1 /* $NetBSD: sockin.c,v 1.21 2010/01/26 17:41:20 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 1.21 2010/01/26 17:41:20 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 57 /* 58 * An inet communication domain which uses the socket interface. 59 * Currently supports only IPv4 UDP, but could easily be extended to 60 * support IPv6 and TCP by adding more stuff to the protosw. 61 */ 62 63 DOMAIN_DEFINE(sockindomain); 64 65 static void sockin_init(void); 66 static int sockin_usrreq(struct socket *, int, struct mbuf *, 67 struct mbuf *, struct mbuf *, struct lwp *); 68 static int sockin_ctloutput(int op, struct socket *, struct sockopt *); 69 70 const struct protosw sockinsw[] = { 71 { 72 .pr_type = SOCK_DGRAM, 73 .pr_domain = &sockindomain, 74 .pr_protocol = IPPROTO_UDP, 75 .pr_flags = PR_ATOMIC|PR_ADDR, 76 .pr_usrreq = sockin_usrreq, 77 .pr_ctloutput = sockin_ctloutput, 78 }, 79 { 80 .pr_type = SOCK_STREAM, 81 .pr_domain = &sockindomain, 82 .pr_protocol = IPPROTO_TCP, 83 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_LISTEN|PR_ABRTACPTDIS, 84 .pr_usrreq = sockin_usrreq, 85 .pr_ctloutput = sockin_ctloutput, 86 }}; 87 88 struct domain sockindomain = { 89 .dom_family = PF_INET, 90 .dom_name = "socket_inet", 91 .dom_init = sockin_init, 92 .dom_externalize = NULL, 93 .dom_dispose = NULL, 94 .dom_protosw = sockinsw, 95 .dom_protoswNPROTOSW = &sockinsw[__arraycount(sockinsw)], 96 .dom_rtattach = rn_inithead, 97 .dom_rtoffset = 32, 98 .dom_maxrtkey = sizeof(struct sockaddr_in), 99 .dom_ifattach = NULL, 100 .dom_ifdetach = NULL, 101 .dom_ifqueues = { NULL }, 102 .dom_link = { NULL }, 103 .dom_mowner = MOWNER_INIT("",""), 104 .dom_rtcache = { NULL }, 105 .dom_sockaddr_cmp = NULL 106 }; 107 108 #define SO2S(so) ((intptr_t)(so->so_internal)) 109 #define SOCKIN_SBSIZE 65536 110 111 struct sockin_unit { 112 struct socket *su_so; 113 114 LIST_ENTRY(sockin_unit) su_entries; 115 }; 116 static LIST_HEAD(, sockin_unit) su_ent = LIST_HEAD_INITIALIZER(su_ent); 117 static kmutex_t su_mtx; 118 static bool rebuild; 119 static int nsock; 120 121 /* XXX: for the bpf hack */ 122 static struct ifnet sockin_if; 123 int ifpromisc(struct ifnet *ifp, int pswitch) { return 0; } 124 125 static int 126 registersock(struct socket *so, int news) 127 { 128 struct sockin_unit *su; 129 130 su = kmem_alloc(sizeof(*su), KM_NOSLEEP); 131 if (!su) 132 return ENOMEM; 133 134 so->so_internal = (void *)(intptr_t)news; 135 su->su_so = so; 136 137 mutex_enter(&su_mtx); 138 LIST_INSERT_HEAD(&su_ent, su, su_entries); 139 nsock++; 140 rebuild = true; 141 mutex_exit(&su_mtx); 142 143 return 0; 144 } 145 146 static void 147 removesock(struct socket *so) 148 { 149 struct sockin_unit *su_iter; 150 int error; 151 152 mutex_enter(&su_mtx); 153 LIST_FOREACH(su_iter, &su_ent, su_entries) { 154 if (su_iter->su_so == so) 155 break; 156 } 157 if (!su_iter) 158 panic("no such socket"); 159 160 LIST_REMOVE(su_iter, su_entries); 161 nsock--; 162 rebuild = true; 163 mutex_exit(&su_mtx); 164 165 rumpuser_close(SO2S(su_iter->su_so), &error); 166 kmem_free(su_iter, sizeof(*su_iter)); 167 } 168 169 static void 170 sockin_process(struct socket *so) 171 { 172 struct sockaddr_in from; 173 struct iovec io; 174 struct msghdr rmsg; 175 struct mbuf *m; 176 ssize_t n; 177 size_t plen; 178 int error; 179 180 m = m_gethdr(M_WAIT, MT_DATA); 181 if (so->so_proto->pr_type == SOCK_DGRAM) { 182 plen = IP_MAXPACKET; 183 MEXTMALLOC(m, plen, M_DONTWAIT); 184 } else { 185 plen = MCLBYTES; 186 MCLGET(m, M_DONTWAIT); 187 } 188 if ((m->m_flags & M_EXT) == 0) { 189 m_freem(m); 190 return; 191 } 192 193 memset(&rmsg, 0, sizeof(rmsg)); 194 io.iov_base = mtod(m, void *); 195 io.iov_len = plen; 196 rmsg.msg_iov = &io; 197 rmsg.msg_iovlen = 1; 198 rmsg.msg_name = (struct sockaddr *)&from; 199 rmsg.msg_namelen = sizeof(from); 200 201 n = rumpuser_net_recvmsg(SO2S(so), &rmsg, 0, &error); 202 if (n <= 0) { 203 m_freem(m); 204 205 /* Treat a TCP socket a goner */ 206 if (error != EAGAIN && so->so_proto->pr_type == SOCK_STREAM) { 207 mutex_enter(softnet_lock); 208 soisdisconnected(so); 209 mutex_exit(softnet_lock); 210 removesock(so); 211 } 212 return; 213 } 214 m->m_len = m->m_pkthdr.len = n; 215 216 if (sockin_if.if_bpf) 217 bpf_ops->bpf_mtap_af(sockin_if.if_bpf, AF_UNSPEC, m); 218 219 mutex_enter(softnet_lock); 220 if (so->so_proto->pr_type == SOCK_DGRAM) { 221 if (!sbappendaddr(&so->so_rcv, rmsg.msg_name, m, NULL)) { 222 m_freem(m); 223 } 224 } else { 225 sbappendstream(&so->so_rcv, m); 226 } 227 228 sorwakeup(so); 229 mutex_exit(softnet_lock); 230 } 231 232 static void 233 sockin_accept(struct socket *so) 234 { 235 struct socket *nso; 236 struct sockaddr_in sin; 237 int news, error, slen; 238 239 slen = sizeof(sin); 240 news = rumpuser_net_accept(SO2S(so), (struct sockaddr *)&sin, 241 &slen, &error); 242 if (news == -1) 243 return; 244 245 mutex_enter(softnet_lock); 246 nso = sonewconn(so, SS_ISCONNECTED); 247 if (nso == NULL) 248 goto errout; 249 if (registersock(nso, news) != 0) 250 goto errout; 251 mutex_exit(softnet_lock); 252 return; 253 254 errout: 255 rumpuser_close(news, &error); 256 if (nso) 257 soclose(nso); 258 mutex_exit(softnet_lock); 259 } 260 261 #define POLLTIMEOUT 100 /* check for new entries every 100ms */ 262 263 /* XXX: doesn't handle socket (kernel) locking properly? */ 264 static void 265 sockinworker(void *arg) 266 { 267 struct pollfd *pfds = NULL, *npfds; 268 struct sockin_unit *su_iter; 269 struct socket *so; 270 int cursock = 0, i, rv, error; 271 272 /* 273 * Loop reading requests. Check for new sockets periodically 274 * (could be smarter, but I'm lazy). 275 */ 276 for (;;) { 277 if (rebuild) { 278 npfds = NULL; 279 mutex_enter(&su_mtx); 280 if (nsock) 281 npfds = kmem_alloc(nsock * sizeof(*npfds), 282 KM_NOSLEEP); 283 if (npfds || nsock == 0) { 284 if (pfds) 285 kmem_free(pfds, cursock*sizeof(*pfds)); 286 pfds = npfds; 287 cursock = nsock; 288 rebuild = false; 289 290 i = 0; 291 LIST_FOREACH(su_iter, &su_ent, su_entries) { 292 pfds[i].fd = SO2S(su_iter->su_so); 293 pfds[i].events = POLLIN; 294 pfds[i].revents = 0; 295 i++; 296 } 297 KASSERT(i == nsock); 298 } 299 mutex_exit(&su_mtx); 300 } 301 302 /* find affected sockets & process */ 303 rv = rumpuser_poll(pfds, cursock, POLLTIMEOUT, &error); 304 for (i = 0; i < cursock && rv > 0; i++) { 305 if (pfds[i].revents & POLLIN) { 306 mutex_enter(&su_mtx); 307 LIST_FOREACH(su_iter, &su_ent, su_entries) { 308 if (SO2S(su_iter->su_so)==pfds[i].fd) { 309 so = su_iter->su_so; 310 mutex_exit(&su_mtx); 311 if(so->so_options&SO_ACCEPTCONN) 312 sockin_accept(so); 313 else 314 sockin_process(so); 315 mutex_enter(&su_mtx); 316 break; 317 } 318 } 319 /* if we can't find it, just wing it */ 320 KASSERT(rebuild || su_iter); 321 mutex_exit(&su_mtx); 322 pfds[i].revents = 0; 323 rv--; 324 i = -1; 325 continue; 326 } 327 328 /* something else? ignore */ 329 if (pfds[i].revents) { 330 pfds[i].revents = 0; 331 rv--; 332 } 333 } 334 KASSERT(rv <= 0); 335 } 336 337 } 338 339 static void 340 sockin_init(void) 341 { 342 int rv; 343 344 if (rump_threads) { 345 if ((rv = kthread_create(PRI_NONE, 0, NULL, sockinworker, 346 NULL, NULL, "sockwork")) != 0) 347 panic("sockin_init: could not create worker thread\n"); 348 } else { 349 printf("sockin_init: no threads => no worker thread\n"); 350 } 351 mutex_init(&su_mtx, MUTEX_DEFAULT, IPL_NONE); 352 strlcpy(sockin_if.if_xname, "sockin0", sizeof(sockin_if.if_xname)); 353 bpf_ops->bpf_attach(&sockin_if, DLT_NULL, 0, &sockin_if.if_bpf); 354 } 355 356 static int 357 sockin_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam, 358 struct mbuf *control, struct lwp *l) 359 { 360 int error = 0, rv; 361 362 switch (req) { 363 case PRU_ATTACH: 364 { 365 int news, dummy; 366 int sbsize; 367 368 sosetlock(so); 369 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 370 error = soreserve(so, SOCKIN_SBSIZE, SOCKIN_SBSIZE); 371 if (error) 372 break; 373 } 374 375 news = rumpuser_net_socket(PF_INET, so->so_proto->pr_type, 376 0, &error); 377 if (news == -1) 378 break; 379 380 /* for UDP sockets, make sure we can send&recv max */ 381 if (so->so_proto->pr_type == SOCK_DGRAM) { 382 sbsize = SOCKIN_SBSIZE; 383 rumpuser_net_setsockopt(news, SOL_SOCKET, SO_SNDBUF, 384 &sbsize, sizeof(sbsize), &error); 385 sbsize = SOCKIN_SBSIZE; 386 rumpuser_net_setsockopt(news, SOL_SOCKET, SO_RCVBUF, 387 &sbsize, sizeof(sbsize), &error); 388 } 389 390 if ((error = registersock(so, news)) != 0) 391 rumpuser_close(news, &dummy); 392 393 break; 394 } 395 396 case PRU_ACCEPT: 397 /* we do all the work in the worker thread */ 398 break; 399 400 case PRU_BIND: 401 rumpuser_net_bind(SO2S(so), mtod(nam, const struct sockaddr *), 402 sizeof(struct sockaddr_in), &error); 403 break; 404 405 case PRU_CONNECT: 406 rv = rumpuser_net_connect(SO2S(so), 407 mtod(nam, struct sockaddr *), sizeof(struct sockaddr_in), 408 &error); 409 if (rv == 0) 410 soisconnected(so); 411 break; 412 413 case PRU_LISTEN: 414 rumpuser_net_listen(SO2S(so), so->so_qlimit, &error); 415 break; 416 417 case PRU_SEND: 418 { 419 struct sockaddr *saddr; 420 struct msghdr mhdr; 421 size_t iov_max, i; 422 struct iovec iov_buf[32], *iov; 423 struct mbuf *m2; 424 size_t tot; 425 int s; 426 427 if (sockin_if.if_bpf) 428 bpf_ops->bpf_mtap_af(sockin_if.if_bpf, AF_UNSPEC, m); 429 430 memset(&mhdr, 0, sizeof(mhdr)); 431 432 iov_max = 0; 433 for (m2 = m; m2 != NULL; m2 = m2->m_next) { 434 iov_max++; 435 } 436 437 if (iov_max <= __arraycount(iov_buf)) { 438 iov = iov_buf; 439 } else { 440 iov = kmem_alloc(sizeof(struct iovec) * iov_max, 441 KM_SLEEP); 442 } 443 444 tot = 0; 445 for (i = 0, m2 = m; m2 != NULL; m2 = m2->m_next, i++) { 446 iov[i].iov_base = m2->m_data; 447 iov[i].iov_len = m2->m_len; 448 tot += m2->m_len; 449 } 450 mhdr.msg_iov = iov; 451 mhdr.msg_iovlen = i; 452 s = SO2S(so); 453 454 if (nam != NULL) { 455 saddr = mtod(nam, struct sockaddr *); 456 mhdr.msg_name = saddr; 457 mhdr.msg_namelen = saddr->sa_len; 458 } 459 460 rumpuser_net_sendmsg(s, &mhdr, 0, &error); 461 462 if (iov != iov_buf) 463 kmem_free(iov, sizeof(struct iovec) * iov_max); 464 465 m_freem(m); 466 m_freem(control); 467 468 /* this assumes too many things to list.. buthey, testing */ 469 if (!rump_threads) 470 sockin_process(so); 471 } 472 break; 473 474 case PRU_SHUTDOWN: 475 removesock(so); 476 break; 477 478 case PRU_SOCKADDR: 479 case PRU_PEERADDR: 480 { 481 int slen = nam->m_len; 482 enum rumpuser_getnametype which; 483 484 if (req == PRU_SOCKADDR) 485 which = RUMPUSER_SOCKNAME; 486 else 487 which = RUMPUSER_PEERNAME; 488 rumpuser_net_getname(SO2S(so), 489 mtod(nam, struct sockaddr *), &slen, which, &error); 490 if (error == 0) 491 nam->m_len = slen; 492 break; 493 } 494 495 default: 496 panic("sockin_usrreq: IMPLEMENT ME, req %d not supported", req); 497 } 498 499 return error; 500 } 501 502 static int 503 sockin_ctloutput(int op, struct socket *so, struct sockopt *sopt) 504 { 505 int error; 506 507 rumpuser_net_setsockopt(SO2S(so), sopt->sopt_level, 508 sopt->sopt_name, sopt->sopt_data, sopt->sopt_size, &error); 509 return error; 510 } 511