// Copyright 2018 The gVisor Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

// +build linux

// Package fdbased provides the implemention of data-link layer endpoints
// backed by boundary-preserving file descriptors (e.g., TUN devices,
// seqpacket/datagram sockets).
//
// FD based endpoints can be used in the networking stack by calling New() to
// create a new endpoint, and then passing it as an argument to
// Stack.CreateNIC().
package fdbased

import (
	"fmt"
	"syscall"

	"gvisor.googlesource.com/gvisor/pkg/tcpip"
	"gvisor.googlesource.com/gvisor/pkg/tcpip/buffer"
	"gvisor.googlesource.com/gvisor/pkg/tcpip/header"
	"gvisor.googlesource.com/gvisor/pkg/tcpip/link/rawfile"
	"gvisor.googlesource.com/gvisor/pkg/tcpip/stack"
)

// linkDispatcher reads packets from the link FD and dispatches them to the
// NetworkDispatcher.
type linkDispatcher interface {
	dispatch() (bool, *tcpip.Error)
}

// PacketDispatchMode are the various supported methods of receiving and
// dispatching packets from the underlying FD.
type PacketDispatchMode int

const (
	// Readv is the default dispatch mode and is the least performant of the
	// dispatch options but the one that is supported by all underlying FD
	// types.
	Readv PacketDispatchMode = iota
	// RecvMMsg enables use of recvmmsg() syscall instead of readv() to
	// read inbound packets. This reduces # of syscalls needed to process
	// packets.
	//
	// NOTE: recvmmsg() is only supported for sockets, so if the underlying
	// FD is not a socket then the code will still fall back to the readv()
	// path.
	RecvMMsg
	// PacketMMap enables use of PACKET_RX_RING to receive packets from the
	// NIC. PacketMMap requires that the underlying FD be an AF_PACKET. The
	// primary use-case for this is runsc which uses an AF_PACKET FD to
	// receive packets from the veth device.
	PacketMMap
)

type endpoint struct {
	// fd is the file descriptor used to send and receive packets.
	fd int

	// mtu (maximum transmission unit) is the maximum size of a packet.
	mtu uint32

	// hdrSize specifies the link-layer header size. If set to 0, no header
	// is added/removed; otherwise an ethernet header is used.
	hdrSize int

	// addr is the address of the endpoint.
	addr tcpip.LinkAddress

	// caps holds the endpoint capabilities.
	caps stack.LinkEndpointCapabilities

	// closed is a function to be called when the FD's peer (if any) closes
	// its end of the communication pipe.
	closed func(*tcpip.Error)

	inboundDispatcher linkDispatcher
	dispatcher        stack.NetworkDispatcher

	// packetDispatchMode controls the packet dispatcher used by this
	// endpoint.
	packetDispatchMode PacketDispatchMode

	// gsoMaxSize is the maximum GSO packet size. It is zero if GSO is
	// disabled.
	gsoMaxSize uint32
}

// Options specify the details about the fd-based endpoint to be created.
type Options struct {
	FD                 int
	MTU                uint32
	EthernetHeader     bool
	ClosedFunc         func(*tcpip.Error)
	Address            tcpip.LinkAddress
	SaveRestore        bool
	DisconnectOk       bool
	GSOMaxSize         uint32
	PacketDispatchMode PacketDispatchMode
	TXChecksumOffload  bool
	RXChecksumOffload  bool
}

// New creates a new fd-based endpoint.
//
// Makes fd non-blocking, but does not take ownership of fd, which must remain
// open for the lifetime of the returned endpoint.
func New(opts *Options) (tcpip.LinkEndpointID, error) {
	if err := syscall.SetNonblock(opts.FD, true); err != nil {
		return 0, fmt.Errorf("syscall.SetNonblock(%v) failed: %v", opts.FD, err)
	}

	caps := stack.LinkEndpointCapabilities(0)
	if opts.RXChecksumOffload {
		caps |= stack.CapabilityRXChecksumOffload
	}

	if opts.TXChecksumOffload {
		caps |= stack.CapabilityTXChecksumOffload
	}

	hdrSize := 0
	if opts.EthernetHeader {
		hdrSize = header.EthernetMinimumSize
		caps |= stack.CapabilityResolutionRequired
	}

	if opts.SaveRestore {
		caps |= stack.CapabilitySaveRestore
	}

	if opts.DisconnectOk {
		caps |= stack.CapabilityDisconnectOk
	}

	e := &endpoint{
		fd:                 opts.FD,
		mtu:                opts.MTU,
		caps:               caps,
		closed:             opts.ClosedFunc,
		addr:               opts.Address,
		hdrSize:            hdrSize,
		packetDispatchMode: opts.PacketDispatchMode,
	}

	isSocket, err := isSocketFD(e.fd)
	if err != nil {
		return 0, err
	}
	if isSocket {
		if opts.GSOMaxSize != 0 {
			e.caps |= stack.CapabilityGSO
			e.gsoMaxSize = opts.GSOMaxSize
		}
	}
	e.inboundDispatcher, err = createInboundDispatcher(e, isSocket)
	if err != nil {
		return 0, fmt.Errorf("createInboundDispatcher(...) = %v", err)
	}

	return stack.RegisterLinkEndpoint(e), nil
}

func createInboundDispatcher(e *endpoint, isSocket bool) (linkDispatcher, error) {
	// By default use the readv() dispatcher as it works with all kinds of
	// FDs (tap/tun/unix domain sockets and af_packet).
	inboundDispatcher, err := newReadVDispatcher(e.fd, e)
	if err != nil {
		return nil, fmt.Errorf("newReadVDispatcher(%d, %+v) = %v", e.fd, e, err)
	}

	if isSocket {
		switch e.packetDispatchMode {
		case PacketMMap:
			inboundDispatcher, err = newPacketMMapDispatcher(e.fd, e)
			if err != nil {
				return nil, fmt.Errorf("newPacketMMapDispatcher(%d, %+v) = %v", e.fd, e, err)
			}
		case RecvMMsg:
			// If the provided FD is a socket then we optimize
			// packet reads by using recvmmsg() instead of read() to
			// read packets in a batch.
			inboundDispatcher, err = newRecvMMsgDispatcher(e.fd, e)
			if err != nil {
				return nil, fmt.Errorf("newRecvMMsgDispatcher(%d, %+v) = %v", e.fd, e, err)
			}
		}
	}
	return inboundDispatcher, nil
}

func isSocketFD(fd int) (bool, error) {
	var stat syscall.Stat_t
	if err := syscall.Fstat(fd, &stat); err != nil {
		return false, fmt.Errorf("syscall.Fstat(%v,...) failed: %v", fd, err)
	}
	return (stat.Mode & syscall.S_IFSOCK) == syscall.S_IFSOCK, nil
}

// Attach launches the goroutine that reads packets from the file descriptor and
// dispatches them via the provided dispatcher.
func (e *endpoint) Attach(dispatcher stack.NetworkDispatcher) {
	e.dispatcher = dispatcher
	// Link endpoints are not savable. When transportation endpoints are
	// saved, they stop sending outgoing packets and all incoming packets
	// are rejected.
	go e.dispatchLoop() // S/R-SAFE: See above.
}

// IsAttached implements stack.LinkEndpoint.IsAttached.
func (e *endpoint) IsAttached() bool {
	return e.dispatcher != nil
}

// MTU implements stack.LinkEndpoint.MTU. It returns the value initialized
// during construction.
func (e *endpoint) MTU() uint32 {
	return e.mtu
}

// Capabilities implements stack.LinkEndpoint.Capabilities.
func (e *endpoint) Capabilities() stack.LinkEndpointCapabilities {
	return e.caps
}

// MaxHeaderLength returns the maximum size of the link-layer header.
func (e *endpoint) MaxHeaderLength() uint16 {
	return uint16(e.hdrSize)
}

// LinkAddress returns the link address of this endpoint.
func (e *endpoint) LinkAddress() tcpip.LinkAddress {
	return e.addr
}

// virtioNetHdr is declared in linux/virtio_net.h.
type virtioNetHdr struct {
	flags      uint8
	gsoType    uint8
	hdrLen     uint16
	gsoSize    uint16
	csumStart  uint16
	csumOffset uint16
}

// These constants are declared in linux/virtio_net.h.
const (
	_VIRTIO_NET_HDR_F_NEEDS_CSUM = 1

	_VIRTIO_NET_HDR_GSO_TCPV4 = 1
	_VIRTIO_NET_HDR_GSO_TCPV6 = 4
)

// WritePacket writes outbound packets to the file descriptor. If it is not
// currently writable, the packet is dropped.
func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, hdr buffer.Prependable, payload buffer.VectorisedView, protocol tcpip.NetworkProtocolNumber) *tcpip.Error {
	if e.hdrSize > 0 {
		// Add ethernet header if needed.
		eth := header.Ethernet(hdr.Prepend(header.EthernetMinimumSize))
		ethHdr := &header.EthernetFields{
			DstAddr: r.RemoteLinkAddress,
			Type:    protocol,
		}

		// Preserve the src address if it's set in the route.
		if r.LocalLinkAddress != "" {
			ethHdr.SrcAddr = r.LocalLinkAddress
		} else {
			ethHdr.SrcAddr = e.addr
		}
		eth.Encode(ethHdr)
	}

	if e.Capabilities()&stack.CapabilityGSO != 0 {
		vnetHdr := virtioNetHdr{}
		vnetHdrBuf := vnetHdrToByteSlice(&vnetHdr)
		if gso != nil {
			vnetHdr.hdrLen = uint16(hdr.UsedLength())
			if gso.NeedsCsum {
				vnetHdr.flags = _VIRTIO_NET_HDR_F_NEEDS_CSUM
				vnetHdr.csumStart = header.EthernetMinimumSize + gso.L3HdrLen
				vnetHdr.csumOffset = gso.CsumOffset
			}
			if gso.Type != stack.GSONone && uint16(payload.Size()) > gso.MSS {
				switch gso.Type {
				case stack.GSOTCPv4:
					vnetHdr.gsoType = _VIRTIO_NET_HDR_GSO_TCPV4
				case stack.GSOTCPv6:
					vnetHdr.gsoType = _VIRTIO_NET_HDR_GSO_TCPV6
				default:
					panic(fmt.Sprintf("Unknown gso type: %v", gso.Type))
				}
				vnetHdr.gsoSize = gso.MSS
			}
		}

		return rawfile.NonBlockingWrite3(e.fd, vnetHdrBuf, hdr.View(), payload.ToView())
	}

	if payload.Size() == 0 {
		return rawfile.NonBlockingWrite(e.fd, hdr.View())
	}

	return rawfile.NonBlockingWrite3(e.fd, hdr.View(), payload.ToView(), nil)
}

// WriteRawPacket writes a raw packet directly to the file descriptor.
func (e *endpoint) WriteRawPacket(dest tcpip.Address, packet []byte) *tcpip.Error {
	return rawfile.NonBlockingWrite(e.fd, packet)
}

// dispatchLoop reads packets from the file descriptor in a loop and dispatches
// them to the network stack.
func (e *endpoint) dispatchLoop() *tcpip.Error {
	for {
		cont, err := e.inboundDispatcher.dispatch()
		if err != nil || !cont {
			if e.closed != nil {
				e.closed(err)
			}
			return err
		}
	}
}

// GSOMaxSize returns the maximum GSO packet size.
func (e *endpoint) GSOMaxSize() uint32 {
	return e.gsoMaxSize
}

// InjectableEndpoint is an injectable fd-based endpoint. The endpoint writes
// to the FD, but does not read from it. All reads come from injected packets.
type InjectableEndpoint struct {
	endpoint

	dispatcher stack.NetworkDispatcher
}

// Attach saves the stack network-layer dispatcher for use later when packets
// are injected.
func (e *InjectableEndpoint) Attach(dispatcher stack.NetworkDispatcher) {
	e.dispatcher = dispatcher
}

// Inject injects an inbound packet.
func (e *InjectableEndpoint) Inject(protocol tcpip.NetworkProtocolNumber, vv buffer.VectorisedView) {
	e.dispatcher.DeliverNetworkPacket(e, "" /* remote */, "" /* local */, protocol, vv)
}

// NewInjectable creates a new fd-based InjectableEndpoint.
func NewInjectable(fd int, mtu uint32, capabilities stack.LinkEndpointCapabilities) (tcpip.LinkEndpointID, *InjectableEndpoint) {
	syscall.SetNonblock(fd, true)

	e := &InjectableEndpoint{endpoint: endpoint{
		fd:   fd,
		mtu:  mtu,
		caps: capabilities,
	}}

	return stack.RegisterLinkEndpoint(e), e
}