Documentation ¶
Overview ¶
Package bpf implements marshaling and unmarshaling of programs for the Berkeley Packet Filter virtual machine, and provides a Go implementation of the virtual machine.
BPF's main use is to specify a packet filter for network taps, so that the kernel doesn't have to expensively copy every packet it sees to userspace. However, it's been repurposed to other areas where running user code in-kernel is needed. For example, Linux's seccomp uses BPF to apply security policies to system calls. For simplicity, this documentation refers only to packets, but other uses of BPF have their own data payloads.
BPF programs run in a restricted virtual machine. It has almost no access to kernel functions, and while conditional branches are allowed, they can only jump forwards, to guarantee that there are no infinite loops.
The virtual machine ¶
The BPF VM is an accumulator machine. Its main register, called register A, is an implicit source and destination in all arithmetic and logic operations. The machine also has 16 scratch registers for temporary storage, and an indirection register (register X) for indirect memory access. All registers are 32 bits wide.
Each run of a BPF program is given one packet, which is placed in the VM's read-only "main memory". LoadAbsolute and LoadIndirect instructions can fetch up to 32 bits at a time into register A for examination.
The goal of a BPF program is to produce and return a verdict (uint32), which tells the kernel what to do with the packet. In the context of packet filtering, the returned value is the number of bytes of the packet to forward to userspace, or 0 to ignore the packet. Other contexts like seccomp define their own return values.
In order to simplify programs, attempts to read past the end of the packet terminate the program execution with a verdict of 0 (ignore packet). This means that the vast majority of BPF programs don't need to do any explicit bounds checking.
In addition to the bytes of the packet, some BPF programs have access to extensions, which are essentially calls to kernel utility functions. Currently, the only extensions supported by this package are the Linux packet filter extensions.
Examples ¶
This packet filter selects all ARP packets.
bpf.Assemble([]bpf.Instruction{ // Load "EtherType" field from the ethernet header. bpf.LoadAbsolute{Off: 12, Size: 2}, // Skip over the next instruction if EtherType is not ARP. bpf.JumpIf{Cond: bpf.JumpNotEqual, Val: 0x0806, SkipTrue: 1}, // Verdict is "send up to 4k of the packet to userspace." bpf.RetConstant{Val: 4096}, // Verdict is "ignore packet." bpf.RetConstant{Val: 0}, })
This packet filter captures a random 1% sample of traffic.
bpf.Assemble([]bpf.Instruction{ // Get a 32-bit random number from the Linux kernel. bpf.LoadExtension{Num: bpf.ExtRand}, // 1% dice roll? bpf.JumpIf{Cond: bpf.JumpLessThan, Val: 2^32/100, SkipFalse: 1}, // Capture. bpf.RetConstant{Val: 4096}, // Ignore. bpf.RetConstant{Val: 0}, })
Index ¶
- type ALUOp
- type ALUOpConstant
- type ALUOpX
- type Extension
- type Instruction
- type Jump
- type JumpIf
- type JumpTest
- type LoadAbsolute
- type LoadConstant
- type LoadExtension
- type LoadIndirect
- type LoadMemShift
- type LoadScratch
- type NegateA
- type RawInstruction
- type Register
- type RetA
- type RetConstant
- type Setter
- type StoreScratch
- type TAX
- type TXA
- type VM
Examples ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type ALUOpConstant ¶
ALUOpConstant executes A = A <Op> Val.
func (ALUOpConstant) Assemble ¶
func (a ALUOpConstant) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
func (ALUOpConstant) String ¶
func (a ALUOpConstant) String() string
String returns the instruction in assembler notation.
type ALUOpX ¶
type ALUOpX struct {
Op ALUOp
}
ALUOpX executes A = A <Op> X
func (ALUOpX) Assemble ¶
func (a ALUOpX) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
type Extension ¶
type Extension int
An Extension is a function call provided by the kernel that performs advanced operations that are expensive or impossible within the BPF virtual machine.
Extensions are only implemented by the Linux kernel.
TODO: should we prune this list? Some of these extensions seem either broken or near-impossible to use correctly, whereas other (len, random, ifindex) are quite useful.
const ( // ExtLen returns the length of the packet. ExtLen Extension = 1 // ExtProto returns the packet's L3 protocol type. ExtProto Extension = 0 // ExtType returns the packet's type (skb->pkt_type in the kernel) // // TODO: better documentation. How nice an API do we want to // provide for these esoteric extensions? ExtType Extension = 4 // ExtPayloadOffset returns the offset of the packet payload, or // the first protocol header that the kernel does not know how to // parse. ExtPayloadOffset Extension = 52 // ExtInterfaceIndex returns the index of the interface on which // the packet was received. ExtInterfaceIndex Extension = 8 // ExtNetlinkAttr returns the netlink attribute of type X at // offset A. ExtNetlinkAttr Extension = 12 // ExtNetlinkAttrNested returns the nested netlink attribute of // type X at offset A. ExtNetlinkAttrNested Extension = 16 // ExtMark returns the packet's mark value. ExtMark Extension = 20 // ExtQueue returns the packet's assigned hardware queue. ExtQueue Extension = 24 // ExtLinkLayerType returns the packet's hardware address type // (e.g. Ethernet, Infiniband). ExtLinkLayerType Extension = 28 // ExtRXHash returns the packets receive hash. // // TODO: figure out what this rxhash actually is. ExtRXHash Extension = 32 // ExtCPUID returns the ID of the CPU processing the current // packet. ExtCPUID Extension = 36 // ExtVLANTag returns the packet's VLAN tag. ExtVLANTag Extension = 44 // ExtVLANTagPresent returns non-zero if the packet has a VLAN // tag. // // TODO: I think this might be a lie: it reads bit 0x1000 of the // VLAN header, which changed meaning in recent revisions of the // spec - this extension may now return meaningless information. ExtVLANTagPresent Extension = 48 // ExtVLANProto returns 0x8100 if the frame has a VLAN header, // 0x88a8 if the frame has a "Q-in-Q" double VLAN header, or some // other value if no VLAN information is present. ExtVLANProto Extension = 60 // ExtRand returns a uniformly random uint32. ExtRand Extension = 56 )
Extension functions available in the Linux kernel.
type Instruction ¶
type Instruction interface { // Assemble assembles the Instruction into a RawInstruction. Assemble() (RawInstruction, error) }
An Instruction is one instruction executed by the BPF virtual machine.
func Disassemble ¶
func Disassemble(raw []RawInstruction) (insts []Instruction, allDecoded bool)
Disassemble attempts to parse raw back into Instructions. Unrecognized RawInstructions are assumed to be an extension not implemented by this package, and are passed through unchanged to the output. The allDecoded value reports whether insts contains no RawInstructions.
type Jump ¶
type Jump struct {
Skip uint32
}
Jump skips the following Skip instructions in the program.
func (Jump) Assemble ¶
func (a Jump) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
type JumpIf ¶
JumpIf skips the following Skip instructions in the program if A <Cond> Val is true.
func (JumpIf) Assemble ¶
func (a JumpIf) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
type LoadAbsolute ¶
LoadAbsolute loads packet[Off:Off+Size] as an integer value into register A.
func (LoadAbsolute) Assemble ¶
func (a LoadAbsolute) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
func (LoadAbsolute) String ¶
func (a LoadAbsolute) String() string
String returns the instruction in assembler notation.
type LoadConstant ¶
LoadConstant loads Val into register Dst.
func (LoadConstant) Assemble ¶
func (a LoadConstant) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
func (LoadConstant) String ¶
func (a LoadConstant) String() string
String returns the instruction in assembler notation.
type LoadExtension ¶
type LoadExtension struct {
Num Extension
}
LoadExtension invokes a linux-specific extension and stores the result in register A.
func (LoadExtension) Assemble ¶
func (a LoadExtension) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
func (LoadExtension) String ¶
func (a LoadExtension) String() string
String returns the instruction in assembler notation.
type LoadIndirect ¶
LoadIndirect loads packet[X+Off:X+Off+Size] as an integer value into register A.
func (LoadIndirect) Assemble ¶
func (a LoadIndirect) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
func (LoadIndirect) String ¶
func (a LoadIndirect) String() string
String returns the instruction in assembler notation.
type LoadMemShift ¶
type LoadMemShift struct {
Off uint32
}
LoadMemShift multiplies the first 4 bits of the byte at packet[Off] by 4 and stores the result in register X.
This instruction is mainly useful to load into X the length of an IPv4 packet header in a single instruction, rather than have to do the arithmetic on the header's first byte by hand.
func (LoadMemShift) Assemble ¶
func (a LoadMemShift) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
func (LoadMemShift) String ¶
func (a LoadMemShift) String() string
String returns the instruction in assembler notation.
type LoadScratch ¶
LoadScratch loads scratch[N] into register Dst.
func (LoadScratch) Assemble ¶
func (a LoadScratch) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
func (LoadScratch) String ¶
func (a LoadScratch) String() string
String returns the instruction in assembler notation.
type NegateA ¶
type NegateA struct{}
NegateA executes A = -A.
func (NegateA) Assemble ¶
func (a NegateA) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
type RawInstruction ¶
type RawInstruction struct { // Operation to execute. Op uint16 // For conditional jump instructions, the number of instructions // to skip if the condition is true/false. Jt uint8 Jf uint8 // Constant parameter. The meaning depends on the Op. K uint32 }
A RawInstruction is a raw BPF virtual machine instruction.
func Assemble ¶
func Assemble(insts []Instruction) ([]RawInstruction, error)
Assemble converts insts into raw instructions suitable for loading into a BPF virtual machine.
Currently, no optimization is attempted, the assembled program flow is exactly as provided.
func (RawInstruction) Assemble ¶
func (ri RawInstruction) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
func (RawInstruction) Disassemble ¶
func (ri RawInstruction) Disassemble() Instruction
Disassemble parses ri into an Instruction and returns it. If ri is not recognized by this package, ri itself is returned.
type RetA ¶
type RetA struct{}
RetA exits the BPF program, returning the value of register A.
func (RetA) Assemble ¶
func (a RetA) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
type RetConstant ¶
type RetConstant struct {
Val uint32
}
RetConstant exits the BPF program, returning a constant value.
func (RetConstant) Assemble ¶
func (a RetConstant) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
func (RetConstant) String ¶
func (a RetConstant) String() string
String returns the instruction in assembler notation.
type Setter ¶
type Setter interface {
SetBPF(filter []RawInstruction) error
}
A Setter is a type which can attach a compiled BPF filter to itself.
type StoreScratch ¶
StoreScratch stores register Src into scratch[N].
func (StoreScratch) Assemble ¶
func (a StoreScratch) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
func (StoreScratch) String ¶
func (a StoreScratch) String() string
String returns the instruction in assembler notation.
type TAX ¶
type TAX struct{}
TAX copies the value of register A to register X.
func (TAX) Assemble ¶
func (a TAX) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
type TXA ¶
type TXA struct{}
TXA copies the value of register X to register A.
func (TXA) Assemble ¶
func (a TXA) Assemble() (RawInstruction, error)
Assemble implements the Instruction Assemble method.
type VM ¶
type VM struct {
// contains filtered or unexported fields
}
A VM is an emulated BPF virtual machine.
func NewVM ¶
func NewVM(filter []Instruction) (*VM, error)
NewVM returns a new VM using the input BPF program.
Example ¶
ExampleNewVM demonstrates usage of a VM, using an Ethernet frame as input and checking its EtherType to determine if it should be accepted.
package main import ( "fmt" "golang.org/x/net/bpf" ) func main() { // Offset | Length | Comment // ------------------------- // 00 | 06 | Ethernet destination MAC address // 06 | 06 | Ethernet source MAC address // 12 | 02 | Ethernet EtherType const ( etOff = 12 etLen = 2 etARP = 0x0806 ) // Set up a VM to filter traffic based on if its EtherType // matches the ARP EtherType. vm, err := bpf.NewVM([]bpf.Instruction{ // Load EtherType value from Ethernet header bpf.LoadAbsolute{ Off: etOff, Size: etLen, }, // If EtherType is equal to the ARP EtherType, jump to allow // packet to be accepted bpf.JumpIf{ Cond: bpf.JumpEqual, Val: etARP, SkipTrue: 1, }, // EtherType does not match the ARP EtherType bpf.RetConstant{ Val: 0, }, // EtherType matches the ARP EtherType, accept up to 1500 // bytes of packet bpf.RetConstant{ Val: 1500, }, }) if err != nil { panic(fmt.Sprintf("failed to load BPF program: %v", err)) } // Create an Ethernet frame with the ARP EtherType for testing frame := []byte{ 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0x00, 0x11, 0x22, 0x33, 0x44, 0x55, 0x08, 0x06, // Payload omitted for brevity } // Run our VM's BPF program using the Ethernet frame as input out, err := vm.Run(frame) if err != nil { panic(fmt.Sprintf("failed to accept Ethernet frame: %v", err)) } // BPF VM can return a byte count greater than the number of input // bytes, so trim the output to match the input byte length if out > len(frame) { out = len(frame) } fmt.Printf("out: %d bytes", out) }
Output: out: 14 bytes