uipc_socket.c revision 1.138
1 /* $NetBSD: uipc_socket.c,v 1.138 2007/04/03 23:44:53 rmind 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 1.138 2007/04/03 23:44:53 rmind 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 uid = kauth_cred_geteuid(l->l_cred); 499 so->so_uidinfo = uid_find(uid); 500 error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0, 501 (struct mbuf *)(long)proto, (struct mbuf *)0, l); 502 if (error) { 503 so->so_state |= SS_NOFDREF; 504 sofree(so); 505 splx(s); 506 return (error); 507 } 508 splx(s); 509 *aso = so; 510 return (0); 511 } 512 513 int 514 sobind(struct socket *so, struct mbuf *nam, struct lwp *l) 515 { 516 int s, error; 517 518 s = splsoftnet(); 519 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0, 520 nam, (struct mbuf *)0, l); 521 splx(s); 522 return (error); 523 } 524 525 int 526 solisten(struct socket *so, int backlog) 527 { 528 int s, error; 529 530 s = splsoftnet(); 531 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0, 532 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 533 if (error) { 534 splx(s); 535 return (error); 536 } 537 if (TAILQ_EMPTY(&so->so_q)) 538 so->so_options |= SO_ACCEPTCONN; 539 if (backlog < 0) 540 backlog = 0; 541 so->so_qlimit = min(backlog, somaxconn); 542 splx(s); 543 return (0); 544 } 545 546 void 547 sofree(struct socket *so) 548 { 549 550 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 551 return; 552 if (so->so_head) { 553 /* 554 * We must not decommission a socket that's on the accept(2) 555 * queue. If we do, then accept(2) may hang after select(2) 556 * indicated that the listening socket was ready. 557 */ 558 if (!soqremque(so, 0)) 559 return; 560 } 561 if (so->so_rcv.sb_hiwat) 562 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 563 RLIM_INFINITY); 564 if (so->so_snd.sb_hiwat) 565 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 566 RLIM_INFINITY); 567 sbrelease(&so->so_snd, so); 568 sorflush(so); 569 pool_put(&socket_pool, so); 570 } 571 572 /* 573 * Close a socket on last file table reference removal. 574 * Initiate disconnect if connected. 575 * Free socket when disconnect complete. 576 */ 577 int 578 soclose(struct socket *so) 579 { 580 struct socket *so2; 581 int s, error; 582 583 error = 0; 584 s = splsoftnet(); /* conservative */ 585 if (so->so_options & SO_ACCEPTCONN) { 586 while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 587 (void) soqremque(so2, 0); 588 (void) soabort(so2); 589 } 590 while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 591 (void) soqremque(so2, 1); 592 (void) soabort(so2); 593 } 594 } 595 if (so->so_pcb == 0) 596 goto discard; 597 if (so->so_state & SS_ISCONNECTED) { 598 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 599 error = sodisconnect(so); 600 if (error) 601 goto drop; 602 } 603 if (so->so_options & SO_LINGER) { 604 if ((so->so_state & SS_ISDISCONNECTING) && 605 (so->so_state & SS_NBIO)) 606 goto drop; 607 while (so->so_state & SS_ISCONNECTED) { 608 error = tsleep((void *)&so->so_timeo, 609 PSOCK | PCATCH, netcls, 610 so->so_linger * hz); 611 if (error) 612 break; 613 } 614 } 615 } 616 drop: 617 if (so->so_pcb) { 618 int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 619 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 620 (struct lwp *)0); 621 if (error == 0) 622 error = error2; 623 } 624 discard: 625 if (so->so_state & SS_NOFDREF) 626 panic("soclose: NOFDREF"); 627 so->so_state |= SS_NOFDREF; 628 sofree(so); 629 splx(s); 630 return (error); 631 } 632 633 /* 634 * Must be called at splsoftnet... 635 */ 636 int 637 soabort(struct socket *so) 638 { 639 640 return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0, 641 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 642 } 643 644 int 645 soaccept(struct socket *so, struct mbuf *nam) 646 { 647 int s, error; 648 649 error = 0; 650 s = splsoftnet(); 651 if ((so->so_state & SS_NOFDREF) == 0) 652 panic("soaccept: !NOFDREF"); 653 so->so_state &= ~SS_NOFDREF; 654 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 655 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 656 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 657 (struct mbuf *)0, nam, (struct mbuf *)0, (struct lwp *)0); 658 else 659 error = ECONNABORTED; 660 661 splx(s); 662 return (error); 663 } 664 665 int 666 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l) 667 { 668 int s, error; 669 670 if (so->so_options & SO_ACCEPTCONN) 671 return (EOPNOTSUPP); 672 s = splsoftnet(); 673 /* 674 * If protocol is connection-based, can only connect once. 675 * Otherwise, if connected, try to disconnect first. 676 * This allows user to disconnect by connecting to, e.g., 677 * a null address. 678 */ 679 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 680 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 681 (error = sodisconnect(so)))) 682 error = EISCONN; 683 else 684 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 685 (struct mbuf *)0, nam, (struct mbuf *)0, l); 686 splx(s); 687 return (error); 688 } 689 690 int 691 soconnect2(struct socket *so1, struct socket *so2) 692 { 693 int s, error; 694 695 s = splsoftnet(); 696 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 697 (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0, 698 (struct lwp *)0); 699 splx(s); 700 return (error); 701 } 702 703 int 704 sodisconnect(struct socket *so) 705 { 706 int s, error; 707 708 s = splsoftnet(); 709 if ((so->so_state & SS_ISCONNECTED) == 0) { 710 error = ENOTCONN; 711 goto bad; 712 } 713 if (so->so_state & SS_ISDISCONNECTING) { 714 error = EALREADY; 715 goto bad; 716 } 717 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 718 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 719 (struct lwp *)0); 720 bad: 721 splx(s); 722 sodopendfree(); 723 return (error); 724 } 725 726 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 727 /* 728 * Send on a socket. 729 * If send must go all at once and message is larger than 730 * send buffering, then hard error. 731 * Lock against other senders. 732 * If must go all at once and not enough room now, then 733 * inform user that this would block and do nothing. 734 * Otherwise, if nonblocking, send as much as possible. 735 * The data to be sent is described by "uio" if nonzero, 736 * otherwise by the mbuf chain "top" (which must be null 737 * if uio is not). Data provided in mbuf chain must be small 738 * enough to send all at once. 739 * 740 * Returns nonzero on error, timeout or signal; callers 741 * must check for short counts if EINTR/ERESTART are returned. 742 * Data and control buffers are freed on return. 743 */ 744 int 745 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 746 struct mbuf *control, int flags, struct lwp *l) 747 { 748 struct mbuf **mp, *m; 749 struct proc *p; 750 long space, len, resid, clen, mlen; 751 int error, s, dontroute, atomic; 752 753 p = l->l_proc; 754 sodopendfree(); 755 756 clen = 0; 757 atomic = sosendallatonce(so) || top; 758 if (uio) 759 resid = uio->uio_resid; 760 else 761 resid = top->m_pkthdr.len; 762 /* 763 * In theory resid should be unsigned. 764 * However, space must be signed, as it might be less than 0 765 * if we over-committed, and we must use a signed comparison 766 * of space and resid. On the other hand, a negative resid 767 * causes us to loop sending 0-length segments to the protocol. 768 */ 769 if (resid < 0) { 770 error = EINVAL; 771 goto out; 772 } 773 dontroute = 774 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 775 (so->so_proto->pr_flags & PR_ATOMIC); 776 if (p) 777 p->p_stats->p_ru.ru_msgsnd++; 778 if (control) 779 clen = control->m_len; 780 #define snderr(errno) { error = errno; splx(s); goto release; } 781 782 restart: 783 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 784 goto out; 785 do { 786 s = splsoftnet(); 787 if (so->so_state & SS_CANTSENDMORE) 788 snderr(EPIPE); 789 if (so->so_error) { 790 error = so->so_error; 791 so->so_error = 0; 792 splx(s); 793 goto release; 794 } 795 if ((so->so_state & SS_ISCONNECTED) == 0) { 796 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 797 if ((so->so_state & SS_ISCONFIRMING) == 0 && 798 !(resid == 0 && clen != 0)) 799 snderr(ENOTCONN); 800 } else if (addr == 0) 801 snderr(EDESTADDRREQ); 802 } 803 space = sbspace(&so->so_snd); 804 if (flags & MSG_OOB) 805 space += 1024; 806 if ((atomic && resid > so->so_snd.sb_hiwat) || 807 clen > so->so_snd.sb_hiwat) 808 snderr(EMSGSIZE); 809 if (space < resid + clen && 810 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 811 if (so->so_state & SS_NBIO) 812 snderr(EWOULDBLOCK); 813 sbunlock(&so->so_snd); 814 error = sbwait(&so->so_snd); 815 splx(s); 816 if (error) 817 goto out; 818 goto restart; 819 } 820 splx(s); 821 mp = ⊤ 822 space -= clen; 823 do { 824 if (uio == NULL) { 825 /* 826 * Data is prepackaged in "top". 827 */ 828 resid = 0; 829 if (flags & MSG_EOR) 830 top->m_flags |= M_EOR; 831 } else do { 832 if (top == 0) { 833 m = m_gethdr(M_WAIT, MT_DATA); 834 mlen = MHLEN; 835 m->m_pkthdr.len = 0; 836 m->m_pkthdr.rcvif = (struct ifnet *)0; 837 } else { 838 m = m_get(M_WAIT, MT_DATA); 839 mlen = MLEN; 840 } 841 MCLAIM(m, so->so_snd.sb_mowner); 842 if (sock_loan_thresh >= 0 && 843 uio->uio_iov->iov_len >= sock_loan_thresh && 844 space >= sock_loan_thresh && 845 (len = sosend_loan(so, uio, m, 846 space)) != 0) { 847 SOSEND_COUNTER_INCR(&sosend_loan_big); 848 space -= len; 849 goto have_data; 850 } 851 if (resid >= MINCLSIZE && space >= MCLBYTES) { 852 SOSEND_COUNTER_INCR(&sosend_copy_big); 853 m_clget(m, M_WAIT); 854 if ((m->m_flags & M_EXT) == 0) 855 goto nopages; 856 mlen = MCLBYTES; 857 if (atomic && top == 0) { 858 len = lmin(MCLBYTES - max_hdr, 859 resid); 860 m->m_data += max_hdr; 861 } else 862 len = lmin(MCLBYTES, resid); 863 space -= len; 864 } else { 865 nopages: 866 SOSEND_COUNTER_INCR(&sosend_copy_small); 867 len = lmin(lmin(mlen, resid), space); 868 space -= len; 869 /* 870 * For datagram protocols, leave room 871 * for protocol headers in first mbuf. 872 */ 873 if (atomic && top == 0 && len < mlen) 874 MH_ALIGN(m, len); 875 } 876 error = uiomove(mtod(m, void *), (int)len, 877 uio); 878 have_data: 879 resid = uio->uio_resid; 880 m->m_len = len; 881 *mp = m; 882 top->m_pkthdr.len += len; 883 if (error) 884 goto release; 885 mp = &m->m_next; 886 if (resid <= 0) { 887 if (flags & MSG_EOR) 888 top->m_flags |= M_EOR; 889 break; 890 } 891 } while (space > 0 && atomic); 892 893 s = splsoftnet(); 894 895 if (so->so_state & SS_CANTSENDMORE) 896 snderr(EPIPE); 897 898 if (dontroute) 899 so->so_options |= SO_DONTROUTE; 900 if (resid > 0) 901 so->so_state |= SS_MORETOCOME; 902 error = (*so->so_proto->pr_usrreq)(so, 903 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 904 top, addr, control, curlwp); /* XXX */ 905 if (dontroute) 906 so->so_options &= ~SO_DONTROUTE; 907 if (resid > 0) 908 so->so_state &= ~SS_MORETOCOME; 909 splx(s); 910 911 clen = 0; 912 control = 0; 913 top = 0; 914 mp = ⊤ 915 if (error) 916 goto release; 917 } while (resid && space > 0); 918 } while (resid); 919 920 release: 921 sbunlock(&so->so_snd); 922 out: 923 if (top) 924 m_freem(top); 925 if (control) 926 m_freem(control); 927 return (error); 928 } 929 930 /* 931 * Implement receive operations on a socket. 932 * We depend on the way that records are added to the sockbuf 933 * by sbappend*. In particular, each record (mbufs linked through m_next) 934 * must begin with an address if the protocol so specifies, 935 * followed by an optional mbuf or mbufs containing ancillary data, 936 * and then zero or more mbufs of data. 937 * In order to avoid blocking network interrupts for the entire time here, 938 * we splx() while doing the actual copy to user space. 939 * Although the sockbuf is locked, new data may still be appended, 940 * and thus we must maintain consistency of the sockbuf during that time. 941 * 942 * The caller may receive the data as a single mbuf chain by supplying 943 * an mbuf **mp0 for use in returning the chain. The uio is then used 944 * only for the count in uio_resid. 945 */ 946 int 947 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 948 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 949 { 950 struct lwp *l = curlwp; 951 struct mbuf *m, **mp; 952 int flags, len, error, s, offset, moff, type, orig_resid; 953 const struct protosw *pr; 954 struct mbuf *nextrecord; 955 int mbuf_removed = 0; 956 957 pr = so->so_proto; 958 mp = mp0; 959 type = 0; 960 orig_resid = uio->uio_resid; 961 962 if (paddr) 963 *paddr = 0; 964 if (controlp) 965 *controlp = 0; 966 if (flagsp) 967 flags = *flagsp &~ MSG_EOR; 968 else 969 flags = 0; 970 971 if ((flags & MSG_DONTWAIT) == 0) 972 sodopendfree(); 973 974 if (flags & MSG_OOB) { 975 m = m_get(M_WAIT, MT_DATA); 976 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 977 (struct mbuf *)(long)(flags & MSG_PEEK), 978 (struct mbuf *)0, l); 979 if (error) 980 goto bad; 981 do { 982 error = uiomove(mtod(m, void *), 983 (int) min(uio->uio_resid, m->m_len), uio); 984 m = m_free(m); 985 } while (uio->uio_resid && error == 0 && m); 986 bad: 987 if (m) 988 m_freem(m); 989 return (error); 990 } 991 if (mp) 992 *mp = (struct mbuf *)0; 993 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 994 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 995 (struct mbuf *)0, (struct mbuf *)0, l); 996 997 restart: 998 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) 999 return (error); 1000 s = splsoftnet(); 1001 1002 m = so->so_rcv.sb_mb; 1003 /* 1004 * If we have less data than requested, block awaiting more 1005 * (subject to any timeout) if: 1006 * 1. the current count is less than the low water mark, 1007 * 2. MSG_WAITALL is set, and it is possible to do the entire 1008 * receive operation at once if we block (resid <= hiwat), or 1009 * 3. MSG_DONTWAIT is not set. 1010 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1011 * we have to do the receive in sections, and thus risk returning 1012 * a short count if a timeout or signal occurs after we start. 1013 */ 1014 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 1015 so->so_rcv.sb_cc < uio->uio_resid) && 1016 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1017 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1018 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1019 #ifdef DIAGNOSTIC 1020 if (m == 0 && so->so_rcv.sb_cc) 1021 panic("receive 1"); 1022 #endif 1023 if (so->so_error) { 1024 if (m) 1025 goto dontblock; 1026 error = so->so_error; 1027 if ((flags & MSG_PEEK) == 0) 1028 so->so_error = 0; 1029 goto release; 1030 } 1031 if (so->so_state & SS_CANTRCVMORE) { 1032 if (m) 1033 goto dontblock; 1034 else 1035 goto release; 1036 } 1037 for (; m; m = m->m_next) 1038 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1039 m = so->so_rcv.sb_mb; 1040 goto dontblock; 1041 } 1042 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1043 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1044 error = ENOTCONN; 1045 goto release; 1046 } 1047 if (uio->uio_resid == 0) 1048 goto release; 1049 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 1050 error = EWOULDBLOCK; 1051 goto release; 1052 } 1053 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1054 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1055 sbunlock(&so->so_rcv); 1056 error = sbwait(&so->so_rcv); 1057 splx(s); 1058 if (error) 1059 return (error); 1060 goto restart; 1061 } 1062 dontblock: 1063 /* 1064 * On entry here, m points to the first record of the socket buffer. 1065 * While we process the initial mbufs containing address and control 1066 * info, we save a copy of m->m_nextpkt into nextrecord. 1067 */ 1068 if (l) 1069 l->l_proc->p_stats->p_ru.ru_msgrcv++; 1070 KASSERT(m == so->so_rcv.sb_mb); 1071 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1072 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1073 nextrecord = m->m_nextpkt; 1074 if (pr->pr_flags & PR_ADDR) { 1075 #ifdef DIAGNOSTIC 1076 if (m->m_type != MT_SONAME) 1077 panic("receive 1a"); 1078 #endif 1079 orig_resid = 0; 1080 if (flags & MSG_PEEK) { 1081 if (paddr) 1082 *paddr = m_copy(m, 0, m->m_len); 1083 m = m->m_next; 1084 } else { 1085 sbfree(&so->so_rcv, m); 1086 mbuf_removed = 1; 1087 if (paddr) { 1088 *paddr = m; 1089 so->so_rcv.sb_mb = m->m_next; 1090 m->m_next = 0; 1091 m = so->so_rcv.sb_mb; 1092 } else { 1093 MFREE(m, so->so_rcv.sb_mb); 1094 m = so->so_rcv.sb_mb; 1095 } 1096 } 1097 } 1098 while (m && m->m_type == MT_CONTROL && error == 0) { 1099 if (flags & MSG_PEEK) { 1100 if (controlp) 1101 *controlp = m_copy(m, 0, m->m_len); 1102 m = m->m_next; 1103 } else { 1104 sbfree(&so->so_rcv, m); 1105 mbuf_removed = 1; 1106 if (controlp) { 1107 struct domain *dom = pr->pr_domain; 1108 if (dom->dom_externalize && l && 1109 mtod(m, struct cmsghdr *)->cmsg_type == 1110 SCM_RIGHTS) 1111 error = (*dom->dom_externalize)(m, l); 1112 *controlp = m; 1113 so->so_rcv.sb_mb = m->m_next; 1114 m->m_next = 0; 1115 m = so->so_rcv.sb_mb; 1116 } else { 1117 /* 1118 * Dispose of any SCM_RIGHTS message that went 1119 * through the read path rather than recv. 1120 */ 1121 if (pr->pr_domain->dom_dispose && 1122 mtod(m, struct cmsghdr *)->cmsg_type == SCM_RIGHTS) 1123 (*pr->pr_domain->dom_dispose)(m); 1124 MFREE(m, so->so_rcv.sb_mb); 1125 m = so->so_rcv.sb_mb; 1126 } 1127 } 1128 if (controlp) { 1129 orig_resid = 0; 1130 controlp = &(*controlp)->m_next; 1131 } 1132 } 1133 1134 /* 1135 * If m is non-NULL, we have some data to read. From now on, 1136 * make sure to keep sb_lastrecord consistent when working on 1137 * the last packet on the chain (nextrecord == NULL) and we 1138 * change m->m_nextpkt. 1139 */ 1140 if (m) { 1141 if ((flags & MSG_PEEK) == 0) { 1142 m->m_nextpkt = nextrecord; 1143 /* 1144 * If nextrecord == NULL (this is a single chain), 1145 * then sb_lastrecord may not be valid here if m 1146 * was changed earlier. 1147 */ 1148 if (nextrecord == NULL) { 1149 KASSERT(so->so_rcv.sb_mb == m); 1150 so->so_rcv.sb_lastrecord = m; 1151 } 1152 } 1153 type = m->m_type; 1154 if (type == MT_OOBDATA) 1155 flags |= MSG_OOB; 1156 } else { 1157 if ((flags & MSG_PEEK) == 0) { 1158 KASSERT(so->so_rcv.sb_mb == m); 1159 so->so_rcv.sb_mb = nextrecord; 1160 SB_EMPTY_FIXUP(&so->so_rcv); 1161 } 1162 } 1163 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1164 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1165 1166 moff = 0; 1167 offset = 0; 1168 while (m && uio->uio_resid > 0 && error == 0) { 1169 if (m->m_type == MT_OOBDATA) { 1170 if (type != MT_OOBDATA) 1171 break; 1172 } else if (type == MT_OOBDATA) 1173 break; 1174 #ifdef DIAGNOSTIC 1175 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1176 panic("receive 3"); 1177 #endif 1178 so->so_state &= ~SS_RCVATMARK; 1179 len = uio->uio_resid; 1180 if (so->so_oobmark && len > so->so_oobmark - offset) 1181 len = so->so_oobmark - offset; 1182 if (len > m->m_len - moff) 1183 len = m->m_len - moff; 1184 /* 1185 * If mp is set, just pass back the mbufs. 1186 * Otherwise copy them out via the uio, then free. 1187 * Sockbuf must be consistent here (points to current mbuf, 1188 * it points to next record) when we drop priority; 1189 * we must note any additions to the sockbuf when we 1190 * block interrupts again. 1191 */ 1192 if (mp == 0) { 1193 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1194 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1195 splx(s); 1196 error = uiomove(mtod(m, char *) + moff, (int)len, uio); 1197 s = splsoftnet(); 1198 if (error) { 1199 /* 1200 * If any part of the record has been removed 1201 * (such as the MT_SONAME mbuf, which will 1202 * happen when PR_ADDR, and thus also 1203 * PR_ATOMIC, is set), then drop the entire 1204 * record to maintain the atomicity of the 1205 * receive operation. 1206 * 1207 * This avoids a later panic("receive 1a") 1208 * when compiled with DIAGNOSTIC. 1209 */ 1210 if (m && mbuf_removed 1211 && (pr->pr_flags & PR_ATOMIC)) 1212 (void) sbdroprecord(&so->so_rcv); 1213 1214 goto release; 1215 } 1216 } else 1217 uio->uio_resid -= len; 1218 if (len == m->m_len - moff) { 1219 if (m->m_flags & M_EOR) 1220 flags |= MSG_EOR; 1221 if (flags & MSG_PEEK) { 1222 m = m->m_next; 1223 moff = 0; 1224 } else { 1225 nextrecord = m->m_nextpkt; 1226 sbfree(&so->so_rcv, m); 1227 if (mp) { 1228 *mp = m; 1229 mp = &m->m_next; 1230 so->so_rcv.sb_mb = m = m->m_next; 1231 *mp = (struct mbuf *)0; 1232 } else { 1233 MFREE(m, so->so_rcv.sb_mb); 1234 m = so->so_rcv.sb_mb; 1235 } 1236 /* 1237 * If m != NULL, we also know that 1238 * so->so_rcv.sb_mb != NULL. 1239 */ 1240 KASSERT(so->so_rcv.sb_mb == m); 1241 if (m) { 1242 m->m_nextpkt = nextrecord; 1243 if (nextrecord == NULL) 1244 so->so_rcv.sb_lastrecord = m; 1245 } else { 1246 so->so_rcv.sb_mb = nextrecord; 1247 SB_EMPTY_FIXUP(&so->so_rcv); 1248 } 1249 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1250 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1251 } 1252 } else { 1253 if (flags & MSG_PEEK) 1254 moff += len; 1255 else { 1256 if (mp) 1257 *mp = m_copym(m, 0, len, M_WAIT); 1258 m->m_data += len; 1259 m->m_len -= len; 1260 so->so_rcv.sb_cc -= len; 1261 } 1262 } 1263 if (so->so_oobmark) { 1264 if ((flags & MSG_PEEK) == 0) { 1265 so->so_oobmark -= len; 1266 if (so->so_oobmark == 0) { 1267 so->so_state |= SS_RCVATMARK; 1268 break; 1269 } 1270 } else { 1271 offset += len; 1272 if (offset == so->so_oobmark) 1273 break; 1274 } 1275 } 1276 if (flags & MSG_EOR) 1277 break; 1278 /* 1279 * If the MSG_WAITALL flag is set (for non-atomic socket), 1280 * we must not quit until "uio->uio_resid == 0" or an error 1281 * termination. If a signal/timeout occurs, return 1282 * with a short count but without error. 1283 * Keep sockbuf locked against other readers. 1284 */ 1285 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1286 !sosendallatonce(so) && !nextrecord) { 1287 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1288 break; 1289 /* 1290 * If we are peeking and the socket receive buffer is 1291 * full, stop since we can't get more data to peek at. 1292 */ 1293 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1294 break; 1295 /* 1296 * If we've drained the socket buffer, tell the 1297 * protocol in case it needs to do something to 1298 * get it filled again. 1299 */ 1300 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1301 (*pr->pr_usrreq)(so, PRU_RCVD, 1302 (struct mbuf *)0, 1303 (struct mbuf *)(long)flags, 1304 (struct mbuf *)0, l); 1305 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1306 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1307 error = sbwait(&so->so_rcv); 1308 if (error) { 1309 sbunlock(&so->so_rcv); 1310 splx(s); 1311 return (0); 1312 } 1313 if ((m = so->so_rcv.sb_mb) != NULL) 1314 nextrecord = m->m_nextpkt; 1315 } 1316 } 1317 1318 if (m && pr->pr_flags & PR_ATOMIC) { 1319 flags |= MSG_TRUNC; 1320 if ((flags & MSG_PEEK) == 0) 1321 (void) sbdroprecord(&so->so_rcv); 1322 } 1323 if ((flags & MSG_PEEK) == 0) { 1324 if (m == 0) { 1325 /* 1326 * First part is an inline SB_EMPTY_FIXUP(). Second 1327 * part makes sure sb_lastrecord is up-to-date if 1328 * there is still data in the socket buffer. 1329 */ 1330 so->so_rcv.sb_mb = nextrecord; 1331 if (so->so_rcv.sb_mb == NULL) { 1332 so->so_rcv.sb_mbtail = NULL; 1333 so->so_rcv.sb_lastrecord = NULL; 1334 } else if (nextrecord->m_nextpkt == NULL) 1335 so->so_rcv.sb_lastrecord = nextrecord; 1336 } 1337 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1338 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1339 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1340 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1341 (struct mbuf *)(long)flags, (struct mbuf *)0, l); 1342 } 1343 if (orig_resid == uio->uio_resid && orig_resid && 1344 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1345 sbunlock(&so->so_rcv); 1346 splx(s); 1347 goto restart; 1348 } 1349 1350 if (flagsp) 1351 *flagsp |= flags; 1352 release: 1353 sbunlock(&so->so_rcv); 1354 splx(s); 1355 return (error); 1356 } 1357 1358 int 1359 soshutdown(struct socket *so, int how) 1360 { 1361 const struct protosw *pr; 1362 1363 pr = so->so_proto; 1364 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1365 return (EINVAL); 1366 1367 if (how == SHUT_RD || how == SHUT_RDWR) 1368 sorflush(so); 1369 if (how == SHUT_WR || how == SHUT_RDWR) 1370 return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0, 1371 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1372 return (0); 1373 } 1374 1375 void 1376 sorflush(struct socket *so) 1377 { 1378 struct sockbuf *sb, asb; 1379 const struct protosw *pr; 1380 int s; 1381 1382 sb = &so->so_rcv; 1383 pr = so->so_proto; 1384 sb->sb_flags |= SB_NOINTR; 1385 (void) sblock(sb, M_WAITOK); 1386 s = splnet(); 1387 socantrcvmore(so); 1388 sbunlock(sb); 1389 asb = *sb; 1390 /* 1391 * Clear most of the sockbuf structure, but leave some of the 1392 * fields valid. 1393 */ 1394 memset(&sb->sb_startzero, 0, 1395 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1396 splx(s); 1397 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1398 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1399 sbrelease(&asb, so); 1400 } 1401 1402 int 1403 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0) 1404 { 1405 int error; 1406 struct mbuf *m; 1407 struct linger *l; 1408 1409 error = 0; 1410 m = m0; 1411 if (level != SOL_SOCKET) { 1412 if (so->so_proto && so->so_proto->pr_ctloutput) 1413 return ((*so->so_proto->pr_ctloutput) 1414 (PRCO_SETOPT, so, level, optname, &m0)); 1415 error = ENOPROTOOPT; 1416 } else { 1417 switch (optname) { 1418 1419 case SO_LINGER: 1420 if (m == NULL || m->m_len != sizeof(struct linger)) { 1421 error = EINVAL; 1422 goto bad; 1423 } 1424 l = mtod(m, struct linger *); 1425 if (l->l_linger < 0 || l->l_linger > USHRT_MAX || 1426 l->l_linger > (INT_MAX / hz)) { 1427 error = EDOM; 1428 goto bad; 1429 } 1430 so->so_linger = l->l_linger; 1431 if (l->l_onoff) 1432 so->so_options |= SO_LINGER; 1433 else 1434 so->so_options &= ~SO_LINGER; 1435 break; 1436 1437 case SO_DEBUG: 1438 case SO_KEEPALIVE: 1439 case SO_DONTROUTE: 1440 case SO_USELOOPBACK: 1441 case SO_BROADCAST: 1442 case SO_REUSEADDR: 1443 case SO_REUSEPORT: 1444 case SO_OOBINLINE: 1445 case SO_TIMESTAMP: 1446 if (m == NULL || m->m_len < sizeof(int)) { 1447 error = EINVAL; 1448 goto bad; 1449 } 1450 if (*mtod(m, int *)) 1451 so->so_options |= optname; 1452 else 1453 so->so_options &= ~optname; 1454 break; 1455 1456 case SO_SNDBUF: 1457 case SO_RCVBUF: 1458 case SO_SNDLOWAT: 1459 case SO_RCVLOWAT: 1460 { 1461 int optval; 1462 1463 if (m == NULL || m->m_len < sizeof(int)) { 1464 error = EINVAL; 1465 goto bad; 1466 } 1467 1468 /* 1469 * Values < 1 make no sense for any of these 1470 * options, so disallow them. 1471 */ 1472 optval = *mtod(m, int *); 1473 if (optval < 1) { 1474 error = EINVAL; 1475 goto bad; 1476 } 1477 1478 switch (optname) { 1479 1480 case SO_SNDBUF: 1481 case SO_RCVBUF: 1482 if (sbreserve(optname == SO_SNDBUF ? 1483 &so->so_snd : &so->so_rcv, 1484 (u_long) optval, so) == 0) { 1485 error = ENOBUFS; 1486 goto bad; 1487 } 1488 break; 1489 1490 /* 1491 * Make sure the low-water is never greater than 1492 * the high-water. 1493 */ 1494 case SO_SNDLOWAT: 1495 so->so_snd.sb_lowat = 1496 (optval > so->so_snd.sb_hiwat) ? 1497 so->so_snd.sb_hiwat : optval; 1498 break; 1499 case SO_RCVLOWAT: 1500 so->so_rcv.sb_lowat = 1501 (optval > so->so_rcv.sb_hiwat) ? 1502 so->so_rcv.sb_hiwat : optval; 1503 break; 1504 } 1505 break; 1506 } 1507 1508 case SO_SNDTIMEO: 1509 case SO_RCVTIMEO: 1510 { 1511 struct timeval *tv; 1512 int val; 1513 1514 if (m == NULL || m->m_len < sizeof(*tv)) { 1515 error = EINVAL; 1516 goto bad; 1517 } 1518 tv = mtod(m, struct timeval *); 1519 if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz) { 1520 error = EDOM; 1521 goto bad; 1522 } 1523 val = tv->tv_sec * hz + tv->tv_usec / tick; 1524 if (val == 0 && tv->tv_usec != 0) 1525 val = 1; 1526 1527 switch (optname) { 1528 1529 case SO_SNDTIMEO: 1530 so->so_snd.sb_timeo = val; 1531 break; 1532 case SO_RCVTIMEO: 1533 so->so_rcv.sb_timeo = val; 1534 break; 1535 } 1536 break; 1537 } 1538 1539 default: 1540 error = ENOPROTOOPT; 1541 break; 1542 } 1543 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1544 (void) ((*so->so_proto->pr_ctloutput) 1545 (PRCO_SETOPT, so, level, optname, &m0)); 1546 m = NULL; /* freed by protocol */ 1547 } 1548 } 1549 bad: 1550 if (m) 1551 (void) m_free(m); 1552 return (error); 1553 } 1554 1555 int 1556 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp) 1557 { 1558 struct mbuf *m; 1559 1560 if (level != SOL_SOCKET) { 1561 if (so->so_proto && so->so_proto->pr_ctloutput) { 1562 return ((*so->so_proto->pr_ctloutput) 1563 (PRCO_GETOPT, so, level, optname, mp)); 1564 } else 1565 return (ENOPROTOOPT); 1566 } else { 1567 m = m_get(M_WAIT, MT_SOOPTS); 1568 m->m_len = sizeof(int); 1569 1570 switch (optname) { 1571 1572 case SO_LINGER: 1573 m->m_len = sizeof(struct linger); 1574 mtod(m, struct linger *)->l_onoff = 1575 (so->so_options & SO_LINGER) ? 1 : 0; 1576 mtod(m, struct linger *)->l_linger = so->so_linger; 1577 break; 1578 1579 case SO_USELOOPBACK: 1580 case SO_DONTROUTE: 1581 case SO_DEBUG: 1582 case SO_KEEPALIVE: 1583 case SO_REUSEADDR: 1584 case SO_REUSEPORT: 1585 case SO_BROADCAST: 1586 case SO_OOBINLINE: 1587 case SO_TIMESTAMP: 1588 *mtod(m, int *) = (so->so_options & optname) ? 1 : 0; 1589 break; 1590 1591 case SO_TYPE: 1592 *mtod(m, int *) = so->so_type; 1593 break; 1594 1595 case SO_ERROR: 1596 *mtod(m, int *) = so->so_error; 1597 so->so_error = 0; 1598 break; 1599 1600 case SO_SNDBUF: 1601 *mtod(m, int *) = so->so_snd.sb_hiwat; 1602 break; 1603 1604 case SO_RCVBUF: 1605 *mtod(m, int *) = so->so_rcv.sb_hiwat; 1606 break; 1607 1608 case SO_SNDLOWAT: 1609 *mtod(m, int *) = so->so_snd.sb_lowat; 1610 break; 1611 1612 case SO_RCVLOWAT: 1613 *mtod(m, int *) = so->so_rcv.sb_lowat; 1614 break; 1615 1616 case SO_SNDTIMEO: 1617 case SO_RCVTIMEO: 1618 { 1619 int val = (optname == SO_SNDTIMEO ? 1620 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1621 1622 m->m_len = sizeof(struct timeval); 1623 mtod(m, struct timeval *)->tv_sec = val / hz; 1624 mtod(m, struct timeval *)->tv_usec = 1625 (val % hz) * tick; 1626 break; 1627 } 1628 1629 case SO_OVERFLOWED: 1630 *mtod(m, int *) = so->so_rcv.sb_overflowed; 1631 break; 1632 1633 default: 1634 (void)m_free(m); 1635 return (ENOPROTOOPT); 1636 } 1637 *mp = m; 1638 return (0); 1639 } 1640 } 1641 1642 void 1643 sohasoutofband(struct socket *so) 1644 { 1645 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 1646 selwakeup(&so->so_rcv.sb_sel); 1647 } 1648 1649 static void 1650 filt_sordetach(struct knote *kn) 1651 { 1652 struct socket *so; 1653 1654 so = (struct socket *)kn->kn_fp->f_data; 1655 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 1656 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 1657 so->so_rcv.sb_flags &= ~SB_KNOTE; 1658 } 1659 1660 /*ARGSUSED*/ 1661 static int 1662 filt_soread(struct knote *kn, long hint) 1663 { 1664 struct socket *so; 1665 1666 so = (struct socket *)kn->kn_fp->f_data; 1667 kn->kn_data = so->so_rcv.sb_cc; 1668 if (so->so_state & SS_CANTRCVMORE) { 1669 kn->kn_flags |= EV_EOF; 1670 kn->kn_fflags = so->so_error; 1671 return (1); 1672 } 1673 if (so->so_error) /* temporary udp error */ 1674 return (1); 1675 if (kn->kn_sfflags & NOTE_LOWAT) 1676 return (kn->kn_data >= kn->kn_sdata); 1677 return (kn->kn_data >= so->so_rcv.sb_lowat); 1678 } 1679 1680 static void 1681 filt_sowdetach(struct knote *kn) 1682 { 1683 struct socket *so; 1684 1685 so = (struct socket *)kn->kn_fp->f_data; 1686 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 1687 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 1688 so->so_snd.sb_flags &= ~SB_KNOTE; 1689 } 1690 1691 /*ARGSUSED*/ 1692 static int 1693 filt_sowrite(struct knote *kn, long hint) 1694 { 1695 struct socket *so; 1696 1697 so = (struct socket *)kn->kn_fp->f_data; 1698 kn->kn_data = sbspace(&so->so_snd); 1699 if (so->so_state & SS_CANTSENDMORE) { 1700 kn->kn_flags |= EV_EOF; 1701 kn->kn_fflags = so->so_error; 1702 return (1); 1703 } 1704 if (so->so_error) /* temporary udp error */ 1705 return (1); 1706 if (((so->so_state & SS_ISCONNECTED) == 0) && 1707 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1708 return (0); 1709 if (kn->kn_sfflags & NOTE_LOWAT) 1710 return (kn->kn_data >= kn->kn_sdata); 1711 return (kn->kn_data >= so->so_snd.sb_lowat); 1712 } 1713 1714 /*ARGSUSED*/ 1715 static int 1716 filt_solisten(struct knote *kn, long hint) 1717 { 1718 struct socket *so; 1719 1720 so = (struct socket *)kn->kn_fp->f_data; 1721 1722 /* 1723 * Set kn_data to number of incoming connections, not 1724 * counting partial (incomplete) connections. 1725 */ 1726 kn->kn_data = so->so_qlen; 1727 return (kn->kn_data > 0); 1728 } 1729 1730 static const struct filterops solisten_filtops = 1731 { 1, NULL, filt_sordetach, filt_solisten }; 1732 static const struct filterops soread_filtops = 1733 { 1, NULL, filt_sordetach, filt_soread }; 1734 static const struct filterops sowrite_filtops = 1735 { 1, NULL, filt_sowdetach, filt_sowrite }; 1736 1737 int 1738 soo_kqfilter(struct file *fp, struct knote *kn) 1739 { 1740 struct socket *so; 1741 struct sockbuf *sb; 1742 1743 so = (struct socket *)kn->kn_fp->f_data; 1744 switch (kn->kn_filter) { 1745 case EVFILT_READ: 1746 if (so->so_options & SO_ACCEPTCONN) 1747 kn->kn_fop = &solisten_filtops; 1748 else 1749 kn->kn_fop = &soread_filtops; 1750 sb = &so->so_rcv; 1751 break; 1752 case EVFILT_WRITE: 1753 kn->kn_fop = &sowrite_filtops; 1754 sb = &so->so_snd; 1755 break; 1756 default: 1757 return (1); 1758 } 1759 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 1760 sb->sb_flags |= SB_KNOTE; 1761 return (0); 1762 } 1763 1764 #include <sys/sysctl.h> 1765 1766 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 1767 1768 /* 1769 * sysctl helper routine for kern.somaxkva. ensures that the given 1770 * value is not too small. 1771 * (XXX should we maybe make sure it's not too large as well?) 1772 */ 1773 static int 1774 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 1775 { 1776 int error, new_somaxkva; 1777 struct sysctlnode node; 1778 1779 new_somaxkva = somaxkva; 1780 node = *rnode; 1781 node.sysctl_data = &new_somaxkva; 1782 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1783 if (error || newp == NULL) 1784 return (error); 1785 1786 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 1787 return (EINVAL); 1788 1789 mutex_enter(&so_pendfree_lock); 1790 somaxkva = new_somaxkva; 1791 cv_broadcast(&socurkva_cv); 1792 mutex_exit(&so_pendfree_lock); 1793 1794 return (error); 1795 } 1796 1797 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup") 1798 { 1799 1800 sysctl_createv(clog, 0, NULL, NULL, 1801 CTLFLAG_PERMANENT, 1802 CTLTYPE_NODE, "kern", NULL, 1803 NULL, 0, NULL, 0, 1804 CTL_KERN, CTL_EOL); 1805 1806 sysctl_createv(clog, 0, NULL, NULL, 1807 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1808 CTLTYPE_INT, "somaxkva", 1809 SYSCTL_DESCR("Maximum amount of kernel memory to be " 1810 "used for socket buffers"), 1811 sysctl_kern_somaxkva, 0, NULL, 0, 1812 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 1813 } 1814
Indexes created Sat Sep 20 22:09:52 GMT 2025