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 2007-2009 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 "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 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 base->evsigbase = kqueueop; 158 159 return (kqueueop); 160 err: 161 if (kqueueop) 162 kqop_free(kqueueop); 163 164 return (NULL); 165 } 166 167 static void 168 kq_sighandler(int sig) 169 { 170 /* Do nothing here */ 171 } 172 173 static void 174 kq_setup_kevent(struct kevent *out, evutil_socket_t fd, int filter, short change) 175 { 176 memset(out, 0, sizeof(out)); 177 out->ident = fd; 178 out->filter = filter; 179 180 if (change & EV_CHANGE_ADD) { 181 out->flags = EV_ADD; 182 if (change & EV_ET) 183 out->flags |= EV_CLEAR; 184 #ifdef NOTE_EOF 185 /* Make it behave like select() and poll() */ 186 if (filter == EVFILT_READ) 187 out->fflags = NOTE_EOF; 188 #endif 189 } else { 190 EVUTIL_ASSERT(change & EV_CHANGE_DEL); 191 out->flags = EV_DELETE; 192 } 193 } 194 195 static int 196 kq_build_changes_list(const struct event_changelist *changelist, 197 struct kqop *kqop) 198 { 199 int i; 200 int n_changes = 0; 201 202 for (i = 0; i < changelist->n_changes; ++i) { 203 struct event_change *in_ch = &changelist->changes[i]; 204 struct kevent *out_ch; 205 if (n_changes >= kqop->changes_size - 1) { 206 int newsize = kqop->changes_size * 2; 207 struct kevent *newchanges; 208 209 newchanges = mm_realloc(kqop->changes, 210 newsize * sizeof(struct kevent)); 211 if (newchanges == NULL) { 212 event_warn("%s: realloc", __func__); 213 return (-1); 214 } 215 kqop->changes = newchanges; 216 kqop->changes_size = newsize; 217 } 218 if (in_ch->read_change) { 219 out_ch = &kqop->changes[n_changes++]; 220 kq_setup_kevent(out_ch, in_ch->fd, EVFILT_READ, 221 in_ch->read_change); 222 } 223 if (in_ch->write_change) { 224 out_ch = &kqop->changes[n_changes++]; 225 kq_setup_kevent(out_ch, in_ch->fd, EVFILT_WRITE, 226 in_ch->write_change); 227 } 228 } 229 return n_changes; 230 } 231 232 static int 233 kq_dispatch(struct event_base *base, struct timeval *tv) 234 { 235 struct kqop *kqop = base->evbase; 236 struct kevent *events = kqop->events; 237 struct kevent *changes; 238 struct timespec ts, *ts_p = NULL; 239 int i, n_changes, res; 240 241 if (tv != NULL) { 242 TIMEVAL_TO_TIMESPEC(tv, &ts); 243 ts_p = &ts; 244 } 245 246 /* Build "changes" from "base->changes" */ 247 EVUTIL_ASSERT(kqop->changes); 248 n_changes = kq_build_changes_list(&base->changelist, kqop); 249 if (n_changes < 0) 250 return -1; 251 252 event_changelist_remove_all(&base->changelist, base); 253 254 /* steal the changes array in case some broken code tries to call 255 * dispatch twice at once. */ 256 changes = kqop->changes; 257 kqop->changes = NULL; 258 259 EVBASE_RELEASE_LOCK(base, th_base_lock); 260 261 res = kevent(kqop->kq, changes, n_changes, 262 events, kqop->events_size, ts_p); 263 264 EVBASE_ACQUIRE_LOCK(base, th_base_lock); 265 266 EVUTIL_ASSERT(kqop->changes == NULL); 267 kqop->changes = changes; 268 269 if (res == -1) { 270 if (errno != EINTR) { 271 event_warn("kevent"); 272 return (-1); 273 } 274 275 return (0); 276 } 277 278 event_debug(("%s: kevent reports %d", __func__, res)); 279 280 for (i = 0; i < res; i++) { 281 int which = 0; 282 283 if (events[i].flags & EV_ERROR) { 284 /* 285 * Error messages that can happen, when a delete fails. 286 * EBADF happens when the file descriptor has been 287 * closed, 288 * ENOENT when the file descriptor was closed and 289 * then reopened. 290 * EINVAL for some reasons not understood; EINVAL 291 * should not be returned ever; but FreeBSD does :-\ 292 * An error is also indicated when a callback deletes 293 * an event we are still processing. In that case 294 * the data field is set to ENOENT. 295 */ 296 if (events[i].data == EBADF || 297 events[i].data == EINVAL || 298 events[i].data == ENOENT) 299 continue; 300 errno = events[i].data; 301 return (-1); 302 } 303 304 if (events[i].filter == EVFILT_READ) { 305 which |= EV_READ; 306 } else if (events[i].filter == EVFILT_WRITE) { 307 which |= EV_WRITE; 308 } else if (events[i].filter == EVFILT_SIGNAL) { 309 which |= EV_SIGNAL; 310 } 311 312 if (!which) 313 continue; 314 315 if (events[i].filter == EVFILT_SIGNAL) { 316 evmap_signal_active(base, events[i].ident, 1); 317 } else { 318 evmap_io_active(base, events[i].ident, which | EV_ET); 319 } 320 } 321 322 if (res == kqop->events_size) { 323 struct kevent *newresult; 324 int size = kqop->events_size; 325 /* We used all the events space that we have. Maybe we should 326 make it bigger. */ 327 size *= 2; 328 newresult = mm_realloc(kqop->events, 329 size * sizeof(struct kevent)); 330 if (newresult) { 331 kqop->events = newresult; 332 kqop->events_size = size; 333 } 334 } 335 336 return (0); 337 } 338 339 static void 340 kqop_free(struct kqop *kqop) 341 { 342 if (kqop->changes) 343 mm_free(kqop->changes); 344 if (kqop->events) 345 mm_free(kqop->events); 346 if (kqop->kq >= 0 && kqop->pid == getpid()) 347 close(kqop->kq); 348 memset(kqop, 0, sizeof(struct kqop)); 349 mm_free(kqop); 350 } 351 352 static void 353 kq_dealloc(struct event_base *base) 354 { 355 struct kqop *kqop = base->evbase; 356 evsig_dealloc(base); 357 kqop_free(kqop); 358 } 359 360 /* signal handling */ 361 static int 362 kq_sig_add(struct event_base *base, int nsignal, short old, short events, void *p) 363 { 364 struct kqop *kqop = base->evbase; 365 struct kevent kev; 366 struct timespec timeout = { 0, 0 }; 367 (void)p; 368 369 EVUTIL_ASSERT(nsignal >= 0 && nsignal < NSIG); 370 371 memset(&kev, 0, sizeof(kev)); 372 kev.ident = nsignal; 373 kev.filter = EVFILT_SIGNAL; 374 kev.flags = EV_ADD; 375 376 /* Be ready for the signal if it is sent any 377 * time between now and the next call to 378 * kq_dispatch. */ 379 if (kevent(kqop->kq, &kev, 1, NULL, 0, &timeout) == -1) 380 return (-1); 381 382 if (_evsig_set_handler(base, nsignal, kq_sighandler) == -1) 383 return (-1); 384 385 return (0); 386 } 387 388 static int 389 kq_sig_del(struct event_base *base, int nsignal, short old, short events, void *p) 390 { 391 struct kqop *kqop = base->evbase; 392 struct kevent kev; 393 394 struct timespec timeout = { 0, 0 }; 395 (void)p; 396 397 EVUTIL_ASSERT(nsignal >= 0 && nsignal < NSIG); 398 399 memset(&kev, 0, sizeof(kev)); 400 kev.ident = nsignal; 401 kev.filter = EVFILT_SIGNAL; 402 kev.flags = EV_DELETE; 403 404 /* Because we insert signal events 405 * immediately, we need to delete them 406 * immediately, too */ 407 if (kevent(kqop->kq, &kev, 1, NULL, 0, &timeout) == -1) 408 return (-1); 409 410 if (_evsig_restore_handler(base, nsignal) == -1) 411 return (-1); 412 413 return (0); 414 } 415