1// Copyright 2009 The Go Authors. All rights reserved.
   2// Use of this source code is governed by a BSD-style
   3// license that can be found in the LICENSE file.
   4
   5// Linux system calls.
   6// This file is compiled as ordinary Go code,
   7// but it is also input to mksyscall,
   8// which parses the //sys lines and generates system call stubs.
   9// Note that sometimes we use a lowercase //sys name and
  10// wrap it in our own nicer implementation.
  11
  12package unix
  13
  14import (
  15	"encoding/binary"
  16	"strconv"
  17	"syscall"
  18	"time"
  19	"unsafe"
  20)
  21
  22/*
  23 * Wrapped
  24 */
  25
  26func Access(path string, mode uint32) (err error) {
  27	return Faccessat(AT_FDCWD, path, mode, 0)
  28}
  29
  30func Chmod(path string, mode uint32) (err error) {
  31	return Fchmodat(AT_FDCWD, path, mode, 0)
  32}
  33
  34func Chown(path string, uid int, gid int) (err error) {
  35	return Fchownat(AT_FDCWD, path, uid, gid, 0)
  36}
  37
  38func Creat(path string, mode uint32) (fd int, err error) {
  39	return Open(path, O_CREAT|O_WRONLY|O_TRUNC, mode)
  40}
  41
  42func EpollCreate(size int) (fd int, err error) {
  43	if size <= 0 {
  44		return -1, EINVAL
  45	}
  46	return EpollCreate1(0)
  47}
  48
  49//sys	FanotifyInit(flags uint, event_f_flags uint) (fd int, err error)
  50//sys	fanotifyMark(fd int, flags uint, mask uint64, dirFd int, pathname *byte) (err error)
  51
  52func FanotifyMark(fd int, flags uint, mask uint64, dirFd int, pathname string) (err error) {
  53	if pathname == "" {
  54		return fanotifyMark(fd, flags, mask, dirFd, nil)
  55	}
  56	p, err := BytePtrFromString(pathname)
  57	if err != nil {
  58		return err
  59	}
  60	return fanotifyMark(fd, flags, mask, dirFd, p)
  61}
  62
  63//sys	fchmodat(dirfd int, path string, mode uint32) (err error)
  64
  65func Fchmodat(dirfd int, path string, mode uint32, flags int) (err error) {
  66	// Linux fchmodat doesn't support the flags parameter. Mimick glibc's behavior
  67	// and check the flags. Otherwise the mode would be applied to the symlink
  68	// destination which is not what the user expects.
  69	if flags&^AT_SYMLINK_NOFOLLOW != 0 {
  70		return EINVAL
  71	} else if flags&AT_SYMLINK_NOFOLLOW != 0 {
  72		return EOPNOTSUPP
  73	}
  74	return fchmodat(dirfd, path, mode)
  75}
  76
  77func InotifyInit() (fd int, err error) {
  78	return InotifyInit1(0)
  79}
  80
  81//sys	ioctl(fd int, req uint, arg uintptr) (err error) = SYS_IOCTL
  82//sys	ioctlPtr(fd int, req uint, arg unsafe.Pointer) (err error) = SYS_IOCTL
  83
  84// ioctl itself should not be exposed directly, but additional get/set functions
  85// for specific types are permissible. These are defined in ioctl.go and
  86// ioctl_linux.go.
  87//
  88// The third argument to ioctl is often a pointer but sometimes an integer.
  89// Callers should use ioctlPtr when the third argument is a pointer and ioctl
  90// when the third argument is an integer.
  91//
  92// TODO: some existing code incorrectly uses ioctl when it should use ioctlPtr.
  93
  94//sys	Linkat(olddirfd int, oldpath string, newdirfd int, newpath string, flags int) (err error)
  95
  96func Link(oldpath string, newpath string) (err error) {
  97	return Linkat(AT_FDCWD, oldpath, AT_FDCWD, newpath, 0)
  98}
  99
 100func Mkdir(path string, mode uint32) (err error) {
 101	return Mkdirat(AT_FDCWD, path, mode)
 102}
 103
 104func Mknod(path string, mode uint32, dev int) (err error) {
 105	return Mknodat(AT_FDCWD, path, mode, dev)
 106}
 107
 108func Open(path string, mode int, perm uint32) (fd int, err error) {
 109	return openat(AT_FDCWD, path, mode|O_LARGEFILE, perm)
 110}
 111
 112//sys	openat(dirfd int, path string, flags int, mode uint32) (fd int, err error)
 113
 114func Openat(dirfd int, path string, flags int, mode uint32) (fd int, err error) {
 115	return openat(dirfd, path, flags|O_LARGEFILE, mode)
 116}
 117
 118//sys	openat2(dirfd int, path string, open_how *OpenHow, size int) (fd int, err error)
 119
 120func Openat2(dirfd int, path string, how *OpenHow) (fd int, err error) {
 121	return openat2(dirfd, path, how, SizeofOpenHow)
 122}
 123
 124func Pipe(p []int) error {
 125	return Pipe2(p, 0)
 126}
 127
 128//sysnb	pipe2(p *[2]_C_int, flags int) (err error)
 129
 130func Pipe2(p []int, flags int) error {
 131	if len(p) != 2 {
 132		return EINVAL
 133	}
 134	var pp [2]_C_int
 135	err := pipe2(&pp, flags)
 136	if err == nil {
 137		p[0] = int(pp[0])
 138		p[1] = int(pp[1])
 139	}
 140	return err
 141}
 142
 143//sys	ppoll(fds *PollFd, nfds int, timeout *Timespec, sigmask *Sigset_t) (n int, err error)
 144
 145func Ppoll(fds []PollFd, timeout *Timespec, sigmask *Sigset_t) (n int, err error) {
 146	if len(fds) == 0 {
 147		return ppoll(nil, 0, timeout, sigmask)
 148	}
 149	return ppoll(&fds[0], len(fds), timeout, sigmask)
 150}
 151
 152func Poll(fds []PollFd, timeout int) (n int, err error) {
 153	var ts *Timespec
 154	if timeout >= 0 {
 155		ts = new(Timespec)
 156		*ts = NsecToTimespec(int64(timeout) * 1e6)
 157	}
 158	return Ppoll(fds, ts, nil)
 159}
 160
 161//sys	Readlinkat(dirfd int, path string, buf []byte) (n int, err error)
 162
 163func Readlink(path string, buf []byte) (n int, err error) {
 164	return Readlinkat(AT_FDCWD, path, buf)
 165}
 166
 167func Rename(oldpath string, newpath string) (err error) {
 168	return Renameat(AT_FDCWD, oldpath, AT_FDCWD, newpath)
 169}
 170
 171func Rmdir(path string) error {
 172	return Unlinkat(AT_FDCWD, path, AT_REMOVEDIR)
 173}
 174
 175//sys	Symlinkat(oldpath string, newdirfd int, newpath string) (err error)
 176
 177func Symlink(oldpath string, newpath string) (err error) {
 178	return Symlinkat(oldpath, AT_FDCWD, newpath)
 179}
 180
 181func Unlink(path string) error {
 182	return Unlinkat(AT_FDCWD, path, 0)
 183}
 184
 185//sys	Unlinkat(dirfd int, path string, flags int) (err error)
 186
 187func Utimes(path string, tv []Timeval) error {
 188	if tv == nil {
 189		err := utimensat(AT_FDCWD, path, nil, 0)
 190		if err != ENOSYS {
 191			return err
 192		}
 193		return utimes(path, nil)
 194	}
 195	if len(tv) != 2 {
 196		return EINVAL
 197	}
 198	var ts [2]Timespec
 199	ts[0] = NsecToTimespec(TimevalToNsec(tv[0]))
 200	ts[1] = NsecToTimespec(TimevalToNsec(tv[1]))
 201	err := utimensat(AT_FDCWD, path, (*[2]Timespec)(unsafe.Pointer(&ts[0])), 0)
 202	if err != ENOSYS {
 203		return err
 204	}
 205	return utimes(path, (*[2]Timeval)(unsafe.Pointer(&tv[0])))
 206}
 207
 208//sys	utimensat(dirfd int, path string, times *[2]Timespec, flags int) (err error)
 209
 210func UtimesNano(path string, ts []Timespec) error {
 211	return UtimesNanoAt(AT_FDCWD, path, ts, 0)
 212}
 213
 214func UtimesNanoAt(dirfd int, path string, ts []Timespec, flags int) error {
 215	if ts == nil {
 216		return utimensat(dirfd, path, nil, flags)
 217	}
 218	if len(ts) != 2 {
 219		return EINVAL
 220	}
 221	return utimensat(dirfd, path, (*[2]Timespec)(unsafe.Pointer(&ts[0])), flags)
 222}
 223
 224func Futimesat(dirfd int, path string, tv []Timeval) error {
 225	if tv == nil {
 226		return futimesat(dirfd, path, nil)
 227	}
 228	if len(tv) != 2 {
 229		return EINVAL
 230	}
 231	return futimesat(dirfd, path, (*[2]Timeval)(unsafe.Pointer(&tv[0])))
 232}
 233
 234func Futimes(fd int, tv []Timeval) (err error) {
 235	// Believe it or not, this is the best we can do on Linux
 236	// (and is what glibc does).
 237	return Utimes("/proc/self/fd/"+strconv.Itoa(fd), tv)
 238}
 239
 240const ImplementsGetwd = true
 241
 242//sys	Getcwd(buf []byte) (n int, err error)
 243
 244func Getwd() (wd string, err error) {
 245	var buf [PathMax]byte
 246	n, err := Getcwd(buf[0:])
 247	if err != nil {
 248		return "", err
 249	}
 250	// Getcwd returns the number of bytes written to buf, including the NUL.
 251	if n < 1 || n > len(buf) || buf[n-1] != 0 {
 252		return "", EINVAL
 253	}
 254	// In some cases, Linux can return a path that starts with the
 255	// "(unreachable)" prefix, which can potentially be a valid relative
 256	// path. To work around that, return ENOENT if path is not absolute.
 257	if buf[0] != '/' {
 258		return "", ENOENT
 259	}
 260
 261	return string(buf[0 : n-1]), nil
 262}
 263
 264func Getgroups() (gids []int, err error) {
 265	n, err := getgroups(0, nil)
 266	if err != nil {
 267		return nil, err
 268	}
 269	if n == 0 {
 270		return nil, nil
 271	}
 272
 273	// Sanity check group count. Max is 1<<16 on Linux.
 274	if n < 0 || n > 1<<20 {
 275		return nil, EINVAL
 276	}
 277
 278	a := make([]_Gid_t, n)
 279	n, err = getgroups(n, &a[0])
 280	if err != nil {
 281		return nil, err
 282	}
 283	gids = make([]int, n)
 284	for i, v := range a[0:n] {
 285		gids[i] = int(v)
 286	}
 287	return
 288}
 289
 290func Setgroups(gids []int) (err error) {
 291	if len(gids) == 0 {
 292		return setgroups(0, nil)
 293	}
 294
 295	a := make([]_Gid_t, len(gids))
 296	for i, v := range gids {
 297		a[i] = _Gid_t(v)
 298	}
 299	return setgroups(len(a), &a[0])
 300}
 301
 302type WaitStatus uint32
 303
 304// Wait status is 7 bits at bottom, either 0 (exited),
 305// 0x7F (stopped), or a signal number that caused an exit.
 306// The 0x80 bit is whether there was a core dump.
 307// An extra number (exit code, signal causing a stop)
 308// is in the high bits. At least that's the idea.
 309// There are various irregularities. For example, the
 310// "continued" status is 0xFFFF, distinguishing itself
 311// from stopped via the core dump bit.
 312
 313const (
 314	mask    = 0x7F
 315	core    = 0x80
 316	exited  = 0x00
 317	stopped = 0x7F
 318	shift   = 8
 319)
 320
 321func (w WaitStatus) Exited() bool { return w&mask == exited }
 322
 323func (w WaitStatus) Signaled() bool { return w&mask != stopped && w&mask != exited }
 324
 325func (w WaitStatus) Stopped() bool { return w&0xFF == stopped }
 326
 327func (w WaitStatus) Continued() bool { return w == 0xFFFF }
 328
 329func (w WaitStatus) CoreDump() bool { return w.Signaled() && w&core != 0 }
 330
 331func (w WaitStatus) ExitStatus() int {
 332	if !w.Exited() {
 333		return -1
 334	}
 335	return int(w>>shift) & 0xFF
 336}
 337
 338func (w WaitStatus) Signal() syscall.Signal {
 339	if !w.Signaled() {
 340		return -1
 341	}
 342	return syscall.Signal(w & mask)
 343}
 344
 345func (w WaitStatus) StopSignal() syscall.Signal {
 346	if !w.Stopped() {
 347		return -1
 348	}
 349	return syscall.Signal(w>>shift) & 0xFF
 350}
 351
 352func (w WaitStatus) TrapCause() int {
 353	if w.StopSignal() != SIGTRAP {
 354		return -1
 355	}
 356	return int(w>>shift) >> 8
 357}
 358
 359//sys	wait4(pid int, wstatus *_C_int, options int, rusage *Rusage) (wpid int, err error)
 360
 361func Wait4(pid int, wstatus *WaitStatus, options int, rusage *Rusage) (wpid int, err error) {
 362	var status _C_int
 363	wpid, err = wait4(pid, &status, options, rusage)
 364	if wstatus != nil {
 365		*wstatus = WaitStatus(status)
 366	}
 367	return
 368}
 369
 370//sys	Waitid(idType int, id int, info *Siginfo, options int, rusage *Rusage) (err error)
 371
 372func Mkfifo(path string, mode uint32) error {
 373	return Mknod(path, mode|S_IFIFO, 0)
 374}
 375
 376func Mkfifoat(dirfd int, path string, mode uint32) error {
 377	return Mknodat(dirfd, path, mode|S_IFIFO, 0)
 378}
 379
 380func (sa *SockaddrInet4) sockaddr() (unsafe.Pointer, _Socklen, error) {
 381	if sa.Port < 0 || sa.Port > 0xFFFF {
 382		return nil, 0, EINVAL
 383	}
 384	sa.raw.Family = AF_INET
 385	p := (*[2]byte)(unsafe.Pointer(&sa.raw.Port))
 386	p[0] = byte(sa.Port >> 8)
 387	p[1] = byte(sa.Port)
 388	sa.raw.Addr = sa.Addr
 389	return unsafe.Pointer(&sa.raw), SizeofSockaddrInet4, nil
 390}
 391
 392func (sa *SockaddrInet6) sockaddr() (unsafe.Pointer, _Socklen, error) {
 393	if sa.Port < 0 || sa.Port > 0xFFFF {
 394		return nil, 0, EINVAL
 395	}
 396	sa.raw.Family = AF_INET6
 397	p := (*[2]byte)(unsafe.Pointer(&sa.raw.Port))
 398	p[0] = byte(sa.Port >> 8)
 399	p[1] = byte(sa.Port)
 400	sa.raw.Scope_id = sa.ZoneId
 401	sa.raw.Addr = sa.Addr
 402	return unsafe.Pointer(&sa.raw), SizeofSockaddrInet6, nil
 403}
 404
 405func (sa *SockaddrUnix) sockaddr() (unsafe.Pointer, _Socklen, error) {
 406	name := sa.Name
 407	n := len(name)
 408	if n >= len(sa.raw.Path) {
 409		return nil, 0, EINVAL
 410	}
 411	sa.raw.Family = AF_UNIX
 412	for i := 0; i < n; i++ {
 413		sa.raw.Path[i] = int8(name[i])
 414	}
 415	// length is family (uint16), name, NUL.
 416	sl := _Socklen(2)
 417	if n > 0 {
 418		sl += _Socklen(n) + 1
 419	}
 420	if sa.raw.Path[0] == '@' {
 421		sa.raw.Path[0] = 0
 422		// Don't count trailing NUL for abstract address.
 423		sl--
 424	}
 425
 426	return unsafe.Pointer(&sa.raw), sl, nil
 427}
 428
 429// SockaddrLinklayer implements the Sockaddr interface for AF_PACKET type sockets.
 430type SockaddrLinklayer struct {
 431	Protocol uint16
 432	Ifindex  int
 433	Hatype   uint16
 434	Pkttype  uint8
 435	Halen    uint8
 436	Addr     [8]byte
 437	raw      RawSockaddrLinklayer
 438}
 439
 440func (sa *SockaddrLinklayer) sockaddr() (unsafe.Pointer, _Socklen, error) {
 441	if sa.Ifindex < 0 || sa.Ifindex > 0x7fffffff {
 442		return nil, 0, EINVAL
 443	}
 444	sa.raw.Family = AF_PACKET
 445	sa.raw.Protocol = sa.Protocol
 446	sa.raw.Ifindex = int32(sa.Ifindex)
 447	sa.raw.Hatype = sa.Hatype
 448	sa.raw.Pkttype = sa.Pkttype
 449	sa.raw.Halen = sa.Halen
 450	sa.raw.Addr = sa.Addr
 451	return unsafe.Pointer(&sa.raw), SizeofSockaddrLinklayer, nil
 452}
 453
 454// SockaddrNetlink implements the Sockaddr interface for AF_NETLINK type sockets.
 455type SockaddrNetlink struct {
 456	Family uint16
 457	Pad    uint16
 458	Pid    uint32
 459	Groups uint32
 460	raw    RawSockaddrNetlink
 461}
 462
 463func (sa *SockaddrNetlink) sockaddr() (unsafe.Pointer, _Socklen, error) {
 464	sa.raw.Family = AF_NETLINK
 465	sa.raw.Pad = sa.Pad
 466	sa.raw.Pid = sa.Pid
 467	sa.raw.Groups = sa.Groups
 468	return unsafe.Pointer(&sa.raw), SizeofSockaddrNetlink, nil
 469}
 470
 471// SockaddrHCI implements the Sockaddr interface for AF_BLUETOOTH type sockets
 472// using the HCI protocol.
 473type SockaddrHCI struct {
 474	Dev     uint16
 475	Channel uint16
 476	raw     RawSockaddrHCI
 477}
 478
 479func (sa *SockaddrHCI) sockaddr() (unsafe.Pointer, _Socklen, error) {
 480	sa.raw.Family = AF_BLUETOOTH
 481	sa.raw.Dev = sa.Dev
 482	sa.raw.Channel = sa.Channel
 483	return unsafe.Pointer(&sa.raw), SizeofSockaddrHCI, nil
 484}
 485
 486// SockaddrL2 implements the Sockaddr interface for AF_BLUETOOTH type sockets
 487// using the L2CAP protocol.
 488type SockaddrL2 struct {
 489	PSM      uint16
 490	CID      uint16
 491	Addr     [6]uint8
 492	AddrType uint8
 493	raw      RawSockaddrL2
 494}
 495
 496func (sa *SockaddrL2) sockaddr() (unsafe.Pointer, _Socklen, error) {
 497	sa.raw.Family = AF_BLUETOOTH
 498	psm := (*[2]byte)(unsafe.Pointer(&sa.raw.Psm))
 499	psm[0] = byte(sa.PSM)
 500	psm[1] = byte(sa.PSM >> 8)
 501	for i := 0; i < len(sa.Addr); i++ {
 502		sa.raw.Bdaddr[i] = sa.Addr[len(sa.Addr)-1-i]
 503	}
 504	cid := (*[2]byte)(unsafe.Pointer(&sa.raw.Cid))
 505	cid[0] = byte(sa.CID)
 506	cid[1] = byte(sa.CID >> 8)
 507	sa.raw.Bdaddr_type = sa.AddrType
 508	return unsafe.Pointer(&sa.raw), SizeofSockaddrL2, nil
 509}
 510
 511// SockaddrRFCOMM implements the Sockaddr interface for AF_BLUETOOTH type sockets
 512// using the RFCOMM protocol.
 513//
 514// Server example:
 515//
 516//	fd, _ := Socket(AF_BLUETOOTH, SOCK_STREAM, BTPROTO_RFCOMM)
 517//	_ = unix.Bind(fd, &unix.SockaddrRFCOMM{
 518//		Channel: 1,
 519//		Addr:    [6]uint8{0, 0, 0, 0, 0, 0}, // BDADDR_ANY or 00:00:00:00:00:00
 520//	})
 521//	_ = Listen(fd, 1)
 522//	nfd, sa, _ := Accept(fd)
 523//	fmt.Printf("conn addr=%v fd=%d", sa.(*unix.SockaddrRFCOMM).Addr, nfd)
 524//	Read(nfd, buf)
 525//
 526// Client example:
 527//
 528//	fd, _ := Socket(AF_BLUETOOTH, SOCK_STREAM, BTPROTO_RFCOMM)
 529//	_ = Connect(fd, &SockaddrRFCOMM{
 530//		Channel: 1,
 531//		Addr:    [6]byte{0x11, 0x22, 0x33, 0xaa, 0xbb, 0xcc}, // CC:BB:AA:33:22:11
 532//	})
 533//	Write(fd, []byte(`hello`))
 534type SockaddrRFCOMM struct {
 535	// Addr represents a bluetooth address, byte ordering is little-endian.
 536	Addr [6]uint8
 537
 538	// Channel is a designated bluetooth channel, only 1-30 are available for use.
 539	// Since Linux 2.6.7 and further zero value is the first available channel.
 540	Channel uint8
 541
 542	raw RawSockaddrRFCOMM
 543}
 544
 545func (sa *SockaddrRFCOMM) sockaddr() (unsafe.Pointer, _Socklen, error) {
 546	sa.raw.Family = AF_BLUETOOTH
 547	sa.raw.Channel = sa.Channel
 548	sa.raw.Bdaddr = sa.Addr
 549	return unsafe.Pointer(&sa.raw), SizeofSockaddrRFCOMM, nil
 550}
 551
 552// SockaddrCAN implements the Sockaddr interface for AF_CAN type sockets.
 553// The RxID and TxID fields are used for transport protocol addressing in
 554// (CAN_TP16, CAN_TP20, CAN_MCNET, and CAN_ISOTP), they can be left with
 555// zero values for CAN_RAW and CAN_BCM sockets as they have no meaning.
 556//
 557// The SockaddrCAN struct must be bound to the socket file descriptor
 558// using Bind before the CAN socket can be used.
 559//
 560//	// Read one raw CAN frame
 561//	fd, _ := Socket(AF_CAN, SOCK_RAW, CAN_RAW)
 562//	addr := &SockaddrCAN{Ifindex: index}
 563//	Bind(fd, addr)
 564//	frame := make([]byte, 16)
 565//	Read(fd, frame)
 566//
 567// The full SocketCAN documentation can be found in the linux kernel
 568// archives at: https://www.kernel.org/doc/Documentation/networking/can.txt
 569type SockaddrCAN struct {
 570	Ifindex int
 571	RxID    uint32
 572	TxID    uint32
 573	raw     RawSockaddrCAN
 574}
 575
 576func (sa *SockaddrCAN) sockaddr() (unsafe.Pointer, _Socklen, error) {
 577	if sa.Ifindex < 0 || sa.Ifindex > 0x7fffffff {
 578		return nil, 0, EINVAL
 579	}
 580	sa.raw.Family = AF_CAN
 581	sa.raw.Ifindex = int32(sa.Ifindex)
 582	rx := (*[4]byte)(unsafe.Pointer(&sa.RxID))
 583	for i := 0; i < 4; i++ {
 584		sa.raw.Addr[i] = rx[i]
 585	}
 586	tx := (*[4]byte)(unsafe.Pointer(&sa.TxID))
 587	for i := 0; i < 4; i++ {
 588		sa.raw.Addr[i+4] = tx[i]
 589	}
 590	return unsafe.Pointer(&sa.raw), SizeofSockaddrCAN, nil
 591}
 592
 593// SockaddrCANJ1939 implements the Sockaddr interface for AF_CAN using J1939
 594// protocol (https://en.wikipedia.org/wiki/SAE_J1939). For more information
 595// on the purposes of the fields, check the official linux kernel documentation
 596// available here: https://www.kernel.org/doc/Documentation/networking/j1939.rst
 597type SockaddrCANJ1939 struct {
 598	Ifindex int
 599	Name    uint64
 600	PGN     uint32
 601	Addr    uint8
 602	raw     RawSockaddrCAN
 603}
 604
 605func (sa *SockaddrCANJ1939) sockaddr() (unsafe.Pointer, _Socklen, error) {
 606	if sa.Ifindex < 0 || sa.Ifindex > 0x7fffffff {
 607		return nil, 0, EINVAL
 608	}
 609	sa.raw.Family = AF_CAN
 610	sa.raw.Ifindex = int32(sa.Ifindex)
 611	n := (*[8]byte)(unsafe.Pointer(&sa.Name))
 612	for i := 0; i < 8; i++ {
 613		sa.raw.Addr[i] = n[i]
 614	}
 615	p := (*[4]byte)(unsafe.Pointer(&sa.PGN))
 616	for i := 0; i < 4; i++ {
 617		sa.raw.Addr[i+8] = p[i]
 618	}
 619	sa.raw.Addr[12] = sa.Addr
 620	return unsafe.Pointer(&sa.raw), SizeofSockaddrCAN, nil
 621}
 622
 623// SockaddrALG implements the Sockaddr interface for AF_ALG type sockets.
 624// SockaddrALG enables userspace access to the Linux kernel's cryptography
 625// subsystem. The Type and Name fields specify which type of hash or cipher
 626// should be used with a given socket.
 627//
 628// To create a file descriptor that provides access to a hash or cipher, both
 629// Bind and Accept must be used. Once the setup process is complete, input
 630// data can be written to the socket, processed by the kernel, and then read
 631// back as hash output or ciphertext.
 632//
 633// Here is an example of using an AF_ALG socket with SHA1 hashing.
 634// The initial socket setup process is as follows:
 635//
 636//	// Open a socket to perform SHA1 hashing.
 637//	fd, _ := unix.Socket(unix.AF_ALG, unix.SOCK_SEQPACKET, 0)
 638//	addr := &unix.SockaddrALG{Type: "hash", Name: "sha1"}
 639//	unix.Bind(fd, addr)
 640//	// Note: unix.Accept does not work at this time; must invoke accept()
 641//	// manually using unix.Syscall.
 642//	hashfd, _, _ := unix.Syscall(unix.SYS_ACCEPT, uintptr(fd), 0, 0)
 643//
 644// Once a file descriptor has been returned from Accept, it may be used to
 645// perform SHA1 hashing. The descriptor is not safe for concurrent use, but
 646// may be re-used repeatedly with subsequent Write and Read operations.
 647//
 648// When hashing a small byte slice or string, a single Write and Read may
 649// be used:
 650//
 651//	// Assume hashfd is already configured using the setup process.
 652//	hash := os.NewFile(hashfd, "sha1")
 653//	// Hash an input string and read the results. Each Write discards
 654//	// previous hash state. Read always reads the current state.
 655//	b := make([]byte, 20)
 656//	for i := 0; i < 2; i++ {
 657//	    io.WriteString(hash, "Hello, world.")
 658//	    hash.Read(b)
 659//	    fmt.Println(hex.EncodeToString(b))
 660//	}
 661//	// Output:
 662//	// 2ae01472317d1935a84797ec1983ae243fc6aa28
 663//	// 2ae01472317d1935a84797ec1983ae243fc6aa28
 664//
 665// For hashing larger byte slices, or byte streams such as those read from
 666// a file or socket, use Sendto with MSG_MORE to instruct the kernel to update
 667// the hash digest instead of creating a new one for a given chunk and finalizing it.
 668//
 669//	// Assume hashfd and addr are already configured using the setup process.
 670//	hash := os.NewFile(hashfd, "sha1")
 671//	// Hash the contents of a file.
 672//	f, _ := os.Open("/tmp/linux-4.10-rc7.tar.xz")
 673//	b := make([]byte, 4096)
 674//	for {
 675//	    n, err := f.Read(b)
 676//	    if err == io.EOF {
 677//	        break
 678//	    }
 679//	    unix.Sendto(hashfd, b[:n], unix.MSG_MORE, addr)
 680//	}
 681//	hash.Read(b)
 682//	fmt.Println(hex.EncodeToString(b))
 683//	// Output: 85cdcad0c06eef66f805ecce353bec9accbeecc5
 684//
 685// For more information, see: http://www.chronox.de/crypto-API/crypto/userspace-if.html.
 686type SockaddrALG struct {
 687	Type    string
 688	Name    string
 689	Feature uint32
 690	Mask    uint32
 691	raw     RawSockaddrALG
 692}
 693
 694func (sa *SockaddrALG) sockaddr() (unsafe.Pointer, _Socklen, error) {
 695	// Leave room for NUL byte terminator.
 696	if len(sa.Type) > 13 {
 697		return nil, 0, EINVAL
 698	}
 699	if len(sa.Name) > 63 {
 700		return nil, 0, EINVAL
 701	}
 702
 703	sa.raw.Family = AF_ALG
 704	sa.raw.Feat = sa.Feature
 705	sa.raw.Mask = sa.Mask
 706
 707	typ, err := ByteSliceFromString(sa.Type)
 708	if err != nil {
 709		return nil, 0, err
 710	}
 711	name, err := ByteSliceFromString(sa.Name)
 712	if err != nil {
 713		return nil, 0, err
 714	}
 715
 716	copy(sa.raw.Type[:], typ)
 717	copy(sa.raw.Name[:], name)
 718
 719	return unsafe.Pointer(&sa.raw), SizeofSockaddrALG, nil
 720}
 721
 722// SockaddrVM implements the Sockaddr interface for AF_VSOCK type sockets.
 723// SockaddrVM provides access to Linux VM sockets: a mechanism that enables
 724// bidirectional communication between a hypervisor and its guest virtual
 725// machines.
 726type SockaddrVM struct {
 727	// CID and Port specify a context ID and port address for a VM socket.
 728	// Guests have a unique CID, and hosts may have a well-known CID of:
 729	//  - VMADDR_CID_HYPERVISOR: refers to the hypervisor process.
 730	//  - VMADDR_CID_LOCAL: refers to local communication (loopback).
 731	//  - VMADDR_CID_HOST: refers to other processes on the host.
 732	CID   uint32
 733	Port  uint32
 734	Flags uint8
 735	raw   RawSockaddrVM
 736}
 737
 738func (sa *SockaddrVM) sockaddr() (unsafe.Pointer, _Socklen, error) {
 739	sa.raw.Family = AF_VSOCK
 740	sa.raw.Port = sa.Port
 741	sa.raw.Cid = sa.CID
 742	sa.raw.Flags = sa.Flags
 743
 744	return unsafe.Pointer(&sa.raw), SizeofSockaddrVM, nil
 745}
 746
 747type SockaddrXDP struct {
 748	Flags        uint16
 749	Ifindex      uint32
 750	QueueID      uint32
 751	SharedUmemFD uint32
 752	raw          RawSockaddrXDP
 753}
 754
 755func (sa *SockaddrXDP) sockaddr() (unsafe.Pointer, _Socklen, error) {
 756	sa.raw.Family = AF_XDP
 757	sa.raw.Flags = sa.Flags
 758	sa.raw.Ifindex = sa.Ifindex
 759	sa.raw.Queue_id = sa.QueueID
 760	sa.raw.Shared_umem_fd = sa.SharedUmemFD
 761
 762	return unsafe.Pointer(&sa.raw), SizeofSockaddrXDP, nil
 763}
 764
 765// This constant mirrors the #define of PX_PROTO_OE in
 766// linux/if_pppox.h. We're defining this by hand here instead of
 767// autogenerating through mkerrors.sh because including
 768// linux/if_pppox.h causes some declaration conflicts with other
 769// includes (linux/if_pppox.h includes linux/in.h, which conflicts
 770// with netinet/in.h). Given that we only need a single zero constant
 771// out of that file, it's cleaner to just define it by hand here.
 772const px_proto_oe = 0
 773
 774type SockaddrPPPoE struct {
 775	SID    uint16
 776	Remote []byte
 777	Dev    string
 778	raw    RawSockaddrPPPoX
 779}
 780
 781func (sa *SockaddrPPPoE) sockaddr() (unsafe.Pointer, _Socklen, error) {
 782	if len(sa.Remote) != 6 {
 783		return nil, 0, EINVAL
 784	}
 785	if len(sa.Dev) > IFNAMSIZ-1 {
 786		return nil, 0, EINVAL
 787	}
 788
 789	*(*uint16)(unsafe.Pointer(&sa.raw[0])) = AF_PPPOX
 790	// This next field is in host-endian byte order. We can't use the
 791	// same unsafe pointer cast as above, because this value is not
 792	// 32-bit aligned and some architectures don't allow unaligned
 793	// access.
 794	//
 795	// However, the value of px_proto_oe is 0, so we can use
 796	// encoding/binary helpers to write the bytes without worrying
 797	// about the ordering.
 798	binary.BigEndian.PutUint32(sa.raw[2:6], px_proto_oe)
 799	// This field is deliberately big-endian, unlike the previous
 800	// one. The kernel expects SID to be in network byte order.
 801	binary.BigEndian.PutUint16(sa.raw[6:8], sa.SID)
 802	copy(sa.raw[8:14], sa.Remote)
 803	for i := 14; i < 14+IFNAMSIZ; i++ {
 804		sa.raw[i] = 0
 805	}
 806	copy(sa.raw[14:], sa.Dev)
 807	return unsafe.Pointer(&sa.raw), SizeofSockaddrPPPoX, nil
 808}
 809
 810// SockaddrTIPC implements the Sockaddr interface for AF_TIPC type sockets.
 811// For more information on TIPC, see: http://tipc.sourceforge.net/.
 812type SockaddrTIPC struct {
 813	// Scope is the publication scopes when binding service/service range.
 814	// Should be set to TIPC_CLUSTER_SCOPE or TIPC_NODE_SCOPE.
 815	Scope int
 816
 817	// Addr is the type of address used to manipulate a socket. Addr must be
 818	// one of:
 819	//  - *TIPCSocketAddr: "id" variant in the C addr union
 820	//  - *TIPCServiceRange: "nameseq" variant in the C addr union
 821	//  - *TIPCServiceName: "name" variant in the C addr union
 822	//
 823	// If nil, EINVAL will be returned when the structure is used.
 824	Addr TIPCAddr
 825
 826	raw RawSockaddrTIPC
 827}
 828
 829// TIPCAddr is implemented by types that can be used as an address for
 830// SockaddrTIPC. It is only implemented by *TIPCSocketAddr, *TIPCServiceRange,
 831// and *TIPCServiceName.
 832type TIPCAddr interface {
 833	tipcAddrtype() uint8
 834	tipcAddr() [12]byte
 835}
 836
 837func (sa *TIPCSocketAddr) tipcAddr() [12]byte {
 838	var out [12]byte
 839	copy(out[:], (*(*[unsafe.Sizeof(TIPCSocketAddr{})]byte)(unsafe.Pointer(sa)))[:])
 840	return out
 841}
 842
 843func (sa *TIPCSocketAddr) tipcAddrtype() uint8 { return TIPC_SOCKET_ADDR }
 844
 845func (sa *TIPCServiceRange) tipcAddr() [12]byte {
 846	var out [12]byte
 847	copy(out[:], (*(*[unsafe.Sizeof(TIPCServiceRange{})]byte)(unsafe.Pointer(sa)))[:])
 848	return out
 849}
 850
 851func (sa *TIPCServiceRange) tipcAddrtype() uint8 { return TIPC_SERVICE_RANGE }
 852
 853func (sa *TIPCServiceName) tipcAddr() [12]byte {
 854	var out [12]byte
 855	copy(out[:], (*(*[unsafe.Sizeof(TIPCServiceName{})]byte)(unsafe.Pointer(sa)))[:])
 856	return out
 857}
 858
 859func (sa *TIPCServiceName) tipcAddrtype() uint8 { return TIPC_SERVICE_ADDR }
 860
 861func (sa *SockaddrTIPC) sockaddr() (unsafe.Pointer, _Socklen, error) {
 862	if sa.Addr == nil {
 863		return nil, 0, EINVAL
 864	}
 865	sa.raw.Family = AF_TIPC
 866	sa.raw.Scope = int8(sa.Scope)
 867	sa.raw.Addrtype = sa.Addr.tipcAddrtype()
 868	sa.raw.Addr = sa.Addr.tipcAddr()
 869	return unsafe.Pointer(&sa.raw), SizeofSockaddrTIPC, nil
 870}
 871
 872// SockaddrL2TPIP implements the Sockaddr interface for IPPROTO_L2TP/AF_INET sockets.
 873type SockaddrL2TPIP struct {
 874	Addr   [4]byte
 875	ConnId uint32
 876	raw    RawSockaddrL2TPIP
 877}
 878
 879func (sa *SockaddrL2TPIP) sockaddr() (unsafe.Pointer, _Socklen, error) {
 880	sa.raw.Family = AF_INET
 881	sa.raw.Conn_id = sa.ConnId
 882	sa.raw.Addr = sa.Addr
 883	return unsafe.Pointer(&sa.raw), SizeofSockaddrL2TPIP, nil
 884}
 885
 886// SockaddrL2TPIP6 implements the Sockaddr interface for IPPROTO_L2TP/AF_INET6 sockets.
 887type SockaddrL2TPIP6 struct {
 888	Addr   [16]byte
 889	ZoneId uint32
 890	ConnId uint32
 891	raw    RawSockaddrL2TPIP6
 892}
 893
 894func (sa *SockaddrL2TPIP6) sockaddr() (unsafe.Pointer, _Socklen, error) {
 895	sa.raw.Family = AF_INET6
 896	sa.raw.Conn_id = sa.ConnId
 897	sa.raw.Scope_id = sa.ZoneId
 898	sa.raw.Addr = sa.Addr
 899	return unsafe.Pointer(&sa.raw), SizeofSockaddrL2TPIP6, nil
 900}
 901
 902// SockaddrIUCV implements the Sockaddr interface for AF_IUCV sockets.
 903type SockaddrIUCV struct {
 904	UserID string
 905	Name   string
 906	raw    RawSockaddrIUCV
 907}
 908
 909func (sa *SockaddrIUCV) sockaddr() (unsafe.Pointer, _Socklen, error) {
 910	sa.raw.Family = AF_IUCV
 911	// These are EBCDIC encoded by the kernel, but we still need to pad them
 912	// with blanks. Initializing with blanks allows the caller to feed in either
 913	// a padded or an unpadded string.
 914	for i := 0; i < 8; i++ {
 915		sa.raw.Nodeid[i] = ' '
 916		sa.raw.User_id[i] = ' '
 917		sa.raw.Name[i] = ' '
 918	}
 919	if len(sa.UserID) > 8 || len(sa.Name) > 8 {
 920		return nil, 0, EINVAL
 921	}
 922	for i, b := range []byte(sa.UserID[:]) {
 923		sa.raw.User_id[i] = int8(b)
 924	}
 925	for i, b := range []byte(sa.Name[:]) {
 926		sa.raw.Name[i] = int8(b)
 927	}
 928	return unsafe.Pointer(&sa.raw), SizeofSockaddrIUCV, nil
 929}
 930
 931type SockaddrNFC struct {
 932	DeviceIdx   uint32
 933	TargetIdx   uint32
 934	NFCProtocol uint32
 935	raw         RawSockaddrNFC
 936}
 937
 938func (sa *SockaddrNFC) sockaddr() (unsafe.Pointer, _Socklen, error) {
 939	sa.raw.Sa_family = AF_NFC
 940	sa.raw.Dev_idx = sa.DeviceIdx
 941	sa.raw.Target_idx = sa.TargetIdx
 942	sa.raw.Nfc_protocol = sa.NFCProtocol
 943	return unsafe.Pointer(&sa.raw), SizeofSockaddrNFC, nil
 944}
 945
 946type SockaddrNFCLLCP struct {
 947	DeviceIdx      uint32
 948	TargetIdx      uint32
 949	NFCProtocol    uint32
 950	DestinationSAP uint8
 951	SourceSAP      uint8
 952	ServiceName    string
 953	raw            RawSockaddrNFCLLCP
 954}
 955
 956func (sa *SockaddrNFCLLCP) sockaddr() (unsafe.Pointer, _Socklen, error) {
 957	sa.raw.Sa_family = AF_NFC
 958	sa.raw.Dev_idx = sa.DeviceIdx
 959	sa.raw.Target_idx = sa.TargetIdx
 960	sa.raw.Nfc_protocol = sa.NFCProtocol
 961	sa.raw.Dsap = sa.DestinationSAP
 962	sa.raw.Ssap = sa.SourceSAP
 963	if len(sa.ServiceName) > len(sa.raw.Service_name) {
 964		return nil, 0, EINVAL
 965	}
 966	copy(sa.raw.Service_name[:], sa.ServiceName)
 967	sa.raw.SetServiceNameLen(len(sa.ServiceName))
 968	return unsafe.Pointer(&sa.raw), SizeofSockaddrNFCLLCP, nil
 969}
 970
 971var socketProtocol = func(fd int) (int, error) {
 972	return GetsockoptInt(fd, SOL_SOCKET, SO_PROTOCOL)
 973}
 974
 975func anyToSockaddr(fd int, rsa *RawSockaddrAny) (Sockaddr, error) {
 976	switch rsa.Addr.Family {
 977	case AF_NETLINK:
 978		pp := (*RawSockaddrNetlink)(unsafe.Pointer(rsa))
 979		sa := new(SockaddrNetlink)
 980		sa.Family = pp.Family
 981		sa.Pad = pp.Pad
 982		sa.Pid = pp.Pid
 983		sa.Groups = pp.Groups
 984		return sa, nil
 985
 986	case AF_PACKET:
 987		pp := (*RawSockaddrLinklayer)(unsafe.Pointer(rsa))
 988		sa := new(SockaddrLinklayer)
 989		sa.Protocol = pp.Protocol
 990		sa.Ifindex = int(pp.Ifindex)
 991		sa.Hatype = pp.Hatype
 992		sa.Pkttype = pp.Pkttype
 993		sa.Halen = pp.Halen
 994		sa.Addr = pp.Addr
 995		return sa, nil
 996
 997	case AF_UNIX:
 998		pp := (*RawSockaddrUnix)(unsafe.Pointer(rsa))
 999		sa := new(SockaddrUnix)
1000		if pp.Path[0] == 0 {
1001			// "Abstract" Unix domain socket.
1002			// Rewrite leading NUL as @ for textual display.
1003			// (This is the standard convention.)
1004			// Not friendly to overwrite in place,
1005			// but the callers below don't care.
1006			pp.Path[0] = '@'
1007		}
1008
1009		// Assume path ends at NUL.
1010		// This is not technically the Linux semantics for
1011		// abstract Unix domain sockets--they are supposed
1012		// to be uninterpreted fixed-size binary blobs--but
1013		// everyone uses this convention.
1014		n := 0
1015		for n < len(pp.Path) && pp.Path[n] != 0 {
1016			n++
1017		}
1018		sa.Name = string(unsafe.Slice((*byte)(unsafe.Pointer(&pp.Path[0])), n))
1019		return sa, nil
1020
1021	case AF_INET:
1022		proto, err := socketProtocol(fd)
1023		if err != nil {
1024			return nil, err
1025		}
1026
1027		switch proto {
1028		case IPPROTO_L2TP:
1029			pp := (*RawSockaddrL2TPIP)(unsafe.Pointer(rsa))
1030			sa := new(SockaddrL2TPIP)
1031			sa.ConnId = pp.Conn_id
1032			sa.Addr = pp.Addr
1033			return sa, nil
1034		default:
1035			pp := (*RawSockaddrInet4)(unsafe.Pointer(rsa))
1036			sa := new(SockaddrInet4)
1037			p := (*[2]byte)(unsafe.Pointer(&pp.Port))
1038			sa.Port = int(p[0])<<8 + int(p[1])
1039			sa.Addr = pp.Addr
1040			return sa, nil
1041		}
1042
1043	case AF_INET6:
1044		proto, err := socketProtocol(fd)
1045		if err != nil {
1046			return nil, err
1047		}
1048
1049		switch proto {
1050		case IPPROTO_L2TP:
1051			pp := (*RawSockaddrL2TPIP6)(unsafe.Pointer(rsa))
1052			sa := new(SockaddrL2TPIP6)
1053			sa.ConnId = pp.Conn_id
1054			sa.ZoneId = pp.Scope_id
1055			sa.Addr = pp.Addr
1056			return sa, nil
1057		default:
1058			pp := (*RawSockaddrInet6)(unsafe.Pointer(rsa))
1059			sa := new(SockaddrInet6)
1060			p := (*[2]byte)(unsafe.Pointer(&pp.Port))
1061			sa.Port = int(p[0])<<8 + int(p[1])
1062			sa.ZoneId = pp.Scope_id
1063			sa.Addr = pp.Addr
1064			return sa, nil
1065		}
1066
1067	case AF_VSOCK:
1068		pp := (*RawSockaddrVM)(unsafe.Pointer(rsa))
1069		sa := &SockaddrVM{
1070			CID:   pp.Cid,
1071			Port:  pp.Port,
1072			Flags: pp.Flags,
1073		}
1074		return sa, nil
1075	case AF_BLUETOOTH:
1076		proto, err := socketProtocol(fd)
1077		if err != nil {
1078			return nil, err
1079		}
1080		// only BTPROTO_L2CAP and BTPROTO_RFCOMM can accept connections
1081		switch proto {
1082		case BTPROTO_L2CAP:
1083			pp := (*RawSockaddrL2)(unsafe.Pointer(rsa))
1084			sa := &SockaddrL2{
1085				PSM:      pp.Psm,
1086				CID:      pp.Cid,
1087				Addr:     pp.Bdaddr,
1088				AddrType: pp.Bdaddr_type,
1089			}
1090			return sa, nil
1091		case BTPROTO_RFCOMM:
1092			pp := (*RawSockaddrRFCOMM)(unsafe.Pointer(rsa))
1093			sa := &SockaddrRFCOMM{
1094				Channel: pp.Channel,
1095				Addr:    pp.Bdaddr,
1096			}
1097			return sa, nil
1098		}
1099	case AF_XDP:
1100		pp := (*RawSockaddrXDP)(unsafe.Pointer(rsa))
1101		sa := &SockaddrXDP{
1102			Flags:        pp.Flags,
1103			Ifindex:      pp.Ifindex,
1104			QueueID:      pp.Queue_id,
1105			SharedUmemFD: pp.Shared_umem_fd,
1106		}
1107		return sa, nil
1108	case AF_PPPOX:
1109		pp := (*RawSockaddrPPPoX)(unsafe.Pointer(rsa))
1110		if binary.BigEndian.Uint32(pp[2:6]) != px_proto_oe {
1111			return nil, EINVAL
1112		}
1113		sa := &SockaddrPPPoE{
1114			SID:    binary.BigEndian.Uint16(pp[6:8]),
1115			Remote: pp[8:14],
1116		}
1117		for i := 14; i < 14+IFNAMSIZ; i++ {
1118			if pp[i] == 0 {
1119				sa.Dev = string(pp[14:i])
1120				break
1121			}
1122		}
1123		return sa, nil
1124	case AF_TIPC:
1125		pp := (*RawSockaddrTIPC)(unsafe.Pointer(rsa))
1126
1127		sa := &SockaddrTIPC{
1128			Scope: int(pp.Scope),
1129		}
1130
1131		// Determine which union variant is present in pp.Addr by checking
1132		// pp.Addrtype.
1133		switch pp.Addrtype {
1134		case TIPC_SERVICE_RANGE:
1135			sa.Addr = (*TIPCServiceRange)(unsafe.Pointer(&pp.Addr))
1136		case TIPC_SERVICE_ADDR:
1137			sa.Addr = (*TIPCServiceName)(unsafe.Pointer(&pp.Addr))
1138		case TIPC_SOCKET_ADDR:
1139			sa.Addr = (*TIPCSocketAddr)(unsafe.Pointer(&pp.Addr))
1140		default:
1141			return nil, EINVAL
1142		}
1143
1144		return sa, nil
1145	case AF_IUCV:
1146		pp := (*RawSockaddrIUCV)(unsafe.Pointer(rsa))
1147
1148		var user [8]byte
1149		var name [8]byte
1150
1151		for i := 0; i < 8; i++ {
1152			user[i] = byte(pp.User_id[i])
1153			name[i] = byte(pp.Name[i])
1154		}
1155
1156		sa := &SockaddrIUCV{
1157			UserID: string(user[:]),
1158			Name:   string(name[:]),
1159		}
1160		return sa, nil
1161
1162	case AF_CAN:
1163		proto, err := socketProtocol(fd)
1164		if err != nil {
1165			return nil, err
1166		}
1167
1168		pp := (*RawSockaddrCAN)(unsafe.Pointer(rsa))
1169
1170		switch proto {
1171		case CAN_J1939:
1172			sa := &SockaddrCANJ1939{
1173				Ifindex: int(pp.Ifindex),
1174			}
1175			name := (*[8]byte)(unsafe.Pointer(&sa.Name))
1176			for i := 0; i < 8; i++ {
1177				name[i] = pp.Addr[i]
1178			}
1179			pgn := (*[4]byte)(unsafe.Pointer(&sa.PGN))
1180			for i := 0; i < 4; i++ {
1181				pgn[i] = pp.Addr[i+8]
1182			}
1183			addr := (*[1]byte)(unsafe.Pointer(&sa.Addr))
1184			addr[0] = pp.Addr[12]
1185			return sa, nil
1186		default:
1187			sa := &SockaddrCAN{
1188				Ifindex: int(pp.Ifindex),
1189			}
1190			rx := (*[4]byte)(unsafe.Pointer(&sa.RxID))
1191			for i := 0; i < 4; i++ {
1192				rx[i] = pp.Addr[i]
1193			}
1194			tx := (*[4]byte)(unsafe.Pointer(&sa.TxID))
1195			for i := 0; i < 4; i++ {
1196				tx[i] = pp.Addr[i+4]
1197			}
1198			return sa, nil
1199		}
1200	case AF_NFC:
1201		proto, err := socketProtocol(fd)
1202		if err != nil {
1203			return nil, err
1204		}
1205		switch proto {
1206		case NFC_SOCKPROTO_RAW:
1207			pp := (*RawSockaddrNFC)(unsafe.Pointer(rsa))
1208			sa := &SockaddrNFC{
1209				DeviceIdx:   pp.Dev_idx,
1210				TargetIdx:   pp.Target_idx,
1211				NFCProtocol: pp.Nfc_protocol,
1212			}
1213			return sa, nil
1214		case NFC_SOCKPROTO_LLCP:
1215			pp := (*RawSockaddrNFCLLCP)(unsafe.Pointer(rsa))
1216			if uint64(pp.Service_name_len) > uint64(len(pp.Service_name)) {
1217				return nil, EINVAL
1218			}
1219			sa := &SockaddrNFCLLCP{
1220				DeviceIdx:      pp.Dev_idx,
1221				TargetIdx:      pp.Target_idx,
1222				NFCProtocol:    pp.Nfc_protocol,
1223				DestinationSAP: pp.Dsap,
1224				SourceSAP:      pp.Ssap,
1225				ServiceName:    string(pp.Service_name[:pp.Service_name_len]),
1226			}
1227			return sa, nil
1228		default:
1229			return nil, EINVAL
1230		}
1231	}
1232	return nil, EAFNOSUPPORT
1233}
1234
1235func Accept(fd int) (nfd int, sa Sockaddr, err error) {
1236	var rsa RawSockaddrAny
1237	var len _Socklen = SizeofSockaddrAny
1238	nfd, err = accept4(fd, &rsa, &len, 0)
1239	if err != nil {
1240		return
1241	}
1242	sa, err = anyToSockaddr(fd, &rsa)
1243	if err != nil {
1244		Close(nfd)
1245		nfd = 0
1246	}
1247	return
1248}
1249
1250func Accept4(fd int, flags int) (nfd int, sa Sockaddr, err error) {
1251	var rsa RawSockaddrAny
1252	var len _Socklen = SizeofSockaddrAny
1253	nfd, err = accept4(fd, &rsa, &len, flags)
1254	if err != nil {
1255		return
1256	}
1257	if len > SizeofSockaddrAny {
1258		panic("RawSockaddrAny too small")
1259	}
1260	sa, err = anyToSockaddr(fd, &rsa)
1261	if err != nil {
1262		Close(nfd)
1263		nfd = 0
1264	}
1265	return
1266}
1267
1268func Getsockname(fd int) (sa Sockaddr, err error) {
1269	var rsa RawSockaddrAny
1270	var len _Socklen = SizeofSockaddrAny
1271	if err = getsockname(fd, &rsa, &len); err != nil {
1272		return
1273	}
1274	return anyToSockaddr(fd, &rsa)
1275}
1276
1277func GetsockoptIPMreqn(fd, level, opt int) (*IPMreqn, error) {
1278	var value IPMreqn
1279	vallen := _Socklen(SizeofIPMreqn)
1280	err := getsockopt(fd, level, opt, unsafe.Pointer(&value), &vallen)
1281	return &value, err
1282}
1283
1284func GetsockoptUcred(fd, level, opt int) (*Ucred, error) {
1285	var value Ucred
1286	vallen := _Socklen(SizeofUcred)
1287	err := getsockopt(fd, level, opt, unsafe.Pointer(&value), &vallen)
1288	return &value, err
1289}
1290
1291func GetsockoptTCPInfo(fd, level, opt int) (*TCPInfo, error) {
1292	var value TCPInfo
1293	vallen := _Socklen(SizeofTCPInfo)
1294	err := getsockopt(fd, level, opt, unsafe.Pointer(&value), &vallen)
1295	return &value, err
1296}
1297
1298// GetsockoptString returns the string value of the socket option opt for the
1299// socket associated with fd at the given socket level.
1300func GetsockoptString(fd, level, opt int) (string, error) {
1301	buf := make([]byte, 256)
1302	vallen := _Socklen(len(buf))
1303	err := getsockopt(fd, level, opt, unsafe.Pointer(&buf[0]), &vallen)
1304	if err != nil {
1305		if err == ERANGE {
1306			buf = make([]byte, vallen)
1307			err = getsockopt(fd, level, opt, unsafe.Pointer(&buf[0]), &vallen)
1308		}
1309		if err != nil {
1310			return "", err
1311		}
1312	}
1313	return string(buf[:vallen-1]), nil
1314}
1315
1316func GetsockoptTpacketStats(fd, level, opt int) (*TpacketStats, error) {
1317	var value TpacketStats
1318	vallen := _Socklen(SizeofTpacketStats)
1319	err := getsockopt(fd, level, opt, unsafe.Pointer(&value), &vallen)
1320	return &value, err
1321}
1322
1323func GetsockoptTpacketStatsV3(fd, level, opt int) (*TpacketStatsV3, error) {
1324	var value TpacketStatsV3
1325	vallen := _Socklen(SizeofTpacketStatsV3)
1326	err := getsockopt(fd, level, opt, unsafe.Pointer(&value), &vallen)
1327	return &value, err
1328}
1329
1330func SetsockoptIPMreqn(fd, level, opt int, mreq *IPMreqn) (err error) {
1331	return setsockopt(fd, level, opt, unsafe.Pointer(mreq), unsafe.Sizeof(*mreq))
1332}
1333
1334func SetsockoptPacketMreq(fd, level, opt int, mreq *PacketMreq) error {
1335	return setsockopt(fd, level, opt, unsafe.Pointer(mreq), unsafe.Sizeof(*mreq))
1336}
1337
1338// SetsockoptSockFprog attaches a classic BPF or an extended BPF program to a
1339// socket to filter incoming packets.  See 'man 7 socket' for usage information.
1340func SetsockoptSockFprog(fd, level, opt int, fprog *SockFprog) error {
1341	return setsockopt(fd, level, opt, unsafe.Pointer(fprog), unsafe.Sizeof(*fprog))
1342}
1343
1344func SetsockoptCanRawFilter(fd, level, opt int, filter []CanFilter) error {
1345	var p unsafe.Pointer
1346	if len(filter) > 0 {
1347		p = unsafe.Pointer(&filter[0])
1348	}
1349	return setsockopt(fd, level, opt, p, uintptr(len(filter)*SizeofCanFilter))
1350}
1351
1352func SetsockoptTpacketReq(fd, level, opt int, tp *TpacketReq) error {
1353	return setsockopt(fd, level, opt, unsafe.Pointer(tp), unsafe.Sizeof(*tp))
1354}
1355
1356func SetsockoptTpacketReq3(fd, level, opt int, tp *TpacketReq3) error {
1357	return setsockopt(fd, level, opt, unsafe.Pointer(tp), unsafe.Sizeof(*tp))
1358}
1359
1360func SetsockoptTCPRepairOpt(fd, level, opt int, o []TCPRepairOpt) (err error) {
1361	if len(o) == 0 {
1362		return EINVAL
1363	}
1364	return setsockopt(fd, level, opt, unsafe.Pointer(&o[0]), uintptr(SizeofTCPRepairOpt*len(o)))
1365}
1366
1367func SetsockoptTCPMD5Sig(fd, level, opt int, s *TCPMD5Sig) error {
1368	return setsockopt(fd, level, opt, unsafe.Pointer(s), unsafe.Sizeof(*s))
1369}
1370
1371// Keyctl Commands (http://man7.org/linux/man-pages/man2/keyctl.2.html)
1372
1373// KeyctlInt calls keyctl commands in which each argument is an int.
1374// These commands are KEYCTL_REVOKE, KEYCTL_CHOWN, KEYCTL_CLEAR, KEYCTL_LINK,
1375// KEYCTL_UNLINK, KEYCTL_NEGATE, KEYCTL_SET_REQKEY_KEYRING, KEYCTL_SET_TIMEOUT,
1376// KEYCTL_ASSUME_AUTHORITY, KEYCTL_SESSION_TO_PARENT, KEYCTL_REJECT,
1377// KEYCTL_INVALIDATE, and KEYCTL_GET_PERSISTENT.
1378//sys	KeyctlInt(cmd int, arg2 int, arg3 int, arg4 int, arg5 int) (ret int, err error) = SYS_KEYCTL
1379
1380// KeyctlBuffer calls keyctl commands in which the third and fourth
1381// arguments are a buffer and its length, respectively.
1382// These commands are KEYCTL_UPDATE, KEYCTL_READ, and KEYCTL_INSTANTIATE.
1383//sys	KeyctlBuffer(cmd int, arg2 int, buf []byte, arg5 int) (ret int, err error) = SYS_KEYCTL
1384
1385// KeyctlString calls keyctl commands which return a string.
1386// These commands are KEYCTL_DESCRIBE and KEYCTL_GET_SECURITY.
1387func KeyctlString(cmd int, id int) (string, error) {
1388	// We must loop as the string data may change in between the syscalls.
1389	// We could allocate a large buffer here to reduce the chance that the
1390	// syscall needs to be called twice; however, this is unnecessary as
1391	// the performance loss is negligible.
1392	var buffer []byte
1393	for {
1394		// Try to fill the buffer with data
1395		length, err := KeyctlBuffer(cmd, id, buffer, 0)
1396		if err != nil {
1397			return "", err
1398		}
1399
1400		// Check if the data was written
1401		if length <= len(buffer) {
1402			// Exclude the null terminator
1403			return string(buffer[:length-1]), nil
1404		}
1405
1406		// Make a bigger buffer if needed
1407		buffer = make([]byte, length)
1408	}
1409}
1410
1411// Keyctl commands with special signatures.
1412
1413// KeyctlGetKeyringID implements the KEYCTL_GET_KEYRING_ID command.
1414// See the full documentation at:
1415// http://man7.org/linux/man-pages/man3/keyctl_get_keyring_ID.3.html
1416func KeyctlGetKeyringID(id int, create bool) (ringid int, err error) {
1417	createInt := 0
1418	if create {
1419		createInt = 1
1420	}
1421	return KeyctlInt(KEYCTL_GET_KEYRING_ID, id, createInt, 0, 0)
1422}
1423
1424// KeyctlSetperm implements the KEYCTL_SETPERM command. The perm value is the
1425// key handle permission mask as described in the "keyctl setperm" section of
1426// http://man7.org/linux/man-pages/man1/keyctl.1.html.
1427// See the full documentation at:
1428// http://man7.org/linux/man-pages/man3/keyctl_setperm.3.html
1429func KeyctlSetperm(id int, perm uint32) error {
1430	_, err := KeyctlInt(KEYCTL_SETPERM, id, int(perm), 0, 0)
1431	return err
1432}
1433
1434//sys	keyctlJoin(cmd int, arg2 string) (ret int, err error) = SYS_KEYCTL
1435
1436// KeyctlJoinSessionKeyring implements the KEYCTL_JOIN_SESSION_KEYRING command.
1437// See the full documentation at:
1438// http://man7.org/linux/man-pages/man3/keyctl_join_session_keyring.3.html
1439func KeyctlJoinSessionKeyring(name string) (ringid int, err error) {
1440	return keyctlJoin(KEYCTL_JOIN_SESSION_KEYRING, name)
1441}
1442
1443//sys	keyctlSearch(cmd int, arg2 int, arg3 string, arg4 string, arg5 int) (ret int, err error) = SYS_KEYCTL
1444
1445// KeyctlSearch implements the KEYCTL_SEARCH command.
1446// See the full documentation at:
1447// http://man7.org/linux/man-pages/man3/keyctl_search.3.html
1448func KeyctlSearch(ringid int, keyType, description string, destRingid int) (id int, err error) {
1449	return keyctlSearch(KEYCTL_SEARCH, ringid, keyType, description, destRingid)
1450}
1451
1452//sys	keyctlIOV(cmd int, arg2 int, payload []Iovec, arg5 int) (err error) = SYS_KEYCTL
1453
1454// KeyctlInstantiateIOV implements the KEYCTL_INSTANTIATE_IOV command. This
1455// command is similar to KEYCTL_INSTANTIATE, except that the payload is a slice
1456// of Iovec (each of which represents a buffer) instead of a single buffer.
1457// See the full documentation at:
1458// http://man7.org/linux/man-pages/man3/keyctl_instantiate_iov.3.html
1459func KeyctlInstantiateIOV(id int, payload []Iovec, ringid int) error {
1460	return keyctlIOV(KEYCTL_INSTANTIATE_IOV, id, payload, ringid)
1461}
1462
1463//sys	keyctlDH(cmd int, arg2 *KeyctlDHParams, buf []byte) (ret int, err error) = SYS_KEYCTL
1464
1465// KeyctlDHCompute implements the KEYCTL_DH_COMPUTE command. This command
1466// computes a Diffie-Hellman shared secret based on the provide params. The
1467// secret is written to the provided buffer and the returned size is the number
1468// of bytes written (returning an error if there is insufficient space in the
1469// buffer). If a nil buffer is passed in, this function returns the minimum
1470// buffer length needed to store the appropriate data. Note that this differs
1471// from KEYCTL_READ's behavior which always returns the requested payload size.
1472// See the full documentation at:
1473// http://man7.org/linux/man-pages/man3/keyctl_dh_compute.3.html
1474func KeyctlDHCompute(params *KeyctlDHParams, buffer []byte) (size int, err error) {
1475	return keyctlDH(KEYCTL_DH_COMPUTE, params, buffer)
1476}
1477
1478// KeyctlRestrictKeyring implements the KEYCTL_RESTRICT_KEYRING command. This
1479// command limits the set of keys that can be linked to the keyring, regardless
1480// of keyring permissions. The command requires the "setattr" permission.
1481//
1482// When called with an empty keyType the command locks the keyring, preventing
1483// any further keys from being linked to the keyring.
1484//
1485// The "asymmetric" keyType defines restrictions requiring key payloads to be
1486// DER encoded X.509 certificates signed by keys in another keyring. Restrictions
1487// for "asymmetric" include "builtin_trusted", "builtin_and_secondary_trusted",
1488// "key_or_keyring:<key>", and "key_or_keyring:<key>:chain".
1489//
1490// As of Linux 4.12, only the "asymmetric" keyType defines type-specific
1491// restrictions.
1492//
1493// See the full documentation at:
1494// http://man7.org/linux/man-pages/man3/keyctl_restrict_keyring.3.html
1495// http://man7.org/linux/man-pages/man2/keyctl.2.html
1496func KeyctlRestrictKeyring(ringid int, keyType string, restriction string) error {
1497	if keyType == "" {
1498		return keyctlRestrictKeyring(KEYCTL_RESTRICT_KEYRING, ringid)
1499	}
1500	return keyctlRestrictKeyringByType(KEYCTL_RESTRICT_KEYRING, ringid, keyType, restriction)
1501}
1502
1503//sys	keyctlRestrictKeyringByType(cmd int, arg2 int, keyType string, restriction string) (err error) = SYS_KEYCTL
1504//sys	keyctlRestrictKeyring(cmd int, arg2 int) (err error) = SYS_KEYCTL
1505
1506func recvmsgRaw(fd int, iov []Iovec, oob []byte, flags int, rsa *RawSockaddrAny) (n, oobn int, recvflags int, err error) {
1507	var msg Msghdr
1508	msg.Name = (*byte)(unsafe.Pointer(rsa))
1509	msg.Namelen = uint32(SizeofSockaddrAny)
1510	var dummy byte
1511	if len(oob) > 0 {
1512		if emptyIovecs(iov) {
1513			var sockType int
1514			sockType, err = GetsockoptInt(fd, SOL_SOCKET, SO_TYPE)
1515			if err != nil {
1516				return
1517			}
1518			// receive at least one normal byte
1519			if sockType != SOCK_DGRAM {
1520				var iova [1]Iovec
1521				iova[0].Base = &dummy
1522				iova[0].SetLen(1)
1523				iov = iova[:]
1524			}
1525		}
1526		msg.Control = &oob[0]
1527		msg.SetControllen(len(oob))
1528	}
1529	if len(iov) > 0 {
1530		msg.Iov = &iov[0]
1531		msg.SetIovlen(len(iov))
1532	}
1533	if n, err = recvmsg(fd, &msg, flags); err != nil {
1534		return
1535	}
1536	oobn = int(msg.Controllen)
1537	recvflags = int(msg.Flags)
1538	return
1539}
1540
1541func sendmsgN(fd int, iov []Iovec, oob []byte, ptr unsafe.Pointer, salen _Socklen, flags int) (n int, err error) {
1542	var msg Msghdr
1543	msg.Name = (*byte)(ptr)
1544	msg.Namelen = uint32(salen)
1545	var dummy byte
1546	var empty bool
1547	if len(oob) > 0 {
1548		empty = emptyIovecs(iov)
1549		if empty {
1550			var sockType int
1551			sockType, err = GetsockoptInt(fd, SOL_SOCKET, SO_TYPE)
1552			if err != nil {
1553				return 0, err
1554			}
1555			// send at least one normal byte
1556			if sockType != SOCK_DGRAM {
1557				var iova [1]Iovec
1558				iova[0].Base = &dummy
1559				iova[0].SetLen(1)
1560				iov = iova[:]
1561			}
1562		}
1563		msg.Control = &oob[0]
1564		msg.SetControllen(len(oob))
1565	}
1566	if len(iov) > 0 {
1567		msg.Iov = &iov[0]
1568		msg.SetIovlen(len(iov))
1569	}
1570	if n, err = sendmsg(fd, &msg, flags); err != nil {
1571		return 0, err
1572	}
1573	if len(oob) > 0 && empty {
1574		n = 0
1575	}
1576	return n, nil
1577}
1578
1579// BindToDevice binds the socket associated with fd to device.
1580func BindToDevice(fd int, device string) (err error) {
1581	return SetsockoptString(fd, SOL_SOCKET, SO_BINDTODEVICE, device)
1582}
1583
1584//sys	ptrace(request int, pid int, addr uintptr, data uintptr) (err error)
1585//sys	ptracePtr(request int, pid int, addr uintptr, data unsafe.Pointer) (err error) = SYS_PTRACE
1586
1587func ptracePeek(req int, pid int, addr uintptr, out []byte) (count int, err error) {
1588	// The peek requests are machine-size oriented, so we wrap it
1589	// to retrieve arbitrary-length data.
1590
1591	// The ptrace syscall differs from glibc's ptrace.
1592	// Peeks returns the word in *data, not as the return value.
1593
1594	var buf [SizeofPtr]byte
1595
1596	// Leading edge. PEEKTEXT/PEEKDATA don't require aligned
1597	// access (PEEKUSER warns that it might), but if we don't
1598	// align our reads, we might straddle an unmapped page
1599	// boundary and not get the bytes leading up to the page
1600	// boundary.
1601	n := 0
1602	if addr%SizeofPtr != 0 {
1603		err = ptracePtr(req, pid, addr-addr%SizeofPtr, unsafe.Pointer(&buf[0]))
1604		if err != nil {
1605			return 0, err
1606		}
1607		n += copy(out, buf[addr%SizeofPtr:])
1608		out = out[n:]
1609	}
1610
1611	// Remainder.
1612	for len(out) > 0 {
1613		// We use an internal buffer to guarantee alignment.
1614		// It's not documented if this is necessary, but we're paranoid.
1615		err = ptracePtr(req, pid, addr+uintptr(n), unsafe.Pointer(&buf[0]))
1616		if err != nil {
1617			return n, err
1618		}
1619		copied := copy(out, buf[0:])
1620		n += copied
1621		out = out[copied:]
1622	}
1623
1624	return n, nil
1625}
1626
1627func PtracePeekText(pid int, addr uintptr, out []byte) (count int, err error) {
1628	return ptracePeek(PTRACE_PEEKTEXT, pid, addr, out)
1629}
1630
1631func PtracePeekData(pid int, addr uintptr, out []byte) (count int, err error) {
1632	return ptracePeek(PTRACE_PEEKDATA, pid, addr, out)
1633}
1634
1635func PtracePeekUser(pid int, addr uintptr, out []byte) (count int, err error) {
1636	return ptracePeek(PTRACE_PEEKUSR, pid, addr, out)
1637}
1638
1639func ptracePoke(pokeReq int, peekReq int, pid int, addr uintptr, data []byte) (count int, err error) {
1640	// As for ptracePeek, we need to align our accesses to deal
1641	// with the possibility of straddling an invalid page.
1642
1643	// Leading edge.
1644	n := 0
1645	if addr%SizeofPtr != 0 {
1646		var buf [SizeofPtr]byte
1647		err = ptracePtr(peekReq, pid, addr-addr%SizeofPtr, unsafe.Pointer(&buf[0]))
1648		if err != nil {
1649			return 0, err
1650		}
1651		n += copy(buf[addr%SizeofPtr:], data)
1652		word := *((*uintptr)(unsafe.Pointer(&buf[0])))
1653		err = ptrace(pokeReq, pid, addr-addr%SizeofPtr, word)
1654		if err != nil {
1655			return 0, err
1656		}
1657		data = data[n:]
1658	}
1659
1660	// Interior.
1661	for len(data) > SizeofPtr {
1662		word := *((*uintptr)(unsafe.Pointer(&data[0])))
1663		err = ptrace(pokeReq, pid, addr+uintptr(n), word)
1664		if err != nil {
1665			return n, err
1666		}
1667		n += SizeofPtr
1668		data = data[SizeofPtr:]
1669	}
1670
1671	// Trailing edge.
1672	if len(data) > 0 {
1673		var buf [SizeofPtr]byte
1674		err = ptracePtr(peekReq, pid, addr+uintptr(n), unsafe.Pointer(&buf[0]))
1675		if err != nil {
1676			return n, err
1677		}
1678		copy(buf[0:], data)
1679		word := *((*uintptr)(unsafe.Pointer(&buf[0])))
1680		err = ptrace(pokeReq, pid, addr+uintptr(n), word)
1681		if err != nil {
1682			return n, err
1683		}
1684		n += len(data)
1685	}
1686
1687	return n, nil
1688}
1689
1690func PtracePokeText(pid int, addr uintptr, data []byte) (count int, err error) {
1691	return ptracePoke(PTRACE_POKETEXT, PTRACE_PEEKTEXT, pid, addr, data)
1692}
1693
1694func PtracePokeData(pid int, addr uintptr, data []byte) (count int, err error) {
1695	return ptracePoke(PTRACE_POKEDATA, PTRACE_PEEKDATA, pid, addr, data)
1696}
1697
1698func PtracePokeUser(pid int, addr uintptr, data []byte) (count int, err error) {
1699	return ptracePoke(PTRACE_POKEUSR, PTRACE_PEEKUSR, pid, addr, data)
1700}
1701
1702func PtraceGetRegs(pid int, regsout *PtraceRegs) (err error) {
1703	return ptracePtr(PTRACE_GETREGS, pid, 0, unsafe.Pointer(regsout))
1704}
1705
1706func PtraceSetRegs(pid int, regs *PtraceRegs) (err error) {
1707	return ptracePtr(PTRACE_SETREGS, pid, 0, unsafe.Pointer(regs))
1708}
1709
1710func PtraceSetOptions(pid int, options int) (err error) {
1711	return ptrace(PTRACE_SETOPTIONS, pid, 0, uintptr(options))
1712}
1713
1714func PtraceGetEventMsg(pid int) (msg uint, err error) {
1715	var data _C_long
1716	err = ptracePtr(PTRACE_GETEVENTMSG, pid, 0, unsafe.Pointer(&data))
1717	msg = uint(data)
1718	return
1719}
1720
1721func PtraceCont(pid int, signal int) (err error) {
1722	return ptrace(PTRACE_CONT, pid, 0, uintptr(signal))
1723}
1724
1725func PtraceSyscall(pid int, signal int) (err error) {
1726	return ptrace(PTRACE_SYSCALL, pid, 0, uintptr(signal))
1727}
1728
1729func PtraceSingleStep(pid int) (err error) { return ptrace(PTRACE_SINGLESTEP, pid, 0, 0) }
1730
1731func PtraceInterrupt(pid int) (err error) { return ptrace(PTRACE_INTERRUPT, pid, 0, 0) }
1732
1733func PtraceAttach(pid int) (err error) { return ptrace(PTRACE_ATTACH, pid, 0, 0) }
1734
1735func PtraceSeize(pid int) (err error) { return ptrace(PTRACE_SEIZE, pid, 0, 0) }
1736
1737func PtraceDetach(pid int) (err error) { return ptrace(PTRACE_DETACH, pid, 0, 0) }
1738
1739//sys	reboot(magic1 uint, magic2 uint, cmd int, arg string) (err error)
1740
1741func Reboot(cmd int) (err error) {
1742	return reboot(LINUX_REBOOT_MAGIC1, LINUX_REBOOT_MAGIC2, cmd, "")
1743}
1744
1745func direntIno(buf []byte) (uint64, bool) {
1746	return readInt(buf, unsafe.Offsetof(Dirent{}.Ino), unsafe.Sizeof(Dirent{}.Ino))
1747}
1748
1749func direntReclen(buf []byte) (uint64, bool) {
1750	return readInt(buf, unsafe.Offsetof(Dirent{}.Reclen), unsafe.Sizeof(Dirent{}.Reclen))
1751}
1752
1753func direntNamlen(buf []byte) (uint64, bool) {
1754	reclen, ok := direntReclen(buf)
1755	if !ok {
1756		return 0, false
1757	}
1758	return reclen - uint64(unsafe.Offsetof(Dirent{}.Name)), true
1759}
1760
1761//sys	mount(source string, target string, fstype string, flags uintptr, data *byte) (err error)
1762
1763func Mount(source string, target string, fstype string, flags uintptr, data string) (err error) {
1764	// Certain file systems get rather angry and EINVAL if you give
1765	// them an empty string of data, rather than NULL.
1766	if data == "" {
1767		return mount(source, target, fstype, flags, nil)
1768	}
1769	datap, err := BytePtrFromString(data)
1770	if err != nil {
1771		return err
1772	}
1773	return mount(source, target, fstype, flags, datap)
1774}
1775
1776//sys	mountSetattr(dirfd int, pathname string, flags uint, attr *MountAttr, size uintptr) (err error) = SYS_MOUNT_SETATTR
1777
1778// MountSetattr is a wrapper for mount_setattr(2).
1779// https://man7.org/linux/man-pages/man2/mount_setattr.2.html
1780//
1781// Requires kernel >= 5.12.
1782func MountSetattr(dirfd int, pathname string, flags uint, attr *MountAttr) error {
1783	return mountSetattr(dirfd, pathname, flags, attr, unsafe.Sizeof(*attr))
1784}
1785
1786func Sendfile(outfd int, infd int, offset *int64, count int) (written int, err error) {
1787	if raceenabled {
1788		raceReleaseMerge(unsafe.Pointer(&ioSync))
1789	}
1790	return sendfile(outfd, infd, offset, count)
1791}
1792
1793// Sendto
1794// Recvfrom
1795// Socketpair
1796
1797/*
1798 * Direct access
1799 */
1800//sys	Acct(path string) (err error)
1801//sys	AddKey(keyType string, description string, payload []byte, ringid int) (id int, err error)
1802//sys	Adjtimex(buf *Timex) (state int, err error)
1803//sysnb	Capget(hdr *CapUserHeader, data *CapUserData) (err error)
1804//sysnb	Capset(hdr *CapUserHeader, data *CapUserData) (err error)
1805//sys	Chdir(path string) (err error)
1806//sys	Chroot(path string) (err error)
1807//sys	ClockAdjtime(clockid int32, buf *Timex) (state int, err error)
1808//sys	ClockGetres(clockid int32, res *Timespec) (err error)
1809//sys	ClockGettime(clockid int32, time *Timespec) (err error)
1810//sys	ClockNanosleep(clockid int32, flags int, request *Timespec, remain *Timespec) (err error)
1811//sys	Close(fd int) (err error)
1812//sys	CloseRange(first uint, last uint, flags uint) (err error)
1813//sys	CopyFileRange(rfd int, roff *int64, wfd int, woff *int64, len int, flags int) (n int, err error)
1814//sys	DeleteModule(name string, flags int) (err error)
1815//sys	Dup(oldfd int) (fd int, err error)
1816
1817func Dup2(oldfd, newfd int) error {
1818	return Dup3(oldfd, newfd, 0)
1819}
1820
1821//sys	Dup3(oldfd int, newfd int, flags int) (err error)
1822//sysnb	EpollCreate1(flag int) (fd int, err error)
1823//sysnb	EpollCtl(epfd int, op int, fd int, event *EpollEvent) (err error)
1824//sys	Eventfd(initval uint, flags int) (fd int, err error) = SYS_EVENTFD2
1825//sys	Exit(code int) = SYS_EXIT_GROUP
1826//sys	Fallocate(fd int, mode uint32, off int64, len int64) (err error)
1827//sys	Fchdir(fd int) (err error)
1828//sys	Fchmod(fd int, mode uint32) (err error)
1829//sys	Fchownat(dirfd int, path string, uid int, gid int, flags int) (err error)
1830//sys	Fdatasync(fd int) (err error)
1831//sys	Fgetxattr(fd int, attr string, dest []byte) (sz int, err error)
1832//sys	FinitModule(fd int, params string, flags int) (err error)
1833//sys	Flistxattr(fd int, dest []byte) (sz int, err error)
1834//sys	Flock(fd int, how int) (err error)
1835//sys	Fremovexattr(fd int, attr string) (err error)
1836//sys	Fsetxattr(fd int, attr string, dest []byte, flags int) (err error)
1837//sys	Fsync(fd int) (err error)
1838//sys	Fsmount(fd int, flags int, mountAttrs int) (fsfd int, err error)
1839//sys	Fsopen(fsName string, flags int) (fd int, err error)
1840//sys	Fspick(dirfd int, pathName string, flags int) (fd int, err error)
1841//sys	Getdents(fd int, buf []byte) (n int, err error) = SYS_GETDENTS64
1842//sysnb	Getpgid(pid int) (pgid int, err error)
1843
1844func Getpgrp() (pid int) {
1845	pid, _ = Getpgid(0)
1846	return
1847}
1848
1849//sysnb	Getpid() (pid int)
1850//sysnb	Getppid() (ppid int)
1851//sys	Getpriority(which int, who int) (prio int, err error)
1852//sys	Getrandom(buf []byte, flags int) (n int, err error)
1853//sysnb	Getrusage(who int, rusage *Rusage) (err error)
1854//sysnb	Getsid(pid int) (sid int, err error)
1855//sysnb	Gettid() (tid int)
1856//sys	Getxattr(path string, attr string, dest []byte) (sz int, err error)
1857//sys	InitModule(moduleImage []byte, params string) (err error)
1858//sys	InotifyAddWatch(fd int, pathname string, mask uint32) (watchdesc int, err error)
1859//sysnb	InotifyInit1(flags int) (fd int, err error)
1860//sysnb	InotifyRmWatch(fd int, watchdesc uint32) (success int, err error)
1861//sysnb	Kill(pid int, sig syscall.Signal) (err error)
1862//sys	Klogctl(typ int, buf []byte) (n int, err error) = SYS_SYSLOG
1863//sys	Lgetxattr(path string, attr string, dest []byte) (sz int, err error)
1864//sys	Listxattr(path string, dest []byte) (sz int, err error)
1865//sys	Llistxattr(path string, dest []byte) (sz int, err error)
1866//sys	Lremovexattr(path string, attr string) (err error)
1867//sys	Lsetxattr(path string, attr string, data []byte, flags int) (err error)
1868//sys	MemfdCreate(name string, flags int) (fd int, err error)
1869//sys	Mkdirat(dirfd int, path string, mode uint32) (err error)
1870//sys	Mknodat(dirfd int, path string, mode uint32, dev int) (err error)
1871//sys	MoveMount(fromDirfd int, fromPathName string, toDirfd int, toPathName string, flags int) (err error)
1872//sys	Nanosleep(time *Timespec, leftover *Timespec) (err error)
1873//sys	OpenTree(dfd int, fileName string, flags uint) (r int, err error)
1874//sys	PerfEventOpen(attr *PerfEventAttr, pid int, cpu int, groupFd int, flags int) (fd int, err error)
1875//sys	PivotRoot(newroot string, putold string) (err error) = SYS_PIVOT_ROOT
1876//sys	Prctl(option int, arg2 uintptr, arg3 uintptr, arg4 uintptr, arg5 uintptr) (err error)
1877//sys	Pselect(nfd int, r *FdSet, w *FdSet, e *FdSet, timeout *Timespec, sigmask *Sigset_t) (n int, err error) = SYS_PSELECT6
1878//sys	read(fd int, p []byte) (n int, err error)
1879//sys	Removexattr(path string, attr string) (err error)
1880//sys	Renameat2(olddirfd int, oldpath string, newdirfd int, newpath string, flags uint) (err error)
1881//sys	RequestKey(keyType string, description string, callback string, destRingid int) (id int, err error)
1882//sys	Setdomainname(p []byte) (err error)
1883//sys	Sethostname(p []byte) (err error)
1884//sysnb	Setpgid(pid int, pgid int) (err error)
1885//sysnb	Setsid() (pid int, err error)
1886//sysnb	Settimeofday(tv *Timeval) (err error)
1887//sys	Setns(fd int, nstype int) (err error)
1888
1889//go:linkname syscall_prlimit syscall.prlimit
1890func syscall_prlimit(pid, resource int, newlimit, old *syscall.Rlimit) error
1891
1892func Prlimit(pid, resource int, newlimit, old *Rlimit) error {
1893	// Just call the syscall version, because as of Go 1.21
1894	// it will affect starting a new process.
1895	return syscall_prlimit(pid, resource, (*syscall.Rlimit)(newlimit), (*syscall.Rlimit)(old))
1896}
1897
1898// PrctlRetInt performs a prctl operation specified by option and further
1899// optional arguments arg2 through arg5 depending on option. It returns a
1900// non-negative integer that is returned by the prctl syscall.
1901func PrctlRetInt(option int, arg2 uintptr, arg3 uintptr, arg4 uintptr, arg5 uintptr) (int, error) {
1902	ret, _, err := Syscall6(SYS_PRCTL, uintptr(option), uintptr(arg2), uintptr(arg3), uintptr(arg4), uintptr(arg5), 0)
1903	if err != 0 {
1904		return 0, err
1905	}
1906	return int(ret), nil
1907}
1908
1909func Setuid(uid int) (err error) {
1910	return syscall.Setuid(uid)
1911}
1912
1913func Setgid(gid int) (err error) {
1914	return syscall.Setgid(gid)
1915}
1916
1917func Setreuid(ruid, euid int) (err error) {
1918	return syscall.Setreuid(ruid, euid)
1919}
1920
1921func Setregid(rgid, egid int) (err error) {
1922	return syscall.Setregid(rgid, egid)
1923}
1924
1925func Setresuid(ruid, euid, suid int) (err error) {
1926	return syscall.Setresuid(ruid, euid, suid)
1927}
1928
1929func Setresgid(rgid, egid, sgid int) (err error) {
1930	return syscall.Setresgid(rgid, egid, sgid)
1931}
1932
1933// SetfsgidRetGid sets fsgid for current thread and returns previous fsgid set.
1934// setfsgid(2) will return a non-nil error only if its caller lacks CAP_SETUID capability.
1935// If the call fails due to other reasons, current fsgid will be returned.
1936func SetfsgidRetGid(gid int) (int, error) {
1937	return setfsgid(gid)
1938}
1939
1940// SetfsuidRetUid sets fsuid for current thread and returns previous fsuid set.
1941// setfsgid(2) will return a non-nil error only if its caller lacks CAP_SETUID capability
1942// If the call fails due to other reasons, current fsuid will be returned.
1943func SetfsuidRetUid(uid int) (int, error) {
1944	return setfsuid(uid)
1945}
1946
1947func Setfsgid(gid int) error {
1948	_, err := setfsgid(gid)
1949	return err
1950}
1951
1952func Setfsuid(uid int) error {
1953	_, err := setfsuid(uid)
1954	return err
1955}
1956
1957func Signalfd(fd int, sigmask *Sigset_t, flags int) (newfd int, err error) {
1958	return signalfd(fd, sigmask, _C__NSIG/8, flags)
1959}
1960
1961//sys	Setpriority(which int, who int, prio int) (err error)
1962//sys	Setxattr(path string, attr string, data []byte, flags int) (err error)
1963//sys	signalfd(fd int, sigmask *Sigset_t, maskSize uintptr, flags int) (newfd int, err error) = SYS_SIGNALFD4
1964//sys	Statx(dirfd int, path string, flags int, mask int, stat *Statx_t) (err error)
1965//sys	Sync()
1966//sys	Syncfs(fd int) (err error)
1967//sysnb	Sysinfo(info *Sysinfo_t) (err error)
1968//sys	Tee(rfd int, wfd int, len int, flags int) (n int64, err error)
1969//sysnb	TimerfdCreate(clockid int, flags int) (fd int, err error)
1970//sysnb	TimerfdGettime(fd int, currValue *ItimerSpec) (err error)
1971//sysnb	TimerfdSettime(fd int, flags int, newValue *ItimerSpec, oldValue *ItimerSpec) (err error)
1972//sysnb	Tgkill(tgid int, tid int, sig syscall.Signal) (err error)
1973//sysnb	Times(tms *Tms) (ticks uintptr, err error)
1974//sysnb	Umask(mask int) (oldmask int)
1975//sysnb	Uname(buf *Utsname) (err error)
1976//sys	Unmount(target string, flags int) (err error) = SYS_UMOUNT2
1977//sys	Unshare(flags int) (err error)
1978//sys	write(fd int, p []byte) (n int, err error)
1979//sys	exitThread(code int) (err error) = SYS_EXIT
1980//sys	readlen(fd int, p *byte, np int) (n int, err error) = SYS_READ
1981//sys	writelen(fd int, p *byte, np int) (n int, err error) = SYS_WRITE
1982//sys	readv(fd int, iovs []Iovec) (n int, err error) = SYS_READV
1983//sys	writev(fd int, iovs []Iovec) (n int, err error) = SYS_WRITEV
1984//sys	preadv(fd int, iovs []Iovec, offs_l uintptr, offs_h uintptr) (n int, err error) = SYS_PREADV
1985//sys	pwritev(fd int, iovs []Iovec, offs_l uintptr, offs_h uintptr) (n int, err error) = SYS_PWRITEV
1986//sys	preadv2(fd int, iovs []Iovec, offs_l uintptr, offs_h uintptr, flags int) (n int, err error) = SYS_PREADV2
1987//sys	pwritev2(fd int, iovs []Iovec, offs_l uintptr, offs_h uintptr, flags int) (n int, err error) = SYS_PWRITEV2
1988
1989// minIovec is the size of the small initial allocation used by
1990// Readv, Writev, etc.
1991//
1992// This small allocation gets stack allocated, which lets the
1993// common use case of len(iovs) <= minIovs avoid more expensive
1994// heap allocations.
1995const minIovec = 8
1996
1997// appendBytes converts bs to Iovecs and appends them to vecs.
1998func appendBytes(vecs []Iovec, bs [][]byte) []Iovec {
1999	for _, b := range bs {
2000		var v Iovec
2001		v.SetLen(len(b))
2002		if len(b) > 0 {
2003			v.Base = &b[0]
2004		} else {
2005			v.Base = (*byte)(unsafe.Pointer(&_zero))
2006		}
2007		vecs = append(vecs, v)
2008	}
2009	return vecs
2010}
2011
2012// offs2lohi splits offs into its low and high order bits.
2013func offs2lohi(offs int64) (lo, hi uintptr) {
2014	const longBits = SizeofLong * 8
2015	return uintptr(offs), uintptr(uint64(offs) >> (longBits - 1) >> 1) // two shifts to avoid false positive in vet
2016}
2017
2018func Readv(fd int, iovs [][]byte) (n int, err error) {
2019	iovecs := make([]Iovec, 0, minIovec)
2020	iovecs = appendBytes(iovecs, iovs)
2021	n, err = readv(fd, iovecs)
2022	readvRacedetect(iovecs, n, err)
2023	return n, err
2024}
2025
2026func Preadv(fd int, iovs [][]byte, offset int64) (n int, err error) {
2027	iovecs := make([]Iovec, 0, minIovec)
2028	iovecs = appendBytes(iovecs, iovs)
2029	lo, hi := offs2lohi(offset)
2030	n, err = preadv(fd, iovecs, lo, hi)
2031	readvRacedetect(iovecs, n, err)
2032	return n, err
2033}
2034
2035func Preadv2(fd int, iovs [][]byte, offset int64, flags int) (n int, err error) {
2036	iovecs := make([]Iovec, 0, minIovec)
2037	iovecs = appendBytes(iovecs, iovs)
2038	lo, hi := offs2lohi(offset)
2039	n, err = preadv2(fd, iovecs, lo, hi, flags)
2040	readvRacedetect(iovecs, n, err)
2041	return n, err
2042}
2043
2044func readvRacedetect(iovecs []Iovec, n int, err error) {
2045	if !raceenabled {
2046		return
2047	}
2048	for i := 0; n > 0 && i < len(iovecs); i++ {
2049		m := int(iovecs[i].Len)
2050		if m > n {
2051			m = n
2052		}
2053		n -= m
2054		if m > 0 {
2055			raceWriteRange(unsafe.Pointer(iovecs[i].Base), m)
2056		}
2057	}
2058	if err == nil {
2059		raceAcquire(unsafe.Pointer(&ioSync))
2060	}
2061}
2062
2063func Writev(fd int, iovs [][]byte) (n int, err error) {
2064	iovecs := make([]Iovec, 0, minIovec)
2065	iovecs = appendBytes(iovecs, iovs)
2066	if raceenabled {
2067		raceReleaseMerge(unsafe.Pointer(&ioSync))
2068	}
2069	n, err = writev(fd, iovecs)
2070	writevRacedetect(iovecs, n)
2071	return n, err
2072}
2073
2074func Pwritev(fd int, iovs [][]byte, offset int64) (n int, err error) {
2075	iovecs := make([]Iovec, 0, minIovec)
2076	iovecs = appendBytes(iovecs, iovs)
2077	if raceenabled {
2078		raceReleaseMerge(unsafe.Pointer(&ioSync))
2079	}
2080	lo, hi := offs2lohi(offset)
2081	n, err = pwritev(fd, iovecs, lo, hi)
2082	writevRacedetect(iovecs, n)
2083	return n, err
2084}
2085
2086func Pwritev2(fd int, iovs [][]byte, offset int64, flags int) (n int, err error) {
2087	iovecs := make([]Iovec, 0, minIovec)
2088	iovecs = appendBytes(iovecs, iovs)
2089	if raceenabled {
2090		raceReleaseMerge(unsafe.Pointer(&ioSync))
2091	}
2092	lo, hi := offs2lohi(offset)
2093	n, err = pwritev2(fd, iovecs, lo, hi, flags)
2094	writevRacedetect(iovecs, n)
2095	return n, err
2096}
2097
2098func writevRacedetect(iovecs []Iovec, n int) {
2099	if !raceenabled {
2100		return
2101	}
2102	for i := 0; n > 0 && i < len(iovecs); i++ {
2103		m := int(iovecs[i].Len)
2104		if m > n {
2105			m = n
2106		}
2107		n -= m
2108		if m > 0 {
2109			raceReadRange(unsafe.Pointer(iovecs[i].Base), m)
2110		}
2111	}
2112}
2113
2114// mmap varies by architecture; see syscall_linux_*.go.
2115//sys	munmap(addr uintptr, length uintptr) (err error)
2116
2117var mapper = &mmapper{
2118	active: make(map[*byte][]byte),
2119	mmap:   mmap,
2120	munmap: munmap,
2121}
2122
2123func Mmap(fd int, offset int64, length int, prot int, flags int) (data []byte, err error) {
2124	return mapper.Mmap(fd, offset, length, prot, flags)
2125}
2126
2127func Munmap(b []byte) (err error) {
2128	return mapper.Munmap(b)
2129}
2130
2131//sys	Madvise(b []byte, advice int) (err error)
2132//sys	Mprotect(b []byte, prot int) (err error)
2133//sys	Mlock(b []byte) (err error)
2134//sys	Mlockall(flags int) (err error)
2135//sys	Msync(b []byte, flags int) (err error)
2136//sys	Munlock(b []byte) (err error)
2137//sys	Munlockall() (err error)
2138
2139// Vmsplice splices user pages from a slice of Iovecs into a pipe specified by fd,
2140// using the specified flags.
2141func Vmsplice(fd int, iovs []Iovec, flags int) (int, error) {
2142	var p unsafe.Pointer
2143	if len(iovs) > 0 {
2144		p = unsafe.Pointer(&iovs[0])
2145	}
2146
2147	n, _, errno := Syscall6(SYS_VMSPLICE, uintptr(fd), uintptr(p), uintptr(len(iovs)), uintptr(flags), 0, 0)
2148	if errno != 0 {
2149		return 0, syscall.Errno(errno)
2150	}
2151
2152	return int(n), nil
2153}
2154
2155func isGroupMember(gid int) bool {
2156	groups, err := Getgroups()
2157	if err != nil {
2158		return false
2159	}
2160
2161	for _, g := range groups {
2162		if g == gid {
2163			return true
2164		}
2165	}
2166	return false
2167}
2168
2169func isCapDacOverrideSet() bool {
2170	hdr := CapUserHeader{Version: LINUX_CAPABILITY_VERSION_3}
2171	data := [2]CapUserData{}
2172	err := Capget(&hdr, &data[0])
2173
2174	return err == nil && data[0].Effective&(1<<CAP_DAC_OVERRIDE) != 0
2175}
2176
2177//sys	faccessat(dirfd int, path string, mode uint32) (err error)
2178//sys	Faccessat2(dirfd int, path string, mode uint32, flags int) (err error)
2179
2180func Faccessat(dirfd int, path string, mode uint32, flags int) (err error) {
2181	if flags == 0 {
2182		return faccessat(dirfd, path, mode)
2183	}
2184
2185	if err := Faccessat2(dirfd, path, mode, flags); err != ENOSYS && err != EPERM {
2186		return err
2187	}
2188
2189	// The Linux kernel faccessat system call does not take any flags.
2190	// The glibc faccessat implements the flags itself; see
2191	// https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/faccessat.c;hb=HEAD
2192	// Because people naturally expect syscall.Faccessat to act
2193	// like C faccessat, we do the same.
2194
2195	if flags & ^(AT_SYMLINK_NOFOLLOW|AT_EACCESS) != 0 {
2196		return EINVAL
2197	}
2198
2199	var st Stat_t
2200	if err := Fstatat(dirfd, path, &st, flags&AT_SYMLINK_NOFOLLOW); err != nil {
2201		return err
2202	}
2203
2204	mode &= 7
2205	if mode == 0 {
2206		return nil
2207	}
2208
2209	var uid int
2210	if flags&AT_EACCESS != 0 {
2211		uid = Geteuid()
2212		if uid != 0 && isCapDacOverrideSet() {
2213			// If CAP_DAC_OVERRIDE is set, file access check is
2214			// done by the kernel in the same way as for root
2215			// (see generic_permission() in the Linux sources).
2216			uid = 0
2217		}
2218	} else {
2219		uid = Getuid()
2220	}
2221
2222	if uid == 0 {
2223		if mode&1 == 0 {
2224			// Root can read and write any file.
2225			return nil
2226		}
2227		if st.Mode&0111 != 0 {
2228			// Root can execute any file that anybody can execute.
2229			return nil
2230		}
2231		return EACCES
2232	}
2233
2234	var fmode uint32
2235	if uint32(uid) == st.Uid {
2236		fmode = (st.Mode >> 6) & 7
2237	} else {
2238		var gid int
2239		if flags&AT_EACCESS != 0 {
2240			gid = Getegid()
2241		} else {
2242			gid = Getgid()
2243		}
2244
2245		if uint32(gid) == st.Gid || isGroupMember(int(st.Gid)) {
2246			fmode = (st.Mode >> 3) & 7
2247		} else {
2248			fmode = st.Mode & 7
2249		}
2250	}
2251
2252	if fmode&mode == mode {
2253		return nil
2254	}
2255
2256	return EACCES
2257}
2258
2259//sys	nameToHandleAt(dirFD int, pathname string, fh *fileHandle, mountID *_C_int, flags int) (err error) = SYS_NAME_TO_HANDLE_AT
2260//sys	openByHandleAt(mountFD int, fh *fileHandle, flags int) (fd int, err error) = SYS_OPEN_BY_HANDLE_AT
2261
2262// fileHandle is the argument to nameToHandleAt and openByHandleAt. We
2263// originally tried to generate it via unix/linux/types.go with "type
2264// fileHandle C.struct_file_handle" but that generated empty structs
2265// for mips64 and mips64le. Instead, hard code it for now (it's the
2266// same everywhere else) until the mips64 generator issue is fixed.
2267type fileHandle struct {
2268	Bytes uint32
2269	Type  int32
2270}
2271
2272// FileHandle represents the C struct file_handle used by
2273// name_to_handle_at (see NameToHandleAt) and open_by_handle_at (see
2274// OpenByHandleAt).
2275type FileHandle struct {
2276	*fileHandle
2277}
2278
2279// NewFileHandle constructs a FileHandle.
2280func NewFileHandle(handleType int32, handle []byte) FileHandle {
2281	const hdrSize = unsafe.Sizeof(fileHandle{})
2282	buf := make([]byte, hdrSize+uintptr(len(handle)))
2283	copy(buf[hdrSize:], handle)
2284	fh := (*fileHandle)(unsafe.Pointer(&buf[0]))
2285	fh.Type = handleType
2286	fh.Bytes = uint32(len(handle))
2287	return FileHandle{fh}
2288}
2289
2290func (fh *FileHandle) Size() int   { return int(fh.fileHandle.Bytes) }
2291func (fh *FileHandle) Type() int32 { return fh.fileHandle.Type }
2292func (fh *FileHandle) Bytes() []byte {
2293	n := fh.Size()
2294	if n == 0 {
2295		return nil
2296	}
2297	return unsafe.Slice((*byte)(unsafe.Pointer(uintptr(unsafe.Pointer(&fh.fileHandle.Type))+4)), n)
2298}
2299
2300// NameToHandleAt wraps the name_to_handle_at system call; it obtains
2301// a handle for a path name.
2302func NameToHandleAt(dirfd int, path string, flags int) (handle FileHandle, mountID int, err error) {
2303	var mid _C_int
2304	// Try first with a small buffer, assuming the handle will
2305	// only be 32 bytes.
2306	size := uint32(32 + unsafe.Sizeof(fileHandle{}))
2307	didResize := false
2308	for {
2309		buf := make([]byte, size)
2310		fh := (*fileHandle)(unsafe.Pointer(&buf[0]))
2311		fh.Bytes = size - uint32(unsafe.Sizeof(fileHandle{}))
2312		err = nameToHandleAt(dirfd, path, fh, &mid, flags)
2313		if err == EOVERFLOW {
2314			if didResize {
2315				// We shouldn't need to resize more than once
2316				return
2317			}
2318			didResize = true
2319			size = fh.Bytes + uint32(unsafe.Sizeof(fileHandle{}))
2320			continue
2321		}
2322		if err != nil {
2323			return
2324		}
2325		return FileHandle{fh}, int(mid), nil
2326	}
2327}
2328
2329// OpenByHandleAt wraps the open_by_handle_at system call; it opens a
2330// file via a handle as previously returned by NameToHandleAt.
2331func OpenByHandleAt(mountFD int, handle FileHandle, flags int) (fd int, err error) {
2332	return openByHandleAt(mountFD, handle.fileHandle, flags)
2333}
2334
2335// Klogset wraps the sys_syslog system call; it sets console_loglevel to
2336// the value specified by arg and passes a dummy pointer to bufp.
2337func Klogset(typ int, arg int) (err error) {
2338	var p unsafe.Pointer
2339	_, _, errno := Syscall(SYS_SYSLOG, uintptr(typ), uintptr(p), uintptr(arg))
2340	if errno != 0 {
2341		return errnoErr(errno)
2342	}
2343	return nil
2344}
2345
2346// RemoteIovec is Iovec with the pointer replaced with an integer.
2347// It is used for ProcessVMReadv and ProcessVMWritev, where the pointer
2348// refers to a location in a different process' address space, which
2349// would confuse the Go garbage collector.
2350type RemoteIovec struct {
2351	Base uintptr
2352	Len  int
2353}
2354
2355//sys	ProcessVMReadv(pid int, localIov []Iovec, remoteIov []RemoteIovec, flags uint) (n int, err error) = SYS_PROCESS_VM_READV
2356//sys	ProcessVMWritev(pid int, localIov []Iovec, remoteIov []RemoteIovec, flags uint) (n int, err error) = SYS_PROCESS_VM_WRITEV
2357
2358//sys	PidfdOpen(pid int, flags int) (fd int, err error) = SYS_PIDFD_OPEN
2359//sys	PidfdGetfd(pidfd int, targetfd int, flags int) (fd int, err error) = SYS_PIDFD_GETFD
2360//sys	PidfdSendSignal(pidfd int, sig Signal, info *Siginfo, flags int) (err error) = SYS_PIDFD_SEND_SIGNAL
2361
2362//sys	shmat(id int, addr uintptr, flag int) (ret uintptr, err error)
2363//sys	shmctl(id int, cmd int, buf *SysvShmDesc) (result int, err error)
2364//sys	shmdt(addr uintptr) (err error)
2365//sys	shmget(key int, size int, flag int) (id int, err error)
2366
2367//sys	getitimer(which int, currValue *Itimerval) (err error)
2368//sys	setitimer(which int, newValue *Itimerval, oldValue *Itimerval) (err error)
2369
2370// MakeItimerval creates an Itimerval from interval and value durations.
2371func MakeItimerval(interval, value time.Duration) Itimerval {
2372	return Itimerval{
2373		Interval: NsecToTimeval(interval.Nanoseconds()),
2374		Value:    NsecToTimeval(value.Nanoseconds()),
2375	}
2376}
2377
2378// A value which may be passed to the which parameter for Getitimer and
2379// Setitimer.
2380type ItimerWhich int
2381
2382// Possible which values for Getitimer and Setitimer.
2383const (
2384	ItimerReal    ItimerWhich = ITIMER_REAL
2385	ItimerVirtual ItimerWhich = ITIMER_VIRTUAL
2386	ItimerProf    ItimerWhich = ITIMER_PROF
2387)
2388
2389// Getitimer wraps getitimer(2) to return the current value of the timer
2390// specified by which.
2391func Getitimer(which ItimerWhich) (Itimerval, error) {
2392	var it Itimerval
2393	if err := getitimer(int(which), &it); err != nil {
2394		return Itimerval{}, err
2395	}
2396
2397	return it, nil
2398}
2399
2400// Setitimer wraps setitimer(2) to arm or disarm the timer specified by which.
2401// It returns the previous value of the timer.
2402//
2403// If the Itimerval argument is the zero value, the timer will be disarmed.
2404func Setitimer(which ItimerWhich, it Itimerval) (Itimerval, error) {
2405	var prev Itimerval
2406	if err := setitimer(int(which), &it, &prev); err != nil {
2407		return Itimerval{}, err
2408	}
2409
2410	return prev, nil
2411}
2412
2413//sysnb	rtSigprocmask(how int, set *Sigset_t, oldset *Sigset_t, sigsetsize uintptr) (err error) = SYS_RT_SIGPROCMASK
2414
2415func PthreadSigmask(how int, set, oldset *Sigset_t) error {
2416	if oldset != nil {
2417		// Explicitly clear in case Sigset_t is larger than _C__NSIG.
2418		*oldset = Sigset_t{}
2419	}
2420	return rtSigprocmask(how, set, oldset, _C__NSIG/8)
2421}
2422
2423/*
2424 * Unimplemented
2425 */
2426// AfsSyscall
2427// ArchPrctl
2428// Brk
2429// ClockNanosleep
2430// ClockSettime
2431// Clone
2432// EpollCtlOld
2433// EpollPwait
2434// EpollWaitOld
2435// Execve
2436// Fork
2437// Futex
2438// GetKernelSyms
2439// GetMempolicy
2440// GetRobustList
2441// GetThreadArea
2442// Getpmsg
2443// IoCancel
2444// IoDestroy
2445// IoGetevents
2446// IoSetup
2447// IoSubmit
2448// IoprioGet
2449// IoprioSet
2450// KexecLoad
2451// LookupDcookie
2452// Mbind
2453// MigratePages
2454// Mincore
2455// ModifyLdt
2456// Mount
2457// MovePages
2458// MqGetsetattr
2459// MqNotify
2460// MqOpen
2461// MqTimedreceive
2462// MqTimedsend
2463// MqUnlink
2464// Mremap
2465// Msgctl
2466// Msgget
2467// Msgrcv
2468// Msgsnd
2469// Nfsservctl
2470// Personality
2471// Pselect6
2472// Ptrace
2473// Putpmsg
2474// Quotactl
2475// Readahead
2476// Readv
2477// RemapFilePages
2478// RestartSyscall
2479// RtSigaction
2480// RtSigpending
2481// RtSigqueueinfo
2482// RtSigreturn
2483// RtSigsuspend
2484// RtSigtimedwait
2485// SchedGetPriorityMax
2486// SchedGetPriorityMin
2487// SchedGetparam
2488// SchedGetscheduler
2489// SchedRrGetInterval
2490// SchedSetparam
2491// SchedYield
2492// Security
2493// Semctl
2494// Semget
2495// Semop
2496// Semtimedop
2497// SetMempolicy
2498// SetRobustList
2499// SetThreadArea
2500// SetTidAddress
2501// Sigaltstack
2502// Swapoff
2503// Swapon
2504// Sysfs
2505// TimerCreate
2506// TimerDelete
2507// TimerGetoverrun
2508// TimerGettime
2509// TimerSettime
2510// Tkill (obsolete)
2511// Tuxcall
2512// Umount2
2513// Uselib
2514// Utimensat
2515// Vfork
2516// Vhangup
2517// Vserver
2518// _Sysctl