ircd/engine_kqueue.c

   1 /*
   2  * IRC - Internet Relay Chat, ircd/engine_kqueue.c
   3  * Copyright (C) 2001 Kevin L. Mitchell <klmitch@mit.edu>
   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 FreeBSD kqueue()/kevent() event engine.
  21  * @version $Id: engine_kqueue.c,v 1.8 2005/03/23 00:25:16 entrope Exp $
  22  */
  23 #include "config.h"
  24
  25 #include "ircd_events.h"
  26
  27 #include "ircd.h"
  28 #include "ircd_alloc.h"
  29 #include "ircd_features.h"
  30 #include "ircd_log.h"
  31 #include "s_debug.h"
  32
  33 /* #include <assert.h> -- Now using assert in ircd_log.h */
  34 #include <errno.h>
  35 #include <signal.h>
  36 #include <sys/types.h>
  37 #include <sys/event.h>
  38 #include <sys/socket.h>
  39 #include <sys/time.h>
  40 #include <time.h>
  41 #include <unistd.h>
  42
  43 #define KQUEUE_ERROR_THRESHOLD  20      /**< after 20 kqueue errors, restart */
  44 #define ERROR_EXPIRE_TIME       3600    /**< expire errors after an hour */
  45
  46 /** Array of active Socket structures, indexed by file descriptor. */
  47 static struct Socket** sockList;
  48 /** Maximum file descriptor supported, plus one. */
  49 static int kqueue_max;
  50 /** File descriptor for kqueue pseudo-file. */
  51 static int kqueue_id;
  52
  53 /** Number of recent errors from kqueue. */
  54 static int errors = 0;
  55 /** Periodic timer to forget errors. */
  56 static struct Timer clear_error;
  57
  58 /** Decrement the error count (once per hour).
  59  * @param[in] ev Expired timer event (ignored).
  60  */
  61 static void
  62 error_clear(struct Event* ev)
  63 {
  64   if (!--errors) /* remove timer when error count reaches 0 */
  65     timer_del(ev_timer(ev));
  66 }
  67
  68 /** Initialize the kqueue engine.
  69  * @param[in] max_sockets Maximum number of file descriptors to support.
  70  * @return Non-zero on success, or zero on failure.
  71  */
  72 static int
  73 engine_init(int max_sockets)
  74 {
  75   int i;
  76
  77   if ((kqueue_id = kqueue()) < 0) { /* initialize... */
  78     log_write(LS_SYSTEM, L_WARNING, 0,
  79               "kqueue() engine cannot initialize: %m");
  80     return 0;
  81   }
  82
  83   /* allocate necessary memory */
  84   sockList = (struct Socket**) MyMalloc(sizeof(struct Socket*) * max_sockets);
  85
  86   /* initialize the data */
  87   for (i = 0; i < max_sockets; i++)
  88     sockList[i] = 0;
  89
  90   kqueue_max = max_sockets; /* number of sockets allocated */
  91
  92   return 1; /* success! */
  93 }
  94
  95 /** Add a signal to the event engine.
  96  * @param[in] sig Signal to add to engine.
  97  */
  98 static void
  99 engine_signal(struct Signal* sig)
 100 {
 101   struct kevent sigevent;
 102   struct sigaction act;
 103
 104   assert(0 != signal);
 105
 106   Debug((DEBUG_ENGINE, "kqueue: Adding filter for signal %d [%p]",
 107          sig_signal(sig), sig));
 108
 109   sigevent.ident = sig_signal(sig); /* set up the kqueue event */
 110   sigevent.filter = EVFILT_SIGNAL; /* looking for signals... */
 111   sigevent.flags = EV_ADD | EV_ENABLE; /* add and enable it */
 112   sigevent.fflags = 0;
 113   sigevent.data = 0;
 114   sigevent.udata = sig; /* store our user data */
 115
 116   if (kevent(kqueue_id, &sigevent, 1, 0, 0, 0) < 0) { /* add event */
 117     log_write(LS_SYSTEM, L_WARNING, 0, "Unable to trap signal %d",
 118               sig_signal(sig));
 119     return;
 120   }
 121
 122   act.sa_handler = SIG_IGN; /* ignore the signal */
 123   act.sa_flags = 0;
 124   sigemptyset(&act.sa_mask);
 125   sigaction(sig_signal(sig), &act, 0);
 126 }
 127
 128 /** Figure out what events go with a given state.
 129  * @param[in] state %Socket state to consider.
 130  * @param[in] events User-specified preferred event set.
 131  * @return Actual set of preferred events.
 132  */
 133 static unsigned int
 134 state_to_events(enum SocketState state, unsigned int events)
 135 {
 136   switch (state) {
 137   case SS_CONNECTING: /* connecting socket */
 138     return SOCK_EVENT_WRITABLE;
 139     break;
 140
 141   case SS_LISTENING: /* listening socket */
 142   case SS_NOTSOCK: /* our signal socket--just in case */
 143     return SOCK_EVENT_READABLE;
 144     break;
 145
 146   case SS_CONNECTED: case SS_DATAGRAM: case SS_CONNECTDG:
 147     return events; /* ordinary socket */
 148     break;
 149   }
 150
 151   /*NOTREACHED*/
 152   return 0;
 153 }
 154
 155 /** Activate kqueue filters as appropriate.
 156  * @param[in] sock Socket structure to operate on.
 157  * @param[in] clear Set of interest events to clear from socket.
 158  * @param[in] set Set of interest events to set on socket.
 159  */
 160 static void
 161 set_or_clear(struct Socket* sock, unsigned int clear, unsigned int set)
 162 {
 163   int i = 0;
 164   struct kevent chglist[2];
 165
 166   assert(0 != sock);
 167   assert(-1 < s_fd(sock));
 168
 169   if ((clear ^ set) & SOCK_EVENT_READABLE) { /* readable has changed */
 170     chglist[i].ident = s_fd(sock); /* set up the change list */
 171     chglist[i].filter = EVFILT_READ; /* readable filter */
 172     chglist[i].flags = EV_ADD; /* adding it */
 173     chglist[i].fflags = 0;
 174     chglist[i].data = 0;
 175     chglist[i].udata = 0; /* I love udata, but it can't really be used here */
 176
 177     if (set & SOCK_EVENT_READABLE) /* it's set */
 178       chglist[i].flags |= EV_ENABLE;
 179     else /* clear it */
 180       chglist[i].flags |= EV_DISABLE;
 181
 182     i++; /* advance to next element */
 183   }
 184
 185   if ((clear ^ set) & SOCK_EVENT_WRITABLE) { /* writable has changed */
 186     chglist[i].ident = s_fd(sock); /* set up the change list */
 187     chglist[i].filter = EVFILT_WRITE; /* writable filter */
 188     chglist[i].flags = EV_ADD; /* adding it */
 189     chglist[i].fflags = 0;
 190     chglist[i].data = 0;
 191     chglist[i].udata = 0;
 192
 193     if (set & SOCK_EVENT_WRITABLE) /* it's set */
 194       chglist[i].flags |= EV_ENABLE;
 195     else /* clear it */
 196       chglist[i].flags |= EV_DISABLE;
 197
 198     i++; /* advance count... */
 199   }
 200
 201   if (kevent(kqueue_id, chglist, i, 0, 0, 0) < 0 && errno != EBADF)
 202     event_generate(ET_ERROR, sock, errno); /* report error */
 203 }
 204
 205 /** Add a socket to the event engine.
 206  * @param[in] sock Socket to add to engine.
 207  * @return Non-zero on success, or zero on error.
 208  */
 209 static int
 210 engine_add(struct Socket* sock)
 211 {
 212   assert(0 != sock);
 213   assert(0 == sockList[s_fd(sock)]);
 214
 215   /* bounds-check... */
 216   if (sock->s_fd >= kqueue_max) {
 217     log_write(LS_SYSTEM, L_ERROR, 0,
 218               "Attempt to add socket %d (> %d) to event engine", s_fd(sock),
 219               kqueue_max);
 220     return 0;
 221   }
 222
 223   sockList[s_fd(sock)] = sock; /* add to list */
 224
 225   Debug((DEBUG_ENGINE, "kqueue: Adding socket %d [%p], state %s, to engine",
 226          s_fd(sock), sock, state_to_name(s_state(sock))));
 227
 228   /* Add socket to queue */
 229   set_or_clear(sock, 0, state_to_events(s_state(sock), s_events(sock)));
 230
 231   return 1; /* success */
 232 }
 233
 234 /** Handle state transition for a socket.
 235  * @param[in] sock Socket changing state.
 236  * @param[in] new_state New state for socket.
 237  */
 238 static void
 239 engine_state(struct Socket* sock, enum SocketState new_state)
 240 {
 241   assert(0 != sock);
 242   assert(sock == sockList[s_fd(sock)]);
 243
 244   Debug((DEBUG_ENGINE, "kqueue: Changing state for socket %p to %s", sock,
 245          state_to_name(new_state)));
 246
 247   /* set the correct events */
 248   set_or_clear(sock,
 249                state_to_events(s_state(sock), s_events(sock)), /* old state */
 250                state_to_events(new_state, s_events(sock))); /* new state */
 251
 252 }
 253
 254 /** Handle change to preferred socket events.
 255  * @param[in] sock Socket getting new interest list.
 256  * @param[in] new_events New set of interesting events for socket.
 257  */
 258 static void
 259 engine_events(struct Socket* sock, unsigned int new_events)
 260 {
 261   assert(0 != sock);
 262   assert(sock == sockList[s_fd(sock)]);
 263
 264   Debug((DEBUG_ENGINE, "kqueue: Changing event mask for socket %p to [%s]",
 265          sock, sock_flags(new_events)));
 266
 267   /* set the correct events */
 268   set_or_clear(sock,
 269                state_to_events(s_state(sock), s_events(sock)), /* old events */
 270                state_to_events(s_state(sock), new_events)); /* new events */
 271 }
 272
 273 /** Remove a socket from the event engine.
 274  * @param[in] sock Socket being destroyed.
 275  */
 276 static void
 277 engine_delete(struct Socket* sock)
 278 {
 279   struct kevent dellist[2];
 280
 281   assert(0 != sock);
 282   assert(sock == sockList[s_fd(sock)]);
 283
 284   Debug((DEBUG_ENGINE, "kqueue: Deleting socket %d [%p], state %s",
 285          s_fd(sock), sock, state_to_name(s_state(sock))));
 286
 287   dellist[0].ident = s_fd(sock); /* set up the delete list */
 288   dellist[0].filter = EVFILT_READ; /* readable filter */
 289   dellist[0].flags = EV_DELETE; /* delete it */
 290   dellist[0].fflags = 0;
 291   dellist[0].data = 0;
 292   dellist[0].udata = 0;
 293
 294   dellist[1].ident = s_fd(sock);
 295   dellist[1].filter = EVFILT_WRITE; /* writable filter */
 296   dellist[1].flags = EV_DELETE; /* delete it */
 297   dellist[1].fflags = 0;
 298   dellist[1].data = 0;
 299   dellist[1].udata = 0;
 300
 301   sockList[s_fd(sock)] = 0;
 302 }
 303
 304 /** Run engine event loop.
 305  * @param[in] gen Lists of generators of various types.
 306  */
 307 static void
 308 engine_loop(struct Generators* gen)
 309 {
 310   struct kevent *events;
 311   int events_count;
 312   struct Socket* sock;
 313   struct timespec wait;
 314   int nevs;
 315   int i;
 316   int errcode;
 317   size_t codesize;
 318
 319   if ((events_count = feature_int(FEAT_POLLS_PER_LOOP)) < 20)
 320     events_count = 20;
 321   events = (struct kevent *)MyMalloc(sizeof(struct kevent) * events_count);
 322
 323   while (running) {
 324     if ((i = feature_int(FEAT_POLLS_PER_LOOP)) >= 20 && i != events_count) {
 325       events = (struct kevent *)MyRealloc(events, sizeof(struct kevent) * i);
 326       events_count = i;
 327     }
 328
 329     /* set up the sleep time */
 330     wait.tv_sec = timer_next(gen) ? (timer_next(gen) - CurrentTime) : -1;
 331     wait.tv_nsec = 0;
 332
 333     Debug((DEBUG_INFO, "kqueue: delay: %Tu (%Tu) %Tu", timer_next(gen),
 334            CurrentTime, wait.tv_sec));
 335
 336     /* check for active events */
 337     nevs = kevent(kqueue_id, 0, 0, events, events_count,
 338                   wait.tv_sec < 0 ? 0 : &wait);
 339
 340     CurrentTime = time(0); /* set current time... */
 341
 342     if (nevs < 0) {
 343       if (errno != EINTR) { /* ignore kevent interrupts */
 344         /* Log the kqueue error */
 345         log_write(LS_SOCKET, L_ERROR, 0, "kevent() error: %m");
 346         if (!errors++)
 347           timer_add(timer_init(&clear_error), error_clear, 0, TT_PERIODIC,
 348                     ERROR_EXPIRE_TIME);
 349         else if (errors > KQUEUE_ERROR_THRESHOLD) /* too many errors... */
 350           server_restart("too many kevent errors");
 351       }
 352       /* old code did a sleep(1) here; with usage these days,
 353        * that may be too expensive
 354        */
 355       continue;
 356     }
 357
 358     for (i = 0; i < nevs; i++) {
 359       if (events[i].filter == EVFILT_SIGNAL) {
 360         /* it's a signal; deal appropriately */
 361         event_generate(ET_SIGNAL, events[i].udata, events[i].ident);
 362         continue; /* skip socket processing loop */
 363       }
 364
 365       assert(events[i].filter == EVFILT_READ ||
 366              events[i].filter == EVFILT_WRITE);
 367
 368       sock = sockList[events[i].ident];
 369       if (!sock) /* slots may become empty while processing events */
 370         continue;
 371
 372       assert(s_fd(sock) == events[i].ident);
 373
 374       gen_ref_inc(sock); /* can't have it going away on us */
 375
 376       Debug((DEBUG_ENGINE, "kqueue: Checking socket %p (fd %d) state %s, "
 377              "events %s", sock, s_fd(sock), state_to_name(s_state(sock)),
 378              sock_flags(s_events(sock))));
 379
 380       if (s_state(sock) != SS_NOTSOCK) {
 381         errcode = 0; /* check for errors on socket */
 382         codesize = sizeof(errcode);
 383         if (getsockopt(s_fd(sock), SOL_SOCKET, SO_ERROR, &errcode,
 384                        &codesize) < 0)
 385           errcode = errno; /* work around Solaris implementation */
 386
 387         if (errcode) { /* an error occurred; generate an event */
 388           Debug((DEBUG_ENGINE, "kqueue: Error %d on fd %d, socket %p", errcode,
 389                  s_fd(sock), sock));
 390           event_generate(ET_ERROR, sock, errcode);
 391           gen_ref_dec(sock); /* careful not to leak reference counts */
 392           continue;
 393         }
 394       }
 395
 396       switch (s_state(sock)) {
 397       case SS_CONNECTING:
 398         if (events[i].filter == EVFILT_WRITE) { /* connection completed */
 399           Debug((DEBUG_ENGINE, "kqueue: Connection completed"));
 400           event_generate(ET_CONNECT, sock, 0);
 401         }
 402         break;
 403
 404       case SS_LISTENING:
 405         if (events[i].filter == EVFILT_READ) { /* connect. to be accept. */
 406           Debug((DEBUG_ENGINE, "kqueue: Ready for accept"));
 407           event_generate(ET_ACCEPT, sock, 0);
 408         }
 409         break;
 410
 411       case SS_NOTSOCK: /* doing nothing socket-specific */
 412       case SS_CONNECTED:
 413         if (events[i].filter == EVFILT_READ) { /* data on socket */
 414           Debug((DEBUG_ENGINE, "kqueue: EOF or data to be read"));
 415           event_generate(events[i].flags & EV_EOF ? ET_EOF : ET_READ, sock, 0);
 416         }
 417         if (events[i].filter == EVFILT_WRITE) { /* socket writable */
 418           Debug((DEBUG_ENGINE, "kqueue: Data can be written"));
 419           event_generate(ET_WRITE, sock, 0);
 420         }
 421         break;
 422
 423       case SS_DATAGRAM: case SS_CONNECTDG:
 424         if (events[i].filter == EVFILT_READ) { /* socket readable */
 425           Debug((DEBUG_ENGINE, "kqueue: Datagram to be read"));
 426           event_generate(ET_READ, sock, 0);
 427         }
 428         if (events[i].filter == EVFILT_WRITE) { /* socket writable */
 429           Debug((DEBUG_ENGINE, "kqueue: Datagram can be written"));
 430           event_generate(ET_WRITE, sock, 0);
 431         }
 432         break;
 433       }
 434
 435       gen_ref_dec(sock); /* we're done with it */
 436     }
 437
 438     timer_run(); /* execute any pending timers */
 439   }
 440 }
 441
 442 /** Descriptor for kqueue() event engine. */
 443 struct Engine engine_kqueue = {
 444   "kqueue()",           /* Engine name */
 445   engine_init,          /* Engine initialization function */
 446   engine_signal,        /* Engine signal registration function */
 447   engine_add,           /* Engine socket registration function */
 448   engine_state,         /* Engine socket state change function */
 449   engine_events,        /* Engine socket events mask function */
 450   engine_delete,        /* Engine socket deletion function */
 451   engine_loop           /* Core engine event loop */
 452 };