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/sysctl.h> 47 #include <sys/mbuf.h> 48 #include <vm/pmap.h> 49 50 #include <net/netmsg2.h> 51 #include <sys/socketvar2.h> 52 53 #include <net/netisr.h> 54 #include <net/netmsg.h> 55 56 static int async_rcvd_drop_race = 0; 57 SYSCTL_INT(_kern_ipc, OID_AUTO, async_rcvd_drop_race, CTLFLAG_RW, 58 &async_rcvd_drop_race, 0, "# of asynchronized pru_rcvd msg drop races"); 59 60 /* 61 * Abort a socket and free it, asynchronously. Called from 62 * soabort_async() only. soabort_async() got a ref on the 63 * socket which we must free on reply. 64 */ 65 void 66 so_pru_abort_async(struct socket *so) 67 { 68 struct netmsg_pru_abort *msg; 69 70 msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_WAITOK | M_ZERO); 71 netmsg_init(&msg->base, so, &netisr_afree_free_so_rport, 72 0, so->so_proto->pr_usrreqs->pru_abort); 73 lwkt_sendmsg(so->so_port, &msg->base.lmsg); 74 } 75 76 /* 77 * Abort a socket and free it. Called from soabort_oncpu() only. 78 * Caller must make sure that the current CPU is inpcb's owner CPU. 79 * soabort_oncpu() got a ref on the socket which we must free. 80 */ 81 void 82 so_pru_abort_direct(struct socket *so) 83 { 84 struct netmsg_pru_abort msg; 85 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_abort; 86 87 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 88 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 89 msg.base.lmsg.ms_flags |= MSGF_SYNC; 90 func((netmsg_t)&msg); 91 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 92 sofree(msg.base.nm_so); 93 } 94 95 int 96 so_pru_accept(struct socket *so, struct sockaddr **nam) 97 { 98 struct netmsg_pru_accept msg; 99 100 netmsg_init(&msg.base, so, &curthread->td_msgport, 101 0, so->so_proto->pr_usrreqs->pru_accept); 102 msg.nm_nam = nam; 103 104 return lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 105 } 106 107 int 108 so_pru_attach(struct socket *so, int proto, struct pru_attach_info *ai) 109 { 110 struct netmsg_pru_attach msg; 111 int error; 112 113 netmsg_init(&msg.base, so, &curthread->td_msgport, 114 0, so->so_proto->pr_usrreqs->pru_attach); 115 msg.nm_proto = proto; 116 msg.nm_ai = ai; 117 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 118 return (error); 119 } 120 121 int 122 so_pru_attach_direct(struct socket *so, int proto, struct pru_attach_info *ai) 123 { 124 struct netmsg_pru_attach msg; 125 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_attach; 126 127 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 128 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 129 msg.base.lmsg.ms_flags |= MSGF_SYNC; 130 msg.nm_proto = proto; 131 msg.nm_ai = ai; 132 func((netmsg_t)&msg); 133 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 134 return(msg.base.lmsg.ms_error); 135 } 136 137 /* 138 * NOTE: If the target port changes the bind operation will deal with it. 139 */ 140 int 141 so_pru_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 142 { 143 struct netmsg_pru_bind msg; 144 int error; 145 146 netmsg_init(&msg.base, so, &curthread->td_msgport, 147 0, so->so_proto->pr_usrreqs->pru_bind); 148 msg.nm_nam = nam; 149 msg.nm_td = td; /* used only for prison_ip() */ 150 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 151 return (error); 152 } 153 154 int 155 so_pru_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 156 { 157 struct netmsg_pru_connect msg; 158 int error; 159 160 netmsg_init(&msg.base, so, &curthread->td_msgport, 161 0, so->so_proto->pr_usrreqs->pru_connect); 162 msg.nm_nam = nam; 163 msg.nm_td = td; 164 msg.nm_m = NULL; 165 msg.nm_sndflags = 0; 166 msg.nm_flags = 0; 167 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 168 return (error); 169 } 170 171 int 172 so_pru_connect_async(struct socket *so, struct sockaddr *nam, struct thread *td) 173 { 174 struct netmsg_pru_connect *msg; 175 int error, flags; 176 177 KASSERT(so->so_proto->pr_usrreqs->pru_preconnect != NULL, 178 ("async pru_connect is not supported")); 179 180 /* NOTE: sockaddr immediately follows netmsg */ 181 msg = kmalloc(sizeof(*msg) + nam->sa_len, M_LWKTMSG, M_NOWAIT); 182 if (msg == NULL) { 183 /* 184 * Fail to allocate message w/o waiting; 185 * fallback to synchronized pru_connect. 186 */ 187 return so_pru_connect(so, nam, td); 188 } 189 190 error = so->so_proto->pr_usrreqs->pru_preconnect(so, nam, td); 191 if (error) { 192 kfree(msg, M_LWKTMSG); 193 return error; 194 } 195 196 flags = PRUC_ASYNC; 197 if (td != NULL && (so->so_proto->pr_flags & PR_ACONN_HOLDTD)) { 198 lwkt_hold(td); 199 flags |= PRUC_HELDTD; 200 } 201 202 netmsg_init(&msg->base, so, &netisr_afree_rport, 0, 203 so->so_proto->pr_usrreqs->pru_connect); 204 msg->nm_nam = (struct sockaddr *)(msg + 1); 205 memcpy(msg->nm_nam, nam, nam->sa_len); 206 msg->nm_td = td; 207 msg->nm_m = NULL; 208 msg->nm_sndflags = 0; 209 msg->nm_flags = flags; 210 lwkt_sendmsg(so->so_port, &msg->base.lmsg); 211 return 0; 212 } 213 214 int 215 so_pru_connect2(struct socket *so1, struct socket *so2) 216 { 217 struct netmsg_pru_connect2 msg; 218 int error; 219 220 netmsg_init(&msg.base, so1, &curthread->td_msgport, 221 0, so1->so_proto->pr_usrreqs->pru_connect2); 222 msg.nm_so1 = so1; 223 msg.nm_so2 = so2; 224 error = lwkt_domsg(so1->so_port, &msg.base.lmsg, 0); 225 return (error); 226 } 227 228 /* 229 * WARNING! Synchronous call from user context. Control function may do 230 * copyin/copyout. 231 */ 232 int 233 so_pru_control_direct(struct socket *so, u_long cmd, caddr_t data, 234 struct ifnet *ifp) 235 { 236 struct netmsg_pru_control msg; 237 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_control; 238 239 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 240 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 241 msg.base.lmsg.ms_flags |= MSGF_SYNC; 242 msg.nm_cmd = cmd; 243 msg.nm_data = data; 244 msg.nm_ifp = ifp; 245 msg.nm_td = curthread; 246 func((netmsg_t)&msg); 247 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 248 return(msg.base.lmsg.ms_error); 249 } 250 251 int 252 so_pru_detach(struct socket *so) 253 { 254 struct netmsg_pru_detach msg; 255 int error; 256 257 netmsg_init(&msg.base, so, &curthread->td_msgport, 258 0, so->so_proto->pr_usrreqs->pru_detach); 259 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 260 return (error); 261 } 262 263 int 264 so_pru_detach_direct(struct socket *so) 265 { 266 struct netmsg_pru_detach msg; 267 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_detach; 268 269 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 270 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 271 msg.base.lmsg.ms_flags |= MSGF_SYNC; 272 func((netmsg_t)&msg); 273 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 274 return(msg.base.lmsg.ms_error); 275 } 276 277 int 278 so_pru_disconnect(struct socket *so) 279 { 280 struct netmsg_pru_disconnect msg; 281 int error; 282 283 netmsg_init(&msg.base, so, &curthread->td_msgport, 284 0, so->so_proto->pr_usrreqs->pru_disconnect); 285 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 286 return (error); 287 } 288 289 void 290 so_pru_disconnect_direct(struct socket *so) 291 { 292 struct netmsg_pru_disconnect msg; 293 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_disconnect; 294 295 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 296 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 297 msg.base.lmsg.ms_flags |= MSGF_SYNC; 298 func((netmsg_t)&msg); 299 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 300 } 301 302 int 303 so_pru_listen(struct socket *so, struct thread *td) 304 { 305 struct netmsg_pru_listen msg; 306 int error; 307 308 netmsg_init(&msg.base, so, &curthread->td_msgport, 309 0, so->so_proto->pr_usrreqs->pru_listen); 310 msg.nm_td = td; /* used only for prison_ip() XXX JH */ 311 msg.nm_flags = 0; 312 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 313 return (error); 314 } 315 316 int 317 so_pru_peeraddr(struct socket *so, struct sockaddr **nam) 318 { 319 struct netmsg_pru_peeraddr msg; 320 int error; 321 322 netmsg_init(&msg.base, so, &curthread->td_msgport, 323 0, so->so_proto->pr_usrreqs->pru_peeraddr); 324 msg.nm_nam = nam; 325 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 326 return (error); 327 } 328 329 int 330 so_pru_rcvd(struct socket *so, int flags) 331 { 332 struct netmsg_pru_rcvd msg; 333 int error; 334 335 netmsg_init(&msg.base, so, &curthread->td_msgport, 336 0, so->so_proto->pr_usrreqs->pru_rcvd); 337 msg.nm_flags = flags; 338 msg.nm_pru_flags = 0; 339 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 340 return (error); 341 } 342 343 void 344 so_pru_rcvd_async(struct socket *so) 345 { 346 lwkt_msg_t lmsg = &so->so_rcvd_msg.base.lmsg; 347 348 KASSERT(so->so_proto->pr_flags & PR_ASYNC_RCVD, 349 ("async pru_rcvd is not supported")); 350 351 /* 352 * WARNING! Spinlock is a bit dodgy, use hacked up sendmsg 353 * to avoid deadlocking. 354 */ 355 spin_lock(&so->so_rcvd_spin); 356 if ((so->so_rcvd_msg.nm_pru_flags & PRUR_DEAD) == 0) { 357 if (lmsg->ms_flags & MSGF_DONE) { 358 lwkt_sendmsg_prepare(so->so_port, lmsg); 359 spin_unlock(&so->so_rcvd_spin); 360 lwkt_sendmsg_start(so->so_port, lmsg); 361 } else { 362 spin_unlock(&so->so_rcvd_spin); 363 } 364 } else { 365 spin_unlock(&so->so_rcvd_spin); 366 } 367 } 368 369 int 370 so_pru_rcvoob(struct socket *so, struct mbuf *m, int flags) 371 { 372 struct netmsg_pru_rcvoob msg; 373 int error; 374 375 netmsg_init(&msg.base, so, &curthread->td_msgport, 376 0, so->so_proto->pr_usrreqs->pru_rcvoob); 377 msg.nm_m = m; 378 msg.nm_flags = flags; 379 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 380 return (error); 381 } 382 383 /* 384 * NOTE: If the target port changes the implied connect will deal with it. 385 */ 386 int 387 so_pru_send(struct socket *so, int flags, struct mbuf *m, 388 struct sockaddr *addr, struct mbuf *control, struct thread *td) 389 { 390 struct netmsg_pru_send msg; 391 int error; 392 393 netmsg_init(&msg.base, so, &curthread->td_msgport, 394 0, so->so_proto->pr_usrreqs->pru_send); 395 msg.nm_flags = flags; 396 msg.nm_m = m; 397 msg.nm_addr = addr; 398 msg.nm_control = control; 399 msg.nm_td = td; 400 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 401 return (error); 402 } 403 404 void 405 so_pru_sync(struct socket *so) 406 { 407 struct netmsg_base msg; 408 409 netmsg_init(&msg, so, &curthread->td_msgport, 0, 410 netmsg_sync_handler); 411 lwkt_domsg(so->so_port, &msg.lmsg, 0); 412 } 413 414 void 415 so_pru_send_async(struct socket *so, int flags, struct mbuf *m, 416 struct sockaddr *addr0, struct mbuf *control, struct thread *td) 417 { 418 struct netmsg_pru_send *msg; 419 struct sockaddr *addr = NULL; 420 421 KASSERT(so->so_proto->pr_flags & PR_ASYNC_SEND, 422 ("async pru_send is not supported")); 423 424 if (addr0 != NULL) { 425 addr = kmalloc(addr0->sa_len, M_SONAME, M_WAITOK | M_NULLOK); 426 if (addr == NULL) { 427 /* 428 * Fail to allocate address; fallback to 429 * synchronized pru_send. 430 */ 431 so_pru_send(so, flags, m, addr0, control, td); 432 return; 433 } 434 memcpy(addr, addr0, addr0->sa_len); 435 flags |= PRUS_FREEADDR; 436 } 437 flags |= PRUS_NOREPLY; 438 439 if (td != NULL && (so->so_proto->pr_flags & PR_ASEND_HOLDTD)) { 440 lwkt_hold(td); 441 flags |= PRUS_HELDTD; 442 } 443 444 msg = &m->m_hdr.mh_sndmsg; 445 netmsg_init(&msg->base, so, &netisr_apanic_rport, 446 0, so->so_proto->pr_usrreqs->pru_send); 447 msg->nm_flags = flags; 448 msg->nm_m = m; 449 msg->nm_addr = addr; 450 msg->nm_control = control; 451 msg->nm_td = td; 452 lwkt_sendmsg(so->so_port, &msg->base.lmsg); 453 } 454 455 int 456 so_pru_sense(struct socket *so, struct stat *sb) 457 { 458 struct netmsg_pru_sense msg; 459 int error; 460 461 netmsg_init(&msg.base, so, &curthread->td_msgport, 462 0, so->so_proto->pr_usrreqs->pru_sense); 463 msg.nm_stat = sb; 464 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 465 return (error); 466 } 467 468 int 469 so_pru_shutdown(struct socket *so) 470 { 471 struct netmsg_pru_shutdown msg; 472 int error; 473 474 netmsg_init(&msg.base, so, &curthread->td_msgport, 475 0, so->so_proto->pr_usrreqs->pru_shutdown); 476 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 477 return (error); 478 } 479 480 int 481 so_pru_sockaddr(struct socket *so, struct sockaddr **nam) 482 { 483 struct netmsg_pru_sockaddr msg; 484 int error; 485 486 netmsg_init(&msg.base, so, &curthread->td_msgport, 487 0, so->so_proto->pr_usrreqs->pru_sockaddr); 488 msg.nm_nam = nam; 489 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 490 return (error); 491 } 492 493 int 494 so_pr_ctloutput(struct socket *so, struct sockopt *sopt) 495 { 496 struct netmsg_pr_ctloutput msg; 497 int error; 498 499 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 500 netmsg_init(&msg.base, so, &curthread->td_msgport, 501 0, so->so_proto->pr_ctloutput); 502 msg.nm_sopt = sopt; 503 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 504 return (error); 505 } 506 507 struct lwkt_port * 508 so_pr_ctlport(struct protosw *pr, int cmd, struct sockaddr *arg, 509 void *extra, int *cpuid) 510 { 511 if (pr->pr_ctlport == NULL) 512 return NULL; 513 KKASSERT(pr->pr_ctlinput != NULL); 514 515 return pr->pr_ctlport(cmd, arg, extra, cpuid); 516 } 517 518 /* 519 * Protocol control input, typically via icmp. 520 * 521 * If the protocol pr_ctlport is not NULL we call it to figure out the 522 * protocol port. If NULL is returned we can just return, otherwise 523 * we issue a netmsg to call pr_ctlinput in the proper thread. 524 * 525 * This must be done synchronously as arg and/or extra may point to 526 * temporary data. 527 */ 528 void 529 so_pr_ctlinput(struct protosw *pr, int cmd, struct sockaddr *arg, void *extra) 530 { 531 struct netmsg_pr_ctlinput msg; 532 lwkt_port_t port; 533 int cpuid; 534 535 port = so_pr_ctlport(pr, cmd, arg, extra, &cpuid); 536 if (port == NULL) 537 return; 538 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 539 0, pr->pr_ctlinput); 540 msg.nm_cmd = cmd; 541 msg.nm_direct = 0; 542 msg.nm_arg = arg; 543 msg.nm_extra = extra; 544 lwkt_domsg(port, &msg.base.lmsg, 0); 545 } 546 547 void 548 so_pr_ctlinput_direct(struct protosw *pr, int cmd, struct sockaddr *arg, 549 void *extra) 550 { 551 struct netmsg_pr_ctlinput msg; 552 netisr_fn_t func; 553 lwkt_port_t port; 554 int cpuid; 555 556 port = so_pr_ctlport(pr, cmd, arg, extra, &cpuid); 557 if (port == NULL) 558 return; 559 if (cpuid != ncpus && cpuid != mycpuid) 560 return; 561 562 func = pr->pr_ctlinput; 563 netmsg_init(&msg.base, NULL, &netisr_adone_rport, 0, func); 564 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 565 msg.base.lmsg.ms_flags |= MSGF_SYNC; 566 msg.nm_cmd = cmd; 567 msg.nm_direct = 1; 568 msg.nm_arg = arg; 569 msg.nm_extra = extra; 570 func((netmsg_t)&msg); 571 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 572 } 573 574 /* 575 * If we convert all the protosw pr_ functions for all the protocols 576 * to take a message directly, this layer can go away. For the moment 577 * our dispatcher ignores the return value, but since we are handling 578 * the replymsg ourselves we return EASYNC by convention. 579 */ 580 581 /* 582 * Handle a predicate event request. This function is only called once 583 * when the predicate message queueing request is received. 584 */ 585 void 586 netmsg_so_notify(netmsg_t msg) 587 { 588 struct lwkt_token *tok; 589 struct signalsockbuf *ssb; 590 591 ssb = (msg->notify.nm_etype & NM_REVENT) ? 592 &msg->base.nm_so->so_rcv : 593 &msg->base.nm_so->so_snd; 594 595 /* 596 * Reply immediately if the event has occured, otherwise queue the 597 * request. 598 * 599 * NOTE: Socket can change if this is an accept predicate so cache 600 * the token. 601 */ 602 tok = lwkt_token_pool_lookup(msg->base.nm_so); 603 lwkt_gettoken(tok); 604 atomic_set_int(&ssb->ssb_flags, SSB_MEVENT); 605 if (msg->notify.nm_predicate(&msg->notify)) { 606 if (TAILQ_EMPTY(&ssb->ssb_kq.ki_mlist)) 607 atomic_clear_int(&ssb->ssb_flags, SSB_MEVENT); 608 lwkt_reltoken(tok); 609 lwkt_replymsg(&msg->base.lmsg, 610 msg->base.lmsg.ms_error); 611 } else { 612 TAILQ_INSERT_TAIL(&ssb->ssb_kq.ki_mlist, &msg->notify, nm_list); 613 /* 614 * NOTE: 615 * If predict ever blocks, 'tok' will be released, so 616 * SSB_MEVENT set beforehand could have been cleared 617 * when we reach here. In case that happens, we set 618 * SSB_MEVENT again, after the notify has been queued. 619 */ 620 atomic_set_int(&ssb->ssb_flags, SSB_MEVENT); 621 lwkt_reltoken(tok); 622 } 623 } 624 625 /* 626 * Called by doio when trying to abort a netmsg_so_notify message. 627 * Unlike the other functions this one is dispatched directly by 628 * the LWKT subsystem, so it takes a lwkt_msg_t as an argument. 629 * 630 * The original message, lmsg, is under the control of the caller and 631 * will not be destroyed until we return so we can safely reference it 632 * in our synchronous abort request. 633 * 634 * This part of the abort request occurs on the originating cpu which 635 * means we may race the message flags and the original message may 636 * not even have been processed by the target cpu yet. 637 */ 638 void 639 netmsg_so_notify_doabort(lwkt_msg_t lmsg) 640 { 641 struct netmsg_so_notify_abort msg; 642 643 if ((lmsg->ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) { 644 const struct netmsg_base *nmsg = 645 (const struct netmsg_base *)lmsg; 646 647 netmsg_init(&msg.base, nmsg->nm_so, &curthread->td_msgport, 648 0, netmsg_so_notify_abort); 649 msg.nm_notifymsg = (void *)lmsg; 650 lwkt_domsg(lmsg->ms_target_port, &msg.base.lmsg, 0); 651 } 652 } 653 654 /* 655 * Predicate requests can be aborted. This function is only called once 656 * and will interlock against processing/reply races (since such races 657 * occur on the same thread that controls the port where the abort is 658 * requeued). 659 * 660 * This part of the abort request occurs on the target cpu. The message 661 * flags must be tested again in case the test that we did on the 662 * originating cpu raced. Since messages are handled in sequence, the 663 * original message will have already been handled by the loop and either 664 * replied to or queued. 665 * 666 * We really only need to interlock with MSGF_REPLY (a bit that is set on 667 * our cpu when we reply). Note that MSGF_DONE is not set until the 668 * reply reaches the originating cpu. Test both bits anyway. 669 */ 670 void 671 netmsg_so_notify_abort(netmsg_t msg) 672 { 673 struct netmsg_so_notify_abort *abrtmsg = &msg->notify_abort; 674 struct netmsg_so_notify *nmsg = abrtmsg->nm_notifymsg; 675 struct signalsockbuf *ssb; 676 677 /* 678 * The original notify message is not destroyed until after the 679 * abort request is returned, so we can check its state. 680 */ 681 lwkt_getpooltoken(nmsg->base.nm_so); 682 if ((nmsg->base.lmsg.ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) { 683 ssb = (nmsg->nm_etype & NM_REVENT) ? 684 &nmsg->base.nm_so->so_rcv : 685 &nmsg->base.nm_so->so_snd; 686 TAILQ_REMOVE(&ssb->ssb_kq.ki_mlist, nmsg, nm_list); 687 lwkt_relpooltoken(nmsg->base.nm_so); 688 lwkt_replymsg(&nmsg->base.lmsg, EINTR); 689 } else { 690 lwkt_relpooltoken(nmsg->base.nm_so); 691 } 692 693 /* 694 * Reply to the abort message 695 */ 696 lwkt_replymsg(&abrtmsg->base.lmsg, 0); 697 } 698 699 void 700 so_async_rcvd_reply(struct socket *so) 701 { 702 /* 703 * Spinlock safe, reply runs to degenerate lwkt_null_replyport() 704 */ 705 spin_lock(&so->so_rcvd_spin); 706 lwkt_replymsg(&so->so_rcvd_msg.base.lmsg, 0); 707 spin_unlock(&so->so_rcvd_spin); 708 } 709 710 void 711 so_async_rcvd_drop(struct socket *so) 712 { 713 lwkt_msg_t lmsg = &so->so_rcvd_msg.base.lmsg; 714 715 /* 716 * Spinlock safe, drop runs to degenerate lwkt_spin_dropmsg() 717 */ 718 spin_lock(&so->so_rcvd_spin); 719 so->so_rcvd_msg.nm_pru_flags |= PRUR_DEAD; 720 again: 721 lwkt_dropmsg(lmsg); 722 if ((lmsg->ms_flags & MSGF_DONE) == 0) { 723 ++async_rcvd_drop_race; 724 ssleep(so, &so->so_rcvd_spin, 0, "soadrop", 1); 725 goto again; 726 } 727 spin_unlock(&so->so_rcvd_spin); 728 } 729