uipc_socket.c revision 1.136.2.1
1 /* $NetBSD: uipc_socket.c,v 1.136.2.1 2007/03/18 00:06:45 reinoud Exp $ */ 2 3 /*- 4 * Copyright (c) 2002, 2007 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 1.136.2.1 2007/03/18 00:06:45 reinoud 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 #include <sys/mutex.h> 96 #include <sys/condvar.h> 97 98 #include <uvm/uvm.h> 99 100 POOL_INIT(socket_pool, sizeof(struct socket), 0, 0, 0, "sockpl", NULL, 101 IPL_SOFTNET); 102 103 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options"); 104 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 105 106 extern int somaxconn; /* patchable (XXX sysctl) */ 107 int somaxconn = SOMAXCONN; 108 109 #ifdef SOSEND_COUNTERS 110 #include <sys/device.h> 111 112 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 113 NULL, "sosend", "loan big"); 114 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 115 NULL, "sosend", "copy big"); 116 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 117 NULL, "sosend", "copy small"); 118 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 119 NULL, "sosend", "kva limit"); 120 121 #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++ 122 123 EVCNT_ATTACH_STATIC(sosend_loan_big); 124 EVCNT_ATTACH_STATIC(sosend_copy_big); 125 EVCNT_ATTACH_STATIC(sosend_copy_small); 126 EVCNT_ATTACH_STATIC(sosend_kvalimit); 127 #else 128 129 #define SOSEND_COUNTER_INCR(ev) /* nothing */ 130 131 #endif /* SOSEND_COUNTERS */ 132 133 static struct callback_entry sokva_reclaimerentry; 134 135 #ifdef SOSEND_NO_LOAN 136 int sock_loan_thresh = -1; 137 #else 138 int sock_loan_thresh = 4096; 139 #endif 140 141 static kmutex_t so_pendfree_lock; 142 static struct mbuf *so_pendfree; 143 144 #ifndef SOMAXKVA 145 #define SOMAXKVA (16 * 1024 * 1024) 146 #endif 147 int somaxkva = SOMAXKVA; 148 static int socurkva; 149 static kcondvar_t socurkva_cv; 150 151 #define SOCK_LOAN_CHUNK 65536 152 153 static size_t sodopendfree(void); 154 static size_t sodopendfreel(void); 155 156 static vsize_t 157 sokvareserve(struct socket *so, vsize_t len) 158 { 159 int error; 160 161 mutex_enter(&so_pendfree_lock); 162 while (socurkva + len > somaxkva) { 163 size_t freed; 164 165 /* 166 * try to do pendfree. 167 */ 168 169 freed = sodopendfreel(); 170 171 /* 172 * if some kva was freed, try again. 173 */ 174 175 if (freed) 176 continue; 177 178 SOSEND_COUNTER_INCR(&sosend_kvalimit); 179 error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock); 180 if (error) { 181 len = 0; 182 break; 183 } 184 } 185 socurkva += len; 186 mutex_exit(&so_pendfree_lock); 187 return len; 188 } 189 190 static void 191 sokvaunreserve(vsize_t len) 192 { 193 194 mutex_enter(&so_pendfree_lock); 195 socurkva -= len; 196 cv_broadcast(&socurkva_cv); 197 mutex_exit(&so_pendfree_lock); 198 } 199 200 /* 201 * sokvaalloc: allocate kva for loan. 202 */ 203 204 vaddr_t 205 sokvaalloc(vsize_t len, struct socket *so) 206 { 207 vaddr_t lva; 208 209 /* 210 * reserve kva. 211 */ 212 213 if (sokvareserve(so, len) == 0) 214 return 0; 215 216 /* 217 * allocate kva. 218 */ 219 220 lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA); 221 if (lva == 0) { 222 sokvaunreserve(len); 223 return (0); 224 } 225 226 return lva; 227 } 228 229 /* 230 * sokvafree: free kva for loan. 231 */ 232 233 void 234 sokvafree(vaddr_t sva, vsize_t len) 235 { 236 237 /* 238 * free kva. 239 */ 240 241 uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY); 242 243 /* 244 * unreserve kva. 245 */ 246 247 sokvaunreserve(len); 248 } 249 250 static void 251 sodoloanfree(struct vm_page **pgs, void *buf, size_t size) 252 { 253 vaddr_t va, sva, eva; 254 vsize_t len; 255 paddr_t pa; 256 int i, npgs; 257 258 eva = round_page((vaddr_t) buf + size); 259 sva = trunc_page((vaddr_t) buf); 260 len = eva - sva; 261 npgs = len >> PAGE_SHIFT; 262 263 if (__predict_false(pgs == NULL)) { 264 pgs = alloca(npgs * sizeof(*pgs)); 265 266 for (i = 0, va = sva; va < eva; i++, va += PAGE_SIZE) { 267 if (pmap_extract(pmap_kernel(), va, &pa) == false) 268 panic("sodoloanfree: va 0x%lx not mapped", va); 269 pgs[i] = PHYS_TO_VM_PAGE(pa); 270 } 271 } 272 273 pmap_kremove(sva, len); 274 pmap_update(pmap_kernel()); 275 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 276 sokvafree(sva, len); 277 } 278 279 static size_t 280 sodopendfree() 281 { 282 size_t rv; 283 284 mutex_enter(&so_pendfree_lock); 285 rv = sodopendfreel(); 286 mutex_exit(&so_pendfree_lock); 287 288 return rv; 289 } 290 291 /* 292 * sodopendfreel: free mbufs on "pendfree" list. 293 * unlock and relock so_pendfree_lock when freeing mbufs. 294 * 295 * => called with so_pendfree_lock held. 296 */ 297 298 static size_t 299 sodopendfreel() 300 { 301 struct mbuf *m, *next; 302 size_t rv = 0; 303 int s; 304 305 KASSERT(mutex_owned(&so_pendfree_lock)); 306 307 while (so_pendfree != NULL) { 308 m = so_pendfree; 309 so_pendfree = NULL; 310 mutex_exit(&so_pendfree_lock); 311 312 for (; m != NULL; m = next) { 313 next = m->m_next; 314 315 rv += m->m_ext.ext_size; 316 sodoloanfree((m->m_flags & M_EXT_PAGES) ? 317 m->m_ext.ext_pgs : NULL, m->m_ext.ext_buf, 318 m->m_ext.ext_size); 319 s = splvm(); 320 pool_cache_put(&mbpool_cache, m); 321 splx(s); 322 } 323 324 mutex_enter(&so_pendfree_lock); 325 } 326 327 return (rv); 328 } 329 330 void 331 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg) 332 { 333 334 if (m == NULL) { 335 336 /* 337 * called from MEXTREMOVE. 338 */ 339 340 sodoloanfree(NULL, buf, size); 341 return; 342 } 343 344 /* 345 * postpone freeing mbuf. 346 * 347 * we can't do it in interrupt context 348 * because we need to put kva back to kernel_map. 349 */ 350 351 mutex_enter(&so_pendfree_lock); 352 m->m_next = so_pendfree; 353 so_pendfree = m; 354 cv_broadcast(&socurkva_cv); 355 mutex_exit(&so_pendfree_lock); 356 } 357 358 static long 359 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 360 { 361 struct iovec *iov = uio->uio_iov; 362 vaddr_t sva, eva; 363 vsize_t len; 364 vaddr_t lva, va; 365 int npgs, i, error; 366 367 if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace)) 368 return (0); 369 370 if (iov->iov_len < (size_t) space) 371 space = iov->iov_len; 372 if (space > SOCK_LOAN_CHUNK) 373 space = SOCK_LOAN_CHUNK; 374 375 eva = round_page((vaddr_t) iov->iov_base + space); 376 sva = trunc_page((vaddr_t) iov->iov_base); 377 len = eva - sva; 378 npgs = len >> PAGE_SHIFT; 379 380 /* XXX KDASSERT */ 381 KASSERT(npgs <= M_EXT_MAXPAGES); 382 383 lva = sokvaalloc(len, so); 384 if (lva == 0) 385 return 0; 386 387 error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len, 388 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 389 if (error) { 390 sokvafree(lva, len); 391 return (0); 392 } 393 394 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 395 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 396 VM_PROT_READ); 397 pmap_update(pmap_kernel()); 398 399 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 400 401 MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so); 402 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 403 404 uio->uio_resid -= space; 405 /* uio_offset not updated, not set/used for write(2) */ 406 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space; 407 uio->uio_iov->iov_len -= space; 408 if (uio->uio_iov->iov_len == 0) { 409 uio->uio_iov++; 410 uio->uio_iovcnt--; 411 } 412 413 return (space); 414 } 415 416 static int 417 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg) 418 { 419 420 KASSERT(ce == &sokva_reclaimerentry); 421 KASSERT(obj == NULL); 422 423 sodopendfree(); 424 if (!vm_map_starved_p(kernel_map)) { 425 return CALLBACK_CHAIN_ABORT; 426 } 427 return CALLBACK_CHAIN_CONTINUE; 428 } 429 430 void 431 soinit(void) 432 { 433 434 mutex_init(&so_pendfree_lock, MUTEX_DRIVER, IPL_VM); 435 cv_init(&socurkva_cv, "sokva"); 436 437 /* Set the initial adjusted socket buffer size. */ 438 if (sb_max_set(sb_max)) 439 panic("bad initial sb_max value: %lu", sb_max); 440 441 callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback, 442 &sokva_reclaimerentry, NULL, sokva_reclaim_callback); 443 } 444 445 /* 446 * Socket operation routines. 447 * These routines are called by the routines in 448 * sys_socket.c or from a system process, and 449 * implement the semantics of socket operations by 450 * switching out to the protocol specific routines. 451 */ 452 /*ARGSUSED*/ 453 int 454 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l) 455 { 456 const struct protosw *prp; 457 struct socket *so; 458 uid_t uid; 459 int error, s; 460 461 error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET, 462 KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type), 463 KAUTH_ARG(proto)); 464 if (error) 465 return (error); 466 467 if (proto) 468 prp = pffindproto(dom, proto, type); 469 else 470 prp = pffindtype(dom, type); 471 if (prp == 0) { 472 /* no support for domain */ 473 if (pffinddomain(dom) == 0) 474 return (EAFNOSUPPORT); 475 /* no support for socket type */ 476 if (proto == 0 && type != 0) 477 return (EPROTOTYPE); 478 return (EPROTONOSUPPORT); 479 } 480 if (prp->pr_usrreq == 0) 481 return (EPROTONOSUPPORT); 482 if (prp->pr_type != type) 483 return (EPROTOTYPE); 484 s = splsoftnet(); 485 so = pool_get(&socket_pool, PR_WAITOK); 486 memset((void *)so, 0, sizeof(*so)); 487 TAILQ_INIT(&so->so_q0); 488 TAILQ_INIT(&so->so_q); 489 so->so_type = type; 490 so->so_proto = prp; 491 so->so_send = sosend; 492 so->so_receive = soreceive; 493 #ifdef MBUFTRACE 494 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 495 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 496 so->so_mowner = &prp->pr_domain->dom_mowner; 497 #endif 498 if (l != NULL) { 499 uid = kauth_cred_geteuid(l->l_cred); 500 } else { 501 uid = 0; 502 } 503 so->so_uidinfo = uid_find(uid); 504 error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0, 505 (struct mbuf *)(long)proto, (struct mbuf *)0, l); 506 if (error) { 507 so->so_state |= SS_NOFDREF; 508 sofree(so); 509 splx(s); 510 return (error); 511 } 512 splx(s); 513 *aso = so; 514 return (0); 515 } 516 517 int 518 sobind(struct socket *so, struct mbuf *nam, struct lwp *l) 519 { 520 int s, error; 521 522 s = splsoftnet(); 523 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0, 524 nam, (struct mbuf *)0, l); 525 splx(s); 526 return (error); 527 } 528 529 int 530 solisten(struct socket *so, int backlog) 531 { 532 int s, error; 533 534 s = splsoftnet(); 535 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0, 536 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 537 if (error) { 538 splx(s); 539 return (error); 540 } 541 if (TAILQ_EMPTY(&so->so_q)) 542 so->so_options |= SO_ACCEPTCONN; 543 if (backlog < 0) 544 backlog = 0; 545 so->so_qlimit = min(backlog, somaxconn); 546 splx(s); 547 return (0); 548 } 549 550 void 551 sofree(struct socket *so) 552 { 553 554 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 555 return; 556 if (so->so_head) { 557 /* 558 * We must not decommission a socket that's on the accept(2) 559 * queue. If we do, then accept(2) may hang after select(2) 560 * indicated that the listening socket was ready. 561 */ 562 if (!soqremque(so, 0)) 563 return; 564 } 565 if (so->so_rcv.sb_hiwat) 566 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 567 RLIM_INFINITY); 568 if (so->so_snd.sb_hiwat) 569 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 570 RLIM_INFINITY); 571 sbrelease(&so->so_snd, so); 572 sorflush(so); 573 pool_put(&socket_pool, so); 574 } 575 576 /* 577 * Close a socket on last file table reference removal. 578 * Initiate disconnect if connected. 579 * Free socket when disconnect complete. 580 */ 581 int 582 soclose(struct socket *so) 583 { 584 struct socket *so2; 585 int s, error; 586 587 error = 0; 588 s = splsoftnet(); /* conservative */ 589 if (so->so_options & SO_ACCEPTCONN) { 590 while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 591 (void) soqremque(so2, 0); 592 (void) soabort(so2); 593 } 594 while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 595 (void) soqremque(so2, 1); 596 (void) soabort(so2); 597 } 598 } 599 if (so->so_pcb == 0) 600 goto discard; 601 if (so->so_state & SS_ISCONNECTED) { 602 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 603 error = sodisconnect(so); 604 if (error) 605 goto drop; 606 } 607 if (so->so_options & SO_LINGER) { 608 if ((so->so_state & SS_ISDISCONNECTING) && 609 (so->so_state & SS_NBIO)) 610 goto drop; 611 while (so->so_state & SS_ISCONNECTED) { 612 error = tsleep((void *)&so->so_timeo, 613 PSOCK | PCATCH, netcls, 614 so->so_linger * hz); 615 if (error) 616 break; 617 } 618 } 619 } 620 drop: 621 if (so->so_pcb) { 622 int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 623 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 624 (struct lwp *)0); 625 if (error == 0) 626 error = error2; 627 } 628 discard: 629 if (so->so_state & SS_NOFDREF) 630 panic("soclose: NOFDREF"); 631 so->so_state |= SS_NOFDREF; 632 sofree(so); 633 splx(s); 634 return (error); 635 } 636 637 /* 638 * Must be called at splsoftnet... 639 */ 640 int 641 soabort(struct socket *so) 642 { 643 644 return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0, 645 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 646 } 647 648 int 649 soaccept(struct socket *so, struct mbuf *nam) 650 { 651 int s, error; 652 653 error = 0; 654 s = splsoftnet(); 655 if ((so->so_state & SS_NOFDREF) == 0) 656 panic("soaccept: !NOFDREF"); 657 so->so_state &= ~SS_NOFDREF; 658 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 659 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 660 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 661 (struct mbuf *)0, nam, (struct mbuf *)0, (struct lwp *)0); 662 else 663 error = ECONNABORTED; 664 665 splx(s); 666 return (error); 667 } 668 669 int 670 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l) 671 { 672 int s, error; 673 674 if (so->so_options & SO_ACCEPTCONN) 675 return (EOPNOTSUPP); 676 s = splsoftnet(); 677 /* 678 * If protocol is connection-based, can only connect once. 679 * Otherwise, if connected, try to disconnect first. 680 * This allows user to disconnect by connecting to, e.g., 681 * a null address. 682 */ 683 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 684 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 685 (error = sodisconnect(so)))) 686 error = EISCONN; 687 else 688 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 689 (struct mbuf *)0, nam, (struct mbuf *)0, l); 690 splx(s); 691 return (error); 692 } 693 694 int 695 soconnect2(struct socket *so1, struct socket *so2) 696 { 697 int s, error; 698 699 s = splsoftnet(); 700 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 701 (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0, 702 (struct lwp *)0); 703 splx(s); 704 return (error); 705 } 706 707 int 708 sodisconnect(struct socket *so) 709 { 710 int s, error; 711 712 s = splsoftnet(); 713 if ((so->so_state & SS_ISCONNECTED) == 0) { 714 error = ENOTCONN; 715 goto bad; 716 } 717 if (so->so_state & SS_ISDISCONNECTING) { 718 error = EALREADY; 719 goto bad; 720 } 721 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 722 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 723 (struct lwp *)0); 724 bad: 725 splx(s); 726 sodopendfree(); 727 return (error); 728 } 729 730 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 731 /* 732 * Send on a socket. 733 * If send must go all at once and message is larger than 734 * send buffering, then hard error. 735 * Lock against other senders. 736 * If must go all at once and not enough room now, then 737 * inform user that this would block and do nothing. 738 * Otherwise, if nonblocking, send as much as possible. 739 * The data to be sent is described by "uio" if nonzero, 740 * otherwise by the mbuf chain "top" (which must be null 741 * if uio is not). Data provided in mbuf chain must be small 742 * enough to send all at once. 743 * 744 * Returns nonzero on error, timeout or signal; callers 745 * must check for short counts if EINTR/ERESTART are returned. 746 * Data and control buffers are freed on return. 747 */ 748 int 749 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 750 struct mbuf *control, int flags, struct lwp *l) 751 { 752 struct mbuf **mp, *m; 753 struct proc *p; 754 long space, len, resid, clen, mlen; 755 int error, s, dontroute, atomic; 756 757 p = l->l_proc; 758 sodopendfree(); 759 760 clen = 0; 761 atomic = sosendallatonce(so) || top; 762 if (uio) 763 resid = uio->uio_resid; 764 else 765 resid = top->m_pkthdr.len; 766 /* 767 * In theory resid should be unsigned. 768 * However, space must be signed, as it might be less than 0 769 * if we over-committed, and we must use a signed comparison 770 * of space and resid. On the other hand, a negative resid 771 * causes us to loop sending 0-length segments to the protocol. 772 */ 773 if (resid < 0) { 774 error = EINVAL; 775 goto out; 776 } 777 dontroute = 778 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 779 (so->so_proto->pr_flags & PR_ATOMIC); 780 if (p) 781 p->p_stats->p_ru.ru_msgsnd++; 782 if (control) 783 clen = control->m_len; 784 #define snderr(errno) { error = errno; splx(s); goto release; } 785 786 restart: 787 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 788 goto out; 789 do { 790 s = splsoftnet(); 791 if (so->so_state & SS_CANTSENDMORE) 792 snderr(EPIPE); 793 if (so->so_error) { 794 error = so->so_error; 795 so->so_error = 0; 796 splx(s); 797 goto release; 798 } 799 if ((so->so_state & SS_ISCONNECTED) == 0) { 800 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 801 if ((so->so_state & SS_ISCONFIRMING) == 0 && 802 !(resid == 0 && clen != 0)) 803 snderr(ENOTCONN); 804 } else if (addr == 0) 805 snderr(EDESTADDRREQ); 806 } 807 space = sbspace(&so->so_snd); 808 if (flags & MSG_OOB) 809 space += 1024; 810 if ((atomic && resid > so->so_snd.sb_hiwat) || 811 clen > so->so_snd.sb_hiwat) 812 snderr(EMSGSIZE); 813 if (space < resid + clen && 814 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 815 if (so->so_state & SS_NBIO) 816 snderr(EWOULDBLOCK); 817 sbunlock(&so->so_snd); 818 error = sbwait(&so->so_snd); 819 splx(s); 820 if (error) 821 goto out; 822 goto restart; 823 } 824 splx(s); 825 mp = ⊤ 826 space -= clen; 827 do { 828 if (uio == NULL) { 829 /* 830 * Data is prepackaged in "top". 831 */ 832 resid = 0; 833 if (flags & MSG_EOR) 834 top->m_flags |= M_EOR; 835 } else do { 836 if (top == 0) { 837 m = m_gethdr(M_WAIT, MT_DATA); 838 mlen = MHLEN; 839 m->m_pkthdr.len = 0; 840 m->m_pkthdr.rcvif = (struct ifnet *)0; 841 } else { 842 m = m_get(M_WAIT, MT_DATA); 843 mlen = MLEN; 844 } 845 MCLAIM(m, so->so_snd.sb_mowner); 846 if (sock_loan_thresh >= 0 && 847 uio->uio_iov->iov_len >= sock_loan_thresh && 848 space >= sock_loan_thresh && 849 (len = sosend_loan(so, uio, m, 850 space)) != 0) { 851 SOSEND_COUNTER_INCR(&sosend_loan_big); 852 space -= len; 853 goto have_data; 854 } 855 if (resid >= MINCLSIZE && space >= MCLBYTES) { 856 SOSEND_COUNTER_INCR(&sosend_copy_big); 857 m_clget(m, M_WAIT); 858 if ((m->m_flags & M_EXT) == 0) 859 goto nopages; 860 mlen = MCLBYTES; 861 if (atomic && top == 0) { 862 len = lmin(MCLBYTES - max_hdr, 863 resid); 864 m->m_data += max_hdr; 865 } else 866 len = lmin(MCLBYTES, resid); 867 space -= len; 868 } else { 869 nopages: 870 SOSEND_COUNTER_INCR(&sosend_copy_small); 871 len = lmin(lmin(mlen, resid), space); 872 space -= len; 873 /* 874 * For datagram protocols, leave room 875 * for protocol headers in first mbuf. 876 */ 877 if (atomic && top == 0 && len < mlen) 878 MH_ALIGN(m, len); 879 } 880 error = uiomove(mtod(m, void *), (int)len, 881 uio); 882 have_data: 883 resid = uio->uio_resid; 884 m->m_len = len; 885 *mp = m; 886 top->m_pkthdr.len += len; 887 if (error) 888 goto release; 889 mp = &m->m_next; 890 if (resid <= 0) { 891 if (flags & MSG_EOR) 892 top->m_flags |= M_EOR; 893 break; 894 } 895 } while (space > 0 && atomic); 896 897 s = splsoftnet(); 898 899 if (so->so_state & SS_CANTSENDMORE) 900 snderr(EPIPE); 901 902 if (dontroute) 903 so->so_options |= SO_DONTROUTE; 904 if (resid > 0) 905 so->so_state |= SS_MORETOCOME; 906 error = (*so->so_proto->pr_usrreq)(so, 907 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 908 top, addr, control, curlwp); /* XXX */ 909 if (dontroute) 910 so->so_options &= ~SO_DONTROUTE; 911 if (resid > 0) 912 so->so_state &= ~SS_MORETOCOME; 913 splx(s); 914 915 clen = 0; 916 control = 0; 917 top = 0; 918 mp = ⊤ 919 if (error) 920 goto release; 921 } while (resid && space > 0); 922 } while (resid); 923 924 release: 925 sbunlock(&so->so_snd); 926 out: 927 if (top) 928 m_freem(top); 929 if (control) 930 m_freem(control); 931 return (error); 932 } 933 934 /* 935 * Implement receive operations on a socket. 936 * We depend on the way that records are added to the sockbuf 937 * by sbappend*. In particular, each record (mbufs linked through m_next) 938 * must begin with an address if the protocol so specifies, 939 * followed by an optional mbuf or mbufs containing ancillary data, 940 * and then zero or more mbufs of data. 941 * In order to avoid blocking network interrupts for the entire time here, 942 * we splx() while doing the actual copy to user space. 943 * Although the sockbuf is locked, new data may still be appended, 944 * and thus we must maintain consistency of the sockbuf during that time. 945 * 946 * The caller may receive the data as a single mbuf chain by supplying 947 * an mbuf **mp0 for use in returning the chain. The uio is then used 948 * only for the count in uio_resid. 949 */ 950 int 951 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 952 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 953 { 954 struct lwp *l = curlwp; 955 struct mbuf *m, **mp; 956 int flags, len, error, s, offset, moff, type, orig_resid; 957 const struct protosw *pr; 958 struct mbuf *nextrecord; 959 int mbuf_removed = 0; 960 961 pr = so->so_proto; 962 mp = mp0; 963 type = 0; 964 orig_resid = uio->uio_resid; 965 966 if (paddr) 967 *paddr = 0; 968 if (controlp) 969 *controlp = 0; 970 if (flagsp) 971 flags = *flagsp &~ MSG_EOR; 972 else 973 flags = 0; 974 975 if ((flags & MSG_DONTWAIT) == 0) 976 sodopendfree(); 977 978 if (flags & MSG_OOB) { 979 m = m_get(M_WAIT, MT_DATA); 980 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 981 (struct mbuf *)(long)(flags & MSG_PEEK), 982 (struct mbuf *)0, l); 983 if (error) 984 goto bad; 985 do { 986 error = uiomove(mtod(m, void *), 987 (int) min(uio->uio_resid, m->m_len), uio); 988 m = m_free(m); 989 } while (uio->uio_resid && error == 0 && m); 990 bad: 991 if (m) 992 m_freem(m); 993 return (error); 994 } 995 if (mp) 996 *mp = (struct mbuf *)0; 997 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 998 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 999 (struct mbuf *)0, (struct mbuf *)0, l); 1000 1001 restart: 1002 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) 1003 return (error); 1004 s = splsoftnet(); 1005 1006 m = so->so_rcv.sb_mb; 1007 /* 1008 * If we have less data than requested, block awaiting more 1009 * (subject to any timeout) if: 1010 * 1. the current count is less than the low water mark, 1011 * 2. MSG_WAITALL is set, and it is possible to do the entire 1012 * receive operation at once if we block (resid <= hiwat), or 1013 * 3. MSG_DONTWAIT is not set. 1014 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1015 * we have to do the receive in sections, and thus risk returning 1016 * a short count if a timeout or signal occurs after we start. 1017 */ 1018 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 1019 so->so_rcv.sb_cc < uio->uio_resid) && 1020 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1021 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1022 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1023 #ifdef DIAGNOSTIC 1024 if (m == 0 && so->so_rcv.sb_cc) 1025 panic("receive 1"); 1026 #endif 1027 if (so->so_error) { 1028 if (m) 1029 goto dontblock; 1030 error = so->so_error; 1031 if ((flags & MSG_PEEK) == 0) 1032 so->so_error = 0; 1033 goto release; 1034 } 1035 if (so->so_state & SS_CANTRCVMORE) { 1036 if (m) 1037 goto dontblock; 1038 else 1039 goto release; 1040 } 1041 for (; m; m = m->m_next) 1042 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1043 m = so->so_rcv.sb_mb; 1044 goto dontblock; 1045 } 1046 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1047 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1048 error = ENOTCONN; 1049 goto release; 1050 } 1051 if (uio->uio_resid == 0) 1052 goto release; 1053 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 1054 error = EWOULDBLOCK; 1055 goto release; 1056 } 1057 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1058 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1059 sbunlock(&so->so_rcv); 1060 error = sbwait(&so->so_rcv); 1061 splx(s); 1062 if (error) 1063 return (error); 1064 goto restart; 1065 } 1066 dontblock: 1067 /* 1068 * On entry here, m points to the first record of the socket buffer. 1069 * While we process the initial mbufs containing address and control 1070 * info, we save a copy of m->m_nextpkt into nextrecord. 1071 */ 1072 if (l) 1073 l->l_proc->p_stats->p_ru.ru_msgrcv++; 1074 KASSERT(m == so->so_rcv.sb_mb); 1075 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1076 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1077 nextrecord = m->m_nextpkt; 1078 if (pr->pr_flags & PR_ADDR) { 1079 #ifdef DIAGNOSTIC 1080 if (m->m_type != MT_SONAME) 1081 panic("receive 1a"); 1082 #endif 1083 orig_resid = 0; 1084 if (flags & MSG_PEEK) { 1085 if (paddr) 1086 *paddr = m_copy(m, 0, m->m_len); 1087 m = m->m_next; 1088 } else { 1089 sbfree(&so->so_rcv, m); 1090 mbuf_removed = 1; 1091 if (paddr) { 1092 *paddr = m; 1093 so->so_rcv.sb_mb = m->m_next; 1094 m->m_next = 0; 1095 m = so->so_rcv.sb_mb; 1096 } else { 1097 MFREE(m, so->so_rcv.sb_mb); 1098 m = so->so_rcv.sb_mb; 1099 } 1100 } 1101 } 1102 while (m && m->m_type == MT_CONTROL && error == 0) { 1103 if (flags & MSG_PEEK) { 1104 if (controlp) 1105 *controlp = m_copy(m, 0, m->m_len); 1106 m = m->m_next; 1107 } else { 1108 sbfree(&so->so_rcv, m); 1109 mbuf_removed = 1; 1110 if (controlp) { 1111 struct domain *dom = pr->pr_domain; 1112 if (dom->dom_externalize && l && 1113 mtod(m, struct cmsghdr *)->cmsg_type == 1114 SCM_RIGHTS) 1115 error = (*dom->dom_externalize)(m, l); 1116 *controlp = m; 1117 so->so_rcv.sb_mb = m->m_next; 1118 m->m_next = 0; 1119 m = so->so_rcv.sb_mb; 1120 } else { 1121 /* 1122 * Dispose of any SCM_RIGHTS message that went 1123 * through the read path rather than recv. 1124 */ 1125 if (pr->pr_domain->dom_dispose && 1126 mtod(m, struct cmsghdr *)->cmsg_type == SCM_RIGHTS) 1127 (*pr->pr_domain->dom_dispose)(m); 1128 MFREE(m, so->so_rcv.sb_mb); 1129 m = so->so_rcv.sb_mb; 1130 } 1131 } 1132 if (controlp) { 1133 orig_resid = 0; 1134 controlp = &(*controlp)->m_next; 1135 } 1136 } 1137 1138 /* 1139 * If m is non-NULL, we have some data to read. From now on, 1140 * make sure to keep sb_lastrecord consistent when working on 1141 * the last packet on the chain (nextrecord == NULL) and we 1142 * change m->m_nextpkt. 1143 */ 1144 if (m) { 1145 if ((flags & MSG_PEEK) == 0) { 1146 m->m_nextpkt = nextrecord; 1147 /* 1148 * If nextrecord == NULL (this is a single chain), 1149 * then sb_lastrecord may not be valid here if m 1150 * was changed earlier. 1151 */ 1152 if (nextrecord == NULL) { 1153 KASSERT(so->so_rcv.sb_mb == m); 1154 so->so_rcv.sb_lastrecord = m; 1155 } 1156 } 1157 type = m->m_type; 1158 if (type == MT_OOBDATA) 1159 flags |= MSG_OOB; 1160 } else { 1161 if ((flags & MSG_PEEK) == 0) { 1162 KASSERT(so->so_rcv.sb_mb == m); 1163 so->so_rcv.sb_mb = nextrecord; 1164 SB_EMPTY_FIXUP(&so->so_rcv); 1165 } 1166 } 1167 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1168 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1169 1170 moff = 0; 1171 offset = 0; 1172 while (m && uio->uio_resid > 0 && error == 0) { 1173 if (m->m_type == MT_OOBDATA) { 1174 if (type != MT_OOBDATA) 1175 break; 1176 } else if (type == MT_OOBDATA) 1177 break; 1178 #ifdef DIAGNOSTIC 1179 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1180 panic("receive 3"); 1181 #endif 1182 so->so_state &= ~SS_RCVATMARK; 1183 len = uio->uio_resid; 1184 if (so->so_oobmark && len > so->so_oobmark - offset) 1185 len = so->so_oobmark - offset; 1186 if (len > m->m_len - moff) 1187 len = m->m_len - moff; 1188 /* 1189 * If mp is set, just pass back the mbufs. 1190 * Otherwise copy them out via the uio, then free. 1191 * Sockbuf must be consistent here (points to current mbuf, 1192 * it points to next record) when we drop priority; 1193 * we must note any additions to the sockbuf when we 1194 * block interrupts again. 1195 */ 1196 if (mp == 0) { 1197 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1198 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1199 splx(s); 1200 error = uiomove(mtod(m, char *) + moff, (int)len, uio); 1201 s = splsoftnet(); 1202 if (error) { 1203 /* 1204 * If any part of the record has been removed 1205 * (such as the MT_SONAME mbuf, which will 1206 * happen when PR_ADDR, and thus also 1207 * PR_ATOMIC, is set), then drop the entire 1208 * record to maintain the atomicity of the 1209 * receive operation. 1210 * 1211 * This avoids a later panic("receive 1a") 1212 * when compiled with DIAGNOSTIC. 1213 */ 1214 if (m && mbuf_removed 1215 && (pr->pr_flags & PR_ATOMIC)) 1216 (void) sbdroprecord(&so->so_rcv); 1217 1218 goto release; 1219 } 1220 } else 1221 uio->uio_resid -= len; 1222 if (len == m->m_len - moff) { 1223 if (m->m_flags & M_EOR) 1224 flags |= MSG_EOR; 1225 if (flags & MSG_PEEK) { 1226 m = m->m_next; 1227 moff = 0; 1228 } else { 1229 nextrecord = m->m_nextpkt; 1230 sbfree(&so->so_rcv, m); 1231 if (mp) { 1232 *mp = m; 1233 mp = &m->m_next; 1234 so->so_rcv.sb_mb = m = m->m_next; 1235 *mp = (struct mbuf *)0; 1236 } else { 1237 MFREE(m, so->so_rcv.sb_mb); 1238 m = so->so_rcv.sb_mb; 1239 } 1240 /* 1241 * If m != NULL, we also know that 1242 * so->so_rcv.sb_mb != NULL. 1243 */ 1244 KASSERT(so->so_rcv.sb_mb == m); 1245 if (m) { 1246 m->m_nextpkt = nextrecord; 1247 if (nextrecord == NULL) 1248 so->so_rcv.sb_lastrecord = m; 1249 } else { 1250 so->so_rcv.sb_mb = nextrecord; 1251 SB_EMPTY_FIXUP(&so->so_rcv); 1252 } 1253 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1254 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1255 } 1256 } else { 1257 if (flags & MSG_PEEK) 1258 moff += len; 1259 else { 1260 if (mp) 1261 *mp = m_copym(m, 0, len, M_WAIT); 1262 m->m_data += len; 1263 m->m_len -= len; 1264 so->so_rcv.sb_cc -= len; 1265 } 1266 } 1267 if (so->so_oobmark) { 1268 if ((flags & MSG_PEEK) == 0) { 1269 so->so_oobmark -= len; 1270 if (so->so_oobmark == 0) { 1271 so->so_state |= SS_RCVATMARK; 1272 break; 1273 } 1274 } else { 1275 offset += len; 1276 if (offset == so->so_oobmark) 1277 break; 1278 } 1279 } 1280 if (flags & MSG_EOR) 1281 break; 1282 /* 1283 * If the MSG_WAITALL flag is set (for non-atomic socket), 1284 * we must not quit until "uio->uio_resid == 0" or an error 1285 * termination. If a signal/timeout occurs, return 1286 * with a short count but without error. 1287 * Keep sockbuf locked against other readers. 1288 */ 1289 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1290 !sosendallatonce(so) && !nextrecord) { 1291 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1292 break; 1293 /* 1294 * If we are peeking and the socket receive buffer is 1295 * full, stop since we can't get more data to peek at. 1296 */ 1297 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1298 break; 1299 /* 1300 * If we've drained the socket buffer, tell the 1301 * protocol in case it needs to do something to 1302 * get it filled again. 1303 */ 1304 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1305 (*pr->pr_usrreq)(so, PRU_RCVD, 1306 (struct mbuf *)0, 1307 (struct mbuf *)(long)flags, 1308 (struct mbuf *)0, l); 1309 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1310 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1311 error = sbwait(&so->so_rcv); 1312 if (error) { 1313 sbunlock(&so->so_rcv); 1314 splx(s); 1315 return (0); 1316 } 1317 if ((m = so->so_rcv.sb_mb) != NULL) 1318 nextrecord = m->m_nextpkt; 1319 } 1320 } 1321 1322 if (m && pr->pr_flags & PR_ATOMIC) { 1323 flags |= MSG_TRUNC; 1324 if ((flags & MSG_PEEK) == 0) 1325 (void) sbdroprecord(&so->so_rcv); 1326 } 1327 if ((flags & MSG_PEEK) == 0) { 1328 if (m == 0) { 1329 /* 1330 * First part is an inline SB_EMPTY_FIXUP(). Second 1331 * part makes sure sb_lastrecord is up-to-date if 1332 * there is still data in the socket buffer. 1333 */ 1334 so->so_rcv.sb_mb = nextrecord; 1335 if (so->so_rcv.sb_mb == NULL) { 1336 so->so_rcv.sb_mbtail = NULL; 1337 so->so_rcv.sb_lastrecord = NULL; 1338 } else if (nextrecord->m_nextpkt == NULL) 1339 so->so_rcv.sb_lastrecord = nextrecord; 1340 } 1341 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1342 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1343 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1344 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1345 (struct mbuf *)(long)flags, (struct mbuf *)0, l); 1346 } 1347 if (orig_resid == uio->uio_resid && orig_resid && 1348 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1349 sbunlock(&so->so_rcv); 1350 splx(s); 1351 goto restart; 1352 } 1353 1354 if (flagsp) 1355 *flagsp |= flags; 1356 release: 1357 sbunlock(&so->so_rcv); 1358 splx(s); 1359 return (error); 1360 } 1361 1362 int 1363 soshutdown(struct socket *so, int how) 1364 { 1365 const struct protosw *pr; 1366 1367 pr = so->so_proto; 1368 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1369 return (EINVAL); 1370 1371 if (how == SHUT_RD || how == SHUT_RDWR) 1372 sorflush(so); 1373 if (how == SHUT_WR || how == SHUT_RDWR) 1374 return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0, 1375 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1376 return (0); 1377 } 1378 1379 void 1380 sorflush(struct socket *so) 1381 { 1382 struct sockbuf *sb, asb; 1383 const struct protosw *pr; 1384 int s; 1385 1386 sb = &so->so_rcv; 1387 pr = so->so_proto; 1388 sb->sb_flags |= SB_NOINTR; 1389 (void) sblock(sb, M_WAITOK); 1390 s = splnet(); 1391 socantrcvmore(so); 1392 sbunlock(sb); 1393 asb = *sb; 1394 /* 1395 * Clear most of the sockbuf structure, but leave some of the 1396 * fields valid. 1397 */ 1398 memset(&sb->sb_startzero, 0, 1399 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1400 splx(s); 1401 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1402 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1403 sbrelease(&asb, so); 1404 } 1405 1406 int 1407 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0) 1408 { 1409 int error; 1410 struct mbuf *m; 1411 struct linger *l; 1412 1413 error = 0; 1414 m = m0; 1415 if (level != SOL_SOCKET) { 1416 if (so->so_proto && so->so_proto->pr_ctloutput) 1417 return ((*so->so_proto->pr_ctloutput) 1418 (PRCO_SETOPT, so, level, optname, &m0)); 1419 error = ENOPROTOOPT; 1420 } else { 1421 switch (optname) { 1422 1423 case SO_LINGER: 1424 if (m == NULL || m->m_len != sizeof(struct linger)) { 1425 error = EINVAL; 1426 goto bad; 1427 } 1428 l = mtod(m, struct linger *); 1429 if (l->l_linger < 0 || l->l_linger > USHRT_MAX || 1430 l->l_linger > (INT_MAX / hz)) { 1431 error = EDOM; 1432 goto bad; 1433 } 1434 so->so_linger = l->l_linger; 1435 if (l->l_onoff) 1436 so->so_options |= SO_LINGER; 1437 else 1438 so->so_options &= ~SO_LINGER; 1439 break; 1440 1441 case SO_DEBUG: 1442 case SO_KEEPALIVE: 1443 case SO_DONTROUTE: 1444 case SO_USELOOPBACK: 1445 case SO_BROADCAST: 1446 case SO_REUSEADDR: 1447 case SO_REUSEPORT: 1448 case SO_OOBINLINE: 1449 case SO_TIMESTAMP: 1450 if (m == NULL || m->m_len < sizeof(int)) { 1451 error = EINVAL; 1452 goto bad; 1453 } 1454 if (*mtod(m, int *)) 1455 so->so_options |= optname; 1456 else 1457 so->so_options &= ~optname; 1458 break; 1459 1460 case SO_SNDBUF: 1461 case SO_RCVBUF: 1462 case SO_SNDLOWAT: 1463 case SO_RCVLOWAT: 1464 { 1465 int optval; 1466 1467 if (m == NULL || m->m_len < sizeof(int)) { 1468 error = EINVAL; 1469 goto bad; 1470 } 1471 1472 /* 1473 * Values < 1 make no sense for any of these 1474 * options, so disallow them. 1475 */ 1476 optval = *mtod(m, int *); 1477 if (optval < 1) { 1478 error = EINVAL; 1479 goto bad; 1480 } 1481 1482 switch (optname) { 1483 1484 case SO_SNDBUF: 1485 case SO_RCVBUF: 1486 if (sbreserve(optname == SO_SNDBUF ? 1487 &so->so_snd : &so->so_rcv, 1488 (u_long) optval, so) == 0) { 1489 error = ENOBUFS; 1490 goto bad; 1491 } 1492 break; 1493 1494 /* 1495 * Make sure the low-water is never greater than 1496 * the high-water. 1497 */ 1498 case SO_SNDLOWAT: 1499 so->so_snd.sb_lowat = 1500 (optval > so->so_snd.sb_hiwat) ? 1501 so->so_snd.sb_hiwat : optval; 1502 break; 1503 case SO_RCVLOWAT: 1504 so->so_rcv.sb_lowat = 1505 (optval > so->so_rcv.sb_hiwat) ? 1506 so->so_rcv.sb_hiwat : optval; 1507 break; 1508 } 1509 break; 1510 } 1511 1512 case SO_SNDTIMEO: 1513 case SO_RCVTIMEO: 1514 { 1515 struct timeval *tv; 1516 int val; 1517 1518 if (m == NULL || m->m_len < sizeof(*tv)) { 1519 error = EINVAL; 1520 goto bad; 1521 } 1522 tv = mtod(m, struct timeval *); 1523 if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz) { 1524 error = EDOM; 1525 goto bad; 1526 } 1527 val = tv->tv_sec * hz + tv->tv_usec / tick; 1528 if (val == 0 && tv->tv_usec != 0) 1529 val = 1; 1530 1531 switch (optname) { 1532 1533 case SO_SNDTIMEO: 1534 so->so_snd.sb_timeo = val; 1535 break; 1536 case SO_RCVTIMEO: 1537 so->so_rcv.sb_timeo = val; 1538 break; 1539 } 1540 break; 1541 } 1542 1543 default: 1544 error = ENOPROTOOPT; 1545 break; 1546 } 1547 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1548 (void) ((*so->so_proto->pr_ctloutput) 1549 (PRCO_SETOPT, so, level, optname, &m0)); 1550 m = NULL; /* freed by protocol */ 1551 } 1552 } 1553 bad: 1554 if (m) 1555 (void) m_free(m); 1556 return (error); 1557 } 1558 1559 int 1560 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp) 1561 { 1562 struct mbuf *m; 1563 1564 if (level != SOL_SOCKET) { 1565 if (so->so_proto && so->so_proto->pr_ctloutput) { 1566 return ((*so->so_proto->pr_ctloutput) 1567 (PRCO_GETOPT, so, level, optname, mp)); 1568 } else 1569 return (ENOPROTOOPT); 1570 } else { 1571 m = m_get(M_WAIT, MT_SOOPTS); 1572 m->m_len = sizeof(int); 1573 1574 switch (optname) { 1575 1576 case SO_LINGER: 1577 m->m_len = sizeof(struct linger); 1578 mtod(m, struct linger *)->l_onoff = 1579 (so->so_options & SO_LINGER) ? 1 : 0; 1580 mtod(m, struct linger *)->l_linger = so->so_linger; 1581 break; 1582 1583 case SO_USELOOPBACK: 1584 case SO_DONTROUTE: 1585 case SO_DEBUG: 1586 case SO_KEEPALIVE: 1587 case SO_REUSEADDR: 1588 case SO_REUSEPORT: 1589 case SO_BROADCAST: 1590 case SO_OOBINLINE: 1591 case SO_TIMESTAMP: 1592 *mtod(m, int *) = (so->so_options & optname) ? 1 : 0; 1593 break; 1594 1595 case SO_TYPE: 1596 *mtod(m, int *) = so->so_type; 1597 break; 1598 1599 case SO_ERROR: 1600 *mtod(m, int *) = so->so_error; 1601 so->so_error = 0; 1602 break; 1603 1604 case SO_SNDBUF: 1605 *mtod(m, int *) = so->so_snd.sb_hiwat; 1606 break; 1607 1608 case SO_RCVBUF: 1609 *mtod(m, int *) = so->so_rcv.sb_hiwat; 1610 break; 1611 1612 case SO_SNDLOWAT: 1613 *mtod(m, int *) = so->so_snd.sb_lowat; 1614 break; 1615 1616 case SO_RCVLOWAT: 1617 *mtod(m, int *) = so->so_rcv.sb_lowat; 1618 break; 1619 1620 case SO_SNDTIMEO: 1621 case SO_RCVTIMEO: 1622 { 1623 int val = (optname == SO_SNDTIMEO ? 1624 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1625 1626 m->m_len = sizeof(struct timeval); 1627 mtod(m, struct timeval *)->tv_sec = val / hz; 1628 mtod(m, struct timeval *)->tv_usec = 1629 (val % hz) * tick; 1630 break; 1631 } 1632 1633 case SO_OVERFLOWED: 1634 *mtod(m, int *) = so->so_rcv.sb_overflowed; 1635 break; 1636 1637 default: 1638 (void)m_free(m); 1639 return (ENOPROTOOPT); 1640 } 1641 *mp = m; 1642 return (0); 1643 } 1644 } 1645 1646 void 1647 sohasoutofband(struct socket *so) 1648 { 1649 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 1650 selwakeup(&so->so_rcv.sb_sel); 1651 } 1652 1653 static void 1654 filt_sordetach(struct knote *kn) 1655 { 1656 struct socket *so; 1657 1658 so = (struct socket *)kn->kn_fp->f_data; 1659 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 1660 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 1661 so->so_rcv.sb_flags &= ~SB_KNOTE; 1662 } 1663 1664 /*ARGSUSED*/ 1665 static int 1666 filt_soread(struct knote *kn, long hint) 1667 { 1668 struct socket *so; 1669 1670 so = (struct socket *)kn->kn_fp->f_data; 1671 kn->kn_data = so->so_rcv.sb_cc; 1672 if (so->so_state & SS_CANTRCVMORE) { 1673 kn->kn_flags |= EV_EOF; 1674 kn->kn_fflags = so->so_error; 1675 return (1); 1676 } 1677 if (so->so_error) /* temporary udp error */ 1678 return (1); 1679 if (kn->kn_sfflags & NOTE_LOWAT) 1680 return (kn->kn_data >= kn->kn_sdata); 1681 return (kn->kn_data >= so->so_rcv.sb_lowat); 1682 } 1683 1684 static void 1685 filt_sowdetach(struct knote *kn) 1686 { 1687 struct socket *so; 1688 1689 so = (struct socket *)kn->kn_fp->f_data; 1690 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 1691 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 1692 so->so_snd.sb_flags &= ~SB_KNOTE; 1693 } 1694 1695 /*ARGSUSED*/ 1696 static int 1697 filt_sowrite(struct knote *kn, long hint) 1698 { 1699 struct socket *so; 1700 1701 so = (struct socket *)kn->kn_fp->f_data; 1702 kn->kn_data = sbspace(&so->so_snd); 1703 if (so->so_state & SS_CANTSENDMORE) { 1704 kn->kn_flags |= EV_EOF; 1705 kn->kn_fflags = so->so_error; 1706 return (1); 1707 } 1708 if (so->so_error) /* temporary udp error */ 1709 return (1); 1710 if (((so->so_state & SS_ISCONNECTED) == 0) && 1711 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1712 return (0); 1713 if (kn->kn_sfflags & NOTE_LOWAT) 1714 return (kn->kn_data >= kn->kn_sdata); 1715 return (kn->kn_data >= so->so_snd.sb_lowat); 1716 } 1717 1718 /*ARGSUSED*/ 1719 static int 1720 filt_solisten(struct knote *kn, long hint) 1721 { 1722 struct socket *so; 1723 1724 so = (struct socket *)kn->kn_fp->f_data; 1725 1726 /* 1727 * Set kn_data to number of incoming connections, not 1728 * counting partial (incomplete) connections. 1729 */ 1730 kn->kn_data = so->so_qlen; 1731 return (kn->kn_data > 0); 1732 } 1733 1734 static const struct filterops solisten_filtops = 1735 { 1, NULL, filt_sordetach, filt_solisten }; 1736 static const struct filterops soread_filtops = 1737 { 1, NULL, filt_sordetach, filt_soread }; 1738 static const struct filterops sowrite_filtops = 1739 { 1, NULL, filt_sowdetach, filt_sowrite }; 1740 1741 int 1742 soo_kqfilter(struct file *fp, struct knote *kn) 1743 { 1744 struct socket *so; 1745 struct sockbuf *sb; 1746 1747 so = (struct socket *)kn->kn_fp->f_data; 1748 switch (kn->kn_filter) { 1749 case EVFILT_READ: 1750 if (so->so_options & SO_ACCEPTCONN) 1751 kn->kn_fop = &solisten_filtops; 1752 else 1753 kn->kn_fop = &soread_filtops; 1754 sb = &so->so_rcv; 1755 break; 1756 case EVFILT_WRITE: 1757 kn->kn_fop = &sowrite_filtops; 1758 sb = &so->so_snd; 1759 break; 1760 default: 1761 return (1); 1762 } 1763 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 1764 sb->sb_flags |= SB_KNOTE; 1765 return (0); 1766 } 1767 1768 #include <sys/sysctl.h> 1769 1770 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 1771 1772 /* 1773 * sysctl helper routine for kern.somaxkva. ensures that the given 1774 * value is not too small. 1775 * (XXX should we maybe make sure it's not too large as well?) 1776 */ 1777 static int 1778 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 1779 { 1780 int error, new_somaxkva; 1781 struct sysctlnode node; 1782 1783 new_somaxkva = somaxkva; 1784 node = *rnode; 1785 node.sysctl_data = &new_somaxkva; 1786 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1787 if (error || newp == NULL) 1788 return (error); 1789 1790 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 1791 return (EINVAL); 1792 1793 mutex_enter(&so_pendfree_lock); 1794 somaxkva = new_somaxkva; 1795 cv_broadcast(&socurkva_cv); 1796 mutex_exit(&so_pendfree_lock); 1797 1798 return (error); 1799 } 1800 1801 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup") 1802 { 1803 1804 sysctl_createv(clog, 0, NULL, NULL, 1805 CTLFLAG_PERMANENT, 1806 CTLTYPE_NODE, "kern", NULL, 1807 NULL, 0, NULL, 0, 1808 CTL_KERN, CTL_EOL); 1809 1810 sysctl_createv(clog, 0, NULL, NULL, 1811 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1812 CTLTYPE_INT, "somaxkva", 1813 SYSCTL_DESCR("Maximum amount of kernel memory to be " 1814 "used for socket buffers"), 1815 sysctl_kern_somaxkva, 0, NULL, 0, 1816 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 1817 } 1818
Indexes created Sat Sep 20 22:09:52 GMT 2025