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