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. Neither the name of the University nor the names of its contributors 47 * may be used to endorse or promote products derived from this software 48 * without specific prior written permission. 49 * 50 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 53 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 60 * SUCH DAMAGE. 61 * 62 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 63 * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.24 2003/11/11 17:18:18 silby Exp $ 64 */ 65 66 #include "opt_inet.h" 67 68 #include <sys/param.h> 69 #include <sys/systm.h> 70 #include <sys/fcntl.h> 71 #include <sys/malloc.h> 72 #include <sys/mbuf.h> 73 #include <sys/domain.h> 74 #include <sys/file.h> /* for struct knote */ 75 #include <sys/kernel.h> 76 #include <sys/event.h> 77 #include <sys/proc.h> 78 #include <sys/protosw.h> 79 #include <sys/socket.h> 80 #include <sys/socketvar.h> 81 #include <sys/socketops.h> 82 #include <sys/resourcevar.h> 83 #include <sys/signalvar.h> 84 #include <sys/sysctl.h> 85 #include <sys/uio.h> 86 #include <sys/jail.h> 87 #include <vm/vm_zone.h> 88 #include <vm/pmap.h> 89 #include <net/netmsg2.h> 90 #include <net/netisr2.h> 91 92 #include <sys/thread2.h> 93 #include <sys/socketvar2.h> 94 #include <sys/spinlock2.h> 95 96 #include <machine/limits.h> 97 98 #ifdef INET 99 extern int tcp_sosend_agglim; 100 extern int tcp_sosend_async; 101 extern int tcp_sosend_jcluster; 102 extern int udp_sosend_async; 103 extern int udp_sosend_prepend; 104 105 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt); 106 #endif /* INET */ 107 108 static void filt_sordetach(struct knote *kn); 109 static int filt_soread(struct knote *kn, long hint); 110 static void filt_sowdetach(struct knote *kn); 111 static int filt_sowrite(struct knote *kn, long hint); 112 static int filt_solisten(struct knote *kn, long hint); 113 114 static int soclose_sync(struct socket *so, int fflag); 115 static void soclose_fast(struct socket *so); 116 117 static struct filterops solisten_filtops = 118 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten }; 119 static struct filterops soread_filtops = 120 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread }; 121 static struct filterops sowrite_filtops = 122 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite }; 123 static struct filterops soexcept_filtops = 124 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread }; 125 126 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct"); 127 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 128 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 129 130 131 static int somaxconn = SOMAXCONN; 132 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW, 133 &somaxconn, 0, "Maximum pending socket connection queue size"); 134 135 static int use_soclose_fast = 1; 136 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW, 137 &use_soclose_fast, 0, "Fast socket close"); 138 139 int use_soaccept_pred_fast = 1; 140 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW, 141 &use_soaccept_pred_fast, 0, "Fast socket accept predication"); 142 143 int use_sendfile_async = 1; 144 SYSCTL_INT(_kern_ipc, OID_AUTO, sendfile_async, CTLFLAG_RW, 145 &use_sendfile_async, 0, "sendfile uses asynchronized pru_send"); 146 147 int use_soconnect_async = 1; 148 SYSCTL_INT(_kern_ipc, OID_AUTO, soconnect_async, CTLFLAG_RW, 149 &use_soconnect_async, 0, "soconnect uses asynchronized pru_connect"); 150 151 static int use_socreate_fast = 1; 152 SYSCTL_INT(_kern_ipc, OID_AUTO, socreate_fast, CTLFLAG_RW, 153 &use_socreate_fast, 0, "Fast socket creation"); 154 155 /* 156 * Socket operation routines. 157 * These routines are called by the routines in 158 * sys_socket.c or from a system process, and 159 * implement the semantics of socket operations by 160 * switching out to the protocol specific routines. 161 */ 162 163 /* 164 * Get a socket structure, and initialize it. 165 * Note that it would probably be better to allocate socket 166 * and PCB at the same time, but I'm not convinced that all 167 * the protocols can be easily modified to do this. 168 */ 169 struct socket * 170 soalloc(int waitok, struct protosw *pr) 171 { 172 struct socket *so; 173 unsigned waitmask; 174 175 waitmask = waitok ? M_WAITOK : M_NOWAIT; 176 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask); 177 if (so) { 178 /* XXX race condition for reentrant kernel */ 179 so->so_proto = pr; 180 TAILQ_INIT(&so->so_aiojobq); 181 TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist); 182 TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist); 183 lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok"); 184 lwkt_token_init(&so->so_snd.ssb_token, "sndtok"); 185 spin_init(&so->so_rcvd_spin, "soalloc"); 186 netmsg_init(&so->so_rcvd_msg.base, so, &netisr_adone_rport, 187 MSGF_DROPABLE | MSGF_PRIORITY, 188 so->so_proto->pr_usrreqs->pru_rcvd); 189 so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC; 190 so->so_state = SS_NOFDREF; 191 so->so_refs = 1; 192 } 193 return so; 194 } 195 196 int 197 socreate(int dom, struct socket **aso, int type, 198 int proto, struct thread *td) 199 { 200 struct proc *p = td->td_proc; 201 struct protosw *prp; 202 struct socket *so; 203 struct pru_attach_info ai; 204 int error; 205 206 if (proto) 207 prp = pffindproto(dom, proto, type); 208 else 209 prp = pffindtype(dom, type); 210 211 if (prp == NULL || prp->pr_usrreqs->pru_attach == 0) 212 return (EPROTONOSUPPORT); 213 214 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only && 215 prp->pr_domain->dom_family != PF_LOCAL && 216 prp->pr_domain->dom_family != PF_INET && 217 prp->pr_domain->dom_family != PF_INET6 && 218 prp->pr_domain->dom_family != PF_ROUTE) { 219 return (EPROTONOSUPPORT); 220 } 221 222 if (prp->pr_type != type) 223 return (EPROTOTYPE); 224 so = soalloc(p != NULL, prp); 225 if (so == NULL) 226 return (ENOBUFS); 227 228 /* 229 * Callers of socreate() presumably will connect up a descriptor 230 * and call soclose() if they cannot. This represents our so_refs 231 * (which should be 1) from soalloc(). 232 */ 233 soclrstate(so, SS_NOFDREF); 234 235 /* 236 * Set a default port for protocol processing. No action will occur 237 * on the socket on this port until an inpcb is attached to it and 238 * is able to match incoming packets, or until the socket becomes 239 * available to userland. 240 * 241 * We normally default the socket to the protocol thread on cpu 0, 242 * if protocol does not provide its own method to initialize the 243 * default port. 244 * 245 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol 246 * thread and all pr_*()/pru_*() calls are executed synchronously. 247 */ 248 if (prp->pr_flags & PR_SYNC_PORT) 249 so->so_port = &netisr_sync_port; 250 else if (prp->pr_initport != NULL) 251 so->so_port = prp->pr_initport(); 252 else 253 so->so_port = netisr_cpuport(0); 254 255 TAILQ_INIT(&so->so_incomp); 256 TAILQ_INIT(&so->so_comp); 257 so->so_type = type; 258 so->so_cred = crhold(p->p_ucred); 259 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE]; 260 ai.p_ucred = p->p_ucred; 261 ai.fd_rdir = p->p_fd->fd_rdir; 262 263 /* 264 * Auto-sizing of socket buffers is managed by the protocols and 265 * the appropriate flags must be set in the pru_attach function. 266 */ 267 if (use_socreate_fast && prp->pr_usrreqs->pru_preattach) 268 error = so_pru_attach_fast(so, proto, &ai); 269 else 270 error = so_pru_attach(so, proto, &ai); 271 if (error) { 272 sosetstate(so, SS_NOFDREF); 273 sofree(so); /* from soalloc */ 274 return error; 275 } 276 277 /* 278 * NOTE: Returns referenced socket. 279 */ 280 *aso = so; 281 return (0); 282 } 283 284 int 285 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 286 { 287 int error; 288 289 error = so_pru_bind(so, nam, td); 290 return (error); 291 } 292 293 static void 294 sodealloc(struct socket *so) 295 { 296 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 0); 297 /* TODO: assert accept queues are empty, after unix socket is fixed */ 298 299 if (so->so_rcv.ssb_hiwat) 300 (void)chgsbsize(so->so_cred->cr_uidinfo, 301 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY); 302 if (so->so_snd.ssb_hiwat) 303 (void)chgsbsize(so->so_cred->cr_uidinfo, 304 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY); 305 #ifdef INET 306 /* remove accept filter if present */ 307 if (so->so_accf != NULL) 308 do_setopt_accept_filter(so, NULL); 309 #endif /* INET */ 310 crfree(so->so_cred); 311 if (so->so_faddr != NULL) 312 kfree(so->so_faddr, M_SONAME); 313 kfree(so, M_SOCKET); 314 } 315 316 int 317 solisten(struct socket *so, int backlog, struct thread *td) 318 { 319 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) 320 return (EINVAL); 321 322 lwkt_gettoken(&so->so_rcv.ssb_token); 323 if (TAILQ_EMPTY(&so->so_comp)) 324 so->so_options |= SO_ACCEPTCONN; 325 lwkt_reltoken(&so->so_rcv.ssb_token); 326 if (backlog < 0 || backlog > somaxconn) 327 backlog = somaxconn; 328 so->so_qlimit = backlog; 329 return so_pru_listen(so, td); 330 } 331 332 static void 333 soqflush(struct socket *so) 334 { 335 lwkt_getpooltoken(so); 336 if (so->so_options & SO_ACCEPTCONN) { 337 struct socket *sp; 338 339 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) { 340 KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) == 341 SS_INCOMP); 342 TAILQ_REMOVE(&so->so_incomp, sp, so_list); 343 so->so_incqlen--; 344 soclrstate(sp, SS_INCOMP); 345 soabort_async(sp, TRUE); 346 } 347 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) { 348 KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) == 349 SS_COMP); 350 TAILQ_REMOVE(&so->so_comp, sp, so_list); 351 so->so_qlen--; 352 soclrstate(sp, SS_COMP); 353 soabort_async(sp, TRUE); 354 } 355 } 356 lwkt_relpooltoken(so); 357 } 358 359 /* 360 * Destroy a disconnected socket. This routine is a NOP if entities 361 * still have a reference on the socket: 362 * 363 * so_pcb - The protocol stack still has a reference 364 * SS_NOFDREF - There is no longer a file pointer reference 365 */ 366 void 367 sofree(struct socket *so) 368 { 369 struct socket *head; 370 371 /* 372 * This is a bit hackish at the moment. We need to interlock 373 * any accept queue we are on before we potentially lose the 374 * last reference to avoid races against a re-reference from 375 * someone operating on the queue. 376 */ 377 while ((head = so->so_head) != NULL) { 378 lwkt_getpooltoken(head); 379 if (so->so_head == head) 380 break; 381 lwkt_relpooltoken(head); 382 } 383 384 /* 385 * Arbitrage the last free. 386 */ 387 KKASSERT(so->so_refs > 0); 388 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) { 389 if (head) 390 lwkt_relpooltoken(head); 391 return; 392 } 393 394 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF)); 395 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0); 396 397 if (head != NULL) { 398 /* 399 * We're done, remove ourselves from the accept queue we are 400 * on, if we are on one. 401 */ 402 if (so->so_state & SS_INCOMP) { 403 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 404 SS_INCOMP); 405 TAILQ_REMOVE(&head->so_incomp, so, so_list); 406 head->so_incqlen--; 407 } else if (so->so_state & SS_COMP) { 408 /* 409 * We must not decommission a socket that's 410 * on the accept(2) queue. If we do, then 411 * accept(2) may hang after select(2) indicated 412 * that the listening socket was ready. 413 */ 414 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 415 SS_COMP); 416 lwkt_relpooltoken(head); 417 return; 418 } else { 419 panic("sofree: not queued"); 420 } 421 soclrstate(so, SS_INCOMP); 422 so->so_head = NULL; 423 lwkt_relpooltoken(head); 424 } else { 425 /* Flush accept queues, if we are accepting. */ 426 soqflush(so); 427 } 428 ssb_release(&so->so_snd, so); 429 sorflush(so); 430 sodealloc(so); 431 } 432 433 /* 434 * Close a socket on last file table reference removal. 435 * Initiate disconnect if connected. 436 * Free socket when disconnect complete. 437 */ 438 int 439 soclose(struct socket *so, int fflag) 440 { 441 int error; 442 443 funsetown(&so->so_sigio); 444 sosetstate(so, SS_ISCLOSING); 445 if (!use_soclose_fast || 446 (so->so_proto->pr_flags & PR_SYNC_PORT) || 447 ((so->so_state & SS_ISCONNECTED) && 448 (so->so_options & SO_LINGER))) { 449 error = soclose_sync(so, fflag); 450 } else { 451 soclose_fast(so); 452 error = 0; 453 } 454 return error; 455 } 456 457 void 458 sodiscard(struct socket *so) 459 { 460 if (so->so_state & SS_NOFDREF) 461 panic("soclose: NOFDREF"); 462 sosetstate(so, SS_NOFDREF); /* take ref */ 463 } 464 465 /* 466 * Append the completed queue of head to head_inh (inherting listen socket). 467 */ 468 void 469 soinherit(struct socket *head, struct socket *head_inh) 470 { 471 boolean_t do_wakeup = FALSE; 472 473 KASSERT(head->so_options & SO_ACCEPTCONN, 474 ("head does not accept connection")); 475 KASSERT(head_inh->so_options & SO_ACCEPTCONN, 476 ("head_inh does not accept connection")); 477 478 lwkt_getpooltoken(head); 479 lwkt_getpooltoken(head_inh); 480 481 if (head->so_qlen > 0) 482 do_wakeup = TRUE; 483 484 while (!TAILQ_EMPTY(&head->so_comp)) { 485 struct ucred *old_cr; 486 struct socket *sp; 487 488 sp = TAILQ_FIRST(&head->so_comp); 489 KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP); 490 491 /* 492 * Remove this socket from the current listen socket 493 * completed queue. 494 */ 495 TAILQ_REMOVE(&head->so_comp, sp, so_list); 496 head->so_qlen--; 497 498 /* Save the old ucred for later free. */ 499 old_cr = sp->so_cred; 500 501 /* 502 * Install this socket to the inheriting listen socket 503 * completed queue. 504 */ 505 sp->so_cred = crhold(head_inh->so_cred); /* non-blocking */ 506 sp->so_head = head_inh; 507 508 TAILQ_INSERT_TAIL(&head_inh->so_comp, sp, so_list); 509 head_inh->so_qlen++; 510 511 /* 512 * NOTE: 513 * crfree() may block and release the tokens temporarily. 514 * However, we are fine here, since the transition is done. 515 */ 516 crfree(old_cr); 517 } 518 519 lwkt_relpooltoken(head_inh); 520 lwkt_relpooltoken(head); 521 522 if (do_wakeup) { 523 /* 524 * "New" connections have arrived 525 */ 526 sorwakeup(head_inh); 527 wakeup(&head_inh->so_timeo); 528 } 529 } 530 531 static int 532 soclose_sync(struct socket *so, int fflag) 533 { 534 int error = 0; 535 536 if ((so->so_proto->pr_flags & PR_SYNC_PORT) == 0) 537 so_pru_sync(so); /* unpend async prus */ 538 539 if (so->so_pcb == NULL) 540 goto discard; 541 542 if (so->so_state & SS_ISCONNECTED) { 543 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 544 error = sodisconnect(so); 545 if (error) 546 goto drop; 547 } 548 if (so->so_options & SO_LINGER) { 549 if ((so->so_state & SS_ISDISCONNECTING) && 550 (fflag & FNONBLOCK)) 551 goto drop; 552 while (so->so_state & SS_ISCONNECTED) { 553 error = tsleep(&so->so_timeo, PCATCH, 554 "soclos", so->so_linger * hz); 555 if (error) 556 break; 557 } 558 } 559 } 560 drop: 561 if (so->so_pcb) { 562 int error2; 563 564 error2 = so_pru_detach(so); 565 if (error2 == EJUSTRETURN) { 566 /* 567 * Protocol will call sodiscard() 568 * and sofree() for us. 569 */ 570 return error; 571 } 572 if (error == 0) 573 error = error2; 574 } 575 discard: 576 sodiscard(so); 577 sofree(so); /* dispose of ref */ 578 579 return (error); 580 } 581 582 static void 583 soclose_fast_handler(netmsg_t msg) 584 { 585 struct socket *so = msg->base.nm_so; 586 587 if (so->so_pcb == NULL) 588 goto discard; 589 590 if ((so->so_state & SS_ISCONNECTED) && 591 (so->so_state & SS_ISDISCONNECTING) == 0) 592 so_pru_disconnect_direct(so); 593 594 if (so->so_pcb) { 595 int error; 596 597 error = so_pru_detach_direct(so); 598 if (error == EJUSTRETURN) { 599 /* 600 * Protocol will call sodiscard() 601 * and sofree() for us. 602 */ 603 return; 604 } 605 } 606 discard: 607 sodiscard(so); 608 sofree(so); 609 } 610 611 static void 612 soclose_fast(struct socket *so) 613 { 614 struct netmsg_base *base = &so->so_clomsg; 615 616 netmsg_init(base, so, &netisr_apanic_rport, 0, 617 soclose_fast_handler); 618 lwkt_sendmsg(so->so_port, &base->lmsg); 619 } 620 621 /* 622 * Abort and destroy a socket. Only one abort can be in progress 623 * at any given moment. 624 */ 625 void 626 soabort_async(struct socket *so, boolean_t clr_head) 627 { 628 /* 629 * Keep a reference before clearing the so_head 630 * to avoid racing socket close in netisr. 631 */ 632 soreference(so); 633 if (clr_head) 634 so->so_head = NULL; 635 so_pru_abort_async(so); 636 } 637 638 void 639 soabort_direct(struct socket *so) 640 { 641 soreference(so); 642 so_pru_abort_direct(so); 643 } 644 645 /* 646 * so is passed in ref'd, which becomes owned by 647 * the cleared SS_NOFDREF flag. 648 */ 649 void 650 soaccept_generic(struct socket *so) 651 { 652 if ((so->so_state & SS_NOFDREF) == 0) 653 panic("soaccept: !NOFDREF"); 654 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */ 655 } 656 657 int 658 soaccept(struct socket *so, struct sockaddr **nam) 659 { 660 int error; 661 662 soaccept_generic(so); 663 error = so_pru_accept(so, nam); 664 return (error); 665 } 666 667 int 668 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td, 669 boolean_t sync) 670 { 671 int error; 672 673 if (so->so_options & SO_ACCEPTCONN) 674 return (EOPNOTSUPP); 675 /* 676 * If protocol is connection-based, can only connect once. 677 * Otherwise, if connected, try to disconnect first. 678 * This allows user to disconnect by connecting to, e.g., 679 * a null address. 680 */ 681 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 682 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 683 (error = sodisconnect(so)))) { 684 error = EISCONN; 685 } else { 686 /* 687 * Prevent accumulated error from previous connection 688 * from biting us. 689 */ 690 so->so_error = 0; 691 if (!sync && so->so_proto->pr_usrreqs->pru_preconnect) 692 error = so_pru_connect_async(so, nam, td); 693 else 694 error = so_pru_connect(so, nam, td); 695 } 696 return (error); 697 } 698 699 int 700 soconnect2(struct socket *so1, struct socket *so2) 701 { 702 int error; 703 704 error = so_pru_connect2(so1, so2); 705 return (error); 706 } 707 708 int 709 sodisconnect(struct socket *so) 710 { 711 int error; 712 713 if ((so->so_state & SS_ISCONNECTED) == 0) { 714 error = ENOTCONN; 715 goto bad; 716 } 717 if (so->so_state & SS_ISDISCONNECTING) { 718 error = EALREADY; 719 goto bad; 720 } 721 error = so_pru_disconnect(so); 722 bad: 723 return (error); 724 } 725 726 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 727 /* 728 * Send on a socket. 729 * If send must go all at once and message is larger than 730 * send buffering, then hard error. 731 * Lock against other senders. 732 * If must go all at once and not enough room now, then 733 * inform user that this would block and do nothing. 734 * Otherwise, if nonblocking, send as much as possible. 735 * The data to be sent is described by "uio" if nonzero, 736 * otherwise by the mbuf chain "top" (which must be null 737 * if uio is not). Data provided in mbuf chain must be small 738 * enough to send all at once. 739 * 740 * Returns nonzero on error, timeout or signal; callers 741 * must check for short counts if EINTR/ERESTART are returned. 742 * Data and control buffers are freed on return. 743 */ 744 int 745 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 746 struct mbuf *top, struct mbuf *control, int flags, 747 struct thread *td) 748 { 749 struct mbuf **mp; 750 struct mbuf *m; 751 size_t resid; 752 int space, len; 753 int clen = 0, error, dontroute, mlen; 754 int atomic = sosendallatonce(so) || top; 755 int pru_flags; 756 757 if (uio) { 758 resid = uio->uio_resid; 759 } else { 760 resid = (size_t)top->m_pkthdr.len; 761 #ifdef INVARIANTS 762 len = 0; 763 for (m = top; m; m = m->m_next) 764 len += m->m_len; 765 KKASSERT(top->m_pkthdr.len == len); 766 #endif 767 } 768 769 /* 770 * WARNING! resid is unsigned, space and len are signed. space 771 * can wind up negative if the sockbuf is overcommitted. 772 * 773 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 774 * type sockets since that's an error. 775 */ 776 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) { 777 error = EINVAL; 778 goto out; 779 } 780 781 dontroute = 782 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 783 (so->so_proto->pr_flags & PR_ATOMIC); 784 if (td->td_lwp != NULL) 785 td->td_lwp->lwp_ru.ru_msgsnd++; 786 if (control) 787 clen = control->m_len; 788 #define gotoerr(errcode) { error = errcode; goto release; } 789 790 restart: 791 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags)); 792 if (error) 793 goto out; 794 795 do { 796 if (so->so_state & SS_CANTSENDMORE) 797 gotoerr(EPIPE); 798 if (so->so_error) { 799 error = so->so_error; 800 so->so_error = 0; 801 goto release; 802 } 803 if ((so->so_state & SS_ISCONNECTED) == 0) { 804 /* 805 * `sendto' and `sendmsg' is allowed on a connection- 806 * based socket if it supports implied connect. 807 * Return ENOTCONN if not connected and no address is 808 * supplied. 809 */ 810 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 811 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 812 if ((so->so_state & SS_ISCONFIRMING) == 0 && 813 !(resid == 0 && clen != 0)) 814 gotoerr(ENOTCONN); 815 } else if (addr == NULL) 816 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ? 817 ENOTCONN : EDESTADDRREQ); 818 } 819 if ((atomic && resid > so->so_snd.ssb_hiwat) || 820 clen > so->so_snd.ssb_hiwat) { 821 gotoerr(EMSGSIZE); 822 } 823 space = ssb_space(&so->so_snd); 824 if (flags & MSG_OOB) 825 space += 1024; 826 if ((space < 0 || (size_t)space < resid + clen) && uio && 827 (atomic || space < so->so_snd.ssb_lowat || space < clen)) { 828 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) 829 gotoerr(EWOULDBLOCK); 830 ssb_unlock(&so->so_snd); 831 error = ssb_wait(&so->so_snd); 832 if (error) 833 goto out; 834 goto restart; 835 } 836 mp = ⊤ 837 space -= clen; 838 do { 839 if (uio == NULL) { 840 /* 841 * Data is prepackaged in "top". 842 */ 843 resid = 0; 844 if (flags & MSG_EOR) 845 top->m_flags |= M_EOR; 846 } else do { 847 if (resid > INT_MAX) 848 resid = INT_MAX; 849 m = m_getl((int)resid, M_WAITOK, MT_DATA, 850 top == NULL ? M_PKTHDR : 0, &mlen); 851 if (top == NULL) { 852 m->m_pkthdr.len = 0; 853 m->m_pkthdr.rcvif = NULL; 854 } 855 len = imin((int)szmin(mlen, resid), space); 856 if (resid < MINCLSIZE) { 857 /* 858 * For datagram protocols, leave room 859 * for protocol headers in first mbuf. 860 */ 861 if (atomic && top == NULL && len < mlen) 862 MH_ALIGN(m, len); 863 } 864 space -= len; 865 error = uiomove(mtod(m, caddr_t), (size_t)len, uio); 866 resid = uio->uio_resid; 867 m->m_len = len; 868 *mp = m; 869 top->m_pkthdr.len += len; 870 if (error) 871 goto release; 872 mp = &m->m_next; 873 if (resid == 0) { 874 if (flags & MSG_EOR) 875 top->m_flags |= M_EOR; 876 break; 877 } 878 } while (space > 0 && atomic); 879 if (dontroute) 880 so->so_options |= SO_DONTROUTE; 881 if (flags & MSG_OOB) { 882 pru_flags = PRUS_OOB; 883 } else if ((flags & MSG_EOF) && 884 (so->so_proto->pr_flags & PR_IMPLOPCL) && 885 (resid == 0)) { 886 /* 887 * If the user set MSG_EOF, the protocol 888 * understands this flag and nothing left to 889 * send then use PRU_SEND_EOF instead of PRU_SEND. 890 */ 891 pru_flags = PRUS_EOF; 892 } else if (resid > 0 && space > 0) { 893 /* If there is more to send, set PRUS_MORETOCOME */ 894 pru_flags = PRUS_MORETOCOME; 895 } else { 896 pru_flags = 0; 897 } 898 /* 899 * XXX all the SS_CANTSENDMORE checks previously 900 * done could be out of date. We could have recieved 901 * a reset packet in an interrupt or maybe we slept 902 * while doing page faults in uiomove() etc. We could 903 * probably recheck again inside the splnet() protection 904 * here, but there are probably other places that this 905 * also happens. We must rethink this. 906 */ 907 error = so_pru_send(so, pru_flags, top, addr, control, td); 908 if (dontroute) 909 so->so_options &= ~SO_DONTROUTE; 910 clen = 0; 911 control = NULL; 912 top = NULL; 913 mp = ⊤ 914 if (error) 915 goto release; 916 } while (resid && space > 0); 917 } while (resid); 918 919 release: 920 ssb_unlock(&so->so_snd); 921 out: 922 if (top) 923 m_freem(top); 924 if (control) 925 m_freem(control); 926 return (error); 927 } 928 929 #ifdef INET 930 /* 931 * A specialization of sosend() for UDP based on protocol-specific knowledge: 932 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that 933 * sosendallatonce() returns true, 934 * the "atomic" variable is true, 935 * and sosendudp() blocks until space is available for the entire send. 936 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or 937 * PR_IMPLOPCL flags set. 938 * UDP has no out-of-band data. 939 * UDP has no control data. 940 * UDP does not support MSG_EOR. 941 */ 942 int 943 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio, 944 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 945 { 946 size_t resid; 947 int error, pru_flags = 0; 948 int space; 949 950 if (td->td_lwp != NULL) 951 td->td_lwp->lwp_ru.ru_msgsnd++; 952 if (control) 953 m_freem(control); 954 955 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp")); 956 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len; 957 958 restart: 959 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags)); 960 if (error) 961 goto out; 962 963 if (so->so_state & SS_CANTSENDMORE) 964 gotoerr(EPIPE); 965 if (so->so_error) { 966 error = so->so_error; 967 so->so_error = 0; 968 goto release; 969 } 970 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL) 971 gotoerr(EDESTADDRREQ); 972 if (resid > so->so_snd.ssb_hiwat) 973 gotoerr(EMSGSIZE); 974 space = ssb_space(&so->so_snd); 975 if (uio && (space < 0 || (size_t)space < resid)) { 976 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) 977 gotoerr(EWOULDBLOCK); 978 ssb_unlock(&so->so_snd); 979 error = ssb_wait(&so->so_snd); 980 if (error) 981 goto out; 982 goto restart; 983 } 984 985 if (uio) { 986 int hdrlen = max_hdr; 987 988 /* 989 * We try to optimize out the additional mbuf 990 * allocations in M_PREPEND() on output path, e.g. 991 * - udp_output(), when it tries to prepend protocol 992 * headers. 993 * - Link layer output function, when it tries to 994 * prepend link layer header. 995 * 996 * This probably will not benefit any data that will 997 * be fragmented, so this optimization is only performed 998 * when the size of data and max size of protocol+link 999 * headers fit into one mbuf cluster. 1000 */ 1001 if (uio->uio_resid > MCLBYTES - hdrlen || 1002 !udp_sosend_prepend) { 1003 top = m_uiomove(uio); 1004 if (top == NULL) 1005 goto release; 1006 } else { 1007 int nsize; 1008 1009 top = m_getl(uio->uio_resid + hdrlen, M_WAITOK, 1010 MT_DATA, M_PKTHDR, &nsize); 1011 KASSERT(nsize >= uio->uio_resid + hdrlen, 1012 ("sosendudp invalid nsize %d, " 1013 "resid %zu, hdrlen %d", 1014 nsize, uio->uio_resid, hdrlen)); 1015 1016 top->m_len = uio->uio_resid; 1017 top->m_pkthdr.len = uio->uio_resid; 1018 top->m_data += hdrlen; 1019 1020 error = uiomove(mtod(top, caddr_t), top->m_len, uio); 1021 if (error) 1022 goto out; 1023 } 1024 } 1025 1026 if (flags & MSG_DONTROUTE) 1027 pru_flags |= PRUS_DONTROUTE; 1028 1029 if (udp_sosend_async && (flags & MSG_SYNC) == 0) { 1030 so_pru_send_async(so, pru_flags, top, addr, NULL, td); 1031 error = 0; 1032 } else { 1033 error = so_pru_send(so, pru_flags, top, addr, NULL, td); 1034 } 1035 top = NULL; /* sent or freed in lower layer */ 1036 1037 release: 1038 ssb_unlock(&so->so_snd); 1039 out: 1040 if (top) 1041 m_freem(top); 1042 return (error); 1043 } 1044 1045 int 1046 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio, 1047 struct mbuf *top, struct mbuf *control, int flags, 1048 struct thread *td) 1049 { 1050 struct mbuf **mp; 1051 struct mbuf *m; 1052 size_t resid; 1053 int space, len; 1054 int error, mlen; 1055 int allatonce; 1056 int pru_flags; 1057 1058 if (uio) { 1059 KKASSERT(top == NULL); 1060 allatonce = 0; 1061 resid = uio->uio_resid; 1062 } else { 1063 allatonce = 1; 1064 resid = (size_t)top->m_pkthdr.len; 1065 #ifdef INVARIANTS 1066 len = 0; 1067 for (m = top; m; m = m->m_next) 1068 len += m->m_len; 1069 KKASSERT(top->m_pkthdr.len == len); 1070 #endif 1071 } 1072 1073 /* 1074 * WARNING! resid is unsigned, space and len are signed. space 1075 * can wind up negative if the sockbuf is overcommitted. 1076 * 1077 * Also check to make sure that MSG_EOR isn't used on TCP 1078 */ 1079 if (flags & MSG_EOR) { 1080 error = EINVAL; 1081 goto out; 1082 } 1083 1084 if (control) { 1085 /* TCP doesn't do control messages (rights, creds, etc) */ 1086 if (control->m_len) { 1087 error = EINVAL; 1088 goto out; 1089 } 1090 m_freem(control); /* empty control, just free it */ 1091 control = NULL; 1092 } 1093 1094 if (td->td_lwp != NULL) 1095 td->td_lwp->lwp_ru.ru_msgsnd++; 1096 1097 #define gotoerr(errcode) { error = errcode; goto release; } 1098 1099 restart: 1100 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags)); 1101 if (error) 1102 goto out; 1103 1104 do { 1105 if (so->so_state & SS_CANTSENDMORE) 1106 gotoerr(EPIPE); 1107 if (so->so_error) { 1108 error = so->so_error; 1109 so->so_error = 0; 1110 goto release; 1111 } 1112 if ((so->so_state & SS_ISCONNECTED) == 0 && 1113 (so->so_state & SS_ISCONFIRMING) == 0) 1114 gotoerr(ENOTCONN); 1115 if (allatonce && resid > so->so_snd.ssb_hiwat) 1116 gotoerr(EMSGSIZE); 1117 1118 space = ssb_space_prealloc(&so->so_snd); 1119 if (flags & MSG_OOB) 1120 space += 1024; 1121 if ((space < 0 || (size_t)space < resid) && !allatonce && 1122 space < so->so_snd.ssb_lowat) { 1123 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) 1124 gotoerr(EWOULDBLOCK); 1125 ssb_unlock(&so->so_snd); 1126 error = ssb_wait(&so->so_snd); 1127 if (error) 1128 goto out; 1129 goto restart; 1130 } 1131 mp = ⊤ 1132 do { 1133 int cnt = 0, async = 0; 1134 1135 if (uio == NULL) { 1136 /* 1137 * Data is prepackaged in "top". 1138 */ 1139 resid = 0; 1140 } else do { 1141 if (resid > INT_MAX) 1142 resid = INT_MAX; 1143 if (tcp_sosend_jcluster) { 1144 m = m_getlj((int)resid, M_WAITOK, MT_DATA, 1145 top == NULL ? M_PKTHDR : 0, &mlen); 1146 } else { 1147 m = m_getl((int)resid, M_WAITOK, MT_DATA, 1148 top == NULL ? M_PKTHDR : 0, &mlen); 1149 } 1150 if (top == NULL) { 1151 m->m_pkthdr.len = 0; 1152 m->m_pkthdr.rcvif = NULL; 1153 } 1154 len = imin((int)szmin(mlen, resid), space); 1155 space -= len; 1156 error = uiomove(mtod(m, caddr_t), (size_t)len, uio); 1157 resid = uio->uio_resid; 1158 m->m_len = len; 1159 *mp = m; 1160 top->m_pkthdr.len += len; 1161 if (error) 1162 goto release; 1163 mp = &m->m_next; 1164 if (resid == 0) 1165 break; 1166 ++cnt; 1167 } while (space > 0 && cnt < tcp_sosend_agglim); 1168 1169 if (tcp_sosend_async) 1170 async = 1; 1171 1172 if (flags & MSG_OOB) { 1173 pru_flags = PRUS_OOB; 1174 async = 0; 1175 } else if ((flags & MSG_EOF) && resid == 0) { 1176 pru_flags = PRUS_EOF; 1177 } else if (resid > 0 && space > 0) { 1178 /* If there is more to send, set PRUS_MORETOCOME */ 1179 pru_flags = PRUS_MORETOCOME; 1180 async = 1; 1181 } else { 1182 pru_flags = 0; 1183 } 1184 1185 if (flags & MSG_SYNC) 1186 async = 0; 1187 1188 /* 1189 * XXX all the SS_CANTSENDMORE checks previously 1190 * done could be out of date. We could have recieved 1191 * a reset packet in an interrupt or maybe we slept 1192 * while doing page faults in uiomove() etc. We could 1193 * probably recheck again inside the splnet() protection 1194 * here, but there are probably other places that this 1195 * also happens. We must rethink this. 1196 */ 1197 for (m = top; m; m = m->m_next) 1198 ssb_preallocstream(&so->so_snd, m); 1199 if (!async) { 1200 error = so_pru_send(so, pru_flags, top, 1201 NULL, NULL, td); 1202 } else { 1203 so_pru_send_async(so, pru_flags, top, 1204 NULL, NULL, td); 1205 error = 0; 1206 } 1207 1208 top = NULL; 1209 mp = ⊤ 1210 if (error) 1211 goto release; 1212 } while (resid && space > 0); 1213 } while (resid); 1214 1215 release: 1216 ssb_unlock(&so->so_snd); 1217 out: 1218 if (top) 1219 m_freem(top); 1220 if (control) 1221 m_freem(control); 1222 return (error); 1223 } 1224 #endif 1225 1226 /* 1227 * Implement receive operations on a socket. 1228 * 1229 * We depend on the way that records are added to the signalsockbuf 1230 * by sbappend*. In particular, each record (mbufs linked through m_next) 1231 * must begin with an address if the protocol so specifies, 1232 * followed by an optional mbuf or mbufs containing ancillary data, 1233 * and then zero or more mbufs of data. 1234 * 1235 * Although the signalsockbuf is locked, new data may still be appended. 1236 * A token inside the ssb_lock deals with MP issues and still allows 1237 * the network to access the socket if we block in a uio. 1238 * 1239 * The caller may receive the data as a single mbuf chain by supplying 1240 * an mbuf **mp0 for use in returning the chain. The uio is then used 1241 * only for the count in uio_resid. 1242 */ 1243 int 1244 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 1245 struct sockbuf *sio, struct mbuf **controlp, int *flagsp) 1246 { 1247 struct mbuf *m, *n; 1248 struct mbuf *free_chain = NULL; 1249 int flags, len, error, offset; 1250 struct protosw *pr = so->so_proto; 1251 int moff, type = 0; 1252 size_t resid, orig_resid; 1253 boolean_t free_rights = FALSE; 1254 1255 if (uio) 1256 resid = uio->uio_resid; 1257 else 1258 resid = (size_t)(sio->sb_climit - sio->sb_cc); 1259 orig_resid = resid; 1260 1261 if (psa) 1262 *psa = NULL; 1263 if (controlp) 1264 *controlp = NULL; 1265 if (flagsp) 1266 flags = *flagsp &~ MSG_EOR; 1267 else 1268 flags = 0; 1269 if (flags & MSG_OOB) { 1270 m = m_get(M_WAITOK, MT_DATA); 1271 if (m == NULL) 1272 return (ENOBUFS); 1273 error = so_pru_rcvoob(so, m, flags & MSG_PEEK); 1274 if (error) 1275 goto bad; 1276 if (sio) { 1277 do { 1278 sbappend(sio, m); 1279 KKASSERT(resid >= (size_t)m->m_len); 1280 resid -= (size_t)m->m_len; 1281 } while (resid > 0 && m); 1282 } else { 1283 do { 1284 uio->uio_resid = resid; 1285 error = uiomove(mtod(m, caddr_t), 1286 (int)szmin(resid, m->m_len), 1287 uio); 1288 resid = uio->uio_resid; 1289 m = m_free(m); 1290 } while (uio->uio_resid && error == 0 && m); 1291 } 1292 bad: 1293 if (m) 1294 m_freem(m); 1295 return (error); 1296 } 1297 if ((so->so_state & SS_ISCONFIRMING) && resid) 1298 so_pru_rcvd(so, 0); 1299 1300 /* 1301 * The token interlocks against the protocol thread while 1302 * ssb_lock is a blocking lock against other userland entities. 1303 */ 1304 lwkt_gettoken(&so->so_rcv.ssb_token); 1305 restart: 1306 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags)); 1307 if (error) 1308 goto done; 1309 1310 m = so->so_rcv.ssb_mb; 1311 /* 1312 * If we have less data than requested, block awaiting more 1313 * (subject to any timeout) if: 1314 * 1. the current count is less than the low water mark, or 1315 * 2. MSG_WAITALL is set, and it is possible to do the entire 1316 * receive operation at once if we block (resid <= hiwat). 1317 * 3. MSG_DONTWAIT is not set 1318 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1319 * we have to do the receive in sections, and thus risk returning 1320 * a short count if a timeout or signal occurs after we start. 1321 */ 1322 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1323 (size_t)so->so_rcv.ssb_cc < resid) && 1324 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat || 1325 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) && 1326 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1327 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1")); 1328 if (so->so_error) { 1329 if (m) 1330 goto dontblock; 1331 error = so->so_error; 1332 if ((flags & MSG_PEEK) == 0) 1333 so->so_error = 0; 1334 goto release; 1335 } 1336 if (so->so_state & SS_CANTRCVMORE) { 1337 if (m) 1338 goto dontblock; 1339 else 1340 goto release; 1341 } 1342 for (; m; m = m->m_next) { 1343 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1344 m = so->so_rcv.ssb_mb; 1345 goto dontblock; 1346 } 1347 } 1348 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1349 (pr->pr_flags & PR_CONNREQUIRED)) { 1350 error = ENOTCONN; 1351 goto release; 1352 } 1353 if (resid == 0) 1354 goto release; 1355 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) { 1356 error = EWOULDBLOCK; 1357 goto release; 1358 } 1359 ssb_unlock(&so->so_rcv); 1360 error = ssb_wait(&so->so_rcv); 1361 if (error) 1362 goto done; 1363 goto restart; 1364 } 1365 dontblock: 1366 if (uio && uio->uio_td && uio->uio_td->td_proc) 1367 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++; 1368 1369 /* 1370 * note: m should be == sb_mb here. Cache the next record while 1371 * cleaning up. Note that calling m_free*() will break out critical 1372 * section. 1373 */ 1374 KKASSERT(m == so->so_rcv.ssb_mb); 1375 1376 /* 1377 * Skip any address mbufs prepending the record. 1378 */ 1379 if (pr->pr_flags & PR_ADDR) { 1380 KASSERT(m->m_type == MT_SONAME, ("receive 1a")); 1381 orig_resid = 0; 1382 if (psa) 1383 *psa = dup_sockaddr(mtod(m, struct sockaddr *)); 1384 if (flags & MSG_PEEK) 1385 m = m->m_next; 1386 else 1387 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1388 } 1389 1390 /* 1391 * Skip any control mbufs prepending the record. 1392 */ 1393 while (m && m->m_type == MT_CONTROL && error == 0) { 1394 if (flags & MSG_PEEK) { 1395 if (controlp) 1396 *controlp = m_copy(m, 0, m->m_len); 1397 m = m->m_next; /* XXX race */ 1398 } else { 1399 const struct cmsghdr *cm = mtod(m, struct cmsghdr *); 1400 1401 if (controlp) { 1402 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL); 1403 if (pr->pr_domain->dom_externalize && 1404 cm->cmsg_level == SOL_SOCKET && 1405 cm->cmsg_type == SCM_RIGHTS) { 1406 error = pr->pr_domain->dom_externalize 1407 (m, flags); 1408 } 1409 *controlp = m; 1410 m = n; 1411 } else { 1412 if (cm->cmsg_level == SOL_SOCKET && 1413 cm->cmsg_type == SCM_RIGHTS) 1414 free_rights = TRUE; 1415 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1416 } 1417 } 1418 if (controlp && *controlp) { 1419 orig_resid = 0; 1420 controlp = &(*controlp)->m_next; 1421 } 1422 } 1423 1424 /* 1425 * flag OOB data. 1426 */ 1427 if (m) { 1428 type = m->m_type; 1429 if (type == MT_OOBDATA) 1430 flags |= MSG_OOB; 1431 } 1432 1433 /* 1434 * Copy to the UIO or mbuf return chain (*mp). 1435 */ 1436 moff = 0; 1437 offset = 0; 1438 while (m && resid > 0 && error == 0) { 1439 if (m->m_type == MT_OOBDATA) { 1440 if (type != MT_OOBDATA) 1441 break; 1442 } else if (type == MT_OOBDATA) 1443 break; 1444 else 1445 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER, 1446 ("receive 3")); 1447 soclrstate(so, SS_RCVATMARK); 1448 len = (resid > INT_MAX) ? INT_MAX : resid; 1449 if (so->so_oobmark && len > so->so_oobmark - offset) 1450 len = so->so_oobmark - offset; 1451 if (len > m->m_len - moff) 1452 len = m->m_len - moff; 1453 1454 /* 1455 * Copy out to the UIO or pass the mbufs back to the SIO. 1456 * The SIO is dealt with when we eat the mbuf, but deal 1457 * with the resid here either way. 1458 */ 1459 if (uio) { 1460 uio->uio_resid = resid; 1461 error = uiomove(mtod(m, caddr_t) + moff, len, uio); 1462 resid = uio->uio_resid; 1463 if (error) 1464 goto release; 1465 } else { 1466 resid -= (size_t)len; 1467 } 1468 1469 /* 1470 * Eat the entire mbuf or just a piece of it 1471 */ 1472 if (len == m->m_len - moff) { 1473 if (m->m_flags & M_EOR) 1474 flags |= MSG_EOR; 1475 if (flags & MSG_PEEK) { 1476 m = m->m_next; 1477 moff = 0; 1478 } else { 1479 if (sio) { 1480 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL); 1481 sbappend(sio, m); 1482 m = n; 1483 } else { 1484 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain); 1485 } 1486 } 1487 } else { 1488 if (flags & MSG_PEEK) { 1489 moff += len; 1490 } else { 1491 if (sio) { 1492 n = m_copym(m, 0, len, M_WAITOK); 1493 if (n) 1494 sbappend(sio, n); 1495 } 1496 m->m_data += len; 1497 m->m_len -= len; 1498 so->so_rcv.ssb_cc -= len; 1499 } 1500 } 1501 if (so->so_oobmark) { 1502 if ((flags & MSG_PEEK) == 0) { 1503 so->so_oobmark -= len; 1504 if (so->so_oobmark == 0) { 1505 sosetstate(so, SS_RCVATMARK); 1506 break; 1507 } 1508 } else { 1509 offset += len; 1510 if (offset == so->so_oobmark) 1511 break; 1512 } 1513 } 1514 if (flags & MSG_EOR) 1515 break; 1516 /* 1517 * If the MSG_WAITALL flag is set (for non-atomic socket), 1518 * we must not quit until resid == 0 or an error 1519 * termination. If a signal/timeout occurs, return 1520 * with a short count but without error. 1521 * Keep signalsockbuf locked against other readers. 1522 */ 1523 while ((flags & MSG_WAITALL) && m == NULL && 1524 resid > 0 && !sosendallatonce(so) && 1525 so->so_rcv.ssb_mb == NULL) { 1526 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1527 break; 1528 /* 1529 * The window might have closed to zero, make 1530 * sure we send an ack now that we've drained 1531 * the buffer or we might end up blocking until 1532 * the idle takes over (5 seconds). 1533 */ 1534 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1535 so_pru_rcvd(so, flags); 1536 error = ssb_wait(&so->so_rcv); 1537 if (error) { 1538 ssb_unlock(&so->so_rcv); 1539 error = 0; 1540 goto done; 1541 } 1542 m = so->so_rcv.ssb_mb; 1543 } 1544 } 1545 1546 /* 1547 * If an atomic read was requested but unread data still remains 1548 * in the record, set MSG_TRUNC. 1549 */ 1550 if (m && pr->pr_flags & PR_ATOMIC) 1551 flags |= MSG_TRUNC; 1552 1553 /* 1554 * Cleanup. If an atomic read was requested drop any unread data. 1555 */ 1556 if ((flags & MSG_PEEK) == 0) { 1557 if (m && (pr->pr_flags & PR_ATOMIC)) 1558 sbdroprecord(&so->so_rcv.sb); 1559 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1560 so_pru_rcvd(so, flags); 1561 } 1562 1563 if (orig_resid == resid && orig_resid && 1564 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1565 ssb_unlock(&so->so_rcv); 1566 goto restart; 1567 } 1568 1569 if (flagsp) 1570 *flagsp |= flags; 1571 release: 1572 ssb_unlock(&so->so_rcv); 1573 done: 1574 lwkt_reltoken(&so->so_rcv.ssb_token); 1575 if (free_chain) { 1576 if (free_rights && (pr->pr_flags & PR_RIGHTS) && 1577 pr->pr_domain->dom_dispose) 1578 pr->pr_domain->dom_dispose(free_chain); 1579 m_freem(free_chain); 1580 } 1581 return (error); 1582 } 1583 1584 int 1585 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio, 1586 struct sockbuf *sio, struct mbuf **controlp, int *flagsp) 1587 { 1588 struct mbuf *m, *n; 1589 struct mbuf *free_chain = NULL; 1590 int flags, len, error, offset; 1591 struct protosw *pr = so->so_proto; 1592 int moff; 1593 int didoob; 1594 size_t resid, orig_resid, restmp; 1595 1596 if (uio) 1597 resid = uio->uio_resid; 1598 else 1599 resid = (size_t)(sio->sb_climit - sio->sb_cc); 1600 orig_resid = resid; 1601 1602 if (psa) 1603 *psa = NULL; 1604 if (controlp) 1605 *controlp = NULL; 1606 if (flagsp) 1607 flags = *flagsp &~ MSG_EOR; 1608 else 1609 flags = 0; 1610 if (flags & MSG_OOB) { 1611 m = m_get(M_WAITOK, MT_DATA); 1612 if (m == NULL) 1613 return (ENOBUFS); 1614 error = so_pru_rcvoob(so, m, flags & MSG_PEEK); 1615 if (error) 1616 goto bad; 1617 if (sio) { 1618 do { 1619 sbappend(sio, m); 1620 KKASSERT(resid >= (size_t)m->m_len); 1621 resid -= (size_t)m->m_len; 1622 } while (resid > 0 && m); 1623 } else { 1624 do { 1625 uio->uio_resid = resid; 1626 error = uiomove(mtod(m, caddr_t), 1627 (int)szmin(resid, m->m_len), 1628 uio); 1629 resid = uio->uio_resid; 1630 m = m_free(m); 1631 } while (uio->uio_resid && error == 0 && m); 1632 } 1633 bad: 1634 if (m) 1635 m_freem(m); 1636 return (error); 1637 } 1638 1639 /* 1640 * The token interlocks against the protocol thread while 1641 * ssb_lock is a blocking lock against other userland entities. 1642 * 1643 * Lock a limited number of mbufs (not all, so sbcompress() still 1644 * works well). The token is used as an interlock for sbwait() so 1645 * release it afterwords. 1646 */ 1647 restart: 1648 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags)); 1649 if (error) 1650 goto done; 1651 1652 lwkt_gettoken(&so->so_rcv.ssb_token); 1653 m = so->so_rcv.ssb_mb; 1654 1655 /* 1656 * If we have less data than requested, block awaiting more 1657 * (subject to any timeout) if: 1658 * 1. the current count is less than the low water mark, or 1659 * 2. MSG_WAITALL is set, and it is possible to do the entire 1660 * receive operation at once if we block (resid <= hiwat). 1661 * 3. MSG_DONTWAIT is not set 1662 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1663 * we have to do the receive in sections, and thus risk returning 1664 * a short count if a timeout or signal occurs after we start. 1665 */ 1666 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1667 (size_t)so->so_rcv.ssb_cc < resid) && 1668 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat || 1669 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) { 1670 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1")); 1671 if (so->so_error) { 1672 if (m) 1673 goto dontblock; 1674 lwkt_reltoken(&so->so_rcv.ssb_token); 1675 error = so->so_error; 1676 if ((flags & MSG_PEEK) == 0) 1677 so->so_error = 0; 1678 goto release; 1679 } 1680 if (so->so_state & SS_CANTRCVMORE) { 1681 if (m) 1682 goto dontblock; 1683 lwkt_reltoken(&so->so_rcv.ssb_token); 1684 goto release; 1685 } 1686 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1687 (pr->pr_flags & PR_CONNREQUIRED)) { 1688 lwkt_reltoken(&so->so_rcv.ssb_token); 1689 error = ENOTCONN; 1690 goto release; 1691 } 1692 if (resid == 0) { 1693 lwkt_reltoken(&so->so_rcv.ssb_token); 1694 goto release; 1695 } 1696 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) { 1697 lwkt_reltoken(&so->so_rcv.ssb_token); 1698 error = EWOULDBLOCK; 1699 goto release; 1700 } 1701 ssb_unlock(&so->so_rcv); 1702 error = ssb_wait(&so->so_rcv); 1703 lwkt_reltoken(&so->so_rcv.ssb_token); 1704 if (error) 1705 goto done; 1706 goto restart; 1707 } 1708 1709 /* 1710 * Token still held 1711 */ 1712 dontblock: 1713 n = m; 1714 restmp = 0; 1715 while (n && restmp < resid) { 1716 n->m_flags |= M_SOLOCKED; 1717 restmp += n->m_len; 1718 if (n->m_next == NULL) 1719 n = n->m_nextpkt; 1720 else 1721 n = n->m_next; 1722 } 1723 1724 /* 1725 * Release token for loop 1726 */ 1727 lwkt_reltoken(&so->so_rcv.ssb_token); 1728 if (uio && uio->uio_td && uio->uio_td->td_proc) 1729 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++; 1730 1731 /* 1732 * note: m should be == sb_mb here. Cache the next record while 1733 * cleaning up. Note that calling m_free*() will break out critical 1734 * section. 1735 */ 1736 KKASSERT(m == so->so_rcv.ssb_mb); 1737 1738 /* 1739 * Copy to the UIO or mbuf return chain (*mp). 1740 * 1741 * NOTE: Token is not held for loop 1742 */ 1743 moff = 0; 1744 offset = 0; 1745 didoob = 0; 1746 1747 while (m && (m->m_flags & M_SOLOCKED) && resid > 0 && error == 0) { 1748 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER, 1749 ("receive 3")); 1750 1751 soclrstate(so, SS_RCVATMARK); 1752 len = (resid > INT_MAX) ? INT_MAX : resid; 1753 if (so->so_oobmark && len > so->so_oobmark - offset) 1754 len = so->so_oobmark - offset; 1755 if (len > m->m_len - moff) 1756 len = m->m_len - moff; 1757 1758 /* 1759 * Copy out to the UIO or pass the mbufs back to the SIO. 1760 * The SIO is dealt with when we eat the mbuf, but deal 1761 * with the resid here either way. 1762 */ 1763 if (uio) { 1764 uio->uio_resid = resid; 1765 error = uiomove(mtod(m, caddr_t) + moff, len, uio); 1766 resid = uio->uio_resid; 1767 if (error) 1768 goto release; 1769 } else { 1770 resid -= (size_t)len; 1771 } 1772 1773 /* 1774 * Eat the entire mbuf or just a piece of it 1775 */ 1776 offset += len; 1777 if (len == m->m_len - moff) { 1778 m = m->m_next; 1779 moff = 0; 1780 } else { 1781 moff += len; 1782 } 1783 1784 /* 1785 * Check oobmark 1786 */ 1787 if (so->so_oobmark && offset == so->so_oobmark) { 1788 didoob = 1; 1789 break; 1790 } 1791 } 1792 1793 /* 1794 * Synchronize sockbuf with data we read. 1795 * 1796 * NOTE: (m) is junk on entry (it could be left over from the 1797 * previous loop). 1798 */ 1799 if ((flags & MSG_PEEK) == 0) { 1800 lwkt_gettoken(&so->so_rcv.ssb_token); 1801 m = so->so_rcv.ssb_mb; 1802 while (m && offset >= m->m_len) { 1803 if (so->so_oobmark) { 1804 so->so_oobmark -= m->m_len; 1805 if (so->so_oobmark == 0) { 1806 sosetstate(so, SS_RCVATMARK); 1807 didoob = 1; 1808 } 1809 } 1810 offset -= m->m_len; 1811 if (sio) { 1812 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL); 1813 sbappend(sio, m); 1814 m = n; 1815 } else { 1816 m = sbunlinkmbuf(&so->so_rcv.sb, 1817 m, &free_chain); 1818 } 1819 } 1820 if (offset) { 1821 KKASSERT(m); 1822 if (sio) { 1823 n = m_copym(m, 0, offset, M_WAITOK); 1824 if (n) 1825 sbappend(sio, n); 1826 } 1827 m->m_data += offset; 1828 m->m_len -= offset; 1829 so->so_rcv.ssb_cc -= offset; 1830 if (so->so_oobmark) { 1831 so->so_oobmark -= offset; 1832 if (so->so_oobmark == 0) { 1833 sosetstate(so, SS_RCVATMARK); 1834 didoob = 1; 1835 } 1836 } 1837 offset = 0; 1838 } 1839 lwkt_reltoken(&so->so_rcv.ssb_token); 1840 } 1841 1842 /* 1843 * If the MSG_WAITALL flag is set (for non-atomic socket), 1844 * we must not quit until resid == 0 or an error termination. 1845 * 1846 * If a signal/timeout occurs, return with a short count but without 1847 * error. 1848 * 1849 * Keep signalsockbuf locked against other readers. 1850 * 1851 * XXX if MSG_PEEK we currently do quit. 1852 */ 1853 if ((flags & MSG_WAITALL) && !(flags & MSG_PEEK) && 1854 didoob == 0 && resid > 0 && 1855 !sosendallatonce(so)) { 1856 lwkt_gettoken(&so->so_rcv.ssb_token); 1857 error = 0; 1858 while ((m = so->so_rcv.ssb_mb) == NULL) { 1859 if (so->so_error || (so->so_state & SS_CANTRCVMORE)) { 1860 error = so->so_error; 1861 break; 1862 } 1863 /* 1864 * The window might have closed to zero, make 1865 * sure we send an ack now that we've drained 1866 * the buffer or we might end up blocking until 1867 * the idle takes over (5 seconds). 1868 */ 1869 if (so->so_pcb) 1870 so_pru_rcvd_async(so); 1871 if (so->so_rcv.ssb_mb == NULL) 1872 error = ssb_wait(&so->so_rcv); 1873 if (error) { 1874 lwkt_reltoken(&so->so_rcv.ssb_token); 1875 ssb_unlock(&so->so_rcv); 1876 error = 0; 1877 goto done; 1878 } 1879 } 1880 if (m && error == 0) 1881 goto dontblock; 1882 lwkt_reltoken(&so->so_rcv.ssb_token); 1883 } 1884 1885 /* 1886 * Token not held here. 1887 * 1888 * Cleanup. If an atomic read was requested drop any unread data XXX 1889 */ 1890 if ((flags & MSG_PEEK) == 0) { 1891 if (so->so_pcb) 1892 so_pru_rcvd_async(so); 1893 } 1894 1895 if (orig_resid == resid && orig_resid && 1896 (so->so_state & SS_CANTRCVMORE) == 0) { 1897 ssb_unlock(&so->so_rcv); 1898 goto restart; 1899 } 1900 1901 if (flagsp) 1902 *flagsp |= flags; 1903 release: 1904 ssb_unlock(&so->so_rcv); 1905 done: 1906 if (free_chain) 1907 m_freem(free_chain); 1908 return (error); 1909 } 1910 1911 /* 1912 * Shut a socket down. Note that we do not get a frontend lock as we 1913 * want to be able to shut the socket down even if another thread is 1914 * blocked in a read(), thus waking it up. 1915 */ 1916 int 1917 soshutdown(struct socket *so, int how) 1918 { 1919 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1920 return (EINVAL); 1921 1922 if (how != SHUT_WR) { 1923 /*ssb_lock(&so->so_rcv, M_WAITOK);*/ 1924 sorflush(so); 1925 /*ssb_unlock(&so->so_rcv);*/ 1926 } 1927 if (how != SHUT_RD) 1928 return (so_pru_shutdown(so)); 1929 return (0); 1930 } 1931 1932 void 1933 sorflush(struct socket *so) 1934 { 1935 struct signalsockbuf *ssb = &so->so_rcv; 1936 struct protosw *pr = so->so_proto; 1937 struct signalsockbuf asb; 1938 1939 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR); 1940 1941 lwkt_gettoken(&ssb->ssb_token); 1942 socantrcvmore(so); 1943 asb = *ssb; 1944 1945 /* 1946 * Can't just blow up the ssb structure here 1947 */ 1948 bzero(&ssb->sb, sizeof(ssb->sb)); 1949 ssb->ssb_timeo = 0; 1950 ssb->ssb_lowat = 0; 1951 ssb->ssb_hiwat = 0; 1952 ssb->ssb_mbmax = 0; 1953 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK); 1954 1955 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose) 1956 (*pr->pr_domain->dom_dispose)(asb.ssb_mb); 1957 ssb_release(&asb, so); 1958 1959 lwkt_reltoken(&ssb->ssb_token); 1960 } 1961 1962 #ifdef INET 1963 static int 1964 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt) 1965 { 1966 struct accept_filter_arg *afap = NULL; 1967 struct accept_filter *afp; 1968 struct so_accf *af = so->so_accf; 1969 int error = 0; 1970 1971 /* do not set/remove accept filters on non listen sockets */ 1972 if ((so->so_options & SO_ACCEPTCONN) == 0) { 1973 error = EINVAL; 1974 goto out; 1975 } 1976 1977 /* removing the filter */ 1978 if (sopt == NULL) { 1979 if (af != NULL) { 1980 if (af->so_accept_filter != NULL && 1981 af->so_accept_filter->accf_destroy != NULL) { 1982 af->so_accept_filter->accf_destroy(so); 1983 } 1984 if (af->so_accept_filter_str != NULL) { 1985 kfree(af->so_accept_filter_str, M_ACCF); 1986 } 1987 kfree(af, M_ACCF); 1988 so->so_accf = NULL; 1989 } 1990 so->so_options &= ~SO_ACCEPTFILTER; 1991 return (0); 1992 } 1993 /* adding a filter */ 1994 /* must remove previous filter first */ 1995 if (af != NULL) { 1996 error = EINVAL; 1997 goto out; 1998 } 1999 /* don't put large objects on the kernel stack */ 2000 afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK); 2001 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap); 2002 afap->af_name[sizeof(afap->af_name)-1] = '\0'; 2003 afap->af_arg[sizeof(afap->af_arg)-1] = '\0'; 2004 if (error) 2005 goto out; 2006 afp = accept_filt_get(afap->af_name); 2007 if (afp == NULL) { 2008 error = ENOENT; 2009 goto out; 2010 } 2011 af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO); 2012 if (afp->accf_create != NULL) { 2013 if (afap->af_name[0] != '\0') { 2014 int len = strlen(afap->af_name) + 1; 2015 2016 af->so_accept_filter_str = kmalloc(len, M_ACCF, 2017 M_WAITOK); 2018 strcpy(af->so_accept_filter_str, afap->af_name); 2019 } 2020 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg); 2021 if (af->so_accept_filter_arg == NULL) { 2022 kfree(af->so_accept_filter_str, M_ACCF); 2023 kfree(af, M_ACCF); 2024 so->so_accf = NULL; 2025 error = EINVAL; 2026 goto out; 2027 } 2028 } 2029 af->so_accept_filter = afp; 2030 so->so_accf = af; 2031 so->so_options |= SO_ACCEPTFILTER; 2032 out: 2033 if (afap != NULL) 2034 kfree(afap, M_TEMP); 2035 return (error); 2036 } 2037 #endif /* INET */ 2038 2039 /* 2040 * Perhaps this routine, and sooptcopyout(), below, ought to come in 2041 * an additional variant to handle the case where the option value needs 2042 * to be some kind of integer, but not a specific size. 2043 * In addition to their use here, these functions are also called by the 2044 * protocol-level pr_ctloutput() routines. 2045 */ 2046 int 2047 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2048 { 2049 return soopt_to_kbuf(sopt, buf, len, minlen); 2050 } 2051 2052 int 2053 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2054 { 2055 size_t valsize; 2056 2057 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 2058 KKASSERT(kva_p(buf)); 2059 2060 /* 2061 * If the user gives us more than we wanted, we ignore it, 2062 * but if we don't get the minimum length the caller 2063 * wants, we return EINVAL. On success, sopt->sopt_valsize 2064 * is set to however much we actually retrieved. 2065 */ 2066 if ((valsize = sopt->sopt_valsize) < minlen) 2067 return EINVAL; 2068 if (valsize > len) 2069 sopt->sopt_valsize = valsize = len; 2070 2071 bcopy(sopt->sopt_val, buf, valsize); 2072 return 0; 2073 } 2074 2075 2076 int 2077 sosetopt(struct socket *so, struct sockopt *sopt) 2078 { 2079 int error, optval; 2080 struct linger l; 2081 struct timeval tv; 2082 u_long val; 2083 struct signalsockbuf *sotmp; 2084 2085 error = 0; 2086 sopt->sopt_dir = SOPT_SET; 2087 if (sopt->sopt_level != SOL_SOCKET) { 2088 if (so->so_proto && so->so_proto->pr_ctloutput) { 2089 return (so_pr_ctloutput(so, sopt)); 2090 } 2091 error = ENOPROTOOPT; 2092 } else { 2093 switch (sopt->sopt_name) { 2094 #ifdef INET 2095 case SO_ACCEPTFILTER: 2096 error = do_setopt_accept_filter(so, sopt); 2097 if (error) 2098 goto bad; 2099 break; 2100 #endif /* INET */ 2101 case SO_LINGER: 2102 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 2103 if (error) 2104 goto bad; 2105 2106 so->so_linger = l.l_linger; 2107 if (l.l_onoff) 2108 so->so_options |= SO_LINGER; 2109 else 2110 so->so_options &= ~SO_LINGER; 2111 break; 2112 2113 case SO_DEBUG: 2114 case SO_KEEPALIVE: 2115 case SO_DONTROUTE: 2116 case SO_USELOOPBACK: 2117 case SO_BROADCAST: 2118 case SO_REUSEADDR: 2119 case SO_REUSEPORT: 2120 case SO_OOBINLINE: 2121 case SO_TIMESTAMP: 2122 case SO_NOSIGPIPE: 2123 error = sooptcopyin(sopt, &optval, sizeof optval, 2124 sizeof optval); 2125 if (error) 2126 goto bad; 2127 if (optval) 2128 so->so_options |= sopt->sopt_name; 2129 else 2130 so->so_options &= ~sopt->sopt_name; 2131 break; 2132 2133 case SO_SNDBUF: 2134 case SO_RCVBUF: 2135 case SO_SNDLOWAT: 2136 case SO_RCVLOWAT: 2137 error = sooptcopyin(sopt, &optval, sizeof optval, 2138 sizeof optval); 2139 if (error) 2140 goto bad; 2141 2142 /* 2143 * Values < 1 make no sense for any of these 2144 * options, so disallow them. 2145 */ 2146 if (optval < 1) { 2147 error = EINVAL; 2148 goto bad; 2149 } 2150 2151 switch (sopt->sopt_name) { 2152 case SO_SNDBUF: 2153 case SO_RCVBUF: 2154 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ? 2155 &so->so_snd : &so->so_rcv, (u_long)optval, 2156 so, 2157 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) { 2158 error = ENOBUFS; 2159 goto bad; 2160 } 2161 sotmp = (sopt->sopt_name == SO_SNDBUF) ? 2162 &so->so_snd : &so->so_rcv; 2163 atomic_clear_int(&sotmp->ssb_flags, 2164 SSB_AUTOSIZE); 2165 break; 2166 2167 /* 2168 * Make sure the low-water is never greater than 2169 * the high-water. 2170 */ 2171 case SO_SNDLOWAT: 2172 so->so_snd.ssb_lowat = 2173 (optval > so->so_snd.ssb_hiwat) ? 2174 so->so_snd.ssb_hiwat : optval; 2175 atomic_clear_int(&so->so_snd.ssb_flags, 2176 SSB_AUTOLOWAT); 2177 break; 2178 case SO_RCVLOWAT: 2179 so->so_rcv.ssb_lowat = 2180 (optval > so->so_rcv.ssb_hiwat) ? 2181 so->so_rcv.ssb_hiwat : optval; 2182 atomic_clear_int(&so->so_rcv.ssb_flags, 2183 SSB_AUTOLOWAT); 2184 break; 2185 } 2186 break; 2187 2188 case SO_SNDTIMEO: 2189 case SO_RCVTIMEO: 2190 error = sooptcopyin(sopt, &tv, sizeof tv, 2191 sizeof tv); 2192 if (error) 2193 goto bad; 2194 2195 /* assert(hz > 0); */ 2196 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz || 2197 tv.tv_usec < 0 || tv.tv_usec >= 1000000) { 2198 error = EDOM; 2199 goto bad; 2200 } 2201 /* assert(tick > 0); */ 2202 /* assert(ULONG_MAX - INT_MAX >= 1000000); */ 2203 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick; 2204 if (val > INT_MAX) { 2205 error = EDOM; 2206 goto bad; 2207 } 2208 if (val == 0 && tv.tv_usec != 0) 2209 val = 1; 2210 2211 switch (sopt->sopt_name) { 2212 case SO_SNDTIMEO: 2213 so->so_snd.ssb_timeo = val; 2214 break; 2215 case SO_RCVTIMEO: 2216 so->so_rcv.ssb_timeo = val; 2217 break; 2218 } 2219 break; 2220 default: 2221 error = ENOPROTOOPT; 2222 break; 2223 } 2224 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 2225 (void) so_pr_ctloutput(so, sopt); 2226 } 2227 } 2228 bad: 2229 return (error); 2230 } 2231 2232 /* Helper routine for getsockopt */ 2233 int 2234 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 2235 { 2236 soopt_from_kbuf(sopt, buf, len); 2237 return 0; 2238 } 2239 2240 void 2241 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len) 2242 { 2243 size_t valsize; 2244 2245 if (len == 0) { 2246 sopt->sopt_valsize = 0; 2247 return; 2248 } 2249 2250 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 2251 KKASSERT(kva_p(buf)); 2252 2253 /* 2254 * Documented get behavior is that we always return a value, 2255 * possibly truncated to fit in the user's buffer. 2256 * Traditional behavior is that we always tell the user 2257 * precisely how much we copied, rather than something useful 2258 * like the total amount we had available for her. 2259 * Note that this interface is not idempotent; the entire answer must 2260 * generated ahead of time. 2261 */ 2262 valsize = szmin(len, sopt->sopt_valsize); 2263 sopt->sopt_valsize = valsize; 2264 if (sopt->sopt_val != 0) { 2265 bcopy(buf, sopt->sopt_val, valsize); 2266 } 2267 } 2268 2269 int 2270 sogetopt(struct socket *so, struct sockopt *sopt) 2271 { 2272 int error, optval; 2273 long optval_l; 2274 struct linger l; 2275 struct timeval tv; 2276 #ifdef INET 2277 struct accept_filter_arg *afap; 2278 #endif 2279 2280 error = 0; 2281 sopt->sopt_dir = SOPT_GET; 2282 if (sopt->sopt_level != SOL_SOCKET) { 2283 if (so->so_proto && so->so_proto->pr_ctloutput) { 2284 return (so_pr_ctloutput(so, sopt)); 2285 } else 2286 return (ENOPROTOOPT); 2287 } else { 2288 switch (sopt->sopt_name) { 2289 #ifdef INET 2290 case SO_ACCEPTFILTER: 2291 if ((so->so_options & SO_ACCEPTCONN) == 0) 2292 return (EINVAL); 2293 afap = kmalloc(sizeof(*afap), M_TEMP, 2294 M_WAITOK | M_ZERO); 2295 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 2296 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name); 2297 if (so->so_accf->so_accept_filter_str != NULL) 2298 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str); 2299 } 2300 error = sooptcopyout(sopt, afap, sizeof(*afap)); 2301 kfree(afap, M_TEMP); 2302 break; 2303 #endif /* INET */ 2304 2305 case SO_LINGER: 2306 l.l_onoff = so->so_options & SO_LINGER; 2307 l.l_linger = so->so_linger; 2308 error = sooptcopyout(sopt, &l, sizeof l); 2309 break; 2310 2311 case SO_USELOOPBACK: 2312 case SO_DONTROUTE: 2313 case SO_DEBUG: 2314 case SO_KEEPALIVE: 2315 case SO_REUSEADDR: 2316 case SO_REUSEPORT: 2317 case SO_BROADCAST: 2318 case SO_OOBINLINE: 2319 case SO_TIMESTAMP: 2320 case SO_NOSIGPIPE: 2321 optval = so->so_options & sopt->sopt_name; 2322 integer: 2323 error = sooptcopyout(sopt, &optval, sizeof optval); 2324 break; 2325 2326 case SO_TYPE: 2327 optval = so->so_type; 2328 goto integer; 2329 2330 case SO_ERROR: 2331 optval = so->so_error; 2332 so->so_error = 0; 2333 goto integer; 2334 2335 case SO_SNDBUF: 2336 optval = so->so_snd.ssb_hiwat; 2337 goto integer; 2338 2339 case SO_RCVBUF: 2340 optval = so->so_rcv.ssb_hiwat; 2341 goto integer; 2342 2343 case SO_SNDLOWAT: 2344 optval = so->so_snd.ssb_lowat; 2345 goto integer; 2346 2347 case SO_RCVLOWAT: 2348 optval = so->so_rcv.ssb_lowat; 2349 goto integer; 2350 2351 case SO_SNDTIMEO: 2352 case SO_RCVTIMEO: 2353 optval = (sopt->sopt_name == SO_SNDTIMEO ? 2354 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo); 2355 2356 tv.tv_sec = optval / hz; 2357 tv.tv_usec = (optval % hz) * ustick; 2358 error = sooptcopyout(sopt, &tv, sizeof tv); 2359 break; 2360 2361 case SO_SNDSPACE: 2362 optval_l = ssb_space(&so->so_snd); 2363 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l)); 2364 break; 2365 2366 case SO_CPUHINT: 2367 optval = -1; /* no hint */ 2368 goto integer; 2369 2370 default: 2371 error = ENOPROTOOPT; 2372 break; 2373 } 2374 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) 2375 so_pr_ctloutput(so, sopt); 2376 return (error); 2377 } 2378 } 2379 2380 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */ 2381 int 2382 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 2383 { 2384 struct mbuf *m, *m_prev; 2385 int sopt_size = sopt->sopt_valsize, msize; 2386 2387 m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA, 2388 0, &msize); 2389 if (m == NULL) 2390 return (ENOBUFS); 2391 m->m_len = min(msize, sopt_size); 2392 sopt_size -= m->m_len; 2393 *mp = m; 2394 m_prev = m; 2395 2396 while (sopt_size > 0) { 2397 m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT, 2398 MT_DATA, 0, &msize); 2399 if (m == NULL) { 2400 m_freem(*mp); 2401 return (ENOBUFS); 2402 } 2403 m->m_len = min(msize, sopt_size); 2404 sopt_size -= m->m_len; 2405 m_prev->m_next = m; 2406 m_prev = m; 2407 } 2408 return (0); 2409 } 2410 2411 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */ 2412 int 2413 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 2414 { 2415 soopt_to_mbuf(sopt, m); 2416 return 0; 2417 } 2418 2419 void 2420 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m) 2421 { 2422 size_t valsize; 2423 void *val; 2424 2425 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 2426 KKASSERT(kva_p(m)); 2427 if (sopt->sopt_val == NULL) 2428 return; 2429 val = sopt->sopt_val; 2430 valsize = sopt->sopt_valsize; 2431 while (m != NULL && valsize >= m->m_len) { 2432 bcopy(val, mtod(m, char *), m->m_len); 2433 valsize -= m->m_len; 2434 val = (caddr_t)val + m->m_len; 2435 m = m->m_next; 2436 } 2437 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 2438 panic("ip6_sooptmcopyin"); 2439 } 2440 2441 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */ 2442 int 2443 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 2444 { 2445 return soopt_from_mbuf(sopt, m); 2446 } 2447 2448 int 2449 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m) 2450 { 2451 struct mbuf *m0 = m; 2452 size_t valsize = 0; 2453 size_t maxsize; 2454 void *val; 2455 2456 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val)); 2457 KKASSERT(kva_p(m)); 2458 if (sopt->sopt_val == NULL) 2459 return 0; 2460 val = sopt->sopt_val; 2461 maxsize = sopt->sopt_valsize; 2462 while (m != NULL && maxsize >= m->m_len) { 2463 bcopy(mtod(m, char *), val, m->m_len); 2464 maxsize -= m->m_len; 2465 val = (caddr_t)val + m->m_len; 2466 valsize += m->m_len; 2467 m = m->m_next; 2468 } 2469 if (m != NULL) { 2470 /* enough soopt buffer should be given from user-land */ 2471 m_freem(m0); 2472 return (EINVAL); 2473 } 2474 sopt->sopt_valsize = valsize; 2475 return 0; 2476 } 2477 2478 void 2479 sohasoutofband(struct socket *so) 2480 { 2481 if (so->so_sigio != NULL) 2482 pgsigio(so->so_sigio, SIGURG, 0); 2483 KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB); 2484 } 2485 2486 int 2487 sokqfilter(struct file *fp, struct knote *kn) 2488 { 2489 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2490 struct signalsockbuf *ssb; 2491 2492 switch (kn->kn_filter) { 2493 case EVFILT_READ: 2494 if (so->so_options & SO_ACCEPTCONN) 2495 kn->kn_fop = &solisten_filtops; 2496 else 2497 kn->kn_fop = &soread_filtops; 2498 ssb = &so->so_rcv; 2499 break; 2500 case EVFILT_WRITE: 2501 kn->kn_fop = &sowrite_filtops; 2502 ssb = &so->so_snd; 2503 break; 2504 case EVFILT_EXCEPT: 2505 kn->kn_fop = &soexcept_filtops; 2506 ssb = &so->so_rcv; 2507 break; 2508 default: 2509 return (EOPNOTSUPP); 2510 } 2511 2512 knote_insert(&ssb->ssb_kq.ki_note, kn); 2513 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE); 2514 return (0); 2515 } 2516 2517 static void 2518 filt_sordetach(struct knote *kn) 2519 { 2520 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2521 2522 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn); 2523 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note)) 2524 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE); 2525 } 2526 2527 /*ARGSUSED*/ 2528 static int 2529 filt_soread(struct knote *kn, long hint) 2530 { 2531 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2532 2533 if (kn->kn_sfflags & NOTE_OOB) { 2534 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) { 2535 kn->kn_fflags |= NOTE_OOB; 2536 return (1); 2537 } 2538 return (0); 2539 } 2540 kn->kn_data = so->so_rcv.ssb_cc; 2541 2542 if (so->so_state & SS_CANTRCVMORE) { 2543 /* 2544 * Only set NODATA if all data has been exhausted. 2545 */ 2546 if (kn->kn_data == 0) 2547 kn->kn_flags |= EV_NODATA; 2548 kn->kn_flags |= EV_EOF; 2549 kn->kn_fflags = so->so_error; 2550 return (1); 2551 } 2552 if (so->so_error) /* temporary udp error */ 2553 return (1); 2554 if (kn->kn_sfflags & NOTE_LOWAT) 2555 return (kn->kn_data >= kn->kn_sdata); 2556 return ((kn->kn_data >= so->so_rcv.ssb_lowat) || 2557 !TAILQ_EMPTY(&so->so_comp)); 2558 } 2559 2560 static void 2561 filt_sowdetach(struct knote *kn) 2562 { 2563 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2564 2565 knote_remove(&so->so_snd.ssb_kq.ki_note, kn); 2566 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note)) 2567 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE); 2568 } 2569 2570 /*ARGSUSED*/ 2571 static int 2572 filt_sowrite(struct knote *kn, long hint) 2573 { 2574 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2575 2576 kn->kn_data = ssb_space(&so->so_snd); 2577 if (so->so_state & SS_CANTSENDMORE) { 2578 kn->kn_flags |= (EV_EOF | EV_NODATA); 2579 kn->kn_fflags = so->so_error; 2580 return (1); 2581 } 2582 if (so->so_error) /* temporary udp error */ 2583 return (1); 2584 if (((so->so_state & SS_ISCONNECTED) == 0) && 2585 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 2586 return (0); 2587 if (kn->kn_sfflags & NOTE_LOWAT) 2588 return (kn->kn_data >= kn->kn_sdata); 2589 return (kn->kn_data >= so->so_snd.ssb_lowat); 2590 } 2591 2592 /*ARGSUSED*/ 2593 static int 2594 filt_solisten(struct knote *kn, long hint) 2595 { 2596 struct socket *so = (struct socket *)kn->kn_fp->f_data; 2597 2598 kn->kn_data = so->so_qlen; 2599 return (! TAILQ_EMPTY(&so->so_comp)); 2600 } 2601