xref: /netbsd-src/external/bsd/libevent/dist/evport.c (revision 805a1ce9000bc0ec0951bf35f7e52b85cafb37e2)
1 /*	$NetBSD: evport.c,v 1.1.1.2 2017/01/31 21:14:53 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 1.1.1.2 2017/01/31 21:14:53 christos Exp $");
60 #include "evconfig-private.h"
61 
62 #ifdef EVENT__HAVE_EVENT_PORTS
63 
64 #include <sys/time.h>
65 #include <sys/queue.h>
66 #include <errno.h>
67 #include <poll.h>
68 #include <port.h>
69 #include <signal.h>
70 #include <stdio.h>
71 #include <stdlib.h>
72 #include <string.h>
73 #include <time.h>
74 #include <unistd.h>
75 
76 #include "event2/thread.h"
77 
78 #include "evthread-internal.h"
79 #include "event-internal.h"
80 #include "log-internal.h"
81 #include "evsignal-internal.h"
82 #include "evmap-internal.h"
83 
84 #define INITIAL_EVENTS_PER_GETN 8
85 #define MAX_EVENTS_PER_GETN 4096
86 
87 /*
88  * Per-file-descriptor information about what events we're subscribed to. These
89  * fields are NULL if no event is subscribed to either of them.
90  */
91 
92 struct fd_info {
93 	/* combinations of EV_READ and EV_WRITE */
94 	short fdi_what;
95 	/* Index of this fd within ed_pending, plus 1.  Zero if this fd is
96 	 * not in ed_pending.  (The +1 is a hack so that memset(0) will set
97 	 * it to a nil index. */
98 	int pending_idx_plus_1;
99 };
100 
101 #define FDI_HAS_READ(fdi)  ((fdi)->fdi_what & EV_READ)
102 #define FDI_HAS_WRITE(fdi) ((fdi)->fdi_what & EV_WRITE)
103 #define FDI_HAS_EVENTS(fdi) (FDI_HAS_READ(fdi) || FDI_HAS_WRITE(fdi))
104 #define FDI_TO_SYSEVENTS(fdi) (FDI_HAS_READ(fdi) ? POLLIN : 0) | \
105     (FDI_HAS_WRITE(fdi) ? POLLOUT : 0)
106 
107 struct evport_data {
108 	int		ed_port;	/* event port for system events  */
109 	/* How many elements of ed_pending should we look at? */
110 	int ed_npending;
111 	/* How many elements are allocated in ed_pending and pevtlist? */
112 	int ed_maxevents;
113 	/* fdi's that we need to reassoc */
114 	int *ed_pending;
115 	/* storage space for incoming events. */
116 	port_event_t *ed_pevtlist;
117 
118 };
119 
120 static void*	evport_init(struct event_base *);
121 static int evport_add(struct event_base *, int fd, short old, short events, void *);
122 static int evport_del(struct event_base *, int fd, short old, short events, void *);
123 static int	evport_dispatch(struct event_base *, struct timeval *);
124 static void	evport_dealloc(struct event_base *);
125 static int	grow(struct evport_data *, int min_events);
126 
127 const struct eventop evportops = {
128 	"evport",
129 	evport_init,
130 	evport_add,
131 	evport_del,
132 	evport_dispatch,
133 	evport_dealloc,
134 	1, /* need reinit */
135 	0, /* features */
136 	sizeof(struct fd_info), /* fdinfo length */
137 };
138 
139 /*
140  * Initialize the event port implementation.
141  */
142 
143 static void*
evport_init(struct event_base * base)144 evport_init(struct event_base *base)
145 {
146 	struct evport_data *evpd;
147 
148 	if (!(evpd = mm_calloc(1, sizeof(struct evport_data))))
149 		return (NULL);
150 
151 	if ((evpd->ed_port = port_create()) == -1) {
152 		mm_free(evpd);
153 		return (NULL);
154 	}
155 
156 	if (grow(evpd, INITIAL_EVENTS_PER_GETN) < 0) {
157 		close(evpd->ed_port);
158 		mm_free(evpd);
159 		return NULL;
160 	}
161 
162 	evpd->ed_npending = 0;
163 
164 	evsig_init_(base);
165 
166 	return (evpd);
167 }
168 
169 static int
grow(struct evport_data * data,int min_events)170 grow(struct evport_data *data, int min_events)
171 {
172 	int newsize;
173 	int *new_pending;
174 	port_event_t *new_pevtlist;
175 	if (data->ed_maxevents) {
176 		newsize = data->ed_maxevents;
177 		do {
178 			newsize *= 2;
179 		} while (newsize < min_events);
180 	} else {
181 		newsize = min_events;
182 	}
183 
184 	new_pending = mm_realloc(data->ed_pending, sizeof(int)*newsize);
185 	if (new_pending == NULL)
186 		return -1;
187 	data->ed_pending = new_pending;
188 	new_pevtlist = mm_realloc(data->ed_pevtlist, sizeof(port_event_t)*newsize);
189 	if (new_pevtlist == NULL)
190 		return -1;
191 	data->ed_pevtlist = new_pevtlist;
192 
193 	data->ed_maxevents = newsize;
194 	return 0;
195 }
196 
197 #ifdef CHECK_INVARIANTS
198 /*
199  * Checks some basic properties about the evport_data structure. Because it
200  * checks all file descriptors, this function can be expensive when the maximum
201  * file descriptor ever used is rather large.
202  */
203 
204 static void
check_evportop(struct evport_data * evpd)205 check_evportop(struct evport_data *evpd)
206 {
207 	EVUTIL_ASSERT(evpd);
208 	EVUTIL_ASSERT(evpd->ed_port > 0);
209 }
210 
211 /*
212  * Verifies very basic integrity of a given port_event.
213  */
214 static void
check_event(port_event_t * pevt)215 check_event(port_event_t* pevt)
216 {
217 	/*
218 	 * We've only registered for PORT_SOURCE_FD events. The only
219 	 * other thing we can legitimately receive is PORT_SOURCE_ALERT,
220 	 * but since we're not using port_alert either, we can assume
221 	 * PORT_SOURCE_FD.
222 	 */
223 	EVUTIL_ASSERT(pevt->portev_source == PORT_SOURCE_FD);
224 }
225 
226 #else
227 #define check_evportop(epop)
228 #define check_event(pevt)
229 #endif /* CHECK_INVARIANTS */
230 
231 /*
232  * (Re)associates the given file descriptor with the event port. The OS events
233  * are specified (implicitly) from the fd_info struct.
234  */
235 static int
reassociate(struct evport_data * epdp,struct fd_info * fdip,int fd)236 reassociate(struct evport_data *epdp, struct fd_info *fdip, int fd)
237 {
238 	int sysevents = FDI_TO_SYSEVENTS(fdip);
239 
240 	if (sysevents != 0) {
241 		if (port_associate(epdp->ed_port, PORT_SOURCE_FD,
242 				   fd, sysevents, fdip) == -1) {
243 			event_warn("port_associate");
244 			return (-1);
245 		}
246 	}
247 
248 	check_evportop(epdp);
249 
250 	return (0);
251 }
252 
253 /*
254  * Main event loop - polls port_getn for some number of events, and processes
255  * them.
256  */
257 
258 static int
evport_dispatch(struct event_base * base,struct timeval * tv)259 evport_dispatch(struct event_base *base, struct timeval *tv)
260 {
261 	int i, res;
262 	struct evport_data *epdp = base->evbase;
263 	port_event_t *pevtlist = epdp->ed_pevtlist;
264 
265 	/*
266 	 * port_getn will block until it has at least nevents events. It will
267 	 * also return how many it's given us (which may be more than we asked
268 	 * for, as long as it's less than our maximum (ed_maxevents)) in
269 	 * nevents.
270 	 */
271 	int nevents = 1;
272 
273 	/*
274 	 * We have to convert a struct timeval to a struct timespec
275 	 * (only difference is nanoseconds vs. microseconds). If no time-based
276 	 * events are active, we should wait for I/O (and tv == NULL).
277 	 */
278 	struct timespec ts;
279 	struct timespec *ts_p = NULL;
280 	if (tv != NULL) {
281 		ts.tv_sec = tv->tv_sec;
282 		ts.tv_nsec = tv->tv_usec * 1000;
283 		ts_p = &ts;
284 	}
285 
286 	/*
287 	 * Before doing anything else, we need to reassociate the events we hit
288 	 * last time which need reassociation. See comment at the end of the
289 	 * loop below.
290 	 */
291 	for (i = 0; i < epdp->ed_npending; ++i) {
292 		struct fd_info *fdi = NULL;
293 		const int fd = epdp->ed_pending[i];
294 		if (fd != -1) {
295 			/* We might have cleared out this event; we need
296 			 * to be sure that it's still set. */
297 			fdi = evmap_io_get_fdinfo_(&base->io, fd);
298 		}
299 
300 		if (fdi != NULL && FDI_HAS_EVENTS(fdi)) {
301 			reassociate(epdp, fdi, fd);
302 			/* epdp->ed_pending[i] = -1; */
303 			fdi->pending_idx_plus_1 = 0;
304 		}
305 	}
306 
307 	EVBASE_RELEASE_LOCK(base, th_base_lock);
308 
309 	res = port_getn(epdp->ed_port, pevtlist, epdp->ed_maxevents,
310 	    (unsigned int *) &nevents, ts_p);
311 
312 	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
313 
314 	if (res == -1) {
315 		if (errno == EINTR || errno == EAGAIN) {
316 			return (0);
317 		} else if (errno == ETIME) {
318 			if (nevents == 0)
319 				return (0);
320 		} else {
321 			event_warn("port_getn");
322 			return (-1);
323 		}
324 	}
325 
326 	event_debug(("%s: port_getn reports %d events", __func__, nevents));
327 
328 	for (i = 0; i < nevents; ++i) {
329 		port_event_t *pevt = &pevtlist[i];
330 		int fd = (int) pevt->portev_object;
331 		struct fd_info *fdi = pevt->portev_user;
332 		/*EVUTIL_ASSERT(evmap_io_get_fdinfo_(&base->io, fd) == fdi);*/
333 
334 		check_evportop(epdp);
335 		check_event(pevt);
336 		epdp->ed_pending[i] = fd;
337 		fdi->pending_idx_plus_1 = i + 1;
338 
339 		/*
340 		 * Figure out what kind of event it was
341 		 * (because we have to pass this to the callback)
342 		 */
343 		res = 0;
344 		if (pevt->portev_events & (POLLERR|POLLHUP)) {
345 			res = EV_READ | EV_WRITE;
346 		} else {
347 			if (pevt->portev_events & POLLIN)
348 				res |= EV_READ;
349 			if (pevt->portev_events & POLLOUT)
350 				res |= EV_WRITE;
351 		}
352 
353 		/*
354 		 * Check for the error situations or a hangup situation
355 		 */
356 		if (pevt->portev_events & (POLLERR|POLLHUP|POLLNVAL))
357 			res |= EV_READ|EV_WRITE;
358 
359 		evmap_io_active_(base, fd, res);
360 	} /* end of all events gotten */
361 	epdp->ed_npending = nevents;
362 
363 	if (nevents == epdp->ed_maxevents &&
364 	    epdp->ed_maxevents < MAX_EVENTS_PER_GETN) {
365 		/* we used all the space this time.  We should be ready
366 		 * for more events next time around. */
367 		grow(epdp, epdp->ed_maxevents * 2);
368 	}
369 
370 	check_evportop(epdp);
371 
372 	return (0);
373 }
374 
375 
376 /*
377  * Adds the given event (so that you will be notified when it happens via
378  * the callback function).
379  */
380 
381 static int
evport_add(struct event_base * base,int fd,short old,short events,void * p)382 evport_add(struct event_base *base, int fd, short old, short events, void *p)
383 {
384 	struct evport_data *evpd = base->evbase;
385 	struct fd_info *fdi = p;
386 
387 	check_evportop(evpd);
388 
389 	fdi->fdi_what |= events;
390 
391 	return reassociate(evpd, fdi, fd);
392 }
393 
394 /*
395  * Removes the given event from the list of events to wait for.
396  */
397 
398 static int
evport_del(struct event_base * base,int fd,short old,short events,void * p)399 evport_del(struct event_base *base, int fd, short old, short events, void *p)
400 {
401 	struct evport_data *evpd = base->evbase;
402 	struct fd_info *fdi = p;
403 	int associated = ! fdi->pending_idx_plus_1;
404 
405 	check_evportop(evpd);
406 
407 	fdi->fdi_what &= ~(events &(EV_READ|EV_WRITE));
408 
409 	if (associated) {
410 		if (!FDI_HAS_EVENTS(fdi) &&
411 		    port_dissociate(evpd->ed_port, PORT_SOURCE_FD, fd) == -1) {
412 			/*
413 			 * Ignore EBADFD error the fd could have been closed
414 			 * before event_del() was called.
415 			 */
416 			if (errno != EBADFD) {
417 				event_warn("port_dissociate");
418 				return (-1);
419 			}
420 		} else {
421 			if (FDI_HAS_EVENTS(fdi)) {
422 				return (reassociate(evpd, fdi, fd));
423 			}
424 		}
425 	} else {
426 		if ((fdi->fdi_what & (EV_READ|EV_WRITE)) == 0) {
427 			const int i = fdi->pending_idx_plus_1 - 1;
428 			EVUTIL_ASSERT(evpd->ed_pending[i] == fd);
429 			evpd->ed_pending[i] = -1;
430 			fdi->pending_idx_plus_1 = 0;
431 		}
432 	}
433 	return 0;
434 }
435 
436 
437 static void
evport_dealloc(struct event_base * base)438 evport_dealloc(struct event_base *base)
439 {
440 	struct evport_data *evpd = base->evbase;
441 
442 	evsig_dealloc_(base);
443 
444 	close(evpd->ed_port);
445 
446 	if (evpd->ed_pending)
447 		mm_free(evpd->ed_pending);
448 	if (evpd->ed_pevtlist)
449 		mm_free(evpd->ed_pevtlist);
450 
451 	mm_free(evpd);
452 }
453 
454 #endif /* EVENT__HAVE_EVENT_PORTS */
455