Home | History | Annotate | Line # | Download | only in libsockin
sockin.c revision 1.10 
      1 /*	$NetBSD: sockin.c,v 1.10 2009/01/26 13:44:51 pooka Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2008, 2009 Antti Kantee.  All Rights Reserved.
      5  *
      6  * Redistribution and use in source and binary forms, with or without
      7  * modification, are permitted provided that the following conditions
      8  * are met:
      9  * 1. Redistributions of source code must retain the above copyright
     10  *    notice, this list of conditions and the following disclaimer.
     11  * 2. Redistributions in binary form must reproduce the above copyright
     12  *    notice, this list of conditions and the following disclaimer in the
     13  *    documentation and/or other materials provided with the distribution.
     14  *
     15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
     16  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
     17  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
     18  * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
     21  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     25  * SUCH DAMAGE.
     26  */
     27 
     28 #include <sys/cdefs.h>
     29 __KERNEL_RCSID(0, "$NetBSD: sockin.c,v 1.10 2009/01/26 13:44:51 pooka Exp $");
     30 
     31 #include <sys/param.h>
     32 #include <sys/condvar.h>
     33 #include <sys/domain.h>
     34 #include <sys/kmem.h>
     35 #include <sys/kthread.h>
     36 #include <sys/mbuf.h>
     37 #include <sys/mutex.h>
     38 #include <sys/poll.h>
     39 #include <sys/protosw.h>
     40 #include <sys/queue.h>
     41 #include <sys/socket.h>
     42 #include <sys/socketvar.h>
     43 #include <sys/time.h>
     44 
     45 #include <net/radix.h>
     46 
     47 #include <netinet/in.h>
     48 #include <netinet/in_systm.h>
     49 #include <netinet/ip.h>
     50 
     51 #include <rump/rumpuser.h>
     52 
     53 #include "rump_private.h"
     54 
     55 /*
     56  * An inet communication domain which uses the socket interface.
     57  * Currently supports only IPv4 UDP, but could easily be extended to
     58  * support IPv6 and TCP by adding more stuff to the protosw.
     59  */
     60 
     61 DOMAIN_DEFINE(sockindomain);
     62 
     63 static void	sockin_init(void);
     64 static int	sockin_usrreq(struct socket *, int, struct mbuf *,
     65 			      struct mbuf *, struct mbuf *, struct lwp *);
     66 static int	sockin_ctloutput(int op, struct socket *, struct sockopt *);
     67 
     68 const struct protosw sockinsw[] = {
     69 {
     70 	.pr_type = SOCK_DGRAM,
     71 	.pr_domain = &sockindomain,
     72 	.pr_protocol = IPPROTO_UDP,
     73 	.pr_flags = PR_ATOMIC|PR_ADDR,
     74 	.pr_usrreq = sockin_usrreq,
     75 	.pr_ctloutput = sockin_ctloutput,
     76 },
     77 {
     78 	.pr_type = SOCK_STREAM,
     79 	.pr_domain = &sockindomain,
     80 	.pr_protocol = IPPROTO_TCP,
     81 	.pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_LISTEN|PR_ABRTACPTDIS,
     82 	.pr_usrreq = sockin_usrreq,
     83 	.pr_ctloutput = sockin_ctloutput,
     84 }};
     85 
     86 struct domain sockindomain = {
     87 	.dom_family = PF_INET,
     88 	.dom_name = "socket_inet",
     89 	.dom_init = sockin_init,
     90 	.dom_externalize = NULL,
     91 	.dom_dispose = NULL,
     92 	.dom_protosw = sockinsw,
     93 	.dom_protoswNPROTOSW = &sockinsw[__arraycount(sockinsw)],
     94 	.dom_rtattach = rn_inithead,
     95 	.dom_rtoffset = 32,
     96 	.dom_maxrtkey = sizeof(struct sockaddr_in),
     97 	.dom_ifattach = NULL,
     98 	.dom_ifdetach = NULL,
     99 	.dom_ifqueues = { NULL },
    100 	.dom_link = { NULL },
    101 	.dom_mowner = MOWNER_INIT("",""),
    102 	.dom_rtcache = { NULL },
    103 	.dom_sockaddr_cmp = NULL
    104 };
    105 
    106 #define SO2S(so) ((intptr_t)(so->so_internal))
    107 #define SOCKIN_SBSIZE 65536
    108 
    109 struct sockin_unit {
    110 	struct socket *su_so;
    111 
    112 	LIST_ENTRY(sockin_unit) su_entries;
    113 };
    114 static LIST_HEAD(, sockin_unit) su_ent = LIST_HEAD_INITIALIZER(su_ent);
    115 static kmutex_t su_mtx;
    116 static bool rebuild;
    117 static int nsock;
    118 
    119 static int
    120 registersock(struct socket *so, int news)
    121 {
    122 	struct sockin_unit *su;
    123 
    124 	su = kmem_alloc(sizeof(*su), KM_NOSLEEP);
    125 	if (!su)
    126 		return ENOMEM;
    127 
    128 	so->so_internal = (void *)(intptr_t)news;
    129 	su->su_so = so;
    130 
    131 	mutex_enter(&su_mtx);
    132 	LIST_INSERT_HEAD(&su_ent, su, su_entries);
    133 	nsock++;
    134 	rebuild = true;
    135 	mutex_exit(&su_mtx);
    136 
    137 	return 0;
    138 }
    139 
    140 static void
    141 removesock(struct socket *so)
    142 {
    143 	struct sockin_unit *su_iter;
    144 	int error;
    145 
    146 	mutex_enter(&su_mtx);
    147 	LIST_FOREACH(su_iter, &su_ent, su_entries) {
    148 		if (su_iter->su_so == so)
    149 			break;
    150 	}
    151 	if (!su_iter)
    152 		panic("no such socket");
    153 
    154 	LIST_REMOVE(su_iter, su_entries);
    155 	nsock--;
    156 	rebuild = true;
    157 	mutex_exit(&su_mtx);
    158 
    159 	rumpuser_close(SO2S(su_iter->su_so), &error);
    160 	kmem_free(su_iter, sizeof(*su_iter));
    161 }
    162 
    163 static void
    164 sockin_process(struct socket *so)
    165 {
    166 	struct sockaddr_in from;
    167 	struct iovec io;
    168 	struct msghdr rmsg;
    169 	struct mbuf *m;
    170 	ssize_t n;
    171 	size_t plen;
    172 	int error;
    173 
    174 	plen = IP_MAXPACKET;
    175 	m = m_gethdr(M_WAIT, MT_DATA);
    176 	MEXTMALLOC(m, plen, M_WAIT);
    177 
    178 	memset(&rmsg, 0, sizeof(rmsg));
    179 	io.iov_base = mtod(m, void *);
    180 	io.iov_len = plen;
    181 	rmsg.msg_iov = &io;
    182 	rmsg.msg_iovlen = 1;
    183 	rmsg.msg_name = (struct sockaddr *)&from;
    184 	rmsg.msg_namelen = sizeof(from);
    185 
    186 	n = rumpuser_net_recvmsg(SO2S(so), &rmsg, 0, &error);
    187 	if (n <= 0) {
    188 		m_freem(m);
    189 
    190 		/* Treat a TCP socket a goner */
    191 		if (so->so_proto->pr_type == SOCK_STREAM && error != EAGAIN) {
    192 			soisdisconnected(so);
    193 			removesock(so);
    194 		}
    195 		return;
    196 	}
    197 	m->m_len = m->m_pkthdr.len = n;
    198 
    199 	if (so->so_proto->pr_type == SOCK_DGRAM) {
    200 		if (!sbappendaddr(&so->so_rcv, rmsg.msg_name, m, NULL)) {
    201 			m_freem(m);
    202 		}
    203 	} else {
    204 		sbappendstream(&so->so_rcv, m);
    205 	}
    206 
    207 	sorwakeup(so);
    208 }
    209 
    210 static void
    211 sockin_accept(struct socket *so)
    212 {
    213 	struct socket *nso;
    214 	struct sockaddr_in sin;
    215 	int news, error, slen;
    216 
    217 	slen = sizeof(sin);
    218 	news = rumpuser_net_accept(SO2S(so), (struct sockaddr *)&sin,
    219 	    &slen, &error);
    220 	if (news == -1)
    221 		return;
    222 
    223 	if ((nso = sonewconn(so, SS_ISCONNECTED)) == NULL)
    224 		goto errout;
    225 	if (registersock(nso, news) != 0)
    226 		goto errout;
    227 	return;
    228 
    229  errout:
    230 	rumpuser_close(news, &error);
    231 	if (nso)
    232 		soclose(nso);
    233 }
    234 
    235 #define POLLTIMEOUT 100	/* check for new entries every 100ms */
    236 
    237 /* XXX: doesn't handle socket (kernel) locking properly? */
    238 static void
    239 sockinworker(void *arg)
    240 {
    241 	struct pollfd *pfds = NULL, *npfds;
    242 	struct sockin_unit *su_iter;
    243 	struct socket *so;
    244 	int cursock = 0, i, rv, error;
    245 
    246 	/*
    247 	 * Loop reading requests.  Check for new sockets periodically
    248 	 * (could be smarter, but I'm lazy).
    249 	 */
    250 	for (;;) {
    251 		if (rebuild) {
    252 			npfds = NULL;
    253 			mutex_enter(&su_mtx);
    254 			if (nsock)
    255 				npfds = kmem_alloc(nsock * sizeof(*npfds),
    256 				    KM_NOSLEEP);
    257 			if (npfds || nsock == 0) {
    258 				if (pfds)
    259 					kmem_free(pfds, cursock*sizeof(*pfds));
    260 				pfds = npfds;
    261 				cursock = nsock;
    262 				rebuild = false;
    263 
    264 				i = 0;
    265 				LIST_FOREACH(su_iter, &su_ent, su_entries) {
    266 					pfds[i].fd = SO2S(su_iter->su_so);
    267 					pfds[i].events = POLLIN;
    268 					pfds[i].revents = 0;
    269 					i++;
    270 				}
    271 				KASSERT(i == nsock);
    272 			}
    273 			mutex_exit(&su_mtx);
    274 		}
    275 
    276 		/* find affected sockets & process */
    277 		rv = rumpuser_poll(pfds, cursock, POLLTIMEOUT, &error);
    278 		for (i = 0; i < cursock && rv > 0; i++) {
    279 			if (pfds[i].revents & POLLIN) {
    280 				mutex_enter(&su_mtx);
    281 				LIST_FOREACH(su_iter, &su_ent, su_entries) {
    282 					if (SO2S(su_iter->su_so)==pfds[i].fd) {
    283 						so = su_iter->su_so;
    284 						mutex_exit(&su_mtx);
    285 						mutex_enter(softnet_lock);
    286 						if(so->so_options&SO_ACCEPTCONN)
    287 							sockin_accept(so);
    288 						else
    289 							sockin_process(so);
    290 						mutex_exit(softnet_lock);
    291 						mutex_enter(&su_mtx);
    292 						break;
    293 					}
    294 				}
    295 				/* if we can't find it, just wing it */
    296 				KASSERT(rebuild || su_iter);
    297 				mutex_exit(&su_mtx);
    298 				pfds[i].revents = 0;
    299 				rv--;
    300 				i = -1;
    301 				continue;
    302 			}
    303 
    304 			/* something else?  ignore */
    305 			if (pfds[i].revents) {
    306 				pfds[i].revents = 0;
    307 				rv--;
    308 			}
    309 		}
    310 		KASSERT(rv <= 0);
    311 	}
    312 
    313 }
    314 
    315 static void
    316 sockin_init()
    317 {
    318 	int rv;
    319 
    320 	if (rump_threads) {
    321 		if ((rv = kthread_create(PRI_NONE, 0, NULL, sockinworker,
    322 		    NULL, NULL, "sockwork")) != 0)
    323 			panic("sockin_init: could not create worker thread\n");
    324 	} else {
    325 		printf("sockin_init: no threads => no worker thread\n");
    326 	}
    327 	mutex_init(&su_mtx, MUTEX_DEFAULT, IPL_NONE);
    328 }
    329 
    330 static int
    331 sockin_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam,
    332 	struct mbuf *control, struct lwp *l)
    333 {
    334 	int error = 0, rv;
    335 
    336 	switch (req) {
    337 	case PRU_ATTACH:
    338 	{
    339 		int news;
    340 
    341 		sosetlock(so);
    342 		if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
    343 			error = soreserve(so, SOCKIN_SBSIZE, SOCKIN_SBSIZE);
    344 			if (error)
    345 				break;
    346 		}
    347 
    348 		news = rumpuser_net_socket(PF_INET, so->so_proto->pr_type,
    349 		    0, &error);
    350 		if (news == -1)
    351 			break;
    352 
    353 		if ((error = registersock(so, news)) != 0)
    354 			rumpuser_close(news, &error);
    355 
    356 		break;
    357 	}
    358 
    359 	case PRU_ACCEPT:
    360 		/* we do all the work in the worker thread */
    361 		break;
    362 
    363 	case PRU_BIND:
    364 		rumpuser_net_bind(SO2S(so), mtod(nam, const struct sockaddr *),
    365 		    sizeof(struct sockaddr_in), &error);
    366 		break;
    367 
    368 	case PRU_CONNECT:
    369 		/* don't bother to connect udp sockets, always sendmsg */
    370 		if (so->so_proto->pr_type == SOCK_DGRAM)
    371 			break;
    372 
    373 		rv = rumpuser_net_connect(SO2S(so),
    374 		    mtod(nam, struct sockaddr *), sizeof(struct sockaddr_in),
    375 		    &error);
    376 		if (rv == 0)
    377 			soisconnected(so);
    378 		break;
    379 
    380 	case PRU_LISTEN:
    381 		rumpuser_net_listen(SO2S(so), so->so_qlimit, &error);
    382 		break;
    383 
    384 	case PRU_SEND:
    385 	{
    386 		struct sockaddr *saddr;
    387 		struct msghdr mhdr;
    388 		struct iovec iov[16];
    389 		struct mbuf *m2;
    390 		size_t tot;
    391 		int i, s;
    392 
    393 		memset(&mhdr, 0, sizeof(mhdr));
    394 
    395 		tot = 0;
    396 		for (i = 0, m2 = m; m2; m2 = m2->m_next, i++) {
    397 			if (i > 16)
    398 				panic("lazy bum");
    399 			iov[i].iov_base = m2->m_data;
    400 			iov[i].iov_len = m2->m_len;
    401 			tot += m2->m_len;
    402 
    403 		}
    404 		mhdr.msg_iov = iov;
    405 		mhdr.msg_iovlen = i;
    406 		s = SO2S(so);
    407 
    408 		if (so->so_proto->pr_type == SOCK_DGRAM) {
    409 			saddr = mtod(nam, struct sockaddr *);
    410 			mhdr.msg_name = saddr;
    411 			mhdr.msg_namelen = saddr->sa_len;
    412 		}
    413 
    414 		rumpuser_net_sendmsg(s, &mhdr, 0, &error);
    415 
    416 		m_freem(m);
    417 		m_freem(control);
    418 
    419 		/* this assumes too many things to list.. buthey, testing */
    420 		if (!rump_threads)
    421 			sockin_process(so);
    422 	}
    423 		break;
    424 
    425 	case PRU_SHUTDOWN:
    426 		removesock(so);
    427 		break;
    428 
    429 	case PRU_SOCKADDR:
    430 	case PRU_PEERADDR:
    431 	{
    432 		int slen = nam->m_len;
    433 		enum rumpuser_getnametype which;
    434 
    435 		if (req == PRU_SOCKADDR)
    436 			which = RUMPUSER_SOCKNAME;
    437 		else
    438 			which = RUMPUSER_PEERNAME;
    439 		rumpuser_net_getname(SO2S(so),
    440 		    mtod(nam, struct sockaddr *), &slen, which, &error);
    441 		if (error == 0)
    442 			nam->m_len = slen;
    443 		break;
    444 	}
    445 
    446 	default:
    447 		panic("sockin_usrreq: IMPLEMENT ME, req %d not supported", req);
    448 	}
    449 
    450 	return error;
    451 }
    452 
    453 static int
    454 sockin_ctloutput(int op, struct socket *so, struct sockopt *sopt)
    455 {
    456 
    457 	/* XXX: we should also do something here */
    458 	return 0;
    459 }
    460