// 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])
}
func (p *pollster) StopWaiting(fd int, bits uint) {
events, already := p.events[fd]
if !already {
print("Epoll unexpected fd=", fd, "\n")
return
}
// If syscall.EPOLLONESHOT is not set, the wait
// is a repeating wait, so don't change it.
if events&syscall.EPOLLONESHOT == 0 {
return
}
// Disable the given bits.
// If we're still waiting for other events, modify the fd
// event in the kernel. Otherwise, delete it.
events &= ^uint32(bits)
if int32(events)&^syscall.EPOLLONESHOT != 0 {
var ev syscall.EpollEvent
ev.Fd = int32(fd)
ev.Events = events
if e := syscall.EpollCtl(p.epfd, syscall.EPOLL_CTL_MOD, fd, &ev); e != 0 {
print("Epoll modify fd=", fd, ": ", os.Errno(e).String(), "\n")
}
p.events[fd] = events
} else {
if e := syscall.EpollCtl(p.epfd, syscall.EPOLL_CTL_DEL, fd, nil); e != 0 {
print("Epoll delete fd=", fd, ": ", os.Errno(e).String(), "\n")
}
p.events[fd] = 0, false
}
}
func (p *pollster) StopWaiting(fd int, bits uint) {
// pollServer is locked.
events, already := p.events[fd]
if !already {
// The fd returned by the kernel may have been
// cancelled already; return silently.
return
}
// If syscall.EPOLLONESHOT is not set, the wait
// is a repeating wait, so don't change it.
if events&syscall.EPOLLONESHOT == 0 {
return
}
// Disable the given bits.
// If we're still waiting for other events, modify the fd
// event in the kernel. Otherwise, delete it.
events &= ^uint32(bits)
if int32(events)&^syscall.EPOLLONESHOT != 0 {
p.ctlEvent.Fd = int32(fd)
p.ctlEvent.Events = events
if err := syscall.EpollCtl(p.epfd, syscall.EPOLL_CTL_MOD, fd, &p.ctlEvent); err != nil {
print("Epoll modify fd=", fd, ": ", err.Error(), "\n")
}
p.events[fd] = events
} else {
if err := syscall.EpollCtl(p.epfd, syscall.EPOLL_CTL_DEL, fd, nil); err != nil {
print("Epoll delete fd=", fd, ": ", err.Error(), "\n")
}
delete(p.events, fd)
}
}
func (io *NetIOManager) ProxyNetAccept(serverinfo syscall.Sockaddr) (sa syscall.Sockaddr, err error) {
var clientfd, serverfd int
// accpet mongodb client connection request
clientfd, clientinfo, err := syscall.Accept(io.proxy_server_fd)
if err != nil {
goto ClientError
}
err = syscall.SetNonblock(clientfd, true)
if err != nil {
goto ClientCleanup
}
err = syscall.EpollCtl(io.epoll_fd, syscall.EPOLL_CTL_ADD, clientfd,
&syscall.EpollEvent{Events: syscall.EPOLLIN | syscall.EPOLLOUT |
syscall.EPOLLRDHUP, Fd: int32(clientfd)})
if err != nil {
goto ClientCleanup
}
// establish connection with mongodb server
serverfd, err = syscall.Socket(syscall.AF_INET, syscall.SOCK_STREAM,
syscall.IPPROTO_TCP)
if err != nil {
goto ServerError
}
err = syscall.Connect(serverfd, serverinfo)
if err != nil {
goto ServerCleanup
}
err = syscall.SetNonblock(serverfd, true)
if err != nil {
goto ServerCleanup
}
err = syscall.EpollCtl(io.epoll_fd, syscall.EPOLL_CTL_ADD, serverfd,
&syscall.EpollEvent{Events: syscall.EPOLLIN | syscall.EPOLLOUT |
syscall.EPOLLRDHUP, Fd: int32(serverfd)})
if err != nil {
goto ServerCleanup
}
// now proxy server becomes a bridge between client <-> server
add_sock_peer(io, clientfd, clientinfo, serverfd, serverinfo)
return clientinfo, nil
ServerCleanup:
syscall.Close(serverfd)
ServerError:
syscall.EpollCtl(io.epoll_fd, syscall.EPOLL_CTL_DEL, clientfd,
&syscall.EpollEvent{Events: syscall.EPOLLIN | syscall.EPOLLOUT |
syscall.EPOLLRDHUP, Fd: int32(clientfd)})
ClientCleanup:
syscall.Close(clientfd)
ClientError:
return nil, err
}
func (fd *FD) closeUnlocked() error {
// Caller MUST already hold the C lock. Take the both R and W
// locks, to exclude Read and Write operations from accessing
// a closed sysfd.
fd.r.cond.L.Lock()
fd.w.cond.L.Lock()
fd.closed = true
// ev is not used by EpollCtl/DEL. Just don't pass a nil
// pointer.
var ev syscall.EpollEvent
err := syscall.EpollCtl(epfd, syscall.EPOLL_CTL_DEL, fd.sysfd, &ev)
if err != nil {
log.Printf("poller: EpollCtl/DEL (fd=%d, sysfd=%d): %s",
fd.id, fd.sysfd, err.Error())
}
if fd.r.timer != nil {
fd.r.timer.Stop()
}
if fd.w.timer != nil {
fd.w.timer.Stop()
}
fdM.DelFD(fd.id)
err = syscall.Close(fd.sysfd)
// Wake up everybody waiting on the FD.
fd.r.cond.Broadcast()
fd.w.cond.Broadcast()
debugf("FD %03d: CL: close(sysfd=%d)", fd.id, fd.sysfd)
fd.w.cond.L.Unlock()
fd.r.cond.L.Unlock()
return err
}
func (p *pollster) AddFD(fd int, mode int, repeat bool) (bool, error) {
// pollServer is locked.
var already bool
p.ctlEvent.Fd = int32(fd)
p.ctlEvent.Events, already = p.events[fd]
if !repeat {
p.ctlEvent.Events |= syscall.EPOLLONESHOT
}
if mode == 'r' {
p.ctlEvent.Events |= readFlags
} else {
p.ctlEvent.Events |= writeFlags
}
var op int
if already {
op = syscall.EPOLL_CTL_MOD
} else {
op = syscall.EPOLL_CTL_ADD
}
if e := syscall.EpollCtl(p.epfd, op, fd, &p.ctlEvent); e != nil {
return false, os.NewSyscallError("epoll_ctl", e)
}
p.events[fd] = p.ctlEvent.Events
return false, nil
}
func (p *pollster) AddFD(fd int, mode int, repeat bool) os.Error {
var ev syscall.EpollEvent
var already bool
ev.Fd = int32(fd)
ev.Events, already = p.events[fd]
if !repeat {
ev.Events |= syscall.EPOLLONESHOT
}
if mode == 'r' {
ev.Events |= readFlags
} else {
ev.Events |= writeFlags
}
var op int
if already {
op = syscall.EPOLL_CTL_MOD
} else {
op = syscall.EPOLL_CTL_ADD
}
if e := syscall.EpollCtl(p.epfd, op, fd, &ev); e != 0 {
return os.NewSyscallError("epoll_ctl", e)
}
p.events[fd] = ev.Events
return nil
}
func (io *NetIOManager) ProxyNetListen(sa syscall.Sockaddr) error {
serverfd, err := syscall.Socket(syscall.AF_INET, syscall.SOCK_STREAM,
syscall.IPPROTO_TCP)
if err != nil {
goto Error
}
err = syscall.Bind(serverfd, sa)
if err != nil {
goto Cleanup
}
err = syscall.Listen(serverfd, io.max_backlog)
if err != nil {
goto Cleanup
}
err = syscall.EpollCtl(io.epoll_fd, syscall.EPOLL_CTL_ADD, serverfd,
&syscall.EpollEvent{Events: syscall.EPOLLIN, Fd: int32(serverfd)})
if err != nil {
goto Cleanup
}
io.proxy_server_fd = serverfd
return nil
Cleanup:
syscall.Close(serverfd)
Error:
return err
}
func registerInterrupt(pin *digitalPin, handler func(embd.DigitalPin)) error {
l := getEpollListenerInstance()
pinFd := int(pin.val.Fd())
l.mu.Lock()
defer l.mu.Unlock()
if _, ok := l.interruptablePins[pinFd]; ok {
return ErrorPinAlreadyRegistered
}
var event syscall.EpollEvent
event.Events = syscall.EPOLLIN | (syscall.EPOLLET & 0xffffffff) | syscall.EPOLLPRI
if err := syscall.SetNonblock(pinFd, true); err != nil {
return err
}
event.Fd = int32(pinFd)
if err := syscall.EpollCtl(l.fd, syscall.EPOLL_CTL_ADD, pinFd, &event); err != nil {
return err
}
l.interruptablePins[pinFd] = &interrupt{pin: pin, handler: handler}
return nil
}
func newFD(sysfd int) (*FD, error) {
// Set sysfd to non-blocking mode
err := syscall.SetNonblock(sysfd, true)
if err != nil {
debugf("FD xxx: NF: sysfd=%d, err=%v", sysfd, err)
return nil, err
}
// Initialize FD
fd := &FD{sysfd: sysfd}
fd.id = fdM.GetID()
fd.r.cond = sync.NewCond(&fd.r.mu)
fd.w.cond = sync.NewCond(&fd.w.mu)
// Add to Epoll set. We may imediatelly start receiving events
// after this. They will be dropped since the FD is not yet in
// fdMap. It's ok. Nobody is waiting on this FD yet, anyway.
ev := syscall.EpollEvent{
Events: syscall.EPOLLIN |
syscall.EPOLLOUT |
syscall.EPOLLRDHUP |
(syscall.EPOLLET & 0xffffffff),
Fd: int32(fd.id)}
err = syscall.EpollCtl(epfd, syscall.EPOLL_CTL_ADD, fd.sysfd, &ev)
if err != nil {
debugf("FD %03d: NF: sysfd=%d, err=%v", fd.id, fd.sysfd, err)
return nil, err
}
// Add to fdMap
fdM.AddFD(fd)
debugf("FD %03d: NF: sysfd=%d", fd.id, fd.sysfd)
return fd, nil
}
func (p *pin) BeginWatch(edge Edge, callback IRQEvent) error {
p.SetMode(ModeInput)
if err := write([]byte(edge), p.edgePath); err != nil {
return err
}
var event syscall.EpollEvent
event.Events = syscall.EPOLLIN | (syscall.EPOLLET & 0xffffffff) | syscall.EPOLLPRI
fd := int(p.valueFile.Fd())
p.callback = callback
watchEventCallbacks[fd] = p
if err := syscall.SetNonblock(fd, true); err != nil {
return err
}
event.Fd = int32(fd)
if err := syscall.EpollCtl(epollFD, syscall.EPOLL_CTL_ADD, fd, &event); err != nil {
return err
}
return nil
}
func (gpio *GPIO) DisableEdgeDetection() {
epollFd := gpio.epollFd.Swap(0)
if epollFd != 0 {
syscall.EpollCtl(epollFd, syscall.EPOLL_CTL_DEL, int(gpio.valueFile.Fd()), new(syscall.EpollEvent))
syscall.Close(epollFd)
}
gpio.setEdge(EDGE_NONE)
}
func sock_close(io *NetIOManager, fd int) {
syscall.EpollCtl(io.epoll_fd, syscall.EPOLL_CTL_DEL, fd,
&syscall.EpollEvent{Events: syscall.EPOLLIN | syscall.EPOLLOUT |
syscall.EPOLLRDHUP, Fd: int32(fd)})
syscall.Close(fd)
delete(io.pending_output_skbs, fd)
delete(io.io_socket_peers, fd)
}
func (ep *EPoll) AddFD(fd int, events uint32) error {
ev := syscall.EpollEvent{
Events: events,
Fd: int32(fd),
}
return syscall.EpollCtl(ep.fd, syscall.EPOLL_CTL_ADD, fd, &ev)
}
// Add registers a socket for sending and/or receiving. The caller can't
// access the socket directly after this. The send channel (if any) should be
// closed by the caller.
func (io *IO) Add(s *zmq.Socket, send <-chan Data, recv chan<- Data) (err error) {
fd, err := s.GetFd()
if err != nil {
return
}
w := newWorker()
io.lock.Lock()
io.workers[int32(fd)] = w
io.lock.Unlock()
defer func() {
if err != nil {
io.lock.Lock()
delete(io.workers, int32(fd))
io.lock.Unlock()
}
}()
e := &syscall.EpollEvent{
Events: syscall.EPOLLIN | syscall.EPOLLET&0xffffffff,
Fd: int32(fd),
}
if err = syscall.EpollCtl(io.epollFd, syscall.EPOLL_CTL_ADD, fd, e); err != nil {
return
}
state, err := s.GetEvents()
if err != nil {
syscall.EpollCtl(io.epollFd, syscall.EPOLL_CTL_DEL, fd, nil)
return
}
go func() {
defer s.Close()
defer syscall.EpollCtl(io.epollFd, syscall.EPOLL_CTL_DEL, fd, nil)
w.socketLoop(s, send, recv, state)
}()
return
}
// 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
}
func (gpio *GPIO) AddEdgeDetect(edge GPIOEdge) (chan bool, error) {
gpio.RemoveEdgeDetect()
err := gpio.SetDirection(GPIO_INPUT)
if err != nil {
return nil, err
}
err = gpio.SetEdge(edge)
if err != nil {
return nil, err
}
err = gpio.openValueFile()
if err != nil {
return nil, err
}
epfd, err := syscall.EpollCreate(1)
if err != nil {
return nil, err
}
event := &syscall.EpollEvent{
Events: syscall.EPOLLIN | _EPOLLET | syscall.EPOLLPRI,
Fd: int32(gpio.value.Fd()),
}
err = syscall.EpollCtl(epfd, syscall.EPOLL_CTL_ADD, int(gpio.value.Fd()), event)
if err != nil {
syscall.Close(epfd)
return nil, err
}
// / first time triggers with current state, so ignore
_, err = syscall.EpollWait(epfd, make([]syscall.EpollEvent, 1), -1)
if err != nil {
syscall.Close(epfd)
return nil, err
}
gpio.epfd.Set(epfd)
valueChan := make(chan bool)
go func() {
for gpio.epfd.Get() != 0 {
n, _ := syscall.EpollWait(epfd, make([]syscall.EpollEvent, 1), -1)
if n > 0 {
value, err := gpio.Value()
if err == nil {
valueChan <- value
}
}
}
}()
return valueChan, 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 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 (m *Monitor) start() {
var events [128]syscall.EpollEvent
for {
n, err := archutils.EpollWait(m.epollFd, events[:], -1)
if err != nil {
if err == syscall.EINTR {
continue
}
logrus.WithField("error", err).Fatal("containerd: epoll wait")
}
// process events
for i := 0; i < n; i++ {
fd := int(events[i].Fd)
m.m.Lock()
r := m.receivers[fd]
switch t := r.(type) {
case runtime.Process:
if events[i].Events == syscall.EPOLLHUP {
delete(m.receivers, fd)
if err = syscall.EpollCtl(m.epollFd, syscall.EPOLL_CTL_DEL, fd, &syscall.EpollEvent{
Events: syscall.EPOLLHUP,
Fd: int32(fd),
}); err != nil {
logrus.WithField("error", err).Error("containerd: epoll remove fd")
}
if err := t.Close(); err != nil {
logrus.WithField("error", err).Error("containerd: close process IO")
}
EpollFdCounter.Dec(1)
m.exits <- t
}
case runtime.OOM:
// always flush the event fd
t.Flush()
if t.Removed() {
delete(m.receivers, fd)
// epoll will remove the fd from its set after it has been closed
t.Close()
EpollFdCounter.Dec(1)
} else {
m.ooms <- t.ContainerID()
}
}
m.m.Unlock()
}
}
}
// EndWatch stops watching the pin
func (p *pin) EndWatch() error {
fd := int(p.valueFile.Fd())
if err := syscall.EpollCtl(epollFD, syscall.EPOLL_CTL_DEL, fd, nil); err != nil {
return err
}
if err := syscall.SetNonblock(fd, false); err != nil {
return err
}
delete(watchEventCallbacks, fd)
return nil
}
func (gpio *GPIO) WaitForEdge(edge Edge) (value Value, err error) {
if err = gpio.setEdge(edge); err != nil {
return 0, err
}
if err = gpio.ensureValueFileIsOpen(); err != nil {
return 0, err
}
epollFd := gpio.epollFd.Get()
if epollFd == 0 {
epollFd, err = syscall.EpollCreate(1)
if err != nil {
return 0, err
}
event := &syscall.EpollEvent{
Events: syscall.EPOLLIN | syscall.EPOLLPRI | _EPOLLET,
Fd: int32(gpio.valueFile.Fd()),
}
err = syscall.EpollCtl(epollFd, syscall.EPOLL_CTL_ADD, int(gpio.valueFile.Fd()), event)
if err != nil {
syscall.Close(epollFd)
return 0, err
}
// first time triggers with current state, so ignore
_, err = syscall.EpollWait(epollFd, dummyEpollEvents, -1)
if err != nil {
syscall.Close(epollFd)
return 0, err
}
gpio.epollFd.Set(epollFd)
}
_, err = syscall.EpollWait(epollFd, dummyEpollEvents, -1)
if err != nil {
return 0, err
}
return gpio.Value()
}
func init_inotify() {
var err error
name_by_wd = make(map[int32]string)
wd_by_name = make(map[string]int32)
var event syscall.InotifyEvent
event_size = int(unsafe.Sizeof(event))
inotify_fd, _ = syscall.InotifyInit()
if -1 == inotify_fd {
bump_message("InotifyInit failed, file changes outside of tabby " +
"will remain unnoticed")
return
}
epoll_fd, err = syscall.EpollCreate(1)
if -1 == epoll_fd {
tabby_log("init_inotify: " + err.Error())
}
var epoll_event syscall.EpollEvent
epoll_event.Events = syscall.EPOLLIN
syscall.EpollCtl(epoll_fd, syscall.EPOLL_CTL_ADD, inotify_fd, &epoll_event)
go inotify_observe()
}
func unregisterInterrupt(pin *digitalPin) error {
l := getEpollListenerInstance()
pinFd := int(pin.val.Fd())
l.mu.Lock()
defer l.mu.Unlock()
if _, ok := l.interruptablePins[pinFd]; !ok {
return nil
}
if err := syscall.EpollCtl(l.fd, syscall.EPOLL_CTL_DEL, pinFd, nil); err != nil {
return err
}
if err := syscall.SetNonblock(pinFd, false); err != nil {
return err
}
delete(l.interruptablePins, pinFd)
return 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
}
}
return err
}},
{
ErrFmt: "Bind epoll fd error [%s]\n",
Action: func(t *Tunnel) error {
return syscall.Bind(t.LFd, &syscall.SockaddrInet4{Port: int(t.EPort), Addr: [4]byte{0, 0, 0, 0}})
}},
{
ErrFmt: "Listen fd error [%s]\n",
Action: func(t *Tunnel) error {
return syscall.Listen(t.LFd, 10)
}},
{
ErrFmt: "Add fd to epoll error [%s]\n",
Action: func(t *Tunnel) error {
return syscall.EpollCtl(epollFd, syscall.EPOLL_CTL_ADD, t.LFd, &syscall.EpollEvent{Events: syscall.EPOLLIN, Fd: int32(t.LFd)})
}},
}
var pathAddStep = [...]TunnelStep{
{
ErrFmt: "Port [%d] accept error [%s]",
Action: func(tc *TunnelConn) (err error) {
tc.EFd, _, err = syscall.Accept(tc.RelTunnel.LFd)
return
}},
{
ErrFmt: "Port [%d] set fd nonblock error [%s]",
Action: func(tc *TunnelConn) (err error) {
return syscall.SetNonblock(tc.EFd, true)
}},
// EpollCtl directly calls syscall.EpollCtl
func EpollCtl(epfd int, op int, fd int, event *syscall.EpollEvent) error {
return syscall.EpollCtl(epfd, op, fd, event)
}
