1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993 3 * The Regents of the University of California. 4 * Copyright (c) 2004 The FreeBSD Foundation 5 * Copyright (c) 2004-2007 Robert N. M. Watson 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 33 */ 34 35 /* 36 * Comments on the socket life cycle: 37 * 38 * soalloc() sets of socket layer state for a socket, called only by 39 * socreate() and sonewconn(). Socket layer private. 40 * 41 * sodealloc() tears down socket layer state for a socket, called only by 42 * sofree() and sonewconn(). Socket layer private. 43 * 44 * pru_attach() associates protocol layer state with an allocated socket; 45 * called only once, may fail, aborting socket allocation. This is called 46 * from socreate() and sonewconn(). Socket layer private. 47 * 48 * pru_detach() disassociates protocol layer state from an attached socket, 49 * and will be called exactly once for sockets in which pru_attach() has 50 * been successfully called. If pru_attach() returned an error, 51 * pru_detach() will not be called. Socket layer private. 52 * 53 * pru_abort() and pru_close() notify the protocol layer that the last 54 * consumer of a socket is starting to tear down the socket, and that the 55 * protocol should terminate the connection. Historically, pru_abort() also 56 * detached protocol state from the socket state, but this is no longer the 57 * case. 58 * 59 * socreate() creates a socket and attaches protocol state. This is a public 60 * interface that may be used by socket layer consumers to create new 61 * sockets. 62 * 63 * sonewconn() creates a socket and attaches protocol state. This is a 64 * public interface that may be used by protocols to create new sockets when 65 * a new connection is received and will be available for accept() on a 66 * listen socket. 67 * 68 * soclose() destroys a socket after possibly waiting for it to disconnect. 69 * This is a public interface that socket consumers should use to close and 70 * release a socket when done with it. 71 * 72 * soabort() destroys a socket without waiting for it to disconnect (used 73 * only for incoming connections that are already partially or fully 74 * connected). This is used internally by the socket layer when clearing 75 * listen socket queues (due to overflow or close on the listen socket), but 76 * is also a public interface protocols may use to abort connections in 77 * their incomplete listen queues should they no longer be required. Sockets 78 * placed in completed connection listen queues should not be aborted for 79 * reasons described in the comment above the soclose() implementation. This 80 * is not a general purpose close routine, and except in the specific 81 * circumstances described here, should not be used. 82 * 83 * sofree() will free a socket and its protocol state if all references on 84 * the socket have been released, and is the public interface to attempt to 85 * free a socket when a reference is removed. This is a socket layer private 86 * interface. 87 * 88 * NOTE: In addition to socreate() and soclose(), which provide a single 89 * socket reference to the consumer to be managed as required, there are two 90 * calls to explicitly manage socket references, soref(), and sorele(). 91 * Currently, these are generally required only when transitioning a socket 92 * from a listen queue to a file descriptor, in order to prevent garbage 93 * collection of the socket at an untimely moment. For a number of reasons, 94 * these interfaces are not preferred, and should be avoided. 95 */ 96 97 #include <sys/cdefs.h> 98 __FBSDID("$FreeBSD$"); 99 100 #include "opt_inet.h" 101 #include "opt_mac.h" 102 #include "opt_zero.h" 103 #include "opt_compat.h" 104 105 #include <sys/param.h> 106 #include <sys/systm.h> 107 #include <sys/fcntl.h> 108 #include <sys/limits.h> 109 #include <sys/lock.h> 110 #include <sys/mac.h> 111 #include <sys/malloc.h> 112 #include <sys/mbuf.h> 113 #include <sys/mutex.h> 114 #include <sys/domain.h> 115 #include <sys/file.h> /* for struct knote */ 116 #include <sys/kernel.h> 117 #include <sys/event.h> 118 #include <sys/eventhandler.h> 119 #include <sys/poll.h> 120 #include <sys/proc.h> 121 #include <sys/protosw.h> 122 #include <sys/socket.h> 123 #include <sys/socketvar.h> 124 #include <sys/resourcevar.h> 125 #include <net/route.h> 126 #include <sys/signalvar.h> 127 #include <sys/stat.h> 128 #include <sys/sx.h> 129 #include <sys/sysctl.h> 130 #include <sys/uio.h> 131 #include <sys/jail.h> 132 133 #include <security/mac/mac_framework.h> 134 135 #include <vm/uma.h> 136 137 #ifdef COMPAT_IA32 138 #include <sys/mount.h> 139 #include <compat/freebsd32/freebsd32.h> 140 141 extern struct sysentvec ia32_freebsd_sysvec; 142 #endif 143 144 static int soreceive_rcvoob(struct socket *so, struct uio *uio, 145 int flags); 146 147 static void filt_sordetach(struct knote *kn); 148 static int filt_soread(struct knote *kn, long hint); 149 static void filt_sowdetach(struct knote *kn); 150 static int filt_sowrite(struct knote *kn, long hint); 151 static int filt_solisten(struct knote *kn, long hint); 152 153 static struct filterops solisten_filtops = 154 { 1, NULL, filt_sordetach, filt_solisten }; 155 static struct filterops soread_filtops = 156 { 1, NULL, filt_sordetach, filt_soread }; 157 static struct filterops sowrite_filtops = 158 { 1, NULL, filt_sowdetach, filt_sowrite }; 159 160 uma_zone_t socket_zone; 161 so_gen_t so_gencnt; /* generation count for sockets */ 162 163 int maxsockets; 164 165 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 166 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 167 168 static int somaxconn = SOMAXCONN; 169 static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS); 170 /* XXX: we dont have SYSCTL_USHORT */ 171 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW, 172 0, sizeof(int), sysctl_somaxconn, "I", "Maximum pending socket connection " 173 "queue size"); 174 static int numopensockets; 175 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD, 176 &numopensockets, 0, "Number of open sockets"); 177 #ifdef ZERO_COPY_SOCKETS 178 /* These aren't static because they're used in other files. */ 179 int so_zero_copy_send = 1; 180 int so_zero_copy_receive = 1; 181 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0, 182 "Zero copy controls"); 183 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW, 184 &so_zero_copy_receive, 0, "Enable zero copy receive"); 185 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW, 186 &so_zero_copy_send, 0, "Enable zero copy send"); 187 #endif /* ZERO_COPY_SOCKETS */ 188 189 /* 190 * accept_mtx locks down per-socket fields relating to accept queues. See 191 * socketvar.h for an annotation of the protected fields of struct socket. 192 */ 193 struct mtx accept_mtx; 194 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF); 195 196 /* 197 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket 198 * so_gencnt field. 199 */ 200 static struct mtx so_global_mtx; 201 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF); 202 203 /* 204 * General IPC sysctl name space, used by sockets and a variety of other IPC 205 * types. 206 */ 207 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); 208 209 /* 210 * Sysctl to get and set the maximum global sockets limit. Notify protocols 211 * of the change so that they can update their dependent limits as required. 212 */ 213 static int 214 sysctl_maxsockets(SYSCTL_HANDLER_ARGS) 215 { 216 int error, newmaxsockets; 217 218 newmaxsockets = maxsockets; 219 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req); 220 if (error == 0 && req->newptr) { 221 if (newmaxsockets > maxsockets) { 222 maxsockets = newmaxsockets; 223 if (maxsockets > ((maxfiles / 4) * 3)) { 224 maxfiles = (maxsockets * 5) / 4; 225 maxfilesperproc = (maxfiles * 9) / 10; 226 } 227 EVENTHANDLER_INVOKE(maxsockets_change); 228 } else 229 error = EINVAL; 230 } 231 return (error); 232 } 233 234 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW, 235 &maxsockets, 0, sysctl_maxsockets, "IU", 236 "Maximum number of sockets avaliable"); 237 238 /* 239 * Initialise maxsockets. 240 */ 241 static void init_maxsockets(void *ignored) 242 { 243 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 244 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters)); 245 } 246 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); 247 248 /* 249 * Socket operation routines. These routines are called by the routines in 250 * sys_socket.c or from a system process, and implement the semantics of 251 * socket operations by switching out to the protocol specific routines. 252 */ 253 254 /* 255 * Get a socket structure from our zone, and initialize it. Note that it 256 * would probably be better to allocate socket and PCB at the same time, but 257 * I'm not convinced that all the protocols can be easily modified to do 258 * this. 259 * 260 * soalloc() returns a socket with a ref count of 0. 261 */ 262 static struct socket * 263 soalloc(void) 264 { 265 struct socket *so; 266 267 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO); 268 if (so == NULL) 269 return (NULL); 270 #ifdef MAC 271 if (mac_socket_init(so, M_NOWAIT) != 0) { 272 uma_zfree(socket_zone, so); 273 return (NULL); 274 } 275 #endif 276 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd"); 277 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv"); 278 sx_init(&so->so_snd.sb_sx, "so_snd_sx"); 279 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx"); 280 TAILQ_INIT(&so->so_aiojobq); 281 mtx_lock(&so_global_mtx); 282 so->so_gencnt = ++so_gencnt; 283 ++numopensockets; 284 mtx_unlock(&so_global_mtx); 285 return (so); 286 } 287 288 /* 289 * Free the storage associated with a socket at the socket layer, tear down 290 * locks, labels, etc. All protocol state is assumed already to have been 291 * torn down (and possibly never set up) by the caller. 292 */ 293 static void 294 sodealloc(struct socket *so) 295 { 296 297 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); 298 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL")); 299 300 mtx_lock(&so_global_mtx); 301 so->so_gencnt = ++so_gencnt; 302 --numopensockets; /* Could be below, but faster here. */ 303 mtx_unlock(&so_global_mtx); 304 if (so->so_rcv.sb_hiwat) 305 (void)chgsbsize(so->so_cred->cr_uidinfo, 306 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 307 if (so->so_snd.sb_hiwat) 308 (void)chgsbsize(so->so_cred->cr_uidinfo, 309 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 310 #ifdef INET 311 /* remove acccept filter if one is present. */ 312 if (so->so_accf != NULL) 313 do_setopt_accept_filter(so, NULL); 314 #endif 315 #ifdef MAC 316 mac_socket_destroy(so); 317 #endif 318 crfree(so->so_cred); 319 sx_destroy(&so->so_snd.sb_sx); 320 sx_destroy(&so->so_rcv.sb_sx); 321 SOCKBUF_LOCK_DESTROY(&so->so_snd); 322 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 323 uma_zfree(socket_zone, so); 324 } 325 326 /* 327 * socreate returns a socket with a ref count of 1. The socket should be 328 * closed with soclose(). 329 */ 330 int 331 socreate(int dom, struct socket **aso, int type, int proto, 332 struct ucred *cred, struct thread *td) 333 { 334 struct protosw *prp; 335 struct socket *so; 336 int error; 337 338 if (proto) 339 prp = pffindproto(dom, proto, type); 340 else 341 prp = pffindtype(dom, type); 342 343 if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL || 344 prp->pr_usrreqs->pru_attach == pru_attach_notsupp) 345 return (EPROTONOSUPPORT); 346 347 if (jailed(cred) && jail_socket_unixiproute_only && 348 prp->pr_domain->dom_family != PF_LOCAL && 349 prp->pr_domain->dom_family != PF_INET && 350 prp->pr_domain->dom_family != PF_ROUTE) { 351 return (EPROTONOSUPPORT); 352 } 353 354 if (prp->pr_type != type) 355 return (EPROTOTYPE); 356 so = soalloc(); 357 if (so == NULL) 358 return (ENOBUFS); 359 360 TAILQ_INIT(&so->so_incomp); 361 TAILQ_INIT(&so->so_comp); 362 so->so_type = type; 363 so->so_cred = crhold(cred); 364 if ((prp->pr_domain->dom_family == PF_INET) || 365 (prp->pr_domain->dom_family == PF_ROUTE)) 366 so->so_fibnum = td->td_proc->p_fibnum; 367 else 368 so->so_fibnum = 0; 369 so->so_proto = prp; 370 #ifdef MAC 371 mac_socket_create(cred, so); 372 #endif 373 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv), 374 NULL, NULL, NULL); 375 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd), 376 NULL, NULL, NULL); 377 so->so_count = 1; 378 /* 379 * Auto-sizing of socket buffers is managed by the protocols and 380 * the appropriate flags must be set in the pru_attach function. 381 */ 382 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); 383 if (error) { 384 KASSERT(so->so_count == 1, ("socreate: so_count %d", 385 so->so_count)); 386 so->so_count = 0; 387 sodealloc(so); 388 return (error); 389 } 390 *aso = so; 391 return (0); 392 } 393 394 #ifdef REGRESSION 395 static int regression_sonewconn_earlytest = 1; 396 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW, 397 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test"); 398 #endif 399 400 /* 401 * When an attempt at a new connection is noted on a socket which accepts 402 * connections, sonewconn is called. If the connection is possible (subject 403 * to space constraints, etc.) then we allocate a new structure, propoerly 404 * linked into the data structure of the original socket, and return this. 405 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. 406 * 407 * Note: the ref count on the socket is 0 on return. 408 */ 409 struct socket * 410 sonewconn(struct socket *head, int connstatus) 411 { 412 struct socket *so; 413 int over; 414 415 ACCEPT_LOCK(); 416 over = (head->so_qlen > 3 * head->so_qlimit / 2); 417 ACCEPT_UNLOCK(); 418 #ifdef REGRESSION 419 if (regression_sonewconn_earlytest && over) 420 #else 421 if (over) 422 #endif 423 return (NULL); 424 so = soalloc(); 425 if (so == NULL) 426 return (NULL); 427 if ((head->so_options & SO_ACCEPTFILTER) != 0) 428 connstatus = 0; 429 so->so_head = head; 430 so->so_type = head->so_type; 431 so->so_options = head->so_options &~ SO_ACCEPTCONN; 432 so->so_linger = head->so_linger; 433 so->so_state = head->so_state | SS_NOFDREF; 434 so->so_proto = head->so_proto; 435 so->so_cred = crhold(head->so_cred); 436 #ifdef MAC 437 SOCK_LOCK(head); 438 mac_socket_newconn(head, so); 439 SOCK_UNLOCK(head); 440 #endif 441 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv), 442 NULL, NULL, NULL); 443 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd), 444 NULL, NULL, NULL); 445 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) || 446 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 447 sodealloc(so); 448 return (NULL); 449 } 450 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 451 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 452 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 453 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 454 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; 455 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; 456 so->so_state |= connstatus; 457 ACCEPT_LOCK(); 458 if (connstatus) { 459 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 460 so->so_qstate |= SQ_COMP; 461 head->so_qlen++; 462 } else { 463 /* 464 * Keep removing sockets from the head until there's room for 465 * us to insert on the tail. In pre-locking revisions, this 466 * was a simple if(), but as we could be racing with other 467 * threads and soabort() requires dropping locks, we must 468 * loop waiting for the condition to be true. 469 */ 470 while (head->so_incqlen > head->so_qlimit) { 471 struct socket *sp; 472 sp = TAILQ_FIRST(&head->so_incomp); 473 TAILQ_REMOVE(&head->so_incomp, sp, so_list); 474 head->so_incqlen--; 475 sp->so_qstate &= ~SQ_INCOMP; 476 sp->so_head = NULL; 477 ACCEPT_UNLOCK(); 478 soabort(sp); 479 ACCEPT_LOCK(); 480 } 481 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); 482 so->so_qstate |= SQ_INCOMP; 483 head->so_incqlen++; 484 } 485 ACCEPT_UNLOCK(); 486 if (connstatus) { 487 sorwakeup(head); 488 wakeup_one(&head->so_timeo); 489 } 490 return (so); 491 } 492 493 int 494 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 495 { 496 497 return ((*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td)); 498 } 499 500 /* 501 * solisten() transitions a socket from a non-listening state to a listening 502 * state, but can also be used to update the listen queue depth on an 503 * existing listen socket. The protocol will call back into the sockets 504 * layer using solisten_proto_check() and solisten_proto() to check and set 505 * socket-layer listen state. Call backs are used so that the protocol can 506 * acquire both protocol and socket layer locks in whatever order is required 507 * by the protocol. 508 * 509 * Protocol implementors are advised to hold the socket lock across the 510 * socket-layer test and set to avoid races at the socket layer. 511 */ 512 int 513 solisten(struct socket *so, int backlog, struct thread *td) 514 { 515 516 return ((*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td)); 517 } 518 519 int 520 solisten_proto_check(struct socket *so) 521 { 522 523 SOCK_LOCK_ASSERT(so); 524 525 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 526 SS_ISDISCONNECTING)) 527 return (EINVAL); 528 return (0); 529 } 530 531 void 532 solisten_proto(struct socket *so, int backlog) 533 { 534 535 SOCK_LOCK_ASSERT(so); 536 537 if (backlog < 0 || backlog > somaxconn) 538 backlog = somaxconn; 539 so->so_qlimit = backlog; 540 so->so_options |= SO_ACCEPTCONN; 541 } 542 543 /* 544 * Attempt to free a socket. This should really be sotryfree(). 545 * 546 * sofree() will succeed if: 547 * 548 * - There are no outstanding file descriptor references or related consumers 549 * (so_count == 0). 550 * 551 * - The socket has been closed by user space, if ever open (SS_NOFDREF). 552 * 553 * - The protocol does not have an outstanding strong reference on the socket 554 * (SS_PROTOREF). 555 * 556 * - The socket is not in a completed connection queue, so a process has been 557 * notified that it is present. If it is removed, the user process may 558 * block in accept() despite select() saying the socket was ready. 559 * 560 * Otherwise, it will quietly abort so that a future call to sofree(), when 561 * conditions are right, can succeed. 562 */ 563 void 564 sofree(struct socket *so) 565 { 566 struct protosw *pr = so->so_proto; 567 struct socket *head; 568 569 ACCEPT_LOCK_ASSERT(); 570 SOCK_LOCK_ASSERT(so); 571 572 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 || 573 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) { 574 SOCK_UNLOCK(so); 575 ACCEPT_UNLOCK(); 576 return; 577 } 578 579 head = so->so_head; 580 if (head != NULL) { 581 KASSERT((so->so_qstate & SQ_COMP) != 0 || 582 (so->so_qstate & SQ_INCOMP) != 0, 583 ("sofree: so_head != NULL, but neither SQ_COMP nor " 584 "SQ_INCOMP")); 585 KASSERT((so->so_qstate & SQ_COMP) == 0 || 586 (so->so_qstate & SQ_INCOMP) == 0, 587 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP")); 588 TAILQ_REMOVE(&head->so_incomp, so, so_list); 589 head->so_incqlen--; 590 so->so_qstate &= ~SQ_INCOMP; 591 so->so_head = NULL; 592 } 593 KASSERT((so->so_qstate & SQ_COMP) == 0 && 594 (so->so_qstate & SQ_INCOMP) == 0, 595 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)", 596 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP)); 597 if (so->so_options & SO_ACCEPTCONN) { 598 KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated")); 599 KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_comp populated")); 600 } 601 SOCK_UNLOCK(so); 602 ACCEPT_UNLOCK(); 603 604 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 605 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb); 606 if (pr->pr_usrreqs->pru_detach != NULL) 607 (*pr->pr_usrreqs->pru_detach)(so); 608 609 /* 610 * From this point on, we assume that no other references to this 611 * socket exist anywhere else in the stack. Therefore, no locks need 612 * to be acquired or held. 613 * 614 * We used to do a lot of socket buffer and socket locking here, as 615 * well as invoke sorflush() and perform wakeups. The direct call to 616 * dom_dispose() and sbrelease_internal() are an inlining of what was 617 * necessary from sorflush(). 618 * 619 * Notice that the socket buffer and kqueue state are torn down 620 * before calling pru_detach. This means that protocols shold not 621 * assume they can perform socket wakeups, etc, in their detach code. 622 */ 623 sbdestroy(&so->so_snd, so); 624 sbdestroy(&so->so_rcv, so); 625 knlist_destroy(&so->so_rcv.sb_sel.si_note); 626 knlist_destroy(&so->so_snd.sb_sel.si_note); 627 sodealloc(so); 628 } 629 630 /* 631 * Close a socket on last file table reference removal. Initiate disconnect 632 * if connected. Free socket when disconnect complete. 633 * 634 * This function will sorele() the socket. Note that soclose() may be called 635 * prior to the ref count reaching zero. The actual socket structure will 636 * not be freed until the ref count reaches zero. 637 */ 638 int 639 soclose(struct socket *so) 640 { 641 int error = 0; 642 643 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter")); 644 645 funsetown(&so->so_sigio); 646 if (so->so_state & SS_ISCONNECTED) { 647 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 648 error = sodisconnect(so); 649 if (error) 650 goto drop; 651 } 652 if (so->so_options & SO_LINGER) { 653 if ((so->so_state & SS_ISDISCONNECTING) && 654 (so->so_state & SS_NBIO)) 655 goto drop; 656 while (so->so_state & SS_ISCONNECTED) { 657 error = tsleep(&so->so_timeo, 658 PSOCK | PCATCH, "soclos", so->so_linger * hz); 659 if (error) 660 break; 661 } 662 } 663 } 664 665 drop: 666 if (so->so_proto->pr_usrreqs->pru_close != NULL) 667 (*so->so_proto->pr_usrreqs->pru_close)(so); 668 if (so->so_options & SO_ACCEPTCONN) { 669 struct socket *sp; 670 ACCEPT_LOCK(); 671 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) { 672 TAILQ_REMOVE(&so->so_incomp, sp, so_list); 673 so->so_incqlen--; 674 sp->so_qstate &= ~SQ_INCOMP; 675 sp->so_head = NULL; 676 ACCEPT_UNLOCK(); 677 soabort(sp); 678 ACCEPT_LOCK(); 679 } 680 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) { 681 TAILQ_REMOVE(&so->so_comp, sp, so_list); 682 so->so_qlen--; 683 sp->so_qstate &= ~SQ_COMP; 684 sp->so_head = NULL; 685 ACCEPT_UNLOCK(); 686 soabort(sp); 687 ACCEPT_LOCK(); 688 } 689 ACCEPT_UNLOCK(); 690 } 691 ACCEPT_LOCK(); 692 SOCK_LOCK(so); 693 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF")); 694 so->so_state |= SS_NOFDREF; 695 sorele(so); 696 return (error); 697 } 698 699 /* 700 * soabort() is used to abruptly tear down a connection, such as when a 701 * resource limit is reached (listen queue depth exceeded), or if a listen 702 * socket is closed while there are sockets waiting to be accepted. 703 * 704 * This interface is tricky, because it is called on an unreferenced socket, 705 * and must be called only by a thread that has actually removed the socket 706 * from the listen queue it was on, or races with other threads are risked. 707 * 708 * This interface will call into the protocol code, so must not be called 709 * with any socket locks held. Protocols do call it while holding their own 710 * recursible protocol mutexes, but this is something that should be subject 711 * to review in the future. 712 */ 713 void 714 soabort(struct socket *so) 715 { 716 717 /* 718 * In as much as is possible, assert that no references to this 719 * socket are held. This is not quite the same as asserting that the 720 * current thread is responsible for arranging for no references, but 721 * is as close as we can get for now. 722 */ 723 KASSERT(so->so_count == 0, ("soabort: so_count")); 724 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF")); 725 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF")); 726 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP")); 727 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP")); 728 729 if (so->so_proto->pr_usrreqs->pru_abort != NULL) 730 (*so->so_proto->pr_usrreqs->pru_abort)(so); 731 ACCEPT_LOCK(); 732 SOCK_LOCK(so); 733 sofree(so); 734 } 735 736 int 737 soaccept(struct socket *so, struct sockaddr **nam) 738 { 739 int error; 740 741 SOCK_LOCK(so); 742 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF")); 743 so->so_state &= ~SS_NOFDREF; 744 SOCK_UNLOCK(so); 745 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); 746 return (error); 747 } 748 749 int 750 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 751 { 752 int error; 753 754 if (so->so_options & SO_ACCEPTCONN) 755 return (EOPNOTSUPP); 756 /* 757 * If protocol is connection-based, can only connect once. 758 * Otherwise, if connected, try to disconnect first. This allows 759 * user to disconnect by connecting to, e.g., a null address. 760 */ 761 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 762 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 763 (error = sodisconnect(so)))) { 764 error = EISCONN; 765 } else { 766 /* 767 * Prevent accumulated error from previous connection from 768 * biting us. 769 */ 770 so->so_error = 0; 771 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td); 772 } 773 774 return (error); 775 } 776 777 int 778 soconnect2(struct socket *so1, struct socket *so2) 779 { 780 781 return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2)); 782 } 783 784 int 785 sodisconnect(struct socket *so) 786 { 787 int error; 788 789 if ((so->so_state & SS_ISCONNECTED) == 0) 790 return (ENOTCONN); 791 if (so->so_state & SS_ISDISCONNECTING) 792 return (EALREADY); 793 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 794 return (error); 795 } 796 797 #ifdef ZERO_COPY_SOCKETS 798 struct so_zerocopy_stats{ 799 int size_ok; 800 int align_ok; 801 int found_ifp; 802 }; 803 struct so_zerocopy_stats so_zerocp_stats = {0,0,0}; 804 #include <netinet/in.h> 805 #include <net/route.h> 806 #include <netinet/in_pcb.h> 807 #include <vm/vm.h> 808 #include <vm/vm_page.h> 809 #include <vm/vm_object.h> 810 811 /* 812 * sosend_copyin() is only used if zero copy sockets are enabled. Otherwise 813 * sosend_dgram() and sosend_generic() use m_uiotombuf(). 814 * 815 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or 816 * all of the data referenced by the uio. If desired, it uses zero-copy. 817 * *space will be updated to reflect data copied in. 818 * 819 * NB: If atomic I/O is requested, the caller must already have checked that 820 * space can hold resid bytes. 821 * 822 * NB: In the event of an error, the caller may need to free the partial 823 * chain pointed to by *mpp. The contents of both *uio and *space may be 824 * modified even in the case of an error. 825 */ 826 static int 827 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space, 828 int flags) 829 { 830 struct mbuf *m, **mp, *top; 831 long len, resid; 832 int error; 833 #ifdef ZERO_COPY_SOCKETS 834 int cow_send; 835 #endif 836 837 *retmp = top = NULL; 838 mp = ⊤ 839 len = 0; 840 resid = uio->uio_resid; 841 error = 0; 842 do { 843 #ifdef ZERO_COPY_SOCKETS 844 cow_send = 0; 845 #endif /* ZERO_COPY_SOCKETS */ 846 if (resid >= MINCLSIZE) { 847 #ifdef ZERO_COPY_SOCKETS 848 if (top == NULL) { 849 m = m_gethdr(M_WAITOK, MT_DATA); 850 m->m_pkthdr.len = 0; 851 m->m_pkthdr.rcvif = NULL; 852 } else 853 m = m_get(M_WAITOK, MT_DATA); 854 if (so_zero_copy_send && 855 resid>=PAGE_SIZE && 856 *space>=PAGE_SIZE && 857 uio->uio_iov->iov_len>=PAGE_SIZE) { 858 so_zerocp_stats.size_ok++; 859 so_zerocp_stats.align_ok++; 860 cow_send = socow_setup(m, uio); 861 len = cow_send; 862 } 863 if (!cow_send) { 864 m_clget(m, M_WAITOK); 865 len = min(min(MCLBYTES, resid), *space); 866 } 867 #else /* ZERO_COPY_SOCKETS */ 868 if (top == NULL) { 869 m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR); 870 m->m_pkthdr.len = 0; 871 m->m_pkthdr.rcvif = NULL; 872 } else 873 m = m_getcl(M_WAIT, MT_DATA, 0); 874 len = min(min(MCLBYTES, resid), *space); 875 #endif /* ZERO_COPY_SOCKETS */ 876 } else { 877 if (top == NULL) { 878 m = m_gethdr(M_WAIT, MT_DATA); 879 m->m_pkthdr.len = 0; 880 m->m_pkthdr.rcvif = NULL; 881 882 len = min(min(MHLEN, resid), *space); 883 /* 884 * For datagram protocols, leave room 885 * for protocol headers in first mbuf. 886 */ 887 if (atomic && m && len < MHLEN) 888 MH_ALIGN(m, len); 889 } else { 890 m = m_get(M_WAIT, MT_DATA); 891 len = min(min(MLEN, resid), *space); 892 } 893 } 894 if (m == NULL) { 895 error = ENOBUFS; 896 goto out; 897 } 898 899 *space -= len; 900 #ifdef ZERO_COPY_SOCKETS 901 if (cow_send) 902 error = 0; 903 else 904 #endif /* ZERO_COPY_SOCKETS */ 905 error = uiomove(mtod(m, void *), (int)len, uio); 906 resid = uio->uio_resid; 907 m->m_len = len; 908 *mp = m; 909 top->m_pkthdr.len += len; 910 if (error) 911 goto out; 912 mp = &m->m_next; 913 if (resid <= 0) { 914 if (flags & MSG_EOR) 915 top->m_flags |= M_EOR; 916 break; 917 } 918 } while (*space > 0 && atomic); 919 out: 920 *retmp = top; 921 return (error); 922 } 923 #endif /*ZERO_COPY_SOCKETS*/ 924 925 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) 926 927 int 928 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 929 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 930 { 931 long space, resid; 932 int clen = 0, error, dontroute; 933 #ifdef ZERO_COPY_SOCKETS 934 int atomic = sosendallatonce(so) || top; 935 #endif 936 937 KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM")); 938 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 939 ("sodgram_send: !PR_ATOMIC")); 940 941 if (uio != NULL) 942 resid = uio->uio_resid; 943 else 944 resid = top->m_pkthdr.len; 945 /* 946 * In theory resid should be unsigned. However, space must be 947 * signed, as it might be less than 0 if we over-committed, and we 948 * must use a signed comparison of space and resid. On the other 949 * hand, a negative resid causes us to loop sending 0-length 950 * segments to the protocol. 951 * 952 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 953 * type sockets since that's an error. 954 */ 955 if (resid < 0) { 956 error = EINVAL; 957 goto out; 958 } 959 960 dontroute = 961 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 962 if (td != NULL) 963 td->td_ru.ru_msgsnd++; 964 if (control != NULL) 965 clen = control->m_len; 966 967 SOCKBUF_LOCK(&so->so_snd); 968 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 969 SOCKBUF_UNLOCK(&so->so_snd); 970 error = EPIPE; 971 goto out; 972 } 973 if (so->so_error) { 974 error = so->so_error; 975 so->so_error = 0; 976 SOCKBUF_UNLOCK(&so->so_snd); 977 goto out; 978 } 979 if ((so->so_state & SS_ISCONNECTED) == 0) { 980 /* 981 * `sendto' and `sendmsg' is allowed on a connection-based 982 * socket if it supports implied connect. Return ENOTCONN if 983 * not connected and no address is supplied. 984 */ 985 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 986 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 987 if ((so->so_state & SS_ISCONFIRMING) == 0 && 988 !(resid == 0 && clen != 0)) { 989 SOCKBUF_UNLOCK(&so->so_snd); 990 error = ENOTCONN; 991 goto out; 992 } 993 } else if (addr == NULL) { 994 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 995 error = ENOTCONN; 996 else 997 error = EDESTADDRREQ; 998 SOCKBUF_UNLOCK(&so->so_snd); 999 goto out; 1000 } 1001 } 1002 1003 /* 1004 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1005 * problem and need fixing. 1006 */ 1007 space = sbspace(&so->so_snd); 1008 if (flags & MSG_OOB) 1009 space += 1024; 1010 space -= clen; 1011 SOCKBUF_UNLOCK(&so->so_snd); 1012 if (resid > space) { 1013 error = EMSGSIZE; 1014 goto out; 1015 } 1016 if (uio == NULL) { 1017 resid = 0; 1018 if (flags & MSG_EOR) 1019 top->m_flags |= M_EOR; 1020 } else { 1021 #ifdef ZERO_COPY_SOCKETS 1022 error = sosend_copyin(uio, &top, atomic, &space, flags); 1023 if (error) 1024 goto out; 1025 #else 1026 /* 1027 * Copy the data from userland into a mbuf chain. 1028 * If no data is to be copied in, a single empty mbuf 1029 * is returned. 1030 */ 1031 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1032 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1033 if (top == NULL) { 1034 error = EFAULT; /* only possible error */ 1035 goto out; 1036 } 1037 space -= resid - uio->uio_resid; 1038 #endif 1039 resid = uio->uio_resid; 1040 } 1041 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1042 /* 1043 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1044 * than with. 1045 */ 1046 if (dontroute) { 1047 SOCK_LOCK(so); 1048 so->so_options |= SO_DONTROUTE; 1049 SOCK_UNLOCK(so); 1050 } 1051 /* 1052 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1053 * of date. We could have recieved a reset packet in an interrupt or 1054 * maybe we slept while doing page faults in uiomove() etc. We could 1055 * probably recheck again inside the locking protection here, but 1056 * there are probably other places that this also happens. We must 1057 * rethink this. 1058 */ 1059 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1060 (flags & MSG_OOB) ? PRUS_OOB : 1061 /* 1062 * If the user set MSG_EOF, the protocol understands this flag and 1063 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1064 */ 1065 ((flags & MSG_EOF) && 1066 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1067 (resid <= 0)) ? 1068 PRUS_EOF : 1069 /* If there is more to send set PRUS_MORETOCOME */ 1070 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1071 top, addr, control, td); 1072 if (dontroute) { 1073 SOCK_LOCK(so); 1074 so->so_options &= ~SO_DONTROUTE; 1075 SOCK_UNLOCK(so); 1076 } 1077 clen = 0; 1078 control = NULL; 1079 top = NULL; 1080 out: 1081 if (top != NULL) 1082 m_freem(top); 1083 if (control != NULL) 1084 m_freem(control); 1085 return (error); 1086 } 1087 1088 /* 1089 * Send on a socket. If send must go all at once and message is larger than 1090 * send buffering, then hard error. Lock against other senders. If must go 1091 * all at once and not enough room now, then inform user that this would 1092 * block and do nothing. Otherwise, if nonblocking, send as much as 1093 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1094 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1095 * in mbuf chain must be small enough to send all at once. 1096 * 1097 * Returns nonzero on error, timeout or signal; callers must check for short 1098 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1099 * on return. 1100 */ 1101 int 1102 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1103 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1104 { 1105 long space, resid; 1106 int clen = 0, error, dontroute; 1107 int atomic = sosendallatonce(so) || top; 1108 1109 if (uio != NULL) 1110 resid = uio->uio_resid; 1111 else 1112 resid = top->m_pkthdr.len; 1113 /* 1114 * In theory resid should be unsigned. However, space must be 1115 * signed, as it might be less than 0 if we over-committed, and we 1116 * must use a signed comparison of space and resid. On the other 1117 * hand, a negative resid causes us to loop sending 0-length 1118 * segments to the protocol. 1119 * 1120 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1121 * type sockets since that's an error. 1122 */ 1123 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1124 error = EINVAL; 1125 goto out; 1126 } 1127 1128 dontroute = 1129 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1130 (so->so_proto->pr_flags & PR_ATOMIC); 1131 if (td != NULL) 1132 td->td_ru.ru_msgsnd++; 1133 if (control != NULL) 1134 clen = control->m_len; 1135 1136 error = sblock(&so->so_snd, SBLOCKWAIT(flags)); 1137 if (error) 1138 goto out; 1139 1140 restart: 1141 do { 1142 SOCKBUF_LOCK(&so->so_snd); 1143 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1144 SOCKBUF_UNLOCK(&so->so_snd); 1145 error = EPIPE; 1146 goto release; 1147 } 1148 if (so->so_error) { 1149 error = so->so_error; 1150 so->so_error = 0; 1151 SOCKBUF_UNLOCK(&so->so_snd); 1152 goto release; 1153 } 1154 if ((so->so_state & SS_ISCONNECTED) == 0) { 1155 /* 1156 * `sendto' and `sendmsg' is allowed on a connection- 1157 * based socket if it supports implied connect. 1158 * Return ENOTCONN if not connected and no address is 1159 * supplied. 1160 */ 1161 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1162 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1163 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1164 !(resid == 0 && clen != 0)) { 1165 SOCKBUF_UNLOCK(&so->so_snd); 1166 error = ENOTCONN; 1167 goto release; 1168 } 1169 } else if (addr == NULL) { 1170 SOCKBUF_UNLOCK(&so->so_snd); 1171 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1172 error = ENOTCONN; 1173 else 1174 error = EDESTADDRREQ; 1175 goto release; 1176 } 1177 } 1178 space = sbspace(&so->so_snd); 1179 if (flags & MSG_OOB) 1180 space += 1024; 1181 if ((atomic && resid > so->so_snd.sb_hiwat) || 1182 clen > so->so_snd.sb_hiwat) { 1183 SOCKBUF_UNLOCK(&so->so_snd); 1184 error = EMSGSIZE; 1185 goto release; 1186 } 1187 if (space < resid + clen && 1188 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1189 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) { 1190 SOCKBUF_UNLOCK(&so->so_snd); 1191 error = EWOULDBLOCK; 1192 goto release; 1193 } 1194 error = sbwait(&so->so_snd); 1195 SOCKBUF_UNLOCK(&so->so_snd); 1196 if (error) 1197 goto release; 1198 goto restart; 1199 } 1200 SOCKBUF_UNLOCK(&so->so_snd); 1201 space -= clen; 1202 do { 1203 if (uio == NULL) { 1204 resid = 0; 1205 if (flags & MSG_EOR) 1206 top->m_flags |= M_EOR; 1207 } else { 1208 #ifdef ZERO_COPY_SOCKETS 1209 error = sosend_copyin(uio, &top, atomic, 1210 &space, flags); 1211 if (error != 0) 1212 goto release; 1213 #else 1214 /* 1215 * Copy the data from userland into a mbuf 1216 * chain. If no data is to be copied in, 1217 * a single empty mbuf is returned. 1218 */ 1219 top = m_uiotombuf(uio, M_WAITOK, space, 1220 (atomic ? max_hdr : 0), 1221 (atomic ? M_PKTHDR : 0) | 1222 ((flags & MSG_EOR) ? M_EOR : 0)); 1223 if (top == NULL) { 1224 error = EFAULT; /* only possible error */ 1225 goto release; 1226 } 1227 space -= resid - uio->uio_resid; 1228 #endif 1229 resid = uio->uio_resid; 1230 } 1231 if (dontroute) { 1232 SOCK_LOCK(so); 1233 so->so_options |= SO_DONTROUTE; 1234 SOCK_UNLOCK(so); 1235 } 1236 /* 1237 * XXX all the SBS_CANTSENDMORE checks previously 1238 * done could be out of date. We could have recieved 1239 * a reset packet in an interrupt or maybe we slept 1240 * while doing page faults in uiomove() etc. We 1241 * could probably recheck again inside the locking 1242 * protection here, but there are probably other 1243 * places that this also happens. We must rethink 1244 * this. 1245 */ 1246 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1247 (flags & MSG_OOB) ? PRUS_OOB : 1248 /* 1249 * If the user set MSG_EOF, the protocol understands 1250 * this flag and nothing left to send then use 1251 * PRU_SEND_EOF instead of PRU_SEND. 1252 */ 1253 ((flags & MSG_EOF) && 1254 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1255 (resid <= 0)) ? 1256 PRUS_EOF : 1257 /* If there is more to send set PRUS_MORETOCOME. */ 1258 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1259 top, addr, control, td); 1260 if (dontroute) { 1261 SOCK_LOCK(so); 1262 so->so_options &= ~SO_DONTROUTE; 1263 SOCK_UNLOCK(so); 1264 } 1265 clen = 0; 1266 control = NULL; 1267 top = NULL; 1268 if (error) 1269 goto release; 1270 } while (resid && space > 0); 1271 } while (resid); 1272 1273 release: 1274 sbunlock(&so->so_snd); 1275 out: 1276 if (top != NULL) 1277 m_freem(top); 1278 if (control != NULL) 1279 m_freem(control); 1280 return (error); 1281 } 1282 1283 int 1284 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1285 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1286 { 1287 1288 /* XXXRW: Temporary debugging. */ 1289 KASSERT(so->so_proto->pr_usrreqs->pru_sosend != sosend, 1290 ("sosend: protocol calls sosend")); 1291 1292 return (so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top, 1293 control, flags, td)); 1294 } 1295 1296 /* 1297 * The part of soreceive() that implements reading non-inline out-of-band 1298 * data from a socket. For more complete comments, see soreceive(), from 1299 * which this code originated. 1300 * 1301 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1302 * unable to return an mbuf chain to the caller. 1303 */ 1304 static int 1305 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1306 { 1307 struct protosw *pr = so->so_proto; 1308 struct mbuf *m; 1309 int error; 1310 1311 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1312 1313 m = m_get(M_WAIT, MT_DATA); 1314 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 1315 if (error) 1316 goto bad; 1317 do { 1318 #ifdef ZERO_COPY_SOCKETS 1319 if (so_zero_copy_receive) { 1320 int disposable; 1321 1322 if ((m->m_flags & M_EXT) 1323 && (m->m_ext.ext_type == EXT_DISPOSABLE)) 1324 disposable = 1; 1325 else 1326 disposable = 0; 1327 1328 error = uiomoveco(mtod(m, void *), 1329 min(uio->uio_resid, m->m_len), 1330 uio, disposable); 1331 } else 1332 #endif /* ZERO_COPY_SOCKETS */ 1333 error = uiomove(mtod(m, void *), 1334 (int) min(uio->uio_resid, m->m_len), uio); 1335 m = m_free(m); 1336 } while (uio->uio_resid && error == 0 && m); 1337 bad: 1338 if (m != NULL) 1339 m_freem(m); 1340 return (error); 1341 } 1342 1343 /* 1344 * Following replacement or removal of the first mbuf on the first mbuf chain 1345 * of a socket buffer, push necessary state changes back into the socket 1346 * buffer so that other consumers see the values consistently. 'nextrecord' 1347 * is the callers locally stored value of the original value of 1348 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1349 * NOTE: 'nextrecord' may be NULL. 1350 */ 1351 static __inline void 1352 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1353 { 1354 1355 SOCKBUF_LOCK_ASSERT(sb); 1356 /* 1357 * First, update for the new value of nextrecord. If necessary, make 1358 * it the first record. 1359 */ 1360 if (sb->sb_mb != NULL) 1361 sb->sb_mb->m_nextpkt = nextrecord; 1362 else 1363 sb->sb_mb = nextrecord; 1364 1365 /* 1366 * Now update any dependent socket buffer fields to reflect the new 1367 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1368 * addition of a second clause that takes care of the case where 1369 * sb_mb has been updated, but remains the last record. 1370 */ 1371 if (sb->sb_mb == NULL) { 1372 sb->sb_mbtail = NULL; 1373 sb->sb_lastrecord = NULL; 1374 } else if (sb->sb_mb->m_nextpkt == NULL) 1375 sb->sb_lastrecord = sb->sb_mb; 1376 } 1377 1378 1379 /* 1380 * Implement receive operations on a socket. We depend on the way that 1381 * records are added to the sockbuf by sbappend. In particular, each record 1382 * (mbufs linked through m_next) must begin with an address if the protocol 1383 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1384 * data, and then zero or more mbufs of data. In order to allow parallelism 1385 * between network receive and copying to user space, as well as avoid 1386 * sleeping with a mutex held, we release the socket buffer mutex during the 1387 * user space copy. Although the sockbuf is locked, new data may still be 1388 * appended, and thus we must maintain consistency of the sockbuf during that 1389 * time. 1390 * 1391 * The caller may receive the data as a single mbuf chain by supplying an 1392 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1393 * the count in uio_resid. 1394 */ 1395 int 1396 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1397 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1398 { 1399 struct mbuf *m, **mp; 1400 int flags, len, error, offset; 1401 struct protosw *pr = so->so_proto; 1402 struct mbuf *nextrecord; 1403 int moff, type = 0; 1404 int orig_resid = uio->uio_resid; 1405 1406 mp = mp0; 1407 if (psa != NULL) 1408 *psa = NULL; 1409 if (controlp != NULL) 1410 *controlp = NULL; 1411 if (flagsp != NULL) 1412 flags = *flagsp &~ MSG_EOR; 1413 else 1414 flags = 0; 1415 if (flags & MSG_OOB) 1416 return (soreceive_rcvoob(so, uio, flags)); 1417 if (mp != NULL) 1418 *mp = NULL; 1419 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 1420 && uio->uio_resid) 1421 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 1422 1423 error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); 1424 if (error) 1425 return (error); 1426 1427 restart: 1428 SOCKBUF_LOCK(&so->so_rcv); 1429 m = so->so_rcv.sb_mb; 1430 /* 1431 * If we have less data than requested, block awaiting more (subject 1432 * to any timeout) if: 1433 * 1. the current count is less than the low water mark, or 1434 * 2. MSG_WAITALL is set, and it is possible to do the entire 1435 * receive operation at once if we block (resid <= hiwat). 1436 * 3. MSG_DONTWAIT is not set 1437 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1438 * we have to do the receive in sections, and thus risk returning a 1439 * short count if a timeout or signal occurs after we start. 1440 */ 1441 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1442 so->so_rcv.sb_cc < uio->uio_resid) && 1443 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1444 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1445 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 1446 KASSERT(m != NULL || !so->so_rcv.sb_cc, 1447 ("receive: m == %p so->so_rcv.sb_cc == %u", 1448 m, so->so_rcv.sb_cc)); 1449 if (so->so_error) { 1450 if (m != NULL) 1451 goto dontblock; 1452 error = so->so_error; 1453 if ((flags & MSG_PEEK) == 0) 1454 so->so_error = 0; 1455 SOCKBUF_UNLOCK(&so->so_rcv); 1456 goto release; 1457 } 1458 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1459 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1460 if (m == NULL) { 1461 SOCKBUF_UNLOCK(&so->so_rcv); 1462 goto release; 1463 } else 1464 goto dontblock; 1465 } 1466 for (; m != NULL; m = m->m_next) 1467 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1468 m = so->so_rcv.sb_mb; 1469 goto dontblock; 1470 } 1471 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1472 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1473 SOCKBUF_UNLOCK(&so->so_rcv); 1474 error = ENOTCONN; 1475 goto release; 1476 } 1477 if (uio->uio_resid == 0) { 1478 SOCKBUF_UNLOCK(&so->so_rcv); 1479 goto release; 1480 } 1481 if ((so->so_state & SS_NBIO) || 1482 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1483 SOCKBUF_UNLOCK(&so->so_rcv); 1484 error = EWOULDBLOCK; 1485 goto release; 1486 } 1487 SBLASTRECORDCHK(&so->so_rcv); 1488 SBLASTMBUFCHK(&so->so_rcv); 1489 error = sbwait(&so->so_rcv); 1490 SOCKBUF_UNLOCK(&so->so_rcv); 1491 if (error) 1492 goto release; 1493 goto restart; 1494 } 1495 dontblock: 1496 /* 1497 * From this point onward, we maintain 'nextrecord' as a cache of the 1498 * pointer to the next record in the socket buffer. We must keep the 1499 * various socket buffer pointers and local stack versions of the 1500 * pointers in sync, pushing out modifications before dropping the 1501 * socket buffer mutex, and re-reading them when picking it up. 1502 * 1503 * Otherwise, we will race with the network stack appending new data 1504 * or records onto the socket buffer by using inconsistent/stale 1505 * versions of the field, possibly resulting in socket buffer 1506 * corruption. 1507 * 1508 * By holding the high-level sblock(), we prevent simultaneous 1509 * readers from pulling off the front of the socket buffer. 1510 */ 1511 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1512 if (uio->uio_td) 1513 uio->uio_td->td_ru.ru_msgrcv++; 1514 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 1515 SBLASTRECORDCHK(&so->so_rcv); 1516 SBLASTMBUFCHK(&so->so_rcv); 1517 nextrecord = m->m_nextpkt; 1518 if (pr->pr_flags & PR_ADDR) { 1519 KASSERT(m->m_type == MT_SONAME, 1520 ("m->m_type == %d", m->m_type)); 1521 orig_resid = 0; 1522 if (psa != NULL) 1523 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 1524 M_NOWAIT); 1525 if (flags & MSG_PEEK) { 1526 m = m->m_next; 1527 } else { 1528 sbfree(&so->so_rcv, m); 1529 so->so_rcv.sb_mb = m_free(m); 1530 m = so->so_rcv.sb_mb; 1531 sockbuf_pushsync(&so->so_rcv, nextrecord); 1532 } 1533 } 1534 1535 /* 1536 * Process one or more MT_CONTROL mbufs present before any data mbufs 1537 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 1538 * just copy the data; if !MSG_PEEK, we call into the protocol to 1539 * perform externalization (or freeing if controlp == NULL). 1540 */ 1541 if (m != NULL && m->m_type == MT_CONTROL) { 1542 struct mbuf *cm = NULL, *cmn; 1543 struct mbuf **cme = &cm; 1544 1545 do { 1546 if (flags & MSG_PEEK) { 1547 if (controlp != NULL) { 1548 *controlp = m_copy(m, 0, m->m_len); 1549 controlp = &(*controlp)->m_next; 1550 } 1551 m = m->m_next; 1552 } else { 1553 sbfree(&so->so_rcv, m); 1554 so->so_rcv.sb_mb = m->m_next; 1555 m->m_next = NULL; 1556 *cme = m; 1557 cme = &(*cme)->m_next; 1558 m = so->so_rcv.sb_mb; 1559 } 1560 } while (m != NULL && m->m_type == MT_CONTROL); 1561 if ((flags & MSG_PEEK) == 0) 1562 sockbuf_pushsync(&so->so_rcv, nextrecord); 1563 while (cm != NULL) { 1564 cmn = cm->m_next; 1565 cm->m_next = NULL; 1566 if (pr->pr_domain->dom_externalize != NULL) { 1567 SOCKBUF_UNLOCK(&so->so_rcv); 1568 error = (*pr->pr_domain->dom_externalize) 1569 (cm, controlp); 1570 SOCKBUF_LOCK(&so->so_rcv); 1571 } else if (controlp != NULL) 1572 *controlp = cm; 1573 else 1574 m_freem(cm); 1575 if (controlp != NULL) { 1576 orig_resid = 0; 1577 while (*controlp != NULL) 1578 controlp = &(*controlp)->m_next; 1579 } 1580 cm = cmn; 1581 } 1582 if (m != NULL) 1583 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 1584 else 1585 nextrecord = so->so_rcv.sb_mb; 1586 orig_resid = 0; 1587 } 1588 if (m != NULL) { 1589 if ((flags & MSG_PEEK) == 0) { 1590 KASSERT(m->m_nextpkt == nextrecord, 1591 ("soreceive: post-control, nextrecord !sync")); 1592 if (nextrecord == NULL) { 1593 KASSERT(so->so_rcv.sb_mb == m, 1594 ("soreceive: post-control, sb_mb!=m")); 1595 KASSERT(so->so_rcv.sb_lastrecord == m, 1596 ("soreceive: post-control, lastrecord!=m")); 1597 } 1598 } 1599 type = m->m_type; 1600 if (type == MT_OOBDATA) 1601 flags |= MSG_OOB; 1602 } else { 1603 if ((flags & MSG_PEEK) == 0) { 1604 KASSERT(so->so_rcv.sb_mb == nextrecord, 1605 ("soreceive: sb_mb != nextrecord")); 1606 if (so->so_rcv.sb_mb == NULL) { 1607 KASSERT(so->so_rcv.sb_lastrecord == NULL, 1608 ("soreceive: sb_lastercord != NULL")); 1609 } 1610 } 1611 } 1612 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1613 SBLASTRECORDCHK(&so->so_rcv); 1614 SBLASTMBUFCHK(&so->so_rcv); 1615 1616 /* 1617 * Now continue to read any data mbufs off of the head of the socket 1618 * buffer until the read request is satisfied. Note that 'type' is 1619 * used to store the type of any mbuf reads that have happened so far 1620 * such that soreceive() can stop reading if the type changes, which 1621 * causes soreceive() to return only one of regular data and inline 1622 * out-of-band data in a single socket receive operation. 1623 */ 1624 moff = 0; 1625 offset = 0; 1626 while (m != NULL && uio->uio_resid > 0 && error == 0) { 1627 /* 1628 * If the type of mbuf has changed since the last mbuf 1629 * examined ('type'), end the receive operation. 1630 */ 1631 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1632 if (m->m_type == MT_OOBDATA) { 1633 if (type != MT_OOBDATA) 1634 break; 1635 } else if (type == MT_OOBDATA) 1636 break; 1637 else 1638 KASSERT(m->m_type == MT_DATA, 1639 ("m->m_type == %d", m->m_type)); 1640 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 1641 len = uio->uio_resid; 1642 if (so->so_oobmark && len > so->so_oobmark - offset) 1643 len = so->so_oobmark - offset; 1644 if (len > m->m_len - moff) 1645 len = m->m_len - moff; 1646 /* 1647 * If mp is set, just pass back the mbufs. Otherwise copy 1648 * them out via the uio, then free. Sockbuf must be 1649 * consistent here (points to current mbuf, it points to next 1650 * record) when we drop priority; we must note any additions 1651 * to the sockbuf when we block interrupts again. 1652 */ 1653 if (mp == NULL) { 1654 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1655 SBLASTRECORDCHK(&so->so_rcv); 1656 SBLASTMBUFCHK(&so->so_rcv); 1657 SOCKBUF_UNLOCK(&so->so_rcv); 1658 #ifdef ZERO_COPY_SOCKETS 1659 if (so_zero_copy_receive) { 1660 int disposable; 1661 1662 if ((m->m_flags & M_EXT) 1663 && (m->m_ext.ext_type == EXT_DISPOSABLE)) 1664 disposable = 1; 1665 else 1666 disposable = 0; 1667 1668 error = uiomoveco(mtod(m, char *) + moff, 1669 (int)len, uio, 1670 disposable); 1671 } else 1672 #endif /* ZERO_COPY_SOCKETS */ 1673 error = uiomove(mtod(m, char *) + moff, (int)len, uio); 1674 SOCKBUF_LOCK(&so->so_rcv); 1675 if (error) { 1676 /* 1677 * The MT_SONAME mbuf has already been removed 1678 * from the record, so it is necessary to 1679 * remove the data mbufs, if any, to preserve 1680 * the invariant in the case of PR_ADDR that 1681 * requires MT_SONAME mbufs at the head of 1682 * each record. 1683 */ 1684 if (m && pr->pr_flags & PR_ATOMIC && 1685 ((flags & MSG_PEEK) == 0)) 1686 (void)sbdroprecord_locked(&so->so_rcv); 1687 SOCKBUF_UNLOCK(&so->so_rcv); 1688 goto release; 1689 } 1690 } else 1691 uio->uio_resid -= len; 1692 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1693 if (len == m->m_len - moff) { 1694 if (m->m_flags & M_EOR) 1695 flags |= MSG_EOR; 1696 if (flags & MSG_PEEK) { 1697 m = m->m_next; 1698 moff = 0; 1699 } else { 1700 nextrecord = m->m_nextpkt; 1701 sbfree(&so->so_rcv, m); 1702 if (mp != NULL) { 1703 *mp = m; 1704 mp = &m->m_next; 1705 so->so_rcv.sb_mb = m = m->m_next; 1706 *mp = NULL; 1707 } else { 1708 so->so_rcv.sb_mb = m_free(m); 1709 m = so->so_rcv.sb_mb; 1710 } 1711 sockbuf_pushsync(&so->so_rcv, nextrecord); 1712 SBLASTRECORDCHK(&so->so_rcv); 1713 SBLASTMBUFCHK(&so->so_rcv); 1714 } 1715 } else { 1716 if (flags & MSG_PEEK) 1717 moff += len; 1718 else { 1719 if (mp != NULL) { 1720 int copy_flag; 1721 1722 if (flags & MSG_DONTWAIT) 1723 copy_flag = M_DONTWAIT; 1724 else 1725 copy_flag = M_WAIT; 1726 if (copy_flag == M_WAIT) 1727 SOCKBUF_UNLOCK(&so->so_rcv); 1728 *mp = m_copym(m, 0, len, copy_flag); 1729 if (copy_flag == M_WAIT) 1730 SOCKBUF_LOCK(&so->so_rcv); 1731 if (*mp == NULL) { 1732 /* 1733 * m_copym() couldn't 1734 * allocate an mbuf. Adjust 1735 * uio_resid back (it was 1736 * adjusted down by len 1737 * bytes, which we didn't end 1738 * up "copying" over). 1739 */ 1740 uio->uio_resid += len; 1741 break; 1742 } 1743 } 1744 m->m_data += len; 1745 m->m_len -= len; 1746 so->so_rcv.sb_cc -= len; 1747 } 1748 } 1749 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1750 if (so->so_oobmark) { 1751 if ((flags & MSG_PEEK) == 0) { 1752 so->so_oobmark -= len; 1753 if (so->so_oobmark == 0) { 1754 so->so_rcv.sb_state |= SBS_RCVATMARK; 1755 break; 1756 } 1757 } else { 1758 offset += len; 1759 if (offset == so->so_oobmark) 1760 break; 1761 } 1762 } 1763 if (flags & MSG_EOR) 1764 break; 1765 /* 1766 * If the MSG_WAITALL flag is set (for non-atomic socket), we 1767 * must not quit until "uio->uio_resid == 0" or an error 1768 * termination. If a signal/timeout occurs, return with a 1769 * short count but without error. Keep sockbuf locked 1770 * against other readers. 1771 */ 1772 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 1773 !sosendallatonce(so) && nextrecord == NULL) { 1774 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1775 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE) 1776 break; 1777 /* 1778 * Notify the protocol that some data has been 1779 * drained before blocking. 1780 */ 1781 if (pr->pr_flags & PR_WANTRCVD) { 1782 SOCKBUF_UNLOCK(&so->so_rcv); 1783 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 1784 SOCKBUF_LOCK(&so->so_rcv); 1785 } 1786 SBLASTRECORDCHK(&so->so_rcv); 1787 SBLASTMBUFCHK(&so->so_rcv); 1788 error = sbwait(&so->so_rcv); 1789 if (error) { 1790 SOCKBUF_UNLOCK(&so->so_rcv); 1791 goto release; 1792 } 1793 m = so->so_rcv.sb_mb; 1794 if (m != NULL) 1795 nextrecord = m->m_nextpkt; 1796 } 1797 } 1798 1799 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1800 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 1801 flags |= MSG_TRUNC; 1802 if ((flags & MSG_PEEK) == 0) 1803 (void) sbdroprecord_locked(&so->so_rcv); 1804 } 1805 if ((flags & MSG_PEEK) == 0) { 1806 if (m == NULL) { 1807 /* 1808 * First part is an inline SB_EMPTY_FIXUP(). Second 1809 * part makes sure sb_lastrecord is up-to-date if 1810 * there is still data in the socket buffer. 1811 */ 1812 so->so_rcv.sb_mb = nextrecord; 1813 if (so->so_rcv.sb_mb == NULL) { 1814 so->so_rcv.sb_mbtail = NULL; 1815 so->so_rcv.sb_lastrecord = NULL; 1816 } else if (nextrecord->m_nextpkt == NULL) 1817 so->so_rcv.sb_lastrecord = nextrecord; 1818 } 1819 SBLASTRECORDCHK(&so->so_rcv); 1820 SBLASTMBUFCHK(&so->so_rcv); 1821 /* 1822 * If soreceive() is being done from the socket callback, 1823 * then don't need to generate ACK to peer to update window, 1824 * since ACK will be generated on return to TCP. 1825 */ 1826 if (!(flags & MSG_SOCALLBCK) && 1827 (pr->pr_flags & PR_WANTRCVD)) { 1828 SOCKBUF_UNLOCK(&so->so_rcv); 1829 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 1830 SOCKBUF_LOCK(&so->so_rcv); 1831 } 1832 } 1833 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1834 if (orig_resid == uio->uio_resid && orig_resid && 1835 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 1836 SOCKBUF_UNLOCK(&so->so_rcv); 1837 goto restart; 1838 } 1839 SOCKBUF_UNLOCK(&so->so_rcv); 1840 1841 if (flagsp != NULL) 1842 *flagsp |= flags; 1843 release: 1844 sbunlock(&so->so_rcv); 1845 return (error); 1846 } 1847 1848 int 1849 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 1850 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1851 { 1852 1853 /* XXXRW: Temporary debugging. */ 1854 KASSERT(so->so_proto->pr_usrreqs->pru_soreceive != soreceive, 1855 ("soreceive: protocol calls soreceive")); 1856 1857 return (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0, 1858 controlp, flagsp)); 1859 } 1860 1861 int 1862 soshutdown(struct socket *so, int how) 1863 { 1864 struct protosw *pr = so->so_proto; 1865 1866 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 1867 return (EINVAL); 1868 if (pr->pr_usrreqs->pru_flush != NULL) { 1869 (*pr->pr_usrreqs->pru_flush)(so, how); 1870 } 1871 if (how != SHUT_WR) 1872 sorflush(so); 1873 if (how != SHUT_RD) 1874 return ((*pr->pr_usrreqs->pru_shutdown)(so)); 1875 return (0); 1876 } 1877 1878 void 1879 sorflush(struct socket *so) 1880 { 1881 struct sockbuf *sb = &so->so_rcv; 1882 struct protosw *pr = so->so_proto; 1883 struct sockbuf asb; 1884 1885 /* 1886 * In order to avoid calling dom_dispose with the socket buffer mutex 1887 * held, and in order to generally avoid holding the lock for a long 1888 * time, we make a copy of the socket buffer and clear the original 1889 * (except locks, state). The new socket buffer copy won't have 1890 * initialized locks so we can only call routines that won't use or 1891 * assert those locks. 1892 * 1893 * Dislodge threads currently blocked in receive and wait to acquire 1894 * a lock against other simultaneous readers before clearing the 1895 * socket buffer. Don't let our acquire be interrupted by a signal 1896 * despite any existing socket disposition on interruptable waiting. 1897 */ 1898 socantrcvmore(so); 1899 (void) sblock(sb, SBL_WAIT | SBL_NOINTR); 1900 1901 /* 1902 * Invalidate/clear most of the sockbuf structure, but leave selinfo 1903 * and mutex data unchanged. 1904 */ 1905 SOCKBUF_LOCK(sb); 1906 bzero(&asb, offsetof(struct sockbuf, sb_startzero)); 1907 bcopy(&sb->sb_startzero, &asb.sb_startzero, 1908 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1909 bzero(&sb->sb_startzero, 1910 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 1911 SOCKBUF_UNLOCK(sb); 1912 sbunlock(sb); 1913 1914 /* 1915 * Dispose of special rights and flush the socket buffer. Don't call 1916 * any unsafe routines (that rely on locks being initialized) on asb. 1917 */ 1918 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 1919 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 1920 sbrelease_internal(&asb, so); 1921 } 1922 1923 /* 1924 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 1925 * additional variant to handle the case where the option value needs to be 1926 * some kind of integer, but not a specific size. In addition to their use 1927 * here, these functions are also called by the protocol-level pr_ctloutput() 1928 * routines. 1929 */ 1930 int 1931 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 1932 { 1933 size_t valsize; 1934 1935 /* 1936 * If the user gives us more than we wanted, we ignore it, but if we 1937 * don't get the minimum length the caller wants, we return EINVAL. 1938 * On success, sopt->sopt_valsize is set to however much we actually 1939 * retrieved. 1940 */ 1941 if ((valsize = sopt->sopt_valsize) < minlen) 1942 return EINVAL; 1943 if (valsize > len) 1944 sopt->sopt_valsize = valsize = len; 1945 1946 if (sopt->sopt_td != NULL) 1947 return (copyin(sopt->sopt_val, buf, valsize)); 1948 1949 bcopy(sopt->sopt_val, buf, valsize); 1950 return (0); 1951 } 1952 1953 /* 1954 * Kernel version of setsockopt(2). 1955 * 1956 * XXX: optlen is size_t, not socklen_t 1957 */ 1958 int 1959 so_setsockopt(struct socket *so, int level, int optname, void *optval, 1960 size_t optlen) 1961 { 1962 struct sockopt sopt; 1963 1964 sopt.sopt_level = level; 1965 sopt.sopt_name = optname; 1966 sopt.sopt_dir = SOPT_SET; 1967 sopt.sopt_val = optval; 1968 sopt.sopt_valsize = optlen; 1969 sopt.sopt_td = NULL; 1970 return (sosetopt(so, &sopt)); 1971 } 1972 1973 int 1974 sosetopt(struct socket *so, struct sockopt *sopt) 1975 { 1976 int error, optval; 1977 struct linger l; 1978 struct timeval tv; 1979 u_long val; 1980 #ifdef MAC 1981 struct mac extmac; 1982 #endif 1983 1984 error = 0; 1985 if (sopt->sopt_level != SOL_SOCKET) { 1986 if (so->so_proto && so->so_proto->pr_ctloutput) 1987 return ((*so->so_proto->pr_ctloutput) 1988 (so, sopt)); 1989 error = ENOPROTOOPT; 1990 } else { 1991 switch (sopt->sopt_name) { 1992 #ifdef INET 1993 case SO_ACCEPTFILTER: 1994 error = do_setopt_accept_filter(so, sopt); 1995 if (error) 1996 goto bad; 1997 break; 1998 #endif 1999 case SO_LINGER: 2000 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 2001 if (error) 2002 goto bad; 2003 2004 SOCK_LOCK(so); 2005 so->so_linger = l.l_linger; 2006 if (l.l_onoff) 2007 so->so_options |= SO_LINGER; 2008 else 2009 so->so_options &= ~SO_LINGER; 2010 SOCK_UNLOCK(so); 2011 break; 2012 2013 case SO_DEBUG: 2014 case SO_KEEPALIVE: 2015 case SO_DONTROUTE: 2016 case SO_USELOOPBACK: 2017 case SO_BROADCAST: 2018 case SO_REUSEADDR: 2019 case SO_REUSEPORT: 2020 case SO_OOBINLINE: 2021 case SO_TIMESTAMP: 2022 case SO_BINTIME: 2023 case SO_NOSIGPIPE: 2024 error = sooptcopyin(sopt, &optval, sizeof optval, 2025 sizeof optval); 2026 if (error) 2027 goto bad; 2028 SOCK_LOCK(so); 2029 if (optval) 2030 so->so_options |= sopt->sopt_name; 2031 else 2032 so->so_options &= ~sopt->sopt_name; 2033 SOCK_UNLOCK(so); 2034 break; 2035 2036 case SO_SETFIB: 2037 error = sooptcopyin(sopt, &optval, sizeof optval, 2038 sizeof optval); 2039 if (optval < 1 || optval > rt_numfibs) { 2040 error = EINVAL; 2041 goto bad; 2042 } 2043 if ((so->so_proto->pr_domain->dom_family == PF_INET) || 2044 (so->so_proto->pr_domain->dom_family == PF_ROUTE)) { 2045 so->so_fibnum = optval; 2046 } else { 2047 so->so_fibnum = 0; 2048 } 2049 break; 2050 case SO_SNDBUF: 2051 case SO_RCVBUF: 2052 case SO_SNDLOWAT: 2053 case SO_RCVLOWAT: 2054 error = sooptcopyin(sopt, &optval, sizeof optval, 2055 sizeof optval); 2056 if (error) 2057 goto bad; 2058 2059 /* 2060 * Values < 1 make no sense for any of these options, 2061 * so disallow them. 2062 */ 2063 if (optval < 1) { 2064 error = EINVAL; 2065 goto bad; 2066 } 2067 2068 switch (sopt->sopt_name) { 2069 case SO_SNDBUF: 2070 case SO_RCVBUF: 2071 if (sbreserve(sopt->sopt_name == SO_SNDBUF ? 2072 &so->so_snd : &so->so_rcv, (u_long)optval, 2073 so, curthread) == 0) { 2074 error = ENOBUFS; 2075 goto bad; 2076 } 2077 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd : 2078 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE; 2079 break; 2080 2081 /* 2082 * Make sure the low-water is never greater than the 2083 * high-water. 2084 */ 2085 case SO_SNDLOWAT: 2086 SOCKBUF_LOCK(&so->so_snd); 2087 so->so_snd.sb_lowat = 2088 (optval > so->so_snd.sb_hiwat) ? 2089 so->so_snd.sb_hiwat : optval; 2090 SOCKBUF_UNLOCK(&so->so_snd); 2091 break; 2092 case SO_RCVLOWAT: 2093 SOCKBUF_LOCK(&so->so_rcv); 2094 so->so_rcv.sb_lowat = 2095 (optval > so->so_rcv.sb_hiwat) ? 2096 so->so_rcv.sb_hiwat : optval; 2097 SOCKBUF_UNLOCK(&so->so_rcv); 2098 break; 2099 } 2100 break; 2101 2102 case SO_SNDTIMEO: 2103 case SO_RCVTIMEO: 2104 #ifdef COMPAT_IA32 2105 if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) { 2106 struct timeval32 tv32; 2107 2108 error = sooptcopyin(sopt, &tv32, sizeof tv32, 2109 sizeof tv32); 2110 CP(tv32, tv, tv_sec); 2111 CP(tv32, tv, tv_usec); 2112 } else 2113 #endif 2114 error = sooptcopyin(sopt, &tv, sizeof tv, 2115 sizeof tv); 2116 if (error) 2117 goto bad; 2118 2119 /* assert(hz > 0); */ 2120 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz || 2121 tv.tv_usec < 0 || tv.tv_usec >= 1000000) { 2122 error = EDOM; 2123 goto bad; 2124 } 2125 /* assert(tick > 0); */ 2126 /* assert(ULONG_MAX - INT_MAX >= 1000000); */ 2127 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick; 2128 if (val > INT_MAX) { 2129 error = EDOM; 2130 goto bad; 2131 } 2132 if (val == 0 && tv.tv_usec != 0) 2133 val = 1; 2134 2135 switch (sopt->sopt_name) { 2136 case SO_SNDTIMEO: 2137 so->so_snd.sb_timeo = val; 2138 break; 2139 case SO_RCVTIMEO: 2140 so->so_rcv.sb_timeo = val; 2141 break; 2142 } 2143 break; 2144 2145 case SO_LABEL: 2146 #ifdef MAC 2147 error = sooptcopyin(sopt, &extmac, sizeof extmac, 2148 sizeof extmac); 2149 if (error) 2150 goto bad; 2151 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 2152 so, &extmac); 2153 #else 2154 error = EOPNOTSUPP; 2155 #endif 2156 break; 2157 2158 default: 2159 error = ENOPROTOOPT; 2160 break; 2161 } 2162 if (error == 0 && so->so_proto != NULL && 2163 so->so_proto->pr_ctloutput != NULL) { 2164 (void) ((*so->so_proto->pr_ctloutput) 2165 (so, sopt)); 2166 } 2167 } 2168 bad: 2169 return (error); 2170 } 2171 2172 /* 2173 * Helper routine for getsockopt. 2174 */ 2175 int 2176 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 2177 { 2178 int error; 2179 size_t valsize; 2180 2181 error = 0; 2182 2183 /* 2184 * Documented get behavior is that we always return a value, possibly 2185 * truncated to fit in the user's buffer. Traditional behavior is 2186 * that we always tell the user precisely how much we copied, rather 2187 * than something useful like the total amount we had available for 2188 * her. Note that this interface is not idempotent; the entire 2189 * answer must generated ahead of time. 2190 */ 2191 valsize = min(len, sopt->sopt_valsize); 2192 sopt->sopt_valsize = valsize; 2193 if (sopt->sopt_val != NULL) { 2194 if (sopt->sopt_td != NULL) 2195 error = copyout(buf, sopt->sopt_val, valsize); 2196 else 2197 bcopy(buf, sopt->sopt_val, valsize); 2198 } 2199 return (error); 2200 } 2201 2202 int 2203 sogetopt(struct socket *so, struct sockopt *sopt) 2204 { 2205 int error, optval; 2206 struct linger l; 2207 struct timeval tv; 2208 #ifdef MAC 2209 struct mac extmac; 2210 #endif 2211 2212 error = 0; 2213 if (sopt->sopt_level != SOL_SOCKET) { 2214 if (so->so_proto && so->so_proto->pr_ctloutput) { 2215 return ((*so->so_proto->pr_ctloutput) 2216 (so, sopt)); 2217 } else 2218 return (ENOPROTOOPT); 2219 } else { 2220 switch (sopt->sopt_name) { 2221 #ifdef INET 2222 case SO_ACCEPTFILTER: 2223 error = do_getopt_accept_filter(so, sopt); 2224 break; 2225 #endif 2226 case SO_LINGER: 2227 SOCK_LOCK(so); 2228 l.l_onoff = so->so_options & SO_LINGER; 2229 l.l_linger = so->so_linger; 2230 SOCK_UNLOCK(so); 2231 error = sooptcopyout(sopt, &l, sizeof l); 2232 break; 2233 2234 case SO_USELOOPBACK: 2235 case SO_DONTROUTE: 2236 case SO_DEBUG: 2237 case SO_KEEPALIVE: 2238 case SO_REUSEADDR: 2239 case SO_REUSEPORT: 2240 case SO_BROADCAST: 2241 case SO_OOBINLINE: 2242 case SO_ACCEPTCONN: 2243 case SO_TIMESTAMP: 2244 case SO_BINTIME: 2245 case SO_NOSIGPIPE: 2246 optval = so->so_options & sopt->sopt_name; 2247 integer: 2248 error = sooptcopyout(sopt, &optval, sizeof optval); 2249 break; 2250 2251 case SO_TYPE: 2252 optval = so->so_type; 2253 goto integer; 2254 2255 case SO_ERROR: 2256 SOCK_LOCK(so); 2257 optval = so->so_error; 2258 so->so_error = 0; 2259 SOCK_UNLOCK(so); 2260 goto integer; 2261 2262 case SO_SNDBUF: 2263 optval = so->so_snd.sb_hiwat; 2264 goto integer; 2265 2266 case SO_RCVBUF: 2267 optval = so->so_rcv.sb_hiwat; 2268 goto integer; 2269 2270 case SO_SNDLOWAT: 2271 optval = so->so_snd.sb_lowat; 2272 goto integer; 2273 2274 case SO_RCVLOWAT: 2275 optval = so->so_rcv.sb_lowat; 2276 goto integer; 2277 2278 case SO_SNDTIMEO: 2279 case SO_RCVTIMEO: 2280 optval = (sopt->sopt_name == SO_SNDTIMEO ? 2281 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 2282 2283 tv.tv_sec = optval / hz; 2284 tv.tv_usec = (optval % hz) * tick; 2285 #ifdef COMPAT_IA32 2286 if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) { 2287 struct timeval32 tv32; 2288 2289 CP(tv, tv32, tv_sec); 2290 CP(tv, tv32, tv_usec); 2291 error = sooptcopyout(sopt, &tv32, sizeof tv32); 2292 } else 2293 #endif 2294 error = sooptcopyout(sopt, &tv, sizeof tv); 2295 break; 2296 2297 case SO_LABEL: 2298 #ifdef MAC 2299 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 2300 sizeof(extmac)); 2301 if (error) 2302 return (error); 2303 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 2304 so, &extmac); 2305 if (error) 2306 return (error); 2307 error = sooptcopyout(sopt, &extmac, sizeof extmac); 2308 #else 2309 error = EOPNOTSUPP; 2310 #endif 2311 break; 2312 2313 case SO_PEERLABEL: 2314 #ifdef MAC 2315 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 2316 sizeof(extmac)); 2317 if (error) 2318 return (error); 2319 error = mac_getsockopt_peerlabel( 2320 sopt->sopt_td->td_ucred, so, &extmac); 2321 if (error) 2322 return (error); 2323 error = sooptcopyout(sopt, &extmac, sizeof extmac); 2324 #else 2325 error = EOPNOTSUPP; 2326 #endif 2327 break; 2328 2329 case SO_LISTENQLIMIT: 2330 optval = so->so_qlimit; 2331 goto integer; 2332 2333 case SO_LISTENQLEN: 2334 optval = so->so_qlen; 2335 goto integer; 2336 2337 case SO_LISTENINCQLEN: 2338 optval = so->so_incqlen; 2339 goto integer; 2340 2341 default: 2342 error = ENOPROTOOPT; 2343 break; 2344 } 2345 return (error); 2346 } 2347 } 2348 2349 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */ 2350 int 2351 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 2352 { 2353 struct mbuf *m, *m_prev; 2354 int sopt_size = sopt->sopt_valsize; 2355 2356 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA); 2357 if (m == NULL) 2358 return ENOBUFS; 2359 if (sopt_size > MLEN) { 2360 MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT); 2361 if ((m->m_flags & M_EXT) == 0) { 2362 m_free(m); 2363 return ENOBUFS; 2364 } 2365 m->m_len = min(MCLBYTES, sopt_size); 2366 } else { 2367 m->m_len = min(MLEN, sopt_size); 2368 } 2369 sopt_size -= m->m_len; 2370 *mp = m; 2371 m_prev = m; 2372 2373 while (sopt_size) { 2374 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA); 2375 if (m == NULL) { 2376 m_freem(*mp); 2377 return ENOBUFS; 2378 } 2379 if (sopt_size > MLEN) { 2380 MCLGET(m, sopt->sopt_td != NULL ? M_WAIT : 2381 M_DONTWAIT); 2382 if ((m->m_flags & M_EXT) == 0) { 2383 m_freem(m); 2384 m_freem(*mp); 2385 return ENOBUFS; 2386 } 2387 m->m_len = min(MCLBYTES, sopt_size); 2388 } else { 2389 m->m_len = min(MLEN, sopt_size); 2390 } 2391 sopt_size -= m->m_len; 2392 m_prev->m_next = m; 2393 m_prev = m; 2394 } 2395 return (0); 2396 } 2397 2398 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */ 2399 int 2400 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 2401 { 2402 struct mbuf *m0 = m; 2403 2404 if (sopt->sopt_val == NULL) 2405 return (0); 2406 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 2407 if (sopt->sopt_td != NULL) { 2408 int error; 2409 2410 error = copyin(sopt->sopt_val, mtod(m, char *), 2411 m->m_len); 2412 if (error != 0) { 2413 m_freem(m0); 2414 return(error); 2415 } 2416 } else 2417 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 2418 sopt->sopt_valsize -= m->m_len; 2419 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 2420 m = m->m_next; 2421 } 2422 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 2423 panic("ip6_sooptmcopyin"); 2424 return (0); 2425 } 2426 2427 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */ 2428 int 2429 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 2430 { 2431 struct mbuf *m0 = m; 2432 size_t valsize = 0; 2433 2434 if (sopt->sopt_val == NULL) 2435 return (0); 2436 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 2437 if (sopt->sopt_td != NULL) { 2438 int error; 2439 2440 error = copyout(mtod(m, char *), sopt->sopt_val, 2441 m->m_len); 2442 if (error != 0) { 2443 m_freem(m0); 2444 return(error); 2445 } 2446 } else 2447 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 2448 sopt->sopt_valsize -= m->m_len; 2449 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 2450 valsize += m->m_len; 2451 m = m->m_next; 2452 } 2453 if (m != NULL) { 2454 /* enough soopt buffer should be given from user-land */ 2455 m_freem(m0); 2456 return(EINVAL); 2457 } 2458 sopt->sopt_valsize = valsize; 2459 return (0); 2460 } 2461 2462 /* 2463 * sohasoutofband(): protocol notifies socket layer of the arrival of new 2464 * out-of-band data, which will then notify socket consumers. 2465 */ 2466 void 2467 sohasoutofband(struct socket *so) 2468 { 2469 2470 if (so->so_sigio != NULL) 2471 pgsigio(&so->so_sigio, SIGURG, 0); 2472 selwakeuppri(&so->so_rcv.sb_sel, PSOCK); 2473 } 2474 2475 int 2476 sopoll(struct socket *so, int events, struct ucred *active_cred, 2477 struct thread *td) 2478 { 2479 2480 /* XXXRW: Temporary debugging. */ 2481 KASSERT(so->so_proto->pr_usrreqs->pru_sopoll != sopoll, 2482 ("sopoll: protocol calls sopoll")); 2483 2484 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, 2485 td)); 2486 } 2487 2488 int 2489 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 2490 struct thread *td) 2491 { 2492 int revents = 0; 2493 2494 SOCKBUF_LOCK(&so->so_snd); 2495 SOCKBUF_LOCK(&so->so_rcv); 2496 if (events & (POLLIN | POLLRDNORM)) 2497 if (soreadable(so)) 2498 revents |= events & (POLLIN | POLLRDNORM); 2499 2500 if (events & POLLINIGNEOF) 2501 if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat || 2502 !TAILQ_EMPTY(&so->so_comp) || so->so_error) 2503 revents |= POLLINIGNEOF; 2504 2505 if (events & (POLLOUT | POLLWRNORM)) 2506 if (sowriteable(so)) 2507 revents |= events & (POLLOUT | POLLWRNORM); 2508 2509 if (events & (POLLPRI | POLLRDBAND)) 2510 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK)) 2511 revents |= events & (POLLPRI | POLLRDBAND); 2512 2513 if (revents == 0) { 2514 if (events & 2515 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM | 2516 POLLRDBAND)) { 2517 selrecord(td, &so->so_rcv.sb_sel); 2518 so->so_rcv.sb_flags |= SB_SEL; 2519 } 2520 2521 if (events & (POLLOUT | POLLWRNORM)) { 2522 selrecord(td, &so->so_snd.sb_sel); 2523 so->so_snd.sb_flags |= SB_SEL; 2524 } 2525 } 2526 2527 SOCKBUF_UNLOCK(&so->so_rcv); 2528 SOCKBUF_UNLOCK(&so->so_snd); 2529 return (revents); 2530 } 2531 2532 int 2533 soo_kqfilter(struct file *fp, struct knote *kn) 2534 { 2535 struct socket *so = kn->kn_fp->f_data; 2536 struct sockbuf *sb; 2537 2538 switch (kn->kn_filter) { 2539 case EVFILT_READ: 2540 if (so->so_options & SO_ACCEPTCONN) 2541 kn->kn_fop = &solisten_filtops; 2542 else 2543 kn->kn_fop = &soread_filtops; 2544 sb = &so->so_rcv; 2545 break; 2546 case EVFILT_WRITE: 2547 kn->kn_fop = &sowrite_filtops; 2548 sb = &so->so_snd; 2549 break; 2550 default: 2551 return (EINVAL); 2552 } 2553 2554 SOCKBUF_LOCK(sb); 2555 knlist_add(&sb->sb_sel.si_note, kn, 1); 2556 sb->sb_flags |= SB_KNOTE; 2557 SOCKBUF_UNLOCK(sb); 2558 return (0); 2559 } 2560 2561 /* 2562 * Some routines that return EOPNOTSUPP for entry points that are not 2563 * supported by a protocol. Fill in as needed. 2564 */ 2565 int 2566 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 2567 { 2568 2569 return EOPNOTSUPP; 2570 } 2571 2572 int 2573 pru_attach_notsupp(struct socket *so, int proto, struct thread *td) 2574 { 2575 2576 return EOPNOTSUPP; 2577 } 2578 2579 int 2580 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 2581 { 2582 2583 return EOPNOTSUPP; 2584 } 2585 2586 int 2587 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 2588 { 2589 2590 return EOPNOTSUPP; 2591 } 2592 2593 int 2594 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 2595 { 2596 2597 return EOPNOTSUPP; 2598 } 2599 2600 int 2601 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 2602 struct ifnet *ifp, struct thread *td) 2603 { 2604 2605 return EOPNOTSUPP; 2606 } 2607 2608 int 2609 pru_disconnect_notsupp(struct socket *so) 2610 { 2611 2612 return EOPNOTSUPP; 2613 } 2614 2615 int 2616 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) 2617 { 2618 2619 return EOPNOTSUPP; 2620 } 2621 2622 int 2623 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) 2624 { 2625 2626 return EOPNOTSUPP; 2627 } 2628 2629 int 2630 pru_rcvd_notsupp(struct socket *so, int flags) 2631 { 2632 2633 return EOPNOTSUPP; 2634 } 2635 2636 int 2637 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 2638 { 2639 2640 return EOPNOTSUPP; 2641 } 2642 2643 int 2644 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, 2645 struct sockaddr *addr, struct mbuf *control, struct thread *td) 2646 { 2647 2648 return EOPNOTSUPP; 2649 } 2650 2651 /* 2652 * This isn't really a ``null'' operation, but it's the default one and 2653 * doesn't do anything destructive. 2654 */ 2655 int 2656 pru_sense_null(struct socket *so, struct stat *sb) 2657 { 2658 2659 sb->st_blksize = so->so_snd.sb_hiwat; 2660 return 0; 2661 } 2662 2663 int 2664 pru_shutdown_notsupp(struct socket *so) 2665 { 2666 2667 return EOPNOTSUPP; 2668 } 2669 2670 int 2671 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) 2672 { 2673 2674 return EOPNOTSUPP; 2675 } 2676 2677 int 2678 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 2679 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 2680 { 2681 2682 return EOPNOTSUPP; 2683 } 2684 2685 int 2686 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 2687 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2688 { 2689 2690 return EOPNOTSUPP; 2691 } 2692 2693 int 2694 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, 2695 struct thread *td) 2696 { 2697 2698 return EOPNOTSUPP; 2699 } 2700 2701 static void 2702 filt_sordetach(struct knote *kn) 2703 { 2704 struct socket *so = kn->kn_fp->f_data; 2705 2706 SOCKBUF_LOCK(&so->so_rcv); 2707 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1); 2708 if (knlist_empty(&so->so_rcv.sb_sel.si_note)) 2709 so->so_rcv.sb_flags &= ~SB_KNOTE; 2710 SOCKBUF_UNLOCK(&so->so_rcv); 2711 } 2712 2713 /*ARGSUSED*/ 2714 static int 2715 filt_soread(struct knote *kn, long hint) 2716 { 2717 struct socket *so; 2718 2719 so = kn->kn_fp->f_data; 2720 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2721 2722 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl; 2723 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2724 kn->kn_flags |= EV_EOF; 2725 kn->kn_fflags = so->so_error; 2726 return (1); 2727 } else if (so->so_error) /* temporary udp error */ 2728 return (1); 2729 else if (kn->kn_sfflags & NOTE_LOWAT) 2730 return (kn->kn_data >= kn->kn_sdata); 2731 else 2732 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat); 2733 } 2734 2735 static void 2736 filt_sowdetach(struct knote *kn) 2737 { 2738 struct socket *so = kn->kn_fp->f_data; 2739 2740 SOCKBUF_LOCK(&so->so_snd); 2741 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1); 2742 if (knlist_empty(&so->so_snd.sb_sel.si_note)) 2743 so->so_snd.sb_flags &= ~SB_KNOTE; 2744 SOCKBUF_UNLOCK(&so->so_snd); 2745 } 2746 2747 /*ARGSUSED*/ 2748 static int 2749 filt_sowrite(struct knote *kn, long hint) 2750 { 2751 struct socket *so; 2752 2753 so = kn->kn_fp->f_data; 2754 SOCKBUF_LOCK_ASSERT(&so->so_snd); 2755 kn->kn_data = sbspace(&so->so_snd); 2756 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 2757 kn->kn_flags |= EV_EOF; 2758 kn->kn_fflags = so->so_error; 2759 return (1); 2760 } else if (so->so_error) /* temporary udp error */ 2761 return (1); 2762 else if (((so->so_state & SS_ISCONNECTED) == 0) && 2763 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 2764 return (0); 2765 else if (kn->kn_sfflags & NOTE_LOWAT) 2766 return (kn->kn_data >= kn->kn_sdata); 2767 else 2768 return (kn->kn_data >= so->so_snd.sb_lowat); 2769 } 2770 2771 /*ARGSUSED*/ 2772 static int 2773 filt_solisten(struct knote *kn, long hint) 2774 { 2775 struct socket *so = kn->kn_fp->f_data; 2776 2777 kn->kn_data = so->so_qlen; 2778 return (! TAILQ_EMPTY(&so->so_comp)); 2779 } 2780 2781 int 2782 socheckuid(struct socket *so, uid_t uid) 2783 { 2784 2785 if (so == NULL) 2786 return (EPERM); 2787 if (so->so_cred->cr_uid != uid) 2788 return (EPERM); 2789 return (0); 2790 } 2791 2792 static int 2793 sysctl_somaxconn(SYSCTL_HANDLER_ARGS) 2794 { 2795 int error; 2796 int val; 2797 2798 val = somaxconn; 2799 error = sysctl_handle_int(oidp, &val, 0, req); 2800 if (error || !req->newptr ) 2801 return (error); 2802 2803 if (val < 1 || val > USHRT_MAX) 2804 return (EINVAL); 2805 2806 somaxconn = val; 2807 return (0); 2808 } 2809 2810 /* 2811 * These functions are used by protocols to notify the socket layer (and its 2812 * consumers) of state changes in the sockets driven by protocol-side events. 2813 */ 2814 2815 /* 2816 * Procedures to manipulate state flags of socket and do appropriate wakeups. 2817 * 2818 * Normal sequence from the active (originating) side is that 2819 * soisconnecting() is called during processing of connect() call, resulting 2820 * in an eventual call to soisconnected() if/when the connection is 2821 * established. When the connection is torn down soisdisconnecting() is 2822 * called during processing of disconnect() call, and soisdisconnected() is 2823 * called when the connection to the peer is totally severed. The semantics 2824 * of these routines are such that connectionless protocols can call 2825 * soisconnected() and soisdisconnected() only, bypassing the in-progress 2826 * calls when setting up a ``connection'' takes no time. 2827 * 2828 * From the passive side, a socket is created with two queues of sockets: 2829 * so_incomp for connections in progress and so_comp for connections already 2830 * made and awaiting user acceptance. As a protocol is preparing incoming 2831 * connections, it creates a socket structure queued on so_incomp by calling 2832 * sonewconn(). When the connection is established, soisconnected() is 2833 * called, and transfers the socket structure to so_comp, making it available 2834 * to accept(). 2835 * 2836 * If a socket is closed with sockets on either so_incomp or so_comp, these 2837 * sockets are dropped. 2838 * 2839 * If higher-level protocols are implemented in the kernel, the wakeups done 2840 * here will sometimes cause software-interrupt process scheduling. 2841 */ 2842 void 2843 soisconnecting(struct socket *so) 2844 { 2845 2846 SOCK_LOCK(so); 2847 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 2848 so->so_state |= SS_ISCONNECTING; 2849 SOCK_UNLOCK(so); 2850 } 2851 2852 void 2853 soisconnected(struct socket *so) 2854 { 2855 struct socket *head; 2856 2857 ACCEPT_LOCK(); 2858 SOCK_LOCK(so); 2859 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 2860 so->so_state |= SS_ISCONNECTED; 2861 head = so->so_head; 2862 if (head != NULL && (so->so_qstate & SQ_INCOMP)) { 2863 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 2864 SOCK_UNLOCK(so); 2865 TAILQ_REMOVE(&head->so_incomp, so, so_list); 2866 head->so_incqlen--; 2867 so->so_qstate &= ~SQ_INCOMP; 2868 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 2869 head->so_qlen++; 2870 so->so_qstate |= SQ_COMP; 2871 ACCEPT_UNLOCK(); 2872 sorwakeup(head); 2873 wakeup_one(&head->so_timeo); 2874 } else { 2875 ACCEPT_UNLOCK(); 2876 so->so_upcall = 2877 head->so_accf->so_accept_filter->accf_callback; 2878 so->so_upcallarg = head->so_accf->so_accept_filter_arg; 2879 so->so_rcv.sb_flags |= SB_UPCALL; 2880 so->so_options &= ~SO_ACCEPTFILTER; 2881 SOCK_UNLOCK(so); 2882 so->so_upcall(so, so->so_upcallarg, M_DONTWAIT); 2883 } 2884 return; 2885 } 2886 SOCK_UNLOCK(so); 2887 ACCEPT_UNLOCK(); 2888 wakeup(&so->so_timeo); 2889 sorwakeup(so); 2890 sowwakeup(so); 2891 } 2892 2893 void 2894 soisdisconnecting(struct socket *so) 2895 { 2896 2897 /* 2898 * Note: This code assumes that SOCK_LOCK(so) and 2899 * SOCKBUF_LOCK(&so->so_rcv) are the same. 2900 */ 2901 SOCKBUF_LOCK(&so->so_rcv); 2902 so->so_state &= ~SS_ISCONNECTING; 2903 so->so_state |= SS_ISDISCONNECTING; 2904 so->so_rcv.sb_state |= SBS_CANTRCVMORE; 2905 sorwakeup_locked(so); 2906 SOCKBUF_LOCK(&so->so_snd); 2907 so->so_snd.sb_state |= SBS_CANTSENDMORE; 2908 sowwakeup_locked(so); 2909 wakeup(&so->so_timeo); 2910 } 2911 2912 void 2913 soisdisconnected(struct socket *so) 2914 { 2915 2916 /* 2917 * Note: This code assumes that SOCK_LOCK(so) and 2918 * SOCKBUF_LOCK(&so->so_rcv) are the same. 2919 */ 2920 SOCKBUF_LOCK(&so->so_rcv); 2921 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 2922 so->so_state |= SS_ISDISCONNECTED; 2923 so->so_rcv.sb_state |= SBS_CANTRCVMORE; 2924 sorwakeup_locked(so); 2925 SOCKBUF_LOCK(&so->so_snd); 2926 so->so_snd.sb_state |= SBS_CANTSENDMORE; 2927 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc); 2928 sowwakeup_locked(so); 2929 wakeup(&so->so_timeo); 2930 } 2931 2932 /* 2933 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 2934 */ 2935 struct sockaddr * 2936 sodupsockaddr(const struct sockaddr *sa, int mflags) 2937 { 2938 struct sockaddr *sa2; 2939 2940 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 2941 if (sa2) 2942 bcopy(sa, sa2, sa->sa_len); 2943 return sa2; 2944 } 2945 2946 /* 2947 * Create an external-format (``xsocket'') structure using the information in 2948 * the kernel-format socket structure pointed to by so. This is done to 2949 * reduce the spew of irrelevant information over this interface, to isolate 2950 * user code from changes in the kernel structure, and potentially to provide 2951 * information-hiding if we decide that some of this information should be 2952 * hidden from users. 2953 */ 2954 void 2955 sotoxsocket(struct socket *so, struct xsocket *xso) 2956 { 2957 2958 xso->xso_len = sizeof *xso; 2959 xso->xso_so = so; 2960 xso->so_type = so->so_type; 2961 xso->so_options = so->so_options; 2962 xso->so_linger = so->so_linger; 2963 xso->so_state = so->so_state; 2964 xso->so_pcb = so->so_pcb; 2965 xso->xso_protocol = so->so_proto->pr_protocol; 2966 xso->xso_family = so->so_proto->pr_domain->dom_family; 2967 xso->so_qlen = so->so_qlen; 2968 xso->so_incqlen = so->so_incqlen; 2969 xso->so_qlimit = so->so_qlimit; 2970 xso->so_timeo = so->so_timeo; 2971 xso->so_error = so->so_error; 2972 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 2973 xso->so_oobmark = so->so_oobmark; 2974 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 2975 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 2976 xso->so_uid = so->so_cred->cr_uid; 2977 } 2978