Home | History | Annotate | Line # | Download | only in libsockin
sockin.c revision 1.32 
      1 /*	$NetBSD: sockin.c,v 1.32 2013/06/01 11:44:37 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 1.32 2013/06/01 11:44:37 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/bpf.h>
     46 #include <net/if.h>
     47 #include <net/radix.h>
     48 
     49 #include <netinet/in.h>
     50 #include <netinet/in_systm.h>
     51 #include <netinet/ip.h>
     52 
     53 #include <rump/rumpuser.h>
     54 
     55 #include "rump_private.h"
     56 #include "rumpcomp_user.h"
     57 
     58 /*
     59  * An inet communication domain which uses the socket interface.
     60  * Currently supports only IPv4 UDP, but could easily be extended to
     61  * support IPv6 and TCP by adding more stuff to the protosw.
     62  */
     63 
     64 DOMAIN_DEFINE(sockindomain);
     65 DOMAIN_DEFINE(sockin6domain);
     66 
     67 
     68 static void	sockin_init(void);
     69 static int	sockin_usrreq(struct socket *, int, struct mbuf *,
     70 			      struct mbuf *, struct mbuf *, struct lwp *);
     71 static int	sockin_ctloutput(int op, struct socket *, struct sockopt *);
     72 
     73 const struct protosw sockinsw[] = {
     74 {
     75 	.pr_type = SOCK_DGRAM,
     76 	.pr_domain = &sockindomain,
     77 	.pr_protocol = IPPROTO_UDP,
     78 	.pr_flags = PR_ATOMIC|PR_ADDR,
     79 	.pr_usrreq = sockin_usrreq,
     80 	.pr_ctloutput = sockin_ctloutput,
     81 },
     82 {
     83 	.pr_type = SOCK_STREAM,
     84 	.pr_domain = &sockindomain,
     85 	.pr_protocol = IPPROTO_TCP,
     86 	.pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_LISTEN|PR_ABRTACPTDIS,
     87 	.pr_usrreq = sockin_usrreq,
     88 	.pr_ctloutput = sockin_ctloutput,
     89 }};
     90 const struct protosw sockin6sw[] = {
     91 {
     92 	.pr_type = SOCK_DGRAM,
     93 	.pr_domain = &sockin6domain,
     94 	.pr_protocol = IPPROTO_UDP,
     95 	.pr_flags = PR_ATOMIC|PR_ADDR,
     96 	.pr_usrreq = sockin_usrreq,
     97 	.pr_ctloutput = sockin_ctloutput,
     98 },
     99 {
    100 	.pr_type = SOCK_STREAM,
    101 	.pr_domain = &sockin6domain,
    102 	.pr_protocol = IPPROTO_TCP,
    103 	.pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_LISTEN|PR_ABRTACPTDIS,
    104 	.pr_usrreq = sockin_usrreq,
    105 	.pr_ctloutput = sockin_ctloutput,
    106 }};
    107 
    108 struct domain sockindomain = {
    109 	.dom_family = PF_INET,
    110 	.dom_name = "socket_inet",
    111 	.dom_init = sockin_init,
    112 	.dom_externalize = NULL,
    113 	.dom_dispose = NULL,
    114 	.dom_protosw = sockinsw,
    115 	.dom_protoswNPROTOSW = &sockinsw[__arraycount(sockinsw)],
    116 	.dom_rtattach = rt_inithead,
    117 	.dom_rtoffset = 32,
    118 	.dom_maxrtkey = sizeof(struct sockaddr_in),
    119 	.dom_ifattach = NULL,
    120 	.dom_ifdetach = NULL,
    121 	.dom_ifqueues = { NULL },
    122 	.dom_link = { NULL },
    123 	.dom_mowner = MOWNER_INIT("",""),
    124 	.dom_rtcache = { NULL },
    125 	.dom_sockaddr_cmp = NULL
    126 };
    127 struct domain sockin6domain = {
    128 	.dom_family = PF_INET6,
    129 	.dom_name = "socket_inet6",
    130 	.dom_init = sockin_init,
    131 	.dom_externalize = NULL,
    132 	.dom_dispose = NULL,
    133 	.dom_protosw = sockin6sw,
    134 	.dom_protoswNPROTOSW = &sockin6sw[__arraycount(sockin6sw)],
    135 	.dom_rtattach = rt_inithead,
    136 	.dom_rtoffset = 32,
    137 	.dom_maxrtkey = sizeof(struct sockaddr_in6),
    138 	.dom_ifattach = NULL,
    139 	.dom_ifdetach = NULL,
    140 	.dom_ifqueues = { NULL },
    141 	.dom_link = { NULL },
    142 	.dom_mowner = MOWNER_INIT("",""),
    143 	.dom_rtcache = { NULL },
    144 	.dom_sockaddr_cmp = NULL
    145 };
    146 
    147 #define SO2S(so) ((intptr_t)(so->so_internal))
    148 #define SOCKIN_SBSIZE 65536
    149 
    150 struct sockin_unit {
    151 	struct socket *su_so;
    152 
    153 	LIST_ENTRY(sockin_unit) su_entries;
    154 };
    155 static LIST_HEAD(, sockin_unit) su_ent = LIST_HEAD_INITIALIZER(su_ent);
    156 static kmutex_t su_mtx;
    157 static bool rebuild;
    158 static int nsock;
    159 
    160 /* XXX: for the bpf hack */
    161 static struct ifnet sockin_if;
    162 int ifpromisc(struct ifnet *ifp, int pswitch) { return 0; }
    163 
    164 static int
    165 registersock(struct socket *so, int news)
    166 {
    167 	struct sockin_unit *su;
    168 
    169 	su = kmem_alloc(sizeof(*su), KM_NOSLEEP);
    170 	if (!su)
    171 		return ENOMEM;
    172 
    173 	so->so_internal = (void *)(intptr_t)news;
    174 	su->su_so = so;
    175 
    176 	mutex_enter(&su_mtx);
    177 	LIST_INSERT_HEAD(&su_ent, su, su_entries);
    178 	nsock++;
    179 	rebuild = true;
    180 	mutex_exit(&su_mtx);
    181 
    182 	return 0;
    183 }
    184 
    185 static void
    186 removesock(struct socket *so)
    187 {
    188 	struct sockin_unit *su_iter;
    189 
    190 	mutex_enter(&su_mtx);
    191 	LIST_FOREACH(su_iter, &su_ent, su_entries) {
    192 		if (su_iter->su_so == so)
    193 			break;
    194 	}
    195 	if (!su_iter)
    196 		panic("no such socket");
    197 
    198 	LIST_REMOVE(su_iter, su_entries);
    199 	nsock--;
    200 	rebuild = true;
    201 	mutex_exit(&su_mtx);
    202 
    203 	rumpuser_close(SO2S(su_iter->su_so));
    204 	kmem_free(su_iter, sizeof(*su_iter));
    205 }
    206 
    207 static void
    208 sockin_process(struct socket *so)
    209 {
    210 	struct sockaddr_in6 from;
    211 	struct iovec io;
    212 	struct msghdr rmsg;
    213 	struct mbuf *m;
    214 	size_t n, plen;
    215 	int error;
    216 
    217 	m = m_gethdr(M_WAIT, MT_DATA);
    218 	if (so->so_proto->pr_type == SOCK_DGRAM) {
    219 		plen = IP_MAXPACKET;
    220 		MEXTMALLOC(m, plen, M_DONTWAIT);
    221 	} else {
    222 		plen = MCLBYTES;
    223 		MCLGET(m, M_DONTWAIT);
    224 	}
    225 	if ((m->m_flags & M_EXT) == 0) {
    226 		m_freem(m);
    227 		return;
    228 	}
    229 
    230 	memset(&rmsg, 0, sizeof(rmsg));
    231 	io.iov_base = mtod(m, void *);
    232 	io.iov_len = plen;
    233 	rmsg.msg_iov = &io;
    234 	rmsg.msg_iovlen = 1;
    235 	rmsg.msg_name = (struct sockaddr *)&from;
    236 	rmsg.msg_namelen = sizeof(from);
    237 
    238 	error = rumpcomp_sockin_recvmsg(SO2S(so), &rmsg, 0, &n);
    239 	if (error) {
    240 		m_freem(m);
    241 
    242 		/* Treat a TCP socket a goner */
    243 		if (error != EAGAIN && so->so_proto->pr_type == SOCK_STREAM) {
    244 			mutex_enter(softnet_lock);
    245 			soisdisconnected(so);
    246 			mutex_exit(softnet_lock);
    247 			removesock(so);
    248 		}
    249 		return;
    250 	}
    251 	m->m_len = m->m_pkthdr.len = n;
    252 
    253 	bpf_mtap_af(&sockin_if, AF_UNSPEC, m);
    254 
    255 	mutex_enter(softnet_lock);
    256 	if (so->so_proto->pr_type == SOCK_DGRAM) {
    257 		if (!sbappendaddr(&so->so_rcv, rmsg.msg_name, m, NULL)) {
    258 			m_freem(m);
    259 		}
    260 	} else {
    261 		sbappendstream(&so->so_rcv, m);
    262 	}
    263 
    264 	sorwakeup(so);
    265 	mutex_exit(softnet_lock);
    266 }
    267 
    268 static void
    269 sockin_accept(struct socket *so)
    270 {
    271 	struct socket *nso;
    272 	struct sockaddr_in6 sin;
    273 	int news, error, slen;
    274 
    275 	slen = sizeof(sin);
    276 	error = rumpcomp_sockin_accept(SO2S(so), (struct sockaddr *)&sin,
    277 	    &slen, &news);
    278 	if (error)
    279 		return;
    280 
    281 	mutex_enter(softnet_lock);
    282 	nso = sonewconn(so, SS_ISCONNECTED);
    283 	if (nso == NULL)
    284 		goto errout;
    285 	if (registersock(nso, news) != 0)
    286 		goto errout;
    287 	mutex_exit(softnet_lock);
    288 	return;
    289 
    290  errout:
    291 	rumpuser_close(news);
    292 	if (nso)
    293 		soclose(nso);
    294 	mutex_exit(softnet_lock);
    295 }
    296 
    297 #define POLLTIMEOUT 100	/* check for new entries every 100ms */
    298 
    299 /* XXX: doesn't handle socket (kernel) locking properly? */
    300 static void
    301 sockinworker(void *arg)
    302 {
    303 	struct pollfd *pfds = NULL, *npfds;
    304 	struct sockin_unit *su_iter;
    305 	struct socket *so;
    306 	int cursock = 0, i, rv, error;
    307 
    308 	/*
    309 	 * Loop reading requests.  Check for new sockets periodically
    310 	 * (could be smarter, but I'm lazy).
    311 	 */
    312 	for (;;) {
    313 		if (rebuild) {
    314 			npfds = NULL;
    315 			mutex_enter(&su_mtx);
    316 			if (nsock)
    317 				npfds = kmem_alloc(nsock * sizeof(*npfds),
    318 				    KM_NOSLEEP);
    319 			if (npfds || nsock == 0) {
    320 				if (pfds)
    321 					kmem_free(pfds, cursock*sizeof(*pfds));
    322 				pfds = npfds;
    323 				cursock = nsock;
    324 				rebuild = false;
    325 
    326 				i = 0;
    327 				LIST_FOREACH(su_iter, &su_ent, su_entries) {
    328 					pfds[i].fd = SO2S(su_iter->su_so);
    329 					pfds[i].events = POLLIN;
    330 					pfds[i].revents = 0;
    331 					i++;
    332 				}
    333 				KASSERT(i == nsock);
    334 			}
    335 			mutex_exit(&su_mtx);
    336 		}
    337 
    338 		/* find affected sockets & process */
    339 		error = rumpcomp_sockin_poll(pfds, cursock, POLLTIMEOUT, &rv);
    340 		for (i = 0; i < cursock && rv > 0 && error == 0; i++) {
    341 			if (pfds[i].revents & POLLIN) {
    342 				mutex_enter(&su_mtx);
    343 				LIST_FOREACH(su_iter, &su_ent, su_entries) {
    344 					if (SO2S(su_iter->su_so)==pfds[i].fd) {
    345 						so = su_iter->su_so;
    346 						mutex_exit(&su_mtx);
    347 						if(so->so_options&SO_ACCEPTCONN)
    348 							sockin_accept(so);
    349 						else
    350 							sockin_process(so);
    351 						mutex_enter(&su_mtx);
    352 						break;
    353 					}
    354 				}
    355 				/* if we can't find it, just wing it */
    356 				KASSERT(rebuild || su_iter);
    357 				mutex_exit(&su_mtx);
    358 				pfds[i].revents = 0;
    359 				rv--;
    360 				i = -1;
    361 				continue;
    362 			}
    363 
    364 			/* something else?  ignore */
    365 			if (pfds[i].revents) {
    366 				pfds[i].revents = 0;
    367 				rv--;
    368 			}
    369 		}
    370 		KASSERT(rv <= 0);
    371 	}
    372 
    373 }
    374 
    375 static void
    376 sockin_init(void)
    377 {
    378 	int rv;
    379 
    380 	if (rump_threads) {
    381 		if ((rv = kthread_create(PRI_NONE, 0, NULL, sockinworker,
    382 		    NULL, NULL, "sockwork")) != 0)
    383 			panic("sockin_init: could not create worker thread\n");
    384 	} else {
    385 		printf("sockin_init: no threads => no worker thread\n");
    386 	}
    387 	mutex_init(&su_mtx, MUTEX_DEFAULT, IPL_NONE);
    388 	strlcpy(sockin_if.if_xname, "sockin0", sizeof(sockin_if.if_xname));
    389 	bpf_attach(&sockin_if, DLT_NULL, 0);
    390 }
    391 
    392 static int
    393 sockin_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam,
    394 	struct mbuf *control, struct lwp *l)
    395 {
    396 	int error = 0;
    397 
    398 	switch (req) {
    399 	case PRU_ATTACH:
    400 	{
    401 		int news;
    402 		int sbsize;
    403 		int family;
    404 
    405 		sosetlock(so);
    406 		if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
    407 			error = soreserve(so, SOCKIN_SBSIZE, SOCKIN_SBSIZE);
    408 			if (error)
    409 				break;
    410 		}
    411 
    412 		family = so->so_proto->pr_domain->dom_family;
    413 		KASSERT(family == PF_INET || family == PF_INET6);
    414 		error = rumpcomp_sockin_socket(family,
    415 		    so->so_proto->pr_type, 0, &news);
    416 		if (error)
    417 			break;
    418 
    419 		/* for UDP sockets, make sure we can send&recv max */
    420 		if (so->so_proto->pr_type == SOCK_DGRAM) {
    421 			sbsize = SOCKIN_SBSIZE;
    422 			error = rumpcomp_sockin_setsockopt(news,
    423 			    SOL_SOCKET, SO_SNDBUF,
    424 			    &sbsize, sizeof(sbsize));
    425 			sbsize = SOCKIN_SBSIZE;
    426 			error = rumpcomp_sockin_setsockopt(news,
    427 			    SOL_SOCKET, SO_RCVBUF,
    428 			    &sbsize, sizeof(sbsize));
    429 		}
    430 
    431 		if ((error = registersock(so, news)) != 0)
    432 			rumpuser_close(news);
    433 
    434 		break;
    435 	}
    436 
    437 	case PRU_ACCEPT:
    438 		/* we do all the work in the worker thread */
    439 		break;
    440 
    441 	case PRU_BIND:
    442 		error = rumpcomp_sockin_bind(SO2S(so),
    443 		    mtod(nam, const struct sockaddr *),
    444 		    nam->m_len);
    445 		break;
    446 
    447 	case PRU_CONNECT:
    448 		error = rumpcomp_sockin_connect(SO2S(so),
    449 		    mtod(nam, struct sockaddr *), nam->m_len);
    450 		if (error == 0)
    451 			soisconnected(so);
    452 		break;
    453 
    454 	case PRU_LISTEN:
    455 		error = rumpcomp_sockin_listen(SO2S(so), so->so_qlimit);
    456 		break;
    457 
    458 	case PRU_SEND:
    459 	{
    460 		struct sockaddr *saddr;
    461 		struct msghdr mhdr;
    462 		size_t iov_max, i;
    463 		struct iovec iov_buf[32], *iov;
    464 		struct mbuf *m2;
    465 		size_t tot, n;
    466 		int s;
    467 
    468 		bpf_mtap_af(&sockin_if, AF_UNSPEC, m);
    469 
    470 		memset(&mhdr, 0, sizeof(mhdr));
    471 
    472 		iov_max = 0;
    473 		for (m2 = m; m2 != NULL; m2 = m2->m_next) {
    474 			iov_max++;
    475 		}
    476 
    477 		if (iov_max <= __arraycount(iov_buf)) {
    478 			iov = iov_buf;
    479 		} else {
    480 			iov = kmem_alloc(sizeof(struct iovec) * iov_max,
    481 			    KM_SLEEP);
    482 		}
    483 
    484 		tot = 0;
    485 		for (i = 0, m2 = m; m2 != NULL; m2 = m2->m_next, i++) {
    486 			iov[i].iov_base = m2->m_data;
    487 			iov[i].iov_len = m2->m_len;
    488 			tot += m2->m_len;
    489 		}
    490 		mhdr.msg_iov = iov;
    491 		mhdr.msg_iovlen = i;
    492 		s = SO2S(so);
    493 
    494 		if (nam != NULL) {
    495 			saddr = mtod(nam, struct sockaddr *);
    496 			mhdr.msg_name = saddr;
    497 			mhdr.msg_namelen = saddr->sa_len;
    498 		}
    499 
    500 		rumpcomp_sockin_sendmsg(s, &mhdr, 0, &n);
    501 
    502 		if (iov != iov_buf)
    503 			kmem_free(iov, sizeof(struct iovec) * iov_max);
    504 
    505 		m_freem(m);
    506 		m_freem(control);
    507 
    508 		/* this assumes too many things to list.. buthey, testing */
    509 		if (!rump_threads)
    510 			sockin_process(so);
    511 	}
    512 		break;
    513 
    514 	case PRU_SHUTDOWN:
    515 		removesock(so);
    516 		break;
    517 
    518 	case PRU_SOCKADDR:
    519 	case PRU_PEERADDR:
    520 	{
    521 		int slen = nam->m_len;
    522 		enum rumpcomp_sockin_getnametype which;
    523 
    524 		if (req == PRU_SOCKADDR)
    525 			which = RUMPCOMP_SOCKIN_SOCKNAME;
    526 		else
    527 			which = RUMPCOMP_SOCKIN_PEERNAME;
    528 		error = rumpcomp_sockin_getname(SO2S(so),
    529 		    mtod(nam, struct sockaddr *), &slen, which);
    530 		if (error == 0)
    531 			nam->m_len = slen;
    532 		break;
    533 	}
    534 
    535 	case PRU_CONTROL:
    536 		error = ENOTTY;
    537 		break;
    538 
    539 	default:
    540 		panic("sockin_usrreq: IMPLEMENT ME, req %d not supported", req);
    541 	}
    542 
    543 	return error;
    544 }
    545 
    546 static int
    547 sockin_ctloutput(int op, struct socket *so, struct sockopt *sopt)
    548 {
    549 
    550 	return rumpcomp_sockin_setsockopt(SO2S(so), sopt->sopt_level,
    551 	    sopt->sopt_name, sopt->sopt_data, sopt->sopt_size);
    552 }
    553