1 /* $NetBSD: uipc_socket.c,v 1.214 2013/03/14 19:13:17 gdt 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 #include <sys/cdefs.h> 66 __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.214 2013/03/14 19:13:17 gdt Exp $"); 67 68 #include "opt_compat_netbsd.h" 69 #include "opt_sock_counters.h" 70 #include "opt_sosend_loan.h" 71 #include "opt_mbuftrace.h" 72 #include "opt_somaxkva.h" 73 #include "opt_multiprocessor.h" /* XXX */ 74 75 #include <sys/param.h> 76 #include <sys/systm.h> 77 #include <sys/proc.h> 78 #include <sys/file.h> 79 #include <sys/filedesc.h> 80 #include <sys/kmem.h> 81 #include <sys/mbuf.h> 82 #include <sys/domain.h> 83 #include <sys/kernel.h> 84 #include <sys/protosw.h> 85 #include <sys/socket.h> 86 #include <sys/socketvar.h> 87 #include <sys/signalvar.h> 88 #include <sys/resourcevar.h> 89 #include <sys/uidinfo.h> 90 #include <sys/event.h> 91 #include <sys/poll.h> 92 #include <sys/kauth.h> 93 #include <sys/mutex.h> 94 #include <sys/condvar.h> 95 #include <sys/kthread.h> 96 97 #ifdef COMPAT_50 98 #include <compat/sys/time.h> 99 #include <compat/sys/socket.h> 100 #endif 101 102 #include <uvm/uvm_extern.h> 103 #include <uvm/uvm_loan.h> 104 #include <uvm/uvm_page.h> 105 106 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options"); 107 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 108 109 extern const struct fileops socketops; 110 111 extern int somaxconn; /* patchable (XXX sysctl) */ 112 int somaxconn = SOMAXCONN; 113 kmutex_t *softnet_lock; 114 115 #ifdef SOSEND_COUNTERS 116 #include <sys/device.h> 117 118 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 119 NULL, "sosend", "loan big"); 120 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 121 NULL, "sosend", "copy big"); 122 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 123 NULL, "sosend", "copy small"); 124 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 125 NULL, "sosend", "kva limit"); 126 127 #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++ 128 129 EVCNT_ATTACH_STATIC(sosend_loan_big); 130 EVCNT_ATTACH_STATIC(sosend_copy_big); 131 EVCNT_ATTACH_STATIC(sosend_copy_small); 132 EVCNT_ATTACH_STATIC(sosend_kvalimit); 133 #else 134 135 #define SOSEND_COUNTER_INCR(ev) /* nothing */ 136 137 #endif /* SOSEND_COUNTERS */ 138 139 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR) 140 int sock_loan_thresh = -1; 141 #else 142 int sock_loan_thresh = 4096; 143 #endif 144 145 static kmutex_t so_pendfree_lock; 146 static struct mbuf *so_pendfree = NULL; 147 148 #ifndef SOMAXKVA 149 #define SOMAXKVA (16 * 1024 * 1024) 150 #endif 151 int somaxkva = SOMAXKVA; 152 static int socurkva; 153 static kcondvar_t socurkva_cv; 154 155 static kauth_listener_t socket_listener; 156 157 #define SOCK_LOAN_CHUNK 65536 158 159 static void sopendfree_thread(void *); 160 static kcondvar_t pendfree_thread_cv; 161 static lwp_t *sopendfree_lwp; 162 163 static void sysctl_kern_socket_setup(void); 164 static struct sysctllog *socket_sysctllog; 165 166 static vsize_t 167 sokvareserve(struct socket *so, vsize_t len) 168 { 169 int error; 170 171 mutex_enter(&so_pendfree_lock); 172 while (socurkva + len > somaxkva) { 173 SOSEND_COUNTER_INCR(&sosend_kvalimit); 174 error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock); 175 if (error) { 176 len = 0; 177 break; 178 } 179 } 180 socurkva += len; 181 mutex_exit(&so_pendfree_lock); 182 return len; 183 } 184 185 static void 186 sokvaunreserve(vsize_t len) 187 { 188 189 mutex_enter(&so_pendfree_lock); 190 socurkva -= len; 191 cv_broadcast(&socurkva_cv); 192 mutex_exit(&so_pendfree_lock); 193 } 194 195 /* 196 * sokvaalloc: allocate kva for loan. 197 */ 198 199 vaddr_t 200 sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so) 201 { 202 vaddr_t lva; 203 204 /* 205 * reserve kva. 206 */ 207 208 if (sokvareserve(so, len) == 0) 209 return 0; 210 211 /* 212 * allocate kva. 213 */ 214 215 lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask, 216 UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA); 217 if (lva == 0) { 218 sokvaunreserve(len); 219 return (0); 220 } 221 222 return lva; 223 } 224 225 /* 226 * sokvafree: free kva for loan. 227 */ 228 229 void 230 sokvafree(vaddr_t sva, vsize_t len) 231 { 232 233 /* 234 * free kva. 235 */ 236 237 uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY); 238 239 /* 240 * unreserve kva. 241 */ 242 243 sokvaunreserve(len); 244 } 245 246 static void 247 sodoloanfree(struct vm_page **pgs, void *buf, size_t size) 248 { 249 vaddr_t sva, eva; 250 vsize_t len; 251 int npgs; 252 253 KASSERT(pgs != NULL); 254 255 eva = round_page((vaddr_t) buf + size); 256 sva = trunc_page((vaddr_t) buf); 257 len = eva - sva; 258 npgs = len >> PAGE_SHIFT; 259 260 pmap_kremove(sva, len); 261 pmap_update(pmap_kernel()); 262 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 263 sokvafree(sva, len); 264 } 265 266 /* 267 * sopendfree_thread: free mbufs on "pendfree" list. 268 * unlock and relock so_pendfree_lock when freeing mbufs. 269 */ 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)) == 0); 289 KASSERT(m->m_ext.ext_refcnt == 0); 290 291 rv += m->m_ext.ext_size; 292 sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf, 293 m->m_ext.ext_size); 294 pool_cache_put(mb_cache, m); 295 } 296 297 mutex_enter(&so_pendfree_lock); 298 } 299 if (rv) 300 cv_broadcast(&socurkva_cv); 301 cv_wait(&pendfree_thread_cv, &so_pendfree_lock); 302 } 303 panic("sopendfree_thread"); 304 /* NOTREACHED */ 305 } 306 307 void 308 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg) 309 { 310 311 KASSERT(m != NULL); 312 313 /* 314 * postpone freeing mbuf. 315 * 316 * we can't do it in interrupt context 317 * because we need to put kva back to kernel_map. 318 */ 319 320 mutex_enter(&so_pendfree_lock); 321 m->m_next = so_pendfree; 322 so_pendfree = m; 323 cv_signal(&pendfree_thread_cv); 324 mutex_exit(&so_pendfree_lock); 325 } 326 327 static long 328 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 329 { 330 struct iovec *iov = uio->uio_iov; 331 vaddr_t sva, eva; 332 vsize_t len; 333 vaddr_t lva; 334 int npgs, error; 335 vaddr_t va; 336 int i; 337 338 if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace)) 339 return (0); 340 341 if (iov->iov_len < (size_t) space) 342 space = iov->iov_len; 343 if (space > SOCK_LOAN_CHUNK) 344 space = SOCK_LOAN_CHUNK; 345 346 eva = round_page((vaddr_t) iov->iov_base + space); 347 sva = trunc_page((vaddr_t) iov->iov_base); 348 len = eva - sva; 349 npgs = len >> PAGE_SHIFT; 350 351 KASSERT(npgs <= M_EXT_MAXPAGES); 352 353 lva = sokvaalloc(sva, len, so); 354 if (lva == 0) 355 return 0; 356 357 error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len, 358 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 359 if (error) { 360 sokvafree(lva, len); 361 return (0); 362 } 363 364 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 365 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 366 VM_PROT_READ, 0); 367 pmap_update(pmap_kernel()); 368 369 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 370 371 MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so); 372 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 373 374 uio->uio_resid -= space; 375 /* uio_offset not updated, not set/used for write(2) */ 376 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space; 377 uio->uio_iov->iov_len -= space; 378 if (uio->uio_iov->iov_len == 0) { 379 uio->uio_iov++; 380 uio->uio_iovcnt--; 381 } 382 383 return (space); 384 } 385 386 struct mbuf * 387 getsombuf(struct socket *so, int type) 388 { 389 struct mbuf *m; 390 391 m = m_get(M_WAIT, type); 392 MCLAIM(m, so->so_mowner); 393 return m; 394 } 395 396 static int 397 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 398 void *arg0, void *arg1, void *arg2, void *arg3) 399 { 400 int result; 401 enum kauth_network_req req; 402 403 result = KAUTH_RESULT_DEFER; 404 req = (enum kauth_network_req)arg0; 405 406 if ((action != KAUTH_NETWORK_SOCKET) && 407 (action != KAUTH_NETWORK_BIND)) 408 return result; 409 410 switch (req) { 411 case KAUTH_REQ_NETWORK_BIND_PORT: 412 result = KAUTH_RESULT_ALLOW; 413 break; 414 415 case KAUTH_REQ_NETWORK_SOCKET_DROP: { 416 /* Normal users can only drop their own connections. */ 417 struct socket *so = (struct socket *)arg1; 418 419 if (proc_uidmatch(cred, so->so_cred)) 420 result = KAUTH_RESULT_ALLOW; 421 422 break; 423 } 424 425 case KAUTH_REQ_NETWORK_SOCKET_OPEN: 426 /* We allow "raw" routing/bluetooth sockets to anyone. */ 427 if ((u_long)arg1 == PF_ROUTE || (u_long)arg1 == PF_OROUTE 428 || (u_long)arg1 == PF_BLUETOOTH) { 429 result = KAUTH_RESULT_ALLOW; 430 } else { 431 /* Privileged, let secmodel handle this. */ 432 if ((u_long)arg2 == SOCK_RAW) 433 break; 434 } 435 436 result = KAUTH_RESULT_ALLOW; 437 438 break; 439 440 case KAUTH_REQ_NETWORK_SOCKET_CANSEE: 441 result = KAUTH_RESULT_ALLOW; 442 443 break; 444 445 default: 446 break; 447 } 448 449 return result; 450 } 451 452 void 453 soinit(void) 454 { 455 456 sysctl_kern_socket_setup(); 457 458 mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM); 459 softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 460 cv_init(&socurkva_cv, "sokva"); 461 cv_init(&pendfree_thread_cv, "sopendfr"); 462 soinit2(); 463 464 /* Set the initial adjusted socket buffer size. */ 465 if (sb_max_set(sb_max)) 466 panic("bad initial sb_max value: %lu", sb_max); 467 468 socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK, 469 socket_listener_cb, NULL); 470 } 471 472 void 473 soinit1(void) 474 { 475 int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, 476 sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree"); 477 if (error) 478 panic("soinit1 %d", error); 479 } 480 481 /* 482 * Socket operation routines. 483 * These routines are called by the routines in 484 * sys_socket.c or from a system process, and 485 * implement the semantics of socket operations by 486 * switching out to the protocol specific routines. 487 */ 488 /*ARGSUSED*/ 489 int 490 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l, 491 struct socket *lockso) 492 { 493 const struct protosw *prp; 494 struct socket *so; 495 uid_t uid; 496 int error; 497 kmutex_t *lock; 498 499 error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET, 500 KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type), 501 KAUTH_ARG(proto)); 502 if (error != 0) 503 return error; 504 505 if (proto) 506 prp = pffindproto(dom, proto, type); 507 else 508 prp = pffindtype(dom, type); 509 if (prp == NULL) { 510 /* no support for domain */ 511 if (pffinddomain(dom) == 0) 512 return EAFNOSUPPORT; 513 /* no support for socket type */ 514 if (proto == 0 && type != 0) 515 return EPROTOTYPE; 516 return EPROTONOSUPPORT; 517 } 518 if (prp->pr_usrreq == NULL) 519 return EPROTONOSUPPORT; 520 if (prp->pr_type != type) 521 return EPROTOTYPE; 522 523 so = soget(true); 524 so->so_type = type; 525 so->so_proto = prp; 526 so->so_send = sosend; 527 so->so_receive = soreceive; 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_cpid = l->l_proc->p_pid; 536 if (lockso != NULL) { 537 /* Caller wants us to share a lock. */ 538 lock = lockso->so_lock; 539 so->so_lock = lock; 540 mutex_obj_hold(lock); 541 /* XXX Why is this not solock, to match sounlock? */ 542 mutex_enter(lock); 543 } else { 544 /* Lock assigned and taken during PRU_ATTACH. */ 545 } 546 error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL, 547 (struct mbuf *)(long)proto, NULL, l); 548 KASSERT(solocked(so)); 549 if (error != 0) { 550 so->so_state |= SS_NOFDREF; 551 sofree(so); 552 return error; 553 } 554 so->so_cred = kauth_cred_dup(l->l_cred); 555 sounlock(so); 556 *aso = so; 557 return 0; 558 } 559 560 /* On success, write file descriptor to fdout and return zero. On 561 * failure, return non-zero; *fdout will be undefined. 562 */ 563 int 564 fsocreate(int domain, struct socket **sop, int type, int protocol, 565 struct lwp *l, int *fdout) 566 { 567 struct socket *so; 568 struct file *fp; 569 int fd, error; 570 int flags = type & SOCK_FLAGS_MASK; 571 572 type &= ~SOCK_FLAGS_MASK; 573 if ((error = fd_allocfile(&fp, &fd)) != 0) 574 return error; 575 fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0); 576 fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)| 577 ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0); 578 fp->f_type = DTYPE_SOCKET; 579 fp->f_ops = &socketops; 580 error = socreate(domain, &so, type, protocol, l, NULL); 581 if (error != 0) { 582 fd_abort(curproc, fp, fd); 583 } else { 584 if (sop != NULL) 585 *sop = so; 586 fp->f_data = so; 587 fd_affix(curproc, fp, fd); 588 *fdout = fd; 589 if (flags & SOCK_NONBLOCK) 590 so->so_state |= SS_NBIO; 591 } 592 return error; 593 } 594 595 int 596 sofamily(const struct socket *so) 597 { 598 const struct protosw *pr; 599 const struct domain *dom; 600 601 if ((pr = so->so_proto) == NULL) 602 return AF_UNSPEC; 603 if ((dom = pr->pr_domain) == NULL) 604 return AF_UNSPEC; 605 return dom->dom_family; 606 } 607 608 int 609 sobind(struct socket *so, struct mbuf *nam, struct lwp *l) 610 { 611 int error; 612 613 solock(so); 614 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, 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; 623 624 solock(so); 625 if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 626 SS_ISDISCONNECTING)) != 0) { 627 sounlock(so); 628 return (EINVAL); 629 } 630 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL, 631 NULL, NULL, l); 632 if (error != 0) { 633 sounlock(so); 634 return error; 635 } 636 if (TAILQ_EMPTY(&so->so_q)) 637 so->so_options |= SO_ACCEPTCONN; 638 if (backlog < 0) 639 backlog = 0; 640 so->so_qlimit = min(backlog, somaxconn); 641 sounlock(so); 642 return 0; 643 } 644 645 void 646 sofree(struct socket *so) 647 { 648 u_int refs; 649 650 KASSERT(solocked(so)); 651 652 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) { 653 sounlock(so); 654 return; 655 } 656 if (so->so_head) { 657 /* 658 * We must not decommission a socket that's on the accept(2) 659 * queue. If we do, then accept(2) may hang after select(2) 660 * indicated that the listening socket was ready. 661 */ 662 if (!soqremque(so, 0)) { 663 sounlock(so); 664 return; 665 } 666 } 667 if (so->so_rcv.sb_hiwat) 668 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 669 RLIM_INFINITY); 670 if (so->so_snd.sb_hiwat) 671 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 672 RLIM_INFINITY); 673 sbrelease(&so->so_snd, so); 674 KASSERT(!cv_has_waiters(&so->so_cv)); 675 KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv)); 676 KASSERT(!cv_has_waiters(&so->so_snd.sb_cv)); 677 sorflush(so); 678 refs = so->so_aborting; /* XXX */ 679 /* Remove acccept filter if one is present. */ 680 if (so->so_accf != NULL) 681 (void)accept_filt_clear(so); 682 sounlock(so); 683 if (refs == 0) /* XXX */ 684 soput(so); 685 } 686 687 /* 688 * Close a socket on last file table reference removal. 689 * Initiate disconnect if connected. 690 * Free socket when disconnect complete. 691 */ 692 int 693 soclose(struct socket *so) 694 { 695 struct socket *so2; 696 int error; 697 int error2; 698 699 error = 0; 700 solock(so); 701 if (so->so_options & SO_ACCEPTCONN) { 702 for (;;) { 703 if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 704 KASSERT(solocked2(so, so2)); 705 (void) soqremque(so2, 0); 706 /* soabort drops the lock. */ 707 (void) soabort(so2); 708 solock(so); 709 continue; 710 } 711 if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 712 KASSERT(solocked2(so, so2)); 713 (void) soqremque(so2, 1); 714 /* soabort drops the lock. */ 715 (void) soabort(so2); 716 solock(so); 717 continue; 718 } 719 break; 720 } 721 } 722 if (so->so_pcb == 0) 723 goto discard; 724 if (so->so_state & SS_ISCONNECTED) { 725 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 726 error = sodisconnect(so); 727 if (error) 728 goto drop; 729 } 730 if (so->so_options & SO_LINGER) { 731 if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) == 732 (SS_ISDISCONNECTING|SS_NBIO)) 733 goto drop; 734 while (so->so_state & SS_ISCONNECTED) { 735 error = sowait(so, true, so->so_linger * hz); 736 if (error) 737 break; 738 } 739 } 740 } 741 drop: 742 if (so->so_pcb) { 743 error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 744 NULL, NULL, NULL, NULL); 745 if (error == 0) 746 error = error2; 747 } 748 discard: 749 if (so->so_state & SS_NOFDREF) 750 panic("soclose: NOFDREF"); 751 kauth_cred_free(so->so_cred); 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_usrreq)(so, PRU_ABORT, NULL, 771 NULL, NULL, NULL); 772 refs = --so->so_aborting; /* XXX */ 773 if (error || (refs == 0)) { 774 sofree(so); 775 } else { 776 sounlock(so); 777 } 778 return error; 779 } 780 781 int 782 soaccept(struct socket *so, struct mbuf *nam) 783 { 784 int error; 785 786 KASSERT(solocked(so)); 787 788 error = 0; 789 if ((so->so_state & SS_NOFDREF) == 0) 790 panic("soaccept: !NOFDREF"); 791 so->so_state &= ~SS_NOFDREF; 792 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 793 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 794 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 795 NULL, nam, NULL, NULL); 796 else 797 error = ECONNABORTED; 798 799 return (error); 800 } 801 802 int 803 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l) 804 { 805 int error; 806 807 KASSERT(solocked(so)); 808 809 if (so->so_options & SO_ACCEPTCONN) 810 return (EOPNOTSUPP); 811 /* 812 * If protocol is connection-based, can only connect once. 813 * Otherwise, if connected, try to disconnect first. 814 * This allows user to disconnect by connecting to, e.g., 815 * a null address. 816 */ 817 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 818 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 819 (error = sodisconnect(so)))) 820 error = EISCONN; 821 else 822 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 823 NULL, nam, NULL, l); 824 return (error); 825 } 826 827 int 828 soconnect2(struct socket *so1, struct socket *so2) 829 { 830 int error; 831 832 KASSERT(solocked2(so1, so2)); 833 834 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 835 NULL, (struct mbuf *)so2, NULL, NULL); 836 return (error); 837 } 838 839 int 840 sodisconnect(struct socket *so) 841 { 842 int error; 843 844 KASSERT(solocked(so)); 845 846 if ((so->so_state & SS_ISCONNECTED) == 0) { 847 error = ENOTCONN; 848 } else if (so->so_state & SS_ISDISCONNECTING) { 849 error = EALREADY; 850 } else { 851 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 852 NULL, NULL, NULL, NULL); 853 } 854 return (error); 855 } 856 857 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 858 /* 859 * Send on a socket. 860 * If send must go all at once and message is larger than 861 * send buffering, then hard error. 862 * Lock against other senders. 863 * If must go all at once and not enough room now, then 864 * inform user that this would block and do nothing. 865 * Otherwise, if nonblocking, send as much as possible. 866 * The data to be sent is described by "uio" if nonzero, 867 * otherwise by the mbuf chain "top" (which must be null 868 * if uio is not). Data provided in mbuf chain must be small 869 * enough to send all at once. 870 * 871 * Returns nonzero on error, timeout or signal; callers 872 * must check for short counts if EINTR/ERESTART are returned. 873 * Data and control buffers are freed on return. 874 */ 875 int 876 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 877 struct mbuf *control, int flags, struct lwp *l) 878 { 879 struct mbuf **mp, *m; 880 struct proc *p; 881 long space, len, resid, clen, mlen; 882 int error, s, dontroute, atomic; 883 short wakeup_state = 0; 884 885 p = l->l_proc; 886 clen = 0; 887 888 /* 889 * solock() provides atomicity of access. splsoftnet() prevents 890 * protocol processing soft interrupts from interrupting us and 891 * blocking (expensive). 892 */ 893 s = splsoftnet(); 894 solock(so); 895 atomic = sosendallatonce(so) || top; 896 if (uio) 897 resid = uio->uio_resid; 898 else 899 resid = top->m_pkthdr.len; 900 /* 901 * In theory resid should be unsigned. 902 * However, space must be signed, as it might be less than 0 903 * if we over-committed, and we must use a signed comparison 904 * of space and resid. On the other hand, a negative resid 905 * causes us to loop sending 0-length segments to the protocol. 906 */ 907 if (resid < 0) { 908 error = EINVAL; 909 goto out; 910 } 911 dontroute = 912 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 913 (so->so_proto->pr_flags & PR_ATOMIC); 914 l->l_ru.ru_msgsnd++; 915 if (control) 916 clen = control->m_len; 917 restart: 918 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 919 goto out; 920 do { 921 if (so->so_state & SS_CANTSENDMORE) { 922 error = EPIPE; 923 goto release; 924 } 925 if (so->so_error) { 926 error = so->so_error; 927 so->so_error = 0; 928 goto release; 929 } 930 if ((so->so_state & SS_ISCONNECTED) == 0) { 931 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 932 if ((so->so_state & SS_ISCONFIRMING) == 0 && 933 !(resid == 0 && clen != 0)) { 934 error = ENOTCONN; 935 goto release; 936 } 937 } else if (addr == 0) { 938 error = EDESTADDRREQ; 939 goto release; 940 } 941 } 942 space = sbspace(&so->so_snd); 943 if (flags & MSG_OOB) 944 space += 1024; 945 if ((atomic && resid > so->so_snd.sb_hiwat) || 946 clen > so->so_snd.sb_hiwat) { 947 error = EMSGSIZE; 948 goto release; 949 } 950 if (space < resid + clen && 951 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 952 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) { 953 error = EWOULDBLOCK; 954 goto release; 955 } 956 sbunlock(&so->so_snd); 957 if (wakeup_state & SS_RESTARTSYS) { 958 error = ERESTART; 959 goto out; 960 } 961 error = sbwait(&so->so_snd); 962 if (error) 963 goto out; 964 wakeup_state = so->so_state; 965 goto restart; 966 } 967 wakeup_state = 0; 968 mp = ⊤ 969 space -= clen; 970 do { 971 if (uio == NULL) { 972 /* 973 * Data is prepackaged in "top". 974 */ 975 resid = 0; 976 if (flags & MSG_EOR) 977 top->m_flags |= M_EOR; 978 } else do { 979 sounlock(so); 980 splx(s); 981 if (top == NULL) { 982 m = m_gethdr(M_WAIT, MT_DATA); 983 mlen = MHLEN; 984 m->m_pkthdr.len = 0; 985 m->m_pkthdr.rcvif = NULL; 986 } else { 987 m = m_get(M_WAIT, MT_DATA); 988 mlen = MLEN; 989 } 990 MCLAIM(m, so->so_snd.sb_mowner); 991 if (sock_loan_thresh >= 0 && 992 uio->uio_iov->iov_len >= sock_loan_thresh && 993 space >= sock_loan_thresh && 994 (len = sosend_loan(so, uio, m, 995 space)) != 0) { 996 SOSEND_COUNTER_INCR(&sosend_loan_big); 997 space -= len; 998 goto have_data; 999 } 1000 if (resid >= MINCLSIZE && space >= MCLBYTES) { 1001 SOSEND_COUNTER_INCR(&sosend_copy_big); 1002 m_clget(m, M_DONTWAIT); 1003 if ((m->m_flags & M_EXT) == 0) 1004 goto nopages; 1005 mlen = MCLBYTES; 1006 if (atomic && top == 0) { 1007 len = lmin(MCLBYTES - max_hdr, 1008 resid); 1009 m->m_data += max_hdr; 1010 } else 1011 len = lmin(MCLBYTES, resid); 1012 space -= len; 1013 } else { 1014 nopages: 1015 SOSEND_COUNTER_INCR(&sosend_copy_small); 1016 len = lmin(lmin(mlen, resid), space); 1017 space -= len; 1018 /* 1019 * For datagram protocols, leave room 1020 * for protocol headers in first mbuf. 1021 */ 1022 if (atomic && top == 0 && len < mlen) 1023 MH_ALIGN(m, len); 1024 } 1025 error = uiomove(mtod(m, void *), (int)len, uio); 1026 have_data: 1027 resid = uio->uio_resid; 1028 m->m_len = len; 1029 *mp = m; 1030 top->m_pkthdr.len += len; 1031 s = splsoftnet(); 1032 solock(so); 1033 if (error != 0) 1034 goto release; 1035 mp = &m->m_next; 1036 if (resid <= 0) { 1037 if (flags & MSG_EOR) 1038 top->m_flags |= M_EOR; 1039 break; 1040 } 1041 } while (space > 0 && atomic); 1042 1043 if (so->so_state & SS_CANTSENDMORE) { 1044 error = EPIPE; 1045 goto release; 1046 } 1047 if (dontroute) 1048 so->so_options |= SO_DONTROUTE; 1049 if (resid > 0) 1050 so->so_state |= SS_MORETOCOME; 1051 error = (*so->so_proto->pr_usrreq)(so, 1052 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 1053 top, addr, control, curlwp); 1054 if (dontroute) 1055 so->so_options &= ~SO_DONTROUTE; 1056 if (resid > 0) 1057 so->so_state &= ~SS_MORETOCOME; 1058 clen = 0; 1059 control = NULL; 1060 top = NULL; 1061 mp = ⊤ 1062 if (error != 0) 1063 goto release; 1064 } while (resid && space > 0); 1065 } while (resid); 1066 1067 release: 1068 sbunlock(&so->so_snd); 1069 out: 1070 sounlock(so); 1071 splx(s); 1072 if (top) 1073 m_freem(top); 1074 if (control) 1075 m_freem(control); 1076 return (error); 1077 } 1078 1079 /* 1080 * Following replacement or removal of the first mbuf on the first 1081 * mbuf chain of a socket buffer, push necessary state changes back 1082 * into the socket buffer so that other consumers see the values 1083 * consistently. 'nextrecord' is the callers locally stored value of 1084 * the original value of sb->sb_mb->m_nextpkt which must be restored 1085 * when the lead mbuf changes. NOTE: 'nextrecord' may be NULL. 1086 */ 1087 static void 1088 sbsync(struct sockbuf *sb, struct mbuf *nextrecord) 1089 { 1090 1091 KASSERT(solocked(sb->sb_so)); 1092 1093 /* 1094 * First, update for the new value of nextrecord. If necessary, 1095 * make it the first record. 1096 */ 1097 if (sb->sb_mb != NULL) 1098 sb->sb_mb->m_nextpkt = nextrecord; 1099 else 1100 sb->sb_mb = nextrecord; 1101 1102 /* 1103 * Now update any dependent socket buffer fields to reflect 1104 * the new state. This is an inline of SB_EMPTY_FIXUP, with 1105 * the addition of a second clause that takes care of the 1106 * case where sb_mb has been updated, but remains the last 1107 * record. 1108 */ 1109 if (sb->sb_mb == NULL) { 1110 sb->sb_mbtail = NULL; 1111 sb->sb_lastrecord = NULL; 1112 } else if (sb->sb_mb->m_nextpkt == NULL) 1113 sb->sb_lastrecord = sb->sb_mb; 1114 } 1115 1116 /* 1117 * Implement receive operations on a socket. 1118 * We depend on the way that records are added to the sockbuf 1119 * by sbappend*. In particular, each record (mbufs linked through m_next) 1120 * must begin with an address if the protocol so specifies, 1121 * followed by an optional mbuf or mbufs containing ancillary data, 1122 * and then zero or more mbufs of data. 1123 * In order to avoid blocking network interrupts for the entire time here, 1124 * we splx() while doing the actual copy to user space. 1125 * Although the sockbuf is locked, new data may still be appended, 1126 * and thus we must maintain consistency of the sockbuf during that time. 1127 * 1128 * The caller may receive the data as a single mbuf chain by supplying 1129 * an mbuf **mp0 for use in returning the chain. The uio is then used 1130 * only for the count in uio_resid. 1131 */ 1132 int 1133 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 1134 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1135 { 1136 struct lwp *l = curlwp; 1137 struct mbuf *m, **mp, *mt; 1138 size_t len, offset, moff, orig_resid; 1139 int atomic, flags, error, s, type; 1140 const struct protosw *pr; 1141 struct mbuf *nextrecord; 1142 int mbuf_removed = 0; 1143 const struct domain *dom; 1144 short wakeup_state = 0; 1145 1146 pr = so->so_proto; 1147 atomic = pr->pr_flags & PR_ATOMIC; 1148 dom = pr->pr_domain; 1149 mp = mp0; 1150 type = 0; 1151 orig_resid = uio->uio_resid; 1152 1153 if (paddr != NULL) 1154 *paddr = NULL; 1155 if (controlp != NULL) 1156 *controlp = NULL; 1157 if (flagsp != NULL) 1158 flags = *flagsp &~ MSG_EOR; 1159 else 1160 flags = 0; 1161 1162 if (flags & MSG_OOB) { 1163 m = m_get(M_WAIT, MT_DATA); 1164 solock(so); 1165 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 1166 (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l); 1167 sounlock(so); 1168 if (error) 1169 goto bad; 1170 do { 1171 error = uiomove(mtod(m, void *), 1172 MIN(uio->uio_resid, m->m_len), uio); 1173 m = m_free(m); 1174 } while (uio->uio_resid > 0 && error == 0 && m); 1175 bad: 1176 if (m != NULL) 1177 m_freem(m); 1178 return error; 1179 } 1180 if (mp != NULL) 1181 *mp = NULL; 1182 1183 /* 1184 * solock() provides atomicity of access. splsoftnet() prevents 1185 * protocol processing soft interrupts from interrupting us and 1186 * blocking (expensive). 1187 */ 1188 s = splsoftnet(); 1189 solock(so); 1190 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 1191 (*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l); 1192 1193 restart: 1194 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) { 1195 sounlock(so); 1196 splx(s); 1197 return error; 1198 } 1199 1200 m = so->so_rcv.sb_mb; 1201 /* 1202 * If we have less data than requested, block awaiting more 1203 * (subject to any timeout) if: 1204 * 1. the current count is less than the low water mark, 1205 * 2. MSG_WAITALL is set, and it is possible to do the entire 1206 * receive operation at once if we block (resid <= hiwat), or 1207 * 3. MSG_DONTWAIT is not set. 1208 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1209 * we have to do the receive in sections, and thus risk returning 1210 * a short count if a timeout or signal occurs after we start. 1211 */ 1212 if (m == NULL || 1213 ((flags & MSG_DONTWAIT) == 0 && 1214 so->so_rcv.sb_cc < uio->uio_resid && 1215 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1216 ((flags & MSG_WAITALL) && 1217 uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1218 m->m_nextpkt == NULL && !atomic)) { 1219 #ifdef DIAGNOSTIC 1220 if (m == NULL && so->so_rcv.sb_cc) 1221 panic("receive 1"); 1222 #endif 1223 if (so->so_error) { 1224 if (m != NULL) 1225 goto dontblock; 1226 error = so->so_error; 1227 if ((flags & MSG_PEEK) == 0) 1228 so->so_error = 0; 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 dontblock: 1270 /* 1271 * On entry here, m points to the first record of the socket buffer. 1272 * From this point onward, we maintain 'nextrecord' as a cache of the 1273 * pointer to the next record in the socket buffer. We must keep the 1274 * various socket buffer pointers and local stack versions of the 1275 * pointers in sync, pushing out modifications before dropping the 1276 * socket lock, and re-reading them when picking it up. 1277 * 1278 * Otherwise, we will race with the network stack appending new data 1279 * or records onto the socket buffer by using inconsistent/stale 1280 * versions of the field, possibly resulting in socket buffer 1281 * corruption. 1282 * 1283 * By holding the high-level sblock(), we prevent simultaneous 1284 * readers from pulling off the front of the socket buffer. 1285 */ 1286 if (l != NULL) 1287 l->l_ru.ru_msgrcv++; 1288 KASSERT(m == so->so_rcv.sb_mb); 1289 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1290 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1291 nextrecord = m->m_nextpkt; 1292 if (pr->pr_flags & PR_ADDR) { 1293 #ifdef DIAGNOSTIC 1294 if (m->m_type != MT_SONAME) 1295 panic("receive 1a"); 1296 #endif 1297 orig_resid = 0; 1298 if (flags & MSG_PEEK) { 1299 if (paddr) 1300 *paddr = m_copy(m, 0, m->m_len); 1301 m = m->m_next; 1302 } else { 1303 sbfree(&so->so_rcv, m); 1304 mbuf_removed = 1; 1305 if (paddr != NULL) { 1306 *paddr = m; 1307 so->so_rcv.sb_mb = m->m_next; 1308 m->m_next = NULL; 1309 m = so->so_rcv.sb_mb; 1310 } else { 1311 MFREE(m, so->so_rcv.sb_mb); 1312 m = so->so_rcv.sb_mb; 1313 } 1314 sbsync(&so->so_rcv, nextrecord); 1315 } 1316 } 1317 1318 /* 1319 * Process one or more MT_CONTROL mbufs present before any data mbufs 1320 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 1321 * just copy the data; if !MSG_PEEK, we call into the protocol to 1322 * perform externalization (or freeing if controlp == NULL). 1323 */ 1324 if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) { 1325 struct mbuf *cm = NULL, *cmn; 1326 struct mbuf **cme = &cm; 1327 1328 do { 1329 if (flags & MSG_PEEK) { 1330 if (controlp != NULL) { 1331 *controlp = m_copy(m, 0, m->m_len); 1332 controlp = &(*controlp)->m_next; 1333 } 1334 m = m->m_next; 1335 } else { 1336 sbfree(&so->so_rcv, m); 1337 so->so_rcv.sb_mb = m->m_next; 1338 m->m_next = NULL; 1339 *cme = m; 1340 cme = &(*cme)->m_next; 1341 m = so->so_rcv.sb_mb; 1342 } 1343 } while (m != NULL && m->m_type == MT_CONTROL); 1344 if ((flags & MSG_PEEK) == 0) 1345 sbsync(&so->so_rcv, nextrecord); 1346 for (; cm != NULL; cm = cmn) { 1347 cmn = cm->m_next; 1348 cm->m_next = NULL; 1349 type = mtod(cm, struct cmsghdr *)->cmsg_type; 1350 if (controlp != NULL) { 1351 if (dom->dom_externalize != NULL && 1352 type == SCM_RIGHTS) { 1353 sounlock(so); 1354 splx(s); 1355 error = (*dom->dom_externalize)(cm, l, 1356 (flags & MSG_CMSG_CLOEXEC) ? 1357 O_CLOEXEC : 0); 1358 s = splsoftnet(); 1359 solock(so); 1360 } 1361 *controlp = cm; 1362 while (*controlp != NULL) 1363 controlp = &(*controlp)->m_next; 1364 } else { 1365 /* 1366 * Dispose of any SCM_RIGHTS message that went 1367 * through the read path rather than recv. 1368 */ 1369 if (dom->dom_dispose != NULL && 1370 type == SCM_RIGHTS) { 1371 sounlock(so); 1372 (*dom->dom_dispose)(cm); 1373 solock(so); 1374 } 1375 m_freem(cm); 1376 } 1377 } 1378 if (m != NULL) 1379 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 1380 else 1381 nextrecord = so->so_rcv.sb_mb; 1382 orig_resid = 0; 1383 } 1384 1385 /* If m is non-NULL, we have some data to read. */ 1386 if (__predict_true(m != NULL)) { 1387 type = m->m_type; 1388 if (type == MT_OOBDATA) 1389 flags |= MSG_OOB; 1390 } 1391 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1392 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1393 1394 moff = 0; 1395 offset = 0; 1396 while (m != NULL && uio->uio_resid > 0 && error == 0) { 1397 if (m->m_type == MT_OOBDATA) { 1398 if (type != MT_OOBDATA) 1399 break; 1400 } else if (type == MT_OOBDATA) 1401 break; 1402 #ifdef DIAGNOSTIC 1403 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1404 panic("receive 3"); 1405 #endif 1406 so->so_state &= ~SS_RCVATMARK; 1407 wakeup_state = 0; 1408 len = uio->uio_resid; 1409 if (so->so_oobmark && len > so->so_oobmark - offset) 1410 len = so->so_oobmark - offset; 1411 if (len > m->m_len - moff) 1412 len = m->m_len - moff; 1413 /* 1414 * If mp is set, just pass back the mbufs. 1415 * Otherwise copy them out via the uio, then free. 1416 * Sockbuf must be consistent here (points to current mbuf, 1417 * it points to next record) when we drop priority; 1418 * we must note any additions to the sockbuf when we 1419 * block interrupts again. 1420 */ 1421 if (mp == NULL) { 1422 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1423 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1424 sounlock(so); 1425 splx(s); 1426 error = uiomove(mtod(m, char *) + moff, len, uio); 1427 s = splsoftnet(); 1428 solock(so); 1429 if (error != 0) { 1430 /* 1431 * If any part of the record has been removed 1432 * (such as the MT_SONAME mbuf, which will 1433 * happen when PR_ADDR, and thus also 1434 * PR_ATOMIC, is set), then drop the entire 1435 * record to maintain the atomicity of the 1436 * receive operation. 1437 * 1438 * This avoids a later panic("receive 1a") 1439 * when compiled with DIAGNOSTIC. 1440 */ 1441 if (m && mbuf_removed && atomic) 1442 (void) sbdroprecord(&so->so_rcv); 1443 1444 goto release; 1445 } 1446 } else 1447 uio->uio_resid -= len; 1448 if (len == m->m_len - moff) { 1449 if (m->m_flags & M_EOR) 1450 flags |= MSG_EOR; 1451 if (flags & MSG_PEEK) { 1452 m = m->m_next; 1453 moff = 0; 1454 } else { 1455 nextrecord = m->m_nextpkt; 1456 sbfree(&so->so_rcv, m); 1457 if (mp) { 1458 *mp = m; 1459 mp = &m->m_next; 1460 so->so_rcv.sb_mb = m = m->m_next; 1461 *mp = NULL; 1462 } else { 1463 MFREE(m, so->so_rcv.sb_mb); 1464 m = so->so_rcv.sb_mb; 1465 } 1466 /* 1467 * If m != NULL, we also know that 1468 * so->so_rcv.sb_mb != NULL. 1469 */ 1470 KASSERT(so->so_rcv.sb_mb == m); 1471 if (m) { 1472 m->m_nextpkt = nextrecord; 1473 if (nextrecord == NULL) 1474 so->so_rcv.sb_lastrecord = m; 1475 } else { 1476 so->so_rcv.sb_mb = nextrecord; 1477 SB_EMPTY_FIXUP(&so->so_rcv); 1478 } 1479 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1480 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1481 } 1482 } else if (flags & MSG_PEEK) 1483 moff += len; 1484 else { 1485 if (mp != NULL) { 1486 mt = m_copym(m, 0, len, M_NOWAIT); 1487 if (__predict_false(mt == NULL)) { 1488 sounlock(so); 1489 mt = m_copym(m, 0, len, M_WAIT); 1490 solock(so); 1491 } 1492 *mp = mt; 1493 } 1494 m->m_data += len; 1495 m->m_len -= len; 1496 so->so_rcv.sb_cc -= len; 1497 } 1498 if (so->so_oobmark) { 1499 if ((flags & MSG_PEEK) == 0) { 1500 so->so_oobmark -= len; 1501 if (so->so_oobmark == 0) { 1502 so->so_state |= SS_RCVATMARK; 1503 break; 1504 } 1505 } else { 1506 offset += len; 1507 if (offset == so->so_oobmark) 1508 break; 1509 } 1510 } 1511 if (flags & MSG_EOR) 1512 break; 1513 /* 1514 * If the MSG_WAITALL flag is set (for non-atomic socket), 1515 * we must not quit until "uio->uio_resid == 0" or an error 1516 * termination. If a signal/timeout occurs, return 1517 * with a short count but without error. 1518 * Keep sockbuf locked against other readers. 1519 */ 1520 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 1521 !sosendallatonce(so) && !nextrecord) { 1522 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1523 break; 1524 /* 1525 * If we are peeking and the socket receive buffer is 1526 * full, stop since we can't get more data to peek at. 1527 */ 1528 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1529 break; 1530 /* 1531 * If we've drained the socket buffer, tell the 1532 * protocol in case it needs to do something to 1533 * get it filled again. 1534 */ 1535 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1536 (*pr->pr_usrreq)(so, PRU_RCVD, 1537 NULL, (struct mbuf *)(long)flags, NULL, l); 1538 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1539 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1540 if (wakeup_state & SS_RESTARTSYS) 1541 error = ERESTART; 1542 else 1543 error = sbwait(&so->so_rcv); 1544 if (error != 0) { 1545 sbunlock(&so->so_rcv); 1546 sounlock(so); 1547 splx(s); 1548 return 0; 1549 } 1550 if ((m = so->so_rcv.sb_mb) != NULL) 1551 nextrecord = m->m_nextpkt; 1552 wakeup_state = so->so_state; 1553 } 1554 } 1555 1556 if (m && atomic) { 1557 flags |= MSG_TRUNC; 1558 if ((flags & MSG_PEEK) == 0) 1559 (void) sbdroprecord(&so->so_rcv); 1560 } 1561 if ((flags & MSG_PEEK) == 0) { 1562 if (m == NULL) { 1563 /* 1564 * First part is an inline SB_EMPTY_FIXUP(). Second 1565 * part makes sure sb_lastrecord is up-to-date if 1566 * there is still data in the socket buffer. 1567 */ 1568 so->so_rcv.sb_mb = nextrecord; 1569 if (so->so_rcv.sb_mb == NULL) { 1570 so->so_rcv.sb_mbtail = NULL; 1571 so->so_rcv.sb_lastrecord = NULL; 1572 } else if (nextrecord->m_nextpkt == NULL) 1573 so->so_rcv.sb_lastrecord = nextrecord; 1574 } 1575 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1576 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1577 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1578 (*pr->pr_usrreq)(so, PRU_RCVD, NULL, 1579 (struct mbuf *)(long)flags, NULL, l); 1580 } 1581 if (orig_resid == uio->uio_resid && orig_resid && 1582 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1583 sbunlock(&so->so_rcv); 1584 goto restart; 1585 } 1586 1587 if (flagsp != NULL) 1588 *flagsp |= flags; 1589 release: 1590 sbunlock(&so->so_rcv); 1591 sounlock(so); 1592 splx(s); 1593 return error; 1594 } 1595 1596 int 1597 soshutdown(struct socket *so, int how) 1598 { 1599 const struct protosw *pr; 1600 int error; 1601 1602 KASSERT(solocked(so)); 1603 1604 pr = so->so_proto; 1605 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1606 return (EINVAL); 1607 1608 if (how == SHUT_RD || how == SHUT_RDWR) { 1609 sorflush(so); 1610 error = 0; 1611 } 1612 if (how == SHUT_WR || how == SHUT_RDWR) 1613 error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL, 1614 NULL, NULL, NULL); 1615 1616 return error; 1617 } 1618 1619 void 1620 sorestart(struct socket *so) 1621 { 1622 /* 1623 * An application has called close() on an fd on which another 1624 * of its threads has called a socket system call. 1625 * Mark this and wake everyone up, and code that would block again 1626 * instead returns ERESTART. 1627 * On system call re-entry the fd is validated and EBADF returned. 1628 * Any other fd will block again on the 2nd syscall. 1629 */ 1630 solock(so); 1631 so->so_state |= SS_RESTARTSYS; 1632 cv_broadcast(&so->so_cv); 1633 cv_broadcast(&so->so_snd.sb_cv); 1634 cv_broadcast(&so->so_rcv.sb_cv); 1635 sounlock(so); 1636 } 1637 1638 void 1639 sorflush(struct socket *so) 1640 { 1641 struct sockbuf *sb, asb; 1642 const struct protosw *pr; 1643 1644 KASSERT(solocked(so)); 1645 1646 sb = &so->so_rcv; 1647 pr = so->so_proto; 1648 socantrcvmore(so); 1649 sb->sb_flags |= SB_NOINTR; 1650 (void )sblock(sb, M_WAITOK); 1651 sbunlock(sb); 1652 asb = *sb; 1653 /* 1654 * Clear most of the sockbuf structure, but leave some of the 1655 * fields valid. 1656 */ 1657 memset(&sb->sb_startzero, 0, 1658 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1659 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) { 1660 sounlock(so); 1661 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1662 solock(so); 1663 } 1664 sbrelease(&asb, so); 1665 } 1666 1667 /* 1668 * internal set SOL_SOCKET options 1669 */ 1670 static int 1671 sosetopt1(struct socket *so, const struct sockopt *sopt) 1672 { 1673 int error = EINVAL, optval, opt; 1674 struct linger l; 1675 struct timeval tv; 1676 1677 switch ((opt = sopt->sopt_name)) { 1678 1679 case SO_ACCEPTFILTER: 1680 error = accept_filt_setopt(so, sopt); 1681 KASSERT(solocked(so)); 1682 break; 1683 1684 case SO_LINGER: 1685 error = sockopt_get(sopt, &l, sizeof(l)); 1686 solock(so); 1687 if (error) 1688 break; 1689 if (l.l_linger < 0 || l.l_linger > USHRT_MAX || 1690 l.l_linger > (INT_MAX / hz)) { 1691 error = EDOM; 1692 break; 1693 } 1694 so->so_linger = l.l_linger; 1695 if (l.l_onoff) 1696 so->so_options |= SO_LINGER; 1697 else 1698 so->so_options &= ~SO_LINGER; 1699 break; 1700 1701 case SO_DEBUG: 1702 case SO_KEEPALIVE: 1703 case SO_DONTROUTE: 1704 case SO_USELOOPBACK: 1705 case SO_BROADCAST: 1706 case SO_REUSEADDR: 1707 case SO_REUSEPORT: 1708 case SO_OOBINLINE: 1709 case SO_TIMESTAMP: 1710 case SO_NOSIGPIPE: 1711 #ifdef SO_OTIMESTAMP 1712 case SO_OTIMESTAMP: 1713 #endif 1714 error = sockopt_getint(sopt, &optval); 1715 solock(so); 1716 if (error) 1717 break; 1718 if (optval) 1719 so->so_options |= opt; 1720 else 1721 so->so_options &= ~opt; 1722 break; 1723 1724 case SO_SNDBUF: 1725 case SO_RCVBUF: 1726 case SO_SNDLOWAT: 1727 case SO_RCVLOWAT: 1728 error = sockopt_getint(sopt, &optval); 1729 solock(so); 1730 if (error) 1731 break; 1732 1733 /* 1734 * Values < 1 make no sense for any of these 1735 * options, so disallow them. 1736 */ 1737 if (optval < 1) { 1738 error = EINVAL; 1739 break; 1740 } 1741 1742 switch (opt) { 1743 case SO_SNDBUF: 1744 if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) { 1745 error = ENOBUFS; 1746 break; 1747 } 1748 so->so_snd.sb_flags &= ~SB_AUTOSIZE; 1749 break; 1750 1751 case SO_RCVBUF: 1752 if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) { 1753 error = ENOBUFS; 1754 break; 1755 } 1756 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1757 break; 1758 1759 /* 1760 * Make sure the low-water is never greater than 1761 * the high-water. 1762 */ 1763 case SO_SNDLOWAT: 1764 if (optval > so->so_snd.sb_hiwat) 1765 optval = so->so_snd.sb_hiwat; 1766 1767 so->so_snd.sb_lowat = optval; 1768 break; 1769 1770 case SO_RCVLOWAT: 1771 if (optval > so->so_rcv.sb_hiwat) 1772 optval = so->so_rcv.sb_hiwat; 1773 1774 so->so_rcv.sb_lowat = optval; 1775 break; 1776 } 1777 break; 1778 1779 #ifdef COMPAT_50 1780 case SO_OSNDTIMEO: 1781 case SO_ORCVTIMEO: { 1782 struct timeval50 otv; 1783 error = sockopt_get(sopt, &otv, sizeof(otv)); 1784 if (error) { 1785 solock(so); 1786 break; 1787 } 1788 timeval50_to_timeval(&otv, &tv); 1789 opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO; 1790 error = 0; 1791 /*FALLTHROUGH*/ 1792 } 1793 #endif /* COMPAT_50 */ 1794 1795 case SO_SNDTIMEO: 1796 case SO_RCVTIMEO: 1797 if (error) 1798 error = sockopt_get(sopt, &tv, sizeof(tv)); 1799 solock(so); 1800 if (error) 1801 break; 1802 1803 if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) { 1804 error = EDOM; 1805 break; 1806 } 1807 1808 optval = tv.tv_sec * hz + tv.tv_usec / tick; 1809 if (optval == 0 && tv.tv_usec != 0) 1810 optval = 1; 1811 1812 switch (opt) { 1813 case SO_SNDTIMEO: 1814 so->so_snd.sb_timeo = optval; 1815 break; 1816 case SO_RCVTIMEO: 1817 so->so_rcv.sb_timeo = optval; 1818 break; 1819 } 1820 break; 1821 1822 default: 1823 solock(so); 1824 error = ENOPROTOOPT; 1825 break; 1826 } 1827 KASSERT(solocked(so)); 1828 return error; 1829 } 1830 1831 int 1832 sosetopt(struct socket *so, struct sockopt *sopt) 1833 { 1834 int error, prerr; 1835 1836 if (sopt->sopt_level == SOL_SOCKET) { 1837 error = sosetopt1(so, sopt); 1838 KASSERT(solocked(so)); 1839 } else { 1840 error = ENOPROTOOPT; 1841 solock(so); 1842 } 1843 1844 if ((error == 0 || error == ENOPROTOOPT) && 1845 so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) { 1846 /* give the protocol stack a shot */ 1847 prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt); 1848 if (prerr == 0) 1849 error = 0; 1850 else if (prerr != ENOPROTOOPT) 1851 error = prerr; 1852 } 1853 sounlock(so); 1854 return error; 1855 } 1856 1857 /* 1858 * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt() 1859 */ 1860 int 1861 so_setsockopt(struct lwp *l, struct socket *so, int level, int name, 1862 const void *val, size_t valsize) 1863 { 1864 struct sockopt sopt; 1865 int error; 1866 1867 KASSERT(valsize == 0 || val != NULL); 1868 1869 sockopt_init(&sopt, level, name, valsize); 1870 sockopt_set(&sopt, val, valsize); 1871 1872 error = sosetopt(so, &sopt); 1873 1874 sockopt_destroy(&sopt); 1875 1876 return error; 1877 } 1878 1879 /* 1880 * internal get SOL_SOCKET options 1881 */ 1882 static int 1883 sogetopt1(struct socket *so, struct sockopt *sopt) 1884 { 1885 int error, optval, opt; 1886 struct linger l; 1887 struct timeval tv; 1888 1889 switch ((opt = sopt->sopt_name)) { 1890 1891 case SO_ACCEPTFILTER: 1892 error = accept_filt_getopt(so, sopt); 1893 break; 1894 1895 case SO_LINGER: 1896 l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0; 1897 l.l_linger = so->so_linger; 1898 1899 error = sockopt_set(sopt, &l, sizeof(l)); 1900 break; 1901 1902 case SO_USELOOPBACK: 1903 case SO_DONTROUTE: 1904 case SO_DEBUG: 1905 case SO_KEEPALIVE: 1906 case SO_REUSEADDR: 1907 case SO_REUSEPORT: 1908 case SO_BROADCAST: 1909 case SO_OOBINLINE: 1910 case SO_TIMESTAMP: 1911 case SO_NOSIGPIPE: 1912 #ifdef SO_OTIMESTAMP 1913 case SO_OTIMESTAMP: 1914 #endif 1915 error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0); 1916 break; 1917 1918 case SO_TYPE: 1919 error = sockopt_setint(sopt, so->so_type); 1920 break; 1921 1922 case SO_ERROR: 1923 error = sockopt_setint(sopt, so->so_error); 1924 so->so_error = 0; 1925 break; 1926 1927 case SO_SNDBUF: 1928 error = sockopt_setint(sopt, so->so_snd.sb_hiwat); 1929 break; 1930 1931 case SO_RCVBUF: 1932 error = sockopt_setint(sopt, so->so_rcv.sb_hiwat); 1933 break; 1934 1935 case SO_SNDLOWAT: 1936 error = sockopt_setint(sopt, so->so_snd.sb_lowat); 1937 break; 1938 1939 case SO_RCVLOWAT: 1940 error = sockopt_setint(sopt, so->so_rcv.sb_lowat); 1941 break; 1942 1943 #ifdef COMPAT_50 1944 case SO_OSNDTIMEO: 1945 case SO_ORCVTIMEO: { 1946 struct timeval50 otv; 1947 1948 optval = (opt == SO_OSNDTIMEO ? 1949 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1950 1951 otv.tv_sec = optval / hz; 1952 otv.tv_usec = (optval % hz) * tick; 1953 1954 error = sockopt_set(sopt, &otv, sizeof(otv)); 1955 break; 1956 } 1957 #endif /* COMPAT_50 */ 1958 1959 case SO_SNDTIMEO: 1960 case SO_RCVTIMEO: 1961 optval = (opt == SO_SNDTIMEO ? 1962 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1963 1964 tv.tv_sec = optval / hz; 1965 tv.tv_usec = (optval % hz) * tick; 1966 1967 error = sockopt_set(sopt, &tv, sizeof(tv)); 1968 break; 1969 1970 case SO_OVERFLOWED: 1971 error = sockopt_setint(sopt, so->so_rcv.sb_overflowed); 1972 break; 1973 1974 default: 1975 error = ENOPROTOOPT; 1976 break; 1977 } 1978 1979 return (error); 1980 } 1981 1982 int 1983 sogetopt(struct socket *so, struct sockopt *sopt) 1984 { 1985 int error; 1986 1987 solock(so); 1988 if (sopt->sopt_level != SOL_SOCKET) { 1989 if (so->so_proto && so->so_proto->pr_ctloutput) { 1990 error = ((*so->so_proto->pr_ctloutput) 1991 (PRCO_GETOPT, so, sopt)); 1992 } else 1993 error = (ENOPROTOOPT); 1994 } else { 1995 error = sogetopt1(so, sopt); 1996 } 1997 sounlock(so); 1998 return (error); 1999 } 2000 2001 /* 2002 * alloc sockopt data buffer buffer 2003 * - will be released at destroy 2004 */ 2005 static int 2006 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag) 2007 { 2008 2009 KASSERT(sopt->sopt_size == 0); 2010 2011 if (len > sizeof(sopt->sopt_buf)) { 2012 sopt->sopt_data = kmem_zalloc(len, kmflag); 2013 if (sopt->sopt_data == NULL) 2014 return ENOMEM; 2015 } else 2016 sopt->sopt_data = sopt->sopt_buf; 2017 2018 sopt->sopt_size = len; 2019 return 0; 2020 } 2021 2022 /* 2023 * initialise sockopt storage 2024 * - MAY sleep during allocation 2025 */ 2026 void 2027 sockopt_init(struct sockopt *sopt, int level, int name, size_t size) 2028 { 2029 2030 memset(sopt, 0, sizeof(*sopt)); 2031 2032 sopt->sopt_level = level; 2033 sopt->sopt_name = name; 2034 (void)sockopt_alloc(sopt, size, KM_SLEEP); 2035 } 2036 2037 /* 2038 * destroy sockopt storage 2039 * - will release any held memory references 2040 */ 2041 void 2042 sockopt_destroy(struct sockopt *sopt) 2043 { 2044 2045 if (sopt->sopt_data != sopt->sopt_buf) 2046 kmem_free(sopt->sopt_data, sopt->sopt_size); 2047 2048 memset(sopt, 0, sizeof(*sopt)); 2049 } 2050 2051 /* 2052 * set sockopt value 2053 * - value is copied into sockopt 2054 * - memory is allocated when necessary, will not sleep 2055 */ 2056 int 2057 sockopt_set(struct sockopt *sopt, const void *buf, size_t len) 2058 { 2059 int error; 2060 2061 if (sopt->sopt_size == 0) { 2062 error = sockopt_alloc(sopt, len, KM_NOSLEEP); 2063 if (error) 2064 return error; 2065 } 2066 2067 KASSERT(sopt->sopt_size == len); 2068 memcpy(sopt->sopt_data, buf, len); 2069 return 0; 2070 } 2071 2072 /* 2073 * common case of set sockopt integer value 2074 */ 2075 int 2076 sockopt_setint(struct sockopt *sopt, int val) 2077 { 2078 2079 return sockopt_set(sopt, &val, sizeof(int)); 2080 } 2081 2082 /* 2083 * get sockopt value 2084 * - correct size must be given 2085 */ 2086 int 2087 sockopt_get(const struct sockopt *sopt, void *buf, size_t len) 2088 { 2089 2090 if (sopt->sopt_size != len) 2091 return EINVAL; 2092 2093 memcpy(buf, sopt->sopt_data, len); 2094 return 0; 2095 } 2096 2097 /* 2098 * common case of get sockopt integer value 2099 */ 2100 int 2101 sockopt_getint(const struct sockopt *sopt, int *valp) 2102 { 2103 2104 return sockopt_get(sopt, valp, sizeof(int)); 2105 } 2106 2107 /* 2108 * set sockopt value from mbuf 2109 * - ONLY for legacy code 2110 * - mbuf is released by sockopt 2111 * - will not sleep 2112 */ 2113 int 2114 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m) 2115 { 2116 size_t len; 2117 int error; 2118 2119 len = m_length(m); 2120 2121 if (sopt->sopt_size == 0) { 2122 error = sockopt_alloc(sopt, len, KM_NOSLEEP); 2123 if (error) 2124 return error; 2125 } 2126 2127 KASSERT(sopt->sopt_size == len); 2128 m_copydata(m, 0, len, sopt->sopt_data); 2129 m_freem(m); 2130 2131 return 0; 2132 } 2133 2134 /* 2135 * get sockopt value into mbuf 2136 * - ONLY for legacy code 2137 * - mbuf to be released by the caller 2138 * - will not sleep 2139 */ 2140 struct mbuf * 2141 sockopt_getmbuf(const struct sockopt *sopt) 2142 { 2143 struct mbuf *m; 2144 2145 if (sopt->sopt_size > MCLBYTES) 2146 return NULL; 2147 2148 m = m_get(M_DONTWAIT, MT_SOOPTS); 2149 if (m == NULL) 2150 return NULL; 2151 2152 if (sopt->sopt_size > MLEN) { 2153 MCLGET(m, M_DONTWAIT); 2154 if ((m->m_flags & M_EXT) == 0) { 2155 m_free(m); 2156 return NULL; 2157 } 2158 } 2159 2160 memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size); 2161 m->m_len = sopt->sopt_size; 2162 2163 return m; 2164 } 2165 2166 void 2167 sohasoutofband(struct socket *so) 2168 { 2169 2170 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 2171 selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT); 2172 } 2173 2174 static void 2175 filt_sordetach(struct knote *kn) 2176 { 2177 struct socket *so; 2178 2179 so = ((file_t *)kn->kn_obj)->f_data; 2180 solock(so); 2181 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 2182 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 2183 so->so_rcv.sb_flags &= ~SB_KNOTE; 2184 sounlock(so); 2185 } 2186 2187 /*ARGSUSED*/ 2188 static int 2189 filt_soread(struct knote *kn, long hint) 2190 { 2191 struct socket *so; 2192 int rv; 2193 2194 so = ((file_t *)kn->kn_obj)->f_data; 2195 if (hint != NOTE_SUBMIT) 2196 solock(so); 2197 kn->kn_data = so->so_rcv.sb_cc; 2198 if (so->so_state & SS_CANTRCVMORE) { 2199 kn->kn_flags |= EV_EOF; 2200 kn->kn_fflags = so->so_error; 2201 rv = 1; 2202 } else if (so->so_error) /* temporary udp error */ 2203 rv = 1; 2204 else if (kn->kn_sfflags & NOTE_LOWAT) 2205 rv = (kn->kn_data >= kn->kn_sdata); 2206 else 2207 rv = (kn->kn_data >= so->so_rcv.sb_lowat); 2208 if (hint != NOTE_SUBMIT) 2209 sounlock(so); 2210 return rv; 2211 } 2212 2213 static void 2214 filt_sowdetach(struct knote *kn) 2215 { 2216 struct socket *so; 2217 2218 so = ((file_t *)kn->kn_obj)->f_data; 2219 solock(so); 2220 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 2221 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 2222 so->so_snd.sb_flags &= ~SB_KNOTE; 2223 sounlock(so); 2224 } 2225 2226 /*ARGSUSED*/ 2227 static int 2228 filt_sowrite(struct knote *kn, long hint) 2229 { 2230 struct socket *so; 2231 int rv; 2232 2233 so = ((file_t *)kn->kn_obj)->f_data; 2234 if (hint != NOTE_SUBMIT) 2235 solock(so); 2236 kn->kn_data = sbspace(&so->so_snd); 2237 if (so->so_state & SS_CANTSENDMORE) { 2238 kn->kn_flags |= EV_EOF; 2239 kn->kn_fflags = so->so_error; 2240 rv = 1; 2241 } else if (so->so_error) /* temporary udp error */ 2242 rv = 1; 2243 else if (((so->so_state & SS_ISCONNECTED) == 0) && 2244 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 2245 rv = 0; 2246 else if (kn->kn_sfflags & NOTE_LOWAT) 2247 rv = (kn->kn_data >= kn->kn_sdata); 2248 else 2249 rv = (kn->kn_data >= so->so_snd.sb_lowat); 2250 if (hint != NOTE_SUBMIT) 2251 sounlock(so); 2252 return rv; 2253 } 2254 2255 /*ARGSUSED*/ 2256 static int 2257 filt_solisten(struct knote *kn, long hint) 2258 { 2259 struct socket *so; 2260 int rv; 2261 2262 so = ((file_t *)kn->kn_obj)->f_data; 2263 2264 /* 2265 * Set kn_data to number of incoming connections, not 2266 * counting partial (incomplete) connections. 2267 */ 2268 if (hint != NOTE_SUBMIT) 2269 solock(so); 2270 kn->kn_data = so->so_qlen; 2271 rv = (kn->kn_data > 0); 2272 if (hint != NOTE_SUBMIT) 2273 sounlock(so); 2274 return rv; 2275 } 2276 2277 static const struct filterops solisten_filtops = 2278 { 1, NULL, filt_sordetach, filt_solisten }; 2279 static const struct filterops soread_filtops = 2280 { 1, NULL, filt_sordetach, filt_soread }; 2281 static const struct filterops sowrite_filtops = 2282 { 1, NULL, filt_sowdetach, filt_sowrite }; 2283 2284 int 2285 soo_kqfilter(struct file *fp, struct knote *kn) 2286 { 2287 struct socket *so; 2288 struct sockbuf *sb; 2289 2290 so = ((file_t *)kn->kn_obj)->f_data; 2291 solock(so); 2292 switch (kn->kn_filter) { 2293 case EVFILT_READ: 2294 if (so->so_options & SO_ACCEPTCONN) 2295 kn->kn_fop = &solisten_filtops; 2296 else 2297 kn->kn_fop = &soread_filtops; 2298 sb = &so->so_rcv; 2299 break; 2300 case EVFILT_WRITE: 2301 kn->kn_fop = &sowrite_filtops; 2302 sb = &so->so_snd; 2303 break; 2304 default: 2305 sounlock(so); 2306 return (EINVAL); 2307 } 2308 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 2309 sb->sb_flags |= SB_KNOTE; 2310 sounlock(so); 2311 return (0); 2312 } 2313 2314 static int 2315 sodopoll(struct socket *so, int events) 2316 { 2317 int revents; 2318 2319 revents = 0; 2320 2321 if (events & (POLLIN | POLLRDNORM)) 2322 if (soreadable(so)) 2323 revents |= events & (POLLIN | POLLRDNORM); 2324 2325 if (events & (POLLOUT | POLLWRNORM)) 2326 if (sowritable(so)) 2327 revents |= events & (POLLOUT | POLLWRNORM); 2328 2329 if (events & (POLLPRI | POLLRDBAND)) 2330 if (so->so_oobmark || (so->so_state & SS_RCVATMARK)) 2331 revents |= events & (POLLPRI | POLLRDBAND); 2332 2333 return revents; 2334 } 2335 2336 int 2337 sopoll(struct socket *so, int events) 2338 { 2339 int revents = 0; 2340 2341 #ifndef DIAGNOSTIC 2342 /* 2343 * Do a quick, unlocked check in expectation that the socket 2344 * will be ready for I/O. Don't do this check if DIAGNOSTIC, 2345 * as the solocked() assertions will fail. 2346 */ 2347 if ((revents = sodopoll(so, events)) != 0) 2348 return revents; 2349 #endif 2350 2351 solock(so); 2352 if ((revents = sodopoll(so, events)) == 0) { 2353 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 2354 selrecord(curlwp, &so->so_rcv.sb_sel); 2355 so->so_rcv.sb_flags |= SB_NOTIFY; 2356 } 2357 2358 if (events & (POLLOUT | POLLWRNORM)) { 2359 selrecord(curlwp, &so->so_snd.sb_sel); 2360 so->so_snd.sb_flags |= SB_NOTIFY; 2361 } 2362 } 2363 sounlock(so); 2364 2365 return revents; 2366 } 2367 2368 2369 #include <sys/sysctl.h> 2370 2371 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 2372 static int sysctl_kern_sbmax(SYSCTLFN_PROTO); 2373 2374 /* 2375 * sysctl helper routine for kern.somaxkva. ensures that the given 2376 * value is not too small. 2377 * (XXX should we maybe make sure it's not too large as well?) 2378 */ 2379 static int 2380 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 2381 { 2382 int error, new_somaxkva; 2383 struct sysctlnode node; 2384 2385 new_somaxkva = somaxkva; 2386 node = *rnode; 2387 node.sysctl_data = &new_somaxkva; 2388 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2389 if (error || newp == NULL) 2390 return (error); 2391 2392 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 2393 return (EINVAL); 2394 2395 mutex_enter(&so_pendfree_lock); 2396 somaxkva = new_somaxkva; 2397 cv_broadcast(&socurkva_cv); 2398 mutex_exit(&so_pendfree_lock); 2399 2400 return (error); 2401 } 2402 2403 /* 2404 * sysctl helper routine for kern.sbmax. Basically just ensures that 2405 * any new value is not too small. 2406 */ 2407 static int 2408 sysctl_kern_sbmax(SYSCTLFN_ARGS) 2409 { 2410 int error, new_sbmax; 2411 struct sysctlnode node; 2412 2413 new_sbmax = sb_max; 2414 node = *rnode; 2415 node.sysctl_data = &new_sbmax; 2416 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2417 if (error || newp == NULL) 2418 return (error); 2419 2420 KERNEL_LOCK(1, NULL); 2421 error = sb_max_set(new_sbmax); 2422 KERNEL_UNLOCK_ONE(NULL); 2423 2424 return (error); 2425 } 2426 2427 static void 2428 sysctl_kern_socket_setup(void) 2429 { 2430 2431 KASSERT(socket_sysctllog == NULL); 2432 sysctl_createv(&socket_sysctllog, 0, NULL, NULL, 2433 CTLFLAG_PERMANENT, 2434 CTLTYPE_NODE, "kern", NULL, 2435 NULL, 0, NULL, 0, 2436 CTL_KERN, CTL_EOL); 2437 2438 sysctl_createv(&socket_sysctllog, 0, NULL, NULL, 2439 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2440 CTLTYPE_INT, "somaxkva", 2441 SYSCTL_DESCR("Maximum amount of kernel memory to be " 2442 "used for socket buffers"), 2443 sysctl_kern_somaxkva, 0, NULL, 0, 2444 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 2445 2446 sysctl_createv(&socket_sysctllog, 0, NULL, NULL, 2447 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2448 CTLTYPE_INT, "sbmax", 2449 SYSCTL_DESCR("Maximum socket buffer size"), 2450 sysctl_kern_sbmax, 0, NULL, 0, 2451 CTL_KERN, KERN_SBMAX, CTL_EOL); 2452 } 2453