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 /* 1350 * If protocol is connection-based, can only connect once. 1351 * Otherwise, if connected, try to disconnect first. This allows 1352 * user to disconnect by connecting to, e.g., a null address. 1353 * 1354 * Note, this check is racy and may need to be re-evaluated at the 1355 * protocol layer. 1356 */ 1357 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 1358 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 1359 (error = sodisconnect(so)))) { 1360 error = EISCONN; 1361 } else { 1362 /* 1363 * Prevent accumulated error from previous connection from 1364 * biting us. 1365 */ 1366 so->so_error = 0; 1367 if (fd == AT_FDCWD) { 1368 error = so->so_proto->pr_connect(so, nam, td); 1369 } else { 1370 error = so->so_proto->pr_connectat(fd, so, nam, td); 1371 } 1372 } 1373 CURVNET_RESTORE(); 1374 1375 return (error); 1376 } 1377 1378 int 1379 soconnect2(struct socket *so1, struct socket *so2) 1380 { 1381 int error; 1382 1383 CURVNET_SET(so1->so_vnet); 1384 error = so1->so_proto->pr_connect2(so1, so2); 1385 CURVNET_RESTORE(); 1386 return (error); 1387 } 1388 1389 int 1390 sodisconnect(struct socket *so) 1391 { 1392 int error; 1393 1394 if ((so->so_state & SS_ISCONNECTED) == 0) 1395 return (ENOTCONN); 1396 if (so->so_state & SS_ISDISCONNECTING) 1397 return (EALREADY); 1398 VNET_SO_ASSERT(so); 1399 error = so->so_proto->pr_disconnect(so); 1400 return (error); 1401 } 1402 1403 int 1404 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1405 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1406 { 1407 long space; 1408 ssize_t resid; 1409 int clen = 0, error, dontroute; 1410 1411 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 1412 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 1413 ("sosend_dgram: !PR_ATOMIC")); 1414 1415 if (uio != NULL) 1416 resid = uio->uio_resid; 1417 else 1418 resid = top->m_pkthdr.len; 1419 /* 1420 * In theory resid should be unsigned. However, space must be 1421 * signed, as it might be less than 0 if we over-committed, and we 1422 * must use a signed comparison of space and resid. On the other 1423 * hand, a negative resid causes us to loop sending 0-length 1424 * segments to the protocol. 1425 */ 1426 if (resid < 0) { 1427 error = EINVAL; 1428 goto out; 1429 } 1430 1431 dontroute = 1432 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 1433 if (td != NULL) 1434 td->td_ru.ru_msgsnd++; 1435 if (control != NULL) 1436 clen = control->m_len; 1437 1438 SOCKBUF_LOCK(&so->so_snd); 1439 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1440 SOCKBUF_UNLOCK(&so->so_snd); 1441 error = EPIPE; 1442 goto out; 1443 } 1444 if (so->so_error) { 1445 error = so->so_error; 1446 so->so_error = 0; 1447 SOCKBUF_UNLOCK(&so->so_snd); 1448 goto out; 1449 } 1450 if ((so->so_state & SS_ISCONNECTED) == 0) { 1451 /* 1452 * `sendto' and `sendmsg' is allowed on a connection-based 1453 * socket if it supports implied connect. Return ENOTCONN if 1454 * not connected and no address is supplied. 1455 */ 1456 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1457 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1458 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1459 !(resid == 0 && clen != 0)) { 1460 SOCKBUF_UNLOCK(&so->so_snd); 1461 error = ENOTCONN; 1462 goto out; 1463 } 1464 } else if (addr == NULL) { 1465 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1466 error = ENOTCONN; 1467 else 1468 error = EDESTADDRREQ; 1469 SOCKBUF_UNLOCK(&so->so_snd); 1470 goto out; 1471 } 1472 } 1473 1474 /* 1475 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1476 * problem and need fixing. 1477 */ 1478 space = sbspace(&so->so_snd); 1479 if (flags & MSG_OOB) 1480 space += 1024; 1481 space -= clen; 1482 SOCKBUF_UNLOCK(&so->so_snd); 1483 if (resid > space) { 1484 error = EMSGSIZE; 1485 goto out; 1486 } 1487 if (uio == NULL) { 1488 resid = 0; 1489 if (flags & MSG_EOR) 1490 top->m_flags |= M_EOR; 1491 } else { 1492 /* 1493 * Copy the data from userland into a mbuf chain. 1494 * If no data is to be copied in, a single empty mbuf 1495 * is returned. 1496 */ 1497 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1498 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1499 if (top == NULL) { 1500 error = EFAULT; /* only possible error */ 1501 goto out; 1502 } 1503 space -= resid - uio->uio_resid; 1504 resid = uio->uio_resid; 1505 } 1506 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1507 /* 1508 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1509 * than with. 1510 */ 1511 if (dontroute) { 1512 SOCK_LOCK(so); 1513 so->so_options |= SO_DONTROUTE; 1514 SOCK_UNLOCK(so); 1515 } 1516 /* 1517 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1518 * of date. We could have received a reset packet in an interrupt or 1519 * maybe we slept while doing page faults in uiomove() etc. We could 1520 * probably recheck again inside the locking protection here, but 1521 * there are probably other places that this also happens. We must 1522 * rethink this. 1523 */ 1524 VNET_SO_ASSERT(so); 1525 error = so->so_proto->pr_send(so, (flags & MSG_OOB) ? PRUS_OOB : 1526 /* 1527 * If the user set MSG_EOF, the protocol understands this flag and 1528 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1529 */ 1530 ((flags & MSG_EOF) && 1531 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1532 (resid <= 0)) ? 1533 PRUS_EOF : 1534 /* If there is more to send set PRUS_MORETOCOME */ 1535 (flags & MSG_MORETOCOME) || 1536 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1537 top, addr, control, td); 1538 if (dontroute) { 1539 SOCK_LOCK(so); 1540 so->so_options &= ~SO_DONTROUTE; 1541 SOCK_UNLOCK(so); 1542 } 1543 clen = 0; 1544 control = NULL; 1545 top = NULL; 1546 out: 1547 if (top != NULL) 1548 m_freem(top); 1549 if (control != NULL) 1550 m_freem(control); 1551 return (error); 1552 } 1553 1554 /* 1555 * Send on a socket. If send must go all at once and message is larger than 1556 * send buffering, then hard error. Lock against other senders. If must go 1557 * all at once and not enough room now, then inform user that this would 1558 * block and do nothing. Otherwise, if nonblocking, send as much as 1559 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1560 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1561 * in mbuf chain must be small enough to send all at once. 1562 * 1563 * Returns nonzero on error, timeout or signal; callers must check for short 1564 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1565 * on return. 1566 */ 1567 int 1568 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1569 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1570 { 1571 long space; 1572 ssize_t resid; 1573 int clen = 0, error, dontroute; 1574 int atomic = sosendallatonce(so) || top; 1575 int pr_send_flag; 1576 #ifdef KERN_TLS 1577 struct ktls_session *tls; 1578 int tls_enq_cnt, tls_send_flag; 1579 uint8_t tls_rtype; 1580 1581 tls = NULL; 1582 tls_rtype = TLS_RLTYPE_APP; 1583 #endif 1584 if (uio != NULL) 1585 resid = uio->uio_resid; 1586 else if ((top->m_flags & M_PKTHDR) != 0) 1587 resid = top->m_pkthdr.len; 1588 else 1589 resid = m_length(top, NULL); 1590 /* 1591 * In theory resid should be unsigned. However, space must be 1592 * signed, as it might be less than 0 if we over-committed, and we 1593 * must use a signed comparison of space and resid. On the other 1594 * hand, a negative resid causes us to loop sending 0-length 1595 * segments to the protocol. 1596 * 1597 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1598 * type sockets since that's an error. 1599 */ 1600 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1601 error = EINVAL; 1602 goto out; 1603 } 1604 1605 dontroute = 1606 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1607 (so->so_proto->pr_flags & PR_ATOMIC); 1608 if (td != NULL) 1609 td->td_ru.ru_msgsnd++; 1610 if (control != NULL) 1611 clen = control->m_len; 1612 1613 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 1614 if (error) 1615 goto out; 1616 1617 #ifdef KERN_TLS 1618 tls_send_flag = 0; 1619 tls = ktls_hold(so->so_snd.sb_tls_info); 1620 if (tls != NULL) { 1621 if (tls->mode == TCP_TLS_MODE_SW) 1622 tls_send_flag = PRUS_NOTREADY; 1623 1624 if (control != NULL) { 1625 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1626 1627 if (clen >= sizeof(*cm) && 1628 cm->cmsg_type == TLS_SET_RECORD_TYPE) { 1629 tls_rtype = *((uint8_t *)CMSG_DATA(cm)); 1630 clen = 0; 1631 m_freem(control); 1632 control = NULL; 1633 atomic = 1; 1634 } 1635 } 1636 1637 if (resid == 0 && !ktls_permit_empty_frames(tls)) { 1638 error = EINVAL; 1639 goto release; 1640 } 1641 } 1642 #endif 1643 1644 restart: 1645 do { 1646 SOCKBUF_LOCK(&so->so_snd); 1647 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1648 SOCKBUF_UNLOCK(&so->so_snd); 1649 error = EPIPE; 1650 goto release; 1651 } 1652 if (so->so_error) { 1653 error = so->so_error; 1654 so->so_error = 0; 1655 SOCKBUF_UNLOCK(&so->so_snd); 1656 goto release; 1657 } 1658 if ((so->so_state & SS_ISCONNECTED) == 0) { 1659 /* 1660 * `sendto' and `sendmsg' is allowed on a connection- 1661 * based socket if it supports implied connect. 1662 * Return ENOTCONN if not connected and no address is 1663 * supplied. 1664 */ 1665 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1666 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1667 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1668 !(resid == 0 && clen != 0)) { 1669 SOCKBUF_UNLOCK(&so->so_snd); 1670 error = ENOTCONN; 1671 goto release; 1672 } 1673 } else if (addr == NULL) { 1674 SOCKBUF_UNLOCK(&so->so_snd); 1675 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1676 error = ENOTCONN; 1677 else 1678 error = EDESTADDRREQ; 1679 goto release; 1680 } 1681 } 1682 space = sbspace(&so->so_snd); 1683 if (flags & MSG_OOB) 1684 space += 1024; 1685 if ((atomic && resid > so->so_snd.sb_hiwat) || 1686 clen > so->so_snd.sb_hiwat) { 1687 SOCKBUF_UNLOCK(&so->so_snd); 1688 error = EMSGSIZE; 1689 goto release; 1690 } 1691 if (space < resid + clen && 1692 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1693 if ((so->so_state & SS_NBIO) || 1694 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) { 1695 SOCKBUF_UNLOCK(&so->so_snd); 1696 error = EWOULDBLOCK; 1697 goto release; 1698 } 1699 error = sbwait(so, SO_SND); 1700 SOCKBUF_UNLOCK(&so->so_snd); 1701 if (error) 1702 goto release; 1703 goto restart; 1704 } 1705 SOCKBUF_UNLOCK(&so->so_snd); 1706 space -= clen; 1707 do { 1708 if (uio == NULL) { 1709 resid = 0; 1710 if (flags & MSG_EOR) 1711 top->m_flags |= M_EOR; 1712 #ifdef KERN_TLS 1713 if (tls != NULL) { 1714 ktls_frame(top, tls, &tls_enq_cnt, 1715 tls_rtype); 1716 tls_rtype = TLS_RLTYPE_APP; 1717 } 1718 #endif 1719 } else { 1720 /* 1721 * Copy the data from userland into a mbuf 1722 * chain. If resid is 0, which can happen 1723 * only if we have control to send, then 1724 * a single empty mbuf is returned. This 1725 * is a workaround to prevent protocol send 1726 * methods to panic. 1727 */ 1728 #ifdef KERN_TLS 1729 if (tls != NULL) { 1730 top = m_uiotombuf(uio, M_WAITOK, space, 1731 tls->params.max_frame_len, 1732 M_EXTPG | 1733 ((flags & MSG_EOR) ? M_EOR : 0)); 1734 if (top != NULL) { 1735 ktls_frame(top, tls, 1736 &tls_enq_cnt, tls_rtype); 1737 } 1738 tls_rtype = TLS_RLTYPE_APP; 1739 } else 1740 #endif 1741 top = m_uiotombuf(uio, M_WAITOK, space, 1742 (atomic ? max_hdr : 0), 1743 (atomic ? M_PKTHDR : 0) | 1744 ((flags & MSG_EOR) ? M_EOR : 0)); 1745 if (top == NULL) { 1746 error = EFAULT; /* only possible error */ 1747 goto release; 1748 } 1749 space -= resid - uio->uio_resid; 1750 resid = uio->uio_resid; 1751 } 1752 if (dontroute) { 1753 SOCK_LOCK(so); 1754 so->so_options |= SO_DONTROUTE; 1755 SOCK_UNLOCK(so); 1756 } 1757 /* 1758 * XXX all the SBS_CANTSENDMORE checks previously 1759 * done could be out of date. We could have received 1760 * a reset packet in an interrupt or maybe we slept 1761 * while doing page faults in uiomove() etc. We 1762 * could probably recheck again inside the locking 1763 * protection here, but there are probably other 1764 * places that this also happens. We must rethink 1765 * this. 1766 */ 1767 VNET_SO_ASSERT(so); 1768 1769 pr_send_flag = (flags & MSG_OOB) ? PRUS_OOB : 1770 /* 1771 * If the user set MSG_EOF, the protocol understands 1772 * this flag and nothing left to send then use 1773 * PRU_SEND_EOF instead of PRU_SEND. 1774 */ 1775 ((flags & MSG_EOF) && 1776 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1777 (resid <= 0)) ? 1778 PRUS_EOF : 1779 /* If there is more to send set PRUS_MORETOCOME. */ 1780 (flags & MSG_MORETOCOME) || 1781 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; 1782 1783 #ifdef KERN_TLS 1784 pr_send_flag |= tls_send_flag; 1785 #endif 1786 1787 error = so->so_proto->pr_send(so, pr_send_flag, top, 1788 addr, control, td); 1789 1790 if (dontroute) { 1791 SOCK_LOCK(so); 1792 so->so_options &= ~SO_DONTROUTE; 1793 SOCK_UNLOCK(so); 1794 } 1795 1796 #ifdef KERN_TLS 1797 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) { 1798 if (error != 0) { 1799 m_freem(top); 1800 top = NULL; 1801 } else { 1802 soref(so); 1803 ktls_enqueue(top, so, tls_enq_cnt); 1804 } 1805 } 1806 #endif 1807 clen = 0; 1808 control = NULL; 1809 top = NULL; 1810 if (error) 1811 goto release; 1812 } while (resid && space > 0); 1813 } while (resid); 1814 1815 release: 1816 SOCK_IO_SEND_UNLOCK(so); 1817 out: 1818 #ifdef KERN_TLS 1819 if (tls != NULL) 1820 ktls_free(tls); 1821 #endif 1822 if (top != NULL) 1823 m_freem(top); 1824 if (control != NULL) 1825 m_freem(control); 1826 return (error); 1827 } 1828 1829 /* 1830 * Send to a socket from a kernel thread. 1831 * 1832 * XXXGL: in almost all cases uio is NULL and the mbuf is supplied. 1833 * Exception is nfs/bootp_subr.c. It is arguable that the VNET context needs 1834 * to be set at all. This function should just boil down to a static inline 1835 * calling the protocol method. 1836 */ 1837 int 1838 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1839 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1840 { 1841 int error; 1842 1843 CURVNET_SET(so->so_vnet); 1844 error = so->so_proto->pr_sosend(so, addr, uio, 1845 top, control, flags, td); 1846 CURVNET_RESTORE(); 1847 return (error); 1848 } 1849 1850 /* 1851 * send(2), write(2) or aio_write(2) on a socket. 1852 */ 1853 int 1854 sousrsend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1855 struct mbuf *control, int flags, struct proc *userproc) 1856 { 1857 struct thread *td; 1858 ssize_t len; 1859 int error; 1860 1861 td = uio->uio_td; 1862 len = uio->uio_resid; 1863 CURVNET_SET(so->so_vnet); 1864 error = so->so_proto->pr_sosend(so, addr, uio, NULL, control, flags, 1865 td); 1866 CURVNET_RESTORE(); 1867 if (error != 0) { 1868 /* 1869 * Clear transient errors for stream protocols if they made 1870 * some progress. Make exclusion for aio(4) that would 1871 * schedule a new write in case of EWOULDBLOCK and clear 1872 * error itself. See soaio_process_job(). 1873 */ 1874 if (uio->uio_resid != len && 1875 (so->so_proto->pr_flags & PR_ATOMIC) == 0 && 1876 userproc == NULL && 1877 (error == ERESTART || error == EINTR || 1878 error == EWOULDBLOCK)) 1879 error = 0; 1880 /* Generation of SIGPIPE can be controlled per socket. */ 1881 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0 && 1882 (flags & MSG_NOSIGNAL) == 0) { 1883 if (userproc != NULL) { 1884 /* aio(4) job */ 1885 PROC_LOCK(userproc); 1886 kern_psignal(userproc, SIGPIPE); 1887 PROC_UNLOCK(userproc); 1888 } else { 1889 PROC_LOCK(td->td_proc); 1890 tdsignal(td, SIGPIPE); 1891 PROC_UNLOCK(td->td_proc); 1892 } 1893 } 1894 } 1895 return (error); 1896 } 1897 1898 /* 1899 * The part of soreceive() that implements reading non-inline out-of-band 1900 * data from a socket. For more complete comments, see soreceive(), from 1901 * which this code originated. 1902 * 1903 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1904 * unable to return an mbuf chain to the caller. 1905 */ 1906 static int 1907 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1908 { 1909 struct protosw *pr = so->so_proto; 1910 struct mbuf *m; 1911 int error; 1912 1913 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1914 VNET_SO_ASSERT(so); 1915 1916 m = m_get(M_WAITOK, MT_DATA); 1917 error = pr->pr_rcvoob(so, m, flags & MSG_PEEK); 1918 if (error) 1919 goto bad; 1920 do { 1921 error = uiomove(mtod(m, void *), 1922 (int) min(uio->uio_resid, m->m_len), uio); 1923 m = m_free(m); 1924 } while (uio->uio_resid && error == 0 && m); 1925 bad: 1926 if (m != NULL) 1927 m_freem(m); 1928 return (error); 1929 } 1930 1931 /* 1932 * Following replacement or removal of the first mbuf on the first mbuf chain 1933 * of a socket buffer, push necessary state changes back into the socket 1934 * buffer so that other consumers see the values consistently. 'nextrecord' 1935 * is the callers locally stored value of the original value of 1936 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1937 * NOTE: 'nextrecord' may be NULL. 1938 */ 1939 static __inline void 1940 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1941 { 1942 1943 SOCKBUF_LOCK_ASSERT(sb); 1944 /* 1945 * First, update for the new value of nextrecord. If necessary, make 1946 * it the first record. 1947 */ 1948 if (sb->sb_mb != NULL) 1949 sb->sb_mb->m_nextpkt = nextrecord; 1950 else 1951 sb->sb_mb = nextrecord; 1952 1953 /* 1954 * Now update any dependent socket buffer fields to reflect the new 1955 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1956 * addition of a second clause that takes care of the case where 1957 * sb_mb has been updated, but remains the last record. 1958 */ 1959 if (sb->sb_mb == NULL) { 1960 sb->sb_mbtail = NULL; 1961 sb->sb_lastrecord = NULL; 1962 } else if (sb->sb_mb->m_nextpkt == NULL) 1963 sb->sb_lastrecord = sb->sb_mb; 1964 } 1965 1966 /* 1967 * Implement receive operations on a socket. We depend on the way that 1968 * records are added to the sockbuf by sbappend. In particular, each record 1969 * (mbufs linked through m_next) must begin with an address if the protocol 1970 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1971 * data, and then zero or more mbufs of data. In order to allow parallelism 1972 * between network receive and copying to user space, as well as avoid 1973 * sleeping with a mutex held, we release the socket buffer mutex during the 1974 * user space copy. Although the sockbuf is locked, new data may still be 1975 * appended, and thus we must maintain consistency of the sockbuf during that 1976 * time. 1977 * 1978 * The caller may receive the data as a single mbuf chain by supplying an 1979 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1980 * the count in uio_resid. 1981 */ 1982 int 1983 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1984 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1985 { 1986 struct mbuf *m, **mp; 1987 int flags, error, offset; 1988 ssize_t len; 1989 struct protosw *pr = so->so_proto; 1990 struct mbuf *nextrecord; 1991 int moff, type = 0; 1992 ssize_t orig_resid = uio->uio_resid; 1993 bool report_real_len = false; 1994 1995 mp = mp0; 1996 if (psa != NULL) 1997 *psa = NULL; 1998 if (controlp != NULL) 1999 *controlp = NULL; 2000 if (flagsp != NULL) { 2001 report_real_len = *flagsp & MSG_TRUNC; 2002 *flagsp &= ~MSG_TRUNC; 2003 flags = *flagsp &~ MSG_EOR; 2004 } else 2005 flags = 0; 2006 if (flags & MSG_OOB) 2007 return (soreceive_rcvoob(so, uio, flags)); 2008 if (mp != NULL) 2009 *mp = NULL; 2010 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 2011 && uio->uio_resid) { 2012 VNET_SO_ASSERT(so); 2013 pr->pr_rcvd(so, 0); 2014 } 2015 2016 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 2017 if (error) 2018 return (error); 2019 2020 restart: 2021 SOCKBUF_LOCK(&so->so_rcv); 2022 m = so->so_rcv.sb_mb; 2023 /* 2024 * If we have less data than requested, block awaiting more (subject 2025 * to any timeout) if: 2026 * 1. the current count is less than the low water mark, or 2027 * 2. MSG_DONTWAIT is not set 2028 */ 2029 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 2030 sbavail(&so->so_rcv) < uio->uio_resid) && 2031 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && 2032 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 2033 KASSERT(m != NULL || !sbavail(&so->so_rcv), 2034 ("receive: m == %p sbavail == %u", 2035 m, sbavail(&so->so_rcv))); 2036 if (so->so_error || so->so_rerror) { 2037 if (m != NULL) 2038 goto dontblock; 2039 if (so->so_error) 2040 error = so->so_error; 2041 else 2042 error = so->so_rerror; 2043 if ((flags & MSG_PEEK) == 0) { 2044 if (so->so_error) 2045 so->so_error = 0; 2046 else 2047 so->so_rerror = 0; 2048 } 2049 SOCKBUF_UNLOCK(&so->so_rcv); 2050 goto release; 2051 } 2052 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2053 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2054 if (m != NULL) 2055 goto dontblock; 2056 #ifdef KERN_TLS 2057 else if (so->so_rcv.sb_tlsdcc == 0 && 2058 so->so_rcv.sb_tlscc == 0) { 2059 #else 2060 else { 2061 #endif 2062 SOCKBUF_UNLOCK(&so->so_rcv); 2063 goto release; 2064 } 2065 } 2066 for (; m != NULL; m = m->m_next) 2067 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 2068 m = so->so_rcv.sb_mb; 2069 goto dontblock; 2070 } 2071 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED | 2072 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 && 2073 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 2074 SOCKBUF_UNLOCK(&so->so_rcv); 2075 error = ENOTCONN; 2076 goto release; 2077 } 2078 if (uio->uio_resid == 0 && !report_real_len) { 2079 SOCKBUF_UNLOCK(&so->so_rcv); 2080 goto release; 2081 } 2082 if ((so->so_state & SS_NBIO) || 2083 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2084 SOCKBUF_UNLOCK(&so->so_rcv); 2085 error = EWOULDBLOCK; 2086 goto release; 2087 } 2088 SBLASTRECORDCHK(&so->so_rcv); 2089 SBLASTMBUFCHK(&so->so_rcv); 2090 error = sbwait(so, SO_RCV); 2091 SOCKBUF_UNLOCK(&so->so_rcv); 2092 if (error) 2093 goto release; 2094 goto restart; 2095 } 2096 dontblock: 2097 /* 2098 * From this point onward, we maintain 'nextrecord' as a cache of the 2099 * pointer to the next record in the socket buffer. We must keep the 2100 * various socket buffer pointers and local stack versions of the 2101 * pointers in sync, pushing out modifications before dropping the 2102 * socket buffer mutex, and re-reading them when picking it up. 2103 * 2104 * Otherwise, we will race with the network stack appending new data 2105 * or records onto the socket buffer by using inconsistent/stale 2106 * versions of the field, possibly resulting in socket buffer 2107 * corruption. 2108 * 2109 * By holding the high-level sblock(), we prevent simultaneous 2110 * readers from pulling off the front of the socket buffer. 2111 */ 2112 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2113 if (uio->uio_td) 2114 uio->uio_td->td_ru.ru_msgrcv++; 2115 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 2116 SBLASTRECORDCHK(&so->so_rcv); 2117 SBLASTMBUFCHK(&so->so_rcv); 2118 nextrecord = m->m_nextpkt; 2119 if (pr->pr_flags & PR_ADDR) { 2120 KASSERT(m->m_type == MT_SONAME, 2121 ("m->m_type == %d", m->m_type)); 2122 orig_resid = 0; 2123 if (psa != NULL) 2124 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2125 M_NOWAIT); 2126 if (flags & MSG_PEEK) { 2127 m = m->m_next; 2128 } else { 2129 sbfree(&so->so_rcv, m); 2130 so->so_rcv.sb_mb = m_free(m); 2131 m = so->so_rcv.sb_mb; 2132 sockbuf_pushsync(&so->so_rcv, nextrecord); 2133 } 2134 } 2135 2136 /* 2137 * Process one or more MT_CONTROL mbufs present before any data mbufs 2138 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 2139 * just copy the data; if !MSG_PEEK, we call into the protocol to 2140 * perform externalization (or freeing if controlp == NULL). 2141 */ 2142 if (m != NULL && m->m_type == MT_CONTROL) { 2143 struct mbuf *cm = NULL, *cmn; 2144 struct mbuf **cme = &cm; 2145 #ifdef KERN_TLS 2146 struct cmsghdr *cmsg; 2147 struct tls_get_record tgr; 2148 2149 /* 2150 * For MSG_TLSAPPDATA, check for an alert record. 2151 * If found, return ENXIO without removing 2152 * it from the receive queue. This allows a subsequent 2153 * call without MSG_TLSAPPDATA to receive it. 2154 * Note that, for TLS, there should only be a single 2155 * control mbuf with the TLS_GET_RECORD message in it. 2156 */ 2157 if (flags & MSG_TLSAPPDATA) { 2158 cmsg = mtod(m, struct cmsghdr *); 2159 if (cmsg->cmsg_type == TLS_GET_RECORD && 2160 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) { 2161 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr)); 2162 if (__predict_false(tgr.tls_type == 2163 TLS_RLTYPE_ALERT)) { 2164 SOCKBUF_UNLOCK(&so->so_rcv); 2165 error = ENXIO; 2166 goto release; 2167 } 2168 } 2169 } 2170 #endif 2171 2172 do { 2173 if (flags & MSG_PEEK) { 2174 if (controlp != NULL) { 2175 *controlp = m_copym(m, 0, m->m_len, 2176 M_NOWAIT); 2177 controlp = &(*controlp)->m_next; 2178 } 2179 m = m->m_next; 2180 } else { 2181 sbfree(&so->so_rcv, m); 2182 so->so_rcv.sb_mb = m->m_next; 2183 m->m_next = NULL; 2184 *cme = m; 2185 cme = &(*cme)->m_next; 2186 m = so->so_rcv.sb_mb; 2187 } 2188 } while (m != NULL && m->m_type == MT_CONTROL); 2189 if ((flags & MSG_PEEK) == 0) 2190 sockbuf_pushsync(&so->so_rcv, nextrecord); 2191 while (cm != NULL) { 2192 cmn = cm->m_next; 2193 cm->m_next = NULL; 2194 if (pr->pr_domain->dom_externalize != NULL) { 2195 SOCKBUF_UNLOCK(&so->so_rcv); 2196 VNET_SO_ASSERT(so); 2197 error = (*pr->pr_domain->dom_externalize) 2198 (cm, controlp, flags); 2199 SOCKBUF_LOCK(&so->so_rcv); 2200 } else if (controlp != NULL) 2201 *controlp = cm; 2202 else 2203 m_freem(cm); 2204 if (controlp != NULL) { 2205 while (*controlp != NULL) 2206 controlp = &(*controlp)->m_next; 2207 } 2208 cm = cmn; 2209 } 2210 if (m != NULL) 2211 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 2212 else 2213 nextrecord = so->so_rcv.sb_mb; 2214 orig_resid = 0; 2215 } 2216 if (m != NULL) { 2217 if ((flags & MSG_PEEK) == 0) { 2218 KASSERT(m->m_nextpkt == nextrecord, 2219 ("soreceive: post-control, nextrecord !sync")); 2220 if (nextrecord == NULL) { 2221 KASSERT(so->so_rcv.sb_mb == m, 2222 ("soreceive: post-control, sb_mb!=m")); 2223 KASSERT(so->so_rcv.sb_lastrecord == m, 2224 ("soreceive: post-control, lastrecord!=m")); 2225 } 2226 } 2227 type = m->m_type; 2228 if (type == MT_OOBDATA) 2229 flags |= MSG_OOB; 2230 } else { 2231 if ((flags & MSG_PEEK) == 0) { 2232 KASSERT(so->so_rcv.sb_mb == nextrecord, 2233 ("soreceive: sb_mb != nextrecord")); 2234 if (so->so_rcv.sb_mb == NULL) { 2235 KASSERT(so->so_rcv.sb_lastrecord == NULL, 2236 ("soreceive: sb_lastercord != NULL")); 2237 } 2238 } 2239 } 2240 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2241 SBLASTRECORDCHK(&so->so_rcv); 2242 SBLASTMBUFCHK(&so->so_rcv); 2243 2244 /* 2245 * Now continue to read any data mbufs off of the head of the socket 2246 * buffer until the read request is satisfied. Note that 'type' is 2247 * used to store the type of any mbuf reads that have happened so far 2248 * such that soreceive() can stop reading if the type changes, which 2249 * causes soreceive() to return only one of regular data and inline 2250 * out-of-band data in a single socket receive operation. 2251 */ 2252 moff = 0; 2253 offset = 0; 2254 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 2255 && error == 0) { 2256 /* 2257 * If the type of mbuf has changed since the last mbuf 2258 * examined ('type'), end the receive operation. 2259 */ 2260 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2261 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 2262 if (type != m->m_type) 2263 break; 2264 } else if (type == MT_OOBDATA) 2265 break; 2266 else 2267 KASSERT(m->m_type == MT_DATA, 2268 ("m->m_type == %d", m->m_type)); 2269 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 2270 len = uio->uio_resid; 2271 if (so->so_oobmark && len > so->so_oobmark - offset) 2272 len = so->so_oobmark - offset; 2273 if (len > m->m_len - moff) 2274 len = m->m_len - moff; 2275 /* 2276 * If mp is set, just pass back the mbufs. Otherwise copy 2277 * them out via the uio, then free. Sockbuf must be 2278 * consistent here (points to current mbuf, it points to next 2279 * record) when we drop priority; we must note any additions 2280 * to the sockbuf when we block interrupts again. 2281 */ 2282 if (mp == NULL) { 2283 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2284 SBLASTRECORDCHK(&so->so_rcv); 2285 SBLASTMBUFCHK(&so->so_rcv); 2286 SOCKBUF_UNLOCK(&so->so_rcv); 2287 if ((m->m_flags & M_EXTPG) != 0) 2288 error = m_unmapped_uiomove(m, moff, uio, 2289 (int)len); 2290 else 2291 error = uiomove(mtod(m, char *) + moff, 2292 (int)len, uio); 2293 SOCKBUF_LOCK(&so->so_rcv); 2294 if (error) { 2295 /* 2296 * The MT_SONAME mbuf has already been removed 2297 * from the record, so it is necessary to 2298 * remove the data mbufs, if any, to preserve 2299 * the invariant in the case of PR_ADDR that 2300 * requires MT_SONAME mbufs at the head of 2301 * each record. 2302 */ 2303 if (pr->pr_flags & PR_ATOMIC && 2304 ((flags & MSG_PEEK) == 0)) 2305 (void)sbdroprecord_locked(&so->so_rcv); 2306 SOCKBUF_UNLOCK(&so->so_rcv); 2307 goto release; 2308 } 2309 } else 2310 uio->uio_resid -= len; 2311 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2312 if (len == m->m_len - moff) { 2313 if (m->m_flags & M_EOR) 2314 flags |= MSG_EOR; 2315 if (flags & MSG_PEEK) { 2316 m = m->m_next; 2317 moff = 0; 2318 } else { 2319 nextrecord = m->m_nextpkt; 2320 sbfree(&so->so_rcv, m); 2321 if (mp != NULL) { 2322 m->m_nextpkt = NULL; 2323 *mp = m; 2324 mp = &m->m_next; 2325 so->so_rcv.sb_mb = m = m->m_next; 2326 *mp = NULL; 2327 } else { 2328 so->so_rcv.sb_mb = m_free(m); 2329 m = so->so_rcv.sb_mb; 2330 } 2331 sockbuf_pushsync(&so->so_rcv, nextrecord); 2332 SBLASTRECORDCHK(&so->so_rcv); 2333 SBLASTMBUFCHK(&so->so_rcv); 2334 } 2335 } else { 2336 if (flags & MSG_PEEK) 2337 moff += len; 2338 else { 2339 if (mp != NULL) { 2340 if (flags & MSG_DONTWAIT) { 2341 *mp = m_copym(m, 0, len, 2342 M_NOWAIT); 2343 if (*mp == NULL) { 2344 /* 2345 * m_copym() couldn't 2346 * allocate an mbuf. 2347 * Adjust uio_resid back 2348 * (it was adjusted 2349 * down by len bytes, 2350 * which we didn't end 2351 * up "copying" over). 2352 */ 2353 uio->uio_resid += len; 2354 break; 2355 } 2356 } else { 2357 SOCKBUF_UNLOCK(&so->so_rcv); 2358 *mp = m_copym(m, 0, len, 2359 M_WAITOK); 2360 SOCKBUF_LOCK(&so->so_rcv); 2361 } 2362 } 2363 sbcut_locked(&so->so_rcv, len); 2364 } 2365 } 2366 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2367 if (so->so_oobmark) { 2368 if ((flags & MSG_PEEK) == 0) { 2369 so->so_oobmark -= len; 2370 if (so->so_oobmark == 0) { 2371 so->so_rcv.sb_state |= SBS_RCVATMARK; 2372 break; 2373 } 2374 } else { 2375 offset += len; 2376 if (offset == so->so_oobmark) 2377 break; 2378 } 2379 } 2380 if (flags & MSG_EOR) 2381 break; 2382 /* 2383 * If the MSG_WAITALL flag is set (for non-atomic socket), we 2384 * must not quit until "uio->uio_resid == 0" or an error 2385 * termination. If a signal/timeout occurs, return with a 2386 * short count but without error. Keep sockbuf locked 2387 * against other readers. 2388 */ 2389 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 2390 !sosendallatonce(so) && nextrecord == NULL) { 2391 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2392 if (so->so_error || so->so_rerror || 2393 so->so_rcv.sb_state & SBS_CANTRCVMORE) 2394 break; 2395 /* 2396 * Notify the protocol that some data has been 2397 * drained before blocking. 2398 */ 2399 if (pr->pr_flags & PR_WANTRCVD) { 2400 SOCKBUF_UNLOCK(&so->so_rcv); 2401 VNET_SO_ASSERT(so); 2402 pr->pr_rcvd(so, flags); 2403 SOCKBUF_LOCK(&so->so_rcv); 2404 } 2405 SBLASTRECORDCHK(&so->so_rcv); 2406 SBLASTMBUFCHK(&so->so_rcv); 2407 /* 2408 * We could receive some data while was notifying 2409 * the protocol. Skip blocking in this case. 2410 */ 2411 if (so->so_rcv.sb_mb == NULL) { 2412 error = sbwait(so, SO_RCV); 2413 if (error) { 2414 SOCKBUF_UNLOCK(&so->so_rcv); 2415 goto release; 2416 } 2417 } 2418 m = so->so_rcv.sb_mb; 2419 if (m != NULL) 2420 nextrecord = m->m_nextpkt; 2421 } 2422 } 2423 2424 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2425 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 2426 if (report_real_len) 2427 uio->uio_resid -= m_length(m, NULL) - moff; 2428 flags |= MSG_TRUNC; 2429 if ((flags & MSG_PEEK) == 0) 2430 (void) sbdroprecord_locked(&so->so_rcv); 2431 } 2432 if ((flags & MSG_PEEK) == 0) { 2433 if (m == NULL) { 2434 /* 2435 * First part is an inline SB_EMPTY_FIXUP(). Second 2436 * part makes sure sb_lastrecord is up-to-date if 2437 * there is still data in the socket buffer. 2438 */ 2439 so->so_rcv.sb_mb = nextrecord; 2440 if (so->so_rcv.sb_mb == NULL) { 2441 so->so_rcv.sb_mbtail = NULL; 2442 so->so_rcv.sb_lastrecord = NULL; 2443 } else if (nextrecord->m_nextpkt == NULL) 2444 so->so_rcv.sb_lastrecord = nextrecord; 2445 } 2446 SBLASTRECORDCHK(&so->so_rcv); 2447 SBLASTMBUFCHK(&so->so_rcv); 2448 /* 2449 * If soreceive() is being done from the socket callback, 2450 * then don't need to generate ACK to peer to update window, 2451 * since ACK will be generated on return to TCP. 2452 */ 2453 if (!(flags & MSG_SOCALLBCK) && 2454 (pr->pr_flags & PR_WANTRCVD)) { 2455 SOCKBUF_UNLOCK(&so->so_rcv); 2456 VNET_SO_ASSERT(so); 2457 pr->pr_rcvd(so, flags); 2458 SOCKBUF_LOCK(&so->so_rcv); 2459 } 2460 } 2461 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2462 if (orig_resid == uio->uio_resid && orig_resid && 2463 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 2464 SOCKBUF_UNLOCK(&so->so_rcv); 2465 goto restart; 2466 } 2467 SOCKBUF_UNLOCK(&so->so_rcv); 2468 2469 if (flagsp != NULL) 2470 *flagsp |= flags; 2471 release: 2472 SOCK_IO_RECV_UNLOCK(so); 2473 return (error); 2474 } 2475 2476 /* 2477 * Optimized version of soreceive() for stream (TCP) sockets. 2478 */ 2479 int 2480 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 2481 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2482 { 2483 int len = 0, error = 0, flags, oresid; 2484 struct sockbuf *sb; 2485 struct mbuf *m, *n = NULL; 2486 2487 /* We only do stream sockets. */ 2488 if (so->so_type != SOCK_STREAM) 2489 return (EINVAL); 2490 if (psa != NULL) 2491 *psa = NULL; 2492 if (flagsp != NULL) 2493 flags = *flagsp &~ MSG_EOR; 2494 else 2495 flags = 0; 2496 if (controlp != NULL) 2497 *controlp = NULL; 2498 if (flags & MSG_OOB) 2499 return (soreceive_rcvoob(so, uio, flags)); 2500 if (mp0 != NULL) 2501 *mp0 = NULL; 2502 2503 sb = &so->so_rcv; 2504 2505 #ifdef KERN_TLS 2506 /* 2507 * KTLS store TLS records as records with a control message to 2508 * describe the framing. 2509 * 2510 * We check once here before acquiring locks to optimize the 2511 * common case. 2512 */ 2513 if (sb->sb_tls_info != NULL) 2514 return (soreceive_generic(so, psa, uio, mp0, controlp, 2515 flagsp)); 2516 #endif 2517 2518 /* Prevent other readers from entering the socket. */ 2519 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 2520 if (error) 2521 return (error); 2522 SOCKBUF_LOCK(sb); 2523 2524 #ifdef KERN_TLS 2525 if (sb->sb_tls_info != NULL) { 2526 SOCKBUF_UNLOCK(sb); 2527 SOCK_IO_RECV_UNLOCK(so); 2528 return (soreceive_generic(so, psa, uio, mp0, controlp, 2529 flagsp)); 2530 } 2531 #endif 2532 2533 /* Easy one, no space to copyout anything. */ 2534 if (uio->uio_resid == 0) { 2535 error = EINVAL; 2536 goto out; 2537 } 2538 oresid = uio->uio_resid; 2539 2540 /* We will never ever get anything unless we are or were connected. */ 2541 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 2542 error = ENOTCONN; 2543 goto out; 2544 } 2545 2546 restart: 2547 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2548 2549 /* Abort if socket has reported problems. */ 2550 if (so->so_error) { 2551 if (sbavail(sb) > 0) 2552 goto deliver; 2553 if (oresid > uio->uio_resid) 2554 goto out; 2555 error = so->so_error; 2556 if (!(flags & MSG_PEEK)) 2557 so->so_error = 0; 2558 goto out; 2559 } 2560 2561 /* Door is closed. Deliver what is left, if any. */ 2562 if (sb->sb_state & SBS_CANTRCVMORE) { 2563 if (sbavail(sb) > 0) 2564 goto deliver; 2565 else 2566 goto out; 2567 } 2568 2569 /* Socket buffer is empty and we shall not block. */ 2570 if (sbavail(sb) == 0 && 2571 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2572 error = EAGAIN; 2573 goto out; 2574 } 2575 2576 /* Socket buffer got some data that we shall deliver now. */ 2577 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 2578 ((so->so_state & SS_NBIO) || 2579 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2580 sbavail(sb) >= sb->sb_lowat || 2581 sbavail(sb) >= uio->uio_resid || 2582 sbavail(sb) >= sb->sb_hiwat) ) { 2583 goto deliver; 2584 } 2585 2586 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2587 if ((flags & MSG_WAITALL) && 2588 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 2589 goto deliver; 2590 2591 /* 2592 * Wait and block until (more) data comes in. 2593 * NB: Drops the sockbuf lock during wait. 2594 */ 2595 error = sbwait(so, SO_RCV); 2596 if (error) 2597 goto out; 2598 goto restart; 2599 2600 deliver: 2601 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2602 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 2603 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2604 2605 /* Statistics. */ 2606 if (uio->uio_td) 2607 uio->uio_td->td_ru.ru_msgrcv++; 2608 2609 /* Fill uio until full or current end of socket buffer is reached. */ 2610 len = min(uio->uio_resid, sbavail(sb)); 2611 if (mp0 != NULL) { 2612 /* Dequeue as many mbufs as possible. */ 2613 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2614 if (*mp0 == NULL) 2615 *mp0 = sb->sb_mb; 2616 else 2617 m_cat(*mp0, sb->sb_mb); 2618 for (m = sb->sb_mb; 2619 m != NULL && m->m_len <= len; 2620 m = m->m_next) { 2621 KASSERT(!(m->m_flags & M_NOTAVAIL), 2622 ("%s: m %p not available", __func__, m)); 2623 len -= m->m_len; 2624 uio->uio_resid -= m->m_len; 2625 sbfree(sb, m); 2626 n = m; 2627 } 2628 n->m_next = NULL; 2629 sb->sb_mb = m; 2630 sb->sb_lastrecord = sb->sb_mb; 2631 if (sb->sb_mb == NULL) 2632 SB_EMPTY_FIXUP(sb); 2633 } 2634 /* Copy the remainder. */ 2635 if (len > 0) { 2636 KASSERT(sb->sb_mb != NULL, 2637 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2638 2639 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2640 if (m == NULL) 2641 len = 0; /* Don't flush data from sockbuf. */ 2642 else 2643 uio->uio_resid -= len; 2644 if (*mp0 != NULL) 2645 m_cat(*mp0, m); 2646 else 2647 *mp0 = m; 2648 if (*mp0 == NULL) { 2649 error = ENOBUFS; 2650 goto out; 2651 } 2652 } 2653 } else { 2654 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2655 SOCKBUF_UNLOCK(sb); 2656 error = m_mbuftouio(uio, sb->sb_mb, len); 2657 SOCKBUF_LOCK(sb); 2658 if (error) 2659 goto out; 2660 } 2661 SBLASTRECORDCHK(sb); 2662 SBLASTMBUFCHK(sb); 2663 2664 /* 2665 * Remove the delivered data from the socket buffer unless we 2666 * were only peeking. 2667 */ 2668 if (!(flags & MSG_PEEK)) { 2669 if (len > 0) 2670 sbdrop_locked(sb, len); 2671 2672 /* Notify protocol that we drained some data. */ 2673 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2674 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2675 !(flags & MSG_SOCALLBCK))) { 2676 SOCKBUF_UNLOCK(sb); 2677 VNET_SO_ASSERT(so); 2678 so->so_proto->pr_rcvd(so, flags); 2679 SOCKBUF_LOCK(sb); 2680 } 2681 } 2682 2683 /* 2684 * For MSG_WAITALL we may have to loop again and wait for 2685 * more data to come in. 2686 */ 2687 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2688 goto restart; 2689 out: 2690 SBLASTRECORDCHK(sb); 2691 SBLASTMBUFCHK(sb); 2692 SOCKBUF_UNLOCK(sb); 2693 SOCK_IO_RECV_UNLOCK(so); 2694 return (error); 2695 } 2696 2697 /* 2698 * Optimized version of soreceive() for simple datagram cases from userspace. 2699 * Unlike in the stream case, we're able to drop a datagram if copyout() 2700 * fails, and because we handle datagrams atomically, we don't need to use a 2701 * sleep lock to prevent I/O interlacing. 2702 */ 2703 int 2704 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2705 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2706 { 2707 struct mbuf *m, *m2; 2708 int flags, error; 2709 ssize_t len; 2710 struct protosw *pr = so->so_proto; 2711 struct mbuf *nextrecord; 2712 2713 if (psa != NULL) 2714 *psa = NULL; 2715 if (controlp != NULL) 2716 *controlp = NULL; 2717 if (flagsp != NULL) 2718 flags = *flagsp &~ MSG_EOR; 2719 else 2720 flags = 0; 2721 2722 /* 2723 * For any complicated cases, fall back to the full 2724 * soreceive_generic(). 2725 */ 2726 if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC))) 2727 return (soreceive_generic(so, psa, uio, mp0, controlp, 2728 flagsp)); 2729 2730 /* 2731 * Enforce restrictions on use. 2732 */ 2733 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2734 ("soreceive_dgram: wantrcvd")); 2735 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2736 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2737 ("soreceive_dgram: SBS_RCVATMARK")); 2738 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2739 ("soreceive_dgram: P_CONNREQUIRED")); 2740 2741 /* 2742 * Loop blocking while waiting for a datagram. 2743 */ 2744 SOCKBUF_LOCK(&so->so_rcv); 2745 while ((m = so->so_rcv.sb_mb) == NULL) { 2746 KASSERT(sbavail(&so->so_rcv) == 0, 2747 ("soreceive_dgram: sb_mb NULL but sbavail %u", 2748 sbavail(&so->so_rcv))); 2749 if (so->so_error) { 2750 error = so->so_error; 2751 so->so_error = 0; 2752 SOCKBUF_UNLOCK(&so->so_rcv); 2753 return (error); 2754 } 2755 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2756 uio->uio_resid == 0) { 2757 SOCKBUF_UNLOCK(&so->so_rcv); 2758 return (0); 2759 } 2760 if ((so->so_state & SS_NBIO) || 2761 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2762 SOCKBUF_UNLOCK(&so->so_rcv); 2763 return (EWOULDBLOCK); 2764 } 2765 SBLASTRECORDCHK(&so->so_rcv); 2766 SBLASTMBUFCHK(&so->so_rcv); 2767 error = sbwait(so, SO_RCV); 2768 if (error) { 2769 SOCKBUF_UNLOCK(&so->so_rcv); 2770 return (error); 2771 } 2772 } 2773 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2774 2775 if (uio->uio_td) 2776 uio->uio_td->td_ru.ru_msgrcv++; 2777 SBLASTRECORDCHK(&so->so_rcv); 2778 SBLASTMBUFCHK(&so->so_rcv); 2779 nextrecord = m->m_nextpkt; 2780 if (nextrecord == NULL) { 2781 KASSERT(so->so_rcv.sb_lastrecord == m, 2782 ("soreceive_dgram: lastrecord != m")); 2783 } 2784 2785 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2786 ("soreceive_dgram: m_nextpkt != nextrecord")); 2787 2788 /* 2789 * Pull 'm' and its chain off the front of the packet queue. 2790 */ 2791 so->so_rcv.sb_mb = NULL; 2792 sockbuf_pushsync(&so->so_rcv, nextrecord); 2793 2794 /* 2795 * Walk 'm's chain and free that many bytes from the socket buffer. 2796 */ 2797 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2798 sbfree(&so->so_rcv, m2); 2799 2800 /* 2801 * Do a few last checks before we let go of the lock. 2802 */ 2803 SBLASTRECORDCHK(&so->so_rcv); 2804 SBLASTMBUFCHK(&so->so_rcv); 2805 SOCKBUF_UNLOCK(&so->so_rcv); 2806 2807 if (pr->pr_flags & PR_ADDR) { 2808 KASSERT(m->m_type == MT_SONAME, 2809 ("m->m_type == %d", m->m_type)); 2810 if (psa != NULL) 2811 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2812 M_NOWAIT); 2813 m = m_free(m); 2814 } 2815 if (m == NULL) { 2816 /* XXXRW: Can this happen? */ 2817 return (0); 2818 } 2819 2820 /* 2821 * Packet to copyout() is now in 'm' and it is disconnected from the 2822 * queue. 2823 * 2824 * Process one or more MT_CONTROL mbufs present before any data mbufs 2825 * in the first mbuf chain on the socket buffer. We call into the 2826 * protocol to perform externalization (or freeing if controlp == 2827 * NULL). In some cases there can be only MT_CONTROL mbufs without 2828 * MT_DATA mbufs. 2829 */ 2830 if (m->m_type == MT_CONTROL) { 2831 struct mbuf *cm = NULL, *cmn; 2832 struct mbuf **cme = &cm; 2833 2834 do { 2835 m2 = m->m_next; 2836 m->m_next = NULL; 2837 *cme = m; 2838 cme = &(*cme)->m_next; 2839 m = m2; 2840 } while (m != NULL && m->m_type == MT_CONTROL); 2841 while (cm != NULL) { 2842 cmn = cm->m_next; 2843 cm->m_next = NULL; 2844 if (pr->pr_domain->dom_externalize != NULL) { 2845 error = (*pr->pr_domain->dom_externalize) 2846 (cm, controlp, flags); 2847 } else if (controlp != NULL) 2848 *controlp = cm; 2849 else 2850 m_freem(cm); 2851 if (controlp != NULL) { 2852 while (*controlp != NULL) 2853 controlp = &(*controlp)->m_next; 2854 } 2855 cm = cmn; 2856 } 2857 } 2858 KASSERT(m == NULL || m->m_type == MT_DATA, 2859 ("soreceive_dgram: !data")); 2860 while (m != NULL && uio->uio_resid > 0) { 2861 len = uio->uio_resid; 2862 if (len > m->m_len) 2863 len = m->m_len; 2864 error = uiomove(mtod(m, char *), (int)len, uio); 2865 if (error) { 2866 m_freem(m); 2867 return (error); 2868 } 2869 if (len == m->m_len) 2870 m = m_free(m); 2871 else { 2872 m->m_data += len; 2873 m->m_len -= len; 2874 } 2875 } 2876 if (m != NULL) { 2877 flags |= MSG_TRUNC; 2878 m_freem(m); 2879 } 2880 if (flagsp != NULL) 2881 *flagsp |= flags; 2882 return (0); 2883 } 2884 2885 int 2886 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2887 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2888 { 2889 int error; 2890 2891 CURVNET_SET(so->so_vnet); 2892 error = so->so_proto->pr_soreceive(so, psa, uio, mp0, controlp, flagsp); 2893 CURVNET_RESTORE(); 2894 return (error); 2895 } 2896 2897 int 2898 soshutdown(struct socket *so, int how) 2899 { 2900 struct protosw *pr; 2901 int error, soerror_enotconn; 2902 2903 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2904 return (EINVAL); 2905 2906 soerror_enotconn = 0; 2907 SOCK_LOCK(so); 2908 if ((so->so_state & 2909 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 2910 /* 2911 * POSIX mandates us to return ENOTCONN when shutdown(2) is 2912 * invoked on a datagram sockets, however historically we would 2913 * actually tear socket down. This is known to be leveraged by 2914 * some applications to unblock process waiting in recvXXX(2) 2915 * by other process that it shares that socket with. Try to meet 2916 * both backward-compatibility and POSIX requirements by forcing 2917 * ENOTCONN but still asking protocol to perform pru_shutdown(). 2918 */ 2919 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) { 2920 SOCK_UNLOCK(so); 2921 return (ENOTCONN); 2922 } 2923 soerror_enotconn = 1; 2924 } 2925 2926 if (SOLISTENING(so)) { 2927 if (how != SHUT_WR) { 2928 so->so_error = ECONNABORTED; 2929 solisten_wakeup(so); /* unlocks so */ 2930 } else { 2931 SOCK_UNLOCK(so); 2932 } 2933 goto done; 2934 } 2935 SOCK_UNLOCK(so); 2936 2937 CURVNET_SET(so->so_vnet); 2938 pr = so->so_proto; 2939 if (pr->pr_flush != NULL) 2940 pr->pr_flush(so, how); 2941 if (how != SHUT_WR) 2942 sorflush(so); 2943 if (how != SHUT_RD) { 2944 error = pr->pr_shutdown(so); 2945 wakeup(&so->so_timeo); 2946 CURVNET_RESTORE(); 2947 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error); 2948 } 2949 wakeup(&so->so_timeo); 2950 CURVNET_RESTORE(); 2951 2952 done: 2953 return (soerror_enotconn ? ENOTCONN : 0); 2954 } 2955 2956 void 2957 sorflush(struct socket *so) 2958 { 2959 struct protosw *pr; 2960 int error; 2961 2962 VNET_SO_ASSERT(so); 2963 2964 /* 2965 * Dislodge threads currently blocked in receive and wait to acquire 2966 * a lock against other simultaneous readers before clearing the 2967 * socket buffer. Don't let our acquire be interrupted by a signal 2968 * despite any existing socket disposition on interruptable waiting. 2969 */ 2970 socantrcvmore(so); 2971 2972 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR); 2973 if (error != 0) { 2974 KASSERT(SOLISTENING(so), 2975 ("%s: soiolock(%p) failed", __func__, so)); 2976 return; 2977 } 2978 2979 pr = so->so_proto; 2980 if (pr->pr_flags & PR_RIGHTS) { 2981 MPASS(pr->pr_domain->dom_dispose != NULL); 2982 (*pr->pr_domain->dom_dispose)(so); 2983 } else { 2984 sbrelease(so, SO_RCV); 2985 SOCK_IO_RECV_UNLOCK(so); 2986 } 2987 2988 } 2989 2990 /* 2991 * Wrapper for Socket established helper hook. 2992 * Parameters: socket, context of the hook point, hook id. 2993 */ 2994 static int inline 2995 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 2996 { 2997 struct socket_hhook_data hhook_data = { 2998 .so = so, 2999 .hctx = hctx, 3000 .m = NULL, 3001 .status = 0 3002 }; 3003 3004 CURVNET_SET(so->so_vnet); 3005 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 3006 CURVNET_RESTORE(); 3007 3008 /* Ugly but needed, since hhooks return void for now */ 3009 return (hhook_data.status); 3010 } 3011 3012 /* 3013 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 3014 * additional variant to handle the case where the option value needs to be 3015 * some kind of integer, but not a specific size. In addition to their use 3016 * here, these functions are also called by the protocol-level pr_ctloutput() 3017 * routines. 3018 */ 3019 int 3020 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 3021 { 3022 size_t valsize; 3023 3024 /* 3025 * If the user gives us more than we wanted, we ignore it, but if we 3026 * don't get the minimum length the caller wants, we return EINVAL. 3027 * On success, sopt->sopt_valsize is set to however much we actually 3028 * retrieved. 3029 */ 3030 if ((valsize = sopt->sopt_valsize) < minlen) 3031 return EINVAL; 3032 if (valsize > len) 3033 sopt->sopt_valsize = valsize = len; 3034 3035 if (sopt->sopt_td != NULL) 3036 return (copyin(sopt->sopt_val, buf, valsize)); 3037 3038 bcopy(sopt->sopt_val, buf, valsize); 3039 return (0); 3040 } 3041 3042 /* 3043 * Kernel version of setsockopt(2). 3044 * 3045 * XXX: optlen is size_t, not socklen_t 3046 */ 3047 int 3048 so_setsockopt(struct socket *so, int level, int optname, void *optval, 3049 size_t optlen) 3050 { 3051 struct sockopt sopt; 3052 3053 sopt.sopt_level = level; 3054 sopt.sopt_name = optname; 3055 sopt.sopt_dir = SOPT_SET; 3056 sopt.sopt_val = optval; 3057 sopt.sopt_valsize = optlen; 3058 sopt.sopt_td = NULL; 3059 return (sosetopt(so, &sopt)); 3060 } 3061 3062 int 3063 sosetopt(struct socket *so, struct sockopt *sopt) 3064 { 3065 int error, optval; 3066 struct linger l; 3067 struct timeval tv; 3068 sbintime_t val, *valp; 3069 uint32_t val32; 3070 #ifdef MAC 3071 struct mac extmac; 3072 #endif 3073 3074 CURVNET_SET(so->so_vnet); 3075 error = 0; 3076 if (sopt->sopt_level != SOL_SOCKET) { 3077 if (so->so_proto->pr_ctloutput != NULL) 3078 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3079 else 3080 error = ENOPROTOOPT; 3081 } else { 3082 switch (sopt->sopt_name) { 3083 case SO_ACCEPTFILTER: 3084 error = accept_filt_setopt(so, sopt); 3085 if (error) 3086 goto bad; 3087 break; 3088 3089 case SO_LINGER: 3090 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 3091 if (error) 3092 goto bad; 3093 if (l.l_linger < 0 || 3094 l.l_linger > USHRT_MAX || 3095 l.l_linger > (INT_MAX / hz)) { 3096 error = EDOM; 3097 goto bad; 3098 } 3099 SOCK_LOCK(so); 3100 so->so_linger = l.l_linger; 3101 if (l.l_onoff) 3102 so->so_options |= SO_LINGER; 3103 else 3104 so->so_options &= ~SO_LINGER; 3105 SOCK_UNLOCK(so); 3106 break; 3107 3108 case SO_DEBUG: 3109 case SO_KEEPALIVE: 3110 case SO_DONTROUTE: 3111 case SO_USELOOPBACK: 3112 case SO_BROADCAST: 3113 case SO_REUSEADDR: 3114 case SO_REUSEPORT: 3115 case SO_REUSEPORT_LB: 3116 case SO_OOBINLINE: 3117 case SO_TIMESTAMP: 3118 case SO_BINTIME: 3119 case SO_NOSIGPIPE: 3120 case SO_NO_DDP: 3121 case SO_NO_OFFLOAD: 3122 case SO_RERROR: 3123 error = sooptcopyin(sopt, &optval, sizeof optval, 3124 sizeof optval); 3125 if (error) 3126 goto bad; 3127 SOCK_LOCK(so); 3128 if (optval) 3129 so->so_options |= sopt->sopt_name; 3130 else 3131 so->so_options &= ~sopt->sopt_name; 3132 SOCK_UNLOCK(so); 3133 break; 3134 3135 case SO_SETFIB: 3136 error = sooptcopyin(sopt, &optval, sizeof optval, 3137 sizeof optval); 3138 if (error) 3139 goto bad; 3140 3141 if (optval < 0 || optval >= rt_numfibs) { 3142 error = EINVAL; 3143 goto bad; 3144 } 3145 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 3146 (so->so_proto->pr_domain->dom_family == PF_INET6) || 3147 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 3148 so->so_fibnum = optval; 3149 else 3150 so->so_fibnum = 0; 3151 break; 3152 3153 case SO_USER_COOKIE: 3154 error = sooptcopyin(sopt, &val32, sizeof val32, 3155 sizeof val32); 3156 if (error) 3157 goto bad; 3158 so->so_user_cookie = val32; 3159 break; 3160 3161 case SO_SNDBUF: 3162 case SO_RCVBUF: 3163 case SO_SNDLOWAT: 3164 case SO_RCVLOWAT: 3165 error = so->so_proto->pr_setsbopt(so, sopt); 3166 if (error) 3167 goto bad; 3168 break; 3169 3170 case SO_SNDTIMEO: 3171 case SO_RCVTIMEO: 3172 #ifdef COMPAT_FREEBSD32 3173 if (SV_CURPROC_FLAG(SV_ILP32)) { 3174 struct timeval32 tv32; 3175 3176 error = sooptcopyin(sopt, &tv32, sizeof tv32, 3177 sizeof tv32); 3178 CP(tv32, tv, tv_sec); 3179 CP(tv32, tv, tv_usec); 3180 } else 3181 #endif 3182 error = sooptcopyin(sopt, &tv, sizeof tv, 3183 sizeof tv); 3184 if (error) 3185 goto bad; 3186 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 3187 tv.tv_usec >= 1000000) { 3188 error = EDOM; 3189 goto bad; 3190 } 3191 if (tv.tv_sec > INT32_MAX) 3192 val = SBT_MAX; 3193 else 3194 val = tvtosbt(tv); 3195 SOCK_LOCK(so); 3196 valp = sopt->sopt_name == SO_SNDTIMEO ? 3197 (SOLISTENING(so) ? &so->sol_sbsnd_timeo : 3198 &so->so_snd.sb_timeo) : 3199 (SOLISTENING(so) ? &so->sol_sbrcv_timeo : 3200 &so->so_rcv.sb_timeo); 3201 *valp = val; 3202 SOCK_UNLOCK(so); 3203 break; 3204 3205 case SO_LABEL: 3206 #ifdef MAC 3207 error = sooptcopyin(sopt, &extmac, sizeof extmac, 3208 sizeof extmac); 3209 if (error) 3210 goto bad; 3211 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 3212 so, &extmac); 3213 #else 3214 error = EOPNOTSUPP; 3215 #endif 3216 break; 3217 3218 case SO_TS_CLOCK: 3219 error = sooptcopyin(sopt, &optval, sizeof optval, 3220 sizeof optval); 3221 if (error) 3222 goto bad; 3223 if (optval < 0 || optval > SO_TS_CLOCK_MAX) { 3224 error = EINVAL; 3225 goto bad; 3226 } 3227 so->so_ts_clock = optval; 3228 break; 3229 3230 case SO_MAX_PACING_RATE: 3231 error = sooptcopyin(sopt, &val32, sizeof(val32), 3232 sizeof(val32)); 3233 if (error) 3234 goto bad; 3235 so->so_max_pacing_rate = val32; 3236 break; 3237 3238 default: 3239 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3240 error = hhook_run_socket(so, sopt, 3241 HHOOK_SOCKET_OPT); 3242 else 3243 error = ENOPROTOOPT; 3244 break; 3245 } 3246 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 3247 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 3248 } 3249 bad: 3250 CURVNET_RESTORE(); 3251 return (error); 3252 } 3253 3254 /* 3255 * Helper routine for getsockopt. 3256 */ 3257 int 3258 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 3259 { 3260 int error; 3261 size_t valsize; 3262 3263 error = 0; 3264 3265 /* 3266 * Documented get behavior is that we always return a value, possibly 3267 * truncated to fit in the user's buffer. Traditional behavior is 3268 * that we always tell the user precisely how much we copied, rather 3269 * than something useful like the total amount we had available for 3270 * her. Note that this interface is not idempotent; the entire 3271 * answer must be generated ahead of time. 3272 */ 3273 valsize = min(len, sopt->sopt_valsize); 3274 sopt->sopt_valsize = valsize; 3275 if (sopt->sopt_val != NULL) { 3276 if (sopt->sopt_td != NULL) 3277 error = copyout(buf, sopt->sopt_val, valsize); 3278 else 3279 bcopy(buf, sopt->sopt_val, valsize); 3280 } 3281 return (error); 3282 } 3283 3284 int 3285 sogetopt(struct socket *so, struct sockopt *sopt) 3286 { 3287 int error, optval; 3288 struct linger l; 3289 struct timeval tv; 3290 #ifdef MAC 3291 struct mac extmac; 3292 #endif 3293 3294 CURVNET_SET(so->so_vnet); 3295 error = 0; 3296 if (sopt->sopt_level != SOL_SOCKET) { 3297 if (so->so_proto->pr_ctloutput != NULL) 3298 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3299 else 3300 error = ENOPROTOOPT; 3301 CURVNET_RESTORE(); 3302 return (error); 3303 } else { 3304 switch (sopt->sopt_name) { 3305 case SO_ACCEPTFILTER: 3306 error = accept_filt_getopt(so, sopt); 3307 break; 3308 3309 case SO_LINGER: 3310 SOCK_LOCK(so); 3311 l.l_onoff = so->so_options & SO_LINGER; 3312 l.l_linger = so->so_linger; 3313 SOCK_UNLOCK(so); 3314 error = sooptcopyout(sopt, &l, sizeof l); 3315 break; 3316 3317 case SO_USELOOPBACK: 3318 case SO_DONTROUTE: 3319 case SO_DEBUG: 3320 case SO_KEEPALIVE: 3321 case SO_REUSEADDR: 3322 case SO_REUSEPORT: 3323 case SO_REUSEPORT_LB: 3324 case SO_BROADCAST: 3325 case SO_OOBINLINE: 3326 case SO_ACCEPTCONN: 3327 case SO_TIMESTAMP: 3328 case SO_BINTIME: 3329 case SO_NOSIGPIPE: 3330 case SO_NO_DDP: 3331 case SO_NO_OFFLOAD: 3332 case SO_RERROR: 3333 optval = so->so_options & sopt->sopt_name; 3334 integer: 3335 error = sooptcopyout(sopt, &optval, sizeof optval); 3336 break; 3337 3338 case SO_DOMAIN: 3339 optval = so->so_proto->pr_domain->dom_family; 3340 goto integer; 3341 3342 case SO_TYPE: 3343 optval = so->so_type; 3344 goto integer; 3345 3346 case SO_PROTOCOL: 3347 optval = so->so_proto->pr_protocol; 3348 goto integer; 3349 3350 case SO_ERROR: 3351 SOCK_LOCK(so); 3352 if (so->so_error) { 3353 optval = so->so_error; 3354 so->so_error = 0; 3355 } else { 3356 optval = so->so_rerror; 3357 so->so_rerror = 0; 3358 } 3359 SOCK_UNLOCK(so); 3360 goto integer; 3361 3362 case SO_SNDBUF: 3363 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : 3364 so->so_snd.sb_hiwat; 3365 goto integer; 3366 3367 case SO_RCVBUF: 3368 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : 3369 so->so_rcv.sb_hiwat; 3370 goto integer; 3371 3372 case SO_SNDLOWAT: 3373 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : 3374 so->so_snd.sb_lowat; 3375 goto integer; 3376 3377 case SO_RCVLOWAT: 3378 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : 3379 so->so_rcv.sb_lowat; 3380 goto integer; 3381 3382 case SO_SNDTIMEO: 3383 case SO_RCVTIMEO: 3384 SOCK_LOCK(so); 3385 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 3386 (SOLISTENING(so) ? so->sol_sbsnd_timeo : 3387 so->so_snd.sb_timeo) : 3388 (SOLISTENING(so) ? so->sol_sbrcv_timeo : 3389 so->so_rcv.sb_timeo)); 3390 SOCK_UNLOCK(so); 3391 #ifdef COMPAT_FREEBSD32 3392 if (SV_CURPROC_FLAG(SV_ILP32)) { 3393 struct timeval32 tv32; 3394 3395 CP(tv, tv32, tv_sec); 3396 CP(tv, tv32, tv_usec); 3397 error = sooptcopyout(sopt, &tv32, sizeof tv32); 3398 } else 3399 #endif 3400 error = sooptcopyout(sopt, &tv, sizeof tv); 3401 break; 3402 3403 case SO_LABEL: 3404 #ifdef MAC 3405 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3406 sizeof(extmac)); 3407 if (error) 3408 goto bad; 3409 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 3410 so, &extmac); 3411 if (error) 3412 goto bad; 3413 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3414 #else 3415 error = EOPNOTSUPP; 3416 #endif 3417 break; 3418 3419 case SO_PEERLABEL: 3420 #ifdef MAC 3421 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3422 sizeof(extmac)); 3423 if (error) 3424 goto bad; 3425 error = mac_getsockopt_peerlabel( 3426 sopt->sopt_td->td_ucred, so, &extmac); 3427 if (error) 3428 goto bad; 3429 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3430 #else 3431 error = EOPNOTSUPP; 3432 #endif 3433 break; 3434 3435 case SO_LISTENQLIMIT: 3436 optval = SOLISTENING(so) ? so->sol_qlimit : 0; 3437 goto integer; 3438 3439 case SO_LISTENQLEN: 3440 optval = SOLISTENING(so) ? so->sol_qlen : 0; 3441 goto integer; 3442 3443 case SO_LISTENINCQLEN: 3444 optval = SOLISTENING(so) ? so->sol_incqlen : 0; 3445 goto integer; 3446 3447 case SO_TS_CLOCK: 3448 optval = so->so_ts_clock; 3449 goto integer; 3450 3451 case SO_MAX_PACING_RATE: 3452 optval = so->so_max_pacing_rate; 3453 goto integer; 3454 3455 default: 3456 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3457 error = hhook_run_socket(so, sopt, 3458 HHOOK_SOCKET_OPT); 3459 else 3460 error = ENOPROTOOPT; 3461 break; 3462 } 3463 } 3464 #ifdef MAC 3465 bad: 3466 #endif 3467 CURVNET_RESTORE(); 3468 return (error); 3469 } 3470 3471 int 3472 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 3473 { 3474 struct mbuf *m, *m_prev; 3475 int sopt_size = sopt->sopt_valsize; 3476 3477 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3478 if (m == NULL) 3479 return ENOBUFS; 3480 if (sopt_size > MLEN) { 3481 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 3482 if ((m->m_flags & M_EXT) == 0) { 3483 m_free(m); 3484 return ENOBUFS; 3485 } 3486 m->m_len = min(MCLBYTES, sopt_size); 3487 } else { 3488 m->m_len = min(MLEN, sopt_size); 3489 } 3490 sopt_size -= m->m_len; 3491 *mp = m; 3492 m_prev = m; 3493 3494 while (sopt_size) { 3495 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3496 if (m == NULL) { 3497 m_freem(*mp); 3498 return ENOBUFS; 3499 } 3500 if (sopt_size > MLEN) { 3501 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 3502 M_NOWAIT); 3503 if ((m->m_flags & M_EXT) == 0) { 3504 m_freem(m); 3505 m_freem(*mp); 3506 return ENOBUFS; 3507 } 3508 m->m_len = min(MCLBYTES, sopt_size); 3509 } else { 3510 m->m_len = min(MLEN, sopt_size); 3511 } 3512 sopt_size -= m->m_len; 3513 m_prev->m_next = m; 3514 m_prev = m; 3515 } 3516 return (0); 3517 } 3518 3519 int 3520 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 3521 { 3522 struct mbuf *m0 = m; 3523 3524 if (sopt->sopt_val == NULL) 3525 return (0); 3526 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3527 if (sopt->sopt_td != NULL) { 3528 int error; 3529 3530 error = copyin(sopt->sopt_val, mtod(m, char *), 3531 m->m_len); 3532 if (error != 0) { 3533 m_freem(m0); 3534 return(error); 3535 } 3536 } else 3537 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 3538 sopt->sopt_valsize -= m->m_len; 3539 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3540 m = m->m_next; 3541 } 3542 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 3543 panic("ip6_sooptmcopyin"); 3544 return (0); 3545 } 3546 3547 int 3548 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 3549 { 3550 struct mbuf *m0 = m; 3551 size_t valsize = 0; 3552 3553 if (sopt->sopt_val == NULL) 3554 return (0); 3555 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3556 if (sopt->sopt_td != NULL) { 3557 int error; 3558 3559 error = copyout(mtod(m, char *), sopt->sopt_val, 3560 m->m_len); 3561 if (error != 0) { 3562 m_freem(m0); 3563 return(error); 3564 } 3565 } else 3566 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 3567 sopt->sopt_valsize -= m->m_len; 3568 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3569 valsize += m->m_len; 3570 m = m->m_next; 3571 } 3572 if (m != NULL) { 3573 /* enough soopt buffer should be given from user-land */ 3574 m_freem(m0); 3575 return(EINVAL); 3576 } 3577 sopt->sopt_valsize = valsize; 3578 return (0); 3579 } 3580 3581 /* 3582 * sohasoutofband(): protocol notifies socket layer of the arrival of new 3583 * out-of-band data, which will then notify socket consumers. 3584 */ 3585 void 3586 sohasoutofband(struct socket *so) 3587 { 3588 3589 if (so->so_sigio != NULL) 3590 pgsigio(&so->so_sigio, SIGURG, 0); 3591 selwakeuppri(&so->so_rdsel, PSOCK); 3592 } 3593 3594 int 3595 sopoll(struct socket *so, int events, struct ucred *active_cred, 3596 struct thread *td) 3597 { 3598 3599 /* 3600 * We do not need to set or assert curvnet as long as everyone uses 3601 * sopoll_generic(). 3602 */ 3603 return (so->so_proto->pr_sopoll(so, events, active_cred, td)); 3604 } 3605 3606 int 3607 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3608 struct thread *td) 3609 { 3610 int revents; 3611 3612 SOCK_LOCK(so); 3613 if (SOLISTENING(so)) { 3614 if (!(events & (POLLIN | POLLRDNORM))) 3615 revents = 0; 3616 else if (!TAILQ_EMPTY(&so->sol_comp)) 3617 revents = events & (POLLIN | POLLRDNORM); 3618 else if ((events & POLLINIGNEOF) == 0 && so->so_error) 3619 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP; 3620 else { 3621 selrecord(td, &so->so_rdsel); 3622 revents = 0; 3623 } 3624 } else { 3625 revents = 0; 3626 SOCK_SENDBUF_LOCK(so); 3627 SOCK_RECVBUF_LOCK(so); 3628 if (events & (POLLIN | POLLRDNORM)) 3629 if (soreadabledata(so)) 3630 revents |= events & (POLLIN | POLLRDNORM); 3631 if (events & (POLLOUT | POLLWRNORM)) 3632 if (sowriteable(so)) 3633 revents |= events & (POLLOUT | POLLWRNORM); 3634 if (events & (POLLPRI | POLLRDBAND)) 3635 if (so->so_oobmark || 3636 (so->so_rcv.sb_state & SBS_RCVATMARK)) 3637 revents |= events & (POLLPRI | POLLRDBAND); 3638 if ((events & POLLINIGNEOF) == 0) { 3639 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3640 revents |= events & (POLLIN | POLLRDNORM); 3641 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3642 revents |= POLLHUP; 3643 } 3644 } 3645 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3646 revents |= events & POLLRDHUP; 3647 if (revents == 0) { 3648 if (events & 3649 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) { 3650 selrecord(td, &so->so_rdsel); 3651 so->so_rcv.sb_flags |= SB_SEL; 3652 } 3653 if (events & (POLLOUT | POLLWRNORM)) { 3654 selrecord(td, &so->so_wrsel); 3655 so->so_snd.sb_flags |= SB_SEL; 3656 } 3657 } 3658 SOCK_RECVBUF_UNLOCK(so); 3659 SOCK_SENDBUF_UNLOCK(so); 3660 } 3661 SOCK_UNLOCK(so); 3662 return (revents); 3663 } 3664 3665 int 3666 soo_kqfilter(struct file *fp, struct knote *kn) 3667 { 3668 struct socket *so = kn->kn_fp->f_data; 3669 struct sockbuf *sb; 3670 sb_which which; 3671 struct knlist *knl; 3672 3673 switch (kn->kn_filter) { 3674 case EVFILT_READ: 3675 kn->kn_fop = &soread_filtops; 3676 knl = &so->so_rdsel.si_note; 3677 sb = &so->so_rcv; 3678 which = SO_RCV; 3679 break; 3680 case EVFILT_WRITE: 3681 kn->kn_fop = &sowrite_filtops; 3682 knl = &so->so_wrsel.si_note; 3683 sb = &so->so_snd; 3684 which = SO_SND; 3685 break; 3686 case EVFILT_EMPTY: 3687 kn->kn_fop = &soempty_filtops; 3688 knl = &so->so_wrsel.si_note; 3689 sb = &so->so_snd; 3690 which = SO_SND; 3691 break; 3692 default: 3693 return (EINVAL); 3694 } 3695 3696 SOCK_LOCK(so); 3697 if (SOLISTENING(so)) { 3698 knlist_add(knl, kn, 1); 3699 } else { 3700 SOCK_BUF_LOCK(so, which); 3701 knlist_add(knl, kn, 1); 3702 sb->sb_flags |= SB_KNOTE; 3703 SOCK_BUF_UNLOCK(so, which); 3704 } 3705 SOCK_UNLOCK(so); 3706 return (0); 3707 } 3708 3709 static void 3710 filt_sordetach(struct knote *kn) 3711 { 3712 struct socket *so = kn->kn_fp->f_data; 3713 3714 so_rdknl_lock(so); 3715 knlist_remove(&so->so_rdsel.si_note, kn, 1); 3716 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note)) 3717 so->so_rcv.sb_flags &= ~SB_KNOTE; 3718 so_rdknl_unlock(so); 3719 } 3720 3721 /*ARGSUSED*/ 3722 static int 3723 filt_soread(struct knote *kn, long hint) 3724 { 3725 struct socket *so; 3726 3727 so = kn->kn_fp->f_data; 3728 3729 if (SOLISTENING(so)) { 3730 SOCK_LOCK_ASSERT(so); 3731 kn->kn_data = so->sol_qlen; 3732 if (so->so_error) { 3733 kn->kn_flags |= EV_EOF; 3734 kn->kn_fflags = so->so_error; 3735 return (1); 3736 } 3737 return (!TAILQ_EMPTY(&so->sol_comp)); 3738 } 3739 3740 SOCK_RECVBUF_LOCK_ASSERT(so); 3741 3742 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 3743 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3744 kn->kn_flags |= EV_EOF; 3745 kn->kn_fflags = so->so_error; 3746 return (1); 3747 } else if (so->so_error || so->so_rerror) 3748 return (1); 3749 3750 if (kn->kn_sfflags & NOTE_LOWAT) { 3751 if (kn->kn_data >= kn->kn_sdata) 3752 return (1); 3753 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) 3754 return (1); 3755 3756 /* This hook returning non-zero indicates an event, not error */ 3757 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); 3758 } 3759 3760 static void 3761 filt_sowdetach(struct knote *kn) 3762 { 3763 struct socket *so = kn->kn_fp->f_data; 3764 3765 so_wrknl_lock(so); 3766 knlist_remove(&so->so_wrsel.si_note, kn, 1); 3767 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note)) 3768 so->so_snd.sb_flags &= ~SB_KNOTE; 3769 so_wrknl_unlock(so); 3770 } 3771 3772 /*ARGSUSED*/ 3773 static int 3774 filt_sowrite(struct knote *kn, long hint) 3775 { 3776 struct socket *so; 3777 3778 so = kn->kn_fp->f_data; 3779 3780 if (SOLISTENING(so)) 3781 return (0); 3782 3783 SOCK_SENDBUF_LOCK_ASSERT(so); 3784 kn->kn_data = sbspace(&so->so_snd); 3785 3786 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); 3787 3788 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3789 kn->kn_flags |= EV_EOF; 3790 kn->kn_fflags = so->so_error; 3791 return (1); 3792 } else if (so->so_error) /* temporary udp error */ 3793 return (1); 3794 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3795 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3796 return (0); 3797 else if (kn->kn_sfflags & NOTE_LOWAT) 3798 return (kn->kn_data >= kn->kn_sdata); 3799 else 3800 return (kn->kn_data >= so->so_snd.sb_lowat); 3801 } 3802 3803 static int 3804 filt_soempty(struct knote *kn, long hint) 3805 { 3806 struct socket *so; 3807 3808 so = kn->kn_fp->f_data; 3809 3810 if (SOLISTENING(so)) 3811 return (1); 3812 3813 SOCK_SENDBUF_LOCK_ASSERT(so); 3814 kn->kn_data = sbused(&so->so_snd); 3815 3816 if (kn->kn_data == 0) 3817 return (1); 3818 else 3819 return (0); 3820 } 3821 3822 int 3823 socheckuid(struct socket *so, uid_t uid) 3824 { 3825 3826 if (so == NULL) 3827 return (EPERM); 3828 if (so->so_cred->cr_uid != uid) 3829 return (EPERM); 3830 return (0); 3831 } 3832 3833 /* 3834 * These functions are used by protocols to notify the socket layer (and its 3835 * consumers) of state changes in the sockets driven by protocol-side events. 3836 */ 3837 3838 /* 3839 * Procedures to manipulate state flags of socket and do appropriate wakeups. 3840 * 3841 * Normal sequence from the active (originating) side is that 3842 * soisconnecting() is called during processing of connect() call, resulting 3843 * in an eventual call to soisconnected() if/when the connection is 3844 * established. When the connection is torn down soisdisconnecting() is 3845 * called during processing of disconnect() call, and soisdisconnected() is 3846 * called when the connection to the peer is totally severed. The semantics 3847 * of these routines are such that connectionless protocols can call 3848 * soisconnected() and soisdisconnected() only, bypassing the in-progress 3849 * calls when setting up a ``connection'' takes no time. 3850 * 3851 * From the passive side, a socket is created with two queues of sockets: 3852 * so_incomp for connections in progress and so_comp for connections already 3853 * made and awaiting user acceptance. As a protocol is preparing incoming 3854 * connections, it creates a socket structure queued on so_incomp by calling 3855 * sonewconn(). When the connection is established, soisconnected() is 3856 * called, and transfers the socket structure to so_comp, making it available 3857 * to accept(). 3858 * 3859 * If a socket is closed with sockets on either so_incomp or so_comp, these 3860 * sockets are dropped. 3861 * 3862 * If higher-level protocols are implemented in the kernel, the wakeups done 3863 * here will sometimes cause software-interrupt process scheduling. 3864 */ 3865 void 3866 soisconnecting(struct socket *so) 3867 { 3868 3869 SOCK_LOCK(so); 3870 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 3871 so->so_state |= SS_ISCONNECTING; 3872 SOCK_UNLOCK(so); 3873 } 3874 3875 void 3876 soisconnected(struct socket *so) 3877 { 3878 bool last __diagused; 3879 3880 SOCK_LOCK(so); 3881 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 3882 so->so_state |= SS_ISCONNECTED; 3883 3884 if (so->so_qstate == SQ_INCOMP) { 3885 struct socket *head = so->so_listen; 3886 int ret; 3887 3888 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so)); 3889 /* 3890 * Promoting a socket from incomplete queue to complete, we 3891 * need to go through reverse order of locking. We first do 3892 * trylock, and if that doesn't succeed, we go the hard way 3893 * leaving a reference and rechecking consistency after proper 3894 * locking. 3895 */ 3896 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) { 3897 soref(head); 3898 SOCK_UNLOCK(so); 3899 SOLISTEN_LOCK(head); 3900 SOCK_LOCK(so); 3901 if (__predict_false(head != so->so_listen)) { 3902 /* 3903 * The socket went off the listen queue, 3904 * should be lost race to close(2) of sol. 3905 * The socket is about to soabort(). 3906 */ 3907 SOCK_UNLOCK(so); 3908 sorele_locked(head); 3909 return; 3910 } 3911 last = refcount_release(&head->so_count); 3912 KASSERT(!last, ("%s: released last reference for %p", 3913 __func__, head)); 3914 } 3915 again: 3916 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 3917 TAILQ_REMOVE(&head->sol_incomp, so, so_list); 3918 head->sol_incqlen--; 3919 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 3920 head->sol_qlen++; 3921 so->so_qstate = SQ_COMP; 3922 SOCK_UNLOCK(so); 3923 solisten_wakeup(head); /* unlocks */ 3924 } else { 3925 SOCK_RECVBUF_LOCK(so); 3926 soupcall_set(so, SO_RCV, 3927 head->sol_accept_filter->accf_callback, 3928 head->sol_accept_filter_arg); 3929 so->so_options &= ~SO_ACCEPTFILTER; 3930 ret = head->sol_accept_filter->accf_callback(so, 3931 head->sol_accept_filter_arg, M_NOWAIT); 3932 if (ret == SU_ISCONNECTED) { 3933 soupcall_clear(so, SO_RCV); 3934 SOCK_RECVBUF_UNLOCK(so); 3935 goto again; 3936 } 3937 SOCK_RECVBUF_UNLOCK(so); 3938 SOCK_UNLOCK(so); 3939 SOLISTEN_UNLOCK(head); 3940 } 3941 return; 3942 } 3943 SOCK_UNLOCK(so); 3944 wakeup(&so->so_timeo); 3945 sorwakeup(so); 3946 sowwakeup(so); 3947 } 3948 3949 void 3950 soisdisconnecting(struct socket *so) 3951 { 3952 3953 SOCK_LOCK(so); 3954 so->so_state &= ~SS_ISCONNECTING; 3955 so->so_state |= SS_ISDISCONNECTING; 3956 3957 if (!SOLISTENING(so)) { 3958 SOCK_RECVBUF_LOCK(so); 3959 socantrcvmore_locked(so); 3960 SOCK_SENDBUF_LOCK(so); 3961 socantsendmore_locked(so); 3962 } 3963 SOCK_UNLOCK(so); 3964 wakeup(&so->so_timeo); 3965 } 3966 3967 void 3968 soisdisconnected(struct socket *so) 3969 { 3970 3971 SOCK_LOCK(so); 3972 3973 /* 3974 * There is at least one reader of so_state that does not 3975 * acquire socket lock, namely soreceive_generic(). Ensure 3976 * that it never sees all flags that track connection status 3977 * cleared, by ordering the update with a barrier semantic of 3978 * our release thread fence. 3979 */ 3980 so->so_state |= SS_ISDISCONNECTED; 3981 atomic_thread_fence_rel(); 3982 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 3983 3984 if (!SOLISTENING(so)) { 3985 SOCK_UNLOCK(so); 3986 SOCK_RECVBUF_LOCK(so); 3987 socantrcvmore_locked(so); 3988 SOCK_SENDBUF_LOCK(so); 3989 sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); 3990 socantsendmore_locked(so); 3991 } else 3992 SOCK_UNLOCK(so); 3993 wakeup(&so->so_timeo); 3994 } 3995 3996 int 3997 soiolock(struct socket *so, struct sx *sx, int flags) 3998 { 3999 int error; 4000 4001 KASSERT((flags & SBL_VALID) == flags, 4002 ("soiolock: invalid flags %#x", flags)); 4003 4004 if ((flags & SBL_WAIT) != 0) { 4005 if ((flags & SBL_NOINTR) != 0) { 4006 sx_xlock(sx); 4007 } else { 4008 error = sx_xlock_sig(sx); 4009 if (error != 0) 4010 return (error); 4011 } 4012 } else if (!sx_try_xlock(sx)) { 4013 return (EWOULDBLOCK); 4014 } 4015 4016 if (__predict_false(SOLISTENING(so))) { 4017 sx_xunlock(sx); 4018 return (ENOTCONN); 4019 } 4020 return (0); 4021 } 4022 4023 void 4024 soiounlock(struct sx *sx) 4025 { 4026 sx_xunlock(sx); 4027 } 4028 4029 /* 4030 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 4031 */ 4032 struct sockaddr * 4033 sodupsockaddr(const struct sockaddr *sa, int mflags) 4034 { 4035 struct sockaddr *sa2; 4036 4037 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 4038 if (sa2) 4039 bcopy(sa, sa2, sa->sa_len); 4040 return sa2; 4041 } 4042 4043 /* 4044 * Register per-socket destructor. 4045 */ 4046 void 4047 sodtor_set(struct socket *so, so_dtor_t *func) 4048 { 4049 4050 SOCK_LOCK_ASSERT(so); 4051 so->so_dtor = func; 4052 } 4053 4054 /* 4055 * Register per-socket buffer upcalls. 4056 */ 4057 void 4058 soupcall_set(struct socket *so, sb_which which, so_upcall_t func, void *arg) 4059 { 4060 struct sockbuf *sb; 4061 4062 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4063 4064 switch (which) { 4065 case SO_RCV: 4066 sb = &so->so_rcv; 4067 break; 4068 case SO_SND: 4069 sb = &so->so_snd; 4070 break; 4071 } 4072 SOCK_BUF_LOCK_ASSERT(so, which); 4073 sb->sb_upcall = func; 4074 sb->sb_upcallarg = arg; 4075 sb->sb_flags |= SB_UPCALL; 4076 } 4077 4078 void 4079 soupcall_clear(struct socket *so, sb_which which) 4080 { 4081 struct sockbuf *sb; 4082 4083 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4084 4085 switch (which) { 4086 case SO_RCV: 4087 sb = &so->so_rcv; 4088 break; 4089 case SO_SND: 4090 sb = &so->so_snd; 4091 break; 4092 } 4093 SOCK_BUF_LOCK_ASSERT(so, which); 4094 KASSERT(sb->sb_upcall != NULL, 4095 ("%s: so %p no upcall to clear", __func__, so)); 4096 sb->sb_upcall = NULL; 4097 sb->sb_upcallarg = NULL; 4098 sb->sb_flags &= ~SB_UPCALL; 4099 } 4100 4101 void 4102 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg) 4103 { 4104 4105 SOLISTEN_LOCK_ASSERT(so); 4106 so->sol_upcall = func; 4107 so->sol_upcallarg = arg; 4108 } 4109 4110 static void 4111 so_rdknl_lock(void *arg) 4112 { 4113 struct socket *so = arg; 4114 4115 retry: 4116 if (SOLISTENING(so)) { 4117 SOLISTEN_LOCK(so); 4118 } else { 4119 SOCK_RECVBUF_LOCK(so); 4120 if (__predict_false(SOLISTENING(so))) { 4121 SOCK_RECVBUF_UNLOCK(so); 4122 goto retry; 4123 } 4124 } 4125 } 4126 4127 static void 4128 so_rdknl_unlock(void *arg) 4129 { 4130 struct socket *so = arg; 4131 4132 if (SOLISTENING(so)) 4133 SOLISTEN_UNLOCK(so); 4134 else 4135 SOCK_RECVBUF_UNLOCK(so); 4136 } 4137 4138 static void 4139 so_rdknl_assert_lock(void *arg, int what) 4140 { 4141 struct socket *so = arg; 4142 4143 if (what == LA_LOCKED) { 4144 if (SOLISTENING(so)) 4145 SOLISTEN_LOCK_ASSERT(so); 4146 else 4147 SOCK_RECVBUF_LOCK_ASSERT(so); 4148 } else { 4149 if (SOLISTENING(so)) 4150 SOLISTEN_UNLOCK_ASSERT(so); 4151 else 4152 SOCK_RECVBUF_UNLOCK_ASSERT(so); 4153 } 4154 } 4155 4156 static void 4157 so_wrknl_lock(void *arg) 4158 { 4159 struct socket *so = arg; 4160 4161 retry: 4162 if (SOLISTENING(so)) { 4163 SOLISTEN_LOCK(so); 4164 } else { 4165 SOCK_SENDBUF_LOCK(so); 4166 if (__predict_false(SOLISTENING(so))) { 4167 SOCK_SENDBUF_UNLOCK(so); 4168 goto retry; 4169 } 4170 } 4171 } 4172 4173 static void 4174 so_wrknl_unlock(void *arg) 4175 { 4176 struct socket *so = arg; 4177 4178 if (SOLISTENING(so)) 4179 SOLISTEN_UNLOCK(so); 4180 else 4181 SOCK_SENDBUF_UNLOCK(so); 4182 } 4183 4184 static void 4185 so_wrknl_assert_lock(void *arg, int what) 4186 { 4187 struct socket *so = arg; 4188 4189 if (what == LA_LOCKED) { 4190 if (SOLISTENING(so)) 4191 SOLISTEN_LOCK_ASSERT(so); 4192 else 4193 SOCK_SENDBUF_LOCK_ASSERT(so); 4194 } else { 4195 if (SOLISTENING(so)) 4196 SOLISTEN_UNLOCK_ASSERT(so); 4197 else 4198 SOCK_SENDBUF_UNLOCK_ASSERT(so); 4199 } 4200 } 4201 4202 /* 4203 * Create an external-format (``xsocket'') structure using the information in 4204 * the kernel-format socket structure pointed to by so. This is done to 4205 * reduce the spew of irrelevant information over this interface, to isolate 4206 * user code from changes in the kernel structure, and potentially to provide 4207 * information-hiding if we decide that some of this information should be 4208 * hidden from users. 4209 */ 4210 void 4211 sotoxsocket(struct socket *so, struct xsocket *xso) 4212 { 4213 4214 bzero(xso, sizeof(*xso)); 4215 xso->xso_len = sizeof *xso; 4216 xso->xso_so = (uintptr_t)so; 4217 xso->so_type = so->so_type; 4218 xso->so_options = so->so_options; 4219 xso->so_linger = so->so_linger; 4220 xso->so_state = so->so_state; 4221 xso->so_pcb = (uintptr_t)so->so_pcb; 4222 xso->xso_protocol = so->so_proto->pr_protocol; 4223 xso->xso_family = so->so_proto->pr_domain->dom_family; 4224 xso->so_timeo = so->so_timeo; 4225 xso->so_error = so->so_error; 4226 xso->so_uid = so->so_cred->cr_uid; 4227 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 4228 if (SOLISTENING(so)) { 4229 xso->so_qlen = so->sol_qlen; 4230 xso->so_incqlen = so->sol_incqlen; 4231 xso->so_qlimit = so->sol_qlimit; 4232 xso->so_oobmark = 0; 4233 } else { 4234 xso->so_state |= so->so_qstate; 4235 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0; 4236 xso->so_oobmark = so->so_oobmark; 4237 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 4238 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 4239 } 4240 } 4241 4242 struct sockbuf * 4243 so_sockbuf_rcv(struct socket *so) 4244 { 4245 4246 return (&so->so_rcv); 4247 } 4248 4249 struct sockbuf * 4250 so_sockbuf_snd(struct socket *so) 4251 { 4252 4253 return (&so->so_snd); 4254 } 4255 4256 int 4257 so_state_get(const struct socket *so) 4258 { 4259 4260 return (so->so_state); 4261 } 4262 4263 void 4264 so_state_set(struct socket *so, int val) 4265 { 4266 4267 so->so_state = val; 4268 } 4269 4270 int 4271 so_options_get(const struct socket *so) 4272 { 4273 4274 return (so->so_options); 4275 } 4276 4277 void 4278 so_options_set(struct socket *so, int val) 4279 { 4280 4281 so->so_options = val; 4282 } 4283 4284 int 4285 so_error_get(const struct socket *so) 4286 { 4287 4288 return (so->so_error); 4289 } 4290 4291 void 4292 so_error_set(struct socket *so, int val) 4293 { 4294 4295 so->so_error = val; 4296 } 4297 4298 int 4299 so_linger_get(const struct socket *so) 4300 { 4301 4302 return (so->so_linger); 4303 } 4304 4305 void 4306 so_linger_set(struct socket *so, int val) 4307 { 4308 4309 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz), 4310 ("%s: val %d out of range", __func__, val)); 4311 4312 so->so_linger = val; 4313 } 4314 4315 struct protosw * 4316 so_protosw_get(const struct socket *so) 4317 { 4318 4319 return (so->so_proto); 4320 } 4321 4322 void 4323 so_protosw_set(struct socket *so, struct protosw *val) 4324 { 4325 4326 so->so_proto = val; 4327 } 4328 4329 void 4330 so_sorwakeup(struct socket *so) 4331 { 4332 4333 sorwakeup(so); 4334 } 4335 4336 void 4337 so_sowwakeup(struct socket *so) 4338 { 4339 4340 sowwakeup(so); 4341 } 4342 4343 void 4344 so_sorwakeup_locked(struct socket *so) 4345 { 4346 4347 sorwakeup_locked(so); 4348 } 4349 4350 void 4351 so_sowwakeup_locked(struct socket *so) 4352 { 4353 4354 sowwakeup_locked(so); 4355 } 4356 4357 void 4358 so_lock(struct socket *so) 4359 { 4360 4361 SOCK_LOCK(so); 4362 } 4363 4364 void 4365 so_unlock(struct socket *so) 4366 { 4367 4368 SOCK_UNLOCK(so); 4369 } 4370