1 /* $OpenBSD: kqueue.c,v 1.5 2002/07/10 14:41:31 art Exp $ */ 2 3 /* 4 * Copyright 2000-2007 Niels Provos <[email protected]> 5 * Copyright 2007-2010 Niels Provos and Nick Mathewson 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. The name of the author may not be used to endorse or promote products 16 * derived from this software without specific prior written permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 */ 29 #include "event2/event-config.h" 30 31 #define _GNU_SOURCE 32 33 #include <sys/types.h> 34 #ifdef _EVENT_HAVE_SYS_TIME_H 35 #include <sys/time.h> 36 #endif 37 #include <sys/queue.h> 38 #include <sys/event.h> 39 #include <signal.h> 40 #include <stdio.h> 41 #include <stdlib.h> 42 #include <string.h> 43 #include <unistd.h> 44 #include <errno.h> 45 #ifdef _EVENT_HAVE_INTTYPES_H 46 #include <inttypes.h> 47 #endif 48 49 /* Some platforms apparently define the udata field of struct kevent as 50 * intptr_t, whereas others define it as void*. There doesn't seem to be an 51 * easy way to tell them apart via autoconf, so we need to use OS macros. */ 52 #if defined(_EVENT_HAVE_INTTYPES_H) && !defined(__OpenBSD__) && !defined(__FreeBSD__) && !defined(__darwin__) && !defined(__APPLE__) 53 #define PTR_TO_UDATA(x) ((intptr_t)(x)) 54 #else 55 #define PTR_TO_UDATA(x) (x) 56 #endif 57 58 #include "event-internal.h" 59 #include "log-internal.h" 60 #include "evmap-internal.h" 61 #include "event2/thread.h" 62 #include "evthread-internal.h" 63 #include "changelist-internal.h" 64 65 #define NEVENT 64 66 67 struct kqop { 68 struct kevent *changes; 69 int changes_size; 70 71 struct kevent *events; 72 int events_size; 73 int kq; 74 pid_t pid; 75 }; 76 77 static void kqop_free(struct kqop *kqop); 78 79 static void *kq_init(struct event_base *); 80 static int kq_sig_add(struct event_base *, int, short, short, void *); 81 static int kq_sig_del(struct event_base *, int, short, short, void *); 82 static int kq_dispatch(struct event_base *, struct timeval *); 83 static void kq_dealloc(struct event_base *); 84 85 const struct eventop kqops = { 86 "kqueue", 87 kq_init, 88 event_changelist_add, 89 event_changelist_del, 90 kq_dispatch, 91 kq_dealloc, 92 1 /* need reinit */, 93 EV_FEATURE_ET|EV_FEATURE_O1|EV_FEATURE_FDS, 94 EVENT_CHANGELIST_FDINFO_SIZE 95 }; 96 97 static const struct eventop kqsigops = { 98 "kqueue_signal", 99 NULL, 100 kq_sig_add, 101 kq_sig_del, 102 NULL, 103 NULL, 104 1 /* need reinit */, 105 0, 106 0 107 }; 108 109 static void * 110 kq_init(struct event_base *base) 111 { 112 int kq = -1; 113 struct kqop *kqueueop = NULL; 114 115 if (!(kqueueop = mm_calloc(1, sizeof(struct kqop)))) 116 return (NULL); 117 118 /* Initialize the kernel queue */ 119 120 if ((kq = kqueue()) == -1) { 121 event_warn("kqueue"); 122 goto err; 123 } 124 125 kqueueop->kq = kq; 126 127 kqueueop->pid = getpid(); 128 129 /* Initialize fields */ 130 kqueueop->changes = mm_calloc(NEVENT, sizeof(struct kevent)); 131 if (kqueueop->changes == NULL) 132 goto err; 133 kqueueop->events = mm_calloc(NEVENT, sizeof(struct kevent)); 134 if (kqueueop->events == NULL) 135 goto err; 136 kqueueop->events_size = kqueueop->changes_size = NEVENT; 137 138 /* Check for Mac OS X kqueue bug. */ 139 memset(&kqueueop->changes[0], 0, sizeof kqueueop->changes[0]); 140 kqueueop->changes[0].ident = -1; 141 kqueueop->changes[0].filter = EVFILT_READ; 142 kqueueop->changes[0].flags = EV_ADD; 143 /* 144 * If kqueue works, then kevent will succeed, and it will 145 * stick an error in events[0]. If kqueue is broken, then 146 * kevent will fail. 147 */ 148 if (kevent(kq, 149 kqueueop->changes, 1, kqueueop->events, NEVENT, NULL) != 1 || 150 (int)kqueueop->events[0].ident != -1 || 151 kqueueop->events[0].flags != EV_ERROR) { 152 event_warn("%s: detected broken kqueue; not using.", __func__); 153 goto err; 154 } 155 156 base->evsigsel = &kqsigops; 157 158 return (kqueueop); 159 err: 160 if (kqueueop) 161 kqop_free(kqueueop); 162 163 return (NULL); 164 } 165 166 static void 167 kq_sighandler(int sig) 168 { 169 /* Do nothing here */ 170 } 171 172 static void 173 kq_setup_kevent(struct kevent *out, evutil_socket_t fd, int filter, short change) 174 { 175 memset(out, 0, sizeof(out)); 176 out->ident = fd; 177 out->filter = filter; 178 179 if (change & EV_CHANGE_ADD) { 180 out->flags = EV_ADD; 181 if (change & EV_ET) 182 out->flags |= EV_CLEAR; 183 #ifdef NOTE_EOF 184 /* Make it behave like select() and poll() */ 185 if (filter == EVFILT_READ) 186 out->fflags = NOTE_EOF; 187 #endif 188 } else { 189 EVUTIL_ASSERT(change & EV_CHANGE_DEL); 190 out->flags = EV_DELETE; 191 } 192 } 193 194 static int 195 kq_build_changes_list(const struct event_changelist *changelist, 196 struct kqop *kqop) 197 { 198 int i; 199 int n_changes = 0; 200 201 for (i = 0; i < changelist->n_changes; ++i) { 202 struct event_change *in_ch = &changelist->changes[i]; 203 struct kevent *out_ch; 204 if (n_changes >= kqop->changes_size - 1) { 205 int newsize = kqop->changes_size * 2; 206 struct kevent *newchanges; 207 208 newchanges = mm_realloc(kqop->changes, 209 newsize * sizeof(struct kevent)); 210 if (newchanges == NULL) { 211 event_warn("%s: realloc", __func__); 212 return (-1); 213 } 214 kqop->changes = newchanges; 215 kqop->changes_size = newsize; 216 } 217 if (in_ch->read_change) { 218 out_ch = &kqop->changes[n_changes++]; 219 kq_setup_kevent(out_ch, in_ch->fd, EVFILT_READ, 220 in_ch->read_change); 221 } 222 if (in_ch->write_change) { 223 out_ch = &kqop->changes[n_changes++]; 224 kq_setup_kevent(out_ch, in_ch->fd, EVFILT_WRITE, 225 in_ch->write_change); 226 } 227 } 228 return n_changes; 229 } 230 231 static int 232 kq_dispatch(struct event_base *base, struct timeval *tv) 233 { 234 struct kqop *kqop = base->evbase; 235 struct kevent *events = kqop->events; 236 struct kevent *changes; 237 struct timespec ts, *ts_p = NULL; 238 int i, n_changes, res; 239 240 if (tv != NULL) { 241 TIMEVAL_TO_TIMESPEC(tv, &ts); 242 ts_p = &ts; 243 } 244 245 /* Build "changes" from "base->changes" */ 246 EVUTIL_ASSERT(kqop->changes); 247 n_changes = kq_build_changes_list(&base->changelist, kqop); 248 if (n_changes < 0) 249 return -1; 250 251 event_changelist_remove_all(&base->changelist, base); 252 253 /* steal the changes array in case some broken code tries to call 254 * dispatch twice at once. */ 255 changes = kqop->changes; 256 kqop->changes = NULL; 257 258 EVBASE_RELEASE_LOCK(base, th_base_lock); 259 260 res = kevent(kqop->kq, changes, n_changes, 261 events, kqop->events_size, ts_p); 262 263 EVBASE_ACQUIRE_LOCK(base, th_base_lock); 264 265 EVUTIL_ASSERT(kqop->changes == NULL); 266 kqop->changes = changes; 267 268 if (res == -1) { 269 if (errno != EINTR) { 270 event_warn("kevent"); 271 return (-1); 272 } 273 274 return (0); 275 } 276 277 event_debug(("%s: kevent reports %d", __func__, res)); 278 279 for (i = 0; i < res; i++) { 280 int which = 0; 281 282 if (events[i].flags & EV_ERROR) { 283 /* 284 * Error messages that can happen, when a delete fails. 285 * EBADF happens when the file descriptor has been 286 * closed, 287 * ENOENT when the file descriptor was closed and 288 * then reopened. 289 * EINVAL for some reasons not understood; EINVAL 290 * should not be returned ever; but FreeBSD does :-\ 291 * An error is also indicated when a callback deletes 292 * an event we are still processing. In that case 293 * the data field is set to ENOENT. 294 */ 295 if (events[i].data == EBADF || 296 events[i].data == EINVAL || 297 events[i].data == ENOENT) 298 continue; 299 errno = events[i].data; 300 return (-1); 301 } 302 303 if (events[i].filter == EVFILT_READ) { 304 which |= EV_READ; 305 } else if (events[i].filter == EVFILT_WRITE) { 306 which |= EV_WRITE; 307 } else if (events[i].filter == EVFILT_SIGNAL) { 308 which |= EV_SIGNAL; 309 } 310 311 if (!which) 312 continue; 313 314 if (events[i].filter == EVFILT_SIGNAL) { 315 evmap_signal_active(base, events[i].ident, 1); 316 } else { 317 evmap_io_active(base, events[i].ident, which | EV_ET); 318 } 319 } 320 321 if (res == kqop->events_size) { 322 struct kevent *newresult; 323 int size = kqop->events_size; 324 /* We used all the events space that we have. Maybe we should 325 make it bigger. */ 326 size *= 2; 327 newresult = mm_realloc(kqop->events, 328 size * sizeof(struct kevent)); 329 if (newresult) { 330 kqop->events = newresult; 331 kqop->events_size = size; 332 } 333 } 334 335 return (0); 336 } 337 338 static void 339 kqop_free(struct kqop *kqop) 340 { 341 if (kqop->changes) 342 mm_free(kqop->changes); 343 if (kqop->events) 344 mm_free(kqop->events); 345 if (kqop->kq >= 0 && kqop->pid == getpid()) 346 close(kqop->kq); 347 memset(kqop, 0, sizeof(struct kqop)); 348 mm_free(kqop); 349 } 350 351 static void 352 kq_dealloc(struct event_base *base) 353 { 354 struct kqop *kqop = base->evbase; 355 evsig_dealloc(base); 356 kqop_free(kqop); 357 } 358 359 /* signal handling */ 360 static int 361 kq_sig_add(struct event_base *base, int nsignal, short old, short events, void *p) 362 { 363 struct kqop *kqop = base->evbase; 364 struct kevent kev; 365 struct timespec timeout = { 0, 0 }; 366 (void)p; 367 368 EVUTIL_ASSERT(nsignal >= 0 && nsignal < NSIG); 369 370 memset(&kev, 0, sizeof(kev)); 371 kev.ident = nsignal; 372 kev.filter = EVFILT_SIGNAL; 373 kev.flags = EV_ADD; 374 375 /* Be ready for the signal if it is sent any 376 * time between now and the next call to 377 * kq_dispatch. */ 378 if (kevent(kqop->kq, &kev, 1, NULL, 0, &timeout) == -1) 379 return (-1); 380 381 /* XXXX The manpage suggest we could use SIG_IGN instead of a 382 * do-nothing handler */ 383 if (_evsig_set_handler(base, nsignal, kq_sighandler) == -1) 384 return (-1); 385 386 return (0); 387 } 388 389 static int 390 kq_sig_del(struct event_base *base, int nsignal, short old, short events, void *p) 391 { 392 struct kqop *kqop = base->evbase; 393 struct kevent kev; 394 395 struct timespec timeout = { 0, 0 }; 396 (void)p; 397 398 EVUTIL_ASSERT(nsignal >= 0 && nsignal < NSIG); 399 400 memset(&kev, 0, sizeof(kev)); 401 kev.ident = nsignal; 402 kev.filter = EVFILT_SIGNAL; 403 kev.flags = EV_DELETE; 404 405 /* Because we insert signal events 406 * immediately, we need to delete them 407 * immediately, too */ 408 if (kevent(kqop->kq, &kev, 1, NULL, 0, &timeout) == -1) 409 return (-1); 410 411 if (_evsig_restore_handler(base, nsignal) == -1) 412 return (-1); 413 414 return (0); 415 } 416