Framework for running BPF tracers with rules on Linux as a daemon. Container aware.
This is not just "yet another tool to trace"...
Since it uses BPF and allows for any implementation of the Tracer
interface you
can use it to do all sorts of things from modifying a file everytime a call to open
is
called on it, to hot patching an internal kernel function to prevent a known vulnerability
without the need to upgrade your kernel.
More use cases with examples coming soon... for now see how it works.
Table of Contents
Tracers retrieve the data... Rules filter the data... Actions perform actions on the data.
The tracers are in the tracer/ folder. The idea is that you can add any tracers you would like and then create rules for the data retrieved from the tracers. Any events with data that passes the filters will be passed on to the specified action.
The tracers that exist today are based off a few bcc-tools tracers.
You could always add your own tracers in a fork if you worry people will reverse engineer the data you are collecting and alerting on.
The current compiled in tracers are:
- dockeropenbreakout: trace when files that are not inside the container rootfs are being accessed
- bashreadline: trace commands being entered into the bash command line
- exec: trace calls to exec binaries
- open: trace calls to open files
These must implement the Tracer
interface:
// Tracer defines the basic capabilities of a tracer.
type Tracer interface {
// Load creates the bpf module and starts collecting the data for the tracer.
Load() error
// Unload closes the bpf module and all the probes that all attached to it.
Unload()
// WatchEvent defines the function to watch the events for the tracer.
WatchEvent() (*grpc.Event, error)
// Start starts the map for the tracer.
Start()
// String returns a string representation of this tracer.
String() string
}
As you can see from above you could technically implement this interface with something other than BPF ;)
The Event
type defines the data returned from the tracer. As you can see
below, the Data
is of type map[string]string
meaning any key value pair can
be returned for the data. The rules then filter using those key value pairs.
// Event defines the data struct for holding event data.
type Event struct {
PID uint32 // Process ID.
TGID uint32 // Task group ID.
UID uint32 // User ID.
GID uint32 // User group ID.
Command string // The command for the process.
ReturnValue int32 // The return value for the function.
Data map[string]string
ContainerRuntime string // Filled in after the tracer is run so you don't need to.
ContainerID string // Filled in after the tracer is run so you don't need to.
Tracer string // Filled in after the tracer is run so you don't need to.
}
These are toml files that hold some logic for what you would like to trace.
You can search for anything returned by a Tracer
in its map[string]string
data struct.
You can also filter based off the container runtime you would like to alert on. The container runtime must be one of the strings defined here.
If you provide no rules for a tracer, then all the events will be passed to actions.
The example below describes a rule file to filter the data returned from the
exec
tracer. Events from exec
will only be returned if the command
matches
one of those values AND the container runtime is docker
or kube
.
tracer = "exec"
actions = ["stdout"]
[filterEvents]
[filterEvents.command]
values = ["sshd", "dbus-daemon-lau", "ping", "ping6", "critical-stack-", "pmmcli", "filemng", "PassengerAgent", "bwrap", "osdetect", "nginxmng", "sw-engine-fpm", "start-stop-daem"]
containerRuntimes = ["docker","kube"]
If you are wondering where the command
key comes from, it's defined in the
exec
tracer here.
Rules can be dynamically controlled via bpfd's gRPC interface.
The cli tool can also be used for creating rules dynamically, see
create
usage.
The protobuf protocol definition is defined in api/grpc/api.proto
To interact with the gRPC api you can use the --gpc-addr
flag
or the default is a sock at /run/bpfd/bpfd.sock
.
Actions do "something" on an event. This way you can send filtered events to Slack, email, or even run arbitrary code. You could kill a container, pause a container, or checkpoint a container to restore it elsewhere without even having to login to a computer.
The current compiled in actions are:
Actions implement the Actions
interface:
// Action performs an action on an event.
type Action interface {
// Do runs the action on an event.
Do(event *grpc.Event) error
// String returns a string representation of this tracer.
String() string
}
To build, you need to have libbcc
installed SEE INSTRUCTIONS HERE
For installation instructions from binaries please visit the Releases Page.
$ go get github.com/genuinetools/bpfd
$ docker run --rm -it \
--name bpfd \
-v /lib/modules:/lib/modules:ro \
-v /usr/src:/usr/src:ro \
--privileged \
r.j3ss.co/bpfd daemon
$ bpfd -h
bpfd - Framework for running BPF tracers with rules on Linux as a daemon.
Usage: bpfd <command>
Flags:
-d, --debug enable debug logging (default: false)
--grpc-addr Address for gRPC api communication (default: /run/bpfd/bpfd.sock)
Commands:
create Create one or more rules.
daemon Start the daemon.
ls List rules.
rm Remove one or more rules.
trace Live trace the events returned after filtering.
version Show the version information.
You can preload rules by passing --rules-dir
to the command or placing
rules in the default directory: /etc/bpfd/rules
.
$ bpfd daemon -h
Usage: bpfd daemon [OPTIONS]
Start the daemon.
Flags:
-d, --debug enable debug logging (default: false)
--grpc-addr Address for gRPC api communication (default: /run/bpfd/bpfd.sock)
--rules-dir Directory that stores the rules files (default: /etc/bpfd/rules)
You can create rules on the fly with the create
command. You can pass more
than one file at a time.
Usage: bpfd create [OPTIONS] RULE_FILE [RULE_FILE...]
Create one or more rules.
Flags:
-d, --debug enable debug logging (default: false)
--grpc-addr Address for gRPC api communication (default: /run/bpfd/bpfd.sock)
You can delete rules with the rm
command. You can pass more than one
rule name at a time.
$ bpfd rm -h
Usage: bpfd rm [OPTIONS] RULE_NAME [RULE_NAME...]
Remove one or more rules.
Flags:
-d, --debug enable debug logging (default: false)
--grpc-addr Address for gRPC api communication (default: /run/bpfd/bpfd.sock)
You can list the rules that the daemon is filtering with by using the ls
command.
$ bpfd ls
NAME TRACER
bashreadline bashreadline
password_files open
setuid_binaries exec
You can live trace the events returned after filtering with the trace
command.
This does not include past events. Consider it like a tail.
$ bpfd trace
INFO[0000] map[string]string{"filename":"/etc/shadow", "command":"sudo", "returnval":"4"} container_id= container_runtime=not-found pid=12893 tracer=open tgid=0
INFO[0000] map[string]string{"command":"sudo", "returnval":"4", "filename":"/etc/sudoers.d/README"} container_id= container_runtime=not-found pid=12893 tracer=open tgid=0
INFO[0000] map[string]string{"command":"sudo", "returnval":"4", "filename":"/etc/sudoers.d"} container_id= container_runtime=not-found pid=12893 tracer=open tgid=0
INFO[0000] map[string]string{"filename":"/etc/sudoers", "command":"sudo", "returnval":"3"} container_id= container_runtime=not-found pid=12893 tracer=open tgid=0
INFO[0000] map[string]string{"command":"sudo bpfd trace"} container_id= container_runtime=not-found pid=23751 tracer=bashreadline tgid=0
INFO[0000] map[string]string{"command":"vim README.md"} container_id= container_runtime=not-found pid=23751 tracer=bashreadline tgid=0
INFO[0000] map[string]string{"filename":"/etc/shadow", "command":"sudo", "returnval":"4"} container_id= container_runtime=not-found pid=12786 tracer=open tgid=0
INFO[0000] map[string]string{"command":"sudo", "returnval":"4", "filename":"/etc/sudoers.d/README"} container_id= container_runtime=not-found pid=12786 tracer=open tgid=0
INFO[0000] map[string]string{"filename":"/etc/sudoers.d", "command":"sudo", "returnval":"4"} container_id= container_runtime=not-found pid=12786 tracer=open tgid=0
INFO[0000] map[string]string{"filename":"/etc/sudoers", "command":"sudo", "returnval":"3"} container_id= container_runtime=not-found pid=12786 tracer=open tgid=0