1 /* 2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * $DragonFly: src/sys/kern/uipc_msg.c,v 1.26 2008/10/27 02:56:30 sephe Exp $ 34 */ 35 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/kernel.h> 39 #include <sys/msgport.h> 40 #include <sys/protosw.h> 41 #include <sys/socket.h> 42 #include <sys/socketvar.h> 43 #include <sys/socketops.h> 44 #include <sys/thread.h> 45 #include <sys/thread2.h> 46 #include <sys/msgport2.h> 47 #include <sys/mbuf.h> 48 #include <vm/pmap.h> 49 #include <net/netmsg2.h> 50 51 #include <net/netisr.h> 52 #include <net/netmsg.h> 53 54 /* 55 * Abort a socket and free it. Called from soabort() only. soabort() 56 * got a ref on the socket which we must free on reply. 57 */ 58 void 59 so_pru_abort(struct socket *so) 60 { 61 struct netmsg_pru_abort msg; 62 63 netmsg_init(&msg.base, so, &curthread->td_msgport, 64 0, so->so_proto->pr_usrreqs->pru_abort); 65 (void)lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 66 sofree(msg.base.nm_so); 67 } 68 69 /* 70 * Abort a socket and free it, asynchronously. Called from 71 * soaborta() only. soaborta() got a ref on the socket which we must 72 * free on reply. 73 */ 74 void 75 so_pru_aborta(struct socket *so) 76 { 77 struct netmsg_pru_abort *msg; 78 79 msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_WAITOK | M_ZERO); 80 netmsg_init(&msg->base, so, &netisr_afree_free_so_rport, 81 0, so->so_proto->pr_usrreqs->pru_abort); 82 lwkt_sendmsg(so->so_port, &msg->base.lmsg); 83 } 84 85 /* 86 * Abort a socket and free it. Called from soabort_oncpu() only. 87 * Caller must make sure that the current CPU is inpcb's owner CPU. 88 */ 89 void 90 so_pru_abort_oncpu(struct socket *so) 91 { 92 struct netmsg_pru_abort msg; 93 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_abort; 94 95 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 96 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 97 msg.base.lmsg.ms_flags |= MSGF_SYNC; 98 func((netmsg_t)&msg); 99 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 100 sofree(msg.base.nm_so); 101 } 102 103 /* 104 * WARNING! Synchronous call from user context 105 */ 106 int 107 so_pru_accept_direct(struct socket *so, struct sockaddr **nam) 108 { 109 struct netmsg_pru_accept msg; 110 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_accept; 111 112 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 113 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 114 msg.base.lmsg.ms_flags |= MSGF_SYNC; 115 msg.nm_nam = nam; 116 func((netmsg_t)&msg); 117 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 118 return(msg.base.lmsg.ms_error); 119 } 120 121 int 122 so_pru_attach(struct socket *so, int proto, struct pru_attach_info *ai) 123 { 124 struct netmsg_pru_attach msg; 125 int error; 126 127 netmsg_init(&msg.base, so, &curthread->td_msgport, 128 0, so->so_proto->pr_usrreqs->pru_attach); 129 msg.nm_proto = proto; 130 msg.nm_ai = ai; 131 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 132 return (error); 133 } 134 135 int 136 so_pru_attach_direct(struct socket *so, int proto, struct pru_attach_info *ai) 137 { 138 struct netmsg_pru_attach msg; 139 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_attach; 140 141 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 142 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 143 msg.base.lmsg.ms_flags |= MSGF_SYNC; 144 msg.nm_proto = proto; 145 msg.nm_ai = ai; 146 func((netmsg_t)&msg); 147 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 148 return(msg.base.lmsg.ms_error); 149 } 150 151 /* 152 * NOTE: If the target port changes the bind operation will deal with it. 153 */ 154 int 155 so_pru_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 156 { 157 struct netmsg_pru_bind msg; 158 int error; 159 160 netmsg_init(&msg.base, so, &curthread->td_msgport, 161 0, so->so_proto->pr_usrreqs->pru_bind); 162 msg.nm_nam = nam; 163 msg.nm_td = td; /* used only for prison_ip() */ 164 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 165 return (error); 166 } 167 168 int 169 so_pru_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 170 { 171 struct netmsg_pru_connect msg; 172 int error; 173 174 netmsg_init(&msg.base, so, &curthread->td_msgport, 175 0, so->so_proto->pr_usrreqs->pru_connect); 176 msg.nm_nam = nam; 177 msg.nm_td = td; 178 msg.nm_m = NULL; 179 msg.nm_flags = 0; 180 msg.nm_reconnect = 0; 181 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 182 return (error); 183 } 184 185 int 186 so_pru_connect2(struct socket *so1, struct socket *so2) 187 { 188 struct netmsg_pru_connect2 msg; 189 int error; 190 191 netmsg_init(&msg.base, so1, &curthread->td_msgport, 192 0, so1->so_proto->pr_usrreqs->pru_connect2); 193 msg.nm_so1 = so1; 194 msg.nm_so2 = so2; 195 error = lwkt_domsg(so1->so_port, &msg.base.lmsg, 0); 196 return (error); 197 } 198 199 /* 200 * WARNING! Synchronous call from user context. Control function may do 201 * copyin/copyout. 202 */ 203 int 204 so_pru_control_direct(struct socket *so, u_long cmd, caddr_t data, 205 struct ifnet *ifp) 206 { 207 struct netmsg_pru_control msg; 208 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_control; 209 210 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 211 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 212 msg.base.lmsg.ms_flags |= MSGF_SYNC; 213 msg.nm_cmd = cmd; 214 msg.nm_data = data; 215 msg.nm_ifp = ifp; 216 msg.nm_td = curthread; 217 func((netmsg_t)&msg); 218 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 219 return(msg.base.lmsg.ms_error); 220 } 221 222 int 223 so_pru_detach(struct socket *so) 224 { 225 struct netmsg_pru_detach msg; 226 int error; 227 228 netmsg_init(&msg.base, so, &curthread->td_msgport, 229 0, so->so_proto->pr_usrreqs->pru_detach); 230 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 231 return (error); 232 } 233 234 int 235 so_pru_disconnect(struct socket *so) 236 { 237 struct netmsg_pru_disconnect msg; 238 int error; 239 240 netmsg_init(&msg.base, so, &curthread->td_msgport, 241 0, so->so_proto->pr_usrreqs->pru_disconnect); 242 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 243 return (error); 244 } 245 246 int 247 so_pru_listen(struct socket *so, struct thread *td) 248 { 249 struct netmsg_pru_listen msg; 250 int error; 251 252 netmsg_init(&msg.base, so, &curthread->td_msgport, 253 0, so->so_proto->pr_usrreqs->pru_listen); 254 msg.nm_td = td; /* used only for prison_ip() XXX JH */ 255 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 256 return (error); 257 } 258 259 int 260 so_pru_peeraddr(struct socket *so, struct sockaddr **nam) 261 { 262 struct netmsg_pru_peeraddr msg; 263 int error; 264 265 netmsg_init(&msg.base, so, &curthread->td_msgport, 266 0, so->so_proto->pr_usrreqs->pru_peeraddr); 267 msg.nm_nam = nam; 268 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 269 return (error); 270 } 271 272 int 273 so_pru_rcvd(struct socket *so, int flags) 274 { 275 struct netmsg_pru_rcvd msg; 276 int error; 277 278 netmsg_init(&msg.base, so, &curthread->td_msgport, 279 0, so->so_proto->pr_usrreqs->pru_rcvd); 280 msg.nm_flags = flags; 281 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 282 return (error); 283 } 284 285 int 286 so_pru_rcvoob(struct socket *so, struct mbuf *m, int flags) 287 { 288 struct netmsg_pru_rcvoob msg; 289 int error; 290 291 netmsg_init(&msg.base, so, &curthread->td_msgport, 292 0, so->so_proto->pr_usrreqs->pru_rcvoob); 293 msg.nm_m = m; 294 msg.nm_flags = flags; 295 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 296 return (error); 297 } 298 299 /* 300 * NOTE: If the target port changes the implied connect will deal with it. 301 */ 302 int 303 so_pru_send(struct socket *so, int flags, struct mbuf *m, 304 struct sockaddr *addr, struct mbuf *control, struct thread *td) 305 { 306 struct netmsg_pru_send msg; 307 int error; 308 309 netmsg_init(&msg.base, so, &curthread->td_msgport, 310 0, so->so_proto->pr_usrreqs->pru_send); 311 msg.nm_flags = flags; 312 msg.nm_m = m; 313 msg.nm_addr = addr; 314 msg.nm_control = control; 315 msg.nm_td = td; 316 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 317 return (error); 318 } 319 320 static void 321 so_pru_sync_handler(netmsg_t msg) 322 { 323 lwkt_replymsg(&msg->lmsg, 0); 324 } 325 326 void 327 so_pru_sync(struct socket *so) 328 { 329 struct netmsg_base msg; 330 331 netmsg_init(&msg, so, &curthread->td_msgport, 0, 332 so_pru_sync_handler); 333 lwkt_domsg(so->so_port, &msg.lmsg, 0); 334 } 335 336 void 337 so_pru_send_async(struct socket *so, int flags, struct mbuf *m, 338 struct sockaddr *addr, struct mbuf *control, struct thread *td) 339 { 340 struct netmsg_pru_send *msg; 341 342 msg = &m->m_hdr.mh_sndmsg; 343 netmsg_init(&msg->base, so, &netisr_apanic_rport, 344 0, so->so_proto->pr_usrreqs->pru_send); 345 msg->nm_flags = flags | PRUS_NOREPLY; 346 msg->nm_m = m; 347 msg->nm_addr = addr; 348 msg->nm_control = control; 349 msg->nm_td = td; 350 lwkt_sendmsg(so->so_port, &msg->base.lmsg); 351 } 352 353 int 354 so_pru_sense(struct socket *so, struct stat *sb) 355 { 356 struct netmsg_pru_sense msg; 357 int error; 358 359 netmsg_init(&msg.base, so, &curthread->td_msgport, 360 0, so->so_proto->pr_usrreqs->pru_sense); 361 msg.nm_stat = sb; 362 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 363 return (error); 364 } 365 366 int 367 so_pru_shutdown(struct socket *so) 368 { 369 struct netmsg_pru_shutdown msg; 370 int error; 371 372 netmsg_init(&msg.base, so, &curthread->td_msgport, 373 0, so->so_proto->pr_usrreqs->pru_shutdown); 374 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 375 return (error); 376 } 377 378 int 379 so_pru_sockaddr(struct socket *so, struct sockaddr **nam) 380 { 381 struct netmsg_pru_sockaddr msg; 382 int error; 383 384 netmsg_init(&msg.base, so, &curthread->td_msgport, 385 0, so->so_proto->pr_usrreqs->pru_sockaddr); 386 msg.nm_nam = nam; 387 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 388 return (error); 389 } 390 391 int 392 so_pr_ctloutput(struct socket *so, struct sockopt *sopt) 393 { 394 struct netmsg_pr_ctloutput msg; 395 int error; 396 397 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 398 netmsg_init(&msg.base, so, &curthread->td_msgport, 399 0, so->so_proto->pr_ctloutput); 400 msg.nm_sopt = sopt; 401 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 402 return (error); 403 } 404 405 /* 406 * Protocol control input, typically via icmp. 407 * 408 * If the protocol pr_ctlport is not NULL we call it to figure out the 409 * protocol port. If NULL is returned we can just return, otherwise 410 * we issue a netmsg to call pr_ctlinput in the proper thread. 411 * 412 * This must be done synchronously as arg and/or extra may point to 413 * temporary data. 414 */ 415 void 416 so_pru_ctlinput(struct protosw *pr, int cmd, struct sockaddr *arg, void *extra) 417 { 418 struct netmsg_pru_ctlinput msg; 419 lwkt_port_t port; 420 421 if (pr->pr_ctlport == NULL) 422 return; 423 KKASSERT(pr->pr_ctlinput != NULL); 424 port = pr->pr_ctlport(cmd, arg, extra); 425 if (port == NULL) 426 return; 427 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 428 0, pr->pr_ctlinput); 429 msg.nm_cmd = cmd; 430 msg.nm_arg = arg; 431 msg.nm_extra = extra; 432 lwkt_domsg(port, &msg.base.lmsg, 0); 433 } 434 435 /* 436 * If we convert all the protosw pr_ functions for all the protocols 437 * to take a message directly, this layer can go away. For the moment 438 * our dispatcher ignores the return value, but since we are handling 439 * the replymsg ourselves we return EASYNC by convention. 440 */ 441 442 /* 443 * Handle a predicate event request. This function is only called once 444 * when the predicate message queueing request is received. 445 */ 446 void 447 netmsg_so_notify(netmsg_t msg) 448 { 449 struct lwkt_token *tok; 450 struct signalsockbuf *ssb; 451 452 ssb = (msg->notify.nm_etype & NM_REVENT) ? 453 &msg->base.nm_so->so_rcv : 454 &msg->base.nm_so->so_snd; 455 456 /* 457 * Reply immediately if the event has occured, otherwise queue the 458 * request. 459 * 460 * NOTE: Socket can change if this is an accept predicate so cache 461 * the token. 462 */ 463 tok = lwkt_token_pool_lookup(msg->base.nm_so); 464 lwkt_gettoken(tok); 465 if (msg->notify.nm_predicate(&msg->notify)) { 466 lwkt_reltoken(tok); 467 lwkt_replymsg(&msg->base.lmsg, 468 msg->base.lmsg.ms_error); 469 } else { 470 TAILQ_INSERT_TAIL(&ssb->ssb_kq.ki_mlist, &msg->notify, nm_list); 471 atomic_set_int(&ssb->ssb_flags, SSB_MEVENT); 472 lwkt_reltoken(tok); 473 } 474 } 475 476 /* 477 * Called by doio when trying to abort a netmsg_so_notify message. 478 * Unlike the other functions this one is dispatched directly by 479 * the LWKT subsystem, so it takes a lwkt_msg_t as an argument. 480 * 481 * The original message, lmsg, is under the control of the caller and 482 * will not be destroyed until we return so we can safely reference it 483 * in our synchronous abort request. 484 * 485 * This part of the abort request occurs on the originating cpu which 486 * means we may race the message flags and the original message may 487 * not even have been processed by the target cpu yet. 488 */ 489 void 490 netmsg_so_notify_doabort(lwkt_msg_t lmsg) 491 { 492 struct netmsg_so_notify_abort msg; 493 494 if ((lmsg->ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) { 495 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 496 0, netmsg_so_notify_abort); 497 msg.nm_notifymsg = (void *)lmsg; 498 lwkt_domsg(lmsg->ms_target_port, &msg.base.lmsg, 0); 499 } 500 } 501 502 /* 503 * Predicate requests can be aborted. This function is only called once 504 * and will interlock against processing/reply races (since such races 505 * occur on the same thread that controls the port where the abort is 506 * requeued). 507 * 508 * This part of the abort request occurs on the target cpu. The message 509 * flags must be tested again in case the test that we did on the 510 * originating cpu raced. Since messages are handled in sequence, the 511 * original message will have already been handled by the loop and either 512 * replied to or queued. 513 * 514 * We really only need to interlock with MSGF_REPLY (a bit that is set on 515 * our cpu when we reply). Note that MSGF_DONE is not set until the 516 * reply reaches the originating cpu. Test both bits anyway. 517 */ 518 void 519 netmsg_so_notify_abort(netmsg_t msg) 520 { 521 struct netmsg_so_notify_abort *abrtmsg = &msg->notify_abort; 522 struct netmsg_so_notify *nmsg = abrtmsg->nm_notifymsg; 523 struct signalsockbuf *ssb; 524 525 /* 526 * The original notify message is not destroyed until after the 527 * abort request is returned, so we can check its state. 528 */ 529 lwkt_getpooltoken(nmsg->base.nm_so); 530 if ((nmsg->base.lmsg.ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) { 531 ssb = (nmsg->nm_etype & NM_REVENT) ? 532 &nmsg->base.nm_so->so_rcv : 533 &nmsg->base.nm_so->so_snd; 534 TAILQ_REMOVE(&ssb->ssb_kq.ki_mlist, nmsg, nm_list); 535 lwkt_relpooltoken(nmsg->base.nm_so); 536 lwkt_replymsg(&nmsg->base.lmsg, EINTR); 537 } else { 538 lwkt_relpooltoken(nmsg->base.nm_so); 539 } 540 541 /* 542 * Reply to the abort message 543 */ 544 lwkt_replymsg(&abrtmsg->base.lmsg, 0); 545 } 546