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