1 /* 2 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * Copyright (c) 1982, 1986, 1988, 1990, 1993 36 * The Regents of the University of California. All rights reserved. 37 * 38 * Redistribution and use in source and binary forms, with or without 39 * modification, are permitted provided that the following conditions 40 * are met: 41 * 1. Redistributions of source code must retain the above copyright 42 * notice, this list of conditions and the following disclaimer. 43 * 2. Redistributions in binary form must reproduce the above copyright 44 * notice, this list of conditions and the following disclaimer in the 45 * documentation and/or other materials provided with the distribution. 46 * 3. All advertising materials mentioning features or use of this software 47 * must display the following acknowledgement: 48 * This product includes software developed by the University of 49 * California, Berkeley and its contributors. 50 * 4. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 67 * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.24 2003/11/11 17:18:18 silby Exp $ 68 */ 69 70 #include "opt_inet.h" 71 #include "opt_sctp.h" 72 73 #include <sys/param.h> 74 #include <sys/systm.h> 75 #include <sys/fcntl.h> 76 #include <sys/malloc.h> 77 #include <sys/mbuf.h> 78 #include <sys/domain.h> 79 #include <sys/file.h> /* for struct knote */ 80 #include <sys/kernel.h> 81 #include <sys/event.h> 82 #include <sys/proc.h> 83 #include <sys/protosw.h> 84 #include <sys/socket.h> 85 #include <sys/socketvar.h> 86 #include <sys/socketops.h> 87 #include <sys/resourcevar.h> 88 #include <sys/signalvar.h> 89 #include <sys/sysctl.h> 90 #include <sys/uio.h> 91 #include <sys/jail.h> 92 #include <vm/vm_zone.h> 93 #include <vm/pmap.h> 94 #include <net/netmsg2.h> 95 96 #include <sys/thread2.h> 97 #include <sys/socketvar2.h> 98 #include <sys/spinlock2.h> 99 100 #include <machine/limits.h> 101 102 extern int tcp_sosend_agglim; 103 extern int tcp_sosend_async; 104 extern int udp_sosend_async; 105 extern int udp_sosend_prepend; 106 107 #ifdef INET 108 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt); 109 #endif /* INET */ 110 111 static void filt_sordetach(struct knote *kn); 112 static int filt_soread(struct knote *kn, long hint); 113 static void filt_sowdetach(struct knote *kn); 114 static int filt_sowrite(struct knote *kn, long hint); 115 static int filt_solisten(struct knote *kn, long hint); 116 117 static void sodiscard(struct socket *so); 118 static int soclose_sync(struct socket *so, int fflag); 119 static void soclose_fast(struct socket *so); 120 121 static struct filterops solisten_filtops = 122 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten }; 123 static struct filterops soread_filtops = 124 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread }; 125 static struct filterops sowrite_filtops = 126 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite }; 127 static struct filterops soexcept_filtops = 128 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread }; 129 130 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct"); 131 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 132 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 133 134 135 static int somaxconn = SOMAXCONN; 136 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW, 137 &somaxconn, 0, "Maximum pending socket connection queue size"); 138 139 static int use_soclose_fast = 1; 140 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW, 141 &use_soclose_fast, 0, "Fast socket close"); 142 143 int use_soaccept_pred_fast = 1; 144 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW, 145 &use_soaccept_pred_fast, 0, "Fast socket accept predication"); 146 147 int use_sendfile_async = 1; 148 SYSCTL_INT(_kern_ipc, OID_AUTO, sendfile_async, CTLFLAG_RW, 149 &use_sendfile_async, 0, "sendfile uses asynchronized pru_send"); 150 151 /* 152 * Socket operation routines. 153 * These routines are called by the routines in 154 * sys_socket.c or from a system process, and 155 * implement the semantics of socket operations by 156 * switching out to the protocol specific routines. 157 */ 158 159 /* 160 * Get a socket structure, and initialize it. 161 * Note that it would probably be better to allocate socket 162 * and PCB at the same time, but I'm not convinced that all 163 * the protocols can be easily modified to do this. 164 */ 165 struct socket * 166 soalloc(int waitok, struct protosw *pr) 167 { 168 struct socket *so; 169 unsigned waitmask; 170 171 waitmask = waitok ? M_WAITOK : M_NOWAIT; 172 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask); 173 if (so) { 174 /* XXX race condition for reentrant kernel */ 175 so->so_proto = pr; 176 TAILQ_INIT(&so->so_aiojobq); 177 TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist); 178 TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist); 179 lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok"); 180 lwkt_token_init(&so->so_snd.ssb_token, "sndtok"); 181 spin_init(&so->so_rcvd_spin); 182 netmsg_init(&so->so_rcvd_msg.base, so, &netisr_adone_rport, 183 MSGF_DROPABLE, so->so_proto->pr_usrreqs->pru_rcvd); 184 so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC; 185 so->so_state = SS_NOFDREF; 186 so->so_refs = 1; 187 } 188 return so; 189 } 190 191 int 192 socreate(int dom, struct socket **aso, int type, 193 int proto, struct thread *td) 194 { 195 struct proc *p = td->td_proc; 196 struct protosw *prp; 197 struct socket *so; 198 struct pru_attach_info ai; 199 int error; 200 201 if (proto) 202 prp = pffindproto(dom, proto, type); 203 else 204 prp = pffindtype(dom, type); 205 206 if (prp == NULL || prp->pr_usrreqs->pru_attach == 0) 207 return (EPROTONOSUPPORT); 208 209 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only && 210 prp->pr_domain->dom_family != PF_LOCAL && 211 prp->pr_domain->dom_family != PF_INET && 212 prp->pr_domain->dom_family != PF_INET6 && 213 prp->pr_domain->dom_family != PF_ROUTE) { 214 return (EPROTONOSUPPORT); 215 } 216 217 if (prp->pr_type != type) 218 return (EPROTOTYPE); 219 so = soalloc(p != NULL, prp); 220 if (so == NULL) 221 return (ENOBUFS); 222 223 /* 224 * Callers of socreate() presumably will connect up a descriptor 225 * and call soclose() if they cannot. This represents our so_refs 226 * (which should be 1) from soalloc(). 227 */ 228 soclrstate(so, SS_NOFDREF); 229 230 /* 231 * Set a default port for protocol processing. No action will occur 232 * on the socket on this port until an inpcb is attached to it and 233 * is able to match incoming packets, or until the socket becomes 234 * available to userland. 235 * 236 * We normally default the socket to the protocol thread on cpu 0. 237 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol 238 * thread and all pr_*()/pru_*() calls are executed synchronously. 239 */ 240 if (prp->pr_flags & PR_SYNC_PORT) 241 so->so_port = &netisr_sync_port; 242 else 243 so->so_port = netisr_portfn(0); 244 245 TAILQ_INIT(&so->so_incomp); 246 TAILQ_INIT(&so->so_comp); 247 so->so_type = type; 248 so->so_cred = crhold(p->p_ucred); 249 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE]; 250 ai.p_ucred = p->p_ucred; 251 ai.fd_rdir = p->p_fd->fd_rdir; 252 253 /* 254 * Auto-sizing of socket buffers is managed by the protocols and 255 * the appropriate flags must be set in the pru_attach function. 256 */ 257 error = so_pru_attach(so, proto, &ai); 258 if (error) { 259 sosetstate(so, SS_NOFDREF); 260 sofree(so); /* from soalloc */ 261 return error; 262 } 263 264 /* 265 * NOTE: Returns referenced socket. 266 */ 267 *aso = so; 268 return (0); 269 } 270 271 int 272 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 273 { 274 int error; 275 276 error = so_pru_bind(so, nam, td); 277 return (error); 278 } 279 280 static void 281 sodealloc(struct socket *so) 282 { 283 if (so->so_rcv.ssb_hiwat) 284 (void)chgsbsize(so->so_cred->cr_uidinfo, 285 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY); 286 if (so->so_snd.ssb_hiwat) 287 (void)chgsbsize(so->so_cred->cr_uidinfo, 288 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY); 289 #ifdef INET 290 /* remove accept filter if present */ 291 if (so->so_accf != NULL) 292 do_setopt_accept_filter(so, NULL); 293 #endif /* INET */ 294 crfree(so->so_cred); 295 if (so->so_faddr != NULL) 296 kfree(so->so_faddr, M_SONAME); 297 kfree(so, M_SOCKET); 298 } 299 300 int 301 solisten(struct socket *so, int backlog, struct thread *td) 302 { 303 int error; 304 #ifdef SCTP 305 short oldopt, oldqlimit; 306 #endif /* SCTP */ 307 308 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) 309 return (EINVAL); 310 311 #ifdef SCTP 312 oldopt = so->so_options; 313 oldqlimit = so->so_qlimit; 314 #endif /* SCTP */ 315 316 lwkt_gettoken(&so->so_rcv.ssb_token); 317 if (TAILQ_EMPTY(&so->so_comp)) 318 so->so_options |= SO_ACCEPTCONN; 319 lwkt_reltoken(&so->so_rcv.ssb_token); 320 if (backlog < 0 || backlog > somaxconn) 321 backlog = somaxconn; 322 so->so_qlimit = backlog; 323 /* SCTP needs to look at tweak both the inbound backlog parameter AND 324 * the so_options (UDP model both connect's and gets inbound 325 * connections .. implicitly). 326 */ 327 error = so_pru_listen(so, td); 328 if (error) { 329 #ifdef SCTP 330 /* Restore the params */ 331 so->so_options = oldopt; 332 so->so_qlimit = oldqlimit; 333 #endif /* SCTP */ 334 return (error); 335 } 336 return (0); 337 } 338 339 /* 340 * Destroy a disconnected socket. This routine is a NOP if entities 341 * still have a reference on the socket: 342 * 343 * so_pcb - The protocol stack still has a reference 344 * SS_NOFDREF - There is no longer a file pointer reference 345 */ 346 void 347 sofree(struct socket *so) 348 { 349 struct socket *head; 350 351 /* 352 * This is a bit hackish at the moment. We need to interlock 353 * any accept queue we are on before we potentially lose the 354 * last reference to avoid races against a re-reference from 355 * someone operating on the queue. 356 */ 357 while ((head = so->so_head) != NULL) { 358 lwkt_getpooltoken(head); 359 if (so->so_head == head) 360 break; 361 lwkt_relpooltoken(head); 362 } 363 364 /* 365 * Arbitrage the last free. 366 */ 367 KKASSERT(so->so_refs > 0); 368 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) { 369 if (head) 370 lwkt_relpooltoken(head); 371 return; 372 } 373 374 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF)); 375 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0); 376 377 /* 378 * We're done, remove ourselves from the accept queue we are 379 * on, if we are on one. 380 */ 381 if (head != NULL) { 382 if (so->so_state & SS_INCOMP) { 383 TAILQ_REMOVE(&head->so_incomp, so, so_list); 384 head->so_incqlen--; 385 } else if (so->so_state & SS_COMP) { 386 /* 387 * We must not decommission a socket that's 388 * on the accept(2) queue. If we do, then 389 * accept(2) may hang after select(2) indicated 390 * that the listening socket was ready. 391 */ 392 lwkt_relpooltoken(head); 393 return; 394 } else { 395 panic("sofree: not queued"); 396 } 397 soclrstate(so, SS_INCOMP); 398 so->so_head = NULL; 399 lwkt_relpooltoken(head); 400 } 401 ssb_release(&so->so_snd, so); 402 sorflush(so); 403 sodealloc(so); 404 } 405 406 /* 407 * Close a socket on last file table reference removal. 408 * Initiate disconnect if connected. 409 * Free socket when disconnect complete. 410 */ 411 int 412 soclose(struct socket *so, int fflag) 413 { 414 int error; 415 416 funsetown(&so->so_sigio); 417 if (!use_soclose_fast || 418 (so->so_proto->pr_flags & PR_SYNC_PORT) || 419 (so->so_options & SO_LINGER)) { 420 error = soclose_sync(so, fflag); 421 } else { 422 soclose_fast(so); 423 error = 0; 424 } 425 return error; 426 } 427 428 static void 429 sodiscard(struct socket *so) 430 { 431 lwkt_getpooltoken(so); 432 if (so->so_options & SO_ACCEPTCONN) { 433 struct socket *sp; 434 435 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) { 436 TAILQ_REMOVE(&so->so_incomp, sp, so_list); 437 soclrstate(sp, SS_INCOMP); 438 sp->so_head = NULL; 439 so->so_incqlen--; 440 soaborta(sp); 441 } 442 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) { 443 TAILQ_REMOVE(&so->so_comp, sp, so_list); 444 soclrstate(sp, SS_COMP); 445 sp->so_head = NULL; 446 so->so_qlen--; 447 soaborta(sp); 448 } 449 } 450 lwkt_relpooltoken(so); 451 452 if (so->so_state & SS_NOFDREF) 453 panic("soclose: NOFDREF"); 454 sosetstate(so, SS_NOFDREF); /* take ref */ 455 } 456 457 static int 458 soclose_sync(struct socket *so, int fflag) 459 { 460 int error = 0; 461 462 if (so->so_pcb == NULL) 463 goto discard; 464 if (so->so_state & SS_ISCONNECTED) { 465 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 466 error = sodisconnect(so); 467 if (error) 468 goto drop; 469 } 470 if (so->so_options & SO_LINGER) { 471 if ((so->so_state & SS_ISDISCONNECTING) && 472 (fflag & FNONBLOCK)) 473 goto drop; 474 while (so->so_state & SS_ISCONNECTED) { 475 error = tsleep(&so->so_timeo, PCATCH, 476 "soclos", so->so_linger * hz); 477 if (error) 478 break; 479 } 480 } 481 } 482 drop: 483 if (so->so_pcb) { 484 int error2; 485 486 error2 = so_pru_detach(so); 487 if (error == 0) 488 error = error2; 489 } 490 discard: 491 sodiscard(so); 492 so_pru_sync(so); /* unpend async sending */ 493 sofree(so); /* dispose of ref */ 494 495 return (error); 496 } 497 498 static void 499 soclose_sofree_async_handler(netmsg_t msg) 500 { 501 sofree(msg->base.nm_so); 502 } 503 504 static void 505 soclose_sofree_async(struct socket *so) 506 { 507 struct netmsg_base *base = &so->so_clomsg; 508 509 netmsg_init(base, so, &netisr_apanic_rport, 0, 510 soclose_sofree_async_handler); 511 lwkt_sendmsg(so->so_port, &base->lmsg); 512 } 513 514 static void 515 soclose_disconn_async_handler(netmsg_t msg) 516 { 517 struct socket *so = msg->base.nm_so; 518 519 if ((so->so_state & SS_ISCONNECTED) && 520 (so->so_state & SS_ISDISCONNECTING) == 0) 521 so_pru_disconnect_direct(so); 522 523 if (so->so_pcb) 524 so_pru_detach_direct(so); 525 526 sodiscard(so); 527 sofree(so); 528 } 529 530 static void 531 soclose_disconn_async(struct socket *so) 532 { 533 struct netmsg_base *base = &so->so_clomsg; 534 535 netmsg_init(base, so, &netisr_apanic_rport, 0, 536 soclose_disconn_async_handler); 537 lwkt_sendmsg(so->so_port, &base->lmsg); 538 } 539 540 static void 541 soclose_detach_async_handler(netmsg_t msg) 542 { 543 struct socket *so = msg->base.nm_so; 544 545 if (so->so_pcb) 546 so_pru_detach_direct(so); 547 548 sodiscard(so); 549 sofree(so); 550 } 551 552 static void 553 soclose_detach_async(struct socket *so) 554 { 555 struct netmsg_base *base = &so->so_clomsg; 556 557 netmsg_init(base, so, &netisr_apanic_rport, 0, 558 soclose_detach_async_handler); 559 lwkt_sendmsg(so->so_port, &base->lmsg); 560 } 561 562 static void 563 soclose_fast(struct socket *so) 564 { 565 if (so->so_pcb == NULL) 566 goto discard; 567 568 if ((so->so_state & SS_ISCONNECTED) && 569 (so->so_state & SS_ISDISCONNECTING) == 0) { 570 soclose_disconn_async(so); 571 return; 572 } 573 574 if (so->so_pcb) { 575 soclose_detach_async(so); 576 return; 577 } 578 579 discard: 580 sodiscard(so); 581 soclose_sofree_async(so); 582 } 583 584 /* 585 * Abort and destroy a socket. Only one abort can be in progress 586 * at any given moment. 587 */ 588 void 589 soabort(struct socket *so) 590 { 591 soreference(so); 592 so_pru_abort(so); 593 } 594 595 void 596 soaborta(struct socket *so) 597 { 598 soreference(so); 599 so_pru_aborta(so); 600 } 601 602 void 603 soabort_oncpu(struct socket *so) 604 { 605 soreference(so); 606 so_pru_abort_oncpu(so); 607 } 608 609 /* 610 * so is passed in ref'd, which becomes owned by 611 * the cleared SS_NOFDREF flag. 612 */ 613 void 614 soaccept_generic(struct socket *so) 615 { 616 if ((so->so_state & SS_NOFDREF) == 0) 617 panic("soaccept: !NOFDREF"); 618 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */ 619 } 620 621 int 622 soaccept(struct socket *so, struct sockaddr **nam) 623 { 624 int error; 625 626 soaccept_generic(so); 627 error = so_pru_accept(so, nam); 628 return (error); 629 } 630 631 int 632 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 633 { 634 int error; 635 636 if (so->so_options & SO_ACCEPTCONN) 637 return (EOPNOTSUPP); 638 /* 639 * If protocol is connection-based, can only connect once. 640 * Otherwise, if connected, try to disconnect first. 641 * This allows user to disconnect by connecting to, e.g., 642 * a null address. 643 */ 644 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 645 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 646 (error = sodisconnect(so)))) { 647 error = EISCONN; 648 } else { 649 /* 650 * Prevent accumulated error from previous connection 651 * from biting us. 652 */ 653 so->so_error = 0; 654 error = so_pru_connect(so, nam, td); 655 } 656 return (error); 657 } 658 659 int 660 soconnect2(struct socket *so1, struct socket *so2) 661 { 662 int error; 663 664 error = so_pru_connect2(so1, so2); 665 return (error); 666 } 667 668 int 669 sodisconnect(struct socket *so) 670 { 671 int error; 672 673 if ((so->so_state & SS_ISCONNECTED) == 0) { 674 error = ENOTCONN; 675 goto bad; 676 } 677 if (so->so_state & SS_ISDISCONNECTING) { 678 error = EALREADY; 679 goto bad; 680 } 681 error = so_pru_disconnect(so); 682 bad: 683 return (error); 684 } 685 686 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 687 /* 688 * Send on a socket. 689 * If send must go all at once and message is larger than 690 * send buffering, then hard error. 691 * Lock against other senders. 692 * If must go all at once and not enough room now, then 693 * inform user that this would block and do nothing. 694 * Otherwise, if nonblocking, send as much as possible. 695 * The data to be sent is described by "uio" if nonzero, 696 * otherwise by the mbuf chain "top" (which must be null 697 * if uio is not). Data provided in mbuf chain must be small 698 * enough to send all at once. 699 * 700 * Returns nonzero on error, timeout or signal; callers 701 * must check for short counts if EINTR/ERESTART are returned. 702 * Data and control buffers are freed on return. 703 */ 704 int 705 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 706 struct mbuf *top, struct mbuf *control, int flags, 707 struct thread *td) 708 { 709 struct mbuf **mp; 710 struct mbuf *m; 711 size_t resid; 712 int space, len; 713 int clen = 0, error, dontroute, mlen; 714 int atomic = sosendallatonce(so) || top; 715 int pru_flags; 716 717 if (uio) { 718 resid = uio->uio_resid; 719 } else { 720 resid = (size_t)top->m_pkthdr.len; 721 #ifdef INVARIANTS 722 len = 0; 723 for (m = top; m; m = m->m_next) 724 len += m->m_len; 725 KKASSERT(top->m_pkthdr.len == len); 726 #endif 727 } 728 729 /* 730 * WARNING! resid is unsigned, space and len are signed. space 731 * can wind up negative if the sockbuf is overcommitted. 732 * 733 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 734 * type sockets since that's an error. 735 */ 736 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) { 737 error = EINVAL; 738 goto out; 739 } 740 741 dontroute = 742 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 743 (so->so_proto->pr_flags & PR_ATOMIC); 744 if (td->td_lwp != NULL) 745 td->td_lwp->lwp_ru.ru_msgsnd++; 746 if (control) 747 clen = control->m_len; 748 #define gotoerr(errcode) { error = errcode; goto release; } 749 750 restart: 751 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags)); 752 if (error) 753 goto out; 754 755 do { 756 if (so->so_state & SS_CANTSENDMORE) 757 gotoerr(EPIPE); 758 if (so->so_error) { 759 error = so->so_error; 760 so->so_error = 0; 761 goto release; 762 } 763 if ((so->so_state & SS_ISCONNECTED) == 0) { 764 /* 765 * `sendto' and `sendmsg' is allowed on a connection- 766 * based socket if it supports implied connect. 767 * Return ENOTCONN if not connected and no address is 768 * supplied. 769 */ 770 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 771 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 772 if ((so->so_state & SS_ISCONFIRMING) == 0 && 773 !(resid == 0 && clen != 0)) 774 gotoerr(ENOTCONN); 775 } else if (addr == NULL) 776 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ? 777 ENOTCONN : EDESTADDRREQ); 778 } 779 if ((atomic && resid > so->so_snd.ssb_hiwat) || 780 clen > so->so_snd.ssb_hiwat) { 781 gotoerr(EMSGSIZE); 782 } 783 space = ssb_space(&so->so_snd); 784 if (flags & MSG_OOB) 785 space += 1024; 786 if ((space < 0 || (size_t)space < resid + clen) && uio && 787 (atomic || space < so->so_snd.ssb_lowat || space < clen)) { 788 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) 789 gotoerr(EWOULDBLOCK); 790 ssb_unlock(&so->so_snd); 791 error = ssb_wait(&so->so_snd); 792 if (error) 793 goto out; 794 goto restart; 795 } 796 mp = ⊤ 797 space -= clen; 798 do { 799 if (uio == NULL) { 800 /* 801 * Data is prepackaged in "top". 802 */ 803 resid = 0; 804 if (flags & MSG_EOR) 805 top->m_flags |= M_EOR; 806 } else do { 807 if (resid > INT_MAX) 808 resid = INT_MAX; 809 m = m_getl((int)resid, MB_WAIT, MT_DATA, 810 top == NULL ? M_PKTHDR : 0, &mlen); 811 if (top == NULL) { 812 m->m_pkthdr.len = 0; 813 m->m_pkthdr.rcvif = NULL; 814 } 815 len = imin((int)szmin(mlen, resid), space); 816 if (resid < MINCLSIZE) { 817 /* 818 * For datagram protocols, leave room 819 * for protocol headers in first mbuf. 820 */ 821 if (atomic && top == NULL && len < mlen) 822 MH_ALIGN(m, len); 823 } 824 space -= len; 825 error = uiomove(mtod(m, caddr_t), (size_t)len, uio); 826 resid = uio->uio_resid; 827 m->m_len = len; 828 *mp = m; 829 top->m_pkthdr.len += len; 830 if (error) 831 goto release; 832 mp = &m->m_next; 833 if (resid == 0) { 834 if (flags & MSG_EOR) 835 top->m_flags |= M_EOR; 836 break; 837 } 838 } while (space > 0 && atomic); 839 if (dontroute) 840 so->so_options |= SO_DONTROUTE; 841 if (flags & MSG_OOB) { 842 pru_flags = PRUS_OOB; 843 } else if ((flags & MSG_EOF) && 844 (so->so_proto->pr_flags & PR_IMPLOPCL) && 845 (resid == 0)) { 846 /* 847 * If the user set MSG_EOF, the protocol 848 * understands this flag and nothing left to 849 * send then use PRU_SEND_EOF instead of PRU_SEND. 850 */ 851 pru_flags = PRUS_EOF; 852 } else if (resid > 0 && space > 0) { 853 /* If there is more to send, set PRUS_MORETOCOME */ 854 pru_flags = PRUS_MORETOCOME; 855 } else { 856 pru_flags = 0; 857 } 858 /* 859 * XXX all the SS_CANTSENDMORE checks previously 860 * done could be out of date. We could have recieved 861 * a reset packet in an interrupt or maybe we slept 862 * while doing page faults in uiomove() etc. We could 863 * probably recheck again inside the splnet() protection 864 * here, but there are probably other places that this 865 * also happens. We must rethink this. 866 */ 867 error = so_pru_send(so, pru_flags, top, addr, control, td); 868 if (dontroute) 869 so->so_options &= ~SO_DONTROUTE; 870 clen = 0; 871 control = NULL; 872 top = NULL; 873 mp = ⊤ 874 if (error) 875 goto release; 876 } while (resid && space > 0); 877 } while (resid); 878 879 release: 880 ssb_unlock(&so->so_snd); 881 out: 882 if (top) 883 m_freem(top); 884 if (control) 885 m_freem(control); 886 return (error); 887 } 888 889 /* 890 * A specialization of sosend() for UDP based on protocol-specific knowledge: 891 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that 892 * sosendallatonce() returns true, 893 * the "atomic" variable is true, 894 * and sosendudp() blocks until space is available for the entire send. 895 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or 896 * PR_IMPLOPCL flags set. 897 * UDP has no out-of-band data. 898 * UDP has no control data. 899 * UDP does not support MSG_EOR. 900 */ 901 int 902 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio, 903 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 904 { 905 size_t resid; 906 int error, pru_flags = 0; 907 int space; 908 909 if (td->td_lwp != NULL) 910 td->td_lwp->lwp_ru.ru_msgsnd++; 911 if (control) 912 m_freem(control); 913 914 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp")); 915 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len; 916 917 restart: 918 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags)); 919 if (error) 920 goto out; 921 922 if (so->so_state & SS_CANTSENDMORE) 923 gotoerr(EPIPE); 924 if (so->so_error) { 925 error = so->so_error; 926 so->so_error = 0; 927 goto release; 928 } 929 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL) 930 gotoerr(EDESTADDRREQ); 931 if (resid > so->so_snd.ssb_hiwat) 932 gotoerr(EMSGSIZE); 933 space = ssb_space(&so->so_snd); 934 if (uio && (space < 0 || (size_t)space < resid)) { 935 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) 936 gotoerr(EWOULDBLOCK); 937 ssb_unlock(&so->so_snd); 938 error = ssb_wait(&so->so_snd); 939 if (error) 940 goto out; 941 goto restart; 942 } 943 944 if (uio) { 945 int hdrlen = max_hdr; 946 947 /* 948 * We try to optimize out the additional mbuf 949 * allocations in M_PREPEND() on output path, e.g. 950 * - udp_output(), when it tries to prepend protocol 951 * headers. 952 * - Link layer output function, when it tries to 953 * prepend link layer header. 954 * 955 * This probably will not benefit any data that will 956 * be fragmented, so this optimization is only performed 957 * when the size of data and max size of protocol+link 958 * headers fit into one mbuf cluster. 959 */ 960 if (uio->uio_resid > MCLBYTES - hdrlen || 961 !udp_sosend_prepend) { 962 top = m_uiomove(uio); 963 if (top == NULL) 964 goto release; 965 } else { 966 int nsize; 967 968 top = m_getl(uio->uio_resid + hdrlen, MB_WAIT, 969 MT_DATA, M_PKTHDR, &nsize); 970 KASSERT(nsize >= uio->uio_resid + hdrlen, 971 ("sosendudp invalid nsize %d, " 972 "resid %zu, hdrlen %d", 973 nsize, uio->uio_resid, hdrlen)); 974 975 top->m_len = uio->uio_resid; 976 top->m_pkthdr.len = uio->uio_resid; 977 top->m_data += hdrlen; 978 979 error = uiomove(mtod(top, caddr_t), top->m_len, uio); 980 if (error) 981 goto out; 982 } 983 } 984 985 if (flags & MSG_DONTROUTE) 986 pru_flags |= PRUS_DONTROUTE; 987 988 if (udp_sosend_async && (flags & MSG_SYNC) == 0) { 989 so_pru_send_async(so, pru_flags, top, addr, NULL, td); 990 error = 0; 991 } else { 992 error = so_pru_send(so, pru_flags, top, addr, NULL, td); 993 } 994 top = NULL; /* sent or freed in lower layer */ 995 996 release: 997 ssb_unlock(&so->so_snd); 998 out: 999 if (top) 1000 m_freem(top); 1001 return (error); 1002 } 1003 1004 int 1005 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio, 1006 struct mbuf *top, struct mbuf *control, int flags, 1007 struct thread *td) 1008 { 1009 struct mbuf **mp; 1010 struct mbuf *m; 1011 size_t resid; 1012 int space, len; 1013 int error, mlen; 1014 int allatonce; 1015 int pru_flags; 1016 1017 if (uio) { 1018 KKASSERT(top == NULL); 1019 allatonce = 0; 1020 resid = uio->uio_resid; 1021 } else { 1022 allatonce = 1; 1023 resid = (size_t)top->m_pkthdr.len; 1024 #ifdef INVARIANTS 1025 len = 0; 1026 for (m = top; m; m = m->m_next) 1027 len += m->m_len; 1028 KKASSERT(top->m_pkthdr.len == len); 1029 #endif 1030 } 1031 1032 /* 1033 * WARNING! resid is unsigned, space and len are signed. space 1034 * can wind up negative if the sockbuf is overcommitted. 1035 * 1036 * Also check to make sure that MSG_EOR isn't used on TCP 1037 */ 1038 if (flags & MSG_EOR) { 1039 error = EINVAL; 1040 goto out; 1041 } 1042 1043 if (control) { 1044 /* TCP doesn't do control messages (rights, creds, etc) */ 1045 if (control->m_len) { 1046 error = EINVAL; 1047 goto out; 1048 } 1049 m_freem(control); /* empty control, just free it */ 1050 control = NULL; 1051 } 1052 1053 if (td->td_lwp != NULL) 1054 td->td_lwp->lwp_ru.ru_msgsnd++; 1055 1056 #define gotoerr(errcode) { error = errcode; goto release; } 1057 1058 restart: 1059 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags)); 1060 if (error) 1061 goto out; 1062 1063 do { 1064 if (so->so_state & SS_CANTSENDMORE) 1065 gotoerr(EPIPE); 1066 if (so->so_error) { 1067 error = so->so_error; 1068 so->so_error = 0; 1069 goto release; 1070 } 1071 if ((so->so_state & SS_ISCONNECTED) == 0 && 1072 (so->so_state & SS_ISCONFIRMING) == 0) 1073 gotoerr(ENOTCONN); 1074 if (allatonce && resid > so->so_snd.ssb_hiwat) 1075 gotoerr(EMSGSIZE); 1076 1077 space = ssb_space_prealloc(&so->so_snd); 1078 if (flags & MSG_OOB) 1079 space += 1024; 1080 if ((space < 0 || (size_t)space < resid) && !allatonce && 1081 space < so->so_snd.ssb_lowat) { 1082 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) 1083 gotoerr(EWOULDBLOCK); 1084 ssb_unlock(&so->so_snd); 1085 error = ssb_wait(&so->so_snd); 1086 if (error) 1087 goto out; 1088 goto restart; 1089 } 1090 mp = ⊤ 1091 do { 1092 int cnt = 0, async = 0; 1093 1094 if (uio == NULL) { 1095 /* 1096 * Data is prepackaged in "top". 1097 */ 1098 resid = 0; 1099 } else do { 1100 if (resid > INT_MAX) 1101 resid = INT_MAX; 1102 m = m_getl((int)resid, MB_WAIT, MT_DATA, 1103 top == NULL ? M_PKTHDR : 0, &mlen); 1104 if (top == NULL) { 1105 m->m_pkthdr.len = 0; 1106 m->m_pkthdr.rcvif = NULL; 1107 } 1108 len = imin((int)szmin(mlen, resid), space); 1109 space -= len; 1110 error = uiomove(mtod(m, caddr_t), (size_t)len, uio); 1111 resid = uio->uio_resid; 1112 m->m_len = len; 1113 *mp = m; 1114 top->m_pkthdr.len += len; 1115 if (error) 1116 goto release; 1117 mp = &m->m_next; 1118 if (resid == 0) 1119 break; 1120 ++cnt; 1121 } while (space > 0 && cnt < tcp_sosend_agglim); 1122 1123 if (tcp_sosend_async) 1124 async = 1; 1125 1126 if (flags & MSG_OOB) { 1127 pru_flags = PRUS_OOB; 1128 async = 0; 1129 } else if ((flags & MSG_EOF) && resid == 0) { 1130 pru_flags = PRUS_EOF; 1131 } else if (resid > 0 && space > 0) { 1132 /* If there is more to send, set PRUS_MORETOCOME */ 1133 pru_flags = PRUS_MORETOCOME; 1134 async = 1; 1135 } else { 1136 pru_flags = 0; 1137 } 1138 1139 if (flags & MSG_SYNC) 1140 async = 0; 1141 1142 /* 1143 * XXX all the SS_CANTSENDMORE checks previously 1144 * done could be out of date. We could have recieved 1145 * a reset packet in an interrupt or maybe we slept 1146 * while doing page faults in uiomove() etc. We could 1147 * probably recheck again inside the splnet() protection 1148 * here, but there are probably other places that this 1149 * also happens. We must rethink this. 1150 */ 1151 for (m = top; m; m = m->m_next) 1152 ssb_preallocstream(&so->so_snd, m); 1153 if (!async) { 1154 error = so_pru_send(so, pru_flags, top, 1155 NULL, NULL, td); 1156 } else { 1157 so_pru_send_async(so, pru_flags, top, 1158 NULL, NULL, td); 1159 error = 0; 1160 } 1161 1162 top = NULL; 1163 mp = ⊤ 1164 if (error) 1165 goto release; 1166 } while (resid && space > 0); 1167 } while (resid); 1168 1169 release: 1170 ssb_unlock(&so->so_snd); 1171 out: 1172 if (top) 1173 m_freem(top); 1174 if (control) 1175 m_freem(control); 1176 return (error); 1177 } 1178 1179 /* 1180 * Implement receive operations on a socket. 1181 * 1182 * We depend on the way that records are added to the signalsockbuf 1183 * by sbappend*. In particular, each record (mbufs linked through m_next) 1184 * must begin with an address if the protocol so specifies, 1185 * followed by an optional mbuf or mbufs containing ancillary data, 1186 * and then zero or more mbufs of data. 1187 * 1188 * Although the signalsockbuf is locked, new data may still be appended. 1189 * A token inside the ssb_lock deals with MP issues and still allows 1190 * the network to access the socket if we block in a uio. 1191 * 1192 * The caller may receive the data as a single mbuf chain by supplying 1193 * an mbuf **mp0 for use in returning the chain. The uio is then used 1194 * only for the count in uio_resid. 1195 */ 1196 int 1197 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 1198 struct sockbuf *sio, struct mbuf **controlp, int *flagsp) 1199 { 1200 struct mbuf *m, *n; 1201 struct mbuf *free_chain = NULL; 1202 int flags, len, error, offset; 1203 struct protosw *pr = so->so_proto; 1204 int moff, type = 0; 1205 size_t resid, orig_resid; 1206 1207 if (uio) 1208 resid = uio->uio_resid; 1209 else 1210 resid = (size_t)(sio->sb_climit - sio->sb_cc); 1211 orig_resid = resid; 1212 1213 if (psa) 1214 *psa = NULL; 1215 if (controlp) 1216 *controlp = NULL; 1217 if (flagsp) 1218 flags = *flagsp &~ MSG_EOR; 1219 else 1220 flags = 0; 1221 if (flags & MSG_OOB) { 1222 m = m_get(MB_WAIT, MT_DATA); 1223 if (m == NULL) 1224 return (ENOBUFS); 1225 error = so_pru_rcvoob(so, m, flags & MSG_PEEK); 1226 if (error) 1227 goto bad; 1228 if (sio) { 1229 do { 1230 sbappend(sio, m); 1231 KKASSERT(resid >= (size_t)m->m_len); 1232 resid -= (size_t)m->m_len; 1233 } while (resid > 0 && m); 1234 } else { 1235 do { 1236 uio->uio_resid = resid; 1237 error = uiomove(mtod(m, caddr_t), 1238 (int)szmin(resid, m->m_len), 1239 uio); 1240 resid = uio->uio_resid; 1241 m = m_free(m); 1242 } while (uio->uio_resid && error == 0 && m); 1243 } 1244 bad: 1245 if (m) 1246 m_freem(m); 1247 return (error); 1248 } 1249 if ((so->so_state & SS_ISCONFIRMING) && resid) 1250 so_pru_rcvd(so, 0); 1251 1252 /* 1253 * The token interlocks against the protocol thread while 1254 * ssb_lock is a blocking lock against other userland entities. 1255 */ 1256 lwkt_gettoken(&so->so_rcv.ssb_token); 1257 restart: 1258 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags)); 1259 if (error) 1260 goto done; 1261 1262 m = so->so_rcv.ssb_mb; 1263 /* 1264 * If we have less data than requested, block awaiting more 1265 * (subject to any timeout) if: 1266 * 1. the current count is less than the low water mark, or 1267 * 2. MSG_WAITALL is set, and it is possible to do the entire 1268 * receive operation at once if we block (resid <= hiwat). 1269 * 3. MSG_DONTWAIT is not set 1270 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1271 * we have to do the receive in sections, and thus risk returning 1272 * a short count if a timeout or signal occurs after we start. 1273 */ 1274 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1275 (size_t)so->so_rcv.ssb_cc < resid) && 1276 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat || 1277 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) && 1278 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1279 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1")); 1280 if (so->so_error) { 1281 if (m) 1282 goto dontblock; 1283 error = so->so_error; 1284 if ((flags & MSG_PEEK) == 0) 1285 so->so_error = 0; 1286 goto release; 1287 } 1288 if (so->so_state & SS_CANTRCVMORE) { 1289 if (m) 1290 goto dontblock; 1291 else 1292 goto release; 1293 } 1294 for (; m; m = m->m_next) { 1295 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1296 m = so->so_rcv.ssb_mb; 1297 goto dontblock; 1298 } 1299 } 1300 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1301 (pr->pr_flags & PR_CONNREQUIRED)) { 1302 error = ENOTCONN; 1303 goto release; 1304 } 1305 if (resid == 0) 1306 goto release; 1307 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) { 1308 error = EWOULDBLOCK; 1309 goto release; 1310 } 1311 ssb_unlock(&so->so_rcv); 1312 error = ssb_wait(&so->so_rcv); 1313 if (error) 1314 goto done; 1315 goto restart; 1316 } 1317 dontblock: 1318 if (uio && uio->uio_td && uio->uio_td->td_proc) 1319 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++; 1320 1321 /* 1322 * note: m should be == sb_mb here. Cache the next record while 1323 * cleaning up. Note that calling m_free*() will break out critical 1324 * section. 1325 */ 1326 KKASSERT(m == so->so_rcv.ssb_mb); 1327 1328 /* 1329 * Skip any address mbufs prepending the record. 1330 */ 1331 if (pr->pr_flags & PR_ADDR) { 1332 KASSERT(m->m_type == MT_SONAME, ("receive 1a")); 1333 orig_resid = 0; 1334 if (psa) 1335 *psa = dup_sockaddr(mtod(m, struct sockaddr *)); 1336 if (flags & MSG_PEEK) 1337 m = m->m_next; 1338 else 1339 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1340 } 1341 1342 /* 1343 * Skip any control mbufs prepending the record. 1344 */ 1345 #ifdef SCTP 1346 if (pr->pr_flags & PR_ADDR_OPT) { 1347 /* 1348 * For SCTP we may be getting a 1349 * whole message OR a partial delivery. 1350 */ 1351 if (m && m->m_type == MT_SONAME) { 1352 orig_resid = 0; 1353 if (psa) 1354 *psa = dup_sockaddr(mtod(m, struct sockaddr *)); 1355 if (flags & MSG_PEEK) 1356 m = m->m_next; 1357 else 1358 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1359 } 1360 } 1361 #endif /* SCTP */ 1362 while (m && m->m_type == MT_CONTROL && error == 0) { 1363 if (flags & MSG_PEEK) { 1364 if (controlp) 1365 *controlp = m_copy(m, 0, m->m_len); 1366 m = m->m_next; /* XXX race */ 1367 } else { 1368 if (controlp) { 1369 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL); 1370 if (pr->pr_domain->dom_externalize && 1371 mtod(m, struct cmsghdr *)->cmsg_type == 1372 SCM_RIGHTS) 1373 error = (*pr->pr_domain->dom_externalize)(m); 1374 *controlp = m; 1375 m = n; 1376 } else { 1377 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1378 } 1379 } 1380 if (controlp && *controlp) { 1381 orig_resid = 0; 1382 controlp = &(*controlp)->m_next; 1383 } 1384 } 1385 1386 /* 1387 * flag OOB data. 1388 */ 1389 if (m) { 1390 type = m->m_type; 1391 if (type == MT_OOBDATA) 1392 flags |= MSG_OOB; 1393 } 1394 1395 /* 1396 * Copy to the UIO or mbuf return chain (*mp). 1397 */ 1398 moff = 0; 1399 offset = 0; 1400 while (m && resid > 0 && error == 0) { 1401 if (m->m_type == MT_OOBDATA) { 1402 if (type != MT_OOBDATA) 1403 break; 1404 } else if (type == MT_OOBDATA) 1405 break; 1406 else 1407 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER, 1408 ("receive 3")); 1409 soclrstate(so, SS_RCVATMARK); 1410 len = (resid > INT_MAX) ? INT_MAX : resid; 1411 if (so->so_oobmark && len > so->so_oobmark - offset) 1412 len = so->so_oobmark - offset; 1413 if (len > m->m_len - moff) 1414 len = m->m_len - moff; 1415 1416 /* 1417 * Copy out to the UIO or pass the mbufs back to the SIO. 1418 * The SIO is dealt with when we eat the mbuf, but deal 1419 * with the resid here either way. 1420 */ 1421 if (uio) { 1422 uio->uio_resid = resid; 1423 error = uiomove(mtod(m, caddr_t) + moff, len, uio); 1424 resid = uio->uio_resid; 1425 if (error) 1426 goto release; 1427 } else { 1428 resid -= (size_t)len; 1429 } 1430 1431 /* 1432 * Eat the entire mbuf or just a piece of it 1433 */ 1434 if (len == m->m_len - moff) { 1435 if (m->m_flags & M_EOR) 1436 flags |= MSG_EOR; 1437 #ifdef SCTP 1438 if (m->m_flags & M_NOTIFICATION) 1439 flags |= MSG_NOTIFICATION; 1440 #endif /* SCTP */ 1441 if (flags & MSG_PEEK) { 1442 m = m->m_next; 1443 moff = 0; 1444 } else { 1445 if (sio) { 1446 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL); 1447 sbappend(sio, m); 1448 m = n; 1449 } else { 1450 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1451 } 1452 } 1453 } else { 1454 if (flags & MSG_PEEK) { 1455 moff += len; 1456 } else { 1457 if (sio) { 1458 n = m_copym(m, 0, len, MB_WAIT); 1459 if (n) 1460 sbappend(sio, n); 1461 } 1462 m->m_data += len; 1463 m->m_len -= len; 1464 so->so_rcv.ssb_cc -= len; 1465 } 1466 } 1467 if (so->so_oobmark) { 1468 if ((flags & MSG_PEEK) == 0) { 1469 so->so_oobmark -= len; 1470 if (so->so_oobmark == 0) { 1471 sosetstate(so, SS_RCVATMARK); 1472 break; 1473 } 1474 } else { 1475 offset += len; 1476 if (offset == so->so_oobmark) 1477 break; 1478 } 1479 } 1480 if (flags & MSG_EOR) 1481 break; 1482 /* 1483 * If the MSG_WAITALL flag is set (for non-atomic socket), 1484 * we must not quit until resid == 0 or an error 1485 * termination. If a signal/timeout occurs, return 1486 * with a short count but without error. 1487 * Keep signalsockbuf locked against other readers. 1488 */ 1489 while ((flags & MSG_WAITALL) && m == NULL && 1490 resid > 0 && !sosendallatonce(so) && 1491 so->so_rcv.ssb_mb == NULL) { 1492 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1493 break; 1494 /* 1495 * The window might have closed to zero, make 1496 * sure we send an ack now that we've drained 1497 * the buffer or we might end up blocking until 1498 * the idle takes over (5 seconds). 1499 */ 1500 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1501 so_pru_rcvd(so, flags); 1502 error = ssb_wait(&so->so_rcv); 1503 if (error) { 1504 ssb_unlock(&so->so_rcv); 1505 error = 0; 1506 goto done; 1507 } 1508 m = so->so_rcv.ssb_mb; 1509 } 1510 } 1511 1512 /* 1513 * If an atomic read was requested but unread data still remains 1514 * in the record, set MSG_TRUNC. 1515 */ 1516 if (m && pr->pr_flags & PR_ATOMIC) 1517 flags |= MSG_TRUNC; 1518 1519 /* 1520 * Cleanup. If an atomic read was requested drop any unread data. 1521 */ 1522 if ((flags & MSG_PEEK) == 0) { 1523 if (m && (pr->pr_flags & PR_ATOMIC)) 1524 sbdroprecord(&so->so_rcv.sb); 1525 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1526 so_pru_rcvd(so, flags); 1527 } 1528 1529 if (orig_resid == resid && orig_resid && 1530 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1531 ssb_unlock(&so->so_rcv); 1532 goto restart; 1533 } 1534 1535 if (flagsp) 1536 *flagsp |= flags; 1537 release: 1538 ssb_unlock(&so->so_rcv); 1539 done: 1540 lwkt_reltoken(&so->so_rcv.ssb_token); 1541 if (free_chain) 1542 m_freem(free_chain); 1543 return (error); 1544 } 1545 1546 int 1547 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio, 1548 struct sockbuf *sio, struct mbuf **controlp, int *flagsp) 1549 { 1550 struct mbuf *m, *n; 1551 struct mbuf *free_chain = NULL; 1552 int flags, len, error, offset; 1553 struct protosw *pr = so->so_proto; 1554 int moff; 1555 size_t resid, orig_resid; 1556 1557 if (uio) 1558 resid = uio->uio_resid; 1559 else 1560 resid = (size_t)(sio->sb_climit - sio->sb_cc); 1561 orig_resid = resid; 1562 1563 if (psa) 1564 *psa = NULL; 1565 if (controlp) 1566 *controlp = NULL; 1567 if (flagsp) 1568 flags = *flagsp &~ MSG_EOR; 1569 else 1570 flags = 0; 1571 if (flags & MSG_OOB) { 1572 m = m_get(MB_WAIT, MT_DATA); 1573 if (m == NULL) 1574 return (ENOBUFS); 1575 error = so_pru_rcvoob(so, m, flags & MSG_PEEK); 1576 if (error) 1577 goto bad; 1578 if (sio) { 1579 do { 1580 sbappend(sio, m); 1581 KKASSERT(resid >= (size_t)m->m_len); 1582 resid -= (size_t)m->m_len; 1583 } while (resid > 0 && m); 1584 } else { 1585 do { 1586 uio->uio_resid = resid; 1587 error = uiomove(mtod(m, caddr_t), 1588 (int)szmin(resid, m->m_len), 1589 uio); 1590 resid = uio->uio_resid; 1591 m = m_free(m); 1592 } while (uio->uio_resid && error == 0 && m); 1593 } 1594 bad: 1595 if (m) 1596 m_freem(m); 1597 return (error); 1598 } 1599 1600 /* 1601 * The token interlocks against the protocol thread while 1602 * ssb_lock is a blocking lock against other userland entities. 1603 */ 1604 lwkt_gettoken(&so->so_rcv.ssb_token); 1605 restart: 1606 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags)); 1607 if (error) 1608 goto done; 1609 1610 m = so->so_rcv.ssb_mb; 1611 /* 1612 * If we have less data than requested, block awaiting more 1613 * (subject to any timeout) if: 1614 * 1. the current count is less than the low water mark, or 1615 * 2. MSG_WAITALL is set, and it is possible to do the entire 1616 * receive operation at once if we block (resid <= hiwat). 1617 * 3. MSG_DONTWAIT is not set 1618 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1619 * we have to do the receive in sections, and thus risk returning 1620 * a short count if a timeout or signal occurs after we start. 1621 */ 1622 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1623 (size_t)so->so_rcv.ssb_cc < resid) && 1624 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat || 1625 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) { 1626 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1")); 1627 if (so->so_error) { 1628 if (m) 1629 goto dontblock; 1630 error = so->so_error; 1631 if ((flags & MSG_PEEK) == 0) 1632 so->so_error = 0; 1633 goto release; 1634 } 1635 if (so->so_state & SS_CANTRCVMORE) { 1636 if (m) 1637 goto dontblock; 1638 else 1639 goto release; 1640 } 1641 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1642 (pr->pr_flags & PR_CONNREQUIRED)) { 1643 error = ENOTCONN; 1644 goto release; 1645 } 1646 if (resid == 0) 1647 goto release; 1648 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) { 1649 error = EWOULDBLOCK; 1650 goto release; 1651 } 1652 ssb_unlock(&so->so_rcv); 1653 error = ssb_wait(&so->so_rcv); 1654 if (error) 1655 goto done; 1656 goto restart; 1657 } 1658 dontblock: 1659 if (uio && uio->uio_td && uio->uio_td->td_proc) 1660 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++; 1661 1662 /* 1663 * note: m should be == sb_mb here. Cache the next record while 1664 * cleaning up. Note that calling m_free*() will break out critical 1665 * section. 1666 */ 1667 KKASSERT(m == so->so_rcv.ssb_mb); 1668 1669 /* 1670 * Copy to the UIO or mbuf return chain (*mp). 1671 */ 1672 moff = 0; 1673 offset = 0; 1674 while (m && resid > 0 && error == 0) { 1675 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER, 1676 ("receive 3")); 1677 1678 soclrstate(so, SS_RCVATMARK); 1679 len = (resid > INT_MAX) ? INT_MAX : resid; 1680 if (so->so_oobmark && len > so->so_oobmark - offset) 1681 len = so->so_oobmark - offset; 1682 if (len > m->m_len - moff) 1683 len = m->m_len - moff; 1684 1685 /* 1686 * Copy out to the UIO or pass the mbufs back to the SIO. 1687 * The SIO is dealt with when we eat the mbuf, but deal 1688 * with the resid here either way. 1689 */ 1690 if (uio) { 1691 uio->uio_resid = resid; 1692 error = uiomove(mtod(m, caddr_t) + moff, len, uio); 1693 resid = uio->uio_resid; 1694 if (error) 1695 goto release; 1696 } else { 1697 resid -= (size_t)len; 1698 } 1699 1700 /* 1701 * Eat the entire mbuf or just a piece of it 1702 */ 1703 if (len == m->m_len - moff) { 1704 if (flags & MSG_PEEK) { 1705 m = m->m_next; 1706 moff = 0; 1707 } else { 1708 if (sio) { 1709 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL); 1710 sbappend(sio, m); 1711 m = n; 1712 } else { 1713 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1714 } 1715 } 1716 } else { 1717 if (flags & MSG_PEEK) { 1718 moff += len; 1719 } else { 1720 if (sio) { 1721 n = m_copym(m, 0, len, MB_WAIT); 1722 if (n) 1723 sbappend(sio, n); 1724 } 1725 m->m_data += len; 1726 m->m_len -= len; 1727 so->so_rcv.ssb_cc -= len; 1728 } 1729 } 1730 if (so->so_oobmark) { 1731 if ((flags & MSG_PEEK) == 0) { 1732 so->so_oobmark -= len; 1733 if (so->so_oobmark == 0) { 1734 sosetstate(so, SS_RCVATMARK); 1735 break; 1736 } 1737 } else { 1738 offset += len; 1739 if (offset == so->so_oobmark) 1740 break; 1741 } 1742 } 1743 /* 1744 * If the MSG_WAITALL flag is set (for non-atomic socket), 1745 * we must not quit until resid == 0 or an error 1746 * termination. If a signal/timeout occurs, return 1747 * with a short count but without error. 1748 * Keep signalsockbuf locked against other readers. 1749 */ 1750 while ((flags & MSG_WAITALL) && m == NULL && 1751 resid > 0 && !sosendallatonce(so) && 1752 so->so_rcv.ssb_mb == NULL) { 1753 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1754 break; 1755 /* 1756 * The window might have closed to zero, make 1757 * sure we send an ack now that we've drained 1758 * the buffer or we might end up blocking until 1759 * the idle takes over (5 seconds). 1760 */ 1761 if (so->so_pcb) 1762 so_pru_rcvd_async(so); 1763 error = ssb_wait(&so->so_rcv); 1764 if (error) { 1765 ssb_unlock(&so->so_rcv); 1766 error = 0; 1767 goto done; 1768 } 1769 m = so->so_rcv.ssb_mb; 1770 } 1771 } 1772 1773 /* 1774 * Cleanup. If an atomic read was requested drop any unread data. 1775 */ 1776 if ((flags & MSG_PEEK) == 0) { 1777 if (so->so_pcb) 1778 so_pru_rcvd_async(so); 1779 } 1780 1781 if (orig_resid == resid && orig_resid && 1782 (so->so_state & SS_CANTRCVMORE) == 0) { 1783 ssb_unlock(&so->so_rcv); 1784 goto restart; 1785 } 1786 1787 if (flagsp) 1788 *flagsp |= flags; 1789 release: 1790 ssb_unlock(&so->so_rcv); 1791 done: 1792 lwkt_reltoken(&so->so_rcv.ssb_token); 1793 if (free_chain) 1794 m_freem(free_chain); 1795 return (error); 1796 } 1797 1798 /* 1799 * Shut a socket down. Note that we do not get a frontend lock as we 1800 * want to be able to shut the socket down even if another thread is 1801 * blocked in a read(), thus waking it up. 1802 */ 1803 int 1804 soshutdown(struct socket *so, int how) 1805 { 1806 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1807 return (EINVAL); 1808 1809 if (how != SHUT_WR) { 1810 /*ssb_lock(&so->so_rcv, M_WAITOK);*/ 1811 sorflush(so); 1812 /*ssb_unlock(&so->so_rcv);*/ 1813 } 1814 if (how != SHUT_RD) 1815 return (so_pru_shutdown(so)); 1816 return (0); 1817 } 1818 1819 void 1820 sorflush(struct socket *so) 1821 { 1822 struct signalsockbuf *ssb = &so->so_rcv; 1823 struct protosw *pr = so->so_proto; 1824 struct signalsockbuf asb; 1825 1826 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR); 1827 1828 lwkt_gettoken(&ssb->ssb_token); 1829 socantrcvmore(so); 1830 asb = *ssb; 1831 1832 /* 1833 * Can't just blow up the ssb structure here 1834 */ 1835 bzero(&ssb->sb, sizeof(ssb->sb)); 1836 ssb->ssb_timeo = 0; 1837 ssb->ssb_lowat = 0; 1838 ssb->ssb_hiwat = 0; 1839 ssb->ssb_mbmax = 0; 1840 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK); 1841 1842 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose) 1843 (*pr->pr_domain->dom_dispose)(asb.ssb_mb); 1844 ssb_release(&asb, so); 1845 1846 lwkt_reltoken(&ssb->ssb_token); 1847 } 1848 1849 #ifdef INET 1850 static int 1851 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt) 1852 { 1853 struct accept_filter_arg *afap = NULL; 1854 struct accept_filter *afp; 1855 struct so_accf *af = so->so_accf; 1856 int error = 0; 1857 1858 /* do not set/remove accept filters on non listen sockets */ 1859 if ((so->so_options & SO_ACCEPTCONN) == 0) { 1860 error = EINVAL; 1861 goto out; 1862 } 1863 1864 /* removing the filter */ 1865 if (sopt == NULL) { 1866 if (af != NULL) { 1867 if (af->so_accept_filter != NULL && 1868 af->so_accept_filter->accf_destroy != NULL) { 1869 af->so_accept_filter->accf_destroy(so); 1870 } 1871 if (af->so_accept_filter_str != NULL) { 1872 kfree(af->so_accept_filter_str, M_ACCF); 1873 } 1874 kfree(af, M_ACCF); 1875 so->so_accf = NULL; 1876 } 1877 so->so_options &= ~SO_ACCEPTFILTER; 1878 return (0); 1879 } 1880 /* adding a filter */ 1881 /* must remove previous filter first */ 1882 if (af != NULL) { 1883 error = EINVAL; 1884 goto out; 1885 } 1886 /* don't put large objects on the kernel stack */ 1887 afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK); 1888 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap); 1889 afap->af_name[sizeof(afap->af_name)-1] = '\0'; 1890 afap->af_arg[sizeof(afap->af_arg)-1] = '\0'; 1891 if (error) 1892 goto out; 1893 afp = accept_filt_get(afap->af_name); 1894 if (afp == NULL) { 1895 error = ENOENT; 1896 goto out; 1897 } 1898 af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO); 1899 if (afp->accf_create != NULL) { 1900 if (afap->af_name[0] != '\0') { 1901 int len = strlen(afap->af_name) + 1; 1902 1903 af->so_accept_filter_str = kmalloc(len, M_ACCF, 1904 M_WAITOK); 1905 strcpy(af->so_accept_filter_str, afap->af_name); 1906 } 1907 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg); 1908 if (af->so_accept_filter_arg == NULL) { 1909 kfree(af->so_accept_filter_str, M_ACCF); 1910 kfree(af, M_ACCF); 1911 so->so_accf = NULL; 1912 error = EINVAL; 1913 goto out; 1914 } 1915 } 1916 af->so_accept_filter = afp; 1917 so->so_accf = af; 1918 so->so_options |= SO_ACCEPTFILTER; 1919 out: 1920 if (afap != NULL) 1921 kfree(afap, M_TEMP); 1922 return (error); 1923 } 1924 #endif /* INET */ 1925 1926 /* 1927 * Perhaps this routine, and sooptcopyout(), below, ought to come in 1928 * an additional variant to handle the case where the option value needs 1929 * to be some kind of integer, but not a specific size. 1930 * In addition to their use here, these functions are also called by the 1931 * protocol-level pr_ctloutput() routines. 1932 */ 1933 int 1934 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 1935 { 1936 return soopt_to_kbuf(sopt, buf, len, minlen); 1937 } 1938 1939 int 1940 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 1941 { 1942 size_t valsize; 1943 1944 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 1945 KKASSERT(kva_p(buf)); 1946 1947 /* 1948 * If the user gives us more than we wanted, we ignore it, 1949 * but if we don't get the minimum length the caller 1950 * wants, we return EINVAL. On success, sopt->sopt_valsize 1951 * is set to however much we actually retrieved. 1952 */ 1953 if ((valsize = sopt->sopt_valsize) < minlen) 1954 return EINVAL; 1955 if (valsize > len) 1956 sopt->sopt_valsize = valsize = len; 1957 1958 bcopy(sopt->sopt_val, buf, valsize); 1959 return 0; 1960 } 1961 1962 1963 int 1964 sosetopt(struct socket *so, struct sockopt *sopt) 1965 { 1966 int error, optval; 1967 struct linger l; 1968 struct timeval tv; 1969 u_long val; 1970 struct signalsockbuf *sotmp; 1971 1972 error = 0; 1973 sopt->sopt_dir = SOPT_SET; 1974 if (sopt->sopt_level != SOL_SOCKET) { 1975 if (so->so_proto && so->so_proto->pr_ctloutput) { 1976 return (so_pr_ctloutput(so, sopt)); 1977 } 1978 error = ENOPROTOOPT; 1979 } else { 1980 switch (sopt->sopt_name) { 1981 #ifdef INET 1982 case SO_ACCEPTFILTER: 1983 error = do_setopt_accept_filter(so, sopt); 1984 if (error) 1985 goto bad; 1986 break; 1987 #endif /* INET */ 1988 case SO_LINGER: 1989 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 1990 if (error) 1991 goto bad; 1992 1993 so->so_linger = l.l_linger; 1994 if (l.l_onoff) 1995 so->so_options |= SO_LINGER; 1996 else 1997 so->so_options &= ~SO_LINGER; 1998 break; 1999 2000 case SO_DEBUG: 2001 case SO_KEEPALIVE: 2002 case SO_DONTROUTE: 2003 case SO_USELOOPBACK: 2004 case SO_BROADCAST: 2005 case SO_REUSEADDR: 2006 case SO_REUSEPORT: 2007 case SO_OOBINLINE: 2008 case SO_TIMESTAMP: 2009 case SO_NOSIGPIPE: 2010 error = sooptcopyin(sopt, &optval, sizeof optval, 2011 sizeof optval); 2012 if (error) 2013 goto bad; 2014 if (optval) 2015 so->so_options |= sopt->sopt_name; 2016 else 2017 so->so_options &= ~sopt->sopt_name; 2018 break; 2019 2020 case SO_SNDBUF: 2021 case SO_RCVBUF: 2022 case SO_SNDLOWAT: 2023 case SO_RCVLOWAT: 2024 error = sooptcopyin(sopt, &optval, sizeof optval, 2025 sizeof optval); 2026 if (error) 2027 goto bad; 2028 2029 /* 2030 * Values < 1 make no sense for any of these 2031 * options, so disallow them. 2032 */ 2033 if (optval < 1) { 2034 error = EINVAL; 2035 goto bad; 2036 } 2037 2038 switch (sopt->sopt_name) { 2039 case SO_SNDBUF: 2040 case SO_RCVBUF: 2041 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ? 2042 &so->so_snd : &so->so_rcv, (u_long)optval, 2043 so, 2044 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) { 2045 error = ENOBUFS; 2046 goto bad; 2047 } 2048 sotmp = (sopt->sopt_name == SO_SNDBUF) ? 2049 &so->so_snd : &so->so_rcv; 2050 atomic_clear_int(&sotmp->ssb_flags, 2051 SSB_AUTOSIZE); 2052 break; 2053 2054 /* 2055 * Make sure the low-water is never greater than 2056 * the high-water. 2057 */ 2058 case SO_SNDLOWAT: 2059 so->so_snd.ssb_lowat = 2060 (optval > so->so_snd.ssb_hiwat) ? 2061 so->so_snd.ssb_hiwat : optval; 2062 atomic_clear_int(&so->so_snd.ssb_flags, 2063 SSB_AUTOLOWAT); 2064 break; 2065 case SO_RCVLOWAT: 2066 so->so_rcv.ssb_lowat = 2067 (optval > so->so_rcv.ssb_hiwat) ? 2068 so->so_rcv.ssb_hiwat : optval; 2069 atomic_clear_int(&so->so_rcv.ssb_flags, 2070 SSB_AUTOLOWAT); 2071 break; 2072 } 2073 break; 2074 2075 case SO_SNDTIMEO: 2076 case SO_RCVTIMEO: 2077 error = sooptcopyin(sopt, &tv, sizeof tv, 2078 sizeof tv); 2079 if (error) 2080 goto bad; 2081 2082 /* assert(hz > 0); */ 2083 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz || 2084 tv.tv_usec < 0 || tv.tv_usec >= 1000000) { 2085 error = EDOM; 2086 goto bad; 2087 } 2088 /* assert(tick > 0); */ 2089 /* assert(ULONG_MAX - INT_MAX >= 1000000); */ 2090 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick; 2091 if (val > INT_MAX) { 2092 error = EDOM; 2093 goto bad; 2094 } 2095 if (val == 0 && tv.tv_usec != 0) 2096 val = 1; 2097 2098 switch (sopt->sopt_name) { 2099 case SO_SNDTIMEO: 2100 so->so_snd.ssb_timeo = val; 2101 break; 2102 case SO_RCVTIMEO: 2103 so->so_rcv.ssb_timeo = val; 2104 break; 2105 } 2106 break; 2107 default: 2108 error = ENOPROTOOPT; 2109 break; 2110 } 2111 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 2112 (void) so_pr_ctloutput(so, sopt); 2113 } 2114 } 2115 bad: 2116 return (error); 2117 } 2118 2119 /* Helper routine for getsockopt */ 2120 int 2121 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 2122 { 2123 soopt_from_kbuf(sopt, buf, len); 2124 return 0; 2125 } 2126 2127 void 2128 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len) 2129 { 2130 size_t valsize; 2131 2132 if (len == 0) { 2133 sopt->sopt_valsize = 0; 2134 return; 2135 } 2136 2137 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 2138 KKASSERT(kva_p(buf)); 2139 2140 /* 2141 * Documented get behavior is that we always return a value, 2142 * possibly truncated to fit in the user's buffer. 2143 * Traditional behavior is that we always tell the user 2144 * precisely how much we copied, rather than something useful 2145 * like the total amount we had available for her. 2146 * Note that this interface is not idempotent; the entire answer must 2147 * generated ahead of time. 2148 */ 2149 valsize = szmin(len, sopt->sopt_valsize); 2150 sopt->sopt_valsize = valsize; 2151 if (sopt->sopt_val != 0) { 2152 bcopy(buf, sopt->sopt_val, valsize); 2153 } 2154 } 2155 2156 int 2157 sogetopt(struct socket *so, struct sockopt *sopt) 2158 { 2159 int error, optval; 2160 long optval_l; 2161 struct linger l; 2162 struct timeval tv; 2163 #ifdef INET 2164 struct accept_filter_arg *afap; 2165 #endif 2166 2167 error = 0; 2168 sopt->sopt_dir = SOPT_GET; 2169 if (sopt->sopt_level != SOL_SOCKET) { 2170 if (so->so_proto && so->so_proto->pr_ctloutput) { 2171 return (so_pr_ctloutput(so, sopt)); 2172 } else 2173 return (ENOPROTOOPT); 2174 } else { 2175 switch (sopt->sopt_name) { 2176 #ifdef INET 2177 case SO_ACCEPTFILTER: 2178 if ((so->so_options & SO_ACCEPTCONN) == 0) 2179 return (EINVAL); 2180 afap = kmalloc(sizeof(*afap), M_TEMP, 2181 M_WAITOK | M_ZERO); 2182 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 2183 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name); 2184 if (so->so_accf->so_accept_filter_str != NULL) 2185 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str); 2186 } 2187 error = sooptcopyout(sopt, afap, sizeof(*afap)); 2188 kfree(afap, M_TEMP); 2189 break; 2190 #endif /* INET */ 2191 2192 case SO_LINGER: 2193 l.l_onoff = so->so_options & SO_LINGER; 2194 l.l_linger = so->so_linger; 2195 error = sooptcopyout(sopt, &l, sizeof l); 2196 break; 2197 2198 case SO_USELOOPBACK: 2199 case SO_DONTROUTE: 2200 case SO_DEBUG: 2201 case SO_KEEPALIVE: 2202 case SO_REUSEADDR: 2203 case SO_REUSEPORT: 2204 case SO_BROADCAST: 2205 case SO_OOBINLINE: 2206 case SO_TIMESTAMP: 2207 case SO_NOSIGPIPE: 2208 optval = so->so_options & sopt->sopt_name; 2209 integer: 2210 error = sooptcopyout(sopt, &optval, sizeof optval); 2211 break; 2212 2213 case SO_TYPE: 2214 optval = so->so_type; 2215 goto integer; 2216 2217 case SO_ERROR: 2218 optval = so->so_error; 2219 so->so_error = 0; 2220 goto integer; 2221 2222 case SO_SNDBUF: 2223 optval = so->so_snd.ssb_hiwat; 2224 goto integer; 2225 2226 case SO_RCVBUF: 2227 optval = so->so_rcv.ssb_hiwat; 2228 goto integer; 2229 2230 case SO_SNDLOWAT: 2231 optval = so->so_snd.ssb_lowat; 2232 goto integer; 2233 2234 case SO_RCVLOWAT: 2235 optval = so->so_rcv.ssb_lowat; 2236 goto integer; 2237 2238 case SO_SNDTIMEO: 2239 case SO_RCVTIMEO: 2240 optval = (sopt->sopt_name == SO_SNDTIMEO ? 2241 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo); 2242 2243 tv.tv_sec = optval / hz; 2244 tv.tv_usec = (optval % hz) * ustick; 2245 error = sooptcopyout(sopt, &tv, sizeof tv); 2246 break; 2247 2248 case SO_SNDSPACE: 2249 optval_l = ssb_space(&so->so_snd); 2250 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l)); 2251 break; 2252 2253 default: 2254 error = ENOPROTOOPT; 2255 break; 2256 } 2257 return (error); 2258 } 2259 } 2260 2261 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */ 2262 int 2263 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 2264 { 2265 struct mbuf *m, *m_prev; 2266 int sopt_size = sopt->sopt_valsize, msize; 2267 2268 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA, 2269 0, &msize); 2270 if (m == NULL) 2271 return (ENOBUFS); 2272 m->m_len = min(msize, sopt_size); 2273 sopt_size -= m->m_len; 2274 *mp = m; 2275 m_prev = m; 2276 2277 while (sopt_size > 0) { 2278 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, 2279 MT_DATA, 0, &msize); 2280 if (m == NULL) { 2281 m_freem(*mp); 2282 return (ENOBUFS); 2283 } 2284 m->m_len = min(msize, sopt_size); 2285 sopt_size -= m->m_len; 2286 m_prev->m_next = m; 2287 m_prev = m; 2288 } 2289 return (0); 2290 } 2291 2292 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */ 2293 int 2294 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 2295 { 2296 soopt_to_mbuf(sopt, m); 2297 return 0; 2298 } 2299 2300 void 2301 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m) 2302 { 2303 size_t valsize; 2304 void *val; 2305 2306 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 2307 KKASSERT(kva_p(m)); 2308 if (sopt->sopt_val == NULL) 2309 return; 2310 val = sopt->sopt_val; 2311 valsize = sopt->sopt_valsize; 2312 while (m != NULL && valsize >= m->m_len) { 2313 bcopy(val, mtod(m, char *), m->m_len); 2314 valsize -= m->m_len; 2315 val = (caddr_t)val + m->m_len; 2316 m = m->m_next; 2317 } 2318 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 2319 panic("ip6_sooptmcopyin"); 2320 } 2321 2322 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */ 2323 int 2324 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 2325 { 2326 return soopt_from_mbuf(sopt, m); 2327 } 2328 2329 int 2330 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m) 2331 { 2332 struct mbuf *m0 = m; 2333 size_t valsize = 0; 2334 size_t maxsize; 2335 void *val; 2336 2337 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 2338 KKASSERT(kva_p(m)); 2339 if (sopt->sopt_val == NULL) 2340 return 0; 2341 val = sopt->sopt_val; 2342 maxsize = sopt->sopt_valsize; 2343 while (m != NULL && maxsize >= m->m_len) { 2344 bcopy(mtod(m, char *), val, m->m_len); 2345 maxsize -= m->m_len; 2346 val = (caddr_t)val + m->m_len; 2347 valsize += m->m_len; 2348 m = m->m_next; 2349 } 2350 if (m != NULL) { 2351 /* enough soopt buffer should be given from user-land */ 2352 m_freem(m0); 2353 return (EINVAL); 2354 } 2355 sopt->sopt_valsize = valsize; 2356 return 0; 2357 } 2358 2359 void 2360 sohasoutofband(struct socket *so) 2361 { 2362 if (so->so_sigio != NULL) 2363 pgsigio(so->so_sigio, SIGURG, 0); 2364 KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB); 2365 } 2366 2367 int 2368 sokqfilter(struct file *fp, struct knote *kn) 2369 { 2370 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2371 struct signalsockbuf *ssb; 2372 2373 switch (kn->kn_filter) { 2374 case EVFILT_READ: 2375 if (so->so_options & SO_ACCEPTCONN) 2376 kn->kn_fop = &solisten_filtops; 2377 else 2378 kn->kn_fop = &soread_filtops; 2379 ssb = &so->so_rcv; 2380 break; 2381 case EVFILT_WRITE: 2382 kn->kn_fop = &sowrite_filtops; 2383 ssb = &so->so_snd; 2384 break; 2385 case EVFILT_EXCEPT: 2386 kn->kn_fop = &soexcept_filtops; 2387 ssb = &so->so_rcv; 2388 break; 2389 default: 2390 return (EOPNOTSUPP); 2391 } 2392 2393 knote_insert(&ssb->ssb_kq.ki_note, kn); 2394 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE); 2395 return (0); 2396 } 2397 2398 static void 2399 filt_sordetach(struct knote *kn) 2400 { 2401 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2402 2403 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn); 2404 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note)) 2405 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE); 2406 } 2407 2408 /*ARGSUSED*/ 2409 static int 2410 filt_soread(struct knote *kn, long hint) 2411 { 2412 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2413 2414 if (kn->kn_sfflags & NOTE_OOB) { 2415 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) { 2416 kn->kn_fflags |= NOTE_OOB; 2417 return (1); 2418 } 2419 return (0); 2420 } 2421 kn->kn_data = so->so_rcv.ssb_cc; 2422 2423 if (so->so_state & SS_CANTRCVMORE) { 2424 /* 2425 * Only set NODATA if all data has been exhausted. 2426 */ 2427 if (kn->kn_data == 0) 2428 kn->kn_flags |= EV_NODATA; 2429 kn->kn_flags |= EV_EOF; 2430 kn->kn_fflags = so->so_error; 2431 return (1); 2432 } 2433 if (so->so_error) /* temporary udp error */ 2434 return (1); 2435 if (kn->kn_sfflags & NOTE_LOWAT) 2436 return (kn->kn_data >= kn->kn_sdata); 2437 return ((kn->kn_data >= so->so_rcv.ssb_lowat) || 2438 !TAILQ_EMPTY(&so->so_comp)); 2439 } 2440 2441 static void 2442 filt_sowdetach(struct knote *kn) 2443 { 2444 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2445 2446 knote_remove(&so->so_snd.ssb_kq.ki_note, kn); 2447 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note)) 2448 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE); 2449 } 2450 2451 /*ARGSUSED*/ 2452 static int 2453 filt_sowrite(struct knote *kn, long hint) 2454 { 2455 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2456 2457 kn->kn_data = ssb_space(&so->so_snd); 2458 if (so->so_state & SS_CANTSENDMORE) { 2459 kn->kn_flags |= (EV_EOF | EV_NODATA); 2460 kn->kn_fflags = so->so_error; 2461 return (1); 2462 } 2463 if (so->so_error) /* temporary udp error */ 2464 return (1); 2465 if (((so->so_state & SS_ISCONNECTED) == 0) && 2466 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 2467 return (0); 2468 if (kn->kn_sfflags & NOTE_LOWAT) 2469 return (kn->kn_data >= kn->kn_sdata); 2470 return (kn->kn_data >= so->so_snd.ssb_lowat); 2471 } 2472 2473 /*ARGSUSED*/ 2474 static int 2475 filt_solisten(struct knote *kn, long hint) 2476 { 2477 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2478 2479 kn->kn_data = so->so_qlen; 2480 return (! TAILQ_EMPTY(&so->so_comp)); 2481 } 2482