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 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket 256 * so_gencnt field. 257 */ 258 static struct mtx so_global_mtx; 259 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF); 260 261 /* 262 * General IPC sysctl name space, used by sockets and a variety of other IPC 263 * types. 264 */ 265 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 266 "IPC"); 267 268 /* 269 * Initialize the socket subsystem and set up the socket 270 * memory allocator. 271 */ 272 static uma_zone_t socket_zone; 273 int maxsockets; 274 275 static void 276 socket_zone_change(void *tag) 277 { 278 279 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 280 } 281 282 static void 283 socket_hhook_register(int subtype) 284 { 285 286 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype, 287 &V_socket_hhh[subtype], 288 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 289 printf("%s: WARNING: unable to register hook\n", __func__); 290 } 291 292 static void 293 socket_hhook_deregister(int subtype) 294 { 295 296 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0) 297 printf("%s: WARNING: unable to deregister hook\n", __func__); 298 } 299 300 static void 301 socket_init(void *tag) 302 { 303 304 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL, 305 NULL, NULL, UMA_ALIGN_PTR, 0); 306 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 307 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached"); 308 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL, 309 EVENTHANDLER_PRI_FIRST); 310 } 311 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL); 312 313 static void 314 socket_vnet_init(const void *unused __unused) 315 { 316 int i; 317 318 /* We expect a contiguous range */ 319 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 320 socket_hhook_register(i); 321 } 322 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 323 socket_vnet_init, NULL); 324 325 static void 326 socket_vnet_uninit(const void *unused __unused) 327 { 328 int i; 329 330 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 331 socket_hhook_deregister(i); 332 } 333 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 334 socket_vnet_uninit, NULL); 335 336 /* 337 * Initialise maxsockets. This SYSINIT must be run after 338 * tunable_mbinit(). 339 */ 340 static void 341 init_maxsockets(void *ignored) 342 { 343 344 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 345 maxsockets = imax(maxsockets, maxfiles); 346 } 347 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); 348 349 /* 350 * Sysctl to get and set the maximum global sockets limit. Notify protocols 351 * of the change so that they can update their dependent limits as required. 352 */ 353 static int 354 sysctl_maxsockets(SYSCTL_HANDLER_ARGS) 355 { 356 int error, newmaxsockets; 357 358 newmaxsockets = maxsockets; 359 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req); 360 if (error == 0 && req->newptr && newmaxsockets != maxsockets) { 361 if (newmaxsockets > maxsockets && 362 newmaxsockets <= maxfiles) { 363 maxsockets = newmaxsockets; 364 EVENTHANDLER_INVOKE(maxsockets_change); 365 } else 366 error = EINVAL; 367 } 368 return (error); 369 } 370 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, 371 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &maxsockets, 0, 372 sysctl_maxsockets, "IU", 373 "Maximum number of sockets available"); 374 375 /* 376 * Socket operation routines. These routines are called by the routines in 377 * sys_socket.c or from a system process, and implement the semantics of 378 * socket operations by switching out to the protocol specific routines. 379 */ 380 381 /* 382 * Get a socket structure from our zone, and initialize it. Note that it 383 * would probably be better to allocate socket and PCB at the same time, but 384 * I'm not convinced that all the protocols can be easily modified to do 385 * this. 386 * 387 * soalloc() returns a socket with a ref count of 0. 388 */ 389 static struct socket * 390 soalloc(struct vnet *vnet) 391 { 392 struct socket *so; 393 394 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO); 395 if (so == NULL) 396 return (NULL); 397 #ifdef MAC 398 if (mac_socket_init(so, M_NOWAIT) != 0) { 399 uma_zfree(socket_zone, so); 400 return (NULL); 401 } 402 #endif 403 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) { 404 uma_zfree(socket_zone, so); 405 return (NULL); 406 } 407 408 /* 409 * The socket locking protocol allows to lock 2 sockets at a time, 410 * however, the first one must be a listening socket. WITNESS lacks 411 * a feature to change class of an existing lock, so we use DUPOK. 412 */ 413 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK); 414 mtx_init(&so->so_snd_mtx, "so_snd", NULL, MTX_DEF); 415 mtx_init(&so->so_rcv_mtx, "so_rcv", NULL, MTX_DEF); 416 so->so_rcv.sb_sel = &so->so_rdsel; 417 so->so_snd.sb_sel = &so->so_wrsel; 418 sx_init(&so->so_snd_sx, "so_snd_sx"); 419 sx_init(&so->so_rcv_sx, "so_rcv_sx"); 420 TAILQ_INIT(&so->so_snd.sb_aiojobq); 421 TAILQ_INIT(&so->so_rcv.sb_aiojobq); 422 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so); 423 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so); 424 #ifdef VIMAGE 425 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p", 426 __func__, __LINE__, so)); 427 so->so_vnet = vnet; 428 #endif 429 /* We shouldn't need the so_global_mtx */ 430 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) { 431 /* Do we need more comprehensive error returns? */ 432 uma_zfree(socket_zone, so); 433 return (NULL); 434 } 435 mtx_lock(&so_global_mtx); 436 so->so_gencnt = ++so_gencnt; 437 ++numopensockets; 438 #ifdef VIMAGE 439 vnet->vnet_sockcnt++; 440 #endif 441 mtx_unlock(&so_global_mtx); 442 443 return (so); 444 } 445 446 /* 447 * Free the storage associated with a socket at the socket layer, tear down 448 * locks, labels, etc. All protocol state is assumed already to have been 449 * torn down (and possibly never set up) by the caller. 450 */ 451 void 452 sodealloc(struct socket *so) 453 { 454 455 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); 456 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL")); 457 458 mtx_lock(&so_global_mtx); 459 so->so_gencnt = ++so_gencnt; 460 --numopensockets; /* Could be below, but faster here. */ 461 #ifdef VIMAGE 462 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p", 463 __func__, __LINE__, so)); 464 so->so_vnet->vnet_sockcnt--; 465 #endif 466 mtx_unlock(&so_global_mtx); 467 #ifdef MAC 468 mac_socket_destroy(so); 469 #endif 470 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE); 471 472 khelp_destroy_osd(&so->osd); 473 if (SOLISTENING(so)) { 474 if (so->sol_accept_filter != NULL) 475 accept_filt_setopt(so, NULL); 476 } else { 477 if (so->so_rcv.sb_hiwat) 478 (void)chgsbsize(so->so_cred->cr_uidinfo, 479 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 480 if (so->so_snd.sb_hiwat) 481 (void)chgsbsize(so->so_cred->cr_uidinfo, 482 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 483 sx_destroy(&so->so_snd_sx); 484 sx_destroy(&so->so_rcv_sx); 485 mtx_destroy(&so->so_snd_mtx); 486 mtx_destroy(&so->so_rcv_mtx); 487 } 488 crfree(so->so_cred); 489 mtx_destroy(&so->so_lock); 490 uma_zfree(socket_zone, so); 491 } 492 493 /* 494 * socreate returns a socket with a ref count of 1 and a file descriptor 495 * reference. The socket should be closed with soclose(). 496 */ 497 int 498 socreate(int dom, struct socket **aso, int type, int proto, 499 struct ucred *cred, struct thread *td) 500 { 501 struct protosw *prp; 502 struct socket *so; 503 int error; 504 505 /* 506 * XXX: divert(4) historically abused PF_INET. Keep this compatibility 507 * shim until all applications have been updated. 508 */ 509 if (__predict_false(dom == PF_INET && type == SOCK_RAW && 510 proto == IPPROTO_DIVERT)) { 511 dom = PF_DIVERT; 512 printf("%s uses obsolete way to create divert(4) socket\n", 513 td->td_proc->p_comm); 514 } 515 516 prp = pffindproto(dom, type, proto); 517 if (prp == NULL) { 518 /* No support for domain. */ 519 if (pffinddomain(dom) == NULL) 520 return (EAFNOSUPPORT); 521 /* No support for socket type. */ 522 if (proto == 0 && type != 0) 523 return (EPROTOTYPE); 524 return (EPROTONOSUPPORT); 525 } 526 527 MPASS(prp->pr_attach); 528 529 if (IN_CAPABILITY_MODE(td) && (prp->pr_flags & PR_CAPATTACH) == 0) 530 return (ECAPMODE); 531 532 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0) 533 return (EPROTONOSUPPORT); 534 535 so = soalloc(CRED_TO_VNET(cred)); 536 if (so == NULL) 537 return (ENOBUFS); 538 539 so->so_type = type; 540 so->so_cred = crhold(cred); 541 if ((prp->pr_domain->dom_family == PF_INET) || 542 (prp->pr_domain->dom_family == PF_INET6) || 543 (prp->pr_domain->dom_family == PF_ROUTE)) 544 so->so_fibnum = td->td_proc->p_fibnum; 545 else 546 so->so_fibnum = 0; 547 so->so_proto = prp; 548 #ifdef MAC 549 mac_socket_create(cred, so); 550 #endif 551 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 552 so_rdknl_assert_lock); 553 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 554 so_wrknl_assert_lock); 555 if ((prp->pr_flags & PR_SOCKBUF) == 0) { 556 so->so_snd.sb_mtx = &so->so_snd_mtx; 557 so->so_rcv.sb_mtx = &so->so_rcv_mtx; 558 } 559 /* 560 * Auto-sizing of socket buffers is managed by the protocols and 561 * the appropriate flags must be set in the pru_attach function. 562 */ 563 CURVNET_SET(so->so_vnet); 564 error = prp->pr_attach(so, proto, td); 565 CURVNET_RESTORE(); 566 if (error) { 567 sodealloc(so); 568 return (error); 569 } 570 soref(so); 571 *aso = so; 572 return (0); 573 } 574 575 #ifdef REGRESSION 576 static int regression_sonewconn_earlytest = 1; 577 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW, 578 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test"); 579 #endif 580 581 static struct timeval overinterval = { 60, 0 }; 582 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW, 583 &overinterval, 584 "Delay in seconds between warnings for listen socket overflows"); 585 586 /* 587 * When an attempt at a new connection is noted on a socket which supports 588 * accept(2), the protocol has two options: 589 * 1) Call legacy sonewconn() function, which would call protocol attach 590 * method, same as used for socket(2). 591 * 2) Call solisten_clone(), do attach that is specific to a cloned connection, 592 * and then call solisten_enqueue(). 593 * 594 * Note: the ref count on the socket is 0 on return. 595 */ 596 struct socket * 597 solisten_clone(struct socket *head) 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 /* 703 * Preserve the historic listen queue overflow log 704 * message, that starts with "sonewconn:". It has 705 * been known to sysadmins for years and also test 706 * sys/kern/sonewconn_overflow checks for it. 707 */ 708 if (head->so_cred == 0) { 709 log(LOG_DEBUG, "sonewconn: pcb %p (%s): " 710 "Listen queue overflow: %i already in " 711 "queue awaiting acceptance (%d " 712 "occurrences)\n", head->so_pcb, 713 sbuf_data(&descrsb), 714 qlen, overcount); 715 } else { 716 log(LOG_DEBUG, "sonewconn: pcb %p (%s): " 717 "Listen queue overflow: " 718 "%i already in queue awaiting acceptance " 719 "(%d occurrences), euid %d, rgid %d, jail %s\n", 720 head->so_pcb, sbuf_data(&descrsb), qlen, 721 overcount, head->so_cred->cr_uid, 722 head->so_cred->cr_rgid, 723 head->so_cred->cr_prison ? 724 head->so_cred->cr_prison->pr_name : 725 "not_jailed"); 726 } 727 sbuf_delete(&descrsb); 728 729 overcount = 0; 730 } 731 732 return (NULL); 733 } 734 SOLISTEN_UNLOCK(head); 735 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL", 736 __func__, head)); 737 so = soalloc(head->so_vnet); 738 if (so == NULL) { 739 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 740 "limit reached or out of memory\n", 741 __func__, head->so_pcb); 742 return (NULL); 743 } 744 so->so_listen = head; 745 so->so_type = head->so_type; 746 so->so_options = head->so_options & ~SO_ACCEPTCONN; 747 so->so_linger = head->so_linger; 748 so->so_state = head->so_state; 749 so->so_fibnum = head->so_fibnum; 750 so->so_proto = head->so_proto; 751 so->so_cred = crhold(head->so_cred); 752 #ifdef MAC 753 mac_socket_newconn(head, so); 754 #endif 755 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 756 so_rdknl_assert_lock); 757 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 758 so_wrknl_assert_lock); 759 VNET_SO_ASSERT(head); 760 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) { 761 sodealloc(so); 762 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 763 __func__, head->so_pcb); 764 return (NULL); 765 } 766 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat; 767 so->so_snd.sb_lowat = head->sol_sbsnd_lowat; 768 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo; 769 so->so_snd.sb_timeo = head->sol_sbsnd_timeo; 770 so->so_rcv.sb_flags = head->sol_sbrcv_flags & SB_AUTOSIZE; 771 so->so_snd.sb_flags = head->sol_sbsnd_flags & SB_AUTOSIZE; 772 if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) { 773 so->so_snd.sb_mtx = &so->so_snd_mtx; 774 so->so_rcv.sb_mtx = &so->so_rcv_mtx; 775 } 776 777 return (so); 778 } 779 780 /* Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED. */ 781 struct socket * 782 sonewconn(struct socket *head, int connstatus) 783 { 784 struct socket *so; 785 786 if ((so = solisten_clone(head)) == NULL) 787 return (NULL); 788 789 if (so->so_proto->pr_attach(so, 0, NULL) != 0) { 790 sodealloc(so); 791 log(LOG_DEBUG, "%s: pcb %p: pr_attach() failed\n", 792 __func__, head->so_pcb); 793 return (NULL); 794 } 795 796 (void)solisten_enqueue(so, connstatus); 797 798 return (so); 799 } 800 801 /* 802 * Enqueue socket cloned by solisten_clone() to the listen queue of the 803 * listener it has been cloned from. 804 * 805 * Return 'true' if socket landed on complete queue, otherwise 'false'. 806 */ 807 bool 808 solisten_enqueue(struct socket *so, int connstatus) 809 { 810 struct socket *head = so->so_listen; 811 812 MPASS(refcount_load(&so->so_count) == 0); 813 refcount_init(&so->so_count, 1); 814 815 SOLISTEN_LOCK(head); 816 if (head->sol_accept_filter != NULL) 817 connstatus = 0; 818 so->so_state |= connstatus; 819 soref(head); /* A socket on (in)complete queue refs head. */ 820 if (connstatus) { 821 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 822 so->so_qstate = SQ_COMP; 823 head->sol_qlen++; 824 solisten_wakeup(head); /* unlocks */ 825 return (true); 826 } else { 827 /* 828 * Keep removing sockets from the head until there's room for 829 * us to insert on the tail. In pre-locking revisions, this 830 * was a simple if(), but as we could be racing with other 831 * threads and soabort() requires dropping locks, we must 832 * loop waiting for the condition to be true. 833 */ 834 while (head->sol_incqlen > head->sol_qlimit) { 835 struct socket *sp; 836 837 sp = TAILQ_FIRST(&head->sol_incomp); 838 TAILQ_REMOVE(&head->sol_incomp, sp, so_list); 839 head->sol_incqlen--; 840 SOCK_LOCK(sp); 841 sp->so_qstate = SQ_NONE; 842 sp->so_listen = NULL; 843 SOCK_UNLOCK(sp); 844 sorele_locked(head); /* does SOLISTEN_UNLOCK, head stays */ 845 soabort(sp); 846 SOLISTEN_LOCK(head); 847 } 848 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list); 849 so->so_qstate = SQ_INCOMP; 850 head->sol_incqlen++; 851 SOLISTEN_UNLOCK(head); 852 return (false); 853 } 854 } 855 856 #if defined(SCTP) || defined(SCTP_SUPPORT) 857 /* 858 * Socket part of sctp_peeloff(). Detach a new socket from an 859 * association. The new socket is returned with a reference. 860 * 861 * XXXGL: reduce copy-paste with solisten_clone(). 862 */ 863 struct socket * 864 sopeeloff(struct socket *head) 865 { 866 struct socket *so; 867 868 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p", 869 __func__, __LINE__, head)); 870 so = soalloc(head->so_vnet); 871 if (so == NULL) { 872 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 873 "limit reached or out of memory\n", 874 __func__, head->so_pcb); 875 return (NULL); 876 } 877 so->so_type = head->so_type; 878 so->so_options = head->so_options; 879 so->so_linger = head->so_linger; 880 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED; 881 so->so_fibnum = head->so_fibnum; 882 so->so_proto = head->so_proto; 883 so->so_cred = crhold(head->so_cred); 884 #ifdef MAC 885 mac_socket_newconn(head, so); 886 #endif 887 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 888 so_rdknl_assert_lock); 889 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 890 so_wrknl_assert_lock); 891 VNET_SO_ASSERT(head); 892 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { 893 sodealloc(so); 894 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 895 __func__, head->so_pcb); 896 return (NULL); 897 } 898 if ((*so->so_proto->pr_attach)(so, 0, NULL)) { 899 sodealloc(so); 900 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 901 __func__, head->so_pcb); 902 return (NULL); 903 } 904 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 905 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 906 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 907 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 908 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; 909 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; 910 911 soref(so); 912 913 return (so); 914 } 915 #endif /* SCTP */ 916 917 int 918 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 919 { 920 int error; 921 922 CURVNET_SET(so->so_vnet); 923 error = so->so_proto->pr_bind(so, nam, td); 924 CURVNET_RESTORE(); 925 return (error); 926 } 927 928 int 929 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 930 { 931 int error; 932 933 CURVNET_SET(so->so_vnet); 934 error = so->so_proto->pr_bindat(fd, so, nam, td); 935 CURVNET_RESTORE(); 936 return (error); 937 } 938 939 /* 940 * solisten() transitions a socket from a non-listening state to a listening 941 * state, but can also be used to update the listen queue depth on an 942 * existing listen socket. The protocol will call back into the sockets 943 * layer using solisten_proto_check() and solisten_proto() to check and set 944 * socket-layer listen state. Call backs are used so that the protocol can 945 * acquire both protocol and socket layer locks in whatever order is required 946 * by the protocol. 947 * 948 * Protocol implementors are advised to hold the socket lock across the 949 * socket-layer test and set to avoid races at the socket layer. 950 */ 951 int 952 solisten(struct socket *so, int backlog, struct thread *td) 953 { 954 int error; 955 956 CURVNET_SET(so->so_vnet); 957 error = so->so_proto->pr_listen(so, backlog, td); 958 CURVNET_RESTORE(); 959 return (error); 960 } 961 962 /* 963 * Prepare for a call to solisten_proto(). Acquire all socket buffer locks in 964 * order to interlock with socket I/O. 965 */ 966 int 967 solisten_proto_check(struct socket *so) 968 { 969 SOCK_LOCK_ASSERT(so); 970 971 if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 972 SS_ISDISCONNECTING)) != 0) 973 return (EINVAL); 974 975 /* 976 * Sleeping is not permitted here, so simply fail if userspace is 977 * attempting to transmit or receive on the socket. This kind of 978 * transient failure is not ideal, but it should occur only if userspace 979 * is misusing the socket interfaces. 980 */ 981 if (!sx_try_xlock(&so->so_snd_sx)) 982 return (EAGAIN); 983 if (!sx_try_xlock(&so->so_rcv_sx)) { 984 sx_xunlock(&so->so_snd_sx); 985 return (EAGAIN); 986 } 987 mtx_lock(&so->so_snd_mtx); 988 mtx_lock(&so->so_rcv_mtx); 989 990 /* Interlock with soo_aio_queue() and KTLS. */ 991 if (!SOLISTENING(so)) { 992 bool ktls; 993 994 #ifdef KERN_TLS 995 ktls = so->so_snd.sb_tls_info != NULL || 996 so->so_rcv.sb_tls_info != NULL; 997 #else 998 ktls = false; 999 #endif 1000 if (ktls || 1001 (so->so_snd.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0 || 1002 (so->so_rcv.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0) { 1003 solisten_proto_abort(so); 1004 return (EINVAL); 1005 } 1006 } 1007 1008 return (0); 1009 } 1010 1011 /* 1012 * Undo the setup done by solisten_proto_check(). 1013 */ 1014 void 1015 solisten_proto_abort(struct socket *so) 1016 { 1017 mtx_unlock(&so->so_snd_mtx); 1018 mtx_unlock(&so->so_rcv_mtx); 1019 sx_xunlock(&so->so_snd_sx); 1020 sx_xunlock(&so->so_rcv_sx); 1021 } 1022 1023 void 1024 solisten_proto(struct socket *so, int backlog) 1025 { 1026 int sbrcv_lowat, sbsnd_lowat; 1027 u_int sbrcv_hiwat, sbsnd_hiwat; 1028 short sbrcv_flags, sbsnd_flags; 1029 sbintime_t sbrcv_timeo, sbsnd_timeo; 1030 1031 SOCK_LOCK_ASSERT(so); 1032 KASSERT((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 1033 SS_ISDISCONNECTING)) == 0, 1034 ("%s: bad socket state %p", __func__, so)); 1035 1036 if (SOLISTENING(so)) 1037 goto listening; 1038 1039 /* 1040 * Change this socket to listening state. 1041 */ 1042 sbrcv_lowat = so->so_rcv.sb_lowat; 1043 sbsnd_lowat = so->so_snd.sb_lowat; 1044 sbrcv_hiwat = so->so_rcv.sb_hiwat; 1045 sbsnd_hiwat = so->so_snd.sb_hiwat; 1046 sbrcv_flags = so->so_rcv.sb_flags; 1047 sbsnd_flags = so->so_snd.sb_flags; 1048 sbrcv_timeo = so->so_rcv.sb_timeo; 1049 sbsnd_timeo = so->so_snd.sb_timeo; 1050 1051 sbdestroy(so, SO_SND); 1052 sbdestroy(so, SO_RCV); 1053 1054 #ifdef INVARIANTS 1055 bzero(&so->so_rcv, 1056 sizeof(struct socket) - offsetof(struct socket, so_rcv)); 1057 #endif 1058 1059 so->sol_sbrcv_lowat = sbrcv_lowat; 1060 so->sol_sbsnd_lowat = sbsnd_lowat; 1061 so->sol_sbrcv_hiwat = sbrcv_hiwat; 1062 so->sol_sbsnd_hiwat = sbsnd_hiwat; 1063 so->sol_sbrcv_flags = sbrcv_flags; 1064 so->sol_sbsnd_flags = sbsnd_flags; 1065 so->sol_sbrcv_timeo = sbrcv_timeo; 1066 so->sol_sbsnd_timeo = sbsnd_timeo; 1067 1068 so->sol_qlen = so->sol_incqlen = 0; 1069 TAILQ_INIT(&so->sol_incomp); 1070 TAILQ_INIT(&so->sol_comp); 1071 1072 so->sol_accept_filter = NULL; 1073 so->sol_accept_filter_arg = NULL; 1074 so->sol_accept_filter_str = NULL; 1075 1076 so->sol_upcall = NULL; 1077 so->sol_upcallarg = NULL; 1078 1079 so->so_options |= SO_ACCEPTCONN; 1080 1081 listening: 1082 if (backlog < 0 || backlog > somaxconn) 1083 backlog = somaxconn; 1084 so->sol_qlimit = backlog; 1085 1086 mtx_unlock(&so->so_snd_mtx); 1087 mtx_unlock(&so->so_rcv_mtx); 1088 sx_xunlock(&so->so_snd_sx); 1089 sx_xunlock(&so->so_rcv_sx); 1090 } 1091 1092 /* 1093 * Wakeup listeners/subsystems once we have a complete connection. 1094 * Enters with lock, returns unlocked. 1095 */ 1096 void 1097 solisten_wakeup(struct socket *sol) 1098 { 1099 1100 if (sol->sol_upcall != NULL) 1101 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT); 1102 else { 1103 selwakeuppri(&sol->so_rdsel, PSOCK); 1104 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0); 1105 } 1106 SOLISTEN_UNLOCK(sol); 1107 wakeup_one(&sol->sol_comp); 1108 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL) 1109 pgsigio(&sol->so_sigio, SIGIO, 0); 1110 } 1111 1112 /* 1113 * Return single connection off a listening socket queue. Main consumer of 1114 * the function is kern_accept4(). Some modules, that do their own accept 1115 * management also use the function. The socket reference held by the 1116 * listen queue is handed to the caller. 1117 * 1118 * Listening socket must be locked on entry and is returned unlocked on 1119 * return. 1120 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT. 1121 */ 1122 int 1123 solisten_dequeue(struct socket *head, struct socket **ret, int flags) 1124 { 1125 struct socket *so; 1126 int error; 1127 1128 SOLISTEN_LOCK_ASSERT(head); 1129 1130 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) && 1131 head->so_error == 0) { 1132 error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH, 1133 "accept", 0); 1134 if (error != 0) { 1135 SOLISTEN_UNLOCK(head); 1136 return (error); 1137 } 1138 } 1139 if (head->so_error) { 1140 error = head->so_error; 1141 head->so_error = 0; 1142 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp)) 1143 error = EWOULDBLOCK; 1144 else 1145 error = 0; 1146 if (error) { 1147 SOLISTEN_UNLOCK(head); 1148 return (error); 1149 } 1150 so = TAILQ_FIRST(&head->sol_comp); 1151 SOCK_LOCK(so); 1152 KASSERT(so->so_qstate == SQ_COMP, 1153 ("%s: so %p not SQ_COMP", __func__, so)); 1154 head->sol_qlen--; 1155 so->so_qstate = SQ_NONE; 1156 so->so_listen = NULL; 1157 TAILQ_REMOVE(&head->sol_comp, so, so_list); 1158 if (flags & ACCEPT4_INHERIT) 1159 so->so_state |= (head->so_state & SS_NBIO); 1160 else 1161 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0; 1162 SOCK_UNLOCK(so); 1163 sorele_locked(head); 1164 1165 *ret = so; 1166 return (0); 1167 } 1168 1169 /* 1170 * Free socket upon release of the very last reference. 1171 */ 1172 static void 1173 sofree(struct socket *so) 1174 { 1175 struct protosw *pr = so->so_proto; 1176 1177 SOCK_LOCK_ASSERT(so); 1178 KASSERT(refcount_load(&so->so_count) == 0, 1179 ("%s: so %p has references", __func__, so)); 1180 KASSERT(SOLISTENING(so) || so->so_qstate == SQ_NONE, 1181 ("%s: so %p is on listen queue", __func__, so)); 1182 1183 SOCK_UNLOCK(so); 1184 1185 if (so->so_dtor != NULL) 1186 so->so_dtor(so); 1187 1188 VNET_SO_ASSERT(so); 1189 if ((pr->pr_flags & PR_RIGHTS) && !SOLISTENING(so)) { 1190 MPASS(pr->pr_domain->dom_dispose != NULL); 1191 (*pr->pr_domain->dom_dispose)(so); 1192 } 1193 if (pr->pr_detach != NULL) 1194 pr->pr_detach(so); 1195 1196 /* 1197 * From this point on, we assume that no other references to this 1198 * socket exist anywhere else in the stack. Therefore, no locks need 1199 * to be acquired or held. 1200 */ 1201 if (!(pr->pr_flags & PR_SOCKBUF) && !SOLISTENING(so)) { 1202 sbdestroy(so, SO_SND); 1203 sbdestroy(so, SO_RCV); 1204 } 1205 seldrain(&so->so_rdsel); 1206 seldrain(&so->so_wrsel); 1207 knlist_destroy(&so->so_rdsel.si_note); 1208 knlist_destroy(&so->so_wrsel.si_note); 1209 sodealloc(so); 1210 } 1211 1212 /* 1213 * Release a reference on a socket while holding the socket lock. 1214 * Unlocks the socket lock before returning. 1215 */ 1216 void 1217 sorele_locked(struct socket *so) 1218 { 1219 SOCK_LOCK_ASSERT(so); 1220 if (refcount_release(&so->so_count)) 1221 sofree(so); 1222 else 1223 SOCK_UNLOCK(so); 1224 } 1225 1226 /* 1227 * Close a socket on last file table reference removal. Initiate disconnect 1228 * if connected. Free socket when disconnect complete. 1229 * 1230 * This function will sorele() the socket. Note that soclose() may be called 1231 * prior to the ref count reaching zero. The actual socket structure will 1232 * not be freed until the ref count reaches zero. 1233 */ 1234 int 1235 soclose(struct socket *so) 1236 { 1237 struct accept_queue lqueue; 1238 int error = 0; 1239 bool listening, last __diagused; 1240 1241 CURVNET_SET(so->so_vnet); 1242 funsetown(&so->so_sigio); 1243 if (so->so_state & SS_ISCONNECTED) { 1244 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 1245 error = sodisconnect(so); 1246 if (error) { 1247 if (error == ENOTCONN) 1248 error = 0; 1249 goto drop; 1250 } 1251 } 1252 1253 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) { 1254 if ((so->so_state & SS_ISDISCONNECTING) && 1255 (so->so_state & SS_NBIO)) 1256 goto drop; 1257 while (so->so_state & SS_ISCONNECTED) { 1258 error = tsleep(&so->so_timeo, 1259 PSOCK | PCATCH, "soclos", 1260 so->so_linger * hz); 1261 if (error) 1262 break; 1263 } 1264 } 1265 } 1266 1267 drop: 1268 if (so->so_proto->pr_close != NULL) 1269 so->so_proto->pr_close(so); 1270 1271 SOCK_LOCK(so); 1272 if ((listening = SOLISTENING(so))) { 1273 struct socket *sp; 1274 1275 TAILQ_INIT(&lqueue); 1276 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list); 1277 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list); 1278 1279 so->sol_qlen = so->sol_incqlen = 0; 1280 1281 TAILQ_FOREACH(sp, &lqueue, so_list) { 1282 SOCK_LOCK(sp); 1283 sp->so_qstate = SQ_NONE; 1284 sp->so_listen = NULL; 1285 SOCK_UNLOCK(sp); 1286 last = refcount_release(&so->so_count); 1287 KASSERT(!last, ("%s: released last reference for %p", 1288 __func__, so)); 1289 } 1290 } 1291 sorele_locked(so); 1292 if (listening) { 1293 struct socket *sp, *tsp; 1294 1295 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) 1296 soabort(sp); 1297 } 1298 CURVNET_RESTORE(); 1299 return (error); 1300 } 1301 1302 /* 1303 * soabort() is used to abruptly tear down a connection, such as when a 1304 * resource limit is reached (listen queue depth exceeded), or if a listen 1305 * socket is closed while there are sockets waiting to be accepted. 1306 * 1307 * This interface is tricky, because it is called on an unreferenced socket, 1308 * and must be called only by a thread that has actually removed the socket 1309 * from the listen queue it was on. Likely this thread holds the last 1310 * reference on the socket and soabort() will proceed with sofree(). But 1311 * it might be not the last, as the sockets on the listen queues are seen 1312 * from the protocol side. 1313 * 1314 * This interface will call into the protocol code, so must not be called 1315 * with any socket locks held. Protocols do call it while holding their own 1316 * recursible protocol mutexes, but this is something that should be subject 1317 * to review in the future. 1318 * 1319 * Usually socket should have a single reference left, but this is not a 1320 * requirement. In the past, when we have had named references for file 1321 * descriptor and protocol, we asserted that none of them are being held. 1322 */ 1323 void 1324 soabort(struct socket *so) 1325 { 1326 1327 VNET_SO_ASSERT(so); 1328 1329 if (so->so_proto->pr_abort != NULL) 1330 so->so_proto->pr_abort(so); 1331 SOCK_LOCK(so); 1332 sorele_locked(so); 1333 } 1334 1335 int 1336 soaccept(struct socket *so, struct sockaddr **nam) 1337 { 1338 int error; 1339 1340 CURVNET_SET(so->so_vnet); 1341 error = so->so_proto->pr_accept(so, nam); 1342 CURVNET_RESTORE(); 1343 return (error); 1344 } 1345 1346 int 1347 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 1348 { 1349 1350 return (soconnectat(AT_FDCWD, so, nam, td)); 1351 } 1352 1353 int 1354 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 1355 { 1356 int error; 1357 1358 CURVNET_SET(so->so_vnet); 1359 1360 /* 1361 * If protocol is connection-based, can only connect once. 1362 * Otherwise, if connected, try to disconnect first. This allows 1363 * user to disconnect by connecting to, e.g., a null address. 1364 * 1365 * Note, this check is racy and may need to be re-evaluated at the 1366 * protocol layer. 1367 */ 1368 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 1369 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 1370 (error = sodisconnect(so)))) { 1371 error = EISCONN; 1372 } else { 1373 /* 1374 * Prevent accumulated error from previous connection from 1375 * biting us. 1376 */ 1377 so->so_error = 0; 1378 if (fd == AT_FDCWD) { 1379 error = so->so_proto->pr_connect(so, nam, td); 1380 } else { 1381 error = so->so_proto->pr_connectat(fd, so, nam, td); 1382 } 1383 } 1384 CURVNET_RESTORE(); 1385 1386 return (error); 1387 } 1388 1389 int 1390 soconnect2(struct socket *so1, struct socket *so2) 1391 { 1392 int error; 1393 1394 CURVNET_SET(so1->so_vnet); 1395 error = so1->so_proto->pr_connect2(so1, so2); 1396 CURVNET_RESTORE(); 1397 return (error); 1398 } 1399 1400 int 1401 sodisconnect(struct socket *so) 1402 { 1403 int error; 1404 1405 if ((so->so_state & SS_ISCONNECTED) == 0) 1406 return (ENOTCONN); 1407 if (so->so_state & SS_ISDISCONNECTING) 1408 return (EALREADY); 1409 VNET_SO_ASSERT(so); 1410 error = so->so_proto->pr_disconnect(so); 1411 return (error); 1412 } 1413 1414 int 1415 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1416 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1417 { 1418 long space; 1419 ssize_t resid; 1420 int clen = 0, error, dontroute; 1421 1422 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 1423 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 1424 ("sosend_dgram: !PR_ATOMIC")); 1425 1426 if (uio != NULL) 1427 resid = uio->uio_resid; 1428 else 1429 resid = top->m_pkthdr.len; 1430 /* 1431 * In theory resid should be unsigned. However, space must be 1432 * signed, as it might be less than 0 if we over-committed, and we 1433 * must use a signed comparison of space and resid. On the other 1434 * hand, a negative resid causes us to loop sending 0-length 1435 * segments to the protocol. 1436 */ 1437 if (resid < 0) { 1438 error = EINVAL; 1439 goto out; 1440 } 1441 1442 dontroute = 1443 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 1444 if (td != NULL) 1445 td->td_ru.ru_msgsnd++; 1446 if (control != NULL) 1447 clen = control->m_len; 1448 1449 SOCKBUF_LOCK(&so->so_snd); 1450 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1451 SOCKBUF_UNLOCK(&so->so_snd); 1452 error = EPIPE; 1453 goto out; 1454 } 1455 if (so->so_error) { 1456 error = so->so_error; 1457 so->so_error = 0; 1458 SOCKBUF_UNLOCK(&so->so_snd); 1459 goto out; 1460 } 1461 if ((so->so_state & SS_ISCONNECTED) == 0) { 1462 /* 1463 * `sendto' and `sendmsg' is allowed on a connection-based 1464 * socket if it supports implied connect. Return ENOTCONN if 1465 * not connected and no address is supplied. 1466 */ 1467 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1468 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1469 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1470 !(resid == 0 && clen != 0)) { 1471 SOCKBUF_UNLOCK(&so->so_snd); 1472 error = ENOTCONN; 1473 goto out; 1474 } 1475 } else if (addr == NULL) { 1476 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1477 error = ENOTCONN; 1478 else 1479 error = EDESTADDRREQ; 1480 SOCKBUF_UNLOCK(&so->so_snd); 1481 goto out; 1482 } 1483 } 1484 1485 /* 1486 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1487 * problem and need fixing. 1488 */ 1489 space = sbspace(&so->so_snd); 1490 if (flags & MSG_OOB) 1491 space += 1024; 1492 space -= clen; 1493 SOCKBUF_UNLOCK(&so->so_snd); 1494 if (resid > space) { 1495 error = EMSGSIZE; 1496 goto out; 1497 } 1498 if (uio == NULL) { 1499 resid = 0; 1500 if (flags & MSG_EOR) 1501 top->m_flags |= M_EOR; 1502 } else { 1503 /* 1504 * Copy the data from userland into a mbuf chain. 1505 * If no data is to be copied in, a single empty mbuf 1506 * is returned. 1507 */ 1508 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1509 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1510 if (top == NULL) { 1511 error = EFAULT; /* only possible error */ 1512 goto out; 1513 } 1514 space -= resid - uio->uio_resid; 1515 resid = uio->uio_resid; 1516 } 1517 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1518 /* 1519 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1520 * than with. 1521 */ 1522 if (dontroute) { 1523 SOCK_LOCK(so); 1524 so->so_options |= SO_DONTROUTE; 1525 SOCK_UNLOCK(so); 1526 } 1527 /* 1528 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1529 * of date. We could have received a reset packet in an interrupt or 1530 * maybe we slept while doing page faults in uiomove() etc. We could 1531 * probably recheck again inside the locking protection here, but 1532 * there are probably other places that this also happens. We must 1533 * rethink this. 1534 */ 1535 VNET_SO_ASSERT(so); 1536 error = so->so_proto->pr_send(so, (flags & MSG_OOB) ? PRUS_OOB : 1537 /* 1538 * If the user set MSG_EOF, the protocol understands this flag and 1539 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1540 */ 1541 ((flags & MSG_EOF) && 1542 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1543 (resid <= 0)) ? 1544 PRUS_EOF : 1545 /* If there is more to send set PRUS_MORETOCOME */ 1546 (flags & MSG_MORETOCOME) || 1547 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1548 top, addr, control, td); 1549 if (dontroute) { 1550 SOCK_LOCK(so); 1551 so->so_options &= ~SO_DONTROUTE; 1552 SOCK_UNLOCK(so); 1553 } 1554 clen = 0; 1555 control = NULL; 1556 top = NULL; 1557 out: 1558 if (top != NULL) 1559 m_freem(top); 1560 if (control != NULL) 1561 m_freem(control); 1562 return (error); 1563 } 1564 1565 /* 1566 * Send on a socket. If send must go all at once and message is larger than 1567 * send buffering, then hard error. Lock against other senders. If must go 1568 * all at once and not enough room now, then inform user that this would 1569 * block and do nothing. Otherwise, if nonblocking, send as much as 1570 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1571 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1572 * in mbuf chain must be small enough to send all at once. 1573 * 1574 * Returns nonzero on error, timeout or signal; callers must check for short 1575 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1576 * on return. 1577 */ 1578 int 1579 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1580 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1581 { 1582 long space; 1583 ssize_t resid; 1584 int clen = 0, error, dontroute; 1585 int atomic = sosendallatonce(so) || top; 1586 int pr_send_flag; 1587 #ifdef KERN_TLS 1588 struct ktls_session *tls; 1589 int tls_enq_cnt, tls_send_flag; 1590 uint8_t tls_rtype; 1591 1592 tls = NULL; 1593 tls_rtype = TLS_RLTYPE_APP; 1594 #endif 1595 if (uio != NULL) 1596 resid = uio->uio_resid; 1597 else if ((top->m_flags & M_PKTHDR) != 0) 1598 resid = top->m_pkthdr.len; 1599 else 1600 resid = m_length(top, NULL); 1601 /* 1602 * In theory resid should be unsigned. However, space must be 1603 * signed, as it might be less than 0 if we over-committed, and we 1604 * must use a signed comparison of space and resid. On the other 1605 * hand, a negative resid causes us to loop sending 0-length 1606 * segments to the protocol. 1607 * 1608 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1609 * type sockets since that's an error. 1610 */ 1611 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1612 error = EINVAL; 1613 goto out; 1614 } 1615 1616 dontroute = 1617 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1618 (so->so_proto->pr_flags & PR_ATOMIC); 1619 if (td != NULL) 1620 td->td_ru.ru_msgsnd++; 1621 if (control != NULL) 1622 clen = control->m_len; 1623 1624 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 1625 if (error) 1626 goto out; 1627 1628 #ifdef KERN_TLS 1629 tls_send_flag = 0; 1630 tls = ktls_hold(so->so_snd.sb_tls_info); 1631 if (tls != NULL) { 1632 if (tls->mode == TCP_TLS_MODE_SW) 1633 tls_send_flag = PRUS_NOTREADY; 1634 1635 if (control != NULL) { 1636 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1637 1638 if (clen >= sizeof(*cm) && 1639 cm->cmsg_type == TLS_SET_RECORD_TYPE) { 1640 tls_rtype = *((uint8_t *)CMSG_DATA(cm)); 1641 clen = 0; 1642 m_freem(control); 1643 control = NULL; 1644 atomic = 1; 1645 } 1646 } 1647 1648 if (resid == 0 && !ktls_permit_empty_frames(tls)) { 1649 error = EINVAL; 1650 goto release; 1651 } 1652 } 1653 #endif 1654 1655 restart: 1656 do { 1657 SOCKBUF_LOCK(&so->so_snd); 1658 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1659 SOCKBUF_UNLOCK(&so->so_snd); 1660 error = EPIPE; 1661 goto release; 1662 } 1663 if (so->so_error) { 1664 error = so->so_error; 1665 so->so_error = 0; 1666 SOCKBUF_UNLOCK(&so->so_snd); 1667 goto release; 1668 } 1669 if ((so->so_state & SS_ISCONNECTED) == 0) { 1670 /* 1671 * `sendto' and `sendmsg' is allowed on a connection- 1672 * based socket if it supports implied connect. 1673 * Return ENOTCONN if not connected and no address is 1674 * supplied. 1675 */ 1676 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1677 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1678 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1679 !(resid == 0 && clen != 0)) { 1680 SOCKBUF_UNLOCK(&so->so_snd); 1681 error = ENOTCONN; 1682 goto release; 1683 } 1684 } else if (addr == NULL) { 1685 SOCKBUF_UNLOCK(&so->so_snd); 1686 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1687 error = ENOTCONN; 1688 else 1689 error = EDESTADDRREQ; 1690 goto release; 1691 } 1692 } 1693 space = sbspace(&so->so_snd); 1694 if (flags & MSG_OOB) 1695 space += 1024; 1696 if ((atomic && resid > so->so_snd.sb_hiwat) || 1697 clen > so->so_snd.sb_hiwat) { 1698 SOCKBUF_UNLOCK(&so->so_snd); 1699 error = EMSGSIZE; 1700 goto release; 1701 } 1702 if (space < resid + clen && 1703 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1704 if ((so->so_state & SS_NBIO) || 1705 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) { 1706 SOCKBUF_UNLOCK(&so->so_snd); 1707 error = EWOULDBLOCK; 1708 goto release; 1709 } 1710 error = sbwait(so, SO_SND); 1711 SOCKBUF_UNLOCK(&so->so_snd); 1712 if (error) 1713 goto release; 1714 goto restart; 1715 } 1716 SOCKBUF_UNLOCK(&so->so_snd); 1717 space -= clen; 1718 do { 1719 if (uio == NULL) { 1720 resid = 0; 1721 if (flags & MSG_EOR) 1722 top->m_flags |= M_EOR; 1723 #ifdef KERN_TLS 1724 if (tls != NULL) { 1725 ktls_frame(top, tls, &tls_enq_cnt, 1726 tls_rtype); 1727 tls_rtype = TLS_RLTYPE_APP; 1728 } 1729 #endif 1730 } else { 1731 /* 1732 * Copy the data from userland into a mbuf 1733 * chain. If resid is 0, which can happen 1734 * only if we have control to send, then 1735 * a single empty mbuf is returned. This 1736 * is a workaround to prevent protocol send 1737 * methods to panic. 1738 */ 1739 #ifdef KERN_TLS 1740 if (tls != NULL) { 1741 top = m_uiotombuf(uio, M_WAITOK, space, 1742 tls->params.max_frame_len, 1743 M_EXTPG | 1744 ((flags & MSG_EOR) ? M_EOR : 0)); 1745 if (top != NULL) { 1746 ktls_frame(top, tls, 1747 &tls_enq_cnt, tls_rtype); 1748 } 1749 tls_rtype = TLS_RLTYPE_APP; 1750 } else 1751 #endif 1752 top = m_uiotombuf(uio, M_WAITOK, space, 1753 (atomic ? max_hdr : 0), 1754 (atomic ? M_PKTHDR : 0) | 1755 ((flags & MSG_EOR) ? M_EOR : 0)); 1756 if (top == NULL) { 1757 error = EFAULT; /* only possible error */ 1758 goto release; 1759 } 1760 space -= resid - uio->uio_resid; 1761 resid = uio->uio_resid; 1762 } 1763 if (dontroute) { 1764 SOCK_LOCK(so); 1765 so->so_options |= SO_DONTROUTE; 1766 SOCK_UNLOCK(so); 1767 } 1768 /* 1769 * XXX all the SBS_CANTSENDMORE checks previously 1770 * done could be out of date. We could have received 1771 * a reset packet in an interrupt or maybe we slept 1772 * while doing page faults in uiomove() etc. We 1773 * could probably recheck again inside the locking 1774 * protection here, but there are probably other 1775 * places that this also happens. We must rethink 1776 * this. 1777 */ 1778 VNET_SO_ASSERT(so); 1779 1780 pr_send_flag = (flags & MSG_OOB) ? PRUS_OOB : 1781 /* 1782 * If the user set MSG_EOF, the protocol understands 1783 * this flag and nothing left to send then use 1784 * PRU_SEND_EOF instead of PRU_SEND. 1785 */ 1786 ((flags & MSG_EOF) && 1787 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1788 (resid <= 0)) ? 1789 PRUS_EOF : 1790 /* If there is more to send set PRUS_MORETOCOME. */ 1791 (flags & MSG_MORETOCOME) || 1792 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; 1793 1794 #ifdef KERN_TLS 1795 pr_send_flag |= tls_send_flag; 1796 #endif 1797 1798 error = so->so_proto->pr_send(so, pr_send_flag, top, 1799 addr, control, td); 1800 1801 if (dontroute) { 1802 SOCK_LOCK(so); 1803 so->so_options &= ~SO_DONTROUTE; 1804 SOCK_UNLOCK(so); 1805 } 1806 1807 #ifdef KERN_TLS 1808 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) { 1809 if (error != 0) { 1810 m_freem(top); 1811 top = NULL; 1812 } else { 1813 soref(so); 1814 ktls_enqueue(top, so, tls_enq_cnt); 1815 } 1816 } 1817 #endif 1818 clen = 0; 1819 control = NULL; 1820 top = NULL; 1821 if (error) 1822 goto release; 1823 } while (resid && space > 0); 1824 } while (resid); 1825 1826 release: 1827 SOCK_IO_SEND_UNLOCK(so); 1828 out: 1829 #ifdef KERN_TLS 1830 if (tls != NULL) 1831 ktls_free(tls); 1832 #endif 1833 if (top != NULL) 1834 m_freem(top); 1835 if (control != NULL) 1836 m_freem(control); 1837 return (error); 1838 } 1839 1840 /* 1841 * Send to a socket from a kernel thread. 1842 * 1843 * XXXGL: in almost all cases uio is NULL and the mbuf is supplied. 1844 * Exception is nfs/bootp_subr.c. It is arguable that the VNET context needs 1845 * to be set at all. This function should just boil down to a static inline 1846 * calling the protocol method. 1847 */ 1848 int 1849 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1850 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1851 { 1852 int error; 1853 1854 CURVNET_SET(so->so_vnet); 1855 error = so->so_proto->pr_sosend(so, addr, uio, 1856 top, control, flags, td); 1857 CURVNET_RESTORE(); 1858 return (error); 1859 } 1860 1861 /* 1862 * send(2), write(2) or aio_write(2) on a socket. 1863 */ 1864 int 1865 sousrsend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1866 struct mbuf *control, int flags, struct proc *userproc) 1867 { 1868 struct thread *td; 1869 ssize_t len; 1870 int error; 1871 1872 td = uio->uio_td; 1873 len = uio->uio_resid; 1874 CURVNET_SET(so->so_vnet); 1875 error = so->so_proto->pr_sosend(so, addr, uio, NULL, control, flags, 1876 td); 1877 CURVNET_RESTORE(); 1878 if (error != 0) { 1879 /* 1880 * Clear transient errors for stream protocols if they made 1881 * some progress. Make exclusion for aio(4) that would 1882 * schedule a new write in case of EWOULDBLOCK and clear 1883 * error itself. See soaio_process_job(). 1884 */ 1885 if (uio->uio_resid != len && 1886 (so->so_proto->pr_flags & PR_ATOMIC) == 0 && 1887 userproc == NULL && 1888 (error == ERESTART || error == EINTR || 1889 error == EWOULDBLOCK)) 1890 error = 0; 1891 /* Generation of SIGPIPE can be controlled per socket. */ 1892 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0 && 1893 (flags & MSG_NOSIGNAL) == 0) { 1894 if (userproc != NULL) { 1895 /* aio(4) job */ 1896 PROC_LOCK(userproc); 1897 kern_psignal(userproc, SIGPIPE); 1898 PROC_UNLOCK(userproc); 1899 } else { 1900 PROC_LOCK(td->td_proc); 1901 tdsignal(td, SIGPIPE); 1902 PROC_UNLOCK(td->td_proc); 1903 } 1904 } 1905 } 1906 return (error); 1907 } 1908 1909 /* 1910 * The part of soreceive() that implements reading non-inline out-of-band 1911 * data from a socket. For more complete comments, see soreceive(), from 1912 * which this code originated. 1913 * 1914 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1915 * unable to return an mbuf chain to the caller. 1916 */ 1917 static int 1918 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1919 { 1920 struct protosw *pr = so->so_proto; 1921 struct mbuf *m; 1922 int error; 1923 1924 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1925 VNET_SO_ASSERT(so); 1926 1927 m = m_get(M_WAITOK, MT_DATA); 1928 error = pr->pr_rcvoob(so, m, flags & MSG_PEEK); 1929 if (error) 1930 goto bad; 1931 do { 1932 error = uiomove(mtod(m, void *), 1933 (int) min(uio->uio_resid, m->m_len), uio); 1934 m = m_free(m); 1935 } while (uio->uio_resid && error == 0 && m); 1936 bad: 1937 if (m != NULL) 1938 m_freem(m); 1939 return (error); 1940 } 1941 1942 /* 1943 * Following replacement or removal of the first mbuf on the first mbuf chain 1944 * of a socket buffer, push necessary state changes back into the socket 1945 * buffer so that other consumers see the values consistently. 'nextrecord' 1946 * is the callers locally stored value of the original value of 1947 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1948 * NOTE: 'nextrecord' may be NULL. 1949 */ 1950 static __inline void 1951 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1952 { 1953 1954 SOCKBUF_LOCK_ASSERT(sb); 1955 /* 1956 * First, update for the new value of nextrecord. If necessary, make 1957 * it the first record. 1958 */ 1959 if (sb->sb_mb != NULL) 1960 sb->sb_mb->m_nextpkt = nextrecord; 1961 else 1962 sb->sb_mb = nextrecord; 1963 1964 /* 1965 * Now update any dependent socket buffer fields to reflect the new 1966 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1967 * addition of a second clause that takes care of the case where 1968 * sb_mb has been updated, but remains the last record. 1969 */ 1970 if (sb->sb_mb == NULL) { 1971 sb->sb_mbtail = NULL; 1972 sb->sb_lastrecord = NULL; 1973 } else if (sb->sb_mb->m_nextpkt == NULL) 1974 sb->sb_lastrecord = sb->sb_mb; 1975 } 1976 1977 /* 1978 * Implement receive operations on a socket. We depend on the way that 1979 * records are added to the sockbuf by sbappend. In particular, each record 1980 * (mbufs linked through m_next) must begin with an address if the protocol 1981 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1982 * data, and then zero or more mbufs of data. In order to allow parallelism 1983 * between network receive and copying to user space, as well as avoid 1984 * sleeping with a mutex held, we release the socket buffer mutex during the 1985 * user space copy. Although the sockbuf is locked, new data may still be 1986 * appended, and thus we must maintain consistency of the sockbuf during that 1987 * time. 1988 * 1989 * The caller may receive the data as a single mbuf chain by supplying an 1990 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1991 * the count in uio_resid. 1992 */ 1993 int 1994 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1995 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1996 { 1997 struct mbuf *m, **mp; 1998 int flags, error, offset; 1999 ssize_t len; 2000 struct protosw *pr = so->so_proto; 2001 struct mbuf *nextrecord; 2002 int moff, type = 0; 2003 ssize_t orig_resid = uio->uio_resid; 2004 bool report_real_len = false; 2005 2006 mp = mp0; 2007 if (psa != NULL) 2008 *psa = NULL; 2009 if (controlp != NULL) 2010 *controlp = NULL; 2011 if (flagsp != NULL) { 2012 report_real_len = *flagsp & MSG_TRUNC; 2013 *flagsp &= ~MSG_TRUNC; 2014 flags = *flagsp &~ MSG_EOR; 2015 } else 2016 flags = 0; 2017 if (flags & MSG_OOB) 2018 return (soreceive_rcvoob(so, uio, flags)); 2019 if (mp != NULL) 2020 *mp = NULL; 2021 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 2022 && uio->uio_resid) { 2023 VNET_SO_ASSERT(so); 2024 pr->pr_rcvd(so, 0); 2025 } 2026 2027 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 2028 if (error) 2029 return (error); 2030 2031 restart: 2032 SOCKBUF_LOCK(&so->so_rcv); 2033 m = so->so_rcv.sb_mb; 2034 /* 2035 * If we have less data than requested, block awaiting more (subject 2036 * to any timeout) if: 2037 * 1. the current count is less than the low water mark, or 2038 * 2. MSG_DONTWAIT is not set 2039 */ 2040 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 2041 sbavail(&so->so_rcv) < uio->uio_resid) && 2042 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && 2043 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 2044 KASSERT(m != NULL || !sbavail(&so->so_rcv), 2045 ("receive: m == %p sbavail == %u", 2046 m, sbavail(&so->so_rcv))); 2047 if (so->so_error || so->so_rerror) { 2048 if (m != NULL) 2049 goto dontblock; 2050 if (so->so_error) 2051 error = so->so_error; 2052 else 2053 error = so->so_rerror; 2054 if ((flags & MSG_PEEK) == 0) { 2055 if (so->so_error) 2056 so->so_error = 0; 2057 else 2058 so->so_rerror = 0; 2059 } 2060 SOCKBUF_UNLOCK(&so->so_rcv); 2061 goto release; 2062 } 2063 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2064 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2065 if (m != NULL) 2066 goto dontblock; 2067 #ifdef KERN_TLS 2068 else if (so->so_rcv.sb_tlsdcc == 0 && 2069 so->so_rcv.sb_tlscc == 0) { 2070 #else 2071 else { 2072 #endif 2073 SOCKBUF_UNLOCK(&so->so_rcv); 2074 goto release; 2075 } 2076 } 2077 for (; m != NULL; m = m->m_next) 2078 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 2079 m = so->so_rcv.sb_mb; 2080 goto dontblock; 2081 } 2082 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED | 2083 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 && 2084 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 2085 SOCKBUF_UNLOCK(&so->so_rcv); 2086 error = ENOTCONN; 2087 goto release; 2088 } 2089 if (uio->uio_resid == 0 && !report_real_len) { 2090 SOCKBUF_UNLOCK(&so->so_rcv); 2091 goto release; 2092 } 2093 if ((so->so_state & SS_NBIO) || 2094 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2095 SOCKBUF_UNLOCK(&so->so_rcv); 2096 error = EWOULDBLOCK; 2097 goto release; 2098 } 2099 SBLASTRECORDCHK(&so->so_rcv); 2100 SBLASTMBUFCHK(&so->so_rcv); 2101 error = sbwait(so, SO_RCV); 2102 SOCKBUF_UNLOCK(&so->so_rcv); 2103 if (error) 2104 goto release; 2105 goto restart; 2106 } 2107 dontblock: 2108 /* 2109 * From this point onward, we maintain 'nextrecord' as a cache of the 2110 * pointer to the next record in the socket buffer. We must keep the 2111 * various socket buffer pointers and local stack versions of the 2112 * pointers in sync, pushing out modifications before dropping the 2113 * socket buffer mutex, and re-reading them when picking it up. 2114 * 2115 * Otherwise, we will race with the network stack appending new data 2116 * or records onto the socket buffer by using inconsistent/stale 2117 * versions of the field, possibly resulting in socket buffer 2118 * corruption. 2119 * 2120 * By holding the high-level sblock(), we prevent simultaneous 2121 * readers from pulling off the front of the socket buffer. 2122 */ 2123 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2124 if (uio->uio_td) 2125 uio->uio_td->td_ru.ru_msgrcv++; 2126 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 2127 SBLASTRECORDCHK(&so->so_rcv); 2128 SBLASTMBUFCHK(&so->so_rcv); 2129 nextrecord = m->m_nextpkt; 2130 if (pr->pr_flags & PR_ADDR) { 2131 KASSERT(m->m_type == MT_SONAME, 2132 ("m->m_type == %d", m->m_type)); 2133 orig_resid = 0; 2134 if (psa != NULL) 2135 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2136 M_NOWAIT); 2137 if (flags & MSG_PEEK) { 2138 m = m->m_next; 2139 } else { 2140 sbfree(&so->so_rcv, m); 2141 so->so_rcv.sb_mb = m_free(m); 2142 m = so->so_rcv.sb_mb; 2143 sockbuf_pushsync(&so->so_rcv, nextrecord); 2144 } 2145 } 2146 2147 /* 2148 * Process one or more MT_CONTROL mbufs present before any data mbufs 2149 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 2150 * just copy the data; if !MSG_PEEK, we call into the protocol to 2151 * perform externalization (or freeing if controlp == NULL). 2152 */ 2153 if (m != NULL && m->m_type == MT_CONTROL) { 2154 struct mbuf *cm = NULL, *cmn; 2155 struct mbuf **cme = &cm; 2156 #ifdef KERN_TLS 2157 struct cmsghdr *cmsg; 2158 struct tls_get_record tgr; 2159 2160 /* 2161 * For MSG_TLSAPPDATA, check for an alert record. 2162 * If found, return ENXIO without removing 2163 * it from the receive queue. This allows a subsequent 2164 * call without MSG_TLSAPPDATA to receive it. 2165 * Note that, for TLS, there should only be a single 2166 * control mbuf with the TLS_GET_RECORD message in it. 2167 */ 2168 if (flags & MSG_TLSAPPDATA) { 2169 cmsg = mtod(m, struct cmsghdr *); 2170 if (cmsg->cmsg_type == TLS_GET_RECORD && 2171 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) { 2172 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr)); 2173 if (__predict_false(tgr.tls_type == 2174 TLS_RLTYPE_ALERT)) { 2175 SOCKBUF_UNLOCK(&so->so_rcv); 2176 error = ENXIO; 2177 goto release; 2178 } 2179 } 2180 } 2181 #endif 2182 2183 do { 2184 if (flags & MSG_PEEK) { 2185 if (controlp != NULL) { 2186 *controlp = m_copym(m, 0, m->m_len, 2187 M_NOWAIT); 2188 controlp = &(*controlp)->m_next; 2189 } 2190 m = m->m_next; 2191 } else { 2192 sbfree(&so->so_rcv, m); 2193 so->so_rcv.sb_mb = m->m_next; 2194 m->m_next = NULL; 2195 *cme = m; 2196 cme = &(*cme)->m_next; 2197 m = so->so_rcv.sb_mb; 2198 } 2199 } while (m != NULL && m->m_type == MT_CONTROL); 2200 if ((flags & MSG_PEEK) == 0) 2201 sockbuf_pushsync(&so->so_rcv, nextrecord); 2202 while (cm != NULL) { 2203 cmn = cm->m_next; 2204 cm->m_next = NULL; 2205 if (pr->pr_domain->dom_externalize != NULL) { 2206 SOCKBUF_UNLOCK(&so->so_rcv); 2207 VNET_SO_ASSERT(so); 2208 error = (*pr->pr_domain->dom_externalize) 2209 (cm, controlp, flags); 2210 SOCKBUF_LOCK(&so->so_rcv); 2211 } else if (controlp != NULL) 2212 *controlp = cm; 2213 else 2214 m_freem(cm); 2215 if (controlp != NULL) { 2216 while (*controlp != NULL) 2217 controlp = &(*controlp)->m_next; 2218 } 2219 cm = cmn; 2220 } 2221 if (m != NULL) 2222 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 2223 else 2224 nextrecord = so->so_rcv.sb_mb; 2225 orig_resid = 0; 2226 } 2227 if (m != NULL) { 2228 if ((flags & MSG_PEEK) == 0) { 2229 KASSERT(m->m_nextpkt == nextrecord, 2230 ("soreceive: post-control, nextrecord !sync")); 2231 if (nextrecord == NULL) { 2232 KASSERT(so->so_rcv.sb_mb == m, 2233 ("soreceive: post-control, sb_mb!=m")); 2234 KASSERT(so->so_rcv.sb_lastrecord == m, 2235 ("soreceive: post-control, lastrecord!=m")); 2236 } 2237 } 2238 type = m->m_type; 2239 if (type == MT_OOBDATA) 2240 flags |= MSG_OOB; 2241 } else { 2242 if ((flags & MSG_PEEK) == 0) { 2243 KASSERT(so->so_rcv.sb_mb == nextrecord, 2244 ("soreceive: sb_mb != nextrecord")); 2245 if (so->so_rcv.sb_mb == NULL) { 2246 KASSERT(so->so_rcv.sb_lastrecord == NULL, 2247 ("soreceive: sb_lastercord != NULL")); 2248 } 2249 } 2250 } 2251 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2252 SBLASTRECORDCHK(&so->so_rcv); 2253 SBLASTMBUFCHK(&so->so_rcv); 2254 2255 /* 2256 * Now continue to read any data mbufs off of the head of the socket 2257 * buffer until the read request is satisfied. Note that 'type' is 2258 * used to store the type of any mbuf reads that have happened so far 2259 * such that soreceive() can stop reading if the type changes, which 2260 * causes soreceive() to return only one of regular data and inline 2261 * out-of-band data in a single socket receive operation. 2262 */ 2263 moff = 0; 2264 offset = 0; 2265 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 2266 && error == 0) { 2267 /* 2268 * If the type of mbuf has changed since the last mbuf 2269 * examined ('type'), end the receive operation. 2270 */ 2271 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2272 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 2273 if (type != m->m_type) 2274 break; 2275 } else if (type == MT_OOBDATA) 2276 break; 2277 else 2278 KASSERT(m->m_type == MT_DATA, 2279 ("m->m_type == %d", m->m_type)); 2280 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 2281 len = uio->uio_resid; 2282 if (so->so_oobmark && len > so->so_oobmark - offset) 2283 len = so->so_oobmark - offset; 2284 if (len > m->m_len - moff) 2285 len = m->m_len - moff; 2286 /* 2287 * If mp is set, just pass back the mbufs. Otherwise copy 2288 * them out via the uio, then free. Sockbuf must be 2289 * consistent here (points to current mbuf, it points to next 2290 * record) when we drop priority; we must note any additions 2291 * to the sockbuf when we block interrupts again. 2292 */ 2293 if (mp == NULL) { 2294 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2295 SBLASTRECORDCHK(&so->so_rcv); 2296 SBLASTMBUFCHK(&so->so_rcv); 2297 SOCKBUF_UNLOCK(&so->so_rcv); 2298 if ((m->m_flags & M_EXTPG) != 0) 2299 error = m_unmapped_uiomove(m, moff, uio, 2300 (int)len); 2301 else 2302 error = uiomove(mtod(m, char *) + moff, 2303 (int)len, uio); 2304 SOCKBUF_LOCK(&so->so_rcv); 2305 if (error) { 2306 /* 2307 * The MT_SONAME mbuf has already been removed 2308 * from the record, so it is necessary to 2309 * remove the data mbufs, if any, to preserve 2310 * the invariant in the case of PR_ADDR that 2311 * requires MT_SONAME mbufs at the head of 2312 * each record. 2313 */ 2314 if (pr->pr_flags & PR_ATOMIC && 2315 ((flags & MSG_PEEK) == 0)) 2316 (void)sbdroprecord_locked(&so->so_rcv); 2317 SOCKBUF_UNLOCK(&so->so_rcv); 2318 goto release; 2319 } 2320 } else 2321 uio->uio_resid -= len; 2322 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2323 if (len == m->m_len - moff) { 2324 if (m->m_flags & M_EOR) 2325 flags |= MSG_EOR; 2326 if (flags & MSG_PEEK) { 2327 m = m->m_next; 2328 moff = 0; 2329 } else { 2330 nextrecord = m->m_nextpkt; 2331 sbfree(&so->so_rcv, m); 2332 if (mp != NULL) { 2333 m->m_nextpkt = NULL; 2334 *mp = m; 2335 mp = &m->m_next; 2336 so->so_rcv.sb_mb = m = m->m_next; 2337 *mp = NULL; 2338 } else { 2339 so->so_rcv.sb_mb = m_free(m); 2340 m = so->so_rcv.sb_mb; 2341 } 2342 sockbuf_pushsync(&so->so_rcv, nextrecord); 2343 SBLASTRECORDCHK(&so->so_rcv); 2344 SBLASTMBUFCHK(&so->so_rcv); 2345 } 2346 } else { 2347 if (flags & MSG_PEEK) 2348 moff += len; 2349 else { 2350 if (mp != NULL) { 2351 if (flags & MSG_DONTWAIT) { 2352 *mp = m_copym(m, 0, len, 2353 M_NOWAIT); 2354 if (*mp == NULL) { 2355 /* 2356 * m_copym() couldn't 2357 * allocate an mbuf. 2358 * Adjust uio_resid back 2359 * (it was adjusted 2360 * down by len bytes, 2361 * which we didn't end 2362 * up "copying" over). 2363 */ 2364 uio->uio_resid += len; 2365 break; 2366 } 2367 } else { 2368 SOCKBUF_UNLOCK(&so->so_rcv); 2369 *mp = m_copym(m, 0, len, 2370 M_WAITOK); 2371 SOCKBUF_LOCK(&so->so_rcv); 2372 } 2373 } 2374 sbcut_locked(&so->so_rcv, len); 2375 } 2376 } 2377 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2378 if (so->so_oobmark) { 2379 if ((flags & MSG_PEEK) == 0) { 2380 so->so_oobmark -= len; 2381 if (so->so_oobmark == 0) { 2382 so->so_rcv.sb_state |= SBS_RCVATMARK; 2383 break; 2384 } 2385 } else { 2386 offset += len; 2387 if (offset == so->so_oobmark) 2388 break; 2389 } 2390 } 2391 if (flags & MSG_EOR) 2392 break; 2393 /* 2394 * If the MSG_WAITALL flag is set (for non-atomic socket), we 2395 * must not quit until "uio->uio_resid == 0" or an error 2396 * termination. If a signal/timeout occurs, return with a 2397 * short count but without error. Keep sockbuf locked 2398 * against other readers. 2399 */ 2400 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 2401 !sosendallatonce(so) && nextrecord == NULL) { 2402 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2403 if (so->so_error || so->so_rerror || 2404 so->so_rcv.sb_state & SBS_CANTRCVMORE) 2405 break; 2406 /* 2407 * Notify the protocol that some data has been 2408 * drained before blocking. 2409 */ 2410 if (pr->pr_flags & PR_WANTRCVD) { 2411 SOCKBUF_UNLOCK(&so->so_rcv); 2412 VNET_SO_ASSERT(so); 2413 pr->pr_rcvd(so, flags); 2414 SOCKBUF_LOCK(&so->so_rcv); 2415 } 2416 SBLASTRECORDCHK(&so->so_rcv); 2417 SBLASTMBUFCHK(&so->so_rcv); 2418 /* 2419 * We could receive some data while was notifying 2420 * the protocol. Skip blocking in this case. 2421 */ 2422 if (so->so_rcv.sb_mb == NULL) { 2423 error = sbwait(so, SO_RCV); 2424 if (error) { 2425 SOCKBUF_UNLOCK(&so->so_rcv); 2426 goto release; 2427 } 2428 } 2429 m = so->so_rcv.sb_mb; 2430 if (m != NULL) 2431 nextrecord = m->m_nextpkt; 2432 } 2433 } 2434 2435 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2436 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 2437 if (report_real_len) 2438 uio->uio_resid -= m_length(m, NULL) - moff; 2439 flags |= MSG_TRUNC; 2440 if ((flags & MSG_PEEK) == 0) 2441 (void) sbdroprecord_locked(&so->so_rcv); 2442 } 2443 if ((flags & MSG_PEEK) == 0) { 2444 if (m == NULL) { 2445 /* 2446 * First part is an inline SB_EMPTY_FIXUP(). Second 2447 * part makes sure sb_lastrecord is up-to-date if 2448 * there is still data in the socket buffer. 2449 */ 2450 so->so_rcv.sb_mb = nextrecord; 2451 if (so->so_rcv.sb_mb == NULL) { 2452 so->so_rcv.sb_mbtail = NULL; 2453 so->so_rcv.sb_lastrecord = NULL; 2454 } else if (nextrecord->m_nextpkt == NULL) 2455 so->so_rcv.sb_lastrecord = nextrecord; 2456 } 2457 SBLASTRECORDCHK(&so->so_rcv); 2458 SBLASTMBUFCHK(&so->so_rcv); 2459 /* 2460 * If soreceive() is being done from the socket callback, 2461 * then don't need to generate ACK to peer to update window, 2462 * since ACK will be generated on return to TCP. 2463 */ 2464 if (!(flags & MSG_SOCALLBCK) && 2465 (pr->pr_flags & PR_WANTRCVD)) { 2466 SOCKBUF_UNLOCK(&so->so_rcv); 2467 VNET_SO_ASSERT(so); 2468 pr->pr_rcvd(so, flags); 2469 SOCKBUF_LOCK(&so->so_rcv); 2470 } 2471 } 2472 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2473 if (orig_resid == uio->uio_resid && orig_resid && 2474 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 2475 SOCKBUF_UNLOCK(&so->so_rcv); 2476 goto restart; 2477 } 2478 SOCKBUF_UNLOCK(&so->so_rcv); 2479 2480 if (flagsp != NULL) 2481 *flagsp |= flags; 2482 release: 2483 SOCK_IO_RECV_UNLOCK(so); 2484 return (error); 2485 } 2486 2487 /* 2488 * Optimized version of soreceive() for stream (TCP) sockets. 2489 */ 2490 int 2491 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 2492 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2493 { 2494 int len = 0, error = 0, flags, oresid; 2495 struct sockbuf *sb; 2496 struct mbuf *m, *n = NULL; 2497 2498 /* We only do stream sockets. */ 2499 if (so->so_type != SOCK_STREAM) 2500 return (EINVAL); 2501 if (psa != NULL) 2502 *psa = NULL; 2503 if (flagsp != NULL) 2504 flags = *flagsp &~ MSG_EOR; 2505 else 2506 flags = 0; 2507 if (controlp != NULL) 2508 *controlp = NULL; 2509 if (flags & MSG_OOB) 2510 return (soreceive_rcvoob(so, uio, flags)); 2511 if (mp0 != NULL) 2512 *mp0 = NULL; 2513 2514 sb = &so->so_rcv; 2515 2516 #ifdef KERN_TLS 2517 /* 2518 * KTLS store TLS records as records with a control message to 2519 * describe the framing. 2520 * 2521 * We check once here before acquiring locks to optimize the 2522 * common case. 2523 */ 2524 if (sb->sb_tls_info != NULL) 2525 return (soreceive_generic(so, psa, uio, mp0, controlp, 2526 flagsp)); 2527 #endif 2528 2529 /* Prevent other readers from entering the socket. */ 2530 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 2531 if (error) 2532 return (error); 2533 SOCKBUF_LOCK(sb); 2534 2535 #ifdef KERN_TLS 2536 if (sb->sb_tls_info != NULL) { 2537 SOCKBUF_UNLOCK(sb); 2538 SOCK_IO_RECV_UNLOCK(so); 2539 return (soreceive_generic(so, psa, uio, mp0, controlp, 2540 flagsp)); 2541 } 2542 #endif 2543 2544 /* Easy one, no space to copyout anything. */ 2545 if (uio->uio_resid == 0) { 2546 error = EINVAL; 2547 goto out; 2548 } 2549 oresid = uio->uio_resid; 2550 2551 /* We will never ever get anything unless we are or were connected. */ 2552 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 2553 error = ENOTCONN; 2554 goto out; 2555 } 2556 2557 restart: 2558 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2559 2560 /* Abort if socket has reported problems. */ 2561 if (so->so_error) { 2562 if (sbavail(sb) > 0) 2563 goto deliver; 2564 if (oresid > uio->uio_resid) 2565 goto out; 2566 error = so->so_error; 2567 if (!(flags & MSG_PEEK)) 2568 so->so_error = 0; 2569 goto out; 2570 } 2571 2572 /* Door is closed. Deliver what is left, if any. */ 2573 if (sb->sb_state & SBS_CANTRCVMORE) { 2574 if (sbavail(sb) > 0) 2575 goto deliver; 2576 else 2577 goto out; 2578 } 2579 2580 /* Socket buffer is empty and we shall not block. */ 2581 if (sbavail(sb) == 0 && 2582 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2583 error = EAGAIN; 2584 goto out; 2585 } 2586 2587 /* Socket buffer got some data that we shall deliver now. */ 2588 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 2589 ((so->so_state & SS_NBIO) || 2590 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2591 sbavail(sb) >= sb->sb_lowat || 2592 sbavail(sb) >= uio->uio_resid || 2593 sbavail(sb) >= sb->sb_hiwat) ) { 2594 goto deliver; 2595 } 2596 2597 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2598 if ((flags & MSG_WAITALL) && 2599 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 2600 goto deliver; 2601 2602 /* 2603 * Wait and block until (more) data comes in. 2604 * NB: Drops the sockbuf lock during wait. 2605 */ 2606 error = sbwait(so, SO_RCV); 2607 if (error) 2608 goto out; 2609 goto restart; 2610 2611 deliver: 2612 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2613 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 2614 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2615 2616 /* Statistics. */ 2617 if (uio->uio_td) 2618 uio->uio_td->td_ru.ru_msgrcv++; 2619 2620 /* Fill uio until full or current end of socket buffer is reached. */ 2621 len = min(uio->uio_resid, sbavail(sb)); 2622 if (mp0 != NULL) { 2623 /* Dequeue as many mbufs as possible. */ 2624 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2625 if (*mp0 == NULL) 2626 *mp0 = sb->sb_mb; 2627 else 2628 m_cat(*mp0, sb->sb_mb); 2629 for (m = sb->sb_mb; 2630 m != NULL && m->m_len <= len; 2631 m = m->m_next) { 2632 KASSERT(!(m->m_flags & M_NOTAVAIL), 2633 ("%s: m %p not available", __func__, m)); 2634 len -= m->m_len; 2635 uio->uio_resid -= m->m_len; 2636 sbfree(sb, m); 2637 n = m; 2638 } 2639 n->m_next = NULL; 2640 sb->sb_mb = m; 2641 sb->sb_lastrecord = sb->sb_mb; 2642 if (sb->sb_mb == NULL) 2643 SB_EMPTY_FIXUP(sb); 2644 } 2645 /* Copy the remainder. */ 2646 if (len > 0) { 2647 KASSERT(sb->sb_mb != NULL, 2648 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2649 2650 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2651 if (m == NULL) 2652 len = 0; /* Don't flush data from sockbuf. */ 2653 else 2654 uio->uio_resid -= len; 2655 if (*mp0 != NULL) 2656 m_cat(*mp0, m); 2657 else 2658 *mp0 = m; 2659 if (*mp0 == NULL) { 2660 error = ENOBUFS; 2661 goto out; 2662 } 2663 } 2664 } else { 2665 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2666 SOCKBUF_UNLOCK(sb); 2667 error = m_mbuftouio(uio, sb->sb_mb, len); 2668 SOCKBUF_LOCK(sb); 2669 if (error) 2670 goto out; 2671 } 2672 SBLASTRECORDCHK(sb); 2673 SBLASTMBUFCHK(sb); 2674 2675 /* 2676 * Remove the delivered data from the socket buffer unless we 2677 * were only peeking. 2678 */ 2679 if (!(flags & MSG_PEEK)) { 2680 if (len > 0) 2681 sbdrop_locked(sb, len); 2682 2683 /* Notify protocol that we drained some data. */ 2684 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2685 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2686 !(flags & MSG_SOCALLBCK))) { 2687 SOCKBUF_UNLOCK(sb); 2688 VNET_SO_ASSERT(so); 2689 so->so_proto->pr_rcvd(so, flags); 2690 SOCKBUF_LOCK(sb); 2691 } 2692 } 2693 2694 /* 2695 * For MSG_WAITALL we may have to loop again and wait for 2696 * more data to come in. 2697 */ 2698 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2699 goto restart; 2700 out: 2701 SBLASTRECORDCHK(sb); 2702 SBLASTMBUFCHK(sb); 2703 SOCKBUF_UNLOCK(sb); 2704 SOCK_IO_RECV_UNLOCK(so); 2705 return (error); 2706 } 2707 2708 /* 2709 * Optimized version of soreceive() for simple datagram cases from userspace. 2710 * Unlike in the stream case, we're able to drop a datagram if copyout() 2711 * fails, and because we handle datagrams atomically, we don't need to use a 2712 * sleep lock to prevent I/O interlacing. 2713 */ 2714 int 2715 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2716 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2717 { 2718 struct mbuf *m, *m2; 2719 int flags, error; 2720 ssize_t len; 2721 struct protosw *pr = so->so_proto; 2722 struct mbuf *nextrecord; 2723 2724 if (psa != NULL) 2725 *psa = NULL; 2726 if (controlp != NULL) 2727 *controlp = NULL; 2728 if (flagsp != NULL) 2729 flags = *flagsp &~ MSG_EOR; 2730 else 2731 flags = 0; 2732 2733 /* 2734 * For any complicated cases, fall back to the full 2735 * soreceive_generic(). 2736 */ 2737 if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC))) 2738 return (soreceive_generic(so, psa, uio, mp0, controlp, 2739 flagsp)); 2740 2741 /* 2742 * Enforce restrictions on use. 2743 */ 2744 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2745 ("soreceive_dgram: wantrcvd")); 2746 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2747 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2748 ("soreceive_dgram: SBS_RCVATMARK")); 2749 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2750 ("soreceive_dgram: P_CONNREQUIRED")); 2751 2752 /* 2753 * Loop blocking while waiting for a datagram. 2754 */ 2755 SOCKBUF_LOCK(&so->so_rcv); 2756 while ((m = so->so_rcv.sb_mb) == NULL) { 2757 KASSERT(sbavail(&so->so_rcv) == 0, 2758 ("soreceive_dgram: sb_mb NULL but sbavail %u", 2759 sbavail(&so->so_rcv))); 2760 if (so->so_error) { 2761 error = so->so_error; 2762 so->so_error = 0; 2763 SOCKBUF_UNLOCK(&so->so_rcv); 2764 return (error); 2765 } 2766 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2767 uio->uio_resid == 0) { 2768 SOCKBUF_UNLOCK(&so->so_rcv); 2769 return (0); 2770 } 2771 if ((so->so_state & SS_NBIO) || 2772 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2773 SOCKBUF_UNLOCK(&so->so_rcv); 2774 return (EWOULDBLOCK); 2775 } 2776 SBLASTRECORDCHK(&so->so_rcv); 2777 SBLASTMBUFCHK(&so->so_rcv); 2778 error = sbwait(so, SO_RCV); 2779 if (error) { 2780 SOCKBUF_UNLOCK(&so->so_rcv); 2781 return (error); 2782 } 2783 } 2784 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2785 2786 if (uio->uio_td) 2787 uio->uio_td->td_ru.ru_msgrcv++; 2788 SBLASTRECORDCHK(&so->so_rcv); 2789 SBLASTMBUFCHK(&so->so_rcv); 2790 nextrecord = m->m_nextpkt; 2791 if (nextrecord == NULL) { 2792 KASSERT(so->so_rcv.sb_lastrecord == m, 2793 ("soreceive_dgram: lastrecord != m")); 2794 } 2795 2796 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2797 ("soreceive_dgram: m_nextpkt != nextrecord")); 2798 2799 /* 2800 * Pull 'm' and its chain off the front of the packet queue. 2801 */ 2802 so->so_rcv.sb_mb = NULL; 2803 sockbuf_pushsync(&so->so_rcv, nextrecord); 2804 2805 /* 2806 * Walk 'm's chain and free that many bytes from the socket buffer. 2807 */ 2808 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2809 sbfree(&so->so_rcv, m2); 2810 2811 /* 2812 * Do a few last checks before we let go of the lock. 2813 */ 2814 SBLASTRECORDCHK(&so->so_rcv); 2815 SBLASTMBUFCHK(&so->so_rcv); 2816 SOCKBUF_UNLOCK(&so->so_rcv); 2817 2818 if (pr->pr_flags & PR_ADDR) { 2819 KASSERT(m->m_type == MT_SONAME, 2820 ("m->m_type == %d", m->m_type)); 2821 if (psa != NULL) 2822 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2823 M_NOWAIT); 2824 m = m_free(m); 2825 } 2826 if (m == NULL) { 2827 /* XXXRW: Can this happen? */ 2828 return (0); 2829 } 2830 2831 /* 2832 * Packet to copyout() is now in 'm' and it is disconnected from the 2833 * queue. 2834 * 2835 * Process one or more MT_CONTROL mbufs present before any data mbufs 2836 * in the first mbuf chain on the socket buffer. We call into the 2837 * protocol to perform externalization (or freeing if controlp == 2838 * NULL). In some cases there can be only MT_CONTROL mbufs without 2839 * MT_DATA mbufs. 2840 */ 2841 if (m->m_type == MT_CONTROL) { 2842 struct mbuf *cm = NULL, *cmn; 2843 struct mbuf **cme = &cm; 2844 2845 do { 2846 m2 = m->m_next; 2847 m->m_next = NULL; 2848 *cme = m; 2849 cme = &(*cme)->m_next; 2850 m = m2; 2851 } while (m != NULL && m->m_type == MT_CONTROL); 2852 while (cm != NULL) { 2853 cmn = cm->m_next; 2854 cm->m_next = NULL; 2855 if (pr->pr_domain->dom_externalize != NULL) { 2856 error = (*pr->pr_domain->dom_externalize) 2857 (cm, controlp, flags); 2858 } else if (controlp != NULL) 2859 *controlp = cm; 2860 else 2861 m_freem(cm); 2862 if (controlp != NULL) { 2863 while (*controlp != NULL) 2864 controlp = &(*controlp)->m_next; 2865 } 2866 cm = cmn; 2867 } 2868 } 2869 KASSERT(m == NULL || m->m_type == MT_DATA, 2870 ("soreceive_dgram: !data")); 2871 while (m != NULL && uio->uio_resid > 0) { 2872 len = uio->uio_resid; 2873 if (len > m->m_len) 2874 len = m->m_len; 2875 error = uiomove(mtod(m, char *), (int)len, uio); 2876 if (error) { 2877 m_freem(m); 2878 return (error); 2879 } 2880 if (len == m->m_len) 2881 m = m_free(m); 2882 else { 2883 m->m_data += len; 2884 m->m_len -= len; 2885 } 2886 } 2887 if (m != NULL) { 2888 flags |= MSG_TRUNC; 2889 m_freem(m); 2890 } 2891 if (flagsp != NULL) 2892 *flagsp |= flags; 2893 return (0); 2894 } 2895 2896 int 2897 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2898 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2899 { 2900 int error; 2901 2902 CURVNET_SET(so->so_vnet); 2903 error = so->so_proto->pr_soreceive(so, psa, uio, mp0, controlp, flagsp); 2904 CURVNET_RESTORE(); 2905 return (error); 2906 } 2907 2908 int 2909 soshutdown(struct socket *so, int how) 2910 { 2911 struct protosw *pr; 2912 int error, soerror_enotconn; 2913 2914 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2915 return (EINVAL); 2916 2917 soerror_enotconn = 0; 2918 SOCK_LOCK(so); 2919 if ((so->so_state & 2920 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 2921 /* 2922 * POSIX mandates us to return ENOTCONN when shutdown(2) is 2923 * invoked on a datagram sockets, however historically we would 2924 * actually tear socket down. This is known to be leveraged by 2925 * some applications to unblock process waiting in recvXXX(2) 2926 * by other process that it shares that socket with. Try to meet 2927 * both backward-compatibility and POSIX requirements by forcing 2928 * ENOTCONN but still asking protocol to perform pru_shutdown(). 2929 */ 2930 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) { 2931 SOCK_UNLOCK(so); 2932 return (ENOTCONN); 2933 } 2934 soerror_enotconn = 1; 2935 } 2936 2937 if (SOLISTENING(so)) { 2938 if (how != SHUT_WR) { 2939 so->so_error = ECONNABORTED; 2940 solisten_wakeup(so); /* unlocks so */ 2941 } else { 2942 SOCK_UNLOCK(so); 2943 } 2944 goto done; 2945 } 2946 SOCK_UNLOCK(so); 2947 2948 CURVNET_SET(so->so_vnet); 2949 pr = so->so_proto; 2950 if (pr->pr_flush != NULL) 2951 pr->pr_flush(so, how); 2952 if (how != SHUT_WR) 2953 sorflush(so); 2954 if (how != SHUT_RD) { 2955 error = pr->pr_shutdown(so); 2956 wakeup(&so->so_timeo); 2957 CURVNET_RESTORE(); 2958 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error); 2959 } 2960 wakeup(&so->so_timeo); 2961 CURVNET_RESTORE(); 2962 2963 done: 2964 return (soerror_enotconn ? ENOTCONN : 0); 2965 } 2966 2967 void 2968 sorflush(struct socket *so) 2969 { 2970 struct protosw *pr; 2971 int error; 2972 2973 VNET_SO_ASSERT(so); 2974 2975 /* 2976 * Dislodge threads currently blocked in receive and wait to acquire 2977 * a lock against other simultaneous readers before clearing the 2978 * socket buffer. Don't let our acquire be interrupted by a signal 2979 * despite any existing socket disposition on interruptable waiting. 2980 */ 2981 socantrcvmore(so); 2982 2983 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR); 2984 if (error != 0) { 2985 KASSERT(SOLISTENING(so), 2986 ("%s: soiolock(%p) failed", __func__, so)); 2987 return; 2988 } 2989 2990 pr = so->so_proto; 2991 if (pr->pr_flags & PR_RIGHTS) { 2992 MPASS(pr->pr_domain->dom_dispose != NULL); 2993 (*pr->pr_domain->dom_dispose)(so); 2994 } else { 2995 sbrelease(so, SO_RCV); 2996 SOCK_IO_RECV_UNLOCK(so); 2997 } 2998 2999 } 3000 3001 /* 3002 * Wrapper for Socket established helper hook. 3003 * Parameters: socket, context of the hook point, hook id. 3004 */ 3005 static int inline 3006 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 3007 { 3008 struct socket_hhook_data hhook_data = { 3009 .so = so, 3010 .hctx = hctx, 3011 .m = NULL, 3012 .status = 0 3013 }; 3014 3015 CURVNET_SET(so->so_vnet); 3016 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 3017 CURVNET_RESTORE(); 3018 3019 /* Ugly but needed, since hhooks return void for now */ 3020 return (hhook_data.status); 3021 } 3022 3023 /* 3024 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 3025 * additional variant to handle the case where the option value needs to be 3026 * some kind of integer, but not a specific size. In addition to their use 3027 * here, these functions are also called by the protocol-level pr_ctloutput() 3028 * routines. 3029 */ 3030 int 3031 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 3032 { 3033 size_t valsize; 3034 3035 /* 3036 * If the user gives us more than we wanted, we ignore it, but if we 3037 * don't get the minimum length the caller wants, we return EINVAL. 3038 * On success, sopt->sopt_valsize is set to however much we actually 3039 * retrieved. 3040 */ 3041 if ((valsize = sopt->sopt_valsize) < minlen) 3042 return EINVAL; 3043 if (valsize > len) 3044 sopt->sopt_valsize = valsize = len; 3045 3046 if (sopt->sopt_td != NULL) 3047 return (copyin(sopt->sopt_val, buf, valsize)); 3048 3049 bcopy(sopt->sopt_val, buf, valsize); 3050 return (0); 3051 } 3052 3053 /* 3054 * Kernel version of setsockopt(2). 3055 * 3056 * XXX: optlen is size_t, not socklen_t 3057 */ 3058 int 3059 so_setsockopt(struct socket *so, int level, int optname, void *optval, 3060 size_t optlen) 3061 { 3062 struct sockopt sopt; 3063 3064 sopt.sopt_level = level; 3065 sopt.sopt_name = optname; 3066 sopt.sopt_dir = SOPT_SET; 3067 sopt.sopt_val = optval; 3068 sopt.sopt_valsize = optlen; 3069 sopt.sopt_td = NULL; 3070 return (sosetopt(so, &sopt)); 3071 } 3072 3073 int 3074 sosetopt(struct socket *so, struct sockopt *sopt) 3075 { 3076 int error, optval; 3077 struct linger l; 3078 struct timeval tv; 3079 sbintime_t val, *valp; 3080 uint32_t val32; 3081 #ifdef MAC 3082 struct mac extmac; 3083 #endif 3084 3085 CURVNET_SET(so->so_vnet); 3086 error = 0; 3087 if (sopt->sopt_level != SOL_SOCKET) { 3088 if (so->so_proto->pr_ctloutput != NULL) 3089 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3090 else 3091 error = ENOPROTOOPT; 3092 } else { 3093 switch (sopt->sopt_name) { 3094 case SO_ACCEPTFILTER: 3095 error = accept_filt_setopt(so, sopt); 3096 if (error) 3097 goto bad; 3098 break; 3099 3100 case SO_LINGER: 3101 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 3102 if (error) 3103 goto bad; 3104 if (l.l_linger < 0 || 3105 l.l_linger > USHRT_MAX || 3106 l.l_linger > (INT_MAX / hz)) { 3107 error = EDOM; 3108 goto bad; 3109 } 3110 SOCK_LOCK(so); 3111 so->so_linger = l.l_linger; 3112 if (l.l_onoff) 3113 so->so_options |= SO_LINGER; 3114 else 3115 so->so_options &= ~SO_LINGER; 3116 SOCK_UNLOCK(so); 3117 break; 3118 3119 case SO_DEBUG: 3120 case SO_KEEPALIVE: 3121 case SO_DONTROUTE: 3122 case SO_USELOOPBACK: 3123 case SO_BROADCAST: 3124 case SO_REUSEADDR: 3125 case SO_REUSEPORT: 3126 case SO_REUSEPORT_LB: 3127 case SO_OOBINLINE: 3128 case SO_TIMESTAMP: 3129 case SO_BINTIME: 3130 case SO_NOSIGPIPE: 3131 case SO_NO_DDP: 3132 case SO_NO_OFFLOAD: 3133 case SO_RERROR: 3134 error = sooptcopyin(sopt, &optval, sizeof optval, 3135 sizeof optval); 3136 if (error) 3137 goto bad; 3138 SOCK_LOCK(so); 3139 if (optval) 3140 so->so_options |= sopt->sopt_name; 3141 else 3142 so->so_options &= ~sopt->sopt_name; 3143 SOCK_UNLOCK(so); 3144 break; 3145 3146 case SO_SETFIB: 3147 error = sooptcopyin(sopt, &optval, sizeof optval, 3148 sizeof optval); 3149 if (error) 3150 goto bad; 3151 3152 if (optval < 0 || optval >= rt_numfibs) { 3153 error = EINVAL; 3154 goto bad; 3155 } 3156 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 3157 (so->so_proto->pr_domain->dom_family == PF_INET6) || 3158 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 3159 so->so_fibnum = optval; 3160 else 3161 so->so_fibnum = 0; 3162 break; 3163 3164 case SO_USER_COOKIE: 3165 error = sooptcopyin(sopt, &val32, sizeof val32, 3166 sizeof val32); 3167 if (error) 3168 goto bad; 3169 so->so_user_cookie = val32; 3170 break; 3171 3172 case SO_SNDBUF: 3173 case SO_RCVBUF: 3174 case SO_SNDLOWAT: 3175 case SO_RCVLOWAT: 3176 error = so->so_proto->pr_setsbopt(so, sopt); 3177 if (error) 3178 goto bad; 3179 break; 3180 3181 case SO_SNDTIMEO: 3182 case SO_RCVTIMEO: 3183 #ifdef COMPAT_FREEBSD32 3184 if (SV_CURPROC_FLAG(SV_ILP32)) { 3185 struct timeval32 tv32; 3186 3187 error = sooptcopyin(sopt, &tv32, sizeof tv32, 3188 sizeof tv32); 3189 CP(tv32, tv, tv_sec); 3190 CP(tv32, tv, tv_usec); 3191 } else 3192 #endif 3193 error = sooptcopyin(sopt, &tv, sizeof tv, 3194 sizeof tv); 3195 if (error) 3196 goto bad; 3197 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 3198 tv.tv_usec >= 1000000) { 3199 error = EDOM; 3200 goto bad; 3201 } 3202 if (tv.tv_sec > INT32_MAX) 3203 val = SBT_MAX; 3204 else 3205 val = tvtosbt(tv); 3206 SOCK_LOCK(so); 3207 valp = sopt->sopt_name == SO_SNDTIMEO ? 3208 (SOLISTENING(so) ? &so->sol_sbsnd_timeo : 3209 &so->so_snd.sb_timeo) : 3210 (SOLISTENING(so) ? &so->sol_sbrcv_timeo : 3211 &so->so_rcv.sb_timeo); 3212 *valp = val; 3213 SOCK_UNLOCK(so); 3214 break; 3215 3216 case SO_LABEL: 3217 #ifdef MAC 3218 error = sooptcopyin(sopt, &extmac, sizeof extmac, 3219 sizeof extmac); 3220 if (error) 3221 goto bad; 3222 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 3223 so, &extmac); 3224 #else 3225 error = EOPNOTSUPP; 3226 #endif 3227 break; 3228 3229 case SO_TS_CLOCK: 3230 error = sooptcopyin(sopt, &optval, sizeof optval, 3231 sizeof optval); 3232 if (error) 3233 goto bad; 3234 if (optval < 0 || optval > SO_TS_CLOCK_MAX) { 3235 error = EINVAL; 3236 goto bad; 3237 } 3238 so->so_ts_clock = optval; 3239 break; 3240 3241 case SO_MAX_PACING_RATE: 3242 error = sooptcopyin(sopt, &val32, sizeof(val32), 3243 sizeof(val32)); 3244 if (error) 3245 goto bad; 3246 so->so_max_pacing_rate = val32; 3247 break; 3248 3249 default: 3250 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3251 error = hhook_run_socket(so, sopt, 3252 HHOOK_SOCKET_OPT); 3253 else 3254 error = ENOPROTOOPT; 3255 break; 3256 } 3257 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 3258 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 3259 } 3260 bad: 3261 CURVNET_RESTORE(); 3262 return (error); 3263 } 3264 3265 /* 3266 * Helper routine for getsockopt. 3267 */ 3268 int 3269 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 3270 { 3271 int error; 3272 size_t valsize; 3273 3274 error = 0; 3275 3276 /* 3277 * Documented get behavior is that we always return a value, possibly 3278 * truncated to fit in the user's buffer. Traditional behavior is 3279 * that we always tell the user precisely how much we copied, rather 3280 * than something useful like the total amount we had available for 3281 * her. Note that this interface is not idempotent; the entire 3282 * answer must be generated ahead of time. 3283 */ 3284 valsize = min(len, sopt->sopt_valsize); 3285 sopt->sopt_valsize = valsize; 3286 if (sopt->sopt_val != NULL) { 3287 if (sopt->sopt_td != NULL) 3288 error = copyout(buf, sopt->sopt_val, valsize); 3289 else 3290 bcopy(buf, sopt->sopt_val, valsize); 3291 } 3292 return (error); 3293 } 3294 3295 int 3296 sogetopt(struct socket *so, struct sockopt *sopt) 3297 { 3298 int error, optval; 3299 struct linger l; 3300 struct timeval tv; 3301 #ifdef MAC 3302 struct mac extmac; 3303 #endif 3304 3305 CURVNET_SET(so->so_vnet); 3306 error = 0; 3307 if (sopt->sopt_level != SOL_SOCKET) { 3308 if (so->so_proto->pr_ctloutput != NULL) 3309 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3310 else 3311 error = ENOPROTOOPT; 3312 CURVNET_RESTORE(); 3313 return (error); 3314 } else { 3315 switch (sopt->sopt_name) { 3316 case SO_ACCEPTFILTER: 3317 error = accept_filt_getopt(so, sopt); 3318 break; 3319 3320 case SO_LINGER: 3321 SOCK_LOCK(so); 3322 l.l_onoff = so->so_options & SO_LINGER; 3323 l.l_linger = so->so_linger; 3324 SOCK_UNLOCK(so); 3325 error = sooptcopyout(sopt, &l, sizeof l); 3326 break; 3327 3328 case SO_USELOOPBACK: 3329 case SO_DONTROUTE: 3330 case SO_DEBUG: 3331 case SO_KEEPALIVE: 3332 case SO_REUSEADDR: 3333 case SO_REUSEPORT: 3334 case SO_REUSEPORT_LB: 3335 case SO_BROADCAST: 3336 case SO_OOBINLINE: 3337 case SO_ACCEPTCONN: 3338 case SO_TIMESTAMP: 3339 case SO_BINTIME: 3340 case SO_NOSIGPIPE: 3341 case SO_NO_DDP: 3342 case SO_NO_OFFLOAD: 3343 case SO_RERROR: 3344 optval = so->so_options & sopt->sopt_name; 3345 integer: 3346 error = sooptcopyout(sopt, &optval, sizeof optval); 3347 break; 3348 3349 case SO_DOMAIN: 3350 optval = so->so_proto->pr_domain->dom_family; 3351 goto integer; 3352 3353 case SO_TYPE: 3354 optval = so->so_type; 3355 goto integer; 3356 3357 case SO_PROTOCOL: 3358 optval = so->so_proto->pr_protocol; 3359 goto integer; 3360 3361 case SO_ERROR: 3362 SOCK_LOCK(so); 3363 if (so->so_error) { 3364 optval = so->so_error; 3365 so->so_error = 0; 3366 } else { 3367 optval = so->so_rerror; 3368 so->so_rerror = 0; 3369 } 3370 SOCK_UNLOCK(so); 3371 goto integer; 3372 3373 case SO_SNDBUF: 3374 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : 3375 so->so_snd.sb_hiwat; 3376 goto integer; 3377 3378 case SO_RCVBUF: 3379 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : 3380 so->so_rcv.sb_hiwat; 3381 goto integer; 3382 3383 case SO_SNDLOWAT: 3384 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : 3385 so->so_snd.sb_lowat; 3386 goto integer; 3387 3388 case SO_RCVLOWAT: 3389 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : 3390 so->so_rcv.sb_lowat; 3391 goto integer; 3392 3393 case SO_SNDTIMEO: 3394 case SO_RCVTIMEO: 3395 SOCK_LOCK(so); 3396 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 3397 (SOLISTENING(so) ? so->sol_sbsnd_timeo : 3398 so->so_snd.sb_timeo) : 3399 (SOLISTENING(so) ? so->sol_sbrcv_timeo : 3400 so->so_rcv.sb_timeo)); 3401 SOCK_UNLOCK(so); 3402 #ifdef COMPAT_FREEBSD32 3403 if (SV_CURPROC_FLAG(SV_ILP32)) { 3404 struct timeval32 tv32; 3405 3406 CP(tv, tv32, tv_sec); 3407 CP(tv, tv32, tv_usec); 3408 error = sooptcopyout(sopt, &tv32, sizeof tv32); 3409 } else 3410 #endif 3411 error = sooptcopyout(sopt, &tv, sizeof tv); 3412 break; 3413 3414 case SO_LABEL: 3415 #ifdef MAC 3416 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3417 sizeof(extmac)); 3418 if (error) 3419 goto bad; 3420 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 3421 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_PEERLABEL: 3431 #ifdef MAC 3432 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3433 sizeof(extmac)); 3434 if (error) 3435 goto bad; 3436 error = mac_getsockopt_peerlabel( 3437 sopt->sopt_td->td_ucred, so, &extmac); 3438 if (error) 3439 goto bad; 3440 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3441 #else 3442 error = EOPNOTSUPP; 3443 #endif 3444 break; 3445 3446 case SO_LISTENQLIMIT: 3447 optval = SOLISTENING(so) ? so->sol_qlimit : 0; 3448 goto integer; 3449 3450 case SO_LISTENQLEN: 3451 optval = SOLISTENING(so) ? so->sol_qlen : 0; 3452 goto integer; 3453 3454 case SO_LISTENINCQLEN: 3455 optval = SOLISTENING(so) ? so->sol_incqlen : 0; 3456 goto integer; 3457 3458 case SO_TS_CLOCK: 3459 optval = so->so_ts_clock; 3460 goto integer; 3461 3462 case SO_MAX_PACING_RATE: 3463 optval = so->so_max_pacing_rate; 3464 goto integer; 3465 3466 default: 3467 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3468 error = hhook_run_socket(so, sopt, 3469 HHOOK_SOCKET_OPT); 3470 else 3471 error = ENOPROTOOPT; 3472 break; 3473 } 3474 } 3475 #ifdef MAC 3476 bad: 3477 #endif 3478 CURVNET_RESTORE(); 3479 return (error); 3480 } 3481 3482 int 3483 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 3484 { 3485 struct mbuf *m, *m_prev; 3486 int sopt_size = sopt->sopt_valsize; 3487 3488 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3489 if (m == NULL) 3490 return ENOBUFS; 3491 if (sopt_size > MLEN) { 3492 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 3493 if ((m->m_flags & M_EXT) == 0) { 3494 m_free(m); 3495 return ENOBUFS; 3496 } 3497 m->m_len = min(MCLBYTES, sopt_size); 3498 } else { 3499 m->m_len = min(MLEN, sopt_size); 3500 } 3501 sopt_size -= m->m_len; 3502 *mp = m; 3503 m_prev = m; 3504 3505 while (sopt_size) { 3506 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3507 if (m == NULL) { 3508 m_freem(*mp); 3509 return ENOBUFS; 3510 } 3511 if (sopt_size > MLEN) { 3512 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 3513 M_NOWAIT); 3514 if ((m->m_flags & M_EXT) == 0) { 3515 m_freem(m); 3516 m_freem(*mp); 3517 return ENOBUFS; 3518 } 3519 m->m_len = min(MCLBYTES, sopt_size); 3520 } else { 3521 m->m_len = min(MLEN, sopt_size); 3522 } 3523 sopt_size -= m->m_len; 3524 m_prev->m_next = m; 3525 m_prev = m; 3526 } 3527 return (0); 3528 } 3529 3530 int 3531 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 3532 { 3533 struct mbuf *m0 = m; 3534 3535 if (sopt->sopt_val == NULL) 3536 return (0); 3537 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3538 if (sopt->sopt_td != NULL) { 3539 int error; 3540 3541 error = copyin(sopt->sopt_val, mtod(m, char *), 3542 m->m_len); 3543 if (error != 0) { 3544 m_freem(m0); 3545 return(error); 3546 } 3547 } else 3548 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 3549 sopt->sopt_valsize -= m->m_len; 3550 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3551 m = m->m_next; 3552 } 3553 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 3554 panic("ip6_sooptmcopyin"); 3555 return (0); 3556 } 3557 3558 int 3559 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 3560 { 3561 struct mbuf *m0 = m; 3562 size_t valsize = 0; 3563 3564 if (sopt->sopt_val == NULL) 3565 return (0); 3566 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3567 if (sopt->sopt_td != NULL) { 3568 int error; 3569 3570 error = copyout(mtod(m, char *), sopt->sopt_val, 3571 m->m_len); 3572 if (error != 0) { 3573 m_freem(m0); 3574 return(error); 3575 } 3576 } else 3577 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 3578 sopt->sopt_valsize -= m->m_len; 3579 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3580 valsize += m->m_len; 3581 m = m->m_next; 3582 } 3583 if (m != NULL) { 3584 /* enough soopt buffer should be given from user-land */ 3585 m_freem(m0); 3586 return(EINVAL); 3587 } 3588 sopt->sopt_valsize = valsize; 3589 return (0); 3590 } 3591 3592 /* 3593 * sohasoutofband(): protocol notifies socket layer of the arrival of new 3594 * out-of-band data, which will then notify socket consumers. 3595 */ 3596 void 3597 sohasoutofband(struct socket *so) 3598 { 3599 3600 if (so->so_sigio != NULL) 3601 pgsigio(&so->so_sigio, SIGURG, 0); 3602 selwakeuppri(&so->so_rdsel, PSOCK); 3603 } 3604 3605 int 3606 sopoll(struct socket *so, int events, struct ucred *active_cred, 3607 struct thread *td) 3608 { 3609 3610 /* 3611 * We do not need to set or assert curvnet as long as everyone uses 3612 * sopoll_generic(). 3613 */ 3614 return (so->so_proto->pr_sopoll(so, events, active_cred, td)); 3615 } 3616 3617 int 3618 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3619 struct thread *td) 3620 { 3621 int revents; 3622 3623 SOCK_LOCK(so); 3624 if (SOLISTENING(so)) { 3625 if (!(events & (POLLIN | POLLRDNORM))) 3626 revents = 0; 3627 else if (!TAILQ_EMPTY(&so->sol_comp)) 3628 revents = events & (POLLIN | POLLRDNORM); 3629 else if ((events & POLLINIGNEOF) == 0 && so->so_error) 3630 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP; 3631 else { 3632 selrecord(td, &so->so_rdsel); 3633 revents = 0; 3634 } 3635 } else { 3636 revents = 0; 3637 SOCK_SENDBUF_LOCK(so); 3638 SOCK_RECVBUF_LOCK(so); 3639 if (events & (POLLIN | POLLRDNORM)) 3640 if (soreadabledata(so)) 3641 revents |= events & (POLLIN | POLLRDNORM); 3642 if (events & (POLLOUT | POLLWRNORM)) 3643 if (sowriteable(so)) 3644 revents |= events & (POLLOUT | POLLWRNORM); 3645 if (events & (POLLPRI | POLLRDBAND)) 3646 if (so->so_oobmark || 3647 (so->so_rcv.sb_state & SBS_RCVATMARK)) 3648 revents |= events & (POLLPRI | POLLRDBAND); 3649 if ((events & POLLINIGNEOF) == 0) { 3650 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3651 revents |= events & (POLLIN | POLLRDNORM); 3652 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3653 revents |= POLLHUP; 3654 } 3655 } 3656 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3657 revents |= events & POLLRDHUP; 3658 if (revents == 0) { 3659 if (events & 3660 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) { 3661 selrecord(td, &so->so_rdsel); 3662 so->so_rcv.sb_flags |= SB_SEL; 3663 } 3664 if (events & (POLLOUT | POLLWRNORM)) { 3665 selrecord(td, &so->so_wrsel); 3666 so->so_snd.sb_flags |= SB_SEL; 3667 } 3668 } 3669 SOCK_RECVBUF_UNLOCK(so); 3670 SOCK_SENDBUF_UNLOCK(so); 3671 } 3672 SOCK_UNLOCK(so); 3673 return (revents); 3674 } 3675 3676 int 3677 soo_kqfilter(struct file *fp, struct knote *kn) 3678 { 3679 struct socket *so = kn->kn_fp->f_data; 3680 struct sockbuf *sb; 3681 sb_which which; 3682 struct knlist *knl; 3683 3684 switch (kn->kn_filter) { 3685 case EVFILT_READ: 3686 kn->kn_fop = &soread_filtops; 3687 knl = &so->so_rdsel.si_note; 3688 sb = &so->so_rcv; 3689 which = SO_RCV; 3690 break; 3691 case EVFILT_WRITE: 3692 kn->kn_fop = &sowrite_filtops; 3693 knl = &so->so_wrsel.si_note; 3694 sb = &so->so_snd; 3695 which = SO_SND; 3696 break; 3697 case EVFILT_EMPTY: 3698 kn->kn_fop = &soempty_filtops; 3699 knl = &so->so_wrsel.si_note; 3700 sb = &so->so_snd; 3701 which = SO_SND; 3702 break; 3703 default: 3704 return (EINVAL); 3705 } 3706 3707 SOCK_LOCK(so); 3708 if (SOLISTENING(so)) { 3709 knlist_add(knl, kn, 1); 3710 } else { 3711 SOCK_BUF_LOCK(so, which); 3712 knlist_add(knl, kn, 1); 3713 sb->sb_flags |= SB_KNOTE; 3714 SOCK_BUF_UNLOCK(so, which); 3715 } 3716 SOCK_UNLOCK(so); 3717 return (0); 3718 } 3719 3720 static void 3721 filt_sordetach(struct knote *kn) 3722 { 3723 struct socket *so = kn->kn_fp->f_data; 3724 3725 so_rdknl_lock(so); 3726 knlist_remove(&so->so_rdsel.si_note, kn, 1); 3727 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note)) 3728 so->so_rcv.sb_flags &= ~SB_KNOTE; 3729 so_rdknl_unlock(so); 3730 } 3731 3732 /*ARGSUSED*/ 3733 static int 3734 filt_soread(struct knote *kn, long hint) 3735 { 3736 struct socket *so; 3737 3738 so = kn->kn_fp->f_data; 3739 3740 if (SOLISTENING(so)) { 3741 SOCK_LOCK_ASSERT(so); 3742 kn->kn_data = so->sol_qlen; 3743 if (so->so_error) { 3744 kn->kn_flags |= EV_EOF; 3745 kn->kn_fflags = so->so_error; 3746 return (1); 3747 } 3748 return (!TAILQ_EMPTY(&so->sol_comp)); 3749 } 3750 3751 SOCK_RECVBUF_LOCK_ASSERT(so); 3752 3753 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 3754 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3755 kn->kn_flags |= EV_EOF; 3756 kn->kn_fflags = so->so_error; 3757 return (1); 3758 } else if (so->so_error || so->so_rerror) 3759 return (1); 3760 3761 if (kn->kn_sfflags & NOTE_LOWAT) { 3762 if (kn->kn_data >= kn->kn_sdata) 3763 return (1); 3764 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) 3765 return (1); 3766 3767 /* This hook returning non-zero indicates an event, not error */ 3768 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); 3769 } 3770 3771 static void 3772 filt_sowdetach(struct knote *kn) 3773 { 3774 struct socket *so = kn->kn_fp->f_data; 3775 3776 so_wrknl_lock(so); 3777 knlist_remove(&so->so_wrsel.si_note, kn, 1); 3778 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note)) 3779 so->so_snd.sb_flags &= ~SB_KNOTE; 3780 so_wrknl_unlock(so); 3781 } 3782 3783 /*ARGSUSED*/ 3784 static int 3785 filt_sowrite(struct knote *kn, long hint) 3786 { 3787 struct socket *so; 3788 3789 so = kn->kn_fp->f_data; 3790 3791 if (SOLISTENING(so)) 3792 return (0); 3793 3794 SOCK_SENDBUF_LOCK_ASSERT(so); 3795 kn->kn_data = sbspace(&so->so_snd); 3796 3797 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); 3798 3799 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3800 kn->kn_flags |= EV_EOF; 3801 kn->kn_fflags = so->so_error; 3802 return (1); 3803 } else if (so->so_error) /* temporary udp error */ 3804 return (1); 3805 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3806 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3807 return (0); 3808 else if (kn->kn_sfflags & NOTE_LOWAT) 3809 return (kn->kn_data >= kn->kn_sdata); 3810 else 3811 return (kn->kn_data >= so->so_snd.sb_lowat); 3812 } 3813 3814 static int 3815 filt_soempty(struct knote *kn, long hint) 3816 { 3817 struct socket *so; 3818 3819 so = kn->kn_fp->f_data; 3820 3821 if (SOLISTENING(so)) 3822 return (1); 3823 3824 SOCK_SENDBUF_LOCK_ASSERT(so); 3825 kn->kn_data = sbused(&so->so_snd); 3826 3827 if (kn->kn_data == 0) 3828 return (1); 3829 else 3830 return (0); 3831 } 3832 3833 int 3834 socheckuid(struct socket *so, uid_t uid) 3835 { 3836 3837 if (so == NULL) 3838 return (EPERM); 3839 if (so->so_cred->cr_uid != uid) 3840 return (EPERM); 3841 return (0); 3842 } 3843 3844 /* 3845 * These functions are used by protocols to notify the socket layer (and its 3846 * consumers) of state changes in the sockets driven by protocol-side events. 3847 */ 3848 3849 /* 3850 * Procedures to manipulate state flags of socket and do appropriate wakeups. 3851 * 3852 * Normal sequence from the active (originating) side is that 3853 * soisconnecting() is called during processing of connect() call, resulting 3854 * in an eventual call to soisconnected() if/when the connection is 3855 * established. When the connection is torn down soisdisconnecting() is 3856 * called during processing of disconnect() call, and soisdisconnected() is 3857 * called when the connection to the peer is totally severed. The semantics 3858 * of these routines are such that connectionless protocols can call 3859 * soisconnected() and soisdisconnected() only, bypassing the in-progress 3860 * calls when setting up a ``connection'' takes no time. 3861 * 3862 * From the passive side, a socket is created with two queues of sockets: 3863 * so_incomp for connections in progress and so_comp for connections already 3864 * made and awaiting user acceptance. As a protocol is preparing incoming 3865 * connections, it creates a socket structure queued on so_incomp by calling 3866 * sonewconn(). When the connection is established, soisconnected() is 3867 * called, and transfers the socket structure to so_comp, making it available 3868 * to accept(). 3869 * 3870 * If a socket is closed with sockets on either so_incomp or so_comp, these 3871 * sockets are dropped. 3872 * 3873 * If higher-level protocols are implemented in the kernel, the wakeups done 3874 * here will sometimes cause software-interrupt process scheduling. 3875 */ 3876 void 3877 soisconnecting(struct socket *so) 3878 { 3879 3880 SOCK_LOCK(so); 3881 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 3882 so->so_state |= SS_ISCONNECTING; 3883 SOCK_UNLOCK(so); 3884 } 3885 3886 void 3887 soisconnected(struct socket *so) 3888 { 3889 bool last __diagused; 3890 3891 SOCK_LOCK(so); 3892 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 3893 so->so_state |= SS_ISCONNECTED; 3894 3895 if (so->so_qstate == SQ_INCOMP) { 3896 struct socket *head = so->so_listen; 3897 int ret; 3898 3899 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so)); 3900 /* 3901 * Promoting a socket from incomplete queue to complete, we 3902 * need to go through reverse order of locking. We first do 3903 * trylock, and if that doesn't succeed, we go the hard way 3904 * leaving a reference and rechecking consistency after proper 3905 * locking. 3906 */ 3907 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) { 3908 soref(head); 3909 SOCK_UNLOCK(so); 3910 SOLISTEN_LOCK(head); 3911 SOCK_LOCK(so); 3912 if (__predict_false(head != so->so_listen)) { 3913 /* 3914 * The socket went off the listen queue, 3915 * should be lost race to close(2) of sol. 3916 * The socket is about to soabort(). 3917 */ 3918 SOCK_UNLOCK(so); 3919 sorele_locked(head); 3920 return; 3921 } 3922 last = refcount_release(&head->so_count); 3923 KASSERT(!last, ("%s: released last reference for %p", 3924 __func__, head)); 3925 } 3926 again: 3927 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 3928 TAILQ_REMOVE(&head->sol_incomp, so, so_list); 3929 head->sol_incqlen--; 3930 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 3931 head->sol_qlen++; 3932 so->so_qstate = SQ_COMP; 3933 SOCK_UNLOCK(so); 3934 solisten_wakeup(head); /* unlocks */ 3935 } else { 3936 SOCK_RECVBUF_LOCK(so); 3937 soupcall_set(so, SO_RCV, 3938 head->sol_accept_filter->accf_callback, 3939 head->sol_accept_filter_arg); 3940 so->so_options &= ~SO_ACCEPTFILTER; 3941 ret = head->sol_accept_filter->accf_callback(so, 3942 head->sol_accept_filter_arg, M_NOWAIT); 3943 if (ret == SU_ISCONNECTED) { 3944 soupcall_clear(so, SO_RCV); 3945 SOCK_RECVBUF_UNLOCK(so); 3946 goto again; 3947 } 3948 SOCK_RECVBUF_UNLOCK(so); 3949 SOCK_UNLOCK(so); 3950 SOLISTEN_UNLOCK(head); 3951 } 3952 return; 3953 } 3954 SOCK_UNLOCK(so); 3955 wakeup(&so->so_timeo); 3956 sorwakeup(so); 3957 sowwakeup(so); 3958 } 3959 3960 void 3961 soisdisconnecting(struct socket *so) 3962 { 3963 3964 SOCK_LOCK(so); 3965 so->so_state &= ~SS_ISCONNECTING; 3966 so->so_state |= SS_ISDISCONNECTING; 3967 3968 if (!SOLISTENING(so)) { 3969 SOCK_RECVBUF_LOCK(so); 3970 socantrcvmore_locked(so); 3971 SOCK_SENDBUF_LOCK(so); 3972 socantsendmore_locked(so); 3973 } 3974 SOCK_UNLOCK(so); 3975 wakeup(&so->so_timeo); 3976 } 3977 3978 void 3979 soisdisconnected(struct socket *so) 3980 { 3981 3982 SOCK_LOCK(so); 3983 3984 /* 3985 * There is at least one reader of so_state that does not 3986 * acquire socket lock, namely soreceive_generic(). Ensure 3987 * that it never sees all flags that track connection status 3988 * cleared, by ordering the update with a barrier semantic of 3989 * our release thread fence. 3990 */ 3991 so->so_state |= SS_ISDISCONNECTED; 3992 atomic_thread_fence_rel(); 3993 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 3994 3995 if (!SOLISTENING(so)) { 3996 SOCK_UNLOCK(so); 3997 SOCK_RECVBUF_LOCK(so); 3998 socantrcvmore_locked(so); 3999 SOCK_SENDBUF_LOCK(so); 4000 sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); 4001 socantsendmore_locked(so); 4002 } else 4003 SOCK_UNLOCK(so); 4004 wakeup(&so->so_timeo); 4005 } 4006 4007 int 4008 soiolock(struct socket *so, struct sx *sx, int flags) 4009 { 4010 int error; 4011 4012 KASSERT((flags & SBL_VALID) == flags, 4013 ("soiolock: invalid flags %#x", flags)); 4014 4015 if ((flags & SBL_WAIT) != 0) { 4016 if ((flags & SBL_NOINTR) != 0) { 4017 sx_xlock(sx); 4018 } else { 4019 error = sx_xlock_sig(sx); 4020 if (error != 0) 4021 return (error); 4022 } 4023 } else if (!sx_try_xlock(sx)) { 4024 return (EWOULDBLOCK); 4025 } 4026 4027 if (__predict_false(SOLISTENING(so))) { 4028 sx_xunlock(sx); 4029 return (ENOTCONN); 4030 } 4031 return (0); 4032 } 4033 4034 void 4035 soiounlock(struct sx *sx) 4036 { 4037 sx_xunlock(sx); 4038 } 4039 4040 /* 4041 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 4042 */ 4043 struct sockaddr * 4044 sodupsockaddr(const struct sockaddr *sa, int mflags) 4045 { 4046 struct sockaddr *sa2; 4047 4048 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 4049 if (sa2) 4050 bcopy(sa, sa2, sa->sa_len); 4051 return sa2; 4052 } 4053 4054 /* 4055 * Register per-socket destructor. 4056 */ 4057 void 4058 sodtor_set(struct socket *so, so_dtor_t *func) 4059 { 4060 4061 SOCK_LOCK_ASSERT(so); 4062 so->so_dtor = func; 4063 } 4064 4065 /* 4066 * Register per-socket buffer upcalls. 4067 */ 4068 void 4069 soupcall_set(struct socket *so, sb_which which, so_upcall_t func, void *arg) 4070 { 4071 struct sockbuf *sb; 4072 4073 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4074 4075 switch (which) { 4076 case SO_RCV: 4077 sb = &so->so_rcv; 4078 break; 4079 case SO_SND: 4080 sb = &so->so_snd; 4081 break; 4082 } 4083 SOCK_BUF_LOCK_ASSERT(so, which); 4084 sb->sb_upcall = func; 4085 sb->sb_upcallarg = arg; 4086 sb->sb_flags |= SB_UPCALL; 4087 } 4088 4089 void 4090 soupcall_clear(struct socket *so, sb_which which) 4091 { 4092 struct sockbuf *sb; 4093 4094 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4095 4096 switch (which) { 4097 case SO_RCV: 4098 sb = &so->so_rcv; 4099 break; 4100 case SO_SND: 4101 sb = &so->so_snd; 4102 break; 4103 } 4104 SOCK_BUF_LOCK_ASSERT(so, which); 4105 KASSERT(sb->sb_upcall != NULL, 4106 ("%s: so %p no upcall to clear", __func__, so)); 4107 sb->sb_upcall = NULL; 4108 sb->sb_upcallarg = NULL; 4109 sb->sb_flags &= ~SB_UPCALL; 4110 } 4111 4112 void 4113 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg) 4114 { 4115 4116 SOLISTEN_LOCK_ASSERT(so); 4117 so->sol_upcall = func; 4118 so->sol_upcallarg = arg; 4119 } 4120 4121 static void 4122 so_rdknl_lock(void *arg) 4123 { 4124 struct socket *so = arg; 4125 4126 retry: 4127 if (SOLISTENING(so)) { 4128 SOLISTEN_LOCK(so); 4129 } else { 4130 SOCK_RECVBUF_LOCK(so); 4131 if (__predict_false(SOLISTENING(so))) { 4132 SOCK_RECVBUF_UNLOCK(so); 4133 goto retry; 4134 } 4135 } 4136 } 4137 4138 static void 4139 so_rdknl_unlock(void *arg) 4140 { 4141 struct socket *so = arg; 4142 4143 if (SOLISTENING(so)) 4144 SOLISTEN_UNLOCK(so); 4145 else 4146 SOCK_RECVBUF_UNLOCK(so); 4147 } 4148 4149 static void 4150 so_rdknl_assert_lock(void *arg, int what) 4151 { 4152 struct socket *so = arg; 4153 4154 if (what == LA_LOCKED) { 4155 if (SOLISTENING(so)) 4156 SOLISTEN_LOCK_ASSERT(so); 4157 else 4158 SOCK_RECVBUF_LOCK_ASSERT(so); 4159 } else { 4160 if (SOLISTENING(so)) 4161 SOLISTEN_UNLOCK_ASSERT(so); 4162 else 4163 SOCK_RECVBUF_UNLOCK_ASSERT(so); 4164 } 4165 } 4166 4167 static void 4168 so_wrknl_lock(void *arg) 4169 { 4170 struct socket *so = arg; 4171 4172 retry: 4173 if (SOLISTENING(so)) { 4174 SOLISTEN_LOCK(so); 4175 } else { 4176 SOCK_SENDBUF_LOCK(so); 4177 if (__predict_false(SOLISTENING(so))) { 4178 SOCK_SENDBUF_UNLOCK(so); 4179 goto retry; 4180 } 4181 } 4182 } 4183 4184 static void 4185 so_wrknl_unlock(void *arg) 4186 { 4187 struct socket *so = arg; 4188 4189 if (SOLISTENING(so)) 4190 SOLISTEN_UNLOCK(so); 4191 else 4192 SOCK_SENDBUF_UNLOCK(so); 4193 } 4194 4195 static void 4196 so_wrknl_assert_lock(void *arg, int what) 4197 { 4198 struct socket *so = arg; 4199 4200 if (what == LA_LOCKED) { 4201 if (SOLISTENING(so)) 4202 SOLISTEN_LOCK_ASSERT(so); 4203 else 4204 SOCK_SENDBUF_LOCK_ASSERT(so); 4205 } else { 4206 if (SOLISTENING(so)) 4207 SOLISTEN_UNLOCK_ASSERT(so); 4208 else 4209 SOCK_SENDBUF_UNLOCK_ASSERT(so); 4210 } 4211 } 4212 4213 /* 4214 * Create an external-format (``xsocket'') structure using the information in 4215 * the kernel-format socket structure pointed to by so. This is done to 4216 * reduce the spew of irrelevant information over this interface, to isolate 4217 * user code from changes in the kernel structure, and potentially to provide 4218 * information-hiding if we decide that some of this information should be 4219 * hidden from users. 4220 */ 4221 void 4222 sotoxsocket(struct socket *so, struct xsocket *xso) 4223 { 4224 4225 bzero(xso, sizeof(*xso)); 4226 xso->xso_len = sizeof *xso; 4227 xso->xso_so = (uintptr_t)so; 4228 xso->so_type = so->so_type; 4229 xso->so_options = so->so_options; 4230 xso->so_linger = so->so_linger; 4231 xso->so_state = so->so_state; 4232 xso->so_pcb = (uintptr_t)so->so_pcb; 4233 xso->xso_protocol = so->so_proto->pr_protocol; 4234 xso->xso_family = so->so_proto->pr_domain->dom_family; 4235 xso->so_timeo = so->so_timeo; 4236 xso->so_error = so->so_error; 4237 xso->so_uid = so->so_cred->cr_uid; 4238 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 4239 if (SOLISTENING(so)) { 4240 xso->so_qlen = so->sol_qlen; 4241 xso->so_incqlen = so->sol_incqlen; 4242 xso->so_qlimit = so->sol_qlimit; 4243 xso->so_oobmark = 0; 4244 } else { 4245 xso->so_state |= so->so_qstate; 4246 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0; 4247 xso->so_oobmark = so->so_oobmark; 4248 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 4249 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 4250 } 4251 } 4252 4253 struct sockbuf * 4254 so_sockbuf_rcv(struct socket *so) 4255 { 4256 4257 return (&so->so_rcv); 4258 } 4259 4260 struct sockbuf * 4261 so_sockbuf_snd(struct socket *so) 4262 { 4263 4264 return (&so->so_snd); 4265 } 4266 4267 int 4268 so_state_get(const struct socket *so) 4269 { 4270 4271 return (so->so_state); 4272 } 4273 4274 void 4275 so_state_set(struct socket *so, int val) 4276 { 4277 4278 so->so_state = val; 4279 } 4280 4281 int 4282 so_options_get(const struct socket *so) 4283 { 4284 4285 return (so->so_options); 4286 } 4287 4288 void 4289 so_options_set(struct socket *so, int val) 4290 { 4291 4292 so->so_options = val; 4293 } 4294 4295 int 4296 so_error_get(const struct socket *so) 4297 { 4298 4299 return (so->so_error); 4300 } 4301 4302 void 4303 so_error_set(struct socket *so, int val) 4304 { 4305 4306 so->so_error = val; 4307 } 4308 4309 int 4310 so_linger_get(const struct socket *so) 4311 { 4312 4313 return (so->so_linger); 4314 } 4315 4316 void 4317 so_linger_set(struct socket *so, int val) 4318 { 4319 4320 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz), 4321 ("%s: val %d out of range", __func__, val)); 4322 4323 so->so_linger = val; 4324 } 4325 4326 struct protosw * 4327 so_protosw_get(const struct socket *so) 4328 { 4329 4330 return (so->so_proto); 4331 } 4332 4333 void 4334 so_protosw_set(struct socket *so, struct protosw *val) 4335 { 4336 4337 so->so_proto = val; 4338 } 4339 4340 void 4341 so_sorwakeup(struct socket *so) 4342 { 4343 4344 sorwakeup(so); 4345 } 4346 4347 void 4348 so_sowwakeup(struct socket *so) 4349 { 4350 4351 sowwakeup(so); 4352 } 4353 4354 void 4355 so_sorwakeup_locked(struct socket *so) 4356 { 4357 4358 sorwakeup_locked(so); 4359 } 4360 4361 void 4362 so_sowwakeup_locked(struct socket *so) 4363 { 4364 4365 sowwakeup_locked(so); 4366 } 4367 4368 void 4369 so_lock(struct socket *so) 4370 { 4371 4372 SOCK_LOCK(so); 4373 } 4374 4375 void 4376 so_unlock(struct socket *so) 4377 { 4378 4379 SOCK_UNLOCK(so); 4380 } 4381