ircd/engine_epoll.c

   1 /*
   2  * IRC - Internet Relay Chat, ircd/engine_epoll.c
   3  * Copyright (C) 2003 Michael Poole <mdpoole@troilus.org>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; either version 1, or (at your option)
   8  * any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write to the Free Software
  17  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  18  */
  19 /** @file
  20  * @brief Linux epoll_*() event engine.
  21  * @version $Id: engine_epoll.c,v 1.12 2005/08/25 01:26:46 entrope Exp $
  22  */
  23 #include "config.h"
  24
  25 #include "ircd.h"
  26 #include "ircd_events.h"
  27 #include "ircd_alloc.h"
  28 #include "ircd_features.h"
  29 #include "ircd_log.h"
  30 #include "s_debug.h"
  31
  32 /* #include <assert.h> -- Now using assert in ircd_log.h */
  33 #include <errno.h>
  34 #include <sys/types.h>
  35 #ifdef HAVE_STDINT_H
  36 #include <stdint.h> /* bah */
  37 #endif
  38 #include <string.h>
  39 #include <sys/epoll.h>
  40 #include <sys/socket.h>
  41 #include <time.h>
  42 #include <linux/unistd.h>
  43
  44 /* The GNU C library may have a valid header but stub implementations
  45  * of the epoll system calls.  If so, provide our own. */
  46 #if defined(__stub_epoll_create) || defined(__stub___epoll_create) || defined(EPOLL_NEED_BODY)
  47
  48 /* Oh, did we mention that some glibc releases do not even define the
  49  * syscall numbers? */
  50 #if !defined(__NR_epoll_create)
  51 #if defined(__ia64__)
  52 #define __NR_epoll_create 1243
  53 #define __NR_epoll_ctl 1244
  54 #define __NR_epoll_wait 1245
  55 #elif defined(__x86_64__)
  56 #define __NR_epoll_create 214
  57 #define __NR_epoll_ctl 233
  58 #define __NR_epoll_wait 232
  59 #elif defined(__sparc64__) || defined(__sparc__)
  60 #define __NR_epoll_create 193
  61 #define __NR_epoll_ctl 194
  62 #define __NR_epoll_wait 195
  63 #elif defined(__s390__) || defined(__m68k__)
  64 #define __NR_epoll_create 249
  65 #define __NR_epoll_ctl 250
  66 #define __NR_epoll_wait 251
  67 #elif defined(__ppc64__) || defined(__ppc__)
  68 #define __NR_epoll_create 236
  69 #define __NR_epoll_ctl 237
  70 #define __NR_epoll_wait 238
  71 #elif defined(__parisc__) || defined(__arm26__) || defined(__arm__)
  72 #define __NR_epoll_create 224
  73 #define __NR_epoll_ctl 225
  74 #define __NR_epoll_wait 226
  75 #elif defined(__alpha__)
  76 #define __NR_epoll_create 407
  77 #define __NR_epoll_ctl 408
  78 #define __NR_epoll_wait 409
  79 #elif defined(__sh64__)
  80 #define __NR_epoll_create 282
  81 #define __NR_epoll_ctl 283
  82 #define __NR_epoll_wait 284
  83 #elif defined(__i386__) || defined(__sh__) || defined(__m32r__) || defined(__h8300__) || defined(__frv__)
  84 #define __NR_epoll_create 254
  85 #define __NR_epoll_ctl 255
  86 #define __NR_epoll_wait 256
  87 #else /* cpu types */
  88 #error No system call numbers defined for epoll family.
  89 #endif /* cpu types */
  90 #endif /* !defined(__NR_epoll_create) */
  91
  92 _syscall1(int, epoll_create, int, size)
  93 _syscall4(int, epoll_ctl, int, epfd, int, op, int, fd, struct epoll_event *, event)
  94 _syscall4(int, epoll_wait, int, epfd, struct epoll_event *, pevents, int, maxevents, int, timeout)
  95
  96 #endif /* epoll_create defined as stub */
  97
  98 #define EPOLL_ERROR_THRESHOLD 20   /**< after 20 epoll errors, restart */
  99 #define ERROR_EXPIRE_TIME     3600 /**< expire errors after an hour */
 100
 101 /** File descriptor for epoll pseudo-file. */
 102 static int epoll_fd;
 103 /** Number of recent epoll errors. */
 104 static int errors;
 105 /** Periodic timer to forget errors. */
 106 static struct Timer clear_error;
 107
 108 /** Decrement the error count (once per hour).
 109  * @param[in] ev Expired timer event (ignored).
 110  */
 111 static void
 112 error_clear(struct Event *ev)
 113 {
 114   if (!--errors)
 115     timer_del(ev_timer(ev));
 116 }
 117
 118 /** Initialize the epoll engine.
 119  * @param[in] max_sockets Maximum number of file descriptors to support.
 120  * @return Non-zero on success, or zero on failure.
 121  */
 122 static int
 123 engine_init(int max_sockets)
 124 {
 125   if ((epoll_fd = epoll_create(max_sockets)) < 0) {
 126     log_write(LS_SYSTEM, L_WARNING, 0,
 127               "epoll() engine cannot initialize: %m");
 128     return 0;
 129   }
 130   return 1;
 131 }
 132
 133 /** Set events for a particular socket.
 134  * @param[in] sock Socket to calculate events for.
 135  * @param[in] state Current socket state.
 136  * @param[in] events User-specified event interest list.
 137  * @param[out] evt epoll event structure for socket.
 138  */
 139 static void
 140 set_events(struct Socket *sock, enum SocketState state, unsigned int events, struct epoll_event *evt)
 141 {
 142   assert(0 != sock);
 143   assert(0 <= s_fd(sock));
 144   memset(evt, 0, sizeof(*evt));
 145
 146   evt->data.ptr = sock;
 147
 148   switch (state) {
 149   case SS_CONNECTING:
 150     evt->events = EPOLLOUT;
 151     break;
 152
 153   case SS_LISTENING:
 154   case SS_NOTSOCK:
 155     evt->events = EPOLLIN;
 156     break;
 157
 158   case SS_CONNECTED:
 159   case SS_DATAGRAM:
 160   case SS_CONNECTDG:
 161     switch (events & SOCK_EVENT_MASK) {
 162     case 0:
 163       evt->events = 0;
 164       break;
 165     case SOCK_EVENT_READABLE:
 166       evt->events = EPOLLIN;
 167       break;
 168     case SOCK_EVENT_WRITABLE:
 169       evt->events = EPOLLOUT;
 170       break;
 171     case SOCK_EVENT_READABLE|SOCK_EVENT_WRITABLE:
 172       evt->events = EPOLLIN|EPOLLOUT;
 173       break;
 174     }
 175     break;
 176   }
 177 }
 178
 179 /** Add a socket to the event engine.
 180  * @param[in] sock Socket to add to engine.
 181  * @return Non-zero on success, or zero on error.
 182  */
 183 static int
 184 engine_add(struct Socket *sock)
 185 {
 186   struct epoll_event evt;
 187
 188   assert(0 != sock);
 189   Debug((DEBUG_ENGINE, "epoll: Adding socket %d [%p], state %s, to engine",
 190          s_fd(sock), sock, state_to_name(s_state(sock))));
 191   set_events(sock, s_state(sock), s_events(sock), &evt);
 192   if (epoll_ctl(epoll_fd, EPOLL_CTL_ADD, s_fd(sock), &evt) < 0) {
 193     event_generate(ET_ERROR, sock, errno);
 194     return 0;
 195   }
 196   return 1;
 197 }
 198
 199 /** Handle state transition for a socket.
 200  * @param[in] sock Socket changing state.
 201  * @param[in] new_state New state for socket.
 202  */
 203 static void
 204 engine_set_state(struct Socket *sock, enum SocketState new_state)
 205 {
 206   struct epoll_event evt;
 207
 208   assert(0 != sock);
 209   Debug((DEBUG_ENGINE, "epoll: Changing state for socket %p to %s",
 210          sock, state_to_name(new_state)));
 211   set_events(sock, new_state, s_events(sock), &evt);
 212   if (epoll_ctl(epoll_fd, EPOLL_CTL_MOD, s_fd(sock), &evt) < 0)
 213     event_generate(ET_ERROR, sock, errno);
 214 }
 215
 216 /** Handle change to preferred socket events.
 217  * @param[in] sock Socket getting new interest list.
 218  * @param[in] new_events New set of interesting events for socket.
 219  */
 220 static void
 221 engine_set_events(struct Socket *sock, unsigned new_events)
 222 {
 223   struct epoll_event evt;
 224
 225   assert(0 != sock);
 226   Debug((DEBUG_ENGINE, "epoll: Changing event mask for socket %p to [%s]",
 227          sock, sock_flags(new_events)));
 228   set_events(sock, s_state(sock), new_events, &evt);
 229   if (epoll_ctl(epoll_fd, EPOLL_CTL_MOD, s_fd(sock), &evt) < 0)
 230     event_generate(ET_ERROR, sock, errno);
 231 }
 232
 233 /** Remove a socket from the event engine.
 234  * @param[in] sock Socket being destroyed.
 235  */
 236 static void
 237 engine_delete(struct Socket *sock)
 238 {
 239   assert(0 != sock);
 240   Debug((DEBUG_ENGINE, "epoll: Deleting socket %d [%p], state %s",
 241          s_fd(sock), sock, state_to_name(s_state(sock))));
 242   /* No action necessary; epoll removes the socket on close(). */
 243 }
 244
 245 /** Run engine event loop.
 246  * @param[in] gen Lists of generators of various types.
 247  */
 248 static void
 249 engine_loop(struct Generators *gen)
 250 {
 251   struct epoll_event *events;
 252   struct Socket *sock;
 253   socklen_t codesize;
 254   int events_count, i, wait, nevs, errcode;
 255
 256   if ((events_count = feature_int(FEAT_POLLS_PER_LOOP)) < 20)
 257     events_count = 20;
 258   events = MyMalloc(sizeof(events[0]) * events_count);
 259   while (running) {
 260     if ((i = feature_int(FEAT_POLLS_PER_LOOP)) >= 20 && i != events_count) {
 261       events = MyRealloc(events, sizeof(events[0]) * i);
 262       events_count = i;
 263     }
 264
 265     wait = timer_next(gen) ? (timer_next(gen) - CurrentTime) * 1000 : -1;
 266     Debug((DEBUG_ENGINE, "epoll: delay: %d (%d) %d", timer_next(gen),
 267            CurrentTime, wait));
 268     nevs = epoll_wait(epoll_fd, events, events_count, wait);
 269     CurrentTime = time(0);
 270
 271     if (nevs < 0) {
 272       if (errno != EINTR) {
 273         log_write(LS_SOCKET, L_ERROR, 0, "epoll() error: %m");
 274         if (!errors++)
 275           timer_add(timer_init(&clear_error), error_clear, 0, TT_PERIODIC,
 276                     ERROR_EXPIRE_TIME);
 277         else if (errors > EPOLL_ERROR_THRESHOLD)
 278           server_restart("too many epoll errors");
 279       }
 280       continue;
 281     }
 282
 283     for (i = 0; i < nevs; i++) {
 284       if (!(sock = events[i].data.ptr))
 285         continue;
 286       gen_ref_inc(sock);
 287       Debug((DEBUG_ENGINE,
 288              "epoll: Checking socket %p (fd %d) state %s, events %s",
 289              sock, s_fd(sock), state_to_name(s_state(sock)),
 290              sock_flags(s_events(sock))));
 291
 292       if (events[i].events & EPOLLERR) {
 293         errcode = 0;
 294         codesize = sizeof(errcode);
 295         if (getsockopt(s_fd(sock), SOL_SOCKET, SO_ERROR, &errcode,
 296                        &codesize) < 0)
 297           errcode = errno;
 298         if (errcode) {
 299           event_generate(ET_ERROR, sock, errcode);
 300           gen_ref_dec(sock);
 301           continue;
 302         }
 303       } else if (events[i].events & EPOLLHUP) {
 304         event_generate(ET_EOF, sock, 0);
 305       } else switch (s_state(sock)) {
 306       case SS_CONNECTING:
 307         if (events[i].events & EPOLLOUT) /* connection completed */
 308           event_generate(ET_CONNECT, sock, 0);
 309         break;
 310
 311       case SS_LISTENING:
 312         if (events[i].events & EPOLLIN) /* incoming connection */
 313           event_generate(ET_ACCEPT, sock, 0);
 314         break;
 315
 316       case SS_NOTSOCK:
 317       case SS_CONNECTED:
 318       case SS_DATAGRAM:
 319       case SS_CONNECTDG:
 320         if (events[i].events & EPOLLIN)
 321           event_generate(ET_READ, sock, 0);
 322         if (events[i].events & EPOLLOUT)
 323           event_generate(ET_WRITE, sock, 0);
 324         break;
 325       }
 326       gen_ref_dec(sock);
 327     }
 328     timer_run();
 329   }
 330   MyFree(events);
 331 }
 332
 333 /** Descriptor for epoll event engine. */
 334 struct Engine engine_epoll = {
 335   "epoll()",
 336   engine_init,
 337   0,
 338   engine_add,
 339   engine_set_state,
 340   engine_set_events,
 341   engine_delete,
 342   engine_loop
 343 };