ebpf

package
v0.0.0-...-914b764 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Apr 29, 2020 License: Apache-2.0, Apache-2.0 Imports: 24 Imported by: 0

README

Note

This package is a fork of the weaveworks tcptracer-bpf package which focused on tracing TCP state events (connect, accept, close) without kernel specific runtime dependencies.

This fork adds support for UDP, as well as collection of metrics like bytes sent/received. It also opts for event collection via polling (using BPF maps) instead of being pushed event updates via perf buffers.

tracer-bpf

tracer-bpf is an eBPF program using kprobes to trace TCP/UDP events (connect, accept, close, send_msg, recv_msg).

The eBPF program is compiled to an ELF object file.

tracer-bpf also provides a Go library that provides a simple API for loading the ELF object file. Internally, it is using the gobpf elf package.

tracer-bpf does not have any run-time dependencies on kernel headers and is not tied to a specific kernel version or kernel configuration. This is quite unusual for eBPF programs using kprobes: for example, eBPF programs using kprobes with bcc are compiled on the fly and depend on kernel headers. And perf tools compiled for one kernel version cannot be used on another kernel version.

To adapt to the currently running kernel at run-time, tracer-bpf creates a series of TCP connections with known parameters (such as known IP addresses and ports) and discovers where those parameters are stored in the kernel struct sock. The offsets of the struct sock fields vary depending on the kernel version and kernel configuration. Since an eBPF programs cannot loop, tracer-bpf does not directly iterate over the possible offsets. It is instead controlled from userspace by the Go library using a state machine.

Development

make nettop will run a small testing program which periodically prints statistics about TCP/UDP traffic.

Documentation

Index

Constants

View Source 
const (
	// DEBUGCLIENT is the ClientID for debugging
	DEBUGCLIENT = "-1"
)
View Source 
const KprobeProfile = "/sys/kernel/debug/tracing/kprobe_profile"

KprobeProfile is the default path to the kprobe_profile file

Variables

View Source 
var (
	// ErrNotImplemented will be returned on non-linux environments like Windows and Mac OSX
	ErrNotImplemented = errors.New("BPF-based system probe not implemented on non-linux systems")
)

Functions

func Asset 

func Asset(name string) ([]byte, error)

Asset loads and returns the asset for the given name. It returns an error if the asset could not be found or could not be loaded.

func AssetDir 

func AssetDir(name string) ([]string, error)

AssetDir returns the file names below a certain directory embedded in the file by go-bindata. For example if you run go-bindata on data/... and data contains the following hierarchy: 

data/
  foo.txt
  img/
    a.png
    b.png

then AssetDir("data") would return []string{"foo.txt", "img"} AssetDir("data/img") would return []string{"a.png", "b.png"} AssetDir("foo.txt") and AssetDir("notexist") would return an error AssetDir("") will return []string{"data"}.

func AssetInfo 

func AssetInfo(name string) (os.FileInfo, error)

AssetInfo loads and returns the asset info for the given name. It returns an error if the asset could not be found or could not be loaded.

func AssetNames 

func AssetNames() []string

AssetNames returns the names of the assets.

func BeautifyKey 

func BeautifyKey(key string) string

BeautifyKey returns a human readable byte key (used for debugging purposes) it should be in sync with ByteKey Note: This is only used in /debug/* endpoints

func ConnectionSummary 

func ConnectionSummary(c ConnectionStats, names map[util.Address][]string) string

ConnectionSummary returns a string summarizing a connection

func CurrentKernelVersion 

func CurrentKernelVersion() (uint32, error)

CurrentKernelVersion is not implemented on this OS for Tracer

func GetProbeStats 

func GetProbeStats() map[string]int64

GetProbeStats gathers stats about the # of kprobes triggered /missed by reading the kprobe_profile file

func IsBlacklistedConnection 

func IsBlacklistedConnection(scf []*ConnectionFilter, dcf []*ConnectionFilter, conn *ConnectionStats) bool

IsBlacklistedConnection returns true if a given connection should be excluded by the tracer based on user defined filters

func IsTracerSupportedByOS 

func IsTracerSupportedByOS(exclusionList []string) (bool, string)

IsTracerSupportedByOS returns whether or not the current kernel version supports tracer functionality along with some context on why it's not supported

func MustAsset 

func MustAsset(name string) []byte

MustAsset is like Asset but panics when Asset would return an error. It simplifies safe initialization of global variables.

func RestoreAsset 

func RestoreAsset(dir, name string) error

RestoreAsset restores an asset under the given directory

func RestoreAssets 

func RestoreAssets(dir, name string) error

RestoreAssets restores an asset under the given directory recursively

Types

type Config 

type Config struct {
	// CollectTCPConns specifies whether the tracer should collect traffic statistics for TCP connections
	CollectTCPConns bool

	// CollectUDPConns specifies whether the tracer should collect traffic statistics for UDP connections
	CollectUDPConns bool

	// CollectIPv6Conns specifics whether the tracer should capture traffic for IPv6 TCP/UDP connections
	CollectIPv6Conns bool

	// CollectLocalDNS specifies whether the tracer should capture traffic for local DNS calls
	CollectLocalDNS bool

	// DNSInspection specifies whether the tracer should enhance connection data with domain names by inspecting DNS traffic
	// Notice this does *not* depend on CollectLocalDNS
	DNSInspection bool

	// CollectDNSStats specifies whether the tracer should enhance connection data with relevant DNS stats
	// It is relevant *only* when DNSInspection is enabled.
	CollectDNSStats bool

	// UDPConnTimeout determines the length of traffic inactivity between two (IP, port)-pairs before declaring a UDP
	// connection as inactive.
	// Note: As UDP traffic is technically "connection-less", for tracking, we consider a UDP connection to be traffic
	//       between a source and destination IP and port.
	UDPConnTimeout time.Duration

	// TCPConnTimeout is like UDPConnTimeout, but for TCP connections. TCP connections are cleared when
	// the BPF module receives a tcp_close call, but TCP connections also age out to catch cases where
	// tcp_close is not intercepted for some reason.
	TCPConnTimeout time.Duration

	// MaxTrackedConnections specifies the maximum number of connections we can track. This determines the size of the eBPF Maps
	MaxTrackedConnections uint

	// MaxClosedConnectionsBuffered represents the maximum number of closed connections we'll buffer in memory. These closed connections
	// get flushed on every client request (default 30s check interval)
	MaxClosedConnectionsBuffered int

	// MaxDNSStatsBufferred represents the maximum number of DNS stats we'll buffer in memory. These stats
	// get flushed on every client request (default 30s check interval)
	MaxDNSStatsBufferred int

	// MaxConnectionsStateBuffered represents the maximum number of state objects that we'll store in memory. These state objects store
	// the stats for a connection so we can accurately determine traffic change between client requests.
	MaxConnectionsStateBuffered int

	// ClientStateExpiry specifies the max time a client (e.g. process-agent)'s state will be stored in memory before being evicted.
	ClientStateExpiry time.Duration

	// ProcRoot is the root path to the proc filesystem
	ProcRoot string

	// BPFDebug enables bpf debug logs
	BPFDebug bool

	// EnableConntrack enables probing conntrack for network address translation via netlink
	EnableConntrack bool

	// ConntrackIgnoreENOBUFS: When set to true, the system-probe will ignore ENOBUF errors
	// and continue processing Conntrack events when/if the netlink recv buffer overruns.
	// Enabling this can have an adverse effect on CPU utilization for high-throughput systems.
	ConntrackIgnoreENOBUFS bool

	// ConntrackMaxStateSize specifies the maximum number of connections with NAT we can track
	ConntrackMaxStateSize int

	// DebugPort specifies a port to run golang's expvar and pprof debug endpoint
	DebugPort int

	// ClosedChannelSize specifies the size for closed channel for the tracer
	ClosedChannelSize int

	// ExcludedSourceConnections is a map of source connections to blacklist
	ExcludedSourceConnections map[string][]string

	// ExcludedDestinationConnections is a map of destination connections to blacklist
	ExcludedDestinationConnections map[string][]string
}

Config stores all flags used by the eBPF tracer

func NewDefaultConfig 

func NewDefaultConfig() *Config

NewDefaultConfig enables traffic collection for all connection types

func (*Config) EnabledKProbes 

func (c *Config) EnabledKProbes(pre410Kernel bool) map[KProbeName]struct{}

EnabledKProbes returns a map of kprobes that are enabled per config settings. This map does not include the probes used exclusively in the offset guessing process.

type ConnTypeFilter 

type ConnTypeFilter struct {
	TCP bool
	UDP bool
}

ConnTypeFilter holds user-defined protocols

type ConnectionDirection 

type ConnectionDirection uint8

ConnectionDirection indicates if the connection is incoming to the host or outbound 

const (
	// INCOMING represents connections inbound to the host
	INCOMING ConnectionDirection = 1

	// OUTGOING represents outbound connections from the host
	OUTGOING ConnectionDirection = 2

	// LOCAL represents connections that don't leave the host
	LOCAL ConnectionDirection = 3

	// NONE represents connections that have no direction (udp, for example)
	NONE ConnectionDirection = 4
)

func (ConnectionDirection) String 

func (d ConnectionDirection) String() string

type ConnectionFamily 

type ConnectionFamily uint8

ConnectionFamily will be either v4 or v6 

const (
	// AFINET represents v4 connections
	AFINET ConnectionFamily = 0

	// AFINET6 represents v6 connections
	AFINET6 ConnectionFamily = 1
)

type ConnectionFilter 

type ConnectionFilter struct {
	IP       *net.IPNet // If nil, then all IPs will be considered matching.
	AllPorts ConnTypeFilter

	Ports map[uint16]ConnTypeFilter
}

ConnectionFilter holds a user-defined blacklisted IP/CIDR, and ports

func ParseConnectionFilters 

func ParseConnectionFilters(filters map[string][]string) (blacklist []*ConnectionFilter)

ParseConnectionFilters takes the user defined blacklist and returns a slice of ConnectionFilters

type ConnectionStats 

type ConnectionStats struct {
	Source util.Address
	Dest   util.Address

	MonotonicSentBytes uint64
	LastSentBytes      uint64

	MonotonicRecvBytes uint64
	LastRecvBytes      uint64

	// Last time the stats for this connection were updated
	LastUpdateEpoch uint64

	MonotonicRetransmits uint32
	LastRetransmits      uint32

	RTT    uint32 // Stored in µs
	RTTVar uint32

	Pid   uint32
	NetNS uint32

	SPort                  uint16
	DPort                  uint16
	Type                   ConnectionType
	Family                 ConnectionFamily
	Direction              ConnectionDirection
	IPTranslation          *netlink.IPTranslation
	IntraHost              bool
	DNSSuccessfulResponses uint32
}

ConnectionStats stores statistics for a single connection. Field order in the struct should be 8-byte aligned

func (ConnectionStats) ByteKey 

func (c ConnectionStats) ByteKey(buffer *bytes.Buffer) ([]byte, error)

ByteKey returns a unique key for this connection represented as a byte array It's as following: 

32b     16b     16b      4b      4b     32/128b      32/128b

| PID | SPORT | DPORT | Family | Type | SrcAddr | DestAddr

func (ConnectionStats) String 

func (c ConnectionStats) String() string

type ConnectionType 

type ConnectionType uint8

ConnectionType will be either TCP or UDP 

const (
	// TCP connection type
	TCP ConnectionType = 0

	// UDP connection type
	UDP ConnectionType = 1
)

func (ConnectionType) String 

func (c ConnectionType) String() string

type Connections 

type Connections struct {
	DNS   map[util.Address][]string
	Conns []ConnectionStats
}

Connections wraps a collection of ConnectionStats

type KProbeName 

type KProbeName string

KProbeName stores the name of the kernel probes setup for tracing 

const (
	// TCPv4DestroySock traces the tcp_v4_destroy_sock system call (called for both ipv4 and ipv6)
	TCPv4DestroySock KProbeName = "kprobe/tcp_v4_destroy_sock"

	// TCPv6Connect traces the v6 connect() system call
	TCPv6Connect KProbeName = "kprobe/tcp_v6_connect"
	// TCPv6ConnectReturn traces the return value for the v6 connect() system call
	TCPv6ConnectReturn KProbeName = "kretprobe/tcp_v6_connect"

	// TCPSendMsg traces the tcp_sendmsg() system call
	TCPSendMsg KProbeName = "kprobe/tcp_sendmsg"

	// TCPSendMsgPre410 traces the tcp_sendmsg() system call on kernels prior to 4.1.0. This is created because
	// we need to load a different kprobe implementation
	TCPSendMsgPre410 KProbeName = "kprobe/tcp_sendmsg/pre_4_1_0"

	// TCPSendMsgReturn traces the return value for the tcp_sendmsg() system call
	// XXX: This is only used for telemetry for now to count the number of errors returned
	// by the tcp_sendmsg func (so we can have a # of tcp sent bytes we miscounted)
	TCPSendMsgReturn KProbeName = "kretprobe/tcp_sendmsg"

	// TCPGetInfo traces the tcp_get_info() system call
	// This probe is used for offset guessing only
	TCPGetInfo KProbeName = "kprobe/tcp_get_info"

	// TCPCleanupRBuf traces the tcp_cleanup_rbuf() system call
	TCPCleanupRBuf KProbeName = "kprobe/tcp_cleanup_rbuf"
	// TCPClose traces the tcp_close() system call
	TCPClose KProbeName = "kprobe/tcp_close"

	// UDPSendMsg traces the udp_sendmsg() system call
	UDPSendMsg KProbeName = "kprobe/udp_sendmsg"
	// UDPSendMsgPre410 traces the udp_sendmsg() system call on kernels prior to 4.1.0
	UDPSendMsgPre410 KProbeName = "kprobe/udp_sendmsg/pre_4_1_0"
	// UDPRecvMsg traces the udp_recvmsg() system call
	UDPRecvMsg KProbeName = "kprobe/udp_recvmsg"
	// UDPRecvMsgPre410 traces the udp_recvmsg() system call on kernels prior to 4.1.0
	UDPRecvMsgPre410 KProbeName = "kprobe/udp_recvmsg/pre_4_1_0"
	// UDPRecvMsgReturn traces the return value for the udp_recvmsg() system call
	UDPRecvMsgReturn KProbeName = "kretprobe/udp_recvmsg"

	// UDPDestroySock traces the udp_destroy_sock() function
	UDPDestroySock KProbeName = "kprobe/udp_destroy_sock"

	// TCPRetransmit traces the return value for the tcp_retransmit_skb() system call
	TCPRetransmit KProbeName = "kprobe/tcp_retransmit_skb"

	// InetCskAcceptReturn traces the return value for the inet_csk_accept syscall
	InetCskAcceptReturn KProbeName = "kretprobe/inet_csk_accept"

	// SysSocket traces calls to the socket kprobe
	SysSocket KProbeName = "kprobe/sys_socket"

	// SysSocketRet is the kretprobe for SysSocket
	SysSocketRet KProbeName = "kretprobe/sys_socket"

	// SysBind is the kprobe the bind() syscall.
	SysBind KProbeName = "kprobe/sys_bind"

	// SysBindRet is the kretprobe for bind().
	SysBindRet KProbeName = "kretprobe/sys_bind"
)

type NetworkState 

type NetworkState interface {
	// Connections returns the list of connections for the given client when provided the latest set of active connections
	Connections(
		clientID string,
		latestTime uint64,
		latestConns []ConnectionStats,
		dns map[dnsKey]dnsStats,
	) []ConnectionStats

	// StoreClosedConnection stores a new closed connection
	StoreClosedConnection(conn ConnectionStats)

	// RemoveClient stops tracking stateful data for a given client
	RemoveClient(clientID string)

	// RemoveExpiredClients removes expired clients from the state
	RemoveExpiredClients(now time.Time)

	// RemoveConnections removes the given keys from the state
	RemoveConnections(keys []string)

	// GetStats returns a map of statistics about the current network state
	GetStats() map[string]interface{}

	// DebugNetworkState returns a map with the current network state for a client ID
	DumpState(clientID string) map[string]interface{}
}

NetworkState takes care of handling the logic for: - closed connections - sent and received bytes per connection

func NewDefaultNetworkState 

func NewDefaultNetworkState() NetworkState

NewDefaultNetworkState creates a new network state with default settings

func NewNetworkState 

func NewNetworkState(clientExpiry time.Duration, maxClosedConns, maxClientStats, maxDNSStats int) NetworkState

NewNetworkState creates a new network state

type PortMapping 

type PortMapping struct {
	sync.RWMutex
	// contains filtered or unexported fields
}

PortMapping tracks which ports a pid is listening on

func NewPortMapping 

func NewPortMapping(procRoot string, config *Config) *PortMapping

NewPortMapping creates a new PortMapping instance

func (*PortMapping) AddMapping 

func (pm *PortMapping) AddMapping(port uint16)

AddMapping indicates that something is listening on the provided port

func (*PortMapping) IsListening 

func (pm *PortMapping) IsListening(port uint16) bool

IsListening returns true if something is listening on the given port

func (*PortMapping) ReadInitialState 

func (pm *PortMapping) ReadInitialState() error

ReadInitialState reads the /proc filesystem and determines which ports are being listened on

func (*PortMapping) ReadInitialUDPState 

func (pm *PortMapping) ReadInitialUDPState() error

ReadInitialUDPState reads the /proc filesystem and determines which ports are being used as UDP server

func (*PortMapping) RemoveMapping 

func (pm *PortMapping) RemoveMapping(port uint16)

RemoveMapping indicates that the provided port is no longer being listened on

type ReverseDNS 

type ReverseDNS interface {
	Resolve([]ConnectionStats) map[util.Address][]string
	GetDNSStats() map[dnsKey]dnsStats
	GetStats() map[string]int64
	Close()
}

ReverseDNS translates IPs to names

type TCPQueueLengthTracer 

type TCPQueueLengthTracer struct{}

TCPQueueLengthTracer is not implemented on non-linux systems

func NewTCPQueueLengthTracer 

func NewTCPQueueLengthTracer() (*TCPQueueLengthTracer, error)

NewTCPQueueLengthTracer is not implemented on non-linux systems

func (*TCPQueueLengthTracer) Close 

func (t *TCPQueueLengthTracer) Close()

Close is not implemented on non-linux systems

func (*TCPQueueLengthTracer) Get 

func (t *TCPQueueLengthTracer) Get() []tcpqueuelength.Stats

Get is not implemented on non-linux systems

func (*TCPQueueLengthTracer) GetAndFlush 

func (t *TCPQueueLengthTracer) GetAndFlush() []tcpqueuelength.Stats

GetAndFlush is not implemented on non-linux systems

type Tracer 

type Tracer struct{}

Tracer is not implemented

func NewTracer 

func NewTracer(_ *Config) (*Tracer, error)

NewTracer is not implemented on this OS for Tracer

func (*Tracer) DebugNetworkMaps 

func (t *Tracer) DebugNetworkMaps() (*Connections, error)

DebugNetworkMaps is not implemented on this OS for Tracer

func (*Tracer) DebugNetworkState 

func (t *Tracer) DebugNetworkState(clientID string) (map[string]interface{}, error)

DebugNetworkState is not implemented on this OS for Tracer

func (*Tracer) GetActiveConnections 

func (t *Tracer) GetActiveConnections(_ string) (*Connections, error)

GetActiveConnections is not implemented on this OS for Tracer

func (*Tracer) GetStats 

func (t *Tracer) GetStats() (map[string]interface{}, error)

GetStats is not implemented on this OS for Tracer

func (*Tracer) Stop 

func (t *Tracer) Stop()

Stop is not implemented on this OS for Tracer

Source Files

View all Source files

Directories

Path Synopsis

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.