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