1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993 5 * The Regents of the University of California. 6 * Copyright (c) 2004 The FreeBSD Foundation 7 * Copyright (c) 2004-2008 Robert N. M. Watson 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 35 */ 36 37 /* 38 * Comments on the socket life cycle: 39 * 40 * soalloc() sets of socket layer state for a socket, called only by 41 * socreate() and sonewconn(). Socket layer private. 42 * 43 * sodealloc() tears down socket layer state for a socket, called only by 44 * sofree() and sonewconn(). Socket layer private. 45 * 46 * pru_attach() associates protocol layer state with an allocated socket; 47 * called only once, may fail, aborting socket allocation. This is called 48 * from socreate() and sonewconn(). Socket layer private. 49 * 50 * pru_detach() disassociates protocol layer state from an attached socket, 51 * and will be called exactly once for sockets in which pru_attach() has 52 * been successfully called. If pru_attach() returned an error, 53 * pru_detach() will not be called. Socket layer private. 54 * 55 * pru_abort() and pru_close() notify the protocol layer that the last 56 * consumer of a socket is starting to tear down the socket, and that the 57 * protocol should terminate the connection. Historically, pru_abort() also 58 * detached protocol state from the socket state, but this is no longer the 59 * case. 60 * 61 * socreate() creates a socket and attaches protocol state. This is a public 62 * interface that may be used by socket layer consumers to create new 63 * sockets. 64 * 65 * sonewconn() creates a socket and attaches protocol state. This is a 66 * public interface that may be used by protocols to create new sockets when 67 * a new connection is received and will be available for accept() on a 68 * listen socket. 69 * 70 * soclose() destroys a socket after possibly waiting for it to disconnect. 71 * This is a public interface that socket consumers should use to close and 72 * release a socket when done with it. 73 * 74 * soabort() destroys a socket without waiting for it to disconnect (used 75 * only for incoming connections that are already partially or fully 76 * connected). This is used internally by the socket layer when clearing 77 * listen socket queues (due to overflow or close on the listen socket), but 78 * is also a public interface protocols may use to abort connections in 79 * their incomplete listen queues should they no longer be required. Sockets 80 * placed in completed connection listen queues should not be aborted for 81 * reasons described in the comment above the soclose() implementation. This 82 * is not a general purpose close routine, and except in the specific 83 * circumstances described here, should not be used. 84 * 85 * sofree() will free a socket and its protocol state if all references on 86 * the socket have been released, and is the public interface to attempt to 87 * free a socket when a reference is removed. This is a socket layer private 88 * interface. 89 * 90 * NOTE: In addition to socreate() and soclose(), which provide a single 91 * socket reference to the consumer to be managed as required, there are two 92 * calls to explicitly manage socket references, soref(), and sorele(). 93 * Currently, these are generally required only when transitioning a socket 94 * from a listen queue to a file descriptor, in order to prevent garbage 95 * collection of the socket at an untimely moment. For a number of reasons, 96 * these interfaces are not preferred, and should be avoided. 97 * 98 * NOTE: With regard to VNETs the general rule is that callers do not set 99 * curvnet. Exceptions to this rule include soabort(), sodisconnect(), 100 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn() 101 * and sorflush(), which are usually called from a pre-set VNET context. 102 * sopoll() currently does not need a VNET context to be set. 103 */ 104 105 #include <sys/cdefs.h> 106 __FBSDID("$FreeBSD$"); 107 108 #include "opt_inet.h" 109 #include "opt_inet6.h" 110 #include "opt_kern_tls.h" 111 #include "opt_sctp.h" 112 113 #include <sys/param.h> 114 #include <sys/systm.h> 115 #include <sys/capsicum.h> 116 #include <sys/fcntl.h> 117 #include <sys/limits.h> 118 #include <sys/lock.h> 119 #include <sys/mac.h> 120 #include <sys/malloc.h> 121 #include <sys/mbuf.h> 122 #include <sys/mutex.h> 123 #include <sys/domain.h> 124 #include <sys/file.h> /* for struct knote */ 125 #include <sys/hhook.h> 126 #include <sys/kernel.h> 127 #include <sys/khelp.h> 128 #include <sys/ktls.h> 129 #include <sys/event.h> 130 #include <sys/eventhandler.h> 131 #include <sys/poll.h> 132 #include <sys/proc.h> 133 #include <sys/protosw.h> 134 #include <sys/sbuf.h> 135 #include <sys/socket.h> 136 #include <sys/socketvar.h> 137 #include <sys/resourcevar.h> 138 #include <net/route.h> 139 #include <sys/signalvar.h> 140 #include <sys/stat.h> 141 #include <sys/sx.h> 142 #include <sys/sysctl.h> 143 #include <sys/taskqueue.h> 144 #include <sys/uio.h> 145 #include <sys/un.h> 146 #include <sys/unpcb.h> 147 #include <sys/jail.h> 148 #include <sys/syslog.h> 149 #include <netinet/in.h> 150 #include <netinet/in_pcb.h> 151 #include <netinet/tcp.h> 152 153 #include <net/vnet.h> 154 155 #include <security/mac/mac_framework.h> 156 157 #include <vm/uma.h> 158 159 #ifdef COMPAT_FREEBSD32 160 #include <sys/mount.h> 161 #include <sys/sysent.h> 162 #include <compat/freebsd32/freebsd32.h> 163 #endif 164 165 static int soreceive_rcvoob(struct socket *so, struct uio *uio, 166 int flags); 167 static void so_rdknl_lock(void *); 168 static void so_rdknl_unlock(void *); 169 static void so_rdknl_assert_lock(void *, int); 170 static void so_wrknl_lock(void *); 171 static void so_wrknl_unlock(void *); 172 static void so_wrknl_assert_lock(void *, int); 173 174 static void filt_sordetach(struct knote *kn); 175 static int filt_soread(struct knote *kn, long hint); 176 static void filt_sowdetach(struct knote *kn); 177 static int filt_sowrite(struct knote *kn, long hint); 178 static int filt_soempty(struct knote *kn, long hint); 179 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id); 180 fo_kqfilter_t soo_kqfilter; 181 182 static struct filterops soread_filtops = { 183 .f_isfd = 1, 184 .f_detach = filt_sordetach, 185 .f_event = filt_soread, 186 }; 187 static struct filterops sowrite_filtops = { 188 .f_isfd = 1, 189 .f_detach = filt_sowdetach, 190 .f_event = filt_sowrite, 191 }; 192 static struct filterops soempty_filtops = { 193 .f_isfd = 1, 194 .f_detach = filt_sowdetach, 195 .f_event = filt_soempty, 196 }; 197 198 so_gen_t so_gencnt; /* generation count for sockets */ 199 200 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 201 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 202 203 #define VNET_SO_ASSERT(so) \ 204 VNET_ASSERT(curvnet != NULL, \ 205 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so))); 206 207 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]); 208 #define V_socket_hhh VNET(socket_hhh) 209 210 /* 211 * Limit on the number of connections in the listen queue waiting 212 * for accept(2). 213 * NB: The original sysctl somaxconn is still available but hidden 214 * to prevent confusion about the actual purpose of this number. 215 */ 216 static u_int somaxconn = SOMAXCONN; 217 218 static int 219 sysctl_somaxconn(SYSCTL_HANDLER_ARGS) 220 { 221 int error; 222 int val; 223 224 val = somaxconn; 225 error = sysctl_handle_int(oidp, &val, 0, req); 226 if (error || !req->newptr ) 227 return (error); 228 229 /* 230 * The purpose of the UINT_MAX / 3 limit, is so that the formula 231 * 3 * so_qlimit / 2 232 * below, will not overflow. 233 */ 234 235 if (val < 1 || val > UINT_MAX / 3) 236 return (EINVAL); 237 238 somaxconn = val; 239 return (0); 240 } 241 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, 242 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int), 243 sysctl_somaxconn, "I", 244 "Maximum listen socket pending connection accept queue size"); 245 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, 246 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, 0, 247 sizeof(int), sysctl_somaxconn, "I", 248 "Maximum listen socket pending connection accept queue size (compat)"); 249 250 static int numopensockets; 251 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD, 252 &numopensockets, 0, "Number of open sockets"); 253 254 /* 255 * accept_mtx locks down per-socket fields relating to accept queues. See 256 * socketvar.h for an annotation of the protected fields of struct socket. 257 */ 258 struct mtx accept_mtx; 259 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF); 260 261 /* 262 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket 263 * so_gencnt field. 264 */ 265 static struct mtx so_global_mtx; 266 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF); 267 268 /* 269 * General IPC sysctl name space, used by sockets and a variety of other IPC 270 * types. 271 */ 272 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 273 "IPC"); 274 275 /* 276 * Initialize the socket subsystem and set up the socket 277 * memory allocator. 278 */ 279 static uma_zone_t socket_zone; 280 int maxsockets; 281 282 static void 283 socket_zone_change(void *tag) 284 { 285 286 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 287 } 288 289 static void 290 socket_hhook_register(int subtype) 291 { 292 293 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype, 294 &V_socket_hhh[subtype], 295 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 296 printf("%s: WARNING: unable to register hook\n", __func__); 297 } 298 299 static void 300 socket_hhook_deregister(int subtype) 301 { 302 303 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0) 304 printf("%s: WARNING: unable to deregister hook\n", __func__); 305 } 306 307 static void 308 socket_init(void *tag) 309 { 310 311 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL, 312 NULL, NULL, UMA_ALIGN_PTR, 0); 313 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 314 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached"); 315 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL, 316 EVENTHANDLER_PRI_FIRST); 317 } 318 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL); 319 320 static void 321 socket_vnet_init(const void *unused __unused) 322 { 323 int i; 324 325 /* We expect a contiguous range */ 326 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 327 socket_hhook_register(i); 328 } 329 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 330 socket_vnet_init, NULL); 331 332 static void 333 socket_vnet_uninit(const void *unused __unused) 334 { 335 int i; 336 337 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 338 socket_hhook_deregister(i); 339 } 340 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 341 socket_vnet_uninit, NULL); 342 343 /* 344 * Initialise maxsockets. This SYSINIT must be run after 345 * tunable_mbinit(). 346 */ 347 static void 348 init_maxsockets(void *ignored) 349 { 350 351 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 352 maxsockets = imax(maxsockets, maxfiles); 353 } 354 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); 355 356 /* 357 * Sysctl to get and set the maximum global sockets limit. Notify protocols 358 * of the change so that they can update their dependent limits as required. 359 */ 360 static int 361 sysctl_maxsockets(SYSCTL_HANDLER_ARGS) 362 { 363 int error, newmaxsockets; 364 365 newmaxsockets = maxsockets; 366 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req); 367 if (error == 0 && req->newptr && newmaxsockets != maxsockets) { 368 if (newmaxsockets > maxsockets && 369 newmaxsockets <= maxfiles) { 370 maxsockets = newmaxsockets; 371 EVENTHANDLER_INVOKE(maxsockets_change); 372 } else 373 error = EINVAL; 374 } 375 return (error); 376 } 377 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, 378 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &maxsockets, 0, 379 sysctl_maxsockets, "IU", 380 "Maximum number of sockets available"); 381 382 /* 383 * Socket operation routines. These routines are called by the routines in 384 * sys_socket.c or from a system process, and implement the semantics of 385 * socket operations by switching out to the protocol specific routines. 386 */ 387 388 /* 389 * Get a socket structure from our zone, and initialize it. Note that it 390 * would probably be better to allocate socket and PCB at the same time, but 391 * I'm not convinced that all the protocols can be easily modified to do 392 * this. 393 * 394 * soalloc() returns a socket with a ref count of 0. 395 */ 396 static struct socket * 397 soalloc(struct vnet *vnet) 398 { 399 struct socket *so; 400 401 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO); 402 if (so == NULL) 403 return (NULL); 404 #ifdef MAC 405 if (mac_socket_init(so, M_NOWAIT) != 0) { 406 uma_zfree(socket_zone, so); 407 return (NULL); 408 } 409 #endif 410 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) { 411 uma_zfree(socket_zone, so); 412 return (NULL); 413 } 414 415 /* 416 * The socket locking protocol allows to lock 2 sockets at a time, 417 * however, the first one must be a listening socket. WITNESS lacks 418 * a feature to change class of an existing lock, so we use DUPOK. 419 */ 420 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK); 421 so->so_snd.sb_mtx = &so->so_snd_mtx; 422 so->so_rcv.sb_mtx = &so->so_rcv_mtx; 423 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd"); 424 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv"); 425 so->so_rcv.sb_sel = &so->so_rdsel; 426 so->so_snd.sb_sel = &so->so_wrsel; 427 sx_init(&so->so_snd_sx, "so_snd_sx"); 428 sx_init(&so->so_rcv_sx, "so_rcv_sx"); 429 TAILQ_INIT(&so->so_snd.sb_aiojobq); 430 TAILQ_INIT(&so->so_rcv.sb_aiojobq); 431 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so); 432 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so); 433 #ifdef VIMAGE 434 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p", 435 __func__, __LINE__, so)); 436 so->so_vnet = vnet; 437 #endif 438 /* We shouldn't need the so_global_mtx */ 439 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) { 440 /* Do we need more comprehensive error returns? */ 441 uma_zfree(socket_zone, so); 442 return (NULL); 443 } 444 mtx_lock(&so_global_mtx); 445 so->so_gencnt = ++so_gencnt; 446 ++numopensockets; 447 #ifdef VIMAGE 448 vnet->vnet_sockcnt++; 449 #endif 450 mtx_unlock(&so_global_mtx); 451 452 return (so); 453 } 454 455 /* 456 * Free the storage associated with a socket at the socket layer, tear down 457 * locks, labels, etc. All protocol state is assumed already to have been 458 * torn down (and possibly never set up) by the caller. 459 */ 460 static void 461 sodealloc(struct socket *so) 462 { 463 464 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); 465 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL")); 466 467 mtx_lock(&so_global_mtx); 468 so->so_gencnt = ++so_gencnt; 469 --numopensockets; /* Could be below, but faster here. */ 470 #ifdef VIMAGE 471 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p", 472 __func__, __LINE__, so)); 473 so->so_vnet->vnet_sockcnt--; 474 #endif 475 mtx_unlock(&so_global_mtx); 476 #ifdef MAC 477 mac_socket_destroy(so); 478 #endif 479 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE); 480 481 khelp_destroy_osd(&so->osd); 482 if (SOLISTENING(so)) { 483 if (so->sol_accept_filter != NULL) 484 accept_filt_setopt(so, NULL); 485 } else { 486 if (so->so_rcv.sb_hiwat) 487 (void)chgsbsize(so->so_cred->cr_uidinfo, 488 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 489 if (so->so_snd.sb_hiwat) 490 (void)chgsbsize(so->so_cred->cr_uidinfo, 491 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 492 sx_destroy(&so->so_snd_sx); 493 sx_destroy(&so->so_rcv_sx); 494 SOCKBUF_LOCK_DESTROY(&so->so_snd); 495 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 496 } 497 crfree(so->so_cred); 498 mtx_destroy(&so->so_lock); 499 uma_zfree(socket_zone, so); 500 } 501 502 /* 503 * socreate returns a socket with a ref count of 1. The socket should be 504 * closed with soclose(). 505 */ 506 int 507 socreate(int dom, struct socket **aso, int type, int proto, 508 struct ucred *cred, struct thread *td) 509 { 510 struct protosw *prp; 511 struct socket *so; 512 int error; 513 514 if (proto) 515 prp = pffindproto(dom, proto, type); 516 else 517 prp = pffindtype(dom, type); 518 519 if (prp == NULL) { 520 /* No support for domain. */ 521 if (pffinddomain(dom) == NULL) 522 return (EAFNOSUPPORT); 523 /* No support for socket type. */ 524 if (proto == 0 && type != 0) 525 return (EPROTOTYPE); 526 return (EPROTONOSUPPORT); 527 } 528 if (prp->pr_usrreqs->pru_attach == NULL || 529 prp->pr_usrreqs->pru_attach == pru_attach_notsupp) 530 return (EPROTONOSUPPORT); 531 532 if (IN_CAPABILITY_MODE(td) && (prp->pr_flags & PR_CAPATTACH) == 0) 533 return (ECAPMODE); 534 535 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0) 536 return (EPROTONOSUPPORT); 537 538 if (prp->pr_type != type) 539 return (EPROTOTYPE); 540 so = soalloc(CRED_TO_VNET(cred)); 541 if (so == NULL) 542 return (ENOBUFS); 543 544 so->so_type = type; 545 so->so_cred = crhold(cred); 546 if ((prp->pr_domain->dom_family == PF_INET) || 547 (prp->pr_domain->dom_family == PF_INET6) || 548 (prp->pr_domain->dom_family == PF_ROUTE)) 549 so->so_fibnum = td->td_proc->p_fibnum; 550 else 551 so->so_fibnum = 0; 552 so->so_proto = prp; 553 #ifdef MAC 554 mac_socket_create(cred, so); 555 #endif 556 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 557 so_rdknl_assert_lock); 558 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 559 so_wrknl_assert_lock); 560 /* 561 * Auto-sizing of socket buffers is managed by the protocols and 562 * the appropriate flags must be set in the pru_attach function. 563 */ 564 CURVNET_SET(so->so_vnet); 565 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); 566 CURVNET_RESTORE(); 567 if (error) { 568 sodealloc(so); 569 return (error); 570 } 571 soref(so); 572 *aso = so; 573 return (0); 574 } 575 576 #ifdef REGRESSION 577 static int regression_sonewconn_earlytest = 1; 578 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW, 579 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test"); 580 #endif 581 582 static struct timeval overinterval = { 60, 0 }; 583 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW, 584 &overinterval, 585 "Delay in seconds between warnings for listen socket overflows"); 586 587 /* 588 * When an attempt at a new connection is noted on a socket which accepts 589 * connections, sonewconn is called. If the connection is possible (subject 590 * to space constraints, etc.) then we allocate a new structure, properly 591 * linked into the data structure of the original socket, and return this. 592 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED. 593 * 594 * Note: the ref count on the socket is 0 on return. 595 */ 596 struct socket * 597 sonewconn(struct socket *head, int connstatus) 598 { 599 struct sbuf descrsb; 600 struct socket *so; 601 int len, overcount; 602 u_int qlen; 603 const char localprefix[] = "local:"; 604 char descrbuf[SUNPATHLEN + sizeof(localprefix)]; 605 #if defined(INET6) 606 char addrbuf[INET6_ADDRSTRLEN]; 607 #elif defined(INET) 608 char addrbuf[INET_ADDRSTRLEN]; 609 #endif 610 bool dolog, over; 611 612 SOLISTEN_LOCK(head); 613 over = (head->sol_qlen > 3 * head->sol_qlimit / 2); 614 #ifdef REGRESSION 615 if (regression_sonewconn_earlytest && over) { 616 #else 617 if (over) { 618 #endif 619 head->sol_overcount++; 620 dolog = !!ratecheck(&head->sol_lastover, &overinterval); 621 622 /* 623 * If we're going to log, copy the overflow count and queue 624 * length from the listen socket before dropping the lock. 625 * Also, reset the overflow count. 626 */ 627 if (dolog) { 628 overcount = head->sol_overcount; 629 head->sol_overcount = 0; 630 qlen = head->sol_qlen; 631 } 632 SOLISTEN_UNLOCK(head); 633 634 if (dolog) { 635 /* 636 * Try to print something descriptive about the 637 * socket for the error message. 638 */ 639 sbuf_new(&descrsb, descrbuf, sizeof(descrbuf), 640 SBUF_FIXEDLEN); 641 switch (head->so_proto->pr_domain->dom_family) { 642 #if defined(INET) || defined(INET6) 643 #ifdef INET 644 case AF_INET: 645 #endif 646 #ifdef INET6 647 case AF_INET6: 648 if (head->so_proto->pr_domain->dom_family == 649 AF_INET6 || 650 (sotoinpcb(head)->inp_inc.inc_flags & 651 INC_ISIPV6)) { 652 ip6_sprintf(addrbuf, 653 &sotoinpcb(head)->inp_inc.inc6_laddr); 654 sbuf_printf(&descrsb, "[%s]", addrbuf); 655 } else 656 #endif 657 { 658 #ifdef INET 659 inet_ntoa_r( 660 sotoinpcb(head)->inp_inc.inc_laddr, 661 addrbuf); 662 sbuf_cat(&descrsb, addrbuf); 663 #endif 664 } 665 sbuf_printf(&descrsb, ":%hu (proto %u)", 666 ntohs(sotoinpcb(head)->inp_inc.inc_lport), 667 head->so_proto->pr_protocol); 668 break; 669 #endif /* INET || INET6 */ 670 case AF_UNIX: 671 sbuf_cat(&descrsb, localprefix); 672 if (sotounpcb(head)->unp_addr != NULL) 673 len = 674 sotounpcb(head)->unp_addr->sun_len - 675 offsetof(struct sockaddr_un, 676 sun_path); 677 else 678 len = 0; 679 if (len > 0) 680 sbuf_bcat(&descrsb, 681 sotounpcb(head)->unp_addr->sun_path, 682 len); 683 else 684 sbuf_cat(&descrsb, "(unknown)"); 685 break; 686 } 687 688 /* 689 * If we can't print something more specific, at least 690 * print the domain name. 691 */ 692 if (sbuf_finish(&descrsb) != 0 || 693 sbuf_len(&descrsb) <= 0) { 694 sbuf_clear(&descrsb); 695 sbuf_cat(&descrsb, 696 head->so_proto->pr_domain->dom_name ?: 697 "unknown"); 698 sbuf_finish(&descrsb); 699 } 700 KASSERT(sbuf_len(&descrsb) > 0, 701 ("%s: sbuf creation failed", __func__)); 702 log(LOG_DEBUG, 703 "%s: pcb %p (%s): Listen queue overflow: " 704 "%i already in queue awaiting acceptance " 705 "(%d occurrences)\n", 706 __func__, head->so_pcb, sbuf_data(&descrsb), 707 qlen, overcount); 708 sbuf_delete(&descrsb); 709 710 overcount = 0; 711 } 712 713 return (NULL); 714 } 715 SOLISTEN_UNLOCK(head); 716 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL", 717 __func__, head)); 718 so = soalloc(head->so_vnet); 719 if (so == NULL) { 720 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 721 "limit reached or out of memory\n", 722 __func__, head->so_pcb); 723 return (NULL); 724 } 725 so->so_listen = head; 726 so->so_type = head->so_type; 727 so->so_options = head->so_options & ~SO_ACCEPTCONN; 728 so->so_linger = head->so_linger; 729 so->so_state = head->so_state | SS_NOFDREF; 730 so->so_fibnum = head->so_fibnum; 731 so->so_proto = head->so_proto; 732 so->so_cred = crhold(head->so_cred); 733 #ifdef MAC 734 mac_socket_newconn(head, so); 735 #endif 736 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 737 so_rdknl_assert_lock); 738 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 739 so_wrknl_assert_lock); 740 VNET_SO_ASSERT(head); 741 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) { 742 sodealloc(so); 743 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 744 __func__, head->so_pcb); 745 return (NULL); 746 } 747 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 748 sodealloc(so); 749 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 750 __func__, head->so_pcb); 751 return (NULL); 752 } 753 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat; 754 so->so_snd.sb_lowat = head->sol_sbsnd_lowat; 755 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo; 756 so->so_snd.sb_timeo = head->sol_sbsnd_timeo; 757 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE; 758 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE; 759 760 SOLISTEN_LOCK(head); 761 if (head->sol_accept_filter != NULL) 762 connstatus = 0; 763 so->so_state |= connstatus; 764 soref(head); /* A socket on (in)complete queue refs head. */ 765 if (connstatus) { 766 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 767 so->so_qstate = SQ_COMP; 768 head->sol_qlen++; 769 solisten_wakeup(head); /* unlocks */ 770 } else { 771 /* 772 * Keep removing sockets from the head until there's room for 773 * us to insert on the tail. In pre-locking revisions, this 774 * was a simple if(), but as we could be racing with other 775 * threads and soabort() requires dropping locks, we must 776 * loop waiting for the condition to be true. 777 */ 778 while (head->sol_incqlen > head->sol_qlimit) { 779 struct socket *sp; 780 781 sp = TAILQ_FIRST(&head->sol_incomp); 782 TAILQ_REMOVE(&head->sol_incomp, sp, so_list); 783 head->sol_incqlen--; 784 SOCK_LOCK(sp); 785 sp->so_qstate = SQ_NONE; 786 sp->so_listen = NULL; 787 SOCK_UNLOCK(sp); 788 sorele_locked(head); /* does SOLISTEN_UNLOCK, head stays */ 789 soabort(sp); 790 SOLISTEN_LOCK(head); 791 } 792 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list); 793 so->so_qstate = SQ_INCOMP; 794 head->sol_incqlen++; 795 SOLISTEN_UNLOCK(head); 796 } 797 return (so); 798 } 799 800 #if defined(SCTP) || defined(SCTP_SUPPORT) 801 /* 802 * Socket part of sctp_peeloff(). Detach a new socket from an 803 * association. The new socket is returned with a reference. 804 */ 805 struct socket * 806 sopeeloff(struct socket *head) 807 { 808 struct socket *so; 809 810 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p", 811 __func__, __LINE__, head)); 812 so = soalloc(head->so_vnet); 813 if (so == NULL) { 814 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 815 "limit reached or out of memory\n", 816 __func__, head->so_pcb); 817 return (NULL); 818 } 819 so->so_type = head->so_type; 820 so->so_options = head->so_options; 821 so->so_linger = head->so_linger; 822 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED; 823 so->so_fibnum = head->so_fibnum; 824 so->so_proto = head->so_proto; 825 so->so_cred = crhold(head->so_cred); 826 #ifdef MAC 827 mac_socket_newconn(head, so); 828 #endif 829 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 830 so_rdknl_assert_lock); 831 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 832 so_wrknl_assert_lock); 833 VNET_SO_ASSERT(head); 834 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { 835 sodealloc(so); 836 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 837 __func__, head->so_pcb); 838 return (NULL); 839 } 840 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 841 sodealloc(so); 842 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 843 __func__, head->so_pcb); 844 return (NULL); 845 } 846 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 847 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 848 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 849 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 850 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; 851 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; 852 853 soref(so); 854 855 return (so); 856 } 857 #endif /* SCTP */ 858 859 int 860 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 861 { 862 int error; 863 864 CURVNET_SET(so->so_vnet); 865 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td); 866 CURVNET_RESTORE(); 867 return (error); 868 } 869 870 int 871 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 872 { 873 int error; 874 875 CURVNET_SET(so->so_vnet); 876 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td); 877 CURVNET_RESTORE(); 878 return (error); 879 } 880 881 /* 882 * solisten() transitions a socket from a non-listening state to a listening 883 * state, but can also be used to update the listen queue depth on an 884 * existing listen socket. The protocol will call back into the sockets 885 * layer using solisten_proto_check() and solisten_proto() to check and set 886 * socket-layer listen state. Call backs are used so that the protocol can 887 * acquire both protocol and socket layer locks in whatever order is required 888 * by the protocol. 889 * 890 * Protocol implementors are advised to hold the socket lock across the 891 * socket-layer test and set to avoid races at the socket layer. 892 */ 893 int 894 solisten(struct socket *so, int backlog, struct thread *td) 895 { 896 int error; 897 898 CURVNET_SET(so->so_vnet); 899 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td); 900 CURVNET_RESTORE(); 901 return (error); 902 } 903 904 /* 905 * Prepare for a call to solisten_proto(). Acquire all socket buffer locks in 906 * order to interlock with socket I/O. 907 */ 908 int 909 solisten_proto_check(struct socket *so) 910 { 911 SOCK_LOCK_ASSERT(so); 912 913 if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 914 SS_ISDISCONNECTING)) != 0) 915 return (EINVAL); 916 917 /* 918 * Sleeping is not permitted here, so simply fail if userspace is 919 * attempting to transmit or receive on the socket. This kind of 920 * transient failure is not ideal, but it should occur only if userspace 921 * is misusing the socket interfaces. 922 */ 923 if (!sx_try_xlock(&so->so_snd_sx)) 924 return (EAGAIN); 925 if (!sx_try_xlock(&so->so_rcv_sx)) { 926 sx_xunlock(&so->so_snd_sx); 927 return (EAGAIN); 928 } 929 mtx_lock(&so->so_snd_mtx); 930 mtx_lock(&so->so_rcv_mtx); 931 932 /* Interlock with soo_aio_queue(). */ 933 if ((so->so_snd.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0 || 934 (so->so_rcv.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0) { 935 solisten_proto_abort(so); 936 return (EINVAL); 937 } 938 return (0); 939 } 940 941 /* 942 * Undo the setup done by solisten_proto_check(). 943 */ 944 void 945 solisten_proto_abort(struct socket *so) 946 { 947 mtx_unlock(&so->so_snd_mtx); 948 mtx_unlock(&so->so_rcv_mtx); 949 sx_xunlock(&so->so_snd_sx); 950 sx_xunlock(&so->so_rcv_sx); 951 } 952 953 void 954 solisten_proto(struct socket *so, int backlog) 955 { 956 int sbrcv_lowat, sbsnd_lowat; 957 u_int sbrcv_hiwat, sbsnd_hiwat; 958 short sbrcv_flags, sbsnd_flags; 959 sbintime_t sbrcv_timeo, sbsnd_timeo; 960 961 SOCK_LOCK_ASSERT(so); 962 KASSERT((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 963 SS_ISDISCONNECTING)) == 0, 964 ("%s: bad socket state %p", __func__, so)); 965 966 if (SOLISTENING(so)) 967 goto listening; 968 969 /* 970 * Change this socket to listening state. 971 */ 972 sbrcv_lowat = so->so_rcv.sb_lowat; 973 sbsnd_lowat = so->so_snd.sb_lowat; 974 sbrcv_hiwat = so->so_rcv.sb_hiwat; 975 sbsnd_hiwat = so->so_snd.sb_hiwat; 976 sbrcv_flags = so->so_rcv.sb_flags; 977 sbsnd_flags = so->so_snd.sb_flags; 978 sbrcv_timeo = so->so_rcv.sb_timeo; 979 sbsnd_timeo = so->so_snd.sb_timeo; 980 981 sbdestroy(&so->so_snd, so); 982 sbdestroy(&so->so_rcv, so); 983 984 #ifdef INVARIANTS 985 bzero(&so->so_rcv, 986 sizeof(struct socket) - offsetof(struct socket, so_rcv)); 987 #endif 988 989 so->sol_sbrcv_lowat = sbrcv_lowat; 990 so->sol_sbsnd_lowat = sbsnd_lowat; 991 so->sol_sbrcv_hiwat = sbrcv_hiwat; 992 so->sol_sbsnd_hiwat = sbsnd_hiwat; 993 so->sol_sbrcv_flags = sbrcv_flags; 994 so->sol_sbsnd_flags = sbsnd_flags; 995 so->sol_sbrcv_timeo = sbrcv_timeo; 996 so->sol_sbsnd_timeo = sbsnd_timeo; 997 998 so->sol_qlen = so->sol_incqlen = 0; 999 TAILQ_INIT(&so->sol_incomp); 1000 TAILQ_INIT(&so->sol_comp); 1001 1002 so->sol_accept_filter = NULL; 1003 so->sol_accept_filter_arg = NULL; 1004 so->sol_accept_filter_str = NULL; 1005 1006 so->sol_upcall = NULL; 1007 so->sol_upcallarg = NULL; 1008 1009 so->so_options |= SO_ACCEPTCONN; 1010 1011 listening: 1012 if (backlog < 0 || backlog > somaxconn) 1013 backlog = somaxconn; 1014 so->sol_qlimit = backlog; 1015 1016 mtx_unlock(&so->so_snd_mtx); 1017 mtx_unlock(&so->so_rcv_mtx); 1018 sx_xunlock(&so->so_snd_sx); 1019 sx_xunlock(&so->so_rcv_sx); 1020 } 1021 1022 /* 1023 * Wakeup listeners/subsystems once we have a complete connection. 1024 * Enters with lock, returns unlocked. 1025 */ 1026 void 1027 solisten_wakeup(struct socket *sol) 1028 { 1029 1030 if (sol->sol_upcall != NULL) 1031 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT); 1032 else { 1033 selwakeuppri(&sol->so_rdsel, PSOCK); 1034 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0); 1035 } 1036 SOLISTEN_UNLOCK(sol); 1037 wakeup_one(&sol->sol_comp); 1038 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL) 1039 pgsigio(&sol->so_sigio, SIGIO, 0); 1040 } 1041 1042 /* 1043 * Return single connection off a listening socket queue. Main consumer of 1044 * the function is kern_accept4(). Some modules, that do their own accept 1045 * management also use the function. 1046 * 1047 * Listening socket must be locked on entry and is returned unlocked on 1048 * return. 1049 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT. 1050 */ 1051 int 1052 solisten_dequeue(struct socket *head, struct socket **ret, int flags) 1053 { 1054 struct socket *so; 1055 int error; 1056 1057 SOLISTEN_LOCK_ASSERT(head); 1058 1059 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) && 1060 head->so_error == 0) { 1061 error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH, 1062 "accept", 0); 1063 if (error != 0) { 1064 SOLISTEN_UNLOCK(head); 1065 return (error); 1066 } 1067 } 1068 if (head->so_error) { 1069 error = head->so_error; 1070 head->so_error = 0; 1071 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp)) 1072 error = EWOULDBLOCK; 1073 else 1074 error = 0; 1075 if (error) { 1076 SOLISTEN_UNLOCK(head); 1077 return (error); 1078 } 1079 so = TAILQ_FIRST(&head->sol_comp); 1080 SOCK_LOCK(so); 1081 KASSERT(so->so_qstate == SQ_COMP, 1082 ("%s: so %p not SQ_COMP", __func__, so)); 1083 soref(so); 1084 head->sol_qlen--; 1085 so->so_qstate = SQ_NONE; 1086 so->so_listen = NULL; 1087 TAILQ_REMOVE(&head->sol_comp, so, so_list); 1088 if (flags & ACCEPT4_INHERIT) 1089 so->so_state |= (head->so_state & SS_NBIO); 1090 else 1091 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0; 1092 SOCK_UNLOCK(so); 1093 sorele_locked(head); 1094 1095 *ret = so; 1096 return (0); 1097 } 1098 1099 /* 1100 * Evaluate the reference count and named references on a socket; if no 1101 * references remain, free it. This should be called whenever a reference is 1102 * released, such as in sorele(), but also when named reference flags are 1103 * cleared in socket or protocol code. 1104 * 1105 * sofree() will free the socket if: 1106 * 1107 * - There are no outstanding file descriptor references or related consumers 1108 * (so_count == 0). 1109 * 1110 * - The socket has been closed by user space, if ever open (SS_NOFDREF). 1111 * 1112 * - The protocol does not have an outstanding strong reference on the socket 1113 * (SS_PROTOREF). 1114 * 1115 * - The socket is not in a completed connection queue, so a process has been 1116 * notified that it is present. If it is removed, the user process may 1117 * block in accept() despite select() saying the socket was ready. 1118 */ 1119 void 1120 sofree(struct socket *so) 1121 { 1122 struct protosw *pr = so->so_proto; 1123 bool last __diagused; 1124 1125 SOCK_LOCK_ASSERT(so); 1126 1127 if ((so->so_state & (SS_NOFDREF | SS_PROTOREF)) != SS_NOFDREF || 1128 refcount_load(&so->so_count) != 0 || so->so_qstate == SQ_COMP) { 1129 SOCK_UNLOCK(so); 1130 return; 1131 } 1132 1133 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) { 1134 struct socket *sol; 1135 1136 sol = so->so_listen; 1137 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so)); 1138 1139 /* 1140 * To solve race between close of a listening socket and 1141 * a socket on its incomplete queue, we need to lock both. 1142 * The order is first listening socket, then regular. 1143 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this 1144 * function and the listening socket are the only pointers 1145 * to so. To preserve so and sol, we reference both and then 1146 * relock. 1147 * After relock the socket may not move to so_comp since it 1148 * doesn't have PCB already, but it may be removed from 1149 * so_incomp. If that happens, we share responsiblity on 1150 * freeing the socket, but soclose() has already removed 1151 * it from queue. 1152 */ 1153 soref(sol); 1154 soref(so); 1155 SOCK_UNLOCK(so); 1156 SOLISTEN_LOCK(sol); 1157 SOCK_LOCK(so); 1158 if (so->so_qstate == SQ_INCOMP) { 1159 KASSERT(so->so_listen == sol, 1160 ("%s: so %p migrated out of sol %p", 1161 __func__, so, sol)); 1162 TAILQ_REMOVE(&sol->sol_incomp, so, so_list); 1163 sol->sol_incqlen--; 1164 last = refcount_release(&sol->so_count); 1165 KASSERT(!last, ("%s: released last reference for %p", 1166 __func__, sol)); 1167 so->so_qstate = SQ_NONE; 1168 so->so_listen = NULL; 1169 } else 1170 KASSERT(so->so_listen == NULL, 1171 ("%s: so %p not on (in)comp with so_listen", 1172 __func__, so)); 1173 sorele_locked(sol); 1174 KASSERT(refcount_load(&so->so_count) == 1, 1175 ("%s: so %p count %u", __func__, so, so->so_count)); 1176 so->so_count = 0; 1177 } 1178 if (SOLISTENING(so)) 1179 so->so_error = ECONNABORTED; 1180 SOCK_UNLOCK(so); 1181 1182 if (so->so_dtor != NULL) 1183 so->so_dtor(so); 1184 1185 VNET_SO_ASSERT(so); 1186 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 1187 (*pr->pr_domain->dom_dispose)(so); 1188 if (pr->pr_usrreqs->pru_detach != NULL) 1189 (*pr->pr_usrreqs->pru_detach)(so); 1190 1191 /* 1192 * From this point on, we assume that no other references to this 1193 * socket exist anywhere else in the stack. Therefore, no locks need 1194 * to be acquired or held. 1195 * 1196 * We used to do a lot of socket buffer and socket locking here, as 1197 * well as invoke sorflush() and perform wakeups. The direct call to 1198 * dom_dispose() and sbdestroy() are an inlining of what was 1199 * necessary from sorflush(). 1200 * 1201 * Notice that the socket buffer and kqueue state are torn down 1202 * before calling pru_detach. This means that protocols shold not 1203 * assume they can perform socket wakeups, etc, in their detach code. 1204 */ 1205 if (!SOLISTENING(so)) { 1206 sbdestroy(&so->so_snd, so); 1207 sbdestroy(&so->so_rcv, so); 1208 } 1209 seldrain(&so->so_rdsel); 1210 seldrain(&so->so_wrsel); 1211 knlist_destroy(&so->so_rdsel.si_note); 1212 knlist_destroy(&so->so_wrsel.si_note); 1213 sodealloc(so); 1214 } 1215 1216 /* 1217 * Release a reference on a socket while holding the socket lock. 1218 * Unlocks the socket lock before returning. 1219 */ 1220 void 1221 sorele_locked(struct socket *so) 1222 { 1223 SOCK_LOCK_ASSERT(so); 1224 if (refcount_release(&so->so_count)) 1225 sofree(so); 1226 else 1227 SOCK_UNLOCK(so); 1228 } 1229 1230 /* 1231 * Close a socket on last file table reference removal. Initiate disconnect 1232 * if connected. Free socket when disconnect complete. 1233 * 1234 * This function will sorele() the socket. Note that soclose() may be called 1235 * prior to the ref count reaching zero. The actual socket structure will 1236 * not be freed until the ref count reaches zero. 1237 */ 1238 int 1239 soclose(struct socket *so) 1240 { 1241 struct accept_queue lqueue; 1242 int error = 0; 1243 bool listening, last __diagused; 1244 1245 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter")); 1246 1247 CURVNET_SET(so->so_vnet); 1248 funsetown(&so->so_sigio); 1249 if (so->so_state & SS_ISCONNECTED) { 1250 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 1251 error = sodisconnect(so); 1252 if (error) { 1253 if (error == ENOTCONN) 1254 error = 0; 1255 goto drop; 1256 } 1257 } 1258 1259 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) { 1260 if ((so->so_state & SS_ISDISCONNECTING) && 1261 (so->so_state & SS_NBIO)) 1262 goto drop; 1263 while (so->so_state & SS_ISCONNECTED) { 1264 error = tsleep(&so->so_timeo, 1265 PSOCK | PCATCH, "soclos", 1266 so->so_linger * hz); 1267 if (error) 1268 break; 1269 } 1270 } 1271 } 1272 1273 drop: 1274 if (so->so_proto->pr_usrreqs->pru_close != NULL) 1275 (*so->so_proto->pr_usrreqs->pru_close)(so); 1276 1277 SOCK_LOCK(so); 1278 if ((listening = SOLISTENING(so))) { 1279 struct socket *sp; 1280 1281 TAILQ_INIT(&lqueue); 1282 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list); 1283 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list); 1284 1285 so->sol_qlen = so->sol_incqlen = 0; 1286 1287 TAILQ_FOREACH(sp, &lqueue, so_list) { 1288 SOCK_LOCK(sp); 1289 sp->so_qstate = SQ_NONE; 1290 sp->so_listen = NULL; 1291 SOCK_UNLOCK(sp); 1292 last = refcount_release(&so->so_count); 1293 KASSERT(!last, ("%s: released last reference for %p", 1294 __func__, so)); 1295 } 1296 } 1297 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF")); 1298 so->so_state |= SS_NOFDREF; 1299 sorele_locked(so); 1300 if (listening) { 1301 struct socket *sp, *tsp; 1302 1303 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) { 1304 SOCK_LOCK(sp); 1305 if (refcount_load(&sp->so_count) == 0) { 1306 SOCK_UNLOCK(sp); 1307 soabort(sp); 1308 } else { 1309 /* See the handling of queued sockets 1310 in sofree(). */ 1311 SOCK_UNLOCK(sp); 1312 } 1313 } 1314 } 1315 CURVNET_RESTORE(); 1316 return (error); 1317 } 1318 1319 /* 1320 * soabort() is used to abruptly tear down a connection, such as when a 1321 * resource limit is reached (listen queue depth exceeded), or if a listen 1322 * socket is closed while there are sockets waiting to be accepted. 1323 * 1324 * This interface is tricky, because it is called on an unreferenced socket, 1325 * and must be called only by a thread that has actually removed the socket 1326 * from the listen queue it was on, or races with other threads are risked. 1327 * 1328 * This interface will call into the protocol code, so must not be called 1329 * with any socket locks held. Protocols do call it while holding their own 1330 * recursible protocol mutexes, but this is something that should be subject 1331 * to review in the future. 1332 */ 1333 void 1334 soabort(struct socket *so) 1335 { 1336 1337 /* 1338 * In as much as is possible, assert that no references to this 1339 * socket are held. This is not quite the same as asserting that the 1340 * current thread is responsible for arranging for no references, but 1341 * is as close as we can get for now. 1342 */ 1343 KASSERT(so->so_count == 0, ("soabort: so_count")); 1344 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF")); 1345 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF")); 1346 VNET_SO_ASSERT(so); 1347 1348 if (so->so_proto->pr_usrreqs->pru_abort != NULL) 1349 (*so->so_proto->pr_usrreqs->pru_abort)(so); 1350 SOCK_LOCK(so); 1351 sofree(so); 1352 } 1353 1354 int 1355 soaccept(struct socket *so, struct sockaddr **nam) 1356 { 1357 int error; 1358 1359 SOCK_LOCK(so); 1360 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF")); 1361 so->so_state &= ~SS_NOFDREF; 1362 SOCK_UNLOCK(so); 1363 1364 CURVNET_SET(so->so_vnet); 1365 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); 1366 CURVNET_RESTORE(); 1367 return (error); 1368 } 1369 1370 int 1371 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 1372 { 1373 1374 return (soconnectat(AT_FDCWD, so, nam, td)); 1375 } 1376 1377 int 1378 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 1379 { 1380 int error; 1381 1382 CURVNET_SET(so->so_vnet); 1383 /* 1384 * If protocol is connection-based, can only connect once. 1385 * Otherwise, if connected, try to disconnect first. This allows 1386 * user to disconnect by connecting to, e.g., a null address. 1387 */ 1388 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 1389 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 1390 (error = sodisconnect(so)))) { 1391 error = EISCONN; 1392 } else { 1393 /* 1394 * Prevent accumulated error from previous connection from 1395 * biting us. 1396 */ 1397 so->so_error = 0; 1398 if (fd == AT_FDCWD) { 1399 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, 1400 nam, td); 1401 } else { 1402 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd, 1403 so, nam, td); 1404 } 1405 } 1406 CURVNET_RESTORE(); 1407 1408 return (error); 1409 } 1410 1411 int 1412 soconnect2(struct socket *so1, struct socket *so2) 1413 { 1414 int error; 1415 1416 CURVNET_SET(so1->so_vnet); 1417 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); 1418 CURVNET_RESTORE(); 1419 return (error); 1420 } 1421 1422 int 1423 sodisconnect(struct socket *so) 1424 { 1425 int error; 1426 1427 if ((so->so_state & SS_ISCONNECTED) == 0) 1428 return (ENOTCONN); 1429 if (so->so_state & SS_ISDISCONNECTING) 1430 return (EALREADY); 1431 VNET_SO_ASSERT(so); 1432 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 1433 return (error); 1434 } 1435 1436 int 1437 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1438 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1439 { 1440 long space; 1441 ssize_t resid; 1442 int clen = 0, error, dontroute; 1443 1444 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 1445 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 1446 ("sosend_dgram: !PR_ATOMIC")); 1447 1448 if (uio != NULL) 1449 resid = uio->uio_resid; 1450 else 1451 resid = top->m_pkthdr.len; 1452 /* 1453 * In theory resid should be unsigned. However, space must be 1454 * signed, as it might be less than 0 if we over-committed, and we 1455 * must use a signed comparison of space and resid. On the other 1456 * hand, a negative resid causes us to loop sending 0-length 1457 * segments to the protocol. 1458 */ 1459 if (resid < 0) { 1460 error = EINVAL; 1461 goto out; 1462 } 1463 1464 dontroute = 1465 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 1466 if (td != NULL) 1467 td->td_ru.ru_msgsnd++; 1468 if (control != NULL) 1469 clen = control->m_len; 1470 1471 SOCKBUF_LOCK(&so->so_snd); 1472 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1473 SOCKBUF_UNLOCK(&so->so_snd); 1474 error = EPIPE; 1475 goto out; 1476 } 1477 if (so->so_error) { 1478 error = so->so_error; 1479 so->so_error = 0; 1480 SOCKBUF_UNLOCK(&so->so_snd); 1481 goto out; 1482 } 1483 if ((so->so_state & SS_ISCONNECTED) == 0) { 1484 /* 1485 * `sendto' and `sendmsg' is allowed on a connection-based 1486 * socket if it supports implied connect. Return ENOTCONN if 1487 * not connected and no address is supplied. 1488 */ 1489 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1490 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1491 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1492 !(resid == 0 && clen != 0)) { 1493 SOCKBUF_UNLOCK(&so->so_snd); 1494 error = ENOTCONN; 1495 goto out; 1496 } 1497 } else if (addr == NULL) { 1498 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1499 error = ENOTCONN; 1500 else 1501 error = EDESTADDRREQ; 1502 SOCKBUF_UNLOCK(&so->so_snd); 1503 goto out; 1504 } 1505 } 1506 1507 /* 1508 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1509 * problem and need fixing. 1510 */ 1511 space = sbspace(&so->so_snd); 1512 if (flags & MSG_OOB) 1513 space += 1024; 1514 space -= clen; 1515 SOCKBUF_UNLOCK(&so->so_snd); 1516 if (resid > space) { 1517 error = EMSGSIZE; 1518 goto out; 1519 } 1520 if (uio == NULL) { 1521 resid = 0; 1522 if (flags & MSG_EOR) 1523 top->m_flags |= M_EOR; 1524 } else { 1525 /* 1526 * Copy the data from userland into a mbuf chain. 1527 * If no data is to be copied in, a single empty mbuf 1528 * is returned. 1529 */ 1530 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1531 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1532 if (top == NULL) { 1533 error = EFAULT; /* only possible error */ 1534 goto out; 1535 } 1536 space -= resid - uio->uio_resid; 1537 resid = uio->uio_resid; 1538 } 1539 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1540 /* 1541 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1542 * than with. 1543 */ 1544 if (dontroute) { 1545 SOCK_LOCK(so); 1546 so->so_options |= SO_DONTROUTE; 1547 SOCK_UNLOCK(so); 1548 } 1549 /* 1550 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1551 * of date. We could have received a reset packet in an interrupt or 1552 * maybe we slept while doing page faults in uiomove() etc. We could 1553 * probably recheck again inside the locking protection here, but 1554 * there are probably other places that this also happens. We must 1555 * rethink this. 1556 */ 1557 VNET_SO_ASSERT(so); 1558 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1559 (flags & MSG_OOB) ? PRUS_OOB : 1560 /* 1561 * If the user set MSG_EOF, the protocol understands this flag and 1562 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1563 */ 1564 ((flags & MSG_EOF) && 1565 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1566 (resid <= 0)) ? 1567 PRUS_EOF : 1568 /* If there is more to send set PRUS_MORETOCOME */ 1569 (flags & MSG_MORETOCOME) || 1570 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1571 top, addr, control, td); 1572 if (dontroute) { 1573 SOCK_LOCK(so); 1574 so->so_options &= ~SO_DONTROUTE; 1575 SOCK_UNLOCK(so); 1576 } 1577 clen = 0; 1578 control = NULL; 1579 top = NULL; 1580 out: 1581 if (top != NULL) 1582 m_freem(top); 1583 if (control != NULL) 1584 m_freem(control); 1585 return (error); 1586 } 1587 1588 /* 1589 * Send on a socket. If send must go all at once and message is larger than 1590 * send buffering, then hard error. Lock against other senders. If must go 1591 * all at once and not enough room now, then inform user that this would 1592 * block and do nothing. Otherwise, if nonblocking, send as much as 1593 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1594 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1595 * in mbuf chain must be small enough to send all at once. 1596 * 1597 * Returns nonzero on error, timeout or signal; callers must check for short 1598 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1599 * on return. 1600 */ 1601 int 1602 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1603 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1604 { 1605 long space; 1606 ssize_t resid; 1607 int clen = 0, error, dontroute; 1608 int atomic = sosendallatonce(so) || top; 1609 int pru_flag; 1610 #ifdef KERN_TLS 1611 struct ktls_session *tls; 1612 int tls_enq_cnt, tls_pruflag; 1613 uint8_t tls_rtype; 1614 1615 tls = NULL; 1616 tls_rtype = TLS_RLTYPE_APP; 1617 #endif 1618 if (uio != NULL) 1619 resid = uio->uio_resid; 1620 else if ((top->m_flags & M_PKTHDR) != 0) 1621 resid = top->m_pkthdr.len; 1622 else 1623 resid = m_length(top, NULL); 1624 /* 1625 * In theory resid should be unsigned. However, space must be 1626 * signed, as it might be less than 0 if we over-committed, and we 1627 * must use a signed comparison of space and resid. On the other 1628 * hand, a negative resid causes us to loop sending 0-length 1629 * segments to the protocol. 1630 * 1631 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1632 * type sockets since that's an error. 1633 */ 1634 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1635 error = EINVAL; 1636 goto out; 1637 } 1638 1639 dontroute = 1640 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1641 (so->so_proto->pr_flags & PR_ATOMIC); 1642 if (td != NULL) 1643 td->td_ru.ru_msgsnd++; 1644 if (control != NULL) 1645 clen = control->m_len; 1646 1647 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 1648 if (error) 1649 goto out; 1650 1651 #ifdef KERN_TLS 1652 tls_pruflag = 0; 1653 tls = ktls_hold(so->so_snd.sb_tls_info); 1654 if (tls != NULL) { 1655 if (tls->mode == TCP_TLS_MODE_SW) 1656 tls_pruflag = PRUS_NOTREADY; 1657 1658 if (control != NULL) { 1659 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1660 1661 if (clen >= sizeof(*cm) && 1662 cm->cmsg_type == TLS_SET_RECORD_TYPE) { 1663 tls_rtype = *((uint8_t *)CMSG_DATA(cm)); 1664 clen = 0; 1665 m_freem(control); 1666 control = NULL; 1667 atomic = 1; 1668 } 1669 } 1670 1671 if (resid == 0 && !ktls_permit_empty_frames(tls)) { 1672 error = EINVAL; 1673 goto release; 1674 } 1675 } 1676 #endif 1677 1678 restart: 1679 do { 1680 SOCKBUF_LOCK(&so->so_snd); 1681 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1682 SOCKBUF_UNLOCK(&so->so_snd); 1683 error = EPIPE; 1684 goto release; 1685 } 1686 if (so->so_error) { 1687 error = so->so_error; 1688 so->so_error = 0; 1689 SOCKBUF_UNLOCK(&so->so_snd); 1690 goto release; 1691 } 1692 if ((so->so_state & SS_ISCONNECTED) == 0) { 1693 /* 1694 * `sendto' and `sendmsg' is allowed on a connection- 1695 * based socket if it supports implied connect. 1696 * Return ENOTCONN if not connected and no address is 1697 * supplied. 1698 */ 1699 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1700 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1701 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1702 !(resid == 0 && clen != 0)) { 1703 SOCKBUF_UNLOCK(&so->so_snd); 1704 error = ENOTCONN; 1705 goto release; 1706 } 1707 } else if (addr == NULL) { 1708 SOCKBUF_UNLOCK(&so->so_snd); 1709 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1710 error = ENOTCONN; 1711 else 1712 error = EDESTADDRREQ; 1713 goto release; 1714 } 1715 } 1716 space = sbspace(&so->so_snd); 1717 if (flags & MSG_OOB) 1718 space += 1024; 1719 if ((atomic && resid > so->so_snd.sb_hiwat) || 1720 clen > so->so_snd.sb_hiwat) { 1721 SOCKBUF_UNLOCK(&so->so_snd); 1722 error = EMSGSIZE; 1723 goto release; 1724 } 1725 if (space < resid + clen && 1726 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1727 if ((so->so_state & SS_NBIO) || 1728 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) { 1729 SOCKBUF_UNLOCK(&so->so_snd); 1730 error = EWOULDBLOCK; 1731 goto release; 1732 } 1733 error = sbwait(&so->so_snd); 1734 SOCKBUF_UNLOCK(&so->so_snd); 1735 if (error) 1736 goto release; 1737 goto restart; 1738 } 1739 SOCKBUF_UNLOCK(&so->so_snd); 1740 space -= clen; 1741 do { 1742 if (uio == NULL) { 1743 resid = 0; 1744 if (flags & MSG_EOR) 1745 top->m_flags |= M_EOR; 1746 #ifdef KERN_TLS 1747 if (tls != NULL) { 1748 ktls_frame(top, tls, &tls_enq_cnt, 1749 tls_rtype); 1750 tls_rtype = TLS_RLTYPE_APP; 1751 } 1752 #endif 1753 } else { 1754 /* 1755 * Copy the data from userland into a mbuf 1756 * chain. If resid is 0, which can happen 1757 * only if we have control to send, then 1758 * a single empty mbuf is returned. This 1759 * is a workaround to prevent protocol send 1760 * methods to panic. 1761 */ 1762 #ifdef KERN_TLS 1763 if (tls != NULL) { 1764 top = m_uiotombuf(uio, M_WAITOK, space, 1765 tls->params.max_frame_len, 1766 M_EXTPG | 1767 ((flags & MSG_EOR) ? M_EOR : 0)); 1768 if (top != NULL) { 1769 ktls_frame(top, tls, 1770 &tls_enq_cnt, tls_rtype); 1771 } 1772 tls_rtype = TLS_RLTYPE_APP; 1773 } else 1774 #endif 1775 top = m_uiotombuf(uio, M_WAITOK, space, 1776 (atomic ? max_hdr : 0), 1777 (atomic ? M_PKTHDR : 0) | 1778 ((flags & MSG_EOR) ? M_EOR : 0)); 1779 if (top == NULL) { 1780 error = EFAULT; /* only possible error */ 1781 goto release; 1782 } 1783 space -= resid - uio->uio_resid; 1784 resid = uio->uio_resid; 1785 } 1786 if (dontroute) { 1787 SOCK_LOCK(so); 1788 so->so_options |= SO_DONTROUTE; 1789 SOCK_UNLOCK(so); 1790 } 1791 /* 1792 * XXX all the SBS_CANTSENDMORE checks previously 1793 * done could be out of date. We could have received 1794 * a reset packet in an interrupt or maybe we slept 1795 * while doing page faults in uiomove() etc. We 1796 * could probably recheck again inside the locking 1797 * protection here, but there are probably other 1798 * places that this also happens. We must rethink 1799 * this. 1800 */ 1801 VNET_SO_ASSERT(so); 1802 1803 pru_flag = (flags & MSG_OOB) ? PRUS_OOB : 1804 /* 1805 * If the user set MSG_EOF, the protocol understands 1806 * this flag and nothing left to send then use 1807 * PRU_SEND_EOF instead of PRU_SEND. 1808 */ 1809 ((flags & MSG_EOF) && 1810 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1811 (resid <= 0)) ? 1812 PRUS_EOF : 1813 /* If there is more to send set PRUS_MORETOCOME. */ 1814 (flags & MSG_MORETOCOME) || 1815 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; 1816 1817 #ifdef KERN_TLS 1818 pru_flag |= tls_pruflag; 1819 #endif 1820 1821 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1822 pru_flag, top, addr, control, td); 1823 1824 if (dontroute) { 1825 SOCK_LOCK(so); 1826 so->so_options &= ~SO_DONTROUTE; 1827 SOCK_UNLOCK(so); 1828 } 1829 1830 #ifdef KERN_TLS 1831 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) { 1832 if (error != 0) { 1833 m_freem(top); 1834 top = NULL; 1835 } else { 1836 soref(so); 1837 ktls_enqueue(top, so, tls_enq_cnt); 1838 } 1839 } 1840 #endif 1841 clen = 0; 1842 control = NULL; 1843 top = NULL; 1844 if (error) 1845 goto release; 1846 } while (resid && space > 0); 1847 } while (resid); 1848 1849 release: 1850 SOCK_IO_SEND_UNLOCK(so); 1851 out: 1852 #ifdef KERN_TLS 1853 if (tls != NULL) 1854 ktls_free(tls); 1855 #endif 1856 if (top != NULL) 1857 m_freem(top); 1858 if (control != NULL) 1859 m_freem(control); 1860 return (error); 1861 } 1862 1863 int 1864 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1865 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1866 { 1867 int error; 1868 1869 CURVNET_SET(so->so_vnet); 1870 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, 1871 top, control, flags, td); 1872 CURVNET_RESTORE(); 1873 return (error); 1874 } 1875 1876 /* 1877 * The part of soreceive() that implements reading non-inline out-of-band 1878 * data from a socket. For more complete comments, see soreceive(), from 1879 * which this code originated. 1880 * 1881 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1882 * unable to return an mbuf chain to the caller. 1883 */ 1884 static int 1885 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1886 { 1887 struct protosw *pr = so->so_proto; 1888 struct mbuf *m; 1889 int error; 1890 1891 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1892 VNET_SO_ASSERT(so); 1893 1894 m = m_get(M_WAITOK, MT_DATA); 1895 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 1896 if (error) 1897 goto bad; 1898 do { 1899 error = uiomove(mtod(m, void *), 1900 (int) min(uio->uio_resid, m->m_len), uio); 1901 m = m_free(m); 1902 } while (uio->uio_resid && error == 0 && m); 1903 bad: 1904 if (m != NULL) 1905 m_freem(m); 1906 return (error); 1907 } 1908 1909 /* 1910 * Following replacement or removal of the first mbuf on the first mbuf chain 1911 * of a socket buffer, push necessary state changes back into the socket 1912 * buffer so that other consumers see the values consistently. 'nextrecord' 1913 * is the callers locally stored value of the original value of 1914 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1915 * NOTE: 'nextrecord' may be NULL. 1916 */ 1917 static __inline void 1918 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1919 { 1920 1921 SOCKBUF_LOCK_ASSERT(sb); 1922 /* 1923 * First, update for the new value of nextrecord. If necessary, make 1924 * it the first record. 1925 */ 1926 if (sb->sb_mb != NULL) 1927 sb->sb_mb->m_nextpkt = nextrecord; 1928 else 1929 sb->sb_mb = nextrecord; 1930 1931 /* 1932 * Now update any dependent socket buffer fields to reflect the new 1933 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1934 * addition of a second clause that takes care of the case where 1935 * sb_mb has been updated, but remains the last record. 1936 */ 1937 if (sb->sb_mb == NULL) { 1938 sb->sb_mbtail = NULL; 1939 sb->sb_lastrecord = NULL; 1940 } else if (sb->sb_mb->m_nextpkt == NULL) 1941 sb->sb_lastrecord = sb->sb_mb; 1942 } 1943 1944 /* 1945 * Implement receive operations on a socket. We depend on the way that 1946 * records are added to the sockbuf by sbappend. In particular, each record 1947 * (mbufs linked through m_next) must begin with an address if the protocol 1948 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1949 * data, and then zero or more mbufs of data. In order to allow parallelism 1950 * between network receive and copying to user space, as well as avoid 1951 * sleeping with a mutex held, we release the socket buffer mutex during the 1952 * user space copy. Although the sockbuf is locked, new data may still be 1953 * appended, and thus we must maintain consistency of the sockbuf during that 1954 * time. 1955 * 1956 * The caller may receive the data as a single mbuf chain by supplying an 1957 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1958 * the count in uio_resid. 1959 */ 1960 int 1961 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1962 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1963 { 1964 struct mbuf *m, **mp; 1965 int flags, error, offset; 1966 ssize_t len; 1967 struct protosw *pr = so->so_proto; 1968 struct mbuf *nextrecord; 1969 int moff, type = 0; 1970 ssize_t orig_resid = uio->uio_resid; 1971 1972 mp = mp0; 1973 if (psa != NULL) 1974 *psa = NULL; 1975 if (controlp != NULL) 1976 *controlp = NULL; 1977 if (flagsp != NULL) 1978 flags = *flagsp &~ MSG_EOR; 1979 else 1980 flags = 0; 1981 if (flags & MSG_OOB) 1982 return (soreceive_rcvoob(so, uio, flags)); 1983 if (mp != NULL) 1984 *mp = NULL; 1985 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 1986 && uio->uio_resid) { 1987 VNET_SO_ASSERT(so); 1988 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 1989 } 1990 1991 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 1992 if (error) 1993 return (error); 1994 1995 restart: 1996 SOCKBUF_LOCK(&so->so_rcv); 1997 m = so->so_rcv.sb_mb; 1998 /* 1999 * If we have less data than requested, block awaiting more (subject 2000 * to any timeout) if: 2001 * 1. the current count is less than the low water mark, or 2002 * 2. MSG_DONTWAIT is not set 2003 */ 2004 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 2005 sbavail(&so->so_rcv) < uio->uio_resid) && 2006 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && 2007 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 2008 KASSERT(m != NULL || !sbavail(&so->so_rcv), 2009 ("receive: m == %p sbavail == %u", 2010 m, sbavail(&so->so_rcv))); 2011 if (so->so_error || so->so_rerror) { 2012 if (m != NULL) 2013 goto dontblock; 2014 if (so->so_error) 2015 error = so->so_error; 2016 else 2017 error = so->so_rerror; 2018 if ((flags & MSG_PEEK) == 0) { 2019 if (so->so_error) 2020 so->so_error = 0; 2021 else 2022 so->so_rerror = 0; 2023 } 2024 SOCKBUF_UNLOCK(&so->so_rcv); 2025 goto release; 2026 } 2027 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2028 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2029 if (m != NULL) 2030 goto dontblock; 2031 #ifdef KERN_TLS 2032 else if (so->so_rcv.sb_tlsdcc == 0 && 2033 so->so_rcv.sb_tlscc == 0) { 2034 #else 2035 else { 2036 #endif 2037 SOCKBUF_UNLOCK(&so->so_rcv); 2038 goto release; 2039 } 2040 } 2041 for (; m != NULL; m = m->m_next) 2042 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 2043 m = so->so_rcv.sb_mb; 2044 goto dontblock; 2045 } 2046 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED | 2047 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 && 2048 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 2049 SOCKBUF_UNLOCK(&so->so_rcv); 2050 error = ENOTCONN; 2051 goto release; 2052 } 2053 if (uio->uio_resid == 0) { 2054 SOCKBUF_UNLOCK(&so->so_rcv); 2055 goto release; 2056 } 2057 if ((so->so_state & SS_NBIO) || 2058 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2059 SOCKBUF_UNLOCK(&so->so_rcv); 2060 error = EWOULDBLOCK; 2061 goto release; 2062 } 2063 SBLASTRECORDCHK(&so->so_rcv); 2064 SBLASTMBUFCHK(&so->so_rcv); 2065 error = sbwait(&so->so_rcv); 2066 SOCKBUF_UNLOCK(&so->so_rcv); 2067 if (error) 2068 goto release; 2069 goto restart; 2070 } 2071 dontblock: 2072 /* 2073 * From this point onward, we maintain 'nextrecord' as a cache of the 2074 * pointer to the next record in the socket buffer. We must keep the 2075 * various socket buffer pointers and local stack versions of the 2076 * pointers in sync, pushing out modifications before dropping the 2077 * socket buffer mutex, and re-reading them when picking it up. 2078 * 2079 * Otherwise, we will race with the network stack appending new data 2080 * or records onto the socket buffer by using inconsistent/stale 2081 * versions of the field, possibly resulting in socket buffer 2082 * corruption. 2083 * 2084 * By holding the high-level sblock(), we prevent simultaneous 2085 * readers from pulling off the front of the socket buffer. 2086 */ 2087 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2088 if (uio->uio_td) 2089 uio->uio_td->td_ru.ru_msgrcv++; 2090 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 2091 SBLASTRECORDCHK(&so->so_rcv); 2092 SBLASTMBUFCHK(&so->so_rcv); 2093 nextrecord = m->m_nextpkt; 2094 if (pr->pr_flags & PR_ADDR) { 2095 KASSERT(m->m_type == MT_SONAME, 2096 ("m->m_type == %d", m->m_type)); 2097 orig_resid = 0; 2098 if (psa != NULL) 2099 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2100 M_NOWAIT); 2101 if (flags & MSG_PEEK) { 2102 m = m->m_next; 2103 } else { 2104 sbfree(&so->so_rcv, m); 2105 so->so_rcv.sb_mb = m_free(m); 2106 m = so->so_rcv.sb_mb; 2107 sockbuf_pushsync(&so->so_rcv, nextrecord); 2108 } 2109 } 2110 2111 /* 2112 * Process one or more MT_CONTROL mbufs present before any data mbufs 2113 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 2114 * just copy the data; if !MSG_PEEK, we call into the protocol to 2115 * perform externalization (or freeing if controlp == NULL). 2116 */ 2117 if (m != NULL && m->m_type == MT_CONTROL) { 2118 struct mbuf *cm = NULL, *cmn; 2119 struct mbuf **cme = &cm; 2120 #ifdef KERN_TLS 2121 struct cmsghdr *cmsg; 2122 struct tls_get_record tgr; 2123 2124 /* 2125 * For MSG_TLSAPPDATA, check for a non-application data 2126 * record. If found, return ENXIO without removing 2127 * it from the receive queue. This allows a subsequent 2128 * call without MSG_TLSAPPDATA to receive it. 2129 * Note that, for TLS, there should only be a single 2130 * control mbuf with the TLS_GET_RECORD message in it. 2131 */ 2132 if (flags & MSG_TLSAPPDATA) { 2133 cmsg = mtod(m, struct cmsghdr *); 2134 if (cmsg->cmsg_type == TLS_GET_RECORD && 2135 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) { 2136 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr)); 2137 /* This will need to change for TLS 1.3. */ 2138 if (tgr.tls_type != TLS_RLTYPE_APP) { 2139 SOCKBUF_UNLOCK(&so->so_rcv); 2140 error = ENXIO; 2141 goto release; 2142 } 2143 } 2144 } 2145 #endif 2146 2147 do { 2148 if (flags & MSG_PEEK) { 2149 if (controlp != NULL) { 2150 *controlp = m_copym(m, 0, m->m_len, 2151 M_NOWAIT); 2152 controlp = &(*controlp)->m_next; 2153 } 2154 m = m->m_next; 2155 } else { 2156 sbfree(&so->so_rcv, m); 2157 so->so_rcv.sb_mb = m->m_next; 2158 m->m_next = NULL; 2159 *cme = m; 2160 cme = &(*cme)->m_next; 2161 m = so->so_rcv.sb_mb; 2162 } 2163 } while (m != NULL && m->m_type == MT_CONTROL); 2164 if ((flags & MSG_PEEK) == 0) 2165 sockbuf_pushsync(&so->so_rcv, nextrecord); 2166 while (cm != NULL) { 2167 cmn = cm->m_next; 2168 cm->m_next = NULL; 2169 if (pr->pr_domain->dom_externalize != NULL) { 2170 SOCKBUF_UNLOCK(&so->so_rcv); 2171 VNET_SO_ASSERT(so); 2172 error = (*pr->pr_domain->dom_externalize) 2173 (cm, controlp, flags); 2174 SOCKBUF_LOCK(&so->so_rcv); 2175 } else if (controlp != NULL) 2176 *controlp = cm; 2177 else 2178 m_freem(cm); 2179 if (controlp != NULL) { 2180 while (*controlp != NULL) 2181 controlp = &(*controlp)->m_next; 2182 } 2183 cm = cmn; 2184 } 2185 if (m != NULL) 2186 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 2187 else 2188 nextrecord = so->so_rcv.sb_mb; 2189 orig_resid = 0; 2190 } 2191 if (m != NULL) { 2192 if ((flags & MSG_PEEK) == 0) { 2193 KASSERT(m->m_nextpkt == nextrecord, 2194 ("soreceive: post-control, nextrecord !sync")); 2195 if (nextrecord == NULL) { 2196 KASSERT(so->so_rcv.sb_mb == m, 2197 ("soreceive: post-control, sb_mb!=m")); 2198 KASSERT(so->so_rcv.sb_lastrecord == m, 2199 ("soreceive: post-control, lastrecord!=m")); 2200 } 2201 } 2202 type = m->m_type; 2203 if (type == MT_OOBDATA) 2204 flags |= MSG_OOB; 2205 } else { 2206 if ((flags & MSG_PEEK) == 0) { 2207 KASSERT(so->so_rcv.sb_mb == nextrecord, 2208 ("soreceive: sb_mb != nextrecord")); 2209 if (so->so_rcv.sb_mb == NULL) { 2210 KASSERT(so->so_rcv.sb_lastrecord == NULL, 2211 ("soreceive: sb_lastercord != NULL")); 2212 } 2213 } 2214 } 2215 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2216 SBLASTRECORDCHK(&so->so_rcv); 2217 SBLASTMBUFCHK(&so->so_rcv); 2218 2219 /* 2220 * Now continue to read any data mbufs off of the head of the socket 2221 * buffer until the read request is satisfied. Note that 'type' is 2222 * used to store the type of any mbuf reads that have happened so far 2223 * such that soreceive() can stop reading if the type changes, which 2224 * causes soreceive() to return only one of regular data and inline 2225 * out-of-band data in a single socket receive operation. 2226 */ 2227 moff = 0; 2228 offset = 0; 2229 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 2230 && error == 0) { 2231 /* 2232 * If the type of mbuf has changed since the last mbuf 2233 * examined ('type'), end the receive operation. 2234 */ 2235 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2236 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 2237 if (type != m->m_type) 2238 break; 2239 } else if (type == MT_OOBDATA) 2240 break; 2241 else 2242 KASSERT(m->m_type == MT_DATA, 2243 ("m->m_type == %d", m->m_type)); 2244 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 2245 len = uio->uio_resid; 2246 if (so->so_oobmark && len > so->so_oobmark - offset) 2247 len = so->so_oobmark - offset; 2248 if (len > m->m_len - moff) 2249 len = m->m_len - moff; 2250 /* 2251 * If mp is set, just pass back the mbufs. Otherwise copy 2252 * them out via the uio, then free. Sockbuf must be 2253 * consistent here (points to current mbuf, it points to next 2254 * record) when we drop priority; we must note any additions 2255 * to the sockbuf when we block interrupts again. 2256 */ 2257 if (mp == NULL) { 2258 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2259 SBLASTRECORDCHK(&so->so_rcv); 2260 SBLASTMBUFCHK(&so->so_rcv); 2261 SOCKBUF_UNLOCK(&so->so_rcv); 2262 if ((m->m_flags & M_EXTPG) != 0) 2263 error = m_unmapped_uiomove(m, moff, uio, 2264 (int)len); 2265 else 2266 error = uiomove(mtod(m, char *) + moff, 2267 (int)len, uio); 2268 SOCKBUF_LOCK(&so->so_rcv); 2269 if (error) { 2270 /* 2271 * The MT_SONAME mbuf has already been removed 2272 * from the record, so it is necessary to 2273 * remove the data mbufs, if any, to preserve 2274 * the invariant in the case of PR_ADDR that 2275 * requires MT_SONAME mbufs at the head of 2276 * each record. 2277 */ 2278 if (pr->pr_flags & PR_ATOMIC && 2279 ((flags & MSG_PEEK) == 0)) 2280 (void)sbdroprecord_locked(&so->so_rcv); 2281 SOCKBUF_UNLOCK(&so->so_rcv); 2282 goto release; 2283 } 2284 } else 2285 uio->uio_resid -= len; 2286 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2287 if (len == m->m_len - moff) { 2288 if (m->m_flags & M_EOR) 2289 flags |= MSG_EOR; 2290 if (flags & MSG_PEEK) { 2291 m = m->m_next; 2292 moff = 0; 2293 } else { 2294 nextrecord = m->m_nextpkt; 2295 sbfree(&so->so_rcv, m); 2296 if (mp != NULL) { 2297 m->m_nextpkt = NULL; 2298 *mp = m; 2299 mp = &m->m_next; 2300 so->so_rcv.sb_mb = m = m->m_next; 2301 *mp = NULL; 2302 } else { 2303 so->so_rcv.sb_mb = m_free(m); 2304 m = so->so_rcv.sb_mb; 2305 } 2306 sockbuf_pushsync(&so->so_rcv, nextrecord); 2307 SBLASTRECORDCHK(&so->so_rcv); 2308 SBLASTMBUFCHK(&so->so_rcv); 2309 } 2310 } else { 2311 if (flags & MSG_PEEK) 2312 moff += len; 2313 else { 2314 if (mp != NULL) { 2315 if (flags & MSG_DONTWAIT) { 2316 *mp = m_copym(m, 0, len, 2317 M_NOWAIT); 2318 if (*mp == NULL) { 2319 /* 2320 * m_copym() couldn't 2321 * allocate an mbuf. 2322 * Adjust uio_resid back 2323 * (it was adjusted 2324 * down by len bytes, 2325 * which we didn't end 2326 * up "copying" over). 2327 */ 2328 uio->uio_resid += len; 2329 break; 2330 } 2331 } else { 2332 SOCKBUF_UNLOCK(&so->so_rcv); 2333 *mp = m_copym(m, 0, len, 2334 M_WAITOK); 2335 SOCKBUF_LOCK(&so->so_rcv); 2336 } 2337 } 2338 sbcut_locked(&so->so_rcv, len); 2339 } 2340 } 2341 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2342 if (so->so_oobmark) { 2343 if ((flags & MSG_PEEK) == 0) { 2344 so->so_oobmark -= len; 2345 if (so->so_oobmark == 0) { 2346 so->so_rcv.sb_state |= SBS_RCVATMARK; 2347 break; 2348 } 2349 } else { 2350 offset += len; 2351 if (offset == so->so_oobmark) 2352 break; 2353 } 2354 } 2355 if (flags & MSG_EOR) 2356 break; 2357 /* 2358 * If the MSG_WAITALL flag is set (for non-atomic socket), we 2359 * must not quit until "uio->uio_resid == 0" or an error 2360 * termination. If a signal/timeout occurs, return with a 2361 * short count but without error. Keep sockbuf locked 2362 * against other readers. 2363 */ 2364 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 2365 !sosendallatonce(so) && nextrecord == NULL) { 2366 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2367 if (so->so_error || so->so_rerror || 2368 so->so_rcv.sb_state & SBS_CANTRCVMORE) 2369 break; 2370 /* 2371 * Notify the protocol that some data has been 2372 * drained before blocking. 2373 */ 2374 if (pr->pr_flags & PR_WANTRCVD) { 2375 SOCKBUF_UNLOCK(&so->so_rcv); 2376 VNET_SO_ASSERT(so); 2377 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2378 SOCKBUF_LOCK(&so->so_rcv); 2379 } 2380 SBLASTRECORDCHK(&so->so_rcv); 2381 SBLASTMBUFCHK(&so->so_rcv); 2382 /* 2383 * We could receive some data while was notifying 2384 * the protocol. Skip blocking in this case. 2385 */ 2386 if (so->so_rcv.sb_mb == NULL) { 2387 error = sbwait(&so->so_rcv); 2388 if (error) { 2389 SOCKBUF_UNLOCK(&so->so_rcv); 2390 goto release; 2391 } 2392 } 2393 m = so->so_rcv.sb_mb; 2394 if (m != NULL) 2395 nextrecord = m->m_nextpkt; 2396 } 2397 } 2398 2399 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2400 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 2401 flags |= MSG_TRUNC; 2402 if ((flags & MSG_PEEK) == 0) 2403 (void) sbdroprecord_locked(&so->so_rcv); 2404 } 2405 if ((flags & MSG_PEEK) == 0) { 2406 if (m == NULL) { 2407 /* 2408 * First part is an inline SB_EMPTY_FIXUP(). Second 2409 * part makes sure sb_lastrecord is up-to-date if 2410 * there is still data in the socket buffer. 2411 */ 2412 so->so_rcv.sb_mb = nextrecord; 2413 if (so->so_rcv.sb_mb == NULL) { 2414 so->so_rcv.sb_mbtail = NULL; 2415 so->so_rcv.sb_lastrecord = NULL; 2416 } else if (nextrecord->m_nextpkt == NULL) 2417 so->so_rcv.sb_lastrecord = nextrecord; 2418 } 2419 SBLASTRECORDCHK(&so->so_rcv); 2420 SBLASTMBUFCHK(&so->so_rcv); 2421 /* 2422 * If soreceive() is being done from the socket callback, 2423 * then don't need to generate ACK to peer to update window, 2424 * since ACK will be generated on return to TCP. 2425 */ 2426 if (!(flags & MSG_SOCALLBCK) && 2427 (pr->pr_flags & PR_WANTRCVD)) { 2428 SOCKBUF_UNLOCK(&so->so_rcv); 2429 VNET_SO_ASSERT(so); 2430 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2431 SOCKBUF_LOCK(&so->so_rcv); 2432 } 2433 } 2434 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2435 if (orig_resid == uio->uio_resid && orig_resid && 2436 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 2437 SOCKBUF_UNLOCK(&so->so_rcv); 2438 goto restart; 2439 } 2440 SOCKBUF_UNLOCK(&so->so_rcv); 2441 2442 if (flagsp != NULL) 2443 *flagsp |= flags; 2444 release: 2445 SOCK_IO_RECV_UNLOCK(so); 2446 return (error); 2447 } 2448 2449 /* 2450 * Optimized version of soreceive() for stream (TCP) sockets. 2451 */ 2452 int 2453 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 2454 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2455 { 2456 int len = 0, error = 0, flags, oresid; 2457 struct sockbuf *sb; 2458 struct mbuf *m, *n = NULL; 2459 2460 /* We only do stream sockets. */ 2461 if (so->so_type != SOCK_STREAM) 2462 return (EINVAL); 2463 if (psa != NULL) 2464 *psa = NULL; 2465 if (flagsp != NULL) 2466 flags = *flagsp &~ MSG_EOR; 2467 else 2468 flags = 0; 2469 if (controlp != NULL) 2470 *controlp = NULL; 2471 if (flags & MSG_OOB) 2472 return (soreceive_rcvoob(so, uio, flags)); 2473 if (mp0 != NULL) 2474 *mp0 = NULL; 2475 2476 sb = &so->so_rcv; 2477 2478 #ifdef KERN_TLS 2479 /* 2480 * KTLS store TLS records as records with a control message to 2481 * describe the framing. 2482 * 2483 * We check once here before acquiring locks to optimize the 2484 * common case. 2485 */ 2486 if (sb->sb_tls_info != NULL) 2487 return (soreceive_generic(so, psa, uio, mp0, controlp, 2488 flagsp)); 2489 #endif 2490 2491 /* Prevent other readers from entering the socket. */ 2492 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 2493 if (error) 2494 return (error); 2495 SOCKBUF_LOCK(sb); 2496 2497 #ifdef KERN_TLS 2498 if (sb->sb_tls_info != NULL) { 2499 SOCKBUF_UNLOCK(sb); 2500 SOCK_IO_RECV_UNLOCK(so); 2501 return (soreceive_generic(so, psa, uio, mp0, controlp, 2502 flagsp)); 2503 } 2504 #endif 2505 2506 /* Easy one, no space to copyout anything. */ 2507 if (uio->uio_resid == 0) { 2508 error = EINVAL; 2509 goto out; 2510 } 2511 oresid = uio->uio_resid; 2512 2513 /* We will never ever get anything unless we are or were connected. */ 2514 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 2515 error = ENOTCONN; 2516 goto out; 2517 } 2518 2519 restart: 2520 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2521 2522 /* Abort if socket has reported problems. */ 2523 if (so->so_error) { 2524 if (sbavail(sb) > 0) 2525 goto deliver; 2526 if (oresid > uio->uio_resid) 2527 goto out; 2528 error = so->so_error; 2529 if (!(flags & MSG_PEEK)) 2530 so->so_error = 0; 2531 goto out; 2532 } 2533 2534 /* Door is closed. Deliver what is left, if any. */ 2535 if (sb->sb_state & SBS_CANTRCVMORE) { 2536 if (sbavail(sb) > 0) 2537 goto deliver; 2538 else 2539 goto out; 2540 } 2541 2542 /* Socket buffer is empty and we shall not block. */ 2543 if (sbavail(sb) == 0 && 2544 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2545 error = EAGAIN; 2546 goto out; 2547 } 2548 2549 /* Socket buffer got some data that we shall deliver now. */ 2550 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 2551 ((so->so_state & SS_NBIO) || 2552 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2553 sbavail(sb) >= sb->sb_lowat || 2554 sbavail(sb) >= uio->uio_resid || 2555 sbavail(sb) >= sb->sb_hiwat) ) { 2556 goto deliver; 2557 } 2558 2559 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2560 if ((flags & MSG_WAITALL) && 2561 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 2562 goto deliver; 2563 2564 /* 2565 * Wait and block until (more) data comes in. 2566 * NB: Drops the sockbuf lock during wait. 2567 */ 2568 error = sbwait(sb); 2569 if (error) 2570 goto out; 2571 goto restart; 2572 2573 deliver: 2574 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2575 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 2576 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2577 2578 /* Statistics. */ 2579 if (uio->uio_td) 2580 uio->uio_td->td_ru.ru_msgrcv++; 2581 2582 /* Fill uio until full or current end of socket buffer is reached. */ 2583 len = min(uio->uio_resid, sbavail(sb)); 2584 if (mp0 != NULL) { 2585 /* Dequeue as many mbufs as possible. */ 2586 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2587 if (*mp0 == NULL) 2588 *mp0 = sb->sb_mb; 2589 else 2590 m_cat(*mp0, sb->sb_mb); 2591 for (m = sb->sb_mb; 2592 m != NULL && m->m_len <= len; 2593 m = m->m_next) { 2594 KASSERT(!(m->m_flags & M_NOTAVAIL), 2595 ("%s: m %p not available", __func__, m)); 2596 len -= m->m_len; 2597 uio->uio_resid -= m->m_len; 2598 sbfree(sb, m); 2599 n = m; 2600 } 2601 n->m_next = NULL; 2602 sb->sb_mb = m; 2603 sb->sb_lastrecord = sb->sb_mb; 2604 if (sb->sb_mb == NULL) 2605 SB_EMPTY_FIXUP(sb); 2606 } 2607 /* Copy the remainder. */ 2608 if (len > 0) { 2609 KASSERT(sb->sb_mb != NULL, 2610 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2611 2612 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2613 if (m == NULL) 2614 len = 0; /* Don't flush data from sockbuf. */ 2615 else 2616 uio->uio_resid -= len; 2617 if (*mp0 != NULL) 2618 m_cat(*mp0, m); 2619 else 2620 *mp0 = m; 2621 if (*mp0 == NULL) { 2622 error = ENOBUFS; 2623 goto out; 2624 } 2625 } 2626 } else { 2627 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2628 SOCKBUF_UNLOCK(sb); 2629 error = m_mbuftouio(uio, sb->sb_mb, len); 2630 SOCKBUF_LOCK(sb); 2631 if (error) 2632 goto out; 2633 } 2634 SBLASTRECORDCHK(sb); 2635 SBLASTMBUFCHK(sb); 2636 2637 /* 2638 * Remove the delivered data from the socket buffer unless we 2639 * were only peeking. 2640 */ 2641 if (!(flags & MSG_PEEK)) { 2642 if (len > 0) 2643 sbdrop_locked(sb, len); 2644 2645 /* Notify protocol that we drained some data. */ 2646 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2647 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2648 !(flags & MSG_SOCALLBCK))) { 2649 SOCKBUF_UNLOCK(sb); 2650 VNET_SO_ASSERT(so); 2651 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags); 2652 SOCKBUF_LOCK(sb); 2653 } 2654 } 2655 2656 /* 2657 * For MSG_WAITALL we may have to loop again and wait for 2658 * more data to come in. 2659 */ 2660 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2661 goto restart; 2662 out: 2663 SBLASTRECORDCHK(sb); 2664 SBLASTMBUFCHK(sb); 2665 SOCKBUF_UNLOCK(sb); 2666 SOCK_IO_RECV_UNLOCK(so); 2667 return (error); 2668 } 2669 2670 /* 2671 * Optimized version of soreceive() for simple datagram cases from userspace. 2672 * Unlike in the stream case, we're able to drop a datagram if copyout() 2673 * fails, and because we handle datagrams atomically, we don't need to use a 2674 * sleep lock to prevent I/O interlacing. 2675 */ 2676 int 2677 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2678 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2679 { 2680 struct mbuf *m, *m2; 2681 int flags, error; 2682 ssize_t len; 2683 struct protosw *pr = so->so_proto; 2684 struct mbuf *nextrecord; 2685 2686 if (psa != NULL) 2687 *psa = NULL; 2688 if (controlp != NULL) 2689 *controlp = NULL; 2690 if (flagsp != NULL) 2691 flags = *flagsp &~ MSG_EOR; 2692 else 2693 flags = 0; 2694 2695 /* 2696 * For any complicated cases, fall back to the full 2697 * soreceive_generic(). 2698 */ 2699 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB)) 2700 return (soreceive_generic(so, psa, uio, mp0, controlp, 2701 flagsp)); 2702 2703 /* 2704 * Enforce restrictions on use. 2705 */ 2706 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2707 ("soreceive_dgram: wantrcvd")); 2708 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2709 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2710 ("soreceive_dgram: SBS_RCVATMARK")); 2711 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2712 ("soreceive_dgram: P_CONNREQUIRED")); 2713 2714 /* 2715 * Loop blocking while waiting for a datagram. 2716 */ 2717 SOCKBUF_LOCK(&so->so_rcv); 2718 while ((m = so->so_rcv.sb_mb) == NULL) { 2719 KASSERT(sbavail(&so->so_rcv) == 0, 2720 ("soreceive_dgram: sb_mb NULL but sbavail %u", 2721 sbavail(&so->so_rcv))); 2722 if (so->so_error) { 2723 error = so->so_error; 2724 so->so_error = 0; 2725 SOCKBUF_UNLOCK(&so->so_rcv); 2726 return (error); 2727 } 2728 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2729 uio->uio_resid == 0) { 2730 SOCKBUF_UNLOCK(&so->so_rcv); 2731 return (0); 2732 } 2733 if ((so->so_state & SS_NBIO) || 2734 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2735 SOCKBUF_UNLOCK(&so->so_rcv); 2736 return (EWOULDBLOCK); 2737 } 2738 SBLASTRECORDCHK(&so->so_rcv); 2739 SBLASTMBUFCHK(&so->so_rcv); 2740 error = sbwait(&so->so_rcv); 2741 if (error) { 2742 SOCKBUF_UNLOCK(&so->so_rcv); 2743 return (error); 2744 } 2745 } 2746 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2747 2748 if (uio->uio_td) 2749 uio->uio_td->td_ru.ru_msgrcv++; 2750 SBLASTRECORDCHK(&so->so_rcv); 2751 SBLASTMBUFCHK(&so->so_rcv); 2752 nextrecord = m->m_nextpkt; 2753 if (nextrecord == NULL) { 2754 KASSERT(so->so_rcv.sb_lastrecord == m, 2755 ("soreceive_dgram: lastrecord != m")); 2756 } 2757 2758 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2759 ("soreceive_dgram: m_nextpkt != nextrecord")); 2760 2761 /* 2762 * Pull 'm' and its chain off the front of the packet queue. 2763 */ 2764 so->so_rcv.sb_mb = NULL; 2765 sockbuf_pushsync(&so->so_rcv, nextrecord); 2766 2767 /* 2768 * Walk 'm's chain and free that many bytes from the socket buffer. 2769 */ 2770 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2771 sbfree(&so->so_rcv, m2); 2772 2773 /* 2774 * Do a few last checks before we let go of the lock. 2775 */ 2776 SBLASTRECORDCHK(&so->so_rcv); 2777 SBLASTMBUFCHK(&so->so_rcv); 2778 SOCKBUF_UNLOCK(&so->so_rcv); 2779 2780 if (pr->pr_flags & PR_ADDR) { 2781 KASSERT(m->m_type == MT_SONAME, 2782 ("m->m_type == %d", m->m_type)); 2783 if (psa != NULL) 2784 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2785 M_NOWAIT); 2786 m = m_free(m); 2787 } 2788 if (m == NULL) { 2789 /* XXXRW: Can this happen? */ 2790 return (0); 2791 } 2792 2793 /* 2794 * Packet to copyout() is now in 'm' and it is disconnected from the 2795 * queue. 2796 * 2797 * Process one or more MT_CONTROL mbufs present before any data mbufs 2798 * in the first mbuf chain on the socket buffer. We call into the 2799 * protocol to perform externalization (or freeing if controlp == 2800 * NULL). In some cases there can be only MT_CONTROL mbufs without 2801 * MT_DATA mbufs. 2802 */ 2803 if (m->m_type == MT_CONTROL) { 2804 struct mbuf *cm = NULL, *cmn; 2805 struct mbuf **cme = &cm; 2806 2807 do { 2808 m2 = m->m_next; 2809 m->m_next = NULL; 2810 *cme = m; 2811 cme = &(*cme)->m_next; 2812 m = m2; 2813 } while (m != NULL && m->m_type == MT_CONTROL); 2814 while (cm != NULL) { 2815 cmn = cm->m_next; 2816 cm->m_next = NULL; 2817 if (pr->pr_domain->dom_externalize != NULL) { 2818 error = (*pr->pr_domain->dom_externalize) 2819 (cm, controlp, flags); 2820 } else if (controlp != NULL) 2821 *controlp = cm; 2822 else 2823 m_freem(cm); 2824 if (controlp != NULL) { 2825 while (*controlp != NULL) 2826 controlp = &(*controlp)->m_next; 2827 } 2828 cm = cmn; 2829 } 2830 } 2831 KASSERT(m == NULL || m->m_type == MT_DATA, 2832 ("soreceive_dgram: !data")); 2833 while (m != NULL && uio->uio_resid > 0) { 2834 len = uio->uio_resid; 2835 if (len > m->m_len) 2836 len = m->m_len; 2837 error = uiomove(mtod(m, char *), (int)len, uio); 2838 if (error) { 2839 m_freem(m); 2840 return (error); 2841 } 2842 if (len == m->m_len) 2843 m = m_free(m); 2844 else { 2845 m->m_data += len; 2846 m->m_len -= len; 2847 } 2848 } 2849 if (m != NULL) { 2850 flags |= MSG_TRUNC; 2851 m_freem(m); 2852 } 2853 if (flagsp != NULL) 2854 *flagsp |= flags; 2855 return (0); 2856 } 2857 2858 int 2859 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2860 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2861 { 2862 int error; 2863 2864 CURVNET_SET(so->so_vnet); 2865 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, 2866 mp0, controlp, flagsp)); 2867 CURVNET_RESTORE(); 2868 return (error); 2869 } 2870 2871 int 2872 soshutdown(struct socket *so, int how) 2873 { 2874 struct protosw *pr; 2875 int error, soerror_enotconn; 2876 2877 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2878 return (EINVAL); 2879 2880 soerror_enotconn = 0; 2881 SOCK_LOCK(so); 2882 if ((so->so_state & 2883 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 2884 /* 2885 * POSIX mandates us to return ENOTCONN when shutdown(2) is 2886 * invoked on a datagram sockets, however historically we would 2887 * actually tear socket down. This is known to be leveraged by 2888 * some applications to unblock process waiting in recvXXX(2) 2889 * by other process that it shares that socket with. Try to meet 2890 * both backward-compatibility and POSIX requirements by forcing 2891 * ENOTCONN but still asking protocol to perform pru_shutdown(). 2892 */ 2893 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) { 2894 SOCK_UNLOCK(so); 2895 return (ENOTCONN); 2896 } 2897 soerror_enotconn = 1; 2898 } 2899 2900 if (SOLISTENING(so)) { 2901 if (how != SHUT_WR) { 2902 so->so_error = ECONNABORTED; 2903 solisten_wakeup(so); /* unlocks so */ 2904 } else { 2905 SOCK_UNLOCK(so); 2906 } 2907 goto done; 2908 } 2909 SOCK_UNLOCK(so); 2910 2911 CURVNET_SET(so->so_vnet); 2912 pr = so->so_proto; 2913 if (pr->pr_usrreqs->pru_flush != NULL) 2914 (*pr->pr_usrreqs->pru_flush)(so, how); 2915 if (how != SHUT_WR) 2916 sorflush(so); 2917 if (how != SHUT_RD) { 2918 error = (*pr->pr_usrreqs->pru_shutdown)(so); 2919 wakeup(&so->so_timeo); 2920 CURVNET_RESTORE(); 2921 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error); 2922 } 2923 wakeup(&so->so_timeo); 2924 CURVNET_RESTORE(); 2925 2926 done: 2927 return (soerror_enotconn ? ENOTCONN : 0); 2928 } 2929 2930 void 2931 sorflush(struct socket *so) 2932 { 2933 struct socket aso; 2934 struct protosw *pr; 2935 int error; 2936 2937 VNET_SO_ASSERT(so); 2938 2939 /* 2940 * In order to avoid calling dom_dispose with the socket buffer mutex 2941 * held, we make a partial copy of the socket buffer and clear the 2942 * original. The new socket buffer copy won't have initialized locks so 2943 * we can only call routines that won't use or assert those locks. 2944 * Ideally calling socantrcvmore() would prevent data from being added 2945 * to the buffer, but currently it merely prevents buffered data from 2946 * being read by userspace. We make this effort to free buffered data 2947 * nonetheless. 2948 * 2949 * Dislodge threads currently blocked in receive and wait to acquire 2950 * a lock against other simultaneous readers before clearing the 2951 * socket buffer. Don't let our acquire be interrupted by a signal 2952 * despite any existing socket disposition on interruptable waiting. 2953 */ 2954 socantrcvmore(so); 2955 2956 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR); 2957 if (error != 0) { 2958 KASSERT(SOLISTENING(so), 2959 ("%s: soiolock(%p) failed", __func__, so)); 2960 return; 2961 } 2962 2963 SOCK_RECVBUF_LOCK(so); 2964 bzero(&aso, sizeof(aso)); 2965 aso.so_pcb = so->so_pcb; 2966 bcopy(&so->so_rcv.sb_startzero, &aso.so_rcv.sb_startzero, 2967 offsetof(struct sockbuf, sb_endzero) - 2968 offsetof(struct sockbuf, sb_startzero)); 2969 bzero(&so->so_rcv.sb_startzero, 2970 offsetof(struct sockbuf, sb_endzero) - 2971 offsetof(struct sockbuf, sb_startzero)); 2972 SOCK_RECVBUF_UNLOCK(so); 2973 SOCK_IO_RECV_UNLOCK(so); 2974 2975 /* 2976 * Dispose of special rights and flush the copied socket. Don't call 2977 * any unsafe routines (that rely on locks being initialized) on aso. 2978 */ 2979 pr = so->so_proto; 2980 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 2981 (*pr->pr_domain->dom_dispose)(&aso); 2982 sbrelease_internal(&aso.so_rcv, so); 2983 } 2984 2985 /* 2986 * Wrapper for Socket established helper hook. 2987 * Parameters: socket, context of the hook point, hook id. 2988 */ 2989 static int inline 2990 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 2991 { 2992 struct socket_hhook_data hhook_data = { 2993 .so = so, 2994 .hctx = hctx, 2995 .m = NULL, 2996 .status = 0 2997 }; 2998 2999 CURVNET_SET(so->so_vnet); 3000 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 3001 CURVNET_RESTORE(); 3002 3003 /* Ugly but needed, since hhooks return void for now */ 3004 return (hhook_data.status); 3005 } 3006 3007 /* 3008 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 3009 * additional variant to handle the case where the option value needs to be 3010 * some kind of integer, but not a specific size. In addition to their use 3011 * here, these functions are also called by the protocol-level pr_ctloutput() 3012 * routines. 3013 */ 3014 int 3015 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 3016 { 3017 size_t valsize; 3018 3019 /* 3020 * If the user gives us more than we wanted, we ignore it, but if we 3021 * don't get the minimum length the caller wants, we return EINVAL. 3022 * On success, sopt->sopt_valsize is set to however much we actually 3023 * retrieved. 3024 */ 3025 if ((valsize = sopt->sopt_valsize) < minlen) 3026 return EINVAL; 3027 if (valsize > len) 3028 sopt->sopt_valsize = valsize = len; 3029 3030 if (sopt->sopt_td != NULL) 3031 return (copyin(sopt->sopt_val, buf, valsize)); 3032 3033 bcopy(sopt->sopt_val, buf, valsize); 3034 return (0); 3035 } 3036 3037 /* 3038 * Kernel version of setsockopt(2). 3039 * 3040 * XXX: optlen is size_t, not socklen_t 3041 */ 3042 int 3043 so_setsockopt(struct socket *so, int level, int optname, void *optval, 3044 size_t optlen) 3045 { 3046 struct sockopt sopt; 3047 3048 sopt.sopt_level = level; 3049 sopt.sopt_name = optname; 3050 sopt.sopt_dir = SOPT_SET; 3051 sopt.sopt_val = optval; 3052 sopt.sopt_valsize = optlen; 3053 sopt.sopt_td = NULL; 3054 return (sosetopt(so, &sopt)); 3055 } 3056 3057 int 3058 sosetopt(struct socket *so, struct sockopt *sopt) 3059 { 3060 int error, optval; 3061 struct linger l; 3062 struct timeval tv; 3063 sbintime_t val; 3064 uint32_t val32; 3065 #ifdef MAC 3066 struct mac extmac; 3067 #endif 3068 3069 CURVNET_SET(so->so_vnet); 3070 error = 0; 3071 if (sopt->sopt_level != SOL_SOCKET) { 3072 if (so->so_proto->pr_ctloutput != NULL) 3073 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3074 else 3075 error = ENOPROTOOPT; 3076 } else { 3077 switch (sopt->sopt_name) { 3078 case SO_ACCEPTFILTER: 3079 error = accept_filt_setopt(so, sopt); 3080 if (error) 3081 goto bad; 3082 break; 3083 3084 case SO_LINGER: 3085 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 3086 if (error) 3087 goto bad; 3088 if (l.l_linger < 0 || 3089 l.l_linger > USHRT_MAX || 3090 l.l_linger > (INT_MAX / hz)) { 3091 error = EDOM; 3092 goto bad; 3093 } 3094 SOCK_LOCK(so); 3095 so->so_linger = l.l_linger; 3096 if (l.l_onoff) 3097 so->so_options |= SO_LINGER; 3098 else 3099 so->so_options &= ~SO_LINGER; 3100 SOCK_UNLOCK(so); 3101 break; 3102 3103 case SO_DEBUG: 3104 case SO_KEEPALIVE: 3105 case SO_DONTROUTE: 3106 case SO_USELOOPBACK: 3107 case SO_BROADCAST: 3108 case SO_REUSEADDR: 3109 case SO_REUSEPORT: 3110 case SO_REUSEPORT_LB: 3111 case SO_OOBINLINE: 3112 case SO_TIMESTAMP: 3113 case SO_BINTIME: 3114 case SO_NOSIGPIPE: 3115 case SO_NO_DDP: 3116 case SO_NO_OFFLOAD: 3117 case SO_RERROR: 3118 error = sooptcopyin(sopt, &optval, sizeof optval, 3119 sizeof optval); 3120 if (error) 3121 goto bad; 3122 SOCK_LOCK(so); 3123 if (optval) 3124 so->so_options |= sopt->sopt_name; 3125 else 3126 so->so_options &= ~sopt->sopt_name; 3127 SOCK_UNLOCK(so); 3128 break; 3129 3130 case SO_SETFIB: 3131 error = sooptcopyin(sopt, &optval, sizeof optval, 3132 sizeof optval); 3133 if (error) 3134 goto bad; 3135 3136 if (optval < 0 || optval >= rt_numfibs) { 3137 error = EINVAL; 3138 goto bad; 3139 } 3140 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 3141 (so->so_proto->pr_domain->dom_family == PF_INET6) || 3142 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 3143 so->so_fibnum = optval; 3144 else 3145 so->so_fibnum = 0; 3146 break; 3147 3148 case SO_USER_COOKIE: 3149 error = sooptcopyin(sopt, &val32, sizeof val32, 3150 sizeof val32); 3151 if (error) 3152 goto bad; 3153 so->so_user_cookie = val32; 3154 break; 3155 3156 case SO_SNDBUF: 3157 case SO_RCVBUF: 3158 case SO_SNDLOWAT: 3159 case SO_RCVLOWAT: 3160 error = sooptcopyin(sopt, &optval, sizeof optval, 3161 sizeof optval); 3162 if (error) 3163 goto bad; 3164 3165 /* 3166 * Values < 1 make no sense for any of these options, 3167 * so disallow them. 3168 */ 3169 if (optval < 1) { 3170 error = EINVAL; 3171 goto bad; 3172 } 3173 3174 error = sbsetopt(so, sopt->sopt_name, optval); 3175 break; 3176 3177 case SO_SNDTIMEO: 3178 case SO_RCVTIMEO: 3179 #ifdef COMPAT_FREEBSD32 3180 if (SV_CURPROC_FLAG(SV_ILP32)) { 3181 struct timeval32 tv32; 3182 3183 error = sooptcopyin(sopt, &tv32, sizeof tv32, 3184 sizeof tv32); 3185 CP(tv32, tv, tv_sec); 3186 CP(tv32, tv, tv_usec); 3187 } else 3188 #endif 3189 error = sooptcopyin(sopt, &tv, sizeof tv, 3190 sizeof tv); 3191 if (error) 3192 goto bad; 3193 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 3194 tv.tv_usec >= 1000000) { 3195 error = EDOM; 3196 goto bad; 3197 } 3198 if (tv.tv_sec > INT32_MAX) 3199 val = SBT_MAX; 3200 else 3201 val = tvtosbt(tv); 3202 switch (sopt->sopt_name) { 3203 case SO_SNDTIMEO: 3204 so->so_snd.sb_timeo = val; 3205 break; 3206 case SO_RCVTIMEO: 3207 so->so_rcv.sb_timeo = val; 3208 break; 3209 } 3210 break; 3211 3212 case SO_LABEL: 3213 #ifdef MAC 3214 error = sooptcopyin(sopt, &extmac, sizeof extmac, 3215 sizeof extmac); 3216 if (error) 3217 goto bad; 3218 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 3219 so, &extmac); 3220 #else 3221 error = EOPNOTSUPP; 3222 #endif 3223 break; 3224 3225 case SO_TS_CLOCK: 3226 error = sooptcopyin(sopt, &optval, sizeof optval, 3227 sizeof optval); 3228 if (error) 3229 goto bad; 3230 if (optval < 0 || optval > SO_TS_CLOCK_MAX) { 3231 error = EINVAL; 3232 goto bad; 3233 } 3234 so->so_ts_clock = optval; 3235 break; 3236 3237 case SO_MAX_PACING_RATE: 3238 error = sooptcopyin(sopt, &val32, sizeof(val32), 3239 sizeof(val32)); 3240 if (error) 3241 goto bad; 3242 so->so_max_pacing_rate = val32; 3243 break; 3244 3245 default: 3246 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3247 error = hhook_run_socket(so, sopt, 3248 HHOOK_SOCKET_OPT); 3249 else 3250 error = ENOPROTOOPT; 3251 break; 3252 } 3253 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 3254 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 3255 } 3256 bad: 3257 CURVNET_RESTORE(); 3258 return (error); 3259 } 3260 3261 /* 3262 * Helper routine for getsockopt. 3263 */ 3264 int 3265 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 3266 { 3267 int error; 3268 size_t valsize; 3269 3270 error = 0; 3271 3272 /* 3273 * Documented get behavior is that we always return a value, possibly 3274 * truncated to fit in the user's buffer. Traditional behavior is 3275 * that we always tell the user precisely how much we copied, rather 3276 * than something useful like the total amount we had available for 3277 * her. Note that this interface is not idempotent; the entire 3278 * answer must be generated ahead of time. 3279 */ 3280 valsize = min(len, sopt->sopt_valsize); 3281 sopt->sopt_valsize = valsize; 3282 if (sopt->sopt_val != NULL) { 3283 if (sopt->sopt_td != NULL) 3284 error = copyout(buf, sopt->sopt_val, valsize); 3285 else 3286 bcopy(buf, sopt->sopt_val, valsize); 3287 } 3288 return (error); 3289 } 3290 3291 int 3292 sogetopt(struct socket *so, struct sockopt *sopt) 3293 { 3294 int error, optval; 3295 struct linger l; 3296 struct timeval tv; 3297 #ifdef MAC 3298 struct mac extmac; 3299 #endif 3300 3301 CURVNET_SET(so->so_vnet); 3302 error = 0; 3303 if (sopt->sopt_level != SOL_SOCKET) { 3304 if (so->so_proto->pr_ctloutput != NULL) 3305 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3306 else 3307 error = ENOPROTOOPT; 3308 CURVNET_RESTORE(); 3309 return (error); 3310 } else { 3311 switch (sopt->sopt_name) { 3312 case SO_ACCEPTFILTER: 3313 error = accept_filt_getopt(so, sopt); 3314 break; 3315 3316 case SO_LINGER: 3317 SOCK_LOCK(so); 3318 l.l_onoff = so->so_options & SO_LINGER; 3319 l.l_linger = so->so_linger; 3320 SOCK_UNLOCK(so); 3321 error = sooptcopyout(sopt, &l, sizeof l); 3322 break; 3323 3324 case SO_USELOOPBACK: 3325 case SO_DONTROUTE: 3326 case SO_DEBUG: 3327 case SO_KEEPALIVE: 3328 case SO_REUSEADDR: 3329 case SO_REUSEPORT: 3330 case SO_REUSEPORT_LB: 3331 case SO_BROADCAST: 3332 case SO_OOBINLINE: 3333 case SO_ACCEPTCONN: 3334 case SO_TIMESTAMP: 3335 case SO_BINTIME: 3336 case SO_NOSIGPIPE: 3337 case SO_NO_DDP: 3338 case SO_NO_OFFLOAD: 3339 case SO_RERROR: 3340 optval = so->so_options & sopt->sopt_name; 3341 integer: 3342 error = sooptcopyout(sopt, &optval, sizeof optval); 3343 break; 3344 3345 case SO_DOMAIN: 3346 optval = so->so_proto->pr_domain->dom_family; 3347 goto integer; 3348 3349 case SO_TYPE: 3350 optval = so->so_type; 3351 goto integer; 3352 3353 case SO_PROTOCOL: 3354 optval = so->so_proto->pr_protocol; 3355 goto integer; 3356 3357 case SO_ERROR: 3358 SOCK_LOCK(so); 3359 if (so->so_error) { 3360 optval = so->so_error; 3361 so->so_error = 0; 3362 } else { 3363 optval = so->so_rerror; 3364 so->so_rerror = 0; 3365 } 3366 SOCK_UNLOCK(so); 3367 goto integer; 3368 3369 case SO_SNDBUF: 3370 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : 3371 so->so_snd.sb_hiwat; 3372 goto integer; 3373 3374 case SO_RCVBUF: 3375 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : 3376 so->so_rcv.sb_hiwat; 3377 goto integer; 3378 3379 case SO_SNDLOWAT: 3380 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : 3381 so->so_snd.sb_lowat; 3382 goto integer; 3383 3384 case SO_RCVLOWAT: 3385 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : 3386 so->so_rcv.sb_lowat; 3387 goto integer; 3388 3389 case SO_SNDTIMEO: 3390 case SO_RCVTIMEO: 3391 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 3392 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 3393 #ifdef COMPAT_FREEBSD32 3394 if (SV_CURPROC_FLAG(SV_ILP32)) { 3395 struct timeval32 tv32; 3396 3397 CP(tv, tv32, tv_sec); 3398 CP(tv, tv32, tv_usec); 3399 error = sooptcopyout(sopt, &tv32, sizeof tv32); 3400 } else 3401 #endif 3402 error = sooptcopyout(sopt, &tv, sizeof tv); 3403 break; 3404 3405 case SO_LABEL: 3406 #ifdef MAC 3407 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3408 sizeof(extmac)); 3409 if (error) 3410 goto bad; 3411 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 3412 so, &extmac); 3413 if (error) 3414 goto bad; 3415 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3416 #else 3417 error = EOPNOTSUPP; 3418 #endif 3419 break; 3420 3421 case SO_PEERLABEL: 3422 #ifdef MAC 3423 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3424 sizeof(extmac)); 3425 if (error) 3426 goto bad; 3427 error = mac_getsockopt_peerlabel( 3428 sopt->sopt_td->td_ucred, so, &extmac); 3429 if (error) 3430 goto bad; 3431 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3432 #else 3433 error = EOPNOTSUPP; 3434 #endif 3435 break; 3436 3437 case SO_LISTENQLIMIT: 3438 optval = SOLISTENING(so) ? so->sol_qlimit : 0; 3439 goto integer; 3440 3441 case SO_LISTENQLEN: 3442 optval = SOLISTENING(so) ? so->sol_qlen : 0; 3443 goto integer; 3444 3445 case SO_LISTENINCQLEN: 3446 optval = SOLISTENING(so) ? so->sol_incqlen : 0; 3447 goto integer; 3448 3449 case SO_TS_CLOCK: 3450 optval = so->so_ts_clock; 3451 goto integer; 3452 3453 case SO_MAX_PACING_RATE: 3454 optval = so->so_max_pacing_rate; 3455 goto integer; 3456 3457 default: 3458 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3459 error = hhook_run_socket(so, sopt, 3460 HHOOK_SOCKET_OPT); 3461 else 3462 error = ENOPROTOOPT; 3463 break; 3464 } 3465 } 3466 #ifdef MAC 3467 bad: 3468 #endif 3469 CURVNET_RESTORE(); 3470 return (error); 3471 } 3472 3473 int 3474 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 3475 { 3476 struct mbuf *m, *m_prev; 3477 int sopt_size = sopt->sopt_valsize; 3478 3479 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3480 if (m == NULL) 3481 return ENOBUFS; 3482 if (sopt_size > MLEN) { 3483 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 3484 if ((m->m_flags & M_EXT) == 0) { 3485 m_free(m); 3486 return ENOBUFS; 3487 } 3488 m->m_len = min(MCLBYTES, sopt_size); 3489 } else { 3490 m->m_len = min(MLEN, sopt_size); 3491 } 3492 sopt_size -= m->m_len; 3493 *mp = m; 3494 m_prev = m; 3495 3496 while (sopt_size) { 3497 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3498 if (m == NULL) { 3499 m_freem(*mp); 3500 return ENOBUFS; 3501 } 3502 if (sopt_size > MLEN) { 3503 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 3504 M_NOWAIT); 3505 if ((m->m_flags & M_EXT) == 0) { 3506 m_freem(m); 3507 m_freem(*mp); 3508 return ENOBUFS; 3509 } 3510 m->m_len = min(MCLBYTES, sopt_size); 3511 } else { 3512 m->m_len = min(MLEN, sopt_size); 3513 } 3514 sopt_size -= m->m_len; 3515 m_prev->m_next = m; 3516 m_prev = m; 3517 } 3518 return (0); 3519 } 3520 3521 int 3522 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 3523 { 3524 struct mbuf *m0 = m; 3525 3526 if (sopt->sopt_val == NULL) 3527 return (0); 3528 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3529 if (sopt->sopt_td != NULL) { 3530 int error; 3531 3532 error = copyin(sopt->sopt_val, mtod(m, char *), 3533 m->m_len); 3534 if (error != 0) { 3535 m_freem(m0); 3536 return(error); 3537 } 3538 } else 3539 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 3540 sopt->sopt_valsize -= m->m_len; 3541 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3542 m = m->m_next; 3543 } 3544 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 3545 panic("ip6_sooptmcopyin"); 3546 return (0); 3547 } 3548 3549 int 3550 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 3551 { 3552 struct mbuf *m0 = m; 3553 size_t valsize = 0; 3554 3555 if (sopt->sopt_val == NULL) 3556 return (0); 3557 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3558 if (sopt->sopt_td != NULL) { 3559 int error; 3560 3561 error = copyout(mtod(m, char *), sopt->sopt_val, 3562 m->m_len); 3563 if (error != 0) { 3564 m_freem(m0); 3565 return(error); 3566 } 3567 } else 3568 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 3569 sopt->sopt_valsize -= m->m_len; 3570 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3571 valsize += m->m_len; 3572 m = m->m_next; 3573 } 3574 if (m != NULL) { 3575 /* enough soopt buffer should be given from user-land */ 3576 m_freem(m0); 3577 return(EINVAL); 3578 } 3579 sopt->sopt_valsize = valsize; 3580 return (0); 3581 } 3582 3583 /* 3584 * sohasoutofband(): protocol notifies socket layer of the arrival of new 3585 * out-of-band data, which will then notify socket consumers. 3586 */ 3587 void 3588 sohasoutofband(struct socket *so) 3589 { 3590 3591 if (so->so_sigio != NULL) 3592 pgsigio(&so->so_sigio, SIGURG, 0); 3593 selwakeuppri(&so->so_rdsel, PSOCK); 3594 } 3595 3596 int 3597 sopoll(struct socket *so, int events, struct ucred *active_cred, 3598 struct thread *td) 3599 { 3600 3601 /* 3602 * We do not need to set or assert curvnet as long as everyone uses 3603 * sopoll_generic(). 3604 */ 3605 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, 3606 td)); 3607 } 3608 3609 int 3610 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3611 struct thread *td) 3612 { 3613 int revents; 3614 3615 SOCK_LOCK(so); 3616 if (SOLISTENING(so)) { 3617 if (!(events & (POLLIN | POLLRDNORM))) 3618 revents = 0; 3619 else if (!TAILQ_EMPTY(&so->sol_comp)) 3620 revents = events & (POLLIN | POLLRDNORM); 3621 else if ((events & POLLINIGNEOF) == 0 && so->so_error) 3622 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP; 3623 else { 3624 selrecord(td, &so->so_rdsel); 3625 revents = 0; 3626 } 3627 } else { 3628 revents = 0; 3629 SOCKBUF_LOCK(&so->so_snd); 3630 SOCKBUF_LOCK(&so->so_rcv); 3631 if (events & (POLLIN | POLLRDNORM)) 3632 if (soreadabledata(so)) 3633 revents |= events & (POLLIN | POLLRDNORM); 3634 if (events & (POLLOUT | POLLWRNORM)) 3635 if (sowriteable(so)) 3636 revents |= events & (POLLOUT | POLLWRNORM); 3637 if (events & (POLLPRI | POLLRDBAND)) 3638 if (so->so_oobmark || 3639 (so->so_rcv.sb_state & SBS_RCVATMARK)) 3640 revents |= events & (POLLPRI | POLLRDBAND); 3641 if ((events & POLLINIGNEOF) == 0) { 3642 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3643 revents |= events & (POLLIN | POLLRDNORM); 3644 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3645 revents |= POLLHUP; 3646 } 3647 } 3648 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3649 revents |= events & POLLRDHUP; 3650 if (revents == 0) { 3651 if (events & 3652 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) { 3653 selrecord(td, &so->so_rdsel); 3654 so->so_rcv.sb_flags |= SB_SEL; 3655 } 3656 if (events & (POLLOUT | POLLWRNORM)) { 3657 selrecord(td, &so->so_wrsel); 3658 so->so_snd.sb_flags |= SB_SEL; 3659 } 3660 } 3661 SOCKBUF_UNLOCK(&so->so_rcv); 3662 SOCKBUF_UNLOCK(&so->so_snd); 3663 } 3664 SOCK_UNLOCK(so); 3665 return (revents); 3666 } 3667 3668 int 3669 soo_kqfilter(struct file *fp, struct knote *kn) 3670 { 3671 struct socket *so = kn->kn_fp->f_data; 3672 struct sockbuf *sb; 3673 struct knlist *knl; 3674 3675 switch (kn->kn_filter) { 3676 case EVFILT_READ: 3677 kn->kn_fop = &soread_filtops; 3678 knl = &so->so_rdsel.si_note; 3679 sb = &so->so_rcv; 3680 break; 3681 case EVFILT_WRITE: 3682 kn->kn_fop = &sowrite_filtops; 3683 knl = &so->so_wrsel.si_note; 3684 sb = &so->so_snd; 3685 break; 3686 case EVFILT_EMPTY: 3687 kn->kn_fop = &soempty_filtops; 3688 knl = &so->so_wrsel.si_note; 3689 sb = &so->so_snd; 3690 break; 3691 default: 3692 return (EINVAL); 3693 } 3694 3695 SOCK_LOCK(so); 3696 if (SOLISTENING(so)) { 3697 knlist_add(knl, kn, 1); 3698 } else { 3699 SOCKBUF_LOCK(sb); 3700 knlist_add(knl, kn, 1); 3701 sb->sb_flags |= SB_KNOTE; 3702 SOCKBUF_UNLOCK(sb); 3703 } 3704 SOCK_UNLOCK(so); 3705 return (0); 3706 } 3707 3708 /* 3709 * Some routines that return EOPNOTSUPP for entry points that are not 3710 * supported by a protocol. Fill in as needed. 3711 */ 3712 int 3713 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 3714 { 3715 3716 return EOPNOTSUPP; 3717 } 3718 3719 int 3720 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job) 3721 { 3722 3723 return EOPNOTSUPP; 3724 } 3725 3726 int 3727 pru_attach_notsupp(struct socket *so, int proto, struct thread *td) 3728 { 3729 3730 return EOPNOTSUPP; 3731 } 3732 3733 int 3734 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3735 { 3736 3737 return EOPNOTSUPP; 3738 } 3739 3740 int 3741 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3742 struct thread *td) 3743 { 3744 3745 return EOPNOTSUPP; 3746 } 3747 3748 int 3749 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3750 { 3751 3752 return EOPNOTSUPP; 3753 } 3754 3755 int 3756 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3757 struct thread *td) 3758 { 3759 3760 return EOPNOTSUPP; 3761 } 3762 3763 int 3764 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 3765 { 3766 3767 return EOPNOTSUPP; 3768 } 3769 3770 int 3771 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 3772 struct ifnet *ifp, struct thread *td) 3773 { 3774 3775 return EOPNOTSUPP; 3776 } 3777 3778 int 3779 pru_disconnect_notsupp(struct socket *so) 3780 { 3781 3782 return EOPNOTSUPP; 3783 } 3784 3785 int 3786 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) 3787 { 3788 3789 return EOPNOTSUPP; 3790 } 3791 3792 int 3793 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) 3794 { 3795 3796 return EOPNOTSUPP; 3797 } 3798 3799 int 3800 pru_rcvd_notsupp(struct socket *so, int flags) 3801 { 3802 3803 return EOPNOTSUPP; 3804 } 3805 3806 int 3807 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 3808 { 3809 3810 return EOPNOTSUPP; 3811 } 3812 3813 int 3814 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, 3815 struct sockaddr *addr, struct mbuf *control, struct thread *td) 3816 { 3817 3818 if (control != NULL) 3819 m_freem(control); 3820 if ((flags & PRUS_NOTREADY) == 0) 3821 m_freem(m); 3822 return (EOPNOTSUPP); 3823 } 3824 3825 int 3826 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count) 3827 { 3828 3829 return (EOPNOTSUPP); 3830 } 3831 3832 /* 3833 * This isn't really a ``null'' operation, but it's the default one and 3834 * doesn't do anything destructive. 3835 */ 3836 int 3837 pru_sense_null(struct socket *so, struct stat *sb) 3838 { 3839 3840 sb->st_blksize = so->so_snd.sb_hiwat; 3841 return 0; 3842 } 3843 3844 int 3845 pru_shutdown_notsupp(struct socket *so) 3846 { 3847 3848 return EOPNOTSUPP; 3849 } 3850 3851 int 3852 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) 3853 { 3854 3855 return EOPNOTSUPP; 3856 } 3857 3858 int 3859 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 3860 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 3861 { 3862 3863 return EOPNOTSUPP; 3864 } 3865 3866 int 3867 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 3868 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 3869 { 3870 3871 return EOPNOTSUPP; 3872 } 3873 3874 int 3875 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, 3876 struct thread *td) 3877 { 3878 3879 return EOPNOTSUPP; 3880 } 3881 3882 static void 3883 filt_sordetach(struct knote *kn) 3884 { 3885 struct socket *so = kn->kn_fp->f_data; 3886 3887 so_rdknl_lock(so); 3888 knlist_remove(&so->so_rdsel.si_note, kn, 1); 3889 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note)) 3890 so->so_rcv.sb_flags &= ~SB_KNOTE; 3891 so_rdknl_unlock(so); 3892 } 3893 3894 /*ARGSUSED*/ 3895 static int 3896 filt_soread(struct knote *kn, long hint) 3897 { 3898 struct socket *so; 3899 3900 so = kn->kn_fp->f_data; 3901 3902 if (SOLISTENING(so)) { 3903 SOCK_LOCK_ASSERT(so); 3904 kn->kn_data = so->sol_qlen; 3905 if (so->so_error) { 3906 kn->kn_flags |= EV_EOF; 3907 kn->kn_fflags = so->so_error; 3908 return (1); 3909 } 3910 return (!TAILQ_EMPTY(&so->sol_comp)); 3911 } 3912 3913 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3914 3915 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 3916 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3917 kn->kn_flags |= EV_EOF; 3918 kn->kn_fflags = so->so_error; 3919 return (1); 3920 } else if (so->so_error || so->so_rerror) 3921 return (1); 3922 3923 if (kn->kn_sfflags & NOTE_LOWAT) { 3924 if (kn->kn_data >= kn->kn_sdata) 3925 return (1); 3926 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) 3927 return (1); 3928 3929 /* This hook returning non-zero indicates an event, not error */ 3930 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); 3931 } 3932 3933 static void 3934 filt_sowdetach(struct knote *kn) 3935 { 3936 struct socket *so = kn->kn_fp->f_data; 3937 3938 so_wrknl_lock(so); 3939 knlist_remove(&so->so_wrsel.si_note, kn, 1); 3940 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note)) 3941 so->so_snd.sb_flags &= ~SB_KNOTE; 3942 so_wrknl_unlock(so); 3943 } 3944 3945 /*ARGSUSED*/ 3946 static int 3947 filt_sowrite(struct knote *kn, long hint) 3948 { 3949 struct socket *so; 3950 3951 so = kn->kn_fp->f_data; 3952 3953 if (SOLISTENING(so)) 3954 return (0); 3955 3956 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3957 kn->kn_data = sbspace(&so->so_snd); 3958 3959 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); 3960 3961 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3962 kn->kn_flags |= EV_EOF; 3963 kn->kn_fflags = so->so_error; 3964 return (1); 3965 } else if (so->so_error) /* temporary udp error */ 3966 return (1); 3967 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3968 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3969 return (0); 3970 else if (kn->kn_sfflags & NOTE_LOWAT) 3971 return (kn->kn_data >= kn->kn_sdata); 3972 else 3973 return (kn->kn_data >= so->so_snd.sb_lowat); 3974 } 3975 3976 static int 3977 filt_soempty(struct knote *kn, long hint) 3978 { 3979 struct socket *so; 3980 3981 so = kn->kn_fp->f_data; 3982 3983 if (SOLISTENING(so)) 3984 return (1); 3985 3986 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3987 kn->kn_data = sbused(&so->so_snd); 3988 3989 if (kn->kn_data == 0) 3990 return (1); 3991 else 3992 return (0); 3993 } 3994 3995 int 3996 socheckuid(struct socket *so, uid_t uid) 3997 { 3998 3999 if (so == NULL) 4000 return (EPERM); 4001 if (so->so_cred->cr_uid != uid) 4002 return (EPERM); 4003 return (0); 4004 } 4005 4006 /* 4007 * These functions are used by protocols to notify the socket layer (and its 4008 * consumers) of state changes in the sockets driven by protocol-side events. 4009 */ 4010 4011 /* 4012 * Procedures to manipulate state flags of socket and do appropriate wakeups. 4013 * 4014 * Normal sequence from the active (originating) side is that 4015 * soisconnecting() is called during processing of connect() call, resulting 4016 * in an eventual call to soisconnected() if/when the connection is 4017 * established. When the connection is torn down soisdisconnecting() is 4018 * called during processing of disconnect() call, and soisdisconnected() is 4019 * called when the connection to the peer is totally severed. The semantics 4020 * of these routines are such that connectionless protocols can call 4021 * soisconnected() and soisdisconnected() only, bypassing the in-progress 4022 * calls when setting up a ``connection'' takes no time. 4023 * 4024 * From the passive side, a socket is created with two queues of sockets: 4025 * so_incomp for connections in progress and so_comp for connections already 4026 * made and awaiting user acceptance. As a protocol is preparing incoming 4027 * connections, it creates a socket structure queued on so_incomp by calling 4028 * sonewconn(). When the connection is established, soisconnected() is 4029 * called, and transfers the socket structure to so_comp, making it available 4030 * to accept(). 4031 * 4032 * If a socket is closed with sockets on either so_incomp or so_comp, these 4033 * sockets are dropped. 4034 * 4035 * If higher-level protocols are implemented in the kernel, the wakeups done 4036 * here will sometimes cause software-interrupt process scheduling. 4037 */ 4038 void 4039 soisconnecting(struct socket *so) 4040 { 4041 4042 SOCK_LOCK(so); 4043 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 4044 so->so_state |= SS_ISCONNECTING; 4045 SOCK_UNLOCK(so); 4046 } 4047 4048 void 4049 soisconnected(struct socket *so) 4050 { 4051 bool last __diagused; 4052 4053 SOCK_LOCK(so); 4054 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 4055 so->so_state |= SS_ISCONNECTED; 4056 4057 if (so->so_qstate == SQ_INCOMP) { 4058 struct socket *head = so->so_listen; 4059 int ret; 4060 4061 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so)); 4062 /* 4063 * Promoting a socket from incomplete queue to complete, we 4064 * need to go through reverse order of locking. We first do 4065 * trylock, and if that doesn't succeed, we go the hard way 4066 * leaving a reference and rechecking consistency after proper 4067 * locking. 4068 */ 4069 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) { 4070 soref(head); 4071 SOCK_UNLOCK(so); 4072 SOLISTEN_LOCK(head); 4073 SOCK_LOCK(so); 4074 if (__predict_false(head != so->so_listen)) { 4075 /* 4076 * The socket went off the listen queue, 4077 * should be lost race to close(2) of sol. 4078 * The socket is about to soabort(). 4079 */ 4080 SOCK_UNLOCK(so); 4081 sorele_locked(head); 4082 return; 4083 } 4084 last = refcount_release(&head->so_count); 4085 KASSERT(!last, ("%s: released last reference for %p", 4086 __func__, head)); 4087 } 4088 again: 4089 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 4090 TAILQ_REMOVE(&head->sol_incomp, so, so_list); 4091 head->sol_incqlen--; 4092 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 4093 head->sol_qlen++; 4094 so->so_qstate = SQ_COMP; 4095 SOCK_UNLOCK(so); 4096 solisten_wakeup(head); /* unlocks */ 4097 } else { 4098 SOCKBUF_LOCK(&so->so_rcv); 4099 soupcall_set(so, SO_RCV, 4100 head->sol_accept_filter->accf_callback, 4101 head->sol_accept_filter_arg); 4102 so->so_options &= ~SO_ACCEPTFILTER; 4103 ret = head->sol_accept_filter->accf_callback(so, 4104 head->sol_accept_filter_arg, M_NOWAIT); 4105 if (ret == SU_ISCONNECTED) { 4106 soupcall_clear(so, SO_RCV); 4107 SOCKBUF_UNLOCK(&so->so_rcv); 4108 goto again; 4109 } 4110 SOCKBUF_UNLOCK(&so->so_rcv); 4111 SOCK_UNLOCK(so); 4112 SOLISTEN_UNLOCK(head); 4113 } 4114 return; 4115 } 4116 SOCK_UNLOCK(so); 4117 wakeup(&so->so_timeo); 4118 sorwakeup(so); 4119 sowwakeup(so); 4120 } 4121 4122 void 4123 soisdisconnecting(struct socket *so) 4124 { 4125 4126 SOCK_LOCK(so); 4127 so->so_state &= ~SS_ISCONNECTING; 4128 so->so_state |= SS_ISDISCONNECTING; 4129 4130 if (!SOLISTENING(so)) { 4131 SOCKBUF_LOCK(&so->so_rcv); 4132 socantrcvmore_locked(so); 4133 SOCKBUF_LOCK(&so->so_snd); 4134 socantsendmore_locked(so); 4135 } 4136 SOCK_UNLOCK(so); 4137 wakeup(&so->so_timeo); 4138 } 4139 4140 void 4141 soisdisconnected(struct socket *so) 4142 { 4143 4144 SOCK_LOCK(so); 4145 4146 /* 4147 * There is at least one reader of so_state that does not 4148 * acquire socket lock, namely soreceive_generic(). Ensure 4149 * that it never sees all flags that track connection status 4150 * cleared, by ordering the update with a barrier semantic of 4151 * our release thread fence. 4152 */ 4153 so->so_state |= SS_ISDISCONNECTED; 4154 atomic_thread_fence_rel(); 4155 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 4156 4157 if (!SOLISTENING(so)) { 4158 SOCK_UNLOCK(so); 4159 SOCKBUF_LOCK(&so->so_rcv); 4160 socantrcvmore_locked(so); 4161 SOCKBUF_LOCK(&so->so_snd); 4162 sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); 4163 socantsendmore_locked(so); 4164 } else 4165 SOCK_UNLOCK(so); 4166 wakeup(&so->so_timeo); 4167 } 4168 4169 int 4170 soiolock(struct socket *so, struct sx *sx, int flags) 4171 { 4172 int error; 4173 4174 KASSERT((flags & SBL_VALID) == flags, 4175 ("soiolock: invalid flags %#x", flags)); 4176 4177 if ((flags & SBL_WAIT) != 0) { 4178 if ((flags & SBL_NOINTR) != 0) { 4179 sx_xlock(sx); 4180 } else { 4181 error = sx_xlock_sig(sx); 4182 if (error != 0) 4183 return (error); 4184 } 4185 } else if (!sx_try_xlock(sx)) { 4186 return (EWOULDBLOCK); 4187 } 4188 4189 if (__predict_false(SOLISTENING(so))) { 4190 sx_xunlock(sx); 4191 return (ENOTCONN); 4192 } 4193 return (0); 4194 } 4195 4196 void 4197 soiounlock(struct sx *sx) 4198 { 4199 sx_xunlock(sx); 4200 } 4201 4202 /* 4203 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 4204 */ 4205 struct sockaddr * 4206 sodupsockaddr(const struct sockaddr *sa, int mflags) 4207 { 4208 struct sockaddr *sa2; 4209 4210 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 4211 if (sa2) 4212 bcopy(sa, sa2, sa->sa_len); 4213 return sa2; 4214 } 4215 4216 /* 4217 * Register per-socket destructor. 4218 */ 4219 void 4220 sodtor_set(struct socket *so, so_dtor_t *func) 4221 { 4222 4223 SOCK_LOCK_ASSERT(so); 4224 so->so_dtor = func; 4225 } 4226 4227 /* 4228 * Register per-socket buffer upcalls. 4229 */ 4230 void 4231 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg) 4232 { 4233 struct sockbuf *sb; 4234 4235 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4236 4237 switch (which) { 4238 case SO_RCV: 4239 sb = &so->so_rcv; 4240 break; 4241 case SO_SND: 4242 sb = &so->so_snd; 4243 break; 4244 default: 4245 panic("soupcall_set: bad which"); 4246 } 4247 SOCKBUF_LOCK_ASSERT(sb); 4248 sb->sb_upcall = func; 4249 sb->sb_upcallarg = arg; 4250 sb->sb_flags |= SB_UPCALL; 4251 } 4252 4253 void 4254 soupcall_clear(struct socket *so, int which) 4255 { 4256 struct sockbuf *sb; 4257 4258 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4259 4260 switch (which) { 4261 case SO_RCV: 4262 sb = &so->so_rcv; 4263 break; 4264 case SO_SND: 4265 sb = &so->so_snd; 4266 break; 4267 default: 4268 panic("soupcall_clear: bad which"); 4269 } 4270 SOCKBUF_LOCK_ASSERT(sb); 4271 KASSERT(sb->sb_upcall != NULL, 4272 ("%s: so %p no upcall to clear", __func__, so)); 4273 sb->sb_upcall = NULL; 4274 sb->sb_upcallarg = NULL; 4275 sb->sb_flags &= ~SB_UPCALL; 4276 } 4277 4278 void 4279 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg) 4280 { 4281 4282 SOLISTEN_LOCK_ASSERT(so); 4283 so->sol_upcall = func; 4284 so->sol_upcallarg = arg; 4285 } 4286 4287 static void 4288 so_rdknl_lock(void *arg) 4289 { 4290 struct socket *so = arg; 4291 4292 if (SOLISTENING(so)) 4293 SOCK_LOCK(so); 4294 else 4295 SOCKBUF_LOCK(&so->so_rcv); 4296 } 4297 4298 static void 4299 so_rdknl_unlock(void *arg) 4300 { 4301 struct socket *so = arg; 4302 4303 if (SOLISTENING(so)) 4304 SOCK_UNLOCK(so); 4305 else 4306 SOCKBUF_UNLOCK(&so->so_rcv); 4307 } 4308 4309 static void 4310 so_rdknl_assert_lock(void *arg, int what) 4311 { 4312 struct socket *so = arg; 4313 4314 if (what == LA_LOCKED) { 4315 if (SOLISTENING(so)) 4316 SOCK_LOCK_ASSERT(so); 4317 else 4318 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 4319 } else { 4320 if (SOLISTENING(so)) 4321 SOCK_UNLOCK_ASSERT(so); 4322 else 4323 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); 4324 } 4325 } 4326 4327 static void 4328 so_wrknl_lock(void *arg) 4329 { 4330 struct socket *so = arg; 4331 4332 if (SOLISTENING(so)) 4333 SOCK_LOCK(so); 4334 else 4335 SOCKBUF_LOCK(&so->so_snd); 4336 } 4337 4338 static void 4339 so_wrknl_unlock(void *arg) 4340 { 4341 struct socket *so = arg; 4342 4343 if (SOLISTENING(so)) 4344 SOCK_UNLOCK(so); 4345 else 4346 SOCKBUF_UNLOCK(&so->so_snd); 4347 } 4348 4349 static void 4350 so_wrknl_assert_lock(void *arg, int what) 4351 { 4352 struct socket *so = arg; 4353 4354 if (what == LA_LOCKED) { 4355 if (SOLISTENING(so)) 4356 SOCK_LOCK_ASSERT(so); 4357 else 4358 SOCKBUF_LOCK_ASSERT(&so->so_snd); 4359 } else { 4360 if (SOLISTENING(so)) 4361 SOCK_UNLOCK_ASSERT(so); 4362 else 4363 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 4364 } 4365 } 4366 4367 /* 4368 * Create an external-format (``xsocket'') structure using the information in 4369 * the kernel-format socket structure pointed to by so. This is done to 4370 * reduce the spew of irrelevant information over this interface, to isolate 4371 * user code from changes in the kernel structure, and potentially to provide 4372 * information-hiding if we decide that some of this information should be 4373 * hidden from users. 4374 */ 4375 void 4376 sotoxsocket(struct socket *so, struct xsocket *xso) 4377 { 4378 4379 bzero(xso, sizeof(*xso)); 4380 xso->xso_len = sizeof *xso; 4381 xso->xso_so = (uintptr_t)so; 4382 xso->so_type = so->so_type; 4383 xso->so_options = so->so_options; 4384 xso->so_linger = so->so_linger; 4385 xso->so_state = so->so_state; 4386 xso->so_pcb = (uintptr_t)so->so_pcb; 4387 xso->xso_protocol = so->so_proto->pr_protocol; 4388 xso->xso_family = so->so_proto->pr_domain->dom_family; 4389 xso->so_timeo = so->so_timeo; 4390 xso->so_error = so->so_error; 4391 xso->so_uid = so->so_cred->cr_uid; 4392 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 4393 if (SOLISTENING(so)) { 4394 xso->so_qlen = so->sol_qlen; 4395 xso->so_incqlen = so->sol_incqlen; 4396 xso->so_qlimit = so->sol_qlimit; 4397 xso->so_oobmark = 0; 4398 } else { 4399 xso->so_state |= so->so_qstate; 4400 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0; 4401 xso->so_oobmark = so->so_oobmark; 4402 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 4403 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 4404 } 4405 } 4406 4407 struct sockbuf * 4408 so_sockbuf_rcv(struct socket *so) 4409 { 4410 4411 return (&so->so_rcv); 4412 } 4413 4414 struct sockbuf * 4415 so_sockbuf_snd(struct socket *so) 4416 { 4417 4418 return (&so->so_snd); 4419 } 4420 4421 int 4422 so_state_get(const struct socket *so) 4423 { 4424 4425 return (so->so_state); 4426 } 4427 4428 void 4429 so_state_set(struct socket *so, int val) 4430 { 4431 4432 so->so_state = val; 4433 } 4434 4435 int 4436 so_options_get(const struct socket *so) 4437 { 4438 4439 return (so->so_options); 4440 } 4441 4442 void 4443 so_options_set(struct socket *so, int val) 4444 { 4445 4446 so->so_options = val; 4447 } 4448 4449 int 4450 so_error_get(const struct socket *so) 4451 { 4452 4453 return (so->so_error); 4454 } 4455 4456 void 4457 so_error_set(struct socket *so, int val) 4458 { 4459 4460 so->so_error = val; 4461 } 4462 4463 int 4464 so_linger_get(const struct socket *so) 4465 { 4466 4467 return (so->so_linger); 4468 } 4469 4470 void 4471 so_linger_set(struct socket *so, int val) 4472 { 4473 4474 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz), 4475 ("%s: val %d out of range", __func__, val)); 4476 4477 so->so_linger = val; 4478 } 4479 4480 struct protosw * 4481 so_protosw_get(const struct socket *so) 4482 { 4483 4484 return (so->so_proto); 4485 } 4486 4487 void 4488 so_protosw_set(struct socket *so, struct protosw *val) 4489 { 4490 4491 so->so_proto = val; 4492 } 4493 4494 void 4495 so_sorwakeup(struct socket *so) 4496 { 4497 4498 sorwakeup(so); 4499 } 4500 4501 void 4502 so_sowwakeup(struct socket *so) 4503 { 4504 4505 sowwakeup(so); 4506 } 4507 4508 void 4509 so_sorwakeup_locked(struct socket *so) 4510 { 4511 4512 sorwakeup_locked(so); 4513 } 4514 4515 void 4516 so_sowwakeup_locked(struct socket *so) 4517 { 4518 4519 sowwakeup_locked(so); 4520 } 4521 4522 void 4523 so_lock(struct socket *so) 4524 { 4525 4526 SOCK_LOCK(so); 4527 } 4528 4529 void 4530 so_unlock(struct socket *so) 4531 { 4532 4533 SOCK_UNLOCK(so); 4534 } 4535