uipc_socket.c revision 1.135
1 /* $NetBSD: uipc_socket.c,v 1.135 2007/03/12 18:18:34 ad Exp $ */ 2 3 /*- 4 * Copyright (c) 2002 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of Wasabi Systems, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 39 /* 40 * Copyright (c) 1982, 1986, 1988, 1990, 1993 41 * The Regents of the University of California. All rights reserved. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. Neither the name of the University nor the names of its contributors 52 * may be used to endorse or promote products derived from this software 53 * without specific prior written permission. 54 * 55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 65 * SUCH DAMAGE. 66 * 67 * @(#)uipc_socket.c 8.6 (Berkeley) 5/2/95 68 */ 69 70 #include <sys/cdefs.h> 71 __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.135 2007/03/12 18:18:34 ad Exp $"); 72 73 #include "opt_sock_counters.h" 74 #include "opt_sosend_loan.h" 75 #include "opt_mbuftrace.h" 76 #include "opt_somaxkva.h" 77 78 #include <sys/param.h> 79 #include <sys/systm.h> 80 #include <sys/proc.h> 81 #include <sys/file.h> 82 #include <sys/malloc.h> 83 #include <sys/mbuf.h> 84 #include <sys/domain.h> 85 #include <sys/kernel.h> 86 #include <sys/protosw.h> 87 #include <sys/socket.h> 88 #include <sys/socketvar.h> 89 #include <sys/signalvar.h> 90 #include <sys/resourcevar.h> 91 #include <sys/pool.h> 92 #include <sys/event.h> 93 #include <sys/poll.h> 94 #include <sys/kauth.h> 95 96 #include <uvm/uvm.h> 97 98 POOL_INIT(socket_pool, sizeof(struct socket), 0, 0, 0, "sockpl", NULL, 99 IPL_SOFTNET); 100 101 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options"); 102 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 103 104 extern int somaxconn; /* patchable (XXX sysctl) */ 105 int somaxconn = SOMAXCONN; 106 107 #ifdef SOSEND_COUNTERS 108 #include <sys/device.h> 109 110 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 111 NULL, "sosend", "loan big"); 112 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 113 NULL, "sosend", "copy big"); 114 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 115 NULL, "sosend", "copy small"); 116 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 117 NULL, "sosend", "kva limit"); 118 119 #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++ 120 121 EVCNT_ATTACH_STATIC(sosend_loan_big); 122 EVCNT_ATTACH_STATIC(sosend_copy_big); 123 EVCNT_ATTACH_STATIC(sosend_copy_small); 124 EVCNT_ATTACH_STATIC(sosend_kvalimit); 125 #else 126 127 #define SOSEND_COUNTER_INCR(ev) /* nothing */ 128 129 #endif /* SOSEND_COUNTERS */ 130 131 static struct callback_entry sokva_reclaimerentry; 132 133 #ifdef SOSEND_NO_LOAN 134 int sock_loan_thresh = -1; 135 #else 136 int sock_loan_thresh = 4096; 137 #endif 138 139 static struct simplelock so_pendfree_slock = SIMPLELOCK_INITIALIZER; 140 static struct mbuf *so_pendfree; 141 142 #ifndef SOMAXKVA 143 #define SOMAXKVA (16 * 1024 * 1024) 144 #endif 145 int somaxkva = SOMAXKVA; 146 static int socurkva; 147 static int sokvawaiters; 148 149 #define SOCK_LOAN_CHUNK 65536 150 151 static size_t sodopendfree(void); 152 static size_t sodopendfreel(void); 153 154 static vsize_t 155 sokvareserve(struct socket *so, vsize_t len) 156 { 157 int s; 158 int error; 159 160 s = splvm(); 161 simple_lock(&so_pendfree_slock); 162 while (socurkva + len > somaxkva) { 163 size_t freed; 164 165 /* 166 * try to do pendfree. 167 */ 168 169 freed = sodopendfreel(); 170 171 /* 172 * if some kva was freed, try again. 173 */ 174 175 if (freed) 176 continue; 177 178 SOSEND_COUNTER_INCR(&sosend_kvalimit); 179 sokvawaiters++; 180 error = ltsleep(&socurkva, PVM | PCATCH, "sokva", 0, 181 &so_pendfree_slock); 182 sokvawaiters--; 183 if (error) { 184 len = 0; 185 break; 186 } 187 } 188 socurkva += len; 189 simple_unlock(&so_pendfree_slock); 190 splx(s); 191 return len; 192 } 193 194 static void 195 sokvaunreserve(vsize_t len) 196 { 197 int s; 198 199 s = splvm(); 200 simple_lock(&so_pendfree_slock); 201 socurkva -= len; 202 if (sokvawaiters) 203 wakeup(&socurkva); 204 simple_unlock(&so_pendfree_slock); 205 splx(s); 206 } 207 208 /* 209 * sokvaalloc: allocate kva for loan. 210 */ 211 212 vaddr_t 213 sokvaalloc(vsize_t len, struct socket *so) 214 { 215 vaddr_t lva; 216 217 /* 218 * reserve kva. 219 */ 220 221 if (sokvareserve(so, len) == 0) 222 return 0; 223 224 /* 225 * allocate kva. 226 */ 227 228 lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA); 229 if (lva == 0) { 230 sokvaunreserve(len); 231 return (0); 232 } 233 234 return lva; 235 } 236 237 /* 238 * sokvafree: free kva for loan. 239 */ 240 241 void 242 sokvafree(vaddr_t sva, vsize_t len) 243 { 244 245 /* 246 * free kva. 247 */ 248 249 uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY); 250 251 /* 252 * unreserve kva. 253 */ 254 255 sokvaunreserve(len); 256 } 257 258 static void 259 sodoloanfree(struct vm_page **pgs, void *buf, size_t size) 260 { 261 vaddr_t va, sva, eva; 262 vsize_t len; 263 paddr_t pa; 264 int i, npgs; 265 266 eva = round_page((vaddr_t) buf + size); 267 sva = trunc_page((vaddr_t) buf); 268 len = eva - sva; 269 npgs = len >> PAGE_SHIFT; 270 271 if (__predict_false(pgs == NULL)) { 272 pgs = alloca(npgs * sizeof(*pgs)); 273 274 for (i = 0, va = sva; va < eva; i++, va += PAGE_SIZE) { 275 if (pmap_extract(pmap_kernel(), va, &pa) == false) 276 panic("sodoloanfree: va 0x%lx not mapped", va); 277 pgs[i] = PHYS_TO_VM_PAGE(pa); 278 } 279 } 280 281 pmap_kremove(sva, len); 282 pmap_update(pmap_kernel()); 283 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 284 sokvafree(sva, len); 285 } 286 287 static size_t 288 sodopendfree() 289 { 290 int s; 291 size_t rv; 292 293 s = splvm(); 294 simple_lock(&so_pendfree_slock); 295 rv = sodopendfreel(); 296 simple_unlock(&so_pendfree_slock); 297 splx(s); 298 299 return rv; 300 } 301 302 /* 303 * sodopendfreel: free mbufs on "pendfree" list. 304 * unlock and relock so_pendfree_slock when freeing mbufs. 305 * 306 * => called with so_pendfree_slock held. 307 * => called at splvm. 308 */ 309 310 static size_t 311 sodopendfreel() 312 { 313 size_t rv = 0; 314 315 LOCK_ASSERT(simple_lock_held(&so_pendfree_slock)); 316 317 for (;;) { 318 struct mbuf *m; 319 struct mbuf *next; 320 321 m = so_pendfree; 322 if (m == NULL) 323 break; 324 so_pendfree = NULL; 325 simple_unlock(&so_pendfree_slock); 326 /* XXX splx */ 327 328 for (; m != NULL; m = next) { 329 next = m->m_next; 330 331 rv += m->m_ext.ext_size; 332 sodoloanfree((m->m_flags & M_EXT_PAGES) ? 333 m->m_ext.ext_pgs : NULL, m->m_ext.ext_buf, 334 m->m_ext.ext_size); 335 pool_cache_put(&mbpool_cache, m); 336 } 337 338 /* XXX splvm */ 339 simple_lock(&so_pendfree_slock); 340 } 341 342 return (rv); 343 } 344 345 void 346 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg) 347 { 348 int s; 349 350 if (m == NULL) { 351 352 /* 353 * called from MEXTREMOVE. 354 */ 355 356 sodoloanfree(NULL, buf, size); 357 return; 358 } 359 360 /* 361 * postpone freeing mbuf. 362 * 363 * we can't do it in interrupt context 364 * because we need to put kva back to kernel_map. 365 */ 366 367 s = splvm(); 368 simple_lock(&so_pendfree_slock); 369 m->m_next = so_pendfree; 370 so_pendfree = m; 371 if (sokvawaiters) 372 wakeup(&socurkva); 373 simple_unlock(&so_pendfree_slock); 374 splx(s); 375 } 376 377 static long 378 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 379 { 380 struct iovec *iov = uio->uio_iov; 381 vaddr_t sva, eva; 382 vsize_t len; 383 vaddr_t lva, va; 384 int npgs, i, error; 385 386 if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace)) 387 return (0); 388 389 if (iov->iov_len < (size_t) space) 390 space = iov->iov_len; 391 if (space > SOCK_LOAN_CHUNK) 392 space = SOCK_LOAN_CHUNK; 393 394 eva = round_page((vaddr_t) iov->iov_base + space); 395 sva = trunc_page((vaddr_t) iov->iov_base); 396 len = eva - sva; 397 npgs = len >> PAGE_SHIFT; 398 399 /* XXX KDASSERT */ 400 KASSERT(npgs <= M_EXT_MAXPAGES); 401 402 lva = sokvaalloc(len, so); 403 if (lva == 0) 404 return 0; 405 406 error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len, 407 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 408 if (error) { 409 sokvafree(lva, len); 410 return (0); 411 } 412 413 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 414 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 415 VM_PROT_READ); 416 pmap_update(pmap_kernel()); 417 418 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 419 420 MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so); 421 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 422 423 uio->uio_resid -= space; 424 /* uio_offset not updated, not set/used for write(2) */ 425 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space; 426 uio->uio_iov->iov_len -= space; 427 if (uio->uio_iov->iov_len == 0) { 428 uio->uio_iov++; 429 uio->uio_iovcnt--; 430 } 431 432 return (space); 433 } 434 435 static int 436 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg) 437 { 438 439 KASSERT(ce == &sokva_reclaimerentry); 440 KASSERT(obj == NULL); 441 442 sodopendfree(); 443 if (!vm_map_starved_p(kernel_map)) { 444 return CALLBACK_CHAIN_ABORT; 445 } 446 return CALLBACK_CHAIN_CONTINUE; 447 } 448 449 void 450 soinit(void) 451 { 452 453 /* Set the initial adjusted socket buffer size. */ 454 if (sb_max_set(sb_max)) 455 panic("bad initial sb_max value: %lu", sb_max); 456 457 callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback, 458 &sokva_reclaimerentry, NULL, sokva_reclaim_callback); 459 } 460 461 /* 462 * Socket operation routines. 463 * These routines are called by the routines in 464 * sys_socket.c or from a system process, and 465 * implement the semantics of socket operations by 466 * switching out to the protocol specific routines. 467 */ 468 /*ARGSUSED*/ 469 int 470 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l) 471 { 472 const struct protosw *prp; 473 struct socket *so; 474 uid_t uid; 475 int error, s; 476 477 error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET, 478 KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type), 479 KAUTH_ARG(proto)); 480 if (error) 481 return (error); 482 483 if (proto) 484 prp = pffindproto(dom, proto, type); 485 else 486 prp = pffindtype(dom, type); 487 if (prp == 0) { 488 /* no support for domain */ 489 if (pffinddomain(dom) == 0) 490 return (EAFNOSUPPORT); 491 /* no support for socket type */ 492 if (proto == 0 && type != 0) 493 return (EPROTOTYPE); 494 return (EPROTONOSUPPORT); 495 } 496 if (prp->pr_usrreq == 0) 497 return (EPROTONOSUPPORT); 498 if (prp->pr_type != type) 499 return (EPROTOTYPE); 500 s = splsoftnet(); 501 so = pool_get(&socket_pool, PR_WAITOK); 502 memset((void *)so, 0, sizeof(*so)); 503 TAILQ_INIT(&so->so_q0); 504 TAILQ_INIT(&so->so_q); 505 so->so_type = type; 506 so->so_proto = prp; 507 so->so_send = sosend; 508 so->so_receive = soreceive; 509 #ifdef MBUFTRACE 510 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 511 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 512 so->so_mowner = &prp->pr_domain->dom_mowner; 513 #endif 514 if (l != NULL) { 515 uid = kauth_cred_geteuid(l->l_cred); 516 } else { 517 uid = 0; 518 } 519 so->so_uidinfo = uid_find(uid); 520 error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0, 521 (struct mbuf *)(long)proto, (struct mbuf *)0, l); 522 if (error) { 523 so->so_state |= SS_NOFDREF; 524 sofree(so); 525 splx(s); 526 return (error); 527 } 528 splx(s); 529 *aso = so; 530 return (0); 531 } 532 533 int 534 sobind(struct socket *so, struct mbuf *nam, struct lwp *l) 535 { 536 int s, error; 537 538 s = splsoftnet(); 539 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0, 540 nam, (struct mbuf *)0, l); 541 splx(s); 542 return (error); 543 } 544 545 int 546 solisten(struct socket *so, int backlog) 547 { 548 int s, error; 549 550 s = splsoftnet(); 551 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0, 552 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 553 if (error) { 554 splx(s); 555 return (error); 556 } 557 if (TAILQ_EMPTY(&so->so_q)) 558 so->so_options |= SO_ACCEPTCONN; 559 if (backlog < 0) 560 backlog = 0; 561 so->so_qlimit = min(backlog, somaxconn); 562 splx(s); 563 return (0); 564 } 565 566 void 567 sofree(struct socket *so) 568 { 569 570 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 571 return; 572 if (so->so_head) { 573 /* 574 * We must not decommission a socket that's on the accept(2) 575 * queue. If we do, then accept(2) may hang after select(2) 576 * indicated that the listening socket was ready. 577 */ 578 if (!soqremque(so, 0)) 579 return; 580 } 581 if (so->so_rcv.sb_hiwat) 582 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 583 RLIM_INFINITY); 584 if (so->so_snd.sb_hiwat) 585 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 586 RLIM_INFINITY); 587 sbrelease(&so->so_snd, so); 588 sorflush(so); 589 pool_put(&socket_pool, so); 590 } 591 592 /* 593 * Close a socket on last file table reference removal. 594 * Initiate disconnect if connected. 595 * Free socket when disconnect complete. 596 */ 597 int 598 soclose(struct socket *so) 599 { 600 struct socket *so2; 601 int s, error; 602 603 error = 0; 604 s = splsoftnet(); /* conservative */ 605 if (so->so_options & SO_ACCEPTCONN) { 606 while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 607 (void) soqremque(so2, 0); 608 (void) soabort(so2); 609 } 610 while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 611 (void) soqremque(so2, 1); 612 (void) soabort(so2); 613 } 614 } 615 if (so->so_pcb == 0) 616 goto discard; 617 if (so->so_state & SS_ISCONNECTED) { 618 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 619 error = sodisconnect(so); 620 if (error) 621 goto drop; 622 } 623 if (so->so_options & SO_LINGER) { 624 if ((so->so_state & SS_ISDISCONNECTING) && 625 (so->so_state & SS_NBIO)) 626 goto drop; 627 while (so->so_state & SS_ISCONNECTED) { 628 error = tsleep((void *)&so->so_timeo, 629 PSOCK | PCATCH, netcls, 630 so->so_linger * hz); 631 if (error) 632 break; 633 } 634 } 635 } 636 drop: 637 if (so->so_pcb) { 638 int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 639 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 640 (struct lwp *)0); 641 if (error == 0) 642 error = error2; 643 } 644 discard: 645 if (so->so_state & SS_NOFDREF) 646 panic("soclose: NOFDREF"); 647 so->so_state |= SS_NOFDREF; 648 sofree(so); 649 splx(s); 650 return (error); 651 } 652 653 /* 654 * Must be called at splsoftnet... 655 */ 656 int 657 soabort(struct socket *so) 658 { 659 660 return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0, 661 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 662 } 663 664 int 665 soaccept(struct socket *so, struct mbuf *nam) 666 { 667 int s, error; 668 669 error = 0; 670 s = splsoftnet(); 671 if ((so->so_state & SS_NOFDREF) == 0) 672 panic("soaccept: !NOFDREF"); 673 so->so_state &= ~SS_NOFDREF; 674 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 675 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 676 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 677 (struct mbuf *)0, nam, (struct mbuf *)0, (struct lwp *)0); 678 else 679 error = ECONNABORTED; 680 681 splx(s); 682 return (error); 683 } 684 685 int 686 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l) 687 { 688 int s, error; 689 690 if (so->so_options & SO_ACCEPTCONN) 691 return (EOPNOTSUPP); 692 s = splsoftnet(); 693 /* 694 * If protocol is connection-based, can only connect once. 695 * Otherwise, if connected, try to disconnect first. 696 * This allows user to disconnect by connecting to, e.g., 697 * a null address. 698 */ 699 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 700 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 701 (error = sodisconnect(so)))) 702 error = EISCONN; 703 else 704 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 705 (struct mbuf *)0, nam, (struct mbuf *)0, l); 706 splx(s); 707 return (error); 708 } 709 710 int 711 soconnect2(struct socket *so1, struct socket *so2) 712 { 713 int s, error; 714 715 s = splsoftnet(); 716 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 717 (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0, 718 (struct lwp *)0); 719 splx(s); 720 return (error); 721 } 722 723 int 724 sodisconnect(struct socket *so) 725 { 726 int s, error; 727 728 s = splsoftnet(); 729 if ((so->so_state & SS_ISCONNECTED) == 0) { 730 error = ENOTCONN; 731 goto bad; 732 } 733 if (so->so_state & SS_ISDISCONNECTING) { 734 error = EALREADY; 735 goto bad; 736 } 737 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 738 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 739 (struct lwp *)0); 740 bad: 741 splx(s); 742 sodopendfree(); 743 return (error); 744 } 745 746 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 747 /* 748 * Send on a socket. 749 * If send must go all at once and message is larger than 750 * send buffering, then hard error. 751 * Lock against other senders. 752 * If must go all at once and not enough room now, then 753 * inform user that this would block and do nothing. 754 * Otherwise, if nonblocking, send as much as possible. 755 * The data to be sent is described by "uio" if nonzero, 756 * otherwise by the mbuf chain "top" (which must be null 757 * if uio is not). Data provided in mbuf chain must be small 758 * enough to send all at once. 759 * 760 * Returns nonzero on error, timeout or signal; callers 761 * must check for short counts if EINTR/ERESTART are returned. 762 * Data and control buffers are freed on return. 763 */ 764 int 765 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 766 struct mbuf *control, int flags, struct lwp *l) 767 { 768 struct mbuf **mp, *m; 769 struct proc *p; 770 long space, len, resid, clen, mlen; 771 int error, s, dontroute, atomic; 772 773 p = l->l_proc; 774 sodopendfree(); 775 776 clen = 0; 777 atomic = sosendallatonce(so) || top; 778 if (uio) 779 resid = uio->uio_resid; 780 else 781 resid = top->m_pkthdr.len; 782 /* 783 * In theory resid should be unsigned. 784 * However, space must be signed, as it might be less than 0 785 * if we over-committed, and we must use a signed comparison 786 * of space and resid. On the other hand, a negative resid 787 * causes us to loop sending 0-length segments to the protocol. 788 */ 789 if (resid < 0) { 790 error = EINVAL; 791 goto out; 792 } 793 dontroute = 794 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 795 (so->so_proto->pr_flags & PR_ATOMIC); 796 if (p) 797 p->p_stats->p_ru.ru_msgsnd++; 798 if (control) 799 clen = control->m_len; 800 #define snderr(errno) { error = errno; splx(s); goto release; } 801 802 restart: 803 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 804 goto out; 805 do { 806 s = splsoftnet(); 807 if (so->so_state & SS_CANTSENDMORE) 808 snderr(EPIPE); 809 if (so->so_error) { 810 error = so->so_error; 811 so->so_error = 0; 812 splx(s); 813 goto release; 814 } 815 if ((so->so_state & SS_ISCONNECTED) == 0) { 816 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 817 if ((so->so_state & SS_ISCONFIRMING) == 0 && 818 !(resid == 0 && clen != 0)) 819 snderr(ENOTCONN); 820 } else if (addr == 0) 821 snderr(EDESTADDRREQ); 822 } 823 space = sbspace(&so->so_snd); 824 if (flags & MSG_OOB) 825 space += 1024; 826 if ((atomic && resid > so->so_snd.sb_hiwat) || 827 clen > so->so_snd.sb_hiwat) 828 snderr(EMSGSIZE); 829 if (space < resid + clen && 830 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 831 if (so->so_state & SS_NBIO) 832 snderr(EWOULDBLOCK); 833 sbunlock(&so->so_snd); 834 error = sbwait(&so->so_snd); 835 splx(s); 836 if (error) 837 goto out; 838 goto restart; 839 } 840 splx(s); 841 mp = ⊤ 842 space -= clen; 843 do { 844 if (uio == NULL) { 845 /* 846 * Data is prepackaged in "top". 847 */ 848 resid = 0; 849 if (flags & MSG_EOR) 850 top->m_flags |= M_EOR; 851 } else do { 852 if (top == 0) { 853 m = m_gethdr(M_WAIT, MT_DATA); 854 mlen = MHLEN; 855 m->m_pkthdr.len = 0; 856 m->m_pkthdr.rcvif = (struct ifnet *)0; 857 } else { 858 m = m_get(M_WAIT, MT_DATA); 859 mlen = MLEN; 860 } 861 MCLAIM(m, so->so_snd.sb_mowner); 862 if (sock_loan_thresh >= 0 && 863 uio->uio_iov->iov_len >= sock_loan_thresh && 864 space >= sock_loan_thresh && 865 (len = sosend_loan(so, uio, m, 866 space)) != 0) { 867 SOSEND_COUNTER_INCR(&sosend_loan_big); 868 space -= len; 869 goto have_data; 870 } 871 if (resid >= MINCLSIZE && space >= MCLBYTES) { 872 SOSEND_COUNTER_INCR(&sosend_copy_big); 873 m_clget(m, M_WAIT); 874 if ((m->m_flags & M_EXT) == 0) 875 goto nopages; 876 mlen = MCLBYTES; 877 if (atomic && top == 0) { 878 len = lmin(MCLBYTES - max_hdr, 879 resid); 880 m->m_data += max_hdr; 881 } else 882 len = lmin(MCLBYTES, resid); 883 space -= len; 884 } else { 885 nopages: 886 SOSEND_COUNTER_INCR(&sosend_copy_small); 887 len = lmin(lmin(mlen, resid), space); 888 space -= len; 889 /* 890 * For datagram protocols, leave room 891 * for protocol headers in first mbuf. 892 */ 893 if (atomic && top == 0 && len < mlen) 894 MH_ALIGN(m, len); 895 } 896 error = uiomove(mtod(m, void *), (int)len, 897 uio); 898 have_data: 899 resid = uio->uio_resid; 900 m->m_len = len; 901 *mp = m; 902 top->m_pkthdr.len += len; 903 if (error) 904 goto release; 905 mp = &m->m_next; 906 if (resid <= 0) { 907 if (flags & MSG_EOR) 908 top->m_flags |= M_EOR; 909 break; 910 } 911 } while (space > 0 && atomic); 912 913 s = splsoftnet(); 914 915 if (so->so_state & SS_CANTSENDMORE) 916 snderr(EPIPE); 917 918 if (dontroute) 919 so->so_options |= SO_DONTROUTE; 920 if (resid > 0) 921 so->so_state |= SS_MORETOCOME; 922 error = (*so->so_proto->pr_usrreq)(so, 923 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 924 top, addr, control, curlwp); /* XXX */ 925 if (dontroute) 926 so->so_options &= ~SO_DONTROUTE; 927 if (resid > 0) 928 so->so_state &= ~SS_MORETOCOME; 929 splx(s); 930 931 clen = 0; 932 control = 0; 933 top = 0; 934 mp = ⊤ 935 if (error) 936 goto release; 937 } while (resid && space > 0); 938 } while (resid); 939 940 release: 941 sbunlock(&so->so_snd); 942 out: 943 if (top) 944 m_freem(top); 945 if (control) 946 m_freem(control); 947 return (error); 948 } 949 950 /* 951 * Implement receive operations on a socket. 952 * We depend on the way that records are added to the sockbuf 953 * by sbappend*. In particular, each record (mbufs linked through m_next) 954 * must begin with an address if the protocol so specifies, 955 * followed by an optional mbuf or mbufs containing ancillary data, 956 * and then zero or more mbufs of data. 957 * In order to avoid blocking network interrupts for the entire time here, 958 * we splx() while doing the actual copy to user space. 959 * Although the sockbuf is locked, new data may still be appended, 960 * and thus we must maintain consistency of the sockbuf during that time. 961 * 962 * The caller may receive the data as a single mbuf chain by supplying 963 * an mbuf **mp0 for use in returning the chain. The uio is then used 964 * only for the count in uio_resid. 965 */ 966 int 967 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 968 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 969 { 970 struct lwp *l = curlwp; 971 struct mbuf *m, **mp; 972 int flags, len, error, s, offset, moff, type, orig_resid; 973 const struct protosw *pr; 974 struct mbuf *nextrecord; 975 int mbuf_removed = 0; 976 977 pr = so->so_proto; 978 mp = mp0; 979 type = 0; 980 orig_resid = uio->uio_resid; 981 982 if (paddr) 983 *paddr = 0; 984 if (controlp) 985 *controlp = 0; 986 if (flagsp) 987 flags = *flagsp &~ MSG_EOR; 988 else 989 flags = 0; 990 991 if ((flags & MSG_DONTWAIT) == 0) 992 sodopendfree(); 993 994 if (flags & MSG_OOB) { 995 m = m_get(M_WAIT, MT_DATA); 996 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 997 (struct mbuf *)(long)(flags & MSG_PEEK), 998 (struct mbuf *)0, l); 999 if (error) 1000 goto bad; 1001 do { 1002 error = uiomove(mtod(m, void *), 1003 (int) min(uio->uio_resid, m->m_len), uio); 1004 m = m_free(m); 1005 } while (uio->uio_resid && error == 0 && m); 1006 bad: 1007 if (m) 1008 m_freem(m); 1009 return (error); 1010 } 1011 if (mp) 1012 *mp = (struct mbuf *)0; 1013 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 1014 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1015 (struct mbuf *)0, (struct mbuf *)0, l); 1016 1017 restart: 1018 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) 1019 return (error); 1020 s = splsoftnet(); 1021 1022 m = so->so_rcv.sb_mb; 1023 /* 1024 * If we have less data than requested, block awaiting more 1025 * (subject to any timeout) if: 1026 * 1. the current count is less than the low water mark, 1027 * 2. MSG_WAITALL is set, and it is possible to do the entire 1028 * receive operation at once if we block (resid <= hiwat), or 1029 * 3. MSG_DONTWAIT is not set. 1030 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1031 * we have to do the receive in sections, and thus risk returning 1032 * a short count if a timeout or signal occurs after we start. 1033 */ 1034 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 1035 so->so_rcv.sb_cc < uio->uio_resid) && 1036 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1037 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1038 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1039 #ifdef DIAGNOSTIC 1040 if (m == 0 && so->so_rcv.sb_cc) 1041 panic("receive 1"); 1042 #endif 1043 if (so->so_error) { 1044 if (m) 1045 goto dontblock; 1046 error = so->so_error; 1047 if ((flags & MSG_PEEK) == 0) 1048 so->so_error = 0; 1049 goto release; 1050 } 1051 if (so->so_state & SS_CANTRCVMORE) { 1052 if (m) 1053 goto dontblock; 1054 else 1055 goto release; 1056 } 1057 for (; m; m = m->m_next) 1058 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1059 m = so->so_rcv.sb_mb; 1060 goto dontblock; 1061 } 1062 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1063 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1064 error = ENOTCONN; 1065 goto release; 1066 } 1067 if (uio->uio_resid == 0) 1068 goto release; 1069 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 1070 error = EWOULDBLOCK; 1071 goto release; 1072 } 1073 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1074 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1075 sbunlock(&so->so_rcv); 1076 error = sbwait(&so->so_rcv); 1077 splx(s); 1078 if (error) 1079 return (error); 1080 goto restart; 1081 } 1082 dontblock: 1083 /* 1084 * On entry here, m points to the first record of the socket buffer. 1085 * While we process the initial mbufs containing address and control 1086 * info, we save a copy of m->m_nextpkt into nextrecord. 1087 */ 1088 if (l) 1089 l->l_proc->p_stats->p_ru.ru_msgrcv++; 1090 KASSERT(m == so->so_rcv.sb_mb); 1091 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1092 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1093 nextrecord = m->m_nextpkt; 1094 if (pr->pr_flags & PR_ADDR) { 1095 #ifdef DIAGNOSTIC 1096 if (m->m_type != MT_SONAME) 1097 panic("receive 1a"); 1098 #endif 1099 orig_resid = 0; 1100 if (flags & MSG_PEEK) { 1101 if (paddr) 1102 *paddr = m_copy(m, 0, m->m_len); 1103 m = m->m_next; 1104 } else { 1105 sbfree(&so->so_rcv, m); 1106 mbuf_removed = 1; 1107 if (paddr) { 1108 *paddr = m; 1109 so->so_rcv.sb_mb = m->m_next; 1110 m->m_next = 0; 1111 m = so->so_rcv.sb_mb; 1112 } else { 1113 MFREE(m, so->so_rcv.sb_mb); 1114 m = so->so_rcv.sb_mb; 1115 } 1116 } 1117 } 1118 while (m && m->m_type == MT_CONTROL && error == 0) { 1119 if (flags & MSG_PEEK) { 1120 if (controlp) 1121 *controlp = m_copy(m, 0, m->m_len); 1122 m = m->m_next; 1123 } else { 1124 sbfree(&so->so_rcv, m); 1125 mbuf_removed = 1; 1126 if (controlp) { 1127 struct domain *dom = pr->pr_domain; 1128 if (dom->dom_externalize && l && 1129 mtod(m, struct cmsghdr *)->cmsg_type == 1130 SCM_RIGHTS) 1131 error = (*dom->dom_externalize)(m, l); 1132 *controlp = m; 1133 so->so_rcv.sb_mb = m->m_next; 1134 m->m_next = 0; 1135 m = so->so_rcv.sb_mb; 1136 } else { 1137 /* 1138 * Dispose of any SCM_RIGHTS message that went 1139 * through the read path rather than recv. 1140 */ 1141 if (pr->pr_domain->dom_dispose && 1142 mtod(m, struct cmsghdr *)->cmsg_type == SCM_RIGHTS) 1143 (*pr->pr_domain->dom_dispose)(m); 1144 MFREE(m, so->so_rcv.sb_mb); 1145 m = so->so_rcv.sb_mb; 1146 } 1147 } 1148 if (controlp) { 1149 orig_resid = 0; 1150 controlp = &(*controlp)->m_next; 1151 } 1152 } 1153 1154 /* 1155 * If m is non-NULL, we have some data to read. From now on, 1156 * make sure to keep sb_lastrecord consistent when working on 1157 * the last packet on the chain (nextrecord == NULL) and we 1158 * change m->m_nextpkt. 1159 */ 1160 if (m) { 1161 if ((flags & MSG_PEEK) == 0) { 1162 m->m_nextpkt = nextrecord; 1163 /* 1164 * If nextrecord == NULL (this is a single chain), 1165 * then sb_lastrecord may not be valid here if m 1166 * was changed earlier. 1167 */ 1168 if (nextrecord == NULL) { 1169 KASSERT(so->so_rcv.sb_mb == m); 1170 so->so_rcv.sb_lastrecord = m; 1171 } 1172 } 1173 type = m->m_type; 1174 if (type == MT_OOBDATA) 1175 flags |= MSG_OOB; 1176 } else { 1177 if ((flags & MSG_PEEK) == 0) { 1178 KASSERT(so->so_rcv.sb_mb == m); 1179 so->so_rcv.sb_mb = nextrecord; 1180 SB_EMPTY_FIXUP(&so->so_rcv); 1181 } 1182 } 1183 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1184 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1185 1186 moff = 0; 1187 offset = 0; 1188 while (m && uio->uio_resid > 0 && error == 0) { 1189 if (m->m_type == MT_OOBDATA) { 1190 if (type != MT_OOBDATA) 1191 break; 1192 } else if (type == MT_OOBDATA) 1193 break; 1194 #ifdef DIAGNOSTIC 1195 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1196 panic("receive 3"); 1197 #endif 1198 so->so_state &= ~SS_RCVATMARK; 1199 len = uio->uio_resid; 1200 if (so->so_oobmark && len > so->so_oobmark - offset) 1201 len = so->so_oobmark - offset; 1202 if (len > m->m_len - moff) 1203 len = m->m_len - moff; 1204 /* 1205 * If mp is set, just pass back the mbufs. 1206 * Otherwise copy them out via the uio, then free. 1207 * Sockbuf must be consistent here (points to current mbuf, 1208 * it points to next record) when we drop priority; 1209 * we must note any additions to the sockbuf when we 1210 * block interrupts again. 1211 */ 1212 if (mp == 0) { 1213 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1214 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1215 splx(s); 1216 error = uiomove(mtod(m, char *) + moff, (int)len, uio); 1217 s = splsoftnet(); 1218 if (error) { 1219 /* 1220 * If any part of the record has been removed 1221 * (such as the MT_SONAME mbuf, which will 1222 * happen when PR_ADDR, and thus also 1223 * PR_ATOMIC, is set), then drop the entire 1224 * record to maintain the atomicity of the 1225 * receive operation. 1226 * 1227 * This avoids a later panic("receive 1a") 1228 * when compiled with DIAGNOSTIC. 1229 */ 1230 if (m && mbuf_removed 1231 && (pr->pr_flags & PR_ATOMIC)) 1232 (void) sbdroprecord(&so->so_rcv); 1233 1234 goto release; 1235 } 1236 } else 1237 uio->uio_resid -= len; 1238 if (len == m->m_len - moff) { 1239 if (m->m_flags & M_EOR) 1240 flags |= MSG_EOR; 1241 if (flags & MSG_PEEK) { 1242 m = m->m_next; 1243 moff = 0; 1244 } else { 1245 nextrecord = m->m_nextpkt; 1246 sbfree(&so->so_rcv, m); 1247 if (mp) { 1248 *mp = m; 1249 mp = &m->m_next; 1250 so->so_rcv.sb_mb = m = m->m_next; 1251 *mp = (struct mbuf *)0; 1252 } else { 1253 MFREE(m, so->so_rcv.sb_mb); 1254 m = so->so_rcv.sb_mb; 1255 } 1256 /* 1257 * If m != NULL, we also know that 1258 * so->so_rcv.sb_mb != NULL. 1259 */ 1260 KASSERT(so->so_rcv.sb_mb == m); 1261 if (m) { 1262 m->m_nextpkt = nextrecord; 1263 if (nextrecord == NULL) 1264 so->so_rcv.sb_lastrecord = m; 1265 } else { 1266 so->so_rcv.sb_mb = nextrecord; 1267 SB_EMPTY_FIXUP(&so->so_rcv); 1268 } 1269 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1270 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1271 } 1272 } else { 1273 if (flags & MSG_PEEK) 1274 moff += len; 1275 else { 1276 if (mp) 1277 *mp = m_copym(m, 0, len, M_WAIT); 1278 m->m_data += len; 1279 m->m_len -= len; 1280 so->so_rcv.sb_cc -= len; 1281 } 1282 } 1283 if (so->so_oobmark) { 1284 if ((flags & MSG_PEEK) == 0) { 1285 so->so_oobmark -= len; 1286 if (so->so_oobmark == 0) { 1287 so->so_state |= SS_RCVATMARK; 1288 break; 1289 } 1290 } else { 1291 offset += len; 1292 if (offset == so->so_oobmark) 1293 break; 1294 } 1295 } 1296 if (flags & MSG_EOR) 1297 break; 1298 /* 1299 * If the MSG_WAITALL flag is set (for non-atomic socket), 1300 * we must not quit until "uio->uio_resid == 0" or an error 1301 * termination. If a signal/timeout occurs, return 1302 * with a short count but without error. 1303 * Keep sockbuf locked against other readers. 1304 */ 1305 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1306 !sosendallatonce(so) && !nextrecord) { 1307 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1308 break; 1309 /* 1310 * If we are peeking and the socket receive buffer is 1311 * full, stop since we can't get more data to peek at. 1312 */ 1313 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1314 break; 1315 /* 1316 * If we've drained the socket buffer, tell the 1317 * protocol in case it needs to do something to 1318 * get it filled again. 1319 */ 1320 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1321 (*pr->pr_usrreq)(so, PRU_RCVD, 1322 (struct mbuf *)0, 1323 (struct mbuf *)(long)flags, 1324 (struct mbuf *)0, l); 1325 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1326 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1327 error = sbwait(&so->so_rcv); 1328 if (error) { 1329 sbunlock(&so->so_rcv); 1330 splx(s); 1331 return (0); 1332 } 1333 if ((m = so->so_rcv.sb_mb) != NULL) 1334 nextrecord = m->m_nextpkt; 1335 } 1336 } 1337 1338 if (m && pr->pr_flags & PR_ATOMIC) { 1339 flags |= MSG_TRUNC; 1340 if ((flags & MSG_PEEK) == 0) 1341 (void) sbdroprecord(&so->so_rcv); 1342 } 1343 if ((flags & MSG_PEEK) == 0) { 1344 if (m == 0) { 1345 /* 1346 * First part is an inline SB_EMPTY_FIXUP(). Second 1347 * part makes sure sb_lastrecord is up-to-date if 1348 * there is still data in the socket buffer. 1349 */ 1350 so->so_rcv.sb_mb = nextrecord; 1351 if (so->so_rcv.sb_mb == NULL) { 1352 so->so_rcv.sb_mbtail = NULL; 1353 so->so_rcv.sb_lastrecord = NULL; 1354 } else if (nextrecord->m_nextpkt == NULL) 1355 so->so_rcv.sb_lastrecord = nextrecord; 1356 } 1357 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1358 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1359 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1360 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1361 (struct mbuf *)(long)flags, (struct mbuf *)0, l); 1362 } 1363 if (orig_resid == uio->uio_resid && orig_resid && 1364 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1365 sbunlock(&so->so_rcv); 1366 splx(s); 1367 goto restart; 1368 } 1369 1370 if (flagsp) 1371 *flagsp |= flags; 1372 release: 1373 sbunlock(&so->so_rcv); 1374 splx(s); 1375 return (error); 1376 } 1377 1378 int 1379 soshutdown(struct socket *so, int how) 1380 { 1381 const struct protosw *pr; 1382 1383 pr = so->so_proto; 1384 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1385 return (EINVAL); 1386 1387 if (how == SHUT_RD || how == SHUT_RDWR) 1388 sorflush(so); 1389 if (how == SHUT_WR || how == SHUT_RDWR) 1390 return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0, 1391 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1392 return (0); 1393 } 1394 1395 void 1396 sorflush(struct socket *so) 1397 { 1398 struct sockbuf *sb, asb; 1399 const struct protosw *pr; 1400 int s; 1401 1402 sb = &so->so_rcv; 1403 pr = so->so_proto; 1404 sb->sb_flags |= SB_NOINTR; 1405 (void) sblock(sb, M_WAITOK); 1406 s = splnet(); 1407 socantrcvmore(so); 1408 sbunlock(sb); 1409 asb = *sb; 1410 /* 1411 * Clear most of the sockbuf structure, but leave some of the 1412 * fields valid. 1413 */ 1414 memset(&sb->sb_startzero, 0, 1415 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1416 splx(s); 1417 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1418 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1419 sbrelease(&asb, so); 1420 } 1421 1422 int 1423 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0) 1424 { 1425 int error; 1426 struct mbuf *m; 1427 struct linger *l; 1428 1429 error = 0; 1430 m = m0; 1431 if (level != SOL_SOCKET) { 1432 if (so->so_proto && so->so_proto->pr_ctloutput) 1433 return ((*so->so_proto->pr_ctloutput) 1434 (PRCO_SETOPT, so, level, optname, &m0)); 1435 error = ENOPROTOOPT; 1436 } else { 1437 switch (optname) { 1438 1439 case SO_LINGER: 1440 if (m == NULL || m->m_len != sizeof(struct linger)) { 1441 error = EINVAL; 1442 goto bad; 1443 } 1444 l = mtod(m, struct linger *); 1445 if (l->l_linger < 0 || l->l_linger > USHRT_MAX || 1446 l->l_linger > (INT_MAX / hz)) { 1447 error = EDOM; 1448 goto bad; 1449 } 1450 so->so_linger = l->l_linger; 1451 if (l->l_onoff) 1452 so->so_options |= SO_LINGER; 1453 else 1454 so->so_options &= ~SO_LINGER; 1455 break; 1456 1457 case SO_DEBUG: 1458 case SO_KEEPALIVE: 1459 case SO_DONTROUTE: 1460 case SO_USELOOPBACK: 1461 case SO_BROADCAST: 1462 case SO_REUSEADDR: 1463 case SO_REUSEPORT: 1464 case SO_OOBINLINE: 1465 case SO_TIMESTAMP: 1466 if (m == NULL || m->m_len < sizeof(int)) { 1467 error = EINVAL; 1468 goto bad; 1469 } 1470 if (*mtod(m, int *)) 1471 so->so_options |= optname; 1472 else 1473 so->so_options &= ~optname; 1474 break; 1475 1476 case SO_SNDBUF: 1477 case SO_RCVBUF: 1478 case SO_SNDLOWAT: 1479 case SO_RCVLOWAT: 1480 { 1481 int optval; 1482 1483 if (m == NULL || m->m_len < sizeof(int)) { 1484 error = EINVAL; 1485 goto bad; 1486 } 1487 1488 /* 1489 * Values < 1 make no sense for any of these 1490 * options, so disallow them. 1491 */ 1492 optval = *mtod(m, int *); 1493 if (optval < 1) { 1494 error = EINVAL; 1495 goto bad; 1496 } 1497 1498 switch (optname) { 1499 1500 case SO_SNDBUF: 1501 case SO_RCVBUF: 1502 if (sbreserve(optname == SO_SNDBUF ? 1503 &so->so_snd : &so->so_rcv, 1504 (u_long) optval, so) == 0) { 1505 error = ENOBUFS; 1506 goto bad; 1507 } 1508 break; 1509 1510 /* 1511 * Make sure the low-water is never greater than 1512 * the high-water. 1513 */ 1514 case SO_SNDLOWAT: 1515 so->so_snd.sb_lowat = 1516 (optval > so->so_snd.sb_hiwat) ? 1517 so->so_snd.sb_hiwat : optval; 1518 break; 1519 case SO_RCVLOWAT: 1520 so->so_rcv.sb_lowat = 1521 (optval > so->so_rcv.sb_hiwat) ? 1522 so->so_rcv.sb_hiwat : optval; 1523 break; 1524 } 1525 break; 1526 } 1527 1528 case SO_SNDTIMEO: 1529 case SO_RCVTIMEO: 1530 { 1531 struct timeval *tv; 1532 int val; 1533 1534 if (m == NULL || m->m_len < sizeof(*tv)) { 1535 error = EINVAL; 1536 goto bad; 1537 } 1538 tv = mtod(m, struct timeval *); 1539 if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz) { 1540 error = EDOM; 1541 goto bad; 1542 } 1543 val = tv->tv_sec * hz + tv->tv_usec / tick; 1544 if (val == 0 && tv->tv_usec != 0) 1545 val = 1; 1546 1547 switch (optname) { 1548 1549 case SO_SNDTIMEO: 1550 so->so_snd.sb_timeo = val; 1551 break; 1552 case SO_RCVTIMEO: 1553 so->so_rcv.sb_timeo = val; 1554 break; 1555 } 1556 break; 1557 } 1558 1559 default: 1560 error = ENOPROTOOPT; 1561 break; 1562 } 1563 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1564 (void) ((*so->so_proto->pr_ctloutput) 1565 (PRCO_SETOPT, so, level, optname, &m0)); 1566 m = NULL; /* freed by protocol */ 1567 } 1568 } 1569 bad: 1570 if (m) 1571 (void) m_free(m); 1572 return (error); 1573 } 1574 1575 int 1576 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp) 1577 { 1578 struct mbuf *m; 1579 1580 if (level != SOL_SOCKET) { 1581 if (so->so_proto && so->so_proto->pr_ctloutput) { 1582 return ((*so->so_proto->pr_ctloutput) 1583 (PRCO_GETOPT, so, level, optname, mp)); 1584 } else 1585 return (ENOPROTOOPT); 1586 } else { 1587 m = m_get(M_WAIT, MT_SOOPTS); 1588 m->m_len = sizeof(int); 1589 1590 switch (optname) { 1591 1592 case SO_LINGER: 1593 m->m_len = sizeof(struct linger); 1594 mtod(m, struct linger *)->l_onoff = 1595 (so->so_options & SO_LINGER) ? 1 : 0; 1596 mtod(m, struct linger *)->l_linger = so->so_linger; 1597 break; 1598 1599 case SO_USELOOPBACK: 1600 case SO_DONTROUTE: 1601 case SO_DEBUG: 1602 case SO_KEEPALIVE: 1603 case SO_REUSEADDR: 1604 case SO_REUSEPORT: 1605 case SO_BROADCAST: 1606 case SO_OOBINLINE: 1607 case SO_TIMESTAMP: 1608 *mtod(m, int *) = (so->so_options & optname) ? 1 : 0; 1609 break; 1610 1611 case SO_TYPE: 1612 *mtod(m, int *) = so->so_type; 1613 break; 1614 1615 case SO_ERROR: 1616 *mtod(m, int *) = so->so_error; 1617 so->so_error = 0; 1618 break; 1619 1620 case SO_SNDBUF: 1621 *mtod(m, int *) = so->so_snd.sb_hiwat; 1622 break; 1623 1624 case SO_RCVBUF: 1625 *mtod(m, int *) = so->so_rcv.sb_hiwat; 1626 break; 1627 1628 case SO_SNDLOWAT: 1629 *mtod(m, int *) = so->so_snd.sb_lowat; 1630 break; 1631 1632 case SO_RCVLOWAT: 1633 *mtod(m, int *) = so->so_rcv.sb_lowat; 1634 break; 1635 1636 case SO_SNDTIMEO: 1637 case SO_RCVTIMEO: 1638 { 1639 int val = (optname == SO_SNDTIMEO ? 1640 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1641 1642 m->m_len = sizeof(struct timeval); 1643 mtod(m, struct timeval *)->tv_sec = val / hz; 1644 mtod(m, struct timeval *)->tv_usec = 1645 (val % hz) * tick; 1646 break; 1647 } 1648 1649 case SO_OVERFLOWED: 1650 *mtod(m, int *) = so->so_rcv.sb_overflowed; 1651 break; 1652 1653 default: 1654 (void)m_free(m); 1655 return (ENOPROTOOPT); 1656 } 1657 *mp = m; 1658 return (0); 1659 } 1660 } 1661 1662 void 1663 sohasoutofband(struct socket *so) 1664 { 1665 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 1666 selwakeup(&so->so_rcv.sb_sel); 1667 } 1668 1669 static void 1670 filt_sordetach(struct knote *kn) 1671 { 1672 struct socket *so; 1673 1674 so = (struct socket *)kn->kn_fp->f_data; 1675 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 1676 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 1677 so->so_rcv.sb_flags &= ~SB_KNOTE; 1678 } 1679 1680 /*ARGSUSED*/ 1681 static int 1682 filt_soread(struct knote *kn, long hint) 1683 { 1684 struct socket *so; 1685 1686 so = (struct socket *)kn->kn_fp->f_data; 1687 kn->kn_data = so->so_rcv.sb_cc; 1688 if (so->so_state & SS_CANTRCVMORE) { 1689 kn->kn_flags |= EV_EOF; 1690 kn->kn_fflags = so->so_error; 1691 return (1); 1692 } 1693 if (so->so_error) /* temporary udp error */ 1694 return (1); 1695 if (kn->kn_sfflags & NOTE_LOWAT) 1696 return (kn->kn_data >= kn->kn_sdata); 1697 return (kn->kn_data >= so->so_rcv.sb_lowat); 1698 } 1699 1700 static void 1701 filt_sowdetach(struct knote *kn) 1702 { 1703 struct socket *so; 1704 1705 so = (struct socket *)kn->kn_fp->f_data; 1706 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 1707 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 1708 so->so_snd.sb_flags &= ~SB_KNOTE; 1709 } 1710 1711 /*ARGSUSED*/ 1712 static int 1713 filt_sowrite(struct knote *kn, long hint) 1714 { 1715 struct socket *so; 1716 1717 so = (struct socket *)kn->kn_fp->f_data; 1718 kn->kn_data = sbspace(&so->so_snd); 1719 if (so->so_state & SS_CANTSENDMORE) { 1720 kn->kn_flags |= EV_EOF; 1721 kn->kn_fflags = so->so_error; 1722 return (1); 1723 } 1724 if (so->so_error) /* temporary udp error */ 1725 return (1); 1726 if (((so->so_state & SS_ISCONNECTED) == 0) && 1727 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1728 return (0); 1729 if (kn->kn_sfflags & NOTE_LOWAT) 1730 return (kn->kn_data >= kn->kn_sdata); 1731 return (kn->kn_data >= so->so_snd.sb_lowat); 1732 } 1733 1734 /*ARGSUSED*/ 1735 static int 1736 filt_solisten(struct knote *kn, long hint) 1737 { 1738 struct socket *so; 1739 1740 so = (struct socket *)kn->kn_fp->f_data; 1741 1742 /* 1743 * Set kn_data to number of incoming connections, not 1744 * counting partial (incomplete) connections. 1745 */ 1746 kn->kn_data = so->so_qlen; 1747 return (kn->kn_data > 0); 1748 } 1749 1750 static const struct filterops solisten_filtops = 1751 { 1, NULL, filt_sordetach, filt_solisten }; 1752 static const struct filterops soread_filtops = 1753 { 1, NULL, filt_sordetach, filt_soread }; 1754 static const struct filterops sowrite_filtops = 1755 { 1, NULL, filt_sowdetach, filt_sowrite }; 1756 1757 int 1758 soo_kqfilter(struct file *fp, struct knote *kn) 1759 { 1760 struct socket *so; 1761 struct sockbuf *sb; 1762 1763 so = (struct socket *)kn->kn_fp->f_data; 1764 switch (kn->kn_filter) { 1765 case EVFILT_READ: 1766 if (so->so_options & SO_ACCEPTCONN) 1767 kn->kn_fop = &solisten_filtops; 1768 else 1769 kn->kn_fop = &soread_filtops; 1770 sb = &so->so_rcv; 1771 break; 1772 case EVFILT_WRITE: 1773 kn->kn_fop = &sowrite_filtops; 1774 sb = &so->so_snd; 1775 break; 1776 default: 1777 return (1); 1778 } 1779 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 1780 sb->sb_flags |= SB_KNOTE; 1781 return (0); 1782 } 1783 1784 #include <sys/sysctl.h> 1785 1786 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 1787 1788 /* 1789 * sysctl helper routine for kern.somaxkva. ensures that the given 1790 * value is not too small. 1791 * (XXX should we maybe make sure it's not too large as well?) 1792 */ 1793 static int 1794 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 1795 { 1796 int error, new_somaxkva; 1797 struct sysctlnode node; 1798 int s; 1799 1800 new_somaxkva = somaxkva; 1801 node = *rnode; 1802 node.sysctl_data = &new_somaxkva; 1803 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1804 if (error || newp == NULL) 1805 return (error); 1806 1807 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 1808 return (EINVAL); 1809 1810 s = splvm(); 1811 simple_lock(&so_pendfree_slock); 1812 somaxkva = new_somaxkva; 1813 wakeup(&socurkva); 1814 simple_unlock(&so_pendfree_slock); 1815 splx(s); 1816 1817 return (error); 1818 } 1819 1820 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup") 1821 { 1822 1823 sysctl_createv(clog, 0, NULL, NULL, 1824 CTLFLAG_PERMANENT, 1825 CTLTYPE_NODE, "kern", NULL, 1826 NULL, 0, NULL, 0, 1827 CTL_KERN, CTL_EOL); 1828 1829 sysctl_createv(clog, 0, NULL, NULL, 1830 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1831 CTLTYPE_INT, "somaxkva", 1832 SYSCTL_DESCR("Maximum amount of kernel memory to be " 1833 "used for socket buffers"), 1834 sysctl_kern_somaxkva, 0, NULL, 0, 1835 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 1836 } 1837
Indexes created Sat Sep 20 22:09:52 GMT 2025