive unit
// TODO: Currently we ignore this, but we probably should use it.
opt_data_r.get_u16b();
break;
case 0x02: // Escaped control character flags
c->hdlc_escape_control_character_flags = opt_data_r.get_u32b();
break;
case 0x05: // Magic-Number
c->hdlc_remote_magic_number = opt_data_r.get_u32b();
break;
case 0x00: // RESERVED
case 0x03: // Authentication protocol
case 0x04: // Quality protocol
case 0x07: // Protocol field compression
case 0x08: // Address and control field compression
throw runtime_error(string_printf("unimplemented LCP option %02hhX (%zu bytes)", opt, opt_data.size()));
default:
throw runtime_error("unknown LCP option");
}
}
// Technically, we should implement the LCP state machine, but I'm too
// lazy to do this right now. In our situation, it should suffice to
// simply always send a Configure-Request to the client with a magic
// number not equal to the one we received.
StringWriter opts_w;
opts_w.put_u8(0x01); // Maximum receive unit
opts_w.put_u8(0x04);
opts_w.put_u16b(1500);
opts_w.put_u8(0x02); // Escaped control character flags (we don't require any)
opts_w.put_u8(0x06);
opts_w.put_u32(0);
opts_w.put_u8(0x03); // Authentication protocol
opts_w.put_u8(0x04);
opts_w.put_u16b(0xC023); // Password authentication protocol
opts_w.put_u8(0x05); // Magic number (bitwise inverse of the remote end's)
opts_w.put_u8(0x06);
opts_w.put_u32b(~c->hdlc_remote_magic_number);
StringWriter request_w;
request_w.put<LCPHeader>(LCPHeader{
.command = 0x01, // Configure-Request
.request_id = fi.lcp->request_id,
.size = static_cast<uint16_t>(sizeof(LCPHeader) + opts_w.size()),
});
request_w.write(opts_w.str());
this->send_layer3_frame(c, FrameInfo::Protocol::LCP, request_w.str());
StringWriter ack_w;
ack_w.put<LCPHeader>(LCPHeader{
.command = 0x02, // Configure-Ack
.request_id = fi.lcp->request_id,
.size = fi.lcp->size,
});
ack_w.write(fi.payload, fi.payload_size);
this->send_layer3_frame(c, FrameInfo::Protocol::LCP, ack_w.str());
break;
}
case 0x05: { // Terminate-Request
c->ipv4_addr = 0;
c->tcp_connections.clear();
string response(reinterpret_cast<const char*>(fi.payload), fi.payload_size);
response.at(0) = 0x06; // Terminate-Ack
this->send_layer3_frame(c, FrameInfo::Protocol::LCP, response);
break;
}
case 0x09: { // Echo-Request
string response(reinterpret_cast<const char*>(fi.payload), fi.payload_size);
response.at(0) = 0x0A; // Echo-Reply
this->send_layer3_frame(c, FrameInfo::Protocol::LCP, response);
break;
}
case 0x0B: // Discard-Request
case 0x02: // Configure-Ack
break;
case 0x03: // Configure-Nak
case 0x04: // Configure-Reject
case 0x06: // Terminate-Ack
case 0x07: // Code-Reject
case 0x08: // Protocol-Reject
case 0x0A: // Echo-Reply
throw runtime_error("unimplemented LCP command");
default:
throw runtime_error("unknown LCP command");
}
}
void IPStackSimulator::on_client_pap_frame(shared_ptr<IPClient> c, const FrameInfo& fi) {
if (fi.pap->command != 0x01) { // Authenticate-Request
throw runtime_error("client sent incorrect PAP command");
}
auto r = fi.read_payload();
string username = r.read(r.get_u8());
string password = r.read(r.get_u8());
ip_stack_simulator_log.info("Client logged in with username \"%s\" and password", username.c_str());
static const string login_message = "newserv PPP simulator";
StringWriter w;
w.put<PAPHeader>(PAPHeader{
.command = 0x02, // Authenticate-Ack
.request_id = fi.pap->request_id,
.size = login_message.size() + sizeof(PAPHeader) + 1,
});
w.put_u8(login_message.size());
w.write(login_message);
this->send_layer3_frame(c, FrameInfo::Protocol::PAP, w.str());
}
void IPStackSimulator::on_client_ipcp_frame(shared_ptr<IPClient> c, const FrameInfo& fi) {
switch (fi.ipcp->command) {
case 0x01: { // Configure-Request
auto opts_r = fi.read_payload();
uint32_t remote_ip = 0;
uint32_t remote_primary_dns = 0;
uint32_t remote_secondary_dns = 0;
StringWriter rejected_opts_w;
while (!opts_r.eof()) {
uint8_t opt = opts_r.get_u8();
string opt_data = opts_r.read(opts_r.get_u8() - 2);
StringReader opt_data_r(opt_data);
switch (opt) {
case 0x01: // IP addresses (deprecated as of 1992; we don't support it at all)
throw runtime_error("IPCP client sent IP-Addresses option");
case 0x02: // IP compression protocol
rejected_opts_w.put_u8(0x02);
rejected_opts_w.put_u8(opt_data_r.size() + 2);
rejected_opts_w.write(opt_data);
break;
case 0x03: // IP address
remote_ip = opt_data_r.get_u32b();
break;
case 0x81: // Primary DNS server address
remote_primary_dns = opt_data_r.get_u32b();
break;
case 0x83: // Secondary DNS server address
remote_secondary_dns = opt_data_r.get_u32b();
break;
case 0x82: // Primary NBNS server address
case 0x84: // Secondary NBNS server address
case 0x04: // Mobile IP address
throw runtime_error(string_printf("unimplemented IPCP option %02hhX (%zu bytes)", opt, opt_data.size()));
default:
throw runtime_error("unknown IPCP option");
}
}
if (!rejected_opts_w.str().empty()) {
// Send a Configure-Reject if the client specified IP header compression
StringWriter reject_w;
reject_w.put<IPCPHeader>(IPCPHeader{
.command = 0x04, // Configure-Reject
.request_id = fi.ipcp->request_id,
.size = sizeof(IPCPHeader) + rejected_opts_w.size(),
});
reject_w.write(rejected_opts_w.str());
this->send_layer3_frame(c, FrameInfo::Protocol::IPCP, reject_w.str());
} else if ((remote_ip != 0x1E1E1E1E) ||
(remote_primary_dns != 0x23232323) ||
(remote_secondary_dns != 0x24242424)) {
// Send a Configure-Nak if the client's request doesn't exactly match
// what we want them to use.
StringWriter opts_w;
opts_w.put_u8(0x03); // IP address
opts_w.put_u8(0x06);
opts_w.put_u32b(0x1E1E1E1E);
opts_w.put_u8(0x81); // Primary DNS server address
opts_w.put_u8(0x06);
opts_w.put_u32b(0x23232323);
opts_w.put_u8(0x83); // Secondary DNS server address
opts_w.put_u8(0x06);
opts_w.put_u32b(0x24242424);
StringWriter nak_w;
nak_w.put<IPCPHeader>(IPCPHeader{
.command = 0x03, // Configure-Nak
.request_id = fi.ipcp->request_id,
.size = static_cast<uint16_t>(opts_w.size() + sizeof(IPCPHeader)),
});
nak_w.write(opts_w.str());
this->send_layer3_frame(c, FrameInfo::Protocol::IPCP, nak_w.str());
} else { // Options OK
c->ipv4_addr = remote_ip;
// As with LCP, we technically should implement the state machine, but I
// continue to be lazy.
StringWriter opts_w;
opts_w.put_u8(0x03); // IP address
opts_w.put_u8(0x06);
opts_w.put_u32b(0x39393939);
opts_w.put_u8(0x81); // Primary DNS server address
opts_w.put_u8(0x06);
opts_w.put_u32b(0x23232323);
opts_w.put_u8(0x83); // Secondary DNS server address
opts_w.put_u8(0x06);
opts_w.put_u32b(0x24242424);
StringWriter request_w;
request_w.put<IPCPHeader>(IPCPHeader{
.command = 0x01, // Configure-Request
.request_id = fi.ipcp->request_id,
.size = static_cast<uint16_t>(opts_w.size() + sizeof(IPCPHeader)),
});
request_w.write(opts_w.str());
this->send_layer3_frame(c, FrameInfo::Protocol::IPCP, request_w.str());
StringWriter ack_w;
ack_w.put<IPCPHeader>(IPCPHeader{
.command = 0x02, // Configure-Ack
.request_id = fi.ipcp->request_id,
.size = fi.ipcp->size,
});
ack_w.write(fi.payload, fi.payload_size);
this->send_layer3_frame(c, FrameInfo::Protocol::IPCP, ack_w.str());
}
break;
}
case 0x05: { // Terminate-Request
c->ipv4_addr = 0;
c->tcp_connections.clear();
string response(reinterpret_cast<const char*>(fi.payload), fi.payload_size);
response.at(0) = 0x06; // Terminate-Ack
this->send_layer3_frame(c, FrameInfo::Protocol::LCP, response);
break;
}
case 0x02: // Configure-Ack
break;
case 0x03: // Configure-Nak
case 0x04: // Configure-Reject
case 0x06: // Terminate-Ack
case 0x07: // Code-Reject
throw runtime_error("unimplemented IPCP command");
default:
throw runtime_error("unknown LCP command");
}
}
void IPStackSimulator::on_client_arp_frame(
shared_ptr<IPClient> c, const FrameInfo& fi) {
if (fi.arp->hwaddr_len != 6 ||
fi.arp->paddr_len != 4 ||
fi.arp->hardware_type != 0x0001 ||
fi.arp->protocol_type != 0x0800) {
throw runtime_error("unsupported ARP parameters");
}
if (fi.payload_size < 20) {
throw runtime_error("ARP payload too small");
}
if (c->ipv4_addr == 0) {
c->ipv4_addr = *reinterpret_cast<const be_uint32_t*>(
reinterpret_cast<const uint8_t*>(fi.payload) + 6);
}
StringWriter w;
w.put<ARPHeader>(ARPHeader{
.hardware_type = fi.arp->hardware_type,
.protocol_type = fi.arp->protocol_type,
.hwaddr_len = 6,
.paddr_len = 4,
.operation = 0x0002,
});
// The incoming payload is:
// uint8_t src_mac[6]; // MAC address of client
// uint8_t src_ip[4]; // IP address of client
// uint8_t dest_mac[6]; // MAC address of host (all zeroes)
// uint8_t dest_ip[4]; // IP address of host
// The outgoing payload is:
// uint8_t dest_mac[6]; // MAC address of host (from configuration)
// uint8_t dest_ip[4]; // IP address of host
// uint8_t src_mac[6]; // MAC address of client
// uint8_t src_ip[4]; // IP address of client
const char* payload_bytes = reinterpret_cast<const char*>(fi.payload);
w.write(this->host_mac_address_bytes.data(), 6);
w.write(payload_bytes + 16, 4);
w.write(payload_bytes, 10);
this->send_layer3_frame(c, FrameInfo::Protocol::ARP, w.str());
}
void IPStackSimulator::on_client_udp_frame(shared_ptr<IPClient> c, const FrameInfo& fi) {
// We only implement DHCP and newserv's DNS server here.
// Every received UDP packet will elicit exactly one UDP response from
// newserv, so we prepare the response headers in advance
IPv4Header r_ipv4;
r_ipv4.version_ihl = 0x45;
r_ipv4.tos = 0;
// r_ipv4.size filled in later
r_ipv4.id = 0;
r_ipv4.frag_offset = 0;
r_ipv4.ttl = 20; // TODO: Does this value actually matter? Looks like it just has to be nonzero
r_ipv4.protocol = 17; // UDP
// r_ipv4.checksum filled in later
r_ipv4.src_addr = fi.ipv4->dest_addr;
r_ipv4.dest_addr = fi.ipv4->src_addr;
UDPHeader r_udp;
r_udp.src_port = fi.udp->dest_port;
r_udp.dest_port = fi.udp->src_port;
// r_udp.size filled in later
// r_udp.checksum filled in later
string r_data;
if (fi.udp->dest_port == 67) { // DHCP
auto r = fi.read_payload();
const auto& dhcp = r.get<DHCPHeader>();
if (dhcp.hardware_type != 1) {
throw runtime_error("unknown DHCP hardware type");
}
if (dhcp.hardware_address_length != 6) {
throw runtime_error("unknown DHCP hardware address length");
}
if (dhcp.magic != 0x63825363) {
throw runtime_error("incorrect DHCP magic cookie");
}
if (dhcp.opcode != 1) { // Request
throw runtime_error("DHCP packet is not a request");
}
unordered_map<uint8_t, string> option_data;
for (;;) {
uint8_t option = r.get_u8();
if (option == 0xFF) {
break;
}
uint8_t size = r.get_u8();
option_data.emplace(option, r.read(size));
}
uint8_t command = 0;
try {
command = option_data.at(53).at(0);
} catch (const out_of_range&) {
throw runtime_error("client did not send a DHCP command option");
}
if (command == 7) {
// Release IP address (we just ignore these)
} else if ((command == 1) || (command == 3)) {
// Populate the client's addresses
c->mac_addr = dhcp.client_hardware_address.data();
c->ipv4_addr = 0x0A000105; // 10.0.1.5
// In this case, the client doesn't know its IPv4 address or ours yet,
// so we overwrite the existing fields with the appropriate addresses.
r_ipv4.src_addr = 0x0A000101; // 10.0.1.1
r_ipv4.dest_addr = c->ipv4_addr;
if ((command != 1) && (command != 3)) {
throw runtime_error("client sent unknown DHCP command option");
}
StringWriter w;
DHCPHeader r_dhcp;
r_dhcp.opcode = 2; // Response
r_dhcp.hardware_type = 1; // Ethernet
r_dhcp.hardware_address_length = 6; // Ethernet
r_dhcp.hops = 0;
r_dhcp.transaction_id = dhcp.transaction_id;
r_dhcp.seconds_elapsed = 0;
r_dhcp.flags = 0;
r_dhcp.client_ip_address = 0;
r_dhcp.your_ip_address = r_ipv4.dest_addr;
r_dhcp.server_ip_address = r_ipv4.src_addr;
r_dhcp.gateway_ip_address = 0;
r_dhcp.client_hardware_address = c->mac_addr;
r_dhcp.unused_bootp_legacy.clear(0);
r_dhcp.magic = 0x63825363;
w.put(r_dhcp);
// DHCP message type option
w.put_u8(53);
w.put_u8(1);
w.put_u8(static_cast<uint8_t>((command == 3) ? 5 : 2)); // Offer or ack
// DHCP server ID option
w.put_u8(54);
w.put_u8(4);
w.put_u32b(0x0A000101); // 10.0.1.1
// Lease time option
w.put_u8(51);
w.put_u8(4);
w.put_u32b(60 * 60 * 24 * 7); // 1 week
// Renewal time option
w.put_u8(58);
w.put_u8(4);
w.put_u32b(60 * 60 * 24 * 7); // 1 week
// Rebind time option
w.put_u8(59);
w.put_u8(4);
w.put_u32b(60 * 60 * 24 * 7); // 1 week
// Subnet mask option
w.put_u8(1);
w.put_u8(4);
w.put_u32b(0xFFFFFF00); // 255.255.255.0
// Broadcast IP option
w.put_u8(28);
w.put_u8(4);
w.put_u32b(c->ipv4_addr | 0x000000FF);
// DNS server option
w.put_u8(6);
w.put_u8(4);
w.put_u32b(0x0A000101); // 10.0.1.1
// Domain name option
w.put_u8(15);
w.put_u8(7);
w.write("newserv");
// Default gateway option
w.put_u8(3);
w.put_u8(4);
w.put_u32b(0x0A000101); // 10.0.1.1
// End option list
w.put_u8(0xFF);
r_data = std::move(w.str());
} else {
throw runtime_error("client sent unknown DHCP command");
}
} else if (fi.udp->dest_port == 53) { // DNS
if (fi.payload_size < 0x0C) {
throw runtime_error("DNS payload too small");
}
uint32_t resolved_address = this->connect_address_for_remote_address(c->ipv4_addr);
r_data = DNSServer::response_for_query(fi.payload, fi.payload_size, resolved_address);
} else { // Not DHCP or DNS
throw runtime_error("UDP packet is not DHCP or DNS");
}
if (!r_data.empty()) {
r_ipv4.size = sizeof(IPv4Header) + sizeof(UDPHeader) + r_data.size();
r_udp.size = sizeof(UDPHeader) + r_data.size();
r_ipv4.checksum = FrameInfo::computed_ipv4_header_checksum(r_ipv4);
r_udp.checksum = FrameInfo::computed_udp4_checksum(
r_ipv4, r_udp, r_data.data(), r_data.size());
if (ip_stack_simulator_log.should_log(LogLevel::DEBUG)) {
string remote_str = this->str_for_ipv4_netloc(fi.ipv4->src_addr, fi.udp->src_port);
ip_stack_simulator_log.debug("Sending UDP response to %s", remote_str.c_str());
print_data(stderr, r_data);
}
StringWriter w;
w.put(r_ipv4);
w.put(r_udp);
w.write(r_data);
this->send_layer3_frame(c, FrameInfo::Protocol::IPV4, w.str());
}
}
uint64_t IPStackSimulator::tcp_conn_key_for_connection(
const IPClient::TCPConnection& conn) {
return (static_cast<uint64_t>(conn.server_addr) << 32) |
(static_cast<uint64_t>(conn.server_port) << 16) |
static_cast<uint64_t>(conn.client_port);
}
uint64_t IPStackSimulator::tcp_conn_key_for_client_frame(
const IPv4Header& ipv4, const TCPHeader& tcp) {
return (static_cast<uint64_t>(ipv4.dest_addr) << 32) |
(static_cast<uint64_t>(tcp.dest_port) << 16) |
static_cast<uint64_t>(tcp.src_port);
}
uint64_t IPStackSimulator::tcp_conn_key_for_client_frame(const FrameInfo& fi) {
if (!fi.ipv4 || !fi.tcp) {
throw logic_error("tcp_conn_key_for_frame called on non-TCP frame");
}
return IPStackSimulator::tcp_conn_key_for_client_frame(*fi.ipv4, *fi.tcp);
}
void IPStackSimulator::on_client_tcp_frame(
shared_ptr<IPClient> c, const FrameInfo& fi) {
ip_stack_simulator_log.debug("Virtual network sent TCP frame (seq=%08" PRIX32 ", ack=%08" PRIX32 ")",
fi.tcp->seq_num.load(), fi.tcp->ack_num.load());
if (fi.tcp->flags & (TCPHeader::Flag::NS | TCPHeader::Flag::CWR | TCPHeader::Flag::ECE | TCPHeader::Flag::URG)) {
throw runtime_error("unsupported flag in TCP packet");
}
if (fi.tcp->flags & TCPHeader::Flag::SYN) {
// We never make connections back to the client, so we should never receive
// a SYN+ACK. Essentially, no other flags should be set in any received SYN.
if ((fi.tcp->flags & 0x0FFF) != TCPHeader::Flag::SYN) {
throw runtime_error("TCP SYN contains extra flags");
}
StringReader options_r(fi.tcp + 1, fi.tcp_options_size);
size_t max_frame_size = 1400;
while (!options_r.eof()) {
uint8_t option = options_r.get_u8();
uint8_t option_size = (option < 2) ? 1 : options_r.get_u8();
switch (option) {
case 0: // End of options list
options_r.go(options_r.size());
break;
case 1: // No option (padding)
break;
case 2: // Max segment size
if (option_size != 4) {
throw runtime_error("incorrect size for TCP max frame size option");
}
max_frame_size = options_r.get_u16b();
break;
case 3: // Window scale (ignored)
if (option_size != 3) {
throw runtime_error("incorrect size for TCP window scale option");
}
options_r.skip(option_size);
break;
case 4: // Selective ACK supported (ignored)
if (option_size != 2) {
throw runtime_error("incorrect size for TCP selective ACK supported option");
}
break;
case 5: // Selective ACK (ignored)
options_r.skip(option_size - 2);
break;
case 8: // Timestamps (ignored)
if (option_size != 10) {
throw runtime_error("incorrect size for TCP timestamps option");
}
options_r.skip(8);
break;
default:
throw runtime_error("invalid TCP option");
}
}
uint64_t key = this->tcp_conn_key_for_client_frame(fi);
auto emplace_ret = c->tcp_connections.emplace(key, IPClient::TCPConnection());
auto& conn = emplace_ret.first->second;
string conn_str;
if (emplace_ret.second) {
// Connection is new; initialize it
conn.client = c;
conn.resend_push_event.reset(event_new(this->base.get(), -1, EV_TIMEOUT,
&IPStackSimulator::dispatch_on_resend_push, &conn));
conn.server_addr = fi.ipv4->dest_addr;
conn.server_port = fi.tcp->dest_port;
conn.client_port = fi.tcp->src_port;
conn.next_client_seq = fi.tcp->seq_num + 1;
conn.acked_server_seq = random_object<uint32_t>();
conn.resend_push_usecs = DEFAULT_RESEND_PUSH_USECS;
conn.next_push_max_frame_size = max_frame_size;
conn.awaiting_first_ack = true;
conn.max_frame_size = max_frame_size;
conn.bytes_received = 0;
conn.bytes_sent = 0;
conn_str = this->str_for_tcp_connection(c, conn);
ip_stack_simulator_log.info("Client opened TCP connection %s (acked_server_seq=%08" PRIX32 ", next_client_seq=%08" PRIX32 ")",
conn_str.c_str(), conn.acked_server_seq, conn.next_client_seq);
} else {
// Connection is NOT new; this is probably a resend of an earlier SYN
if (!conn.awaiting_first_ack) {
throw logic_error("SYN received on already-open connection after initial phase");
}
// TODO: We should check the syn/ack numbers here instead of just assuming
// they're correct
conn_str = this->str_for_tcp_connection(c, conn);
ip_stack_simulator_log.debug("Client resent SYN for TCP connection %s",
conn_str.c_str());
}
// Send a SYN+ACK (send_tcp_frame always adds the ACK flag)
this->send_tcp_frame(c, conn, TCPHeader::Flag::SYN);
ip_stack_simulator_log.debug("Sent SYN+ACK on %s (acked_server_seq=%08" PRIX32 ", next_client_seq=%08" PRIX32 ")",
conn_str.c_str(), conn.acked_server_seq, conn.next_client_seq);
} else {
// This frame isn't a SYN, so a connection object should already exist
uint64_t key = this->tcp_conn_key_for_client_frame(fi);
IPClient::TCPConnection* conn;
try {
conn = &c->tcp_connections.at(key);
} catch (const out_of_range&) {
throw runtime_error("non-SYN frame does not correspond to any open TCP connection");
}
bool conn_valid = true;
bool acked_seq_changed = false;
if (fi.tcp->flags & TCPHeader::Flag::ACK) {
ip_stack_simulator_log.debug("Client sent ACK %08" PRIX32, fi.tcp->ack_num.load());
if (conn->awaiting_first_ack) {
if (fi.tcp->ack_num != conn->acked_server_seq + 1) {
throw runtime_error("first ack_num was not acked_server_seq + 1");
}
conn->acked_server_seq++;
conn->awaiting_first_ack = false;
} else {
if (seq_num_greater(fi.tcp->ack_num, conn->acked_server_seq)) {
ip_stack_simulator_log.debug("Advancing acked_server_seq from %08" PRIX32, conn->acked_server_seq);
uint32_t ack_delta = fi.tcp->ack_num - conn->acked_server_seq;
size_t pending_bytes = evbuffer_get_length(conn->pending_data.get());
if (pending_bytes < ack_delta) {
throw runtime_error("client acknowledged beyond end of sent data");
}
evbuffer_drain(conn->pending_data.get(), ack_delta);
conn->acked_server_seq += ack_delta;
conn->resend_push_usecs = DEFAULT_RESEND_PUSH_USECS;
conn->next_push_max_frame_size = conn->max_frame_size;
acked_seq_changed = true;
ip_stack_simulator_log.debug("Removed %08" PRIX32 " bytes from pending buffer and advanced acked_server_seq to %08" PRIX32,
ack_delta, conn->acked_server_seq);
} else if (seq_num_less(fi.tcp->ack_num, conn->acked_server_seq)) {
throw runtime_error("client sent lower ack num than previous frame");
}
}
if (!conn->server_bev.get()) {
this->open_server_connection(c, *conn);
}
}
if (fi.tcp->flags & (TCPHeader::Flag::RST | TCPHeader::Flag::FIN)) {
bool is_rst = (fi.tcp->flags & TCPHeader::Flag::RST);
if (is_rst && (fi.tcp->flags & TCPHeader::Flag::FIN)) {
throw runtime_error("client sent TCP FIN+RST");
}
string conn_str = this->str_for_tcp_connection(c, *conn);
ip_stack_simulator_log.info("Client closed TCP connection %s", conn_str.c_str());
// TODO: Are we supposed to send a response to an RST? Here we do, and the
// client probably just ignores it anyway
this->send_tcp_frame(c, *conn, fi.tcp->flags & (TCPHeader::Flag::RST | TCPHeader::Flag::FIN));
// Delete the connection object. The unique_ptr destructor flushes the
// bufferevent, and thereby sends an EOF to the server's end.
c->tcp_connections.erase(key);
conn_valid = false;
// Note: The PSH flag isn't required to be set on all packets that contain
// data. The PSH flag just means "tell the application that data is
// available", so some senders only set the PSH flag on the last frame of a
// large segment of data, since the application wouldn't be able to process
// the segment until all of it is available. newserv can handle incomplete
// commands, so we just ignore the PSH flag and forward any data to the
// server immediately.
} else if (fi.payload_size != 0) {
string conn_str = ip_stack_simulator_log.should_log(LogLevel::WARNING)
? this->str_for_tcp_connection(c, *conn)
: "";
size_t payload_skip_bytes;
if (fi.tcp->seq_num == conn->next_client_seq) {
payload_skip_bytes = 0;
} else if (seq_num_less(fi.tcp->seq_num, conn->next_client_seq)) {
// If the frame overlaps an existing boundary, we'll accept some of the
// data; otherwise we'll ignore it entirely (but still send an ACK)
uint32_t end_seq = fi.tcp->seq_num + fi.payload_size;
if (seq_num_less_or_equal(end_seq, conn->next_client_seq)) { // Fully "in the past"
payload_skip_bytes = fi.payload_size;
} else { // Partially "in the past"
payload_skip_bytes = fi.payload_size - (end_seq - conn->next_client_seq);
}
} else {
// Payload is in the future - we must have missed a data frame. We'll
// ignore it (but warn) and send an ACK later, and the client should
// retransmit the lost data
ip_stack_simulator_log.warning(
"Client sent out-of-order sequence number (expected %08" PRIX32 ", received %08" PRIX32 ", 0x%zX data bytes)",
conn->next_client_seq, fi.tcp->seq_num.load(), fi.payload_size);
payload_skip_bytes = fi.payload_size;
}
if (payload_skip_bytes > fi.payload_size) {
throw logic_error("payload skip bytes too large");
}
if (payload_skip_bytes < fi.payload_size) {
const void* payload = reinterpret_cast<const uint8_t*>(fi.payload) + payload_skip_bytes;
size_t payload_size = fi.payload_size - payload_skip_bytes;
bool was_logged;
if (payload_skip_bytes) {
was_logged = ip_stack_simulator_log.debug("Client sent data on TCP connection %s, overlapping existing ack'ed data (0x%zX bytes ignored)",
conn_str.c_str(), payload_skip_bytes);
} else {
was_logged = ip_stack_simulator_log.debug("Client sent data on TCP connection %s",
conn_str.c_str());
}
if (was_logged) {
print_data(stderr, payload, payload_size);
}
// Send the new data to the server
struct evbuffer* server_out_buf = bufferevent_get_output(
conn->server_bev.get());
evbuffer_add(server_out_buf, payload, payload_size);
// Update the sequence number and stats
conn->next_client_seq += payload_size;
conn->bytes_received += payload_size;
if (conn->next_client_seq < payload_size) {
ip_stack_simulator_log.warning("Client sequence number has wrapped (next=%08" PRIX32 ", bytes=%zX)",
fi.tcp->seq_num.load(), payload_size);
}
}
// Send an ACK
this->send_tcp_frame(c, *conn);
ip_stack_simulator_log.debug("Sent PSH ACK on %s (acked_server_seq=%08" PRIX32 ", next_client_seq=%08" PRIX32 ", bytes_received=0x%zX)",
conn_str.c_str(), conn->acked_server_seq, conn->next_client_seq, conn->bytes_received);
}
if (conn_valid && acked_seq_changed) {
// Try to send some more data if the client is waiting on it
this->send_pending_push_frame(c, *conn, true);
}
}
}
void IPStackSimulator::open_server_connection(shared_ptr<IPClient> c, IPClient::TCPConnection& conn) {
if (conn.server_bev.get()) {
throw logic_error("server connection is already open");
}
struct bufferevent* bevs[2];
bufferevent_pair_new(this->base.get(), 0, bevs);
// Set up the IPStackSimulator end of the virtual connection
bufferevent_setcb(bevs[0], &IPStackSimulator::dispatch_on_server_input,
nullptr, &IPStackSimulator::dispatch_on_server_error, &conn);
bufferevent_enable(bevs[0], EV_READ | EV_WRITE);
conn.server_bev.reset(bevs[0]);
// Link the client to the server - the server sees this as a normal TCP
// connection and treats it as if the client connected to one of its listening
// sockets
shared_ptr<const PortConfiguration> port_config;
try {
port_config = this->state->number_to_port_config.at(conn.server_port);
} catch (const out_of_range&) {
bufferevent_free(bevs[1]);
throw logic_error("client connected to port missing from configuration");
}
string conn_str = this->str_for_tcp_connection(c, conn);
if (port_config->behavior == ServerBehavior::PROXY_SERVER) {
if (!this->state->proxy_server.get()) {
ip_stack_simulator_log.error("TCP connection %s is to non-running proxy server",
conn_str.c_str());
flush_and_free_bufferevent(bevs[1]);
} else {
this->state->proxy_server->connect_client(bevs[1], conn.server_port);
ip_stack_simulator_log.info("Connected TCP connection %s to proxy server",
conn_str.c_str());
}
} else if (this->state->game_server.get()) {
this->state->game_server->connect_client(bevs[1], c->ipv4_addr,
conn.client_port, conn.server_port, port_config->version,
port_config->behavior);
ip_stack_simulator_log.info("Connected TCP connection %s to game server",
conn_str.c_str());
} else {
ip_stack_simulator_log.error("No server available for TCP connection %s",
conn_str.c_str());
flush_and_free_bufferevent(bevs[1]);
}
}
void IPStackSimulator::send_pending_push_frame(
shared_ptr<IPClient> c, IPClient::TCPConnection& conn, bool always_send) {
size_t pending_bytes = evbuffer_get_length(conn.pending_data.get());
if (!pending_bytes) {
event_del(conn.resend_push_event.get());
return;
}
// If we're waiting to receive an ACK from the client, don't send another PSH
// until we get the ACK (unless this is a resend of a previous PSH due to a
// timeout)
if (!always_send && event_pending(conn.resend_push_event.get(), EV_TIMEOUT, nullptr)) {
return;
}
size_t bytes_to_send = min<size_t>(pending_bytes, conn.next_push_max_frame_size);
if (c->protocol == Protocol::HDLC_TAPSERVER) {
// There is a bug in Dolphin's modem implementation (which I wrote, so it's
// my fault) that causes commands to be dropped when too much data is sent
// at once. To work around this, we only send up to 200 bytes in each push
// frame.
bytes_to_send = min<size_t>(bytes_to_send, 200);
}
ip_stack_simulator_log.debug("Sending PSH frame with seq_num %08" PRIX32 ", 0x%zX/0x%zX data bytes",
conn.acked_server_seq, bytes_to_send, pending_bytes);
this->send_tcp_frame(c, conn, TCPHeader::Flag::PSH, conn.pending_data.get(), bytes_to_send);
struct timeval resend_push_timeout = usecs_to_timeval(conn.resend_push_usecs);
event_add(conn.resend_push_event.get(), &resend_push_timeout);
// If the client isn't responding to our PSHes, back off exponentially up to
// a limit of 5 seconds between PSH frames. This window is reset when
// acked_server_seq changes (that is, when the client has acknowledged any new
// data). It seems some situations cause GameCube clients to drop packets more
// often; to alleviate this, we also try to resend less data.
conn.resend_push_usecs *= 2;
if (conn.resend_push_usecs > 5000000) {
conn.resend_push_usecs = 5000000;
}
conn.next_push_max_frame_size = max<size_t>(0x100, conn.next_push_max_frame_size - 0x100);
}
void IPStackSimulator::send_tcp_frame(
shared_ptr<IPClient> c,
IPClient::TCPConnection& conn,
uint16_t flags,
struct evbuffer* src_buf,
size_t src_bytes) {
if (!src_bytes != !(flags & TCPHeader::Flag::PSH)) {
throw logic_error("data should be given if and only if PSH is given");
}
IPv4Header ipv4;
ipv4.version_ihl = 0x45;
ipv4.tos = 0;
// ipv4.size filled in later
ipv4.id = 0;
ipv4.frag_offset = 0;
ipv4.ttl = 20;
ipv4.protocol = 6; // TCP
// ipv4.checksum filled in later
ipv4.src_addr = conn.server_addr;
ipv4.dest_addr = c->ipv4_addr;
TCPHeader tcp;
tcp.src_port = conn.server_port;
tcp.dest_port = conn.client_port;
tcp.seq_num = conn.acked_server_seq;
tcp.ack_num = conn.next_client_seq;
tcp.flags = (5 << 12) | TCPHeader::Flag::ACK | flags;
tcp.window = 0x1000;
tcp.urgent_ptr = 0;
// tcp.checksum filled in later
ipv4.size = sizeof(IPv4Header) + sizeof(TCPHeader) + src_bytes;
ipv4.checksum = FrameInfo::computed_ipv4_header_checksum(ipv4);
const void* linear_data = src_bytes ? evbuffer_pullup(src_buf, src_bytes) : nullptr;
tcp.checksum = FrameInfo::computed_tcp4_checksum(ipv4, tcp, linear_data, src_bytes);
StringWriter w;
w.put(ipv4);
w.put(tcp);
if (src_bytes) {
w.write(linear_data, src_bytes);
}
this->send_layer3_frame(c, FrameInfo::Protocol::IPV4, w.str());
}
void IPStackSimulator::dispatch_on_resend_push(evutil_socket_t, short, void* ctx) {
auto* conn = reinterpret_cast<IPClient::TCPConnection*>(ctx);
auto c = conn->client.lock();
if (!c.get()) {
ip_stack_simulator_log.warning("Resend push event triggered for deleted client; ignoring");
} else {
auto sim = c->sim.lock();
if (!sim) {
ip_stack_simulator_log.warning("Resend push event triggered for client on deleted simulator; ignoring");
} else {
sim->send_pending_push_frame(c, *conn, true);
}
}
}
void IPStackSimulator::dispatch_on_server_input(struct bufferevent*, void* ctx) {
auto* conn = reinterpret_cast<IPClient::TCPConnection*>(ctx);
auto c = conn->client.lock();
if (!c.get()) {
ip_stack_simulator_log.warning("Server input event triggered for deleted client; ignoring");
} else {
auto sim = c->sim.lock();
if (!sim) {
ip_stack_simulator_log.warning("Server input event triggered for client on deleted simulator; ignoring");
} else {
sim->on_server_input(c, *conn);
}
}
}
void IPStackSimulator::on_server_input(shared_ptr<IPClient> c, IPClient::TCPConnection& conn) {
struct evbuffer* buf = bufferevent_get_input(conn.server_bev.get());
ip_stack_simulator_log.debug("Server input event: 0x%zX bytes to read",
evbuffer_get_length(buf));
auto sim = c->sim.lock();
uint64_t idle_timeout_usecs = sim ? sim->state->client_idle_timeout_usecs : 60000000;
struct timeval tv = usecs_to_timeval(idle_timeout_usecs);
event_add(c->idle_timeout_event.get(), &tv);
evbuffer_add_buffer(conn.pending_data.get(), buf);
this->send_pending_push_frame(c, conn, false);
}
void IPStackSimulator::dispatch_on_server_error(
struct bufferevent*, short events, void* ctx) {
auto* conn = reinterpret_cast<IPClient::TCPConnection*>(ctx);
auto c = conn->client.lock();
if (!c.get()) {
ip_stack_simulator_log.warning("Server error event triggered for deleted client; ignoring");
} else {
auto sim = c->sim.lock();
if (!sim) {
ip_stack_simulator_log.warning("Server error event triggered for client on deleted simulator; ignoring");
} else {
sim->on_server_error(c, *conn, events);
}
}
}
void IPStackSimulator::on_server_error(
shared_ptr<IPClient> c, IPClient::TCPConnection& conn, short events) {
if (events & BEV_EVENT_ERROR) {
int err = EVUTIL_SOCKET_ERROR();
ip_stack_simulator_log.warning("Received error %d from virtual connection (%s)", err,
evutil_socket_error_to_string(err));
}
if (events & (BEV_EVENT_EOF | BEV_EVENT_ERROR)) {
// Send an RST to the client. Kind of rude (we really should use FIN) but
// the PSO network stack always sends an RST to us when disconnecting, so
// whatever
this->send_tcp_frame(c, conn, TCPHeader::Flag::RST);
// Delete the connection object (this also flushes and frees the server
// virtual connection bufferevent)
string conn_str = this->str_for_tcp_connection(c, conn);
ip_stack_simulator_log.info("Server closed TCP connection %s", conn_str.c_str());
c->tcp_connections.erase(this->tcp_conn_key_for_connection(conn));
}
}
void IPStackSimulator::log_frame(const string& data) const {
if (this->pcap_text_log_file) {
print_data(this->pcap_text_log_file, data, 0, nullptr,
PrintDataFlags::SKIP_SEPARATOR);
fputc('\n', this->pcap_text_log_file);
fflush(this->pcap_text_log_file);
}
}
Reference in New Issue View Git Blame Copy Permalink