uipc_socket.c revision 1.111.2.6
1 /* $NetBSD: uipc_socket.c,v 1.111.2.6 2007/02/26 09:11:20 yamt 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 1.111.2.6 2007/02/26 09:11:20 yamt 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, caddr_t buf, size_t size, bool mapped) 259 { 260 vaddr_t sva, eva; 261 vsize_t len; 262 int npgs; 263 264 KASSERT(pgs != NULL); 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 (mapped) { 272 pmap_kremove(sva, len); 273 pmap_update(pmap_kernel()); 274 } 275 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 276 sokvafree(sva, len); 277 } 278 279 static size_t 280 sodopendfree() 281 { 282 int s; 283 size_t rv; 284 285 s = splvm(); 286 simple_lock(&so_pendfree_slock); 287 rv = sodopendfreel(); 288 simple_unlock(&so_pendfree_slock); 289 splx(s); 290 291 return rv; 292 } 293 294 /* 295 * sodopendfreel: free mbufs on "pendfree" list. 296 * unlock and relock so_pendfree_slock when freeing mbufs. 297 * 298 * => called with so_pendfree_slock held. 299 * => called at splvm. 300 */ 301 302 static size_t 303 sodopendfreel() 304 { 305 size_t rv = 0; 306 307 LOCK_ASSERT(simple_lock_held(&so_pendfree_slock)); 308 309 for (;;) { 310 struct mbuf *m; 311 struct mbuf *next; 312 313 m = so_pendfree; 314 if (m == NULL) 315 break; 316 so_pendfree = NULL; 317 simple_unlock(&so_pendfree_slock); 318 /* XXX splx */ 319 320 for (; m != NULL; m = next) { 321 next = m->m_next; 322 KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0); 323 KASSERT(m->m_ext.ext_refcnt == 0); 324 325 rv += m->m_ext.ext_size; 326 sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf, 327 m->m_ext.ext_size, 328 (m->m_ext.ext_flags & M_EXT_LAZY) == 0); 329 pool_cache_put(&mbpool_cache, m); 330 } 331 332 /* XXX splvm */ 333 simple_lock(&so_pendfree_slock); 334 } 335 336 return (rv); 337 } 338 339 void 340 soloanfree(struct mbuf *m, caddr_t buf, size_t size, void *arg) 341 { 342 int s; 343 344 KASSERT(m != NULL); 345 346 /* 347 * postpone freeing mbuf. 348 * 349 * we can't do it in interrupt context 350 * because we need to put kva back to kernel_map. 351 */ 352 353 s = splvm(); 354 simple_lock(&so_pendfree_slock); 355 m->m_next = so_pendfree; 356 so_pendfree = m; 357 if (sokvawaiters) 358 wakeup(&socurkva); 359 simple_unlock(&so_pendfree_slock); 360 splx(s); 361 } 362 363 static long 364 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 365 { 366 struct iovec *iov = uio->uio_iov; 367 vaddr_t sva, eva; 368 vsize_t len; 369 vaddr_t lva; 370 int npgs, error; 371 #if !defined(__HAVE_LAZY_MBUF) 372 vaddr_t va; 373 int i; 374 #endif /* !defined(__HAVE_LAZY_MBUF) */ 375 376 if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace)) 377 return (0); 378 379 if (iov->iov_len < (size_t) space) 380 space = iov->iov_len; 381 if (space > SOCK_LOAN_CHUNK) 382 space = SOCK_LOAN_CHUNK; 383 384 eva = round_page((vaddr_t) iov->iov_base + space); 385 sva = trunc_page((vaddr_t) iov->iov_base); 386 len = eva - sva; 387 npgs = len >> PAGE_SHIFT; 388 389 /* XXX KDASSERT */ 390 KASSERT(npgs <= M_EXT_MAXPAGES); 391 392 lva = sokvaalloc(len, so); 393 if (lva == 0) 394 return 0; 395 396 error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len, 397 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 398 if (error) { 399 sokvafree(lva, len); 400 return (0); 401 } 402 403 #if !defined(__HAVE_LAZY_MBUF) 404 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 405 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 406 VM_PROT_READ); 407 pmap_update(pmap_kernel()); 408 #endif /* !defined(__HAVE_LAZY_MBUF) */ 409 410 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 411 412 MEXTADD(m, (caddr_t) lva, space, M_MBUF, soloanfree, so); 413 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 414 415 #if defined(__HAVE_LAZY_MBUF) 416 m->m_flags |= M_EXT_LAZY; 417 m->m_ext.ext_flags |= M_EXT_LAZY; 418 #endif /* defined(__HAVE_LAZY_MBUF) */ 419 420 uio->uio_resid -= space; 421 /* uio_offset not updated, not set/used for write(2) */ 422 uio->uio_iov->iov_base = (caddr_t) uio->uio_iov->iov_base + space; 423 uio->uio_iov->iov_len -= space; 424 if (uio->uio_iov->iov_len == 0) { 425 uio->uio_iov++; 426 uio->uio_iovcnt--; 427 } 428 429 return (space); 430 } 431 432 static int 433 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg) 434 { 435 436 KASSERT(ce == &sokva_reclaimerentry); 437 KASSERT(obj == NULL); 438 439 sodopendfree(); 440 if (!vm_map_starved_p(kernel_map)) { 441 return CALLBACK_CHAIN_ABORT; 442 } 443 return CALLBACK_CHAIN_CONTINUE; 444 } 445 446 void 447 soinit(void) 448 { 449 450 /* Set the initial adjusted socket buffer size. */ 451 if (sb_max_set(sb_max)) 452 panic("bad initial sb_max value: %lu", sb_max); 453 454 callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback, 455 &sokva_reclaimerentry, NULL, sokva_reclaim_callback); 456 } 457 458 /* 459 * Socket operation routines. 460 * These routines are called by the routines in 461 * sys_socket.c or from a system process, and 462 * implement the semantics of socket operations by 463 * switching out to the protocol specific routines. 464 */ 465 /*ARGSUSED*/ 466 int 467 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l) 468 { 469 const struct protosw *prp; 470 struct socket *so; 471 uid_t uid; 472 int error, s; 473 474 error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET, 475 KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type), 476 KAUTH_ARG(proto)); 477 if (error) 478 return (error); 479 480 if (proto) 481 prp = pffindproto(dom, proto, type); 482 else 483 prp = pffindtype(dom, type); 484 if (prp == 0) { 485 /* no support for domain */ 486 if (pffinddomain(dom) == 0) 487 return (EAFNOSUPPORT); 488 /* no support for socket type */ 489 if (proto == 0 && type != 0) 490 return (EPROTOTYPE); 491 return (EPROTONOSUPPORT); 492 } 493 if (prp->pr_usrreq == 0) 494 return (EPROTONOSUPPORT); 495 if (prp->pr_type != type) 496 return (EPROTOTYPE); 497 s = splsoftnet(); 498 so = pool_get(&socket_pool, PR_WAITOK); 499 memset((caddr_t)so, 0, sizeof(*so)); 500 TAILQ_INIT(&so->so_q0); 501 TAILQ_INIT(&so->so_q); 502 so->so_type = type; 503 so->so_proto = prp; 504 so->so_send = sosend; 505 so->so_receive = soreceive; 506 #ifdef MBUFTRACE 507 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 508 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 509 so->so_mowner = &prp->pr_domain->dom_mowner; 510 #endif 511 if (l != NULL) { 512 uid = kauth_cred_geteuid(l->l_cred); 513 } else { 514 uid = 0; 515 } 516 so->so_uidinfo = uid_find(uid); 517 error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0, 518 (struct mbuf *)(long)proto, (struct mbuf *)0, l); 519 if (error) { 520 so->so_state |= SS_NOFDREF; 521 sofree(so); 522 splx(s); 523 return (error); 524 } 525 splx(s); 526 *aso = so; 527 return (0); 528 } 529 530 int 531 sobind(struct socket *so, struct mbuf *nam, struct lwp *l) 532 { 533 int s, error; 534 535 s = splsoftnet(); 536 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0, 537 nam, (struct mbuf *)0, l); 538 splx(s); 539 return (error); 540 } 541 542 int 543 solisten(struct socket *so, int backlog) 544 { 545 int s, error; 546 547 s = splsoftnet(); 548 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0, 549 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 550 if (error) { 551 splx(s); 552 return (error); 553 } 554 if (TAILQ_EMPTY(&so->so_q)) 555 so->so_options |= SO_ACCEPTCONN; 556 if (backlog < 0) 557 backlog = 0; 558 so->so_qlimit = min(backlog, somaxconn); 559 splx(s); 560 return (0); 561 } 562 563 void 564 sofree(struct socket *so) 565 { 566 567 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 568 return; 569 if (so->so_head) { 570 /* 571 * We must not decommission a socket that's on the accept(2) 572 * queue. If we do, then accept(2) may hang after select(2) 573 * indicated that the listening socket was ready. 574 */ 575 if (!soqremque(so, 0)) 576 return; 577 } 578 if (so->so_rcv.sb_hiwat) 579 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 580 RLIM_INFINITY); 581 if (so->so_snd.sb_hiwat) 582 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 583 RLIM_INFINITY); 584 sbrelease(&so->so_snd, so); 585 sorflush(so); 586 pool_put(&socket_pool, so); 587 } 588 589 /* 590 * Close a socket on last file table reference removal. 591 * Initiate disconnect if connected. 592 * Free socket when disconnect complete. 593 */ 594 int 595 soclose(struct socket *so) 596 { 597 struct socket *so2; 598 int s, error; 599 600 error = 0; 601 s = splsoftnet(); /* conservative */ 602 if (so->so_options & SO_ACCEPTCONN) { 603 while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 604 (void) soqremque(so2, 0); 605 (void) soabort(so2); 606 } 607 while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 608 (void) soqremque(so2, 1); 609 (void) soabort(so2); 610 } 611 } 612 if (so->so_pcb == 0) 613 goto discard; 614 if (so->so_state & SS_ISCONNECTED) { 615 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 616 error = sodisconnect(so); 617 if (error) 618 goto drop; 619 } 620 if (so->so_options & SO_LINGER) { 621 if ((so->so_state & SS_ISDISCONNECTING) && 622 (so->so_state & SS_NBIO)) 623 goto drop; 624 while (so->so_state & SS_ISCONNECTED) { 625 error = tsleep((caddr_t)&so->so_timeo, 626 PSOCK | PCATCH, netcls, 627 so->so_linger * hz); 628 if (error) 629 break; 630 } 631 } 632 } 633 drop: 634 if (so->so_pcb) { 635 int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 636 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 637 (struct lwp *)0); 638 if (error == 0) 639 error = error2; 640 } 641 discard: 642 if (so->so_state & SS_NOFDREF) 643 panic("soclose: NOFDREF"); 644 so->so_state |= SS_NOFDREF; 645 sofree(so); 646 splx(s); 647 return (error); 648 } 649 650 /* 651 * Must be called at splsoftnet... 652 */ 653 int 654 soabort(struct socket *so) 655 { 656 657 return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0, 658 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 659 } 660 661 int 662 soaccept(struct socket *so, struct mbuf *nam) 663 { 664 int s, error; 665 666 error = 0; 667 s = splsoftnet(); 668 if ((so->so_state & SS_NOFDREF) == 0) 669 panic("soaccept: !NOFDREF"); 670 so->so_state &= ~SS_NOFDREF; 671 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 672 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 673 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 674 (struct mbuf *)0, nam, (struct mbuf *)0, (struct lwp *)0); 675 else 676 error = ECONNABORTED; 677 678 splx(s); 679 return (error); 680 } 681 682 int 683 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l) 684 { 685 int s, error; 686 687 if (so->so_options & SO_ACCEPTCONN) 688 return (EOPNOTSUPP); 689 s = splsoftnet(); 690 /* 691 * If protocol is connection-based, can only connect once. 692 * Otherwise, if connected, try to disconnect first. 693 * This allows user to disconnect by connecting to, e.g., 694 * a null address. 695 */ 696 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 697 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 698 (error = sodisconnect(so)))) 699 error = EISCONN; 700 else 701 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 702 (struct mbuf *)0, nam, (struct mbuf *)0, l); 703 splx(s); 704 return (error); 705 } 706 707 int 708 soconnect2(struct socket *so1, struct socket *so2) 709 { 710 int s, error; 711 712 s = splsoftnet(); 713 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 714 (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0, 715 (struct lwp *)0); 716 splx(s); 717 return (error); 718 } 719 720 int 721 sodisconnect(struct socket *so) 722 { 723 int s, error; 724 725 s = splsoftnet(); 726 if ((so->so_state & SS_ISCONNECTED) == 0) { 727 error = ENOTCONN; 728 goto bad; 729 } 730 if (so->so_state & SS_ISDISCONNECTING) { 731 error = EALREADY; 732 goto bad; 733 } 734 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 735 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 736 (struct lwp *)0); 737 bad: 738 splx(s); 739 sodopendfree(); 740 return (error); 741 } 742 743 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 744 /* 745 * Send on a socket. 746 * If send must go all at once and message is larger than 747 * send buffering, then hard error. 748 * Lock against other senders. 749 * If must go all at once and not enough room now, then 750 * inform user that this would block and do nothing. 751 * Otherwise, if nonblocking, send as much as possible. 752 * The data to be sent is described by "uio" if nonzero, 753 * otherwise by the mbuf chain "top" (which must be null 754 * if uio is not). Data provided in mbuf chain must be small 755 * enough to send all at once. 756 * 757 * Returns nonzero on error, timeout or signal; callers 758 * must check for short counts if EINTR/ERESTART are returned. 759 * Data and control buffers are freed on return. 760 */ 761 int 762 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 763 struct mbuf *control, int flags, struct lwp *l) 764 { 765 struct mbuf **mp, *m; 766 struct proc *p; 767 long space, len, resid, clen, mlen; 768 int error, s, dontroute, atomic; 769 770 p = l->l_proc; 771 sodopendfree(); 772 773 clen = 0; 774 atomic = sosendallatonce(so) || top; 775 if (uio) 776 resid = uio->uio_resid; 777 else 778 resid = top->m_pkthdr.len; 779 /* 780 * In theory resid should be unsigned. 781 * However, space must be signed, as it might be less than 0 782 * if we over-committed, and we must use a signed comparison 783 * of space and resid. On the other hand, a negative resid 784 * causes us to loop sending 0-length segments to the protocol. 785 */ 786 if (resid < 0) { 787 error = EINVAL; 788 goto out; 789 } 790 dontroute = 791 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 792 (so->so_proto->pr_flags & PR_ATOMIC); 793 if (p) 794 p->p_stats->p_ru.ru_msgsnd++; 795 if (control) 796 clen = control->m_len; 797 #define snderr(errno) { error = errno; splx(s); goto release; } 798 799 restart: 800 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 801 goto out; 802 do { 803 s = splsoftnet(); 804 if (so->so_state & SS_CANTSENDMORE) 805 snderr(EPIPE); 806 if (so->so_error) { 807 error = so->so_error; 808 so->so_error = 0; 809 splx(s); 810 goto release; 811 } 812 if ((so->so_state & SS_ISCONNECTED) == 0) { 813 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 814 if ((so->so_state & SS_ISCONFIRMING) == 0 && 815 !(resid == 0 && clen != 0)) 816 snderr(ENOTCONN); 817 } else if (addr == 0) 818 snderr(EDESTADDRREQ); 819 } 820 space = sbspace(&so->so_snd); 821 if (flags & MSG_OOB) 822 space += 1024; 823 if ((atomic && resid > so->so_snd.sb_hiwat) || 824 clen > so->so_snd.sb_hiwat) 825 snderr(EMSGSIZE); 826 if (space < resid + clen && 827 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 828 if (so->so_state & SS_NBIO) 829 snderr(EWOULDBLOCK); 830 sbunlock(&so->so_snd); 831 error = sbwait(&so->so_snd); 832 splx(s); 833 if (error) 834 goto out; 835 goto restart; 836 } 837 splx(s); 838 mp = ⊤ 839 space -= clen; 840 do { 841 if (uio == NULL) { 842 /* 843 * Data is prepackaged in "top". 844 */ 845 resid = 0; 846 if (flags & MSG_EOR) 847 top->m_flags |= M_EOR; 848 } else do { 849 if (top == 0) { 850 m = m_gethdr(M_WAIT, MT_DATA); 851 mlen = MHLEN; 852 m->m_pkthdr.len = 0; 853 m->m_pkthdr.rcvif = (struct ifnet *)0; 854 } else { 855 m = m_get(M_WAIT, MT_DATA); 856 mlen = MLEN; 857 } 858 MCLAIM(m, so->so_snd.sb_mowner); 859 if (sock_loan_thresh >= 0 && 860 uio->uio_iov->iov_len >= sock_loan_thresh && 861 space >= sock_loan_thresh && 862 (len = sosend_loan(so, uio, m, 863 space)) != 0) { 864 SOSEND_COUNTER_INCR(&sosend_loan_big); 865 space -= len; 866 goto have_data; 867 } 868 if (resid >= MINCLSIZE && space >= MCLBYTES) { 869 SOSEND_COUNTER_INCR(&sosend_copy_big); 870 m_clget(m, M_WAIT); 871 if ((m->m_flags & M_EXT) == 0) 872 goto nopages; 873 mlen = MCLBYTES; 874 if (atomic && top == 0) { 875 len = lmin(MCLBYTES - max_hdr, 876 resid); 877 m->m_data += max_hdr; 878 } else 879 len = lmin(MCLBYTES, resid); 880 space -= len; 881 } else { 882 nopages: 883 SOSEND_COUNTER_INCR(&sosend_copy_small); 884 len = lmin(lmin(mlen, resid), space); 885 space -= len; 886 /* 887 * For datagram protocols, leave room 888 * for protocol headers in first mbuf. 889 */ 890 if (atomic && top == 0 && len < mlen) 891 MH_ALIGN(m, len); 892 } 893 error = uiomove(mtod(m, caddr_t), (int)len, 894 uio); 895 have_data: 896 resid = uio->uio_resid; 897 m->m_len = len; 898 *mp = m; 899 top->m_pkthdr.len += len; 900 if (error) 901 goto release; 902 mp = &m->m_next; 903 if (resid <= 0) { 904 if (flags & MSG_EOR) 905 top->m_flags |= M_EOR; 906 break; 907 } 908 } while (space > 0 && atomic); 909 910 s = splsoftnet(); 911 912 if (so->so_state & SS_CANTSENDMORE) 913 snderr(EPIPE); 914 915 if (dontroute) 916 so->so_options |= SO_DONTROUTE; 917 if (resid > 0) 918 so->so_state |= SS_MORETOCOME; 919 error = (*so->so_proto->pr_usrreq)(so, 920 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 921 top, addr, control, curlwp); /* XXX */ 922 if (dontroute) 923 so->so_options &= ~SO_DONTROUTE; 924 if (resid > 0) 925 so->so_state &= ~SS_MORETOCOME; 926 splx(s); 927 928 clen = 0; 929 control = 0; 930 top = 0; 931 mp = ⊤ 932 if (error) 933 goto release; 934 } while (resid && space > 0); 935 } while (resid); 936 937 release: 938 sbunlock(&so->so_snd); 939 out: 940 if (top) 941 m_freem(top); 942 if (control) 943 m_freem(control); 944 return (error); 945 } 946 947 /* 948 * Implement receive operations on a socket. 949 * We depend on the way that records are added to the sockbuf 950 * by sbappend*. In particular, each record (mbufs linked through m_next) 951 * must begin with an address if the protocol so specifies, 952 * followed by an optional mbuf or mbufs containing ancillary data, 953 * and then zero or more mbufs of data. 954 * In order to avoid blocking network interrupts for the entire time here, 955 * we splx() while doing the actual copy to user space. 956 * Although the sockbuf is locked, new data may still be appended, 957 * and thus we must maintain consistency of the sockbuf during that time. 958 * 959 * The caller may receive the data as a single mbuf chain by supplying 960 * an mbuf **mp0 for use in returning the chain. The uio is then used 961 * only for the count in uio_resid. 962 */ 963 int 964 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 965 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 966 { 967 struct lwp *l = curlwp; 968 struct mbuf *m, **mp; 969 int flags, len, error, s, offset, moff, type, orig_resid; 970 const struct protosw *pr; 971 struct mbuf *nextrecord; 972 int mbuf_removed = 0; 973 974 pr = so->so_proto; 975 mp = mp0; 976 type = 0; 977 orig_resid = uio->uio_resid; 978 979 if (paddr) 980 *paddr = 0; 981 if (controlp) 982 *controlp = 0; 983 if (flagsp) 984 flags = *flagsp &~ MSG_EOR; 985 else 986 flags = 0; 987 988 if ((flags & MSG_DONTWAIT) == 0) 989 sodopendfree(); 990 991 if (flags & MSG_OOB) { 992 m = m_get(M_WAIT, MT_DATA); 993 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 994 (struct mbuf *)(long)(flags & MSG_PEEK), 995 (struct mbuf *)0, l); 996 if (error) 997 goto bad; 998 do { 999 error = uiomove(mtod(m, caddr_t), 1000 (int) min(uio->uio_resid, m->m_len), uio); 1001 m = m_free(m); 1002 } while (uio->uio_resid && error == 0 && m); 1003 bad: 1004 if (m) 1005 m_freem(m); 1006 return (error); 1007 } 1008 if (mp) 1009 *mp = (struct mbuf *)0; 1010 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 1011 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1012 (struct mbuf *)0, (struct mbuf *)0, l); 1013 1014 restart: 1015 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) 1016 return (error); 1017 s = splsoftnet(); 1018 1019 m = so->so_rcv.sb_mb; 1020 /* 1021 * If we have less data than requested, block awaiting more 1022 * (subject to any timeout) if: 1023 * 1. the current count is less than the low water mark, 1024 * 2. MSG_WAITALL is set, and it is possible to do the entire 1025 * receive operation at once if we block (resid <= hiwat), or 1026 * 3. MSG_DONTWAIT is not set. 1027 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1028 * we have to do the receive in sections, and thus risk returning 1029 * a short count if a timeout or signal occurs after we start. 1030 */ 1031 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 1032 so->so_rcv.sb_cc < uio->uio_resid) && 1033 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1034 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1035 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1036 #ifdef DIAGNOSTIC 1037 if (m == 0 && so->so_rcv.sb_cc) 1038 panic("receive 1"); 1039 #endif 1040 if (so->so_error) { 1041 if (m) 1042 goto dontblock; 1043 error = so->so_error; 1044 if ((flags & MSG_PEEK) == 0) 1045 so->so_error = 0; 1046 goto release; 1047 } 1048 if (so->so_state & SS_CANTRCVMORE) { 1049 if (m) 1050 goto dontblock; 1051 else 1052 goto release; 1053 } 1054 for (; m; m = m->m_next) 1055 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1056 m = so->so_rcv.sb_mb; 1057 goto dontblock; 1058 } 1059 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1060 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1061 error = ENOTCONN; 1062 goto release; 1063 } 1064 if (uio->uio_resid == 0) 1065 goto release; 1066 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 1067 error = EWOULDBLOCK; 1068 goto release; 1069 } 1070 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1071 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1072 sbunlock(&so->so_rcv); 1073 error = sbwait(&so->so_rcv); 1074 splx(s); 1075 if (error) 1076 return (error); 1077 goto restart; 1078 } 1079 dontblock: 1080 /* 1081 * On entry here, m points to the first record of the socket buffer. 1082 * While we process the initial mbufs containing address and control 1083 * info, we save a copy of m->m_nextpkt into nextrecord. 1084 */ 1085 if (l) 1086 l->l_proc->p_stats->p_ru.ru_msgrcv++; 1087 KASSERT(m == so->so_rcv.sb_mb); 1088 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1089 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1090 nextrecord = m->m_nextpkt; 1091 if (pr->pr_flags & PR_ADDR) { 1092 #ifdef DIAGNOSTIC 1093 if (m->m_type != MT_SONAME) 1094 panic("receive 1a"); 1095 #endif 1096 orig_resid = 0; 1097 if (flags & MSG_PEEK) { 1098 if (paddr) 1099 *paddr = m_copy(m, 0, m->m_len); 1100 m = m->m_next; 1101 } else { 1102 sbfree(&so->so_rcv, m); 1103 mbuf_removed = 1; 1104 if (paddr) { 1105 *paddr = m; 1106 so->so_rcv.sb_mb = m->m_next; 1107 m->m_next = 0; 1108 m = so->so_rcv.sb_mb; 1109 } else { 1110 MFREE(m, so->so_rcv.sb_mb); 1111 m = so->so_rcv.sb_mb; 1112 } 1113 } 1114 } 1115 while (m && m->m_type == MT_CONTROL && error == 0) { 1116 if (flags & MSG_PEEK) { 1117 if (controlp) 1118 *controlp = m_copy(m, 0, m->m_len); 1119 m = m->m_next; 1120 } else { 1121 sbfree(&so->so_rcv, m); 1122 mbuf_removed = 1; 1123 if (controlp) { 1124 struct domain *dom = pr->pr_domain; 1125 if (dom->dom_externalize && l && 1126 mtod(m, struct cmsghdr *)->cmsg_type == 1127 SCM_RIGHTS) 1128 error = (*dom->dom_externalize)(m, l); 1129 *controlp = m; 1130 so->so_rcv.sb_mb = m->m_next; 1131 m->m_next = 0; 1132 m = so->so_rcv.sb_mb; 1133 } else { 1134 /* 1135 * Dispose of any SCM_RIGHTS message that went 1136 * through the read path rather than recv. 1137 */ 1138 if (pr->pr_domain->dom_dispose && 1139 mtod(m, struct cmsghdr *)->cmsg_type == SCM_RIGHTS) 1140 (*pr->pr_domain->dom_dispose)(m); 1141 MFREE(m, so->so_rcv.sb_mb); 1142 m = so->so_rcv.sb_mb; 1143 } 1144 } 1145 if (controlp) { 1146 orig_resid = 0; 1147 controlp = &(*controlp)->m_next; 1148 } 1149 } 1150 1151 /* 1152 * If m is non-NULL, we have some data to read. From now on, 1153 * make sure to keep sb_lastrecord consistent when working on 1154 * the last packet on the chain (nextrecord == NULL) and we 1155 * change m->m_nextpkt. 1156 */ 1157 if (m) { 1158 if ((flags & MSG_PEEK) == 0) { 1159 m->m_nextpkt = nextrecord; 1160 /* 1161 * If nextrecord == NULL (this is a single chain), 1162 * then sb_lastrecord may not be valid here if m 1163 * was changed earlier. 1164 */ 1165 if (nextrecord == NULL) { 1166 KASSERT(so->so_rcv.sb_mb == m); 1167 so->so_rcv.sb_lastrecord = m; 1168 } 1169 } 1170 type = m->m_type; 1171 if (type == MT_OOBDATA) 1172 flags |= MSG_OOB; 1173 } else { 1174 if ((flags & MSG_PEEK) == 0) { 1175 KASSERT(so->so_rcv.sb_mb == m); 1176 so->so_rcv.sb_mb = nextrecord; 1177 SB_EMPTY_FIXUP(&so->so_rcv); 1178 } 1179 } 1180 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1181 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1182 1183 moff = 0; 1184 offset = 0; 1185 while (m && uio->uio_resid > 0 && error == 0) { 1186 if (m->m_type == MT_OOBDATA) { 1187 if (type != MT_OOBDATA) 1188 break; 1189 } else if (type == MT_OOBDATA) 1190 break; 1191 #ifdef DIAGNOSTIC 1192 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1193 panic("receive 3"); 1194 #endif 1195 so->so_state &= ~SS_RCVATMARK; 1196 len = uio->uio_resid; 1197 if (so->so_oobmark && len > so->so_oobmark - offset) 1198 len = so->so_oobmark - offset; 1199 if (len > m->m_len - moff) 1200 len = m->m_len - moff; 1201 /* 1202 * If mp is set, just pass back the mbufs. 1203 * Otherwise copy them out via the uio, then free. 1204 * Sockbuf must be consistent here (points to current mbuf, 1205 * it points to next record) when we drop priority; 1206 * we must note any additions to the sockbuf when we 1207 * block interrupts again. 1208 */ 1209 if (mp == 0) { 1210 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1211 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1212 splx(s); 1213 error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio); 1214 s = splsoftnet(); 1215 if (error) { 1216 /* 1217 * If any part of the record has been removed 1218 * (such as the MT_SONAME mbuf, which will 1219 * happen when PR_ADDR, and thus also 1220 * PR_ATOMIC, is set), then drop the entire 1221 * record to maintain the atomicity of the 1222 * receive operation. 1223 * 1224 * This avoids a later panic("receive 1a") 1225 * when compiled with DIAGNOSTIC. 1226 */ 1227 if (m && mbuf_removed 1228 && (pr->pr_flags & PR_ATOMIC)) 1229 (void) sbdroprecord(&so->so_rcv); 1230 1231 goto release; 1232 } 1233 } else 1234 uio->uio_resid -= len; 1235 if (len == m->m_len - moff) { 1236 if (m->m_flags & M_EOR) 1237 flags |= MSG_EOR; 1238 if (flags & MSG_PEEK) { 1239 m = m->m_next; 1240 moff = 0; 1241 } else { 1242 nextrecord = m->m_nextpkt; 1243 sbfree(&so->so_rcv, m); 1244 if (mp) { 1245 *mp = m; 1246 mp = &m->m_next; 1247 so->so_rcv.sb_mb = m = m->m_next; 1248 *mp = (struct mbuf *)0; 1249 } else { 1250 MFREE(m, so->so_rcv.sb_mb); 1251 m = so->so_rcv.sb_mb; 1252 } 1253 /* 1254 * If m != NULL, we also know that 1255 * so->so_rcv.sb_mb != NULL. 1256 */ 1257 KASSERT(so->so_rcv.sb_mb == m); 1258 if (m) { 1259 m->m_nextpkt = nextrecord; 1260 if (nextrecord == NULL) 1261 so->so_rcv.sb_lastrecord = m; 1262 } else { 1263 so->so_rcv.sb_mb = nextrecord; 1264 SB_EMPTY_FIXUP(&so->so_rcv); 1265 } 1266 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1267 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1268 } 1269 } else { 1270 if (flags & MSG_PEEK) 1271 moff += len; 1272 else { 1273 if (mp) 1274 *mp = m_copym(m, 0, len, M_WAIT); 1275 m->m_data += len; 1276 m->m_len -= len; 1277 so->so_rcv.sb_cc -= len; 1278 } 1279 } 1280 if (so->so_oobmark) { 1281 if ((flags & MSG_PEEK) == 0) { 1282 so->so_oobmark -= len; 1283 if (so->so_oobmark == 0) { 1284 so->so_state |= SS_RCVATMARK; 1285 break; 1286 } 1287 } else { 1288 offset += len; 1289 if (offset == so->so_oobmark) 1290 break; 1291 } 1292 } 1293 if (flags & MSG_EOR) 1294 break; 1295 /* 1296 * If the MSG_WAITALL flag is set (for non-atomic socket), 1297 * we must not quit until "uio->uio_resid == 0" or an error 1298 * termination. If a signal/timeout occurs, return 1299 * with a short count but without error. 1300 * Keep sockbuf locked against other readers. 1301 */ 1302 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1303 !sosendallatonce(so) && !nextrecord) { 1304 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1305 break; 1306 /* 1307 * If we are peeking and the socket receive buffer is 1308 * full, stop since we can't get more data to peek at. 1309 */ 1310 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1311 break; 1312 /* 1313 * If we've drained the socket buffer, tell the 1314 * protocol in case it needs to do something to 1315 * get it filled again. 1316 */ 1317 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1318 (*pr->pr_usrreq)(so, PRU_RCVD, 1319 (struct mbuf *)0, 1320 (struct mbuf *)(long)flags, 1321 (struct mbuf *)0, l); 1322 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1323 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1324 error = sbwait(&so->so_rcv); 1325 if (error) { 1326 sbunlock(&so->so_rcv); 1327 splx(s); 1328 return (0); 1329 } 1330 if ((m = so->so_rcv.sb_mb) != NULL) 1331 nextrecord = m->m_nextpkt; 1332 } 1333 } 1334 1335 if (m && pr->pr_flags & PR_ATOMIC) { 1336 flags |= MSG_TRUNC; 1337 if ((flags & MSG_PEEK) == 0) 1338 (void) sbdroprecord(&so->so_rcv); 1339 } 1340 if ((flags & MSG_PEEK) == 0) { 1341 if (m == 0) { 1342 /* 1343 * First part is an inline SB_EMPTY_FIXUP(). Second 1344 * part makes sure sb_lastrecord is up-to-date if 1345 * there is still data in the socket buffer. 1346 */ 1347 so->so_rcv.sb_mb = nextrecord; 1348 if (so->so_rcv.sb_mb == NULL) { 1349 so->so_rcv.sb_mbtail = NULL; 1350 so->so_rcv.sb_lastrecord = NULL; 1351 } else if (nextrecord->m_nextpkt == NULL) 1352 so->so_rcv.sb_lastrecord = nextrecord; 1353 } 1354 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1355 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1356 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1357 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1358 (struct mbuf *)(long)flags, (struct mbuf *)0, l); 1359 } 1360 if (orig_resid == uio->uio_resid && orig_resid && 1361 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1362 sbunlock(&so->so_rcv); 1363 splx(s); 1364 goto restart; 1365 } 1366 1367 if (flagsp) 1368 *flagsp |= flags; 1369 release: 1370 sbunlock(&so->so_rcv); 1371 splx(s); 1372 return (error); 1373 } 1374 1375 int 1376 soshutdown(struct socket *so, int how) 1377 { 1378 const struct protosw *pr; 1379 1380 pr = so->so_proto; 1381 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1382 return (EINVAL); 1383 1384 if (how == SHUT_RD || how == SHUT_RDWR) 1385 sorflush(so); 1386 if (how == SHUT_WR || how == SHUT_RDWR) 1387 return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0, 1388 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1389 return (0); 1390 } 1391 1392 void 1393 sorflush(struct socket *so) 1394 { 1395 struct sockbuf *sb, asb; 1396 const struct protosw *pr; 1397 int s; 1398 1399 sb = &so->so_rcv; 1400 pr = so->so_proto; 1401 sb->sb_flags |= SB_NOINTR; 1402 (void) sblock(sb, M_WAITOK); 1403 s = splnet(); 1404 socantrcvmore(so); 1405 sbunlock(sb); 1406 asb = *sb; 1407 /* 1408 * Clear most of the sockbuf structure, but leave some of the 1409 * fields valid. 1410 */ 1411 memset(&sb->sb_startzero, 0, 1412 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1413 splx(s); 1414 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1415 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1416 sbrelease(&asb, so); 1417 } 1418 1419 int 1420 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0) 1421 { 1422 int error; 1423 struct mbuf *m; 1424 struct linger *l; 1425 1426 error = 0; 1427 m = m0; 1428 if (level != SOL_SOCKET) { 1429 if (so->so_proto && so->so_proto->pr_ctloutput) 1430 return ((*so->so_proto->pr_ctloutput) 1431 (PRCO_SETOPT, so, level, optname, &m0)); 1432 error = ENOPROTOOPT; 1433 } else { 1434 switch (optname) { 1435 1436 case SO_LINGER: 1437 if (m == NULL || m->m_len != sizeof(struct linger)) { 1438 error = EINVAL; 1439 goto bad; 1440 } 1441 l = mtod(m, struct linger *); 1442 if (l->l_linger < 0 || l->l_linger > USHRT_MAX || 1443 l->l_linger > (INT_MAX / hz)) { 1444 error = EDOM; 1445 goto bad; 1446 } 1447 so->so_linger = l->l_linger; 1448 if (l->l_onoff) 1449 so->so_options |= SO_LINGER; 1450 else 1451 so->so_options &= ~SO_LINGER; 1452 break; 1453 1454 case SO_DEBUG: 1455 case SO_KEEPALIVE: 1456 case SO_DONTROUTE: 1457 case SO_USELOOPBACK: 1458 case SO_BROADCAST: 1459 case SO_REUSEADDR: 1460 case SO_REUSEPORT: 1461 case SO_OOBINLINE: 1462 case SO_TIMESTAMP: 1463 if (m == NULL || m->m_len < sizeof(int)) { 1464 error = EINVAL; 1465 goto bad; 1466 } 1467 if (*mtod(m, int *)) 1468 so->so_options |= optname; 1469 else 1470 so->so_options &= ~optname; 1471 break; 1472 1473 case SO_SNDBUF: 1474 case SO_RCVBUF: 1475 case SO_SNDLOWAT: 1476 case SO_RCVLOWAT: 1477 { 1478 int optval; 1479 1480 if (m == NULL || m->m_len < sizeof(int)) { 1481 error = EINVAL; 1482 goto bad; 1483 } 1484 1485 /* 1486 * Values < 1 make no sense for any of these 1487 * options, so disallow them. 1488 */ 1489 optval = *mtod(m, int *); 1490 if (optval < 1) { 1491 error = EINVAL; 1492 goto bad; 1493 } 1494 1495 switch (optname) { 1496 1497 case SO_SNDBUF: 1498 case SO_RCVBUF: 1499 if (sbreserve(optname == SO_SNDBUF ? 1500 &so->so_snd : &so->so_rcv, 1501 (u_long) optval, so) == 0) { 1502 error = ENOBUFS; 1503 goto bad; 1504 } 1505 break; 1506 1507 /* 1508 * Make sure the low-water is never greater than 1509 * the high-water. 1510 */ 1511 case SO_SNDLOWAT: 1512 so->so_snd.sb_lowat = 1513 (optval > so->so_snd.sb_hiwat) ? 1514 so->so_snd.sb_hiwat : optval; 1515 break; 1516 case SO_RCVLOWAT: 1517 so->so_rcv.sb_lowat = 1518 (optval > so->so_rcv.sb_hiwat) ? 1519 so->so_rcv.sb_hiwat : optval; 1520 break; 1521 } 1522 break; 1523 } 1524 1525 case SO_SNDTIMEO: 1526 case SO_RCVTIMEO: 1527 { 1528 struct timeval *tv; 1529 int val; 1530 1531 if (m == NULL || m->m_len < sizeof(*tv)) { 1532 error = EINVAL; 1533 goto bad; 1534 } 1535 tv = mtod(m, struct timeval *); 1536 if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz) { 1537 error = EDOM; 1538 goto bad; 1539 } 1540 val = tv->tv_sec * hz + tv->tv_usec / tick; 1541 if (val == 0 && tv->tv_usec != 0) 1542 val = 1; 1543 1544 switch (optname) { 1545 1546 case SO_SNDTIMEO: 1547 so->so_snd.sb_timeo = val; 1548 break; 1549 case SO_RCVTIMEO: 1550 so->so_rcv.sb_timeo = val; 1551 break; 1552 } 1553 break; 1554 } 1555 1556 default: 1557 error = ENOPROTOOPT; 1558 break; 1559 } 1560 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1561 (void) ((*so->so_proto->pr_ctloutput) 1562 (PRCO_SETOPT, so, level, optname, &m0)); 1563 m = NULL; /* freed by protocol */ 1564 } 1565 } 1566 bad: 1567 if (m) 1568 (void) m_free(m); 1569 return (error); 1570 } 1571 1572 int 1573 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp) 1574 { 1575 struct mbuf *m; 1576 1577 if (level != SOL_SOCKET) { 1578 if (so->so_proto && so->so_proto->pr_ctloutput) { 1579 return ((*so->so_proto->pr_ctloutput) 1580 (PRCO_GETOPT, so, level, optname, mp)); 1581 } else 1582 return (ENOPROTOOPT); 1583 } else { 1584 m = m_get(M_WAIT, MT_SOOPTS); 1585 m->m_len = sizeof(int); 1586 1587 switch (optname) { 1588 1589 case SO_LINGER: 1590 m->m_len = sizeof(struct linger); 1591 mtod(m, struct linger *)->l_onoff = 1592 (so->so_options & SO_LINGER) ? 1 : 0; 1593 mtod(m, struct linger *)->l_linger = so->so_linger; 1594 break; 1595 1596 case SO_USELOOPBACK: 1597 case SO_DONTROUTE: 1598 case SO_DEBUG: 1599 case SO_KEEPALIVE: 1600 case SO_REUSEADDR: 1601 case SO_REUSEPORT: 1602 case SO_BROADCAST: 1603 case SO_OOBINLINE: 1604 case SO_TIMESTAMP: 1605 *mtod(m, int *) = (so->so_options & optname) ? 1 : 0; 1606 break; 1607 1608 case SO_TYPE: 1609 *mtod(m, int *) = so->so_type; 1610 break; 1611 1612 case SO_ERROR: 1613 *mtod(m, int *) = so->so_error; 1614 so->so_error = 0; 1615 break; 1616 1617 case SO_SNDBUF: 1618 *mtod(m, int *) = so->so_snd.sb_hiwat; 1619 break; 1620 1621 case SO_RCVBUF: 1622 *mtod(m, int *) = so->so_rcv.sb_hiwat; 1623 break; 1624 1625 case SO_SNDLOWAT: 1626 *mtod(m, int *) = so->so_snd.sb_lowat; 1627 break; 1628 1629 case SO_RCVLOWAT: 1630 *mtod(m, int *) = so->so_rcv.sb_lowat; 1631 break; 1632 1633 case SO_SNDTIMEO: 1634 case SO_RCVTIMEO: 1635 { 1636 int val = (optname == SO_SNDTIMEO ? 1637 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1638 1639 m->m_len = sizeof(struct timeval); 1640 mtod(m, struct timeval *)->tv_sec = val / hz; 1641 mtod(m, struct timeval *)->tv_usec = 1642 (val % hz) * tick; 1643 break; 1644 } 1645 1646 case SO_OVERFLOWED: 1647 *mtod(m, int *) = so->so_rcv.sb_overflowed; 1648 break; 1649 1650 default: 1651 (void)m_free(m); 1652 return (ENOPROTOOPT); 1653 } 1654 *mp = m; 1655 return (0); 1656 } 1657 } 1658 1659 void 1660 sohasoutofband(struct socket *so) 1661 { 1662 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 1663 selwakeup(&so->so_rcv.sb_sel); 1664 } 1665 1666 static void 1667 filt_sordetach(struct knote *kn) 1668 { 1669 struct socket *so; 1670 1671 so = (struct socket *)kn->kn_fp->f_data; 1672 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 1673 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 1674 so->so_rcv.sb_flags &= ~SB_KNOTE; 1675 } 1676 1677 /*ARGSUSED*/ 1678 static int 1679 filt_soread(struct knote *kn, long hint) 1680 { 1681 struct socket *so; 1682 1683 so = (struct socket *)kn->kn_fp->f_data; 1684 kn->kn_data = so->so_rcv.sb_cc; 1685 if (so->so_state & SS_CANTRCVMORE) { 1686 kn->kn_flags |= EV_EOF; 1687 kn->kn_fflags = so->so_error; 1688 return (1); 1689 } 1690 if (so->so_error) /* temporary udp error */ 1691 return (1); 1692 if (kn->kn_sfflags & NOTE_LOWAT) 1693 return (kn->kn_data >= kn->kn_sdata); 1694 return (kn->kn_data >= so->so_rcv.sb_lowat); 1695 } 1696 1697 static void 1698 filt_sowdetach(struct knote *kn) 1699 { 1700 struct socket *so; 1701 1702 so = (struct socket *)kn->kn_fp->f_data; 1703 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 1704 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 1705 so->so_snd.sb_flags &= ~SB_KNOTE; 1706 } 1707 1708 /*ARGSUSED*/ 1709 static int 1710 filt_sowrite(struct knote *kn, long hint) 1711 { 1712 struct socket *so; 1713 1714 so = (struct socket *)kn->kn_fp->f_data; 1715 kn->kn_data = sbspace(&so->so_snd); 1716 if (so->so_state & SS_CANTSENDMORE) { 1717 kn->kn_flags |= EV_EOF; 1718 kn->kn_fflags = so->so_error; 1719 return (1); 1720 } 1721 if (so->so_error) /* temporary udp error */ 1722 return (1); 1723 if (((so->so_state & SS_ISCONNECTED) == 0) && 1724 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1725 return (0); 1726 if (kn->kn_sfflags & NOTE_LOWAT) 1727 return (kn->kn_data >= kn->kn_sdata); 1728 return (kn->kn_data >= so->so_snd.sb_lowat); 1729 } 1730 1731 /*ARGSUSED*/ 1732 static int 1733 filt_solisten(struct knote *kn, long hint) 1734 { 1735 struct socket *so; 1736 1737 so = (struct socket *)kn->kn_fp->f_data; 1738 1739 /* 1740 * Set kn_data to number of incoming connections, not 1741 * counting partial (incomplete) connections. 1742 */ 1743 kn->kn_data = so->so_qlen; 1744 return (kn->kn_data > 0); 1745 } 1746 1747 static const struct filterops solisten_filtops = 1748 { 1, NULL, filt_sordetach, filt_solisten }; 1749 static const struct filterops soread_filtops = 1750 { 1, NULL, filt_sordetach, filt_soread }; 1751 static const struct filterops sowrite_filtops = 1752 { 1, NULL, filt_sowdetach, filt_sowrite }; 1753 1754 int 1755 soo_kqfilter(struct file *fp, struct knote *kn) 1756 { 1757 struct socket *so; 1758 struct sockbuf *sb; 1759 1760 so = (struct socket *)kn->kn_fp->f_data; 1761 switch (kn->kn_filter) { 1762 case EVFILT_READ: 1763 if (so->so_options & SO_ACCEPTCONN) 1764 kn->kn_fop = &solisten_filtops; 1765 else 1766 kn->kn_fop = &soread_filtops; 1767 sb = &so->so_rcv; 1768 break; 1769 case EVFILT_WRITE: 1770 kn->kn_fop = &sowrite_filtops; 1771 sb = &so->so_snd; 1772 break; 1773 default: 1774 return (1); 1775 } 1776 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 1777 sb->sb_flags |= SB_KNOTE; 1778 return (0); 1779 } 1780 1781 #include <sys/sysctl.h> 1782 1783 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 1784 1785 /* 1786 * sysctl helper routine for kern.somaxkva. ensures that the given 1787 * value is not too small. 1788 * (XXX should we maybe make sure it's not too large as well?) 1789 */ 1790 static int 1791 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 1792 { 1793 int error, new_somaxkva; 1794 struct sysctlnode node; 1795 int s; 1796 1797 new_somaxkva = somaxkva; 1798 node = *rnode; 1799 node.sysctl_data = &new_somaxkva; 1800 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1801 if (error || newp == NULL) 1802 return (error); 1803 1804 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 1805 return (EINVAL); 1806 1807 s = splvm(); 1808 simple_lock(&so_pendfree_slock); 1809 somaxkva = new_somaxkva; 1810 wakeup(&socurkva); 1811 simple_unlock(&so_pendfree_slock); 1812 splx(s); 1813 1814 return (error); 1815 } 1816 1817 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup") 1818 { 1819 1820 sysctl_createv(clog, 0, NULL, NULL, 1821 CTLFLAG_PERMANENT, 1822 CTLTYPE_NODE, "kern", NULL, 1823 NULL, 0, NULL, 0, 1824 CTL_KERN, CTL_EOL); 1825 1826 sysctl_createv(clog, 0, NULL, NULL, 1827 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1828 CTLTYPE_INT, "somaxkva", 1829 SYSCTL_DESCR("Maximum amount of kernel memory to be " 1830 "used for socket buffers"), 1831 sysctl_kern_somaxkva, 0, NULL, 0, 1832 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 1833 } 1834
Indexes created Sat Sep 20 22:09:52 GMT 2025