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 */ 35 36 #include <sys/cdefs.h> 37 __FBSDID("$FreeBSD$"); 38 39 #include "opt_inet.h" 40 #include "opt_mac.h" 41 #include "opt_zero.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/fcntl.h> 46 #include <sys/limits.h> 47 #include <sys/lock.h> 48 #include <sys/mac.h> 49 #include <sys/malloc.h> 50 #include <sys/mbuf.h> 51 #include <sys/mutex.h> 52 #include <sys/domain.h> 53 #include <sys/file.h> /* for struct knote */ 54 #include <sys/kernel.h> 55 #include <sys/event.h> 56 #include <sys/poll.h> 57 #include <sys/proc.h> 58 #include <sys/protosw.h> 59 #include <sys/socket.h> 60 #include <sys/socketvar.h> 61 #include <sys/resourcevar.h> 62 #include <sys/signalvar.h> 63 #include <sys/sysctl.h> 64 #include <sys/uio.h> 65 #include <sys/jail.h> 66 67 #include <vm/uma.h> 68 69 70 #ifdef INET 71 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt); 72 #endif 73 74 static void filt_sordetach(struct knote *kn); 75 static int filt_soread(struct knote *kn, long hint); 76 static void filt_sowdetach(struct knote *kn); 77 static int filt_sowrite(struct knote *kn, long hint); 78 static int filt_solisten(struct knote *kn, long hint); 79 80 static struct filterops solisten_filtops = 81 { 1, NULL, filt_sordetach, filt_solisten }; 82 static struct filterops soread_filtops = 83 { 1, NULL, filt_sordetach, filt_soread }; 84 static struct filterops sowrite_filtops = 85 { 1, NULL, filt_sowdetach, filt_sowrite }; 86 87 uma_zone_t socket_zone; 88 so_gen_t so_gencnt; /* generation count for sockets */ 89 90 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 91 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 92 93 SYSCTL_DECL(_kern_ipc); 94 95 static int somaxconn = SOMAXCONN; 96 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW, 97 &somaxconn, 0, "Maximum pending socket connection queue size"); 98 static int numopensockets; 99 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD, 100 &numopensockets, 0, "Number of open sockets"); 101 #ifdef ZERO_COPY_SOCKETS 102 /* These aren't static because they're used in other files. */ 103 int so_zero_copy_send = 1; 104 int so_zero_copy_receive = 1; 105 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0, 106 "Zero copy controls"); 107 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW, 108 &so_zero_copy_receive, 0, "Enable zero copy receive"); 109 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW, 110 &so_zero_copy_send, 0, "Enable zero copy send"); 111 #endif /* ZERO_COPY_SOCKETS */ 112 113 114 /* 115 * Socket operation routines. 116 * These routines are called by the routines in 117 * sys_socket.c or from a system process, and 118 * implement the semantics of socket operations by 119 * switching out to the protocol specific routines. 120 */ 121 122 /* 123 * Get a socket structure from our zone, and initialize it. 124 * Note that it would probably be better to allocate socket 125 * and PCB at the same time, but I'm not convinced that all 126 * the protocols can be easily modified to do this. 127 * 128 * soalloc() returns a socket with a ref count of 0. 129 */ 130 struct socket * 131 soalloc(int mflags) 132 { 133 struct socket *so; 134 #ifdef MAC 135 int error; 136 #endif 137 138 so = uma_zalloc(socket_zone, mflags | M_ZERO); 139 if (so) { 140 #ifdef MAC 141 error = mac_init_socket(so, mflags); 142 if (error != 0) { 143 uma_zfree(socket_zone, so); 144 so = NULL; 145 return so; 146 } 147 #endif 148 /* XXX race condition for reentrant kernel */ 149 so->so_gencnt = ++so_gencnt; 150 /* sx_init(&so->so_sxlock, "socket sxlock"); */ 151 TAILQ_INIT(&so->so_aiojobq); 152 ++numopensockets; 153 } 154 return so; 155 } 156 157 /* 158 * socreate returns a socket with a ref count of 1. The socket should be 159 * closed with soclose(). 160 */ 161 int 162 socreate(dom, aso, type, proto, cred, td) 163 int dom; 164 struct socket **aso; 165 int type; 166 int proto; 167 struct ucred *cred; 168 struct thread *td; 169 { 170 struct protosw *prp; 171 struct socket *so; 172 int error; 173 174 if (proto) 175 prp = pffindproto(dom, proto, type); 176 else 177 prp = pffindtype(dom, type); 178 179 if (prp == 0 || prp->pr_usrreqs->pru_attach == 0) 180 return (EPROTONOSUPPORT); 181 182 if (jailed(cred) && jail_socket_unixiproute_only && 183 prp->pr_domain->dom_family != PF_LOCAL && 184 prp->pr_domain->dom_family != PF_INET && 185 prp->pr_domain->dom_family != PF_ROUTE) { 186 return (EPROTONOSUPPORT); 187 } 188 189 if (prp->pr_type != type) 190 return (EPROTOTYPE); 191 so = soalloc(M_WAITOK); 192 if (so == NULL) 193 return (ENOBUFS); 194 195 TAILQ_INIT(&so->so_incomp); 196 TAILQ_INIT(&so->so_comp); 197 so->so_type = type; 198 so->so_cred = crhold(cred); 199 so->so_proto = prp; 200 #ifdef MAC 201 mac_create_socket(cred, so); 202 #endif 203 soref(so); 204 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); 205 if (error) { 206 so->so_state |= SS_NOFDREF; 207 sorele(so); 208 return (error); 209 } 210 *aso = so; 211 return (0); 212 } 213 214 int 215 sobind(so, nam, td) 216 struct socket *so; 217 struct sockaddr *nam; 218 struct thread *td; 219 { 220 int s = splnet(); 221 int error; 222 223 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td); 224 splx(s); 225 return (error); 226 } 227 228 void 229 sodealloc(struct socket *so) 230 { 231 232 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); 233 so->so_gencnt = ++so_gencnt; 234 if (so->so_rcv.sb_hiwat) 235 (void)chgsbsize(so->so_cred->cr_uidinfo, 236 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 237 if (so->so_snd.sb_hiwat) 238 (void)chgsbsize(so->so_cred->cr_uidinfo, 239 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 240 #ifdef INET 241 /* remove acccept filter if one is present. */ 242 if (so->so_accf != NULL) 243 do_setopt_accept_filter(so, NULL); 244 #endif 245 #ifdef MAC 246 mac_destroy_socket(so); 247 #endif 248 crfree(so->so_cred); 249 /* sx_destroy(&so->so_sxlock); */ 250 uma_zfree(socket_zone, so); 251 --numopensockets; 252 } 253 254 int 255 solisten(so, backlog, td) 256 struct socket *so; 257 int backlog; 258 struct thread *td; 259 { 260 int s, error; 261 262 s = splnet(); 263 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 264 SS_ISDISCONNECTING)) { 265 splx(s); 266 return (EINVAL); 267 } 268 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, td); 269 if (error) { 270 splx(s); 271 return (error); 272 } 273 if (TAILQ_EMPTY(&so->so_comp)) 274 so->so_options |= SO_ACCEPTCONN; 275 if (backlog < 0 || backlog > somaxconn) 276 backlog = somaxconn; 277 so->so_qlimit = backlog; 278 splx(s); 279 return (0); 280 } 281 282 void 283 sofree(so) 284 struct socket *so; 285 { 286 struct socket *head = so->so_head; 287 int s; 288 289 KASSERT(so->so_count == 0, ("socket %p so_count not 0", so)); 290 291 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 292 return; 293 if (head != NULL) { 294 if (so->so_state & SS_INCOMP) { 295 TAILQ_REMOVE(&head->so_incomp, so, so_list); 296 head->so_incqlen--; 297 } else if (so->so_state & SS_COMP) { 298 /* 299 * We must not decommission a socket that's 300 * on the accept(2) queue. If we do, then 301 * accept(2) may hang after select(2) indicated 302 * that the listening socket was ready. 303 */ 304 return; 305 } else { 306 panic("sofree: not queued"); 307 } 308 so->so_state &= ~SS_INCOMP; 309 so->so_head = NULL; 310 } 311 so->so_snd.sb_flags |= SB_NOINTR; 312 (void)sblock(&so->so_snd, M_WAITOK); 313 s = splimp(); 314 socantsendmore(so); 315 splx(s); 316 sbunlock(&so->so_snd); 317 sbrelease(&so->so_snd, so); 318 sorflush(so); 319 sodealloc(so); 320 } 321 322 /* 323 * Close a socket on last file table reference removal. 324 * Initiate disconnect if connected. 325 * Free socket when disconnect complete. 326 * 327 * This function will sorele() the socket. Note that soclose() may be 328 * called prior to the ref count reaching zero. The actual socket 329 * structure will not be freed until the ref count reaches zero. 330 */ 331 int 332 soclose(so) 333 struct socket *so; 334 { 335 int s = splnet(); /* conservative */ 336 int error = 0; 337 338 funsetown(&so->so_sigio); 339 if (so->so_options & SO_ACCEPTCONN) { 340 struct socket *sp, *sonext; 341 342 sp = TAILQ_FIRST(&so->so_incomp); 343 for (; sp != NULL; sp = sonext) { 344 sonext = TAILQ_NEXT(sp, so_list); 345 (void) soabort(sp); 346 } 347 for (sp = TAILQ_FIRST(&so->so_comp); sp != NULL; sp = sonext) { 348 sonext = TAILQ_NEXT(sp, so_list); 349 /* Dequeue from so_comp since sofree() won't do it */ 350 TAILQ_REMOVE(&so->so_comp, sp, so_list); 351 so->so_qlen--; 352 sp->so_state &= ~SS_COMP; 353 sp->so_head = NULL; 354 (void) soabort(sp); 355 } 356 } 357 if (so->so_pcb == 0) 358 goto discard; 359 if (so->so_state & SS_ISCONNECTED) { 360 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 361 error = sodisconnect(so); 362 if (error) 363 goto drop; 364 } 365 if (so->so_options & SO_LINGER) { 366 if ((so->so_state & SS_ISDISCONNECTING) && 367 (so->so_state & SS_NBIO)) 368 goto drop; 369 while (so->so_state & SS_ISCONNECTED) { 370 error = tsleep(&so->so_timeo, 371 PSOCK | PCATCH, "soclos", so->so_linger * hz); 372 if (error) 373 break; 374 } 375 } 376 } 377 drop: 378 if (so->so_pcb) { 379 int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so); 380 if (error == 0) 381 error = error2; 382 } 383 discard: 384 if (so->so_state & SS_NOFDREF) 385 panic("soclose: NOFDREF"); 386 so->so_state |= SS_NOFDREF; 387 sorele(so); 388 splx(s); 389 return (error); 390 } 391 392 /* 393 * Must be called at splnet... 394 */ 395 int 396 soabort(so) 397 struct socket *so; 398 { 399 int error; 400 401 error = (*so->so_proto->pr_usrreqs->pru_abort)(so); 402 if (error) { 403 sotryfree(so); /* note: does not decrement the ref count */ 404 return error; 405 } 406 return (0); 407 } 408 409 int 410 soaccept(so, nam) 411 struct socket *so; 412 struct sockaddr **nam; 413 { 414 int s = splnet(); 415 int error; 416 417 if ((so->so_state & SS_NOFDREF) == 0) 418 panic("soaccept: !NOFDREF"); 419 so->so_state &= ~SS_NOFDREF; 420 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); 421 splx(s); 422 return (error); 423 } 424 425 int 426 soconnect(so, nam, td) 427 struct socket *so; 428 struct sockaddr *nam; 429 struct thread *td; 430 { 431 int s; 432 int error; 433 434 if (so->so_options & SO_ACCEPTCONN) 435 return (EOPNOTSUPP); 436 s = splnet(); 437 /* 438 * If protocol is connection-based, can only connect once. 439 * Otherwise, if connected, try to disconnect first. 440 * This allows user to disconnect by connecting to, e.g., 441 * a null address. 442 */ 443 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 444 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 445 (error = sodisconnect(so)))) 446 error = EISCONN; 447 else 448 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td); 449 splx(s); 450 return (error); 451 } 452 453 int 454 soconnect2(so1, so2) 455 struct socket *so1; 456 struct socket *so2; 457 { 458 int s = splnet(); 459 int error; 460 461 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); 462 splx(s); 463 return (error); 464 } 465 466 int 467 sodisconnect(so) 468 struct socket *so; 469 { 470 int s = splnet(); 471 int error; 472 473 if ((so->so_state & SS_ISCONNECTED) == 0) { 474 error = ENOTCONN; 475 goto bad; 476 } 477 if (so->so_state & SS_ISDISCONNECTING) { 478 error = EALREADY; 479 goto bad; 480 } 481 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 482 bad: 483 splx(s); 484 return (error); 485 } 486 487 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 488 /* 489 * Send on a socket. 490 * If send must go all at once and message is larger than 491 * send buffering, then hard error. 492 * Lock against other senders. 493 * If must go all at once and not enough room now, then 494 * inform user that this would block and do nothing. 495 * Otherwise, if nonblocking, send as much as possible. 496 * The data to be sent is described by "uio" if nonzero, 497 * otherwise by the mbuf chain "top" (which must be null 498 * if uio is not). Data provided in mbuf chain must be small 499 * enough to send all at once. 500 * 501 * Returns nonzero on error, timeout or signal; callers 502 * must check for short counts if EINTR/ERESTART are returned. 503 * Data and control buffers are freed on return. 504 */ 505 506 #ifdef ZERO_COPY_SOCKETS 507 struct so_zerocopy_stats{ 508 int size_ok; 509 int align_ok; 510 int found_ifp; 511 }; 512 struct so_zerocopy_stats so_zerocp_stats = {0,0,0}; 513 #include <netinet/in.h> 514 #include <net/route.h> 515 #include <netinet/in_pcb.h> 516 #include <vm/vm.h> 517 #include <vm/vm_page.h> 518 #include <vm/vm_object.h> 519 #endif /*ZERO_COPY_SOCKETS*/ 520 521 int 522 sosend(so, addr, uio, top, control, flags, td) 523 struct socket *so; 524 struct sockaddr *addr; 525 struct uio *uio; 526 struct mbuf *top; 527 struct mbuf *control; 528 int flags; 529 struct thread *td; 530 { 531 struct mbuf **mp; 532 struct mbuf *m; 533 long space, len, resid; 534 int clen = 0, error, s, dontroute, mlen; 535 int atomic = sosendallatonce(so) || top; 536 #ifdef ZERO_COPY_SOCKETS 537 int cow_send; 538 #endif /* ZERO_COPY_SOCKETS */ 539 540 if (uio) 541 resid = uio->uio_resid; 542 else 543 resid = top->m_pkthdr.len; 544 /* 545 * In theory resid should be unsigned. 546 * However, space must be signed, as it might be less than 0 547 * if we over-committed, and we must use a signed comparison 548 * of space and resid. On the other hand, a negative resid 549 * causes us to loop sending 0-length segments to the protocol. 550 * 551 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 552 * type sockets since that's an error. 553 */ 554 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 555 error = EINVAL; 556 goto out; 557 } 558 559 dontroute = 560 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 561 (so->so_proto->pr_flags & PR_ATOMIC); 562 if (td) 563 td->td_proc->p_stats->p_ru.ru_msgsnd++; 564 if (control) 565 clen = control->m_len; 566 #define snderr(errno) { error = (errno); splx(s); goto release; } 567 568 restart: 569 error = sblock(&so->so_snd, SBLOCKWAIT(flags)); 570 if (error) 571 goto out; 572 do { 573 s = splnet(); 574 if (so->so_state & SS_CANTSENDMORE) 575 snderr(EPIPE); 576 if (so->so_error) { 577 error = so->so_error; 578 so->so_error = 0; 579 splx(s); 580 goto release; 581 } 582 if ((so->so_state & SS_ISCONNECTED) == 0) { 583 /* 584 * `sendto' and `sendmsg' is allowed on a connection- 585 * based socket if it supports implied connect. 586 * Return ENOTCONN if not connected and no address is 587 * supplied. 588 */ 589 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 590 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 591 if ((so->so_state & SS_ISCONFIRMING) == 0 && 592 !(resid == 0 && clen != 0)) 593 snderr(ENOTCONN); 594 } else if (addr == 0) 595 snderr(so->so_proto->pr_flags & PR_CONNREQUIRED ? 596 ENOTCONN : EDESTADDRREQ); 597 } 598 space = sbspace(&so->so_snd); 599 if (flags & MSG_OOB) 600 space += 1024; 601 if ((atomic && resid > so->so_snd.sb_hiwat) || 602 clen > so->so_snd.sb_hiwat) 603 snderr(EMSGSIZE); 604 if (space < resid + clen && 605 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 606 if (so->so_state & SS_NBIO) 607 snderr(EWOULDBLOCK); 608 sbunlock(&so->so_snd); 609 error = sbwait(&so->so_snd); 610 splx(s); 611 if (error) 612 goto out; 613 goto restart; 614 } 615 splx(s); 616 mp = ⊤ 617 space -= clen; 618 do { 619 if (uio == NULL) { 620 /* 621 * Data is prepackaged in "top". 622 */ 623 resid = 0; 624 if (flags & MSG_EOR) 625 top->m_flags |= M_EOR; 626 } else do { 627 #ifdef ZERO_COPY_SOCKETS 628 cow_send = 0; 629 #endif /* ZERO_COPY_SOCKETS */ 630 if (top == 0) { 631 MGETHDR(m, M_TRYWAIT, MT_DATA); 632 if (m == NULL) { 633 error = ENOBUFS; 634 goto release; 635 } 636 mlen = MHLEN; 637 m->m_pkthdr.len = 0; 638 m->m_pkthdr.rcvif = (struct ifnet *)0; 639 } else { 640 MGET(m, M_TRYWAIT, MT_DATA); 641 if (m == NULL) { 642 error = ENOBUFS; 643 goto release; 644 } 645 mlen = MLEN; 646 } 647 if (resid >= MINCLSIZE) { 648 #ifdef ZERO_COPY_SOCKETS 649 if (so_zero_copy_send && 650 resid>=PAGE_SIZE && 651 space>=PAGE_SIZE && 652 uio->uio_iov->iov_len>=PAGE_SIZE) { 653 so_zerocp_stats.size_ok++; 654 if (!((vm_offset_t) 655 uio->uio_iov->iov_base & PAGE_MASK)){ 656 so_zerocp_stats.align_ok++; 657 cow_send = socow_setup(m, uio); 658 } 659 } 660 if (!cow_send){ 661 #endif /* ZERO_COPY_SOCKETS */ 662 MCLGET(m, M_TRYWAIT); 663 if ((m->m_flags & M_EXT) == 0) 664 goto nopages; 665 mlen = MCLBYTES; 666 len = min(min(mlen, resid), space); 667 } else { 668 #ifdef ZERO_COPY_SOCKETS 669 len = PAGE_SIZE; 670 } 671 672 } else { 673 #endif /* ZERO_COPY_SOCKETS */ 674 nopages: 675 len = min(min(mlen, resid), space); 676 /* 677 * For datagram protocols, leave room 678 * for protocol headers in first mbuf. 679 */ 680 if (atomic && top == 0 && len < mlen) 681 MH_ALIGN(m, len); 682 } 683 space -= len; 684 #ifdef ZERO_COPY_SOCKETS 685 if (cow_send) 686 error = 0; 687 else 688 #endif /* ZERO_COPY_SOCKETS */ 689 error = uiomove(mtod(m, void *), (int)len, uio); 690 resid = uio->uio_resid; 691 m->m_len = len; 692 *mp = m; 693 top->m_pkthdr.len += len; 694 if (error) 695 goto release; 696 mp = &m->m_next; 697 if (resid <= 0) { 698 if (flags & MSG_EOR) 699 top->m_flags |= M_EOR; 700 break; 701 } 702 } while (space > 0 && atomic); 703 if (dontroute) 704 so->so_options |= SO_DONTROUTE; 705 s = splnet(); /* XXX */ 706 /* 707 * XXX all the SS_CANTSENDMORE checks previously 708 * done could be out of date. We could have recieved 709 * a reset packet in an interrupt or maybe we slept 710 * while doing page faults in uiomove() etc. We could 711 * probably recheck again inside the splnet() protection 712 * here, but there are probably other places that this 713 * also happens. We must rethink this. 714 */ 715 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 716 (flags & MSG_OOB) ? PRUS_OOB : 717 /* 718 * If the user set MSG_EOF, the protocol 719 * understands this flag and nothing left to 720 * send then use PRU_SEND_EOF instead of PRU_SEND. 721 */ 722 ((flags & MSG_EOF) && 723 (so->so_proto->pr_flags & PR_IMPLOPCL) && 724 (resid <= 0)) ? 725 PRUS_EOF : 726 /* If there is more to send set PRUS_MORETOCOME */ 727 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 728 top, addr, control, td); 729 splx(s); 730 if (dontroute) 731 so->so_options &= ~SO_DONTROUTE; 732 clen = 0; 733 control = 0; 734 top = 0; 735 mp = ⊤ 736 if (error) 737 goto release; 738 } while (resid && space > 0); 739 } while (resid); 740 741 release: 742 sbunlock(&so->so_snd); 743 out: 744 if (top) 745 m_freem(top); 746 if (control) 747 m_freem(control); 748 return (error); 749 } 750 751 /* 752 * Implement receive operations on a socket. 753 * We depend on the way that records are added to the sockbuf 754 * by sbappend*. In particular, each record (mbufs linked through m_next) 755 * must begin with an address if the protocol so specifies, 756 * followed by an optional mbuf or mbufs containing ancillary data, 757 * and then zero or more mbufs of data. 758 * In order to avoid blocking network interrupts for the entire time here, 759 * we splx() while doing the actual copy to user space. 760 * Although the sockbuf is locked, new data may still be appended, 761 * and thus we must maintain consistency of the sockbuf during that time. 762 * 763 * The caller may receive the data as a single mbuf chain by supplying 764 * an mbuf **mp0 for use in returning the chain. The uio is then used 765 * only for the count in uio_resid. 766 */ 767 int 768 soreceive(so, psa, uio, mp0, controlp, flagsp) 769 struct socket *so; 770 struct sockaddr **psa; 771 struct uio *uio; 772 struct mbuf **mp0; 773 struct mbuf **controlp; 774 int *flagsp; 775 { 776 struct mbuf *m, **mp; 777 int flags, len, error, s, offset; 778 struct protosw *pr = so->so_proto; 779 struct mbuf *nextrecord; 780 int moff, type = 0; 781 int orig_resid = uio->uio_resid; 782 783 mp = mp0; 784 if (psa) 785 *psa = 0; 786 if (controlp) 787 *controlp = 0; 788 if (flagsp) 789 flags = *flagsp &~ MSG_EOR; 790 else 791 flags = 0; 792 if (flags & MSG_OOB) { 793 m = m_get(M_TRYWAIT, MT_DATA); 794 if (m == NULL) 795 return (ENOBUFS); 796 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 797 if (error) 798 goto bad; 799 do { 800 #ifdef ZERO_COPY_SOCKETS 801 if (so_zero_copy_receive) { 802 vm_page_t pg; 803 int disposable; 804 805 if ((m->m_flags & M_EXT) 806 && (m->m_ext.ext_type == EXT_DISPOSABLE)) 807 disposable = 1; 808 else 809 disposable = 0; 810 811 pg = PHYS_TO_VM_PAGE(vtophys(mtod(m, caddr_t))); 812 if (uio->uio_offset == -1) 813 uio->uio_offset =IDX_TO_OFF(pg->pindex); 814 815 error = uiomoveco(mtod(m, void *), 816 min(uio->uio_resid, m->m_len), 817 uio, pg->object, 818 disposable); 819 } else 820 #endif /* ZERO_COPY_SOCKETS */ 821 error = uiomove(mtod(m, void *), 822 (int) min(uio->uio_resid, m->m_len), uio); 823 m = m_free(m); 824 } while (uio->uio_resid && error == 0 && m); 825 bad: 826 if (m) 827 m_freem(m); 828 return (error); 829 } 830 if (mp) 831 *mp = (struct mbuf *)0; 832 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 833 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 834 835 restart: 836 error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); 837 if (error) 838 return (error); 839 s = splnet(); 840 841 m = so->so_rcv.sb_mb; 842 /* 843 * If we have less data than requested, block awaiting more 844 * (subject to any timeout) if: 845 * 1. the current count is less than the low water mark, or 846 * 2. MSG_WAITALL is set, and it is possible to do the entire 847 * receive operation at once if we block (resid <= hiwat). 848 * 3. MSG_DONTWAIT is not set 849 * If MSG_WAITALL is set but resid is larger than the receive buffer, 850 * we have to do the receive in sections, and thus risk returning 851 * a short count if a timeout or signal occurs after we start. 852 */ 853 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 854 so->so_rcv.sb_cc < uio->uio_resid) && 855 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 856 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 857 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 858 KASSERT(m != 0 || !so->so_rcv.sb_cc, 859 ("receive: m == %p so->so_rcv.sb_cc == %u", 860 m, so->so_rcv.sb_cc)); 861 if (so->so_error) { 862 if (m) 863 goto dontblock; 864 error = so->so_error; 865 if ((flags & MSG_PEEK) == 0) 866 so->so_error = 0; 867 goto release; 868 } 869 if (so->so_state & SS_CANTRCVMORE) { 870 if (m) 871 goto dontblock; 872 else 873 goto release; 874 } 875 for (; m; m = m->m_next) 876 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 877 m = so->so_rcv.sb_mb; 878 goto dontblock; 879 } 880 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 881 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 882 error = ENOTCONN; 883 goto release; 884 } 885 if (uio->uio_resid == 0) 886 goto release; 887 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 888 error = EWOULDBLOCK; 889 goto release; 890 } 891 SBLASTRECORDCHK(&so->so_rcv); 892 SBLASTMBUFCHK(&so->so_rcv); 893 sbunlock(&so->so_rcv); 894 error = sbwait(&so->so_rcv); 895 splx(s); 896 if (error) 897 return (error); 898 goto restart; 899 } 900 dontblock: 901 if (uio->uio_td) 902 uio->uio_td->td_proc->p_stats->p_ru.ru_msgrcv++; 903 SBLASTRECORDCHK(&so->so_rcv); 904 SBLASTMBUFCHK(&so->so_rcv); 905 nextrecord = m->m_nextpkt; 906 if (pr->pr_flags & PR_ADDR) { 907 KASSERT(m->m_type == MT_SONAME, 908 ("m->m_type == %d", m->m_type)); 909 orig_resid = 0; 910 if (psa) 911 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 912 mp0 == NULL ? M_WAITOK : M_NOWAIT); 913 if (flags & MSG_PEEK) { 914 m = m->m_next; 915 } else { 916 sbfree(&so->so_rcv, m); 917 so->so_rcv.sb_mb = m_free(m); 918 m = so->so_rcv.sb_mb; 919 } 920 } 921 while (m && m->m_type == MT_CONTROL && error == 0) { 922 if (flags & MSG_PEEK) { 923 if (controlp) 924 *controlp = m_copy(m, 0, m->m_len); 925 m = m->m_next; 926 } else { 927 sbfree(&so->so_rcv, m); 928 so->so_rcv.sb_mb = m->m_next; 929 m->m_next = NULL; 930 if (pr->pr_domain->dom_externalize) 931 error = 932 (*pr->pr_domain->dom_externalize)(m, controlp); 933 else if (controlp) 934 *controlp = m; 935 else 936 m_freem(m); 937 m = so->so_rcv.sb_mb; 938 } 939 if (controlp) { 940 orig_resid = 0; 941 while (*controlp != NULL) 942 controlp = &(*controlp)->m_next; 943 } 944 } 945 if (m) { 946 if ((flags & MSG_PEEK) == 0) { 947 m->m_nextpkt = nextrecord; 948 /* 949 * If nextrecord == NULL (this is a single chain), 950 * then sb_lastrecord may not be valid here if m 951 * was changed earlier. 952 */ 953 if (nextrecord == NULL) { 954 KASSERT(so->so_rcv.sb_mb == m, 955 ("receive tailq 1")); 956 so->so_rcv.sb_lastrecord = m; 957 } 958 } 959 type = m->m_type; 960 if (type == MT_OOBDATA) 961 flags |= MSG_OOB; 962 } else { 963 if ((flags & MSG_PEEK) == 0) { 964 KASSERT(so->so_rcv.sb_mb == m,("receive tailq 2")); 965 so->so_rcv.sb_mb = nextrecord; 966 SB_EMPTY_FIXUP(&so->so_rcv); 967 } 968 } 969 SBLASTRECORDCHK(&so->so_rcv); 970 SBLASTMBUFCHK(&so->so_rcv); 971 972 moff = 0; 973 offset = 0; 974 while (m && uio->uio_resid > 0 && error == 0) { 975 if (m->m_type == MT_OOBDATA) { 976 if (type != MT_OOBDATA) 977 break; 978 } else if (type == MT_OOBDATA) 979 break; 980 else 981 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER, 982 ("m->m_type == %d", m->m_type)); 983 so->so_state &= ~SS_RCVATMARK; 984 len = uio->uio_resid; 985 if (so->so_oobmark && len > so->so_oobmark - offset) 986 len = so->so_oobmark - offset; 987 if (len > m->m_len - moff) 988 len = m->m_len - moff; 989 /* 990 * If mp is set, just pass back the mbufs. 991 * Otherwise copy them out via the uio, then free. 992 * Sockbuf must be consistent here (points to current mbuf, 993 * it points to next record) when we drop priority; 994 * we must note any additions to the sockbuf when we 995 * block interrupts again. 996 */ 997 if (mp == 0) { 998 SBLASTRECORDCHK(&so->so_rcv); 999 SBLASTMBUFCHK(&so->so_rcv); 1000 splx(s); 1001 #ifdef ZERO_COPY_SOCKETS 1002 if (so_zero_copy_receive) { 1003 vm_page_t pg; 1004 int disposable; 1005 1006 if ((m->m_flags & M_EXT) 1007 && (m->m_ext.ext_type == EXT_DISPOSABLE)) 1008 disposable = 1; 1009 else 1010 disposable = 0; 1011 1012 pg = PHYS_TO_VM_PAGE(vtophys(mtod(m, caddr_t) + 1013 moff)); 1014 1015 if (uio->uio_offset == -1) 1016 uio->uio_offset =IDX_TO_OFF(pg->pindex); 1017 1018 error = uiomoveco(mtod(m, char *) + moff, 1019 (int)len, uio,pg->object, 1020 disposable); 1021 } else 1022 #endif /* ZERO_COPY_SOCKETS */ 1023 error = uiomove(mtod(m, char *) + moff, (int)len, uio); 1024 s = splnet(); 1025 if (error) 1026 goto release; 1027 } else 1028 uio->uio_resid -= len; 1029 if (len == m->m_len - moff) { 1030 if (m->m_flags & M_EOR) 1031 flags |= MSG_EOR; 1032 if (flags & MSG_PEEK) { 1033 m = m->m_next; 1034 moff = 0; 1035 } else { 1036 nextrecord = m->m_nextpkt; 1037 sbfree(&so->so_rcv, m); 1038 if (mp) { 1039 *mp = m; 1040 mp = &m->m_next; 1041 so->so_rcv.sb_mb = m = m->m_next; 1042 *mp = (struct mbuf *)0; 1043 } else { 1044 so->so_rcv.sb_mb = m_free(m); 1045 m = so->so_rcv.sb_mb; 1046 } 1047 if (m) { 1048 m->m_nextpkt = nextrecord; 1049 if (nextrecord == NULL) 1050 so->so_rcv.sb_lastrecord = m; 1051 } else { 1052 so->so_rcv.sb_mb = nextrecord; 1053 SB_EMPTY_FIXUP(&so->so_rcv); 1054 } 1055 SBLASTRECORDCHK(&so->so_rcv); 1056 SBLASTMBUFCHK(&so->so_rcv); 1057 } 1058 } else { 1059 if (flags & MSG_PEEK) 1060 moff += len; 1061 else { 1062 if (mp) 1063 *mp = m_copym(m, 0, len, M_TRYWAIT); 1064 m->m_data += len; 1065 m->m_len -= len; 1066 so->so_rcv.sb_cc -= len; 1067 } 1068 } 1069 if (so->so_oobmark) { 1070 if ((flags & MSG_PEEK) == 0) { 1071 so->so_oobmark -= len; 1072 if (so->so_oobmark == 0) { 1073 so->so_state |= SS_RCVATMARK; 1074 break; 1075 } 1076 } else { 1077 offset += len; 1078 if (offset == so->so_oobmark) 1079 break; 1080 } 1081 } 1082 if (flags & MSG_EOR) 1083 break; 1084 /* 1085 * If the MSG_WAITALL flag is set (for non-atomic socket), 1086 * we must not quit until "uio->uio_resid == 0" or an error 1087 * termination. If a signal/timeout occurs, return 1088 * with a short count but without error. 1089 * Keep sockbuf locked against other readers. 1090 */ 1091 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1092 !sosendallatonce(so) && !nextrecord) { 1093 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1094 break; 1095 /* 1096 * Notify the protocol that some data has been 1097 * drained before blocking. 1098 */ 1099 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1100 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 1101 SBLASTRECORDCHK(&so->so_rcv); 1102 SBLASTMBUFCHK(&so->so_rcv); 1103 error = sbwait(&so->so_rcv); 1104 if (error) { 1105 sbunlock(&so->so_rcv); 1106 splx(s); 1107 return (0); 1108 } 1109 m = so->so_rcv.sb_mb; 1110 if (m) 1111 nextrecord = m->m_nextpkt; 1112 } 1113 } 1114 1115 if (m && pr->pr_flags & PR_ATOMIC) { 1116 flags |= MSG_TRUNC; 1117 if ((flags & MSG_PEEK) == 0) 1118 (void) sbdroprecord(&so->so_rcv); 1119 } 1120 if ((flags & MSG_PEEK) == 0) { 1121 if (m == 0) { 1122 /* 1123 * First part is an inline SB_EMPTY_FIXUP(). Second 1124 * part makes sure sb_lastrecord is up-to-date if 1125 * there is still data in the socket buffer. 1126 */ 1127 so->so_rcv.sb_mb = nextrecord; 1128 if (so->so_rcv.sb_mb == NULL) { 1129 so->so_rcv.sb_mbtail = NULL; 1130 so->so_rcv.sb_lastrecord = NULL; 1131 } else if (nextrecord->m_nextpkt == NULL) 1132 so->so_rcv.sb_lastrecord = nextrecord; 1133 } 1134 SBLASTRECORDCHK(&so->so_rcv); 1135 SBLASTMBUFCHK(&so->so_rcv); 1136 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1137 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 1138 } 1139 if (orig_resid == uio->uio_resid && orig_resid && 1140 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1141 sbunlock(&so->so_rcv); 1142 splx(s); 1143 goto restart; 1144 } 1145 1146 if (flagsp) 1147 *flagsp |= flags; 1148 release: 1149 sbunlock(&so->so_rcv); 1150 splx(s); 1151 return (error); 1152 } 1153 1154 int 1155 soshutdown(so, how) 1156 struct socket *so; 1157 int how; 1158 { 1159 struct protosw *pr = so->so_proto; 1160 1161 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1162 return (EINVAL); 1163 1164 if (how != SHUT_WR) 1165 sorflush(so); 1166 if (how != SHUT_RD) 1167 return ((*pr->pr_usrreqs->pru_shutdown)(so)); 1168 return (0); 1169 } 1170 1171 void 1172 sorflush(so) 1173 struct socket *so; 1174 { 1175 struct sockbuf *sb = &so->so_rcv; 1176 struct protosw *pr = so->so_proto; 1177 int s; 1178 struct sockbuf asb; 1179 1180 sb->sb_flags |= SB_NOINTR; 1181 (void) sblock(sb, M_WAITOK); 1182 s = splimp(); 1183 socantrcvmore(so); 1184 sbunlock(sb); 1185 asb = *sb; 1186 /* 1187 * Invalidate/clear most of the sockbuf structure, but keep 1188 * its selinfo structure valid. 1189 */ 1190 bzero(&sb->sb_startzero, 1191 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1192 splx(s); 1193 1194 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1195 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1196 sbrelease(&asb, so); 1197 } 1198 1199 #ifdef INET 1200 static int 1201 do_setopt_accept_filter(so, sopt) 1202 struct socket *so; 1203 struct sockopt *sopt; 1204 { 1205 struct accept_filter_arg *afap = NULL; 1206 struct accept_filter *afp; 1207 struct so_accf *af = so->so_accf; 1208 int error = 0; 1209 1210 /* do not set/remove accept filters on non listen sockets */ 1211 if ((so->so_options & SO_ACCEPTCONN) == 0) { 1212 error = EINVAL; 1213 goto out; 1214 } 1215 1216 /* removing the filter */ 1217 if (sopt == NULL) { 1218 if (af != NULL) { 1219 if (af->so_accept_filter != NULL && 1220 af->so_accept_filter->accf_destroy != NULL) { 1221 af->so_accept_filter->accf_destroy(so); 1222 } 1223 if (af->so_accept_filter_str != NULL) { 1224 FREE(af->so_accept_filter_str, M_ACCF); 1225 } 1226 FREE(af, M_ACCF); 1227 so->so_accf = NULL; 1228 } 1229 so->so_options &= ~SO_ACCEPTFILTER; 1230 return (0); 1231 } 1232 /* adding a filter */ 1233 /* must remove previous filter first */ 1234 if (af != NULL) { 1235 error = EINVAL; 1236 goto out; 1237 } 1238 /* don't put large objects on the kernel stack */ 1239 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK); 1240 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap); 1241 afap->af_name[sizeof(afap->af_name)-1] = '\0'; 1242 afap->af_arg[sizeof(afap->af_arg)-1] = '\0'; 1243 if (error) 1244 goto out; 1245 afp = accept_filt_get(afap->af_name); 1246 if (afp == NULL) { 1247 error = ENOENT; 1248 goto out; 1249 } 1250 MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO); 1251 if (afp->accf_create != NULL) { 1252 if (afap->af_name[0] != '\0') { 1253 int len = strlen(afap->af_name) + 1; 1254 1255 MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK); 1256 strcpy(af->so_accept_filter_str, afap->af_name); 1257 } 1258 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg); 1259 if (af->so_accept_filter_arg == NULL) { 1260 FREE(af->so_accept_filter_str, M_ACCF); 1261 FREE(af, M_ACCF); 1262 so->so_accf = NULL; 1263 error = EINVAL; 1264 goto out; 1265 } 1266 } 1267 af->so_accept_filter = afp; 1268 so->so_accf = af; 1269 so->so_options |= SO_ACCEPTFILTER; 1270 out: 1271 if (afap != NULL) 1272 FREE(afap, M_TEMP); 1273 return (error); 1274 } 1275 #endif /* INET */ 1276 1277 /* 1278 * Perhaps this routine, and sooptcopyout(), below, ought to come in 1279 * an additional variant to handle the case where the option value needs 1280 * to be some kind of integer, but not a specific size. 1281 * In addition to their use here, these functions are also called by the 1282 * protocol-level pr_ctloutput() routines. 1283 */ 1284 int 1285 sooptcopyin(sopt, buf, len, minlen) 1286 struct sockopt *sopt; 1287 void *buf; 1288 size_t len; 1289 size_t minlen; 1290 { 1291 size_t valsize; 1292 1293 /* 1294 * If the user gives us more than we wanted, we ignore it, 1295 * but if we don't get the minimum length the caller 1296 * wants, we return EINVAL. On success, sopt->sopt_valsize 1297 * is set to however much we actually retrieved. 1298 */ 1299 if ((valsize = sopt->sopt_valsize) < minlen) 1300 return EINVAL; 1301 if (valsize > len) 1302 sopt->sopt_valsize = valsize = len; 1303 1304 if (sopt->sopt_td != 0) 1305 return (copyin(sopt->sopt_val, buf, valsize)); 1306 1307 bcopy(sopt->sopt_val, buf, valsize); 1308 return 0; 1309 } 1310 1311 int 1312 sosetopt(so, sopt) 1313 struct socket *so; 1314 struct sockopt *sopt; 1315 { 1316 int error, optval; 1317 struct linger l; 1318 struct timeval tv; 1319 u_long val; 1320 #ifdef MAC 1321 struct mac extmac; 1322 #endif 1323 1324 error = 0; 1325 if (sopt->sopt_level != SOL_SOCKET) { 1326 if (so->so_proto && so->so_proto->pr_ctloutput) 1327 return ((*so->so_proto->pr_ctloutput) 1328 (so, sopt)); 1329 error = ENOPROTOOPT; 1330 } else { 1331 switch (sopt->sopt_name) { 1332 #ifdef INET 1333 case SO_ACCEPTFILTER: 1334 error = do_setopt_accept_filter(so, sopt); 1335 if (error) 1336 goto bad; 1337 break; 1338 #endif 1339 case SO_LINGER: 1340 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 1341 if (error) 1342 goto bad; 1343 1344 so->so_linger = l.l_linger; 1345 if (l.l_onoff) 1346 so->so_options |= SO_LINGER; 1347 else 1348 so->so_options &= ~SO_LINGER; 1349 break; 1350 1351 case SO_DEBUG: 1352 case SO_KEEPALIVE: 1353 case SO_DONTROUTE: 1354 case SO_USELOOPBACK: 1355 case SO_BROADCAST: 1356 case SO_REUSEADDR: 1357 case SO_REUSEPORT: 1358 case SO_OOBINLINE: 1359 case SO_TIMESTAMP: 1360 case SO_BINTIME: 1361 case SO_NOSIGPIPE: 1362 error = sooptcopyin(sopt, &optval, sizeof optval, 1363 sizeof optval); 1364 if (error) 1365 goto bad; 1366 if (optval) 1367 so->so_options |= sopt->sopt_name; 1368 else 1369 so->so_options &= ~sopt->sopt_name; 1370 break; 1371 1372 case SO_SNDBUF: 1373 case SO_RCVBUF: 1374 case SO_SNDLOWAT: 1375 case SO_RCVLOWAT: 1376 error = sooptcopyin(sopt, &optval, sizeof optval, 1377 sizeof optval); 1378 if (error) 1379 goto bad; 1380 1381 /* 1382 * Values < 1 make no sense for any of these 1383 * options, so disallow them. 1384 */ 1385 if (optval < 1) { 1386 error = EINVAL; 1387 goto bad; 1388 } 1389 1390 switch (sopt->sopt_name) { 1391 case SO_SNDBUF: 1392 case SO_RCVBUF: 1393 if (sbreserve(sopt->sopt_name == SO_SNDBUF ? 1394 &so->so_snd : &so->so_rcv, (u_long)optval, 1395 so, curthread) == 0) { 1396 error = ENOBUFS; 1397 goto bad; 1398 } 1399 break; 1400 1401 /* 1402 * Make sure the low-water is never greater than 1403 * the high-water. 1404 */ 1405 case SO_SNDLOWAT: 1406 so->so_snd.sb_lowat = 1407 (optval > so->so_snd.sb_hiwat) ? 1408 so->so_snd.sb_hiwat : optval; 1409 break; 1410 case SO_RCVLOWAT: 1411 so->so_rcv.sb_lowat = 1412 (optval > so->so_rcv.sb_hiwat) ? 1413 so->so_rcv.sb_hiwat : optval; 1414 break; 1415 } 1416 break; 1417 1418 case SO_SNDTIMEO: 1419 case SO_RCVTIMEO: 1420 error = sooptcopyin(sopt, &tv, sizeof tv, 1421 sizeof tv); 1422 if (error) 1423 goto bad; 1424 1425 /* assert(hz > 0); */ 1426 if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz || 1427 tv.tv_usec < 0 || tv.tv_usec >= 1000000) { 1428 error = EDOM; 1429 goto bad; 1430 } 1431 /* assert(tick > 0); */ 1432 /* assert(ULONG_MAX - SHRT_MAX >= 1000000); */ 1433 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick; 1434 if (val > SHRT_MAX) { 1435 error = EDOM; 1436 goto bad; 1437 } 1438 if (val == 0 && tv.tv_usec != 0) 1439 val = 1; 1440 1441 switch (sopt->sopt_name) { 1442 case SO_SNDTIMEO: 1443 so->so_snd.sb_timeo = val; 1444 break; 1445 case SO_RCVTIMEO: 1446 so->so_rcv.sb_timeo = val; 1447 break; 1448 } 1449 break; 1450 case SO_LABEL: 1451 #ifdef MAC 1452 error = sooptcopyin(sopt, &extmac, sizeof extmac, 1453 sizeof extmac); 1454 if (error) 1455 goto bad; 1456 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 1457 so, &extmac); 1458 #else 1459 error = EOPNOTSUPP; 1460 #endif 1461 break; 1462 default: 1463 error = ENOPROTOOPT; 1464 break; 1465 } 1466 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1467 (void) ((*so->so_proto->pr_ctloutput) 1468 (so, sopt)); 1469 } 1470 } 1471 bad: 1472 return (error); 1473 } 1474 1475 /* Helper routine for getsockopt */ 1476 int 1477 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 1478 { 1479 int error; 1480 size_t valsize; 1481 1482 error = 0; 1483 1484 /* 1485 * Documented get behavior is that we always return a value, 1486 * possibly truncated to fit in the user's buffer. 1487 * Traditional behavior is that we always tell the user 1488 * precisely how much we copied, rather than something useful 1489 * like the total amount we had available for her. 1490 * Note that this interface is not idempotent; the entire answer must 1491 * generated ahead of time. 1492 */ 1493 valsize = min(len, sopt->sopt_valsize); 1494 sopt->sopt_valsize = valsize; 1495 if (sopt->sopt_val != 0) { 1496 if (sopt->sopt_td != 0) 1497 error = copyout(buf, sopt->sopt_val, valsize); 1498 else 1499 bcopy(buf, sopt->sopt_val, valsize); 1500 } 1501 return error; 1502 } 1503 1504 int 1505 sogetopt(so, sopt) 1506 struct socket *so; 1507 struct sockopt *sopt; 1508 { 1509 int error, optval; 1510 struct linger l; 1511 struct timeval tv; 1512 #ifdef INET 1513 struct accept_filter_arg *afap; 1514 #endif 1515 #ifdef MAC 1516 struct mac extmac; 1517 #endif 1518 1519 error = 0; 1520 if (sopt->sopt_level != SOL_SOCKET) { 1521 if (so->so_proto && so->so_proto->pr_ctloutput) { 1522 return ((*so->so_proto->pr_ctloutput) 1523 (so, sopt)); 1524 } else 1525 return (ENOPROTOOPT); 1526 } else { 1527 switch (sopt->sopt_name) { 1528 #ifdef INET 1529 case SO_ACCEPTFILTER: 1530 if ((so->so_options & SO_ACCEPTCONN) == 0) 1531 return (EINVAL); 1532 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), 1533 M_TEMP, M_WAITOK | M_ZERO); 1534 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 1535 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name); 1536 if (so->so_accf->so_accept_filter_str != NULL) 1537 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str); 1538 } 1539 error = sooptcopyout(sopt, afap, sizeof(*afap)); 1540 FREE(afap, M_TEMP); 1541 break; 1542 #endif 1543 1544 case SO_LINGER: 1545 l.l_onoff = so->so_options & SO_LINGER; 1546 l.l_linger = so->so_linger; 1547 error = sooptcopyout(sopt, &l, sizeof l); 1548 break; 1549 1550 case SO_USELOOPBACK: 1551 case SO_DONTROUTE: 1552 case SO_DEBUG: 1553 case SO_KEEPALIVE: 1554 case SO_REUSEADDR: 1555 case SO_REUSEPORT: 1556 case SO_BROADCAST: 1557 case SO_OOBINLINE: 1558 case SO_TIMESTAMP: 1559 case SO_BINTIME: 1560 case SO_NOSIGPIPE: 1561 optval = so->so_options & sopt->sopt_name; 1562 integer: 1563 error = sooptcopyout(sopt, &optval, sizeof optval); 1564 break; 1565 1566 case SO_TYPE: 1567 optval = so->so_type; 1568 goto integer; 1569 1570 case SO_ERROR: 1571 optval = so->so_error; 1572 so->so_error = 0; 1573 goto integer; 1574 1575 case SO_SNDBUF: 1576 optval = so->so_snd.sb_hiwat; 1577 goto integer; 1578 1579 case SO_RCVBUF: 1580 optval = so->so_rcv.sb_hiwat; 1581 goto integer; 1582 1583 case SO_SNDLOWAT: 1584 optval = so->so_snd.sb_lowat; 1585 goto integer; 1586 1587 case SO_RCVLOWAT: 1588 optval = so->so_rcv.sb_lowat; 1589 goto integer; 1590 1591 case SO_SNDTIMEO: 1592 case SO_RCVTIMEO: 1593 optval = (sopt->sopt_name == SO_SNDTIMEO ? 1594 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1595 1596 tv.tv_sec = optval / hz; 1597 tv.tv_usec = (optval % hz) * tick; 1598 error = sooptcopyout(sopt, &tv, sizeof tv); 1599 break; 1600 case SO_LABEL: 1601 #ifdef MAC 1602 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 1603 sizeof(extmac)); 1604 if (error) 1605 return (error); 1606 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 1607 so, &extmac); 1608 if (error) 1609 return (error); 1610 error = sooptcopyout(sopt, &extmac, sizeof extmac); 1611 #else 1612 error = EOPNOTSUPP; 1613 #endif 1614 break; 1615 case SO_PEERLABEL: 1616 #ifdef MAC 1617 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 1618 sizeof(extmac)); 1619 if (error) 1620 return (error); 1621 error = mac_getsockopt_peerlabel( 1622 sopt->sopt_td->td_ucred, so, &extmac); 1623 if (error) 1624 return (error); 1625 error = sooptcopyout(sopt, &extmac, sizeof extmac); 1626 #else 1627 error = EOPNOTSUPP; 1628 #endif 1629 break; 1630 default: 1631 error = ENOPROTOOPT; 1632 break; 1633 } 1634 return (error); 1635 } 1636 } 1637 1638 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */ 1639 int 1640 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 1641 { 1642 struct mbuf *m, *m_prev; 1643 int sopt_size = sopt->sopt_valsize; 1644 1645 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA); 1646 if (m == 0) 1647 return ENOBUFS; 1648 if (sopt_size > MLEN) { 1649 MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT); 1650 if ((m->m_flags & M_EXT) == 0) { 1651 m_free(m); 1652 return ENOBUFS; 1653 } 1654 m->m_len = min(MCLBYTES, sopt_size); 1655 } else { 1656 m->m_len = min(MLEN, sopt_size); 1657 } 1658 sopt_size -= m->m_len; 1659 *mp = m; 1660 m_prev = m; 1661 1662 while (sopt_size) { 1663 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA); 1664 if (m == 0) { 1665 m_freem(*mp); 1666 return ENOBUFS; 1667 } 1668 if (sopt_size > MLEN) { 1669 MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT); 1670 if ((m->m_flags & M_EXT) == 0) { 1671 m_freem(m); 1672 m_freem(*mp); 1673 return ENOBUFS; 1674 } 1675 m->m_len = min(MCLBYTES, sopt_size); 1676 } else { 1677 m->m_len = min(MLEN, sopt_size); 1678 } 1679 sopt_size -= m->m_len; 1680 m_prev->m_next = m; 1681 m_prev = m; 1682 } 1683 return 0; 1684 } 1685 1686 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */ 1687 int 1688 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 1689 { 1690 struct mbuf *m0 = m; 1691 1692 if (sopt->sopt_val == NULL) 1693 return 0; 1694 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 1695 if (sopt->sopt_td != NULL) { 1696 int error; 1697 1698 error = copyin(sopt->sopt_val, mtod(m, char *), 1699 m->m_len); 1700 if (error != 0) { 1701 m_freem(m0); 1702 return(error); 1703 } 1704 } else 1705 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 1706 sopt->sopt_valsize -= m->m_len; 1707 (caddr_t)sopt->sopt_val += m->m_len; 1708 m = m->m_next; 1709 } 1710 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 1711 panic("ip6_sooptmcopyin"); 1712 return 0; 1713 } 1714 1715 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */ 1716 int 1717 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 1718 { 1719 struct mbuf *m0 = m; 1720 size_t valsize = 0; 1721 1722 if (sopt->sopt_val == NULL) 1723 return 0; 1724 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 1725 if (sopt->sopt_td != NULL) { 1726 int error; 1727 1728 error = copyout(mtod(m, char *), sopt->sopt_val, 1729 m->m_len); 1730 if (error != 0) { 1731 m_freem(m0); 1732 return(error); 1733 } 1734 } else 1735 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 1736 sopt->sopt_valsize -= m->m_len; 1737 (caddr_t)sopt->sopt_val += m->m_len; 1738 valsize += m->m_len; 1739 m = m->m_next; 1740 } 1741 if (m != NULL) { 1742 /* enough soopt buffer should be given from user-land */ 1743 m_freem(m0); 1744 return(EINVAL); 1745 } 1746 sopt->sopt_valsize = valsize; 1747 return 0; 1748 } 1749 1750 void 1751 sohasoutofband(so) 1752 struct socket *so; 1753 { 1754 if (so->so_sigio != NULL) 1755 pgsigio(&so->so_sigio, SIGURG, 0); 1756 selwakeuppri(&so->so_rcv.sb_sel, PSOCK); 1757 } 1758 1759 int 1760 sopoll(struct socket *so, int events, struct ucred *active_cred, 1761 struct thread *td) 1762 { 1763 int revents = 0; 1764 int s = splnet(); 1765 1766 if (events & (POLLIN | POLLRDNORM)) 1767 if (soreadable(so)) 1768 revents |= events & (POLLIN | POLLRDNORM); 1769 1770 if (events & POLLINIGNEOF) 1771 if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat || 1772 !TAILQ_EMPTY(&so->so_comp) || so->so_error) 1773 revents |= POLLINIGNEOF; 1774 1775 if (events & (POLLOUT | POLLWRNORM)) 1776 if (sowriteable(so)) 1777 revents |= events & (POLLOUT | POLLWRNORM); 1778 1779 if (events & (POLLPRI | POLLRDBAND)) 1780 if (so->so_oobmark || (so->so_state & SS_RCVATMARK)) 1781 revents |= events & (POLLPRI | POLLRDBAND); 1782 1783 if (revents == 0) { 1784 if (events & 1785 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM | 1786 POLLRDBAND)) { 1787 selrecord(td, &so->so_rcv.sb_sel); 1788 so->so_rcv.sb_flags |= SB_SEL; 1789 } 1790 1791 if (events & (POLLOUT | POLLWRNORM)) { 1792 selrecord(td, &so->so_snd.sb_sel); 1793 so->so_snd.sb_flags |= SB_SEL; 1794 } 1795 } 1796 1797 splx(s); 1798 return (revents); 1799 } 1800 1801 int 1802 soo_kqfilter(struct file *fp, struct knote *kn) 1803 { 1804 struct socket *so = kn->kn_fp->f_data; 1805 struct sockbuf *sb; 1806 int s; 1807 1808 switch (kn->kn_filter) { 1809 case EVFILT_READ: 1810 if (so->so_options & SO_ACCEPTCONN) 1811 kn->kn_fop = &solisten_filtops; 1812 else 1813 kn->kn_fop = &soread_filtops; 1814 sb = &so->so_rcv; 1815 break; 1816 case EVFILT_WRITE: 1817 kn->kn_fop = &sowrite_filtops; 1818 sb = &so->so_snd; 1819 break; 1820 default: 1821 return (1); 1822 } 1823 1824 s = splnet(); 1825 SLIST_INSERT_HEAD(&sb->sb_sel.si_note, kn, kn_selnext); 1826 sb->sb_flags |= SB_KNOTE; 1827 splx(s); 1828 return (0); 1829 } 1830 1831 static void 1832 filt_sordetach(struct knote *kn) 1833 { 1834 struct socket *so = kn->kn_fp->f_data; 1835 int s = splnet(); 1836 1837 SLIST_REMOVE(&so->so_rcv.sb_sel.si_note, kn, knote, kn_selnext); 1838 if (SLIST_EMPTY(&so->so_rcv.sb_sel.si_note)) 1839 so->so_rcv.sb_flags &= ~SB_KNOTE; 1840 splx(s); 1841 } 1842 1843 /*ARGSUSED*/ 1844 static int 1845 filt_soread(struct knote *kn, long hint) 1846 { 1847 struct socket *so = kn->kn_fp->f_data; 1848 1849 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl; 1850 if (so->so_state & SS_CANTRCVMORE) { 1851 kn->kn_flags |= EV_EOF; 1852 kn->kn_fflags = so->so_error; 1853 return (1); 1854 } 1855 if (so->so_error) /* temporary udp error */ 1856 return (1); 1857 if (kn->kn_sfflags & NOTE_LOWAT) 1858 return (kn->kn_data >= kn->kn_sdata); 1859 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat); 1860 } 1861 1862 static void 1863 filt_sowdetach(struct knote *kn) 1864 { 1865 struct socket *so = kn->kn_fp->f_data; 1866 int s = splnet(); 1867 1868 SLIST_REMOVE(&so->so_snd.sb_sel.si_note, kn, knote, kn_selnext); 1869 if (SLIST_EMPTY(&so->so_snd.sb_sel.si_note)) 1870 so->so_snd.sb_flags &= ~SB_KNOTE; 1871 splx(s); 1872 } 1873 1874 /*ARGSUSED*/ 1875 static int 1876 filt_sowrite(struct knote *kn, long hint) 1877 { 1878 struct socket *so = kn->kn_fp->f_data; 1879 1880 kn->kn_data = sbspace(&so->so_snd); 1881 if (so->so_state & SS_CANTSENDMORE) { 1882 kn->kn_flags |= EV_EOF; 1883 kn->kn_fflags = so->so_error; 1884 return (1); 1885 } 1886 if (so->so_error) /* temporary udp error */ 1887 return (1); 1888 if (((so->so_state & SS_ISCONNECTED) == 0) && 1889 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1890 return (0); 1891 if (kn->kn_sfflags & NOTE_LOWAT) 1892 return (kn->kn_data >= kn->kn_sdata); 1893 return (kn->kn_data >= so->so_snd.sb_lowat); 1894 } 1895 1896 /*ARGSUSED*/ 1897 static int 1898 filt_solisten(struct knote *kn, long hint) 1899 { 1900 struct socket *so = kn->kn_fp->f_data; 1901 1902 kn->kn_data = so->so_qlen; 1903 return (! TAILQ_EMPTY(&so->so_comp)); 1904 } 1905 1906 int 1907 socheckuid(struct socket *so, uid_t uid) 1908 { 1909 1910 if (so == NULL) 1911 return (EPERM); 1912 if (so->so_cred->cr_uid == uid) 1913 return (0); 1914 return (EPERM); 1915 } 1916