/*      $NetBSD: epoll.c,v 1.6 2024/08/18 20:47:21 christos Exp $       */

/*
* Copyright 2000-2007 Niels Provos <[email protected]>
* Copyright 2007-2012 Niels Provos, Nick Mathewson
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
*    notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
*    notice, this list of conditions and the following disclaimer in the
*    documentation and/or other materials provided with the distribution.
* 3. The name of the author may not be used to endorse or promote products
*    derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include "event2/event-config.h"
#include "evconfig-private.h"

#ifdef EVENT__HAVE_EPOLL

#include <stdint.h>
#include <sys/types.h>
#include <sys/resource.h>
#ifdef EVENT__HAVE_SYS_TIME_H
#include <sys/time.h>
#endif
#include <sys/queue.h>
#include <sys/epoll.h>
#include <signal.h>
#include <limits.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <errno.h>
#ifdef EVENT__HAVE_FCNTL_H
#include <fcntl.h>
#endif
#ifdef EVENT__HAVE_SYS_TIMERFD_H
#include <sys/timerfd.h>
#endif

#include "event-internal.h"
#include "evsignal-internal.h"
#include "event2/thread.h"
#include "evthread-internal.h"
#include "log-internal.h"
#include "evmap-internal.h"
#include "changelist-internal.h"
#include "time-internal.h"

/* Since Linux 2.6.17, epoll is able to report about peer half-closed connection
  using special EPOLLRDHUP flag on a read event.
*/
#if !defined(EPOLLRDHUP)
#define EPOLLRDHUP 0
#define EARLY_CLOSE_IF_HAVE_RDHUP 0
#else
#define EARLY_CLOSE_IF_HAVE_RDHUP EV_FEATURE_EARLY_CLOSE
#endif

#include "epolltable-internal.h"

#if defined(EVENT__HAVE_SYS_TIMERFD_H) &&                         \
       defined(EVENT__HAVE_TIMERFD_CREATE) &&                    \
       defined(HAVE_POSIX_MONOTONIC) && defined(TFD_NONBLOCK) && \
       defined(TFD_CLOEXEC)
/* Note that we only use timerfd if TFD_NONBLOCK and TFD_CLOEXEC are available
  and working.  This means that we can't support it on 2.6.25 (where timerfd
  was introduced) or 2.6.26, since 2.6.27 introduced those flags.
*/
#define USING_TIMERFD
#endif

struct epollop {
       struct epoll_event *events;
       int nevents;
       int epfd;
#ifdef USING_TIMERFD
       int timerfd;
#endif
};

static void *epoll_init(struct event_base *);
static int epoll_dispatch(struct event_base *, struct timeval *);
static void epoll_dealloc(struct event_base *);

static const struct eventop epollops_changelist = {
       "epoll (with changelist)",
       epoll_init,
       event_changelist_add_,
       event_changelist_del_,
       epoll_dispatch,
       epoll_dealloc,
       1, /* need reinit */
       EV_FEATURE_ET|EV_FEATURE_O1| EARLY_CLOSE_IF_HAVE_RDHUP,
       EVENT_CHANGELIST_FDINFO_SIZE
};


static int epoll_nochangelist_add(struct event_base *base, evutil_socket_t fd,
   short old, short events, void *p);
static int epoll_nochangelist_del(struct event_base *base, evutil_socket_t fd,
   short old, short events, void *p);

const struct eventop epollops = {
       "epoll",
       epoll_init,
       epoll_nochangelist_add,
       epoll_nochangelist_del,
       epoll_dispatch,
       epoll_dealloc,
       1, /* need reinit */
       EV_FEATURE_ET|EV_FEATURE_O1|EV_FEATURE_EARLY_CLOSE,
       0
};

#define INITIAL_NEVENT 32
#define MAX_NEVENT 4096

/* On Linux kernels at least up to 2.6.24.4, epoll can't handle timeout
* values bigger than (LONG_MAX - 999ULL)/HZ.  HZ in the wild can be
* as big as 1000, and LONG_MAX can be as small as (1<<31)-1, so the
* largest number of msec we can support here is 2147482.  Let's
* round that down by 47 seconds.
*/
#define MAX_EPOLL_TIMEOUT_MSEC (35*60*1000)

static void *
epoll_init(struct event_base *base)
{
       int epfd = -1;
       struct epollop *epollop;

#ifdef EVENT__HAVE_EPOLL_CREATE1
       /* First, try the shiny new epoll_create1 interface, if we have it. */
       epfd = epoll_create1(EPOLL_CLOEXEC);
#endif
       if (epfd == -1) {
               /* Initialize the kernel queue using the old interface.  (The
               size field is ignored   since 2.6.8.) */
               if ((epfd = epoll_create(32000)) == -1) {
                       if (errno != ENOSYS)
                               event_warn("epoll_create");
                       return (NULL);
               }
               evutil_make_socket_closeonexec(epfd);
       }

       if (!(epollop = mm_calloc(1, sizeof(struct epollop)))) {
               close(epfd);
               return (NULL);
       }

       epollop->epfd = epfd;

       /* Initialize fields */
       epollop->events = mm_calloc(INITIAL_NEVENT, sizeof(struct epoll_event));
       if (epollop->events == NULL) {
               mm_free(epollop);
               close(epfd);
               return (NULL);
       }
       epollop->nevents = INITIAL_NEVENT;

       if ((base->flags & EVENT_BASE_FLAG_EPOLL_USE_CHANGELIST) != 0 ||
           ((base->flags & EVENT_BASE_FLAG_IGNORE_ENV) == 0 &&
               evutil_getenv_("EVENT_EPOLL_USE_CHANGELIST") != NULL)) {

               base->evsel = &epollops_changelist;
       }

#ifdef USING_TIMERFD
       /*
         The epoll interface ordinarily gives us one-millisecond precision,
         so on Linux it makes perfect sense to use the CLOCK_MONOTONIC_COARSE
         timer.  But when the user has set the new PRECISE_TIMER flag for an
         event_base, we can try to use timerfd to give them finer granularity.
       */
       if ((base->flags & EVENT_BASE_FLAG_PRECISE_TIMER) &&
           base->monotonic_timer.monotonic_clock == CLOCK_MONOTONIC) {
               int fd;
               fd = epollop->timerfd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC);
               if (epollop->timerfd >= 0) {
                       struct epoll_event epev;
                       memset(&epev, 0, sizeof(epev));
                       epev.data.fd = epollop->timerfd;
                       epev.events = EPOLLIN;
                       if (epoll_ctl(epollop->epfd, EPOLL_CTL_ADD, fd, &epev) < 0) {
                               event_warn("epoll_ctl(timerfd)");
                               close(fd);
                               epollop->timerfd = -1;
                       }
               } else {
                       if (errno != EINVAL && errno != ENOSYS) {
                               /* These errors probably mean that we were
                                * compiled with timerfd/TFD_* support, but
                                * we're running on a kernel that lacks those.
                                */
                               event_warn("timerfd_create");
                       }
                       epollop->timerfd = -1;
               }
       } else {
               epollop->timerfd = -1;
       }
#endif

       evsig_init_(base);

       return (epollop);
}

static const char *
change_to_string(int change)
{
       change &= (EV_CHANGE_ADD|EV_CHANGE_DEL);
       if (change == EV_CHANGE_ADD) {
               return "add";
       } else if (change == EV_CHANGE_DEL) {
               return "del";
       } else if (change == 0) {
               return "none";
       } else {
               return "???";
       }
}

static const char *
epoll_op_to_string(int op)
{
       return op == EPOLL_CTL_ADD?"ADD":
           op == EPOLL_CTL_DEL?"DEL":
           op == EPOLL_CTL_MOD?"MOD":
           "???";
}

#define PRINT_CHANGES(op, events, ch, status)  \
       "Epoll %s(%d) on fd %d " status ". "       \
       "Old events were %d; "                     \
       "read change was %d (%s); "                \
       "write change was %d (%s); "               \
       "close change was %d (%s)",                \
       epoll_op_to_string(op),                    \
       events,                                    \
       ch->fd,                                    \
       ch->old_events,                            \
       ch->read_change,                           \
       change_to_string(ch->read_change),         \
       ch->write_change,                          \
       change_to_string(ch->write_change),        \
       ch->close_change,                          \
       change_to_string(ch->close_change)

static int
epoll_apply_one_change(struct event_base *base,
   struct epollop *epollop,
   const struct event_change *ch)
{
       struct epoll_event epev;
       int op, events = 0;
       int idx;

       idx = EPOLL_OP_TABLE_INDEX(ch);
       op = epoll_op_table[idx].op;
       events = epoll_op_table[idx].events;

       if (!events) {
               EVUTIL_ASSERT(op == 0);
               return 0;
       }

       if ((ch->read_change|ch->write_change|ch->close_change) & EV_CHANGE_ET)
               events |= EPOLLET;

       memset(&epev, 0, sizeof(epev));
       epev.data.fd = ch->fd;
       epev.events = events;
       if (epoll_ctl(epollop->epfd, op, ch->fd, &epev) == 0) {
               event_debug((PRINT_CHANGES(op, epev.events, ch, "okay")));
               return 0;
       }

       switch (op) {
       case EPOLL_CTL_MOD:
               if (errno == ENOENT) {
                       /* If a MOD operation fails with ENOENT, the
                        * fd was probably closed and re-opened.  We
                        * should retry the operation as an ADD.
                        */
                       if (epoll_ctl(epollop->epfd, EPOLL_CTL_ADD, ch->fd, &epev) == -1) {
                               event_warn("Epoll MOD(%d) on %d retried as ADD; that failed too",
                                   (int)epev.events, ch->fd);
                               return -1;
                       } else {
                               event_debug(("Epoll MOD(%d) on %d retried as ADD; succeeded.",
                                       (int)epev.events,
                                       ch->fd));
                               return 0;
                       }
               }
               break;
       case EPOLL_CTL_ADD:
               if (errno == EEXIST) {
                       /* If an ADD operation fails with EEXIST,
                        * either the operation was redundant (as with a
                        * precautionary add), or we ran into a fun
                        * kernel bug where using dup*() to duplicate the
                        * same file into the same fd gives you the same epitem
                        * rather than a fresh one.  For the second case,
                        * we must retry with MOD. */
                       if (epoll_ctl(epollop->epfd, EPOLL_CTL_MOD, ch->fd, &epev) == -1) {
                               event_warn("Epoll ADD(%d) on %d retried as MOD; that failed too",
                                   (int)epev.events, ch->fd);
                               return -1;
                       } else {
                               event_debug(("Epoll ADD(%d) on %d retried as MOD; succeeded.",
                                       (int)epev.events,
                                       ch->fd));
                               return 0;
                       }
               }
               break;
       case EPOLL_CTL_DEL:
               if (errno == ENOENT || errno == EBADF || errno == EPERM) {
                       /* If a delete fails with one of these errors,
                        * that's fine too: we closed the fd before we
                        * got around to calling epoll_dispatch. */
                       event_debug(("Epoll DEL(%d) on fd %d gave %s: DEL was unnecessary.",
                               (int)epev.events,
                               ch->fd,
                               strerror(errno)));
                       return 0;
               }
               break;
       default:
               break;
       }

       event_warn(PRINT_CHANGES(op, epev.events, ch, "failed"));
       return -1;
}

static int
epoll_apply_changes(struct event_base *base)
{
       struct event_changelist *changelist = &base->changelist;
       struct epollop *epollop = base->evbase;
       struct event_change *ch;

       int r = 0;
       int i;

       for (i = 0; i < changelist->n_changes; ++i) {
               ch = &changelist->changes[i];
               if (epoll_apply_one_change(base, epollop, ch) < 0)
                       r = -1;
       }

       return (r);
}

static int
epoll_nochangelist_add(struct event_base *base, evutil_socket_t fd,
   short old, short events, void *p)
{
       struct event_change ch;
       ch.fd = fd;
       ch.old_events = old;
       ch.read_change = ch.write_change = ch.close_change = 0;
       if (events & EV_WRITE)
               ch.write_change = EV_CHANGE_ADD |
                   (events & EV_ET);
       if (events & EV_READ)
               ch.read_change = EV_CHANGE_ADD |
                   (events & EV_ET);
       if (events & EV_CLOSED)
               ch.close_change = EV_CHANGE_ADD |
                   (events & EV_ET);

       return epoll_apply_one_change(base, base->evbase, &ch);
}

static int
epoll_nochangelist_del(struct event_base *base, evutil_socket_t fd,
   short old, short events, void *p)
{
       struct event_change ch;
       ch.fd = fd;
       ch.old_events = old;
       ch.read_change = ch.write_change = ch.close_change = 0;
       if (events & EV_WRITE)
               ch.write_change = EV_CHANGE_DEL |
                   (events & EV_ET);
       if (events & EV_READ)
               ch.read_change = EV_CHANGE_DEL |
                   (events & EV_ET);
       if (events & EV_CLOSED)
               ch.close_change = EV_CHANGE_DEL |
                   (events & EV_ET);

       return epoll_apply_one_change(base, base->evbase, &ch);
}

static int
epoll_dispatch(struct event_base *base, struct timeval *tv)
{
       struct epollop *epollop = base->evbase;
       struct epoll_event *events = epollop->events;
       int i, res;
       long timeout = -1;

#ifdef USING_TIMERFD
       if (epollop->timerfd >= 0) {
               struct itimerspec is;
               is.it_interval.tv_sec = 0;
               is.it_interval.tv_nsec = 0;
               if (tv == NULL) {
                       /* No timeout; disarm the timer. */
                       is.it_value.tv_sec = 0;
                       is.it_value.tv_nsec = 0;
               } else {
                       if (tv->tv_sec == 0 && tv->tv_usec == 0) {
                               /* we need to exit immediately; timerfd can't
                                * do that. */
                               timeout = 0;
                       }
                       is.it_value.tv_sec = tv->tv_sec;
                       is.it_value.tv_nsec = tv->tv_usec * 1000;
               }
               /* TODO: we could avoid unnecessary syscalls here by only
                  calling timerfd_settime when the top timeout changes, or
                  when we're called with a different timeval.
               */
               if (timerfd_settime(epollop->timerfd, 0, &is, NULL) < 0) {
                       event_warn("timerfd_settime");
               }
       } else
#endif
       if (tv != NULL) {
               timeout = evutil_tv_to_msec_(tv);
               if (timeout < 0 || timeout > MAX_EPOLL_TIMEOUT_MSEC) {
                       /* Linux kernels can wait forever if the timeout is
                        * too big; see comment on MAX_EPOLL_TIMEOUT_MSEC. */
                       timeout = MAX_EPOLL_TIMEOUT_MSEC;
               }
       }

       epoll_apply_changes(base);
       event_changelist_remove_all_(&base->changelist, base);

       EVBASE_RELEASE_LOCK(base, th_base_lock);

       res = epoll_wait(epollop->epfd, events, epollop->nevents, timeout);

       EVBASE_ACQUIRE_LOCK(base, th_base_lock);

       if (res == -1) {
               if (errno != EINTR) {
                       event_warn("epoll_wait");
                       return (-1);
               }

               return (0);
       }

       event_debug(("%s: epoll_wait reports %d", __func__, res));
       EVUTIL_ASSERT(res <= epollop->nevents);

       for (i = 0; i < res; i++) {
               int what = events[i].events;
               short ev = 0;
#ifdef USING_TIMERFD
               if (events[i].data.fd == epollop->timerfd)
                       continue;
#endif

               if (what & EPOLLERR) {
                       ev = EV_READ | EV_WRITE;
               } else if ((what & EPOLLHUP) && !(what & EPOLLRDHUP)) {
                       ev = EV_READ | EV_WRITE;
               } else {
                       if (what & EPOLLIN)
                               ev |= EV_READ;
                       if (what & EPOLLOUT)
                               ev |= EV_WRITE;
                       if (what & EPOLLRDHUP)
                               ev |= EV_CLOSED;
               }

               if (!ev)
                       continue;

               evmap_io_active_(base, events[i].data.fd, ev | EV_ET);
       }

       if (res == epollop->nevents && epollop->nevents < MAX_NEVENT) {
               /* We used all of the event space this time.  We should
                  be ready for more events next time. */
               int new_nevents = epollop->nevents * 2;
               struct epoll_event *new_events;

               new_events = mm_realloc(epollop->events,
                   new_nevents * sizeof(struct epoll_event));
               if (new_events) {
                       epollop->events = new_events;
                       epollop->nevents = new_nevents;
               }
       }

       return (0);
}


static void
epoll_dealloc(struct event_base *base)
{
       struct epollop *epollop = base->evbase;

       evsig_dealloc_(base);
       if (epollop->events)
               mm_free(epollop->events);
       if (epollop->epfd >= 0)
               close(epollop->epfd);
#ifdef USING_TIMERFD
       if (epollop->timerfd >= 0)
               close(epollop->timerfd);
#endif

       memset(epollop, 0, sizeof(struct epollop));
       mm_free(epollop);
}

#endif /* EVENT__HAVE_EPOLL */

You are viewing proxied material from ftp.icm.edu.pl. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.