1 /* 2 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 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 /* 35 * Copyright (c) 1982, 1986, 1988, 1990, 1993 36 * The Regents of the University of California. All rights reserved. 37 * 38 * Redistribution and use in source and binary forms, with or without 39 * modification, are permitted provided that the following conditions 40 * are met: 41 * 1. Redistributions of source code must retain the above copyright 42 * notice, this list of conditions and the following disclaimer. 43 * 2. Redistributions in binary form must reproduce the above copyright 44 * notice, this list of conditions and the following disclaimer in the 45 * documentation and/or other materials provided with the distribution. 46 * 3. Neither the name of the University nor the names of its contributors 47 * may be used to endorse or promote products derived from this software 48 * without specific prior written permission. 49 * 50 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 53 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 60 * SUCH DAMAGE. 61 * 62 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 63 * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.24 2003/11/11 17:18:18 silby Exp $ 64 */ 65 66 #include "opt_inet.h" 67 #include "opt_sctp.h" 68 69 #include <sys/param.h> 70 #include <sys/systm.h> 71 #include <sys/fcntl.h> 72 #include <sys/malloc.h> 73 #include <sys/mbuf.h> 74 #include <sys/domain.h> 75 #include <sys/file.h> /* for struct knote */ 76 #include <sys/kernel.h> 77 #include <sys/event.h> 78 #include <sys/proc.h> 79 #include <sys/protosw.h> 80 #include <sys/socket.h> 81 #include <sys/socketvar.h> 82 #include <sys/socketops.h> 83 #include <sys/resourcevar.h> 84 #include <sys/signalvar.h> 85 #include <sys/sysctl.h> 86 #include <sys/uio.h> 87 #include <sys/jail.h> 88 #include <vm/vm_zone.h> 89 #include <vm/pmap.h> 90 #include <net/netmsg2.h> 91 #include <net/netisr2.h> 92 93 #include <sys/thread2.h> 94 #include <sys/socketvar2.h> 95 #include <sys/spinlock2.h> 96 97 #include <machine/limits.h> 98 99 #ifdef INET 100 extern int tcp_sosend_agglim; 101 extern int tcp_sosend_async; 102 extern int tcp_sosend_jcluster; 103 extern int udp_sosend_async; 104 extern int udp_sosend_prepend; 105 106 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt); 107 #endif /* INET */ 108 109 static void filt_sordetach(struct knote *kn); 110 static int filt_soread(struct knote *kn, long hint); 111 static void filt_sowdetach(struct knote *kn); 112 static int filt_sowrite(struct knote *kn, long hint); 113 static int filt_solisten(struct knote *kn, long hint); 114 115 static int soclose_sync(struct socket *so, int fflag); 116 static void soclose_fast(struct socket *so); 117 118 static struct filterops solisten_filtops = 119 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten }; 120 static struct filterops soread_filtops = 121 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread }; 122 static struct filterops sowrite_filtops = 123 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite }; 124 static struct filterops soexcept_filtops = 125 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread }; 126 127 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct"); 128 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 129 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 130 131 132 static int somaxconn = SOMAXCONN; 133 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW, 134 &somaxconn, 0, "Maximum pending socket connection queue size"); 135 136 static int use_soclose_fast = 1; 137 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW, 138 &use_soclose_fast, 0, "Fast socket close"); 139 140 int use_soaccept_pred_fast = 1; 141 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW, 142 &use_soaccept_pred_fast, 0, "Fast socket accept predication"); 143 144 int use_sendfile_async = 1; 145 SYSCTL_INT(_kern_ipc, OID_AUTO, sendfile_async, CTLFLAG_RW, 146 &use_sendfile_async, 0, "sendfile uses asynchronized pru_send"); 147 148 int use_soconnect_async = 1; 149 SYSCTL_INT(_kern_ipc, OID_AUTO, soconnect_async, CTLFLAG_RW, 150 &use_soconnect_async, 0, "soconnect uses asynchronized pru_connect"); 151 152 /* 153 * Socket operation routines. 154 * These routines are called by the routines in 155 * sys_socket.c or from a system process, and 156 * implement the semantics of socket operations by 157 * switching out to the protocol specific routines. 158 */ 159 160 /* 161 * Get a socket structure, and initialize it. 162 * Note that it would probably be better to allocate socket 163 * and PCB at the same time, but I'm not convinced that all 164 * the protocols can be easily modified to do this. 165 */ 166 struct socket * 167 soalloc(int waitok, struct protosw *pr) 168 { 169 struct socket *so; 170 unsigned waitmask; 171 172 waitmask = waitok ? M_WAITOK : M_NOWAIT; 173 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask); 174 if (so) { 175 /* XXX race condition for reentrant kernel */ 176 so->so_proto = pr; 177 TAILQ_INIT(&so->so_aiojobq); 178 TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist); 179 TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist); 180 lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok"); 181 lwkt_token_init(&so->so_snd.ssb_token, "sndtok"); 182 spin_init(&so->so_rcvd_spin, "soalloc"); 183 netmsg_init(&so->so_rcvd_msg.base, so, &netisr_adone_rport, 184 MSGF_DROPABLE | MSGF_PRIORITY, 185 so->so_proto->pr_usrreqs->pru_rcvd); 186 so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC; 187 so->so_state = SS_NOFDREF; 188 so->so_refs = 1; 189 } 190 return so; 191 } 192 193 int 194 socreate(int dom, struct socket **aso, int type, 195 int proto, struct thread *td) 196 { 197 struct proc *p = td->td_proc; 198 struct protosw *prp; 199 struct socket *so; 200 struct pru_attach_info ai; 201 int error; 202 203 if (proto) 204 prp = pffindproto(dom, proto, type); 205 else 206 prp = pffindtype(dom, type); 207 208 if (prp == NULL || prp->pr_usrreqs->pru_attach == 0) 209 return (EPROTONOSUPPORT); 210 211 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only && 212 prp->pr_domain->dom_family != PF_LOCAL && 213 prp->pr_domain->dom_family != PF_INET && 214 prp->pr_domain->dom_family != PF_INET6 && 215 prp->pr_domain->dom_family != PF_ROUTE) { 216 return (EPROTONOSUPPORT); 217 } 218 219 if (prp->pr_type != type) 220 return (EPROTOTYPE); 221 so = soalloc(p != NULL, prp); 222 if (so == NULL) 223 return (ENOBUFS); 224 225 /* 226 * Callers of socreate() presumably will connect up a descriptor 227 * and call soclose() if they cannot. This represents our so_refs 228 * (which should be 1) from soalloc(). 229 */ 230 soclrstate(so, SS_NOFDREF); 231 232 /* 233 * Set a default port for protocol processing. No action will occur 234 * on the socket on this port until an inpcb is attached to it and 235 * is able to match incoming packets, or until the socket becomes 236 * available to userland. 237 * 238 * We normally default the socket to the protocol thread on cpu 0, 239 * if protocol does not provide its own method to initialize the 240 * default port. 241 * 242 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol 243 * thread and all pr_*()/pru_*() calls are executed synchronously. 244 */ 245 if (prp->pr_flags & PR_SYNC_PORT) 246 so->so_port = &netisr_sync_port; 247 else if (prp->pr_initport != NULL) 248 so->so_port = prp->pr_initport(); 249 else 250 so->so_port = netisr_cpuport(0); 251 252 TAILQ_INIT(&so->so_incomp); 253 TAILQ_INIT(&so->so_comp); 254 so->so_type = type; 255 so->so_cred = crhold(p->p_ucred); 256 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE]; 257 ai.p_ucred = p->p_ucred; 258 ai.fd_rdir = p->p_fd->fd_rdir; 259 260 /* 261 * Auto-sizing of socket buffers is managed by the protocols and 262 * the appropriate flags must be set in the pru_attach function. 263 */ 264 error = so_pru_attach(so, proto, &ai); 265 if (error) { 266 sosetstate(so, SS_NOFDREF); 267 sofree(so); /* from soalloc */ 268 return error; 269 } 270 271 /* 272 * NOTE: Returns referenced socket. 273 */ 274 *aso = so; 275 return (0); 276 } 277 278 int 279 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 280 { 281 int error; 282 283 error = so_pru_bind(so, nam, td); 284 return (error); 285 } 286 287 static void 288 sodealloc(struct socket *so) 289 { 290 if (so->so_rcv.ssb_hiwat) 291 (void)chgsbsize(so->so_cred->cr_uidinfo, 292 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY); 293 if (so->so_snd.ssb_hiwat) 294 (void)chgsbsize(so->so_cred->cr_uidinfo, 295 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY); 296 #ifdef INET 297 /* remove accept filter if present */ 298 if (so->so_accf != NULL) 299 do_setopt_accept_filter(so, NULL); 300 #endif /* INET */ 301 crfree(so->so_cred); 302 if (so->so_faddr != NULL) 303 kfree(so->so_faddr, M_SONAME); 304 kfree(so, M_SOCKET); 305 } 306 307 int 308 solisten(struct socket *so, int backlog, struct thread *td) 309 { 310 int error; 311 #ifdef SCTP 312 short oldopt, oldqlimit; 313 #endif /* SCTP */ 314 315 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) 316 return (EINVAL); 317 318 #ifdef SCTP 319 oldopt = so->so_options; 320 oldqlimit = so->so_qlimit; 321 #endif /* SCTP */ 322 323 lwkt_gettoken(&so->so_rcv.ssb_token); 324 if (TAILQ_EMPTY(&so->so_comp)) 325 so->so_options |= SO_ACCEPTCONN; 326 lwkt_reltoken(&so->so_rcv.ssb_token); 327 if (backlog < 0 || backlog > somaxconn) 328 backlog = somaxconn; 329 so->so_qlimit = backlog; 330 /* SCTP needs to look at tweak both the inbound backlog parameter AND 331 * the so_options (UDP model both connect's and gets inbound 332 * connections .. implicitly). 333 */ 334 error = so_pru_listen(so, td); 335 if (error) { 336 #ifdef SCTP 337 /* Restore the params */ 338 so->so_options = oldopt; 339 so->so_qlimit = oldqlimit; 340 #endif /* SCTP */ 341 return (error); 342 } 343 return (0); 344 } 345 346 /* 347 * Destroy a disconnected socket. This routine is a NOP if entities 348 * still have a reference on the socket: 349 * 350 * so_pcb - The protocol stack still has a reference 351 * SS_NOFDREF - There is no longer a file pointer reference 352 */ 353 void 354 sofree(struct socket *so) 355 { 356 struct socket *head; 357 358 /* 359 * This is a bit hackish at the moment. We need to interlock 360 * any accept queue we are on before we potentially lose the 361 * last reference to avoid races against a re-reference from 362 * someone operating on the queue. 363 */ 364 while ((head = so->so_head) != NULL) { 365 lwkt_getpooltoken(head); 366 if (so->so_head == head) 367 break; 368 lwkt_relpooltoken(head); 369 } 370 371 /* 372 * Arbitrage the last free. 373 */ 374 KKASSERT(so->so_refs > 0); 375 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) { 376 if (head) 377 lwkt_relpooltoken(head); 378 return; 379 } 380 381 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF)); 382 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0); 383 384 /* 385 * We're done, remove ourselves from the accept queue we are 386 * on, if we are on one. 387 */ 388 if (head != NULL) { 389 if (so->so_state & SS_INCOMP) { 390 TAILQ_REMOVE(&head->so_incomp, so, so_list); 391 head->so_incqlen--; 392 } else if (so->so_state & SS_COMP) { 393 /* 394 * We must not decommission a socket that's 395 * on the accept(2) queue. If we do, then 396 * accept(2) may hang after select(2) indicated 397 * that the listening socket was ready. 398 */ 399 lwkt_relpooltoken(head); 400 return; 401 } else { 402 panic("sofree: not queued"); 403 } 404 soclrstate(so, SS_INCOMP); 405 so->so_head = NULL; 406 lwkt_relpooltoken(head); 407 } 408 ssb_release(&so->so_snd, so); 409 sorflush(so); 410 sodealloc(so); 411 } 412 413 /* 414 * Close a socket on last file table reference removal. 415 * Initiate disconnect if connected. 416 * Free socket when disconnect complete. 417 */ 418 int 419 soclose(struct socket *so, int fflag) 420 { 421 int error; 422 423 funsetown(&so->so_sigio); 424 sosetstate(so, SS_ISCLOSING); 425 if (!use_soclose_fast || 426 (so->so_proto->pr_flags & PR_SYNC_PORT) || 427 ((so->so_state & SS_ISCONNECTED) && 428 (so->so_options & SO_LINGER))) { 429 error = soclose_sync(so, fflag); 430 } else { 431 soclose_fast(so); 432 error = 0; 433 } 434 return error; 435 } 436 437 void 438 sodiscard(struct socket *so) 439 { 440 lwkt_getpooltoken(so); 441 if (so->so_options & SO_ACCEPTCONN) { 442 struct socket *sp; 443 444 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) { 445 TAILQ_REMOVE(&so->so_incomp, sp, so_list); 446 soclrstate(sp, SS_INCOMP); 447 sp->so_head = NULL; 448 so->so_incqlen--; 449 soabort_async(sp); 450 } 451 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) { 452 TAILQ_REMOVE(&so->so_comp, sp, so_list); 453 soclrstate(sp, SS_COMP); 454 sp->so_head = NULL; 455 so->so_qlen--; 456 soabort_async(sp); 457 } 458 } 459 lwkt_relpooltoken(so); 460 461 if (so->so_state & SS_NOFDREF) 462 panic("soclose: NOFDREF"); 463 sosetstate(so, SS_NOFDREF); /* take ref */ 464 } 465 466 void 467 soinherit(struct socket *so, struct socket *so_inh) 468 { 469 TAILQ_HEAD(, socket) comp, incomp; 470 struct socket *sp; 471 int qlen, incqlen; 472 473 KASSERT(so->so_options & SO_ACCEPTCONN, 474 ("so does not accept connection")); 475 KASSERT(so_inh->so_options & SO_ACCEPTCONN, 476 ("so_inh does not accept connection")); 477 478 TAILQ_INIT(&comp); 479 TAILQ_INIT(&incomp); 480 481 lwkt_getpooltoken(so); 482 lwkt_getpooltoken(so_inh); 483 484 /* 485 * Save completed queue and incompleted queue 486 */ 487 TAILQ_CONCAT(&comp, &so->so_comp, so_list); 488 qlen = so->so_qlen; 489 so->so_qlen = 0; 490 491 TAILQ_CONCAT(&incomp, &so->so_incomp, so_list); 492 incqlen = so->so_incqlen; 493 so->so_incqlen = 0; 494 495 /* 496 * Append the saved completed queue and incompleted 497 * queue to the socket inherits them. 498 * 499 * XXX 500 * This may temporarily break the inheriting socket's 501 * so_qlimit. 502 */ 503 TAILQ_FOREACH(sp, &comp, so_list) { 504 sp->so_head = so_inh; 505 crfree(sp->so_cred); 506 sp->so_cred = crhold(so_inh->so_cred); 507 } 508 509 TAILQ_FOREACH(sp, &incomp, so_list) { 510 sp->so_head = so_inh; 511 crfree(sp->so_cred); 512 sp->so_cred = crhold(so_inh->so_cred); 513 } 514 515 TAILQ_CONCAT(&so_inh->so_comp, &comp, so_list); 516 so_inh->so_qlen += qlen; 517 518 TAILQ_CONCAT(&so_inh->so_incomp, &incomp, so_list); 519 so_inh->so_incqlen += incqlen; 520 521 lwkt_relpooltoken(so_inh); 522 lwkt_relpooltoken(so); 523 524 if (qlen) { 525 /* 526 * "New" connections have arrived 527 */ 528 sorwakeup(so_inh); 529 wakeup(&so_inh->so_timeo); 530 } 531 } 532 533 static int 534 soclose_sync(struct socket *so, int fflag) 535 { 536 int error = 0; 537 538 if (so->so_pcb == NULL) 539 goto discard; 540 if (so->so_state & SS_ISCONNECTED) { 541 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 542 error = sodisconnect(so); 543 if (error) 544 goto drop; 545 } 546 if (so->so_options & SO_LINGER) { 547 if ((so->so_state & SS_ISDISCONNECTING) && 548 (fflag & FNONBLOCK)) 549 goto drop; 550 while (so->so_state & SS_ISCONNECTED) { 551 error = tsleep(&so->so_timeo, PCATCH, 552 "soclos", so->so_linger * hz); 553 if (error) 554 break; 555 } 556 } 557 } 558 drop: 559 if (so->so_pcb) { 560 int error2; 561 562 error2 = so_pru_detach(so); 563 if (error2 == EJUSTRETURN) { 564 /* 565 * Protocol will call sodiscard() 566 * and sofree() for us. 567 */ 568 return error; 569 } 570 if (error == 0) 571 error = error2; 572 } 573 discard: 574 sodiscard(so); 575 so_pru_sync(so); /* unpend async sending */ 576 sofree(so); /* dispose of ref */ 577 578 return (error); 579 } 580 581 static void 582 soclose_sofree_async_handler(netmsg_t msg) 583 { 584 sofree(msg->base.nm_so); 585 } 586 587 static void 588 soclose_sofree_async(struct socket *so) 589 { 590 struct netmsg_base *base = &so->so_clomsg; 591 592 netmsg_init(base, so, &netisr_apanic_rport, 0, 593 soclose_sofree_async_handler); 594 lwkt_sendmsg(so->so_port, &base->lmsg); 595 } 596 597 static void 598 soclose_disconn_async_handler(netmsg_t msg) 599 { 600 struct socket *so = msg->base.nm_so; 601 602 if ((so->so_state & SS_ISCONNECTED) && 603 (so->so_state & SS_ISDISCONNECTING) == 0) 604 so_pru_disconnect_direct(so); 605 606 if (so->so_pcb) { 607 int error; 608 609 error = so_pru_detach_direct(so); 610 if (error == EJUSTRETURN) { 611 /* 612 * Protocol will call sodiscard() 613 * and sofree() for us. 614 */ 615 return; 616 } 617 } 618 619 sodiscard(so); 620 sofree(so); 621 } 622 623 static void 624 soclose_disconn_async(struct socket *so) 625 { 626 struct netmsg_base *base = &so->so_clomsg; 627 628 netmsg_init(base, so, &netisr_apanic_rport, 0, 629 soclose_disconn_async_handler); 630 lwkt_sendmsg(so->so_port, &base->lmsg); 631 } 632 633 static void 634 soclose_detach_async_handler(netmsg_t msg) 635 { 636 struct socket *so = msg->base.nm_so; 637 638 if (so->so_pcb) { 639 int error; 640 641 error = so_pru_detach_direct(so); 642 if (error == EJUSTRETURN) { 643 /* 644 * Protocol will call sodiscard() 645 * and sofree() for us. 646 */ 647 return; 648 } 649 } 650 651 sodiscard(so); 652 sofree(so); 653 } 654 655 static void 656 soclose_detach_async(struct socket *so) 657 { 658 struct netmsg_base *base = &so->so_clomsg; 659 660 netmsg_init(base, so, &netisr_apanic_rport, 0, 661 soclose_detach_async_handler); 662 lwkt_sendmsg(so->so_port, &base->lmsg); 663 } 664 665 static void 666 soclose_fast(struct socket *so) 667 { 668 if (so->so_pcb == NULL) 669 goto discard; 670 671 if ((so->so_state & SS_ISCONNECTED) && 672 (so->so_state & SS_ISDISCONNECTING) == 0) { 673 soclose_disconn_async(so); 674 return; 675 } 676 677 if (so->so_pcb) { 678 soclose_detach_async(so); 679 return; 680 } 681 682 discard: 683 sodiscard(so); 684 soclose_sofree_async(so); 685 } 686 687 /* 688 * Abort and destroy a socket. Only one abort can be in progress 689 * at any given moment. 690 */ 691 void 692 soabort(struct socket *so) 693 { 694 soreference(so); 695 so_pru_abort(so); 696 } 697 698 void 699 soabort_async(struct socket *so) 700 { 701 soreference(so); 702 so_pru_abort_async(so); 703 } 704 705 void 706 soabort_oncpu(struct socket *so) 707 { 708 soreference(so); 709 so_pru_abort_direct(so); 710 } 711 712 /* 713 * so is passed in ref'd, which becomes owned by 714 * the cleared SS_NOFDREF flag. 715 */ 716 void 717 soaccept_generic(struct socket *so) 718 { 719 if ((so->so_state & SS_NOFDREF) == 0) 720 panic("soaccept: !NOFDREF"); 721 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */ 722 } 723 724 int 725 soaccept(struct socket *so, struct sockaddr **nam) 726 { 727 int error; 728 729 soaccept_generic(so); 730 error = so_pru_accept(so, nam); 731 return (error); 732 } 733 734 int 735 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td, 736 boolean_t sync) 737 { 738 int error; 739 740 if (so->so_options & SO_ACCEPTCONN) 741 return (EOPNOTSUPP); 742 /* 743 * If protocol is connection-based, can only connect once. 744 * Otherwise, if connected, try to disconnect first. 745 * This allows user to disconnect by connecting to, e.g., 746 * a null address. 747 */ 748 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 749 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 750 (error = sodisconnect(so)))) { 751 error = EISCONN; 752 } else { 753 /* 754 * Prevent accumulated error from previous connection 755 * from biting us. 756 */ 757 so->so_error = 0; 758 if (!sync && so->so_proto->pr_usrreqs->pru_preconnect) 759 error = so_pru_connect_async(so, nam, td); 760 else 761 error = so_pru_connect(so, nam, td); 762 } 763 return (error); 764 } 765 766 int 767 soconnect2(struct socket *so1, struct socket *so2) 768 { 769 int error; 770 771 error = so_pru_connect2(so1, so2); 772 return (error); 773 } 774 775 int 776 sodisconnect(struct socket *so) 777 { 778 int error; 779 780 if ((so->so_state & SS_ISCONNECTED) == 0) { 781 error = ENOTCONN; 782 goto bad; 783 } 784 if (so->so_state & SS_ISDISCONNECTING) { 785 error = EALREADY; 786 goto bad; 787 } 788 error = so_pru_disconnect(so); 789 bad: 790 return (error); 791 } 792 793 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 794 /* 795 * Send on a socket. 796 * If send must go all at once and message is larger than 797 * send buffering, then hard error. 798 * Lock against other senders. 799 * If must go all at once and not enough room now, then 800 * inform user that this would block and do nothing. 801 * Otherwise, if nonblocking, send as much as possible. 802 * The data to be sent is described by "uio" if nonzero, 803 * otherwise by the mbuf chain "top" (which must be null 804 * if uio is not). Data provided in mbuf chain must be small 805 * enough to send all at once. 806 * 807 * Returns nonzero on error, timeout or signal; callers 808 * must check for short counts if EINTR/ERESTART are returned. 809 * Data and control buffers are freed on return. 810 */ 811 int 812 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 813 struct mbuf *top, struct mbuf *control, int flags, 814 struct thread *td) 815 { 816 struct mbuf **mp; 817 struct mbuf *m; 818 size_t resid; 819 int space, len; 820 int clen = 0, error, dontroute, mlen; 821 int atomic = sosendallatonce(so) || top; 822 int pru_flags; 823 824 if (uio) { 825 resid = uio->uio_resid; 826 } else { 827 resid = (size_t)top->m_pkthdr.len; 828 #ifdef INVARIANTS 829 len = 0; 830 for (m = top; m; m = m->m_next) 831 len += m->m_len; 832 KKASSERT(top->m_pkthdr.len == len); 833 #endif 834 } 835 836 /* 837 * WARNING! resid is unsigned, space and len are signed. space 838 * can wind up negative if the sockbuf is overcommitted. 839 * 840 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 841 * type sockets since that's an error. 842 */ 843 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) { 844 error = EINVAL; 845 goto out; 846 } 847 848 dontroute = 849 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 850 (so->so_proto->pr_flags & PR_ATOMIC); 851 if (td->td_lwp != NULL) 852 td->td_lwp->lwp_ru.ru_msgsnd++; 853 if (control) 854 clen = control->m_len; 855 #define gotoerr(errcode) { error = errcode; goto release; } 856 857 restart: 858 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags)); 859 if (error) 860 goto out; 861 862 do { 863 if (so->so_state & SS_CANTSENDMORE) 864 gotoerr(EPIPE); 865 if (so->so_error) { 866 error = so->so_error; 867 so->so_error = 0; 868 goto release; 869 } 870 if ((so->so_state & SS_ISCONNECTED) == 0) { 871 /* 872 * `sendto' and `sendmsg' is allowed on a connection- 873 * based socket if it supports implied connect. 874 * Return ENOTCONN if not connected and no address is 875 * supplied. 876 */ 877 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 878 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 879 if ((so->so_state & SS_ISCONFIRMING) == 0 && 880 !(resid == 0 && clen != 0)) 881 gotoerr(ENOTCONN); 882 } else if (addr == NULL) 883 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ? 884 ENOTCONN : EDESTADDRREQ); 885 } 886 if ((atomic && resid > so->so_snd.ssb_hiwat) || 887 clen > so->so_snd.ssb_hiwat) { 888 gotoerr(EMSGSIZE); 889 } 890 space = ssb_space(&so->so_snd); 891 if (flags & MSG_OOB) 892 space += 1024; 893 if ((space < 0 || (size_t)space < resid + clen) && uio && 894 (atomic || space < so->so_snd.ssb_lowat || space < clen)) { 895 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) 896 gotoerr(EWOULDBLOCK); 897 ssb_unlock(&so->so_snd); 898 error = ssb_wait(&so->so_snd); 899 if (error) 900 goto out; 901 goto restart; 902 } 903 mp = ⊤ 904 space -= clen; 905 do { 906 if (uio == NULL) { 907 /* 908 * Data is prepackaged in "top". 909 */ 910 resid = 0; 911 if (flags & MSG_EOR) 912 top->m_flags |= M_EOR; 913 } else do { 914 if (resid > INT_MAX) 915 resid = INT_MAX; 916 m = m_getl((int)resid, MB_WAIT, MT_DATA, 917 top == NULL ? M_PKTHDR : 0, &mlen); 918 if (top == NULL) { 919 m->m_pkthdr.len = 0; 920 m->m_pkthdr.rcvif = NULL; 921 } 922 len = imin((int)szmin(mlen, resid), space); 923 if (resid < MINCLSIZE) { 924 /* 925 * For datagram protocols, leave room 926 * for protocol headers in first mbuf. 927 */ 928 if (atomic && top == NULL && len < mlen) 929 MH_ALIGN(m, len); 930 } 931 space -= len; 932 error = uiomove(mtod(m, caddr_t), (size_t)len, uio); 933 resid = uio->uio_resid; 934 m->m_len = len; 935 *mp = m; 936 top->m_pkthdr.len += len; 937 if (error) 938 goto release; 939 mp = &m->m_next; 940 if (resid == 0) { 941 if (flags & MSG_EOR) 942 top->m_flags |= M_EOR; 943 break; 944 } 945 } while (space > 0 && atomic); 946 if (dontroute) 947 so->so_options |= SO_DONTROUTE; 948 if (flags & MSG_OOB) { 949 pru_flags = PRUS_OOB; 950 } else if ((flags & MSG_EOF) && 951 (so->so_proto->pr_flags & PR_IMPLOPCL) && 952 (resid == 0)) { 953 /* 954 * If the user set MSG_EOF, the protocol 955 * understands this flag and nothing left to 956 * send then use PRU_SEND_EOF instead of PRU_SEND. 957 */ 958 pru_flags = PRUS_EOF; 959 } else if (resid > 0 && space > 0) { 960 /* If there is more to send, set PRUS_MORETOCOME */ 961 pru_flags = PRUS_MORETOCOME; 962 } else { 963 pru_flags = 0; 964 } 965 /* 966 * XXX all the SS_CANTSENDMORE checks previously 967 * done could be out of date. We could have recieved 968 * a reset packet in an interrupt or maybe we slept 969 * while doing page faults in uiomove() etc. We could 970 * probably recheck again inside the splnet() protection 971 * here, but there are probably other places that this 972 * also happens. We must rethink this. 973 */ 974 error = so_pru_send(so, pru_flags, top, addr, control, td); 975 if (dontroute) 976 so->so_options &= ~SO_DONTROUTE; 977 clen = 0; 978 control = NULL; 979 top = NULL; 980 mp = ⊤ 981 if (error) 982 goto release; 983 } while (resid && space > 0); 984 } while (resid); 985 986 release: 987 ssb_unlock(&so->so_snd); 988 out: 989 if (top) 990 m_freem(top); 991 if (control) 992 m_freem(control); 993 return (error); 994 } 995 996 #ifdef INET 997 /* 998 * A specialization of sosend() for UDP based on protocol-specific knowledge: 999 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that 1000 * sosendallatonce() returns true, 1001 * the "atomic" variable is true, 1002 * and sosendudp() blocks until space is available for the entire send. 1003 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or 1004 * PR_IMPLOPCL flags set. 1005 * UDP has no out-of-band data. 1006 * UDP has no control data. 1007 * UDP does not support MSG_EOR. 1008 */ 1009 int 1010 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio, 1011 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1012 { 1013 size_t resid; 1014 int error, pru_flags = 0; 1015 int space; 1016 1017 if (td->td_lwp != NULL) 1018 td->td_lwp->lwp_ru.ru_msgsnd++; 1019 if (control) 1020 m_freem(control); 1021 1022 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp")); 1023 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len; 1024 1025 restart: 1026 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags)); 1027 if (error) 1028 goto out; 1029 1030 if (so->so_state & SS_CANTSENDMORE) 1031 gotoerr(EPIPE); 1032 if (so->so_error) { 1033 error = so->so_error; 1034 so->so_error = 0; 1035 goto release; 1036 } 1037 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL) 1038 gotoerr(EDESTADDRREQ); 1039 if (resid > so->so_snd.ssb_hiwat) 1040 gotoerr(EMSGSIZE); 1041 space = ssb_space(&so->so_snd); 1042 if (uio && (space < 0 || (size_t)space < resid)) { 1043 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) 1044 gotoerr(EWOULDBLOCK); 1045 ssb_unlock(&so->so_snd); 1046 error = ssb_wait(&so->so_snd); 1047 if (error) 1048 goto out; 1049 goto restart; 1050 } 1051 1052 if (uio) { 1053 int hdrlen = max_hdr; 1054 1055 /* 1056 * We try to optimize out the additional mbuf 1057 * allocations in M_PREPEND() on output path, e.g. 1058 * - udp_output(), when it tries to prepend protocol 1059 * headers. 1060 * - Link layer output function, when it tries to 1061 * prepend link layer header. 1062 * 1063 * This probably will not benefit any data that will 1064 * be fragmented, so this optimization is only performed 1065 * when the size of data and max size of protocol+link 1066 * headers fit into one mbuf cluster. 1067 */ 1068 if (uio->uio_resid > MCLBYTES - hdrlen || 1069 !udp_sosend_prepend) { 1070 top = m_uiomove(uio); 1071 if (top == NULL) 1072 goto release; 1073 } else { 1074 int nsize; 1075 1076 top = m_getl(uio->uio_resid + hdrlen, MB_WAIT, 1077 MT_DATA, M_PKTHDR, &nsize); 1078 KASSERT(nsize >= uio->uio_resid + hdrlen, 1079 ("sosendudp invalid nsize %d, " 1080 "resid %zu, hdrlen %d", 1081 nsize, uio->uio_resid, hdrlen)); 1082 1083 top->m_len = uio->uio_resid; 1084 top->m_pkthdr.len = uio->uio_resid; 1085 top->m_data += hdrlen; 1086 1087 error = uiomove(mtod(top, caddr_t), top->m_len, uio); 1088 if (error) 1089 goto out; 1090 } 1091 } 1092 1093 if (flags & MSG_DONTROUTE) 1094 pru_flags |= PRUS_DONTROUTE; 1095 1096 if (udp_sosend_async && (flags & MSG_SYNC) == 0) { 1097 so_pru_send_async(so, pru_flags, top, addr, NULL, td); 1098 error = 0; 1099 } else { 1100 error = so_pru_send(so, pru_flags, top, addr, NULL, td); 1101 } 1102 top = NULL; /* sent or freed in lower layer */ 1103 1104 release: 1105 ssb_unlock(&so->so_snd); 1106 out: 1107 if (top) 1108 m_freem(top); 1109 return (error); 1110 } 1111 1112 int 1113 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio, 1114 struct mbuf *top, struct mbuf *control, int flags, 1115 struct thread *td) 1116 { 1117 struct mbuf **mp; 1118 struct mbuf *m; 1119 size_t resid; 1120 int space, len; 1121 int error, mlen; 1122 int allatonce; 1123 int pru_flags; 1124 1125 if (uio) { 1126 KKASSERT(top == NULL); 1127 allatonce = 0; 1128 resid = uio->uio_resid; 1129 } else { 1130 allatonce = 1; 1131 resid = (size_t)top->m_pkthdr.len; 1132 #ifdef INVARIANTS 1133 len = 0; 1134 for (m = top; m; m = m->m_next) 1135 len += m->m_len; 1136 KKASSERT(top->m_pkthdr.len == len); 1137 #endif 1138 } 1139 1140 /* 1141 * WARNING! resid is unsigned, space and len are signed. space 1142 * can wind up negative if the sockbuf is overcommitted. 1143 * 1144 * Also check to make sure that MSG_EOR isn't used on TCP 1145 */ 1146 if (flags & MSG_EOR) { 1147 error = EINVAL; 1148 goto out; 1149 } 1150 1151 if (control) { 1152 /* TCP doesn't do control messages (rights, creds, etc) */ 1153 if (control->m_len) { 1154 error = EINVAL; 1155 goto out; 1156 } 1157 m_freem(control); /* empty control, just free it */ 1158 control = NULL; 1159 } 1160 1161 if (td->td_lwp != NULL) 1162 td->td_lwp->lwp_ru.ru_msgsnd++; 1163 1164 #define gotoerr(errcode) { error = errcode; goto release; } 1165 1166 restart: 1167 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags)); 1168 if (error) 1169 goto out; 1170 1171 do { 1172 if (so->so_state & SS_CANTSENDMORE) 1173 gotoerr(EPIPE); 1174 if (so->so_error) { 1175 error = so->so_error; 1176 so->so_error = 0; 1177 goto release; 1178 } 1179 if ((so->so_state & SS_ISCONNECTED) == 0 && 1180 (so->so_state & SS_ISCONFIRMING) == 0) 1181 gotoerr(ENOTCONN); 1182 if (allatonce && resid > so->so_snd.ssb_hiwat) 1183 gotoerr(EMSGSIZE); 1184 1185 space = ssb_space_prealloc(&so->so_snd); 1186 if (flags & MSG_OOB) 1187 space += 1024; 1188 if ((space < 0 || (size_t)space < resid) && !allatonce && 1189 space < so->so_snd.ssb_lowat) { 1190 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) 1191 gotoerr(EWOULDBLOCK); 1192 ssb_unlock(&so->so_snd); 1193 error = ssb_wait(&so->so_snd); 1194 if (error) 1195 goto out; 1196 goto restart; 1197 } 1198 mp = ⊤ 1199 do { 1200 int cnt = 0, async = 0; 1201 1202 if (uio == NULL) { 1203 /* 1204 * Data is prepackaged in "top". 1205 */ 1206 resid = 0; 1207 } else do { 1208 if (resid > INT_MAX) 1209 resid = INT_MAX; 1210 if (tcp_sosend_jcluster) { 1211 m = m_getlj((int)resid, MB_WAIT, MT_DATA, 1212 top == NULL ? M_PKTHDR : 0, &mlen); 1213 } else { 1214 m = m_getl((int)resid, MB_WAIT, MT_DATA, 1215 top == NULL ? M_PKTHDR : 0, &mlen); 1216 } 1217 if (top == NULL) { 1218 m->m_pkthdr.len = 0; 1219 m->m_pkthdr.rcvif = NULL; 1220 } 1221 len = imin((int)szmin(mlen, resid), space); 1222 space -= len; 1223 error = uiomove(mtod(m, caddr_t), (size_t)len, uio); 1224 resid = uio->uio_resid; 1225 m->m_len = len; 1226 *mp = m; 1227 top->m_pkthdr.len += len; 1228 if (error) 1229 goto release; 1230 mp = &m->m_next; 1231 if (resid == 0) 1232 break; 1233 ++cnt; 1234 } while (space > 0 && cnt < tcp_sosend_agglim); 1235 1236 if (tcp_sosend_async) 1237 async = 1; 1238 1239 if (flags & MSG_OOB) { 1240 pru_flags = PRUS_OOB; 1241 async = 0; 1242 } else if ((flags & MSG_EOF) && resid == 0) { 1243 pru_flags = PRUS_EOF; 1244 } else if (resid > 0 && space > 0) { 1245 /* If there is more to send, set PRUS_MORETOCOME */ 1246 pru_flags = PRUS_MORETOCOME; 1247 async = 1; 1248 } else { 1249 pru_flags = 0; 1250 } 1251 1252 if (flags & MSG_SYNC) 1253 async = 0; 1254 1255 /* 1256 * XXX all the SS_CANTSENDMORE checks previously 1257 * done could be out of date. We could have recieved 1258 * a reset packet in an interrupt or maybe we slept 1259 * while doing page faults in uiomove() etc. We could 1260 * probably recheck again inside the splnet() protection 1261 * here, but there are probably other places that this 1262 * also happens. We must rethink this. 1263 */ 1264 for (m = top; m; m = m->m_next) 1265 ssb_preallocstream(&so->so_snd, m); 1266 if (!async) { 1267 error = so_pru_send(so, pru_flags, top, 1268 NULL, NULL, td); 1269 } else { 1270 so_pru_send_async(so, pru_flags, top, 1271 NULL, NULL, td); 1272 error = 0; 1273 } 1274 1275 top = NULL; 1276 mp = ⊤ 1277 if (error) 1278 goto release; 1279 } while (resid && space > 0); 1280 } while (resid); 1281 1282 release: 1283 ssb_unlock(&so->so_snd); 1284 out: 1285 if (top) 1286 m_freem(top); 1287 if (control) 1288 m_freem(control); 1289 return (error); 1290 } 1291 #endif 1292 1293 /* 1294 * Implement receive operations on a socket. 1295 * 1296 * We depend on the way that records are added to the signalsockbuf 1297 * by sbappend*. In particular, each record (mbufs linked through m_next) 1298 * must begin with an address if the protocol so specifies, 1299 * followed by an optional mbuf or mbufs containing ancillary data, 1300 * and then zero or more mbufs of data. 1301 * 1302 * Although the signalsockbuf is locked, new data may still be appended. 1303 * A token inside the ssb_lock deals with MP issues and still allows 1304 * the network to access the socket if we block in a uio. 1305 * 1306 * The caller may receive the data as a single mbuf chain by supplying 1307 * an mbuf **mp0 for use in returning the chain. The uio is then used 1308 * only for the count in uio_resid. 1309 */ 1310 int 1311 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 1312 struct sockbuf *sio, struct mbuf **controlp, int *flagsp) 1313 { 1314 struct mbuf *m, *n; 1315 struct mbuf *free_chain = NULL; 1316 int flags, len, error, offset; 1317 struct protosw *pr = so->so_proto; 1318 int moff, type = 0; 1319 size_t resid, orig_resid; 1320 1321 if (uio) 1322 resid = uio->uio_resid; 1323 else 1324 resid = (size_t)(sio->sb_climit - sio->sb_cc); 1325 orig_resid = resid; 1326 1327 if (psa) 1328 *psa = NULL; 1329 if (controlp) 1330 *controlp = NULL; 1331 if (flagsp) 1332 flags = *flagsp &~ MSG_EOR; 1333 else 1334 flags = 0; 1335 if (flags & MSG_OOB) { 1336 m = m_get(MB_WAIT, MT_DATA); 1337 if (m == NULL) 1338 return (ENOBUFS); 1339 error = so_pru_rcvoob(so, m, flags & MSG_PEEK); 1340 if (error) 1341 goto bad; 1342 if (sio) { 1343 do { 1344 sbappend(sio, m); 1345 KKASSERT(resid >= (size_t)m->m_len); 1346 resid -= (size_t)m->m_len; 1347 } while (resid > 0 && m); 1348 } else { 1349 do { 1350 uio->uio_resid = resid; 1351 error = uiomove(mtod(m, caddr_t), 1352 (int)szmin(resid, m->m_len), 1353 uio); 1354 resid = uio->uio_resid; 1355 m = m_free(m); 1356 } while (uio->uio_resid && error == 0 && m); 1357 } 1358 bad: 1359 if (m) 1360 m_freem(m); 1361 return (error); 1362 } 1363 if ((so->so_state & SS_ISCONFIRMING) && resid) 1364 so_pru_rcvd(so, 0); 1365 1366 /* 1367 * The token interlocks against the protocol thread while 1368 * ssb_lock is a blocking lock against other userland entities. 1369 */ 1370 lwkt_gettoken(&so->so_rcv.ssb_token); 1371 restart: 1372 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags)); 1373 if (error) 1374 goto done; 1375 1376 m = so->so_rcv.ssb_mb; 1377 /* 1378 * If we have less data than requested, block awaiting more 1379 * (subject to any timeout) if: 1380 * 1. the current count is less than the low water mark, or 1381 * 2. MSG_WAITALL is set, and it is possible to do the entire 1382 * receive operation at once if we block (resid <= hiwat). 1383 * 3. MSG_DONTWAIT is not set 1384 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1385 * we have to do the receive in sections, and thus risk returning 1386 * a short count if a timeout or signal occurs after we start. 1387 */ 1388 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1389 (size_t)so->so_rcv.ssb_cc < resid) && 1390 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat || 1391 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) && 1392 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1393 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1")); 1394 if (so->so_error) { 1395 if (m) 1396 goto dontblock; 1397 error = so->so_error; 1398 if ((flags & MSG_PEEK) == 0) 1399 so->so_error = 0; 1400 goto release; 1401 } 1402 if (so->so_state & SS_CANTRCVMORE) { 1403 if (m) 1404 goto dontblock; 1405 else 1406 goto release; 1407 } 1408 for (; m; m = m->m_next) { 1409 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1410 m = so->so_rcv.ssb_mb; 1411 goto dontblock; 1412 } 1413 } 1414 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1415 (pr->pr_flags & PR_CONNREQUIRED)) { 1416 error = ENOTCONN; 1417 goto release; 1418 } 1419 if (resid == 0) 1420 goto release; 1421 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) { 1422 error = EWOULDBLOCK; 1423 goto release; 1424 } 1425 ssb_unlock(&so->so_rcv); 1426 error = ssb_wait(&so->so_rcv); 1427 if (error) 1428 goto done; 1429 goto restart; 1430 } 1431 dontblock: 1432 if (uio && uio->uio_td && uio->uio_td->td_proc) 1433 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++; 1434 1435 /* 1436 * note: m should be == sb_mb here. Cache the next record while 1437 * cleaning up. Note that calling m_free*() will break out critical 1438 * section. 1439 */ 1440 KKASSERT(m == so->so_rcv.ssb_mb); 1441 1442 /* 1443 * Skip any address mbufs prepending the record. 1444 */ 1445 if (pr->pr_flags & PR_ADDR) { 1446 KASSERT(m->m_type == MT_SONAME, ("receive 1a")); 1447 orig_resid = 0; 1448 if (psa) 1449 *psa = dup_sockaddr(mtod(m, struct sockaddr *)); 1450 if (flags & MSG_PEEK) 1451 m = m->m_next; 1452 else 1453 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1454 } 1455 1456 /* 1457 * Skip any control mbufs prepending the record. 1458 */ 1459 #ifdef SCTP 1460 if (pr->pr_flags & PR_ADDR_OPT) { 1461 /* 1462 * For SCTP we may be getting a 1463 * whole message OR a partial delivery. 1464 */ 1465 if (m && m->m_type == MT_SONAME) { 1466 orig_resid = 0; 1467 if (psa) 1468 *psa = dup_sockaddr(mtod(m, struct sockaddr *)); 1469 if (flags & MSG_PEEK) 1470 m = m->m_next; 1471 else 1472 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1473 } 1474 } 1475 #endif /* SCTP */ 1476 while (m && m->m_type == MT_CONTROL && error == 0) { 1477 if (flags & MSG_PEEK) { 1478 if (controlp) 1479 *controlp = m_copy(m, 0, m->m_len); 1480 m = m->m_next; /* XXX race */ 1481 } else { 1482 if (controlp) { 1483 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL); 1484 if (pr->pr_domain->dom_externalize && 1485 mtod(m, struct cmsghdr *)->cmsg_type == 1486 SCM_RIGHTS) 1487 error = (*pr->pr_domain->dom_externalize)(m); 1488 *controlp = m; 1489 m = n; 1490 } else { 1491 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1492 } 1493 } 1494 if (controlp && *controlp) { 1495 orig_resid = 0; 1496 controlp = &(*controlp)->m_next; 1497 } 1498 } 1499 1500 /* 1501 * flag OOB data. 1502 */ 1503 if (m) { 1504 type = m->m_type; 1505 if (type == MT_OOBDATA) 1506 flags |= MSG_OOB; 1507 } 1508 1509 /* 1510 * Copy to the UIO or mbuf return chain (*mp). 1511 */ 1512 moff = 0; 1513 offset = 0; 1514 while (m && resid > 0 && error == 0) { 1515 if (m->m_type == MT_OOBDATA) { 1516 if (type != MT_OOBDATA) 1517 break; 1518 } else if (type == MT_OOBDATA) 1519 break; 1520 else 1521 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER, 1522 ("receive 3")); 1523 soclrstate(so, SS_RCVATMARK); 1524 len = (resid > INT_MAX) ? INT_MAX : resid; 1525 if (so->so_oobmark && len > so->so_oobmark - offset) 1526 len = so->so_oobmark - offset; 1527 if (len > m->m_len - moff) 1528 len = m->m_len - moff; 1529 1530 /* 1531 * Copy out to the UIO or pass the mbufs back to the SIO. 1532 * The SIO is dealt with when we eat the mbuf, but deal 1533 * with the resid here either way. 1534 */ 1535 if (uio) { 1536 uio->uio_resid = resid; 1537 error = uiomove(mtod(m, caddr_t) + moff, len, uio); 1538 resid = uio->uio_resid; 1539 if (error) 1540 goto release; 1541 } else { 1542 resid -= (size_t)len; 1543 } 1544 1545 /* 1546 * Eat the entire mbuf or just a piece of it 1547 */ 1548 if (len == m->m_len - moff) { 1549 if (m->m_flags & M_EOR) 1550 flags |= MSG_EOR; 1551 #ifdef SCTP 1552 if (m->m_flags & M_NOTIFICATION) 1553 flags |= MSG_NOTIFICATION; 1554 #endif /* SCTP */ 1555 if (flags & MSG_PEEK) { 1556 m = m->m_next; 1557 moff = 0; 1558 } else { 1559 if (sio) { 1560 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL); 1561 sbappend(sio, m); 1562 m = n; 1563 } else { 1564 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1565 } 1566 } 1567 } else { 1568 if (flags & MSG_PEEK) { 1569 moff += len; 1570 } else { 1571 if (sio) { 1572 n = m_copym(m, 0, len, MB_WAIT); 1573 if (n) 1574 sbappend(sio, n); 1575 } 1576 m->m_data += len; 1577 m->m_len -= len; 1578 so->so_rcv.ssb_cc -= len; 1579 } 1580 } 1581 if (so->so_oobmark) { 1582 if ((flags & MSG_PEEK) == 0) { 1583 so->so_oobmark -= len; 1584 if (so->so_oobmark == 0) { 1585 sosetstate(so, SS_RCVATMARK); 1586 break; 1587 } 1588 } else { 1589 offset += len; 1590 if (offset == so->so_oobmark) 1591 break; 1592 } 1593 } 1594 if (flags & MSG_EOR) 1595 break; 1596 /* 1597 * If the MSG_WAITALL flag is set (for non-atomic socket), 1598 * we must not quit until resid == 0 or an error 1599 * termination. If a signal/timeout occurs, return 1600 * with a short count but without error. 1601 * Keep signalsockbuf locked against other readers. 1602 */ 1603 while ((flags & MSG_WAITALL) && m == NULL && 1604 resid > 0 && !sosendallatonce(so) && 1605 so->so_rcv.ssb_mb == NULL) { 1606 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1607 break; 1608 /* 1609 * The window might have closed to zero, make 1610 * sure we send an ack now that we've drained 1611 * the buffer or we might end up blocking until 1612 * the idle takes over (5 seconds). 1613 */ 1614 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1615 so_pru_rcvd(so, flags); 1616 error = ssb_wait(&so->so_rcv); 1617 if (error) { 1618 ssb_unlock(&so->so_rcv); 1619 error = 0; 1620 goto done; 1621 } 1622 m = so->so_rcv.ssb_mb; 1623 } 1624 } 1625 1626 /* 1627 * If an atomic read was requested but unread data still remains 1628 * in the record, set MSG_TRUNC. 1629 */ 1630 if (m && pr->pr_flags & PR_ATOMIC) 1631 flags |= MSG_TRUNC; 1632 1633 /* 1634 * Cleanup. If an atomic read was requested drop any unread data. 1635 */ 1636 if ((flags & MSG_PEEK) == 0) { 1637 if (m && (pr->pr_flags & PR_ATOMIC)) 1638 sbdroprecord(&so->so_rcv.sb); 1639 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1640 so_pru_rcvd(so, flags); 1641 } 1642 1643 if (orig_resid == resid && orig_resid && 1644 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1645 ssb_unlock(&so->so_rcv); 1646 goto restart; 1647 } 1648 1649 if (flagsp) 1650 *flagsp |= flags; 1651 release: 1652 ssb_unlock(&so->so_rcv); 1653 done: 1654 lwkt_reltoken(&so->so_rcv.ssb_token); 1655 if (free_chain) 1656 m_freem(free_chain); 1657 return (error); 1658 } 1659 1660 int 1661 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio, 1662 struct sockbuf *sio, struct mbuf **controlp, int *flagsp) 1663 { 1664 struct mbuf *m, *n; 1665 struct mbuf *free_chain = NULL; 1666 int flags, len, error, offset; 1667 struct protosw *pr = so->so_proto; 1668 int moff; 1669 int didoob; 1670 size_t resid, orig_resid, restmp; 1671 1672 if (uio) 1673 resid = uio->uio_resid; 1674 else 1675 resid = (size_t)(sio->sb_climit - sio->sb_cc); 1676 orig_resid = resid; 1677 1678 if (psa) 1679 *psa = NULL; 1680 if (controlp) 1681 *controlp = NULL; 1682 if (flagsp) 1683 flags = *flagsp &~ MSG_EOR; 1684 else 1685 flags = 0; 1686 if (flags & MSG_OOB) { 1687 m = m_get(MB_WAIT, MT_DATA); 1688 if (m == NULL) 1689 return (ENOBUFS); 1690 error = so_pru_rcvoob(so, m, flags & MSG_PEEK); 1691 if (error) 1692 goto bad; 1693 if (sio) { 1694 do { 1695 sbappend(sio, m); 1696 KKASSERT(resid >= (size_t)m->m_len); 1697 resid -= (size_t)m->m_len; 1698 } while (resid > 0 && m); 1699 } else { 1700 do { 1701 uio->uio_resid = resid; 1702 error = uiomove(mtod(m, caddr_t), 1703 (int)szmin(resid, m->m_len), 1704 uio); 1705 resid = uio->uio_resid; 1706 m = m_free(m); 1707 } while (uio->uio_resid && error == 0 && m); 1708 } 1709 bad: 1710 if (m) 1711 m_freem(m); 1712 return (error); 1713 } 1714 1715 /* 1716 * The token interlocks against the protocol thread while 1717 * ssb_lock is a blocking lock against other userland entities. 1718 * 1719 * Lock a limited number of mbufs (not all, so sbcompress() still 1720 * works well). The token is used as an interlock for sbwait() so 1721 * release it afterwords. 1722 */ 1723 restart: 1724 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags)); 1725 if (error) 1726 goto done; 1727 1728 lwkt_gettoken(&so->so_rcv.ssb_token); 1729 m = so->so_rcv.ssb_mb; 1730 1731 /* 1732 * If we have less data than requested, block awaiting more 1733 * (subject to any timeout) if: 1734 * 1. the current count is less than the low water mark, or 1735 * 2. MSG_WAITALL is set, and it is possible to do the entire 1736 * receive operation at once if we block (resid <= hiwat). 1737 * 3. MSG_DONTWAIT is not set 1738 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1739 * we have to do the receive in sections, and thus risk returning 1740 * a short count if a timeout or signal occurs after we start. 1741 */ 1742 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1743 (size_t)so->so_rcv.ssb_cc < resid) && 1744 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat || 1745 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) { 1746 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1")); 1747 if (so->so_error) { 1748 if (m) 1749 goto dontblock; 1750 lwkt_reltoken(&so->so_rcv.ssb_token); 1751 error = so->so_error; 1752 if ((flags & MSG_PEEK) == 0) 1753 so->so_error = 0; 1754 goto release; 1755 } 1756 if (so->so_state & SS_CANTRCVMORE) { 1757 if (m) 1758 goto dontblock; 1759 lwkt_reltoken(&so->so_rcv.ssb_token); 1760 goto release; 1761 } 1762 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1763 (pr->pr_flags & PR_CONNREQUIRED)) { 1764 lwkt_reltoken(&so->so_rcv.ssb_token); 1765 error = ENOTCONN; 1766 goto release; 1767 } 1768 if (resid == 0) { 1769 lwkt_reltoken(&so->so_rcv.ssb_token); 1770 goto release; 1771 } 1772 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) { 1773 lwkt_reltoken(&so->so_rcv.ssb_token); 1774 error = EWOULDBLOCK; 1775 goto release; 1776 } 1777 ssb_unlock(&so->so_rcv); 1778 error = ssb_wait(&so->so_rcv); 1779 lwkt_reltoken(&so->so_rcv.ssb_token); 1780 if (error) 1781 goto done; 1782 goto restart; 1783 } 1784 1785 /* 1786 * Token still held 1787 */ 1788 dontblock: 1789 n = m; 1790 restmp = 0; 1791 while (n && restmp < resid) { 1792 n->m_flags |= M_SOLOCKED; 1793 restmp += n->m_len; 1794 if (n->m_next == NULL) 1795 n = n->m_nextpkt; 1796 else 1797 n = n->m_next; 1798 } 1799 1800 /* 1801 * Release token for loop 1802 */ 1803 lwkt_reltoken(&so->so_rcv.ssb_token); 1804 if (uio && uio->uio_td && uio->uio_td->td_proc) 1805 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++; 1806 1807 /* 1808 * note: m should be == sb_mb here. Cache the next record while 1809 * cleaning up. Note that calling m_free*() will break out critical 1810 * section. 1811 */ 1812 KKASSERT(m == so->so_rcv.ssb_mb); 1813 1814 /* 1815 * Copy to the UIO or mbuf return chain (*mp). 1816 * 1817 * NOTE: Token is not held for loop 1818 */ 1819 moff = 0; 1820 offset = 0; 1821 didoob = 0; 1822 1823 while (m && (m->m_flags & M_SOLOCKED) && resid > 0 && error == 0) { 1824 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER, 1825 ("receive 3")); 1826 1827 soclrstate(so, SS_RCVATMARK); 1828 len = (resid > INT_MAX) ? INT_MAX : resid; 1829 if (so->so_oobmark && len > so->so_oobmark - offset) 1830 len = so->so_oobmark - offset; 1831 if (len > m->m_len - moff) 1832 len = m->m_len - moff; 1833 1834 /* 1835 * Copy out to the UIO or pass the mbufs back to the SIO. 1836 * The SIO is dealt with when we eat the mbuf, but deal 1837 * with the resid here either way. 1838 */ 1839 if (uio) { 1840 uio->uio_resid = resid; 1841 error = uiomove(mtod(m, caddr_t) + moff, len, uio); 1842 resid = uio->uio_resid; 1843 if (error) 1844 goto release; 1845 } else { 1846 resid -= (size_t)len; 1847 } 1848 1849 /* 1850 * Eat the entire mbuf or just a piece of it 1851 */ 1852 offset += len; 1853 if (len == m->m_len - moff) { 1854 m = m->m_next; 1855 moff = 0; 1856 } else { 1857 moff += len; 1858 } 1859 1860 /* 1861 * Check oobmark 1862 */ 1863 if (so->so_oobmark && offset == so->so_oobmark) { 1864 didoob = 1; 1865 break; 1866 } 1867 } 1868 1869 /* 1870 * Synchronize sockbuf with data we read. 1871 * 1872 * NOTE: (m) is junk on entry (it could be left over from the 1873 * previous loop). 1874 */ 1875 if ((flags & MSG_PEEK) == 0) { 1876 lwkt_gettoken(&so->so_rcv.ssb_token); 1877 m = so->so_rcv.ssb_mb; 1878 while (m && offset >= m->m_len) { 1879 if (so->so_oobmark) { 1880 so->so_oobmark -= m->m_len; 1881 if (so->so_oobmark == 0) { 1882 sosetstate(so, SS_RCVATMARK); 1883 didoob = 1; 1884 } 1885 } 1886 offset -= m->m_len; 1887 if (sio) { 1888 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL); 1889 sbappend(sio, m); 1890 m = n; 1891 } else { 1892 m = sbunlinkmbuf(&so->so_rcv.sb, 1893 m, &free_chain); 1894 } 1895 } 1896 if (offset) { 1897 KKASSERT(m); 1898 if (sio) { 1899 n = m_copym(m, 0, offset, MB_WAIT); 1900 if (n) 1901 sbappend(sio, n); 1902 } 1903 m->m_data += offset; 1904 m->m_len -= offset; 1905 so->so_rcv.ssb_cc -= offset; 1906 if (so->so_oobmark) { 1907 so->so_oobmark -= offset; 1908 if (so->so_oobmark == 0) { 1909 sosetstate(so, SS_RCVATMARK); 1910 didoob = 1; 1911 } 1912 } 1913 offset = 0; 1914 } 1915 lwkt_reltoken(&so->so_rcv.ssb_token); 1916 } 1917 1918 /* 1919 * If the MSG_WAITALL flag is set (for non-atomic socket), 1920 * we must not quit until resid == 0 or an error termination. 1921 * 1922 * If a signal/timeout occurs, return with a short count but without 1923 * error. 1924 * 1925 * Keep signalsockbuf locked against other readers. 1926 * 1927 * XXX if MSG_PEEK we currently do quit. 1928 */ 1929 if ((flags & MSG_WAITALL) && !(flags & MSG_PEEK) && 1930 didoob == 0 && resid > 0 && 1931 !sosendallatonce(so)) { 1932 lwkt_gettoken(&so->so_rcv.ssb_token); 1933 error = 0; 1934 while ((m = so->so_rcv.ssb_mb) == NULL) { 1935 if (so->so_error || (so->so_state & SS_CANTRCVMORE)) { 1936 error = so->so_error; 1937 break; 1938 } 1939 /* 1940 * The window might have closed to zero, make 1941 * sure we send an ack now that we've drained 1942 * the buffer or we might end up blocking until 1943 * the idle takes over (5 seconds). 1944 */ 1945 if (so->so_pcb) 1946 so_pru_rcvd_async(so); 1947 if (so->so_rcv.ssb_mb == NULL) 1948 error = ssb_wait(&so->so_rcv); 1949 if (error) { 1950 lwkt_reltoken(&so->so_rcv.ssb_token); 1951 ssb_unlock(&so->so_rcv); 1952 error = 0; 1953 goto done; 1954 } 1955 } 1956 if (m && error == 0) 1957 goto dontblock; 1958 lwkt_reltoken(&so->so_rcv.ssb_token); 1959 } 1960 1961 /* 1962 * Token not held here. 1963 * 1964 * Cleanup. If an atomic read was requested drop any unread data XXX 1965 */ 1966 if ((flags & MSG_PEEK) == 0) { 1967 if (so->so_pcb) 1968 so_pru_rcvd_async(so); 1969 } 1970 1971 if (orig_resid == resid && orig_resid && 1972 (so->so_state & SS_CANTRCVMORE) == 0) { 1973 ssb_unlock(&so->so_rcv); 1974 goto restart; 1975 } 1976 1977 if (flagsp) 1978 *flagsp |= flags; 1979 release: 1980 ssb_unlock(&so->so_rcv); 1981 done: 1982 if (free_chain) 1983 m_freem(free_chain); 1984 return (error); 1985 } 1986 1987 /* 1988 * Shut a socket down. Note that we do not get a frontend lock as we 1989 * want to be able to shut the socket down even if another thread is 1990 * blocked in a read(), thus waking it up. 1991 */ 1992 int 1993 soshutdown(struct socket *so, int how) 1994 { 1995 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1996 return (EINVAL); 1997 1998 if (how != SHUT_WR) { 1999 /*ssb_lock(&so->so_rcv, M_WAITOK);*/ 2000 sorflush(so); 2001 /*ssb_unlock(&so->so_rcv);*/ 2002 } 2003 if (how != SHUT_RD) 2004 return (so_pru_shutdown(so)); 2005 return (0); 2006 } 2007 2008 void 2009 sorflush(struct socket *so) 2010 { 2011 struct signalsockbuf *ssb = &so->so_rcv; 2012 struct protosw *pr = so->so_proto; 2013 struct signalsockbuf asb; 2014 2015 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR); 2016 2017 lwkt_gettoken(&ssb->ssb_token); 2018 socantrcvmore(so); 2019 asb = *ssb; 2020 2021 /* 2022 * Can't just blow up the ssb structure here 2023 */ 2024 bzero(&ssb->sb, sizeof(ssb->sb)); 2025 ssb->ssb_timeo = 0; 2026 ssb->ssb_lowat = 0; 2027 ssb->ssb_hiwat = 0; 2028 ssb->ssb_mbmax = 0; 2029 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK); 2030 2031 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose) 2032 (*pr->pr_domain->dom_dispose)(asb.ssb_mb); 2033 ssb_release(&asb, so); 2034 2035 lwkt_reltoken(&ssb->ssb_token); 2036 } 2037 2038 #ifdef INET 2039 static int 2040 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt) 2041 { 2042 struct accept_filter_arg *afap = NULL; 2043 struct accept_filter *afp; 2044 struct so_accf *af = so->so_accf; 2045 int error = 0; 2046 2047 /* do not set/remove accept filters on non listen sockets */ 2048 if ((so->so_options & SO_ACCEPTCONN) == 0) { 2049 error = EINVAL; 2050 goto out; 2051 } 2052 2053 /* removing the filter */ 2054 if (sopt == NULL) { 2055 if (af != NULL) { 2056 if (af->so_accept_filter != NULL && 2057 af->so_accept_filter->accf_destroy != NULL) { 2058 af->so_accept_filter->accf_destroy(so); 2059 } 2060 if (af->so_accept_filter_str != NULL) { 2061 kfree(af->so_accept_filter_str, M_ACCF); 2062 } 2063 kfree(af, M_ACCF); 2064 so->so_accf = NULL; 2065 } 2066 so->so_options &= ~SO_ACCEPTFILTER; 2067 return (0); 2068 } 2069 /* adding a filter */ 2070 /* must remove previous filter first */ 2071 if (af != NULL) { 2072 error = EINVAL; 2073 goto out; 2074 } 2075 /* don't put large objects on the kernel stack */ 2076 afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK); 2077 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap); 2078 afap->af_name[sizeof(afap->af_name)-1] = '\0'; 2079 afap->af_arg[sizeof(afap->af_arg)-1] = '\0'; 2080 if (error) 2081 goto out; 2082 afp = accept_filt_get(afap->af_name); 2083 if (afp == NULL) { 2084 error = ENOENT; 2085 goto out; 2086 } 2087 af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO); 2088 if (afp->accf_create != NULL) { 2089 if (afap->af_name[0] != '\0') { 2090 int len = strlen(afap->af_name) + 1; 2091 2092 af->so_accept_filter_str = kmalloc(len, M_ACCF, 2093 M_WAITOK); 2094 strcpy(af->so_accept_filter_str, afap->af_name); 2095 } 2096 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg); 2097 if (af->so_accept_filter_arg == NULL) { 2098 kfree(af->so_accept_filter_str, M_ACCF); 2099 kfree(af, M_ACCF); 2100 so->so_accf = NULL; 2101 error = EINVAL; 2102 goto out; 2103 } 2104 } 2105 af->so_accept_filter = afp; 2106 so->so_accf = af; 2107 so->so_options |= SO_ACCEPTFILTER; 2108 out: 2109 if (afap != NULL) 2110 kfree(afap, M_TEMP); 2111 return (error); 2112 } 2113 #endif /* INET */ 2114 2115 /* 2116 * Perhaps this routine, and sooptcopyout(), below, ought to come in 2117 * an additional variant to handle the case where the option value needs 2118 * to be some kind of integer, but not a specific size. 2119 * In addition to their use here, these functions are also called by the 2120 * protocol-level pr_ctloutput() routines. 2121 */ 2122 int 2123 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2124 { 2125 return soopt_to_kbuf(sopt, buf, len, minlen); 2126 } 2127 2128 int 2129 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2130 { 2131 size_t valsize; 2132 2133 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 2134 KKASSERT(kva_p(buf)); 2135 2136 /* 2137 * If the user gives us more than we wanted, we ignore it, 2138 * but if we don't get the minimum length the caller 2139 * wants, we return EINVAL. On success, sopt->sopt_valsize 2140 * is set to however much we actually retrieved. 2141 */ 2142 if ((valsize = sopt->sopt_valsize) < minlen) 2143 return EINVAL; 2144 if (valsize > len) 2145 sopt->sopt_valsize = valsize = len; 2146 2147 bcopy(sopt->sopt_val, buf, valsize); 2148 return 0; 2149 } 2150 2151 2152 int 2153 sosetopt(struct socket *so, struct sockopt *sopt) 2154 { 2155 int error, optval; 2156 struct linger l; 2157 struct timeval tv; 2158 u_long val; 2159 struct signalsockbuf *sotmp; 2160 2161 error = 0; 2162 sopt->sopt_dir = SOPT_SET; 2163 if (sopt->sopt_level != SOL_SOCKET) { 2164 if (so->so_proto && so->so_proto->pr_ctloutput) { 2165 return (so_pr_ctloutput(so, sopt)); 2166 } 2167 error = ENOPROTOOPT; 2168 } else { 2169 switch (sopt->sopt_name) { 2170 #ifdef INET 2171 case SO_ACCEPTFILTER: 2172 error = do_setopt_accept_filter(so, sopt); 2173 if (error) 2174 goto bad; 2175 break; 2176 #endif /* INET */ 2177 case SO_LINGER: 2178 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 2179 if (error) 2180 goto bad; 2181 2182 so->so_linger = l.l_linger; 2183 if (l.l_onoff) 2184 so->so_options |= SO_LINGER; 2185 else 2186 so->so_options &= ~SO_LINGER; 2187 break; 2188 2189 case SO_DEBUG: 2190 case SO_KEEPALIVE: 2191 case SO_DONTROUTE: 2192 case SO_USELOOPBACK: 2193 case SO_BROADCAST: 2194 case SO_REUSEADDR: 2195 case SO_REUSEPORT: 2196 case SO_OOBINLINE: 2197 case SO_TIMESTAMP: 2198 case SO_NOSIGPIPE: 2199 error = sooptcopyin(sopt, &optval, sizeof optval, 2200 sizeof optval); 2201 if (error) 2202 goto bad; 2203 if (optval) 2204 so->so_options |= sopt->sopt_name; 2205 else 2206 so->so_options &= ~sopt->sopt_name; 2207 break; 2208 2209 case SO_SNDBUF: 2210 case SO_RCVBUF: 2211 case SO_SNDLOWAT: 2212 case SO_RCVLOWAT: 2213 error = sooptcopyin(sopt, &optval, sizeof optval, 2214 sizeof optval); 2215 if (error) 2216 goto bad; 2217 2218 /* 2219 * Values < 1 make no sense for any of these 2220 * options, so disallow them. 2221 */ 2222 if (optval < 1) { 2223 error = EINVAL; 2224 goto bad; 2225 } 2226 2227 switch (sopt->sopt_name) { 2228 case SO_SNDBUF: 2229 case SO_RCVBUF: 2230 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ? 2231 &so->so_snd : &so->so_rcv, (u_long)optval, 2232 so, 2233 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) { 2234 error = ENOBUFS; 2235 goto bad; 2236 } 2237 sotmp = (sopt->sopt_name == SO_SNDBUF) ? 2238 &so->so_snd : &so->so_rcv; 2239 atomic_clear_int(&sotmp->ssb_flags, 2240 SSB_AUTOSIZE); 2241 break; 2242 2243 /* 2244 * Make sure the low-water is never greater than 2245 * the high-water. 2246 */ 2247 case SO_SNDLOWAT: 2248 so->so_snd.ssb_lowat = 2249 (optval > so->so_snd.ssb_hiwat) ? 2250 so->so_snd.ssb_hiwat : optval; 2251 atomic_clear_int(&so->so_snd.ssb_flags, 2252 SSB_AUTOLOWAT); 2253 break; 2254 case SO_RCVLOWAT: 2255 so->so_rcv.ssb_lowat = 2256 (optval > so->so_rcv.ssb_hiwat) ? 2257 so->so_rcv.ssb_hiwat : optval; 2258 atomic_clear_int(&so->so_rcv.ssb_flags, 2259 SSB_AUTOLOWAT); 2260 break; 2261 } 2262 break; 2263 2264 case SO_SNDTIMEO: 2265 case SO_RCVTIMEO: 2266 error = sooptcopyin(sopt, &tv, sizeof tv, 2267 sizeof tv); 2268 if (error) 2269 goto bad; 2270 2271 /* assert(hz > 0); */ 2272 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz || 2273 tv.tv_usec < 0 || tv.tv_usec >= 1000000) { 2274 error = EDOM; 2275 goto bad; 2276 } 2277 /* assert(tick > 0); */ 2278 /* assert(ULONG_MAX - INT_MAX >= 1000000); */ 2279 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick; 2280 if (val > INT_MAX) { 2281 error = EDOM; 2282 goto bad; 2283 } 2284 if (val == 0 && tv.tv_usec != 0) 2285 val = 1; 2286 2287 switch (sopt->sopt_name) { 2288 case SO_SNDTIMEO: 2289 so->so_snd.ssb_timeo = val; 2290 break; 2291 case SO_RCVTIMEO: 2292 so->so_rcv.ssb_timeo = val; 2293 break; 2294 } 2295 break; 2296 default: 2297 error = ENOPROTOOPT; 2298 break; 2299 } 2300 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 2301 (void) so_pr_ctloutput(so, sopt); 2302 } 2303 } 2304 bad: 2305 return (error); 2306 } 2307 2308 /* Helper routine for getsockopt */ 2309 int 2310 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 2311 { 2312 soopt_from_kbuf(sopt, buf, len); 2313 return 0; 2314 } 2315 2316 void 2317 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len) 2318 { 2319 size_t valsize; 2320 2321 if (len == 0) { 2322 sopt->sopt_valsize = 0; 2323 return; 2324 } 2325 2326 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 2327 KKASSERT(kva_p(buf)); 2328 2329 /* 2330 * Documented get behavior is that we always return a value, 2331 * possibly truncated to fit in the user's buffer. 2332 * Traditional behavior is that we always tell the user 2333 * precisely how much we copied, rather than something useful 2334 * like the total amount we had available for her. 2335 * Note that this interface is not idempotent; the entire answer must 2336 * generated ahead of time. 2337 */ 2338 valsize = szmin(len, sopt->sopt_valsize); 2339 sopt->sopt_valsize = valsize; 2340 if (sopt->sopt_val != 0) { 2341 bcopy(buf, sopt->sopt_val, valsize); 2342 } 2343 } 2344 2345 int 2346 sogetopt(struct socket *so, struct sockopt *sopt) 2347 { 2348 int error, optval; 2349 long optval_l; 2350 struct linger l; 2351 struct timeval tv; 2352 #ifdef INET 2353 struct accept_filter_arg *afap; 2354 #endif 2355 2356 error = 0; 2357 sopt->sopt_dir = SOPT_GET; 2358 if (sopt->sopt_level != SOL_SOCKET) { 2359 if (so->so_proto && so->so_proto->pr_ctloutput) { 2360 return (so_pr_ctloutput(so, sopt)); 2361 } else 2362 return (ENOPROTOOPT); 2363 } else { 2364 switch (sopt->sopt_name) { 2365 #ifdef INET 2366 case SO_ACCEPTFILTER: 2367 if ((so->so_options & SO_ACCEPTCONN) == 0) 2368 return (EINVAL); 2369 afap = kmalloc(sizeof(*afap), M_TEMP, 2370 M_WAITOK | M_ZERO); 2371 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 2372 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name); 2373 if (so->so_accf->so_accept_filter_str != NULL) 2374 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str); 2375 } 2376 error = sooptcopyout(sopt, afap, sizeof(*afap)); 2377 kfree(afap, M_TEMP); 2378 break; 2379 #endif /* INET */ 2380 2381 case SO_LINGER: 2382 l.l_onoff = so->so_options & SO_LINGER; 2383 l.l_linger = so->so_linger; 2384 error = sooptcopyout(sopt, &l, sizeof l); 2385 break; 2386 2387 case SO_USELOOPBACK: 2388 case SO_DONTROUTE: 2389 case SO_DEBUG: 2390 case SO_KEEPALIVE: 2391 case SO_REUSEADDR: 2392 case SO_REUSEPORT: 2393 case SO_BROADCAST: 2394 case SO_OOBINLINE: 2395 case SO_TIMESTAMP: 2396 case SO_NOSIGPIPE: 2397 optval = so->so_options & sopt->sopt_name; 2398 integer: 2399 error = sooptcopyout(sopt, &optval, sizeof optval); 2400 break; 2401 2402 case SO_TYPE: 2403 optval = so->so_type; 2404 goto integer; 2405 2406 case SO_ERROR: 2407 optval = so->so_error; 2408 so->so_error = 0; 2409 goto integer; 2410 2411 case SO_SNDBUF: 2412 optval = so->so_snd.ssb_hiwat; 2413 goto integer; 2414 2415 case SO_RCVBUF: 2416 optval = so->so_rcv.ssb_hiwat; 2417 goto integer; 2418 2419 case SO_SNDLOWAT: 2420 optval = so->so_snd.ssb_lowat; 2421 goto integer; 2422 2423 case SO_RCVLOWAT: 2424 optval = so->so_rcv.ssb_lowat; 2425 goto integer; 2426 2427 case SO_SNDTIMEO: 2428 case SO_RCVTIMEO: 2429 optval = (sopt->sopt_name == SO_SNDTIMEO ? 2430 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo); 2431 2432 tv.tv_sec = optval / hz; 2433 tv.tv_usec = (optval % hz) * ustick; 2434 error = sooptcopyout(sopt, &tv, sizeof tv); 2435 break; 2436 2437 case SO_SNDSPACE: 2438 optval_l = ssb_space(&so->so_snd); 2439 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l)); 2440 break; 2441 2442 case SO_CPUHINT: 2443 optval = -1; /* no hint */ 2444 goto integer; 2445 2446 default: 2447 error = ENOPROTOOPT; 2448 break; 2449 } 2450 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) 2451 so_pr_ctloutput(so, sopt); 2452 return (error); 2453 } 2454 } 2455 2456 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */ 2457 int 2458 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 2459 { 2460 struct mbuf *m, *m_prev; 2461 int sopt_size = sopt->sopt_valsize, msize; 2462 2463 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA, 2464 0, &msize); 2465 if (m == NULL) 2466 return (ENOBUFS); 2467 m->m_len = min(msize, sopt_size); 2468 sopt_size -= m->m_len; 2469 *mp = m; 2470 m_prev = m; 2471 2472 while (sopt_size > 0) { 2473 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, 2474 MT_DATA, 0, &msize); 2475 if (m == NULL) { 2476 m_freem(*mp); 2477 return (ENOBUFS); 2478 } 2479 m->m_len = min(msize, sopt_size); 2480 sopt_size -= m->m_len; 2481 m_prev->m_next = m; 2482 m_prev = m; 2483 } 2484 return (0); 2485 } 2486 2487 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */ 2488 int 2489 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 2490 { 2491 soopt_to_mbuf(sopt, m); 2492 return 0; 2493 } 2494 2495 void 2496 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m) 2497 { 2498 size_t valsize; 2499 void *val; 2500 2501 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 2502 KKASSERT(kva_p(m)); 2503 if (sopt->sopt_val == NULL) 2504 return; 2505 val = sopt->sopt_val; 2506 valsize = sopt->sopt_valsize; 2507 while (m != NULL && valsize >= m->m_len) { 2508 bcopy(val, mtod(m, char *), m->m_len); 2509 valsize -= m->m_len; 2510 val = (caddr_t)val + m->m_len; 2511 m = m->m_next; 2512 } 2513 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 2514 panic("ip6_sooptmcopyin"); 2515 } 2516 2517 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */ 2518 int 2519 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 2520 { 2521 return soopt_from_mbuf(sopt, m); 2522 } 2523 2524 int 2525 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m) 2526 { 2527 struct mbuf *m0 = m; 2528 size_t valsize = 0; 2529 size_t maxsize; 2530 void *val; 2531 2532 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 2533 KKASSERT(kva_p(m)); 2534 if (sopt->sopt_val == NULL) 2535 return 0; 2536 val = sopt->sopt_val; 2537 maxsize = sopt->sopt_valsize; 2538 while (m != NULL && maxsize >= m->m_len) { 2539 bcopy(mtod(m, char *), val, m->m_len); 2540 maxsize -= m->m_len; 2541 val = (caddr_t)val + m->m_len; 2542 valsize += m->m_len; 2543 m = m->m_next; 2544 } 2545 if (m != NULL) { 2546 /* enough soopt buffer should be given from user-land */ 2547 m_freem(m0); 2548 return (EINVAL); 2549 } 2550 sopt->sopt_valsize = valsize; 2551 return 0; 2552 } 2553 2554 void 2555 sohasoutofband(struct socket *so) 2556 { 2557 if (so->so_sigio != NULL) 2558 pgsigio(so->so_sigio, SIGURG, 0); 2559 KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB); 2560 } 2561 2562 int 2563 sokqfilter(struct file *fp, struct knote *kn) 2564 { 2565 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2566 struct signalsockbuf *ssb; 2567 2568 switch (kn->kn_filter) { 2569 case EVFILT_READ: 2570 if (so->so_options & SO_ACCEPTCONN) 2571 kn->kn_fop = &solisten_filtops; 2572 else 2573 kn->kn_fop = &soread_filtops; 2574 ssb = &so->so_rcv; 2575 break; 2576 case EVFILT_WRITE: 2577 kn->kn_fop = &sowrite_filtops; 2578 ssb = &so->so_snd; 2579 break; 2580 case EVFILT_EXCEPT: 2581 kn->kn_fop = &soexcept_filtops; 2582 ssb = &so->so_rcv; 2583 break; 2584 default: 2585 return (EOPNOTSUPP); 2586 } 2587 2588 knote_insert(&ssb->ssb_kq.ki_note, kn); 2589 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE); 2590 return (0); 2591 } 2592 2593 static void 2594 filt_sordetach(struct knote *kn) 2595 { 2596 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2597 2598 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn); 2599 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note)) 2600 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE); 2601 } 2602 2603 /*ARGSUSED*/ 2604 static int 2605 filt_soread(struct knote *kn, long hint) 2606 { 2607 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2608 2609 if (kn->kn_sfflags & NOTE_OOB) { 2610 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) { 2611 kn->kn_fflags |= NOTE_OOB; 2612 return (1); 2613 } 2614 return (0); 2615 } 2616 kn->kn_data = so->so_rcv.ssb_cc; 2617 2618 if (so->so_state & SS_CANTRCVMORE) { 2619 /* 2620 * Only set NODATA if all data has been exhausted. 2621 */ 2622 if (kn->kn_data == 0) 2623 kn->kn_flags |= EV_NODATA; 2624 kn->kn_flags |= EV_EOF; 2625 kn->kn_fflags = so->so_error; 2626 return (1); 2627 } 2628 if (so->so_error) /* temporary udp error */ 2629 return (1); 2630 if (kn->kn_sfflags & NOTE_LOWAT) 2631 return (kn->kn_data >= kn->kn_sdata); 2632 return ((kn->kn_data >= so->so_rcv.ssb_lowat) || 2633 !TAILQ_EMPTY(&so->so_comp)); 2634 } 2635 2636 static void 2637 filt_sowdetach(struct knote *kn) 2638 { 2639 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2640 2641 knote_remove(&so->so_snd.ssb_kq.ki_note, kn); 2642 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note)) 2643 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE); 2644 } 2645 2646 /*ARGSUSED*/ 2647 static int 2648 filt_sowrite(struct knote *kn, long hint) 2649 { 2650 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2651 2652 kn->kn_data = ssb_space(&so->so_snd); 2653 if (so->so_state & SS_CANTSENDMORE) { 2654 kn->kn_flags |= (EV_EOF | EV_NODATA); 2655 kn->kn_fflags = so->so_error; 2656 return (1); 2657 } 2658 if (so->so_error) /* temporary udp error */ 2659 return (1); 2660 if (((so->so_state & SS_ISCONNECTED) == 0) && 2661 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 2662 return (0); 2663 if (kn->kn_sfflags & NOTE_LOWAT) 2664 return (kn->kn_data >= kn->kn_sdata); 2665 return (kn->kn_data >= so->so_snd.ssb_lowat); 2666 } 2667 2668 /*ARGSUSED*/ 2669 static int 2670 filt_solisten(struct knote *kn, long hint) 2671 { 2672 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2673 2674 kn->kn_data = so->so_qlen; 2675 return (! TAILQ_EMPTY(&so->so_comp)); 2676 } 2677