1 /* $NetBSD: uipc_socket.c,v 1.114 2005/12/11 12:24:30 christos 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 1.114 2005/12/11 12:24:30 christos 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_lwp != NULL); 409 410 lva = sokvaalloc(len, so); 411 if (lva == 0) 412 return 0; 413 414 error = uvm_loan(&uio->uio_lwp->l_proc->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 lwp *l) 453 { 454 const struct protosw *prp; 455 struct socket *so; 456 struct proc *p; 457 int error, s; 458 459 p = l->l_proc; 460 if (proto) 461 prp = pffindproto(dom, proto, type); 462 else 463 prp = pffindtype(dom, type); 464 if (prp == 0 || prp->pr_usrreq == 0) 465 return (EPROTONOSUPPORT); 466 if (prp->pr_type != type) 467 return (EPROTOTYPE); 468 s = splsoftnet(); 469 so = pool_get(&socket_pool, PR_WAITOK); 470 memset((caddr_t)so, 0, sizeof(*so)); 471 TAILQ_INIT(&so->so_q0); 472 TAILQ_INIT(&so->so_q); 473 so->so_type = type; 474 so->so_proto = prp; 475 so->so_send = sosend; 476 so->so_receive = soreceive; 477 #ifdef MBUFTRACE 478 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 479 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 480 so->so_mowner = &prp->pr_domain->dom_mowner; 481 #endif 482 if (p != 0) 483 so->so_uidinfo = uid_find(p->p_ucred->cr_uid); 484 else 485 so->so_uidinfo = uid_find(0); 486 error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0, 487 (struct mbuf *)(long)proto, (struct mbuf *)0, l); 488 if (error) { 489 so->so_state |= SS_NOFDREF; 490 sofree(so); 491 splx(s); 492 return (error); 493 } 494 splx(s); 495 *aso = so; 496 return (0); 497 } 498 499 int 500 sobind(struct socket *so, struct mbuf *nam, struct lwp *l) 501 { 502 int s, error; 503 504 s = splsoftnet(); 505 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0, 506 nam, (struct mbuf *)0, l); 507 splx(s); 508 return (error); 509 } 510 511 int 512 solisten(struct socket *so, int backlog) 513 { 514 int s, error; 515 516 s = splsoftnet(); 517 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0, 518 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 519 if (error) { 520 splx(s); 521 return (error); 522 } 523 if (TAILQ_EMPTY(&so->so_q)) 524 so->so_options |= SO_ACCEPTCONN; 525 if (backlog < 0) 526 backlog = 0; 527 so->so_qlimit = min(backlog, somaxconn); 528 splx(s); 529 return (0); 530 } 531 532 void 533 sofree(struct socket *so) 534 { 535 536 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 537 return; 538 if (so->so_head) { 539 /* 540 * We must not decommission a socket that's on the accept(2) 541 * queue. If we do, then accept(2) may hang after select(2) 542 * indicated that the listening socket was ready. 543 */ 544 if (!soqremque(so, 0)) 545 return; 546 } 547 if (so->so_rcv.sb_hiwat) 548 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 549 RLIM_INFINITY); 550 if (so->so_snd.sb_hiwat) 551 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 552 RLIM_INFINITY); 553 sbrelease(&so->so_snd, so); 554 sorflush(so); 555 pool_put(&socket_pool, so); 556 } 557 558 /* 559 * Close a socket on last file table reference removal. 560 * Initiate disconnect if connected. 561 * Free socket when disconnect complete. 562 */ 563 int 564 soclose(struct socket *so) 565 { 566 struct socket *so2; 567 int s, error; 568 569 error = 0; 570 s = splsoftnet(); /* conservative */ 571 if (so->so_options & SO_ACCEPTCONN) { 572 while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 573 (void) soqremque(so2, 0); 574 (void) soabort(so2); 575 } 576 while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 577 (void) soqremque(so2, 1); 578 (void) soabort(so2); 579 } 580 } 581 if (so->so_pcb == 0) 582 goto discard; 583 if (so->so_state & SS_ISCONNECTED) { 584 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 585 error = sodisconnect(so); 586 if (error) 587 goto drop; 588 } 589 if (so->so_options & SO_LINGER) { 590 if ((so->so_state & SS_ISDISCONNECTING) && 591 (so->so_state & SS_NBIO)) 592 goto drop; 593 while (so->so_state & SS_ISCONNECTED) { 594 error = tsleep((caddr_t)&so->so_timeo, 595 PSOCK | PCATCH, netcls, 596 so->so_linger * hz); 597 if (error) 598 break; 599 } 600 } 601 } 602 drop: 603 if (so->so_pcb) { 604 int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 605 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 606 (struct lwp *)0); 607 if (error == 0) 608 error = error2; 609 } 610 discard: 611 if (so->so_state & SS_NOFDREF) 612 panic("soclose: NOFDREF"); 613 so->so_state |= SS_NOFDREF; 614 sofree(so); 615 splx(s); 616 return (error); 617 } 618 619 /* 620 * Must be called at splsoftnet... 621 */ 622 int 623 soabort(struct socket *so) 624 { 625 626 return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0, 627 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 628 } 629 630 int 631 soaccept(struct socket *so, struct mbuf *nam) 632 { 633 int s, error; 634 635 error = 0; 636 s = splsoftnet(); 637 if ((so->so_state & SS_NOFDREF) == 0) 638 panic("soaccept: !NOFDREF"); 639 so->so_state &= ~SS_NOFDREF; 640 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 641 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 642 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 643 (struct mbuf *)0, nam, (struct mbuf *)0, (struct lwp *)0); 644 else 645 error = ECONNABORTED; 646 647 splx(s); 648 return (error); 649 } 650 651 int 652 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l) 653 { 654 int s, error; 655 656 if (so->so_options & SO_ACCEPTCONN) 657 return (EOPNOTSUPP); 658 s = splsoftnet(); 659 /* 660 * If protocol is connection-based, can only connect once. 661 * Otherwise, if connected, try to disconnect first. 662 * This allows user to disconnect by connecting to, e.g., 663 * a null address. 664 */ 665 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 666 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 667 (error = sodisconnect(so)))) 668 error = EISCONN; 669 else 670 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 671 (struct mbuf *)0, nam, (struct mbuf *)0, l); 672 splx(s); 673 return (error); 674 } 675 676 int 677 soconnect2(struct socket *so1, struct socket *so2) 678 { 679 int s, error; 680 681 s = splsoftnet(); 682 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 683 (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0, 684 (struct lwp *)0); 685 splx(s); 686 return (error); 687 } 688 689 int 690 sodisconnect(struct socket *so) 691 { 692 int s, error; 693 694 s = splsoftnet(); 695 if ((so->so_state & SS_ISCONNECTED) == 0) { 696 error = ENOTCONN; 697 goto bad; 698 } 699 if (so->so_state & SS_ISDISCONNECTING) { 700 error = EALREADY; 701 goto bad; 702 } 703 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 704 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 705 (struct lwp *)0); 706 bad: 707 splx(s); 708 sodopendfree(so); 709 return (error); 710 } 711 712 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 713 /* 714 * Send on a socket. 715 * If send must go all at once and message is larger than 716 * send buffering, then hard error. 717 * Lock against other senders. 718 * If must go all at once and not enough room now, then 719 * inform user that this would block and do nothing. 720 * Otherwise, if nonblocking, send as much as possible. 721 * The data to be sent is described by "uio" if nonzero, 722 * otherwise by the mbuf chain "top" (which must be null 723 * if uio is not). Data provided in mbuf chain must be small 724 * enough to send all at once. 725 * 726 * Returns nonzero on error, timeout or signal; callers 727 * must check for short counts if EINTR/ERESTART are returned. 728 * Data and control buffers are freed on return. 729 */ 730 int 731 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 732 struct mbuf *control, int flags, struct lwp *l) 733 { 734 struct mbuf **mp, *m; 735 struct proc *p; 736 long space, len, resid, clen, mlen; 737 int error, s, dontroute, atomic; 738 739 p = l->l_proc; 740 sodopendfree(so); 741 742 clen = 0; 743 atomic = sosendallatonce(so) || top; 744 if (uio) 745 resid = uio->uio_resid; 746 else 747 resid = top->m_pkthdr.len; 748 /* 749 * In theory resid should be unsigned. 750 * However, space must be signed, as it might be less than 0 751 * if we over-committed, and we must use a signed comparison 752 * of space and resid. On the other hand, a negative resid 753 * causes us to loop sending 0-length segments to the protocol. 754 */ 755 if (resid < 0) { 756 error = EINVAL; 757 goto out; 758 } 759 dontroute = 760 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 761 (so->so_proto->pr_flags & PR_ATOMIC); 762 if (p) 763 p->p_stats->p_ru.ru_msgsnd++; 764 if (control) 765 clen = control->m_len; 766 #define snderr(errno) { error = errno; splx(s); goto release; } 767 768 restart: 769 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 770 goto out; 771 do { 772 s = splsoftnet(); 773 if (so->so_state & SS_CANTSENDMORE) 774 snderr(EPIPE); 775 if (so->so_error) { 776 error = so->so_error; 777 so->so_error = 0; 778 splx(s); 779 goto release; 780 } 781 if ((so->so_state & SS_ISCONNECTED) == 0) { 782 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 783 if ((so->so_state & SS_ISCONFIRMING) == 0 && 784 !(resid == 0 && clen != 0)) 785 snderr(ENOTCONN); 786 } else if (addr == 0) 787 snderr(EDESTADDRREQ); 788 } 789 space = sbspace(&so->so_snd); 790 if (flags & MSG_OOB) 791 space += 1024; 792 if ((atomic && resid > so->so_snd.sb_hiwat) || 793 clen > so->so_snd.sb_hiwat) 794 snderr(EMSGSIZE); 795 if (space < resid + clen && 796 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 797 if (so->so_state & SS_NBIO) 798 snderr(EWOULDBLOCK); 799 sbunlock(&so->so_snd); 800 error = sbwait(&so->so_snd); 801 splx(s); 802 if (error) 803 goto out; 804 goto restart; 805 } 806 splx(s); 807 mp = ⊤ 808 space -= clen; 809 do { 810 if (uio == NULL) { 811 /* 812 * Data is prepackaged in "top". 813 */ 814 resid = 0; 815 if (flags & MSG_EOR) 816 top->m_flags |= M_EOR; 817 } else do { 818 if (top == 0) { 819 m = m_gethdr(M_WAIT, MT_DATA); 820 mlen = MHLEN; 821 m->m_pkthdr.len = 0; 822 m->m_pkthdr.rcvif = (struct ifnet *)0; 823 } else { 824 m = m_get(M_WAIT, MT_DATA); 825 mlen = MLEN; 826 } 827 MCLAIM(m, so->so_snd.sb_mowner); 828 if (use_sosend_loan && 829 uio->uio_iov->iov_len >= SOCK_LOAN_THRESH && 830 space >= SOCK_LOAN_THRESH && 831 (len = sosend_loan(so, uio, m, 832 space)) != 0) { 833 SOSEND_COUNTER_INCR(&sosend_loan_big); 834 space -= len; 835 goto have_data; 836 } 837 if (resid >= MINCLSIZE && space >= MCLBYTES) { 838 SOSEND_COUNTER_INCR(&sosend_copy_big); 839 m_clget(m, M_WAIT); 840 if ((m->m_flags & M_EXT) == 0) 841 goto nopages; 842 mlen = MCLBYTES; 843 if (atomic && top == 0) { 844 len = lmin(MCLBYTES - max_hdr, 845 resid); 846 m->m_data += max_hdr; 847 } else 848 len = lmin(MCLBYTES, resid); 849 space -= len; 850 } else { 851 nopages: 852 SOSEND_COUNTER_INCR(&sosend_copy_small); 853 len = lmin(lmin(mlen, resid), space); 854 space -= len; 855 /* 856 * For datagram protocols, leave room 857 * for protocol headers in first mbuf. 858 */ 859 if (atomic && top == 0 && len < mlen) 860 MH_ALIGN(m, len); 861 } 862 error = uiomove(mtod(m, caddr_t), (int)len, 863 uio); 864 have_data: 865 resid = uio->uio_resid; 866 m->m_len = len; 867 *mp = m; 868 top->m_pkthdr.len += len; 869 if (error) 870 goto release; 871 mp = &m->m_next; 872 if (resid <= 0) { 873 if (flags & MSG_EOR) 874 top->m_flags |= M_EOR; 875 break; 876 } 877 } while (space > 0 && atomic); 878 879 s = splsoftnet(); 880 881 if (so->so_state & SS_CANTSENDMORE) 882 snderr(EPIPE); 883 884 if (dontroute) 885 so->so_options |= SO_DONTROUTE; 886 if (resid > 0) 887 so->so_state |= SS_MORETOCOME; 888 error = (*so->so_proto->pr_usrreq)(so, 889 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 890 top, addr, control, curlwp); /* XXX */ 891 if (dontroute) 892 so->so_options &= ~SO_DONTROUTE; 893 if (resid > 0) 894 so->so_state &= ~SS_MORETOCOME; 895 splx(s); 896 897 clen = 0; 898 control = 0; 899 top = 0; 900 mp = ⊤ 901 if (error) 902 goto release; 903 } while (resid && space > 0); 904 } while (resid); 905 906 release: 907 sbunlock(&so->so_snd); 908 out: 909 if (top) 910 m_freem(top); 911 if (control) 912 m_freem(control); 913 return (error); 914 } 915 916 /* 917 * Implement receive operations on a socket. 918 * We depend on the way that records are added to the sockbuf 919 * by sbappend*. In particular, each record (mbufs linked through m_next) 920 * must begin with an address if the protocol so specifies, 921 * followed by an optional mbuf or mbufs containing ancillary data, 922 * and then zero or more mbufs of data. 923 * In order to avoid blocking network interrupts for the entire time here, 924 * we splx() while doing the actual copy to user space. 925 * Although the sockbuf is locked, new data may still be appended, 926 * and thus we must maintain consistency of the sockbuf during that time. 927 * 928 * The caller may receive the data as a single mbuf chain by supplying 929 * an mbuf **mp0 for use in returning the chain. The uio is then used 930 * only for the count in uio_resid. 931 */ 932 int 933 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 934 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 935 { 936 struct lwp *l; 937 struct mbuf *m, **mp; 938 int flags, len, error, s, offset, moff, type, orig_resid; 939 const struct protosw *pr; 940 struct mbuf *nextrecord; 941 int mbuf_removed = 0; 942 943 pr = so->so_proto; 944 mp = mp0; 945 type = 0; 946 orig_resid = uio->uio_resid; 947 l = uio->uio_lwp; 948 949 if (paddr) 950 *paddr = 0; 951 if (controlp) 952 *controlp = 0; 953 if (flagsp) 954 flags = *flagsp &~ MSG_EOR; 955 else 956 flags = 0; 957 958 if ((flags & MSG_DONTWAIT) == 0) 959 sodopendfree(so); 960 961 if (flags & MSG_OOB) { 962 m = m_get(M_WAIT, MT_DATA); 963 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 964 (struct mbuf *)(long)(flags & MSG_PEEK), 965 (struct mbuf *)0, l); 966 if (error) 967 goto bad; 968 do { 969 error = uiomove(mtod(m, caddr_t), 970 (int) min(uio->uio_resid, m->m_len), uio); 971 m = m_free(m); 972 } while (uio->uio_resid && error == 0 && m); 973 bad: 974 if (m) 975 m_freem(m); 976 return (error); 977 } 978 if (mp) 979 *mp = (struct mbuf *)0; 980 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 981 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 982 (struct mbuf *)0, (struct mbuf *)0, l); 983 984 restart: 985 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) 986 return (error); 987 s = splsoftnet(); 988 989 m = so->so_rcv.sb_mb; 990 /* 991 * If we have less data than requested, block awaiting more 992 * (subject to any timeout) if: 993 * 1. the current count is less than the low water mark, 994 * 2. MSG_WAITALL is set, and it is possible to do the entire 995 * receive operation at once if we block (resid <= hiwat), or 996 * 3. MSG_DONTWAIT is not set. 997 * If MSG_WAITALL is set but resid is larger than the receive buffer, 998 * we have to do the receive in sections, and thus risk returning 999 * a short count if a timeout or signal occurs after we start. 1000 */ 1001 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 1002 so->so_rcv.sb_cc < uio->uio_resid) && 1003 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1004 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1005 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1006 #ifdef DIAGNOSTIC 1007 if (m == 0 && so->so_rcv.sb_cc) 1008 panic("receive 1"); 1009 #endif 1010 if (so->so_error) { 1011 if (m) 1012 goto dontblock; 1013 error = so->so_error; 1014 if ((flags & MSG_PEEK) == 0) 1015 so->so_error = 0; 1016 goto release; 1017 } 1018 if (so->so_state & SS_CANTRCVMORE) { 1019 if (m) 1020 goto dontblock; 1021 else 1022 goto release; 1023 } 1024 for (; m; m = m->m_next) 1025 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1026 m = so->so_rcv.sb_mb; 1027 goto dontblock; 1028 } 1029 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1030 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1031 error = ENOTCONN; 1032 goto release; 1033 } 1034 if (uio->uio_resid == 0) 1035 goto release; 1036 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 1037 error = EWOULDBLOCK; 1038 goto release; 1039 } 1040 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1041 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1042 sbunlock(&so->so_rcv); 1043 error = sbwait(&so->so_rcv); 1044 splx(s); 1045 if (error) 1046 return (error); 1047 goto restart; 1048 } 1049 dontblock: 1050 /* 1051 * On entry here, m points to the first record of the socket buffer. 1052 * While we process the initial mbufs containing address and control 1053 * info, we save a copy of m->m_nextpkt into nextrecord. 1054 */ 1055 if (l) 1056 l->l_proc->p_stats->p_ru.ru_msgrcv++; 1057 KASSERT(m == so->so_rcv.sb_mb); 1058 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1059 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1060 nextrecord = m->m_nextpkt; 1061 if (pr->pr_flags & PR_ADDR) { 1062 #ifdef DIAGNOSTIC 1063 if (m->m_type != MT_SONAME) 1064 panic("receive 1a"); 1065 #endif 1066 orig_resid = 0; 1067 if (flags & MSG_PEEK) { 1068 if (paddr) 1069 *paddr = m_copy(m, 0, m->m_len); 1070 m = m->m_next; 1071 } else { 1072 sbfree(&so->so_rcv, m); 1073 mbuf_removed = 1; 1074 if (paddr) { 1075 *paddr = m; 1076 so->so_rcv.sb_mb = m->m_next; 1077 m->m_next = 0; 1078 m = so->so_rcv.sb_mb; 1079 } else { 1080 MFREE(m, so->so_rcv.sb_mb); 1081 m = so->so_rcv.sb_mb; 1082 } 1083 } 1084 } 1085 while (m && m->m_type == MT_CONTROL && error == 0) { 1086 if (flags & MSG_PEEK) { 1087 if (controlp) 1088 *controlp = m_copy(m, 0, m->m_len); 1089 m = m->m_next; 1090 } else { 1091 sbfree(&so->so_rcv, m); 1092 mbuf_removed = 1; 1093 if (controlp) { 1094 struct domain *dom = pr->pr_domain; 1095 if (dom->dom_externalize && l && 1096 mtod(m, struct cmsghdr *)->cmsg_type == 1097 SCM_RIGHTS) 1098 error = (*dom->dom_externalize)(m, l); 1099 *controlp = m; 1100 so->so_rcv.sb_mb = m->m_next; 1101 m->m_next = 0; 1102 m = so->so_rcv.sb_mb; 1103 } else { 1104 /* 1105 * Dispose of any SCM_RIGHTS message that went 1106 * through the read path rather than recv. 1107 */ 1108 if (pr->pr_domain->dom_dispose && 1109 mtod(m, struct cmsghdr *)->cmsg_type == SCM_RIGHTS) 1110 (*pr->pr_domain->dom_dispose)(m); 1111 MFREE(m, so->so_rcv.sb_mb); 1112 m = so->so_rcv.sb_mb; 1113 } 1114 } 1115 if (controlp) { 1116 orig_resid = 0; 1117 controlp = &(*controlp)->m_next; 1118 } 1119 } 1120 1121 /* 1122 * If m is non-NULL, we have some data to read. From now on, 1123 * make sure to keep sb_lastrecord consistent when working on 1124 * the last packet on the chain (nextrecord == NULL) and we 1125 * change m->m_nextpkt. 1126 */ 1127 if (m) { 1128 if ((flags & MSG_PEEK) == 0) { 1129 m->m_nextpkt = nextrecord; 1130 /* 1131 * If nextrecord == NULL (this is a single chain), 1132 * then sb_lastrecord may not be valid here if m 1133 * was changed earlier. 1134 */ 1135 if (nextrecord == NULL) { 1136 KASSERT(so->so_rcv.sb_mb == m); 1137 so->so_rcv.sb_lastrecord = m; 1138 } 1139 } 1140 type = m->m_type; 1141 if (type == MT_OOBDATA) 1142 flags |= MSG_OOB; 1143 } else { 1144 if ((flags & MSG_PEEK) == 0) { 1145 KASSERT(so->so_rcv.sb_mb == m); 1146 so->so_rcv.sb_mb = nextrecord; 1147 SB_EMPTY_FIXUP(&so->so_rcv); 1148 } 1149 } 1150 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1151 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1152 1153 moff = 0; 1154 offset = 0; 1155 while (m && uio->uio_resid > 0 && error == 0) { 1156 if (m->m_type == MT_OOBDATA) { 1157 if (type != MT_OOBDATA) 1158 break; 1159 } else if (type == MT_OOBDATA) 1160 break; 1161 #ifdef DIAGNOSTIC 1162 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1163 panic("receive 3"); 1164 #endif 1165 so->so_state &= ~SS_RCVATMARK; 1166 len = uio->uio_resid; 1167 if (so->so_oobmark && len > so->so_oobmark - offset) 1168 len = so->so_oobmark - offset; 1169 if (len > m->m_len - moff) 1170 len = m->m_len - moff; 1171 /* 1172 * If mp is set, just pass back the mbufs. 1173 * Otherwise copy them out via the uio, then free. 1174 * Sockbuf must be consistent here (points to current mbuf, 1175 * it points to next record) when we drop priority; 1176 * we must note any additions to the sockbuf when we 1177 * block interrupts again. 1178 */ 1179 if (mp == 0) { 1180 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1181 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1182 splx(s); 1183 error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio); 1184 s = splsoftnet(); 1185 if (error) { 1186 /* 1187 * If any part of the record has been removed 1188 * (such as the MT_SONAME mbuf, which will 1189 * happen when PR_ADDR, and thus also 1190 * PR_ATOMIC, is set), then drop the entire 1191 * record to maintain the atomicity of the 1192 * receive operation. 1193 * 1194 * This avoids a later panic("receive 1a") 1195 * when compiled with DIAGNOSTIC. 1196 */ 1197 if (m && mbuf_removed 1198 && (pr->pr_flags & PR_ATOMIC)) 1199 (void) sbdroprecord(&so->so_rcv); 1200 1201 goto release; 1202 } 1203 } else 1204 uio->uio_resid -= len; 1205 if (len == m->m_len - moff) { 1206 if (m->m_flags & M_EOR) 1207 flags |= MSG_EOR; 1208 if (flags & MSG_PEEK) { 1209 m = m->m_next; 1210 moff = 0; 1211 } else { 1212 nextrecord = m->m_nextpkt; 1213 sbfree(&so->so_rcv, m); 1214 if (mp) { 1215 *mp = m; 1216 mp = &m->m_next; 1217 so->so_rcv.sb_mb = m = m->m_next; 1218 *mp = (struct mbuf *)0; 1219 } else { 1220 MFREE(m, so->so_rcv.sb_mb); 1221 m = so->so_rcv.sb_mb; 1222 } 1223 /* 1224 * If m != NULL, we also know that 1225 * so->so_rcv.sb_mb != NULL. 1226 */ 1227 KASSERT(so->so_rcv.sb_mb == m); 1228 if (m) { 1229 m->m_nextpkt = nextrecord; 1230 if (nextrecord == NULL) 1231 so->so_rcv.sb_lastrecord = m; 1232 } else { 1233 so->so_rcv.sb_mb = nextrecord; 1234 SB_EMPTY_FIXUP(&so->so_rcv); 1235 } 1236 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1237 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1238 } 1239 } else { 1240 if (flags & MSG_PEEK) 1241 moff += len; 1242 else { 1243 if (mp) 1244 *mp = m_copym(m, 0, len, M_WAIT); 1245 m->m_data += len; 1246 m->m_len -= len; 1247 so->so_rcv.sb_cc -= len; 1248 } 1249 } 1250 if (so->so_oobmark) { 1251 if ((flags & MSG_PEEK) == 0) { 1252 so->so_oobmark -= len; 1253 if (so->so_oobmark == 0) { 1254 so->so_state |= SS_RCVATMARK; 1255 break; 1256 } 1257 } else { 1258 offset += len; 1259 if (offset == so->so_oobmark) 1260 break; 1261 } 1262 } 1263 if (flags & MSG_EOR) 1264 break; 1265 /* 1266 * If the MSG_WAITALL flag is set (for non-atomic socket), 1267 * we must not quit until "uio->uio_resid == 0" or an error 1268 * termination. If a signal/timeout occurs, return 1269 * with a short count but without error. 1270 * Keep sockbuf locked against other readers. 1271 */ 1272 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1273 !sosendallatonce(so) && !nextrecord) { 1274 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1275 break; 1276 /* 1277 * If we are peeking and the socket receive buffer is 1278 * full, stop since we can't get more data to peek at. 1279 */ 1280 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1281 break; 1282 /* 1283 * If we've drained the socket buffer, tell the 1284 * protocol in case it needs to do something to 1285 * get it filled again. 1286 */ 1287 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1288 (*pr->pr_usrreq)(so, PRU_RCVD, 1289 (struct mbuf *)0, 1290 (struct mbuf *)(long)flags, 1291 (struct mbuf *)0, l); 1292 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1293 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1294 error = sbwait(&so->so_rcv); 1295 if (error) { 1296 sbunlock(&so->so_rcv); 1297 splx(s); 1298 return (0); 1299 } 1300 if ((m = so->so_rcv.sb_mb) != NULL) 1301 nextrecord = m->m_nextpkt; 1302 } 1303 } 1304 1305 if (m && pr->pr_flags & PR_ATOMIC) { 1306 flags |= MSG_TRUNC; 1307 if ((flags & MSG_PEEK) == 0) 1308 (void) sbdroprecord(&so->so_rcv); 1309 } 1310 if ((flags & MSG_PEEK) == 0) { 1311 if (m == 0) { 1312 /* 1313 * First part is an inline SB_EMPTY_FIXUP(). Second 1314 * part makes sure sb_lastrecord is up-to-date if 1315 * there is still data in the socket buffer. 1316 */ 1317 so->so_rcv.sb_mb = nextrecord; 1318 if (so->so_rcv.sb_mb == NULL) { 1319 so->so_rcv.sb_mbtail = NULL; 1320 so->so_rcv.sb_lastrecord = NULL; 1321 } else if (nextrecord->m_nextpkt == NULL) 1322 so->so_rcv.sb_lastrecord = nextrecord; 1323 } 1324 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1325 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1326 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1327 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1328 (struct mbuf *)(long)flags, (struct mbuf *)0, l); 1329 } 1330 if (orig_resid == uio->uio_resid && orig_resid && 1331 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1332 sbunlock(&so->so_rcv); 1333 splx(s); 1334 goto restart; 1335 } 1336 1337 if (flagsp) 1338 *flagsp |= flags; 1339 release: 1340 sbunlock(&so->so_rcv); 1341 splx(s); 1342 return (error); 1343 } 1344 1345 int 1346 soshutdown(struct socket *so, int how) 1347 { 1348 const struct protosw *pr; 1349 1350 pr = so->so_proto; 1351 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1352 return (EINVAL); 1353 1354 if (how == SHUT_RD || how == SHUT_RDWR) 1355 sorflush(so); 1356 if (how == SHUT_WR || how == SHUT_RDWR) 1357 return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0, 1358 (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0); 1359 return (0); 1360 } 1361 1362 void 1363 sorflush(struct socket *so) 1364 { 1365 struct sockbuf *sb, asb; 1366 const struct protosw *pr; 1367 int s; 1368 1369 sb = &so->so_rcv; 1370 pr = so->so_proto; 1371 sb->sb_flags |= SB_NOINTR; 1372 (void) sblock(sb, M_WAITOK); 1373 s = splnet(); 1374 socantrcvmore(so); 1375 sbunlock(sb); 1376 asb = *sb; 1377 /* 1378 * Clear most of the sockbuf structure, but leave some of the 1379 * fields valid. 1380 */ 1381 memset(&sb->sb_startzero, 0, 1382 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1383 splx(s); 1384 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1385 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1386 sbrelease(&asb, so); 1387 } 1388 1389 int 1390 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0) 1391 { 1392 int error; 1393 struct mbuf *m; 1394 1395 error = 0; 1396 m = m0; 1397 if (level != SOL_SOCKET) { 1398 if (so->so_proto && so->so_proto->pr_ctloutput) 1399 return ((*so->so_proto->pr_ctloutput) 1400 (PRCO_SETOPT, so, level, optname, &m0)); 1401 error = ENOPROTOOPT; 1402 } else { 1403 switch (optname) { 1404 1405 case SO_LINGER: 1406 if (m == NULL || m->m_len != sizeof(struct linger)) { 1407 error = EINVAL; 1408 goto bad; 1409 } 1410 if (mtod(m, struct linger *)->l_linger < 0 || 1411 mtod(m, struct linger *)->l_linger > (INT_MAX / hz)) { 1412 error = EDOM; 1413 goto bad; 1414 } 1415 so->so_linger = mtod(m, struct linger *)->l_linger; 1416 /* fall thru... */ 1417 1418 case SO_DEBUG: 1419 case SO_KEEPALIVE: 1420 case SO_DONTROUTE: 1421 case SO_USELOOPBACK: 1422 case SO_BROADCAST: 1423 case SO_REUSEADDR: 1424 case SO_REUSEPORT: 1425 case SO_OOBINLINE: 1426 case SO_TIMESTAMP: 1427 if (m == NULL || m->m_len < sizeof(int)) { 1428 error = EINVAL; 1429 goto bad; 1430 } 1431 if (*mtod(m, int *)) 1432 so->so_options |= optname; 1433 else 1434 so->so_options &= ~optname; 1435 break; 1436 1437 case SO_SNDBUF: 1438 case SO_RCVBUF: 1439 case SO_SNDLOWAT: 1440 case SO_RCVLOWAT: 1441 { 1442 int optval; 1443 1444 if (m == NULL || m->m_len < sizeof(int)) { 1445 error = EINVAL; 1446 goto bad; 1447 } 1448 1449 /* 1450 * Values < 1 make no sense for any of these 1451 * options, so disallow them. 1452 */ 1453 optval = *mtod(m, int *); 1454 if (optval < 1) { 1455 error = EINVAL; 1456 goto bad; 1457 } 1458 1459 switch (optname) { 1460 1461 case SO_SNDBUF: 1462 case SO_RCVBUF: 1463 if (sbreserve(optname == SO_SNDBUF ? 1464 &so->so_snd : &so->so_rcv, 1465 (u_long) optval, so) == 0) { 1466 error = ENOBUFS; 1467 goto bad; 1468 } 1469 break; 1470 1471 /* 1472 * Make sure the low-water is never greater than 1473 * the high-water. 1474 */ 1475 case SO_SNDLOWAT: 1476 so->so_snd.sb_lowat = 1477 (optval > so->so_snd.sb_hiwat) ? 1478 so->so_snd.sb_hiwat : optval; 1479 break; 1480 case SO_RCVLOWAT: 1481 so->so_rcv.sb_lowat = 1482 (optval > so->so_rcv.sb_hiwat) ? 1483 so->so_rcv.sb_hiwat : optval; 1484 break; 1485 } 1486 break; 1487 } 1488 1489 case SO_SNDTIMEO: 1490 case SO_RCVTIMEO: 1491 { 1492 struct timeval *tv; 1493 int val; 1494 1495 if (m == NULL || m->m_len < sizeof(*tv)) { 1496 error = EINVAL; 1497 goto bad; 1498 } 1499 tv = mtod(m, struct timeval *); 1500 if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz) { 1501 error = EDOM; 1502 goto bad; 1503 } 1504 val = tv->tv_sec * hz + tv->tv_usec / tick; 1505 if (val == 0 && tv->tv_usec != 0) 1506 val = 1; 1507 1508 switch (optname) { 1509 1510 case SO_SNDTIMEO: 1511 so->so_snd.sb_timeo = val; 1512 break; 1513 case SO_RCVTIMEO: 1514 so->so_rcv.sb_timeo = val; 1515 break; 1516 } 1517 break; 1518 } 1519 1520 default: 1521 error = ENOPROTOOPT; 1522 break; 1523 } 1524 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1525 (void) ((*so->so_proto->pr_ctloutput) 1526 (PRCO_SETOPT, so, level, optname, &m0)); 1527 m = NULL; /* freed by protocol */ 1528 } 1529 } 1530 bad: 1531 if (m) 1532 (void) m_free(m); 1533 return (error); 1534 } 1535 1536 int 1537 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp) 1538 { 1539 struct mbuf *m; 1540 1541 if (level != SOL_SOCKET) { 1542 if (so->so_proto && so->so_proto->pr_ctloutput) { 1543 return ((*so->so_proto->pr_ctloutput) 1544 (PRCO_GETOPT, so, level, optname, mp)); 1545 } else 1546 return (ENOPROTOOPT); 1547 } else { 1548 m = m_get(M_WAIT, MT_SOOPTS); 1549 m->m_len = sizeof(int); 1550 1551 switch (optname) { 1552 1553 case SO_LINGER: 1554 m->m_len = sizeof(struct linger); 1555 mtod(m, struct linger *)->l_onoff = 1556 so->so_options & SO_LINGER; 1557 mtod(m, struct linger *)->l_linger = so->so_linger; 1558 break; 1559 1560 case SO_USELOOPBACK: 1561 case SO_DONTROUTE: 1562 case SO_DEBUG: 1563 case SO_KEEPALIVE: 1564 case SO_REUSEADDR: 1565 case SO_REUSEPORT: 1566 case SO_BROADCAST: 1567 case SO_OOBINLINE: 1568 case SO_TIMESTAMP: 1569 *mtod(m, int *) = so->so_options & optname; 1570 break; 1571 1572 case SO_TYPE: 1573 *mtod(m, int *) = so->so_type; 1574 break; 1575 1576 case SO_ERROR: 1577 *mtod(m, int *) = so->so_error; 1578 so->so_error = 0; 1579 break; 1580 1581 case SO_SNDBUF: 1582 *mtod(m, int *) = so->so_snd.sb_hiwat; 1583 break; 1584 1585 case SO_RCVBUF: 1586 *mtod(m, int *) = so->so_rcv.sb_hiwat; 1587 break; 1588 1589 case SO_SNDLOWAT: 1590 *mtod(m, int *) = so->so_snd.sb_lowat; 1591 break; 1592 1593 case SO_RCVLOWAT: 1594 *mtod(m, int *) = so->so_rcv.sb_lowat; 1595 break; 1596 1597 case SO_SNDTIMEO: 1598 case SO_RCVTIMEO: 1599 { 1600 int val = (optname == SO_SNDTIMEO ? 1601 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1602 1603 m->m_len = sizeof(struct timeval); 1604 mtod(m, struct timeval *)->tv_sec = val / hz; 1605 mtod(m, struct timeval *)->tv_usec = 1606 (val % hz) * tick; 1607 break; 1608 } 1609 1610 case SO_OVERFLOWED: 1611 *mtod(m, int *) = so->so_rcv.sb_overflowed; 1612 break; 1613 1614 default: 1615 (void)m_free(m); 1616 return (ENOPROTOOPT); 1617 } 1618 *mp = m; 1619 return (0); 1620 } 1621 } 1622 1623 void 1624 sohasoutofband(struct socket *so) 1625 { 1626 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 1627 selwakeup(&so->so_rcv.sb_sel); 1628 } 1629 1630 static void 1631 filt_sordetach(struct knote *kn) 1632 { 1633 struct socket *so; 1634 1635 so = (struct socket *)kn->kn_fp->f_data; 1636 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 1637 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 1638 so->so_rcv.sb_flags &= ~SB_KNOTE; 1639 } 1640 1641 /*ARGSUSED*/ 1642 static int 1643 filt_soread(struct knote *kn, long hint) 1644 { 1645 struct socket *so; 1646 1647 so = (struct socket *)kn->kn_fp->f_data; 1648 kn->kn_data = so->so_rcv.sb_cc; 1649 if (so->so_state & SS_CANTRCVMORE) { 1650 kn->kn_flags |= EV_EOF; 1651 kn->kn_fflags = so->so_error; 1652 return (1); 1653 } 1654 if (so->so_error) /* temporary udp error */ 1655 return (1); 1656 if (kn->kn_sfflags & NOTE_LOWAT) 1657 return (kn->kn_data >= kn->kn_sdata); 1658 return (kn->kn_data >= so->so_rcv.sb_lowat); 1659 } 1660 1661 static void 1662 filt_sowdetach(struct knote *kn) 1663 { 1664 struct socket *so; 1665 1666 so = (struct socket *)kn->kn_fp->f_data; 1667 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 1668 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 1669 so->so_snd.sb_flags &= ~SB_KNOTE; 1670 } 1671 1672 /*ARGSUSED*/ 1673 static int 1674 filt_sowrite(struct knote *kn, long hint) 1675 { 1676 struct socket *so; 1677 1678 so = (struct socket *)kn->kn_fp->f_data; 1679 kn->kn_data = sbspace(&so->so_snd); 1680 if (so->so_state & SS_CANTSENDMORE) { 1681 kn->kn_flags |= EV_EOF; 1682 kn->kn_fflags = so->so_error; 1683 return (1); 1684 } 1685 if (so->so_error) /* temporary udp error */ 1686 return (1); 1687 if (((so->so_state & SS_ISCONNECTED) == 0) && 1688 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1689 return (0); 1690 if (kn->kn_sfflags & NOTE_LOWAT) 1691 return (kn->kn_data >= kn->kn_sdata); 1692 return (kn->kn_data >= so->so_snd.sb_lowat); 1693 } 1694 1695 /*ARGSUSED*/ 1696 static int 1697 filt_solisten(struct knote *kn, long hint) 1698 { 1699 struct socket *so; 1700 1701 so = (struct socket *)kn->kn_fp->f_data; 1702 1703 /* 1704 * Set kn_data to number of incoming connections, not 1705 * counting partial (incomplete) connections. 1706 */ 1707 kn->kn_data = so->so_qlen; 1708 return (kn->kn_data > 0); 1709 } 1710 1711 static const struct filterops solisten_filtops = 1712 { 1, NULL, filt_sordetach, filt_solisten }; 1713 static const struct filterops soread_filtops = 1714 { 1, NULL, filt_sordetach, filt_soread }; 1715 static const struct filterops sowrite_filtops = 1716 { 1, NULL, filt_sowdetach, filt_sowrite }; 1717 1718 int 1719 soo_kqfilter(struct file *fp, struct knote *kn) 1720 { 1721 struct socket *so; 1722 struct sockbuf *sb; 1723 1724 so = (struct socket *)kn->kn_fp->f_data; 1725 switch (kn->kn_filter) { 1726 case EVFILT_READ: 1727 if (so->so_options & SO_ACCEPTCONN) 1728 kn->kn_fop = &solisten_filtops; 1729 else 1730 kn->kn_fop = &soread_filtops; 1731 sb = &so->so_rcv; 1732 break; 1733 case EVFILT_WRITE: 1734 kn->kn_fop = &sowrite_filtops; 1735 sb = &so->so_snd; 1736 break; 1737 default: 1738 return (1); 1739 } 1740 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 1741 sb->sb_flags |= SB_KNOTE; 1742 return (0); 1743 } 1744 1745 #include <sys/sysctl.h> 1746 1747 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 1748 1749 /* 1750 * sysctl helper routine for kern.somaxkva. ensures that the given 1751 * value is not too small. 1752 * (XXX should we maybe make sure it's not too large as well?) 1753 */ 1754 static int 1755 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 1756 { 1757 int error, new_somaxkva; 1758 struct sysctlnode node; 1759 int s; 1760 1761 new_somaxkva = somaxkva; 1762 node = *rnode; 1763 node.sysctl_data = &new_somaxkva; 1764 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1765 if (error || newp == NULL) 1766 return (error); 1767 1768 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 1769 return (EINVAL); 1770 1771 s = splvm(); 1772 simple_lock(&so_pendfree_slock); 1773 somaxkva = new_somaxkva; 1774 wakeup(&socurkva); 1775 simple_unlock(&so_pendfree_slock); 1776 splx(s); 1777 1778 return (error); 1779 } 1780 1781 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup") 1782 { 1783 1784 sysctl_createv(clog, 0, NULL, NULL, 1785 CTLFLAG_PERMANENT, 1786 CTLTYPE_NODE, "kern", NULL, 1787 NULL, 0, NULL, 0, 1788 CTL_KERN, CTL_EOL); 1789 1790 sysctl_createv(clog, 0, NULL, NULL, 1791 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1792 CTLTYPE_INT, "somaxkva", 1793 SYSCTL_DESCR("Maximum amount of kernel memory to be " 1794 "used for socket buffers"), 1795 sysctl_kern_somaxkva, 0, NULL, 0, 1796 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 1797 } 1798