// epollinit opens an epoll file descriptor and creates a pipe which will be
// used to wake up the epoll_wait(2) function. Then, file descriptor associated
// with inotify event queue and the read end of the pipe are added to epoll set.
// Note that `fd` member must be set before this function is called.
func (i *inotify) epollinit() (err error) {
if i.epfd, err = syscall.EpollCreate1(0); err != nil {
return
}
if err = syscall.Pipe(i.pipefd); err != nil {
return
}
i.epes = []syscall.EpollEvent{
{Events: syscall.EPOLLIN, Fd: i.fd},
{Events: syscall.EPOLLIN, Fd: int32(i.pipefd[0])},
}
if err = syscall.EpollCtl(i.epfd, syscall.EPOLL_CTL_ADD, int(i.fd), &i.epes[0]); err != nil {
return
}
return syscall.EpollCtl(i.epfd, syscall.EPOLL_CTL_ADD, i.pipefd[0], &i.epes[1])
}
// setupEpoll sets up epoll for use
func setupEpoll() {
var err error
epollFD, err = syscall.EpollCreate1(0)
if err != nil {
fmt.Println("Unable to create epoll FD: ", err.Error())
os.Exit(1)
}
go func() {
var epollEvents [GPIOCount]syscall.EpollEvent
for {
numEvents, err := syscall.EpollWait(epollFD, epollEvents[:], -1)
if err != nil {
panic(fmt.Sprintf("EpollWait error: %s", err.Error()))
}
for i := 0; i < numEvents; i++ {
if eventPin, exists := watchEventCallbacks[int(epollEvents[i].Fd)]; exists {
if eventPin.initial {
eventPin.initial = false
} else {
eventPin.callback()
}
}
}
}
}()
}
/*-----------------------------------------------------------------------------
-- FUNCTION: listen
--
-- DATE: February 6, 2016
--
-- REVISIONS: February 10, 2016 - Modified for EPoll
--
-- DESIGNER: Marc Vouve
--
-- PROGRAMMER: Marc Vouve
--
-- INTERFACE: func listen(srvInfo serverInfo)
-- srvInfo: information about the overall application, IPC and networking
--
-- RETURNS: void doesn't actually return.
--
-- NOTES: This function handles Epoll in a foreverloop. This function never
-- exits
------------------------------------------------------------------------------*/
func listen(srvInfo serverInfo) {
client := make(map[int]connectionInfo)
epollFd, _ := syscall.EpollCreate1(0)
events := make([]syscall.EpollEvent, epollQueueLen)
addConnectionToEPoll(epollFd, srvInfo.listener)
for {
n, _ := syscall.EpollWait(epollFd, events[:], -1)
for i := 0; i < n; i++ {
if events[i].Events&(syscall.EPOLLHUP|syscall.EPOLLERR) != 0 {
fmt.Println("Error on epoll")
endConnection(srvInfo, client[int(events[i].Fd)])
}
if events[i].Fd == int32(srvInfo.listener) { // new connection
newClient, err := newConnection(srvInfo.listener)
if err == nil {
srvInfo.serverConnection <- newConnectionConst
client[newClient.FileDescriptor] = newClient
addConnectionToEPoll(epollFd, newClient.FileDescriptor)
}
} else { // data to read from connection
// client[int(ev.Fd)] can not be used as an argument in handleData
conn := client[int(events[i].Fd)]
err := handleData(&conn)
client[int(events[i].Fd)] = conn
if err != nil {
endConnection(srvInfo, client[int(events[i].Fd)])
}
}
}
}
}
func initEpollListener() *epollListener {
fd, err := syscall.EpollCreate1(0)
if err != nil {
panic(fmt.Sprintf("Unable to create epoll: %v", err))
}
listener := &epollListener{fd: fd, interruptablePins: make(map[int]*interrupt)}
go func() {
var epollEvents [MaxGPIOInterrupt]syscall.EpollEvent
for {
n, err := syscall.EpollWait(listener.fd, epollEvents[:], -1)
if err != nil {
panic(fmt.Sprintf("EpollWait error: %v", err))
}
listener.mu.Lock()
for i := 0; i < n; i++ {
if irq, ok := listener.interruptablePins[int(epollEvents[i].Fd)]; ok {
irq.Signal()
}
}
listener.mu.Unlock()
}
}()
return listener
}
func init() {
fdM.Init(128)
fd, err := syscall.EpollCreate1(syscall.EPOLL_CLOEXEC)
if err != nil {
log.Panicf("poller: EpollCreate1: %s", err.Error())
}
epfd = fd
go poller()
}
func NewPerCpuEvents(config *PerfEventConfig) (*PerCpuEvents, error) {
var err error
e := &PerCpuEvents{
cpus: config.NumCpus,
npages: config.NumPages,
pagesize: os.Getpagesize(),
event: make(map[int]*PerfEvent),
}
defer func() {
if err != nil {
e.CloseAll()
}
}()
e.poll.fd, err = syscall.EpollCreate1(0)
if err != nil {
return nil, err
}
e.eventMap, err = openMap(config.MapPath)
if err != nil {
return nil, err
}
for cpu := int(0); cpu < config.NumCpus; cpu++ {
event, err := PerfEventOpen(config, -1, cpu, -1, 0)
if err != nil {
return nil, err
} else {
e.event[event.Fd] = event
}
if err := e.poll.AddFD(event.Fd, syscall.EPOLLIN); err != nil {
return nil, err
}
if err = event.Mmap(e.pagesize, e.npages); err != nil {
return nil, err
}
if err = event.Enable(); err != nil {
return nil, err
}
}
for _, event := range e.event {
// FIXME: Not sure what to do here, the map has already been updated and we can't
// fully restore it.
if err := e.eventMap.Update(event); err != nil {
return nil, err
}
}
return e, nil
}
// NewIO allocates resources which will not be released before program
// termination. You shouldn't need many instances. Panics on error.
func NewIO() (io *IO) {
epollFd, err := syscall.EpollCreate1(syscall.EPOLL_CLOEXEC)
if err != nil {
panic(err)
}
io = &IO{
epollFd: epollFd,
workers: make(map[int32]*worker),
}
go io.eventLoop()
return
}
func New(fd int) (p *Poller, err error) {
p = &Poller{}
if p.epfd, err = syscall.EpollCreate1(0); err != nil {
return nil, err
}
p.event.Events = syscall.EPOLLOUT
p.event.Fd = int32(fd)
if err = syscall.EpollCtl(p.epfd, syscall.EPOLL_CTL_ADD, fd, &p.event); err != nil {
p.Close()
return nil, err
}
return p, nil
}
func newpollster() (p *pollster, err error) {
p = new(pollster)
if p.epfd, err = syscall.EpollCreate1(syscall.EPOLL_CLOEXEC); err != nil {
if err != syscall.ENOSYS {
return nil, os.NewSyscallError("epoll_create1", err)
}
// The arg to epoll_create is a hint to the kernel
// about the number of FDs we will care about.
// We don't know, and since 2.6.8 the kernel ignores it anyhow.
if p.epfd, err = syscall.EpollCreate(16); err != nil {
return nil, os.NewSyscallError("epoll_create", err)
}
syscall.CloseOnExec(p.epfd)
}
p.events = make(map[int]uint32)
return p, nil
}
// Create a new inotify poller.
// This creates an inotify handler, and an epoll handler.
func newFdPoller(fd int) (*fdPoller, error) {
var errno error
poller := emptyPoller(fd)
defer func() {
if errno != nil {
poller.close()
}
}()
poller.fd = fd
// Create epoll fd
poller.epfd, errno = syscall.EpollCreate1(0)
if poller.epfd == -1 {
return nil, errno
}
// Create pipe; pipe[0] is the read end, pipe[1] the write end.
errno = syscall.Pipe2(poller.pipe[:], syscall.O_NONBLOCK)
if errno != nil {
return nil, errno
}
// Register inotify fd with epoll
event := syscall.EpollEvent{
Fd: int32(poller.fd),
Events: syscall.EPOLLIN,
}
errno = syscall.EpollCtl(poller.epfd, syscall.EPOLL_CTL_ADD, poller.fd, &event)
if errno != nil {
return nil, errno
}
// Register pipe fd with epoll
event = syscall.EpollEvent{
Fd: int32(poller.pipe[0]),
Events: syscall.EPOLLIN,
}
errno = syscall.EpollCtl(poller.epfd, syscall.EPOLL_CTL_ADD, poller.pipe[0], &event)
if errno != nil {
return nil, errno
}
return poller, nil
}
// NewWatcher establishes a new watcher with the underlying OS and begins waiting for events.
func NewWatcher() (*Watcher, error) {
fd, errno := syscall.InotifyInit()
if fd == -1 {
return nil, os.NewSyscallError("inotify_init", errno)
}
w := &Watcher{
fd: fd,
watches: make(map[string]*watch),
paths: make(map[int]string),
Events: make(chan Event),
Errors: make(chan error),
done: make(chan bool),
closed: make(chan bool),
}
w.cv = sync.NewCond(&w.mu)
rp, wp, err := os.Pipe() // for done
if err != nil {
return nil, err
}
epfd, err := syscall.EpollCreate1(0)
if err != nil {
return nil, os.NewSyscallError("epoll_create1", err)
}
event := &syscall.EpollEvent{syscall.EPOLLIN, int32(w.fd), 0}
if err = syscall.EpollCtl(epfd, syscall.EPOLL_CTL_ADD, w.fd, event); err != nil {
return nil, os.NewSyscallError("epoll_ctl", err)
}
event = &syscall.EpollEvent{syscall.EPOLLIN, int32(rp.Fd()), 0}
if err = syscall.EpollCtl(epfd, syscall.EPOLL_CTL_ADD, int(rp.Fd()), event); err != nil {
return nil, os.NewSyscallError("epoll_ctl", err)
}
go func() {
<-w.done
wp.Close() // make rp readable
}()
go w.epollEvents(epfd, rp)
return w, nil
}
func DialTCPTimeoutWithMD5Sig(host string, port int, localAddr, key string, msec int) (*net.TCPConn, error) {
var family int
var ra, la syscall.Sockaddr
ip, err := net.ResolveIPAddr("ip", host)
if err != nil {
return nil, fmt.Errorf("invalid ip: %s", err)
}
l, err := net.ResolveIPAddr("ip", localAddr)
if l == nil {
return nil, fmt.Errorf("invalid local ip: %s", err)
}
if (ip.IP.To4() != nil) != (l.IP.To4() != nil) {
return nil, fmt.Errorf("remote and local ip address family is not same")
}
switch {
case ip.IP.To4() != nil:
family = syscall.AF_INET
i := &syscall.SockaddrInet4{
Port: port,
}
for idx, _ := range i.Addr {
i.Addr[idx] = ip.IP.To4()[idx]
}
ra = i
j := &syscall.SockaddrInet4{}
for idx, _ := range j.Addr {
j.Addr[idx] = l.IP.To4()[idx]
}
la = j
default:
family = syscall.AF_INET6
i := &syscall.SockaddrInet6{
Port: port,
}
for idx, _ := range i.Addr {
i.Addr[idx] = ip.IP[idx]
}
ra = i
var zone uint32
if l.Zone != "" {
intf, err := net.InterfaceByName(l.Zone)
if err != nil {
return nil, err
}
zone = uint32(intf.Index)
}
j := &syscall.SockaddrInet6{
ZoneId: zone,
}
for idx, _ := range j.Addr {
j.Addr[idx] = l.IP[idx]
}
la = j
}
sotype := syscall.SOCK_STREAM | syscall.SOCK_CLOEXEC | syscall.SOCK_NONBLOCK
proto := 0
fd, err := syscall.Socket(family, sotype, proto)
if err != nil {
return nil, err
}
fi := os.NewFile(uintptr(fd), "")
defer fi.Close()
t, err := buildTcpMD5Sig(host, key)
if err != nil {
return nil, err
}
if _, _, e := syscall.Syscall6(syscall.SYS_SETSOCKOPT, uintptr(fd),
uintptr(syscall.IPPROTO_TCP), uintptr(TCP_MD5SIG),
uintptr(unsafe.Pointer(&t)), unsafe.Sizeof(t), 0); e > 0 {
return nil, os.NewSyscallError("setsockopt", e)
}
if err = syscall.SetsockoptInt(fd, syscall.SOL_SOCKET, syscall.SO_BROADCAST, 1); err != nil {
return nil, os.NewSyscallError("setsockopt", err)
}
if err = syscall.SetsockoptInt(fd, syscall.IPPROTO_TCP, syscall.TCP_NODELAY, 1); err != nil {
return nil, os.NewSyscallError("setsockopt", err)
}
if err = syscall.Bind(fd, la); err != nil {
return nil, os.NewSyscallError("bind", err)
}
tcpconn := func(fi *os.File) (*net.TCPConn, error) {
conn, err := net.FileConn(fi)
return conn.(*net.TCPConn), err
}
err = syscall.Connect(fd, ra)
switch err {
case syscall.EINPROGRESS, syscall.EALREADY, syscall.EINTR:
// do timeout handling
case nil, syscall.EISCONN:
return tcpconn(fi)
default:
return nil, os.NewSyscallError("connect", err)
}
epfd, e := syscall.EpollCreate1(syscall.EPOLL_CLOEXEC)
if e != nil {
return nil, e
}
defer syscall.Close(epfd)
var event syscall.EpollEvent
events := make([]syscall.EpollEvent, 1)
event.Events = syscall.EPOLLIN | syscall.EPOLLOUT | syscall.EPOLLPRI
event.Fd = int32(fd)
if e = syscall.EpollCtl(epfd, syscall.EPOLL_CTL_ADD, fd, &event); e != nil {
return nil, e
}
for {
nevents, e := syscall.EpollWait(epfd, events, msec)
if e != nil {
return nil, e
}
if nevents == 0 {
return nil, fmt.Errorf("timeout")
} else if nevents == 1 && events[0].Fd == int32(fd) {
nerr, err := syscall.GetsockoptInt(fd, syscall.SOL_SOCKET, syscall.SO_ERROR)
if err != nil {
return nil, os.NewSyscallError("getsockopt", err)
}
switch err := syscall.Errno(nerr); err {
case syscall.EINPROGRESS, syscall.EALREADY, syscall.EINTR:
case syscall.Errno(0), syscall.EISCONN:
return tcpconn(fi)
default:
return nil, os.NewSyscallError("getsockopt", err)
}
} else {
return nil, fmt.Errorf("unexpected epoll behavior")
}
}
}
func (u *NetLinkProbe) start() {
s, err := nl.Subscribe(syscall.NETLINK_ROUTE, syscall.RTNLGRP_LINK)
if err != nil {
logging.GetLogger().Errorf("Failed to subscribe to netlink RTNLGRP_LINK messages: %s", err.Error())
return
}
u.nlSocket = s
defer u.nlSocket.Close()
fd := u.nlSocket.GetFd()
err = syscall.SetNonblock(fd, true)
if err != nil {
logging.GetLogger().Errorf("Failed to set the netlink fd as non-blocking: %s", err.Error())
return
}
epfd, e := syscall.EpollCreate1(0)
if e != nil {
logging.GetLogger().Errorf("Failed to create epoll: %s", err.Error())
return
}
defer syscall.Close(epfd)
u.initialize()
event := syscall.EpollEvent{Events: syscall.EPOLLIN, Fd: int32(fd)}
if e = syscall.EpollCtl(epfd, syscall.EPOLL_CTL_ADD, fd, &event); e != nil {
logging.GetLogger().Errorf("Failed to control epoll: %s", err.Error())
return
}
events := make([]syscall.EpollEvent, maxEpollEvents)
for u.running.Load() == true {
n, err := syscall.EpollWait(epfd, events[:], 1000)
if err != nil {
errno, ok := err.(syscall.Errno)
if ok && errno != syscall.EINTR {
logging.GetLogger().Errorf("Failed to receive from events from netlink: %s", err.Error())
}
continue
}
if n == 0 {
continue
}
msgs, err := s.Receive()
if err != nil {
logging.GetLogger().Errorf("Failed to receive from netlink messages: %s", err.Error())
time.Sleep(1 * time.Second)
continue
}
for _, msg := range msgs {
switch msg.Header.Type {
case syscall.RTM_NEWLINK:
ifmsg := nl.DeserializeIfInfomsg(msg.Data)
u.onLinkAdded(int(ifmsg.Index))
case syscall.RTM_DELLINK:
ifmsg := nl.DeserializeIfInfomsg(msg.Data)
u.onLinkDeleted(int(ifmsg.Index))
}
}
}
}
func doParent(mpid int, stdin_fd, stdout_fd, stderr_fd [2]int) resultPack {
syscall.Close(stdin_fd[0])
syscall.Close(stdout_fd[1])
syscall.Close(stderr_fd[1])
exit_chan := make(chan resultPack)
if err := SetupScope(uint32(mpid)); err != nil {
poePanic(err, "setup systemd scope unit failed")
}
go func() {
runtime.LockOSThread()
sig_chan := make(chan os.Signal, 1)
signal.Notify(sig_chan, syscall.SIGCHLD)
signal.Notify(sig_chan, syscall.SIGTERM)
signal.Notify(sig_chan, syscall.SIGINT)
if err := PtraceSeize(mpid, syscall.PTRACE_O_TRACECLONE|syscall.PTRACE_O_TRACEFORK|PTRACE_O_TRACESECCOMP|syscall.PTRACE_O_TRACEVFORK); err != nil {
poePanic(error(err), "PTRACE_SEIZE failed")
}
for {
sig := <-sig_chan
switch sig {
case syscall.SIGCHLD:
if res := handleSigchld(mpid); res != nil {
exit_chan <- *res
}
case syscall.SIGTERM, syscall.SIGINT:
exit_chan <- resultPack{POE_TIMEDOUT, -1, "Supervisor terminated"}
}
}
}()
// TODO: output order is not preserved
out_chan := make(chan childOutput, 1)
go func() {
epfd, err := syscall.EpollCreate1(0)
if err != nil {
poePanic(err, "epoll_create1 failed")
}
defer syscall.Close(epfd)
event1 := syscall.EpollEvent{Events: syscall.EPOLLIN, Fd: int32(stdout_fd[0]), Pad: 1} // Fd/Pad is epoll_data_t (userdata)
if err := syscall.EpollCtl(epfd, syscall.EPOLL_CTL_ADD, stdout_fd[0], &event1); err != nil {
poePanic(err, "epoll_ctl (EPOLL_CTL_ADD) failed")
}
event2 := syscall.EpollEvent{Events: syscall.EPOLLIN, Fd: int32(stderr_fd[0]), Pad: 2}
if err := syscall.EpollCtl(epfd, syscall.EPOLL_CTL_ADD, stderr_fd[0], &event2); err != nil {
poePanic(err, "epoll_ctl (EPOLL_CTL_ADD) failed")
}
var events [32]syscall.EpollEvent
for {
en, err := syscall.EpollWait(epfd, events[:], -1)
if err != nil {
poePanic(err, "epoll_wait failed")
}
var buf [65536]byte // TODO: how to get PIPE_BUF?
for ev := 0; ev < en; ev++ {
for {
n, err := syscall.Read(int(events[ev].Fd), buf[:])
if err != nil {
break
}
if n > 0 {
nbuf := make([]byte, n)
copy(nbuf, buf[:n])
out_chan <- childOutput{int(events[ev].Pad), n, nbuf}
}
}
}
}
}()
go func() {
for {
out := <-out_chan
var outbuf bytes.Buffer
binary.Write(&outbuf, binary.LittleEndian, int32(out.fd))
binary.Write(&outbuf, binary.LittleEndian, int32(out.n))
if _, err := os.Stdout.Write(outbuf.Bytes()); err != nil {
poePanic(err, "stdout write failed")
}
if _, err := os.Stdout.Write(out.buf); err != nil {
poePanic(err, "stdout write failed")
}
}
}()
go func() {
<-time.After(3 * time.Second)
exit_chan <- resultPack{POE_TIMEDOUT, -1, ""}
}()
if _, err := syscall.Write(stdin_fd[1], []byte{0}); err != nil {
poePanic(err, "write to stdin failed")
}
return <-exit_chan
}
func main() {
flag.Parse()
if SHOWVER {
fmt.Println("Version:", VERSION)
return
}
debug("Phoenix Guest Agent started [pid=" + fmt.Sprintf("%d", os.Getpid()) + "]")
port, err := OpenPort(PORTPATH)
if err != nil {
log.Fatalln("Failed to open character device:", err)
}
defer port.Close()
epollFd, err := syscall.EpollCreate1(0)
if err != nil {
log.Fatalln("Error creating epoll:", err)
}
defer syscall.Close(epollFd)
ctlEvent := syscall.EpollEvent{Events: syscall.EPOLLIN, Fd: int32(port.fd)}
if err := syscall.EpollCtl(epollFd, syscall.EPOLL_CTL_ADD, int(port.fd), &ctlEvent); err != nil {
log.Fatalln("Error registering epoll event:", err)
}
reader := bufio.NewReader(port.f)
cReq := make(chan []byte)
cResp := make(chan *Response, 1)
lock := false
for {
if !lock {
lock = true
go listenPort(cReq, epollFd, reader)
}
select {
case jsonReq := <-cReq:
lock = false
req := &Request{}
if err := json.Unmarshal(jsonReq, &req); err != nil {
debug("JSON parse error:", err)
port.SendError(fmt.Errorf("JSON parse error: %s", err), "")
continue
}
switch req.Command {
case "ping":
port.SendResponse(VERSION, req.Tag)
continue
case "agent-shutdown":
debug("Shutdown command received from client")
return
case "get-commands":
go GetCommandList(cResp, req.Tag)
continue
}
if FROZEN && req.Command != "get-freeze-status" && req.Command != "fs-unfreeze" {
debug("All filesystems are frozen. Cannot execute:", req.Command)
port.SendError(fmt.Errorf("All filesystems are frozen. Cannot execute: %s", req.Command), req.Tag)
continue
}
if _, ok := Commands[req.Command]; !ok {
debug("Unknown command:", req.Command)
port.SendError(fmt.Errorf("Unknown command: %s", req.Command), req.Tag)
continue
}
debug("Processing command:", req.Command+", tag =", req.Tag)
go Commands[req.Command](cResp, req.RawArgs, req.Tag)
case resp := <-cResp:
if resp.Err != nil {
port.SendError(resp.Err, resp.Tag)
} else {
port.SendResponse(resp.Value, resp.Tag)
}
} // end of select
}
}
// EpollCreate1 directly calls syscall.EpollCreate1
func EpollCreate1(flag int) (int, error) {
return syscall.EpollCreate1(flag)
}
func NewFdset() *Fdset {
self := &Fdset{
rListener: make(map[int]fdListener),
wListener: make(map[int]fdListener),
rDeadline: make(map[int]time.Time),
wDeadline: make(map[int]time.Time),
}
efd, e1 := syscall.EpollCreate1(syscall.EPOLL_CLOEXEC)
if e1 != nil {
return nil
}
ctls, e2 := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_STREAM|syscall.SOCK_NONBLOCK|syscall.SOCK_CLOEXEC, 0)
if e2 != nil {
return nil
}
self.ctl = ctls[0]
self.rListener[ctls[1]] = fdDrain(ctls[1])
go func() {
defer func() {
syscall.Close(ctls[1])
self.Close()
}()
evs := make([]syscall.EpollEvent, 8)
for {
fds := make(map[int]uint32)
for fd, _ := range self.rListener {
fds[fd] |= syscall.EPOLLIN
}
for fd, _ := range self.wListener {
fds[fd] |= syscall.EPOLLOUT
}
if len(fds) == 0 {
break
}
have_tm := false
var tm time.Time
for _, t := range self.rDeadline {
if !have_tm || t.Before(tm) {
tm = t
}
have_tm = true
}
for _, t := range self.wDeadline {
if !have_tm || t.Before(tm) {
tm = t
}
have_tm = true
}
for fd, flags := range fds {
if err := syscall.EpollCtl(efd, syscall.EPOLL_CTL_ADD, fd, &syscall.EpollEvent{
Events: flags,
Fd: int32(fd),
}); err != nil {
if fd == int(ctls[1]) {
// maybe ctrl closed
return
} else {
log.Print(err)
delete(fds, fd)
}
}
}
dur := time.Hour
if have_tm {
dur = tm.Sub(time.Now())
}
evLen := 0
if dur > 0 {
if n, err := syscall.EpollWait(efd, evs, int(dur/time.Millisecond)); err != nil {
if eno, ok := err.(syscall.Errno); !ok || !eno.Temporary() {
// notify fdset poll loop error
for _, li := range self.wListener {
li.FdListen(err)
}
for _, li := range self.rListener {
li.FdListen(err)
}
return
}
} else {
evLen = n
}
}
if len(self.wDeadline) > 0 || len(self.rDeadline) > 0 {
tm := time.Now()
for fd, t := range self.wDeadline {
if li, ok := self.wListener[fd]; ok {
if t.Before(tm) {
if !li.FdListen(Timeout(t)) {
delete(self.wListener, fd)
}
delete(self.wDeadline, fd)
}
}
}
for fd, t := range self.rDeadline {
if li, ok := self.rListener[fd]; ok {
if t.Before(tm) {
if !li.FdListen(Timeout(t)) {
delete(self.rListener, fd)
}
delete(self.rDeadline, fd)
}
}
}
}
for _, ev := range evs[:evLen] {
if ev.Events&syscall.EPOLLOUT != 0 {
if li, ok := self.wListener[int(ev.Fd)]; ok {
if !li.FdListen(nil) {
delete(self.wListener, int(ev.Fd))
}
delete(self.wDeadline, int(ev.Fd))
}
}
if ev.Events&syscall.EPOLLIN != 0 {
if li, ok := self.rListener[int(ev.Fd)]; ok {
if !li.FdListen(nil) {
delete(self.rListener, int(ev.Fd))
}
delete(self.rDeadline, int(ev.Fd))
}
}
}
for fd, _ := range fds {
if err := syscall.EpollCtl(efd, syscall.EPOLL_CTL_DEL, fd, nil); err != nil {
panic(err)
}
}
}
}()
return self
}
func (watcher *PinWatcher) pinHandler() {
exf, exf_err := os.OpenFile("/sys/class/gpio/export", syscall.O_WRONLY, 0400)
if exf_err != nil {
log.Panicln("Error opening /sys/class/gpio/export:", exf_err)
}
_, ex_err := exf.WriteString(fmt.Sprintf("%d", watcher.pin))
if ex_err != nil {
log.Panicln("Error writing to /sys/class/gpio/export:", ex_err)
}
exf.Close()
time.Sleep(time.Microsecond)
edge_file := fmt.Sprintf("/sys/class/gpio/gpio%d/edge", watcher.pin)
edgef, edgef_err := os.OpenFile(edge_file, syscall.O_WRONLY, 0400)
if edgef_err != nil {
log.Panicf("Error opening %s: %s\n", edge_file, edgef_err)
}
_, edge_err := edgef.WriteString("both")
if edge_err != nil {
log.Panicf("Error writing to %s: %s\n", edge_file, edge_err)
}
edgef.Close()
time.Sleep(time.Microsecond)
value_file := fmt.Sprintf("/sys/class/gpio/gpio%d/value", watcher.pin)
irq_fd, irq_err := syscall.Open(value_file, syscall.O_RDONLY|syscall.O_NONBLOCK, syscall.S_IREAD)
if irq_err != nil {
log.Panicln("Error opening %s: %s\n", value_file, irq_err)
}
epfd, eperr := syscall.EpollCreate1(syscall.EPOLL_CLOEXEC)
if eperr != nil {
log.Panicln("Error creating epoll:", eperr)
}
event := new(syscall.EpollEvent)
event.Fd = int32(irq_fd)
event.Events |= syscall.EPOLLPRI
if ctlerr := syscall.EpollCtl(epfd, syscall.EPOLL_CTL_ADD, irq_fd, event); ctlerr != nil {
log.Panicln("Error on epoll control operation:", ctlerr)
}
var events_buffer [10]syscall.EpollEvent
var buf [1]byte
for !watcher.shutdown {
n, werr := syscall.EpollWait(epfd, events_buffer[0:], 4000)
if werr != nil {
log.Println("Epoll error:", werr)
} else if n == 1 {
syscall.Seek(irq_fd, 0, 0)
syscall.Read(irq_fd, buf[0:])
log.Println("Interrupt!", events_buffer[0].Fd, events_buffer[0].Events, buf)
} else {
log.Println("Timeout")
}
}
syscall.Close(irq_fd)
unexf, unexf_err := os.OpenFile("/sys/class/gpio/unexport", syscall.O_WRONLY|syscall.O_SYNC, 0400)
if unexf_err != nil {
log.Panicln("Error opening /sys/class/gpio/unexport:", unexf_err)
}
_, unex_err := unexf.WriteString(fmt.Sprintf("%d", watcher.pin))
if unex_err != nil {
log.Panicln("Error writing to /sys/class/gpio/unexport:", unex_err)
}
unexf.Close()
watcher.wg.Done()
}
func (u *NetLinkTopoUpdater) start() {
logging.GetLogger().Debug("Start NetLink Topo Updater for container: %s", u.Container.ID)
s, err := nl.Subscribe(syscall.NETLINK_ROUTE, syscall.RTNLGRP_LINK)
if err != nil {
logging.GetLogger().Error("Failed to subscribe to netlink RTNLGRP_LINK messages: %s", err.Error())
return
}
u.nlSocket = s
fd := u.nlSocket.GetFd()
err = syscall.SetNonblock(fd, true)
if err != nil {
logging.GetLogger().Error("Failed to set the netlink fd as non-blocking: %s", err.Error())
return
}
epfd, e := syscall.EpollCreate1(0)
if e != nil {
logging.GetLogger().Error("Failed to set the netlink fd as non-blocking: %s", err.Error())
return
}
defer syscall.Close(epfd)
u.initialize()
event := syscall.EpollEvent{Events: syscall.EPOLLIN, Fd: int32(fd)}
if e = syscall.EpollCtl(epfd, syscall.EPOLL_CTL_ADD, fd, &event); e != nil {
logging.GetLogger().Error("Failed to set the netlink fd as non-blocking: %s", err.Error())
return
}
events := make([]syscall.EpollEvent, maxEpollEvents)
Loop:
for {
n, err := syscall.EpollWait(epfd, events[:], 1000)
if err != nil {
logging.GetLogger().Error("Failed to receive from netlink messages: %s", err.Error())
continue
}
if n == 0 {
select {
case <-u.doneChan:
logging.GetLogger().Debug("WHOU")
break Loop
default:
continue
}
}
msgs, err := s.Receive()
if err != nil {
logging.GetLogger().Error("Failed to receive from netlink messages: %s", err.Error())
time.Sleep(1 * time.Second)
continue
}
for _, msg := range msgs {
switch msg.Header.Type {
case syscall.RTM_NEWLINK:
ifmsg := nl.DeserializeIfInfomsg(msg.Data)
u.onLinkAdded(int(ifmsg.Index))
case syscall.RTM_DELLINK:
ifmsg := nl.DeserializeIfInfomsg(msg.Data)
u.onLinkDeleted(int(ifmsg.Index))
}
}
}
u.nlSocket.Close()
}
func (self HciDev) Request(opcode OpCode, params ...Parameter) (Parameters, error) {
req := make([]byte, 4)
req[0] = HCI_COMMAND_PKT
binary.LittleEndian.PutUint16(req[1:], uint16(opcode))
if pbuf, err := Parameters(params).MarshalBinary(); err != nil {
return nil, err
} else {
req[3] = uint8(len(pbuf))
req = append(req, pbuf...)
}
if filter, err := GetsockoptHciFilter(int(self), SOL_HCI, HCI_FILTER); err != nil {
return nil, err
} else {
defer SetsockoptHciFilter(int(self), SOL_HCI, HCI_FILTER, filter)
}
filter := &HciFilter{
Type_mask: 1 << HCI_EVENT_PKT,
Event_mask: FilterEventMask(EVT_CMD_STATUS, EVT_CMD_COMPLETE),
Opcode: opcode.Native(),
}
if opcode.Ogf() == OGF_LE_CTL {
filter.Event_mask = FilterEventMask(EVT_CMD_STATUS, EVT_CMD_COMPLETE, EVT_LE_META_EVENT)
}
if err := SetsockoptHciFilter(int(self), SOL_HCI, HCI_FILTER, filter); err != nil {
return nil, err
}
if n, err := self.Write(req); err != nil {
return nil, err
} else if n != len(req) {
return nil, fmt.Errorf("write incomplete")
}
if efd, err := syscall.EpollCreate1(syscall.EPOLL_CLOEXEC); err != nil {
return nil, err
} else {
defer syscall.Close(efd)
var events [1]syscall.EpollEvent
syscall.EpollCtl(efd, syscall.EPOLL_CTL_ADD, int(self), &syscall.EpollEvent{
Events: syscall.EPOLLIN,
Fd: int32(self),
})
capture := make([]byte, 0, 258)
buf := make([]byte, 258)
for {
if n, err := syscall.EpollWait(efd, events[:], 100); err != nil {
return nil, err
} else if n == 0 {
continue
}
// todo: operation timeout
if n, _, _, _, err := syscall.Recvmsg(int(self), buf, nil, syscall.MSG_DONTWAIT); err != nil {
if errno, ok := err.(syscall.Errno); ok && errno.Temporary() {
continue
}
return nil, err
} else if n == 0 {
continue
} else {
capture = append(capture, buf[:n]...)
}
if pkt, step := Parse(capture); step == 0 {
continue
} else if p, err := pkt.(EventPkt).Parse(); err != nil {
continue
} else {
switch ev := p.(type) {
case EvtCmdComplete:
if ev.OpCode == uint16(opcode) {
return opcode.Response(ev.Params)
}
case EvtCmdStatus:
if ev.OpCode == uint16(opcode) && ev.Status != 0 {
return nil, HciError(ev.Status)
}
case EvtLeMetaEvent:
// todo: OGF_LE_CTL opcode expects this
}
capture = capture[step:]
}
}
return nil, fmt.Errorf("should not reach")
}
}
func (u *NetLinkProbe) start(nsPath string) {
var context *common.NetNSContext
var err error
// Enter the network namespace if necessary
if nsPath != "" {
context, err = common.NewNetNsContext(nsPath)
if err != nil {
logging.GetLogger().Errorf("Failed to switch namespace: %s", err.Error())
return
}
}
// Both NewHandle and Subscribe need to done in the network namespace.
h, err := netlink.NewHandle(syscall.NETLINK_ROUTE)
if err != nil {
logging.GetLogger().Errorf("Failed to create netlink handle: %s", err.Error())
context.Close()
return
}
defer h.Delete()
s, err := nl.Subscribe(syscall.NETLINK_ROUTE, syscall.RTNLGRP_LINK, syscall.RTNLGRP_IPV4_IFADDR, syscall.RTNLGRP_IPV6_IFADDR)
if err != nil {
logging.GetLogger().Errorf("Failed to subscribe to netlink messages: %s", err.Error())
context.Close()
return
}
defer s.Close()
u.ethtool, err = ethtool.NewEthtool()
if err != nil {
logging.GetLogger().Errorf("Failed to create ethtool object: %s", err.Error())
context.Close()
return
}
defer u.ethtool.Close()
epfd, e := syscall.EpollCreate1(0)
if e != nil {
logging.GetLogger().Errorf("Failed to create epoll: %s", err.Error())
return
}
defer syscall.Close(epfd)
// Leave the network namespace
context.Close()
u.wg.Add(1)
defer u.wg.Done()
atomic.StoreInt64(&u.state, common.RunningState)
defer atomic.StoreInt64(&u.state, common.StoppedState)
u.netlink = h
u.initialize()
fd := s.GetFd()
err = syscall.SetNonblock(fd, true)
if err != nil {
logging.GetLogger().Errorf("Failed to set the netlink fd as non-blocking: %s", err.Error())
return
}
event := syscall.EpollEvent{Events: syscall.EPOLLIN, Fd: int32(fd)}
if err = syscall.EpollCtl(epfd, syscall.EPOLL_CTL_ADD, fd, &event); err != nil {
logging.GetLogger().Errorf("Failed to control epoll: %s", err.Error())
return
}
events := make([]syscall.EpollEvent, maxEpollEvents)
for atomic.LoadInt64(&u.state) == common.RunningState {
n, err := syscall.EpollWait(epfd, events[:], 1000)
if err != nil {
errno, ok := err.(syscall.Errno)
if ok && errno != syscall.EINTR {
logging.GetLogger().Errorf("Failed to receive from events from netlink: %s", err.Error())
}
continue
}
if n == 0 {
continue
}
msgs, err := s.Receive()
if err != nil {
if errno, ok := err.(syscall.Errno); !ok || !errno.Temporary() {
logging.GetLogger().Errorf("Failed to receive from netlink messages: %s", err.Error())
return
}
time.Sleep(1 * time.Second)
continue
}
for _, msg := range msgs {
switch msg.Header.Type {
case syscall.RTM_NEWLINK:
link, err := netlink.LinkDeserialize(msg.Data)
if err != nil {
logging.GetLogger().Warningf("Failed to deserialize netlink message: %s", err.Error())
continue
}
u.onLinkAdded(link)
case syscall.RTM_DELLINK:
link, err := netlink.LinkDeserialize(msg.Data)
if err != nil {
logging.GetLogger().Warningf("Failed to deserialize netlink message: %s", err.Error())
continue
}
u.onLinkDeleted(link)
case syscall.RTM_NEWADDR:
addr, family, ifindex, err := parseAddr(msg.Data)
if err != nil {
logging.GetLogger().Warningf("Failed to parse newlink message: %s", err.Error())
continue
}
u.onAddressAdded(addr, family, ifindex)
case syscall.RTM_DELADDR:
addr, family, ifindex, err := parseAddr(msg.Data)
if err != nil {
logging.GetLogger().Warningf("Failed to parse newlink message: %s", err.Error())
continue
}
u.onAddressDeleted(addr, family, ifindex)
}
}
}
}
