1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 34 * $FreeBSD$ 35 */ 36 37 #include "opt_inet.h" 38 #include "opt_mac.h" 39 #include "opt_zero.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/fcntl.h> 44 #include <sys/lock.h> 45 #include <sys/mac.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #include <sys/mutex.h> 49 #include <sys/domain.h> 50 #include <sys/file.h> /* for struct knote */ 51 #include <sys/kernel.h> 52 #include <sys/malloc.h> 53 #include <sys/event.h> 54 #include <sys/poll.h> 55 #include <sys/proc.h> 56 #include <sys/protosw.h> 57 #include <sys/socket.h> 58 #include <sys/socketvar.h> 59 #include <sys/resourcevar.h> 60 #include <sys/signalvar.h> 61 #include <sys/sysctl.h> 62 #include <sys/uio.h> 63 #include <sys/jail.h> 64 65 #include <vm/uma.h> 66 67 #include <machine/limits.h> 68 69 #ifdef INET 70 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt); 71 #endif 72 73 static void filt_sordetach(struct knote *kn); 74 static int filt_soread(struct knote *kn, long hint); 75 static void filt_sowdetach(struct knote *kn); 76 static int filt_sowrite(struct knote *kn, long hint); 77 static int filt_solisten(struct knote *kn, long hint); 78 79 static struct filterops solisten_filtops = 80 { 1, NULL, filt_sordetach, filt_solisten }; 81 static struct filterops soread_filtops = 82 { 1, NULL, filt_sordetach, filt_soread }; 83 static struct filterops sowrite_filtops = 84 { 1, NULL, filt_sowdetach, filt_sowrite }; 85 86 uma_zone_t socket_zone; 87 so_gen_t so_gencnt; /* generation count for sockets */ 88 89 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 90 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 91 92 SYSCTL_DECL(_kern_ipc); 93 94 static int somaxconn = SOMAXCONN; 95 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW, 96 &somaxconn, 0, "Maximum pending socket connection queue size"); 97 static int numopensockets; 98 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD, 99 &numopensockets, 0, "Number of open sockets"); 100 #ifdef ZERO_COPY_SOCKETS 101 /* These aren't static because they're used in other files. */ 102 int so_zero_copy_send = 1; 103 int so_zero_copy_receive = 1; 104 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0, 105 "Zero copy controls"); 106 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW, 107 &so_zero_copy_receive, 0, "Enable zero copy receive"); 108 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW, 109 &so_zero_copy_send, 0, "Enable zero copy send"); 110 #endif /* ZERO_COPY_SOCKETS */ 111 112 113 /* 114 * Socket operation routines. 115 * These routines are called by the routines in 116 * sys_socket.c or from a system process, and 117 * implement the semantics of socket operations by 118 * switching out to the protocol specific routines. 119 */ 120 121 /* 122 * Get a socket structure from our zone, and initialize it. 123 * Note that it would probably be better to allocate socket 124 * and PCB at the same time, but I'm not convinced that all 125 * the protocols can be easily modified to do this. 126 * 127 * soalloc() returns a socket with a ref count of 0. 128 */ 129 struct socket * 130 soalloc(waitok) 131 int waitok; 132 { 133 struct socket *so; 134 int flag; 135 136 if (waitok == 1) 137 flag = M_WAITOK; 138 else 139 flag = M_NOWAIT; 140 flag |= M_ZERO; 141 so = uma_zalloc(socket_zone, flag); 142 if (so) { 143 /* XXX race condition for reentrant kernel */ 144 so->so_gencnt = ++so_gencnt; 145 /* sx_init(&so->so_sxlock, "socket sxlock"); */ 146 TAILQ_INIT(&so->so_aiojobq); 147 ++numopensockets; 148 #ifdef MAC 149 mac_init_socket(so); 150 #endif 151 } 152 return so; 153 } 154 155 /* 156 * socreate returns a socket with a ref count of 1. The socket should be 157 * closed with soclose(). 158 */ 159 int 160 socreate(dom, aso, type, proto, cred, td) 161 int dom; 162 struct socket **aso; 163 register int type; 164 int proto; 165 struct ucred *cred; 166 struct thread *td; 167 { 168 register struct protosw *prp; 169 register struct socket *so; 170 register int error; 171 172 if (proto) 173 prp = pffindproto(dom, proto, type); 174 else 175 prp = pffindtype(dom, type); 176 177 if (prp == 0 || prp->pr_usrreqs->pru_attach == 0) 178 return (EPROTONOSUPPORT); 179 180 if (jailed(cred) && jail_socket_unixiproute_only && 181 prp->pr_domain->dom_family != PF_LOCAL && 182 prp->pr_domain->dom_family != PF_INET && 183 prp->pr_domain->dom_family != PF_ROUTE) { 184 return (EPROTONOSUPPORT); 185 } 186 187 if (prp->pr_type != type) 188 return (EPROTOTYPE); 189 so = soalloc(M_NOWAIT); 190 if (so == NULL) 191 return (ENOBUFS); 192 193 TAILQ_INIT(&so->so_incomp); 194 TAILQ_INIT(&so->so_comp); 195 so->so_type = type; 196 so->so_cred = crhold(cred); 197 so->so_proto = prp; 198 #ifdef MAC 199 mac_create_socket(cred, so); 200 #endif 201 soref(so); 202 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); 203 if (error) { 204 so->so_state |= SS_NOFDREF; 205 sorele(so); 206 return (error); 207 } 208 *aso = so; 209 return (0); 210 } 211 212 int 213 sobind(so, nam, td) 214 struct socket *so; 215 struct sockaddr *nam; 216 struct thread *td; 217 { 218 int s = splnet(); 219 int error; 220 221 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td); 222 splx(s); 223 return (error); 224 } 225 226 static void 227 sodealloc(struct socket *so) 228 { 229 230 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); 231 so->so_gencnt = ++so_gencnt; 232 if (so->so_rcv.sb_hiwat) 233 (void)chgsbsize(so->so_cred->cr_uidinfo, 234 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 235 if (so->so_snd.sb_hiwat) 236 (void)chgsbsize(so->so_cred->cr_uidinfo, 237 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 238 #ifdef INET 239 if (so->so_accf != NULL) { 240 if (so->so_accf->so_accept_filter != NULL && 241 so->so_accf->so_accept_filter->accf_destroy != NULL) { 242 so->so_accf->so_accept_filter->accf_destroy(so); 243 } 244 if (so->so_accf->so_accept_filter_str != NULL) 245 FREE(so->so_accf->so_accept_filter_str, M_ACCF); 246 FREE(so->so_accf, M_ACCF); 247 } 248 #endif 249 #ifdef MAC 250 mac_destroy_socket(so); 251 #endif 252 crfree(so->so_cred); 253 /* sx_destroy(&so->so_sxlock); */ 254 uma_zfree(socket_zone, so); 255 --numopensockets; 256 } 257 258 int 259 solisten(so, backlog, td) 260 register struct socket *so; 261 int backlog; 262 struct thread *td; 263 { 264 int s, error; 265 266 s = splnet(); 267 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, td); 268 if (error) { 269 splx(s); 270 return (error); 271 } 272 if (TAILQ_EMPTY(&so->so_comp)) 273 so->so_options |= SO_ACCEPTCONN; 274 if (backlog < 0 || backlog > somaxconn) 275 backlog = somaxconn; 276 so->so_qlimit = backlog; 277 splx(s); 278 return (0); 279 } 280 281 void 282 sofree(so) 283 register struct socket *so; 284 { 285 struct socket *head = so->so_head; 286 287 KASSERT(so->so_count == 0, ("socket %p so_count not 0", so)); 288 289 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 290 return; 291 if (head != NULL) { 292 if (so->so_state & SS_INCOMP) { 293 TAILQ_REMOVE(&head->so_incomp, so, so_list); 294 head->so_incqlen--; 295 } else if (so->so_state & SS_COMP) { 296 /* 297 * We must not decommission a socket that's 298 * on the accept(2) queue. If we do, then 299 * accept(2) may hang after select(2) indicated 300 * that the listening socket was ready. 301 */ 302 return; 303 } else { 304 panic("sofree: not queued"); 305 } 306 so->so_state &= ~SS_INCOMP; 307 so->so_head = NULL; 308 } 309 sbrelease(&so->so_snd, so); 310 sorflush(so); 311 sodealloc(so); 312 } 313 314 /* 315 * Close a socket on last file table reference removal. 316 * Initiate disconnect if connected. 317 * Free socket when disconnect complete. 318 * 319 * This function will sorele() the socket. Note that soclose() may be 320 * called prior to the ref count reaching zero. The actual socket 321 * structure will not be freed until the ref count reaches zero. 322 */ 323 int 324 soclose(so) 325 register struct socket *so; 326 { 327 int s = splnet(); /* conservative */ 328 int error = 0; 329 330 funsetown(&so->so_sigio); 331 if (so->so_options & SO_ACCEPTCONN) { 332 struct socket *sp, *sonext; 333 334 sp = TAILQ_FIRST(&so->so_incomp); 335 for (; sp != NULL; sp = sonext) { 336 sonext = TAILQ_NEXT(sp, so_list); 337 (void) soabort(sp); 338 } 339 for (sp = TAILQ_FIRST(&so->so_comp); sp != NULL; sp = sonext) { 340 sonext = TAILQ_NEXT(sp, so_list); 341 /* Dequeue from so_comp since sofree() won't do it */ 342 TAILQ_REMOVE(&so->so_comp, sp, so_list); 343 so->so_qlen--; 344 sp->so_state &= ~SS_COMP; 345 sp->so_head = NULL; 346 (void) soabort(sp); 347 } 348 } 349 if (so->so_pcb == 0) 350 goto discard; 351 if (so->so_state & SS_ISCONNECTED) { 352 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 353 error = sodisconnect(so); 354 if (error) 355 goto drop; 356 } 357 if (so->so_options & SO_LINGER) { 358 if ((so->so_state & SS_ISDISCONNECTING) && 359 (so->so_state & SS_NBIO)) 360 goto drop; 361 while (so->so_state & SS_ISCONNECTED) { 362 error = tsleep(&so->so_timeo, 363 PSOCK | PCATCH, "soclos", so->so_linger * hz); 364 if (error) 365 break; 366 } 367 } 368 } 369 drop: 370 if (so->so_pcb) { 371 int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so); 372 if (error == 0) 373 error = error2; 374 } 375 discard: 376 if (so->so_state & SS_NOFDREF) 377 panic("soclose: NOFDREF"); 378 so->so_state |= SS_NOFDREF; 379 sorele(so); 380 splx(s); 381 return (error); 382 } 383 384 /* 385 * Must be called at splnet... 386 */ 387 int 388 soabort(so) 389 struct socket *so; 390 { 391 int error; 392 393 error = (*so->so_proto->pr_usrreqs->pru_abort)(so); 394 if (error) { 395 sotryfree(so); /* note: does not decrement the ref count */ 396 return error; 397 } 398 return (0); 399 } 400 401 int 402 soaccept(so, nam) 403 register struct socket *so; 404 struct sockaddr **nam; 405 { 406 int s = splnet(); 407 int error; 408 409 if ((so->so_state & SS_NOFDREF) == 0) 410 panic("soaccept: !NOFDREF"); 411 so->so_state &= ~SS_NOFDREF; 412 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); 413 splx(s); 414 return (error); 415 } 416 417 int 418 soconnect(so, nam, td) 419 register struct socket *so; 420 struct sockaddr *nam; 421 struct thread *td; 422 { 423 int s; 424 int error; 425 426 if (so->so_options & SO_ACCEPTCONN) 427 return (EOPNOTSUPP); 428 s = splnet(); 429 /* 430 * If protocol is connection-based, can only connect once. 431 * Otherwise, if connected, try to disconnect first. 432 * This allows user to disconnect by connecting to, e.g., 433 * a null address. 434 */ 435 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 436 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 437 (error = sodisconnect(so)))) 438 error = EISCONN; 439 else 440 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td); 441 splx(s); 442 return (error); 443 } 444 445 int 446 soconnect2(so1, so2) 447 register struct socket *so1; 448 struct socket *so2; 449 { 450 int s = splnet(); 451 int error; 452 453 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); 454 splx(s); 455 return (error); 456 } 457 458 int 459 sodisconnect(so) 460 register struct socket *so; 461 { 462 int s = splnet(); 463 int error; 464 465 if ((so->so_state & SS_ISCONNECTED) == 0) { 466 error = ENOTCONN; 467 goto bad; 468 } 469 if (so->so_state & SS_ISDISCONNECTING) { 470 error = EALREADY; 471 goto bad; 472 } 473 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 474 bad: 475 splx(s); 476 return (error); 477 } 478 479 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 480 /* 481 * Send on a socket. 482 * If send must go all at once and message is larger than 483 * send buffering, then hard error. 484 * Lock against other senders. 485 * If must go all at once and not enough room now, then 486 * inform user that this would block and do nothing. 487 * Otherwise, if nonblocking, send as much as possible. 488 * The data to be sent is described by "uio" if nonzero, 489 * otherwise by the mbuf chain "top" (which must be null 490 * if uio is not). Data provided in mbuf chain must be small 491 * enough to send all at once. 492 * 493 * Returns nonzero on error, timeout or signal; callers 494 * must check for short counts if EINTR/ERESTART are returned. 495 * Data and control buffers are freed on return. 496 */ 497 498 #ifdef ZERO_COPY_SOCKETS 499 struct so_zerocopy_stats{ 500 int size_ok; 501 int align_ok; 502 int found_ifp; 503 }; 504 struct so_zerocopy_stats so_zerocp_stats = {0,0,0}; 505 #include <netinet/in.h> 506 #include <net/route.h> 507 #include <netinet/in_pcb.h> 508 #include <vm/vm.h> 509 #include <vm/vm_page.h> 510 #include <vm/vm_object.h> 511 #endif /*ZERO_COPY_SOCKETS*/ 512 513 int 514 sosend(so, addr, uio, top, control, flags, td) 515 register struct socket *so; 516 struct sockaddr *addr; 517 struct uio *uio; 518 struct mbuf *top; 519 struct mbuf *control; 520 int flags; 521 struct thread *td; 522 { 523 struct mbuf **mp; 524 register struct mbuf *m; 525 register long space, len, resid; 526 int clen = 0, error, s, dontroute, mlen; 527 int atomic = sosendallatonce(so) || top; 528 #ifdef ZERO_COPY_SOCKETS 529 int cow_send; 530 #endif /* ZERO_COPY_SOCKETS */ 531 532 if (uio) 533 resid = uio->uio_resid; 534 else 535 resid = top->m_pkthdr.len; 536 /* 537 * In theory resid should be unsigned. 538 * However, space must be signed, as it might be less than 0 539 * if we over-committed, and we must use a signed comparison 540 * of space and resid. On the other hand, a negative resid 541 * causes us to loop sending 0-length segments to the protocol. 542 * 543 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 544 * type sockets since that's an error. 545 */ 546 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 547 error = EINVAL; 548 goto out; 549 } 550 551 dontroute = 552 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 553 (so->so_proto->pr_flags & PR_ATOMIC); 554 if (td) 555 td->td_proc->p_stats->p_ru.ru_msgsnd++; 556 if (control) 557 clen = control->m_len; 558 #define snderr(errno) { error = errno; splx(s); goto release; } 559 560 restart: 561 error = sblock(&so->so_snd, SBLOCKWAIT(flags)); 562 if (error) 563 goto out; 564 do { 565 s = splnet(); 566 if (so->so_state & SS_CANTSENDMORE) 567 snderr(EPIPE); 568 if (so->so_error) { 569 error = so->so_error; 570 so->so_error = 0; 571 splx(s); 572 goto release; 573 } 574 if ((so->so_state & SS_ISCONNECTED) == 0) { 575 /* 576 * `sendto' and `sendmsg' is allowed on a connection- 577 * based socket if it supports implied connect. 578 * Return ENOTCONN if not connected and no address is 579 * supplied. 580 */ 581 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 582 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 583 if ((so->so_state & SS_ISCONFIRMING) == 0 && 584 !(resid == 0 && clen != 0)) 585 snderr(ENOTCONN); 586 } else if (addr == 0) 587 snderr(so->so_proto->pr_flags & PR_CONNREQUIRED ? 588 ENOTCONN : EDESTADDRREQ); 589 } 590 space = sbspace(&so->so_snd); 591 if (flags & MSG_OOB) 592 space += 1024; 593 if ((atomic && resid > so->so_snd.sb_hiwat) || 594 clen > so->so_snd.sb_hiwat) 595 snderr(EMSGSIZE); 596 if (space < resid + clen && 597 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 598 if (so->so_state & SS_NBIO) 599 snderr(EWOULDBLOCK); 600 sbunlock(&so->so_snd); 601 error = sbwait(&so->so_snd); 602 splx(s); 603 if (error) 604 goto out; 605 goto restart; 606 } 607 splx(s); 608 mp = ⊤ 609 space -= clen; 610 do { 611 if (uio == NULL) { 612 /* 613 * Data is prepackaged in "top". 614 */ 615 resid = 0; 616 if (flags & MSG_EOR) 617 top->m_flags |= M_EOR; 618 } else do { 619 #ifdef ZERO_COPY_SOCKETS 620 cow_send = 0; 621 #endif /* ZERO_COPY_SOCKETS */ 622 if (top == 0) { 623 MGETHDR(m, M_TRYWAIT, MT_DATA); 624 if (m == NULL) { 625 error = ENOBUFS; 626 goto release; 627 } 628 mlen = MHLEN; 629 m->m_pkthdr.len = 0; 630 m->m_pkthdr.rcvif = (struct ifnet *)0; 631 } else { 632 MGET(m, M_TRYWAIT, MT_DATA); 633 if (m == NULL) { 634 error = ENOBUFS; 635 goto release; 636 } 637 mlen = MLEN; 638 } 639 if (resid >= MINCLSIZE) { 640 #ifdef ZERO_COPY_SOCKETS 641 if (so_zero_copy_send && 642 resid>=PAGE_SIZE && 643 space>=PAGE_SIZE && 644 uio->uio_iov->iov_len>=PAGE_SIZE) { 645 so_zerocp_stats.size_ok++; 646 if (!((vm_offset_t) 647 uio->uio_iov->iov_base & PAGE_MASK)){ 648 so_zerocp_stats.align_ok++; 649 cow_send = socow_setup(m, uio); 650 } 651 } 652 if (!cow_send){ 653 #endif /* ZERO_COPY_SOCKETS */ 654 MCLGET(m, M_TRYWAIT); 655 if ((m->m_flags & M_EXT) == 0) 656 goto nopages; 657 mlen = MCLBYTES; 658 len = min(min(mlen, resid), space); 659 } else { 660 #ifdef ZERO_COPY_SOCKETS 661 len = PAGE_SIZE; 662 } 663 664 } else { 665 #endif /* ZERO_COPY_SOCKETS */ 666 nopages: 667 len = min(min(mlen, resid), space); 668 /* 669 * For datagram protocols, leave room 670 * for protocol headers in first mbuf. 671 */ 672 if (atomic && top == 0 && len < mlen) 673 MH_ALIGN(m, len); 674 } 675 space -= len; 676 #ifdef ZERO_COPY_SOCKETS 677 if (cow_send) 678 error = 0; 679 else 680 #endif /* ZERO_COPY_SOCKETS */ 681 error = uiomove(mtod(m, caddr_t), (int)len, uio); 682 resid = uio->uio_resid; 683 m->m_len = len; 684 *mp = m; 685 top->m_pkthdr.len += len; 686 if (error) 687 goto release; 688 mp = &m->m_next; 689 if (resid <= 0) { 690 if (flags & MSG_EOR) 691 top->m_flags |= M_EOR; 692 break; 693 } 694 } while (space > 0 && atomic); 695 if (dontroute) 696 so->so_options |= SO_DONTROUTE; 697 s = splnet(); /* XXX */ 698 /* 699 * XXX all the SS_CANTSENDMORE checks previously 700 * done could be out of date. We could have recieved 701 * a reset packet in an interrupt or maybe we slept 702 * while doing page faults in uiomove() etc. We could 703 * probably recheck again inside the splnet() protection 704 * here, but there are probably other places that this 705 * also happens. We must rethink this. 706 */ 707 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 708 (flags & MSG_OOB) ? PRUS_OOB : 709 /* 710 * If the user set MSG_EOF, the protocol 711 * understands this flag and nothing left to 712 * send then use PRU_SEND_EOF instead of PRU_SEND. 713 */ 714 ((flags & MSG_EOF) && 715 (so->so_proto->pr_flags & PR_IMPLOPCL) && 716 (resid <= 0)) ? 717 PRUS_EOF : 718 /* If there is more to send set PRUS_MORETOCOME */ 719 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 720 top, addr, control, td); 721 splx(s); 722 if (dontroute) 723 so->so_options &= ~SO_DONTROUTE; 724 clen = 0; 725 control = 0; 726 top = 0; 727 mp = ⊤ 728 if (error) 729 goto release; 730 } while (resid && space > 0); 731 } while (resid); 732 733 release: 734 sbunlock(&so->so_snd); 735 out: 736 if (top) 737 m_freem(top); 738 if (control) 739 m_freem(control); 740 return (error); 741 } 742 743 /* 744 * Implement receive operations on a socket. 745 * We depend on the way that records are added to the sockbuf 746 * by sbappend*. In particular, each record (mbufs linked through m_next) 747 * must begin with an address if the protocol so specifies, 748 * followed by an optional mbuf or mbufs containing ancillary data, 749 * and then zero or more mbufs of data. 750 * In order to avoid blocking network interrupts for the entire time here, 751 * we splx() while doing the actual copy to user space. 752 * Although the sockbuf is locked, new data may still be appended, 753 * and thus we must maintain consistency of the sockbuf during that time. 754 * 755 * The caller may receive the data as a single mbuf chain by supplying 756 * an mbuf **mp0 for use in returning the chain. The uio is then used 757 * only for the count in uio_resid. 758 */ 759 int 760 soreceive(so, psa, uio, mp0, controlp, flagsp) 761 register struct socket *so; 762 struct sockaddr **psa; 763 struct uio *uio; 764 struct mbuf **mp0; 765 struct mbuf **controlp; 766 int *flagsp; 767 { 768 struct mbuf *m, **mp; 769 register int flags, len, error, s, offset; 770 struct protosw *pr = so->so_proto; 771 struct mbuf *nextrecord; 772 int moff, type = 0; 773 int orig_resid = uio->uio_resid; 774 775 mp = mp0; 776 if (psa) 777 *psa = 0; 778 if (controlp) 779 *controlp = 0; 780 if (flagsp) 781 flags = *flagsp &~ MSG_EOR; 782 else 783 flags = 0; 784 if (flags & MSG_OOB) { 785 m = m_get(M_TRYWAIT, MT_DATA); 786 if (m == NULL) 787 return (ENOBUFS); 788 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 789 if (error) 790 goto bad; 791 do { 792 #ifdef ZERO_COPY_SOCKETS 793 if (so_zero_copy_receive) { 794 vm_page_t pg; 795 int disposable; 796 797 if ((m->m_flags & M_EXT) 798 && (m->m_ext.ext_type == EXT_DISPOSABLE)) 799 disposable = 1; 800 else 801 disposable = 0; 802 803 pg = PHYS_TO_VM_PAGE(vtophys(mtod(m, caddr_t))); 804 if (uio->uio_offset == -1) 805 uio->uio_offset =IDX_TO_OFF(pg->pindex); 806 807 error = uiomoveco(mtod(m, caddr_t), 808 min(uio->uio_resid, m->m_len), 809 uio, pg->object, 810 disposable); 811 } else 812 #endif /* ZERO_COPY_SOCKETS */ 813 error = uiomove(mtod(m, caddr_t), 814 (int) min(uio->uio_resid, m->m_len), uio); 815 m = m_free(m); 816 } while (uio->uio_resid && error == 0 && m); 817 bad: 818 if (m) 819 m_freem(m); 820 return (error); 821 } 822 if (mp) 823 *mp = (struct mbuf *)0; 824 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 825 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 826 827 restart: 828 error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); 829 if (error) 830 return (error); 831 s = splnet(); 832 833 m = so->so_rcv.sb_mb; 834 /* 835 * If we have less data than requested, block awaiting more 836 * (subject to any timeout) if: 837 * 1. the current count is less than the low water mark, or 838 * 2. MSG_WAITALL is set, and it is possible to do the entire 839 * receive operation at once if we block (resid <= hiwat). 840 * 3. MSG_DONTWAIT is not set 841 * If MSG_WAITALL is set but resid is larger than the receive buffer, 842 * we have to do the receive in sections, and thus risk returning 843 * a short count if a timeout or signal occurs after we start. 844 */ 845 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 846 so->so_rcv.sb_cc < uio->uio_resid) && 847 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 848 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 849 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 850 KASSERT(m != 0 || !so->so_rcv.sb_cc, 851 ("receive: m == %p so->so_rcv.sb_cc == %u", 852 m, so->so_rcv.sb_cc)); 853 if (so->so_error) { 854 if (m) 855 goto dontblock; 856 error = so->so_error; 857 if ((flags & MSG_PEEK) == 0) 858 so->so_error = 0; 859 goto release; 860 } 861 if (so->so_state & SS_CANTRCVMORE) { 862 if (m) 863 goto dontblock; 864 else 865 goto release; 866 } 867 for (; m; m = m->m_next) 868 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 869 m = so->so_rcv.sb_mb; 870 goto dontblock; 871 } 872 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 873 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 874 error = ENOTCONN; 875 goto release; 876 } 877 if (uio->uio_resid == 0) 878 goto release; 879 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 880 error = EWOULDBLOCK; 881 goto release; 882 } 883 sbunlock(&so->so_rcv); 884 error = sbwait(&so->so_rcv); 885 splx(s); 886 if (error) 887 return (error); 888 goto restart; 889 } 890 dontblock: 891 if (uio->uio_td) 892 uio->uio_td->td_proc->p_stats->p_ru.ru_msgrcv++; 893 nextrecord = m->m_nextpkt; 894 if (pr->pr_flags & PR_ADDR) { 895 KASSERT(m->m_type == MT_SONAME, 896 ("m->m_type == %d", m->m_type)); 897 orig_resid = 0; 898 if (psa) 899 *psa = dup_sockaddr(mtod(m, struct sockaddr *), 900 mp0 == 0); 901 if (flags & MSG_PEEK) { 902 m = m->m_next; 903 } else { 904 sbfree(&so->so_rcv, m); 905 so->so_rcv.sb_mb = m_free(m); 906 m = so->so_rcv.sb_mb; 907 } 908 } 909 while (m && m->m_type == MT_CONTROL && error == 0) { 910 if (flags & MSG_PEEK) { 911 if (controlp) 912 *controlp = m_copy(m, 0, m->m_len); 913 m = m->m_next; 914 } else { 915 sbfree(&so->so_rcv, m); 916 so->so_rcv.sb_mb = m->m_next; 917 m->m_next = NULL; 918 if (pr->pr_domain->dom_externalize) 919 error = 920 (*pr->pr_domain->dom_externalize)(m, controlp); 921 else if (controlp) 922 *controlp = m; 923 else 924 m_freem(m); 925 m = so->so_rcv.sb_mb; 926 } 927 if (controlp) { 928 orig_resid = 0; 929 do 930 controlp = &(*controlp)->m_next; 931 while (*controlp != NULL); 932 } 933 } 934 if (m) { 935 if ((flags & MSG_PEEK) == 0) 936 m->m_nextpkt = nextrecord; 937 type = m->m_type; 938 if (type == MT_OOBDATA) 939 flags |= MSG_OOB; 940 } 941 moff = 0; 942 offset = 0; 943 while (m && uio->uio_resid > 0 && error == 0) { 944 if (m->m_type == MT_OOBDATA) { 945 if (type != MT_OOBDATA) 946 break; 947 } else if (type == MT_OOBDATA) 948 break; 949 else 950 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER, 951 ("m->m_type == %d", m->m_type)); 952 so->so_state &= ~SS_RCVATMARK; 953 len = uio->uio_resid; 954 if (so->so_oobmark && len > so->so_oobmark - offset) 955 len = so->so_oobmark - offset; 956 if (len > m->m_len - moff) 957 len = m->m_len - moff; 958 /* 959 * If mp is set, just pass back the mbufs. 960 * Otherwise copy them out via the uio, then free. 961 * Sockbuf must be consistent here (points to current mbuf, 962 * it points to next record) when we drop priority; 963 * we must note any additions to the sockbuf when we 964 * block interrupts again. 965 */ 966 if (mp == 0) { 967 splx(s); 968 #ifdef ZERO_COPY_SOCKETS 969 if (so_zero_copy_receive) { 970 vm_page_t pg; 971 int disposable; 972 973 if ((m->m_flags & M_EXT) 974 && (m->m_ext.ext_type == EXT_DISPOSABLE)) 975 disposable = 1; 976 else 977 disposable = 0; 978 979 pg = PHYS_TO_VM_PAGE(vtophys(mtod(m, caddr_t) + 980 moff)); 981 982 if (uio->uio_offset == -1) 983 uio->uio_offset =IDX_TO_OFF(pg->pindex); 984 985 error = uiomoveco(mtod(m, caddr_t) + moff, 986 (int)len, uio,pg->object, 987 disposable); 988 } else 989 #endif /* ZERO_COPY_SOCKETS */ 990 error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio); 991 s = splnet(); 992 if (error) 993 goto release; 994 } else 995 uio->uio_resid -= len; 996 if (len == m->m_len - moff) { 997 if (m->m_flags & M_EOR) 998 flags |= MSG_EOR; 999 if (flags & MSG_PEEK) { 1000 m = m->m_next; 1001 moff = 0; 1002 } else { 1003 nextrecord = m->m_nextpkt; 1004 sbfree(&so->so_rcv, m); 1005 if (mp) { 1006 *mp = m; 1007 mp = &m->m_next; 1008 so->so_rcv.sb_mb = m = m->m_next; 1009 *mp = (struct mbuf *)0; 1010 } else { 1011 so->so_rcv.sb_mb = m_free(m); 1012 m = so->so_rcv.sb_mb; 1013 } 1014 if (m) 1015 m->m_nextpkt = nextrecord; 1016 } 1017 } else { 1018 if (flags & MSG_PEEK) 1019 moff += len; 1020 else { 1021 if (mp) 1022 *mp = m_copym(m, 0, len, M_TRYWAIT); 1023 m->m_data += len; 1024 m->m_len -= len; 1025 so->so_rcv.sb_cc -= len; 1026 } 1027 } 1028 if (so->so_oobmark) { 1029 if ((flags & MSG_PEEK) == 0) { 1030 so->so_oobmark -= len; 1031 if (so->so_oobmark == 0) { 1032 so->so_state |= SS_RCVATMARK; 1033 break; 1034 } 1035 } else { 1036 offset += len; 1037 if (offset == so->so_oobmark) 1038 break; 1039 } 1040 } 1041 if (flags & MSG_EOR) 1042 break; 1043 /* 1044 * If the MSG_WAITALL flag is set (for non-atomic socket), 1045 * we must not quit until "uio->uio_resid == 0" or an error 1046 * termination. If a signal/timeout occurs, return 1047 * with a short count but without error. 1048 * Keep sockbuf locked against other readers. 1049 */ 1050 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1051 !sosendallatonce(so) && !nextrecord) { 1052 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1053 break; 1054 /* 1055 * Notify the protocol that some data has been 1056 * drained before blocking. 1057 */ 1058 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1059 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 1060 error = sbwait(&so->so_rcv); 1061 if (error) { 1062 sbunlock(&so->so_rcv); 1063 splx(s); 1064 return (0); 1065 } 1066 m = so->so_rcv.sb_mb; 1067 if (m) 1068 nextrecord = m->m_nextpkt; 1069 } 1070 } 1071 1072 if (m && pr->pr_flags & PR_ATOMIC) { 1073 flags |= MSG_TRUNC; 1074 if ((flags & MSG_PEEK) == 0) 1075 (void) sbdroprecord(&so->so_rcv); 1076 } 1077 if ((flags & MSG_PEEK) == 0) { 1078 if (m == 0) 1079 so->so_rcv.sb_mb = nextrecord; 1080 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1081 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 1082 } 1083 if (orig_resid == uio->uio_resid && orig_resid && 1084 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1085 sbunlock(&so->so_rcv); 1086 splx(s); 1087 goto restart; 1088 } 1089 1090 if (flagsp) 1091 *flagsp |= flags; 1092 release: 1093 sbunlock(&so->so_rcv); 1094 splx(s); 1095 return (error); 1096 } 1097 1098 int 1099 soshutdown(so, how) 1100 register struct socket *so; 1101 register int how; 1102 { 1103 register struct protosw *pr = so->so_proto; 1104 1105 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1106 return (EINVAL); 1107 1108 if (how != SHUT_WR) 1109 sorflush(so); 1110 if (how != SHUT_RD) 1111 return ((*pr->pr_usrreqs->pru_shutdown)(so)); 1112 return (0); 1113 } 1114 1115 void 1116 sorflush(so) 1117 register struct socket *so; 1118 { 1119 register struct sockbuf *sb = &so->so_rcv; 1120 register struct protosw *pr = so->so_proto; 1121 register int s; 1122 struct sockbuf asb; 1123 1124 sb->sb_flags |= SB_NOINTR; 1125 (void) sblock(sb, M_WAITOK); 1126 s = splimp(); 1127 socantrcvmore(so); 1128 sbunlock(sb); 1129 asb = *sb; 1130 bzero(sb, sizeof (*sb)); 1131 splx(s); 1132 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1133 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1134 sbrelease(&asb, so); 1135 } 1136 1137 #ifdef INET 1138 static int 1139 do_setopt_accept_filter(so, sopt) 1140 struct socket *so; 1141 struct sockopt *sopt; 1142 { 1143 struct accept_filter_arg *afap = NULL; 1144 struct accept_filter *afp; 1145 struct so_accf *af = so->so_accf; 1146 int error = 0; 1147 1148 /* do not set/remove accept filters on non listen sockets */ 1149 if ((so->so_options & SO_ACCEPTCONN) == 0) { 1150 error = EINVAL; 1151 goto out; 1152 } 1153 1154 /* removing the filter */ 1155 if (sopt == NULL) { 1156 if (af != NULL) { 1157 if (af->so_accept_filter != NULL && 1158 af->so_accept_filter->accf_destroy != NULL) { 1159 af->so_accept_filter->accf_destroy(so); 1160 } 1161 if (af->so_accept_filter_str != NULL) { 1162 FREE(af->so_accept_filter_str, M_ACCF); 1163 } 1164 FREE(af, M_ACCF); 1165 so->so_accf = NULL; 1166 } 1167 so->so_options &= ~SO_ACCEPTFILTER; 1168 return (0); 1169 } 1170 /* adding a filter */ 1171 /* must remove previous filter first */ 1172 if (af != NULL) { 1173 error = EINVAL; 1174 goto out; 1175 } 1176 /* don't put large objects on the kernel stack */ 1177 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK); 1178 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap); 1179 afap->af_name[sizeof(afap->af_name)-1] = '\0'; 1180 afap->af_arg[sizeof(afap->af_arg)-1] = '\0'; 1181 if (error) 1182 goto out; 1183 afp = accept_filt_get(afap->af_name); 1184 if (afp == NULL) { 1185 error = ENOENT; 1186 goto out; 1187 } 1188 MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO); 1189 if (afp->accf_create != NULL) { 1190 if (afap->af_name[0] != '\0') { 1191 int len = strlen(afap->af_name) + 1; 1192 1193 MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK); 1194 strcpy(af->so_accept_filter_str, afap->af_name); 1195 } 1196 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg); 1197 if (af->so_accept_filter_arg == NULL) { 1198 FREE(af->so_accept_filter_str, M_ACCF); 1199 FREE(af, M_ACCF); 1200 so->so_accf = NULL; 1201 error = EINVAL; 1202 goto out; 1203 } 1204 } 1205 af->so_accept_filter = afp; 1206 so->so_accf = af; 1207 so->so_options |= SO_ACCEPTFILTER; 1208 out: 1209 if (afap != NULL) 1210 FREE(afap, M_TEMP); 1211 return (error); 1212 } 1213 #endif /* INET */ 1214 1215 /* 1216 * Perhaps this routine, and sooptcopyout(), below, ought to come in 1217 * an additional variant to handle the case where the option value needs 1218 * to be some kind of integer, but not a specific size. 1219 * In addition to their use here, these functions are also called by the 1220 * protocol-level pr_ctloutput() routines. 1221 */ 1222 int 1223 sooptcopyin(sopt, buf, len, minlen) 1224 struct sockopt *sopt; 1225 void *buf; 1226 size_t len; 1227 size_t minlen; 1228 { 1229 size_t valsize; 1230 1231 /* 1232 * If the user gives us more than we wanted, we ignore it, 1233 * but if we don't get the minimum length the caller 1234 * wants, we return EINVAL. On success, sopt->sopt_valsize 1235 * is set to however much we actually retrieved. 1236 */ 1237 if ((valsize = sopt->sopt_valsize) < minlen) 1238 return EINVAL; 1239 if (valsize > len) 1240 sopt->sopt_valsize = valsize = len; 1241 1242 if (sopt->sopt_td != 0) 1243 return (copyin(sopt->sopt_val, buf, valsize)); 1244 1245 bcopy(sopt->sopt_val, buf, valsize); 1246 return 0; 1247 } 1248 1249 int 1250 sosetopt(so, sopt) 1251 struct socket *so; 1252 struct sockopt *sopt; 1253 { 1254 int error, optval; 1255 struct linger l; 1256 struct timeval tv; 1257 u_long val; 1258 #ifdef MAC 1259 struct mac extmac; 1260 #endif /* MAC */ 1261 1262 error = 0; 1263 if (sopt->sopt_level != SOL_SOCKET) { 1264 if (so->so_proto && so->so_proto->pr_ctloutput) 1265 return ((*so->so_proto->pr_ctloutput) 1266 (so, sopt)); 1267 error = ENOPROTOOPT; 1268 } else { 1269 switch (sopt->sopt_name) { 1270 #ifdef INET 1271 case SO_ACCEPTFILTER: 1272 error = do_setopt_accept_filter(so, sopt); 1273 if (error) 1274 goto bad; 1275 break; 1276 #endif 1277 case SO_LINGER: 1278 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 1279 if (error) 1280 goto bad; 1281 1282 so->so_linger = l.l_linger; 1283 if (l.l_onoff) 1284 so->so_options |= SO_LINGER; 1285 else 1286 so->so_options &= ~SO_LINGER; 1287 break; 1288 1289 case SO_DEBUG: 1290 case SO_KEEPALIVE: 1291 case SO_DONTROUTE: 1292 case SO_USELOOPBACK: 1293 case SO_BROADCAST: 1294 case SO_REUSEADDR: 1295 case SO_REUSEPORT: 1296 case SO_OOBINLINE: 1297 case SO_TIMESTAMP: 1298 case SO_NOSIGPIPE: 1299 error = sooptcopyin(sopt, &optval, sizeof optval, 1300 sizeof optval); 1301 if (error) 1302 goto bad; 1303 if (optval) 1304 so->so_options |= sopt->sopt_name; 1305 else 1306 so->so_options &= ~sopt->sopt_name; 1307 break; 1308 1309 case SO_SNDBUF: 1310 case SO_RCVBUF: 1311 case SO_SNDLOWAT: 1312 case SO_RCVLOWAT: 1313 error = sooptcopyin(sopt, &optval, sizeof optval, 1314 sizeof optval); 1315 if (error) 1316 goto bad; 1317 1318 /* 1319 * Values < 1 make no sense for any of these 1320 * options, so disallow them. 1321 */ 1322 if (optval < 1) { 1323 error = EINVAL; 1324 goto bad; 1325 } 1326 1327 switch (sopt->sopt_name) { 1328 case SO_SNDBUF: 1329 case SO_RCVBUF: 1330 if (sbreserve(sopt->sopt_name == SO_SNDBUF ? 1331 &so->so_snd : &so->so_rcv, (u_long)optval, 1332 so, curthread) == 0) { 1333 error = ENOBUFS; 1334 goto bad; 1335 } 1336 break; 1337 1338 /* 1339 * Make sure the low-water is never greater than 1340 * the high-water. 1341 */ 1342 case SO_SNDLOWAT: 1343 so->so_snd.sb_lowat = 1344 (optval > so->so_snd.sb_hiwat) ? 1345 so->so_snd.sb_hiwat : optval; 1346 break; 1347 case SO_RCVLOWAT: 1348 so->so_rcv.sb_lowat = 1349 (optval > so->so_rcv.sb_hiwat) ? 1350 so->so_rcv.sb_hiwat : optval; 1351 break; 1352 } 1353 break; 1354 1355 case SO_SNDTIMEO: 1356 case SO_RCVTIMEO: 1357 error = sooptcopyin(sopt, &tv, sizeof tv, 1358 sizeof tv); 1359 if (error) 1360 goto bad; 1361 1362 /* assert(hz > 0); */ 1363 if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz || 1364 tv.tv_usec < 0 || tv.tv_usec >= 1000000) { 1365 error = EDOM; 1366 goto bad; 1367 } 1368 /* assert(tick > 0); */ 1369 /* assert(ULONG_MAX - SHRT_MAX >= 1000000); */ 1370 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick; 1371 if (val > SHRT_MAX) { 1372 error = EDOM; 1373 goto bad; 1374 } 1375 1376 switch (sopt->sopt_name) { 1377 case SO_SNDTIMEO: 1378 so->so_snd.sb_timeo = val; 1379 break; 1380 case SO_RCVTIMEO: 1381 so->so_rcv.sb_timeo = val; 1382 break; 1383 } 1384 break; 1385 case SO_LABEL: 1386 #ifdef MAC 1387 error = sooptcopyin(sopt, &extmac, sizeof extmac, 1388 sizeof extmac); 1389 if (error) 1390 goto bad; 1391 1392 error = mac_setsockopt_label_set( 1393 sopt->sopt_td->td_ucred, so, &extmac); 1394 1395 #else /* MAC */ 1396 error = EOPNOTSUPP; 1397 #endif /* MAC */ 1398 break; 1399 default: 1400 error = ENOPROTOOPT; 1401 break; 1402 } 1403 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1404 (void) ((*so->so_proto->pr_ctloutput) 1405 (so, sopt)); 1406 } 1407 } 1408 bad: 1409 return (error); 1410 } 1411 1412 /* Helper routine for getsockopt */ 1413 int 1414 sooptcopyout(sopt, buf, len) 1415 struct sockopt *sopt; 1416 void *buf; 1417 size_t len; 1418 { 1419 int error; 1420 size_t valsize; 1421 1422 error = 0; 1423 1424 /* 1425 * Documented get behavior is that we always return a value, 1426 * possibly truncated to fit in the user's buffer. 1427 * Traditional behavior is that we always tell the user 1428 * precisely how much we copied, rather than something useful 1429 * like the total amount we had available for her. 1430 * Note that this interface is not idempotent; the entire answer must 1431 * generated ahead of time. 1432 */ 1433 valsize = min(len, sopt->sopt_valsize); 1434 sopt->sopt_valsize = valsize; 1435 if (sopt->sopt_val != 0) { 1436 if (sopt->sopt_td != 0) 1437 error = copyout(buf, sopt->sopt_val, valsize); 1438 else 1439 bcopy(buf, sopt->sopt_val, valsize); 1440 } 1441 return error; 1442 } 1443 1444 int 1445 sogetopt(so, sopt) 1446 struct socket *so; 1447 struct sockopt *sopt; 1448 { 1449 int error, optval; 1450 struct linger l; 1451 struct timeval tv; 1452 #ifdef INET 1453 struct accept_filter_arg *afap; 1454 #endif 1455 #ifdef MAC 1456 struct mac extmac; 1457 #endif /* MAC */ 1458 1459 error = 0; 1460 if (sopt->sopt_level != SOL_SOCKET) { 1461 if (so->so_proto && so->so_proto->pr_ctloutput) { 1462 return ((*so->so_proto->pr_ctloutput) 1463 (so, sopt)); 1464 } else 1465 return (ENOPROTOOPT); 1466 } else { 1467 switch (sopt->sopt_name) { 1468 #ifdef INET 1469 case SO_ACCEPTFILTER: 1470 if ((so->so_options & SO_ACCEPTCONN) == 0) 1471 return (EINVAL); 1472 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), 1473 M_TEMP, M_WAITOK | M_ZERO); 1474 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 1475 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name); 1476 if (so->so_accf->so_accept_filter_str != NULL) 1477 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str); 1478 } 1479 error = sooptcopyout(sopt, afap, sizeof(*afap)); 1480 FREE(afap, M_TEMP); 1481 break; 1482 #endif 1483 1484 case SO_LINGER: 1485 l.l_onoff = so->so_options & SO_LINGER; 1486 l.l_linger = so->so_linger; 1487 error = sooptcopyout(sopt, &l, sizeof l); 1488 break; 1489 1490 case SO_USELOOPBACK: 1491 case SO_DONTROUTE: 1492 case SO_DEBUG: 1493 case SO_KEEPALIVE: 1494 case SO_REUSEADDR: 1495 case SO_REUSEPORT: 1496 case SO_BROADCAST: 1497 case SO_OOBINLINE: 1498 case SO_TIMESTAMP: 1499 case SO_NOSIGPIPE: 1500 optval = so->so_options & sopt->sopt_name; 1501 integer: 1502 error = sooptcopyout(sopt, &optval, sizeof optval); 1503 break; 1504 1505 case SO_TYPE: 1506 optval = so->so_type; 1507 goto integer; 1508 1509 case SO_ERROR: 1510 optval = so->so_error; 1511 so->so_error = 0; 1512 goto integer; 1513 1514 case SO_SNDBUF: 1515 optval = so->so_snd.sb_hiwat; 1516 goto integer; 1517 1518 case SO_RCVBUF: 1519 optval = so->so_rcv.sb_hiwat; 1520 goto integer; 1521 1522 case SO_SNDLOWAT: 1523 optval = so->so_snd.sb_lowat; 1524 goto integer; 1525 1526 case SO_RCVLOWAT: 1527 optval = so->so_rcv.sb_lowat; 1528 goto integer; 1529 1530 case SO_SNDTIMEO: 1531 case SO_RCVTIMEO: 1532 optval = (sopt->sopt_name == SO_SNDTIMEO ? 1533 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1534 1535 tv.tv_sec = optval / hz; 1536 tv.tv_usec = (optval % hz) * tick; 1537 error = sooptcopyout(sopt, &tv, sizeof tv); 1538 break; 1539 case SO_LABEL: 1540 #ifdef MAC 1541 error = mac_getsockopt_label_get( 1542 sopt->sopt_td->td_ucred, so, &extmac); 1543 if (error) 1544 return (error); 1545 error = sooptcopyout(sopt, &extmac, sizeof extmac); 1546 #else /* MAC */ 1547 error = EOPNOTSUPP; 1548 #endif /* MAC */ 1549 break; 1550 case SO_PEERLABEL: 1551 #ifdef MAC 1552 error = mac_getsockopt_peerlabel_get( 1553 sopt->sopt_td->td_ucred, so, &extmac); 1554 if (error) 1555 return (error); 1556 error = sooptcopyout(sopt, &extmac, sizeof extmac); 1557 #else /* MAC */ 1558 error = EOPNOTSUPP; 1559 #endif /* MAC */ 1560 break; 1561 default: 1562 error = ENOPROTOOPT; 1563 break; 1564 } 1565 return (error); 1566 } 1567 } 1568 1569 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */ 1570 int 1571 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 1572 { 1573 struct mbuf *m, *m_prev; 1574 int sopt_size = sopt->sopt_valsize; 1575 1576 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA); 1577 if (m == 0) 1578 return ENOBUFS; 1579 if (sopt_size > MLEN) { 1580 MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT); 1581 if ((m->m_flags & M_EXT) == 0) { 1582 m_free(m); 1583 return ENOBUFS; 1584 } 1585 m->m_len = min(MCLBYTES, sopt_size); 1586 } else { 1587 m->m_len = min(MLEN, sopt_size); 1588 } 1589 sopt_size -= m->m_len; 1590 *mp = m; 1591 m_prev = m; 1592 1593 while (sopt_size) { 1594 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA); 1595 if (m == 0) { 1596 m_freem(*mp); 1597 return ENOBUFS; 1598 } 1599 if (sopt_size > MLEN) { 1600 MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT); 1601 if ((m->m_flags & M_EXT) == 0) { 1602 m_freem(*mp); 1603 return ENOBUFS; 1604 } 1605 m->m_len = min(MCLBYTES, sopt_size); 1606 } else { 1607 m->m_len = min(MLEN, sopt_size); 1608 } 1609 sopt_size -= m->m_len; 1610 m_prev->m_next = m; 1611 m_prev = m; 1612 } 1613 return 0; 1614 } 1615 1616 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */ 1617 int 1618 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 1619 { 1620 struct mbuf *m0 = m; 1621 1622 if (sopt->sopt_val == NULL) 1623 return 0; 1624 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 1625 if (sopt->sopt_td != NULL) { 1626 int error; 1627 1628 error = copyin(sopt->sopt_val, mtod(m, char *), 1629 m->m_len); 1630 if (error != 0) { 1631 m_freem(m0); 1632 return(error); 1633 } 1634 } else 1635 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 1636 sopt->sopt_valsize -= m->m_len; 1637 (caddr_t)sopt->sopt_val += m->m_len; 1638 m = m->m_next; 1639 } 1640 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 1641 panic("ip6_sooptmcopyin"); 1642 return 0; 1643 } 1644 1645 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */ 1646 int 1647 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 1648 { 1649 struct mbuf *m0 = m; 1650 size_t valsize = 0; 1651 1652 if (sopt->sopt_val == NULL) 1653 return 0; 1654 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 1655 if (sopt->sopt_td != NULL) { 1656 int error; 1657 1658 error = copyout(mtod(m, char *), sopt->sopt_val, 1659 m->m_len); 1660 if (error != 0) { 1661 m_freem(m0); 1662 return(error); 1663 } 1664 } else 1665 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 1666 sopt->sopt_valsize -= m->m_len; 1667 (caddr_t)sopt->sopt_val += m->m_len; 1668 valsize += m->m_len; 1669 m = m->m_next; 1670 } 1671 if (m != NULL) { 1672 /* enough soopt buffer should be given from user-land */ 1673 m_freem(m0); 1674 return(EINVAL); 1675 } 1676 sopt->sopt_valsize = valsize; 1677 return 0; 1678 } 1679 1680 void 1681 sohasoutofband(so) 1682 register struct socket *so; 1683 { 1684 if (so->so_sigio != NULL) 1685 pgsigio(&so->so_sigio, SIGURG, 0); 1686 selwakeup(&so->so_rcv.sb_sel); 1687 } 1688 1689 int 1690 sopoll(struct socket *so, int events, struct ucred *active_cred, 1691 struct thread *td) 1692 { 1693 int revents = 0; 1694 int s = splnet(); 1695 1696 if (events & (POLLIN | POLLRDNORM)) 1697 if (soreadable(so)) 1698 revents |= events & (POLLIN | POLLRDNORM); 1699 1700 if (events & POLLINIGNEOF) 1701 if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat || 1702 !TAILQ_EMPTY(&so->so_comp) || so->so_error) 1703 revents |= POLLINIGNEOF; 1704 1705 if (events & (POLLOUT | POLLWRNORM)) 1706 if (sowriteable(so)) 1707 revents |= events & (POLLOUT | POLLWRNORM); 1708 1709 if (events & (POLLPRI | POLLRDBAND)) 1710 if (so->so_oobmark || (so->so_state & SS_RCVATMARK)) 1711 revents |= events & (POLLPRI | POLLRDBAND); 1712 1713 if (revents == 0) { 1714 if (events & 1715 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM | 1716 POLLRDBAND)) { 1717 selrecord(td, &so->so_rcv.sb_sel); 1718 so->so_rcv.sb_flags |= SB_SEL; 1719 } 1720 1721 if (events & (POLLOUT | POLLWRNORM)) { 1722 selrecord(td, &so->so_snd.sb_sel); 1723 so->so_snd.sb_flags |= SB_SEL; 1724 } 1725 } 1726 1727 splx(s); 1728 return (revents); 1729 } 1730 1731 int 1732 sokqfilter(struct file *fp, struct knote *kn) 1733 { 1734 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1735 struct sockbuf *sb; 1736 int s; 1737 1738 switch (kn->kn_filter) { 1739 case EVFILT_READ: 1740 if (so->so_options & SO_ACCEPTCONN) 1741 kn->kn_fop = &solisten_filtops; 1742 else 1743 kn->kn_fop = &soread_filtops; 1744 sb = &so->so_rcv; 1745 break; 1746 case EVFILT_WRITE: 1747 kn->kn_fop = &sowrite_filtops; 1748 sb = &so->so_snd; 1749 break; 1750 default: 1751 return (1); 1752 } 1753 1754 s = splnet(); 1755 SLIST_INSERT_HEAD(&sb->sb_sel.si_note, kn, kn_selnext); 1756 sb->sb_flags |= SB_KNOTE; 1757 splx(s); 1758 return (0); 1759 } 1760 1761 static void 1762 filt_sordetach(struct knote *kn) 1763 { 1764 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1765 int s = splnet(); 1766 1767 SLIST_REMOVE(&so->so_rcv.sb_sel.si_note, kn, knote, kn_selnext); 1768 if (SLIST_EMPTY(&so->so_rcv.sb_sel.si_note)) 1769 so->so_rcv.sb_flags &= ~SB_KNOTE; 1770 splx(s); 1771 } 1772 1773 /*ARGSUSED*/ 1774 static int 1775 filt_soread(struct knote *kn, long hint) 1776 { 1777 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1778 1779 kn->kn_data = so->so_rcv.sb_cc; 1780 if (so->so_state & SS_CANTRCVMORE) { 1781 kn->kn_flags |= EV_EOF; 1782 kn->kn_fflags = so->so_error; 1783 return (1); 1784 } 1785 if (so->so_error) /* temporary udp error */ 1786 return (1); 1787 if (kn->kn_sfflags & NOTE_LOWAT) 1788 return (kn->kn_data >= kn->kn_sdata); 1789 return (kn->kn_data >= so->so_rcv.sb_lowat); 1790 } 1791 1792 static void 1793 filt_sowdetach(struct knote *kn) 1794 { 1795 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1796 int s = splnet(); 1797 1798 SLIST_REMOVE(&so->so_snd.sb_sel.si_note, kn, knote, kn_selnext); 1799 if (SLIST_EMPTY(&so->so_snd.sb_sel.si_note)) 1800 so->so_snd.sb_flags &= ~SB_KNOTE; 1801 splx(s); 1802 } 1803 1804 /*ARGSUSED*/ 1805 static int 1806 filt_sowrite(struct knote *kn, long hint) 1807 { 1808 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1809 1810 kn->kn_data = sbspace(&so->so_snd); 1811 if (so->so_state & SS_CANTSENDMORE) { 1812 kn->kn_flags |= EV_EOF; 1813 kn->kn_fflags = so->so_error; 1814 return (1); 1815 } 1816 if (so->so_error) /* temporary udp error */ 1817 return (1); 1818 if (((so->so_state & SS_ISCONNECTED) == 0) && 1819 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1820 return (0); 1821 if (kn->kn_sfflags & NOTE_LOWAT) 1822 return (kn->kn_data >= kn->kn_sdata); 1823 return (kn->kn_data >= so->so_snd.sb_lowat); 1824 } 1825 1826 /*ARGSUSED*/ 1827 static int 1828 filt_solisten(struct knote *kn, long hint) 1829 { 1830 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1831 1832 kn->kn_data = so->so_qlen; 1833 return (! TAILQ_EMPTY(&so->so_comp)); 1834 } 1835 1836 int 1837 socheckuid(struct socket *so, uid_t uid) 1838 { 1839 1840 if (so == NULL) 1841 return (EPERM); 1842 if (so->so_cred->cr_uid == uid) 1843 return (0); 1844 return (EPERM); 1845 } 1846