diff options
Diffstat (limited to 'vendor/github.com/cilium/ebpf/link')
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/cgroup.go | 24 | ||||
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/iter.go | 86 | ||||
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/kprobe.go | 568 | ||||
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/link.go | 251 | ||||
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/netns.go | 28 | ||||
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/perf_event.go | 394 | ||||
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/platform.go | 25 | ||||
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/program.go | 14 | ||||
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/raw_tracepoint.go | 62 | ||||
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/socket_filter.go | 40 | ||||
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/syscalls.go | 108 | ||||
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/tracepoint.go | 77 | ||||
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/tracing.go | 141 | ||||
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/uprobe.go | 373 | ||||
-rw-r--r-- | vendor/github.com/cilium/ebpf/link/xdp.go | 54 |
15 files changed, 1971 insertions, 274 deletions
diff --git a/vendor/github.com/cilium/ebpf/link/cgroup.go b/vendor/github.com/cilium/ebpf/link/cgroup.go index 16a943930..003b0638e 100644 --- a/vendor/github.com/cilium/ebpf/link/cgroup.go +++ b/vendor/github.com/cilium/ebpf/link/cgroup.go @@ -56,16 +56,6 @@ func AttachCgroup(opts CgroupOptions) (Link, error) { return cg, nil } -// LoadPinnedCgroup loads a pinned cgroup from a bpffs. -func LoadPinnedCgroup(fileName string) (Link, error) { - link, err := LoadPinnedRawLink(fileName) - if err != nil { - return nil, err - } - - return &linkCgroup{link}, nil -} - type progAttachCgroup struct { cgroup *os.File current *ebpf.Program @@ -147,14 +137,20 @@ 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) +} + +func (cg *progAttachCgroup) Info() (*Info, error) { + return nil, fmt.Errorf("can't get cgroup info: %w", ErrNotSupported) +} + type linkCgroup struct { - *RawLink + RawLink } var _ Link = (*linkCgroup)(nil) -func (cg *linkCgroup) isLink() {} - func newLinkCgroup(cgroup *os.File, attach ebpf.AttachType, prog *ebpf.Program) (*linkCgroup, error) { link, err := AttachRawLink(RawLinkOptions{ Target: int(cgroup.Fd()), @@ -165,5 +161,5 @@ func newLinkCgroup(cgroup *os.File, attach ebpf.AttachType, prog *ebpf.Program) return nil, err } - return &linkCgroup{link}, err + return &linkCgroup{*link}, err } diff --git a/vendor/github.com/cilium/ebpf/link/iter.go b/vendor/github.com/cilium/ebpf/link/iter.go index 2b5f2846a..d2b32ef33 100644 --- a/vendor/github.com/cilium/ebpf/link/iter.go +++ b/vendor/github.com/cilium/ebpf/link/iter.go @@ -3,8 +3,10 @@ package link import ( "fmt" "io" + "unsafe" "github.com/cilium/ebpf" + "github.com/cilium/ebpf/internal/sys" ) type IterOptions struct { @@ -15,77 +17,69 @@ type IterOptions struct { // 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) { - link, err := AttachRawLink(RawLinkOptions{ - Program: opts.Program, - Attach: ebpf.AttachTraceIter, - }) - if err != nil { - return nil, fmt.Errorf("can't link iterator: %w", err) + if err := haveBPFLink(); err != nil { + return nil, err } - return &Iter{link}, err -} + progFd := opts.Program.FD() + if progFd < 0 { + return nil, fmt.Errorf("invalid program: %s", sys.ErrClosedFd) + } + + var info bpfIterLinkInfoMap + if opts.Map != nil { + mapFd := opts.Map.FD() + if mapFd < 0 { + return nil, fmt.Errorf("invalid map: %w", sys.ErrClosedFd) + } + info.map_fd = uint32(mapFd) + } + + attr := sys.LinkCreateIterAttr{ + ProgFd: uint32(progFd), + AttachType: sys.AttachType(ebpf.AttachTraceIter), + IterInfo: sys.NewPointer(unsafe.Pointer(&info)), + IterInfoLen: uint32(unsafe.Sizeof(info)), + } -// LoadPinnedIter loads a pinned iterator from a bpffs. -func LoadPinnedIter(fileName string) (*Iter, error) { - link, err := LoadPinnedRawLink(fileName) + fd, err := sys.LinkCreateIter(&attr) if err != nil { - return nil, err + return nil, fmt.Errorf("can't link iterator: %w", err) } - return &Iter{link}, err + return &Iter{RawLink{fd, ""}}, err } // Iter represents an attached bpf_iter. type Iter struct { - link *RawLink -} - -var _ Link = (*Iter)(nil) - -func (it *Iter) isLink() {} - -// FD returns the underlying file descriptor. -func (it *Iter) FD() int { - return it.link.FD() -} - -// Close implements Link. -func (it *Iter) Close() error { - return it.link.Close() -} - -// Pin implements Link. -func (it *Iter) Pin(fileName string) error { - return it.link.Pin(fileName) -} - -// Update implements Link. -func (it *Iter) Update(new *ebpf.Program) error { - return it.link.Update(new) + 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.link.fd.Value() - if err != nil { - return nil, err + attr := &sys.IterCreateAttr{ + LinkFd: it.fd.Uint(), } - attr := &bpfIterCreateAttr{ - linkFd: linkFd, - } - - fd, err := bpfIterCreate(attr) + fd, err := sys.IterCreate(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 +} diff --git a/vendor/github.com/cilium/ebpf/link/kprobe.go b/vendor/github.com/cilium/ebpf/link/kprobe.go new file mode 100644 index 000000000..fdf622a0c --- /dev/null +++ b/vendor/github.com/cilium/ebpf/link/kprobe.go @@ -0,0 +1,568 @@ +package link + +import ( + "bytes" + "crypto/rand" + "errors" + "fmt" + "os" + "path/filepath" + "runtime" + "strings" + "sync" + "syscall" + "unsafe" + + "github.com/cilium/ebpf" + "github.com/cilium/ebpf/internal/sys" + "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 + +type probeArgs struct { + symbol, group, path string + offset, refCtrOffset, cookie uint64 + pid int + ret bool +} + +// KprobeOptions defines additional parameters that will be used +// when loading Kprobes. +type KprobeOptions struct { + // Arbitrary value that can be fetched from an eBPF program + // via `bpf_get_attach_cookie()`. + // + // Needs kernel 5.15+. + Cookie uint64 + // Offset of the kprobe relative to the traced symbol. + // Can be used to insert kprobes at arbitrary offsets in kernel functions, + // e.g. in places where functions have been inlined. + Offset uint64 +} + +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, nil) +// +// 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, opts *KprobeOptions) (Link, error) { + k, err := kprobe(symbol, prog, opts, false) + if err != nil { + return nil, err + } + + lnk, err := attachPerfEvent(k, prog) + if err != nil { + k.Close() + return nil, err + } + + return lnk, 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, nil) +// +// 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, opts *KprobeOptions) (Link, error) { + k, err := kprobe(symbol, prog, opts, true) + if err != nil { + return nil, err + } + + lnk, err := attachPerfEvent(k, prog) + if err != nil { + k.Close() + return nil, err + } + + return lnk, nil +} + +// isValidKprobeSymbol implements the equivalent of a regex match +// against "^[a-zA-Z_][0-9a-zA-Z_.]*$". +func isValidKprobeSymbol(s string) bool { + if len(s) < 1 { + return false + } + + for i, c := range []byte(s) { + switch { + case c >= 'a' && c <= 'z': + case c >= 'A' && c <= 'Z': + case c == '_': + case i > 0 && c >= '0' && c <= '9': + + // Allow `.` in symbol name. GCC-compiled kernel may change symbol name + // to have a `.isra.$n` suffix, like `udp_send_skb.isra.52`. + // See: https://gcc.gnu.org/gcc-10/changes.html + case i > 0 && c == '.': + + default: + return false + } + } + + return true +} + +// 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, opts *KprobeOptions, 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 !isValidKprobeSymbol(symbol) { + return nil, fmt.Errorf("symbol '%s' must be a valid symbol in /proc/kallsyms: %w", symbol, errInvalidInput) + } + if prog.Type() != ebpf.Kprobe { + return nil, fmt.Errorf("eBPF program type %s is not a Kprobe: %w", prog.Type(), errInvalidInput) + } + + args := probeArgs{ + pid: perfAllThreads, + symbol: symbol, + ret: ret, + } + + if opts != nil { + args.cookie = opts.Cookie + args.offset = opts.Offset + } + + // Use kprobe PMU if the kernel has it available. + tp, err := pmuKprobe(args) + if errors.Is(err, os.ErrNotExist) { + args.symbol = platformPrefix(symbol) + tp, err = pmuKprobe(args) + } + 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. + args.symbol = symbol + tp, err = tracefsKprobe(args) + if errors.Is(err, os.ErrNotExist) { + args.symbol = platformPrefix(symbol) + tp, err = tracefsKprobe(args) + } + 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(args probeArgs) (*perfEvent, error) { + return pmuProbe(kprobeType, args) +} + +// 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, args probeArgs) (*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 args.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(args.symbol) + if err != nil { + return nil, err + } + + attr = unix.PerfEventAttr{ + // The minimum size required for PMU kprobes is PERF_ATTR_SIZE_VER1, + // since it added the config2 (Ext2) field. Use Ext2 as probe_offset. + Size: unix.PERF_ATTR_SIZE_VER1, + Type: uint32(et), // PMU event type read from sysfs + Ext1: uint64(uintptr(sp)), // Kernel symbol to trace + Ext2: args.offset, // Kernel symbol offset + Config: config, // Retprobe flag + } + case uprobeType: + sp, err = unsafeStringPtr(args.path) + if err != nil { + return nil, err + } + + if args.refCtrOffset != 0 { + config |= args.refCtrOffset << uprobeRefCtrOffsetShift + } + + 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: args.offset, // Uprobe offset + Config: config, // RefCtrOffset, Retprobe flag + } + } + + rawFd, err := unix.PerfEventOpen(&attr, args.pid, 0, -1, unix.PERF_FLAG_FD_CLOEXEC) + + // On some old kernels, kprobe PMU doesn't allow `.` in symbol names and + // return -EINVAL. Return ErrNotSupported to allow falling back to tracefs. + // https://github.com/torvalds/linux/blob/94710cac0ef4/kernel/trace/trace_kprobe.c#L340-L343 + if errors.Is(err, unix.EINVAL) && strings.Contains(args.symbol, ".") { + return nil, fmt.Errorf("symbol '%s+%#x': older kernels don't accept dots: %w", args.symbol, args.offset, ErrNotSupported) + } + // 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+%#x' not found: %w", args.symbol, args.offset, os.ErrNotExist) + } + // Since commit ab105a4fb894, -EILSEQ is returned when a kprobe sym+offset is resolved + // to an invalid insn boundary. + if errors.Is(err, syscall.EILSEQ) { + return nil, fmt.Errorf("symbol '%s+%#x' not found (bad insn boundary): %w", args.symbol, args.offset, 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", args.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) + + fd, err := sys.NewFD(rawFd) + if err != nil { + return nil, err + } + + // Kernel has perf_[k,u]probe PMU available, initialize perf event. + return &perfEvent{ + typ: typ.PerfEventType(args.ret), + name: args.symbol, + pmuID: et, + cookie: args.cookie, + fd: fd, + }, nil +} + +// tracefsKprobe creates a Kprobe tracefs entry. +func tracefsKprobe(args probeArgs) (*perfEvent, error) { + return tracefsProbe(kprobeType, args) +} + +// 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, args probeArgs) (_ *perfEvent, err 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) + } + args.group = group + + // 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, args.symbol) + if err == nil { + return nil, fmt.Errorf("trace event already exists: %s/%s", group, args.symbol) + } + if err != nil && !errors.Is(err, os.ErrNotExist) { + return nil, fmt.Errorf("checking trace event %s/%s: %w", group, args.symbol, err) + } + + // Create the [k,u]probe trace event using tracefs. + if err := createTraceFSProbeEvent(typ, args); err != nil { + return nil, fmt.Errorf("creating probe entry on tracefs: %w", err) + } + defer func() { + if err != nil { + // Make sure we clean up the created tracefs event when we return error. + // If a livepatch handler is already active on the symbol, the write to + // tracefs will succeed, a trace event will show up, but creating the + // perf event will fail with EBUSY. + _ = closeTraceFSProbeEvent(typ, args.group, args.symbol) + } + }() + + // Get the newly-created trace event's id. + tid, err := getTraceEventID(group, args.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, args.pid) + if err != nil { + return nil, err + } + + return &perfEvent{ + typ: typ.PerfEventType(args.ret), + group: group, + name: args.symbol, + tracefsID: tid, + cookie: args.cookie, + fd: fd, + }, 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, args probeArgs) 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, token 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. + token = kprobeToken(args) + pe = fmt.Sprintf("%s:%s/%s %s", probePrefix(args.ret), args.group, sanitizeSymbol(args.symbol), token) + 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(0x123) + // + // See Documentation/trace/uprobetracer.txt for more details. + token = uprobeToken(args) + pe = fmt.Sprintf("%s:%s/%s %s", probePrefix(args.ret), args.group, args.symbol, token) + } + _, 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. + // EINVAL is also returned on pre-5.2 kernels when the `SYM[+offs]` token + // is resolved to an invalid insn boundary. + if errors.Is(err, os.ErrNotExist) || errors.Is(err, unix.EINVAL) { + return fmt.Errorf("token %s: %w", token, os.ErrNotExist) + } + // Since commit ab105a4fb894, -EILSEQ is returned when a kprobe sym+offset is resolved + // to an invalid insn boundary. + if errors.Is(err, syscall.EILSEQ) { + return fmt.Errorf("token %s: bad insn boundary: %w", token, os.ErrNotExist) + } + // ERANGE is returned when the `SYM[+offs]` token is too big and cannot + // be resolved. + if errors.Is(err, syscall.ERANGE) { + return fmt.Errorf("token %s: offset too big: %w", token, 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, sanitizeSymbol(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 isValidTraceID. +func randomGroup(prefix string) (string, error) { + if !isValidTraceID(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 +} + +// kprobeToken creates the SYM[+offs] token for the tracefs api. +func kprobeToken(args probeArgs) string { + po := args.symbol + + if args.offset != 0 { + po += fmt.Sprintf("+%#x", args.offset) + } + + return po +} diff --git a/vendor/github.com/cilium/ebpf/link/link.go b/vendor/github.com/cilium/ebpf/link/link.go index 48f1a5529..067d0101a 100644 --- a/vendor/github.com/cilium/ebpf/link/link.go +++ b/vendor/github.com/cilium/ebpf/link/link.go @@ -1,11 +1,14 @@ package link import ( + "bytes" + "encoding/binary" "fmt" - "unsafe" "github.com/cilium/ebpf" + "github.com/cilium/ebpf/btf" "github.com/cilium/ebpf/internal" + "github.com/cilium/ebpf/internal/sys" ) var ErrNotSupported = internal.ErrNotSupported @@ -22,19 +25,65 @@ type Link interface { // 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 pinned. A link - // may continue past the lifetime of the process if Close is + // 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 + // Info returns metadata on a link. + // + // May return an error wrapping ErrNotSupported. + Info() (*Info, error) + // Prevent external users from implementing this interface. isLink() } +// LoadPinnedLink loads a link that was persisted into a bpffs. +func LoadPinnedLink(fileName string, opts *ebpf.LoadPinOptions) (Link, error) { + raw, err := loadPinnedRawLink(fileName, opts) + if err != nil { + return nil, err + } + + return wrapRawLink(raw) +} + +// wrap a RawLink in a more specific type if possible. +// +// The function takes ownership of raw and closes it on error. +func wrapRawLink(raw *RawLink) (Link, error) { + info, err := raw.Info() + if err != nil { + raw.Close() + return nil, err + } + + switch info.Type { + case RawTracepointType: + return &rawTracepoint{*raw}, nil + case TracingType: + return &tracing{*raw}, nil + case CgroupType: + return &linkCgroup{*raw}, nil + case IterType: + return &Iter{*raw}, nil + case NetNsType: + return &NetNsLink{*raw}, nil + default: + return raw, nil + } +} + // ID uniquely identifies a BPF link. -type ID uint32 +type ID = sys.LinkID // RawLinkOptions control the creation of a raw link. type RawLinkOptions struct { @@ -44,13 +93,55 @@ type RawLinkOptions struct { 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 + // Flags control the attach behaviour. + Flags uint32 } -// RawLinkInfo contains metadata on a link. -type RawLinkInfo struct { +// Info contains metadata on a link. +type Info struct { Type Type ID ID Program ebpf.ProgramID + extra interface{} +} + +type TracingInfo sys.TracingLinkInfo +type CgroupInfo sys.CgroupLinkInfo +type NetNsInfo sys.NetNsLinkInfo +type XDPInfo sys.XDPLinkInfo + +// Tracing returns tracing type-specific link info. +// +// Returns nil if the type-specific link info isn't available. +func (r Info) Tracing() *TracingInfo { + e, _ := r.extra.(*TracingInfo) + return e +} + +// Cgroup returns cgroup type-specific link info. +// +// Returns nil if the type-specific link info isn't available. +func (r Info) Cgroup() *CgroupInfo { + e, _ := r.extra.(*CgroupInfo) + return e +} + +// NetNs returns netns type-specific link info. +// +// Returns nil if the type-specific link info isn't available. +func (r Info) NetNs() *NetNsInfo { + e, _ := r.extra.(*NetNsInfo) + return e +} + +// ExtraNetNs returns XDP type-specific link info. +// +// Returns nil if the type-specific link info isn't available. +func (r Info) XDP() *XDPInfo { + e, _ := r.extra.(*XDPInfo) + return e } // RawLink is the low-level API to bpf_link. @@ -58,7 +149,8 @@ type RawLinkInfo struct { // You should consider using the higher level interfaces in this // package instead. type RawLink struct { - fd *internal.FD + fd *sys.FD + pinnedPath string } // AttachRawLink creates a raw link. @@ -68,66 +160,46 @@ func AttachRawLink(opts RawLinkOptions) (*RawLink, error) { } if opts.Target < 0 { - return nil, fmt.Errorf("invalid target: %s", internal.ErrClosedFd) + return nil, fmt.Errorf("invalid target: %s", sys.ErrClosedFd) } progFd := opts.Program.FD() if progFd < 0 { - return nil, fmt.Errorf("invalid program: %s", internal.ErrClosedFd) + return nil, fmt.Errorf("invalid program: %s", sys.ErrClosedFd) } - attr := bpfLinkCreateAttr{ - targetFd: uint32(opts.Target), - progFd: uint32(progFd), - attachType: opts.Attach, + attr := sys.LinkCreateAttr{ + TargetFd: uint32(opts.Target), + ProgFd: uint32(progFd), + AttachType: sys.AttachType(opts.Attach), + TargetBtfId: uint32(opts.BTF), + Flags: opts.Flags, } - fd, err := bpfLinkCreate(&attr) + fd, err := sys.LinkCreate(&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. -func LoadPinnedRawLink(fileName string) (*RawLink, error) { - return loadPinnedRawLink(fileName, UnspecifiedType) + return &RawLink{fd, ""}, nil } -func loadPinnedRawLink(fileName string, typ Type) (*RawLink, error) { - fd, err := internal.BPFObjGet(fileName) - if err != nil { - return nil, fmt.Errorf("load pinned link: %s", err) - } - - link := &RawLink{fd} - if typ == UnspecifiedType { - return link, nil - } - - info, err := link.Info() +func loadPinnedRawLink(fileName string, opts *ebpf.LoadPinOptions) (*RawLink, error) { + fd, err := sys.ObjGet(&sys.ObjGetAttr{ + Pathname: sys.NewStringPointer(fileName), + FileFlags: opts.Marshal(), + }) if err != nil { - link.Close() - return nil, fmt.Errorf("get pinned link info: %s", err) - } - - if info.Type != typ { - link.Close() - return nil, fmt.Errorf("link type %v doesn't match %v", info.Type, typ) + return nil, fmt.Errorf("load pinned link: %w", err) } - return link, nil + return &RawLink{fd, fileName}, 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) + return l.fd.Int() } // Close breaks the link. @@ -142,13 +214,23 @@ func (l *RawLink) Close() error { // 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.BPFObjPin(fileName, l.fd); err != nil { - return fmt.Errorf("can't pin link: %s", err) + if err := internal.Pin(l.pinnedPath, fileName, l.fd); err != nil { + return err } + l.pinnedPath = fileName return nil } -// Update implements Link. +// 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, @@ -166,49 +248,66 @@ type RawLinkUpdateOptions struct { func (l *RawLink) UpdateArgs(opts RawLinkUpdateOptions) error { newFd := opts.New.FD() if newFd < 0 { - return fmt.Errorf("invalid program: %s", internal.ErrClosedFd) + return fmt.Errorf("invalid program: %s", sys.ErrClosedFd) } var oldFd int if opts.Old != nil { oldFd = opts.Old.FD() if oldFd < 0 { - return fmt.Errorf("invalid replacement program: %s", internal.ErrClosedFd) + return fmt.Errorf("invalid replacement program: %s", sys.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, + attr := sys.LinkUpdateAttr{ + LinkFd: l.fd.Uint(), + 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 + return sys.LinkUpdate(&attr) } // 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 { +func (l *RawLink) Info() (*Info, error) { + var info sys.LinkInfo + + if err := sys.ObjInfo(l.fd, &info); err != nil { return nil, fmt.Errorf("link info: %s", err) } - return &RawLinkInfo{ - Type(info.typ), - ID(info.id), - ebpf.ProgramID(info.prog_id), + var extra interface{} + switch info.Type { + case CgroupType: + extra = &CgroupInfo{} + case IterType: + // not supported + case NetNsType: + extra = &NetNsInfo{} + case RawTracepointType: + // not supported + case TracingType: + extra = &TracingInfo{} + case XDPType: + extra = &XDPInfo{} + case PerfEventType: + // no extra + default: + return nil, fmt.Errorf("unknown link info type: %d", info.Type) + } + + if info.Type != RawTracepointType && info.Type != IterType && info.Type != PerfEventType { + buf := bytes.NewReader(info.Extra[:]) + err := binary.Read(buf, internal.NativeEndian, extra) + if err != nil { + return nil, fmt.Errorf("can not read extra link info: %w", err) + } + } + + return &Info{ + info.Type, + info.Id, + ebpf.ProgramID(info.ProgId), + extra, }, nil } diff --git a/vendor/github.com/cilium/ebpf/link/netns.go b/vendor/github.com/cilium/ebpf/link/netns.go index 3533ff0fa..344ecced6 100644 --- a/vendor/github.com/cilium/ebpf/link/netns.go +++ b/vendor/github.com/cilium/ebpf/link/netns.go @@ -6,14 +6,9 @@ import ( "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 + RawLink } // AttachNetNs attaches a program to a network namespace. @@ -37,24 +32,5 @@ func AttachNetNs(ns int, prog *ebpf.Program) (*NetNsLink, error) { return nil, err } - return &NetNsLink{link}, nil -} - -// LoadPinnedNetNs loads a network namespace link from bpffs. -func LoadPinnedNetNs(fileName string) (*NetNsLink, error) { - link, err := loadPinnedRawLink(fileName, NetNsType) - 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 + return &NetNsLink{*link}, nil } diff --git a/vendor/github.com/cilium/ebpf/link/perf_event.go b/vendor/github.com/cilium/ebpf/link/perf_event.go new file mode 100644 index 000000000..0e5bd4791 --- /dev/null +++ b/vendor/github.com/cilium/ebpf/link/perf_event.go @@ -0,0 +1,394 @@ +package link + +import ( + "bytes" + "errors" + "fmt" + "os" + "path/filepath" + "runtime" + "strconv" + "strings" + "unsafe" + + "github.com/cilium/ebpf" + "github.com/cilium/ebpf/asm" + "github.com/cilium/ebpf/internal" + "github.com/cilium/ebpf/internal/sys" + "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" + + 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 { + // The event type determines the types of programs that can be attached. + typ perfEventType + + // 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 + + // User provided arbitrary value. + cookie uint64 + + // This is the perf event FD. + fd *sys.FD +} + +func (pe *perfEvent) Close() error { + if err := pe.fd.Close(); 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 +} + +// perfEventLink represents a bpf perf link. +type perfEventLink struct { + RawLink + pe *perfEvent +} + +func (pl *perfEventLink) isLink() {} + +// Pinning requires the underlying perf event FD to stay open. +// +// | PerfEvent FD | BpfLink FD | Works | +// |--------------|------------|-------| +// | Open | Open | Yes | +// | Closed | Open | No | +// | Open | Closed | No (Pin() -> EINVAL) | +// | Closed | Closed | No (Pin() -> EINVAL) | +// +// There is currently no pretty way to recover the perf event FD +// when loading a pinned link, so leave as not supported for now. +func (pl *perfEventLink) Pin(string) error { + return fmt.Errorf("perf event link pin: %w", ErrNotSupported) +} + +func (pl *perfEventLink) Unpin() error { + return fmt.Errorf("perf event link unpin: %w", ErrNotSupported) +} + +func (pl *perfEventLink) Close() error { + if err := pl.pe.Close(); err != nil { + return fmt.Errorf("perf event link close: %w", err) + } + return pl.fd.Close() +} + +func (pl *perfEventLink) Update(prog *ebpf.Program) error { + return fmt.Errorf("perf event link update: %w", ErrNotSupported) +} + +// perfEventIoctl implements Link and handles the perf event lifecycle +// via ioctl(). +type perfEventIoctl struct { + *perfEvent +} + +func (pi *perfEventIoctl) isLink() {} + +// 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 (pi *perfEventIoctl) Update(prog *ebpf.Program) error { + return fmt.Errorf("perf event ioctl update: %w", ErrNotSupported) +} + +func (pi *perfEventIoctl) Pin(string) error { + return fmt.Errorf("perf event ioctl pin: %w", ErrNotSupported) +} + +func (pi *perfEventIoctl) Unpin() error { + return fmt.Errorf("perf event ioctl unpin: %w", ErrNotSupported) +} + +func (pi *perfEventIoctl) Info() (*Info, error) { + return nil, fmt.Errorf("perf event ioctl info: %w", ErrNotSupported) +} + +// 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 attachPerfEvent(pe *perfEvent, prog *ebpf.Program) (Link, error) { + if prog == nil { + return nil, errors.New("cannot attach a nil program") + } + if prog.FD() < 0 { + return nil, fmt.Errorf("invalid program: %w", sys.ErrClosedFd) + } + + switch pe.typ { + case kprobeEvent, kretprobeEvent, uprobeEvent, uretprobeEvent: + if t := prog.Type(); t != ebpf.Kprobe { + return nil, fmt.Errorf("invalid program type (expected %s): %s", ebpf.Kprobe, t) + } + case tracepointEvent: + if t := prog.Type(); t != ebpf.TracePoint { + return nil, fmt.Errorf("invalid program type (expected %s): %s", ebpf.TracePoint, t) + } + default: + return nil, fmt.Errorf("unknown perf event type: %d", pe.typ) + } + + if err := haveBPFLinkPerfEvent(); err == nil { + return attachPerfEventLink(pe, prog) + } + return attachPerfEventIoctl(pe, prog) +} + +func attachPerfEventIoctl(pe *perfEvent, prog *ebpf.Program) (*perfEventIoctl, error) { + if pe.cookie != 0 { + return nil, fmt.Errorf("cookies are not supported: %w", ErrNotSupported) + } + + // Assign the eBPF program to the perf event. + err := unix.IoctlSetInt(pe.fd.Int(), unix.PERF_EVENT_IOC_SET_BPF, prog.FD()) + if err != nil { + return nil, fmt.Errorf("setting perf event bpf program: %w", err) + } + + // PERF_EVENT_IOC_ENABLE and _DISABLE ignore their given values. + if err := unix.IoctlSetInt(pe.fd.Int(), unix.PERF_EVENT_IOC_ENABLE, 0); err != nil { + return nil, fmt.Errorf("enable perf event: %s", err) + } + + pi := &perfEventIoctl{pe} + + // Close the perf event when its reference is lost to avoid leaking system resources. + runtime.SetFinalizer(pi, (*perfEventIoctl).Close) + return pi, nil +} + +// Use the bpf api to attach the perf event (BPF_LINK_TYPE_PERF_EVENT, 5.15+). +// +// https://github.com/torvalds/linux/commit/b89fbfbb854c9afc3047e8273cc3a694650b802e +func attachPerfEventLink(pe *perfEvent, prog *ebpf.Program) (*perfEventLink, error) { + fd, err := sys.LinkCreatePerfEvent(&sys.LinkCreatePerfEventAttr{ + ProgFd: uint32(prog.FD()), + TargetFd: pe.fd.Uint(), + AttachType: sys.BPF_PERF_EVENT, + BpfCookie: pe.cookie, + }) + if err != nil { + return nil, fmt.Errorf("cannot create bpf perf link: %v", err) + } + + pl := &perfEventLink{RawLink{fd: fd}, pe} + + // Close the perf event when its reference is lost to avoid leaking system resources. + runtime.SetFinalizer(pl, (*perfEventLink).Close) + return pl, 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. +// The kernel requires group and name to be alphanumeric or underscore. +// +// name automatically has its invalid symbols converted to underscores so the caller +// can pass a raw symbol name, e.g. a kernel symbol containing dots. +func getTraceEventID(group, name string) (uint64, error) { + name = sanitizeSymbol(name) + 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) (*sys.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 sys.NewFD(fd) +} + +// 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) +} + +// Probe BPF perf link. +// +// https://elixir.bootlin.com/linux/v5.16.8/source/kernel/bpf/syscall.c#L4307 +// https://github.com/torvalds/linux/commit/b89fbfbb854c9afc3047e8273cc3a694650b802e +var haveBPFLinkPerfEvent = internal.FeatureTest("bpf_link_perf_event", "5.15", func() error { + prog, err := ebpf.NewProgram(&ebpf.ProgramSpec{ + Name: "probe_bpf_perf_link", + Type: ebpf.Kprobe, + Instructions: asm.Instructions{ + asm.Mov.Imm(asm.R0, 0), + asm.Return(), + }, + License: "MIT", + }) + if err != nil { + return err + } + defer prog.Close() + + _, err = sys.LinkCreatePerfEvent(&sys.LinkCreatePerfEventAttr{ + ProgFd: uint32(prog.FD()), + AttachType: sys.BPF_PERF_EVENT, + }) + if errors.Is(err, unix.EINVAL) { + return internal.ErrNotSupported + } + if errors.Is(err, unix.EBADF) { + return nil + } + return err +}) + +// isValidTraceID implements the equivalent of a regex match +// against "^[a-zA-Z_][0-9a-zA-Z_]*$". +// +// 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. +func isValidTraceID(s string) bool { + if len(s) < 1 { + return false + } + for i, c := range []byte(s) { + switch { + case c >= 'a' && c <= 'z': + case c >= 'A' && c <= 'Z': + case c == '_': + case i > 0 && c >= '0' && c <= '9': + + default: + return false + } + } + + return true +} diff --git a/vendor/github.com/cilium/ebpf/link/platform.go b/vendor/github.com/cilium/ebpf/link/platform.go new file mode 100644 index 000000000..eb6f7b7a3 --- /dev/null +++ b/vendor/github.com/cilium/ebpf/link/platform.go @@ -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) +} diff --git a/vendor/github.com/cilium/ebpf/link/program.go b/vendor/github.com/cilium/ebpf/link/program.go index 0fe9d37c4..ea3181737 100644 --- a/vendor/github.com/cilium/ebpf/link/program.go +++ b/vendor/github.com/cilium/ebpf/link/program.go @@ -4,7 +4,7 @@ import ( "fmt" "github.com/cilium/ebpf" - "github.com/cilium/ebpf/internal" + "github.com/cilium/ebpf/internal/sys" ) type RawAttachProgramOptions struct { @@ -34,7 +34,7 @@ func RawAttachProgram(opts RawAttachProgramOptions) error { replaceFd = uint32(opts.Replace.FD()) } - attr := internal.BPFProgAttachAttr{ + attr := sys.ProgAttachAttr{ TargetFd: uint32(opts.Target), AttachBpfFd: uint32(opts.Program.FD()), ReplaceBpfFd: replaceFd, @@ -42,8 +42,8 @@ func RawAttachProgram(opts RawAttachProgramOptions) error { AttachFlags: uint32(opts.Flags), } - if err := internal.BPFProgAttach(&attr); err != nil { - return fmt.Errorf("can't attach program: %s", err) + if err := sys.ProgAttach(&attr); err != nil { + return fmt.Errorf("can't attach program: %w", err) } return nil } @@ -63,13 +63,13 @@ func RawDetachProgram(opts RawDetachProgramOptions) error { return err } - attr := internal.BPFProgDetachAttr{ + attr := sys.ProgDetachAttr{ 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: %s", err) + if err := sys.ProgDetach(&attr); err != nil { + return fmt.Errorf("can't detach program: %w", err) } return nil diff --git a/vendor/github.com/cilium/ebpf/link/raw_tracepoint.go b/vendor/github.com/cilium/ebpf/link/raw_tracepoint.go index 65652486f..925e621cb 100644 --- a/vendor/github.com/cilium/ebpf/link/raw_tracepoint.go +++ b/vendor/github.com/cilium/ebpf/link/raw_tracepoint.go @@ -1,10 +1,11 @@ package link import ( + "errors" "fmt" "github.com/cilium/ebpf" - "github.com/cilium/ebpf/internal" + "github.com/cilium/ebpf/internal/sys" ) type RawTracepointOptions struct { @@ -22,36 +23,65 @@ func AttachRawTracepoint(opts RawTracepointOptions) (Link, error) { 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) + return nil, fmt.Errorf("invalid program: %w", sys.ErrClosedFd) } - fd, err := bpfRawTracepointOpen(&bpfRawTracepointOpenAttr{ - name: internal.NewStringPointer(opts.Name), - fd: uint32(opts.Program.FD()), + fd, err := sys.RawTracepointOpen(&sys.RawTracepointOpenAttr{ + Name: sys.NewStringPointer(opts.Name), + ProgFd: uint32(opts.Program.FD()), }) if err != nil { return nil, err } - return &progAttachRawTracepoint{fd: fd}, nil + err = haveBPFLink() + if errors.Is(err, ErrNotSupported) { + // Prior to commit 70ed506c3bbc ("bpf: Introduce pinnable bpf_link abstraction") + // raw_tracepoints are just a plain fd. + return &simpleRawTracepoint{fd}, nil + } + + if err != nil { + return nil, err + } + + return &rawTracepoint{RawLink{fd: fd}}, nil } -type progAttachRawTracepoint struct { - fd *internal.FD +type simpleRawTracepoint struct { + fd *sys.FD } -var _ Link = (*progAttachRawTracepoint)(nil) +var _ Link = (*simpleRawTracepoint)(nil) -func (rt *progAttachRawTracepoint) isLink() {} +func (frt *simpleRawTracepoint) isLink() {} + +func (frt *simpleRawTracepoint) Close() error { + return frt.fd.Close() +} -func (rt *progAttachRawTracepoint) Close() error { - return rt.fd.Close() +func (frt *simpleRawTracepoint) Update(_ *ebpf.Program) error { + return fmt.Errorf("update raw_tracepoint: %w", ErrNotSupported) } -func (rt *progAttachRawTracepoint) Update(_ *ebpf.Program) error { - return fmt.Errorf("can't update raw_tracepoint: %w", ErrNotSupported) +func (frt *simpleRawTracepoint) Pin(string) error { + return fmt.Errorf("pin raw_tracepoint: %w", ErrNotSupported) } -func (rt *progAttachRawTracepoint) Pin(_ string) error { - return fmt.Errorf("can't pin raw_tracepoint: %w", ErrNotSupported) +func (frt *simpleRawTracepoint) Unpin() error { + return fmt.Errorf("unpin raw_tracepoint: %w", ErrNotSupported) +} + +func (frt *simpleRawTracepoint) Info() (*Info, error) { + return nil, fmt.Errorf("can't get raw_tracepoint info: %w", ErrNotSupported) +} + +type rawTracepoint struct { + RawLink +} + +var _ Link = (*rawTracepoint)(nil) + +func (rt *rawTracepoint) Update(_ *ebpf.Program) error { + return fmt.Errorf("update raw_tracepoint: %w", ErrNotSupported) } diff --git a/vendor/github.com/cilium/ebpf/link/socket_filter.go b/vendor/github.com/cilium/ebpf/link/socket_filter.go new file mode 100644 index 000000000..94f3958cc --- /dev/null +++ b/vendor/github.com/cilium/ebpf/link/socket_filter.go @@ -0,0 +1,40 @@ +package link + +import ( + "syscall" + + "github.com/cilium/ebpf" + "github.com/cilium/ebpf/internal/unix" +) + +// AttachSocketFilter attaches a SocketFilter BPF program to a socket. +func AttachSocketFilter(conn syscall.Conn, program *ebpf.Program) error { + rawConn, err := conn.SyscallConn() + if err != nil { + return err + } + var ssoErr error + err = rawConn.Control(func(fd uintptr) { + ssoErr = syscall.SetsockoptInt(int(fd), unix.SOL_SOCKET, unix.SO_ATTACH_BPF, program.FD()) + }) + if ssoErr != nil { + return ssoErr + } + return err +} + +// DetachSocketFilter detaches a SocketFilter BPF program from a socket. +func DetachSocketFilter(conn syscall.Conn) error { + rawConn, err := conn.SyscallConn() + if err != nil { + return err + } + var ssoErr error + err = rawConn.Control(func(fd uintptr) { + ssoErr = syscall.SetsockoptInt(int(fd), unix.SOL_SOCKET, unix.SO_DETACH_BPF, 0) + }) + if ssoErr != nil { + return ssoErr + } + return err +} diff --git a/vendor/github.com/cilium/ebpf/link/syscalls.go b/vendor/github.com/cilium/ebpf/link/syscalls.go index 19326c8af..a661395b3 100644 --- a/vendor/github.com/cilium/ebpf/link/syscalls.go +++ b/vendor/github.com/cilium/ebpf/link/syscalls.go @@ -2,35 +2,33 @@ package link import ( "errors" - "unsafe" "github.com/cilium/ebpf" "github.com/cilium/ebpf/asm" "github.com/cilium/ebpf/internal" + "github.com/cilium/ebpf/internal/sys" "github.com/cilium/ebpf/internal/unix" ) // Type is the kind of link. -type Type uint32 +type Type = sys.LinkType // Valid link types. -// -// Equivalent to enum bpf_link_type. const ( - UnspecifiedType Type = iota - RawTracepointType - TracingType - CgroupType - IterType - NetNsType - XDPType + UnspecifiedType = sys.BPF_LINK_TYPE_UNSPEC + RawTracepointType = sys.BPF_LINK_TYPE_RAW_TRACEPOINT + TracingType = sys.BPF_LINK_TYPE_TRACING + CgroupType = sys.BPF_LINK_TYPE_CGROUP + IterType = sys.BPF_LINK_TYPE_ITER + NetNsType = sys.BPF_LINK_TYPE_NETNS + XDPType = sys.BPF_LINK_TYPE_XDP + PerfEventType = sys.BPF_LINK_TYPE_PERF_EVENT ) 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", + Type: ebpf.CGroupSKB, + License: "MIT", Instructions: asm.Instructions{ asm.Mov.Imm(asm.R0, 0), asm.Return(), @@ -69,7 +67,7 @@ var haveProgAttachReplace = internal.FeatureTest("BPF_PROG_ATTACH atomic replace // 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{ + attr := sys.ProgAttachAttr{ // We rely on this being checked after attachFlags. TargetFd: ^uint32(0), AttachBpfFd: uint32(prog.FD()), @@ -77,7 +75,7 @@ var haveProgAttachReplace = internal.FeatureTest("BPF_PROG_ATTACH atomic replace AttachFlags: uint32(flagReplace), } - err = internal.BPFProgAttach(&attr) + err = sys.ProgAttach(&attr) if errors.Is(err, unix.EINVAL) { return internal.ErrNotSupported } @@ -87,55 +85,14 @@ var haveProgAttachReplace = internal.FeatureTest("BPF_PROG_ATTACH atomic replace return err }) -type bpfLinkCreateAttr struct { - progFd uint32 - targetFd uint32 - attachType ebpf.AttachType - flags 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 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{ + attr := sys.LinkCreateAttr{ // This is a hopefully invalid file descriptor, which triggers EBADF. - targetFd: ^uint32(0), - progFd: uint32(prog.FD()), - attachType: ebpf.AttachCGroupInetIngress, + TargetFd: ^uint32(0), + ProgFd: ^uint32(0), + AttachType: sys.AttachType(ebpf.AttachCGroupInetIngress), } - _, err = bpfLinkCreate(&attr) + _, err := sys.LinkCreate(&attr) if errors.Is(err, unix.EINVAL) { return internal.ErrNotSupported } @@ -144,30 +101,3 @@ var haveBPFLink = internal.FeatureTest("bpf_link", "5.7", func() error { } 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 -} diff --git a/vendor/github.com/cilium/ebpf/link/tracepoint.go b/vendor/github.com/cilium/ebpf/link/tracepoint.go new file mode 100644 index 000000000..a59ef9d1c --- /dev/null +++ b/vendor/github.com/cilium/ebpf/link/tracepoint.go @@ -0,0 +1,77 @@ +package link + +import ( + "fmt" + + "github.com/cilium/ebpf" +) + +// TracepointOptions defines additional parameters that will be used +// when loading Tracepoints. +type TracepointOptions struct { + // Arbitrary value that can be fetched from an eBPF program + // via `bpf_get_attach_cookie()`. + // + // Needs kernel 5.15+. + Cookie uint64 +} + +// 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, nil) +// +// 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, opts *TracepointOptions) (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 !isValidTraceID(group) || !isValidTraceID(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 + } + + var cookie uint64 + if opts != nil { + cookie = opts.Cookie + } + + pe := &perfEvent{ + typ: tracepointEvent, + group: group, + name: name, + tracefsID: tid, + cookie: cookie, + fd: fd, + } + + lnk, err := attachPerfEvent(pe, prog) + if err != nil { + pe.Close() + return nil, err + } + + return lnk, nil +} diff --git a/vendor/github.com/cilium/ebpf/link/tracing.go b/vendor/github.com/cilium/ebpf/link/tracing.go new file mode 100644 index 000000000..e47e61a3b --- /dev/null +++ b/vendor/github.com/cilium/ebpf/link/tracing.go @@ -0,0 +1,141 @@ +package link + +import ( + "fmt" + + "github.com/cilium/ebpf" + "github.com/cilium/ebpf/btf" + "github.com/cilium/ebpf/internal/sys" +) + +type tracing struct { + RawLink +} + +func (f *tracing) Update(new *ebpf.Program) error { + return fmt.Errorf("tracing update: %w", ErrNotSupported) +} + +// 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) (Link, 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 { + btfHandle, err := targetProg.Handle() + if err != nil { + return nil, err + } + defer btfHandle.Close() + + spec, err := btfHandle.Spec(nil) + if err != nil { + return nil, err + } + + var function *btf.Func + if err := spec.TypeByName(name, &function); err != nil { + return nil, err + } + + target = targetProg.FD() + typeID, err = spec.TypeID(function) + if err != nil { + return nil, err + } + } + + link, err := AttachRawLink(RawLinkOptions{ + Target: target, + Program: prog, + Attach: ebpf.AttachNone, + BTF: typeID, + }) + if err != nil { + return nil, err + } + + return &tracing{*link}, nil +} + +type TracingOptions struct { + // Program must be of type Tracing with attach type + // AttachTraceFEntry/AttachTraceFExit/AttachModifyReturn or + // AttachTraceRawTp. + Program *ebpf.Program +} + +type LSMOptions struct { + // Program must be of type LSM with attach type + // AttachLSMMac. + Program *ebpf.Program +} + +// attachBTFID links all BPF program types (Tracing/LSM) that they attach to a btf_id. +func attachBTFID(program *ebpf.Program) (Link, error) { + if program.FD() < 0 { + return nil, fmt.Errorf("invalid program %w", sys.ErrClosedFd) + } + + fd, err := sys.RawTracepointOpen(&sys.RawTracepointOpenAttr{ + ProgFd: uint32(program.FD()), + }) + if err != nil { + return nil, err + } + + raw := RawLink{fd: fd} + info, err := raw.Info() + if err != nil { + raw.Close() + return nil, err + } + + if info.Type == RawTracepointType { + // Sadness upon sadness: a Tracing program with AttachRawTp returns + // a raw_tracepoint link. Other types return a tracing link. + return &rawTracepoint{raw}, nil + } + + return &tracing{RawLink: RawLink{fd: fd}}, nil +} + +// AttachTracing links a tracing (fentry/fexit/fmod_ret) BPF program or +// a BTF-powered raw tracepoint (tp_btf) BPF Program to a BPF hook defined +// in kernel modules. +func AttachTracing(opts TracingOptions) (Link, error) { + if t := opts.Program.Type(); t != ebpf.Tracing { + return nil, fmt.Errorf("invalid program type %s, expected Tracing", t) + } + + return attachBTFID(opts.Program) +} + +// AttachLSM links a Linux security module (LSM) BPF Program to a BPF +// hook defined in kernel modules. +func AttachLSM(opts LSMOptions) (Link, error) { + if t := opts.Program.Type(); t != ebpf.LSM { + return nil, fmt.Errorf("invalid program type %s, expected LSM", t) + } + + return attachBTFID(opts.Program) +} diff --git a/vendor/github.com/cilium/ebpf/link/uprobe.go b/vendor/github.com/cilium/ebpf/link/uprobe.go new file mode 100644 index 000000000..edf925b57 --- /dev/null +++ b/vendor/github.com/cilium/ebpf/link/uprobe.go @@ -0,0 +1,373 @@ +package link + +import ( + "debug/elf" + "errors" + "fmt" + "os" + "path/filepath" + "strings" + "sync" + + "github.com/cilium/ebpf" + "github.com/cilium/ebpf/internal" +) + +var ( + uprobeEventsPath = filepath.Join(tracefsPath, "uprobe_events") + + uprobeRetprobeBit = struct { + once sync.Once + value uint64 + err error + }{} + + uprobeRefCtrOffsetPMUPath = "/sys/bus/event_source/devices/uprobe/format/ref_ctr_offset" + // elixir.bootlin.com/linux/v5.15-rc7/source/kernel/events/core.c#L9799 + uprobeRefCtrOffsetShift = 32 + haveRefCtrOffsetPMU = internal.FeatureTest("RefCtrOffsetPMU", "4.20", func() error { + _, err := os.Stat(uprobeRefCtrOffsetPMUPath) + if err != nil { + return internal.ErrNotSupported + } + return nil + }) + + // 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 and dynamic symbols' addresses. + addresses map[string]uint64 +} + +// UprobeOptions defines additional parameters that will be used +// when loading Uprobes. +type UprobeOptions struct { + // Symbol address. Must be provided in case of external symbols (shared libs). + // If set, overrides the address eventually parsed from the executable. + Address uint64 + // The offset relative to given symbol. Useful when tracing an arbitrary point + // inside the frame of given symbol. + // + // Note: this field changed from being an absolute offset to being relative + // to Address. + 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 + // Automatically manage SDT reference counts (semaphores). + // + // If this field is set, the Kernel will increment/decrement the + // semaphore located in the process memory at the provided address on + // probe attach/detach. + // + // See also: + // sourceware.org/systemtap/wiki/UserSpaceProbeImplementation (Semaphore Handling) + // github.com/torvalds/linux/commit/1cc33161a83d + // github.com/torvalds/linux/commit/a6ca88b241d5 + RefCtrOffset uint64 + // Arbitrary value that can be fetched from an eBPF program + // via `bpf_get_attach_cookie()`. + // + // Needs kernel 5.15+. + Cookie uint64 +} + +// 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, + addresses: 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 + } + + address := 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 + address = s.Value - prog.Vaddr + prog.Off + break + } + } + + ex.addresses[s.Name] = address + } + + return nil +} + +// address calculates the address of a symbol in the executable. +// +// opts must not be nil. +func (ex *Executable) address(symbol string, opts *UprobeOptions) (uint64, error) { + if opts.Address > 0 { + return opts.Address + opts.Offset, nil + } + + address, ok := ex.addresses[symbol] + if !ok { + return 0, fmt.Errorf("symbol %s: %w", symbol, ErrNoSymbol) + } + + // 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 address == 0 { + return 0, fmt.Errorf("cannot resolve %s library call '%s': %w "+ + "(consider providing UprobeOptions.Address)", ex.path, symbol, ErrNotSupported) + } + + return address + opts.Offset, nil +} + +// 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}) +// +// Note: Setting the Offset field in the options supersedes the symbol's offset. +// +// 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 + } + + lnk, err := attachPerfEvent(u, prog) + if err != nil { + u.Close() + return nil, err + } + + return lnk, 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}) +// +// Note: Setting the Offset field in the options supersedes the symbol's offset. +// +// 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 + } + + lnk, err := attachPerfEvent(u, prog) + if err != nil { + u.Close() + return nil, err + } + + return lnk, 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) + } + if opts == nil { + opts = &UprobeOptions{} + } + + offset, err := ex.address(symbol, opts) + if err != nil { + return nil, err + } + + pid := opts.PID + if pid == 0 { + pid = perfAllThreads + } + + if opts.RefCtrOffset != 0 { + if err := haveRefCtrOffsetPMU(); err != nil { + return nil, fmt.Errorf("uprobe ref_ctr_offset: %w", err) + } + } + + args := probeArgs{ + symbol: symbol, + path: ex.path, + offset: offset, + pid: pid, + refCtrOffset: opts.RefCtrOffset, + ret: ret, + cookie: opts.Cookie, + } + + // Use uprobe PMU if the kernel has it available. + tp, err := pmuUprobe(args) + 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. + args.symbol = sanitizeSymbol(symbol) + tp, err = tracefsUprobe(args) + 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(args probeArgs) (*perfEvent, error) { + return pmuProbe(uprobeType, args) +} + +// tracefsUprobe creates a Uprobe tracefs entry. +func tracefsUprobe(args probeArgs) (*perfEvent, error) { + return tracefsProbe(uprobeType, args) +} + +// sanitizeSymbol replaces every invalid character for the tracefs api with an underscore. +// It is equivalent to calling regexp.MustCompile("[^a-zA-Z0-9]+").ReplaceAllString("_"). +func sanitizeSymbol(s string) string { + var b strings.Builder + b.Grow(len(s)) + var skip bool + for _, c := range []byte(s) { + switch { + case c >= 'a' && c <= 'z', + c >= 'A' && c <= 'Z', + c >= '0' && c <= '9': + skip = false + b.WriteByte(c) + + default: + if !skip { + b.WriteByte('_') + skip = true + } + } + } + + return b.String() +} + +// uprobeToken creates the PATH:OFFSET(REF_CTR_OFFSET) token for the tracefs api. +func uprobeToken(args probeArgs) string { + po := fmt.Sprintf("%s:%#x", args.path, args.offset) + + if args.refCtrOffset != 0 { + // This is not documented in Documentation/trace/uprobetracer.txt. + // elixir.bootlin.com/linux/v5.15-rc7/source/kernel/trace/trace.c#L5564 + po += fmt.Sprintf("(%#x)", args.refCtrOffset) + } + + return po +} + +func uretprobeBit() (uint64, error) { + uprobeRetprobeBit.once.Do(func() { + uprobeRetprobeBit.value, uprobeRetprobeBit.err = determineRetprobeBit(uprobeType) + }) + return uprobeRetprobeBit.value, uprobeRetprobeBit.err +} diff --git a/vendor/github.com/cilium/ebpf/link/xdp.go b/vendor/github.com/cilium/ebpf/link/xdp.go new file mode 100644 index 000000000..aa8dd3a4c --- /dev/null +++ b/vendor/github.com/cilium/ebpf/link/xdp.go @@ -0,0 +1,54 @@ +package link + +import ( + "fmt" + + "github.com/cilium/ebpf" +) + +// XDPAttachFlags represents how XDP program will be attached to interface. +type XDPAttachFlags uint32 + +const ( + // XDPGenericMode (SKB) links XDP BPF program for drivers which do + // not yet support native XDP. + XDPGenericMode XDPAttachFlags = 1 << (iota + 1) + // XDPDriverMode links XDP BPF program into the driver’s receive path. + XDPDriverMode + // XDPOffloadMode offloads the entire XDP BPF program into hardware. + XDPOffloadMode +) + +type XDPOptions struct { + // Program must be an XDP BPF program. + Program *ebpf.Program + + // Interface is the interface index to attach program to. + Interface int + + // Flags is one of XDPAttachFlags (optional). + // + // Only one XDP mode should be set, without flag defaults + // to driver/generic mode (best effort). + Flags XDPAttachFlags +} + +// AttachXDP links an XDP BPF program to an XDP hook. +func AttachXDP(opts XDPOptions) (Link, error) { + if t := opts.Program.Type(); t != ebpf.XDP { + return nil, fmt.Errorf("invalid program type %s, expected XDP", t) + } + + if opts.Interface < 1 { + return nil, fmt.Errorf("invalid interface index: %d", opts.Interface) + } + + rawLink, err := AttachRawLink(RawLinkOptions{ + Program: opts.Program, + Attach: ebpf.AttachXDP, + Target: opts.Interface, + Flags: uint32(opts.Flags), + }) + + return rawLink, err +} |