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 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/kernel.h> 37 #include <sys/msgport.h> 38 #include <sys/protosw.h> 39 #include <sys/socket.h> 40 #include <sys/socketvar.h> 41 #include <sys/socketops.h> 42 #include <sys/thread.h> 43 #include <sys/thread2.h> 44 #include <sys/msgport2.h> 45 #include <sys/spinlock2.h> 46 #include <sys/mbuf.h> 47 #include <vm/pmap.h> 48 #include <net/netmsg2.h> 49 50 #include <net/netisr.h> 51 #include <net/netmsg.h> 52 53 /* 54 * Abort a socket and free it. Called from soabort() only. soabort() 55 * got a ref on the socket which we must free on reply. 56 */ 57 void 58 so_pru_abort(struct socket *so) 59 { 60 struct netmsg_pru_abort msg; 61 62 netmsg_init(&msg.base, so, &curthread->td_msgport, 63 0, so->so_proto->pr_usrreqs->pru_abort); 64 (void)lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 65 sofree(msg.base.nm_so); 66 } 67 68 /* 69 * Abort a socket and free it, asynchronously. Called from 70 * soaborta() only. soaborta() got a ref on the socket which we must 71 * free on reply. 72 */ 73 void 74 so_pru_aborta(struct socket *so) 75 { 76 struct netmsg_pru_abort *msg; 77 78 msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_WAITOK | M_ZERO); 79 netmsg_init(&msg->base, so, &netisr_afree_free_so_rport, 80 0, so->so_proto->pr_usrreqs->pru_abort); 81 lwkt_sendmsg(so->so_port, &msg->base.lmsg); 82 } 83 84 /* 85 * Abort a socket and free it. Called from soabort_oncpu() only. 86 * Caller must make sure that the current CPU is inpcb's owner CPU. 87 */ 88 void 89 so_pru_abort_oncpu(struct socket *so) 90 { 91 struct netmsg_pru_abort msg; 92 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_abort; 93 94 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 95 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 96 msg.base.lmsg.ms_flags |= MSGF_SYNC; 97 func((netmsg_t)&msg); 98 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 99 sofree(msg.base.nm_so); 100 } 101 102 int 103 so_pru_accept(struct socket *so, struct sockaddr **nam) 104 { 105 struct netmsg_pru_accept msg; 106 107 netmsg_init(&msg.base, so, &curthread->td_msgport, 108 0, so->so_proto->pr_usrreqs->pru_accept); 109 msg.nm_nam = nam; 110 111 return lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 112 } 113 114 int 115 so_pru_attach(struct socket *so, int proto, struct pru_attach_info *ai) 116 { 117 struct netmsg_pru_attach msg; 118 int error; 119 120 netmsg_init(&msg.base, so, &curthread->td_msgport, 121 0, so->so_proto->pr_usrreqs->pru_attach); 122 msg.nm_proto = proto; 123 msg.nm_ai = ai; 124 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 125 return (error); 126 } 127 128 int 129 so_pru_attach_direct(struct socket *so, int proto, struct pru_attach_info *ai) 130 { 131 struct netmsg_pru_attach msg; 132 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_attach; 133 134 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 135 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 136 msg.base.lmsg.ms_flags |= MSGF_SYNC; 137 msg.nm_proto = proto; 138 msg.nm_ai = ai; 139 func((netmsg_t)&msg); 140 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 141 return(msg.base.lmsg.ms_error); 142 } 143 144 /* 145 * NOTE: If the target port changes the bind operation will deal with it. 146 */ 147 int 148 so_pru_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 149 { 150 struct netmsg_pru_bind msg; 151 int error; 152 153 netmsg_init(&msg.base, so, &curthread->td_msgport, 154 0, so->so_proto->pr_usrreqs->pru_bind); 155 msg.nm_nam = nam; 156 msg.nm_td = td; /* used only for prison_ip() */ 157 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 158 return (error); 159 } 160 161 int 162 so_pru_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 163 { 164 struct netmsg_pru_connect msg; 165 int error; 166 167 netmsg_init(&msg.base, so, &curthread->td_msgport, 168 0, so->so_proto->pr_usrreqs->pru_connect); 169 msg.nm_nam = nam; 170 msg.nm_td = td; 171 msg.nm_m = NULL; 172 msg.nm_flags = 0; 173 msg.nm_reconnect = 0; 174 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 175 return (error); 176 } 177 178 int 179 so_pru_connect2(struct socket *so1, struct socket *so2) 180 { 181 struct netmsg_pru_connect2 msg; 182 int error; 183 184 netmsg_init(&msg.base, so1, &curthread->td_msgport, 185 0, so1->so_proto->pr_usrreqs->pru_connect2); 186 msg.nm_so1 = so1; 187 msg.nm_so2 = so2; 188 error = lwkt_domsg(so1->so_port, &msg.base.lmsg, 0); 189 return (error); 190 } 191 192 /* 193 * WARNING! Synchronous call from user context. Control function may do 194 * copyin/copyout. 195 */ 196 int 197 so_pru_control_direct(struct socket *so, u_long cmd, caddr_t data, 198 struct ifnet *ifp) 199 { 200 struct netmsg_pru_control msg; 201 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_control; 202 203 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 204 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 205 msg.base.lmsg.ms_flags |= MSGF_SYNC; 206 msg.nm_cmd = cmd; 207 msg.nm_data = data; 208 msg.nm_ifp = ifp; 209 msg.nm_td = curthread; 210 func((netmsg_t)&msg); 211 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 212 return(msg.base.lmsg.ms_error); 213 } 214 215 int 216 so_pru_detach(struct socket *so) 217 { 218 struct netmsg_pru_detach msg; 219 int error; 220 221 netmsg_init(&msg.base, so, &curthread->td_msgport, 222 0, so->so_proto->pr_usrreqs->pru_detach); 223 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 224 return (error); 225 } 226 227 void 228 so_pru_detach_direct(struct socket *so) 229 { 230 struct netmsg_pru_detach msg; 231 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_detach; 232 233 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 234 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 235 msg.base.lmsg.ms_flags |= MSGF_SYNC; 236 func((netmsg_t)&msg); 237 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 238 } 239 240 int 241 so_pru_disconnect(struct socket *so) 242 { 243 struct netmsg_pru_disconnect msg; 244 int error; 245 246 netmsg_init(&msg.base, so, &curthread->td_msgport, 247 0, so->so_proto->pr_usrreqs->pru_disconnect); 248 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 249 return (error); 250 } 251 252 void 253 so_pru_disconnect_direct(struct socket *so) 254 { 255 struct netmsg_pru_disconnect msg; 256 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_disconnect; 257 258 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 259 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 260 msg.base.lmsg.ms_flags |= MSGF_SYNC; 261 func((netmsg_t)&msg); 262 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 263 } 264 265 int 266 so_pru_listen(struct socket *so, struct thread *td) 267 { 268 struct netmsg_pru_listen msg; 269 int error; 270 271 netmsg_init(&msg.base, so, &curthread->td_msgport, 272 0, so->so_proto->pr_usrreqs->pru_listen); 273 msg.nm_td = td; /* used only for prison_ip() XXX JH */ 274 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 275 return (error); 276 } 277 278 int 279 so_pru_peeraddr(struct socket *so, struct sockaddr **nam) 280 { 281 struct netmsg_pru_peeraddr msg; 282 int error; 283 284 netmsg_init(&msg.base, so, &curthread->td_msgport, 285 0, so->so_proto->pr_usrreqs->pru_peeraddr); 286 msg.nm_nam = nam; 287 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 288 return (error); 289 } 290 291 int 292 so_pru_rcvd(struct socket *so, int flags) 293 { 294 struct netmsg_pru_rcvd msg; 295 int error; 296 297 netmsg_init(&msg.base, so, &curthread->td_msgport, 298 0, so->so_proto->pr_usrreqs->pru_rcvd); 299 msg.nm_flags = flags; 300 msg.nm_pru_flags = 0; 301 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 302 return (error); 303 } 304 305 void 306 so_pru_rcvd_async(struct socket *so) 307 { 308 lwkt_msg_t lmsg = &so->so_rcvd_msg.base.lmsg; 309 310 KASSERT(so->so_proto->pr_flags & PR_ASYNC_RCVD, 311 ("async pru_rcvd is not supported")); 312 313 spin_lock(&so->so_rcvd_spin); 314 if (lmsg->ms_flags & MSGF_DONE) 315 lwkt_sendmsg(so->so_port, lmsg); 316 spin_unlock(&so->so_rcvd_spin); 317 } 318 319 int 320 so_pru_rcvoob(struct socket *so, struct mbuf *m, int flags) 321 { 322 struct netmsg_pru_rcvoob msg; 323 int error; 324 325 netmsg_init(&msg.base, so, &curthread->td_msgport, 326 0, so->so_proto->pr_usrreqs->pru_rcvoob); 327 msg.nm_m = m; 328 msg.nm_flags = flags; 329 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 330 return (error); 331 } 332 333 /* 334 * NOTE: If the target port changes the implied connect will deal with it. 335 */ 336 int 337 so_pru_send(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 int error; 342 343 netmsg_init(&msg.base, so, &curthread->td_msgport, 344 0, so->so_proto->pr_usrreqs->pru_send); 345 msg.nm_flags = flags; 346 msg.nm_m = m; 347 msg.nm_addr = addr; 348 msg.nm_control = control; 349 msg.nm_td = td; 350 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 351 return (error); 352 } 353 354 void 355 so_pru_sync(struct socket *so) 356 { 357 struct netmsg_base msg; 358 359 netmsg_init(&msg, so, &curthread->td_msgport, 0, 360 netmsg_sync_handler); 361 lwkt_domsg(so->so_port, &msg.lmsg, 0); 362 } 363 364 void 365 so_pru_send_async(struct socket *so, int flags, struct mbuf *m, 366 struct sockaddr *addr0, struct mbuf *control, struct thread *td) 367 { 368 struct netmsg_pru_send *msg; 369 struct sockaddr *addr = NULL; 370 371 KASSERT(so->so_proto->pr_flags & PR_ASYNC_SEND, 372 ("async pru_send is not supported")); 373 374 flags |= PRUS_NOREPLY; 375 if (addr0 != NULL) { 376 addr = kmalloc(addr0->sa_len, M_SONAME, M_WAITOK); 377 memcpy(addr, addr0, addr0->sa_len); 378 flags |= PRUS_FREEADDR; 379 } 380 381 msg = &m->m_hdr.mh_sndmsg; 382 netmsg_init(&msg->base, so, &netisr_apanic_rport, 383 0, so->so_proto->pr_usrreqs->pru_send); 384 msg->nm_flags = flags; 385 msg->nm_m = m; 386 msg->nm_addr = addr; 387 msg->nm_control = control; 388 msg->nm_td = td; 389 lwkt_sendmsg(so->so_port, &msg->base.lmsg); 390 } 391 392 int 393 so_pru_sense(struct socket *so, struct stat *sb) 394 { 395 struct netmsg_pru_sense msg; 396 int error; 397 398 netmsg_init(&msg.base, so, &curthread->td_msgport, 399 0, so->so_proto->pr_usrreqs->pru_sense); 400 msg.nm_stat = sb; 401 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 402 return (error); 403 } 404 405 int 406 so_pru_shutdown(struct socket *so) 407 { 408 struct netmsg_pru_shutdown msg; 409 int error; 410 411 netmsg_init(&msg.base, so, &curthread->td_msgport, 412 0, so->so_proto->pr_usrreqs->pru_shutdown); 413 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 414 return (error); 415 } 416 417 int 418 so_pru_sockaddr(struct socket *so, struct sockaddr **nam) 419 { 420 struct netmsg_pru_sockaddr msg; 421 int error; 422 423 netmsg_init(&msg.base, so, &curthread->td_msgport, 424 0, so->so_proto->pr_usrreqs->pru_sockaddr); 425 msg.nm_nam = nam; 426 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 427 return (error); 428 } 429 430 int 431 so_pr_ctloutput(struct socket *so, struct sockopt *sopt) 432 { 433 struct netmsg_pr_ctloutput msg; 434 int error; 435 436 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 437 netmsg_init(&msg.base, so, &curthread->td_msgport, 438 0, so->so_proto->pr_ctloutput); 439 msg.nm_sopt = sopt; 440 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 441 return (error); 442 } 443 444 /* 445 * Protocol control input, typically via icmp. 446 * 447 * If the protocol pr_ctlport is not NULL we call it to figure out the 448 * protocol port. If NULL is returned we can just return, otherwise 449 * we issue a netmsg to call pr_ctlinput in the proper thread. 450 * 451 * This must be done synchronously as arg and/or extra may point to 452 * temporary data. 453 */ 454 void 455 so_pru_ctlinput(struct protosw *pr, int cmd, struct sockaddr *arg, void *extra) 456 { 457 struct netmsg_pru_ctlinput msg; 458 lwkt_port_t port; 459 460 if (pr->pr_ctlport == NULL) 461 return; 462 KKASSERT(pr->pr_ctlinput != NULL); 463 port = pr->pr_ctlport(cmd, arg, extra); 464 if (port == NULL) 465 return; 466 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 467 0, pr->pr_ctlinput); 468 msg.nm_cmd = cmd; 469 msg.nm_arg = arg; 470 msg.nm_extra = extra; 471 lwkt_domsg(port, &msg.base.lmsg, 0); 472 } 473 474 /* 475 * If we convert all the protosw pr_ functions for all the protocols 476 * to take a message directly, this layer can go away. For the moment 477 * our dispatcher ignores the return value, but since we are handling 478 * the replymsg ourselves we return EASYNC by convention. 479 */ 480 481 /* 482 * Handle a predicate event request. This function is only called once 483 * when the predicate message queueing request is received. 484 */ 485 void 486 netmsg_so_notify(netmsg_t msg) 487 { 488 struct lwkt_token *tok; 489 struct signalsockbuf *ssb; 490 491 ssb = (msg->notify.nm_etype & NM_REVENT) ? 492 &msg->base.nm_so->so_rcv : 493 &msg->base.nm_so->so_snd; 494 495 /* 496 * Reply immediately if the event has occured, otherwise queue the 497 * request. 498 * 499 * NOTE: Socket can change if this is an accept predicate so cache 500 * the token. 501 */ 502 tok = lwkt_token_pool_lookup(msg->base.nm_so); 503 lwkt_gettoken(tok); 504 if (msg->notify.nm_predicate(&msg->notify)) { 505 lwkt_reltoken(tok); 506 lwkt_replymsg(&msg->base.lmsg, 507 msg->base.lmsg.ms_error); 508 } else { 509 TAILQ_INSERT_TAIL(&ssb->ssb_kq.ki_mlist, &msg->notify, nm_list); 510 atomic_set_int(&ssb->ssb_flags, SSB_MEVENT); 511 lwkt_reltoken(tok); 512 } 513 } 514 515 /* 516 * Called by doio when trying to abort a netmsg_so_notify message. 517 * Unlike the other functions this one is dispatched directly by 518 * the LWKT subsystem, so it takes a lwkt_msg_t as an argument. 519 * 520 * The original message, lmsg, is under the control of the caller and 521 * will not be destroyed until we return so we can safely reference it 522 * in our synchronous abort request. 523 * 524 * This part of the abort request occurs on the originating cpu which 525 * means we may race the message flags and the original message may 526 * not even have been processed by the target cpu yet. 527 */ 528 void 529 netmsg_so_notify_doabort(lwkt_msg_t lmsg) 530 { 531 struct netmsg_so_notify_abort msg; 532 533 if ((lmsg->ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) { 534 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 535 0, netmsg_so_notify_abort); 536 msg.nm_notifymsg = (void *)lmsg; 537 lwkt_domsg(lmsg->ms_target_port, &msg.base.lmsg, 0); 538 } 539 } 540 541 /* 542 * Predicate requests can be aborted. This function is only called once 543 * and will interlock against processing/reply races (since such races 544 * occur on the same thread that controls the port where the abort is 545 * requeued). 546 * 547 * This part of the abort request occurs on the target cpu. The message 548 * flags must be tested again in case the test that we did on the 549 * originating cpu raced. Since messages are handled in sequence, the 550 * original message will have already been handled by the loop and either 551 * replied to or queued. 552 * 553 * We really only need to interlock with MSGF_REPLY (a bit that is set on 554 * our cpu when we reply). Note that MSGF_DONE is not set until the 555 * reply reaches the originating cpu. Test both bits anyway. 556 */ 557 void 558 netmsg_so_notify_abort(netmsg_t msg) 559 { 560 struct netmsg_so_notify_abort *abrtmsg = &msg->notify_abort; 561 struct netmsg_so_notify *nmsg = abrtmsg->nm_notifymsg; 562 struct signalsockbuf *ssb; 563 564 /* 565 * The original notify message is not destroyed until after the 566 * abort request is returned, so we can check its state. 567 */ 568 lwkt_getpooltoken(nmsg->base.nm_so); 569 if ((nmsg->base.lmsg.ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) { 570 ssb = (nmsg->nm_etype & NM_REVENT) ? 571 &nmsg->base.nm_so->so_rcv : 572 &nmsg->base.nm_so->so_snd; 573 TAILQ_REMOVE(&ssb->ssb_kq.ki_mlist, nmsg, nm_list); 574 lwkt_relpooltoken(nmsg->base.nm_so); 575 lwkt_replymsg(&nmsg->base.lmsg, EINTR); 576 } else { 577 lwkt_relpooltoken(nmsg->base.nm_so); 578 } 579 580 /* 581 * Reply to the abort message 582 */ 583 lwkt_replymsg(&abrtmsg->base.lmsg, 0); 584 } 585 586 void 587 so_async_rcvd_reply(struct socket *so) 588 { 589 spin_lock(&so->so_rcvd_spin); 590 lwkt_replymsg(&so->so_rcvd_msg.base.lmsg, 0); 591 spin_unlock(&so->so_rcvd_spin); 592 } 593 594 void 595 so_async_rcvd_drop(struct socket *so) 596 { 597 lwkt_msg_t lmsg = &so->so_rcvd_msg.base.lmsg; 598 599 spin_lock(&so->so_rcvd_spin); 600 if ((lmsg->ms_flags & MSGF_DONE) == 0) 601 lwkt_dropmsg(lmsg); 602 spin_unlock(&so->so_rcvd_spin); 603 } 604