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_compat.h" 109 110 #include <sys/param.h> 111 #include <sys/systm.h> 112 #include <sys/fcntl.h> 113 #include <sys/limits.h> 114 #include <sys/lock.h> 115 #include <sys/mac.h> 116 #include <sys/malloc.h> 117 #include <sys/mbuf.h> 118 #include <sys/mutex.h> 119 #include <sys/domain.h> 120 #include <sys/file.h> /* for struct knote */ 121 #include <sys/hhook.h> 122 #include <sys/kernel.h> 123 #include <sys/khelp.h> 124 #include <sys/event.h> 125 #include <sys/eventhandler.h> 126 #include <sys/poll.h> 127 #include <sys/proc.h> 128 #include <sys/protosw.h> 129 #include <sys/socket.h> 130 #include <sys/socketvar.h> 131 #include <sys/resourcevar.h> 132 #include <net/route.h> 133 #include <sys/signalvar.h> 134 #include <sys/stat.h> 135 #include <sys/sx.h> 136 #include <sys/sysctl.h> 137 #include <sys/uio.h> 138 #include <sys/jail.h> 139 #include <sys/syslog.h> 140 #include <netinet/in.h> 141 142 #include <net/vnet.h> 143 144 #include <security/mac/mac_framework.h> 145 146 #include <vm/uma.h> 147 148 #ifdef COMPAT_FREEBSD32 149 #include <sys/mount.h> 150 #include <sys/sysent.h> 151 #include <compat/freebsd32/freebsd32.h> 152 #endif 153 154 static int soreceive_rcvoob(struct socket *so, struct uio *uio, 155 int flags); 156 157 static void filt_sordetach(struct knote *kn); 158 static int filt_soread(struct knote *kn, long hint); 159 static void filt_sowdetach(struct knote *kn); 160 static int filt_sowrite(struct knote *kn, long hint); 161 static int filt_solisten(struct knote *kn, long hint); 162 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id); 163 fo_kqfilter_t soo_kqfilter; 164 165 static struct filterops solisten_filtops = { 166 .f_isfd = 1, 167 .f_detach = filt_sordetach, 168 .f_event = filt_solisten, 169 }; 170 static struct filterops soread_filtops = { 171 .f_isfd = 1, 172 .f_detach = filt_sordetach, 173 .f_event = filt_soread, 174 }; 175 static struct filterops sowrite_filtops = { 176 .f_isfd = 1, 177 .f_detach = filt_sowdetach, 178 .f_event = filt_sowrite, 179 }; 180 181 so_gen_t so_gencnt; /* generation count for sockets */ 182 183 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 184 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 185 186 #define VNET_SO_ASSERT(so) \ 187 VNET_ASSERT(curvnet != NULL, \ 188 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so))); 189 190 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]); 191 #define V_socket_hhh VNET(socket_hhh) 192 193 /* 194 * Limit on the number of connections in the listen queue waiting 195 * for accept(2). 196 * NB: The orginal sysctl somaxconn is still available but hidden 197 * to prevent confusion about the actual purpose of this number. 198 */ 199 static int somaxconn = SOMAXCONN; 200 201 static int 202 sysctl_somaxconn(SYSCTL_HANDLER_ARGS) 203 { 204 int error; 205 int val; 206 207 val = somaxconn; 208 error = sysctl_handle_int(oidp, &val, 0, req); 209 if (error || !req->newptr ) 210 return (error); 211 212 if (val < 1 || val > USHRT_MAX) 213 return (EINVAL); 214 215 somaxconn = val; 216 return (0); 217 } 218 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW, 219 0, sizeof(int), sysctl_somaxconn, "I", 220 "Maximum listen socket pending connection accept queue size"); 221 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, 222 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP, 223 0, sizeof(int), sysctl_somaxconn, "I", 224 "Maximum listen socket pending connection accept queue size (compat)"); 225 226 static int numopensockets; 227 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD, 228 &numopensockets, 0, "Number of open sockets"); 229 230 /* 231 * accept_mtx locks down per-socket fields relating to accept queues. See 232 * socketvar.h for an annotation of the protected fields of struct socket. 233 */ 234 struct mtx accept_mtx; 235 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF); 236 237 /* 238 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket 239 * so_gencnt field. 240 */ 241 static struct mtx so_global_mtx; 242 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF); 243 244 /* 245 * General IPC sysctl name space, used by sockets and a variety of other IPC 246 * types. 247 */ 248 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); 249 250 /* 251 * Initialize the socket subsystem and set up the socket 252 * memory allocator. 253 */ 254 static uma_zone_t socket_zone; 255 int maxsockets; 256 257 static void 258 socket_zone_change(void *tag) 259 { 260 261 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 262 } 263 264 static void 265 socket_hhook_register(int subtype) 266 { 267 268 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype, 269 &V_socket_hhh[subtype], 270 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 271 printf("%s: WARNING: unable to register hook\n", __func__); 272 } 273 274 static void 275 socket_hhook_deregister(int subtype) 276 { 277 278 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0) 279 printf("%s: WARNING: unable to deregister hook\n", __func__); 280 } 281 282 static void 283 socket_init(void *tag) 284 { 285 286 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL, 287 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 288 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 289 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached"); 290 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL, 291 EVENTHANDLER_PRI_FIRST); 292 } 293 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL); 294 295 static void 296 socket_vnet_init(const void *unused __unused) 297 { 298 int i; 299 300 /* We expect a contiguous range */ 301 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 302 socket_hhook_register(i); 303 } 304 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 305 socket_vnet_init, NULL); 306 307 static void 308 socket_vnet_uninit(const void *unused __unused) 309 { 310 int i; 311 312 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 313 socket_hhook_deregister(i); 314 } 315 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 316 socket_vnet_uninit, NULL); 317 318 /* 319 * Initialise maxsockets. This SYSINIT must be run after 320 * tunable_mbinit(). 321 */ 322 static void 323 init_maxsockets(void *ignored) 324 { 325 326 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 327 maxsockets = imax(maxsockets, maxfiles); 328 } 329 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); 330 331 /* 332 * Sysctl to get and set the maximum global sockets limit. Notify protocols 333 * of the change so that they can update their dependent limits as required. 334 */ 335 static int 336 sysctl_maxsockets(SYSCTL_HANDLER_ARGS) 337 { 338 int error, newmaxsockets; 339 340 newmaxsockets = maxsockets; 341 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req); 342 if (error == 0 && req->newptr) { 343 if (newmaxsockets > maxsockets && 344 newmaxsockets <= maxfiles) { 345 maxsockets = newmaxsockets; 346 EVENTHANDLER_INVOKE(maxsockets_change); 347 } else 348 error = EINVAL; 349 } 350 return (error); 351 } 352 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW, 353 &maxsockets, 0, sysctl_maxsockets, "IU", 354 "Maximum number of sockets avaliable"); 355 356 /* 357 * Socket operation routines. These routines are called by the routines in 358 * sys_socket.c or from a system process, and implement the semantics of 359 * socket operations by switching out to the protocol specific routines. 360 */ 361 362 /* 363 * Get a socket structure from our zone, and initialize it. Note that it 364 * would probably be better to allocate socket and PCB at the same time, but 365 * I'm not convinced that all the protocols can be easily modified to do 366 * this. 367 * 368 * soalloc() returns a socket with a ref count of 0. 369 */ 370 static struct socket * 371 soalloc(struct vnet *vnet) 372 { 373 struct socket *so; 374 375 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO); 376 if (so == NULL) 377 return (NULL); 378 #ifdef MAC 379 if (mac_socket_init(so, M_NOWAIT) != 0) { 380 uma_zfree(socket_zone, so); 381 return (NULL); 382 } 383 #endif 384 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) { 385 uma_zfree(socket_zone, so); 386 return (NULL); 387 } 388 389 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd"); 390 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv"); 391 sx_init(&so->so_snd.sb_sx, "so_snd_sx"); 392 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx"); 393 TAILQ_INIT(&so->so_aiojobq); 394 #ifdef VIMAGE 395 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p", 396 __func__, __LINE__, so)); 397 so->so_vnet = vnet; 398 #endif 399 /* We shouldn't need the so_global_mtx */ 400 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) { 401 /* Do we need more comprehensive error returns? */ 402 uma_zfree(socket_zone, so); 403 return (NULL); 404 } 405 mtx_lock(&so_global_mtx); 406 so->so_gencnt = ++so_gencnt; 407 ++numopensockets; 408 #ifdef VIMAGE 409 vnet->vnet_sockcnt++; 410 #endif 411 mtx_unlock(&so_global_mtx); 412 413 return (so); 414 } 415 416 /* 417 * Free the storage associated with a socket at the socket layer, tear down 418 * locks, labels, etc. All protocol state is assumed already to have been 419 * torn down (and possibly never set up) by the caller. 420 */ 421 static void 422 sodealloc(struct socket *so) 423 { 424 425 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); 426 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL")); 427 428 mtx_lock(&so_global_mtx); 429 so->so_gencnt = ++so_gencnt; 430 --numopensockets; /* Could be below, but faster here. */ 431 #ifdef VIMAGE 432 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p", 433 __func__, __LINE__, so)); 434 so->so_vnet->vnet_sockcnt--; 435 #endif 436 mtx_unlock(&so_global_mtx); 437 if (so->so_rcv.sb_hiwat) 438 (void)chgsbsize(so->so_cred->cr_uidinfo, 439 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 440 if (so->so_snd.sb_hiwat) 441 (void)chgsbsize(so->so_cred->cr_uidinfo, 442 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 443 /* remove acccept filter if one is present. */ 444 if (so->so_accf != NULL) 445 do_setopt_accept_filter(so, NULL); 446 #ifdef MAC 447 mac_socket_destroy(so); 448 #endif 449 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE); 450 451 crfree(so->so_cred); 452 khelp_destroy_osd(&so->osd); 453 sx_destroy(&so->so_snd.sb_sx); 454 sx_destroy(&so->so_rcv.sb_sx); 455 SOCKBUF_LOCK_DESTROY(&so->so_snd); 456 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 457 uma_zfree(socket_zone, so); 458 } 459 460 /* 461 * socreate returns a socket with a ref count of 1. The socket should be 462 * closed with soclose(). 463 */ 464 int 465 socreate(int dom, struct socket **aso, int type, int proto, 466 struct ucred *cred, struct thread *td) 467 { 468 struct protosw *prp; 469 struct socket *so; 470 int error; 471 472 if (proto) 473 prp = pffindproto(dom, proto, type); 474 else 475 prp = pffindtype(dom, type); 476 477 if (prp == NULL) { 478 /* No support for domain. */ 479 if (pffinddomain(dom) == NULL) 480 return (EAFNOSUPPORT); 481 /* No support for socket type. */ 482 if (proto == 0 && type != 0) 483 return (EPROTOTYPE); 484 return (EPROTONOSUPPORT); 485 } 486 if (prp->pr_usrreqs->pru_attach == NULL || 487 prp->pr_usrreqs->pru_attach == pru_attach_notsupp) 488 return (EPROTONOSUPPORT); 489 490 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0) 491 return (EPROTONOSUPPORT); 492 493 if (prp->pr_type != type) 494 return (EPROTOTYPE); 495 so = soalloc(CRED_TO_VNET(cred)); 496 if (so == NULL) 497 return (ENOBUFS); 498 499 TAILQ_INIT(&so->so_incomp); 500 TAILQ_INIT(&so->so_comp); 501 so->so_type = type; 502 so->so_cred = crhold(cred); 503 if ((prp->pr_domain->dom_family == PF_INET) || 504 (prp->pr_domain->dom_family == PF_INET6) || 505 (prp->pr_domain->dom_family == PF_ROUTE)) 506 so->so_fibnum = td->td_proc->p_fibnum; 507 else 508 so->so_fibnum = 0; 509 so->so_proto = prp; 510 #ifdef MAC 511 mac_socket_create(cred, so); 512 #endif 513 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv)); 514 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd)); 515 so->so_count = 1; 516 /* 517 * Auto-sizing of socket buffers is managed by the protocols and 518 * the appropriate flags must be set in the pru_attach function. 519 */ 520 CURVNET_SET(so->so_vnet); 521 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); 522 CURVNET_RESTORE(); 523 if (error) { 524 KASSERT(so->so_count == 1, ("socreate: so_count %d", 525 so->so_count)); 526 so->so_count = 0; 527 sodealloc(so); 528 return (error); 529 } 530 *aso = so; 531 return (0); 532 } 533 534 #ifdef REGRESSION 535 static int regression_sonewconn_earlytest = 1; 536 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW, 537 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test"); 538 #endif 539 540 /* 541 * When an attempt at a new connection is noted on a socket which accepts 542 * connections, sonewconn is called. If the connection is possible (subject 543 * to space constraints, etc.) then we allocate a new structure, propoerly 544 * linked into the data structure of the original socket, and return this. 545 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED. 546 * 547 * Note: the ref count on the socket is 0 on return. 548 */ 549 struct socket * 550 sonewconn(struct socket *head, int connstatus) 551 { 552 static struct timeval lastover; 553 static struct timeval overinterval = { 60, 0 }; 554 static int overcount; 555 556 struct socket *so; 557 int over; 558 559 ACCEPT_LOCK(); 560 over = (head->so_qlen > 3 * head->so_qlimit / 2); 561 ACCEPT_UNLOCK(); 562 #ifdef REGRESSION 563 if (regression_sonewconn_earlytest && over) { 564 #else 565 if (over) { 566 #endif 567 overcount++; 568 569 if (ratecheck(&lastover, &overinterval)) { 570 log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: " 571 "%i already in queue awaiting acceptance " 572 "(%d occurrences)\n", 573 __func__, head->so_pcb, head->so_qlen, overcount); 574 575 overcount = 0; 576 } 577 578 return (NULL); 579 } 580 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p", 581 __func__, __LINE__, head)); 582 so = soalloc(head->so_vnet); 583 if (so == NULL) { 584 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 585 "limit reached or out of memory\n", 586 __func__, head->so_pcb); 587 return (NULL); 588 } 589 if ((head->so_options & SO_ACCEPTFILTER) != 0) 590 connstatus = 0; 591 so->so_head = head; 592 so->so_type = head->so_type; 593 so->so_options = head->so_options &~ SO_ACCEPTCONN; 594 so->so_linger = head->so_linger; 595 so->so_state = head->so_state | SS_NOFDREF; 596 so->so_fibnum = head->so_fibnum; 597 so->so_proto = head->so_proto; 598 so->so_cred = crhold(head->so_cred); 599 #ifdef MAC 600 mac_socket_newconn(head, so); 601 #endif 602 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv)); 603 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd)); 604 VNET_SO_ASSERT(head); 605 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { 606 sodealloc(so); 607 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 608 __func__, head->so_pcb); 609 return (NULL); 610 } 611 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 612 sodealloc(so); 613 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 614 __func__, head->so_pcb); 615 return (NULL); 616 } 617 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 618 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 619 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 620 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 621 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; 622 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; 623 so->so_state |= connstatus; 624 ACCEPT_LOCK(); 625 /* 626 * The accept socket may be tearing down but we just 627 * won a race on the ACCEPT_LOCK. 628 * However, if sctp_peeloff() is called on a 1-to-many 629 * style socket, the SO_ACCEPTCONN doesn't need to be set. 630 */ 631 if (!(head->so_options & SO_ACCEPTCONN) && 632 ((head->so_proto->pr_protocol != IPPROTO_SCTP) || 633 (head->so_type != SOCK_SEQPACKET))) { 634 SOCK_LOCK(so); 635 so->so_head = NULL; 636 sofree(so); /* NB: returns ACCEPT_UNLOCK'ed. */ 637 return (NULL); 638 } 639 if (connstatus) { 640 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 641 so->so_qstate |= SQ_COMP; 642 head->so_qlen++; 643 } else { 644 /* 645 * Keep removing sockets from the head until there's room for 646 * us to insert on the tail. In pre-locking revisions, this 647 * was a simple if(), but as we could be racing with other 648 * threads and soabort() requires dropping locks, we must 649 * loop waiting for the condition to be true. 650 */ 651 while (head->so_incqlen > head->so_qlimit) { 652 struct socket *sp; 653 sp = TAILQ_FIRST(&head->so_incomp); 654 TAILQ_REMOVE(&head->so_incomp, sp, so_list); 655 head->so_incqlen--; 656 sp->so_qstate &= ~SQ_INCOMP; 657 sp->so_head = NULL; 658 ACCEPT_UNLOCK(); 659 soabort(sp); 660 ACCEPT_LOCK(); 661 } 662 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); 663 so->so_qstate |= SQ_INCOMP; 664 head->so_incqlen++; 665 } 666 ACCEPT_UNLOCK(); 667 if (connstatus) { 668 sorwakeup(head); 669 wakeup_one(&head->so_timeo); 670 } 671 return (so); 672 } 673 674 int 675 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 676 { 677 int error; 678 679 CURVNET_SET(so->so_vnet); 680 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td); 681 CURVNET_RESTORE(); 682 return (error); 683 } 684 685 int 686 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 687 { 688 int error; 689 690 CURVNET_SET(so->so_vnet); 691 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td); 692 CURVNET_RESTORE(); 693 return (error); 694 } 695 696 /* 697 * solisten() transitions a socket from a non-listening state to a listening 698 * state, but can also be used to update the listen queue depth on an 699 * existing listen socket. The protocol will call back into the sockets 700 * layer using solisten_proto_check() and solisten_proto() to check and set 701 * socket-layer listen state. Call backs are used so that the protocol can 702 * acquire both protocol and socket layer locks in whatever order is required 703 * by the protocol. 704 * 705 * Protocol implementors are advised to hold the socket lock across the 706 * socket-layer test and set to avoid races at the socket layer. 707 */ 708 int 709 solisten(struct socket *so, int backlog, struct thread *td) 710 { 711 int error; 712 713 CURVNET_SET(so->so_vnet); 714 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td); 715 CURVNET_RESTORE(); 716 return (error); 717 } 718 719 int 720 solisten_proto_check(struct socket *so) 721 { 722 723 SOCK_LOCK_ASSERT(so); 724 725 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 726 SS_ISDISCONNECTING)) 727 return (EINVAL); 728 return (0); 729 } 730 731 void 732 solisten_proto(struct socket *so, int backlog) 733 { 734 735 SOCK_LOCK_ASSERT(so); 736 737 if (backlog < 0 || backlog > somaxconn) 738 backlog = somaxconn; 739 so->so_qlimit = backlog; 740 so->so_options |= SO_ACCEPTCONN; 741 } 742 743 /* 744 * Evaluate the reference count and named references on a socket; if no 745 * references remain, free it. This should be called whenever a reference is 746 * released, such as in sorele(), but also when named reference flags are 747 * cleared in socket or protocol code. 748 * 749 * sofree() will free the socket if: 750 * 751 * - There are no outstanding file descriptor references or related consumers 752 * (so_count == 0). 753 * 754 * - The socket has been closed by user space, if ever open (SS_NOFDREF). 755 * 756 * - The protocol does not have an outstanding strong reference on the socket 757 * (SS_PROTOREF). 758 * 759 * - The socket is not in a completed connection queue, so a process has been 760 * notified that it is present. If it is removed, the user process may 761 * block in accept() despite select() saying the socket was ready. 762 */ 763 void 764 sofree(struct socket *so) 765 { 766 struct protosw *pr = so->so_proto; 767 struct socket *head; 768 769 ACCEPT_LOCK_ASSERT(); 770 SOCK_LOCK_ASSERT(so); 771 772 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 || 773 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) { 774 SOCK_UNLOCK(so); 775 ACCEPT_UNLOCK(); 776 return; 777 } 778 779 head = so->so_head; 780 if (head != NULL) { 781 KASSERT((so->so_qstate & SQ_COMP) != 0 || 782 (so->so_qstate & SQ_INCOMP) != 0, 783 ("sofree: so_head != NULL, but neither SQ_COMP nor " 784 "SQ_INCOMP")); 785 KASSERT((so->so_qstate & SQ_COMP) == 0 || 786 (so->so_qstate & SQ_INCOMP) == 0, 787 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP")); 788 TAILQ_REMOVE(&head->so_incomp, so, so_list); 789 head->so_incqlen--; 790 so->so_qstate &= ~SQ_INCOMP; 791 so->so_head = NULL; 792 } 793 KASSERT((so->so_qstate & SQ_COMP) == 0 && 794 (so->so_qstate & SQ_INCOMP) == 0, 795 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)", 796 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP)); 797 if (so->so_options & SO_ACCEPTCONN) { 798 KASSERT((TAILQ_EMPTY(&so->so_comp)), 799 ("sofree: so_comp populated")); 800 KASSERT((TAILQ_EMPTY(&so->so_incomp)), 801 ("sofree: so_incomp populated")); 802 } 803 SOCK_UNLOCK(so); 804 ACCEPT_UNLOCK(); 805 806 VNET_SO_ASSERT(so); 807 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 808 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb); 809 if (pr->pr_usrreqs->pru_detach != NULL) 810 (*pr->pr_usrreqs->pru_detach)(so); 811 812 /* 813 * From this point on, we assume that no other references to this 814 * socket exist anywhere else in the stack. Therefore, no locks need 815 * to be acquired or held. 816 * 817 * We used to do a lot of socket buffer and socket locking here, as 818 * well as invoke sorflush() and perform wakeups. The direct call to 819 * dom_dispose() and sbrelease_internal() are an inlining of what was 820 * necessary from sorflush(). 821 * 822 * Notice that the socket buffer and kqueue state are torn down 823 * before calling pru_detach. This means that protocols shold not 824 * assume they can perform socket wakeups, etc, in their detach code. 825 */ 826 sbdestroy(&so->so_snd, so); 827 sbdestroy(&so->so_rcv, so); 828 seldrain(&so->so_snd.sb_sel); 829 seldrain(&so->so_rcv.sb_sel); 830 knlist_destroy(&so->so_rcv.sb_sel.si_note); 831 knlist_destroy(&so->so_snd.sb_sel.si_note); 832 sodealloc(so); 833 } 834 835 /* 836 * Close a socket on last file table reference removal. Initiate disconnect 837 * if connected. Free socket when disconnect complete. 838 * 839 * This function will sorele() the socket. Note that soclose() may be called 840 * prior to the ref count reaching zero. The actual socket structure will 841 * not be freed until the ref count reaches zero. 842 */ 843 int 844 soclose(struct socket *so) 845 { 846 int error = 0; 847 848 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter")); 849 850 CURVNET_SET(so->so_vnet); 851 funsetown(&so->so_sigio); 852 if (so->so_state & SS_ISCONNECTED) { 853 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 854 error = sodisconnect(so); 855 if (error) { 856 if (error == ENOTCONN) 857 error = 0; 858 goto drop; 859 } 860 } 861 if (so->so_options & SO_LINGER) { 862 if ((so->so_state & SS_ISDISCONNECTING) && 863 (so->so_state & SS_NBIO)) 864 goto drop; 865 while (so->so_state & SS_ISCONNECTED) { 866 error = tsleep(&so->so_timeo, 867 PSOCK | PCATCH, "soclos", 868 so->so_linger * hz); 869 if (error) 870 break; 871 } 872 } 873 } 874 875 drop: 876 if (so->so_proto->pr_usrreqs->pru_close != NULL) 877 (*so->so_proto->pr_usrreqs->pru_close)(so); 878 ACCEPT_LOCK(); 879 if (so->so_options & SO_ACCEPTCONN) { 880 struct socket *sp; 881 /* 882 * Prevent new additions to the accept queues due 883 * to ACCEPT_LOCK races while we are draining them. 884 */ 885 so->so_options &= ~SO_ACCEPTCONN; 886 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) { 887 TAILQ_REMOVE(&so->so_incomp, sp, so_list); 888 so->so_incqlen--; 889 sp->so_qstate &= ~SQ_INCOMP; 890 sp->so_head = NULL; 891 ACCEPT_UNLOCK(); 892 soabort(sp); 893 ACCEPT_LOCK(); 894 } 895 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) { 896 TAILQ_REMOVE(&so->so_comp, sp, so_list); 897 so->so_qlen--; 898 sp->so_qstate &= ~SQ_COMP; 899 sp->so_head = NULL; 900 ACCEPT_UNLOCK(); 901 soabort(sp); 902 ACCEPT_LOCK(); 903 } 904 KASSERT((TAILQ_EMPTY(&so->so_comp)), 905 ("%s: so_comp populated", __func__)); 906 KASSERT((TAILQ_EMPTY(&so->so_incomp)), 907 ("%s: so_incomp populated", __func__)); 908 } 909 SOCK_LOCK(so); 910 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF")); 911 so->so_state |= SS_NOFDREF; 912 sorele(so); /* NB: Returns with ACCEPT_UNLOCK(). */ 913 CURVNET_RESTORE(); 914 return (error); 915 } 916 917 /* 918 * soabort() is used to abruptly tear down a connection, such as when a 919 * resource limit is reached (listen queue depth exceeded), or if a listen 920 * socket is closed while there are sockets waiting to be accepted. 921 * 922 * This interface is tricky, because it is called on an unreferenced socket, 923 * and must be called only by a thread that has actually removed the socket 924 * from the listen queue it was on, or races with other threads are risked. 925 * 926 * This interface will call into the protocol code, so must not be called 927 * with any socket locks held. Protocols do call it while holding their own 928 * recursible protocol mutexes, but this is something that should be subject 929 * to review in the future. 930 */ 931 void 932 soabort(struct socket *so) 933 { 934 935 /* 936 * In as much as is possible, assert that no references to this 937 * socket are held. This is not quite the same as asserting that the 938 * current thread is responsible for arranging for no references, but 939 * is as close as we can get for now. 940 */ 941 KASSERT(so->so_count == 0, ("soabort: so_count")); 942 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF")); 943 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF")); 944 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP")); 945 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP")); 946 VNET_SO_ASSERT(so); 947 948 if (so->so_proto->pr_usrreqs->pru_abort != NULL) 949 (*so->so_proto->pr_usrreqs->pru_abort)(so); 950 ACCEPT_LOCK(); 951 SOCK_LOCK(so); 952 sofree(so); 953 } 954 955 int 956 soaccept(struct socket *so, struct sockaddr **nam) 957 { 958 int error; 959 960 SOCK_LOCK(so); 961 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF")); 962 so->so_state &= ~SS_NOFDREF; 963 SOCK_UNLOCK(so); 964 965 CURVNET_SET(so->so_vnet); 966 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); 967 CURVNET_RESTORE(); 968 return (error); 969 } 970 971 int 972 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 973 { 974 975 return (soconnectat(AT_FDCWD, so, nam, td)); 976 } 977 978 int 979 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 980 { 981 int error; 982 983 if (so->so_options & SO_ACCEPTCONN) 984 return (EOPNOTSUPP); 985 986 CURVNET_SET(so->so_vnet); 987 /* 988 * If protocol is connection-based, can only connect once. 989 * Otherwise, if connected, try to disconnect first. This allows 990 * user to disconnect by connecting to, e.g., a null address. 991 */ 992 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 993 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 994 (error = sodisconnect(so)))) { 995 error = EISCONN; 996 } else { 997 /* 998 * Prevent accumulated error from previous connection from 999 * biting us. 1000 */ 1001 so->so_error = 0; 1002 if (fd == AT_FDCWD) { 1003 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, 1004 nam, td); 1005 } else { 1006 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd, 1007 so, nam, td); 1008 } 1009 } 1010 CURVNET_RESTORE(); 1011 1012 return (error); 1013 } 1014 1015 int 1016 soconnect2(struct socket *so1, struct socket *so2) 1017 { 1018 int error; 1019 1020 CURVNET_SET(so1->so_vnet); 1021 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); 1022 CURVNET_RESTORE(); 1023 return (error); 1024 } 1025 1026 int 1027 sodisconnect(struct socket *so) 1028 { 1029 int error; 1030 1031 if ((so->so_state & SS_ISCONNECTED) == 0) 1032 return (ENOTCONN); 1033 if (so->so_state & SS_ISDISCONNECTING) 1034 return (EALREADY); 1035 VNET_SO_ASSERT(so); 1036 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 1037 return (error); 1038 } 1039 1040 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) 1041 1042 int 1043 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1044 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1045 { 1046 long space; 1047 ssize_t resid; 1048 int clen = 0, error, dontroute; 1049 1050 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 1051 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 1052 ("sosend_dgram: !PR_ATOMIC")); 1053 1054 if (uio != NULL) 1055 resid = uio->uio_resid; 1056 else 1057 resid = top->m_pkthdr.len; 1058 /* 1059 * In theory resid should be unsigned. However, space must be 1060 * signed, as it might be less than 0 if we over-committed, and we 1061 * must use a signed comparison of space and resid. On the other 1062 * hand, a negative resid causes us to loop sending 0-length 1063 * segments to the protocol. 1064 */ 1065 if (resid < 0) { 1066 error = EINVAL; 1067 goto out; 1068 } 1069 1070 dontroute = 1071 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 1072 if (td != NULL) 1073 td->td_ru.ru_msgsnd++; 1074 if (control != NULL) 1075 clen = control->m_len; 1076 1077 SOCKBUF_LOCK(&so->so_snd); 1078 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1079 SOCKBUF_UNLOCK(&so->so_snd); 1080 error = EPIPE; 1081 goto out; 1082 } 1083 if (so->so_error) { 1084 error = so->so_error; 1085 so->so_error = 0; 1086 SOCKBUF_UNLOCK(&so->so_snd); 1087 goto out; 1088 } 1089 if ((so->so_state & SS_ISCONNECTED) == 0) { 1090 /* 1091 * `sendto' and `sendmsg' is allowed on a connection-based 1092 * socket if it supports implied connect. Return ENOTCONN if 1093 * not connected and no address is supplied. 1094 */ 1095 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1096 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1097 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1098 !(resid == 0 && clen != 0)) { 1099 SOCKBUF_UNLOCK(&so->so_snd); 1100 error = ENOTCONN; 1101 goto out; 1102 } 1103 } else if (addr == NULL) { 1104 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1105 error = ENOTCONN; 1106 else 1107 error = EDESTADDRREQ; 1108 SOCKBUF_UNLOCK(&so->so_snd); 1109 goto out; 1110 } 1111 } 1112 1113 /* 1114 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1115 * problem and need fixing. 1116 */ 1117 space = sbspace(&so->so_snd); 1118 if (flags & MSG_OOB) 1119 space += 1024; 1120 space -= clen; 1121 SOCKBUF_UNLOCK(&so->so_snd); 1122 if (resid > space) { 1123 error = EMSGSIZE; 1124 goto out; 1125 } 1126 if (uio == NULL) { 1127 resid = 0; 1128 if (flags & MSG_EOR) 1129 top->m_flags |= M_EOR; 1130 } else { 1131 /* 1132 * Copy the data from userland into a mbuf chain. 1133 * If no data is to be copied in, a single empty mbuf 1134 * is returned. 1135 */ 1136 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1137 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1138 if (top == NULL) { 1139 error = EFAULT; /* only possible error */ 1140 goto out; 1141 } 1142 space -= resid - uio->uio_resid; 1143 resid = uio->uio_resid; 1144 } 1145 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1146 /* 1147 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1148 * than with. 1149 */ 1150 if (dontroute) { 1151 SOCK_LOCK(so); 1152 so->so_options |= SO_DONTROUTE; 1153 SOCK_UNLOCK(so); 1154 } 1155 /* 1156 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1157 * of date. We could have recieved a reset packet in an interrupt or 1158 * maybe we slept while doing page faults in uiomove() etc. We could 1159 * probably recheck again inside the locking protection here, but 1160 * there are probably other places that this also happens. We must 1161 * rethink this. 1162 */ 1163 VNET_SO_ASSERT(so); 1164 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1165 (flags & MSG_OOB) ? PRUS_OOB : 1166 /* 1167 * If the user set MSG_EOF, the protocol understands this flag and 1168 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1169 */ 1170 ((flags & MSG_EOF) && 1171 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1172 (resid <= 0)) ? 1173 PRUS_EOF : 1174 /* If there is more to send set PRUS_MORETOCOME */ 1175 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1176 top, addr, control, td); 1177 if (dontroute) { 1178 SOCK_LOCK(so); 1179 so->so_options &= ~SO_DONTROUTE; 1180 SOCK_UNLOCK(so); 1181 } 1182 clen = 0; 1183 control = NULL; 1184 top = NULL; 1185 out: 1186 if (top != NULL) 1187 m_freem(top); 1188 if (control != NULL) 1189 m_freem(control); 1190 return (error); 1191 } 1192 1193 /* 1194 * Send on a socket. If send must go all at once and message is larger than 1195 * send buffering, then hard error. Lock against other senders. If must go 1196 * all at once and not enough room now, then inform user that this would 1197 * block and do nothing. Otherwise, if nonblocking, send as much as 1198 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1199 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1200 * in mbuf chain must be small enough to send all at once. 1201 * 1202 * Returns nonzero on error, timeout or signal; callers must check for short 1203 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1204 * on return. 1205 */ 1206 int 1207 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1208 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1209 { 1210 long space; 1211 ssize_t resid; 1212 int clen = 0, error, dontroute; 1213 int atomic = sosendallatonce(so) || top; 1214 1215 if (uio != NULL) 1216 resid = uio->uio_resid; 1217 else 1218 resid = top->m_pkthdr.len; 1219 /* 1220 * In theory resid should be unsigned. However, space must be 1221 * signed, as it might be less than 0 if we over-committed, and we 1222 * must use a signed comparison of space and resid. On the other 1223 * hand, a negative resid causes us to loop sending 0-length 1224 * segments to the protocol. 1225 * 1226 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1227 * type sockets since that's an error. 1228 */ 1229 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1230 error = EINVAL; 1231 goto out; 1232 } 1233 1234 dontroute = 1235 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1236 (so->so_proto->pr_flags & PR_ATOMIC); 1237 if (td != NULL) 1238 td->td_ru.ru_msgsnd++; 1239 if (control != NULL) 1240 clen = control->m_len; 1241 1242 error = sblock(&so->so_snd, SBLOCKWAIT(flags)); 1243 if (error) 1244 goto out; 1245 1246 restart: 1247 do { 1248 SOCKBUF_LOCK(&so->so_snd); 1249 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1250 SOCKBUF_UNLOCK(&so->so_snd); 1251 error = EPIPE; 1252 goto release; 1253 } 1254 if (so->so_error) { 1255 error = so->so_error; 1256 so->so_error = 0; 1257 SOCKBUF_UNLOCK(&so->so_snd); 1258 goto release; 1259 } 1260 if ((so->so_state & SS_ISCONNECTED) == 0) { 1261 /* 1262 * `sendto' and `sendmsg' is allowed on a connection- 1263 * based socket if it supports implied connect. 1264 * Return ENOTCONN if not connected and no address is 1265 * supplied. 1266 */ 1267 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1268 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1269 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1270 !(resid == 0 && clen != 0)) { 1271 SOCKBUF_UNLOCK(&so->so_snd); 1272 error = ENOTCONN; 1273 goto release; 1274 } 1275 } else if (addr == NULL) { 1276 SOCKBUF_UNLOCK(&so->so_snd); 1277 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1278 error = ENOTCONN; 1279 else 1280 error = EDESTADDRREQ; 1281 goto release; 1282 } 1283 } 1284 space = sbspace(&so->so_snd); 1285 if (flags & MSG_OOB) 1286 space += 1024; 1287 if ((atomic && resid > so->so_snd.sb_hiwat) || 1288 clen > so->so_snd.sb_hiwat) { 1289 SOCKBUF_UNLOCK(&so->so_snd); 1290 error = EMSGSIZE; 1291 goto release; 1292 } 1293 if (space < resid + clen && 1294 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1295 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) { 1296 SOCKBUF_UNLOCK(&so->so_snd); 1297 error = EWOULDBLOCK; 1298 goto release; 1299 } 1300 error = sbwait(&so->so_snd); 1301 SOCKBUF_UNLOCK(&so->so_snd); 1302 if (error) 1303 goto release; 1304 goto restart; 1305 } 1306 SOCKBUF_UNLOCK(&so->so_snd); 1307 space -= clen; 1308 do { 1309 if (uio == NULL) { 1310 resid = 0; 1311 if (flags & MSG_EOR) 1312 top->m_flags |= M_EOR; 1313 } else { 1314 /* 1315 * Copy the data from userland into a mbuf 1316 * chain. If resid is 0, which can happen 1317 * only if we have control to send, then 1318 * a single empty mbuf is returned. This 1319 * is a workaround to prevent protocol send 1320 * methods to panic. 1321 */ 1322 top = m_uiotombuf(uio, M_WAITOK, space, 1323 (atomic ? max_hdr : 0), 1324 (atomic ? M_PKTHDR : 0) | 1325 ((flags & MSG_EOR) ? M_EOR : 0)); 1326 if (top == NULL) { 1327 error = EFAULT; /* only possible error */ 1328 goto release; 1329 } 1330 space -= resid - uio->uio_resid; 1331 resid = uio->uio_resid; 1332 } 1333 if (dontroute) { 1334 SOCK_LOCK(so); 1335 so->so_options |= SO_DONTROUTE; 1336 SOCK_UNLOCK(so); 1337 } 1338 /* 1339 * XXX all the SBS_CANTSENDMORE checks previously 1340 * done could be out of date. We could have recieved 1341 * a reset packet in an interrupt or maybe we slept 1342 * while doing page faults in uiomove() etc. We 1343 * could probably recheck again inside the locking 1344 * protection here, but there are probably other 1345 * places that this also happens. We must rethink 1346 * this. 1347 */ 1348 VNET_SO_ASSERT(so); 1349 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1350 (flags & MSG_OOB) ? PRUS_OOB : 1351 /* 1352 * If the user set MSG_EOF, the protocol understands 1353 * this flag and nothing left to send then use 1354 * PRU_SEND_EOF instead of PRU_SEND. 1355 */ 1356 ((flags & MSG_EOF) && 1357 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1358 (resid <= 0)) ? 1359 PRUS_EOF : 1360 /* If there is more to send set PRUS_MORETOCOME. */ 1361 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1362 top, addr, control, td); 1363 if (dontroute) { 1364 SOCK_LOCK(so); 1365 so->so_options &= ~SO_DONTROUTE; 1366 SOCK_UNLOCK(so); 1367 } 1368 clen = 0; 1369 control = NULL; 1370 top = NULL; 1371 if (error) 1372 goto release; 1373 } while (resid && space > 0); 1374 } while (resid); 1375 1376 release: 1377 sbunlock(&so->so_snd); 1378 out: 1379 if (top != NULL) 1380 m_freem(top); 1381 if (control != NULL) 1382 m_freem(control); 1383 return (error); 1384 } 1385 1386 int 1387 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1388 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1389 { 1390 int error; 1391 1392 CURVNET_SET(so->so_vnet); 1393 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top, 1394 control, flags, td); 1395 CURVNET_RESTORE(); 1396 return (error); 1397 } 1398 1399 /* 1400 * The part of soreceive() that implements reading non-inline out-of-band 1401 * data from a socket. For more complete comments, see soreceive(), from 1402 * which this code originated. 1403 * 1404 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1405 * unable to return an mbuf chain to the caller. 1406 */ 1407 static int 1408 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1409 { 1410 struct protosw *pr = so->so_proto; 1411 struct mbuf *m; 1412 int error; 1413 1414 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1415 VNET_SO_ASSERT(so); 1416 1417 m = m_get(M_WAITOK, MT_DATA); 1418 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 1419 if (error) 1420 goto bad; 1421 do { 1422 error = uiomove(mtod(m, void *), 1423 (int) min(uio->uio_resid, m->m_len), uio); 1424 m = m_free(m); 1425 } while (uio->uio_resid && error == 0 && m); 1426 bad: 1427 if (m != NULL) 1428 m_freem(m); 1429 return (error); 1430 } 1431 1432 /* 1433 * Following replacement or removal of the first mbuf on the first mbuf chain 1434 * of a socket buffer, push necessary state changes back into the socket 1435 * buffer so that other consumers see the values consistently. 'nextrecord' 1436 * is the callers locally stored value of the original value of 1437 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1438 * NOTE: 'nextrecord' may be NULL. 1439 */ 1440 static __inline void 1441 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1442 { 1443 1444 SOCKBUF_LOCK_ASSERT(sb); 1445 /* 1446 * First, update for the new value of nextrecord. If necessary, make 1447 * it the first record. 1448 */ 1449 if (sb->sb_mb != NULL) 1450 sb->sb_mb->m_nextpkt = nextrecord; 1451 else 1452 sb->sb_mb = nextrecord; 1453 1454 /* 1455 * Now update any dependent socket buffer fields to reflect the new 1456 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1457 * addition of a second clause that takes care of the case where 1458 * sb_mb has been updated, but remains the last record. 1459 */ 1460 if (sb->sb_mb == NULL) { 1461 sb->sb_mbtail = NULL; 1462 sb->sb_lastrecord = NULL; 1463 } else if (sb->sb_mb->m_nextpkt == NULL) 1464 sb->sb_lastrecord = sb->sb_mb; 1465 } 1466 1467 /* 1468 * Implement receive operations on a socket. We depend on the way that 1469 * records are added to the sockbuf by sbappend. In particular, each record 1470 * (mbufs linked through m_next) must begin with an address if the protocol 1471 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1472 * data, and then zero or more mbufs of data. In order to allow parallelism 1473 * between network receive and copying to user space, as well as avoid 1474 * sleeping with a mutex held, we release the socket buffer mutex during the 1475 * user space copy. Although the sockbuf is locked, new data may still be 1476 * appended, and thus we must maintain consistency of the sockbuf during that 1477 * time. 1478 * 1479 * The caller may receive the data as a single mbuf chain by supplying an 1480 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1481 * the count in uio_resid. 1482 */ 1483 int 1484 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1485 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1486 { 1487 struct mbuf *m, **mp; 1488 int flags, error, offset; 1489 ssize_t len; 1490 struct protosw *pr = so->so_proto; 1491 struct mbuf *nextrecord; 1492 int moff, type = 0; 1493 ssize_t orig_resid = uio->uio_resid; 1494 1495 mp = mp0; 1496 if (psa != NULL) 1497 *psa = NULL; 1498 if (controlp != NULL) 1499 *controlp = NULL; 1500 if (flagsp != NULL) 1501 flags = *flagsp &~ MSG_EOR; 1502 else 1503 flags = 0; 1504 if (flags & MSG_OOB) 1505 return (soreceive_rcvoob(so, uio, flags)); 1506 if (mp != NULL) 1507 *mp = NULL; 1508 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 1509 && uio->uio_resid) { 1510 VNET_SO_ASSERT(so); 1511 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 1512 } 1513 1514 error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); 1515 if (error) 1516 return (error); 1517 1518 restart: 1519 SOCKBUF_LOCK(&so->so_rcv); 1520 m = so->so_rcv.sb_mb; 1521 /* 1522 * If we have less data than requested, block awaiting more (subject 1523 * to any timeout) if: 1524 * 1. the current count is less than the low water mark, or 1525 * 2. MSG_DONTWAIT is not set 1526 */ 1527 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1528 sbavail(&so->so_rcv) < uio->uio_resid) && 1529 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && 1530 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 1531 KASSERT(m != NULL || !sbavail(&so->so_rcv), 1532 ("receive: m == %p sbavail == %u", 1533 m, sbavail(&so->so_rcv))); 1534 if (so->so_error) { 1535 if (m != NULL) 1536 goto dontblock; 1537 error = so->so_error; 1538 if ((flags & MSG_PEEK) == 0) 1539 so->so_error = 0; 1540 SOCKBUF_UNLOCK(&so->so_rcv); 1541 goto release; 1542 } 1543 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1544 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1545 if (m == NULL) { 1546 SOCKBUF_UNLOCK(&so->so_rcv); 1547 goto release; 1548 } else 1549 goto dontblock; 1550 } 1551 for (; m != NULL; m = m->m_next) 1552 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1553 m = so->so_rcv.sb_mb; 1554 goto dontblock; 1555 } 1556 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1557 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1558 SOCKBUF_UNLOCK(&so->so_rcv); 1559 error = ENOTCONN; 1560 goto release; 1561 } 1562 if (uio->uio_resid == 0) { 1563 SOCKBUF_UNLOCK(&so->so_rcv); 1564 goto release; 1565 } 1566 if ((so->so_state & SS_NBIO) || 1567 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1568 SOCKBUF_UNLOCK(&so->so_rcv); 1569 error = EWOULDBLOCK; 1570 goto release; 1571 } 1572 SBLASTRECORDCHK(&so->so_rcv); 1573 SBLASTMBUFCHK(&so->so_rcv); 1574 error = sbwait(&so->so_rcv); 1575 SOCKBUF_UNLOCK(&so->so_rcv); 1576 if (error) 1577 goto release; 1578 goto restart; 1579 } 1580 dontblock: 1581 /* 1582 * From this point onward, we maintain 'nextrecord' as a cache of the 1583 * pointer to the next record in the socket buffer. We must keep the 1584 * various socket buffer pointers and local stack versions of the 1585 * pointers in sync, pushing out modifications before dropping the 1586 * socket buffer mutex, and re-reading them when picking it up. 1587 * 1588 * Otherwise, we will race with the network stack appending new data 1589 * or records onto the socket buffer by using inconsistent/stale 1590 * versions of the field, possibly resulting in socket buffer 1591 * corruption. 1592 * 1593 * By holding the high-level sblock(), we prevent simultaneous 1594 * readers from pulling off the front of the socket buffer. 1595 */ 1596 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1597 if (uio->uio_td) 1598 uio->uio_td->td_ru.ru_msgrcv++; 1599 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 1600 SBLASTRECORDCHK(&so->so_rcv); 1601 SBLASTMBUFCHK(&so->so_rcv); 1602 nextrecord = m->m_nextpkt; 1603 if (pr->pr_flags & PR_ADDR) { 1604 KASSERT(m->m_type == MT_SONAME, 1605 ("m->m_type == %d", m->m_type)); 1606 orig_resid = 0; 1607 if (psa != NULL) 1608 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 1609 M_NOWAIT); 1610 if (flags & MSG_PEEK) { 1611 m = m->m_next; 1612 } else { 1613 sbfree(&so->so_rcv, m); 1614 so->so_rcv.sb_mb = m_free(m); 1615 m = so->so_rcv.sb_mb; 1616 sockbuf_pushsync(&so->so_rcv, nextrecord); 1617 } 1618 } 1619 1620 /* 1621 * Process one or more MT_CONTROL mbufs present before any data mbufs 1622 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 1623 * just copy the data; if !MSG_PEEK, we call into the protocol to 1624 * perform externalization (or freeing if controlp == NULL). 1625 */ 1626 if (m != NULL && m->m_type == MT_CONTROL) { 1627 struct mbuf *cm = NULL, *cmn; 1628 struct mbuf **cme = &cm; 1629 1630 do { 1631 if (flags & MSG_PEEK) { 1632 if (controlp != NULL) { 1633 *controlp = m_copy(m, 0, m->m_len); 1634 controlp = &(*controlp)->m_next; 1635 } 1636 m = m->m_next; 1637 } else { 1638 sbfree(&so->so_rcv, m); 1639 so->so_rcv.sb_mb = m->m_next; 1640 m->m_next = NULL; 1641 *cme = m; 1642 cme = &(*cme)->m_next; 1643 m = so->so_rcv.sb_mb; 1644 } 1645 } while (m != NULL && m->m_type == MT_CONTROL); 1646 if ((flags & MSG_PEEK) == 0) 1647 sockbuf_pushsync(&so->so_rcv, nextrecord); 1648 while (cm != NULL) { 1649 cmn = cm->m_next; 1650 cm->m_next = NULL; 1651 if (pr->pr_domain->dom_externalize != NULL) { 1652 SOCKBUF_UNLOCK(&so->so_rcv); 1653 VNET_SO_ASSERT(so); 1654 error = (*pr->pr_domain->dom_externalize) 1655 (cm, controlp, flags); 1656 SOCKBUF_LOCK(&so->so_rcv); 1657 } else if (controlp != NULL) 1658 *controlp = cm; 1659 else 1660 m_freem(cm); 1661 if (controlp != NULL) { 1662 orig_resid = 0; 1663 while (*controlp != NULL) 1664 controlp = &(*controlp)->m_next; 1665 } 1666 cm = cmn; 1667 } 1668 if (m != NULL) 1669 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 1670 else 1671 nextrecord = so->so_rcv.sb_mb; 1672 orig_resid = 0; 1673 } 1674 if (m != NULL) { 1675 if ((flags & MSG_PEEK) == 0) { 1676 KASSERT(m->m_nextpkt == nextrecord, 1677 ("soreceive: post-control, nextrecord !sync")); 1678 if (nextrecord == NULL) { 1679 KASSERT(so->so_rcv.sb_mb == m, 1680 ("soreceive: post-control, sb_mb!=m")); 1681 KASSERT(so->so_rcv.sb_lastrecord == m, 1682 ("soreceive: post-control, lastrecord!=m")); 1683 } 1684 } 1685 type = m->m_type; 1686 if (type == MT_OOBDATA) 1687 flags |= MSG_OOB; 1688 } else { 1689 if ((flags & MSG_PEEK) == 0) { 1690 KASSERT(so->so_rcv.sb_mb == nextrecord, 1691 ("soreceive: sb_mb != nextrecord")); 1692 if (so->so_rcv.sb_mb == NULL) { 1693 KASSERT(so->so_rcv.sb_lastrecord == NULL, 1694 ("soreceive: sb_lastercord != NULL")); 1695 } 1696 } 1697 } 1698 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1699 SBLASTRECORDCHK(&so->so_rcv); 1700 SBLASTMBUFCHK(&so->so_rcv); 1701 1702 /* 1703 * Now continue to read any data mbufs off of the head of the socket 1704 * buffer until the read request is satisfied. Note that 'type' is 1705 * used to store the type of any mbuf reads that have happened so far 1706 * such that soreceive() can stop reading if the type changes, which 1707 * causes soreceive() to return only one of regular data and inline 1708 * out-of-band data in a single socket receive operation. 1709 */ 1710 moff = 0; 1711 offset = 0; 1712 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 1713 && error == 0) { 1714 /* 1715 * If the type of mbuf has changed since the last mbuf 1716 * examined ('type'), end the receive operation. 1717 */ 1718 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1719 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 1720 if (type != m->m_type) 1721 break; 1722 } else if (type == MT_OOBDATA) 1723 break; 1724 else 1725 KASSERT(m->m_type == MT_DATA, 1726 ("m->m_type == %d", m->m_type)); 1727 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 1728 len = uio->uio_resid; 1729 if (so->so_oobmark && len > so->so_oobmark - offset) 1730 len = so->so_oobmark - offset; 1731 if (len > m->m_len - moff) 1732 len = m->m_len - moff; 1733 /* 1734 * If mp is set, just pass back the mbufs. Otherwise copy 1735 * them out via the uio, then free. Sockbuf must be 1736 * consistent here (points to current mbuf, it points to next 1737 * record) when we drop priority; we must note any additions 1738 * to the sockbuf when we block interrupts again. 1739 */ 1740 if (mp == NULL) { 1741 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1742 SBLASTRECORDCHK(&so->so_rcv); 1743 SBLASTMBUFCHK(&so->so_rcv); 1744 SOCKBUF_UNLOCK(&so->so_rcv); 1745 error = uiomove(mtod(m, char *) + moff, (int)len, uio); 1746 SOCKBUF_LOCK(&so->so_rcv); 1747 if (error) { 1748 /* 1749 * The MT_SONAME mbuf has already been removed 1750 * from the record, so it is necessary to 1751 * remove the data mbufs, if any, to preserve 1752 * the invariant in the case of PR_ADDR that 1753 * requires MT_SONAME mbufs at the head of 1754 * each record. 1755 */ 1756 if (m && pr->pr_flags & PR_ATOMIC && 1757 ((flags & MSG_PEEK) == 0)) 1758 (void)sbdroprecord_locked(&so->so_rcv); 1759 SOCKBUF_UNLOCK(&so->so_rcv); 1760 goto release; 1761 } 1762 } else 1763 uio->uio_resid -= len; 1764 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1765 if (len == m->m_len - moff) { 1766 if (m->m_flags & M_EOR) 1767 flags |= MSG_EOR; 1768 if (flags & MSG_PEEK) { 1769 m = m->m_next; 1770 moff = 0; 1771 } else { 1772 nextrecord = m->m_nextpkt; 1773 sbfree(&so->so_rcv, m); 1774 if (mp != NULL) { 1775 m->m_nextpkt = NULL; 1776 *mp = m; 1777 mp = &m->m_next; 1778 so->so_rcv.sb_mb = m = m->m_next; 1779 *mp = NULL; 1780 } else { 1781 so->so_rcv.sb_mb = m_free(m); 1782 m = so->so_rcv.sb_mb; 1783 } 1784 sockbuf_pushsync(&so->so_rcv, nextrecord); 1785 SBLASTRECORDCHK(&so->so_rcv); 1786 SBLASTMBUFCHK(&so->so_rcv); 1787 } 1788 } else { 1789 if (flags & MSG_PEEK) 1790 moff += len; 1791 else { 1792 if (mp != NULL) { 1793 if (flags & MSG_DONTWAIT) { 1794 *mp = m_copym(m, 0, len, 1795 M_NOWAIT); 1796 if (*mp == NULL) { 1797 /* 1798 * m_copym() couldn't 1799 * allocate an mbuf. 1800 * Adjust uio_resid back 1801 * (it was adjusted 1802 * down by len bytes, 1803 * which we didn't end 1804 * up "copying" over). 1805 */ 1806 uio->uio_resid += len; 1807 break; 1808 } 1809 } else { 1810 SOCKBUF_UNLOCK(&so->so_rcv); 1811 *mp = m_copym(m, 0, len, 1812 M_WAITOK); 1813 SOCKBUF_LOCK(&so->so_rcv); 1814 } 1815 } 1816 sbcut_locked(&so->so_rcv, len); 1817 } 1818 } 1819 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1820 if (so->so_oobmark) { 1821 if ((flags & MSG_PEEK) == 0) { 1822 so->so_oobmark -= len; 1823 if (so->so_oobmark == 0) { 1824 so->so_rcv.sb_state |= SBS_RCVATMARK; 1825 break; 1826 } 1827 } else { 1828 offset += len; 1829 if (offset == so->so_oobmark) 1830 break; 1831 } 1832 } 1833 if (flags & MSG_EOR) 1834 break; 1835 /* 1836 * If the MSG_WAITALL flag is set (for non-atomic socket), we 1837 * must not quit until "uio->uio_resid == 0" or an error 1838 * termination. If a signal/timeout occurs, return with a 1839 * short count but without error. Keep sockbuf locked 1840 * against other readers. 1841 */ 1842 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 1843 !sosendallatonce(so) && nextrecord == NULL) { 1844 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1845 if (so->so_error || 1846 so->so_rcv.sb_state & SBS_CANTRCVMORE) 1847 break; 1848 /* 1849 * Notify the protocol that some data has been 1850 * drained before blocking. 1851 */ 1852 if (pr->pr_flags & PR_WANTRCVD) { 1853 SOCKBUF_UNLOCK(&so->so_rcv); 1854 VNET_SO_ASSERT(so); 1855 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 1856 SOCKBUF_LOCK(&so->so_rcv); 1857 } 1858 SBLASTRECORDCHK(&so->so_rcv); 1859 SBLASTMBUFCHK(&so->so_rcv); 1860 /* 1861 * We could receive some data while was notifying 1862 * the protocol. Skip blocking in this case. 1863 */ 1864 if (so->so_rcv.sb_mb == NULL) { 1865 error = sbwait(&so->so_rcv); 1866 if (error) { 1867 SOCKBUF_UNLOCK(&so->so_rcv); 1868 goto release; 1869 } 1870 } 1871 m = so->so_rcv.sb_mb; 1872 if (m != NULL) 1873 nextrecord = m->m_nextpkt; 1874 } 1875 } 1876 1877 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1878 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 1879 flags |= MSG_TRUNC; 1880 if ((flags & MSG_PEEK) == 0) 1881 (void) sbdroprecord_locked(&so->so_rcv); 1882 } 1883 if ((flags & MSG_PEEK) == 0) { 1884 if (m == NULL) { 1885 /* 1886 * First part is an inline SB_EMPTY_FIXUP(). Second 1887 * part makes sure sb_lastrecord is up-to-date if 1888 * there is still data in the socket buffer. 1889 */ 1890 so->so_rcv.sb_mb = nextrecord; 1891 if (so->so_rcv.sb_mb == NULL) { 1892 so->so_rcv.sb_mbtail = NULL; 1893 so->so_rcv.sb_lastrecord = NULL; 1894 } else if (nextrecord->m_nextpkt == NULL) 1895 so->so_rcv.sb_lastrecord = nextrecord; 1896 } 1897 SBLASTRECORDCHK(&so->so_rcv); 1898 SBLASTMBUFCHK(&so->so_rcv); 1899 /* 1900 * If soreceive() is being done from the socket callback, 1901 * then don't need to generate ACK to peer to update window, 1902 * since ACK will be generated on return to TCP. 1903 */ 1904 if (!(flags & MSG_SOCALLBCK) && 1905 (pr->pr_flags & PR_WANTRCVD)) { 1906 SOCKBUF_UNLOCK(&so->so_rcv); 1907 VNET_SO_ASSERT(so); 1908 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 1909 SOCKBUF_LOCK(&so->so_rcv); 1910 } 1911 } 1912 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1913 if (orig_resid == uio->uio_resid && orig_resid && 1914 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 1915 SOCKBUF_UNLOCK(&so->so_rcv); 1916 goto restart; 1917 } 1918 SOCKBUF_UNLOCK(&so->so_rcv); 1919 1920 if (flagsp != NULL) 1921 *flagsp |= flags; 1922 release: 1923 sbunlock(&so->so_rcv); 1924 return (error); 1925 } 1926 1927 /* 1928 * Optimized version of soreceive() for stream (TCP) sockets. 1929 * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled. 1930 */ 1931 int 1932 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 1933 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1934 { 1935 int len = 0, error = 0, flags, oresid; 1936 struct sockbuf *sb; 1937 struct mbuf *m, *n = NULL; 1938 1939 /* We only do stream sockets. */ 1940 if (so->so_type != SOCK_STREAM) 1941 return (EINVAL); 1942 if (psa != NULL) 1943 *psa = NULL; 1944 if (controlp != NULL) 1945 return (EINVAL); 1946 if (flagsp != NULL) 1947 flags = *flagsp &~ MSG_EOR; 1948 else 1949 flags = 0; 1950 if (flags & MSG_OOB) 1951 return (soreceive_rcvoob(so, uio, flags)); 1952 if (mp0 != NULL) 1953 *mp0 = NULL; 1954 1955 sb = &so->so_rcv; 1956 1957 /* Prevent other readers from entering the socket. */ 1958 error = sblock(sb, SBLOCKWAIT(flags)); 1959 if (error) 1960 goto out; 1961 SOCKBUF_LOCK(sb); 1962 1963 /* Easy one, no space to copyout anything. */ 1964 if (uio->uio_resid == 0) { 1965 error = EINVAL; 1966 goto out; 1967 } 1968 oresid = uio->uio_resid; 1969 1970 /* We will never ever get anything unless we are or were connected. */ 1971 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 1972 error = ENOTCONN; 1973 goto out; 1974 } 1975 1976 restart: 1977 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1978 1979 /* Abort if socket has reported problems. */ 1980 if (so->so_error) { 1981 if (sbavail(sb) > 0) 1982 goto deliver; 1983 if (oresid > uio->uio_resid) 1984 goto out; 1985 error = so->so_error; 1986 if (!(flags & MSG_PEEK)) 1987 so->so_error = 0; 1988 goto out; 1989 } 1990 1991 /* Door is closed. Deliver what is left, if any. */ 1992 if (sb->sb_state & SBS_CANTRCVMORE) { 1993 if (sbavail(sb) > 0) 1994 goto deliver; 1995 else 1996 goto out; 1997 } 1998 1999 /* Socket buffer is empty and we shall not block. */ 2000 if (sbavail(sb) == 0 && 2001 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2002 error = EAGAIN; 2003 goto out; 2004 } 2005 2006 /* Socket buffer got some data that we shall deliver now. */ 2007 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 2008 ((so->so_state & SS_NBIO) || 2009 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2010 sbavail(sb) >= sb->sb_lowat || 2011 sbavail(sb) >= uio->uio_resid || 2012 sbavail(sb) >= sb->sb_hiwat) ) { 2013 goto deliver; 2014 } 2015 2016 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2017 if ((flags & MSG_WAITALL) && 2018 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 2019 goto deliver; 2020 2021 /* 2022 * Wait and block until (more) data comes in. 2023 * NB: Drops the sockbuf lock during wait. 2024 */ 2025 error = sbwait(sb); 2026 if (error) 2027 goto out; 2028 goto restart; 2029 2030 deliver: 2031 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2032 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 2033 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2034 2035 /* Statistics. */ 2036 if (uio->uio_td) 2037 uio->uio_td->td_ru.ru_msgrcv++; 2038 2039 /* Fill uio until full or current end of socket buffer is reached. */ 2040 len = min(uio->uio_resid, sbavail(sb)); 2041 if (mp0 != NULL) { 2042 /* Dequeue as many mbufs as possible. */ 2043 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2044 if (*mp0 == NULL) 2045 *mp0 = sb->sb_mb; 2046 else 2047 m_cat(*mp0, sb->sb_mb); 2048 for (m = sb->sb_mb; 2049 m != NULL && m->m_len <= len; 2050 m = m->m_next) { 2051 KASSERT(!(m->m_flags & M_NOTAVAIL), 2052 ("%s: m %p not available", __func__, m)); 2053 len -= m->m_len; 2054 uio->uio_resid -= m->m_len; 2055 sbfree(sb, m); 2056 n = m; 2057 } 2058 n->m_next = NULL; 2059 sb->sb_mb = m; 2060 sb->sb_lastrecord = sb->sb_mb; 2061 if (sb->sb_mb == NULL) 2062 SB_EMPTY_FIXUP(sb); 2063 } 2064 /* Copy the remainder. */ 2065 if (len > 0) { 2066 KASSERT(sb->sb_mb != NULL, 2067 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2068 2069 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2070 if (m == NULL) 2071 len = 0; /* Don't flush data from sockbuf. */ 2072 else 2073 uio->uio_resid -= len; 2074 if (*mp0 != NULL) 2075 m_cat(*mp0, m); 2076 else 2077 *mp0 = m; 2078 if (*mp0 == NULL) { 2079 error = ENOBUFS; 2080 goto out; 2081 } 2082 } 2083 } else { 2084 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2085 SOCKBUF_UNLOCK(sb); 2086 error = m_mbuftouio(uio, sb->sb_mb, len); 2087 SOCKBUF_LOCK(sb); 2088 if (error) 2089 goto out; 2090 } 2091 SBLASTRECORDCHK(sb); 2092 SBLASTMBUFCHK(sb); 2093 2094 /* 2095 * Remove the delivered data from the socket buffer unless we 2096 * were only peeking. 2097 */ 2098 if (!(flags & MSG_PEEK)) { 2099 if (len > 0) 2100 sbdrop_locked(sb, len); 2101 2102 /* Notify protocol that we drained some data. */ 2103 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2104 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2105 !(flags & MSG_SOCALLBCK))) { 2106 SOCKBUF_UNLOCK(sb); 2107 VNET_SO_ASSERT(so); 2108 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags); 2109 SOCKBUF_LOCK(sb); 2110 } 2111 } 2112 2113 /* 2114 * For MSG_WAITALL we may have to loop again and wait for 2115 * more data to come in. 2116 */ 2117 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2118 goto restart; 2119 out: 2120 SOCKBUF_LOCK_ASSERT(sb); 2121 SBLASTRECORDCHK(sb); 2122 SBLASTMBUFCHK(sb); 2123 SOCKBUF_UNLOCK(sb); 2124 sbunlock(sb); 2125 return (error); 2126 } 2127 2128 /* 2129 * Optimized version of soreceive() for simple datagram cases from userspace. 2130 * Unlike in the stream case, we're able to drop a datagram if copyout() 2131 * fails, and because we handle datagrams atomically, we don't need to use a 2132 * sleep lock to prevent I/O interlacing. 2133 */ 2134 int 2135 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2136 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2137 { 2138 struct mbuf *m, *m2; 2139 int flags, error; 2140 ssize_t len; 2141 struct protosw *pr = so->so_proto; 2142 struct mbuf *nextrecord; 2143 2144 if (psa != NULL) 2145 *psa = NULL; 2146 if (controlp != NULL) 2147 *controlp = NULL; 2148 if (flagsp != NULL) 2149 flags = *flagsp &~ MSG_EOR; 2150 else 2151 flags = 0; 2152 2153 /* 2154 * For any complicated cases, fall back to the full 2155 * soreceive_generic(). 2156 */ 2157 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB)) 2158 return (soreceive_generic(so, psa, uio, mp0, controlp, 2159 flagsp)); 2160 2161 /* 2162 * Enforce restrictions on use. 2163 */ 2164 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2165 ("soreceive_dgram: wantrcvd")); 2166 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2167 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2168 ("soreceive_dgram: SBS_RCVATMARK")); 2169 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2170 ("soreceive_dgram: P_CONNREQUIRED")); 2171 2172 /* 2173 * Loop blocking while waiting for a datagram. 2174 */ 2175 SOCKBUF_LOCK(&so->so_rcv); 2176 while ((m = so->so_rcv.sb_mb) == NULL) { 2177 KASSERT(sbavail(&so->so_rcv) == 0, 2178 ("soreceive_dgram: sb_mb NULL but sbavail %u", 2179 sbavail(&so->so_rcv))); 2180 if (so->so_error) { 2181 error = so->so_error; 2182 so->so_error = 0; 2183 SOCKBUF_UNLOCK(&so->so_rcv); 2184 return (error); 2185 } 2186 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2187 uio->uio_resid == 0) { 2188 SOCKBUF_UNLOCK(&so->so_rcv); 2189 return (0); 2190 } 2191 if ((so->so_state & SS_NBIO) || 2192 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2193 SOCKBUF_UNLOCK(&so->so_rcv); 2194 return (EWOULDBLOCK); 2195 } 2196 SBLASTRECORDCHK(&so->so_rcv); 2197 SBLASTMBUFCHK(&so->so_rcv); 2198 error = sbwait(&so->so_rcv); 2199 if (error) { 2200 SOCKBUF_UNLOCK(&so->so_rcv); 2201 return (error); 2202 } 2203 } 2204 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2205 2206 if (uio->uio_td) 2207 uio->uio_td->td_ru.ru_msgrcv++; 2208 SBLASTRECORDCHK(&so->so_rcv); 2209 SBLASTMBUFCHK(&so->so_rcv); 2210 nextrecord = m->m_nextpkt; 2211 if (nextrecord == NULL) { 2212 KASSERT(so->so_rcv.sb_lastrecord == m, 2213 ("soreceive_dgram: lastrecord != m")); 2214 } 2215 2216 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2217 ("soreceive_dgram: m_nextpkt != nextrecord")); 2218 2219 /* 2220 * Pull 'm' and its chain off the front of the packet queue. 2221 */ 2222 so->so_rcv.sb_mb = NULL; 2223 sockbuf_pushsync(&so->so_rcv, nextrecord); 2224 2225 /* 2226 * Walk 'm's chain and free that many bytes from the socket buffer. 2227 */ 2228 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2229 sbfree(&so->so_rcv, m2); 2230 2231 /* 2232 * Do a few last checks before we let go of the lock. 2233 */ 2234 SBLASTRECORDCHK(&so->so_rcv); 2235 SBLASTMBUFCHK(&so->so_rcv); 2236 SOCKBUF_UNLOCK(&so->so_rcv); 2237 2238 if (pr->pr_flags & PR_ADDR) { 2239 KASSERT(m->m_type == MT_SONAME, 2240 ("m->m_type == %d", m->m_type)); 2241 if (psa != NULL) 2242 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2243 M_NOWAIT); 2244 m = m_free(m); 2245 } 2246 if (m == NULL) { 2247 /* XXXRW: Can this happen? */ 2248 return (0); 2249 } 2250 2251 /* 2252 * Packet to copyout() is now in 'm' and it is disconnected from the 2253 * queue. 2254 * 2255 * Process one or more MT_CONTROL mbufs present before any data mbufs 2256 * in the first mbuf chain on the socket buffer. We call into the 2257 * protocol to perform externalization (or freeing if controlp == 2258 * NULL). In some cases there can be only MT_CONTROL mbufs without 2259 * MT_DATA mbufs. 2260 */ 2261 if (m->m_type == MT_CONTROL) { 2262 struct mbuf *cm = NULL, *cmn; 2263 struct mbuf **cme = &cm; 2264 2265 do { 2266 m2 = m->m_next; 2267 m->m_next = NULL; 2268 *cme = m; 2269 cme = &(*cme)->m_next; 2270 m = m2; 2271 } while (m != NULL && m->m_type == MT_CONTROL); 2272 while (cm != NULL) { 2273 cmn = cm->m_next; 2274 cm->m_next = NULL; 2275 if (pr->pr_domain->dom_externalize != NULL) { 2276 error = (*pr->pr_domain->dom_externalize) 2277 (cm, controlp, flags); 2278 } else if (controlp != NULL) 2279 *controlp = cm; 2280 else 2281 m_freem(cm); 2282 if (controlp != NULL) { 2283 while (*controlp != NULL) 2284 controlp = &(*controlp)->m_next; 2285 } 2286 cm = cmn; 2287 } 2288 } 2289 KASSERT(m == NULL || m->m_type == MT_DATA, 2290 ("soreceive_dgram: !data")); 2291 while (m != NULL && uio->uio_resid > 0) { 2292 len = uio->uio_resid; 2293 if (len > m->m_len) 2294 len = m->m_len; 2295 error = uiomove(mtod(m, char *), (int)len, uio); 2296 if (error) { 2297 m_freem(m); 2298 return (error); 2299 } 2300 if (len == m->m_len) 2301 m = m_free(m); 2302 else { 2303 m->m_data += len; 2304 m->m_len -= len; 2305 } 2306 } 2307 if (m != NULL) { 2308 flags |= MSG_TRUNC; 2309 m_freem(m); 2310 } 2311 if (flagsp != NULL) 2312 *flagsp |= flags; 2313 return (0); 2314 } 2315 2316 int 2317 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2318 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2319 { 2320 int error; 2321 2322 CURVNET_SET(so->so_vnet); 2323 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0, 2324 controlp, flagsp)); 2325 CURVNET_RESTORE(); 2326 return (error); 2327 } 2328 2329 int 2330 soshutdown(struct socket *so, int how) 2331 { 2332 struct protosw *pr = so->so_proto; 2333 int error; 2334 2335 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2336 return (EINVAL); 2337 2338 CURVNET_SET(so->so_vnet); 2339 if (pr->pr_usrreqs->pru_flush != NULL) 2340 (*pr->pr_usrreqs->pru_flush)(so, how); 2341 if (how != SHUT_WR) 2342 sorflush(so); 2343 if (how != SHUT_RD) { 2344 error = (*pr->pr_usrreqs->pru_shutdown)(so); 2345 wakeup(&so->so_timeo); 2346 CURVNET_RESTORE(); 2347 return (error); 2348 } 2349 wakeup(&so->so_timeo); 2350 CURVNET_RESTORE(); 2351 return (0); 2352 } 2353 2354 void 2355 sorflush(struct socket *so) 2356 { 2357 struct sockbuf *sb = &so->so_rcv; 2358 struct protosw *pr = so->so_proto; 2359 struct sockbuf asb; 2360 2361 VNET_SO_ASSERT(so); 2362 2363 /* 2364 * In order to avoid calling dom_dispose with the socket buffer mutex 2365 * held, and in order to generally avoid holding the lock for a long 2366 * time, we make a copy of the socket buffer and clear the original 2367 * (except locks, state). The new socket buffer copy won't have 2368 * initialized locks so we can only call routines that won't use or 2369 * assert those locks. 2370 * 2371 * Dislodge threads currently blocked in receive and wait to acquire 2372 * a lock against other simultaneous readers before clearing the 2373 * socket buffer. Don't let our acquire be interrupted by a signal 2374 * despite any existing socket disposition on interruptable waiting. 2375 */ 2376 socantrcvmore(so); 2377 (void) sblock(sb, SBL_WAIT | SBL_NOINTR); 2378 2379 /* 2380 * Invalidate/clear most of the sockbuf structure, but leave selinfo 2381 * and mutex data unchanged. 2382 */ 2383 SOCKBUF_LOCK(sb); 2384 bzero(&asb, offsetof(struct sockbuf, sb_startzero)); 2385 bcopy(&sb->sb_startzero, &asb.sb_startzero, 2386 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2387 bzero(&sb->sb_startzero, 2388 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2389 SOCKBUF_UNLOCK(sb); 2390 sbunlock(sb); 2391 2392 /* 2393 * Dispose of special rights and flush the socket buffer. Don't call 2394 * any unsafe routines (that rely on locks being initialized) on asb. 2395 */ 2396 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 2397 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 2398 sbrelease_internal(&asb, so); 2399 } 2400 2401 /* 2402 * Wrapper for Socket established helper hook. 2403 * Parameters: socket, context of the hook point, hook id. 2404 */ 2405 static int inline 2406 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 2407 { 2408 struct socket_hhook_data hhook_data = { 2409 .so = so, 2410 .hctx = hctx, 2411 .m = NULL, 2412 .status = 0 2413 }; 2414 2415 CURVNET_SET(so->so_vnet); 2416 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 2417 CURVNET_RESTORE(); 2418 2419 /* Ugly but needed, since hhooks return void for now */ 2420 return (hhook_data.status); 2421 } 2422 2423 /* 2424 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 2425 * additional variant to handle the case where the option value needs to be 2426 * some kind of integer, but not a specific size. In addition to their use 2427 * here, these functions are also called by the protocol-level pr_ctloutput() 2428 * routines. 2429 */ 2430 int 2431 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2432 { 2433 size_t valsize; 2434 2435 /* 2436 * If the user gives us more than we wanted, we ignore it, but if we 2437 * don't get the minimum length the caller wants, we return EINVAL. 2438 * On success, sopt->sopt_valsize is set to however much we actually 2439 * retrieved. 2440 */ 2441 if ((valsize = sopt->sopt_valsize) < minlen) 2442 return EINVAL; 2443 if (valsize > len) 2444 sopt->sopt_valsize = valsize = len; 2445 2446 if (sopt->sopt_td != NULL) 2447 return (copyin(sopt->sopt_val, buf, valsize)); 2448 2449 bcopy(sopt->sopt_val, buf, valsize); 2450 return (0); 2451 } 2452 2453 /* 2454 * Kernel version of setsockopt(2). 2455 * 2456 * XXX: optlen is size_t, not socklen_t 2457 */ 2458 int 2459 so_setsockopt(struct socket *so, int level, int optname, void *optval, 2460 size_t optlen) 2461 { 2462 struct sockopt sopt; 2463 2464 sopt.sopt_level = level; 2465 sopt.sopt_name = optname; 2466 sopt.sopt_dir = SOPT_SET; 2467 sopt.sopt_val = optval; 2468 sopt.sopt_valsize = optlen; 2469 sopt.sopt_td = NULL; 2470 return (sosetopt(so, &sopt)); 2471 } 2472 2473 int 2474 sosetopt(struct socket *so, struct sockopt *sopt) 2475 { 2476 int error, optval; 2477 struct linger l; 2478 struct timeval tv; 2479 sbintime_t val; 2480 uint32_t val32; 2481 #ifdef MAC 2482 struct mac extmac; 2483 #endif 2484 2485 CURVNET_SET(so->so_vnet); 2486 error = 0; 2487 if (sopt->sopt_level != SOL_SOCKET) { 2488 if (so->so_proto->pr_ctloutput != NULL) { 2489 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2490 CURVNET_RESTORE(); 2491 return (error); 2492 } 2493 error = ENOPROTOOPT; 2494 } else { 2495 switch (sopt->sopt_name) { 2496 case SO_ACCEPTFILTER: 2497 error = do_setopt_accept_filter(so, sopt); 2498 if (error) 2499 goto bad; 2500 break; 2501 2502 case SO_LINGER: 2503 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 2504 if (error) 2505 goto bad; 2506 2507 SOCK_LOCK(so); 2508 so->so_linger = l.l_linger; 2509 if (l.l_onoff) 2510 so->so_options |= SO_LINGER; 2511 else 2512 so->so_options &= ~SO_LINGER; 2513 SOCK_UNLOCK(so); 2514 break; 2515 2516 case SO_DEBUG: 2517 case SO_KEEPALIVE: 2518 case SO_DONTROUTE: 2519 case SO_USELOOPBACK: 2520 case SO_BROADCAST: 2521 case SO_REUSEADDR: 2522 case SO_REUSEPORT: 2523 case SO_OOBINLINE: 2524 case SO_TIMESTAMP: 2525 case SO_BINTIME: 2526 case SO_NOSIGPIPE: 2527 case SO_NO_DDP: 2528 case SO_NO_OFFLOAD: 2529 error = sooptcopyin(sopt, &optval, sizeof optval, 2530 sizeof optval); 2531 if (error) 2532 goto bad; 2533 SOCK_LOCK(so); 2534 if (optval) 2535 so->so_options |= sopt->sopt_name; 2536 else 2537 so->so_options &= ~sopt->sopt_name; 2538 SOCK_UNLOCK(so); 2539 break; 2540 2541 case SO_SETFIB: 2542 error = sooptcopyin(sopt, &optval, sizeof optval, 2543 sizeof optval); 2544 if (error) 2545 goto bad; 2546 2547 if (optval < 0 || optval >= rt_numfibs) { 2548 error = EINVAL; 2549 goto bad; 2550 } 2551 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 2552 (so->so_proto->pr_domain->dom_family == PF_INET6) || 2553 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 2554 so->so_fibnum = optval; 2555 else 2556 so->so_fibnum = 0; 2557 break; 2558 2559 case SO_USER_COOKIE: 2560 error = sooptcopyin(sopt, &val32, sizeof val32, 2561 sizeof val32); 2562 if (error) 2563 goto bad; 2564 so->so_user_cookie = val32; 2565 break; 2566 2567 case SO_SNDBUF: 2568 case SO_RCVBUF: 2569 case SO_SNDLOWAT: 2570 case SO_RCVLOWAT: 2571 error = sooptcopyin(sopt, &optval, sizeof optval, 2572 sizeof optval); 2573 if (error) 2574 goto bad; 2575 2576 /* 2577 * Values < 1 make no sense for any of these options, 2578 * so disallow them. 2579 */ 2580 if (optval < 1) { 2581 error = EINVAL; 2582 goto bad; 2583 } 2584 2585 switch (sopt->sopt_name) { 2586 case SO_SNDBUF: 2587 case SO_RCVBUF: 2588 if (sbreserve(sopt->sopt_name == SO_SNDBUF ? 2589 &so->so_snd : &so->so_rcv, (u_long)optval, 2590 so, curthread) == 0) { 2591 error = ENOBUFS; 2592 goto bad; 2593 } 2594 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd : 2595 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE; 2596 break; 2597 2598 /* 2599 * Make sure the low-water is never greater than the 2600 * high-water. 2601 */ 2602 case SO_SNDLOWAT: 2603 SOCKBUF_LOCK(&so->so_snd); 2604 so->so_snd.sb_lowat = 2605 (optval > so->so_snd.sb_hiwat) ? 2606 so->so_snd.sb_hiwat : optval; 2607 SOCKBUF_UNLOCK(&so->so_snd); 2608 break; 2609 case SO_RCVLOWAT: 2610 SOCKBUF_LOCK(&so->so_rcv); 2611 so->so_rcv.sb_lowat = 2612 (optval > so->so_rcv.sb_hiwat) ? 2613 so->so_rcv.sb_hiwat : optval; 2614 SOCKBUF_UNLOCK(&so->so_rcv); 2615 break; 2616 } 2617 break; 2618 2619 case SO_SNDTIMEO: 2620 case SO_RCVTIMEO: 2621 #ifdef COMPAT_FREEBSD32 2622 if (SV_CURPROC_FLAG(SV_ILP32)) { 2623 struct timeval32 tv32; 2624 2625 error = sooptcopyin(sopt, &tv32, sizeof tv32, 2626 sizeof tv32); 2627 CP(tv32, tv, tv_sec); 2628 CP(tv32, tv, tv_usec); 2629 } else 2630 #endif 2631 error = sooptcopyin(sopt, &tv, sizeof tv, 2632 sizeof tv); 2633 if (error) 2634 goto bad; 2635 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 2636 tv.tv_usec >= 1000000) { 2637 error = EDOM; 2638 goto bad; 2639 } 2640 if (tv.tv_sec > INT32_MAX) 2641 val = SBT_MAX; 2642 else 2643 val = tvtosbt(tv); 2644 switch (sopt->sopt_name) { 2645 case SO_SNDTIMEO: 2646 so->so_snd.sb_timeo = val; 2647 break; 2648 case SO_RCVTIMEO: 2649 so->so_rcv.sb_timeo = val; 2650 break; 2651 } 2652 break; 2653 2654 case SO_LABEL: 2655 #ifdef MAC 2656 error = sooptcopyin(sopt, &extmac, sizeof extmac, 2657 sizeof extmac); 2658 if (error) 2659 goto bad; 2660 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 2661 so, &extmac); 2662 #else 2663 error = EOPNOTSUPP; 2664 #endif 2665 break; 2666 2667 default: 2668 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 2669 error = hhook_run_socket(so, sopt, 2670 HHOOK_SOCKET_OPT); 2671 else 2672 error = ENOPROTOOPT; 2673 break; 2674 } 2675 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 2676 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 2677 } 2678 bad: 2679 CURVNET_RESTORE(); 2680 return (error); 2681 } 2682 2683 /* 2684 * Helper routine for getsockopt. 2685 */ 2686 int 2687 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 2688 { 2689 int error; 2690 size_t valsize; 2691 2692 error = 0; 2693 2694 /* 2695 * Documented get behavior is that we always return a value, possibly 2696 * truncated to fit in the user's buffer. Traditional behavior is 2697 * that we always tell the user precisely how much we copied, rather 2698 * than something useful like the total amount we had available for 2699 * her. Note that this interface is not idempotent; the entire 2700 * answer must generated ahead of time. 2701 */ 2702 valsize = min(len, sopt->sopt_valsize); 2703 sopt->sopt_valsize = valsize; 2704 if (sopt->sopt_val != NULL) { 2705 if (sopt->sopt_td != NULL) 2706 error = copyout(buf, sopt->sopt_val, valsize); 2707 else 2708 bcopy(buf, sopt->sopt_val, valsize); 2709 } 2710 return (error); 2711 } 2712 2713 int 2714 sogetopt(struct socket *so, struct sockopt *sopt) 2715 { 2716 int error, optval; 2717 struct linger l; 2718 struct timeval tv; 2719 #ifdef MAC 2720 struct mac extmac; 2721 #endif 2722 2723 CURVNET_SET(so->so_vnet); 2724 error = 0; 2725 if (sopt->sopt_level != SOL_SOCKET) { 2726 if (so->so_proto->pr_ctloutput != NULL) 2727 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2728 else 2729 error = ENOPROTOOPT; 2730 CURVNET_RESTORE(); 2731 return (error); 2732 } else { 2733 switch (sopt->sopt_name) { 2734 case SO_ACCEPTFILTER: 2735 error = do_getopt_accept_filter(so, sopt); 2736 break; 2737 2738 case SO_LINGER: 2739 SOCK_LOCK(so); 2740 l.l_onoff = so->so_options & SO_LINGER; 2741 l.l_linger = so->so_linger; 2742 SOCK_UNLOCK(so); 2743 error = sooptcopyout(sopt, &l, sizeof l); 2744 break; 2745 2746 case SO_USELOOPBACK: 2747 case SO_DONTROUTE: 2748 case SO_DEBUG: 2749 case SO_KEEPALIVE: 2750 case SO_REUSEADDR: 2751 case SO_REUSEPORT: 2752 case SO_BROADCAST: 2753 case SO_OOBINLINE: 2754 case SO_ACCEPTCONN: 2755 case SO_TIMESTAMP: 2756 case SO_BINTIME: 2757 case SO_NOSIGPIPE: 2758 optval = so->so_options & sopt->sopt_name; 2759 integer: 2760 error = sooptcopyout(sopt, &optval, sizeof optval); 2761 break; 2762 2763 case SO_TYPE: 2764 optval = so->so_type; 2765 goto integer; 2766 2767 case SO_PROTOCOL: 2768 optval = so->so_proto->pr_protocol; 2769 goto integer; 2770 2771 case SO_ERROR: 2772 SOCK_LOCK(so); 2773 optval = so->so_error; 2774 so->so_error = 0; 2775 SOCK_UNLOCK(so); 2776 goto integer; 2777 2778 case SO_SNDBUF: 2779 optval = so->so_snd.sb_hiwat; 2780 goto integer; 2781 2782 case SO_RCVBUF: 2783 optval = so->so_rcv.sb_hiwat; 2784 goto integer; 2785 2786 case SO_SNDLOWAT: 2787 optval = so->so_snd.sb_lowat; 2788 goto integer; 2789 2790 case SO_RCVLOWAT: 2791 optval = so->so_rcv.sb_lowat; 2792 goto integer; 2793 2794 case SO_SNDTIMEO: 2795 case SO_RCVTIMEO: 2796 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 2797 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 2798 #ifdef COMPAT_FREEBSD32 2799 if (SV_CURPROC_FLAG(SV_ILP32)) { 2800 struct timeval32 tv32; 2801 2802 CP(tv, tv32, tv_sec); 2803 CP(tv, tv32, tv_usec); 2804 error = sooptcopyout(sopt, &tv32, sizeof tv32); 2805 } else 2806 #endif 2807 error = sooptcopyout(sopt, &tv, sizeof tv); 2808 break; 2809 2810 case SO_LABEL: 2811 #ifdef MAC 2812 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 2813 sizeof(extmac)); 2814 if (error) 2815 goto bad; 2816 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 2817 so, &extmac); 2818 if (error) 2819 goto bad; 2820 error = sooptcopyout(sopt, &extmac, sizeof extmac); 2821 #else 2822 error = EOPNOTSUPP; 2823 #endif 2824 break; 2825 2826 case SO_PEERLABEL: 2827 #ifdef MAC 2828 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 2829 sizeof(extmac)); 2830 if (error) 2831 goto bad; 2832 error = mac_getsockopt_peerlabel( 2833 sopt->sopt_td->td_ucred, so, &extmac); 2834 if (error) 2835 goto bad; 2836 error = sooptcopyout(sopt, &extmac, sizeof extmac); 2837 #else 2838 error = EOPNOTSUPP; 2839 #endif 2840 break; 2841 2842 case SO_LISTENQLIMIT: 2843 optval = so->so_qlimit; 2844 goto integer; 2845 2846 case SO_LISTENQLEN: 2847 optval = so->so_qlen; 2848 goto integer; 2849 2850 case SO_LISTENINCQLEN: 2851 optval = so->so_incqlen; 2852 goto integer; 2853 2854 default: 2855 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 2856 error = hhook_run_socket(so, sopt, 2857 HHOOK_SOCKET_OPT); 2858 else 2859 error = ENOPROTOOPT; 2860 break; 2861 } 2862 } 2863 #ifdef MAC 2864 bad: 2865 #endif 2866 CURVNET_RESTORE(); 2867 return (error); 2868 } 2869 2870 int 2871 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 2872 { 2873 struct mbuf *m, *m_prev; 2874 int sopt_size = sopt->sopt_valsize; 2875 2876 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 2877 if (m == NULL) 2878 return ENOBUFS; 2879 if (sopt_size > MLEN) { 2880 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 2881 if ((m->m_flags & M_EXT) == 0) { 2882 m_free(m); 2883 return ENOBUFS; 2884 } 2885 m->m_len = min(MCLBYTES, sopt_size); 2886 } else { 2887 m->m_len = min(MLEN, sopt_size); 2888 } 2889 sopt_size -= m->m_len; 2890 *mp = m; 2891 m_prev = m; 2892 2893 while (sopt_size) { 2894 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 2895 if (m == NULL) { 2896 m_freem(*mp); 2897 return ENOBUFS; 2898 } 2899 if (sopt_size > MLEN) { 2900 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 2901 M_NOWAIT); 2902 if ((m->m_flags & M_EXT) == 0) { 2903 m_freem(m); 2904 m_freem(*mp); 2905 return ENOBUFS; 2906 } 2907 m->m_len = min(MCLBYTES, sopt_size); 2908 } else { 2909 m->m_len = min(MLEN, sopt_size); 2910 } 2911 sopt_size -= m->m_len; 2912 m_prev->m_next = m; 2913 m_prev = m; 2914 } 2915 return (0); 2916 } 2917 2918 int 2919 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 2920 { 2921 struct mbuf *m0 = m; 2922 2923 if (sopt->sopt_val == NULL) 2924 return (0); 2925 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 2926 if (sopt->sopt_td != NULL) { 2927 int error; 2928 2929 error = copyin(sopt->sopt_val, mtod(m, char *), 2930 m->m_len); 2931 if (error != 0) { 2932 m_freem(m0); 2933 return(error); 2934 } 2935 } else 2936 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 2937 sopt->sopt_valsize -= m->m_len; 2938 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 2939 m = m->m_next; 2940 } 2941 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 2942 panic("ip6_sooptmcopyin"); 2943 return (0); 2944 } 2945 2946 int 2947 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 2948 { 2949 struct mbuf *m0 = m; 2950 size_t valsize = 0; 2951 2952 if (sopt->sopt_val == NULL) 2953 return (0); 2954 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 2955 if (sopt->sopt_td != NULL) { 2956 int error; 2957 2958 error = copyout(mtod(m, char *), sopt->sopt_val, 2959 m->m_len); 2960 if (error != 0) { 2961 m_freem(m0); 2962 return(error); 2963 } 2964 } else 2965 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 2966 sopt->sopt_valsize -= m->m_len; 2967 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 2968 valsize += m->m_len; 2969 m = m->m_next; 2970 } 2971 if (m != NULL) { 2972 /* enough soopt buffer should be given from user-land */ 2973 m_freem(m0); 2974 return(EINVAL); 2975 } 2976 sopt->sopt_valsize = valsize; 2977 return (0); 2978 } 2979 2980 /* 2981 * sohasoutofband(): protocol notifies socket layer of the arrival of new 2982 * out-of-band data, which will then notify socket consumers. 2983 */ 2984 void 2985 sohasoutofband(struct socket *so) 2986 { 2987 2988 if (so->so_sigio != NULL) 2989 pgsigio(&so->so_sigio, SIGURG, 0); 2990 selwakeuppri(&so->so_rcv.sb_sel, PSOCK); 2991 } 2992 2993 int 2994 sopoll(struct socket *so, int events, struct ucred *active_cred, 2995 struct thread *td) 2996 { 2997 2998 /* 2999 * We do not need to set or assert curvnet as long as everyone uses 3000 * sopoll_generic(). 3001 */ 3002 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, 3003 td)); 3004 } 3005 3006 int 3007 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3008 struct thread *td) 3009 { 3010 int revents = 0; 3011 3012 SOCKBUF_LOCK(&so->so_snd); 3013 SOCKBUF_LOCK(&so->so_rcv); 3014 if (events & (POLLIN | POLLRDNORM)) 3015 if (soreadabledata(so)) 3016 revents |= events & (POLLIN | POLLRDNORM); 3017 3018 if (events & (POLLOUT | POLLWRNORM)) 3019 if (sowriteable(so)) 3020 revents |= events & (POLLOUT | POLLWRNORM); 3021 3022 if (events & (POLLPRI | POLLRDBAND)) 3023 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK)) 3024 revents |= events & (POLLPRI | POLLRDBAND); 3025 3026 if ((events & POLLINIGNEOF) == 0) { 3027 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3028 revents |= events & (POLLIN | POLLRDNORM); 3029 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3030 revents |= POLLHUP; 3031 } 3032 } 3033 3034 if (revents == 0) { 3035 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 3036 selrecord(td, &so->so_rcv.sb_sel); 3037 so->so_rcv.sb_flags |= SB_SEL; 3038 } 3039 3040 if (events & (POLLOUT | POLLWRNORM)) { 3041 selrecord(td, &so->so_snd.sb_sel); 3042 so->so_snd.sb_flags |= SB_SEL; 3043 } 3044 } 3045 3046 SOCKBUF_UNLOCK(&so->so_rcv); 3047 SOCKBUF_UNLOCK(&so->so_snd); 3048 return (revents); 3049 } 3050 3051 int 3052 soo_kqfilter(struct file *fp, struct knote *kn) 3053 { 3054 struct socket *so = kn->kn_fp->f_data; 3055 struct sockbuf *sb; 3056 3057 switch (kn->kn_filter) { 3058 case EVFILT_READ: 3059 if (so->so_options & SO_ACCEPTCONN) 3060 kn->kn_fop = &solisten_filtops; 3061 else 3062 kn->kn_fop = &soread_filtops; 3063 sb = &so->so_rcv; 3064 break; 3065 case EVFILT_WRITE: 3066 kn->kn_fop = &sowrite_filtops; 3067 sb = &so->so_snd; 3068 break; 3069 default: 3070 return (EINVAL); 3071 } 3072 3073 SOCKBUF_LOCK(sb); 3074 knlist_add(&sb->sb_sel.si_note, kn, 1); 3075 sb->sb_flags |= SB_KNOTE; 3076 SOCKBUF_UNLOCK(sb); 3077 return (0); 3078 } 3079 3080 /* 3081 * Some routines that return EOPNOTSUPP for entry points that are not 3082 * supported by a protocol. Fill in as needed. 3083 */ 3084 int 3085 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 3086 { 3087 3088 return EOPNOTSUPP; 3089 } 3090 3091 int 3092 pru_attach_notsupp(struct socket *so, int proto, struct thread *td) 3093 { 3094 3095 return EOPNOTSUPP; 3096 } 3097 3098 int 3099 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3100 { 3101 3102 return EOPNOTSUPP; 3103 } 3104 3105 int 3106 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3107 struct thread *td) 3108 { 3109 3110 return EOPNOTSUPP; 3111 } 3112 3113 int 3114 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3115 { 3116 3117 return EOPNOTSUPP; 3118 } 3119 3120 int 3121 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3122 struct thread *td) 3123 { 3124 3125 return EOPNOTSUPP; 3126 } 3127 3128 int 3129 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 3130 { 3131 3132 return EOPNOTSUPP; 3133 } 3134 3135 int 3136 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 3137 struct ifnet *ifp, struct thread *td) 3138 { 3139 3140 return EOPNOTSUPP; 3141 } 3142 3143 int 3144 pru_disconnect_notsupp(struct socket *so) 3145 { 3146 3147 return EOPNOTSUPP; 3148 } 3149 3150 int 3151 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) 3152 { 3153 3154 return EOPNOTSUPP; 3155 } 3156 3157 int 3158 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) 3159 { 3160 3161 return EOPNOTSUPP; 3162 } 3163 3164 int 3165 pru_rcvd_notsupp(struct socket *so, int flags) 3166 { 3167 3168 return EOPNOTSUPP; 3169 } 3170 3171 int 3172 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 3173 { 3174 3175 return EOPNOTSUPP; 3176 } 3177 3178 int 3179 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, 3180 struct sockaddr *addr, struct mbuf *control, struct thread *td) 3181 { 3182 3183 return EOPNOTSUPP; 3184 } 3185 3186 int 3187 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count) 3188 { 3189 3190 return (EOPNOTSUPP); 3191 } 3192 3193 /* 3194 * This isn't really a ``null'' operation, but it's the default one and 3195 * doesn't do anything destructive. 3196 */ 3197 int 3198 pru_sense_null(struct socket *so, struct stat *sb) 3199 { 3200 3201 sb->st_blksize = so->so_snd.sb_hiwat; 3202 return 0; 3203 } 3204 3205 int 3206 pru_shutdown_notsupp(struct socket *so) 3207 { 3208 3209 return EOPNOTSUPP; 3210 } 3211 3212 int 3213 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) 3214 { 3215 3216 return EOPNOTSUPP; 3217 } 3218 3219 int 3220 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 3221 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 3222 { 3223 3224 return EOPNOTSUPP; 3225 } 3226 3227 int 3228 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 3229 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 3230 { 3231 3232 return EOPNOTSUPP; 3233 } 3234 3235 int 3236 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, 3237 struct thread *td) 3238 { 3239 3240 return EOPNOTSUPP; 3241 } 3242 3243 static void 3244 filt_sordetach(struct knote *kn) 3245 { 3246 struct socket *so = kn->kn_fp->f_data; 3247 3248 SOCKBUF_LOCK(&so->so_rcv); 3249 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1); 3250 if (knlist_empty(&so->so_rcv.sb_sel.si_note)) 3251 so->so_rcv.sb_flags &= ~SB_KNOTE; 3252 SOCKBUF_UNLOCK(&so->so_rcv); 3253 } 3254 3255 /*ARGSUSED*/ 3256 static int 3257 filt_soread(struct knote *kn, long hint) 3258 { 3259 struct socket *so; 3260 3261 so = kn->kn_fp->f_data; 3262 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3263 3264 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 3265 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3266 kn->kn_flags |= EV_EOF; 3267 kn->kn_fflags = so->so_error; 3268 return (1); 3269 } else if (so->so_error) /* temporary udp error */ 3270 return (1); 3271 3272 if (kn->kn_sfflags & NOTE_LOWAT) { 3273 if (kn->kn_data >= kn->kn_sdata) 3274 return 1; 3275 } else { 3276 if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) 3277 return 1; 3278 } 3279 3280 /* This hook returning non-zero indicates an event, not error */ 3281 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); 3282 } 3283 3284 static void 3285 filt_sowdetach(struct knote *kn) 3286 { 3287 struct socket *so = kn->kn_fp->f_data; 3288 3289 SOCKBUF_LOCK(&so->so_snd); 3290 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1); 3291 if (knlist_empty(&so->so_snd.sb_sel.si_note)) 3292 so->so_snd.sb_flags &= ~SB_KNOTE; 3293 SOCKBUF_UNLOCK(&so->so_snd); 3294 } 3295 3296 /*ARGSUSED*/ 3297 static int 3298 filt_sowrite(struct knote *kn, long hint) 3299 { 3300 struct socket *so; 3301 3302 so = kn->kn_fp->f_data; 3303 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3304 kn->kn_data = sbspace(&so->so_snd); 3305 3306 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); 3307 3308 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3309 kn->kn_flags |= EV_EOF; 3310 kn->kn_fflags = so->so_error; 3311 return (1); 3312 } else if (so->so_error) /* temporary udp error */ 3313 return (1); 3314 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3315 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3316 return (0); 3317 else if (kn->kn_sfflags & NOTE_LOWAT) 3318 return (kn->kn_data >= kn->kn_sdata); 3319 else 3320 return (kn->kn_data >= so->so_snd.sb_lowat); 3321 } 3322 3323 /*ARGSUSED*/ 3324 static int 3325 filt_solisten(struct knote *kn, long hint) 3326 { 3327 struct socket *so = kn->kn_fp->f_data; 3328 3329 kn->kn_data = so->so_qlen; 3330 return (!TAILQ_EMPTY(&so->so_comp)); 3331 } 3332 3333 int 3334 socheckuid(struct socket *so, uid_t uid) 3335 { 3336 3337 if (so == NULL) 3338 return (EPERM); 3339 if (so->so_cred->cr_uid != uid) 3340 return (EPERM); 3341 return (0); 3342 } 3343 3344 /* 3345 * These functions are used by protocols to notify the socket layer (and its 3346 * consumers) of state changes in the sockets driven by protocol-side events. 3347 */ 3348 3349 /* 3350 * Procedures to manipulate state flags of socket and do appropriate wakeups. 3351 * 3352 * Normal sequence from the active (originating) side is that 3353 * soisconnecting() is called during processing of connect() call, resulting 3354 * in an eventual call to soisconnected() if/when the connection is 3355 * established. When the connection is torn down soisdisconnecting() is 3356 * called during processing of disconnect() call, and soisdisconnected() is 3357 * called when the connection to the peer is totally severed. The semantics 3358 * of these routines are such that connectionless protocols can call 3359 * soisconnected() and soisdisconnected() only, bypassing the in-progress 3360 * calls when setting up a ``connection'' takes no time. 3361 * 3362 * From the passive side, a socket is created with two queues of sockets: 3363 * so_incomp for connections in progress and so_comp for connections already 3364 * made and awaiting user acceptance. As a protocol is preparing incoming 3365 * connections, it creates a socket structure queued on so_incomp by calling 3366 * sonewconn(). When the connection is established, soisconnected() is 3367 * called, and transfers the socket structure to so_comp, making it available 3368 * to accept(). 3369 * 3370 * If a socket is closed with sockets on either so_incomp or so_comp, these 3371 * sockets are dropped. 3372 * 3373 * If higher-level protocols are implemented in the kernel, the wakeups done 3374 * here will sometimes cause software-interrupt process scheduling. 3375 */ 3376 void 3377 soisconnecting(struct socket *so) 3378 { 3379 3380 SOCK_LOCK(so); 3381 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 3382 so->so_state |= SS_ISCONNECTING; 3383 SOCK_UNLOCK(so); 3384 } 3385 3386 void 3387 soisconnected(struct socket *so) 3388 { 3389 struct socket *head; 3390 int ret; 3391 3392 restart: 3393 ACCEPT_LOCK(); 3394 SOCK_LOCK(so); 3395 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 3396 so->so_state |= SS_ISCONNECTED; 3397 head = so->so_head; 3398 if (head != NULL && (so->so_qstate & SQ_INCOMP)) { 3399 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 3400 SOCK_UNLOCK(so); 3401 TAILQ_REMOVE(&head->so_incomp, so, so_list); 3402 head->so_incqlen--; 3403 so->so_qstate &= ~SQ_INCOMP; 3404 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 3405 head->so_qlen++; 3406 so->so_qstate |= SQ_COMP; 3407 ACCEPT_UNLOCK(); 3408 sorwakeup(head); 3409 wakeup_one(&head->so_timeo); 3410 } else { 3411 ACCEPT_UNLOCK(); 3412 soupcall_set(so, SO_RCV, 3413 head->so_accf->so_accept_filter->accf_callback, 3414 head->so_accf->so_accept_filter_arg); 3415 so->so_options &= ~SO_ACCEPTFILTER; 3416 ret = head->so_accf->so_accept_filter->accf_callback(so, 3417 head->so_accf->so_accept_filter_arg, M_NOWAIT); 3418 if (ret == SU_ISCONNECTED) 3419 soupcall_clear(so, SO_RCV); 3420 SOCK_UNLOCK(so); 3421 if (ret == SU_ISCONNECTED) 3422 goto restart; 3423 } 3424 return; 3425 } 3426 SOCK_UNLOCK(so); 3427 ACCEPT_UNLOCK(); 3428 wakeup(&so->so_timeo); 3429 sorwakeup(so); 3430 sowwakeup(so); 3431 } 3432 3433 void 3434 soisdisconnecting(struct socket *so) 3435 { 3436 3437 /* 3438 * Note: This code assumes that SOCK_LOCK(so) and 3439 * SOCKBUF_LOCK(&so->so_rcv) are the same. 3440 */ 3441 SOCKBUF_LOCK(&so->so_rcv); 3442 so->so_state &= ~SS_ISCONNECTING; 3443 so->so_state |= SS_ISDISCONNECTING; 3444 socantrcvmore_locked(so); 3445 SOCKBUF_LOCK(&so->so_snd); 3446 socantsendmore_locked(so); 3447 wakeup(&so->so_timeo); 3448 } 3449 3450 void 3451 soisdisconnected(struct socket *so) 3452 { 3453 3454 /* 3455 * Note: This code assumes that SOCK_LOCK(so) and 3456 * SOCKBUF_LOCK(&so->so_rcv) are the same. 3457 */ 3458 SOCKBUF_LOCK(&so->so_rcv); 3459 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 3460 so->so_state |= SS_ISDISCONNECTED; 3461 socantrcvmore_locked(so); 3462 SOCKBUF_LOCK(&so->so_snd); 3463 sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); 3464 socantsendmore_locked(so); 3465 wakeup(&so->so_timeo); 3466 } 3467 3468 /* 3469 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 3470 */ 3471 struct sockaddr * 3472 sodupsockaddr(const struct sockaddr *sa, int mflags) 3473 { 3474 struct sockaddr *sa2; 3475 3476 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 3477 if (sa2) 3478 bcopy(sa, sa2, sa->sa_len); 3479 return sa2; 3480 } 3481 3482 /* 3483 * Register per-socket buffer upcalls. 3484 */ 3485 void 3486 soupcall_set(struct socket *so, int which, 3487 int (*func)(struct socket *, void *, int), void *arg) 3488 { 3489 struct sockbuf *sb; 3490 3491 switch (which) { 3492 case SO_RCV: 3493 sb = &so->so_rcv; 3494 break; 3495 case SO_SND: 3496 sb = &so->so_snd; 3497 break; 3498 default: 3499 panic("soupcall_set: bad which"); 3500 } 3501 SOCKBUF_LOCK_ASSERT(sb); 3502 #if 0 3503 /* XXX: accf_http actually wants to do this on purpose. */ 3504 KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall")); 3505 #endif 3506 sb->sb_upcall = func; 3507 sb->sb_upcallarg = arg; 3508 sb->sb_flags |= SB_UPCALL; 3509 } 3510 3511 void 3512 soupcall_clear(struct socket *so, int which) 3513 { 3514 struct sockbuf *sb; 3515 3516 switch (which) { 3517 case SO_RCV: 3518 sb = &so->so_rcv; 3519 break; 3520 case SO_SND: 3521 sb = &so->so_snd; 3522 break; 3523 default: 3524 panic("soupcall_clear: bad which"); 3525 } 3526 SOCKBUF_LOCK_ASSERT(sb); 3527 KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear")); 3528 sb->sb_upcall = NULL; 3529 sb->sb_upcallarg = NULL; 3530 sb->sb_flags &= ~SB_UPCALL; 3531 } 3532 3533 /* 3534 * Create an external-format (``xsocket'') structure using the information in 3535 * the kernel-format socket structure pointed to by so. This is done to 3536 * reduce the spew of irrelevant information over this interface, to isolate 3537 * user code from changes in the kernel structure, and potentially to provide 3538 * information-hiding if we decide that some of this information should be 3539 * hidden from users. 3540 */ 3541 void 3542 sotoxsocket(struct socket *so, struct xsocket *xso) 3543 { 3544 3545 xso->xso_len = sizeof *xso; 3546 xso->xso_so = so; 3547 xso->so_type = so->so_type; 3548 xso->so_options = so->so_options; 3549 xso->so_linger = so->so_linger; 3550 xso->so_state = so->so_state; 3551 xso->so_pcb = so->so_pcb; 3552 xso->xso_protocol = so->so_proto->pr_protocol; 3553 xso->xso_family = so->so_proto->pr_domain->dom_family; 3554 xso->so_qlen = so->so_qlen; 3555 xso->so_incqlen = so->so_incqlen; 3556 xso->so_qlimit = so->so_qlimit; 3557 xso->so_timeo = so->so_timeo; 3558 xso->so_error = so->so_error; 3559 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 3560 xso->so_oobmark = so->so_oobmark; 3561 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 3562 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 3563 xso->so_uid = so->so_cred->cr_uid; 3564 } 3565 3566 3567 /* 3568 * Socket accessor functions to provide external consumers with 3569 * a safe interface to socket state 3570 * 3571 */ 3572 3573 void 3574 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), 3575 void *arg) 3576 { 3577 3578 TAILQ_FOREACH(so, &so->so_comp, so_list) 3579 func(so, arg); 3580 } 3581 3582 struct sockbuf * 3583 so_sockbuf_rcv(struct socket *so) 3584 { 3585 3586 return (&so->so_rcv); 3587 } 3588 3589 struct sockbuf * 3590 so_sockbuf_snd(struct socket *so) 3591 { 3592 3593 return (&so->so_snd); 3594 } 3595 3596 int 3597 so_state_get(const struct socket *so) 3598 { 3599 3600 return (so->so_state); 3601 } 3602 3603 void 3604 so_state_set(struct socket *so, int val) 3605 { 3606 3607 so->so_state = val; 3608 } 3609 3610 int 3611 so_options_get(const struct socket *so) 3612 { 3613 3614 return (so->so_options); 3615 } 3616 3617 void 3618 so_options_set(struct socket *so, int val) 3619 { 3620 3621 so->so_options = val; 3622 } 3623 3624 int 3625 so_error_get(const struct socket *so) 3626 { 3627 3628 return (so->so_error); 3629 } 3630 3631 void 3632 so_error_set(struct socket *so, int val) 3633 { 3634 3635 so->so_error = val; 3636 } 3637 3638 int 3639 so_linger_get(const struct socket *so) 3640 { 3641 3642 return (so->so_linger); 3643 } 3644 3645 void 3646 so_linger_set(struct socket *so, int val) 3647 { 3648 3649 so->so_linger = val; 3650 } 3651 3652 struct protosw * 3653 so_protosw_get(const struct socket *so) 3654 { 3655 3656 return (so->so_proto); 3657 } 3658 3659 void 3660 so_protosw_set(struct socket *so, struct protosw *val) 3661 { 3662 3663 so->so_proto = val; 3664 } 3665 3666 void 3667 so_sorwakeup(struct socket *so) 3668 { 3669 3670 sorwakeup(so); 3671 } 3672 3673 void 3674 so_sowwakeup(struct socket *so) 3675 { 3676 3677 sowwakeup(so); 3678 } 3679 3680 void 3681 so_sorwakeup_locked(struct socket *so) 3682 { 3683 3684 sorwakeup_locked(so); 3685 } 3686 3687 void 3688 so_sowwakeup_locked(struct socket *so) 3689 { 3690 3691 sowwakeup_locked(so); 3692 } 3693 3694 void 3695 so_lock(struct socket *so) 3696 { 3697 3698 SOCK_LOCK(so); 3699 } 3700 3701 void 3702 so_unlock(struct socket *so) 3703 { 3704 3705 SOCK_UNLOCK(so); 3706 } 3707