uipc_socket.c revision 1.264.2.2
1 /* $NetBSD: uipc_socket.c,v 1.264.2.2 2020/04/08 14:08:52 martin 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 1.264.2.2 2020/04/08 14:08:52 martin 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 #include <sys/compat_stub.h> 108 109 #include <compat/sys/time.h> 110 #include <compat/sys/socket.h> 111 112 #include <uvm/uvm_extern.h> 113 #include <uvm/uvm_loan.h> 114 #include <uvm/uvm_page.h> 115 116 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 117 118 extern const struct fileops socketops; 119 120 static int sooptions; 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 vaddr_t 209 sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so) 210 { 211 vaddr_t lva; 212 213 if (sokvareserve(so, len) == 0) 214 return 0; 215 216 lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask, 217 UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA); 218 if (lva == 0) { 219 sokvaunreserve(len); 220 return 0; 221 } 222 223 return lva; 224 } 225 226 /* 227 * sokvafree: free kva for loan. 228 */ 229 void 230 sokvafree(vaddr_t sva, vsize_t len) 231 { 232 233 uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY); 234 sokvaunreserve(len); 235 } 236 237 static void 238 sodoloanfree(struct vm_page **pgs, void *buf, size_t size) 239 { 240 vaddr_t sva, eva; 241 vsize_t len; 242 int npgs; 243 244 KASSERT(pgs != NULL); 245 246 eva = round_page((vaddr_t) buf + size); 247 sva = trunc_page((vaddr_t) buf); 248 len = eva - sva; 249 npgs = len >> PAGE_SHIFT; 250 251 pmap_kremove(sva, len); 252 pmap_update(pmap_kernel()); 253 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 254 sokvafree(sva, len); 255 } 256 257 /* 258 * sopendfree_thread: free mbufs on "pendfree" list. Unlock and relock 259 * so_pendfree_lock when freeing mbufs. 260 */ 261 static void 262 sopendfree_thread(void *v) 263 { 264 struct mbuf *m, *next; 265 size_t rv; 266 267 mutex_enter(&so_pendfree_lock); 268 269 for (;;) { 270 rv = 0; 271 while (so_pendfree != NULL) { 272 m = so_pendfree; 273 so_pendfree = NULL; 274 mutex_exit(&so_pendfree_lock); 275 276 for (; m != NULL; m = next) { 277 next = m->m_next; 278 KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 279 0); 280 KASSERT(m->m_ext.ext_refcnt == 0); 281 282 rv += m->m_ext.ext_size; 283 sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf, 284 m->m_ext.ext_size); 285 pool_cache_put(mb_cache, m); 286 } 287 288 mutex_enter(&so_pendfree_lock); 289 } 290 if (rv) 291 cv_broadcast(&socurkva_cv); 292 cv_wait(&pendfree_thread_cv, &so_pendfree_lock); 293 } 294 panic("sopendfree_thread"); 295 /* NOTREACHED */ 296 } 297 298 void 299 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg) 300 { 301 302 KASSERT(m != NULL); 303 304 /* 305 * postpone freeing mbuf. 306 * 307 * we can't do it in interrupt context 308 * because we need to put kva back to kernel_map. 309 */ 310 311 mutex_enter(&so_pendfree_lock); 312 m->m_next = so_pendfree; 313 so_pendfree = m; 314 cv_signal(&pendfree_thread_cv); 315 mutex_exit(&so_pendfree_lock); 316 } 317 318 static long 319 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 320 { 321 struct iovec *iov = uio->uio_iov; 322 vaddr_t sva, eva; 323 vsize_t len; 324 vaddr_t lva; 325 int npgs, error; 326 vaddr_t va; 327 int i; 328 329 if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace)) 330 return 0; 331 332 if (iov->iov_len < (size_t) space) 333 space = iov->iov_len; 334 if (space > SOCK_LOAN_CHUNK) 335 space = SOCK_LOAN_CHUNK; 336 337 eva = round_page((vaddr_t) iov->iov_base + space); 338 sva = trunc_page((vaddr_t) iov->iov_base); 339 len = eva - sva; 340 npgs = len >> PAGE_SHIFT; 341 342 KASSERT(npgs <= M_EXT_MAXPAGES); 343 344 lva = sokvaalloc(sva, len, so); 345 if (lva == 0) 346 return 0; 347 348 error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len, 349 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 350 if (error) { 351 sokvafree(lva, len); 352 return 0; 353 } 354 355 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 356 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 357 VM_PROT_READ, 0); 358 pmap_update(pmap_kernel()); 359 360 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 361 362 MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so); 363 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 364 365 uio->uio_resid -= space; 366 /* uio_offset not updated, not set/used for write(2) */ 367 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space; 368 uio->uio_iov->iov_len -= space; 369 if (uio->uio_iov->iov_len == 0) { 370 uio->uio_iov++; 371 uio->uio_iovcnt--; 372 } 373 374 return space; 375 } 376 377 static int 378 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 379 void *arg0, void *arg1, void *arg2, void *arg3) 380 { 381 int result; 382 enum kauth_network_req req; 383 384 result = KAUTH_RESULT_DEFER; 385 req = (enum kauth_network_req)(uintptr_t)arg0; 386 387 if ((action != KAUTH_NETWORK_SOCKET) && 388 (action != KAUTH_NETWORK_BIND)) 389 return result; 390 391 switch (req) { 392 case KAUTH_REQ_NETWORK_BIND_PORT: 393 result = KAUTH_RESULT_ALLOW; 394 break; 395 396 case KAUTH_REQ_NETWORK_SOCKET_DROP: { 397 /* Normal users can only drop their own connections. */ 398 struct socket *so = (struct socket *)arg1; 399 400 if (so->so_cred && proc_uidmatch(cred, so->so_cred) == 0) 401 result = KAUTH_RESULT_ALLOW; 402 403 break; 404 } 405 406 case KAUTH_REQ_NETWORK_SOCKET_OPEN: 407 /* We allow "raw" routing/bluetooth sockets to anyone. */ 408 switch ((u_long)arg1) { 409 case PF_ROUTE: 410 case PF_OROUTE: 411 case PF_BLUETOOTH: 412 case PF_CAN: 413 result = KAUTH_RESULT_ALLOW; 414 break; 415 default: 416 /* Privileged, let secmodel handle this. */ 417 if ((u_long)arg2 == SOCK_RAW) 418 break; 419 result = KAUTH_RESULT_ALLOW; 420 break; 421 } 422 break; 423 424 case KAUTH_REQ_NETWORK_SOCKET_CANSEE: 425 result = KAUTH_RESULT_ALLOW; 426 427 break; 428 429 default: 430 break; 431 } 432 433 return result; 434 } 435 436 void 437 soinit(void) 438 { 439 440 sysctl_kern_socket_setup(); 441 442 mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM); 443 softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 444 cv_init(&socurkva_cv, "sokva"); 445 cv_init(&pendfree_thread_cv, "sopendfr"); 446 soinit2(); 447 448 /* Set the initial adjusted socket buffer size. */ 449 if (sb_max_set(sb_max)) 450 panic("bad initial sb_max value: %lu", sb_max); 451 452 socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK, 453 socket_listener_cb, NULL); 454 } 455 456 void 457 soinit1(void) 458 { 459 int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, 460 sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree"); 461 if (error) 462 panic("soinit1 %d", error); 463 } 464 465 /* 466 * socreate: create a new socket of the specified type and the protocol. 467 * 468 * => Caller may specify another socket for lock sharing (must not be held). 469 * => Returns the new socket without lock held. 470 */ 471 int 472 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l, 473 struct socket *lockso) 474 { 475 const struct protosw *prp; 476 struct socket *so; 477 uid_t uid; 478 int error; 479 kmutex_t *lock; 480 481 error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET, 482 KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type), 483 KAUTH_ARG(proto)); 484 if (error != 0) 485 return error; 486 487 if (proto) 488 prp = pffindproto(dom, proto, type); 489 else 490 prp = pffindtype(dom, type); 491 if (prp == NULL) { 492 /* no support for domain */ 493 if (pffinddomain(dom) == 0) 494 return EAFNOSUPPORT; 495 /* no support for socket type */ 496 if (proto == 0 && type != 0) 497 return EPROTOTYPE; 498 return EPROTONOSUPPORT; 499 } 500 if (prp->pr_usrreqs == NULL) 501 return EPROTONOSUPPORT; 502 if (prp->pr_type != type) 503 return EPROTOTYPE; 504 505 so = soget(true); 506 so->so_type = type; 507 so->so_proto = prp; 508 so->so_send = sosend; 509 so->so_receive = soreceive; 510 so->so_options = sooptions; 511 #ifdef MBUFTRACE 512 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 513 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 514 so->so_mowner = &prp->pr_domain->dom_mowner; 515 #endif 516 uid = kauth_cred_geteuid(l->l_cred); 517 so->so_uidinfo = uid_find(uid); 518 so->so_cpid = l->l_proc->p_pid; 519 520 /* 521 * Lock assigned and taken during PCB attach, unless we share 522 * the lock with another socket, e.g. socketpair(2) case. 523 */ 524 if (lockso) { 525 lock = lockso->so_lock; 526 so->so_lock = lock; 527 mutex_obj_hold(lock); 528 mutex_enter(lock); 529 } 530 531 /* Attach the PCB (returns with the socket lock held). */ 532 error = (*prp->pr_usrreqs->pr_attach)(so, proto); 533 KASSERT(solocked(so)); 534 535 if (error) { 536 KASSERT(so->so_pcb == NULL); 537 so->so_state |= SS_NOFDREF; 538 sofree(so); 539 return error; 540 } 541 so->so_cred = kauth_cred_dup(l->l_cred); 542 sounlock(so); 543 544 *aso = so; 545 return 0; 546 } 547 548 /* 549 * fsocreate: create a socket and a file descriptor associated with it. 550 * 551 * => On success, write file descriptor to fdout and return zero. 552 * => On failure, return non-zero; *fdout will be undefined. 553 */ 554 int 555 fsocreate(int domain, struct socket **sop, int type, int proto, int *fdout) 556 { 557 lwp_t *l = curlwp; 558 int error, fd, flags; 559 struct socket *so; 560 struct file *fp; 561 562 if ((error = fd_allocfile(&fp, &fd)) != 0) { 563 return error; 564 } 565 flags = type & SOCK_FLAGS_MASK; 566 fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0); 567 fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)| 568 ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0); 569 fp->f_type = DTYPE_SOCKET; 570 fp->f_ops = &socketops; 571 572 type &= ~SOCK_FLAGS_MASK; 573 error = socreate(domain, &so, type, proto, l, NULL); 574 if (error) { 575 fd_abort(curproc, fp, fd); 576 return error; 577 } 578 if (flags & SOCK_NONBLOCK) { 579 so->so_state |= SS_NBIO; 580 } 581 fp->f_socket = so; 582 fd_affix(curproc, fp, fd); 583 584 if (sop != NULL) { 585 *sop = so; 586 } 587 *fdout = fd; 588 return error; 589 } 590 591 int 592 sofamily(const struct socket *so) 593 { 594 const struct protosw *pr; 595 const struct domain *dom; 596 597 if ((pr = so->so_proto) == NULL) 598 return AF_UNSPEC; 599 if ((dom = pr->pr_domain) == NULL) 600 return AF_UNSPEC; 601 return dom->dom_family; 602 } 603 604 int 605 sobind(struct socket *so, struct sockaddr *nam, struct lwp *l) 606 { 607 int error; 608 609 solock(so); 610 if (nam->sa_family != so->so_proto->pr_domain->dom_family) { 611 sounlock(so); 612 return EAFNOSUPPORT; 613 } 614 error = (*so->so_proto->pr_usrreqs->pr_bind)(so, nam, l); 615 sounlock(so); 616 return error; 617 } 618 619 int 620 solisten(struct socket *so, int backlog, struct lwp *l) 621 { 622 int error; 623 short oldopt, oldqlimit; 624 625 solock(so); 626 if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 627 SS_ISDISCONNECTING)) != 0) { 628 sounlock(so); 629 return EINVAL; 630 } 631 oldopt = so->so_options; 632 oldqlimit = so->so_qlimit; 633 if (TAILQ_EMPTY(&so->so_q)) 634 so->so_options |= SO_ACCEPTCONN; 635 if (backlog < 0) 636 backlog = 0; 637 so->so_qlimit = uimin(backlog, somaxconn); 638 639 error = (*so->so_proto->pr_usrreqs->pr_listen)(so, l); 640 if (error != 0) { 641 so->so_options = oldopt; 642 so->so_qlimit = oldqlimit; 643 sounlock(so); 644 return error; 645 } 646 sounlock(so); 647 return 0; 648 } 649 650 void 651 sofree(struct socket *so) 652 { 653 u_int refs; 654 655 KASSERT(solocked(so)); 656 657 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) { 658 sounlock(so); 659 return; 660 } 661 if (so->so_head) { 662 /* 663 * We must not decommission a socket that's on the accept(2) 664 * queue. If we do, then accept(2) may hang after select(2) 665 * indicated that the listening socket was ready. 666 */ 667 if (!soqremque(so, 0)) { 668 sounlock(so); 669 return; 670 } 671 } 672 if (so->so_rcv.sb_hiwat) 673 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 674 RLIM_INFINITY); 675 if (so->so_snd.sb_hiwat) 676 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 677 RLIM_INFINITY); 678 sbrelease(&so->so_snd, so); 679 KASSERT(!cv_has_waiters(&so->so_cv)); 680 KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv)); 681 KASSERT(!cv_has_waiters(&so->so_snd.sb_cv)); 682 sorflush(so); 683 refs = so->so_aborting; /* XXX */ 684 /* Remove acccept filter if one is present. */ 685 if (so->so_accf != NULL) 686 (void)accept_filt_clear(so); 687 sounlock(so); 688 if (refs == 0) /* XXX */ 689 soput(so); 690 } 691 692 /* 693 * soclose: close a socket on last file table reference removal. 694 * Initiate disconnect if connected. Free socket when disconnect complete. 695 */ 696 int 697 soclose(struct socket *so) 698 { 699 struct socket *so2; 700 int error = 0; 701 702 solock(so); 703 if (so->so_options & SO_ACCEPTCONN) { 704 for (;;) { 705 if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 706 KASSERT(solocked2(so, so2)); 707 (void) soqremque(so2, 0); 708 /* soabort drops the lock. */ 709 (void) soabort(so2); 710 solock(so); 711 continue; 712 } 713 if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 714 KASSERT(solocked2(so, so2)); 715 (void) soqremque(so2, 1); 716 /* soabort drops the lock. */ 717 (void) soabort(so2); 718 solock(so); 719 continue; 720 } 721 break; 722 } 723 } 724 if (so->so_pcb == NULL) 725 goto discard; 726 if (so->so_state & SS_ISCONNECTED) { 727 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 728 error = sodisconnect(so); 729 if (error) 730 goto drop; 731 } 732 if (so->so_options & SO_LINGER) { 733 if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) == 734 (SS_ISDISCONNECTING|SS_NBIO)) 735 goto drop; 736 while (so->so_state & SS_ISCONNECTED) { 737 error = sowait(so, true, so->so_linger * hz); 738 if (error) 739 break; 740 } 741 } 742 } 743 drop: 744 if (so->so_pcb) { 745 KASSERT(solocked(so)); 746 (*so->so_proto->pr_usrreqs->pr_detach)(so); 747 } 748 discard: 749 KASSERT((so->so_state & SS_NOFDREF) == 0); 750 kauth_cred_free(so->so_cred); 751 so->so_cred = NULL; 752 so->so_state |= SS_NOFDREF; 753 sofree(so); 754 return error; 755 } 756 757 /* 758 * Must be called with the socket locked.. Will return with it unlocked. 759 */ 760 int 761 soabort(struct socket *so) 762 { 763 u_int refs; 764 int error; 765 766 KASSERT(solocked(so)); 767 KASSERT(so->so_head == NULL); 768 769 so->so_aborting++; /* XXX */ 770 error = (*so->so_proto->pr_usrreqs->pr_abort)(so); 771 refs = --so->so_aborting; /* XXX */ 772 if (error || (refs == 0)) { 773 sofree(so); 774 } else { 775 sounlock(so); 776 } 777 return error; 778 } 779 780 int 781 soaccept(struct socket *so, struct sockaddr *nam) 782 { 783 int error; 784 785 KASSERT(solocked(so)); 786 KASSERT((so->so_state & SS_NOFDREF) != 0); 787 788 so->so_state &= ~SS_NOFDREF; 789 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 790 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 791 error = (*so->so_proto->pr_usrreqs->pr_accept)(so, nam); 792 else 793 error = ECONNABORTED; 794 795 return error; 796 } 797 798 int 799 soconnect(struct socket *so, struct sockaddr *nam, struct lwp *l) 800 { 801 int error; 802 803 KASSERT(solocked(so)); 804 805 if (so->so_options & SO_ACCEPTCONN) 806 return EOPNOTSUPP; 807 /* 808 * If protocol is connection-based, can only connect once. 809 * Otherwise, if connected, try to disconnect first. 810 * This allows user to disconnect by connecting to, e.g., 811 * a null address. 812 */ 813 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 814 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 815 (error = sodisconnect(so)))) { 816 error = EISCONN; 817 } else { 818 if (nam->sa_family != so->so_proto->pr_domain->dom_family) { 819 return EAFNOSUPPORT; 820 } 821 error = (*so->so_proto->pr_usrreqs->pr_connect)(so, nam, l); 822 } 823 824 return error; 825 } 826 827 int 828 soconnect2(struct socket *so1, struct socket *so2) 829 { 830 KASSERT(solocked2(so1, so2)); 831 832 return (*so1->so_proto->pr_usrreqs->pr_connect2)(so1, so2); 833 } 834 835 int 836 sodisconnect(struct socket *so) 837 { 838 int error; 839 840 KASSERT(solocked(so)); 841 842 if ((so->so_state & SS_ISCONNECTED) == 0) { 843 error = ENOTCONN; 844 } else if (so->so_state & SS_ISDISCONNECTING) { 845 error = EALREADY; 846 } else { 847 error = (*so->so_proto->pr_usrreqs->pr_disconnect)(so); 848 } 849 return error; 850 } 851 852 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 853 /* 854 * Send on a socket. 855 * If send must go all at once and message is larger than 856 * send buffering, then hard error. 857 * Lock against other senders. 858 * If must go all at once and not enough room now, then 859 * inform user that this would block and do nothing. 860 * Otherwise, if nonblocking, send as much as possible. 861 * The data to be sent is described by "uio" if nonzero, 862 * otherwise by the mbuf chain "top" (which must be null 863 * if uio is not). Data provided in mbuf chain must be small 864 * enough to send all at once. 865 * 866 * Returns nonzero on error, timeout or signal; callers 867 * must check for short counts if EINTR/ERESTART are returned. 868 * Data and control buffers are freed on return. 869 */ 870 int 871 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 872 struct mbuf *top, struct mbuf *control, int flags, struct lwp *l) 873 { 874 struct mbuf **mp, *m; 875 long space, len, resid, clen, mlen; 876 int error, s, dontroute, atomic; 877 short wakeup_state = 0; 878 879 clen = 0; 880 881 /* 882 * solock() provides atomicity of access. splsoftnet() prevents 883 * protocol processing soft interrupts from interrupting us and 884 * blocking (expensive). 885 */ 886 s = splsoftnet(); 887 solock(so); 888 atomic = sosendallatonce(so) || top; 889 if (uio) 890 resid = uio->uio_resid; 891 else 892 resid = top->m_pkthdr.len; 893 /* 894 * In theory resid should be unsigned. 895 * However, space must be signed, as it might be less than 0 896 * if we over-committed, and we must use a signed comparison 897 * of space and resid. On the other hand, a negative resid 898 * causes us to loop sending 0-length segments to the protocol. 899 */ 900 if (resid < 0) { 901 error = EINVAL; 902 goto out; 903 } 904 dontroute = 905 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 906 (so->so_proto->pr_flags & PR_ATOMIC); 907 l->l_ru.ru_msgsnd++; 908 if (control) 909 clen = control->m_len; 910 restart: 911 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 912 goto out; 913 do { 914 if (so->so_state & SS_CANTSENDMORE) { 915 error = EPIPE; 916 goto release; 917 } 918 if (so->so_error) { 919 error = so->so_error; 920 so->so_error = 0; 921 goto release; 922 } 923 if ((so->so_state & SS_ISCONNECTED) == 0) { 924 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 925 if (resid || clen == 0) { 926 error = ENOTCONN; 927 goto release; 928 } 929 } else if (addr == NULL) { 930 error = EDESTADDRREQ; 931 goto release; 932 } 933 } 934 space = sbspace(&so->so_snd); 935 if (flags & MSG_OOB) 936 space += 1024; 937 if ((atomic && resid > so->so_snd.sb_hiwat) || 938 clen > so->so_snd.sb_hiwat) { 939 error = EMSGSIZE; 940 goto release; 941 } 942 if (space < resid + clen && 943 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 944 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) { 945 error = EWOULDBLOCK; 946 goto release; 947 } 948 sbunlock(&so->so_snd); 949 if (wakeup_state & SS_RESTARTSYS) { 950 error = ERESTART; 951 goto out; 952 } 953 error = sbwait(&so->so_snd); 954 if (error) 955 goto out; 956 wakeup_state = so->so_state; 957 goto restart; 958 } 959 wakeup_state = 0; 960 mp = ⊤ 961 space -= clen; 962 do { 963 if (uio == NULL) { 964 /* 965 * Data is prepackaged in "top". 966 */ 967 resid = 0; 968 if (flags & MSG_EOR) 969 top->m_flags |= M_EOR; 970 } else do { 971 sounlock(so); 972 splx(s); 973 if (top == NULL) { 974 m = m_gethdr(M_WAIT, MT_DATA); 975 mlen = MHLEN; 976 m->m_pkthdr.len = 0; 977 m_reset_rcvif(m); 978 } else { 979 m = m_get(M_WAIT, MT_DATA); 980 mlen = MLEN; 981 } 982 MCLAIM(m, so->so_snd.sb_mowner); 983 if (sock_loan_thresh >= 0 && 984 uio->uio_iov->iov_len >= sock_loan_thresh && 985 space >= sock_loan_thresh && 986 (len = sosend_loan(so, uio, m, 987 space)) != 0) { 988 SOSEND_COUNTER_INCR(&sosend_loan_big); 989 space -= len; 990 goto have_data; 991 } 992 if (resid >= MINCLSIZE && space >= MCLBYTES) { 993 SOSEND_COUNTER_INCR(&sosend_copy_big); 994 m_clget(m, M_DONTWAIT); 995 if ((m->m_flags & M_EXT) == 0) 996 goto nopages; 997 mlen = MCLBYTES; 998 if (atomic && top == 0) { 999 len = lmin(MCLBYTES - max_hdr, 1000 resid); 1001 m->m_data += max_hdr; 1002 } else 1003 len = lmin(MCLBYTES, resid); 1004 space -= len; 1005 } else { 1006 nopages: 1007 SOSEND_COUNTER_INCR(&sosend_copy_small); 1008 len = lmin(lmin(mlen, resid), space); 1009 space -= len; 1010 /* 1011 * For datagram protocols, leave room 1012 * for protocol headers in first mbuf. 1013 */ 1014 if (atomic && top == 0 && len < mlen) 1015 m_align(m, len); 1016 } 1017 error = uiomove(mtod(m, void *), (int)len, uio); 1018 have_data: 1019 resid = uio->uio_resid; 1020 m->m_len = len; 1021 *mp = m; 1022 top->m_pkthdr.len += len; 1023 s = splsoftnet(); 1024 solock(so); 1025 if (error != 0) 1026 goto release; 1027 mp = &m->m_next; 1028 if (resid <= 0) { 1029 if (flags & MSG_EOR) 1030 top->m_flags |= M_EOR; 1031 break; 1032 } 1033 } while (space > 0 && atomic); 1034 1035 if (so->so_state & SS_CANTSENDMORE) { 1036 error = EPIPE; 1037 goto release; 1038 } 1039 if (dontroute) 1040 so->so_options |= SO_DONTROUTE; 1041 if (resid > 0) 1042 so->so_state |= SS_MORETOCOME; 1043 if (flags & MSG_OOB) { 1044 error = (*so->so_proto->pr_usrreqs->pr_sendoob)( 1045 so, top, control); 1046 } else { 1047 error = (*so->so_proto->pr_usrreqs->pr_send)(so, 1048 top, addr, control, l); 1049 } 1050 if (dontroute) 1051 so->so_options &= ~SO_DONTROUTE; 1052 if (resid > 0) 1053 so->so_state &= ~SS_MORETOCOME; 1054 clen = 0; 1055 control = NULL; 1056 top = NULL; 1057 mp = ⊤ 1058 if (error != 0) 1059 goto release; 1060 } while (resid && space > 0); 1061 } while (resid); 1062 1063 release: 1064 sbunlock(&so->so_snd); 1065 out: 1066 sounlock(so); 1067 splx(s); 1068 if (top) 1069 m_freem(top); 1070 if (control) 1071 m_freem(control); 1072 return error; 1073 } 1074 1075 /* 1076 * Following replacement or removal of the first mbuf on the first 1077 * mbuf chain of a socket buffer, push necessary state changes back 1078 * into the socket buffer so that other consumers see the values 1079 * consistently. 'nextrecord' is the caller's locally stored value of 1080 * the original value of sb->sb_mb->m_nextpkt which must be restored 1081 * when the lead mbuf changes. NOTE: 'nextrecord' may be NULL. 1082 */ 1083 static void 1084 sbsync(struct sockbuf *sb, struct mbuf *nextrecord) 1085 { 1086 1087 KASSERT(solocked(sb->sb_so)); 1088 1089 /* 1090 * First, update for the new value of nextrecord. If necessary, 1091 * make it the first record. 1092 */ 1093 if (sb->sb_mb != NULL) 1094 sb->sb_mb->m_nextpkt = nextrecord; 1095 else 1096 sb->sb_mb = nextrecord; 1097 1098 /* 1099 * Now update any dependent socket buffer fields to reflect 1100 * the new state. This is an inline of SB_EMPTY_FIXUP, with 1101 * the addition of a second clause that takes care of the 1102 * case where sb_mb has been updated, but remains the last 1103 * record. 1104 */ 1105 if (sb->sb_mb == NULL) { 1106 sb->sb_mbtail = NULL; 1107 sb->sb_lastrecord = NULL; 1108 } else if (sb->sb_mb->m_nextpkt == NULL) 1109 sb->sb_lastrecord = sb->sb_mb; 1110 } 1111 1112 /* 1113 * Implement receive operations on a socket. 1114 * 1115 * We depend on the way that records are added to the sockbuf by sbappend*. In 1116 * particular, each record (mbufs linked through m_next) must begin with an 1117 * address if the protocol so specifies, followed by an optional mbuf or mbufs 1118 * containing ancillary data, and then zero or more mbufs of data. 1119 * 1120 * In order to avoid blocking network interrupts for the entire time here, we 1121 * splx() while doing the actual copy to user space. Although the sockbuf is 1122 * locked, new data may still be appended, and thus we must maintain 1123 * consistency of the sockbuf during that time. 1124 * 1125 * The caller may receive the data as a single mbuf chain by supplying an mbuf 1126 * **mp0 for use in returning the chain. The uio is then used only for the 1127 * count in uio_resid. 1128 */ 1129 int 1130 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 1131 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1132 { 1133 struct lwp *l = curlwp; 1134 struct mbuf *m, **mp, *mt; 1135 size_t len, offset, moff, orig_resid; 1136 int atomic, flags, error, s, type; 1137 const struct protosw *pr; 1138 struct mbuf *nextrecord; 1139 int mbuf_removed = 0; 1140 const struct domain *dom; 1141 short wakeup_state = 0; 1142 1143 pr = so->so_proto; 1144 atomic = pr->pr_flags & PR_ATOMIC; 1145 dom = pr->pr_domain; 1146 mp = mp0; 1147 type = 0; 1148 orig_resid = uio->uio_resid; 1149 1150 if (paddr != NULL) 1151 *paddr = NULL; 1152 if (controlp != NULL) 1153 *controlp = NULL; 1154 if (flagsp != NULL) 1155 flags = *flagsp &~ MSG_EOR; 1156 else 1157 flags = 0; 1158 1159 if (flags & MSG_OOB) { 1160 m = m_get(M_WAIT, MT_DATA); 1161 solock(so); 1162 error = (*pr->pr_usrreqs->pr_recvoob)(so, m, flags & MSG_PEEK); 1163 sounlock(so); 1164 if (error) 1165 goto bad; 1166 do { 1167 error = uiomove(mtod(m, void *), 1168 MIN(uio->uio_resid, m->m_len), uio); 1169 m = m_free(m); 1170 } while (uio->uio_resid > 0 && error == 0 && m); 1171 /* We consumed the oob data, no more oobmark. */ 1172 so->so_oobmark = 0; 1173 so->so_state &= ~SS_RCVATMARK; 1174 bad: 1175 if (m != NULL) 1176 m_freem(m); 1177 return error; 1178 } 1179 if (mp != NULL) 1180 *mp = NULL; 1181 1182 /* 1183 * solock() provides atomicity of access. splsoftnet() prevents 1184 * protocol processing soft interrupts from interrupting us and 1185 * blocking (expensive). 1186 */ 1187 s = splsoftnet(); 1188 solock(so); 1189 restart: 1190 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) { 1191 sounlock(so); 1192 splx(s); 1193 return error; 1194 } 1195 m = so->so_rcv.sb_mb; 1196 1197 /* 1198 * If we have less data than requested, block awaiting more 1199 * (subject to any timeout) if: 1200 * 1. the current count is less than the low water mark, 1201 * 2. MSG_WAITALL is set, and it is possible to do the entire 1202 * receive operation at once if we block (resid <= hiwat), or 1203 * 3. MSG_DONTWAIT is not set. 1204 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1205 * we have to do the receive in sections, and thus risk returning 1206 * a short count if a timeout or signal occurs after we start. 1207 */ 1208 if (m == NULL || 1209 ((flags & MSG_DONTWAIT) == 0 && 1210 so->so_rcv.sb_cc < uio->uio_resid && 1211 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1212 ((flags & MSG_WAITALL) && 1213 uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1214 m->m_nextpkt == NULL && !atomic)) { 1215 #ifdef DIAGNOSTIC 1216 if (m == NULL && so->so_rcv.sb_cc) 1217 panic("receive 1"); 1218 #endif 1219 if (so->so_error || so->so_rerror) { 1220 if (m != NULL) 1221 goto dontblock; 1222 if (so->so_error) { 1223 error = so->so_error; 1224 so->so_error = 0; 1225 } else { 1226 error = so->so_rerror; 1227 so->so_rerror = 0; 1228 } 1229 goto release; 1230 } 1231 if (so->so_state & SS_CANTRCVMORE) { 1232 if (m != NULL) 1233 goto dontblock; 1234 else 1235 goto release; 1236 } 1237 for (; m != NULL; m = m->m_next) 1238 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1239 m = so->so_rcv.sb_mb; 1240 goto dontblock; 1241 } 1242 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1243 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1244 error = ENOTCONN; 1245 goto release; 1246 } 1247 if (uio->uio_resid == 0) 1248 goto release; 1249 if ((so->so_state & SS_NBIO) || 1250 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1251 error = EWOULDBLOCK; 1252 goto release; 1253 } 1254 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1255 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1256 sbunlock(&so->so_rcv); 1257 if (wakeup_state & SS_RESTARTSYS) 1258 error = ERESTART; 1259 else 1260 error = sbwait(&so->so_rcv); 1261 if (error != 0) { 1262 sounlock(so); 1263 splx(s); 1264 return error; 1265 } 1266 wakeup_state = so->so_state; 1267 goto restart; 1268 } 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 1294 if (pr->pr_flags & PR_ADDR) { 1295 KASSERT(m->m_type == MT_SONAME); 1296 orig_resid = 0; 1297 if (flags & MSG_PEEK) { 1298 if (paddr) 1299 *paddr = m_copym(m, 0, m->m_len, M_DONTWAIT); 1300 m = m->m_next; 1301 } else { 1302 sbfree(&so->so_rcv, m); 1303 mbuf_removed = 1; 1304 if (paddr != NULL) { 1305 *paddr = m; 1306 so->so_rcv.sb_mb = m->m_next; 1307 m->m_next = NULL; 1308 m = so->so_rcv.sb_mb; 1309 } else { 1310 m = so->so_rcv.sb_mb = m_free(m); 1311 } 1312 sbsync(&so->so_rcv, nextrecord); 1313 } 1314 } 1315 1316 if (pr->pr_flags & PR_ADDR_OPT) { 1317 /* 1318 * For SCTP we may be getting a whole message OR a partial 1319 * delivery. 1320 */ 1321 if (m->m_type == MT_SONAME) { 1322 orig_resid = 0; 1323 if (flags & MSG_PEEK) { 1324 if (paddr) 1325 *paddr = m_copym(m, 0, m->m_len, M_DONTWAIT); 1326 m = m->m_next; 1327 } else { 1328 sbfree(&so->so_rcv, m); 1329 /* XXX XXX XXX: should set mbuf_removed? */ 1330 if (paddr) { 1331 *paddr = m; 1332 so->so_rcv.sb_mb = m->m_next; 1333 m->m_next = 0; 1334 m = so->so_rcv.sb_mb; 1335 } else { 1336 m = so->so_rcv.sb_mb = m_free(m); 1337 } 1338 /* XXX XXX XXX: isn't there an sbsync() 1339 * missing here? */ 1340 } 1341 } 1342 } 1343 1344 /* 1345 * Process one or more MT_CONTROL mbufs present before any data mbufs 1346 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 1347 * just copy the data; if !MSG_PEEK, we call into the protocol to 1348 * perform externalization (or freeing if controlp == NULL). 1349 */ 1350 if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) { 1351 struct mbuf *cm = NULL, *cmn; 1352 struct mbuf **cme = &cm; 1353 1354 do { 1355 if (flags & MSG_PEEK) { 1356 if (controlp != NULL) { 1357 *controlp = m_copym(m, 0, m->m_len, M_DONTWAIT); 1358 controlp = &(*controlp)->m_next; 1359 } 1360 m = m->m_next; 1361 } else { 1362 sbfree(&so->so_rcv, m); 1363 so->so_rcv.sb_mb = m->m_next; 1364 m->m_next = NULL; 1365 *cme = m; 1366 cme = &(*cme)->m_next; 1367 m = so->so_rcv.sb_mb; 1368 } 1369 } while (m != NULL && m->m_type == MT_CONTROL); 1370 if ((flags & MSG_PEEK) == 0) 1371 sbsync(&so->so_rcv, nextrecord); 1372 1373 for (; cm != NULL; cm = cmn) { 1374 cmn = cm->m_next; 1375 cm->m_next = NULL; 1376 type = mtod(cm, struct cmsghdr *)->cmsg_type; 1377 if (controlp != NULL) { 1378 if (dom->dom_externalize != NULL && 1379 type == SCM_RIGHTS) { 1380 sounlock(so); 1381 splx(s); 1382 error = (*dom->dom_externalize)(cm, l, 1383 (flags & MSG_CMSG_CLOEXEC) ? 1384 O_CLOEXEC : 0); 1385 s = splsoftnet(); 1386 solock(so); 1387 } 1388 *controlp = cm; 1389 while (*controlp != NULL) 1390 controlp = &(*controlp)->m_next; 1391 } else { 1392 /* 1393 * Dispose of any SCM_RIGHTS message that went 1394 * through the read path rather than recv. 1395 */ 1396 if (dom->dom_dispose != NULL && 1397 type == SCM_RIGHTS) { 1398 sounlock(so); 1399 (*dom->dom_dispose)(cm); 1400 solock(so); 1401 } 1402 m_freem(cm); 1403 } 1404 } 1405 if (m != NULL) 1406 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 1407 else 1408 nextrecord = so->so_rcv.sb_mb; 1409 orig_resid = 0; 1410 } 1411 1412 /* If m is non-NULL, we have some data to read. */ 1413 if (__predict_true(m != NULL)) { 1414 type = m->m_type; 1415 if (type == MT_OOBDATA) 1416 flags |= MSG_OOB; 1417 } 1418 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1419 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1420 1421 moff = 0; 1422 offset = 0; 1423 while (m != NULL && uio->uio_resid > 0 && error == 0) { 1424 /* 1425 * If the type of mbuf has changed, end the receive 1426 * operation and do a short read. 1427 */ 1428 if (m->m_type == MT_OOBDATA) { 1429 if (type != MT_OOBDATA) 1430 break; 1431 } else if (type == MT_OOBDATA) { 1432 break; 1433 } else if (m->m_type == MT_CONTROL) { 1434 break; 1435 } 1436 #ifdef DIAGNOSTIC 1437 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) { 1438 panic("%s: m_type=%d", __func__, m->m_type); 1439 } 1440 #endif 1441 1442 so->so_state &= ~SS_RCVATMARK; 1443 wakeup_state = 0; 1444 len = uio->uio_resid; 1445 if (so->so_oobmark && len > so->so_oobmark - offset) 1446 len = so->so_oobmark - offset; 1447 if (len > m->m_len - moff) 1448 len = m->m_len - moff; 1449 1450 /* 1451 * If mp is set, just pass back the mbufs. 1452 * Otherwise copy them out via the uio, then free. 1453 * Sockbuf must be consistent here (points to current mbuf, 1454 * it points to next record) when we drop priority; 1455 * we must note any additions to the sockbuf when we 1456 * block interrupts again. 1457 */ 1458 if (mp == NULL) { 1459 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1460 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1461 sounlock(so); 1462 splx(s); 1463 error = uiomove(mtod(m, char *) + moff, len, uio); 1464 s = splsoftnet(); 1465 solock(so); 1466 if (error != 0) { 1467 /* 1468 * If any part of the record has been removed 1469 * (such as the MT_SONAME mbuf, which will 1470 * happen when PR_ADDR, and thus also 1471 * PR_ATOMIC, is set), then drop the entire 1472 * record to maintain the atomicity of the 1473 * receive operation. 1474 * 1475 * This avoids a later panic("receive 1a") 1476 * when compiled with DIAGNOSTIC. 1477 */ 1478 if (m && mbuf_removed && atomic) 1479 (void) sbdroprecord(&so->so_rcv); 1480 1481 goto release; 1482 } 1483 } else { 1484 uio->uio_resid -= len; 1485 } 1486 1487 if (len == m->m_len - moff) { 1488 if (m->m_flags & M_EOR) 1489 flags |= MSG_EOR; 1490 #ifdef SCTP 1491 if (m->m_flags & M_NOTIFICATION) 1492 flags |= MSG_NOTIFICATION; 1493 #endif 1494 if (flags & MSG_PEEK) { 1495 m = m->m_next; 1496 moff = 0; 1497 } else { 1498 nextrecord = m->m_nextpkt; 1499 sbfree(&so->so_rcv, m); 1500 if (mp) { 1501 *mp = m; 1502 mp = &m->m_next; 1503 so->so_rcv.sb_mb = m = m->m_next; 1504 *mp = NULL; 1505 } else { 1506 m = so->so_rcv.sb_mb = m_free(m); 1507 } 1508 /* 1509 * If m != NULL, we also know that 1510 * so->so_rcv.sb_mb != NULL. 1511 */ 1512 KASSERT(so->so_rcv.sb_mb == m); 1513 if (m) { 1514 m->m_nextpkt = nextrecord; 1515 if (nextrecord == NULL) 1516 so->so_rcv.sb_lastrecord = m; 1517 } else { 1518 so->so_rcv.sb_mb = nextrecord; 1519 SB_EMPTY_FIXUP(&so->so_rcv); 1520 } 1521 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1522 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1523 } 1524 } else if (flags & MSG_PEEK) { 1525 moff += len; 1526 } else { 1527 if (mp != NULL) { 1528 mt = m_copym(m, 0, len, M_NOWAIT); 1529 if (__predict_false(mt == NULL)) { 1530 sounlock(so); 1531 mt = m_copym(m, 0, len, M_WAIT); 1532 solock(so); 1533 } 1534 *mp = mt; 1535 } 1536 m->m_data += len; 1537 m->m_len -= len; 1538 so->so_rcv.sb_cc -= len; 1539 } 1540 1541 if (so->so_oobmark) { 1542 if ((flags & MSG_PEEK) == 0) { 1543 so->so_oobmark -= len; 1544 if (so->so_oobmark == 0) { 1545 so->so_state |= SS_RCVATMARK; 1546 break; 1547 } 1548 } else { 1549 offset += len; 1550 if (offset == so->so_oobmark) 1551 break; 1552 } 1553 } 1554 if (flags & MSG_EOR) 1555 break; 1556 1557 /* 1558 * If the MSG_WAITALL flag is set (for non-atomic socket), 1559 * we must not quit until "uio->uio_resid == 0" or an error 1560 * termination. If a signal/timeout occurs, return 1561 * with a short count but without error. 1562 * Keep sockbuf locked against other readers. 1563 */ 1564 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 1565 !sosendallatonce(so) && !nextrecord) { 1566 if (so->so_error || so->so_rerror || 1567 so->so_state & SS_CANTRCVMORE) 1568 break; 1569 /* 1570 * If we are peeking and the socket receive buffer is 1571 * full, stop since we can't get more data to peek at. 1572 */ 1573 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1574 break; 1575 /* 1576 * If we've drained the socket buffer, tell the 1577 * protocol in case it needs to do something to 1578 * get it filled again. 1579 */ 1580 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1581 (*pr->pr_usrreqs->pr_rcvd)(so, flags, l); 1582 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1583 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1584 if (wakeup_state & SS_RESTARTSYS) 1585 error = ERESTART; 1586 else 1587 error = sbwait(&so->so_rcv); 1588 if (error != 0) { 1589 sbunlock(&so->so_rcv); 1590 sounlock(so); 1591 splx(s); 1592 return 0; 1593 } 1594 if ((m = so->so_rcv.sb_mb) != NULL) 1595 nextrecord = m->m_nextpkt; 1596 wakeup_state = so->so_state; 1597 } 1598 } 1599 1600 if (m && atomic) { 1601 flags |= MSG_TRUNC; 1602 if ((flags & MSG_PEEK) == 0) 1603 (void) sbdroprecord(&so->so_rcv); 1604 } 1605 if ((flags & MSG_PEEK) == 0) { 1606 if (m == NULL) { 1607 /* 1608 * First part is an inline SB_EMPTY_FIXUP(). Second 1609 * part makes sure sb_lastrecord is up-to-date if 1610 * there is still data in the socket buffer. 1611 */ 1612 so->so_rcv.sb_mb = nextrecord; 1613 if (so->so_rcv.sb_mb == NULL) { 1614 so->so_rcv.sb_mbtail = NULL; 1615 so->so_rcv.sb_lastrecord = NULL; 1616 } else if (nextrecord->m_nextpkt == NULL) 1617 so->so_rcv.sb_lastrecord = nextrecord; 1618 } 1619 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1620 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1621 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1622 (*pr->pr_usrreqs->pr_rcvd)(so, flags, l); 1623 } 1624 if (orig_resid == uio->uio_resid && orig_resid && 1625 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1626 sbunlock(&so->so_rcv); 1627 goto restart; 1628 } 1629 1630 if (flagsp != NULL) 1631 *flagsp |= flags; 1632 release: 1633 sbunlock(&so->so_rcv); 1634 sounlock(so); 1635 splx(s); 1636 return error; 1637 } 1638 1639 int 1640 soshutdown(struct socket *so, int how) 1641 { 1642 const struct protosw *pr; 1643 int error; 1644 1645 KASSERT(solocked(so)); 1646 1647 pr = so->so_proto; 1648 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1649 return EINVAL; 1650 1651 if (how == SHUT_RD || how == SHUT_RDWR) { 1652 sorflush(so); 1653 error = 0; 1654 } 1655 if (how == SHUT_WR || how == SHUT_RDWR) 1656 error = (*pr->pr_usrreqs->pr_shutdown)(so); 1657 1658 return error; 1659 } 1660 1661 void 1662 sorestart(struct socket *so) 1663 { 1664 /* 1665 * An application has called close() on an fd on which another 1666 * of its threads has called a socket system call. 1667 * Mark this and wake everyone up, and code that would block again 1668 * instead returns ERESTART. 1669 * On system call re-entry the fd is validated and EBADF returned. 1670 * Any other fd will block again on the 2nd syscall. 1671 */ 1672 solock(so); 1673 so->so_state |= SS_RESTARTSYS; 1674 cv_broadcast(&so->so_cv); 1675 cv_broadcast(&so->so_snd.sb_cv); 1676 cv_broadcast(&so->so_rcv.sb_cv); 1677 sounlock(so); 1678 } 1679 1680 void 1681 sorflush(struct socket *so) 1682 { 1683 struct sockbuf *sb, asb; 1684 const struct protosw *pr; 1685 1686 KASSERT(solocked(so)); 1687 1688 sb = &so->so_rcv; 1689 pr = so->so_proto; 1690 socantrcvmore(so); 1691 sb->sb_flags |= SB_NOINTR; 1692 (void )sblock(sb, M_WAITOK); 1693 sbunlock(sb); 1694 asb = *sb; 1695 /* 1696 * Clear most of the sockbuf structure, but leave some of the 1697 * fields valid. 1698 */ 1699 memset(&sb->sb_startzero, 0, 1700 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1701 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) { 1702 sounlock(so); 1703 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1704 solock(so); 1705 } 1706 sbrelease(&asb, so); 1707 } 1708 1709 /* 1710 * internal set SOL_SOCKET options 1711 */ 1712 static int 1713 sosetopt1(struct socket *so, const struct sockopt *sopt) 1714 { 1715 int error, opt; 1716 int optval = 0; /* XXX: gcc */ 1717 struct linger l; 1718 struct timeval tv; 1719 1720 opt = sopt->sopt_name; 1721 1722 switch (opt) { 1723 1724 case SO_ACCEPTFILTER: 1725 error = accept_filt_setopt(so, sopt); 1726 KASSERT(solocked(so)); 1727 break; 1728 1729 case SO_LINGER: 1730 error = sockopt_get(sopt, &l, sizeof(l)); 1731 solock(so); 1732 if (error) 1733 break; 1734 if (l.l_linger < 0 || l.l_linger > USHRT_MAX || 1735 l.l_linger > (INT_MAX / hz)) { 1736 error = EDOM; 1737 break; 1738 } 1739 so->so_linger = l.l_linger; 1740 if (l.l_onoff) 1741 so->so_options |= SO_LINGER; 1742 else 1743 so->so_options &= ~SO_LINGER; 1744 break; 1745 1746 case SO_DEBUG: 1747 case SO_KEEPALIVE: 1748 case SO_DONTROUTE: 1749 case SO_USELOOPBACK: 1750 case SO_BROADCAST: 1751 case SO_REUSEADDR: 1752 case SO_REUSEPORT: 1753 case SO_OOBINLINE: 1754 case SO_TIMESTAMP: 1755 case SO_NOSIGPIPE: 1756 case SO_RERROR: 1757 error = sockopt_getint(sopt, &optval); 1758 solock(so); 1759 if (error) 1760 break; 1761 if (optval) 1762 so->so_options |= opt; 1763 else 1764 so->so_options &= ~opt; 1765 break; 1766 1767 case SO_SNDBUF: 1768 case SO_RCVBUF: 1769 case SO_SNDLOWAT: 1770 case SO_RCVLOWAT: 1771 error = sockopt_getint(sopt, &optval); 1772 solock(so); 1773 if (error) 1774 break; 1775 1776 /* 1777 * Values < 1 make no sense for any of these 1778 * options, so disallow them. 1779 */ 1780 if (optval < 1) { 1781 error = EINVAL; 1782 break; 1783 } 1784 1785 switch (opt) { 1786 case SO_SNDBUF: 1787 if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) { 1788 error = ENOBUFS; 1789 break; 1790 } 1791 so->so_snd.sb_flags &= ~SB_AUTOSIZE; 1792 break; 1793 1794 case SO_RCVBUF: 1795 if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) { 1796 error = ENOBUFS; 1797 break; 1798 } 1799 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1800 break; 1801 1802 /* 1803 * Make sure the low-water is never greater than 1804 * the high-water. 1805 */ 1806 case SO_SNDLOWAT: 1807 if (optval > so->so_snd.sb_hiwat) 1808 optval = so->so_snd.sb_hiwat; 1809 1810 so->so_snd.sb_lowat = optval; 1811 break; 1812 1813 case SO_RCVLOWAT: 1814 if (optval > so->so_rcv.sb_hiwat) 1815 optval = so->so_rcv.sb_hiwat; 1816 1817 so->so_rcv.sb_lowat = optval; 1818 break; 1819 } 1820 break; 1821 1822 case SO_SNDTIMEO: 1823 case SO_RCVTIMEO: 1824 solock(so); 1825 error = sockopt_get(sopt, &tv, sizeof(tv)); 1826 if (error) 1827 break; 1828 1829 if (tv.tv_sec < 0 || tv.tv_usec < 0 || tv.tv_usec >= 1000000) { 1830 error = EDOM; 1831 break; 1832 } 1833 if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) { 1834 error = EDOM; 1835 break; 1836 } 1837 1838 optval = tv.tv_sec * hz + tv.tv_usec / tick; 1839 if (optval == 0 && tv.tv_usec != 0) 1840 optval = 1; 1841 1842 switch (opt) { 1843 case SO_SNDTIMEO: 1844 so->so_snd.sb_timeo = optval; 1845 break; 1846 case SO_RCVTIMEO: 1847 so->so_rcv.sb_timeo = optval; 1848 break; 1849 } 1850 break; 1851 1852 default: 1853 MODULE_HOOK_CALL(uipc_socket_50_setopt1_hook, 1854 (opt, so, sopt), enosys(), error); 1855 if (error == ENOSYS || error == EPASSTHROUGH) { 1856 solock(so); 1857 error = ENOPROTOOPT; 1858 } 1859 break; 1860 } 1861 KASSERT(solocked(so)); 1862 return error; 1863 } 1864 1865 int 1866 sosetopt(struct socket *so, struct sockopt *sopt) 1867 { 1868 int error, prerr; 1869 1870 if (sopt->sopt_level == SOL_SOCKET) { 1871 error = sosetopt1(so, sopt); 1872 KASSERT(solocked(so)); 1873 } else { 1874 error = ENOPROTOOPT; 1875 solock(so); 1876 } 1877 1878 if ((error == 0 || error == ENOPROTOOPT) && 1879 so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) { 1880 /* give the protocol stack a shot */ 1881 prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt); 1882 if (prerr == 0) 1883 error = 0; 1884 else if (prerr != ENOPROTOOPT) 1885 error = prerr; 1886 } 1887 sounlock(so); 1888 return error; 1889 } 1890 1891 /* 1892 * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt() 1893 */ 1894 int 1895 so_setsockopt(struct lwp *l, struct socket *so, int level, int name, 1896 const void *val, size_t valsize) 1897 { 1898 struct sockopt sopt; 1899 int error; 1900 1901 KASSERT(valsize == 0 || val != NULL); 1902 1903 sockopt_init(&sopt, level, name, valsize); 1904 sockopt_set(&sopt, val, valsize); 1905 1906 error = sosetopt(so, &sopt); 1907 1908 sockopt_destroy(&sopt); 1909 1910 return error; 1911 } 1912 1913 /* 1914 * internal get SOL_SOCKET options 1915 */ 1916 static int 1917 sogetopt1(struct socket *so, struct sockopt *sopt) 1918 { 1919 int error, optval, opt; 1920 struct linger l; 1921 struct timeval tv; 1922 1923 switch ((opt = sopt->sopt_name)) { 1924 1925 case SO_ACCEPTFILTER: 1926 error = accept_filt_getopt(so, sopt); 1927 break; 1928 1929 case SO_LINGER: 1930 l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0; 1931 l.l_linger = so->so_linger; 1932 1933 error = sockopt_set(sopt, &l, sizeof(l)); 1934 break; 1935 1936 case SO_USELOOPBACK: 1937 case SO_DONTROUTE: 1938 case SO_DEBUG: 1939 case SO_KEEPALIVE: 1940 case SO_REUSEADDR: 1941 case SO_REUSEPORT: 1942 case SO_BROADCAST: 1943 case SO_OOBINLINE: 1944 case SO_TIMESTAMP: 1945 case SO_NOSIGPIPE: 1946 case SO_RERROR: 1947 case SO_ACCEPTCONN: 1948 error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0); 1949 break; 1950 1951 case SO_TYPE: 1952 error = sockopt_setint(sopt, so->so_type); 1953 break; 1954 1955 case SO_ERROR: 1956 if (so->so_error == 0) { 1957 so->so_error = so->so_rerror; 1958 so->so_rerror = 0; 1959 } 1960 error = sockopt_setint(sopt, so->so_error); 1961 so->so_error = 0; 1962 break; 1963 1964 case SO_SNDBUF: 1965 error = sockopt_setint(sopt, so->so_snd.sb_hiwat); 1966 break; 1967 1968 case SO_RCVBUF: 1969 error = sockopt_setint(sopt, so->so_rcv.sb_hiwat); 1970 break; 1971 1972 case SO_SNDLOWAT: 1973 error = sockopt_setint(sopt, so->so_snd.sb_lowat); 1974 break; 1975 1976 case SO_RCVLOWAT: 1977 error = sockopt_setint(sopt, so->so_rcv.sb_lowat); 1978 break; 1979 1980 case SO_SNDTIMEO: 1981 case SO_RCVTIMEO: 1982 optval = (opt == SO_SNDTIMEO ? 1983 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1984 1985 memset(&tv, 0, sizeof(tv)); 1986 tv.tv_sec = optval / hz; 1987 tv.tv_usec = (optval % hz) * tick; 1988 1989 error = sockopt_set(sopt, &tv, sizeof(tv)); 1990 break; 1991 1992 case SO_OVERFLOWED: 1993 error = sockopt_setint(sopt, so->so_rcv.sb_overflowed); 1994 break; 1995 1996 default: 1997 MODULE_HOOK_CALL(uipc_socket_50_getopt1_hook, 1998 (opt, so, sopt), enosys(), error); 1999 if (error) 2000 error = ENOPROTOOPT; 2001 break; 2002 } 2003 2004 return error; 2005 } 2006 2007 int 2008 sogetopt(struct socket *so, struct sockopt *sopt) 2009 { 2010 int error; 2011 2012 solock(so); 2013 if (sopt->sopt_level != SOL_SOCKET) { 2014 if (so->so_proto && so->so_proto->pr_ctloutput) { 2015 error = ((*so->so_proto->pr_ctloutput) 2016 (PRCO_GETOPT, so, sopt)); 2017 } else 2018 error = (ENOPROTOOPT); 2019 } else { 2020 error = sogetopt1(so, sopt); 2021 } 2022 sounlock(so); 2023 return error; 2024 } 2025 2026 /* 2027 * alloc sockopt data buffer buffer 2028 * - will be released at destroy 2029 */ 2030 static int 2031 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag) 2032 { 2033 2034 KASSERT(sopt->sopt_size == 0); 2035 2036 if (len > sizeof(sopt->sopt_buf)) { 2037 sopt->sopt_data = kmem_zalloc(len, kmflag); 2038 if (sopt->sopt_data == NULL) 2039 return ENOMEM; 2040 } else 2041 sopt->sopt_data = sopt->sopt_buf; 2042 2043 sopt->sopt_size = len; 2044 return 0; 2045 } 2046 2047 /* 2048 * initialise sockopt storage 2049 * - MAY sleep during allocation 2050 */ 2051 void 2052 sockopt_init(struct sockopt *sopt, int level, int name, size_t size) 2053 { 2054 2055 memset(sopt, 0, sizeof(*sopt)); 2056 2057 sopt->sopt_level = level; 2058 sopt->sopt_name = name; 2059 (void)sockopt_alloc(sopt, size, KM_SLEEP); 2060 } 2061 2062 /* 2063 * destroy sockopt storage 2064 * - will release any held memory references 2065 */ 2066 void 2067 sockopt_destroy(struct sockopt *sopt) 2068 { 2069 2070 if (sopt->sopt_data != sopt->sopt_buf) 2071 kmem_free(sopt->sopt_data, sopt->sopt_size); 2072 2073 memset(sopt, 0, sizeof(*sopt)); 2074 } 2075 2076 /* 2077 * set sockopt value 2078 * - value is copied into sockopt 2079 * - memory is allocated when necessary, will not sleep 2080 */ 2081 int 2082 sockopt_set(struct sockopt *sopt, const void *buf, size_t len) 2083 { 2084 int error; 2085 2086 if (sopt->sopt_size == 0) { 2087 error = sockopt_alloc(sopt, len, KM_NOSLEEP); 2088 if (error) 2089 return error; 2090 } 2091 2092 sopt->sopt_retsize = MIN(sopt->sopt_size, len); 2093 memcpy(sopt->sopt_data, buf, sopt->sopt_retsize); 2094 2095 return 0; 2096 } 2097 2098 /* 2099 * common case of set sockopt integer value 2100 */ 2101 int 2102 sockopt_setint(struct sockopt *sopt, int val) 2103 { 2104 2105 return sockopt_set(sopt, &val, sizeof(int)); 2106 } 2107 2108 /* 2109 * get sockopt value 2110 * - correct size must be given 2111 */ 2112 int 2113 sockopt_get(const struct sockopt *sopt, void *buf, size_t len) 2114 { 2115 2116 if (sopt->sopt_size != len) 2117 return EINVAL; 2118 2119 memcpy(buf, sopt->sopt_data, len); 2120 return 0; 2121 } 2122 2123 /* 2124 * common case of get sockopt integer value 2125 */ 2126 int 2127 sockopt_getint(const struct sockopt *sopt, int *valp) 2128 { 2129 2130 return sockopt_get(sopt, valp, sizeof(int)); 2131 } 2132 2133 /* 2134 * set sockopt value from mbuf 2135 * - ONLY for legacy code 2136 * - mbuf is released by sockopt 2137 * - will not sleep 2138 */ 2139 int 2140 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m) 2141 { 2142 size_t len; 2143 int error; 2144 2145 len = m_length(m); 2146 2147 if (sopt->sopt_size == 0) { 2148 error = sockopt_alloc(sopt, len, KM_NOSLEEP); 2149 if (error) 2150 return error; 2151 } 2152 2153 sopt->sopt_retsize = MIN(sopt->sopt_size, len); 2154 m_copydata(m, 0, sopt->sopt_retsize, sopt->sopt_data); 2155 m_freem(m); 2156 2157 return 0; 2158 } 2159 2160 /* 2161 * get sockopt value into mbuf 2162 * - ONLY for legacy code 2163 * - mbuf to be released by the caller 2164 * - will not sleep 2165 */ 2166 struct mbuf * 2167 sockopt_getmbuf(const struct sockopt *sopt) 2168 { 2169 struct mbuf *m; 2170 2171 if (sopt->sopt_size > MCLBYTES) 2172 return NULL; 2173 2174 m = m_get(M_DONTWAIT, MT_SOOPTS); 2175 if (m == NULL) 2176 return NULL; 2177 2178 if (sopt->sopt_size > MLEN) { 2179 MCLGET(m, M_DONTWAIT); 2180 if ((m->m_flags & M_EXT) == 0) { 2181 m_free(m); 2182 return NULL; 2183 } 2184 } 2185 2186 memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size); 2187 m->m_len = sopt->sopt_size; 2188 2189 return m; 2190 } 2191 2192 void 2193 sohasoutofband(struct socket *so) 2194 { 2195 2196 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 2197 selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT); 2198 } 2199 2200 static void 2201 filt_sordetach(struct knote *kn) 2202 { 2203 struct socket *so; 2204 2205 so = ((file_t *)kn->kn_obj)->f_socket; 2206 solock(so); 2207 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 2208 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 2209 so->so_rcv.sb_flags &= ~SB_KNOTE; 2210 sounlock(so); 2211 } 2212 2213 /*ARGSUSED*/ 2214 static int 2215 filt_soread(struct knote *kn, long hint) 2216 { 2217 struct socket *so; 2218 int rv; 2219 2220 so = ((file_t *)kn->kn_obj)->f_socket; 2221 if (hint != NOTE_SUBMIT) 2222 solock(so); 2223 kn->kn_data = so->so_rcv.sb_cc; 2224 if (so->so_state & SS_CANTRCVMORE) { 2225 kn->kn_flags |= EV_EOF; 2226 kn->kn_fflags = so->so_error; 2227 rv = 1; 2228 } else if (so->so_error || so->so_rerror) 2229 rv = 1; 2230 else if (kn->kn_sfflags & NOTE_LOWAT) 2231 rv = (kn->kn_data >= kn->kn_sdata); 2232 else 2233 rv = (kn->kn_data >= so->so_rcv.sb_lowat); 2234 if (hint != NOTE_SUBMIT) 2235 sounlock(so); 2236 return rv; 2237 } 2238 2239 static void 2240 filt_sowdetach(struct knote *kn) 2241 { 2242 struct socket *so; 2243 2244 so = ((file_t *)kn->kn_obj)->f_socket; 2245 solock(so); 2246 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 2247 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 2248 so->so_snd.sb_flags &= ~SB_KNOTE; 2249 sounlock(so); 2250 } 2251 2252 /*ARGSUSED*/ 2253 static int 2254 filt_sowrite(struct knote *kn, long hint) 2255 { 2256 struct socket *so; 2257 int rv; 2258 2259 so = ((file_t *)kn->kn_obj)->f_socket; 2260 if (hint != NOTE_SUBMIT) 2261 solock(so); 2262 kn->kn_data = sbspace(&so->so_snd); 2263 if (so->so_state & SS_CANTSENDMORE) { 2264 kn->kn_flags |= EV_EOF; 2265 kn->kn_fflags = so->so_error; 2266 rv = 1; 2267 } else if (so->so_error) 2268 rv = 1; 2269 else if (((so->so_state & SS_ISCONNECTED) == 0) && 2270 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 2271 rv = 0; 2272 else if (kn->kn_sfflags & NOTE_LOWAT) 2273 rv = (kn->kn_data >= kn->kn_sdata); 2274 else 2275 rv = (kn->kn_data >= so->so_snd.sb_lowat); 2276 if (hint != NOTE_SUBMIT) 2277 sounlock(so); 2278 return rv; 2279 } 2280 2281 /*ARGSUSED*/ 2282 static int 2283 filt_solisten(struct knote *kn, long hint) 2284 { 2285 struct socket *so; 2286 int rv; 2287 2288 so = ((file_t *)kn->kn_obj)->f_socket; 2289 2290 /* 2291 * Set kn_data to number of incoming connections, not 2292 * counting partial (incomplete) connections. 2293 */ 2294 if (hint != NOTE_SUBMIT) 2295 solock(so); 2296 kn->kn_data = so->so_qlen; 2297 rv = (kn->kn_data > 0); 2298 if (hint != NOTE_SUBMIT) 2299 sounlock(so); 2300 return rv; 2301 } 2302 2303 static const struct filterops solisten_filtops = { 2304 .f_isfd = 1, 2305 .f_attach = NULL, 2306 .f_detach = filt_sordetach, 2307 .f_event = filt_solisten, 2308 }; 2309 2310 static const struct filterops soread_filtops = { 2311 .f_isfd = 1, 2312 .f_attach = NULL, 2313 .f_detach = filt_sordetach, 2314 .f_event = filt_soread, 2315 }; 2316 2317 static const struct filterops sowrite_filtops = { 2318 .f_isfd = 1, 2319 .f_attach = NULL, 2320 .f_detach = filt_sowdetach, 2321 .f_event = filt_sowrite, 2322 }; 2323 2324 int 2325 soo_kqfilter(struct file *fp, struct knote *kn) 2326 { 2327 struct socket *so; 2328 struct sockbuf *sb; 2329 2330 so = ((file_t *)kn->kn_obj)->f_socket; 2331 solock(so); 2332 switch (kn->kn_filter) { 2333 case EVFILT_READ: 2334 if (so->so_options & SO_ACCEPTCONN) 2335 kn->kn_fop = &solisten_filtops; 2336 else 2337 kn->kn_fop = &soread_filtops; 2338 sb = &so->so_rcv; 2339 break; 2340 case EVFILT_WRITE: 2341 kn->kn_fop = &sowrite_filtops; 2342 sb = &so->so_snd; 2343 break; 2344 default: 2345 sounlock(so); 2346 return EINVAL; 2347 } 2348 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 2349 sb->sb_flags |= SB_KNOTE; 2350 sounlock(so); 2351 return 0; 2352 } 2353 2354 static int 2355 sodopoll(struct socket *so, int events) 2356 { 2357 int revents; 2358 2359 revents = 0; 2360 2361 if (events & (POLLIN | POLLRDNORM)) 2362 if (soreadable(so)) 2363 revents |= events & (POLLIN | POLLRDNORM); 2364 2365 if (events & (POLLOUT | POLLWRNORM)) 2366 if (sowritable(so)) 2367 revents |= events & (POLLOUT | POLLWRNORM); 2368 2369 if (events & (POLLPRI | POLLRDBAND)) 2370 if (so->so_oobmark || (so->so_state & SS_RCVATMARK)) 2371 revents |= events & (POLLPRI | POLLRDBAND); 2372 2373 return revents; 2374 } 2375 2376 int 2377 sopoll(struct socket *so, int events) 2378 { 2379 int revents = 0; 2380 2381 #ifndef DIAGNOSTIC 2382 /* 2383 * Do a quick, unlocked check in expectation that the socket 2384 * will be ready for I/O. Don't do this check if DIAGNOSTIC, 2385 * as the solocked() assertions will fail. 2386 */ 2387 if ((revents = sodopoll(so, events)) != 0) 2388 return revents; 2389 #endif 2390 2391 solock(so); 2392 if ((revents = sodopoll(so, events)) == 0) { 2393 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 2394 selrecord(curlwp, &so->so_rcv.sb_sel); 2395 so->so_rcv.sb_flags |= SB_NOTIFY; 2396 } 2397 2398 if (events & (POLLOUT | POLLWRNORM)) { 2399 selrecord(curlwp, &so->so_snd.sb_sel); 2400 so->so_snd.sb_flags |= SB_NOTIFY; 2401 } 2402 } 2403 sounlock(so); 2404 2405 return revents; 2406 } 2407 2408 struct mbuf ** 2409 sbsavetimestamp(int opt, struct mbuf **mp) 2410 { 2411 struct timeval tv; 2412 int error; 2413 2414 microtime(&tv); 2415 2416 MODULE_HOOK_CALL(uipc_socket_50_sbts_hook, (opt, mp), enosys(), error); 2417 if (error == 0) 2418 return mp; 2419 2420 if (opt & SO_TIMESTAMP) { 2421 *mp = sbcreatecontrol(&tv, sizeof(tv), 2422 SCM_TIMESTAMP, SOL_SOCKET); 2423 if (*mp) 2424 mp = &(*mp)->m_next; 2425 } 2426 return mp; 2427 } 2428 2429 2430 #include <sys/sysctl.h> 2431 2432 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 2433 static int sysctl_kern_sbmax(SYSCTLFN_PROTO); 2434 2435 /* 2436 * sysctl helper routine for kern.somaxkva. ensures that the given 2437 * value is not too small. 2438 * (XXX should we maybe make sure it's not too large as well?) 2439 */ 2440 static int 2441 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 2442 { 2443 int error, new_somaxkva; 2444 struct sysctlnode node; 2445 2446 new_somaxkva = somaxkva; 2447 node = *rnode; 2448 node.sysctl_data = &new_somaxkva; 2449 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2450 if (error || newp == NULL) 2451 return error; 2452 2453 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 2454 return EINVAL; 2455 2456 mutex_enter(&so_pendfree_lock); 2457 somaxkva = new_somaxkva; 2458 cv_broadcast(&socurkva_cv); 2459 mutex_exit(&so_pendfree_lock); 2460 2461 return error; 2462 } 2463 2464 /* 2465 * sysctl helper routine for kern.sbmax. Basically just ensures that 2466 * any new value is not too small. 2467 */ 2468 static int 2469 sysctl_kern_sbmax(SYSCTLFN_ARGS) 2470 { 2471 int error, new_sbmax; 2472 struct sysctlnode node; 2473 2474 new_sbmax = sb_max; 2475 node = *rnode; 2476 node.sysctl_data = &new_sbmax; 2477 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2478 if (error || newp == NULL) 2479 return error; 2480 2481 KERNEL_LOCK(1, NULL); 2482 error = sb_max_set(new_sbmax); 2483 KERNEL_UNLOCK_ONE(NULL); 2484 2485 return error; 2486 } 2487 2488 /* 2489 * sysctl helper routine for kern.sooptions. Ensures that only allowed 2490 * options can be set. 2491 */ 2492 static int 2493 sysctl_kern_sooptions(SYSCTLFN_ARGS) 2494 { 2495 int error, new_options; 2496 struct sysctlnode node; 2497 2498 new_options = sooptions; 2499 node = *rnode; 2500 node.sysctl_data = &new_options; 2501 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2502 if (error || newp == NULL) 2503 return error; 2504 2505 if (new_options & ~SO_DEFOPTS) 2506 return EINVAL; 2507 2508 sooptions = new_options; 2509 2510 return 0; 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 sysctl_createv(&socket_sysctllog, 0, NULL, NULL, 2535 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2536 CTLTYPE_INT, "sooptions", 2537 SYSCTL_DESCR("Default socket options"), 2538 sysctl_kern_sooptions, 0, NULL, 0, 2539 CTL_KERN, CTL_CREATE, CTL_EOL); 2540 } 2541
Indexes created Sat Sep 20 22:09:52 GMT 2025