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) 2004-2008 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 * NOTE: With regard to VNETs the general rule is that callers do not set 97 * curvnet. Exceptions to this rule include soabort(), sodisconnect(), 98 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn() 99 * and sorflush(), which are usually called from a pre-set VNET context. 100 * sopoll() currently does not need a VNET context to be set. 101 */ 102 103 #include <sys/cdefs.h> 104 __FBSDID("$FreeBSD$"); 105 106 #include "opt_inet.h" 107 #include "opt_inet6.h" 108 #include "opt_zero.h" 109 #include "opt_compat.h" 110 111 #include <sys/param.h> 112 #include <sys/systm.h> 113 #include <sys/fcntl.h> 114 #include <sys/limits.h> 115 #include <sys/lock.h> 116 #include <sys/mac.h> 117 #include <sys/malloc.h> 118 #include <sys/mbuf.h> 119 #include <sys/mutex.h> 120 #include <sys/domain.h> 121 #include <sys/file.h> /* for struct knote */ 122 #include <sys/kernel.h> 123 #include <sys/event.h> 124 #include <sys/eventhandler.h> 125 #include <sys/poll.h> 126 #include <sys/proc.h> 127 #include <sys/protosw.h> 128 #include <sys/socket.h> 129 #include <sys/socketvar.h> 130 #include <sys/resourcevar.h> 131 #include <net/route.h> 132 #include <sys/signalvar.h> 133 #include <sys/stat.h> 134 #include <sys/sx.h> 135 #include <sys/sysctl.h> 136 #include <sys/uio.h> 137 #include <sys/jail.h> 138 #include <sys/syslog.h> 139 #include <netinet/in.h> 140 141 #include <net/vnet.h> 142 143 #include <security/mac/mac_framework.h> 144 145 #include <vm/uma.h> 146 147 #ifdef COMPAT_FREEBSD32 148 #include <sys/mount.h> 149 #include <sys/sysent.h> 150 #include <compat/freebsd32/freebsd32.h> 151 #endif 152 153 static int soreceive_rcvoob(struct socket *so, struct uio *uio, 154 int flags); 155 156 static void filt_sordetach(struct knote *kn); 157 static int filt_soread(struct knote *kn, long hint); 158 static void filt_sowdetach(struct knote *kn); 159 static int filt_sowrite(struct knote *kn, long hint); 160 static int filt_solisten(struct knote *kn, long hint); 161 162 static struct filterops solisten_filtops = { 163 .f_isfd = 1, 164 .f_detach = filt_sordetach, 165 .f_event = filt_solisten, 166 }; 167 static struct filterops soread_filtops = { 168 .f_isfd = 1, 169 .f_detach = filt_sordetach, 170 .f_event = filt_soread, 171 }; 172 static struct filterops sowrite_filtops = { 173 .f_isfd = 1, 174 .f_detach = filt_sowdetach, 175 .f_event = filt_sowrite, 176 }; 177 178 so_gen_t so_gencnt; /* generation count for sockets */ 179 180 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 181 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 182 183 #define VNET_SO_ASSERT(so) \ 184 VNET_ASSERT(curvnet != NULL, \ 185 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so))); 186 187 /* 188 * Limit on the number of connections in the listen queue waiting 189 * for accept(2). 190 * NB: The orginal sysctl somaxconn is still available but hidden 191 * to prevent confusion about the actual purpose of this number. 192 */ 193 static int somaxconn = SOMAXCONN; 194 195 static int 196 sysctl_somaxconn(SYSCTL_HANDLER_ARGS) 197 { 198 int error; 199 int val; 200 201 val = somaxconn; 202 error = sysctl_handle_int(oidp, &val, 0, req); 203 if (error || !req->newptr ) 204 return (error); 205 206 if (val < 1 || val > USHRT_MAX) 207 return (EINVAL); 208 209 somaxconn = val; 210 return (0); 211 } 212 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW, 213 0, sizeof(int), sysctl_somaxconn, "I", 214 "Maximum listen socket pending connection accept queue size"); 215 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, 216 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP, 217 0, sizeof(int), sysctl_somaxconn, "I", 218 "Maximum listen socket pending connection accept queue size (compat)"); 219 220 static int numopensockets; 221 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD, 222 &numopensockets, 0, "Number of open sockets"); 223 224 #if defined(SOCKET_SEND_COW) || defined(SOCKET_RECV_PFLIP) 225 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0, 226 "Zero copy controls"); 227 #ifdef SOCKET_RECV_PFLIP 228 int so_zero_copy_receive = 1; 229 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW, 230 &so_zero_copy_receive, 0, "Enable zero copy receive"); 231 #endif 232 #ifdef SOCKET_SEND_COW 233 int so_zero_copy_send = 1; 234 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW, 235 &so_zero_copy_send, 0, "Enable zero copy send"); 236 #endif /* SOCKET_SEND_COW */ 237 #endif /* SOCKET_SEND_COW || SOCKET_RECV_PFLIP */ 238 239 /* 240 * accept_mtx locks down per-socket fields relating to accept queues. See 241 * socketvar.h for an annotation of the protected fields of struct socket. 242 */ 243 struct mtx accept_mtx; 244 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF); 245 246 /* 247 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket 248 * so_gencnt field. 249 */ 250 static struct mtx so_global_mtx; 251 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF); 252 253 /* 254 * General IPC sysctl name space, used by sockets and a variety of other IPC 255 * types. 256 */ 257 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); 258 259 /* 260 * Initialize the socket subsystem and set up the socket 261 * memory allocator. 262 */ 263 static uma_zone_t socket_zone; 264 int maxsockets; 265 266 static void 267 socket_zone_change(void *tag) 268 { 269 270 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 271 } 272 273 static void 274 socket_init(void *tag) 275 { 276 277 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL, 278 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 279 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 280 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached"); 281 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL, 282 EVENTHANDLER_PRI_FIRST); 283 } 284 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL); 285 286 /* 287 * Initialise maxsockets. This SYSINIT must be run after 288 * tunable_mbinit(). 289 */ 290 static void 291 init_maxsockets(void *ignored) 292 { 293 294 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 295 maxsockets = imax(maxsockets, maxfiles); 296 } 297 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); 298 299 /* 300 * Sysctl to get and set the maximum global sockets limit. Notify protocols 301 * of the change so that they can update their dependent limits as required. 302 */ 303 static int 304 sysctl_maxsockets(SYSCTL_HANDLER_ARGS) 305 { 306 int error, newmaxsockets; 307 308 newmaxsockets = maxsockets; 309 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req); 310 if (error == 0 && req->newptr) { 311 if (newmaxsockets > maxsockets && 312 newmaxsockets <= maxfiles) { 313 maxsockets = newmaxsockets; 314 EVENTHANDLER_INVOKE(maxsockets_change); 315 } else 316 error = EINVAL; 317 } 318 return (error); 319 } 320 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW, 321 &maxsockets, 0, sysctl_maxsockets, "IU", 322 "Maximum number of sockets avaliable"); 323 324 /* 325 * Socket operation routines. These routines are called by the routines in 326 * sys_socket.c or from a system process, and implement the semantics of 327 * socket operations by switching out to the protocol specific routines. 328 */ 329 330 /* 331 * Get a socket structure from our zone, and initialize it. Note that it 332 * would probably be better to allocate socket and PCB at the same time, but 333 * I'm not convinced that all the protocols can be easily modified to do 334 * this. 335 * 336 * soalloc() returns a socket with a ref count of 0. 337 */ 338 static struct socket * 339 soalloc(struct vnet *vnet) 340 { 341 struct socket *so; 342 343 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO); 344 if (so == NULL) 345 return (NULL); 346 #ifdef MAC 347 if (mac_socket_init(so, M_NOWAIT) != 0) { 348 uma_zfree(socket_zone, so); 349 return (NULL); 350 } 351 #endif 352 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd"); 353 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv"); 354 sx_init(&so->so_snd.sb_sx, "so_snd_sx"); 355 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx"); 356 TAILQ_INIT(&so->so_aiojobq); 357 mtx_lock(&so_global_mtx); 358 so->so_gencnt = ++so_gencnt; 359 ++numopensockets; 360 #ifdef VIMAGE 361 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p", 362 __func__, __LINE__, so)); 363 vnet->vnet_sockcnt++; 364 so->so_vnet = vnet; 365 #endif 366 mtx_unlock(&so_global_mtx); 367 return (so); 368 } 369 370 /* 371 * Free the storage associated with a socket at the socket layer, tear down 372 * locks, labels, etc. All protocol state is assumed already to have been 373 * torn down (and possibly never set up) by the caller. 374 */ 375 static void 376 sodealloc(struct socket *so) 377 { 378 379 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); 380 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL")); 381 382 mtx_lock(&so_global_mtx); 383 so->so_gencnt = ++so_gencnt; 384 --numopensockets; /* Could be below, but faster here. */ 385 #ifdef VIMAGE 386 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p", 387 __func__, __LINE__, so)); 388 so->so_vnet->vnet_sockcnt--; 389 #endif 390 mtx_unlock(&so_global_mtx); 391 if (so->so_rcv.sb_hiwat) 392 (void)chgsbsize(so->so_cred->cr_uidinfo, 393 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 394 if (so->so_snd.sb_hiwat) 395 (void)chgsbsize(so->so_cred->cr_uidinfo, 396 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 397 #ifdef INET 398 /* remove acccept filter if one is present. */ 399 if (so->so_accf != NULL) 400 do_setopt_accept_filter(so, NULL); 401 #endif 402 #ifdef MAC 403 mac_socket_destroy(so); 404 #endif 405 crfree(so->so_cred); 406 sx_destroy(&so->so_snd.sb_sx); 407 sx_destroy(&so->so_rcv.sb_sx); 408 SOCKBUF_LOCK_DESTROY(&so->so_snd); 409 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 410 uma_zfree(socket_zone, so); 411 } 412 413 /* 414 * socreate returns a socket with a ref count of 1. The socket should be 415 * closed with soclose(). 416 */ 417 int 418 socreate(int dom, struct socket **aso, int type, int proto, 419 struct ucred *cred, struct thread *td) 420 { 421 struct protosw *prp; 422 struct socket *so; 423 int error; 424 425 if (proto) 426 prp = pffindproto(dom, proto, type); 427 else 428 prp = pffindtype(dom, type); 429 430 if (prp == NULL) { 431 /* No support for domain. */ 432 if (pffinddomain(dom) == NULL) 433 return (EAFNOSUPPORT); 434 /* No support for socket type. */ 435 if (proto == 0 && type != 0) 436 return (EPROTOTYPE); 437 return (EPROTONOSUPPORT); 438 } 439 if (prp->pr_usrreqs->pru_attach == NULL || 440 prp->pr_usrreqs->pru_attach == pru_attach_notsupp) 441 return (EPROTONOSUPPORT); 442 443 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0) 444 return (EPROTONOSUPPORT); 445 446 if (prp->pr_type != type) 447 return (EPROTOTYPE); 448 so = soalloc(CRED_TO_VNET(cred)); 449 if (so == NULL) 450 return (ENOBUFS); 451 452 TAILQ_INIT(&so->so_incomp); 453 TAILQ_INIT(&so->so_comp); 454 so->so_type = type; 455 so->so_cred = crhold(cred); 456 if ((prp->pr_domain->dom_family == PF_INET) || 457 (prp->pr_domain->dom_family == PF_INET6) || 458 (prp->pr_domain->dom_family == PF_ROUTE)) 459 so->so_fibnum = td->td_proc->p_fibnum; 460 else 461 so->so_fibnum = 0; 462 so->so_proto = prp; 463 #ifdef MAC 464 mac_socket_create(cred, so); 465 #endif 466 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv)); 467 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd)); 468 so->so_count = 1; 469 /* 470 * Auto-sizing of socket buffers is managed by the protocols and 471 * the appropriate flags must be set in the pru_attach function. 472 */ 473 CURVNET_SET(so->so_vnet); 474 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); 475 CURVNET_RESTORE(); 476 if (error) { 477 KASSERT(so->so_count == 1, ("socreate: so_count %d", 478 so->so_count)); 479 so->so_count = 0; 480 sodealloc(so); 481 return (error); 482 } 483 *aso = so; 484 return (0); 485 } 486 487 #ifdef REGRESSION 488 static int regression_sonewconn_earlytest = 1; 489 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW, 490 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test"); 491 #endif 492 493 /* 494 * When an attempt at a new connection is noted on a socket which accepts 495 * connections, sonewconn is called. If the connection is possible (subject 496 * to space constraints, etc.) then we allocate a new structure, propoerly 497 * linked into the data structure of the original socket, and return this. 498 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. 499 * 500 * Note: the ref count on the socket is 0 on return. 501 */ 502 struct socket * 503 sonewconn(struct socket *head, int connstatus) 504 { 505 struct socket *so; 506 int over; 507 508 ACCEPT_LOCK(); 509 over = (head->so_qlen > 3 * head->so_qlimit / 2); 510 ACCEPT_UNLOCK(); 511 #ifdef REGRESSION 512 if (regression_sonewconn_earlytest && over) { 513 #else 514 if (over) { 515 #endif 516 log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: " 517 "%i already in queue awaiting acceptance\n", 518 __func__, head->so_pcb, over); 519 return (NULL); 520 } 521 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p", 522 __func__, __LINE__, head)); 523 so = soalloc(head->so_vnet); 524 if (so == NULL) { 525 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 526 "limit reached or out of memory\n", 527 __func__, head->so_pcb); 528 return (NULL); 529 } 530 if ((head->so_options & SO_ACCEPTFILTER) != 0) 531 connstatus = 0; 532 so->so_head = head; 533 so->so_type = head->so_type; 534 so->so_options = head->so_options &~ SO_ACCEPTCONN; 535 so->so_linger = head->so_linger; 536 so->so_state = head->so_state | SS_NOFDREF; 537 so->so_fibnum = head->so_fibnum; 538 so->so_proto = head->so_proto; 539 so->so_cred = crhold(head->so_cred); 540 #ifdef MAC 541 mac_socket_newconn(head, so); 542 #endif 543 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv)); 544 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd)); 545 VNET_SO_ASSERT(head); 546 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { 547 sodealloc(so); 548 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 549 __func__, head->so_pcb); 550 return (NULL); 551 } 552 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 553 sodealloc(so); 554 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 555 __func__, head->so_pcb); 556 return (NULL); 557 } 558 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 559 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 560 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 561 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 562 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; 563 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; 564 so->so_state |= connstatus; 565 ACCEPT_LOCK(); 566 /* 567 * The accept socket may be tearing down but we just 568 * won a race on the ACCEPT_LOCK. 569 * However, if sctp_peeloff() is called on a 1-to-many 570 * style socket, the SO_ACCEPTCONN doesn't need to be set. 571 */ 572 if (!(head->so_options & SO_ACCEPTCONN) && 573 ((head->so_proto->pr_protocol != IPPROTO_SCTP) || 574 (head->so_type != SOCK_SEQPACKET))) { 575 SOCK_LOCK(so); 576 so->so_head = NULL; 577 sofree(so); /* NB: returns ACCEPT_UNLOCK'ed. */ 578 return (NULL); 579 } 580 if (connstatus) { 581 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 582 so->so_qstate |= SQ_COMP; 583 head->so_qlen++; 584 } else { 585 /* 586 * Keep removing sockets from the head until there's room for 587 * us to insert on the tail. In pre-locking revisions, this 588 * was a simple if(), but as we could be racing with other 589 * threads and soabort() requires dropping locks, we must 590 * loop waiting for the condition to be true. 591 */ 592 while (head->so_incqlen > head->so_qlimit) { 593 struct socket *sp; 594 sp = TAILQ_FIRST(&head->so_incomp); 595 TAILQ_REMOVE(&head->so_incomp, sp, so_list); 596 head->so_incqlen--; 597 sp->so_qstate &= ~SQ_INCOMP; 598 sp->so_head = NULL; 599 ACCEPT_UNLOCK(); 600 soabort(sp); 601 ACCEPT_LOCK(); 602 } 603 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); 604 so->so_qstate |= SQ_INCOMP; 605 head->so_incqlen++; 606 } 607 ACCEPT_UNLOCK(); 608 if (connstatus) { 609 sorwakeup(head); 610 wakeup_one(&head->so_timeo); 611 } 612 return (so); 613 } 614 615 int 616 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 617 { 618 int error; 619 620 CURVNET_SET(so->so_vnet); 621 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td); 622 CURVNET_RESTORE(); 623 return (error); 624 } 625 626 int 627 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 628 { 629 int error; 630 631 CURVNET_SET(so->so_vnet); 632 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td); 633 CURVNET_RESTORE(); 634 return (error); 635 } 636 637 /* 638 * solisten() transitions a socket from a non-listening state to a listening 639 * state, but can also be used to update the listen queue depth on an 640 * existing listen socket. The protocol will call back into the sockets 641 * layer using solisten_proto_check() and solisten_proto() to check and set 642 * socket-layer listen state. Call backs are used so that the protocol can 643 * acquire both protocol and socket layer locks in whatever order is required 644 * by the protocol. 645 * 646 * Protocol implementors are advised to hold the socket lock across the 647 * socket-layer test and set to avoid races at the socket layer. 648 */ 649 int 650 solisten(struct socket *so, int backlog, struct thread *td) 651 { 652 int error; 653 654 CURVNET_SET(so->so_vnet); 655 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td); 656 CURVNET_RESTORE(); 657 return (error); 658 } 659 660 int 661 solisten_proto_check(struct socket *so) 662 { 663 664 SOCK_LOCK_ASSERT(so); 665 666 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 667 SS_ISDISCONNECTING)) 668 return (EINVAL); 669 return (0); 670 } 671 672 void 673 solisten_proto(struct socket *so, int backlog) 674 { 675 676 SOCK_LOCK_ASSERT(so); 677 678 if (backlog < 0 || backlog > somaxconn) 679 backlog = somaxconn; 680 so->so_qlimit = backlog; 681 so->so_options |= SO_ACCEPTCONN; 682 } 683 684 /* 685 * Evaluate the reference count and named references on a socket; if no 686 * references remain, free it. This should be called whenever a reference is 687 * released, such as in sorele(), but also when named reference flags are 688 * cleared in socket or protocol code. 689 * 690 * sofree() will free the socket if: 691 * 692 * - There are no outstanding file descriptor references or related consumers 693 * (so_count == 0). 694 * 695 * - The socket has been closed by user space, if ever open (SS_NOFDREF). 696 * 697 * - The protocol does not have an outstanding strong reference on the socket 698 * (SS_PROTOREF). 699 * 700 * - The socket is not in a completed connection queue, so a process has been 701 * notified that it is present. If it is removed, the user process may 702 * block in accept() despite select() saying the socket was ready. 703 */ 704 void 705 sofree(struct socket *so) 706 { 707 struct protosw *pr = so->so_proto; 708 struct socket *head; 709 710 ACCEPT_LOCK_ASSERT(); 711 SOCK_LOCK_ASSERT(so); 712 713 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 || 714 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) { 715 SOCK_UNLOCK(so); 716 ACCEPT_UNLOCK(); 717 return; 718 } 719 720 head = so->so_head; 721 if (head != NULL) { 722 KASSERT((so->so_qstate & SQ_COMP) != 0 || 723 (so->so_qstate & SQ_INCOMP) != 0, 724 ("sofree: so_head != NULL, but neither SQ_COMP nor " 725 "SQ_INCOMP")); 726 KASSERT((so->so_qstate & SQ_COMP) == 0 || 727 (so->so_qstate & SQ_INCOMP) == 0, 728 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP")); 729 TAILQ_REMOVE(&head->so_incomp, so, so_list); 730 head->so_incqlen--; 731 so->so_qstate &= ~SQ_INCOMP; 732 so->so_head = NULL; 733 } 734 KASSERT((so->so_qstate & SQ_COMP) == 0 && 735 (so->so_qstate & SQ_INCOMP) == 0, 736 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)", 737 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP)); 738 if (so->so_options & SO_ACCEPTCONN) { 739 KASSERT((TAILQ_EMPTY(&so->so_comp)), 740 ("sofree: so_comp populated")); 741 KASSERT((TAILQ_EMPTY(&so->so_incomp)), 742 ("sofree: so_incomp populated")); 743 } 744 SOCK_UNLOCK(so); 745 ACCEPT_UNLOCK(); 746 747 VNET_SO_ASSERT(so); 748 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 749 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb); 750 if (pr->pr_usrreqs->pru_detach != NULL) 751 (*pr->pr_usrreqs->pru_detach)(so); 752 753 /* 754 * From this point on, we assume that no other references to this 755 * socket exist anywhere else in the stack. Therefore, no locks need 756 * to be acquired or held. 757 * 758 * We used to do a lot of socket buffer and socket locking here, as 759 * well as invoke sorflush() and perform wakeups. The direct call to 760 * dom_dispose() and sbrelease_internal() are an inlining of what was 761 * necessary from sorflush(). 762 * 763 * Notice that the socket buffer and kqueue state are torn down 764 * before calling pru_detach. This means that protocols shold not 765 * assume they can perform socket wakeups, etc, in their detach code. 766 */ 767 sbdestroy(&so->so_snd, so); 768 sbdestroy(&so->so_rcv, so); 769 seldrain(&so->so_snd.sb_sel); 770 seldrain(&so->so_rcv.sb_sel); 771 knlist_destroy(&so->so_rcv.sb_sel.si_note); 772 knlist_destroy(&so->so_snd.sb_sel.si_note); 773 sodealloc(so); 774 } 775 776 /* 777 * Close a socket on last file table reference removal. Initiate disconnect 778 * if connected. Free socket when disconnect complete. 779 * 780 * This function will sorele() the socket. Note that soclose() may be called 781 * prior to the ref count reaching zero. The actual socket structure will 782 * not be freed until the ref count reaches zero. 783 */ 784 int 785 soclose(struct socket *so) 786 { 787 int error = 0; 788 789 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter")); 790 791 CURVNET_SET(so->so_vnet); 792 funsetown(&so->so_sigio); 793 if (so->so_state & SS_ISCONNECTED) { 794 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 795 error = sodisconnect(so); 796 if (error) { 797 if (error == ENOTCONN) 798 error = 0; 799 goto drop; 800 } 801 } 802 if (so->so_options & SO_LINGER) { 803 if ((so->so_state & SS_ISDISCONNECTING) && 804 (so->so_state & SS_NBIO)) 805 goto drop; 806 while (so->so_state & SS_ISCONNECTED) { 807 error = tsleep(&so->so_timeo, 808 PSOCK | PCATCH, "soclos", 809 so->so_linger * hz); 810 if (error) 811 break; 812 } 813 } 814 } 815 816 drop: 817 if (so->so_proto->pr_usrreqs->pru_close != NULL) 818 (*so->so_proto->pr_usrreqs->pru_close)(so); 819 ACCEPT_LOCK(); 820 if (so->so_options & SO_ACCEPTCONN) { 821 struct socket *sp; 822 /* 823 * Prevent new additions to the accept queues due 824 * to ACCEPT_LOCK races while we are draining them. 825 */ 826 so->so_options &= ~SO_ACCEPTCONN; 827 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) { 828 TAILQ_REMOVE(&so->so_incomp, sp, so_list); 829 so->so_incqlen--; 830 sp->so_qstate &= ~SQ_INCOMP; 831 sp->so_head = NULL; 832 ACCEPT_UNLOCK(); 833 soabort(sp); 834 ACCEPT_LOCK(); 835 } 836 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) { 837 TAILQ_REMOVE(&so->so_comp, sp, so_list); 838 so->so_qlen--; 839 sp->so_qstate &= ~SQ_COMP; 840 sp->so_head = NULL; 841 ACCEPT_UNLOCK(); 842 soabort(sp); 843 ACCEPT_LOCK(); 844 } 845 KASSERT((TAILQ_EMPTY(&so->so_comp)), 846 ("%s: so_comp populated", __func__)); 847 KASSERT((TAILQ_EMPTY(&so->so_incomp)), 848 ("%s: so_incomp populated", __func__)); 849 } 850 SOCK_LOCK(so); 851 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF")); 852 so->so_state |= SS_NOFDREF; 853 sorele(so); /* NB: Returns with ACCEPT_UNLOCK(). */ 854 CURVNET_RESTORE(); 855 return (error); 856 } 857 858 /* 859 * soabort() is used to abruptly tear down a connection, such as when a 860 * resource limit is reached (listen queue depth exceeded), or if a listen 861 * socket is closed while there are sockets waiting to be accepted. 862 * 863 * This interface is tricky, because it is called on an unreferenced socket, 864 * and must be called only by a thread that has actually removed the socket 865 * from the listen queue it was on, or races with other threads are risked. 866 * 867 * This interface will call into the protocol code, so must not be called 868 * with any socket locks held. Protocols do call it while holding their own 869 * recursible protocol mutexes, but this is something that should be subject 870 * to review in the future. 871 */ 872 void 873 soabort(struct socket *so) 874 { 875 876 /* 877 * In as much as is possible, assert that no references to this 878 * socket are held. This is not quite the same as asserting that the 879 * current thread is responsible for arranging for no references, but 880 * is as close as we can get for now. 881 */ 882 KASSERT(so->so_count == 0, ("soabort: so_count")); 883 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF")); 884 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF")); 885 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP")); 886 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP")); 887 VNET_SO_ASSERT(so); 888 889 if (so->so_proto->pr_usrreqs->pru_abort != NULL) 890 (*so->so_proto->pr_usrreqs->pru_abort)(so); 891 ACCEPT_LOCK(); 892 SOCK_LOCK(so); 893 sofree(so); 894 } 895 896 int 897 soaccept(struct socket *so, struct sockaddr **nam) 898 { 899 int error; 900 901 SOCK_LOCK(so); 902 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF")); 903 so->so_state &= ~SS_NOFDREF; 904 SOCK_UNLOCK(so); 905 906 CURVNET_SET(so->so_vnet); 907 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); 908 CURVNET_RESTORE(); 909 return (error); 910 } 911 912 int 913 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 914 { 915 916 return (soconnectat(AT_FDCWD, so, nam, td)); 917 } 918 919 int 920 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 921 { 922 int error; 923 924 if (so->so_options & SO_ACCEPTCONN) 925 return (EOPNOTSUPP); 926 927 CURVNET_SET(so->so_vnet); 928 /* 929 * If protocol is connection-based, can only connect once. 930 * Otherwise, if connected, try to disconnect first. This allows 931 * user to disconnect by connecting to, e.g., a null address. 932 */ 933 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 934 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 935 (error = sodisconnect(so)))) { 936 error = EISCONN; 937 } else { 938 /* 939 * Prevent accumulated error from previous connection from 940 * biting us. 941 */ 942 so->so_error = 0; 943 if (fd == AT_FDCWD) { 944 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, 945 nam, td); 946 } else { 947 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd, 948 so, nam, td); 949 } 950 } 951 CURVNET_RESTORE(); 952 953 return (error); 954 } 955 956 int 957 soconnect2(struct socket *so1, struct socket *so2) 958 { 959 int error; 960 961 CURVNET_SET(so1->so_vnet); 962 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); 963 CURVNET_RESTORE(); 964 return (error); 965 } 966 967 int 968 sodisconnect(struct socket *so) 969 { 970 int error; 971 972 if ((so->so_state & SS_ISCONNECTED) == 0) 973 return (ENOTCONN); 974 if (so->so_state & SS_ISDISCONNECTING) 975 return (EALREADY); 976 VNET_SO_ASSERT(so); 977 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 978 return (error); 979 } 980 981 #ifdef SOCKET_SEND_COW 982 struct so_zerocopy_stats{ 983 int size_ok; 984 int align_ok; 985 int found_ifp; 986 }; 987 struct so_zerocopy_stats so_zerocp_stats = {0,0,0}; 988 989 /* 990 * sosend_copyin() is only used if zero copy sockets are enabled. Otherwise 991 * sosend_dgram() and sosend_generic() use m_uiotombuf(). 992 * 993 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or 994 * all of the data referenced by the uio. If desired, it uses zero-copy. 995 * *space will be updated to reflect data copied in. 996 * 997 * NB: If atomic I/O is requested, the caller must already have checked that 998 * space can hold resid bytes. 999 * 1000 * NB: In the event of an error, the caller may need to free the partial 1001 * chain pointed to by *mpp. The contents of both *uio and *space may be 1002 * modified even in the case of an error. 1003 */ 1004 static int 1005 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space, 1006 int flags) 1007 { 1008 struct mbuf *m, **mp, *top; 1009 long len; 1010 ssize_t resid; 1011 int error; 1012 int cow_send; 1013 1014 *retmp = top = NULL; 1015 mp = ⊤ 1016 len = 0; 1017 resid = uio->uio_resid; 1018 error = 0; 1019 do { 1020 cow_send = 0; 1021 if (resid >= MINCLSIZE) { 1022 if (top == NULL) { 1023 m = m_gethdr(M_WAITOK, MT_DATA); 1024 m->m_pkthdr.len = 0; 1025 m->m_pkthdr.rcvif = NULL; 1026 } else 1027 m = m_get(M_WAITOK, MT_DATA); 1028 if (so_zero_copy_send && 1029 resid >= PAGE_SIZE && 1030 *space >= PAGE_SIZE && 1031 uio->uio_iov->iov_len >= PAGE_SIZE) { 1032 so_zerocp_stats.size_ok++; 1033 so_zerocp_stats.align_ok++; 1034 cow_send = socow_setup(m, uio); 1035 len = cow_send; 1036 } 1037 if (!cow_send) { 1038 m_clget(m, M_WAITOK); 1039 len = min(min(MCLBYTES, resid), *space); 1040 } 1041 } else { 1042 if (top == NULL) { 1043 m = m_gethdr(M_WAITOK, MT_DATA); 1044 m->m_pkthdr.len = 0; 1045 m->m_pkthdr.rcvif = NULL; 1046 1047 len = min(min(MHLEN, resid), *space); 1048 /* 1049 * For datagram protocols, leave room 1050 * for protocol headers in first mbuf. 1051 */ 1052 if (atomic && m && len < MHLEN) 1053 MH_ALIGN(m, len); 1054 } else { 1055 m = m_get(M_WAITOK, MT_DATA); 1056 len = min(min(MLEN, resid), *space); 1057 } 1058 } 1059 if (m == NULL) { 1060 error = ENOBUFS; 1061 goto out; 1062 } 1063 1064 *space -= len; 1065 if (cow_send) 1066 error = 0; 1067 else 1068 error = uiomove(mtod(m, void *), (int)len, uio); 1069 resid = uio->uio_resid; 1070 m->m_len = len; 1071 *mp = m; 1072 top->m_pkthdr.len += len; 1073 if (error) 1074 goto out; 1075 mp = &m->m_next; 1076 if (resid <= 0) { 1077 if (flags & MSG_EOR) 1078 top->m_flags |= M_EOR; 1079 break; 1080 } 1081 } while (*space > 0 && atomic); 1082 out: 1083 *retmp = top; 1084 return (error); 1085 } 1086 #endif /* SOCKET_SEND_COW */ 1087 1088 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) 1089 1090 int 1091 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1092 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1093 { 1094 long space; 1095 ssize_t resid; 1096 int clen = 0, error, dontroute; 1097 #ifdef SOCKET_SEND_COW 1098 int atomic = sosendallatonce(so) || top; 1099 #endif 1100 1101 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 1102 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 1103 ("sosend_dgram: !PR_ATOMIC")); 1104 1105 if (uio != NULL) 1106 resid = uio->uio_resid; 1107 else 1108 resid = top->m_pkthdr.len; 1109 /* 1110 * In theory resid should be unsigned. However, space must be 1111 * signed, as it might be less than 0 if we over-committed, and we 1112 * must use a signed comparison of space and resid. On the other 1113 * hand, a negative resid causes us to loop sending 0-length 1114 * segments to the protocol. 1115 */ 1116 if (resid < 0) { 1117 error = EINVAL; 1118 goto out; 1119 } 1120 1121 dontroute = 1122 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 1123 if (td != NULL) 1124 td->td_ru.ru_msgsnd++; 1125 if (control != NULL) 1126 clen = control->m_len; 1127 1128 SOCKBUF_LOCK(&so->so_snd); 1129 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1130 SOCKBUF_UNLOCK(&so->so_snd); 1131 error = EPIPE; 1132 goto out; 1133 } 1134 if (so->so_error) { 1135 error = so->so_error; 1136 so->so_error = 0; 1137 SOCKBUF_UNLOCK(&so->so_snd); 1138 goto out; 1139 } 1140 if ((so->so_state & SS_ISCONNECTED) == 0) { 1141 /* 1142 * `sendto' and `sendmsg' is allowed on a connection-based 1143 * socket if it supports implied connect. Return ENOTCONN if 1144 * not connected and no address is supplied. 1145 */ 1146 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1147 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1148 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1149 !(resid == 0 && clen != 0)) { 1150 SOCKBUF_UNLOCK(&so->so_snd); 1151 error = ENOTCONN; 1152 goto out; 1153 } 1154 } else if (addr == NULL) { 1155 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1156 error = ENOTCONN; 1157 else 1158 error = EDESTADDRREQ; 1159 SOCKBUF_UNLOCK(&so->so_snd); 1160 goto out; 1161 } 1162 } 1163 1164 /* 1165 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1166 * problem and need fixing. 1167 */ 1168 space = sbspace(&so->so_snd); 1169 if (flags & MSG_OOB) 1170 space += 1024; 1171 space -= clen; 1172 SOCKBUF_UNLOCK(&so->so_snd); 1173 if (resid > space) { 1174 error = EMSGSIZE; 1175 goto out; 1176 } 1177 if (uio == NULL) { 1178 resid = 0; 1179 if (flags & MSG_EOR) 1180 top->m_flags |= M_EOR; 1181 } else { 1182 #ifdef SOCKET_SEND_COW 1183 error = sosend_copyin(uio, &top, atomic, &space, flags); 1184 if (error) 1185 goto out; 1186 #else 1187 /* 1188 * Copy the data from userland into a mbuf chain. 1189 * If no data is to be copied in, a single empty mbuf 1190 * is returned. 1191 */ 1192 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1193 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1194 if (top == NULL) { 1195 error = EFAULT; /* only possible error */ 1196 goto out; 1197 } 1198 space -= resid - uio->uio_resid; 1199 #endif /* SOCKET_SEND_COW */ 1200 resid = uio->uio_resid; 1201 } 1202 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1203 /* 1204 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1205 * than with. 1206 */ 1207 if (dontroute) { 1208 SOCK_LOCK(so); 1209 so->so_options |= SO_DONTROUTE; 1210 SOCK_UNLOCK(so); 1211 } 1212 /* 1213 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1214 * of date. We could have recieved a reset packet in an interrupt or 1215 * maybe we slept while doing page faults in uiomove() etc. We could 1216 * probably recheck again inside the locking protection here, but 1217 * there are probably other places that this also happens. We must 1218 * rethink this. 1219 */ 1220 VNET_SO_ASSERT(so); 1221 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1222 (flags & MSG_OOB) ? PRUS_OOB : 1223 /* 1224 * If the user set MSG_EOF, the protocol understands this flag and 1225 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1226 */ 1227 ((flags & MSG_EOF) && 1228 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1229 (resid <= 0)) ? 1230 PRUS_EOF : 1231 /* If there is more to send set PRUS_MORETOCOME */ 1232 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1233 top, addr, control, td); 1234 if (dontroute) { 1235 SOCK_LOCK(so); 1236 so->so_options &= ~SO_DONTROUTE; 1237 SOCK_UNLOCK(so); 1238 } 1239 clen = 0; 1240 control = NULL; 1241 top = NULL; 1242 out: 1243 if (top != NULL) 1244 m_freem(top); 1245 if (control != NULL) 1246 m_freem(control); 1247 return (error); 1248 } 1249 1250 /* 1251 * Send on a socket. If send must go all at once and message is larger than 1252 * send buffering, then hard error. Lock against other senders. If must go 1253 * all at once and not enough room now, then inform user that this would 1254 * block and do nothing. Otherwise, if nonblocking, send as much as 1255 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1256 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1257 * in mbuf chain must be small enough to send all at once. 1258 * 1259 * Returns nonzero on error, timeout or signal; callers must check for short 1260 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1261 * on return. 1262 */ 1263 int 1264 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1265 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1266 { 1267 long space; 1268 ssize_t resid; 1269 int clen = 0, error, dontroute; 1270 int atomic = sosendallatonce(so) || top; 1271 1272 if (uio != NULL) 1273 resid = uio->uio_resid; 1274 else 1275 resid = top->m_pkthdr.len; 1276 /* 1277 * In theory resid should be unsigned. However, space must be 1278 * signed, as it might be less than 0 if we over-committed, and we 1279 * must use a signed comparison of space and resid. On the other 1280 * hand, a negative resid causes us to loop sending 0-length 1281 * segments to the protocol. 1282 * 1283 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1284 * type sockets since that's an error. 1285 */ 1286 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1287 error = EINVAL; 1288 goto out; 1289 } 1290 1291 dontroute = 1292 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1293 (so->so_proto->pr_flags & PR_ATOMIC); 1294 if (td != NULL) 1295 td->td_ru.ru_msgsnd++; 1296 if (control != NULL) 1297 clen = control->m_len; 1298 1299 error = sblock(&so->so_snd, SBLOCKWAIT(flags)); 1300 if (error) 1301 goto out; 1302 1303 restart: 1304 do { 1305 SOCKBUF_LOCK(&so->so_snd); 1306 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1307 SOCKBUF_UNLOCK(&so->so_snd); 1308 error = EPIPE; 1309 goto release; 1310 } 1311 if (so->so_error) { 1312 error = so->so_error; 1313 so->so_error = 0; 1314 SOCKBUF_UNLOCK(&so->so_snd); 1315 goto release; 1316 } 1317 if ((so->so_state & SS_ISCONNECTED) == 0) { 1318 /* 1319 * `sendto' and `sendmsg' is allowed on a connection- 1320 * based socket if it supports implied connect. 1321 * Return ENOTCONN if not connected and no address is 1322 * supplied. 1323 */ 1324 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1325 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1326 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1327 !(resid == 0 && clen != 0)) { 1328 SOCKBUF_UNLOCK(&so->so_snd); 1329 error = ENOTCONN; 1330 goto release; 1331 } 1332 } else if (addr == NULL) { 1333 SOCKBUF_UNLOCK(&so->so_snd); 1334 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1335 error = ENOTCONN; 1336 else 1337 error = EDESTADDRREQ; 1338 goto release; 1339 } 1340 } 1341 space = sbspace(&so->so_snd); 1342 if (flags & MSG_OOB) 1343 space += 1024; 1344 if ((atomic && resid > so->so_snd.sb_hiwat) || 1345 clen > so->so_snd.sb_hiwat) { 1346 SOCKBUF_UNLOCK(&so->so_snd); 1347 error = EMSGSIZE; 1348 goto release; 1349 } 1350 if (space < resid + clen && 1351 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1352 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) { 1353 SOCKBUF_UNLOCK(&so->so_snd); 1354 error = EWOULDBLOCK; 1355 goto release; 1356 } 1357 error = sbwait(&so->so_snd); 1358 SOCKBUF_UNLOCK(&so->so_snd); 1359 if (error) 1360 goto release; 1361 goto restart; 1362 } 1363 SOCKBUF_UNLOCK(&so->so_snd); 1364 space -= clen; 1365 do { 1366 if (uio == NULL) { 1367 resid = 0; 1368 if (flags & MSG_EOR) 1369 top->m_flags |= M_EOR; 1370 } else { 1371 #ifdef SOCKET_SEND_COW 1372 error = sosend_copyin(uio, &top, atomic, 1373 &space, flags); 1374 if (error != 0) 1375 goto release; 1376 #else 1377 /* 1378 * Copy the data from userland into a mbuf 1379 * chain. If no data is to be copied in, 1380 * a single empty mbuf is returned. 1381 */ 1382 top = m_uiotombuf(uio, M_WAITOK, space, 1383 (atomic ? max_hdr : 0), 1384 (atomic ? M_PKTHDR : 0) | 1385 ((flags & MSG_EOR) ? M_EOR : 0)); 1386 if (top == NULL) { 1387 error = EFAULT; /* only possible error */ 1388 goto release; 1389 } 1390 space -= resid - uio->uio_resid; 1391 #endif /* SOCKET_SEND_COW */ 1392 resid = uio->uio_resid; 1393 } 1394 if (dontroute) { 1395 SOCK_LOCK(so); 1396 so->so_options |= SO_DONTROUTE; 1397 SOCK_UNLOCK(so); 1398 } 1399 /* 1400 * XXX all the SBS_CANTSENDMORE checks previously 1401 * done could be out of date. We could have recieved 1402 * a reset packet in an interrupt or maybe we slept 1403 * while doing page faults in uiomove() etc. We 1404 * could probably recheck again inside the locking 1405 * protection here, but there are probably other 1406 * places that this also happens. We must rethink 1407 * this. 1408 */ 1409 VNET_SO_ASSERT(so); 1410 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1411 (flags & MSG_OOB) ? PRUS_OOB : 1412 /* 1413 * If the user set MSG_EOF, the protocol understands 1414 * this flag and nothing left to send then use 1415 * PRU_SEND_EOF instead of PRU_SEND. 1416 */ 1417 ((flags & MSG_EOF) && 1418 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1419 (resid <= 0)) ? 1420 PRUS_EOF : 1421 /* If there is more to send set PRUS_MORETOCOME. */ 1422 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1423 top, addr, control, td); 1424 if (dontroute) { 1425 SOCK_LOCK(so); 1426 so->so_options &= ~SO_DONTROUTE; 1427 SOCK_UNLOCK(so); 1428 } 1429 clen = 0; 1430 control = NULL; 1431 top = NULL; 1432 if (error) 1433 goto release; 1434 } while (resid && space > 0); 1435 } while (resid); 1436 1437 release: 1438 sbunlock(&so->so_snd); 1439 out: 1440 if (top != NULL) 1441 m_freem(top); 1442 if (control != NULL) 1443 m_freem(control); 1444 return (error); 1445 } 1446 1447 int 1448 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1449 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1450 { 1451 int error; 1452 1453 CURVNET_SET(so->so_vnet); 1454 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top, 1455 control, flags, td); 1456 CURVNET_RESTORE(); 1457 return (error); 1458 } 1459 1460 /* 1461 * The part of soreceive() that implements reading non-inline out-of-band 1462 * data from a socket. For more complete comments, see soreceive(), from 1463 * which this code originated. 1464 * 1465 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1466 * unable to return an mbuf chain to the caller. 1467 */ 1468 static int 1469 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1470 { 1471 struct protosw *pr = so->so_proto; 1472 struct mbuf *m; 1473 int error; 1474 1475 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1476 VNET_SO_ASSERT(so); 1477 1478 m = m_get(M_WAITOK, MT_DATA); 1479 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 1480 if (error) 1481 goto bad; 1482 do { 1483 #ifdef SOCKET_RECV_PFLIP 1484 if (so_zero_copy_receive) { 1485 int disposable; 1486 1487 if ((m->m_flags & M_EXT) 1488 && (m->m_ext.ext_type == EXT_DISPOSABLE)) 1489 disposable = 1; 1490 else 1491 disposable = 0; 1492 1493 error = uiomoveco(mtod(m, void *), 1494 min(uio->uio_resid, m->m_len), uio, disposable); 1495 } else 1496 #endif /* SOCKET_RECV_PFLIP */ 1497 error = uiomove(mtod(m, void *), 1498 (int) min(uio->uio_resid, m->m_len), uio); 1499 m = m_free(m); 1500 } while (uio->uio_resid && error == 0 && m); 1501 bad: 1502 if (m != NULL) 1503 m_freem(m); 1504 return (error); 1505 } 1506 1507 /* 1508 * Following replacement or removal of the first mbuf on the first mbuf chain 1509 * of a socket buffer, push necessary state changes back into the socket 1510 * buffer so that other consumers see the values consistently. 'nextrecord' 1511 * is the callers locally stored value of the original value of 1512 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1513 * NOTE: 'nextrecord' may be NULL. 1514 */ 1515 static __inline void 1516 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1517 { 1518 1519 SOCKBUF_LOCK_ASSERT(sb); 1520 /* 1521 * First, update for the new value of nextrecord. If necessary, make 1522 * it the first record. 1523 */ 1524 if (sb->sb_mb != NULL) 1525 sb->sb_mb->m_nextpkt = nextrecord; 1526 else 1527 sb->sb_mb = nextrecord; 1528 1529 /* 1530 * Now update any dependent socket buffer fields to reflect the new 1531 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1532 * addition of a second clause that takes care of the case where 1533 * sb_mb has been updated, but remains the last record. 1534 */ 1535 if (sb->sb_mb == NULL) { 1536 sb->sb_mbtail = NULL; 1537 sb->sb_lastrecord = NULL; 1538 } else if (sb->sb_mb->m_nextpkt == NULL) 1539 sb->sb_lastrecord = sb->sb_mb; 1540 } 1541 1542 /* 1543 * Implement receive operations on a socket. We depend on the way that 1544 * records are added to the sockbuf by sbappend. In particular, each record 1545 * (mbufs linked through m_next) must begin with an address if the protocol 1546 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1547 * data, and then zero or more mbufs of data. In order to allow parallelism 1548 * between network receive and copying to user space, as well as avoid 1549 * sleeping with a mutex held, we release the socket buffer mutex during the 1550 * user space copy. Although the sockbuf is locked, new data may still be 1551 * appended, and thus we must maintain consistency of the sockbuf during that 1552 * time. 1553 * 1554 * The caller may receive the data as a single mbuf chain by supplying an 1555 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1556 * the count in uio_resid. 1557 */ 1558 int 1559 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1560 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1561 { 1562 struct mbuf *m, **mp; 1563 int flags, error, offset; 1564 ssize_t len; 1565 struct protosw *pr = so->so_proto; 1566 struct mbuf *nextrecord; 1567 int moff, type = 0; 1568 ssize_t orig_resid = uio->uio_resid; 1569 1570 mp = mp0; 1571 if (psa != NULL) 1572 *psa = NULL; 1573 if (controlp != NULL) 1574 *controlp = NULL; 1575 if (flagsp != NULL) 1576 flags = *flagsp &~ MSG_EOR; 1577 else 1578 flags = 0; 1579 if (flags & MSG_OOB) 1580 return (soreceive_rcvoob(so, uio, flags)); 1581 if (mp != NULL) 1582 *mp = NULL; 1583 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 1584 && uio->uio_resid) { 1585 VNET_SO_ASSERT(so); 1586 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 1587 } 1588 1589 error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); 1590 if (error) 1591 return (error); 1592 1593 restart: 1594 SOCKBUF_LOCK(&so->so_rcv); 1595 m = so->so_rcv.sb_mb; 1596 /* 1597 * If we have less data than requested, block awaiting more (subject 1598 * to any timeout) if: 1599 * 1. the current count is less than the low water mark, or 1600 * 2. MSG_DONTWAIT is not set 1601 */ 1602 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1603 so->so_rcv.sb_cc < uio->uio_resid) && 1604 so->so_rcv.sb_cc < so->so_rcv.sb_lowat && 1605 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 1606 KASSERT(m != NULL || !so->so_rcv.sb_cc, 1607 ("receive: m == %p so->so_rcv.sb_cc == %u", 1608 m, so->so_rcv.sb_cc)); 1609 if (so->so_error) { 1610 if (m != NULL) 1611 goto dontblock; 1612 error = so->so_error; 1613 if ((flags & MSG_PEEK) == 0) 1614 so->so_error = 0; 1615 SOCKBUF_UNLOCK(&so->so_rcv); 1616 goto release; 1617 } 1618 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1619 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1620 if (m == NULL) { 1621 SOCKBUF_UNLOCK(&so->so_rcv); 1622 goto release; 1623 } else 1624 goto dontblock; 1625 } 1626 for (; m != NULL; m = m->m_next) 1627 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1628 m = so->so_rcv.sb_mb; 1629 goto dontblock; 1630 } 1631 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1632 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1633 SOCKBUF_UNLOCK(&so->so_rcv); 1634 error = ENOTCONN; 1635 goto release; 1636 } 1637 if (uio->uio_resid == 0) { 1638 SOCKBUF_UNLOCK(&so->so_rcv); 1639 goto release; 1640 } 1641 if ((so->so_state & SS_NBIO) || 1642 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1643 SOCKBUF_UNLOCK(&so->so_rcv); 1644 error = EWOULDBLOCK; 1645 goto release; 1646 } 1647 SBLASTRECORDCHK(&so->so_rcv); 1648 SBLASTMBUFCHK(&so->so_rcv); 1649 error = sbwait(&so->so_rcv); 1650 SOCKBUF_UNLOCK(&so->so_rcv); 1651 if (error) 1652 goto release; 1653 goto restart; 1654 } 1655 dontblock: 1656 /* 1657 * From this point onward, we maintain 'nextrecord' as a cache of the 1658 * pointer to the next record in the socket buffer. We must keep the 1659 * various socket buffer pointers and local stack versions of the 1660 * pointers in sync, pushing out modifications before dropping the 1661 * socket buffer mutex, and re-reading them when picking it up. 1662 * 1663 * Otherwise, we will race with the network stack appending new data 1664 * or records onto the socket buffer by using inconsistent/stale 1665 * versions of the field, possibly resulting in socket buffer 1666 * corruption. 1667 * 1668 * By holding the high-level sblock(), we prevent simultaneous 1669 * readers from pulling off the front of the socket buffer. 1670 */ 1671 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1672 if (uio->uio_td) 1673 uio->uio_td->td_ru.ru_msgrcv++; 1674 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 1675 SBLASTRECORDCHK(&so->so_rcv); 1676 SBLASTMBUFCHK(&so->so_rcv); 1677 nextrecord = m->m_nextpkt; 1678 if (pr->pr_flags & PR_ADDR) { 1679 KASSERT(m->m_type == MT_SONAME, 1680 ("m->m_type == %d", m->m_type)); 1681 orig_resid = 0; 1682 if (psa != NULL) 1683 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 1684 M_NOWAIT); 1685 if (flags & MSG_PEEK) { 1686 m = m->m_next; 1687 } else { 1688 sbfree(&so->so_rcv, m); 1689 so->so_rcv.sb_mb = m_free(m); 1690 m = so->so_rcv.sb_mb; 1691 sockbuf_pushsync(&so->so_rcv, nextrecord); 1692 } 1693 } 1694 1695 /* 1696 * Process one or more MT_CONTROL mbufs present before any data mbufs 1697 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 1698 * just copy the data; if !MSG_PEEK, we call into the protocol to 1699 * perform externalization (or freeing if controlp == NULL). 1700 */ 1701 if (m != NULL && m->m_type == MT_CONTROL) { 1702 struct mbuf *cm = NULL, *cmn; 1703 struct mbuf **cme = &cm; 1704 1705 do { 1706 if (flags & MSG_PEEK) { 1707 if (controlp != NULL) { 1708 *controlp = m_copy(m, 0, m->m_len); 1709 controlp = &(*controlp)->m_next; 1710 } 1711 m = m->m_next; 1712 } else { 1713 sbfree(&so->so_rcv, m); 1714 so->so_rcv.sb_mb = m->m_next; 1715 m->m_next = NULL; 1716 *cme = m; 1717 cme = &(*cme)->m_next; 1718 m = so->so_rcv.sb_mb; 1719 } 1720 } while (m != NULL && m->m_type == MT_CONTROL); 1721 if ((flags & MSG_PEEK) == 0) 1722 sockbuf_pushsync(&so->so_rcv, nextrecord); 1723 while (cm != NULL) { 1724 cmn = cm->m_next; 1725 cm->m_next = NULL; 1726 if (pr->pr_domain->dom_externalize != NULL) { 1727 SOCKBUF_UNLOCK(&so->so_rcv); 1728 VNET_SO_ASSERT(so); 1729 error = (*pr->pr_domain->dom_externalize) 1730 (cm, controlp, flags); 1731 SOCKBUF_LOCK(&so->so_rcv); 1732 } else if (controlp != NULL) 1733 *controlp = cm; 1734 else 1735 m_freem(cm); 1736 if (controlp != NULL) { 1737 orig_resid = 0; 1738 while (*controlp != NULL) 1739 controlp = &(*controlp)->m_next; 1740 } 1741 cm = cmn; 1742 } 1743 if (m != NULL) 1744 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 1745 else 1746 nextrecord = so->so_rcv.sb_mb; 1747 orig_resid = 0; 1748 } 1749 if (m != NULL) { 1750 if ((flags & MSG_PEEK) == 0) { 1751 KASSERT(m->m_nextpkt == nextrecord, 1752 ("soreceive: post-control, nextrecord !sync")); 1753 if (nextrecord == NULL) { 1754 KASSERT(so->so_rcv.sb_mb == m, 1755 ("soreceive: post-control, sb_mb!=m")); 1756 KASSERT(so->so_rcv.sb_lastrecord == m, 1757 ("soreceive: post-control, lastrecord!=m")); 1758 } 1759 } 1760 type = m->m_type; 1761 if (type == MT_OOBDATA) 1762 flags |= MSG_OOB; 1763 } else { 1764 if ((flags & MSG_PEEK) == 0) { 1765 KASSERT(so->so_rcv.sb_mb == nextrecord, 1766 ("soreceive: sb_mb != nextrecord")); 1767 if (so->so_rcv.sb_mb == NULL) { 1768 KASSERT(so->so_rcv.sb_lastrecord == NULL, 1769 ("soreceive: sb_lastercord != NULL")); 1770 } 1771 } 1772 } 1773 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1774 SBLASTRECORDCHK(&so->so_rcv); 1775 SBLASTMBUFCHK(&so->so_rcv); 1776 1777 /* 1778 * Now continue to read any data mbufs off of the head of the socket 1779 * buffer until the read request is satisfied. Note that 'type' is 1780 * used to store the type of any mbuf reads that have happened so far 1781 * such that soreceive() can stop reading if the type changes, which 1782 * causes soreceive() to return only one of regular data and inline 1783 * out-of-band data in a single socket receive operation. 1784 */ 1785 moff = 0; 1786 offset = 0; 1787 while (m != NULL && uio->uio_resid > 0 && error == 0) { 1788 /* 1789 * If the type of mbuf has changed since the last mbuf 1790 * examined ('type'), end the receive operation. 1791 */ 1792 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1793 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 1794 if (type != m->m_type) 1795 break; 1796 } else if (type == MT_OOBDATA) 1797 break; 1798 else 1799 KASSERT(m->m_type == MT_DATA, 1800 ("m->m_type == %d", m->m_type)); 1801 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 1802 len = uio->uio_resid; 1803 if (so->so_oobmark && len > so->so_oobmark - offset) 1804 len = so->so_oobmark - offset; 1805 if (len > m->m_len - moff) 1806 len = m->m_len - moff; 1807 /* 1808 * If mp is set, just pass back the mbufs. Otherwise copy 1809 * them out via the uio, then free. Sockbuf must be 1810 * consistent here (points to current mbuf, it points to next 1811 * record) when we drop priority; we must note any additions 1812 * to the sockbuf when we block interrupts again. 1813 */ 1814 if (mp == NULL) { 1815 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1816 SBLASTRECORDCHK(&so->so_rcv); 1817 SBLASTMBUFCHK(&so->so_rcv); 1818 SOCKBUF_UNLOCK(&so->so_rcv); 1819 #ifdef SOCKET_RECV_PFLIP 1820 if (so_zero_copy_receive) { 1821 int disposable; 1822 1823 if ((m->m_flags & M_EXT) 1824 && (m->m_ext.ext_type == EXT_DISPOSABLE)) 1825 disposable = 1; 1826 else 1827 disposable = 0; 1828 1829 error = uiomoveco(mtod(m, char *) + moff, 1830 (int)len, uio, disposable); 1831 } else 1832 #endif /* SOCKET_RECV_PFLIP */ 1833 error = uiomove(mtod(m, char *) + moff, (int)len, uio); 1834 SOCKBUF_LOCK(&so->so_rcv); 1835 if (error) { 1836 /* 1837 * The MT_SONAME mbuf has already been removed 1838 * from the record, so it is necessary to 1839 * remove the data mbufs, if any, to preserve 1840 * the invariant in the case of PR_ADDR that 1841 * requires MT_SONAME mbufs at the head of 1842 * each record. 1843 */ 1844 if (m && pr->pr_flags & PR_ATOMIC && 1845 ((flags & MSG_PEEK) == 0)) 1846 (void)sbdroprecord_locked(&so->so_rcv); 1847 SOCKBUF_UNLOCK(&so->so_rcv); 1848 goto release; 1849 } 1850 } else 1851 uio->uio_resid -= len; 1852 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1853 if (len == m->m_len - moff) { 1854 if (m->m_flags & M_EOR) 1855 flags |= MSG_EOR; 1856 if (flags & MSG_PEEK) { 1857 m = m->m_next; 1858 moff = 0; 1859 } else { 1860 nextrecord = m->m_nextpkt; 1861 sbfree(&so->so_rcv, m); 1862 if (mp != NULL) { 1863 m->m_nextpkt = NULL; 1864 *mp = m; 1865 mp = &m->m_next; 1866 so->so_rcv.sb_mb = m = m->m_next; 1867 *mp = NULL; 1868 } else { 1869 so->so_rcv.sb_mb = m_free(m); 1870 m = so->so_rcv.sb_mb; 1871 } 1872 sockbuf_pushsync(&so->so_rcv, nextrecord); 1873 SBLASTRECORDCHK(&so->so_rcv); 1874 SBLASTMBUFCHK(&so->so_rcv); 1875 } 1876 } else { 1877 if (flags & MSG_PEEK) 1878 moff += len; 1879 else { 1880 if (mp != NULL) { 1881 int copy_flag; 1882 1883 if (flags & MSG_DONTWAIT) 1884 copy_flag = M_NOWAIT; 1885 else 1886 copy_flag = M_WAIT; 1887 if (copy_flag == M_WAITOK) 1888 SOCKBUF_UNLOCK(&so->so_rcv); 1889 *mp = m_copym(m, 0, len, copy_flag); 1890 if (copy_flag == M_WAITOK) 1891 SOCKBUF_LOCK(&so->so_rcv); 1892 if (*mp == NULL) { 1893 /* 1894 * m_copym() couldn't 1895 * allocate an mbuf. Adjust 1896 * uio_resid back (it was 1897 * adjusted down by len 1898 * bytes, which we didn't end 1899 * up "copying" over). 1900 */ 1901 uio->uio_resid += len; 1902 break; 1903 } 1904 } 1905 m->m_data += len; 1906 m->m_len -= len; 1907 so->so_rcv.sb_cc -= len; 1908 } 1909 } 1910 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1911 if (so->so_oobmark) { 1912 if ((flags & MSG_PEEK) == 0) { 1913 so->so_oobmark -= len; 1914 if (so->so_oobmark == 0) { 1915 so->so_rcv.sb_state |= SBS_RCVATMARK; 1916 break; 1917 } 1918 } else { 1919 offset += len; 1920 if (offset == so->so_oobmark) 1921 break; 1922 } 1923 } 1924 if (flags & MSG_EOR) 1925 break; 1926 /* 1927 * If the MSG_WAITALL flag is set (for non-atomic socket), we 1928 * must not quit until "uio->uio_resid == 0" or an error 1929 * termination. If a signal/timeout occurs, return with a 1930 * short count but without error. Keep sockbuf locked 1931 * against other readers. 1932 */ 1933 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 1934 !sosendallatonce(so) && nextrecord == NULL) { 1935 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1936 if (so->so_error || 1937 so->so_rcv.sb_state & SBS_CANTRCVMORE) 1938 break; 1939 /* 1940 * Notify the protocol that some data has been 1941 * drained before blocking. 1942 */ 1943 if (pr->pr_flags & PR_WANTRCVD) { 1944 SOCKBUF_UNLOCK(&so->so_rcv); 1945 VNET_SO_ASSERT(so); 1946 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 1947 SOCKBUF_LOCK(&so->so_rcv); 1948 } 1949 SBLASTRECORDCHK(&so->so_rcv); 1950 SBLASTMBUFCHK(&so->so_rcv); 1951 /* 1952 * We could receive some data while was notifying 1953 * the protocol. Skip blocking in this case. 1954 */ 1955 if (so->so_rcv.sb_mb == NULL) { 1956 error = sbwait(&so->so_rcv); 1957 if (error) { 1958 SOCKBUF_UNLOCK(&so->so_rcv); 1959 goto release; 1960 } 1961 } 1962 m = so->so_rcv.sb_mb; 1963 if (m != NULL) 1964 nextrecord = m->m_nextpkt; 1965 } 1966 } 1967 1968 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1969 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 1970 flags |= MSG_TRUNC; 1971 if ((flags & MSG_PEEK) == 0) 1972 (void) sbdroprecord_locked(&so->so_rcv); 1973 } 1974 if ((flags & MSG_PEEK) == 0) { 1975 if (m == NULL) { 1976 /* 1977 * First part is an inline SB_EMPTY_FIXUP(). Second 1978 * part makes sure sb_lastrecord is up-to-date if 1979 * there is still data in the socket buffer. 1980 */ 1981 so->so_rcv.sb_mb = nextrecord; 1982 if (so->so_rcv.sb_mb == NULL) { 1983 so->so_rcv.sb_mbtail = NULL; 1984 so->so_rcv.sb_lastrecord = NULL; 1985 } else if (nextrecord->m_nextpkt == NULL) 1986 so->so_rcv.sb_lastrecord = nextrecord; 1987 } 1988 SBLASTRECORDCHK(&so->so_rcv); 1989 SBLASTMBUFCHK(&so->so_rcv); 1990 /* 1991 * If soreceive() is being done from the socket callback, 1992 * then don't need to generate ACK to peer to update window, 1993 * since ACK will be generated on return to TCP. 1994 */ 1995 if (!(flags & MSG_SOCALLBCK) && 1996 (pr->pr_flags & PR_WANTRCVD)) { 1997 SOCKBUF_UNLOCK(&so->so_rcv); 1998 VNET_SO_ASSERT(so); 1999 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2000 SOCKBUF_LOCK(&so->so_rcv); 2001 } 2002 } 2003 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2004 if (orig_resid == uio->uio_resid && orig_resid && 2005 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 2006 SOCKBUF_UNLOCK(&so->so_rcv); 2007 goto restart; 2008 } 2009 SOCKBUF_UNLOCK(&so->so_rcv); 2010 2011 if (flagsp != NULL) 2012 *flagsp |= flags; 2013 release: 2014 sbunlock(&so->so_rcv); 2015 return (error); 2016 } 2017 2018 /* 2019 * Optimized version of soreceive() for stream (TCP) sockets. 2020 * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled. 2021 */ 2022 int 2023 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 2024 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2025 { 2026 int len = 0, error = 0, flags, oresid; 2027 struct sockbuf *sb; 2028 struct mbuf *m, *n = NULL; 2029 2030 /* We only do stream sockets. */ 2031 if (so->so_type != SOCK_STREAM) 2032 return (EINVAL); 2033 if (psa != NULL) 2034 *psa = NULL; 2035 if (controlp != NULL) 2036 return (EINVAL); 2037 if (flagsp != NULL) 2038 flags = *flagsp &~ MSG_EOR; 2039 else 2040 flags = 0; 2041 if (flags & MSG_OOB) 2042 return (soreceive_rcvoob(so, uio, flags)); 2043 if (mp0 != NULL) 2044 *mp0 = NULL; 2045 2046 sb = &so->so_rcv; 2047 2048 /* Prevent other readers from entering the socket. */ 2049 error = sblock(sb, SBLOCKWAIT(flags)); 2050 if (error) 2051 goto out; 2052 SOCKBUF_LOCK(sb); 2053 2054 /* Easy one, no space to copyout anything. */ 2055 if (uio->uio_resid == 0) { 2056 error = EINVAL; 2057 goto out; 2058 } 2059 oresid = uio->uio_resid; 2060 2061 /* We will never ever get anything unless we are or were connected. */ 2062 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 2063 error = ENOTCONN; 2064 goto out; 2065 } 2066 2067 restart: 2068 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2069 2070 /* Abort if socket has reported problems. */ 2071 if (so->so_error) { 2072 if (sb->sb_cc > 0) 2073 goto deliver; 2074 if (oresid > uio->uio_resid) 2075 goto out; 2076 error = so->so_error; 2077 if (!(flags & MSG_PEEK)) 2078 so->so_error = 0; 2079 goto out; 2080 } 2081 2082 /* Door is closed. Deliver what is left, if any. */ 2083 if (sb->sb_state & SBS_CANTRCVMORE) { 2084 if (sb->sb_cc > 0) 2085 goto deliver; 2086 else 2087 goto out; 2088 } 2089 2090 /* Socket buffer is empty and we shall not block. */ 2091 if (sb->sb_cc == 0 && 2092 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2093 error = EAGAIN; 2094 goto out; 2095 } 2096 2097 /* Socket buffer got some data that we shall deliver now. */ 2098 if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) && 2099 ((sb->sb_flags & SS_NBIO) || 2100 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2101 sb->sb_cc >= sb->sb_lowat || 2102 sb->sb_cc >= uio->uio_resid || 2103 sb->sb_cc >= sb->sb_hiwat) ) { 2104 goto deliver; 2105 } 2106 2107 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2108 if ((flags & MSG_WAITALL) && 2109 (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_hiwat)) 2110 goto deliver; 2111 2112 /* 2113 * Wait and block until (more) data comes in. 2114 * NB: Drops the sockbuf lock during wait. 2115 */ 2116 error = sbwait(sb); 2117 if (error) 2118 goto out; 2119 goto restart; 2120 2121 deliver: 2122 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2123 KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__)); 2124 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2125 2126 /* Statistics. */ 2127 if (uio->uio_td) 2128 uio->uio_td->td_ru.ru_msgrcv++; 2129 2130 /* Fill uio until full or current end of socket buffer is reached. */ 2131 len = min(uio->uio_resid, sb->sb_cc); 2132 if (mp0 != NULL) { 2133 /* Dequeue as many mbufs as possible. */ 2134 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2135 if (*mp0 == NULL) 2136 *mp0 = sb->sb_mb; 2137 else 2138 m_cat(*mp0, sb->sb_mb); 2139 for (m = sb->sb_mb; 2140 m != NULL && m->m_len <= len; 2141 m = m->m_next) { 2142 len -= m->m_len; 2143 uio->uio_resid -= m->m_len; 2144 sbfree(sb, m); 2145 n = m; 2146 } 2147 n->m_next = NULL; 2148 sb->sb_mb = m; 2149 sb->sb_lastrecord = sb->sb_mb; 2150 if (sb->sb_mb == NULL) 2151 SB_EMPTY_FIXUP(sb); 2152 } 2153 /* Copy the remainder. */ 2154 if (len > 0) { 2155 KASSERT(sb->sb_mb != NULL, 2156 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2157 2158 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2159 if (m == NULL) 2160 len = 0; /* Don't flush data from sockbuf. */ 2161 else 2162 uio->uio_resid -= len; 2163 if (*mp0 != NULL) 2164 m_cat(*mp0, m); 2165 else 2166 *mp0 = m; 2167 if (*mp0 == NULL) { 2168 error = ENOBUFS; 2169 goto out; 2170 } 2171 } 2172 } else { 2173 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2174 SOCKBUF_UNLOCK(sb); 2175 error = m_mbuftouio(uio, sb->sb_mb, len); 2176 SOCKBUF_LOCK(sb); 2177 if (error) 2178 goto out; 2179 } 2180 SBLASTRECORDCHK(sb); 2181 SBLASTMBUFCHK(sb); 2182 2183 /* 2184 * Remove the delivered data from the socket buffer unless we 2185 * were only peeking. 2186 */ 2187 if (!(flags & MSG_PEEK)) { 2188 if (len > 0) 2189 sbdrop_locked(sb, len); 2190 2191 /* Notify protocol that we drained some data. */ 2192 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2193 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2194 !(flags & MSG_SOCALLBCK))) { 2195 SOCKBUF_UNLOCK(sb); 2196 VNET_SO_ASSERT(so); 2197 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags); 2198 SOCKBUF_LOCK(sb); 2199 } 2200 } 2201 2202 /* 2203 * For MSG_WAITALL we may have to loop again and wait for 2204 * more data to come in. 2205 */ 2206 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2207 goto restart; 2208 out: 2209 SOCKBUF_LOCK_ASSERT(sb); 2210 SBLASTRECORDCHK(sb); 2211 SBLASTMBUFCHK(sb); 2212 SOCKBUF_UNLOCK(sb); 2213 sbunlock(sb); 2214 return (error); 2215 } 2216 2217 /* 2218 * Optimized version of soreceive() for simple datagram cases from userspace. 2219 * Unlike in the stream case, we're able to drop a datagram if copyout() 2220 * fails, and because we handle datagrams atomically, we don't need to use a 2221 * sleep lock to prevent I/O interlacing. 2222 */ 2223 int 2224 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2225 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2226 { 2227 struct mbuf *m, *m2; 2228 int flags, error; 2229 ssize_t len; 2230 struct protosw *pr = so->so_proto; 2231 struct mbuf *nextrecord; 2232 2233 if (psa != NULL) 2234 *psa = NULL; 2235 if (controlp != NULL) 2236 *controlp = NULL; 2237 if (flagsp != NULL) 2238 flags = *flagsp &~ MSG_EOR; 2239 else 2240 flags = 0; 2241 2242 /* 2243 * For any complicated cases, fall back to the full 2244 * soreceive_generic(). 2245 */ 2246 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB)) 2247 return (soreceive_generic(so, psa, uio, mp0, controlp, 2248 flagsp)); 2249 2250 /* 2251 * Enforce restrictions on use. 2252 */ 2253 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2254 ("soreceive_dgram: wantrcvd")); 2255 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2256 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2257 ("soreceive_dgram: SBS_RCVATMARK")); 2258 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2259 ("soreceive_dgram: P_CONNREQUIRED")); 2260 2261 /* 2262 * Loop blocking while waiting for a datagram. 2263 */ 2264 SOCKBUF_LOCK(&so->so_rcv); 2265 while ((m = so->so_rcv.sb_mb) == NULL) { 2266 KASSERT(so->so_rcv.sb_cc == 0, 2267 ("soreceive_dgram: sb_mb NULL but sb_cc %u", 2268 so->so_rcv.sb_cc)); 2269 if (so->so_error) { 2270 error = so->so_error; 2271 so->so_error = 0; 2272 SOCKBUF_UNLOCK(&so->so_rcv); 2273 return (error); 2274 } 2275 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2276 uio->uio_resid == 0) { 2277 SOCKBUF_UNLOCK(&so->so_rcv); 2278 return (0); 2279 } 2280 if ((so->so_state & SS_NBIO) || 2281 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2282 SOCKBUF_UNLOCK(&so->so_rcv); 2283 return (EWOULDBLOCK); 2284 } 2285 SBLASTRECORDCHK(&so->so_rcv); 2286 SBLASTMBUFCHK(&so->so_rcv); 2287 error = sbwait(&so->so_rcv); 2288 if (error) { 2289 SOCKBUF_UNLOCK(&so->so_rcv); 2290 return (error); 2291 } 2292 } 2293 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2294 2295 if (uio->uio_td) 2296 uio->uio_td->td_ru.ru_msgrcv++; 2297 SBLASTRECORDCHK(&so->so_rcv); 2298 SBLASTMBUFCHK(&so->so_rcv); 2299 nextrecord = m->m_nextpkt; 2300 if (nextrecord == NULL) { 2301 KASSERT(so->so_rcv.sb_lastrecord == m, 2302 ("soreceive_dgram: lastrecord != m")); 2303 } 2304 2305 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2306 ("soreceive_dgram: m_nextpkt != nextrecord")); 2307 2308 /* 2309 * Pull 'm' and its chain off the front of the packet queue. 2310 */ 2311 so->so_rcv.sb_mb = NULL; 2312 sockbuf_pushsync(&so->so_rcv, nextrecord); 2313 2314 /* 2315 * Walk 'm's chain and free that many bytes from the socket buffer. 2316 */ 2317 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2318 sbfree(&so->so_rcv, m2); 2319 2320 /* 2321 * Do a few last checks before we let go of the lock. 2322 */ 2323 SBLASTRECORDCHK(&so->so_rcv); 2324 SBLASTMBUFCHK(&so->so_rcv); 2325 SOCKBUF_UNLOCK(&so->so_rcv); 2326 2327 if (pr->pr_flags & PR_ADDR) { 2328 KASSERT(m->m_type == MT_SONAME, 2329 ("m->m_type == %d", m->m_type)); 2330 if (psa != NULL) 2331 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2332 M_NOWAIT); 2333 m = m_free(m); 2334 } 2335 if (m == NULL) { 2336 /* XXXRW: Can this happen? */ 2337 return (0); 2338 } 2339 2340 /* 2341 * Packet to copyout() is now in 'm' and it is disconnected from the 2342 * queue. 2343 * 2344 * Process one or more MT_CONTROL mbufs present before any data mbufs 2345 * in the first mbuf chain on the socket buffer. We call into the 2346 * protocol to perform externalization (or freeing if controlp == 2347 * NULL). 2348 */ 2349 if (m->m_type == MT_CONTROL) { 2350 struct mbuf *cm = NULL, *cmn; 2351 struct mbuf **cme = &cm; 2352 2353 do { 2354 m2 = m->m_next; 2355 m->m_next = NULL; 2356 *cme = m; 2357 cme = &(*cme)->m_next; 2358 m = m2; 2359 } while (m != NULL && m->m_type == MT_CONTROL); 2360 while (cm != NULL) { 2361 cmn = cm->m_next; 2362 cm->m_next = NULL; 2363 if (pr->pr_domain->dom_externalize != NULL) { 2364 error = (*pr->pr_domain->dom_externalize) 2365 (cm, controlp, flags); 2366 } else if (controlp != NULL) 2367 *controlp = cm; 2368 else 2369 m_freem(cm); 2370 if (controlp != NULL) { 2371 while (*controlp != NULL) 2372 controlp = &(*controlp)->m_next; 2373 } 2374 cm = cmn; 2375 } 2376 } 2377 KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data")); 2378 2379 while (m != NULL && uio->uio_resid > 0) { 2380 len = uio->uio_resid; 2381 if (len > m->m_len) 2382 len = m->m_len; 2383 error = uiomove(mtod(m, char *), (int)len, uio); 2384 if (error) { 2385 m_freem(m); 2386 return (error); 2387 } 2388 if (len == m->m_len) 2389 m = m_free(m); 2390 else { 2391 m->m_data += len; 2392 m->m_len -= len; 2393 } 2394 } 2395 if (m != NULL) 2396 flags |= MSG_TRUNC; 2397 m_freem(m); 2398 if (flagsp != NULL) 2399 *flagsp |= flags; 2400 return (0); 2401 } 2402 2403 int 2404 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2405 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2406 { 2407 int error; 2408 2409 CURVNET_SET(so->so_vnet); 2410 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0, 2411 controlp, flagsp)); 2412 CURVNET_RESTORE(); 2413 return (error); 2414 } 2415 2416 int 2417 soshutdown(struct socket *so, int how) 2418 { 2419 struct protosw *pr = so->so_proto; 2420 int error; 2421 2422 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2423 return (EINVAL); 2424 2425 CURVNET_SET(so->so_vnet); 2426 if (pr->pr_usrreqs->pru_flush != NULL) 2427 (*pr->pr_usrreqs->pru_flush)(so, how); 2428 if (how != SHUT_WR) 2429 sorflush(so); 2430 if (how != SHUT_RD) { 2431 error = (*pr->pr_usrreqs->pru_shutdown)(so); 2432 wakeup(&so->so_timeo); 2433 CURVNET_RESTORE(); 2434 return (error); 2435 } 2436 wakeup(&so->so_timeo); 2437 CURVNET_RESTORE(); 2438 return (0); 2439 } 2440 2441 void 2442 sorflush(struct socket *so) 2443 { 2444 struct sockbuf *sb = &so->so_rcv; 2445 struct protosw *pr = so->so_proto; 2446 struct sockbuf asb; 2447 2448 VNET_SO_ASSERT(so); 2449 2450 /* 2451 * In order to avoid calling dom_dispose with the socket buffer mutex 2452 * held, and in order to generally avoid holding the lock for a long 2453 * time, we make a copy of the socket buffer and clear the original 2454 * (except locks, state). The new socket buffer copy won't have 2455 * initialized locks so we can only call routines that won't use or 2456 * assert those locks. 2457 * 2458 * Dislodge threads currently blocked in receive and wait to acquire 2459 * a lock against other simultaneous readers before clearing the 2460 * socket buffer. Don't let our acquire be interrupted by a signal 2461 * despite any existing socket disposition on interruptable waiting. 2462 */ 2463 socantrcvmore(so); 2464 (void) sblock(sb, SBL_WAIT | SBL_NOINTR); 2465 2466 /* 2467 * Invalidate/clear most of the sockbuf structure, but leave selinfo 2468 * and mutex data unchanged. 2469 */ 2470 SOCKBUF_LOCK(sb); 2471 bzero(&asb, offsetof(struct sockbuf, sb_startzero)); 2472 bcopy(&sb->sb_startzero, &asb.sb_startzero, 2473 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2474 bzero(&sb->sb_startzero, 2475 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2476 SOCKBUF_UNLOCK(sb); 2477 sbunlock(sb); 2478 2479 /* 2480 * Dispose of special rights and flush the socket buffer. Don't call 2481 * any unsafe routines (that rely on locks being initialized) on asb. 2482 */ 2483 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 2484 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 2485 sbrelease_internal(&asb, so); 2486 } 2487 2488 /* 2489 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 2490 * additional variant to handle the case where the option value needs to be 2491 * some kind of integer, but not a specific size. In addition to their use 2492 * here, these functions are also called by the protocol-level pr_ctloutput() 2493 * routines. 2494 */ 2495 int 2496 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2497 { 2498 size_t valsize; 2499 2500 /* 2501 * If the user gives us more than we wanted, we ignore it, but if we 2502 * don't get the minimum length the caller wants, we return EINVAL. 2503 * On success, sopt->sopt_valsize is set to however much we actually 2504 * retrieved. 2505 */ 2506 if ((valsize = sopt->sopt_valsize) < minlen) 2507 return EINVAL; 2508 if (valsize > len) 2509 sopt->sopt_valsize = valsize = len; 2510 2511 if (sopt->sopt_td != NULL) 2512 return (copyin(sopt->sopt_val, buf, valsize)); 2513 2514 bcopy(sopt->sopt_val, buf, valsize); 2515 return (0); 2516 } 2517 2518 /* 2519 * Kernel version of setsockopt(2). 2520 * 2521 * XXX: optlen is size_t, not socklen_t 2522 */ 2523 int 2524 so_setsockopt(struct socket *so, int level, int optname, void *optval, 2525 size_t optlen) 2526 { 2527 struct sockopt sopt; 2528 2529 sopt.sopt_level = level; 2530 sopt.sopt_name = optname; 2531 sopt.sopt_dir = SOPT_SET; 2532 sopt.sopt_val = optval; 2533 sopt.sopt_valsize = optlen; 2534 sopt.sopt_td = NULL; 2535 return (sosetopt(so, &sopt)); 2536 } 2537 2538 int 2539 sosetopt(struct socket *so, struct sockopt *sopt) 2540 { 2541 int error, optval; 2542 struct linger l; 2543 struct timeval tv; 2544 u_long val; 2545 uint32_t val32; 2546 #ifdef MAC 2547 struct mac extmac; 2548 #endif 2549 2550 CURVNET_SET(so->so_vnet); 2551 error = 0; 2552 if (sopt->sopt_level != SOL_SOCKET) { 2553 if (so->so_proto->pr_ctloutput != NULL) { 2554 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2555 CURVNET_RESTORE(); 2556 return (error); 2557 } 2558 error = ENOPROTOOPT; 2559 } else { 2560 switch (sopt->sopt_name) { 2561 #ifdef INET 2562 case SO_ACCEPTFILTER: 2563 error = do_setopt_accept_filter(so, sopt); 2564 if (error) 2565 goto bad; 2566 break; 2567 #endif 2568 case SO_LINGER: 2569 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 2570 if (error) 2571 goto bad; 2572 2573 SOCK_LOCK(so); 2574 so->so_linger = l.l_linger; 2575 if (l.l_onoff) 2576 so->so_options |= SO_LINGER; 2577 else 2578 so->so_options &= ~SO_LINGER; 2579 SOCK_UNLOCK(so); 2580 break; 2581 2582 case SO_DEBUG: 2583 case SO_KEEPALIVE: 2584 case SO_DONTROUTE: 2585 case SO_USELOOPBACK: 2586 case SO_BROADCAST: 2587 case SO_REUSEADDR: 2588 case SO_REUSEPORT: 2589 case SO_OOBINLINE: 2590 case SO_TIMESTAMP: 2591 case SO_BINTIME: 2592 case SO_NOSIGPIPE: 2593 case SO_NO_DDP: 2594 case SO_NO_OFFLOAD: 2595 error = sooptcopyin(sopt, &optval, sizeof optval, 2596 sizeof optval); 2597 if (error) 2598 goto bad; 2599 SOCK_LOCK(so); 2600 if (optval) 2601 so->so_options |= sopt->sopt_name; 2602 else 2603 so->so_options &= ~sopt->sopt_name; 2604 SOCK_UNLOCK(so); 2605 break; 2606 2607 case SO_SETFIB: 2608 error = sooptcopyin(sopt, &optval, sizeof optval, 2609 sizeof optval); 2610 if (error) 2611 goto bad; 2612 2613 if (optval < 0 || optval >= rt_numfibs) { 2614 error = EINVAL; 2615 goto bad; 2616 } 2617 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 2618 (so->so_proto->pr_domain->dom_family == PF_INET6) || 2619 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 2620 so->so_fibnum = optval; 2621 else 2622 so->so_fibnum = 0; 2623 break; 2624 2625 case SO_USER_COOKIE: 2626 error = sooptcopyin(sopt, &val32, sizeof val32, 2627 sizeof val32); 2628 if (error) 2629 goto bad; 2630 so->so_user_cookie = val32; 2631 break; 2632 2633 case SO_SNDBUF: 2634 case SO_RCVBUF: 2635 case SO_SNDLOWAT: 2636 case SO_RCVLOWAT: 2637 error = sooptcopyin(sopt, &optval, sizeof optval, 2638 sizeof optval); 2639 if (error) 2640 goto bad; 2641 2642 /* 2643 * Values < 1 make no sense for any of these options, 2644 * so disallow them. 2645 */ 2646 if (optval < 1) { 2647 error = EINVAL; 2648 goto bad; 2649 } 2650 2651 switch (sopt->sopt_name) { 2652 case SO_SNDBUF: 2653 case SO_RCVBUF: 2654 if (sbreserve(sopt->sopt_name == SO_SNDBUF ? 2655 &so->so_snd : &so->so_rcv, (u_long)optval, 2656 so, curthread) == 0) { 2657 error = ENOBUFS; 2658 goto bad; 2659 } 2660 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd : 2661 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE; 2662 break; 2663 2664 /* 2665 * Make sure the low-water is never greater than the 2666 * high-water. 2667 */ 2668 case SO_SNDLOWAT: 2669 SOCKBUF_LOCK(&so->so_snd); 2670 so->so_snd.sb_lowat = 2671 (optval > so->so_snd.sb_hiwat) ? 2672 so->so_snd.sb_hiwat : optval; 2673 SOCKBUF_UNLOCK(&so->so_snd); 2674 break; 2675 case SO_RCVLOWAT: 2676 SOCKBUF_LOCK(&so->so_rcv); 2677 so->so_rcv.sb_lowat = 2678 (optval > so->so_rcv.sb_hiwat) ? 2679 so->so_rcv.sb_hiwat : optval; 2680 SOCKBUF_UNLOCK(&so->so_rcv); 2681 break; 2682 } 2683 break; 2684 2685 case SO_SNDTIMEO: 2686 case SO_RCVTIMEO: 2687 #ifdef COMPAT_FREEBSD32 2688 if (SV_CURPROC_FLAG(SV_ILP32)) { 2689 struct timeval32 tv32; 2690 2691 error = sooptcopyin(sopt, &tv32, sizeof tv32, 2692 sizeof tv32); 2693 CP(tv32, tv, tv_sec); 2694 CP(tv32, tv, tv_usec); 2695 } else 2696 #endif 2697 error = sooptcopyin(sopt, &tv, sizeof tv, 2698 sizeof tv); 2699 if (error) 2700 goto bad; 2701 2702 /* assert(hz > 0); */ 2703 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz || 2704 tv.tv_usec < 0 || tv.tv_usec >= 1000000) { 2705 error = EDOM; 2706 goto bad; 2707 } 2708 /* assert(tick > 0); */ 2709 /* assert(ULONG_MAX - INT_MAX >= 1000000); */ 2710 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick; 2711 if (val > INT_MAX) { 2712 error = EDOM; 2713 goto bad; 2714 } 2715 if (val == 0 && tv.tv_usec != 0) 2716 val = 1; 2717 2718 switch (sopt->sopt_name) { 2719 case SO_SNDTIMEO: 2720 so->so_snd.sb_timeo = val; 2721 break; 2722 case SO_RCVTIMEO: 2723 so->so_rcv.sb_timeo = val; 2724 break; 2725 } 2726 break; 2727 2728 case SO_LABEL: 2729 #ifdef MAC 2730 error = sooptcopyin(sopt, &extmac, sizeof extmac, 2731 sizeof extmac); 2732 if (error) 2733 goto bad; 2734 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 2735 so, &extmac); 2736 #else 2737 error = EOPNOTSUPP; 2738 #endif 2739 break; 2740 2741 default: 2742 error = ENOPROTOOPT; 2743 break; 2744 } 2745 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 2746 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 2747 } 2748 bad: 2749 CURVNET_RESTORE(); 2750 return (error); 2751 } 2752 2753 /* 2754 * Helper routine for getsockopt. 2755 */ 2756 int 2757 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 2758 { 2759 int error; 2760 size_t valsize; 2761 2762 error = 0; 2763 2764 /* 2765 * Documented get behavior is that we always return a value, possibly 2766 * truncated to fit in the user's buffer. Traditional behavior is 2767 * that we always tell the user precisely how much we copied, rather 2768 * than something useful like the total amount we had available for 2769 * her. Note that this interface is not idempotent; the entire 2770 * answer must generated ahead of time. 2771 */ 2772 valsize = min(len, sopt->sopt_valsize); 2773 sopt->sopt_valsize = valsize; 2774 if (sopt->sopt_val != NULL) { 2775 if (sopt->sopt_td != NULL) 2776 error = copyout(buf, sopt->sopt_val, valsize); 2777 else 2778 bcopy(buf, sopt->sopt_val, valsize); 2779 } 2780 return (error); 2781 } 2782 2783 int 2784 sogetopt(struct socket *so, struct sockopt *sopt) 2785 { 2786 int error, optval; 2787 struct linger l; 2788 struct timeval tv; 2789 #ifdef MAC 2790 struct mac extmac; 2791 #endif 2792 2793 CURVNET_SET(so->so_vnet); 2794 error = 0; 2795 if (sopt->sopt_level != SOL_SOCKET) { 2796 if (so->so_proto->pr_ctloutput != NULL) 2797 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2798 else 2799 error = ENOPROTOOPT; 2800 CURVNET_RESTORE(); 2801 return (error); 2802 } else { 2803 switch (sopt->sopt_name) { 2804 #ifdef INET 2805 case SO_ACCEPTFILTER: 2806 error = do_getopt_accept_filter(so, sopt); 2807 break; 2808 #endif 2809 case SO_LINGER: 2810 SOCK_LOCK(so); 2811 l.l_onoff = so->so_options & SO_LINGER; 2812 l.l_linger = so->so_linger; 2813 SOCK_UNLOCK(so); 2814 error = sooptcopyout(sopt, &l, sizeof l); 2815 break; 2816 2817 case SO_USELOOPBACK: 2818 case SO_DONTROUTE: 2819 case SO_DEBUG: 2820 case SO_KEEPALIVE: 2821 case SO_REUSEADDR: 2822 case SO_REUSEPORT: 2823 case SO_BROADCAST: 2824 case SO_OOBINLINE: 2825 case SO_ACCEPTCONN: 2826 case SO_TIMESTAMP: 2827 case SO_BINTIME: 2828 case SO_NOSIGPIPE: 2829 optval = so->so_options & sopt->sopt_name; 2830 integer: 2831 error = sooptcopyout(sopt, &optval, sizeof optval); 2832 break; 2833 2834 case SO_TYPE: 2835 optval = so->so_type; 2836 goto integer; 2837 2838 case SO_PROTOCOL: 2839 optval = so->so_proto->pr_protocol; 2840 goto integer; 2841 2842 case SO_ERROR: 2843 SOCK_LOCK(so); 2844 optval = so->so_error; 2845 so->so_error = 0; 2846 SOCK_UNLOCK(so); 2847 goto integer; 2848 2849 case SO_SNDBUF: 2850 optval = so->so_snd.sb_hiwat; 2851 goto integer; 2852 2853 case SO_RCVBUF: 2854 optval = so->so_rcv.sb_hiwat; 2855 goto integer; 2856 2857 case SO_SNDLOWAT: 2858 optval = so->so_snd.sb_lowat; 2859 goto integer; 2860 2861 case SO_RCVLOWAT: 2862 optval = so->so_rcv.sb_lowat; 2863 goto integer; 2864 2865 case SO_SNDTIMEO: 2866 case SO_RCVTIMEO: 2867 optval = (sopt->sopt_name == SO_SNDTIMEO ? 2868 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 2869 2870 tv.tv_sec = optval / hz; 2871 tv.tv_usec = (optval % hz) * tick; 2872 #ifdef COMPAT_FREEBSD32 2873 if (SV_CURPROC_FLAG(SV_ILP32)) { 2874 struct timeval32 tv32; 2875 2876 CP(tv, tv32, tv_sec); 2877 CP(tv, tv32, tv_usec); 2878 error = sooptcopyout(sopt, &tv32, sizeof tv32); 2879 } else 2880 #endif 2881 error = sooptcopyout(sopt, &tv, sizeof tv); 2882 break; 2883 2884 case SO_LABEL: 2885 #ifdef MAC 2886 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 2887 sizeof(extmac)); 2888 if (error) 2889 goto bad; 2890 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 2891 so, &extmac); 2892 if (error) 2893 goto bad; 2894 error = sooptcopyout(sopt, &extmac, sizeof extmac); 2895 #else 2896 error = EOPNOTSUPP; 2897 #endif 2898 break; 2899 2900 case SO_PEERLABEL: 2901 #ifdef MAC 2902 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 2903 sizeof(extmac)); 2904 if (error) 2905 goto bad; 2906 error = mac_getsockopt_peerlabel( 2907 sopt->sopt_td->td_ucred, so, &extmac); 2908 if (error) 2909 goto bad; 2910 error = sooptcopyout(sopt, &extmac, sizeof extmac); 2911 #else 2912 error = EOPNOTSUPP; 2913 #endif 2914 break; 2915 2916 case SO_LISTENQLIMIT: 2917 optval = so->so_qlimit; 2918 goto integer; 2919 2920 case SO_LISTENQLEN: 2921 optval = so->so_qlen; 2922 goto integer; 2923 2924 case SO_LISTENINCQLEN: 2925 optval = so->so_incqlen; 2926 goto integer; 2927 2928 default: 2929 error = ENOPROTOOPT; 2930 break; 2931 } 2932 } 2933 #ifdef MAC 2934 bad: 2935 #endif 2936 CURVNET_RESTORE(); 2937 return (error); 2938 } 2939 2940 int 2941 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 2942 { 2943 struct mbuf *m, *m_prev; 2944 int sopt_size = sopt->sopt_valsize; 2945 2946 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 2947 if (m == NULL) 2948 return ENOBUFS; 2949 if (sopt_size > MLEN) { 2950 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 2951 if ((m->m_flags & M_EXT) == 0) { 2952 m_free(m); 2953 return ENOBUFS; 2954 } 2955 m->m_len = min(MCLBYTES, sopt_size); 2956 } else { 2957 m->m_len = min(MLEN, sopt_size); 2958 } 2959 sopt_size -= m->m_len; 2960 *mp = m; 2961 m_prev = m; 2962 2963 while (sopt_size) { 2964 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 2965 if (m == NULL) { 2966 m_freem(*mp); 2967 return ENOBUFS; 2968 } 2969 if (sopt_size > MLEN) { 2970 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 2971 M_NOWAIT); 2972 if ((m->m_flags & M_EXT) == 0) { 2973 m_freem(m); 2974 m_freem(*mp); 2975 return ENOBUFS; 2976 } 2977 m->m_len = min(MCLBYTES, sopt_size); 2978 } else { 2979 m->m_len = min(MLEN, sopt_size); 2980 } 2981 sopt_size -= m->m_len; 2982 m_prev->m_next = m; 2983 m_prev = m; 2984 } 2985 return (0); 2986 } 2987 2988 int 2989 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 2990 { 2991 struct mbuf *m0 = m; 2992 2993 if (sopt->sopt_val == NULL) 2994 return (0); 2995 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 2996 if (sopt->sopt_td != NULL) { 2997 int error; 2998 2999 error = copyin(sopt->sopt_val, mtod(m, char *), 3000 m->m_len); 3001 if (error != 0) { 3002 m_freem(m0); 3003 return(error); 3004 } 3005 } else 3006 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 3007 sopt->sopt_valsize -= m->m_len; 3008 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3009 m = m->m_next; 3010 } 3011 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 3012 panic("ip6_sooptmcopyin"); 3013 return (0); 3014 } 3015 3016 int 3017 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 3018 { 3019 struct mbuf *m0 = m; 3020 size_t valsize = 0; 3021 3022 if (sopt->sopt_val == NULL) 3023 return (0); 3024 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3025 if (sopt->sopt_td != NULL) { 3026 int error; 3027 3028 error = copyout(mtod(m, char *), sopt->sopt_val, 3029 m->m_len); 3030 if (error != 0) { 3031 m_freem(m0); 3032 return(error); 3033 } 3034 } else 3035 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 3036 sopt->sopt_valsize -= m->m_len; 3037 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3038 valsize += m->m_len; 3039 m = m->m_next; 3040 } 3041 if (m != NULL) { 3042 /* enough soopt buffer should be given from user-land */ 3043 m_freem(m0); 3044 return(EINVAL); 3045 } 3046 sopt->sopt_valsize = valsize; 3047 return (0); 3048 } 3049 3050 /* 3051 * sohasoutofband(): protocol notifies socket layer of the arrival of new 3052 * out-of-band data, which will then notify socket consumers. 3053 */ 3054 void 3055 sohasoutofband(struct socket *so) 3056 { 3057 3058 if (so->so_sigio != NULL) 3059 pgsigio(&so->so_sigio, SIGURG, 0); 3060 selwakeuppri(&so->so_rcv.sb_sel, PSOCK); 3061 } 3062 3063 int 3064 sopoll(struct socket *so, int events, struct ucred *active_cred, 3065 struct thread *td) 3066 { 3067 3068 /* 3069 * We do not need to set or assert curvnet as long as everyone uses 3070 * sopoll_generic(). 3071 */ 3072 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, 3073 td)); 3074 } 3075 3076 int 3077 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3078 struct thread *td) 3079 { 3080 int revents = 0; 3081 3082 SOCKBUF_LOCK(&so->so_snd); 3083 SOCKBUF_LOCK(&so->so_rcv); 3084 if (events & (POLLIN | POLLRDNORM)) 3085 if (soreadabledata(so)) 3086 revents |= events & (POLLIN | POLLRDNORM); 3087 3088 if (events & (POLLOUT | POLLWRNORM)) 3089 if (sowriteable(so)) 3090 revents |= events & (POLLOUT | POLLWRNORM); 3091 3092 if (events & (POLLPRI | POLLRDBAND)) 3093 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK)) 3094 revents |= events & (POLLPRI | POLLRDBAND); 3095 3096 if ((events & POLLINIGNEOF) == 0) { 3097 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3098 revents |= events & (POLLIN | POLLRDNORM); 3099 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3100 revents |= POLLHUP; 3101 } 3102 } 3103 3104 if (revents == 0) { 3105 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 3106 selrecord(td, &so->so_rcv.sb_sel); 3107 so->so_rcv.sb_flags |= SB_SEL; 3108 } 3109 3110 if (events & (POLLOUT | POLLWRNORM)) { 3111 selrecord(td, &so->so_snd.sb_sel); 3112 so->so_snd.sb_flags |= SB_SEL; 3113 } 3114 } 3115 3116 SOCKBUF_UNLOCK(&so->so_rcv); 3117 SOCKBUF_UNLOCK(&so->so_snd); 3118 return (revents); 3119 } 3120 3121 int 3122 soo_kqfilter(struct file *fp, struct knote *kn) 3123 { 3124 struct socket *so = kn->kn_fp->f_data; 3125 struct sockbuf *sb; 3126 3127 switch (kn->kn_filter) { 3128 case EVFILT_READ: 3129 if (so->so_options & SO_ACCEPTCONN) 3130 kn->kn_fop = &solisten_filtops; 3131 else 3132 kn->kn_fop = &soread_filtops; 3133 sb = &so->so_rcv; 3134 break; 3135 case EVFILT_WRITE: 3136 kn->kn_fop = &sowrite_filtops; 3137 sb = &so->so_snd; 3138 break; 3139 default: 3140 return (EINVAL); 3141 } 3142 3143 SOCKBUF_LOCK(sb); 3144 knlist_add(&sb->sb_sel.si_note, kn, 1); 3145 sb->sb_flags |= SB_KNOTE; 3146 SOCKBUF_UNLOCK(sb); 3147 return (0); 3148 } 3149 3150 /* 3151 * Some routines that return EOPNOTSUPP for entry points that are not 3152 * supported by a protocol. Fill in as needed. 3153 */ 3154 int 3155 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 3156 { 3157 3158 return EOPNOTSUPP; 3159 } 3160 3161 int 3162 pru_attach_notsupp(struct socket *so, int proto, struct thread *td) 3163 { 3164 3165 return EOPNOTSUPP; 3166 } 3167 3168 int 3169 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3170 { 3171 3172 return EOPNOTSUPP; 3173 } 3174 3175 int 3176 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3177 struct thread *td) 3178 { 3179 3180 return EOPNOTSUPP; 3181 } 3182 3183 int 3184 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3185 { 3186 3187 return EOPNOTSUPP; 3188 } 3189 3190 int 3191 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3192 struct thread *td) 3193 { 3194 3195 return EOPNOTSUPP; 3196 } 3197 3198 int 3199 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 3200 { 3201 3202 return EOPNOTSUPP; 3203 } 3204 3205 int 3206 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 3207 struct ifnet *ifp, struct thread *td) 3208 { 3209 3210 return EOPNOTSUPP; 3211 } 3212 3213 int 3214 pru_disconnect_notsupp(struct socket *so) 3215 { 3216 3217 return EOPNOTSUPP; 3218 } 3219 3220 int 3221 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) 3222 { 3223 3224 return EOPNOTSUPP; 3225 } 3226 3227 int 3228 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) 3229 { 3230 3231 return EOPNOTSUPP; 3232 } 3233 3234 int 3235 pru_rcvd_notsupp(struct socket *so, int flags) 3236 { 3237 3238 return EOPNOTSUPP; 3239 } 3240 3241 int 3242 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 3243 { 3244 3245 return EOPNOTSUPP; 3246 } 3247 3248 int 3249 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, 3250 struct sockaddr *addr, struct mbuf *control, struct thread *td) 3251 { 3252 3253 return EOPNOTSUPP; 3254 } 3255 3256 /* 3257 * This isn't really a ``null'' operation, but it's the default one and 3258 * doesn't do anything destructive. 3259 */ 3260 int 3261 pru_sense_null(struct socket *so, struct stat *sb) 3262 { 3263 3264 sb->st_blksize = so->so_snd.sb_hiwat; 3265 return 0; 3266 } 3267 3268 int 3269 pru_shutdown_notsupp(struct socket *so) 3270 { 3271 3272 return EOPNOTSUPP; 3273 } 3274 3275 int 3276 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) 3277 { 3278 3279 return EOPNOTSUPP; 3280 } 3281 3282 int 3283 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 3284 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 3285 { 3286 3287 return EOPNOTSUPP; 3288 } 3289 3290 int 3291 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 3292 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 3293 { 3294 3295 return EOPNOTSUPP; 3296 } 3297 3298 int 3299 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, 3300 struct thread *td) 3301 { 3302 3303 return EOPNOTSUPP; 3304 } 3305 3306 static void 3307 filt_sordetach(struct knote *kn) 3308 { 3309 struct socket *so = kn->kn_fp->f_data; 3310 3311 SOCKBUF_LOCK(&so->so_rcv); 3312 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1); 3313 if (knlist_empty(&so->so_rcv.sb_sel.si_note)) 3314 so->so_rcv.sb_flags &= ~SB_KNOTE; 3315 SOCKBUF_UNLOCK(&so->so_rcv); 3316 } 3317 3318 /*ARGSUSED*/ 3319 static int 3320 filt_soread(struct knote *kn, long hint) 3321 { 3322 struct socket *so; 3323 3324 so = kn->kn_fp->f_data; 3325 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3326 3327 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl; 3328 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3329 kn->kn_flags |= EV_EOF; 3330 kn->kn_fflags = so->so_error; 3331 return (1); 3332 } else if (so->so_error) /* temporary udp error */ 3333 return (1); 3334 else if (kn->kn_sfflags & NOTE_LOWAT) 3335 return (kn->kn_data >= kn->kn_sdata); 3336 else 3337 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat); 3338 } 3339 3340 static void 3341 filt_sowdetach(struct knote *kn) 3342 { 3343 struct socket *so = kn->kn_fp->f_data; 3344 3345 SOCKBUF_LOCK(&so->so_snd); 3346 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1); 3347 if (knlist_empty(&so->so_snd.sb_sel.si_note)) 3348 so->so_snd.sb_flags &= ~SB_KNOTE; 3349 SOCKBUF_UNLOCK(&so->so_snd); 3350 } 3351 3352 /*ARGSUSED*/ 3353 static int 3354 filt_sowrite(struct knote *kn, long hint) 3355 { 3356 struct socket *so; 3357 3358 so = kn->kn_fp->f_data; 3359 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3360 kn->kn_data = sbspace(&so->so_snd); 3361 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3362 kn->kn_flags |= EV_EOF; 3363 kn->kn_fflags = so->so_error; 3364 return (1); 3365 } else if (so->so_error) /* temporary udp error */ 3366 return (1); 3367 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3368 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3369 return (0); 3370 else if (kn->kn_sfflags & NOTE_LOWAT) 3371 return (kn->kn_data >= kn->kn_sdata); 3372 else 3373 return (kn->kn_data >= so->so_snd.sb_lowat); 3374 } 3375 3376 /*ARGSUSED*/ 3377 static int 3378 filt_solisten(struct knote *kn, long hint) 3379 { 3380 struct socket *so = kn->kn_fp->f_data; 3381 3382 kn->kn_data = so->so_qlen; 3383 return (!TAILQ_EMPTY(&so->so_comp)); 3384 } 3385 3386 int 3387 socheckuid(struct socket *so, uid_t uid) 3388 { 3389 3390 if (so == NULL) 3391 return (EPERM); 3392 if (so->so_cred->cr_uid != uid) 3393 return (EPERM); 3394 return (0); 3395 } 3396 3397 /* 3398 * These functions are used by protocols to notify the socket layer (and its 3399 * consumers) of state changes in the sockets driven by protocol-side events. 3400 */ 3401 3402 /* 3403 * Procedures to manipulate state flags of socket and do appropriate wakeups. 3404 * 3405 * Normal sequence from the active (originating) side is that 3406 * soisconnecting() is called during processing of connect() call, resulting 3407 * in an eventual call to soisconnected() if/when the connection is 3408 * established. When the connection is torn down soisdisconnecting() is 3409 * called during processing of disconnect() call, and soisdisconnected() is 3410 * called when the connection to the peer is totally severed. The semantics 3411 * of these routines are such that connectionless protocols can call 3412 * soisconnected() and soisdisconnected() only, bypassing the in-progress 3413 * calls when setting up a ``connection'' takes no time. 3414 * 3415 * From the passive side, a socket is created with two queues of sockets: 3416 * so_incomp for connections in progress and so_comp for connections already 3417 * made and awaiting user acceptance. As a protocol is preparing incoming 3418 * connections, it creates a socket structure queued on so_incomp by calling 3419 * sonewconn(). When the connection is established, soisconnected() is 3420 * called, and transfers the socket structure to so_comp, making it available 3421 * to accept(). 3422 * 3423 * If a socket is closed with sockets on either so_incomp or so_comp, these 3424 * sockets are dropped. 3425 * 3426 * If higher-level protocols are implemented in the kernel, the wakeups done 3427 * here will sometimes cause software-interrupt process scheduling. 3428 */ 3429 void 3430 soisconnecting(struct socket *so) 3431 { 3432 3433 SOCK_LOCK(so); 3434 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 3435 so->so_state |= SS_ISCONNECTING; 3436 SOCK_UNLOCK(so); 3437 } 3438 3439 void 3440 soisconnected(struct socket *so) 3441 { 3442 struct socket *head; 3443 int ret; 3444 3445 restart: 3446 ACCEPT_LOCK(); 3447 SOCK_LOCK(so); 3448 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 3449 so->so_state |= SS_ISCONNECTED; 3450 head = so->so_head; 3451 if (head != NULL && (so->so_qstate & SQ_INCOMP)) { 3452 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 3453 SOCK_UNLOCK(so); 3454 TAILQ_REMOVE(&head->so_incomp, so, so_list); 3455 head->so_incqlen--; 3456 so->so_qstate &= ~SQ_INCOMP; 3457 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 3458 head->so_qlen++; 3459 so->so_qstate |= SQ_COMP; 3460 ACCEPT_UNLOCK(); 3461 sorwakeup(head); 3462 wakeup_one(&head->so_timeo); 3463 } else { 3464 ACCEPT_UNLOCK(); 3465 soupcall_set(so, SO_RCV, 3466 head->so_accf->so_accept_filter->accf_callback, 3467 head->so_accf->so_accept_filter_arg); 3468 so->so_options &= ~SO_ACCEPTFILTER; 3469 ret = head->so_accf->so_accept_filter->accf_callback(so, 3470 head->so_accf->so_accept_filter_arg, M_NOWAIT); 3471 if (ret == SU_ISCONNECTED) 3472 soupcall_clear(so, SO_RCV); 3473 SOCK_UNLOCK(so); 3474 if (ret == SU_ISCONNECTED) 3475 goto restart; 3476 } 3477 return; 3478 } 3479 SOCK_UNLOCK(so); 3480 ACCEPT_UNLOCK(); 3481 wakeup(&so->so_timeo); 3482 sorwakeup(so); 3483 sowwakeup(so); 3484 } 3485 3486 void 3487 soisdisconnecting(struct socket *so) 3488 { 3489 3490 /* 3491 * Note: This code assumes that SOCK_LOCK(so) and 3492 * SOCKBUF_LOCK(&so->so_rcv) are the same. 3493 */ 3494 SOCKBUF_LOCK(&so->so_rcv); 3495 so->so_state &= ~SS_ISCONNECTING; 3496 so->so_state |= SS_ISDISCONNECTING; 3497 so->so_rcv.sb_state |= SBS_CANTRCVMORE; 3498 sorwakeup_locked(so); 3499 SOCKBUF_LOCK(&so->so_snd); 3500 so->so_snd.sb_state |= SBS_CANTSENDMORE; 3501 sowwakeup_locked(so); 3502 wakeup(&so->so_timeo); 3503 } 3504 3505 void 3506 soisdisconnected(struct socket *so) 3507 { 3508 3509 /* 3510 * Note: This code assumes that SOCK_LOCK(so) and 3511 * SOCKBUF_LOCK(&so->so_rcv) are the same. 3512 */ 3513 SOCKBUF_LOCK(&so->so_rcv); 3514 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 3515 so->so_state |= SS_ISDISCONNECTED; 3516 so->so_rcv.sb_state |= SBS_CANTRCVMORE; 3517 sorwakeup_locked(so); 3518 SOCKBUF_LOCK(&so->so_snd); 3519 so->so_snd.sb_state |= SBS_CANTSENDMORE; 3520 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc); 3521 sowwakeup_locked(so); 3522 wakeup(&so->so_timeo); 3523 } 3524 3525 /* 3526 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 3527 */ 3528 struct sockaddr * 3529 sodupsockaddr(const struct sockaddr *sa, int mflags) 3530 { 3531 struct sockaddr *sa2; 3532 3533 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 3534 if (sa2) 3535 bcopy(sa, sa2, sa->sa_len); 3536 return sa2; 3537 } 3538 3539 /* 3540 * Register per-socket buffer upcalls. 3541 */ 3542 void 3543 soupcall_set(struct socket *so, int which, 3544 int (*func)(struct socket *, void *, int), void *arg) 3545 { 3546 struct sockbuf *sb; 3547 3548 switch (which) { 3549 case SO_RCV: 3550 sb = &so->so_rcv; 3551 break; 3552 case SO_SND: 3553 sb = &so->so_snd; 3554 break; 3555 default: 3556 panic("soupcall_set: bad which"); 3557 } 3558 SOCKBUF_LOCK_ASSERT(sb); 3559 #if 0 3560 /* XXX: accf_http actually wants to do this on purpose. */ 3561 KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall")); 3562 #endif 3563 sb->sb_upcall = func; 3564 sb->sb_upcallarg = arg; 3565 sb->sb_flags |= SB_UPCALL; 3566 } 3567 3568 void 3569 soupcall_clear(struct socket *so, int which) 3570 { 3571 struct sockbuf *sb; 3572 3573 switch (which) { 3574 case SO_RCV: 3575 sb = &so->so_rcv; 3576 break; 3577 case SO_SND: 3578 sb = &so->so_snd; 3579 break; 3580 default: 3581 panic("soupcall_clear: bad which"); 3582 } 3583 SOCKBUF_LOCK_ASSERT(sb); 3584 KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear")); 3585 sb->sb_upcall = NULL; 3586 sb->sb_upcallarg = NULL; 3587 sb->sb_flags &= ~SB_UPCALL; 3588 } 3589 3590 /* 3591 * Create an external-format (``xsocket'') structure using the information in 3592 * the kernel-format socket structure pointed to by so. This is done to 3593 * reduce the spew of irrelevant information over this interface, to isolate 3594 * user code from changes in the kernel structure, and potentially to provide 3595 * information-hiding if we decide that some of this information should be 3596 * hidden from users. 3597 */ 3598 void 3599 sotoxsocket(struct socket *so, struct xsocket *xso) 3600 { 3601 3602 xso->xso_len = sizeof *xso; 3603 xso->xso_so = so; 3604 xso->so_type = so->so_type; 3605 xso->so_options = so->so_options; 3606 xso->so_linger = so->so_linger; 3607 xso->so_state = so->so_state; 3608 xso->so_pcb = so->so_pcb; 3609 xso->xso_protocol = so->so_proto->pr_protocol; 3610 xso->xso_family = so->so_proto->pr_domain->dom_family; 3611 xso->so_qlen = so->so_qlen; 3612 xso->so_incqlen = so->so_incqlen; 3613 xso->so_qlimit = so->so_qlimit; 3614 xso->so_timeo = so->so_timeo; 3615 xso->so_error = so->so_error; 3616 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 3617 xso->so_oobmark = so->so_oobmark; 3618 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 3619 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 3620 xso->so_uid = so->so_cred->cr_uid; 3621 } 3622 3623 3624 /* 3625 * Socket accessor functions to provide external consumers with 3626 * a safe interface to socket state 3627 * 3628 */ 3629 3630 void 3631 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), 3632 void *arg) 3633 { 3634 3635 TAILQ_FOREACH(so, &so->so_comp, so_list) 3636 func(so, arg); 3637 } 3638 3639 struct sockbuf * 3640 so_sockbuf_rcv(struct socket *so) 3641 { 3642 3643 return (&so->so_rcv); 3644 } 3645 3646 struct sockbuf * 3647 so_sockbuf_snd(struct socket *so) 3648 { 3649 3650 return (&so->so_snd); 3651 } 3652 3653 int 3654 so_state_get(const struct socket *so) 3655 { 3656 3657 return (so->so_state); 3658 } 3659 3660 void 3661 so_state_set(struct socket *so, int val) 3662 { 3663 3664 so->so_state = val; 3665 } 3666 3667 int 3668 so_options_get(const struct socket *so) 3669 { 3670 3671 return (so->so_options); 3672 } 3673 3674 void 3675 so_options_set(struct socket *so, int val) 3676 { 3677 3678 so->so_options = val; 3679 } 3680 3681 int 3682 so_error_get(const struct socket *so) 3683 { 3684 3685 return (so->so_error); 3686 } 3687 3688 void 3689 so_error_set(struct socket *so, int val) 3690 { 3691 3692 so->so_error = val; 3693 } 3694 3695 int 3696 so_linger_get(const struct socket *so) 3697 { 3698 3699 return (so->so_linger); 3700 } 3701 3702 void 3703 so_linger_set(struct socket *so, int val) 3704 { 3705 3706 so->so_linger = val; 3707 } 3708 3709 struct protosw * 3710 so_protosw_get(const struct socket *so) 3711 { 3712 3713 return (so->so_proto); 3714 } 3715 3716 void 3717 so_protosw_set(struct socket *so, struct protosw *val) 3718 { 3719 3720 so->so_proto = val; 3721 } 3722 3723 void 3724 so_sorwakeup(struct socket *so) 3725 { 3726 3727 sorwakeup(so); 3728 } 3729 3730 void 3731 so_sowwakeup(struct socket *so) 3732 { 3733 3734 sowwakeup(so); 3735 } 3736 3737 void 3738 so_sorwakeup_locked(struct socket *so) 3739 { 3740 3741 sorwakeup_locked(so); 3742 } 3743 3744 void 3745 so_sowwakeup_locked(struct socket *so) 3746 { 3747 3748 sowwakeup_locked(so); 3749 } 3750 3751 void 3752 so_lock(struct socket *so) 3753 { 3754 3755 SOCK_LOCK(so); 3756 } 3757 3758 void 3759 so_unlock(struct socket *so) 3760 { 3761 3762 SOCK_UNLOCK(so); 3763 } 3764