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_direct() only. 78 * Caller must make sure that the current CPU is inpcb's owner CPU. 79 * soabort_direct() 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 int 138 so_pru_attach_fast(struct socket *so, int proto, struct pru_attach_info *ai) 139 { 140 struct netmsg_pru_attach *msg; 141 int error; 142 143 error = so->so_proto->pr_usrreqs->pru_preattach(so, proto, ai); 144 if (error) 145 return error; 146 147 msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_WAITOK | M_NULLOK); 148 if (msg == NULL) { 149 /* 150 * Fail to allocate message; fallback to 151 * synchronized pru_attach. 152 */ 153 return so_pru_attach(so, proto, NULL /* postattach */); 154 } 155 156 netmsg_init(&msg->base, so, &netisr_afree_rport, 0, 157 so->so_proto->pr_usrreqs->pru_attach); 158 msg->nm_proto = proto; 159 msg->nm_ai = NULL; /* postattach */ 160 lwkt_sendmsg(so->so_port, &msg->base.lmsg); 161 162 return 0; 163 } 164 165 /* 166 * NOTE: If the target port changes the bind operation will deal with it. 167 */ 168 int 169 so_pru_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 170 { 171 struct netmsg_pru_bind msg; 172 int error; 173 174 netmsg_init(&msg.base, so, &curthread->td_msgport, 175 0, so->so_proto->pr_usrreqs->pru_bind); 176 msg.nm_nam = nam; 177 msg.nm_td = td; /* used only for prison_ip() */ 178 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 179 return (error); 180 } 181 182 int 183 so_pru_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 184 { 185 struct netmsg_pru_connect msg; 186 int error; 187 188 netmsg_init(&msg.base, so, &curthread->td_msgport, 189 0, so->so_proto->pr_usrreqs->pru_connect); 190 msg.nm_nam = nam; 191 msg.nm_td = td; 192 msg.nm_m = NULL; 193 msg.nm_sndflags = 0; 194 msg.nm_flags = 0; 195 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 196 return (error); 197 } 198 199 int 200 so_pru_connect_async(struct socket *so, struct sockaddr *nam, struct thread *td) 201 { 202 struct netmsg_pru_connect *msg; 203 int error, flags; 204 205 KASSERT(so->so_proto->pr_usrreqs->pru_preconnect != NULL, 206 ("async pru_connect is not supported")); 207 208 /* NOTE: sockaddr immediately follows netmsg */ 209 msg = kmalloc(sizeof(*msg) + nam->sa_len, M_LWKTMSG, 210 M_WAITOK | M_NULLOK); 211 if (msg == NULL) { 212 /* 213 * Fail to allocate message; fallback to 214 * synchronized pru_connect. 215 */ 216 return so_pru_connect(so, nam, td); 217 } 218 219 error = so->so_proto->pr_usrreqs->pru_preconnect(so, nam, td); 220 if (error) { 221 kfree(msg, M_LWKTMSG); 222 return error; 223 } 224 225 flags = PRUC_ASYNC; 226 if (td != NULL && (so->so_proto->pr_flags & PR_ACONN_HOLDTD)) { 227 lwkt_hold(td); 228 flags |= PRUC_HELDTD; 229 } 230 231 netmsg_init(&msg->base, so, &netisr_afree_rport, 0, 232 so->so_proto->pr_usrreqs->pru_connect); 233 msg->nm_nam = (struct sockaddr *)(msg + 1); 234 memcpy(msg->nm_nam, nam, nam->sa_len); 235 msg->nm_td = td; 236 msg->nm_m = NULL; 237 msg->nm_sndflags = 0; 238 msg->nm_flags = flags; 239 lwkt_sendmsg(so->so_port, &msg->base.lmsg); 240 return 0; 241 } 242 243 int 244 so_pru_connect2(struct socket *so1, struct socket *so2) 245 { 246 struct netmsg_pru_connect2 msg; 247 int error; 248 249 netmsg_init(&msg.base, so1, &curthread->td_msgport, 250 0, so1->so_proto->pr_usrreqs->pru_connect2); 251 msg.nm_so1 = so1; 252 msg.nm_so2 = so2; 253 error = lwkt_domsg(so1->so_port, &msg.base.lmsg, 0); 254 return (error); 255 } 256 257 /* 258 * WARNING! Synchronous call from user context. Control function may do 259 * copyin/copyout. 260 */ 261 int 262 so_pru_control_direct(struct socket *so, u_long cmd, caddr_t data, 263 struct ifnet *ifp) 264 { 265 struct netmsg_pru_control msg; 266 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_control; 267 268 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 269 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 270 msg.base.lmsg.ms_flags |= MSGF_SYNC; 271 msg.nm_cmd = cmd; 272 msg.nm_data = data; 273 msg.nm_ifp = ifp; 274 msg.nm_td = curthread; 275 func((netmsg_t)&msg); 276 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 277 return(msg.base.lmsg.ms_error); 278 } 279 280 int 281 so_pru_detach(struct socket *so) 282 { 283 struct netmsg_pru_detach msg; 284 int error; 285 286 netmsg_init(&msg.base, so, &curthread->td_msgport, 287 0, so->so_proto->pr_usrreqs->pru_detach); 288 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 289 return (error); 290 } 291 292 int 293 so_pru_detach_direct(struct socket *so) 294 { 295 struct netmsg_pru_detach msg; 296 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_detach; 297 298 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 299 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 300 msg.base.lmsg.ms_flags |= MSGF_SYNC; 301 func((netmsg_t)&msg); 302 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 303 return(msg.base.lmsg.ms_error); 304 } 305 306 int 307 so_pru_disconnect(struct socket *so) 308 { 309 struct netmsg_pru_disconnect msg; 310 int error; 311 312 netmsg_init(&msg.base, so, &curthread->td_msgport, 313 0, so->so_proto->pr_usrreqs->pru_disconnect); 314 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 315 return (error); 316 } 317 318 void 319 so_pru_disconnect_direct(struct socket *so) 320 { 321 struct netmsg_pru_disconnect msg; 322 netisr_fn_t func = so->so_proto->pr_usrreqs->pru_disconnect; 323 324 netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func); 325 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 326 msg.base.lmsg.ms_flags |= MSGF_SYNC; 327 func((netmsg_t)&msg); 328 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 329 } 330 331 int 332 so_pru_listen(struct socket *so, struct thread *td) 333 { 334 struct netmsg_pru_listen msg; 335 int error; 336 337 netmsg_init(&msg.base, so, &curthread->td_msgport, 338 0, so->so_proto->pr_usrreqs->pru_listen); 339 msg.nm_td = td; /* used only for prison_ip() XXX JH */ 340 msg.nm_flags = 0; 341 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 342 return (error); 343 } 344 345 int 346 so_pru_peeraddr(struct socket *so, struct sockaddr **nam) 347 { 348 struct netmsg_pru_peeraddr msg; 349 int error; 350 351 netmsg_init(&msg.base, so, &curthread->td_msgport, 352 0, so->so_proto->pr_usrreqs->pru_peeraddr); 353 msg.nm_nam = nam; 354 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 355 return (error); 356 } 357 358 int 359 so_pru_rcvd(struct socket *so, int flags) 360 { 361 struct netmsg_pru_rcvd msg; 362 int error; 363 364 netmsg_init(&msg.base, so, &curthread->td_msgport, 365 0, so->so_proto->pr_usrreqs->pru_rcvd); 366 msg.nm_flags = flags; 367 msg.nm_pru_flags = 0; 368 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 369 return (error); 370 } 371 372 void 373 so_pru_rcvd_async(struct socket *so) 374 { 375 lwkt_msg_t lmsg = &so->so_rcvd_msg.base.lmsg; 376 377 KASSERT(so->so_proto->pr_flags & PR_ASYNC_RCVD, 378 ("async pru_rcvd is not supported")); 379 380 /* 381 * WARNING! Spinlock is a bit dodgy, use hacked up sendmsg 382 * to avoid deadlocking. 383 */ 384 spin_lock(&so->so_rcvd_spin); 385 if ((so->so_rcvd_msg.nm_pru_flags & PRUR_DEAD) == 0) { 386 if (lmsg->ms_flags & MSGF_DONE) { 387 lwkt_sendmsg_prepare(so->so_port, lmsg); 388 spin_unlock(&so->so_rcvd_spin); 389 lwkt_sendmsg_start(so->so_port, lmsg); 390 } else { 391 spin_unlock(&so->so_rcvd_spin); 392 } 393 } else { 394 spin_unlock(&so->so_rcvd_spin); 395 } 396 } 397 398 int 399 so_pru_rcvoob(struct socket *so, struct mbuf *m, int flags) 400 { 401 struct netmsg_pru_rcvoob msg; 402 int error; 403 404 netmsg_init(&msg.base, so, &curthread->td_msgport, 405 0, so->so_proto->pr_usrreqs->pru_rcvoob); 406 msg.nm_m = m; 407 msg.nm_flags = flags; 408 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 409 return (error); 410 } 411 412 /* 413 * NOTE: If the target port changes the implied connect will deal with it. 414 */ 415 int 416 so_pru_send(struct socket *so, int flags, struct mbuf *m, 417 struct sockaddr *addr, struct mbuf *control, struct thread *td) 418 { 419 struct netmsg_pru_send msg; 420 int error; 421 422 netmsg_init(&msg.base, so, &curthread->td_msgport, 423 0, so->so_proto->pr_usrreqs->pru_send); 424 msg.nm_flags = flags; 425 msg.nm_m = m; 426 msg.nm_addr = addr; 427 msg.nm_control = control; 428 msg.nm_td = td; 429 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 430 return (error); 431 } 432 433 void 434 so_pru_sync(struct socket *so) 435 { 436 struct netmsg_base msg; 437 438 netmsg_init(&msg, so, &curthread->td_msgport, 0, 439 netmsg_sync_handler); 440 lwkt_domsg(so->so_port, &msg.lmsg, 0); 441 } 442 443 void 444 so_pru_send_async(struct socket *so, int flags, struct mbuf *m, 445 struct sockaddr *addr0, struct mbuf *control, struct thread *td) 446 { 447 struct netmsg_pru_send *msg; 448 struct sockaddr *addr = NULL; 449 450 KASSERT(so->so_proto->pr_flags & PR_ASYNC_SEND, 451 ("async pru_send is not supported")); 452 453 if (addr0 != NULL) { 454 addr = kmalloc(addr0->sa_len, M_SONAME, M_WAITOK | M_NULLOK); 455 if (addr == NULL) { 456 /* 457 * Fail to allocate address; fallback to 458 * synchronized pru_send. 459 */ 460 so_pru_send(so, flags, m, addr0, control, td); 461 return; 462 } 463 memcpy(addr, addr0, addr0->sa_len); 464 flags |= PRUS_FREEADDR; 465 } 466 flags |= PRUS_NOREPLY; 467 468 if (td != NULL && (so->so_proto->pr_flags & PR_ASEND_HOLDTD)) { 469 lwkt_hold(td); 470 flags |= PRUS_HELDTD; 471 } 472 473 msg = &m->m_hdr.mh_sndmsg; 474 netmsg_init(&msg->base, so, &netisr_apanic_rport, 475 0, so->so_proto->pr_usrreqs->pru_send); 476 msg->nm_flags = flags; 477 msg->nm_m = m; 478 msg->nm_addr = addr; 479 msg->nm_control = control; 480 msg->nm_td = td; 481 lwkt_sendmsg(so->so_port, &msg->base.lmsg); 482 } 483 484 int 485 so_pru_sense(struct socket *so, struct stat *sb) 486 { 487 struct netmsg_pru_sense msg; 488 int error; 489 490 netmsg_init(&msg.base, so, &curthread->td_msgport, 491 0, so->so_proto->pr_usrreqs->pru_sense); 492 msg.nm_stat = sb; 493 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 494 return (error); 495 } 496 497 int 498 so_pru_shutdown(struct socket *so) 499 { 500 struct netmsg_pru_shutdown msg; 501 int error; 502 503 netmsg_init(&msg.base, so, &curthread->td_msgport, 504 0, so->so_proto->pr_usrreqs->pru_shutdown); 505 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 506 return (error); 507 } 508 509 int 510 so_pru_sockaddr(struct socket *so, struct sockaddr **nam) 511 { 512 struct netmsg_pru_sockaddr msg; 513 int error; 514 515 netmsg_init(&msg.base, so, &curthread->td_msgport, 516 0, so->so_proto->pr_usrreqs->pru_sockaddr); 517 msg.nm_nam = nam; 518 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 519 return (error); 520 } 521 522 int 523 so_pr_ctloutput(struct socket *so, struct sockopt *sopt) 524 { 525 struct netmsg_pr_ctloutput msg; 526 int error; 527 528 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 529 netmsg_init(&msg.base, so, &curthread->td_msgport, 530 0, so->so_proto->pr_ctloutput); 531 msg.nm_sopt = sopt; 532 error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0); 533 return (error); 534 } 535 536 struct lwkt_port * 537 so_pr_ctlport(struct protosw *pr, int cmd, struct sockaddr *arg, 538 void *extra, int *cpuid) 539 { 540 if (pr->pr_ctlport == NULL) 541 return NULL; 542 KKASSERT(pr->pr_ctlinput != NULL); 543 544 return pr->pr_ctlport(cmd, arg, extra, cpuid); 545 } 546 547 /* 548 * Protocol control input, typically via icmp. 549 * 550 * If the protocol pr_ctlport is not NULL we call it to figure out the 551 * protocol port. If NULL is returned we can just return, otherwise 552 * we issue a netmsg to call pr_ctlinput in the proper thread. 553 * 554 * This must be done synchronously as arg and/or extra may point to 555 * temporary data. 556 */ 557 void 558 so_pr_ctlinput(struct protosw *pr, int cmd, struct sockaddr *arg, void *extra) 559 { 560 struct netmsg_pr_ctlinput msg; 561 lwkt_port_t port; 562 int cpuid; 563 564 port = so_pr_ctlport(pr, cmd, arg, extra, &cpuid); 565 if (port == NULL) 566 return; 567 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 568 0, pr->pr_ctlinput); 569 msg.nm_cmd = cmd; 570 msg.nm_direct = 0; 571 msg.nm_arg = arg; 572 msg.nm_extra = extra; 573 lwkt_domsg(port, &msg.base.lmsg, 0); 574 } 575 576 void 577 so_pr_ctlinput_direct(struct protosw *pr, int cmd, struct sockaddr *arg, 578 void *extra) 579 { 580 struct netmsg_pr_ctlinput msg; 581 netisr_fn_t func; 582 lwkt_port_t port; 583 int cpuid; 584 585 port = so_pr_ctlport(pr, cmd, arg, extra, &cpuid); 586 if (port == NULL) 587 return; 588 if (cpuid != ncpus && cpuid != mycpuid) 589 return; 590 591 func = pr->pr_ctlinput; 592 netmsg_init(&msg.base, NULL, &netisr_adone_rport, 0, func); 593 msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE); 594 msg.base.lmsg.ms_flags |= MSGF_SYNC; 595 msg.nm_cmd = cmd; 596 msg.nm_direct = 1; 597 msg.nm_arg = arg; 598 msg.nm_extra = extra; 599 func((netmsg_t)&msg); 600 KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE); 601 } 602 603 /* 604 * If we convert all the protosw pr_ functions for all the protocols 605 * to take a message directly, this layer can go away. For the moment 606 * our dispatcher ignores the return value, but since we are handling 607 * the replymsg ourselves we return EASYNC by convention. 608 */ 609 610 /* 611 * Handle a predicate event request. This function is only called once 612 * when the predicate message queueing request is received. 613 */ 614 void 615 netmsg_so_notify(netmsg_t msg) 616 { 617 struct lwkt_token *tok; 618 struct signalsockbuf *ssb; 619 620 ssb = (msg->notify.nm_etype & NM_REVENT) ? 621 &msg->base.nm_so->so_rcv : 622 &msg->base.nm_so->so_snd; 623 624 /* 625 * Reply immediately if the event has occured, otherwise queue the 626 * request. 627 * 628 * NOTE: Socket can change if this is an accept predicate so cache 629 * the token. 630 */ 631 tok = lwkt_token_pool_lookup(msg->base.nm_so); 632 lwkt_gettoken(tok); 633 atomic_set_int(&ssb->ssb_flags, SSB_MEVENT); 634 if (msg->notify.nm_predicate(&msg->notify)) { 635 if (TAILQ_EMPTY(&ssb->ssb_kq.ki_mlist)) 636 atomic_clear_int(&ssb->ssb_flags, SSB_MEVENT); 637 lwkt_reltoken(tok); 638 lwkt_replymsg(&msg->base.lmsg, 639 msg->base.lmsg.ms_error); 640 } else { 641 TAILQ_INSERT_TAIL(&ssb->ssb_kq.ki_mlist, &msg->notify, nm_list); 642 /* 643 * NOTE: 644 * If predict ever blocks, 'tok' will be released, so 645 * SSB_MEVENT set beforehand could have been cleared 646 * when we reach here. In case that happens, we set 647 * SSB_MEVENT again, after the notify has been queued. 648 */ 649 atomic_set_int(&ssb->ssb_flags, SSB_MEVENT); 650 lwkt_reltoken(tok); 651 } 652 } 653 654 /* 655 * Called by doio when trying to abort a netmsg_so_notify message. 656 * Unlike the other functions this one is dispatched directly by 657 * the LWKT subsystem, so it takes a lwkt_msg_t as an argument. 658 * 659 * The original message, lmsg, is under the control of the caller and 660 * will not be destroyed until we return so we can safely reference it 661 * in our synchronous abort request. 662 * 663 * This part of the abort request occurs on the originating cpu which 664 * means we may race the message flags and the original message may 665 * not even have been processed by the target cpu yet. 666 */ 667 void 668 netmsg_so_notify_doabort(lwkt_msg_t lmsg) 669 { 670 struct netmsg_so_notify_abort msg; 671 672 if ((lmsg->ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) { 673 const struct netmsg_base *nmsg = 674 (const struct netmsg_base *)lmsg; 675 676 netmsg_init(&msg.base, nmsg->nm_so, &curthread->td_msgport, 677 0, netmsg_so_notify_abort); 678 msg.nm_notifymsg = (void *)lmsg; 679 lwkt_domsg(lmsg->ms_target_port, &msg.base.lmsg, 0); 680 } 681 } 682 683 /* 684 * Predicate requests can be aborted. This function is only called once 685 * and will interlock against processing/reply races (since such races 686 * occur on the same thread that controls the port where the abort is 687 * requeued). 688 * 689 * This part of the abort request occurs on the target cpu. The message 690 * flags must be tested again in case the test that we did on the 691 * originating cpu raced. Since messages are handled in sequence, the 692 * original message will have already been handled by the loop and either 693 * replied to or queued. 694 * 695 * We really only need to interlock with MSGF_REPLY (a bit that is set on 696 * our cpu when we reply). Note that MSGF_DONE is not set until the 697 * reply reaches the originating cpu. Test both bits anyway. 698 */ 699 void 700 netmsg_so_notify_abort(netmsg_t msg) 701 { 702 struct netmsg_so_notify_abort *abrtmsg = &msg->notify_abort; 703 struct netmsg_so_notify *nmsg = abrtmsg->nm_notifymsg; 704 struct signalsockbuf *ssb; 705 706 /* 707 * The original notify message is not destroyed until after the 708 * abort request is returned, so we can check its state. 709 */ 710 lwkt_getpooltoken(nmsg->base.nm_so); 711 if ((nmsg->base.lmsg.ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) { 712 ssb = (nmsg->nm_etype & NM_REVENT) ? 713 &nmsg->base.nm_so->so_rcv : 714 &nmsg->base.nm_so->so_snd; 715 TAILQ_REMOVE(&ssb->ssb_kq.ki_mlist, nmsg, nm_list); 716 lwkt_relpooltoken(nmsg->base.nm_so); 717 lwkt_replymsg(&nmsg->base.lmsg, EINTR); 718 } else { 719 lwkt_relpooltoken(nmsg->base.nm_so); 720 } 721 722 /* 723 * Reply to the abort message 724 */ 725 lwkt_replymsg(&abrtmsg->base.lmsg, 0); 726 } 727 728 void 729 so_async_rcvd_reply(struct socket *so) 730 { 731 /* 732 * Spinlock safe, reply runs to degenerate lwkt_null_replyport() 733 */ 734 spin_lock(&so->so_rcvd_spin); 735 lwkt_replymsg(&so->so_rcvd_msg.base.lmsg, 0); 736 spin_unlock(&so->so_rcvd_spin); 737 } 738 739 void 740 so_async_rcvd_drop(struct socket *so) 741 { 742 lwkt_msg_t lmsg = &so->so_rcvd_msg.base.lmsg; 743 744 /* 745 * Spinlock safe, drop runs to degenerate lwkt_spin_dropmsg() 746 */ 747 spin_lock(&so->so_rcvd_spin); 748 so->so_rcvd_msg.nm_pru_flags |= PRUR_DEAD; 749 again: 750 lwkt_dropmsg(lmsg); 751 if ((lmsg->ms_flags & MSGF_DONE) == 0) { 752 ++async_rcvd_drop_race; 753 ssleep(so, &so->so_rcvd_spin, 0, "soadrop", 1); 754 goto again; 755 } 756 spin_unlock(&so->so_rcvd_spin); 757 } 758