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