1 /* $NetBSD: uipc_socket.c,v 1.187 2009/03/15 17:14:40 cegger Exp $ */ 2 3 /*- 4 * Copyright (c) 2002, 2007, 2008 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. 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 1.187 2009/03/15 17:14:40 cegger 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 96 #ifdef COMPAT_50 97 #include <compat/sys/time.h> 98 #include <compat/sys/socket.h> 99 #endif 100 101 #include <uvm/uvm.h> 102 103 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options"); 104 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 105 106 extern const struct fileops socketops; 107 108 extern int somaxconn; /* patchable (XXX sysctl) */ 109 int somaxconn = SOMAXCONN; 110 kmutex_t *softnet_lock; 111 112 #ifdef SOSEND_COUNTERS 113 #include <sys/device.h> 114 115 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 116 NULL, "sosend", "loan big"); 117 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 118 NULL, "sosend", "copy big"); 119 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 120 NULL, "sosend", "copy small"); 121 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, 122 NULL, "sosend", "kva limit"); 123 124 #define SOSEND_COUNTER_INCR(ev) (ev)->ev_count++ 125 126 EVCNT_ATTACH_STATIC(sosend_loan_big); 127 EVCNT_ATTACH_STATIC(sosend_copy_big); 128 EVCNT_ATTACH_STATIC(sosend_copy_small); 129 EVCNT_ATTACH_STATIC(sosend_kvalimit); 130 #else 131 132 #define SOSEND_COUNTER_INCR(ev) /* nothing */ 133 134 #endif /* SOSEND_COUNTERS */ 135 136 static struct callback_entry sokva_reclaimerentry; 137 138 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR) 139 int sock_loan_thresh = -1; 140 #else 141 int sock_loan_thresh = 4096; 142 #endif 143 144 static kmutex_t so_pendfree_lock; 145 static struct mbuf *so_pendfree; 146 147 #ifndef SOMAXKVA 148 #define SOMAXKVA (16 * 1024 * 1024) 149 #endif 150 int somaxkva = SOMAXKVA; 151 static int socurkva; 152 static kcondvar_t socurkva_cv; 153 154 #define SOCK_LOAN_CHUNK 65536 155 156 static size_t sodopendfree(void); 157 static size_t sodopendfreel(void); 158 159 static void sysctl_kern_somaxkva_setup(void); 160 static struct sysctllog *socket_sysctllog; 161 162 static vsize_t 163 sokvareserve(struct socket *so, vsize_t len) 164 { 165 int error; 166 167 mutex_enter(&so_pendfree_lock); 168 while (socurkva + len > somaxkva) { 169 size_t freed; 170 171 /* 172 * try to do pendfree. 173 */ 174 175 freed = sodopendfreel(); 176 177 /* 178 * if some kva was freed, try again. 179 */ 180 181 if (freed) 182 continue; 183 184 SOSEND_COUNTER_INCR(&sosend_kvalimit); 185 error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock); 186 if (error) { 187 len = 0; 188 break; 189 } 190 } 191 socurkva += len; 192 mutex_exit(&so_pendfree_lock); 193 return len; 194 } 195 196 static void 197 sokvaunreserve(vsize_t len) 198 { 199 200 mutex_enter(&so_pendfree_lock); 201 socurkva -= len; 202 cv_broadcast(&socurkva_cv); 203 mutex_exit(&so_pendfree_lock); 204 } 205 206 /* 207 * sokvaalloc: allocate kva for loan. 208 */ 209 210 vaddr_t 211 sokvaalloc(vsize_t len, struct socket *so) 212 { 213 vaddr_t lva; 214 215 /* 216 * reserve kva. 217 */ 218 219 if (sokvareserve(so, len) == 0) 220 return 0; 221 222 /* 223 * allocate kva. 224 */ 225 226 lva = uvm_km_alloc(kernel_map, len, 0, UVM_KMF_VAONLY | UVM_KMF_WAITVA); 227 if (lva == 0) { 228 sokvaunreserve(len); 229 return (0); 230 } 231 232 return lva; 233 } 234 235 /* 236 * sokvafree: free kva for loan. 237 */ 238 239 void 240 sokvafree(vaddr_t sva, vsize_t len) 241 { 242 243 /* 244 * free kva. 245 */ 246 247 uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY); 248 249 /* 250 * unreserve kva. 251 */ 252 253 sokvaunreserve(len); 254 } 255 256 static void 257 sodoloanfree(struct vm_page **pgs, void *buf, size_t size) 258 { 259 vaddr_t sva, eva; 260 vsize_t len; 261 int npgs; 262 263 KASSERT(pgs != NULL); 264 265 eva = round_page((vaddr_t) buf + size); 266 sva = trunc_page((vaddr_t) buf); 267 len = eva - sva; 268 npgs = len >> PAGE_SHIFT; 269 270 pmap_kremove(sva, len); 271 pmap_update(pmap_kernel()); 272 uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE); 273 sokvafree(sva, len); 274 } 275 276 static size_t 277 sodopendfree(void) 278 { 279 size_t rv; 280 281 if (__predict_true(so_pendfree == NULL)) 282 return 0; 283 284 mutex_enter(&so_pendfree_lock); 285 rv = sodopendfreel(); 286 mutex_exit(&so_pendfree_lock); 287 288 return rv; 289 } 290 291 /* 292 * sodopendfreel: free mbufs on "pendfree" list. 293 * unlock and relock so_pendfree_lock when freeing mbufs. 294 * 295 * => called with so_pendfree_lock held. 296 */ 297 298 static size_t 299 sodopendfreel(void) 300 { 301 struct mbuf *m, *next; 302 size_t rv = 0; 303 304 KASSERT(mutex_owned(&so_pendfree_lock)); 305 306 while (so_pendfree != NULL) { 307 m = so_pendfree; 308 so_pendfree = NULL; 309 mutex_exit(&so_pendfree_lock); 310 311 for (; m != NULL; m = next) { 312 next = m->m_next; 313 KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) == 0); 314 KASSERT(m->m_ext.ext_refcnt == 0); 315 316 rv += m->m_ext.ext_size; 317 sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf, 318 m->m_ext.ext_size); 319 pool_cache_put(mb_cache, m); 320 } 321 322 mutex_enter(&so_pendfree_lock); 323 } 324 325 return (rv); 326 } 327 328 void 329 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg) 330 { 331 332 KASSERT(m != NULL); 333 334 /* 335 * postpone freeing mbuf. 336 * 337 * we can't do it in interrupt context 338 * because we need to put kva back to kernel_map. 339 */ 340 341 mutex_enter(&so_pendfree_lock); 342 m->m_next = so_pendfree; 343 so_pendfree = m; 344 cv_broadcast(&socurkva_cv); 345 mutex_exit(&so_pendfree_lock); 346 } 347 348 static long 349 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space) 350 { 351 struct iovec *iov = uio->uio_iov; 352 vaddr_t sva, eva; 353 vsize_t len; 354 vaddr_t lva; 355 int npgs, error; 356 vaddr_t va; 357 int i; 358 359 if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace)) 360 return (0); 361 362 if (iov->iov_len < (size_t) space) 363 space = iov->iov_len; 364 if (space > SOCK_LOAN_CHUNK) 365 space = SOCK_LOAN_CHUNK; 366 367 eva = round_page((vaddr_t) iov->iov_base + space); 368 sva = trunc_page((vaddr_t) iov->iov_base); 369 len = eva - sva; 370 npgs = len >> PAGE_SHIFT; 371 372 KASSERT(npgs <= M_EXT_MAXPAGES); 373 374 lva = sokvaalloc(len, so); 375 if (lva == 0) 376 return 0; 377 378 error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len, 379 m->m_ext.ext_pgs, UVM_LOAN_TOPAGE); 380 if (error) { 381 sokvafree(lva, len); 382 return (0); 383 } 384 385 for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE) 386 pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]), 387 VM_PROT_READ); 388 pmap_update(pmap_kernel()); 389 390 lva += (vaddr_t) iov->iov_base & PAGE_MASK; 391 392 MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so); 393 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP; 394 395 uio->uio_resid -= space; 396 /* uio_offset not updated, not set/used for write(2) */ 397 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space; 398 uio->uio_iov->iov_len -= space; 399 if (uio->uio_iov->iov_len == 0) { 400 uio->uio_iov++; 401 uio->uio_iovcnt--; 402 } 403 404 return (space); 405 } 406 407 static int 408 sokva_reclaim_callback(struct callback_entry *ce, void *obj, void *arg) 409 { 410 411 KASSERT(ce == &sokva_reclaimerentry); 412 KASSERT(obj == NULL); 413 414 sodopendfree(); 415 if (!vm_map_starved_p(kernel_map)) { 416 return CALLBACK_CHAIN_ABORT; 417 } 418 return CALLBACK_CHAIN_CONTINUE; 419 } 420 421 struct mbuf * 422 getsombuf(struct socket *so, int type) 423 { 424 struct mbuf *m; 425 426 m = m_get(M_WAIT, type); 427 MCLAIM(m, so->so_mowner); 428 return m; 429 } 430 431 void 432 soinit(void) 433 { 434 435 sysctl_kern_somaxkva_setup(); 436 437 mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM); 438 softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 439 cv_init(&socurkva_cv, "sokva"); 440 soinit2(); 441 442 /* Set the initial adjusted socket buffer size. */ 443 if (sb_max_set(sb_max)) 444 panic("bad initial sb_max value: %lu", sb_max); 445 446 callback_register(&vm_map_to_kernel(kernel_map)->vmk_reclaim_callback, 447 &sokva_reclaimerentry, NULL, sokva_reclaim_callback); 448 } 449 450 /* 451 * Socket operation routines. 452 * These routines are called by the routines in 453 * sys_socket.c or from a system process, and 454 * implement the semantics of socket operations by 455 * switching out to the protocol specific routines. 456 */ 457 /*ARGSUSED*/ 458 int 459 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l, 460 struct socket *lockso) 461 { 462 const struct protosw *prp; 463 struct socket *so; 464 uid_t uid; 465 int error; 466 kmutex_t *lock; 467 468 error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET, 469 KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type), 470 KAUTH_ARG(proto)); 471 if (error != 0) 472 return error; 473 474 if (proto) 475 prp = pffindproto(dom, proto, type); 476 else 477 prp = pffindtype(dom, type); 478 if (prp == NULL) { 479 /* no support for domain */ 480 if (pffinddomain(dom) == 0) 481 return EAFNOSUPPORT; 482 /* no support for socket type */ 483 if (proto == 0 && type != 0) 484 return EPROTOTYPE; 485 return EPROTONOSUPPORT; 486 } 487 if (prp->pr_usrreq == NULL) 488 return EPROTONOSUPPORT; 489 if (prp->pr_type != type) 490 return EPROTOTYPE; 491 492 so = soget(true); 493 so->so_type = type; 494 so->so_proto = prp; 495 so->so_send = sosend; 496 so->so_receive = soreceive; 497 #ifdef MBUFTRACE 498 so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner; 499 so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner; 500 so->so_mowner = &prp->pr_domain->dom_mowner; 501 #endif 502 uid = kauth_cred_geteuid(l->l_cred); 503 so->so_uidinfo = uid_find(uid); 504 so->so_egid = kauth_cred_getegid(l->l_cred); 505 so->so_cpid = l->l_proc->p_pid; 506 if (lockso != NULL) { 507 /* Caller wants us to share a lock. */ 508 lock = lockso->so_lock; 509 so->so_lock = lock; 510 mutex_obj_hold(lock); 511 mutex_enter(lock); 512 } else { 513 /* Lock assigned and taken during PRU_ATTACH. */ 514 } 515 error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL, 516 (struct mbuf *)(long)proto, NULL, l); 517 KASSERT(solocked(so)); 518 if (error != 0) { 519 so->so_state |= SS_NOFDREF; 520 sofree(so); 521 return error; 522 } 523 sounlock(so); 524 *aso = so; 525 return 0; 526 } 527 528 /* On success, write file descriptor to fdout and return zero. On 529 * failure, return non-zero; *fdout will be undefined. 530 */ 531 int 532 fsocreate(int domain, struct socket **sop, int type, int protocol, 533 struct lwp *l, int *fdout) 534 { 535 struct socket *so; 536 struct file *fp; 537 int fd, error; 538 539 if ((error = fd_allocfile(&fp, &fd)) != 0) 540 return (error); 541 fp->f_flag = FREAD|FWRITE; 542 fp->f_type = DTYPE_SOCKET; 543 fp->f_ops = &socketops; 544 error = socreate(domain, &so, type, protocol, l, NULL); 545 if (error != 0) { 546 fd_abort(curproc, fp, fd); 547 } else { 548 if (sop != NULL) 549 *sop = so; 550 fp->f_data = so; 551 fd_affix(curproc, fp, fd); 552 *fdout = fd; 553 } 554 return error; 555 } 556 557 int 558 sobind(struct socket *so, struct mbuf *nam, struct lwp *l) 559 { 560 int error; 561 562 solock(so); 563 error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l); 564 sounlock(so); 565 return error; 566 } 567 568 int 569 solisten(struct socket *so, int backlog, struct lwp *l) 570 { 571 int error; 572 573 solock(so); 574 if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 575 SS_ISDISCONNECTING)) != 0) { 576 sounlock(so); 577 return (EOPNOTSUPP); 578 } 579 error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL, 580 NULL, NULL, l); 581 if (error != 0) { 582 sounlock(so); 583 return error; 584 } 585 if (TAILQ_EMPTY(&so->so_q)) 586 so->so_options |= SO_ACCEPTCONN; 587 if (backlog < 0) 588 backlog = 0; 589 so->so_qlimit = min(backlog, somaxconn); 590 sounlock(so); 591 return 0; 592 } 593 594 void 595 sofree(struct socket *so) 596 { 597 u_int refs; 598 599 KASSERT(solocked(so)); 600 601 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) { 602 sounlock(so); 603 return; 604 } 605 if (so->so_head) { 606 /* 607 * We must not decommission a socket that's on the accept(2) 608 * queue. If we do, then accept(2) may hang after select(2) 609 * indicated that the listening socket was ready. 610 */ 611 if (!soqremque(so, 0)) { 612 sounlock(so); 613 return; 614 } 615 } 616 if (so->so_rcv.sb_hiwat) 617 (void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0, 618 RLIM_INFINITY); 619 if (so->so_snd.sb_hiwat) 620 (void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0, 621 RLIM_INFINITY); 622 sbrelease(&so->so_snd, so); 623 KASSERT(!cv_has_waiters(&so->so_cv)); 624 KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv)); 625 KASSERT(!cv_has_waiters(&so->so_snd.sb_cv)); 626 sorflush(so); 627 refs = so->so_aborting; /* XXX */ 628 /* Remove acccept filter if one is present. */ 629 if (so->so_accf != NULL) 630 (void)accept_filt_clear(so); 631 sounlock(so); 632 if (refs == 0) /* XXX */ 633 soput(so); 634 } 635 636 /* 637 * Close a socket on last file table reference removal. 638 * Initiate disconnect if connected. 639 * Free socket when disconnect complete. 640 */ 641 int 642 soclose(struct socket *so) 643 { 644 struct socket *so2; 645 int error; 646 int error2; 647 648 error = 0; 649 solock(so); 650 if (so->so_options & SO_ACCEPTCONN) { 651 for (;;) { 652 if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) { 653 KASSERT(solocked2(so, so2)); 654 (void) soqremque(so2, 0); 655 /* soabort drops the lock. */ 656 (void) soabort(so2); 657 solock(so); 658 continue; 659 } 660 if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) { 661 KASSERT(solocked2(so, so2)); 662 (void) soqremque(so2, 1); 663 /* soabort drops the lock. */ 664 (void) soabort(so2); 665 solock(so); 666 continue; 667 } 668 break; 669 } 670 } 671 if (so->so_pcb == 0) 672 goto discard; 673 if (so->so_state & SS_ISCONNECTED) { 674 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 675 error = sodisconnect(so); 676 if (error) 677 goto drop; 678 } 679 if (so->so_options & SO_LINGER) { 680 if ((so->so_state & SS_ISDISCONNECTING) && so->so_nbio) 681 goto drop; 682 while (so->so_state & SS_ISCONNECTED) { 683 error = sowait(so, true, so->so_linger * hz); 684 if (error) 685 break; 686 } 687 } 688 } 689 drop: 690 if (so->so_pcb) { 691 error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH, 692 NULL, NULL, NULL, NULL); 693 if (error == 0) 694 error = error2; 695 } 696 discard: 697 if (so->so_state & SS_NOFDREF) 698 panic("soclose: NOFDREF"); 699 so->so_state |= SS_NOFDREF; 700 sofree(so); 701 return (error); 702 } 703 704 /* 705 * Must be called with the socket locked.. Will return with it unlocked. 706 */ 707 int 708 soabort(struct socket *so) 709 { 710 u_int refs; 711 int error; 712 713 KASSERT(solocked(so)); 714 KASSERT(so->so_head == NULL); 715 716 so->so_aborting++; /* XXX */ 717 error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL, 718 NULL, NULL, NULL); 719 refs = --so->so_aborting; /* XXX */ 720 if (error || (refs == 0)) { 721 sofree(so); 722 } else { 723 sounlock(so); 724 } 725 return error; 726 } 727 728 int 729 soaccept(struct socket *so, struct mbuf *nam) 730 { 731 int error; 732 733 KASSERT(solocked(so)); 734 735 error = 0; 736 if ((so->so_state & SS_NOFDREF) == 0) 737 panic("soaccept: !NOFDREF"); 738 so->so_state &= ~SS_NOFDREF; 739 if ((so->so_state & SS_ISDISCONNECTED) == 0 || 740 (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0) 741 error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT, 742 NULL, nam, NULL, NULL); 743 else 744 error = ECONNABORTED; 745 746 return (error); 747 } 748 749 int 750 soconnect(struct socket *so, struct mbuf *nam, struct lwp *l) 751 { 752 int error; 753 754 KASSERT(solocked(so)); 755 756 if (so->so_options & SO_ACCEPTCONN) 757 return (EOPNOTSUPP); 758 /* 759 * If protocol is connection-based, can only connect once. 760 * Otherwise, if connected, try to disconnect first. 761 * This allows user to disconnect by connecting to, e.g., 762 * a null address. 763 */ 764 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 765 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 766 (error = sodisconnect(so)))) 767 error = EISCONN; 768 else 769 error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT, 770 NULL, nam, NULL, l); 771 return (error); 772 } 773 774 int 775 soconnect2(struct socket *so1, struct socket *so2) 776 { 777 int error; 778 779 KASSERT(solocked2(so1, so2)); 780 781 error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2, 782 NULL, (struct mbuf *)so2, NULL, NULL); 783 return (error); 784 } 785 786 int 787 sodisconnect(struct socket *so) 788 { 789 int error; 790 791 KASSERT(solocked(so)); 792 793 if ((so->so_state & SS_ISCONNECTED) == 0) { 794 error = ENOTCONN; 795 } else if (so->so_state & SS_ISDISCONNECTING) { 796 error = EALREADY; 797 } else { 798 error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT, 799 NULL, NULL, NULL, NULL); 800 } 801 sodopendfree(); 802 return (error); 803 } 804 805 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 806 /* 807 * Send on a socket. 808 * If send must go all at once and message is larger than 809 * send buffering, then hard error. 810 * Lock against other senders. 811 * If must go all at once and not enough room now, then 812 * inform user that this would block and do nothing. 813 * Otherwise, if nonblocking, send as much as possible. 814 * The data to be sent is described by "uio" if nonzero, 815 * otherwise by the mbuf chain "top" (which must be null 816 * if uio is not). Data provided in mbuf chain must be small 817 * enough to send all at once. 818 * 819 * Returns nonzero on error, timeout or signal; callers 820 * must check for short counts if EINTR/ERESTART are returned. 821 * Data and control buffers are freed on return. 822 */ 823 int 824 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top, 825 struct mbuf *control, int flags, struct lwp *l) 826 { 827 struct mbuf **mp, *m; 828 struct proc *p; 829 long space, len, resid, clen, mlen; 830 int error, s, dontroute, atomic; 831 832 p = l->l_proc; 833 sodopendfree(); 834 clen = 0; 835 836 /* 837 * solock() provides atomicity of access. splsoftnet() prevents 838 * protocol processing soft interrupts from interrupting us and 839 * blocking (expensive). 840 */ 841 s = splsoftnet(); 842 solock(so); 843 atomic = sosendallatonce(so) || top; 844 if (uio) 845 resid = uio->uio_resid; 846 else 847 resid = top->m_pkthdr.len; 848 /* 849 * In theory resid should be unsigned. 850 * However, space must be signed, as it might be less than 0 851 * if we over-committed, and we must use a signed comparison 852 * of space and resid. On the other hand, a negative resid 853 * causes us to loop sending 0-length segments to the protocol. 854 */ 855 if (resid < 0) { 856 error = EINVAL; 857 goto out; 858 } 859 dontroute = 860 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 861 (so->so_proto->pr_flags & PR_ATOMIC); 862 l->l_ru.ru_msgsnd++; 863 if (control) 864 clen = control->m_len; 865 restart: 866 if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0) 867 goto out; 868 do { 869 if (so->so_state & SS_CANTSENDMORE) { 870 error = EPIPE; 871 goto release; 872 } 873 if (so->so_error) { 874 error = so->so_error; 875 so->so_error = 0; 876 goto release; 877 } 878 if ((so->so_state & SS_ISCONNECTED) == 0) { 879 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 880 if ((so->so_state & SS_ISCONFIRMING) == 0 && 881 !(resid == 0 && clen != 0)) { 882 error = ENOTCONN; 883 goto release; 884 } 885 } else if (addr == 0) { 886 error = EDESTADDRREQ; 887 goto release; 888 } 889 } 890 space = sbspace(&so->so_snd); 891 if (flags & MSG_OOB) 892 space += 1024; 893 if ((atomic && resid > so->so_snd.sb_hiwat) || 894 clen > so->so_snd.sb_hiwat) { 895 error = EMSGSIZE; 896 goto release; 897 } 898 if (space < resid + clen && 899 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 900 if (so->so_nbio) { 901 error = EWOULDBLOCK; 902 goto release; 903 } 904 sbunlock(&so->so_snd); 905 error = sbwait(&so->so_snd); 906 if (error) 907 goto out; 908 goto restart; 909 } 910 mp = ⊤ 911 space -= clen; 912 do { 913 if (uio == NULL) { 914 /* 915 * Data is prepackaged in "top". 916 */ 917 resid = 0; 918 if (flags & MSG_EOR) 919 top->m_flags |= M_EOR; 920 } else do { 921 sounlock(so); 922 splx(s); 923 if (top == NULL) { 924 m = m_gethdr(M_WAIT, MT_DATA); 925 mlen = MHLEN; 926 m->m_pkthdr.len = 0; 927 m->m_pkthdr.rcvif = NULL; 928 } else { 929 m = m_get(M_WAIT, MT_DATA); 930 mlen = MLEN; 931 } 932 MCLAIM(m, so->so_snd.sb_mowner); 933 if (sock_loan_thresh >= 0 && 934 uio->uio_iov->iov_len >= sock_loan_thresh && 935 space >= sock_loan_thresh && 936 (len = sosend_loan(so, uio, m, 937 space)) != 0) { 938 SOSEND_COUNTER_INCR(&sosend_loan_big); 939 space -= len; 940 goto have_data; 941 } 942 if (resid >= MINCLSIZE && space >= MCLBYTES) { 943 SOSEND_COUNTER_INCR(&sosend_copy_big); 944 m_clget(m, M_WAIT); 945 if ((m->m_flags & M_EXT) == 0) 946 goto nopages; 947 mlen = MCLBYTES; 948 if (atomic && top == 0) { 949 len = lmin(MCLBYTES - max_hdr, 950 resid); 951 m->m_data += max_hdr; 952 } else 953 len = lmin(MCLBYTES, resid); 954 space -= len; 955 } else { 956 nopages: 957 SOSEND_COUNTER_INCR(&sosend_copy_small); 958 len = lmin(lmin(mlen, resid), space); 959 space -= len; 960 /* 961 * For datagram protocols, leave room 962 * for protocol headers in first mbuf. 963 */ 964 if (atomic && top == 0 && len < mlen) 965 MH_ALIGN(m, len); 966 } 967 error = uiomove(mtod(m, void *), (int)len, uio); 968 have_data: 969 resid = uio->uio_resid; 970 m->m_len = len; 971 *mp = m; 972 top->m_pkthdr.len += len; 973 s = splsoftnet(); 974 solock(so); 975 if (error != 0) 976 goto release; 977 mp = &m->m_next; 978 if (resid <= 0) { 979 if (flags & MSG_EOR) 980 top->m_flags |= M_EOR; 981 break; 982 } 983 } while (space > 0 && atomic); 984 985 if (so->so_state & SS_CANTSENDMORE) { 986 error = EPIPE; 987 goto release; 988 } 989 if (dontroute) 990 so->so_options |= SO_DONTROUTE; 991 if (resid > 0) 992 so->so_state |= SS_MORETOCOME; 993 error = (*so->so_proto->pr_usrreq)(so, 994 (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND, 995 top, addr, control, curlwp); 996 if (dontroute) 997 so->so_options &= ~SO_DONTROUTE; 998 if (resid > 0) 999 so->so_state &= ~SS_MORETOCOME; 1000 clen = 0; 1001 control = NULL; 1002 top = NULL; 1003 mp = ⊤ 1004 if (error != 0) 1005 goto release; 1006 } while (resid && space > 0); 1007 } while (resid); 1008 1009 release: 1010 sbunlock(&so->so_snd); 1011 out: 1012 sounlock(so); 1013 splx(s); 1014 if (top) 1015 m_freem(top); 1016 if (control) 1017 m_freem(control); 1018 return (error); 1019 } 1020 1021 /* 1022 * Following replacement or removal of the first mbuf on the first 1023 * mbuf chain of a socket buffer, push necessary state changes back 1024 * into the socket buffer so that other consumers see the values 1025 * consistently. 'nextrecord' is the callers locally stored value of 1026 * the original value of sb->sb_mb->m_nextpkt which must be restored 1027 * when the lead mbuf changes. NOTE: 'nextrecord' may be NULL. 1028 */ 1029 static void 1030 sbsync(struct sockbuf *sb, struct mbuf *nextrecord) 1031 { 1032 1033 KASSERT(solocked(sb->sb_so)); 1034 1035 /* 1036 * First, update for the new value of nextrecord. If necessary, 1037 * make it the first record. 1038 */ 1039 if (sb->sb_mb != NULL) 1040 sb->sb_mb->m_nextpkt = nextrecord; 1041 else 1042 sb->sb_mb = nextrecord; 1043 1044 /* 1045 * Now update any dependent socket buffer fields to reflect 1046 * the new state. This is an inline of SB_EMPTY_FIXUP, with 1047 * the addition of a second clause that takes care of the 1048 * case where sb_mb has been updated, but remains the last 1049 * record. 1050 */ 1051 if (sb->sb_mb == NULL) { 1052 sb->sb_mbtail = NULL; 1053 sb->sb_lastrecord = NULL; 1054 } else if (sb->sb_mb->m_nextpkt == NULL) 1055 sb->sb_lastrecord = sb->sb_mb; 1056 } 1057 1058 /* 1059 * Implement receive operations on a socket. 1060 * We depend on the way that records are added to the sockbuf 1061 * by sbappend*. In particular, each record (mbufs linked through m_next) 1062 * must begin with an address if the protocol so specifies, 1063 * followed by an optional mbuf or mbufs containing ancillary data, 1064 * and then zero or more mbufs of data. 1065 * In order to avoid blocking network interrupts for the entire time here, 1066 * we splx() while doing the actual copy to user space. 1067 * Although the sockbuf is locked, new data may still be appended, 1068 * and thus we must maintain consistency of the sockbuf during that time. 1069 * 1070 * The caller may receive the data as a single mbuf chain by supplying 1071 * an mbuf **mp0 for use in returning the chain. The uio is then used 1072 * only for the count in uio_resid. 1073 */ 1074 int 1075 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio, 1076 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1077 { 1078 struct lwp *l = curlwp; 1079 struct mbuf *m, **mp, *mt; 1080 int atomic, flags, len, error, s, offset, moff, type, orig_resid; 1081 const struct protosw *pr; 1082 struct mbuf *nextrecord; 1083 int mbuf_removed = 0; 1084 const struct domain *dom; 1085 1086 pr = so->so_proto; 1087 atomic = pr->pr_flags & PR_ATOMIC; 1088 dom = pr->pr_domain; 1089 mp = mp0; 1090 type = 0; 1091 orig_resid = uio->uio_resid; 1092 1093 if (paddr != NULL) 1094 *paddr = NULL; 1095 if (controlp != NULL) 1096 *controlp = NULL; 1097 if (flagsp != NULL) 1098 flags = *flagsp &~ MSG_EOR; 1099 else 1100 flags = 0; 1101 1102 if ((flags & MSG_DONTWAIT) == 0) 1103 sodopendfree(); 1104 1105 if (flags & MSG_OOB) { 1106 m = m_get(M_WAIT, MT_DATA); 1107 solock(so); 1108 error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m, 1109 (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l); 1110 sounlock(so); 1111 if (error) 1112 goto bad; 1113 do { 1114 error = uiomove(mtod(m, void *), 1115 (int) min(uio->uio_resid, m->m_len), uio); 1116 m = m_free(m); 1117 } while (uio->uio_resid > 0 && error == 0 && m); 1118 bad: 1119 if (m != NULL) 1120 m_freem(m); 1121 return error; 1122 } 1123 if (mp != NULL) 1124 *mp = NULL; 1125 1126 /* 1127 * solock() provides atomicity of access. splsoftnet() prevents 1128 * protocol processing soft interrupts from interrupting us and 1129 * blocking (expensive). 1130 */ 1131 s = splsoftnet(); 1132 solock(so); 1133 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 1134 (*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l); 1135 1136 restart: 1137 if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) { 1138 sounlock(so); 1139 splx(s); 1140 return error; 1141 } 1142 1143 m = so->so_rcv.sb_mb; 1144 /* 1145 * If we have less data than requested, block awaiting more 1146 * (subject to any timeout) if: 1147 * 1. the current count is less than the low water mark, 1148 * 2. MSG_WAITALL is set, and it is possible to do the entire 1149 * receive operation at once if we block (resid <= hiwat), or 1150 * 3. MSG_DONTWAIT is not set. 1151 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1152 * we have to do the receive in sections, and thus risk returning 1153 * a short count if a timeout or signal occurs after we start. 1154 */ 1155 if (m == NULL || 1156 ((flags & MSG_DONTWAIT) == 0 && 1157 so->so_rcv.sb_cc < uio->uio_resid && 1158 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1159 ((flags & MSG_WAITALL) && 1160 uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1161 m->m_nextpkt == NULL && !atomic)) { 1162 #ifdef DIAGNOSTIC 1163 if (m == NULL && so->so_rcv.sb_cc) 1164 panic("receive 1"); 1165 #endif 1166 if (so->so_error) { 1167 if (m != NULL) 1168 goto dontblock; 1169 error = so->so_error; 1170 if ((flags & MSG_PEEK) == 0) 1171 so->so_error = 0; 1172 goto release; 1173 } 1174 if (so->so_state & SS_CANTRCVMORE) { 1175 if (m != NULL) 1176 goto dontblock; 1177 else 1178 goto release; 1179 } 1180 for (; m != NULL; m = m->m_next) 1181 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1182 m = so->so_rcv.sb_mb; 1183 goto dontblock; 1184 } 1185 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1186 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1187 error = ENOTCONN; 1188 goto release; 1189 } 1190 if (uio->uio_resid == 0) 1191 goto release; 1192 if (so->so_nbio || (flags & MSG_DONTWAIT)) { 1193 error = EWOULDBLOCK; 1194 goto release; 1195 } 1196 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1"); 1197 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1"); 1198 sbunlock(&so->so_rcv); 1199 error = sbwait(&so->so_rcv); 1200 if (error != 0) { 1201 sounlock(so); 1202 splx(s); 1203 return error; 1204 } 1205 goto restart; 1206 } 1207 dontblock: 1208 /* 1209 * On entry here, m points to the first record of the socket buffer. 1210 * From this point onward, we maintain 'nextrecord' as a cache of the 1211 * pointer to the next record in the socket buffer. We must keep the 1212 * various socket buffer pointers and local stack versions of the 1213 * pointers in sync, pushing out modifications before dropping the 1214 * socket lock, and re-reading them when picking it up. 1215 * 1216 * Otherwise, we will race with the network stack appending new data 1217 * or records onto the socket buffer by using inconsistent/stale 1218 * versions of the field, possibly resulting in socket buffer 1219 * corruption. 1220 * 1221 * By holding the high-level sblock(), we prevent simultaneous 1222 * readers from pulling off the front of the socket buffer. 1223 */ 1224 if (l != NULL) 1225 l->l_ru.ru_msgrcv++; 1226 KASSERT(m == so->so_rcv.sb_mb); 1227 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1"); 1228 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1"); 1229 nextrecord = m->m_nextpkt; 1230 if (pr->pr_flags & PR_ADDR) { 1231 #ifdef DIAGNOSTIC 1232 if (m->m_type != MT_SONAME) 1233 panic("receive 1a"); 1234 #endif 1235 orig_resid = 0; 1236 if (flags & MSG_PEEK) { 1237 if (paddr) 1238 *paddr = m_copy(m, 0, m->m_len); 1239 m = m->m_next; 1240 } else { 1241 sbfree(&so->so_rcv, m); 1242 mbuf_removed = 1; 1243 if (paddr != NULL) { 1244 *paddr = m; 1245 so->so_rcv.sb_mb = m->m_next; 1246 m->m_next = NULL; 1247 m = so->so_rcv.sb_mb; 1248 } else { 1249 MFREE(m, so->so_rcv.sb_mb); 1250 m = so->so_rcv.sb_mb; 1251 } 1252 sbsync(&so->so_rcv, nextrecord); 1253 } 1254 } 1255 1256 /* 1257 * Process one or more MT_CONTROL mbufs present before any data mbufs 1258 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 1259 * just copy the data; if !MSG_PEEK, we call into the protocol to 1260 * perform externalization (or freeing if controlp == NULL). 1261 */ 1262 if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) { 1263 struct mbuf *cm = NULL, *cmn; 1264 struct mbuf **cme = &cm; 1265 1266 do { 1267 if (flags & MSG_PEEK) { 1268 if (controlp != NULL) { 1269 *controlp = m_copy(m, 0, m->m_len); 1270 controlp = &(*controlp)->m_next; 1271 } 1272 m = m->m_next; 1273 } else { 1274 sbfree(&so->so_rcv, m); 1275 so->so_rcv.sb_mb = m->m_next; 1276 m->m_next = NULL; 1277 *cme = m; 1278 cme = &(*cme)->m_next; 1279 m = so->so_rcv.sb_mb; 1280 } 1281 } while (m != NULL && m->m_type == MT_CONTROL); 1282 if ((flags & MSG_PEEK) == 0) 1283 sbsync(&so->so_rcv, nextrecord); 1284 for (; cm != NULL; cm = cmn) { 1285 cmn = cm->m_next; 1286 cm->m_next = NULL; 1287 type = mtod(cm, struct cmsghdr *)->cmsg_type; 1288 if (controlp != NULL) { 1289 if (dom->dom_externalize != NULL && 1290 type == SCM_RIGHTS) { 1291 sounlock(so); 1292 splx(s); 1293 error = (*dom->dom_externalize)(cm, l); 1294 s = splsoftnet(); 1295 solock(so); 1296 } 1297 *controlp = cm; 1298 while (*controlp != NULL) 1299 controlp = &(*controlp)->m_next; 1300 } else { 1301 /* 1302 * Dispose of any SCM_RIGHTS message that went 1303 * through the read path rather than recv. 1304 */ 1305 if (dom->dom_dispose != NULL && 1306 type == SCM_RIGHTS) { 1307 sounlock(so); 1308 (*dom->dom_dispose)(cm); 1309 solock(so); 1310 } 1311 m_freem(cm); 1312 } 1313 } 1314 if (m != NULL) 1315 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 1316 else 1317 nextrecord = so->so_rcv.sb_mb; 1318 orig_resid = 0; 1319 } 1320 1321 /* If m is non-NULL, we have some data to read. */ 1322 if (__predict_true(m != NULL)) { 1323 type = m->m_type; 1324 if (type == MT_OOBDATA) 1325 flags |= MSG_OOB; 1326 } 1327 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2"); 1328 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2"); 1329 1330 moff = 0; 1331 offset = 0; 1332 while (m != NULL && uio->uio_resid > 0 && error == 0) { 1333 if (m->m_type == MT_OOBDATA) { 1334 if (type != MT_OOBDATA) 1335 break; 1336 } else if (type == MT_OOBDATA) 1337 break; 1338 #ifdef DIAGNOSTIC 1339 else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) 1340 panic("receive 3"); 1341 #endif 1342 so->so_state &= ~SS_RCVATMARK; 1343 len = uio->uio_resid; 1344 if (so->so_oobmark && len > so->so_oobmark - offset) 1345 len = so->so_oobmark - offset; 1346 if (len > m->m_len - moff) 1347 len = m->m_len - moff; 1348 /* 1349 * If mp is set, just pass back the mbufs. 1350 * Otherwise copy them out via the uio, then free. 1351 * Sockbuf must be consistent here (points to current mbuf, 1352 * it points to next record) when we drop priority; 1353 * we must note any additions to the sockbuf when we 1354 * block interrupts again. 1355 */ 1356 if (mp == NULL) { 1357 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove"); 1358 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove"); 1359 sounlock(so); 1360 splx(s); 1361 error = uiomove(mtod(m, char *) + moff, (int)len, uio); 1362 s = splsoftnet(); 1363 solock(so); 1364 if (error != 0) { 1365 /* 1366 * If any part of the record has been removed 1367 * (such as the MT_SONAME mbuf, which will 1368 * happen when PR_ADDR, and thus also 1369 * PR_ATOMIC, is set), then drop the entire 1370 * record to maintain the atomicity of the 1371 * receive operation. 1372 * 1373 * This avoids a later panic("receive 1a") 1374 * when compiled with DIAGNOSTIC. 1375 */ 1376 if (m && mbuf_removed && atomic) 1377 (void) sbdroprecord(&so->so_rcv); 1378 1379 goto release; 1380 } 1381 } else 1382 uio->uio_resid -= len; 1383 if (len == m->m_len - moff) { 1384 if (m->m_flags & M_EOR) 1385 flags |= MSG_EOR; 1386 if (flags & MSG_PEEK) { 1387 m = m->m_next; 1388 moff = 0; 1389 } else { 1390 nextrecord = m->m_nextpkt; 1391 sbfree(&so->so_rcv, m); 1392 if (mp) { 1393 *mp = m; 1394 mp = &m->m_next; 1395 so->so_rcv.sb_mb = m = m->m_next; 1396 *mp = NULL; 1397 } else { 1398 MFREE(m, so->so_rcv.sb_mb); 1399 m = so->so_rcv.sb_mb; 1400 } 1401 /* 1402 * If m != NULL, we also know that 1403 * so->so_rcv.sb_mb != NULL. 1404 */ 1405 KASSERT(so->so_rcv.sb_mb == m); 1406 if (m) { 1407 m->m_nextpkt = nextrecord; 1408 if (nextrecord == NULL) 1409 so->so_rcv.sb_lastrecord = m; 1410 } else { 1411 so->so_rcv.sb_mb = nextrecord; 1412 SB_EMPTY_FIXUP(&so->so_rcv); 1413 } 1414 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3"); 1415 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3"); 1416 } 1417 } else if (flags & MSG_PEEK) 1418 moff += len; 1419 else { 1420 if (mp != NULL) { 1421 mt = m_copym(m, 0, len, M_NOWAIT); 1422 if (__predict_false(mt == NULL)) { 1423 sounlock(so); 1424 mt = m_copym(m, 0, len, M_WAIT); 1425 solock(so); 1426 } 1427 *mp = mt; 1428 } 1429 m->m_data += len; 1430 m->m_len -= len; 1431 so->so_rcv.sb_cc -= len; 1432 } 1433 if (so->so_oobmark) { 1434 if ((flags & MSG_PEEK) == 0) { 1435 so->so_oobmark -= len; 1436 if (so->so_oobmark == 0) { 1437 so->so_state |= SS_RCVATMARK; 1438 break; 1439 } 1440 } else { 1441 offset += len; 1442 if (offset == so->so_oobmark) 1443 break; 1444 } 1445 } 1446 if (flags & MSG_EOR) 1447 break; 1448 /* 1449 * If the MSG_WAITALL flag is set (for non-atomic socket), 1450 * we must not quit until "uio->uio_resid == 0" or an error 1451 * termination. If a signal/timeout occurs, return 1452 * with a short count but without error. 1453 * Keep sockbuf locked against other readers. 1454 */ 1455 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 1456 !sosendallatonce(so) && !nextrecord) { 1457 if (so->so_error || so->so_state & SS_CANTRCVMORE) 1458 break; 1459 /* 1460 * If we are peeking and the socket receive buffer is 1461 * full, stop since we can't get more data to peek at. 1462 */ 1463 if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0) 1464 break; 1465 /* 1466 * If we've drained the socket buffer, tell the 1467 * protocol in case it needs to do something to 1468 * get it filled again. 1469 */ 1470 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) 1471 (*pr->pr_usrreq)(so, PRU_RCVD, 1472 NULL, (struct mbuf *)(long)flags, NULL, l); 1473 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2"); 1474 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2"); 1475 error = sbwait(&so->so_rcv); 1476 if (error != 0) { 1477 sbunlock(&so->so_rcv); 1478 sounlock(so); 1479 splx(s); 1480 return 0; 1481 } 1482 if ((m = so->so_rcv.sb_mb) != NULL) 1483 nextrecord = m->m_nextpkt; 1484 } 1485 } 1486 1487 if (m && atomic) { 1488 flags |= MSG_TRUNC; 1489 if ((flags & MSG_PEEK) == 0) 1490 (void) sbdroprecord(&so->so_rcv); 1491 } 1492 if ((flags & MSG_PEEK) == 0) { 1493 if (m == NULL) { 1494 /* 1495 * First part is an inline SB_EMPTY_FIXUP(). Second 1496 * part makes sure sb_lastrecord is up-to-date if 1497 * there is still data in the socket buffer. 1498 */ 1499 so->so_rcv.sb_mb = nextrecord; 1500 if (so->so_rcv.sb_mb == NULL) { 1501 so->so_rcv.sb_mbtail = NULL; 1502 so->so_rcv.sb_lastrecord = NULL; 1503 } else if (nextrecord->m_nextpkt == NULL) 1504 so->so_rcv.sb_lastrecord = nextrecord; 1505 } 1506 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4"); 1507 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4"); 1508 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) 1509 (*pr->pr_usrreq)(so, PRU_RCVD, NULL, 1510 (struct mbuf *)(long)flags, NULL, l); 1511 } 1512 if (orig_resid == uio->uio_resid && orig_resid && 1513 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { 1514 sbunlock(&so->so_rcv); 1515 goto restart; 1516 } 1517 1518 if (flagsp != NULL) 1519 *flagsp |= flags; 1520 release: 1521 sbunlock(&so->so_rcv); 1522 sounlock(so); 1523 splx(s); 1524 return error; 1525 } 1526 1527 int 1528 soshutdown(struct socket *so, int how) 1529 { 1530 const struct protosw *pr; 1531 int error; 1532 1533 KASSERT(solocked(so)); 1534 1535 pr = so->so_proto; 1536 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1537 return (EINVAL); 1538 1539 if (how == SHUT_RD || how == SHUT_RDWR) { 1540 sorflush(so); 1541 error = 0; 1542 } 1543 if (how == SHUT_WR || how == SHUT_RDWR) 1544 error = (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL, 1545 NULL, NULL, NULL); 1546 1547 return error; 1548 } 1549 1550 void 1551 sorflush(struct socket *so) 1552 { 1553 struct sockbuf *sb, asb; 1554 const struct protosw *pr; 1555 1556 KASSERT(solocked(so)); 1557 1558 sb = &so->so_rcv; 1559 pr = so->so_proto; 1560 socantrcvmore(so); 1561 sb->sb_flags |= SB_NOINTR; 1562 (void )sblock(sb, M_WAITOK); 1563 sbunlock(sb); 1564 asb = *sb; 1565 /* 1566 * Clear most of the sockbuf structure, but leave some of the 1567 * fields valid. 1568 */ 1569 memset(&sb->sb_startzero, 0, 1570 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1571 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) { 1572 sounlock(so); 1573 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1574 solock(so); 1575 } 1576 sbrelease(&asb, so); 1577 } 1578 1579 /* 1580 * internal set SOL_SOCKET options 1581 */ 1582 static int 1583 sosetopt1(struct socket *so, const struct sockopt *sopt) 1584 { 1585 int error = EINVAL, optval, opt; 1586 struct linger l; 1587 struct timeval tv; 1588 1589 switch ((opt = sopt->sopt_name)) { 1590 1591 case SO_ACCEPTFILTER: 1592 error = accept_filt_setopt(so, sopt); 1593 KASSERT(solocked(so)); 1594 break; 1595 1596 case SO_LINGER: 1597 error = sockopt_get(sopt, &l, sizeof(l)); 1598 solock(so); 1599 if (error) 1600 break; 1601 if (l.l_linger < 0 || l.l_linger > USHRT_MAX || 1602 l.l_linger > (INT_MAX / hz)) { 1603 error = EDOM; 1604 break; 1605 } 1606 so->so_linger = l.l_linger; 1607 if (l.l_onoff) 1608 so->so_options |= SO_LINGER; 1609 else 1610 so->so_options &= ~SO_LINGER; 1611 break; 1612 1613 case SO_DEBUG: 1614 case SO_KEEPALIVE: 1615 case SO_DONTROUTE: 1616 case SO_USELOOPBACK: 1617 case SO_BROADCAST: 1618 case SO_REUSEADDR: 1619 case SO_REUSEPORT: 1620 case SO_OOBINLINE: 1621 case SO_TIMESTAMP: 1622 #ifdef SO_OTIMESTAMP 1623 case SO_OTIMESTAMP: 1624 #endif 1625 error = sockopt_getint(sopt, &optval); 1626 solock(so); 1627 if (error) 1628 break; 1629 if (optval) 1630 so->so_options |= opt; 1631 else 1632 so->so_options &= ~opt; 1633 break; 1634 1635 case SO_SNDBUF: 1636 case SO_RCVBUF: 1637 case SO_SNDLOWAT: 1638 case SO_RCVLOWAT: 1639 error = sockopt_getint(sopt, &optval); 1640 solock(so); 1641 if (error) 1642 break; 1643 1644 /* 1645 * Values < 1 make no sense for any of these 1646 * options, so disallow them. 1647 */ 1648 if (optval < 1) { 1649 error = EINVAL; 1650 break; 1651 } 1652 1653 switch (opt) { 1654 case SO_SNDBUF: 1655 if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) { 1656 error = ENOBUFS; 1657 break; 1658 } 1659 so->so_snd.sb_flags &= ~SB_AUTOSIZE; 1660 break; 1661 1662 case SO_RCVBUF: 1663 if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) { 1664 error = ENOBUFS; 1665 break; 1666 } 1667 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1668 break; 1669 1670 /* 1671 * Make sure the low-water is never greater than 1672 * the high-water. 1673 */ 1674 case SO_SNDLOWAT: 1675 if (optval > so->so_snd.sb_hiwat) 1676 optval = so->so_snd.sb_hiwat; 1677 1678 so->so_snd.sb_lowat = optval; 1679 break; 1680 1681 case SO_RCVLOWAT: 1682 if (optval > so->so_rcv.sb_hiwat) 1683 optval = so->so_rcv.sb_hiwat; 1684 1685 so->so_rcv.sb_lowat = optval; 1686 break; 1687 } 1688 break; 1689 1690 #ifdef COMPAT_50 1691 case SO_OSNDTIMEO: 1692 case SO_ORCVTIMEO: { 1693 struct timeval50 otv; 1694 error = sockopt_get(sopt, &otv, sizeof(otv)); 1695 if (error) { 1696 solock(so); 1697 break; 1698 } 1699 timeval50_to_timeval(&otv, &tv); 1700 opt = opt == SO_OSNDTIMEO ? SO_SNDTIMEO : SO_RCVTIMEO; 1701 error = 0; 1702 /*FALLTHROUGH*/ 1703 } 1704 #endif /* COMPAT_50 */ 1705 1706 case SO_SNDTIMEO: 1707 case SO_RCVTIMEO: 1708 if (error) 1709 error = sockopt_get(sopt, &tv, sizeof(tv)); 1710 solock(so); 1711 if (error) 1712 break; 1713 1714 if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) { 1715 error = EDOM; 1716 break; 1717 } 1718 1719 optval = tv.tv_sec * hz + tv.tv_usec / tick; 1720 if (optval == 0 && tv.tv_usec != 0) 1721 optval = 1; 1722 1723 switch (opt) { 1724 case SO_SNDTIMEO: 1725 so->so_snd.sb_timeo = optval; 1726 break; 1727 case SO_RCVTIMEO: 1728 so->so_rcv.sb_timeo = optval; 1729 break; 1730 } 1731 break; 1732 1733 default: 1734 solock(so); 1735 error = ENOPROTOOPT; 1736 break; 1737 } 1738 KASSERT(solocked(so)); 1739 return error; 1740 } 1741 1742 int 1743 sosetopt(struct socket *so, struct sockopt *sopt) 1744 { 1745 int error, prerr; 1746 1747 if (sopt->sopt_level == SOL_SOCKET) { 1748 error = sosetopt1(so, sopt); 1749 KASSERT(solocked(so)); 1750 } else { 1751 error = ENOPROTOOPT; 1752 solock(so); 1753 } 1754 1755 if ((error == 0 || error == ENOPROTOOPT) && 1756 so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) { 1757 /* give the protocol stack a shot */ 1758 prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt); 1759 if (prerr == 0) 1760 error = 0; 1761 else if (prerr != ENOPROTOOPT) 1762 error = prerr; 1763 } 1764 sounlock(so); 1765 return error; 1766 } 1767 1768 /* 1769 * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt() 1770 */ 1771 int 1772 so_setsockopt(struct lwp *l, struct socket *so, int level, int name, 1773 const void *val, size_t valsize) 1774 { 1775 struct sockopt sopt; 1776 int error; 1777 1778 KASSERT(valsize == 0 || val != NULL); 1779 1780 sockopt_init(&sopt, level, name, valsize); 1781 sockopt_set(&sopt, val, valsize); 1782 1783 error = sosetopt(so, &sopt); 1784 1785 sockopt_destroy(&sopt); 1786 1787 return error; 1788 } 1789 1790 /* 1791 * internal get SOL_SOCKET options 1792 */ 1793 static int 1794 sogetopt1(struct socket *so, struct sockopt *sopt) 1795 { 1796 int error, optval, opt; 1797 struct linger l; 1798 struct timeval tv; 1799 1800 switch ((opt = sopt->sopt_name)) { 1801 1802 case SO_ACCEPTFILTER: 1803 error = accept_filt_getopt(so, sopt); 1804 break; 1805 1806 case SO_LINGER: 1807 l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0; 1808 l.l_linger = so->so_linger; 1809 1810 error = sockopt_set(sopt, &l, sizeof(l)); 1811 break; 1812 1813 case SO_USELOOPBACK: 1814 case SO_DONTROUTE: 1815 case SO_DEBUG: 1816 case SO_KEEPALIVE: 1817 case SO_REUSEADDR: 1818 case SO_REUSEPORT: 1819 case SO_BROADCAST: 1820 case SO_OOBINLINE: 1821 case SO_TIMESTAMP: 1822 #ifdef SO_OTIMESTAMP 1823 case SO_OTIMESTAMP: 1824 #endif 1825 error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0); 1826 break; 1827 1828 case SO_TYPE: 1829 error = sockopt_setint(sopt, so->so_type); 1830 break; 1831 1832 case SO_ERROR: 1833 error = sockopt_setint(sopt, so->so_error); 1834 so->so_error = 0; 1835 break; 1836 1837 case SO_SNDBUF: 1838 error = sockopt_setint(sopt, so->so_snd.sb_hiwat); 1839 break; 1840 1841 case SO_RCVBUF: 1842 error = sockopt_setint(sopt, so->so_rcv.sb_hiwat); 1843 break; 1844 1845 case SO_SNDLOWAT: 1846 error = sockopt_setint(sopt, so->so_snd.sb_lowat); 1847 break; 1848 1849 case SO_RCVLOWAT: 1850 error = sockopt_setint(sopt, so->so_rcv.sb_lowat); 1851 break; 1852 1853 #ifdef COMPAT_50 1854 case SO_OSNDTIMEO: 1855 case SO_ORCVTIMEO: { 1856 struct timeval50 otv; 1857 1858 optval = (opt == SO_OSNDTIMEO ? 1859 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1860 1861 otv.tv_sec = optval / hz; 1862 otv.tv_usec = (optval % hz) * tick; 1863 1864 error = sockopt_set(sopt, &otv, sizeof(otv)); 1865 break; 1866 } 1867 #endif /* COMPAT_50 */ 1868 1869 case SO_SNDTIMEO: 1870 case SO_RCVTIMEO: 1871 optval = (opt == SO_SNDTIMEO ? 1872 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1873 1874 tv.tv_sec = optval / hz; 1875 tv.tv_usec = (optval % hz) * tick; 1876 1877 error = sockopt_set(sopt, &tv, sizeof(tv)); 1878 break; 1879 1880 case SO_OVERFLOWED: 1881 error = sockopt_setint(sopt, so->so_rcv.sb_overflowed); 1882 break; 1883 1884 default: 1885 error = ENOPROTOOPT; 1886 break; 1887 } 1888 1889 return (error); 1890 } 1891 1892 int 1893 sogetopt(struct socket *so, struct sockopt *sopt) 1894 { 1895 int error; 1896 1897 solock(so); 1898 if (sopt->sopt_level != SOL_SOCKET) { 1899 if (so->so_proto && so->so_proto->pr_ctloutput) { 1900 error = ((*so->so_proto->pr_ctloutput) 1901 (PRCO_GETOPT, so, sopt)); 1902 } else 1903 error = (ENOPROTOOPT); 1904 } else { 1905 error = sogetopt1(so, sopt); 1906 } 1907 sounlock(so); 1908 return (error); 1909 } 1910 1911 /* 1912 * alloc sockopt data buffer buffer 1913 * - will be released at destroy 1914 */ 1915 static int 1916 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag) 1917 { 1918 1919 KASSERT(sopt->sopt_size == 0); 1920 1921 if (len > sizeof(sopt->sopt_buf)) { 1922 sopt->sopt_data = kmem_zalloc(len, kmflag); 1923 if (sopt->sopt_data == NULL) 1924 return ENOMEM; 1925 } else 1926 sopt->sopt_data = sopt->sopt_buf; 1927 1928 sopt->sopt_size = len; 1929 return 0; 1930 } 1931 1932 /* 1933 * initialise sockopt storage 1934 * - MAY sleep during allocation 1935 */ 1936 void 1937 sockopt_init(struct sockopt *sopt, int level, int name, size_t size) 1938 { 1939 1940 memset(sopt, 0, sizeof(*sopt)); 1941 1942 sopt->sopt_level = level; 1943 sopt->sopt_name = name; 1944 (void)sockopt_alloc(sopt, size, KM_SLEEP); 1945 } 1946 1947 /* 1948 * destroy sockopt storage 1949 * - will release any held memory references 1950 */ 1951 void 1952 sockopt_destroy(struct sockopt *sopt) 1953 { 1954 1955 if (sopt->sopt_data != sopt->sopt_buf) 1956 kmem_free(sopt->sopt_data, sopt->sopt_size); 1957 1958 memset(sopt, 0, sizeof(*sopt)); 1959 } 1960 1961 /* 1962 * set sockopt value 1963 * - value is copied into sockopt 1964 * - memory is allocated when necessary, will not sleep 1965 */ 1966 int 1967 sockopt_set(struct sockopt *sopt, const void *buf, size_t len) 1968 { 1969 int error; 1970 1971 if (sopt->sopt_size == 0) { 1972 error = sockopt_alloc(sopt, len, KM_NOSLEEP); 1973 if (error) 1974 return error; 1975 } 1976 1977 KASSERT(sopt->sopt_size == len); 1978 memcpy(sopt->sopt_data, buf, len); 1979 return 0; 1980 } 1981 1982 /* 1983 * common case of set sockopt integer value 1984 */ 1985 int 1986 sockopt_setint(struct sockopt *sopt, int val) 1987 { 1988 1989 return sockopt_set(sopt, &val, sizeof(int)); 1990 } 1991 1992 /* 1993 * get sockopt value 1994 * - correct size must be given 1995 */ 1996 int 1997 sockopt_get(const struct sockopt *sopt, void *buf, size_t len) 1998 { 1999 2000 if (sopt->sopt_size != len) 2001 return EINVAL; 2002 2003 memcpy(buf, sopt->sopt_data, len); 2004 return 0; 2005 } 2006 2007 /* 2008 * common case of get sockopt integer value 2009 */ 2010 int 2011 sockopt_getint(const struct sockopt *sopt, int *valp) 2012 { 2013 2014 return sockopt_get(sopt, valp, sizeof(int)); 2015 } 2016 2017 /* 2018 * set sockopt value from mbuf 2019 * - ONLY for legacy code 2020 * - mbuf is released by sockopt 2021 * - will not sleep 2022 */ 2023 int 2024 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m) 2025 { 2026 size_t len; 2027 int error; 2028 2029 len = m_length(m); 2030 2031 if (sopt->sopt_size == 0) { 2032 error = sockopt_alloc(sopt, len, KM_NOSLEEP); 2033 if (error) 2034 return error; 2035 } 2036 2037 KASSERT(sopt->sopt_size == len); 2038 m_copydata(m, 0, len, sopt->sopt_data); 2039 m_freem(m); 2040 2041 return 0; 2042 } 2043 2044 /* 2045 * get sockopt value into mbuf 2046 * - ONLY for legacy code 2047 * - mbuf to be released by the caller 2048 * - will not sleep 2049 */ 2050 struct mbuf * 2051 sockopt_getmbuf(const struct sockopt *sopt) 2052 { 2053 struct mbuf *m; 2054 2055 if (sopt->sopt_size > MCLBYTES) 2056 return NULL; 2057 2058 m = m_get(M_DONTWAIT, MT_SOOPTS); 2059 if (m == NULL) 2060 return NULL; 2061 2062 if (sopt->sopt_size > MLEN) { 2063 MCLGET(m, M_DONTWAIT); 2064 if ((m->m_flags & M_EXT) == 0) { 2065 m_free(m); 2066 return NULL; 2067 } 2068 } 2069 2070 memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size); 2071 m->m_len = sopt->sopt_size; 2072 2073 return m; 2074 } 2075 2076 void 2077 sohasoutofband(struct socket *so) 2078 { 2079 2080 fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so); 2081 selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, 0); 2082 } 2083 2084 static void 2085 filt_sordetach(struct knote *kn) 2086 { 2087 struct socket *so; 2088 2089 so = ((file_t *)kn->kn_obj)->f_data; 2090 solock(so); 2091 SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext); 2092 if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist)) 2093 so->so_rcv.sb_flags &= ~SB_KNOTE; 2094 sounlock(so); 2095 } 2096 2097 /*ARGSUSED*/ 2098 static int 2099 filt_soread(struct knote *kn, long hint) 2100 { 2101 struct socket *so; 2102 int rv; 2103 2104 so = ((file_t *)kn->kn_obj)->f_data; 2105 if (hint != NOTE_SUBMIT) 2106 solock(so); 2107 kn->kn_data = so->so_rcv.sb_cc; 2108 if (so->so_state & SS_CANTRCVMORE) { 2109 kn->kn_flags |= EV_EOF; 2110 kn->kn_fflags = so->so_error; 2111 rv = 1; 2112 } else if (so->so_error) /* temporary udp error */ 2113 rv = 1; 2114 else if (kn->kn_sfflags & NOTE_LOWAT) 2115 rv = (kn->kn_data >= kn->kn_sdata); 2116 else 2117 rv = (kn->kn_data >= so->so_rcv.sb_lowat); 2118 if (hint != NOTE_SUBMIT) 2119 sounlock(so); 2120 return rv; 2121 } 2122 2123 static void 2124 filt_sowdetach(struct knote *kn) 2125 { 2126 struct socket *so; 2127 2128 so = ((file_t *)kn->kn_obj)->f_data; 2129 solock(so); 2130 SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext); 2131 if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist)) 2132 so->so_snd.sb_flags &= ~SB_KNOTE; 2133 sounlock(so); 2134 } 2135 2136 /*ARGSUSED*/ 2137 static int 2138 filt_sowrite(struct knote *kn, long hint) 2139 { 2140 struct socket *so; 2141 int rv; 2142 2143 so = ((file_t *)kn->kn_obj)->f_data; 2144 if (hint != NOTE_SUBMIT) 2145 solock(so); 2146 kn->kn_data = sbspace(&so->so_snd); 2147 if (so->so_state & SS_CANTSENDMORE) { 2148 kn->kn_flags |= EV_EOF; 2149 kn->kn_fflags = so->so_error; 2150 rv = 1; 2151 } else if (so->so_error) /* temporary udp error */ 2152 rv = 1; 2153 else if (((so->so_state & SS_ISCONNECTED) == 0) && 2154 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 2155 rv = 0; 2156 else if (kn->kn_sfflags & NOTE_LOWAT) 2157 rv = (kn->kn_data >= kn->kn_sdata); 2158 else 2159 rv = (kn->kn_data >= so->so_snd.sb_lowat); 2160 if (hint != NOTE_SUBMIT) 2161 sounlock(so); 2162 return rv; 2163 } 2164 2165 /*ARGSUSED*/ 2166 static int 2167 filt_solisten(struct knote *kn, long hint) 2168 { 2169 struct socket *so; 2170 int rv; 2171 2172 so = ((file_t *)kn->kn_obj)->f_data; 2173 2174 /* 2175 * Set kn_data to number of incoming connections, not 2176 * counting partial (incomplete) connections. 2177 */ 2178 if (hint != NOTE_SUBMIT) 2179 solock(so); 2180 kn->kn_data = so->so_qlen; 2181 rv = (kn->kn_data > 0); 2182 if (hint != NOTE_SUBMIT) 2183 sounlock(so); 2184 return rv; 2185 } 2186 2187 static const struct filterops solisten_filtops = 2188 { 1, NULL, filt_sordetach, filt_solisten }; 2189 static const struct filterops soread_filtops = 2190 { 1, NULL, filt_sordetach, filt_soread }; 2191 static const struct filterops sowrite_filtops = 2192 { 1, NULL, filt_sowdetach, filt_sowrite }; 2193 2194 int 2195 soo_kqfilter(struct file *fp, struct knote *kn) 2196 { 2197 struct socket *so; 2198 struct sockbuf *sb; 2199 2200 so = ((file_t *)kn->kn_obj)->f_data; 2201 solock(so); 2202 switch (kn->kn_filter) { 2203 case EVFILT_READ: 2204 if (so->so_options & SO_ACCEPTCONN) 2205 kn->kn_fop = &solisten_filtops; 2206 else 2207 kn->kn_fop = &soread_filtops; 2208 sb = &so->so_rcv; 2209 break; 2210 case EVFILT_WRITE: 2211 kn->kn_fop = &sowrite_filtops; 2212 sb = &so->so_snd; 2213 break; 2214 default: 2215 sounlock(so); 2216 return (EINVAL); 2217 } 2218 SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext); 2219 sb->sb_flags |= SB_KNOTE; 2220 sounlock(so); 2221 return (0); 2222 } 2223 2224 static int 2225 sodopoll(struct socket *so, int events) 2226 { 2227 int revents; 2228 2229 revents = 0; 2230 2231 if (events & (POLLIN | POLLRDNORM)) 2232 if (soreadable(so)) 2233 revents |= events & (POLLIN | POLLRDNORM); 2234 2235 if (events & (POLLOUT | POLLWRNORM)) 2236 if (sowritable(so)) 2237 revents |= events & (POLLOUT | POLLWRNORM); 2238 2239 if (events & (POLLPRI | POLLRDBAND)) 2240 if (so->so_oobmark || (so->so_state & SS_RCVATMARK)) 2241 revents |= events & (POLLPRI | POLLRDBAND); 2242 2243 return revents; 2244 } 2245 2246 int 2247 sopoll(struct socket *so, int events) 2248 { 2249 int revents = 0; 2250 2251 #ifndef DIAGNOSTIC 2252 /* 2253 * Do a quick, unlocked check in expectation that the socket 2254 * will be ready for I/O. Don't do this check if DIAGNOSTIC, 2255 * as the solocked() assertions will fail. 2256 */ 2257 if ((revents = sodopoll(so, events)) != 0) 2258 return revents; 2259 #endif 2260 2261 solock(so); 2262 if ((revents = sodopoll(so, events)) == 0) { 2263 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 2264 selrecord(curlwp, &so->so_rcv.sb_sel); 2265 so->so_rcv.sb_flags |= SB_NOTIFY; 2266 } 2267 2268 if (events & (POLLOUT | POLLWRNORM)) { 2269 selrecord(curlwp, &so->so_snd.sb_sel); 2270 so->so_snd.sb_flags |= SB_NOTIFY; 2271 } 2272 } 2273 sounlock(so); 2274 2275 return revents; 2276 } 2277 2278 2279 #include <sys/sysctl.h> 2280 2281 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO); 2282 2283 /* 2284 * sysctl helper routine for kern.somaxkva. ensures that the given 2285 * value is not too small. 2286 * (XXX should we maybe make sure it's not too large as well?) 2287 */ 2288 static int 2289 sysctl_kern_somaxkva(SYSCTLFN_ARGS) 2290 { 2291 int error, new_somaxkva; 2292 struct sysctlnode node; 2293 2294 new_somaxkva = somaxkva; 2295 node = *rnode; 2296 node.sysctl_data = &new_somaxkva; 2297 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2298 if (error || newp == NULL) 2299 return (error); 2300 2301 if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */ 2302 return (EINVAL); 2303 2304 mutex_enter(&so_pendfree_lock); 2305 somaxkva = new_somaxkva; 2306 cv_broadcast(&socurkva_cv); 2307 mutex_exit(&so_pendfree_lock); 2308 2309 return (error); 2310 } 2311 2312 static void 2313 sysctl_kern_somaxkva_setup(void) 2314 { 2315 2316 KASSERT(socket_sysctllog == NULL); 2317 sysctl_createv(&socket_sysctllog, 0, NULL, NULL, 2318 CTLFLAG_PERMANENT, 2319 CTLTYPE_NODE, "kern", NULL, 2320 NULL, 0, NULL, 0, 2321 CTL_KERN, CTL_EOL); 2322 2323 sysctl_createv(&socket_sysctllog, 0, NULL, NULL, 2324 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2325 CTLTYPE_INT, "somaxkva", 2326 SYSCTL_DESCR("Maximum amount of kernel memory to be " 2327 "used for socket buffers"), 2328 sysctl_kern_somaxkva, 0, NULL, 0, 2329 CTL_KERN, KERN_SOMAXKVA, CTL_EOL); 2330 } 2331