uipc_socket.c revision 1.111.2.3
1 /* $NetBSD: uipc_socket.c,v 1.111.2.3 2006/06/21 15:09:39 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 1.111.2.3 2006/06/21 15:09:39 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(!MCLISREFERENCED(m)); 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 (proto) 475 prp = pffindproto(dom, proto, type); 476 else 477 prp = pffindtype(dom, type); 478 if (prp == 0) { 479 /* no support for domain */ 480 if (pffinddomain(dom) == 0) 481 return (EAFNOSUPPORT); 482 /* no support for socket type */ 483 if (proto == 0 && type != 0) 484 return (EPROTOTYPE); 485 return (EPROTONOSUPPORT); 486 } 487 if (prp->pr_usrreq == 0) 488 return (EPROTONOSUPPORT); 489 if (prp->pr_type != type) 490 return (EPROTOTYPE); 491 s = splsoftnet(); 492 so = pool_get(&socket_pool, PR_WAITOK); 493 memset((caddr_t)so, 0, sizeof(*so)); 494 TAILQ_INIT(&so->so_q0); 495 TAILQ_INIT(&so->so_q); 496 so->so_type = type; 497 so->so_proto = prp; 498 so->so_send = sosend; 499 so->so_receive = soreceive; 500 #ifdef MBUFTRACE 501 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 502 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 503 so->so_mowner = &prp->pr_domain->dom_mowner; 504 #endif 505 if (l != NULL) { 506 uid = kauth_cred_geteuid(l->l_proc->p_cred); 507 } else { 508 uid = 0; 509 } 510 so->so_uidinfo = uid_find(uid); 511 error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0, 512 (struct mbuf *)(long)proto, (struct mbuf *)0, l); 513 if (error) { 514 so->so_state |= SS_NOFDREF; 515 sofree(so); 516 splx(s); 517 return (error); 518 } 519 splx(s); 520 *aso = so; 521 return (0); 522 } 523 524 int 525 sobind(struct socket *so, struct mbuf *nam, struct lwp *l) 526 { 527 int s, error; 528 529 s = splsoftnet(); 530 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0, 531 nam, (struct mbuf *)0, l); 532 splx(s); 533 return (error); 534 } 535 536 int 537 solisten(struct socket *so, int backlog) 538 { 539 int s, error; 540 541 s = splsoftnet(); 542 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0, 543 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 544 if (error) { 545 splx(s); 546 return (error); 547 } 548 if (TAILQ_EMPTY(&so->so_q)) 549 so->so_options |= SO_ACCEPTCONN; 550 if (backlog < 0) 551 backlog = 0; 552 so->so_qlimit = min(backlog, somaxconn); 553 splx(s); 554 return (0); 555 } 556 557 void 558 sofree(struct socket *so) 559 { 560 561 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 562 return; 563 if (so->so_head) { 564 /* 565 * We must not decommission a socket that's on the accept(2) 566 * queue. If we do, then accept(2) may hang after select(2) 567 * indicated that the listening socket was ready. 568 */ 569 if (!soqremque(so, 0)) 570 return; 571 } 572 if (so->so_rcv.sb_hiwat) 573 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 574 RLIM_INFINITY); 575 if (so->so_snd.sb_hiwat) 576 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 577 RLIM_INFINITY); 578 sbrelease(&so->so_snd, so); 579 sorflush(so); 580 pool_put(&socket_pool, so); 581 } 582 583 /* 584 * Close a socket on last file table reference removal. 585 * Initiate disconnect if connected. 586 * Free socket when disconnect complete. 587 */ 588 int 589 soclose(struct socket *so) 590 { 591 struct socket *so2; 592 int s, error; 593 594 error = 0; 595 s = splsoftnet(); /* conservative */ 596 if (so->so_options & SO_ACCEPTCONN) { 597 while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 598 (void) soqremque(so2, 0); 599 (void) soabort(so2); 600 } 601 while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 602 (void) soqremque(so2, 1); 603 (void) soabort(so2); 604 } 605 } 606 if (so->so_pcb == 0) 607 goto discard; 608 if (so->so_state & SS_ISCONNECTED) { 609 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 610 error = sodisconnect(so); 611 if (error) 612 goto drop; 613 } 614 if (so->so_options & SO_LINGER) { 615 if ((so->so_state & SS_ISDISCONNECTING) && 616 (so->so_state & SS_NBIO)) 617 goto drop; 618 while (so->so_state & SS_ISCONNECTED) { 619 error = tsleep((caddr_t)&so->so_timeo, 620 PSOCK | PCATCH, netcls, 621 so->so_linger * hz); 622 if (error) 623 break; 624 } 625 } 626 } 627 drop: 628 if (so->so_pcb) { 629 int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 630 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 631 (struct lwp *)0); 632 if (error == 0) 633 error = error2; 634 } 635 discard: 636 if (so->so_state & SS_NOFDREF) 637 panic("soclose: NOFDREF"); 638 so->so_state |= SS_NOFDREF; 639 sofree(so); 640 splx(s); 641 return (error); 642 } 643 644 /* 645 * Must be called at splsoftnet... 646 */ 647 int 648 soabort(struct socket *so) 649 { 650 651 return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0, 652 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 653 } 654 655 int 656 soaccept(struct socket *so, struct mbuf *nam) 657 { 658 int s, error; 659 660 error = 0; 661 s = splsoftnet(); 662 if ((so->so_state & SS_NOFDREF) == 0) 663 panic("soaccept: !NOFDREF"); 664 so->so_state &= ~SS_NOFDREF; 665 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 666 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 667 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 668 (struct mbuf *)0, nam, (struct mbuf *)0, (struct lwp *)0); 669 else 670 error = ECONNABORTED; 671 672 splx(s); 673 return (error); 674 } 675 676 int 677 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l) 678 { 679 int s, error; 680 681 if (so->so_options & SO_ACCEPTCONN) 682 return (EOPNOTSUPP); 683 s = splsoftnet(); 684 /* 685 * If protocol is connection-based, can only connect once. 686 * Otherwise, if connected, try to disconnect first. 687 * This allows user to disconnect by connecting to, e.g., 688 * a null address. 689 */ 690 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 691 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 692 (error = sodisconnect(so)))) 693 error = EISCONN; 694 else 695 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 696 (struct mbuf *)0, nam, (struct mbuf *)0, l); 697 splx(s); 698 return (error); 699 } 700 701 int 702 soconnect2(struct socket *so1, struct socket *so2) 703 { 704 int s, error; 705 706 s = splsoftnet(); 707 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 708 (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0, 709 (struct lwp *)0); 710 splx(s); 711 return (error); 712 } 713 714 int 715 sodisconnect(struct socket *so) 716 { 717 int s, error; 718 719 s = splsoftnet(); 720 if ((so->so_state & SS_ISCONNECTED) == 0) { 721 error = ENOTCONN; 722 goto bad; 723 } 724 if (so->so_state & SS_ISDISCONNECTING) { 725 error = EALREADY; 726 goto bad; 727 } 728 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 729 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 730 (struct lwp *)0); 731 bad: 732 splx(s); 733 sodopendfree(); 734 return (error); 735 } 736 737 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 738 /* 739 * Send on a socket. 740 * If send must go all at once and message is larger than 741 * send buffering, then hard error. 742 * Lock against other senders. 743 * If must go all at once and not enough room now, then 744 * inform user that this would block and do nothing. 745 * Otherwise, if nonblocking, send as much as possible. 746 * The data to be sent is described by "uio" if nonzero, 747 * otherwise by the mbuf chain "top" (which must be null 748 * if uio is not). Data provided in mbuf chain must be small 749 * enough to send all at once. 750 * 751 * Returns nonzero on error, timeout or signal; callers 752 * must check for short counts if EINTR/ERESTART are returned. 753 * Data and control buffers are freed on return. 754 */ 755 int 756 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 757 struct mbuf *control, int flags, struct lwp *l) 758 { 759 struct mbuf **mp, *m; 760 struct proc *p; 761 long space, len, resid, clen, mlen; 762 int error, s, dontroute, atomic; 763 764 p = l->l_proc; 765 sodopendfree(); 766 767 clen = 0; 768 atomic = sosendallatonce(so) || top; 769 if (uio) 770 resid = uio->uio_resid; 771 else 772 resid = top->m_pkthdr.len; 773 /* 774 * In theory resid should be unsigned. 775 * However, space must be signed, as it might be less than 0 776 * if we over-committed, and we must use a signed comparison 777 * of space and resid. On the other hand, a negative resid 778 * causes us to loop sending 0-length segments to the protocol. 779 */ 780 if (resid < 0) { 781 error = EINVAL; 782 goto out; 783 } 784 dontroute = 785 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 786 (so->so_proto->pr_flags & PR_ATOMIC); 787 if (p) 788 p->p_stats->p_ru.ru_msgsnd++; 789 if (control) 790 clen = control->m_len; 791 #define snderr(errno) { error = errno; splx(s); goto release; } 792 793 restart: 794 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 795 goto out; 796 do { 797 s = splsoftnet(); 798 if (so->so_state & SS_CANTSENDMORE) 799 snderr(EPIPE); 800 if (so->so_error) { 801 error = so->so_error; 802 so->so_error = 0; 803 splx(s); 804 goto release; 805 } 806 if ((so->so_state & SS_ISCONNECTED) == 0) { 807 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 808 if ((so->so_state & SS_ISCONFIRMING) == 0 && 809 !(resid == 0 && clen != 0)) 810 snderr(ENOTCONN); 811 } else if (addr == 0) 812 snderr(EDESTADDRREQ); 813 } 814 space = sbspace(&so->so_snd); 815 if (flags & MSG_OOB) 816 space += 1024; 817 if ((atomic && resid > so->so_snd.sb_hiwat) || 818 clen > so->so_snd.sb_hiwat) 819 snderr(EMSGSIZE); 820 if (space < resid + clen && 821 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 822 if (so->so_state & SS_NBIO) 823 snderr(EWOULDBLOCK); 824 sbunlock(&so->so_snd); 825 error = sbwait(&so->so_snd); 826 splx(s); 827 if (error) 828 goto out; 829 goto restart; 830 } 831 splx(s); 832 mp = ⊤ 833 space -= clen; 834 do { 835 if (uio == NULL) { 836 /* 837 * Data is prepackaged in "top". 838 */ 839 resid = 0; 840 if (flags & MSG_EOR) 841 top->m_flags |= M_EOR; 842 } else do { 843 if (top == 0) { 844 m = m_gethdr(M_WAIT, MT_DATA); 845 mlen = MHLEN; 846 m->m_pkthdr.len = 0; 847 m->m_pkthdr.rcvif = (struct ifnet *)0; 848 } else { 849 m = m_get(M_WAIT, MT_DATA); 850 mlen = MLEN; 851 } 852 MCLAIM(m, so->so_snd.sb_mowner); 853 if (sock_loan_thresh >= 0 && 854 uio->uio_iov->iov_len >= sock_loan_thresh && 855 space >= sock_loan_thresh && 856 (len = sosend_loan(so, uio, m, 857 space)) != 0) { 858 SOSEND_COUNTER_INCR(&sosend_loan_big); 859 space -= len; 860 goto have_data; 861 } 862 if (resid >= MINCLSIZE && space >= MCLBYTES) { 863 SOSEND_COUNTER_INCR(&sosend_copy_big); 864 m_clget(m, M_WAIT); 865 if ((m->m_flags & M_EXT) == 0) 866 goto nopages; 867 mlen = MCLBYTES; 868 if (atomic && top == 0) { 869 len = lmin(MCLBYTES - max_hdr, 870 resid); 871 m->m_data += max_hdr; 872 } else 873 len = lmin(MCLBYTES, resid); 874 space -= len; 875 } else { 876 nopages: 877 SOSEND_COUNTER_INCR(&sosend_copy_small); 878 len = lmin(lmin(mlen, resid), space); 879 space -= len; 880 /* 881 * For datagram protocols, leave room 882 * for protocol headers in first mbuf. 883 */ 884 if (atomic && top == 0 && len < mlen) 885 MH_ALIGN(m, len); 886 } 887 error = uiomove(mtod(m, caddr_t), (int)len, 888 uio); 889 have_data: 890 resid = uio->uio_resid; 891 m->m_len = len; 892 *mp = m; 893 top->m_pkthdr.len += len; 894 if (error) 895 goto release; 896 mp = &m->m_next; 897 if (resid <= 0) { 898 if (flags & MSG_EOR) 899 top->m_flags |= M_EOR; 900 break; 901 } 902 } while (space > 0 && atomic); 903 904 s = splsoftnet(); 905 906 if (so->so_state & SS_CANTSENDMORE) 907 snderr(EPIPE); 908 909 if (dontroute) 910 so->so_options |= SO_DONTROUTE; 911 if (resid > 0) 912 so->so_state |= SS_MORETOCOME; 913 error = (*so->so_proto->pr_usrreq)(so, 914 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 915 top, addr, control, curlwp); /* XXX */ 916 if (dontroute) 917 so->so_options &= ~SO_DONTROUTE; 918 if (resid > 0) 919 so->so_state &= ~SS_MORETOCOME; 920 splx(s); 921 922 clen = 0; 923 control = 0; 924 top = 0; 925 mp = ⊤ 926 if (error) 927 goto release; 928 } while (resid && space > 0); 929 } while (resid); 930 931 release: 932 sbunlock(&so->so_snd); 933 out: 934 if (top) 935 m_freem(top); 936 if (control) 937 m_freem(control); 938 return (error); 939 } 940 941 /* 942 * Implement receive operations on a socket. 943 * We depend on the way that records are added to the sockbuf 944 * by sbappend*. In particular, each record (mbufs linked through m_next) 945 * must begin with an address if the protocol so specifies, 946 * followed by an optional mbuf or mbufs containing ancillary data, 947 * and then zero or more mbufs of data. 948 * In order to avoid blocking network interrupts for the entire time here, 949 * we splx() while doing the actual copy to user space. 950 * Although the sockbuf is locked, new data may still be appended, 951 * and thus we must maintain consistency of the sockbuf during that time. 952 * 953 * The caller may receive the data as a single mbuf chain by supplying 954 * an mbuf **mp0 for use in returning the chain. The uio is then used 955 * only for the count in uio_resid. 956 */ 957 int 958 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 959 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 960 { 961 struct lwp *l = curlwp; 962 struct mbuf *m, **mp; 963 int flags, len, error, s, offset, moff, type, orig_resid; 964 const struct protosw *pr; 965 struct mbuf *nextrecord; 966 int mbuf_removed = 0; 967 968 pr = so->so_proto; 969 mp = mp0; 970 type = 0; 971 orig_resid = uio->uio_resid; 972 973 if (paddr) 974 *paddr = 0; 975 if (controlp) 976 *controlp = 0; 977 if (flagsp) 978 flags = *flagsp &~ MSG_EOR; 979 else 980 flags = 0; 981 982 if ((flags & MSG_DONTWAIT) == 0) 983 sodopendfree(); 984 985 if (flags & MSG_OOB) { 986 m = m_get(M_WAIT, MT_DATA); 987 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 988 (struct mbuf *)(long)(flags & MSG_PEEK), 989 (struct mbuf *)0, l); 990 if (error) 991 goto bad; 992 do { 993 error = uiomove(mtod(m, caddr_t), 994 (int) min(uio->uio_resid, m->m_len), uio); 995 m = m_free(m); 996 } while (uio->uio_resid && error == 0 && m); 997 bad: 998 if (m) 999 m_freem(m); 1000 return (error); 1001 } 1002 if (mp) 1003 *mp = (struct mbuf *)0; 1004 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 1005 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1006 (struct mbuf *)0, (struct mbuf *)0, l); 1007 1008 restart: 1009 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) 1010 return (error); 1011 s = splsoftnet(); 1012 1013 m = so->so_rcv.sb_mb; 1014 /* 1015 * If we have less data than requested, block awaiting more 1016 * (subject to any timeout) if: 1017 * 1. the current count is less than the low water mark, 1018 * 2. MSG_WAITALL is set, and it is possible to do the entire 1019 * receive operation at once if we block (resid <= hiwat), or 1020 * 3. MSG_DONTWAIT is not set. 1021 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1022 * we have to do the receive in sections, and thus risk returning 1023 * a short count if a timeout or signal occurs after we start. 1024 */ 1025 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 1026 so->so_rcv.sb_cc < uio->uio_resid) && 1027 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1028 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1029 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1030 #ifdef DIAGNOSTIC 1031 if (m == 0 && so->so_rcv.sb_cc) 1032 panic("receive 1"); 1033 #endif 1034 if (so->so_error) { 1035 if (m) 1036 goto dontblock; 1037 error = so->so_error; 1038 if ((flags & MSG_PEEK) == 0) 1039 so->so_error = 0; 1040 goto release; 1041 } 1042 if (so->so_state & SS_CANTRCVMORE) { 1043 if (m) 1044 goto dontblock; 1045 else 1046 goto release; 1047 } 1048 for (; m; m = m->m_next) 1049 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1050 m = so->so_rcv.sb_mb; 1051 goto dontblock; 1052 } 1053 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1054 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1055 error = ENOTCONN; 1056 goto release; 1057 } 1058 if (uio->uio_resid == 0) 1059 goto release; 1060 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 1061 error = EWOULDBLOCK; 1062 goto release; 1063 } 1064 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1065 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1066 sbunlock(&so->so_rcv); 1067 error = sbwait(&so->so_rcv); 1068 splx(s); 1069 if (error) 1070 return (error); 1071 goto restart; 1072 } 1073 dontblock: 1074 /* 1075 * On entry here, m points to the first record of the socket buffer. 1076 * While we process the initial mbufs containing address and control 1077 * info, we save a copy of m->m_nextpkt into nextrecord. 1078 */ 1079 if (l) 1080 l->l_proc->p_stats->p_ru.ru_msgrcv++; 1081 KASSERT(m == so->so_rcv.sb_mb); 1082 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1083 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1084 nextrecord = m->m_nextpkt; 1085 if (pr->pr_flags & PR_ADDR) { 1086 #ifdef DIAGNOSTIC 1087 if (m->m_type != MT_SONAME) 1088 panic("receive 1a"); 1089 #endif 1090 orig_resid = 0; 1091 if (flags & MSG_PEEK) { 1092 if (paddr) 1093 *paddr = m_copy(m, 0, m->m_len); 1094 m = m->m_next; 1095 } else { 1096 sbfree(&so->so_rcv, m); 1097 mbuf_removed = 1; 1098 if (paddr) { 1099 *paddr = m; 1100 so->so_rcv.sb_mb = m->m_next; 1101 m->m_next = 0; 1102 m = so->so_rcv.sb_mb; 1103 } else { 1104 MFREE(m, so->so_rcv.sb_mb); 1105 m = so->so_rcv.sb_mb; 1106 } 1107 } 1108 } 1109 while (m && m->m_type == MT_CONTROL && error == 0) { 1110 if (flags & MSG_PEEK) { 1111 if (controlp) 1112 *controlp = m_copy(m, 0, m->m_len); 1113 m = m->m_next; 1114 } else { 1115 sbfree(&so->so_rcv, m); 1116 mbuf_removed = 1; 1117 if (controlp) { 1118 struct domain *dom = pr->pr_domain; 1119 if (dom->dom_externalize && l && 1120 mtod(m, struct cmsghdr *)->cmsg_type == 1121 SCM_RIGHTS) 1122 error = (*dom->dom_externalize)(m, l); 1123 *controlp = m; 1124 so->so_rcv.sb_mb = m->m_next; 1125 m->m_next = 0; 1126 m = so->so_rcv.sb_mb; 1127 } else { 1128 /* 1129 * Dispose of any SCM_RIGHTS message that went 1130 * through the read path rather than recv. 1131 */ 1132 if (pr->pr_domain->dom_dispose && 1133 mtod(m, struct cmsghdr *)->cmsg_type == SCM_RIGHTS) 1134 (*pr->pr_domain->dom_dispose)(m); 1135 MFREE(m, so->so_rcv.sb_mb); 1136 m = so->so_rcv.sb_mb; 1137 } 1138 } 1139 if (controlp) { 1140 orig_resid = 0; 1141 controlp = &(*controlp)->m_next; 1142 } 1143 } 1144 1145 /* 1146 * If m is non-NULL, we have some data to read. From now on, 1147 * make sure to keep sb_lastrecord consistent when working on 1148 * the last packet on the chain (nextrecord == NULL) and we 1149 * change m->m_nextpkt. 1150 */ 1151 if (m) { 1152 if ((flags & MSG_PEEK) == 0) { 1153 m->m_nextpkt = nextrecord; 1154 /* 1155 * If nextrecord == NULL (this is a single chain), 1156 * then sb_lastrecord may not be valid here if m 1157 * was changed earlier. 1158 */ 1159 if (nextrecord == NULL) { 1160 KASSERT(so->so_rcv.sb_mb == m); 1161 so->so_rcv.sb_lastrecord = m; 1162 } 1163 } 1164 type = m->m_type; 1165 if (type == MT_OOBDATA) 1166 flags |= MSG_OOB; 1167 } else { 1168 if ((flags & MSG_PEEK) == 0) { 1169 KASSERT(so->so_rcv.sb_mb == m); 1170 so->so_rcv.sb_mb = nextrecord; 1171 SB_EMPTY_FIXUP(&so->so_rcv); 1172 } 1173 } 1174 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1175 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1176 1177 moff = 0; 1178 offset = 0; 1179 while (m && uio->uio_resid > 0 && error == 0) { 1180 if (m->m_type == MT_OOBDATA) { 1181 if (type != MT_OOBDATA) 1182 break; 1183 } else if (type == MT_OOBDATA) 1184 break; 1185 #ifdef DIAGNOSTIC 1186 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1187 panic("receive 3"); 1188 #endif 1189 so->so_state &= ~SS_RCVATMARK; 1190 len = uio->uio_resid; 1191 if (so->so_oobmark && len > so->so_oobmark - offset) 1192 len = so->so_oobmark - offset; 1193 if (len > m->m_len - moff) 1194 len = m->m_len - moff; 1195 /* 1196 * If mp is set, just pass back the mbufs. 1197 * Otherwise copy them out via the uio, then free. 1198 * Sockbuf must be consistent here (points to current mbuf, 1199 * it points to next record) when we drop priority; 1200 * we must note any additions to the sockbuf when we 1201 * block interrupts again. 1202 */ 1203 if (mp == 0) { 1204 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1205 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1206 splx(s); 1207 error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio); 1208 s = splsoftnet(); 1209 if (error) { 1210 /* 1211 * If any part of the record has been removed 1212 * (such as the MT_SONAME mbuf, which will 1213 * happen when PR_ADDR, and thus also 1214 * PR_ATOMIC, is set), then drop the entire 1215 * record to maintain the atomicity of the 1216 * receive operation. 1217 * 1218 * This avoids a later panic("receive 1a") 1219 * when compiled with DIAGNOSTIC. 1220 */ 1221 if (m && mbuf_removed 1222 && (pr->pr_flags & PR_ATOMIC)) 1223 (void) sbdroprecord(&so->so_rcv); 1224 1225 goto release; 1226 } 1227 } else 1228 uio->uio_resid -= len; 1229 if (len == m->m_len - moff) { 1230 if (m->m_flags & M_EOR) 1231 flags |= MSG_EOR; 1232 if (flags & MSG_PEEK) { 1233 m = m->m_next; 1234 moff = 0; 1235 } else { 1236 nextrecord = m->m_nextpkt; 1237 sbfree(&so->so_rcv, m); 1238 if (mp) { 1239 *mp = m; 1240 mp = &m->m_next; 1241 so->so_rcv.sb_mb = m = m->m_next; 1242 *mp = (struct mbuf *)0; 1243 } else { 1244 MFREE(m, so->so_rcv.sb_mb); 1245 m = so->so_rcv.sb_mb; 1246 } 1247 /* 1248 * If m != NULL, we also know that 1249 * so->so_rcv.sb_mb != NULL. 1250 */ 1251 KASSERT(so->so_rcv.sb_mb == m); 1252 if (m) { 1253 m->m_nextpkt = nextrecord; 1254 if (nextrecord == NULL) 1255 so->so_rcv.sb_lastrecord = m; 1256 } else { 1257 so->so_rcv.sb_mb = nextrecord; 1258 SB_EMPTY_FIXUP(&so->so_rcv); 1259 } 1260 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1261 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1262 } 1263 } else { 1264 if (flags & MSG_PEEK) 1265 moff += len; 1266 else { 1267 if (mp) 1268 *mp = m_copym(m, 0, len, M_WAIT); 1269 m->m_data += len; 1270 m->m_len -= len; 1271 so->so_rcv.sb_cc -= len; 1272 } 1273 } 1274 if (so->so_oobmark) { 1275 if ((flags & MSG_PEEK) == 0) { 1276 so->so_oobmark -= len; 1277 if (so->so_oobmark == 0) { 1278 so->so_state |= SS_RCVATMARK; 1279 break; 1280 } 1281 } else { 1282 offset += len; 1283 if (offset == so->so_oobmark) 1284 break; 1285 } 1286 } 1287 if (flags & MSG_EOR) 1288 break; 1289 /* 1290 * If the MSG_WAITALL flag is set (for non-atomic socket), 1291 * we must not quit until "uio->uio_resid == 0" or an error 1292 * termination. If a signal/timeout occurs, return 1293 * with a short count but without error. 1294 * Keep sockbuf locked against other readers. 1295 */ 1296 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1297 !sosendallatonce(so) && !nextrecord) { 1298 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1299 break; 1300 /* 1301 * If we are peeking and the socket receive buffer is 1302 * full, stop since we can't get more data to peek at. 1303 */ 1304 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1305 break; 1306 /* 1307 * If we've drained the socket buffer, tell the 1308 * protocol in case it needs to do something to 1309 * get it filled again. 1310 */ 1311 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1312 (*pr->pr_usrreq)(so, PRU_RCVD, 1313 (struct mbuf *)0, 1314 (struct mbuf *)(long)flags, 1315 (struct mbuf *)0, l); 1316 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1317 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1318 error = sbwait(&so->so_rcv); 1319 if (error) { 1320 sbunlock(&so->so_rcv); 1321 splx(s); 1322 return (0); 1323 } 1324 if ((m = so->so_rcv.sb_mb) != NULL) 1325 nextrecord = m->m_nextpkt; 1326 } 1327 } 1328 1329 if (m && pr->pr_flags & PR_ATOMIC) { 1330 flags |= MSG_TRUNC; 1331 if ((flags & MSG_PEEK) == 0) 1332 (void) sbdroprecord(&so->so_rcv); 1333 } 1334 if ((flags & MSG_PEEK) == 0) { 1335 if (m == 0) { 1336 /* 1337 * First part is an inline SB_EMPTY_FIXUP(). Second 1338 * part makes sure sb_lastrecord is up-to-date if 1339 * there is still data in the socket buffer. 1340 */ 1341 so->so_rcv.sb_mb = nextrecord; 1342 if (so->so_rcv.sb_mb == NULL) { 1343 so->so_rcv.sb_mbtail = NULL; 1344 so->so_rcv.sb_lastrecord = NULL; 1345 } else if (nextrecord->m_nextpkt == NULL) 1346 so->so_rcv.sb_lastrecord = nextrecord; 1347 } 1348 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1349 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1350 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1351 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1352 (struct mbuf *)(long)flags, (struct mbuf *)0, l); 1353 } 1354 if (orig_resid == uio->uio_resid && orig_resid && 1355 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1356 sbunlock(&so->so_rcv); 1357 splx(s); 1358 goto restart; 1359 } 1360 1361 if (flagsp) 1362 *flagsp |= flags; 1363 release: 1364 sbunlock(&so->so_rcv); 1365 splx(s); 1366 return (error); 1367 } 1368 1369 int 1370 soshutdown(struct socket *so, int how) 1371 { 1372 const struct protosw *pr; 1373 1374 pr = so->so_proto; 1375 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1376 return (EINVAL); 1377 1378 if (how == SHUT_RD || how == SHUT_RDWR) 1379 sorflush(so); 1380 if (how == SHUT_WR || how == SHUT_RDWR) 1381 return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0, 1382 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1383 return (0); 1384 } 1385 1386 void 1387 sorflush(struct socket *so) 1388 { 1389 struct sockbuf *sb, asb; 1390 const struct protosw *pr; 1391 int s; 1392 1393 sb = &so->so_rcv; 1394 pr = so->so_proto; 1395 sb->sb_flags |= SB_NOINTR; 1396 (void) sblock(sb, M_WAITOK); 1397 s = splnet(); 1398 socantrcvmore(so); 1399 sbunlock(sb); 1400 asb = *sb; 1401 /* 1402 * Clear most of the sockbuf structure, but leave some of the 1403 * fields valid. 1404 */ 1405 memset(&sb->sb_startzero, 0, 1406 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1407 splx(s); 1408 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1409 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1410 sbrelease(&asb, so); 1411 } 1412 1413 int 1414 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0) 1415 { 1416 int error; 1417 struct mbuf *m; 1418 1419 error = 0; 1420 m = m0; 1421 if (level != SOL_SOCKET) { 1422 if (so->so_proto && so->so_proto->pr_ctloutput) 1423 return ((*so->so_proto->pr_ctloutput) 1424 (PRCO_SETOPT, so, level, optname, &m0)); 1425 error = ENOPROTOOPT; 1426 } else { 1427 switch (optname) { 1428 1429 case SO_LINGER: 1430 if (m == NULL || m->m_len != sizeof(struct linger)) { 1431 error = EINVAL; 1432 goto bad; 1433 } 1434 if (mtod(m, struct linger *)->l_linger < 0 || 1435 mtod(m, struct linger *)->l_linger > (INT_MAX / hz)) { 1436 error = EDOM; 1437 goto bad; 1438 } 1439 so->so_linger = mtod(m, struct linger *)->l_linger; 1440 /* fall thru... */ 1441 1442 case SO_DEBUG: 1443 case SO_KEEPALIVE: 1444 case SO_DONTROUTE: 1445 case SO_USELOOPBACK: 1446 case SO_BROADCAST: 1447 case SO_REUSEADDR: 1448 case SO_REUSEPORT: 1449 case SO_OOBINLINE: 1450 case SO_TIMESTAMP: 1451 if (m == NULL || m->m_len < sizeof(int)) { 1452 error = EINVAL; 1453 goto bad; 1454 } 1455 if (*mtod(m, int *)) 1456 so->so_options |= optname; 1457 else 1458 so->so_options &= ~optname; 1459 break; 1460 1461 case SO_SNDBUF: 1462 case SO_RCVBUF: 1463 case SO_SNDLOWAT: 1464 case SO_RCVLOWAT: 1465 { 1466 int optval; 1467 1468 if (m == NULL || m->m_len < sizeof(int)) { 1469 error = EINVAL; 1470 goto bad; 1471 } 1472 1473 /* 1474 * Values < 1 make no sense for any of these 1475 * options, so disallow them. 1476 */ 1477 optval = *mtod(m, int *); 1478 if (optval < 1) { 1479 error = EINVAL; 1480 goto bad; 1481 } 1482 1483 switch (optname) { 1484 1485 case SO_SNDBUF: 1486 case SO_RCVBUF: 1487 if (sbreserve(optname == SO_SNDBUF ? 1488 &so->so_snd : &so->so_rcv, 1489 (u_long) optval, so) == 0) { 1490 error = ENOBUFS; 1491 goto bad; 1492 } 1493 break; 1494 1495 /* 1496 * Make sure the low-water is never greater than 1497 * the high-water. 1498 */ 1499 case SO_SNDLOWAT: 1500 so->so_snd.sb_lowat = 1501 (optval > so->so_snd.sb_hiwat) ? 1502 so->so_snd.sb_hiwat : optval; 1503 break; 1504 case SO_RCVLOWAT: 1505 so->so_rcv.sb_lowat = 1506 (optval > so->so_rcv.sb_hiwat) ? 1507 so->so_rcv.sb_hiwat : optval; 1508 break; 1509 } 1510 break; 1511 } 1512 1513 case SO_SNDTIMEO: 1514 case SO_RCVTIMEO: 1515 { 1516 struct timeval *tv; 1517 int val; 1518 1519 if (m == NULL || m->m_len < sizeof(*tv)) { 1520 error = EINVAL; 1521 goto bad; 1522 } 1523 tv = mtod(m, struct timeval *); 1524 if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz) { 1525 error = EDOM; 1526 goto bad; 1527 } 1528 val = tv->tv_sec * hz + tv->tv_usec / tick; 1529 if (val == 0 && tv->tv_usec != 0) 1530 val = 1; 1531 1532 switch (optname) { 1533 1534 case SO_SNDTIMEO: 1535 so->so_snd.sb_timeo = val; 1536 break; 1537 case SO_RCVTIMEO: 1538 so->so_rcv.sb_timeo = val; 1539 break; 1540 } 1541 break; 1542 } 1543 1544 default: 1545 error = ENOPROTOOPT; 1546 break; 1547 } 1548 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1549 (void) ((*so->so_proto->pr_ctloutput) 1550 (PRCO_SETOPT, so, level, optname, &m0)); 1551 m = NULL; /* freed by protocol */ 1552 } 1553 } 1554 bad: 1555 if (m) 1556 (void) m_free(m); 1557 return (error); 1558 } 1559 1560 int 1561 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp) 1562 { 1563 struct mbuf *m; 1564 1565 if (level != SOL_SOCKET) { 1566 if (so->so_proto && so->so_proto->pr_ctloutput) { 1567 return ((*so->so_proto->pr_ctloutput) 1568 (PRCO_GETOPT, so, level, optname, mp)); 1569 } else 1570 return (ENOPROTOOPT); 1571 } else { 1572 m = m_get(M_WAIT, MT_SOOPTS); 1573 m->m_len = sizeof(int); 1574 1575 switch (optname) { 1576 1577 case SO_LINGER: 1578 m->m_len = sizeof(struct linger); 1579 mtod(m, struct linger *)->l_onoff = 1580 so->so_options & SO_LINGER; 1581 mtod(m, struct linger *)->l_linger = so->so_linger; 1582 break; 1583 1584 case SO_USELOOPBACK: 1585 case SO_DONTROUTE: 1586 case SO_DEBUG: 1587 case SO_KEEPALIVE: 1588 case SO_REUSEADDR: 1589 case SO_REUSEPORT: 1590 case SO_BROADCAST: 1591 case SO_OOBINLINE: 1592 case SO_TIMESTAMP: 1593 *mtod(m, int *) = so->so_options & optname; 1594 break; 1595 1596 case SO_TYPE: 1597 *mtod(m, int *) = so->so_type; 1598 break; 1599 1600 case SO_ERROR: 1601 *mtod(m, int *) = so->so_error; 1602 so->so_error = 0; 1603 break; 1604 1605 case SO_SNDBUF: 1606 *mtod(m, int *) = so->so_snd.sb_hiwat; 1607 break; 1608 1609 case SO_RCVBUF: 1610 *mtod(m, int *) = so->so_rcv.sb_hiwat; 1611 break; 1612 1613 case SO_SNDLOWAT: 1614 *mtod(m, int *) = so->so_snd.sb_lowat; 1615 break; 1616 1617 case SO_RCVLOWAT: 1618 *mtod(m, int *) = so->so_rcv.sb_lowat; 1619 break; 1620 1621 case SO_SNDTIMEO: 1622 case SO_RCVTIMEO: 1623 { 1624 int val = (optname == SO_SNDTIMEO ? 1625 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1626 1627 m->m_len = sizeof(struct timeval); 1628 mtod(m, struct timeval *)->tv_sec = val / hz; 1629 mtod(m, struct timeval *)->tv_usec = 1630 (val % hz) * tick; 1631 break; 1632 } 1633 1634 case SO_OVERFLOWED: 1635 *mtod(m, int *) = so->so_rcv.sb_overflowed; 1636 break; 1637 1638 default: 1639 (void)m_free(m); 1640 return (ENOPROTOOPT); 1641 } 1642 *mp = m; 1643 return (0); 1644 } 1645 } 1646 1647 void 1648 sohasoutofband(struct socket *so) 1649 { 1650 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 1651 selwakeup(&so->so_rcv.sb_sel); 1652 } 1653 1654 static void 1655 filt_sordetach(struct knote *kn) 1656 { 1657 struct socket *so; 1658 1659 so = (struct socket *)kn->kn_fp->f_data; 1660 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 1661 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 1662 so->so_rcv.sb_flags &= ~SB_KNOTE; 1663 } 1664 1665 /*ARGSUSED*/ 1666 static int 1667 filt_soread(struct knote *kn, long hint) 1668 { 1669 struct socket *so; 1670 1671 so = (struct socket *)kn->kn_fp->f_data; 1672 kn->kn_data = so->so_rcv.sb_cc; 1673 if (so->so_state & SS_CANTRCVMORE) { 1674 kn->kn_flags |= EV_EOF; 1675 kn->kn_fflags = so->so_error; 1676 return (1); 1677 } 1678 if (so->so_error) /* temporary udp error */ 1679 return (1); 1680 if (kn->kn_sfflags & NOTE_LOWAT) 1681 return (kn->kn_data >= kn->kn_sdata); 1682 return (kn->kn_data >= so->so_rcv.sb_lowat); 1683 } 1684 1685 static void 1686 filt_sowdetach(struct knote *kn) 1687 { 1688 struct socket *so; 1689 1690 so = (struct socket *)kn->kn_fp->f_data; 1691 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 1692 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 1693 so->so_snd.sb_flags &= ~SB_KNOTE; 1694 } 1695 1696 /*ARGSUSED*/ 1697 static int 1698 filt_sowrite(struct knote *kn, long hint) 1699 { 1700 struct socket *so; 1701 1702 so = (struct socket *)kn->kn_fp->f_data; 1703 kn->kn_data = sbspace(&so->so_snd); 1704 if (so->so_state & SS_CANTSENDMORE) { 1705 kn->kn_flags |= EV_EOF; 1706 kn->kn_fflags = so->so_error; 1707 return (1); 1708 } 1709 if (so->so_error) /* temporary udp error */ 1710 return (1); 1711 if (((so->so_state & SS_ISCONNECTED) == 0) && 1712 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1713 return (0); 1714 if (kn->kn_sfflags & NOTE_LOWAT) 1715 return (kn->kn_data >= kn->kn_sdata); 1716 return (kn->kn_data >= so->so_snd.sb_lowat); 1717 } 1718 1719 /*ARGSUSED*/ 1720 static int 1721 filt_solisten(struct knote *kn, long hint) 1722 { 1723 struct socket *so; 1724 1725 so = (struct socket *)kn->kn_fp->f_data; 1726 1727 /* 1728 * Set kn_data to number of incoming connections, not 1729 * counting partial (incomplete) connections. 1730 */ 1731 kn->kn_data = so->so_qlen; 1732 return (kn->kn_data > 0); 1733 } 1734 1735 static const struct filterops solisten_filtops = 1736 { 1, NULL, filt_sordetach, filt_solisten }; 1737 static const struct filterops soread_filtops = 1738 { 1, NULL, filt_sordetach, filt_soread }; 1739 static const struct filterops sowrite_filtops = 1740 { 1, NULL, filt_sowdetach, filt_sowrite }; 1741 1742 int 1743 soo_kqfilter(struct file *fp, struct knote *kn) 1744 { 1745 struct socket *so; 1746 struct sockbuf *sb; 1747 1748 so = (struct socket *)kn->kn_fp->f_data; 1749 switch (kn->kn_filter) { 1750 case EVFILT_READ: 1751 if (so->so_options & SO_ACCEPTCONN) 1752 kn->kn_fop = &solisten_filtops; 1753 else 1754 kn->kn_fop = &soread_filtops; 1755 sb = &so->so_rcv; 1756 break; 1757 case EVFILT_WRITE: 1758 kn->kn_fop = &sowrite_filtops; 1759 sb = &so->so_snd; 1760 break; 1761 default: 1762 return (1); 1763 } 1764 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 1765 sb->sb_flags |= SB_KNOTE; 1766 return (0); 1767 } 1768 1769 #include <sys/sysctl.h> 1770 1771 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 1772 1773 /* 1774 * sysctl helper routine for kern.somaxkva. ensures that the given 1775 * value is not too small. 1776 * (XXX should we maybe make sure it's not too large as well?) 1777 */ 1778 static int 1779 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 1780 { 1781 int error, new_somaxkva; 1782 struct sysctlnode node; 1783 int s; 1784 1785 new_somaxkva = somaxkva; 1786 node = *rnode; 1787 node.sysctl_data = &new_somaxkva; 1788 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1789 if (error || newp == NULL) 1790 return (error); 1791 1792 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 1793 return (EINVAL); 1794 1795 s = splvm(); 1796 simple_lock(&so_pendfree_slock); 1797 somaxkva = new_somaxkva; 1798 wakeup(&socurkva); 1799 simple_unlock(&so_pendfree_slock); 1800 splx(s); 1801 1802 return (error); 1803 } 1804 1805 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup") 1806 { 1807 1808 sysctl_createv(clog, 0, NULL, NULL, 1809 CTLFLAG_PERMANENT, 1810 CTLTYPE_NODE, "kern", NULL, 1811 NULL, 0, NULL, 0, 1812 CTL_KERN, CTL_EOL); 1813 1814 sysctl_createv(clog, 0, NULL, NULL, 1815 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1816 CTLTYPE_INT, "somaxkva", 1817 SYSCTL_DESCR("Maximum amount of kernel memory to be " 1818 "used for socket buffers"), 1819 sysctl_kern_somaxkva, 0, NULL, 0, 1820 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 1821 } 1822
Indexes created Sat Sep 20 22:09:52 GMT 2025