1 /* 2 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * Copyright (c) 1982, 1986, 1988, 1990, 1993 36 * The Regents of the University of California. All rights reserved. 37 * 38 * Redistribution and use in source and binary forms, with or without 39 * modification, are permitted provided that the following conditions 40 * are met: 41 * 1. Redistributions of source code must retain the above copyright 42 * notice, this list of conditions and the following disclaimer. 43 * 2. Redistributions in binary form must reproduce the above copyright 44 * notice, this list of conditions and the following disclaimer in the 45 * documentation and/or other materials provided with the distribution. 46 * 3. All advertising materials mentioning features or use of this software 47 * must display the following acknowledgement: 48 * This product includes software developed by the University of 49 * California, Berkeley and its contributors. 50 * 4. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 67 * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.24 2003/11/11 17:18:18 silby Exp $ 68 * $DragonFly: src/sys/kern/uipc_socket.c,v 1.55 2008/09/02 16:17:52 dillon Exp $ 69 */ 70 71 #include "opt_inet.h" 72 #include "opt_sctp.h" 73 74 #include <sys/param.h> 75 #include <sys/systm.h> 76 #include <sys/fcntl.h> 77 #include <sys/malloc.h> 78 #include <sys/mbuf.h> 79 #include <sys/domain.h> 80 #include <sys/file.h> /* for struct knote */ 81 #include <sys/kernel.h> 82 #include <sys/malloc.h> 83 #include <sys/event.h> 84 #include <sys/proc.h> 85 #include <sys/protosw.h> 86 #include <sys/socket.h> 87 #include <sys/socketvar.h> 88 #include <sys/socketops.h> 89 #include <sys/resourcevar.h> 90 #include <sys/signalvar.h> 91 #include <sys/sysctl.h> 92 #include <sys/uio.h> 93 #include <sys/jail.h> 94 #include <vm/vm_zone.h> 95 #include <vm/pmap.h> 96 97 #include <sys/thread2.h> 98 #include <sys/socketvar2.h> 99 100 #include <machine/limits.h> 101 102 #ifdef INET 103 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt); 104 #endif /* INET */ 105 106 static void filt_sordetach(struct knote *kn); 107 static int filt_soread(struct knote *kn, long hint); 108 static void filt_sowdetach(struct knote *kn); 109 static int filt_sowrite(struct knote *kn, long hint); 110 static int filt_solisten(struct knote *kn, long hint); 111 112 static struct filterops solisten_filtops = 113 { FILTEROP_ISFD, NULL, filt_sordetach, filt_solisten }; 114 static struct filterops soread_filtops = 115 { FILTEROP_ISFD, NULL, filt_sordetach, filt_soread }; 116 static struct filterops sowrite_filtops = 117 { FILTEROP_ISFD, NULL, filt_sowdetach, filt_sowrite }; 118 static struct filterops soexcept_filtops = 119 { FILTEROP_ISFD, NULL, filt_sordetach, filt_soread }; 120 121 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct"); 122 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 123 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 124 125 126 static int somaxconn = SOMAXCONN; 127 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW, 128 &somaxconn, 0, "Maximum pending socket connection queue size"); 129 130 /* 131 * Socket operation routines. 132 * These routines are called by the routines in 133 * sys_socket.c or from a system process, and 134 * implement the semantics of socket operations by 135 * switching out to the protocol specific routines. 136 */ 137 138 /* 139 * Get a socket structure, and initialize it. 140 * Note that it would probably be better to allocate socket 141 * and PCB at the same time, but I'm not convinced that all 142 * the protocols can be easily modified to do this. 143 */ 144 struct socket * 145 soalloc(int waitok) 146 { 147 struct socket *so; 148 unsigned waitmask; 149 150 waitmask = waitok ? M_WAITOK : M_NOWAIT; 151 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask); 152 if (so) { 153 /* XXX race condition for reentrant kernel */ 154 TAILQ_INIT(&so->so_aiojobq); 155 TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist); 156 TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist); 157 lwkt_token_init(&so->so_rcv.ssb_token, 1, "rcvtok"); 158 lwkt_token_init(&so->so_snd.ssb_token, 1, "sndtok"); 159 so->so_state = SS_NOFDREF; 160 so->so_refs = 1; 161 } 162 return so; 163 } 164 165 int 166 socreate(int dom, struct socket **aso, int type, 167 int proto, struct thread *td) 168 { 169 struct proc *p = td->td_proc; 170 struct protosw *prp; 171 struct socket *so; 172 struct pru_attach_info ai; 173 int error; 174 175 if (proto) 176 prp = pffindproto(dom, proto, type); 177 else 178 prp = pffindtype(dom, type); 179 180 if (prp == 0 || prp->pr_usrreqs->pru_attach == 0) 181 return (EPROTONOSUPPORT); 182 183 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only && 184 prp->pr_domain->dom_family != PF_LOCAL && 185 prp->pr_domain->dom_family != PF_INET && 186 prp->pr_domain->dom_family != PF_INET6 && 187 prp->pr_domain->dom_family != PF_ROUTE) { 188 return (EPROTONOSUPPORT); 189 } 190 191 if (prp->pr_type != type) 192 return (EPROTOTYPE); 193 so = soalloc(p != 0); 194 if (so == NULL) 195 return (ENOBUFS); 196 197 /* 198 * Callers of socreate() presumably will connect up a descriptor 199 * and call soclose() if they cannot. This represents our so_refs 200 * (which should be 1) from soalloc(). 201 */ 202 soclrstate(so, SS_NOFDREF); 203 204 /* 205 * Set a default port for protocol processing. No action will occur 206 * on the socket on this port until an inpcb is attached to it and 207 * is able to match incoming packets, or until the socket becomes 208 * available to userland. 209 * 210 * We normally default the socket to the protocol thread on cpu 0. 211 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol 212 * thread and all pr_*()/pru_*() calls are executed synchronously. 213 */ 214 if (prp->pr_flags & PR_SYNC_PORT) 215 so->so_port = &netisr_sync_port; 216 else 217 so->so_port = cpu_portfn(0); 218 219 TAILQ_INIT(&so->so_incomp); 220 TAILQ_INIT(&so->so_comp); 221 so->so_type = type; 222 so->so_cred = crhold(p->p_ucred); 223 so->so_proto = prp; 224 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE]; 225 ai.p_ucred = p->p_ucred; 226 ai.fd_rdir = p->p_fd->fd_rdir; 227 228 /* 229 * Auto-sizing of socket buffers is managed by the protocols and 230 * the appropriate flags must be set in the pru_attach function. 231 */ 232 error = so_pru_attach(so, proto, &ai); 233 if (error) { 234 sosetstate(so, SS_NOFDREF); 235 sofree(so); /* from soalloc */ 236 return error; 237 } 238 239 /* 240 * NOTE: Returns referenced socket. 241 */ 242 *aso = so; 243 return (0); 244 } 245 246 int 247 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 248 { 249 int error; 250 251 error = so_pru_bind(so, nam, td); 252 return (error); 253 } 254 255 static void 256 sodealloc(struct socket *so) 257 { 258 if (so->so_rcv.ssb_hiwat) 259 (void)chgsbsize(so->so_cred->cr_uidinfo, 260 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY); 261 if (so->so_snd.ssb_hiwat) 262 (void)chgsbsize(so->so_cred->cr_uidinfo, 263 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY); 264 #ifdef INET 265 /* remove accept filter if present */ 266 if (so->so_accf != NULL) 267 do_setopt_accept_filter(so, NULL); 268 #endif /* INET */ 269 crfree(so->so_cred); 270 kfree(so, M_SOCKET); 271 } 272 273 int 274 solisten(struct socket *so, int backlog, struct thread *td) 275 { 276 int error; 277 #ifdef SCTP 278 short oldopt, oldqlimit; 279 #endif /* SCTP */ 280 281 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) 282 return (EINVAL); 283 284 #ifdef SCTP 285 oldopt = so->so_options; 286 oldqlimit = so->so_qlimit; 287 #endif /* SCTP */ 288 289 lwkt_gettoken(&so->so_rcv.ssb_token); 290 if (TAILQ_EMPTY(&so->so_comp)) 291 so->so_options |= SO_ACCEPTCONN; 292 lwkt_reltoken(&so->so_rcv.ssb_token); 293 if (backlog < 0 || backlog > somaxconn) 294 backlog = somaxconn; 295 so->so_qlimit = backlog; 296 /* SCTP needs to look at tweak both the inbound backlog parameter AND 297 * the so_options (UDP model both connect's and gets inbound 298 * connections .. implicitly). 299 */ 300 error = so_pru_listen(so, td); 301 if (error) { 302 #ifdef SCTP 303 /* Restore the params */ 304 so->so_options = oldopt; 305 so->so_qlimit = oldqlimit; 306 #endif /* SCTP */ 307 return (error); 308 } 309 return (0); 310 } 311 312 /* 313 * Destroy a disconnected socket. This routine is a NOP if entities 314 * still have a reference on the socket: 315 * 316 * so_pcb - The protocol stack still has a reference 317 * SS_NOFDREF - There is no longer a file pointer reference 318 */ 319 void 320 sofree(struct socket *so) 321 { 322 struct socket *head; 323 324 /* 325 * This is a bit hackish at the moment. We need to interlock 326 * any accept queue we are on before we potentially lose the 327 * last reference to avoid races against a re-reference from 328 * someone operating on the queue. 329 */ 330 while ((head = so->so_head) != NULL) { 331 lwkt_getpooltoken(head); 332 if (so->so_head == head) 333 break; 334 lwkt_relpooltoken(head); 335 } 336 337 /* 338 * Arbitrage the last free. 339 */ 340 KKASSERT(so->so_refs > 0); 341 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) { 342 if (head) 343 lwkt_relpooltoken(head); 344 return; 345 } 346 347 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF)); 348 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0); 349 350 /* 351 * We're done, remove ourselves from the accept queue we are 352 * on, if we are on one. 353 */ 354 if (head != NULL) { 355 if (so->so_state & SS_INCOMP) { 356 TAILQ_REMOVE(&head->so_incomp, so, so_list); 357 head->so_incqlen--; 358 } else if (so->so_state & SS_COMP) { 359 /* 360 * We must not decommission a socket that's 361 * on the accept(2) queue. If we do, then 362 * accept(2) may hang after select(2) indicated 363 * that the listening socket was ready. 364 */ 365 lwkt_relpooltoken(head); 366 return; 367 } else { 368 panic("sofree: not queued"); 369 } 370 soclrstate(so, SS_INCOMP); 371 so->so_head = NULL; 372 lwkt_relpooltoken(head); 373 } 374 ssb_release(&so->so_snd, so); 375 sorflush(so); 376 sodealloc(so); 377 } 378 379 /* 380 * Close a socket on last file table reference removal. 381 * Initiate disconnect if connected. 382 * Free socket when disconnect complete. 383 */ 384 int 385 soclose(struct socket *so, int fflag) 386 { 387 int error = 0; 388 389 funsetown(so->so_sigio); 390 if (so->so_pcb == NULL) 391 goto discard; 392 if (so->so_state & SS_ISCONNECTED) { 393 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 394 error = sodisconnect(so); 395 if (error) 396 goto drop; 397 } 398 if (so->so_options & SO_LINGER) { 399 if ((so->so_state & SS_ISDISCONNECTING) && 400 (fflag & FNONBLOCK)) 401 goto drop; 402 while (so->so_state & SS_ISCONNECTED) { 403 error = tsleep(&so->so_timeo, PCATCH, 404 "soclos", so->so_linger * hz); 405 if (error) 406 break; 407 } 408 } 409 } 410 drop: 411 if (so->so_pcb) { 412 int error2; 413 414 error2 = so_pru_detach(so); 415 if (error == 0) 416 error = error2; 417 } 418 discard: 419 lwkt_getpooltoken(so); 420 if (so->so_options & SO_ACCEPTCONN) { 421 struct socket *sp; 422 423 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) { 424 TAILQ_REMOVE(&so->so_incomp, sp, so_list); 425 soclrstate(sp, SS_INCOMP); 426 sp->so_head = NULL; 427 so->so_incqlen--; 428 soaborta(sp); 429 } 430 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) { 431 TAILQ_REMOVE(&so->so_comp, sp, so_list); 432 soclrstate(sp, SS_COMP); 433 sp->so_head = NULL; 434 so->so_qlen--; 435 soaborta(sp); 436 } 437 } 438 lwkt_relpooltoken(so); 439 if (so->so_state & SS_NOFDREF) 440 panic("soclose: NOFDREF"); 441 sosetstate(so, SS_NOFDREF); /* take ref */ 442 sofree(so); /* dispose of ref */ 443 return (error); 444 } 445 446 /* 447 * Abort and destroy a socket. Only one abort can be in progress 448 * at any given moment. 449 */ 450 void 451 soabort(struct socket *so) 452 { 453 soreference(so); 454 so_pru_abort(so); 455 } 456 457 void 458 soaborta(struct socket *so) 459 { 460 soreference(so); 461 so_pru_aborta(so); 462 } 463 464 void 465 soabort_oncpu(struct socket *so) 466 { 467 soreference(so); 468 so_pru_abort_oncpu(so); 469 } 470 471 /* 472 * so is passed in ref'd, which becomes owned by 473 * the cleared SS_NOFDREF flag. 474 */ 475 int 476 soaccept(struct socket *so, struct sockaddr **nam) 477 { 478 int error; 479 480 if ((so->so_state & SS_NOFDREF) == 0) 481 panic("soaccept: !NOFDREF"); 482 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */ 483 error = so_pru_accept_direct(so, nam); 484 return (error); 485 } 486 487 int 488 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 489 { 490 int error; 491 492 if (so->so_options & SO_ACCEPTCONN) 493 return (EOPNOTSUPP); 494 /* 495 * If protocol is connection-based, can only connect once. 496 * Otherwise, if connected, try to disconnect first. 497 * This allows user to disconnect by connecting to, e.g., 498 * a null address. 499 */ 500 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 501 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 502 (error = sodisconnect(so)))) { 503 error = EISCONN; 504 } else { 505 /* 506 * Prevent accumulated error from previous connection 507 * from biting us. 508 */ 509 so->so_error = 0; 510 error = so_pru_connect(so, nam, td); 511 } 512 return (error); 513 } 514 515 int 516 soconnect2(struct socket *so1, struct socket *so2) 517 { 518 int error; 519 520 error = so_pru_connect2(so1, so2); 521 return (error); 522 } 523 524 int 525 sodisconnect(struct socket *so) 526 { 527 int error; 528 529 if ((so->so_state & SS_ISCONNECTED) == 0) { 530 error = ENOTCONN; 531 goto bad; 532 } 533 if (so->so_state & SS_ISDISCONNECTING) { 534 error = EALREADY; 535 goto bad; 536 } 537 error = so_pru_disconnect(so); 538 bad: 539 return (error); 540 } 541 542 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 543 /* 544 * Send on a socket. 545 * If send must go all at once and message is larger than 546 * send buffering, then hard error. 547 * Lock against other senders. 548 * If must go all at once and not enough room now, then 549 * inform user that this would block and do nothing. 550 * Otherwise, if nonblocking, send as much as possible. 551 * The data to be sent is described by "uio" if nonzero, 552 * otherwise by the mbuf chain "top" (which must be null 553 * if uio is not). Data provided in mbuf chain must be small 554 * enough to send all at once. 555 * 556 * Returns nonzero on error, timeout or signal; callers 557 * must check for short counts if EINTR/ERESTART are returned. 558 * Data and control buffers are freed on return. 559 */ 560 int 561 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 562 struct mbuf *top, struct mbuf *control, int flags, 563 struct thread *td) 564 { 565 struct mbuf **mp; 566 struct mbuf *m; 567 size_t resid; 568 int space, len; 569 int clen = 0, error, dontroute, mlen; 570 int atomic = sosendallatonce(so) || top; 571 int pru_flags; 572 573 if (uio) { 574 resid = uio->uio_resid; 575 } else { 576 resid = (size_t)top->m_pkthdr.len; 577 #ifdef INVARIANTS 578 len = 0; 579 for (m = top; m; m = m->m_next) 580 len += m->m_len; 581 KKASSERT(top->m_pkthdr.len == len); 582 #endif 583 } 584 585 /* 586 * WARNING! resid is unsigned, space and len are signed. space 587 * can wind up negative if the sockbuf is overcommitted. 588 * 589 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 590 * type sockets since that's an error. 591 */ 592 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) { 593 error = EINVAL; 594 goto out; 595 } 596 597 dontroute = 598 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 599 (so->so_proto->pr_flags & PR_ATOMIC); 600 if (td->td_lwp != NULL) 601 td->td_lwp->lwp_ru.ru_msgsnd++; 602 if (control) 603 clen = control->m_len; 604 #define gotoerr(errcode) { error = errcode; goto release; } 605 606 restart: 607 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags)); 608 if (error) 609 goto out; 610 611 do { 612 if (so->so_state & SS_CANTSENDMORE) 613 gotoerr(EPIPE); 614 if (so->so_error) { 615 error = so->so_error; 616 so->so_error = 0; 617 goto release; 618 } 619 if ((so->so_state & SS_ISCONNECTED) == 0) { 620 /* 621 * `sendto' and `sendmsg' is allowed on a connection- 622 * based socket if it supports implied connect. 623 * Return ENOTCONN if not connected and no address is 624 * supplied. 625 */ 626 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 627 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 628 if ((so->so_state & SS_ISCONFIRMING) == 0 && 629 !(resid == 0 && clen != 0)) 630 gotoerr(ENOTCONN); 631 } else if (addr == 0) 632 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ? 633 ENOTCONN : EDESTADDRREQ); 634 } 635 if ((atomic && resid > so->so_snd.ssb_hiwat) || 636 clen > so->so_snd.ssb_hiwat) { 637 gotoerr(EMSGSIZE); 638 } 639 space = ssb_space(&so->so_snd); 640 if (flags & MSG_OOB) 641 space += 1024; 642 if ((space < 0 || (size_t)space < resid + clen) && uio && 643 (atomic || space < so->so_snd.ssb_lowat || space < clen)) { 644 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) 645 gotoerr(EWOULDBLOCK); 646 ssb_unlock(&so->so_snd); 647 error = ssb_wait(&so->so_snd); 648 if (error) 649 goto out; 650 goto restart; 651 } 652 mp = ⊤ 653 space -= clen; 654 do { 655 if (uio == NULL) { 656 /* 657 * Data is prepackaged in "top". 658 */ 659 resid = 0; 660 if (flags & MSG_EOR) 661 top->m_flags |= M_EOR; 662 } else do { 663 if (resid > INT_MAX) 664 resid = INT_MAX; 665 m = m_getl((int)resid, MB_WAIT, MT_DATA, 666 top == NULL ? M_PKTHDR : 0, &mlen); 667 if (top == NULL) { 668 m->m_pkthdr.len = 0; 669 m->m_pkthdr.rcvif = NULL; 670 } 671 len = imin((int)szmin(mlen, resid), space); 672 if (resid < MINCLSIZE) { 673 /* 674 * For datagram protocols, leave room 675 * for protocol headers in first mbuf. 676 */ 677 if (atomic && top == 0 && len < mlen) 678 MH_ALIGN(m, len); 679 } 680 space -= len; 681 error = uiomove(mtod(m, caddr_t), (size_t)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 if (flags & MSG_OOB) { 698 pru_flags = PRUS_OOB; 699 } else if ((flags & MSG_EOF) && 700 (so->so_proto->pr_flags & PR_IMPLOPCL) && 701 (resid == 0)) { 702 /* 703 * If the user set MSG_EOF, the protocol 704 * understands this flag and nothing left to 705 * send then use PRU_SEND_EOF instead of PRU_SEND. 706 */ 707 pru_flags = PRUS_EOF; 708 } else if (resid > 0 && space > 0) { 709 /* If there is more to send, set PRUS_MORETOCOME */ 710 pru_flags = PRUS_MORETOCOME; 711 } else { 712 pru_flags = 0; 713 } 714 /* 715 * XXX all the SS_CANTSENDMORE checks previously 716 * done could be out of date. We could have recieved 717 * a reset packet in an interrupt or maybe we slept 718 * while doing page faults in uiomove() etc. We could 719 * probably recheck again inside the splnet() protection 720 * here, but there are probably other places that this 721 * also happens. We must rethink this. 722 */ 723 error = so_pru_send(so, pru_flags, top, addr, control, td); 724 if (dontroute) 725 so->so_options &= ~SO_DONTROUTE; 726 clen = 0; 727 control = 0; 728 top = NULL; 729 mp = ⊤ 730 if (error) 731 goto release; 732 } while (resid && space > 0); 733 } while (resid); 734 735 release: 736 ssb_unlock(&so->so_snd); 737 out: 738 if (top) 739 m_freem(top); 740 if (control) 741 m_freem(control); 742 return (error); 743 } 744 745 /* 746 * A specialization of sosend() for UDP based on protocol-specific knowledge: 747 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that 748 * sosendallatonce() returns true, 749 * the "atomic" variable is true, 750 * and sosendudp() blocks until space is available for the entire send. 751 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or 752 * PR_IMPLOPCL flags set. 753 * UDP has no out-of-band data. 754 * UDP has no control data. 755 * UDP does not support MSG_EOR. 756 */ 757 int 758 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio, 759 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 760 { 761 boolean_t dontroute; /* temporary SO_DONTROUTE setting */ 762 size_t resid; 763 int error; 764 int space; 765 766 if (td->td_lwp != NULL) 767 td->td_lwp->lwp_ru.ru_msgsnd++; 768 if (control) 769 m_freem(control); 770 771 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp")); 772 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len; 773 774 restart: 775 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags)); 776 if (error) 777 goto out; 778 779 if (so->so_state & SS_CANTSENDMORE) 780 gotoerr(EPIPE); 781 if (so->so_error) { 782 error = so->so_error; 783 so->so_error = 0; 784 goto release; 785 } 786 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL) 787 gotoerr(EDESTADDRREQ); 788 if (resid > so->so_snd.ssb_hiwat) 789 gotoerr(EMSGSIZE); 790 space = ssb_space(&so->so_snd); 791 if (uio && (space < 0 || (size_t)space < resid)) { 792 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) 793 gotoerr(EWOULDBLOCK); 794 ssb_unlock(&so->so_snd); 795 error = ssb_wait(&so->so_snd); 796 if (error) 797 goto out; 798 goto restart; 799 } 800 801 if (uio) { 802 top = m_uiomove(uio); 803 if (top == NULL) 804 goto release; 805 } 806 807 dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE); 808 if (dontroute) 809 so->so_options |= SO_DONTROUTE; 810 811 error = so_pru_send(so, 0, top, addr, NULL, td); 812 top = NULL; /* sent or freed in lower layer */ 813 814 if (dontroute) 815 so->so_options &= ~SO_DONTROUTE; 816 817 release: 818 ssb_unlock(&so->so_snd); 819 out: 820 if (top) 821 m_freem(top); 822 return (error); 823 } 824 825 /* 826 * Implement receive operations on a socket. 827 * 828 * We depend on the way that records are added to the signalsockbuf 829 * by sbappend*. In particular, each record (mbufs linked through m_next) 830 * must begin with an address if the protocol so specifies, 831 * followed by an optional mbuf or mbufs containing ancillary data, 832 * and then zero or more mbufs of data. 833 * 834 * Although the signalsockbuf is locked, new data may still be appended. 835 * A token inside the ssb_lock deals with MP issues and still allows 836 * the network to access the socket if we block in a uio. 837 * 838 * The caller may receive the data as a single mbuf chain by supplying 839 * an mbuf **mp0 for use in returning the chain. The uio is then used 840 * only for the count in uio_resid. 841 */ 842 int 843 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 844 struct sockbuf *sio, struct mbuf **controlp, int *flagsp) 845 { 846 struct mbuf *m, *n; 847 struct mbuf *free_chain = NULL; 848 int flags, len, error, offset; 849 struct protosw *pr = so->so_proto; 850 int moff, type = 0; 851 size_t resid, orig_resid; 852 853 if (uio) 854 resid = uio->uio_resid; 855 else 856 resid = (size_t)(sio->sb_climit - sio->sb_cc); 857 orig_resid = resid; 858 859 if (psa) 860 *psa = NULL; 861 if (controlp) 862 *controlp = NULL; 863 if (flagsp) 864 flags = *flagsp &~ MSG_EOR; 865 else 866 flags = 0; 867 if (flags & MSG_OOB) { 868 m = m_get(MB_WAIT, MT_DATA); 869 if (m == NULL) 870 return (ENOBUFS); 871 error = so_pru_rcvoob(so, m, flags & MSG_PEEK); 872 if (error) 873 goto bad; 874 if (sio) { 875 do { 876 sbappend(sio, m); 877 KKASSERT(resid >= (size_t)m->m_len); 878 resid -= (size_t)m->m_len; 879 } while (resid > 0 && m); 880 } else { 881 do { 882 uio->uio_resid = resid; 883 error = uiomove(mtod(m, caddr_t), 884 (int)szmin(resid, m->m_len), 885 uio); 886 resid = uio->uio_resid; 887 m = m_free(m); 888 } while (uio->uio_resid && error == 0 && m); 889 } 890 bad: 891 if (m) 892 m_freem(m); 893 return (error); 894 } 895 if ((so->so_state & SS_ISCONFIRMING) && resid) 896 so_pru_rcvd(so, 0); 897 898 /* 899 * The token interlocks against the protocol thread while 900 * ssb_lock is a blocking lock against other userland entities. 901 */ 902 lwkt_gettoken(&so->so_rcv.ssb_token); 903 restart: 904 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags)); 905 if (error) 906 goto done; 907 908 m = so->so_rcv.ssb_mb; 909 /* 910 * If we have less data than requested, block awaiting more 911 * (subject to any timeout) if: 912 * 1. the current count is less than the low water mark, or 913 * 2. MSG_WAITALL is set, and it is possible to do the entire 914 * receive operation at once if we block (resid <= hiwat). 915 * 3. MSG_DONTWAIT is not set 916 * If MSG_WAITALL is set but resid is larger than the receive buffer, 917 * we have to do the receive in sections, and thus risk returning 918 * a short count if a timeout or signal occurs after we start. 919 */ 920 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 921 (size_t)so->so_rcv.ssb_cc < resid) && 922 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat || 923 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) && 924 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 925 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1")); 926 if (so->so_error) { 927 if (m) 928 goto dontblock; 929 error = so->so_error; 930 if ((flags & MSG_PEEK) == 0) 931 so->so_error = 0; 932 goto release; 933 } 934 if (so->so_state & SS_CANTRCVMORE) { 935 if (m) 936 goto dontblock; 937 else 938 goto release; 939 } 940 for (; m; m = m->m_next) { 941 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 942 m = so->so_rcv.ssb_mb; 943 goto dontblock; 944 } 945 } 946 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 947 (pr->pr_flags & PR_CONNREQUIRED)) { 948 error = ENOTCONN; 949 goto release; 950 } 951 if (resid == 0) 952 goto release; 953 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) { 954 error = EWOULDBLOCK; 955 goto release; 956 } 957 ssb_unlock(&so->so_rcv); 958 error = ssb_wait(&so->so_rcv); 959 if (error) 960 goto done; 961 goto restart; 962 } 963 dontblock: 964 if (uio && uio->uio_td && uio->uio_td->td_proc) 965 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++; 966 967 /* 968 * note: m should be == sb_mb here. Cache the next record while 969 * cleaning up. Note that calling m_free*() will break out critical 970 * section. 971 */ 972 KKASSERT(m == so->so_rcv.ssb_mb); 973 974 /* 975 * Skip any address mbufs prepending the record. 976 */ 977 if (pr->pr_flags & PR_ADDR) { 978 KASSERT(m->m_type == MT_SONAME, ("receive 1a")); 979 orig_resid = 0; 980 if (psa) 981 *psa = dup_sockaddr(mtod(m, struct sockaddr *)); 982 if (flags & MSG_PEEK) 983 m = m->m_next; 984 else 985 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 986 } 987 988 /* 989 * Skip any control mbufs prepending the record. 990 */ 991 #ifdef SCTP 992 if (pr->pr_flags & PR_ADDR_OPT) { 993 /* 994 * For SCTP we may be getting a 995 * whole message OR a partial delivery. 996 */ 997 if (m && m->m_type == MT_SONAME) { 998 orig_resid = 0; 999 if (psa) 1000 *psa = dup_sockaddr(mtod(m, struct sockaddr *)); 1001 if (flags & MSG_PEEK) 1002 m = m->m_next; 1003 else 1004 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1005 } 1006 } 1007 #endif /* SCTP */ 1008 while (m && m->m_type == MT_CONTROL && error == 0) { 1009 if (flags & MSG_PEEK) { 1010 if (controlp) 1011 *controlp = m_copy(m, 0, m->m_len); 1012 m = m->m_next; /* XXX race */ 1013 } else { 1014 if (controlp) { 1015 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL); 1016 if (pr->pr_domain->dom_externalize && 1017 mtod(m, struct cmsghdr *)->cmsg_type == 1018 SCM_RIGHTS) 1019 error = (*pr->pr_domain->dom_externalize)(m); 1020 *controlp = m; 1021 m = n; 1022 } else { 1023 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1024 } 1025 } 1026 if (controlp && *controlp) { 1027 orig_resid = 0; 1028 controlp = &(*controlp)->m_next; 1029 } 1030 } 1031 1032 /* 1033 * flag OOB data. 1034 */ 1035 if (m) { 1036 type = m->m_type; 1037 if (type == MT_OOBDATA) 1038 flags |= MSG_OOB; 1039 } 1040 1041 /* 1042 * Copy to the UIO or mbuf return chain (*mp). 1043 */ 1044 moff = 0; 1045 offset = 0; 1046 while (m && resid > 0 && error == 0) { 1047 if (m->m_type == MT_OOBDATA) { 1048 if (type != MT_OOBDATA) 1049 break; 1050 } else if (type == MT_OOBDATA) 1051 break; 1052 else 1053 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER, 1054 ("receive 3")); 1055 soclrstate(so, SS_RCVATMARK); 1056 len = (resid > INT_MAX) ? INT_MAX : resid; 1057 if (so->so_oobmark && len > so->so_oobmark - offset) 1058 len = so->so_oobmark - offset; 1059 if (len > m->m_len - moff) 1060 len = m->m_len - moff; 1061 1062 /* 1063 * Copy out to the UIO or pass the mbufs back to the SIO. 1064 * The SIO is dealt with when we eat the mbuf, but deal 1065 * with the resid here either way. 1066 */ 1067 if (uio) { 1068 uio->uio_resid = resid; 1069 error = uiomove(mtod(m, caddr_t) + moff, len, uio); 1070 resid = uio->uio_resid; 1071 if (error) 1072 goto release; 1073 } else { 1074 resid -= (size_t)len; 1075 } 1076 1077 /* 1078 * Eat the entire mbuf or just a piece of it 1079 */ 1080 if (len == m->m_len - moff) { 1081 if (m->m_flags & M_EOR) 1082 flags |= MSG_EOR; 1083 #ifdef SCTP 1084 if (m->m_flags & M_NOTIFICATION) 1085 flags |= MSG_NOTIFICATION; 1086 #endif /* SCTP */ 1087 if (flags & MSG_PEEK) { 1088 m = m->m_next; 1089 moff = 0; 1090 } else { 1091 if (sio) { 1092 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL); 1093 sbappend(sio, m); 1094 m = n; 1095 } else { 1096 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1097 } 1098 } 1099 } else { 1100 if (flags & MSG_PEEK) { 1101 moff += len; 1102 } else { 1103 if (sio) { 1104 n = m_copym(m, 0, len, MB_WAIT); 1105 if (n) 1106 sbappend(sio, n); 1107 } 1108 m->m_data += len; 1109 m->m_len -= len; 1110 so->so_rcv.ssb_cc -= len; 1111 } 1112 } 1113 if (so->so_oobmark) { 1114 if ((flags & MSG_PEEK) == 0) { 1115 so->so_oobmark -= len; 1116 if (so->so_oobmark == 0) { 1117 sosetstate(so, SS_RCVATMARK); 1118 break; 1119 } 1120 } else { 1121 offset += len; 1122 if (offset == so->so_oobmark) 1123 break; 1124 } 1125 } 1126 if (flags & MSG_EOR) 1127 break; 1128 /* 1129 * If the MSG_WAITALL flag is set (for non-atomic socket), 1130 * we must not quit until resid == 0 or an error 1131 * termination. If a signal/timeout occurs, return 1132 * with a short count but without error. 1133 * Keep signalsockbuf locked against other readers. 1134 */ 1135 while ((flags & MSG_WAITALL) && m == NULL && 1136 resid > 0 && !sosendallatonce(so) && 1137 so->so_rcv.ssb_mb == NULL) { 1138 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1139 break; 1140 /* 1141 * The window might have closed to zero, make 1142 * sure we send an ack now that we've drained 1143 * the buffer or we might end up blocking until 1144 * the idle takes over (5 seconds). 1145 */ 1146 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1147 so_pru_rcvd(so, flags); 1148 error = ssb_wait(&so->so_rcv); 1149 if (error) { 1150 ssb_unlock(&so->so_rcv); 1151 error = 0; 1152 goto done; 1153 } 1154 m = so->so_rcv.ssb_mb; 1155 } 1156 } 1157 1158 /* 1159 * If an atomic read was requested but unread data still remains 1160 * in the record, set MSG_TRUNC. 1161 */ 1162 if (m && pr->pr_flags & PR_ATOMIC) 1163 flags |= MSG_TRUNC; 1164 1165 /* 1166 * Cleanup. If an atomic read was requested drop any unread data. 1167 */ 1168 if ((flags & MSG_PEEK) == 0) { 1169 if (m && (pr->pr_flags & PR_ATOMIC)) 1170 sbdroprecord(&so->so_rcv.sb); 1171 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1172 so_pru_rcvd(so, flags); 1173 } 1174 1175 if (orig_resid == resid && orig_resid && 1176 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1177 ssb_unlock(&so->so_rcv); 1178 goto restart; 1179 } 1180 1181 if (flagsp) 1182 *flagsp |= flags; 1183 release: 1184 ssb_unlock(&so->so_rcv); 1185 done: 1186 lwkt_reltoken(&so->so_rcv.ssb_token); 1187 if (free_chain) 1188 m_freem(free_chain); 1189 return (error); 1190 } 1191 1192 /* 1193 * Shut a socket down. Note that we do not get a frontend lock as we 1194 * want to be able to shut the socket down even if another thread is 1195 * blocked in a read(), thus waking it up. 1196 */ 1197 int 1198 soshutdown(struct socket *so, int how) 1199 { 1200 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1201 return (EINVAL); 1202 1203 if (how != SHUT_WR) { 1204 /*ssb_lock(&so->so_rcv, M_WAITOK);*/ 1205 sorflush(so); 1206 /*ssb_unlock(&so->so_rcv);*/ 1207 } 1208 if (how != SHUT_RD) 1209 return (so_pru_shutdown(so)); 1210 return (0); 1211 } 1212 1213 void 1214 sorflush(struct socket *so) 1215 { 1216 struct signalsockbuf *ssb = &so->so_rcv; 1217 struct protosw *pr = so->so_proto; 1218 struct signalsockbuf asb; 1219 1220 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR); 1221 1222 lwkt_gettoken(&ssb->ssb_token); 1223 socantrcvmore(so); 1224 asb = *ssb; 1225 1226 /* 1227 * Can't just blow up the ssb structure here 1228 */ 1229 bzero(&ssb->sb, sizeof(ssb->sb)); 1230 ssb->ssb_timeo = 0; 1231 ssb->ssb_unused01 = 0; 1232 ssb->ssb_lowat = 0; 1233 ssb->ssb_hiwat = 0; 1234 ssb->ssb_mbmax = 0; 1235 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK); 1236 1237 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose) 1238 (*pr->pr_domain->dom_dispose)(asb.ssb_mb); 1239 ssb_release(&asb, so); 1240 1241 lwkt_reltoken(&ssb->ssb_token); 1242 } 1243 1244 #ifdef INET 1245 static int 1246 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt) 1247 { 1248 struct accept_filter_arg *afap = NULL; 1249 struct accept_filter *afp; 1250 struct so_accf *af = so->so_accf; 1251 int error = 0; 1252 1253 /* do not set/remove accept filters on non listen sockets */ 1254 if ((so->so_options & SO_ACCEPTCONN) == 0) { 1255 error = EINVAL; 1256 goto out; 1257 } 1258 1259 /* removing the filter */ 1260 if (sopt == NULL) { 1261 if (af != NULL) { 1262 if (af->so_accept_filter != NULL && 1263 af->so_accept_filter->accf_destroy != NULL) { 1264 af->so_accept_filter->accf_destroy(so); 1265 } 1266 if (af->so_accept_filter_str != NULL) { 1267 FREE(af->so_accept_filter_str, M_ACCF); 1268 } 1269 FREE(af, M_ACCF); 1270 so->so_accf = NULL; 1271 } 1272 so->so_options &= ~SO_ACCEPTFILTER; 1273 return (0); 1274 } 1275 /* adding a filter */ 1276 /* must remove previous filter first */ 1277 if (af != NULL) { 1278 error = EINVAL; 1279 goto out; 1280 } 1281 /* don't put large objects on the kernel stack */ 1282 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK); 1283 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap); 1284 afap->af_name[sizeof(afap->af_name)-1] = '\0'; 1285 afap->af_arg[sizeof(afap->af_arg)-1] = '\0'; 1286 if (error) 1287 goto out; 1288 afp = accept_filt_get(afap->af_name); 1289 if (afp == NULL) { 1290 error = ENOENT; 1291 goto out; 1292 } 1293 MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO); 1294 if (afp->accf_create != NULL) { 1295 if (afap->af_name[0] != '\0') { 1296 int len = strlen(afap->af_name) + 1; 1297 1298 MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK); 1299 strcpy(af->so_accept_filter_str, afap->af_name); 1300 } 1301 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg); 1302 if (af->so_accept_filter_arg == NULL) { 1303 FREE(af->so_accept_filter_str, M_ACCF); 1304 FREE(af, M_ACCF); 1305 so->so_accf = NULL; 1306 error = EINVAL; 1307 goto out; 1308 } 1309 } 1310 af->so_accept_filter = afp; 1311 so->so_accf = af; 1312 so->so_options |= SO_ACCEPTFILTER; 1313 out: 1314 if (afap != NULL) 1315 FREE(afap, M_TEMP); 1316 return (error); 1317 } 1318 #endif /* INET */ 1319 1320 /* 1321 * Perhaps this routine, and sooptcopyout(), below, ought to come in 1322 * an additional variant to handle the case where the option value needs 1323 * to be some kind of integer, but not a specific size. 1324 * In addition to their use here, these functions are also called by the 1325 * protocol-level pr_ctloutput() routines. 1326 */ 1327 int 1328 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 1329 { 1330 return soopt_to_kbuf(sopt, buf, len, minlen); 1331 } 1332 1333 int 1334 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 1335 { 1336 size_t valsize; 1337 1338 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 1339 KKASSERT(kva_p(buf)); 1340 1341 /* 1342 * If the user gives us more than we wanted, we ignore it, 1343 * but if we don't get the minimum length the caller 1344 * wants, we return EINVAL. On success, sopt->sopt_valsize 1345 * is set to however much we actually retrieved. 1346 */ 1347 if ((valsize = sopt->sopt_valsize) < minlen) 1348 return EINVAL; 1349 if (valsize > len) 1350 sopt->sopt_valsize = valsize = len; 1351 1352 bcopy(sopt->sopt_val, buf, valsize); 1353 return 0; 1354 } 1355 1356 1357 int 1358 sosetopt(struct socket *so, struct sockopt *sopt) 1359 { 1360 int error, optval; 1361 struct linger l; 1362 struct timeval tv; 1363 u_long val; 1364 struct signalsockbuf *sotmp; 1365 1366 error = 0; 1367 sopt->sopt_dir = SOPT_SET; 1368 if (sopt->sopt_level != SOL_SOCKET) { 1369 if (so->so_proto && so->so_proto->pr_ctloutput) { 1370 return (so_pr_ctloutput(so, sopt)); 1371 } 1372 error = ENOPROTOOPT; 1373 } else { 1374 switch (sopt->sopt_name) { 1375 #ifdef INET 1376 case SO_ACCEPTFILTER: 1377 error = do_setopt_accept_filter(so, sopt); 1378 if (error) 1379 goto bad; 1380 break; 1381 #endif /* INET */ 1382 case SO_LINGER: 1383 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 1384 if (error) 1385 goto bad; 1386 1387 so->so_linger = l.l_linger; 1388 if (l.l_onoff) 1389 so->so_options |= SO_LINGER; 1390 else 1391 so->so_options &= ~SO_LINGER; 1392 break; 1393 1394 case SO_DEBUG: 1395 case SO_KEEPALIVE: 1396 case SO_DONTROUTE: 1397 case SO_USELOOPBACK: 1398 case SO_BROADCAST: 1399 case SO_REUSEADDR: 1400 case SO_REUSEPORT: 1401 case SO_OOBINLINE: 1402 case SO_TIMESTAMP: 1403 error = sooptcopyin(sopt, &optval, sizeof optval, 1404 sizeof optval); 1405 if (error) 1406 goto bad; 1407 if (optval) 1408 so->so_options |= sopt->sopt_name; 1409 else 1410 so->so_options &= ~sopt->sopt_name; 1411 break; 1412 1413 case SO_SNDBUF: 1414 case SO_RCVBUF: 1415 case SO_SNDLOWAT: 1416 case SO_RCVLOWAT: 1417 error = sooptcopyin(sopt, &optval, sizeof optval, 1418 sizeof optval); 1419 if (error) 1420 goto bad; 1421 1422 /* 1423 * Values < 1 make no sense for any of these 1424 * options, so disallow them. 1425 */ 1426 if (optval < 1) { 1427 error = EINVAL; 1428 goto bad; 1429 } 1430 1431 switch (sopt->sopt_name) { 1432 case SO_SNDBUF: 1433 case SO_RCVBUF: 1434 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ? 1435 &so->so_snd : &so->so_rcv, (u_long)optval, 1436 so, 1437 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) { 1438 error = ENOBUFS; 1439 goto bad; 1440 } 1441 sotmp = (sopt->sopt_name == SO_SNDBUF) ? 1442 &so->so_snd : &so->so_rcv; 1443 atomic_clear_int(&sotmp->ssb_flags, 1444 SSB_AUTOSIZE); 1445 break; 1446 1447 /* 1448 * Make sure the low-water is never greater than 1449 * the high-water. 1450 */ 1451 case SO_SNDLOWAT: 1452 so->so_snd.ssb_lowat = 1453 (optval > so->so_snd.ssb_hiwat) ? 1454 so->so_snd.ssb_hiwat : optval; 1455 atomic_clear_int(&so->so_snd.ssb_flags, 1456 SSB_AUTOLOWAT); 1457 break; 1458 case SO_RCVLOWAT: 1459 so->so_rcv.ssb_lowat = 1460 (optval > so->so_rcv.ssb_hiwat) ? 1461 so->so_rcv.ssb_hiwat : optval; 1462 atomic_clear_int(&so->so_rcv.ssb_flags, 1463 SSB_AUTOLOWAT); 1464 break; 1465 } 1466 break; 1467 1468 case SO_SNDTIMEO: 1469 case SO_RCVTIMEO: 1470 error = sooptcopyin(sopt, &tv, sizeof tv, 1471 sizeof tv); 1472 if (error) 1473 goto bad; 1474 1475 /* assert(hz > 0); */ 1476 if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz || 1477 tv.tv_usec < 0 || tv.tv_usec >= 1000000) { 1478 error = EDOM; 1479 goto bad; 1480 } 1481 /* assert(tick > 0); */ 1482 /* assert(ULONG_MAX - SHRT_MAX >= 1000000); */ 1483 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick; 1484 if (val > SHRT_MAX) { 1485 error = EDOM; 1486 goto bad; 1487 } 1488 if (val == 0 && tv.tv_usec != 0) 1489 val = 1; 1490 1491 switch (sopt->sopt_name) { 1492 case SO_SNDTIMEO: 1493 so->so_snd.ssb_timeo = val; 1494 break; 1495 case SO_RCVTIMEO: 1496 so->so_rcv.ssb_timeo = val; 1497 break; 1498 } 1499 break; 1500 default: 1501 error = ENOPROTOOPT; 1502 break; 1503 } 1504 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1505 (void) so_pr_ctloutput(so, sopt); 1506 } 1507 } 1508 bad: 1509 return (error); 1510 } 1511 1512 /* Helper routine for getsockopt */ 1513 int 1514 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 1515 { 1516 soopt_from_kbuf(sopt, buf, len); 1517 return 0; 1518 } 1519 1520 void 1521 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len) 1522 { 1523 size_t valsize; 1524 1525 if (len == 0) { 1526 sopt->sopt_valsize = 0; 1527 return; 1528 } 1529 1530 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 1531 KKASSERT(kva_p(buf)); 1532 1533 /* 1534 * Documented get behavior is that we always return a value, 1535 * possibly truncated to fit in the user's buffer. 1536 * Traditional behavior is that we always tell the user 1537 * precisely how much we copied, rather than something useful 1538 * like the total amount we had available for her. 1539 * Note that this interface is not idempotent; the entire answer must 1540 * generated ahead of time. 1541 */ 1542 valsize = szmin(len, sopt->sopt_valsize); 1543 sopt->sopt_valsize = valsize; 1544 if (sopt->sopt_val != 0) { 1545 bcopy(buf, sopt->sopt_val, valsize); 1546 } 1547 } 1548 1549 int 1550 sogetopt(struct socket *so, struct sockopt *sopt) 1551 { 1552 int error, optval; 1553 struct linger l; 1554 struct timeval tv; 1555 #ifdef INET 1556 struct accept_filter_arg *afap; 1557 #endif 1558 1559 error = 0; 1560 sopt->sopt_dir = SOPT_GET; 1561 if (sopt->sopt_level != SOL_SOCKET) { 1562 if (so->so_proto && so->so_proto->pr_ctloutput) { 1563 return (so_pr_ctloutput(so, sopt)); 1564 } else 1565 return (ENOPROTOOPT); 1566 } else { 1567 switch (sopt->sopt_name) { 1568 #ifdef INET 1569 case SO_ACCEPTFILTER: 1570 if ((so->so_options & SO_ACCEPTCONN) == 0) 1571 return (EINVAL); 1572 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), 1573 M_TEMP, M_WAITOK | M_ZERO); 1574 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 1575 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name); 1576 if (so->so_accf->so_accept_filter_str != NULL) 1577 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str); 1578 } 1579 error = sooptcopyout(sopt, afap, sizeof(*afap)); 1580 FREE(afap, M_TEMP); 1581 break; 1582 #endif /* INET */ 1583 1584 case SO_LINGER: 1585 l.l_onoff = so->so_options & SO_LINGER; 1586 l.l_linger = so->so_linger; 1587 error = sooptcopyout(sopt, &l, sizeof l); 1588 break; 1589 1590 case SO_USELOOPBACK: 1591 case SO_DONTROUTE: 1592 case SO_DEBUG: 1593 case SO_KEEPALIVE: 1594 case SO_REUSEADDR: 1595 case SO_REUSEPORT: 1596 case SO_BROADCAST: 1597 case SO_OOBINLINE: 1598 case SO_TIMESTAMP: 1599 optval = so->so_options & sopt->sopt_name; 1600 integer: 1601 error = sooptcopyout(sopt, &optval, sizeof optval); 1602 break; 1603 1604 case SO_TYPE: 1605 optval = so->so_type; 1606 goto integer; 1607 1608 case SO_ERROR: 1609 optval = so->so_error; 1610 so->so_error = 0; 1611 goto integer; 1612 1613 case SO_SNDBUF: 1614 optval = so->so_snd.ssb_hiwat; 1615 goto integer; 1616 1617 case SO_RCVBUF: 1618 optval = so->so_rcv.ssb_hiwat; 1619 goto integer; 1620 1621 case SO_SNDLOWAT: 1622 optval = so->so_snd.ssb_lowat; 1623 goto integer; 1624 1625 case SO_RCVLOWAT: 1626 optval = so->so_rcv.ssb_lowat; 1627 goto integer; 1628 1629 case SO_SNDTIMEO: 1630 case SO_RCVTIMEO: 1631 optval = (sopt->sopt_name == SO_SNDTIMEO ? 1632 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo); 1633 1634 tv.tv_sec = optval / hz; 1635 tv.tv_usec = (optval % hz) * ustick; 1636 error = sooptcopyout(sopt, &tv, sizeof tv); 1637 break; 1638 1639 default: 1640 error = ENOPROTOOPT; 1641 break; 1642 } 1643 return (error); 1644 } 1645 } 1646 1647 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */ 1648 int 1649 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 1650 { 1651 struct mbuf *m, *m_prev; 1652 int sopt_size = sopt->sopt_valsize, msize; 1653 1654 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA, 1655 0, &msize); 1656 if (m == NULL) 1657 return (ENOBUFS); 1658 m->m_len = min(msize, sopt_size); 1659 sopt_size -= m->m_len; 1660 *mp = m; 1661 m_prev = m; 1662 1663 while (sopt_size > 0) { 1664 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, 1665 MT_DATA, 0, &msize); 1666 if (m == NULL) { 1667 m_freem(*mp); 1668 return (ENOBUFS); 1669 } 1670 m->m_len = min(msize, sopt_size); 1671 sopt_size -= m->m_len; 1672 m_prev->m_next = m; 1673 m_prev = m; 1674 } 1675 return (0); 1676 } 1677 1678 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */ 1679 int 1680 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 1681 { 1682 soopt_to_mbuf(sopt, m); 1683 return 0; 1684 } 1685 1686 void 1687 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m) 1688 { 1689 size_t valsize; 1690 void *val; 1691 1692 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 1693 KKASSERT(kva_p(m)); 1694 if (sopt->sopt_val == NULL) 1695 return; 1696 val = sopt->sopt_val; 1697 valsize = sopt->sopt_valsize; 1698 while (m != NULL && valsize >= m->m_len) { 1699 bcopy(val, mtod(m, char *), m->m_len); 1700 valsize -= m->m_len; 1701 val = (caddr_t)val + m->m_len; 1702 m = m->m_next; 1703 } 1704 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 1705 panic("ip6_sooptmcopyin"); 1706 } 1707 1708 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */ 1709 int 1710 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 1711 { 1712 return soopt_from_mbuf(sopt, m); 1713 } 1714 1715 int 1716 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m) 1717 { 1718 struct mbuf *m0 = m; 1719 size_t valsize = 0; 1720 size_t maxsize; 1721 void *val; 1722 1723 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 1724 KKASSERT(kva_p(m)); 1725 if (sopt->sopt_val == NULL) 1726 return 0; 1727 val = sopt->sopt_val; 1728 maxsize = sopt->sopt_valsize; 1729 while (m != NULL && maxsize >= m->m_len) { 1730 bcopy(mtod(m, char *), val, m->m_len); 1731 maxsize -= m->m_len; 1732 val = (caddr_t)val + m->m_len; 1733 valsize += m->m_len; 1734 m = m->m_next; 1735 } 1736 if (m != NULL) { 1737 /* enough soopt buffer should be given from user-land */ 1738 m_freem(m0); 1739 return (EINVAL); 1740 } 1741 sopt->sopt_valsize = valsize; 1742 return 0; 1743 } 1744 1745 void 1746 sohasoutofband(struct socket *so) 1747 { 1748 if (so->so_sigio != NULL) 1749 pgsigio(so->so_sigio, SIGURG, 0); 1750 KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB); 1751 } 1752 1753 int 1754 sokqfilter(struct file *fp, struct knote *kn) 1755 { 1756 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1757 struct signalsockbuf *ssb; 1758 1759 switch (kn->kn_filter) { 1760 case EVFILT_READ: 1761 if (so->so_options & SO_ACCEPTCONN) 1762 kn->kn_fop = &solisten_filtops; 1763 else 1764 kn->kn_fop = &soread_filtops; 1765 ssb = &so->so_rcv; 1766 break; 1767 case EVFILT_WRITE: 1768 kn->kn_fop = &sowrite_filtops; 1769 ssb = &so->so_snd; 1770 break; 1771 case EVFILT_EXCEPT: 1772 kn->kn_fop = &soexcept_filtops; 1773 ssb = &so->so_rcv; 1774 break; 1775 default: 1776 return (EOPNOTSUPP); 1777 } 1778 1779 knote_insert(&ssb->ssb_kq.ki_note, kn); 1780 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE); 1781 return (0); 1782 } 1783 1784 static void 1785 filt_sordetach(struct knote *kn) 1786 { 1787 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1788 1789 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn); 1790 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note)) 1791 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE); 1792 } 1793 1794 /*ARGSUSED*/ 1795 static int 1796 filt_soread(struct knote *kn, long hint) 1797 { 1798 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1799 1800 if (kn->kn_sfflags & NOTE_OOB) { 1801 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) { 1802 kn->kn_fflags |= NOTE_OOB; 1803 return (1); 1804 } 1805 return (0); 1806 } 1807 kn->kn_data = so->so_rcv.ssb_cc; 1808 1809 /* 1810 * Only set EOF if all data has been exhausted. 1811 */ 1812 if ((so->so_state & SS_CANTRCVMORE) && kn->kn_data == 0) { 1813 kn->kn_flags |= EV_EOF; 1814 kn->kn_fflags = so->so_error; 1815 return (1); 1816 } 1817 if (so->so_error) /* temporary udp error */ 1818 return (1); 1819 if (kn->kn_sfflags & NOTE_LOWAT) 1820 return (kn->kn_data >= kn->kn_sdata); 1821 return ((kn->kn_data >= so->so_rcv.ssb_lowat) || 1822 !TAILQ_EMPTY(&so->so_comp)); 1823 } 1824 1825 static void 1826 filt_sowdetach(struct knote *kn) 1827 { 1828 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1829 1830 knote_remove(&so->so_snd.ssb_kq.ki_note, kn); 1831 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note)) 1832 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE); 1833 } 1834 1835 /*ARGSUSED*/ 1836 static int 1837 filt_sowrite(struct knote *kn, long hint) 1838 { 1839 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1840 1841 kn->kn_data = ssb_space(&so->so_snd); 1842 if (so->so_state & SS_CANTSENDMORE) { 1843 kn->kn_flags |= EV_EOF; 1844 kn->kn_fflags = so->so_error; 1845 return (1); 1846 } 1847 if (so->so_error) /* temporary udp error */ 1848 return (1); 1849 if (((so->so_state & SS_ISCONNECTED) == 0) && 1850 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1851 return (0); 1852 if (kn->kn_sfflags & NOTE_LOWAT) 1853 return (kn->kn_data >= kn->kn_sdata); 1854 return (kn->kn_data >= so->so_snd.ssb_lowat); 1855 } 1856 1857 /*ARGSUSED*/ 1858 static int 1859 filt_solisten(struct knote *kn, long hint) 1860 { 1861 struct socket *so = (struct socket *)kn->kn_fp->f_data; 1862 1863 kn->kn_data = so->so_qlen; 1864 return (! TAILQ_EMPTY(&so->so_comp)); 1865 } 1866