1 /* $NetBSD: sockin.c,v 1.30 2013/04/30 00:12:35 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 1.30 2013/04/30 00:12:35 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 66 static void sockin_init(void); 67 static int sockin_usrreq(struct socket *, int, struct mbuf *, 68 struct mbuf *, struct mbuf *, struct lwp *); 69 static int sockin_ctloutput(int op, struct socket *, struct sockopt *); 70 71 const struct protosw sockinsw[] = { 72 { 73 .pr_type = SOCK_DGRAM, 74 .pr_domain = &sockindomain, 75 .pr_protocol = IPPROTO_UDP, 76 .pr_flags = PR_ATOMIC|PR_ADDR, 77 .pr_usrreq = sockin_usrreq, 78 .pr_ctloutput = sockin_ctloutput, 79 }, 80 { 81 .pr_type = SOCK_STREAM, 82 .pr_domain = &sockindomain, 83 .pr_protocol = IPPROTO_TCP, 84 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_LISTEN|PR_ABRTACPTDIS, 85 .pr_usrreq = sockin_usrreq, 86 .pr_ctloutput = sockin_ctloutput, 87 }}; 88 89 struct domain sockindomain = { 90 .dom_family = PF_INET, 91 .dom_name = "socket_inet", 92 .dom_init = sockin_init, 93 .dom_externalize = NULL, 94 .dom_dispose = NULL, 95 .dom_protosw = sockinsw, 96 .dom_protoswNPROTOSW = &sockinsw[__arraycount(sockinsw)], 97 .dom_rtattach = rt_inithead, 98 .dom_rtoffset = 32, 99 .dom_maxrtkey = sizeof(struct sockaddr_in), 100 .dom_ifattach = NULL, 101 .dom_ifdetach = NULL, 102 .dom_ifqueues = { NULL }, 103 .dom_link = { NULL }, 104 .dom_mowner = MOWNER_INIT("",""), 105 .dom_rtcache = { NULL }, 106 .dom_sockaddr_cmp = NULL 107 }; 108 109 #define SO2S(so) ((intptr_t)(so->so_internal)) 110 #define SOCKIN_SBSIZE 65536 111 112 struct sockin_unit { 113 struct socket *su_so; 114 115 LIST_ENTRY(sockin_unit) su_entries; 116 }; 117 static LIST_HEAD(, sockin_unit) su_ent = LIST_HEAD_INITIALIZER(su_ent); 118 static kmutex_t su_mtx; 119 static bool rebuild; 120 static int nsock; 121 122 /* XXX: for the bpf hack */ 123 static struct ifnet sockin_if; 124 int ifpromisc(struct ifnet *ifp, int pswitch) { return 0; } 125 126 static int 127 registersock(struct socket *so, int news) 128 { 129 struct sockin_unit *su; 130 131 su = kmem_alloc(sizeof(*su), KM_NOSLEEP); 132 if (!su) 133 return ENOMEM; 134 135 so->so_internal = (void *)(intptr_t)news; 136 su->su_so = so; 137 138 mutex_enter(&su_mtx); 139 LIST_INSERT_HEAD(&su_ent, su, su_entries); 140 nsock++; 141 rebuild = true; 142 mutex_exit(&su_mtx); 143 144 return 0; 145 } 146 147 static void 148 removesock(struct socket *so) 149 { 150 struct sockin_unit *su_iter; 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)); 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 size_t n, plen; 177 int error; 178 179 m = m_gethdr(M_WAIT, MT_DATA); 180 if (so->so_proto->pr_type == SOCK_DGRAM) { 181 plen = IP_MAXPACKET; 182 MEXTMALLOC(m, plen, M_DONTWAIT); 183 } else { 184 plen = MCLBYTES; 185 MCLGET(m, M_DONTWAIT); 186 } 187 if ((m->m_flags & M_EXT) == 0) { 188 m_freem(m); 189 return; 190 } 191 192 memset(&rmsg, 0, sizeof(rmsg)); 193 io.iov_base = mtod(m, void *); 194 io.iov_len = plen; 195 rmsg.msg_iov = &io; 196 rmsg.msg_iovlen = 1; 197 rmsg.msg_name = (struct sockaddr *)&from; 198 rmsg.msg_namelen = sizeof(from); 199 200 error = rumpcomp_sockin_recvmsg(SO2S(so), &rmsg, 0, &n); 201 if (error) { 202 m_freem(m); 203 204 /* Treat a TCP socket a goner */ 205 if (error != EAGAIN && so->so_proto->pr_type == SOCK_STREAM) { 206 mutex_enter(softnet_lock); 207 soisdisconnected(so); 208 mutex_exit(softnet_lock); 209 removesock(so); 210 } 211 return; 212 } 213 m->m_len = m->m_pkthdr.len = n; 214 215 bpf_mtap_af(&sockin_if, AF_UNSPEC, m); 216 217 mutex_enter(softnet_lock); 218 if (so->so_proto->pr_type == SOCK_DGRAM) { 219 if (!sbappendaddr(&so->so_rcv, rmsg.msg_name, m, NULL)) { 220 m_freem(m); 221 } 222 } else { 223 sbappendstream(&so->so_rcv, m); 224 } 225 226 sorwakeup(so); 227 mutex_exit(softnet_lock); 228 } 229 230 static void 231 sockin_accept(struct socket *so) 232 { 233 struct socket *nso; 234 struct sockaddr_in sin; 235 int news, error, slen; 236 237 slen = sizeof(sin); 238 error = rumpcomp_sockin_accept(SO2S(so), (struct sockaddr *)&sin, 239 &slen, &news); 240 if (error) 241 return; 242 243 mutex_enter(softnet_lock); 244 nso = sonewconn(so, SS_ISCONNECTED); 245 if (nso == NULL) 246 goto errout; 247 if (registersock(nso, news) != 0) 248 goto errout; 249 mutex_exit(softnet_lock); 250 return; 251 252 errout: 253 rumpuser_close(news); 254 if (nso) 255 soclose(nso); 256 mutex_exit(softnet_lock); 257 } 258 259 #define POLLTIMEOUT 100 /* check for new entries every 100ms */ 260 261 /* XXX: doesn't handle socket (kernel) locking properly? */ 262 static void 263 sockinworker(void *arg) 264 { 265 struct pollfd *pfds = NULL, *npfds; 266 struct sockin_unit *su_iter; 267 struct socket *so; 268 int cursock = 0, i, rv, error; 269 270 /* 271 * Loop reading requests. Check for new sockets periodically 272 * (could be smarter, but I'm lazy). 273 */ 274 for (;;) { 275 if (rebuild) { 276 npfds = NULL; 277 mutex_enter(&su_mtx); 278 if (nsock) 279 npfds = kmem_alloc(nsock * sizeof(*npfds), 280 KM_NOSLEEP); 281 if (npfds || nsock == 0) { 282 if (pfds) 283 kmem_free(pfds, cursock*sizeof(*pfds)); 284 pfds = npfds; 285 cursock = nsock; 286 rebuild = false; 287 288 i = 0; 289 LIST_FOREACH(su_iter, &su_ent, su_entries) { 290 pfds[i].fd = SO2S(su_iter->su_so); 291 pfds[i].events = POLLIN; 292 pfds[i].revents = 0; 293 i++; 294 } 295 KASSERT(i == nsock); 296 } 297 mutex_exit(&su_mtx); 298 } 299 300 /* find affected sockets & process */ 301 error = rumpcomp_sockin_poll(pfds, cursock, POLLTIMEOUT, &rv); 302 for (i = 0; i < cursock && rv > 0 && error == 0; i++) { 303 if (pfds[i].revents & POLLIN) { 304 mutex_enter(&su_mtx); 305 LIST_FOREACH(su_iter, &su_ent, su_entries) { 306 if (SO2S(su_iter->su_so)==pfds[i].fd) { 307 so = su_iter->su_so; 308 mutex_exit(&su_mtx); 309 if(so->so_options&SO_ACCEPTCONN) 310 sockin_accept(so); 311 else 312 sockin_process(so); 313 mutex_enter(&su_mtx); 314 break; 315 } 316 } 317 /* if we can't find it, just wing it */ 318 KASSERT(rebuild || su_iter); 319 mutex_exit(&su_mtx); 320 pfds[i].revents = 0; 321 rv--; 322 i = -1; 323 continue; 324 } 325 326 /* something else? ignore */ 327 if (pfds[i].revents) { 328 pfds[i].revents = 0; 329 rv--; 330 } 331 } 332 KASSERT(rv <= 0); 333 } 334 335 } 336 337 static void 338 sockin_init(void) 339 { 340 int rv; 341 342 if (rump_threads) { 343 if ((rv = kthread_create(PRI_NONE, 0, NULL, sockinworker, 344 NULL, NULL, "sockwork")) != 0) 345 panic("sockin_init: could not create worker thread\n"); 346 } else { 347 printf("sockin_init: no threads => no worker thread\n"); 348 } 349 mutex_init(&su_mtx, MUTEX_DEFAULT, IPL_NONE); 350 strlcpy(sockin_if.if_xname, "sockin0", sizeof(sockin_if.if_xname)); 351 bpf_attach(&sockin_if, DLT_NULL, 0); 352 } 353 354 static int 355 sockin_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam, 356 struct mbuf *control, struct lwp *l) 357 { 358 int error = 0; 359 360 switch (req) { 361 case PRU_ATTACH: 362 { 363 int news; 364 int sbsize; 365 366 sosetlock(so); 367 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 368 error = soreserve(so, SOCKIN_SBSIZE, SOCKIN_SBSIZE); 369 if (error) 370 break; 371 } 372 373 error = rumpcomp_sockin_socket(PF_INET, so->so_proto->pr_type, 374 0, &news); 375 if (error) 376 break; 377 378 /* for UDP sockets, make sure we can send&recv max */ 379 if (so->so_proto->pr_type == SOCK_DGRAM) { 380 sbsize = SOCKIN_SBSIZE; 381 error = rumpcomp_sockin_setsockopt(news, 382 SOL_SOCKET, SO_SNDBUF, 383 &sbsize, sizeof(sbsize)); 384 sbsize = SOCKIN_SBSIZE; 385 error = rumpcomp_sockin_setsockopt(news, 386 SOL_SOCKET, SO_RCVBUF, 387 &sbsize, sizeof(sbsize)); 388 } 389 390 if ((error = registersock(so, news)) != 0) 391 rumpuser_close(news); 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 error = rumpcomp_sockin_bind(SO2S(so), 402 mtod(nam, const struct sockaddr *), 403 sizeof(struct sockaddr_in)); 404 break; 405 406 case PRU_CONNECT: 407 error = rumpcomp_sockin_connect(SO2S(so), 408 mtod(nam, struct sockaddr *), sizeof(struct sockaddr_in)); 409 if (error == 0) 410 soisconnected(so); 411 break; 412 413 case PRU_LISTEN: 414 error = rumpcomp_sockin_listen(SO2S(so), so->so_qlimit); 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, n; 425 int s; 426 427 bpf_mtap_af(&sockin_if, AF_UNSPEC, m); 428 429 memset(&mhdr, 0, sizeof(mhdr)); 430 431 iov_max = 0; 432 for (m2 = m; m2 != NULL; m2 = m2->m_next) { 433 iov_max++; 434 } 435 436 if (iov_max <= __arraycount(iov_buf)) { 437 iov = iov_buf; 438 } else { 439 iov = kmem_alloc(sizeof(struct iovec) * iov_max, 440 KM_SLEEP); 441 } 442 443 tot = 0; 444 for (i = 0, m2 = m; m2 != NULL; m2 = m2->m_next, i++) { 445 iov[i].iov_base = m2->m_data; 446 iov[i].iov_len = m2->m_len; 447 tot += m2->m_len; 448 } 449 mhdr.msg_iov = iov; 450 mhdr.msg_iovlen = i; 451 s = SO2S(so); 452 453 if (nam != NULL) { 454 saddr = mtod(nam, struct sockaddr *); 455 mhdr.msg_name = saddr; 456 mhdr.msg_namelen = saddr->sa_len; 457 } 458 459 rumpcomp_sockin_sendmsg(s, &mhdr, 0, &n); 460 461 if (iov != iov_buf) 462 kmem_free(iov, sizeof(struct iovec) * iov_max); 463 464 m_freem(m); 465 m_freem(control); 466 467 /* this assumes too many things to list.. buthey, testing */ 468 if (!rump_threads) 469 sockin_process(so); 470 } 471 break; 472 473 case PRU_SHUTDOWN: 474 removesock(so); 475 break; 476 477 case PRU_SOCKADDR: 478 case PRU_PEERADDR: 479 { 480 int slen = nam->m_len; 481 enum rumpcomp_sockin_getnametype which; 482 483 if (req == PRU_SOCKADDR) 484 which = RUMPCOMP_SOCKIN_SOCKNAME; 485 else 486 which = RUMPCOMP_SOCKIN_PEERNAME; 487 error = rumpcomp_sockin_getname(SO2S(so), 488 mtod(nam, struct sockaddr *), &slen, which); 489 if (error == 0) 490 nam->m_len = slen; 491 break; 492 } 493 494 case PRU_CONTROL: 495 error = ENOTTY; 496 break; 497 498 default: 499 panic("sockin_usrreq: IMPLEMENT ME, req %d not supported", req); 500 } 501 502 return error; 503 } 504 505 static int 506 sockin_ctloutput(int op, struct socket *so, struct sockopt *sopt) 507 { 508 509 return rumpcomp_sockin_setsockopt(SO2S(so), sopt->sopt_level, 510 sopt->sopt_name, sopt->sopt_data, sopt->sopt_size); 511 } 512