lokinet/llarp/quic/endpoint.hpp

258 lines
10 KiB
C++
Raw Normal View History

2021-03-10 15:11:42 +00:00
#pragma once
#include "address.hpp"
#include "connection.hpp"
#include "io_result.hpp"
#include "null_crypto.hpp"
#include "packet.hpp"
#include "stream.hpp"
QUIC lokinet integration refactor Refactors how quic packets get handled: the actual tunnels now live in tunnel.hpp's TunnelManager which holds and manages all the quic<->tcp tunnelling. service::Endpoint now holds a TunnelManager rather than a quic::Server. We only need one quic server, but we need a separate quic client instance per outgoing quic tunnel, and TunnelManager handles all that glue now. Adds QUIC packet handling to get to the right tunnel code. This required multiplexing incoming quic packets, as follows: Adds a very small quic tunnel packet header of 4 bytes: [1, SPORT, ECN] for client->server packets, where SPORT is our source "port" (really: just a uint16_t unique quic instance identifier) or [2, DPORT, ECN] for server->client packets where the DPORT is the SPORT from above. (This also reworks ECN bits to get properly carried over lokinet.) We don't need a destination/source port for the server-side because there is only ever one quic server (and we know we're going to it when the first byte of the header is 1). Removes the config option for quic exposing ports; a full lokinet will simply accept anything incoming on quic and tunnel it to the requested port on the the local endpoint IP (this handler will come in a following commit). Replace ConvoTags with full addresses: we need to carry the port, as well, which the ConvoTag can't give us, so change those to more general SockAddrs from which we can extract both the ConvoTag *and* the port. Add a pending connection queue along with new quic-side handlers to call when a stream becomes available (TunnelManager uses this to wire up pending incoming conns with quic streams as streams open up). Completely get rid of tunnel_server/tunnel_client.cpp code; it is now moved to tunnel.hpp. Add listen()/forget() methods in TunnelManager for setting up quic listening sockets (for liblokinet usage). Add open()/close() methods in TunnelManager for spinning up new quic clients for outgoing quic connections.
2021-03-23 19:26:32 +00:00
#include <llarp/net/ip_packet.hpp>
2021-03-10 15:11:42 +00:00
#include <chrono>
#include <map>
#include <memory>
#include <queue>
#include <unordered_map>
#include <vector>
QUIC lokinet integration refactor Refactors how quic packets get handled: the actual tunnels now live in tunnel.hpp's TunnelManager which holds and manages all the quic<->tcp tunnelling. service::Endpoint now holds a TunnelManager rather than a quic::Server. We only need one quic server, but we need a separate quic client instance per outgoing quic tunnel, and TunnelManager handles all that glue now. Adds QUIC packet handling to get to the right tunnel code. This required multiplexing incoming quic packets, as follows: Adds a very small quic tunnel packet header of 4 bytes: [1, SPORT, ECN] for client->server packets, where SPORT is our source "port" (really: just a uint16_t unique quic instance identifier) or [2, DPORT, ECN] for server->client packets where the DPORT is the SPORT from above. (This also reworks ECN bits to get properly carried over lokinet.) We don't need a destination/source port for the server-side because there is only ever one quic server (and we know we're going to it when the first byte of the header is 1). Removes the config option for quic exposing ports; a full lokinet will simply accept anything incoming on quic and tunnel it to the requested port on the the local endpoint IP (this handler will come in a following commit). Replace ConvoTags with full addresses: we need to carry the port, as well, which the ConvoTag can't give us, so change those to more general SockAddrs from which we can extract both the ConvoTag *and* the port. Add a pending connection queue along with new quic-side handlers to call when a stream becomes available (TunnelManager uses this to wire up pending incoming conns with quic streams as streams open up). Completely get rid of tunnel_server/tunnel_client.cpp code; it is now moved to tunnel.hpp. Add listen()/forget() methods in TunnelManager for setting up quic listening sockets (for liblokinet usage). Add open()/close() methods in TunnelManager for spinning up new quic clients for outgoing quic connections.
2021-03-23 19:26:32 +00:00
#include <uvw/async.h>
2021-03-10 15:11:42 +00:00
#include <uvw/timer.h>
2021-03-26 13:16:43 +00:00
namespace llarp
{
2021-03-26 13:16:43 +00:00
class EndpointBase;
} // namespace llarp
2021-03-10 15:11:42 +00:00
namespace llarp::quic
{
using namespace std::literals;
inline constexpr auto IDLE_TIMEOUT = 5min;
QUIC lokinet integration refactor Refactors how quic packets get handled: the actual tunnels now live in tunnel.hpp's TunnelManager which holds and manages all the quic<->tcp tunnelling. service::Endpoint now holds a TunnelManager rather than a quic::Server. We only need one quic server, but we need a separate quic client instance per outgoing quic tunnel, and TunnelManager handles all that glue now. Adds QUIC packet handling to get to the right tunnel code. This required multiplexing incoming quic packets, as follows: Adds a very small quic tunnel packet header of 4 bytes: [1, SPORT, ECN] for client->server packets, where SPORT is our source "port" (really: just a uint16_t unique quic instance identifier) or [2, DPORT, ECN] for server->client packets where the DPORT is the SPORT from above. (This also reworks ECN bits to get properly carried over lokinet.) We don't need a destination/source port for the server-side because there is only ever one quic server (and we know we're going to it when the first byte of the header is 1). Removes the config option for quic exposing ports; a full lokinet will simply accept anything incoming on quic and tunnel it to the requested port on the the local endpoint IP (this handler will come in a following commit). Replace ConvoTags with full addresses: we need to carry the port, as well, which the ConvoTag can't give us, so change those to more general SockAddrs from which we can extract both the ConvoTag *and* the port. Add a pending connection queue along with new quic-side handlers to call when a stream becomes available (TunnelManager uses this to wire up pending incoming conns with quic streams as streams open up). Completely get rid of tunnel_server/tunnel_client.cpp code; it is now moved to tunnel.hpp. Add listen()/forget() methods in TunnelManager for setting up quic listening sockets (for liblokinet usage). Add open()/close() methods in TunnelManager for spinning up new quic clients for outgoing quic connections.
2021-03-23 19:26:32 +00:00
inline constexpr std::byte CLIENT_TO_SERVER{1};
inline constexpr std::byte SERVER_TO_CLIENT{2};
/// QUIC Tunnel Endpoint; this is the class that implements either end of a quic tunnel for both
/// servers and clients.
2021-03-10 15:11:42 +00:00
class Endpoint
{
QUIC lokinet integration refactor Refactors how quic packets get handled: the actual tunnels now live in tunnel.hpp's TunnelManager which holds and manages all the quic<->tcp tunnelling. service::Endpoint now holds a TunnelManager rather than a quic::Server. We only need one quic server, but we need a separate quic client instance per outgoing quic tunnel, and TunnelManager handles all that glue now. Adds QUIC packet handling to get to the right tunnel code. This required multiplexing incoming quic packets, as follows: Adds a very small quic tunnel packet header of 4 bytes: [1, SPORT, ECN] for client->server packets, where SPORT is our source "port" (really: just a uint16_t unique quic instance identifier) or [2, DPORT, ECN] for server->client packets where the DPORT is the SPORT from above. (This also reworks ECN bits to get properly carried over lokinet.) We don't need a destination/source port for the server-side because there is only ever one quic server (and we know we're going to it when the first byte of the header is 1). Removes the config option for quic exposing ports; a full lokinet will simply accept anything incoming on quic and tunnel it to the requested port on the the local endpoint IP (this handler will come in a following commit). Replace ConvoTags with full addresses: we need to carry the port, as well, which the ConvoTag can't give us, so change those to more general SockAddrs from which we can extract both the ConvoTag *and* the port. Add a pending connection queue along with new quic-side handlers to call when a stream becomes available (TunnelManager uses this to wire up pending incoming conns with quic streams as streams open up). Completely get rid of tunnel_server/tunnel_client.cpp code; it is now moved to tunnel.hpp. Add listen()/forget() methods in TunnelManager for setting up quic listening sockets (for liblokinet usage). Add open()/close() methods in TunnelManager for spinning up new quic clients for outgoing quic connections.
2021-03-23 19:26:32 +00:00
public:
/// Called from tun code via TunnelManager to deliver an incoming packet to us.
///
/// \param src - the source address; this may be a tun interface address, or may be a fake IPv6
/// address based on the convo tag. The port is not used.
/// \param ecn - the packet ecn parameter
void
receive_packet(const SockAddr& src, uint8_t ecn, bstring_view data);
/// Returns a shared pointer to the uvw loop.
std::shared_ptr<uvw::Loop>
get_loop();
2021-03-10 15:11:42 +00:00
protected:
/// the service endpoint we are owned by
2021-03-26 13:16:43 +00:00
EndpointBase& service_endpoint;
2021-03-10 15:11:42 +00:00
QUIC lokinet integration refactor Refactors how quic packets get handled: the actual tunnels now live in tunnel.hpp's TunnelManager which holds and manages all the quic<->tcp tunnelling. service::Endpoint now holds a TunnelManager rather than a quic::Server. We only need one quic server, but we need a separate quic client instance per outgoing quic tunnel, and TunnelManager handles all that glue now. Adds QUIC packet handling to get to the right tunnel code. This required multiplexing incoming quic packets, as follows: Adds a very small quic tunnel packet header of 4 bytes: [1, SPORT, ECN] for client->server packets, where SPORT is our source "port" (really: just a uint16_t unique quic instance identifier) or [2, DPORT, ECN] for server->client packets where the DPORT is the SPORT from above. (This also reworks ECN bits to get properly carried over lokinet.) We don't need a destination/source port for the server-side because there is only ever one quic server (and we know we're going to it when the first byte of the header is 1). Removes the config option for quic exposing ports; a full lokinet will simply accept anything incoming on quic and tunnel it to the requested port on the the local endpoint IP (this handler will come in a following commit). Replace ConvoTags with full addresses: we need to carry the port, as well, which the ConvoTag can't give us, so change those to more general SockAddrs from which we can extract both the ConvoTag *and* the port. Add a pending connection queue along with new quic-side handlers to call when a stream becomes available (TunnelManager uses this to wire up pending incoming conns with quic streams as streams open up). Completely get rid of tunnel_server/tunnel_client.cpp code; it is now moved to tunnel.hpp. Add listen()/forget() methods in TunnelManager for setting up quic listening sockets (for liblokinet usage). Add open()/close() methods in TunnelManager for spinning up new quic clients for outgoing quic connections.
2021-03-23 19:26:32 +00:00
/// local "address" is the IPv6 unspecified address since we don't have (or care about) the
/// actual local address for building quic packets. The port of this address must be set to our
/// local pseudo-port, for clients, and 0 for a server.
Address local_addr{in6addr_any};
2021-03-10 15:11:42 +00:00
std::shared_ptr<uvw::TimerHandle> expiry_timer;
QUIC lokinet integration refactor Refactors how quic packets get handled: the actual tunnels now live in tunnel.hpp's TunnelManager which holds and manages all the quic<->tcp tunnelling. service::Endpoint now holds a TunnelManager rather than a quic::Server. We only need one quic server, but we need a separate quic client instance per outgoing quic tunnel, and TunnelManager handles all that glue now. Adds QUIC packet handling to get to the right tunnel code. This required multiplexing incoming quic packets, as follows: Adds a very small quic tunnel packet header of 4 bytes: [1, SPORT, ECN] for client->server packets, where SPORT is our source "port" (really: just a uint16_t unique quic instance identifier) or [2, DPORT, ECN] for server->client packets where the DPORT is the SPORT from above. (This also reworks ECN bits to get properly carried over lokinet.) We don't need a destination/source port for the server-side because there is only ever one quic server (and we know we're going to it when the first byte of the header is 1). Removes the config option for quic exposing ports; a full lokinet will simply accept anything incoming on quic and tunnel it to the requested port on the the local endpoint IP (this handler will come in a following commit). Replace ConvoTags with full addresses: we need to carry the port, as well, which the ConvoTag can't give us, so change those to more general SockAddrs from which we can extract both the ConvoTag *and* the port. Add a pending connection queue along with new quic-side handlers to call when a stream becomes available (TunnelManager uses this to wire up pending incoming conns with quic streams as streams open up). Completely get rid of tunnel_server/tunnel_client.cpp code; it is now moved to tunnel.hpp. Add listen()/forget() methods in TunnelManager for setting up quic listening sockets (for liblokinet usage). Add open()/close() methods in TunnelManager for spinning up new quic clients for outgoing quic connections.
2021-03-23 19:26:32 +00:00
std::vector<std::byte> buf;
2021-03-10 15:11:42 +00:00
// Max theoretical size of a UDP packet is 2^16-1 minus IP/UDP header overhead
static constexpr size_t max_buf_size = 64 * 1024;
// Max size of a UDP packet that we'll send
2021-09-24 21:15:50 +00:00
static constexpr size_t max_pkt_size_v4 = NGTCP2_MAX_UDP_PAYLOAD_SIZE;
static constexpr size_t max_pkt_size_v6 = NGTCP2_MAX_UDP_PAYLOAD_SIZE;
2021-03-10 15:11:42 +00:00
using primary_conn_ptr = std::shared_ptr<Connection>;
using alias_conn_ptr = std::weak_ptr<Connection>;
// Connections. When a client establishes a new connection it chooses its own source connection
// ID and a destination connection ID and sends them to the server.
//
// This container stores the primary Connection instance as a shared_ptr, and any connection
// aliases as weak_ptrs referencing the primary instance (so that we don't have to double a
// double-hash lookup on incoming packets, since those frequently use aliases).
//
// The destination connection ID should be entirely random and can be up to 160 bits, but the
// source connection ID does not have to be (i.e. it can encode some information, if desired).
//
// The server is going to include in the response:
// - destination connection ID equal to the client's source connection ID
// - a new random source connection ID. (We don't use the client's destination ID but generate
// our own). Like the clients source ID, this can contain embedded info.
//
// The client stores this, and so we end up with client-scid == server-dcid, and client-dcid ==
// server-scid, where each side chose its own source connection ID.
//
// Ultimately, we store here our own {source connection ID -> Connection} pairs (or
// equivalently, on incoming packets, the key will be the packet's dest conn ID).
std::unordered_map<ConnectionID, std::variant<primary_conn_ptr, alias_conn_ptr>> conns;
using conns_iterator = decltype(conns)::iterator;
// Connections that are draining (i.e. we are dropping, but need to keep around for a while
// to catch and drop lagged packets). The time point is the scheduled removal time.
std::queue<std::pair<ConnectionID, std::chrono::steady_clock::time_point>> draining;
NullCrypto null_crypto;
// Random data that we hash together with a CID to make a stateless reset token
std::array<std::byte, 32> static_secret;
friend class Connection;
2021-03-26 13:16:43 +00:00
explicit Endpoint(EndpointBase& service_endpoint_);
2021-03-10 15:11:42 +00:00
virtual ~Endpoint();
// Version & connection id info that we can potentially extract when decoding a packet
struct version_info
{
uint32_t version;
const uint8_t* dcid;
size_t dcid_len;
const uint8_t* scid;
size_t scid_len;
};
// Called to handle an incoming packet
QUIC lokinet integration refactor Refactors how quic packets get handled: the actual tunnels now live in tunnel.hpp's TunnelManager which holds and manages all the quic<->tcp tunnelling. service::Endpoint now holds a TunnelManager rather than a quic::Server. We only need one quic server, but we need a separate quic client instance per outgoing quic tunnel, and TunnelManager handles all that glue now. Adds QUIC packet handling to get to the right tunnel code. This required multiplexing incoming quic packets, as follows: Adds a very small quic tunnel packet header of 4 bytes: [1, SPORT, ECN] for client->server packets, where SPORT is our source "port" (really: just a uint16_t unique quic instance identifier) or [2, DPORT, ECN] for server->client packets where the DPORT is the SPORT from above. (This also reworks ECN bits to get properly carried over lokinet.) We don't need a destination/source port for the server-side because there is only ever one quic server (and we know we're going to it when the first byte of the header is 1). Removes the config option for quic exposing ports; a full lokinet will simply accept anything incoming on quic and tunnel it to the requested port on the the local endpoint IP (this handler will come in a following commit). Replace ConvoTags with full addresses: we need to carry the port, as well, which the ConvoTag can't give us, so change those to more general SockAddrs from which we can extract both the ConvoTag *and* the port. Add a pending connection queue along with new quic-side handlers to call when a stream becomes available (TunnelManager uses this to wire up pending incoming conns with quic streams as streams open up). Completely get rid of tunnel_server/tunnel_client.cpp code; it is now moved to tunnel.hpp. Add listen()/forget() methods in TunnelManager for setting up quic listening sockets (for liblokinet usage). Add open()/close() methods in TunnelManager for spinning up new quic clients for outgoing quic connections.
2021-03-23 19:26:32 +00:00
void
handle_packet(const Packet& p);
2021-03-10 15:11:42 +00:00
// Internal method: handles initial common packet decoding, returns the connection ID or nullopt
// if decoding failed.
std::optional<ConnectionID>
handle_packet_init(const Packet& p);
// Internal method: handles a packet sent to the given connection
void
handle_conn_packet(Connection& c, const Packet& p);
QUIC lokinet integration refactor Refactors how quic packets get handled: the actual tunnels now live in tunnel.hpp's TunnelManager which holds and manages all the quic<->tcp tunnelling. service::Endpoint now holds a TunnelManager rather than a quic::Server. We only need one quic server, but we need a separate quic client instance per outgoing quic tunnel, and TunnelManager handles all that glue now. Adds QUIC packet handling to get to the right tunnel code. This required multiplexing incoming quic packets, as follows: Adds a very small quic tunnel packet header of 4 bytes: [1, SPORT, ECN] for client->server packets, where SPORT is our source "port" (really: just a uint16_t unique quic instance identifier) or [2, DPORT, ECN] for server->client packets where the DPORT is the SPORT from above. (This also reworks ECN bits to get properly carried over lokinet.) We don't need a destination/source port for the server-side because there is only ever one quic server (and we know we're going to it when the first byte of the header is 1). Removes the config option for quic exposing ports; a full lokinet will simply accept anything incoming on quic and tunnel it to the requested port on the the local endpoint IP (this handler will come in a following commit). Replace ConvoTags with full addresses: we need to carry the port, as well, which the ConvoTag can't give us, so change those to more general SockAddrs from which we can extract both the ConvoTag *and* the port. Add a pending connection queue along with new quic-side handlers to call when a stream becomes available (TunnelManager uses this to wire up pending incoming conns with quic streams as streams open up). Completely get rid of tunnel_server/tunnel_client.cpp code; it is now moved to tunnel.hpp. Add listen()/forget() methods in TunnelManager for setting up quic listening sockets (for liblokinet usage). Add open()/close() methods in TunnelManager for spinning up new quic clients for outgoing quic connections.
2021-03-23 19:26:32 +00:00
// Accept a new incoming connection, i.e. pre-handshake. Returns a nullptr if the connection
// can't be created (e.g. because of invalid initial data), or if incoming connections are not
// accepted by this endpoint (i.e. because it is not a Server instance, or because there are no
// registered listen handlers). The base class default returns nullptr.
virtual std::shared_ptr<Connection>
accept_initial_connection(const Packet&)
{
return nullptr;
}
2021-03-10 15:11:42 +00:00
// Reads a packet and handles various error conditions. Returns an io_result. Note that it is
// possible for the conn_it to be erased from `conns` if the error code is anything other than
// success (0) or NGTCP2_ERR_RETRY.
io_result
read_packet(const Packet& p, Connection& conn);
QUIC lokinet integration refactor Refactors how quic packets get handled: the actual tunnels now live in tunnel.hpp's TunnelManager which holds and manages all the quic<->tcp tunnelling. service::Endpoint now holds a TunnelManager rather than a quic::Server. We only need one quic server, but we need a separate quic client instance per outgoing quic tunnel, and TunnelManager handles all that glue now. Adds QUIC packet handling to get to the right tunnel code. This required multiplexing incoming quic packets, as follows: Adds a very small quic tunnel packet header of 4 bytes: [1, SPORT, ECN] for client->server packets, where SPORT is our source "port" (really: just a uint16_t unique quic instance identifier) or [2, DPORT, ECN] for server->client packets where the DPORT is the SPORT from above. (This also reworks ECN bits to get properly carried over lokinet.) We don't need a destination/source port for the server-side because there is only ever one quic server (and we know we're going to it when the first byte of the header is 1). Removes the config option for quic exposing ports; a full lokinet will simply accept anything incoming on quic and tunnel it to the requested port on the the local endpoint IP (this handler will come in a following commit). Replace ConvoTags with full addresses: we need to carry the port, as well, which the ConvoTag can't give us, so change those to more general SockAddrs from which we can extract both the ConvoTag *and* the port. Add a pending connection queue along with new quic-side handlers to call when a stream becomes available (TunnelManager uses this to wire up pending incoming conns with quic streams as streams open up). Completely get rid of tunnel_server/tunnel_client.cpp code; it is now moved to tunnel.hpp. Add listen()/forget() methods in TunnelManager for setting up quic listening sockets (for liblokinet usage). Add open()/close() methods in TunnelManager for spinning up new quic clients for outgoing quic connections.
2021-03-23 19:26:32 +00:00
// Writes the lokinet packet header to the beginning of `buf_`; the header is prepend to quic
// packets to identify which quic server the packet should be delivered to and consists of:
// - type [1 byte]: 1 for client->server packets; 2 for server->client packets (other values
// reserved)
// - port [2 bytes, network order]: client pseudoport (i.e. either a source or destination port
// depending on type)
// - ecn value [1 byte]: provided by ngtcp2. (Only the lower 2 bits are actually used).
//
// \param psuedo_port - the remote's pseudo-port (will be 0 if the remote is a server, > 0 for
// a client remote)
// \param ecn - the ecn value from ngtcp2
//
// Returns the number of bytes written to buf_.
virtual size_t
write_packet_header(nuint16_t pseudo_port, uint8_t ecn) = 0;
2021-03-10 15:11:42 +00:00
// Sends a packet to `to` containing `data`. Returns a non-error io_result on success,
// an io_result with .error_code set to the errno of the failure on failure.
io_result
QUIC lokinet integration refactor Refactors how quic packets get handled: the actual tunnels now live in tunnel.hpp's TunnelManager which holds and manages all the quic<->tcp tunnelling. service::Endpoint now holds a TunnelManager rather than a quic::Server. We only need one quic server, but we need a separate quic client instance per outgoing quic tunnel, and TunnelManager handles all that glue now. Adds QUIC packet handling to get to the right tunnel code. This required multiplexing incoming quic packets, as follows: Adds a very small quic tunnel packet header of 4 bytes: [1, SPORT, ECN] for client->server packets, where SPORT is our source "port" (really: just a uint16_t unique quic instance identifier) or [2, DPORT, ECN] for server->client packets where the DPORT is the SPORT from above. (This also reworks ECN bits to get properly carried over lokinet.) We don't need a destination/source port for the server-side because there is only ever one quic server (and we know we're going to it when the first byte of the header is 1). Removes the config option for quic exposing ports; a full lokinet will simply accept anything incoming on quic and tunnel it to the requested port on the the local endpoint IP (this handler will come in a following commit). Replace ConvoTags with full addresses: we need to carry the port, as well, which the ConvoTag can't give us, so change those to more general SockAddrs from which we can extract both the ConvoTag *and* the port. Add a pending connection queue along with new quic-side handlers to call when a stream becomes available (TunnelManager uses this to wire up pending incoming conns with quic streams as streams open up). Completely get rid of tunnel_server/tunnel_client.cpp code; it is now moved to tunnel.hpp. Add listen()/forget() methods in TunnelManager for setting up quic listening sockets (for liblokinet usage). Add open()/close() methods in TunnelManager for spinning up new quic clients for outgoing quic connections.
2021-03-23 19:26:32 +00:00
send_packet(const Address& to, bstring_view data, uint8_t ecn);
2021-03-10 15:11:42 +00:00
// Wrapper around the above that takes a regular std::string_view (i.e. of chars) and recasts
// it to an string_view of std::bytes.
io_result
QUIC lokinet integration refactor Refactors how quic packets get handled: the actual tunnels now live in tunnel.hpp's TunnelManager which holds and manages all the quic<->tcp tunnelling. service::Endpoint now holds a TunnelManager rather than a quic::Server. We only need one quic server, but we need a separate quic client instance per outgoing quic tunnel, and TunnelManager handles all that glue now. Adds QUIC packet handling to get to the right tunnel code. This required multiplexing incoming quic packets, as follows: Adds a very small quic tunnel packet header of 4 bytes: [1, SPORT, ECN] for client->server packets, where SPORT is our source "port" (really: just a uint16_t unique quic instance identifier) or [2, DPORT, ECN] for server->client packets where the DPORT is the SPORT from above. (This also reworks ECN bits to get properly carried over lokinet.) We don't need a destination/source port for the server-side because there is only ever one quic server (and we know we're going to it when the first byte of the header is 1). Removes the config option for quic exposing ports; a full lokinet will simply accept anything incoming on quic and tunnel it to the requested port on the the local endpoint IP (this handler will come in a following commit). Replace ConvoTags with full addresses: we need to carry the port, as well, which the ConvoTag can't give us, so change those to more general SockAddrs from which we can extract both the ConvoTag *and* the port. Add a pending connection queue along with new quic-side handlers to call when a stream becomes available (TunnelManager uses this to wire up pending incoming conns with quic streams as streams open up). Completely get rid of tunnel_server/tunnel_client.cpp code; it is now moved to tunnel.hpp. Add listen()/forget() methods in TunnelManager for setting up quic listening sockets (for liblokinet usage). Add open()/close() methods in TunnelManager for spinning up new quic clients for outgoing quic connections.
2021-03-23 19:26:32 +00:00
send_packet(const Address& to, std::string_view data, uint8_t ecn)
2021-03-10 15:11:42 +00:00
{
return send_packet(
to, bstring_view{reinterpret_cast<const std::byte*>(data.data()), data.size()}, ecn);
}
// Another wrapper taking a vector
io_result
QUIC lokinet integration refactor Refactors how quic packets get handled: the actual tunnels now live in tunnel.hpp's TunnelManager which holds and manages all the quic<->tcp tunnelling. service::Endpoint now holds a TunnelManager rather than a quic::Server. We only need one quic server, but we need a separate quic client instance per outgoing quic tunnel, and TunnelManager handles all that glue now. Adds QUIC packet handling to get to the right tunnel code. This required multiplexing incoming quic packets, as follows: Adds a very small quic tunnel packet header of 4 bytes: [1, SPORT, ECN] for client->server packets, where SPORT is our source "port" (really: just a uint16_t unique quic instance identifier) or [2, DPORT, ECN] for server->client packets where the DPORT is the SPORT from above. (This also reworks ECN bits to get properly carried over lokinet.) We don't need a destination/source port for the server-side because there is only ever one quic server (and we know we're going to it when the first byte of the header is 1). Removes the config option for quic exposing ports; a full lokinet will simply accept anything incoming on quic and tunnel it to the requested port on the the local endpoint IP (this handler will come in a following commit). Replace ConvoTags with full addresses: we need to carry the port, as well, which the ConvoTag can't give us, so change those to more general SockAddrs from which we can extract both the ConvoTag *and* the port. Add a pending connection queue along with new quic-side handlers to call when a stream becomes available (TunnelManager uses this to wire up pending incoming conns with quic streams as streams open up). Completely get rid of tunnel_server/tunnel_client.cpp code; it is now moved to tunnel.hpp. Add listen()/forget() methods in TunnelManager for setting up quic listening sockets (for liblokinet usage). Add open()/close() methods in TunnelManager for spinning up new quic clients for outgoing quic connections.
2021-03-23 19:26:32 +00:00
send_packet(const Address& to, const std::vector<std::byte>& data, uint8_t ecn)
2021-03-10 15:11:42 +00:00
{
return send_packet(to, bstring_view{data.data(), data.size()}, ecn);
}
void
send_version_negotiation(const version_info& vi, const Address& source);
// Looks up a connection. Returns a shared_ptr (either copied for a primary connection, or
// locked from an alias's weak pointer) if the connection was found or nullptr if not; and a
// bool indicating whether this connection ID was an alias (true) or not (false). [Note: the
// alias value can be true even if the shared_ptr is null in the case of an expired alias that
// hasn't yet been cleaned up].
std::pair<std::shared_ptr<Connection>, bool>
get_conn(const ConnectionID& cid);
// Called to start closing (or continue closing) a connection by sending a connection close
// response to any incoming packets.
//
// Takes the iterator to the connection pair from `conns` and optional error parameters: if
// `application` is false (the default) then we do a hard connection close because of transport
// error, if true we do a graceful application close. `close_reason` can be provided for
// propagating reason for close to remote, defaults to empty string. For application closes the
// code is application-defined; for hard closes the code should be one of the NGTCP2_*_ERROR
// values.
2021-03-10 15:11:42 +00:00
void
close_connection(
Connection& conn,
uint64_t code = NGTCP2_NO_ERROR,
bool application = false,
std::string_view close_reason = ""sv);
2021-03-10 15:11:42 +00:00
/// Puts a connection into draining mode (i.e. after getting a connection close). This will
/// keep the connection registered for the recommended 3*Probe Timeout, during which we drop
/// packets that use the connection id and after which we will forget about it.
void
start_draining(Connection& conn);
void
check_timeouts();
/// Deletes a connection from `conns`; if the connecion is a primary connection shared pointer
/// then it is removed and clean_alias_conns() is immediately called to remove any aliases to
/// the connection. If the given connection is an alias connection then it is removed but no
/// cleanup is performed. Returns true if something was removed, false if the connection was
/// not found.
bool
delete_conn(const ConnectionID& cid);
/// Removes any connection id aliases that no longer have associated Connections.
void
clean_alias_conns();
/// Creates a new, unused connection ID alias for the given connection; adds the alias to
/// `conns` and returns the ConnectionID.
ConnectionID
add_connection_id(Connection& conn, size_t cid_length = ConnectionID::max_size());
public:
// Makes a deterministic stateless reset token for the given connection ID. Writes it to dest
// (which must have NGTCP2_STATELESS_RESET_TOKENLEN bytes available).
void
make_stateless_reset_token(const ConnectionID& cid, unsigned char* dest);
// Default stream buffer size for streams opened through this endpoint.
size_t default_stream_buffer_size = 64 * 1024;
QUIC lokinet integration refactor Refactors how quic packets get handled: the actual tunnels now live in tunnel.hpp's TunnelManager which holds and manages all the quic<->tcp tunnelling. service::Endpoint now holds a TunnelManager rather than a quic::Server. We only need one quic server, but we need a separate quic client instance per outgoing quic tunnel, and TunnelManager handles all that glue now. Adds QUIC packet handling to get to the right tunnel code. This required multiplexing incoming quic packets, as follows: Adds a very small quic tunnel packet header of 4 bytes: [1, SPORT, ECN] for client->server packets, where SPORT is our source "port" (really: just a uint16_t unique quic instance identifier) or [2, DPORT, ECN] for server->client packets where the DPORT is the SPORT from above. (This also reworks ECN bits to get properly carried over lokinet.) We don't need a destination/source port for the server-side because there is only ever one quic server (and we know we're going to it when the first byte of the header is 1). Removes the config option for quic exposing ports; a full lokinet will simply accept anything incoming on quic and tunnel it to the requested port on the the local endpoint IP (this handler will come in a following commit). Replace ConvoTags with full addresses: we need to carry the port, as well, which the ConvoTag can't give us, so change those to more general SockAddrs from which we can extract both the ConvoTag *and* the port. Add a pending connection queue along with new quic-side handlers to call when a stream becomes available (TunnelManager uses this to wire up pending incoming conns with quic streams as streams open up). Completely get rid of tunnel_server/tunnel_client.cpp code; it is now moved to tunnel.hpp. Add listen()/forget() methods in TunnelManager for setting up quic listening sockets (for liblokinet usage). Add open()/close() methods in TunnelManager for spinning up new quic clients for outgoing quic connections.
2021-03-23 19:26:32 +00:00
// Packet buffer we use when constructing custom packets to fire over lokinet
std::array<std::byte, net::IPPacket::MaxSize> buf_;
// Non-copyable, non-movable
Endpoint(const Endpoint&) = delete;
Endpoint(Endpoint&&) = delete;
2021-03-10 15:11:42 +00:00
};
} // namespace llarp::quic