1 /* $NetBSD: uipc_socket.c,v 1.310 2024/07/05 04:31:53 rin Exp $ */ 2 3 /* 4 * Copyright (c) 2002, 2007, 2008, 2009, 2023 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 1.310 2024/07/05 04:31:53 rin 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 #include "opt_pipe.h" 85 #endif 86 87 #include <sys/param.h> 88 #include <sys/systm.h> 89 #include <sys/proc.h> 90 #include <sys/file.h> 91 #include <sys/filedesc.h> 92 #include <sys/kmem.h> 93 #include <sys/mbuf.h> 94 #include <sys/domain.h> 95 #include <sys/kernel.h> 96 #include <sys/protosw.h> 97 #include <sys/socket.h> 98 #include <sys/socketvar.h> 99 #include <sys/signalvar.h> 100 #include <sys/resourcevar.h> 101 #include <sys/uidinfo.h> 102 #include <sys/event.h> 103 #include <sys/poll.h> 104 #include <sys/kauth.h> 105 #include <sys/mutex.h> 106 #include <sys/condvar.h> 107 #include <sys/kthread.h> 108 #include <sys/compat_stub.h> 109 110 #include <compat/sys/time.h> 111 #include <compat/sys/socket.h> 112 113 #include <uvm/uvm_extern.h> 114 #include <uvm/uvm_loan.h> 115 #include <uvm/uvm_page.h> 116 117 #ifdef SCTP 118 #include <netinet/sctp_route.h> 119 #endif 120 121 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 122 123 extern const struct fileops socketops; 124 125 static int sooptions; 126 extern int somaxconn; /* patchable (XXX sysctl) */ 127 int somaxconn = SOMAXCONN; 128 kmutex_t *softnet_lock; 129 130 #ifdef SOSEND_COUNTERS 131 #include <sys/device.h> 132 133 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 134 NULL, "sosend", "loan big"); 135 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 136 NULL, "sosend", "copy big"); 137 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 138 NULL, "sosend", "copy small"); 139 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 140 NULL, "sosend", "kva limit"); 141 142 #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++ 143 144 EVCNT_ATTACH_STATIC(sosend_loan_big); 145 EVCNT_ATTACH_STATIC(sosend_copy_big); 146 EVCNT_ATTACH_STATIC(sosend_copy_small); 147 EVCNT_ATTACH_STATIC(sosend_kvalimit); 148 #else 149 150 #define SOSEND_COUNTER_INCR(ev) /* nothing */ 151 152 #endif /* SOSEND_COUNTERS */ 153 154 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR) 155 int sock_loan_thresh = -1; 156 #else 157 int sock_loan_thresh = 4096; 158 #endif 159 160 static kmutex_t so_pendfree_lock; 161 static struct mbuf *so_pendfree = NULL; 162 163 #ifndef SOMAXKVA 164 #define SOMAXKVA (16 * 1024 * 1024) 165 #endif 166 int somaxkva = SOMAXKVA; 167 static int socurkva; 168 static kcondvar_t socurkva_cv; 169 170 #ifndef SOFIXEDBUF 171 #define SOFIXEDBUF true 172 #endif 173 bool sofixedbuf = SOFIXEDBUF; 174 175 static kauth_listener_t socket_listener; 176 177 #define SOCK_LOAN_CHUNK 65536 178 179 static void sopendfree_thread(void *); 180 static kcondvar_t pendfree_thread_cv; 181 static lwp_t *sopendfree_lwp; 182 183 static void sysctl_kern_socket_setup(void); 184 static struct sysctllog *socket_sysctllog; 185 186 static vsize_t 187 sokvareserve(struct socket *so, vsize_t len) 188 { 189 int error; 190 191 mutex_enter(&so_pendfree_lock); 192 while (socurkva + len > somaxkva) { 193 SOSEND_COUNTER_INCR(&sosend_kvalimit); 194 error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock); 195 if (error) { 196 len = 0; 197 break; 198 } 199 } 200 socurkva += len; 201 mutex_exit(&so_pendfree_lock); 202 return len; 203 } 204 205 static void 206 sokvaunreserve(vsize_t len) 207 { 208 209 mutex_enter(&so_pendfree_lock); 210 socurkva -= len; 211 cv_broadcast(&socurkva_cv); 212 mutex_exit(&so_pendfree_lock); 213 } 214 215 /* 216 * sokvaalloc: allocate kva for loan. 217 */ 218 vaddr_t 219 sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so) 220 { 221 vaddr_t lva; 222 223 if (sokvareserve(so, len) == 0) 224 return 0; 225 226 lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask, 227 UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA); 228 if (lva == 0) { 229 sokvaunreserve(len); 230 return 0; 231 } 232 233 return lva; 234 } 235 236 /* 237 * sokvafree: free kva for loan. 238 */ 239 void 240 sokvafree(vaddr_t sva, vsize_t len) 241 { 242 243 uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY); 244 sokvaunreserve(len); 245 } 246 247 static void 248 sodoloanfree(struct vm_page **pgs, void *buf, size_t size) 249 { 250 vaddr_t sva, eva; 251 vsize_t len; 252 int npgs; 253 254 KASSERT(pgs != NULL); 255 256 eva = round_page((vaddr_t) buf + size); 257 sva = trunc_page((vaddr_t) buf); 258 len = eva - sva; 259 npgs = len >> PAGE_SHIFT; 260 261 pmap_kremove(sva, len); 262 pmap_update(pmap_kernel()); 263 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 264 sokvafree(sva, len); 265 } 266 267 /* 268 * sopendfree_thread: free mbufs on "pendfree" list. Unlock and relock 269 * so_pendfree_lock when freeing mbufs. 270 */ 271 static void 272 sopendfree_thread(void *v) 273 { 274 struct mbuf *m, *next; 275 size_t rv; 276 277 mutex_enter(&so_pendfree_lock); 278 279 for (;;) { 280 rv = 0; 281 while (so_pendfree != NULL) { 282 m = so_pendfree; 283 so_pendfree = NULL; 284 mutex_exit(&so_pendfree_lock); 285 286 for (; m != NULL; m = next) { 287 next = m->m_next; 288 KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 289 0); 290 KASSERT(m->m_ext.ext_refcnt == 0); 291 292 rv += m->m_ext.ext_size; 293 sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf, 294 m->m_ext.ext_size); 295 pool_cache_put(mb_cache, m); 296 } 297 298 mutex_enter(&so_pendfree_lock); 299 } 300 if (rv) 301 cv_broadcast(&socurkva_cv); 302 cv_wait(&pendfree_thread_cv, &so_pendfree_lock); 303 } 304 panic("sopendfree_thread"); 305 /* NOTREACHED */ 306 } 307 308 void 309 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg) 310 { 311 312 KASSERT(m != NULL); 313 314 /* 315 * postpone freeing mbuf. 316 * 317 * we can't do it in interrupt context 318 * because we need to put kva back to kernel_map. 319 */ 320 321 mutex_enter(&so_pendfree_lock); 322 m->m_next = so_pendfree; 323 so_pendfree = m; 324 cv_signal(&pendfree_thread_cv); 325 mutex_exit(&so_pendfree_lock); 326 } 327 328 static long 329 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 330 { 331 struct iovec *iov = uio->uio_iov; 332 vaddr_t sva, eva; 333 vsize_t len; 334 vaddr_t lva; 335 int npgs, error; 336 vaddr_t va; 337 int i; 338 339 if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace)) 340 return 0; 341 342 if (iov->iov_len < (size_t) space) 343 space = iov->iov_len; 344 if (space > SOCK_LOAN_CHUNK) 345 space = SOCK_LOAN_CHUNK; 346 347 eva = round_page((vaddr_t) iov->iov_base + space); 348 sva = trunc_page((vaddr_t) iov->iov_base); 349 len = eva - sva; 350 npgs = len >> PAGE_SHIFT; 351 352 KASSERT(npgs <= M_EXT_MAXPAGES); 353 354 lva = sokvaalloc(sva, len, so); 355 if (lva == 0) 356 return 0; 357 358 error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len, 359 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 360 if (error) { 361 sokvafree(lva, len); 362 return 0; 363 } 364 365 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 366 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 367 VM_PROT_READ, 0); 368 pmap_update(pmap_kernel()); 369 370 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 371 372 MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so); 373 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 374 375 uio->uio_resid -= space; 376 /* uio_offset not updated, not set/used for write(2) */ 377 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space; 378 uio->uio_iov->iov_len -= space; 379 if (uio->uio_iov->iov_len == 0) { 380 uio->uio_iov++; 381 uio->uio_iovcnt--; 382 } 383 384 return space; 385 } 386 387 static int 388 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 389 void *arg0, void *arg1, void *arg2, void *arg3) 390 { 391 int result; 392 enum kauth_network_req req; 393 394 result = KAUTH_RESULT_DEFER; 395 req = (enum kauth_network_req)(uintptr_t)arg0; 396 397 if ((action != KAUTH_NETWORK_SOCKET) && 398 (action != KAUTH_NETWORK_BIND)) 399 return result; 400 401 switch (req) { 402 case KAUTH_REQ_NETWORK_BIND_PORT: 403 result = KAUTH_RESULT_ALLOW; 404 break; 405 406 case KAUTH_REQ_NETWORK_SOCKET_DROP: { 407 /* Normal users can only drop their own connections. */ 408 struct socket *so = (struct socket *)arg1; 409 410 if (so->so_cred && proc_uidmatch(cred, so->so_cred) == 0) 411 result = KAUTH_RESULT_ALLOW; 412 413 break; 414 } 415 416 case KAUTH_REQ_NETWORK_SOCKET_OPEN: 417 /* We allow "raw" routing/bluetooth sockets to anyone. */ 418 switch ((u_long)arg1) { 419 case PF_ROUTE: 420 case PF_OROUTE: 421 case PF_BLUETOOTH: 422 case PF_CAN: 423 result = KAUTH_RESULT_ALLOW; 424 break; 425 default: 426 /* Privileged, let secmodel handle this. */ 427 if ((u_long)arg2 == SOCK_RAW) 428 break; 429 result = KAUTH_RESULT_ALLOW; 430 break; 431 } 432 break; 433 434 case KAUTH_REQ_NETWORK_SOCKET_CANSEE: 435 result = KAUTH_RESULT_ALLOW; 436 437 break; 438 439 default: 440 break; 441 } 442 443 return result; 444 } 445 446 void 447 soinit(void) 448 { 449 450 sysctl_kern_socket_setup(); 451 452 #ifdef SCTP 453 /* Update the SCTP function hooks if necessary*/ 454 455 vec_sctp_add_ip_address = sctp_add_ip_address; 456 vec_sctp_delete_ip_address = sctp_delete_ip_address; 457 #endif 458 459 mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM); 460 softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 461 cv_init(&socurkva_cv, "sokva"); 462 cv_init(&pendfree_thread_cv, "sopendfr"); 463 soinit2(); 464 465 /* Set the initial adjusted socket buffer size. */ 466 if (sb_max_set(sb_max)) 467 panic("bad initial sb_max value: %lu", sb_max); 468 469 socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK, 470 socket_listener_cb, NULL); 471 } 472 473 void 474 soinit1(void) 475 { 476 int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, 477 sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree"); 478 if (error) 479 panic("soinit1 %d", error); 480 } 481 482 /* 483 * socreate: create a new socket of the specified type and the protocol. 484 * 485 * => Caller may specify another socket for lock sharing (must not be held). 486 * => Returns the new socket without lock held. 487 */ 488 int 489 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l, 490 struct socket *lockso) 491 { 492 const struct protosw *prp; 493 struct socket *so; 494 uid_t uid; 495 int error; 496 kmutex_t *lock; 497 498 error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET, 499 KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type), 500 KAUTH_ARG(proto)); 501 if (error != 0) 502 return error; 503 504 if (proto) 505 prp = pffindproto(dom, proto, type); 506 else 507 prp = pffindtype(dom, type); 508 if (prp == NULL) { 509 /* no support for domain */ 510 if (pffinddomain(dom) == 0) 511 return EAFNOSUPPORT; 512 /* no support for socket type */ 513 if (proto == 0 && type != 0) 514 return EPROTOTYPE; 515 return EPROTONOSUPPORT; 516 } 517 if (prp->pr_usrreqs == NULL) 518 return EPROTONOSUPPORT; 519 if (prp->pr_type != type) 520 return EPROTOTYPE; 521 522 so = soget(true); 523 so->so_type = type; 524 so->so_proto = prp; 525 so->so_send = sosend; 526 so->so_receive = soreceive; 527 so->so_options = sooptions; 528 #ifdef MBUFTRACE 529 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 530 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 531 so->so_mowner = &prp->pr_domain->dom_mowner; 532 #endif 533 uid = kauth_cred_geteuid(l->l_cred); 534 so->so_uidinfo = uid_find(uid); 535 so->so_egid = kauth_cred_getegid(l->l_cred); 536 so->so_cpid = l->l_proc->p_pid; 537 538 /* 539 * Lock assigned and taken during PCB attach, unless we share 540 * the lock with another socket, e.g. socketpair(2) case. 541 */ 542 if (lockso) { 543 /* 544 * lockso->so_lock should be stable at this point, so 545 * no need for atomic_load_*. 546 */ 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_hold(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 * Returns the allocated file structure in *fpp, but the descriptor 573 * is not visible yet for the process. 574 * Caller is responsible for calling fd_affix() for the returned *fpp once 575 * it's socket initialization is finished successfully, or fd_abort() if it's 576 * initialization fails. 577 * 578 * 579 * => On success, write file descriptor to *fdout and *fpp and return zero. 580 * => On failure, return non-zero; *fdout and *fpp will be undefined. 581 */ 582 int 583 fsocreate(int domain, struct socket **sop, int type, int proto, int *fdout, 584 file_t **fpp, struct socket *lockso) 585 { 586 lwp_t *l = curlwp; 587 int error, fd, flags; 588 struct socket *so; 589 file_t *fp; 590 591 flags = type & SOCK_FLAGS_MASK; 592 type &= ~SOCK_FLAGS_MASK; 593 error = socreate(domain, &so, type, proto, l, lockso); 594 if (error) { 595 return error; 596 } 597 598 if ((error = fd_allocfile(&fp, &fd)) != 0) { 599 soclose(so); 600 return error; 601 } 602 fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0); 603 fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)| 604 ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0); 605 fp->f_type = DTYPE_SOCKET; 606 fp->f_ops = &socketops; 607 if (flags & SOCK_NONBLOCK) { 608 so->so_state |= SS_NBIO; 609 } 610 fp->f_socket = so; 611 612 if (sop != NULL) { 613 *sop = so; 614 } 615 *fdout = fd; 616 *fpp = fp; 617 return error; 618 } 619 620 int 621 sofamily(const struct socket *so) 622 { 623 const struct protosw *pr; 624 const struct domain *dom; 625 626 if ((pr = so->so_proto) == NULL) 627 return AF_UNSPEC; 628 if ((dom = pr->pr_domain) == NULL) 629 return AF_UNSPEC; 630 return dom->dom_family; 631 } 632 633 int 634 sobind(struct socket *so, struct sockaddr *nam, struct lwp *l) 635 { 636 int error; 637 638 solock(so); 639 if (nam->sa_family != so->so_proto->pr_domain->dom_family) { 640 sounlock(so); 641 return EAFNOSUPPORT; 642 } 643 error = (*so->so_proto->pr_usrreqs->pr_bind)(so, nam, l); 644 sounlock(so); 645 return error; 646 } 647 648 int 649 solisten(struct socket *so, int backlog, struct lwp *l) 650 { 651 int error; 652 short oldopt, oldqlimit; 653 654 solock(so); 655 if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 656 SS_ISDISCONNECTING)) != 0) { 657 sounlock(so); 658 return EINVAL; 659 } 660 oldopt = so->so_options; 661 oldqlimit = so->so_qlimit; 662 if (TAILQ_EMPTY(&so->so_q)) 663 so->so_options |= SO_ACCEPTCONN; 664 if (backlog < 0) 665 backlog = 0; 666 so->so_qlimit = uimin(backlog, somaxconn); 667 668 error = (*so->so_proto->pr_usrreqs->pr_listen)(so, l); 669 if (error != 0) { 670 so->so_options = oldopt; 671 so->so_qlimit = oldqlimit; 672 sounlock(so); 673 return error; 674 } 675 sounlock(so); 676 return 0; 677 } 678 679 void 680 sofree(struct socket *so) 681 { 682 u_int refs; 683 684 KASSERT(solocked(so)); 685 686 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) { 687 sounlock(so); 688 return; 689 } 690 if (so->so_head) { 691 /* 692 * We must not decommission a socket that's on the accept(2) 693 * queue. If we do, then accept(2) may hang after select(2) 694 * indicated that the listening socket was ready. 695 */ 696 if (!soqremque(so, 0)) { 697 sounlock(so); 698 return; 699 } 700 } 701 if (so->so_rcv.sb_hiwat) 702 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 703 RLIM_INFINITY); 704 if (so->so_snd.sb_hiwat) 705 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 706 RLIM_INFINITY); 707 sbrelease(&so->so_snd, so); 708 KASSERT(!cv_has_waiters(&so->so_cv)); 709 KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv)); 710 KASSERT(!cv_has_waiters(&so->so_snd.sb_cv)); 711 sorflush(so); 712 refs = so->so_aborting; /* XXX */ 713 /* Remove accept filter if one is present. */ 714 if (so->so_accf != NULL) 715 (void)accept_filt_clear(so); 716 sounlock(so); 717 if (refs == 0) /* XXX */ 718 soput(so); 719 } 720 721 /* 722 * soclose: close a socket on last file table reference removal. 723 * Initiate disconnect if connected. Free socket when disconnect complete. 724 */ 725 int 726 soclose(struct socket *so) 727 { 728 struct socket *so2; 729 int error = 0; 730 731 solock(so); 732 if (so->so_options & SO_ACCEPTCONN) { 733 for (;;) { 734 if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 735 KASSERT(solocked2(so, so2)); 736 (void) soqremque(so2, 0); 737 /* soabort drops the lock. */ 738 (void) soabort(so2); 739 solock(so); 740 continue; 741 } 742 if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 743 KASSERT(solocked2(so, so2)); 744 (void) soqremque(so2, 1); 745 /* soabort drops the lock. */ 746 (void) soabort(so2); 747 solock(so); 748 continue; 749 } 750 break; 751 } 752 } 753 if (so->so_pcb == NULL) 754 goto discard; 755 if (so->so_state & SS_ISCONNECTED) { 756 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 757 error = sodisconnect(so); 758 if (error) 759 goto drop; 760 } 761 if (so->so_options & SO_LINGER) { 762 if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) == 763 (SS_ISDISCONNECTING|SS_NBIO)) 764 goto drop; 765 while (so->so_state & SS_ISCONNECTED) { 766 error = sowait(so, true, so->so_linger * hz); 767 if (error) 768 break; 769 } 770 } 771 } 772 drop: 773 if (so->so_pcb) { 774 KASSERT(solocked(so)); 775 (*so->so_proto->pr_usrreqs->pr_detach)(so); 776 } 777 discard: 778 KASSERT((so->so_state & SS_NOFDREF) == 0); 779 kauth_cred_free(so->so_cred); 780 so->so_cred = NULL; 781 so->so_state |= SS_NOFDREF; 782 sofree(so); 783 return error; 784 } 785 786 /* 787 * Must be called with the socket locked.. Will return with it unlocked. 788 */ 789 int 790 soabort(struct socket *so) 791 { 792 u_int refs; 793 int error; 794 795 KASSERT(solocked(so)); 796 KASSERT(so->so_head == NULL); 797 798 so->so_aborting++; /* XXX */ 799 error = (*so->so_proto->pr_usrreqs->pr_abort)(so); 800 refs = --so->so_aborting; /* XXX */ 801 if (error || (refs == 0)) { 802 sofree(so); 803 } else { 804 sounlock(so); 805 } 806 return error; 807 } 808 809 int 810 soaccept(struct socket *so, struct sockaddr *nam) 811 { 812 int error; 813 814 KASSERT(solocked(so)); 815 KASSERT((so->so_state & SS_NOFDREF) != 0); 816 817 so->so_state &= ~SS_NOFDREF; 818 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 819 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 820 error = (*so->so_proto->pr_usrreqs->pr_accept)(so, nam); 821 else 822 error = ECONNABORTED; 823 824 return error; 825 } 826 827 int 828 soconnect(struct socket *so, struct sockaddr *nam, struct lwp *l) 829 { 830 int error; 831 832 KASSERT(solocked(so)); 833 834 if (so->so_options & SO_ACCEPTCONN) 835 return EOPNOTSUPP; 836 /* 837 * If protocol is connection-based, can only connect once. 838 * Otherwise, if connected, try to disconnect first. 839 * This allows user to disconnect by connecting to, e.g., 840 * a null address. 841 */ 842 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 843 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 844 (error = sodisconnect(so)))) { 845 error = EISCONN; 846 } else { 847 if (nam->sa_family != so->so_proto->pr_domain->dom_family) { 848 return EAFNOSUPPORT; 849 } 850 error = (*so->so_proto->pr_usrreqs->pr_connect)(so, nam, l); 851 } 852 853 return error; 854 } 855 856 int 857 soconnect2(struct socket *so1, struct socket *so2) 858 { 859 KASSERT(solocked2(so1, so2)); 860 861 return (*so1->so_proto->pr_usrreqs->pr_connect2)(so1, so2); 862 } 863 864 int 865 sodisconnect(struct socket *so) 866 { 867 int error; 868 869 KASSERT(solocked(so)); 870 871 if ((so->so_state & SS_ISCONNECTED) == 0) { 872 error = ENOTCONN; 873 } else if (so->so_state & SS_ISDISCONNECTING) { 874 error = EALREADY; 875 } else { 876 error = (*so->so_proto->pr_usrreqs->pr_disconnect)(so); 877 } 878 return error; 879 } 880 881 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 882 /* 883 * Send on a socket. 884 * If send must go all at once and message is larger than 885 * send buffering, then hard error. 886 * Lock against other senders. 887 * If must go all at once and not enough room now, then 888 * inform user that this would block and do nothing. 889 * Otherwise, if nonblocking, send as much as possible. 890 * The data to be sent is described by "uio" if nonzero, 891 * otherwise by the mbuf chain "top" (which must be null 892 * if uio is not). Data provided in mbuf chain must be small 893 * enough to send all at once. 894 * 895 * Returns nonzero on error, timeout or signal; callers 896 * must check for short counts if EINTR/ERESTART are returned. 897 * Data and control buffers are freed on return. 898 */ 899 int 900 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 901 struct mbuf *top, struct mbuf *control, int flags, struct lwp *l) 902 { 903 struct mbuf **mp, *m; 904 long space, len, resid, clen, mlen; 905 int error, s, dontroute, atomic; 906 short wakeup_state = 0; 907 908 clen = 0; 909 910 /* 911 * solock() provides atomicity of access. splsoftnet() prevents 912 * protocol processing soft interrupts from interrupting us and 913 * blocking (expensive). 914 */ 915 s = splsoftnet(); 916 solock(so); 917 atomic = sosendallatonce(so) || top; 918 if (uio) 919 resid = uio->uio_resid; 920 else 921 resid = top->m_pkthdr.len; 922 /* 923 * In theory resid should be unsigned. 924 * However, space must be signed, as it might be less than 0 925 * if we over-committed, and we must use a signed comparison 926 * of space and resid. On the other hand, a negative resid 927 * causes us to loop sending 0-length segments to the protocol. 928 */ 929 if (resid < 0) { 930 error = EINVAL; 931 goto out; 932 } 933 dontroute = 934 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 935 (so->so_proto->pr_flags & PR_ATOMIC); 936 l->l_ru.ru_msgsnd++; 937 if (control) 938 clen = control->m_len; 939 restart: 940 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 941 goto out; 942 do { 943 if (so->so_state & SS_CANTSENDMORE) { 944 error = EPIPE; 945 goto release; 946 } 947 if (so->so_error) { 948 error = so->so_error; 949 if ((flags & MSG_PEEK) == 0) 950 so->so_error = 0; 951 goto release; 952 } 953 if ((so->so_state & SS_ISCONNECTED) == 0) { 954 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 955 if (resid || clen == 0) { 956 error = ENOTCONN; 957 goto release; 958 } 959 } else if (addr == NULL) { 960 error = EDESTADDRREQ; 961 goto release; 962 } 963 } 964 space = sbspace(&so->so_snd); 965 if (flags & MSG_OOB) 966 space += 1024; 967 if ((atomic && resid > so->so_snd.sb_hiwat) || 968 clen > so->so_snd.sb_hiwat) { 969 error = EMSGSIZE; 970 goto release; 971 } 972 if (space < resid + clen && 973 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 974 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) { 975 error = EWOULDBLOCK; 976 goto release; 977 } 978 sbunlock(&so->so_snd); 979 if (wakeup_state & SS_RESTARTSYS) { 980 error = ERESTART; 981 goto out; 982 } 983 error = sbwait(&so->so_snd); 984 if (error) 985 goto out; 986 wakeup_state = so->so_state; 987 goto restart; 988 } 989 wakeup_state = 0; 990 mp = ⊤ 991 space -= clen; 992 do { 993 if (uio == NULL) { 994 /* 995 * Data is prepackaged in "top". 996 */ 997 resid = 0; 998 if (flags & MSG_EOR) 999 top->m_flags |= M_EOR; 1000 } else do { 1001 sounlock(so); 1002 splx(s); 1003 if (top == NULL) { 1004 m = m_gethdr(M_WAIT, MT_DATA); 1005 mlen = MHLEN; 1006 m->m_pkthdr.len = 0; 1007 m_reset_rcvif(m); 1008 } else { 1009 m = m_get(M_WAIT, MT_DATA); 1010 mlen = MLEN; 1011 } 1012 MCLAIM(m, so->so_snd.sb_mowner); 1013 if (sock_loan_thresh >= 0 && 1014 uio->uio_iov->iov_len >= sock_loan_thresh && 1015 space >= sock_loan_thresh && 1016 (len = sosend_loan(so, uio, m, 1017 space)) != 0) { 1018 SOSEND_COUNTER_INCR(&sosend_loan_big); 1019 space -= len; 1020 goto have_data; 1021 } 1022 if (resid >= MINCLSIZE && space >= MCLBYTES) { 1023 SOSEND_COUNTER_INCR(&sosend_copy_big); 1024 m_clget(m, M_DONTWAIT); 1025 if ((m->m_flags & M_EXT) == 0) 1026 goto nopages; 1027 mlen = MCLBYTES; 1028 if (atomic && top == 0) { 1029 len = lmin(MCLBYTES - max_hdr, 1030 resid); 1031 m->m_data += max_hdr; 1032 } else 1033 len = lmin(MCLBYTES, resid); 1034 space -= len; 1035 } else { 1036 nopages: 1037 SOSEND_COUNTER_INCR(&sosend_copy_small); 1038 len = lmin(lmin(mlen, resid), space); 1039 space -= len; 1040 /* 1041 * For datagram protocols, leave room 1042 * for protocol headers in first mbuf. 1043 */ 1044 if (atomic && top == 0 && len < mlen) 1045 m_align(m, len); 1046 } 1047 error = uiomove(mtod(m, void *), (int)len, uio); 1048 have_data: 1049 resid = uio->uio_resid; 1050 m->m_len = len; 1051 *mp = m; 1052 top->m_pkthdr.len += len; 1053 s = splsoftnet(); 1054 solock(so); 1055 if (error != 0) 1056 goto release; 1057 mp = &m->m_next; 1058 if (resid <= 0) { 1059 if (flags & MSG_EOR) 1060 top->m_flags |= M_EOR; 1061 break; 1062 } 1063 } while (space > 0 && atomic); 1064 1065 if (so->so_state & SS_CANTSENDMORE) { 1066 error = EPIPE; 1067 goto release; 1068 } 1069 if (dontroute) 1070 so->so_options |= SO_DONTROUTE; 1071 if (resid > 0) 1072 so->so_state |= SS_MORETOCOME; 1073 if (flags & MSG_OOB) { 1074 error = (*so->so_proto->pr_usrreqs->pr_sendoob)( 1075 so, top, control); 1076 } else { 1077 error = (*so->so_proto->pr_usrreqs->pr_send)(so, 1078 top, addr, control, l); 1079 } 1080 if (dontroute) 1081 so->so_options &= ~SO_DONTROUTE; 1082 if (resid > 0) 1083 so->so_state &= ~SS_MORETOCOME; 1084 clen = 0; 1085 control = NULL; 1086 top = NULL; 1087 mp = ⊤ 1088 if (error != 0) 1089 goto release; 1090 } while (resid && space > 0); 1091 } while (resid); 1092 1093 release: 1094 sbunlock(&so->so_snd); 1095 out: 1096 sounlock(so); 1097 splx(s); 1098 m_freem(top); 1099 m_freem(control); 1100 return error; 1101 } 1102 1103 /* 1104 * Following replacement or removal of the first mbuf on the first 1105 * mbuf chain of a socket buffer, push necessary state changes back 1106 * into the socket buffer so that other consumers see the values 1107 * consistently. 'nextrecord' is the caller's locally stored value of 1108 * the original value of sb->sb_mb->m_nextpkt which must be restored 1109 * when the lead mbuf changes. NOTE: 'nextrecord' may be NULL. 1110 */ 1111 static void 1112 sbsync(struct sockbuf *sb, struct mbuf *nextrecord) 1113 { 1114 1115 KASSERT(solocked(sb->sb_so)); 1116 1117 /* 1118 * First, update for the new value of nextrecord. If necessary, 1119 * make it the first record. 1120 */ 1121 if (sb->sb_mb != NULL) 1122 sb->sb_mb->m_nextpkt = nextrecord; 1123 else 1124 sb->sb_mb = nextrecord; 1125 1126 /* 1127 * Now update any dependent socket buffer fields to reflect 1128 * the new state. This is an inline of SB_EMPTY_FIXUP, with 1129 * the addition of a second clause that takes care of the 1130 * case where sb_mb has been updated, but remains the last 1131 * record. 1132 */ 1133 if (sb->sb_mb == NULL) { 1134 sb->sb_mbtail = NULL; 1135 sb->sb_lastrecord = NULL; 1136 } else if (sb->sb_mb->m_nextpkt == NULL) 1137 sb->sb_lastrecord = sb->sb_mb; 1138 } 1139 1140 /* 1141 * Implement receive operations on a socket. 1142 * 1143 * We depend on the way that records are added to the sockbuf by sbappend*. In 1144 * particular, each record (mbufs linked through m_next) must begin with an 1145 * address if the protocol so specifies, followed by an optional mbuf or mbufs 1146 * containing ancillary data, and then zero or more mbufs of data. 1147 * 1148 * In order to avoid blocking network interrupts for the entire time here, we 1149 * splx() while doing the actual copy to user space. Although the sockbuf is 1150 * locked, new data may still be appended, and thus we must maintain 1151 * consistency of the sockbuf during that time. 1152 * 1153 * The caller may receive the data as a single mbuf chain by supplying an mbuf 1154 * **mp0 for use in returning the chain. The uio is then used only for the 1155 * count in uio_resid. 1156 */ 1157 int 1158 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 1159 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1160 { 1161 struct lwp *l = curlwp; 1162 struct mbuf *m, **mp, *mt; 1163 size_t len, offset, moff, orig_resid; 1164 int atomic, flags, error, s, type; 1165 const struct protosw *pr; 1166 struct mbuf *nextrecord; 1167 int mbuf_removed = 0; 1168 const struct domain *dom; 1169 short wakeup_state = 0; 1170 1171 pr = so->so_proto; 1172 atomic = pr->pr_flags & PR_ATOMIC; 1173 dom = pr->pr_domain; 1174 mp = mp0; 1175 type = 0; 1176 orig_resid = uio->uio_resid; 1177 1178 if (paddr != NULL) 1179 *paddr = NULL; 1180 if (controlp != NULL) 1181 *controlp = NULL; 1182 if (flagsp != NULL) 1183 flags = *flagsp &~ MSG_EOR; 1184 else 1185 flags = 0; 1186 1187 if (flags & MSG_OOB) { 1188 m = m_get(M_WAIT, MT_DATA); 1189 solock(so); 1190 error = (*pr->pr_usrreqs->pr_recvoob)(so, m, flags & MSG_PEEK); 1191 sounlock(so); 1192 if (error) 1193 goto bad; 1194 do { 1195 error = uiomove(mtod(m, void *), 1196 MIN(uio->uio_resid, m->m_len), uio); 1197 m = m_free(m); 1198 } while (uio->uio_resid > 0 && error == 0 && m); 1199 bad: 1200 m_freem(m); 1201 return error; 1202 } 1203 if (mp != NULL) 1204 *mp = NULL; 1205 1206 /* 1207 * solock() provides atomicity of access. splsoftnet() prevents 1208 * protocol processing soft interrupts from interrupting us and 1209 * blocking (expensive). 1210 */ 1211 s = splsoftnet(); 1212 solock(so); 1213 restart: 1214 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) { 1215 sounlock(so); 1216 splx(s); 1217 return error; 1218 } 1219 m = so->so_rcv.sb_mb; 1220 1221 /* 1222 * If we have less data than requested, block awaiting more 1223 * (subject to any timeout) if: 1224 * 1. the current count is less than the low water mark, 1225 * 2. MSG_WAITALL is set, and it is possible to do the entire 1226 * receive operation at once if we block (resid <= hiwat), or 1227 * 3. MSG_DONTWAIT is not set. 1228 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1229 * we have to do the receive in sections, and thus risk returning 1230 * a short count if a timeout or signal occurs after we start. 1231 */ 1232 if (m == NULL || 1233 ((flags & MSG_DONTWAIT) == 0 && 1234 so->so_rcv.sb_cc < uio->uio_resid && 1235 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1236 ((flags & MSG_WAITALL) && 1237 uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1238 m->m_nextpkt == NULL && !atomic)) { 1239 #ifdef DIAGNOSTIC 1240 if (m == NULL && so->so_rcv.sb_cc) 1241 panic("receive 1"); 1242 #endif 1243 if (so->so_error || so->so_rerror) { 1244 u_short *e; 1245 if (m != NULL) 1246 goto dontblock; 1247 e = so->so_error ? &so->so_error : &so->so_rerror; 1248 error = *e; 1249 if ((flags & MSG_PEEK) == 0) 1250 *e = 0; 1251 goto release; 1252 } 1253 if (so->so_state & SS_CANTRCVMORE) { 1254 if (m != NULL) 1255 goto dontblock; 1256 else 1257 goto release; 1258 } 1259 for (; m != NULL; m = m->m_next) 1260 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1261 m = so->so_rcv.sb_mb; 1262 goto dontblock; 1263 } 1264 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1265 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1266 error = ENOTCONN; 1267 goto release; 1268 } 1269 if (uio->uio_resid == 0) 1270 goto release; 1271 if ((so->so_state & SS_NBIO) || 1272 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1273 error = EWOULDBLOCK; 1274 goto release; 1275 } 1276 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1277 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1278 sbunlock(&so->so_rcv); 1279 if (wakeup_state & SS_RESTARTSYS) 1280 error = ERESTART; 1281 else 1282 error = sbwait(&so->so_rcv); 1283 if (error != 0) { 1284 sounlock(so); 1285 splx(s); 1286 return error; 1287 } 1288 wakeup_state = so->so_state; 1289 goto restart; 1290 } 1291 1292 dontblock: 1293 /* 1294 * On entry here, m points to the first record of the socket buffer. 1295 * From this point onward, we maintain 'nextrecord' as a cache of the 1296 * pointer to the next record in the socket buffer. We must keep the 1297 * various socket buffer pointers and local stack versions of the 1298 * pointers in sync, pushing out modifications before dropping the 1299 * socket lock, and re-reading them when picking it up. 1300 * 1301 * Otherwise, we will race with the network stack appending new data 1302 * or records onto the socket buffer by using inconsistent/stale 1303 * versions of the field, possibly resulting in socket buffer 1304 * corruption. 1305 * 1306 * By holding the high-level sblock(), we prevent simultaneous 1307 * readers from pulling off the front of the socket buffer. 1308 */ 1309 if (l != NULL) 1310 l->l_ru.ru_msgrcv++; 1311 KASSERT(m == so->so_rcv.sb_mb); 1312 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1313 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1314 nextrecord = m->m_nextpkt; 1315 1316 if (pr->pr_flags & PR_ADDR) { 1317 KASSERT(m->m_type == MT_SONAME); 1318 orig_resid = 0; 1319 if (flags & MSG_PEEK) { 1320 if (paddr) 1321 *paddr = m_copym(m, 0, m->m_len, M_DONTWAIT); 1322 m = m->m_next; 1323 } else { 1324 sbfree(&so->so_rcv, m); 1325 mbuf_removed = 1; 1326 if (paddr != NULL) { 1327 *paddr = m; 1328 so->so_rcv.sb_mb = m->m_next; 1329 m->m_next = NULL; 1330 m = so->so_rcv.sb_mb; 1331 } else { 1332 m = so->so_rcv.sb_mb = m_free(m); 1333 } 1334 sbsync(&so->so_rcv, nextrecord); 1335 } 1336 } 1337 1338 if (pr->pr_flags & PR_ADDR_OPT) { 1339 /* 1340 * For SCTP we may be getting a whole message OR a partial 1341 * delivery. 1342 */ 1343 if (m->m_type == MT_SONAME) { 1344 orig_resid = 0; 1345 if (flags & MSG_PEEK) { 1346 if (paddr) 1347 *paddr = m_copym(m, 0, m->m_len, M_DONTWAIT); 1348 m = m->m_next; 1349 } else { 1350 sbfree(&so->so_rcv, m); 1351 mbuf_removed = 1; 1352 if (paddr) { 1353 *paddr = m; 1354 so->so_rcv.sb_mb = m->m_next; 1355 m->m_next = 0; 1356 m = so->so_rcv.sb_mb; 1357 } else { 1358 m = so->so_rcv.sb_mb = m_free(m); 1359 } 1360 sbsync(&so->so_rcv, nextrecord); 1361 } 1362 } 1363 } 1364 1365 /* 1366 * Process one or more MT_CONTROL mbufs present before any data mbufs 1367 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 1368 * just copy the data; if !MSG_PEEK, we call into the protocol to 1369 * perform externalization (or freeing if controlp == NULL). 1370 */ 1371 if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) { 1372 struct mbuf *cm = NULL, *cmn; 1373 struct mbuf **cme = &cm; 1374 1375 do { 1376 if (flags & MSG_PEEK) { 1377 if (controlp != NULL) { 1378 *controlp = m_copym(m, 0, m->m_len, M_DONTWAIT); 1379 controlp = (*controlp == NULL ? NULL : 1380 &(*controlp)->m_next); 1381 } 1382 m = m->m_next; 1383 } else { 1384 sbfree(&so->so_rcv, m); 1385 so->so_rcv.sb_mb = m->m_next; 1386 m->m_next = NULL; 1387 *cme = m; 1388 cme = &(*cme)->m_next; 1389 m = so->so_rcv.sb_mb; 1390 } 1391 } while (m != NULL && m->m_type == MT_CONTROL); 1392 if ((flags & MSG_PEEK) == 0) 1393 sbsync(&so->so_rcv, nextrecord); 1394 1395 for (; cm != NULL; cm = cmn) { 1396 cmn = cm->m_next; 1397 cm->m_next = NULL; 1398 type = mtod(cm, struct cmsghdr *)->cmsg_type; 1399 if (controlp != NULL) { 1400 if (dom->dom_externalize != NULL && 1401 type == SCM_RIGHTS) { 1402 sounlock(so); 1403 splx(s); 1404 error = (*dom->dom_externalize)(cm, l, 1405 (flags & MSG_CMSG_CLOEXEC) ? 1406 O_CLOEXEC : 0); 1407 s = splsoftnet(); 1408 solock(so); 1409 } 1410 *controlp = cm; 1411 while (*controlp != NULL) 1412 controlp = &(*controlp)->m_next; 1413 } else { 1414 /* 1415 * Dispose of any SCM_RIGHTS message that went 1416 * through the read path rather than recv. 1417 */ 1418 if (dom->dom_dispose != NULL && 1419 type == SCM_RIGHTS) { 1420 sounlock(so); 1421 (*dom->dom_dispose)(cm); 1422 solock(so); 1423 } 1424 m_freem(cm); 1425 } 1426 } 1427 if (m != NULL) 1428 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 1429 else 1430 nextrecord = so->so_rcv.sb_mb; 1431 orig_resid = 0; 1432 } 1433 1434 /* If m is non-NULL, we have some data to read. */ 1435 if (__predict_true(m != NULL)) { 1436 type = m->m_type; 1437 if (type == MT_OOBDATA) 1438 flags |= MSG_OOB; 1439 } 1440 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1441 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1442 1443 moff = 0; 1444 offset = 0; 1445 while (m != NULL && uio->uio_resid > 0 && error == 0) { 1446 /* 1447 * If the type of mbuf has changed, end the receive 1448 * operation and do a short read. 1449 */ 1450 if (m->m_type == MT_OOBDATA) { 1451 if (type != MT_OOBDATA) 1452 break; 1453 } else if (type == MT_OOBDATA) { 1454 break; 1455 } else if (m->m_type == MT_CONTROL) { 1456 break; 1457 } 1458 #ifdef DIAGNOSTIC 1459 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) { 1460 panic("%s: m_type=%d", __func__, m->m_type); 1461 } 1462 #endif 1463 1464 so->so_state &= ~SS_RCVATMARK; 1465 wakeup_state = 0; 1466 len = uio->uio_resid; 1467 if (so->so_oobmark && len > so->so_oobmark - offset) 1468 len = so->so_oobmark - offset; 1469 if (len > m->m_len - moff) 1470 len = m->m_len - moff; 1471 1472 /* 1473 * If mp is set, just pass back the mbufs. 1474 * Otherwise copy them out via the uio, then free. 1475 * Sockbuf must be consistent here (points to current mbuf, 1476 * it points to next record) when we drop priority; 1477 * we must note any additions to the sockbuf when we 1478 * block interrupts again. 1479 */ 1480 if (mp == NULL) { 1481 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1482 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1483 sounlock(so); 1484 splx(s); 1485 error = uiomove(mtod(m, char *) + moff, len, uio); 1486 s = splsoftnet(); 1487 solock(so); 1488 if (error != 0) { 1489 /* 1490 * If any part of the record has been removed 1491 * (such as the MT_SONAME mbuf, which will 1492 * happen when PR_ADDR, and thus also 1493 * PR_ATOMIC, is set), then drop the entire 1494 * record to maintain the atomicity of the 1495 * receive operation. 1496 * 1497 * This avoids a later panic("receive 1a") 1498 * when compiled with DIAGNOSTIC. 1499 */ 1500 if (m && mbuf_removed && atomic) 1501 (void) sbdroprecord(&so->so_rcv); 1502 1503 goto release; 1504 } 1505 } else { 1506 uio->uio_resid -= len; 1507 } 1508 1509 if (len == m->m_len - moff) { 1510 if (m->m_flags & M_EOR) 1511 flags |= MSG_EOR; 1512 #ifdef SCTP 1513 if (m->m_flags & M_NOTIFICATION) 1514 flags |= MSG_NOTIFICATION; 1515 #endif 1516 if (flags & MSG_PEEK) { 1517 m = m->m_next; 1518 moff = 0; 1519 } else { 1520 nextrecord = m->m_nextpkt; 1521 sbfree(&so->so_rcv, m); 1522 if (mp) { 1523 *mp = m; 1524 mp = &m->m_next; 1525 so->so_rcv.sb_mb = m = m->m_next; 1526 *mp = NULL; 1527 } else { 1528 m = so->so_rcv.sb_mb = m_free(m); 1529 } 1530 /* 1531 * If m != NULL, we also know that 1532 * so->so_rcv.sb_mb != NULL. 1533 */ 1534 KASSERT(so->so_rcv.sb_mb == m); 1535 if (m) { 1536 m->m_nextpkt = nextrecord; 1537 if (nextrecord == NULL) 1538 so->so_rcv.sb_lastrecord = m; 1539 } else { 1540 so->so_rcv.sb_mb = nextrecord; 1541 SB_EMPTY_FIXUP(&so->so_rcv); 1542 } 1543 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1544 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1545 } 1546 } else if (flags & MSG_PEEK) { 1547 moff += len; 1548 } else { 1549 if (mp != NULL) { 1550 mt = m_copym(m, 0, len, M_NOWAIT); 1551 if (__predict_false(mt == NULL)) { 1552 sounlock(so); 1553 mt = m_copym(m, 0, len, M_WAIT); 1554 solock(so); 1555 } 1556 *mp = mt; 1557 } 1558 m->m_data += len; 1559 m->m_len -= len; 1560 so->so_rcv.sb_cc -= len; 1561 } 1562 1563 if (so->so_oobmark) { 1564 if ((flags & MSG_PEEK) == 0) { 1565 so->so_oobmark -= len; 1566 if (so->so_oobmark == 0) { 1567 so->so_state |= SS_RCVATMARK; 1568 break; 1569 } 1570 } else { 1571 offset += len; 1572 if (offset == so->so_oobmark) 1573 break; 1574 } 1575 } else { 1576 so->so_state &= ~SS_POLLRDBAND; 1577 } 1578 if (flags & MSG_EOR) 1579 break; 1580 1581 /* 1582 * If the MSG_WAITALL flag is set (for non-atomic socket), 1583 * we must not quit until "uio->uio_resid == 0" or an error 1584 * termination. If a signal/timeout occurs, return 1585 * with a short count but without error. 1586 * Keep sockbuf locked against other readers. 1587 */ 1588 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 1589 !sosendallatonce(so) && !nextrecord) { 1590 if (so->so_error || so->so_rerror || 1591 so->so_state & SS_CANTRCVMORE) 1592 break; 1593 /* 1594 * If we are peeking and the socket receive buffer is 1595 * full, stop since we can't get more data to peek at. 1596 */ 1597 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1598 break; 1599 /* 1600 * If we've drained the socket buffer, tell the 1601 * protocol in case it needs to do something to 1602 * get it filled again. 1603 */ 1604 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1605 (*pr->pr_usrreqs->pr_rcvd)(so, flags, l); 1606 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1607 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1608 if (wakeup_state & SS_RESTARTSYS) 1609 error = ERESTART; 1610 else 1611 error = sbwait(&so->so_rcv); 1612 if (error != 0) { 1613 sbunlock(&so->so_rcv); 1614 sounlock(so); 1615 splx(s); 1616 return 0; 1617 } 1618 if ((m = so->so_rcv.sb_mb) != NULL) 1619 nextrecord = m->m_nextpkt; 1620 wakeup_state = so->so_state; 1621 } 1622 } 1623 1624 if (m && atomic) { 1625 flags |= MSG_TRUNC; 1626 if ((flags & MSG_PEEK) == 0) 1627 (void) sbdroprecord(&so->so_rcv); 1628 } 1629 if ((flags & MSG_PEEK) == 0) { 1630 if (m == NULL) { 1631 /* 1632 * First part is an inline SB_EMPTY_FIXUP(). Second 1633 * part makes sure sb_lastrecord is up-to-date if 1634 * there is still data in the socket buffer. 1635 */ 1636 so->so_rcv.sb_mb = nextrecord; 1637 if (so->so_rcv.sb_mb == NULL) { 1638 so->so_rcv.sb_mbtail = NULL; 1639 so->so_rcv.sb_lastrecord = NULL; 1640 } else if (nextrecord->m_nextpkt == NULL) 1641 so->so_rcv.sb_lastrecord = nextrecord; 1642 } 1643 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1644 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1645 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1646 (*pr->pr_usrreqs->pr_rcvd)(so, flags, l); 1647 } 1648 if (orig_resid == uio->uio_resid && orig_resid && 1649 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1650 sbunlock(&so->so_rcv); 1651 goto restart; 1652 } 1653 1654 if (flagsp != NULL) 1655 *flagsp |= flags; 1656 release: 1657 sbunlock(&so->so_rcv); 1658 sounlock(so); 1659 splx(s); 1660 return error; 1661 } 1662 1663 int 1664 soshutdown(struct socket *so, int how) 1665 { 1666 const struct protosw *pr; 1667 int error; 1668 1669 KASSERT(solocked(so)); 1670 1671 pr = so->so_proto; 1672 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1673 return EINVAL; 1674 1675 if (how == SHUT_RD || how == SHUT_RDWR) { 1676 sorflush(so); 1677 error = 0; 1678 } 1679 if (how == SHUT_WR || how == SHUT_RDWR) 1680 error = (*pr->pr_usrreqs->pr_shutdown)(so); 1681 1682 return error; 1683 } 1684 1685 void 1686 sorestart(struct socket *so) 1687 { 1688 /* 1689 * An application has called close() on an fd on which another 1690 * of its threads has called a socket system call. 1691 * Mark this and wake everyone up, and code that would block again 1692 * instead returns ERESTART. 1693 * On system call re-entry the fd is validated and EBADF returned. 1694 * Any other fd will block again on the 2nd syscall. 1695 */ 1696 solock(so); 1697 so->so_state |= SS_RESTARTSYS; 1698 cv_broadcast(&so->so_cv); 1699 cv_broadcast(&so->so_snd.sb_cv); 1700 cv_broadcast(&so->so_rcv.sb_cv); 1701 sounlock(so); 1702 } 1703 1704 void 1705 sorflush(struct socket *so) 1706 { 1707 struct sockbuf *sb, asb; 1708 const struct protosw *pr; 1709 1710 KASSERT(solocked(so)); 1711 1712 sb = &so->so_rcv; 1713 pr = so->so_proto; 1714 socantrcvmore(so); 1715 sb->sb_flags |= SB_NOINTR; 1716 (void )sblock(sb, M_WAITOK); 1717 sbunlock(sb); 1718 asb = *sb; 1719 /* 1720 * Clear most of the sockbuf structure, but leave some of the 1721 * fields valid. 1722 */ 1723 memset(&sb->sb_startzero, 0, 1724 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1725 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) { 1726 sounlock(so); 1727 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1728 solock(so); 1729 } 1730 sbrelease(&asb, so); 1731 } 1732 1733 /* 1734 * internal set SOL_SOCKET options 1735 */ 1736 static int 1737 sosetopt1(struct socket *so, const struct sockopt *sopt) 1738 { 1739 int error, opt; 1740 int optval = 0; /* XXX: gcc */ 1741 struct linger l; 1742 struct timeval tv; 1743 1744 opt = sopt->sopt_name; 1745 1746 switch (opt) { 1747 1748 case SO_ACCEPTFILTER: 1749 error = accept_filt_setopt(so, sopt); 1750 KASSERT(solocked(so)); 1751 break; 1752 1753 case SO_LINGER: 1754 error = sockopt_get(sopt, &l, sizeof(l)); 1755 solock(so); 1756 if (error) 1757 break; 1758 if (l.l_linger < 0 || l.l_linger > USHRT_MAX || 1759 l.l_linger > (INT_MAX / hz)) { 1760 error = EDOM; 1761 break; 1762 } 1763 so->so_linger = l.l_linger; 1764 if (l.l_onoff) 1765 so->so_options |= SO_LINGER; 1766 else 1767 so->so_options &= ~SO_LINGER; 1768 break; 1769 1770 case SO_DEBUG: 1771 case SO_KEEPALIVE: 1772 case SO_DONTROUTE: 1773 case SO_USELOOPBACK: 1774 case SO_BROADCAST: 1775 case SO_REUSEADDR: 1776 case SO_REUSEPORT: 1777 case SO_OOBINLINE: 1778 case SO_TIMESTAMP: 1779 case SO_NOSIGPIPE: 1780 case SO_RERROR: 1781 error = sockopt_getint(sopt, &optval); 1782 solock(so); 1783 if (error) 1784 break; 1785 if (optval) 1786 so->so_options |= opt; 1787 else 1788 so->so_options &= ~opt; 1789 break; 1790 1791 case SO_SNDBUF: 1792 case SO_RCVBUF: 1793 case SO_SNDLOWAT: 1794 case SO_RCVLOWAT: 1795 error = sockopt_getint(sopt, &optval); 1796 solock(so); 1797 if (error) 1798 break; 1799 1800 /* 1801 * Values < 1 make no sense for any of these 1802 * options, so disallow them. 1803 */ 1804 if (optval < 1) { 1805 error = EINVAL; 1806 break; 1807 } 1808 1809 switch (opt) { 1810 case SO_SNDBUF: 1811 if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) { 1812 error = ENOBUFS; 1813 break; 1814 } 1815 if (sofixedbuf) 1816 so->so_snd.sb_flags &= ~SB_AUTOSIZE; 1817 break; 1818 1819 case SO_RCVBUF: 1820 if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) { 1821 error = ENOBUFS; 1822 break; 1823 } 1824 if (sofixedbuf) 1825 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1826 break; 1827 1828 /* 1829 * Make sure the low-water is never greater than 1830 * the high-water. 1831 */ 1832 case SO_SNDLOWAT: 1833 if (optval > so->so_snd.sb_hiwat) 1834 optval = so->so_snd.sb_hiwat; 1835 1836 so->so_snd.sb_lowat = optval; 1837 break; 1838 1839 case SO_RCVLOWAT: 1840 if (optval > so->so_rcv.sb_hiwat) 1841 optval = so->so_rcv.sb_hiwat; 1842 1843 so->so_rcv.sb_lowat = optval; 1844 break; 1845 } 1846 break; 1847 1848 case SO_SNDTIMEO: 1849 case SO_RCVTIMEO: 1850 solock(so); 1851 error = sockopt_get(sopt, &tv, sizeof(tv)); 1852 if (error) 1853 break; 1854 1855 if (tv.tv_sec < 0 || tv.tv_usec < 0 || tv.tv_usec >= 1000000) { 1856 error = EDOM; 1857 break; 1858 } 1859 if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) { 1860 error = EDOM; 1861 break; 1862 } 1863 1864 optval = tv.tv_sec * hz + tv.tv_usec / tick; 1865 if (optval == 0 && tv.tv_usec != 0) 1866 optval = 1; 1867 1868 switch (opt) { 1869 case SO_SNDTIMEO: 1870 so->so_snd.sb_timeo = optval; 1871 break; 1872 case SO_RCVTIMEO: 1873 so->so_rcv.sb_timeo = optval; 1874 break; 1875 } 1876 break; 1877 1878 default: 1879 MODULE_HOOK_CALL(uipc_socket_50_setopt1_hook, 1880 (opt, so, sopt), enosys(), error); 1881 if (error == ENOSYS || error == EPASSTHROUGH) { 1882 solock(so); 1883 error = ENOPROTOOPT; 1884 } 1885 break; 1886 } 1887 KASSERT(solocked(so)); 1888 return error; 1889 } 1890 1891 int 1892 sosetopt(struct socket *so, struct sockopt *sopt) 1893 { 1894 int error, prerr; 1895 1896 if (sopt->sopt_level == SOL_SOCKET) { 1897 error = sosetopt1(so, sopt); 1898 KASSERT(solocked(so)); 1899 } else { 1900 error = ENOPROTOOPT; 1901 solock(so); 1902 } 1903 1904 if ((error == 0 || error == ENOPROTOOPT) && 1905 so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) { 1906 /* give the protocol stack a shot */ 1907 prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt); 1908 if (prerr == 0) 1909 error = 0; 1910 else if (prerr != ENOPROTOOPT) 1911 error = prerr; 1912 } 1913 sounlock(so); 1914 return error; 1915 } 1916 1917 /* 1918 * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt() 1919 */ 1920 int 1921 so_setsockopt(struct lwp *l, struct socket *so, int level, int name, 1922 const void *val, size_t valsize) 1923 { 1924 struct sockopt sopt; 1925 int error; 1926 1927 KASSERT(valsize == 0 || val != NULL); 1928 1929 sockopt_init(&sopt, level, name, valsize); 1930 sockopt_set(&sopt, val, valsize); 1931 1932 error = sosetopt(so, &sopt); 1933 1934 sockopt_destroy(&sopt); 1935 1936 return error; 1937 } 1938 1939 /* 1940 * internal get SOL_SOCKET options 1941 */ 1942 static int 1943 sogetopt1(struct socket *so, struct sockopt *sopt) 1944 { 1945 int error, optval, opt; 1946 struct linger l; 1947 struct timeval tv; 1948 1949 switch ((opt = sopt->sopt_name)) { 1950 1951 case SO_ACCEPTFILTER: 1952 error = accept_filt_getopt(so, sopt); 1953 break; 1954 1955 case SO_LINGER: 1956 l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0; 1957 l.l_linger = so->so_linger; 1958 1959 error = sockopt_set(sopt, &l, sizeof(l)); 1960 break; 1961 1962 case SO_USELOOPBACK: 1963 case SO_DONTROUTE: 1964 case SO_DEBUG: 1965 case SO_KEEPALIVE: 1966 case SO_REUSEADDR: 1967 case SO_REUSEPORT: 1968 case SO_BROADCAST: 1969 case SO_OOBINLINE: 1970 case SO_TIMESTAMP: 1971 case SO_NOSIGPIPE: 1972 case SO_RERROR: 1973 case SO_ACCEPTCONN: 1974 error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0); 1975 break; 1976 1977 case SO_TYPE: 1978 error = sockopt_setint(sopt, so->so_type); 1979 break; 1980 1981 case SO_ERROR: 1982 if (so->so_error == 0) { 1983 so->so_error = so->so_rerror; 1984 so->so_rerror = 0; 1985 } 1986 error = sockopt_setint(sopt, so->so_error); 1987 so->so_error = 0; 1988 break; 1989 1990 case SO_SNDBUF: 1991 error = sockopt_setint(sopt, so->so_snd.sb_hiwat); 1992 break; 1993 1994 case SO_RCVBUF: 1995 error = sockopt_setint(sopt, so->so_rcv.sb_hiwat); 1996 break; 1997 1998 case SO_SNDLOWAT: 1999 error = sockopt_setint(sopt, so->so_snd.sb_lowat); 2000 break; 2001 2002 case SO_RCVLOWAT: 2003 error = sockopt_setint(sopt, so->so_rcv.sb_lowat); 2004 break; 2005 2006 case SO_SNDTIMEO: 2007 case SO_RCVTIMEO: 2008 optval = (opt == SO_SNDTIMEO ? 2009 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 2010 2011 memset(&tv, 0, sizeof(tv)); 2012 tv.tv_sec = optval / hz; 2013 tv.tv_usec = (optval % hz) * tick; 2014 2015 error = sockopt_set(sopt, &tv, sizeof(tv)); 2016 break; 2017 2018 case SO_OVERFLOWED: 2019 error = sockopt_setint(sopt, so->so_rcv.sb_overflowed); 2020 break; 2021 2022 default: 2023 MODULE_HOOK_CALL(uipc_socket_50_getopt1_hook, 2024 (opt, so, sopt), enosys(), error); 2025 if (error) 2026 error = ENOPROTOOPT; 2027 break; 2028 } 2029 2030 return error; 2031 } 2032 2033 int 2034 sogetopt(struct socket *so, struct sockopt *sopt) 2035 { 2036 int error; 2037 2038 solock(so); 2039 if (sopt->sopt_level != SOL_SOCKET) { 2040 if (so->so_proto && so->so_proto->pr_ctloutput) { 2041 error = ((*so->so_proto->pr_ctloutput) 2042 (PRCO_GETOPT, so, sopt)); 2043 } else 2044 error = (ENOPROTOOPT); 2045 } else { 2046 error = sogetopt1(so, sopt); 2047 } 2048 sounlock(so); 2049 return error; 2050 } 2051 2052 /* 2053 * alloc sockopt data buffer buffer 2054 * - will be released at destroy 2055 */ 2056 static int 2057 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag) 2058 { 2059 void *data; 2060 2061 KASSERT(sopt->sopt_size == 0); 2062 2063 if (len > sizeof(sopt->sopt_buf)) { 2064 data = kmem_zalloc(len, kmflag); 2065 if (data == NULL) 2066 return ENOMEM; 2067 sopt->sopt_data = data; 2068 } else 2069 sopt->sopt_data = sopt->sopt_buf; 2070 2071 sopt->sopt_size = len; 2072 return 0; 2073 } 2074 2075 /* 2076 * initialise sockopt storage 2077 * - MAY sleep during allocation 2078 */ 2079 void 2080 sockopt_init(struct sockopt *sopt, int level, int name, size_t size) 2081 { 2082 2083 memset(sopt, 0, sizeof(*sopt)); 2084 2085 sopt->sopt_level = level; 2086 sopt->sopt_name = name; 2087 (void)sockopt_alloc(sopt, size, KM_SLEEP); 2088 } 2089 2090 /* 2091 * destroy sockopt storage 2092 * - will release any held memory references 2093 */ 2094 void 2095 sockopt_destroy(struct sockopt *sopt) 2096 { 2097 2098 if (sopt->sopt_data != sopt->sopt_buf) 2099 kmem_free(sopt->sopt_data, sopt->sopt_size); 2100 2101 memset(sopt, 0, sizeof(*sopt)); 2102 } 2103 2104 /* 2105 * set sockopt value 2106 * - value is copied into sockopt 2107 * - memory is allocated when necessary, will not sleep 2108 */ 2109 int 2110 sockopt_set(struct sockopt *sopt, const void *buf, size_t len) 2111 { 2112 int error; 2113 2114 if (sopt->sopt_size == 0) { 2115 error = sockopt_alloc(sopt, len, KM_NOSLEEP); 2116 if (error) 2117 return error; 2118 } 2119 2120 sopt->sopt_retsize = MIN(sopt->sopt_size, len); 2121 if (sopt->sopt_retsize > 0) { 2122 memcpy(sopt->sopt_data, buf, sopt->sopt_retsize); 2123 } 2124 2125 return 0; 2126 } 2127 2128 /* 2129 * common case of set sockopt integer value 2130 */ 2131 int 2132 sockopt_setint(struct sockopt *sopt, int val) 2133 { 2134 2135 return sockopt_set(sopt, &val, sizeof(int)); 2136 } 2137 2138 /* 2139 * get sockopt value 2140 * - correct size must be given 2141 */ 2142 int 2143 sockopt_get(const struct sockopt *sopt, void *buf, size_t len) 2144 { 2145 2146 if (sopt->sopt_size != len) 2147 return EINVAL; 2148 2149 memcpy(buf, sopt->sopt_data, len); 2150 return 0; 2151 } 2152 2153 /* 2154 * common case of get sockopt integer value 2155 */ 2156 int 2157 sockopt_getint(const struct sockopt *sopt, int *valp) 2158 { 2159 2160 return sockopt_get(sopt, valp, sizeof(int)); 2161 } 2162 2163 /* 2164 * set sockopt value from mbuf 2165 * - ONLY for legacy code 2166 * - mbuf is released by sockopt 2167 * - will not sleep 2168 */ 2169 int 2170 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m) 2171 { 2172 size_t len; 2173 int error; 2174 2175 len = m_length(m); 2176 2177 if (sopt->sopt_size == 0) { 2178 error = sockopt_alloc(sopt, len, KM_NOSLEEP); 2179 if (error) 2180 return error; 2181 } 2182 2183 sopt->sopt_retsize = MIN(sopt->sopt_size, len); 2184 m_copydata(m, 0, sopt->sopt_retsize, sopt->sopt_data); 2185 m_freem(m); 2186 2187 return 0; 2188 } 2189 2190 /* 2191 * get sockopt value into mbuf 2192 * - ONLY for legacy code 2193 * - mbuf to be released by the caller 2194 * - will not sleep 2195 */ 2196 struct mbuf * 2197 sockopt_getmbuf(const struct sockopt *sopt) 2198 { 2199 struct mbuf *m; 2200 2201 if (sopt->sopt_size > MCLBYTES) 2202 return NULL; 2203 2204 m = m_get(M_DONTWAIT, MT_SOOPTS); 2205 if (m == NULL) 2206 return NULL; 2207 2208 if (sopt->sopt_size > MLEN) { 2209 MCLGET(m, M_DONTWAIT); 2210 if ((m->m_flags & M_EXT) == 0) { 2211 m_free(m); 2212 return NULL; 2213 } 2214 } 2215 2216 memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size); 2217 m->m_len = sopt->sopt_size; 2218 2219 return m; 2220 } 2221 2222 void 2223 sohasoutofband(struct socket *so) 2224 { 2225 2226 so->so_state |= SS_POLLRDBAND; 2227 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 2228 selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT); 2229 } 2230 2231 static void 2232 filt_sordetach(struct knote *kn) 2233 { 2234 struct socket *so; 2235 2236 so = ((file_t *)kn->kn_obj)->f_socket; 2237 solock(so); 2238 if (selremove_knote(&so->so_rcv.sb_sel, kn)) 2239 so->so_rcv.sb_flags &= ~SB_KNOTE; 2240 sounlock(so); 2241 } 2242 2243 /*ARGSUSED*/ 2244 static int 2245 filt_soread(struct knote *kn, long hint) 2246 { 2247 struct socket *so; 2248 int rv; 2249 2250 so = ((file_t *)kn->kn_obj)->f_socket; 2251 if (hint != NOTE_SUBMIT) 2252 solock(so); 2253 kn->kn_data = so->so_rcv.sb_cc; 2254 if (so->so_state & SS_CANTRCVMORE) { 2255 knote_set_eof(kn, 0); 2256 kn->kn_fflags = so->so_error; 2257 rv = 1; 2258 } else if (so->so_error || so->so_rerror) 2259 rv = 1; 2260 else if (kn->kn_sfflags & NOTE_LOWAT) 2261 rv = (kn->kn_data >= kn->kn_sdata); 2262 else 2263 rv = (kn->kn_data >= so->so_rcv.sb_lowat); 2264 if (hint != NOTE_SUBMIT) 2265 sounlock(so); 2266 return rv; 2267 } 2268 2269 static void 2270 filt_sowdetach(struct knote *kn) 2271 { 2272 struct socket *so; 2273 2274 so = ((file_t *)kn->kn_obj)->f_socket; 2275 solock(so); 2276 if (selremove_knote(&so->so_snd.sb_sel, kn)) 2277 so->so_snd.sb_flags &= ~SB_KNOTE; 2278 sounlock(so); 2279 } 2280 2281 /*ARGSUSED*/ 2282 static int 2283 filt_sowrite(struct knote *kn, long hint) 2284 { 2285 struct socket *so; 2286 int rv; 2287 2288 so = ((file_t *)kn->kn_obj)->f_socket; 2289 if (hint != NOTE_SUBMIT) 2290 solock(so); 2291 kn->kn_data = sbspace(&so->so_snd); 2292 if (so->so_state & SS_CANTSENDMORE) { 2293 knote_set_eof(kn, 0); 2294 kn->kn_fflags = so->so_error; 2295 rv = 1; 2296 } else if (so->so_error) 2297 rv = 1; 2298 else if (((so->so_state & SS_ISCONNECTED) == 0) && 2299 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 2300 rv = 0; 2301 else if (kn->kn_sfflags & NOTE_LOWAT) 2302 rv = (kn->kn_data >= kn->kn_sdata); 2303 else 2304 rv = (kn->kn_data >= so->so_snd.sb_lowat); 2305 if (hint != NOTE_SUBMIT) 2306 sounlock(so); 2307 return rv; 2308 } 2309 2310 static int 2311 filt_soempty(struct knote *kn, long hint) 2312 { 2313 struct socket *so; 2314 int rv; 2315 2316 so = ((file_t *)kn->kn_obj)->f_socket; 2317 if (hint != NOTE_SUBMIT) 2318 solock(so); 2319 rv = (kn->kn_data = sbused(&so->so_snd)) == 0 || 2320 (so->so_options & SO_ACCEPTCONN) != 0; 2321 if (hint != NOTE_SUBMIT) 2322 sounlock(so); 2323 return rv; 2324 } 2325 2326 /*ARGSUSED*/ 2327 static int 2328 filt_solisten(struct knote *kn, long hint) 2329 { 2330 struct socket *so; 2331 int rv; 2332 2333 so = ((file_t *)kn->kn_obj)->f_socket; 2334 2335 /* 2336 * Set kn_data to number of incoming connections, not 2337 * counting partial (incomplete) connections. 2338 */ 2339 if (hint != NOTE_SUBMIT) 2340 solock(so); 2341 kn->kn_data = so->so_qlen; 2342 rv = (kn->kn_data > 0); 2343 if (hint != NOTE_SUBMIT) 2344 sounlock(so); 2345 return rv; 2346 } 2347 2348 static const struct filterops solisten_filtops = { 2349 .f_flags = FILTEROP_ISFD | FILTEROP_MPSAFE, 2350 .f_attach = NULL, 2351 .f_detach = filt_sordetach, 2352 .f_event = filt_solisten, 2353 }; 2354 2355 static const struct filterops soread_filtops = { 2356 .f_flags = FILTEROP_ISFD | FILTEROP_MPSAFE, 2357 .f_attach = NULL, 2358 .f_detach = filt_sordetach, 2359 .f_event = filt_soread, 2360 }; 2361 2362 static const struct filterops sowrite_filtops = { 2363 .f_flags = FILTEROP_ISFD | FILTEROP_MPSAFE, 2364 .f_attach = NULL, 2365 .f_detach = filt_sowdetach, 2366 .f_event = filt_sowrite, 2367 }; 2368 2369 static const struct filterops soempty_filtops = { 2370 .f_flags = FILTEROP_ISFD | FILTEROP_MPSAFE, 2371 .f_attach = NULL, 2372 .f_detach = filt_sowdetach, 2373 .f_event = filt_soempty, 2374 }; 2375 2376 int 2377 soo_kqfilter(struct file *fp, struct knote *kn) 2378 { 2379 struct socket *so; 2380 struct sockbuf *sb; 2381 2382 so = ((file_t *)kn->kn_obj)->f_socket; 2383 solock(so); 2384 switch (kn->kn_filter) { 2385 case EVFILT_READ: 2386 if (so->so_options & SO_ACCEPTCONN) 2387 kn->kn_fop = &solisten_filtops; 2388 else 2389 kn->kn_fop = &soread_filtops; 2390 sb = &so->so_rcv; 2391 break; 2392 case EVFILT_WRITE: 2393 kn->kn_fop = &sowrite_filtops; 2394 sb = &so->so_snd; 2395 2396 #ifdef PIPE_SOCKETPAIR 2397 if (so->so_state & SS_ISAPIPE) { 2398 /* Other end of pipe has been closed. */ 2399 if (so->so_state & SS_ISDISCONNECTED) { 2400 sounlock(so); 2401 return EBADF; 2402 } 2403 } 2404 #endif 2405 break; 2406 case EVFILT_EMPTY: 2407 kn->kn_fop = &soempty_filtops; 2408 sb = &so->so_snd; 2409 break; 2410 default: 2411 sounlock(so); 2412 return EINVAL; 2413 } 2414 selrecord_knote(&sb->sb_sel, kn); 2415 sb->sb_flags |= SB_KNOTE; 2416 sounlock(so); 2417 return 0; 2418 } 2419 2420 static int 2421 sodopoll(struct socket *so, int events) 2422 { 2423 int revents; 2424 2425 revents = 0; 2426 2427 if (events & (POLLIN | POLLRDNORM)) 2428 if (soreadable(so)) 2429 revents |= events & (POLLIN | POLLRDNORM); 2430 2431 if (events & (POLLOUT | POLLWRNORM)) 2432 if (sowritable(so)) 2433 revents |= events & (POLLOUT | POLLWRNORM); 2434 2435 if (events & (POLLPRI | POLLRDBAND)) 2436 if (so->so_state & SS_POLLRDBAND) 2437 revents |= events & (POLLPRI | POLLRDBAND); 2438 2439 return revents; 2440 } 2441 2442 int 2443 sopoll(struct socket *so, int events) 2444 { 2445 int revents = 0; 2446 2447 #ifndef DIAGNOSTIC 2448 /* 2449 * Do a quick, unlocked check in expectation that the socket 2450 * will be ready for I/O. Don't do this check if DIAGNOSTIC, 2451 * as the solocked() assertions will fail. 2452 */ 2453 if ((revents = sodopoll(so, events)) != 0) 2454 return revents; 2455 #endif 2456 2457 solock(so); 2458 if ((revents = sodopoll(so, events)) == 0) { 2459 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 2460 selrecord(curlwp, &so->so_rcv.sb_sel); 2461 so->so_rcv.sb_flags |= SB_NOTIFY; 2462 } 2463 2464 if (events & (POLLOUT | POLLWRNORM)) { 2465 selrecord(curlwp, &so->so_snd.sb_sel); 2466 so->so_snd.sb_flags |= SB_NOTIFY; 2467 } 2468 } 2469 sounlock(so); 2470 2471 return revents; 2472 } 2473 2474 struct mbuf ** 2475 sbsavetimestamp(int opt, struct mbuf **mp) 2476 { 2477 struct timeval tv; 2478 int error; 2479 2480 memset(&tv, 0, sizeof(tv)); 2481 microtime(&tv); 2482 2483 MODULE_HOOK_CALL(uipc_socket_50_sbts_hook, (opt, &mp), enosys(), error); 2484 if (error == 0) 2485 return mp; 2486 2487 if (opt & SO_TIMESTAMP) { 2488 *mp = sbcreatecontrol(&tv, sizeof(tv), 2489 SCM_TIMESTAMP, SOL_SOCKET); 2490 if (*mp) 2491 mp = &(*mp)->m_next; 2492 } 2493 return mp; 2494 } 2495 2496 2497 #include <sys/sysctl.h> 2498 2499 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 2500 static int sysctl_kern_sbmax(SYSCTLFN_PROTO); 2501 2502 /* 2503 * sysctl helper routine for kern.somaxkva. ensures that the given 2504 * value is not too small. 2505 * (XXX should we maybe make sure it's not too large as well?) 2506 */ 2507 static int 2508 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 2509 { 2510 int error, new_somaxkva; 2511 struct sysctlnode node; 2512 2513 new_somaxkva = somaxkva; 2514 node = *rnode; 2515 node.sysctl_data = &new_somaxkva; 2516 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2517 if (error || newp == NULL) 2518 return error; 2519 2520 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 2521 return EINVAL; 2522 2523 mutex_enter(&so_pendfree_lock); 2524 somaxkva = new_somaxkva; 2525 cv_broadcast(&socurkva_cv); 2526 mutex_exit(&so_pendfree_lock); 2527 2528 return error; 2529 } 2530 2531 /* 2532 * sysctl helper routine for kern.sbmax. Basically just ensures that 2533 * any new value is not too small. 2534 */ 2535 static int 2536 sysctl_kern_sbmax(SYSCTLFN_ARGS) 2537 { 2538 int error, new_sbmax; 2539 struct sysctlnode node; 2540 2541 new_sbmax = sb_max; 2542 node = *rnode; 2543 node.sysctl_data = &new_sbmax; 2544 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2545 if (error || newp == NULL) 2546 return error; 2547 2548 KERNEL_LOCK(1, NULL); 2549 error = sb_max_set(new_sbmax); 2550 KERNEL_UNLOCK_ONE(NULL); 2551 2552 return error; 2553 } 2554 2555 /* 2556 * sysctl helper routine for kern.sooptions. Ensures that only allowed 2557 * options can be set. 2558 */ 2559 static int 2560 sysctl_kern_sooptions(SYSCTLFN_ARGS) 2561 { 2562 int error, new_options; 2563 struct sysctlnode node; 2564 2565 new_options = sooptions; 2566 node = *rnode; 2567 node.sysctl_data = &new_options; 2568 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2569 if (error || newp == NULL) 2570 return error; 2571 2572 if (new_options & ~SO_DEFOPTS) 2573 return EINVAL; 2574 2575 sooptions = new_options; 2576 2577 return 0; 2578 } 2579 2580 static void 2581 sysctl_kern_socket_setup(void) 2582 { 2583 2584 KASSERT(socket_sysctllog == NULL); 2585 2586 sysctl_createv(&socket_sysctllog, 0, NULL, NULL, 2587 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2588 CTLTYPE_INT, "somaxkva", 2589 SYSCTL_DESCR("Maximum amount of kernel memory to be " 2590 "used for socket buffers"), 2591 sysctl_kern_somaxkva, 0, NULL, 0, 2592 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 2593 2594 sysctl_createv(&socket_sysctllog, 0, NULL, NULL, 2595 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2596 CTLTYPE_BOOL, "sofixedbuf", 2597 SYSCTL_DESCR("Prevent scaling of fixed socket buffers"), 2598 NULL, 0, &sofixedbuf, 0, 2599 CTL_KERN, KERN_SOFIXEDBUF, CTL_EOL); 2600 2601 sysctl_createv(&socket_sysctllog, 0, NULL, NULL, 2602 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2603 CTLTYPE_INT, "sbmax", 2604 SYSCTL_DESCR("Maximum socket buffer size"), 2605 sysctl_kern_sbmax, 0, NULL, 0, 2606 CTL_KERN, KERN_SBMAX, CTL_EOL); 2607 2608 sysctl_createv(&socket_sysctllog, 0, NULL, NULL, 2609 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2610 CTLTYPE_INT, "sooptions", 2611 SYSCTL_DESCR("Default socket options"), 2612 sysctl_kern_sooptions, 0, NULL, 0, 2613 CTL_KERN, CTL_CREATE, CTL_EOL); 2614 } 2615