1/*
2 * Submitted by David Pacheco (dp.spambait@gmail.com)
3 *
4 * Copyright 2006-2007 Niels Provos
5 * Copyright 2007-2012 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 SUN MICROSYSTEMS, INC. ``AS IS'' AND ANY
19 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
20 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
21 * DISCLAIMED. IN NO EVENT SHALL SUN MICROSYSTEMS, INC. BE LIABLE FOR ANY
22 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
23 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
24 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
25 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
27 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 */
29
30/*
31 * Copyright (c) 2007 Sun Microsystems. All rights reserved.
32 * Use is subject to license terms.
33 */
34
35/*
36 * evport.c: event backend using Solaris 10 event ports. See port_create(3C).
37 * This implementation is loosely modeled after the one used for select(2) (in
38 * select.c).
39 *
40 * The outstanding events are tracked in a data structure called evport_data.
41 * Each entry in the ed_fds array corresponds to a file descriptor, and contains
42 * pointers to the read and write events that correspond to that fd. (That is,
43 * when the file is readable, the "read" event should handle it, etc.)
44 *
45 * evport_add and evport_del update this data structure. evport_dispatch uses it
46 * to determine where to callback when an event occurs (which it gets from
47 * port_getn).
48 *
49 * Helper functions are used: grow() grows the file descriptor array as
50 * necessary when large fd's come in. reassociate() takes care of maintaining
51 * the proper file-descriptor/event-port associations.
52 *
53 * As in the select(2) implementation, signals are handled by evsignal.
54 */
55
56#include "event2/event-config.h"
57#include "evconfig-private.h"
58
59#ifdef EVENT__HAVE_EVENT_PORTS
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#define INITIAL_EVENTS_PER_GETN 8
82#define MAX_EVENTS_PER_GETN 4096
83
84/*
85 * Per-file-descriptor information about what events we're subscribed to. These
86 * fields are NULL if no event is subscribed to either of them.
87 */
88
89struct fd_info {
90	/* combinations of EV_READ and EV_WRITE */
91	short fdi_what;
92	/* Index of this fd within ed_pending, plus 1.  Zero if this fd is
93	 * not in ed_pending.  (The +1 is a hack so that memset(0) will set
94	 * it to a nil index. */
95	int pending_idx_plus_1;
96};
97
98#define FDI_HAS_READ(fdi)  ((fdi)->fdi_what & EV_READ)
99#define FDI_HAS_WRITE(fdi) ((fdi)->fdi_what & EV_WRITE)
100#define FDI_HAS_EVENTS(fdi) (FDI_HAS_READ(fdi) || FDI_HAS_WRITE(fdi))
101#define FDI_TO_SYSEVENTS(fdi) (FDI_HAS_READ(fdi) ? POLLIN : 0) | \
102    (FDI_HAS_WRITE(fdi) ? POLLOUT : 0)
103
104struct evport_data {
105	int		ed_port;	/* event port for system events  */
106	/* How many elements of ed_pending should we look at? */
107	int ed_npending;
108	/* How many elements are allocated in ed_pending and pevtlist? */
109	int ed_maxevents;
110	/* fdi's that we need to reassoc */
111	int *ed_pending;
112	/* storage space for incoming events. */
113	port_event_t *ed_pevtlist;
114
115};
116
117static void*	evport_init(struct event_base *);
118static int evport_add(struct event_base *, int fd, short old, short events, void *);
119static int evport_del(struct event_base *, int fd, short old, short events, void *);
120static int	evport_dispatch(struct event_base *, struct timeval *);
121static void	evport_dealloc(struct event_base *);
122static int	grow(struct evport_data *, int min_events);
123
124const struct eventop evportops = {
125	"evport",
126	evport_init,
127	evport_add,
128	evport_del,
129	evport_dispatch,
130	evport_dealloc,
131	1, /* need reinit */
132	0, /* features */
133	sizeof(struct fd_info), /* fdinfo length */
134};
135
136/*
137 * Initialize the event port implementation.
138 */
139
140static void*
141evport_init(struct event_base *base)
142{
143	struct evport_data *evpd;
144
145	if (!(evpd = mm_calloc(1, sizeof(struct evport_data))))
146		return (NULL);
147
148	if ((evpd->ed_port = port_create()) == -1) {
149		mm_free(evpd);
150		return (NULL);
151	}
152
153	if (grow(evpd, INITIAL_EVENTS_PER_GETN) < 0) {
154		close(evpd->ed_port);
155		mm_free(evpd);
156		return NULL;
157	}
158
159	evpd->ed_npending = 0;
160
161	evsig_init_(base);
162
163	return (evpd);
164}
165
166static int
167grow(struct evport_data *data, int min_events)
168{
169	int newsize;
170	int *new_pending;
171	port_event_t *new_pevtlist;
172	if (data->ed_maxevents) {
173		newsize = data->ed_maxevents;
174		do {
175			newsize *= 2;
176		} while (newsize < min_events);
177	} else {
178		newsize = min_events;
179	}
180
181	new_pending = mm_realloc(data->ed_pending, sizeof(int)*newsize);
182	if (new_pending == NULL)
183		return -1;
184	data->ed_pending = new_pending;
185	new_pevtlist = mm_realloc(data->ed_pevtlist, sizeof(port_event_t)*newsize);
186	if (new_pevtlist == NULL)
187		return -1;
188	data->ed_pevtlist = new_pevtlist;
189
190	data->ed_maxevents = newsize;
191	return 0;
192}
193
194#ifdef CHECK_INVARIANTS
195/*
196 * Checks some basic properties about the evport_data structure. Because it
197 * checks all file descriptors, this function can be expensive when the maximum
198 * file descriptor ever used is rather large.
199 */
200
201static void
202check_evportop(struct evport_data *evpd)
203{
204	EVUTIL_ASSERT(evpd);
205	EVUTIL_ASSERT(evpd->ed_port > 0);
206}
207
208/*
209 * Verifies very basic integrity of a given port_event.
210 */
211static void
212check_event(port_event_t* pevt)
213{
214	/*
215	 * We've only registered for PORT_SOURCE_FD events. The only
216	 * other thing we can legitimately receive is PORT_SOURCE_ALERT,
217	 * but since we're not using port_alert either, we can assume
218	 * PORT_SOURCE_FD.
219	 */
220	EVUTIL_ASSERT(pevt->portev_source == PORT_SOURCE_FD);
221}
222
223#else
224#define check_evportop(epop)
225#define check_event(pevt)
226#endif /* CHECK_INVARIANTS */
227
228/*
229 * (Re)associates the given file descriptor with the event port. The OS events
230 * are specified (implicitly) from the fd_info struct.
231 */
232static int
233reassociate(struct evport_data *epdp, struct fd_info *fdip, int fd)
234{
235	int sysevents = FDI_TO_SYSEVENTS(fdip);
236
237	if (sysevents != 0) {
238		if (port_associate(epdp->ed_port, PORT_SOURCE_FD,
239				   fd, sysevents, fdip) == -1) {
240			event_warn("port_associate");
241			return (-1);
242		}
243	}
244
245	check_evportop(epdp);
246
247	return (0);
248}
249
250/*
251 * Main event loop - polls port_getn for some number of events, and processes
252 * them.
253 */
254
255static int
256evport_dispatch(struct event_base *base, struct timeval *tv)
257{
258	int i, res;
259	struct evport_data *epdp = base->evbase;
260	port_event_t *pevtlist = epdp->ed_pevtlist;
261
262	/*
263	 * port_getn will block until it has at least nevents events. It will
264	 * also return how many it's given us (which may be more than we asked
265	 * for, as long as it's less than our maximum (ed_maxevents)) in
266	 * nevents.
267	 */
268	int nevents = 1;
269
270	/*
271	 * We have to convert a struct timeval to a struct timespec
272	 * (only difference is nanoseconds vs. microseconds). If no time-based
273	 * events are active, we should wait for I/O (and tv == NULL).
274	 */
275	struct timespec ts;
276	struct timespec *ts_p = NULL;
277	if (tv != NULL) {
278		ts.tv_sec = tv->tv_sec;
279		ts.tv_nsec = tv->tv_usec * 1000;
280		ts_p = &ts;
281	}
282
283	/*
284	 * Before doing anything else, we need to reassociate the events we hit
285	 * last time which need reassociation. See comment at the end of the
286	 * loop below.
287	 */
288	for (i = 0; i < epdp->ed_npending; ++i) {
289		struct fd_info *fdi = NULL;
290		const int fd = epdp->ed_pending[i];
291		if (fd != -1) {
292			/* We might have cleared out this event; we need
293			 * to be sure that it's still set. */
294			fdi = evmap_io_get_fdinfo_(&base->io, fd);
295		}
296
297		if (fdi != NULL && FDI_HAS_EVENTS(fdi)) {
298			reassociate(epdp, fdi, fd);
299			/* epdp->ed_pending[i] = -1; */
300			fdi->pending_idx_plus_1 = 0;
301		}
302	}
303
304	EVBASE_RELEASE_LOCK(base, th_base_lock);
305
306	res = port_getn(epdp->ed_port, pevtlist, epdp->ed_maxevents,
307	    (unsigned int *) &nevents, ts_p);
308
309	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
310
311	if (res == -1) {
312		if (errno == EINTR || errno == EAGAIN) {
313			return (0);
314		} else if (errno == ETIME) {
315			if (nevents == 0)
316				return (0);
317		} else {
318			event_warn("port_getn");
319			return (-1);
320		}
321	}
322
323	event_debug(("%s: port_getn reports %d events", __func__, nevents));
324
325	for (i = 0; i < nevents; ++i) {
326		port_event_t *pevt = &pevtlist[i];
327		int fd = (int) pevt->portev_object;
328		struct fd_info *fdi = pevt->portev_user;
329		/*EVUTIL_ASSERT(evmap_io_get_fdinfo_(&base->io, fd) == fdi);*/
330
331		check_evportop(epdp);
332		check_event(pevt);
333		epdp->ed_pending[i] = fd;
334		fdi->pending_idx_plus_1 = i + 1;
335
336		/*
337		 * Figure out what kind of event it was
338		 * (because we have to pass this to the callback)
339		 */
340		res = 0;
341		if (pevt->portev_events & (POLLERR|POLLHUP)) {
342			res = EV_READ | EV_WRITE;
343		} else {
344			if (pevt->portev_events & POLLIN)
345				res |= EV_READ;
346			if (pevt->portev_events & POLLOUT)
347				res |= EV_WRITE;
348		}
349
350		/*
351		 * Check for the error situations or a hangup situation
352		 */
353		if (pevt->portev_events & (POLLERR|POLLHUP|POLLNVAL))
354			res |= EV_READ|EV_WRITE;
355
356		evmap_io_active_(base, fd, res);
357	} /* end of all events gotten */
358	epdp->ed_npending = nevents;
359
360	if (nevents == epdp->ed_maxevents &&
361	    epdp->ed_maxevents < MAX_EVENTS_PER_GETN) {
362		/* we used all the space this time.  We should be ready
363		 * for more events next time around. */
364		grow(epdp, epdp->ed_maxevents * 2);
365	}
366
367	check_evportop(epdp);
368
369	return (0);
370}
371
372
373/*
374 * Adds the given event (so that you will be notified when it happens via
375 * the callback function).
376 */
377
378static int
379evport_add(struct event_base *base, int fd, short old, short events, void *p)
380{
381	struct evport_data *evpd = base->evbase;
382	struct fd_info *fdi = p;
383
384	check_evportop(evpd);
385
386	fdi->fdi_what |= events;
387
388	return reassociate(evpd, fdi, fd);
389}
390
391/*
392 * Removes the given event from the list of events to wait for.
393 */
394
395static int
396evport_del(struct event_base *base, int fd, short old, short events, void *p)
397{
398	struct evport_data *evpd = base->evbase;
399	struct fd_info *fdi = p;
400	int associated = ! fdi->pending_idx_plus_1;
401
402	check_evportop(evpd);
403
404	fdi->fdi_what &= ~(events &(EV_READ|EV_WRITE));
405
406	if (associated) {
407		if (!FDI_HAS_EVENTS(fdi) &&
408		    port_dissociate(evpd->ed_port, PORT_SOURCE_FD, fd) == -1) {
409			/*
410			 * Ignore EBADFD error the fd could have been closed
411			 * before event_del() was called.
412			 */
413			if (errno != EBADFD) {
414				event_warn("port_dissociate");
415				return (-1);
416			}
417		} else {
418			if (FDI_HAS_EVENTS(fdi)) {
419				return (reassociate(evpd, fdi, fd));
420			}
421		}
422	} else {
423		if ((fdi->fdi_what & (EV_READ|EV_WRITE)) == 0) {
424			const int i = fdi->pending_idx_plus_1 - 1;
425			EVUTIL_ASSERT(evpd->ed_pending[i] == fd);
426			evpd->ed_pending[i] = -1;
427			fdi->pending_idx_plus_1 = 0;
428		}
429	}
430	return 0;
431}
432
433
434static void
435evport_dealloc(struct event_base *base)
436{
437	struct evport_data *evpd = base->evbase;
438
439	evsig_dealloc_(base);
440
441	close(evpd->ed_port);
442
443	if (evpd->ed_pending)
444		mm_free(evpd->ed_pending);
445	if (evpd->ed_pevtlist)
446		mm_free(evpd->ed_pevtlist);
447
448	mm_free(evpd);
449}
450
451#endif /* EVENT__HAVE_EVENT_PORTS */
452