uipc_socket.c revision 1.108.2.2
1 /* $NetBSD: uipc_socket.c,v 1.108.2.2 2006/10/25 07:53:19 ghen Exp $ */ 2 3 /*- 4 * Copyright (c) 2002 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of Wasabi Systems, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 39 /* 40 * Copyright (c) 1982, 1986, 1988, 1990, 1993 41 * The Regents of the University of California. All rights reserved. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. Neither the name of the University nor the names of its contributors 52 * may be used to endorse or promote products derived from this software 53 * without specific prior written permission. 54 * 55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 65 * SUCH DAMAGE. 66 * 67 * @(#)uipc_socket.c 8.6 (Berkeley) 5/2/95 68 */ 69 70 #include <sys/cdefs.h> 71 __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.108.2.2 2006/10/25 07:53:19 ghen 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 95 #include <uvm/uvm.h> 96 97 POOL_INIT(socket_pool, sizeof(struct socket), 0, 0, 0, "sockpl", NULL); 98 99 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options"); 100 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 101 102 extern int somaxconn; /* patchable (XXX sysctl) */ 103 int somaxconn = SOMAXCONN; 104 105 #ifdef SOSEND_COUNTERS 106 #include <sys/device.h> 107 108 struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 109 NULL, "sosend", "loan big"); 110 struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 111 NULL, "sosend", "copy big"); 112 struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 113 NULL, "sosend", "copy small"); 114 struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 115 NULL, "sosend", "kva limit"); 116 117 #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++ 118 119 EVCNT_ATTACH_STATIC(sosend_loan_big); 120 EVCNT_ATTACH_STATIC(sosend_copy_big); 121 EVCNT_ATTACH_STATIC(sosend_copy_small); 122 EVCNT_ATTACH_STATIC(sosend_kvalimit); 123 #else 124 125 #define SOSEND_COUNTER_INCR(ev) /* nothing */ 126 127 #endif /* SOSEND_COUNTERS */ 128 129 void 130 soinit(void) 131 { 132 133 /* Set the initial adjusted socket buffer size. */ 134 if (sb_max_set(sb_max)) 135 panic("bad initial sb_max value: %lu\n", sb_max); 136 137 } 138 139 #ifdef SOSEND_NO_LOAN 140 int use_sosend_loan = 0; 141 #else 142 int use_sosend_loan = 1; 143 #endif 144 145 struct simplelock so_pendfree_slock = SIMPLELOCK_INITIALIZER; 146 struct mbuf *so_pendfree; 147 148 #ifndef SOMAXKVA 149 #define SOMAXKVA (16 * 1024 * 1024) 150 #endif 151 int somaxkva = SOMAXKVA; 152 int socurkva; 153 int sokvawaiters; 154 155 #define SOCK_LOAN_THRESH 4096 156 #define SOCK_LOAN_CHUNK 65536 157 158 static size_t sodopendfree(struct socket *); 159 static size_t sodopendfreel(struct socket *); 160 static __inline vsize_t sokvareserve(struct socket *, vsize_t); 161 static __inline void sokvaunreserve(vsize_t); 162 163 static __inline vsize_t 164 sokvareserve(struct socket *so, vsize_t len) 165 { 166 int s; 167 int error; 168 169 s = splvm(); 170 simple_lock(&so_pendfree_slock); 171 while (socurkva + len > somaxkva) { 172 size_t freed; 173 174 /* 175 * try to do pendfree. 176 */ 177 178 freed = sodopendfreel(so); 179 180 /* 181 * if some kva was freed, try again. 182 */ 183 184 if (freed) 185 continue; 186 187 SOSEND_COUNTER_INCR(&sosend_kvalimit); 188 sokvawaiters++; 189 error = ltsleep(&socurkva, PVM | PCATCH, "sokva", 0, 190 &so_pendfree_slock); 191 sokvawaiters--; 192 if (error) { 193 len = 0; 194 break; 195 } 196 } 197 socurkva += len; 198 simple_unlock(&so_pendfree_slock); 199 splx(s); 200 return len; 201 } 202 203 static __inline void 204 sokvaunreserve(vsize_t len) 205 { 206 int s; 207 208 s = splvm(); 209 simple_lock(&so_pendfree_slock); 210 socurkva -= len; 211 if (sokvawaiters) 212 wakeup(&socurkva); 213 simple_unlock(&so_pendfree_slock); 214 splx(s); 215 } 216 217 /* 218 * sokvaalloc: allocate kva for loan. 219 */ 220 221 vaddr_t 222 sokvaalloc(vsize_t len, struct socket *so) 223 { 224 vaddr_t lva; 225 226 /* 227 * reserve kva. 228 */ 229 230 if (sokvareserve(so, len) == 0) 231 return 0; 232 233 /* 234 * allocate kva. 235 */ 236 237 lva = uvm_km_valloc_wait(kernel_map, len); 238 if (lva == 0) { 239 sokvaunreserve(len); 240 return (0); 241 } 242 243 return lva; 244 } 245 246 /* 247 * sokvafree: free kva for loan. 248 */ 249 250 void 251 sokvafree(vaddr_t sva, vsize_t len) 252 { 253 254 /* 255 * free kva. 256 */ 257 258 uvm_km_free(kernel_map, sva, len); 259 260 /* 261 * unreserve kva. 262 */ 263 264 sokvaunreserve(len); 265 } 266 267 static void 268 sodoloanfree(struct vm_page **pgs, caddr_t buf, size_t size) 269 { 270 vaddr_t va, sva, eva; 271 vsize_t len; 272 paddr_t pa; 273 int i, npgs; 274 275 eva = round_page((vaddr_t) buf + size); 276 sva = trunc_page((vaddr_t) buf); 277 len = eva - sva; 278 npgs = len >> PAGE_SHIFT; 279 280 if (__predict_false(pgs == NULL)) { 281 pgs = alloca(npgs * sizeof(*pgs)); 282 283 for (i = 0, va = sva; va < eva; i++, va += PAGE_SIZE) { 284 if (pmap_extract(pmap_kernel(), va, &pa) == FALSE) 285 panic("sodoloanfree: va 0x%lx not mapped", va); 286 pgs[i] = PHYS_TO_VM_PAGE(pa); 287 } 288 } 289 290 pmap_kremove(sva, len); 291 pmap_update(pmap_kernel()); 292 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 293 sokvafree(sva, len); 294 } 295 296 static size_t 297 sodopendfree(struct socket *so) 298 { 299 int s; 300 size_t rv; 301 302 s = splvm(); 303 simple_lock(&so_pendfree_slock); 304 rv = sodopendfreel(so); 305 simple_unlock(&so_pendfree_slock); 306 splx(s); 307 308 return rv; 309 } 310 311 /* 312 * sodopendfreel: free mbufs on "pendfree" list. 313 * unlock and relock so_pendfree_slock when freeing mbufs. 314 * 315 * => called with so_pendfree_slock held. 316 * => called at splvm. 317 */ 318 319 static size_t 320 sodopendfreel(struct socket *so) 321 { 322 size_t rv = 0; 323 324 LOCK_ASSERT(simple_lock_held(&so_pendfree_slock)); 325 326 for (;;) { 327 struct mbuf *m; 328 struct mbuf *next; 329 330 m = so_pendfree; 331 if (m == NULL) 332 break; 333 so_pendfree = NULL; 334 simple_unlock(&so_pendfree_slock); 335 /* XXX splx */ 336 337 for (; m != NULL; m = next) { 338 next = m->m_next; 339 340 rv += m->m_ext.ext_size; 341 sodoloanfree((m->m_flags & M_EXT_PAGES) ? 342 m->m_ext.ext_pgs : NULL, m->m_ext.ext_buf, 343 m->m_ext.ext_size); 344 pool_cache_put(&mbpool_cache, m); 345 } 346 347 /* XXX splvm */ 348 simple_lock(&so_pendfree_slock); 349 } 350 351 return (rv); 352 } 353 354 void 355 soloanfree(struct mbuf *m, caddr_t buf, size_t size, void *arg) 356 { 357 int s; 358 359 if (m == NULL) { 360 361 /* 362 * called from MEXTREMOVE. 363 */ 364 365 sodoloanfree(NULL, buf, size); 366 return; 367 } 368 369 /* 370 * postpone freeing mbuf. 371 * 372 * we can't do it in interrupt context 373 * because we need to put kva back to kernel_map. 374 */ 375 376 s = splvm(); 377 simple_lock(&so_pendfree_slock); 378 m->m_next = so_pendfree; 379 so_pendfree = m; 380 if (sokvawaiters) 381 wakeup(&socurkva); 382 simple_unlock(&so_pendfree_slock); 383 splx(s); 384 } 385 386 static long 387 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 388 { 389 struct iovec *iov = uio->uio_iov; 390 vaddr_t sva, eva; 391 vsize_t len; 392 vaddr_t lva, va; 393 int npgs, i, error; 394 395 if (uio->uio_segflg != UIO_USERSPACE) 396 return (0); 397 398 if (iov->iov_len < (size_t) space) 399 space = iov->iov_len; 400 if (space > SOCK_LOAN_CHUNK) 401 space = SOCK_LOAN_CHUNK; 402 403 eva = round_page((vaddr_t) iov->iov_base + space); 404 sva = trunc_page((vaddr_t) iov->iov_base); 405 len = eva - sva; 406 npgs = len >> PAGE_SHIFT; 407 408 /* XXX KDASSERT */ 409 KASSERT(npgs <= M_EXT_MAXPAGES); 410 KASSERT(uio->uio_procp != NULL); 411 412 lva = sokvaalloc(len, so); 413 if (lva == 0) 414 return 0; 415 416 error = uvm_loan(&uio->uio_procp->p_vmspace->vm_map, sva, len, 417 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 418 if (error) { 419 sokvafree(lva, len); 420 return (0); 421 } 422 423 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 424 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 425 VM_PROT_READ); 426 pmap_update(pmap_kernel()); 427 428 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 429 430 MEXTADD(m, (caddr_t) lva, space, M_MBUF, soloanfree, so); 431 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 432 433 uio->uio_resid -= space; 434 /* uio_offset not updated, not set/used for write(2) */ 435 uio->uio_iov->iov_base = (caddr_t) uio->uio_iov->iov_base + space; 436 uio->uio_iov->iov_len -= space; 437 if (uio->uio_iov->iov_len == 0) { 438 uio->uio_iov++; 439 uio->uio_iovcnt--; 440 } 441 442 return (space); 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 proc *p) 455 { 456 const struct protosw *prp; 457 struct socket *so; 458 int error, s; 459 460 if (proto) 461 prp = pffindproto(dom, proto, type); 462 else 463 prp = pffindtype(dom, type); 464 if (prp == 0) { 465 /* no support for domain */ 466 if (pffinddomain(dom) == 0) 467 return (EAFNOSUPPORT); 468 /* no support for socket type */ 469 if (proto == 0 && type != 0) 470 return (EPROTOTYPE); 471 return (EPROTONOSUPPORT); 472 } 473 if (prp->pr_usrreq == 0) 474 return (EPROTONOSUPPORT); 475 if (prp->pr_type != type) 476 return (EPROTOTYPE); 477 s = splsoftnet(); 478 so = pool_get(&socket_pool, PR_WAITOK); 479 memset((caddr_t)so, 0, sizeof(*so)); 480 TAILQ_INIT(&so->so_q0); 481 TAILQ_INIT(&so->so_q); 482 so->so_type = type; 483 so->so_proto = prp; 484 so->so_send = sosend; 485 so->so_receive = soreceive; 486 #ifdef MBUFTRACE 487 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 488 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 489 so->so_mowner = &prp->pr_domain->dom_mowner; 490 #endif 491 if (p != 0) 492 so->so_uid = p->p_ucred->cr_uid; 493 else 494 so->so_uid = UID_MAX; 495 error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0, 496 (struct mbuf *)(long)proto, (struct mbuf *)0, p); 497 if (error) { 498 so->so_state |= SS_NOFDREF; 499 sofree(so); 500 splx(s); 501 return (error); 502 } 503 splx(s); 504 *aso = so; 505 return (0); 506 } 507 508 int 509 sobind(struct socket *so, struct mbuf *nam, struct proc *p) 510 { 511 int s, error; 512 513 s = splsoftnet(); 514 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0, 515 nam, (struct mbuf *)0, p); 516 splx(s); 517 return (error); 518 } 519 520 int 521 solisten(struct socket *so, int backlog) 522 { 523 int s, error; 524 525 s = splsoftnet(); 526 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0, 527 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 528 if (error) { 529 splx(s); 530 return (error); 531 } 532 if (TAILQ_EMPTY(&so->so_q)) 533 so->so_options |= SO_ACCEPTCONN; 534 if (backlog < 0) 535 backlog = 0; 536 so->so_qlimit = min(backlog, somaxconn); 537 splx(s); 538 return (0); 539 } 540 541 void 542 sofree(struct socket *so) 543 { 544 545 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 546 return; 547 if (so->so_head) { 548 /* 549 * We must not decommission a socket that's on the accept(2) 550 * queue. If we do, then accept(2) may hang after select(2) 551 * indicated that the listening socket was ready. 552 */ 553 if (!soqremque(so, 0)) 554 return; 555 } 556 if (so->so_rcv.sb_hiwat) 557 (void)chgsbsize(so->so_uid, &so->so_rcv.sb_hiwat, 0, 558 RLIM_INFINITY); 559 if (so->so_snd.sb_hiwat) 560 (void)chgsbsize(so->so_uid, &so->so_snd.sb_hiwat, 0, 561 RLIM_INFINITY); 562 sbrelease(&so->so_snd, so); 563 sorflush(so); 564 pool_put(&socket_pool, so); 565 } 566 567 /* 568 * Close a socket on last file table reference removal. 569 * Initiate disconnect if connected. 570 * Free socket when disconnect complete. 571 */ 572 int 573 soclose(struct socket *so) 574 { 575 struct socket *so2; 576 int s, error; 577 578 error = 0; 579 s = splsoftnet(); /* conservative */ 580 if (so->so_options & SO_ACCEPTCONN) { 581 while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 582 (void) soqremque(so2, 0); 583 (void) soabort(so2); 584 } 585 while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 586 (void) soqremque(so2, 1); 587 (void) soabort(so2); 588 } 589 } 590 if (so->so_pcb == 0) 591 goto discard; 592 if (so->so_state & SS_ISCONNECTED) { 593 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 594 error = sodisconnect(so); 595 if (error) 596 goto drop; 597 } 598 if (so->so_options & SO_LINGER) { 599 if ((so->so_state & SS_ISDISCONNECTING) && 600 (so->so_state & SS_NBIO)) 601 goto drop; 602 while (so->so_state & SS_ISCONNECTED) { 603 error = tsleep((caddr_t)&so->so_timeo, 604 PSOCK | PCATCH, netcls, 605 so->so_linger * hz); 606 if (error) 607 break; 608 } 609 } 610 } 611 drop: 612 if (so->so_pcb) { 613 int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 614 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 615 (struct proc *)0); 616 if (error == 0) 617 error = error2; 618 } 619 discard: 620 if (so->so_state & SS_NOFDREF) 621 panic("soclose: NOFDREF"); 622 so->so_state |= SS_NOFDREF; 623 sofree(so); 624 splx(s); 625 return (error); 626 } 627 628 /* 629 * Must be called at splsoftnet... 630 */ 631 int 632 soabort(struct socket *so) 633 { 634 635 return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0, 636 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 637 } 638 639 int 640 soaccept(struct socket *so, struct mbuf *nam) 641 { 642 int s, error; 643 644 error = 0; 645 s = splsoftnet(); 646 if ((so->so_state & SS_NOFDREF) == 0) 647 panic("soaccept: !NOFDREF"); 648 so->so_state &= ~SS_NOFDREF; 649 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 650 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 651 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 652 (struct mbuf *)0, nam, (struct mbuf *)0, (struct proc *)0); 653 else 654 error = ECONNABORTED; 655 656 splx(s); 657 return (error); 658 } 659 660 int 661 soconnect(struct socket *so, struct mbuf *nam, struct proc *p) 662 { 663 int s, error; 664 665 if (so->so_options & SO_ACCEPTCONN) 666 return (EOPNOTSUPP); 667 s = splsoftnet(); 668 /* 669 * If protocol is connection-based, can only connect once. 670 * Otherwise, if connected, try to disconnect first. 671 * This allows user to disconnect by connecting to, e.g., 672 * a null address. 673 */ 674 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 675 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 676 (error = sodisconnect(so)))) 677 error = EISCONN; 678 else 679 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 680 (struct mbuf *)0, nam, (struct mbuf *)0, p); 681 splx(s); 682 return (error); 683 } 684 685 int 686 soconnect2(struct socket *so1, struct socket *so2) 687 { 688 int s, error; 689 690 s = splsoftnet(); 691 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 692 (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0, 693 (struct proc *)0); 694 splx(s); 695 return (error); 696 } 697 698 int 699 sodisconnect(struct socket *so) 700 { 701 int s, error; 702 703 s = splsoftnet(); 704 if ((so->so_state & SS_ISCONNECTED) == 0) { 705 error = ENOTCONN; 706 goto bad; 707 } 708 if (so->so_state & SS_ISDISCONNECTING) { 709 error = EALREADY; 710 goto bad; 711 } 712 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 713 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 714 (struct proc *)0); 715 bad: 716 splx(s); 717 sodopendfree(so); 718 return (error); 719 } 720 721 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 722 /* 723 * Send on a socket. 724 * If send must go all at once and message is larger than 725 * send buffering, then hard error. 726 * Lock against other senders. 727 * If must go all at once and not enough room now, then 728 * inform user that this would block and do nothing. 729 * Otherwise, if nonblocking, send as much as possible. 730 * The data to be sent is described by "uio" if nonzero, 731 * otherwise by the mbuf chain "top" (which must be null 732 * if uio is not). Data provided in mbuf chain must be small 733 * enough to send all at once. 734 * 735 * Returns nonzero on error, timeout or signal; callers 736 * must check for short counts if EINTR/ERESTART are returned. 737 * Data and control buffers are freed on return. 738 */ 739 int 740 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 741 struct mbuf *control, int flags, struct proc *p) 742 { 743 struct mbuf **mp, *m; 744 long space, len, resid, clen, mlen; 745 int error, s, dontroute, atomic; 746 747 sodopendfree(so); 748 749 clen = 0; 750 atomic = sosendallatonce(so) || top; 751 if (uio) 752 resid = uio->uio_resid; 753 else 754 resid = top->m_pkthdr.len; 755 /* 756 * In theory resid should be unsigned. 757 * However, space must be signed, as it might be less than 0 758 * if we over-committed, and we must use a signed comparison 759 * of space and resid. On the other hand, a negative resid 760 * causes us to loop sending 0-length segments to the protocol. 761 */ 762 if (resid < 0) { 763 error = EINVAL; 764 goto out; 765 } 766 dontroute = 767 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 768 (so->so_proto->pr_flags & PR_ATOMIC); 769 if (p) 770 p->p_stats->p_ru.ru_msgsnd++; 771 if (control) 772 clen = control->m_len; 773 #define snderr(errno) { error = errno; splx(s); goto release; } 774 775 restart: 776 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 777 goto out; 778 do { 779 s = splsoftnet(); 780 if (so->so_state & SS_CANTSENDMORE) 781 snderr(EPIPE); 782 if (so->so_error) { 783 error = so->so_error; 784 so->so_error = 0; 785 splx(s); 786 goto release; 787 } 788 if ((so->so_state & SS_ISCONNECTED) == 0) { 789 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 790 if ((so->so_state & SS_ISCONFIRMING) == 0 && 791 !(resid == 0 && clen != 0)) 792 snderr(ENOTCONN); 793 } else if (addr == 0) 794 snderr(EDESTADDRREQ); 795 } 796 space = sbspace(&so->so_snd); 797 if (flags & MSG_OOB) 798 space += 1024; 799 if ((atomic && resid > so->so_snd.sb_hiwat) || 800 clen > so->so_snd.sb_hiwat) 801 snderr(EMSGSIZE); 802 if (space < resid + clen && 803 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 804 if (so->so_state & SS_NBIO) 805 snderr(EWOULDBLOCK); 806 sbunlock(&so->so_snd); 807 error = sbwait(&so->so_snd); 808 splx(s); 809 if (error) 810 goto out; 811 goto restart; 812 } 813 splx(s); 814 mp = ⊤ 815 space -= clen; 816 do { 817 if (uio == NULL) { 818 /* 819 * Data is prepackaged in "top". 820 */ 821 resid = 0; 822 if (flags & MSG_EOR) 823 top->m_flags |= M_EOR; 824 } else do { 825 if (top == 0) { 826 m = m_gethdr(M_WAIT, MT_DATA); 827 mlen = MHLEN; 828 m->m_pkthdr.len = 0; 829 m->m_pkthdr.rcvif = (struct ifnet *)0; 830 } else { 831 m = m_get(M_WAIT, MT_DATA); 832 mlen = MLEN; 833 } 834 MCLAIM(m, so->so_snd.sb_mowner); 835 if (use_sosend_loan && 836 uio->uio_iov->iov_len >= SOCK_LOAN_THRESH && 837 space >= SOCK_LOAN_THRESH && 838 (len = sosend_loan(so, uio, m, 839 space)) != 0) { 840 SOSEND_COUNTER_INCR(&sosend_loan_big); 841 space -= len; 842 goto have_data; 843 } 844 if (resid >= MINCLSIZE && space >= MCLBYTES) { 845 SOSEND_COUNTER_INCR(&sosend_copy_big); 846 m_clget(m, M_WAIT); 847 if ((m->m_flags & M_EXT) == 0) 848 goto nopages; 849 mlen = MCLBYTES; 850 if (atomic && top == 0) { 851 len = lmin(MCLBYTES - max_hdr, 852 resid); 853 m->m_data += max_hdr; 854 } else 855 len = lmin(MCLBYTES, resid); 856 space -= len; 857 } else { 858 nopages: 859 SOSEND_COUNTER_INCR(&sosend_copy_small); 860 len = lmin(lmin(mlen, resid), space); 861 space -= len; 862 /* 863 * For datagram protocols, leave room 864 * for protocol headers in first mbuf. 865 */ 866 if (atomic && top == 0 && len < mlen) 867 MH_ALIGN(m, len); 868 } 869 error = uiomove(mtod(m, caddr_t), (int)len, 870 uio); 871 have_data: 872 resid = uio->uio_resid; 873 m->m_len = len; 874 *mp = m; 875 top->m_pkthdr.len += len; 876 if (error) 877 goto release; 878 mp = &m->m_next; 879 if (resid <= 0) { 880 if (flags & MSG_EOR) 881 top->m_flags |= M_EOR; 882 break; 883 } 884 } while (space > 0 && atomic); 885 886 s = splsoftnet(); 887 888 if (so->so_state & SS_CANTSENDMORE) 889 snderr(EPIPE); 890 891 if (dontroute) 892 so->so_options |= SO_DONTROUTE; 893 if (resid > 0) 894 so->so_state |= SS_MORETOCOME; 895 error = (*so->so_proto->pr_usrreq)(so, 896 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 897 top, addr, control, p); 898 if (dontroute) 899 so->so_options &= ~SO_DONTROUTE; 900 if (resid > 0) 901 so->so_state &= ~SS_MORETOCOME; 902 splx(s); 903 904 clen = 0; 905 control = 0; 906 top = 0; 907 mp = ⊤ 908 if (error) 909 goto release; 910 } while (resid && space > 0); 911 } while (resid); 912 913 release: 914 sbunlock(&so->so_snd); 915 out: 916 if (top) 917 m_freem(top); 918 if (control) 919 m_freem(control); 920 return (error); 921 } 922 923 /* 924 * Implement receive operations on a socket. 925 * We depend on the way that records are added to the sockbuf 926 * by sbappend*. In particular, each record (mbufs linked through m_next) 927 * must begin with an address if the protocol so specifies, 928 * followed by an optional mbuf or mbufs containing ancillary data, 929 * and then zero or more mbufs of data. 930 * In order to avoid blocking network interrupts for the entire time here, 931 * we splx() while doing the actual copy to user space. 932 * Although the sockbuf is locked, new data may still be appended, 933 * and thus we must maintain consistency of the sockbuf during that time. 934 * 935 * The caller may receive the data as a single mbuf chain by supplying 936 * an mbuf **mp0 for use in returning the chain. The uio is then used 937 * only for the count in uio_resid. 938 */ 939 int 940 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 941 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 942 { 943 struct proc * p; 944 struct mbuf *m, **mp; 945 int flags, len, error, s, offset, moff, type, orig_resid; 946 const struct protosw *pr; 947 struct mbuf *nextrecord; 948 int mbuf_removed = 0; 949 950 pr = so->so_proto; 951 mp = mp0; 952 type = 0; 953 orig_resid = uio->uio_resid; 954 p = uio->uio_procp; 955 956 if (paddr) 957 *paddr = 0; 958 if (controlp) 959 *controlp = 0; 960 if (flagsp) 961 flags = *flagsp &~ MSG_EOR; 962 else 963 flags = 0; 964 965 if ((flags & MSG_DONTWAIT) == 0) 966 sodopendfree(so); 967 968 if (flags & MSG_OOB) { 969 m = m_get(M_WAIT, MT_DATA); 970 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 971 (struct mbuf *)(long)(flags & MSG_PEEK), 972 (struct mbuf *)0, p); 973 if (error) 974 goto bad; 975 do { 976 error = uiomove(mtod(m, caddr_t), 977 (int) min(uio->uio_resid, m->m_len), uio); 978 m = m_free(m); 979 } while (uio->uio_resid && error == 0 && m); 980 bad: 981 if (m) 982 m_freem(m); 983 return (error); 984 } 985 if (mp) 986 *mp = (struct mbuf *)0; 987 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 988 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 989 (struct mbuf *)0, (struct mbuf *)0, p); 990 991 restart: 992 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) 993 return (error); 994 s = splsoftnet(); 995 996 m = so->so_rcv.sb_mb; 997 /* 998 * If we have less data than requested, block awaiting more 999 * (subject to any timeout) if: 1000 * 1. the current count is less than the low water mark, 1001 * 2. MSG_WAITALL is set, and it is possible to do the entire 1002 * receive operation at once if we block (resid <= hiwat), or 1003 * 3. MSG_DONTWAIT is not set. 1004 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1005 * we have to do the receive in sections, and thus risk returning 1006 * a short count if a timeout or signal occurs after we start. 1007 */ 1008 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 1009 so->so_rcv.sb_cc < uio->uio_resid) && 1010 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1011 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1012 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1013 #ifdef DIAGNOSTIC 1014 if (m == 0 && so->so_rcv.sb_cc) 1015 panic("receive 1"); 1016 #endif 1017 if (so->so_error) { 1018 if (m) 1019 goto dontblock; 1020 error = so->so_error; 1021 if ((flags & MSG_PEEK) == 0) 1022 so->so_error = 0; 1023 goto release; 1024 } 1025 if (so->so_state & SS_CANTRCVMORE) { 1026 if (m) 1027 goto dontblock; 1028 else 1029 goto release; 1030 } 1031 for (; m; m = m->m_next) 1032 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1033 m = so->so_rcv.sb_mb; 1034 goto dontblock; 1035 } 1036 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1037 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1038 error = ENOTCONN; 1039 goto release; 1040 } 1041 if (uio->uio_resid == 0) 1042 goto release; 1043 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 1044 error = EWOULDBLOCK; 1045 goto release; 1046 } 1047 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1048 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1049 sbunlock(&so->so_rcv); 1050 error = sbwait(&so->so_rcv); 1051 splx(s); 1052 if (error) 1053 return (error); 1054 goto restart; 1055 } 1056 dontblock: 1057 /* 1058 * On entry here, m points to the first record of the socket buffer. 1059 * While we process the initial mbufs containing address and control 1060 * info, we save a copy of m->m_nextpkt into nextrecord. 1061 */ 1062 if (p) 1063 p->p_stats->p_ru.ru_msgrcv++; 1064 KASSERT(m == so->so_rcv.sb_mb); 1065 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1066 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1067 nextrecord = m->m_nextpkt; 1068 if (pr->pr_flags & PR_ADDR) { 1069 #ifdef DIAGNOSTIC 1070 if (m->m_type != MT_SONAME) 1071 panic("receive 1a"); 1072 #endif 1073 orig_resid = 0; 1074 if (flags & MSG_PEEK) { 1075 if (paddr) 1076 *paddr = m_copy(m, 0, m->m_len); 1077 m = m->m_next; 1078 } else { 1079 sbfree(&so->so_rcv, m); 1080 mbuf_removed = 1; 1081 if (paddr) { 1082 *paddr = m; 1083 so->so_rcv.sb_mb = m->m_next; 1084 m->m_next = 0; 1085 m = so->so_rcv.sb_mb; 1086 } else { 1087 MFREE(m, so->so_rcv.sb_mb); 1088 m = so->so_rcv.sb_mb; 1089 } 1090 } 1091 } 1092 while (m && m->m_type == MT_CONTROL && error == 0) { 1093 if (flags & MSG_PEEK) { 1094 if (controlp) 1095 *controlp = m_copy(m, 0, m->m_len); 1096 m = m->m_next; 1097 } else { 1098 sbfree(&so->so_rcv, m); 1099 mbuf_removed = 1; 1100 if (controlp) { 1101 struct domain *dom = pr->pr_domain; 1102 if (dom->dom_externalize && p && 1103 mtod(m, struct cmsghdr *)->cmsg_type == 1104 SCM_RIGHTS) 1105 error = (*dom->dom_externalize)(m, p); 1106 *controlp = m; 1107 so->so_rcv.sb_mb = m->m_next; 1108 m->m_next = 0; 1109 m = so->so_rcv.sb_mb; 1110 } else { 1111 /* 1112 * Dispose of any SCM_RIGHTS message that went 1113 * through the read path rather than recv. 1114 */ 1115 if (pr->pr_domain->dom_dispose && 1116 mtod(m, struct cmsghdr *)->cmsg_type == SCM_RIGHTS) 1117 (*pr->pr_domain->dom_dispose)(m); 1118 MFREE(m, so->so_rcv.sb_mb); 1119 m = so->so_rcv.sb_mb; 1120 } 1121 } 1122 if (controlp) { 1123 orig_resid = 0; 1124 controlp = &(*controlp)->m_next; 1125 } 1126 } 1127 1128 /* 1129 * If m is non-NULL, we have some data to read. From now on, 1130 * make sure to keep sb_lastrecord consistent when working on 1131 * the last packet on the chain (nextrecord == NULL) and we 1132 * change m->m_nextpkt. 1133 */ 1134 if (m) { 1135 if ((flags & MSG_PEEK) == 0) { 1136 m->m_nextpkt = nextrecord; 1137 /* 1138 * If nextrecord == NULL (this is a single chain), 1139 * then sb_lastrecord may not be valid here if m 1140 * was changed earlier. 1141 */ 1142 if (nextrecord == NULL) { 1143 KASSERT(so->so_rcv.sb_mb == m); 1144 so->so_rcv.sb_lastrecord = m; 1145 } 1146 } 1147 type = m->m_type; 1148 if (type == MT_OOBDATA) 1149 flags |= MSG_OOB; 1150 } else { 1151 if ((flags & MSG_PEEK) == 0) { 1152 KASSERT(so->so_rcv.sb_mb == m); 1153 so->so_rcv.sb_mb = nextrecord; 1154 SB_EMPTY_FIXUP(&so->so_rcv); 1155 } 1156 } 1157 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1158 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1159 1160 moff = 0; 1161 offset = 0; 1162 while (m && uio->uio_resid > 0 && error == 0) { 1163 if (m->m_type == MT_OOBDATA) { 1164 if (type != MT_OOBDATA) 1165 break; 1166 } else if (type == MT_OOBDATA) 1167 break; 1168 #ifdef DIAGNOSTIC 1169 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1170 panic("receive 3"); 1171 #endif 1172 so->so_state &= ~SS_RCVATMARK; 1173 len = uio->uio_resid; 1174 if (so->so_oobmark && len > so->so_oobmark - offset) 1175 len = so->so_oobmark - offset; 1176 if (len > m->m_len - moff) 1177 len = m->m_len - moff; 1178 /* 1179 * If mp is set, just pass back the mbufs. 1180 * Otherwise copy them out via the uio, then free. 1181 * Sockbuf must be consistent here (points to current mbuf, 1182 * it points to next record) when we drop priority; 1183 * we must note any additions to the sockbuf when we 1184 * block interrupts again. 1185 */ 1186 if (mp == 0) { 1187 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1188 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1189 splx(s); 1190 error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio); 1191 s = splsoftnet(); 1192 if (error) { 1193 /* 1194 * If any part of the record has been removed 1195 * (such as the MT_SONAME mbuf, which will 1196 * happen when PR_ADDR, and thus also 1197 * PR_ATOMIC, is set), then drop the entire 1198 * record to maintain the atomicity of the 1199 * receive operation. 1200 * 1201 * This avoids a later panic("receive 1a") 1202 * when compiled with DIAGNOSTIC. 1203 */ 1204 if (m && mbuf_removed 1205 && (pr->pr_flags & PR_ATOMIC)) 1206 (void) sbdroprecord(&so->so_rcv); 1207 1208 goto release; 1209 } 1210 } else 1211 uio->uio_resid -= len; 1212 if (len == m->m_len - moff) { 1213 if (m->m_flags & M_EOR) 1214 flags |= MSG_EOR; 1215 if (flags & MSG_PEEK) { 1216 m = m->m_next; 1217 moff = 0; 1218 } else { 1219 nextrecord = m->m_nextpkt; 1220 sbfree(&so->so_rcv, m); 1221 if (mp) { 1222 *mp = m; 1223 mp = &m->m_next; 1224 so->so_rcv.sb_mb = m = m->m_next; 1225 *mp = (struct mbuf *)0; 1226 } else { 1227 MFREE(m, so->so_rcv.sb_mb); 1228 m = so->so_rcv.sb_mb; 1229 } 1230 /* 1231 * If m != NULL, we also know that 1232 * so->so_rcv.sb_mb != NULL. 1233 */ 1234 KASSERT(so->so_rcv.sb_mb == m); 1235 if (m) { 1236 m->m_nextpkt = nextrecord; 1237 if (nextrecord == NULL) 1238 so->so_rcv.sb_lastrecord = m; 1239 } else { 1240 so->so_rcv.sb_mb = nextrecord; 1241 SB_EMPTY_FIXUP(&so->so_rcv); 1242 } 1243 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1244 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1245 } 1246 } else { 1247 if (flags & MSG_PEEK) 1248 moff += len; 1249 else { 1250 if (mp) 1251 *mp = m_copym(m, 0, len, M_WAIT); 1252 m->m_data += len; 1253 m->m_len -= len; 1254 so->so_rcv.sb_cc -= len; 1255 } 1256 } 1257 if (so->so_oobmark) { 1258 if ((flags & MSG_PEEK) == 0) { 1259 so->so_oobmark -= len; 1260 if (so->so_oobmark == 0) { 1261 so->so_state |= SS_RCVATMARK; 1262 break; 1263 } 1264 } else { 1265 offset += len; 1266 if (offset == so->so_oobmark) 1267 break; 1268 } 1269 } 1270 if (flags & MSG_EOR) 1271 break; 1272 /* 1273 * If the MSG_WAITALL flag is set (for non-atomic socket), 1274 * we must not quit until "uio->uio_resid == 0" or an error 1275 * termination. If a signal/timeout occurs, return 1276 * with a short count but without error. 1277 * Keep sockbuf locked against other readers. 1278 */ 1279 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1280 !sosendallatonce(so) && !nextrecord) { 1281 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1282 break; 1283 /* 1284 * If we are peeking and the socket receive buffer is 1285 * full, stop since we can't get more data to peek at. 1286 */ 1287 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1288 break; 1289 /* 1290 * If we've drained the socket buffer, tell the 1291 * protocol in case it needs to do something to 1292 * get it filled again. 1293 */ 1294 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1295 (*pr->pr_usrreq)(so, PRU_RCVD, 1296 (struct mbuf *)0, 1297 (struct mbuf *)(long)flags, 1298 (struct mbuf *)0, p); 1299 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1300 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1301 error = sbwait(&so->so_rcv); 1302 if (error) { 1303 sbunlock(&so->so_rcv); 1304 splx(s); 1305 return (0); 1306 } 1307 if ((m = so->so_rcv.sb_mb) != NULL) 1308 nextrecord = m->m_nextpkt; 1309 } 1310 } 1311 1312 if (m && pr->pr_flags & PR_ATOMIC) { 1313 flags |= MSG_TRUNC; 1314 if ((flags & MSG_PEEK) == 0) 1315 (void) sbdroprecord(&so->so_rcv); 1316 } 1317 if ((flags & MSG_PEEK) == 0) { 1318 if (m == 0) { 1319 /* 1320 * First part is an inline SB_EMPTY_FIXUP(). Second 1321 * part makes sure sb_lastrecord is up-to-date if 1322 * there is still data in the socket buffer. 1323 */ 1324 so->so_rcv.sb_mb = nextrecord; 1325 if (so->so_rcv.sb_mb == NULL) { 1326 so->so_rcv.sb_mbtail = NULL; 1327 so->so_rcv.sb_lastrecord = NULL; 1328 } else if (nextrecord->m_nextpkt == NULL) 1329 so->so_rcv.sb_lastrecord = nextrecord; 1330 } 1331 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1332 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1333 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1334 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1335 (struct mbuf *)(long)flags, (struct mbuf *)0, p); 1336 } 1337 if (orig_resid == uio->uio_resid && orig_resid && 1338 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1339 sbunlock(&so->so_rcv); 1340 splx(s); 1341 goto restart; 1342 } 1343 1344 if (flagsp) 1345 *flagsp |= flags; 1346 release: 1347 sbunlock(&so->so_rcv); 1348 splx(s); 1349 return (error); 1350 } 1351 1352 int 1353 soshutdown(struct socket *so, int how) 1354 { 1355 const struct protosw *pr; 1356 1357 pr = so->so_proto; 1358 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1359 return (EINVAL); 1360 1361 if (how == SHUT_RD || how == SHUT_RDWR) 1362 sorflush(so); 1363 if (how == SHUT_WR || how == SHUT_RDWR) 1364 return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0, 1365 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1366 return (0); 1367 } 1368 1369 void 1370 sorflush(struct socket *so) 1371 { 1372 struct sockbuf *sb, asb; 1373 const struct protosw *pr; 1374 int s; 1375 1376 sb = &so->so_rcv; 1377 pr = so->so_proto; 1378 sb->sb_flags |= SB_NOINTR; 1379 (void) sblock(sb, M_WAITOK); 1380 s = splnet(); 1381 socantrcvmore(so); 1382 sbunlock(sb); 1383 asb = *sb; 1384 /* 1385 * Clear most of the sockbuf structure, but leave some of the 1386 * fields valid. 1387 */ 1388 memset(&sb->sb_startzero, 0, 1389 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1390 splx(s); 1391 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1392 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1393 sbrelease(&asb, so); 1394 } 1395 1396 int 1397 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0) 1398 { 1399 int error; 1400 struct mbuf *m; 1401 1402 error = 0; 1403 m = m0; 1404 if (level != SOL_SOCKET) { 1405 if (so->so_proto && so->so_proto->pr_ctloutput) 1406 return ((*so->so_proto->pr_ctloutput) 1407 (PRCO_SETOPT, so, level, optname, &m0)); 1408 error = ENOPROTOOPT; 1409 } else { 1410 switch (optname) { 1411 1412 case SO_LINGER: 1413 if (m == NULL || m->m_len != sizeof(struct linger)) { 1414 error = EINVAL; 1415 goto bad; 1416 } 1417 if (mtod(m, struct linger *)->l_linger < 0 || 1418 mtod(m, struct linger *)->l_linger > (INT_MAX / hz)) { 1419 error = EDOM; 1420 goto bad; 1421 } 1422 so->so_linger = mtod(m, struct linger *)->l_linger; 1423 /* fall thru... */ 1424 1425 case SO_DEBUG: 1426 case SO_KEEPALIVE: 1427 case SO_DONTROUTE: 1428 case SO_USELOOPBACK: 1429 case SO_BROADCAST: 1430 case SO_REUSEADDR: 1431 case SO_REUSEPORT: 1432 case SO_OOBINLINE: 1433 case SO_TIMESTAMP: 1434 if (m == NULL || m->m_len < sizeof(int)) { 1435 error = EINVAL; 1436 goto bad; 1437 } 1438 if (*mtod(m, int *)) 1439 so->so_options |= optname; 1440 else 1441 so->so_options &= ~optname; 1442 break; 1443 1444 case SO_SNDBUF: 1445 case SO_RCVBUF: 1446 case SO_SNDLOWAT: 1447 case SO_RCVLOWAT: 1448 { 1449 int optval; 1450 1451 if (m == NULL || m->m_len < sizeof(int)) { 1452 error = EINVAL; 1453 goto bad; 1454 } 1455 1456 /* 1457 * Values < 1 make no sense for any of these 1458 * options, so disallow them. 1459 */ 1460 optval = *mtod(m, int *); 1461 if (optval < 1) { 1462 error = EINVAL; 1463 goto bad; 1464 } 1465 1466 switch (optname) { 1467 1468 case SO_SNDBUF: 1469 case SO_RCVBUF: 1470 if (sbreserve(optname == SO_SNDBUF ? 1471 &so->so_snd : &so->so_rcv, 1472 (u_long) optval, so) == 0) { 1473 error = ENOBUFS; 1474 goto bad; 1475 } 1476 break; 1477 1478 /* 1479 * Make sure the low-water is never greater than 1480 * the high-water. 1481 */ 1482 case SO_SNDLOWAT: 1483 so->so_snd.sb_lowat = 1484 (optval > so->so_snd.sb_hiwat) ? 1485 so->so_snd.sb_hiwat : optval; 1486 break; 1487 case SO_RCVLOWAT: 1488 so->so_rcv.sb_lowat = 1489 (optval > so->so_rcv.sb_hiwat) ? 1490 so->so_rcv.sb_hiwat : optval; 1491 break; 1492 } 1493 break; 1494 } 1495 1496 case SO_SNDTIMEO: 1497 case SO_RCVTIMEO: 1498 { 1499 struct timeval *tv; 1500 int val; 1501 1502 if (m == NULL || m->m_len < sizeof(*tv)) { 1503 error = EINVAL; 1504 goto bad; 1505 } 1506 tv = mtod(m, struct timeval *); 1507 if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz) { 1508 error = EDOM; 1509 goto bad; 1510 } 1511 val = tv->tv_sec * hz + tv->tv_usec / tick; 1512 if (val == 0 && tv->tv_usec != 0) 1513 val = 1; 1514 1515 switch (optname) { 1516 1517 case SO_SNDTIMEO: 1518 so->so_snd.sb_timeo = val; 1519 break; 1520 case SO_RCVTIMEO: 1521 so->so_rcv.sb_timeo = val; 1522 break; 1523 } 1524 break; 1525 } 1526 1527 default: 1528 error = ENOPROTOOPT; 1529 break; 1530 } 1531 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1532 (void) ((*so->so_proto->pr_ctloutput) 1533 (PRCO_SETOPT, so, level, optname, &m0)); 1534 m = NULL; /* freed by protocol */ 1535 } 1536 } 1537 bad: 1538 if (m) 1539 (void) m_free(m); 1540 return (error); 1541 } 1542 1543 int 1544 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp) 1545 { 1546 struct mbuf *m; 1547 1548 if (level != SOL_SOCKET) { 1549 if (so->so_proto && so->so_proto->pr_ctloutput) { 1550 return ((*so->so_proto->pr_ctloutput) 1551 (PRCO_GETOPT, so, level, optname, mp)); 1552 } else 1553 return (ENOPROTOOPT); 1554 } else { 1555 m = m_get(M_WAIT, MT_SOOPTS); 1556 m->m_len = sizeof(int); 1557 1558 switch (optname) { 1559 1560 case SO_LINGER: 1561 m->m_len = sizeof(struct linger); 1562 mtod(m, struct linger *)->l_onoff = 1563 so->so_options & SO_LINGER; 1564 mtod(m, struct linger *)->l_linger = so->so_linger; 1565 break; 1566 1567 case SO_USELOOPBACK: 1568 case SO_DONTROUTE: 1569 case SO_DEBUG: 1570 case SO_KEEPALIVE: 1571 case SO_REUSEADDR: 1572 case SO_REUSEPORT: 1573 case SO_BROADCAST: 1574 case SO_OOBINLINE: 1575 case SO_TIMESTAMP: 1576 *mtod(m, int *) = so->so_options & optname; 1577 break; 1578 1579 case SO_TYPE: 1580 *mtod(m, int *) = so->so_type; 1581 break; 1582 1583 case SO_ERROR: 1584 *mtod(m, int *) = so->so_error; 1585 so->so_error = 0; 1586 break; 1587 1588 case SO_SNDBUF: 1589 *mtod(m, int *) = so->so_snd.sb_hiwat; 1590 break; 1591 1592 case SO_RCVBUF: 1593 *mtod(m, int *) = so->so_rcv.sb_hiwat; 1594 break; 1595 1596 case SO_SNDLOWAT: 1597 *mtod(m, int *) = so->so_snd.sb_lowat; 1598 break; 1599 1600 case SO_RCVLOWAT: 1601 *mtod(m, int *) = so->so_rcv.sb_lowat; 1602 break; 1603 1604 case SO_SNDTIMEO: 1605 case SO_RCVTIMEO: 1606 { 1607 int val = (optname == SO_SNDTIMEO ? 1608 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1609 1610 m->m_len = sizeof(struct timeval); 1611 mtod(m, struct timeval *)->tv_sec = val / hz; 1612 mtod(m, struct timeval *)->tv_usec = 1613 (val % hz) * tick; 1614 break; 1615 } 1616 1617 case SO_OVERFLOWED: 1618 *mtod(m, int *) = so->so_rcv.sb_overflowed; 1619 break; 1620 1621 default: 1622 (void)m_free(m); 1623 return (ENOPROTOOPT); 1624 } 1625 *mp = m; 1626 return (0); 1627 } 1628 } 1629 1630 void 1631 sohasoutofband(struct socket *so) 1632 { 1633 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 1634 selwakeup(&so->so_rcv.sb_sel); 1635 } 1636 1637 static void 1638 filt_sordetach(struct knote *kn) 1639 { 1640 struct socket *so; 1641 1642 so = (struct socket *)kn->kn_fp->f_data; 1643 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 1644 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 1645 so->so_rcv.sb_flags &= ~SB_KNOTE; 1646 } 1647 1648 /*ARGSUSED*/ 1649 static int 1650 filt_soread(struct knote *kn, long hint) 1651 { 1652 struct socket *so; 1653 1654 so = (struct socket *)kn->kn_fp->f_data; 1655 kn->kn_data = so->so_rcv.sb_cc; 1656 if (so->so_state & SS_CANTRCVMORE) { 1657 kn->kn_flags |= EV_EOF; 1658 kn->kn_fflags = so->so_error; 1659 return (1); 1660 } 1661 if (so->so_error) /* temporary udp error */ 1662 return (1); 1663 if (kn->kn_sfflags & NOTE_LOWAT) 1664 return (kn->kn_data >= kn->kn_sdata); 1665 return (kn->kn_data >= so->so_rcv.sb_lowat); 1666 } 1667 1668 static void 1669 filt_sowdetach(struct knote *kn) 1670 { 1671 struct socket *so; 1672 1673 so = (struct socket *)kn->kn_fp->f_data; 1674 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 1675 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 1676 so->so_snd.sb_flags &= ~SB_KNOTE; 1677 } 1678 1679 /*ARGSUSED*/ 1680 static int 1681 filt_sowrite(struct knote *kn, long hint) 1682 { 1683 struct socket *so; 1684 1685 so = (struct socket *)kn->kn_fp->f_data; 1686 kn->kn_data = sbspace(&so->so_snd); 1687 if (so->so_state & SS_CANTSENDMORE) { 1688 kn->kn_flags |= EV_EOF; 1689 kn->kn_fflags = so->so_error; 1690 return (1); 1691 } 1692 if (so->so_error) /* temporary udp error */ 1693 return (1); 1694 if (((so->so_state & SS_ISCONNECTED) == 0) && 1695 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1696 return (0); 1697 if (kn->kn_sfflags & NOTE_LOWAT) 1698 return (kn->kn_data >= kn->kn_sdata); 1699 return (kn->kn_data >= so->so_snd.sb_lowat); 1700 } 1701 1702 /*ARGSUSED*/ 1703 static int 1704 filt_solisten(struct knote *kn, long hint) 1705 { 1706 struct socket *so; 1707 1708 so = (struct socket *)kn->kn_fp->f_data; 1709 1710 /* 1711 * Set kn_data to number of incoming connections, not 1712 * counting partial (incomplete) connections. 1713 */ 1714 kn->kn_data = so->so_qlen; 1715 return (kn->kn_data > 0); 1716 } 1717 1718 static const struct filterops solisten_filtops = 1719 { 1, NULL, filt_sordetach, filt_solisten }; 1720 static const struct filterops soread_filtops = 1721 { 1, NULL, filt_sordetach, filt_soread }; 1722 static const struct filterops sowrite_filtops = 1723 { 1, NULL, filt_sowdetach, filt_sowrite }; 1724 1725 int 1726 soo_kqfilter(struct file *fp, struct knote *kn) 1727 { 1728 struct socket *so; 1729 struct sockbuf *sb; 1730 1731 so = (struct socket *)kn->kn_fp->f_data; 1732 switch (kn->kn_filter) { 1733 case EVFILT_READ: 1734 if (so->so_options & SO_ACCEPTCONN) 1735 kn->kn_fop = &solisten_filtops; 1736 else 1737 kn->kn_fop = &soread_filtops; 1738 sb = &so->so_rcv; 1739 break; 1740 case EVFILT_WRITE: 1741 kn->kn_fop = &sowrite_filtops; 1742 sb = &so->so_snd; 1743 break; 1744 default: 1745 return (1); 1746 } 1747 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 1748 sb->sb_flags |= SB_KNOTE; 1749 return (0); 1750 } 1751 1752 #include <sys/sysctl.h> 1753 1754 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 1755 1756 /* 1757 * sysctl helper routine for kern.somaxkva. ensures that the given 1758 * value is not too small. 1759 * (XXX should we maybe make sure it's not too large as well?) 1760 */ 1761 static int 1762 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 1763 { 1764 int error, new_somaxkva; 1765 struct sysctlnode node; 1766 int s; 1767 1768 new_somaxkva = somaxkva; 1769 node = *rnode; 1770 node.sysctl_data = &new_somaxkva; 1771 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1772 if (error || newp == NULL) 1773 return (error); 1774 1775 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 1776 return (EINVAL); 1777 1778 s = splvm(); 1779 simple_lock(&so_pendfree_slock); 1780 somaxkva = new_somaxkva; 1781 wakeup(&socurkva); 1782 simple_unlock(&so_pendfree_slock); 1783 splx(s); 1784 1785 return (error); 1786 } 1787 1788 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup") 1789 { 1790 1791 sysctl_createv(clog, 0, NULL, NULL, 1792 CTLFLAG_PERMANENT, 1793 CTLTYPE_NODE, "kern", NULL, 1794 NULL, 0, NULL, 0, 1795 CTL_KERN, CTL_EOL); 1796 1797 sysctl_createv(clog, 0, NULL, NULL, 1798 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1799 CTLTYPE_INT, "somaxkva", 1800 SYSCTL_DESCR("Maximum amount of kernel memory to be " 1801 "used for socket buffers"), 1802 sysctl_kern_somaxkva, 0, NULL, 0, 1803 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 1804 } 1805
Indexes created Sat Sep 20 22:09:52 GMT 2025