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 int 104 so_pru_accept(struct socket *so, struct sockaddr **nam) 105 { 106 struct netmsg_pru_accept msg; 107 108 netmsg_init(&msg.base, so, &curthread->td_msgport, 109 0, so->so_proto->pr_usrreqs->pru_accept); 110 msg.nm_nam = nam; 111 112 return lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 113 } 114 115 int 116 so_pru_attach(struct socket *so, int proto, struct pru_attach_info *ai) 117 { 118 struct netmsg_pru_attach msg; 119 int error; 120 121 netmsg_init(&msg.base, so, &curthread->td_msgport, 122 0, so->so_proto->pr_usrreqs->pru_attach); 123 msg.nm_proto = proto; 124 msg.nm_ai = ai; 125 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 126 return (error); 127 } 128 129 int 130 so_pru_attach_direct(struct socket *so, int proto, struct pru_attach_info *ai) 131 { 132 struct netmsg_pru_attach msg; 133 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_attach; 134 135 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 136 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 137 msg.base.lmsg.ms_flags |= MSGF_SYNC; 138 msg.nm_proto = proto; 139 msg.nm_ai = ai; 140 func((netmsg_t)&msg); 141 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 142 return(msg.base.lmsg.ms_error); 143 } 144 145 /* 146 * NOTE: If the target port changes the bind operation will deal with it. 147 */ 148 int 149 so_pru_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 150 { 151 struct netmsg_pru_bind msg; 152 int error; 153 154 netmsg_init(&msg.base, so, &curthread->td_msgport, 155 0, so->so_proto->pr_usrreqs->pru_bind); 156 msg.nm_nam = nam; 157 msg.nm_td = td; /* used only for prison_ip() */ 158 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 159 return (error); 160 } 161 162 int 163 so_pru_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 164 { 165 struct netmsg_pru_connect msg; 166 int error; 167 168 netmsg_init(&msg.base, so, &curthread->td_msgport, 169 0, so->so_proto->pr_usrreqs->pru_connect); 170 msg.nm_nam = nam; 171 msg.nm_td = td; 172 msg.nm_m = NULL; 173 msg.nm_flags = 0; 174 msg.nm_reconnect = 0; 175 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 176 return (error); 177 } 178 179 int 180 so_pru_connect2(struct socket *so1, struct socket *so2) 181 { 182 struct netmsg_pru_connect2 msg; 183 int error; 184 185 netmsg_init(&msg.base, so1, &curthread->td_msgport, 186 0, so1->so_proto->pr_usrreqs->pru_connect2); 187 msg.nm_so1 = so1; 188 msg.nm_so2 = so2; 189 error = lwkt_domsg(so1->so_port, &msg.base.lmsg, 0); 190 return (error); 191 } 192 193 /* 194 * WARNING! Synchronous call from user context. Control function may do 195 * copyin/copyout. 196 */ 197 int 198 so_pru_control_direct(struct socket *so, u_long cmd, caddr_t data, 199 struct ifnet *ifp) 200 { 201 struct netmsg_pru_control msg; 202 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_control; 203 204 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 205 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 206 msg.base.lmsg.ms_flags |= MSGF_SYNC; 207 msg.nm_cmd = cmd; 208 msg.nm_data = data; 209 msg.nm_ifp = ifp; 210 msg.nm_td = curthread; 211 func((netmsg_t)&msg); 212 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 213 return(msg.base.lmsg.ms_error); 214 } 215 216 int 217 so_pru_detach(struct socket *so) 218 { 219 struct netmsg_pru_detach msg; 220 int error; 221 222 netmsg_init(&msg.base, so, &curthread->td_msgport, 223 0, so->so_proto->pr_usrreqs->pru_detach); 224 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 225 return (error); 226 } 227 228 void 229 so_pru_detach_direct(struct socket *so) 230 { 231 struct netmsg_pru_detach msg; 232 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_detach; 233 234 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 235 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 236 msg.base.lmsg.ms_flags |= MSGF_SYNC; 237 func((netmsg_t)&msg); 238 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 239 } 240 241 int 242 so_pru_disconnect(struct socket *so) 243 { 244 struct netmsg_pru_disconnect msg; 245 int error; 246 247 netmsg_init(&msg.base, so, &curthread->td_msgport, 248 0, so->so_proto->pr_usrreqs->pru_disconnect); 249 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 250 return (error); 251 } 252 253 void 254 so_pru_disconnect_direct(struct socket *so) 255 { 256 struct netmsg_pru_disconnect msg; 257 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_disconnect; 258 259 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 260 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 261 msg.base.lmsg.ms_flags |= MSGF_SYNC; 262 func((netmsg_t)&msg); 263 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 264 } 265 266 int 267 so_pru_listen(struct socket *so, struct thread *td) 268 { 269 struct netmsg_pru_listen msg; 270 int error; 271 272 netmsg_init(&msg.base, so, &curthread->td_msgport, 273 0, so->so_proto->pr_usrreqs->pru_listen); 274 msg.nm_td = td; /* used only for prison_ip() XXX JH */ 275 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 276 return (error); 277 } 278 279 int 280 so_pru_peeraddr(struct socket *so, struct sockaddr **nam) 281 { 282 struct netmsg_pru_peeraddr msg; 283 int error; 284 285 netmsg_init(&msg.base, so, &curthread->td_msgport, 286 0, so->so_proto->pr_usrreqs->pru_peeraddr); 287 msg.nm_nam = nam; 288 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 289 return (error); 290 } 291 292 int 293 so_pru_rcvd(struct socket *so, int flags) 294 { 295 struct netmsg_pru_rcvd msg; 296 int error; 297 298 netmsg_init(&msg.base, so, &curthread->td_msgport, 299 0, so->so_proto->pr_usrreqs->pru_rcvd); 300 msg.nm_flags = flags; 301 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 302 return (error); 303 } 304 305 int 306 so_pru_rcvoob(struct socket *so, struct mbuf *m, int flags) 307 { 308 struct netmsg_pru_rcvoob msg; 309 int error; 310 311 netmsg_init(&msg.base, so, &curthread->td_msgport, 312 0, so->so_proto->pr_usrreqs->pru_rcvoob); 313 msg.nm_m = m; 314 msg.nm_flags = flags; 315 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 316 return (error); 317 } 318 319 /* 320 * NOTE: If the target port changes the implied connect will deal with it. 321 */ 322 int 323 so_pru_send(struct socket *so, int flags, struct mbuf *m, 324 struct sockaddr *addr, struct mbuf *control, struct thread *td) 325 { 326 struct netmsg_pru_send msg; 327 int error; 328 329 netmsg_init(&msg.base, so, &curthread->td_msgport, 330 0, so->so_proto->pr_usrreqs->pru_send); 331 msg.nm_flags = flags; 332 msg.nm_m = m; 333 msg.nm_addr = addr; 334 msg.nm_control = control; 335 msg.nm_td = td; 336 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 337 return (error); 338 } 339 340 void 341 so_pru_sync(struct socket *so) 342 { 343 struct netmsg_base msg; 344 345 netmsg_init(&msg, so, &curthread->td_msgport, 0, 346 netmsg_sync_handler); 347 lwkt_domsg(so->so_port, &msg.lmsg, 0); 348 } 349 350 void 351 so_pru_send_async(struct socket *so, int flags, struct mbuf *m, 352 struct sockaddr *addr0, struct mbuf *control, struct thread *td) 353 { 354 struct netmsg_pru_send *msg; 355 struct sockaddr *addr = NULL; 356 357 KASSERT(so->so_proto->pr_flags & PR_ASYNC_SEND, 358 ("async pru_send is not supported\n")); 359 360 flags |= PRUS_NOREPLY; 361 if (addr0 != NULL) { 362 addr = kmalloc(addr0->sa_len, M_SONAME, M_WAITOK); 363 memcpy(addr, addr0, addr0->sa_len); 364 flags |= PRUS_FREEADDR; 365 } 366 367 msg = &m->m_hdr.mh_sndmsg; 368 netmsg_init(&msg->base, so, &netisr_apanic_rport, 369 0, so->so_proto->pr_usrreqs->pru_send); 370 msg->nm_flags = flags; 371 msg->nm_m = m; 372 msg->nm_addr = addr; 373 msg->nm_control = control; 374 msg->nm_td = td; 375 lwkt_sendmsg(so->so_port, &msg->base.lmsg); 376 } 377 378 int 379 so_pru_sense(struct socket *so, struct stat *sb) 380 { 381 struct netmsg_pru_sense msg; 382 int error; 383 384 netmsg_init(&msg.base, so, &curthread->td_msgport, 385 0, so->so_proto->pr_usrreqs->pru_sense); 386 msg.nm_stat = sb; 387 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 388 return (error); 389 } 390 391 int 392 so_pru_shutdown(struct socket *so) 393 { 394 struct netmsg_pru_shutdown msg; 395 int error; 396 397 netmsg_init(&msg.base, so, &curthread->td_msgport, 398 0, so->so_proto->pr_usrreqs->pru_shutdown); 399 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 400 return (error); 401 } 402 403 int 404 so_pru_sockaddr(struct socket *so, struct sockaddr **nam) 405 { 406 struct netmsg_pru_sockaddr msg; 407 int error; 408 409 netmsg_init(&msg.base, so, &curthread->td_msgport, 410 0, so->so_proto->pr_usrreqs->pru_sockaddr); 411 msg.nm_nam = nam; 412 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 413 return (error); 414 } 415 416 int 417 so_pr_ctloutput(struct socket *so, struct sockopt *sopt) 418 { 419 struct netmsg_pr_ctloutput msg; 420 int error; 421 422 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 423 netmsg_init(&msg.base, so, &curthread->td_msgport, 424 0, so->so_proto->pr_ctloutput); 425 msg.nm_sopt = sopt; 426 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 427 return (error); 428 } 429 430 /* 431 * Protocol control input, typically via icmp. 432 * 433 * If the protocol pr_ctlport is not NULL we call it to figure out the 434 * protocol port. If NULL is returned we can just return, otherwise 435 * we issue a netmsg to call pr_ctlinput in the proper thread. 436 * 437 * This must be done synchronously as arg and/or extra may point to 438 * temporary data. 439 */ 440 void 441 so_pru_ctlinput(struct protosw *pr, int cmd, struct sockaddr *arg, void *extra) 442 { 443 struct netmsg_pru_ctlinput msg; 444 lwkt_port_t port; 445 446 if (pr->pr_ctlport == NULL) 447 return; 448 KKASSERT(pr->pr_ctlinput != NULL); 449 port = pr->pr_ctlport(cmd, arg, extra); 450 if (port == NULL) 451 return; 452 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 453 0, pr->pr_ctlinput); 454 msg.nm_cmd = cmd; 455 msg.nm_arg = arg; 456 msg.nm_extra = extra; 457 lwkt_domsg(port, &msg.base.lmsg, 0); 458 } 459 460 /* 461 * If we convert all the protosw pr_ functions for all the protocols 462 * to take a message directly, this layer can go away. For the moment 463 * our dispatcher ignores the return value, but since we are handling 464 * the replymsg ourselves we return EASYNC by convention. 465 */ 466 467 /* 468 * Handle a predicate event request. This function is only called once 469 * when the predicate message queueing request is received. 470 */ 471 void 472 netmsg_so_notify(netmsg_t msg) 473 { 474 struct lwkt_token *tok; 475 struct signalsockbuf *ssb; 476 477 ssb = (msg->notify.nm_etype & NM_REVENT) ? 478 &msg->base.nm_so->so_rcv : 479 &msg->base.nm_so->so_snd; 480 481 /* 482 * Reply immediately if the event has occured, otherwise queue the 483 * request. 484 * 485 * NOTE: Socket can change if this is an accept predicate so cache 486 * the token. 487 */ 488 tok = lwkt_token_pool_lookup(msg->base.nm_so); 489 lwkt_gettoken(tok); 490 if (msg->notify.nm_predicate(&msg->notify)) { 491 lwkt_reltoken(tok); 492 lwkt_replymsg(&msg->base.lmsg, 493 msg->base.lmsg.ms_error); 494 } else { 495 TAILQ_INSERT_TAIL(&ssb->ssb_kq.ki_mlist, &msg->notify, nm_list); 496 atomic_set_int(&ssb->ssb_flags, SSB_MEVENT); 497 lwkt_reltoken(tok); 498 } 499 } 500 501 /* 502 * Called by doio when trying to abort a netmsg_so_notify message. 503 * Unlike the other functions this one is dispatched directly by 504 * the LWKT subsystem, so it takes a lwkt_msg_t as an argument. 505 * 506 * The original message, lmsg, is under the control of the caller and 507 * will not be destroyed until we return so we can safely reference it 508 * in our synchronous abort request. 509 * 510 * This part of the abort request occurs on the originating cpu which 511 * means we may race the message flags and the original message may 512 * not even have been processed by the target cpu yet. 513 */ 514 void 515 netmsg_so_notify_doabort(lwkt_msg_t lmsg) 516 { 517 struct netmsg_so_notify_abort msg; 518 519 if ((lmsg->ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) { 520 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 521 0, netmsg_so_notify_abort); 522 msg.nm_notifymsg = (void *)lmsg; 523 lwkt_domsg(lmsg->ms_target_port, &msg.base.lmsg, 0); 524 } 525 } 526 527 /* 528 * Predicate requests can be aborted. This function is only called once 529 * and will interlock against processing/reply races (since such races 530 * occur on the same thread that controls the port where the abort is 531 * requeued). 532 * 533 * This part of the abort request occurs on the target cpu. The message 534 * flags must be tested again in case the test that we did on the 535 * originating cpu raced. Since messages are handled in sequence, the 536 * original message will have already been handled by the loop and either 537 * replied to or queued. 538 * 539 * We really only need to interlock with MSGF_REPLY (a bit that is set on 540 * our cpu when we reply). Note that MSGF_DONE is not set until the 541 * reply reaches the originating cpu. Test both bits anyway. 542 */ 543 void 544 netmsg_so_notify_abort(netmsg_t msg) 545 { 546 struct netmsg_so_notify_abort *abrtmsg = &msg->notify_abort; 547 struct netmsg_so_notify *nmsg = abrtmsg->nm_notifymsg; 548 struct signalsockbuf *ssb; 549 550 /* 551 * The original notify message is not destroyed until after the 552 * abort request is returned, so we can check its state. 553 */ 554 lwkt_getpooltoken(nmsg->base.nm_so); 555 if ((nmsg->base.lmsg.ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) { 556 ssb = (nmsg->nm_etype & NM_REVENT) ? 557 &nmsg->base.nm_so->so_rcv : 558 &nmsg->base.nm_so->so_snd; 559 TAILQ_REMOVE(&ssb->ssb_kq.ki_mlist, nmsg, nm_list); 560 lwkt_relpooltoken(nmsg->base.nm_so); 561 lwkt_replymsg(&nmsg->base.lmsg, EINTR); 562 } else { 563 lwkt_relpooltoken(nmsg->base.nm_so); 564 } 565 566 /* 567 * Reply to the abort message 568 */ 569 lwkt_replymsg(&abrtmsg->base.lmsg, 0); 570 } 571