Home | History | Annotate | Line # | Download | only in libsockin
sockin.c revision 1.9 
      1 /*	$NetBSD: sockin.c,v 1.9 2009/01/26 10:43:21 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 1.9 2009/01/26 10:43:21 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 sockin_process(struct socket *so)
    142 {
    143 	struct sockaddr_in from;
    144 	struct iovec io;
    145 	struct msghdr rmsg;
    146 	struct mbuf *m;
    147 	ssize_t n;
    148 	size_t plen;
    149 	int error;
    150 
    151 	plen = IP_MAXPACKET;
    152 	m = m_gethdr(M_WAIT, MT_DATA);
    153 	MEXTMALLOC(m, plen, M_WAIT);
    154 
    155 	memset(&rmsg, 0, sizeof(rmsg));
    156 	io.iov_base = mtod(m, void *);
    157 	io.iov_len = plen;
    158 	rmsg.msg_iov = &io;
    159 	rmsg.msg_iovlen = 1;
    160 	rmsg.msg_name = (struct sockaddr *)&from;
    161 	rmsg.msg_namelen = sizeof(from);
    162 
    163 	n = rumpuser_net_recvmsg(SO2S(so), &rmsg, 0, &error);
    164 	if (n <= 0) {
    165 		m_freem(m);
    166 		return;
    167 	}
    168 	m->m_len = m->m_pkthdr.len = n;
    169 
    170 	if (so->so_proto->pr_type == SOCK_DGRAM) {
    171 		if (!sbappendaddr(&so->so_rcv, rmsg.msg_name, m, NULL)) {
    172 			m_freem(m);
    173 		}
    174 	} else {
    175 		sbappendstream(&so->so_rcv, m);
    176 	}
    177 
    178 	sorwakeup(so);
    179 }
    180 
    181 static void
    182 sockin_accept(struct socket *so)
    183 {
    184 	struct socket *nso;
    185 	struct sockaddr_in sin;
    186 	int news, error, slen;
    187 
    188 	slen = sizeof(sin);
    189 	news = rumpuser_net_accept(SO2S(so), (struct sockaddr *)&sin,
    190 	    &slen, &error);
    191 	if (news == -1)
    192 		return;
    193 
    194 	if ((nso = sonewconn(so, SS_ISCONNECTED)) == NULL)
    195 		goto errout;
    196 	if (registersock(nso, news) != 0)
    197 		goto errout;
    198 	return;
    199 
    200  errout:
    201 	rumpuser_close(news, &error);
    202 	if (nso)
    203 		soclose(nso);
    204 }
    205 
    206 #define POLLTIMEOUT 100	/* check for new entries every 100ms */
    207 
    208 /* XXX: doesn't handle socket (kernel) locking properly? */
    209 static void
    210 sockinworker(void *arg)
    211 {
    212 	struct pollfd *pfds = NULL, *npfds;
    213 	struct sockin_unit *su_iter;
    214 	struct socket *so;
    215 	int cursock = 0, i, rv, error;
    216 
    217 	/*
    218 	 * Loop reading requests.  Check for new sockets periodically
    219 	 * (could be smarter, but I'm lazy).
    220 	 */
    221 	for (;;) {
    222 		if (rebuild) {
    223 			npfds = NULL;
    224 			mutex_enter(&su_mtx);
    225 			if (nsock)
    226 				npfds = kmem_alloc(nsock * sizeof(*npfds),
    227 				    KM_NOSLEEP);
    228 			if (npfds || nsock == 0) {
    229 				if (pfds)
    230 					kmem_free(pfds, cursock*sizeof(*pfds));
    231 				pfds = npfds;
    232 				cursock = nsock;
    233 				rebuild = false;
    234 
    235 				i = 0;
    236 				LIST_FOREACH(su_iter, &su_ent, su_entries) {
    237 					pfds[i].fd = SO2S(su_iter->su_so);
    238 					pfds[i].events = POLLIN;
    239 					pfds[i].revents = 0;
    240 					i++;
    241 				}
    242 				KASSERT(i == nsock);
    243 			}
    244 			mutex_exit(&su_mtx);
    245 		}
    246 
    247 		/* find affected sockets & process */
    248 		rv = rumpuser_poll(pfds, cursock, POLLTIMEOUT, &error);
    249 		for (i = 0; i < cursock && rv > 0; i++) {
    250 			if (pfds[i].revents & POLLIN) {
    251 				mutex_enter(&su_mtx);
    252 				LIST_FOREACH(su_iter, &su_ent, su_entries) {
    253 					if (SO2S(su_iter->su_so)==pfds[i].fd) {
    254 						so = su_iter->su_so;
    255 						mutex_exit(&su_mtx);
    256 						mutex_enter(softnet_lock);
    257 						if(so->so_options&SO_ACCEPTCONN)
    258 							sockin_accept(so);
    259 						else
    260 							sockin_process(so);
    261 						mutex_exit(softnet_lock);
    262 						mutex_enter(&su_mtx);
    263 						break;
    264 					}
    265 				}
    266 				/* if we can't find it, just wing it */
    267 				KASSERT(rebuild || su_iter);
    268 				mutex_exit(&su_mtx);
    269 				pfds[i].revents = 0;
    270 				rv--;
    271 				i = -1;
    272 				continue;
    273 			}
    274 
    275 			/* something else?  ignore */
    276 			if (pfds[i].revents) {
    277 				pfds[i].revents = 0;
    278 				rv--;
    279 			}
    280 		}
    281 		KASSERT(rv <= 0);
    282 	}
    283 
    284 }
    285 
    286 static void
    287 sockin_init()
    288 {
    289 	int rv;
    290 
    291 	if (rump_threads) {
    292 		if ((rv = kthread_create(PRI_NONE, 0, NULL, sockinworker,
    293 		    NULL, NULL, "sockwork")) != 0)
    294 			panic("sockin_init: could not create worker thread\n");
    295 	} else {
    296 		printf("sockin_init: no threads => no worker thread\n");
    297 	}
    298 	mutex_init(&su_mtx, MUTEX_DEFAULT, IPL_NONE);
    299 }
    300 
    301 static int
    302 sockin_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam,
    303 	struct mbuf *control, struct lwp *l)
    304 {
    305 	int error = 0, rv;
    306 
    307 	switch (req) {
    308 	case PRU_ATTACH:
    309 	{
    310 		int news;
    311 
    312 		sosetlock(so);
    313 		if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
    314 			error = soreserve(so, SOCKIN_SBSIZE, SOCKIN_SBSIZE);
    315 			if (error)
    316 				break;
    317 		}
    318 
    319 		news = rumpuser_net_socket(PF_INET, so->so_proto->pr_type,
    320 		    0, &error);
    321 		if (news == -1)
    322 			break;
    323 
    324 		if ((error = registersock(so, news)) != 0)
    325 			rumpuser_close(news, &error);
    326 
    327 		break;
    328 	}
    329 
    330 	case PRU_ACCEPT:
    331 		/* we do all the work in the worker thread */
    332 		break;
    333 
    334 	case PRU_BIND:
    335 		rumpuser_net_bind(SO2S(so), mtod(nam, const struct sockaddr *),
    336 		    sizeof(struct sockaddr_in), &error);
    337 		break;
    338 
    339 	case PRU_CONNECT:
    340 		/* don't bother to connect udp sockets, always sendmsg */
    341 		if (so->so_proto->pr_type == SOCK_DGRAM)
    342 			break;
    343 
    344 		rv = rumpuser_net_connect(SO2S(so),
    345 		    mtod(nam, struct sockaddr *), sizeof(struct sockaddr_in),
    346 		    &error);
    347 		if (rv == 0)
    348 		soisconnected(so);
    349 		break;
    350 
    351 	case PRU_LISTEN:
    352 		rumpuser_net_listen(SO2S(so), so->so_qlimit, &error);
    353 		break;
    354 
    355 	case PRU_SEND:
    356 	{
    357 		struct sockaddr *saddr;
    358 		struct msghdr mhdr;
    359 		struct iovec iov[16];
    360 		struct mbuf *m2;
    361 		size_t tot;
    362 		int i, s;
    363 
    364 		memset(&mhdr, 0, sizeof(mhdr));
    365 
    366 		tot = 0;
    367 		for (i = 0, m2 = m; m2; m2 = m2->m_next, i++) {
    368 			if (i > 16)
    369 				panic("lazy bum");
    370 			iov[i].iov_base = m2->m_data;
    371 			iov[i].iov_len = m2->m_len;
    372 			tot += m2->m_len;
    373 
    374 		}
    375 		mhdr.msg_iov = iov;
    376 		mhdr.msg_iovlen = i;
    377 		s = SO2S(so);
    378 
    379 		if (so->so_proto->pr_type == SOCK_DGRAM) {
    380 			saddr = mtod(nam, struct sockaddr *);
    381 			mhdr.msg_name = saddr;
    382 			mhdr.msg_namelen = saddr->sa_len;
    383 		}
    384 
    385 		rumpuser_net_sendmsg(s, &mhdr, 0, &error);
    386 
    387 		m_freem(m);
    388 		m_freem(control);
    389 
    390 		/* this assumes too many things to list.. buthey, testing */
    391 		if (!rump_threads)
    392 			sockin_process(so);
    393 	}
    394 		break;
    395 
    396 	case PRU_SHUTDOWN:
    397 	{
    398 		struct sockin_unit *su_iter;
    399 
    400 		mutex_enter(&su_mtx);
    401 		LIST_FOREACH(su_iter, &su_ent, su_entries) {
    402 			if (su_iter->su_so == so)
    403 				break;
    404 		}
    405 		if (!su_iter)
    406 			panic("no such socket");
    407 
    408 		LIST_REMOVE(su_iter, su_entries);
    409 		nsock--;
    410 		rebuild = true;
    411 		mutex_exit(&su_mtx);
    412 
    413 		rumpuser_close(SO2S(su_iter->su_so), &error);
    414 		kmem_free(su_iter, sizeof(*su_iter));
    415 	}
    416 		break;
    417 
    418 	default:
    419 		panic("sockin_usrreq: IMPLEMENT ME, req %d not supported", req);
    420 	}
    421 
    422 	return error;
    423 }
    424 
    425 static int
    426 sockin_ctloutput(int op, struct socket *so, struct sockopt *sopt)
    427 {
    428 
    429 	/* XXX: we should also do something here */
    430 	return 0;
    431 }
    432