uipc_socket.c revision 1.111.2.1
1 /* $NetBSD: uipc_socket.c,v 1.111.2.1 2005/07/07 12:07:38 yamt 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 1.111.2.1 2005/07/07 12:07:38 yamt Exp $"); 72 73 #include "opt_sock_counters.h" 74 #include "opt_sosend_loan.h" 75 #include "opt_mbuftrace.h" 76 #include "opt_somaxkva.h" 77 78 #include <sys/param.h> 79 #include <sys/systm.h> 80 #include <sys/proc.h> 81 #include <sys/file.h> 82 #include <sys/malloc.h> 83 #include <sys/mbuf.h> 84 #include <sys/domain.h> 85 #include <sys/kernel.h> 86 #include <sys/protosw.h> 87 #include <sys/socket.h> 88 #include <sys/socketvar.h> 89 #include <sys/signalvar.h> 90 #include <sys/resourcevar.h> 91 #include <sys/pool.h> 92 #include <sys/event.h> 93 #include <sys/poll.h> 94 95 #include <uvm/uvm.h> 96 97 POOL_INIT(socket_pool, sizeof(struct socket), 0, 0, 0, "sockpl", NULL); 98 99 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options"); 100 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 101 102 extern int somaxconn; /* patchable (XXX sysctl) */ 103 int somaxconn = SOMAXCONN; 104 105 #ifdef SOSEND_COUNTERS 106 #include <sys/device.h> 107 108 struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 109 NULL, "sosend", "loan big"); 110 struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 111 NULL, "sosend", "copy big"); 112 struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 113 NULL, "sosend", "copy small"); 114 struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 115 NULL, "sosend", "kva limit"); 116 117 #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++ 118 119 EVCNT_ATTACH_STATIC(sosend_loan_big); 120 EVCNT_ATTACH_STATIC(sosend_copy_big); 121 EVCNT_ATTACH_STATIC(sosend_copy_small); 122 EVCNT_ATTACH_STATIC(sosend_kvalimit); 123 #else 124 125 #define SOSEND_COUNTER_INCR(ev) /* nothing */ 126 127 #endif /* SOSEND_COUNTERS */ 128 129 void 130 soinit(void) 131 { 132 133 /* Set the initial adjusted socket buffer size. */ 134 if (sb_max_set(sb_max)) 135 panic("bad initial sb_max value: %lu", sb_max); 136 137 } 138 139 #ifdef SOSEND_NO_LOAN 140 int use_sosend_loan = 0; 141 #else 142 int use_sosend_loan = 1; 143 #endif 144 145 struct simplelock so_pendfree_slock = SIMPLELOCK_INITIALIZER; 146 struct mbuf *so_pendfree; 147 148 #ifndef SOMAXKVA 149 #define SOMAXKVA (16 * 1024 * 1024) 150 #endif 151 int somaxkva = SOMAXKVA; 152 int socurkva; 153 int sokvawaiters; 154 155 #define SOCK_LOAN_THRESH 4096 156 #define SOCK_LOAN_CHUNK 65536 157 158 static size_t sodopendfree(struct socket *); 159 static size_t sodopendfreel(struct socket *); 160 static __inline vsize_t sokvareserve(struct socket *, vsize_t); 161 static __inline void sokvaunreserve(vsize_t); 162 163 static __inline vsize_t 164 sokvareserve(struct socket *so, vsize_t len) 165 { 166 int s; 167 int error; 168 169 s = splvm(); 170 simple_lock(&so_pendfree_slock); 171 while (socurkva + len > somaxkva) { 172 size_t freed; 173 174 /* 175 * try to do pendfree. 176 */ 177 178 freed = sodopendfreel(so); 179 180 /* 181 * if some kva was freed, try again. 182 */ 183 184 if (freed) 185 continue; 186 187 SOSEND_COUNTER_INCR(&sosend_kvalimit); 188 sokvawaiters++; 189 error = ltsleep(&socurkva, PVM | PCATCH, "sokva", 0, 190 &so_pendfree_slock); 191 sokvawaiters--; 192 if (error) { 193 len = 0; 194 break; 195 } 196 } 197 socurkva += len; 198 simple_unlock(&so_pendfree_slock); 199 splx(s); 200 return len; 201 } 202 203 static __inline void 204 sokvaunreserve(vsize_t len) 205 { 206 int s; 207 208 s = splvm(); 209 simple_lock(&so_pendfree_slock); 210 socurkva -= len; 211 if (sokvawaiters) 212 wakeup(&socurkva); 213 simple_unlock(&so_pendfree_slock); 214 splx(s); 215 } 216 217 /* 218 * sokvaalloc: allocate kva for loan. 219 */ 220 221 vaddr_t 222 sokvaalloc(vsize_t len, struct socket *so) 223 { 224 vaddr_t lva; 225 226 /* 227 * reserve kva. 228 */ 229 230 if (sokvareserve(so, len) == 0) 231 return 0; 232 233 /* 234 * allocate kva. 235 */ 236 237 lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA); 238 if (lva == 0) { 239 sokvaunreserve(len); 240 return (0); 241 } 242 243 return lva; 244 } 245 246 /* 247 * sokvafree: free kva for loan. 248 */ 249 250 void 251 sokvafree(vaddr_t sva, vsize_t len) 252 { 253 254 /* 255 * free kva. 256 */ 257 258 uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY); 259 260 /* 261 * unreserve kva. 262 */ 263 264 sokvaunreserve(len); 265 } 266 267 static void 268 sodoloanfree(struct vm_page **pgs, caddr_t buf, size_t size, boolean_t mapped) 269 { 270 vaddr_t sva, eva; 271 vsize_t len; 272 int npgs; 273 274 KASSERT(pgs != NULL); 275 276 eva = round_page((vaddr_t) buf + size); 277 sva = trunc_page((vaddr_t) buf); 278 len = eva - sva; 279 npgs = len >> PAGE_SHIFT; 280 281 if (mapped) { 282 pmap_kremove(sva, len); 283 pmap_update(pmap_kernel()); 284 } 285 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 286 sokvafree(sva, len); 287 } 288 289 static size_t 290 sodopendfree(struct socket *so) 291 { 292 int s; 293 size_t rv; 294 295 s = splvm(); 296 simple_lock(&so_pendfree_slock); 297 rv = sodopendfreel(so); 298 simple_unlock(&so_pendfree_slock); 299 splx(s); 300 301 return rv; 302 } 303 304 /* 305 * sodopendfreel: free mbufs on "pendfree" list. 306 * unlock and relock so_pendfree_slock when freeing mbufs. 307 * 308 * => called with so_pendfree_slock held. 309 * => called at splvm. 310 */ 311 312 static size_t 313 sodopendfreel(struct socket *so) 314 { 315 size_t rv = 0; 316 317 LOCK_ASSERT(simple_lock_held(&so_pendfree_slock)); 318 319 for (;;) { 320 struct mbuf *m; 321 struct mbuf *next; 322 323 m = so_pendfree; 324 if (m == NULL) 325 break; 326 so_pendfree = NULL; 327 simple_unlock(&so_pendfree_slock); 328 /* XXX splx */ 329 330 for (; m != NULL; m = next) { 331 next = m->m_next; 332 KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0); 333 KASSERT(!MCLISREFERENCED(m)); 334 335 rv += m->m_ext.ext_size; 336 sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf, 337 m->m_ext.ext_size, 338 (m->m_ext.ext_flags & M_EXT_LAZY) == 0); 339 pool_cache_put(&mbpool_cache, m); 340 } 341 342 /* XXX splvm */ 343 simple_lock(&so_pendfree_slock); 344 } 345 346 return (rv); 347 } 348 349 void 350 soloanfree(struct mbuf *m, caddr_t buf, size_t size, void *arg) 351 { 352 int s; 353 354 KASSERT(m != NULL); 355 356 /* 357 * postpone freeing mbuf. 358 * 359 * we can't do it in interrupt context 360 * because we need to put kva back to kernel_map. 361 */ 362 363 s = splvm(); 364 simple_lock(&so_pendfree_slock); 365 m->m_next = so_pendfree; 366 so_pendfree = m; 367 if (sokvawaiters) 368 wakeup(&socurkva); 369 simple_unlock(&so_pendfree_slock); 370 splx(s); 371 } 372 373 static long 374 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 375 { 376 struct iovec *iov = uio->uio_iov; 377 vaddr_t sva, eva; 378 vsize_t len; 379 vaddr_t lva; 380 int npgs, error; 381 #if 0 382 vaddr_t va; 383 int i; 384 #endif 385 386 if (uio->uio_segflg != UIO_USERSPACE) 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 KASSERT(uio->uio_procp != NULL); 402 403 lva = sokvaalloc(len, so); 404 if (lva == 0) 405 return 0; 406 407 error = uvm_loan(&uio->uio_procp->p_vmspace->vm_map, sva, len, 408 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 409 if (error) { 410 sokvafree(lva, len); 411 return (0); 412 } 413 414 #if 0 415 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 416 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 417 VM_PROT_READ); 418 pmap_update(pmap_kernel()); 419 #endif 420 421 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 422 423 MEXTADD(m, (caddr_t) lva, space, M_MBUF, soloanfree, so); 424 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 425 426 m->m_flags |= M_EXT_LAZY; 427 m->m_ext.ext_flags |= M_EXT_LAZY; 428 429 uio->uio_resid -= space; 430 /* uio_offset not updated, not set/used for write(2) */ 431 uio->uio_iov->iov_base = (caddr_t) uio->uio_iov->iov_base + space; 432 uio->uio_iov->iov_len -= space; 433 if (uio->uio_iov->iov_len == 0) { 434 uio->uio_iov++; 435 uio->uio_iovcnt--; 436 } 437 438 return (space); 439 } 440 441 /* 442 * Socket operation routines. 443 * These routines are called by the routines in 444 * sys_socket.c or from a system process, and 445 * implement the semantics of socket operations by 446 * switching out to the protocol specific routines. 447 */ 448 /*ARGSUSED*/ 449 int 450 socreate(int dom, struct socket **aso, int type, int proto, struct proc *p) 451 { 452 const struct protosw *prp; 453 struct socket *so; 454 int error, s; 455 456 if (proto) 457 prp = pffindproto(dom, proto, type); 458 else 459 prp = pffindtype(dom, type); 460 if (prp == 0 || prp->pr_usrreq == 0) 461 return (EPROTONOSUPPORT); 462 if (prp->pr_type != type) 463 return (EPROTOTYPE); 464 s = splsoftnet(); 465 so = pool_get(&socket_pool, PR_WAITOK); 466 memset((caddr_t)so, 0, sizeof(*so)); 467 TAILQ_INIT(&so->so_q0); 468 TAILQ_INIT(&so->so_q); 469 so->so_type = type; 470 so->so_proto = prp; 471 so->so_send = sosend; 472 so->so_receive = soreceive; 473 #ifdef MBUFTRACE 474 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 475 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 476 so->so_mowner = &prp->pr_domain->dom_mowner; 477 #endif 478 if (p != 0) 479 so->so_uidinfo = uid_find(p->p_ucred->cr_uid); 480 else 481 so->so_uidinfo = uid_find(0); 482 error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0, 483 (struct mbuf *)(long)proto, (struct mbuf *)0, p); 484 if (error) { 485 so->so_state |= SS_NOFDREF; 486 sofree(so); 487 splx(s); 488 return (error); 489 } 490 splx(s); 491 *aso = so; 492 return (0); 493 } 494 495 int 496 sobind(struct socket *so, struct mbuf *nam, struct proc *p) 497 { 498 int s, error; 499 500 s = splsoftnet(); 501 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0, 502 nam, (struct mbuf *)0, p); 503 splx(s); 504 return (error); 505 } 506 507 int 508 solisten(struct socket *so, int backlog) 509 { 510 int s, error; 511 512 s = splsoftnet(); 513 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0, 514 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 515 if (error) { 516 splx(s); 517 return (error); 518 } 519 if (TAILQ_EMPTY(&so->so_q)) 520 so->so_options |= SO_ACCEPTCONN; 521 if (backlog < 0) 522 backlog = 0; 523 so->so_qlimit = min(backlog, somaxconn); 524 splx(s); 525 return (0); 526 } 527 528 void 529 sofree(struct socket *so) 530 { 531 532 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) 533 return; 534 if (so->so_head) { 535 /* 536 * We must not decommission a socket that's on the accept(2) 537 * queue. If we do, then accept(2) may hang after select(2) 538 * indicated that the listening socket was ready. 539 */ 540 if (!soqremque(so, 0)) 541 return; 542 } 543 if (so->so_rcv.sb_hiwat) 544 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 545 RLIM_INFINITY); 546 if (so->so_snd.sb_hiwat) 547 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 548 RLIM_INFINITY); 549 sbrelease(&so->so_snd, so); 550 sorflush(so); 551 pool_put(&socket_pool, so); 552 } 553 554 /* 555 * Close a socket on last file table reference removal. 556 * Initiate disconnect if connected. 557 * Free socket when disconnect complete. 558 */ 559 int 560 soclose(struct socket *so) 561 { 562 struct socket *so2; 563 int s, error; 564 565 error = 0; 566 s = splsoftnet(); /* conservative */ 567 if (so->so_options & SO_ACCEPTCONN) { 568 while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 569 (void) soqremque(so2, 0); 570 (void) soabort(so2); 571 } 572 while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 573 (void) soqremque(so2, 1); 574 (void) soabort(so2); 575 } 576 } 577 if (so->so_pcb == 0) 578 goto discard; 579 if (so->so_state & SS_ISCONNECTED) { 580 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 581 error = sodisconnect(so); 582 if (error) 583 goto drop; 584 } 585 if (so->so_options & SO_LINGER) { 586 if ((so->so_state & SS_ISDISCONNECTING) && 587 (so->so_state & SS_NBIO)) 588 goto drop; 589 while (so->so_state & SS_ISCONNECTED) { 590 error = tsleep((caddr_t)&so->so_timeo, 591 PSOCK | PCATCH, netcls, 592 so->so_linger * hz); 593 if (error) 594 break; 595 } 596 } 597 } 598 drop: 599 if (so->so_pcb) { 600 int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 601 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 602 (struct proc *)0); 603 if (error == 0) 604 error = error2; 605 } 606 discard: 607 if (so->so_state & SS_NOFDREF) 608 panic("soclose: NOFDREF"); 609 so->so_state |= SS_NOFDREF; 610 sofree(so); 611 splx(s); 612 return (error); 613 } 614 615 /* 616 * Must be called at splsoftnet... 617 */ 618 int 619 soabort(struct socket *so) 620 { 621 622 return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0, 623 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 624 } 625 626 int 627 soaccept(struct socket *so, struct mbuf *nam) 628 { 629 int s, error; 630 631 error = 0; 632 s = splsoftnet(); 633 if ((so->so_state & SS_NOFDREF) == 0) 634 panic("soaccept: !NOFDREF"); 635 so->so_state &= ~SS_NOFDREF; 636 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 637 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 638 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 639 (struct mbuf *)0, nam, (struct mbuf *)0, (struct proc *)0); 640 else 641 error = ECONNABORTED; 642 643 splx(s); 644 return (error); 645 } 646 647 int 648 soconnect(struct socket *so, struct mbuf *nam, struct proc *p) 649 { 650 int s, error; 651 652 if (so->so_options & SO_ACCEPTCONN) 653 return (EOPNOTSUPP); 654 s = splsoftnet(); 655 /* 656 * If protocol is connection-based, can only connect once. 657 * Otherwise, if connected, try to disconnect first. 658 * This allows user to disconnect by connecting to, e.g., 659 * a null address. 660 */ 661 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 662 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 663 (error = sodisconnect(so)))) 664 error = EISCONN; 665 else 666 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 667 (struct mbuf *)0, nam, (struct mbuf *)0, p); 668 splx(s); 669 return (error); 670 } 671 672 int 673 soconnect2(struct socket *so1, struct socket *so2) 674 { 675 int s, error; 676 677 s = splsoftnet(); 678 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 679 (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0, 680 (struct proc *)0); 681 splx(s); 682 return (error); 683 } 684 685 int 686 sodisconnect(struct socket *so) 687 { 688 int s, error; 689 690 s = splsoftnet(); 691 if ((so->so_state & SS_ISCONNECTED) == 0) { 692 error = ENOTCONN; 693 goto bad; 694 } 695 if (so->so_state & SS_ISDISCONNECTING) { 696 error = EALREADY; 697 goto bad; 698 } 699 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 700 (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0, 701 (struct proc *)0); 702 bad: 703 splx(s); 704 sodopendfree(so); 705 return (error); 706 } 707 708 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 709 /* 710 * Send on a socket. 711 * If send must go all at once and message is larger than 712 * send buffering, then hard error. 713 * Lock against other senders. 714 * If must go all at once and not enough room now, then 715 * inform user that this would block and do nothing. 716 * Otherwise, if nonblocking, send as much as possible. 717 * The data to be sent is described by "uio" if nonzero, 718 * otherwise by the mbuf chain "top" (which must be null 719 * if uio is not). Data provided in mbuf chain must be small 720 * enough to send all at once. 721 * 722 * Returns nonzero on error, timeout or signal; callers 723 * must check for short counts if EINTR/ERESTART are returned. 724 * Data and control buffers are freed on return. 725 */ 726 int 727 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 728 struct mbuf *control, int flags, struct proc *p) 729 { 730 struct mbuf **mp, *m; 731 long space, len, resid, clen, mlen; 732 int error, s, dontroute, atomic; 733 734 sodopendfree(so); 735 736 clen = 0; 737 atomic = sosendallatonce(so) || top; 738 if (uio) 739 resid = uio->uio_resid; 740 else 741 resid = top->m_pkthdr.len; 742 /* 743 * In theory resid should be unsigned. 744 * However, space must be signed, as it might be less than 0 745 * if we over-committed, and we must use a signed comparison 746 * of space and resid. On the other hand, a negative resid 747 * causes us to loop sending 0-length segments to the protocol. 748 */ 749 if (resid < 0) { 750 error = EINVAL; 751 goto out; 752 } 753 dontroute = 754 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 755 (so->so_proto->pr_flags & PR_ATOMIC); 756 if (p) 757 p->p_stats->p_ru.ru_msgsnd++; 758 if (control) 759 clen = control->m_len; 760 #define snderr(errno) { error = errno; splx(s); goto release; } 761 762 restart: 763 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 764 goto out; 765 do { 766 s = splsoftnet(); 767 if (so->so_state & SS_CANTSENDMORE) 768 snderr(EPIPE); 769 if (so->so_error) { 770 error = so->so_error; 771 so->so_error = 0; 772 splx(s); 773 goto release; 774 } 775 if ((so->so_state & SS_ISCONNECTED) == 0) { 776 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 777 if ((so->so_state & SS_ISCONFIRMING) == 0 && 778 !(resid == 0 && clen != 0)) 779 snderr(ENOTCONN); 780 } else if (addr == 0) 781 snderr(EDESTADDRREQ); 782 } 783 space = sbspace(&so->so_snd); 784 if (flags & MSG_OOB) 785 space += 1024; 786 if ((atomic && resid > so->so_snd.sb_hiwat) || 787 clen > so->so_snd.sb_hiwat) 788 snderr(EMSGSIZE); 789 if (space < resid + clen && 790 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 791 if (so->so_state & SS_NBIO) 792 snderr(EWOULDBLOCK); 793 sbunlock(&so->so_snd); 794 error = sbwait(&so->so_snd); 795 splx(s); 796 if (error) 797 goto out; 798 goto restart; 799 } 800 splx(s); 801 mp = ⊤ 802 space -= clen; 803 do { 804 if (uio == NULL) { 805 /* 806 * Data is prepackaged in "top". 807 */ 808 resid = 0; 809 if (flags & MSG_EOR) 810 top->m_flags |= M_EOR; 811 } else do { 812 if (top == 0) { 813 m = m_gethdr(M_WAIT, MT_DATA); 814 mlen = MHLEN; 815 m->m_pkthdr.len = 0; 816 m->m_pkthdr.rcvif = (struct ifnet *)0; 817 } else { 818 m = m_get(M_WAIT, MT_DATA); 819 mlen = MLEN; 820 } 821 MCLAIM(m, so->so_snd.sb_mowner); 822 if (use_sosend_loan && 823 uio->uio_iov->iov_len >= SOCK_LOAN_THRESH && 824 space >= SOCK_LOAN_THRESH && 825 (len = sosend_loan(so, uio, m, 826 space)) != 0) { 827 SOSEND_COUNTER_INCR(&sosend_loan_big); 828 space -= len; 829 goto have_data; 830 } 831 if (resid >= MINCLSIZE && space >= MCLBYTES) { 832 SOSEND_COUNTER_INCR(&sosend_copy_big); 833 m_clget(m, M_WAIT); 834 if ((m->m_flags & M_EXT) == 0) 835 goto nopages; 836 mlen = MCLBYTES; 837 if (atomic && top == 0) { 838 len = lmin(MCLBYTES - max_hdr, 839 resid); 840 m->m_data += max_hdr; 841 } else 842 len = lmin(MCLBYTES, resid); 843 space -= len; 844 } else { 845 nopages: 846 SOSEND_COUNTER_INCR(&sosend_copy_small); 847 len = lmin(lmin(mlen, resid), space); 848 space -= len; 849 /* 850 * For datagram protocols, leave room 851 * for protocol headers in first mbuf. 852 */ 853 if (atomic && top == 0 && len < mlen) 854 MH_ALIGN(m, len); 855 } 856 error = uiomove(mtod(m, caddr_t), (int)len, 857 uio); 858 have_data: 859 resid = uio->uio_resid; 860 m->m_len = len; 861 *mp = m; 862 top->m_pkthdr.len += len; 863 if (error) 864 goto release; 865 mp = &m->m_next; 866 if (resid <= 0) { 867 if (flags & MSG_EOR) 868 top->m_flags |= M_EOR; 869 break; 870 } 871 } while (space > 0 && atomic); 872 873 s = splsoftnet(); 874 875 if (so->so_state & SS_CANTSENDMORE) 876 snderr(EPIPE); 877 878 if (dontroute) 879 so->so_options |= SO_DONTROUTE; 880 if (resid > 0) 881 so->so_state |= SS_MORETOCOME; 882 error = (*so->so_proto->pr_usrreq)(so, 883 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 884 top, addr, control, p); 885 if (dontroute) 886 so->so_options &= ~SO_DONTROUTE; 887 if (resid > 0) 888 so->so_state &= ~SS_MORETOCOME; 889 splx(s); 890 891 clen = 0; 892 control = 0; 893 top = 0; 894 mp = ⊤ 895 if (error) 896 goto release; 897 } while (resid && space > 0); 898 } while (resid); 899 900 release: 901 sbunlock(&so->so_snd); 902 out: 903 if (top) 904 m_freem(top); 905 if (control) 906 m_freem(control); 907 return (error); 908 } 909 910 /* 911 * Implement receive operations on a socket. 912 * We depend on the way that records are added to the sockbuf 913 * by sbappend*. In particular, each record (mbufs linked through m_next) 914 * must begin with an address if the protocol so specifies, 915 * followed by an optional mbuf or mbufs containing ancillary data, 916 * and then zero or more mbufs of data. 917 * In order to avoid blocking network interrupts for the entire time here, 918 * we splx() while doing the actual copy to user space. 919 * Although the sockbuf is locked, new data may still be appended, 920 * and thus we must maintain consistency of the sockbuf during that time. 921 * 922 * The caller may receive the data as a single mbuf chain by supplying 923 * an mbuf **mp0 for use in returning the chain. The uio is then used 924 * only for the count in uio_resid. 925 */ 926 int 927 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 928 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 929 { 930 struct proc * p; 931 struct mbuf *m, **mp; 932 int flags, len, error, s, offset, moff, type, orig_resid; 933 const struct protosw *pr; 934 struct mbuf *nextrecord; 935 int mbuf_removed = 0; 936 937 pr = so->so_proto; 938 mp = mp0; 939 type = 0; 940 orig_resid = uio->uio_resid; 941 p = uio->uio_procp; 942 943 if (paddr) 944 *paddr = 0; 945 if (controlp) 946 *controlp = 0; 947 if (flagsp) 948 flags = *flagsp &~ MSG_EOR; 949 else 950 flags = 0; 951 952 if ((flags & MSG_DONTWAIT) == 0) 953 sodopendfree(so); 954 955 if (flags & MSG_OOB) { 956 m = m_get(M_WAIT, MT_DATA); 957 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 958 (struct mbuf *)(long)(flags & MSG_PEEK), 959 (struct mbuf *)0, p); 960 if (error) 961 goto bad; 962 do { 963 error = uiomove(mtod(m, caddr_t), 964 (int) min(uio->uio_resid, m->m_len), uio); 965 m = m_free(m); 966 } while (uio->uio_resid && error == 0 && m); 967 bad: 968 if (m) 969 m_freem(m); 970 return (error); 971 } 972 if (mp) 973 *mp = (struct mbuf *)0; 974 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 975 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 976 (struct mbuf *)0, (struct mbuf *)0, p); 977 978 restart: 979 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) 980 return (error); 981 s = splsoftnet(); 982 983 m = so->so_rcv.sb_mb; 984 /* 985 * If we have less data than requested, block awaiting more 986 * (subject to any timeout) if: 987 * 1. the current count is less than the low water mark, 988 * 2. MSG_WAITALL is set, and it is possible to do the entire 989 * receive operation at once if we block (resid <= hiwat), or 990 * 3. MSG_DONTWAIT is not set. 991 * If MSG_WAITALL is set but resid is larger than the receive buffer, 992 * we have to do the receive in sections, and thus risk returning 993 * a short count if a timeout or signal occurs after we start. 994 */ 995 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 && 996 so->so_rcv.sb_cc < uio->uio_resid) && 997 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 998 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 999 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { 1000 #ifdef DIAGNOSTIC 1001 if (m == 0 && so->so_rcv.sb_cc) 1002 panic("receive 1"); 1003 #endif 1004 if (so->so_error) { 1005 if (m) 1006 goto dontblock; 1007 error = so->so_error; 1008 if ((flags & MSG_PEEK) == 0) 1009 so->so_error = 0; 1010 goto release; 1011 } 1012 if (so->so_state & SS_CANTRCVMORE) { 1013 if (m) 1014 goto dontblock; 1015 else 1016 goto release; 1017 } 1018 for (; m; m = m->m_next) 1019 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1020 m = so->so_rcv.sb_mb; 1021 goto dontblock; 1022 } 1023 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1024 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1025 error = ENOTCONN; 1026 goto release; 1027 } 1028 if (uio->uio_resid == 0) 1029 goto release; 1030 if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) { 1031 error = EWOULDBLOCK; 1032 goto release; 1033 } 1034 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1035 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1036 sbunlock(&so->so_rcv); 1037 error = sbwait(&so->so_rcv); 1038 splx(s); 1039 if (error) 1040 return (error); 1041 goto restart; 1042 } 1043 dontblock: 1044 /* 1045 * On entry here, m points to the first record of the socket buffer. 1046 * While we process the initial mbufs containing address and control 1047 * info, we save a copy of m->m_nextpkt into nextrecord. 1048 */ 1049 if (p) 1050 p->p_stats->p_ru.ru_msgrcv++; 1051 KASSERT(m == so->so_rcv.sb_mb); 1052 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1053 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1054 nextrecord = m->m_nextpkt; 1055 if (pr->pr_flags & PR_ADDR) { 1056 #ifdef DIAGNOSTIC 1057 if (m->m_type != MT_SONAME) 1058 panic("receive 1a"); 1059 #endif 1060 orig_resid = 0; 1061 if (flags & MSG_PEEK) { 1062 if (paddr) 1063 *paddr = m_copy(m, 0, m->m_len); 1064 m = m->m_next; 1065 } else { 1066 sbfree(&so->so_rcv, m); 1067 mbuf_removed = 1; 1068 if (paddr) { 1069 *paddr = m; 1070 so->so_rcv.sb_mb = m->m_next; 1071 m->m_next = 0; 1072 m = so->so_rcv.sb_mb; 1073 } else { 1074 MFREE(m, so->so_rcv.sb_mb); 1075 m = so->so_rcv.sb_mb; 1076 } 1077 } 1078 } 1079 while (m && m->m_type == MT_CONTROL && error == 0) { 1080 if (flags & MSG_PEEK) { 1081 if (controlp) 1082 *controlp = m_copy(m, 0, m->m_len); 1083 m = m->m_next; 1084 } else { 1085 sbfree(&so->so_rcv, m); 1086 mbuf_removed = 1; 1087 if (controlp) { 1088 struct domain *dom = pr->pr_domain; 1089 if (dom->dom_externalize && p && 1090 mtod(m, struct cmsghdr *)->cmsg_type == 1091 SCM_RIGHTS) 1092 error = (*dom->dom_externalize)(m, p); 1093 *controlp = 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 /* 1099 * Dispose of any SCM_RIGHTS message that went 1100 * through the read path rather than recv. 1101 */ 1102 if (pr->pr_domain->dom_dispose && 1103 mtod(m, struct cmsghdr *)->cmsg_type == SCM_RIGHTS) 1104 (*pr->pr_domain->dom_dispose)(m); 1105 MFREE(m, so->so_rcv.sb_mb); 1106 m = so->so_rcv.sb_mb; 1107 } 1108 } 1109 if (controlp) { 1110 orig_resid = 0; 1111 controlp = &(*controlp)->m_next; 1112 } 1113 } 1114 1115 /* 1116 * If m is non-NULL, we have some data to read. From now on, 1117 * make sure to keep sb_lastrecord consistent when working on 1118 * the last packet on the chain (nextrecord == NULL) and we 1119 * change m->m_nextpkt. 1120 */ 1121 if (m) { 1122 if ((flags & MSG_PEEK) == 0) { 1123 m->m_nextpkt = nextrecord; 1124 /* 1125 * If nextrecord == NULL (this is a single chain), 1126 * then sb_lastrecord may not be valid here if m 1127 * was changed earlier. 1128 */ 1129 if (nextrecord == NULL) { 1130 KASSERT(so->so_rcv.sb_mb == m); 1131 so->so_rcv.sb_lastrecord = m; 1132 } 1133 } 1134 type = m->m_type; 1135 if (type == MT_OOBDATA) 1136 flags |= MSG_OOB; 1137 } else { 1138 if ((flags & MSG_PEEK) == 0) { 1139 KASSERT(so->so_rcv.sb_mb == m); 1140 so->so_rcv.sb_mb = nextrecord; 1141 SB_EMPTY_FIXUP(&so->so_rcv); 1142 } 1143 } 1144 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1145 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1146 1147 moff = 0; 1148 offset = 0; 1149 while (m && uio->uio_resid > 0 && error == 0) { 1150 if (m->m_type == MT_OOBDATA) { 1151 if (type != MT_OOBDATA) 1152 break; 1153 } else if (type == MT_OOBDATA) 1154 break; 1155 #ifdef DIAGNOSTIC 1156 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1157 panic("receive 3"); 1158 #endif 1159 so->so_state &= ~SS_RCVATMARK; 1160 len = uio->uio_resid; 1161 if (so->so_oobmark && len > so->so_oobmark - offset) 1162 len = so->so_oobmark - offset; 1163 if (len > m->m_len - moff) 1164 len = m->m_len - moff; 1165 /* 1166 * If mp is set, just pass back the mbufs. 1167 * Otherwise copy them out via the uio, then free. 1168 * Sockbuf must be consistent here (points to current mbuf, 1169 * it points to next record) when we drop priority; 1170 * we must note any additions to the sockbuf when we 1171 * block interrupts again. 1172 */ 1173 if (mp == 0) { 1174 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1175 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1176 splx(s); 1177 error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio); 1178 s = splsoftnet(); 1179 if (error) { 1180 /* 1181 * If any part of the record has been removed 1182 * (such as the MT_SONAME mbuf, which will 1183 * happen when PR_ADDR, and thus also 1184 * PR_ATOMIC, is set), then drop the entire 1185 * record to maintain the atomicity of the 1186 * receive operation. 1187 * 1188 * This avoids a later panic("receive 1a") 1189 * when compiled with DIAGNOSTIC. 1190 */ 1191 if (m && mbuf_removed 1192 && (pr->pr_flags & PR_ATOMIC)) 1193 (void) sbdroprecord(&so->so_rcv); 1194 1195 goto release; 1196 } 1197 } else 1198 uio->uio_resid -= len; 1199 if (len == m->m_len - moff) { 1200 if (m->m_flags & M_EOR) 1201 flags |= MSG_EOR; 1202 if (flags & MSG_PEEK) { 1203 m = m->m_next; 1204 moff = 0; 1205 } else { 1206 nextrecord = m->m_nextpkt; 1207 sbfree(&so->so_rcv, m); 1208 if (mp) { 1209 *mp = m; 1210 mp = &m->m_next; 1211 so->so_rcv.sb_mb = m = m->m_next; 1212 *mp = (struct mbuf *)0; 1213 } else { 1214 MFREE(m, so->so_rcv.sb_mb); 1215 m = so->so_rcv.sb_mb; 1216 } 1217 /* 1218 * If m != NULL, we also know that 1219 * so->so_rcv.sb_mb != NULL. 1220 */ 1221 KASSERT(so->so_rcv.sb_mb == m); 1222 if (m) { 1223 m->m_nextpkt = nextrecord; 1224 if (nextrecord == NULL) 1225 so->so_rcv.sb_lastrecord = m; 1226 } else { 1227 so->so_rcv.sb_mb = nextrecord; 1228 SB_EMPTY_FIXUP(&so->so_rcv); 1229 } 1230 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1231 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1232 } 1233 } else { 1234 if (flags & MSG_PEEK) 1235 moff += len; 1236 else { 1237 if (mp) 1238 *mp = m_copym(m, 0, len, M_WAIT); 1239 m->m_data += len; 1240 m->m_len -= len; 1241 so->so_rcv.sb_cc -= len; 1242 } 1243 } 1244 if (so->so_oobmark) { 1245 if ((flags & MSG_PEEK) == 0) { 1246 so->so_oobmark -= len; 1247 if (so->so_oobmark == 0) { 1248 so->so_state |= SS_RCVATMARK; 1249 break; 1250 } 1251 } else { 1252 offset += len; 1253 if (offset == so->so_oobmark) 1254 break; 1255 } 1256 } 1257 if (flags & MSG_EOR) 1258 break; 1259 /* 1260 * If the MSG_WAITALL flag is set (for non-atomic socket), 1261 * we must not quit until "uio->uio_resid == 0" or an error 1262 * termination. If a signal/timeout occurs, return 1263 * with a short count but without error. 1264 * Keep sockbuf locked against other readers. 1265 */ 1266 while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 && 1267 !sosendallatonce(so) && !nextrecord) { 1268 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1269 break; 1270 /* 1271 * If we are peeking and the socket receive buffer is 1272 * full, stop since we can't get more data to peek at. 1273 */ 1274 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1275 break; 1276 /* 1277 * If we've drained the socket buffer, tell the 1278 * protocol in case it needs to do something to 1279 * get it filled again. 1280 */ 1281 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1282 (*pr->pr_usrreq)(so, PRU_RCVD, 1283 (struct mbuf *)0, 1284 (struct mbuf *)(long)flags, 1285 (struct mbuf *)0, p); 1286 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1287 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1288 error = sbwait(&so->so_rcv); 1289 if (error) { 1290 sbunlock(&so->so_rcv); 1291 splx(s); 1292 return (0); 1293 } 1294 if ((m = so->so_rcv.sb_mb) != NULL) 1295 nextrecord = m->m_nextpkt; 1296 } 1297 } 1298 1299 if (m && pr->pr_flags & PR_ATOMIC) { 1300 flags |= MSG_TRUNC; 1301 if ((flags & MSG_PEEK) == 0) 1302 (void) sbdroprecord(&so->so_rcv); 1303 } 1304 if ((flags & MSG_PEEK) == 0) { 1305 if (m == 0) { 1306 /* 1307 * First part is an inline SB_EMPTY_FIXUP(). Second 1308 * part makes sure sb_lastrecord is up-to-date if 1309 * there is still data in the socket buffer. 1310 */ 1311 so->so_rcv.sb_mb = nextrecord; 1312 if (so->so_rcv.sb_mb == NULL) { 1313 so->so_rcv.sb_mbtail = NULL; 1314 so->so_rcv.sb_lastrecord = NULL; 1315 } else if (nextrecord->m_nextpkt == NULL) 1316 so->so_rcv.sb_lastrecord = nextrecord; 1317 } 1318 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1319 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1320 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1321 (*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0, 1322 (struct mbuf *)(long)flags, (struct mbuf *)0, p); 1323 } 1324 if (orig_resid == uio->uio_resid && orig_resid && 1325 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1326 sbunlock(&so->so_rcv); 1327 splx(s); 1328 goto restart; 1329 } 1330 1331 if (flagsp) 1332 *flagsp |= flags; 1333 release: 1334 sbunlock(&so->so_rcv); 1335 splx(s); 1336 return (error); 1337 } 1338 1339 int 1340 soshutdown(struct socket *so, int how) 1341 { 1342 const struct protosw *pr; 1343 1344 pr = so->so_proto; 1345 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1346 return (EINVAL); 1347 1348 if (how == SHUT_RD || how == SHUT_RDWR) 1349 sorflush(so); 1350 if (how == SHUT_WR || how == SHUT_RDWR) 1351 return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0, 1352 (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0); 1353 return (0); 1354 } 1355 1356 void 1357 sorflush(struct socket *so) 1358 { 1359 struct sockbuf *sb, asb; 1360 const struct protosw *pr; 1361 int s; 1362 1363 sb = &so->so_rcv; 1364 pr = so->so_proto; 1365 sb->sb_flags |= SB_NOINTR; 1366 (void) sblock(sb, M_WAITOK); 1367 s = splnet(); 1368 socantrcvmore(so); 1369 sbunlock(sb); 1370 asb = *sb; 1371 /* 1372 * Clear most of the sockbuf structure, but leave some of the 1373 * fields valid. 1374 */ 1375 memset(&sb->sb_startzero, 0, 1376 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1377 splx(s); 1378 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) 1379 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1380 sbrelease(&asb, so); 1381 } 1382 1383 int 1384 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0) 1385 { 1386 int error; 1387 struct mbuf *m; 1388 1389 error = 0; 1390 m = m0; 1391 if (level != SOL_SOCKET) { 1392 if (so->so_proto && so->so_proto->pr_ctloutput) 1393 return ((*so->so_proto->pr_ctloutput) 1394 (PRCO_SETOPT, so, level, optname, &m0)); 1395 error = ENOPROTOOPT; 1396 } else { 1397 switch (optname) { 1398 1399 case SO_LINGER: 1400 if (m == NULL || m->m_len != sizeof(struct linger)) { 1401 error = EINVAL; 1402 goto bad; 1403 } 1404 so->so_linger = mtod(m, struct linger *)->l_linger; 1405 /* fall thru... */ 1406 1407 case SO_DEBUG: 1408 case SO_KEEPALIVE: 1409 case SO_DONTROUTE: 1410 case SO_USELOOPBACK: 1411 case SO_BROADCAST: 1412 case SO_REUSEADDR: 1413 case SO_REUSEPORT: 1414 case SO_OOBINLINE: 1415 case SO_TIMESTAMP: 1416 if (m == NULL || m->m_len < sizeof(int)) { 1417 error = EINVAL; 1418 goto bad; 1419 } 1420 if (*mtod(m, int *)) 1421 so->so_options |= optname; 1422 else 1423 so->so_options &= ~optname; 1424 break; 1425 1426 case SO_SNDBUF: 1427 case SO_RCVBUF: 1428 case SO_SNDLOWAT: 1429 case SO_RCVLOWAT: 1430 { 1431 int optval; 1432 1433 if (m == NULL || m->m_len < sizeof(int)) { 1434 error = EINVAL; 1435 goto bad; 1436 } 1437 1438 /* 1439 * Values < 1 make no sense for any of these 1440 * options, so disallow them. 1441 */ 1442 optval = *mtod(m, int *); 1443 if (optval < 1) { 1444 error = EINVAL; 1445 goto bad; 1446 } 1447 1448 switch (optname) { 1449 1450 case SO_SNDBUF: 1451 case SO_RCVBUF: 1452 if (sbreserve(optname == SO_SNDBUF ? 1453 &so->so_snd : &so->so_rcv, 1454 (u_long) optval, so) == 0) { 1455 error = ENOBUFS; 1456 goto bad; 1457 } 1458 break; 1459 1460 /* 1461 * Make sure the low-water is never greater than 1462 * the high-water. 1463 */ 1464 case SO_SNDLOWAT: 1465 so->so_snd.sb_lowat = 1466 (optval > so->so_snd.sb_hiwat) ? 1467 so->so_snd.sb_hiwat : optval; 1468 break; 1469 case SO_RCVLOWAT: 1470 so->so_rcv.sb_lowat = 1471 (optval > so->so_rcv.sb_hiwat) ? 1472 so->so_rcv.sb_hiwat : optval; 1473 break; 1474 } 1475 break; 1476 } 1477 1478 case SO_SNDTIMEO: 1479 case SO_RCVTIMEO: 1480 { 1481 struct timeval *tv; 1482 int val; 1483 1484 if (m == NULL || m->m_len < sizeof(*tv)) { 1485 error = EINVAL; 1486 goto bad; 1487 } 1488 tv = mtod(m, struct timeval *); 1489 if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz) { 1490 error = EDOM; 1491 goto bad; 1492 } 1493 val = tv->tv_sec * hz + tv->tv_usec / tick; 1494 if (val == 0 && tv->tv_usec != 0) 1495 val = 1; 1496 1497 switch (optname) { 1498 1499 case SO_SNDTIMEO: 1500 so->so_snd.sb_timeo = val; 1501 break; 1502 case SO_RCVTIMEO: 1503 so->so_rcv.sb_timeo = val; 1504 break; 1505 } 1506 break; 1507 } 1508 1509 default: 1510 error = ENOPROTOOPT; 1511 break; 1512 } 1513 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) { 1514 (void) ((*so->so_proto->pr_ctloutput) 1515 (PRCO_SETOPT, so, level, optname, &m0)); 1516 m = NULL; /* freed by protocol */ 1517 } 1518 } 1519 bad: 1520 if (m) 1521 (void) m_free(m); 1522 return (error); 1523 } 1524 1525 int 1526 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp) 1527 { 1528 struct mbuf *m; 1529 1530 if (level != SOL_SOCKET) { 1531 if (so->so_proto && so->so_proto->pr_ctloutput) { 1532 return ((*so->so_proto->pr_ctloutput) 1533 (PRCO_GETOPT, so, level, optname, mp)); 1534 } else 1535 return (ENOPROTOOPT); 1536 } else { 1537 m = m_get(M_WAIT, MT_SOOPTS); 1538 m->m_len = sizeof(int); 1539 1540 switch (optname) { 1541 1542 case SO_LINGER: 1543 m->m_len = sizeof(struct linger); 1544 mtod(m, struct linger *)->l_onoff = 1545 so->so_options & SO_LINGER; 1546 mtod(m, struct linger *)->l_linger = so->so_linger; 1547 break; 1548 1549 case SO_USELOOPBACK: 1550 case SO_DONTROUTE: 1551 case SO_DEBUG: 1552 case SO_KEEPALIVE: 1553 case SO_REUSEADDR: 1554 case SO_REUSEPORT: 1555 case SO_BROADCAST: 1556 case SO_OOBINLINE: 1557 case SO_TIMESTAMP: 1558 *mtod(m, int *) = so->so_options & optname; 1559 break; 1560 1561 case SO_TYPE: 1562 *mtod(m, int *) = so->so_type; 1563 break; 1564 1565 case SO_ERROR: 1566 *mtod(m, int *) = so->so_error; 1567 so->so_error = 0; 1568 break; 1569 1570 case SO_SNDBUF: 1571 *mtod(m, int *) = so->so_snd.sb_hiwat; 1572 break; 1573 1574 case SO_RCVBUF: 1575 *mtod(m, int *) = so->so_rcv.sb_hiwat; 1576 break; 1577 1578 case SO_SNDLOWAT: 1579 *mtod(m, int *) = so->so_snd.sb_lowat; 1580 break; 1581 1582 case SO_RCVLOWAT: 1583 *mtod(m, int *) = so->so_rcv.sb_lowat; 1584 break; 1585 1586 case SO_SNDTIMEO: 1587 case SO_RCVTIMEO: 1588 { 1589 int val = (optname == SO_SNDTIMEO ? 1590 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1591 1592 m->m_len = sizeof(struct timeval); 1593 mtod(m, struct timeval *)->tv_sec = val / hz; 1594 mtod(m, struct timeval *)->tv_usec = 1595 (val % hz) * tick; 1596 break; 1597 } 1598 1599 case SO_OVERFLOWED: 1600 *mtod(m, int *) = so->so_rcv.sb_overflowed; 1601 break; 1602 1603 default: 1604 (void)m_free(m); 1605 return (ENOPROTOOPT); 1606 } 1607 *mp = m; 1608 return (0); 1609 } 1610 } 1611 1612 void 1613 sohasoutofband(struct socket *so) 1614 { 1615 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 1616 selwakeup(&so->so_rcv.sb_sel); 1617 } 1618 1619 static void 1620 filt_sordetach(struct knote *kn) 1621 { 1622 struct socket *so; 1623 1624 so = (struct socket *)kn->kn_fp->f_data; 1625 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 1626 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 1627 so->so_rcv.sb_flags &= ~SB_KNOTE; 1628 } 1629 1630 /*ARGSUSED*/ 1631 static int 1632 filt_soread(struct knote *kn, long hint) 1633 { 1634 struct socket *so; 1635 1636 so = (struct socket *)kn->kn_fp->f_data; 1637 kn->kn_data = so->so_rcv.sb_cc; 1638 if (so->so_state & SS_CANTRCVMORE) { 1639 kn->kn_flags |= EV_EOF; 1640 kn->kn_fflags = so->so_error; 1641 return (1); 1642 } 1643 if (so->so_error) /* temporary udp error */ 1644 return (1); 1645 if (kn->kn_sfflags & NOTE_LOWAT) 1646 return (kn->kn_data >= kn->kn_sdata); 1647 return (kn->kn_data >= so->so_rcv.sb_lowat); 1648 } 1649 1650 static void 1651 filt_sowdetach(struct knote *kn) 1652 { 1653 struct socket *so; 1654 1655 so = (struct socket *)kn->kn_fp->f_data; 1656 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 1657 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 1658 so->so_snd.sb_flags &= ~SB_KNOTE; 1659 } 1660 1661 /*ARGSUSED*/ 1662 static int 1663 filt_sowrite(struct knote *kn, long hint) 1664 { 1665 struct socket *so; 1666 1667 so = (struct socket *)kn->kn_fp->f_data; 1668 kn->kn_data = sbspace(&so->so_snd); 1669 if (so->so_state & SS_CANTSENDMORE) { 1670 kn->kn_flags |= EV_EOF; 1671 kn->kn_fflags = so->so_error; 1672 return (1); 1673 } 1674 if (so->so_error) /* temporary udp error */ 1675 return (1); 1676 if (((so->so_state & SS_ISCONNECTED) == 0) && 1677 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 1678 return (0); 1679 if (kn->kn_sfflags & NOTE_LOWAT) 1680 return (kn->kn_data >= kn->kn_sdata); 1681 return (kn->kn_data >= so->so_snd.sb_lowat); 1682 } 1683 1684 /*ARGSUSED*/ 1685 static int 1686 filt_solisten(struct knote *kn, long hint) 1687 { 1688 struct socket *so; 1689 1690 so = (struct socket *)kn->kn_fp->f_data; 1691 1692 /* 1693 * Set kn_data to number of incoming connections, not 1694 * counting partial (incomplete) connections. 1695 */ 1696 kn->kn_data = so->so_qlen; 1697 return (kn->kn_data > 0); 1698 } 1699 1700 static const struct filterops solisten_filtops = 1701 { 1, NULL, filt_sordetach, filt_solisten }; 1702 static const struct filterops soread_filtops = 1703 { 1, NULL, filt_sordetach, filt_soread }; 1704 static const struct filterops sowrite_filtops = 1705 { 1, NULL, filt_sowdetach, filt_sowrite }; 1706 1707 int 1708 soo_kqfilter(struct file *fp, struct knote *kn) 1709 { 1710 struct socket *so; 1711 struct sockbuf *sb; 1712 1713 so = (struct socket *)kn->kn_fp->f_data; 1714 switch (kn->kn_filter) { 1715 case EVFILT_READ: 1716 if (so->so_options & SO_ACCEPTCONN) 1717 kn->kn_fop = &solisten_filtops; 1718 else 1719 kn->kn_fop = &soread_filtops; 1720 sb = &so->so_rcv; 1721 break; 1722 case EVFILT_WRITE: 1723 kn->kn_fop = &sowrite_filtops; 1724 sb = &so->so_snd; 1725 break; 1726 default: 1727 return (1); 1728 } 1729 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 1730 sb->sb_flags |= SB_KNOTE; 1731 return (0); 1732 } 1733 1734 #include <sys/sysctl.h> 1735 1736 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 1737 1738 /* 1739 * sysctl helper routine for kern.somaxkva. ensures that the given 1740 * value is not too small. 1741 * (XXX should we maybe make sure it's not too large as well?) 1742 */ 1743 static int 1744 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 1745 { 1746 int error, new_somaxkva; 1747 struct sysctlnode node; 1748 int s; 1749 1750 new_somaxkva = somaxkva; 1751 node = *rnode; 1752 node.sysctl_data = &new_somaxkva; 1753 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1754 if (error || newp == NULL) 1755 return (error); 1756 1757 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 1758 return (EINVAL); 1759 1760 s = splvm(); 1761 simple_lock(&so_pendfree_slock); 1762 somaxkva = new_somaxkva; 1763 wakeup(&socurkva); 1764 simple_unlock(&so_pendfree_slock); 1765 splx(s); 1766 1767 return (error); 1768 } 1769 1770 SYSCTL_SETUP(sysctl_kern_somaxkva_setup, "sysctl kern.somaxkva setup") 1771 { 1772 1773 sysctl_createv(clog, 0, NULL, NULL, 1774 CTLFLAG_PERMANENT, 1775 CTLTYPE_NODE, "kern", NULL, 1776 NULL, 0, NULL, 0, 1777 CTL_KERN, CTL_EOL); 1778 1779 sysctl_createv(clog, 0, NULL, NULL, 1780 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1781 CTLTYPE_INT, "somaxkva", 1782 SYSCTL_DESCR("Maximum amount of kernel memory to be " 1783 "used for socket buffers"), 1784 sysctl_kern_somaxkva, 0, NULL, 0, 1785 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 1786 } 1787
Indexes created Sat Sep 20 22:09:52 GMT 2025