Create the interfaces for PACKET_MMAP endpoints to implement.

PiperOrigin-RevId: 715019134

pkg/abi/linux/socket.go

@@ -144,6 +144,72 @@ const (
 	PACKET_OUTGOING  = 4 // Outgoing of any type
 )
 
+// Packet socket options from <linux/if_packet.h>
+const (
+	PACKET_RX_RING = 5
+)
+
+// Statuses for a frame in a packet_mmap ring buffer from <linux/if_packet.h>.
+const (
+	TP_STATUS_KERNEL          = 0
+	TP_STATUS_USER            = 0x1
+	TP_STATUS_COPY            = 0x2
+	TP_STATUS_LOSING          = 0x4
+	TP_STATUS_CSUM_NOT_READY  = 0x8
+	TP_STATUS_VLAN_VALID      = 0x10
+	TP_STATUS_BLK_TMO         = 0x20
+	TP_STATUS_VLAN_TPID_VALID = 0x40
+	TP_STATUS_CSUM_VALID      = 0x80
+	TP_STATUS_GSO_TCP         = 0x100
+)
+
+// TpacketReq is the request for a packet_mmap ring buffer from
+// <linux/if_packet.h>.
+//
+// +marshal
+type TpacketReq struct {
+	TpBlockSize uint32
+	TpBlockNr   uint32
+	TpFrameSize uint32
+	TpFrameNr   uint32
+}
+
+// TpacketHdr is the header for a frame in a packet_mmap ring buffer from
+// <linux/if_packet.h>.
+//
+// +marshal
+type TpacketHdr struct {
+	TpStatus  uint64
+	TpLen     uint32
+	TpSnaplen uint32
+	TpMac     uint16
+	TpNet     uint16
+	TpSec     uint32
+	TpUsec    uint32 `marshal:"unaligned"`
+}
+
+// TpacketAlignment is the alignment of a frame in a packet_mmap ring buffer
+// from <linux/if_packet.h>.
+const (
+	TPACKET_ALIGNMENT = 16
+)
+
+// TPACKET_V1 is the version of a packet_mmap ring buffer from
+// <linux/if_packet.h> that is implemented in gVisor.
+const (
+	TPACKET_V1 = iota
+)
+
+// TPACKET_HDRLEN is the length of a TpacketHdr from <linux/if_packet.h>.
+var (
+	TPACKET_HDRLEN = uint32((*TpacketHdr)(nil).SizeBytes()) + uint32((*SockAddrLink)(nil).SizeBytes())
+)
+
+// TPacketAlign aligns a value to the alignment of a TPacket.
+func TPacketAlign(x uint32) uint32 {
+	return (x + TPACKET_ALIGNMENT - 1) &^ (TPACKET_ALIGNMENT - 1)
+}
+
 // Socket options from socket.h.
 const (
 	SO_DEBUG                 = 1

pkg/sentry/socket/netstack/BUILD

@@ -41,6 +41,7 @@ go_library(
         "//pkg/sentry/kernel",
         "//pkg/sentry/kernel/auth",
         "//pkg/sentry/ktime",
+        "//pkg/sentry/memmap",
         "//pkg/sentry/socket",
         "//pkg/sentry/socket/netfilter",
         "//pkg/sentry/socket/netlink/nlmsg",

pkg/sentry/socket/netstack/netstack.go

@@ -51,6 +51,7 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/kernel"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
 	"gvisor.dev/gvisor/pkg/sentry/ktime"
+	"gvisor.dev/gvisor/pkg/sentry/memmap"
 	"gvisor.dev/gvisor/pkg/sentry/socket"
 	"gvisor.dev/gvisor/pkg/sentry/socket/netfilter"
 	epb "gvisor.dev/gvisor/pkg/sentry/socket/netstack/events_go_proto"
@@ -318,6 +319,28 @@ const sizeOfInt32 int = 4
 
 var errStackType = syserr.New("expected but did not receive a netstack.Stack", errno.EINVAL)
 
+// mappableEndpoint is the interface implemented by endpoints that support
+// mmap.
+type mappableEndpoint interface {
+	// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap.
+	ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error
+
+	// AddMapping implements memmap.Mappable.AddMapping.
+	AddMapping(context.Context, memmap.MappingSpace, hostarch.AddrRange, uint64, bool) error
+
+	// RemoveMapping implements memmap.Mappable.RemoveMapping.
+	RemoveMapping(context.Context, memmap.MappingSpace, hostarch.AddrRange, uint64, bool)
+
+	// CopyMapping implements memmap.Mappable.CopyMapping.
+	CopyMapping(context.Context, memmap.MappingSpace, hostarch.AddrRange, hostarch.AddrRange, uint64, bool) error
+
+	// Translate implements memmap.Mappable.Translate.
+	Translate(context.Context, memmap.MappableRange, memmap.MappableRange, hostarch.AccessType) ([]memmap.Translation, error)
+
+	// InvalidateUnsavable implements memmap.Mappable.InvalidateUnsavable.
+	InvalidateUnsavable(context.Context) error
+}
+
 // commonEndpoint represents the intersection of a tcpip.Endpoint and a
 // transport.Endpoint.
 type commonEndpoint interface {

pkg/syserr/netstack.go

@@ -55,6 +55,7 @@ var (
 	ErrMulticastInputCannotBeOutput = New((&tcpip.ErrMulticastInputCannotBeOutput{}).String(), errno.EINVAL)
 	ErrMissingRequiredFields        = New((&tcpip.ErrMissingRequiredFields{}).String(), errno.EINVAL)
 	ErrNoNet                        = New((&tcpip.ErrNoNet{}).String(), errno.ENONET)
+	ErrEndpointBusy                 = New((&tcpip.ErrEndpointBusy{}).String(), errno.EBUSY)
 )
 
 // TranslateNetstackError converts an error from the tcpip package to a sentry
@@ -149,6 +150,8 @@ func TranslateNetstackError(err tcpip.Error) *Error {
 		return ErrMulticastInputCannotBeOutput
 	case *tcpip.ErrMissingRequiredFields:
 		return ErrMissingRequiredFields
+	case *tcpip.ErrEndpointBusy:
+		return ErrEndpointBusy
 	default:
 		panic(fmt.Sprintf("unknown error %T", err))
 	}

pkg/tcpip/errors.go

@@ -620,4 +620,22 @@ func (*ErrMulticastInputCannotBeOutput) IgnoreStats() bool {
 }
 func (*ErrMulticastInputCannotBeOutput) String() string { return "output cannot contain input" }
 
+// ErrEndpointBusy indicates that the operation cannot be completed because the
+// endpoint is busy.
+//
+// +stateify savable
+type ErrEndpointBusy struct{}
+
+// isError implements Error.
+func (*ErrEndpointBusy) isError() {}
+
+// IgnoreStats implements Error.
+func (*ErrEndpointBusy) IgnoreStats() bool {
+	return true
+}
+
+func (*ErrEndpointBusy) String() string {
+	return "operation cannot be completed because the endpoint is busy"
+}
+
 // LINT.ThenChange(../syserr/netstack.go)

pkg/tcpip/stack/registration.go

@@ -171,6 +171,58 @@ type PacketEndpoint interface {
 	HandlePacket(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, pkt *PacketBuffer)
 }
 
+// MappablePacketEndpoint is a packet endpoint that supports forwarding its
+// packets to a PacketMMapEndpoint.
+type MappablePacketEndpoint interface {
+	PacketEndpoint
+
+	// ConfigurePacketMMap configures the endpoint to use a memory mapped endpoint
+	// handler ep. All packets sent or received via this endpoint will be handled
+	// by the PacketMMapEndpoint.
+	ConfigurePacketMMap(ep PacketMMapEndpoint)
+}
+
+// PacketMMapCopyHandler is a function that is called when a packet received is
+// too large for the buffer size specified for the memory mapped endpoint. In
+// this case, the packet is copied and passed to the PacketMMapCopyHandler.
+type PacketMMapCopyHandler func(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, pkt *PacketBuffer)
+
+// InitPacketMMapOpts are the options for initializing a PacketMMapEndpoint.
+type InitPacketMMapOpts struct {
+	RxReq, TxReq *tcpip.TpacketReq
+	Cooked       bool
+	Stack        *Stack
+	Stats        *tcpip.TransportEndpointStats
+	Wq           *waiter.Queue
+	CopyHandler  PacketMMapCopyHandler
+}
+
+// PacketMMapEndpoint is the interface implemented by endpoints to handle memory
+// mapped packets over the packet transport protocol (PACKET_MMAP).
+type PacketMMapEndpoint interface {
+	// HandlePacket is called by the stack when new packets arrive that
+	// match the endpoint.
+	//
+	// Implementers should treat packet as immutable and should copy it
+	// before before modification.
+	//
+	// linkHeader may have a length of 0, in which case the PacketEndpoint
+	// should construct its own ethernet header for applications.
+	//
+	// HandlePacket may modify pkt.
+	HandlePacket(nicID tcpip.NICID, netProto tcpip.NetworkProtocolNumber, pkt *PacketBuffer)
+
+	// Close releases any resources associated with the endpoint.
+	Close()
+
+	// Readiness returns the events that the endpoint is ready for.
+	Readiness(mask waiter.EventMask) waiter.EventMask
+
+	// InitPacketMMap initializes the endpoint. It must be called before any other
+	// methods.
+	InitPacketMMap(opts InitPacketMMapOpts) tcpip.Error
+}
+
 // UnknownDestinationPacketDisposition enumerates the possible return values from
 // HandleUnknownDestinationPacket().
 type UnknownDestinationPacketDisposition int

pkg/tcpip/tcpip.go

@@ -1185,6 +1185,19 @@ func (*ICMPv6Filter) isGettableSocketOption() {}
 
 func (*ICMPv6Filter) isSettableSocketOption() {}
 
+// TpacketReq is the tpacket_req structure as described in
+// https://www.kernel.org/doc/Documentation/networking/packet_mmap.txt
+//
+// +stateify savable
+type TpacketReq struct {
+	TpBlockSize uint32
+	TpBlockNr   uint32
+	TpFrameSize uint32
+	TpFrameNr   uint32
+}
+
+func (*TpacketReq) isSettableSocketOption() {}
+
 // EndpointState represents the state of an endpoint.
 type EndpointState uint8