iocore/net/UnixNet.cc

/** @file

  A brief file description

  @section license License

  Licensed to the Apache Software Foundation (ASF) under one
  or more contributor license agreements.  See the NOTICE file
  distributed with this work for additional information
  regarding copyright ownership.  The ASF licenses this file
  to you under the Apache License, Version 2.0 (the
  "License"); you may not use this file except in compliance
  with the License.  You may obtain a copy of the License at

      http://www.apache.org/licenses/LICENSE-2.0

  Unless required by applicable law or agreed to in writing, software
  distributed under the License is distributed on an "AS IS" BASIS,
  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  See the License for the specific language governing permissions and
  limitations under the License.
 */

#include "P_Net.h"

ink_hrtime last_throttle_warning;
ink_hrtime last_shedding_warning;
ink_hrtime emergency_throttle_time;
int net_connections_throttle;
int fds_throttle;
bool throttle_enabled;
int fds_limit = 8000;
ink_hrtime last_transient_accept_error;

extern "C" void fd_reify(struct ev_loop *);


#ifndef INACTIVITY_TIMEOUT
// INKqa10496
// One Inactivity cop runs on each thread once every second and
// loops through the list of NetVCs and calls the timeouts
struct InactivityCop : public Continuation {
  InactivityCop(ProxyMutex *m):Continuation(m) {
    SET_HANDLER(&InactivityCop::check_inactivity);
  }
  int check_inactivity(int event, Event *e) {
    (void) event;
    ink_hrtime now = ink_get_hrtime();
    NetHandler *nh = get_NetHandler(this_ethread());
    // Copy the list and use pop() to catch any closes caused by callbacks.
    forl_LL(UnixNetVConnection, vc, nh->open_list)
      nh->cop_list.push(vc);
    while (UnixNetVConnection *vc = nh->cop_list.pop()) {
      if (vc->closed) {
        close_UnixNetVConnection(vc, e->ethread);
        continue;
      } 
      if (vc->next_inactivity_timeout_at && vc->next_inactivity_timeout_at < now)
        vc->handleEvent(EVENT_IMMEDIATE, e);
    }
    return 0;
  }
};
#endif

PollCont::PollCont(ProxyMutex *m, int pt):Continuation(m), net_handler(NULL), poll_timeout(pt) {
  pollDescriptor = NEW(new PollDescriptor);
  pollDescriptor->init();
  SET_HANDLER(&PollCont::pollEvent);
}

PollCont::PollCont(ProxyMutex *m, NetHandler *nh, int pt):Continuation(m), net_handler(nh), poll_timeout(pt)
{
  pollDescriptor = NEW(new PollDescriptor);
  pollDescriptor->init();
  SET_HANDLER(&PollCont::pollEvent);
}

PollCont::~PollCont() {
  delete pollDescriptor;
}

//
// PollCont continuation which does the epoll_wait
// and stores the resultant events in ePoll_Triggered_Events
//
int
PollCont::pollEvent(int event, Event *e) {
  (void) event;
  (void) e;

  if (likely(net_handler)) {
    /* checking to see whether there are connections on the ready_queue (either read or write) that need processing [ebalsa] */
    if (likely
        (!net_handler->read_ready_list.empty() || !net_handler->read_ready_list.empty() ||
         !net_handler->read_enable_list.empty() || !net_handler->write_enable_list.empty())) {
      NetDebug("iocore_net_poll", "rrq: %d, wrq: %d, rel: %d, wel: %d", net_handler->read_ready_list.empty(),
               net_handler->write_ready_list.empty(), net_handler->read_enable_list.empty(),
               net_handler->write_enable_list.empty());
      poll_timeout = 0;         //poll immediately returns -- we have triggered stuff to process right now
    } else {
      poll_timeout = net_config_poll_timeout;
    }
  }
  // wait for fd's to tigger, or don't wait if timeout is 0
#if TS_USE_LIBEV
  struct ev_loop *eio = pollDescriptor->eio;
  double pt = (double)poll_timeout/1000.0;
  fd_reify(eio);
  eio->backend_poll(eio, pt);
  pollDescriptor->result = eio->pendingcnt[0];
  NetDebug("iocore_net_poll", "[PollCont::pollEvent] backend_poll(%d,%f), result=%d", eio->backend_fd,pt,pollDescriptor->result);
#elif TS_USE_EPOLL
  pollDescriptor->result = epoll_wait(pollDescriptor->epoll_fd,
                                      pollDescriptor->ePoll_Triggered_Events, POLL_DESCRIPTOR_SIZE, poll_timeout);
  NetDebug("iocore_net_poll", "[PollCont::pollEvent] epoll_fd: %d, timeout: %d, results: %d", pollDescriptor->epoll_fd, poll_timeout,
           pollDescriptor->result);
#elif TS_USE_KQUEUE
  struct timespec tv;
  tv.tv_sec = poll_timeout / 1000;
  tv.tv_nsec = 1000000 * (poll_timeout % 1000);
  pollDescriptor->result = kevent(pollDescriptor->kqueue_fd, NULL, 0,
                                  pollDescriptor->kq_Triggered_Events,
                                  POLL_DESCRIPTOR_SIZE,
                                  &tv);
  NetDebug("iocore_net_poll", "[PollCont::pollEvent] kueue_fd: %d, timeout: %d, results: %d", pollDescriptor->kqueue_fd, poll_timeout,
           pollDescriptor->result);
#elif TS_USE_PORT
  int retval;
  timespec_t ptimeout;
  ptimeout.tv_sec = poll_timeout / 1000;
  ptimeout.tv_nsec = 1000000 * (poll_timeout % 1000);
  unsigned nget = 1;
  if((retval = port_getn(pollDescriptor->port_fd,
                         pollDescriptor->Port_Triggered_Events,
                         POLL_DESCRIPTOR_SIZE, &nget, &ptimeout)) < 0) {
    pollDescriptor->result = 0;
    switch(errno) {
    case EINTR:
    case EAGAIN:
    case ETIME:
      if (nget > 0) {
        pollDescriptor->result = (int)nget;
      }
      break;
    default:
      ink_assert(!"unhandled port_getn() case:");
      break;
    }
  } else {
    pollDescriptor->result = (int)nget;
  }
  NetDebug("iocore_net_poll", "[PollCont::pollEvent] %d[%s]=port_getn(%d,%p,%d,%d,%d),results(%d)",
           retval,retval < 0 ? strerror(errno) : "ok",
           pollDescriptor->port_fd, pollDescriptor->Port_Triggered_Events,
           POLL_DESCRIPTOR_SIZE, nget, poll_timeout, pollDescriptor->result);
#else
#error port me
#endif
  return EVENT_CONT;
}

static void
net_signal_hook_callback(EThread *thread) {
#if TS_HAS_EVENTFD
  uint64_t counter;
  NOWARN_UNUSED_RETURN(read(thread->evfd, &counter, sizeof(uint64_t)));
#else
  char dummy[1024];
  NOWARN_UNUSED_RETURN(read(thread->evpipe[0], &dummy[0], 1024));
#endif
}

static void
net_signal_hook_function(EThread *thread) {
#if TS_HAS_EVENTFD
  uint64_t counter = 1;
  NOWARN_UNUSED_RETURN(write(thread->evfd, &counter, sizeof(uint64_t)));
#else
  char dummy = 1;
  NOWARN_UNUSED_RETURN(write(thread->evpipe[1], &dummy, 1));
#endif
}

void
initialize_thread_for_net(EThread *thread, int thread_index)
{
  NOWARN_UNUSED(thread_index);

  new((ink_dummy_for_new *) get_NetHandler(thread)) NetHandler();
  new((ink_dummy_for_new *) get_PollCont(thread)) PollCont(thread->mutex, get_NetHandler(thread));
  get_NetHandler(thread)->mutex = new_ProxyMutex();
  PollCont *pc = get_PollCont(thread);
  PollDescriptor *pd = pc->pollDescriptor;
#if TS_USE_LIBEV
  if (!thread_index)
    pd->eio = ev_default_loop(LIBEV_BACKEND_LIST);
  else
    pd->eio = ev_loop_new(LIBEV_BACKEND_LIST);
#endif
  thread->schedule_imm(get_NetHandler(thread));

#ifndef INACTIVITY_TIMEOUT
  InactivityCop *inactivityCop = NEW(new InactivityCop(get_NetHandler(thread)->mutex));
  thread->schedule_every(inactivityCop, HRTIME_SECONDS(1));
#endif

  thread->signal_hook = net_signal_hook_function;
  thread->ep = (EventIO*)malloc(sizeof(EventIO));
  thread->ep->type = EVENTIO_ASYNC_SIGNAL;
#if TS_HAS_EVENTFD
  thread->ep->start(pd, thread->evfd, 0, EVENTIO_READ);
#else
  thread->ep->start(pd, thread->evpipe[0], 0, EVENTIO_READ);
#endif
}

// NetHandler method definitions

NetHandler::NetHandler():Continuation(NULL), trigger_event(0)
{
  SET_HANDLER((NetContHandler) & NetHandler::startNetEvent);
}

//
// Initialization here, in the thread in which we will be executing
// from now on.
//
int
NetHandler::startNetEvent(int event, Event *e)
{
  (void) event;
  SET_HANDLER((NetContHandler) & NetHandler::mainNetEvent);
  e->schedule_every(NET_PERIOD);
  trigger_event = e;
  return EVENT_CONT;
}

//
// Move VC's enabled on a different thread to the ready list
//
void
NetHandler::process_enabled_list(NetHandler *nh, EThread *t)
{
  NOWARN_UNUSED(t);
  UnixNetVConnection *vc = NULL;

  SListM(UnixNetVConnection, NetState, read, enable_link) rq(nh->read_enable_list.popall());
  while ((vc = rq.pop())) {
    vc->ep.modify(EVENTIO_READ);
    vc->ep.refresh(EVENTIO_READ);
    vc->read.in_enabled_list = 0;
    if ((vc->read.enabled && vc->read.triggered) || vc->closed)
      nh->read_ready_list.in_or_enqueue(vc);
  }

  SListM(UnixNetVConnection, NetState, write, enable_link) wq(nh->write_enable_list.popall());
  while ((vc = wq.pop())) {
    vc->ep.modify(EVENTIO_WRITE);
    vc->ep.refresh(EVENTIO_WRITE);
    vc->write.in_enabled_list = 0;
    if ((vc->write.enabled && vc->write.triggered) || vc->closed)
      nh->write_ready_list.in_or_enqueue(vc);
  }
}


//
// The main event for NetHandler
// This is called every NET_PERIOD, and handles all IO operations scheduled
// for this period.
//
int
NetHandler::mainNetEvent(int event, Event *e)
{
  ink_assert(trigger_event == e && (event == EVENT_INTERVAL || event == EVENT_POLL));
  (void) event;
  (void) e;
  EventIO *epd = NULL;
  int poll_timeout = net_config_poll_timeout;

  NET_INCREMENT_DYN_STAT(net_handler_run_stat);

  process_enabled_list(this, e->ethread);
  if (likely(!read_ready_list.empty() || !write_ready_list.empty() || !read_enable_list.empty() || !write_enable_list.empty()))
    poll_timeout = 0; // poll immediately returns -- we have triggered stuff to process right now
  else
    poll_timeout = net_config_poll_timeout;

  PollDescriptor *pd = get_PollDescriptor(trigger_event->ethread);
  UnixNetVConnection *vc = NULL;
#if TS_USE_LIBEV
  struct ev_loop *eio = pd->eio;
  double pt = (double)poll_timeout/1000.0;
  fd_reify(eio);
  eio->backend_poll(eio, pt);
  pd->result = eio->pendingcnt[0];
  NetDebug("iocore_net_main_poll", "[NetHandler::mainNetEvent] backend_poll(%d,%f), result=%d", eio->backend_fd,pt,pd->result);
#elif TS_USE_EPOLL
  pd->result = epoll_wait(pd->epoll_fd, pd->ePoll_Triggered_Events, POLL_DESCRIPTOR_SIZE, poll_timeout);
  NetDebug("iocore_net_main_poll", "[NetHandler::mainNetEvent] epoll_wait(%d,%f), result=%d", pd->epoll_fd,poll_timeout,pd->result);
#elif TS_USE_KQUEUE
  struct timespec tv;
  tv.tv_sec = poll_timeout / 1000;
  tv.tv_nsec = 1000000 * (poll_timeout % 1000);
  pd->result = kevent(pd->kqueue_fd, NULL, 0, pd->kq_Triggered_Events, POLL_DESCRIPTOR_SIZE, &tv);
  NetDebug("iocore_net_main_poll", "[NetHandler::mainNetEvent] kevent(%d,%f), result=%d", pd->kqueue_fd,poll_timeout,pd->result);
#elif TS_USE_PORT
  int retval;
  timespec_t ptimeout;
  ptimeout.tv_sec = poll_timeout / 1000;
  ptimeout.tv_nsec = 1000000 * (poll_timeout % 1000);
  unsigned nget = 1;
  if((retval = port_getn(pd->port_fd, pd->Port_Triggered_Events, POLL_DESCRIPTOR_SIZE, &nget, &ptimeout)) < 0) {
    pd->result = 0;
    switch(errno) {
    case EINTR:
    case EAGAIN:
    case ETIME:
      if (nget > 0) {
        pd->result = (int)nget;
      }
      break;
    default:
      ink_assert(!"unhandled port_getn() case:");
      break;
    }
  } else {
    pd->result = (int)nget;
  }
  NetDebug("iocore_net_main_poll", "[NetHandler::mainNetEvent] %d[%s]=port_getn(%d,%p,%d,%d,%d),results(%d)",
           retval,retval < 0 ? strerror(errno) : "ok",
           pd->port_fd, pd->Port_Triggered_Events,
           POLL_DESCRIPTOR_SIZE, nget, poll_timeout, pd->result);

#else
#error port me
#endif

  vc = NULL;
  for (int x = 0; x < pd->result; x++) {
    epd = (EventIO*) get_ev_data(pd,x);
    if (epd->type == EVENTIO_READWRITE_VC) {
      vc = epd->data.vc;
      if (get_ev_events(pd,x) & (EVENTIO_READ|EVENTIO_ERROR)) {
        vc->read.triggered = 1;
        if (!read_ready_list.in(vc))
          read_ready_list.enqueue(vc);
        else if (get_ev_events(pd,x) & EVENTIO_ERROR) {
          // check for unhandled epoll events that should be handled
          Debug("iocore_net_main", "Unhandled epoll event on read: 0x%04x read.enabled=%d closed=%d read.netready_queue=%d",
                get_ev_events(pd,x), vc->read.enabled, vc->closed, read_ready_list.in(vc));
        }
      }
      vc = epd->data.vc;
      if (get_ev_events(pd,x) & (EVENTIO_WRITE|EVENTIO_ERROR)) {
        vc->write.triggered = 1;
        if (!write_ready_list.in(vc))
          write_ready_list.enqueue(vc);
        else if (get_ev_events(pd,x) & EVENTIO_ERROR) {
          // check for unhandled epoll events that should be handled
          Debug("iocore_net_main",
                "Unhandled epoll event on write: 0x%04x write.enabled=%d closed=%d write.netready_queue=%d",
                get_ev_events(pd,x), vc->write.enabled, vc->closed, write_ready_list.in(vc));
        }
      } else if (!get_ev_events(pd,x) & EVENTIO_ERROR) {
        Debug("iocore_net_main", "Unhandled epoll event: 0x%04x", get_ev_events(pd,x));
      }
    } else if (epd->type == EVENTIO_DNS_CONNECTION) {
      if (epd->data.dnscon != NULL) {
        epd->data.dnscon->trigger(); // Make sure the DNSHandler for this con knows we triggered
#if defined(USE_EDGE_TRIGGER)
        epd->refresh(EVENTIO_READ);
#endif
      }
    } else if (epd->type == EVENTIO_ASYNC_SIGNAL)
      net_signal_hook_callback(trigger_event->ethread);
    ev_next_event(pd,x);
  }

  pd->result = 0;

#if defined(USE_EDGE_TRIGGER)
 // UnixNetVConnection *
  while ((vc = read_ready_list.dequeue())) {
    if (vc->closed)
      close_UnixNetVConnection(vc, trigger_event->ethread);
    else if (vc->read.enabled && vc->read.triggered)
      vc->net_read_io(this, trigger_event->ethread);
    else if (!vc->read.enabled) {
      read_ready_list.remove(vc);
#if defined(solaris)
      if (vc->read.triggered && vc->write.enabled) {
        vc->ep.modify(-EVENTIO_READ);
        vc->ep.refresh(EVENTIO_WRITE);
        vc->writeReschedule(this);
      }
#endif
    }
  }
  while ((vc = write_ready_list.dequeue())) {
    if (vc->closed)
      close_UnixNetVConnection(vc, trigger_event->ethread);
    else if (vc->write.enabled && vc->write.triggered)
      write_to_net(this, vc, pd, trigger_event->ethread);
    else if (!vc->write.enabled) {
      write_ready_list.remove(vc);
#if defined(solaris)
      if (vc->write.triggered && vc->read.enabled) {
        vc->ep.modify(-EVENTIO_WRITE);
        vc->ep.refresh(EVENTIO_READ);
        vc->readReschedule(this);
      }
#endif
    }
  }
#else /* !USE_EDGE_TRIGGER */
  while ((vc = read_ready_list.dequeue())) {
    if (vc->closed)
      close_UnixNetVConnection(vc, trigger_event->ethread);
    else if (vc->read.enabled && vc->read.triggered)
      vc->net_read_io(this, trigger_event->ethread);
    else if (!vc->read.enabled)
      vc->ep.modify(-EVENTIO_READ);
  }
  while ((vc = write_ready_list.dequeue())) {
    if (vc->closed)
      close_UnixNetVConnection(vc, trigger_event->ethread);
    else if (vc->write.enabled && vc->write.triggered)
      write_to_net(this, vc, pd, trigger_event->ethread);
    else if (!vc->write.enabled)
      vc->ep.modify(-EVENTIO_WRITE);
  }
#endif /* !USE_EDGE_TRIGGER */

  return EVENT_CONT;
}