michael@0: /*	$OpenBSD: select.c,v 1.2 2002/06/25 15:50:15 mickey Exp $	*/
michael@0: 
michael@0: /*
michael@0:  * Copyright 2000-2007 Niels Provos <provos@citi.umich.edu>
michael@0:  * Copyright 2007-2012 Niels Provos and Nick Mathewson
michael@0:  *
michael@0:  * Redistribution and use in source and binary forms, with or without
michael@0:  * modification, are permitted provided that the following conditions
michael@0:  * are met:
michael@0:  * 1. Redistributions of source code must retain the above copyright
michael@0:  *    notice, this list of conditions and the following disclaimer.
michael@0:  * 2. Redistributions in binary form must reproduce the above copyright
michael@0:  *    notice, this list of conditions and the following disclaimer in the
michael@0:  *    documentation and/or other materials provided with the distribution.
michael@0:  * 3. The name of the author may not be used to endorse or promote products
michael@0:  *    derived from this software without specific prior written permission.
michael@0:  *
michael@0:  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
michael@0:  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
michael@0:  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
michael@0:  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
michael@0:  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
michael@0:  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
michael@0:  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
michael@0:  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
michael@0:  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
michael@0:  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
michael@0:  */
michael@0: #include "event2/event-config.h"
michael@0: 
michael@0: #include <sys/types.h>
michael@0: #ifdef _EVENT_HAVE_SYS_TIME_H
michael@0: #include <sys/time.h>
michael@0: #endif
michael@0: #ifdef _EVENT_HAVE_SYS_SELECT_H
michael@0: #include <sys/select.h>
michael@0: #endif
michael@0: #include <sys/queue.h>
michael@0: #include <signal.h>
michael@0: #include <stdio.h>
michael@0: #include <stdlib.h>
michael@0: #include <string.h>
michael@0: #include <unistd.h>
michael@0: #include <errno.h>
michael@0: 
michael@0: #include "event-internal.h"
michael@0: #include "evsignal-internal.h"
michael@0: #include "event2/thread.h"
michael@0: #include "evthread-internal.h"
michael@0: #include "log-internal.h"
michael@0: #include "evmap-internal.h"
michael@0: 
michael@0: #ifndef _EVENT_HAVE_FD_MASK
michael@0: /* This type is mandatory, but Android doesn't define it. */
michael@0: typedef unsigned long fd_mask;
michael@0: #endif
michael@0: 
michael@0: #ifndef NFDBITS
michael@0: #define NFDBITS (sizeof(fd_mask)*8)
michael@0: #endif
michael@0: 
michael@0: /* Divide positive x by y, rounding up. */
michael@0: #define DIV_ROUNDUP(x, y)   (((x)+((y)-1))/(y))
michael@0: 
michael@0: /* How many bytes to allocate for N fds? */
michael@0: #define SELECT_ALLOC_SIZE(n) \
michael@0: 	(DIV_ROUNDUP(n, NFDBITS) * sizeof(fd_mask))
michael@0: 
michael@0: struct selectop {
michael@0: 	int event_fds;		/* Highest fd in fd set */
michael@0: 	int event_fdsz;
michael@0: 	int resize_out_sets;
michael@0: 	fd_set *event_readset_in;
michael@0: 	fd_set *event_writeset_in;
michael@0: 	fd_set *event_readset_out;
michael@0: 	fd_set *event_writeset_out;
michael@0: };
michael@0: 
michael@0: static void *select_init(struct event_base *);
michael@0: static int select_add(struct event_base *, int, short old, short events, void*);
michael@0: static int select_del(struct event_base *, int, short old, short events, void*);
michael@0: static int select_dispatch(struct event_base *, struct timeval *);
michael@0: static void select_dealloc(struct event_base *);
michael@0: 
michael@0: const struct eventop selectops = {
michael@0: 	"select",
michael@0: 	select_init,
michael@0: 	select_add,
michael@0: 	select_del,
michael@0: 	select_dispatch,
michael@0: 	select_dealloc,
michael@0: 	0, /* doesn't need reinit. */
michael@0: 	EV_FEATURE_FDS,
michael@0: 	0,
michael@0: };
michael@0: 
michael@0: static int select_resize(struct selectop *sop, int fdsz);
michael@0: static void select_free_selectop(struct selectop *sop);
michael@0: 
michael@0: static void *
michael@0: select_init(struct event_base *base)
michael@0: {
michael@0: 	struct selectop *sop;
michael@0: 
michael@0: 	if (!(sop = mm_calloc(1, sizeof(struct selectop))))
michael@0: 		return (NULL);
michael@0: 
michael@0: 	if (select_resize(sop, SELECT_ALLOC_SIZE(32 + 1))) {
michael@0: 		select_free_selectop(sop);
michael@0: 		return (NULL);
michael@0: 	}
michael@0: 
michael@0: 	evsig_init(base);
michael@0: 
michael@0: 	return (sop);
michael@0: }
michael@0: 
michael@0: #ifdef CHECK_INVARIANTS
michael@0: static void
michael@0: check_selectop(struct selectop *sop)
michael@0: {
michael@0: 	/* nothing to be done here */
michael@0: }
michael@0: #else
michael@0: #define check_selectop(sop) do { (void) sop; } while (0)
michael@0: #endif
michael@0: 
michael@0: static int
michael@0: select_dispatch(struct event_base *base, struct timeval *tv)
michael@0: {
michael@0: 	int res=0, i, j, nfds;
michael@0: 	struct selectop *sop = base->evbase;
michael@0: 
michael@0: 	check_selectop(sop);
michael@0: 	if (sop->resize_out_sets) {
michael@0: 		fd_set *readset_out=NULL, *writeset_out=NULL;
michael@0: 		size_t sz = sop->event_fdsz;
michael@0: 		if (!(readset_out = mm_realloc(sop->event_readset_out, sz)))
michael@0: 			return (-1);
michael@0: 		sop->event_readset_out = readset_out;
michael@0: 		if (!(writeset_out = mm_realloc(sop->event_writeset_out, sz))) {
michael@0: 			/* We don't free readset_out here, since it was
michael@0: 			 * already successfully reallocated. The next time
michael@0: 			 * we call select_dispatch, the realloc will be a
michael@0: 			 * no-op. */
michael@0: 			return (-1);
michael@0: 		}
michael@0: 		sop->event_writeset_out = writeset_out;
michael@0: 		sop->resize_out_sets = 0;
michael@0: 	}
michael@0: 
michael@0: 	memcpy(sop->event_readset_out, sop->event_readset_in,
michael@0: 	       sop->event_fdsz);
michael@0: 	memcpy(sop->event_writeset_out, sop->event_writeset_in,
michael@0: 	       sop->event_fdsz);
michael@0: 
michael@0: 	nfds = sop->event_fds+1;
michael@0: 
michael@0: 	EVBASE_RELEASE_LOCK(base, th_base_lock);
michael@0: 
michael@0: 	res = select(nfds, sop->event_readset_out,
michael@0: 	    sop->event_writeset_out, NULL, tv);
michael@0: 
michael@0: 	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
michael@0: 
michael@0: 	check_selectop(sop);
michael@0: 
michael@0: 	if (res == -1) {
michael@0: 		if (errno != EINTR) {
michael@0: 			event_warn("select");
michael@0: 			return (-1);
michael@0: 		}
michael@0: 
michael@0: 		return (0);
michael@0: 	}
michael@0: 
michael@0: 	event_debug(("%s: select reports %d", __func__, res));
michael@0: 
michael@0: 	check_selectop(sop);
michael@0: 	i = random() % nfds;
michael@0: 	for (j = 0; j < nfds; ++j) {
michael@0: 		if (++i >= nfds)
michael@0: 			i = 0;
michael@0: 		res = 0;
michael@0: 		if (FD_ISSET(i, sop->event_readset_out))
michael@0: 			res |= EV_READ;
michael@0: 		if (FD_ISSET(i, sop->event_writeset_out))
michael@0: 			res |= EV_WRITE;
michael@0: 
michael@0: 		if (res == 0)
michael@0: 			continue;
michael@0: 
michael@0: 		evmap_io_active(base, i, res);
michael@0: 	}
michael@0: 	check_selectop(sop);
michael@0: 
michael@0: 	return (0);
michael@0: }
michael@0: 
michael@0: static int
michael@0: select_resize(struct selectop *sop, int fdsz)
michael@0: {
michael@0: 	fd_set *readset_in = NULL;
michael@0: 	fd_set *writeset_in = NULL;
michael@0: 
michael@0: 	if (sop->event_readset_in)
michael@0: 		check_selectop(sop);
michael@0: 
michael@0: 	if ((readset_in = mm_realloc(sop->event_readset_in, fdsz)) == NULL)
michael@0: 		goto error;
michael@0: 	sop->event_readset_in = readset_in;
michael@0: 	if ((writeset_in = mm_realloc(sop->event_writeset_in, fdsz)) == NULL) {
michael@0: 		/* Note that this will leave event_readset_in expanded.
michael@0: 		 * That's okay; we wouldn't want to free it, since that would
michael@0: 		 * change the semantics of select_resize from "expand the
michael@0: 		 * readset_in and writeset_in, or return -1" to "expand the
michael@0: 		 * *set_in members, or trash them and return -1."
michael@0: 		 */
michael@0: 		goto error;
michael@0: 	}
michael@0: 	sop->event_writeset_in = writeset_in;
michael@0: 	sop->resize_out_sets = 1;
michael@0: 
michael@0: 	memset((char *)sop->event_readset_in + sop->event_fdsz, 0,
michael@0: 	    fdsz - sop->event_fdsz);
michael@0: 	memset((char *)sop->event_writeset_in + sop->event_fdsz, 0,
michael@0: 	    fdsz - sop->event_fdsz);
michael@0: 
michael@0: 	sop->event_fdsz = fdsz;
michael@0: 	check_selectop(sop);
michael@0: 
michael@0: 	return (0);
michael@0: 
michael@0:  error:
michael@0: 	event_warn("malloc");
michael@0: 	return (-1);
michael@0: }
michael@0: 
michael@0: 
michael@0: static int
michael@0: select_add(struct event_base *base, int fd, short old, short events, void *p)
michael@0: {
michael@0: 	struct selectop *sop = base->evbase;
michael@0: 	(void) p;
michael@0: 
michael@0: 	EVUTIL_ASSERT((events & EV_SIGNAL) == 0);
michael@0: 	check_selectop(sop);
michael@0: 	/*
michael@0: 	 * Keep track of the highest fd, so that we can calculate the size
michael@0: 	 * of the fd_sets for select(2)
michael@0: 	 */
michael@0: 	if (sop->event_fds < fd) {
michael@0: 		int fdsz = sop->event_fdsz;
michael@0: 
michael@0: 		if (fdsz < (int)sizeof(fd_mask))
michael@0: 			fdsz = (int)sizeof(fd_mask);
michael@0: 
michael@0: 		/* In theory we should worry about overflow here.  In
michael@0: 		 * reality, though, the highest fd on a unixy system will
michael@0: 		 * not overflow here. XXXX */
michael@0: 		while (fdsz < (int) SELECT_ALLOC_SIZE(fd + 1))
michael@0: 			fdsz *= 2;
michael@0: 
michael@0: 		if (fdsz != sop->event_fdsz) {
michael@0: 			if (select_resize(sop, fdsz)) {
michael@0: 				check_selectop(sop);
michael@0: 				return (-1);
michael@0: 			}
michael@0: 		}
michael@0: 
michael@0: 		sop->event_fds = fd;
michael@0: 	}
michael@0: 
michael@0: 	if (events & EV_READ)
michael@0: 		FD_SET(fd, sop->event_readset_in);
michael@0: 	if (events & EV_WRITE)
michael@0: 		FD_SET(fd, sop->event_writeset_in);
michael@0: 	check_selectop(sop);
michael@0: 
michael@0: 	return (0);
michael@0: }
michael@0: 
michael@0: /*
michael@0:  * Nothing to be done here.
michael@0:  */
michael@0: 
michael@0: static int
michael@0: select_del(struct event_base *base, int fd, short old, short events, void *p)
michael@0: {
michael@0: 	struct selectop *sop = base->evbase;
michael@0: 	(void)p;
michael@0: 
michael@0: 	EVUTIL_ASSERT((events & EV_SIGNAL) == 0);
michael@0: 	check_selectop(sop);
michael@0: 
michael@0: 	if (sop->event_fds < fd) {
michael@0: 		check_selectop(sop);
michael@0: 		return (0);
michael@0: 	}
michael@0: 
michael@0: 	if (events & EV_READ)
michael@0: 		FD_CLR(fd, sop->event_readset_in);
michael@0: 
michael@0: 	if (events & EV_WRITE)
michael@0: 		FD_CLR(fd, sop->event_writeset_in);
michael@0: 
michael@0: 	check_selectop(sop);
michael@0: 	return (0);
michael@0: }
michael@0: 
michael@0: static void
michael@0: select_free_selectop(struct selectop *sop)
michael@0: {
michael@0: 	if (sop->event_readset_in)
michael@0: 		mm_free(sop->event_readset_in);
michael@0: 	if (sop->event_writeset_in)
michael@0: 		mm_free(sop->event_writeset_in);
michael@0: 	if (sop->event_readset_out)
michael@0: 		mm_free(sop->event_readset_out);
michael@0: 	if (sop->event_writeset_out)
michael@0: 		mm_free(sop->event_writeset_out);
michael@0: 
michael@0: 	memset(sop, 0, sizeof(struct selectop));
michael@0: 	mm_free(sop);
michael@0: }
michael@0: 
michael@0: static void
michael@0: select_dealloc(struct event_base *base)
michael@0: {
michael@0: 	evsig_dealloc(base);
michael@0: 
michael@0: 	select_free_selectop(base->evbase);
michael@0: }