1 /* $NetBSD: uipc_socket.c,v 1.257 2017/10/25 08:12:39 maya Exp $ */ 2 3 /*- 4 * Copyright (c) 2002, 2007, 2008, 2009 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, and by Andrew Doran. 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 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * Copyright (c) 2004 The FreeBSD Foundation 34 * Copyright (c) 2004 Robert Watson 35 * Copyright (c) 1982, 1986, 1988, 1990, 1993 36 * The Regents of the University of California. All rights reserved. 37 * 38 * Redistribution and use in source and binary forms, with or without 39 * modification, are permitted provided that the following conditions 40 * are met: 41 * 1. Redistributions of source code must retain the above copyright 42 * notice, this list of conditions and the following disclaimer. 43 * 2. Redistributions in binary form must reproduce the above copyright 44 * notice, this list of conditions and the following disclaimer in the 45 * documentation and/or other materials provided with the distribution. 46 * 3. Neither the name of the University nor the names of its contributors 47 * may be used to endorse or promote products derived from this software 48 * without specific prior written permission. 49 * 50 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 53 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 60 * SUCH DAMAGE. 61 * 62 * @(#)uipc_socket.c 8.6 (Berkeley) 5/2/95 63 */ 64 65 /* 66 * Socket operation routines. 67 * 68 * These routines are called by the routines in sys_socket.c or from a 69 * system process, and implement the semantics of socket operations by 70 * switching out to the protocol specific routines. 71 */ 72 73 #include <sys/cdefs.h> 74 __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.257 2017/10/25 08:12:39 maya Exp $"); 75 76 #ifdef _KERNEL_OPT 77 #include "opt_compat_netbsd.h" 78 #include "opt_sock_counters.h" 79 #include "opt_sosend_loan.h" 80 #include "opt_mbuftrace.h" 81 #include "opt_somaxkva.h" 82 #include "opt_multiprocessor.h" /* XXX */ 83 #include "opt_sctp.h" 84 #endif 85 86 #include <sys/param.h> 87 #include <sys/systm.h> 88 #include <sys/proc.h> 89 #include <sys/file.h> 90 #include <sys/filedesc.h> 91 #include <sys/kmem.h> 92 #include <sys/mbuf.h> 93 #include <sys/domain.h> 94 #include <sys/kernel.h> 95 #include <sys/protosw.h> 96 #include <sys/socket.h> 97 #include <sys/socketvar.h> 98 #include <sys/signalvar.h> 99 #include <sys/resourcevar.h> 100 #include <sys/uidinfo.h> 101 #include <sys/event.h> 102 #include <sys/poll.h> 103 #include <sys/kauth.h> 104 #include <sys/mutex.h> 105 #include <sys/condvar.h> 106 #include <sys/kthread.h> 107 108 #ifdef COMPAT_50 109 #include <compat/sys/time.h> 110 #include <compat/sys/socket.h> 111 #endif 112 113 #include <uvm/uvm_extern.h> 114 #include <uvm/uvm_loan.h> 115 #include <uvm/uvm_page.h> 116 117 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 118 119 extern const struct fileops socketops; 120 121 extern int somaxconn; /* patchable (XXX sysctl) */ 122 int somaxconn = SOMAXCONN; 123 kmutex_t *softnet_lock; 124 125 #ifdef SOSEND_COUNTERS 126 #include <sys/device.h> 127 128 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 129 NULL, "sosend", "loan big"); 130 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 131 NULL, "sosend", "copy big"); 132 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 133 NULL, "sosend", "copy small"); 134 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 135 NULL, "sosend", "kva limit"); 136 137 #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++ 138 139 EVCNT_ATTACH_STATIC(sosend_loan_big); 140 EVCNT_ATTACH_STATIC(sosend_copy_big); 141 EVCNT_ATTACH_STATIC(sosend_copy_small); 142 EVCNT_ATTACH_STATIC(sosend_kvalimit); 143 #else 144 145 #define SOSEND_COUNTER_INCR(ev) /* nothing */ 146 147 #endif /* SOSEND_COUNTERS */ 148 149 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR) 150 int sock_loan_thresh = -1; 151 #else 152 int sock_loan_thresh = 4096; 153 #endif 154 155 static kmutex_t so_pendfree_lock; 156 static struct mbuf *so_pendfree = NULL; 157 158 #ifndef SOMAXKVA 159 #define SOMAXKVA (16 * 1024 * 1024) 160 #endif 161 int somaxkva = SOMAXKVA; 162 static int socurkva; 163 static kcondvar_t socurkva_cv; 164 165 static kauth_listener_t socket_listener; 166 167 #define SOCK_LOAN_CHUNK 65536 168 169 static void sopendfree_thread(void *); 170 static kcondvar_t pendfree_thread_cv; 171 static lwp_t *sopendfree_lwp; 172 173 static void sysctl_kern_socket_setup(void); 174 static struct sysctllog *socket_sysctllog; 175 176 static vsize_t 177 sokvareserve(struct socket *so, vsize_t len) 178 { 179 int error; 180 181 mutex_enter(&so_pendfree_lock); 182 while (socurkva + len > somaxkva) { 183 SOSEND_COUNTER_INCR(&sosend_kvalimit); 184 error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock); 185 if (error) { 186 len = 0; 187 break; 188 } 189 } 190 socurkva += len; 191 mutex_exit(&so_pendfree_lock); 192 return len; 193 } 194 195 static void 196 sokvaunreserve(vsize_t len) 197 { 198 199 mutex_enter(&so_pendfree_lock); 200 socurkva -= len; 201 cv_broadcast(&socurkva_cv); 202 mutex_exit(&so_pendfree_lock); 203 } 204 205 /* 206 * sokvaalloc: allocate kva for loan. 207 */ 208 209 vaddr_t 210 sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so) 211 { 212 vaddr_t lva; 213 214 /* 215 * reserve kva. 216 */ 217 218 if (sokvareserve(so, len) == 0) 219 return 0; 220 221 /* 222 * allocate kva. 223 */ 224 225 lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask, 226 UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA); 227 if (lva == 0) { 228 sokvaunreserve(len); 229 return (0); 230 } 231 232 return lva; 233 } 234 235 /* 236 * sokvafree: free kva for loan. 237 */ 238 239 void 240 sokvafree(vaddr_t sva, vsize_t len) 241 { 242 243 /* 244 * free kva. 245 */ 246 247 uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY); 248 249 /* 250 * unreserve kva. 251 */ 252 253 sokvaunreserve(len); 254 } 255 256 static void 257 sodoloanfree(struct vm_page **pgs, void *buf, size_t size) 258 { 259 vaddr_t sva, eva; 260 vsize_t len; 261 int npgs; 262 263 KASSERT(pgs != NULL); 264 265 eva = round_page((vaddr_t) buf + size); 266 sva = trunc_page((vaddr_t) buf); 267 len = eva - sva; 268 npgs = len >> PAGE_SHIFT; 269 270 pmap_kremove(sva, len); 271 pmap_update(pmap_kernel()); 272 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 273 sokvafree(sva, len); 274 } 275 276 /* 277 * sopendfree_thread: free mbufs on "pendfree" list. 278 * unlock and relock so_pendfree_lock when freeing mbufs. 279 */ 280 281 static void 282 sopendfree_thread(void *v) 283 { 284 struct mbuf *m, *next; 285 size_t rv; 286 287 mutex_enter(&so_pendfree_lock); 288 289 for (;;) { 290 rv = 0; 291 while (so_pendfree != NULL) { 292 m = so_pendfree; 293 so_pendfree = NULL; 294 mutex_exit(&so_pendfree_lock); 295 296 for (; m != NULL; m = next) { 297 next = m->m_next; 298 KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 299 0); 300 KASSERT(m->m_ext.ext_refcnt == 0); 301 302 rv += m->m_ext.ext_size; 303 sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf, 304 m->m_ext.ext_size); 305 pool_cache_put(mb_cache, m); 306 } 307 308 mutex_enter(&so_pendfree_lock); 309 } 310 if (rv) 311 cv_broadcast(&socurkva_cv); 312 cv_wait(&pendfree_thread_cv, &so_pendfree_lock); 313 } 314 panic("sopendfree_thread"); 315 /* NOTREACHED */ 316 } 317 318 void 319 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg) 320 { 321 322 KASSERT(m != NULL); 323 324 /* 325 * postpone freeing mbuf. 326 * 327 * we can't do it in interrupt context 328 * because we need to put kva back to kernel_map. 329 */ 330 331 mutex_enter(&so_pendfree_lock); 332 m->m_next = so_pendfree; 333 so_pendfree = m; 334 cv_signal(&pendfree_thread_cv); 335 mutex_exit(&so_pendfree_lock); 336 } 337 338 static long 339 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 340 { 341 struct iovec *iov = uio->uio_iov; 342 vaddr_t sva, eva; 343 vsize_t len; 344 vaddr_t lva; 345 int npgs, error; 346 vaddr_t va; 347 int i; 348 349 if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace)) 350 return (0); 351 352 if (iov->iov_len < (size_t) space) 353 space = iov->iov_len; 354 if (space > SOCK_LOAN_CHUNK) 355 space = SOCK_LOAN_CHUNK; 356 357 eva = round_page((vaddr_t) iov->iov_base + space); 358 sva = trunc_page((vaddr_t) iov->iov_base); 359 len = eva - sva; 360 npgs = len >> PAGE_SHIFT; 361 362 KASSERT(npgs <= M_EXT_MAXPAGES); 363 364 lva = sokvaalloc(sva, len, so); 365 if (lva == 0) 366 return 0; 367 368 error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len, 369 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 370 if (error) { 371 sokvafree(lva, len); 372 return (0); 373 } 374 375 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 376 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 377 VM_PROT_READ, 0); 378 pmap_update(pmap_kernel()); 379 380 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 381 382 MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so); 383 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 384 385 uio->uio_resid -= space; 386 /* uio_offset not updated, not set/used for write(2) */ 387 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space; 388 uio->uio_iov->iov_len -= space; 389 if (uio->uio_iov->iov_len == 0) { 390 uio->uio_iov++; 391 uio->uio_iovcnt--; 392 } 393 394 return (space); 395 } 396 397 struct mbuf * 398 getsombuf(struct socket *so, int type) 399 { 400 struct mbuf *m; 401 402 m = m_get(M_WAIT, type); 403 MCLAIM(m, so->so_mowner); 404 return m; 405 } 406 407 static int 408 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 409 void *arg0, void *arg1, void *arg2, void *arg3) 410 { 411 int result; 412 enum kauth_network_req req; 413 414 result = KAUTH_RESULT_DEFER; 415 req = (enum kauth_network_req)arg0; 416 417 if ((action != KAUTH_NETWORK_SOCKET) && 418 (action != KAUTH_NETWORK_BIND)) 419 return result; 420 421 switch (req) { 422 case KAUTH_REQ_NETWORK_BIND_PORT: 423 result = KAUTH_RESULT_ALLOW; 424 break; 425 426 case KAUTH_REQ_NETWORK_SOCKET_DROP: { 427 /* Normal users can only drop their own connections. */ 428 struct socket *so = (struct socket *)arg1; 429 430 if (so->so_cred && proc_uidmatch(cred, so->so_cred) == 0) 431 result = KAUTH_RESULT_ALLOW; 432 433 break; 434 } 435 436 case KAUTH_REQ_NETWORK_SOCKET_OPEN: 437 /* We allow "raw" routing/bluetooth sockets to anyone. */ 438 switch ((u_long)arg1) { 439 case PF_ROUTE: 440 case PF_OROUTE: 441 case PF_BLUETOOTH: 442 case PF_CAN: 443 result = KAUTH_RESULT_ALLOW; 444 break; 445 default: 446 /* Privileged, let secmodel handle this. */ 447 if ((u_long)arg2 == SOCK_RAW) 448 break; 449 result = KAUTH_RESULT_ALLOW; 450 break; 451 } 452 break; 453 454 case KAUTH_REQ_NETWORK_SOCKET_CANSEE: 455 result = KAUTH_RESULT_ALLOW; 456 457 break; 458 459 default: 460 break; 461 } 462 463 return result; 464 } 465 466 void 467 soinit(void) 468 { 469 470 sysctl_kern_socket_setup(); 471 472 mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM); 473 softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 474 cv_init(&socurkva_cv, "sokva"); 475 cv_init(&pendfree_thread_cv, "sopendfr"); 476 soinit2(); 477 478 /* Set the initial adjusted socket buffer size. */ 479 if (sb_max_set(sb_max)) 480 panic("bad initial sb_max value: %lu", sb_max); 481 482 socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK, 483 socket_listener_cb, NULL); 484 } 485 486 void 487 soinit1(void) 488 { 489 int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, 490 sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree"); 491 if (error) 492 panic("soinit1 %d", error); 493 } 494 495 /* 496 * socreate: create a new socket of the specified type and the protocol. 497 * 498 * => Caller may specify another socket for lock sharing (must not be held). 499 * => Returns the new socket without lock held. 500 */ 501 int 502 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l, 503 struct socket *lockso) 504 { 505 const struct protosw *prp; 506 struct socket *so; 507 uid_t uid; 508 int error; 509 kmutex_t *lock; 510 511 error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET, 512 KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type), 513 KAUTH_ARG(proto)); 514 if (error != 0) 515 return error; 516 517 if (proto) 518 prp = pffindproto(dom, proto, type); 519 else 520 prp = pffindtype(dom, type); 521 if (prp == NULL) { 522 /* no support for domain */ 523 if (pffinddomain(dom) == 0) 524 return EAFNOSUPPORT; 525 /* no support for socket type */ 526 if (proto == 0 && type != 0) 527 return EPROTOTYPE; 528 return EPROTONOSUPPORT; 529 } 530 if (prp->pr_usrreqs == NULL) 531 return EPROTONOSUPPORT; 532 if (prp->pr_type != type) 533 return EPROTOTYPE; 534 535 so = soget(true); 536 so->so_type = type; 537 so->so_proto = prp; 538 so->so_send = sosend; 539 so->so_receive = soreceive; 540 #ifdef MBUFTRACE 541 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 542 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 543 so->so_mowner = &prp->pr_domain->dom_mowner; 544 #endif 545 uid = kauth_cred_geteuid(l->l_cred); 546 so->so_uidinfo = uid_find(uid); 547 so->so_cpid = l->l_proc->p_pid; 548 549 /* 550 * Lock assigned and taken during PCB attach, unless we share 551 * the lock with another socket, e.g. socketpair(2) case. 552 */ 553 if (lockso) { 554 lock = lockso->so_lock; 555 so->so_lock = lock; 556 mutex_obj_hold(lock); 557 mutex_enter(lock); 558 } 559 560 /* Attach the PCB (returns with the socket lock held). */ 561 error = (*prp->pr_usrreqs->pr_attach)(so, proto); 562 KASSERT(solocked(so)); 563 564 if (error) { 565 KASSERT(so->so_pcb == NULL); 566 so->so_state |= SS_NOFDREF; 567 sofree(so); 568 return error; 569 } 570 so->so_cred = kauth_cred_dup(l->l_cred); 571 sounlock(so); 572 573 *aso = so; 574 return 0; 575 } 576 577 /* 578 * fsocreate: create a socket and a file descriptor associated with it. 579 * 580 * => On success, write file descriptor to fdout and return zero. 581 * => On failure, return non-zero; *fdout will be undefined. 582 */ 583 int 584 fsocreate(int domain, struct socket **sop, int type, int proto, int *fdout) 585 { 586 lwp_t *l = curlwp; 587 int error, fd, flags; 588 struct socket *so; 589 struct file *fp; 590 591 if ((error = fd_allocfile(&fp, &fd)) != 0) { 592 return error; 593 } 594 flags = type & SOCK_FLAGS_MASK; 595 fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0); 596 fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)| 597 ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0); 598 fp->f_type = DTYPE_SOCKET; 599 fp->f_ops = &socketops; 600 601 type &= ~SOCK_FLAGS_MASK; 602 error = socreate(domain, &so, type, proto, l, NULL); 603 if (error) { 604 fd_abort(curproc, fp, fd); 605 return error; 606 } 607 if (flags & SOCK_NONBLOCK) { 608 so->so_state |= SS_NBIO; 609 } 610 fp->f_socket = so; 611 fd_affix(curproc, fp, fd); 612 613 if (sop != NULL) { 614 *sop = so; 615 } 616 *fdout = fd; 617 return error; 618 } 619 620 int 621 sofamily(const struct socket *so) 622 { 623 const struct protosw *pr; 624 const struct domain *dom; 625 626 if ((pr = so->so_proto) == NULL) 627 return AF_UNSPEC; 628 if ((dom = pr->pr_domain) == NULL) 629 return AF_UNSPEC; 630 return dom->dom_family; 631 } 632 633 int 634 sobind(struct socket *so, struct sockaddr *nam, struct lwp *l) 635 { 636 int error; 637 638 solock(so); 639 if (nam->sa_family != so->so_proto->pr_domain->dom_family) { 640 sounlock(so); 641 return EAFNOSUPPORT; 642 } 643 error = (*so->so_proto->pr_usrreqs->pr_bind)(so, nam, l); 644 sounlock(so); 645 return error; 646 } 647 648 int 649 solisten(struct socket *so, int backlog, struct lwp *l) 650 { 651 int error; 652 short oldopt, oldqlimit; 653 654 solock(so); 655 if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 656 SS_ISDISCONNECTING)) != 0) { 657 sounlock(so); 658 return EINVAL; 659 } 660 oldopt = so->so_options; 661 oldqlimit = so->so_qlimit; 662 if (TAILQ_EMPTY(&so->so_q)) 663 so->so_options |= SO_ACCEPTCONN; 664 if (backlog < 0) 665 backlog = 0; 666 so->so_qlimit = min(backlog, somaxconn); 667 668 error = (*so->so_proto->pr_usrreqs->pr_listen)(so, l); 669 if (error != 0) { 670 so->so_options = oldopt; 671 so->so_qlimit = oldqlimit; 672 sounlock(so); 673 return error; 674 } 675 sounlock(so); 676 return 0; 677 } 678 679 void 680 sofree(struct socket *so) 681 { 682 u_int refs; 683 684 KASSERT(solocked(so)); 685 686 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) { 687 sounlock(so); 688 return; 689 } 690 if (so->so_head) { 691 /* 692 * We must not decommission a socket that's on the accept(2) 693 * queue. If we do, then accept(2) may hang after select(2) 694 * indicated that the listening socket was ready. 695 */ 696 if (!soqremque(so, 0)) { 697 sounlock(so); 698 return; 699 } 700 } 701 if (so->so_rcv.sb_hiwat) 702 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 703 RLIM_INFINITY); 704 if (so->so_snd.sb_hiwat) 705 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 706 RLIM_INFINITY); 707 sbrelease(&so->so_snd, so); 708 KASSERT(!cv_has_waiters(&so->so_cv)); 709 KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv)); 710 KASSERT(!cv_has_waiters(&so->so_snd.sb_cv)); 711 sorflush(so); 712 refs = so->so_aborting; /* XXX */ 713 /* Remove acccept filter if one is present. */ 714 if (so->so_accf != NULL) 715 (void)accept_filt_clear(so); 716 sounlock(so); 717 if (refs == 0) /* XXX */ 718 soput(so); 719 } 720 721 /* 722 * soclose: close a socket on last file table reference removal. 723 * Initiate disconnect if connected. Free socket when disconnect complete. 724 */ 725 int 726 soclose(struct socket *so) 727 { 728 struct socket *so2; 729 int error = 0; 730 731 solock(so); 732 if (so->so_options & SO_ACCEPTCONN) { 733 for (;;) { 734 if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 735 KASSERT(solocked2(so, so2)); 736 (void) soqremque(so2, 0); 737 /* soabort drops the lock. */ 738 (void) soabort(so2); 739 solock(so); 740 continue; 741 } 742 if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 743 KASSERT(solocked2(so, so2)); 744 (void) soqremque(so2, 1); 745 /* soabort drops the lock. */ 746 (void) soabort(so2); 747 solock(so); 748 continue; 749 } 750 break; 751 } 752 } 753 if (so->so_pcb == NULL) 754 goto discard; 755 if (so->so_state & SS_ISCONNECTED) { 756 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 757 error = sodisconnect(so); 758 if (error) 759 goto drop; 760 } 761 if (so->so_options & SO_LINGER) { 762 if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) == 763 (SS_ISDISCONNECTING|SS_NBIO)) 764 goto drop; 765 while (so->so_state & SS_ISCONNECTED) { 766 error = sowait(so, true, so->so_linger * hz); 767 if (error) 768 break; 769 } 770 } 771 } 772 drop: 773 if (so->so_pcb) { 774 KASSERT(solocked(so)); 775 (*so->so_proto->pr_usrreqs->pr_detach)(so); 776 } 777 discard: 778 KASSERT((so->so_state & SS_NOFDREF) == 0); 779 kauth_cred_free(so->so_cred); 780 so->so_state |= SS_NOFDREF; 781 sofree(so); 782 return error; 783 } 784 785 /* 786 * Must be called with the socket locked.. Will return with it unlocked. 787 */ 788 int 789 soabort(struct socket *so) 790 { 791 u_int refs; 792 int error; 793 794 KASSERT(solocked(so)); 795 KASSERT(so->so_head == NULL); 796 797 so->so_aborting++; /* XXX */ 798 error = (*so->so_proto->pr_usrreqs->pr_abort)(so); 799 refs = --so->so_aborting; /* XXX */ 800 if (error || (refs == 0)) { 801 sofree(so); 802 } else { 803 sounlock(so); 804 } 805 return error; 806 } 807 808 int 809 soaccept(struct socket *so, struct sockaddr *nam) 810 { 811 int error; 812 813 KASSERT(solocked(so)); 814 KASSERT((so->so_state & SS_NOFDREF) != 0); 815 816 so->so_state &= ~SS_NOFDREF; 817 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 818 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 819 error = (*so->so_proto->pr_usrreqs->pr_accept)(so, nam); 820 else 821 error = ECONNABORTED; 822 823 return error; 824 } 825 826 int 827 soconnect(struct socket *so, struct sockaddr *nam, struct lwp *l) 828 { 829 int error; 830 831 KASSERT(solocked(so)); 832 833 if (so->so_options & SO_ACCEPTCONN) 834 return EOPNOTSUPP; 835 /* 836 * If protocol is connection-based, can only connect once. 837 * Otherwise, if connected, try to disconnect first. 838 * This allows user to disconnect by connecting to, e.g., 839 * a null address. 840 */ 841 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 842 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 843 (error = sodisconnect(so)))) { 844 error = EISCONN; 845 } else { 846 if (nam->sa_family != so->so_proto->pr_domain->dom_family) { 847 return EAFNOSUPPORT; 848 } 849 error = (*so->so_proto->pr_usrreqs->pr_connect)(so, nam, l); 850 } 851 852 return error; 853 } 854 855 int 856 soconnect2(struct socket *so1, struct socket *so2) 857 { 858 KASSERT(solocked2(so1, so2)); 859 860 return (*so1->so_proto->pr_usrreqs->pr_connect2)(so1, so2); 861 } 862 863 int 864 sodisconnect(struct socket *so) 865 { 866 int error; 867 868 KASSERT(solocked(so)); 869 870 if ((so->so_state & SS_ISCONNECTED) == 0) { 871 error = ENOTCONN; 872 } else if (so->so_state & SS_ISDISCONNECTING) { 873 error = EALREADY; 874 } else { 875 error = (*so->so_proto->pr_usrreqs->pr_disconnect)(so); 876 } 877 return (error); 878 } 879 880 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 881 /* 882 * Send on a socket. 883 * If send must go all at once and message is larger than 884 * send buffering, then hard error. 885 * Lock against other senders. 886 * If must go all at once and not enough room now, then 887 * inform user that this would block and do nothing. 888 * Otherwise, if nonblocking, send as much as possible. 889 * The data to be sent is described by "uio" if nonzero, 890 * otherwise by the mbuf chain "top" (which must be null 891 * if uio is not). Data provided in mbuf chain must be small 892 * enough to send all at once. 893 * 894 * Returns nonzero on error, timeout or signal; callers 895 * must check for short counts if EINTR/ERESTART are returned. 896 * Data and control buffers are freed on return. 897 */ 898 int 899 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 900 struct mbuf *top, struct mbuf *control, int flags, struct lwp *l) 901 { 902 struct mbuf **mp, *m; 903 long space, len, resid, clen, mlen; 904 int error, s, dontroute, atomic; 905 short wakeup_state = 0; 906 907 clen = 0; 908 909 /* 910 * solock() provides atomicity of access. splsoftnet() prevents 911 * protocol processing soft interrupts from interrupting us and 912 * blocking (expensive). 913 */ 914 s = splsoftnet(); 915 solock(so); 916 atomic = sosendallatonce(so) || top; 917 if (uio) 918 resid = uio->uio_resid; 919 else 920 resid = top->m_pkthdr.len; 921 /* 922 * In theory resid should be unsigned. 923 * However, space must be signed, as it might be less than 0 924 * if we over-committed, and we must use a signed comparison 925 * of space and resid. On the other hand, a negative resid 926 * causes us to loop sending 0-length segments to the protocol. 927 */ 928 if (resid < 0) { 929 error = EINVAL; 930 goto out; 931 } 932 dontroute = 933 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 934 (so->so_proto->pr_flags & PR_ATOMIC); 935 l->l_ru.ru_msgsnd++; 936 if (control) 937 clen = control->m_len; 938 restart: 939 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 940 goto out; 941 do { 942 if (so->so_state & SS_CANTSENDMORE) { 943 error = EPIPE; 944 goto release; 945 } 946 if (so->so_error) { 947 error = so->so_error; 948 so->so_error = 0; 949 goto release; 950 } 951 if ((so->so_state & SS_ISCONNECTED) == 0) { 952 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 953 if (resid || clen == 0) { 954 error = ENOTCONN; 955 goto release; 956 } 957 } else if (addr == NULL) { 958 error = EDESTADDRREQ; 959 goto release; 960 } 961 } 962 space = sbspace(&so->so_snd); 963 if (flags & MSG_OOB) 964 space += 1024; 965 if ((atomic && resid > so->so_snd.sb_hiwat) || 966 clen > so->so_snd.sb_hiwat) { 967 error = EMSGSIZE; 968 goto release; 969 } 970 if (space < resid + clen && 971 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 972 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) { 973 error = EWOULDBLOCK; 974 goto release; 975 } 976 sbunlock(&so->so_snd); 977 if (wakeup_state & SS_RESTARTSYS) { 978 error = ERESTART; 979 goto out; 980 } 981 error = sbwait(&so->so_snd); 982 if (error) 983 goto out; 984 wakeup_state = so->so_state; 985 goto restart; 986 } 987 wakeup_state = 0; 988 mp = ⊤ 989 space -= clen; 990 do { 991 if (uio == NULL) { 992 /* 993 * Data is prepackaged in "top". 994 */ 995 resid = 0; 996 if (flags & MSG_EOR) 997 top->m_flags |= M_EOR; 998 } else do { 999 sounlock(so); 1000 splx(s); 1001 if (top == NULL) { 1002 m = m_gethdr(M_WAIT, MT_DATA); 1003 mlen = MHLEN; 1004 m->m_pkthdr.len = 0; 1005 m_reset_rcvif(m); 1006 } else { 1007 m = m_get(M_WAIT, MT_DATA); 1008 mlen = MLEN; 1009 } 1010 MCLAIM(m, so->so_snd.sb_mowner); 1011 if (sock_loan_thresh >= 0 && 1012 uio->uio_iov->iov_len >= sock_loan_thresh && 1013 space >= sock_loan_thresh && 1014 (len = sosend_loan(so, uio, m, 1015 space)) != 0) { 1016 SOSEND_COUNTER_INCR(&sosend_loan_big); 1017 space -= len; 1018 goto have_data; 1019 } 1020 if (resid >= MINCLSIZE && space >= MCLBYTES) { 1021 SOSEND_COUNTER_INCR(&sosend_copy_big); 1022 m_clget(m, M_DONTWAIT); 1023 if ((m->m_flags & M_EXT) == 0) 1024 goto nopages; 1025 mlen = MCLBYTES; 1026 if (atomic && top == 0) { 1027 len = lmin(MCLBYTES - max_hdr, 1028 resid); 1029 m->m_data += max_hdr; 1030 } else 1031 len = lmin(MCLBYTES, resid); 1032 space -= len; 1033 } else { 1034 nopages: 1035 SOSEND_COUNTER_INCR(&sosend_copy_small); 1036 len = lmin(lmin(mlen, resid), space); 1037 space -= len; 1038 /* 1039 * For datagram protocols, leave room 1040 * for protocol headers in first mbuf. 1041 */ 1042 if (atomic && top == 0 && len < mlen) 1043 MH_ALIGN(m, len); 1044 } 1045 error = uiomove(mtod(m, void *), (int)len, uio); 1046 have_data: 1047 resid = uio->uio_resid; 1048 m->m_len = len; 1049 *mp = m; 1050 top->m_pkthdr.len += len; 1051 s = splsoftnet(); 1052 solock(so); 1053 if (error != 0) 1054 goto release; 1055 mp = &m->m_next; 1056 if (resid <= 0) { 1057 if (flags & MSG_EOR) 1058 top->m_flags |= M_EOR; 1059 break; 1060 } 1061 } while (space > 0 && atomic); 1062 1063 if (so->so_state & SS_CANTSENDMORE) { 1064 error = EPIPE; 1065 goto release; 1066 } 1067 if (dontroute) 1068 so->so_options |= SO_DONTROUTE; 1069 if (resid > 0) 1070 so->so_state |= SS_MORETOCOME; 1071 if (flags & MSG_OOB) { 1072 error = (*so->so_proto->pr_usrreqs->pr_sendoob)( 1073 so, top, control); 1074 } else { 1075 error = (*so->so_proto->pr_usrreqs->pr_send)(so, 1076 top, addr, control, l); 1077 } 1078 if (dontroute) 1079 so->so_options &= ~SO_DONTROUTE; 1080 if (resid > 0) 1081 so->so_state &= ~SS_MORETOCOME; 1082 clen = 0; 1083 control = NULL; 1084 top = NULL; 1085 mp = ⊤ 1086 if (error != 0) 1087 goto release; 1088 } while (resid && space > 0); 1089 } while (resid); 1090 1091 release: 1092 sbunlock(&so->so_snd); 1093 out: 1094 sounlock(so); 1095 splx(s); 1096 if (top) 1097 m_freem(top); 1098 if (control) 1099 m_freem(control); 1100 return (error); 1101 } 1102 1103 /* 1104 * Following replacement or removal of the first mbuf on the first 1105 * mbuf chain of a socket buffer, push necessary state changes back 1106 * into the socket buffer so that other consumers see the values 1107 * consistently. 'nextrecord' is the callers locally stored value of 1108 * the original value of sb->sb_mb->m_nextpkt which must be restored 1109 * when the lead mbuf changes. NOTE: 'nextrecord' may be NULL. 1110 */ 1111 static void 1112 sbsync(struct sockbuf *sb, struct mbuf *nextrecord) 1113 { 1114 1115 KASSERT(solocked(sb->sb_so)); 1116 1117 /* 1118 * First, update for the new value of nextrecord. If necessary, 1119 * make it the first record. 1120 */ 1121 if (sb->sb_mb != NULL) 1122 sb->sb_mb->m_nextpkt = nextrecord; 1123 else 1124 sb->sb_mb = nextrecord; 1125 1126 /* 1127 * Now update any dependent socket buffer fields to reflect 1128 * the new state. This is an inline of SB_EMPTY_FIXUP, with 1129 * the addition of a second clause that takes care of the 1130 * case where sb_mb has been updated, but remains the last 1131 * record. 1132 */ 1133 if (sb->sb_mb == NULL) { 1134 sb->sb_mbtail = NULL; 1135 sb->sb_lastrecord = NULL; 1136 } else if (sb->sb_mb->m_nextpkt == NULL) 1137 sb->sb_lastrecord = sb->sb_mb; 1138 } 1139 1140 /* 1141 * Implement receive operations on a socket. 1142 * We depend on the way that records are added to the sockbuf 1143 * by sbappend*. In particular, each record (mbufs linked through m_next) 1144 * must begin with an address if the protocol so specifies, 1145 * followed by an optional mbuf or mbufs containing ancillary data, 1146 * and then zero or more mbufs of data. 1147 * In order to avoid blocking network interrupts for the entire time here, 1148 * we splx() while doing the actual copy to user space. 1149 * Although the sockbuf is locked, new data may still be appended, 1150 * and thus we must maintain consistency of the sockbuf during that time. 1151 * 1152 * The caller may receive the data as a single mbuf chain by supplying 1153 * an mbuf **mp0 for use in returning the chain. The uio is then used 1154 * only for the count in uio_resid. 1155 */ 1156 int 1157 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 1158 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1159 { 1160 struct lwp *l = curlwp; 1161 struct mbuf *m, **mp, *mt; 1162 size_t len, offset, moff, orig_resid; 1163 int atomic, flags, error, s, type; 1164 const struct protosw *pr; 1165 struct mbuf *nextrecord; 1166 int mbuf_removed = 0; 1167 const struct domain *dom; 1168 short wakeup_state = 0; 1169 1170 pr = so->so_proto; 1171 atomic = pr->pr_flags & PR_ATOMIC; 1172 dom = pr->pr_domain; 1173 mp = mp0; 1174 type = 0; 1175 orig_resid = uio->uio_resid; 1176 1177 if (paddr != NULL) 1178 *paddr = NULL; 1179 if (controlp != NULL) 1180 *controlp = NULL; 1181 if (flagsp != NULL) 1182 flags = *flagsp &~ MSG_EOR; 1183 else 1184 flags = 0; 1185 1186 if (flags & MSG_OOB) { 1187 m = m_get(M_WAIT, MT_DATA); 1188 solock(so); 1189 error = (*pr->pr_usrreqs->pr_recvoob)(so, m, flags & MSG_PEEK); 1190 sounlock(so); 1191 if (error) 1192 goto bad; 1193 do { 1194 error = uiomove(mtod(m, void *), 1195 MIN(uio->uio_resid, m->m_len), uio); 1196 m = m_free(m); 1197 } while (uio->uio_resid > 0 && error == 0 && m); 1198 bad: 1199 if (m != NULL) 1200 m_freem(m); 1201 return error; 1202 } 1203 if (mp != NULL) 1204 *mp = NULL; 1205 1206 /* 1207 * solock() provides atomicity of access. splsoftnet() prevents 1208 * protocol processing soft interrupts from interrupting us and 1209 * blocking (expensive). 1210 */ 1211 s = splsoftnet(); 1212 solock(so); 1213 restart: 1214 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) { 1215 sounlock(so); 1216 splx(s); 1217 return error; 1218 } 1219 1220 m = so->so_rcv.sb_mb; 1221 /* 1222 * If we have less data than requested, block awaiting more 1223 * (subject to any timeout) if: 1224 * 1. the current count is less than the low water mark, 1225 * 2. MSG_WAITALL is set, and it is possible to do the entire 1226 * receive operation at once if we block (resid <= hiwat), or 1227 * 3. MSG_DONTWAIT is not set. 1228 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1229 * we have to do the receive in sections, and thus risk returning 1230 * a short count if a timeout or signal occurs after we start. 1231 */ 1232 if (m == NULL || 1233 ((flags & MSG_DONTWAIT) == 0 && 1234 so->so_rcv.sb_cc < uio->uio_resid && 1235 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1236 ((flags & MSG_WAITALL) && 1237 uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1238 m->m_nextpkt == NULL && !atomic)) { 1239 #ifdef DIAGNOSTIC 1240 if (m == NULL && so->so_rcv.sb_cc) 1241 panic("receive 1"); 1242 #endif 1243 if (so->so_error) { 1244 if (m != NULL) 1245 goto dontblock; 1246 error = so->so_error; 1247 if ((flags & MSG_PEEK) == 0) 1248 so->so_error = 0; 1249 goto release; 1250 } 1251 if (so->so_state & SS_CANTRCVMORE) { 1252 if (m != NULL) 1253 goto dontblock; 1254 else 1255 goto release; 1256 } 1257 for (; m != NULL; m = m->m_next) 1258 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1259 m = so->so_rcv.sb_mb; 1260 goto dontblock; 1261 } 1262 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1263 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1264 error = ENOTCONN; 1265 goto release; 1266 } 1267 if (uio->uio_resid == 0) 1268 goto release; 1269 if ((so->so_state & SS_NBIO) || 1270 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1271 error = EWOULDBLOCK; 1272 goto release; 1273 } 1274 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1275 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1276 sbunlock(&so->so_rcv); 1277 if (wakeup_state & SS_RESTARTSYS) 1278 error = ERESTART; 1279 else 1280 error = sbwait(&so->so_rcv); 1281 if (error != 0) { 1282 sounlock(so); 1283 splx(s); 1284 return error; 1285 } 1286 wakeup_state = so->so_state; 1287 goto restart; 1288 } 1289 dontblock: 1290 /* 1291 * On entry here, m points to the first record of the socket buffer. 1292 * From this point onward, we maintain 'nextrecord' as a cache of the 1293 * pointer to the next record in the socket buffer. We must keep the 1294 * various socket buffer pointers and local stack versions of the 1295 * pointers in sync, pushing out modifications before dropping the 1296 * socket lock, and re-reading them when picking it up. 1297 * 1298 * Otherwise, we will race with the network stack appending new data 1299 * or records onto the socket buffer by using inconsistent/stale 1300 * versions of the field, possibly resulting in socket buffer 1301 * corruption. 1302 * 1303 * By holding the high-level sblock(), we prevent simultaneous 1304 * readers from pulling off the front of the socket buffer. 1305 */ 1306 if (l != NULL) 1307 l->l_ru.ru_msgrcv++; 1308 KASSERT(m == so->so_rcv.sb_mb); 1309 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1310 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1311 nextrecord = m->m_nextpkt; 1312 if (pr->pr_flags & PR_ADDR) { 1313 #ifdef DIAGNOSTIC 1314 if (m->m_type != MT_SONAME) 1315 panic("receive 1a"); 1316 #endif 1317 orig_resid = 0; 1318 if (flags & MSG_PEEK) { 1319 if (paddr) 1320 *paddr = m_copy(m, 0, m->m_len); 1321 m = m->m_next; 1322 } else { 1323 sbfree(&so->so_rcv, m); 1324 mbuf_removed = 1; 1325 if (paddr != NULL) { 1326 *paddr = m; 1327 so->so_rcv.sb_mb = m->m_next; 1328 m->m_next = NULL; 1329 m = so->so_rcv.sb_mb; 1330 } else { 1331 m = so->so_rcv.sb_mb = m_free(m); 1332 } 1333 sbsync(&so->so_rcv, nextrecord); 1334 } 1335 } 1336 if (pr->pr_flags & PR_ADDR_OPT) { 1337 /* 1338 * For SCTP we may be getting a 1339 * whole message OR a partial delivery. 1340 */ 1341 if (m->m_type == MT_SONAME) { 1342 orig_resid = 0; 1343 if (flags & MSG_PEEK) { 1344 if (paddr) 1345 *paddr = m_copy(m, 0, m->m_len); 1346 m = m->m_next; 1347 } else { 1348 sbfree(&so->so_rcv, m); 1349 if (paddr) { 1350 *paddr = m; 1351 so->so_rcv.sb_mb = m->m_next; 1352 m->m_next = 0; 1353 m = so->so_rcv.sb_mb; 1354 } else { 1355 m = so->so_rcv.sb_mb = m_free(m); 1356 } 1357 } 1358 } 1359 } 1360 1361 /* 1362 * Process one or more MT_CONTROL mbufs present before any data mbufs 1363 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 1364 * just copy the data; if !MSG_PEEK, we call into the protocol to 1365 * perform externalization (or freeing if controlp == NULL). 1366 */ 1367 if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) { 1368 struct mbuf *cm = NULL, *cmn; 1369 struct mbuf **cme = &cm; 1370 1371 do { 1372 if (flags & MSG_PEEK) { 1373 if (controlp != NULL) { 1374 *controlp = m_copy(m, 0, m->m_len); 1375 controlp = &(*controlp)->m_next; 1376 } 1377 m = m->m_next; 1378 } else { 1379 sbfree(&so->so_rcv, m); 1380 so->so_rcv.sb_mb = m->m_next; 1381 m->m_next = NULL; 1382 *cme = m; 1383 cme = &(*cme)->m_next; 1384 m = so->so_rcv.sb_mb; 1385 } 1386 } while (m != NULL && m->m_type == MT_CONTROL); 1387 if ((flags & MSG_PEEK) == 0) 1388 sbsync(&so->so_rcv, nextrecord); 1389 for (; cm != NULL; cm = cmn) { 1390 cmn = cm->m_next; 1391 cm->m_next = NULL; 1392 type = mtod(cm, struct cmsghdr *)->cmsg_type; 1393 if (controlp != NULL) { 1394 if (dom->dom_externalize != NULL && 1395 type == SCM_RIGHTS) { 1396 sounlock(so); 1397 splx(s); 1398 error = (*dom->dom_externalize)(cm, l, 1399 (flags & MSG_CMSG_CLOEXEC) ? 1400 O_CLOEXEC : 0); 1401 s = splsoftnet(); 1402 solock(so); 1403 } 1404 *controlp = cm; 1405 while (*controlp != NULL) 1406 controlp = &(*controlp)->m_next; 1407 } else { 1408 /* 1409 * Dispose of any SCM_RIGHTS message that went 1410 * through the read path rather than recv. 1411 */ 1412 if (dom->dom_dispose != NULL && 1413 type == SCM_RIGHTS) { 1414 sounlock(so); 1415 (*dom->dom_dispose)(cm); 1416 solock(so); 1417 } 1418 m_freem(cm); 1419 } 1420 } 1421 if (m != NULL) 1422 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 1423 else 1424 nextrecord = so->so_rcv.sb_mb; 1425 orig_resid = 0; 1426 } 1427 1428 /* If m is non-NULL, we have some data to read. */ 1429 if (__predict_true(m != NULL)) { 1430 type = m->m_type; 1431 if (type == MT_OOBDATA) 1432 flags |= MSG_OOB; 1433 } 1434 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1435 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1436 1437 moff = 0; 1438 offset = 0; 1439 while (m != NULL && uio->uio_resid > 0 && error == 0) { 1440 if (m->m_type == MT_OOBDATA) { 1441 if (type != MT_OOBDATA) 1442 break; 1443 } else if (type == MT_OOBDATA) 1444 break; 1445 #ifdef DIAGNOSTIC 1446 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1447 panic("receive 3"); 1448 #endif 1449 so->so_state &= ~SS_RCVATMARK; 1450 wakeup_state = 0; 1451 len = uio->uio_resid; 1452 if (so->so_oobmark && len > so->so_oobmark - offset) 1453 len = so->so_oobmark - offset; 1454 if (len > m->m_len - moff) 1455 len = m->m_len - moff; 1456 /* 1457 * If mp is set, just pass back the mbufs. 1458 * Otherwise copy them out via the uio, then free. 1459 * Sockbuf must be consistent here (points to current mbuf, 1460 * it points to next record) when we drop priority; 1461 * we must note any additions to the sockbuf when we 1462 * block interrupts again. 1463 */ 1464 if (mp == NULL) { 1465 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1466 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1467 sounlock(so); 1468 splx(s); 1469 error = uiomove(mtod(m, char *) + moff, len, uio); 1470 s = splsoftnet(); 1471 solock(so); 1472 if (error != 0) { 1473 /* 1474 * If any part of the record has been removed 1475 * (such as the MT_SONAME mbuf, which will 1476 * happen when PR_ADDR, and thus also 1477 * PR_ATOMIC, is set), then drop the entire 1478 * record to maintain the atomicity of the 1479 * receive operation. 1480 * 1481 * This avoids a later panic("receive 1a") 1482 * when compiled with DIAGNOSTIC. 1483 */ 1484 if (m && mbuf_removed && atomic) 1485 (void) sbdroprecord(&so->so_rcv); 1486 1487 goto release; 1488 } 1489 } else 1490 uio->uio_resid -= len; 1491 if (len == m->m_len - moff) { 1492 if (m->m_flags & M_EOR) 1493 flags |= MSG_EOR; 1494 #ifdef SCTP 1495 if (m->m_flags & M_NOTIFICATION) 1496 flags |= MSG_NOTIFICATION; 1497 #endif /* SCTP */ 1498 if (flags & MSG_PEEK) { 1499 m = m->m_next; 1500 moff = 0; 1501 } else { 1502 nextrecord = m->m_nextpkt; 1503 sbfree(&so->so_rcv, m); 1504 if (mp) { 1505 *mp = m; 1506 mp = &m->m_next; 1507 so->so_rcv.sb_mb = m = m->m_next; 1508 *mp = NULL; 1509 } else { 1510 m = so->so_rcv.sb_mb = m_free(m); 1511 } 1512 /* 1513 * If m != NULL, we also know that 1514 * so->so_rcv.sb_mb != NULL. 1515 */ 1516 KASSERT(so->so_rcv.sb_mb == m); 1517 if (m) { 1518 m->m_nextpkt = nextrecord; 1519 if (nextrecord == NULL) 1520 so->so_rcv.sb_lastrecord = m; 1521 } else { 1522 so->so_rcv.sb_mb = nextrecord; 1523 SB_EMPTY_FIXUP(&so->so_rcv); 1524 } 1525 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1526 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1527 } 1528 } else if (flags & MSG_PEEK) 1529 moff += len; 1530 else { 1531 if (mp != NULL) { 1532 mt = m_copym(m, 0, len, M_NOWAIT); 1533 if (__predict_false(mt == NULL)) { 1534 sounlock(so); 1535 mt = m_copym(m, 0, len, M_WAIT); 1536 solock(so); 1537 } 1538 *mp = mt; 1539 } 1540 m->m_data += len; 1541 m->m_len -= len; 1542 so->so_rcv.sb_cc -= len; 1543 } 1544 if (so->so_oobmark) { 1545 if ((flags & MSG_PEEK) == 0) { 1546 so->so_oobmark -= len; 1547 if (so->so_oobmark == 0) { 1548 so->so_state |= SS_RCVATMARK; 1549 break; 1550 } 1551 } else { 1552 offset += len; 1553 if (offset == so->so_oobmark) 1554 break; 1555 } 1556 } 1557 if (flags & MSG_EOR) 1558 break; 1559 /* 1560 * If the MSG_WAITALL flag is set (for non-atomic socket), 1561 * we must not quit until "uio->uio_resid == 0" or an error 1562 * termination. If a signal/timeout occurs, return 1563 * with a short count but without error. 1564 * Keep sockbuf locked against other readers. 1565 */ 1566 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 1567 !sosendallatonce(so) && !nextrecord) { 1568 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1569 break; 1570 /* 1571 * If we are peeking and the socket receive buffer is 1572 * full, stop since we can't get more data to peek at. 1573 */ 1574 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1575 break; 1576 /* 1577 * If we've drained the socket buffer, tell the 1578 * protocol in case it needs to do something to 1579 * get it filled again. 1580 */ 1581 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1582 (*pr->pr_usrreqs->pr_rcvd)(so, flags, l); 1583 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1584 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1585 if (wakeup_state & SS_RESTARTSYS) 1586 error = ERESTART; 1587 else 1588 error = sbwait(&so->so_rcv); 1589 if (error != 0) { 1590 sbunlock(&so->so_rcv); 1591 sounlock(so); 1592 splx(s); 1593 return 0; 1594 } 1595 if ((m = so->so_rcv.sb_mb) != NULL) 1596 nextrecord = m->m_nextpkt; 1597 wakeup_state = so->so_state; 1598 } 1599 } 1600 1601 if (m && atomic) { 1602 flags |= MSG_TRUNC; 1603 if ((flags & MSG_PEEK) == 0) 1604 (void) sbdroprecord(&so->so_rcv); 1605 } 1606 if ((flags & MSG_PEEK) == 0) { 1607 if (m == NULL) { 1608 /* 1609 * First part is an inline SB_EMPTY_FIXUP(). Second 1610 * part makes sure sb_lastrecord is up-to-date if 1611 * there is still data in the socket buffer. 1612 */ 1613 so->so_rcv.sb_mb = nextrecord; 1614 if (so->so_rcv.sb_mb == NULL) { 1615 so->so_rcv.sb_mbtail = NULL; 1616 so->so_rcv.sb_lastrecord = NULL; 1617 } else if (nextrecord->m_nextpkt == NULL) 1618 so->so_rcv.sb_lastrecord = nextrecord; 1619 } 1620 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1621 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1622 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1623 (*pr->pr_usrreqs->pr_rcvd)(so, flags, l); 1624 } 1625 if (orig_resid == uio->uio_resid && orig_resid && 1626 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1627 sbunlock(&so->so_rcv); 1628 goto restart; 1629 } 1630 1631 if (flagsp != NULL) 1632 *flagsp |= flags; 1633 release: 1634 sbunlock(&so->so_rcv); 1635 sounlock(so); 1636 splx(s); 1637 return error; 1638 } 1639 1640 int 1641 soshutdown(struct socket *so, int how) 1642 { 1643 const struct protosw *pr; 1644 int error; 1645 1646 KASSERT(solocked(so)); 1647 1648 pr = so->so_proto; 1649 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1650 return (EINVAL); 1651 1652 if (how == SHUT_RD || how == SHUT_RDWR) { 1653 sorflush(so); 1654 error = 0; 1655 } 1656 if (how == SHUT_WR || how == SHUT_RDWR) 1657 error = (*pr->pr_usrreqs->pr_shutdown)(so); 1658 1659 return error; 1660 } 1661 1662 void 1663 sorestart(struct socket *so) 1664 { 1665 /* 1666 * An application has called close() on an fd on which another 1667 * of its threads has called a socket system call. 1668 * Mark this and wake everyone up, and code that would block again 1669 * instead returns ERESTART. 1670 * On system call re-entry the fd is validated and EBADF returned. 1671 * Any other fd will block again on the 2nd syscall. 1672 */ 1673 solock(so); 1674 so->so_state |= SS_RESTARTSYS; 1675 cv_broadcast(&so->so_cv); 1676 cv_broadcast(&so->so_snd.sb_cv); 1677 cv_broadcast(&so->so_rcv.sb_cv); 1678 sounlock(so); 1679 } 1680 1681 void 1682 sorflush(struct socket *so) 1683 { 1684 struct sockbuf *sb, asb; 1685 const struct protosw *pr; 1686 1687 KASSERT(solocked(so)); 1688 1689 sb = &so->so_rcv; 1690 pr = so->so_proto; 1691 socantrcvmore(so); 1692 sb->sb_flags |= SB_NOINTR; 1693 (void )sblock(sb, M_WAITOK); 1694 sbunlock(sb); 1695 asb = *sb; 1696 /* 1697 * Clear most of the sockbuf structure, but leave some of the 1698 * fields valid. 1699 */ 1700 memset(&sb->sb_startzero, 0, 1701 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1702 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) { 1703 sounlock(so); 1704 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1705 solock(so); 1706 } 1707 sbrelease(&asb, so); 1708 } 1709 1710 /* 1711 * internal set SOL_SOCKET options 1712 */ 1713 static int 1714 sosetopt1(struct socket *so, const struct sockopt *sopt) 1715 { 1716 int error = EINVAL, opt; 1717 int optval = 0; /* XXX: gcc */ 1718 struct linger l; 1719 struct timeval tv; 1720 1721 switch ((opt = sopt->sopt_name)) { 1722 1723 case SO_ACCEPTFILTER: 1724 error = accept_filt_setopt(so, sopt); 1725 KASSERT(solocked(so)); 1726 break; 1727 1728 case SO_LINGER: 1729 error = sockopt_get(sopt, &l, sizeof(l)); 1730 solock(so); 1731 if (error) 1732 break; 1733 if (l.l_linger < 0 || l.l_linger > USHRT_MAX || 1734 l.l_linger > (INT_MAX / hz)) { 1735 error = EDOM; 1736 break; 1737 } 1738 so->so_linger = l.l_linger; 1739 if (l.l_onoff) 1740 so->so_options |= SO_LINGER; 1741 else 1742 so->so_options &= ~SO_LINGER; 1743 break; 1744 1745 case SO_DEBUG: 1746 case SO_KEEPALIVE: 1747 case SO_DONTROUTE: 1748 case SO_USELOOPBACK: 1749 case SO_BROADCAST: 1750 case SO_REUSEADDR: 1751 case SO_REUSEPORT: 1752 case SO_OOBINLINE: 1753 case SO_TIMESTAMP: 1754 case SO_NOSIGPIPE: 1755 #ifdef SO_OTIMESTAMP 1756 case SO_OTIMESTAMP: 1757 #endif 1758 error = sockopt_getint(sopt, &optval); 1759 solock(so); 1760 if (error) 1761 break; 1762 if (optval) 1763 so->so_options |= opt; 1764 else 1765 so->so_options &= ~opt; 1766 break; 1767 1768 case SO_SNDBUF: 1769 case SO_RCVBUF: 1770 case SO_SNDLOWAT: 1771 case SO_RCVLOWAT: 1772 error = sockopt_getint(sopt, &optval); 1773 solock(so); 1774 if (error) 1775 break; 1776 1777 /* 1778 * Values < 1 make no sense for any of these 1779 * options, so disallow them. 1780 */ 1781 if (optval < 1) { 1782 error = EINVAL; 1783 break; 1784 } 1785 1786 switch (opt) { 1787 case SO_SNDBUF: 1788 if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) { 1789 error = ENOBUFS; 1790 break; 1791 } 1792 so->so_snd.sb_flags &= ~SB_AUTOSIZE; 1793 break; 1794 1795 case SO_RCVBUF: 1796 if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) { 1797 error = ENOBUFS; 1798 break; 1799 } 1800 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1801 break; 1802 1803 /* 1804 * Make sure the low-water is never greater than 1805 * the high-water. 1806 */ 1807 case SO_SNDLOWAT: 1808 if (optval > so->so_snd.sb_hiwat) 1809 optval = so->so_snd.sb_hiwat; 1810 1811 so->so_snd.sb_lowat = optval; 1812 break; 1813 1814 case SO_RCVLOWAT: 1815 if (optval > so->so_rcv.sb_hiwat) 1816 optval = so->so_rcv.sb_hiwat; 1817 1818 so->so_rcv.sb_lowat = optval; 1819 break; 1820 } 1821 break; 1822 1823 #ifdef COMPAT_50 1824 case SO_OSNDTIMEO: 1825 case SO_ORCVTIMEO: { 1826 struct timeval50 otv; 1827 error = sockopt_get(sopt, &otv, sizeof(otv)); 1828 if (error) { 1829 solock(so); 1830 break; 1831 } 1832 timeval50_to_timeval(&otv, &tv); 1833 opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO; 1834 error = 0; 1835 /*FALLTHROUGH*/ 1836 } 1837 #endif /* COMPAT_50 */ 1838 1839 case SO_SNDTIMEO: 1840 case SO_RCVTIMEO: 1841 if (error) 1842 error = sockopt_get(sopt, &tv, sizeof(tv)); 1843 solock(so); 1844 if (error) 1845 break; 1846 1847 if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) { 1848 error = EDOM; 1849 break; 1850 } 1851 1852 optval = tv.tv_sec * hz + tv.tv_usec / tick; 1853 if (optval == 0 && tv.tv_usec != 0) 1854 optval = 1; 1855 1856 switch (opt) { 1857 case SO_SNDTIMEO: 1858 so->so_snd.sb_timeo = optval; 1859 break; 1860 case SO_RCVTIMEO: 1861 so->so_rcv.sb_timeo = optval; 1862 break; 1863 } 1864 break; 1865 1866 default: 1867 solock(so); 1868 error = ENOPROTOOPT; 1869 break; 1870 } 1871 KASSERT(solocked(so)); 1872 return error; 1873 } 1874 1875 int 1876 sosetopt(struct socket *so, struct sockopt *sopt) 1877 { 1878 int error, prerr; 1879 1880 if (sopt->sopt_level == SOL_SOCKET) { 1881 error = sosetopt1(so, sopt); 1882 KASSERT(solocked(so)); 1883 } else { 1884 error = ENOPROTOOPT; 1885 solock(so); 1886 } 1887 1888 if ((error == 0 || error == ENOPROTOOPT) && 1889 so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) { 1890 /* give the protocol stack a shot */ 1891 prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt); 1892 if (prerr == 0) 1893 error = 0; 1894 else if (prerr != ENOPROTOOPT) 1895 error = prerr; 1896 } 1897 sounlock(so); 1898 return error; 1899 } 1900 1901 /* 1902 * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt() 1903 */ 1904 int 1905 so_setsockopt(struct lwp *l, struct socket *so, int level, int name, 1906 const void *val, size_t valsize) 1907 { 1908 struct sockopt sopt; 1909 int error; 1910 1911 KASSERT(valsize == 0 || val != NULL); 1912 1913 sockopt_init(&sopt, level, name, valsize); 1914 sockopt_set(&sopt, val, valsize); 1915 1916 error = sosetopt(so, &sopt); 1917 1918 sockopt_destroy(&sopt); 1919 1920 return error; 1921 } 1922 1923 /* 1924 * internal get SOL_SOCKET options 1925 */ 1926 static int 1927 sogetopt1(struct socket *so, struct sockopt *sopt) 1928 { 1929 int error, optval, opt; 1930 struct linger l; 1931 struct timeval tv; 1932 1933 switch ((opt = sopt->sopt_name)) { 1934 1935 case SO_ACCEPTFILTER: 1936 error = accept_filt_getopt(so, sopt); 1937 break; 1938 1939 case SO_LINGER: 1940 l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0; 1941 l.l_linger = so->so_linger; 1942 1943 error = sockopt_set(sopt, &l, sizeof(l)); 1944 break; 1945 1946 case SO_USELOOPBACK: 1947 case SO_DONTROUTE: 1948 case SO_DEBUG: 1949 case SO_KEEPALIVE: 1950 case SO_REUSEADDR: 1951 case SO_REUSEPORT: 1952 case SO_BROADCAST: 1953 case SO_OOBINLINE: 1954 case SO_TIMESTAMP: 1955 case SO_NOSIGPIPE: 1956 #ifdef SO_OTIMESTAMP 1957 case SO_OTIMESTAMP: 1958 #endif 1959 case SO_ACCEPTCONN: 1960 error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0); 1961 break; 1962 1963 case SO_TYPE: 1964 error = sockopt_setint(sopt, so->so_type); 1965 break; 1966 1967 case SO_ERROR: 1968 error = sockopt_setint(sopt, so->so_error); 1969 so->so_error = 0; 1970 break; 1971 1972 case SO_SNDBUF: 1973 error = sockopt_setint(sopt, so->so_snd.sb_hiwat); 1974 break; 1975 1976 case SO_RCVBUF: 1977 error = sockopt_setint(sopt, so->so_rcv.sb_hiwat); 1978 break; 1979 1980 case SO_SNDLOWAT: 1981 error = sockopt_setint(sopt, so->so_snd.sb_lowat); 1982 break; 1983 1984 case SO_RCVLOWAT: 1985 error = sockopt_setint(sopt, so->so_rcv.sb_lowat); 1986 break; 1987 1988 #ifdef COMPAT_50 1989 case SO_OSNDTIMEO: 1990 case SO_ORCVTIMEO: { 1991 struct timeval50 otv; 1992 1993 optval = (opt == SO_OSNDTIMEO ? 1994 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1995 1996 otv.tv_sec = optval / hz; 1997 otv.tv_usec = (optval % hz) * tick; 1998 1999 error = sockopt_set(sopt, &otv, sizeof(otv)); 2000 break; 2001 } 2002 #endif /* COMPAT_50 */ 2003 2004 case SO_SNDTIMEO: 2005 case SO_RCVTIMEO: 2006 optval = (opt == SO_SNDTIMEO ? 2007 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 2008 2009 tv.tv_sec = optval / hz; 2010 tv.tv_usec = (optval % hz) * tick; 2011 2012 error = sockopt_set(sopt, &tv, sizeof(tv)); 2013 break; 2014 2015 case SO_OVERFLOWED: 2016 error = sockopt_setint(sopt, so->so_rcv.sb_overflowed); 2017 break; 2018 2019 default: 2020 error = ENOPROTOOPT; 2021 break; 2022 } 2023 2024 return (error); 2025 } 2026 2027 int 2028 sogetopt(struct socket *so, struct sockopt *sopt) 2029 { 2030 int error; 2031 2032 solock(so); 2033 if (sopt->sopt_level != SOL_SOCKET) { 2034 if (so->so_proto && so->so_proto->pr_ctloutput) { 2035 error = ((*so->so_proto->pr_ctloutput) 2036 (PRCO_GETOPT, so, sopt)); 2037 } else 2038 error = (ENOPROTOOPT); 2039 } else { 2040 error = sogetopt1(so, sopt); 2041 } 2042 sounlock(so); 2043 return (error); 2044 } 2045 2046 /* 2047 * alloc sockopt data buffer buffer 2048 * - will be released at destroy 2049 */ 2050 static int 2051 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag) 2052 { 2053 2054 KASSERT(sopt->sopt_size == 0); 2055 2056 if (len > sizeof(sopt->sopt_buf)) { 2057 sopt->sopt_data = kmem_zalloc(len, kmflag); 2058 if (sopt->sopt_data == NULL) 2059 return ENOMEM; 2060 } else 2061 sopt->sopt_data = sopt->sopt_buf; 2062 2063 sopt->sopt_size = len; 2064 return 0; 2065 } 2066 2067 /* 2068 * initialise sockopt storage 2069 * - MAY sleep during allocation 2070 */ 2071 void 2072 sockopt_init(struct sockopt *sopt, int level, int name, size_t size) 2073 { 2074 2075 memset(sopt, 0, sizeof(*sopt)); 2076 2077 sopt->sopt_level = level; 2078 sopt->sopt_name = name; 2079 (void)sockopt_alloc(sopt, size, KM_SLEEP); 2080 } 2081 2082 /* 2083 * destroy sockopt storage 2084 * - will release any held memory references 2085 */ 2086 void 2087 sockopt_destroy(struct sockopt *sopt) 2088 { 2089 2090 if (sopt->sopt_data != sopt->sopt_buf) 2091 kmem_free(sopt->sopt_data, sopt->sopt_size); 2092 2093 memset(sopt, 0, sizeof(*sopt)); 2094 } 2095 2096 /* 2097 * set sockopt value 2098 * - value is copied into sockopt 2099 * - memory is allocated when necessary, will not sleep 2100 */ 2101 int 2102 sockopt_set(struct sockopt *sopt, const void *buf, size_t len) 2103 { 2104 int error; 2105 2106 if (sopt->sopt_size == 0) { 2107 error = sockopt_alloc(sopt, len, KM_NOSLEEP); 2108 if (error) 2109 return error; 2110 } 2111 2112 KASSERT(sopt->sopt_size == len); 2113 memcpy(sopt->sopt_data, buf, len); 2114 return 0; 2115 } 2116 2117 /* 2118 * common case of set sockopt integer value 2119 */ 2120 int 2121 sockopt_setint(struct sockopt *sopt, int val) 2122 { 2123 2124 return sockopt_set(sopt, &val, sizeof(int)); 2125 } 2126 2127 /* 2128 * get sockopt value 2129 * - correct size must be given 2130 */ 2131 int 2132 sockopt_get(const struct sockopt *sopt, void *buf, size_t len) 2133 { 2134 2135 if (sopt->sopt_size != len) 2136 return EINVAL; 2137 2138 memcpy(buf, sopt->sopt_data, len); 2139 return 0; 2140 } 2141 2142 /* 2143 * common case of get sockopt integer value 2144 */ 2145 int 2146 sockopt_getint(const struct sockopt *sopt, int *valp) 2147 { 2148 2149 return sockopt_get(sopt, valp, sizeof(int)); 2150 } 2151 2152 /* 2153 * set sockopt value from mbuf 2154 * - ONLY for legacy code 2155 * - mbuf is released by sockopt 2156 * - will not sleep 2157 */ 2158 int 2159 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m) 2160 { 2161 size_t len; 2162 int error; 2163 2164 len = m_length(m); 2165 2166 if (sopt->sopt_size == 0) { 2167 error = sockopt_alloc(sopt, len, KM_NOSLEEP); 2168 if (error) 2169 return error; 2170 } 2171 2172 KASSERT(sopt->sopt_size == len); 2173 m_copydata(m, 0, len, sopt->sopt_data); 2174 m_freem(m); 2175 2176 return 0; 2177 } 2178 2179 /* 2180 * get sockopt value into mbuf 2181 * - ONLY for legacy code 2182 * - mbuf to be released by the caller 2183 * - will not sleep 2184 */ 2185 struct mbuf * 2186 sockopt_getmbuf(const struct sockopt *sopt) 2187 { 2188 struct mbuf *m; 2189 2190 if (sopt->sopt_size > MCLBYTES) 2191 return NULL; 2192 2193 m = m_get(M_DONTWAIT, MT_SOOPTS); 2194 if (m == NULL) 2195 return NULL; 2196 2197 if (sopt->sopt_size > MLEN) { 2198 MCLGET(m, M_DONTWAIT); 2199 if ((m->m_flags & M_EXT) == 0) { 2200 m_free(m); 2201 return NULL; 2202 } 2203 } 2204 2205 memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size); 2206 m->m_len = sopt->sopt_size; 2207 2208 return m; 2209 } 2210 2211 void 2212 sohasoutofband(struct socket *so) 2213 { 2214 2215 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 2216 selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT); 2217 } 2218 2219 static void 2220 filt_sordetach(struct knote *kn) 2221 { 2222 struct socket *so; 2223 2224 so = ((file_t *)kn->kn_obj)->f_socket; 2225 solock(so); 2226 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 2227 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 2228 so->so_rcv.sb_flags &= ~SB_KNOTE; 2229 sounlock(so); 2230 } 2231 2232 /*ARGSUSED*/ 2233 static int 2234 filt_soread(struct knote *kn, long hint) 2235 { 2236 struct socket *so; 2237 int rv; 2238 2239 so = ((file_t *)kn->kn_obj)->f_socket; 2240 if (hint != NOTE_SUBMIT) 2241 solock(so); 2242 kn->kn_data = so->so_rcv.sb_cc; 2243 if (so->so_state & SS_CANTRCVMORE) { 2244 kn->kn_flags |= EV_EOF; 2245 kn->kn_fflags = so->so_error; 2246 rv = 1; 2247 } else if (so->so_error) /* temporary udp error */ 2248 rv = 1; 2249 else if (kn->kn_sfflags & NOTE_LOWAT) 2250 rv = (kn->kn_data >= kn->kn_sdata); 2251 else 2252 rv = (kn->kn_data >= so->so_rcv.sb_lowat); 2253 if (hint != NOTE_SUBMIT) 2254 sounlock(so); 2255 return rv; 2256 } 2257 2258 static void 2259 filt_sowdetach(struct knote *kn) 2260 { 2261 struct socket *so; 2262 2263 so = ((file_t *)kn->kn_obj)->f_socket; 2264 solock(so); 2265 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 2266 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 2267 so->so_snd.sb_flags &= ~SB_KNOTE; 2268 sounlock(so); 2269 } 2270 2271 /*ARGSUSED*/ 2272 static int 2273 filt_sowrite(struct knote *kn, long hint) 2274 { 2275 struct socket *so; 2276 int rv; 2277 2278 so = ((file_t *)kn->kn_obj)->f_socket; 2279 if (hint != NOTE_SUBMIT) 2280 solock(so); 2281 kn->kn_data = sbspace(&so->so_snd); 2282 if (so->so_state & SS_CANTSENDMORE) { 2283 kn->kn_flags |= EV_EOF; 2284 kn->kn_fflags = so->so_error; 2285 rv = 1; 2286 } else if (so->so_error) /* temporary udp error */ 2287 rv = 1; 2288 else if (((so->so_state & SS_ISCONNECTED) == 0) && 2289 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 2290 rv = 0; 2291 else if (kn->kn_sfflags & NOTE_LOWAT) 2292 rv = (kn->kn_data >= kn->kn_sdata); 2293 else 2294 rv = (kn->kn_data >= so->so_snd.sb_lowat); 2295 if (hint != NOTE_SUBMIT) 2296 sounlock(so); 2297 return rv; 2298 } 2299 2300 /*ARGSUSED*/ 2301 static int 2302 filt_solisten(struct knote *kn, long hint) 2303 { 2304 struct socket *so; 2305 int rv; 2306 2307 so = ((file_t *)kn->kn_obj)->f_socket; 2308 2309 /* 2310 * Set kn_data to number of incoming connections, not 2311 * counting partial (incomplete) connections. 2312 */ 2313 if (hint != NOTE_SUBMIT) 2314 solock(so); 2315 kn->kn_data = so->so_qlen; 2316 rv = (kn->kn_data > 0); 2317 if (hint != NOTE_SUBMIT) 2318 sounlock(so); 2319 return rv; 2320 } 2321 2322 static const struct filterops solisten_filtops = { 2323 .f_isfd = 1, 2324 .f_attach = NULL, 2325 .f_detach = filt_sordetach, 2326 .f_event = filt_solisten, 2327 }; 2328 2329 static const struct filterops soread_filtops = { 2330 .f_isfd = 1, 2331 .f_attach = NULL, 2332 .f_detach = filt_sordetach, 2333 .f_event = filt_soread, 2334 }; 2335 2336 static const struct filterops sowrite_filtops = { 2337 .f_isfd = 1, 2338 .f_attach = NULL, 2339 .f_detach = filt_sowdetach, 2340 .f_event = filt_sowrite, 2341 }; 2342 2343 int 2344 soo_kqfilter(struct file *fp, struct knote *kn) 2345 { 2346 struct socket *so; 2347 struct sockbuf *sb; 2348 2349 so = ((file_t *)kn->kn_obj)->f_socket; 2350 solock(so); 2351 switch (kn->kn_filter) { 2352 case EVFILT_READ: 2353 if (so->so_options & SO_ACCEPTCONN) 2354 kn->kn_fop = &solisten_filtops; 2355 else 2356 kn->kn_fop = &soread_filtops; 2357 sb = &so->so_rcv; 2358 break; 2359 case EVFILT_WRITE: 2360 kn->kn_fop = &sowrite_filtops; 2361 sb = &so->so_snd; 2362 break; 2363 default: 2364 sounlock(so); 2365 return (EINVAL); 2366 } 2367 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 2368 sb->sb_flags |= SB_KNOTE; 2369 sounlock(so); 2370 return (0); 2371 } 2372 2373 static int 2374 sodopoll(struct socket *so, int events) 2375 { 2376 int revents; 2377 2378 revents = 0; 2379 2380 if (events & (POLLIN | POLLRDNORM)) 2381 if (soreadable(so)) 2382 revents |= events & (POLLIN | POLLRDNORM); 2383 2384 if (events & (POLLOUT | POLLWRNORM)) 2385 if (sowritable(so)) 2386 revents |= events & (POLLOUT | POLLWRNORM); 2387 2388 if (events & (POLLPRI | POLLRDBAND)) 2389 if (so->so_oobmark || (so->so_state & SS_RCVATMARK)) 2390 revents |= events & (POLLPRI | POLLRDBAND); 2391 2392 return revents; 2393 } 2394 2395 int 2396 sopoll(struct socket *so, int events) 2397 { 2398 int revents = 0; 2399 2400 #ifndef DIAGNOSTIC 2401 /* 2402 * Do a quick, unlocked check in expectation that the socket 2403 * will be ready for I/O. Don't do this check if DIAGNOSTIC, 2404 * as the solocked() assertions will fail. 2405 */ 2406 if ((revents = sodopoll(so, events)) != 0) 2407 return revents; 2408 #endif 2409 2410 solock(so); 2411 if ((revents = sodopoll(so, events)) == 0) { 2412 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 2413 selrecord(curlwp, &so->so_rcv.sb_sel); 2414 so->so_rcv.sb_flags |= SB_NOTIFY; 2415 } 2416 2417 if (events & (POLLOUT | POLLWRNORM)) { 2418 selrecord(curlwp, &so->so_snd.sb_sel); 2419 so->so_snd.sb_flags |= SB_NOTIFY; 2420 } 2421 } 2422 sounlock(so); 2423 2424 return revents; 2425 } 2426 2427 struct mbuf ** 2428 sbsavetimestamp(int opt, struct mbuf *m, struct mbuf **mp) 2429 { 2430 struct timeval tv; 2431 microtime(&tv); 2432 2433 #ifdef SO_OTIMESTAMP 2434 if (opt & SO_OTIMESTAMP) { 2435 struct timeval50 tv50; 2436 2437 timeval_to_timeval50(&tv, &tv50); 2438 *mp = sbcreatecontrol(&tv50, sizeof(tv50), 2439 SCM_OTIMESTAMP, SOL_SOCKET); 2440 if (*mp) 2441 mp = &(*mp)->m_next; 2442 } else 2443 #endif 2444 2445 if (opt & SO_TIMESTAMP) { 2446 *mp = sbcreatecontrol(&tv, sizeof(tv), 2447 SCM_TIMESTAMP, SOL_SOCKET); 2448 if (*mp) 2449 mp = &(*mp)->m_next; 2450 } 2451 return mp; 2452 } 2453 2454 2455 #include <sys/sysctl.h> 2456 2457 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 2458 static int sysctl_kern_sbmax(SYSCTLFN_PROTO); 2459 2460 /* 2461 * sysctl helper routine for kern.somaxkva. ensures that the given 2462 * value is not too small. 2463 * (XXX should we maybe make sure it's not too large as well?) 2464 */ 2465 static int 2466 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 2467 { 2468 int error, new_somaxkva; 2469 struct sysctlnode node; 2470 2471 new_somaxkva = somaxkva; 2472 node = *rnode; 2473 node.sysctl_data = &new_somaxkva; 2474 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2475 if (error || newp == NULL) 2476 return (error); 2477 2478 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 2479 return (EINVAL); 2480 2481 mutex_enter(&so_pendfree_lock); 2482 somaxkva = new_somaxkva; 2483 cv_broadcast(&socurkva_cv); 2484 mutex_exit(&so_pendfree_lock); 2485 2486 return (error); 2487 } 2488 2489 /* 2490 * sysctl helper routine for kern.sbmax. Basically just ensures that 2491 * any new value is not too small. 2492 */ 2493 static int 2494 sysctl_kern_sbmax(SYSCTLFN_ARGS) 2495 { 2496 int error, new_sbmax; 2497 struct sysctlnode node; 2498 2499 new_sbmax = sb_max; 2500 node = *rnode; 2501 node.sysctl_data = &new_sbmax; 2502 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2503 if (error || newp == NULL) 2504 return (error); 2505 2506 KERNEL_LOCK(1, NULL); 2507 error = sb_max_set(new_sbmax); 2508 KERNEL_UNLOCK_ONE(NULL); 2509 2510 return (error); 2511 } 2512 2513 static void 2514 sysctl_kern_socket_setup(void) 2515 { 2516 2517 KASSERT(socket_sysctllog == NULL); 2518 2519 sysctl_createv(&socket_sysctllog, 0, NULL, NULL, 2520 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2521 CTLTYPE_INT, "somaxkva", 2522 SYSCTL_DESCR("Maximum amount of kernel memory to be " 2523 "used for socket buffers"), 2524 sysctl_kern_somaxkva, 0, NULL, 0, 2525 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 2526 2527 sysctl_createv(&socket_sysctllog, 0, NULL, NULL, 2528 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2529 CTLTYPE_INT, "sbmax", 2530 SYSCTL_DESCR("Maximum socket buffer size"), 2531 sysctl_kern_sbmax, 0, NULL, 0, 2532 CTL_KERN, KERN_SBMAX, CTL_EOL); 2533 } 2534