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 <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. 55 */ 56 void 57 so_pru_abort(struct socket *so) 58 { 59 struct netmsg_pru_abort msg; 60 61 netmsg_init(&msg.base, so, &curthread->td_msgport, 62 0, so->so_proto->pr_usrreqs->pru_abort); 63 (void)lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 64 } 65 66 /* 67 * Abort a socket and free it, asynchronously. Called from 68 * soaborta() only. 69 */ 70 void 71 so_pru_aborta(struct socket *so) 72 { 73 struct netmsg_pru_abort *msg; 74 75 msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_WAITOK | M_ZERO); 76 netmsg_init(&msg->base, so, &netisr_afree_rport, 77 0, so->so_proto->pr_usrreqs->pru_abort); 78 lwkt_sendmsg(so->so_port, &msg->base.lmsg); 79 } 80 81 /* 82 * Abort a socket and free it. Called from soabort_oncpu() only. 83 * Caller must make sure that the current CPU is inpcb's owner CPU. 84 */ 85 void 86 so_pru_abort_oncpu(struct socket *so) 87 { 88 struct netmsg_pru_abort msg; 89 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_abort; 90 91 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 92 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 93 msg.base.lmsg.ms_flags |= MSGF_SYNC; 94 func((netmsg_t)&msg); 95 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 96 } 97 98 /* 99 * WARNING! Synchronous call from user context 100 */ 101 int 102 so_pru_accept_direct(struct socket *so, struct sockaddr **nam) 103 { 104 struct netmsg_pru_accept msg; 105 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_accept; 106 107 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 108 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 109 msg.base.lmsg.ms_flags |= MSGF_SYNC; 110 msg.nm_nam = nam; 111 func((netmsg_t)&msg); 112 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 113 return(msg.base.lmsg.ms_error); 114 } 115 116 int 117 so_pru_attach(struct socket *so, int proto, struct pru_attach_info *ai) 118 { 119 struct netmsg_pru_attach msg; 120 int error; 121 122 netmsg_init(&msg.base, so, &curthread->td_msgport, 123 0, so->so_proto->pr_usrreqs->pru_attach); 124 msg.nm_proto = proto; 125 msg.nm_ai = ai; 126 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 127 return (error); 128 } 129 130 int 131 so_pru_attach_direct(struct socket *so, int proto, struct pru_attach_info *ai) 132 { 133 struct netmsg_pru_attach msg; 134 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_attach; 135 136 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 137 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 138 msg.base.lmsg.ms_flags |= MSGF_SYNC; 139 msg.nm_proto = proto; 140 msg.nm_ai = ai; 141 func((netmsg_t)&msg); 142 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 143 return(msg.base.lmsg.ms_error); 144 } 145 146 /* 147 * NOTE: If the target port changes the bind operation will deal with it. 148 */ 149 int 150 so_pru_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 151 { 152 struct netmsg_pru_bind msg; 153 int error; 154 155 netmsg_init(&msg.base, so, &curthread->td_msgport, 156 0, so->so_proto->pr_usrreqs->pru_bind); 157 msg.nm_nam = nam; 158 msg.nm_td = td; /* used only for prison_ip() */ 159 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 160 return (error); 161 } 162 163 int 164 so_pru_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 165 { 166 struct netmsg_pru_connect msg; 167 int error; 168 169 netmsg_init(&msg.base, so, &curthread->td_msgport, 170 0, so->so_proto->pr_usrreqs->pru_connect); 171 msg.nm_nam = nam; 172 msg.nm_td = td; 173 msg.nm_m = NULL; 174 msg.nm_flags = 0; 175 msg.nm_reconnect = 0; 176 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 177 return (error); 178 } 179 180 int 181 so_pru_connect2(struct socket *so1, struct socket *so2) 182 { 183 struct netmsg_pru_connect2 msg; 184 int error; 185 186 netmsg_init(&msg.base, so1, &curthread->td_msgport, 187 0, so1->so_proto->pr_usrreqs->pru_connect2); 188 msg.nm_so1 = so1; 189 msg.nm_so2 = so2; 190 error = lwkt_domsg(so1->so_port, &msg.base.lmsg, 0); 191 return (error); 192 } 193 194 /* 195 * WARNING! Synchronous call from user context. Control function may do 196 * copyin/copyout. 197 */ 198 int 199 so_pru_control_direct(struct socket *so, u_long cmd, caddr_t data, 200 struct ifnet *ifp) 201 { 202 struct netmsg_pru_control msg; 203 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_control; 204 205 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 206 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 207 msg.base.lmsg.ms_flags |= MSGF_SYNC; 208 msg.nm_cmd = cmd; 209 msg.nm_data = data; 210 msg.nm_ifp = ifp; 211 msg.nm_td = curthread; 212 func((netmsg_t)&msg); 213 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 214 return(msg.base.lmsg.ms_error); 215 } 216 217 int 218 so_pru_detach(struct socket *so) 219 { 220 struct netmsg_pru_detach msg; 221 int error; 222 223 netmsg_init(&msg.base, so, &curthread->td_msgport, 224 0, so->so_proto->pr_usrreqs->pru_detach); 225 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 226 return (error); 227 } 228 229 int 230 so_pru_disconnect(struct socket *so) 231 { 232 struct netmsg_pru_disconnect msg; 233 int error; 234 235 netmsg_init(&msg.base, so, &curthread->td_msgport, 236 0, so->so_proto->pr_usrreqs->pru_disconnect); 237 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 238 return (error); 239 } 240 241 int 242 so_pru_listen(struct socket *so, struct thread *td) 243 { 244 struct netmsg_pru_listen msg; 245 int error; 246 247 netmsg_init(&msg.base, so, &curthread->td_msgport, 248 0, so->so_proto->pr_usrreqs->pru_listen); 249 msg.nm_td = td; /* used only for prison_ip() XXX JH */ 250 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 251 return (error); 252 } 253 254 int 255 so_pru_peeraddr(struct socket *so, struct sockaddr **nam) 256 { 257 struct netmsg_pru_peeraddr msg; 258 int error; 259 260 netmsg_init(&msg.base, so, &curthread->td_msgport, 261 0, so->so_proto->pr_usrreqs->pru_peeraddr); 262 msg.nm_nam = nam; 263 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 264 return (error); 265 } 266 267 int 268 so_pru_rcvd(struct socket *so, int flags) 269 { 270 struct netmsg_pru_rcvd msg; 271 int error; 272 273 netmsg_init(&msg.base, so, &curthread->td_msgport, 274 0, so->so_proto->pr_usrreqs->pru_rcvd); 275 msg.nm_flags = flags; 276 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 277 return (error); 278 } 279 280 int 281 so_pru_rcvoob(struct socket *so, struct mbuf *m, int flags) 282 { 283 struct netmsg_pru_rcvoob msg; 284 int error; 285 286 netmsg_init(&msg.base, so, &curthread->td_msgport, 287 0, so->so_proto->pr_usrreqs->pru_rcvoob); 288 msg.nm_m = m; 289 msg.nm_flags = flags; 290 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 291 return (error); 292 } 293 294 /* 295 * NOTE: If the target port changes the implied connect will deal with it. 296 */ 297 int 298 so_pru_send(struct socket *so, int flags, struct mbuf *m, 299 struct sockaddr *addr, struct mbuf *control, struct thread *td) 300 { 301 struct netmsg_pru_send msg; 302 int error; 303 304 netmsg_init(&msg.base, so, &curthread->td_msgport, 305 0, so->so_proto->pr_usrreqs->pru_send); 306 msg.nm_flags = flags; 307 msg.nm_m = m; 308 msg.nm_addr = addr; 309 msg.nm_control = control; 310 msg.nm_td = td; 311 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 312 return (error); 313 } 314 315 int 316 so_pru_sense(struct socket *so, struct stat *sb) 317 { 318 struct netmsg_pru_sense msg; 319 int error; 320 321 netmsg_init(&msg.base, so, &curthread->td_msgport, 322 0, so->so_proto->pr_usrreqs->pru_sense); 323 msg.nm_stat = sb; 324 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 325 return (error); 326 } 327 328 int 329 so_pru_shutdown(struct socket *so) 330 { 331 struct netmsg_pru_shutdown msg; 332 int error; 333 334 netmsg_init(&msg.base, so, &curthread->td_msgport, 335 0, so->so_proto->pr_usrreqs->pru_shutdown); 336 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 337 return (error); 338 } 339 340 int 341 so_pru_sockaddr(struct socket *so, struct sockaddr **nam) 342 { 343 struct netmsg_pru_sockaddr msg; 344 int error; 345 346 netmsg_init(&msg.base, so, &curthread->td_msgport, 347 0, so->so_proto->pr_usrreqs->pru_sockaddr); 348 msg.nm_nam = nam; 349 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 350 return (error); 351 } 352 353 int 354 so_pr_ctloutput(struct socket *so, struct sockopt *sopt) 355 { 356 struct netmsg_pr_ctloutput msg; 357 int error; 358 359 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 360 netmsg_init(&msg.base, so, &curthread->td_msgport, 361 0, so->so_proto->pr_ctloutput); 362 msg.nm_sopt = sopt; 363 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 364 return (error); 365 } 366 367 /* 368 * Protocol control input, typically via icmp. 369 * 370 * If the protocol pr_ctlport is not NULL we call it to figure out the 371 * protocol port. If NULL is returned we can just return, otherwise 372 * we issue a netmsg to call pr_ctlinput in the proper thread. 373 * 374 * This must be done synchronously as arg and/or extra may point to 375 * temporary data. 376 */ 377 void 378 so_pru_ctlinput(struct protosw *pr, int cmd, struct sockaddr *arg, void *extra) 379 { 380 struct netmsg_pru_ctlinput msg; 381 lwkt_port_t port; 382 383 if (pr->pr_ctlport == NULL) 384 return; 385 KKASSERT(pr->pr_ctlinput != NULL); 386 port = pr->pr_ctlport(cmd, arg, extra); 387 if (port == NULL) 388 return; 389 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 390 0, pr->pr_ctlinput); 391 msg.nm_cmd = cmd; 392 msg.nm_arg = arg; 393 msg.nm_extra = extra; 394 lwkt_domsg(port, &msg.base.lmsg, 0); 395 } 396 397 /* 398 * If we convert all the protosw pr_ functions for all the protocols 399 * to take a message directly, this layer can go away. For the moment 400 * our dispatcher ignores the return value, but since we are handling 401 * the replymsg ourselves we return EASYNC by convention. 402 */ 403 404 /* 405 * Handle a predicate event request. This function is only called once 406 * when the predicate message queueing request is received. 407 */ 408 void 409 netmsg_so_notify(netmsg_t msg) 410 { 411 struct signalsockbuf *ssb; 412 413 ssb = (msg->notify.nm_etype & NM_REVENT) ? 414 &msg->base.nm_so->so_rcv : 415 &msg->base.nm_so->so_snd; 416 417 /* 418 * Reply immediately if the event has occured, otherwise queue the 419 * request. 420 */ 421 if (msg->notify.nm_predicate(&msg->notify)) { 422 lwkt_replymsg(&msg->base.lmsg, 423 msg->base.lmsg.ms_error); 424 } else { 425 lwkt_gettoken(&kq_token); 426 TAILQ_INSERT_TAIL(&ssb->ssb_kq.ki_mlist, &msg->notify, nm_list); 427 atomic_set_int(&ssb->ssb_flags, SSB_MEVENT); 428 lwkt_reltoken(&kq_token); 429 } 430 } 431 432 /* 433 * Called by doio when trying to abort a netmsg_so_notify message. 434 * Unlike the other functions this one is dispatched directly by 435 * the LWKT subsystem, so it takes a lwkt_msg_t as an argument. 436 * 437 * The original message, lmsg, is under the control of the caller and 438 * will not be destroyed until we return so we can safely reference it 439 * in our synchronous abort request. 440 * 441 * This part of the abort request occurs on the originating cpu which 442 * means we may race the message flags and the original message may 443 * not even have been processed by the target cpu yet. 444 */ 445 void 446 netmsg_so_notify_doabort(lwkt_msg_t lmsg) 447 { 448 struct netmsg_so_notify_abort msg; 449 450 if ((lmsg->ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) { 451 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 452 0, netmsg_so_notify_abort); 453 msg.nm_notifymsg = (void *)lmsg; 454 lwkt_domsg(lmsg->ms_target_port, &msg.base.lmsg, 0); 455 } 456 } 457 458 /* 459 * Predicate requests can be aborted. This function is only called once 460 * and will interlock against processing/reply races (since such races 461 * occur on the same thread that controls the port where the abort is 462 * requeued). 463 * 464 * This part of the abort request occurs on the target cpu. The message 465 * flags must be tested again in case the test that we did on the 466 * originating cpu raced. Since messages are handled in sequence, the 467 * original message will have already been handled by the loop and either 468 * replied to or queued. 469 * 470 * We really only need to interlock with MSGF_REPLY (a bit that is set on 471 * our cpu when we reply). Note that MSGF_DONE is not set until the 472 * reply reaches the originating cpu. Test both bits anyway. 473 */ 474 void 475 netmsg_so_notify_abort(netmsg_t msg) 476 { 477 struct netmsg_so_notify_abort *abrtmsg = &msg->notify_abort; 478 struct netmsg_so_notify *nmsg = abrtmsg->nm_notifymsg; 479 struct signalsockbuf *ssb; 480 481 /* 482 * The original notify message is not destroyed until after the 483 * abort request is returned, so we can check its state. 484 */ 485 if ((nmsg->base.lmsg.ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) { 486 ssb = (nmsg->nm_etype & NM_REVENT) ? 487 &nmsg->base.nm_so->so_rcv : 488 &nmsg->base.nm_so->so_snd; 489 lwkt_gettoken(&kq_token); 490 TAILQ_REMOVE(&ssb->ssb_kq.ki_mlist, nmsg, nm_list); 491 lwkt_reltoken(&kq_token); 492 lwkt_replymsg(&nmsg->base.lmsg, EINTR); 493 } 494 495 /* 496 * Reply to the abort message 497 */ 498 lwkt_replymsg(&abrtmsg->base.lmsg, 0); 499 } 500