uipc_socket.c revision 1.228
1 /* $NetBSD: uipc_socket.c,v 1.228 2014/07/30 10:04:26 rtr 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 1.228 2014/07/30 10:04:26 rtr Exp $"); 75 76 #include "opt_compat_netbsd.h" 77 #include "opt_sock_counters.h" 78 #include "opt_sosend_loan.h" 79 #include "opt_mbuftrace.h" 80 #include "opt_somaxkva.h" 81 #include "opt_multiprocessor.h" /* XXX */ 82 83 #include <sys/param.h> 84 #include <sys/systm.h> 85 #include <sys/proc.h> 86 #include <sys/file.h> 87 #include <sys/filedesc.h> 88 #include <sys/kmem.h> 89 #include <sys/mbuf.h> 90 #include <sys/domain.h> 91 #include <sys/kernel.h> 92 #include <sys/protosw.h> 93 #include <sys/socket.h> 94 #include <sys/socketvar.h> 95 #include <sys/signalvar.h> 96 #include <sys/resourcevar.h> 97 #include <sys/uidinfo.h> 98 #include <sys/event.h> 99 #include <sys/poll.h> 100 #include <sys/kauth.h> 101 #include <sys/mutex.h> 102 #include <sys/condvar.h> 103 #include <sys/kthread.h> 104 105 #ifdef COMPAT_50 106 #include <compat/sys/time.h> 107 #include <compat/sys/socket.h> 108 #endif 109 110 #include <uvm/uvm_extern.h> 111 #include <uvm/uvm_loan.h> 112 #include <uvm/uvm_page.h> 113 114 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 115 116 extern const struct fileops socketops; 117 118 extern int somaxconn; /* patchable (XXX sysctl) */ 119 int somaxconn = SOMAXCONN; 120 kmutex_t *softnet_lock; 121 122 #ifdef SOSEND_COUNTERS 123 #include <sys/device.h> 124 125 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 126 NULL, "sosend", "loan big"); 127 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 128 NULL, "sosend", "copy big"); 129 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 130 NULL, "sosend", "copy small"); 131 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 132 NULL, "sosend", "kva limit"); 133 134 #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++ 135 136 EVCNT_ATTACH_STATIC(sosend_loan_big); 137 EVCNT_ATTACH_STATIC(sosend_copy_big); 138 EVCNT_ATTACH_STATIC(sosend_copy_small); 139 EVCNT_ATTACH_STATIC(sosend_kvalimit); 140 #else 141 142 #define SOSEND_COUNTER_INCR(ev) /* nothing */ 143 144 #endif /* SOSEND_COUNTERS */ 145 146 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR) 147 int sock_loan_thresh = -1; 148 #else 149 int sock_loan_thresh = 4096; 150 #endif 151 152 static kmutex_t so_pendfree_lock; 153 static struct mbuf *so_pendfree = NULL; 154 155 #ifndef SOMAXKVA 156 #define SOMAXKVA (16 * 1024 * 1024) 157 #endif 158 int somaxkva = SOMAXKVA; 159 static int socurkva; 160 static kcondvar_t socurkva_cv; 161 162 static kauth_listener_t socket_listener; 163 164 #define SOCK_LOAN_CHUNK 65536 165 166 static void sopendfree_thread(void *); 167 static kcondvar_t pendfree_thread_cv; 168 static lwp_t *sopendfree_lwp; 169 170 static void sysctl_kern_socket_setup(void); 171 static struct sysctllog *socket_sysctllog; 172 173 static vsize_t 174 sokvareserve(struct socket *so, vsize_t len) 175 { 176 int error; 177 178 mutex_enter(&so_pendfree_lock); 179 while (socurkva + len > somaxkva) { 180 SOSEND_COUNTER_INCR(&sosend_kvalimit); 181 error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock); 182 if (error) { 183 len = 0; 184 break; 185 } 186 } 187 socurkva += len; 188 mutex_exit(&so_pendfree_lock); 189 return len; 190 } 191 192 static void 193 sokvaunreserve(vsize_t len) 194 { 195 196 mutex_enter(&so_pendfree_lock); 197 socurkva -= len; 198 cv_broadcast(&socurkva_cv); 199 mutex_exit(&so_pendfree_lock); 200 } 201 202 /* 203 * sokvaalloc: allocate kva for loan. 204 */ 205 206 vaddr_t 207 sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so) 208 { 209 vaddr_t lva; 210 211 /* 212 * reserve kva. 213 */ 214 215 if (sokvareserve(so, len) == 0) 216 return 0; 217 218 /* 219 * allocate kva. 220 */ 221 222 lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask, 223 UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA); 224 if (lva == 0) { 225 sokvaunreserve(len); 226 return (0); 227 } 228 229 return lva; 230 } 231 232 /* 233 * sokvafree: free kva for loan. 234 */ 235 236 void 237 sokvafree(vaddr_t sva, vsize_t len) 238 { 239 240 /* 241 * free kva. 242 */ 243 244 uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY); 245 246 /* 247 * unreserve kva. 248 */ 249 250 sokvaunreserve(len); 251 } 252 253 static void 254 sodoloanfree(struct vm_page **pgs, void *buf, size_t size) 255 { 256 vaddr_t sva, eva; 257 vsize_t len; 258 int npgs; 259 260 KASSERT(pgs != NULL); 261 262 eva = round_page((vaddr_t) buf + size); 263 sva = trunc_page((vaddr_t) buf); 264 len = eva - sva; 265 npgs = len >> PAGE_SHIFT; 266 267 pmap_kremove(sva, len); 268 pmap_update(pmap_kernel()); 269 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 270 sokvafree(sva, len); 271 } 272 273 /* 274 * sopendfree_thread: free mbufs on "pendfree" list. 275 * unlock and relock so_pendfree_lock when freeing mbufs. 276 */ 277 278 static void 279 sopendfree_thread(void *v) 280 { 281 struct mbuf *m, *next; 282 size_t rv; 283 284 mutex_enter(&so_pendfree_lock); 285 286 for (;;) { 287 rv = 0; 288 while (so_pendfree != NULL) { 289 m = so_pendfree; 290 so_pendfree = NULL; 291 mutex_exit(&so_pendfree_lock); 292 293 for (; m != NULL; m = next) { 294 next = m->m_next; 295 KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0); 296 KASSERT(m->m_ext.ext_refcnt == 0); 297 298 rv += m->m_ext.ext_size; 299 sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf, 300 m->m_ext.ext_size); 301 pool_cache_put(mb_cache, m); 302 } 303 304 mutex_enter(&so_pendfree_lock); 305 } 306 if (rv) 307 cv_broadcast(&socurkva_cv); 308 cv_wait(&pendfree_thread_cv, &so_pendfree_lock); 309 } 310 panic("sopendfree_thread"); 311 /* NOTREACHED */ 312 } 313 314 void 315 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg) 316 { 317 318 KASSERT(m != NULL); 319 320 /* 321 * postpone freeing mbuf. 322 * 323 * we can't do it in interrupt context 324 * because we need to put kva back to kernel_map. 325 */ 326 327 mutex_enter(&so_pendfree_lock); 328 m->m_next = so_pendfree; 329 so_pendfree = m; 330 cv_signal(&pendfree_thread_cv); 331 mutex_exit(&so_pendfree_lock); 332 } 333 334 static long 335 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 336 { 337 struct iovec *iov = uio->uio_iov; 338 vaddr_t sva, eva; 339 vsize_t len; 340 vaddr_t lva; 341 int npgs, error; 342 vaddr_t va; 343 int i; 344 345 if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace)) 346 return (0); 347 348 if (iov->iov_len < (size_t) space) 349 space = iov->iov_len; 350 if (space > SOCK_LOAN_CHUNK) 351 space = SOCK_LOAN_CHUNK; 352 353 eva = round_page((vaddr_t) iov->iov_base + space); 354 sva = trunc_page((vaddr_t) iov->iov_base); 355 len = eva - sva; 356 npgs = len >> PAGE_SHIFT; 357 358 KASSERT(npgs <= M_EXT_MAXPAGES); 359 360 lva = sokvaalloc(sva, len, so); 361 if (lva == 0) 362 return 0; 363 364 error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len, 365 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 366 if (error) { 367 sokvafree(lva, len); 368 return (0); 369 } 370 371 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 372 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 373 VM_PROT_READ, 0); 374 pmap_update(pmap_kernel()); 375 376 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 377 378 MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so); 379 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 380 381 uio->uio_resid -= space; 382 /* uio_offset not updated, not set/used for write(2) */ 383 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space; 384 uio->uio_iov->iov_len -= space; 385 if (uio->uio_iov->iov_len == 0) { 386 uio->uio_iov++; 387 uio->uio_iovcnt--; 388 } 389 390 return (space); 391 } 392 393 struct mbuf * 394 getsombuf(struct socket *so, int type) 395 { 396 struct mbuf *m; 397 398 m = m_get(M_WAIT, type); 399 MCLAIM(m, so->so_mowner); 400 return m; 401 } 402 403 static int 404 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 405 void *arg0, void *arg1, void *arg2, void *arg3) 406 { 407 int result; 408 enum kauth_network_req req; 409 410 result = KAUTH_RESULT_DEFER; 411 req = (enum kauth_network_req)arg0; 412 413 if ((action != KAUTH_NETWORK_SOCKET) && 414 (action != KAUTH_NETWORK_BIND)) 415 return result; 416 417 switch (req) { 418 case KAUTH_REQ_NETWORK_BIND_PORT: 419 result = KAUTH_RESULT_ALLOW; 420 break; 421 422 case KAUTH_REQ_NETWORK_SOCKET_DROP: { 423 /* Normal users can only drop their own connections. */ 424 struct socket *so = (struct socket *)arg1; 425 426 if (so->so_cred && proc_uidmatch(cred, so->so_cred) == 0) 427 result = KAUTH_RESULT_ALLOW; 428 429 break; 430 } 431 432 case KAUTH_REQ_NETWORK_SOCKET_OPEN: 433 /* We allow "raw" routing/bluetooth sockets to anyone. */ 434 if ((u_long)arg1 == PF_ROUTE || (u_long)arg1 == PF_OROUTE 435 || (u_long)arg1 == PF_BLUETOOTH) { 436 result = KAUTH_RESULT_ALLOW; 437 } else { 438 /* Privileged, let secmodel handle this. */ 439 if ((u_long)arg2 == SOCK_RAW) 440 break; 441 } 442 443 result = KAUTH_RESULT_ALLOW; 444 445 break; 446 447 case KAUTH_REQ_NETWORK_SOCKET_CANSEE: 448 result = KAUTH_RESULT_ALLOW; 449 450 break; 451 452 default: 453 break; 454 } 455 456 return result; 457 } 458 459 void 460 soinit(void) 461 { 462 463 sysctl_kern_socket_setup(); 464 465 mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM); 466 softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 467 cv_init(&socurkva_cv, "sokva"); 468 cv_init(&pendfree_thread_cv, "sopendfr"); 469 soinit2(); 470 471 /* Set the initial adjusted socket buffer size. */ 472 if (sb_max_set(sb_max)) 473 panic("bad initial sb_max value: %lu", sb_max); 474 475 socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK, 476 socket_listener_cb, NULL); 477 } 478 479 void 480 soinit1(void) 481 { 482 int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, 483 sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree"); 484 if (error) 485 panic("soinit1 %d", error); 486 } 487 488 /* 489 * socreate: create a new socket of the specified type and the protocol. 490 * 491 * => Caller may specify another socket for lock sharing (must not be held). 492 * => Returns the new socket without lock held. 493 */ 494 int 495 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l, 496 struct socket *lockso) 497 { 498 const struct protosw *prp; 499 struct socket *so; 500 uid_t uid; 501 int error; 502 kmutex_t *lock; 503 504 error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET, 505 KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type), 506 KAUTH_ARG(proto)); 507 if (error != 0) 508 return error; 509 510 if (proto) 511 prp = pffindproto(dom, proto, type); 512 else 513 prp = pffindtype(dom, type); 514 if (prp == NULL) { 515 /* no support for domain */ 516 if (pffinddomain(dom) == 0) 517 return EAFNOSUPPORT; 518 /* no support for socket type */ 519 if (proto == 0 && type != 0) 520 return EPROTOTYPE; 521 return EPROTONOSUPPORT; 522 } 523 if (prp->pr_usrreqs == NULL) 524 return EPROTONOSUPPORT; 525 if (prp->pr_type != type) 526 return EPROTOTYPE; 527 528 so = soget(true); 529 so->so_type = type; 530 so->so_proto = prp; 531 so->so_send = sosend; 532 so->so_receive = soreceive; 533 #ifdef MBUFTRACE 534 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 535 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 536 so->so_mowner = &prp->pr_domain->dom_mowner; 537 #endif 538 uid = kauth_cred_geteuid(l->l_cred); 539 so->so_uidinfo = uid_find(uid); 540 so->so_cpid = l->l_proc->p_pid; 541 542 /* 543 * Lock assigned and taken during PCB attach, unless we share 544 * the lock with another socket, e.g. socketpair(2) case. 545 */ 546 if (lockso) { 547 lock = lockso->so_lock; 548 so->so_lock = lock; 549 mutex_obj_hold(lock); 550 mutex_enter(lock); 551 } 552 553 /* Attach the PCB (returns with the socket lock held). */ 554 error = (*prp->pr_usrreqs->pr_attach)(so, proto); 555 KASSERT(solocked(so)); 556 557 if (error) { 558 KASSERT(so->so_pcb == NULL); 559 so->so_state |= SS_NOFDREF; 560 sofree(so); 561 return error; 562 } 563 so->so_cred = kauth_cred_dup(l->l_cred); 564 sounlock(so); 565 566 *aso = so; 567 return 0; 568 } 569 570 /* 571 * fsocreate: create a socket and a file descriptor associated with it. 572 * 573 * => On success, write file descriptor to fdout and return zero. 574 * => On failure, return non-zero; *fdout will be undefined. 575 */ 576 int 577 fsocreate(int domain, struct socket **sop, int type, int proto, int *fdout) 578 { 579 lwp_t *l = curlwp; 580 int error, fd, flags; 581 struct socket *so; 582 struct file *fp; 583 584 if ((error = fd_allocfile(&fp, &fd)) != 0) { 585 return error; 586 } 587 flags = type & SOCK_FLAGS_MASK; 588 fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0); 589 fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)| 590 ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0); 591 fp->f_type = DTYPE_SOCKET; 592 fp->f_ops = &socketops; 593 594 type &= ~SOCK_FLAGS_MASK; 595 error = socreate(domain, &so, type, proto, l, NULL); 596 if (error) { 597 fd_abort(curproc, fp, fd); 598 return error; 599 } 600 if (flags & SOCK_NONBLOCK) { 601 so->so_state |= SS_NBIO; 602 } 603 fp->f_data = so; 604 fd_affix(curproc, fp, fd); 605 606 if (sop != NULL) { 607 *sop = so; 608 } 609 *fdout = fd; 610 return error; 611 } 612 613 int 614 sofamily(const struct socket *so) 615 { 616 const struct protosw *pr; 617 const struct domain *dom; 618 619 if ((pr = so->so_proto) == NULL) 620 return AF_UNSPEC; 621 if ((dom = pr->pr_domain) == NULL) 622 return AF_UNSPEC; 623 return dom->dom_family; 624 } 625 626 int 627 sobind(struct socket *so, struct mbuf *nam, struct lwp *l) 628 { 629 int error; 630 631 solock(so); 632 error = (*so->so_proto->pr_usrreqs->pr_bind)(so, nam); 633 sounlock(so); 634 return error; 635 } 636 637 int 638 solisten(struct socket *so, int backlog, struct lwp *l) 639 { 640 int error; 641 642 solock(so); 643 if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 644 SS_ISDISCONNECTING)) != 0) { 645 sounlock(so); 646 return EINVAL; 647 } 648 error = (*so->so_proto->pr_usrreqs->pr_listen)(so); 649 if (error != 0) { 650 sounlock(so); 651 return error; 652 } 653 if (TAILQ_EMPTY(&so->so_q)) 654 so->so_options |= SO_ACCEPTCONN; 655 if (backlog < 0) 656 backlog = 0; 657 so->so_qlimit = min(backlog, somaxconn); 658 sounlock(so); 659 return 0; 660 } 661 662 void 663 sofree(struct socket *so) 664 { 665 u_int refs; 666 667 KASSERT(solocked(so)); 668 669 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) { 670 sounlock(so); 671 return; 672 } 673 if (so->so_head) { 674 /* 675 * We must not decommission a socket that's on the accept(2) 676 * queue. If we do, then accept(2) may hang after select(2) 677 * indicated that the listening socket was ready. 678 */ 679 if (!soqremque(so, 0)) { 680 sounlock(so); 681 return; 682 } 683 } 684 if (so->so_rcv.sb_hiwat) 685 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 686 RLIM_INFINITY); 687 if (so->so_snd.sb_hiwat) 688 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 689 RLIM_INFINITY); 690 sbrelease(&so->so_snd, so); 691 KASSERT(!cv_has_waiters(&so->so_cv)); 692 KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv)); 693 KASSERT(!cv_has_waiters(&so->so_snd.sb_cv)); 694 sorflush(so); 695 refs = so->so_aborting; /* XXX */ 696 /* Remove acccept filter if one is present. */ 697 if (so->so_accf != NULL) 698 (void)accept_filt_clear(so); 699 sounlock(so); 700 if (refs == 0) /* XXX */ 701 soput(so); 702 } 703 704 /* 705 * soclose: close a socket on last file table reference removal. 706 * Initiate disconnect if connected. Free socket when disconnect complete. 707 */ 708 int 709 soclose(struct socket *so) 710 { 711 struct socket *so2; 712 int error = 0; 713 714 solock(so); 715 if (so->so_options & SO_ACCEPTCONN) { 716 for (;;) { 717 if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 718 KASSERT(solocked2(so, so2)); 719 (void) soqremque(so2, 0); 720 /* soabort drops the lock. */ 721 (void) soabort(so2); 722 solock(so); 723 continue; 724 } 725 if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 726 KASSERT(solocked2(so, so2)); 727 (void) soqremque(so2, 1); 728 /* soabort drops the lock. */ 729 (void) soabort(so2); 730 solock(so); 731 continue; 732 } 733 break; 734 } 735 } 736 if (so->so_pcb == NULL) 737 goto discard; 738 if (so->so_state & SS_ISCONNECTED) { 739 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 740 error = sodisconnect(so); 741 if (error) 742 goto drop; 743 } 744 if (so->so_options & SO_LINGER) { 745 if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) == 746 (SS_ISDISCONNECTING|SS_NBIO)) 747 goto drop; 748 while (so->so_state & SS_ISCONNECTED) { 749 error = sowait(so, true, so->so_linger * hz); 750 if (error) 751 break; 752 } 753 } 754 } 755 drop: 756 if (so->so_pcb) { 757 KASSERT(solocked(so)); 758 (*so->so_proto->pr_usrreqs->pr_detach)(so); 759 } 760 discard: 761 KASSERT((so->so_state & SS_NOFDREF) == 0); 762 kauth_cred_free(so->so_cred); 763 so->so_state |= SS_NOFDREF; 764 sofree(so); 765 return error; 766 } 767 768 /* 769 * Must be called with the socket locked.. Will return with it unlocked. 770 */ 771 int 772 soabort(struct socket *so) 773 { 774 u_int refs; 775 int error; 776 777 KASSERT(solocked(so)); 778 KASSERT(so->so_head == NULL); 779 780 so->so_aborting++; /* XXX */ 781 error = (*so->so_proto->pr_usrreqs->pr_generic)(so, 782 PRU_ABORT, NULL, NULL, NULL, NULL); 783 refs = --so->so_aborting; /* XXX */ 784 if (error || (refs == 0)) { 785 sofree(so); 786 } else { 787 sounlock(so); 788 } 789 return error; 790 } 791 792 int 793 soaccept(struct socket *so, struct mbuf *nam) 794 { 795 int error; 796 797 KASSERT(solocked(so)); 798 KASSERT((so->so_state & SS_NOFDREF) != 0); 799 800 so->so_state &= ~SS_NOFDREF; 801 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 802 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 803 error = (*so->so_proto->pr_usrreqs->pr_accept)(so, nam); 804 else 805 error = ECONNABORTED; 806 807 return error; 808 } 809 810 int 811 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l) 812 { 813 int error; 814 815 KASSERT(solocked(so)); 816 817 if (so->so_options & SO_ACCEPTCONN) 818 return EOPNOTSUPP; 819 /* 820 * If protocol is connection-based, can only connect once. 821 * Otherwise, if connected, try to disconnect first. 822 * This allows user to disconnect by connecting to, e.g., 823 * a null address. 824 */ 825 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 826 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 827 (error = sodisconnect(so)))) 828 error = EISCONN; 829 else 830 error = (*so->so_proto->pr_usrreqs->pr_connect)(so, nam); 831 832 return error; 833 } 834 835 int 836 soconnect2(struct socket *so1, struct socket *so2) 837 { 838 KASSERT(solocked2(so1, so2)); 839 840 return (*so1->so_proto->pr_usrreqs->pr_generic)(so1, 841 PRU_CONNECT2, NULL, (struct mbuf *)so2, NULL, NULL); 842 } 843 844 int 845 sodisconnect(struct socket *so) 846 { 847 int error; 848 849 KASSERT(solocked(so)); 850 851 if ((so->so_state & SS_ISCONNECTED) == 0) { 852 error = ENOTCONN; 853 } else if (so->so_state & SS_ISDISCONNECTING) { 854 error = EALREADY; 855 } else { 856 error = (*so->so_proto->pr_usrreqs->pr_generic)(so, 857 PRU_DISCONNECT, NULL, NULL, NULL, NULL); 858 } 859 return (error); 860 } 861 862 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 863 /* 864 * Send on a socket. 865 * If send must go all at once and message is larger than 866 * send buffering, then hard error. 867 * Lock against other senders. 868 * If must go all at once and not enough room now, then 869 * inform user that this would block and do nothing. 870 * Otherwise, if nonblocking, send as much as possible. 871 * The data to be sent is described by "uio" if nonzero, 872 * otherwise by the mbuf chain "top" (which must be null 873 * if uio is not). Data provided in mbuf chain must be small 874 * enough to send all at once. 875 * 876 * Returns nonzero on error, timeout or signal; callers 877 * must check for short counts if EINTR/ERESTART are returned. 878 * Data and control buffers are freed on return. 879 */ 880 int 881 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 882 struct mbuf *control, int flags, struct lwp *l) 883 { 884 struct mbuf **mp, *m; 885 long space, len, resid, clen, mlen; 886 int error, s, dontroute, atomic; 887 short wakeup_state = 0; 888 889 clen = 0; 890 891 /* 892 * solock() provides atomicity of access. splsoftnet() prevents 893 * protocol processing soft interrupts from interrupting us and 894 * blocking (expensive). 895 */ 896 s = splsoftnet(); 897 solock(so); 898 atomic = sosendallatonce(so) || top; 899 if (uio) 900 resid = uio->uio_resid; 901 else 902 resid = top->m_pkthdr.len; 903 /* 904 * In theory resid should be unsigned. 905 * However, space must be signed, as it might be less than 0 906 * if we over-committed, and we must use a signed comparison 907 * of space and resid. On the other hand, a negative resid 908 * causes us to loop sending 0-length segments to the protocol. 909 */ 910 if (resid < 0) { 911 error = EINVAL; 912 goto out; 913 } 914 dontroute = 915 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 916 (so->so_proto->pr_flags & PR_ATOMIC); 917 l->l_ru.ru_msgsnd++; 918 if (control) 919 clen = control->m_len; 920 restart: 921 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 922 goto out; 923 do { 924 if (so->so_state & SS_CANTSENDMORE) { 925 error = EPIPE; 926 goto release; 927 } 928 if (so->so_error) { 929 error = so->so_error; 930 so->so_error = 0; 931 goto release; 932 } 933 if ((so->so_state & SS_ISCONNECTED) == 0) { 934 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 935 if (resid || clen == 0) { 936 error = ENOTCONN; 937 goto release; 938 } 939 } else if (addr == 0) { 940 error = EDESTADDRREQ; 941 goto release; 942 } 943 } 944 space = sbspace(&so->so_snd); 945 if (flags & MSG_OOB) 946 space += 1024; 947 if ((atomic && resid > so->so_snd.sb_hiwat) || 948 clen > so->so_snd.sb_hiwat) { 949 error = EMSGSIZE; 950 goto release; 951 } 952 if (space < resid + clen && 953 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 954 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) { 955 error = EWOULDBLOCK; 956 goto release; 957 } 958 sbunlock(&so->so_snd); 959 if (wakeup_state & SS_RESTARTSYS) { 960 error = ERESTART; 961 goto out; 962 } 963 error = sbwait(&so->so_snd); 964 if (error) 965 goto out; 966 wakeup_state = so->so_state; 967 goto restart; 968 } 969 wakeup_state = 0; 970 mp = ⊤ 971 space -= clen; 972 do { 973 if (uio == NULL) { 974 /* 975 * Data is prepackaged in "top". 976 */ 977 resid = 0; 978 if (flags & MSG_EOR) 979 top->m_flags |= M_EOR; 980 } else do { 981 sounlock(so); 982 splx(s); 983 if (top == NULL) { 984 m = m_gethdr(M_WAIT, MT_DATA); 985 mlen = MHLEN; 986 m->m_pkthdr.len = 0; 987 m->m_pkthdr.rcvif = NULL; 988 } else { 989 m = m_get(M_WAIT, MT_DATA); 990 mlen = MLEN; 991 } 992 MCLAIM(m, so->so_snd.sb_mowner); 993 if (sock_loan_thresh >= 0 && 994 uio->uio_iov->iov_len >= sock_loan_thresh && 995 space >= sock_loan_thresh && 996 (len = sosend_loan(so, uio, m, 997 space)) != 0) { 998 SOSEND_COUNTER_INCR(&sosend_loan_big); 999 space -= len; 1000 goto have_data; 1001 } 1002 if (resid >= MINCLSIZE && space >= MCLBYTES) { 1003 SOSEND_COUNTER_INCR(&sosend_copy_big); 1004 m_clget(m, M_DONTWAIT); 1005 if ((m->m_flags & M_EXT) == 0) 1006 goto nopages; 1007 mlen = MCLBYTES; 1008 if (atomic && top == 0) { 1009 len = lmin(MCLBYTES - max_hdr, 1010 resid); 1011 m->m_data += max_hdr; 1012 } else 1013 len = lmin(MCLBYTES, resid); 1014 space -= len; 1015 } else { 1016 nopages: 1017 SOSEND_COUNTER_INCR(&sosend_copy_small); 1018 len = lmin(lmin(mlen, resid), space); 1019 space -= len; 1020 /* 1021 * For datagram protocols, leave room 1022 * for protocol headers in first mbuf. 1023 */ 1024 if (atomic && top == 0 && len < mlen) 1025 MH_ALIGN(m, len); 1026 } 1027 error = uiomove(mtod(m, void *), (int)len, uio); 1028 have_data: 1029 resid = uio->uio_resid; 1030 m->m_len = len; 1031 *mp = m; 1032 top->m_pkthdr.len += len; 1033 s = splsoftnet(); 1034 solock(so); 1035 if (error != 0) 1036 goto release; 1037 mp = &m->m_next; 1038 if (resid <= 0) { 1039 if (flags & MSG_EOR) 1040 top->m_flags |= M_EOR; 1041 break; 1042 } 1043 } while (space > 0 && atomic); 1044 1045 if (so->so_state & SS_CANTSENDMORE) { 1046 error = EPIPE; 1047 goto release; 1048 } 1049 if (dontroute) 1050 so->so_options |= SO_DONTROUTE; 1051 if (resid > 0) 1052 so->so_state |= SS_MORETOCOME; 1053 if (flags & MSG_OOB) 1054 error = (*so->so_proto->pr_usrreqs->pr_sendoob)(so, 1055 top, control); 1056 else 1057 error = (*so->so_proto->pr_usrreqs->pr_generic)(so, 1058 PRU_SEND, top, addr, control, curlwp); 1059 if (dontroute) 1060 so->so_options &= ~SO_DONTROUTE; 1061 if (resid > 0) 1062 so->so_state &= ~SS_MORETOCOME; 1063 clen = 0; 1064 control = NULL; 1065 top = NULL; 1066 mp = ⊤ 1067 if (error != 0) 1068 goto release; 1069 } while (resid && space > 0); 1070 } while (resid); 1071 1072 release: 1073 sbunlock(&so->so_snd); 1074 out: 1075 sounlock(so); 1076 splx(s); 1077 if (top) 1078 m_freem(top); 1079 if (control) 1080 m_freem(control); 1081 return (error); 1082 } 1083 1084 /* 1085 * Following replacement or removal of the first mbuf on the first 1086 * mbuf chain of a socket buffer, push necessary state changes back 1087 * into the socket buffer so that other consumers see the values 1088 * consistently. 'nextrecord' is the callers locally stored value of 1089 * the original value of sb->sb_mb->m_nextpkt which must be restored 1090 * when the lead mbuf changes. NOTE: 'nextrecord' may be NULL. 1091 */ 1092 static void 1093 sbsync(struct sockbuf *sb, struct mbuf *nextrecord) 1094 { 1095 1096 KASSERT(solocked(sb->sb_so)); 1097 1098 /* 1099 * First, update for the new value of nextrecord. If necessary, 1100 * make it the first record. 1101 */ 1102 if (sb->sb_mb != NULL) 1103 sb->sb_mb->m_nextpkt = nextrecord; 1104 else 1105 sb->sb_mb = nextrecord; 1106 1107 /* 1108 * Now update any dependent socket buffer fields to reflect 1109 * the new state. This is an inline of SB_EMPTY_FIXUP, with 1110 * the addition of a second clause that takes care of the 1111 * case where sb_mb has been updated, but remains the last 1112 * record. 1113 */ 1114 if (sb->sb_mb == NULL) { 1115 sb->sb_mbtail = NULL; 1116 sb->sb_lastrecord = NULL; 1117 } else if (sb->sb_mb->m_nextpkt == NULL) 1118 sb->sb_lastrecord = sb->sb_mb; 1119 } 1120 1121 /* 1122 * Implement receive operations on a socket. 1123 * We depend on the way that records are added to the sockbuf 1124 * by sbappend*. In particular, each record (mbufs linked through m_next) 1125 * must begin with an address if the protocol so specifies, 1126 * followed by an optional mbuf or mbufs containing ancillary data, 1127 * and then zero or more mbufs of data. 1128 * In order to avoid blocking network interrupts for the entire time here, 1129 * we splx() while doing the actual copy to user space. 1130 * Although the sockbuf is locked, new data may still be appended, 1131 * and thus we must maintain consistency of the sockbuf during that time. 1132 * 1133 * The caller may receive the data as a single mbuf chain by supplying 1134 * an mbuf **mp0 for use in returning the chain. The uio is then used 1135 * only for the count in uio_resid. 1136 */ 1137 int 1138 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 1139 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1140 { 1141 struct lwp *l = curlwp; 1142 struct mbuf *m, **mp, *mt; 1143 size_t len, offset, moff, orig_resid; 1144 int atomic, flags, error, s, type; 1145 const struct protosw *pr; 1146 struct mbuf *nextrecord; 1147 int mbuf_removed = 0; 1148 const struct domain *dom; 1149 short wakeup_state = 0; 1150 1151 pr = so->so_proto; 1152 atomic = pr->pr_flags & PR_ATOMIC; 1153 dom = pr->pr_domain; 1154 mp = mp0; 1155 type = 0; 1156 orig_resid = uio->uio_resid; 1157 1158 if (paddr != NULL) 1159 *paddr = NULL; 1160 if (controlp != NULL) 1161 *controlp = NULL; 1162 if (flagsp != NULL) 1163 flags = *flagsp &~ MSG_EOR; 1164 else 1165 flags = 0; 1166 1167 if (flags & MSG_OOB) { 1168 m = m_get(M_WAIT, MT_DATA); 1169 solock(so); 1170 error = (*pr->pr_usrreqs->pr_recvoob)(so, m, flags & MSG_PEEK); 1171 sounlock(so); 1172 if (error) 1173 goto bad; 1174 do { 1175 error = uiomove(mtod(m, void *), 1176 MIN(uio->uio_resid, m->m_len), uio); 1177 m = m_free(m); 1178 } while (uio->uio_resid > 0 && error == 0 && m); 1179 bad: 1180 if (m != NULL) 1181 m_freem(m); 1182 return error; 1183 } 1184 if (mp != NULL) 1185 *mp = NULL; 1186 1187 /* 1188 * solock() provides atomicity of access. splsoftnet() prevents 1189 * protocol processing soft interrupts from interrupting us and 1190 * blocking (expensive). 1191 */ 1192 s = splsoftnet(); 1193 solock(so); 1194 restart: 1195 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) { 1196 sounlock(so); 1197 splx(s); 1198 return error; 1199 } 1200 1201 m = so->so_rcv.sb_mb; 1202 /* 1203 * If we have less data than requested, block awaiting more 1204 * (subject to any timeout) if: 1205 * 1. the current count is less than the low water mark, 1206 * 2. MSG_WAITALL is set, and it is possible to do the entire 1207 * receive operation at once if we block (resid <= hiwat), or 1208 * 3. MSG_DONTWAIT is not set. 1209 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1210 * we have to do the receive in sections, and thus risk returning 1211 * a short count if a timeout or signal occurs after we start. 1212 */ 1213 if (m == NULL || 1214 ((flags & MSG_DONTWAIT) == 0 && 1215 so->so_rcv.sb_cc < uio->uio_resid && 1216 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1217 ((flags & MSG_WAITALL) && 1218 uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1219 m->m_nextpkt == NULL && !atomic)) { 1220 #ifdef DIAGNOSTIC 1221 if (m == NULL && so->so_rcv.sb_cc) 1222 panic("receive 1"); 1223 #endif 1224 if (so->so_error) { 1225 if (m != NULL) 1226 goto dontblock; 1227 error = so->so_error; 1228 if ((flags & MSG_PEEK) == 0) 1229 so->so_error = 0; 1230 goto release; 1231 } 1232 if (so->so_state & SS_CANTRCVMORE) { 1233 if (m != NULL) 1234 goto dontblock; 1235 else 1236 goto release; 1237 } 1238 for (; m != NULL; m = m->m_next) 1239 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1240 m = so->so_rcv.sb_mb; 1241 goto dontblock; 1242 } 1243 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1244 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1245 error = ENOTCONN; 1246 goto release; 1247 } 1248 if (uio->uio_resid == 0) 1249 goto release; 1250 if ((so->so_state & SS_NBIO) || 1251 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1252 error = EWOULDBLOCK; 1253 goto release; 1254 } 1255 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1256 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1257 sbunlock(&so->so_rcv); 1258 if (wakeup_state & SS_RESTARTSYS) 1259 error = ERESTART; 1260 else 1261 error = sbwait(&so->so_rcv); 1262 if (error != 0) { 1263 sounlock(so); 1264 splx(s); 1265 return error; 1266 } 1267 wakeup_state = so->so_state; 1268 goto restart; 1269 } 1270 dontblock: 1271 /* 1272 * On entry here, m points to the first record of the socket buffer. 1273 * From this point onward, we maintain 'nextrecord' as a cache of the 1274 * pointer to the next record in the socket buffer. We must keep the 1275 * various socket buffer pointers and local stack versions of the 1276 * pointers in sync, pushing out modifications before dropping the 1277 * socket lock, and re-reading them when picking it up. 1278 * 1279 * Otherwise, we will race with the network stack appending new data 1280 * or records onto the socket buffer by using inconsistent/stale 1281 * versions of the field, possibly resulting in socket buffer 1282 * corruption. 1283 * 1284 * By holding the high-level sblock(), we prevent simultaneous 1285 * readers from pulling off the front of the socket buffer. 1286 */ 1287 if (l != NULL) 1288 l->l_ru.ru_msgrcv++; 1289 KASSERT(m == so->so_rcv.sb_mb); 1290 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1291 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1292 nextrecord = m->m_nextpkt; 1293 if (pr->pr_flags & PR_ADDR) { 1294 #ifdef DIAGNOSTIC 1295 if (m->m_type != MT_SONAME) 1296 panic("receive 1a"); 1297 #endif 1298 orig_resid = 0; 1299 if (flags & MSG_PEEK) { 1300 if (paddr) 1301 *paddr = m_copy(m, 0, m->m_len); 1302 m = m->m_next; 1303 } else { 1304 sbfree(&so->so_rcv, m); 1305 mbuf_removed = 1; 1306 if (paddr != NULL) { 1307 *paddr = m; 1308 so->so_rcv.sb_mb = m->m_next; 1309 m->m_next = NULL; 1310 m = so->so_rcv.sb_mb; 1311 } else { 1312 MFREE(m, so->so_rcv.sb_mb); 1313 m = so->so_rcv.sb_mb; 1314 } 1315 sbsync(&so->so_rcv, nextrecord); 1316 } 1317 } 1318 1319 /* 1320 * Process one or more MT_CONTROL mbufs present before any data mbufs 1321 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 1322 * just copy the data; if !MSG_PEEK, we call into the protocol to 1323 * perform externalization (or freeing if controlp == NULL). 1324 */ 1325 if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) { 1326 struct mbuf *cm = NULL, *cmn; 1327 struct mbuf **cme = &cm; 1328 1329 do { 1330 if (flags & MSG_PEEK) { 1331 if (controlp != NULL) { 1332 *controlp = m_copy(m, 0, m->m_len); 1333 controlp = &(*controlp)->m_next; 1334 } 1335 m = m->m_next; 1336 } else { 1337 sbfree(&so->so_rcv, m); 1338 so->so_rcv.sb_mb = m->m_next; 1339 m->m_next = NULL; 1340 *cme = m; 1341 cme = &(*cme)->m_next; 1342 m = so->so_rcv.sb_mb; 1343 } 1344 } while (m != NULL && m->m_type == MT_CONTROL); 1345 if ((flags & MSG_PEEK) == 0) 1346 sbsync(&so->so_rcv, nextrecord); 1347 for (; cm != NULL; cm = cmn) { 1348 cmn = cm->m_next; 1349 cm->m_next = NULL; 1350 type = mtod(cm, struct cmsghdr *)->cmsg_type; 1351 if (controlp != NULL) { 1352 if (dom->dom_externalize != NULL && 1353 type == SCM_RIGHTS) { 1354 sounlock(so); 1355 splx(s); 1356 error = (*dom->dom_externalize)(cm, l, 1357 (flags & MSG_CMSG_CLOEXEC) ? 1358 O_CLOEXEC : 0); 1359 s = splsoftnet(); 1360 solock(so); 1361 } 1362 *controlp = cm; 1363 while (*controlp != NULL) 1364 controlp = &(*controlp)->m_next; 1365 } else { 1366 /* 1367 * Dispose of any SCM_RIGHTS message that went 1368 * through the read path rather than recv. 1369 */ 1370 if (dom->dom_dispose != NULL && 1371 type == SCM_RIGHTS) { 1372 sounlock(so); 1373 (*dom->dom_dispose)(cm); 1374 solock(so); 1375 } 1376 m_freem(cm); 1377 } 1378 } 1379 if (m != NULL) 1380 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 1381 else 1382 nextrecord = so->so_rcv.sb_mb; 1383 orig_resid = 0; 1384 } 1385 1386 /* If m is non-NULL, we have some data to read. */ 1387 if (__predict_true(m != NULL)) { 1388 type = m->m_type; 1389 if (type == MT_OOBDATA) 1390 flags |= MSG_OOB; 1391 } 1392 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1393 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1394 1395 moff = 0; 1396 offset = 0; 1397 while (m != NULL && uio->uio_resid > 0 && error == 0) { 1398 if (m->m_type == MT_OOBDATA) { 1399 if (type != MT_OOBDATA) 1400 break; 1401 } else if (type == MT_OOBDATA) 1402 break; 1403 #ifdef DIAGNOSTIC 1404 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1405 panic("receive 3"); 1406 #endif 1407 so->so_state &= ~SS_RCVATMARK; 1408 wakeup_state = 0; 1409 len = uio->uio_resid; 1410 if (so->so_oobmark && len > so->so_oobmark - offset) 1411 len = so->so_oobmark - offset; 1412 if (len > m->m_len - moff) 1413 len = m->m_len - moff; 1414 /* 1415 * If mp is set, just pass back the mbufs. 1416 * Otherwise copy them out via the uio, then free. 1417 * Sockbuf must be consistent here (points to current mbuf, 1418 * it points to next record) when we drop priority; 1419 * we must note any additions to the sockbuf when we 1420 * block interrupts again. 1421 */ 1422 if (mp == NULL) { 1423 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1424 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1425 sounlock(so); 1426 splx(s); 1427 error = uiomove(mtod(m, char *) + moff, len, uio); 1428 s = splsoftnet(); 1429 solock(so); 1430 if (error != 0) { 1431 /* 1432 * If any part of the record has been removed 1433 * (such as the MT_SONAME mbuf, which will 1434 * happen when PR_ADDR, and thus also 1435 * PR_ATOMIC, is set), then drop the entire 1436 * record to maintain the atomicity of the 1437 * receive operation. 1438 * 1439 * This avoids a later panic("receive 1a") 1440 * when compiled with DIAGNOSTIC. 1441 */ 1442 if (m && mbuf_removed && atomic) 1443 (void) sbdroprecord(&so->so_rcv); 1444 1445 goto release; 1446 } 1447 } else 1448 uio->uio_resid -= len; 1449 if (len == m->m_len - moff) { 1450 if (m->m_flags & M_EOR) 1451 flags |= MSG_EOR; 1452 if (flags & MSG_PEEK) { 1453 m = m->m_next; 1454 moff = 0; 1455 } else { 1456 nextrecord = m->m_nextpkt; 1457 sbfree(&so->so_rcv, m); 1458 if (mp) { 1459 *mp = m; 1460 mp = &m->m_next; 1461 so->so_rcv.sb_mb = m = m->m_next; 1462 *mp = NULL; 1463 } else { 1464 MFREE(m, so->so_rcv.sb_mb); 1465 m = so->so_rcv.sb_mb; 1466 } 1467 /* 1468 * If m != NULL, we also know that 1469 * so->so_rcv.sb_mb != NULL. 1470 */ 1471 KASSERT(so->so_rcv.sb_mb == m); 1472 if (m) { 1473 m->m_nextpkt = nextrecord; 1474 if (nextrecord == NULL) 1475 so->so_rcv.sb_lastrecord = m; 1476 } else { 1477 so->so_rcv.sb_mb = nextrecord; 1478 SB_EMPTY_FIXUP(&so->so_rcv); 1479 } 1480 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1481 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1482 } 1483 } else if (flags & MSG_PEEK) 1484 moff += len; 1485 else { 1486 if (mp != NULL) { 1487 mt = m_copym(m, 0, len, M_NOWAIT); 1488 if (__predict_false(mt == NULL)) { 1489 sounlock(so); 1490 mt = m_copym(m, 0, len, M_WAIT); 1491 solock(so); 1492 } 1493 *mp = mt; 1494 } 1495 m->m_data += len; 1496 m->m_len -= len; 1497 so->so_rcv.sb_cc -= len; 1498 } 1499 if (so->so_oobmark) { 1500 if ((flags & MSG_PEEK) == 0) { 1501 so->so_oobmark -= len; 1502 if (so->so_oobmark == 0) { 1503 so->so_state |= SS_RCVATMARK; 1504 break; 1505 } 1506 } else { 1507 offset += len; 1508 if (offset == so->so_oobmark) 1509 break; 1510 } 1511 } 1512 if (flags & MSG_EOR) 1513 break; 1514 /* 1515 * If the MSG_WAITALL flag is set (for non-atomic socket), 1516 * we must not quit until "uio->uio_resid == 0" or an error 1517 * termination. If a signal/timeout occurs, return 1518 * with a short count but without error. 1519 * Keep sockbuf locked against other readers. 1520 */ 1521 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 1522 !sosendallatonce(so) && !nextrecord) { 1523 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1524 break; 1525 /* 1526 * If we are peeking and the socket receive buffer is 1527 * full, stop since we can't get more data to peek at. 1528 */ 1529 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1530 break; 1531 /* 1532 * If we've drained the socket buffer, tell the 1533 * protocol in case it needs to do something to 1534 * get it filled again. 1535 */ 1536 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1537 (*pr->pr_usrreqs->pr_generic)(so, PRU_RCVD, 1538 NULL, (struct mbuf *)(long)flags, NULL, l); 1539 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1540 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1541 if (wakeup_state & SS_RESTARTSYS) 1542 error = ERESTART; 1543 else 1544 error = sbwait(&so->so_rcv); 1545 if (error != 0) { 1546 sbunlock(&so->so_rcv); 1547 sounlock(so); 1548 splx(s); 1549 return 0; 1550 } 1551 if ((m = so->so_rcv.sb_mb) != NULL) 1552 nextrecord = m->m_nextpkt; 1553 wakeup_state = so->so_state; 1554 } 1555 } 1556 1557 if (m && atomic) { 1558 flags |= MSG_TRUNC; 1559 if ((flags & MSG_PEEK) == 0) 1560 (void) sbdroprecord(&so->so_rcv); 1561 } 1562 if ((flags & MSG_PEEK) == 0) { 1563 if (m == NULL) { 1564 /* 1565 * First part is an inline SB_EMPTY_FIXUP(). Second 1566 * part makes sure sb_lastrecord is up-to-date if 1567 * there is still data in the socket buffer. 1568 */ 1569 so->so_rcv.sb_mb = nextrecord; 1570 if (so->so_rcv.sb_mb == NULL) { 1571 so->so_rcv.sb_mbtail = NULL; 1572 so->so_rcv.sb_lastrecord = NULL; 1573 } else if (nextrecord->m_nextpkt == NULL) 1574 so->so_rcv.sb_lastrecord = nextrecord; 1575 } 1576 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1577 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1578 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1579 (*pr->pr_usrreqs->pr_generic)(so, PRU_RCVD, NULL, 1580 (struct mbuf *)(long)flags, NULL, l); 1581 } 1582 if (orig_resid == uio->uio_resid && orig_resid && 1583 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1584 sbunlock(&so->so_rcv); 1585 goto restart; 1586 } 1587 1588 if (flagsp != NULL) 1589 *flagsp |= flags; 1590 release: 1591 sbunlock(&so->so_rcv); 1592 sounlock(so); 1593 splx(s); 1594 return error; 1595 } 1596 1597 int 1598 soshutdown(struct socket *so, int how) 1599 { 1600 const struct protosw *pr; 1601 int error; 1602 1603 KASSERT(solocked(so)); 1604 1605 pr = so->so_proto; 1606 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1607 return (EINVAL); 1608 1609 if (how == SHUT_RD || how == SHUT_RDWR) { 1610 sorflush(so); 1611 error = 0; 1612 } 1613 if (how == SHUT_WR || how == SHUT_RDWR) 1614 error = (*pr->pr_usrreqs->pr_generic)(so, 1615 PRU_SHUTDOWN, NULL, NULL, NULL, NULL); 1616 1617 return error; 1618 } 1619 1620 void 1621 sorestart(struct socket *so) 1622 { 1623 /* 1624 * An application has called close() on an fd on which another 1625 * of its threads has called a socket system call. 1626 * Mark this and wake everyone up, and code that would block again 1627 * instead returns ERESTART. 1628 * On system call re-entry the fd is validated and EBADF returned. 1629 * Any other fd will block again on the 2nd syscall. 1630 */ 1631 solock(so); 1632 so->so_state |= SS_RESTARTSYS; 1633 cv_broadcast(&so->so_cv); 1634 cv_broadcast(&so->so_snd.sb_cv); 1635 cv_broadcast(&so->so_rcv.sb_cv); 1636 sounlock(so); 1637 } 1638 1639 void 1640 sorflush(struct socket *so) 1641 { 1642 struct sockbuf *sb, asb; 1643 const struct protosw *pr; 1644 1645 KASSERT(solocked(so)); 1646 1647 sb = &so->so_rcv; 1648 pr = so->so_proto; 1649 socantrcvmore(so); 1650 sb->sb_flags |= SB_NOINTR; 1651 (void )sblock(sb, M_WAITOK); 1652 sbunlock(sb); 1653 asb = *sb; 1654 /* 1655 * Clear most of the sockbuf structure, but leave some of the 1656 * fields valid. 1657 */ 1658 memset(&sb->sb_startzero, 0, 1659 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1660 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) { 1661 sounlock(so); 1662 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1663 solock(so); 1664 } 1665 sbrelease(&asb, so); 1666 } 1667 1668 /* 1669 * internal set SOL_SOCKET options 1670 */ 1671 static int 1672 sosetopt1(struct socket *so, const struct sockopt *sopt) 1673 { 1674 int error = EINVAL, opt; 1675 int optval = 0; /* XXX: gcc */ 1676 struct linger l; 1677 struct timeval tv; 1678 1679 switch ((opt = sopt->sopt_name)) { 1680 1681 case SO_ACCEPTFILTER: 1682 error = accept_filt_setopt(so, sopt); 1683 KASSERT(solocked(so)); 1684 break; 1685 1686 case SO_LINGER: 1687 error = sockopt_get(sopt, &l, sizeof(l)); 1688 solock(so); 1689 if (error) 1690 break; 1691 if (l.l_linger < 0 || l.l_linger > USHRT_MAX || 1692 l.l_linger > (INT_MAX / hz)) { 1693 error = EDOM; 1694 break; 1695 } 1696 so->so_linger = l.l_linger; 1697 if (l.l_onoff) 1698 so->so_options |= SO_LINGER; 1699 else 1700 so->so_options &= ~SO_LINGER; 1701 break; 1702 1703 case SO_DEBUG: 1704 case SO_KEEPALIVE: 1705 case SO_DONTROUTE: 1706 case SO_USELOOPBACK: 1707 case SO_BROADCAST: 1708 case SO_REUSEADDR: 1709 case SO_REUSEPORT: 1710 case SO_OOBINLINE: 1711 case SO_TIMESTAMP: 1712 case SO_NOSIGPIPE: 1713 #ifdef SO_OTIMESTAMP 1714 case SO_OTIMESTAMP: 1715 #endif 1716 error = sockopt_getint(sopt, &optval); 1717 solock(so); 1718 if (error) 1719 break; 1720 if (optval) 1721 so->so_options |= opt; 1722 else 1723 so->so_options &= ~opt; 1724 break; 1725 1726 case SO_SNDBUF: 1727 case SO_RCVBUF: 1728 case SO_SNDLOWAT: 1729 case SO_RCVLOWAT: 1730 error = sockopt_getint(sopt, &optval); 1731 solock(so); 1732 if (error) 1733 break; 1734 1735 /* 1736 * Values < 1 make no sense for any of these 1737 * options, so disallow them. 1738 */ 1739 if (optval < 1) { 1740 error = EINVAL; 1741 break; 1742 } 1743 1744 switch (opt) { 1745 case SO_SNDBUF: 1746 if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) { 1747 error = ENOBUFS; 1748 break; 1749 } 1750 so->so_snd.sb_flags &= ~SB_AUTOSIZE; 1751 break; 1752 1753 case SO_RCVBUF: 1754 if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) { 1755 error = ENOBUFS; 1756 break; 1757 } 1758 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1759 break; 1760 1761 /* 1762 * Make sure the low-water is never greater than 1763 * the high-water. 1764 */ 1765 case SO_SNDLOWAT: 1766 if (optval > so->so_snd.sb_hiwat) 1767 optval = so->so_snd.sb_hiwat; 1768 1769 so->so_snd.sb_lowat = optval; 1770 break; 1771 1772 case SO_RCVLOWAT: 1773 if (optval > so->so_rcv.sb_hiwat) 1774 optval = so->so_rcv.sb_hiwat; 1775 1776 so->so_rcv.sb_lowat = optval; 1777 break; 1778 } 1779 break; 1780 1781 #ifdef COMPAT_50 1782 case SO_OSNDTIMEO: 1783 case SO_ORCVTIMEO: { 1784 struct timeval50 otv; 1785 error = sockopt_get(sopt, &otv, sizeof(otv)); 1786 if (error) { 1787 solock(so); 1788 break; 1789 } 1790 timeval50_to_timeval(&otv, &tv); 1791 opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO; 1792 error = 0; 1793 /*FALLTHROUGH*/ 1794 } 1795 #endif /* COMPAT_50 */ 1796 1797 case SO_SNDTIMEO: 1798 case SO_RCVTIMEO: 1799 if (error) 1800 error = sockopt_get(sopt, &tv, sizeof(tv)); 1801 solock(so); 1802 if (error) 1803 break; 1804 1805 if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) { 1806 error = EDOM; 1807 break; 1808 } 1809 1810 optval = tv.tv_sec * hz + tv.tv_usec / tick; 1811 if (optval == 0 && tv.tv_usec != 0) 1812 optval = 1; 1813 1814 switch (opt) { 1815 case SO_SNDTIMEO: 1816 so->so_snd.sb_timeo = optval; 1817 break; 1818 case SO_RCVTIMEO: 1819 so->so_rcv.sb_timeo = optval; 1820 break; 1821 } 1822 break; 1823 1824 default: 1825 solock(so); 1826 error = ENOPROTOOPT; 1827 break; 1828 } 1829 KASSERT(solocked(so)); 1830 return error; 1831 } 1832 1833 int 1834 sosetopt(struct socket *so, struct sockopt *sopt) 1835 { 1836 int error, prerr; 1837 1838 if (sopt->sopt_level == SOL_SOCKET) { 1839 error = sosetopt1(so, sopt); 1840 KASSERT(solocked(so)); 1841 } else { 1842 error = ENOPROTOOPT; 1843 solock(so); 1844 } 1845 1846 if ((error == 0 || error == ENOPROTOOPT) && 1847 so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) { 1848 /* give the protocol stack a shot */ 1849 prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt); 1850 if (prerr == 0) 1851 error = 0; 1852 else if (prerr != ENOPROTOOPT) 1853 error = prerr; 1854 } 1855 sounlock(so); 1856 return error; 1857 } 1858 1859 /* 1860 * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt() 1861 */ 1862 int 1863 so_setsockopt(struct lwp *l, struct socket *so, int level, int name, 1864 const void *val, size_t valsize) 1865 { 1866 struct sockopt sopt; 1867 int error; 1868 1869 KASSERT(valsize == 0 || val != NULL); 1870 1871 sockopt_init(&sopt, level, name, valsize); 1872 sockopt_set(&sopt, val, valsize); 1873 1874 error = sosetopt(so, &sopt); 1875 1876 sockopt_destroy(&sopt); 1877 1878 return error; 1879 } 1880 1881 /* 1882 * internal get SOL_SOCKET options 1883 */ 1884 static int 1885 sogetopt1(struct socket *so, struct sockopt *sopt) 1886 { 1887 int error, optval, opt; 1888 struct linger l; 1889 struct timeval tv; 1890 1891 switch ((opt = sopt->sopt_name)) { 1892 1893 case SO_ACCEPTFILTER: 1894 error = accept_filt_getopt(so, sopt); 1895 break; 1896 1897 case SO_LINGER: 1898 l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0; 1899 l.l_linger = so->so_linger; 1900 1901 error = sockopt_set(sopt, &l, sizeof(l)); 1902 break; 1903 1904 case SO_USELOOPBACK: 1905 case SO_DONTROUTE: 1906 case SO_DEBUG: 1907 case SO_KEEPALIVE: 1908 case SO_REUSEADDR: 1909 case SO_REUSEPORT: 1910 case SO_BROADCAST: 1911 case SO_OOBINLINE: 1912 case SO_TIMESTAMP: 1913 case SO_NOSIGPIPE: 1914 #ifdef SO_OTIMESTAMP 1915 case SO_OTIMESTAMP: 1916 #endif 1917 case SO_ACCEPTCONN: 1918 error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0); 1919 break; 1920 1921 case SO_TYPE: 1922 error = sockopt_setint(sopt, so->so_type); 1923 break; 1924 1925 case SO_ERROR: 1926 error = sockopt_setint(sopt, so->so_error); 1927 so->so_error = 0; 1928 break; 1929 1930 case SO_SNDBUF: 1931 error = sockopt_setint(sopt, so->so_snd.sb_hiwat); 1932 break; 1933 1934 case SO_RCVBUF: 1935 error = sockopt_setint(sopt, so->so_rcv.sb_hiwat); 1936 break; 1937 1938 case SO_SNDLOWAT: 1939 error = sockopt_setint(sopt, so->so_snd.sb_lowat); 1940 break; 1941 1942 case SO_RCVLOWAT: 1943 error = sockopt_setint(sopt, so->so_rcv.sb_lowat); 1944 break; 1945 1946 #ifdef COMPAT_50 1947 case SO_OSNDTIMEO: 1948 case SO_ORCVTIMEO: { 1949 struct timeval50 otv; 1950 1951 optval = (opt == SO_OSNDTIMEO ? 1952 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1953 1954 otv.tv_sec = optval / hz; 1955 otv.tv_usec = (optval % hz) * tick; 1956 1957 error = sockopt_set(sopt, &otv, sizeof(otv)); 1958 break; 1959 } 1960 #endif /* COMPAT_50 */ 1961 1962 case SO_SNDTIMEO: 1963 case SO_RCVTIMEO: 1964 optval = (opt == SO_SNDTIMEO ? 1965 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1966 1967 tv.tv_sec = optval / hz; 1968 tv.tv_usec = (optval % hz) * tick; 1969 1970 error = sockopt_set(sopt, &tv, sizeof(tv)); 1971 break; 1972 1973 case SO_OVERFLOWED: 1974 error = sockopt_setint(sopt, so->so_rcv.sb_overflowed); 1975 break; 1976 1977 default: 1978 error = ENOPROTOOPT; 1979 break; 1980 } 1981 1982 return (error); 1983 } 1984 1985 int 1986 sogetopt(struct socket *so, struct sockopt *sopt) 1987 { 1988 int error; 1989 1990 solock(so); 1991 if (sopt->sopt_level != SOL_SOCKET) { 1992 if (so->so_proto && so->so_proto->pr_ctloutput) { 1993 error = ((*so->so_proto->pr_ctloutput) 1994 (PRCO_GETOPT, so, sopt)); 1995 } else 1996 error = (ENOPROTOOPT); 1997 } else { 1998 error = sogetopt1(so, sopt); 1999 } 2000 sounlock(so); 2001 return (error); 2002 } 2003 2004 /* 2005 * alloc sockopt data buffer buffer 2006 * - will be released at destroy 2007 */ 2008 static int 2009 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag) 2010 { 2011 2012 KASSERT(sopt->sopt_size == 0); 2013 2014 if (len > sizeof(sopt->sopt_buf)) { 2015 sopt->sopt_data = kmem_zalloc(len, kmflag); 2016 if (sopt->sopt_data == NULL) 2017 return ENOMEM; 2018 } else 2019 sopt->sopt_data = sopt->sopt_buf; 2020 2021 sopt->sopt_size = len; 2022 return 0; 2023 } 2024 2025 /* 2026 * initialise sockopt storage 2027 * - MAY sleep during allocation 2028 */ 2029 void 2030 sockopt_init(struct sockopt *sopt, int level, int name, size_t size) 2031 { 2032 2033 memset(sopt, 0, sizeof(*sopt)); 2034 2035 sopt->sopt_level = level; 2036 sopt->sopt_name = name; 2037 (void)sockopt_alloc(sopt, size, KM_SLEEP); 2038 } 2039 2040 /* 2041 * destroy sockopt storage 2042 * - will release any held memory references 2043 */ 2044 void 2045 sockopt_destroy(struct sockopt *sopt) 2046 { 2047 2048 if (sopt->sopt_data != sopt->sopt_buf) 2049 kmem_free(sopt->sopt_data, sopt->sopt_size); 2050 2051 memset(sopt, 0, sizeof(*sopt)); 2052 } 2053 2054 /* 2055 * set sockopt value 2056 * - value is copied into sockopt 2057 * - memory is allocated when necessary, will not sleep 2058 */ 2059 int 2060 sockopt_set(struct sockopt *sopt, const void *buf, size_t len) 2061 { 2062 int error; 2063 2064 if (sopt->sopt_size == 0) { 2065 error = sockopt_alloc(sopt, len, KM_NOSLEEP); 2066 if (error) 2067 return error; 2068 } 2069 2070 KASSERT(sopt->sopt_size == len); 2071 memcpy(sopt->sopt_data, buf, len); 2072 return 0; 2073 } 2074 2075 /* 2076 * common case of set sockopt integer value 2077 */ 2078 int 2079 sockopt_setint(struct sockopt *sopt, int val) 2080 { 2081 2082 return sockopt_set(sopt, &val, sizeof(int)); 2083 } 2084 2085 /* 2086 * get sockopt value 2087 * - correct size must be given 2088 */ 2089 int 2090 sockopt_get(const struct sockopt *sopt, void *buf, size_t len) 2091 { 2092 2093 if (sopt->sopt_size != len) 2094 return EINVAL; 2095 2096 memcpy(buf, sopt->sopt_data, len); 2097 return 0; 2098 } 2099 2100 /* 2101 * common case of get sockopt integer value 2102 */ 2103 int 2104 sockopt_getint(const struct sockopt *sopt, int *valp) 2105 { 2106 2107 return sockopt_get(sopt, valp, sizeof(int)); 2108 } 2109 2110 /* 2111 * set sockopt value from mbuf 2112 * - ONLY for legacy code 2113 * - mbuf is released by sockopt 2114 * - will not sleep 2115 */ 2116 int 2117 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m) 2118 { 2119 size_t len; 2120 int error; 2121 2122 len = m_length(m); 2123 2124 if (sopt->sopt_size == 0) { 2125 error = sockopt_alloc(sopt, len, KM_NOSLEEP); 2126 if (error) 2127 return error; 2128 } 2129 2130 KASSERT(sopt->sopt_size == len); 2131 m_copydata(m, 0, len, sopt->sopt_data); 2132 m_freem(m); 2133 2134 return 0; 2135 } 2136 2137 /* 2138 * get sockopt value into mbuf 2139 * - ONLY for legacy code 2140 * - mbuf to be released by the caller 2141 * - will not sleep 2142 */ 2143 struct mbuf * 2144 sockopt_getmbuf(const struct sockopt *sopt) 2145 { 2146 struct mbuf *m; 2147 2148 if (sopt->sopt_size > MCLBYTES) 2149 return NULL; 2150 2151 m = m_get(M_DONTWAIT, MT_SOOPTS); 2152 if (m == NULL) 2153 return NULL; 2154 2155 if (sopt->sopt_size > MLEN) { 2156 MCLGET(m, M_DONTWAIT); 2157 if ((m->m_flags & M_EXT) == 0) { 2158 m_free(m); 2159 return NULL; 2160 } 2161 } 2162 2163 memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size); 2164 m->m_len = sopt->sopt_size; 2165 2166 return m; 2167 } 2168 2169 void 2170 sohasoutofband(struct socket *so) 2171 { 2172 2173 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 2174 selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT); 2175 } 2176 2177 static void 2178 filt_sordetach(struct knote *kn) 2179 { 2180 struct socket *so; 2181 2182 so = ((file_t *)kn->kn_obj)->f_data; 2183 solock(so); 2184 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 2185 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 2186 so->so_rcv.sb_flags &= ~SB_KNOTE; 2187 sounlock(so); 2188 } 2189 2190 /*ARGSUSED*/ 2191 static int 2192 filt_soread(struct knote *kn, long hint) 2193 { 2194 struct socket *so; 2195 int rv; 2196 2197 so = ((file_t *)kn->kn_obj)->f_data; 2198 if (hint != NOTE_SUBMIT) 2199 solock(so); 2200 kn->kn_data = so->so_rcv.sb_cc; 2201 if (so->so_state & SS_CANTRCVMORE) { 2202 kn->kn_flags |= EV_EOF; 2203 kn->kn_fflags = so->so_error; 2204 rv = 1; 2205 } else if (so->so_error) /* temporary udp error */ 2206 rv = 1; 2207 else if (kn->kn_sfflags & NOTE_LOWAT) 2208 rv = (kn->kn_data >= kn->kn_sdata); 2209 else 2210 rv = (kn->kn_data >= so->so_rcv.sb_lowat); 2211 if (hint != NOTE_SUBMIT) 2212 sounlock(so); 2213 return rv; 2214 } 2215 2216 static void 2217 filt_sowdetach(struct knote *kn) 2218 { 2219 struct socket *so; 2220 2221 so = ((file_t *)kn->kn_obj)->f_data; 2222 solock(so); 2223 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 2224 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 2225 so->so_snd.sb_flags &= ~SB_KNOTE; 2226 sounlock(so); 2227 } 2228 2229 /*ARGSUSED*/ 2230 static int 2231 filt_sowrite(struct knote *kn, long hint) 2232 { 2233 struct socket *so; 2234 int rv; 2235 2236 so = ((file_t *)kn->kn_obj)->f_data; 2237 if (hint != NOTE_SUBMIT) 2238 solock(so); 2239 kn->kn_data = sbspace(&so->so_snd); 2240 if (so->so_state & SS_CANTSENDMORE) { 2241 kn->kn_flags |= EV_EOF; 2242 kn->kn_fflags = so->so_error; 2243 rv = 1; 2244 } else if (so->so_error) /* temporary udp error */ 2245 rv = 1; 2246 else if (((so->so_state & SS_ISCONNECTED) == 0) && 2247 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 2248 rv = 0; 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_snd.sb_lowat); 2253 if (hint != NOTE_SUBMIT) 2254 sounlock(so); 2255 return rv; 2256 } 2257 2258 /*ARGSUSED*/ 2259 static int 2260 filt_solisten(struct knote *kn, long hint) 2261 { 2262 struct socket *so; 2263 int rv; 2264 2265 so = ((file_t *)kn->kn_obj)->f_data; 2266 2267 /* 2268 * Set kn_data to number of incoming connections, not 2269 * counting partial (incomplete) connections. 2270 */ 2271 if (hint != NOTE_SUBMIT) 2272 solock(so); 2273 kn->kn_data = so->so_qlen; 2274 rv = (kn->kn_data > 0); 2275 if (hint != NOTE_SUBMIT) 2276 sounlock(so); 2277 return rv; 2278 } 2279 2280 static const struct filterops solisten_filtops = 2281 { 1, NULL, filt_sordetach, filt_solisten }; 2282 static const struct filterops soread_filtops = 2283 { 1, NULL, filt_sordetach, filt_soread }; 2284 static const struct filterops sowrite_filtops = 2285 { 1, NULL, filt_sowdetach, filt_sowrite }; 2286 2287 int 2288 soo_kqfilter(struct file *fp, struct knote *kn) 2289 { 2290 struct socket *so; 2291 struct sockbuf *sb; 2292 2293 so = ((file_t *)kn->kn_obj)->f_data; 2294 solock(so); 2295 switch (kn->kn_filter) { 2296 case EVFILT_READ: 2297 if (so->so_options & SO_ACCEPTCONN) 2298 kn->kn_fop = &solisten_filtops; 2299 else 2300 kn->kn_fop = &soread_filtops; 2301 sb = &so->so_rcv; 2302 break; 2303 case EVFILT_WRITE: 2304 kn->kn_fop = &sowrite_filtops; 2305 sb = &so->so_snd; 2306 break; 2307 default: 2308 sounlock(so); 2309 return (EINVAL); 2310 } 2311 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 2312 sb->sb_flags |= SB_KNOTE; 2313 sounlock(so); 2314 return (0); 2315 } 2316 2317 static int 2318 sodopoll(struct socket *so, int events) 2319 { 2320 int revents; 2321 2322 revents = 0; 2323 2324 if (events & (POLLIN | POLLRDNORM)) 2325 if (soreadable(so)) 2326 revents |= events & (POLLIN | POLLRDNORM); 2327 2328 if (events & (POLLOUT | POLLWRNORM)) 2329 if (sowritable(so)) 2330 revents |= events & (POLLOUT | POLLWRNORM); 2331 2332 if (events & (POLLPRI | POLLRDBAND)) 2333 if (so->so_oobmark || (so->so_state & SS_RCVATMARK)) 2334 revents |= events & (POLLPRI | POLLRDBAND); 2335 2336 return revents; 2337 } 2338 2339 int 2340 sopoll(struct socket *so, int events) 2341 { 2342 int revents = 0; 2343 2344 #ifndef DIAGNOSTIC 2345 /* 2346 * Do a quick, unlocked check in expectation that the socket 2347 * will be ready for I/O. Don't do this check if DIAGNOSTIC, 2348 * as the solocked() assertions will fail. 2349 */ 2350 if ((revents = sodopoll(so, events)) != 0) 2351 return revents; 2352 #endif 2353 2354 solock(so); 2355 if ((revents = sodopoll(so, events)) == 0) { 2356 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 2357 selrecord(curlwp, &so->so_rcv.sb_sel); 2358 so->so_rcv.sb_flags |= SB_NOTIFY; 2359 } 2360 2361 if (events & (POLLOUT | POLLWRNORM)) { 2362 selrecord(curlwp, &so->so_snd.sb_sel); 2363 so->so_snd.sb_flags |= SB_NOTIFY; 2364 } 2365 } 2366 sounlock(so); 2367 2368 return revents; 2369 } 2370 2371 2372 #include <sys/sysctl.h> 2373 2374 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 2375 static int sysctl_kern_sbmax(SYSCTLFN_PROTO); 2376 2377 /* 2378 * sysctl helper routine for kern.somaxkva. ensures that the given 2379 * value is not too small. 2380 * (XXX should we maybe make sure it's not too large as well?) 2381 */ 2382 static int 2383 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 2384 { 2385 int error, new_somaxkva; 2386 struct sysctlnode node; 2387 2388 new_somaxkva = somaxkva; 2389 node = *rnode; 2390 node.sysctl_data = &new_somaxkva; 2391 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2392 if (error || newp == NULL) 2393 return (error); 2394 2395 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 2396 return (EINVAL); 2397 2398 mutex_enter(&so_pendfree_lock); 2399 somaxkva = new_somaxkva; 2400 cv_broadcast(&socurkva_cv); 2401 mutex_exit(&so_pendfree_lock); 2402 2403 return (error); 2404 } 2405 2406 /* 2407 * sysctl helper routine for kern.sbmax. Basically just ensures that 2408 * any new value is not too small. 2409 */ 2410 static int 2411 sysctl_kern_sbmax(SYSCTLFN_ARGS) 2412 { 2413 int error, new_sbmax; 2414 struct sysctlnode node; 2415 2416 new_sbmax = sb_max; 2417 node = *rnode; 2418 node.sysctl_data = &new_sbmax; 2419 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2420 if (error || newp == NULL) 2421 return (error); 2422 2423 KERNEL_LOCK(1, NULL); 2424 error = sb_max_set(new_sbmax); 2425 KERNEL_UNLOCK_ONE(NULL); 2426 2427 return (error); 2428 } 2429 2430 static void 2431 sysctl_kern_socket_setup(void) 2432 { 2433 2434 KASSERT(socket_sysctllog == NULL); 2435 2436 sysctl_createv(&socket_sysctllog, 0, NULL, NULL, 2437 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2438 CTLTYPE_INT, "somaxkva", 2439 SYSCTL_DESCR("Maximum amount of kernel memory to be " 2440 "used for socket buffers"), 2441 sysctl_kern_somaxkva, 0, NULL, 0, 2442 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 2443 2444 sysctl_createv(&socket_sysctllog, 0, NULL, NULL, 2445 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2446 CTLTYPE_INT, "sbmax", 2447 SYSCTL_DESCR("Maximum socket buffer size"), 2448 sysctl_kern_sbmax, 0, NULL, 0, 2449 CTL_KERN, KERN_SBMAX, CTL_EOL); 2450 } 2451
Indexes created Sat Sep 20 22:09:52 GMT 2025