uipc_socket.c revision 1.101
1 /* $NetBSD: uipc_socket.c,v 1.101 2004/05/01 02:24:38 matt 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 1.101 2004/05/01 02:24:38 matt 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); 410 411 lva = sokvaalloc(len, so); 412 if (lva == 0) 413 return 0; 414 415 error = uvm_loan(&uio->uio_procp->p_vmspace->vm_map, sva, len, 416 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 417 if (error) { 418 sokvafree(lva, len); 419 return (0); 420 } 421 422 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 423 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 424 VM_PROT_READ); 425 pmap_update(pmap_kernel()); 426 427 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 428 429 MEXTADD(m, (caddr_t) lva, space, M_MBUF, soloanfree, so); 430 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 431 432 uio->uio_resid -= space; 433 /* uio_offset not updated, not set/used for write(2) */ 434 uio->uio_iov->iov_base = (caddr_t) uio->uio_iov->iov_base + space; 435 uio->uio_iov->iov_len -= space; 436 if (uio->uio_iov->iov_len == 0) { 437 uio->uio_iov++; 438 uio->uio_iovcnt--; 439 } 440 441 return (space); 442 } 443 444 /* 445 * Socket operation routines. 446 * These routines are called by the routines in 447 * sys_socket.c or from a system process, and 448 * implement the semantics of socket operations by 449 * switching out to the protocol specific routines. 450 */ 451 /*ARGSUSED*/ 452 int 453 socreate(int dom, struct socket **aso, int type, int proto) 454 { 455 struct proc *p; 456 const struct protosw *prp; 457 struct socket *so; 458 int error, s; 459 460 p = curproc; /* XXX */ 461 if (proto) 462 prp = pffindproto(dom, proto, type); 463 else 464 prp = pffindtype(dom, type); 465 if (prp == 0 || prp->pr_usrreq == 0) 466 return (EPROTONOSUPPORT); 467 if (prp->pr_type != type) 468 return (EPROTOTYPE); 469 s = splsoftnet(); 470 so = pool_get(&socket_pool, PR_WAITOK); 471 memset((caddr_t)so, 0, sizeof(*so)); 472 TAILQ_INIT(&so->so_q0); 473 TAILQ_INIT(&so->so_q); 474 so->so_type = type; 475 so->so_proto = prp; 476 so->so_send = sosend; 477 so->so_receive = soreceive; 478 #ifdef MBUFTRACE 479 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 480 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 481 so->so_mowner = &prp->pr_domain->dom_mowner; 482 #endif 483 if (p != 0) 484 so->so_uid = p->p_ucred->cr_uid; 485 else 486 so->so_uid = UID_MAX; 487 error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0, 488 (struct mbuf *)(long)proto, (struct mbuf *)0, p); 489 if (error) { 490 so->so_state |= SS_NOFDREF; 491 sofree(so); 492 splx(s); 493 return (error); 494 } 495 splx(s); 496 *aso = so; 497 return (0); 498 } 499 500 int 501 sobind(struct socket *so, struct mbuf *nam, struct proc *p) 502 { 503 int s, error; 504 505 s = splsoftnet(); 506 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0, 507 nam, (struct mbuf *)0, p); 508 splx(s); 509 return (error); 510 } 511 512 int 513 solisten(struct socket *so, int backlog) 514 { 515 int s, error; 516 517 s = splsoftnet(); 518 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0, 519 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 520 if (error) { 521 splx(s); 522 return (error); 523 } 524 if (TAILQ_EMPTY(&so->so_q)) 525 so->so_options |= SO_ACCEPTCONN; 526 if (backlog < 0) 527 backlog = 0; 528 so->so_qlimit = min(backlog, somaxconn); 529 splx(s); 530 return (0); 531 } 532 533 void 534 sofree(struct socket *so) 535 { 536 537 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 538 return; 539 if (so->so_head) { 540 /* 541 * We must not decommission a socket that's on the accept(2) 542 * queue. If we do, then accept(2) may hang after select(2) 543 * indicated that the listening socket was ready. 544 */ 545 if (!soqremque(so, 0)) 546 return; 547 } 548 if (so->so_rcv.sb_hiwat) 549 (void)chgsbsize(so->so_uid, &so->so_rcv.sb_hiwat, 0, 550 RLIM_INFINITY); 551 if (so->so_snd.sb_hiwat) 552 (void)chgsbsize(so->so_uid, &so->so_snd.sb_hiwat, 0, 553 RLIM_INFINITY); 554 sbrelease(&so->so_snd, so); 555 sorflush(so); 556 pool_put(&socket_pool, so); 557 } 558 559 /* 560 * Close a socket on last file table reference removal. 561 * Initiate disconnect if connected. 562 * Free socket when disconnect complete. 563 */ 564 int 565 soclose(struct socket *so) 566 { 567 struct socket *so2; 568 int s, error; 569 570 error = 0; 571 s = splsoftnet(); /* conservative */ 572 if (so->so_options & SO_ACCEPTCONN) { 573 while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 574 (void) soqremque(so2, 0); 575 (void) soabort(so2); 576 } 577 while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 578 (void) soqremque(so2, 1); 579 (void) soabort(so2); 580 } 581 } 582 if (so->so_pcb == 0) 583 goto discard; 584 if (so->so_state & SS_ISCONNECTED) { 585 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 586 error = sodisconnect(so); 587 if (error) 588 goto drop; 589 } 590 if (so->so_options & SO_LINGER) { 591 if ((so->so_state & SS_ISDISCONNECTING) && 592 (so->so_state & SS_NBIO)) 593 goto drop; 594 while (so->so_state & SS_ISCONNECTED) { 595 error = tsleep((caddr_t)&so->so_timeo, 596 PSOCK | PCATCH, netcls, 597 so->so_linger * hz); 598 if (error) 599 break; 600 } 601 } 602 } 603 drop: 604 if (so->so_pcb) { 605 int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 606 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 607 (struct proc *)0); 608 if (error == 0) 609 error = error2; 610 } 611 discard: 612 if (so->so_state & SS_NOFDREF) 613 panic("soclose: NOFDREF"); 614 so->so_state |= SS_NOFDREF; 615 sofree(so); 616 splx(s); 617 return (error); 618 } 619 620 /* 621 * Must be called at splsoftnet... 622 */ 623 int 624 soabort(struct socket *so) 625 { 626 627 return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0, 628 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 629 } 630 631 int 632 soaccept(struct socket *so, struct mbuf *nam) 633 { 634 int s, error; 635 636 error = 0; 637 s = splsoftnet(); 638 if ((so->so_state & SS_NOFDREF) == 0) 639 panic("soaccept: !NOFDREF"); 640 so->so_state &= ~SS_NOFDREF; 641 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 642 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 643 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 644 (struct mbuf *)0, nam, (struct mbuf *)0, (struct proc *)0); 645 else 646 error = ECONNABORTED; 647 648 splx(s); 649 return (error); 650 } 651 652 int 653 soconnect(struct socket *so, struct mbuf *nam) 654 { 655 struct proc *p; 656 int s, error; 657 658 p = curproc; /* XXX */ 659 if (so->so_options & SO_ACCEPTCONN) 660 return (EOPNOTSUPP); 661 s = splsoftnet(); 662 /* 663 * If protocol is connection-based, can only connect once. 664 * Otherwise, if connected, try to disconnect first. 665 * This allows user to disconnect by connecting to, e.g., 666 * a null address. 667 */ 668 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 669 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 670 (error = sodisconnect(so)))) 671 error = EISCONN; 672 else 673 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 674 (struct mbuf *)0, nam, (struct mbuf *)0, p); 675 splx(s); 676 return (error); 677 } 678 679 int 680 soconnect2(struct socket *so1, struct socket *so2) 681 { 682 int s, error; 683 684 s = splsoftnet(); 685 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 686 (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0, 687 (struct proc *)0); 688 splx(s); 689 return (error); 690 } 691 692 int 693 sodisconnect(struct socket *so) 694 { 695 int s, error; 696 697 s = splsoftnet(); 698 if ((so->so_state & SS_ISCONNECTED) == 0) { 699 error = ENOTCONN; 700 goto bad; 701 } 702 if (so->so_state & SS_ISDISCONNECTING) { 703 error = EALREADY; 704 goto bad; 705 } 706 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 707 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 708 (struct proc *)0); 709 bad: 710 splx(s); 711 sodopendfree(so); 712 return (error); 713 } 714 715 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 716 /* 717 * Send on a socket. 718 * If send must go all at once and message is larger than 719 * send buffering, then hard error. 720 * Lock against other senders. 721 * If must go all at once and not enough room now, then 722 * inform user that this would block and do nothing. 723 * Otherwise, if nonblocking, send as much as possible. 724 * The data to be sent is described by "uio" if nonzero, 725 * otherwise by the mbuf chain "top" (which must be null 726 * if uio is not). Data provided in mbuf chain must be small 727 * enough to send all at once. 728 * 729 * Returns nonzero on error, timeout or signal; callers 730 * must check for short counts if EINTR/ERESTART are returned. 731 * Data and control buffers are freed on return. 732 */ 733 int 734 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 735 struct mbuf *control, int flags) 736 { 737 struct proc *p; 738 struct mbuf **mp, *m; 739 long space, len, resid, clen, mlen; 740 int error, s, dontroute, atomic; 741 742 sodopendfree(so); 743 744 p = curproc; /* XXX */ 745 clen = 0; 746 atomic = sosendallatonce(so) || top; 747 if (uio) 748 resid = uio->uio_resid; 749 else 750 resid = top->m_pkthdr.len; 751 /* 752 * In theory resid should be unsigned. 753 * However, space must be signed, as it might be less than 0 754 * if we over-committed, and we must use a signed comparison 755 * of space and resid. On the other hand, a negative resid 756 * causes us to loop sending 0-length segments to the protocol. 757 */ 758 if (resid < 0) { 759 error = EINVAL; 760 goto out; 761 } 762 dontroute = 763 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 764 (so->so_proto->pr_flags & PR_ATOMIC); 765 p->p_stats->p_ru.ru_msgsnd++; 766 if (control) 767 clen = control->m_len; 768 #define snderr(errno) { error = errno; splx(s); goto release; } 769 770 restart: 771 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 772 goto out; 773 do { 774 s = splsoftnet(); 775 if (so->so_state & SS_CANTSENDMORE) 776 snderr(EPIPE); 777 if (so->so_error) { 778 error = so->so_error; 779 so->so_error = 0; 780 splx(s); 781 goto release; 782 } 783 if ((so->so_state & SS_ISCONNECTED) == 0) { 784 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 785 if ((so->so_state & SS_ISCONFIRMING) == 0 && 786 !(resid == 0 && clen != 0)) 787 snderr(ENOTCONN); 788 } else if (addr == 0) 789 snderr(EDESTADDRREQ); 790 } 791 space = sbspace(&so->so_snd); 792 if (flags & MSG_OOB) 793 space += 1024; 794 if ((atomic && resid > so->so_snd.sb_hiwat) || 795 clen > so->so_snd.sb_hiwat) 796 snderr(EMSGSIZE); 797 if (space < resid + clen && 798 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 799 if (so->so_state & SS_NBIO) 800 snderr(EWOULDBLOCK); 801 sbunlock(&so->so_snd); 802 error = sbwait(&so->so_snd); 803 splx(s); 804 if (error) 805 goto out; 806 goto restart; 807 } 808 splx(s); 809 mp = ⊤ 810 space -= clen; 811 do { 812 if (uio == NULL) { 813 /* 814 * Data is prepackaged in "top". 815 */ 816 resid = 0; 817 if (flags & MSG_EOR) 818 top->m_flags |= M_EOR; 819 } else do { 820 if (top == 0) { 821 m = m_gethdr(M_WAIT, MT_DATA); 822 mlen = MHLEN; 823 m->m_pkthdr.len = 0; 824 m->m_pkthdr.rcvif = (struct ifnet *)0; 825 } else { 826 m = m_get(M_WAIT, MT_DATA); 827 mlen = MLEN; 828 } 829 MCLAIM(m, so->so_snd.sb_mowner); 830 if (use_sosend_loan && 831 uio->uio_iov->iov_len >= SOCK_LOAN_THRESH && 832 space >= SOCK_LOAN_THRESH && 833 (len = sosend_loan(so, uio, m, 834 space)) != 0) { 835 SOSEND_COUNTER_INCR(&sosend_loan_big); 836 space -= len; 837 goto have_data; 838 } 839 if (resid >= MINCLSIZE && space >= MCLBYTES) { 840 SOSEND_COUNTER_INCR(&sosend_copy_big); 841 m_clget(m, M_WAIT); 842 if ((m->m_flags & M_EXT) == 0) 843 goto nopages; 844 mlen = MCLBYTES; 845 if (atomic && top == 0) { 846 len = lmin(MCLBYTES - max_hdr, 847 resid); 848 m->m_data += max_hdr; 849 } else 850 len = lmin(MCLBYTES, resid); 851 space -= len; 852 } else { 853 nopages: 854 SOSEND_COUNTER_INCR(&sosend_copy_small); 855 len = lmin(lmin(mlen, resid), space); 856 space -= len; 857 /* 858 * For datagram protocols, leave room 859 * for protocol headers in first mbuf. 860 */ 861 if (atomic && top == 0 && len < mlen) 862 MH_ALIGN(m, len); 863 } 864 error = uiomove(mtod(m, caddr_t), (int)len, 865 uio); 866 have_data: 867 resid = uio->uio_resid; 868 m->m_len = len; 869 *mp = m; 870 top->m_pkthdr.len += len; 871 if (error) 872 goto release; 873 mp = &m->m_next; 874 if (resid <= 0) { 875 if (flags & MSG_EOR) 876 top->m_flags |= M_EOR; 877 break; 878 } 879 } while (space > 0 && atomic); 880 881 s = splsoftnet(); 882 883 if (so->so_state & SS_CANTSENDMORE) 884 snderr(EPIPE); 885 886 if (dontroute) 887 so->so_options |= SO_DONTROUTE; 888 if (resid > 0) 889 so->so_state |= SS_MORETOCOME; 890 error = (*so->so_proto->pr_usrreq)(so, 891 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 892 top, addr, control, p); 893 if (dontroute) 894 so->so_options &= ~SO_DONTROUTE; 895 if (resid > 0) 896 so->so_state &= ~SS_MORETOCOME; 897 splx(s); 898 899 clen = 0; 900 control = 0; 901 top = 0; 902 mp = ⊤ 903 if (error) 904 goto release; 905 } while (resid && space > 0); 906 } while (resid); 907 908 release: 909 sbunlock(&so->so_snd); 910 out: 911 if (top) 912 m_freem(top); 913 if (control) 914 m_freem(control); 915 return (error); 916 } 917 918 /* 919 * Implement receive operations on a socket. 920 * We depend on the way that records are added to the sockbuf 921 * by sbappend*. In particular, each record (mbufs linked through m_next) 922 * must begin with an address if the protocol so specifies, 923 * followed by an optional mbuf or mbufs containing ancillary data, 924 * and then zero or more mbufs of data. 925 * In order to avoid blocking network interrupts for the entire time here, 926 * we splx() while doing the actual copy to user space. 927 * Although the sockbuf is locked, new data may still be appended, 928 * and thus we must maintain consistency of the sockbuf during that time. 929 * 930 * The caller may receive the data as a single mbuf chain by supplying 931 * an mbuf **mp0 for use in returning the chain. The uio is then used 932 * only for the count in uio_resid. 933 */ 934 int 935 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 936 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 937 { 938 struct mbuf *m, **mp; 939 int flags, len, error, s, offset, moff, type, orig_resid; 940 const struct protosw *pr; 941 struct mbuf *nextrecord; 942 int mbuf_removed = 0; 943 944 pr = so->so_proto; 945 mp = mp0; 946 type = 0; 947 orig_resid = uio->uio_resid; 948 if (paddr) 949 *paddr = 0; 950 if (controlp) 951 *controlp = 0; 952 if (flagsp) 953 flags = *flagsp &~ MSG_EOR; 954 else 955 flags = 0; 956 957 if ((flags & MSG_DONTWAIT) == 0) 958 sodopendfree(so); 959 960 if (flags & MSG_OOB) { 961 m = m_get(M_WAIT, MT_DATA); 962 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 963 (struct mbuf *)(long)(flags & MSG_PEEK), (struct mbuf *)0, 964 (struct proc *)0); 965 if (error) 966 goto bad; 967 do { 968 error = uiomove(mtod(m, caddr_t), 969 (int) min(uio->uio_resid, m->m_len), uio); 970 m = m_free(m); 971 } while (uio->uio_resid && error == 0 && m); 972 bad: 973 if (m) 974 m_freem(m); 975 return (error); 976 } 977 if (mp) 978 *mp = (struct mbuf *)0; 979 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 980 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 981 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 982 983 restart: 984 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) 985 return (error); 986 s = splsoftnet(); 987 988 m = so->so_rcv.sb_mb; 989 /* 990 * If we have less data than requested, block awaiting more 991 * (subject to any timeout) if: 992 * 1. the current count is less than the low water mark, 993 * 2. MSG_WAITALL is set, and it is possible to do the entire 994 * receive operation at once if we block (resid <= hiwat), or 995 * 3. MSG_DONTWAIT is not set. 996 * If MSG_WAITALL is set but resid is larger than the receive buffer, 997 * we have to do the receive in sections, and thus risk returning 998 * a short count if a timeout or signal occurs after we start. 999 */ 1000 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 1001 so->so_rcv.sb_cc < uio->uio_resid) && 1002 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1003 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1004 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1005 #ifdef DIAGNOSTIC 1006 if (m == 0 && so->so_rcv.sb_cc) 1007 panic("receive 1"); 1008 #endif 1009 if (so->so_error) { 1010 if (m) 1011 goto dontblock; 1012 error = so->so_error; 1013 if ((flags & MSG_PEEK) == 0) 1014 so->so_error = 0; 1015 goto release; 1016 } 1017 if (so->so_state & SS_CANTRCVMORE) { 1018 if (m) 1019 goto dontblock; 1020 else 1021 goto release; 1022 } 1023 for (; m; m = m->m_next) 1024 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1025 m = so->so_rcv.sb_mb; 1026 goto dontblock; 1027 } 1028 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1029 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1030 error = ENOTCONN; 1031 goto release; 1032 } 1033 if (uio->uio_resid == 0) 1034 goto release; 1035 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 1036 error = EWOULDBLOCK; 1037 goto release; 1038 } 1039 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1040 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1041 sbunlock(&so->so_rcv); 1042 error = sbwait(&so->so_rcv); 1043 splx(s); 1044 if (error) 1045 return (error); 1046 goto restart; 1047 } 1048 dontblock: 1049 /* 1050 * On entry here, m points to the first record of the socket buffer. 1051 * While we process the initial mbufs containing address and control 1052 * info, we save a copy of m->m_nextpkt into nextrecord. 1053 */ 1054 #ifdef notyet /* XXXX */ 1055 if (uio->uio_procp) 1056 uio->uio_procp->p_stats->p_ru.ru_msgrcv++; 1057 #endif 1058 KASSERT(m == so->so_rcv.sb_mb); 1059 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1060 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1061 nextrecord = m->m_nextpkt; 1062 if (pr->pr_flags & PR_ADDR) { 1063 #ifdef DIAGNOSTIC 1064 if (m->m_type != MT_SONAME) 1065 panic("receive 1a"); 1066 #endif 1067 orig_resid = 0; 1068 if (flags & MSG_PEEK) { 1069 if (paddr) 1070 *paddr = m_copy(m, 0, m->m_len); 1071 m = m->m_next; 1072 } else { 1073 sbfree(&so->so_rcv, m); 1074 mbuf_removed = 1; 1075 if (paddr) { 1076 *paddr = m; 1077 so->so_rcv.sb_mb = m->m_next; 1078 m->m_next = 0; 1079 m = so->so_rcv.sb_mb; 1080 } else { 1081 MFREE(m, so->so_rcv.sb_mb); 1082 m = so->so_rcv.sb_mb; 1083 } 1084 } 1085 } 1086 while (m && m->m_type == MT_CONTROL && error == 0) { 1087 if (flags & MSG_PEEK) { 1088 if (controlp) 1089 *controlp = m_copy(m, 0, m->m_len); 1090 m = m->m_next; 1091 } else { 1092 sbfree(&so->so_rcv, m); 1093 mbuf_removed = 1; 1094 if (controlp) { 1095 if (pr->pr_domain->dom_externalize && 1096 mtod(m, struct cmsghdr *)->cmsg_type == 1097 SCM_RIGHTS) 1098 error = (*pr->pr_domain->dom_externalize)(m); 1099 *controlp = m; 1100 so->so_rcv.sb_mb = m->m_next; 1101 m->m_next = 0; 1102 m = so->so_rcv.sb_mb; 1103 } else { 1104 MFREE(m, so->so_rcv.sb_mb); 1105 m = so->so_rcv.sb_mb; 1106 } 1107 } 1108 if (controlp) { 1109 orig_resid = 0; 1110 controlp = &(*controlp)->m_next; 1111 } 1112 } 1113 1114 /* 1115 * If m is non-NULL, we have some data to read. From now on, 1116 * make sure to keep sb_lastrecord consistent when working on 1117 * the last packet on the chain (nextrecord == NULL) and we 1118 * change m->m_nextpkt. 1119 */ 1120 if (m) { 1121 if ((flags & MSG_PEEK) == 0) { 1122 m->m_nextpkt = nextrecord; 1123 /* 1124 * If nextrecord == NULL (this is a single chain), 1125 * then sb_lastrecord may not be valid here if m 1126 * was changed earlier. 1127 */ 1128 if (nextrecord == NULL) { 1129 KASSERT(so->so_rcv.sb_mb == m); 1130 so->so_rcv.sb_lastrecord = m; 1131 } 1132 } 1133 type = m->m_type; 1134 if (type == MT_OOBDATA) 1135 flags |= MSG_OOB; 1136 } else { 1137 if ((flags & MSG_PEEK) == 0) { 1138 KASSERT(so->so_rcv.sb_mb == m); 1139 so->so_rcv.sb_mb = nextrecord; 1140 SB_EMPTY_FIXUP(&so->so_rcv); 1141 } 1142 } 1143 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1144 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1145 1146 moff = 0; 1147 offset = 0; 1148 while (m && uio->uio_resid > 0 && error == 0) { 1149 if (m->m_type == MT_OOBDATA) { 1150 if (type != MT_OOBDATA) 1151 break; 1152 } else if (type == MT_OOBDATA) 1153 break; 1154 #ifdef DIAGNOSTIC 1155 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1156 panic("receive 3"); 1157 #endif 1158 so->so_state &= ~SS_RCVATMARK; 1159 len = uio->uio_resid; 1160 if (so->so_oobmark && len > so->so_oobmark - offset) 1161 len = so->so_oobmark - offset; 1162 if (len > m->m_len - moff) 1163 len = m->m_len - moff; 1164 /* 1165 * If mp is set, just pass back the mbufs. 1166 * Otherwise copy them out via the uio, then free. 1167 * Sockbuf must be consistent here (points to current mbuf, 1168 * it points to next record) when we drop priority; 1169 * we must note any additions to the sockbuf when we 1170 * block interrupts again. 1171 */ 1172 if (mp == 0) { 1173 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1174 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1175 splx(s); 1176 error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio); 1177 s = splsoftnet(); 1178 if (error) { 1179 /* 1180 * If any part of the record has been removed 1181 * (such as the MT_SONAME mbuf, which will 1182 * happen when PR_ADDR, and thus also 1183 * PR_ATOMIC, is set), then drop the entire 1184 * record to maintain the atomicity of the 1185 * receive operation. 1186 * 1187 * This avoids a later panic("receive 1a") 1188 * when compiled with DIAGNOSTIC. 1189 */ 1190 if (m && mbuf_removed 1191 && (pr->pr_flags & PR_ATOMIC)) 1192 (void) sbdroprecord(&so->so_rcv); 1193 1194 goto release; 1195 } 1196 } else 1197 uio->uio_resid -= len; 1198 if (len == m->m_len - moff) { 1199 if (m->m_flags & M_EOR) 1200 flags |= MSG_EOR; 1201 if (flags & MSG_PEEK) { 1202 m = m->m_next; 1203 moff = 0; 1204 } else { 1205 nextrecord = m->m_nextpkt; 1206 sbfree(&so->so_rcv, m); 1207 if (mp) { 1208 *mp = m; 1209 mp = &m->m_next; 1210 so->so_rcv.sb_mb = m = m->m_next; 1211 *mp = (struct mbuf *)0; 1212 } else { 1213 MFREE(m, so->so_rcv.sb_mb); 1214 m = so->so_rcv.sb_mb; 1215 } 1216 /* 1217 * If m != NULL, we also know that 1218 * so->so_rcv.sb_mb != NULL. 1219 */ 1220 KASSERT(so->so_rcv.sb_mb == m); 1221 if (m) { 1222 m->m_nextpkt = nextrecord; 1223 if (nextrecord == NULL) 1224 so->so_rcv.sb_lastrecord = m; 1225 } else { 1226 so->so_rcv.sb_mb = nextrecord; 1227 SB_EMPTY_FIXUP(&so->so_rcv); 1228 } 1229 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1230 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1231 } 1232 } else { 1233 if (flags & MSG_PEEK) 1234 moff += len; 1235 else { 1236 if (mp) 1237 *mp = m_copym(m, 0, len, M_WAIT); 1238 m->m_data += len; 1239 m->m_len -= len; 1240 so->so_rcv.sb_cc -= len; 1241 } 1242 } 1243 if (so->so_oobmark) { 1244 if ((flags & MSG_PEEK) == 0) { 1245 so->so_oobmark -= len; 1246 if (so->so_oobmark == 0) { 1247 so->so_state |= SS_RCVATMARK; 1248 break; 1249 } 1250 } else { 1251 offset += len; 1252 if (offset == so->so_oobmark) 1253 break; 1254 } 1255 } 1256 if (flags & MSG_EOR) 1257 break; 1258 /* 1259 * If the MSG_WAITALL flag is set (for non-atomic socket), 1260 * we must not quit until "uio->uio_resid == 0" or an error 1261 * termination. If a signal/timeout occurs, return 1262 * with a short count but without error. 1263 * Keep sockbuf locked against other readers. 1264 */ 1265 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1266 !sosendallatonce(so) && !nextrecord) { 1267 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1268 break; 1269 /* 1270 * If we are peeking and the socket receive buffer is 1271 * full, stop since we can't get more data to peek at. 1272 */ 1273 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1274 break; 1275 /* 1276 * If we've drained the socket buffer, tell the 1277 * protocol in case it needs to do something to 1278 * get it filled again. 1279 */ 1280 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1281 (*pr->pr_usrreq)(so, PRU_RCVD, 1282 (struct mbuf *)0, 1283 (struct mbuf *)(long)flags, 1284 (struct mbuf *)0, 1285 (struct proc *)0); 1286 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1287 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1288 error = sbwait(&so->so_rcv); 1289 if (error) { 1290 sbunlock(&so->so_rcv); 1291 splx(s); 1292 return (0); 1293 } 1294 if ((m = so->so_rcv.sb_mb) != NULL) 1295 nextrecord = m->m_nextpkt; 1296 } 1297 } 1298 1299 if (m && pr->pr_flags & PR_ATOMIC) { 1300 flags |= MSG_TRUNC; 1301 if ((flags & MSG_PEEK) == 0) 1302 (void) sbdroprecord(&so->so_rcv); 1303 } 1304 if ((flags & MSG_PEEK) == 0) { 1305 if (m == 0) { 1306 /* 1307 * First part is an inline SB_EMPTY_FIXUP(). Second 1308 * part makes sure sb_lastrecord is up-to-date if 1309 * there is still data in the socket buffer. 1310 */ 1311 so->so_rcv.sb_mb = nextrecord; 1312 if (so->so_rcv.sb_mb == NULL) { 1313 so->so_rcv.sb_mbtail = NULL; 1314 so->so_rcv.sb_lastrecord = NULL; 1315 } else if (nextrecord->m_nextpkt == NULL) 1316 so->so_rcv.sb_lastrecord = nextrecord; 1317 } 1318 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1319 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1320 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1321 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1322 (struct mbuf *)(long)flags, (struct mbuf *)0, 1323 (struct proc *)0); 1324 } 1325 if (orig_resid == uio->uio_resid && orig_resid && 1326 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1327 sbunlock(&so->so_rcv); 1328 splx(s); 1329 goto restart; 1330 } 1331 1332 if (flagsp) 1333 *flagsp |= flags; 1334 release: 1335 sbunlock(&so->so_rcv); 1336 splx(s); 1337 return (error); 1338 } 1339 1340 int 1341 soshutdown(struct socket *so, int how) 1342 { 1343 const struct protosw *pr; 1344 1345 pr = so->so_proto; 1346 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1347 return (EINVAL); 1348 1349 if (how == SHUT_RD || how == SHUT_RDWR) 1350 sorflush(so); 1351 if (how == SHUT_WR || how == SHUT_RDWR) 1352 return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0, 1353 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1354 return (0); 1355 } 1356 1357 void 1358 sorflush(struct socket *so) 1359 { 1360 struct sockbuf *sb, asb; 1361 const struct protosw *pr; 1362 int s; 1363 1364 sb = &so->so_rcv; 1365 pr = so->so_proto; 1366 sb->sb_flags |= SB_NOINTR; 1367 (void) sblock(sb, M_WAITOK); 1368 s = splnet(); 1369 socantrcvmore(so); 1370 sbunlock(sb); 1371 asb = *sb; 1372 /* 1373 * Clear most of the sockbuf structure, but leave some of the 1374 * fields valid. 1375 */ 1376 memset(&sb->sb_startzero, 0, 1377 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1378 splx(s); 1379 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1380 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1381 sbrelease(&asb, so); 1382 } 1383 1384 int 1385 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0) 1386 { 1387 int error; 1388 struct mbuf *m; 1389 1390 error = 0; 1391 m = m0; 1392 if (level != SOL_SOCKET) { 1393 if (so->so_proto && so->so_proto->pr_ctloutput) 1394 return ((*so->so_proto->pr_ctloutput) 1395 (PRCO_SETOPT, so, level, optname, &m0)); 1396 error = ENOPROTOOPT; 1397 } else { 1398 switch (optname) { 1399 1400 case SO_LINGER: 1401 if (m == NULL || m->m_len != sizeof(struct linger)) { 1402 error = EINVAL; 1403 goto bad; 1404 } 1405 so->so_linger = mtod(m, struct linger *)->l_linger; 1406 /* fall thru... */ 1407 1408 case SO_DEBUG: 1409 case SO_KEEPALIVE: 1410 case SO_DONTROUTE: 1411 case SO_USELOOPBACK: 1412 case SO_BROADCAST: 1413 case SO_REUSEADDR: 1414 case SO_REUSEPORT: 1415 case SO_OOBINLINE: 1416 case SO_TIMESTAMP: 1417 if (m == NULL || m->m_len < sizeof(int)) { 1418 error = EINVAL; 1419 goto bad; 1420 } 1421 if (*mtod(m, int *)) 1422 so->so_options |= optname; 1423 else 1424 so->so_options &= ~optname; 1425 break; 1426 1427 case SO_SNDBUF: 1428 case SO_RCVBUF: 1429 case SO_SNDLOWAT: 1430 case SO_RCVLOWAT: 1431 { 1432 int optval; 1433 1434 if (m == NULL || m->m_len < sizeof(int)) { 1435 error = EINVAL; 1436 goto bad; 1437 } 1438 1439 /* 1440 * Values < 1 make no sense for any of these 1441 * options, so disallow them. 1442 */ 1443 optval = *mtod(m, int *); 1444 if (optval < 1) { 1445 error = EINVAL; 1446 goto bad; 1447 } 1448 1449 switch (optname) { 1450 1451 case SO_SNDBUF: 1452 case SO_RCVBUF: 1453 if (sbreserve(optname == SO_SNDBUF ? 1454 &so->so_snd : &so->so_rcv, 1455 (u_long) optval, so) == 0) { 1456 error = ENOBUFS; 1457 goto bad; 1458 } 1459 break; 1460 1461 /* 1462 * Make sure the low-water is never greater than 1463 * the high-water. 1464 */ 1465 case SO_SNDLOWAT: 1466 so->so_snd.sb_lowat = 1467 (optval > so->so_snd.sb_hiwat) ? 1468 so->so_snd.sb_hiwat : optval; 1469 break; 1470 case SO_RCVLOWAT: 1471 so->so_rcv.sb_lowat = 1472 (optval > so->so_rcv.sb_hiwat) ? 1473 so->so_rcv.sb_hiwat : optval; 1474 break; 1475 } 1476 break; 1477 } 1478 1479 case SO_SNDTIMEO: 1480 case SO_RCVTIMEO: 1481 { 1482 struct timeval *tv; 1483 short val; 1484 1485 if (m == NULL || m->m_len < sizeof(*tv)) { 1486 error = EINVAL; 1487 goto bad; 1488 } 1489 tv = mtod(m, struct timeval *); 1490 if (tv->tv_sec > (SHRT_MAX - tv->tv_usec / tick) / hz) { 1491 error = EDOM; 1492 goto bad; 1493 } 1494 val = tv->tv_sec * hz + tv->tv_usec / tick; 1495 if (val == 0 && tv->tv_usec != 0) 1496 val = 1; 1497 1498 switch (optname) { 1499 1500 case SO_SNDTIMEO: 1501 so->so_snd.sb_timeo = val; 1502 break; 1503 case SO_RCVTIMEO: 1504 so->so_rcv.sb_timeo = val; 1505 break; 1506 } 1507 break; 1508 } 1509 1510 default: 1511 error = ENOPROTOOPT; 1512 break; 1513 } 1514 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1515 (void) ((*so->so_proto->pr_ctloutput) 1516 (PRCO_SETOPT, so, level, optname, &m0)); 1517 m = NULL; /* freed by protocol */ 1518 } 1519 } 1520 bad: 1521 if (m) 1522 (void) m_free(m); 1523 return (error); 1524 } 1525 1526 int 1527 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp) 1528 { 1529 struct mbuf *m; 1530 1531 if (level != SOL_SOCKET) { 1532 if (so->so_proto && so->so_proto->pr_ctloutput) { 1533 return ((*so->so_proto->pr_ctloutput) 1534 (PRCO_GETOPT, so, level, optname, mp)); 1535 } else 1536 return (ENOPROTOOPT); 1537 } else { 1538 m = m_get(M_WAIT, MT_SOOPTS); 1539 m->m_len = sizeof(int); 1540 1541 switch (optname) { 1542 1543 case SO_LINGER: 1544 m->m_len = sizeof(struct linger); 1545 mtod(m, struct linger *)->l_onoff = 1546 so->so_options & SO_LINGER; 1547 mtod(m, struct linger *)->l_linger = so->so_linger; 1548 break; 1549 1550 case SO_USELOOPBACK: 1551 case SO_DONTROUTE: 1552 case SO_DEBUG: 1553 case SO_KEEPALIVE: 1554 case SO_REUSEADDR: 1555 case SO_REUSEPORT: 1556 case SO_BROADCAST: 1557 case SO_OOBINLINE: 1558 case SO_TIMESTAMP: 1559 *mtod(m, int *) = so->so_options & optname; 1560 break; 1561 1562 case SO_TYPE: 1563 *mtod(m, int *) = so->so_type; 1564 break; 1565 1566 case SO_ERROR: 1567 *mtod(m, int *) = so->so_error; 1568 so->so_error = 0; 1569 break; 1570 1571 case SO_SNDBUF: 1572 *mtod(m, int *) = so->so_snd.sb_hiwat; 1573 break; 1574 1575 case SO_RCVBUF: 1576 *mtod(m, int *) = so->so_rcv.sb_hiwat; 1577 break; 1578 1579 case SO_SNDLOWAT: 1580 *mtod(m, int *) = so->so_snd.sb_lowat; 1581 break; 1582 1583 case SO_RCVLOWAT: 1584 *mtod(m, int *) = so->so_rcv.sb_lowat; 1585 break; 1586 1587 case SO_SNDTIMEO: 1588 case SO_RCVTIMEO: 1589 { 1590 int val = (optname == SO_SNDTIMEO ? 1591 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1592 1593 m->m_len = sizeof(struct timeval); 1594 mtod(m, struct timeval *)->tv_sec = val / hz; 1595 mtod(m, struct timeval *)->tv_usec = 1596 (val % hz) * tick; 1597 break; 1598 } 1599 1600 default: 1601 (void)m_free(m); 1602 return (ENOPROTOOPT); 1603 } 1604 *mp = m; 1605 return (0); 1606 } 1607 } 1608 1609 void 1610 sohasoutofband(struct socket *so) 1611 { 1612 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 1613 selwakeup(&so->so_rcv.sb_sel); 1614 } 1615 1616 static void 1617 filt_sordetach(struct knote *kn) 1618 { 1619 struct socket *so; 1620 1621 so = (struct socket *)kn->kn_fp->f_data; 1622 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 1623 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 1624 so->so_rcv.sb_flags &= ~SB_KNOTE; 1625 } 1626 1627 /*ARGSUSED*/ 1628 static int 1629 filt_soread(struct knote *kn, long hint) 1630 { 1631 struct socket *so; 1632 1633 so = (struct socket *)kn->kn_fp->f_data; 1634 kn->kn_data = so->so_rcv.sb_cc; 1635 if (so->so_state & SS_CANTRCVMORE) { 1636 kn->kn_flags |= EV_EOF; 1637 kn->kn_fflags = so->so_error; 1638 return (1); 1639 } 1640 if (so->so_error) /* temporary udp error */ 1641 return (1); 1642 if (kn->kn_sfflags & NOTE_LOWAT) 1643 return (kn->kn_data >= kn->kn_sdata); 1644 return (kn->kn_data >= so->so_rcv.sb_lowat); 1645 } 1646 1647 static void 1648 filt_sowdetach(struct knote *kn) 1649 { 1650 struct socket *so; 1651 1652 so = (struct socket *)kn->kn_fp->f_data; 1653 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 1654 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 1655 so->so_snd.sb_flags &= ~SB_KNOTE; 1656 } 1657 1658 /*ARGSUSED*/ 1659 static int 1660 filt_sowrite(struct knote *kn, long hint) 1661 { 1662 struct socket *so; 1663 1664 so = (struct socket *)kn->kn_fp->f_data; 1665 kn->kn_data = sbspace(&so->so_snd); 1666 if (so->so_state & SS_CANTSENDMORE) { 1667 kn->kn_flags |= EV_EOF; 1668 kn->kn_fflags = so->so_error; 1669 return (1); 1670 } 1671 if (so->so_error) /* temporary udp error */ 1672 return (1); 1673 if (((so->so_state & SS_ISCONNECTED) == 0) && 1674 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1675 return (0); 1676 if (kn->kn_sfflags & NOTE_LOWAT) 1677 return (kn->kn_data >= kn->kn_sdata); 1678 return (kn->kn_data >= so->so_snd.sb_lowat); 1679 } 1680 1681 /*ARGSUSED*/ 1682 static int 1683 filt_solisten(struct knote *kn, long hint) 1684 { 1685 struct socket *so; 1686 1687 so = (struct socket *)kn->kn_fp->f_data; 1688 1689 /* 1690 * Set kn_data to number of incoming connections, not 1691 * counting partial (incomplete) connections. 1692 */ 1693 kn->kn_data = so->so_qlen; 1694 return (kn->kn_data > 0); 1695 } 1696 1697 static const struct filterops solisten_filtops = 1698 { 1, NULL, filt_sordetach, filt_solisten }; 1699 static const struct filterops soread_filtops = 1700 { 1, NULL, filt_sordetach, filt_soread }; 1701 static const struct filterops sowrite_filtops = 1702 { 1, NULL, filt_sowdetach, filt_sowrite }; 1703 1704 int 1705 soo_kqfilter(struct file *fp, struct knote *kn) 1706 { 1707 struct socket *so; 1708 struct sockbuf *sb; 1709 1710 so = (struct socket *)kn->kn_fp->f_data; 1711 switch (kn->kn_filter) { 1712 case EVFILT_READ: 1713 if (so->so_options & SO_ACCEPTCONN) 1714 kn->kn_fop = &solisten_filtops; 1715 else 1716 kn->kn_fop = &soread_filtops; 1717 sb = &so->so_rcv; 1718 break; 1719 case EVFILT_WRITE: 1720 kn->kn_fop = &sowrite_filtops; 1721 sb = &so->so_snd; 1722 break; 1723 default: 1724 return (1); 1725 } 1726 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 1727 sb->sb_flags |= SB_KNOTE; 1728 return (0); 1729 } 1730 1731 #include <sys/sysctl.h> 1732 1733 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 1734 1735 /* 1736 * sysctl helper routine for kern.somaxkva. ensures that the given 1737 * value is not too small. 1738 * (XXX should we maybe make sure it's not too large as well?) 1739 */ 1740 static int 1741 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 1742 { 1743 int error, new_somaxkva; 1744 struct sysctlnode node; 1745 int s; 1746 1747 new_somaxkva = somaxkva; 1748 node = *rnode; 1749 node.sysctl_data = &new_somaxkva; 1750 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1751 if (error || newp == NULL) 1752 return (error); 1753 1754 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 1755 return (EINVAL); 1756 1757 s = splvm(); 1758 simple_lock(&so_pendfree_slock); 1759 somaxkva = new_somaxkva; 1760 wakeup(&socurkva); 1761 simple_unlock(&so_pendfree_slock); 1762 splx(s); 1763 1764 return (error); 1765 } 1766 1767 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup") 1768 { 1769 1770 sysctl_createv(clog, 0, NULL, NULL, 1771 CTLFLAG_PERMANENT, 1772 CTLTYPE_NODE, "kern", NULL, 1773 NULL, 0, NULL, 0, 1774 CTL_KERN, CTL_EOL); 1775 1776 sysctl_createv(clog, 0, NULL, NULL, 1777 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1778 CTLTYPE_INT, "somaxkva", NULL, 1779 sysctl_kern_somaxkva, 0, NULL, 0, 1780 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 1781 } 1782
Indexes created Sat Sep 20 22:09:52 GMT 2025