uipc_socket.c revision 1.115.6.2
1 /* $NetBSD: uipc_socket.c,v 1.115.6.2 2006/06/01 22:38:09 kardel 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 1.115.6.2 2006/06/01 22:38:09 kardel Exp $"); 72 73 #include "opt_sock_counters.h" 74 #include "opt_sosend_loan.h" 75 #include "opt_mbuftrace.h" 76 #include "opt_somaxkva.h" 77 78 #include <sys/param.h> 79 #include <sys/systm.h> 80 #include <sys/proc.h> 81 #include <sys/file.h> 82 #include <sys/malloc.h> 83 #include <sys/mbuf.h> 84 #include <sys/domain.h> 85 #include <sys/kernel.h> 86 #include <sys/protosw.h> 87 #include <sys/socket.h> 88 #include <sys/socketvar.h> 89 #include <sys/signalvar.h> 90 #include <sys/resourcevar.h> 91 #include <sys/pool.h> 92 #include <sys/event.h> 93 #include <sys/poll.h> 94 #include <sys/kauth.h> 95 96 #include <uvm/uvm.h> 97 98 POOL_INIT(socket_pool, sizeof(struct socket), 0, 0, 0, "sockpl", NULL); 99 100 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options"); 101 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 102 103 extern int somaxconn; /* patchable (XXX sysctl) */ 104 int somaxconn = SOMAXCONN; 105 106 #ifdef SOSEND_COUNTERS 107 #include <sys/device.h> 108 109 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 110 NULL, "sosend", "loan big"); 111 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 112 NULL, "sosend", "copy big"); 113 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 114 NULL, "sosend", "copy small"); 115 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 116 NULL, "sosend", "kva limit"); 117 118 #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++ 119 120 EVCNT_ATTACH_STATIC(sosend_loan_big); 121 EVCNT_ATTACH_STATIC(sosend_copy_big); 122 EVCNT_ATTACH_STATIC(sosend_copy_small); 123 EVCNT_ATTACH_STATIC(sosend_kvalimit); 124 #else 125 126 #define SOSEND_COUNTER_INCR(ev) /* nothing */ 127 128 #endif /* SOSEND_COUNTERS */ 129 130 static struct callback_entry sokva_reclaimerentry; 131 132 #ifdef SOSEND_NO_LOAN 133 int use_sosend_loan = 0; 134 #else 135 int use_sosend_loan = 1; 136 #endif 137 138 static struct simplelock so_pendfree_slock = SIMPLELOCK_INITIALIZER; 139 static struct mbuf *so_pendfree; 140 141 #ifndef SOMAXKVA 142 #define SOMAXKVA (16 * 1024 * 1024) 143 #endif 144 int somaxkva = SOMAXKVA; 145 static int socurkva; 146 static int sokvawaiters; 147 148 #define SOCK_LOAN_THRESH 4096 149 #define SOCK_LOAN_CHUNK 65536 150 151 static size_t sodopendfree(void); 152 static size_t sodopendfreel(void); 153 154 static vsize_t 155 sokvareserve(struct socket *so, vsize_t len) 156 { 157 int s; 158 int error; 159 160 s = splvm(); 161 simple_lock(&so_pendfree_slock); 162 while (socurkva + len > somaxkva) { 163 size_t freed; 164 165 /* 166 * try to do pendfree. 167 */ 168 169 freed = sodopendfreel(); 170 171 /* 172 * if some kva was freed, try again. 173 */ 174 175 if (freed) 176 continue; 177 178 SOSEND_COUNTER_INCR(&sosend_kvalimit); 179 sokvawaiters++; 180 error = ltsleep(&socurkva, PVM | PCATCH, "sokva", 0, 181 &so_pendfree_slock); 182 sokvawaiters--; 183 if (error) { 184 len = 0; 185 break; 186 } 187 } 188 socurkva += len; 189 simple_unlock(&so_pendfree_slock); 190 splx(s); 191 return len; 192 } 193 194 static void 195 sokvaunreserve(vsize_t len) 196 { 197 int s; 198 199 s = splvm(); 200 simple_lock(&so_pendfree_slock); 201 socurkva -= len; 202 if (sokvawaiters) 203 wakeup(&socurkva); 204 simple_unlock(&so_pendfree_slock); 205 splx(s); 206 } 207 208 /* 209 * sokvaalloc: allocate kva for loan. 210 */ 211 212 vaddr_t 213 sokvaalloc(vsize_t len, struct socket *so) 214 { 215 vaddr_t lva; 216 217 /* 218 * reserve kva. 219 */ 220 221 if (sokvareserve(so, len) == 0) 222 return 0; 223 224 /* 225 * allocate kva. 226 */ 227 228 lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA); 229 if (lva == 0) { 230 sokvaunreserve(len); 231 return (0); 232 } 233 234 return lva; 235 } 236 237 /* 238 * sokvafree: free kva for loan. 239 */ 240 241 void 242 sokvafree(vaddr_t sva, vsize_t len) 243 { 244 245 /* 246 * free kva. 247 */ 248 249 uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY); 250 251 /* 252 * unreserve kva. 253 */ 254 255 sokvaunreserve(len); 256 } 257 258 static void 259 sodoloanfree(struct vm_page **pgs, caddr_t buf, size_t size) 260 { 261 vaddr_t va, sva, eva; 262 vsize_t len; 263 paddr_t pa; 264 int i, npgs; 265 266 eva = round_page((vaddr_t) buf + size); 267 sva = trunc_page((vaddr_t) buf); 268 len = eva - sva; 269 npgs = len >> PAGE_SHIFT; 270 271 if (__predict_false(pgs == NULL)) { 272 pgs = alloca(npgs * sizeof(*pgs)); 273 274 for (i = 0, va = sva; va < eva; i++, va += PAGE_SIZE) { 275 if (pmap_extract(pmap_kernel(), va, &pa) == FALSE) 276 panic("sodoloanfree: va 0x%lx not mapped", va); 277 pgs[i] = PHYS_TO_VM_PAGE(pa); 278 } 279 } 280 281 pmap_kremove(sva, len); 282 pmap_update(pmap_kernel()); 283 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 284 sokvafree(sva, len); 285 } 286 287 static size_t 288 sodopendfree() 289 { 290 int s; 291 size_t rv; 292 293 s = splvm(); 294 simple_lock(&so_pendfree_slock); 295 rv = sodopendfreel(); 296 simple_unlock(&so_pendfree_slock); 297 splx(s); 298 299 return rv; 300 } 301 302 /* 303 * sodopendfreel: free mbufs on "pendfree" list. 304 * unlock and relock so_pendfree_slock when freeing mbufs. 305 * 306 * => called with so_pendfree_slock held. 307 * => called at splvm. 308 */ 309 310 static size_t 311 sodopendfreel() 312 { 313 size_t rv = 0; 314 315 LOCK_ASSERT(simple_lock_held(&so_pendfree_slock)); 316 317 for (;;) { 318 struct mbuf *m; 319 struct mbuf *next; 320 321 m = so_pendfree; 322 if (m == NULL) 323 break; 324 so_pendfree = NULL; 325 simple_unlock(&so_pendfree_slock); 326 /* XXX splx */ 327 328 for (; m != NULL; m = next) { 329 next = m->m_next; 330 331 rv += m->m_ext.ext_size; 332 sodoloanfree((m->m_flags & M_EXT_PAGES) ? 333 m->m_ext.ext_pgs : NULL, m->m_ext.ext_buf, 334 m->m_ext.ext_size); 335 pool_cache_put(&mbpool_cache, m); 336 } 337 338 /* XXX splvm */ 339 simple_lock(&so_pendfree_slock); 340 } 341 342 return (rv); 343 } 344 345 void 346 soloanfree(struct mbuf *m, caddr_t buf, size_t size, void *arg) 347 { 348 int s; 349 350 if (m == NULL) { 351 352 /* 353 * called from MEXTREMOVE. 354 */ 355 356 sodoloanfree(NULL, buf, size); 357 return; 358 } 359 360 /* 361 * postpone freeing mbuf. 362 * 363 * we can't do it in interrupt context 364 * because we need to put kva back to kernel_map. 365 */ 366 367 s = splvm(); 368 simple_lock(&so_pendfree_slock); 369 m->m_next = so_pendfree; 370 so_pendfree = m; 371 if (sokvawaiters) 372 wakeup(&socurkva); 373 simple_unlock(&so_pendfree_slock); 374 splx(s); 375 } 376 377 static long 378 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 379 { 380 struct iovec *iov = uio->uio_iov; 381 vaddr_t sva, eva; 382 vsize_t len; 383 vaddr_t lva, va; 384 int npgs, i, error; 385 386 if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace)) 387 return (0); 388 389 if (iov->iov_len < (size_t) space) 390 space = iov->iov_len; 391 if (space > SOCK_LOAN_CHUNK) 392 space = SOCK_LOAN_CHUNK; 393 394 eva = round_page((vaddr_t) iov->iov_base + space); 395 sva = trunc_page((vaddr_t) iov->iov_base); 396 len = eva - sva; 397 npgs = len >> PAGE_SHIFT; 398 399 /* XXX KDASSERT */ 400 KASSERT(npgs <= M_EXT_MAXPAGES); 401 402 lva = sokvaalloc(len, so); 403 if (lva == 0) 404 return 0; 405 406 error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len, 407 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 408 if (error) { 409 sokvafree(lva, len); 410 return (0); 411 } 412 413 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 414 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 415 VM_PROT_READ); 416 pmap_update(pmap_kernel()); 417 418 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 419 420 MEXTADD(m, (caddr_t) lva, space, M_MBUF, soloanfree, so); 421 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 422 423 uio->uio_resid -= space; 424 /* uio_offset not updated, not set/used for write(2) */ 425 uio->uio_iov->iov_base = (caddr_t) uio->uio_iov->iov_base + space; 426 uio->uio_iov->iov_len -= space; 427 if (uio->uio_iov->iov_len == 0) { 428 uio->uio_iov++; 429 uio->uio_iovcnt--; 430 } 431 432 return (space); 433 } 434 435 static int 436 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg) 437 { 438 439 KASSERT(ce == &sokva_reclaimerentry); 440 KASSERT(obj == NULL); 441 442 sodopendfree(); 443 if (!vm_map_starved_p(kernel_map)) { 444 return CALLBACK_CHAIN_ABORT; 445 } 446 return CALLBACK_CHAIN_CONTINUE; 447 } 448 449 void 450 soinit(void) 451 { 452 453 /* Set the initial adjusted socket buffer size. */ 454 if (sb_max_set(sb_max)) 455 panic("bad initial sb_max value: %lu", sb_max); 456 457 callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback, 458 &sokva_reclaimerentry, NULL, sokva_reclaim_callback); 459 } 460 461 /* 462 * Socket operation routines. 463 * These routines are called by the routines in 464 * sys_socket.c or from a system process, and 465 * implement the semantics of socket operations by 466 * switching out to the protocol specific routines. 467 */ 468 /*ARGSUSED*/ 469 int 470 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l) 471 { 472 const struct protosw *prp; 473 struct socket *so; 474 uid_t uid; 475 int error, s; 476 477 if (proto) 478 prp = pffindproto(dom, proto, type); 479 else 480 prp = pffindtype(dom, type); 481 if (prp == 0 || prp->pr_usrreq == 0) 482 return (EPROTONOSUPPORT); 483 if (prp->pr_type != type) 484 return (EPROTOTYPE); 485 s = splsoftnet(); 486 so = pool_get(&socket_pool, PR_WAITOK); 487 memset((caddr_t)so, 0, sizeof(*so)); 488 TAILQ_INIT(&so->so_q0); 489 TAILQ_INIT(&so->so_q); 490 so->so_type = type; 491 so->so_proto = prp; 492 so->so_send = sosend; 493 so->so_receive = soreceive; 494 #ifdef MBUFTRACE 495 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 496 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 497 so->so_mowner = &prp->pr_domain->dom_mowner; 498 #endif 499 if (l != NULL) { 500 uid = kauth_cred_geteuid(l->l_proc->p_cred); 501 } else { 502 uid = 0; 503 } 504 so->so_uidinfo = uid_find(uid); 505 error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0, 506 (struct mbuf *)(long)proto, (struct mbuf *)0, l); 507 if (error) { 508 so->so_state |= SS_NOFDREF; 509 sofree(so); 510 splx(s); 511 return (error); 512 } 513 splx(s); 514 *aso = so; 515 return (0); 516 } 517 518 int 519 sobind(struct socket *so, struct mbuf *nam, struct lwp *l) 520 { 521 int s, error; 522 523 s = splsoftnet(); 524 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0, 525 nam, (struct mbuf *)0, l); 526 splx(s); 527 return (error); 528 } 529 530 int 531 solisten(struct socket *so, int backlog) 532 { 533 int s, error; 534 535 s = splsoftnet(); 536 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0, 537 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 538 if (error) { 539 splx(s); 540 return (error); 541 } 542 if (TAILQ_EMPTY(&so->so_q)) 543 so->so_options |= SO_ACCEPTCONN; 544 if (backlog < 0) 545 backlog = 0; 546 so->so_qlimit = min(backlog, somaxconn); 547 splx(s); 548 return (0); 549 } 550 551 void 552 sofree(struct socket *so) 553 { 554 555 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 556 return; 557 if (so->so_head) { 558 /* 559 * We must not decommission a socket that's on the accept(2) 560 * queue. If we do, then accept(2) may hang after select(2) 561 * indicated that the listening socket was ready. 562 */ 563 if (!soqremque(so, 0)) 564 return; 565 } 566 if (so->so_rcv.sb_hiwat) 567 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 568 RLIM_INFINITY); 569 if (so->so_snd.sb_hiwat) 570 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 571 RLIM_INFINITY); 572 sbrelease(&so->so_snd, so); 573 sorflush(so); 574 pool_put(&socket_pool, so); 575 } 576 577 /* 578 * Close a socket on last file table reference removal. 579 * Initiate disconnect if connected. 580 * Free socket when disconnect complete. 581 */ 582 int 583 soclose(struct socket *so) 584 { 585 struct socket *so2; 586 int s, error; 587 588 error = 0; 589 s = splsoftnet(); /* conservative */ 590 if (so->so_options & SO_ACCEPTCONN) { 591 while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 592 (void) soqremque(so2, 0); 593 (void) soabort(so2); 594 } 595 while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 596 (void) soqremque(so2, 1); 597 (void) soabort(so2); 598 } 599 } 600 if (so->so_pcb == 0) 601 goto discard; 602 if (so->so_state & SS_ISCONNECTED) { 603 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 604 error = sodisconnect(so); 605 if (error) 606 goto drop; 607 } 608 if (so->so_options & SO_LINGER) { 609 if ((so->so_state & SS_ISDISCONNECTING) && 610 (so->so_state & SS_NBIO)) 611 goto drop; 612 while (so->so_state & SS_ISCONNECTED) { 613 error = tsleep((caddr_t)&so->so_timeo, 614 PSOCK | PCATCH, netcls, 615 so->so_linger * hz); 616 if (error) 617 break; 618 } 619 } 620 } 621 drop: 622 if (so->so_pcb) { 623 int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 624 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 625 (struct lwp *)0); 626 if (error == 0) 627 error = error2; 628 } 629 discard: 630 if (so->so_state & SS_NOFDREF) 631 panic("soclose: NOFDREF"); 632 so->so_state |= SS_NOFDREF; 633 sofree(so); 634 splx(s); 635 return (error); 636 } 637 638 /* 639 * Must be called at splsoftnet... 640 */ 641 int 642 soabort(struct socket *so) 643 { 644 645 return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0, 646 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 647 } 648 649 int 650 soaccept(struct socket *so, struct mbuf *nam) 651 { 652 int s, error; 653 654 error = 0; 655 s = splsoftnet(); 656 if ((so->so_state & SS_NOFDREF) == 0) 657 panic("soaccept: !NOFDREF"); 658 so->so_state &= ~SS_NOFDREF; 659 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 660 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 661 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 662 (struct mbuf *)0, nam, (struct mbuf *)0, (struct lwp *)0); 663 else 664 error = ECONNABORTED; 665 666 splx(s); 667 return (error); 668 } 669 670 int 671 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l) 672 { 673 int s, error; 674 675 if (so->so_options & SO_ACCEPTCONN) 676 return (EOPNOTSUPP); 677 s = splsoftnet(); 678 /* 679 * If protocol is connection-based, can only connect once. 680 * Otherwise, if connected, try to disconnect first. 681 * This allows user to disconnect by connecting to, e.g., 682 * a null address. 683 */ 684 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 685 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 686 (error = sodisconnect(so)))) 687 error = EISCONN; 688 else 689 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 690 (struct mbuf *)0, nam, (struct mbuf *)0, l); 691 splx(s); 692 return (error); 693 } 694 695 int 696 soconnect2(struct socket *so1, struct socket *so2) 697 { 698 int s, error; 699 700 s = splsoftnet(); 701 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 702 (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0, 703 (struct lwp *)0); 704 splx(s); 705 return (error); 706 } 707 708 int 709 sodisconnect(struct socket *so) 710 { 711 int s, error; 712 713 s = splsoftnet(); 714 if ((so->so_state & SS_ISCONNECTED) == 0) { 715 error = ENOTCONN; 716 goto bad; 717 } 718 if (so->so_state & SS_ISDISCONNECTING) { 719 error = EALREADY; 720 goto bad; 721 } 722 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 723 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 724 (struct lwp *)0); 725 bad: 726 splx(s); 727 sodopendfree(); 728 return (error); 729 } 730 731 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 732 /* 733 * Send on a socket. 734 * If send must go all at once and message is larger than 735 * send buffering, then hard error. 736 * Lock against other senders. 737 * If must go all at once and not enough room now, then 738 * inform user that this would block and do nothing. 739 * Otherwise, if nonblocking, send as much as possible. 740 * The data to be sent is described by "uio" if nonzero, 741 * otherwise by the mbuf chain "top" (which must be null 742 * if uio is not). Data provided in mbuf chain must be small 743 * enough to send all at once. 744 * 745 * Returns nonzero on error, timeout or signal; callers 746 * must check for short counts if EINTR/ERESTART are returned. 747 * Data and control buffers are freed on return. 748 */ 749 int 750 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 751 struct mbuf *control, int flags, struct lwp *l) 752 { 753 struct mbuf **mp, *m; 754 struct proc *p; 755 long space, len, resid, clen, mlen; 756 int error, s, dontroute, atomic; 757 758 p = l->l_proc; 759 sodopendfree(); 760 761 clen = 0; 762 atomic = sosendallatonce(so) || top; 763 if (uio) 764 resid = uio->uio_resid; 765 else 766 resid = top->m_pkthdr.len; 767 /* 768 * In theory resid should be unsigned. 769 * However, space must be signed, as it might be less than 0 770 * if we over-committed, and we must use a signed comparison 771 * of space and resid. On the other hand, a negative resid 772 * causes us to loop sending 0-length segments to the protocol. 773 */ 774 if (resid < 0) { 775 error = EINVAL; 776 goto out; 777 } 778 dontroute = 779 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 780 (so->so_proto->pr_flags & PR_ATOMIC); 781 if (p) 782 p->p_stats->p_ru.ru_msgsnd++; 783 if (control) 784 clen = control->m_len; 785 #define snderr(errno) { error = errno; splx(s); goto release; } 786 787 restart: 788 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 789 goto out; 790 do { 791 s = splsoftnet(); 792 if (so->so_state & SS_CANTSENDMORE) 793 snderr(EPIPE); 794 if (so->so_error) { 795 error = so->so_error; 796 so->so_error = 0; 797 splx(s); 798 goto release; 799 } 800 if ((so->so_state & SS_ISCONNECTED) == 0) { 801 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 802 if ((so->so_state & SS_ISCONFIRMING) == 0 && 803 !(resid == 0 && clen != 0)) 804 snderr(ENOTCONN); 805 } else if (addr == 0) 806 snderr(EDESTADDRREQ); 807 } 808 space = sbspace(&so->so_snd); 809 if (flags & MSG_OOB) 810 space += 1024; 811 if ((atomic && resid > so->so_snd.sb_hiwat) || 812 clen > so->so_snd.sb_hiwat) 813 snderr(EMSGSIZE); 814 if (space < resid + clen && 815 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 816 if (so->so_state & SS_NBIO) 817 snderr(EWOULDBLOCK); 818 sbunlock(&so->so_snd); 819 error = sbwait(&so->so_snd); 820 splx(s); 821 if (error) 822 goto out; 823 goto restart; 824 } 825 splx(s); 826 mp = ⊤ 827 space -= clen; 828 do { 829 if (uio == NULL) { 830 /* 831 * Data is prepackaged in "top". 832 */ 833 resid = 0; 834 if (flags & MSG_EOR) 835 top->m_flags |= M_EOR; 836 } else do { 837 if (top == 0) { 838 m = m_gethdr(M_WAIT, MT_DATA); 839 mlen = MHLEN; 840 m->m_pkthdr.len = 0; 841 m->m_pkthdr.rcvif = (struct ifnet *)0; 842 } else { 843 m = m_get(M_WAIT, MT_DATA); 844 mlen = MLEN; 845 } 846 MCLAIM(m, so->so_snd.sb_mowner); 847 if (use_sosend_loan && 848 uio->uio_iov->iov_len >= SOCK_LOAN_THRESH && 849 space >= SOCK_LOAN_THRESH && 850 (len = sosend_loan(so, uio, m, 851 space)) != 0) { 852 SOSEND_COUNTER_INCR(&sosend_loan_big); 853 space -= len; 854 goto have_data; 855 } 856 if (resid >= MINCLSIZE && space >= MCLBYTES) { 857 SOSEND_COUNTER_INCR(&sosend_copy_big); 858 m_clget(m, M_WAIT); 859 if ((m->m_flags & M_EXT) == 0) 860 goto nopages; 861 mlen = MCLBYTES; 862 if (atomic && top == 0) { 863 len = lmin(MCLBYTES - max_hdr, 864 resid); 865 m->m_data += max_hdr; 866 } else 867 len = lmin(MCLBYTES, resid); 868 space -= len; 869 } else { 870 nopages: 871 SOSEND_COUNTER_INCR(&sosend_copy_small); 872 len = lmin(lmin(mlen, resid), space); 873 space -= len; 874 /* 875 * For datagram protocols, leave room 876 * for protocol headers in first mbuf. 877 */ 878 if (atomic && top == 0 && len < mlen) 879 MH_ALIGN(m, len); 880 } 881 error = uiomove(mtod(m, caddr_t), (int)len, 882 uio); 883 have_data: 884 resid = uio->uio_resid; 885 m->m_len = len; 886 *mp = m; 887 top->m_pkthdr.len += len; 888 if (error) 889 goto release; 890 mp = &m->m_next; 891 if (resid <= 0) { 892 if (flags & MSG_EOR) 893 top->m_flags |= M_EOR; 894 break; 895 } 896 } while (space > 0 && atomic); 897 898 s = splsoftnet(); 899 900 if (so->so_state & SS_CANTSENDMORE) 901 snderr(EPIPE); 902 903 if (dontroute) 904 so->so_options |= SO_DONTROUTE; 905 if (resid > 0) 906 so->so_state |= SS_MORETOCOME; 907 error = (*so->so_proto->pr_usrreq)(so, 908 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 909 top, addr, control, curlwp); /* XXX */ 910 if (dontroute) 911 so->so_options &= ~SO_DONTROUTE; 912 if (resid > 0) 913 so->so_state &= ~SS_MORETOCOME; 914 splx(s); 915 916 clen = 0; 917 control = 0; 918 top = 0; 919 mp = ⊤ 920 if (error) 921 goto release; 922 } while (resid && space > 0); 923 } while (resid); 924 925 release: 926 sbunlock(&so->so_snd); 927 out: 928 if (top) 929 m_freem(top); 930 if (control) 931 m_freem(control); 932 return (error); 933 } 934 935 /* 936 * Implement receive operations on a socket. 937 * We depend on the way that records are added to the sockbuf 938 * by sbappend*. In particular, each record (mbufs linked through m_next) 939 * must begin with an address if the protocol so specifies, 940 * followed by an optional mbuf or mbufs containing ancillary data, 941 * and then zero or more mbufs of data. 942 * In order to avoid blocking network interrupts for the entire time here, 943 * we splx() while doing the actual copy to user space. 944 * Although the sockbuf is locked, new data may still be appended, 945 * and thus we must maintain consistency of the sockbuf during that time. 946 * 947 * The caller may receive the data as a single mbuf chain by supplying 948 * an mbuf **mp0 for use in returning the chain. The uio is then used 949 * only for the count in uio_resid. 950 */ 951 int 952 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 953 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 954 { 955 struct lwp *l = curlwp; 956 struct mbuf *m, **mp; 957 int flags, len, error, s, offset, moff, type, orig_resid; 958 const struct protosw *pr; 959 struct mbuf *nextrecord; 960 int mbuf_removed = 0; 961 962 pr = so->so_proto; 963 mp = mp0; 964 type = 0; 965 orig_resid = uio->uio_resid; 966 967 if (paddr) 968 *paddr = 0; 969 if (controlp) 970 *controlp = 0; 971 if (flagsp) 972 flags = *flagsp &~ MSG_EOR; 973 else 974 flags = 0; 975 976 if ((flags & MSG_DONTWAIT) == 0) 977 sodopendfree(); 978 979 if (flags & MSG_OOB) { 980 m = m_get(M_WAIT, MT_DATA); 981 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 982 (struct mbuf *)(long)(flags & MSG_PEEK), 983 (struct mbuf *)0, l); 984 if (error) 985 goto bad; 986 do { 987 error = uiomove(mtod(m, caddr_t), 988 (int) min(uio->uio_resid, m->m_len), uio); 989 m = m_free(m); 990 } while (uio->uio_resid && error == 0 && m); 991 bad: 992 if (m) 993 m_freem(m); 994 return (error); 995 } 996 if (mp) 997 *mp = (struct mbuf *)0; 998 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 999 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1000 (struct mbuf *)0, (struct mbuf *)0, l); 1001 1002 restart: 1003 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) 1004 return (error); 1005 s = splsoftnet(); 1006 1007 m = so->so_rcv.sb_mb; 1008 /* 1009 * If we have less data than requested, block awaiting more 1010 * (subject to any timeout) if: 1011 * 1. the current count is less than the low water mark, 1012 * 2. MSG_WAITALL is set, and it is possible to do the entire 1013 * receive operation at once if we block (resid <= hiwat), or 1014 * 3. MSG_DONTWAIT is not set. 1015 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1016 * we have to do the receive in sections, and thus risk returning 1017 * a short count if a timeout or signal occurs after we start. 1018 */ 1019 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 1020 so->so_rcv.sb_cc < uio->uio_resid) && 1021 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1022 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1023 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1024 #ifdef DIAGNOSTIC 1025 if (m == 0 && so->so_rcv.sb_cc) 1026 panic("receive 1"); 1027 #endif 1028 if (so->so_error) { 1029 if (m) 1030 goto dontblock; 1031 error = so->so_error; 1032 if ((flags & MSG_PEEK) == 0) 1033 so->so_error = 0; 1034 goto release; 1035 } 1036 if (so->so_state & SS_CANTRCVMORE) { 1037 if (m) 1038 goto dontblock; 1039 else 1040 goto release; 1041 } 1042 for (; m; m = m->m_next) 1043 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1044 m = so->so_rcv.sb_mb; 1045 goto dontblock; 1046 } 1047 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1048 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1049 error = ENOTCONN; 1050 goto release; 1051 } 1052 if (uio->uio_resid == 0) 1053 goto release; 1054 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 1055 error = EWOULDBLOCK; 1056 goto release; 1057 } 1058 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1059 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1060 sbunlock(&so->so_rcv); 1061 error = sbwait(&so->so_rcv); 1062 splx(s); 1063 if (error) 1064 return (error); 1065 goto restart; 1066 } 1067 dontblock: 1068 /* 1069 * On entry here, m points to the first record of the socket buffer. 1070 * While we process the initial mbufs containing address and control 1071 * info, we save a copy of m->m_nextpkt into nextrecord. 1072 */ 1073 if (l) 1074 l->l_proc->p_stats->p_ru.ru_msgrcv++; 1075 KASSERT(m == so->so_rcv.sb_mb); 1076 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1077 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1078 nextrecord = m->m_nextpkt; 1079 if (pr->pr_flags & PR_ADDR) { 1080 #ifdef DIAGNOSTIC 1081 if (m->m_type != MT_SONAME) 1082 panic("receive 1a"); 1083 #endif 1084 orig_resid = 0; 1085 if (flags & MSG_PEEK) { 1086 if (paddr) 1087 *paddr = m_copy(m, 0, m->m_len); 1088 m = m->m_next; 1089 } else { 1090 sbfree(&so->so_rcv, m); 1091 mbuf_removed = 1; 1092 if (paddr) { 1093 *paddr = m; 1094 so->so_rcv.sb_mb = m->m_next; 1095 m->m_next = 0; 1096 m = so->so_rcv.sb_mb; 1097 } else { 1098 MFREE(m, so->so_rcv.sb_mb); 1099 m = so->so_rcv.sb_mb; 1100 } 1101 } 1102 } 1103 while (m && m->m_type == MT_CONTROL && error == 0) { 1104 if (flags & MSG_PEEK) { 1105 if (controlp) 1106 *controlp = m_copy(m, 0, m->m_len); 1107 m = m->m_next; 1108 } else { 1109 sbfree(&so->so_rcv, m); 1110 mbuf_removed = 1; 1111 if (controlp) { 1112 struct domain *dom = pr->pr_domain; 1113 if (dom->dom_externalize && l && 1114 mtod(m, struct cmsghdr *)->cmsg_type == 1115 SCM_RIGHTS) 1116 error = (*dom->dom_externalize)(m, l); 1117 *controlp = m; 1118 so->so_rcv.sb_mb = m->m_next; 1119 m->m_next = 0; 1120 m = so->so_rcv.sb_mb; 1121 } else { 1122 /* 1123 * Dispose of any SCM_RIGHTS message that went 1124 * through the read path rather than recv. 1125 */ 1126 if (pr->pr_domain->dom_dispose && 1127 mtod(m, struct cmsghdr *)->cmsg_type == SCM_RIGHTS) 1128 (*pr->pr_domain->dom_dispose)(m); 1129 MFREE(m, so->so_rcv.sb_mb); 1130 m = so->so_rcv.sb_mb; 1131 } 1132 } 1133 if (controlp) { 1134 orig_resid = 0; 1135 controlp = &(*controlp)->m_next; 1136 } 1137 } 1138 1139 /* 1140 * If m is non-NULL, we have some data to read. From now on, 1141 * make sure to keep sb_lastrecord consistent when working on 1142 * the last packet on the chain (nextrecord == NULL) and we 1143 * change m->m_nextpkt. 1144 */ 1145 if (m) { 1146 if ((flags & MSG_PEEK) == 0) { 1147 m->m_nextpkt = nextrecord; 1148 /* 1149 * If nextrecord == NULL (this is a single chain), 1150 * then sb_lastrecord may not be valid here if m 1151 * was changed earlier. 1152 */ 1153 if (nextrecord == NULL) { 1154 KASSERT(so->so_rcv.sb_mb == m); 1155 so->so_rcv.sb_lastrecord = m; 1156 } 1157 } 1158 type = m->m_type; 1159 if (type == MT_OOBDATA) 1160 flags |= MSG_OOB; 1161 } else { 1162 if ((flags & MSG_PEEK) == 0) { 1163 KASSERT(so->so_rcv.sb_mb == m); 1164 so->so_rcv.sb_mb = nextrecord; 1165 SB_EMPTY_FIXUP(&so->so_rcv); 1166 } 1167 } 1168 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1169 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1170 1171 moff = 0; 1172 offset = 0; 1173 while (m && uio->uio_resid > 0 && error == 0) { 1174 if (m->m_type == MT_OOBDATA) { 1175 if (type != MT_OOBDATA) 1176 break; 1177 } else if (type == MT_OOBDATA) 1178 break; 1179 #ifdef DIAGNOSTIC 1180 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1181 panic("receive 3"); 1182 #endif 1183 so->so_state &= ~SS_RCVATMARK; 1184 len = uio->uio_resid; 1185 if (so->so_oobmark && len > so->so_oobmark - offset) 1186 len = so->so_oobmark - offset; 1187 if (len > m->m_len - moff) 1188 len = m->m_len - moff; 1189 /* 1190 * If mp is set, just pass back the mbufs. 1191 * Otherwise copy them out via the uio, then free. 1192 * Sockbuf must be consistent here (points to current mbuf, 1193 * it points to next record) when we drop priority; 1194 * we must note any additions to the sockbuf when we 1195 * block interrupts again. 1196 */ 1197 if (mp == 0) { 1198 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1199 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1200 splx(s); 1201 error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio); 1202 s = splsoftnet(); 1203 if (error) { 1204 /* 1205 * If any part of the record has been removed 1206 * (such as the MT_SONAME mbuf, which will 1207 * happen when PR_ADDR, and thus also 1208 * PR_ATOMIC, is set), then drop the entire 1209 * record to maintain the atomicity of the 1210 * receive operation. 1211 * 1212 * This avoids a later panic("receive 1a") 1213 * when compiled with DIAGNOSTIC. 1214 */ 1215 if (m && mbuf_removed 1216 && (pr->pr_flags & PR_ATOMIC)) 1217 (void) sbdroprecord(&so->so_rcv); 1218 1219 goto release; 1220 } 1221 } else 1222 uio->uio_resid -= len; 1223 if (len == m->m_len - moff) { 1224 if (m->m_flags & M_EOR) 1225 flags |= MSG_EOR; 1226 if (flags & MSG_PEEK) { 1227 m = m->m_next; 1228 moff = 0; 1229 } else { 1230 nextrecord = m->m_nextpkt; 1231 sbfree(&so->so_rcv, m); 1232 if (mp) { 1233 *mp = m; 1234 mp = &m->m_next; 1235 so->so_rcv.sb_mb = m = m->m_next; 1236 *mp = (struct mbuf *)0; 1237 } else { 1238 MFREE(m, so->so_rcv.sb_mb); 1239 m = so->so_rcv.sb_mb; 1240 } 1241 /* 1242 * If m != NULL, we also know that 1243 * so->so_rcv.sb_mb != NULL. 1244 */ 1245 KASSERT(so->so_rcv.sb_mb == m); 1246 if (m) { 1247 m->m_nextpkt = nextrecord; 1248 if (nextrecord == NULL) 1249 so->so_rcv.sb_lastrecord = m; 1250 } else { 1251 so->so_rcv.sb_mb = nextrecord; 1252 SB_EMPTY_FIXUP(&so->so_rcv); 1253 } 1254 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1255 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1256 } 1257 } else { 1258 if (flags & MSG_PEEK) 1259 moff += len; 1260 else { 1261 if (mp) 1262 *mp = m_copym(m, 0, len, M_WAIT); 1263 m->m_data += len; 1264 m->m_len -= len; 1265 so->so_rcv.sb_cc -= len; 1266 } 1267 } 1268 if (so->so_oobmark) { 1269 if ((flags & MSG_PEEK) == 0) { 1270 so->so_oobmark -= len; 1271 if (so->so_oobmark == 0) { 1272 so->so_state |= SS_RCVATMARK; 1273 break; 1274 } 1275 } else { 1276 offset += len; 1277 if (offset == so->so_oobmark) 1278 break; 1279 } 1280 } 1281 if (flags & MSG_EOR) 1282 break; 1283 /* 1284 * If the MSG_WAITALL flag is set (for non-atomic socket), 1285 * we must not quit until "uio->uio_resid == 0" or an error 1286 * termination. If a signal/timeout occurs, return 1287 * with a short count but without error. 1288 * Keep sockbuf locked against other readers. 1289 */ 1290 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1291 !sosendallatonce(so) && !nextrecord) { 1292 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1293 break; 1294 /* 1295 * If we are peeking and the socket receive buffer is 1296 * full, stop since we can't get more data to peek at. 1297 */ 1298 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1299 break; 1300 /* 1301 * If we've drained the socket buffer, tell the 1302 * protocol in case it needs to do something to 1303 * get it filled again. 1304 */ 1305 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1306 (*pr->pr_usrreq)(so, PRU_RCVD, 1307 (struct mbuf *)0, 1308 (struct mbuf *)(long)flags, 1309 (struct mbuf *)0, l); 1310 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1311 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1312 error = sbwait(&so->so_rcv); 1313 if (error) { 1314 sbunlock(&so->so_rcv); 1315 splx(s); 1316 return (0); 1317 } 1318 if ((m = so->so_rcv.sb_mb) != NULL) 1319 nextrecord = m->m_nextpkt; 1320 } 1321 } 1322 1323 if (m && pr->pr_flags & PR_ATOMIC) { 1324 flags |= MSG_TRUNC; 1325 if ((flags & MSG_PEEK) == 0) 1326 (void) sbdroprecord(&so->so_rcv); 1327 } 1328 if ((flags & MSG_PEEK) == 0) { 1329 if (m == 0) { 1330 /* 1331 * First part is an inline SB_EMPTY_FIXUP(). Second 1332 * part makes sure sb_lastrecord is up-to-date if 1333 * there is still data in the socket buffer. 1334 */ 1335 so->so_rcv.sb_mb = nextrecord; 1336 if (so->so_rcv.sb_mb == NULL) { 1337 so->so_rcv.sb_mbtail = NULL; 1338 so->so_rcv.sb_lastrecord = NULL; 1339 } else if (nextrecord->m_nextpkt == NULL) 1340 so->so_rcv.sb_lastrecord = nextrecord; 1341 } 1342 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1343 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1344 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1345 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1346 (struct mbuf *)(long)flags, (struct mbuf *)0, l); 1347 } 1348 if (orig_resid == uio->uio_resid && orig_resid && 1349 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1350 sbunlock(&so->so_rcv); 1351 splx(s); 1352 goto restart; 1353 } 1354 1355 if (flagsp) 1356 *flagsp |= flags; 1357 release: 1358 sbunlock(&so->so_rcv); 1359 splx(s); 1360 return (error); 1361 } 1362 1363 int 1364 soshutdown(struct socket *so, int how) 1365 { 1366 const struct protosw *pr; 1367 1368 pr = so->so_proto; 1369 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1370 return (EINVAL); 1371 1372 if (how == SHUT_RD || how == SHUT_RDWR) 1373 sorflush(so); 1374 if (how == SHUT_WR || how == SHUT_RDWR) 1375 return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0, 1376 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1377 return (0); 1378 } 1379 1380 void 1381 sorflush(struct socket *so) 1382 { 1383 struct sockbuf *sb, asb; 1384 const struct protosw *pr; 1385 int s; 1386 1387 sb = &so->so_rcv; 1388 pr = so->so_proto; 1389 sb->sb_flags |= SB_NOINTR; 1390 (void) sblock(sb, M_WAITOK); 1391 s = splnet(); 1392 socantrcvmore(so); 1393 sbunlock(sb); 1394 asb = *sb; 1395 /* 1396 * Clear most of the sockbuf structure, but leave some of the 1397 * fields valid. 1398 */ 1399 memset(&sb->sb_startzero, 0, 1400 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1401 splx(s); 1402 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1403 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1404 sbrelease(&asb, so); 1405 } 1406 1407 int 1408 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0) 1409 { 1410 int error; 1411 struct mbuf *m; 1412 1413 error = 0; 1414 m = m0; 1415 if (level != SOL_SOCKET) { 1416 if (so->so_proto && so->so_proto->pr_ctloutput) 1417 return ((*so->so_proto->pr_ctloutput) 1418 (PRCO_SETOPT, so, level, optname, &m0)); 1419 error = ENOPROTOOPT; 1420 } else { 1421 switch (optname) { 1422 1423 case SO_LINGER: 1424 if (m == NULL || m->m_len != sizeof(struct linger)) { 1425 error = EINVAL; 1426 goto bad; 1427 } 1428 if (mtod(m, struct linger *)->l_linger < 0 || 1429 mtod(m, struct linger *)->l_linger > (INT_MAX / hz)) { 1430 error = EDOM; 1431 goto bad; 1432 } 1433 so->so_linger = mtod(m, struct linger *)->l_linger; 1434 /* fall thru... */ 1435 1436 case SO_DEBUG: 1437 case SO_KEEPALIVE: 1438 case SO_DONTROUTE: 1439 case SO_USELOOPBACK: 1440 case SO_BROADCAST: 1441 case SO_REUSEADDR: 1442 case SO_REUSEPORT: 1443 case SO_OOBINLINE: 1444 case SO_TIMESTAMP: 1445 if (m == NULL || m->m_len < sizeof(int)) { 1446 error = EINVAL; 1447 goto bad; 1448 } 1449 if (*mtod(m, int *)) 1450 so->so_options |= optname; 1451 else 1452 so->so_options &= ~optname; 1453 break; 1454 1455 case SO_SNDBUF: 1456 case SO_RCVBUF: 1457 case SO_SNDLOWAT: 1458 case SO_RCVLOWAT: 1459 { 1460 int optval; 1461 1462 if (m == NULL || m->m_len < sizeof(int)) { 1463 error = EINVAL; 1464 goto bad; 1465 } 1466 1467 /* 1468 * Values < 1 make no sense for any of these 1469 * options, so disallow them. 1470 */ 1471 optval = *mtod(m, int *); 1472 if (optval < 1) { 1473 error = EINVAL; 1474 goto bad; 1475 } 1476 1477 switch (optname) { 1478 1479 case SO_SNDBUF: 1480 case SO_RCVBUF: 1481 if (sbreserve(optname == SO_SNDBUF ? 1482 &so->so_snd : &so->so_rcv, 1483 (u_long) optval, so) == 0) { 1484 error = ENOBUFS; 1485 goto bad; 1486 } 1487 break; 1488 1489 /* 1490 * Make sure the low-water is never greater than 1491 * the high-water. 1492 */ 1493 case SO_SNDLOWAT: 1494 so->so_snd.sb_lowat = 1495 (optval > so->so_snd.sb_hiwat) ? 1496 so->so_snd.sb_hiwat : optval; 1497 break; 1498 case SO_RCVLOWAT: 1499 so->so_rcv.sb_lowat = 1500 (optval > so->so_rcv.sb_hiwat) ? 1501 so->so_rcv.sb_hiwat : optval; 1502 break; 1503 } 1504 break; 1505 } 1506 1507 case SO_SNDTIMEO: 1508 case SO_RCVTIMEO: 1509 { 1510 struct timeval *tv; 1511 int val; 1512 1513 if (m == NULL || m->m_len < sizeof(*tv)) { 1514 error = EINVAL; 1515 goto bad; 1516 } 1517 tv = mtod(m, struct timeval *); 1518 if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz) { 1519 error = EDOM; 1520 goto bad; 1521 } 1522 val = tv->tv_sec * hz + tv->tv_usec / tick; 1523 if (val == 0 && tv->tv_usec != 0) 1524 val = 1; 1525 1526 switch (optname) { 1527 1528 case SO_SNDTIMEO: 1529 so->so_snd.sb_timeo = val; 1530 break; 1531 case SO_RCVTIMEO: 1532 so->so_rcv.sb_timeo = val; 1533 break; 1534 } 1535 break; 1536 } 1537 1538 default: 1539 error = ENOPROTOOPT; 1540 break; 1541 } 1542 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1543 (void) ((*so->so_proto->pr_ctloutput) 1544 (PRCO_SETOPT, so, level, optname, &m0)); 1545 m = NULL; /* freed by protocol */ 1546 } 1547 } 1548 bad: 1549 if (m) 1550 (void) m_free(m); 1551 return (error); 1552 } 1553 1554 int 1555 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp) 1556 { 1557 struct mbuf *m; 1558 1559 if (level != SOL_SOCKET) { 1560 if (so->so_proto && so->so_proto->pr_ctloutput) { 1561 return ((*so->so_proto->pr_ctloutput) 1562 (PRCO_GETOPT, so, level, optname, mp)); 1563 } else 1564 return (ENOPROTOOPT); 1565 } else { 1566 m = m_get(M_WAIT, MT_SOOPTS); 1567 m->m_len = sizeof(int); 1568 1569 switch (optname) { 1570 1571 case SO_LINGER: 1572 m->m_len = sizeof(struct linger); 1573 mtod(m, struct linger *)->l_onoff = 1574 so->so_options & SO_LINGER; 1575 mtod(m, struct linger *)->l_linger = so->so_linger; 1576 break; 1577 1578 case SO_USELOOPBACK: 1579 case SO_DONTROUTE: 1580 case SO_DEBUG: 1581 case SO_KEEPALIVE: 1582 case SO_REUSEADDR: 1583 case SO_REUSEPORT: 1584 case SO_BROADCAST: 1585 case SO_OOBINLINE: 1586 case SO_TIMESTAMP: 1587 *mtod(m, int *) = so->so_options & optname; 1588 break; 1589 1590 case SO_TYPE: 1591 *mtod(m, int *) = so->so_type; 1592 break; 1593 1594 case SO_ERROR: 1595 *mtod(m, int *) = so->so_error; 1596 so->so_error = 0; 1597 break; 1598 1599 case SO_SNDBUF: 1600 *mtod(m, int *) = so->so_snd.sb_hiwat; 1601 break; 1602 1603 case SO_RCVBUF: 1604 *mtod(m, int *) = so->so_rcv.sb_hiwat; 1605 break; 1606 1607 case SO_SNDLOWAT: 1608 *mtod(m, int *) = so->so_snd.sb_lowat; 1609 break; 1610 1611 case SO_RCVLOWAT: 1612 *mtod(m, int *) = so->so_rcv.sb_lowat; 1613 break; 1614 1615 case SO_SNDTIMEO: 1616 case SO_RCVTIMEO: 1617 { 1618 int val = (optname == SO_SNDTIMEO ? 1619 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1620 1621 m->m_len = sizeof(struct timeval); 1622 mtod(m, struct timeval *)->tv_sec = val / hz; 1623 mtod(m, struct timeval *)->tv_usec = 1624 (val % hz) * tick; 1625 break; 1626 } 1627 1628 case SO_OVERFLOWED: 1629 *mtod(m, int *) = so->so_rcv.sb_overflowed; 1630 break; 1631 1632 default: 1633 (void)m_free(m); 1634 return (ENOPROTOOPT); 1635 } 1636 *mp = m; 1637 return (0); 1638 } 1639 } 1640 1641 void 1642 sohasoutofband(struct socket *so) 1643 { 1644 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 1645 selwakeup(&so->so_rcv.sb_sel); 1646 } 1647 1648 static void 1649 filt_sordetach(struct knote *kn) 1650 { 1651 struct socket *so; 1652 1653 so = (struct socket *)kn->kn_fp->f_data; 1654 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 1655 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 1656 so->so_rcv.sb_flags &= ~SB_KNOTE; 1657 } 1658 1659 /*ARGSUSED*/ 1660 static int 1661 filt_soread(struct knote *kn, long hint) 1662 { 1663 struct socket *so; 1664 1665 so = (struct socket *)kn->kn_fp->f_data; 1666 kn->kn_data = so->so_rcv.sb_cc; 1667 if (so->so_state & SS_CANTRCVMORE) { 1668 kn->kn_flags |= EV_EOF; 1669 kn->kn_fflags = so->so_error; 1670 return (1); 1671 } 1672 if (so->so_error) /* temporary udp error */ 1673 return (1); 1674 if (kn->kn_sfflags & NOTE_LOWAT) 1675 return (kn->kn_data >= kn->kn_sdata); 1676 return (kn->kn_data >= so->so_rcv.sb_lowat); 1677 } 1678 1679 static void 1680 filt_sowdetach(struct knote *kn) 1681 { 1682 struct socket *so; 1683 1684 so = (struct socket *)kn->kn_fp->f_data; 1685 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 1686 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 1687 so->so_snd.sb_flags &= ~SB_KNOTE; 1688 } 1689 1690 /*ARGSUSED*/ 1691 static int 1692 filt_sowrite(struct knote *kn, long hint) 1693 { 1694 struct socket *so; 1695 1696 so = (struct socket *)kn->kn_fp->f_data; 1697 kn->kn_data = sbspace(&so->so_snd); 1698 if (so->so_state & SS_CANTSENDMORE) { 1699 kn->kn_flags |= EV_EOF; 1700 kn->kn_fflags = so->so_error; 1701 return (1); 1702 } 1703 if (so->so_error) /* temporary udp error */ 1704 return (1); 1705 if (((so->so_state & SS_ISCONNECTED) == 0) && 1706 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1707 return (0); 1708 if (kn->kn_sfflags & NOTE_LOWAT) 1709 return (kn->kn_data >= kn->kn_sdata); 1710 return (kn->kn_data >= so->so_snd.sb_lowat); 1711 } 1712 1713 /*ARGSUSED*/ 1714 static int 1715 filt_solisten(struct knote *kn, long hint) 1716 { 1717 struct socket *so; 1718 1719 so = (struct socket *)kn->kn_fp->f_data; 1720 1721 /* 1722 * Set kn_data to number of incoming connections, not 1723 * counting partial (incomplete) connections. 1724 */ 1725 kn->kn_data = so->so_qlen; 1726 return (kn->kn_data > 0); 1727 } 1728 1729 static const struct filterops solisten_filtops = 1730 { 1, NULL, filt_sordetach, filt_solisten }; 1731 static const struct filterops soread_filtops = 1732 { 1, NULL, filt_sordetach, filt_soread }; 1733 static const struct filterops sowrite_filtops = 1734 { 1, NULL, filt_sowdetach, filt_sowrite }; 1735 1736 int 1737 soo_kqfilter(struct file *fp, struct knote *kn) 1738 { 1739 struct socket *so; 1740 struct sockbuf *sb; 1741 1742 so = (struct socket *)kn->kn_fp->f_data; 1743 switch (kn->kn_filter) { 1744 case EVFILT_READ: 1745 if (so->so_options & SO_ACCEPTCONN) 1746 kn->kn_fop = &solisten_filtops; 1747 else 1748 kn->kn_fop = &soread_filtops; 1749 sb = &so->so_rcv; 1750 break; 1751 case EVFILT_WRITE: 1752 kn->kn_fop = &sowrite_filtops; 1753 sb = &so->so_snd; 1754 break; 1755 default: 1756 return (1); 1757 } 1758 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 1759 sb->sb_flags |= SB_KNOTE; 1760 return (0); 1761 } 1762 1763 #include <sys/sysctl.h> 1764 1765 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 1766 1767 /* 1768 * sysctl helper routine for kern.somaxkva. ensures that the given 1769 * value is not too small. 1770 * (XXX should we maybe make sure it's not too large as well?) 1771 */ 1772 static int 1773 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 1774 { 1775 int error, new_somaxkva; 1776 struct sysctlnode node; 1777 int s; 1778 1779 new_somaxkva = somaxkva; 1780 node = *rnode; 1781 node.sysctl_data = &new_somaxkva; 1782 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1783 if (error || newp == NULL) 1784 return (error); 1785 1786 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 1787 return (EINVAL); 1788 1789 s = splvm(); 1790 simple_lock(&so_pendfree_slock); 1791 somaxkva = new_somaxkva; 1792 wakeup(&socurkva); 1793 simple_unlock(&so_pendfree_slock); 1794 splx(s); 1795 1796 return (error); 1797 } 1798 1799 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup") 1800 { 1801 1802 sysctl_createv(clog, 0, NULL, NULL, 1803 CTLFLAG_PERMANENT, 1804 CTLTYPE_NODE, "kern", NULL, 1805 NULL, 0, NULL, 0, 1806 CTL_KERN, CTL_EOL); 1807 1808 sysctl_createv(clog, 0, NULL, NULL, 1809 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1810 CTLTYPE_INT, "somaxkva", 1811 SYSCTL_DESCR("Maximum amount of kernel memory to be " 1812 "used for socket buffers"), 1813 sysctl_kern_somaxkva, 0, NULL, 0, 1814 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 1815 } 1816
Indexes created Sat Sep 20 22:09:52 GMT 2025