uipc_socket.c revision 1.111.2.5
1 /* $NetBSD: uipc_socket.c,v 1.111.2.5 2006/12/30 20:50:07 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 1.111.2.5 2006/12/30 20:50:07 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, boolean_t 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 if (kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET, 475 KAUTH_REQ_NETWORK_SOCKET_OPEN, (void *)(u_long)dom, 476 (void *)(u_long)type, (void *)(u_long)proto) != 0) 477 return (EPERM); 478 479 if (proto) 480 prp = pffindproto(dom, proto, type); 481 else 482 prp = pffindtype(dom, type); 483 if (prp == 0) { 484 /* no support for domain */ 485 if (pffinddomain(dom) == 0) 486 return (EAFNOSUPPORT); 487 /* no support for socket type */ 488 if (proto == 0 && type != 0) 489 return (EPROTOTYPE); 490 return (EPROTONOSUPPORT); 491 } 492 if (prp->pr_usrreq == 0) 493 return (EPROTONOSUPPORT); 494 if (prp->pr_type != type) 495 return (EPROTOTYPE); 496 s = splsoftnet(); 497 so = pool_get(&socket_pool, PR_WAITOK); 498 memset((caddr_t)so, 0, sizeof(*so)); 499 TAILQ_INIT(&so->so_q0); 500 TAILQ_INIT(&so->so_q); 501 so->so_type = type; 502 so->so_proto = prp; 503 so->so_send = sosend; 504 so->so_receive = soreceive; 505 #ifdef MBUFTRACE 506 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 507 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 508 so->so_mowner = &prp->pr_domain->dom_mowner; 509 #endif 510 if (l != NULL) { 511 uid = kauth_cred_geteuid(l->l_cred); 512 } else { 513 uid = 0; 514 } 515 so->so_uidinfo = uid_find(uid); 516 error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0, 517 (struct mbuf *)(long)proto, (struct mbuf *)0, l); 518 if (error) { 519 so->so_state |= SS_NOFDREF; 520 sofree(so); 521 splx(s); 522 return (error); 523 } 524 splx(s); 525 *aso = so; 526 return (0); 527 } 528 529 int 530 sobind(struct socket *so, struct mbuf *nam, struct lwp *l) 531 { 532 int s, error; 533 534 s = splsoftnet(); 535 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0, 536 nam, (struct mbuf *)0, l); 537 splx(s); 538 return (error); 539 } 540 541 int 542 solisten(struct socket *so, int backlog) 543 { 544 int s, error; 545 546 s = splsoftnet(); 547 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0, 548 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 549 if (error) { 550 splx(s); 551 return (error); 552 } 553 if (TAILQ_EMPTY(&so->so_q)) 554 so->so_options |= SO_ACCEPTCONN; 555 if (backlog < 0) 556 backlog = 0; 557 so->so_qlimit = min(backlog, somaxconn); 558 splx(s); 559 return (0); 560 } 561 562 void 563 sofree(struct socket *so) 564 { 565 566 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 567 return; 568 if (so->so_head) { 569 /* 570 * We must not decommission a socket that's on the accept(2) 571 * queue. If we do, then accept(2) may hang after select(2) 572 * indicated that the listening socket was ready. 573 */ 574 if (!soqremque(so, 0)) 575 return; 576 } 577 if (so->so_rcv.sb_hiwat) 578 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 579 RLIM_INFINITY); 580 if (so->so_snd.sb_hiwat) 581 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 582 RLIM_INFINITY); 583 sbrelease(&so->so_snd, so); 584 sorflush(so); 585 pool_put(&socket_pool, so); 586 } 587 588 /* 589 * Close a socket on last file table reference removal. 590 * Initiate disconnect if connected. 591 * Free socket when disconnect complete. 592 */ 593 int 594 soclose(struct socket *so) 595 { 596 struct socket *so2; 597 int s, error; 598 599 error = 0; 600 s = splsoftnet(); /* conservative */ 601 if (so->so_options & SO_ACCEPTCONN) { 602 while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 603 (void) soqremque(so2, 0); 604 (void) soabort(so2); 605 } 606 while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 607 (void) soqremque(so2, 1); 608 (void) soabort(so2); 609 } 610 } 611 if (so->so_pcb == 0) 612 goto discard; 613 if (so->so_state & SS_ISCONNECTED) { 614 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 615 error = sodisconnect(so); 616 if (error) 617 goto drop; 618 } 619 if (so->so_options & SO_LINGER) { 620 if ((so->so_state & SS_ISDISCONNECTING) && 621 (so->so_state & SS_NBIO)) 622 goto drop; 623 while (so->so_state & SS_ISCONNECTED) { 624 error = tsleep((caddr_t)&so->so_timeo, 625 PSOCK | PCATCH, netcls, 626 so->so_linger * hz); 627 if (error) 628 break; 629 } 630 } 631 } 632 drop: 633 if (so->so_pcb) { 634 int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 635 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 636 (struct lwp *)0); 637 if (error == 0) 638 error = error2; 639 } 640 discard: 641 if (so->so_state & SS_NOFDREF) 642 panic("soclose: NOFDREF"); 643 so->so_state |= SS_NOFDREF; 644 sofree(so); 645 splx(s); 646 return (error); 647 } 648 649 /* 650 * Must be called at splsoftnet... 651 */ 652 int 653 soabort(struct socket *so) 654 { 655 656 return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0, 657 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 658 } 659 660 int 661 soaccept(struct socket *so, struct mbuf *nam) 662 { 663 int s, error; 664 665 error = 0; 666 s = splsoftnet(); 667 if ((so->so_state & SS_NOFDREF) == 0) 668 panic("soaccept: !NOFDREF"); 669 so->so_state &= ~SS_NOFDREF; 670 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 671 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 672 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 673 (struct mbuf *)0, nam, (struct mbuf *)0, (struct lwp *)0); 674 else 675 error = ECONNABORTED; 676 677 splx(s); 678 return (error); 679 } 680 681 int 682 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l) 683 { 684 int s, error; 685 686 if (so->so_options & SO_ACCEPTCONN) 687 return (EOPNOTSUPP); 688 s = splsoftnet(); 689 /* 690 * If protocol is connection-based, can only connect once. 691 * Otherwise, if connected, try to disconnect first. 692 * This allows user to disconnect by connecting to, e.g., 693 * a null address. 694 */ 695 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 696 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 697 (error = sodisconnect(so)))) 698 error = EISCONN; 699 else 700 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 701 (struct mbuf *)0, nam, (struct mbuf *)0, l); 702 splx(s); 703 return (error); 704 } 705 706 int 707 soconnect2(struct socket *so1, struct socket *so2) 708 { 709 int s, error; 710 711 s = splsoftnet(); 712 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 713 (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0, 714 (struct lwp *)0); 715 splx(s); 716 return (error); 717 } 718 719 int 720 sodisconnect(struct socket *so) 721 { 722 int s, error; 723 724 s = splsoftnet(); 725 if ((so->so_state & SS_ISCONNECTED) == 0) { 726 error = ENOTCONN; 727 goto bad; 728 } 729 if (so->so_state & SS_ISDISCONNECTING) { 730 error = EALREADY; 731 goto bad; 732 } 733 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 734 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 735 (struct lwp *)0); 736 bad: 737 splx(s); 738 sodopendfree(); 739 return (error); 740 } 741 742 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 743 /* 744 * Send on a socket. 745 * If send must go all at once and message is larger than 746 * send buffering, then hard error. 747 * Lock against other senders. 748 * If must go all at once and not enough room now, then 749 * inform user that this would block and do nothing. 750 * Otherwise, if nonblocking, send as much as possible. 751 * The data to be sent is described by "uio" if nonzero, 752 * otherwise by the mbuf chain "top" (which must be null 753 * if uio is not). Data provided in mbuf chain must be small 754 * enough to send all at once. 755 * 756 * Returns nonzero on error, timeout or signal; callers 757 * must check for short counts if EINTR/ERESTART are returned. 758 * Data and control buffers are freed on return. 759 */ 760 int 761 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 762 struct mbuf *control, int flags, struct lwp *l) 763 { 764 struct mbuf **mp, *m; 765 struct proc *p; 766 long space, len, resid, clen, mlen; 767 int error, s, dontroute, atomic; 768 769 p = l->l_proc; 770 sodopendfree(); 771 772 clen = 0; 773 atomic = sosendallatonce(so) || top; 774 if (uio) 775 resid = uio->uio_resid; 776 else 777 resid = top->m_pkthdr.len; 778 /* 779 * In theory resid should be unsigned. 780 * However, space must be signed, as it might be less than 0 781 * if we over-committed, and we must use a signed comparison 782 * of space and resid. On the other hand, a negative resid 783 * causes us to loop sending 0-length segments to the protocol. 784 */ 785 if (resid < 0) { 786 error = EINVAL; 787 goto out; 788 } 789 dontroute = 790 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 791 (so->so_proto->pr_flags & PR_ATOMIC); 792 if (p) 793 p->p_stats->p_ru.ru_msgsnd++; 794 if (control) 795 clen = control->m_len; 796 #define snderr(errno) { error = errno; splx(s); goto release; } 797 798 restart: 799 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 800 goto out; 801 do { 802 s = splsoftnet(); 803 if (so->so_state & SS_CANTSENDMORE) 804 snderr(EPIPE); 805 if (so->so_error) { 806 error = so->so_error; 807 so->so_error = 0; 808 splx(s); 809 goto release; 810 } 811 if ((so->so_state & SS_ISCONNECTED) == 0) { 812 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 813 if ((so->so_state & SS_ISCONFIRMING) == 0 && 814 !(resid == 0 && clen != 0)) 815 snderr(ENOTCONN); 816 } else if (addr == 0) 817 snderr(EDESTADDRREQ); 818 } 819 space = sbspace(&so->so_snd); 820 if (flags & MSG_OOB) 821 space += 1024; 822 if ((atomic && resid > so->so_snd.sb_hiwat) || 823 clen > so->so_snd.sb_hiwat) 824 snderr(EMSGSIZE); 825 if (space < resid + clen && 826 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 827 if (so->so_state & SS_NBIO) 828 snderr(EWOULDBLOCK); 829 sbunlock(&so->so_snd); 830 error = sbwait(&so->so_snd); 831 splx(s); 832 if (error) 833 goto out; 834 goto restart; 835 } 836 splx(s); 837 mp = ⊤ 838 space -= clen; 839 do { 840 if (uio == NULL) { 841 /* 842 * Data is prepackaged in "top". 843 */ 844 resid = 0; 845 if (flags & MSG_EOR) 846 top->m_flags |= M_EOR; 847 } else do { 848 if (top == 0) { 849 m = m_gethdr(M_WAIT, MT_DATA); 850 mlen = MHLEN; 851 m->m_pkthdr.len = 0; 852 m->m_pkthdr.rcvif = (struct ifnet *)0; 853 } else { 854 m = m_get(M_WAIT, MT_DATA); 855 mlen = MLEN; 856 } 857 MCLAIM(m, so->so_snd.sb_mowner); 858 if (sock_loan_thresh >= 0 && 859 uio->uio_iov->iov_len >= sock_loan_thresh && 860 space >= sock_loan_thresh && 861 (len = sosend_loan(so, uio, m, 862 space)) != 0) { 863 SOSEND_COUNTER_INCR(&sosend_loan_big); 864 space -= len; 865 goto have_data; 866 } 867 if (resid >= MINCLSIZE && space >= MCLBYTES) { 868 SOSEND_COUNTER_INCR(&sosend_copy_big); 869 m_clget(m, M_WAIT); 870 if ((m->m_flags & M_EXT) == 0) 871 goto nopages; 872 mlen = MCLBYTES; 873 if (atomic && top == 0) { 874 len = lmin(MCLBYTES - max_hdr, 875 resid); 876 m->m_data += max_hdr; 877 } else 878 len = lmin(MCLBYTES, resid); 879 space -= len; 880 } else { 881 nopages: 882 SOSEND_COUNTER_INCR(&sosend_copy_small); 883 len = lmin(lmin(mlen, resid), space); 884 space -= len; 885 /* 886 * For datagram protocols, leave room 887 * for protocol headers in first mbuf. 888 */ 889 if (atomic && top == 0 && len < mlen) 890 MH_ALIGN(m, len); 891 } 892 error = uiomove(mtod(m, caddr_t), (int)len, 893 uio); 894 have_data: 895 resid = uio->uio_resid; 896 m->m_len = len; 897 *mp = m; 898 top->m_pkthdr.len += len; 899 if (error) 900 goto release; 901 mp = &m->m_next; 902 if (resid <= 0) { 903 if (flags & MSG_EOR) 904 top->m_flags |= M_EOR; 905 break; 906 } 907 } while (space > 0 && atomic); 908 909 s = splsoftnet(); 910 911 if (so->so_state & SS_CANTSENDMORE) 912 snderr(EPIPE); 913 914 if (dontroute) 915 so->so_options |= SO_DONTROUTE; 916 if (resid > 0) 917 so->so_state |= SS_MORETOCOME; 918 error = (*so->so_proto->pr_usrreq)(so, 919 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 920 top, addr, control, curlwp); /* XXX */ 921 if (dontroute) 922 so->so_options &= ~SO_DONTROUTE; 923 if (resid > 0) 924 so->so_state &= ~SS_MORETOCOME; 925 splx(s); 926 927 clen = 0; 928 control = 0; 929 top = 0; 930 mp = ⊤ 931 if (error) 932 goto release; 933 } while (resid && space > 0); 934 } while (resid); 935 936 release: 937 sbunlock(&so->so_snd); 938 out: 939 if (top) 940 m_freem(top); 941 if (control) 942 m_freem(control); 943 return (error); 944 } 945 946 /* 947 * Implement receive operations on a socket. 948 * We depend on the way that records are added to the sockbuf 949 * by sbappend*. In particular, each record (mbufs linked through m_next) 950 * must begin with an address if the protocol so specifies, 951 * followed by an optional mbuf or mbufs containing ancillary data, 952 * and then zero or more mbufs of data. 953 * In order to avoid blocking network interrupts for the entire time here, 954 * we splx() while doing the actual copy to user space. 955 * Although the sockbuf is locked, new data may still be appended, 956 * and thus we must maintain consistency of the sockbuf during that time. 957 * 958 * The caller may receive the data as a single mbuf chain by supplying 959 * an mbuf **mp0 for use in returning the chain. The uio is then used 960 * only for the count in uio_resid. 961 */ 962 int 963 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 964 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 965 { 966 struct lwp *l = curlwp; 967 struct mbuf *m, **mp; 968 int flags, len, error, s, offset, moff, type, orig_resid; 969 const struct protosw *pr; 970 struct mbuf *nextrecord; 971 int mbuf_removed = 0; 972 973 pr = so->so_proto; 974 mp = mp0; 975 type = 0; 976 orig_resid = uio->uio_resid; 977 978 if (paddr) 979 *paddr = 0; 980 if (controlp) 981 *controlp = 0; 982 if (flagsp) 983 flags = *flagsp &~ MSG_EOR; 984 else 985 flags = 0; 986 987 if ((flags & MSG_DONTWAIT) == 0) 988 sodopendfree(); 989 990 if (flags & MSG_OOB) { 991 m = m_get(M_WAIT, MT_DATA); 992 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 993 (struct mbuf *)(long)(flags & MSG_PEEK), 994 (struct mbuf *)0, l); 995 if (error) 996 goto bad; 997 do { 998 error = uiomove(mtod(m, caddr_t), 999 (int) min(uio->uio_resid, m->m_len), uio); 1000 m = m_free(m); 1001 } while (uio->uio_resid && error == 0 && m); 1002 bad: 1003 if (m) 1004 m_freem(m); 1005 return (error); 1006 } 1007 if (mp) 1008 *mp = (struct mbuf *)0; 1009 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 1010 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1011 (struct mbuf *)0, (struct mbuf *)0, l); 1012 1013 restart: 1014 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) 1015 return (error); 1016 s = splsoftnet(); 1017 1018 m = so->so_rcv.sb_mb; 1019 /* 1020 * If we have less data than requested, block awaiting more 1021 * (subject to any timeout) if: 1022 * 1. the current count is less than the low water mark, 1023 * 2. MSG_WAITALL is set, and it is possible to do the entire 1024 * receive operation at once if we block (resid <= hiwat), or 1025 * 3. MSG_DONTWAIT is not set. 1026 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1027 * we have to do the receive in sections, and thus risk returning 1028 * a short count if a timeout or signal occurs after we start. 1029 */ 1030 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 1031 so->so_rcv.sb_cc < uio->uio_resid) && 1032 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1033 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1034 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1035 #ifdef DIAGNOSTIC 1036 if (m == 0 && so->so_rcv.sb_cc) 1037 panic("receive 1"); 1038 #endif 1039 if (so->so_error) { 1040 if (m) 1041 goto dontblock; 1042 error = so->so_error; 1043 if ((flags & MSG_PEEK) == 0) 1044 so->so_error = 0; 1045 goto release; 1046 } 1047 if (so->so_state & SS_CANTRCVMORE) { 1048 if (m) 1049 goto dontblock; 1050 else 1051 goto release; 1052 } 1053 for (; m; m = m->m_next) 1054 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1055 m = so->so_rcv.sb_mb; 1056 goto dontblock; 1057 } 1058 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1059 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1060 error = ENOTCONN; 1061 goto release; 1062 } 1063 if (uio->uio_resid == 0) 1064 goto release; 1065 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 1066 error = EWOULDBLOCK; 1067 goto release; 1068 } 1069 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1070 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1071 sbunlock(&so->so_rcv); 1072 error = sbwait(&so->so_rcv); 1073 splx(s); 1074 if (error) 1075 return (error); 1076 goto restart; 1077 } 1078 dontblock: 1079 /* 1080 * On entry here, m points to the first record of the socket buffer. 1081 * While we process the initial mbufs containing address and control 1082 * info, we save a copy of m->m_nextpkt into nextrecord. 1083 */ 1084 if (l) 1085 l->l_proc->p_stats->p_ru.ru_msgrcv++; 1086 KASSERT(m == so->so_rcv.sb_mb); 1087 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1088 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1089 nextrecord = m->m_nextpkt; 1090 if (pr->pr_flags & PR_ADDR) { 1091 #ifdef DIAGNOSTIC 1092 if (m->m_type != MT_SONAME) 1093 panic("receive 1a"); 1094 #endif 1095 orig_resid = 0; 1096 if (flags & MSG_PEEK) { 1097 if (paddr) 1098 *paddr = m_copy(m, 0, m->m_len); 1099 m = m->m_next; 1100 } else { 1101 sbfree(&so->so_rcv, m); 1102 mbuf_removed = 1; 1103 if (paddr) { 1104 *paddr = m; 1105 so->so_rcv.sb_mb = m->m_next; 1106 m->m_next = 0; 1107 m = so->so_rcv.sb_mb; 1108 } else { 1109 MFREE(m, so->so_rcv.sb_mb); 1110 m = so->so_rcv.sb_mb; 1111 } 1112 } 1113 } 1114 while (m && m->m_type == MT_CONTROL && error == 0) { 1115 if (flags & MSG_PEEK) { 1116 if (controlp) 1117 *controlp = m_copy(m, 0, m->m_len); 1118 m = m->m_next; 1119 } else { 1120 sbfree(&so->so_rcv, m); 1121 mbuf_removed = 1; 1122 if (controlp) { 1123 struct domain *dom = pr->pr_domain; 1124 if (dom->dom_externalize && l && 1125 mtod(m, struct cmsghdr *)->cmsg_type == 1126 SCM_RIGHTS) 1127 error = (*dom->dom_externalize)(m, l); 1128 *controlp = m; 1129 so->so_rcv.sb_mb = m->m_next; 1130 m->m_next = 0; 1131 m = so->so_rcv.sb_mb; 1132 } else { 1133 /* 1134 * Dispose of any SCM_RIGHTS message that went 1135 * through the read path rather than recv. 1136 */ 1137 if (pr->pr_domain->dom_dispose && 1138 mtod(m, struct cmsghdr *)->cmsg_type == SCM_RIGHTS) 1139 (*pr->pr_domain->dom_dispose)(m); 1140 MFREE(m, so->so_rcv.sb_mb); 1141 m = so->so_rcv.sb_mb; 1142 } 1143 } 1144 if (controlp) { 1145 orig_resid = 0; 1146 controlp = &(*controlp)->m_next; 1147 } 1148 } 1149 1150 /* 1151 * If m is non-NULL, we have some data to read. From now on, 1152 * make sure to keep sb_lastrecord consistent when working on 1153 * the last packet on the chain (nextrecord == NULL) and we 1154 * change m->m_nextpkt. 1155 */ 1156 if (m) { 1157 if ((flags & MSG_PEEK) == 0) { 1158 m->m_nextpkt = nextrecord; 1159 /* 1160 * If nextrecord == NULL (this is a single chain), 1161 * then sb_lastrecord may not be valid here if m 1162 * was changed earlier. 1163 */ 1164 if (nextrecord == NULL) { 1165 KASSERT(so->so_rcv.sb_mb == m); 1166 so->so_rcv.sb_lastrecord = m; 1167 } 1168 } 1169 type = m->m_type; 1170 if (type == MT_OOBDATA) 1171 flags |= MSG_OOB; 1172 } else { 1173 if ((flags & MSG_PEEK) == 0) { 1174 KASSERT(so->so_rcv.sb_mb == m); 1175 so->so_rcv.sb_mb = nextrecord; 1176 SB_EMPTY_FIXUP(&so->so_rcv); 1177 } 1178 } 1179 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1180 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1181 1182 moff = 0; 1183 offset = 0; 1184 while (m && uio->uio_resid > 0 && error == 0) { 1185 if (m->m_type == MT_OOBDATA) { 1186 if (type != MT_OOBDATA) 1187 break; 1188 } else if (type == MT_OOBDATA) 1189 break; 1190 #ifdef DIAGNOSTIC 1191 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1192 panic("receive 3"); 1193 #endif 1194 so->so_state &= ~SS_RCVATMARK; 1195 len = uio->uio_resid; 1196 if (so->so_oobmark && len > so->so_oobmark - offset) 1197 len = so->so_oobmark - offset; 1198 if (len > m->m_len - moff) 1199 len = m->m_len - moff; 1200 /* 1201 * If mp is set, just pass back the mbufs. 1202 * Otherwise copy them out via the uio, then free. 1203 * Sockbuf must be consistent here (points to current mbuf, 1204 * it points to next record) when we drop priority; 1205 * we must note any additions to the sockbuf when we 1206 * block interrupts again. 1207 */ 1208 if (mp == 0) { 1209 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1210 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1211 splx(s); 1212 error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio); 1213 s = splsoftnet(); 1214 if (error) { 1215 /* 1216 * If any part of the record has been removed 1217 * (such as the MT_SONAME mbuf, which will 1218 * happen when PR_ADDR, and thus also 1219 * PR_ATOMIC, is set), then drop the entire 1220 * record to maintain the atomicity of the 1221 * receive operation. 1222 * 1223 * This avoids a later panic("receive 1a") 1224 * when compiled with DIAGNOSTIC. 1225 */ 1226 if (m && mbuf_removed 1227 && (pr->pr_flags & PR_ATOMIC)) 1228 (void) sbdroprecord(&so->so_rcv); 1229 1230 goto release; 1231 } 1232 } else 1233 uio->uio_resid -= len; 1234 if (len == m->m_len - moff) { 1235 if (m->m_flags & M_EOR) 1236 flags |= MSG_EOR; 1237 if (flags & MSG_PEEK) { 1238 m = m->m_next; 1239 moff = 0; 1240 } else { 1241 nextrecord = m->m_nextpkt; 1242 sbfree(&so->so_rcv, m); 1243 if (mp) { 1244 *mp = m; 1245 mp = &m->m_next; 1246 so->so_rcv.sb_mb = m = m->m_next; 1247 *mp = (struct mbuf *)0; 1248 } else { 1249 MFREE(m, so->so_rcv.sb_mb); 1250 m = so->so_rcv.sb_mb; 1251 } 1252 /* 1253 * If m != NULL, we also know that 1254 * so->so_rcv.sb_mb != NULL. 1255 */ 1256 KASSERT(so->so_rcv.sb_mb == m); 1257 if (m) { 1258 m->m_nextpkt = nextrecord; 1259 if (nextrecord == NULL) 1260 so->so_rcv.sb_lastrecord = m; 1261 } else { 1262 so->so_rcv.sb_mb = nextrecord; 1263 SB_EMPTY_FIXUP(&so->so_rcv); 1264 } 1265 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1266 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1267 } 1268 } else { 1269 if (flags & MSG_PEEK) 1270 moff += len; 1271 else { 1272 if (mp) 1273 *mp = m_copym(m, 0, len, M_WAIT); 1274 m->m_data += len; 1275 m->m_len -= len; 1276 so->so_rcv.sb_cc -= len; 1277 } 1278 } 1279 if (so->so_oobmark) { 1280 if ((flags & MSG_PEEK) == 0) { 1281 so->so_oobmark -= len; 1282 if (so->so_oobmark == 0) { 1283 so->so_state |= SS_RCVATMARK; 1284 break; 1285 } 1286 } else { 1287 offset += len; 1288 if (offset == so->so_oobmark) 1289 break; 1290 } 1291 } 1292 if (flags & MSG_EOR) 1293 break; 1294 /* 1295 * If the MSG_WAITALL flag is set (for non-atomic socket), 1296 * we must not quit until "uio->uio_resid == 0" or an error 1297 * termination. If a signal/timeout occurs, return 1298 * with a short count but without error. 1299 * Keep sockbuf locked against other readers. 1300 */ 1301 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1302 !sosendallatonce(so) && !nextrecord) { 1303 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1304 break; 1305 /* 1306 * If we are peeking and the socket receive buffer is 1307 * full, stop since we can't get more data to peek at. 1308 */ 1309 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1310 break; 1311 /* 1312 * If we've drained the socket buffer, tell the 1313 * protocol in case it needs to do something to 1314 * get it filled again. 1315 */ 1316 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1317 (*pr->pr_usrreq)(so, PRU_RCVD, 1318 (struct mbuf *)0, 1319 (struct mbuf *)(long)flags, 1320 (struct mbuf *)0, l); 1321 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1322 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1323 error = sbwait(&so->so_rcv); 1324 if (error) { 1325 sbunlock(&so->so_rcv); 1326 splx(s); 1327 return (0); 1328 } 1329 if ((m = so->so_rcv.sb_mb) != NULL) 1330 nextrecord = m->m_nextpkt; 1331 } 1332 } 1333 1334 if (m && pr->pr_flags & PR_ATOMIC) { 1335 flags |= MSG_TRUNC; 1336 if ((flags & MSG_PEEK) == 0) 1337 (void) sbdroprecord(&so->so_rcv); 1338 } 1339 if ((flags & MSG_PEEK) == 0) { 1340 if (m == 0) { 1341 /* 1342 * First part is an inline SB_EMPTY_FIXUP(). Second 1343 * part makes sure sb_lastrecord is up-to-date if 1344 * there is still data in the socket buffer. 1345 */ 1346 so->so_rcv.sb_mb = nextrecord; 1347 if (so->so_rcv.sb_mb == NULL) { 1348 so->so_rcv.sb_mbtail = NULL; 1349 so->so_rcv.sb_lastrecord = NULL; 1350 } else if (nextrecord->m_nextpkt == NULL) 1351 so->so_rcv.sb_lastrecord = nextrecord; 1352 } 1353 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1354 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1355 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1356 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1357 (struct mbuf *)(long)flags, (struct mbuf *)0, l); 1358 } 1359 if (orig_resid == uio->uio_resid && orig_resid && 1360 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1361 sbunlock(&so->so_rcv); 1362 splx(s); 1363 goto restart; 1364 } 1365 1366 if (flagsp) 1367 *flagsp |= flags; 1368 release: 1369 sbunlock(&so->so_rcv); 1370 splx(s); 1371 return (error); 1372 } 1373 1374 int 1375 soshutdown(struct socket *so, int how) 1376 { 1377 const struct protosw *pr; 1378 1379 pr = so->so_proto; 1380 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1381 return (EINVAL); 1382 1383 if (how == SHUT_RD || how == SHUT_RDWR) 1384 sorflush(so); 1385 if (how == SHUT_WR || how == SHUT_RDWR) 1386 return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0, 1387 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1388 return (0); 1389 } 1390 1391 void 1392 sorflush(struct socket *so) 1393 { 1394 struct sockbuf *sb, asb; 1395 const struct protosw *pr; 1396 int s; 1397 1398 sb = &so->so_rcv; 1399 pr = so->so_proto; 1400 sb->sb_flags |= SB_NOINTR; 1401 (void) sblock(sb, M_WAITOK); 1402 s = splnet(); 1403 socantrcvmore(so); 1404 sbunlock(sb); 1405 asb = *sb; 1406 /* 1407 * Clear most of the sockbuf structure, but leave some of the 1408 * fields valid. 1409 */ 1410 memset(&sb->sb_startzero, 0, 1411 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1412 splx(s); 1413 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1414 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1415 sbrelease(&asb, so); 1416 } 1417 1418 int 1419 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0) 1420 { 1421 int error; 1422 struct mbuf *m; 1423 struct linger *l; 1424 1425 error = 0; 1426 m = m0; 1427 if (level != SOL_SOCKET) { 1428 if (so->so_proto && so->so_proto->pr_ctloutput) 1429 return ((*so->so_proto->pr_ctloutput) 1430 (PRCO_SETOPT, so, level, optname, &m0)); 1431 error = ENOPROTOOPT; 1432 } else { 1433 switch (optname) { 1434 1435 case SO_LINGER: 1436 if (m == NULL || m->m_len != sizeof(struct linger)) { 1437 error = EINVAL; 1438 goto bad; 1439 } 1440 l = mtod(m, struct linger *); 1441 if (l->l_linger < 0 || l->l_linger > USHRT_MAX || 1442 l->l_linger > (INT_MAX / hz)) { 1443 error = EDOM; 1444 goto bad; 1445 } 1446 so->so_linger = l->l_linger; 1447 if (l->l_onoff) 1448 so->so_options |= SO_LINGER; 1449 else 1450 so->so_options &= ~SO_LINGER; 1451 break; 1452 1453 case SO_DEBUG: 1454 case SO_KEEPALIVE: 1455 case SO_DONTROUTE: 1456 case SO_USELOOPBACK: 1457 case SO_BROADCAST: 1458 case SO_REUSEADDR: 1459 case SO_REUSEPORT: 1460 case SO_OOBINLINE: 1461 case SO_TIMESTAMP: 1462 if (m == NULL || m->m_len < sizeof(int)) { 1463 error = EINVAL; 1464 goto bad; 1465 } 1466 if (*mtod(m, int *)) 1467 so->so_options |= optname; 1468 else 1469 so->so_options &= ~optname; 1470 break; 1471 1472 case SO_SNDBUF: 1473 case SO_RCVBUF: 1474 case SO_SNDLOWAT: 1475 case SO_RCVLOWAT: 1476 { 1477 int optval; 1478 1479 if (m == NULL || m->m_len < sizeof(int)) { 1480 error = EINVAL; 1481 goto bad; 1482 } 1483 1484 /* 1485 * Values < 1 make no sense for any of these 1486 * options, so disallow them. 1487 */ 1488 optval = *mtod(m, int *); 1489 if (optval < 1) { 1490 error = EINVAL; 1491 goto bad; 1492 } 1493 1494 switch (optname) { 1495 1496 case SO_SNDBUF: 1497 case SO_RCVBUF: 1498 if (sbreserve(optname == SO_SNDBUF ? 1499 &so->so_snd : &so->so_rcv, 1500 (u_long) optval, so) == 0) { 1501 error = ENOBUFS; 1502 goto bad; 1503 } 1504 break; 1505 1506 /* 1507 * Make sure the low-water is never greater than 1508 * the high-water. 1509 */ 1510 case SO_SNDLOWAT: 1511 so->so_snd.sb_lowat = 1512 (optval > so->so_snd.sb_hiwat) ? 1513 so->so_snd.sb_hiwat : optval; 1514 break; 1515 case SO_RCVLOWAT: 1516 so->so_rcv.sb_lowat = 1517 (optval > so->so_rcv.sb_hiwat) ? 1518 so->so_rcv.sb_hiwat : optval; 1519 break; 1520 } 1521 break; 1522 } 1523 1524 case SO_SNDTIMEO: 1525 case SO_RCVTIMEO: 1526 { 1527 struct timeval *tv; 1528 int val; 1529 1530 if (m == NULL || m->m_len < sizeof(*tv)) { 1531 error = EINVAL; 1532 goto bad; 1533 } 1534 tv = mtod(m, struct timeval *); 1535 if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz) { 1536 error = EDOM; 1537 goto bad; 1538 } 1539 val = tv->tv_sec * hz + tv->tv_usec / tick; 1540 if (val == 0 && tv->tv_usec != 0) 1541 val = 1; 1542 1543 switch (optname) { 1544 1545 case SO_SNDTIMEO: 1546 so->so_snd.sb_timeo = val; 1547 break; 1548 case SO_RCVTIMEO: 1549 so->so_rcv.sb_timeo = val; 1550 break; 1551 } 1552 break; 1553 } 1554 1555 default: 1556 error = ENOPROTOOPT; 1557 break; 1558 } 1559 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1560 (void) ((*so->so_proto->pr_ctloutput) 1561 (PRCO_SETOPT, so, level, optname, &m0)); 1562 m = NULL; /* freed by protocol */ 1563 } 1564 } 1565 bad: 1566 if (m) 1567 (void) m_free(m); 1568 return (error); 1569 } 1570 1571 int 1572 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp) 1573 { 1574 struct mbuf *m; 1575 1576 if (level != SOL_SOCKET) { 1577 if (so->so_proto && so->so_proto->pr_ctloutput) { 1578 return ((*so->so_proto->pr_ctloutput) 1579 (PRCO_GETOPT, so, level, optname, mp)); 1580 } else 1581 return (ENOPROTOOPT); 1582 } else { 1583 m = m_get(M_WAIT, MT_SOOPTS); 1584 m->m_len = sizeof(int); 1585 1586 switch (optname) { 1587 1588 case SO_LINGER: 1589 m->m_len = sizeof(struct linger); 1590 mtod(m, struct linger *)->l_onoff = 1591 (so->so_options & SO_LINGER) ? 1 : 0; 1592 mtod(m, struct linger *)->l_linger = so->so_linger; 1593 break; 1594 1595 case SO_USELOOPBACK: 1596 case SO_DONTROUTE: 1597 case SO_DEBUG: 1598 case SO_KEEPALIVE: 1599 case SO_REUSEADDR: 1600 case SO_REUSEPORT: 1601 case SO_BROADCAST: 1602 case SO_OOBINLINE: 1603 case SO_TIMESTAMP: 1604 *mtod(m, int *) = (so->so_options & optname) ? 1 : 0; 1605 break; 1606 1607 case SO_TYPE: 1608 *mtod(m, int *) = so->so_type; 1609 break; 1610 1611 case SO_ERROR: 1612 *mtod(m, int *) = so->so_error; 1613 so->so_error = 0; 1614 break; 1615 1616 case SO_SNDBUF: 1617 *mtod(m, int *) = so->so_snd.sb_hiwat; 1618 break; 1619 1620 case SO_RCVBUF: 1621 *mtod(m, int *) = so->so_rcv.sb_hiwat; 1622 break; 1623 1624 case SO_SNDLOWAT: 1625 *mtod(m, int *) = so->so_snd.sb_lowat; 1626 break; 1627 1628 case SO_RCVLOWAT: 1629 *mtod(m, int *) = so->so_rcv.sb_lowat; 1630 break; 1631 1632 case SO_SNDTIMEO: 1633 case SO_RCVTIMEO: 1634 { 1635 int val = (optname == SO_SNDTIMEO ? 1636 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1637 1638 m->m_len = sizeof(struct timeval); 1639 mtod(m, struct timeval *)->tv_sec = val / hz; 1640 mtod(m, struct timeval *)->tv_usec = 1641 (val % hz) * tick; 1642 break; 1643 } 1644 1645 case SO_OVERFLOWED: 1646 *mtod(m, int *) = so->so_rcv.sb_overflowed; 1647 break; 1648 1649 default: 1650 (void)m_free(m); 1651 return (ENOPROTOOPT); 1652 } 1653 *mp = m; 1654 return (0); 1655 } 1656 } 1657 1658 void 1659 sohasoutofband(struct socket *so) 1660 { 1661 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 1662 selwakeup(&so->so_rcv.sb_sel); 1663 } 1664 1665 static void 1666 filt_sordetach(struct knote *kn) 1667 { 1668 struct socket *so; 1669 1670 so = (struct socket *)kn->kn_fp->f_data; 1671 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 1672 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 1673 so->so_rcv.sb_flags &= ~SB_KNOTE; 1674 } 1675 1676 /*ARGSUSED*/ 1677 static int 1678 filt_soread(struct knote *kn, long hint) 1679 { 1680 struct socket *so; 1681 1682 so = (struct socket *)kn->kn_fp->f_data; 1683 kn->kn_data = so->so_rcv.sb_cc; 1684 if (so->so_state & SS_CANTRCVMORE) { 1685 kn->kn_flags |= EV_EOF; 1686 kn->kn_fflags = so->so_error; 1687 return (1); 1688 } 1689 if (so->so_error) /* temporary udp error */ 1690 return (1); 1691 if (kn->kn_sfflags & NOTE_LOWAT) 1692 return (kn->kn_data >= kn->kn_sdata); 1693 return (kn->kn_data >= so->so_rcv.sb_lowat); 1694 } 1695 1696 static void 1697 filt_sowdetach(struct knote *kn) 1698 { 1699 struct socket *so; 1700 1701 so = (struct socket *)kn->kn_fp->f_data; 1702 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 1703 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 1704 so->so_snd.sb_flags &= ~SB_KNOTE; 1705 } 1706 1707 /*ARGSUSED*/ 1708 static int 1709 filt_sowrite(struct knote *kn, long hint) 1710 { 1711 struct socket *so; 1712 1713 so = (struct socket *)kn->kn_fp->f_data; 1714 kn->kn_data = sbspace(&so->so_snd); 1715 if (so->so_state & SS_CANTSENDMORE) { 1716 kn->kn_flags |= EV_EOF; 1717 kn->kn_fflags = so->so_error; 1718 return (1); 1719 } 1720 if (so->so_error) /* temporary udp error */ 1721 return (1); 1722 if (((so->so_state & SS_ISCONNECTED) == 0) && 1723 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1724 return (0); 1725 if (kn->kn_sfflags & NOTE_LOWAT) 1726 return (kn->kn_data >= kn->kn_sdata); 1727 return (kn->kn_data >= so->so_snd.sb_lowat); 1728 } 1729 1730 /*ARGSUSED*/ 1731 static int 1732 filt_solisten(struct knote *kn, long hint) 1733 { 1734 struct socket *so; 1735 1736 so = (struct socket *)kn->kn_fp->f_data; 1737 1738 /* 1739 * Set kn_data to number of incoming connections, not 1740 * counting partial (incomplete) connections. 1741 */ 1742 kn->kn_data = so->so_qlen; 1743 return (kn->kn_data > 0); 1744 } 1745 1746 static const struct filterops solisten_filtops = 1747 { 1, NULL, filt_sordetach, filt_solisten }; 1748 static const struct filterops soread_filtops = 1749 { 1, NULL, filt_sordetach, filt_soread }; 1750 static const struct filterops sowrite_filtops = 1751 { 1, NULL, filt_sowdetach, filt_sowrite }; 1752 1753 int 1754 soo_kqfilter(struct file *fp, struct knote *kn) 1755 { 1756 struct socket *so; 1757 struct sockbuf *sb; 1758 1759 so = (struct socket *)kn->kn_fp->f_data; 1760 switch (kn->kn_filter) { 1761 case EVFILT_READ: 1762 if (so->so_options & SO_ACCEPTCONN) 1763 kn->kn_fop = &solisten_filtops; 1764 else 1765 kn->kn_fop = &soread_filtops; 1766 sb = &so->so_rcv; 1767 break; 1768 case EVFILT_WRITE: 1769 kn->kn_fop = &sowrite_filtops; 1770 sb = &so->so_snd; 1771 break; 1772 default: 1773 return (1); 1774 } 1775 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 1776 sb->sb_flags |= SB_KNOTE; 1777 return (0); 1778 } 1779 1780 #include <sys/sysctl.h> 1781 1782 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 1783 1784 /* 1785 * sysctl helper routine for kern.somaxkva. ensures that the given 1786 * value is not too small. 1787 * (XXX should we maybe make sure it's not too large as well?) 1788 */ 1789 static int 1790 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 1791 { 1792 int error, new_somaxkva; 1793 struct sysctlnode node; 1794 int s; 1795 1796 new_somaxkva = somaxkva; 1797 node = *rnode; 1798 node.sysctl_data = &new_somaxkva; 1799 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1800 if (error || newp == NULL) 1801 return (error); 1802 1803 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 1804 return (EINVAL); 1805 1806 s = splvm(); 1807 simple_lock(&so_pendfree_slock); 1808 somaxkva = new_somaxkva; 1809 wakeup(&socurkva); 1810 simple_unlock(&so_pendfree_slock); 1811 splx(s); 1812 1813 return (error); 1814 } 1815 1816 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup") 1817 { 1818 1819 sysctl_createv(clog, 0, NULL, NULL, 1820 CTLFLAG_PERMANENT, 1821 CTLTYPE_NODE, "kern", NULL, 1822 NULL, 0, NULL, 0, 1823 CTL_KERN, CTL_EOL); 1824 1825 sysctl_createv(clog, 0, NULL, NULL, 1826 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1827 CTLTYPE_INT, "somaxkva", 1828 SYSCTL_DESCR("Maximum amount of kernel memory to be " 1829 "used for socket buffers"), 1830 sysctl_kern_somaxkva, 0, NULL, 0, 1831 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 1832 } 1833
Indexes created Sat Sep 20 22:09:52 GMT 2025