switch to go vendoring
This commit is contained in:
+171
@@ -0,0 +1,171 @@
|
||||
package link
|
||||
|
||||
import (
|
||||
"errors"
|
||||
"fmt"
|
||||
"os"
|
||||
|
||||
"github.com/cilium/ebpf"
|
||||
)
|
||||
|
||||
type cgroupAttachFlags uint32
|
||||
|
||||
// cgroup attach flags
|
||||
const (
|
||||
flagAllowOverride cgroupAttachFlags = 1 << iota
|
||||
flagAllowMulti
|
||||
flagReplace
|
||||
)
|
||||
|
||||
type CgroupOptions struct {
|
||||
// Path to a cgroupv2 folder.
|
||||
Path string
|
||||
// One of the AttachCgroup* constants
|
||||
Attach ebpf.AttachType
|
||||
// Program must be of type CGroup*, and the attach type must match Attach.
|
||||
Program *ebpf.Program
|
||||
}
|
||||
|
||||
// AttachCgroup links a BPF program to a cgroup.
|
||||
func AttachCgroup(opts CgroupOptions) (Link, error) {
|
||||
cgroup, err := os.Open(opts.Path)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("can't open cgroup: %s", err)
|
||||
}
|
||||
|
||||
clone, err := opts.Program.Clone()
|
||||
if err != nil {
|
||||
cgroup.Close()
|
||||
return nil, err
|
||||
}
|
||||
|
||||
var cg Link
|
||||
cg, err = newLinkCgroup(cgroup, opts.Attach, clone)
|
||||
if errors.Is(err, ErrNotSupported) {
|
||||
cg, err = newProgAttachCgroup(cgroup, opts.Attach, clone, flagAllowMulti)
|
||||
}
|
||||
if errors.Is(err, ErrNotSupported) {
|
||||
cg, err = newProgAttachCgroup(cgroup, opts.Attach, clone, flagAllowOverride)
|
||||
}
|
||||
if err != nil {
|
||||
cgroup.Close()
|
||||
clone.Close()
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return cg, nil
|
||||
}
|
||||
|
||||
// LoadPinnedCgroup loads a pinned cgroup from a bpffs.
|
||||
func LoadPinnedCgroup(fileName string, opts *ebpf.LoadPinOptions) (Link, error) {
|
||||
link, err := LoadPinnedRawLink(fileName, CgroupType, opts)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return &linkCgroup{*link}, nil
|
||||
}
|
||||
|
||||
type progAttachCgroup struct {
|
||||
cgroup *os.File
|
||||
current *ebpf.Program
|
||||
attachType ebpf.AttachType
|
||||
flags cgroupAttachFlags
|
||||
}
|
||||
|
||||
var _ Link = (*progAttachCgroup)(nil)
|
||||
|
||||
func (cg *progAttachCgroup) isLink() {}
|
||||
|
||||
func newProgAttachCgroup(cgroup *os.File, attach ebpf.AttachType, prog *ebpf.Program, flags cgroupAttachFlags) (*progAttachCgroup, error) {
|
||||
if flags&flagAllowMulti > 0 {
|
||||
if err := haveProgAttachReplace(); err != nil {
|
||||
return nil, fmt.Errorf("can't support multiple programs: %w", err)
|
||||
}
|
||||
}
|
||||
|
||||
err := RawAttachProgram(RawAttachProgramOptions{
|
||||
Target: int(cgroup.Fd()),
|
||||
Program: prog,
|
||||
Flags: uint32(flags),
|
||||
Attach: attach,
|
||||
})
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("cgroup: %w", err)
|
||||
}
|
||||
|
||||
return &progAttachCgroup{cgroup, prog, attach, flags}, nil
|
||||
}
|
||||
|
||||
func (cg *progAttachCgroup) Close() error {
|
||||
defer cg.cgroup.Close()
|
||||
defer cg.current.Close()
|
||||
|
||||
err := RawDetachProgram(RawDetachProgramOptions{
|
||||
Target: int(cg.cgroup.Fd()),
|
||||
Program: cg.current,
|
||||
Attach: cg.attachType,
|
||||
})
|
||||
if err != nil {
|
||||
return fmt.Errorf("close cgroup: %s", err)
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
func (cg *progAttachCgroup) Update(prog *ebpf.Program) error {
|
||||
new, err := prog.Clone()
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
args := RawAttachProgramOptions{
|
||||
Target: int(cg.cgroup.Fd()),
|
||||
Program: prog,
|
||||
Attach: cg.attachType,
|
||||
Flags: uint32(cg.flags),
|
||||
}
|
||||
|
||||
if cg.flags&flagAllowMulti > 0 {
|
||||
// Atomically replacing multiple programs requires at least
|
||||
// 5.5 (commit 7dd68b3279f17921 "bpf: Support replacing cgroup-bpf
|
||||
// program in MULTI mode")
|
||||
args.Flags |= uint32(flagReplace)
|
||||
args.Replace = cg.current
|
||||
}
|
||||
|
||||
if err := RawAttachProgram(args); err != nil {
|
||||
new.Close()
|
||||
return fmt.Errorf("can't update cgroup: %s", err)
|
||||
}
|
||||
|
||||
cg.current.Close()
|
||||
cg.current = new
|
||||
return nil
|
||||
}
|
||||
|
||||
func (cg *progAttachCgroup) Pin(string) error {
|
||||
return fmt.Errorf("can't pin cgroup: %w", ErrNotSupported)
|
||||
}
|
||||
|
||||
func (cg *progAttachCgroup) Unpin() error {
|
||||
return fmt.Errorf("can't pin cgroup: %w", ErrNotSupported)
|
||||
}
|
||||
|
||||
type linkCgroup struct {
|
||||
RawLink
|
||||
}
|
||||
|
||||
var _ Link = (*linkCgroup)(nil)
|
||||
|
||||
func newLinkCgroup(cgroup *os.File, attach ebpf.AttachType, prog *ebpf.Program) (*linkCgroup, error) {
|
||||
link, err := AttachRawLink(RawLinkOptions{
|
||||
Target: int(cgroup.Fd()),
|
||||
Program: prog,
|
||||
Attach: attach,
|
||||
})
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return &linkCgroup{*link}, err
|
||||
}
|
||||
+2
@@ -0,0 +1,2 @@
|
||||
// Package link allows attaching eBPF programs to various kernel hooks.
|
||||
package link
|
||||
+88
@@ -0,0 +1,88 @@
|
||||
package link
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
|
||||
"github.com/cilium/ebpf"
|
||||
"github.com/cilium/ebpf/internal/btf"
|
||||
)
|
||||
|
||||
type FreplaceLink struct {
|
||||
RawLink
|
||||
}
|
||||
|
||||
// AttachFreplace attaches the given eBPF program to the function it replaces.
|
||||
//
|
||||
// The program and name can either be provided at link time, or can be provided
|
||||
// at program load time. If they were provided at load time, they should be nil
|
||||
// and empty respectively here, as they will be ignored by the kernel.
|
||||
// Examples:
|
||||
//
|
||||
// AttachFreplace(dispatcher, "function", replacement)
|
||||
// AttachFreplace(nil, "", replacement)
|
||||
func AttachFreplace(targetProg *ebpf.Program, name string, prog *ebpf.Program) (*FreplaceLink, error) {
|
||||
if (name == "") != (targetProg == nil) {
|
||||
return nil, fmt.Errorf("must provide both or neither of name and targetProg: %w", errInvalidInput)
|
||||
}
|
||||
if prog == nil {
|
||||
return nil, fmt.Errorf("prog cannot be nil: %w", errInvalidInput)
|
||||
}
|
||||
if prog.Type() != ebpf.Extension {
|
||||
return nil, fmt.Errorf("eBPF program type %s is not an Extension: %w", prog.Type(), errInvalidInput)
|
||||
}
|
||||
|
||||
var (
|
||||
target int
|
||||
typeID btf.TypeID
|
||||
)
|
||||
if targetProg != nil {
|
||||
info, err := targetProg.Info()
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
btfID, ok := info.BTFID()
|
||||
if !ok {
|
||||
return nil, fmt.Errorf("could not get BTF ID for program %s: %w", info.Name, errInvalidInput)
|
||||
}
|
||||
btfHandle, err := btf.NewHandleFromID(btfID)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
defer btfHandle.Close()
|
||||
|
||||
var function *btf.Func
|
||||
if err := btfHandle.Spec().FindType(name, &function); err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
target = targetProg.FD()
|
||||
typeID = function.ID()
|
||||
}
|
||||
|
||||
link, err := AttachRawLink(RawLinkOptions{
|
||||
Target: target,
|
||||
Program: prog,
|
||||
Attach: ebpf.AttachNone,
|
||||
BTF: typeID,
|
||||
})
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return &FreplaceLink{*link}, nil
|
||||
}
|
||||
|
||||
// Update implements the Link interface.
|
||||
func (f *FreplaceLink) Update(new *ebpf.Program) error {
|
||||
return fmt.Errorf("freplace update: %w", ErrNotSupported)
|
||||
}
|
||||
|
||||
// LoadPinnedFreplace loads a pinned iterator from a bpffs.
|
||||
func LoadPinnedFreplace(fileName string, opts *ebpf.LoadPinOptions) (*FreplaceLink, error) {
|
||||
link, err := LoadPinnedRawLink(fileName, TracingType, opts)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return &FreplaceLink{*link}, err
|
||||
}
|
||||
+100
@@ -0,0 +1,100 @@
|
||||
package link
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"io"
|
||||
"unsafe"
|
||||
|
||||
"github.com/cilium/ebpf"
|
||||
"github.com/cilium/ebpf/internal"
|
||||
)
|
||||
|
||||
type IterOptions struct {
|
||||
// Program must be of type Tracing with attach type
|
||||
// AttachTraceIter. The kind of iterator to attach to is
|
||||
// determined at load time via the AttachTo field.
|
||||
//
|
||||
// AttachTo requires the kernel to include BTF of itself,
|
||||
// and it to be compiled with a recent pahole (>= 1.16).
|
||||
Program *ebpf.Program
|
||||
|
||||
// Map specifies the target map for bpf_map_elem and sockmap iterators.
|
||||
// It may be nil.
|
||||
Map *ebpf.Map
|
||||
}
|
||||
|
||||
// AttachIter attaches a BPF seq_file iterator.
|
||||
func AttachIter(opts IterOptions) (*Iter, error) {
|
||||
if err := haveBPFLink(); err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
progFd := opts.Program.FD()
|
||||
if progFd < 0 {
|
||||
return nil, fmt.Errorf("invalid program: %s", internal.ErrClosedFd)
|
||||
}
|
||||
|
||||
var info bpfIterLinkInfoMap
|
||||
if opts.Map != nil {
|
||||
mapFd := opts.Map.FD()
|
||||
if mapFd < 0 {
|
||||
return nil, fmt.Errorf("invalid map: %w", internal.ErrClosedFd)
|
||||
}
|
||||
info.map_fd = uint32(mapFd)
|
||||
}
|
||||
|
||||
attr := bpfLinkCreateIterAttr{
|
||||
prog_fd: uint32(progFd),
|
||||
attach_type: ebpf.AttachTraceIter,
|
||||
iter_info: internal.NewPointer(unsafe.Pointer(&info)),
|
||||
iter_info_len: uint32(unsafe.Sizeof(info)),
|
||||
}
|
||||
|
||||
fd, err := bpfLinkCreateIter(&attr)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("can't link iterator: %w", err)
|
||||
}
|
||||
|
||||
return &Iter{RawLink{fd, ""}}, err
|
||||
}
|
||||
|
||||
// LoadPinnedIter loads a pinned iterator from a bpffs.
|
||||
func LoadPinnedIter(fileName string, opts *ebpf.LoadPinOptions) (*Iter, error) {
|
||||
link, err := LoadPinnedRawLink(fileName, IterType, opts)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return &Iter{*link}, err
|
||||
}
|
||||
|
||||
// Iter represents an attached bpf_iter.
|
||||
type Iter struct {
|
||||
RawLink
|
||||
}
|
||||
|
||||
// Open creates a new instance of the iterator.
|
||||
//
|
||||
// Reading from the returned reader triggers the BPF program.
|
||||
func (it *Iter) Open() (io.ReadCloser, error) {
|
||||
linkFd, err := it.fd.Value()
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
attr := &bpfIterCreateAttr{
|
||||
linkFd: linkFd,
|
||||
}
|
||||
|
||||
fd, err := bpfIterCreate(attr)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("can't create iterator: %w", err)
|
||||
}
|
||||
|
||||
return fd.File("bpf_iter"), nil
|
||||
}
|
||||
|
||||
// union bpf_iter_link_info.map
|
||||
type bpfIterLinkInfoMap struct {
|
||||
map_fd uint32
|
||||
}
|
||||
+444
@@ -0,0 +1,444 @@
|
||||
package link
|
||||
|
||||
import (
|
||||
"bytes"
|
||||
"crypto/rand"
|
||||
"errors"
|
||||
"fmt"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"runtime"
|
||||
"sync"
|
||||
"unsafe"
|
||||
|
||||
"github.com/cilium/ebpf"
|
||||
"github.com/cilium/ebpf/internal"
|
||||
"github.com/cilium/ebpf/internal/unix"
|
||||
)
|
||||
|
||||
var (
|
||||
kprobeEventsPath = filepath.Join(tracefsPath, "kprobe_events")
|
||||
|
||||
kprobeRetprobeBit = struct {
|
||||
once sync.Once
|
||||
value uint64
|
||||
err error
|
||||
}{}
|
||||
)
|
||||
|
||||
type probeType uint8
|
||||
|
||||
const (
|
||||
kprobeType probeType = iota
|
||||
uprobeType
|
||||
)
|
||||
|
||||
func (pt probeType) String() string {
|
||||
if pt == kprobeType {
|
||||
return "kprobe"
|
||||
}
|
||||
return "uprobe"
|
||||
}
|
||||
|
||||
func (pt probeType) EventsPath() string {
|
||||
if pt == kprobeType {
|
||||
return kprobeEventsPath
|
||||
}
|
||||
return uprobeEventsPath
|
||||
}
|
||||
|
||||
func (pt probeType) PerfEventType(ret bool) perfEventType {
|
||||
if pt == kprobeType {
|
||||
if ret {
|
||||
return kretprobeEvent
|
||||
}
|
||||
return kprobeEvent
|
||||
}
|
||||
if ret {
|
||||
return uretprobeEvent
|
||||
}
|
||||
return uprobeEvent
|
||||
}
|
||||
|
||||
func (pt probeType) RetprobeBit() (uint64, error) {
|
||||
if pt == kprobeType {
|
||||
return kretprobeBit()
|
||||
}
|
||||
return uretprobeBit()
|
||||
}
|
||||
|
||||
// Kprobe attaches the given eBPF program to a perf event that fires when the
|
||||
// given kernel symbol starts executing. See /proc/kallsyms for available
|
||||
// symbols. For example, printk():
|
||||
//
|
||||
// kp, err := Kprobe("printk", prog)
|
||||
//
|
||||
// Losing the reference to the resulting Link (kp) will close the Kprobe
|
||||
// and prevent further execution of prog. The Link must be Closed during
|
||||
// program shutdown to avoid leaking system resources.
|
||||
func Kprobe(symbol string, prog *ebpf.Program) (Link, error) {
|
||||
k, err := kprobe(symbol, prog, false)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
err = k.attach(prog)
|
||||
if err != nil {
|
||||
k.Close()
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return k, nil
|
||||
}
|
||||
|
||||
// Kretprobe attaches the given eBPF program to a perf event that fires right
|
||||
// before the given kernel symbol exits, with the function stack left intact.
|
||||
// See /proc/kallsyms for available symbols. For example, printk():
|
||||
//
|
||||
// kp, err := Kretprobe("printk", prog)
|
||||
//
|
||||
// Losing the reference to the resulting Link (kp) will close the Kretprobe
|
||||
// and prevent further execution of prog. The Link must be Closed during
|
||||
// program shutdown to avoid leaking system resources.
|
||||
func Kretprobe(symbol string, prog *ebpf.Program) (Link, error) {
|
||||
k, err := kprobe(symbol, prog, true)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
err = k.attach(prog)
|
||||
if err != nil {
|
||||
k.Close()
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return k, nil
|
||||
}
|
||||
|
||||
// kprobe opens a perf event on the given symbol and attaches prog to it.
|
||||
// If ret is true, create a kretprobe.
|
||||
func kprobe(symbol string, prog *ebpf.Program, ret bool) (*perfEvent, error) {
|
||||
if symbol == "" {
|
||||
return nil, fmt.Errorf("symbol name cannot be empty: %w", errInvalidInput)
|
||||
}
|
||||
if prog == nil {
|
||||
return nil, fmt.Errorf("prog cannot be nil: %w", errInvalidInput)
|
||||
}
|
||||
if !rgxTraceEvent.MatchString(symbol) {
|
||||
return nil, fmt.Errorf("symbol '%s' must be alphanumeric or underscore: %w", symbol, errInvalidInput)
|
||||
}
|
||||
if prog.Type() != ebpf.Kprobe {
|
||||
return nil, fmt.Errorf("eBPF program type %s is not a Kprobe: %w", prog.Type(), errInvalidInput)
|
||||
}
|
||||
|
||||
// Use kprobe PMU if the kernel has it available.
|
||||
tp, err := pmuKprobe(platformPrefix(symbol), ret)
|
||||
if errors.Is(err, os.ErrNotExist) {
|
||||
tp, err = pmuKprobe(symbol, ret)
|
||||
}
|
||||
if err == nil {
|
||||
return tp, nil
|
||||
}
|
||||
if err != nil && !errors.Is(err, ErrNotSupported) {
|
||||
return nil, fmt.Errorf("creating perf_kprobe PMU: %w", err)
|
||||
}
|
||||
|
||||
// Use tracefs if kprobe PMU is missing.
|
||||
tp, err = tracefsKprobe(platformPrefix(symbol), ret)
|
||||
if errors.Is(err, os.ErrNotExist) {
|
||||
tp, err = tracefsKprobe(symbol, ret)
|
||||
}
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("creating trace event '%s' in tracefs: %w", symbol, err)
|
||||
}
|
||||
|
||||
return tp, nil
|
||||
}
|
||||
|
||||
// pmuKprobe opens a perf event based on the kprobe PMU.
|
||||
// Returns os.ErrNotExist if the given symbol does not exist in the kernel.
|
||||
func pmuKprobe(symbol string, ret bool) (*perfEvent, error) {
|
||||
return pmuProbe(kprobeType, symbol, "", 0, perfAllThreads, ret)
|
||||
}
|
||||
|
||||
// pmuProbe opens a perf event based on a Performance Monitoring Unit.
|
||||
//
|
||||
// Requires at least a 4.17 kernel.
|
||||
// e12f03d7031a "perf/core: Implement the 'perf_kprobe' PMU"
|
||||
// 33ea4b24277b "perf/core: Implement the 'perf_uprobe' PMU"
|
||||
//
|
||||
// Returns ErrNotSupported if the kernel doesn't support perf_[k,u]probe PMU
|
||||
func pmuProbe(typ probeType, symbol, path string, offset uint64, pid int, ret bool) (*perfEvent, error) {
|
||||
// Getting the PMU type will fail if the kernel doesn't support
|
||||
// the perf_[k,u]probe PMU.
|
||||
et, err := getPMUEventType(typ)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
var config uint64
|
||||
if ret {
|
||||
bit, err := typ.RetprobeBit()
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
config |= 1 << bit
|
||||
}
|
||||
|
||||
var (
|
||||
attr unix.PerfEventAttr
|
||||
sp unsafe.Pointer
|
||||
)
|
||||
switch typ {
|
||||
case kprobeType:
|
||||
// Create a pointer to a NUL-terminated string for the kernel.
|
||||
sp, err = unsafeStringPtr(symbol)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
attr = unix.PerfEventAttr{
|
||||
Type: uint32(et), // PMU event type read from sysfs
|
||||
Ext1: uint64(uintptr(sp)), // Kernel symbol to trace
|
||||
Config: config, // Retprobe flag
|
||||
}
|
||||
case uprobeType:
|
||||
sp, err = unsafeStringPtr(path)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
attr = unix.PerfEventAttr{
|
||||
// The minimum size required for PMU uprobes is PERF_ATTR_SIZE_VER1,
|
||||
// since it added the config2 (Ext2) field. The Size field controls the
|
||||
// size of the internal buffer the kernel allocates for reading the
|
||||
// perf_event_attr argument from userspace.
|
||||
Size: unix.PERF_ATTR_SIZE_VER1,
|
||||
Type: uint32(et), // PMU event type read from sysfs
|
||||
Ext1: uint64(uintptr(sp)), // Uprobe path
|
||||
Ext2: offset, // Uprobe offset
|
||||
Config: config, // Retprobe flag
|
||||
}
|
||||
}
|
||||
|
||||
fd, err := unix.PerfEventOpen(&attr, pid, 0, -1, unix.PERF_FLAG_FD_CLOEXEC)
|
||||
|
||||
// Since commit 97c753e62e6c, ENOENT is correctly returned instead of EINVAL
|
||||
// when trying to create a kretprobe for a missing symbol. Make sure ENOENT
|
||||
// is returned to the caller.
|
||||
if errors.Is(err, os.ErrNotExist) || errors.Is(err, unix.EINVAL) {
|
||||
return nil, fmt.Errorf("symbol '%s' not found: %w", symbol, os.ErrNotExist)
|
||||
}
|
||||
// Since at least commit cb9a19fe4aa51, ENOTSUPP is returned
|
||||
// when attempting to set a uprobe on a trap instruction.
|
||||
if errors.Is(err, unix.ENOTSUPP) {
|
||||
return nil, fmt.Errorf("failed setting uprobe on offset %#x (possible trap insn): %w", offset, err)
|
||||
}
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("opening perf event: %w", err)
|
||||
}
|
||||
|
||||
// Ensure the string pointer is not collected before PerfEventOpen returns.
|
||||
runtime.KeepAlive(sp)
|
||||
|
||||
// Kernel has perf_[k,u]probe PMU available, initialize perf event.
|
||||
return &perfEvent{
|
||||
fd: internal.NewFD(uint32(fd)),
|
||||
pmuID: et,
|
||||
name: symbol,
|
||||
typ: typ.PerfEventType(ret),
|
||||
}, nil
|
||||
}
|
||||
|
||||
// tracefsKprobe creates a Kprobe tracefs entry.
|
||||
func tracefsKprobe(symbol string, ret bool) (*perfEvent, error) {
|
||||
return tracefsProbe(kprobeType, symbol, "", 0, perfAllThreads, ret)
|
||||
}
|
||||
|
||||
// tracefsProbe creates a trace event by writing an entry to <tracefs>/[k,u]probe_events.
|
||||
// A new trace event group name is generated on every call to support creating
|
||||
// multiple trace events for the same kernel or userspace symbol.
|
||||
// Path and offset are only set in the case of uprobe(s) and are used to set
|
||||
// the executable/library path on the filesystem and the offset where the probe is inserted.
|
||||
// A perf event is then opened on the newly-created trace event and returned to the caller.
|
||||
func tracefsProbe(typ probeType, symbol, path string, offset uint64, pid int, ret bool) (*perfEvent, error) {
|
||||
// Generate a random string for each trace event we attempt to create.
|
||||
// This value is used as the 'group' token in tracefs to allow creating
|
||||
// multiple kprobe trace events with the same name.
|
||||
group, err := randomGroup("ebpf")
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("randomizing group name: %w", err)
|
||||
}
|
||||
|
||||
// Before attempting to create a trace event through tracefs,
|
||||
// check if an event with the same group and name already exists.
|
||||
// Kernels 4.x and earlier don't return os.ErrExist on writing a duplicate
|
||||
// entry, so we need to rely on reads for detecting uniqueness.
|
||||
_, err = getTraceEventID(group, symbol)
|
||||
if err == nil {
|
||||
return nil, fmt.Errorf("trace event already exists: %s/%s", group, symbol)
|
||||
}
|
||||
if err != nil && !errors.Is(err, os.ErrNotExist) {
|
||||
return nil, fmt.Errorf("checking trace event %s/%s: %w", group, symbol, err)
|
||||
}
|
||||
|
||||
// Create the [k,u]probe trace event using tracefs.
|
||||
if err := createTraceFSProbeEvent(typ, group, symbol, path, offset, ret); err != nil {
|
||||
return nil, fmt.Errorf("creating probe entry on tracefs: %w", err)
|
||||
}
|
||||
|
||||
// Get the newly-created trace event's id.
|
||||
tid, err := getTraceEventID(group, symbol)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("getting trace event id: %w", err)
|
||||
}
|
||||
|
||||
// Kprobes are ephemeral tracepoints and share the same perf event type.
|
||||
fd, err := openTracepointPerfEvent(tid, pid)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return &perfEvent{
|
||||
fd: fd,
|
||||
group: group,
|
||||
name: symbol,
|
||||
tracefsID: tid,
|
||||
typ: typ.PerfEventType(ret),
|
||||
}, nil
|
||||
}
|
||||
|
||||
// createTraceFSProbeEvent creates a new ephemeral trace event by writing to
|
||||
// <tracefs>/[k,u]probe_events. Returns os.ErrNotExist if symbol is not a valid
|
||||
// kernel symbol, or if it is not traceable with kprobes. Returns os.ErrExist
|
||||
// if a probe with the same group and symbol already exists.
|
||||
func createTraceFSProbeEvent(typ probeType, group, symbol, path string, offset uint64, ret bool) error {
|
||||
// Open the kprobe_events file in tracefs.
|
||||
f, err := os.OpenFile(typ.EventsPath(), os.O_APPEND|os.O_WRONLY, 0666)
|
||||
if err != nil {
|
||||
return fmt.Errorf("error opening '%s': %w", typ.EventsPath(), err)
|
||||
}
|
||||
defer f.Close()
|
||||
|
||||
var pe string
|
||||
switch typ {
|
||||
case kprobeType:
|
||||
// The kprobe_events syntax is as follows (see Documentation/trace/kprobetrace.txt):
|
||||
// p[:[GRP/]EVENT] [MOD:]SYM[+offs]|MEMADDR [FETCHARGS] : Set a probe
|
||||
// r[MAXACTIVE][:[GRP/]EVENT] [MOD:]SYM[+0] [FETCHARGS] : Set a return probe
|
||||
// -:[GRP/]EVENT : Clear a probe
|
||||
//
|
||||
// Some examples:
|
||||
// r:ebpf_1234/r_my_kretprobe nf_conntrack_destroy
|
||||
// p:ebpf_5678/p_my_kprobe __x64_sys_execve
|
||||
//
|
||||
// Leaving the kretprobe's MAXACTIVE set to 0 (or absent) will make the
|
||||
// kernel default to NR_CPUS. This is desired in most eBPF cases since
|
||||
// subsampling or rate limiting logic can be more accurately implemented in
|
||||
// the eBPF program itself.
|
||||
// See Documentation/kprobes.txt for more details.
|
||||
pe = fmt.Sprintf("%s:%s/%s %s", probePrefix(ret), group, symbol, symbol)
|
||||
case uprobeType:
|
||||
// The uprobe_events syntax is as follows:
|
||||
// p[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS] : Set a probe
|
||||
// r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS] : Set a return probe
|
||||
// -:[GRP/]EVENT : Clear a probe
|
||||
//
|
||||
// Some examples:
|
||||
// r:ebpf_1234/readline /bin/bash:0x12345
|
||||
// p:ebpf_5678/main_mySymbol /bin/mybin:0x12345
|
||||
//
|
||||
// See Documentation/trace/uprobetracer.txt for more details.
|
||||
pathOffset := uprobePathOffset(path, offset)
|
||||
pe = fmt.Sprintf("%s:%s/%s %s", probePrefix(ret), group, symbol, pathOffset)
|
||||
}
|
||||
_, err = f.WriteString(pe)
|
||||
// Since commit 97c753e62e6c, ENOENT is correctly returned instead of EINVAL
|
||||
// when trying to create a kretprobe for a missing symbol. Make sure ENOENT
|
||||
// is returned to the caller.
|
||||
if errors.Is(err, os.ErrNotExist) || errors.Is(err, unix.EINVAL) {
|
||||
return fmt.Errorf("symbol %s not found: %w", symbol, os.ErrNotExist)
|
||||
}
|
||||
if err != nil {
|
||||
return fmt.Errorf("writing '%s' to '%s': %w", pe, typ.EventsPath(), err)
|
||||
}
|
||||
|
||||
return nil
|
||||
}
|
||||
|
||||
// closeTraceFSProbeEvent removes the [k,u]probe with the given type, group and symbol
|
||||
// from <tracefs>/[k,u]probe_events.
|
||||
func closeTraceFSProbeEvent(typ probeType, group, symbol string) error {
|
||||
f, err := os.OpenFile(typ.EventsPath(), os.O_APPEND|os.O_WRONLY, 0666)
|
||||
if err != nil {
|
||||
return fmt.Errorf("error opening %s: %w", typ.EventsPath(), err)
|
||||
}
|
||||
defer f.Close()
|
||||
|
||||
// See [k,u]probe_events syntax above. The probe type does not need to be specified
|
||||
// for removals.
|
||||
pe := fmt.Sprintf("-:%s/%s", group, symbol)
|
||||
if _, err = f.WriteString(pe); err != nil {
|
||||
return fmt.Errorf("writing '%s' to '%s': %w", pe, typ.EventsPath(), err)
|
||||
}
|
||||
|
||||
return nil
|
||||
}
|
||||
|
||||
// randomGroup generates a pseudorandom string for use as a tracefs group name.
|
||||
// Returns an error when the output string would exceed 63 characters (kernel
|
||||
// limitation), when rand.Read() fails or when prefix contains characters not
|
||||
// allowed by rgxTraceEvent.
|
||||
func randomGroup(prefix string) (string, error) {
|
||||
if !rgxTraceEvent.MatchString(prefix) {
|
||||
return "", fmt.Errorf("prefix '%s' must be alphanumeric or underscore: %w", prefix, errInvalidInput)
|
||||
}
|
||||
|
||||
b := make([]byte, 8)
|
||||
if _, err := rand.Read(b); err != nil {
|
||||
return "", fmt.Errorf("reading random bytes: %w", err)
|
||||
}
|
||||
|
||||
group := fmt.Sprintf("%s_%x", prefix, b)
|
||||
if len(group) > 63 {
|
||||
return "", fmt.Errorf("group name '%s' cannot be longer than 63 characters: %w", group, errInvalidInput)
|
||||
}
|
||||
|
||||
return group, nil
|
||||
}
|
||||
|
||||
func probePrefix(ret bool) string {
|
||||
if ret {
|
||||
return "r"
|
||||
}
|
||||
return "p"
|
||||
}
|
||||
|
||||
// determineRetprobeBit reads a Performance Monitoring Unit's retprobe bit
|
||||
// from /sys/bus/event_source/devices/<pmu>/format/retprobe.
|
||||
func determineRetprobeBit(typ probeType) (uint64, error) {
|
||||
p := filepath.Join("/sys/bus/event_source/devices/", typ.String(), "/format/retprobe")
|
||||
|
||||
data, err := os.ReadFile(p)
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
|
||||
var rp uint64
|
||||
n, err := fmt.Sscanf(string(bytes.TrimSpace(data)), "config:%d", &rp)
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("parse retprobe bit: %w", err)
|
||||
}
|
||||
if n != 1 {
|
||||
return 0, fmt.Errorf("parse retprobe bit: expected 1 item, got %d", n)
|
||||
}
|
||||
|
||||
return rp, nil
|
||||
}
|
||||
|
||||
func kretprobeBit() (uint64, error) {
|
||||
kprobeRetprobeBit.once.Do(func() {
|
||||
kprobeRetprobeBit.value, kprobeRetprobeBit.err = determineRetprobeBit(kprobeType)
|
||||
})
|
||||
return kprobeRetprobeBit.value, kprobeRetprobeBit.err
|
||||
}
|
||||
+233
@@ -0,0 +1,233 @@
|
||||
package link
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"unsafe"
|
||||
|
||||
"github.com/cilium/ebpf"
|
||||
"github.com/cilium/ebpf/internal"
|
||||
"github.com/cilium/ebpf/internal/btf"
|
||||
)
|
||||
|
||||
var ErrNotSupported = internal.ErrNotSupported
|
||||
|
||||
// Link represents a Program attached to a BPF hook.
|
||||
type Link interface {
|
||||
// Replace the current program with a new program.
|
||||
//
|
||||
// Passing a nil program is an error. May return an error wrapping ErrNotSupported.
|
||||
Update(*ebpf.Program) error
|
||||
|
||||
// Persist a link by pinning it into a bpffs.
|
||||
//
|
||||
// May return an error wrapping ErrNotSupported.
|
||||
Pin(string) error
|
||||
|
||||
// Undo a previous call to Pin.
|
||||
//
|
||||
// May return an error wrapping ErrNotSupported.
|
||||
Unpin() error
|
||||
|
||||
// Close frees resources.
|
||||
//
|
||||
// The link will be broken unless it has been successfully pinned.
|
||||
// A link may continue past the lifetime of the process if Close is
|
||||
// not called.
|
||||
Close() error
|
||||
|
||||
// Prevent external users from implementing this interface.
|
||||
isLink()
|
||||
}
|
||||
|
||||
// ID uniquely identifies a BPF link.
|
||||
type ID uint32
|
||||
|
||||
// RawLinkOptions control the creation of a raw link.
|
||||
type RawLinkOptions struct {
|
||||
// File descriptor to attach to. This differs for each attach type.
|
||||
Target int
|
||||
// Program to attach.
|
||||
Program *ebpf.Program
|
||||
// Attach must match the attach type of Program.
|
||||
Attach ebpf.AttachType
|
||||
// BTF is the BTF of the attachment target.
|
||||
BTF btf.TypeID
|
||||
}
|
||||
|
||||
// RawLinkInfo contains metadata on a link.
|
||||
type RawLinkInfo struct {
|
||||
Type Type
|
||||
ID ID
|
||||
Program ebpf.ProgramID
|
||||
}
|
||||
|
||||
// RawLink is the low-level API to bpf_link.
|
||||
//
|
||||
// You should consider using the higher level interfaces in this
|
||||
// package instead.
|
||||
type RawLink struct {
|
||||
fd *internal.FD
|
||||
pinnedPath string
|
||||
}
|
||||
|
||||
// AttachRawLink creates a raw link.
|
||||
func AttachRawLink(opts RawLinkOptions) (*RawLink, error) {
|
||||
if err := haveBPFLink(); err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
if opts.Target < 0 {
|
||||
return nil, fmt.Errorf("invalid target: %s", internal.ErrClosedFd)
|
||||
}
|
||||
|
||||
progFd := opts.Program.FD()
|
||||
if progFd < 0 {
|
||||
return nil, fmt.Errorf("invalid program: %s", internal.ErrClosedFd)
|
||||
}
|
||||
|
||||
attr := bpfLinkCreateAttr{
|
||||
targetFd: uint32(opts.Target),
|
||||
progFd: uint32(progFd),
|
||||
attachType: opts.Attach,
|
||||
targetBTFID: uint32(opts.BTF),
|
||||
}
|
||||
fd, err := bpfLinkCreate(&attr)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("can't create link: %s", err)
|
||||
}
|
||||
|
||||
return &RawLink{fd, ""}, nil
|
||||
}
|
||||
|
||||
// LoadPinnedRawLink loads a persisted link from a bpffs.
|
||||
//
|
||||
// Returns an error if the pinned link type doesn't match linkType. Pass
|
||||
// UnspecifiedType to disable this behaviour.
|
||||
func LoadPinnedRawLink(fileName string, linkType Type, opts *ebpf.LoadPinOptions) (*RawLink, error) {
|
||||
fd, err := internal.BPFObjGet(fileName, opts.Marshal())
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("load pinned link: %w", err)
|
||||
}
|
||||
|
||||
link := &RawLink{fd, fileName}
|
||||
if linkType == UnspecifiedType {
|
||||
return link, nil
|
||||
}
|
||||
|
||||
info, err := link.Info()
|
||||
if err != nil {
|
||||
link.Close()
|
||||
return nil, fmt.Errorf("get pinned link info: %s", err)
|
||||
}
|
||||
|
||||
if info.Type != linkType {
|
||||
link.Close()
|
||||
return nil, fmt.Errorf("link type %v doesn't match %v", info.Type, linkType)
|
||||
}
|
||||
|
||||
return link, nil
|
||||
}
|
||||
|
||||
func (l *RawLink) isLink() {}
|
||||
|
||||
// FD returns the raw file descriptor.
|
||||
func (l *RawLink) FD() int {
|
||||
fd, err := l.fd.Value()
|
||||
if err != nil {
|
||||
return -1
|
||||
}
|
||||
return int(fd)
|
||||
}
|
||||
|
||||
// Close breaks the link.
|
||||
//
|
||||
// Use Pin if you want to make the link persistent.
|
||||
func (l *RawLink) Close() error {
|
||||
return l.fd.Close()
|
||||
}
|
||||
|
||||
// Pin persists a link past the lifetime of the process.
|
||||
//
|
||||
// Calling Close on a pinned Link will not break the link
|
||||
// until the pin is removed.
|
||||
func (l *RawLink) Pin(fileName string) error {
|
||||
if err := internal.Pin(l.pinnedPath, fileName, l.fd); err != nil {
|
||||
return err
|
||||
}
|
||||
l.pinnedPath = fileName
|
||||
return nil
|
||||
}
|
||||
|
||||
// Unpin implements the Link interface.
|
||||
func (l *RawLink) Unpin() error {
|
||||
if err := internal.Unpin(l.pinnedPath); err != nil {
|
||||
return err
|
||||
}
|
||||
l.pinnedPath = ""
|
||||
return nil
|
||||
}
|
||||
|
||||
// Update implements the Link interface.
|
||||
func (l *RawLink) Update(new *ebpf.Program) error {
|
||||
return l.UpdateArgs(RawLinkUpdateOptions{
|
||||
New: new,
|
||||
})
|
||||
}
|
||||
|
||||
// RawLinkUpdateOptions control the behaviour of RawLink.UpdateArgs.
|
||||
type RawLinkUpdateOptions struct {
|
||||
New *ebpf.Program
|
||||
Old *ebpf.Program
|
||||
Flags uint32
|
||||
}
|
||||
|
||||
// UpdateArgs updates a link based on args.
|
||||
func (l *RawLink) UpdateArgs(opts RawLinkUpdateOptions) error {
|
||||
newFd := opts.New.FD()
|
||||
if newFd < 0 {
|
||||
return fmt.Errorf("invalid program: %s", internal.ErrClosedFd)
|
||||
}
|
||||
|
||||
var oldFd int
|
||||
if opts.Old != nil {
|
||||
oldFd = opts.Old.FD()
|
||||
if oldFd < 0 {
|
||||
return fmt.Errorf("invalid replacement program: %s", internal.ErrClosedFd)
|
||||
}
|
||||
}
|
||||
|
||||
linkFd, err := l.fd.Value()
|
||||
if err != nil {
|
||||
return fmt.Errorf("can't update link: %s", err)
|
||||
}
|
||||
|
||||
attr := bpfLinkUpdateAttr{
|
||||
linkFd: linkFd,
|
||||
newProgFd: uint32(newFd),
|
||||
oldProgFd: uint32(oldFd),
|
||||
flags: opts.Flags,
|
||||
}
|
||||
return bpfLinkUpdate(&attr)
|
||||
}
|
||||
|
||||
// struct bpf_link_info
|
||||
type bpfLinkInfo struct {
|
||||
typ uint32
|
||||
id uint32
|
||||
prog_id uint32
|
||||
}
|
||||
|
||||
// Info returns metadata about the link.
|
||||
func (l *RawLink) Info() (*RawLinkInfo, error) {
|
||||
var info bpfLinkInfo
|
||||
err := internal.BPFObjGetInfoByFD(l.fd, unsafe.Pointer(&info), unsafe.Sizeof(info))
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("link info: %s", err)
|
||||
}
|
||||
|
||||
return &RawLinkInfo{
|
||||
Type(info.typ),
|
||||
ID(info.id),
|
||||
ebpf.ProgramID(info.prog_id),
|
||||
}, nil
|
||||
}
|
||||
+60
@@ -0,0 +1,60 @@
|
||||
package link
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
|
||||
"github.com/cilium/ebpf"
|
||||
)
|
||||
|
||||
// NetNsInfo contains metadata about a network namespace link.
|
||||
type NetNsInfo struct {
|
||||
RawLinkInfo
|
||||
}
|
||||
|
||||
// NetNsLink is a program attached to a network namespace.
|
||||
type NetNsLink struct {
|
||||
*RawLink
|
||||
}
|
||||
|
||||
// AttachNetNs attaches a program to a network namespace.
|
||||
func AttachNetNs(ns int, prog *ebpf.Program) (*NetNsLink, error) {
|
||||
var attach ebpf.AttachType
|
||||
switch t := prog.Type(); t {
|
||||
case ebpf.FlowDissector:
|
||||
attach = ebpf.AttachFlowDissector
|
||||
case ebpf.SkLookup:
|
||||
attach = ebpf.AttachSkLookup
|
||||
default:
|
||||
return nil, fmt.Errorf("can't attach %v to network namespace", t)
|
||||
}
|
||||
|
||||
link, err := AttachRawLink(RawLinkOptions{
|
||||
Target: ns,
|
||||
Program: prog,
|
||||
Attach: attach,
|
||||
})
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return &NetNsLink{link}, nil
|
||||
}
|
||||
|
||||
// LoadPinnedNetNs loads a network namespace link from bpffs.
|
||||
func LoadPinnedNetNs(fileName string, opts *ebpf.LoadPinOptions) (*NetNsLink, error) {
|
||||
link, err := LoadPinnedRawLink(fileName, NetNsType, opts)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return &NetNsLink{link}, nil
|
||||
}
|
||||
|
||||
// Info returns information about the link.
|
||||
func (nns *NetNsLink) Info() (*NetNsInfo, error) {
|
||||
info, err := nns.RawLink.Info()
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return &NetNsInfo{*info}, nil
|
||||
}
|
||||
+272
@@ -0,0 +1,272 @@
|
||||
package link
|
||||
|
||||
import (
|
||||
"bytes"
|
||||
"errors"
|
||||
"fmt"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"regexp"
|
||||
"runtime"
|
||||
"strconv"
|
||||
"strings"
|
||||
"unsafe"
|
||||
|
||||
"github.com/cilium/ebpf"
|
||||
"github.com/cilium/ebpf/internal"
|
||||
"github.com/cilium/ebpf/internal/unix"
|
||||
)
|
||||
|
||||
// Getting the terminology right is usually the hardest part. For posterity and
|
||||
// for staying sane during implementation:
|
||||
//
|
||||
// - trace event: Representation of a kernel runtime hook. Filesystem entries
|
||||
// under <tracefs>/events. Can be tracepoints (static), kprobes or uprobes.
|
||||
// Can be instantiated into perf events (see below).
|
||||
// - tracepoint: A predetermined hook point in the kernel. Exposed as trace
|
||||
// events in (sub)directories under <tracefs>/events. Cannot be closed or
|
||||
// removed, they are static.
|
||||
// - k(ret)probe: Ephemeral trace events based on entry or exit points of
|
||||
// exported kernel symbols. kprobe-based (tracefs) trace events can be
|
||||
// created system-wide by writing to the <tracefs>/kprobe_events file, or
|
||||
// they can be scoped to the current process by creating PMU perf events.
|
||||
// - u(ret)probe: Ephemeral trace events based on user provides ELF binaries
|
||||
// and offsets. uprobe-based (tracefs) trace events can be
|
||||
// created system-wide by writing to the <tracefs>/uprobe_events file, or
|
||||
// they can be scoped to the current process by creating PMU perf events.
|
||||
// - perf event: An object instantiated based on an existing trace event or
|
||||
// kernel symbol. Referred to by fd in userspace.
|
||||
// Exactly one eBPF program can be attached to a perf event. Multiple perf
|
||||
// events can be created from a single trace event. Closing a perf event
|
||||
// stops any further invocations of the attached eBPF program.
|
||||
|
||||
var (
|
||||
tracefsPath = "/sys/kernel/debug/tracing"
|
||||
|
||||
// Trace event groups, names and kernel symbols must adhere to this set
|
||||
// of characters. Non-empty, first character must not be a number, all
|
||||
// characters must be alphanumeric or underscore.
|
||||
rgxTraceEvent = regexp.MustCompile("^[a-zA-Z_][0-9a-zA-Z_]*$")
|
||||
|
||||
errInvalidInput = errors.New("invalid input")
|
||||
)
|
||||
|
||||
const (
|
||||
perfAllThreads = -1
|
||||
)
|
||||
|
||||
type perfEventType uint8
|
||||
|
||||
const (
|
||||
tracepointEvent perfEventType = iota
|
||||
kprobeEvent
|
||||
kretprobeEvent
|
||||
uprobeEvent
|
||||
uretprobeEvent
|
||||
)
|
||||
|
||||
// A perfEvent represents a perf event kernel object. Exactly one eBPF program
|
||||
// can be attached to it. It is created based on a tracefs trace event or a
|
||||
// Performance Monitoring Unit (PMU).
|
||||
type perfEvent struct {
|
||||
|
||||
// Group and name of the tracepoint/kprobe/uprobe.
|
||||
group string
|
||||
name string
|
||||
|
||||
// PMU event ID read from sysfs. Valid IDs are non-zero.
|
||||
pmuID uint64
|
||||
// ID of the trace event read from tracefs. Valid IDs are non-zero.
|
||||
tracefsID uint64
|
||||
|
||||
// The event type determines the types of programs that can be attached.
|
||||
typ perfEventType
|
||||
|
||||
fd *internal.FD
|
||||
}
|
||||
|
||||
func (pe *perfEvent) isLink() {}
|
||||
|
||||
func (pe *perfEvent) Pin(string) error {
|
||||
return fmt.Errorf("pin perf event: %w", ErrNotSupported)
|
||||
}
|
||||
|
||||
func (pe *perfEvent) Unpin() error {
|
||||
return fmt.Errorf("unpin perf event: %w", ErrNotSupported)
|
||||
}
|
||||
|
||||
// Since 4.15 (e87c6bc3852b "bpf: permit multiple bpf attachments for a single perf event"),
|
||||
// calling PERF_EVENT_IOC_SET_BPF appends the given program to a prog_array
|
||||
// owned by the perf event, which means multiple programs can be attached
|
||||
// simultaneously.
|
||||
//
|
||||
// Before 4.15, calling PERF_EVENT_IOC_SET_BPF more than once on a perf event
|
||||
// returns EEXIST.
|
||||
//
|
||||
// Detaching a program from a perf event is currently not possible, so a
|
||||
// program replacement mechanism cannot be implemented for perf events.
|
||||
func (pe *perfEvent) Update(prog *ebpf.Program) error {
|
||||
return fmt.Errorf("can't replace eBPF program in perf event: %w", ErrNotSupported)
|
||||
}
|
||||
|
||||
func (pe *perfEvent) Close() error {
|
||||
if pe.fd == nil {
|
||||
return nil
|
||||
}
|
||||
|
||||
pfd, err := pe.fd.Value()
|
||||
if err != nil {
|
||||
return fmt.Errorf("getting perf event fd: %w", err)
|
||||
}
|
||||
|
||||
err = unix.IoctlSetInt(int(pfd), unix.PERF_EVENT_IOC_DISABLE, 0)
|
||||
if err != nil {
|
||||
return fmt.Errorf("disabling perf event: %w", err)
|
||||
}
|
||||
|
||||
err = pe.fd.Close()
|
||||
if err != nil {
|
||||
return fmt.Errorf("closing perf event fd: %w", err)
|
||||
}
|
||||
|
||||
switch pe.typ {
|
||||
case kprobeEvent, kretprobeEvent:
|
||||
// Clean up kprobe tracefs entry.
|
||||
if pe.tracefsID != 0 {
|
||||
return closeTraceFSProbeEvent(kprobeType, pe.group, pe.name)
|
||||
}
|
||||
case uprobeEvent, uretprobeEvent:
|
||||
// Clean up uprobe tracefs entry.
|
||||
if pe.tracefsID != 0 {
|
||||
return closeTraceFSProbeEvent(uprobeType, pe.group, pe.name)
|
||||
}
|
||||
case tracepointEvent:
|
||||
// Tracepoint trace events don't hold any extra resources.
|
||||
return nil
|
||||
}
|
||||
|
||||
return nil
|
||||
}
|
||||
|
||||
// attach the given eBPF prog to the perf event stored in pe.
|
||||
// pe must contain a valid perf event fd.
|
||||
// prog's type must match the program type stored in pe.
|
||||
func (pe *perfEvent) attach(prog *ebpf.Program) error {
|
||||
if prog == nil {
|
||||
return errors.New("cannot attach a nil program")
|
||||
}
|
||||
if pe.fd == nil {
|
||||
return errors.New("cannot attach to nil perf event")
|
||||
}
|
||||
if prog.FD() < 0 {
|
||||
return fmt.Errorf("invalid program: %w", internal.ErrClosedFd)
|
||||
}
|
||||
switch pe.typ {
|
||||
case kprobeEvent, kretprobeEvent, uprobeEvent, uretprobeEvent:
|
||||
if t := prog.Type(); t != ebpf.Kprobe {
|
||||
return fmt.Errorf("invalid program type (expected %s): %s", ebpf.Kprobe, t)
|
||||
}
|
||||
case tracepointEvent:
|
||||
if t := prog.Type(); t != ebpf.TracePoint {
|
||||
return fmt.Errorf("invalid program type (expected %s): %s", ebpf.TracePoint, t)
|
||||
}
|
||||
default:
|
||||
return fmt.Errorf("unknown perf event type: %d", pe.typ)
|
||||
}
|
||||
|
||||
// The ioctl below will fail when the fd is invalid.
|
||||
kfd, _ := pe.fd.Value()
|
||||
|
||||
// Assign the eBPF program to the perf event.
|
||||
err := unix.IoctlSetInt(int(kfd), unix.PERF_EVENT_IOC_SET_BPF, prog.FD())
|
||||
if err != nil {
|
||||
return fmt.Errorf("setting perf event bpf program: %w", err)
|
||||
}
|
||||
|
||||
// PERF_EVENT_IOC_ENABLE and _DISABLE ignore their given values.
|
||||
if err := unix.IoctlSetInt(int(kfd), unix.PERF_EVENT_IOC_ENABLE, 0); err != nil {
|
||||
return fmt.Errorf("enable perf event: %s", err)
|
||||
}
|
||||
|
||||
// Close the perf event when its reference is lost to avoid leaking system resources.
|
||||
runtime.SetFinalizer(pe, (*perfEvent).Close)
|
||||
return nil
|
||||
}
|
||||
|
||||
// unsafeStringPtr returns an unsafe.Pointer to a NUL-terminated copy of str.
|
||||
func unsafeStringPtr(str string) (unsafe.Pointer, error) {
|
||||
p, err := unix.BytePtrFromString(str)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return unsafe.Pointer(p), nil
|
||||
}
|
||||
|
||||
// getTraceEventID reads a trace event's ID from tracefs given its group and name.
|
||||
// group and name must be alphanumeric or underscore, as required by the kernel.
|
||||
func getTraceEventID(group, name string) (uint64, error) {
|
||||
tid, err := uint64FromFile(tracefsPath, "events", group, name, "id")
|
||||
if errors.Is(err, os.ErrNotExist) {
|
||||
return 0, fmt.Errorf("trace event %s/%s: %w", group, name, os.ErrNotExist)
|
||||
}
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("reading trace event ID of %s/%s: %w", group, name, err)
|
||||
}
|
||||
|
||||
return tid, nil
|
||||
}
|
||||
|
||||
// getPMUEventType reads a Performance Monitoring Unit's type (numeric identifier)
|
||||
// from /sys/bus/event_source/devices/<pmu>/type.
|
||||
//
|
||||
// Returns ErrNotSupported if the pmu type is not supported.
|
||||
func getPMUEventType(typ probeType) (uint64, error) {
|
||||
et, err := uint64FromFile("/sys/bus/event_source/devices", typ.String(), "type")
|
||||
if errors.Is(err, os.ErrNotExist) {
|
||||
return 0, fmt.Errorf("pmu type %s: %w", typ, ErrNotSupported)
|
||||
}
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("reading pmu type %s: %w", typ, err)
|
||||
}
|
||||
|
||||
return et, nil
|
||||
}
|
||||
|
||||
// openTracepointPerfEvent opens a tracepoint-type perf event. System-wide
|
||||
// [k,u]probes created by writing to <tracefs>/[k,u]probe_events are tracepoints
|
||||
// behind the scenes, and can be attached to using these perf events.
|
||||
func openTracepointPerfEvent(tid uint64, pid int) (*internal.FD, error) {
|
||||
attr := unix.PerfEventAttr{
|
||||
Type: unix.PERF_TYPE_TRACEPOINT,
|
||||
Config: tid,
|
||||
Sample_type: unix.PERF_SAMPLE_RAW,
|
||||
Sample: 1,
|
||||
Wakeup: 1,
|
||||
}
|
||||
|
||||
fd, err := unix.PerfEventOpen(&attr, pid, 0, -1, unix.PERF_FLAG_FD_CLOEXEC)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("opening tracepoint perf event: %w", err)
|
||||
}
|
||||
|
||||
return internal.NewFD(uint32(fd)), nil
|
||||
}
|
||||
|
||||
// uint64FromFile reads a uint64 from a file. All elements of path are sanitized
|
||||
// and joined onto base. Returns error if base no longer prefixes the path after
|
||||
// joining all components.
|
||||
func uint64FromFile(base string, path ...string) (uint64, error) {
|
||||
l := filepath.Join(path...)
|
||||
p := filepath.Join(base, l)
|
||||
if !strings.HasPrefix(p, base) {
|
||||
return 0, fmt.Errorf("path '%s' attempts to escape base path '%s': %w", l, base, errInvalidInput)
|
||||
}
|
||||
|
||||
data, err := os.ReadFile(p)
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("reading file %s: %w", p, err)
|
||||
}
|
||||
|
||||
et := bytes.TrimSpace(data)
|
||||
return strconv.ParseUint(string(et), 10, 64)
|
||||
}
|
||||
+25
@@ -0,0 +1,25 @@
|
||||
package link
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"runtime"
|
||||
)
|
||||
|
||||
func platformPrefix(symbol string) string {
|
||||
|
||||
prefix := runtime.GOARCH
|
||||
|
||||
// per https://github.com/golang/go/blob/master/src/go/build/syslist.go
|
||||
switch prefix {
|
||||
case "386":
|
||||
prefix = "ia32"
|
||||
case "amd64", "amd64p32":
|
||||
prefix = "x64"
|
||||
case "arm64", "arm64be":
|
||||
prefix = "arm64"
|
||||
default:
|
||||
return symbol
|
||||
}
|
||||
|
||||
return fmt.Sprintf("__%s_%s", prefix, symbol)
|
||||
}
|
||||
+76
@@ -0,0 +1,76 @@
|
||||
package link
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
|
||||
"github.com/cilium/ebpf"
|
||||
"github.com/cilium/ebpf/internal"
|
||||
)
|
||||
|
||||
type RawAttachProgramOptions struct {
|
||||
// File descriptor to attach to. This differs for each attach type.
|
||||
Target int
|
||||
// Program to attach.
|
||||
Program *ebpf.Program
|
||||
// Program to replace (cgroups).
|
||||
Replace *ebpf.Program
|
||||
// Attach must match the attach type of Program (and Replace).
|
||||
Attach ebpf.AttachType
|
||||
// Flags control the attach behaviour. This differs for each attach type.
|
||||
Flags uint32
|
||||
}
|
||||
|
||||
// RawAttachProgram is a low level wrapper around BPF_PROG_ATTACH.
|
||||
//
|
||||
// You should use one of the higher level abstractions available in this
|
||||
// package if possible.
|
||||
func RawAttachProgram(opts RawAttachProgramOptions) error {
|
||||
if err := haveProgAttach(); err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
var replaceFd uint32
|
||||
if opts.Replace != nil {
|
||||
replaceFd = uint32(opts.Replace.FD())
|
||||
}
|
||||
|
||||
attr := internal.BPFProgAttachAttr{
|
||||
TargetFd: uint32(opts.Target),
|
||||
AttachBpfFd: uint32(opts.Program.FD()),
|
||||
ReplaceBpfFd: replaceFd,
|
||||
AttachType: uint32(opts.Attach),
|
||||
AttachFlags: uint32(opts.Flags),
|
||||
}
|
||||
|
||||
if err := internal.BPFProgAttach(&attr); err != nil {
|
||||
return fmt.Errorf("can't attach program: %w", err)
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
type RawDetachProgramOptions struct {
|
||||
Target int
|
||||
Program *ebpf.Program
|
||||
Attach ebpf.AttachType
|
||||
}
|
||||
|
||||
// RawDetachProgram is a low level wrapper around BPF_PROG_DETACH.
|
||||
//
|
||||
// You should use one of the higher level abstractions available in this
|
||||
// package if possible.
|
||||
func RawDetachProgram(opts RawDetachProgramOptions) error {
|
||||
if err := haveProgAttach(); err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
attr := internal.BPFProgDetachAttr{
|
||||
TargetFd: uint32(opts.Target),
|
||||
AttachBpfFd: uint32(opts.Program.FD()),
|
||||
AttachType: uint32(opts.Attach),
|
||||
}
|
||||
if err := internal.BPFProgDetach(&attr); err != nil {
|
||||
return fmt.Errorf("can't detach program: %w", err)
|
||||
}
|
||||
|
||||
return nil
|
||||
}
|
||||
+61
@@ -0,0 +1,61 @@
|
||||
package link
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
|
||||
"github.com/cilium/ebpf"
|
||||
"github.com/cilium/ebpf/internal"
|
||||
)
|
||||
|
||||
type RawTracepointOptions struct {
|
||||
// Tracepoint name.
|
||||
Name string
|
||||
// Program must be of type RawTracepoint*
|
||||
Program *ebpf.Program
|
||||
}
|
||||
|
||||
// AttachRawTracepoint links a BPF program to a raw_tracepoint.
|
||||
//
|
||||
// Requires at least Linux 4.17.
|
||||
func AttachRawTracepoint(opts RawTracepointOptions) (Link, error) {
|
||||
if t := opts.Program.Type(); t != ebpf.RawTracepoint && t != ebpf.RawTracepointWritable {
|
||||
return nil, fmt.Errorf("invalid program type %s, expected RawTracepoint(Writable)", t)
|
||||
}
|
||||
if opts.Program.FD() < 0 {
|
||||
return nil, fmt.Errorf("invalid program: %w", internal.ErrClosedFd)
|
||||
}
|
||||
|
||||
fd, err := bpfRawTracepointOpen(&bpfRawTracepointOpenAttr{
|
||||
name: internal.NewStringPointer(opts.Name),
|
||||
fd: uint32(opts.Program.FD()),
|
||||
})
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return &progAttachRawTracepoint{fd: fd}, nil
|
||||
}
|
||||
|
||||
type progAttachRawTracepoint struct {
|
||||
fd *internal.FD
|
||||
}
|
||||
|
||||
var _ Link = (*progAttachRawTracepoint)(nil)
|
||||
|
||||
func (rt *progAttachRawTracepoint) isLink() {}
|
||||
|
||||
func (rt *progAttachRawTracepoint) Close() error {
|
||||
return rt.fd.Close()
|
||||
}
|
||||
|
||||
func (rt *progAttachRawTracepoint) Update(_ *ebpf.Program) error {
|
||||
return fmt.Errorf("can't update raw_tracepoint: %w", ErrNotSupported)
|
||||
}
|
||||
|
||||
func (rt *progAttachRawTracepoint) Pin(_ string) error {
|
||||
return fmt.Errorf("can't pin raw_tracepoint: %w", ErrNotSupported)
|
||||
}
|
||||
|
||||
func (rt *progAttachRawTracepoint) Unpin() error {
|
||||
return fmt.Errorf("unpin raw_tracepoint: %w", ErrNotSupported)
|
||||
}
|
||||
+191
@@ -0,0 +1,191 @@
|
||||
package link
|
||||
|
||||
import (
|
||||
"errors"
|
||||
"unsafe"
|
||||
|
||||
"github.com/cilium/ebpf"
|
||||
"github.com/cilium/ebpf/asm"
|
||||
"github.com/cilium/ebpf/internal"
|
||||
"github.com/cilium/ebpf/internal/unix"
|
||||
)
|
||||
|
||||
// Type is the kind of link.
|
||||
type Type uint32
|
||||
|
||||
// Valid link types.
|
||||
//
|
||||
// Equivalent to enum bpf_link_type.
|
||||
const (
|
||||
UnspecifiedType Type = iota
|
||||
RawTracepointType
|
||||
TracingType
|
||||
CgroupType
|
||||
IterType
|
||||
NetNsType
|
||||
XDPType
|
||||
)
|
||||
|
||||
var haveProgAttach = internal.FeatureTest("BPF_PROG_ATTACH", "4.10", func() error {
|
||||
prog, err := ebpf.NewProgram(&ebpf.ProgramSpec{
|
||||
Type: ebpf.CGroupSKB,
|
||||
AttachType: ebpf.AttachCGroupInetIngress,
|
||||
License: "MIT",
|
||||
Instructions: asm.Instructions{
|
||||
asm.Mov.Imm(asm.R0, 0),
|
||||
asm.Return(),
|
||||
},
|
||||
})
|
||||
if err != nil {
|
||||
return internal.ErrNotSupported
|
||||
}
|
||||
|
||||
// BPF_PROG_ATTACH was introduced at the same time as CGgroupSKB,
|
||||
// so being able to load the program is enough to infer that we
|
||||
// have the syscall.
|
||||
prog.Close()
|
||||
return nil
|
||||
})
|
||||
|
||||
var haveProgAttachReplace = internal.FeatureTest("BPF_PROG_ATTACH atomic replacement", "5.5", func() error {
|
||||
if err := haveProgAttach(); err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
prog, err := ebpf.NewProgram(&ebpf.ProgramSpec{
|
||||
Type: ebpf.CGroupSKB,
|
||||
AttachType: ebpf.AttachCGroupInetIngress,
|
||||
License: "MIT",
|
||||
Instructions: asm.Instructions{
|
||||
asm.Mov.Imm(asm.R0, 0),
|
||||
asm.Return(),
|
||||
},
|
||||
})
|
||||
if err != nil {
|
||||
return internal.ErrNotSupported
|
||||
}
|
||||
defer prog.Close()
|
||||
|
||||
// We know that we have BPF_PROG_ATTACH since we can load CGroupSKB programs.
|
||||
// If passing BPF_F_REPLACE gives us EINVAL we know that the feature isn't
|
||||
// present.
|
||||
attr := internal.BPFProgAttachAttr{
|
||||
// We rely on this being checked after attachFlags.
|
||||
TargetFd: ^uint32(0),
|
||||
AttachBpfFd: uint32(prog.FD()),
|
||||
AttachType: uint32(ebpf.AttachCGroupInetIngress),
|
||||
AttachFlags: uint32(flagReplace),
|
||||
}
|
||||
|
||||
err = internal.BPFProgAttach(&attr)
|
||||
if errors.Is(err, unix.EINVAL) {
|
||||
return internal.ErrNotSupported
|
||||
}
|
||||
if errors.Is(err, unix.EBADF) {
|
||||
return nil
|
||||
}
|
||||
return err
|
||||
})
|
||||
|
||||
type bpfLinkCreateAttr struct {
|
||||
progFd uint32
|
||||
targetFd uint32
|
||||
attachType ebpf.AttachType
|
||||
flags uint32
|
||||
targetBTFID uint32
|
||||
}
|
||||
|
||||
func bpfLinkCreate(attr *bpfLinkCreateAttr) (*internal.FD, error) {
|
||||
ptr, err := internal.BPF(internal.BPF_LINK_CREATE, unsafe.Pointer(attr), unsafe.Sizeof(*attr))
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return internal.NewFD(uint32(ptr)), nil
|
||||
}
|
||||
|
||||
type bpfLinkCreateIterAttr struct {
|
||||
prog_fd uint32
|
||||
target_fd uint32
|
||||
attach_type ebpf.AttachType
|
||||
flags uint32
|
||||
iter_info internal.Pointer
|
||||
iter_info_len uint32
|
||||
}
|
||||
|
||||
func bpfLinkCreateIter(attr *bpfLinkCreateIterAttr) (*internal.FD, error) {
|
||||
ptr, err := internal.BPF(internal.BPF_LINK_CREATE, unsafe.Pointer(attr), unsafe.Sizeof(*attr))
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return internal.NewFD(uint32(ptr)), nil
|
||||
}
|
||||
|
||||
type bpfLinkUpdateAttr struct {
|
||||
linkFd uint32
|
||||
newProgFd uint32
|
||||
flags uint32
|
||||
oldProgFd uint32
|
||||
}
|
||||
|
||||
func bpfLinkUpdate(attr *bpfLinkUpdateAttr) error {
|
||||
_, err := internal.BPF(internal.BPF_LINK_UPDATE, unsafe.Pointer(attr), unsafe.Sizeof(*attr))
|
||||
return err
|
||||
}
|
||||
|
||||
var haveBPFLink = internal.FeatureTest("bpf_link", "5.7", func() error {
|
||||
prog, err := ebpf.NewProgram(&ebpf.ProgramSpec{
|
||||
Type: ebpf.CGroupSKB,
|
||||
AttachType: ebpf.AttachCGroupInetIngress,
|
||||
License: "MIT",
|
||||
Instructions: asm.Instructions{
|
||||
asm.Mov.Imm(asm.R0, 0),
|
||||
asm.Return(),
|
||||
},
|
||||
})
|
||||
if err != nil {
|
||||
return internal.ErrNotSupported
|
||||
}
|
||||
defer prog.Close()
|
||||
|
||||
attr := bpfLinkCreateAttr{
|
||||
// This is a hopefully invalid file descriptor, which triggers EBADF.
|
||||
targetFd: ^uint32(0),
|
||||
progFd: uint32(prog.FD()),
|
||||
attachType: ebpf.AttachCGroupInetIngress,
|
||||
}
|
||||
_, err = bpfLinkCreate(&attr)
|
||||
if errors.Is(err, unix.EINVAL) {
|
||||
return internal.ErrNotSupported
|
||||
}
|
||||
if errors.Is(err, unix.EBADF) {
|
||||
return nil
|
||||
}
|
||||
return err
|
||||
})
|
||||
|
||||
type bpfIterCreateAttr struct {
|
||||
linkFd uint32
|
||||
flags uint32
|
||||
}
|
||||
|
||||
func bpfIterCreate(attr *bpfIterCreateAttr) (*internal.FD, error) {
|
||||
ptr, err := internal.BPF(internal.BPF_ITER_CREATE, unsafe.Pointer(attr), unsafe.Sizeof(*attr))
|
||||
if err == nil {
|
||||
return internal.NewFD(uint32(ptr)), nil
|
||||
}
|
||||
return nil, err
|
||||
}
|
||||
|
||||
type bpfRawTracepointOpenAttr struct {
|
||||
name internal.Pointer
|
||||
fd uint32
|
||||
_ uint32
|
||||
}
|
||||
|
||||
func bpfRawTracepointOpen(attr *bpfRawTracepointOpenAttr) (*internal.FD, error) {
|
||||
ptr, err := internal.BPF(internal.BPF_RAW_TRACEPOINT_OPEN, unsafe.Pointer(attr), unsafe.Sizeof(*attr))
|
||||
if err == nil {
|
||||
return internal.NewFD(uint32(ptr)), nil
|
||||
}
|
||||
return nil, err
|
||||
}
|
||||
+60
@@ -0,0 +1,60 @@
|
||||
package link
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
|
||||
"github.com/cilium/ebpf"
|
||||
)
|
||||
|
||||
// Tracepoint attaches the given eBPF program to the tracepoint with the given
|
||||
// group and name. See /sys/kernel/debug/tracing/events to find available
|
||||
// tracepoints. The top-level directory is the group, the event's subdirectory
|
||||
// is the name. Example:
|
||||
//
|
||||
// tp, err := Tracepoint("syscalls", "sys_enter_fork", prog)
|
||||
//
|
||||
// Losing the reference to the resulting Link (tp) will close the Tracepoint
|
||||
// and prevent further execution of prog. The Link must be Closed during
|
||||
// program shutdown to avoid leaking system resources.
|
||||
//
|
||||
// Note that attaching eBPF programs to syscalls (sys_enter_*/sys_exit_*) is
|
||||
// only possible as of kernel 4.14 (commit cf5f5ce).
|
||||
func Tracepoint(group, name string, prog *ebpf.Program) (Link, error) {
|
||||
if group == "" || name == "" {
|
||||
return nil, fmt.Errorf("group and name cannot be empty: %w", errInvalidInput)
|
||||
}
|
||||
if prog == nil {
|
||||
return nil, fmt.Errorf("prog cannot be nil: %w", errInvalidInput)
|
||||
}
|
||||
if !rgxTraceEvent.MatchString(group) || !rgxTraceEvent.MatchString(name) {
|
||||
return nil, fmt.Errorf("group and name '%s/%s' must be alphanumeric or underscore: %w", group, name, errInvalidInput)
|
||||
}
|
||||
if prog.Type() != ebpf.TracePoint {
|
||||
return nil, fmt.Errorf("eBPF program type %s is not a Tracepoint: %w", prog.Type(), errInvalidInput)
|
||||
}
|
||||
|
||||
tid, err := getTraceEventID(group, name)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
fd, err := openTracepointPerfEvent(tid, perfAllThreads)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
pe := &perfEvent{
|
||||
fd: fd,
|
||||
tracefsID: tid,
|
||||
group: group,
|
||||
name: name,
|
||||
typ: tracepointEvent,
|
||||
}
|
||||
|
||||
if err := pe.attach(prog); err != nil {
|
||||
pe.Close()
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return pe, nil
|
||||
}
|
||||
+288
@@ -0,0 +1,288 @@
|
||||
package link
|
||||
|
||||
import (
|
||||
"debug/elf"
|
||||
"errors"
|
||||
"fmt"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"regexp"
|
||||
"sync"
|
||||
|
||||
"github.com/cilium/ebpf"
|
||||
"github.com/cilium/ebpf/internal"
|
||||
)
|
||||
|
||||
var (
|
||||
uprobeEventsPath = filepath.Join(tracefsPath, "uprobe_events")
|
||||
|
||||
// rgxUprobeSymbol is used to strip invalid characters from the uprobe symbol
|
||||
// as they are not allowed to be used as the EVENT token in tracefs.
|
||||
rgxUprobeSymbol = regexp.MustCompile("[^a-zA-Z0-9]+")
|
||||
|
||||
uprobeRetprobeBit = struct {
|
||||
once sync.Once
|
||||
value uint64
|
||||
err error
|
||||
}{}
|
||||
|
||||
// ErrNoSymbol indicates that the given symbol was not found
|
||||
// in the ELF symbols table.
|
||||
ErrNoSymbol = errors.New("not found")
|
||||
)
|
||||
|
||||
// Executable defines an executable program on the filesystem.
|
||||
type Executable struct {
|
||||
// Path of the executable on the filesystem.
|
||||
path string
|
||||
// Parsed ELF symbols and dynamic symbols offsets.
|
||||
offsets map[string]uint64
|
||||
}
|
||||
|
||||
// UprobeOptions defines additional parameters that will be used
|
||||
// when loading Uprobes.
|
||||
type UprobeOptions struct {
|
||||
// Symbol offset. Must be provided in case of external symbols (shared libs).
|
||||
// If set, overrides the offset eventually parsed from the executable.
|
||||
Offset uint64
|
||||
// Only set the uprobe on the given process ID. Useful when tracing
|
||||
// shared library calls or programs that have many running instances.
|
||||
PID int
|
||||
}
|
||||
|
||||
// To open a new Executable, use:
|
||||
//
|
||||
// OpenExecutable("/bin/bash")
|
||||
//
|
||||
// The returned value can then be used to open Uprobe(s).
|
||||
func OpenExecutable(path string) (*Executable, error) {
|
||||
if path == "" {
|
||||
return nil, fmt.Errorf("path cannot be empty")
|
||||
}
|
||||
|
||||
f, err := os.Open(path)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("open file '%s': %w", path, err)
|
||||
}
|
||||
defer f.Close()
|
||||
|
||||
se, err := internal.NewSafeELFFile(f)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("parse ELF file: %w", err)
|
||||
}
|
||||
|
||||
if se.Type != elf.ET_EXEC && se.Type != elf.ET_DYN {
|
||||
// ELF is not an executable or a shared object.
|
||||
return nil, errors.New("the given file is not an executable or a shared object")
|
||||
}
|
||||
|
||||
ex := Executable{
|
||||
path: path,
|
||||
offsets: make(map[string]uint64),
|
||||
}
|
||||
|
||||
if err := ex.load(se); err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return &ex, nil
|
||||
}
|
||||
|
||||
func (ex *Executable) load(f *internal.SafeELFFile) error {
|
||||
syms, err := f.Symbols()
|
||||
if err != nil && !errors.Is(err, elf.ErrNoSymbols) {
|
||||
return err
|
||||
}
|
||||
|
||||
dynsyms, err := f.DynamicSymbols()
|
||||
if err != nil && !errors.Is(err, elf.ErrNoSymbols) {
|
||||
return err
|
||||
}
|
||||
|
||||
syms = append(syms, dynsyms...)
|
||||
|
||||
for _, s := range syms {
|
||||
if elf.ST_TYPE(s.Info) != elf.STT_FUNC {
|
||||
// Symbol not associated with a function or other executable code.
|
||||
continue
|
||||
}
|
||||
|
||||
off := s.Value
|
||||
|
||||
// Loop over ELF segments.
|
||||
for _, prog := range f.Progs {
|
||||
// Skip uninteresting segments.
|
||||
if prog.Type != elf.PT_LOAD || (prog.Flags&elf.PF_X) == 0 {
|
||||
continue
|
||||
}
|
||||
|
||||
if prog.Vaddr <= s.Value && s.Value < (prog.Vaddr+prog.Memsz) {
|
||||
// If the symbol value is contained in the segment, calculate
|
||||
// the symbol offset.
|
||||
//
|
||||
// fn symbol offset = fn symbol VA - .text VA + .text offset
|
||||
//
|
||||
// stackoverflow.com/a/40249502
|
||||
off = s.Value - prog.Vaddr + prog.Off
|
||||
break
|
||||
}
|
||||
}
|
||||
|
||||
ex.offsets[s.Name] = off
|
||||
}
|
||||
|
||||
return nil
|
||||
}
|
||||
|
||||
func (ex *Executable) offset(symbol string) (uint64, error) {
|
||||
if off, ok := ex.offsets[symbol]; ok {
|
||||
// Symbols with location 0 from section undef are shared library calls and
|
||||
// are relocated before the binary is executed. Dynamic linking is not
|
||||
// implemented by the library, so mark this as unsupported for now.
|
||||
//
|
||||
// Since only offset values are stored and not elf.Symbol, if the value is 0,
|
||||
// assume it's an external symbol.
|
||||
if off == 0 {
|
||||
return 0, fmt.Errorf("cannot resolve %s library call '%s', "+
|
||||
"consider providing the offset via options: %w", ex.path, symbol, ErrNotSupported)
|
||||
}
|
||||
return off, nil
|
||||
}
|
||||
return 0, fmt.Errorf("symbol %s: %w", symbol, ErrNoSymbol)
|
||||
}
|
||||
|
||||
// Uprobe attaches the given eBPF program to a perf event that fires when the
|
||||
// given symbol starts executing in the given Executable.
|
||||
// For example, /bin/bash::main():
|
||||
//
|
||||
// ex, _ = OpenExecutable("/bin/bash")
|
||||
// ex.Uprobe("main", prog, nil)
|
||||
//
|
||||
// When using symbols which belongs to shared libraries,
|
||||
// an offset must be provided via options:
|
||||
//
|
||||
// up, err := ex.Uprobe("main", prog, &UprobeOptions{Offset: 0x123})
|
||||
//
|
||||
// Losing the reference to the resulting Link (up) will close the Uprobe
|
||||
// and prevent further execution of prog. The Link must be Closed during
|
||||
// program shutdown to avoid leaking system resources.
|
||||
//
|
||||
// Functions provided by shared libraries can currently not be traced and
|
||||
// will result in an ErrNotSupported.
|
||||
func (ex *Executable) Uprobe(symbol string, prog *ebpf.Program, opts *UprobeOptions) (Link, error) {
|
||||
u, err := ex.uprobe(symbol, prog, opts, false)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
err = u.attach(prog)
|
||||
if err != nil {
|
||||
u.Close()
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return u, nil
|
||||
}
|
||||
|
||||
// Uretprobe attaches the given eBPF program to a perf event that fires right
|
||||
// before the given symbol exits. For example, /bin/bash::main():
|
||||
//
|
||||
// ex, _ = OpenExecutable("/bin/bash")
|
||||
// ex.Uretprobe("main", prog, nil)
|
||||
//
|
||||
// When using symbols which belongs to shared libraries,
|
||||
// an offset must be provided via options:
|
||||
//
|
||||
// up, err := ex.Uretprobe("main", prog, &UprobeOptions{Offset: 0x123})
|
||||
//
|
||||
// Losing the reference to the resulting Link (up) will close the Uprobe
|
||||
// and prevent further execution of prog. The Link must be Closed during
|
||||
// program shutdown to avoid leaking system resources.
|
||||
//
|
||||
// Functions provided by shared libraries can currently not be traced and
|
||||
// will result in an ErrNotSupported.
|
||||
func (ex *Executable) Uretprobe(symbol string, prog *ebpf.Program, opts *UprobeOptions) (Link, error) {
|
||||
u, err := ex.uprobe(symbol, prog, opts, true)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
err = u.attach(prog)
|
||||
if err != nil {
|
||||
u.Close()
|
||||
return nil, err
|
||||
}
|
||||
|
||||
return u, nil
|
||||
}
|
||||
|
||||
// uprobe opens a perf event for the given binary/symbol and attaches prog to it.
|
||||
// If ret is true, create a uretprobe.
|
||||
func (ex *Executable) uprobe(symbol string, prog *ebpf.Program, opts *UprobeOptions, ret bool) (*perfEvent, error) {
|
||||
if prog == nil {
|
||||
return nil, fmt.Errorf("prog cannot be nil: %w", errInvalidInput)
|
||||
}
|
||||
if prog.Type() != ebpf.Kprobe {
|
||||
return nil, fmt.Errorf("eBPF program type %s is not Kprobe: %w", prog.Type(), errInvalidInput)
|
||||
}
|
||||
|
||||
var offset uint64
|
||||
if opts != nil && opts.Offset != 0 {
|
||||
offset = opts.Offset
|
||||
} else {
|
||||
off, err := ex.offset(symbol)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
offset = off
|
||||
}
|
||||
|
||||
pid := perfAllThreads
|
||||
if opts != nil && opts.PID != 0 {
|
||||
pid = opts.PID
|
||||
}
|
||||
|
||||
// Use uprobe PMU if the kernel has it available.
|
||||
tp, err := pmuUprobe(symbol, ex.path, offset, pid, ret)
|
||||
if err == nil {
|
||||
return tp, nil
|
||||
}
|
||||
if err != nil && !errors.Is(err, ErrNotSupported) {
|
||||
return nil, fmt.Errorf("creating perf_uprobe PMU: %w", err)
|
||||
}
|
||||
|
||||
// Use tracefs if uprobe PMU is missing.
|
||||
tp, err = tracefsUprobe(uprobeSanitizedSymbol(symbol), ex.path, offset, pid, ret)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("creating trace event '%s:%s' in tracefs: %w", ex.path, symbol, err)
|
||||
}
|
||||
|
||||
return tp, nil
|
||||
}
|
||||
|
||||
// pmuUprobe opens a perf event based on the uprobe PMU.
|
||||
func pmuUprobe(symbol, path string, offset uint64, pid int, ret bool) (*perfEvent, error) {
|
||||
return pmuProbe(uprobeType, symbol, path, offset, pid, ret)
|
||||
}
|
||||
|
||||
// tracefsUprobe creates a Uprobe tracefs entry.
|
||||
func tracefsUprobe(symbol, path string, offset uint64, pid int, ret bool) (*perfEvent, error) {
|
||||
return tracefsProbe(uprobeType, symbol, path, offset, pid, ret)
|
||||
}
|
||||
|
||||
// uprobeSanitizedSymbol replaces every invalid characted for the tracefs api with an underscore.
|
||||
func uprobeSanitizedSymbol(symbol string) string {
|
||||
return rgxUprobeSymbol.ReplaceAllString(symbol, "_")
|
||||
}
|
||||
|
||||
// uprobePathOffset creates the PATH:OFFSET token for the tracefs api.
|
||||
func uprobePathOffset(path string, offset uint64) string {
|
||||
return fmt.Sprintf("%s:%#x", path, offset)
|
||||
}
|
||||
|
||||
func uretprobeBit() (uint64, error) {
|
||||
uprobeRetprobeBit.once.Do(func() {
|
||||
uprobeRetprobeBit.value, uprobeRetprobeBit.err = determineRetprobeBit(uprobeType)
|
||||
})
|
||||
return uprobeRetprobeBit.value, uprobeRetprobeBit.err
|
||||
}
|
||||
Reference in New Issue
Block a user