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 * $DragonFly: src/sys/kern/uipc_socket.c,v 1.55 2008/09/02 16:17:52 dillon Exp $ 69 */ 70 71 #include "opt_inet.h" 72 #include "opt_sctp.h" 73 74 #include <sys/param.h> 75 #include <sys/systm.h> 76 #include <sys/fcntl.h> 77 #include <sys/malloc.h> 78 #include <sys/mbuf.h> 79 #include <sys/domain.h> 80 #include <sys/file.h> /* for struct knote */ 81 #include <sys/kernel.h> 82 #include <sys/malloc.h> 83 #include <sys/event.h> 84 #include <sys/poll.h> 85 #include <sys/proc.h> 86 #include <sys/protosw.h> 87 #include <sys/socket.h> 88 #include <sys/socketvar.h> 89 #include <sys/socketops.h> 90 #include <sys/resourcevar.h> 91 #include <sys/signalvar.h> 92 #include <sys/sysctl.h> 93 #include <sys/uio.h> 94 #include <sys/jail.h> 95 #include <vm/vm_zone.h> 96 #include <vm/pmap.h> 97 98 #include <sys/thread2.h> 99 #include <sys/socketvar2.h> 100 101 #include <machine/limits.h> 102 103 #ifdef INET 104 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt); 105 #endif /* INET */ 106 107 static void filt_sordetach(struct knote *kn); 108 static int filt_soread(struct knote *kn, long hint); 109 static void filt_sowdetach(struct knote *kn); 110 static int filt_sowrite(struct knote *kn, long hint); 111 static int filt_solisten(struct knote *kn, long hint); 112 113 static struct filterops solisten_filtops = 114 { 1, NULL, filt_sordetach, filt_solisten }; 115 static struct filterops soread_filtops = 116 { 1, NULL, filt_sordetach, filt_soread }; 117 static struct filterops sowrite_filtops = 118 { 1, NULL, filt_sowdetach, filt_sowrite }; 119 120 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct"); 121 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 122 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 123 124 125 static int somaxconn = SOMAXCONN; 126 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW, 127 &somaxconn, 0, "Maximum pending socket connection queue size"); 128 129 /* 130 * Socket operation routines. 131 * These routines are called by the routines in 132 * sys_socket.c or from a system process, and 133 * implement the semantics of socket operations by 134 * switching out to the protocol specific routines. 135 */ 136 137 /* 138 * Get a socket structure, and initialize it. 139 * Note that it would probably be better to allocate socket 140 * and PCB at the same time, but I'm not convinced that all 141 * the protocols can be easily modified to do this. 142 */ 143 struct socket * 144 soalloc(int waitok) 145 { 146 struct socket *so; 147 unsigned waitmask; 148 149 waitmask = waitok ? M_WAITOK : M_NOWAIT; 150 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask); 151 if (so) { 152 /* XXX race condition for reentrant kernel */ 153 TAILQ_INIT(&so->so_aiojobq); 154 TAILQ_INIT(&so->so_rcv.ssb_sel.si_mlist); 155 TAILQ_INIT(&so->so_snd.ssb_sel.si_mlist); 156 } 157 return so; 158 } 159 160 int 161 socreate(int dom, struct socket **aso, int type, 162 int proto, struct thread *td) 163 { 164 struct proc *p = td->td_proc; 165 struct protosw *prp; 166 struct socket *so; 167 struct pru_attach_info ai; 168 int error; 169 170 if (proto) 171 prp = pffindproto(dom, proto, type); 172 else 173 prp = pffindtype(dom, type); 174 175 if (prp == 0 || prp->pr_usrreqs->pru_attach == 0) 176 return (EPROTONOSUPPORT); 177 178 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only && 179 prp->pr_domain->dom_family != PF_LOCAL && 180 prp->pr_domain->dom_family != PF_INET && 181 prp->pr_domain->dom_family != PF_INET6 && 182 prp->pr_domain->dom_family != PF_ROUTE) { 183 return (EPROTONOSUPPORT); 184 } 185 186 if (prp->pr_type != type) 187 return (EPROTOTYPE); 188 so = soalloc(p != 0); 189 if (so == 0) 190 return (ENOBUFS); 191 192 TAILQ_INIT(&so->so_incomp); 193 TAILQ_INIT(&so->so_comp); 194 so->so_type = type; 195 so->so_cred = crhold(p->p_ucred); 196 so->so_proto = prp; 197 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE]; 198 ai.p_ucred = p->p_ucred; 199 ai.fd_rdir = p->p_fd->fd_rdir; 200 error = so_pru_attach(so, proto, &ai); 201 if (error) { 202 so->so_state |= SS_NOFDREF; 203 sofree(so); 204 return (error); 205 } 206 *aso = so; 207 return (0); 208 } 209 210 int 211 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 212 { 213 int error; 214 215 crit_enter(); 216 error = so_pru_bind(so, nam, td); 217 crit_exit(); 218 return (error); 219 } 220 221 void 222 sodealloc(struct socket *so) 223 { 224 if (so->so_rcv.ssb_hiwat) 225 (void)chgsbsize(so->so_cred->cr_uidinfo, 226 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY); 227 if (so->so_snd.ssb_hiwat) 228 (void)chgsbsize(so->so_cred->cr_uidinfo, 229 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY); 230 #ifdef INET 231 /* remove accept filter if present */ 232 if (so->so_accf != NULL) 233 do_setopt_accept_filter(so, NULL); 234 #endif /* INET */ 235 crfree(so->so_cred); 236 kfree(so, M_SOCKET); 237 } 238 239 int 240 solisten(struct socket *so, int backlog, struct thread *td) 241 { 242 int error; 243 #ifdef SCTP 244 short oldopt, oldqlimit; 245 #endif /* SCTP */ 246 247 crit_enter(); 248 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) { 249 crit_exit(); 250 return (EINVAL); 251 } 252 253 #ifdef SCTP 254 oldopt = so->so_options; 255 oldqlimit = so->so_qlimit; 256 #endif /* SCTP */ 257 258 if (TAILQ_EMPTY(&so->so_comp)) 259 so->so_options |= SO_ACCEPTCONN; 260 if (backlog < 0 || backlog > somaxconn) 261 backlog = somaxconn; 262 so->so_qlimit = backlog; 263 /* SCTP needs to look at tweak both the inbound backlog parameter AND 264 * the so_options (UDP model both connect's and gets inbound 265 * connections .. implicitly). 266 */ 267 error = so_pru_listen(so, td); 268 if (error) { 269 #ifdef SCTP 270 /* Restore the params */ 271 so->so_options = oldopt; 272 so->so_qlimit = oldqlimit; 273 #endif /* SCTP */ 274 crit_exit(); 275 return (error); 276 } 277 crit_exit(); 278 return (0); 279 } 280 281 /* 282 * Destroy a disconnected socket. This routine is a NOP if entities 283 * still have a reference on the socket: 284 * 285 * so_pcb - The protocol stack still has a reference 286 * SS_NOFDREF - There is no longer a file pointer reference 287 * SS_ABORTING - An abort netmsg is in-flight 288 */ 289 void 290 sofree(struct socket *so) 291 { 292 struct socket *head = so->so_head; 293 294 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 295 return; 296 if (so->so_state & SS_ABORTING) 297 return; 298 if (head != NULL) { 299 if (so->so_state & SS_INCOMP) { 300 TAILQ_REMOVE(&head->so_incomp, so, so_list); 301 head->so_incqlen--; 302 } else if (so->so_state & SS_COMP) { 303 /* 304 * We must not decommission a socket that's 305 * on the accept(2) queue. If we do, then 306 * accept(2) may hang after select(2) indicated 307 * that the listening socket was ready. 308 */ 309 return; 310 } else { 311 panic("sofree: not queued"); 312 } 313 so->so_state &= ~SS_INCOMP; 314 so->so_head = NULL; 315 } 316 ssb_release(&so->so_snd, so); 317 sorflush(so); 318 sodealloc(so); 319 } 320 321 /* 322 * Close a socket on last file table reference removal. 323 * Initiate disconnect if connected. 324 * Free socket when disconnect complete. 325 */ 326 int 327 soclose(struct socket *so, int fflag) 328 { 329 int error = 0; 330 331 crit_enter(); 332 funsetown(so->so_sigio); 333 if (so->so_pcb == NULL) 334 goto discard; 335 if (so->so_state & SS_ISCONNECTED) { 336 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 337 error = sodisconnect(so); 338 if (error) 339 goto drop; 340 } 341 if (so->so_options & SO_LINGER) { 342 if ((so->so_state & SS_ISDISCONNECTING) && 343 (fflag & FNONBLOCK)) 344 goto drop; 345 while (so->so_state & SS_ISCONNECTED) { 346 error = tsleep((caddr_t)&so->so_timeo, 347 PCATCH, "soclos", so->so_linger * hz); 348 if (error) 349 break; 350 } 351 } 352 } 353 drop: 354 if (so->so_pcb) { 355 int error2; 356 357 error2 = so_pru_detach(so); 358 if (error == 0) 359 error = error2; 360 } 361 discard: 362 if (so->so_options & SO_ACCEPTCONN) { 363 struct socket *sp; 364 365 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) { 366 TAILQ_REMOVE(&so->so_incomp, sp, so_list); 367 sp->so_state &= ~SS_INCOMP; 368 sp->so_head = NULL; 369 so->so_incqlen--; 370 soaborta(sp); 371 } 372 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) { 373 TAILQ_REMOVE(&so->so_comp, sp, so_list); 374 sp->so_state &= ~SS_COMP; 375 sp->so_head = NULL; 376 so->so_qlen--; 377 soaborta(sp); 378 } 379 } 380 if (so->so_state & SS_NOFDREF) 381 panic("soclose: NOFDREF"); 382 so->so_state |= SS_NOFDREF; 383 sofree(so); 384 crit_exit(); 385 return (error); 386 } 387 388 /* 389 * Abort and destroy a socket. Only one abort can be in progress 390 * at any given moment. 391 */ 392 void 393 soabort(struct socket *so) 394 { 395 if ((so->so_state & SS_ABORTING) == 0) { 396 so->so_state |= SS_ABORTING; 397 so_pru_abort(so); 398 } 399 } 400 401 void 402 soaborta(struct socket *so) 403 { 404 if ((so->so_state & SS_ABORTING) == 0) { 405 so->so_state |= SS_ABORTING; 406 so_pru_aborta(so); 407 } 408 } 409 410 int 411 soaccept(struct socket *so, struct sockaddr **nam) 412 { 413 int error; 414 415 crit_enter(); 416 if ((so->so_state & SS_NOFDREF) == 0) 417 panic("soaccept: !NOFDREF"); 418 so->so_state &= ~SS_NOFDREF; 419 error = so_pru_accept(so, nam); 420 crit_exit(); 421 return (error); 422 } 423 424 int 425 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 426 { 427 int error; 428 429 if (so->so_options & SO_ACCEPTCONN) 430 return (EOPNOTSUPP); 431 crit_enter(); 432 /* 433 * If protocol is connection-based, can only connect once. 434 * Otherwise, if connected, try to disconnect first. 435 * This allows user to disconnect by connecting to, e.g., 436 * a null address. 437 */ 438 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 439 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 440 (error = sodisconnect(so)))) { 441 error = EISCONN; 442 } else { 443 /* 444 * Prevent accumulated error from previous connection 445 * from biting us. 446 */ 447 so->so_error = 0; 448 error = so_pru_connect(so, nam, td); 449 } 450 crit_exit(); 451 return (error); 452 } 453 454 int 455 soconnect2(struct socket *so1, struct socket *so2) 456 { 457 int error; 458 459 crit_enter(); 460 error = so_pru_connect2(so1, so2); 461 crit_exit(); 462 return (error); 463 } 464 465 int 466 sodisconnect(struct socket *so) 467 { 468 int error; 469 470 crit_enter(); 471 if ((so->so_state & SS_ISCONNECTED) == 0) { 472 error = ENOTCONN; 473 goto bad; 474 } 475 if (so->so_state & SS_ISDISCONNECTING) { 476 error = EALREADY; 477 goto bad; 478 } 479 error = so_pru_disconnect(so); 480 bad: 481 crit_exit(); 482 return (error); 483 } 484 485 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 486 /* 487 * Send on a socket. 488 * If send must go all at once and message is larger than 489 * send buffering, then hard error. 490 * Lock against other senders. 491 * If must go all at once and not enough room now, then 492 * inform user that this would block and do nothing. 493 * Otherwise, if nonblocking, send as much as possible. 494 * The data to be sent is described by "uio" if nonzero, 495 * otherwise by the mbuf chain "top" (which must be null 496 * if uio is not). Data provided in mbuf chain must be small 497 * enough to send all at once. 498 * 499 * Returns nonzero on error, timeout or signal; callers 500 * must check for short counts if EINTR/ERESTART are returned. 501 * Data and control buffers are freed on return. 502 */ 503 int 504 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 505 struct mbuf *top, struct mbuf *control, int flags, 506 struct thread *td) 507 { 508 struct mbuf **mp; 509 struct mbuf *m; 510 long space, len, resid; 511 int clen = 0, error, dontroute, mlen; 512 int atomic = sosendallatonce(so) || top; 513 int pru_flags; 514 515 if (uio) 516 resid = uio->uio_resid; 517 else 518 resid = top->m_pkthdr.len; 519 /* 520 * In theory resid should be unsigned. 521 * However, space must be signed, as it might be less than 0 522 * if we over-committed, and we must use a signed comparison 523 * of space and resid. On the other hand, a negative resid 524 * causes us to loop sending 0-length segments to the protocol. 525 * 526 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 527 * type sockets since that's an error. 528 */ 529 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 530 error = EINVAL; 531 goto out; 532 } 533 534 dontroute = 535 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 536 (so->so_proto->pr_flags & PR_ATOMIC); 537 if (td->td_lwp != NULL) 538 td->td_lwp->lwp_ru.ru_msgsnd++; 539 if (control) 540 clen = control->m_len; 541 #define gotoerr(errcode) { error = errcode; crit_exit(); goto release; } 542 543 restart: 544 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags)); 545 if (error) 546 goto out; 547 do { 548 crit_enter(); 549 if (so->so_state & SS_CANTSENDMORE) 550 gotoerr(EPIPE); 551 if (so->so_error) { 552 error = so->so_error; 553 so->so_error = 0; 554 crit_exit(); 555 goto release; 556 } 557 if ((so->so_state & SS_ISCONNECTED) == 0) { 558 /* 559 * `sendto' and `sendmsg' is allowed on a connection- 560 * based socket if it supports implied connect. 561 * Return ENOTCONN if not connected and no address is 562 * supplied. 563 */ 564 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 565 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 566 if ((so->so_state & SS_ISCONFIRMING) == 0 && 567 !(resid == 0 && clen != 0)) 568 gotoerr(ENOTCONN); 569 } else if (addr == 0) 570 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ? 571 ENOTCONN : EDESTADDRREQ); 572 } 573 if ((atomic && resid > so->so_snd.ssb_hiwat) || 574 clen > so->so_snd.ssb_hiwat) { 575 gotoerr(EMSGSIZE); 576 } 577 space = ssb_space(&so->so_snd); 578 if (flags & MSG_OOB) 579 space += 1024; 580 if (space < resid + clen && uio && 581 (atomic || space < so->so_snd.ssb_lowat || space < clen)) { 582 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) 583 gotoerr(EWOULDBLOCK); 584 ssb_unlock(&so->so_snd); 585 error = ssb_wait(&so->so_snd); 586 crit_exit(); 587 if (error) 588 goto out; 589 goto restart; 590 } 591 crit_exit(); 592 mp = ⊤ 593 space -= clen; 594 do { 595 if (uio == NULL) { 596 /* 597 * Data is prepackaged in "top". 598 */ 599 resid = 0; 600 if (flags & MSG_EOR) 601 top->m_flags |= M_EOR; 602 } else do { 603 m = m_getl(resid, MB_WAIT, MT_DATA, 604 top == NULL ? M_PKTHDR : 0, &mlen); 605 if (top == NULL) { 606 m->m_pkthdr.len = 0; 607 m->m_pkthdr.rcvif = (struct ifnet *)0; 608 } 609 len = min(min(mlen, resid), space); 610 if (resid < MINCLSIZE) { 611 /* 612 * For datagram protocols, leave room 613 * for protocol headers in first mbuf. 614 */ 615 if (atomic && top == 0 && len < mlen) 616 MH_ALIGN(m, len); 617 } 618 space -= len; 619 error = uiomove(mtod(m, caddr_t), (int)len, uio); 620 resid = uio->uio_resid; 621 m->m_len = len; 622 *mp = m; 623 top->m_pkthdr.len += len; 624 if (error) 625 goto release; 626 mp = &m->m_next; 627 if (resid <= 0) { 628 if (flags & MSG_EOR) 629 top->m_flags |= M_EOR; 630 break; 631 } 632 } while (space > 0 && atomic); 633 if (dontroute) 634 so->so_options |= SO_DONTROUTE; 635 if (flags & MSG_OOB) { 636 pru_flags = PRUS_OOB; 637 } else if ((flags & MSG_EOF) && 638 (so->so_proto->pr_flags & PR_IMPLOPCL) && 639 (resid <= 0)) { 640 /* 641 * If the user set MSG_EOF, the protocol 642 * understands this flag and nothing left to 643 * send then use PRU_SEND_EOF instead of PRU_SEND. 644 */ 645 pru_flags = PRUS_EOF; 646 } else if (resid > 0 && space > 0) { 647 /* If there is more to send, set PRUS_MORETOCOME */ 648 pru_flags = PRUS_MORETOCOME; 649 } else { 650 pru_flags = 0; 651 } 652 crit_enter(); 653 /* 654 * XXX all the SS_CANTSENDMORE checks previously 655 * done could be out of date. We could have recieved 656 * a reset packet in an interrupt or maybe we slept 657 * while doing page faults in uiomove() etc. We could 658 * probably recheck again inside the splnet() protection 659 * here, but there are probably other places that this 660 * also happens. We must rethink this. 661 */ 662 error = so_pru_send(so, pru_flags, top, addr, control, td); 663 crit_exit(); 664 if (dontroute) 665 so->so_options &= ~SO_DONTROUTE; 666 clen = 0; 667 control = 0; 668 top = 0; 669 mp = ⊤ 670 if (error) 671 goto release; 672 } while (resid && space > 0); 673 } while (resid); 674 675 release: 676 ssb_unlock(&so->so_snd); 677 out: 678 if (top) 679 m_freem(top); 680 if (control) 681 m_freem(control); 682 return (error); 683 } 684 685 /* 686 * A specialization of sosend() for UDP based on protocol-specific knowledge: 687 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that 688 * sosendallatonce() returns true, 689 * the "atomic" variable is true, 690 * and sosendudp() blocks until space is available for the entire send. 691 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or 692 * PR_IMPLOPCL flags set. 693 * UDP has no out-of-band data. 694 * UDP has no control data. 695 * UDP does not support MSG_EOR. 696 */ 697 int 698 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio, 699 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 700 { 701 int resid, error; 702 boolean_t dontroute; /* temporary SO_DONTROUTE setting */ 703 704 if (td->td_lwp != NULL) 705 td->td_lwp->lwp_ru.ru_msgsnd++; 706 if (control) 707 m_freem(control); 708 709 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp")); 710 resid = uio ? uio->uio_resid : top->m_pkthdr.len; 711 712 restart: 713 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags)); 714 if (error) 715 goto out; 716 717 crit_enter(); 718 if (so->so_state & SS_CANTSENDMORE) 719 gotoerr(EPIPE); 720 if (so->so_error) { 721 error = so->so_error; 722 so->so_error = 0; 723 crit_exit(); 724 goto release; 725 } 726 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL) 727 gotoerr(EDESTADDRREQ); 728 if (resid > so->so_snd.ssb_hiwat) 729 gotoerr(EMSGSIZE); 730 if (uio && ssb_space(&so->so_snd) < resid) { 731 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) 732 gotoerr(EWOULDBLOCK); 733 ssb_unlock(&so->so_snd); 734 error = ssb_wait(&so->so_snd); 735 crit_exit(); 736 if (error) 737 goto out; 738 goto restart; 739 } 740 crit_exit(); 741 742 if (uio) { 743 top = m_uiomove(uio); 744 if (top == NULL) 745 goto release; 746 } 747 748 dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE); 749 if (dontroute) 750 so->so_options |= SO_DONTROUTE; 751 752 error = so_pru_send(so, 0, top, addr, NULL, td); 753 top = NULL; /* sent or freed in lower layer */ 754 755 if (dontroute) 756 so->so_options &= ~SO_DONTROUTE; 757 758 release: 759 ssb_unlock(&so->so_snd); 760 out: 761 if (top) 762 m_freem(top); 763 return (error); 764 } 765 766 /* 767 * Implement receive operations on a socket. 768 * We depend on the way that records are added to the signalsockbuf 769 * by sbappend*. In particular, each record (mbufs linked through m_next) 770 * must begin with an address if the protocol so specifies, 771 * followed by an optional mbuf or mbufs containing ancillary data, 772 * and then zero or more mbufs of data. 773 * In order to avoid blocking network interrupts for the entire time here, 774 * we exit the critical section while doing the actual copy to user space. 775 * Although the signalsockbuf is locked, new data may still be appended, 776 * and thus we must maintain consistency of the signalsockbuf during that time. 777 * 778 * The caller may receive the data as a single mbuf chain by supplying 779 * an mbuf **mp0 for use in returning the chain. The uio is then used 780 * only for the count in uio_resid. 781 */ 782 int 783 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 784 struct sockbuf *sio, struct mbuf **controlp, int *flagsp) 785 { 786 struct mbuf *m, *n; 787 struct mbuf *free_chain = NULL; 788 int flags, len, error, offset; 789 struct protosw *pr = so->so_proto; 790 int moff, type = 0; 791 int resid, orig_resid; 792 793 if (uio) 794 resid = uio->uio_resid; 795 else 796 resid = (int)(sio->sb_climit - sio->sb_cc); 797 orig_resid = resid; 798 799 if (psa) 800 *psa = NULL; 801 if (controlp) 802 *controlp = NULL; 803 if (flagsp) 804 flags = *flagsp &~ MSG_EOR; 805 else 806 flags = 0; 807 if (flags & MSG_OOB) { 808 m = m_get(MB_WAIT, MT_DATA); 809 if (m == NULL) 810 return (ENOBUFS); 811 error = so_pru_rcvoob(so, m, flags & MSG_PEEK); 812 if (error) 813 goto bad; 814 if (sio) { 815 do { 816 sbappend(sio, m); 817 resid -= m->m_len; 818 } while (resid > 0 && m); 819 } else { 820 do { 821 uio->uio_resid = resid; 822 error = uiomove(mtod(m, caddr_t), 823 (int)min(resid, m->m_len), uio); 824 resid = uio->uio_resid; 825 m = m_free(m); 826 } while (uio->uio_resid && error == 0 && m); 827 } 828 bad: 829 if (m) 830 m_freem(m); 831 return (error); 832 } 833 if (so->so_state & SS_ISCONFIRMING && resid) 834 so_pru_rcvd(so, 0); 835 836 restart: 837 crit_enter(); 838 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags)); 839 if (error) 840 goto done; 841 842 m = so->so_rcv.ssb_mb; 843 /* 844 * If we have less data than requested, block awaiting more 845 * (subject to any timeout) if: 846 * 1. the current count is less than the low water mark, or 847 * 2. MSG_WAITALL is set, and it is possible to do the entire 848 * receive operation at once if we block (resid <= hiwat). 849 * 3. MSG_DONTWAIT is not set 850 * If MSG_WAITALL is set but resid is larger than the receive buffer, 851 * we have to do the receive in sections, and thus risk returning 852 * a short count if a timeout or signal occurs after we start. 853 */ 854 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 855 so->so_rcv.ssb_cc < resid) && 856 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat || 857 ((flags & MSG_WAITALL) && resid <= so->so_rcv.ssb_hiwat)) && 858 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 859 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1")); 860 if (so->so_error) { 861 if (m) 862 goto dontblock; 863 error = so->so_error; 864 if ((flags & MSG_PEEK) == 0) 865 so->so_error = 0; 866 goto release; 867 } 868 if (so->so_state & SS_CANTRCVMORE) { 869 if (m) 870 goto dontblock; 871 else 872 goto release; 873 } 874 for (; m; m = m->m_next) { 875 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 876 m = so->so_rcv.ssb_mb; 877 goto dontblock; 878 } 879 } 880 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 881 (pr->pr_flags & PR_CONNREQUIRED)) { 882 error = ENOTCONN; 883 goto release; 884 } 885 if (resid == 0) 886 goto release; 887 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) { 888 error = EWOULDBLOCK; 889 goto release; 890 } 891 ssb_unlock(&so->so_rcv); 892 error = ssb_wait(&so->so_rcv); 893 if (error) 894 goto done; 895 crit_exit(); 896 goto restart; 897 } 898 dontblock: 899 if (uio && uio->uio_td && uio->uio_td->td_proc) 900 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++; 901 902 /* 903 * note: m should be == sb_mb here. Cache the next record while 904 * cleaning up. Note that calling m_free*() will break out critical 905 * section. 906 */ 907 KKASSERT(m == so->so_rcv.ssb_mb); 908 909 /* 910 * Skip any address mbufs prepending the record. 911 */ 912 if (pr->pr_flags & PR_ADDR) { 913 KASSERT(m->m_type == MT_SONAME, ("receive 1a")); 914 orig_resid = 0; 915 if (psa) 916 *psa = dup_sockaddr(mtod(m, struct sockaddr *)); 917 if (flags & MSG_PEEK) 918 m = m->m_next; 919 else 920 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 921 } 922 923 /* 924 * Skip any control mbufs prepending the record. 925 */ 926 #ifdef SCTP 927 if (pr->pr_flags & PR_ADDR_OPT) { 928 /* 929 * For SCTP we may be getting a 930 * whole message OR a partial delivery. 931 */ 932 if (m && m->m_type == MT_SONAME) { 933 orig_resid = 0; 934 if (psa) 935 *psa = dup_sockaddr(mtod(m, struct sockaddr *)); 936 if (flags & MSG_PEEK) 937 m = m->m_next; 938 else 939 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 940 } 941 } 942 #endif /* SCTP */ 943 while (m && m->m_type == MT_CONTROL && error == 0) { 944 if (flags & MSG_PEEK) { 945 if (controlp) 946 *controlp = m_copy(m, 0, m->m_len); 947 m = m->m_next; /* XXX race */ 948 } else { 949 if (controlp) { 950 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL); 951 if (pr->pr_domain->dom_externalize && 952 mtod(m, struct cmsghdr *)->cmsg_type == 953 SCM_RIGHTS) 954 error = (*pr->pr_domain->dom_externalize)(m); 955 *controlp = m; 956 m = n; 957 } else { 958 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 959 } 960 } 961 if (controlp && *controlp) { 962 orig_resid = 0; 963 controlp = &(*controlp)->m_next; 964 } 965 } 966 967 /* 968 * flag OOB data. 969 */ 970 if (m) { 971 type = m->m_type; 972 if (type == MT_OOBDATA) 973 flags |= MSG_OOB; 974 } 975 976 /* 977 * Copy to the UIO or mbuf return chain (*mp). 978 */ 979 moff = 0; 980 offset = 0; 981 while (m && resid > 0 && error == 0) { 982 if (m->m_type == MT_OOBDATA) { 983 if (type != MT_OOBDATA) 984 break; 985 } else if (type == MT_OOBDATA) 986 break; 987 else 988 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER, 989 ("receive 3")); 990 so->so_state &= ~SS_RCVATMARK; 991 len = resid; 992 if (so->so_oobmark && len > so->so_oobmark - offset) 993 len = so->so_oobmark - offset; 994 if (len > m->m_len - moff) 995 len = m->m_len - moff; 996 997 /* 998 * Copy out to the UIO or pass the mbufs back to the SIO. 999 * The SIO is dealt with when we eat the mbuf, but deal 1000 * with the resid here either way. 1001 */ 1002 if (uio) { 1003 crit_exit(); 1004 uio->uio_resid = resid; 1005 error = uiomove(mtod(m, caddr_t) + moff, len, uio); 1006 resid = uio->uio_resid; 1007 crit_enter(); 1008 if (error) 1009 goto release; 1010 } else { 1011 resid -= len; 1012 } 1013 1014 /* 1015 * Eat the entire mbuf or just a piece of it 1016 */ 1017 if (len == m->m_len - moff) { 1018 if (m->m_flags & M_EOR) 1019 flags |= MSG_EOR; 1020 #ifdef SCTP 1021 if (m->m_flags & M_NOTIFICATION) 1022 flags |= MSG_NOTIFICATION; 1023 #endif /* SCTP */ 1024 if (flags & MSG_PEEK) { 1025 m = m->m_next; 1026 moff = 0; 1027 } else { 1028 if (sio) { 1029 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL); 1030 sbappend(sio, m); 1031 m = n; 1032 } else { 1033 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1034 } 1035 } 1036 } else { 1037 if (flags & MSG_PEEK) { 1038 moff += len; 1039 } else { 1040 if (sio) { 1041 n = m_copym(m, 0, len, MB_WAIT); 1042 if (n) 1043 sbappend(sio, n); 1044 } 1045 m->m_data += len; 1046 m->m_len -= len; 1047 so->so_rcv.ssb_cc -= len; 1048 } 1049 } 1050 if (so->so_oobmark) { 1051 if ((flags & MSG_PEEK) == 0) { 1052 so->so_oobmark -= len; 1053 if (so->so_oobmark == 0) { 1054 so->so_state |= SS_RCVATMARK; 1055 break; 1056 } 1057 } else { 1058 offset += len; 1059 if (offset == so->so_oobmark) 1060 break; 1061 } 1062 } 1063 if (flags & MSG_EOR) 1064 break; 1065 /* 1066 * If the MSG_WAITALL flag is set (for non-atomic socket), 1067 * we must not quit until resid == 0 or an error 1068 * termination. If a signal/timeout occurs, return 1069 * with a short count but without error. 1070 * Keep signalsockbuf locked against other readers. 1071 */ 1072 while ((flags & MSG_WAITALL) && m == NULL && 1073 resid > 0 && !sosendallatonce(so) && 1074 so->so_rcv.ssb_mb == NULL) { 1075 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1076 break; 1077 /* 1078 * The window might have closed to zero, make 1079 * sure we send an ack now that we've drained 1080 * the buffer or we might end up blocking until 1081 * the idle takes over (5 seconds). 1082 */ 1083 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1084 so_pru_rcvd(so, flags); 1085 error = ssb_wait(&so->so_rcv); 1086 if (error) { 1087 ssb_unlock(&so->so_rcv); 1088 error = 0; 1089 goto done; 1090 } 1091 m = so->so_rcv.ssb_mb; 1092 } 1093 } 1094 1095 /* 1096 * If an atomic read was requested but unread data still remains 1097 * in the record, set MSG_TRUNC. 1098 */ 1099 if (m && pr->pr_flags & PR_ATOMIC) 1100 flags |= MSG_TRUNC; 1101 1102 /* 1103 * Cleanup. If an atomic read was requested drop any unread data. 1104 */ 1105 if ((flags & MSG_PEEK) == 0) { 1106 if (m && (pr->pr_flags & PR_ATOMIC)) 1107 sbdroprecord(&so->so_rcv.sb); 1108 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1109 so_pru_rcvd(so, flags); 1110 } 1111 1112 if (orig_resid == resid && orig_resid && 1113 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1114 ssb_unlock(&so->so_rcv); 1115 crit_exit(); 1116 goto restart; 1117 } 1118 1119 if (flagsp) 1120 *flagsp |= flags; 1121 release: 1122 ssb_unlock(&so->so_rcv); 1123 done: 1124 crit_exit(); 1125 if (free_chain) 1126 m_freem(free_chain); 1127 return (error); 1128 } 1129 1130 int 1131 soshutdown(struct socket *so, int how) 1132 { 1133 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1134 return (EINVAL); 1135 1136 if (how != SHUT_WR) 1137 sorflush(so); 1138 if (how != SHUT_RD) 1139 return (so_pru_shutdown(so)); 1140 return (0); 1141 } 1142 1143 void 1144 sorflush(struct socket *so) 1145 { 1146 struct signalsockbuf *ssb = &so->so_rcv; 1147 struct protosw *pr = so->so_proto; 1148 struct signalsockbuf asb; 1149 1150 ssb->ssb_flags |= SSB_NOINTR; 1151 (void) ssb_lock(ssb, M_WAITOK); 1152 1153 crit_enter(); 1154 socantrcvmore(so); 1155 ssb_unlock(ssb); 1156 asb = *ssb; 1157 bzero((caddr_t)ssb, sizeof (*ssb)); 1158 if (asb.ssb_flags & SSB_KNOTE) { 1159 ssb->ssb_sel.si_note = asb.ssb_sel.si_note; 1160 ssb->ssb_flags = SSB_KNOTE; 1161 } 1162 crit_exit(); 1163 1164 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1165 (*pr->pr_domain->dom_dispose)(asb.ssb_mb); 1166 ssb_release(&asb, so); 1167 } 1168 1169 #ifdef INET 1170 static int 1171 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt) 1172 { 1173 struct accept_filter_arg *afap = NULL; 1174 struct accept_filter *afp; 1175 struct so_accf *af = so->so_accf; 1176 int error = 0; 1177 1178 /* do not set/remove accept filters on non listen sockets */ 1179 if ((so->so_options & SO_ACCEPTCONN) == 0) { 1180 error = EINVAL; 1181 goto out; 1182 } 1183 1184 /* removing the filter */ 1185 if (sopt == NULL) { 1186 if (af != NULL) { 1187 if (af->so_accept_filter != NULL && 1188 af->so_accept_filter->accf_destroy != NULL) { 1189 af->so_accept_filter->accf_destroy(so); 1190 } 1191 if (af->so_accept_filter_str != NULL) { 1192 FREE(af->so_accept_filter_str, M_ACCF); 1193 } 1194 FREE(af, M_ACCF); 1195 so->so_accf = NULL; 1196 } 1197 so->so_options &= ~SO_ACCEPTFILTER; 1198 return (0); 1199 } 1200 /* adding a filter */ 1201 /* must remove previous filter first */ 1202 if (af != NULL) { 1203 error = EINVAL; 1204 goto out; 1205 } 1206 /* don't put large objects on the kernel stack */ 1207 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK); 1208 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap); 1209 afap->af_name[sizeof(afap->af_name)-1] = '\0'; 1210 afap->af_arg[sizeof(afap->af_arg)-1] = '\0'; 1211 if (error) 1212 goto out; 1213 afp = accept_filt_get(afap->af_name); 1214 if (afp == NULL) { 1215 error = ENOENT; 1216 goto out; 1217 } 1218 MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO); 1219 if (afp->accf_create != NULL) { 1220 if (afap->af_name[0] != '\0') { 1221 int len = strlen(afap->af_name) + 1; 1222 1223 MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK); 1224 strcpy(af->so_accept_filter_str, afap->af_name); 1225 } 1226 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg); 1227 if (af->so_accept_filter_arg == NULL) { 1228 FREE(af->so_accept_filter_str, M_ACCF); 1229 FREE(af, M_ACCF); 1230 so->so_accf = NULL; 1231 error = EINVAL; 1232 goto out; 1233 } 1234 } 1235 af->so_accept_filter = afp; 1236 so->so_accf = af; 1237 so->so_options |= SO_ACCEPTFILTER; 1238 out: 1239 if (afap != NULL) 1240 FREE(afap, M_TEMP); 1241 return (error); 1242 } 1243 #endif /* INET */ 1244 1245 /* 1246 * Perhaps this routine, and sooptcopyout(), below, ought to come in 1247 * an additional variant to handle the case where the option value needs 1248 * to be some kind of integer, but not a specific size. 1249 * In addition to their use here, these functions are also called by the 1250 * protocol-level pr_ctloutput() routines. 1251 */ 1252 int 1253 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 1254 { 1255 return soopt_to_kbuf(sopt, buf, len, minlen); 1256 } 1257 1258 int 1259 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 1260 { 1261 size_t valsize; 1262 1263 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 1264 KKASSERT(kva_p(buf)); 1265 1266 /* 1267 * If the user gives us more than we wanted, we ignore it, 1268 * but if we don't get the minimum length the caller 1269 * wants, we return EINVAL. On success, sopt->sopt_valsize 1270 * is set to however much we actually retrieved. 1271 */ 1272 if ((valsize = sopt->sopt_valsize) < minlen) 1273 return EINVAL; 1274 if (valsize > len) 1275 sopt->sopt_valsize = valsize = len; 1276 1277 bcopy(sopt->sopt_val, buf, valsize); 1278 return 0; 1279 } 1280 1281 1282 int 1283 sosetopt(struct socket *so, struct sockopt *sopt) 1284 { 1285 int error, optval; 1286 struct linger l; 1287 struct timeval tv; 1288 u_long val; 1289 1290 error = 0; 1291 sopt->sopt_dir = SOPT_SET; 1292 if (sopt->sopt_level != SOL_SOCKET) { 1293 if (so->so_proto && so->so_proto->pr_ctloutput) { 1294 return (so_pru_ctloutput(so, sopt)); 1295 } 1296 error = ENOPROTOOPT; 1297 } else { 1298 switch (sopt->sopt_name) { 1299 #ifdef INET 1300 case SO_ACCEPTFILTER: 1301 error = do_setopt_accept_filter(so, sopt); 1302 if (error) 1303 goto bad; 1304 break; 1305 #endif /* INET */ 1306 case SO_LINGER: 1307 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 1308 if (error) 1309 goto bad; 1310 1311 so->so_linger = l.l_linger; 1312 if (l.l_onoff) 1313 so->so_options |= SO_LINGER; 1314 else 1315 so->so_options &= ~SO_LINGER; 1316 break; 1317 1318 case SO_DEBUG: 1319 case SO_KEEPALIVE: 1320 case SO_DONTROUTE: 1321 case SO_USELOOPBACK: 1322 case SO_BROADCAST: 1323 case SO_REUSEADDR: 1324 case SO_REUSEPORT: 1325 case SO_OOBINLINE: 1326 case SO_TIMESTAMP: 1327 error = sooptcopyin(sopt, &optval, sizeof optval, 1328 sizeof optval); 1329 if (error) 1330 goto bad; 1331 if (optval) 1332 so->so_options |= sopt->sopt_name; 1333 else 1334 so->so_options &= ~sopt->sopt_name; 1335 break; 1336 1337 case SO_SNDBUF: 1338 case SO_RCVBUF: 1339 case SO_SNDLOWAT: 1340 case SO_RCVLOWAT: 1341 error = sooptcopyin(sopt, &optval, sizeof optval, 1342 sizeof optval); 1343 if (error) 1344 goto bad; 1345 1346 /* 1347 * Values < 1 make no sense for any of these 1348 * options, so disallow them. 1349 */ 1350 if (optval < 1) { 1351 error = EINVAL; 1352 goto bad; 1353 } 1354 1355 switch (sopt->sopt_name) { 1356 case SO_SNDBUF: 1357 case SO_RCVBUF: 1358 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ? 1359 &so->so_snd : &so->so_rcv, (u_long)optval, 1360 so, 1361 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) { 1362 error = ENOBUFS; 1363 goto bad; 1364 } 1365 break; 1366 1367 /* 1368 * Make sure the low-water is never greater than 1369 * the high-water. 1370 */ 1371 case SO_SNDLOWAT: 1372 so->so_snd.ssb_lowat = 1373 (optval > so->so_snd.ssb_hiwat) ? 1374 so->so_snd.ssb_hiwat : optval; 1375 break; 1376 case SO_RCVLOWAT: 1377 so->so_rcv.ssb_lowat = 1378 (optval > so->so_rcv.ssb_hiwat) ? 1379 so->so_rcv.ssb_hiwat : optval; 1380 break; 1381 } 1382 break; 1383 1384 case SO_SNDTIMEO: 1385 case SO_RCVTIMEO: 1386 error = sooptcopyin(sopt, &tv, sizeof tv, 1387 sizeof tv); 1388 if (error) 1389 goto bad; 1390 1391 /* assert(hz > 0); */ 1392 if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz || 1393 tv.tv_usec < 0 || tv.tv_usec >= 1000000) { 1394 error = EDOM; 1395 goto bad; 1396 } 1397 /* assert(tick > 0); */ 1398 /* assert(ULONG_MAX - SHRT_MAX >= 1000000); */ 1399 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick; 1400 if (val > SHRT_MAX) { 1401 error = EDOM; 1402 goto bad; 1403 } 1404 if (val == 0 && tv.tv_usec != 0) 1405 val = 1; 1406 1407 switch (sopt->sopt_name) { 1408 case SO_SNDTIMEO: 1409 so->so_snd.ssb_timeo = val; 1410 break; 1411 case SO_RCVTIMEO: 1412 so->so_rcv.ssb_timeo = val; 1413 break; 1414 } 1415 break; 1416 default: 1417 error = ENOPROTOOPT; 1418 break; 1419 } 1420 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1421 (void) so_pru_ctloutput(so, sopt); 1422 } 1423 } 1424 bad: 1425 return (error); 1426 } 1427 1428 /* Helper routine for getsockopt */ 1429 int 1430 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 1431 { 1432 soopt_from_kbuf(sopt, buf, len); 1433 return 0; 1434 } 1435 1436 void 1437 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len) 1438 { 1439 size_t valsize; 1440 1441 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 1442 KKASSERT(kva_p(buf)); 1443 1444 /* 1445 * Documented get behavior is that we always return a value, 1446 * possibly truncated to fit in the user's buffer. 1447 * Traditional behavior is that we always tell the user 1448 * precisely how much we copied, rather than something useful 1449 * like the total amount we had available for her. 1450 * Note that this interface is not idempotent; the entire answer must 1451 * generated ahead of time. 1452 */ 1453 valsize = min(len, sopt->sopt_valsize); 1454 sopt->sopt_valsize = valsize; 1455 if (sopt->sopt_val != 0) { 1456 bcopy(buf, sopt->sopt_val, valsize); 1457 } 1458 } 1459 1460 int 1461 sogetopt(struct socket *so, struct sockopt *sopt) 1462 { 1463 int error, optval; 1464 struct linger l; 1465 struct timeval tv; 1466 #ifdef INET 1467 struct accept_filter_arg *afap; 1468 #endif 1469 1470 error = 0; 1471 sopt->sopt_dir = SOPT_GET; 1472 if (sopt->sopt_level != SOL_SOCKET) { 1473 if (so->so_proto && so->so_proto->pr_ctloutput) { 1474 return (so_pru_ctloutput(so, sopt)); 1475 } else 1476 return (ENOPROTOOPT); 1477 } else { 1478 switch (sopt->sopt_name) { 1479 #ifdef INET 1480 case SO_ACCEPTFILTER: 1481 if ((so->so_options & SO_ACCEPTCONN) == 0) 1482 return (EINVAL); 1483 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), 1484 M_TEMP, M_WAITOK | M_ZERO); 1485 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 1486 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name); 1487 if (so->so_accf->so_accept_filter_str != NULL) 1488 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str); 1489 } 1490 error = sooptcopyout(sopt, afap, sizeof(*afap)); 1491 FREE(afap, M_TEMP); 1492 break; 1493 #endif /* INET */ 1494 1495 case SO_LINGER: 1496 l.l_onoff = so->so_options & SO_LINGER; 1497 l.l_linger = so->so_linger; 1498 error = sooptcopyout(sopt, &l, sizeof l); 1499 break; 1500 1501 case SO_USELOOPBACK: 1502 case SO_DONTROUTE: 1503 case SO_DEBUG: 1504 case SO_KEEPALIVE: 1505 case SO_REUSEADDR: 1506 case SO_REUSEPORT: 1507 case SO_BROADCAST: 1508 case SO_OOBINLINE: 1509 case SO_TIMESTAMP: 1510 optval = so->so_options & sopt->sopt_name; 1511 integer: 1512 error = sooptcopyout(sopt, &optval, sizeof optval); 1513 break; 1514 1515 case SO_TYPE: 1516 optval = so->so_type; 1517 goto integer; 1518 1519 case SO_ERROR: 1520 optval = so->so_error; 1521 so->so_error = 0; 1522 goto integer; 1523 1524 case SO_SNDBUF: 1525 optval = so->so_snd.ssb_hiwat; 1526 goto integer; 1527 1528 case SO_RCVBUF: 1529 optval = so->so_rcv.ssb_hiwat; 1530 goto integer; 1531 1532 case SO_SNDLOWAT: 1533 optval = so->so_snd.ssb_lowat; 1534 goto integer; 1535 1536 case SO_RCVLOWAT: 1537 optval = so->so_rcv.ssb_lowat; 1538 goto integer; 1539 1540 case SO_SNDTIMEO: 1541 case SO_RCVTIMEO: 1542 optval = (sopt->sopt_name == SO_SNDTIMEO ? 1543 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo); 1544 1545 tv.tv_sec = optval / hz; 1546 tv.tv_usec = (optval % hz) * tick; 1547 error = sooptcopyout(sopt, &tv, sizeof tv); 1548 break; 1549 1550 default: 1551 error = ENOPROTOOPT; 1552 break; 1553 } 1554 return (error); 1555 } 1556 } 1557 1558 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */ 1559 int 1560 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 1561 { 1562 struct mbuf *m, *m_prev; 1563 int sopt_size = sopt->sopt_valsize, msize; 1564 1565 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA, 1566 0, &msize); 1567 if (m == NULL) 1568 return (ENOBUFS); 1569 m->m_len = min(msize, sopt_size); 1570 sopt_size -= m->m_len; 1571 *mp = m; 1572 m_prev = m; 1573 1574 while (sopt_size > 0) { 1575 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, 1576 MT_DATA, 0, &msize); 1577 if (m == NULL) { 1578 m_freem(*mp); 1579 return (ENOBUFS); 1580 } 1581 m->m_len = min(msize, sopt_size); 1582 sopt_size -= m->m_len; 1583 m_prev->m_next = m; 1584 m_prev = m; 1585 } 1586 return (0); 1587 } 1588 1589 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */ 1590 int 1591 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 1592 { 1593 soopt_to_mbuf(sopt, m); 1594 return 0; 1595 } 1596 1597 void 1598 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m) 1599 { 1600 size_t valsize; 1601 void *val; 1602 1603 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 1604 KKASSERT(kva_p(m)); 1605 if (sopt->sopt_val == NULL) 1606 return; 1607 val = sopt->sopt_val; 1608 valsize = sopt->sopt_valsize; 1609 while (m != NULL && valsize >= m->m_len) { 1610 bcopy(val, mtod(m, char *), m->m_len); 1611 valsize -= m->m_len; 1612 val = (caddr_t)val + m->m_len; 1613 m = m->m_next; 1614 } 1615 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 1616 panic("ip6_sooptmcopyin"); 1617 } 1618 1619 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */ 1620 int 1621 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 1622 { 1623 return soopt_from_mbuf(sopt, m); 1624 } 1625 1626 int 1627 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m) 1628 { 1629 struct mbuf *m0 = m; 1630 size_t valsize = 0; 1631 size_t maxsize; 1632 void *val; 1633 1634 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 1635 KKASSERT(kva_p(m)); 1636 if (sopt->sopt_val == NULL) 1637 return 0; 1638 val = sopt->sopt_val; 1639 maxsize = sopt->sopt_valsize; 1640 while (m != NULL && maxsize >= m->m_len) { 1641 bcopy(mtod(m, char *), val, m->m_len); 1642 maxsize -= m->m_len; 1643 val = (caddr_t)val + m->m_len; 1644 valsize += m->m_len; 1645 m = m->m_next; 1646 } 1647 if (m != NULL) { 1648 /* enough soopt buffer should be given from user-land */ 1649 m_freem(m0); 1650 return (EINVAL); 1651 } 1652 sopt->sopt_valsize = valsize; 1653 return 0; 1654 } 1655 1656 void 1657 sohasoutofband(struct socket *so) 1658 { 1659 if (so->so_sigio != NULL) 1660 pgsigio(so->so_sigio, SIGURG, 0); 1661 selwakeup(&so->so_rcv.ssb_sel); 1662 } 1663 1664 int 1665 sopoll(struct socket *so, int events, struct ucred *cred, struct thread *td) 1666 { 1667 int revents = 0; 1668 1669 crit_enter(); 1670 1671 if (events & (POLLIN | POLLRDNORM)) 1672 if (soreadable(so)) 1673 revents |= events & (POLLIN | POLLRDNORM); 1674 1675 if (events & POLLINIGNEOF) 1676 if (so->so_rcv.ssb_cc >= so->so_rcv.ssb_lowat || 1677 !TAILQ_EMPTY(&so->so_comp) || so->so_error) 1678 revents |= POLLINIGNEOF; 1679 1680 if (events & (POLLOUT | POLLWRNORM)) 1681 if (sowriteable(so)) 1682 revents |= events & (POLLOUT | POLLWRNORM); 1683 1684 if (events & (POLLPRI | POLLRDBAND)) 1685 if (so->so_oobmark || (so->so_state & SS_RCVATMARK)) 1686 revents |= events & (POLLPRI | POLLRDBAND); 1687 1688 if (revents == 0) { 1689 if (events & 1690 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM | 1691 POLLRDBAND)) { 1692 selrecord(td, &so->so_rcv.ssb_sel); 1693 so->so_rcv.ssb_flags |= SSB_SEL; 1694 } 1695 1696 if (events & (POLLOUT | POLLWRNORM)) { 1697 selrecord(td, &so->so_snd.ssb_sel); 1698 so->so_snd.ssb_flags |= SSB_SEL; 1699 } 1700 } 1701 1702 crit_exit(); 1703 return (revents); 1704 } 1705 1706 int 1707 sokqfilter(struct file *fp, struct knote *kn) 1708 { 1709 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1710 struct signalsockbuf *ssb; 1711 1712 switch (kn->kn_filter) { 1713 case EVFILT_READ: 1714 if (so->so_options & SO_ACCEPTCONN) 1715 kn->kn_fop = &solisten_filtops; 1716 else 1717 kn->kn_fop = &soread_filtops; 1718 ssb = &so->so_rcv; 1719 break; 1720 case EVFILT_WRITE: 1721 kn->kn_fop = &sowrite_filtops; 1722 ssb = &so->so_snd; 1723 break; 1724 default: 1725 return (1); 1726 } 1727 1728 crit_enter(); 1729 SLIST_INSERT_HEAD(&ssb->ssb_sel.si_note, kn, kn_selnext); 1730 ssb->ssb_flags |= SSB_KNOTE; 1731 crit_exit(); 1732 return (0); 1733 } 1734 1735 static void 1736 filt_sordetach(struct knote *kn) 1737 { 1738 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1739 1740 crit_enter(); 1741 SLIST_REMOVE(&so->so_rcv.ssb_sel.si_note, kn, knote, kn_selnext); 1742 if (SLIST_EMPTY(&so->so_rcv.ssb_sel.si_note)) 1743 so->so_rcv.ssb_flags &= ~SSB_KNOTE; 1744 crit_exit(); 1745 } 1746 1747 /*ARGSUSED*/ 1748 static int 1749 filt_soread(struct knote *kn, long hint) 1750 { 1751 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1752 1753 kn->kn_data = so->so_rcv.ssb_cc; 1754 if (so->so_state & SS_CANTRCVMORE) { 1755 kn->kn_flags |= EV_EOF; 1756 kn->kn_fflags = so->so_error; 1757 return (1); 1758 } 1759 if (so->so_error) /* temporary udp error */ 1760 return (1); 1761 if (kn->kn_sfflags & NOTE_LOWAT) 1762 return (kn->kn_data >= kn->kn_sdata); 1763 return (kn->kn_data >= so->so_rcv.ssb_lowat); 1764 } 1765 1766 static void 1767 filt_sowdetach(struct knote *kn) 1768 { 1769 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1770 1771 crit_enter(); 1772 SLIST_REMOVE(&so->so_snd.ssb_sel.si_note, kn, knote, kn_selnext); 1773 if (SLIST_EMPTY(&so->so_snd.ssb_sel.si_note)) 1774 so->so_snd.ssb_flags &= ~SSB_KNOTE; 1775 crit_exit(); 1776 } 1777 1778 /*ARGSUSED*/ 1779 static int 1780 filt_sowrite(struct knote *kn, long hint) 1781 { 1782 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1783 1784 kn->kn_data = ssb_space(&so->so_snd); 1785 if (so->so_state & SS_CANTSENDMORE) { 1786 kn->kn_flags |= EV_EOF; 1787 kn->kn_fflags = so->so_error; 1788 return (1); 1789 } 1790 if (so->so_error) /* temporary udp error */ 1791 return (1); 1792 if (((so->so_state & SS_ISCONNECTED) == 0) && 1793 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1794 return (0); 1795 if (kn->kn_sfflags & NOTE_LOWAT) 1796 return (kn->kn_data >= kn->kn_sdata); 1797 return (kn->kn_data >= so->so_snd.ssb_lowat); 1798 } 1799 1800 /*ARGSUSED*/ 1801 static int 1802 filt_solisten(struct knote *kn, long hint) 1803 { 1804 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1805 1806 kn->kn_data = so->so_qlen; 1807 return (! TAILQ_EMPTY(&so->so_comp)); 1808 } 1809