1/*	$OpenBSD: poll.c,v 1.2 2002/06/25 15:50:15 mickey Exp $	*/
2
3/*
4 * Copyright 2000-2007 Niels Provos <provos@citi.umich.edu>
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 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#include "evconfig-private.h"
31
32#ifdef EVENT__HAVE_POLL
33
34#include <sys/types.h>
35#ifdef EVENT__HAVE_SYS_TIME_H
36#include <sys/time.h>
37#endif
38#include <sys/queue.h>
39#include <poll.h>
40#include <signal.h>
41#include <limits.h>
42#include <stdio.h>
43#include <stdlib.h>
44#include <string.h>
45#include <unistd.h>
46#include <errno.h>
47
48#include "event-internal.h"
49#include "evsignal-internal.h"
50#include "log-internal.h"
51#include "evmap-internal.h"
52#include "event2/thread.h"
53#include "evthread-internal.h"
54#include "time-internal.h"
55
56/* Since Linux 2.6.17, poll is able to report about peer half-closed connection
57   using special POLLRDHUP flag on a read event.
58*/
59#if !defined(POLLRDHUP)
60#define POLLRDHUP 0
61#define EARLY_CLOSE_IF_HAVE_RDHUP 0
62#else
63#define EARLY_CLOSE_IF_HAVE_RDHUP EV_FEATURE_EARLY_CLOSE
64#endif
65
66
67struct pollidx {
68	int idxplus1;
69};
70
71struct pollop {
72	int event_count;		/* Highest number alloc */
73	int nfds;			/* Highest number used */
74	int realloc_copy;		/* True iff we must realloc
75					 * event_set_copy */
76	struct pollfd *event_set;
77	struct pollfd *event_set_copy;
78};
79
80static void *poll_init(struct event_base *);
81static int poll_add(struct event_base *, int, short old, short events, void *idx);
82static int poll_del(struct event_base *, int, short old, short events, void *idx);
83static int poll_dispatch(struct event_base *, struct timeval *);
84static void poll_dealloc(struct event_base *);
85
86const struct eventop pollops = {
87	"poll",
88	poll_init,
89	poll_add,
90	poll_del,
91	poll_dispatch,
92	poll_dealloc,
93	1, /* need_reinit */
94	EV_FEATURE_FDS|EARLY_CLOSE_IF_HAVE_RDHUP,
95	sizeof(struct pollidx),
96};
97
98static void *
99poll_init(struct event_base *base)
100{
101	struct pollop *pollop;
102
103	if (!(pollop = mm_calloc(1, sizeof(struct pollop))))
104		return (NULL);
105
106	evsig_init_(base);
107
108	evutil_weakrand_seed_(&base->weakrand_seed, 0);
109
110	return (pollop);
111}
112
113#ifdef CHECK_INVARIANTS
114static void
115poll_check_ok(struct pollop *pop)
116{
117	int i, idx;
118	struct event *ev;
119
120	for (i = 0; i < pop->fd_count; ++i) {
121		idx = pop->idxplus1_by_fd[i]-1;
122		if (idx < 0)
123			continue;
124		EVUTIL_ASSERT(pop->event_set[idx].fd == i);
125	}
126	for (i = 0; i < pop->nfds; ++i) {
127		struct pollfd *pfd = &pop->event_set[i];
128		EVUTIL_ASSERT(pop->idxplus1_by_fd[pfd->fd] == i+1);
129	}
130}
131#else
132#define poll_check_ok(pop)
133#endif
134
135static int
136poll_dispatch(struct event_base *base, struct timeval *tv)
137{
138	int res, i, j, nfds;
139	long msec = -1;
140	struct pollop *pop = base->evbase;
141	struct pollfd *event_set;
142
143	poll_check_ok(pop);
144
145	nfds = pop->nfds;
146
147#ifndef EVENT__DISABLE_THREAD_SUPPORT
148	if (base->th_base_lock) {
149		/* If we're using this backend in a multithreaded setting,
150		 * then we need to work on a copy of event_set, so that we can
151		 * let other threads modify the main event_set while we're
152		 * polling. If we're not multithreaded, then we'll skip the
153		 * copy step here to save memory and time. */
154		if (pop->realloc_copy) {
155			struct pollfd *tmp = mm_realloc(pop->event_set_copy,
156			    pop->event_count * sizeof(struct pollfd));
157			if (tmp == NULL) {
158				event_warn("realloc");
159				return -1;
160			}
161			pop->event_set_copy = tmp;
162			pop->realloc_copy = 0;
163		}
164		memcpy(pop->event_set_copy, pop->event_set,
165		    sizeof(struct pollfd)*nfds);
166		event_set = pop->event_set_copy;
167	} else {
168		event_set = pop->event_set;
169	}
170#else
171	event_set = pop->event_set;
172#endif
173
174	if (tv != NULL) {
175		msec = evutil_tv_to_msec_(tv);
176		if (msec < 0 || msec > INT_MAX)
177			msec = INT_MAX;
178	}
179
180	EVBASE_RELEASE_LOCK(base, th_base_lock);
181
182	res = poll(event_set, nfds, msec);
183
184	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
185
186	if (res == -1) {
187		if (errno != EINTR) {
188			event_warn("poll");
189			return (-1);
190		}
191
192		return (0);
193	}
194
195	event_debug(("%s: poll reports %d", __func__, res));
196
197	if (res == 0 || nfds == 0)
198		return (0);
199
200	i = evutil_weakrand_range_(&base->weakrand_seed, nfds);
201	for (j = 0; j < nfds; j++) {
202		int what;
203		if (++i == nfds)
204			i = 0;
205		what = event_set[i].revents;
206		if (!what)
207			continue;
208
209		res = 0;
210
211		/* If the file gets closed notify */
212		if (what & (POLLHUP|POLLERR|POLLNVAL))
213			what |= POLLIN|POLLOUT;
214		if (what & POLLIN)
215			res |= EV_READ;
216		if (what & POLLOUT)
217			res |= EV_WRITE;
218		if (what & POLLRDHUP)
219			res |= EV_CLOSED;
220		if (res == 0)
221			continue;
222
223		evmap_io_active_(base, event_set[i].fd, res);
224	}
225
226	return (0);
227}
228
229static int
230poll_add(struct event_base *base, int fd, short old, short events, void *idx_)
231{
232	struct pollop *pop = base->evbase;
233	struct pollfd *pfd = NULL;
234	struct pollidx *idx = idx_;
235	int i;
236
237	EVUTIL_ASSERT((events & EV_SIGNAL) == 0);
238	if (!(events & (EV_READ|EV_WRITE|EV_CLOSED)))
239		return (0);
240
241	poll_check_ok(pop);
242	if (pop->nfds + 1 >= pop->event_count) {
243		struct pollfd *tmp_event_set;
244		int tmp_event_count;
245
246		if (pop->event_count < 32)
247			tmp_event_count = 32;
248		else
249			tmp_event_count = pop->event_count * 2;
250
251		/* We need more file descriptors */
252		tmp_event_set = mm_realloc(pop->event_set,
253				 tmp_event_count * sizeof(struct pollfd));
254		if (tmp_event_set == NULL) {
255			event_warn("realloc");
256			return (-1);
257		}
258		pop->event_set = tmp_event_set;
259
260		pop->event_count = tmp_event_count;
261		pop->realloc_copy = 1;
262	}
263
264	i = idx->idxplus1 - 1;
265
266	if (i >= 0) {
267		pfd = &pop->event_set[i];
268	} else {
269		i = pop->nfds++;
270		pfd = &pop->event_set[i];
271		pfd->events = 0;
272		pfd->fd = fd;
273		idx->idxplus1 = i + 1;
274	}
275
276	pfd->revents = 0;
277	if (events & EV_WRITE)
278		pfd->events |= POLLOUT;
279	if (events & EV_READ)
280		pfd->events |= POLLIN;
281	if (events & EV_CLOSED)
282		pfd->events |= POLLRDHUP;
283	poll_check_ok(pop);
284
285	return (0);
286}
287
288/*
289 * Nothing to be done here.
290 */
291
292static int
293poll_del(struct event_base *base, int fd, short old, short events, void *idx_)
294{
295	struct pollop *pop = base->evbase;
296	struct pollfd *pfd = NULL;
297	struct pollidx *idx = idx_;
298	int i;
299
300	EVUTIL_ASSERT((events & EV_SIGNAL) == 0);
301	if (!(events & (EV_READ|EV_WRITE|EV_CLOSED)))
302		return (0);
303
304	poll_check_ok(pop);
305	i = idx->idxplus1 - 1;
306	if (i < 0)
307		return (-1);
308
309	/* Do we still want to read or write? */
310	pfd = &pop->event_set[i];
311	if (events & EV_READ)
312		pfd->events &= ~POLLIN;
313	if (events & EV_WRITE)
314		pfd->events &= ~POLLOUT;
315	if (events & EV_CLOSED)
316		pfd->events &= ~POLLRDHUP;
317	poll_check_ok(pop);
318	if (pfd->events)
319		/* Another event cares about that fd. */
320		return (0);
321
322	/* Okay, so we aren't interested in that fd anymore. */
323	idx->idxplus1 = 0;
324
325	--pop->nfds;
326	if (i != pop->nfds) {
327		/*
328		 * Shift the last pollfd down into the now-unoccupied
329		 * position.
330		 */
331		memcpy(&pop->event_set[i], &pop->event_set[pop->nfds],
332		       sizeof(struct pollfd));
333		idx = evmap_io_get_fdinfo_(&base->io, pop->event_set[i].fd);
334		EVUTIL_ASSERT(idx);
335		EVUTIL_ASSERT(idx->idxplus1 == pop->nfds + 1);
336		idx->idxplus1 = i + 1;
337	}
338
339	poll_check_ok(pop);
340	return (0);
341}
342
343static void
344poll_dealloc(struct event_base *base)
345{
346	struct pollop *pop = base->evbase;
347
348	evsig_dealloc_(base);
349	if (pop->event_set)
350		mm_free(pop->event_set);
351	if (pop->event_set_copy)
352		mm_free(pop->event_set_copy);
353
354	memset(pop, 0, sizeof(struct pollop));
355	mm_free(pop);
356}
357
358#endif /* EVENT__HAVE_POLL */
359