uipc_socket.c revision 1.134
1 /* $NetBSD: uipc_socket.c,v 1.134 2007/03/04 06:03:11 christos 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 1.134 2007/03/04 06:03:11 christos 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); 99 100 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options"); 101 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 102 103 extern int somaxconn; /* patchable (XXX sysctl) */ 104 int somaxconn = SOMAXCONN; 105 106 #ifdef SOSEND_COUNTERS 107 #include <sys/device.h> 108 109 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 110 NULL, "sosend", "loan big"); 111 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 112 NULL, "sosend", "copy big"); 113 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 114 NULL, "sosend", "copy small"); 115 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 116 NULL, "sosend", "kva limit"); 117 118 #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++ 119 120 EVCNT_ATTACH_STATIC(sosend_loan_big); 121 EVCNT_ATTACH_STATIC(sosend_copy_big); 122 EVCNT_ATTACH_STATIC(sosend_copy_small); 123 EVCNT_ATTACH_STATIC(sosend_kvalimit); 124 #else 125 126 #define SOSEND_COUNTER_INCR(ev) /* nothing */ 127 128 #endif /* SOSEND_COUNTERS */ 129 130 static struct callback_entry sokva_reclaimerentry; 131 132 #ifdef SOSEND_NO_LOAN 133 int sock_loan_thresh = -1; 134 #else 135 int sock_loan_thresh = 4096; 136 #endif 137 138 static struct simplelock so_pendfree_slock = SIMPLELOCK_INITIALIZER; 139 static struct mbuf *so_pendfree; 140 141 #ifndef SOMAXKVA 142 #define SOMAXKVA (16 * 1024 * 1024) 143 #endif 144 int somaxkva = SOMAXKVA; 145 static int socurkva; 146 static int sokvawaiters; 147 148 #define SOCK_LOAN_CHUNK 65536 149 150 static size_t sodopendfree(void); 151 static size_t sodopendfreel(void); 152 153 static vsize_t 154 sokvareserve(struct socket *so, vsize_t len) 155 { 156 int s; 157 int error; 158 159 s = splvm(); 160 simple_lock(&so_pendfree_slock); 161 while (socurkva + len > somaxkva) { 162 size_t freed; 163 164 /* 165 * try to do pendfree. 166 */ 167 168 freed = sodopendfreel(); 169 170 /* 171 * if some kva was freed, try again. 172 */ 173 174 if (freed) 175 continue; 176 177 SOSEND_COUNTER_INCR(&sosend_kvalimit); 178 sokvawaiters++; 179 error = ltsleep(&socurkva, PVM | PCATCH, "sokva", 0, 180 &so_pendfree_slock); 181 sokvawaiters--; 182 if (error) { 183 len = 0; 184 break; 185 } 186 } 187 socurkva += len; 188 simple_unlock(&so_pendfree_slock); 189 splx(s); 190 return len; 191 } 192 193 static void 194 sokvaunreserve(vsize_t len) 195 { 196 int s; 197 198 s = splvm(); 199 simple_lock(&so_pendfree_slock); 200 socurkva -= len; 201 if (sokvawaiters) 202 wakeup(&socurkva); 203 simple_unlock(&so_pendfree_slock); 204 splx(s); 205 } 206 207 /* 208 * sokvaalloc: allocate kva for loan. 209 */ 210 211 vaddr_t 212 sokvaalloc(vsize_t len, struct socket *so) 213 { 214 vaddr_t lva; 215 216 /* 217 * reserve kva. 218 */ 219 220 if (sokvareserve(so, len) == 0) 221 return 0; 222 223 /* 224 * allocate kva. 225 */ 226 227 lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA); 228 if (lva == 0) { 229 sokvaunreserve(len); 230 return (0); 231 } 232 233 return lva; 234 } 235 236 /* 237 * sokvafree: free kva for loan. 238 */ 239 240 void 241 sokvafree(vaddr_t sva, vsize_t len) 242 { 243 244 /* 245 * free kva. 246 */ 247 248 uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY); 249 250 /* 251 * unreserve kva. 252 */ 253 254 sokvaunreserve(len); 255 } 256 257 static void 258 sodoloanfree(struct vm_page **pgs, void *buf, size_t size) 259 { 260 vaddr_t va, sva, eva; 261 vsize_t len; 262 paddr_t pa; 263 int i, npgs; 264 265 eva = round_page((vaddr_t) buf + size); 266 sva = trunc_page((vaddr_t) buf); 267 len = eva - sva; 268 npgs = len >> PAGE_SHIFT; 269 270 if (__predict_false(pgs == NULL)) { 271 pgs = alloca(npgs * sizeof(*pgs)); 272 273 for (i = 0, va = sva; va < eva; i++, va += PAGE_SIZE) { 274 if (pmap_extract(pmap_kernel(), va, &pa) == false) 275 panic("sodoloanfree: va 0x%lx not mapped", va); 276 pgs[i] = PHYS_TO_VM_PAGE(pa); 277 } 278 } 279 280 pmap_kremove(sva, len); 281 pmap_update(pmap_kernel()); 282 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 283 sokvafree(sva, len); 284 } 285 286 static size_t 287 sodopendfree() 288 { 289 int s; 290 size_t rv; 291 292 s = splvm(); 293 simple_lock(&so_pendfree_slock); 294 rv = sodopendfreel(); 295 simple_unlock(&so_pendfree_slock); 296 splx(s); 297 298 return rv; 299 } 300 301 /* 302 * sodopendfreel: free mbufs on "pendfree" list. 303 * unlock and relock so_pendfree_slock when freeing mbufs. 304 * 305 * => called with so_pendfree_slock held. 306 * => called at splvm. 307 */ 308 309 static size_t 310 sodopendfreel() 311 { 312 size_t rv = 0; 313 314 LOCK_ASSERT(simple_lock_held(&so_pendfree_slock)); 315 316 for (;;) { 317 struct mbuf *m; 318 struct mbuf *next; 319 320 m = so_pendfree; 321 if (m == NULL) 322 break; 323 so_pendfree = NULL; 324 simple_unlock(&so_pendfree_slock); 325 /* XXX splx */ 326 327 for (; m != NULL; m = next) { 328 next = m->m_next; 329 330 rv += m->m_ext.ext_size; 331 sodoloanfree((m->m_flags & M_EXT_PAGES) ? 332 m->m_ext.ext_pgs : NULL, m->m_ext.ext_buf, 333 m->m_ext.ext_size); 334 pool_cache_put(&mbpool_cache, m); 335 } 336 337 /* XXX splvm */ 338 simple_lock(&so_pendfree_slock); 339 } 340 341 return (rv); 342 } 343 344 void 345 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg) 346 { 347 int s; 348 349 if (m == NULL) { 350 351 /* 352 * called from MEXTREMOVE. 353 */ 354 355 sodoloanfree(NULL, buf, size); 356 return; 357 } 358 359 /* 360 * postpone freeing mbuf. 361 * 362 * we can't do it in interrupt context 363 * because we need to put kva back to kernel_map. 364 */ 365 366 s = splvm(); 367 simple_lock(&so_pendfree_slock); 368 m->m_next = so_pendfree; 369 so_pendfree = m; 370 if (sokvawaiters) 371 wakeup(&socurkva); 372 simple_unlock(&so_pendfree_slock); 373 splx(s); 374 } 375 376 static long 377 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 378 { 379 struct iovec *iov = uio->uio_iov; 380 vaddr_t sva, eva; 381 vsize_t len; 382 vaddr_t lva, va; 383 int npgs, i, error; 384 385 if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace)) 386 return (0); 387 388 if (iov->iov_len < (size_t) space) 389 space = iov->iov_len; 390 if (space > SOCK_LOAN_CHUNK) 391 space = SOCK_LOAN_CHUNK; 392 393 eva = round_page((vaddr_t) iov->iov_base + space); 394 sva = trunc_page((vaddr_t) iov->iov_base); 395 len = eva - sva; 396 npgs = len >> PAGE_SHIFT; 397 398 /* XXX KDASSERT */ 399 KASSERT(npgs <= M_EXT_MAXPAGES); 400 401 lva = sokvaalloc(len, so); 402 if (lva == 0) 403 return 0; 404 405 error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len, 406 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 407 if (error) { 408 sokvafree(lva, len); 409 return (0); 410 } 411 412 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 413 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 414 VM_PROT_READ); 415 pmap_update(pmap_kernel()); 416 417 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 418 419 MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so); 420 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 421 422 uio->uio_resid -= space; 423 /* uio_offset not updated, not set/used for write(2) */ 424 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space; 425 uio->uio_iov->iov_len -= space; 426 if (uio->uio_iov->iov_len == 0) { 427 uio->uio_iov++; 428 uio->uio_iovcnt--; 429 } 430 431 return (space); 432 } 433 434 static int 435 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg) 436 { 437 438 KASSERT(ce == &sokva_reclaimerentry); 439 KASSERT(obj == NULL); 440 441 sodopendfree(); 442 if (!vm_map_starved_p(kernel_map)) { 443 return CALLBACK_CHAIN_ABORT; 444 } 445 return CALLBACK_CHAIN_CONTINUE; 446 } 447 448 void 449 soinit(void) 450 { 451 452 /* Set the initial adjusted socket buffer size. */ 453 if (sb_max_set(sb_max)) 454 panic("bad initial sb_max value: %lu", sb_max); 455 456 callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback, 457 &sokva_reclaimerentry, NULL, sokva_reclaim_callback); 458 } 459 460 /* 461 * Socket operation routines. 462 * These routines are called by the routines in 463 * sys_socket.c or from a system process, and 464 * implement the semantics of socket operations by 465 * switching out to the protocol specific routines. 466 */ 467 /*ARGSUSED*/ 468 int 469 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l) 470 { 471 const struct protosw *prp; 472 struct socket *so; 473 uid_t uid; 474 int error, s; 475 476 error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET, 477 KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type), 478 KAUTH_ARG(proto)); 479 if (error) 480 return (error); 481 482 if (proto) 483 prp = pffindproto(dom, proto, type); 484 else 485 prp = pffindtype(dom, type); 486 if (prp == 0) { 487 /* no support for domain */ 488 if (pffinddomain(dom) == 0) 489 return (EAFNOSUPPORT); 490 /* no support for socket type */ 491 if (proto == 0 && type != 0) 492 return (EPROTOTYPE); 493 return (EPROTONOSUPPORT); 494 } 495 if (prp->pr_usrreq == 0) 496 return (EPROTONOSUPPORT); 497 if (prp->pr_type != type) 498 return (EPROTOTYPE); 499 s = splsoftnet(); 500 so = pool_get(&socket_pool, PR_WAITOK); 501 memset((void *)so, 0, sizeof(*so)); 502 TAILQ_INIT(&so->so_q0); 503 TAILQ_INIT(&so->so_q); 504 so->so_type = type; 505 so->so_proto = prp; 506 so->so_send = sosend; 507 so->so_receive = soreceive; 508 #ifdef MBUFTRACE 509 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 510 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 511 so->so_mowner = &prp->pr_domain->dom_mowner; 512 #endif 513 if (l != NULL) { 514 uid = kauth_cred_geteuid(l->l_cred); 515 } else { 516 uid = 0; 517 } 518 so->so_uidinfo = uid_find(uid); 519 error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0, 520 (struct mbuf *)(long)proto, (struct mbuf *)0, l); 521 if (error) { 522 so->so_state |= SS_NOFDREF; 523 sofree(so); 524 splx(s); 525 return (error); 526 } 527 splx(s); 528 *aso = so; 529 return (0); 530 } 531 532 int 533 sobind(struct socket *so, struct mbuf *nam, struct lwp *l) 534 { 535 int s, error; 536 537 s = splsoftnet(); 538 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0, 539 nam, (struct mbuf *)0, l); 540 splx(s); 541 return (error); 542 } 543 544 int 545 solisten(struct socket *so, int backlog) 546 { 547 int s, error; 548 549 s = splsoftnet(); 550 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0, 551 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 552 if (error) { 553 splx(s); 554 return (error); 555 } 556 if (TAILQ_EMPTY(&so->so_q)) 557 so->so_options |= SO_ACCEPTCONN; 558 if (backlog < 0) 559 backlog = 0; 560 so->so_qlimit = min(backlog, somaxconn); 561 splx(s); 562 return (0); 563 } 564 565 void 566 sofree(struct socket *so) 567 { 568 569 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 570 return; 571 if (so->so_head) { 572 /* 573 * We must not decommission a socket that's on the accept(2) 574 * queue. If we do, then accept(2) may hang after select(2) 575 * indicated that the listening socket was ready. 576 */ 577 if (!soqremque(so, 0)) 578 return; 579 } 580 if (so->so_rcv.sb_hiwat) 581 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 582 RLIM_INFINITY); 583 if (so->so_snd.sb_hiwat) 584 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 585 RLIM_INFINITY); 586 sbrelease(&so->so_snd, so); 587 sorflush(so); 588 pool_put(&socket_pool, so); 589 } 590 591 /* 592 * Close a socket on last file table reference removal. 593 * Initiate disconnect if connected. 594 * Free socket when disconnect complete. 595 */ 596 int 597 soclose(struct socket *so) 598 { 599 struct socket *so2; 600 int s, error; 601 602 error = 0; 603 s = splsoftnet(); /* conservative */ 604 if (so->so_options & SO_ACCEPTCONN) { 605 while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 606 (void) soqremque(so2, 0); 607 (void) soabort(so2); 608 } 609 while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 610 (void) soqremque(so2, 1); 611 (void) soabort(so2); 612 } 613 } 614 if (so->so_pcb == 0) 615 goto discard; 616 if (so->so_state & SS_ISCONNECTED) { 617 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 618 error = sodisconnect(so); 619 if (error) 620 goto drop; 621 } 622 if (so->so_options & SO_LINGER) { 623 if ((so->so_state & SS_ISDISCONNECTING) && 624 (so->so_state & SS_NBIO)) 625 goto drop; 626 while (so->so_state & SS_ISCONNECTED) { 627 error = tsleep((void *)&so->so_timeo, 628 PSOCK | PCATCH, netcls, 629 so->so_linger * hz); 630 if (error) 631 break; 632 } 633 } 634 } 635 drop: 636 if (so->so_pcb) { 637 int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 638 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 639 (struct lwp *)0); 640 if (error == 0) 641 error = error2; 642 } 643 discard: 644 if (so->so_state & SS_NOFDREF) 645 panic("soclose: NOFDREF"); 646 so->so_state |= SS_NOFDREF; 647 sofree(so); 648 splx(s); 649 return (error); 650 } 651 652 /* 653 * Must be called at splsoftnet... 654 */ 655 int 656 soabort(struct socket *so) 657 { 658 659 return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0, 660 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 661 } 662 663 int 664 soaccept(struct socket *so, struct mbuf *nam) 665 { 666 int s, error; 667 668 error = 0; 669 s = splsoftnet(); 670 if ((so->so_state & SS_NOFDREF) == 0) 671 panic("soaccept: !NOFDREF"); 672 so->so_state &= ~SS_NOFDREF; 673 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 674 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 675 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 676 (struct mbuf *)0, nam, (struct mbuf *)0, (struct lwp *)0); 677 else 678 error = ECONNABORTED; 679 680 splx(s); 681 return (error); 682 } 683 684 int 685 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l) 686 { 687 int s, error; 688 689 if (so->so_options & SO_ACCEPTCONN) 690 return (EOPNOTSUPP); 691 s = splsoftnet(); 692 /* 693 * If protocol is connection-based, can only connect once. 694 * Otherwise, if connected, try to disconnect first. 695 * This allows user to disconnect by connecting to, e.g., 696 * a null address. 697 */ 698 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 699 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 700 (error = sodisconnect(so)))) 701 error = EISCONN; 702 else 703 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 704 (struct mbuf *)0, nam, (struct mbuf *)0, l); 705 splx(s); 706 return (error); 707 } 708 709 int 710 soconnect2(struct socket *so1, struct socket *so2) 711 { 712 int s, error; 713 714 s = splsoftnet(); 715 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 716 (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0, 717 (struct lwp *)0); 718 splx(s); 719 return (error); 720 } 721 722 int 723 sodisconnect(struct socket *so) 724 { 725 int s, error; 726 727 s = splsoftnet(); 728 if ((so->so_state & SS_ISCONNECTED) == 0) { 729 error = ENOTCONN; 730 goto bad; 731 } 732 if (so->so_state & SS_ISDISCONNECTING) { 733 error = EALREADY; 734 goto bad; 735 } 736 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 737 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 738 (struct lwp *)0); 739 bad: 740 splx(s); 741 sodopendfree(); 742 return (error); 743 } 744 745 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 746 /* 747 * Send on a socket. 748 * If send must go all at once and message is larger than 749 * send buffering, then hard error. 750 * Lock against other senders. 751 * If must go all at once and not enough room now, then 752 * inform user that this would block and do nothing. 753 * Otherwise, if nonblocking, send as much as possible. 754 * The data to be sent is described by "uio" if nonzero, 755 * otherwise by the mbuf chain "top" (which must be null 756 * if uio is not). Data provided in mbuf chain must be small 757 * enough to send all at once. 758 * 759 * Returns nonzero on error, timeout or signal; callers 760 * must check for short counts if EINTR/ERESTART are returned. 761 * Data and control buffers are freed on return. 762 */ 763 int 764 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 765 struct mbuf *control, int flags, struct lwp *l) 766 { 767 struct mbuf **mp, *m; 768 struct proc *p; 769 long space, len, resid, clen, mlen; 770 int error, s, dontroute, atomic; 771 772 p = l->l_proc; 773 sodopendfree(); 774 775 clen = 0; 776 atomic = sosendallatonce(so) || top; 777 if (uio) 778 resid = uio->uio_resid; 779 else 780 resid = top->m_pkthdr.len; 781 /* 782 * In theory resid should be unsigned. 783 * However, space must be signed, as it might be less than 0 784 * if we over-committed, and we must use a signed comparison 785 * of space and resid. On the other hand, a negative resid 786 * causes us to loop sending 0-length segments to the protocol. 787 */ 788 if (resid < 0) { 789 error = EINVAL; 790 goto out; 791 } 792 dontroute = 793 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 794 (so->so_proto->pr_flags & PR_ATOMIC); 795 if (p) 796 p->p_stats->p_ru.ru_msgsnd++; 797 if (control) 798 clen = control->m_len; 799 #define snderr(errno) { error = errno; splx(s); goto release; } 800 801 restart: 802 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 803 goto out; 804 do { 805 s = splsoftnet(); 806 if (so->so_state & SS_CANTSENDMORE) 807 snderr(EPIPE); 808 if (so->so_error) { 809 error = so->so_error; 810 so->so_error = 0; 811 splx(s); 812 goto release; 813 } 814 if ((so->so_state & SS_ISCONNECTED) == 0) { 815 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 816 if ((so->so_state & SS_ISCONFIRMING) == 0 && 817 !(resid == 0 && clen != 0)) 818 snderr(ENOTCONN); 819 } else if (addr == 0) 820 snderr(EDESTADDRREQ); 821 } 822 space = sbspace(&so->so_snd); 823 if (flags & MSG_OOB) 824 space += 1024; 825 if ((atomic && resid > so->so_snd.sb_hiwat) || 826 clen > so->so_snd.sb_hiwat) 827 snderr(EMSGSIZE); 828 if (space < resid + clen && 829 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 830 if (so->so_state & SS_NBIO) 831 snderr(EWOULDBLOCK); 832 sbunlock(&so->so_snd); 833 error = sbwait(&so->so_snd); 834 splx(s); 835 if (error) 836 goto out; 837 goto restart; 838 } 839 splx(s); 840 mp = ⊤ 841 space -= clen; 842 do { 843 if (uio == NULL) { 844 /* 845 * Data is prepackaged in "top". 846 */ 847 resid = 0; 848 if (flags & MSG_EOR) 849 top->m_flags |= M_EOR; 850 } else do { 851 if (top == 0) { 852 m = m_gethdr(M_WAIT, MT_DATA); 853 mlen = MHLEN; 854 m->m_pkthdr.len = 0; 855 m->m_pkthdr.rcvif = (struct ifnet *)0; 856 } else { 857 m = m_get(M_WAIT, MT_DATA); 858 mlen = MLEN; 859 } 860 MCLAIM(m, so->so_snd.sb_mowner); 861 if (sock_loan_thresh >= 0 && 862 uio->uio_iov->iov_len >= sock_loan_thresh && 863 space >= sock_loan_thresh && 864 (len = sosend_loan(so, uio, m, 865 space)) != 0) { 866 SOSEND_COUNTER_INCR(&sosend_loan_big); 867 space -= len; 868 goto have_data; 869 } 870 if (resid >= MINCLSIZE && space >= MCLBYTES) { 871 SOSEND_COUNTER_INCR(&sosend_copy_big); 872 m_clget(m, M_WAIT); 873 if ((m->m_flags & M_EXT) == 0) 874 goto nopages; 875 mlen = MCLBYTES; 876 if (atomic && top == 0) { 877 len = lmin(MCLBYTES - max_hdr, 878 resid); 879 m->m_data += max_hdr; 880 } else 881 len = lmin(MCLBYTES, resid); 882 space -= len; 883 } else { 884 nopages: 885 SOSEND_COUNTER_INCR(&sosend_copy_small); 886 len = lmin(lmin(mlen, resid), space); 887 space -= len; 888 /* 889 * For datagram protocols, leave room 890 * for protocol headers in first mbuf. 891 */ 892 if (atomic && top == 0 && len < mlen) 893 MH_ALIGN(m, len); 894 } 895 error = uiomove(mtod(m, void *), (int)len, 896 uio); 897 have_data: 898 resid = uio->uio_resid; 899 m->m_len = len; 900 *mp = m; 901 top->m_pkthdr.len += len; 902 if (error) 903 goto release; 904 mp = &m->m_next; 905 if (resid <= 0) { 906 if (flags & MSG_EOR) 907 top->m_flags |= M_EOR; 908 break; 909 } 910 } while (space > 0 && atomic); 911 912 s = splsoftnet(); 913 914 if (so->so_state & SS_CANTSENDMORE) 915 snderr(EPIPE); 916 917 if (dontroute) 918 so->so_options |= SO_DONTROUTE; 919 if (resid > 0) 920 so->so_state |= SS_MORETOCOME; 921 error = (*so->so_proto->pr_usrreq)(so, 922 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 923 top, addr, control, curlwp); /* XXX */ 924 if (dontroute) 925 so->so_options &= ~SO_DONTROUTE; 926 if (resid > 0) 927 so->so_state &= ~SS_MORETOCOME; 928 splx(s); 929 930 clen = 0; 931 control = 0; 932 top = 0; 933 mp = ⊤ 934 if (error) 935 goto release; 936 } while (resid && space > 0); 937 } while (resid); 938 939 release: 940 sbunlock(&so->so_snd); 941 out: 942 if (top) 943 m_freem(top); 944 if (control) 945 m_freem(control); 946 return (error); 947 } 948 949 /* 950 * Implement receive operations on a socket. 951 * We depend on the way that records are added to the sockbuf 952 * by sbappend*. In particular, each record (mbufs linked through m_next) 953 * must begin with an address if the protocol so specifies, 954 * followed by an optional mbuf or mbufs containing ancillary data, 955 * and then zero or more mbufs of data. 956 * In order to avoid blocking network interrupts for the entire time here, 957 * we splx() while doing the actual copy to user space. 958 * Although the sockbuf is locked, new data may still be appended, 959 * and thus we must maintain consistency of the sockbuf during that time. 960 * 961 * The caller may receive the data as a single mbuf chain by supplying 962 * an mbuf **mp0 for use in returning the chain. The uio is then used 963 * only for the count in uio_resid. 964 */ 965 int 966 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 967 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 968 { 969 struct lwp *l = curlwp; 970 struct mbuf *m, **mp; 971 int flags, len, error, s, offset, moff, type, orig_resid; 972 const struct protosw *pr; 973 struct mbuf *nextrecord; 974 int mbuf_removed = 0; 975 976 pr = so->so_proto; 977 mp = mp0; 978 type = 0; 979 orig_resid = uio->uio_resid; 980 981 if (paddr) 982 *paddr = 0; 983 if (controlp) 984 *controlp = 0; 985 if (flagsp) 986 flags = *flagsp &~ MSG_EOR; 987 else 988 flags = 0; 989 990 if ((flags & MSG_DONTWAIT) == 0) 991 sodopendfree(); 992 993 if (flags & MSG_OOB) { 994 m = m_get(M_WAIT, MT_DATA); 995 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 996 (struct mbuf *)(long)(flags & MSG_PEEK), 997 (struct mbuf *)0, l); 998 if (error) 999 goto bad; 1000 do { 1001 error = uiomove(mtod(m, void *), 1002 (int) min(uio->uio_resid, m->m_len), uio); 1003 m = m_free(m); 1004 } while (uio->uio_resid && error == 0 && m); 1005 bad: 1006 if (m) 1007 m_freem(m); 1008 return (error); 1009 } 1010 if (mp) 1011 *mp = (struct mbuf *)0; 1012 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 1013 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1014 (struct mbuf *)0, (struct mbuf *)0, l); 1015 1016 restart: 1017 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) 1018 return (error); 1019 s = splsoftnet(); 1020 1021 m = so->so_rcv.sb_mb; 1022 /* 1023 * If we have less data than requested, block awaiting more 1024 * (subject to any timeout) if: 1025 * 1. the current count is less than the low water mark, 1026 * 2. MSG_WAITALL is set, and it is possible to do the entire 1027 * receive operation at once if we block (resid <= hiwat), or 1028 * 3. MSG_DONTWAIT is not set. 1029 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1030 * we have to do the receive in sections, and thus risk returning 1031 * a short count if a timeout or signal occurs after we start. 1032 */ 1033 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 1034 so->so_rcv.sb_cc < uio->uio_resid) && 1035 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1036 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1037 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1038 #ifdef DIAGNOSTIC 1039 if (m == 0 && so->so_rcv.sb_cc) 1040 panic("receive 1"); 1041 #endif 1042 if (so->so_error) { 1043 if (m) 1044 goto dontblock; 1045 error = so->so_error; 1046 if ((flags & MSG_PEEK) == 0) 1047 so->so_error = 0; 1048 goto release; 1049 } 1050 if (so->so_state & SS_CANTRCVMORE) { 1051 if (m) 1052 goto dontblock; 1053 else 1054 goto release; 1055 } 1056 for (; m; m = m->m_next) 1057 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1058 m = so->so_rcv.sb_mb; 1059 goto dontblock; 1060 } 1061 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1062 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1063 error = ENOTCONN; 1064 goto release; 1065 } 1066 if (uio->uio_resid == 0) 1067 goto release; 1068 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 1069 error = EWOULDBLOCK; 1070 goto release; 1071 } 1072 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1073 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1074 sbunlock(&so->so_rcv); 1075 error = sbwait(&so->so_rcv); 1076 splx(s); 1077 if (error) 1078 return (error); 1079 goto restart; 1080 } 1081 dontblock: 1082 /* 1083 * On entry here, m points to the first record of the socket buffer. 1084 * While we process the initial mbufs containing address and control 1085 * info, we save a copy of m->m_nextpkt into nextrecord. 1086 */ 1087 if (l) 1088 l->l_proc->p_stats->p_ru.ru_msgrcv++; 1089 KASSERT(m == so->so_rcv.sb_mb); 1090 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1091 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1092 nextrecord = m->m_nextpkt; 1093 if (pr->pr_flags & PR_ADDR) { 1094 #ifdef DIAGNOSTIC 1095 if (m->m_type != MT_SONAME) 1096 panic("receive 1a"); 1097 #endif 1098 orig_resid = 0; 1099 if (flags & MSG_PEEK) { 1100 if (paddr) 1101 *paddr = m_copy(m, 0, m->m_len); 1102 m = m->m_next; 1103 } else { 1104 sbfree(&so->so_rcv, m); 1105 mbuf_removed = 1; 1106 if (paddr) { 1107 *paddr = m; 1108 so->so_rcv.sb_mb = m->m_next; 1109 m->m_next = 0; 1110 m = so->so_rcv.sb_mb; 1111 } else { 1112 MFREE(m, so->so_rcv.sb_mb); 1113 m = so->so_rcv.sb_mb; 1114 } 1115 } 1116 } 1117 while (m && m->m_type == MT_CONTROL && error == 0) { 1118 if (flags & MSG_PEEK) { 1119 if (controlp) 1120 *controlp = m_copy(m, 0, m->m_len); 1121 m = m->m_next; 1122 } else { 1123 sbfree(&so->so_rcv, m); 1124 mbuf_removed = 1; 1125 if (controlp) { 1126 struct domain *dom = pr->pr_domain; 1127 if (dom->dom_externalize && l && 1128 mtod(m, struct cmsghdr *)->cmsg_type == 1129 SCM_RIGHTS) 1130 error = (*dom->dom_externalize)(m, l); 1131 *controlp = m; 1132 so->so_rcv.sb_mb = m->m_next; 1133 m->m_next = 0; 1134 m = so->so_rcv.sb_mb; 1135 } else { 1136 /* 1137 * Dispose of any SCM_RIGHTS message that went 1138 * through the read path rather than recv. 1139 */ 1140 if (pr->pr_domain->dom_dispose && 1141 mtod(m, struct cmsghdr *)->cmsg_type == SCM_RIGHTS) 1142 (*pr->pr_domain->dom_dispose)(m); 1143 MFREE(m, so->so_rcv.sb_mb); 1144 m = so->so_rcv.sb_mb; 1145 } 1146 } 1147 if (controlp) { 1148 orig_resid = 0; 1149 controlp = &(*controlp)->m_next; 1150 } 1151 } 1152 1153 /* 1154 * If m is non-NULL, we have some data to read. From now on, 1155 * make sure to keep sb_lastrecord consistent when working on 1156 * the last packet on the chain (nextrecord == NULL) and we 1157 * change m->m_nextpkt. 1158 */ 1159 if (m) { 1160 if ((flags & MSG_PEEK) == 0) { 1161 m->m_nextpkt = nextrecord; 1162 /* 1163 * If nextrecord == NULL (this is a single chain), 1164 * then sb_lastrecord may not be valid here if m 1165 * was changed earlier. 1166 */ 1167 if (nextrecord == NULL) { 1168 KASSERT(so->so_rcv.sb_mb == m); 1169 so->so_rcv.sb_lastrecord = m; 1170 } 1171 } 1172 type = m->m_type; 1173 if (type == MT_OOBDATA) 1174 flags |= MSG_OOB; 1175 } else { 1176 if ((flags & MSG_PEEK) == 0) { 1177 KASSERT(so->so_rcv.sb_mb == m); 1178 so->so_rcv.sb_mb = nextrecord; 1179 SB_EMPTY_FIXUP(&so->so_rcv); 1180 } 1181 } 1182 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1183 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1184 1185 moff = 0; 1186 offset = 0; 1187 while (m && uio->uio_resid > 0 && error == 0) { 1188 if (m->m_type == MT_OOBDATA) { 1189 if (type != MT_OOBDATA) 1190 break; 1191 } else if (type == MT_OOBDATA) 1192 break; 1193 #ifdef DIAGNOSTIC 1194 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1195 panic("receive 3"); 1196 #endif 1197 so->so_state &= ~SS_RCVATMARK; 1198 len = uio->uio_resid; 1199 if (so->so_oobmark && len > so->so_oobmark - offset) 1200 len = so->so_oobmark - offset; 1201 if (len > m->m_len - moff) 1202 len = m->m_len - moff; 1203 /* 1204 * If mp is set, just pass back the mbufs. 1205 * Otherwise copy them out via the uio, then free. 1206 * Sockbuf must be consistent here (points to current mbuf, 1207 * it points to next record) when we drop priority; 1208 * we must note any additions to the sockbuf when we 1209 * block interrupts again. 1210 */ 1211 if (mp == 0) { 1212 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1213 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1214 splx(s); 1215 error = uiomove(mtod(m, char *) + moff, (int)len, uio); 1216 s = splsoftnet(); 1217 if (error) { 1218 /* 1219 * If any part of the record has been removed 1220 * (such as the MT_SONAME mbuf, which will 1221 * happen when PR_ADDR, and thus also 1222 * PR_ATOMIC, is set), then drop the entire 1223 * record to maintain the atomicity of the 1224 * receive operation. 1225 * 1226 * This avoids a later panic("receive 1a") 1227 * when compiled with DIAGNOSTIC. 1228 */ 1229 if (m && mbuf_removed 1230 && (pr->pr_flags & PR_ATOMIC)) 1231 (void) sbdroprecord(&so->so_rcv); 1232 1233 goto release; 1234 } 1235 } else 1236 uio->uio_resid -= len; 1237 if (len == m->m_len - moff) { 1238 if (m->m_flags & M_EOR) 1239 flags |= MSG_EOR; 1240 if (flags & MSG_PEEK) { 1241 m = m->m_next; 1242 moff = 0; 1243 } else { 1244 nextrecord = m->m_nextpkt; 1245 sbfree(&so->so_rcv, m); 1246 if (mp) { 1247 *mp = m; 1248 mp = &m->m_next; 1249 so->so_rcv.sb_mb = m = m->m_next; 1250 *mp = (struct mbuf *)0; 1251 } else { 1252 MFREE(m, so->so_rcv.sb_mb); 1253 m = so->so_rcv.sb_mb; 1254 } 1255 /* 1256 * If m != NULL, we also know that 1257 * so->so_rcv.sb_mb != NULL. 1258 */ 1259 KASSERT(so->so_rcv.sb_mb == m); 1260 if (m) { 1261 m->m_nextpkt = nextrecord; 1262 if (nextrecord == NULL) 1263 so->so_rcv.sb_lastrecord = m; 1264 } else { 1265 so->so_rcv.sb_mb = nextrecord; 1266 SB_EMPTY_FIXUP(&so->so_rcv); 1267 } 1268 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1269 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1270 } 1271 } else { 1272 if (flags & MSG_PEEK) 1273 moff += len; 1274 else { 1275 if (mp) 1276 *mp = m_copym(m, 0, len, M_WAIT); 1277 m->m_data += len; 1278 m->m_len -= len; 1279 so->so_rcv.sb_cc -= len; 1280 } 1281 } 1282 if (so->so_oobmark) { 1283 if ((flags & MSG_PEEK) == 0) { 1284 so->so_oobmark -= len; 1285 if (so->so_oobmark == 0) { 1286 so->so_state |= SS_RCVATMARK; 1287 break; 1288 } 1289 } else { 1290 offset += len; 1291 if (offset == so->so_oobmark) 1292 break; 1293 } 1294 } 1295 if (flags & MSG_EOR) 1296 break; 1297 /* 1298 * If the MSG_WAITALL flag is set (for non-atomic socket), 1299 * we must not quit until "uio->uio_resid == 0" or an error 1300 * termination. If a signal/timeout occurs, return 1301 * with a short count but without error. 1302 * Keep sockbuf locked against other readers. 1303 */ 1304 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1305 !sosendallatonce(so) && !nextrecord) { 1306 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1307 break; 1308 /* 1309 * If we are peeking and the socket receive buffer is 1310 * full, stop since we can't get more data to peek at. 1311 */ 1312 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1313 break; 1314 /* 1315 * If we've drained the socket buffer, tell the 1316 * protocol in case it needs to do something to 1317 * get it filled again. 1318 */ 1319 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1320 (*pr->pr_usrreq)(so, PRU_RCVD, 1321 (struct mbuf *)0, 1322 (struct mbuf *)(long)flags, 1323 (struct mbuf *)0, l); 1324 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1325 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1326 error = sbwait(&so->so_rcv); 1327 if (error) { 1328 sbunlock(&so->so_rcv); 1329 splx(s); 1330 return (0); 1331 } 1332 if ((m = so->so_rcv.sb_mb) != NULL) 1333 nextrecord = m->m_nextpkt; 1334 } 1335 } 1336 1337 if (m && pr->pr_flags & PR_ATOMIC) { 1338 flags |= MSG_TRUNC; 1339 if ((flags & MSG_PEEK) == 0) 1340 (void) sbdroprecord(&so->so_rcv); 1341 } 1342 if ((flags & MSG_PEEK) == 0) { 1343 if (m == 0) { 1344 /* 1345 * First part is an inline SB_EMPTY_FIXUP(). Second 1346 * part makes sure sb_lastrecord is up-to-date if 1347 * there is still data in the socket buffer. 1348 */ 1349 so->so_rcv.sb_mb = nextrecord; 1350 if (so->so_rcv.sb_mb == NULL) { 1351 so->so_rcv.sb_mbtail = NULL; 1352 so->so_rcv.sb_lastrecord = NULL; 1353 } else if (nextrecord->m_nextpkt == NULL) 1354 so->so_rcv.sb_lastrecord = nextrecord; 1355 } 1356 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1357 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1358 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1359 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1360 (struct mbuf *)(long)flags, (struct mbuf *)0, l); 1361 } 1362 if (orig_resid == uio->uio_resid && orig_resid && 1363 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1364 sbunlock(&so->so_rcv); 1365 splx(s); 1366 goto restart; 1367 } 1368 1369 if (flagsp) 1370 *flagsp |= flags; 1371 release: 1372 sbunlock(&so->so_rcv); 1373 splx(s); 1374 return (error); 1375 } 1376 1377 int 1378 soshutdown(struct socket *so, int how) 1379 { 1380 const struct protosw *pr; 1381 1382 pr = so->so_proto; 1383 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1384 return (EINVAL); 1385 1386 if (how == SHUT_RD || how == SHUT_RDWR) 1387 sorflush(so); 1388 if (how == SHUT_WR || how == SHUT_RDWR) 1389 return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0, 1390 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1391 return (0); 1392 } 1393 1394 void 1395 sorflush(struct socket *so) 1396 { 1397 struct sockbuf *sb, asb; 1398 const struct protosw *pr; 1399 int s; 1400 1401 sb = &so->so_rcv; 1402 pr = so->so_proto; 1403 sb->sb_flags |= SB_NOINTR; 1404 (void) sblock(sb, M_WAITOK); 1405 s = splnet(); 1406 socantrcvmore(so); 1407 sbunlock(sb); 1408 asb = *sb; 1409 /* 1410 * Clear most of the sockbuf structure, but leave some of the 1411 * fields valid. 1412 */ 1413 memset(&sb->sb_startzero, 0, 1414 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1415 splx(s); 1416 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1417 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1418 sbrelease(&asb, so); 1419 } 1420 1421 int 1422 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0) 1423 { 1424 int error; 1425 struct mbuf *m; 1426 struct linger *l; 1427 1428 error = 0; 1429 m = m0; 1430 if (level != SOL_SOCKET) { 1431 if (so->so_proto && so->so_proto->pr_ctloutput) 1432 return ((*so->so_proto->pr_ctloutput) 1433 (PRCO_SETOPT, so, level, optname, &m0)); 1434 error = ENOPROTOOPT; 1435 } else { 1436 switch (optname) { 1437 1438 case SO_LINGER: 1439 if (m == NULL || m->m_len != sizeof(struct linger)) { 1440 error = EINVAL; 1441 goto bad; 1442 } 1443 l = mtod(m, struct linger *); 1444 if (l->l_linger < 0 || l->l_linger > USHRT_MAX || 1445 l->l_linger > (INT_MAX / hz)) { 1446 error = EDOM; 1447 goto bad; 1448 } 1449 so->so_linger = l->l_linger; 1450 if (l->l_onoff) 1451 so->so_options |= SO_LINGER; 1452 else 1453 so->so_options &= ~SO_LINGER; 1454 break; 1455 1456 case SO_DEBUG: 1457 case SO_KEEPALIVE: 1458 case SO_DONTROUTE: 1459 case SO_USELOOPBACK: 1460 case SO_BROADCAST: 1461 case SO_REUSEADDR: 1462 case SO_REUSEPORT: 1463 case SO_OOBINLINE: 1464 case SO_TIMESTAMP: 1465 if (m == NULL || m->m_len < sizeof(int)) { 1466 error = EINVAL; 1467 goto bad; 1468 } 1469 if (*mtod(m, int *)) 1470 so->so_options |= optname; 1471 else 1472 so->so_options &= ~optname; 1473 break; 1474 1475 case SO_SNDBUF: 1476 case SO_RCVBUF: 1477 case SO_SNDLOWAT: 1478 case SO_RCVLOWAT: 1479 { 1480 int optval; 1481 1482 if (m == NULL || m->m_len < sizeof(int)) { 1483 error = EINVAL; 1484 goto bad; 1485 } 1486 1487 /* 1488 * Values < 1 make no sense for any of these 1489 * options, so disallow them. 1490 */ 1491 optval = *mtod(m, int *); 1492 if (optval < 1) { 1493 error = EINVAL; 1494 goto bad; 1495 } 1496 1497 switch (optname) { 1498 1499 case SO_SNDBUF: 1500 case SO_RCVBUF: 1501 if (sbreserve(optname == SO_SNDBUF ? 1502 &so->so_snd : &so->so_rcv, 1503 (u_long) optval, so) == 0) { 1504 error = ENOBUFS; 1505 goto bad; 1506 } 1507 break; 1508 1509 /* 1510 * Make sure the low-water is never greater than 1511 * the high-water. 1512 */ 1513 case SO_SNDLOWAT: 1514 so->so_snd.sb_lowat = 1515 (optval > so->so_snd.sb_hiwat) ? 1516 so->so_snd.sb_hiwat : optval; 1517 break; 1518 case SO_RCVLOWAT: 1519 so->so_rcv.sb_lowat = 1520 (optval > so->so_rcv.sb_hiwat) ? 1521 so->so_rcv.sb_hiwat : optval; 1522 break; 1523 } 1524 break; 1525 } 1526 1527 case SO_SNDTIMEO: 1528 case SO_RCVTIMEO: 1529 { 1530 struct timeval *tv; 1531 int val; 1532 1533 if (m == NULL || m->m_len < sizeof(*tv)) { 1534 error = EINVAL; 1535 goto bad; 1536 } 1537 tv = mtod(m, struct timeval *); 1538 if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz) { 1539 error = EDOM; 1540 goto bad; 1541 } 1542 val = tv->tv_sec * hz + tv->tv_usec / tick; 1543 if (val == 0 && tv->tv_usec != 0) 1544 val = 1; 1545 1546 switch (optname) { 1547 1548 case SO_SNDTIMEO: 1549 so->so_snd.sb_timeo = val; 1550 break; 1551 case SO_RCVTIMEO: 1552 so->so_rcv.sb_timeo = val; 1553 break; 1554 } 1555 break; 1556 } 1557 1558 default: 1559 error = ENOPROTOOPT; 1560 break; 1561 } 1562 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1563 (void) ((*so->so_proto->pr_ctloutput) 1564 (PRCO_SETOPT, so, level, optname, &m0)); 1565 m = NULL; /* freed by protocol */ 1566 } 1567 } 1568 bad: 1569 if (m) 1570 (void) m_free(m); 1571 return (error); 1572 } 1573 1574 int 1575 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp) 1576 { 1577 struct mbuf *m; 1578 1579 if (level != SOL_SOCKET) { 1580 if (so->so_proto && so->so_proto->pr_ctloutput) { 1581 return ((*so->so_proto->pr_ctloutput) 1582 (PRCO_GETOPT, so, level, optname, mp)); 1583 } else 1584 return (ENOPROTOOPT); 1585 } else { 1586 m = m_get(M_WAIT, MT_SOOPTS); 1587 m->m_len = sizeof(int); 1588 1589 switch (optname) { 1590 1591 case SO_LINGER: 1592 m->m_len = sizeof(struct linger); 1593 mtod(m, struct linger *)->l_onoff = 1594 (so->so_options & SO_LINGER) ? 1 : 0; 1595 mtod(m, struct linger *)->l_linger = so->so_linger; 1596 break; 1597 1598 case SO_USELOOPBACK: 1599 case SO_DONTROUTE: 1600 case SO_DEBUG: 1601 case SO_KEEPALIVE: 1602 case SO_REUSEADDR: 1603 case SO_REUSEPORT: 1604 case SO_BROADCAST: 1605 case SO_OOBINLINE: 1606 case SO_TIMESTAMP: 1607 *mtod(m, int *) = (so->so_options & optname) ? 1 : 0; 1608 break; 1609 1610 case SO_TYPE: 1611 *mtod(m, int *) = so->so_type; 1612 break; 1613 1614 case SO_ERROR: 1615 *mtod(m, int *) = so->so_error; 1616 so->so_error = 0; 1617 break; 1618 1619 case SO_SNDBUF: 1620 *mtod(m, int *) = so->so_snd.sb_hiwat; 1621 break; 1622 1623 case SO_RCVBUF: 1624 *mtod(m, int *) = so->so_rcv.sb_hiwat; 1625 break; 1626 1627 case SO_SNDLOWAT: 1628 *mtod(m, int *) = so->so_snd.sb_lowat; 1629 break; 1630 1631 case SO_RCVLOWAT: 1632 *mtod(m, int *) = so->so_rcv.sb_lowat; 1633 break; 1634 1635 case SO_SNDTIMEO: 1636 case SO_RCVTIMEO: 1637 { 1638 int val = (optname == SO_SNDTIMEO ? 1639 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1640 1641 m->m_len = sizeof(struct timeval); 1642 mtod(m, struct timeval *)->tv_sec = val / hz; 1643 mtod(m, struct timeval *)->tv_usec = 1644 (val % hz) * tick; 1645 break; 1646 } 1647 1648 case SO_OVERFLOWED: 1649 *mtod(m, int *) = so->so_rcv.sb_overflowed; 1650 break; 1651 1652 default: 1653 (void)m_free(m); 1654 return (ENOPROTOOPT); 1655 } 1656 *mp = m; 1657 return (0); 1658 } 1659 } 1660 1661 void 1662 sohasoutofband(struct socket *so) 1663 { 1664 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 1665 selwakeup(&so->so_rcv.sb_sel); 1666 } 1667 1668 static void 1669 filt_sordetach(struct knote *kn) 1670 { 1671 struct socket *so; 1672 1673 so = (struct socket *)kn->kn_fp->f_data; 1674 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 1675 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 1676 so->so_rcv.sb_flags &= ~SB_KNOTE; 1677 } 1678 1679 /*ARGSUSED*/ 1680 static int 1681 filt_soread(struct knote *kn, long hint) 1682 { 1683 struct socket *so; 1684 1685 so = (struct socket *)kn->kn_fp->f_data; 1686 kn->kn_data = so->so_rcv.sb_cc; 1687 if (so->so_state & SS_CANTRCVMORE) { 1688 kn->kn_flags |= EV_EOF; 1689 kn->kn_fflags = so->so_error; 1690 return (1); 1691 } 1692 if (so->so_error) /* temporary udp error */ 1693 return (1); 1694 if (kn->kn_sfflags & NOTE_LOWAT) 1695 return (kn->kn_data >= kn->kn_sdata); 1696 return (kn->kn_data >= so->so_rcv.sb_lowat); 1697 } 1698 1699 static void 1700 filt_sowdetach(struct knote *kn) 1701 { 1702 struct socket *so; 1703 1704 so = (struct socket *)kn->kn_fp->f_data; 1705 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 1706 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 1707 so->so_snd.sb_flags &= ~SB_KNOTE; 1708 } 1709 1710 /*ARGSUSED*/ 1711 static int 1712 filt_sowrite(struct knote *kn, long hint) 1713 { 1714 struct socket *so; 1715 1716 so = (struct socket *)kn->kn_fp->f_data; 1717 kn->kn_data = sbspace(&so->so_snd); 1718 if (so->so_state & SS_CANTSENDMORE) { 1719 kn->kn_flags |= EV_EOF; 1720 kn->kn_fflags = so->so_error; 1721 return (1); 1722 } 1723 if (so->so_error) /* temporary udp error */ 1724 return (1); 1725 if (((so->so_state & SS_ISCONNECTED) == 0) && 1726 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1727 return (0); 1728 if (kn->kn_sfflags & NOTE_LOWAT) 1729 return (kn->kn_data >= kn->kn_sdata); 1730 return (kn->kn_data >= so->so_snd.sb_lowat); 1731 } 1732 1733 /*ARGSUSED*/ 1734 static int 1735 filt_solisten(struct knote *kn, long hint) 1736 { 1737 struct socket *so; 1738 1739 so = (struct socket *)kn->kn_fp->f_data; 1740 1741 /* 1742 * Set kn_data to number of incoming connections, not 1743 * counting partial (incomplete) connections. 1744 */ 1745 kn->kn_data = so->so_qlen; 1746 return (kn->kn_data > 0); 1747 } 1748 1749 static const struct filterops solisten_filtops = 1750 { 1, NULL, filt_sordetach, filt_solisten }; 1751 static const struct filterops soread_filtops = 1752 { 1, NULL, filt_sordetach, filt_soread }; 1753 static const struct filterops sowrite_filtops = 1754 { 1, NULL, filt_sowdetach, filt_sowrite }; 1755 1756 int 1757 soo_kqfilter(struct file *fp, struct knote *kn) 1758 { 1759 struct socket *so; 1760 struct sockbuf *sb; 1761 1762 so = (struct socket *)kn->kn_fp->f_data; 1763 switch (kn->kn_filter) { 1764 case EVFILT_READ: 1765 if (so->so_options & SO_ACCEPTCONN) 1766 kn->kn_fop = &solisten_filtops; 1767 else 1768 kn->kn_fop = &soread_filtops; 1769 sb = &so->so_rcv; 1770 break; 1771 case EVFILT_WRITE: 1772 kn->kn_fop = &sowrite_filtops; 1773 sb = &so->so_snd; 1774 break; 1775 default: 1776 return (1); 1777 } 1778 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 1779 sb->sb_flags |= SB_KNOTE; 1780 return (0); 1781 } 1782 1783 #include <sys/sysctl.h> 1784 1785 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 1786 1787 /* 1788 * sysctl helper routine for kern.somaxkva. ensures that the given 1789 * value is not too small. 1790 * (XXX should we maybe make sure it's not too large as well?) 1791 */ 1792 static int 1793 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 1794 { 1795 int error, new_somaxkva; 1796 struct sysctlnode node; 1797 int s; 1798 1799 new_somaxkva = somaxkva; 1800 node = *rnode; 1801 node.sysctl_data = &new_somaxkva; 1802 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1803 if (error || newp == NULL) 1804 return (error); 1805 1806 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 1807 return (EINVAL); 1808 1809 s = splvm(); 1810 simple_lock(&so_pendfree_slock); 1811 somaxkva = new_somaxkva; 1812 wakeup(&socurkva); 1813 simple_unlock(&so_pendfree_slock); 1814 splx(s); 1815 1816 return (error); 1817 } 1818 1819 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup") 1820 { 1821 1822 sysctl_createv(clog, 0, NULL, NULL, 1823 CTLFLAG_PERMANENT, 1824 CTLTYPE_NODE, "kern", NULL, 1825 NULL, 0, NULL, 0, 1826 CTL_KERN, CTL_EOL); 1827 1828 sysctl_createv(clog, 0, NULL, NULL, 1829 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1830 CTLTYPE_INT, "somaxkva", 1831 SYSCTL_DESCR("Maximum amount of kernel memory to be " 1832 "used for socket buffers"), 1833 sysctl_kern_somaxkva, 0, NULL, 0, 1834 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 1835 } 1836
Indexes created Sat Sep 20 22:09:52 GMT 2025