uipc_socket.c revision 1.113
1 /* $NetBSD: uipc_socket.c,v 1.113 2005/12/08 03:13:18 thorpej 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 1.113 2005/12/08 03:13:18 thorpej 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 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 109 NULL, "sosend", "loan big"); 110 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 111 NULL, "sosend", "copy big"); 112 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 113 NULL, "sosend", "copy small"); 114 static 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", 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 static struct simplelock so_pendfree_slock = SIMPLELOCK_INITIALIZER; 146 static struct mbuf *so_pendfree; 147 148 #ifndef SOMAXKVA 149 #define SOMAXKVA (16 * 1024 * 1024) 150 #endif 151 int somaxkva = SOMAXKVA; 152 static int socurkva; 153 static 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 *); 160 161 static vsize_t 162 sokvareserve(struct socket *so, vsize_t len) 163 { 164 int s; 165 int error; 166 167 s = splvm(); 168 simple_lock(&so_pendfree_slock); 169 while (socurkva + len > somaxkva) { 170 size_t freed; 171 172 /* 173 * try to do pendfree. 174 */ 175 176 freed = sodopendfreel(so); 177 178 /* 179 * if some kva was freed, try again. 180 */ 181 182 if (freed) 183 continue; 184 185 SOSEND_COUNTER_INCR(&sosend_kvalimit); 186 sokvawaiters++; 187 error = ltsleep(&socurkva, PVM | PCATCH, "sokva", 0, 188 &so_pendfree_slock); 189 sokvawaiters--; 190 if (error) { 191 len = 0; 192 break; 193 } 194 } 195 socurkva += len; 196 simple_unlock(&so_pendfree_slock); 197 splx(s); 198 return len; 199 } 200 201 static void 202 sokvaunreserve(vsize_t len) 203 { 204 int s; 205 206 s = splvm(); 207 simple_lock(&so_pendfree_slock); 208 socurkva -= len; 209 if (sokvawaiters) 210 wakeup(&socurkva); 211 simple_unlock(&so_pendfree_slock); 212 splx(s); 213 } 214 215 /* 216 * sokvaalloc: allocate kva for loan. 217 */ 218 219 vaddr_t 220 sokvaalloc(vsize_t len, struct socket *so) 221 { 222 vaddr_t lva; 223 224 /* 225 * reserve kva. 226 */ 227 228 if (sokvareserve(so, len) == 0) 229 return 0; 230 231 /* 232 * allocate kva. 233 */ 234 235 lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA); 236 if (lva == 0) { 237 sokvaunreserve(len); 238 return (0); 239 } 240 241 return lva; 242 } 243 244 /* 245 * sokvafree: free kva for loan. 246 */ 247 248 void 249 sokvafree(vaddr_t sva, vsize_t len) 250 { 251 252 /* 253 * free kva. 254 */ 255 256 uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY); 257 258 /* 259 * unreserve kva. 260 */ 261 262 sokvaunreserve(len); 263 } 264 265 static void 266 sodoloanfree(struct vm_page **pgs, caddr_t buf, size_t size) 267 { 268 vaddr_t va, sva, eva; 269 vsize_t len; 270 paddr_t pa; 271 int i, npgs; 272 273 eva = round_page((vaddr_t) buf + size); 274 sva = trunc_page((vaddr_t) buf); 275 len = eva - sva; 276 npgs = len >> PAGE_SHIFT; 277 278 if (__predict_false(pgs == NULL)) { 279 pgs = alloca(npgs * sizeof(*pgs)); 280 281 for (i = 0, va = sva; va < eva; i++, va += PAGE_SIZE) { 282 if (pmap_extract(pmap_kernel(), va, &pa) == FALSE) 283 panic("sodoloanfree: va 0x%lx not mapped", va); 284 pgs[i] = PHYS_TO_VM_PAGE(pa); 285 } 286 } 287 288 pmap_kremove(sva, len); 289 pmap_update(pmap_kernel()); 290 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 291 sokvafree(sva, len); 292 } 293 294 static size_t 295 sodopendfree(struct socket *so) 296 { 297 int s; 298 size_t rv; 299 300 s = splvm(); 301 simple_lock(&so_pendfree_slock); 302 rv = sodopendfreel(so); 303 simple_unlock(&so_pendfree_slock); 304 splx(s); 305 306 return rv; 307 } 308 309 /* 310 * sodopendfreel: free mbufs on "pendfree" list. 311 * unlock and relock so_pendfree_slock when freeing mbufs. 312 * 313 * => called with so_pendfree_slock held. 314 * => called at splvm. 315 */ 316 317 static size_t 318 sodopendfreel(struct socket *so) 319 { 320 size_t rv = 0; 321 322 LOCK_ASSERT(simple_lock_held(&so_pendfree_slock)); 323 324 for (;;) { 325 struct mbuf *m; 326 struct mbuf *next; 327 328 m = so_pendfree; 329 if (m == NULL) 330 break; 331 so_pendfree = NULL; 332 simple_unlock(&so_pendfree_slock); 333 /* XXX splx */ 334 335 for (; m != NULL; m = next) { 336 next = m->m_next; 337 338 rv += m->m_ext.ext_size; 339 sodoloanfree((m->m_flags & M_EXT_PAGES) ? 340 m->m_ext.ext_pgs : NULL, m->m_ext.ext_buf, 341 m->m_ext.ext_size); 342 pool_cache_put(&mbpool_cache, m); 343 } 344 345 /* XXX splvm */ 346 simple_lock(&so_pendfree_slock); 347 } 348 349 return (rv); 350 } 351 352 void 353 soloanfree(struct mbuf *m, caddr_t buf, size_t size, void *arg) 354 { 355 int s; 356 357 if (m == NULL) { 358 359 /* 360 * called from MEXTREMOVE. 361 */ 362 363 sodoloanfree(NULL, buf, size); 364 return; 365 } 366 367 /* 368 * postpone freeing mbuf. 369 * 370 * we can't do it in interrupt context 371 * because we need to put kva back to kernel_map. 372 */ 373 374 s = splvm(); 375 simple_lock(&so_pendfree_slock); 376 m->m_next = so_pendfree; 377 so_pendfree = m; 378 if (sokvawaiters) 379 wakeup(&socurkva); 380 simple_unlock(&so_pendfree_slock); 381 splx(s); 382 } 383 384 static long 385 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 386 { 387 struct iovec *iov = uio->uio_iov; 388 vaddr_t sva, eva; 389 vsize_t len; 390 vaddr_t lva, va; 391 int npgs, i, error; 392 393 if (uio->uio_segflg != UIO_USERSPACE) 394 return (0); 395 396 if (iov->iov_len < (size_t) space) 397 space = iov->iov_len; 398 if (space > SOCK_LOAN_CHUNK) 399 space = SOCK_LOAN_CHUNK; 400 401 eva = round_page((vaddr_t) iov->iov_base + space); 402 sva = trunc_page((vaddr_t) iov->iov_base); 403 len = eva - sva; 404 npgs = len >> PAGE_SHIFT; 405 406 /* XXX KDASSERT */ 407 KASSERT(npgs <= M_EXT_MAXPAGES); 408 KASSERT(uio->uio_procp != NULL); 409 410 lva = sokvaalloc(len, so); 411 if (lva == 0) 412 return 0; 413 414 error = uvm_loan(&uio->uio_procp->p_vmspace->vm_map, sva, len, 415 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 416 if (error) { 417 sokvafree(lva, len); 418 return (0); 419 } 420 421 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 422 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 423 VM_PROT_READ); 424 pmap_update(pmap_kernel()); 425 426 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 427 428 MEXTADD(m, (caddr_t) lva, space, M_MBUF, soloanfree, so); 429 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 430 431 uio->uio_resid -= space; 432 /* uio_offset not updated, not set/used for write(2) */ 433 uio->uio_iov->iov_base = (caddr_t) uio->uio_iov->iov_base + space; 434 uio->uio_iov->iov_len -= space; 435 if (uio->uio_iov->iov_len == 0) { 436 uio->uio_iov++; 437 uio->uio_iovcnt--; 438 } 439 440 return (space); 441 } 442 443 /* 444 * Socket operation routines. 445 * These routines are called by the routines in 446 * sys_socket.c or from a system process, and 447 * implement the semantics of socket operations by 448 * switching out to the protocol specific routines. 449 */ 450 /*ARGSUSED*/ 451 int 452 socreate(int dom, struct socket **aso, int type, int proto, struct proc *p) 453 { 454 const struct protosw *prp; 455 struct socket *so; 456 int error, s; 457 458 if (proto) 459 prp = pffindproto(dom, proto, type); 460 else 461 prp = pffindtype(dom, type); 462 if (prp == 0 || prp->pr_usrreq == 0) 463 return (EPROTONOSUPPORT); 464 if (prp->pr_type != type) 465 return (EPROTOTYPE); 466 s = splsoftnet(); 467 so = pool_get(&socket_pool, PR_WAITOK); 468 memset((caddr_t)so, 0, sizeof(*so)); 469 TAILQ_INIT(&so->so_q0); 470 TAILQ_INIT(&so->so_q); 471 so->so_type = type; 472 so->so_proto = prp; 473 so->so_send = sosend; 474 so->so_receive = soreceive; 475 #ifdef MBUFTRACE 476 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 477 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 478 so->so_mowner = &prp->pr_domain->dom_mowner; 479 #endif 480 if (p != 0) 481 so->so_uidinfo = uid_find(p->p_ucred->cr_uid); 482 else 483 so->so_uidinfo = uid_find(0); 484 error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0, 485 (struct mbuf *)(long)proto, (struct mbuf *)0, p); 486 if (error) { 487 so->so_state |= SS_NOFDREF; 488 sofree(so); 489 splx(s); 490 return (error); 491 } 492 splx(s); 493 *aso = so; 494 return (0); 495 } 496 497 int 498 sobind(struct socket *so, struct mbuf *nam, struct proc *p) 499 { 500 int s, error; 501 502 s = splsoftnet(); 503 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0, 504 nam, (struct mbuf *)0, p); 505 splx(s); 506 return (error); 507 } 508 509 int 510 solisten(struct socket *so, int backlog) 511 { 512 int s, error; 513 514 s = splsoftnet(); 515 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0, 516 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 517 if (error) { 518 splx(s); 519 return (error); 520 } 521 if (TAILQ_EMPTY(&so->so_q)) 522 so->so_options |= SO_ACCEPTCONN; 523 if (backlog < 0) 524 backlog = 0; 525 so->so_qlimit = min(backlog, somaxconn); 526 splx(s); 527 return (0); 528 } 529 530 void 531 sofree(struct socket *so) 532 { 533 534 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 535 return; 536 if (so->so_head) { 537 /* 538 * We must not decommission a socket that's on the accept(2) 539 * queue. If we do, then accept(2) may hang after select(2) 540 * indicated that the listening socket was ready. 541 */ 542 if (!soqremque(so, 0)) 543 return; 544 } 545 if (so->so_rcv.sb_hiwat) 546 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 547 RLIM_INFINITY); 548 if (so->so_snd.sb_hiwat) 549 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 550 RLIM_INFINITY); 551 sbrelease(&so->so_snd, so); 552 sorflush(so); 553 pool_put(&socket_pool, so); 554 } 555 556 /* 557 * Close a socket on last file table reference removal. 558 * Initiate disconnect if connected. 559 * Free socket when disconnect complete. 560 */ 561 int 562 soclose(struct socket *so) 563 { 564 struct socket *so2; 565 int s, error; 566 567 error = 0; 568 s = splsoftnet(); /* conservative */ 569 if (so->so_options & SO_ACCEPTCONN) { 570 while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 571 (void) soqremque(so2, 0); 572 (void) soabort(so2); 573 } 574 while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 575 (void) soqremque(so2, 1); 576 (void) soabort(so2); 577 } 578 } 579 if (so->so_pcb == 0) 580 goto discard; 581 if (so->so_state & SS_ISCONNECTED) { 582 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 583 error = sodisconnect(so); 584 if (error) 585 goto drop; 586 } 587 if (so->so_options & SO_LINGER) { 588 if ((so->so_state & SS_ISDISCONNECTING) && 589 (so->so_state & SS_NBIO)) 590 goto drop; 591 while (so->so_state & SS_ISCONNECTED) { 592 error = tsleep((caddr_t)&so->so_timeo, 593 PSOCK | PCATCH, netcls, 594 so->so_linger * hz); 595 if (error) 596 break; 597 } 598 } 599 } 600 drop: 601 if (so->so_pcb) { 602 int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 603 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 604 (struct proc *)0); 605 if (error == 0) 606 error = error2; 607 } 608 discard: 609 if (so->so_state & SS_NOFDREF) 610 panic("soclose: NOFDREF"); 611 so->so_state |= SS_NOFDREF; 612 sofree(so); 613 splx(s); 614 return (error); 615 } 616 617 /* 618 * Must be called at splsoftnet... 619 */ 620 int 621 soabort(struct socket *so) 622 { 623 624 return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0, 625 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 626 } 627 628 int 629 soaccept(struct socket *so, struct mbuf *nam) 630 { 631 int s, error; 632 633 error = 0; 634 s = splsoftnet(); 635 if ((so->so_state & SS_NOFDREF) == 0) 636 panic("soaccept: !NOFDREF"); 637 so->so_state &= ~SS_NOFDREF; 638 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 639 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 640 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 641 (struct mbuf *)0, nam, (struct mbuf *)0, (struct proc *)0); 642 else 643 error = ECONNABORTED; 644 645 splx(s); 646 return (error); 647 } 648 649 int 650 soconnect(struct socket *so, struct mbuf *nam, struct proc *p) 651 { 652 int s, error; 653 654 if (so->so_options & SO_ACCEPTCONN) 655 return (EOPNOTSUPP); 656 s = splsoftnet(); 657 /* 658 * If protocol is connection-based, can only connect once. 659 * Otherwise, if connected, try to disconnect first. 660 * This allows user to disconnect by connecting to, e.g., 661 * a null address. 662 */ 663 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 664 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 665 (error = sodisconnect(so)))) 666 error = EISCONN; 667 else 668 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 669 (struct mbuf *)0, nam, (struct mbuf *)0, p); 670 splx(s); 671 return (error); 672 } 673 674 int 675 soconnect2(struct socket *so1, struct socket *so2) 676 { 677 int s, error; 678 679 s = splsoftnet(); 680 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 681 (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0, 682 (struct proc *)0); 683 splx(s); 684 return (error); 685 } 686 687 int 688 sodisconnect(struct socket *so) 689 { 690 int s, error; 691 692 s = splsoftnet(); 693 if ((so->so_state & SS_ISCONNECTED) == 0) { 694 error = ENOTCONN; 695 goto bad; 696 } 697 if (so->so_state & SS_ISDISCONNECTING) { 698 error = EALREADY; 699 goto bad; 700 } 701 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 702 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 703 (struct proc *)0); 704 bad: 705 splx(s); 706 sodopendfree(so); 707 return (error); 708 } 709 710 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 711 /* 712 * Send on a socket. 713 * If send must go all at once and message is larger than 714 * send buffering, then hard error. 715 * Lock against other senders. 716 * If must go all at once and not enough room now, then 717 * inform user that this would block and do nothing. 718 * Otherwise, if nonblocking, send as much as possible. 719 * The data to be sent is described by "uio" if nonzero, 720 * otherwise by the mbuf chain "top" (which must be null 721 * if uio is not). Data provided in mbuf chain must be small 722 * enough to send all at once. 723 * 724 * Returns nonzero on error, timeout or signal; callers 725 * must check for short counts if EINTR/ERESTART are returned. 726 * Data and control buffers are freed on return. 727 */ 728 int 729 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 730 struct mbuf *control, int flags, struct proc *p) 731 { 732 struct mbuf **mp, *m; 733 long space, len, resid, clen, mlen; 734 int error, s, dontroute, atomic; 735 736 sodopendfree(so); 737 738 clen = 0; 739 atomic = sosendallatonce(so) || top; 740 if (uio) 741 resid = uio->uio_resid; 742 else 743 resid = top->m_pkthdr.len; 744 /* 745 * In theory resid should be unsigned. 746 * However, space must be signed, as it might be less than 0 747 * if we over-committed, and we must use a signed comparison 748 * of space and resid. On the other hand, a negative resid 749 * causes us to loop sending 0-length segments to the protocol. 750 */ 751 if (resid < 0) { 752 error = EINVAL; 753 goto out; 754 } 755 dontroute = 756 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 757 (so->so_proto->pr_flags & PR_ATOMIC); 758 if (p) 759 p->p_stats->p_ru.ru_msgsnd++; 760 if (control) 761 clen = control->m_len; 762 #define snderr(errno) { error = errno; splx(s); goto release; } 763 764 restart: 765 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 766 goto out; 767 do { 768 s = splsoftnet(); 769 if (so->so_state & SS_CANTSENDMORE) 770 snderr(EPIPE); 771 if (so->so_error) { 772 error = so->so_error; 773 so->so_error = 0; 774 splx(s); 775 goto release; 776 } 777 if ((so->so_state & SS_ISCONNECTED) == 0) { 778 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 779 if ((so->so_state & SS_ISCONFIRMING) == 0 && 780 !(resid == 0 && clen != 0)) 781 snderr(ENOTCONN); 782 } else if (addr == 0) 783 snderr(EDESTADDRREQ); 784 } 785 space = sbspace(&so->so_snd); 786 if (flags & MSG_OOB) 787 space += 1024; 788 if ((atomic && resid > so->so_snd.sb_hiwat) || 789 clen > so->so_snd.sb_hiwat) 790 snderr(EMSGSIZE); 791 if (space < resid + clen && 792 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 793 if (so->so_state & SS_NBIO) 794 snderr(EWOULDBLOCK); 795 sbunlock(&so->so_snd); 796 error = sbwait(&so->so_snd); 797 splx(s); 798 if (error) 799 goto out; 800 goto restart; 801 } 802 splx(s); 803 mp = ⊤ 804 space -= clen; 805 do { 806 if (uio == NULL) { 807 /* 808 * Data is prepackaged in "top". 809 */ 810 resid = 0; 811 if (flags & MSG_EOR) 812 top->m_flags |= M_EOR; 813 } else do { 814 if (top == 0) { 815 m = m_gethdr(M_WAIT, MT_DATA); 816 mlen = MHLEN; 817 m->m_pkthdr.len = 0; 818 m->m_pkthdr.rcvif = (struct ifnet *)0; 819 } else { 820 m = m_get(M_WAIT, MT_DATA); 821 mlen = MLEN; 822 } 823 MCLAIM(m, so->so_snd.sb_mowner); 824 if (use_sosend_loan && 825 uio->uio_iov->iov_len >= SOCK_LOAN_THRESH && 826 space >= SOCK_LOAN_THRESH && 827 (len = sosend_loan(so, uio, m, 828 space)) != 0) { 829 SOSEND_COUNTER_INCR(&sosend_loan_big); 830 space -= len; 831 goto have_data; 832 } 833 if (resid >= MINCLSIZE && space >= MCLBYTES) { 834 SOSEND_COUNTER_INCR(&sosend_copy_big); 835 m_clget(m, M_WAIT); 836 if ((m->m_flags & M_EXT) == 0) 837 goto nopages; 838 mlen = MCLBYTES; 839 if (atomic && top == 0) { 840 len = lmin(MCLBYTES - max_hdr, 841 resid); 842 m->m_data += max_hdr; 843 } else 844 len = lmin(MCLBYTES, resid); 845 space -= len; 846 } else { 847 nopages: 848 SOSEND_COUNTER_INCR(&sosend_copy_small); 849 len = lmin(lmin(mlen, resid), space); 850 space -= len; 851 /* 852 * For datagram protocols, leave room 853 * for protocol headers in first mbuf. 854 */ 855 if (atomic && top == 0 && len < mlen) 856 MH_ALIGN(m, len); 857 } 858 error = uiomove(mtod(m, caddr_t), (int)len, 859 uio); 860 have_data: 861 resid = uio->uio_resid; 862 m->m_len = len; 863 *mp = m; 864 top->m_pkthdr.len += len; 865 if (error) 866 goto release; 867 mp = &m->m_next; 868 if (resid <= 0) { 869 if (flags & MSG_EOR) 870 top->m_flags |= M_EOR; 871 break; 872 } 873 } while (space > 0 && atomic); 874 875 s = splsoftnet(); 876 877 if (so->so_state & SS_CANTSENDMORE) 878 snderr(EPIPE); 879 880 if (dontroute) 881 so->so_options |= SO_DONTROUTE; 882 if (resid > 0) 883 so->so_state |= SS_MORETOCOME; 884 error = (*so->so_proto->pr_usrreq)(so, 885 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 886 top, addr, control, p); 887 if (dontroute) 888 so->so_options &= ~SO_DONTROUTE; 889 if (resid > 0) 890 so->so_state &= ~SS_MORETOCOME; 891 splx(s); 892 893 clen = 0; 894 control = 0; 895 top = 0; 896 mp = ⊤ 897 if (error) 898 goto release; 899 } while (resid && space > 0); 900 } while (resid); 901 902 release: 903 sbunlock(&so->so_snd); 904 out: 905 if (top) 906 m_freem(top); 907 if (control) 908 m_freem(control); 909 return (error); 910 } 911 912 /* 913 * Implement receive operations on a socket. 914 * We depend on the way that records are added to the sockbuf 915 * by sbappend*. In particular, each record (mbufs linked through m_next) 916 * must begin with an address if the protocol so specifies, 917 * followed by an optional mbuf or mbufs containing ancillary data, 918 * and then zero or more mbufs of data. 919 * In order to avoid blocking network interrupts for the entire time here, 920 * we splx() while doing the actual copy to user space. 921 * Although the sockbuf is locked, new data may still be appended, 922 * and thus we must maintain consistency of the sockbuf during that time. 923 * 924 * The caller may receive the data as a single mbuf chain by supplying 925 * an mbuf **mp0 for use in returning the chain. The uio is then used 926 * only for the count in uio_resid. 927 */ 928 int 929 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 930 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 931 { 932 struct proc * p; 933 struct mbuf *m, **mp; 934 int flags, len, error, s, offset, moff, type, orig_resid; 935 const struct protosw *pr; 936 struct mbuf *nextrecord; 937 int mbuf_removed = 0; 938 939 pr = so->so_proto; 940 mp = mp0; 941 type = 0; 942 orig_resid = uio->uio_resid; 943 p = uio->uio_procp; 944 945 if (paddr) 946 *paddr = 0; 947 if (controlp) 948 *controlp = 0; 949 if (flagsp) 950 flags = *flagsp &~ MSG_EOR; 951 else 952 flags = 0; 953 954 if ((flags & MSG_DONTWAIT) == 0) 955 sodopendfree(so); 956 957 if (flags & MSG_OOB) { 958 m = m_get(M_WAIT, MT_DATA); 959 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 960 (struct mbuf *)(long)(flags & MSG_PEEK), 961 (struct mbuf *)0, p); 962 if (error) 963 goto bad; 964 do { 965 error = uiomove(mtod(m, caddr_t), 966 (int) min(uio->uio_resid, m->m_len), uio); 967 m = m_free(m); 968 } while (uio->uio_resid && error == 0 && m); 969 bad: 970 if (m) 971 m_freem(m); 972 return (error); 973 } 974 if (mp) 975 *mp = (struct mbuf *)0; 976 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 977 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 978 (struct mbuf *)0, (struct mbuf *)0, p); 979 980 restart: 981 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) 982 return (error); 983 s = splsoftnet(); 984 985 m = so->so_rcv.sb_mb; 986 /* 987 * If we have less data than requested, block awaiting more 988 * (subject to any timeout) if: 989 * 1. the current count is less than the low water mark, 990 * 2. MSG_WAITALL is set, and it is possible to do the entire 991 * receive operation at once if we block (resid <= hiwat), or 992 * 3. MSG_DONTWAIT is not set. 993 * If MSG_WAITALL is set but resid is larger than the receive buffer, 994 * we have to do the receive in sections, and thus risk returning 995 * a short count if a timeout or signal occurs after we start. 996 */ 997 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 998 so->so_rcv.sb_cc < uio->uio_resid) && 999 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1000 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1001 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1002 #ifdef DIAGNOSTIC 1003 if (m == 0 && so->so_rcv.sb_cc) 1004 panic("receive 1"); 1005 #endif 1006 if (so->so_error) { 1007 if (m) 1008 goto dontblock; 1009 error = so->so_error; 1010 if ((flags & MSG_PEEK) == 0) 1011 so->so_error = 0; 1012 goto release; 1013 } 1014 if (so->so_state & SS_CANTRCVMORE) { 1015 if (m) 1016 goto dontblock; 1017 else 1018 goto release; 1019 } 1020 for (; m; m = m->m_next) 1021 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1022 m = so->so_rcv.sb_mb; 1023 goto dontblock; 1024 } 1025 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1026 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1027 error = ENOTCONN; 1028 goto release; 1029 } 1030 if (uio->uio_resid == 0) 1031 goto release; 1032 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 1033 error = EWOULDBLOCK; 1034 goto release; 1035 } 1036 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1037 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1038 sbunlock(&so->so_rcv); 1039 error = sbwait(&so->so_rcv); 1040 splx(s); 1041 if (error) 1042 return (error); 1043 goto restart; 1044 } 1045 dontblock: 1046 /* 1047 * On entry here, m points to the first record of the socket buffer. 1048 * While we process the initial mbufs containing address and control 1049 * info, we save a copy of m->m_nextpkt into nextrecord. 1050 */ 1051 if (p) 1052 p->p_stats->p_ru.ru_msgrcv++; 1053 KASSERT(m == so->so_rcv.sb_mb); 1054 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1055 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1056 nextrecord = m->m_nextpkt; 1057 if (pr->pr_flags & PR_ADDR) { 1058 #ifdef DIAGNOSTIC 1059 if (m->m_type != MT_SONAME) 1060 panic("receive 1a"); 1061 #endif 1062 orig_resid = 0; 1063 if (flags & MSG_PEEK) { 1064 if (paddr) 1065 *paddr = m_copy(m, 0, m->m_len); 1066 m = m->m_next; 1067 } else { 1068 sbfree(&so->so_rcv, m); 1069 mbuf_removed = 1; 1070 if (paddr) { 1071 *paddr = m; 1072 so->so_rcv.sb_mb = m->m_next; 1073 m->m_next = 0; 1074 m = so->so_rcv.sb_mb; 1075 } else { 1076 MFREE(m, so->so_rcv.sb_mb); 1077 m = so->so_rcv.sb_mb; 1078 } 1079 } 1080 } 1081 while (m && m->m_type == MT_CONTROL && error == 0) { 1082 if (flags & MSG_PEEK) { 1083 if (controlp) 1084 *controlp = m_copy(m, 0, m->m_len); 1085 m = m->m_next; 1086 } else { 1087 sbfree(&so->so_rcv, m); 1088 mbuf_removed = 1; 1089 if (controlp) { 1090 struct domain *dom = pr->pr_domain; 1091 if (dom->dom_externalize && p && 1092 mtod(m, struct cmsghdr *)->cmsg_type == 1093 SCM_RIGHTS) 1094 error = (*dom->dom_externalize)(m, p); 1095 *controlp = m; 1096 so->so_rcv.sb_mb = m->m_next; 1097 m->m_next = 0; 1098 m = so->so_rcv.sb_mb; 1099 } else { 1100 /* 1101 * Dispose of any SCM_RIGHTS message that went 1102 * through the read path rather than recv. 1103 */ 1104 if (pr->pr_domain->dom_dispose && 1105 mtod(m, struct cmsghdr *)->cmsg_type == SCM_RIGHTS) 1106 (*pr->pr_domain->dom_dispose)(m); 1107 MFREE(m, so->so_rcv.sb_mb); 1108 m = so->so_rcv.sb_mb; 1109 } 1110 } 1111 if (controlp) { 1112 orig_resid = 0; 1113 controlp = &(*controlp)->m_next; 1114 } 1115 } 1116 1117 /* 1118 * If m is non-NULL, we have some data to read. From now on, 1119 * make sure to keep sb_lastrecord consistent when working on 1120 * the last packet on the chain (nextrecord == NULL) and we 1121 * change m->m_nextpkt. 1122 */ 1123 if (m) { 1124 if ((flags & MSG_PEEK) == 0) { 1125 m->m_nextpkt = nextrecord; 1126 /* 1127 * If nextrecord == NULL (this is a single chain), 1128 * then sb_lastrecord may not be valid here if m 1129 * was changed earlier. 1130 */ 1131 if (nextrecord == NULL) { 1132 KASSERT(so->so_rcv.sb_mb == m); 1133 so->so_rcv.sb_lastrecord = m; 1134 } 1135 } 1136 type = m->m_type; 1137 if (type == MT_OOBDATA) 1138 flags |= MSG_OOB; 1139 } else { 1140 if ((flags & MSG_PEEK) == 0) { 1141 KASSERT(so->so_rcv.sb_mb == m); 1142 so->so_rcv.sb_mb = nextrecord; 1143 SB_EMPTY_FIXUP(&so->so_rcv); 1144 } 1145 } 1146 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1147 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1148 1149 moff = 0; 1150 offset = 0; 1151 while (m && uio->uio_resid > 0 && error == 0) { 1152 if (m->m_type == MT_OOBDATA) { 1153 if (type != MT_OOBDATA) 1154 break; 1155 } else if (type == MT_OOBDATA) 1156 break; 1157 #ifdef DIAGNOSTIC 1158 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1159 panic("receive 3"); 1160 #endif 1161 so->so_state &= ~SS_RCVATMARK; 1162 len = uio->uio_resid; 1163 if (so->so_oobmark && len > so->so_oobmark - offset) 1164 len = so->so_oobmark - offset; 1165 if (len > m->m_len - moff) 1166 len = m->m_len - moff; 1167 /* 1168 * If mp is set, just pass back the mbufs. 1169 * Otherwise copy them out via the uio, then free. 1170 * Sockbuf must be consistent here (points to current mbuf, 1171 * it points to next record) when we drop priority; 1172 * we must note any additions to the sockbuf when we 1173 * block interrupts again. 1174 */ 1175 if (mp == 0) { 1176 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1177 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1178 splx(s); 1179 error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio); 1180 s = splsoftnet(); 1181 if (error) { 1182 /* 1183 * If any part of the record has been removed 1184 * (such as the MT_SONAME mbuf, which will 1185 * happen when PR_ADDR, and thus also 1186 * PR_ATOMIC, is set), then drop the entire 1187 * record to maintain the atomicity of the 1188 * receive operation. 1189 * 1190 * This avoids a later panic("receive 1a") 1191 * when compiled with DIAGNOSTIC. 1192 */ 1193 if (m && mbuf_removed 1194 && (pr->pr_flags & PR_ATOMIC)) 1195 (void) sbdroprecord(&so->so_rcv); 1196 1197 goto release; 1198 } 1199 } else 1200 uio->uio_resid -= len; 1201 if (len == m->m_len - moff) { 1202 if (m->m_flags & M_EOR) 1203 flags |= MSG_EOR; 1204 if (flags & MSG_PEEK) { 1205 m = m->m_next; 1206 moff = 0; 1207 } else { 1208 nextrecord = m->m_nextpkt; 1209 sbfree(&so->so_rcv, m); 1210 if (mp) { 1211 *mp = m; 1212 mp = &m->m_next; 1213 so->so_rcv.sb_mb = m = m->m_next; 1214 *mp = (struct mbuf *)0; 1215 } else { 1216 MFREE(m, so->so_rcv.sb_mb); 1217 m = so->so_rcv.sb_mb; 1218 } 1219 /* 1220 * If m != NULL, we also know that 1221 * so->so_rcv.sb_mb != NULL. 1222 */ 1223 KASSERT(so->so_rcv.sb_mb == m); 1224 if (m) { 1225 m->m_nextpkt = nextrecord; 1226 if (nextrecord == NULL) 1227 so->so_rcv.sb_lastrecord = m; 1228 } else { 1229 so->so_rcv.sb_mb = nextrecord; 1230 SB_EMPTY_FIXUP(&so->so_rcv); 1231 } 1232 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1233 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1234 } 1235 } else { 1236 if (flags & MSG_PEEK) 1237 moff += len; 1238 else { 1239 if (mp) 1240 *mp = m_copym(m, 0, len, M_WAIT); 1241 m->m_data += len; 1242 m->m_len -= len; 1243 so->so_rcv.sb_cc -= len; 1244 } 1245 } 1246 if (so->so_oobmark) { 1247 if ((flags & MSG_PEEK) == 0) { 1248 so->so_oobmark -= len; 1249 if (so->so_oobmark == 0) { 1250 so->so_state |= SS_RCVATMARK; 1251 break; 1252 } 1253 } else { 1254 offset += len; 1255 if (offset == so->so_oobmark) 1256 break; 1257 } 1258 } 1259 if (flags & MSG_EOR) 1260 break; 1261 /* 1262 * If the MSG_WAITALL flag is set (for non-atomic socket), 1263 * we must not quit until "uio->uio_resid == 0" or an error 1264 * termination. If a signal/timeout occurs, return 1265 * with a short count but without error. 1266 * Keep sockbuf locked against other readers. 1267 */ 1268 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1269 !sosendallatonce(so) && !nextrecord) { 1270 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1271 break; 1272 /* 1273 * If we are peeking and the socket receive buffer is 1274 * full, stop since we can't get more data to peek at. 1275 */ 1276 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1277 break; 1278 /* 1279 * If we've drained the socket buffer, tell the 1280 * protocol in case it needs to do something to 1281 * get it filled again. 1282 */ 1283 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1284 (*pr->pr_usrreq)(so, PRU_RCVD, 1285 (struct mbuf *)0, 1286 (struct mbuf *)(long)flags, 1287 (struct mbuf *)0, p); 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, p); 1325 } 1326 if (orig_resid == uio->uio_resid && orig_resid && 1327 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1328 sbunlock(&so->so_rcv); 1329 splx(s); 1330 goto restart; 1331 } 1332 1333 if (flagsp) 1334 *flagsp |= flags; 1335 release: 1336 sbunlock(&so->so_rcv); 1337 splx(s); 1338 return (error); 1339 } 1340 1341 int 1342 soshutdown(struct socket *so, int how) 1343 { 1344 const struct protosw *pr; 1345 1346 pr = so->so_proto; 1347 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1348 return (EINVAL); 1349 1350 if (how == SHUT_RD || how == SHUT_RDWR) 1351 sorflush(so); 1352 if (how == SHUT_WR || how == SHUT_RDWR) 1353 return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0, 1354 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1355 return (0); 1356 } 1357 1358 void 1359 sorflush(struct socket *so) 1360 { 1361 struct sockbuf *sb, asb; 1362 const struct protosw *pr; 1363 int s; 1364 1365 sb = &so->so_rcv; 1366 pr = so->so_proto; 1367 sb->sb_flags |= SB_NOINTR; 1368 (void) sblock(sb, M_WAITOK); 1369 s = splnet(); 1370 socantrcvmore(so); 1371 sbunlock(sb); 1372 asb = *sb; 1373 /* 1374 * Clear most of the sockbuf structure, but leave some of the 1375 * fields valid. 1376 */ 1377 memset(&sb->sb_startzero, 0, 1378 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1379 splx(s); 1380 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1381 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1382 sbrelease(&asb, so); 1383 } 1384 1385 int 1386 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0) 1387 { 1388 int error; 1389 struct mbuf *m; 1390 1391 error = 0; 1392 m = m0; 1393 if (level != SOL_SOCKET) { 1394 if (so->so_proto && so->so_proto->pr_ctloutput) 1395 return ((*so->so_proto->pr_ctloutput) 1396 (PRCO_SETOPT, so, level, optname, &m0)); 1397 error = ENOPROTOOPT; 1398 } else { 1399 switch (optname) { 1400 1401 case SO_LINGER: 1402 if (m == NULL || m->m_len != sizeof(struct linger)) { 1403 error = EINVAL; 1404 goto bad; 1405 } 1406 if (mtod(m, struct linger *)->l_linger < 0 || 1407 mtod(m, struct linger *)->l_linger > (INT_MAX / hz)) { 1408 error = EDOM; 1409 goto bad; 1410 } 1411 so->so_linger = mtod(m, struct linger *)->l_linger; 1412 /* fall thru... */ 1413 1414 case SO_DEBUG: 1415 case SO_KEEPALIVE: 1416 case SO_DONTROUTE: 1417 case SO_USELOOPBACK: 1418 case SO_BROADCAST: 1419 case SO_REUSEADDR: 1420 case SO_REUSEPORT: 1421 case SO_OOBINLINE: 1422 case SO_TIMESTAMP: 1423 if (m == NULL || m->m_len < sizeof(int)) { 1424 error = EINVAL; 1425 goto bad; 1426 } 1427 if (*mtod(m, int *)) 1428 so->so_options |= optname; 1429 else 1430 so->so_options &= ~optname; 1431 break; 1432 1433 case SO_SNDBUF: 1434 case SO_RCVBUF: 1435 case SO_SNDLOWAT: 1436 case SO_RCVLOWAT: 1437 { 1438 int optval; 1439 1440 if (m == NULL || m->m_len < sizeof(int)) { 1441 error = EINVAL; 1442 goto bad; 1443 } 1444 1445 /* 1446 * Values < 1 make no sense for any of these 1447 * options, so disallow them. 1448 */ 1449 optval = *mtod(m, int *); 1450 if (optval < 1) { 1451 error = EINVAL; 1452 goto bad; 1453 } 1454 1455 switch (optname) { 1456 1457 case SO_SNDBUF: 1458 case SO_RCVBUF: 1459 if (sbreserve(optname == SO_SNDBUF ? 1460 &so->so_snd : &so->so_rcv, 1461 (u_long) optval, so) == 0) { 1462 error = ENOBUFS; 1463 goto bad; 1464 } 1465 break; 1466 1467 /* 1468 * Make sure the low-water is never greater than 1469 * the high-water. 1470 */ 1471 case SO_SNDLOWAT: 1472 so->so_snd.sb_lowat = 1473 (optval > so->so_snd.sb_hiwat) ? 1474 so->so_snd.sb_hiwat : optval; 1475 break; 1476 case SO_RCVLOWAT: 1477 so->so_rcv.sb_lowat = 1478 (optval > so->so_rcv.sb_hiwat) ? 1479 so->so_rcv.sb_hiwat : optval; 1480 break; 1481 } 1482 break; 1483 } 1484 1485 case SO_SNDTIMEO: 1486 case SO_RCVTIMEO: 1487 { 1488 struct timeval *tv; 1489 int val; 1490 1491 if (m == NULL || m->m_len < sizeof(*tv)) { 1492 error = EINVAL; 1493 goto bad; 1494 } 1495 tv = mtod(m, struct timeval *); 1496 if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz) { 1497 error = EDOM; 1498 goto bad; 1499 } 1500 val = tv->tv_sec * hz + tv->tv_usec / tick; 1501 if (val == 0 && tv->tv_usec != 0) 1502 val = 1; 1503 1504 switch (optname) { 1505 1506 case SO_SNDTIMEO: 1507 so->so_snd.sb_timeo = val; 1508 break; 1509 case SO_RCVTIMEO: 1510 so->so_rcv.sb_timeo = val; 1511 break; 1512 } 1513 break; 1514 } 1515 1516 default: 1517 error = ENOPROTOOPT; 1518 break; 1519 } 1520 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1521 (void) ((*so->so_proto->pr_ctloutput) 1522 (PRCO_SETOPT, so, level, optname, &m0)); 1523 m = NULL; /* freed by protocol */ 1524 } 1525 } 1526 bad: 1527 if (m) 1528 (void) m_free(m); 1529 return (error); 1530 } 1531 1532 int 1533 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp) 1534 { 1535 struct mbuf *m; 1536 1537 if (level != SOL_SOCKET) { 1538 if (so->so_proto && so->so_proto->pr_ctloutput) { 1539 return ((*so->so_proto->pr_ctloutput) 1540 (PRCO_GETOPT, so, level, optname, mp)); 1541 } else 1542 return (ENOPROTOOPT); 1543 } else { 1544 m = m_get(M_WAIT, MT_SOOPTS); 1545 m->m_len = sizeof(int); 1546 1547 switch (optname) { 1548 1549 case SO_LINGER: 1550 m->m_len = sizeof(struct linger); 1551 mtod(m, struct linger *)->l_onoff = 1552 so->so_options & SO_LINGER; 1553 mtod(m, struct linger *)->l_linger = so->so_linger; 1554 break; 1555 1556 case SO_USELOOPBACK: 1557 case SO_DONTROUTE: 1558 case SO_DEBUG: 1559 case SO_KEEPALIVE: 1560 case SO_REUSEADDR: 1561 case SO_REUSEPORT: 1562 case SO_BROADCAST: 1563 case SO_OOBINLINE: 1564 case SO_TIMESTAMP: 1565 *mtod(m, int *) = so->so_options & optname; 1566 break; 1567 1568 case SO_TYPE: 1569 *mtod(m, int *) = so->so_type; 1570 break; 1571 1572 case SO_ERROR: 1573 *mtod(m, int *) = so->so_error; 1574 so->so_error = 0; 1575 break; 1576 1577 case SO_SNDBUF: 1578 *mtod(m, int *) = so->so_snd.sb_hiwat; 1579 break; 1580 1581 case SO_RCVBUF: 1582 *mtod(m, int *) = so->so_rcv.sb_hiwat; 1583 break; 1584 1585 case SO_SNDLOWAT: 1586 *mtod(m, int *) = so->so_snd.sb_lowat; 1587 break; 1588 1589 case SO_RCVLOWAT: 1590 *mtod(m, int *) = so->so_rcv.sb_lowat; 1591 break; 1592 1593 case SO_SNDTIMEO: 1594 case SO_RCVTIMEO: 1595 { 1596 int val = (optname == SO_SNDTIMEO ? 1597 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1598 1599 m->m_len = sizeof(struct timeval); 1600 mtod(m, struct timeval *)->tv_sec = val / hz; 1601 mtod(m, struct timeval *)->tv_usec = 1602 (val % hz) * tick; 1603 break; 1604 } 1605 1606 case SO_OVERFLOWED: 1607 *mtod(m, int *) = so->so_rcv.sb_overflowed; 1608 break; 1609 1610 default: 1611 (void)m_free(m); 1612 return (ENOPROTOOPT); 1613 } 1614 *mp = m; 1615 return (0); 1616 } 1617 } 1618 1619 void 1620 sohasoutofband(struct socket *so) 1621 { 1622 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 1623 selwakeup(&so->so_rcv.sb_sel); 1624 } 1625 1626 static void 1627 filt_sordetach(struct knote *kn) 1628 { 1629 struct socket *so; 1630 1631 so = (struct socket *)kn->kn_fp->f_data; 1632 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 1633 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 1634 so->so_rcv.sb_flags &= ~SB_KNOTE; 1635 } 1636 1637 /*ARGSUSED*/ 1638 static int 1639 filt_soread(struct knote *kn, long hint) 1640 { 1641 struct socket *so; 1642 1643 so = (struct socket *)kn->kn_fp->f_data; 1644 kn->kn_data = so->so_rcv.sb_cc; 1645 if (so->so_state & SS_CANTRCVMORE) { 1646 kn->kn_flags |= EV_EOF; 1647 kn->kn_fflags = so->so_error; 1648 return (1); 1649 } 1650 if (so->so_error) /* temporary udp error */ 1651 return (1); 1652 if (kn->kn_sfflags & NOTE_LOWAT) 1653 return (kn->kn_data >= kn->kn_sdata); 1654 return (kn->kn_data >= so->so_rcv.sb_lowat); 1655 } 1656 1657 static void 1658 filt_sowdetach(struct knote *kn) 1659 { 1660 struct socket *so; 1661 1662 so = (struct socket *)kn->kn_fp->f_data; 1663 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 1664 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 1665 so->so_snd.sb_flags &= ~SB_KNOTE; 1666 } 1667 1668 /*ARGSUSED*/ 1669 static int 1670 filt_sowrite(struct knote *kn, long hint) 1671 { 1672 struct socket *so; 1673 1674 so = (struct socket *)kn->kn_fp->f_data; 1675 kn->kn_data = sbspace(&so->so_snd); 1676 if (so->so_state & SS_CANTSENDMORE) { 1677 kn->kn_flags |= EV_EOF; 1678 kn->kn_fflags = so->so_error; 1679 return (1); 1680 } 1681 if (so->so_error) /* temporary udp error */ 1682 return (1); 1683 if (((so->so_state & SS_ISCONNECTED) == 0) && 1684 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1685 return (0); 1686 if (kn->kn_sfflags & NOTE_LOWAT) 1687 return (kn->kn_data >= kn->kn_sdata); 1688 return (kn->kn_data >= so->so_snd.sb_lowat); 1689 } 1690 1691 /*ARGSUSED*/ 1692 static int 1693 filt_solisten(struct knote *kn, long hint) 1694 { 1695 struct socket *so; 1696 1697 so = (struct socket *)kn->kn_fp->f_data; 1698 1699 /* 1700 * Set kn_data to number of incoming connections, not 1701 * counting partial (incomplete) connections. 1702 */ 1703 kn->kn_data = so->so_qlen; 1704 return (kn->kn_data > 0); 1705 } 1706 1707 static const struct filterops solisten_filtops = 1708 { 1, NULL, filt_sordetach, filt_solisten }; 1709 static const struct filterops soread_filtops = 1710 { 1, NULL, filt_sordetach, filt_soread }; 1711 static const struct filterops sowrite_filtops = 1712 { 1, NULL, filt_sowdetach, filt_sowrite }; 1713 1714 int 1715 soo_kqfilter(struct file *fp, struct knote *kn) 1716 { 1717 struct socket *so; 1718 struct sockbuf *sb; 1719 1720 so = (struct socket *)kn->kn_fp->f_data; 1721 switch (kn->kn_filter) { 1722 case EVFILT_READ: 1723 if (so->so_options & SO_ACCEPTCONN) 1724 kn->kn_fop = &solisten_filtops; 1725 else 1726 kn->kn_fop = &soread_filtops; 1727 sb = &so->so_rcv; 1728 break; 1729 case EVFILT_WRITE: 1730 kn->kn_fop = &sowrite_filtops; 1731 sb = &so->so_snd; 1732 break; 1733 default: 1734 return (1); 1735 } 1736 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 1737 sb->sb_flags |= SB_KNOTE; 1738 return (0); 1739 } 1740 1741 #include <sys/sysctl.h> 1742 1743 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 1744 1745 /* 1746 * sysctl helper routine for kern.somaxkva. ensures that the given 1747 * value is not too small. 1748 * (XXX should we maybe make sure it's not too large as well?) 1749 */ 1750 static int 1751 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 1752 { 1753 int error, new_somaxkva; 1754 struct sysctlnode node; 1755 int s; 1756 1757 new_somaxkva = somaxkva; 1758 node = *rnode; 1759 node.sysctl_data = &new_somaxkva; 1760 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1761 if (error || newp == NULL) 1762 return (error); 1763 1764 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 1765 return (EINVAL); 1766 1767 s = splvm(); 1768 simple_lock(&so_pendfree_slock); 1769 somaxkva = new_somaxkva; 1770 wakeup(&socurkva); 1771 simple_unlock(&so_pendfree_slock); 1772 splx(s); 1773 1774 return (error); 1775 } 1776 1777 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup") 1778 { 1779 1780 sysctl_createv(clog, 0, NULL, NULL, 1781 CTLFLAG_PERMANENT, 1782 CTLTYPE_NODE, "kern", NULL, 1783 NULL, 0, NULL, 0, 1784 CTL_KERN, CTL_EOL); 1785 1786 sysctl_createv(clog, 0, NULL, NULL, 1787 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1788 CTLTYPE_INT, "somaxkva", 1789 SYSCTL_DESCR("Maximum amount of kernel memory to be " 1790 "used for socket buffers"), 1791 sysctl_kern_somaxkva, 0, NULL, 0, 1792 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 1793 } 1794
Indexes created Sat Sep 20 22:09:52 GMT 2025