Files
weave-scope/render/filters.go
Iago López Galeiras 9920c4ea48 Add eBPF connection tracking without dependencies on kernel headers
Based on work from Lorenzo, updated by Iago, Alban, Alessandro and
Michael.

This PR adds connection tracking using eBPF. This feature is not enabled by default.
For now, you can enable it by launching scope with the following command:

```
sudo ./scope launch --probe.ebpf.connections=true
```

This patch allows scope to get notified of every connection event,
without relying on the parsing of /proc/$pid/net/tcp{,6} and
/proc/$pid/fd/*, and therefore improve performance.

We vendor https://github.com/iovisor/gobpf in Scope to load the
pre-compiled ebpf program and https://github.com/weaveworks/tcptracer-bpf
to guess the offsets of the structures we need in the kernel. In this
way we don't need a different pre-compiled ebpf object file per kernel.
The pre-compiled ebpf program is included in the vendoring of
tcptracer-bpf.

The ebpf program uses kprobes/kretprobes on the following kernel functions:
- tcp_v4_connect
- tcp_v6_connect
- tcp_set_state
- inet_csk_accept
- tcp_close

It generates "connect", "accept" and "close" events containing the
connection tuple but also pid and netns.
Note: the IPv6 events are not supported in Scope and thus not passed on.

probe/endpoint/ebpf.go maintains the list of connections. Similarly to
conntrack, it also keeps the dead connections for one iteration in order
to report short-lived connections.

The code for parsing /proc/$pid/net/tcp{,6} and /proc/$pid/fd/* is still
there and still used at start-up because eBPF only brings us the events
and not the initial state. However, the /proc parsing for the initial
state is now done in foreground instead of background, via
newForegroundReader().

NAT resolution on connections from eBPF works in the same way as it did
on connections from /proc: by using conntrack. One of the two conntrack
instances is only started to get the initial state and then it is
stopped since eBPF detects short-lived connections.

The Scope Docker image size comparison:
- weaveworks/scope in current master:  22 MB (compressed),  68 MB
  (uncompressed)
- weaveworks/scope with this patchset: 23 MB (compressed), 69 MB
  (uncompressed)

Fixes #1168 (walking /proc to obtain connections is very expensive)

Fixes #1260 (Short-lived connections not tracked for containers in
shared networking namespaces)

Fixes #1962 (Port ebpf tracker to Go)

Fixes #1961 (Remove runtime kernel header dependency from ebpf tracker)
2017-03-08 22:11:12 +01:00

348 lines
9.8 KiB
Go

package render
import (
"strings"
"github.com/weaveworks/common/mtime"
"github.com/weaveworks/scope/probe/docker"
"github.com/weaveworks/scope/probe/endpoint"
"github.com/weaveworks/scope/probe/kubernetes"
"github.com/weaveworks/scope/report"
)
// PreciousNodeRenderer ensures a node is never filtered out by decorators
type PreciousNodeRenderer struct {
PreciousNodeID string
Renderer
}
// Render implements Renderer
func (p PreciousNodeRenderer) Render(rpt report.Report, dct Decorator) report.Nodes {
undecoratedNodes := p.Renderer.Render(rpt, nil)
preciousNode, foundBeforeDecoration := undecoratedNodes[p.PreciousNodeID]
finalNodes := applyDecorator{ConstantRenderer(undecoratedNodes)}.Render(rpt, dct)
if _, ok := finalNodes[p.PreciousNodeID]; !ok && foundBeforeDecoration {
finalNodes[p.PreciousNodeID] = preciousNode
}
return finalNodes
}
// Stats implements Renderer
func (p PreciousNodeRenderer) Stats(rpt report.Report, dct Decorator) Stats {
// default to the underlying renderer
// TODO: we don't take into account the precious node, so we may be off by one
return p.Renderer.Stats(rpt, dct)
}
// CustomRenderer allow for mapping functions that received the entire topology
// in one call - useful for functions that need to consider the entire graph.
// We should minimise the use of this renderer type, as it is very inflexible.
type CustomRenderer struct {
RenderFunc func(report.Nodes) report.Nodes
Renderer
}
// Render implements Renderer
func (c CustomRenderer) Render(rpt report.Report, dct Decorator) report.Nodes {
return c.RenderFunc(c.Renderer.Render(rpt, dct))
}
// ColorConnected colors nodes with the IsConnected key if
// they have edges to or from them. Edges to/from yourself
// are not counted here (see #656).
func ColorConnected(r Renderer) Renderer {
return CustomRenderer{
Renderer: r,
RenderFunc: func(input report.Nodes) report.Nodes {
connected := map[string]struct{}{}
void := struct{}{}
for id, node := range input {
if len(node.Adjacency) == 0 {
continue
}
for _, adj := range node.Adjacency {
if adj != id {
connected[id] = void
connected[adj] = void
}
}
}
output := input.Copy()
for id := range connected {
output[id] = output[id].WithLatest(IsConnected, mtime.Now(), "true")
}
return output
},
}
}
// FilterFunc is the function type used by Filters
type FilterFunc func(report.Node) bool
// AnyFilterFunc checks if any of the filterfuncs matches.
func AnyFilterFunc(fs ...FilterFunc) FilterFunc {
return func(n report.Node) bool {
for _, f := range fs {
if f(n) {
return true
}
}
return false
}
}
// ComposeFilterFuncs composes filterfuncs into a single FilterFunc checking all.
func ComposeFilterFuncs(fs ...FilterFunc) FilterFunc {
return func(n report.Node) bool {
for _, f := range fs {
if !f(n) {
return false
}
}
return true
}
}
// Filter removes nodes from a view based on a predicate.
type Filter struct {
Renderer
FilterFunc FilterFunc
}
// MakeFilter makes a new Filter (that ignores pseudo nodes).
func MakeFilter(f FilterFunc, r Renderer) Renderer {
return Memoise(&Filter{
Renderer: r,
FilterFunc: func(n report.Node) bool {
return n.Topology == Pseudo || f(n)
},
})
}
// MakeFilterPseudo makes a new Filter that will not ignore pseudo nodes.
func MakeFilterPseudo(f FilterFunc, r Renderer) Renderer {
return Memoise(&Filter{
Renderer: r,
FilterFunc: f,
})
}
// MakeFilterDecorator makes a decorator that filters out non-pseudo nodes
// which match the predicate.
func MakeFilterDecorator(f FilterFunc) Decorator {
return func(renderer Renderer) Renderer {
return MakeFilter(f, renderer)
}
}
// MakeFilterPseudoDecorator makes a decorator that filters out all nodes
// (including pseudo nodes) which match the predicate.
func MakeFilterPseudoDecorator(f FilterFunc) Decorator {
return func(renderer Renderer) Renderer {
return MakeFilterPseudo(f, renderer)
}
}
// Render implements Renderer
func (f *Filter) Render(rpt report.Report, dct Decorator) report.Nodes {
nodes, _ := f.render(rpt, dct)
return nodes
}
func (f *Filter) render(rpt report.Report, dct Decorator) (report.Nodes, int) {
output := report.Nodes{}
inDegrees := map[string]int{}
filtered := 0
for id, node := range f.Renderer.Render(rpt, dct) {
if f.FilterFunc(node) {
output[id] = node
inDegrees[id] = 0
} else {
filtered++
}
}
// Deleted nodes also need to be cut as destinations in adjacency lists.
for id, node := range output {
newAdjacency := report.MakeIDList()
for _, dstID := range node.Adjacency {
if _, ok := output[dstID]; ok {
newAdjacency = newAdjacency.Add(dstID)
inDegrees[dstID]++
}
}
node.Adjacency = newAdjacency
output[id] = node
}
// Remove unconnected pseudo nodes, see #483.
for id, inDegree := range inDegrees {
if inDegree > 0 {
continue
}
node := output[id]
if node.Topology != Pseudo || len(node.Adjacency) > 0 {
continue
}
delete(output, id)
filtered++
}
return output, filtered
}
// Stats implements Renderer. General logic is to take the first (i.e.
// highest-level) stats we find, so upstream stats are ignored. This means that
// if we want to count the stats from multiple filters we need to compose their
// FilterFuncs, into a single Filter.
func (f Filter) Stats(rpt report.Report, dct Decorator) Stats {
_, filtered := f.render(rpt, dct)
return Stats{FilteredNodes: filtered}
}
// IsConnected is the key added to Node.Metadata by ColorConnected
// to indicate a node has an edge pointing to it or from it
const IsConnected = "is_connected"
// Complement takes a FilterFunc f and returns a FilterFunc that has the same
// effects, if any, and returns the opposite truth value.
func Complement(f FilterFunc) FilterFunc {
return func(node report.Node) bool { return !f(node) }
}
// FilterUnconnected produces a renderer that filters unconnected nodes
// from the given renderer
func FilterUnconnected(r Renderer) Renderer {
return MakeFilter(
func(node report.Node) bool {
_, ok := node.Latest.Lookup(IsConnected)
return ok
},
ColorConnected(r),
)
}
// Noop allows all nodes through
func Noop(_ report.Node) bool { return true }
// IsRunning checks if the node is a running docker container
func IsRunning(n report.Node) bool {
state, ok := n.Latest.Lookup(docker.ContainerState)
return !ok || (state == docker.StateRunning || state == docker.StateRestarting || state == docker.StatePaused)
}
// IsStopped checks if the node is *not* a running docker container
var IsStopped = Complement(IsRunning)
func nonProcspiedFilter(node report.Node) bool {
_, ok := node.Latest.Lookup(endpoint.Procspied)
return ok
}
func nonEBPFFilter(node report.Node) bool {
_, ok := node.Latest.Lookup(endpoint.EBPF)
return ok
}
// FilterNonProcspied removes endpoints which were not found in procspy.
func FilterNonProcspied(r Renderer) Renderer {
return MakeFilter(nonProcspiedFilter, r)
}
// FilterNonEBPF removes endpoints which were not found via eBPF.
func FilterNonEBPF(r Renderer) Renderer {
return MakeFilter(nonEBPFFilter, r)
}
// FilterNonProcspiedNorEBPF removes endpoints which were not found in procspy
// nor via eBPF.
func FilterNonProcspiedNorEBPF(r Renderer) Renderer {
return MakeFilter(AnyFilterFunc(nonProcspiedFilter, nonEBPFFilter), r)
}
// IsApplication checks if the node is an "application" node
func IsApplication(n report.Node) bool {
containerName, _ := n.Latest.Lookup(docker.ContainerName)
if _, ok := systemContainerNames[containerName]; ok {
return false
}
imageName, _ := n.Latest.Lookup(docker.ImageName)
imagePrefix := strings.SplitN(imageName, ":", 2)[0] // :(
if _, ok := systemImagePrefixes[imagePrefix]; ok || kubernetes.IsPauseImageName(imagePrefix) {
return false
}
roleLabel, _ := n.Latest.Lookup(docker.LabelPrefix + "works.weave.role")
if roleLabel == "system" {
return false
}
roleLabel, _ = n.Latest.Lookup(docker.ImageLabelPrefix + "works.weave.role")
if roleLabel == "system" {
return false
}
namespace, _ := n.Latest.Lookup(docker.LabelPrefix + "io.kubernetes.pod.namespace")
if namespace == "kube-system" {
return false
}
podName, _ := n.Latest.Lookup(docker.LabelPrefix + "io.kubernetes.pod.name")
if strings.HasPrefix(podName, "kube-system/") {
return false
}
return true
}
// IsSystem checks if the node is a "system" node
var IsSystem = Complement(IsApplication)
// HasLabel checks if the node has the desired docker label
func HasLabel(labelKey string, labelValue string) FilterFunc {
return func(n report.Node) bool {
value, _ := n.Latest.Lookup(docker.LabelPrefix + labelKey)
if value == labelValue {
return true
}
return false
}
}
// DoesNotHaveLabel checks if the node does NOT have the specified docker label
func DoesNotHaveLabel(labelKey string, labelValue string) FilterFunc {
return Complement(HasLabel(labelKey, labelValue))
}
// IsNotPseudo returns true if the node is not a pseudo node
// or internet/service nodes.
func IsNotPseudo(n report.Node) bool {
return n.Topology != Pseudo || strings.HasSuffix(n.ID, TheInternetID) || strings.HasPrefix(n.ID, ServiceNodeIDPrefix)
}
// IsNamespace checks if the node is a pod/service in the specified namespace
func IsNamespace(namespace string) FilterFunc {
return func(n report.Node) bool {
gotNamespace, _ := n.Latest.Lookup(kubernetes.Namespace)
return namespace == gotNamespace
}
}
var systemContainerNames = map[string]struct{}{
"weavescope": {},
"weavedns": {},
"weave": {},
"weaveproxy": {},
"weaveexec": {},
"ecs-agent": {},
}
var systemImagePrefixes = map[string]struct{}{
"swarm": {},
"weaveworks/scope": {},
"weaveworks/weavedns": {},
"weaveworks/weave": {},
"weaveworks/weaveproxy": {},
"weaveworks/weaveexec": {},
"amazon/amazon-ecs-agent": {},
"openshift/origin-pod": {},
"docker.io/openshift/origin-pod": {},
}