uipc_socket.c revision 1.100
1 /* $NetBSD: uipc_socket.c,v 1.100 2004/04/25 16:42:41 simonb 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 1.100 2004/04/25 16:42:41 simonb 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 #else 120 121 #define SOSEND_COUNTER_INCR(ev) /* nothing */ 122 123 #endif /* SOSEND_COUNTERS */ 124 125 void 126 soinit(void) 127 { 128 129 /* Set the initial adjusted socket buffer size. */ 130 if (sb_max_set(sb_max)) 131 panic("bad initial sb_max value: %lu\n", sb_max); 132 133 #ifdef SOSEND_COUNTERS 134 evcnt_attach_static(&sosend_loan_big); 135 evcnt_attach_static(&sosend_copy_big); 136 evcnt_attach_static(&sosend_copy_small); 137 evcnt_attach_static(&sosend_kvalimit); 138 #endif /* SOSEND_COUNTERS */ 139 } 140 141 #ifdef SOSEND_NO_LOAN 142 int use_sosend_loan = 0; 143 #else 144 int use_sosend_loan = 1; 145 #endif 146 147 struct simplelock so_pendfree_slock = SIMPLELOCK_INITIALIZER; 148 struct mbuf *so_pendfree; 149 150 #ifndef SOMAXKVA 151 #define SOMAXKVA (16 * 1024 * 1024) 152 #endif 153 int somaxkva = SOMAXKVA; 154 int socurkva; 155 int sokvawaiters; 156 157 #define SOCK_LOAN_THRESH 4096 158 #define SOCK_LOAN_CHUNK 65536 159 160 static size_t sodopendfree(struct socket *); 161 static size_t sodopendfreel(struct socket *); 162 static __inline vsize_t sokvareserve(struct socket *, vsize_t); 163 static __inline void sokvaunreserve(vsize_t); 164 165 static __inline vsize_t 166 sokvareserve(struct socket *so, vsize_t len) 167 { 168 int s; 169 int error; 170 171 s = splvm(); 172 simple_lock(&so_pendfree_slock); 173 while (socurkva + len > somaxkva) { 174 size_t freed; 175 176 /* 177 * try to do pendfree. 178 */ 179 180 freed = sodopendfreel(so); 181 182 /* 183 * if some kva was freed, try again. 184 */ 185 186 if (freed) 187 continue; 188 189 SOSEND_COUNTER_INCR(&sosend_kvalimit); 190 sokvawaiters++; 191 error = ltsleep(&socurkva, PVM | PCATCH, "sokva", 0, 192 &so_pendfree_slock); 193 sokvawaiters--; 194 if (error) { 195 len = 0; 196 break; 197 } 198 } 199 socurkva += len; 200 simple_unlock(&so_pendfree_slock); 201 splx(s); 202 return len; 203 } 204 205 static __inline void 206 sokvaunreserve(vsize_t len) 207 { 208 int s; 209 210 s = splvm(); 211 simple_lock(&so_pendfree_slock); 212 socurkva -= len; 213 if (sokvawaiters) 214 wakeup(&socurkva); 215 simple_unlock(&so_pendfree_slock); 216 splx(s); 217 } 218 219 /* 220 * sokvaalloc: allocate kva for loan. 221 */ 222 223 vaddr_t 224 sokvaalloc(vsize_t len, struct socket *so) 225 { 226 vaddr_t lva; 227 228 /* 229 * reserve kva. 230 */ 231 232 if (sokvareserve(so, len) == 0) 233 return 0; 234 235 /* 236 * allocate kva. 237 */ 238 239 lva = uvm_km_valloc_wait(kernel_map, len); 240 if (lva == 0) { 241 sokvaunreserve(len); 242 return (0); 243 } 244 245 return lva; 246 } 247 248 /* 249 * sokvafree: free kva for loan. 250 */ 251 252 void 253 sokvafree(vaddr_t sva, vsize_t len) 254 { 255 256 /* 257 * free kva. 258 */ 259 260 uvm_km_free(kernel_map, sva, len); 261 262 /* 263 * unreserve kva. 264 */ 265 266 sokvaunreserve(len); 267 } 268 269 static void 270 sodoloanfree(struct vm_page **pgs, caddr_t buf, size_t size) 271 { 272 vaddr_t va, sva, eva; 273 vsize_t len; 274 paddr_t pa; 275 int i, npgs; 276 277 eva = round_page((vaddr_t) buf + size); 278 sva = trunc_page((vaddr_t) buf); 279 len = eva - sva; 280 npgs = len >> PAGE_SHIFT; 281 282 if (__predict_false(pgs == NULL)) { 283 pgs = alloca(npgs * sizeof(*pgs)); 284 285 for (i = 0, va = sva; va < eva; i++, va += PAGE_SIZE) { 286 if (pmap_extract(pmap_kernel(), va, &pa) == FALSE) 287 panic("sodoloanfree: va 0x%lx not mapped", va); 288 pgs[i] = PHYS_TO_VM_PAGE(pa); 289 } 290 } 291 292 pmap_kremove(sva, len); 293 pmap_update(pmap_kernel()); 294 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 295 sokvafree(sva, len); 296 } 297 298 static size_t 299 sodopendfree(struct socket *so) 300 { 301 int s; 302 size_t rv; 303 304 s = splvm(); 305 simple_lock(&so_pendfree_slock); 306 rv = sodopendfreel(so); 307 simple_unlock(&so_pendfree_slock); 308 splx(s); 309 310 return rv; 311 } 312 313 /* 314 * sodopendfreel: free mbufs on "pendfree" list. 315 * unlock and relock so_pendfree_slock when freeing mbufs. 316 * 317 * => called with so_pendfree_slock held. 318 * => called at splvm. 319 */ 320 321 static size_t 322 sodopendfreel(struct socket *so) 323 { 324 size_t rv = 0; 325 326 LOCK_ASSERT(simple_lock_held(&so_pendfree_slock)); 327 328 for (;;) { 329 struct mbuf *m; 330 struct mbuf *next; 331 332 m = so_pendfree; 333 if (m == NULL) 334 break; 335 so_pendfree = NULL; 336 simple_unlock(&so_pendfree_slock); 337 /* XXX splx */ 338 339 for (; m != NULL; m = next) { 340 next = m->m_next; 341 342 rv += m->m_ext.ext_size; 343 sodoloanfree((m->m_flags & M_EXT_PAGES) ? 344 m->m_ext.ext_pgs : NULL, m->m_ext.ext_buf, 345 m->m_ext.ext_size); 346 pool_cache_put(&mbpool_cache, m); 347 } 348 349 /* XXX splvm */ 350 simple_lock(&so_pendfree_slock); 351 } 352 353 return (rv); 354 } 355 356 void 357 soloanfree(struct mbuf *m, caddr_t buf, size_t size, void *arg) 358 { 359 int s; 360 361 if (m == NULL) { 362 363 /* 364 * called from MEXTREMOVE. 365 */ 366 367 sodoloanfree(NULL, buf, size); 368 return; 369 } 370 371 /* 372 * postpone freeing mbuf. 373 * 374 * we can't do it in interrupt context 375 * because we need to put kva back to kernel_map. 376 */ 377 378 s = splvm(); 379 simple_lock(&so_pendfree_slock); 380 m->m_next = so_pendfree; 381 so_pendfree = m; 382 if (sokvawaiters) 383 wakeup(&socurkva); 384 simple_unlock(&so_pendfree_slock); 385 splx(s); 386 } 387 388 static long 389 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 390 { 391 struct iovec *iov = uio->uio_iov; 392 vaddr_t sva, eva; 393 vsize_t len; 394 vaddr_t lva, va; 395 int npgs, i, error; 396 397 if (uio->uio_segflg != UIO_USERSPACE) 398 return (0); 399 400 if (iov->iov_len < (size_t) space) 401 space = iov->iov_len; 402 if (space > SOCK_LOAN_CHUNK) 403 space = SOCK_LOAN_CHUNK; 404 405 eva = round_page((vaddr_t) iov->iov_base + space); 406 sva = trunc_page((vaddr_t) iov->iov_base); 407 len = eva - sva; 408 npgs = len >> PAGE_SHIFT; 409 410 /* XXX KDASSERT */ 411 KASSERT(npgs <= M_EXT_MAXPAGES); 412 413 lva = sokvaalloc(len, so); 414 if (lva == 0) 415 return 0; 416 417 error = uvm_loan(&uio->uio_procp->p_vmspace->vm_map, sva, len, 418 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 419 if (error) { 420 sokvafree(lva, len); 421 return (0); 422 } 423 424 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 425 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 426 VM_PROT_READ); 427 pmap_update(pmap_kernel()); 428 429 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 430 431 MEXTADD(m, (caddr_t) lva, space, M_MBUF, soloanfree, so); 432 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 433 434 uio->uio_resid -= space; 435 /* uio_offset not updated, not set/used for write(2) */ 436 uio->uio_iov->iov_base = (caddr_t) uio->uio_iov->iov_base + space; 437 uio->uio_iov->iov_len -= space; 438 if (uio->uio_iov->iov_len == 0) { 439 uio->uio_iov++; 440 uio->uio_iovcnt--; 441 } 442 443 return (space); 444 } 445 446 /* 447 * Socket operation routines. 448 * These routines are called by the routines in 449 * sys_socket.c or from a system process, and 450 * implement the semantics of socket operations by 451 * switching out to the protocol specific routines. 452 */ 453 /*ARGSUSED*/ 454 int 455 socreate(int dom, struct socket **aso, int type, int proto) 456 { 457 struct proc *p; 458 const struct protosw *prp; 459 struct socket *so; 460 int error, s; 461 462 p = curproc; /* XXX */ 463 if (proto) 464 prp = pffindproto(dom, proto, type); 465 else 466 prp = pffindtype(dom, type); 467 if (prp == 0 || prp->pr_usrreq == 0) 468 return (EPROTONOSUPPORT); 469 if (prp->pr_type != type) 470 return (EPROTOTYPE); 471 s = splsoftnet(); 472 so = pool_get(&socket_pool, PR_WAITOK); 473 memset((caddr_t)so, 0, sizeof(*so)); 474 TAILQ_INIT(&so->so_q0); 475 TAILQ_INIT(&so->so_q); 476 so->so_type = type; 477 so->so_proto = prp; 478 so->so_send = sosend; 479 so->so_receive = soreceive; 480 #ifdef MBUFTRACE 481 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 482 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 483 so->so_mowner = &prp->pr_domain->dom_mowner; 484 #endif 485 if (p != 0) 486 so->so_uid = p->p_ucred->cr_uid; 487 else 488 so->so_uid = UID_MAX; 489 error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0, 490 (struct mbuf *)(long)proto, (struct mbuf *)0, p); 491 if (error) { 492 so->so_state |= SS_NOFDREF; 493 sofree(so); 494 splx(s); 495 return (error); 496 } 497 splx(s); 498 *aso = so; 499 return (0); 500 } 501 502 int 503 sobind(struct socket *so, struct mbuf *nam, struct proc *p) 504 { 505 int s, error; 506 507 s = splsoftnet(); 508 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0, 509 nam, (struct mbuf *)0, p); 510 splx(s); 511 return (error); 512 } 513 514 int 515 solisten(struct socket *so, int backlog) 516 { 517 int s, error; 518 519 s = splsoftnet(); 520 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0, 521 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 522 if (error) { 523 splx(s); 524 return (error); 525 } 526 if (TAILQ_EMPTY(&so->so_q)) 527 so->so_options |= SO_ACCEPTCONN; 528 if (backlog < 0) 529 backlog = 0; 530 so->so_qlimit = min(backlog, somaxconn); 531 splx(s); 532 return (0); 533 } 534 535 void 536 sofree(struct socket *so) 537 { 538 539 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 540 return; 541 if (so->so_head) { 542 /* 543 * We must not decommission a socket that's on the accept(2) 544 * queue. If we do, then accept(2) may hang after select(2) 545 * indicated that the listening socket was ready. 546 */ 547 if (!soqremque(so, 0)) 548 return; 549 } 550 if (so->so_rcv.sb_hiwat) 551 (void)chgsbsize(so->so_uid, &so->so_rcv.sb_hiwat, 0, 552 RLIM_INFINITY); 553 if (so->so_snd.sb_hiwat) 554 (void)chgsbsize(so->so_uid, &so->so_snd.sb_hiwat, 0, 555 RLIM_INFINITY); 556 sbrelease(&so->so_snd, so); 557 sorflush(so); 558 pool_put(&socket_pool, so); 559 } 560 561 /* 562 * Close a socket on last file table reference removal. 563 * Initiate disconnect if connected. 564 * Free socket when disconnect complete. 565 */ 566 int 567 soclose(struct socket *so) 568 { 569 struct socket *so2; 570 int s, error; 571 572 error = 0; 573 s = splsoftnet(); /* conservative */ 574 if (so->so_options & SO_ACCEPTCONN) { 575 while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 576 (void) soqremque(so2, 0); 577 (void) soabort(so2); 578 } 579 while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 580 (void) soqremque(so2, 1); 581 (void) soabort(so2); 582 } 583 } 584 if (so->so_pcb == 0) 585 goto discard; 586 if (so->so_state & SS_ISCONNECTED) { 587 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 588 error = sodisconnect(so); 589 if (error) 590 goto drop; 591 } 592 if (so->so_options & SO_LINGER) { 593 if ((so->so_state & SS_ISDISCONNECTING) && 594 (so->so_state & SS_NBIO)) 595 goto drop; 596 while (so->so_state & SS_ISCONNECTED) { 597 error = tsleep((caddr_t)&so->so_timeo, 598 PSOCK | PCATCH, netcls, 599 so->so_linger * hz); 600 if (error) 601 break; 602 } 603 } 604 } 605 drop: 606 if (so->so_pcb) { 607 int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 608 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 609 (struct proc *)0); 610 if (error == 0) 611 error = error2; 612 } 613 discard: 614 if (so->so_state & SS_NOFDREF) 615 panic("soclose: NOFDREF"); 616 so->so_state |= SS_NOFDREF; 617 sofree(so); 618 splx(s); 619 return (error); 620 } 621 622 /* 623 * Must be called at splsoftnet... 624 */ 625 int 626 soabort(struct socket *so) 627 { 628 629 return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0, 630 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 631 } 632 633 int 634 soaccept(struct socket *so, struct mbuf *nam) 635 { 636 int s, error; 637 638 error = 0; 639 s = splsoftnet(); 640 if ((so->so_state & SS_NOFDREF) == 0) 641 panic("soaccept: !NOFDREF"); 642 so->so_state &= ~SS_NOFDREF; 643 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 644 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 645 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 646 (struct mbuf *)0, nam, (struct mbuf *)0, (struct proc *)0); 647 else 648 error = ECONNABORTED; 649 650 splx(s); 651 return (error); 652 } 653 654 int 655 soconnect(struct socket *so, struct mbuf *nam) 656 { 657 struct proc *p; 658 int s, error; 659 660 p = curproc; /* XXX */ 661 if (so->so_options & SO_ACCEPTCONN) 662 return (EOPNOTSUPP); 663 s = splsoftnet(); 664 /* 665 * If protocol is connection-based, can only connect once. 666 * Otherwise, if connected, try to disconnect first. 667 * This allows user to disconnect by connecting to, e.g., 668 * a null address. 669 */ 670 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 671 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 672 (error = sodisconnect(so)))) 673 error = EISCONN; 674 else 675 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 676 (struct mbuf *)0, nam, (struct mbuf *)0, p); 677 splx(s); 678 return (error); 679 } 680 681 int 682 soconnect2(struct socket *so1, struct socket *so2) 683 { 684 int s, error; 685 686 s = splsoftnet(); 687 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 688 (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0, 689 (struct proc *)0); 690 splx(s); 691 return (error); 692 } 693 694 int 695 sodisconnect(struct socket *so) 696 { 697 int s, error; 698 699 s = splsoftnet(); 700 if ((so->so_state & SS_ISCONNECTED) == 0) { 701 error = ENOTCONN; 702 goto bad; 703 } 704 if (so->so_state & SS_ISDISCONNECTING) { 705 error = EALREADY; 706 goto bad; 707 } 708 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 709 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 710 (struct proc *)0); 711 bad: 712 splx(s); 713 sodopendfree(so); 714 return (error); 715 } 716 717 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 718 /* 719 * Send on a socket. 720 * If send must go all at once and message is larger than 721 * send buffering, then hard error. 722 * Lock against other senders. 723 * If must go all at once and not enough room now, then 724 * inform user that this would block and do nothing. 725 * Otherwise, if nonblocking, send as much as possible. 726 * The data to be sent is described by "uio" if nonzero, 727 * otherwise by the mbuf chain "top" (which must be null 728 * if uio is not). Data provided in mbuf chain must be small 729 * enough to send all at once. 730 * 731 * Returns nonzero on error, timeout or signal; callers 732 * must check for short counts if EINTR/ERESTART are returned. 733 * Data and control buffers are freed on return. 734 */ 735 int 736 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 737 struct mbuf *control, int flags) 738 { 739 struct proc *p; 740 struct mbuf **mp, *m; 741 long space, len, resid, clen, mlen; 742 int error, s, dontroute, atomic; 743 744 sodopendfree(so); 745 746 p = curproc; /* XXX */ 747 clen = 0; 748 atomic = sosendallatonce(so) || top; 749 if (uio) 750 resid = uio->uio_resid; 751 else 752 resid = top->m_pkthdr.len; 753 /* 754 * In theory resid should be unsigned. 755 * However, space must be signed, as it might be less than 0 756 * if we over-committed, and we must use a signed comparison 757 * of space and resid. On the other hand, a negative resid 758 * causes us to loop sending 0-length segments to the protocol. 759 */ 760 if (resid < 0) { 761 error = EINVAL; 762 goto out; 763 } 764 dontroute = 765 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 766 (so->so_proto->pr_flags & PR_ATOMIC); 767 p->p_stats->p_ru.ru_msgsnd++; 768 if (control) 769 clen = control->m_len; 770 #define snderr(errno) { error = errno; splx(s); goto release; } 771 772 restart: 773 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 774 goto out; 775 do { 776 s = splsoftnet(); 777 if (so->so_state & SS_CANTSENDMORE) 778 snderr(EPIPE); 779 if (so->so_error) { 780 error = so->so_error; 781 so->so_error = 0; 782 splx(s); 783 goto release; 784 } 785 if ((so->so_state & SS_ISCONNECTED) == 0) { 786 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 787 if ((so->so_state & SS_ISCONFIRMING) == 0 && 788 !(resid == 0 && clen != 0)) 789 snderr(ENOTCONN); 790 } else if (addr == 0) 791 snderr(EDESTADDRREQ); 792 } 793 space = sbspace(&so->so_snd); 794 if (flags & MSG_OOB) 795 space += 1024; 796 if ((atomic && resid > so->so_snd.sb_hiwat) || 797 clen > so->so_snd.sb_hiwat) 798 snderr(EMSGSIZE); 799 if (space < resid + clen && 800 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 801 if (so->so_state & SS_NBIO) 802 snderr(EWOULDBLOCK); 803 sbunlock(&so->so_snd); 804 error = sbwait(&so->so_snd); 805 splx(s); 806 if (error) 807 goto out; 808 goto restart; 809 } 810 splx(s); 811 mp = ⊤ 812 space -= clen; 813 do { 814 if (uio == NULL) { 815 /* 816 * Data is prepackaged in "top". 817 */ 818 resid = 0; 819 if (flags & MSG_EOR) 820 top->m_flags |= M_EOR; 821 } else do { 822 if (top == 0) { 823 m = m_gethdr(M_WAIT, MT_DATA); 824 mlen = MHLEN; 825 m->m_pkthdr.len = 0; 826 m->m_pkthdr.rcvif = (struct ifnet *)0; 827 } else { 828 m = m_get(M_WAIT, MT_DATA); 829 mlen = MLEN; 830 } 831 MCLAIM(m, so->so_snd.sb_mowner); 832 if (use_sosend_loan && 833 uio->uio_iov->iov_len >= SOCK_LOAN_THRESH && 834 space >= SOCK_LOAN_THRESH && 835 (len = sosend_loan(so, uio, m, 836 space)) != 0) { 837 SOSEND_COUNTER_INCR(&sosend_loan_big); 838 space -= len; 839 goto have_data; 840 } 841 if (resid >= MINCLSIZE && space >= MCLBYTES) { 842 SOSEND_COUNTER_INCR(&sosend_copy_big); 843 m_clget(m, M_WAIT); 844 if ((m->m_flags & M_EXT) == 0) 845 goto nopages; 846 mlen = MCLBYTES; 847 if (atomic && top == 0) { 848 len = lmin(MCLBYTES - max_hdr, 849 resid); 850 m->m_data += max_hdr; 851 } else 852 len = lmin(MCLBYTES, resid); 853 space -= len; 854 } else { 855 nopages: 856 SOSEND_COUNTER_INCR(&sosend_copy_small); 857 len = lmin(lmin(mlen, resid), space); 858 space -= len; 859 /* 860 * For datagram protocols, leave room 861 * for protocol headers in first mbuf. 862 */ 863 if (atomic && top == 0 && len < mlen) 864 MH_ALIGN(m, len); 865 } 866 error = uiomove(mtod(m, caddr_t), (int)len, 867 uio); 868 have_data: 869 resid = uio->uio_resid; 870 m->m_len = len; 871 *mp = m; 872 top->m_pkthdr.len += len; 873 if (error) 874 goto release; 875 mp = &m->m_next; 876 if (resid <= 0) { 877 if (flags & MSG_EOR) 878 top->m_flags |= M_EOR; 879 break; 880 } 881 } while (space > 0 && atomic); 882 883 s = splsoftnet(); 884 885 if (so->so_state & SS_CANTSENDMORE) 886 snderr(EPIPE); 887 888 if (dontroute) 889 so->so_options |= SO_DONTROUTE; 890 if (resid > 0) 891 so->so_state |= SS_MORETOCOME; 892 error = (*so->so_proto->pr_usrreq)(so, 893 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 894 top, addr, control, p); 895 if (dontroute) 896 so->so_options &= ~SO_DONTROUTE; 897 if (resid > 0) 898 so->so_state &= ~SS_MORETOCOME; 899 splx(s); 900 901 clen = 0; 902 control = 0; 903 top = 0; 904 mp = ⊤ 905 if (error) 906 goto release; 907 } while (resid && space > 0); 908 } while (resid); 909 910 release: 911 sbunlock(&so->so_snd); 912 out: 913 if (top) 914 m_freem(top); 915 if (control) 916 m_freem(control); 917 return (error); 918 } 919 920 /* 921 * Implement receive operations on a socket. 922 * We depend on the way that records are added to the sockbuf 923 * by sbappend*. In particular, each record (mbufs linked through m_next) 924 * must begin with an address if the protocol so specifies, 925 * followed by an optional mbuf or mbufs containing ancillary data, 926 * and then zero or more mbufs of data. 927 * In order to avoid blocking network interrupts for the entire time here, 928 * we splx() while doing the actual copy to user space. 929 * Although the sockbuf is locked, new data may still be appended, 930 * and thus we must maintain consistency of the sockbuf during that time. 931 * 932 * The caller may receive the data as a single mbuf chain by supplying 933 * an mbuf **mp0 for use in returning the chain. The uio is then used 934 * only for the count in uio_resid. 935 */ 936 int 937 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 938 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 939 { 940 struct mbuf *m, **mp; 941 int flags, len, error, s, offset, moff, type, orig_resid; 942 const struct protosw *pr; 943 struct mbuf *nextrecord; 944 int mbuf_removed = 0; 945 946 pr = so->so_proto; 947 mp = mp0; 948 type = 0; 949 orig_resid = uio->uio_resid; 950 if (paddr) 951 *paddr = 0; 952 if (controlp) 953 *controlp = 0; 954 if (flagsp) 955 flags = *flagsp &~ MSG_EOR; 956 else 957 flags = 0; 958 959 if ((flags & MSG_DONTWAIT) == 0) 960 sodopendfree(so); 961 962 if (flags & MSG_OOB) { 963 m = m_get(M_WAIT, MT_DATA); 964 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 965 (struct mbuf *)(long)(flags & MSG_PEEK), (struct mbuf *)0, 966 (struct proc *)0); 967 if (error) 968 goto bad; 969 do { 970 error = uiomove(mtod(m, caddr_t), 971 (int) min(uio->uio_resid, m->m_len), uio); 972 m = m_free(m); 973 } while (uio->uio_resid && error == 0 && m); 974 bad: 975 if (m) 976 m_freem(m); 977 return (error); 978 } 979 if (mp) 980 *mp = (struct mbuf *)0; 981 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 982 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 983 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 984 985 restart: 986 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) 987 return (error); 988 s = splsoftnet(); 989 990 m = so->so_rcv.sb_mb; 991 /* 992 * If we have less data than requested, block awaiting more 993 * (subject to any timeout) if: 994 * 1. the current count is less than the low water mark, 995 * 2. MSG_WAITALL is set, and it is possible to do the entire 996 * receive operation at once if we block (resid <= hiwat), or 997 * 3. MSG_DONTWAIT is not set. 998 * If MSG_WAITALL is set but resid is larger than the receive buffer, 999 * we have to do the receive in sections, and thus risk returning 1000 * a short count if a timeout or signal occurs after we start. 1001 */ 1002 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 1003 so->so_rcv.sb_cc < uio->uio_resid) && 1004 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1005 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1006 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1007 #ifdef DIAGNOSTIC 1008 if (m == 0 && so->so_rcv.sb_cc) 1009 panic("receive 1"); 1010 #endif 1011 if (so->so_error) { 1012 if (m) 1013 goto dontblock; 1014 error = so->so_error; 1015 if ((flags & MSG_PEEK) == 0) 1016 so->so_error = 0; 1017 goto release; 1018 } 1019 if (so->so_state & SS_CANTRCVMORE) { 1020 if (m) 1021 goto dontblock; 1022 else 1023 goto release; 1024 } 1025 for (; m; m = m->m_next) 1026 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1027 m = so->so_rcv.sb_mb; 1028 goto dontblock; 1029 } 1030 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1031 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1032 error = ENOTCONN; 1033 goto release; 1034 } 1035 if (uio->uio_resid == 0) 1036 goto release; 1037 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 1038 error = EWOULDBLOCK; 1039 goto release; 1040 } 1041 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1042 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1043 sbunlock(&so->so_rcv); 1044 error = sbwait(&so->so_rcv); 1045 splx(s); 1046 if (error) 1047 return (error); 1048 goto restart; 1049 } 1050 dontblock: 1051 /* 1052 * On entry here, m points to the first record of the socket buffer. 1053 * While we process the initial mbufs containing address and control 1054 * info, we save a copy of m->m_nextpkt into nextrecord. 1055 */ 1056 #ifdef notyet /* XXXX */ 1057 if (uio->uio_procp) 1058 uio->uio_procp->p_stats->p_ru.ru_msgrcv++; 1059 #endif 1060 KASSERT(m == so->so_rcv.sb_mb); 1061 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1062 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1063 nextrecord = m->m_nextpkt; 1064 if (pr->pr_flags & PR_ADDR) { 1065 #ifdef DIAGNOSTIC 1066 if (m->m_type != MT_SONAME) 1067 panic("receive 1a"); 1068 #endif 1069 orig_resid = 0; 1070 if (flags & MSG_PEEK) { 1071 if (paddr) 1072 *paddr = m_copy(m, 0, m->m_len); 1073 m = m->m_next; 1074 } else { 1075 sbfree(&so->so_rcv, m); 1076 mbuf_removed = 1; 1077 if (paddr) { 1078 *paddr = m; 1079 so->so_rcv.sb_mb = m->m_next; 1080 m->m_next = 0; 1081 m = so->so_rcv.sb_mb; 1082 } else { 1083 MFREE(m, so->so_rcv.sb_mb); 1084 m = so->so_rcv.sb_mb; 1085 } 1086 } 1087 } 1088 while (m && m->m_type == MT_CONTROL && error == 0) { 1089 if (flags & MSG_PEEK) { 1090 if (controlp) 1091 *controlp = m_copy(m, 0, m->m_len); 1092 m = m->m_next; 1093 } else { 1094 sbfree(&so->so_rcv, m); 1095 mbuf_removed = 1; 1096 if (controlp) { 1097 if (pr->pr_domain->dom_externalize && 1098 mtod(m, struct cmsghdr *)->cmsg_type == 1099 SCM_RIGHTS) 1100 error = (*pr->pr_domain->dom_externalize)(m); 1101 *controlp = m; 1102 so->so_rcv.sb_mb = m->m_next; 1103 m->m_next = 0; 1104 m = so->so_rcv.sb_mb; 1105 } else { 1106 MFREE(m, so->so_rcv.sb_mb); 1107 m = so->so_rcv.sb_mb; 1108 } 1109 } 1110 if (controlp) { 1111 orig_resid = 0; 1112 controlp = &(*controlp)->m_next; 1113 } 1114 } 1115 1116 /* 1117 * If m is non-NULL, we have some data to read. From now on, 1118 * make sure to keep sb_lastrecord consistent when working on 1119 * the last packet on the chain (nextrecord == NULL) and we 1120 * change m->m_nextpkt. 1121 */ 1122 if (m) { 1123 if ((flags & MSG_PEEK) == 0) { 1124 m->m_nextpkt = nextrecord; 1125 /* 1126 * If nextrecord == NULL (this is a single chain), 1127 * then sb_lastrecord may not be valid here if m 1128 * was changed earlier. 1129 */ 1130 if (nextrecord == NULL) { 1131 KASSERT(so->so_rcv.sb_mb == m); 1132 so->so_rcv.sb_lastrecord = m; 1133 } 1134 } 1135 type = m->m_type; 1136 if (type == MT_OOBDATA) 1137 flags |= MSG_OOB; 1138 } else { 1139 if ((flags & MSG_PEEK) == 0) { 1140 KASSERT(so->so_rcv.sb_mb == m); 1141 so->so_rcv.sb_mb = nextrecord; 1142 SB_EMPTY_FIXUP(&so->so_rcv); 1143 } 1144 } 1145 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1146 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1147 1148 moff = 0; 1149 offset = 0; 1150 while (m && uio->uio_resid > 0 && error == 0) { 1151 if (m->m_type == MT_OOBDATA) { 1152 if (type != MT_OOBDATA) 1153 break; 1154 } else if (type == MT_OOBDATA) 1155 break; 1156 #ifdef DIAGNOSTIC 1157 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1158 panic("receive 3"); 1159 #endif 1160 so->so_state &= ~SS_RCVATMARK; 1161 len = uio->uio_resid; 1162 if (so->so_oobmark && len > so->so_oobmark - offset) 1163 len = so->so_oobmark - offset; 1164 if (len > m->m_len - moff) 1165 len = m->m_len - moff; 1166 /* 1167 * If mp is set, just pass back the mbufs. 1168 * Otherwise copy them out via the uio, then free. 1169 * Sockbuf must be consistent here (points to current mbuf, 1170 * it points to next record) when we drop priority; 1171 * we must note any additions to the sockbuf when we 1172 * block interrupts again. 1173 */ 1174 if (mp == 0) { 1175 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1176 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1177 splx(s); 1178 error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio); 1179 s = splsoftnet(); 1180 if (error) { 1181 /* 1182 * If any part of the record has been removed 1183 * (such as the MT_SONAME mbuf, which will 1184 * happen when PR_ADDR, and thus also 1185 * PR_ATOMIC, is set), then drop the entire 1186 * record to maintain the atomicity of the 1187 * receive operation. 1188 * 1189 * This avoids a later panic("receive 1a") 1190 * when compiled with DIAGNOSTIC. 1191 */ 1192 if (m && mbuf_removed 1193 && (pr->pr_flags & PR_ATOMIC)) 1194 (void) sbdroprecord(&so->so_rcv); 1195 1196 goto release; 1197 } 1198 } else 1199 uio->uio_resid -= len; 1200 if (len == m->m_len - moff) { 1201 if (m->m_flags & M_EOR) 1202 flags |= MSG_EOR; 1203 if (flags & MSG_PEEK) { 1204 m = m->m_next; 1205 moff = 0; 1206 } else { 1207 nextrecord = m->m_nextpkt; 1208 sbfree(&so->so_rcv, m); 1209 if (mp) { 1210 *mp = m; 1211 mp = &m->m_next; 1212 so->so_rcv.sb_mb = m = m->m_next; 1213 *mp = (struct mbuf *)0; 1214 } else { 1215 MFREE(m, so->so_rcv.sb_mb); 1216 m = so->so_rcv.sb_mb; 1217 } 1218 /* 1219 * If m != NULL, we also know that 1220 * so->so_rcv.sb_mb != NULL. 1221 */ 1222 KASSERT(so->so_rcv.sb_mb == m); 1223 if (m) { 1224 m->m_nextpkt = nextrecord; 1225 if (nextrecord == NULL) 1226 so->so_rcv.sb_lastrecord = m; 1227 } else { 1228 so->so_rcv.sb_mb = nextrecord; 1229 SB_EMPTY_FIXUP(&so->so_rcv); 1230 } 1231 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1232 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1233 } 1234 } else { 1235 if (flags & MSG_PEEK) 1236 moff += len; 1237 else { 1238 if (mp) 1239 *mp = m_copym(m, 0, len, M_WAIT); 1240 m->m_data += len; 1241 m->m_len -= len; 1242 so->so_rcv.sb_cc -= len; 1243 } 1244 } 1245 if (so->so_oobmark) { 1246 if ((flags & MSG_PEEK) == 0) { 1247 so->so_oobmark -= len; 1248 if (so->so_oobmark == 0) { 1249 so->so_state |= SS_RCVATMARK; 1250 break; 1251 } 1252 } else { 1253 offset += len; 1254 if (offset == so->so_oobmark) 1255 break; 1256 } 1257 } 1258 if (flags & MSG_EOR) 1259 break; 1260 /* 1261 * If the MSG_WAITALL flag is set (for non-atomic socket), 1262 * we must not quit until "uio->uio_resid == 0" or an error 1263 * termination. If a signal/timeout occurs, return 1264 * with a short count but without error. 1265 * Keep sockbuf locked against other readers. 1266 */ 1267 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1268 !sosendallatonce(so) && !nextrecord) { 1269 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1270 break; 1271 /* 1272 * If we are peeking and the socket receive buffer is 1273 * full, stop since we can't get more data to peek at. 1274 */ 1275 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1276 break; 1277 /* 1278 * If we've drained the socket buffer, tell the 1279 * protocol in case it needs to do something to 1280 * get it filled again. 1281 */ 1282 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1283 (*pr->pr_usrreq)(so, PRU_RCVD, 1284 (struct mbuf *)0, 1285 (struct mbuf *)(long)flags, 1286 (struct mbuf *)0, 1287 (struct proc *)0); 1288 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1289 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1290 error = sbwait(&so->so_rcv); 1291 if (error) { 1292 sbunlock(&so->so_rcv); 1293 splx(s); 1294 return (0); 1295 } 1296 if ((m = so->so_rcv.sb_mb) != NULL) 1297 nextrecord = m->m_nextpkt; 1298 } 1299 } 1300 1301 if (m && pr->pr_flags & PR_ATOMIC) { 1302 flags |= MSG_TRUNC; 1303 if ((flags & MSG_PEEK) == 0) 1304 (void) sbdroprecord(&so->so_rcv); 1305 } 1306 if ((flags & MSG_PEEK) == 0) { 1307 if (m == 0) { 1308 /* 1309 * First part is an inline SB_EMPTY_FIXUP(). Second 1310 * part makes sure sb_lastrecord is up-to-date if 1311 * there is still data in the socket buffer. 1312 */ 1313 so->so_rcv.sb_mb = nextrecord; 1314 if (so->so_rcv.sb_mb == NULL) { 1315 so->so_rcv.sb_mbtail = NULL; 1316 so->so_rcv.sb_lastrecord = NULL; 1317 } else if (nextrecord->m_nextpkt == NULL) 1318 so->so_rcv.sb_lastrecord = nextrecord; 1319 } 1320 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1321 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1322 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1323 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1324 (struct mbuf *)(long)flags, (struct mbuf *)0, 1325 (struct proc *)0); 1326 } 1327 if (orig_resid == uio->uio_resid && orig_resid && 1328 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1329 sbunlock(&so->so_rcv); 1330 splx(s); 1331 goto restart; 1332 } 1333 1334 if (flagsp) 1335 *flagsp |= flags; 1336 release: 1337 sbunlock(&so->so_rcv); 1338 splx(s); 1339 return (error); 1340 } 1341 1342 int 1343 soshutdown(struct socket *so, int how) 1344 { 1345 const struct protosw *pr; 1346 1347 pr = so->so_proto; 1348 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1349 return (EINVAL); 1350 1351 if (how == SHUT_RD || how == SHUT_RDWR) 1352 sorflush(so); 1353 if (how == SHUT_WR || how == SHUT_RDWR) 1354 return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0, 1355 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1356 return (0); 1357 } 1358 1359 void 1360 sorflush(struct socket *so) 1361 { 1362 struct sockbuf *sb, asb; 1363 const struct protosw *pr; 1364 int s; 1365 1366 sb = &so->so_rcv; 1367 pr = so->so_proto; 1368 sb->sb_flags |= SB_NOINTR; 1369 (void) sblock(sb, M_WAITOK); 1370 s = splnet(); 1371 socantrcvmore(so); 1372 sbunlock(sb); 1373 asb = *sb; 1374 /* 1375 * Clear most of the sockbuf structure, but leave some of the 1376 * fields valid. 1377 */ 1378 memset(&sb->sb_startzero, 0, 1379 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1380 splx(s); 1381 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1382 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1383 sbrelease(&asb, so); 1384 } 1385 1386 int 1387 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0) 1388 { 1389 int error; 1390 struct mbuf *m; 1391 1392 error = 0; 1393 m = m0; 1394 if (level != SOL_SOCKET) { 1395 if (so->so_proto && so->so_proto->pr_ctloutput) 1396 return ((*so->so_proto->pr_ctloutput) 1397 (PRCO_SETOPT, so, level, optname, &m0)); 1398 error = ENOPROTOOPT; 1399 } else { 1400 switch (optname) { 1401 1402 case SO_LINGER: 1403 if (m == NULL || m->m_len != sizeof(struct linger)) { 1404 error = EINVAL; 1405 goto bad; 1406 } 1407 so->so_linger = mtod(m, struct linger *)->l_linger; 1408 /* fall thru... */ 1409 1410 case SO_DEBUG: 1411 case SO_KEEPALIVE: 1412 case SO_DONTROUTE: 1413 case SO_USELOOPBACK: 1414 case SO_BROADCAST: 1415 case SO_REUSEADDR: 1416 case SO_REUSEPORT: 1417 case SO_OOBINLINE: 1418 case SO_TIMESTAMP: 1419 if (m == NULL || m->m_len < sizeof(int)) { 1420 error = EINVAL; 1421 goto bad; 1422 } 1423 if (*mtod(m, int *)) 1424 so->so_options |= optname; 1425 else 1426 so->so_options &= ~optname; 1427 break; 1428 1429 case SO_SNDBUF: 1430 case SO_RCVBUF: 1431 case SO_SNDLOWAT: 1432 case SO_RCVLOWAT: 1433 { 1434 int optval; 1435 1436 if (m == NULL || m->m_len < sizeof(int)) { 1437 error = EINVAL; 1438 goto bad; 1439 } 1440 1441 /* 1442 * Values < 1 make no sense for any of these 1443 * options, so disallow them. 1444 */ 1445 optval = *mtod(m, int *); 1446 if (optval < 1) { 1447 error = EINVAL; 1448 goto bad; 1449 } 1450 1451 switch (optname) { 1452 1453 case SO_SNDBUF: 1454 case SO_RCVBUF: 1455 if (sbreserve(optname == SO_SNDBUF ? 1456 &so->so_snd : &so->so_rcv, 1457 (u_long) optval, so) == 0) { 1458 error = ENOBUFS; 1459 goto bad; 1460 } 1461 break; 1462 1463 /* 1464 * Make sure the low-water is never greater than 1465 * the high-water. 1466 */ 1467 case SO_SNDLOWAT: 1468 so->so_snd.sb_lowat = 1469 (optval > so->so_snd.sb_hiwat) ? 1470 so->so_snd.sb_hiwat : optval; 1471 break; 1472 case SO_RCVLOWAT: 1473 so->so_rcv.sb_lowat = 1474 (optval > so->so_rcv.sb_hiwat) ? 1475 so->so_rcv.sb_hiwat : optval; 1476 break; 1477 } 1478 break; 1479 } 1480 1481 case SO_SNDTIMEO: 1482 case SO_RCVTIMEO: 1483 { 1484 struct timeval *tv; 1485 short val; 1486 1487 if (m == NULL || m->m_len < sizeof(*tv)) { 1488 error = EINVAL; 1489 goto bad; 1490 } 1491 tv = mtod(m, struct timeval *); 1492 if (tv->tv_sec > (SHRT_MAX - tv->tv_usec / tick) / hz) { 1493 error = EDOM; 1494 goto bad; 1495 } 1496 val = tv->tv_sec * hz + tv->tv_usec / tick; 1497 if (val == 0 && tv->tv_usec != 0) 1498 val = 1; 1499 1500 switch (optname) { 1501 1502 case SO_SNDTIMEO: 1503 so->so_snd.sb_timeo = val; 1504 break; 1505 case SO_RCVTIMEO: 1506 so->so_rcv.sb_timeo = val; 1507 break; 1508 } 1509 break; 1510 } 1511 1512 default: 1513 error = ENOPROTOOPT; 1514 break; 1515 } 1516 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1517 (void) ((*so->so_proto->pr_ctloutput) 1518 (PRCO_SETOPT, so, level, optname, &m0)); 1519 m = NULL; /* freed by protocol */ 1520 } 1521 } 1522 bad: 1523 if (m) 1524 (void) m_free(m); 1525 return (error); 1526 } 1527 1528 int 1529 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp) 1530 { 1531 struct mbuf *m; 1532 1533 if (level != SOL_SOCKET) { 1534 if (so->so_proto && so->so_proto->pr_ctloutput) { 1535 return ((*so->so_proto->pr_ctloutput) 1536 (PRCO_GETOPT, so, level, optname, mp)); 1537 } else 1538 return (ENOPROTOOPT); 1539 } else { 1540 m = m_get(M_WAIT, MT_SOOPTS); 1541 m->m_len = sizeof(int); 1542 1543 switch (optname) { 1544 1545 case SO_LINGER: 1546 m->m_len = sizeof(struct linger); 1547 mtod(m, struct linger *)->l_onoff = 1548 so->so_options & SO_LINGER; 1549 mtod(m, struct linger *)->l_linger = so->so_linger; 1550 break; 1551 1552 case SO_USELOOPBACK: 1553 case SO_DONTROUTE: 1554 case SO_DEBUG: 1555 case SO_KEEPALIVE: 1556 case SO_REUSEADDR: 1557 case SO_REUSEPORT: 1558 case SO_BROADCAST: 1559 case SO_OOBINLINE: 1560 case SO_TIMESTAMP: 1561 *mtod(m, int *) = so->so_options & optname; 1562 break; 1563 1564 case SO_TYPE: 1565 *mtod(m, int *) = so->so_type; 1566 break; 1567 1568 case SO_ERROR: 1569 *mtod(m, int *) = so->so_error; 1570 so->so_error = 0; 1571 break; 1572 1573 case SO_SNDBUF: 1574 *mtod(m, int *) = so->so_snd.sb_hiwat; 1575 break; 1576 1577 case SO_RCVBUF: 1578 *mtod(m, int *) = so->so_rcv.sb_hiwat; 1579 break; 1580 1581 case SO_SNDLOWAT: 1582 *mtod(m, int *) = so->so_snd.sb_lowat; 1583 break; 1584 1585 case SO_RCVLOWAT: 1586 *mtod(m, int *) = so->so_rcv.sb_lowat; 1587 break; 1588 1589 case SO_SNDTIMEO: 1590 case SO_RCVTIMEO: 1591 { 1592 int val = (optname == SO_SNDTIMEO ? 1593 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1594 1595 m->m_len = sizeof(struct timeval); 1596 mtod(m, struct timeval *)->tv_sec = val / hz; 1597 mtod(m, struct timeval *)->tv_usec = 1598 (val % hz) * tick; 1599 break; 1600 } 1601 1602 default: 1603 (void)m_free(m); 1604 return (ENOPROTOOPT); 1605 } 1606 *mp = m; 1607 return (0); 1608 } 1609 } 1610 1611 void 1612 sohasoutofband(struct socket *so) 1613 { 1614 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 1615 selwakeup(&so->so_rcv.sb_sel); 1616 } 1617 1618 static void 1619 filt_sordetach(struct knote *kn) 1620 { 1621 struct socket *so; 1622 1623 so = (struct socket *)kn->kn_fp->f_data; 1624 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 1625 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 1626 so->so_rcv.sb_flags &= ~SB_KNOTE; 1627 } 1628 1629 /*ARGSUSED*/ 1630 static int 1631 filt_soread(struct knote *kn, long hint) 1632 { 1633 struct socket *so; 1634 1635 so = (struct socket *)kn->kn_fp->f_data; 1636 kn->kn_data = so->so_rcv.sb_cc; 1637 if (so->so_state & SS_CANTRCVMORE) { 1638 kn->kn_flags |= EV_EOF; 1639 kn->kn_fflags = so->so_error; 1640 return (1); 1641 } 1642 if (so->so_error) /* temporary udp error */ 1643 return (1); 1644 if (kn->kn_sfflags & NOTE_LOWAT) 1645 return (kn->kn_data >= kn->kn_sdata); 1646 return (kn->kn_data >= so->so_rcv.sb_lowat); 1647 } 1648 1649 static void 1650 filt_sowdetach(struct knote *kn) 1651 { 1652 struct socket *so; 1653 1654 so = (struct socket *)kn->kn_fp->f_data; 1655 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 1656 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 1657 so->so_snd.sb_flags &= ~SB_KNOTE; 1658 } 1659 1660 /*ARGSUSED*/ 1661 static int 1662 filt_sowrite(struct knote *kn, long hint) 1663 { 1664 struct socket *so; 1665 1666 so = (struct socket *)kn->kn_fp->f_data; 1667 kn->kn_data = sbspace(&so->so_snd); 1668 if (so->so_state & SS_CANTSENDMORE) { 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 (((so->so_state & SS_ISCONNECTED) == 0) && 1676 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1677 return (0); 1678 if (kn->kn_sfflags & NOTE_LOWAT) 1679 return (kn->kn_data >= kn->kn_sdata); 1680 return (kn->kn_data >= so->so_snd.sb_lowat); 1681 } 1682 1683 /*ARGSUSED*/ 1684 static int 1685 filt_solisten(struct knote *kn, long hint) 1686 { 1687 struct socket *so; 1688 1689 so = (struct socket *)kn->kn_fp->f_data; 1690 1691 /* 1692 * Set kn_data to number of incoming connections, not 1693 * counting partial (incomplete) connections. 1694 */ 1695 kn->kn_data = so->so_qlen; 1696 return (kn->kn_data > 0); 1697 } 1698 1699 static const struct filterops solisten_filtops = 1700 { 1, NULL, filt_sordetach, filt_solisten }; 1701 static const struct filterops soread_filtops = 1702 { 1, NULL, filt_sordetach, filt_soread }; 1703 static const struct filterops sowrite_filtops = 1704 { 1, NULL, filt_sowdetach, filt_sowrite }; 1705 1706 int 1707 soo_kqfilter(struct file *fp, struct knote *kn) 1708 { 1709 struct socket *so; 1710 struct sockbuf *sb; 1711 1712 so = (struct socket *)kn->kn_fp->f_data; 1713 switch (kn->kn_filter) { 1714 case EVFILT_READ: 1715 if (so->so_options & SO_ACCEPTCONN) 1716 kn->kn_fop = &solisten_filtops; 1717 else 1718 kn->kn_fop = &soread_filtops; 1719 sb = &so->so_rcv; 1720 break; 1721 case EVFILT_WRITE: 1722 kn->kn_fop = &sowrite_filtops; 1723 sb = &so->so_snd; 1724 break; 1725 default: 1726 return (1); 1727 } 1728 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 1729 sb->sb_flags |= SB_KNOTE; 1730 return (0); 1731 } 1732 1733 #include <sys/sysctl.h> 1734 1735 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 1736 1737 /* 1738 * sysctl helper routine for kern.somaxkva. ensures that the given 1739 * value is not too small. 1740 * (XXX should we maybe make sure it's not too large as well?) 1741 */ 1742 static int 1743 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 1744 { 1745 int error, new_somaxkva; 1746 struct sysctlnode node; 1747 int s; 1748 1749 new_somaxkva = somaxkva; 1750 node = *rnode; 1751 node.sysctl_data = &new_somaxkva; 1752 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1753 if (error || newp == NULL) 1754 return (error); 1755 1756 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 1757 return (EINVAL); 1758 1759 s = splvm(); 1760 simple_lock(&so_pendfree_slock); 1761 somaxkva = new_somaxkva; 1762 wakeup(&socurkva); 1763 simple_unlock(&so_pendfree_slock); 1764 splx(s); 1765 1766 return (error); 1767 } 1768 1769 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup") 1770 { 1771 1772 sysctl_createv(clog, 0, NULL, NULL, 1773 CTLFLAG_PERMANENT, 1774 CTLTYPE_NODE, "kern", NULL, 1775 NULL, 0, NULL, 0, 1776 CTL_KERN, CTL_EOL); 1777 1778 sysctl_createv(clog, 0, NULL, NULL, 1779 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1780 CTLTYPE_INT, "somaxkva", NULL, 1781 sysctl_kern_somaxkva, 0, NULL, 0, 1782 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 1783 } 1784
Indexes created Sat Sep 20 22:09:52 GMT 2025