xref: /minix/external/bsd/libevent/dist/evport.c (revision e985b929)
1 /*	$NetBSD: evport.c,v 1.1.1.1 2013/04/11 16:43:25 christos Exp $	*/
2 /*
3  * Submitted by David Pacheco (dp.spambait@gmail.com)
4  *
5  * Copyright 2006-2007 Niels Provos
6  * Copyright 2007-2012 Niels Provos and Nick Mathewson
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY SUN MICROSYSTEMS, INC. ``AS IS'' AND ANY
20  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
21  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
22  * DISCLAIMED. IN NO EVENT SHALL SUN MICROSYSTEMS, INC. BE LIABLE FOR ANY
23  * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
24  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
25  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
26  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
28  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 /*
32  * Copyright (c) 2007 Sun Microsystems. All rights reserved.
33  * Use is subject to license terms.
34  */
35 
36 /*
37  * evport.c: event backend using Solaris 10 event ports. See port_create(3C).
38  * This implementation is loosely modeled after the one used for select(2) (in
39  * select.c).
40  *
41  * The outstanding events are tracked in a data structure called evport_data.
42  * Each entry in the ed_fds array corresponds to a file descriptor, and contains
43  * pointers to the read and write events that correspond to that fd. (That is,
44  * when the file is readable, the "read" event should handle it, etc.)
45  *
46  * evport_add and evport_del update this data structure. evport_dispatch uses it
47  * to determine where to callback when an event occurs (which it gets from
48  * port_getn).
49  *
50  * Helper functions are used: grow() grows the file descriptor array as
51  * necessary when large fd's come in. reassociate() takes care of maintaining
52  * the proper file-descriptor/event-port associations.
53  *
54  * As in the select(2) implementation, signals are handled by evsignal.
55  */
56 
57 #include "event2/event-config.h"
58 #include <sys/cdefs.h>
59 __RCSID("$NetBSD: evport.c,v 1.1.1.1 2013/04/11 16:43:25 christos Exp $");
60 
61 #include <sys/time.h>
62 #include <sys/queue.h>
63 #include <errno.h>
64 #include <poll.h>
65 #include <port.h>
66 #include <signal.h>
67 #include <stdio.h>
68 #include <stdlib.h>
69 #include <string.h>
70 #include <time.h>
71 #include <unistd.h>
72 
73 #include "event2/thread.h"
74 
75 #include "evthread-internal.h"
76 #include "event-internal.h"
77 #include "log-internal.h"
78 #include "evsignal-internal.h"
79 #include "evmap-internal.h"
80 
81 /*
82  * Default value for ed_nevents, which is the maximum file descriptor number we
83  * can handle. If an event comes in for a file descriptor F > nevents, we will
84  * grow the array of file descriptors, doubling its size.
85  */
86 #define DEFAULT_NFDS	16
87 
88 
89 /*
90  * EVENTS_PER_GETN is the maximum number of events to retrieve from port_getn on
91  * any particular call. You can speed things up by increasing this, but it will
92  * (obviously) require more memory.
93  */
94 #define EVENTS_PER_GETN 8
95 
96 /*
97  * Per-file-descriptor information about what events we're subscribed to. These
98  * fields are NULL if no event is subscribed to either of them.
99  */
100 
101 struct fd_info {
102 	short fdi_what;		/* combinations of EV_READ and EV_WRITE */
103 };
104 
105 #define FDI_HAS_READ(fdi)  ((fdi)->fdi_what & EV_READ)
106 #define FDI_HAS_WRITE(fdi) ((fdi)->fdi_what & EV_WRITE)
107 #define FDI_HAS_EVENTS(fdi) (FDI_HAS_READ(fdi) || FDI_HAS_WRITE(fdi))
108 #define FDI_TO_SYSEVENTS(fdi) (FDI_HAS_READ(fdi) ? POLLIN : 0) | \
109     (FDI_HAS_WRITE(fdi) ? POLLOUT : 0)
110 
111 struct evport_data {
112 	int		ed_port;	/* event port for system events  */
113 	int		ed_nevents;	/* number of allocated fdi's	 */
114 	struct fd_info *ed_fds;		/* allocated fdi table		 */
115 	/* fdi's that we need to reassoc */
116 	int ed_pending[EVENTS_PER_GETN]; /* fd's with pending events */
117 };
118 
119 static void*	evport_init(struct event_base *);
120 static int evport_add(struct event_base *, int fd, short old, short events, void *);
121 static int evport_del(struct event_base *, int fd, short old, short events, void *);
122 static int	evport_dispatch(struct event_base *, struct timeval *);
123 static void	evport_dealloc(struct event_base *);
124 
125 const struct eventop evportops = {
126 	"evport",
127 	evport_init,
128 	evport_add,
129 	evport_del,
130 	evport_dispatch,
131 	evport_dealloc,
132 	1, /* need reinit */
133 	0, /* features */
134 	0, /* fdinfo length */
135 };
136 
137 /*
138  * Initialize the event port implementation.
139  */
140 
141 static void*
evport_init(struct event_base * base)142 evport_init(struct event_base *base)
143 {
144 	struct evport_data *evpd;
145 	int i;
146 
147 	if (!(evpd = mm_calloc(1, sizeof(struct evport_data))))
148 		return (NULL);
149 
150 	if ((evpd->ed_port = port_create()) == -1) {
151 		mm_free(evpd);
152 		return (NULL);
153 	}
154 
155 	/*
156 	 * Initialize file descriptor structure
157 	 */
158 	evpd->ed_fds = mm_calloc(DEFAULT_NFDS, sizeof(struct fd_info));
159 	if (evpd->ed_fds == NULL) {
160 		close(evpd->ed_port);
161 		mm_free(evpd);
162 		return (NULL);
163 	}
164 	evpd->ed_nevents = DEFAULT_NFDS;
165 	for (i = 0; i < EVENTS_PER_GETN; i++)
166 		evpd->ed_pending[i] = -1;
167 
168 	evsig_init(base);
169 
170 	return (evpd);
171 }
172 
173 #ifdef CHECK_INVARIANTS
174 /*
175  * Checks some basic properties about the evport_data structure. Because it
176  * checks all file descriptors, this function can be expensive when the maximum
177  * file descriptor ever used is rather large.
178  */
179 
180 static void
check_evportop(struct evport_data * evpd)181 check_evportop(struct evport_data *evpd)
182 {
183 	EVUTIL_ASSERT(evpd);
184 	EVUTIL_ASSERT(evpd->ed_nevents > 0);
185 	EVUTIL_ASSERT(evpd->ed_port > 0);
186 	EVUTIL_ASSERT(evpd->ed_fds > 0);
187 }
188 
189 /*
190  * Verifies very basic integrity of a given port_event.
191  */
192 static void
check_event(port_event_t * pevt)193 check_event(port_event_t* pevt)
194 {
195 	/*
196 	 * We've only registered for PORT_SOURCE_FD events. The only
197 	 * other thing we can legitimately receive is PORT_SOURCE_ALERT,
198 	 * but since we're not using port_alert either, we can assume
199 	 * PORT_SOURCE_FD.
200 	 */
201 	EVUTIL_ASSERT(pevt->portev_source == PORT_SOURCE_FD);
202 	EVUTIL_ASSERT(pevt->portev_user == NULL);
203 }
204 
205 #else
206 #define check_evportop(epop)
207 #define check_event(pevt)
208 #endif /* CHECK_INVARIANTS */
209 
210 /*
211  * Doubles the size of the allocated file descriptor array.
212  */
213 static int
grow(struct evport_data * epdp,int factor)214 grow(struct evport_data *epdp, int factor)
215 {
216 	struct fd_info *tmp;
217 	int oldsize = epdp->ed_nevents;
218 	int newsize = factor * oldsize;
219 	EVUTIL_ASSERT(factor > 1);
220 
221 	check_evportop(epdp);
222 
223 	tmp = mm_realloc(epdp->ed_fds, sizeof(struct fd_info) * newsize);
224 	if (NULL == tmp)
225 		return -1;
226 	epdp->ed_fds = tmp;
227 	memset((char*) (epdp->ed_fds + oldsize), 0,
228 	    (newsize - oldsize)*sizeof(struct fd_info));
229 	epdp->ed_nevents = newsize;
230 
231 	check_evportop(epdp);
232 
233 	return 0;
234 }
235 
236 
237 /*
238  * (Re)associates the given file descriptor with the event port. The OS events
239  * are specified (implicitly) from the fd_info struct.
240  */
241 static int
reassociate(struct evport_data * epdp,struct fd_info * fdip,int fd)242 reassociate(struct evport_data *epdp, struct fd_info *fdip, int fd)
243 {
244 	int sysevents = FDI_TO_SYSEVENTS(fdip);
245 
246 	if (sysevents != 0) {
247 		if (port_associate(epdp->ed_port, PORT_SOURCE_FD,
248 				   fd, sysevents, NULL) == -1) {
249 			event_warn("port_associate");
250 			return (-1);
251 		}
252 	}
253 
254 	check_evportop(epdp);
255 
256 	return (0);
257 }
258 
259 /*
260  * Main event loop - polls port_getn for some number of events, and processes
261  * them.
262  */
263 
264 static int
evport_dispatch(struct event_base * base,struct timeval * tv)265 evport_dispatch(struct event_base *base, struct timeval *tv)
266 {
267 	int i, res;
268 	struct evport_data *epdp = base->evbase;
269 	port_event_t pevtlist[EVENTS_PER_GETN];
270 
271 	/*
272 	 * port_getn will block until it has at least nevents events. It will
273 	 * also return how many it's given us (which may be more than we asked
274 	 * for, as long as it's less than our maximum (EVENTS_PER_GETN)) in
275 	 * nevents.
276 	 */
277 	int nevents = 1;
278 
279 	/*
280 	 * We have to convert a struct timeval to a struct timespec
281 	 * (only difference is nanoseconds vs. microseconds). If no time-based
282 	 * events are active, we should wait for I/O (and tv == NULL).
283 	 */
284 	struct timespec ts;
285 	struct timespec *ts_p = NULL;
286 	if (tv != NULL) {
287 		ts.tv_sec = tv->tv_sec;
288 		ts.tv_nsec = tv->tv_usec * 1000;
289 		ts_p = &ts;
290 	}
291 
292 	/*
293 	 * Before doing anything else, we need to reassociate the events we hit
294 	 * last time which need reassociation. See comment at the end of the
295 	 * loop below.
296 	 */
297 	for (i = 0; i < EVENTS_PER_GETN; ++i) {
298 		struct fd_info *fdi = NULL;
299 		if (epdp->ed_pending[i] != -1) {
300 			fdi = &(epdp->ed_fds[epdp->ed_pending[i]]);
301 		}
302 
303 		if (fdi != NULL && FDI_HAS_EVENTS(fdi)) {
304 			int fd = epdp->ed_pending[i];
305 			reassociate(epdp, fdi, fd);
306 			epdp->ed_pending[i] = -1;
307 		}
308 	}
309 
310 	EVBASE_RELEASE_LOCK(base, th_base_lock);
311 
312 	res = port_getn(epdp->ed_port, pevtlist, EVENTS_PER_GETN,
313 	    (unsigned int *) &nevents, ts_p);
314 
315 	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
316 
317 	if (res == -1) {
318 		if (errno == EINTR || errno == EAGAIN) {
319 			return (0);
320 		} else if (errno == ETIME) {
321 			if (nevents == 0)
322 				return (0);
323 		} else {
324 			event_warn("port_getn");
325 			return (-1);
326 		}
327 	}
328 
329 	event_debug(("%s: port_getn reports %d events", __func__, nevents));
330 
331 	for (i = 0; i < nevents; ++i) {
332 		struct fd_info *fdi;
333 		port_event_t *pevt = &pevtlist[i];
334 		int fd = (int) pevt->portev_object;
335 
336 		check_evportop(epdp);
337 		check_event(pevt);
338 		epdp->ed_pending[i] = fd;
339 
340 		/*
341 		 * Figure out what kind of event it was
342 		 * (because we have to pass this to the callback)
343 		 */
344 		res = 0;
345 		if (pevt->portev_events & (POLLERR|POLLHUP)) {
346 			res = EV_READ | EV_WRITE;
347 		} else {
348 			if (pevt->portev_events & POLLIN)
349 				res |= EV_READ;
350 			if (pevt->portev_events & POLLOUT)
351 				res |= EV_WRITE;
352 		}
353 
354 		/*
355 		 * Check for the error situations or a hangup situation
356 		 */
357 		if (pevt->portev_events & (POLLERR|POLLHUP|POLLNVAL))
358 			res |= EV_READ|EV_WRITE;
359 
360 		EVUTIL_ASSERT(epdp->ed_nevents > fd);
361 		fdi = &(epdp->ed_fds[fd]);
362 
363 		evmap_io_active(base, fd, res);
364 	} /* end of all events gotten */
365 
366 	check_evportop(epdp);
367 
368 	return (0);
369 }
370 
371 
372 /*
373  * Adds the given event (so that you will be notified when it happens via
374  * the callback function).
375  */
376 
377 static int
evport_add(struct event_base * base,int fd,short old,short events,void * p)378 evport_add(struct event_base *base, int fd, short old, short events, void *p)
379 {
380 	struct evport_data *evpd = base->evbase;
381 	struct fd_info *fdi;
382 	int factor;
383 	(void)p;
384 
385 	check_evportop(evpd);
386 
387 	/*
388 	 * If necessary, grow the file descriptor info table
389 	 */
390 
391 	factor = 1;
392 	while (fd >= factor * evpd->ed_nevents)
393 		factor *= 2;
394 
395 	if (factor > 1) {
396 		if (-1 == grow(evpd, factor)) {
397 			return (-1);
398 		}
399 	}
400 
401 	fdi = &evpd->ed_fds[fd];
402 	fdi->fdi_what |= events;
403 
404 	return reassociate(evpd, fdi, fd);
405 }
406 
407 /*
408  * Removes the given event from the list of events to wait for.
409  */
410 
411 static int
evport_del(struct event_base * base,int fd,short old,short events,void * p)412 evport_del(struct event_base *base, int fd, short old, short events, void *p)
413 {
414 	struct evport_data *evpd = base->evbase;
415 	struct fd_info *fdi;
416 	int i;
417 	int associated = 1;
418 	(void)p;
419 
420 	check_evportop(evpd);
421 
422 	if (evpd->ed_nevents < fd) {
423 		return (-1);
424 	}
425 
426 	for (i = 0; i < EVENTS_PER_GETN; ++i) {
427 		if (evpd->ed_pending[i] == fd) {
428 			associated = 0;
429 			break;
430 		}
431 	}
432 
433 	fdi = &evpd->ed_fds[fd];
434 	if (events & EV_READ)
435 		fdi->fdi_what &= ~EV_READ;
436 	if (events & EV_WRITE)
437 		fdi->fdi_what &= ~EV_WRITE;
438 
439 	if (associated) {
440 		if (!FDI_HAS_EVENTS(fdi) &&
441 		    port_dissociate(evpd->ed_port, PORT_SOURCE_FD, fd) == -1) {
442 			/*
443 			 * Ignore EBADFD error the fd could have been closed
444 			 * before event_del() was called.
445 			 */
446 			if (errno != EBADFD) {
447 				event_warn("port_dissociate");
448 				return (-1);
449 			}
450 		} else {
451 			if (FDI_HAS_EVENTS(fdi)) {
452 				return (reassociate(evpd, fdi, fd));
453 			}
454 		}
455 	} else {
456 		if ((fdi->fdi_what & (EV_READ|EV_WRITE)) == 0) {
457 			evpd->ed_pending[i] = -1;
458 		}
459 	}
460 	return 0;
461 }
462 
463 
464 static void
evport_dealloc(struct event_base * base)465 evport_dealloc(struct event_base *base)
466 {
467 	struct evport_data *evpd = base->evbase;
468 
469 	evsig_dealloc(base);
470 
471 	close(evpd->ed_port);
472 
473 	if (evpd->ed_fds)
474 		mm_free(evpd->ed_fds);
475 	mm_free(evpd);
476 }
477