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_NEEDGIANT, 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_NEEDGIANT, 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, UMA_ZONE_NOFREE); 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) { 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_NEEDGIANT, &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 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd"); 422 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv"); 423 so->so_rcv.sb_sel = &so->so_rdsel; 424 so->so_snd.sb_sel = &so->so_wrsel; 425 sx_init(&so->so_snd.sb_sx, "so_snd_sx"); 426 sx_init(&so->so_rcv.sb_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 static 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.sb_sx); 491 sx_destroy(&so->so_rcv.sb_sx); 492 SOCKBUF_LOCK_DESTROY(&so->so_snd); 493 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 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. The socket should be 502 * 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 /* 559 * Auto-sizing of socket buffers is managed by the protocols and 560 * the appropriate flags must be set in the pru_attach function. 561 */ 562 CURVNET_SET(so->so_vnet); 563 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); 564 CURVNET_RESTORE(); 565 if (error) { 566 sodealloc(so); 567 return (error); 568 } 569 soref(so); 570 *aso = so; 571 return (0); 572 } 573 574 #ifdef REGRESSION 575 static int regression_sonewconn_earlytest = 1; 576 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW, 577 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test"); 578 #endif 579 580 static struct timeval overinterval = { 60, 0 }; 581 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW, 582 &overinterval, 583 "Delay in seconds between warnings for listen socket overflows"); 584 585 /* 586 * When an attempt at a new connection is noted on a socket which accepts 587 * connections, sonewconn is called. If the connection is possible (subject 588 * to space constraints, etc.) then we allocate a new structure, properly 589 * linked into the data structure of the original socket, and return this. 590 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED. 591 * 592 * Note: the ref count on the socket is 0 on return. 593 */ 594 struct socket * 595 sonewconn(struct socket *head, int connstatus) 596 { 597 struct sbuf descrsb; 598 struct socket *so; 599 int len, overcount; 600 u_int qlen; 601 const char localprefix[] = "local:"; 602 char descrbuf[SUNPATHLEN + sizeof(localprefix)]; 603 #if defined(INET6) 604 char addrbuf[INET6_ADDRSTRLEN]; 605 #elif defined(INET) 606 char addrbuf[INET_ADDRSTRLEN]; 607 #endif 608 bool dolog, over; 609 610 SOLISTEN_LOCK(head); 611 over = (head->sol_qlen > 3 * head->sol_qlimit / 2); 612 #ifdef REGRESSION 613 if (regression_sonewconn_earlytest && over) { 614 #else 615 if (over) { 616 #endif 617 head->sol_overcount++; 618 dolog = !!ratecheck(&head->sol_lastover, &overinterval); 619 620 /* 621 * If we're going to log, copy the overflow count and queue 622 * length from the listen socket before dropping the lock. 623 * Also, reset the overflow count. 624 */ 625 if (dolog) { 626 overcount = head->sol_overcount; 627 head->sol_overcount = 0; 628 qlen = head->sol_qlen; 629 } 630 SOLISTEN_UNLOCK(head); 631 632 if (dolog) { 633 /* 634 * Try to print something descriptive about the 635 * socket for the error message. 636 */ 637 sbuf_new(&descrsb, descrbuf, sizeof(descrbuf), 638 SBUF_FIXEDLEN); 639 switch (head->so_proto->pr_domain->dom_family) { 640 #if defined(INET) || defined(INET6) 641 #ifdef INET 642 case AF_INET: 643 #endif 644 #ifdef INET6 645 case AF_INET6: 646 if (head->so_proto->pr_domain->dom_family == 647 AF_INET6 || 648 (sotoinpcb(head)->inp_inc.inc_flags & 649 INC_ISIPV6)) { 650 ip6_sprintf(addrbuf, 651 &sotoinpcb(head)->inp_inc.inc6_laddr); 652 sbuf_printf(&descrsb, "[%s]", addrbuf); 653 } else 654 #endif 655 { 656 #ifdef INET 657 inet_ntoa_r( 658 sotoinpcb(head)->inp_inc.inc_laddr, 659 addrbuf); 660 sbuf_cat(&descrsb, addrbuf); 661 #endif 662 } 663 sbuf_printf(&descrsb, ":%hu (proto %u)", 664 ntohs(sotoinpcb(head)->inp_inc.inc_lport), 665 head->so_proto->pr_protocol); 666 break; 667 #endif /* INET || INET6 */ 668 case AF_UNIX: 669 sbuf_cat(&descrsb, localprefix); 670 if (sotounpcb(head)->unp_addr != NULL) 671 len = 672 sotounpcb(head)->unp_addr->sun_len - 673 offsetof(struct sockaddr_un, 674 sun_path); 675 else 676 len = 0; 677 if (len > 0) 678 sbuf_bcat(&descrsb, 679 sotounpcb(head)->unp_addr->sun_path, 680 len); 681 else 682 sbuf_cat(&descrsb, "(unknown)"); 683 break; 684 } 685 686 /* 687 * If we can't print something more specific, at least 688 * print the domain name. 689 */ 690 if (sbuf_finish(&descrsb) != 0 || 691 sbuf_len(&descrsb) <= 0) { 692 sbuf_clear(&descrsb); 693 sbuf_cat(&descrsb, 694 head->so_proto->pr_domain->dom_name ?: 695 "unknown"); 696 sbuf_finish(&descrsb); 697 } 698 KASSERT(sbuf_len(&descrsb) > 0, 699 ("%s: sbuf creation failed", __func__)); 700 log(LOG_DEBUG, 701 "%s: pcb %p (%s): Listen queue overflow: " 702 "%i already in queue awaiting acceptance " 703 "(%d occurrences)\n", 704 __func__, head->so_pcb, sbuf_data(&descrsb), 705 qlen, overcount); 706 sbuf_delete(&descrsb); 707 708 overcount = 0; 709 } 710 711 return (NULL); 712 } 713 SOLISTEN_UNLOCK(head); 714 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL", 715 __func__, head)); 716 so = soalloc(head->so_vnet); 717 if (so == NULL) { 718 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 719 "limit reached or out of memory\n", 720 __func__, head->so_pcb); 721 return (NULL); 722 } 723 so->so_listen = head; 724 so->so_type = head->so_type; 725 so->so_options = head->so_options & ~SO_ACCEPTCONN; 726 so->so_linger = head->so_linger; 727 so->so_state = head->so_state | SS_NOFDREF; 728 so->so_fibnum = head->so_fibnum; 729 so->so_proto = head->so_proto; 730 so->so_cred = crhold(head->so_cred); 731 #ifdef MAC 732 mac_socket_newconn(head, so); 733 #endif 734 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 735 so_rdknl_assert_lock); 736 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 737 so_wrknl_assert_lock); 738 VNET_SO_ASSERT(head); 739 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) { 740 sodealloc(so); 741 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 742 __func__, head->so_pcb); 743 return (NULL); 744 } 745 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 746 sodealloc(so); 747 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 748 __func__, head->so_pcb); 749 return (NULL); 750 } 751 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat; 752 so->so_snd.sb_lowat = head->sol_sbsnd_lowat; 753 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo; 754 so->so_snd.sb_timeo = head->sol_sbsnd_timeo; 755 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE; 756 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE; 757 758 SOLISTEN_LOCK(head); 759 if (head->sol_accept_filter != NULL) 760 connstatus = 0; 761 so->so_state |= connstatus; 762 soref(head); /* A socket on (in)complete queue refs head. */ 763 if (connstatus) { 764 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 765 so->so_qstate = SQ_COMP; 766 head->sol_qlen++; 767 solisten_wakeup(head); /* unlocks */ 768 } else { 769 /* 770 * Keep removing sockets from the head until there's room for 771 * us to insert on the tail. In pre-locking revisions, this 772 * was a simple if(), but as we could be racing with other 773 * threads and soabort() requires dropping locks, we must 774 * loop waiting for the condition to be true. 775 */ 776 while (head->sol_incqlen > head->sol_qlimit) { 777 struct socket *sp; 778 779 sp = TAILQ_FIRST(&head->sol_incomp); 780 TAILQ_REMOVE(&head->sol_incomp, sp, so_list); 781 head->sol_incqlen--; 782 SOCK_LOCK(sp); 783 sp->so_qstate = SQ_NONE; 784 sp->so_listen = NULL; 785 SOCK_UNLOCK(sp); 786 sorele(head); /* does SOLISTEN_UNLOCK, head stays */ 787 soabort(sp); 788 SOLISTEN_LOCK(head); 789 } 790 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list); 791 so->so_qstate = SQ_INCOMP; 792 head->sol_incqlen++; 793 SOLISTEN_UNLOCK(head); 794 } 795 return (so); 796 } 797 798 #if defined(SCTP) || defined(SCTP_SUPPORT) 799 /* 800 * Socket part of sctp_peeloff(). Detach a new socket from an 801 * association. The new socket is returned with a reference. 802 */ 803 struct socket * 804 sopeeloff(struct socket *head) 805 { 806 struct socket *so; 807 808 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p", 809 __func__, __LINE__, head)); 810 so = soalloc(head->so_vnet); 811 if (so == NULL) { 812 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 813 "limit reached or out of memory\n", 814 __func__, head->so_pcb); 815 return (NULL); 816 } 817 so->so_type = head->so_type; 818 so->so_options = head->so_options; 819 so->so_linger = head->so_linger; 820 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED; 821 so->so_fibnum = head->so_fibnum; 822 so->so_proto = head->so_proto; 823 so->so_cred = crhold(head->so_cred); 824 #ifdef MAC 825 mac_socket_newconn(head, so); 826 #endif 827 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 828 so_rdknl_assert_lock); 829 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 830 so_wrknl_assert_lock); 831 VNET_SO_ASSERT(head); 832 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { 833 sodealloc(so); 834 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 835 __func__, head->so_pcb); 836 return (NULL); 837 } 838 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 839 sodealloc(so); 840 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 841 __func__, head->so_pcb); 842 return (NULL); 843 } 844 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 845 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 846 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 847 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 848 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; 849 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; 850 851 soref(so); 852 853 return (so); 854 } 855 #endif /* SCTP */ 856 857 int 858 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 859 { 860 int error; 861 862 CURVNET_SET(so->so_vnet); 863 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td); 864 CURVNET_RESTORE(); 865 return (error); 866 } 867 868 int 869 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 870 { 871 int error; 872 873 CURVNET_SET(so->so_vnet); 874 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td); 875 CURVNET_RESTORE(); 876 return (error); 877 } 878 879 /* 880 * solisten() transitions a socket from a non-listening state to a listening 881 * state, but can also be used to update the listen queue depth on an 882 * existing listen socket. The protocol will call back into the sockets 883 * layer using solisten_proto_check() and solisten_proto() to check and set 884 * socket-layer listen state. Call backs are used so that the protocol can 885 * acquire both protocol and socket layer locks in whatever order is required 886 * by the protocol. 887 * 888 * Protocol implementors are advised to hold the socket lock across the 889 * socket-layer test and set to avoid races at the socket layer. 890 */ 891 int 892 solisten(struct socket *so, int backlog, struct thread *td) 893 { 894 int error; 895 896 CURVNET_SET(so->so_vnet); 897 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td); 898 CURVNET_RESTORE(); 899 return (error); 900 } 901 902 int 903 solisten_proto_check(struct socket *so) 904 { 905 906 SOCK_LOCK_ASSERT(so); 907 908 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 909 SS_ISDISCONNECTING)) 910 return (EINVAL); 911 return (0); 912 } 913 914 void 915 solisten_proto(struct socket *so, int backlog) 916 { 917 int sbrcv_lowat, sbsnd_lowat; 918 u_int sbrcv_hiwat, sbsnd_hiwat; 919 short sbrcv_flags, sbsnd_flags; 920 sbintime_t sbrcv_timeo, sbsnd_timeo; 921 922 SOCK_LOCK_ASSERT(so); 923 924 if (SOLISTENING(so)) 925 goto listening; 926 927 /* 928 * Change this socket to listening state. 929 */ 930 sbrcv_lowat = so->so_rcv.sb_lowat; 931 sbsnd_lowat = so->so_snd.sb_lowat; 932 sbrcv_hiwat = so->so_rcv.sb_hiwat; 933 sbsnd_hiwat = so->so_snd.sb_hiwat; 934 sbrcv_flags = so->so_rcv.sb_flags; 935 sbsnd_flags = so->so_snd.sb_flags; 936 sbrcv_timeo = so->so_rcv.sb_timeo; 937 sbsnd_timeo = so->so_snd.sb_timeo; 938 939 sbdestroy(&so->so_snd, so); 940 sbdestroy(&so->so_rcv, so); 941 sx_destroy(&so->so_snd.sb_sx); 942 sx_destroy(&so->so_rcv.sb_sx); 943 SOCKBUF_LOCK_DESTROY(&so->so_snd); 944 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 945 946 #ifdef INVARIANTS 947 bzero(&so->so_rcv, 948 sizeof(struct socket) - offsetof(struct socket, so_rcv)); 949 #endif 950 951 so->sol_sbrcv_lowat = sbrcv_lowat; 952 so->sol_sbsnd_lowat = sbsnd_lowat; 953 so->sol_sbrcv_hiwat = sbrcv_hiwat; 954 so->sol_sbsnd_hiwat = sbsnd_hiwat; 955 so->sol_sbrcv_flags = sbrcv_flags; 956 so->sol_sbsnd_flags = sbsnd_flags; 957 so->sol_sbrcv_timeo = sbrcv_timeo; 958 so->sol_sbsnd_timeo = sbsnd_timeo; 959 960 so->sol_qlen = so->sol_incqlen = 0; 961 TAILQ_INIT(&so->sol_incomp); 962 TAILQ_INIT(&so->sol_comp); 963 964 so->sol_accept_filter = NULL; 965 so->sol_accept_filter_arg = NULL; 966 so->sol_accept_filter_str = NULL; 967 968 so->sol_upcall = NULL; 969 so->sol_upcallarg = NULL; 970 971 so->so_options |= SO_ACCEPTCONN; 972 973 listening: 974 if (backlog < 0 || backlog > somaxconn) 975 backlog = somaxconn; 976 so->sol_qlimit = backlog; 977 } 978 979 /* 980 * Wakeup listeners/subsystems once we have a complete connection. 981 * Enters with lock, returns unlocked. 982 */ 983 void 984 solisten_wakeup(struct socket *sol) 985 { 986 987 if (sol->sol_upcall != NULL) 988 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT); 989 else { 990 selwakeuppri(&sol->so_rdsel, PSOCK); 991 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0); 992 } 993 SOLISTEN_UNLOCK(sol); 994 wakeup_one(&sol->sol_comp); 995 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL) 996 pgsigio(&sol->so_sigio, SIGIO, 0); 997 } 998 999 /* 1000 * Return single connection off a listening socket queue. Main consumer of 1001 * the function is kern_accept4(). Some modules, that do their own accept 1002 * management also use the function. 1003 * 1004 * Listening socket must be locked on entry and is returned unlocked on 1005 * return. 1006 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT. 1007 */ 1008 int 1009 solisten_dequeue(struct socket *head, struct socket **ret, int flags) 1010 { 1011 struct socket *so; 1012 int error; 1013 1014 SOLISTEN_LOCK_ASSERT(head); 1015 1016 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) && 1017 head->so_error == 0) { 1018 error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH, 1019 "accept", 0); 1020 if (error != 0) { 1021 SOLISTEN_UNLOCK(head); 1022 return (error); 1023 } 1024 } 1025 if (head->so_error) { 1026 error = head->so_error; 1027 head->so_error = 0; 1028 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp)) 1029 error = EWOULDBLOCK; 1030 else 1031 error = 0; 1032 if (error) { 1033 SOLISTEN_UNLOCK(head); 1034 return (error); 1035 } 1036 so = TAILQ_FIRST(&head->sol_comp); 1037 SOCK_LOCK(so); 1038 KASSERT(so->so_qstate == SQ_COMP, 1039 ("%s: so %p not SQ_COMP", __func__, so)); 1040 soref(so); 1041 head->sol_qlen--; 1042 so->so_qstate = SQ_NONE; 1043 so->so_listen = NULL; 1044 TAILQ_REMOVE(&head->sol_comp, so, so_list); 1045 if (flags & ACCEPT4_INHERIT) 1046 so->so_state |= (head->so_state & SS_NBIO); 1047 else 1048 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0; 1049 SOCK_UNLOCK(so); 1050 sorele(head); 1051 1052 *ret = so; 1053 return (0); 1054 } 1055 1056 /* 1057 * Evaluate the reference count and named references on a socket; if no 1058 * references remain, free it. This should be called whenever a reference is 1059 * released, such as in sorele(), but also when named reference flags are 1060 * cleared in socket or protocol code. 1061 * 1062 * sofree() will free the socket if: 1063 * 1064 * - There are no outstanding file descriptor references or related consumers 1065 * (so_count == 0). 1066 * 1067 * - The socket has been closed by user space, if ever open (SS_NOFDREF). 1068 * 1069 * - The protocol does not have an outstanding strong reference on the socket 1070 * (SS_PROTOREF). 1071 * 1072 * - The socket is not in a completed connection queue, so a process has been 1073 * notified that it is present. If it is removed, the user process may 1074 * block in accept() despite select() saying the socket was ready. 1075 */ 1076 void 1077 sofree(struct socket *so) 1078 { 1079 struct protosw *pr = so->so_proto; 1080 1081 SOCK_LOCK_ASSERT(so); 1082 1083 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 || 1084 (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) { 1085 SOCK_UNLOCK(so); 1086 return; 1087 } 1088 1089 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) { 1090 struct socket *sol; 1091 1092 sol = so->so_listen; 1093 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so)); 1094 1095 /* 1096 * To solve race between close of a listening socket and 1097 * a socket on its incomplete queue, we need to lock both. 1098 * The order is first listening socket, then regular. 1099 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this 1100 * function and the listening socket are the only pointers 1101 * to so. To preserve so and sol, we reference both and then 1102 * relock. 1103 * After relock the socket may not move to so_comp since it 1104 * doesn't have PCB already, but it may be removed from 1105 * so_incomp. If that happens, we share responsiblity on 1106 * freeing the socket, but soclose() has already removed 1107 * it from queue. 1108 */ 1109 soref(sol); 1110 soref(so); 1111 SOCK_UNLOCK(so); 1112 SOLISTEN_LOCK(sol); 1113 SOCK_LOCK(so); 1114 if (so->so_qstate == SQ_INCOMP) { 1115 KASSERT(so->so_listen == sol, 1116 ("%s: so %p migrated out of sol %p", 1117 __func__, so, sol)); 1118 TAILQ_REMOVE(&sol->sol_incomp, so, so_list); 1119 sol->sol_incqlen--; 1120 /* This is guarenteed not to be the last. */ 1121 refcount_release(&sol->so_count); 1122 so->so_qstate = SQ_NONE; 1123 so->so_listen = NULL; 1124 } else 1125 KASSERT(so->so_listen == NULL, 1126 ("%s: so %p not on (in)comp with so_listen", 1127 __func__, so)); 1128 sorele(sol); 1129 KASSERT(so->so_count == 1, 1130 ("%s: so %p count %u", __func__, so, so->so_count)); 1131 so->so_count = 0; 1132 } 1133 if (SOLISTENING(so)) 1134 so->so_error = ECONNABORTED; 1135 SOCK_UNLOCK(so); 1136 1137 if (so->so_dtor != NULL) 1138 so->so_dtor(so); 1139 1140 VNET_SO_ASSERT(so); 1141 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 1142 (*pr->pr_domain->dom_dispose)(so); 1143 if (pr->pr_usrreqs->pru_detach != NULL) 1144 (*pr->pr_usrreqs->pru_detach)(so); 1145 1146 /* 1147 * From this point on, we assume that no other references to this 1148 * socket exist anywhere else in the stack. Therefore, no locks need 1149 * to be acquired or held. 1150 * 1151 * We used to do a lot of socket buffer and socket locking here, as 1152 * well as invoke sorflush() and perform wakeups. The direct call to 1153 * dom_dispose() and sbdestroy() are an inlining of what was 1154 * necessary from sorflush(). 1155 * 1156 * Notice that the socket buffer and kqueue state are torn down 1157 * before calling pru_detach. This means that protocols shold not 1158 * assume they can perform socket wakeups, etc, in their detach code. 1159 */ 1160 if (!SOLISTENING(so)) { 1161 sbdestroy(&so->so_snd, so); 1162 sbdestroy(&so->so_rcv, so); 1163 } 1164 seldrain(&so->so_rdsel); 1165 seldrain(&so->so_wrsel); 1166 knlist_destroy(&so->so_rdsel.si_note); 1167 knlist_destroy(&so->so_wrsel.si_note); 1168 sodealloc(so); 1169 } 1170 1171 /* 1172 * Close a socket on last file table reference removal. Initiate disconnect 1173 * if connected. Free socket when disconnect complete. 1174 * 1175 * This function will sorele() the socket. Note that soclose() may be called 1176 * prior to the ref count reaching zero. The actual socket structure will 1177 * not be freed until the ref count reaches zero. 1178 */ 1179 int 1180 soclose(struct socket *so) 1181 { 1182 struct accept_queue lqueue; 1183 int error = 0; 1184 1185 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter")); 1186 1187 CURVNET_SET(so->so_vnet); 1188 funsetown(&so->so_sigio); 1189 if (so->so_state & SS_ISCONNECTED) { 1190 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 1191 error = sodisconnect(so); 1192 if (error) { 1193 if (error == ENOTCONN) 1194 error = 0; 1195 goto drop; 1196 } 1197 } 1198 1199 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) { 1200 if ((so->so_state & SS_ISDISCONNECTING) && 1201 (so->so_state & SS_NBIO)) 1202 goto drop; 1203 while (so->so_state & SS_ISCONNECTED) { 1204 error = tsleep(&so->so_timeo, 1205 PSOCK | PCATCH, "soclos", 1206 so->so_linger * hz); 1207 if (error) 1208 break; 1209 } 1210 } 1211 } 1212 1213 drop: 1214 if (so->so_proto->pr_usrreqs->pru_close != NULL) 1215 (*so->so_proto->pr_usrreqs->pru_close)(so); 1216 1217 SOCK_LOCK(so); 1218 if (SOLISTENING(so)) { 1219 struct socket *sp; 1220 1221 TAILQ_INIT(&lqueue); 1222 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list); 1223 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list); 1224 1225 so->sol_qlen = so->sol_incqlen = 0; 1226 1227 TAILQ_FOREACH(sp, &lqueue, so_list) { 1228 SOCK_LOCK(sp); 1229 sp->so_qstate = SQ_NONE; 1230 sp->so_listen = NULL; 1231 SOCK_UNLOCK(sp); 1232 /* Guaranteed not to be the last. */ 1233 refcount_release(&so->so_count); 1234 } 1235 } 1236 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF")); 1237 so->so_state |= SS_NOFDREF; 1238 sorele(so); 1239 if (SOLISTENING(so)) { 1240 struct socket *sp, *tsp; 1241 1242 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) { 1243 SOCK_LOCK(sp); 1244 if (sp->so_count == 0) { 1245 SOCK_UNLOCK(sp); 1246 soabort(sp); 1247 } else 1248 /* sp is now in sofree() */ 1249 SOCK_UNLOCK(sp); 1250 } 1251 } 1252 CURVNET_RESTORE(); 1253 return (error); 1254 } 1255 1256 /* 1257 * soabort() is used to abruptly tear down a connection, such as when a 1258 * resource limit is reached (listen queue depth exceeded), or if a listen 1259 * socket is closed while there are sockets waiting to be accepted. 1260 * 1261 * This interface is tricky, because it is called on an unreferenced socket, 1262 * and must be called only by a thread that has actually removed the socket 1263 * from the listen queue it was on, or races with other threads are risked. 1264 * 1265 * This interface will call into the protocol code, so must not be called 1266 * with any socket locks held. Protocols do call it while holding their own 1267 * recursible protocol mutexes, but this is something that should be subject 1268 * to review in the future. 1269 */ 1270 void 1271 soabort(struct socket *so) 1272 { 1273 1274 /* 1275 * In as much as is possible, assert that no references to this 1276 * socket are held. This is not quite the same as asserting that the 1277 * current thread is responsible for arranging for no references, but 1278 * is as close as we can get for now. 1279 */ 1280 KASSERT(so->so_count == 0, ("soabort: so_count")); 1281 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF")); 1282 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF")); 1283 VNET_SO_ASSERT(so); 1284 1285 if (so->so_proto->pr_usrreqs->pru_abort != NULL) 1286 (*so->so_proto->pr_usrreqs->pru_abort)(so); 1287 SOCK_LOCK(so); 1288 sofree(so); 1289 } 1290 1291 int 1292 soaccept(struct socket *so, struct sockaddr **nam) 1293 { 1294 int error; 1295 1296 SOCK_LOCK(so); 1297 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF")); 1298 so->so_state &= ~SS_NOFDREF; 1299 SOCK_UNLOCK(so); 1300 1301 CURVNET_SET(so->so_vnet); 1302 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); 1303 CURVNET_RESTORE(); 1304 return (error); 1305 } 1306 1307 int 1308 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 1309 { 1310 1311 return (soconnectat(AT_FDCWD, so, nam, td)); 1312 } 1313 1314 int 1315 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 1316 { 1317 int error; 1318 1319 /* XXXMJ racy */ 1320 if (SOLISTENING(so)) 1321 return (EOPNOTSUPP); 1322 1323 CURVNET_SET(so->so_vnet); 1324 /* 1325 * If protocol is connection-based, can only connect once. 1326 * Otherwise, if connected, try to disconnect first. This allows 1327 * user to disconnect by connecting to, e.g., a null address. 1328 */ 1329 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 1330 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 1331 (error = sodisconnect(so)))) { 1332 error = EISCONN; 1333 } else { 1334 /* 1335 * Prevent accumulated error from previous connection from 1336 * biting us. 1337 */ 1338 so->so_error = 0; 1339 if (fd == AT_FDCWD) { 1340 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, 1341 nam, td); 1342 } else { 1343 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd, 1344 so, nam, td); 1345 } 1346 } 1347 CURVNET_RESTORE(); 1348 1349 return (error); 1350 } 1351 1352 int 1353 soconnect2(struct socket *so1, struct socket *so2) 1354 { 1355 int error; 1356 1357 CURVNET_SET(so1->so_vnet); 1358 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); 1359 CURVNET_RESTORE(); 1360 return (error); 1361 } 1362 1363 int 1364 sodisconnect(struct socket *so) 1365 { 1366 int error; 1367 1368 if ((so->so_state & SS_ISCONNECTED) == 0) 1369 return (ENOTCONN); 1370 if (so->so_state & SS_ISDISCONNECTING) 1371 return (EALREADY); 1372 VNET_SO_ASSERT(so); 1373 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 1374 return (error); 1375 } 1376 1377 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) 1378 1379 int 1380 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1381 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1382 { 1383 long space; 1384 ssize_t resid; 1385 int clen = 0, error, dontroute; 1386 1387 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 1388 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 1389 ("sosend_dgram: !PR_ATOMIC")); 1390 1391 if (uio != NULL) 1392 resid = uio->uio_resid; 1393 else 1394 resid = top->m_pkthdr.len; 1395 /* 1396 * In theory resid should be unsigned. However, space must be 1397 * signed, as it might be less than 0 if we over-committed, and we 1398 * must use a signed comparison of space and resid. On the other 1399 * hand, a negative resid causes us to loop sending 0-length 1400 * segments to the protocol. 1401 */ 1402 if (resid < 0) { 1403 error = EINVAL; 1404 goto out; 1405 } 1406 1407 dontroute = 1408 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 1409 if (td != NULL) 1410 td->td_ru.ru_msgsnd++; 1411 if (control != NULL) 1412 clen = control->m_len; 1413 1414 SOCKBUF_LOCK(&so->so_snd); 1415 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1416 SOCKBUF_UNLOCK(&so->so_snd); 1417 error = EPIPE; 1418 goto out; 1419 } 1420 if (so->so_error) { 1421 error = so->so_error; 1422 so->so_error = 0; 1423 SOCKBUF_UNLOCK(&so->so_snd); 1424 goto out; 1425 } 1426 if ((so->so_state & SS_ISCONNECTED) == 0) { 1427 /* 1428 * `sendto' and `sendmsg' is allowed on a connection-based 1429 * socket if it supports implied connect. Return ENOTCONN if 1430 * not connected and no address is supplied. 1431 */ 1432 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1433 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1434 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1435 !(resid == 0 && clen != 0)) { 1436 SOCKBUF_UNLOCK(&so->so_snd); 1437 error = ENOTCONN; 1438 goto out; 1439 } 1440 } else if (addr == NULL) { 1441 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1442 error = ENOTCONN; 1443 else 1444 error = EDESTADDRREQ; 1445 SOCKBUF_UNLOCK(&so->so_snd); 1446 goto out; 1447 } 1448 } 1449 1450 /* 1451 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1452 * problem and need fixing. 1453 */ 1454 space = sbspace(&so->so_snd); 1455 if (flags & MSG_OOB) 1456 space += 1024; 1457 space -= clen; 1458 SOCKBUF_UNLOCK(&so->so_snd); 1459 if (resid > space) { 1460 error = EMSGSIZE; 1461 goto out; 1462 } 1463 if (uio == NULL) { 1464 resid = 0; 1465 if (flags & MSG_EOR) 1466 top->m_flags |= M_EOR; 1467 } else { 1468 /* 1469 * Copy the data from userland into a mbuf chain. 1470 * If no data is to be copied in, a single empty mbuf 1471 * is returned. 1472 */ 1473 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1474 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1475 if (top == NULL) { 1476 error = EFAULT; /* only possible error */ 1477 goto out; 1478 } 1479 space -= resid - uio->uio_resid; 1480 resid = uio->uio_resid; 1481 } 1482 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1483 /* 1484 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1485 * than with. 1486 */ 1487 if (dontroute) { 1488 SOCK_LOCK(so); 1489 so->so_options |= SO_DONTROUTE; 1490 SOCK_UNLOCK(so); 1491 } 1492 /* 1493 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1494 * of date. We could have received a reset packet in an interrupt or 1495 * maybe we slept while doing page faults in uiomove() etc. We could 1496 * probably recheck again inside the locking protection here, but 1497 * there are probably other places that this also happens. We must 1498 * rethink this. 1499 */ 1500 VNET_SO_ASSERT(so); 1501 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1502 (flags & MSG_OOB) ? PRUS_OOB : 1503 /* 1504 * If the user set MSG_EOF, the protocol understands this flag and 1505 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1506 */ 1507 ((flags & MSG_EOF) && 1508 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1509 (resid <= 0)) ? 1510 PRUS_EOF : 1511 /* If there is more to send set PRUS_MORETOCOME */ 1512 (flags & MSG_MORETOCOME) || 1513 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1514 top, addr, control, td); 1515 if (dontroute) { 1516 SOCK_LOCK(so); 1517 so->so_options &= ~SO_DONTROUTE; 1518 SOCK_UNLOCK(so); 1519 } 1520 clen = 0; 1521 control = NULL; 1522 top = NULL; 1523 out: 1524 if (top != NULL) 1525 m_freem(top); 1526 if (control != NULL) 1527 m_freem(control); 1528 return (error); 1529 } 1530 1531 /* 1532 * Send on a socket. If send must go all at once and message is larger than 1533 * send buffering, then hard error. Lock against other senders. If must go 1534 * all at once and not enough room now, then inform user that this would 1535 * block and do nothing. Otherwise, if nonblocking, send as much as 1536 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1537 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1538 * in mbuf chain must be small enough to send all at once. 1539 * 1540 * Returns nonzero on error, timeout or signal; callers must check for short 1541 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1542 * on return. 1543 */ 1544 int 1545 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1546 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1547 { 1548 long space; 1549 ssize_t resid; 1550 int clen = 0, error, dontroute; 1551 int atomic = sosendallatonce(so) || top; 1552 int pru_flag; 1553 #ifdef KERN_TLS 1554 struct ktls_session *tls; 1555 int tls_enq_cnt, tls_pruflag; 1556 uint8_t tls_rtype; 1557 1558 tls = NULL; 1559 tls_rtype = TLS_RLTYPE_APP; 1560 #endif 1561 if (uio != NULL) 1562 resid = uio->uio_resid; 1563 else if ((top->m_flags & M_PKTHDR) != 0) 1564 resid = top->m_pkthdr.len; 1565 else 1566 resid = m_length(top, NULL); 1567 /* 1568 * In theory resid should be unsigned. However, space must be 1569 * signed, as it might be less than 0 if we over-committed, and we 1570 * must use a signed comparison of space and resid. On the other 1571 * hand, a negative resid causes us to loop sending 0-length 1572 * segments to the protocol. 1573 * 1574 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1575 * type sockets since that's an error. 1576 */ 1577 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1578 error = EINVAL; 1579 goto out; 1580 } 1581 1582 dontroute = 1583 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1584 (so->so_proto->pr_flags & PR_ATOMIC); 1585 if (td != NULL) 1586 td->td_ru.ru_msgsnd++; 1587 if (control != NULL) 1588 clen = control->m_len; 1589 1590 error = sblock(&so->so_snd, SBLOCKWAIT(flags)); 1591 if (error) 1592 goto out; 1593 1594 #ifdef KERN_TLS 1595 tls_pruflag = 0; 1596 tls = ktls_hold(so->so_snd.sb_tls_info); 1597 if (tls != NULL) { 1598 if (tls->mode == TCP_TLS_MODE_SW) 1599 tls_pruflag = PRUS_NOTREADY; 1600 1601 if (control != NULL) { 1602 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1603 1604 if (clen >= sizeof(*cm) && 1605 cm->cmsg_type == TLS_SET_RECORD_TYPE) { 1606 tls_rtype = *((uint8_t *)CMSG_DATA(cm)); 1607 clen = 0; 1608 m_freem(control); 1609 control = NULL; 1610 atomic = 1; 1611 } 1612 } 1613 } 1614 #endif 1615 1616 restart: 1617 do { 1618 SOCKBUF_LOCK(&so->so_snd); 1619 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1620 SOCKBUF_UNLOCK(&so->so_snd); 1621 error = EPIPE; 1622 goto release; 1623 } 1624 if (so->so_error) { 1625 error = so->so_error; 1626 so->so_error = 0; 1627 SOCKBUF_UNLOCK(&so->so_snd); 1628 goto release; 1629 } 1630 if ((so->so_state & SS_ISCONNECTED) == 0) { 1631 /* 1632 * `sendto' and `sendmsg' is allowed on a connection- 1633 * based socket if it supports implied connect. 1634 * Return ENOTCONN if not connected and no address is 1635 * supplied. 1636 */ 1637 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1638 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1639 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1640 !(resid == 0 && clen != 0)) { 1641 SOCKBUF_UNLOCK(&so->so_snd); 1642 error = ENOTCONN; 1643 goto release; 1644 } 1645 } else if (addr == NULL) { 1646 SOCKBUF_UNLOCK(&so->so_snd); 1647 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1648 error = ENOTCONN; 1649 else 1650 error = EDESTADDRREQ; 1651 goto release; 1652 } 1653 } 1654 space = sbspace(&so->so_snd); 1655 if (flags & MSG_OOB) 1656 space += 1024; 1657 if ((atomic && resid > so->so_snd.sb_hiwat) || 1658 clen > so->so_snd.sb_hiwat) { 1659 SOCKBUF_UNLOCK(&so->so_snd); 1660 error = EMSGSIZE; 1661 goto release; 1662 } 1663 if (space < resid + clen && 1664 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1665 if ((so->so_state & SS_NBIO) || 1666 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) { 1667 SOCKBUF_UNLOCK(&so->so_snd); 1668 error = EWOULDBLOCK; 1669 goto release; 1670 } 1671 error = sbwait(&so->so_snd); 1672 SOCKBUF_UNLOCK(&so->so_snd); 1673 if (error) 1674 goto release; 1675 goto restart; 1676 } 1677 SOCKBUF_UNLOCK(&so->so_snd); 1678 space -= clen; 1679 do { 1680 if (uio == NULL) { 1681 resid = 0; 1682 if (flags & MSG_EOR) 1683 top->m_flags |= M_EOR; 1684 #ifdef KERN_TLS 1685 if (tls != NULL) { 1686 ktls_frame(top, tls, &tls_enq_cnt, 1687 tls_rtype); 1688 tls_rtype = TLS_RLTYPE_APP; 1689 } 1690 #endif 1691 } else { 1692 /* 1693 * Copy the data from userland into a mbuf 1694 * chain. If resid is 0, which can happen 1695 * only if we have control to send, then 1696 * a single empty mbuf is returned. This 1697 * is a workaround to prevent protocol send 1698 * methods to panic. 1699 */ 1700 #ifdef KERN_TLS 1701 if (tls != NULL) { 1702 top = m_uiotombuf(uio, M_WAITOK, space, 1703 tls->params.max_frame_len, 1704 M_EXTPG | 1705 ((flags & MSG_EOR) ? M_EOR : 0)); 1706 if (top != NULL) { 1707 ktls_frame(top, tls, 1708 &tls_enq_cnt, tls_rtype); 1709 } 1710 tls_rtype = TLS_RLTYPE_APP; 1711 } else 1712 #endif 1713 top = m_uiotombuf(uio, M_WAITOK, space, 1714 (atomic ? max_hdr : 0), 1715 (atomic ? M_PKTHDR : 0) | 1716 ((flags & MSG_EOR) ? M_EOR : 0)); 1717 if (top == NULL) { 1718 error = EFAULT; /* only possible error */ 1719 goto release; 1720 } 1721 space -= resid - uio->uio_resid; 1722 resid = uio->uio_resid; 1723 } 1724 if (dontroute) { 1725 SOCK_LOCK(so); 1726 so->so_options |= SO_DONTROUTE; 1727 SOCK_UNLOCK(so); 1728 } 1729 /* 1730 * XXX all the SBS_CANTSENDMORE checks previously 1731 * done could be out of date. We could have received 1732 * a reset packet in an interrupt or maybe we slept 1733 * while doing page faults in uiomove() etc. We 1734 * could probably recheck again inside the locking 1735 * protection here, but there are probably other 1736 * places that this also happens. We must rethink 1737 * this. 1738 */ 1739 VNET_SO_ASSERT(so); 1740 1741 pru_flag = (flags & MSG_OOB) ? PRUS_OOB : 1742 /* 1743 * If the user set MSG_EOF, the protocol understands 1744 * this flag and nothing left to send then use 1745 * PRU_SEND_EOF instead of PRU_SEND. 1746 */ 1747 ((flags & MSG_EOF) && 1748 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1749 (resid <= 0)) ? 1750 PRUS_EOF : 1751 /* If there is more to send set PRUS_MORETOCOME. */ 1752 (flags & MSG_MORETOCOME) || 1753 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; 1754 1755 #ifdef KERN_TLS 1756 pru_flag |= tls_pruflag; 1757 #endif 1758 1759 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1760 pru_flag, top, addr, control, td); 1761 1762 if (dontroute) { 1763 SOCK_LOCK(so); 1764 so->so_options &= ~SO_DONTROUTE; 1765 SOCK_UNLOCK(so); 1766 } 1767 1768 #ifdef KERN_TLS 1769 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) { 1770 if (error != 0) { 1771 m_freem(top); 1772 top = NULL; 1773 } else { 1774 soref(so); 1775 ktls_enqueue(top, so, tls_enq_cnt); 1776 } 1777 } 1778 #endif 1779 clen = 0; 1780 control = NULL; 1781 top = NULL; 1782 if (error) 1783 goto release; 1784 } while (resid && space > 0); 1785 } while (resid); 1786 1787 release: 1788 sbunlock(&so->so_snd); 1789 out: 1790 #ifdef KERN_TLS 1791 if (tls != NULL) 1792 ktls_free(tls); 1793 #endif 1794 if (top != NULL) 1795 m_freem(top); 1796 if (control != NULL) 1797 m_freem(control); 1798 return (error); 1799 } 1800 1801 int 1802 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1803 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1804 { 1805 int error; 1806 1807 CURVNET_SET(so->so_vnet); 1808 if (!SOLISTENING(so)) 1809 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, 1810 top, control, flags, td); 1811 else { 1812 m_freem(top); 1813 m_freem(control); 1814 error = ENOTCONN; 1815 } 1816 CURVNET_RESTORE(); 1817 return (error); 1818 } 1819 1820 /* 1821 * The part of soreceive() that implements reading non-inline out-of-band 1822 * data from a socket. For more complete comments, see soreceive(), from 1823 * which this code originated. 1824 * 1825 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1826 * unable to return an mbuf chain to the caller. 1827 */ 1828 static int 1829 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1830 { 1831 struct protosw *pr = so->so_proto; 1832 struct mbuf *m; 1833 int error; 1834 1835 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1836 VNET_SO_ASSERT(so); 1837 1838 m = m_get(M_WAITOK, MT_DATA); 1839 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 1840 if (error) 1841 goto bad; 1842 do { 1843 error = uiomove(mtod(m, void *), 1844 (int) min(uio->uio_resid, m->m_len), uio); 1845 m = m_free(m); 1846 } while (uio->uio_resid && error == 0 && m); 1847 bad: 1848 if (m != NULL) 1849 m_freem(m); 1850 return (error); 1851 } 1852 1853 /* 1854 * Following replacement or removal of the first mbuf on the first mbuf chain 1855 * of a socket buffer, push necessary state changes back into the socket 1856 * buffer so that other consumers see the values consistently. 'nextrecord' 1857 * is the callers locally stored value of the original value of 1858 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1859 * NOTE: 'nextrecord' may be NULL. 1860 */ 1861 static __inline void 1862 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1863 { 1864 1865 SOCKBUF_LOCK_ASSERT(sb); 1866 /* 1867 * First, update for the new value of nextrecord. If necessary, make 1868 * it the first record. 1869 */ 1870 if (sb->sb_mb != NULL) 1871 sb->sb_mb->m_nextpkt = nextrecord; 1872 else 1873 sb->sb_mb = nextrecord; 1874 1875 /* 1876 * Now update any dependent socket buffer fields to reflect the new 1877 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1878 * addition of a second clause that takes care of the case where 1879 * sb_mb has been updated, but remains the last record. 1880 */ 1881 if (sb->sb_mb == NULL) { 1882 sb->sb_mbtail = NULL; 1883 sb->sb_lastrecord = NULL; 1884 } else if (sb->sb_mb->m_nextpkt == NULL) 1885 sb->sb_lastrecord = sb->sb_mb; 1886 } 1887 1888 /* 1889 * Implement receive operations on a socket. We depend on the way that 1890 * records are added to the sockbuf by sbappend. In particular, each record 1891 * (mbufs linked through m_next) must begin with an address if the protocol 1892 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1893 * data, and then zero or more mbufs of data. In order to allow parallelism 1894 * between network receive and copying to user space, as well as avoid 1895 * sleeping with a mutex held, we release the socket buffer mutex during the 1896 * user space copy. Although the sockbuf is locked, new data may still be 1897 * appended, and thus we must maintain consistency of the sockbuf during that 1898 * time. 1899 * 1900 * The caller may receive the data as a single mbuf chain by supplying an 1901 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1902 * the count in uio_resid. 1903 */ 1904 int 1905 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1906 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1907 { 1908 struct mbuf *m, **mp; 1909 int flags, error, offset; 1910 ssize_t len; 1911 struct protosw *pr = so->so_proto; 1912 struct mbuf *nextrecord; 1913 int moff, type = 0; 1914 ssize_t orig_resid = uio->uio_resid; 1915 1916 mp = mp0; 1917 if (psa != NULL) 1918 *psa = NULL; 1919 if (controlp != NULL) 1920 *controlp = NULL; 1921 if (flagsp != NULL) 1922 flags = *flagsp &~ MSG_EOR; 1923 else 1924 flags = 0; 1925 if (flags & MSG_OOB) 1926 return (soreceive_rcvoob(so, uio, flags)); 1927 if (mp != NULL) 1928 *mp = NULL; 1929 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 1930 && uio->uio_resid) { 1931 VNET_SO_ASSERT(so); 1932 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 1933 } 1934 1935 error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); 1936 if (error) 1937 return (error); 1938 1939 restart: 1940 SOCKBUF_LOCK(&so->so_rcv); 1941 m = so->so_rcv.sb_mb; 1942 /* 1943 * If we have less data than requested, block awaiting more (subject 1944 * to any timeout) if: 1945 * 1. the current count is less than the low water mark, or 1946 * 2. MSG_DONTWAIT is not set 1947 */ 1948 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1949 sbavail(&so->so_rcv) < uio->uio_resid) && 1950 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && 1951 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 1952 KASSERT(m != NULL || !sbavail(&so->so_rcv), 1953 ("receive: m == %p sbavail == %u", 1954 m, sbavail(&so->so_rcv))); 1955 if (so->so_error) { 1956 if (m != NULL) 1957 goto dontblock; 1958 error = so->so_error; 1959 if ((flags & MSG_PEEK) == 0) 1960 so->so_error = 0; 1961 SOCKBUF_UNLOCK(&so->so_rcv); 1962 goto release; 1963 } 1964 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1965 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1966 if (m != NULL) 1967 goto dontblock; 1968 #ifdef KERN_TLS 1969 else if (so->so_rcv.sb_tlsdcc == 0 && 1970 so->so_rcv.sb_tlscc == 0) { 1971 #else 1972 else { 1973 #endif 1974 SOCKBUF_UNLOCK(&so->so_rcv); 1975 goto release; 1976 } 1977 } 1978 for (; m != NULL; m = m->m_next) 1979 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1980 m = so->so_rcv.sb_mb; 1981 goto dontblock; 1982 } 1983 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED | 1984 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 && 1985 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 1986 SOCKBUF_UNLOCK(&so->so_rcv); 1987 error = ENOTCONN; 1988 goto release; 1989 } 1990 if (uio->uio_resid == 0) { 1991 SOCKBUF_UNLOCK(&so->so_rcv); 1992 goto release; 1993 } 1994 if ((so->so_state & SS_NBIO) || 1995 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1996 SOCKBUF_UNLOCK(&so->so_rcv); 1997 error = EWOULDBLOCK; 1998 goto release; 1999 } 2000 SBLASTRECORDCHK(&so->so_rcv); 2001 SBLASTMBUFCHK(&so->so_rcv); 2002 error = sbwait(&so->so_rcv); 2003 SOCKBUF_UNLOCK(&so->so_rcv); 2004 if (error) 2005 goto release; 2006 goto restart; 2007 } 2008 dontblock: 2009 /* 2010 * From this point onward, we maintain 'nextrecord' as a cache of the 2011 * pointer to the next record in the socket buffer. We must keep the 2012 * various socket buffer pointers and local stack versions of the 2013 * pointers in sync, pushing out modifications before dropping the 2014 * socket buffer mutex, and re-reading them when picking it up. 2015 * 2016 * Otherwise, we will race with the network stack appending new data 2017 * or records onto the socket buffer by using inconsistent/stale 2018 * versions of the field, possibly resulting in socket buffer 2019 * corruption. 2020 * 2021 * By holding the high-level sblock(), we prevent simultaneous 2022 * readers from pulling off the front of the socket buffer. 2023 */ 2024 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2025 if (uio->uio_td) 2026 uio->uio_td->td_ru.ru_msgrcv++; 2027 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 2028 SBLASTRECORDCHK(&so->so_rcv); 2029 SBLASTMBUFCHK(&so->so_rcv); 2030 nextrecord = m->m_nextpkt; 2031 if (pr->pr_flags & PR_ADDR) { 2032 KASSERT(m->m_type == MT_SONAME, 2033 ("m->m_type == %d", m->m_type)); 2034 orig_resid = 0; 2035 if (psa != NULL) 2036 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2037 M_NOWAIT); 2038 if (flags & MSG_PEEK) { 2039 m = m->m_next; 2040 } else { 2041 sbfree(&so->so_rcv, m); 2042 so->so_rcv.sb_mb = m_free(m); 2043 m = so->so_rcv.sb_mb; 2044 sockbuf_pushsync(&so->so_rcv, nextrecord); 2045 } 2046 } 2047 2048 /* 2049 * Process one or more MT_CONTROL mbufs present before any data mbufs 2050 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 2051 * just copy the data; if !MSG_PEEK, we call into the protocol to 2052 * perform externalization (or freeing if controlp == NULL). 2053 */ 2054 if (m != NULL && m->m_type == MT_CONTROL) { 2055 struct mbuf *cm = NULL, *cmn; 2056 struct mbuf **cme = &cm; 2057 #ifdef KERN_TLS 2058 struct cmsghdr *cmsg; 2059 struct tls_get_record tgr; 2060 2061 /* 2062 * For MSG_TLSAPPDATA, check for a non-application data 2063 * record. If found, return ENXIO without removing 2064 * it from the receive queue. This allows a subsequent 2065 * call without MSG_TLSAPPDATA to receive it. 2066 * Note that, for TLS, there should only be a single 2067 * control mbuf with the TLS_GET_RECORD message in it. 2068 */ 2069 if (flags & MSG_TLSAPPDATA) { 2070 cmsg = mtod(m, struct cmsghdr *); 2071 if (cmsg->cmsg_type == TLS_GET_RECORD && 2072 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) { 2073 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr)); 2074 /* This will need to change for TLS 1.3. */ 2075 if (tgr.tls_type != TLS_RLTYPE_APP) { 2076 SOCKBUF_UNLOCK(&so->so_rcv); 2077 error = ENXIO; 2078 goto release; 2079 } 2080 } 2081 } 2082 #endif 2083 2084 do { 2085 if (flags & MSG_PEEK) { 2086 if (controlp != NULL) { 2087 *controlp = m_copym(m, 0, m->m_len, 2088 M_NOWAIT); 2089 controlp = &(*controlp)->m_next; 2090 } 2091 m = m->m_next; 2092 } else { 2093 sbfree(&so->so_rcv, m); 2094 so->so_rcv.sb_mb = m->m_next; 2095 m->m_next = NULL; 2096 *cme = m; 2097 cme = &(*cme)->m_next; 2098 m = so->so_rcv.sb_mb; 2099 } 2100 } while (m != NULL && m->m_type == MT_CONTROL); 2101 if ((flags & MSG_PEEK) == 0) 2102 sockbuf_pushsync(&so->so_rcv, nextrecord); 2103 while (cm != NULL) { 2104 cmn = cm->m_next; 2105 cm->m_next = NULL; 2106 if (pr->pr_domain->dom_externalize != NULL) { 2107 SOCKBUF_UNLOCK(&so->so_rcv); 2108 VNET_SO_ASSERT(so); 2109 error = (*pr->pr_domain->dom_externalize) 2110 (cm, controlp, flags); 2111 SOCKBUF_LOCK(&so->so_rcv); 2112 } else if (controlp != NULL) 2113 *controlp = cm; 2114 else 2115 m_freem(cm); 2116 if (controlp != NULL) { 2117 orig_resid = 0; 2118 while (*controlp != NULL) 2119 controlp = &(*controlp)->m_next; 2120 } 2121 cm = cmn; 2122 } 2123 if (m != NULL) 2124 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 2125 else 2126 nextrecord = so->so_rcv.sb_mb; 2127 orig_resid = 0; 2128 } 2129 if (m != NULL) { 2130 if ((flags & MSG_PEEK) == 0) { 2131 KASSERT(m->m_nextpkt == nextrecord, 2132 ("soreceive: post-control, nextrecord !sync")); 2133 if (nextrecord == NULL) { 2134 KASSERT(so->so_rcv.sb_mb == m, 2135 ("soreceive: post-control, sb_mb!=m")); 2136 KASSERT(so->so_rcv.sb_lastrecord == m, 2137 ("soreceive: post-control, lastrecord!=m")); 2138 } 2139 } 2140 type = m->m_type; 2141 if (type == MT_OOBDATA) 2142 flags |= MSG_OOB; 2143 } else { 2144 if ((flags & MSG_PEEK) == 0) { 2145 KASSERT(so->so_rcv.sb_mb == nextrecord, 2146 ("soreceive: sb_mb != nextrecord")); 2147 if (so->so_rcv.sb_mb == NULL) { 2148 KASSERT(so->so_rcv.sb_lastrecord == NULL, 2149 ("soreceive: sb_lastercord != NULL")); 2150 } 2151 } 2152 } 2153 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2154 SBLASTRECORDCHK(&so->so_rcv); 2155 SBLASTMBUFCHK(&so->so_rcv); 2156 2157 /* 2158 * Now continue to read any data mbufs off of the head of the socket 2159 * buffer until the read request is satisfied. Note that 'type' is 2160 * used to store the type of any mbuf reads that have happened so far 2161 * such that soreceive() can stop reading if the type changes, which 2162 * causes soreceive() to return only one of regular data and inline 2163 * out-of-band data in a single socket receive operation. 2164 */ 2165 moff = 0; 2166 offset = 0; 2167 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 2168 && error == 0) { 2169 /* 2170 * If the type of mbuf has changed since the last mbuf 2171 * examined ('type'), end the receive operation. 2172 */ 2173 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2174 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 2175 if (type != m->m_type) 2176 break; 2177 } else if (type == MT_OOBDATA) 2178 break; 2179 else 2180 KASSERT(m->m_type == MT_DATA, 2181 ("m->m_type == %d", m->m_type)); 2182 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 2183 len = uio->uio_resid; 2184 if (so->so_oobmark && len > so->so_oobmark - offset) 2185 len = so->so_oobmark - offset; 2186 if (len > m->m_len - moff) 2187 len = m->m_len - moff; 2188 /* 2189 * If mp is set, just pass back the mbufs. Otherwise copy 2190 * them out via the uio, then free. Sockbuf must be 2191 * consistent here (points to current mbuf, it points to next 2192 * record) when we drop priority; we must note any additions 2193 * to the sockbuf when we block interrupts again. 2194 */ 2195 if (mp == NULL) { 2196 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2197 SBLASTRECORDCHK(&so->so_rcv); 2198 SBLASTMBUFCHK(&so->so_rcv); 2199 SOCKBUF_UNLOCK(&so->so_rcv); 2200 if ((m->m_flags & M_EXTPG) != 0) 2201 error = m_unmapped_uiomove(m, moff, uio, 2202 (int)len); 2203 else 2204 error = uiomove(mtod(m, char *) + moff, 2205 (int)len, uio); 2206 SOCKBUF_LOCK(&so->so_rcv); 2207 if (error) { 2208 /* 2209 * The MT_SONAME mbuf has already been removed 2210 * from the record, so it is necessary to 2211 * remove the data mbufs, if any, to preserve 2212 * the invariant in the case of PR_ADDR that 2213 * requires MT_SONAME mbufs at the head of 2214 * each record. 2215 */ 2216 if (pr->pr_flags & PR_ATOMIC && 2217 ((flags & MSG_PEEK) == 0)) 2218 (void)sbdroprecord_locked(&so->so_rcv); 2219 SOCKBUF_UNLOCK(&so->so_rcv); 2220 goto release; 2221 } 2222 } else 2223 uio->uio_resid -= len; 2224 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2225 if (len == m->m_len - moff) { 2226 if (m->m_flags & M_EOR) 2227 flags |= MSG_EOR; 2228 if (flags & MSG_PEEK) { 2229 m = m->m_next; 2230 moff = 0; 2231 } else { 2232 nextrecord = m->m_nextpkt; 2233 sbfree(&so->so_rcv, m); 2234 if (mp != NULL) { 2235 m->m_nextpkt = NULL; 2236 *mp = m; 2237 mp = &m->m_next; 2238 so->so_rcv.sb_mb = m = m->m_next; 2239 *mp = NULL; 2240 } else { 2241 so->so_rcv.sb_mb = m_free(m); 2242 m = so->so_rcv.sb_mb; 2243 } 2244 sockbuf_pushsync(&so->so_rcv, nextrecord); 2245 SBLASTRECORDCHK(&so->so_rcv); 2246 SBLASTMBUFCHK(&so->so_rcv); 2247 } 2248 } else { 2249 if (flags & MSG_PEEK) 2250 moff += len; 2251 else { 2252 if (mp != NULL) { 2253 if (flags & MSG_DONTWAIT) { 2254 *mp = m_copym(m, 0, len, 2255 M_NOWAIT); 2256 if (*mp == NULL) { 2257 /* 2258 * m_copym() couldn't 2259 * allocate an mbuf. 2260 * Adjust uio_resid back 2261 * (it was adjusted 2262 * down by len bytes, 2263 * which we didn't end 2264 * up "copying" over). 2265 */ 2266 uio->uio_resid += len; 2267 break; 2268 } 2269 } else { 2270 SOCKBUF_UNLOCK(&so->so_rcv); 2271 *mp = m_copym(m, 0, len, 2272 M_WAITOK); 2273 SOCKBUF_LOCK(&so->so_rcv); 2274 } 2275 } 2276 sbcut_locked(&so->so_rcv, len); 2277 } 2278 } 2279 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2280 if (so->so_oobmark) { 2281 if ((flags & MSG_PEEK) == 0) { 2282 so->so_oobmark -= len; 2283 if (so->so_oobmark == 0) { 2284 so->so_rcv.sb_state |= SBS_RCVATMARK; 2285 break; 2286 } 2287 } else { 2288 offset += len; 2289 if (offset == so->so_oobmark) 2290 break; 2291 } 2292 } 2293 if (flags & MSG_EOR) 2294 break; 2295 /* 2296 * If the MSG_WAITALL flag is set (for non-atomic socket), we 2297 * must not quit until "uio->uio_resid == 0" or an error 2298 * termination. If a signal/timeout occurs, return with a 2299 * short count but without error. Keep sockbuf locked 2300 * against other readers. 2301 */ 2302 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 2303 !sosendallatonce(so) && nextrecord == NULL) { 2304 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2305 if (so->so_error || 2306 so->so_rcv.sb_state & SBS_CANTRCVMORE) 2307 break; 2308 /* 2309 * Notify the protocol that some data has been 2310 * drained before blocking. 2311 */ 2312 if (pr->pr_flags & PR_WANTRCVD) { 2313 SOCKBUF_UNLOCK(&so->so_rcv); 2314 VNET_SO_ASSERT(so); 2315 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2316 SOCKBUF_LOCK(&so->so_rcv); 2317 } 2318 SBLASTRECORDCHK(&so->so_rcv); 2319 SBLASTMBUFCHK(&so->so_rcv); 2320 /* 2321 * We could receive some data while was notifying 2322 * the protocol. Skip blocking in this case. 2323 */ 2324 if (so->so_rcv.sb_mb == NULL) { 2325 error = sbwait(&so->so_rcv); 2326 if (error) { 2327 SOCKBUF_UNLOCK(&so->so_rcv); 2328 goto release; 2329 } 2330 } 2331 m = so->so_rcv.sb_mb; 2332 if (m != NULL) 2333 nextrecord = m->m_nextpkt; 2334 } 2335 } 2336 2337 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2338 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 2339 flags |= MSG_TRUNC; 2340 if ((flags & MSG_PEEK) == 0) 2341 (void) sbdroprecord_locked(&so->so_rcv); 2342 } 2343 if ((flags & MSG_PEEK) == 0) { 2344 if (m == NULL) { 2345 /* 2346 * First part is an inline SB_EMPTY_FIXUP(). Second 2347 * part makes sure sb_lastrecord is up-to-date if 2348 * there is still data in the socket buffer. 2349 */ 2350 so->so_rcv.sb_mb = nextrecord; 2351 if (so->so_rcv.sb_mb == NULL) { 2352 so->so_rcv.sb_mbtail = NULL; 2353 so->so_rcv.sb_lastrecord = NULL; 2354 } else if (nextrecord->m_nextpkt == NULL) 2355 so->so_rcv.sb_lastrecord = nextrecord; 2356 } 2357 SBLASTRECORDCHK(&so->so_rcv); 2358 SBLASTMBUFCHK(&so->so_rcv); 2359 /* 2360 * If soreceive() is being done from the socket callback, 2361 * then don't need to generate ACK to peer to update window, 2362 * since ACK will be generated on return to TCP. 2363 */ 2364 if (!(flags & MSG_SOCALLBCK) && 2365 (pr->pr_flags & PR_WANTRCVD)) { 2366 SOCKBUF_UNLOCK(&so->so_rcv); 2367 VNET_SO_ASSERT(so); 2368 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2369 SOCKBUF_LOCK(&so->so_rcv); 2370 } 2371 } 2372 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2373 if (orig_resid == uio->uio_resid && orig_resid && 2374 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 2375 SOCKBUF_UNLOCK(&so->so_rcv); 2376 goto restart; 2377 } 2378 SOCKBUF_UNLOCK(&so->so_rcv); 2379 2380 if (flagsp != NULL) 2381 *flagsp |= flags; 2382 release: 2383 sbunlock(&so->so_rcv); 2384 return (error); 2385 } 2386 2387 /* 2388 * Optimized version of soreceive() for stream (TCP) sockets. 2389 */ 2390 int 2391 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 2392 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2393 { 2394 int len = 0, error = 0, flags, oresid; 2395 struct sockbuf *sb; 2396 struct mbuf *m, *n = NULL; 2397 2398 /* We only do stream sockets. */ 2399 if (so->so_type != SOCK_STREAM) 2400 return (EINVAL); 2401 if (psa != NULL) 2402 *psa = NULL; 2403 if (flagsp != NULL) 2404 flags = *flagsp &~ MSG_EOR; 2405 else 2406 flags = 0; 2407 if (controlp != NULL) 2408 *controlp = NULL; 2409 if (flags & MSG_OOB) 2410 return (soreceive_rcvoob(so, uio, flags)); 2411 if (mp0 != NULL) 2412 *mp0 = NULL; 2413 2414 sb = &so->so_rcv; 2415 2416 #ifdef KERN_TLS 2417 /* 2418 * KTLS store TLS records as records with a control message to 2419 * describe the framing. 2420 * 2421 * We check once here before acquiring locks to optimize the 2422 * common case. 2423 */ 2424 if (sb->sb_tls_info != NULL) 2425 return (soreceive_generic(so, psa, uio, mp0, controlp, 2426 flagsp)); 2427 #endif 2428 2429 /* Prevent other readers from entering the socket. */ 2430 error = sblock(sb, SBLOCKWAIT(flags)); 2431 if (error) 2432 return (error); 2433 SOCKBUF_LOCK(sb); 2434 2435 #ifdef KERN_TLS 2436 if (sb->sb_tls_info != NULL) { 2437 SOCKBUF_UNLOCK(sb); 2438 sbunlock(sb); 2439 return (soreceive_generic(so, psa, uio, mp0, controlp, 2440 flagsp)); 2441 } 2442 #endif 2443 2444 /* Easy one, no space to copyout anything. */ 2445 if (uio->uio_resid == 0) { 2446 error = EINVAL; 2447 goto out; 2448 } 2449 oresid = uio->uio_resid; 2450 2451 /* We will never ever get anything unless we are or were connected. */ 2452 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 2453 error = ENOTCONN; 2454 goto out; 2455 } 2456 2457 restart: 2458 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2459 2460 /* Abort if socket has reported problems. */ 2461 if (so->so_error) { 2462 if (sbavail(sb) > 0) 2463 goto deliver; 2464 if (oresid > uio->uio_resid) 2465 goto out; 2466 error = so->so_error; 2467 if (!(flags & MSG_PEEK)) 2468 so->so_error = 0; 2469 goto out; 2470 } 2471 2472 /* Door is closed. Deliver what is left, if any. */ 2473 if (sb->sb_state & SBS_CANTRCVMORE) { 2474 if (sbavail(sb) > 0) 2475 goto deliver; 2476 else 2477 goto out; 2478 } 2479 2480 /* Socket buffer is empty and we shall not block. */ 2481 if (sbavail(sb) == 0 && 2482 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2483 error = EAGAIN; 2484 goto out; 2485 } 2486 2487 /* Socket buffer got some data that we shall deliver now. */ 2488 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 2489 ((so->so_state & SS_NBIO) || 2490 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2491 sbavail(sb) >= sb->sb_lowat || 2492 sbavail(sb) >= uio->uio_resid || 2493 sbavail(sb) >= sb->sb_hiwat) ) { 2494 goto deliver; 2495 } 2496 2497 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2498 if ((flags & MSG_WAITALL) && 2499 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 2500 goto deliver; 2501 2502 /* 2503 * Wait and block until (more) data comes in. 2504 * NB: Drops the sockbuf lock during wait. 2505 */ 2506 error = sbwait(sb); 2507 if (error) 2508 goto out; 2509 goto restart; 2510 2511 deliver: 2512 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2513 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 2514 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2515 2516 /* Statistics. */ 2517 if (uio->uio_td) 2518 uio->uio_td->td_ru.ru_msgrcv++; 2519 2520 /* Fill uio until full or current end of socket buffer is reached. */ 2521 len = min(uio->uio_resid, sbavail(sb)); 2522 if (mp0 != NULL) { 2523 /* Dequeue as many mbufs as possible. */ 2524 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2525 if (*mp0 == NULL) 2526 *mp0 = sb->sb_mb; 2527 else 2528 m_cat(*mp0, sb->sb_mb); 2529 for (m = sb->sb_mb; 2530 m != NULL && m->m_len <= len; 2531 m = m->m_next) { 2532 KASSERT(!(m->m_flags & M_NOTAVAIL), 2533 ("%s: m %p not available", __func__, m)); 2534 len -= m->m_len; 2535 uio->uio_resid -= m->m_len; 2536 sbfree(sb, m); 2537 n = m; 2538 } 2539 n->m_next = NULL; 2540 sb->sb_mb = m; 2541 sb->sb_lastrecord = sb->sb_mb; 2542 if (sb->sb_mb == NULL) 2543 SB_EMPTY_FIXUP(sb); 2544 } 2545 /* Copy the remainder. */ 2546 if (len > 0) { 2547 KASSERT(sb->sb_mb != NULL, 2548 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2549 2550 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2551 if (m == NULL) 2552 len = 0; /* Don't flush data from sockbuf. */ 2553 else 2554 uio->uio_resid -= len; 2555 if (*mp0 != NULL) 2556 m_cat(*mp0, m); 2557 else 2558 *mp0 = m; 2559 if (*mp0 == NULL) { 2560 error = ENOBUFS; 2561 goto out; 2562 } 2563 } 2564 } else { 2565 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2566 SOCKBUF_UNLOCK(sb); 2567 error = m_mbuftouio(uio, sb->sb_mb, len); 2568 SOCKBUF_LOCK(sb); 2569 if (error) 2570 goto out; 2571 } 2572 SBLASTRECORDCHK(sb); 2573 SBLASTMBUFCHK(sb); 2574 2575 /* 2576 * Remove the delivered data from the socket buffer unless we 2577 * were only peeking. 2578 */ 2579 if (!(flags & MSG_PEEK)) { 2580 if (len > 0) 2581 sbdrop_locked(sb, len); 2582 2583 /* Notify protocol that we drained some data. */ 2584 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2585 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2586 !(flags & MSG_SOCALLBCK))) { 2587 SOCKBUF_UNLOCK(sb); 2588 VNET_SO_ASSERT(so); 2589 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags); 2590 SOCKBUF_LOCK(sb); 2591 } 2592 } 2593 2594 /* 2595 * For MSG_WAITALL we may have to loop again and wait for 2596 * more data to come in. 2597 */ 2598 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2599 goto restart; 2600 out: 2601 SOCKBUF_LOCK_ASSERT(sb); 2602 SBLASTRECORDCHK(sb); 2603 SBLASTMBUFCHK(sb); 2604 SOCKBUF_UNLOCK(sb); 2605 sbunlock(sb); 2606 return (error); 2607 } 2608 2609 /* 2610 * Optimized version of soreceive() for simple datagram cases from userspace. 2611 * Unlike in the stream case, we're able to drop a datagram if copyout() 2612 * fails, and because we handle datagrams atomically, we don't need to use a 2613 * sleep lock to prevent I/O interlacing. 2614 */ 2615 int 2616 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2617 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2618 { 2619 struct mbuf *m, *m2; 2620 int flags, error; 2621 ssize_t len; 2622 struct protosw *pr = so->so_proto; 2623 struct mbuf *nextrecord; 2624 2625 if (psa != NULL) 2626 *psa = NULL; 2627 if (controlp != NULL) 2628 *controlp = NULL; 2629 if (flagsp != NULL) 2630 flags = *flagsp &~ MSG_EOR; 2631 else 2632 flags = 0; 2633 2634 /* 2635 * For any complicated cases, fall back to the full 2636 * soreceive_generic(). 2637 */ 2638 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB)) 2639 return (soreceive_generic(so, psa, uio, mp0, controlp, 2640 flagsp)); 2641 2642 /* 2643 * Enforce restrictions on use. 2644 */ 2645 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2646 ("soreceive_dgram: wantrcvd")); 2647 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2648 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2649 ("soreceive_dgram: SBS_RCVATMARK")); 2650 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2651 ("soreceive_dgram: P_CONNREQUIRED")); 2652 2653 /* 2654 * Loop blocking while waiting for a datagram. 2655 */ 2656 SOCKBUF_LOCK(&so->so_rcv); 2657 while ((m = so->so_rcv.sb_mb) == NULL) { 2658 KASSERT(sbavail(&so->so_rcv) == 0, 2659 ("soreceive_dgram: sb_mb NULL but sbavail %u", 2660 sbavail(&so->so_rcv))); 2661 if (so->so_error) { 2662 error = so->so_error; 2663 so->so_error = 0; 2664 SOCKBUF_UNLOCK(&so->so_rcv); 2665 return (error); 2666 } 2667 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2668 uio->uio_resid == 0) { 2669 SOCKBUF_UNLOCK(&so->so_rcv); 2670 return (0); 2671 } 2672 if ((so->so_state & SS_NBIO) || 2673 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2674 SOCKBUF_UNLOCK(&so->so_rcv); 2675 return (EWOULDBLOCK); 2676 } 2677 SBLASTRECORDCHK(&so->so_rcv); 2678 SBLASTMBUFCHK(&so->so_rcv); 2679 error = sbwait(&so->so_rcv); 2680 if (error) { 2681 SOCKBUF_UNLOCK(&so->so_rcv); 2682 return (error); 2683 } 2684 } 2685 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2686 2687 if (uio->uio_td) 2688 uio->uio_td->td_ru.ru_msgrcv++; 2689 SBLASTRECORDCHK(&so->so_rcv); 2690 SBLASTMBUFCHK(&so->so_rcv); 2691 nextrecord = m->m_nextpkt; 2692 if (nextrecord == NULL) { 2693 KASSERT(so->so_rcv.sb_lastrecord == m, 2694 ("soreceive_dgram: lastrecord != m")); 2695 } 2696 2697 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2698 ("soreceive_dgram: m_nextpkt != nextrecord")); 2699 2700 /* 2701 * Pull 'm' and its chain off the front of the packet queue. 2702 */ 2703 so->so_rcv.sb_mb = NULL; 2704 sockbuf_pushsync(&so->so_rcv, nextrecord); 2705 2706 /* 2707 * Walk 'm's chain and free that many bytes from the socket buffer. 2708 */ 2709 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2710 sbfree(&so->so_rcv, m2); 2711 2712 /* 2713 * Do a few last checks before we let go of the lock. 2714 */ 2715 SBLASTRECORDCHK(&so->so_rcv); 2716 SBLASTMBUFCHK(&so->so_rcv); 2717 SOCKBUF_UNLOCK(&so->so_rcv); 2718 2719 if (pr->pr_flags & PR_ADDR) { 2720 KASSERT(m->m_type == MT_SONAME, 2721 ("m->m_type == %d", m->m_type)); 2722 if (psa != NULL) 2723 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2724 M_NOWAIT); 2725 m = m_free(m); 2726 } 2727 if (m == NULL) { 2728 /* XXXRW: Can this happen? */ 2729 return (0); 2730 } 2731 2732 /* 2733 * Packet to copyout() is now in 'm' and it is disconnected from the 2734 * queue. 2735 * 2736 * Process one or more MT_CONTROL mbufs present before any data mbufs 2737 * in the first mbuf chain on the socket buffer. We call into the 2738 * protocol to perform externalization (or freeing if controlp == 2739 * NULL). In some cases there can be only MT_CONTROL mbufs without 2740 * MT_DATA mbufs. 2741 */ 2742 if (m->m_type == MT_CONTROL) { 2743 struct mbuf *cm = NULL, *cmn; 2744 struct mbuf **cme = &cm; 2745 2746 do { 2747 m2 = m->m_next; 2748 m->m_next = NULL; 2749 *cme = m; 2750 cme = &(*cme)->m_next; 2751 m = m2; 2752 } while (m != NULL && m->m_type == MT_CONTROL); 2753 while (cm != NULL) { 2754 cmn = cm->m_next; 2755 cm->m_next = NULL; 2756 if (pr->pr_domain->dom_externalize != NULL) { 2757 error = (*pr->pr_domain->dom_externalize) 2758 (cm, controlp, flags); 2759 } else if (controlp != NULL) 2760 *controlp = cm; 2761 else 2762 m_freem(cm); 2763 if (controlp != NULL) { 2764 while (*controlp != NULL) 2765 controlp = &(*controlp)->m_next; 2766 } 2767 cm = cmn; 2768 } 2769 } 2770 KASSERT(m == NULL || m->m_type == MT_DATA, 2771 ("soreceive_dgram: !data")); 2772 while (m != NULL && uio->uio_resid > 0) { 2773 len = uio->uio_resid; 2774 if (len > m->m_len) 2775 len = m->m_len; 2776 error = uiomove(mtod(m, char *), (int)len, uio); 2777 if (error) { 2778 m_freem(m); 2779 return (error); 2780 } 2781 if (len == m->m_len) 2782 m = m_free(m); 2783 else { 2784 m->m_data += len; 2785 m->m_len -= len; 2786 } 2787 } 2788 if (m != NULL) { 2789 flags |= MSG_TRUNC; 2790 m_freem(m); 2791 } 2792 if (flagsp != NULL) 2793 *flagsp |= flags; 2794 return (0); 2795 } 2796 2797 int 2798 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2799 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2800 { 2801 int error; 2802 2803 CURVNET_SET(so->so_vnet); 2804 if (!SOLISTENING(so)) 2805 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, 2806 mp0, controlp, flagsp)); 2807 else 2808 error = ENOTCONN; 2809 CURVNET_RESTORE(); 2810 return (error); 2811 } 2812 2813 int 2814 soshutdown(struct socket *so, int how) 2815 { 2816 struct protosw *pr = so->so_proto; 2817 int error, soerror_enotconn; 2818 2819 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2820 return (EINVAL); 2821 2822 soerror_enotconn = 0; 2823 if ((so->so_state & 2824 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 2825 /* 2826 * POSIX mandates us to return ENOTCONN when shutdown(2) is 2827 * invoked on a datagram sockets, however historically we would 2828 * actually tear socket down. This is known to be leveraged by 2829 * some applications to unblock process waiting in recvXXX(2) 2830 * by other process that it shares that socket with. Try to meet 2831 * both backward-compatibility and POSIX requirements by forcing 2832 * ENOTCONN but still asking protocol to perform pru_shutdown(). 2833 */ 2834 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) 2835 return (ENOTCONN); 2836 soerror_enotconn = 1; 2837 } 2838 2839 if (SOLISTENING(so)) { 2840 if (how != SHUT_WR) { 2841 SOLISTEN_LOCK(so); 2842 so->so_error = ECONNABORTED; 2843 solisten_wakeup(so); /* unlocks so */ 2844 } 2845 goto done; 2846 } 2847 2848 CURVNET_SET(so->so_vnet); 2849 if (pr->pr_usrreqs->pru_flush != NULL) 2850 (*pr->pr_usrreqs->pru_flush)(so, how); 2851 if (how != SHUT_WR) 2852 sorflush(so); 2853 if (how != SHUT_RD) { 2854 error = (*pr->pr_usrreqs->pru_shutdown)(so); 2855 wakeup(&so->so_timeo); 2856 CURVNET_RESTORE(); 2857 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error); 2858 } 2859 wakeup(&so->so_timeo); 2860 CURVNET_RESTORE(); 2861 2862 done: 2863 return (soerror_enotconn ? ENOTCONN : 0); 2864 } 2865 2866 void 2867 sorflush(struct socket *so) 2868 { 2869 struct sockbuf *sb = &so->so_rcv; 2870 struct protosw *pr = so->so_proto; 2871 struct socket aso; 2872 2873 VNET_SO_ASSERT(so); 2874 2875 /* 2876 * In order to avoid calling dom_dispose with the socket buffer mutex 2877 * held, and in order to generally avoid holding the lock for a long 2878 * time, we make a copy of the socket buffer and clear the original 2879 * (except locks, state). The new socket buffer copy won't have 2880 * initialized locks so we can only call routines that won't use or 2881 * assert those locks. 2882 * 2883 * Dislodge threads currently blocked in receive and wait to acquire 2884 * a lock against other simultaneous readers before clearing the 2885 * socket buffer. Don't let our acquire be interrupted by a signal 2886 * despite any existing socket disposition on interruptable waiting. 2887 */ 2888 socantrcvmore(so); 2889 (void) sblock(sb, SBL_WAIT | SBL_NOINTR); 2890 2891 /* 2892 * Invalidate/clear most of the sockbuf structure, but leave selinfo 2893 * and mutex data unchanged. 2894 */ 2895 SOCKBUF_LOCK(sb); 2896 bzero(&aso, sizeof(aso)); 2897 aso.so_pcb = so->so_pcb; 2898 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero, 2899 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2900 bzero(&sb->sb_startzero, 2901 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2902 SOCKBUF_UNLOCK(sb); 2903 sbunlock(sb); 2904 2905 /* 2906 * Dispose of special rights and flush the copied socket. Don't call 2907 * any unsafe routines (that rely on locks being initialized) on aso. 2908 */ 2909 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 2910 (*pr->pr_domain->dom_dispose)(&aso); 2911 sbrelease_internal(&aso.so_rcv, so); 2912 } 2913 2914 /* 2915 * Wrapper for Socket established helper hook. 2916 * Parameters: socket, context of the hook point, hook id. 2917 */ 2918 static int inline 2919 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 2920 { 2921 struct socket_hhook_data hhook_data = { 2922 .so = so, 2923 .hctx = hctx, 2924 .m = NULL, 2925 .status = 0 2926 }; 2927 2928 CURVNET_SET(so->so_vnet); 2929 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 2930 CURVNET_RESTORE(); 2931 2932 /* Ugly but needed, since hhooks return void for now */ 2933 return (hhook_data.status); 2934 } 2935 2936 /* 2937 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 2938 * additional variant to handle the case where the option value needs to be 2939 * some kind of integer, but not a specific size. In addition to their use 2940 * here, these functions are also called by the protocol-level pr_ctloutput() 2941 * routines. 2942 */ 2943 int 2944 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2945 { 2946 size_t valsize; 2947 2948 /* 2949 * If the user gives us more than we wanted, we ignore it, but if we 2950 * don't get the minimum length the caller wants, we return EINVAL. 2951 * On success, sopt->sopt_valsize is set to however much we actually 2952 * retrieved. 2953 */ 2954 if ((valsize = sopt->sopt_valsize) < minlen) 2955 return EINVAL; 2956 if (valsize > len) 2957 sopt->sopt_valsize = valsize = len; 2958 2959 if (sopt->sopt_td != NULL) 2960 return (copyin(sopt->sopt_val, buf, valsize)); 2961 2962 bcopy(sopt->sopt_val, buf, valsize); 2963 return (0); 2964 } 2965 2966 /* 2967 * Kernel version of setsockopt(2). 2968 * 2969 * XXX: optlen is size_t, not socklen_t 2970 */ 2971 int 2972 so_setsockopt(struct socket *so, int level, int optname, void *optval, 2973 size_t optlen) 2974 { 2975 struct sockopt sopt; 2976 2977 sopt.sopt_level = level; 2978 sopt.sopt_name = optname; 2979 sopt.sopt_dir = SOPT_SET; 2980 sopt.sopt_val = optval; 2981 sopt.sopt_valsize = optlen; 2982 sopt.sopt_td = NULL; 2983 return (sosetopt(so, &sopt)); 2984 } 2985 2986 int 2987 sosetopt(struct socket *so, struct sockopt *sopt) 2988 { 2989 int error, optval; 2990 struct linger l; 2991 struct timeval tv; 2992 sbintime_t val; 2993 uint32_t val32; 2994 #ifdef MAC 2995 struct mac extmac; 2996 #endif 2997 2998 CURVNET_SET(so->so_vnet); 2999 error = 0; 3000 if (sopt->sopt_level != SOL_SOCKET) { 3001 if (so->so_proto->pr_ctloutput != NULL) 3002 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3003 else 3004 error = ENOPROTOOPT; 3005 } else { 3006 switch (sopt->sopt_name) { 3007 case SO_ACCEPTFILTER: 3008 error = accept_filt_setopt(so, sopt); 3009 if (error) 3010 goto bad; 3011 break; 3012 3013 case SO_LINGER: 3014 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 3015 if (error) 3016 goto bad; 3017 if (l.l_linger < 0 || 3018 l.l_linger > USHRT_MAX || 3019 l.l_linger > (INT_MAX / hz)) { 3020 error = EDOM; 3021 goto bad; 3022 } 3023 SOCK_LOCK(so); 3024 so->so_linger = l.l_linger; 3025 if (l.l_onoff) 3026 so->so_options |= SO_LINGER; 3027 else 3028 so->so_options &= ~SO_LINGER; 3029 SOCK_UNLOCK(so); 3030 break; 3031 3032 case SO_DEBUG: 3033 case SO_KEEPALIVE: 3034 case SO_DONTROUTE: 3035 case SO_USELOOPBACK: 3036 case SO_BROADCAST: 3037 case SO_REUSEADDR: 3038 case SO_REUSEPORT: 3039 case SO_REUSEPORT_LB: 3040 case SO_OOBINLINE: 3041 case SO_TIMESTAMP: 3042 case SO_BINTIME: 3043 case SO_NOSIGPIPE: 3044 case SO_NO_DDP: 3045 case SO_NO_OFFLOAD: 3046 error = sooptcopyin(sopt, &optval, sizeof optval, 3047 sizeof optval); 3048 if (error) 3049 goto bad; 3050 SOCK_LOCK(so); 3051 if (optval) 3052 so->so_options |= sopt->sopt_name; 3053 else 3054 so->so_options &= ~sopt->sopt_name; 3055 SOCK_UNLOCK(so); 3056 break; 3057 3058 case SO_SETFIB: 3059 error = sooptcopyin(sopt, &optval, sizeof optval, 3060 sizeof optval); 3061 if (error) 3062 goto bad; 3063 3064 if (optval < 0 || optval >= rt_numfibs) { 3065 error = EINVAL; 3066 goto bad; 3067 } 3068 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 3069 (so->so_proto->pr_domain->dom_family == PF_INET6) || 3070 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 3071 so->so_fibnum = optval; 3072 else 3073 so->so_fibnum = 0; 3074 break; 3075 3076 case SO_USER_COOKIE: 3077 error = sooptcopyin(sopt, &val32, sizeof val32, 3078 sizeof val32); 3079 if (error) 3080 goto bad; 3081 so->so_user_cookie = val32; 3082 break; 3083 3084 case SO_SNDBUF: 3085 case SO_RCVBUF: 3086 case SO_SNDLOWAT: 3087 case SO_RCVLOWAT: 3088 error = sooptcopyin(sopt, &optval, sizeof optval, 3089 sizeof optval); 3090 if (error) 3091 goto bad; 3092 3093 /* 3094 * Values < 1 make no sense for any of these options, 3095 * so disallow them. 3096 */ 3097 if (optval < 1) { 3098 error = EINVAL; 3099 goto bad; 3100 } 3101 3102 error = sbsetopt(so, sopt->sopt_name, optval); 3103 break; 3104 3105 case SO_SNDTIMEO: 3106 case SO_RCVTIMEO: 3107 #ifdef COMPAT_FREEBSD32 3108 if (SV_CURPROC_FLAG(SV_ILP32)) { 3109 struct timeval32 tv32; 3110 3111 error = sooptcopyin(sopt, &tv32, sizeof tv32, 3112 sizeof tv32); 3113 CP(tv32, tv, tv_sec); 3114 CP(tv32, tv, tv_usec); 3115 } else 3116 #endif 3117 error = sooptcopyin(sopt, &tv, sizeof tv, 3118 sizeof tv); 3119 if (error) 3120 goto bad; 3121 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 3122 tv.tv_usec >= 1000000) { 3123 error = EDOM; 3124 goto bad; 3125 } 3126 if (tv.tv_sec > INT32_MAX) 3127 val = SBT_MAX; 3128 else 3129 val = tvtosbt(tv); 3130 switch (sopt->sopt_name) { 3131 case SO_SNDTIMEO: 3132 so->so_snd.sb_timeo = val; 3133 break; 3134 case SO_RCVTIMEO: 3135 so->so_rcv.sb_timeo = val; 3136 break; 3137 } 3138 break; 3139 3140 case SO_LABEL: 3141 #ifdef MAC 3142 error = sooptcopyin(sopt, &extmac, sizeof extmac, 3143 sizeof extmac); 3144 if (error) 3145 goto bad; 3146 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 3147 so, &extmac); 3148 #else 3149 error = EOPNOTSUPP; 3150 #endif 3151 break; 3152 3153 case SO_TS_CLOCK: 3154 error = sooptcopyin(sopt, &optval, sizeof optval, 3155 sizeof optval); 3156 if (error) 3157 goto bad; 3158 if (optval < 0 || optval > SO_TS_CLOCK_MAX) { 3159 error = EINVAL; 3160 goto bad; 3161 } 3162 so->so_ts_clock = optval; 3163 break; 3164 3165 case SO_MAX_PACING_RATE: 3166 error = sooptcopyin(sopt, &val32, sizeof(val32), 3167 sizeof(val32)); 3168 if (error) 3169 goto bad; 3170 so->so_max_pacing_rate = val32; 3171 break; 3172 3173 default: 3174 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3175 error = hhook_run_socket(so, sopt, 3176 HHOOK_SOCKET_OPT); 3177 else 3178 error = ENOPROTOOPT; 3179 break; 3180 } 3181 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 3182 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 3183 } 3184 bad: 3185 CURVNET_RESTORE(); 3186 return (error); 3187 } 3188 3189 /* 3190 * Helper routine for getsockopt. 3191 */ 3192 int 3193 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 3194 { 3195 int error; 3196 size_t valsize; 3197 3198 error = 0; 3199 3200 /* 3201 * Documented get behavior is that we always return a value, possibly 3202 * truncated to fit in the user's buffer. Traditional behavior is 3203 * that we always tell the user precisely how much we copied, rather 3204 * than something useful like the total amount we had available for 3205 * her. Note that this interface is not idempotent; the entire 3206 * answer must be generated ahead of time. 3207 */ 3208 valsize = min(len, sopt->sopt_valsize); 3209 sopt->sopt_valsize = valsize; 3210 if (sopt->sopt_val != NULL) { 3211 if (sopt->sopt_td != NULL) 3212 error = copyout(buf, sopt->sopt_val, valsize); 3213 else 3214 bcopy(buf, sopt->sopt_val, valsize); 3215 } 3216 return (error); 3217 } 3218 3219 int 3220 sogetopt(struct socket *so, struct sockopt *sopt) 3221 { 3222 int error, optval; 3223 struct linger l; 3224 struct timeval tv; 3225 #ifdef MAC 3226 struct mac extmac; 3227 #endif 3228 3229 CURVNET_SET(so->so_vnet); 3230 error = 0; 3231 if (sopt->sopt_level != SOL_SOCKET) { 3232 if (so->so_proto->pr_ctloutput != NULL) 3233 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3234 else 3235 error = ENOPROTOOPT; 3236 CURVNET_RESTORE(); 3237 return (error); 3238 } else { 3239 switch (sopt->sopt_name) { 3240 case SO_ACCEPTFILTER: 3241 error = accept_filt_getopt(so, sopt); 3242 break; 3243 3244 case SO_LINGER: 3245 SOCK_LOCK(so); 3246 l.l_onoff = so->so_options & SO_LINGER; 3247 l.l_linger = so->so_linger; 3248 SOCK_UNLOCK(so); 3249 error = sooptcopyout(sopt, &l, sizeof l); 3250 break; 3251 3252 case SO_USELOOPBACK: 3253 case SO_DONTROUTE: 3254 case SO_DEBUG: 3255 case SO_KEEPALIVE: 3256 case SO_REUSEADDR: 3257 case SO_REUSEPORT: 3258 case SO_REUSEPORT_LB: 3259 case SO_BROADCAST: 3260 case SO_OOBINLINE: 3261 case SO_ACCEPTCONN: 3262 case SO_TIMESTAMP: 3263 case SO_BINTIME: 3264 case SO_NOSIGPIPE: 3265 case SO_NO_DDP: 3266 case SO_NO_OFFLOAD: 3267 optval = so->so_options & sopt->sopt_name; 3268 integer: 3269 error = sooptcopyout(sopt, &optval, sizeof optval); 3270 break; 3271 3272 case SO_DOMAIN: 3273 optval = so->so_proto->pr_domain->dom_family; 3274 goto integer; 3275 3276 case SO_TYPE: 3277 optval = so->so_type; 3278 goto integer; 3279 3280 case SO_PROTOCOL: 3281 optval = so->so_proto->pr_protocol; 3282 goto integer; 3283 3284 case SO_ERROR: 3285 SOCK_LOCK(so); 3286 optval = so->so_error; 3287 so->so_error = 0; 3288 SOCK_UNLOCK(so); 3289 goto integer; 3290 3291 case SO_SNDBUF: 3292 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : 3293 so->so_snd.sb_hiwat; 3294 goto integer; 3295 3296 case SO_RCVBUF: 3297 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : 3298 so->so_rcv.sb_hiwat; 3299 goto integer; 3300 3301 case SO_SNDLOWAT: 3302 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : 3303 so->so_snd.sb_lowat; 3304 goto integer; 3305 3306 case SO_RCVLOWAT: 3307 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : 3308 so->so_rcv.sb_lowat; 3309 goto integer; 3310 3311 case SO_SNDTIMEO: 3312 case SO_RCVTIMEO: 3313 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 3314 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 3315 #ifdef COMPAT_FREEBSD32 3316 if (SV_CURPROC_FLAG(SV_ILP32)) { 3317 struct timeval32 tv32; 3318 3319 CP(tv, tv32, tv_sec); 3320 CP(tv, tv32, tv_usec); 3321 error = sooptcopyout(sopt, &tv32, sizeof tv32); 3322 } else 3323 #endif 3324 error = sooptcopyout(sopt, &tv, sizeof tv); 3325 break; 3326 3327 case SO_LABEL: 3328 #ifdef MAC 3329 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3330 sizeof(extmac)); 3331 if (error) 3332 goto bad; 3333 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 3334 so, &extmac); 3335 if (error) 3336 goto bad; 3337 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3338 #else 3339 error = EOPNOTSUPP; 3340 #endif 3341 break; 3342 3343 case SO_PEERLABEL: 3344 #ifdef MAC 3345 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3346 sizeof(extmac)); 3347 if (error) 3348 goto bad; 3349 error = mac_getsockopt_peerlabel( 3350 sopt->sopt_td->td_ucred, so, &extmac); 3351 if (error) 3352 goto bad; 3353 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3354 #else 3355 error = EOPNOTSUPP; 3356 #endif 3357 break; 3358 3359 case SO_LISTENQLIMIT: 3360 optval = SOLISTENING(so) ? so->sol_qlimit : 0; 3361 goto integer; 3362 3363 case SO_LISTENQLEN: 3364 optval = SOLISTENING(so) ? so->sol_qlen : 0; 3365 goto integer; 3366 3367 case SO_LISTENINCQLEN: 3368 optval = SOLISTENING(so) ? so->sol_incqlen : 0; 3369 goto integer; 3370 3371 case SO_TS_CLOCK: 3372 optval = so->so_ts_clock; 3373 goto integer; 3374 3375 case SO_MAX_PACING_RATE: 3376 optval = so->so_max_pacing_rate; 3377 goto integer; 3378 3379 default: 3380 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3381 error = hhook_run_socket(so, sopt, 3382 HHOOK_SOCKET_OPT); 3383 else 3384 error = ENOPROTOOPT; 3385 break; 3386 } 3387 } 3388 #ifdef MAC 3389 bad: 3390 #endif 3391 CURVNET_RESTORE(); 3392 return (error); 3393 } 3394 3395 int 3396 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 3397 { 3398 struct mbuf *m, *m_prev; 3399 int sopt_size = sopt->sopt_valsize; 3400 3401 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3402 if (m == NULL) 3403 return ENOBUFS; 3404 if (sopt_size > MLEN) { 3405 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 3406 if ((m->m_flags & M_EXT) == 0) { 3407 m_free(m); 3408 return ENOBUFS; 3409 } 3410 m->m_len = min(MCLBYTES, sopt_size); 3411 } else { 3412 m->m_len = min(MLEN, sopt_size); 3413 } 3414 sopt_size -= m->m_len; 3415 *mp = m; 3416 m_prev = m; 3417 3418 while (sopt_size) { 3419 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3420 if (m == NULL) { 3421 m_freem(*mp); 3422 return ENOBUFS; 3423 } 3424 if (sopt_size > MLEN) { 3425 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 3426 M_NOWAIT); 3427 if ((m->m_flags & M_EXT) == 0) { 3428 m_freem(m); 3429 m_freem(*mp); 3430 return ENOBUFS; 3431 } 3432 m->m_len = min(MCLBYTES, sopt_size); 3433 } else { 3434 m->m_len = min(MLEN, sopt_size); 3435 } 3436 sopt_size -= m->m_len; 3437 m_prev->m_next = m; 3438 m_prev = m; 3439 } 3440 return (0); 3441 } 3442 3443 int 3444 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 3445 { 3446 struct mbuf *m0 = m; 3447 3448 if (sopt->sopt_val == NULL) 3449 return (0); 3450 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3451 if (sopt->sopt_td != NULL) { 3452 int error; 3453 3454 error = copyin(sopt->sopt_val, mtod(m, char *), 3455 m->m_len); 3456 if (error != 0) { 3457 m_freem(m0); 3458 return(error); 3459 } 3460 } else 3461 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 3462 sopt->sopt_valsize -= m->m_len; 3463 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3464 m = m->m_next; 3465 } 3466 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 3467 panic("ip6_sooptmcopyin"); 3468 return (0); 3469 } 3470 3471 int 3472 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 3473 { 3474 struct mbuf *m0 = m; 3475 size_t valsize = 0; 3476 3477 if (sopt->sopt_val == NULL) 3478 return (0); 3479 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3480 if (sopt->sopt_td != NULL) { 3481 int error; 3482 3483 error = copyout(mtod(m, char *), sopt->sopt_val, 3484 m->m_len); 3485 if (error != 0) { 3486 m_freem(m0); 3487 return(error); 3488 } 3489 } else 3490 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 3491 sopt->sopt_valsize -= m->m_len; 3492 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3493 valsize += m->m_len; 3494 m = m->m_next; 3495 } 3496 if (m != NULL) { 3497 /* enough soopt buffer should be given from user-land */ 3498 m_freem(m0); 3499 return(EINVAL); 3500 } 3501 sopt->sopt_valsize = valsize; 3502 return (0); 3503 } 3504 3505 /* 3506 * sohasoutofband(): protocol notifies socket layer of the arrival of new 3507 * out-of-band data, which will then notify socket consumers. 3508 */ 3509 void 3510 sohasoutofband(struct socket *so) 3511 { 3512 3513 if (so->so_sigio != NULL) 3514 pgsigio(&so->so_sigio, SIGURG, 0); 3515 selwakeuppri(&so->so_rdsel, PSOCK); 3516 } 3517 3518 int 3519 sopoll(struct socket *so, int events, struct ucred *active_cred, 3520 struct thread *td) 3521 { 3522 3523 /* 3524 * We do not need to set or assert curvnet as long as everyone uses 3525 * sopoll_generic(). 3526 */ 3527 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, 3528 td)); 3529 } 3530 3531 int 3532 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3533 struct thread *td) 3534 { 3535 int revents; 3536 3537 SOCK_LOCK(so); 3538 if (SOLISTENING(so)) { 3539 if (!(events & (POLLIN | POLLRDNORM))) 3540 revents = 0; 3541 else if (!TAILQ_EMPTY(&so->sol_comp)) 3542 revents = events & (POLLIN | POLLRDNORM); 3543 else if ((events & POLLINIGNEOF) == 0 && so->so_error) 3544 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP; 3545 else { 3546 selrecord(td, &so->so_rdsel); 3547 revents = 0; 3548 } 3549 } else { 3550 revents = 0; 3551 SOCKBUF_LOCK(&so->so_snd); 3552 SOCKBUF_LOCK(&so->so_rcv); 3553 if (events & (POLLIN | POLLRDNORM)) 3554 if (soreadabledata(so)) 3555 revents |= events & (POLLIN | POLLRDNORM); 3556 if (events & (POLLOUT | POLLWRNORM)) 3557 if (sowriteable(so)) 3558 revents |= events & (POLLOUT | POLLWRNORM); 3559 if (events & (POLLPRI | POLLRDBAND)) 3560 if (so->so_oobmark || 3561 (so->so_rcv.sb_state & SBS_RCVATMARK)) 3562 revents |= events & (POLLPRI | POLLRDBAND); 3563 if ((events & POLLINIGNEOF) == 0) { 3564 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3565 revents |= events & (POLLIN | POLLRDNORM); 3566 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3567 revents |= POLLHUP; 3568 } 3569 } 3570 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3571 revents |= events & POLLRDHUP; 3572 if (revents == 0) { 3573 if (events & 3574 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) { 3575 selrecord(td, &so->so_rdsel); 3576 so->so_rcv.sb_flags |= SB_SEL; 3577 } 3578 if (events & (POLLOUT | POLLWRNORM)) { 3579 selrecord(td, &so->so_wrsel); 3580 so->so_snd.sb_flags |= SB_SEL; 3581 } 3582 } 3583 SOCKBUF_UNLOCK(&so->so_rcv); 3584 SOCKBUF_UNLOCK(&so->so_snd); 3585 } 3586 SOCK_UNLOCK(so); 3587 return (revents); 3588 } 3589 3590 int 3591 soo_kqfilter(struct file *fp, struct knote *kn) 3592 { 3593 struct socket *so = kn->kn_fp->f_data; 3594 struct sockbuf *sb; 3595 struct knlist *knl; 3596 3597 switch (kn->kn_filter) { 3598 case EVFILT_READ: 3599 kn->kn_fop = &soread_filtops; 3600 knl = &so->so_rdsel.si_note; 3601 sb = &so->so_rcv; 3602 break; 3603 case EVFILT_WRITE: 3604 kn->kn_fop = &sowrite_filtops; 3605 knl = &so->so_wrsel.si_note; 3606 sb = &so->so_snd; 3607 break; 3608 case EVFILT_EMPTY: 3609 kn->kn_fop = &soempty_filtops; 3610 knl = &so->so_wrsel.si_note; 3611 sb = &so->so_snd; 3612 break; 3613 default: 3614 return (EINVAL); 3615 } 3616 3617 SOCK_LOCK(so); 3618 if (SOLISTENING(so)) { 3619 knlist_add(knl, kn, 1); 3620 } else { 3621 SOCKBUF_LOCK(sb); 3622 knlist_add(knl, kn, 1); 3623 sb->sb_flags |= SB_KNOTE; 3624 SOCKBUF_UNLOCK(sb); 3625 } 3626 SOCK_UNLOCK(so); 3627 return (0); 3628 } 3629 3630 /* 3631 * Some routines that return EOPNOTSUPP for entry points that are not 3632 * supported by a protocol. Fill in as needed. 3633 */ 3634 int 3635 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 3636 { 3637 3638 return EOPNOTSUPP; 3639 } 3640 3641 int 3642 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job) 3643 { 3644 3645 return EOPNOTSUPP; 3646 } 3647 3648 int 3649 pru_attach_notsupp(struct socket *so, int proto, struct thread *td) 3650 { 3651 3652 return EOPNOTSUPP; 3653 } 3654 3655 int 3656 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3657 { 3658 3659 return EOPNOTSUPP; 3660 } 3661 3662 int 3663 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3664 struct thread *td) 3665 { 3666 3667 return EOPNOTSUPP; 3668 } 3669 3670 int 3671 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3672 { 3673 3674 return EOPNOTSUPP; 3675 } 3676 3677 int 3678 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3679 struct thread *td) 3680 { 3681 3682 return EOPNOTSUPP; 3683 } 3684 3685 int 3686 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 3687 { 3688 3689 return EOPNOTSUPP; 3690 } 3691 3692 int 3693 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 3694 struct ifnet *ifp, struct thread *td) 3695 { 3696 3697 return EOPNOTSUPP; 3698 } 3699 3700 int 3701 pru_disconnect_notsupp(struct socket *so) 3702 { 3703 3704 return EOPNOTSUPP; 3705 } 3706 3707 int 3708 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) 3709 { 3710 3711 return EOPNOTSUPP; 3712 } 3713 3714 int 3715 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) 3716 { 3717 3718 return EOPNOTSUPP; 3719 } 3720 3721 int 3722 pru_rcvd_notsupp(struct socket *so, int flags) 3723 { 3724 3725 return EOPNOTSUPP; 3726 } 3727 3728 int 3729 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 3730 { 3731 3732 return EOPNOTSUPP; 3733 } 3734 3735 int 3736 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, 3737 struct sockaddr *addr, struct mbuf *control, struct thread *td) 3738 { 3739 3740 if (control != NULL) 3741 m_freem(control); 3742 if ((flags & PRUS_NOTREADY) == 0) 3743 m_freem(m); 3744 return (EOPNOTSUPP); 3745 } 3746 3747 int 3748 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count) 3749 { 3750 3751 return (EOPNOTSUPP); 3752 } 3753 3754 /* 3755 * This isn't really a ``null'' operation, but it's the default one and 3756 * doesn't do anything destructive. 3757 */ 3758 int 3759 pru_sense_null(struct socket *so, struct stat *sb) 3760 { 3761 3762 sb->st_blksize = so->so_snd.sb_hiwat; 3763 return 0; 3764 } 3765 3766 int 3767 pru_shutdown_notsupp(struct socket *so) 3768 { 3769 3770 return EOPNOTSUPP; 3771 } 3772 3773 int 3774 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) 3775 { 3776 3777 return EOPNOTSUPP; 3778 } 3779 3780 int 3781 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 3782 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 3783 { 3784 3785 return EOPNOTSUPP; 3786 } 3787 3788 int 3789 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 3790 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 3791 { 3792 3793 return EOPNOTSUPP; 3794 } 3795 3796 int 3797 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, 3798 struct thread *td) 3799 { 3800 3801 return EOPNOTSUPP; 3802 } 3803 3804 static void 3805 filt_sordetach(struct knote *kn) 3806 { 3807 struct socket *so = kn->kn_fp->f_data; 3808 3809 so_rdknl_lock(so); 3810 knlist_remove(&so->so_rdsel.si_note, kn, 1); 3811 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note)) 3812 so->so_rcv.sb_flags &= ~SB_KNOTE; 3813 so_rdknl_unlock(so); 3814 } 3815 3816 /*ARGSUSED*/ 3817 static int 3818 filt_soread(struct knote *kn, long hint) 3819 { 3820 struct socket *so; 3821 3822 so = kn->kn_fp->f_data; 3823 3824 if (SOLISTENING(so)) { 3825 SOCK_LOCK_ASSERT(so); 3826 kn->kn_data = so->sol_qlen; 3827 if (so->so_error) { 3828 kn->kn_flags |= EV_EOF; 3829 kn->kn_fflags = so->so_error; 3830 return (1); 3831 } 3832 return (!TAILQ_EMPTY(&so->sol_comp)); 3833 } 3834 3835 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3836 3837 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 3838 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3839 kn->kn_flags |= EV_EOF; 3840 kn->kn_fflags = so->so_error; 3841 return (1); 3842 } else if (so->so_error) /* temporary udp error */ 3843 return (1); 3844 3845 if (kn->kn_sfflags & NOTE_LOWAT) { 3846 if (kn->kn_data >= kn->kn_sdata) 3847 return (1); 3848 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) 3849 return (1); 3850 3851 /* This hook returning non-zero indicates an event, not error */ 3852 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); 3853 } 3854 3855 static void 3856 filt_sowdetach(struct knote *kn) 3857 { 3858 struct socket *so = kn->kn_fp->f_data; 3859 3860 so_wrknl_lock(so); 3861 knlist_remove(&so->so_wrsel.si_note, kn, 1); 3862 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note)) 3863 so->so_snd.sb_flags &= ~SB_KNOTE; 3864 so_wrknl_unlock(so); 3865 } 3866 3867 /*ARGSUSED*/ 3868 static int 3869 filt_sowrite(struct knote *kn, long hint) 3870 { 3871 struct socket *so; 3872 3873 so = kn->kn_fp->f_data; 3874 3875 if (SOLISTENING(so)) 3876 return (0); 3877 3878 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3879 kn->kn_data = sbspace(&so->so_snd); 3880 3881 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); 3882 3883 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3884 kn->kn_flags |= EV_EOF; 3885 kn->kn_fflags = so->so_error; 3886 return (1); 3887 } else if (so->so_error) /* temporary udp error */ 3888 return (1); 3889 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3890 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3891 return (0); 3892 else if (kn->kn_sfflags & NOTE_LOWAT) 3893 return (kn->kn_data >= kn->kn_sdata); 3894 else 3895 return (kn->kn_data >= so->so_snd.sb_lowat); 3896 } 3897 3898 static int 3899 filt_soempty(struct knote *kn, long hint) 3900 { 3901 struct socket *so; 3902 3903 so = kn->kn_fp->f_data; 3904 3905 if (SOLISTENING(so)) 3906 return (1); 3907 3908 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3909 kn->kn_data = sbused(&so->so_snd); 3910 3911 if (kn->kn_data == 0) 3912 return (1); 3913 else 3914 return (0); 3915 } 3916 3917 int 3918 socheckuid(struct socket *so, uid_t uid) 3919 { 3920 3921 if (so == NULL) 3922 return (EPERM); 3923 if (so->so_cred->cr_uid != uid) 3924 return (EPERM); 3925 return (0); 3926 } 3927 3928 /* 3929 * These functions are used by protocols to notify the socket layer (and its 3930 * consumers) of state changes in the sockets driven by protocol-side events. 3931 */ 3932 3933 /* 3934 * Procedures to manipulate state flags of socket and do appropriate wakeups. 3935 * 3936 * Normal sequence from the active (originating) side is that 3937 * soisconnecting() is called during processing of connect() call, resulting 3938 * in an eventual call to soisconnected() if/when the connection is 3939 * established. When the connection is torn down soisdisconnecting() is 3940 * called during processing of disconnect() call, and soisdisconnected() is 3941 * called when the connection to the peer is totally severed. The semantics 3942 * of these routines are such that connectionless protocols can call 3943 * soisconnected() and soisdisconnected() only, bypassing the in-progress 3944 * calls when setting up a ``connection'' takes no time. 3945 * 3946 * From the passive side, a socket is created with two queues of sockets: 3947 * so_incomp for connections in progress and so_comp for connections already 3948 * made and awaiting user acceptance. As a protocol is preparing incoming 3949 * connections, it creates a socket structure queued on so_incomp by calling 3950 * sonewconn(). When the connection is established, soisconnected() is 3951 * called, and transfers the socket structure to so_comp, making it available 3952 * to accept(). 3953 * 3954 * If a socket is closed with sockets on either so_incomp or so_comp, these 3955 * sockets are dropped. 3956 * 3957 * If higher-level protocols are implemented in the kernel, the wakeups done 3958 * here will sometimes cause software-interrupt process scheduling. 3959 */ 3960 void 3961 soisconnecting(struct socket *so) 3962 { 3963 3964 SOCK_LOCK(so); 3965 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 3966 so->so_state |= SS_ISCONNECTING; 3967 SOCK_UNLOCK(so); 3968 } 3969 3970 void 3971 soisconnected(struct socket *so) 3972 { 3973 3974 SOCK_LOCK(so); 3975 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 3976 so->so_state |= SS_ISCONNECTED; 3977 3978 if (so->so_qstate == SQ_INCOMP) { 3979 struct socket *head = so->so_listen; 3980 int ret; 3981 3982 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so)); 3983 /* 3984 * Promoting a socket from incomplete queue to complete, we 3985 * need to go through reverse order of locking. We first do 3986 * trylock, and if that doesn't succeed, we go the hard way 3987 * leaving a reference and rechecking consistency after proper 3988 * locking. 3989 */ 3990 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) { 3991 soref(head); 3992 SOCK_UNLOCK(so); 3993 SOLISTEN_LOCK(head); 3994 SOCK_LOCK(so); 3995 if (__predict_false(head != so->so_listen)) { 3996 /* 3997 * The socket went off the listen queue, 3998 * should be lost race to close(2) of sol. 3999 * The socket is about to soabort(). 4000 */ 4001 SOCK_UNLOCK(so); 4002 sorele(head); 4003 return; 4004 } 4005 /* Not the last one, as so holds a ref. */ 4006 refcount_release(&head->so_count); 4007 } 4008 again: 4009 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 4010 TAILQ_REMOVE(&head->sol_incomp, so, so_list); 4011 head->sol_incqlen--; 4012 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 4013 head->sol_qlen++; 4014 so->so_qstate = SQ_COMP; 4015 SOCK_UNLOCK(so); 4016 solisten_wakeup(head); /* unlocks */ 4017 } else { 4018 SOCKBUF_LOCK(&so->so_rcv); 4019 soupcall_set(so, SO_RCV, 4020 head->sol_accept_filter->accf_callback, 4021 head->sol_accept_filter_arg); 4022 so->so_options &= ~SO_ACCEPTFILTER; 4023 ret = head->sol_accept_filter->accf_callback(so, 4024 head->sol_accept_filter_arg, M_NOWAIT); 4025 if (ret == SU_ISCONNECTED) { 4026 soupcall_clear(so, SO_RCV); 4027 SOCKBUF_UNLOCK(&so->so_rcv); 4028 goto again; 4029 } 4030 SOCKBUF_UNLOCK(&so->so_rcv); 4031 SOCK_UNLOCK(so); 4032 SOLISTEN_UNLOCK(head); 4033 } 4034 return; 4035 } 4036 SOCK_UNLOCK(so); 4037 wakeup(&so->so_timeo); 4038 sorwakeup(so); 4039 sowwakeup(so); 4040 } 4041 4042 void 4043 soisdisconnecting(struct socket *so) 4044 { 4045 4046 SOCK_LOCK(so); 4047 so->so_state &= ~SS_ISCONNECTING; 4048 so->so_state |= SS_ISDISCONNECTING; 4049 4050 if (!SOLISTENING(so)) { 4051 SOCKBUF_LOCK(&so->so_rcv); 4052 socantrcvmore_locked(so); 4053 SOCKBUF_LOCK(&so->so_snd); 4054 socantsendmore_locked(so); 4055 } 4056 SOCK_UNLOCK(so); 4057 wakeup(&so->so_timeo); 4058 } 4059 4060 void 4061 soisdisconnected(struct socket *so) 4062 { 4063 4064 SOCK_LOCK(so); 4065 4066 /* 4067 * There is at least one reader of so_state that does not 4068 * acquire socket lock, namely soreceive_generic(). Ensure 4069 * that it never sees all flags that track connection status 4070 * cleared, by ordering the update with a barrier semantic of 4071 * our release thread fence. 4072 */ 4073 so->so_state |= SS_ISDISCONNECTED; 4074 atomic_thread_fence_rel(); 4075 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 4076 4077 if (!SOLISTENING(so)) { 4078 SOCK_UNLOCK(so); 4079 SOCKBUF_LOCK(&so->so_rcv); 4080 socantrcvmore_locked(so); 4081 SOCKBUF_LOCK(&so->so_snd); 4082 sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); 4083 socantsendmore_locked(so); 4084 } else 4085 SOCK_UNLOCK(so); 4086 wakeup(&so->so_timeo); 4087 } 4088 4089 /* 4090 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 4091 */ 4092 struct sockaddr * 4093 sodupsockaddr(const struct sockaddr *sa, int mflags) 4094 { 4095 struct sockaddr *sa2; 4096 4097 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 4098 if (sa2) 4099 bcopy(sa, sa2, sa->sa_len); 4100 return sa2; 4101 } 4102 4103 /* 4104 * Register per-socket destructor. 4105 */ 4106 void 4107 sodtor_set(struct socket *so, so_dtor_t *func) 4108 { 4109 4110 SOCK_LOCK_ASSERT(so); 4111 so->so_dtor = func; 4112 } 4113 4114 /* 4115 * Register per-socket buffer upcalls. 4116 */ 4117 void 4118 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg) 4119 { 4120 struct sockbuf *sb; 4121 4122 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4123 4124 switch (which) { 4125 case SO_RCV: 4126 sb = &so->so_rcv; 4127 break; 4128 case SO_SND: 4129 sb = &so->so_snd; 4130 break; 4131 default: 4132 panic("soupcall_set: bad which"); 4133 } 4134 SOCKBUF_LOCK_ASSERT(sb); 4135 sb->sb_upcall = func; 4136 sb->sb_upcallarg = arg; 4137 sb->sb_flags |= SB_UPCALL; 4138 } 4139 4140 void 4141 soupcall_clear(struct socket *so, int which) 4142 { 4143 struct sockbuf *sb; 4144 4145 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4146 4147 switch (which) { 4148 case SO_RCV: 4149 sb = &so->so_rcv; 4150 break; 4151 case SO_SND: 4152 sb = &so->so_snd; 4153 break; 4154 default: 4155 panic("soupcall_clear: bad which"); 4156 } 4157 SOCKBUF_LOCK_ASSERT(sb); 4158 KASSERT(sb->sb_upcall != NULL, 4159 ("%s: so %p no upcall to clear", __func__, so)); 4160 sb->sb_upcall = NULL; 4161 sb->sb_upcallarg = NULL; 4162 sb->sb_flags &= ~SB_UPCALL; 4163 } 4164 4165 void 4166 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg) 4167 { 4168 4169 SOLISTEN_LOCK_ASSERT(so); 4170 so->sol_upcall = func; 4171 so->sol_upcallarg = arg; 4172 } 4173 4174 static void 4175 so_rdknl_lock(void *arg) 4176 { 4177 struct socket *so = arg; 4178 4179 if (SOLISTENING(so)) 4180 SOCK_LOCK(so); 4181 else 4182 SOCKBUF_LOCK(&so->so_rcv); 4183 } 4184 4185 static void 4186 so_rdknl_unlock(void *arg) 4187 { 4188 struct socket *so = arg; 4189 4190 if (SOLISTENING(so)) 4191 SOCK_UNLOCK(so); 4192 else 4193 SOCKBUF_UNLOCK(&so->so_rcv); 4194 } 4195 4196 static void 4197 so_rdknl_assert_lock(void *arg, int what) 4198 { 4199 struct socket *so = arg; 4200 4201 if (what == LA_LOCKED) { 4202 if (SOLISTENING(so)) 4203 SOCK_LOCK_ASSERT(so); 4204 else 4205 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 4206 } else { 4207 if (SOLISTENING(so)) 4208 SOCK_UNLOCK_ASSERT(so); 4209 else 4210 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); 4211 } 4212 } 4213 4214 static void 4215 so_wrknl_lock(void *arg) 4216 { 4217 struct socket *so = arg; 4218 4219 if (SOLISTENING(so)) 4220 SOCK_LOCK(so); 4221 else 4222 SOCKBUF_LOCK(&so->so_snd); 4223 } 4224 4225 static void 4226 so_wrknl_unlock(void *arg) 4227 { 4228 struct socket *so = arg; 4229 4230 if (SOLISTENING(so)) 4231 SOCK_UNLOCK(so); 4232 else 4233 SOCKBUF_UNLOCK(&so->so_snd); 4234 } 4235 4236 static void 4237 so_wrknl_assert_lock(void *arg, int what) 4238 { 4239 struct socket *so = arg; 4240 4241 if (what == LA_LOCKED) { 4242 if (SOLISTENING(so)) 4243 SOCK_LOCK_ASSERT(so); 4244 else 4245 SOCKBUF_LOCK_ASSERT(&so->so_snd); 4246 } else { 4247 if (SOLISTENING(so)) 4248 SOCK_UNLOCK_ASSERT(so); 4249 else 4250 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 4251 } 4252 } 4253 4254 /* 4255 * Create an external-format (``xsocket'') structure using the information in 4256 * the kernel-format socket structure pointed to by so. This is done to 4257 * reduce the spew of irrelevant information over this interface, to isolate 4258 * user code from changes in the kernel structure, and potentially to provide 4259 * information-hiding if we decide that some of this information should be 4260 * hidden from users. 4261 */ 4262 void 4263 sotoxsocket(struct socket *so, struct xsocket *xso) 4264 { 4265 4266 bzero(xso, sizeof(*xso)); 4267 xso->xso_len = sizeof *xso; 4268 xso->xso_so = (uintptr_t)so; 4269 xso->so_type = so->so_type; 4270 xso->so_options = so->so_options; 4271 xso->so_linger = so->so_linger; 4272 xso->so_state = so->so_state; 4273 xso->so_pcb = (uintptr_t)so->so_pcb; 4274 xso->xso_protocol = so->so_proto->pr_protocol; 4275 xso->xso_family = so->so_proto->pr_domain->dom_family; 4276 xso->so_timeo = so->so_timeo; 4277 xso->so_error = so->so_error; 4278 xso->so_uid = so->so_cred->cr_uid; 4279 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 4280 if (SOLISTENING(so)) { 4281 xso->so_qlen = so->sol_qlen; 4282 xso->so_incqlen = so->sol_incqlen; 4283 xso->so_qlimit = so->sol_qlimit; 4284 xso->so_oobmark = 0; 4285 } else { 4286 xso->so_state |= so->so_qstate; 4287 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0; 4288 xso->so_oobmark = so->so_oobmark; 4289 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 4290 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 4291 } 4292 } 4293 4294 struct sockbuf * 4295 so_sockbuf_rcv(struct socket *so) 4296 { 4297 4298 return (&so->so_rcv); 4299 } 4300 4301 struct sockbuf * 4302 so_sockbuf_snd(struct socket *so) 4303 { 4304 4305 return (&so->so_snd); 4306 } 4307 4308 int 4309 so_state_get(const struct socket *so) 4310 { 4311 4312 return (so->so_state); 4313 } 4314 4315 void 4316 so_state_set(struct socket *so, int val) 4317 { 4318 4319 so->so_state = val; 4320 } 4321 4322 int 4323 so_options_get(const struct socket *so) 4324 { 4325 4326 return (so->so_options); 4327 } 4328 4329 void 4330 so_options_set(struct socket *so, int val) 4331 { 4332 4333 so->so_options = val; 4334 } 4335 4336 int 4337 so_error_get(const struct socket *so) 4338 { 4339 4340 return (so->so_error); 4341 } 4342 4343 void 4344 so_error_set(struct socket *so, int val) 4345 { 4346 4347 so->so_error = val; 4348 } 4349 4350 int 4351 so_linger_get(const struct socket *so) 4352 { 4353 4354 return (so->so_linger); 4355 } 4356 4357 void 4358 so_linger_set(struct socket *so, int val) 4359 { 4360 4361 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz), 4362 ("%s: val %d out of range", __func__, val)); 4363 4364 so->so_linger = val; 4365 } 4366 4367 struct protosw * 4368 so_protosw_get(const struct socket *so) 4369 { 4370 4371 return (so->so_proto); 4372 } 4373 4374 void 4375 so_protosw_set(struct socket *so, struct protosw *val) 4376 { 4377 4378 so->so_proto = val; 4379 } 4380 4381 void 4382 so_sorwakeup(struct socket *so) 4383 { 4384 4385 sorwakeup(so); 4386 } 4387 4388 void 4389 so_sowwakeup(struct socket *so) 4390 { 4391 4392 sowwakeup(so); 4393 } 4394 4395 void 4396 so_sorwakeup_locked(struct socket *so) 4397 { 4398 4399 sorwakeup_locked(so); 4400 } 4401 4402 void 4403 so_sowwakeup_locked(struct socket *so) 4404 { 4405 4406 sowwakeup_locked(so); 4407 } 4408 4409 void 4410 so_lock(struct socket *so) 4411 { 4412 4413 SOCK_LOCK(so); 4414 } 4415 4416 void 4417 so_unlock(struct socket *so) 4418 { 4419 4420 SOCK_UNLOCK(so); 4421 } 4422