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(). */ 991 if (!SOLISTENING(so) && 992 ((so->so_snd.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0 || 993 (so->so_rcv.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0)) { 994 solisten_proto_abort(so); 995 return (EINVAL); 996 } 997 return (0); 998 } 999 1000 /* 1001 * Undo the setup done by solisten_proto_check(). 1002 */ 1003 void 1004 solisten_proto_abort(struct socket *so) 1005 { 1006 mtx_unlock(&so->so_snd_mtx); 1007 mtx_unlock(&so->so_rcv_mtx); 1008 sx_xunlock(&so->so_snd_sx); 1009 sx_xunlock(&so->so_rcv_sx); 1010 } 1011 1012 void 1013 solisten_proto(struct socket *so, int backlog) 1014 { 1015 int sbrcv_lowat, sbsnd_lowat; 1016 u_int sbrcv_hiwat, sbsnd_hiwat; 1017 short sbrcv_flags, sbsnd_flags; 1018 sbintime_t sbrcv_timeo, sbsnd_timeo; 1019 1020 SOCK_LOCK_ASSERT(so); 1021 KASSERT((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 1022 SS_ISDISCONNECTING)) == 0, 1023 ("%s: bad socket state %p", __func__, so)); 1024 1025 if (SOLISTENING(so)) 1026 goto listening; 1027 1028 /* 1029 * Change this socket to listening state. 1030 */ 1031 sbrcv_lowat = so->so_rcv.sb_lowat; 1032 sbsnd_lowat = so->so_snd.sb_lowat; 1033 sbrcv_hiwat = so->so_rcv.sb_hiwat; 1034 sbsnd_hiwat = so->so_snd.sb_hiwat; 1035 sbrcv_flags = so->so_rcv.sb_flags; 1036 sbsnd_flags = so->so_snd.sb_flags; 1037 sbrcv_timeo = so->so_rcv.sb_timeo; 1038 sbsnd_timeo = so->so_snd.sb_timeo; 1039 1040 sbdestroy(so, SO_SND); 1041 sbdestroy(so, SO_RCV); 1042 1043 #ifdef INVARIANTS 1044 bzero(&so->so_rcv, 1045 sizeof(struct socket) - offsetof(struct socket, so_rcv)); 1046 #endif 1047 1048 so->sol_sbrcv_lowat = sbrcv_lowat; 1049 so->sol_sbsnd_lowat = sbsnd_lowat; 1050 so->sol_sbrcv_hiwat = sbrcv_hiwat; 1051 so->sol_sbsnd_hiwat = sbsnd_hiwat; 1052 so->sol_sbrcv_flags = sbrcv_flags; 1053 so->sol_sbsnd_flags = sbsnd_flags; 1054 so->sol_sbrcv_timeo = sbrcv_timeo; 1055 so->sol_sbsnd_timeo = sbsnd_timeo; 1056 1057 so->sol_qlen = so->sol_incqlen = 0; 1058 TAILQ_INIT(&so->sol_incomp); 1059 TAILQ_INIT(&so->sol_comp); 1060 1061 so->sol_accept_filter = NULL; 1062 so->sol_accept_filter_arg = NULL; 1063 so->sol_accept_filter_str = NULL; 1064 1065 so->sol_upcall = NULL; 1066 so->sol_upcallarg = NULL; 1067 1068 so->so_options |= SO_ACCEPTCONN; 1069 1070 listening: 1071 if (backlog < 0 || backlog > somaxconn) 1072 backlog = somaxconn; 1073 so->sol_qlimit = backlog; 1074 1075 mtx_unlock(&so->so_snd_mtx); 1076 mtx_unlock(&so->so_rcv_mtx); 1077 sx_xunlock(&so->so_snd_sx); 1078 sx_xunlock(&so->so_rcv_sx); 1079 } 1080 1081 /* 1082 * Wakeup listeners/subsystems once we have a complete connection. 1083 * Enters with lock, returns unlocked. 1084 */ 1085 void 1086 solisten_wakeup(struct socket *sol) 1087 { 1088 1089 if (sol->sol_upcall != NULL) 1090 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT); 1091 else { 1092 selwakeuppri(&sol->so_rdsel, PSOCK); 1093 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0); 1094 } 1095 SOLISTEN_UNLOCK(sol); 1096 wakeup_one(&sol->sol_comp); 1097 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL) 1098 pgsigio(&sol->so_sigio, SIGIO, 0); 1099 } 1100 1101 /* 1102 * Return single connection off a listening socket queue. Main consumer of 1103 * the function is kern_accept4(). Some modules, that do their own accept 1104 * management also use the function. The socket reference held by the 1105 * listen queue is handed to the caller. 1106 * 1107 * Listening socket must be locked on entry and is returned unlocked on 1108 * return. 1109 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT. 1110 */ 1111 int 1112 solisten_dequeue(struct socket *head, struct socket **ret, int flags) 1113 { 1114 struct socket *so; 1115 int error; 1116 1117 SOLISTEN_LOCK_ASSERT(head); 1118 1119 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) && 1120 head->so_error == 0) { 1121 error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH, 1122 "accept", 0); 1123 if (error != 0) { 1124 SOLISTEN_UNLOCK(head); 1125 return (error); 1126 } 1127 } 1128 if (head->so_error) { 1129 error = head->so_error; 1130 head->so_error = 0; 1131 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp)) 1132 error = EWOULDBLOCK; 1133 else 1134 error = 0; 1135 if (error) { 1136 SOLISTEN_UNLOCK(head); 1137 return (error); 1138 } 1139 so = TAILQ_FIRST(&head->sol_comp); 1140 SOCK_LOCK(so); 1141 KASSERT(so->so_qstate == SQ_COMP, 1142 ("%s: so %p not SQ_COMP", __func__, so)); 1143 head->sol_qlen--; 1144 so->so_qstate = SQ_NONE; 1145 so->so_listen = NULL; 1146 TAILQ_REMOVE(&head->sol_comp, so, so_list); 1147 if (flags & ACCEPT4_INHERIT) 1148 so->so_state |= (head->so_state & SS_NBIO); 1149 else 1150 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0; 1151 SOCK_UNLOCK(so); 1152 sorele_locked(head); 1153 1154 *ret = so; 1155 return (0); 1156 } 1157 1158 /* 1159 * Free socket upon release of the very last reference. 1160 */ 1161 static void 1162 sofree(struct socket *so) 1163 { 1164 struct protosw *pr = so->so_proto; 1165 1166 SOCK_LOCK_ASSERT(so); 1167 KASSERT(refcount_load(&so->so_count) == 0, 1168 ("%s: so %p has references", __func__, so)); 1169 KASSERT(SOLISTENING(so) || so->so_qstate == SQ_NONE, 1170 ("%s: so %p is on listen queue", __func__, so)); 1171 1172 SOCK_UNLOCK(so); 1173 1174 if (so->so_dtor != NULL) 1175 so->so_dtor(so); 1176 1177 VNET_SO_ASSERT(so); 1178 if ((pr->pr_flags & PR_RIGHTS) && !SOLISTENING(so)) { 1179 MPASS(pr->pr_domain->dom_dispose != NULL); 1180 (*pr->pr_domain->dom_dispose)(so); 1181 } 1182 if (pr->pr_detach != NULL) 1183 pr->pr_detach(so); 1184 1185 /* 1186 * From this point on, we assume that no other references to this 1187 * socket exist anywhere else in the stack. Therefore, no locks need 1188 * to be acquired or held. 1189 */ 1190 if (!(pr->pr_flags & PR_SOCKBUF) && !SOLISTENING(so)) { 1191 sbdestroy(so, SO_SND); 1192 sbdestroy(so, SO_RCV); 1193 } 1194 seldrain(&so->so_rdsel); 1195 seldrain(&so->so_wrsel); 1196 knlist_destroy(&so->so_rdsel.si_note); 1197 knlist_destroy(&so->so_wrsel.si_note); 1198 sodealloc(so); 1199 } 1200 1201 /* 1202 * Release a reference on a socket while holding the socket lock. 1203 * Unlocks the socket lock before returning. 1204 */ 1205 void 1206 sorele_locked(struct socket *so) 1207 { 1208 SOCK_LOCK_ASSERT(so); 1209 if (refcount_release(&so->so_count)) 1210 sofree(so); 1211 else 1212 SOCK_UNLOCK(so); 1213 } 1214 1215 /* 1216 * Close a socket on last file table reference removal. Initiate disconnect 1217 * if connected. Free socket when disconnect complete. 1218 * 1219 * This function will sorele() the socket. Note that soclose() may be called 1220 * prior to the ref count reaching zero. The actual socket structure will 1221 * not be freed until the ref count reaches zero. 1222 */ 1223 int 1224 soclose(struct socket *so) 1225 { 1226 struct accept_queue lqueue; 1227 int error = 0; 1228 bool listening, last __diagused; 1229 1230 CURVNET_SET(so->so_vnet); 1231 funsetown(&so->so_sigio); 1232 if (so->so_state & SS_ISCONNECTED) { 1233 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 1234 error = sodisconnect(so); 1235 if (error) { 1236 if (error == ENOTCONN) 1237 error = 0; 1238 goto drop; 1239 } 1240 } 1241 1242 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) { 1243 if ((so->so_state & SS_ISDISCONNECTING) && 1244 (so->so_state & SS_NBIO)) 1245 goto drop; 1246 while (so->so_state & SS_ISCONNECTED) { 1247 error = tsleep(&so->so_timeo, 1248 PSOCK | PCATCH, "soclos", 1249 so->so_linger * hz); 1250 if (error) 1251 break; 1252 } 1253 } 1254 } 1255 1256 drop: 1257 if (so->so_proto->pr_close != NULL) 1258 so->so_proto->pr_close(so); 1259 1260 SOCK_LOCK(so); 1261 if ((listening = SOLISTENING(so))) { 1262 struct socket *sp; 1263 1264 TAILQ_INIT(&lqueue); 1265 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list); 1266 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list); 1267 1268 so->sol_qlen = so->sol_incqlen = 0; 1269 1270 TAILQ_FOREACH(sp, &lqueue, so_list) { 1271 SOCK_LOCK(sp); 1272 sp->so_qstate = SQ_NONE; 1273 sp->so_listen = NULL; 1274 SOCK_UNLOCK(sp); 1275 last = refcount_release(&so->so_count); 1276 KASSERT(!last, ("%s: released last reference for %p", 1277 __func__, so)); 1278 } 1279 } 1280 sorele_locked(so); 1281 if (listening) { 1282 struct socket *sp, *tsp; 1283 1284 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) 1285 soabort(sp); 1286 } 1287 CURVNET_RESTORE(); 1288 return (error); 1289 } 1290 1291 /* 1292 * soabort() is used to abruptly tear down a connection, such as when a 1293 * resource limit is reached (listen queue depth exceeded), or if a listen 1294 * socket is closed while there are sockets waiting to be accepted. 1295 * 1296 * This interface is tricky, because it is called on an unreferenced socket, 1297 * and must be called only by a thread that has actually removed the socket 1298 * from the listen queue it was on. Likely this thread holds the last 1299 * reference on the socket and soabort() will proceed with sofree(). But 1300 * it might be not the last, as the sockets on the listen queues are seen 1301 * from the protocol side. 1302 * 1303 * This interface will call into the protocol code, so must not be called 1304 * with any socket locks held. Protocols do call it while holding their own 1305 * recursible protocol mutexes, but this is something that should be subject 1306 * to review in the future. 1307 * 1308 * Usually socket should have a single reference left, but this is not a 1309 * requirement. In the past, when we have had named references for file 1310 * descriptor and protocol, we asserted that none of them are being held. 1311 */ 1312 void 1313 soabort(struct socket *so) 1314 { 1315 1316 VNET_SO_ASSERT(so); 1317 1318 if (so->so_proto->pr_abort != NULL) 1319 so->so_proto->pr_abort(so); 1320 SOCK_LOCK(so); 1321 sorele_locked(so); 1322 } 1323 1324 int 1325 soaccept(struct socket *so, struct sockaddr **nam) 1326 { 1327 int error; 1328 1329 CURVNET_SET(so->so_vnet); 1330 error = so->so_proto->pr_accept(so, nam); 1331 CURVNET_RESTORE(); 1332 return (error); 1333 } 1334 1335 int 1336 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 1337 { 1338 1339 return (soconnectat(AT_FDCWD, so, nam, td)); 1340 } 1341 1342 int 1343 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 1344 { 1345 int error; 1346 1347 CURVNET_SET(so->so_vnet); 1348 /* 1349 * If protocol is connection-based, can only connect once. 1350 * Otherwise, if connected, try to disconnect first. This allows 1351 * user to disconnect by connecting to, e.g., a null address. 1352 */ 1353 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 1354 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 1355 (error = sodisconnect(so)))) { 1356 error = EISCONN; 1357 } else { 1358 /* 1359 * Prevent accumulated error from previous connection from 1360 * biting us. 1361 */ 1362 so->so_error = 0; 1363 if (fd == AT_FDCWD) { 1364 error = so->so_proto->pr_connect(so, nam, td); 1365 } else { 1366 error = so->so_proto->pr_connectat(fd, so, nam, td); 1367 } 1368 } 1369 CURVNET_RESTORE(); 1370 1371 return (error); 1372 } 1373 1374 int 1375 soconnect2(struct socket *so1, struct socket *so2) 1376 { 1377 int error; 1378 1379 CURVNET_SET(so1->so_vnet); 1380 error = so1->so_proto->pr_connect2(so1, so2); 1381 CURVNET_RESTORE(); 1382 return (error); 1383 } 1384 1385 int 1386 sodisconnect(struct socket *so) 1387 { 1388 int error; 1389 1390 if ((so->so_state & SS_ISCONNECTED) == 0) 1391 return (ENOTCONN); 1392 if (so->so_state & SS_ISDISCONNECTING) 1393 return (EALREADY); 1394 VNET_SO_ASSERT(so); 1395 error = so->so_proto->pr_disconnect(so); 1396 return (error); 1397 } 1398 1399 int 1400 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1401 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1402 { 1403 long space; 1404 ssize_t resid; 1405 int clen = 0, error, dontroute; 1406 1407 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 1408 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 1409 ("sosend_dgram: !PR_ATOMIC")); 1410 1411 if (uio != NULL) 1412 resid = uio->uio_resid; 1413 else 1414 resid = top->m_pkthdr.len; 1415 /* 1416 * In theory resid should be unsigned. However, space must be 1417 * signed, as it might be less than 0 if we over-committed, and we 1418 * must use a signed comparison of space and resid. On the other 1419 * hand, a negative resid causes us to loop sending 0-length 1420 * segments to the protocol. 1421 */ 1422 if (resid < 0) { 1423 error = EINVAL; 1424 goto out; 1425 } 1426 1427 dontroute = 1428 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 1429 if (td != NULL) 1430 td->td_ru.ru_msgsnd++; 1431 if (control != NULL) 1432 clen = control->m_len; 1433 1434 SOCKBUF_LOCK(&so->so_snd); 1435 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1436 SOCKBUF_UNLOCK(&so->so_snd); 1437 error = EPIPE; 1438 goto out; 1439 } 1440 if (so->so_error) { 1441 error = so->so_error; 1442 so->so_error = 0; 1443 SOCKBUF_UNLOCK(&so->so_snd); 1444 goto out; 1445 } 1446 if ((so->so_state & SS_ISCONNECTED) == 0) { 1447 /* 1448 * `sendto' and `sendmsg' is allowed on a connection-based 1449 * socket if it supports implied connect. Return ENOTCONN if 1450 * not connected and no address is supplied. 1451 */ 1452 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1453 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1454 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1455 !(resid == 0 && clen != 0)) { 1456 SOCKBUF_UNLOCK(&so->so_snd); 1457 error = ENOTCONN; 1458 goto out; 1459 } 1460 } else if (addr == NULL) { 1461 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1462 error = ENOTCONN; 1463 else 1464 error = EDESTADDRREQ; 1465 SOCKBUF_UNLOCK(&so->so_snd); 1466 goto out; 1467 } 1468 } 1469 1470 /* 1471 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1472 * problem and need fixing. 1473 */ 1474 space = sbspace(&so->so_snd); 1475 if (flags & MSG_OOB) 1476 space += 1024; 1477 space -= clen; 1478 SOCKBUF_UNLOCK(&so->so_snd); 1479 if (resid > space) { 1480 error = EMSGSIZE; 1481 goto out; 1482 } 1483 if (uio == NULL) { 1484 resid = 0; 1485 if (flags & MSG_EOR) 1486 top->m_flags |= M_EOR; 1487 } else { 1488 /* 1489 * Copy the data from userland into a mbuf chain. 1490 * If no data is to be copied in, a single empty mbuf 1491 * is returned. 1492 */ 1493 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1494 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1495 if (top == NULL) { 1496 error = EFAULT; /* only possible error */ 1497 goto out; 1498 } 1499 space -= resid - uio->uio_resid; 1500 resid = uio->uio_resid; 1501 } 1502 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1503 /* 1504 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1505 * than with. 1506 */ 1507 if (dontroute) { 1508 SOCK_LOCK(so); 1509 so->so_options |= SO_DONTROUTE; 1510 SOCK_UNLOCK(so); 1511 } 1512 /* 1513 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1514 * of date. We could have received a reset packet in an interrupt or 1515 * maybe we slept while doing page faults in uiomove() etc. We could 1516 * probably recheck again inside the locking protection here, but 1517 * there are probably other places that this also happens. We must 1518 * rethink this. 1519 */ 1520 VNET_SO_ASSERT(so); 1521 error = so->so_proto->pr_send(so, (flags & MSG_OOB) ? PRUS_OOB : 1522 /* 1523 * If the user set MSG_EOF, the protocol understands this flag and 1524 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1525 */ 1526 ((flags & MSG_EOF) && 1527 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1528 (resid <= 0)) ? 1529 PRUS_EOF : 1530 /* If there is more to send set PRUS_MORETOCOME */ 1531 (flags & MSG_MORETOCOME) || 1532 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1533 top, addr, control, td); 1534 if (dontroute) { 1535 SOCK_LOCK(so); 1536 so->so_options &= ~SO_DONTROUTE; 1537 SOCK_UNLOCK(so); 1538 } 1539 clen = 0; 1540 control = NULL; 1541 top = NULL; 1542 out: 1543 if (top != NULL) 1544 m_freem(top); 1545 if (control != NULL) 1546 m_freem(control); 1547 return (error); 1548 } 1549 1550 /* 1551 * Send on a socket. If send must go all at once and message is larger than 1552 * send buffering, then hard error. Lock against other senders. If must go 1553 * all at once and not enough room now, then inform user that this would 1554 * block and do nothing. Otherwise, if nonblocking, send as much as 1555 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1556 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1557 * in mbuf chain must be small enough to send all at once. 1558 * 1559 * Returns nonzero on error, timeout or signal; callers must check for short 1560 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1561 * on return. 1562 */ 1563 int 1564 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1565 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1566 { 1567 long space; 1568 ssize_t resid; 1569 int clen = 0, error, dontroute; 1570 int atomic = sosendallatonce(so) || top; 1571 int pr_send_flag; 1572 #ifdef KERN_TLS 1573 struct ktls_session *tls; 1574 int tls_enq_cnt, tls_send_flag; 1575 uint8_t tls_rtype; 1576 1577 tls = NULL; 1578 tls_rtype = TLS_RLTYPE_APP; 1579 #endif 1580 if (uio != NULL) 1581 resid = uio->uio_resid; 1582 else if ((top->m_flags & M_PKTHDR) != 0) 1583 resid = top->m_pkthdr.len; 1584 else 1585 resid = m_length(top, NULL); 1586 /* 1587 * In theory resid should be unsigned. However, space must be 1588 * signed, as it might be less than 0 if we over-committed, and we 1589 * must use a signed comparison of space and resid. On the other 1590 * hand, a negative resid causes us to loop sending 0-length 1591 * segments to the protocol. 1592 * 1593 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1594 * type sockets since that's an error. 1595 */ 1596 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1597 error = EINVAL; 1598 goto out; 1599 } 1600 1601 dontroute = 1602 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1603 (so->so_proto->pr_flags & PR_ATOMIC); 1604 if (td != NULL) 1605 td->td_ru.ru_msgsnd++; 1606 if (control != NULL) 1607 clen = control->m_len; 1608 1609 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 1610 if (error) 1611 goto out; 1612 1613 #ifdef KERN_TLS 1614 tls_send_flag = 0; 1615 tls = ktls_hold(so->so_snd.sb_tls_info); 1616 if (tls != NULL) { 1617 if (tls->mode == TCP_TLS_MODE_SW) 1618 tls_send_flag = PRUS_NOTREADY; 1619 1620 if (control != NULL) { 1621 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1622 1623 if (clen >= sizeof(*cm) && 1624 cm->cmsg_type == TLS_SET_RECORD_TYPE) { 1625 tls_rtype = *((uint8_t *)CMSG_DATA(cm)); 1626 clen = 0; 1627 m_freem(control); 1628 control = NULL; 1629 atomic = 1; 1630 } 1631 } 1632 1633 if (resid == 0 && !ktls_permit_empty_frames(tls)) { 1634 error = EINVAL; 1635 goto release; 1636 } 1637 } 1638 #endif 1639 1640 restart: 1641 do { 1642 SOCKBUF_LOCK(&so->so_snd); 1643 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1644 SOCKBUF_UNLOCK(&so->so_snd); 1645 error = EPIPE; 1646 goto release; 1647 } 1648 if (so->so_error) { 1649 error = so->so_error; 1650 so->so_error = 0; 1651 SOCKBUF_UNLOCK(&so->so_snd); 1652 goto release; 1653 } 1654 if ((so->so_state & SS_ISCONNECTED) == 0) { 1655 /* 1656 * `sendto' and `sendmsg' is allowed on a connection- 1657 * based socket if it supports implied connect. 1658 * Return ENOTCONN if not connected and no address is 1659 * supplied. 1660 */ 1661 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1662 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1663 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1664 !(resid == 0 && clen != 0)) { 1665 SOCKBUF_UNLOCK(&so->so_snd); 1666 error = ENOTCONN; 1667 goto release; 1668 } 1669 } else if (addr == NULL) { 1670 SOCKBUF_UNLOCK(&so->so_snd); 1671 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1672 error = ENOTCONN; 1673 else 1674 error = EDESTADDRREQ; 1675 goto release; 1676 } 1677 } 1678 space = sbspace(&so->so_snd); 1679 if (flags & MSG_OOB) 1680 space += 1024; 1681 if ((atomic && resid > so->so_snd.sb_hiwat) || 1682 clen > so->so_snd.sb_hiwat) { 1683 SOCKBUF_UNLOCK(&so->so_snd); 1684 error = EMSGSIZE; 1685 goto release; 1686 } 1687 if (space < resid + clen && 1688 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1689 if ((so->so_state & SS_NBIO) || 1690 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) { 1691 SOCKBUF_UNLOCK(&so->so_snd); 1692 error = EWOULDBLOCK; 1693 goto release; 1694 } 1695 error = sbwait(so, SO_SND); 1696 SOCKBUF_UNLOCK(&so->so_snd); 1697 if (error) 1698 goto release; 1699 goto restart; 1700 } 1701 SOCKBUF_UNLOCK(&so->so_snd); 1702 space -= clen; 1703 do { 1704 if (uio == NULL) { 1705 resid = 0; 1706 if (flags & MSG_EOR) 1707 top->m_flags |= M_EOR; 1708 #ifdef KERN_TLS 1709 if (tls != NULL) { 1710 ktls_frame(top, tls, &tls_enq_cnt, 1711 tls_rtype); 1712 tls_rtype = TLS_RLTYPE_APP; 1713 } 1714 #endif 1715 } else { 1716 /* 1717 * Copy the data from userland into a mbuf 1718 * chain. If resid is 0, which can happen 1719 * only if we have control to send, then 1720 * a single empty mbuf is returned. This 1721 * is a workaround to prevent protocol send 1722 * methods to panic. 1723 */ 1724 #ifdef KERN_TLS 1725 if (tls != NULL) { 1726 top = m_uiotombuf(uio, M_WAITOK, space, 1727 tls->params.max_frame_len, 1728 M_EXTPG | 1729 ((flags & MSG_EOR) ? M_EOR : 0)); 1730 if (top != NULL) { 1731 ktls_frame(top, tls, 1732 &tls_enq_cnt, tls_rtype); 1733 } 1734 tls_rtype = TLS_RLTYPE_APP; 1735 } else 1736 #endif 1737 top = m_uiotombuf(uio, M_WAITOK, space, 1738 (atomic ? max_hdr : 0), 1739 (atomic ? M_PKTHDR : 0) | 1740 ((flags & MSG_EOR) ? M_EOR : 0)); 1741 if (top == NULL) { 1742 error = EFAULT; /* only possible error */ 1743 goto release; 1744 } 1745 space -= resid - uio->uio_resid; 1746 resid = uio->uio_resid; 1747 } 1748 if (dontroute) { 1749 SOCK_LOCK(so); 1750 so->so_options |= SO_DONTROUTE; 1751 SOCK_UNLOCK(so); 1752 } 1753 /* 1754 * XXX all the SBS_CANTSENDMORE checks previously 1755 * done could be out of date. We could have received 1756 * a reset packet in an interrupt or maybe we slept 1757 * while doing page faults in uiomove() etc. We 1758 * could probably recheck again inside the locking 1759 * protection here, but there are probably other 1760 * places that this also happens. We must rethink 1761 * this. 1762 */ 1763 VNET_SO_ASSERT(so); 1764 1765 pr_send_flag = (flags & MSG_OOB) ? PRUS_OOB : 1766 /* 1767 * If the user set MSG_EOF, the protocol understands 1768 * this flag and nothing left to send then use 1769 * PRU_SEND_EOF instead of PRU_SEND. 1770 */ 1771 ((flags & MSG_EOF) && 1772 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1773 (resid <= 0)) ? 1774 PRUS_EOF : 1775 /* If there is more to send set PRUS_MORETOCOME. */ 1776 (flags & MSG_MORETOCOME) || 1777 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; 1778 1779 #ifdef KERN_TLS 1780 pr_send_flag |= tls_send_flag; 1781 #endif 1782 1783 error = so->so_proto->pr_send(so, pr_send_flag, top, 1784 addr, control, td); 1785 1786 if (dontroute) { 1787 SOCK_LOCK(so); 1788 so->so_options &= ~SO_DONTROUTE; 1789 SOCK_UNLOCK(so); 1790 } 1791 1792 #ifdef KERN_TLS 1793 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) { 1794 if (error != 0) { 1795 m_freem(top); 1796 top = NULL; 1797 } else { 1798 soref(so); 1799 ktls_enqueue(top, so, tls_enq_cnt); 1800 } 1801 } 1802 #endif 1803 clen = 0; 1804 control = NULL; 1805 top = NULL; 1806 if (error) 1807 goto release; 1808 } while (resid && space > 0); 1809 } while (resid); 1810 1811 release: 1812 SOCK_IO_SEND_UNLOCK(so); 1813 out: 1814 #ifdef KERN_TLS 1815 if (tls != NULL) 1816 ktls_free(tls); 1817 #endif 1818 if (top != NULL) 1819 m_freem(top); 1820 if (control != NULL) 1821 m_freem(control); 1822 return (error); 1823 } 1824 1825 /* 1826 * Send to a socket from a kernel thread. 1827 * 1828 * XXXGL: in almost all cases uio is NULL and the mbuf is supplied. 1829 * Exception is nfs/bootp_subr.c. It is arguable that the VNET context needs 1830 * to be set at all. This function should just boil down to a static inline 1831 * calling the protocol method. 1832 */ 1833 int 1834 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1835 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1836 { 1837 int error; 1838 1839 CURVNET_SET(so->so_vnet); 1840 error = so->so_proto->pr_sosend(so, addr, uio, 1841 top, control, flags, td); 1842 CURVNET_RESTORE(); 1843 return (error); 1844 } 1845 1846 /* 1847 * send(2), write(2) or aio_write(2) on a socket. 1848 */ 1849 int 1850 sousrsend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1851 struct mbuf *control, int flags, struct proc *userproc) 1852 { 1853 struct thread *td; 1854 ssize_t len; 1855 int error; 1856 1857 td = uio->uio_td; 1858 len = uio->uio_resid; 1859 CURVNET_SET(so->so_vnet); 1860 error = so->so_proto->pr_sosend(so, addr, uio, NULL, control, flags, 1861 td); 1862 CURVNET_RESTORE(); 1863 if (error != 0) { 1864 /* 1865 * Clear transient errors for stream protocols if they made 1866 * some progress. Make exclusion for aio(4) that would 1867 * schedule a new write in case of EWOULDBLOCK and clear 1868 * error itself. See soaio_process_job(). 1869 */ 1870 if (uio->uio_resid != len && 1871 (so->so_proto->pr_flags & PR_ATOMIC) == 0 && 1872 userproc == NULL && 1873 (error == ERESTART || error == EINTR || 1874 error == EWOULDBLOCK)) 1875 error = 0; 1876 /* Generation of SIGPIPE can be controlled per socket. */ 1877 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0 && 1878 (flags & MSG_NOSIGNAL) == 0) { 1879 if (userproc != NULL) { 1880 /* aio(4) job */ 1881 PROC_LOCK(userproc); 1882 kern_psignal(userproc, SIGPIPE); 1883 PROC_UNLOCK(userproc); 1884 } else { 1885 PROC_LOCK(td->td_proc); 1886 tdsignal(td, SIGPIPE); 1887 PROC_UNLOCK(td->td_proc); 1888 } 1889 } 1890 } 1891 return (error); 1892 } 1893 1894 /* 1895 * The part of soreceive() that implements reading non-inline out-of-band 1896 * data from a socket. For more complete comments, see soreceive(), from 1897 * which this code originated. 1898 * 1899 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1900 * unable to return an mbuf chain to the caller. 1901 */ 1902 static int 1903 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1904 { 1905 struct protosw *pr = so->so_proto; 1906 struct mbuf *m; 1907 int error; 1908 1909 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1910 VNET_SO_ASSERT(so); 1911 1912 m = m_get(M_WAITOK, MT_DATA); 1913 error = pr->pr_rcvoob(so, m, flags & MSG_PEEK); 1914 if (error) 1915 goto bad; 1916 do { 1917 error = uiomove(mtod(m, void *), 1918 (int) min(uio->uio_resid, m->m_len), uio); 1919 m = m_free(m); 1920 } while (uio->uio_resid && error == 0 && m); 1921 bad: 1922 if (m != NULL) 1923 m_freem(m); 1924 return (error); 1925 } 1926 1927 /* 1928 * Following replacement or removal of the first mbuf on the first mbuf chain 1929 * of a socket buffer, push necessary state changes back into the socket 1930 * buffer so that other consumers see the values consistently. 'nextrecord' 1931 * is the callers locally stored value of the original value of 1932 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1933 * NOTE: 'nextrecord' may be NULL. 1934 */ 1935 static __inline void 1936 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1937 { 1938 1939 SOCKBUF_LOCK_ASSERT(sb); 1940 /* 1941 * First, update for the new value of nextrecord. If necessary, make 1942 * it the first record. 1943 */ 1944 if (sb->sb_mb != NULL) 1945 sb->sb_mb->m_nextpkt = nextrecord; 1946 else 1947 sb->sb_mb = nextrecord; 1948 1949 /* 1950 * Now update any dependent socket buffer fields to reflect the new 1951 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1952 * addition of a second clause that takes care of the case where 1953 * sb_mb has been updated, but remains the last record. 1954 */ 1955 if (sb->sb_mb == NULL) { 1956 sb->sb_mbtail = NULL; 1957 sb->sb_lastrecord = NULL; 1958 } else if (sb->sb_mb->m_nextpkt == NULL) 1959 sb->sb_lastrecord = sb->sb_mb; 1960 } 1961 1962 /* 1963 * Implement receive operations on a socket. We depend on the way that 1964 * records are added to the sockbuf by sbappend. In particular, each record 1965 * (mbufs linked through m_next) must begin with an address if the protocol 1966 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1967 * data, and then zero or more mbufs of data. In order to allow parallelism 1968 * between network receive and copying to user space, as well as avoid 1969 * sleeping with a mutex held, we release the socket buffer mutex during the 1970 * user space copy. Although the sockbuf is locked, new data may still be 1971 * appended, and thus we must maintain consistency of the sockbuf during that 1972 * time. 1973 * 1974 * The caller may receive the data as a single mbuf chain by supplying an 1975 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1976 * the count in uio_resid. 1977 */ 1978 int 1979 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1980 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1981 { 1982 struct mbuf *m, **mp; 1983 int flags, error, offset; 1984 ssize_t len; 1985 struct protosw *pr = so->so_proto; 1986 struct mbuf *nextrecord; 1987 int moff, type = 0; 1988 ssize_t orig_resid = uio->uio_resid; 1989 bool report_real_len = false; 1990 1991 mp = mp0; 1992 if (psa != NULL) 1993 *psa = NULL; 1994 if (controlp != NULL) 1995 *controlp = NULL; 1996 if (flagsp != NULL) { 1997 report_real_len = *flagsp & MSG_TRUNC; 1998 *flagsp &= ~MSG_TRUNC; 1999 flags = *flagsp &~ MSG_EOR; 2000 } else 2001 flags = 0; 2002 if (flags & MSG_OOB) 2003 return (soreceive_rcvoob(so, uio, flags)); 2004 if (mp != NULL) 2005 *mp = NULL; 2006 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 2007 && uio->uio_resid) { 2008 VNET_SO_ASSERT(so); 2009 pr->pr_rcvd(so, 0); 2010 } 2011 2012 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 2013 if (error) 2014 return (error); 2015 2016 restart: 2017 SOCKBUF_LOCK(&so->so_rcv); 2018 m = so->so_rcv.sb_mb; 2019 /* 2020 * If we have less data than requested, block awaiting more (subject 2021 * to any timeout) if: 2022 * 1. the current count is less than the low water mark, or 2023 * 2. MSG_DONTWAIT is not set 2024 */ 2025 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 2026 sbavail(&so->so_rcv) < uio->uio_resid) && 2027 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && 2028 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 2029 KASSERT(m != NULL || !sbavail(&so->so_rcv), 2030 ("receive: m == %p sbavail == %u", 2031 m, sbavail(&so->so_rcv))); 2032 if (so->so_error || so->so_rerror) { 2033 if (m != NULL) 2034 goto dontblock; 2035 if (so->so_error) 2036 error = so->so_error; 2037 else 2038 error = so->so_rerror; 2039 if ((flags & MSG_PEEK) == 0) { 2040 if (so->so_error) 2041 so->so_error = 0; 2042 else 2043 so->so_rerror = 0; 2044 } 2045 SOCKBUF_UNLOCK(&so->so_rcv); 2046 goto release; 2047 } 2048 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2049 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2050 if (m != NULL) 2051 goto dontblock; 2052 #ifdef KERN_TLS 2053 else if (so->so_rcv.sb_tlsdcc == 0 && 2054 so->so_rcv.sb_tlscc == 0) { 2055 #else 2056 else { 2057 #endif 2058 SOCKBUF_UNLOCK(&so->so_rcv); 2059 goto release; 2060 } 2061 } 2062 for (; m != NULL; m = m->m_next) 2063 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 2064 m = so->so_rcv.sb_mb; 2065 goto dontblock; 2066 } 2067 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED | 2068 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 && 2069 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 2070 SOCKBUF_UNLOCK(&so->so_rcv); 2071 error = ENOTCONN; 2072 goto release; 2073 } 2074 if (uio->uio_resid == 0 && !report_real_len) { 2075 SOCKBUF_UNLOCK(&so->so_rcv); 2076 goto release; 2077 } 2078 if ((so->so_state & SS_NBIO) || 2079 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2080 SOCKBUF_UNLOCK(&so->so_rcv); 2081 error = EWOULDBLOCK; 2082 goto release; 2083 } 2084 SBLASTRECORDCHK(&so->so_rcv); 2085 SBLASTMBUFCHK(&so->so_rcv); 2086 error = sbwait(so, SO_RCV); 2087 SOCKBUF_UNLOCK(&so->so_rcv); 2088 if (error) 2089 goto release; 2090 goto restart; 2091 } 2092 dontblock: 2093 /* 2094 * From this point onward, we maintain 'nextrecord' as a cache of the 2095 * pointer to the next record in the socket buffer. We must keep the 2096 * various socket buffer pointers and local stack versions of the 2097 * pointers in sync, pushing out modifications before dropping the 2098 * socket buffer mutex, and re-reading them when picking it up. 2099 * 2100 * Otherwise, we will race with the network stack appending new data 2101 * or records onto the socket buffer by using inconsistent/stale 2102 * versions of the field, possibly resulting in socket buffer 2103 * corruption. 2104 * 2105 * By holding the high-level sblock(), we prevent simultaneous 2106 * readers from pulling off the front of the socket buffer. 2107 */ 2108 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2109 if (uio->uio_td) 2110 uio->uio_td->td_ru.ru_msgrcv++; 2111 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 2112 SBLASTRECORDCHK(&so->so_rcv); 2113 SBLASTMBUFCHK(&so->so_rcv); 2114 nextrecord = m->m_nextpkt; 2115 if (pr->pr_flags & PR_ADDR) { 2116 KASSERT(m->m_type == MT_SONAME, 2117 ("m->m_type == %d", m->m_type)); 2118 orig_resid = 0; 2119 if (psa != NULL) 2120 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2121 M_NOWAIT); 2122 if (flags & MSG_PEEK) { 2123 m = m->m_next; 2124 } else { 2125 sbfree(&so->so_rcv, m); 2126 so->so_rcv.sb_mb = m_free(m); 2127 m = so->so_rcv.sb_mb; 2128 sockbuf_pushsync(&so->so_rcv, nextrecord); 2129 } 2130 } 2131 2132 /* 2133 * Process one or more MT_CONTROL mbufs present before any data mbufs 2134 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 2135 * just copy the data; if !MSG_PEEK, we call into the protocol to 2136 * perform externalization (or freeing if controlp == NULL). 2137 */ 2138 if (m != NULL && m->m_type == MT_CONTROL) { 2139 struct mbuf *cm = NULL, *cmn; 2140 struct mbuf **cme = &cm; 2141 #ifdef KERN_TLS 2142 struct cmsghdr *cmsg; 2143 struct tls_get_record tgr; 2144 2145 /* 2146 * For MSG_TLSAPPDATA, check for an alert record. 2147 * If found, return ENXIO without removing 2148 * it from the receive queue. This allows a subsequent 2149 * call without MSG_TLSAPPDATA to receive it. 2150 * Note that, for TLS, there should only be a single 2151 * control mbuf with the TLS_GET_RECORD message in it. 2152 */ 2153 if (flags & MSG_TLSAPPDATA) { 2154 cmsg = mtod(m, struct cmsghdr *); 2155 if (cmsg->cmsg_type == TLS_GET_RECORD && 2156 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) { 2157 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr)); 2158 if (__predict_false(tgr.tls_type == 2159 TLS_RLTYPE_ALERT)) { 2160 SOCKBUF_UNLOCK(&so->so_rcv); 2161 error = ENXIO; 2162 goto release; 2163 } 2164 } 2165 } 2166 #endif 2167 2168 do { 2169 if (flags & MSG_PEEK) { 2170 if (controlp != NULL) { 2171 *controlp = m_copym(m, 0, m->m_len, 2172 M_NOWAIT); 2173 controlp = &(*controlp)->m_next; 2174 } 2175 m = m->m_next; 2176 } else { 2177 sbfree(&so->so_rcv, m); 2178 so->so_rcv.sb_mb = m->m_next; 2179 m->m_next = NULL; 2180 *cme = m; 2181 cme = &(*cme)->m_next; 2182 m = so->so_rcv.sb_mb; 2183 } 2184 } while (m != NULL && m->m_type == MT_CONTROL); 2185 if ((flags & MSG_PEEK) == 0) 2186 sockbuf_pushsync(&so->so_rcv, nextrecord); 2187 while (cm != NULL) { 2188 cmn = cm->m_next; 2189 cm->m_next = NULL; 2190 if (pr->pr_domain->dom_externalize != NULL) { 2191 SOCKBUF_UNLOCK(&so->so_rcv); 2192 VNET_SO_ASSERT(so); 2193 error = (*pr->pr_domain->dom_externalize) 2194 (cm, controlp, flags); 2195 SOCKBUF_LOCK(&so->so_rcv); 2196 } else if (controlp != NULL) 2197 *controlp = cm; 2198 else 2199 m_freem(cm); 2200 if (controlp != NULL) { 2201 while (*controlp != NULL) 2202 controlp = &(*controlp)->m_next; 2203 } 2204 cm = cmn; 2205 } 2206 if (m != NULL) 2207 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 2208 else 2209 nextrecord = so->so_rcv.sb_mb; 2210 orig_resid = 0; 2211 } 2212 if (m != NULL) { 2213 if ((flags & MSG_PEEK) == 0) { 2214 KASSERT(m->m_nextpkt == nextrecord, 2215 ("soreceive: post-control, nextrecord !sync")); 2216 if (nextrecord == NULL) { 2217 KASSERT(so->so_rcv.sb_mb == m, 2218 ("soreceive: post-control, sb_mb!=m")); 2219 KASSERT(so->so_rcv.sb_lastrecord == m, 2220 ("soreceive: post-control, lastrecord!=m")); 2221 } 2222 } 2223 type = m->m_type; 2224 if (type == MT_OOBDATA) 2225 flags |= MSG_OOB; 2226 } else { 2227 if ((flags & MSG_PEEK) == 0) { 2228 KASSERT(so->so_rcv.sb_mb == nextrecord, 2229 ("soreceive: sb_mb != nextrecord")); 2230 if (so->so_rcv.sb_mb == NULL) { 2231 KASSERT(so->so_rcv.sb_lastrecord == NULL, 2232 ("soreceive: sb_lastercord != NULL")); 2233 } 2234 } 2235 } 2236 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2237 SBLASTRECORDCHK(&so->so_rcv); 2238 SBLASTMBUFCHK(&so->so_rcv); 2239 2240 /* 2241 * Now continue to read any data mbufs off of the head of the socket 2242 * buffer until the read request is satisfied. Note that 'type' is 2243 * used to store the type of any mbuf reads that have happened so far 2244 * such that soreceive() can stop reading if the type changes, which 2245 * causes soreceive() to return only one of regular data and inline 2246 * out-of-band data in a single socket receive operation. 2247 */ 2248 moff = 0; 2249 offset = 0; 2250 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 2251 && error == 0) { 2252 /* 2253 * If the type of mbuf has changed since the last mbuf 2254 * examined ('type'), end the receive operation. 2255 */ 2256 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2257 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 2258 if (type != m->m_type) 2259 break; 2260 } else if (type == MT_OOBDATA) 2261 break; 2262 else 2263 KASSERT(m->m_type == MT_DATA, 2264 ("m->m_type == %d", m->m_type)); 2265 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 2266 len = uio->uio_resid; 2267 if (so->so_oobmark && len > so->so_oobmark - offset) 2268 len = so->so_oobmark - offset; 2269 if (len > m->m_len - moff) 2270 len = m->m_len - moff; 2271 /* 2272 * If mp is set, just pass back the mbufs. Otherwise copy 2273 * them out via the uio, then free. Sockbuf must be 2274 * consistent here (points to current mbuf, it points to next 2275 * record) when we drop priority; we must note any additions 2276 * to the sockbuf when we block interrupts again. 2277 */ 2278 if (mp == NULL) { 2279 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2280 SBLASTRECORDCHK(&so->so_rcv); 2281 SBLASTMBUFCHK(&so->so_rcv); 2282 SOCKBUF_UNLOCK(&so->so_rcv); 2283 if ((m->m_flags & M_EXTPG) != 0) 2284 error = m_unmapped_uiomove(m, moff, uio, 2285 (int)len); 2286 else 2287 error = uiomove(mtod(m, char *) + moff, 2288 (int)len, uio); 2289 SOCKBUF_LOCK(&so->so_rcv); 2290 if (error) { 2291 /* 2292 * The MT_SONAME mbuf has already been removed 2293 * from the record, so it is necessary to 2294 * remove the data mbufs, if any, to preserve 2295 * the invariant in the case of PR_ADDR that 2296 * requires MT_SONAME mbufs at the head of 2297 * each record. 2298 */ 2299 if (pr->pr_flags & PR_ATOMIC && 2300 ((flags & MSG_PEEK) == 0)) 2301 (void)sbdroprecord_locked(&so->so_rcv); 2302 SOCKBUF_UNLOCK(&so->so_rcv); 2303 goto release; 2304 } 2305 } else 2306 uio->uio_resid -= len; 2307 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2308 if (len == m->m_len - moff) { 2309 if (m->m_flags & M_EOR) 2310 flags |= MSG_EOR; 2311 if (flags & MSG_PEEK) { 2312 m = m->m_next; 2313 moff = 0; 2314 } else { 2315 nextrecord = m->m_nextpkt; 2316 sbfree(&so->so_rcv, m); 2317 if (mp != NULL) { 2318 m->m_nextpkt = NULL; 2319 *mp = m; 2320 mp = &m->m_next; 2321 so->so_rcv.sb_mb = m = m->m_next; 2322 *mp = NULL; 2323 } else { 2324 so->so_rcv.sb_mb = m_free(m); 2325 m = so->so_rcv.sb_mb; 2326 } 2327 sockbuf_pushsync(&so->so_rcv, nextrecord); 2328 SBLASTRECORDCHK(&so->so_rcv); 2329 SBLASTMBUFCHK(&so->so_rcv); 2330 } 2331 } else { 2332 if (flags & MSG_PEEK) 2333 moff += len; 2334 else { 2335 if (mp != NULL) { 2336 if (flags & MSG_DONTWAIT) { 2337 *mp = m_copym(m, 0, len, 2338 M_NOWAIT); 2339 if (*mp == NULL) { 2340 /* 2341 * m_copym() couldn't 2342 * allocate an mbuf. 2343 * Adjust uio_resid back 2344 * (it was adjusted 2345 * down by len bytes, 2346 * which we didn't end 2347 * up "copying" over). 2348 */ 2349 uio->uio_resid += len; 2350 break; 2351 } 2352 } else { 2353 SOCKBUF_UNLOCK(&so->so_rcv); 2354 *mp = m_copym(m, 0, len, 2355 M_WAITOK); 2356 SOCKBUF_LOCK(&so->so_rcv); 2357 } 2358 } 2359 sbcut_locked(&so->so_rcv, len); 2360 } 2361 } 2362 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2363 if (so->so_oobmark) { 2364 if ((flags & MSG_PEEK) == 0) { 2365 so->so_oobmark -= len; 2366 if (so->so_oobmark == 0) { 2367 so->so_rcv.sb_state |= SBS_RCVATMARK; 2368 break; 2369 } 2370 } else { 2371 offset += len; 2372 if (offset == so->so_oobmark) 2373 break; 2374 } 2375 } 2376 if (flags & MSG_EOR) 2377 break; 2378 /* 2379 * If the MSG_WAITALL flag is set (for non-atomic socket), we 2380 * must not quit until "uio->uio_resid == 0" or an error 2381 * termination. If a signal/timeout occurs, return with a 2382 * short count but without error. Keep sockbuf locked 2383 * against other readers. 2384 */ 2385 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 2386 !sosendallatonce(so) && nextrecord == NULL) { 2387 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2388 if (so->so_error || so->so_rerror || 2389 so->so_rcv.sb_state & SBS_CANTRCVMORE) 2390 break; 2391 /* 2392 * Notify the protocol that some data has been 2393 * drained before blocking. 2394 */ 2395 if (pr->pr_flags & PR_WANTRCVD) { 2396 SOCKBUF_UNLOCK(&so->so_rcv); 2397 VNET_SO_ASSERT(so); 2398 pr->pr_rcvd(so, flags); 2399 SOCKBUF_LOCK(&so->so_rcv); 2400 } 2401 SBLASTRECORDCHK(&so->so_rcv); 2402 SBLASTMBUFCHK(&so->so_rcv); 2403 /* 2404 * We could receive some data while was notifying 2405 * the protocol. Skip blocking in this case. 2406 */ 2407 if (so->so_rcv.sb_mb == NULL) { 2408 error = sbwait(so, SO_RCV); 2409 if (error) { 2410 SOCKBUF_UNLOCK(&so->so_rcv); 2411 goto release; 2412 } 2413 } 2414 m = so->so_rcv.sb_mb; 2415 if (m != NULL) 2416 nextrecord = m->m_nextpkt; 2417 } 2418 } 2419 2420 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2421 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 2422 if (report_real_len) 2423 uio->uio_resid -= m_length(m, NULL) - moff; 2424 flags |= MSG_TRUNC; 2425 if ((flags & MSG_PEEK) == 0) 2426 (void) sbdroprecord_locked(&so->so_rcv); 2427 } 2428 if ((flags & MSG_PEEK) == 0) { 2429 if (m == NULL) { 2430 /* 2431 * First part is an inline SB_EMPTY_FIXUP(). Second 2432 * part makes sure sb_lastrecord is up-to-date if 2433 * there is still data in the socket buffer. 2434 */ 2435 so->so_rcv.sb_mb = nextrecord; 2436 if (so->so_rcv.sb_mb == NULL) { 2437 so->so_rcv.sb_mbtail = NULL; 2438 so->so_rcv.sb_lastrecord = NULL; 2439 } else if (nextrecord->m_nextpkt == NULL) 2440 so->so_rcv.sb_lastrecord = nextrecord; 2441 } 2442 SBLASTRECORDCHK(&so->so_rcv); 2443 SBLASTMBUFCHK(&so->so_rcv); 2444 /* 2445 * If soreceive() is being done from the socket callback, 2446 * then don't need to generate ACK to peer to update window, 2447 * since ACK will be generated on return to TCP. 2448 */ 2449 if (!(flags & MSG_SOCALLBCK) && 2450 (pr->pr_flags & PR_WANTRCVD)) { 2451 SOCKBUF_UNLOCK(&so->so_rcv); 2452 VNET_SO_ASSERT(so); 2453 pr->pr_rcvd(so, flags); 2454 SOCKBUF_LOCK(&so->so_rcv); 2455 } 2456 } 2457 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2458 if (orig_resid == uio->uio_resid && orig_resid && 2459 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 2460 SOCKBUF_UNLOCK(&so->so_rcv); 2461 goto restart; 2462 } 2463 SOCKBUF_UNLOCK(&so->so_rcv); 2464 2465 if (flagsp != NULL) 2466 *flagsp |= flags; 2467 release: 2468 SOCK_IO_RECV_UNLOCK(so); 2469 return (error); 2470 } 2471 2472 /* 2473 * Optimized version of soreceive() for stream (TCP) sockets. 2474 */ 2475 int 2476 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 2477 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2478 { 2479 int len = 0, error = 0, flags, oresid; 2480 struct sockbuf *sb; 2481 struct mbuf *m, *n = NULL; 2482 2483 /* We only do stream sockets. */ 2484 if (so->so_type != SOCK_STREAM) 2485 return (EINVAL); 2486 if (psa != NULL) 2487 *psa = NULL; 2488 if (flagsp != NULL) 2489 flags = *flagsp &~ MSG_EOR; 2490 else 2491 flags = 0; 2492 if (controlp != NULL) 2493 *controlp = NULL; 2494 if (flags & MSG_OOB) 2495 return (soreceive_rcvoob(so, uio, flags)); 2496 if (mp0 != NULL) 2497 *mp0 = NULL; 2498 2499 sb = &so->so_rcv; 2500 2501 #ifdef KERN_TLS 2502 /* 2503 * KTLS store TLS records as records with a control message to 2504 * describe the framing. 2505 * 2506 * We check once here before acquiring locks to optimize the 2507 * common case. 2508 */ 2509 if (sb->sb_tls_info != NULL) 2510 return (soreceive_generic(so, psa, uio, mp0, controlp, 2511 flagsp)); 2512 #endif 2513 2514 /* Prevent other readers from entering the socket. */ 2515 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 2516 if (error) 2517 return (error); 2518 SOCKBUF_LOCK(sb); 2519 2520 #ifdef KERN_TLS 2521 if (sb->sb_tls_info != NULL) { 2522 SOCKBUF_UNLOCK(sb); 2523 SOCK_IO_RECV_UNLOCK(so); 2524 return (soreceive_generic(so, psa, uio, mp0, controlp, 2525 flagsp)); 2526 } 2527 #endif 2528 2529 /* Easy one, no space to copyout anything. */ 2530 if (uio->uio_resid == 0) { 2531 error = EINVAL; 2532 goto out; 2533 } 2534 oresid = uio->uio_resid; 2535 2536 /* We will never ever get anything unless we are or were connected. */ 2537 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 2538 error = ENOTCONN; 2539 goto out; 2540 } 2541 2542 restart: 2543 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2544 2545 /* Abort if socket has reported problems. */ 2546 if (so->so_error) { 2547 if (sbavail(sb) > 0) 2548 goto deliver; 2549 if (oresid > uio->uio_resid) 2550 goto out; 2551 error = so->so_error; 2552 if (!(flags & MSG_PEEK)) 2553 so->so_error = 0; 2554 goto out; 2555 } 2556 2557 /* Door is closed. Deliver what is left, if any. */ 2558 if (sb->sb_state & SBS_CANTRCVMORE) { 2559 if (sbavail(sb) > 0) 2560 goto deliver; 2561 else 2562 goto out; 2563 } 2564 2565 /* Socket buffer is empty and we shall not block. */ 2566 if (sbavail(sb) == 0 && 2567 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2568 error = EAGAIN; 2569 goto out; 2570 } 2571 2572 /* Socket buffer got some data that we shall deliver now. */ 2573 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 2574 ((so->so_state & SS_NBIO) || 2575 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2576 sbavail(sb) >= sb->sb_lowat || 2577 sbavail(sb) >= uio->uio_resid || 2578 sbavail(sb) >= sb->sb_hiwat) ) { 2579 goto deliver; 2580 } 2581 2582 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2583 if ((flags & MSG_WAITALL) && 2584 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 2585 goto deliver; 2586 2587 /* 2588 * Wait and block until (more) data comes in. 2589 * NB: Drops the sockbuf lock during wait. 2590 */ 2591 error = sbwait(so, SO_RCV); 2592 if (error) 2593 goto out; 2594 goto restart; 2595 2596 deliver: 2597 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2598 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 2599 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2600 2601 /* Statistics. */ 2602 if (uio->uio_td) 2603 uio->uio_td->td_ru.ru_msgrcv++; 2604 2605 /* Fill uio until full or current end of socket buffer is reached. */ 2606 len = min(uio->uio_resid, sbavail(sb)); 2607 if (mp0 != NULL) { 2608 /* Dequeue as many mbufs as possible. */ 2609 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2610 if (*mp0 == NULL) 2611 *mp0 = sb->sb_mb; 2612 else 2613 m_cat(*mp0, sb->sb_mb); 2614 for (m = sb->sb_mb; 2615 m != NULL && m->m_len <= len; 2616 m = m->m_next) { 2617 KASSERT(!(m->m_flags & M_NOTAVAIL), 2618 ("%s: m %p not available", __func__, m)); 2619 len -= m->m_len; 2620 uio->uio_resid -= m->m_len; 2621 sbfree(sb, m); 2622 n = m; 2623 } 2624 n->m_next = NULL; 2625 sb->sb_mb = m; 2626 sb->sb_lastrecord = sb->sb_mb; 2627 if (sb->sb_mb == NULL) 2628 SB_EMPTY_FIXUP(sb); 2629 } 2630 /* Copy the remainder. */ 2631 if (len > 0) { 2632 KASSERT(sb->sb_mb != NULL, 2633 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2634 2635 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2636 if (m == NULL) 2637 len = 0; /* Don't flush data from sockbuf. */ 2638 else 2639 uio->uio_resid -= len; 2640 if (*mp0 != NULL) 2641 m_cat(*mp0, m); 2642 else 2643 *mp0 = m; 2644 if (*mp0 == NULL) { 2645 error = ENOBUFS; 2646 goto out; 2647 } 2648 } 2649 } else { 2650 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2651 SOCKBUF_UNLOCK(sb); 2652 error = m_mbuftouio(uio, sb->sb_mb, len); 2653 SOCKBUF_LOCK(sb); 2654 if (error) 2655 goto out; 2656 } 2657 SBLASTRECORDCHK(sb); 2658 SBLASTMBUFCHK(sb); 2659 2660 /* 2661 * Remove the delivered data from the socket buffer unless we 2662 * were only peeking. 2663 */ 2664 if (!(flags & MSG_PEEK)) { 2665 if (len > 0) 2666 sbdrop_locked(sb, len); 2667 2668 /* Notify protocol that we drained some data. */ 2669 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2670 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2671 !(flags & MSG_SOCALLBCK))) { 2672 SOCKBUF_UNLOCK(sb); 2673 VNET_SO_ASSERT(so); 2674 so->so_proto->pr_rcvd(so, flags); 2675 SOCKBUF_LOCK(sb); 2676 } 2677 } 2678 2679 /* 2680 * For MSG_WAITALL we may have to loop again and wait for 2681 * more data to come in. 2682 */ 2683 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2684 goto restart; 2685 out: 2686 SBLASTRECORDCHK(sb); 2687 SBLASTMBUFCHK(sb); 2688 SOCKBUF_UNLOCK(sb); 2689 SOCK_IO_RECV_UNLOCK(so); 2690 return (error); 2691 } 2692 2693 /* 2694 * Optimized version of soreceive() for simple datagram cases from userspace. 2695 * Unlike in the stream case, we're able to drop a datagram if copyout() 2696 * fails, and because we handle datagrams atomically, we don't need to use a 2697 * sleep lock to prevent I/O interlacing. 2698 */ 2699 int 2700 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2701 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2702 { 2703 struct mbuf *m, *m2; 2704 int flags, error; 2705 ssize_t len; 2706 struct protosw *pr = so->so_proto; 2707 struct mbuf *nextrecord; 2708 2709 if (psa != NULL) 2710 *psa = NULL; 2711 if (controlp != NULL) 2712 *controlp = NULL; 2713 if (flagsp != NULL) 2714 flags = *flagsp &~ MSG_EOR; 2715 else 2716 flags = 0; 2717 2718 /* 2719 * For any complicated cases, fall back to the full 2720 * soreceive_generic(). 2721 */ 2722 if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC))) 2723 return (soreceive_generic(so, psa, uio, mp0, controlp, 2724 flagsp)); 2725 2726 /* 2727 * Enforce restrictions on use. 2728 */ 2729 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2730 ("soreceive_dgram: wantrcvd")); 2731 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2732 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2733 ("soreceive_dgram: SBS_RCVATMARK")); 2734 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2735 ("soreceive_dgram: P_CONNREQUIRED")); 2736 2737 /* 2738 * Loop blocking while waiting for a datagram. 2739 */ 2740 SOCKBUF_LOCK(&so->so_rcv); 2741 while ((m = so->so_rcv.sb_mb) == NULL) { 2742 KASSERT(sbavail(&so->so_rcv) == 0, 2743 ("soreceive_dgram: sb_mb NULL but sbavail %u", 2744 sbavail(&so->so_rcv))); 2745 if (so->so_error) { 2746 error = so->so_error; 2747 so->so_error = 0; 2748 SOCKBUF_UNLOCK(&so->so_rcv); 2749 return (error); 2750 } 2751 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2752 uio->uio_resid == 0) { 2753 SOCKBUF_UNLOCK(&so->so_rcv); 2754 return (0); 2755 } 2756 if ((so->so_state & SS_NBIO) || 2757 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2758 SOCKBUF_UNLOCK(&so->so_rcv); 2759 return (EWOULDBLOCK); 2760 } 2761 SBLASTRECORDCHK(&so->so_rcv); 2762 SBLASTMBUFCHK(&so->so_rcv); 2763 error = sbwait(so, SO_RCV); 2764 if (error) { 2765 SOCKBUF_UNLOCK(&so->so_rcv); 2766 return (error); 2767 } 2768 } 2769 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2770 2771 if (uio->uio_td) 2772 uio->uio_td->td_ru.ru_msgrcv++; 2773 SBLASTRECORDCHK(&so->so_rcv); 2774 SBLASTMBUFCHK(&so->so_rcv); 2775 nextrecord = m->m_nextpkt; 2776 if (nextrecord == NULL) { 2777 KASSERT(so->so_rcv.sb_lastrecord == m, 2778 ("soreceive_dgram: lastrecord != m")); 2779 } 2780 2781 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2782 ("soreceive_dgram: m_nextpkt != nextrecord")); 2783 2784 /* 2785 * Pull 'm' and its chain off the front of the packet queue. 2786 */ 2787 so->so_rcv.sb_mb = NULL; 2788 sockbuf_pushsync(&so->so_rcv, nextrecord); 2789 2790 /* 2791 * Walk 'm's chain and free that many bytes from the socket buffer. 2792 */ 2793 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2794 sbfree(&so->so_rcv, m2); 2795 2796 /* 2797 * Do a few last checks before we let go of the lock. 2798 */ 2799 SBLASTRECORDCHK(&so->so_rcv); 2800 SBLASTMBUFCHK(&so->so_rcv); 2801 SOCKBUF_UNLOCK(&so->so_rcv); 2802 2803 if (pr->pr_flags & PR_ADDR) { 2804 KASSERT(m->m_type == MT_SONAME, 2805 ("m->m_type == %d", m->m_type)); 2806 if (psa != NULL) 2807 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2808 M_NOWAIT); 2809 m = m_free(m); 2810 } 2811 if (m == NULL) { 2812 /* XXXRW: Can this happen? */ 2813 return (0); 2814 } 2815 2816 /* 2817 * Packet to copyout() is now in 'm' and it is disconnected from the 2818 * queue. 2819 * 2820 * Process one or more MT_CONTROL mbufs present before any data mbufs 2821 * in the first mbuf chain on the socket buffer. We call into the 2822 * protocol to perform externalization (or freeing if controlp == 2823 * NULL). In some cases there can be only MT_CONTROL mbufs without 2824 * MT_DATA mbufs. 2825 */ 2826 if (m->m_type == MT_CONTROL) { 2827 struct mbuf *cm = NULL, *cmn; 2828 struct mbuf **cme = &cm; 2829 2830 do { 2831 m2 = m->m_next; 2832 m->m_next = NULL; 2833 *cme = m; 2834 cme = &(*cme)->m_next; 2835 m = m2; 2836 } while (m != NULL && m->m_type == MT_CONTROL); 2837 while (cm != NULL) { 2838 cmn = cm->m_next; 2839 cm->m_next = NULL; 2840 if (pr->pr_domain->dom_externalize != NULL) { 2841 error = (*pr->pr_domain->dom_externalize) 2842 (cm, controlp, flags); 2843 } else if (controlp != NULL) 2844 *controlp = cm; 2845 else 2846 m_freem(cm); 2847 if (controlp != NULL) { 2848 while (*controlp != NULL) 2849 controlp = &(*controlp)->m_next; 2850 } 2851 cm = cmn; 2852 } 2853 } 2854 KASSERT(m == NULL || m->m_type == MT_DATA, 2855 ("soreceive_dgram: !data")); 2856 while (m != NULL && uio->uio_resid > 0) { 2857 len = uio->uio_resid; 2858 if (len > m->m_len) 2859 len = m->m_len; 2860 error = uiomove(mtod(m, char *), (int)len, uio); 2861 if (error) { 2862 m_freem(m); 2863 return (error); 2864 } 2865 if (len == m->m_len) 2866 m = m_free(m); 2867 else { 2868 m->m_data += len; 2869 m->m_len -= len; 2870 } 2871 } 2872 if (m != NULL) { 2873 flags |= MSG_TRUNC; 2874 m_freem(m); 2875 } 2876 if (flagsp != NULL) 2877 *flagsp |= flags; 2878 return (0); 2879 } 2880 2881 int 2882 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2883 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2884 { 2885 int error; 2886 2887 CURVNET_SET(so->so_vnet); 2888 error = so->so_proto->pr_soreceive(so, psa, uio, mp0, controlp, flagsp); 2889 CURVNET_RESTORE(); 2890 return (error); 2891 } 2892 2893 int 2894 soshutdown(struct socket *so, int how) 2895 { 2896 struct protosw *pr; 2897 int error, soerror_enotconn; 2898 2899 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2900 return (EINVAL); 2901 2902 soerror_enotconn = 0; 2903 SOCK_LOCK(so); 2904 if ((so->so_state & 2905 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 2906 /* 2907 * POSIX mandates us to return ENOTCONN when shutdown(2) is 2908 * invoked on a datagram sockets, however historically we would 2909 * actually tear socket down. This is known to be leveraged by 2910 * some applications to unblock process waiting in recvXXX(2) 2911 * by other process that it shares that socket with. Try to meet 2912 * both backward-compatibility and POSIX requirements by forcing 2913 * ENOTCONN but still asking protocol to perform pru_shutdown(). 2914 */ 2915 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) { 2916 SOCK_UNLOCK(so); 2917 return (ENOTCONN); 2918 } 2919 soerror_enotconn = 1; 2920 } 2921 2922 if (SOLISTENING(so)) { 2923 if (how != SHUT_WR) { 2924 so->so_error = ECONNABORTED; 2925 solisten_wakeup(so); /* unlocks so */ 2926 } else { 2927 SOCK_UNLOCK(so); 2928 } 2929 goto done; 2930 } 2931 SOCK_UNLOCK(so); 2932 2933 CURVNET_SET(so->so_vnet); 2934 pr = so->so_proto; 2935 if (pr->pr_flush != NULL) 2936 pr->pr_flush(so, how); 2937 if (how != SHUT_WR) 2938 sorflush(so); 2939 if (how != SHUT_RD) { 2940 error = pr->pr_shutdown(so); 2941 wakeup(&so->so_timeo); 2942 CURVNET_RESTORE(); 2943 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error); 2944 } 2945 wakeup(&so->so_timeo); 2946 CURVNET_RESTORE(); 2947 2948 done: 2949 return (soerror_enotconn ? ENOTCONN : 0); 2950 } 2951 2952 void 2953 sorflush(struct socket *so) 2954 { 2955 struct protosw *pr; 2956 int error; 2957 2958 VNET_SO_ASSERT(so); 2959 2960 /* 2961 * Dislodge threads currently blocked in receive and wait to acquire 2962 * a lock against other simultaneous readers before clearing the 2963 * socket buffer. Don't let our acquire be interrupted by a signal 2964 * despite any existing socket disposition on interruptable waiting. 2965 */ 2966 socantrcvmore(so); 2967 2968 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR); 2969 if (error != 0) { 2970 KASSERT(SOLISTENING(so), 2971 ("%s: soiolock(%p) failed", __func__, so)); 2972 return; 2973 } 2974 2975 pr = so->so_proto; 2976 if (pr->pr_flags & PR_RIGHTS) { 2977 MPASS(pr->pr_domain->dom_dispose != NULL); 2978 (*pr->pr_domain->dom_dispose)(so); 2979 } else { 2980 sbrelease(so, SO_RCV); 2981 SOCK_IO_RECV_UNLOCK(so); 2982 } 2983 2984 } 2985 2986 /* 2987 * Wrapper for Socket established helper hook. 2988 * Parameters: socket, context of the hook point, hook id. 2989 */ 2990 static int inline 2991 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 2992 { 2993 struct socket_hhook_data hhook_data = { 2994 .so = so, 2995 .hctx = hctx, 2996 .m = NULL, 2997 .status = 0 2998 }; 2999 3000 CURVNET_SET(so->so_vnet); 3001 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 3002 CURVNET_RESTORE(); 3003 3004 /* Ugly but needed, since hhooks return void for now */ 3005 return (hhook_data.status); 3006 } 3007 3008 /* 3009 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 3010 * additional variant to handle the case where the option value needs to be 3011 * some kind of integer, but not a specific size. In addition to their use 3012 * here, these functions are also called by the protocol-level pr_ctloutput() 3013 * routines. 3014 */ 3015 int 3016 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 3017 { 3018 size_t valsize; 3019 3020 /* 3021 * If the user gives us more than we wanted, we ignore it, but if we 3022 * don't get the minimum length the caller wants, we return EINVAL. 3023 * On success, sopt->sopt_valsize is set to however much we actually 3024 * retrieved. 3025 */ 3026 if ((valsize = sopt->sopt_valsize) < minlen) 3027 return EINVAL; 3028 if (valsize > len) 3029 sopt->sopt_valsize = valsize = len; 3030 3031 if (sopt->sopt_td != NULL) 3032 return (copyin(sopt->sopt_val, buf, valsize)); 3033 3034 bcopy(sopt->sopt_val, buf, valsize); 3035 return (0); 3036 } 3037 3038 /* 3039 * Kernel version of setsockopt(2). 3040 * 3041 * XXX: optlen is size_t, not socklen_t 3042 */ 3043 int 3044 so_setsockopt(struct socket *so, int level, int optname, void *optval, 3045 size_t optlen) 3046 { 3047 struct sockopt sopt; 3048 3049 sopt.sopt_level = level; 3050 sopt.sopt_name = optname; 3051 sopt.sopt_dir = SOPT_SET; 3052 sopt.sopt_val = optval; 3053 sopt.sopt_valsize = optlen; 3054 sopt.sopt_td = NULL; 3055 return (sosetopt(so, &sopt)); 3056 } 3057 3058 int 3059 sosetopt(struct socket *so, struct sockopt *sopt) 3060 { 3061 int error, optval; 3062 struct linger l; 3063 struct timeval tv; 3064 sbintime_t val, *valp; 3065 uint32_t val32; 3066 #ifdef MAC 3067 struct mac extmac; 3068 #endif 3069 3070 CURVNET_SET(so->so_vnet); 3071 error = 0; 3072 if (sopt->sopt_level != SOL_SOCKET) { 3073 if (so->so_proto->pr_ctloutput != NULL) 3074 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3075 else 3076 error = ENOPROTOOPT; 3077 } else { 3078 switch (sopt->sopt_name) { 3079 case SO_ACCEPTFILTER: 3080 error = accept_filt_setopt(so, sopt); 3081 if (error) 3082 goto bad; 3083 break; 3084 3085 case SO_LINGER: 3086 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 3087 if (error) 3088 goto bad; 3089 if (l.l_linger < 0 || 3090 l.l_linger > USHRT_MAX || 3091 l.l_linger > (INT_MAX / hz)) { 3092 error = EDOM; 3093 goto bad; 3094 } 3095 SOCK_LOCK(so); 3096 so->so_linger = l.l_linger; 3097 if (l.l_onoff) 3098 so->so_options |= SO_LINGER; 3099 else 3100 so->so_options &= ~SO_LINGER; 3101 SOCK_UNLOCK(so); 3102 break; 3103 3104 case SO_DEBUG: 3105 case SO_KEEPALIVE: 3106 case SO_DONTROUTE: 3107 case SO_USELOOPBACK: 3108 case SO_BROADCAST: 3109 case SO_REUSEADDR: 3110 case SO_REUSEPORT: 3111 case SO_REUSEPORT_LB: 3112 case SO_OOBINLINE: 3113 case SO_TIMESTAMP: 3114 case SO_BINTIME: 3115 case SO_NOSIGPIPE: 3116 case SO_NO_DDP: 3117 case SO_NO_OFFLOAD: 3118 case SO_RERROR: 3119 error = sooptcopyin(sopt, &optval, sizeof optval, 3120 sizeof optval); 3121 if (error) 3122 goto bad; 3123 SOCK_LOCK(so); 3124 if (optval) 3125 so->so_options |= sopt->sopt_name; 3126 else 3127 so->so_options &= ~sopt->sopt_name; 3128 SOCK_UNLOCK(so); 3129 break; 3130 3131 case SO_SETFIB: 3132 error = sooptcopyin(sopt, &optval, sizeof optval, 3133 sizeof optval); 3134 if (error) 3135 goto bad; 3136 3137 if (optval < 0 || optval >= rt_numfibs) { 3138 error = EINVAL; 3139 goto bad; 3140 } 3141 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 3142 (so->so_proto->pr_domain->dom_family == PF_INET6) || 3143 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 3144 so->so_fibnum = optval; 3145 else 3146 so->so_fibnum = 0; 3147 break; 3148 3149 case SO_USER_COOKIE: 3150 error = sooptcopyin(sopt, &val32, sizeof val32, 3151 sizeof val32); 3152 if (error) 3153 goto bad; 3154 so->so_user_cookie = val32; 3155 break; 3156 3157 case SO_SNDBUF: 3158 case SO_RCVBUF: 3159 case SO_SNDLOWAT: 3160 case SO_RCVLOWAT: 3161 error = so->so_proto->pr_setsbopt(so, sopt); 3162 if (error) 3163 goto bad; 3164 break; 3165 3166 case SO_SNDTIMEO: 3167 case SO_RCVTIMEO: 3168 #ifdef COMPAT_FREEBSD32 3169 if (SV_CURPROC_FLAG(SV_ILP32)) { 3170 struct timeval32 tv32; 3171 3172 error = sooptcopyin(sopt, &tv32, sizeof tv32, 3173 sizeof tv32); 3174 CP(tv32, tv, tv_sec); 3175 CP(tv32, tv, tv_usec); 3176 } else 3177 #endif 3178 error = sooptcopyin(sopt, &tv, sizeof tv, 3179 sizeof tv); 3180 if (error) 3181 goto bad; 3182 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 3183 tv.tv_usec >= 1000000) { 3184 error = EDOM; 3185 goto bad; 3186 } 3187 if (tv.tv_sec > INT32_MAX) 3188 val = SBT_MAX; 3189 else 3190 val = tvtosbt(tv); 3191 SOCK_LOCK(so); 3192 valp = sopt->sopt_name == SO_SNDTIMEO ? 3193 (SOLISTENING(so) ? &so->sol_sbsnd_timeo : 3194 &so->so_snd.sb_timeo) : 3195 (SOLISTENING(so) ? &so->sol_sbrcv_timeo : 3196 &so->so_rcv.sb_timeo); 3197 *valp = val; 3198 SOCK_UNLOCK(so); 3199 break; 3200 3201 case SO_LABEL: 3202 #ifdef MAC 3203 error = sooptcopyin(sopt, &extmac, sizeof extmac, 3204 sizeof extmac); 3205 if (error) 3206 goto bad; 3207 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 3208 so, &extmac); 3209 #else 3210 error = EOPNOTSUPP; 3211 #endif 3212 break; 3213 3214 case SO_TS_CLOCK: 3215 error = sooptcopyin(sopt, &optval, sizeof optval, 3216 sizeof optval); 3217 if (error) 3218 goto bad; 3219 if (optval < 0 || optval > SO_TS_CLOCK_MAX) { 3220 error = EINVAL; 3221 goto bad; 3222 } 3223 so->so_ts_clock = optval; 3224 break; 3225 3226 case SO_MAX_PACING_RATE: 3227 error = sooptcopyin(sopt, &val32, sizeof(val32), 3228 sizeof(val32)); 3229 if (error) 3230 goto bad; 3231 so->so_max_pacing_rate = val32; 3232 break; 3233 3234 default: 3235 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3236 error = hhook_run_socket(so, sopt, 3237 HHOOK_SOCKET_OPT); 3238 else 3239 error = ENOPROTOOPT; 3240 break; 3241 } 3242 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 3243 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 3244 } 3245 bad: 3246 CURVNET_RESTORE(); 3247 return (error); 3248 } 3249 3250 /* 3251 * Helper routine for getsockopt. 3252 */ 3253 int 3254 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 3255 { 3256 int error; 3257 size_t valsize; 3258 3259 error = 0; 3260 3261 /* 3262 * Documented get behavior is that we always return a value, possibly 3263 * truncated to fit in the user's buffer. Traditional behavior is 3264 * that we always tell the user precisely how much we copied, rather 3265 * than something useful like the total amount we had available for 3266 * her. Note that this interface is not idempotent; the entire 3267 * answer must be generated ahead of time. 3268 */ 3269 valsize = min(len, sopt->sopt_valsize); 3270 sopt->sopt_valsize = valsize; 3271 if (sopt->sopt_val != NULL) { 3272 if (sopt->sopt_td != NULL) 3273 error = copyout(buf, sopt->sopt_val, valsize); 3274 else 3275 bcopy(buf, sopt->sopt_val, valsize); 3276 } 3277 return (error); 3278 } 3279 3280 int 3281 sogetopt(struct socket *so, struct sockopt *sopt) 3282 { 3283 int error, optval; 3284 struct linger l; 3285 struct timeval tv; 3286 #ifdef MAC 3287 struct mac extmac; 3288 #endif 3289 3290 CURVNET_SET(so->so_vnet); 3291 error = 0; 3292 if (sopt->sopt_level != SOL_SOCKET) { 3293 if (so->so_proto->pr_ctloutput != NULL) 3294 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3295 else 3296 error = ENOPROTOOPT; 3297 CURVNET_RESTORE(); 3298 return (error); 3299 } else { 3300 switch (sopt->sopt_name) { 3301 case SO_ACCEPTFILTER: 3302 error = accept_filt_getopt(so, sopt); 3303 break; 3304 3305 case SO_LINGER: 3306 SOCK_LOCK(so); 3307 l.l_onoff = so->so_options & SO_LINGER; 3308 l.l_linger = so->so_linger; 3309 SOCK_UNLOCK(so); 3310 error = sooptcopyout(sopt, &l, sizeof l); 3311 break; 3312 3313 case SO_USELOOPBACK: 3314 case SO_DONTROUTE: 3315 case SO_DEBUG: 3316 case SO_KEEPALIVE: 3317 case SO_REUSEADDR: 3318 case SO_REUSEPORT: 3319 case SO_REUSEPORT_LB: 3320 case SO_BROADCAST: 3321 case SO_OOBINLINE: 3322 case SO_ACCEPTCONN: 3323 case SO_TIMESTAMP: 3324 case SO_BINTIME: 3325 case SO_NOSIGPIPE: 3326 case SO_NO_DDP: 3327 case SO_NO_OFFLOAD: 3328 case SO_RERROR: 3329 optval = so->so_options & sopt->sopt_name; 3330 integer: 3331 error = sooptcopyout(sopt, &optval, sizeof optval); 3332 break; 3333 3334 case SO_DOMAIN: 3335 optval = so->so_proto->pr_domain->dom_family; 3336 goto integer; 3337 3338 case SO_TYPE: 3339 optval = so->so_type; 3340 goto integer; 3341 3342 case SO_PROTOCOL: 3343 optval = so->so_proto->pr_protocol; 3344 goto integer; 3345 3346 case SO_ERROR: 3347 SOCK_LOCK(so); 3348 if (so->so_error) { 3349 optval = so->so_error; 3350 so->so_error = 0; 3351 } else { 3352 optval = so->so_rerror; 3353 so->so_rerror = 0; 3354 } 3355 SOCK_UNLOCK(so); 3356 goto integer; 3357 3358 case SO_SNDBUF: 3359 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : 3360 so->so_snd.sb_hiwat; 3361 goto integer; 3362 3363 case SO_RCVBUF: 3364 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : 3365 so->so_rcv.sb_hiwat; 3366 goto integer; 3367 3368 case SO_SNDLOWAT: 3369 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : 3370 so->so_snd.sb_lowat; 3371 goto integer; 3372 3373 case SO_RCVLOWAT: 3374 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : 3375 so->so_rcv.sb_lowat; 3376 goto integer; 3377 3378 case SO_SNDTIMEO: 3379 case SO_RCVTIMEO: 3380 SOCK_LOCK(so); 3381 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 3382 (SOLISTENING(so) ? so->sol_sbsnd_timeo : 3383 so->so_snd.sb_timeo) : 3384 (SOLISTENING(so) ? so->sol_sbrcv_timeo : 3385 so->so_rcv.sb_timeo)); 3386 SOCK_UNLOCK(so); 3387 #ifdef COMPAT_FREEBSD32 3388 if (SV_CURPROC_FLAG(SV_ILP32)) { 3389 struct timeval32 tv32; 3390 3391 CP(tv, tv32, tv_sec); 3392 CP(tv, tv32, tv_usec); 3393 error = sooptcopyout(sopt, &tv32, sizeof tv32); 3394 } else 3395 #endif 3396 error = sooptcopyout(sopt, &tv, sizeof tv); 3397 break; 3398 3399 case SO_LABEL: 3400 #ifdef MAC 3401 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3402 sizeof(extmac)); 3403 if (error) 3404 goto bad; 3405 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 3406 so, &extmac); 3407 if (error) 3408 goto bad; 3409 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3410 #else 3411 error = EOPNOTSUPP; 3412 #endif 3413 break; 3414 3415 case SO_PEERLABEL: 3416 #ifdef MAC 3417 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3418 sizeof(extmac)); 3419 if (error) 3420 goto bad; 3421 error = mac_getsockopt_peerlabel( 3422 sopt->sopt_td->td_ucred, so, &extmac); 3423 if (error) 3424 goto bad; 3425 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3426 #else 3427 error = EOPNOTSUPP; 3428 #endif 3429 break; 3430 3431 case SO_LISTENQLIMIT: 3432 optval = SOLISTENING(so) ? so->sol_qlimit : 0; 3433 goto integer; 3434 3435 case SO_LISTENQLEN: 3436 optval = SOLISTENING(so) ? so->sol_qlen : 0; 3437 goto integer; 3438 3439 case SO_LISTENINCQLEN: 3440 optval = SOLISTENING(so) ? so->sol_incqlen : 0; 3441 goto integer; 3442 3443 case SO_TS_CLOCK: 3444 optval = so->so_ts_clock; 3445 goto integer; 3446 3447 case SO_MAX_PACING_RATE: 3448 optval = so->so_max_pacing_rate; 3449 goto integer; 3450 3451 default: 3452 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3453 error = hhook_run_socket(so, sopt, 3454 HHOOK_SOCKET_OPT); 3455 else 3456 error = ENOPROTOOPT; 3457 break; 3458 } 3459 } 3460 #ifdef MAC 3461 bad: 3462 #endif 3463 CURVNET_RESTORE(); 3464 return (error); 3465 } 3466 3467 int 3468 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 3469 { 3470 struct mbuf *m, *m_prev; 3471 int sopt_size = sopt->sopt_valsize; 3472 3473 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3474 if (m == NULL) 3475 return ENOBUFS; 3476 if (sopt_size > MLEN) { 3477 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 3478 if ((m->m_flags & M_EXT) == 0) { 3479 m_free(m); 3480 return ENOBUFS; 3481 } 3482 m->m_len = min(MCLBYTES, sopt_size); 3483 } else { 3484 m->m_len = min(MLEN, sopt_size); 3485 } 3486 sopt_size -= m->m_len; 3487 *mp = m; 3488 m_prev = m; 3489 3490 while (sopt_size) { 3491 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3492 if (m == NULL) { 3493 m_freem(*mp); 3494 return ENOBUFS; 3495 } 3496 if (sopt_size > MLEN) { 3497 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 3498 M_NOWAIT); 3499 if ((m->m_flags & M_EXT) == 0) { 3500 m_freem(m); 3501 m_freem(*mp); 3502 return ENOBUFS; 3503 } 3504 m->m_len = min(MCLBYTES, sopt_size); 3505 } else { 3506 m->m_len = min(MLEN, sopt_size); 3507 } 3508 sopt_size -= m->m_len; 3509 m_prev->m_next = m; 3510 m_prev = m; 3511 } 3512 return (0); 3513 } 3514 3515 int 3516 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 3517 { 3518 struct mbuf *m0 = m; 3519 3520 if (sopt->sopt_val == NULL) 3521 return (0); 3522 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3523 if (sopt->sopt_td != NULL) { 3524 int error; 3525 3526 error = copyin(sopt->sopt_val, mtod(m, char *), 3527 m->m_len); 3528 if (error != 0) { 3529 m_freem(m0); 3530 return(error); 3531 } 3532 } else 3533 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 3534 sopt->sopt_valsize -= m->m_len; 3535 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3536 m = m->m_next; 3537 } 3538 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 3539 panic("ip6_sooptmcopyin"); 3540 return (0); 3541 } 3542 3543 int 3544 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 3545 { 3546 struct mbuf *m0 = m; 3547 size_t valsize = 0; 3548 3549 if (sopt->sopt_val == NULL) 3550 return (0); 3551 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3552 if (sopt->sopt_td != NULL) { 3553 int error; 3554 3555 error = copyout(mtod(m, char *), sopt->sopt_val, 3556 m->m_len); 3557 if (error != 0) { 3558 m_freem(m0); 3559 return(error); 3560 } 3561 } else 3562 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 3563 sopt->sopt_valsize -= m->m_len; 3564 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3565 valsize += m->m_len; 3566 m = m->m_next; 3567 } 3568 if (m != NULL) { 3569 /* enough soopt buffer should be given from user-land */ 3570 m_freem(m0); 3571 return(EINVAL); 3572 } 3573 sopt->sopt_valsize = valsize; 3574 return (0); 3575 } 3576 3577 /* 3578 * sohasoutofband(): protocol notifies socket layer of the arrival of new 3579 * out-of-band data, which will then notify socket consumers. 3580 */ 3581 void 3582 sohasoutofband(struct socket *so) 3583 { 3584 3585 if (so->so_sigio != NULL) 3586 pgsigio(&so->so_sigio, SIGURG, 0); 3587 selwakeuppri(&so->so_rdsel, PSOCK); 3588 } 3589 3590 int 3591 sopoll(struct socket *so, int events, struct ucred *active_cred, 3592 struct thread *td) 3593 { 3594 3595 /* 3596 * We do not need to set or assert curvnet as long as everyone uses 3597 * sopoll_generic(). 3598 */ 3599 return (so->so_proto->pr_sopoll(so, events, active_cred, td)); 3600 } 3601 3602 int 3603 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3604 struct thread *td) 3605 { 3606 int revents; 3607 3608 SOCK_LOCK(so); 3609 if (SOLISTENING(so)) { 3610 if (!(events & (POLLIN | POLLRDNORM))) 3611 revents = 0; 3612 else if (!TAILQ_EMPTY(&so->sol_comp)) 3613 revents = events & (POLLIN | POLLRDNORM); 3614 else if ((events & POLLINIGNEOF) == 0 && so->so_error) 3615 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP; 3616 else { 3617 selrecord(td, &so->so_rdsel); 3618 revents = 0; 3619 } 3620 } else { 3621 revents = 0; 3622 SOCK_SENDBUF_LOCK(so); 3623 SOCK_RECVBUF_LOCK(so); 3624 if (events & (POLLIN | POLLRDNORM)) 3625 if (soreadabledata(so)) 3626 revents |= events & (POLLIN | POLLRDNORM); 3627 if (events & (POLLOUT | POLLWRNORM)) 3628 if (sowriteable(so)) 3629 revents |= events & (POLLOUT | POLLWRNORM); 3630 if (events & (POLLPRI | POLLRDBAND)) 3631 if (so->so_oobmark || 3632 (so->so_rcv.sb_state & SBS_RCVATMARK)) 3633 revents |= events & (POLLPRI | POLLRDBAND); 3634 if ((events & POLLINIGNEOF) == 0) { 3635 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3636 revents |= events & (POLLIN | POLLRDNORM); 3637 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3638 revents |= POLLHUP; 3639 } 3640 } 3641 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3642 revents |= events & POLLRDHUP; 3643 if (revents == 0) { 3644 if (events & 3645 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) { 3646 selrecord(td, &so->so_rdsel); 3647 so->so_rcv.sb_flags |= SB_SEL; 3648 } 3649 if (events & (POLLOUT | POLLWRNORM)) { 3650 selrecord(td, &so->so_wrsel); 3651 so->so_snd.sb_flags |= SB_SEL; 3652 } 3653 } 3654 SOCK_RECVBUF_UNLOCK(so); 3655 SOCK_SENDBUF_UNLOCK(so); 3656 } 3657 SOCK_UNLOCK(so); 3658 return (revents); 3659 } 3660 3661 int 3662 soo_kqfilter(struct file *fp, struct knote *kn) 3663 { 3664 struct socket *so = kn->kn_fp->f_data; 3665 struct sockbuf *sb; 3666 sb_which which; 3667 struct knlist *knl; 3668 3669 switch (kn->kn_filter) { 3670 case EVFILT_READ: 3671 kn->kn_fop = &soread_filtops; 3672 knl = &so->so_rdsel.si_note; 3673 sb = &so->so_rcv; 3674 which = SO_RCV; 3675 break; 3676 case EVFILT_WRITE: 3677 kn->kn_fop = &sowrite_filtops; 3678 knl = &so->so_wrsel.si_note; 3679 sb = &so->so_snd; 3680 which = SO_SND; 3681 break; 3682 case EVFILT_EMPTY: 3683 kn->kn_fop = &soempty_filtops; 3684 knl = &so->so_wrsel.si_note; 3685 sb = &so->so_snd; 3686 which = SO_SND; 3687 break; 3688 default: 3689 return (EINVAL); 3690 } 3691 3692 SOCK_LOCK(so); 3693 if (SOLISTENING(so)) { 3694 knlist_add(knl, kn, 1); 3695 } else { 3696 SOCK_BUF_LOCK(so, which); 3697 knlist_add(knl, kn, 1); 3698 sb->sb_flags |= SB_KNOTE; 3699 SOCK_BUF_UNLOCK(so, which); 3700 } 3701 SOCK_UNLOCK(so); 3702 return (0); 3703 } 3704 3705 static void 3706 filt_sordetach(struct knote *kn) 3707 { 3708 struct socket *so = kn->kn_fp->f_data; 3709 3710 so_rdknl_lock(so); 3711 knlist_remove(&so->so_rdsel.si_note, kn, 1); 3712 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note)) 3713 so->so_rcv.sb_flags &= ~SB_KNOTE; 3714 so_rdknl_unlock(so); 3715 } 3716 3717 /*ARGSUSED*/ 3718 static int 3719 filt_soread(struct knote *kn, long hint) 3720 { 3721 struct socket *so; 3722 3723 so = kn->kn_fp->f_data; 3724 3725 if (SOLISTENING(so)) { 3726 SOCK_LOCK_ASSERT(so); 3727 kn->kn_data = so->sol_qlen; 3728 if (so->so_error) { 3729 kn->kn_flags |= EV_EOF; 3730 kn->kn_fflags = so->so_error; 3731 return (1); 3732 } 3733 return (!TAILQ_EMPTY(&so->sol_comp)); 3734 } 3735 3736 SOCK_RECVBUF_LOCK_ASSERT(so); 3737 3738 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 3739 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3740 kn->kn_flags |= EV_EOF; 3741 kn->kn_fflags = so->so_error; 3742 return (1); 3743 } else if (so->so_error || so->so_rerror) 3744 return (1); 3745 3746 if (kn->kn_sfflags & NOTE_LOWAT) { 3747 if (kn->kn_data >= kn->kn_sdata) 3748 return (1); 3749 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) 3750 return (1); 3751 3752 /* This hook returning non-zero indicates an event, not error */ 3753 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); 3754 } 3755 3756 static void 3757 filt_sowdetach(struct knote *kn) 3758 { 3759 struct socket *so = kn->kn_fp->f_data; 3760 3761 so_wrknl_lock(so); 3762 knlist_remove(&so->so_wrsel.si_note, kn, 1); 3763 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note)) 3764 so->so_snd.sb_flags &= ~SB_KNOTE; 3765 so_wrknl_unlock(so); 3766 } 3767 3768 /*ARGSUSED*/ 3769 static int 3770 filt_sowrite(struct knote *kn, long hint) 3771 { 3772 struct socket *so; 3773 3774 so = kn->kn_fp->f_data; 3775 3776 if (SOLISTENING(so)) 3777 return (0); 3778 3779 SOCK_SENDBUF_LOCK_ASSERT(so); 3780 kn->kn_data = sbspace(&so->so_snd); 3781 3782 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); 3783 3784 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3785 kn->kn_flags |= EV_EOF; 3786 kn->kn_fflags = so->so_error; 3787 return (1); 3788 } else if (so->so_error) /* temporary udp error */ 3789 return (1); 3790 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3791 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3792 return (0); 3793 else if (kn->kn_sfflags & NOTE_LOWAT) 3794 return (kn->kn_data >= kn->kn_sdata); 3795 else 3796 return (kn->kn_data >= so->so_snd.sb_lowat); 3797 } 3798 3799 static int 3800 filt_soempty(struct knote *kn, long hint) 3801 { 3802 struct socket *so; 3803 3804 so = kn->kn_fp->f_data; 3805 3806 if (SOLISTENING(so)) 3807 return (1); 3808 3809 SOCK_SENDBUF_LOCK_ASSERT(so); 3810 kn->kn_data = sbused(&so->so_snd); 3811 3812 if (kn->kn_data == 0) 3813 return (1); 3814 else 3815 return (0); 3816 } 3817 3818 int 3819 socheckuid(struct socket *so, uid_t uid) 3820 { 3821 3822 if (so == NULL) 3823 return (EPERM); 3824 if (so->so_cred->cr_uid != uid) 3825 return (EPERM); 3826 return (0); 3827 } 3828 3829 /* 3830 * These functions are used by protocols to notify the socket layer (and its 3831 * consumers) of state changes in the sockets driven by protocol-side events. 3832 */ 3833 3834 /* 3835 * Procedures to manipulate state flags of socket and do appropriate wakeups. 3836 * 3837 * Normal sequence from the active (originating) side is that 3838 * soisconnecting() is called during processing of connect() call, resulting 3839 * in an eventual call to soisconnected() if/when the connection is 3840 * established. When the connection is torn down soisdisconnecting() is 3841 * called during processing of disconnect() call, and soisdisconnected() is 3842 * called when the connection to the peer is totally severed. The semantics 3843 * of these routines are such that connectionless protocols can call 3844 * soisconnected() and soisdisconnected() only, bypassing the in-progress 3845 * calls when setting up a ``connection'' takes no time. 3846 * 3847 * From the passive side, a socket is created with two queues of sockets: 3848 * so_incomp for connections in progress and so_comp for connections already 3849 * made and awaiting user acceptance. As a protocol is preparing incoming 3850 * connections, it creates a socket structure queued on so_incomp by calling 3851 * sonewconn(). When the connection is established, soisconnected() is 3852 * called, and transfers the socket structure to so_comp, making it available 3853 * to accept(). 3854 * 3855 * If a socket is closed with sockets on either so_incomp or so_comp, these 3856 * sockets are dropped. 3857 * 3858 * If higher-level protocols are implemented in the kernel, the wakeups done 3859 * here will sometimes cause software-interrupt process scheduling. 3860 */ 3861 void 3862 soisconnecting(struct socket *so) 3863 { 3864 3865 SOCK_LOCK(so); 3866 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 3867 so->so_state |= SS_ISCONNECTING; 3868 SOCK_UNLOCK(so); 3869 } 3870 3871 void 3872 soisconnected(struct socket *so) 3873 { 3874 bool last __diagused; 3875 3876 SOCK_LOCK(so); 3877 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 3878 so->so_state |= SS_ISCONNECTED; 3879 3880 if (so->so_qstate == SQ_INCOMP) { 3881 struct socket *head = so->so_listen; 3882 int ret; 3883 3884 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so)); 3885 /* 3886 * Promoting a socket from incomplete queue to complete, we 3887 * need to go through reverse order of locking. We first do 3888 * trylock, and if that doesn't succeed, we go the hard way 3889 * leaving a reference and rechecking consistency after proper 3890 * locking. 3891 */ 3892 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) { 3893 soref(head); 3894 SOCK_UNLOCK(so); 3895 SOLISTEN_LOCK(head); 3896 SOCK_LOCK(so); 3897 if (__predict_false(head != so->so_listen)) { 3898 /* 3899 * The socket went off the listen queue, 3900 * should be lost race to close(2) of sol. 3901 * The socket is about to soabort(). 3902 */ 3903 SOCK_UNLOCK(so); 3904 sorele_locked(head); 3905 return; 3906 } 3907 last = refcount_release(&head->so_count); 3908 KASSERT(!last, ("%s: released last reference for %p", 3909 __func__, head)); 3910 } 3911 again: 3912 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 3913 TAILQ_REMOVE(&head->sol_incomp, so, so_list); 3914 head->sol_incqlen--; 3915 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 3916 head->sol_qlen++; 3917 so->so_qstate = SQ_COMP; 3918 SOCK_UNLOCK(so); 3919 solisten_wakeup(head); /* unlocks */ 3920 } else { 3921 SOCK_RECVBUF_LOCK(so); 3922 soupcall_set(so, SO_RCV, 3923 head->sol_accept_filter->accf_callback, 3924 head->sol_accept_filter_arg); 3925 so->so_options &= ~SO_ACCEPTFILTER; 3926 ret = head->sol_accept_filter->accf_callback(so, 3927 head->sol_accept_filter_arg, M_NOWAIT); 3928 if (ret == SU_ISCONNECTED) { 3929 soupcall_clear(so, SO_RCV); 3930 SOCK_RECVBUF_UNLOCK(so); 3931 goto again; 3932 } 3933 SOCK_RECVBUF_UNLOCK(so); 3934 SOCK_UNLOCK(so); 3935 SOLISTEN_UNLOCK(head); 3936 } 3937 return; 3938 } 3939 SOCK_UNLOCK(so); 3940 wakeup(&so->so_timeo); 3941 sorwakeup(so); 3942 sowwakeup(so); 3943 } 3944 3945 void 3946 soisdisconnecting(struct socket *so) 3947 { 3948 3949 SOCK_LOCK(so); 3950 so->so_state &= ~SS_ISCONNECTING; 3951 so->so_state |= SS_ISDISCONNECTING; 3952 3953 if (!SOLISTENING(so)) { 3954 SOCK_RECVBUF_LOCK(so); 3955 socantrcvmore_locked(so); 3956 SOCK_SENDBUF_LOCK(so); 3957 socantsendmore_locked(so); 3958 } 3959 SOCK_UNLOCK(so); 3960 wakeup(&so->so_timeo); 3961 } 3962 3963 void 3964 soisdisconnected(struct socket *so) 3965 { 3966 3967 SOCK_LOCK(so); 3968 3969 /* 3970 * There is at least one reader of so_state that does not 3971 * acquire socket lock, namely soreceive_generic(). Ensure 3972 * that it never sees all flags that track connection status 3973 * cleared, by ordering the update with a barrier semantic of 3974 * our release thread fence. 3975 */ 3976 so->so_state |= SS_ISDISCONNECTED; 3977 atomic_thread_fence_rel(); 3978 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 3979 3980 if (!SOLISTENING(so)) { 3981 SOCK_UNLOCK(so); 3982 SOCK_RECVBUF_LOCK(so); 3983 socantrcvmore_locked(so); 3984 SOCK_SENDBUF_LOCK(so); 3985 sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); 3986 socantsendmore_locked(so); 3987 } else 3988 SOCK_UNLOCK(so); 3989 wakeup(&so->so_timeo); 3990 } 3991 3992 int 3993 soiolock(struct socket *so, struct sx *sx, int flags) 3994 { 3995 int error; 3996 3997 KASSERT((flags & SBL_VALID) == flags, 3998 ("soiolock: invalid flags %#x", flags)); 3999 4000 if ((flags & SBL_WAIT) != 0) { 4001 if ((flags & SBL_NOINTR) != 0) { 4002 sx_xlock(sx); 4003 } else { 4004 error = sx_xlock_sig(sx); 4005 if (error != 0) 4006 return (error); 4007 } 4008 } else if (!sx_try_xlock(sx)) { 4009 return (EWOULDBLOCK); 4010 } 4011 4012 if (__predict_false(SOLISTENING(so))) { 4013 sx_xunlock(sx); 4014 return (ENOTCONN); 4015 } 4016 return (0); 4017 } 4018 4019 void 4020 soiounlock(struct sx *sx) 4021 { 4022 sx_xunlock(sx); 4023 } 4024 4025 /* 4026 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 4027 */ 4028 struct sockaddr * 4029 sodupsockaddr(const struct sockaddr *sa, int mflags) 4030 { 4031 struct sockaddr *sa2; 4032 4033 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 4034 if (sa2) 4035 bcopy(sa, sa2, sa->sa_len); 4036 return sa2; 4037 } 4038 4039 /* 4040 * Register per-socket destructor. 4041 */ 4042 void 4043 sodtor_set(struct socket *so, so_dtor_t *func) 4044 { 4045 4046 SOCK_LOCK_ASSERT(so); 4047 so->so_dtor = func; 4048 } 4049 4050 /* 4051 * Register per-socket buffer upcalls. 4052 */ 4053 void 4054 soupcall_set(struct socket *so, sb_which which, so_upcall_t func, void *arg) 4055 { 4056 struct sockbuf *sb; 4057 4058 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4059 4060 switch (which) { 4061 case SO_RCV: 4062 sb = &so->so_rcv; 4063 break; 4064 case SO_SND: 4065 sb = &so->so_snd; 4066 break; 4067 } 4068 SOCK_BUF_LOCK_ASSERT(so, which); 4069 sb->sb_upcall = func; 4070 sb->sb_upcallarg = arg; 4071 sb->sb_flags |= SB_UPCALL; 4072 } 4073 4074 void 4075 soupcall_clear(struct socket *so, sb_which which) 4076 { 4077 struct sockbuf *sb; 4078 4079 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4080 4081 switch (which) { 4082 case SO_RCV: 4083 sb = &so->so_rcv; 4084 break; 4085 case SO_SND: 4086 sb = &so->so_snd; 4087 break; 4088 } 4089 SOCK_BUF_LOCK_ASSERT(so, which); 4090 KASSERT(sb->sb_upcall != NULL, 4091 ("%s: so %p no upcall to clear", __func__, so)); 4092 sb->sb_upcall = NULL; 4093 sb->sb_upcallarg = NULL; 4094 sb->sb_flags &= ~SB_UPCALL; 4095 } 4096 4097 void 4098 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg) 4099 { 4100 4101 SOLISTEN_LOCK_ASSERT(so); 4102 so->sol_upcall = func; 4103 so->sol_upcallarg = arg; 4104 } 4105 4106 static void 4107 so_rdknl_lock(void *arg) 4108 { 4109 struct socket *so = arg; 4110 4111 retry: 4112 if (SOLISTENING(so)) { 4113 SOLISTEN_LOCK(so); 4114 } else { 4115 SOCK_RECVBUF_LOCK(so); 4116 if (__predict_false(SOLISTENING(so))) { 4117 SOCK_RECVBUF_UNLOCK(so); 4118 goto retry; 4119 } 4120 } 4121 } 4122 4123 static void 4124 so_rdknl_unlock(void *arg) 4125 { 4126 struct socket *so = arg; 4127 4128 if (SOLISTENING(so)) 4129 SOLISTEN_UNLOCK(so); 4130 else 4131 SOCK_RECVBUF_UNLOCK(so); 4132 } 4133 4134 static void 4135 so_rdknl_assert_lock(void *arg, int what) 4136 { 4137 struct socket *so = arg; 4138 4139 if (what == LA_LOCKED) { 4140 if (SOLISTENING(so)) 4141 SOLISTEN_LOCK_ASSERT(so); 4142 else 4143 SOCK_RECVBUF_LOCK_ASSERT(so); 4144 } else { 4145 if (SOLISTENING(so)) 4146 SOLISTEN_UNLOCK_ASSERT(so); 4147 else 4148 SOCK_RECVBUF_UNLOCK_ASSERT(so); 4149 } 4150 } 4151 4152 static void 4153 so_wrknl_lock(void *arg) 4154 { 4155 struct socket *so = arg; 4156 4157 retry: 4158 if (SOLISTENING(so)) { 4159 SOLISTEN_LOCK(so); 4160 } else { 4161 SOCK_SENDBUF_LOCK(so); 4162 if (__predict_false(SOLISTENING(so))) { 4163 SOCK_SENDBUF_UNLOCK(so); 4164 goto retry; 4165 } 4166 } 4167 } 4168 4169 static void 4170 so_wrknl_unlock(void *arg) 4171 { 4172 struct socket *so = arg; 4173 4174 if (SOLISTENING(so)) 4175 SOLISTEN_UNLOCK(so); 4176 else 4177 SOCK_SENDBUF_UNLOCK(so); 4178 } 4179 4180 static void 4181 so_wrknl_assert_lock(void *arg, int what) 4182 { 4183 struct socket *so = arg; 4184 4185 if (what == LA_LOCKED) { 4186 if (SOLISTENING(so)) 4187 SOLISTEN_LOCK_ASSERT(so); 4188 else 4189 SOCK_SENDBUF_LOCK_ASSERT(so); 4190 } else { 4191 if (SOLISTENING(so)) 4192 SOLISTEN_UNLOCK_ASSERT(so); 4193 else 4194 SOCK_SENDBUF_UNLOCK_ASSERT(so); 4195 } 4196 } 4197 4198 /* 4199 * Create an external-format (``xsocket'') structure using the information in 4200 * the kernel-format socket structure pointed to by so. This is done to 4201 * reduce the spew of irrelevant information over this interface, to isolate 4202 * user code from changes in the kernel structure, and potentially to provide 4203 * information-hiding if we decide that some of this information should be 4204 * hidden from users. 4205 */ 4206 void 4207 sotoxsocket(struct socket *so, struct xsocket *xso) 4208 { 4209 4210 bzero(xso, sizeof(*xso)); 4211 xso->xso_len = sizeof *xso; 4212 xso->xso_so = (uintptr_t)so; 4213 xso->so_type = so->so_type; 4214 xso->so_options = so->so_options; 4215 xso->so_linger = so->so_linger; 4216 xso->so_state = so->so_state; 4217 xso->so_pcb = (uintptr_t)so->so_pcb; 4218 xso->xso_protocol = so->so_proto->pr_protocol; 4219 xso->xso_family = so->so_proto->pr_domain->dom_family; 4220 xso->so_timeo = so->so_timeo; 4221 xso->so_error = so->so_error; 4222 xso->so_uid = so->so_cred->cr_uid; 4223 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 4224 if (SOLISTENING(so)) { 4225 xso->so_qlen = so->sol_qlen; 4226 xso->so_incqlen = so->sol_incqlen; 4227 xso->so_qlimit = so->sol_qlimit; 4228 xso->so_oobmark = 0; 4229 } else { 4230 xso->so_state |= so->so_qstate; 4231 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0; 4232 xso->so_oobmark = so->so_oobmark; 4233 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 4234 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 4235 } 4236 } 4237 4238 struct sockbuf * 4239 so_sockbuf_rcv(struct socket *so) 4240 { 4241 4242 return (&so->so_rcv); 4243 } 4244 4245 struct sockbuf * 4246 so_sockbuf_snd(struct socket *so) 4247 { 4248 4249 return (&so->so_snd); 4250 } 4251 4252 int 4253 so_state_get(const struct socket *so) 4254 { 4255 4256 return (so->so_state); 4257 } 4258 4259 void 4260 so_state_set(struct socket *so, int val) 4261 { 4262 4263 so->so_state = val; 4264 } 4265 4266 int 4267 so_options_get(const struct socket *so) 4268 { 4269 4270 return (so->so_options); 4271 } 4272 4273 void 4274 so_options_set(struct socket *so, int val) 4275 { 4276 4277 so->so_options = val; 4278 } 4279 4280 int 4281 so_error_get(const struct socket *so) 4282 { 4283 4284 return (so->so_error); 4285 } 4286 4287 void 4288 so_error_set(struct socket *so, int val) 4289 { 4290 4291 so->so_error = val; 4292 } 4293 4294 int 4295 so_linger_get(const struct socket *so) 4296 { 4297 4298 return (so->so_linger); 4299 } 4300 4301 void 4302 so_linger_set(struct socket *so, int val) 4303 { 4304 4305 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz), 4306 ("%s: val %d out of range", __func__, val)); 4307 4308 so->so_linger = val; 4309 } 4310 4311 struct protosw * 4312 so_protosw_get(const struct socket *so) 4313 { 4314 4315 return (so->so_proto); 4316 } 4317 4318 void 4319 so_protosw_set(struct socket *so, struct protosw *val) 4320 { 4321 4322 so->so_proto = val; 4323 } 4324 4325 void 4326 so_sorwakeup(struct socket *so) 4327 { 4328 4329 sorwakeup(so); 4330 } 4331 4332 void 4333 so_sowwakeup(struct socket *so) 4334 { 4335 4336 sowwakeup(so); 4337 } 4338 4339 void 4340 so_sorwakeup_locked(struct socket *so) 4341 { 4342 4343 sorwakeup_locked(so); 4344 } 4345 4346 void 4347 so_sowwakeup_locked(struct socket *so) 4348 { 4349 4350 sowwakeup_locked(so); 4351 } 4352 4353 void 4354 so_lock(struct socket *so) 4355 { 4356 4357 SOCK_LOCK(so); 4358 } 4359 4360 void 4361 so_unlock(struct socket *so) 4362 { 4363 4364 SOCK_UNLOCK(so); 4365 } 4366