uipc_socket.c revision 131005
1/* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 30 */ 31 32#include <sys/cdefs.h> 33__FBSDID("$FreeBSD: head/sys/kern/uipc_socket.c 131005 2004-06-24 00:54:26Z rwatson $"); 34 35#include "opt_inet.h" 36#include "opt_mac.h" 37#include "opt_zero.h" 38 39#include <sys/param.h> 40#include <sys/systm.h> 41#include <sys/fcntl.h> 42#include <sys/limits.h> 43#include <sys/lock.h> 44#include <sys/mac.h> 45#include <sys/malloc.h> 46#include <sys/mbuf.h> 47#include <sys/mutex.h> 48#include <sys/domain.h> 49#include <sys/file.h> /* for struct knote */ 50#include <sys/kernel.h> 51#include <sys/event.h> 52#include <sys/poll.h> 53#include <sys/proc.h> 54#include <sys/protosw.h> 55#include <sys/socket.h> 56#include <sys/socketvar.h> 57#include <sys/resourcevar.h> 58#include <sys/signalvar.h> 59#include <sys/sysctl.h> 60#include <sys/uio.h> 61#include <sys/jail.h> 62 63#include <vm/uma.h> 64 65 66#ifdef INET 67static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt); 68#endif 69 70static void filt_sordetach(struct knote *kn); 71static int filt_soread(struct knote *kn, long hint); 72static void filt_sowdetach(struct knote *kn); 73static int filt_sowrite(struct knote *kn, long hint); 74static int filt_solisten(struct knote *kn, long hint); 75 76static struct filterops solisten_filtops = 77 { 1, NULL, filt_sordetach, filt_solisten }; 78static struct filterops soread_filtops = 79 { 1, NULL, filt_sordetach, filt_soread }; 80static struct filterops sowrite_filtops = 81 { 1, NULL, filt_sowdetach, filt_sowrite }; 82 83uma_zone_t socket_zone; 84so_gen_t so_gencnt; /* generation count for sockets */ 85 86MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 87MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 88 89SYSCTL_DECL(_kern_ipc); 90 91static int somaxconn = SOMAXCONN; 92SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW, 93 &somaxconn, 0, "Maximum pending socket connection queue size"); 94static int numopensockets; 95SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD, 96 &numopensockets, 0, "Number of open sockets"); 97#ifdef ZERO_COPY_SOCKETS 98/* These aren't static because they're used in other files. */ 99int so_zero_copy_send = 1; 100int so_zero_copy_receive = 1; 101SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0, 102 "Zero copy controls"); 103SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW, 104 &so_zero_copy_receive, 0, "Enable zero copy receive"); 105SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW, 106 &so_zero_copy_send, 0, "Enable zero copy send"); 107#endif /* ZERO_COPY_SOCKETS */ 108 109struct mtx accept_mtx; 110MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF); 111 112 113/* 114 * Socket operation routines. 115 * These routines are called by the routines in 116 * sys_socket.c or from a system process, and 117 * implement the semantics of socket operations by 118 * switching out to the protocol specific routines. 119 */ 120 121/* 122 * Get a socket structure from our zone, and initialize it. 123 * Note that it would probably be better to allocate socket 124 * and PCB at the same time, but I'm not convinced that all 125 * the protocols can be easily modified to do this. 126 * 127 * soalloc() returns a socket with a ref count of 0. 128 */ 129struct socket * 130soalloc(int mflags) 131{ 132 struct socket *so; 133#ifdef MAC 134 int error; 135#endif 136 137 so = uma_zalloc(socket_zone, mflags | M_ZERO); 138 if (so != NULL) { 139#ifdef MAC 140 error = mac_init_socket(so, mflags); 141 if (error != 0) { 142 uma_zfree(socket_zone, so); 143 so = NULL; 144 return so; 145 } 146#endif 147 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd"); 148 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv"); 149 /* XXX race condition for reentrant kernel */ 150 so->so_gencnt = ++so_gencnt; 151 /* sx_init(&so->so_sxlock, "socket sxlock"); */ 152 TAILQ_INIT(&so->so_aiojobq); 153 ++numopensockets; 154 } 155 return so; 156} 157 158/* 159 * socreate returns a socket with a ref count of 1. The socket should be 160 * closed with soclose(). 161 */ 162int 163socreate(dom, aso, type, proto, cred, td) 164 int dom; 165 struct socket **aso; 166 int type; 167 int proto; 168 struct ucred *cred; 169 struct thread *td; 170{ 171 struct protosw *prp; 172 struct socket *so; 173 int error; 174 175 if (proto) 176 prp = pffindproto(dom, proto, type); 177 else 178 prp = pffindtype(dom, type); 179 180 if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL) 181 return (EPROTONOSUPPORT); 182 183 if (jailed(cred) && jail_socket_unixiproute_only && 184 prp->pr_domain->dom_family != PF_LOCAL && 185 prp->pr_domain->dom_family != PF_INET && 186 prp->pr_domain->dom_family != PF_ROUTE) { 187 return (EPROTONOSUPPORT); 188 } 189 190 if (prp->pr_type != type) 191 return (EPROTOTYPE); 192 so = soalloc(M_WAITOK); 193 if (so == NULL) 194 return (ENOBUFS); 195 196 TAILQ_INIT(&so->so_incomp); 197 TAILQ_INIT(&so->so_comp); 198 so->so_type = type; 199 so->so_cred = crhold(cred); 200 so->so_proto = prp; 201#ifdef MAC 202 mac_create_socket(cred, so); 203#endif 204 SOCK_LOCK(so); 205 soref(so); 206 SOCK_UNLOCK(so); 207 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); 208 if (error) { 209 SOCK_LOCK(so); 210 so->so_state |= SS_NOFDREF; 211 sorele(so); 212 return (error); 213 } 214 *aso = so; 215 return (0); 216} 217 218int 219sobind(so, nam, td) 220 struct socket *so; 221 struct sockaddr *nam; 222 struct thread *td; 223{ 224 225 return ((*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td)); 226} 227 228void 229sodealloc(struct socket *so) 230{ 231 232 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); 233 so->so_gencnt = ++so_gencnt; 234 if (so->so_rcv.sb_hiwat) 235 (void)chgsbsize(so->so_cred->cr_uidinfo, 236 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 237 if (so->so_snd.sb_hiwat) 238 (void)chgsbsize(so->so_cred->cr_uidinfo, 239 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 240#ifdef INET 241 /* remove acccept filter if one is present. */ 242 if (so->so_accf != NULL) 243 do_setopt_accept_filter(so, NULL); 244#endif 245#ifdef MAC 246 mac_destroy_socket(so); 247#endif 248 crfree(so->so_cred); 249 SOCKBUF_LOCK_DESTROY(&so->so_snd); 250 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 251 /* sx_destroy(&so->so_sxlock); */ 252 uma_zfree(socket_zone, so); 253 --numopensockets; 254} 255 256int 257solisten(so, backlog, td) 258 struct socket *so; 259 int backlog; 260 struct thread *td; 261{ 262 int error; 263 264 /* 265 * XXXRW: Ordering issue here -- perhaps we need to set 266 * SO_ACCEPTCONN before the call to pru_listen()? 267 * XXXRW: General atomic test-and-set concerns here also. 268 */ 269 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 270 SS_ISDISCONNECTING)) 271 return (EINVAL); 272 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, td); 273 if (error) 274 return (error); 275 ACCEPT_LOCK(); 276 if (TAILQ_EMPTY(&so->so_comp)) { 277 SOCK_LOCK(so); 278 so->so_options |= SO_ACCEPTCONN; 279 SOCK_UNLOCK(so); 280 } 281 if (backlog < 0 || backlog > somaxconn) 282 backlog = somaxconn; 283 so->so_qlimit = backlog; 284 ACCEPT_UNLOCK(); 285 return (0); 286} 287 288void 289sofree(so) 290 struct socket *so; 291{ 292 struct socket *head; 293 294 KASSERT(so->so_count == 0, ("socket %p so_count not 0", so)); 295 SOCK_LOCK_ASSERT(so); 296 297 if (so->so_pcb != NULL || (so->so_state & SS_NOFDREF) == 0) { 298 SOCK_UNLOCK(so); 299 return; 300 } 301 302 SOCK_UNLOCK(so); 303 ACCEPT_LOCK(); 304 head = so->so_head; 305 if (head != NULL) { 306 KASSERT((so->so_qstate & SQ_COMP) != 0 || 307 (so->so_qstate & SQ_INCOMP) != 0, 308 ("sofree: so_head != NULL, but neither SQ_COMP nor " 309 "SQ_INCOMP")); 310 KASSERT((so->so_qstate & SQ_COMP) == 0 || 311 (so->so_qstate & SQ_INCOMP) == 0, 312 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP")); 313 /* 314 * accept(2) is responsible draining the completed 315 * connection queue and freeing those sockets, so 316 * we just return here if this socket is currently 317 * on the completed connection queue. Otherwise, 318 * accept(2) may hang after select(2) has indicating 319 * that a listening socket was ready. If it's an 320 * incomplete connection, we remove it from the queue 321 * and free it; otherwise, it won't be released until 322 * the listening socket is closed. 323 */ 324 if ((so->so_qstate & SQ_COMP) != 0) { 325 ACCEPT_UNLOCK(); 326 return; 327 } 328 TAILQ_REMOVE(&head->so_incomp, so, so_list); 329 head->so_incqlen--; 330 so->so_qstate &= ~SQ_INCOMP; 331 so->so_head = NULL; 332 } 333 KASSERT((so->so_qstate & SQ_COMP) == 0 && 334 (so->so_qstate & SQ_INCOMP) == 0, 335 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)", 336 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP)); 337 ACCEPT_UNLOCK(); 338 SOCKBUF_LOCK(&so->so_snd); 339 so->so_snd.sb_flags |= SB_NOINTR; 340 (void)sblock(&so->so_snd, M_WAITOK); 341 /* 342 * socantsendmore_locked() drops the socket buffer mutex so that it 343 * can safely perform wakeups. Re-acquire the mutex before 344 * continuing. 345 */ 346 socantsendmore_locked(so); 347 SOCKBUF_LOCK(&so->so_snd); 348 sbunlock(&so->so_snd); 349 sbrelease_locked(&so->so_snd, so); 350 SOCKBUF_UNLOCK(&so->so_snd); 351 sorflush(so); 352 sodealloc(so); 353} 354 355/* 356 * Close a socket on last file table reference removal. 357 * Initiate disconnect if connected. 358 * Free socket when disconnect complete. 359 * 360 * This function will sorele() the socket. Note that soclose() may be 361 * called prior to the ref count reaching zero. The actual socket 362 * structure will not be freed until the ref count reaches zero. 363 */ 364int 365soclose(so) 366 struct socket *so; 367{ 368 int error = 0; 369 370 funsetown(&so->so_sigio); 371 if (so->so_options & SO_ACCEPTCONN) { 372 struct socket *sp; 373 ACCEPT_LOCK(); 374 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) { 375 TAILQ_REMOVE(&so->so_incomp, sp, so_list); 376 so->so_incqlen--; 377 sp->so_qstate &= ~SQ_INCOMP; 378 sp->so_head = NULL; 379 ACCEPT_UNLOCK(); 380 (void) soabort(sp); 381 ACCEPT_LOCK(); 382 } 383 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) { 384 TAILQ_REMOVE(&so->so_comp, sp, so_list); 385 so->so_qlen--; 386 sp->so_qstate &= ~SQ_COMP; 387 sp->so_head = NULL; 388 ACCEPT_UNLOCK(); 389 (void) soabort(sp); 390 ACCEPT_LOCK(); 391 } 392 ACCEPT_UNLOCK(); 393 } 394 if (so->so_pcb == NULL) 395 goto discard; 396 if (so->so_state & SS_ISCONNECTED) { 397 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 398 error = sodisconnect(so); 399 if (error) 400 goto drop; 401 } 402 if (so->so_options & SO_LINGER) { 403 if ((so->so_state & SS_ISDISCONNECTING) && 404 (so->so_state & SS_NBIO)) 405 goto drop; 406 while (so->so_state & SS_ISCONNECTED) { 407 error = tsleep(&so->so_timeo, 408 PSOCK | PCATCH, "soclos", so->so_linger * hz); 409 if (error) 410 break; 411 } 412 } 413 } 414drop: 415 if (so->so_pcb != NULL) { 416 int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so); 417 if (error == 0) 418 error = error2; 419 } 420discard: 421 SOCK_LOCK(so); 422 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF")); 423 so->so_state |= SS_NOFDREF; 424 sorele(so); 425 return (error); 426} 427 428/* 429 * Must be called at splnet... 430 */ 431int 432soabort(so) 433 struct socket *so; 434{ 435 int error; 436 437 error = (*so->so_proto->pr_usrreqs->pru_abort)(so); 438 if (error) { 439 SOCK_LOCK(so); 440 sotryfree(so); /* note: does not decrement the ref count */ 441 return error; 442 } 443 return (0); 444} 445 446int 447soaccept(so, nam) 448 struct socket *so; 449 struct sockaddr **nam; 450{ 451 int error; 452 453 SOCK_LOCK(so); 454 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF")); 455 so->so_state &= ~SS_NOFDREF; 456 SOCK_UNLOCK(so); 457 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); 458 return (error); 459} 460 461int 462soconnect(so, nam, td) 463 struct socket *so; 464 struct sockaddr *nam; 465 struct thread *td; 466{ 467 int error; 468 469 if (so->so_options & SO_ACCEPTCONN) 470 return (EOPNOTSUPP); 471 /* 472 * If protocol is connection-based, can only connect once. 473 * Otherwise, if connected, try to disconnect first. 474 * This allows user to disconnect by connecting to, e.g., 475 * a null address. 476 */ 477 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 478 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 479 (error = sodisconnect(so)))) 480 error = EISCONN; 481 else 482 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td); 483 return (error); 484} 485 486int 487soconnect2(so1, so2) 488 struct socket *so1; 489 struct socket *so2; 490{ 491 492 return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2)); 493} 494 495int 496sodisconnect(so) 497 struct socket *so; 498{ 499 int error; 500 501 if ((so->so_state & SS_ISCONNECTED) == 0) 502 return (ENOTCONN); 503 if (so->so_state & SS_ISDISCONNECTING) 504 return (EALREADY); 505 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 506 return (error); 507} 508 509#define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK) 510/* 511 * Send on a socket. 512 * If send must go all at once and message is larger than 513 * send buffering, then hard error. 514 * Lock against other senders. 515 * If must go all at once and not enough room now, then 516 * inform user that this would block and do nothing. 517 * Otherwise, if nonblocking, send as much as possible. 518 * The data to be sent is described by "uio" if nonzero, 519 * otherwise by the mbuf chain "top" (which must be null 520 * if uio is not). Data provided in mbuf chain must be small 521 * enough to send all at once. 522 * 523 * Returns nonzero on error, timeout or signal; callers 524 * must check for short counts if EINTR/ERESTART are returned. 525 * Data and control buffers are freed on return. 526 */ 527 528#ifdef ZERO_COPY_SOCKETS 529struct so_zerocopy_stats{ 530 int size_ok; 531 int align_ok; 532 int found_ifp; 533}; 534struct so_zerocopy_stats so_zerocp_stats = {0,0,0}; 535#include <netinet/in.h> 536#include <net/route.h> 537#include <netinet/in_pcb.h> 538#include <vm/vm.h> 539#include <vm/vm_page.h> 540#include <vm/vm_object.h> 541#endif /*ZERO_COPY_SOCKETS*/ 542 543int 544sosend(so, addr, uio, top, control, flags, td) 545 struct socket *so; 546 struct sockaddr *addr; 547 struct uio *uio; 548 struct mbuf *top; 549 struct mbuf *control; 550 int flags; 551 struct thread *td; 552{ 553 struct mbuf **mp; 554 struct mbuf *m; 555 long space, len = 0, resid; 556 int clen = 0, error, dontroute; 557 int atomic = sosendallatonce(so) || top; 558#ifdef ZERO_COPY_SOCKETS 559 int cow_send; 560#endif /* ZERO_COPY_SOCKETS */ 561 562 if (uio != NULL) 563 resid = uio->uio_resid; 564 else 565 resid = top->m_pkthdr.len; 566 /* 567 * In theory resid should be unsigned. 568 * However, space must be signed, as it might be less than 0 569 * if we over-committed, and we must use a signed comparison 570 * of space and resid. On the other hand, a negative resid 571 * causes us to loop sending 0-length segments to the protocol. 572 * 573 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 574 * type sockets since that's an error. 575 */ 576 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 577 error = EINVAL; 578 goto out; 579 } 580 581 dontroute = 582 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 583 (so->so_proto->pr_flags & PR_ATOMIC); 584 if (td != NULL) 585 td->td_proc->p_stats->p_ru.ru_msgsnd++; 586 if (control != NULL) 587 clen = control->m_len; 588#define snderr(errno) { error = (errno); goto release; } 589 590 SOCKBUF_LOCK(&so->so_snd); 591restart: 592 SOCKBUF_LOCK_ASSERT(&so->so_snd); 593 error = sblock(&so->so_snd, SBLOCKWAIT(flags)); 594 if (error) 595 goto out_locked; 596 do { 597 SOCKBUF_LOCK_ASSERT(&so->so_snd); 598 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 599 snderr(EPIPE); 600 if (so->so_error) { 601 error = so->so_error; 602 so->so_error = 0; 603 goto release; 604 } 605 if ((so->so_state & SS_ISCONNECTED) == 0) { 606 /* 607 * `sendto' and `sendmsg' is allowed on a connection- 608 * based socket if it supports implied connect. 609 * Return ENOTCONN if not connected and no address is 610 * supplied. 611 */ 612 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 613 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 614 if ((so->so_state & SS_ISCONFIRMING) == 0 && 615 !(resid == 0 && clen != 0)) 616 snderr(ENOTCONN); 617 } else if (addr == NULL) 618 snderr(so->so_proto->pr_flags & PR_CONNREQUIRED ? 619 ENOTCONN : EDESTADDRREQ); 620 } 621 space = sbspace(&so->so_snd); 622 if (flags & MSG_OOB) 623 space += 1024; 624 if ((atomic && resid > so->so_snd.sb_hiwat) || 625 clen > so->so_snd.sb_hiwat) 626 snderr(EMSGSIZE); 627 if (space < resid + clen && 628 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 629 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) 630 snderr(EWOULDBLOCK); 631 sbunlock(&so->so_snd); 632 error = sbwait(&so->so_snd); 633 if (error) 634 goto out_locked; 635 goto restart; 636 } 637 SOCKBUF_UNLOCK(&so->so_snd); 638 mp = ⊤ 639 space -= clen; 640 do { 641 if (uio == NULL) { 642 /* 643 * Data is prepackaged in "top". 644 */ 645 resid = 0; 646 if (flags & MSG_EOR) 647 top->m_flags |= M_EOR; 648 } else do { 649#ifdef ZERO_COPY_SOCKETS 650 cow_send = 0; 651#endif /* ZERO_COPY_SOCKETS */ 652 if (resid >= MINCLSIZE) { 653#ifdef ZERO_COPY_SOCKETS 654 if (top == NULL) { 655 MGETHDR(m, M_TRYWAIT, MT_DATA); 656 if (m == NULL) { 657 error = ENOBUFS; 658 SOCKBUF_LOCK(&so->so_snd); 659 goto release; 660 } 661 m->m_pkthdr.len = 0; 662 m->m_pkthdr.rcvif = (struct ifnet *)0; 663 } else { 664 MGET(m, M_TRYWAIT, MT_DATA); 665 if (m == NULL) { 666 error = ENOBUFS; 667 SOCKBUF_LOCK(&so->so_snd); 668 goto release; 669 } 670 } 671 if (so_zero_copy_send && 672 resid>=PAGE_SIZE && 673 space>=PAGE_SIZE && 674 uio->uio_iov->iov_len>=PAGE_SIZE) { 675 so_zerocp_stats.size_ok++; 676 if (!((vm_offset_t) 677 uio->uio_iov->iov_base & PAGE_MASK)){ 678 so_zerocp_stats.align_ok++; 679 cow_send = socow_setup(m, uio); 680 } 681 } 682 if (!cow_send) { 683 MCLGET(m, M_TRYWAIT); 684 if ((m->m_flags & M_EXT) == 0) { 685 m_free(m); 686 m = NULL; 687 } else { 688 len = min(min(MCLBYTES, resid), space); 689 } 690 } else 691 len = PAGE_SIZE; 692#else /* ZERO_COPY_SOCKETS */ 693 if (top == NULL) { 694 m = m_getcl(M_TRYWAIT, MT_DATA, M_PKTHDR); 695 m->m_pkthdr.len = 0; 696 m->m_pkthdr.rcvif = (struct ifnet *)0; 697 } else 698 m = m_getcl(M_TRYWAIT, MT_DATA, 0); 699 len = min(min(MCLBYTES, resid), space); 700#endif /* ZERO_COPY_SOCKETS */ 701 } else { 702 if (top == NULL) { 703 m = m_gethdr(M_TRYWAIT, MT_DATA); 704 m->m_pkthdr.len = 0; 705 m->m_pkthdr.rcvif = (struct ifnet *)0; 706 707 len = min(min(MHLEN, resid), space); 708 /* 709 * For datagram protocols, leave room 710 * for protocol headers in first mbuf. 711 */ 712 if (atomic && m && len < MHLEN) 713 MH_ALIGN(m, len); 714 } else { 715 m = m_get(M_TRYWAIT, MT_DATA); 716 len = min(min(MLEN, resid), space); 717 } 718 } 719 if (m == NULL) { 720 error = ENOBUFS; 721 SOCKBUF_LOCK(&so->so_snd); 722 goto release; 723 } 724 725 space -= len; 726#ifdef ZERO_COPY_SOCKETS 727 if (cow_send) 728 error = 0; 729 else 730#endif /* ZERO_COPY_SOCKETS */ 731 error = uiomove(mtod(m, void *), (int)len, uio); 732 resid = uio->uio_resid; 733 m->m_len = len; 734 *mp = m; 735 top->m_pkthdr.len += len; 736 if (error) { 737 SOCKBUF_LOCK(&so->so_snd); 738 goto release; 739 } 740 mp = &m->m_next; 741 if (resid <= 0) { 742 if (flags & MSG_EOR) 743 top->m_flags |= M_EOR; 744 break; 745 } 746 } while (space > 0 && atomic); 747 if (dontroute) { 748 SOCK_LOCK(so); 749 so->so_options |= SO_DONTROUTE; 750 SOCK_UNLOCK(so); 751 } 752 /* 753 * XXX all the SBS_CANTSENDMORE checks previously 754 * done could be out of date. We could have recieved 755 * a reset packet in an interrupt or maybe we slept 756 * while doing page faults in uiomove() etc. We could 757 * probably recheck again inside the splnet() protection 758 * here, but there are probably other places that this 759 * also happens. We must rethink this. 760 */ 761 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 762 (flags & MSG_OOB) ? PRUS_OOB : 763 /* 764 * If the user set MSG_EOF, the protocol 765 * understands this flag and nothing left to 766 * send then use PRU_SEND_EOF instead of PRU_SEND. 767 */ 768 ((flags & MSG_EOF) && 769 (so->so_proto->pr_flags & PR_IMPLOPCL) && 770 (resid <= 0)) ? 771 PRUS_EOF : 772 /* If there is more to send set PRUS_MORETOCOME */ 773 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 774 top, addr, control, td); 775 if (dontroute) { 776 SOCK_LOCK(so); 777 so->so_options &= ~SO_DONTROUTE; 778 SOCK_UNLOCK(so); 779 } 780 clen = 0; 781 control = NULL; 782 top = NULL; 783 mp = ⊤ 784 if (error) { 785 SOCKBUF_LOCK(&so->so_snd); 786 goto release; 787 } 788 } while (resid && space > 0); 789 SOCKBUF_LOCK(&so->so_snd); 790 } while (resid); 791 792release: 793 SOCKBUF_LOCK_ASSERT(&so->so_snd); 794 sbunlock(&so->so_snd); 795out_locked: 796 SOCKBUF_LOCK_ASSERT(&so->so_snd); 797 SOCKBUF_UNLOCK(&so->so_snd); 798out: 799 if (top != NULL) 800 m_freem(top); 801 if (control != NULL) 802 m_freem(control); 803 return (error); 804} 805 806/* 807 * Implement receive operations on a socket. 808 * We depend on the way that records are added to the sockbuf 809 * by sbappend*. In particular, each record (mbufs linked through m_next) 810 * must begin with an address if the protocol so specifies, 811 * followed by an optional mbuf or mbufs containing ancillary data, 812 * and then zero or more mbufs of data. 813 * In order to avoid blocking network interrupts for the entire time here, 814 * we splx() while doing the actual copy to user space. 815 * Although the sockbuf is locked, new data may still be appended, 816 * and thus we must maintain consistency of the sockbuf during that time. 817 * 818 * The caller may receive the data as a single mbuf chain by supplying 819 * an mbuf **mp0 for use in returning the chain. The uio is then used 820 * only for the count in uio_resid. 821 */ 822int 823soreceive(so, psa, uio, mp0, controlp, flagsp) 824 struct socket *so; 825 struct sockaddr **psa; 826 struct uio *uio; 827 struct mbuf **mp0; 828 struct mbuf **controlp; 829 int *flagsp; 830{ 831 struct mbuf *m, **mp; 832 int flags, len, error, offset; 833 struct protosw *pr = so->so_proto; 834 struct mbuf *nextrecord; 835 int moff, type = 0; 836 int orig_resid = uio->uio_resid; 837 838 mp = mp0; 839 if (psa != NULL) 840 *psa = 0; 841 if (controlp != NULL) 842 *controlp = 0; 843 if (flagsp != NULL) 844 flags = *flagsp &~ MSG_EOR; 845 else 846 flags = 0; 847 if (flags & MSG_OOB) { 848 m = m_get(M_TRYWAIT, MT_DATA); 849 if (m == NULL) 850 return (ENOBUFS); 851 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 852 if (error) 853 goto bad; 854 do { 855#ifdef ZERO_COPY_SOCKETS 856 if (so_zero_copy_receive) { 857 vm_page_t pg; 858 int disposable; 859 860 if ((m->m_flags & M_EXT) 861 && (m->m_ext.ext_type == EXT_DISPOSABLE)) 862 disposable = 1; 863 else 864 disposable = 0; 865 866 pg = PHYS_TO_VM_PAGE(vtophys(mtod(m, caddr_t))); 867 if (uio->uio_offset == -1) 868 uio->uio_offset =IDX_TO_OFF(pg->pindex); 869 870 error = uiomoveco(mtod(m, void *), 871 min(uio->uio_resid, m->m_len), 872 uio, pg->object, 873 disposable); 874 } else 875#endif /* ZERO_COPY_SOCKETS */ 876 error = uiomove(mtod(m, void *), 877 (int) min(uio->uio_resid, m->m_len), uio); 878 m = m_free(m); 879 } while (uio->uio_resid && error == 0 && m); 880bad: 881 if (m != NULL) 882 m_freem(m); 883 return (error); 884 } 885 if (mp != NULL) 886 *mp = NULL; 887 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) 888 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 889 890 SOCKBUF_LOCK(&so->so_rcv); 891restart: 892 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 893 error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); 894 if (error) 895 goto out; 896 897 m = so->so_rcv.sb_mb; 898 /* 899 * If we have less data than requested, block awaiting more 900 * (subject to any timeout) if: 901 * 1. the current count is less than the low water mark, or 902 * 2. MSG_WAITALL is set, and it is possible to do the entire 903 * receive operation at once if we block (resid <= hiwat). 904 * 3. MSG_DONTWAIT is not set 905 * If MSG_WAITALL is set but resid is larger than the receive buffer, 906 * we have to do the receive in sections, and thus risk returning 907 * a short count if a timeout or signal occurs after we start. 908 */ 909 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 910 so->so_rcv.sb_cc < uio->uio_resid) && 911 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 912 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 913 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 914 KASSERT(m != NULL || !so->so_rcv.sb_cc, 915 ("receive: m == %p so->so_rcv.sb_cc == %u", 916 m, so->so_rcv.sb_cc)); 917 if (so->so_error) { 918 if (m != NULL) 919 goto dontblock; 920 error = so->so_error; 921 if ((flags & MSG_PEEK) == 0) 922 so->so_error = 0; 923 goto release; 924 } 925 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 926 if (m) 927 goto dontblock; 928 else 929 goto release; 930 } 931 for (; m != NULL; m = m->m_next) 932 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 933 m = so->so_rcv.sb_mb; 934 goto dontblock; 935 } 936 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 937 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 938 error = ENOTCONN; 939 goto release; 940 } 941 if (uio->uio_resid == 0) 942 goto release; 943 if ((so->so_state & SS_NBIO) || 944 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 945 error = EWOULDBLOCK; 946 goto release; 947 } 948 SBLASTRECORDCHK(&so->so_rcv); 949 SBLASTMBUFCHK(&so->so_rcv); 950 sbunlock(&so->so_rcv); 951 error = sbwait(&so->so_rcv); 952 if (error) 953 goto out; 954 goto restart; 955 } 956dontblock: 957 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 958 if (uio->uio_td) 959 uio->uio_td->td_proc->p_stats->p_ru.ru_msgrcv++; 960 SBLASTRECORDCHK(&so->so_rcv); 961 SBLASTMBUFCHK(&so->so_rcv); 962 nextrecord = m->m_nextpkt; 963 if (pr->pr_flags & PR_ADDR) { 964 KASSERT(m->m_type == MT_SONAME, 965 ("m->m_type == %d", m->m_type)); 966 orig_resid = 0; 967 if (psa != NULL) 968 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 969 M_NOWAIT); 970 if (flags & MSG_PEEK) { 971 m = m->m_next; 972 } else { 973 sbfree(&so->so_rcv, m); 974 so->so_rcv.sb_mb = m_free(m); 975 m = so->so_rcv.sb_mb; 976 } 977 } 978 while (m != NULL && m->m_type == MT_CONTROL && error == 0) { 979 if (flags & MSG_PEEK) { 980 if (controlp != NULL) 981 *controlp = m_copy(m, 0, m->m_len); 982 m = m->m_next; 983 } else { 984 sbfree(&so->so_rcv, m); 985 so->so_rcv.sb_mb = m->m_next; 986 m->m_next = NULL; 987 if (pr->pr_domain->dom_externalize) { 988 SOCKBUF_UNLOCK(&so->so_rcv); 989 error = (*pr->pr_domain->dom_externalize) 990 (m, controlp); 991 SOCKBUF_LOCK(&so->so_rcv); 992 } else if (controlp != NULL) 993 *controlp = m; 994 else 995 m_freem(m); 996 m = so->so_rcv.sb_mb; 997 } 998 if (controlp != NULL) { 999 orig_resid = 0; 1000 while (*controlp != NULL) 1001 controlp = &(*controlp)->m_next; 1002 } 1003 } 1004 if (m != NULL) { 1005 if ((flags & MSG_PEEK) == 0) { 1006 m->m_nextpkt = nextrecord; 1007 /* 1008 * If nextrecord == NULL (this is a single chain), 1009 * then sb_lastrecord may not be valid here if m 1010 * was changed earlier. 1011 */ 1012 if (nextrecord == NULL) { 1013 KASSERT(so->so_rcv.sb_mb == m, 1014 ("receive tailq 1")); 1015 so->so_rcv.sb_lastrecord = m; 1016 } 1017 } 1018 type = m->m_type; 1019 if (type == MT_OOBDATA) 1020 flags |= MSG_OOB; 1021 } else { 1022 if ((flags & MSG_PEEK) == 0) { 1023 KASSERT(so->so_rcv.sb_mb == m,("receive tailq 2")); 1024 so->so_rcv.sb_mb = nextrecord; 1025 SB_EMPTY_FIXUP(&so->so_rcv); 1026 } 1027 } 1028 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1029 SBLASTRECORDCHK(&so->so_rcv); 1030 SBLASTMBUFCHK(&so->so_rcv); 1031 1032 moff = 0; 1033 offset = 0; 1034 while (m != NULL && uio->uio_resid > 0 && error == 0) { 1035 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1036 if (m->m_type == MT_OOBDATA) { 1037 if (type != MT_OOBDATA) 1038 break; 1039 } else if (type == MT_OOBDATA) 1040 break; 1041 else 1042 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER, 1043 ("m->m_type == %d", m->m_type)); 1044 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 1045 len = uio->uio_resid; 1046 if (so->so_oobmark && len > so->so_oobmark - offset) 1047 len = so->so_oobmark - offset; 1048 if (len > m->m_len - moff) 1049 len = m->m_len - moff; 1050 /* 1051 * If mp is set, just pass back the mbufs. 1052 * Otherwise copy them out via the uio, then free. 1053 * Sockbuf must be consistent here (points to current mbuf, 1054 * it points to next record) when we drop priority; 1055 * we must note any additions to the sockbuf when we 1056 * block interrupts again. 1057 */ 1058 if (mp == NULL) { 1059 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1060 SBLASTRECORDCHK(&so->so_rcv); 1061 SBLASTMBUFCHK(&so->so_rcv); 1062 SOCKBUF_UNLOCK(&so->so_rcv); 1063#ifdef ZERO_COPY_SOCKETS 1064 if (so_zero_copy_receive) { 1065 vm_page_t pg; 1066 int disposable; 1067 1068 if ((m->m_flags & M_EXT) 1069 && (m->m_ext.ext_type == EXT_DISPOSABLE)) 1070 disposable = 1; 1071 else 1072 disposable = 0; 1073 1074 pg = PHYS_TO_VM_PAGE(vtophys(mtod(m, caddr_t) + 1075 moff)); 1076 1077 if (uio->uio_offset == -1) 1078 uio->uio_offset =IDX_TO_OFF(pg->pindex); 1079 1080 error = uiomoveco(mtod(m, char *) + moff, 1081 (int)len, uio,pg->object, 1082 disposable); 1083 } else 1084#endif /* ZERO_COPY_SOCKETS */ 1085 error = uiomove(mtod(m, char *) + moff, (int)len, uio); 1086 SOCKBUF_LOCK(&so->so_rcv); 1087 if (error) 1088 goto release; 1089 } else 1090 uio->uio_resid -= len; 1091 if (len == m->m_len - moff) { 1092 if (m->m_flags & M_EOR) 1093 flags |= MSG_EOR; 1094 if (flags & MSG_PEEK) { 1095 m = m->m_next; 1096 moff = 0; 1097 } else { 1098 nextrecord = m->m_nextpkt; 1099 sbfree(&so->so_rcv, m); 1100 if (mp != NULL) { 1101 *mp = m; 1102 mp = &m->m_next; 1103 so->so_rcv.sb_mb = m = m->m_next; 1104 *mp = NULL; 1105 } else { 1106 so->so_rcv.sb_mb = m_free(m); 1107 m = so->so_rcv.sb_mb; 1108 } 1109 if (m != NULL) { 1110 m->m_nextpkt = nextrecord; 1111 if (nextrecord == NULL) 1112 so->so_rcv.sb_lastrecord = m; 1113 } else { 1114 so->so_rcv.sb_mb = nextrecord; 1115 SB_EMPTY_FIXUP(&so->so_rcv); 1116 } 1117 SBLASTRECORDCHK(&so->so_rcv); 1118 SBLASTMBUFCHK(&so->so_rcv); 1119 } 1120 } else { 1121 if (flags & MSG_PEEK) 1122 moff += len; 1123 else { 1124 if (mp != NULL) 1125 *mp = m_copym(m, 0, len, M_TRYWAIT); 1126 m->m_data += len; 1127 m->m_len -= len; 1128 so->so_rcv.sb_cc -= len; 1129 } 1130 } 1131 if (so->so_oobmark) { 1132 if ((flags & MSG_PEEK) == 0) { 1133 so->so_oobmark -= len; 1134 if (so->so_oobmark == 0) { 1135 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1136 so->so_rcv.sb_state |= SBS_RCVATMARK; 1137 break; 1138 } 1139 } else { 1140 offset += len; 1141 if (offset == so->so_oobmark) 1142 break; 1143 } 1144 } 1145 if (flags & MSG_EOR) 1146 break; 1147 /* 1148 * If the MSG_WAITALL flag is set (for non-atomic socket), 1149 * we must not quit until "uio->uio_resid == 0" or an error 1150 * termination. If a signal/timeout occurs, return 1151 * with a short count but without error. 1152 * Keep sockbuf locked against other readers. 1153 */ 1154 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 1155 !sosendallatonce(so) && nextrecord == NULL) { 1156 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1157 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE) 1158 break; 1159 /* 1160 * Notify the protocol that some data has been 1161 * drained before blocking. 1162 */ 1163 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb != NULL) { 1164 SOCKBUF_UNLOCK(&so->so_rcv); 1165 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 1166 SOCKBUF_LOCK(&so->so_rcv); 1167 } 1168 SBLASTRECORDCHK(&so->so_rcv); 1169 SBLASTMBUFCHK(&so->so_rcv); 1170 error = sbwait(&so->so_rcv); 1171 if (error) 1172 goto release; 1173 m = so->so_rcv.sb_mb; 1174 if (m != NULL) 1175 nextrecord = m->m_nextpkt; 1176 } 1177 } 1178 1179 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 1180 flags |= MSG_TRUNC; 1181 if ((flags & MSG_PEEK) == 0) { 1182 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1183 (void) sbdroprecord_locked(&so->so_rcv); 1184 } 1185 } 1186 if ((flags & MSG_PEEK) == 0) { 1187 if (m == NULL) { 1188 /* 1189 * First part is an inline SB_EMPTY_FIXUP(). Second 1190 * part makes sure sb_lastrecord is up-to-date if 1191 * there is still data in the socket buffer. 1192 */ 1193 so->so_rcv.sb_mb = nextrecord; 1194 if (so->so_rcv.sb_mb == NULL) { 1195 so->so_rcv.sb_mbtail = NULL; 1196 so->so_rcv.sb_lastrecord = NULL; 1197 } else if (nextrecord->m_nextpkt == NULL) 1198 so->so_rcv.sb_lastrecord = nextrecord; 1199 } 1200 SBLASTRECORDCHK(&so->so_rcv); 1201 SBLASTMBUFCHK(&so->so_rcv); 1202 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) { 1203 SOCKBUF_UNLOCK(&so->so_rcv); 1204 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 1205 SOCKBUF_LOCK(&so->so_rcv); 1206 } 1207 } 1208 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1209 if (orig_resid == uio->uio_resid && orig_resid && 1210 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 1211 sbunlock(&so->so_rcv); 1212 goto restart; 1213 } 1214 1215 if (flagsp != NULL) 1216 *flagsp |= flags; 1217release: 1218 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1219 sbunlock(&so->so_rcv); 1220out: 1221 SOCKBUF_UNLOCK(&so->so_rcv); 1222 return (error); 1223} 1224 1225int 1226soshutdown(so, how) 1227 struct socket *so; 1228 int how; 1229{ 1230 struct protosw *pr = so->so_proto; 1231 1232 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1233 return (EINVAL); 1234 1235 if (how != SHUT_WR) 1236 sorflush(so); 1237 if (how != SHUT_RD) 1238 return ((*pr->pr_usrreqs->pru_shutdown)(so)); 1239 return (0); 1240} 1241 1242void 1243sorflush(so) 1244 struct socket *so; 1245{ 1246 struct sockbuf *sb = &so->so_rcv; 1247 struct protosw *pr = so->so_proto; 1248 struct sockbuf asb; 1249 1250 /* 1251 * XXXRW: This is quite ugly. The existing code made a copy of the 1252 * socket buffer, then zero'd the original to clear the buffer 1253 * fields. However, with mutexes in the socket buffer, this causes 1254 * problems. We only clear the zeroable bits of the original; 1255 * however, we have to initialize and destroy the mutex in the copy 1256 * so that dom_dispose() and sbrelease() can lock t as needed. 1257 */ 1258 SOCKBUF_LOCK(sb); 1259 sb->sb_flags |= SB_NOINTR; 1260 (void) sblock(sb, M_WAITOK); 1261 /* 1262 * socantrcvmore_locked() drops the socket buffer mutex so that it 1263 * can safely perform wakeups. Re-acquire the mutex before 1264 * continuing. 1265 */ 1266 socantrcvmore_locked(so); 1267 SOCKBUF_LOCK(sb); 1268 sbunlock(sb); 1269 /* 1270 * Invalidate/clear most of the sockbuf structure, but leave 1271 * selinfo and mutex data unchanged. 1272 */ 1273 bzero(&asb, offsetof(struct sockbuf, sb_startzero)); 1274 bcopy(&sb->sb_startzero, &asb.sb_startzero, 1275 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1276 bzero(&sb->sb_startzero, 1277 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1278 SOCKBUF_UNLOCK(sb); 1279 1280 SOCKBUF_LOCK_INIT(&asb, "so_rcv"); 1281 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 1282 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1283 sbrelease(&asb, so); 1284 SOCKBUF_LOCK_DESTROY(&asb); 1285} 1286 1287#ifdef INET 1288static int 1289do_setopt_accept_filter(so, sopt) 1290 struct socket *so; 1291 struct sockopt *sopt; 1292{ 1293 struct accept_filter_arg *afap = NULL; 1294 struct accept_filter *afp; 1295 struct so_accf *af = so->so_accf; 1296 int error = 0; 1297 1298 /* do not set/remove accept filters on non listen sockets */ 1299 if ((so->so_options & SO_ACCEPTCONN) == 0) { 1300 error = EINVAL; 1301 goto out; 1302 } 1303 1304 /* removing the filter */ 1305 if (sopt == NULL) { 1306 if (af != NULL) { 1307 if (af->so_accept_filter != NULL && 1308 af->so_accept_filter->accf_destroy != NULL) { 1309 af->so_accept_filter->accf_destroy(so); 1310 } 1311 if (af->so_accept_filter_str != NULL) { 1312 FREE(af->so_accept_filter_str, M_ACCF); 1313 } 1314 FREE(af, M_ACCF); 1315 so->so_accf = NULL; 1316 } 1317 so->so_options &= ~SO_ACCEPTFILTER; 1318 return (0); 1319 } 1320 /* adding a filter */ 1321 /* must remove previous filter first */ 1322 if (af != NULL) { 1323 error = EINVAL; 1324 goto out; 1325 } 1326 /* don't put large objects on the kernel stack */ 1327 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK); 1328 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap); 1329 afap->af_name[sizeof(afap->af_name)-1] = '\0'; 1330 afap->af_arg[sizeof(afap->af_arg)-1] = '\0'; 1331 if (error) 1332 goto out; 1333 afp = accept_filt_get(afap->af_name); 1334 if (afp == NULL) { 1335 error = ENOENT; 1336 goto out; 1337 } 1338 MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO); 1339 if (afp->accf_create != NULL) { 1340 if (afap->af_name[0] != '\0') { 1341 int len = strlen(afap->af_name) + 1; 1342 1343 MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK); 1344 strcpy(af->so_accept_filter_str, afap->af_name); 1345 } 1346 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg); 1347 if (af->so_accept_filter_arg == NULL) { 1348 FREE(af->so_accept_filter_str, M_ACCF); 1349 FREE(af, M_ACCF); 1350 so->so_accf = NULL; 1351 error = EINVAL; 1352 goto out; 1353 } 1354 } 1355 af->so_accept_filter = afp; 1356 so->so_accf = af; 1357 so->so_options |= SO_ACCEPTFILTER; 1358out: 1359 if (afap != NULL) 1360 FREE(afap, M_TEMP); 1361 return (error); 1362} 1363#endif /* INET */ 1364 1365/* 1366 * Perhaps this routine, and sooptcopyout(), below, ought to come in 1367 * an additional variant to handle the case where the option value needs 1368 * to be some kind of integer, but not a specific size. 1369 * In addition to their use here, these functions are also called by the 1370 * protocol-level pr_ctloutput() routines. 1371 */ 1372int 1373sooptcopyin(sopt, buf, len, minlen) 1374 struct sockopt *sopt; 1375 void *buf; 1376 size_t len; 1377 size_t minlen; 1378{ 1379 size_t valsize; 1380 1381 /* 1382 * If the user gives us more than we wanted, we ignore it, 1383 * but if we don't get the minimum length the caller 1384 * wants, we return EINVAL. On success, sopt->sopt_valsize 1385 * is set to however much we actually retrieved. 1386 */ 1387 if ((valsize = sopt->sopt_valsize) < minlen) 1388 return EINVAL; 1389 if (valsize > len) 1390 sopt->sopt_valsize = valsize = len; 1391 1392 if (sopt->sopt_td != NULL) 1393 return (copyin(sopt->sopt_val, buf, valsize)); 1394 1395 bcopy(sopt->sopt_val, buf, valsize); 1396 return 0; 1397} 1398 1399int 1400sosetopt(so, sopt) 1401 struct socket *so; 1402 struct sockopt *sopt; 1403{ 1404 int error, optval; 1405 struct linger l; 1406 struct timeval tv; 1407 u_long val; 1408#ifdef MAC 1409 struct mac extmac; 1410#endif 1411 1412 error = 0; 1413 if (sopt->sopt_level != SOL_SOCKET) { 1414 if (so->so_proto && so->so_proto->pr_ctloutput) 1415 return ((*so->so_proto->pr_ctloutput) 1416 (so, sopt)); 1417 error = ENOPROTOOPT; 1418 } else { 1419 switch (sopt->sopt_name) { 1420#ifdef INET 1421 case SO_ACCEPTFILTER: 1422 error = do_setopt_accept_filter(so, sopt); 1423 if (error) 1424 goto bad; 1425 break; 1426#endif 1427 case SO_LINGER: 1428 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 1429 if (error) 1430 goto bad; 1431 1432 SOCK_LOCK(so); 1433 so->so_linger = l.l_linger; 1434 if (l.l_onoff) 1435 so->so_options |= SO_LINGER; 1436 else 1437 so->so_options &= ~SO_LINGER; 1438 SOCK_UNLOCK(so); 1439 break; 1440 1441 case SO_DEBUG: 1442 case SO_KEEPALIVE: 1443 case SO_DONTROUTE: 1444 case SO_USELOOPBACK: 1445 case SO_BROADCAST: 1446 case SO_REUSEADDR: 1447 case SO_REUSEPORT: 1448 case SO_OOBINLINE: 1449 case SO_TIMESTAMP: 1450 case SO_BINTIME: 1451 case SO_NOSIGPIPE: 1452 error = sooptcopyin(sopt, &optval, sizeof optval, 1453 sizeof optval); 1454 if (error) 1455 goto bad; 1456 SOCK_LOCK(so); 1457 if (optval) 1458 so->so_options |= sopt->sopt_name; 1459 else 1460 so->so_options &= ~sopt->sopt_name; 1461 SOCK_UNLOCK(so); 1462 break; 1463 1464 case SO_SNDBUF: 1465 case SO_RCVBUF: 1466 case SO_SNDLOWAT: 1467 case SO_RCVLOWAT: 1468 error = sooptcopyin(sopt, &optval, sizeof optval, 1469 sizeof optval); 1470 if (error) 1471 goto bad; 1472 1473 /* 1474 * Values < 1 make no sense for any of these 1475 * options, so disallow them. 1476 */ 1477 if (optval < 1) { 1478 error = EINVAL; 1479 goto bad; 1480 } 1481 1482 switch (sopt->sopt_name) { 1483 case SO_SNDBUF: 1484 case SO_RCVBUF: 1485 if (sbreserve(sopt->sopt_name == SO_SNDBUF ? 1486 &so->so_snd : &so->so_rcv, (u_long)optval, 1487 so, curthread) == 0) { 1488 error = ENOBUFS; 1489 goto bad; 1490 } 1491 break; 1492 1493 /* 1494 * Make sure the low-water is never greater than 1495 * the high-water. 1496 */ 1497 case SO_SNDLOWAT: 1498 so->so_snd.sb_lowat = 1499 (optval > so->so_snd.sb_hiwat) ? 1500 so->so_snd.sb_hiwat : optval; 1501 break; 1502 case SO_RCVLOWAT: 1503 so->so_rcv.sb_lowat = 1504 (optval > so->so_rcv.sb_hiwat) ? 1505 so->so_rcv.sb_hiwat : optval; 1506 break; 1507 } 1508 break; 1509 1510 case SO_SNDTIMEO: 1511 case SO_RCVTIMEO: 1512 error = sooptcopyin(sopt, &tv, sizeof tv, 1513 sizeof tv); 1514 if (error) 1515 goto bad; 1516 1517 /* assert(hz > 0); */ 1518 if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz || 1519 tv.tv_usec < 0 || tv.tv_usec >= 1000000) { 1520 error = EDOM; 1521 goto bad; 1522 } 1523 /* assert(tick > 0); */ 1524 /* assert(ULONG_MAX - SHRT_MAX >= 1000000); */ 1525 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick; 1526 if (val > SHRT_MAX) { 1527 error = EDOM; 1528 goto bad; 1529 } 1530 if (val == 0 && tv.tv_usec != 0) 1531 val = 1; 1532 1533 switch (sopt->sopt_name) { 1534 case SO_SNDTIMEO: 1535 so->so_snd.sb_timeo = val; 1536 break; 1537 case SO_RCVTIMEO: 1538 so->so_rcv.sb_timeo = val; 1539 break; 1540 } 1541 break; 1542 case SO_LABEL: 1543#ifdef MAC 1544 error = sooptcopyin(sopt, &extmac, sizeof extmac, 1545 sizeof extmac); 1546 if (error) 1547 goto bad; 1548 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 1549 so, &extmac); 1550#else 1551 error = EOPNOTSUPP; 1552#endif 1553 break; 1554 default: 1555 error = ENOPROTOOPT; 1556 break; 1557 } 1558 if (error == 0 && so->so_proto != NULL && 1559 so->so_proto->pr_ctloutput != NULL) { 1560 (void) ((*so->so_proto->pr_ctloutput) 1561 (so, sopt)); 1562 } 1563 } 1564bad: 1565 return (error); 1566} 1567 1568/* Helper routine for getsockopt */ 1569int 1570sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 1571{ 1572 int error; 1573 size_t valsize; 1574 1575 error = 0; 1576 1577 /* 1578 * Documented get behavior is that we always return a value, 1579 * possibly truncated to fit in the user's buffer. 1580 * Traditional behavior is that we always tell the user 1581 * precisely how much we copied, rather than something useful 1582 * like the total amount we had available for her. 1583 * Note that this interface is not idempotent; the entire answer must 1584 * generated ahead of time. 1585 */ 1586 valsize = min(len, sopt->sopt_valsize); 1587 sopt->sopt_valsize = valsize; 1588 if (sopt->sopt_val != NULL) { 1589 if (sopt->sopt_td != NULL) 1590 error = copyout(buf, sopt->sopt_val, valsize); 1591 else 1592 bcopy(buf, sopt->sopt_val, valsize); 1593 } 1594 return error; 1595} 1596 1597int 1598sogetopt(so, sopt) 1599 struct socket *so; 1600 struct sockopt *sopt; 1601{ 1602 int error, optval; 1603 struct linger l; 1604 struct timeval tv; 1605#ifdef INET 1606 struct accept_filter_arg *afap; 1607#endif 1608#ifdef MAC 1609 struct mac extmac; 1610#endif 1611 1612 error = 0; 1613 if (sopt->sopt_level != SOL_SOCKET) { 1614 if (so->so_proto && so->so_proto->pr_ctloutput) { 1615 return ((*so->so_proto->pr_ctloutput) 1616 (so, sopt)); 1617 } else 1618 return (ENOPROTOOPT); 1619 } else { 1620 switch (sopt->sopt_name) { 1621#ifdef INET 1622 case SO_ACCEPTFILTER: 1623 if ((so->so_options & SO_ACCEPTCONN) == 0) 1624 return (EINVAL); 1625 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), 1626 M_TEMP, M_WAITOK | M_ZERO); 1627 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 1628 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name); 1629 if (so->so_accf->so_accept_filter_str != NULL) 1630 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str); 1631 } 1632 error = sooptcopyout(sopt, afap, sizeof(*afap)); 1633 FREE(afap, M_TEMP); 1634 break; 1635#endif 1636 1637 case SO_LINGER: 1638 /* 1639 * XXXRW: We grab the lock here to get a consistent 1640 * snapshot of both fields. This may not really 1641 * be necessary. 1642 */ 1643 SOCK_LOCK(so); 1644 l.l_onoff = so->so_options & SO_LINGER; 1645 l.l_linger = so->so_linger; 1646 SOCK_UNLOCK(so); 1647 error = sooptcopyout(sopt, &l, sizeof l); 1648 break; 1649 1650 case SO_USELOOPBACK: 1651 case SO_DONTROUTE: 1652 case SO_DEBUG: 1653 case SO_KEEPALIVE: 1654 case SO_REUSEADDR: 1655 case SO_REUSEPORT: 1656 case SO_BROADCAST: 1657 case SO_OOBINLINE: 1658 case SO_TIMESTAMP: 1659 case SO_BINTIME: 1660 case SO_NOSIGPIPE: 1661 optval = so->so_options & sopt->sopt_name; 1662integer: 1663 error = sooptcopyout(sopt, &optval, sizeof optval); 1664 break; 1665 1666 case SO_TYPE: 1667 optval = so->so_type; 1668 goto integer; 1669 1670 case SO_ERROR: 1671 optval = so->so_error; 1672 so->so_error = 0; 1673 goto integer; 1674 1675 case SO_SNDBUF: 1676 optval = so->so_snd.sb_hiwat; 1677 goto integer; 1678 1679 case SO_RCVBUF: 1680 optval = so->so_rcv.sb_hiwat; 1681 goto integer; 1682 1683 case SO_SNDLOWAT: 1684 optval = so->so_snd.sb_lowat; 1685 goto integer; 1686 1687 case SO_RCVLOWAT: 1688 optval = so->so_rcv.sb_lowat; 1689 goto integer; 1690 1691 case SO_SNDTIMEO: 1692 case SO_RCVTIMEO: 1693 optval = (sopt->sopt_name == SO_SNDTIMEO ? 1694 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 1695 1696 tv.tv_sec = optval / hz; 1697 tv.tv_usec = (optval % hz) * tick; 1698 error = sooptcopyout(sopt, &tv, sizeof tv); 1699 break; 1700 case SO_LABEL: 1701#ifdef MAC 1702 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 1703 sizeof(extmac)); 1704 if (error) 1705 return (error); 1706 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 1707 so, &extmac); 1708 if (error) 1709 return (error); 1710 error = sooptcopyout(sopt, &extmac, sizeof extmac); 1711#else 1712 error = EOPNOTSUPP; 1713#endif 1714 break; 1715 case SO_PEERLABEL: 1716#ifdef MAC 1717 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 1718 sizeof(extmac)); 1719 if (error) 1720 return (error); 1721 error = mac_getsockopt_peerlabel( 1722 sopt->sopt_td->td_ucred, so, &extmac); 1723 if (error) 1724 return (error); 1725 error = sooptcopyout(sopt, &extmac, sizeof extmac); 1726#else 1727 error = EOPNOTSUPP; 1728#endif 1729 break; 1730 default: 1731 error = ENOPROTOOPT; 1732 break; 1733 } 1734 return (error); 1735 } 1736} 1737 1738/* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */ 1739int 1740soopt_getm(struct sockopt *sopt, struct mbuf **mp) 1741{ 1742 struct mbuf *m, *m_prev; 1743 int sopt_size = sopt->sopt_valsize; 1744 1745 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA); 1746 if (m == NULL) 1747 return ENOBUFS; 1748 if (sopt_size > MLEN) { 1749 MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT); 1750 if ((m->m_flags & M_EXT) == 0) { 1751 m_free(m); 1752 return ENOBUFS; 1753 } 1754 m->m_len = min(MCLBYTES, sopt_size); 1755 } else { 1756 m->m_len = min(MLEN, sopt_size); 1757 } 1758 sopt_size -= m->m_len; 1759 *mp = m; 1760 m_prev = m; 1761 1762 while (sopt_size) { 1763 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA); 1764 if (m == NULL) { 1765 m_freem(*mp); 1766 return ENOBUFS; 1767 } 1768 if (sopt_size > MLEN) { 1769 MCLGET(m, sopt->sopt_td != NULL ? M_TRYWAIT : 1770 M_DONTWAIT); 1771 if ((m->m_flags & M_EXT) == 0) { 1772 m_freem(m); 1773 m_freem(*mp); 1774 return ENOBUFS; 1775 } 1776 m->m_len = min(MCLBYTES, sopt_size); 1777 } else { 1778 m->m_len = min(MLEN, sopt_size); 1779 } 1780 sopt_size -= m->m_len; 1781 m_prev->m_next = m; 1782 m_prev = m; 1783 } 1784 return 0; 1785} 1786 1787/* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */ 1788int 1789soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 1790{ 1791 struct mbuf *m0 = m; 1792 1793 if (sopt->sopt_val == NULL) 1794 return 0; 1795 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 1796 if (sopt->sopt_td != NULL) { 1797 int error; 1798 1799 error = copyin(sopt->sopt_val, mtod(m, char *), 1800 m->m_len); 1801 if (error != 0) { 1802 m_freem(m0); 1803 return(error); 1804 } 1805 } else 1806 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 1807 sopt->sopt_valsize -= m->m_len; 1808 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 1809 m = m->m_next; 1810 } 1811 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 1812 panic("ip6_sooptmcopyin"); 1813 return 0; 1814} 1815 1816/* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */ 1817int 1818soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 1819{ 1820 struct mbuf *m0 = m; 1821 size_t valsize = 0; 1822 1823 if (sopt->sopt_val == NULL) 1824 return 0; 1825 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 1826 if (sopt->sopt_td != NULL) { 1827 int error; 1828 1829 error = copyout(mtod(m, char *), sopt->sopt_val, 1830 m->m_len); 1831 if (error != 0) { 1832 m_freem(m0); 1833 return(error); 1834 } 1835 } else 1836 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 1837 sopt->sopt_valsize -= m->m_len; 1838 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 1839 valsize += m->m_len; 1840 m = m->m_next; 1841 } 1842 if (m != NULL) { 1843 /* enough soopt buffer should be given from user-land */ 1844 m_freem(m0); 1845 return(EINVAL); 1846 } 1847 sopt->sopt_valsize = valsize; 1848 return 0; 1849} 1850 1851void 1852sohasoutofband(so) 1853 struct socket *so; 1854{ 1855 if (so->so_sigio != NULL) 1856 pgsigio(&so->so_sigio, SIGURG, 0); 1857 selwakeuppri(&so->so_rcv.sb_sel, PSOCK); 1858} 1859 1860int 1861sopoll(struct socket *so, int events, struct ucred *active_cred, 1862 struct thread *td) 1863{ 1864 int revents = 0; 1865 1866 if (events & (POLLIN | POLLRDNORM)) 1867 if (soreadable(so)) 1868 revents |= events & (POLLIN | POLLRDNORM); 1869 1870 if (events & POLLINIGNEOF) 1871 if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat || 1872 !TAILQ_EMPTY(&so->so_comp) || so->so_error) 1873 revents |= POLLINIGNEOF; 1874 1875 if (events & (POLLOUT | POLLWRNORM)) 1876 if (sowriteable(so)) 1877 revents |= events & (POLLOUT | POLLWRNORM); 1878 1879 if (events & (POLLPRI | POLLRDBAND)) 1880 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK)) 1881 revents |= events & (POLLPRI | POLLRDBAND); 1882 1883 if (revents == 0) { 1884 if (events & 1885 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM | 1886 POLLRDBAND)) { 1887 SOCKBUF_LOCK(&so->so_rcv); 1888 selrecord(td, &so->so_rcv.sb_sel); 1889 so->so_rcv.sb_flags |= SB_SEL; 1890 SOCKBUF_UNLOCK(&so->so_rcv); 1891 } 1892 1893 if (events & (POLLOUT | POLLWRNORM)) { 1894 SOCKBUF_LOCK(&so->so_snd); 1895 selrecord(td, &so->so_snd.sb_sel); 1896 so->so_snd.sb_flags |= SB_SEL; 1897 SOCKBUF_UNLOCK(&so->so_snd); 1898 } 1899 } 1900 1901 return (revents); 1902} 1903 1904int 1905soo_kqfilter(struct file *fp, struct knote *kn) 1906{ 1907 struct socket *so = kn->kn_fp->f_data; 1908 struct sockbuf *sb; 1909 1910 switch (kn->kn_filter) { 1911 case EVFILT_READ: 1912 if (so->so_options & SO_ACCEPTCONN) 1913 kn->kn_fop = &solisten_filtops; 1914 else 1915 kn->kn_fop = &soread_filtops; 1916 sb = &so->so_rcv; 1917 break; 1918 case EVFILT_WRITE: 1919 kn->kn_fop = &sowrite_filtops; 1920 sb = &so->so_snd; 1921 break; 1922 default: 1923 return (1); 1924 } 1925 1926 SOCKBUF_LOCK(sb); 1927 SLIST_INSERT_HEAD(&sb->sb_sel.si_note, kn, kn_selnext); 1928 sb->sb_flags |= SB_KNOTE; 1929 SOCKBUF_UNLOCK(sb); 1930 return (0); 1931} 1932 1933static void 1934filt_sordetach(struct knote *kn) 1935{ 1936 struct socket *so = kn->kn_fp->f_data; 1937 1938 SOCKBUF_LOCK(&so->so_rcv); 1939 SLIST_REMOVE(&so->so_rcv.sb_sel.si_note, kn, knote, kn_selnext); 1940 if (SLIST_EMPTY(&so->so_rcv.sb_sel.si_note)) 1941 so->so_rcv.sb_flags &= ~SB_KNOTE; 1942 SOCKBUF_UNLOCK(&so->so_rcv); 1943} 1944 1945/*ARGSUSED*/ 1946static int 1947filt_soread(struct knote *kn, long hint) 1948{ 1949 struct socket *so = kn->kn_fp->f_data; 1950 int need_lock, result; 1951 1952 /* 1953 * XXXRW: Conditional locking because filt_soread() can be called 1954 * either from KNOTE() in the socket context where the socket buffer 1955 * lock is already held, or from kqueue() itself. 1956 */ 1957 need_lock = !SOCKBUF_OWNED(&so->so_rcv); 1958 if (need_lock) 1959 SOCKBUF_LOCK(&so->so_rcv); 1960 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl; 1961 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1962 kn->kn_flags |= EV_EOF; 1963 kn->kn_fflags = so->so_error; 1964 result = 1; 1965 } else if (so->so_error) /* temporary udp error */ 1966 result = 1; 1967 else if (kn->kn_sfflags & NOTE_LOWAT) 1968 result = (kn->kn_data >= kn->kn_sdata); 1969 else 1970 result = (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat); 1971 if (need_lock) 1972 SOCKBUF_UNLOCK(&so->so_rcv); 1973 return (result); 1974} 1975 1976static void 1977filt_sowdetach(struct knote *kn) 1978{ 1979 struct socket *so = kn->kn_fp->f_data; 1980 1981 SOCKBUF_LOCK(&so->so_snd); 1982 SLIST_REMOVE(&so->so_snd.sb_sel.si_note, kn, knote, kn_selnext); 1983 if (SLIST_EMPTY(&so->so_snd.sb_sel.si_note)) 1984 so->so_snd.sb_flags &= ~SB_KNOTE; 1985 SOCKBUF_UNLOCK(&so->so_snd); 1986} 1987 1988/*ARGSUSED*/ 1989static int 1990filt_sowrite(struct knote *kn, long hint) 1991{ 1992 struct socket *so = kn->kn_fp->f_data; 1993 int need_lock, result; 1994 1995 /* 1996 * XXXRW: Conditional locking because filt_soread() can be called 1997 * either from KNOTE() in the socket context where the socket buffer 1998 * lock is already held, or from kqueue() itself. 1999 */ 2000 need_lock = !SOCKBUF_OWNED(&so->so_snd); 2001 if (need_lock) 2002 SOCKBUF_LOCK(&so->so_snd); 2003 kn->kn_data = sbspace(&so->so_snd); 2004 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 2005 kn->kn_flags |= EV_EOF; 2006 kn->kn_fflags = so->so_error; 2007 result = 1; 2008 } else if (so->so_error) /* temporary udp error */ 2009 result = 1; 2010 else if (((so->so_state & SS_ISCONNECTED) == 0) && 2011 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 2012 result = 0; 2013 else if (kn->kn_sfflags & NOTE_LOWAT) 2014 result = (kn->kn_data >= kn->kn_sdata); 2015 else 2016 result = (kn->kn_data >= so->so_snd.sb_lowat); 2017 if (need_lock) 2018 SOCKBUF_UNLOCK(&so->so_snd); 2019 return (result); 2020} 2021 2022/*ARGSUSED*/ 2023static int 2024filt_solisten(struct knote *kn, long hint) 2025{ 2026 struct socket *so = kn->kn_fp->f_data; 2027 2028 kn->kn_data = so->so_qlen; 2029 return (! TAILQ_EMPTY(&so->so_comp)); 2030} 2031 2032int 2033socheckuid(struct socket *so, uid_t uid) 2034{ 2035 2036 if (so == NULL) 2037 return (EPERM); 2038 if (so->so_cred->cr_uid == uid) 2039 return (0); 2040 return (EPERM); 2041} 2042