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/fcntl.h> 116 #include <sys/limits.h> 117 #include <sys/lock.h> 118 #include <sys/mac.h> 119 #include <sys/malloc.h> 120 #include <sys/mbuf.h> 121 #include <sys/mutex.h> 122 #include <sys/domain.h> 123 #include <sys/file.h> /* for struct knote */ 124 #include <sys/hhook.h> 125 #include <sys/kernel.h> 126 #include <sys/khelp.h> 127 #include <sys/ktls.h> 128 #include <sys/event.h> 129 #include <sys/eventhandler.h> 130 #include <sys/poll.h> 131 #include <sys/proc.h> 132 #include <sys/protosw.h> 133 #include <sys/sbuf.h> 134 #include <sys/socket.h> 135 #include <sys/socketvar.h> 136 #include <sys/resourcevar.h> 137 #include <net/route.h> 138 #include <sys/signalvar.h> 139 #include <sys/stat.h> 140 #include <sys/sx.h> 141 #include <sys/sysctl.h> 142 #include <sys/taskqueue.h> 143 #include <sys/uio.h> 144 #include <sys/un.h> 145 #include <sys/unpcb.h> 146 #include <sys/jail.h> 147 #include <sys/syslog.h> 148 #include <netinet/in.h> 149 #include <netinet/in_pcb.h> 150 #include <netinet/tcp.h> 151 152 #include <net/vnet.h> 153 154 #include <security/mac/mac_framework.h> 155 156 #include <vm/uma.h> 157 158 #ifdef COMPAT_FREEBSD32 159 #include <sys/mount.h> 160 #include <sys/sysent.h> 161 #include <compat/freebsd32/freebsd32.h> 162 #endif 163 164 static int soreceive_rcvoob(struct socket *so, struct uio *uio, 165 int flags); 166 static void so_rdknl_lock(void *); 167 static void so_rdknl_unlock(void *); 168 static void so_rdknl_assert_lock(void *, int); 169 static void so_wrknl_lock(void *); 170 static void so_wrknl_unlock(void *); 171 static void so_wrknl_assert_lock(void *, int); 172 173 static void filt_sordetach(struct knote *kn); 174 static int filt_soread(struct knote *kn, long hint); 175 static void filt_sowdetach(struct knote *kn); 176 static int filt_sowrite(struct knote *kn, long hint); 177 static int filt_soempty(struct knote *kn, long hint); 178 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id); 179 fo_kqfilter_t soo_kqfilter; 180 181 static struct filterops soread_filtops = { 182 .f_isfd = 1, 183 .f_detach = filt_sordetach, 184 .f_event = filt_soread, 185 }; 186 static struct filterops sowrite_filtops = { 187 .f_isfd = 1, 188 .f_detach = filt_sowdetach, 189 .f_event = filt_sowrite, 190 }; 191 static struct filterops soempty_filtops = { 192 .f_isfd = 1, 193 .f_detach = filt_sowdetach, 194 .f_event = filt_soempty, 195 }; 196 197 so_gen_t so_gencnt; /* generation count for sockets */ 198 199 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 200 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 201 202 #define VNET_SO_ASSERT(so) \ 203 VNET_ASSERT(curvnet != NULL, \ 204 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so))); 205 206 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]); 207 #define V_socket_hhh VNET(socket_hhh) 208 209 /* 210 * Limit on the number of connections in the listen queue waiting 211 * for accept(2). 212 * NB: The original sysctl somaxconn is still available but hidden 213 * to prevent confusion about the actual purpose of this number. 214 */ 215 static u_int somaxconn = SOMAXCONN; 216 217 static int 218 sysctl_somaxconn(SYSCTL_HANDLER_ARGS) 219 { 220 int error; 221 int val; 222 223 val = somaxconn; 224 error = sysctl_handle_int(oidp, &val, 0, req); 225 if (error || !req->newptr ) 226 return (error); 227 228 /* 229 * The purpose of the UINT_MAX / 3 limit, is so that the formula 230 * 3 * so_qlimit / 2 231 * below, will not overflow. 232 */ 233 234 if (val < 1 || val > UINT_MAX / 3) 235 return (EINVAL); 236 237 somaxconn = val; 238 return (0); 239 } 240 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, 241 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, sizeof(int), 242 sysctl_somaxconn, "I", 243 "Maximum listen socket pending connection accept queue size"); 244 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, 245 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, 0, 246 sizeof(int), sysctl_somaxconn, "I", 247 "Maximum listen socket pending connection accept queue size (compat)"); 248 249 static int numopensockets; 250 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD, 251 &numopensockets, 0, "Number of open sockets"); 252 253 /* 254 * accept_mtx locks down per-socket fields relating to accept queues. See 255 * socketvar.h for an annotation of the protected fields of struct socket. 256 */ 257 struct mtx accept_mtx; 258 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF); 259 260 /* 261 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket 262 * so_gencnt field. 263 */ 264 static struct mtx so_global_mtx; 265 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF); 266 267 /* 268 * General IPC sysctl name space, used by sockets and a variety of other IPC 269 * types. 270 */ 271 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 272 "IPC"); 273 274 /* 275 * Initialize the socket subsystem and set up the socket 276 * memory allocator. 277 */ 278 static uma_zone_t socket_zone; 279 int maxsockets; 280 281 static void 282 socket_zone_change(void *tag) 283 { 284 285 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 286 } 287 288 static void 289 socket_hhook_register(int subtype) 290 { 291 292 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype, 293 &V_socket_hhh[subtype], 294 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 295 printf("%s: WARNING: unable to register hook\n", __func__); 296 } 297 298 static void 299 socket_hhook_deregister(int subtype) 300 { 301 302 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0) 303 printf("%s: WARNING: unable to deregister hook\n", __func__); 304 } 305 306 static void 307 socket_init(void *tag) 308 { 309 310 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL, 311 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 312 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 313 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached"); 314 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL, 315 EVENTHANDLER_PRI_FIRST); 316 } 317 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL); 318 319 static void 320 socket_vnet_init(const void *unused __unused) 321 { 322 int i; 323 324 /* We expect a contiguous range */ 325 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 326 socket_hhook_register(i); 327 } 328 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 329 socket_vnet_init, NULL); 330 331 static void 332 socket_vnet_uninit(const void *unused __unused) 333 { 334 int i; 335 336 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 337 socket_hhook_deregister(i); 338 } 339 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 340 socket_vnet_uninit, NULL); 341 342 /* 343 * Initialise maxsockets. This SYSINIT must be run after 344 * tunable_mbinit(). 345 */ 346 static void 347 init_maxsockets(void *ignored) 348 { 349 350 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 351 maxsockets = imax(maxsockets, maxfiles); 352 } 353 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); 354 355 /* 356 * Sysctl to get and set the maximum global sockets limit. Notify protocols 357 * of the change so that they can update their dependent limits as required. 358 */ 359 static int 360 sysctl_maxsockets(SYSCTL_HANDLER_ARGS) 361 { 362 int error, newmaxsockets; 363 364 newmaxsockets = maxsockets; 365 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req); 366 if (error == 0 && req->newptr) { 367 if (newmaxsockets > maxsockets && 368 newmaxsockets <= maxfiles) { 369 maxsockets = newmaxsockets; 370 EVENTHANDLER_INVOKE(maxsockets_change); 371 } else 372 error = EINVAL; 373 } 374 return (error); 375 } 376 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, 377 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &maxsockets, 0, 378 sysctl_maxsockets, "IU", 379 "Maximum number of sockets available"); 380 381 /* 382 * Socket operation routines. These routines are called by the routines in 383 * sys_socket.c or from a system process, and implement the semantics of 384 * socket operations by switching out to the protocol specific routines. 385 */ 386 387 /* 388 * Get a socket structure from our zone, and initialize it. Note that it 389 * would probably be better to allocate socket and PCB at the same time, but 390 * I'm not convinced that all the protocols can be easily modified to do 391 * this. 392 * 393 * soalloc() returns a socket with a ref count of 0. 394 */ 395 static struct socket * 396 soalloc(struct vnet *vnet) 397 { 398 struct socket *so; 399 400 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO); 401 if (so == NULL) 402 return (NULL); 403 #ifdef MAC 404 if (mac_socket_init(so, M_NOWAIT) != 0) { 405 uma_zfree(socket_zone, so); 406 return (NULL); 407 } 408 #endif 409 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) { 410 uma_zfree(socket_zone, so); 411 return (NULL); 412 } 413 414 /* 415 * The socket locking protocol allows to lock 2 sockets at a time, 416 * however, the first one must be a listening socket. WITNESS lacks 417 * a feature to change class of an existing lock, so we use DUPOK. 418 */ 419 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK); 420 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd"); 421 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv"); 422 so->so_rcv.sb_sel = &so->so_rdsel; 423 so->so_snd.sb_sel = &so->so_wrsel; 424 sx_init(&so->so_snd.sb_sx, "so_snd_sx"); 425 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx"); 426 TAILQ_INIT(&so->so_snd.sb_aiojobq); 427 TAILQ_INIT(&so->so_rcv.sb_aiojobq); 428 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so); 429 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so); 430 #ifdef VIMAGE 431 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p", 432 __func__, __LINE__, so)); 433 so->so_vnet = vnet; 434 #endif 435 /* We shouldn't need the so_global_mtx */ 436 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) { 437 /* Do we need more comprehensive error returns? */ 438 uma_zfree(socket_zone, so); 439 return (NULL); 440 } 441 mtx_lock(&so_global_mtx); 442 so->so_gencnt = ++so_gencnt; 443 ++numopensockets; 444 #ifdef VIMAGE 445 vnet->vnet_sockcnt++; 446 #endif 447 mtx_unlock(&so_global_mtx); 448 449 return (so); 450 } 451 452 /* 453 * Free the storage associated with a socket at the socket layer, tear down 454 * locks, labels, etc. All protocol state is assumed already to have been 455 * torn down (and possibly never set up) by the caller. 456 */ 457 static void 458 sodealloc(struct socket *so) 459 { 460 461 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); 462 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL")); 463 464 mtx_lock(&so_global_mtx); 465 so->so_gencnt = ++so_gencnt; 466 --numopensockets; /* Could be below, but faster here. */ 467 #ifdef VIMAGE 468 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p", 469 __func__, __LINE__, so)); 470 so->so_vnet->vnet_sockcnt--; 471 #endif 472 mtx_unlock(&so_global_mtx); 473 #ifdef MAC 474 mac_socket_destroy(so); 475 #endif 476 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE); 477 478 khelp_destroy_osd(&so->osd); 479 if (SOLISTENING(so)) { 480 if (so->sol_accept_filter != NULL) 481 accept_filt_setopt(so, NULL); 482 } else { 483 if (so->so_rcv.sb_hiwat) 484 (void)chgsbsize(so->so_cred->cr_uidinfo, 485 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 486 if (so->so_snd.sb_hiwat) 487 (void)chgsbsize(so->so_cred->cr_uidinfo, 488 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 489 sx_destroy(&so->so_snd.sb_sx); 490 sx_destroy(&so->so_rcv.sb_sx); 491 SOCKBUF_LOCK_DESTROY(&so->so_snd); 492 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 493 } 494 crfree(so->so_cred); 495 mtx_destroy(&so->so_lock); 496 uma_zfree(socket_zone, so); 497 } 498 499 /* 500 * socreate returns a socket with a ref count of 1. The socket should be 501 * closed with soclose(). 502 */ 503 int 504 socreate(int dom, struct socket **aso, int type, int proto, 505 struct ucred *cred, struct thread *td) 506 { 507 struct protosw *prp; 508 struct socket *so; 509 int error; 510 511 if (proto) 512 prp = pffindproto(dom, proto, type); 513 else 514 prp = pffindtype(dom, type); 515 516 if (prp == NULL) { 517 /* No support for domain. */ 518 if (pffinddomain(dom) == NULL) 519 return (EAFNOSUPPORT); 520 /* No support for socket type. */ 521 if (proto == 0 && type != 0) 522 return (EPROTOTYPE); 523 return (EPROTONOSUPPORT); 524 } 525 if (prp->pr_usrreqs->pru_attach == NULL || 526 prp->pr_usrreqs->pru_attach == pru_attach_notsupp) 527 return (EPROTONOSUPPORT); 528 529 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0) 530 return (EPROTONOSUPPORT); 531 532 if (prp->pr_type != type) 533 return (EPROTOTYPE); 534 so = soalloc(CRED_TO_VNET(cred)); 535 if (so == NULL) 536 return (ENOBUFS); 537 538 so->so_type = type; 539 so->so_cred = crhold(cred); 540 if ((prp->pr_domain->dom_family == PF_INET) || 541 (prp->pr_domain->dom_family == PF_INET6) || 542 (prp->pr_domain->dom_family == PF_ROUTE)) 543 so->so_fibnum = td->td_proc->p_fibnum; 544 else 545 so->so_fibnum = 0; 546 so->so_proto = prp; 547 #ifdef MAC 548 mac_socket_create(cred, so); 549 #endif 550 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 551 so_rdknl_assert_lock); 552 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 553 so_wrknl_assert_lock); 554 /* 555 * Auto-sizing of socket buffers is managed by the protocols and 556 * the appropriate flags must be set in the pru_attach function. 557 */ 558 CURVNET_SET(so->so_vnet); 559 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); 560 CURVNET_RESTORE(); 561 if (error) { 562 sodealloc(so); 563 return (error); 564 } 565 soref(so); 566 *aso = so; 567 return (0); 568 } 569 570 #ifdef REGRESSION 571 static int regression_sonewconn_earlytest = 1; 572 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW, 573 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test"); 574 #endif 575 576 static struct timeval overinterval = { 60, 0 }; 577 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW, 578 &overinterval, 579 "Delay in seconds between warnings for listen socket overflows"); 580 581 /* 582 * When an attempt at a new connection is noted on a socket which accepts 583 * connections, sonewconn is called. If the connection is possible (subject 584 * to space constraints, etc.) then we allocate a new structure, properly 585 * linked into the data structure of the original socket, and return this. 586 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED. 587 * 588 * Note: the ref count on the socket is 0 on return. 589 */ 590 struct socket * 591 sonewconn(struct socket *head, int connstatus) 592 { 593 struct sbuf descrsb; 594 struct socket *so; 595 int len, overcount; 596 u_int qlen; 597 const char localprefix[] = "local:"; 598 char descrbuf[SUNPATHLEN + sizeof(localprefix)]; 599 #if defined(INET6) 600 char addrbuf[INET6_ADDRSTRLEN]; 601 #elif defined(INET) 602 char addrbuf[INET_ADDRSTRLEN]; 603 #endif 604 bool dolog, over; 605 606 SOLISTEN_LOCK(head); 607 over = (head->sol_qlen > 3 * head->sol_qlimit / 2); 608 #ifdef REGRESSION 609 if (regression_sonewconn_earlytest && over) { 610 #else 611 if (over) { 612 #endif 613 head->sol_overcount++; 614 dolog = !!ratecheck(&head->sol_lastover, &overinterval); 615 616 /* 617 * If we're going to log, copy the overflow count and queue 618 * length from the listen socket before dropping the lock. 619 * Also, reset the overflow count. 620 */ 621 if (dolog) { 622 overcount = head->sol_overcount; 623 head->sol_overcount = 0; 624 qlen = head->sol_qlen; 625 } 626 SOLISTEN_UNLOCK(head); 627 628 if (dolog) { 629 /* 630 * Try to print something descriptive about the 631 * socket for the error message. 632 */ 633 sbuf_new(&descrsb, descrbuf, sizeof(descrbuf), 634 SBUF_FIXEDLEN); 635 switch (head->so_proto->pr_domain->dom_family) { 636 #if defined(INET) || defined(INET6) 637 #ifdef INET 638 case AF_INET: 639 #endif 640 #ifdef INET6 641 case AF_INET6: 642 if (head->so_proto->pr_domain->dom_family == 643 AF_INET6 || 644 (sotoinpcb(head)->inp_inc.inc_flags & 645 INC_ISIPV6)) { 646 ip6_sprintf(addrbuf, 647 &sotoinpcb(head)->inp_inc.inc6_laddr); 648 sbuf_printf(&descrsb, "[%s]", addrbuf); 649 } else 650 #endif 651 { 652 #ifdef INET 653 inet_ntoa_r( 654 sotoinpcb(head)->inp_inc.inc_laddr, 655 addrbuf); 656 sbuf_cat(&descrsb, addrbuf); 657 #endif 658 } 659 sbuf_printf(&descrsb, ":%hu (proto %u)", 660 ntohs(sotoinpcb(head)->inp_inc.inc_lport), 661 head->so_proto->pr_protocol); 662 break; 663 #endif /* INET || INET6 */ 664 case AF_UNIX: 665 sbuf_cat(&descrsb, localprefix); 666 if (sotounpcb(head)->unp_addr != NULL) 667 len = 668 sotounpcb(head)->unp_addr->sun_len - 669 offsetof(struct sockaddr_un, 670 sun_path); 671 else 672 len = 0; 673 if (len > 0) 674 sbuf_bcat(&descrsb, 675 sotounpcb(head)->unp_addr->sun_path, 676 len); 677 else 678 sbuf_cat(&descrsb, "(unknown)"); 679 break; 680 } 681 682 /* 683 * If we can't print something more specific, at least 684 * print the domain name. 685 */ 686 if (sbuf_finish(&descrsb) != 0 || 687 sbuf_len(&descrsb) <= 0) { 688 sbuf_clear(&descrsb); 689 sbuf_cat(&descrsb, 690 head->so_proto->pr_domain->dom_name ?: 691 "unknown"); 692 sbuf_finish(&descrsb); 693 } 694 KASSERT(sbuf_len(&descrsb) > 0, 695 ("%s: sbuf creation failed", __func__)); 696 log(LOG_DEBUG, 697 "%s: pcb %p (%s): Listen queue overflow: " 698 "%i already in queue awaiting acceptance " 699 "(%d occurrences)\n", 700 __func__, head->so_pcb, sbuf_data(&descrsb), 701 qlen, overcount); 702 sbuf_delete(&descrsb); 703 704 overcount = 0; 705 } 706 707 return (NULL); 708 } 709 SOLISTEN_UNLOCK(head); 710 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL", 711 __func__, head)); 712 so = soalloc(head->so_vnet); 713 if (so == NULL) { 714 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 715 "limit reached or out of memory\n", 716 __func__, head->so_pcb); 717 return (NULL); 718 } 719 so->so_listen = head; 720 so->so_type = head->so_type; 721 so->so_options = head->so_options & ~SO_ACCEPTCONN; 722 so->so_linger = head->so_linger; 723 so->so_state = head->so_state | SS_NOFDREF; 724 so->so_fibnum = head->so_fibnum; 725 so->so_proto = head->so_proto; 726 so->so_cred = crhold(head->so_cred); 727 #ifdef MAC 728 mac_socket_newconn(head, so); 729 #endif 730 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 731 so_rdknl_assert_lock); 732 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 733 so_wrknl_assert_lock); 734 VNET_SO_ASSERT(head); 735 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) { 736 sodealloc(so); 737 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 738 __func__, head->so_pcb); 739 return (NULL); 740 } 741 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 742 sodealloc(so); 743 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 744 __func__, head->so_pcb); 745 return (NULL); 746 } 747 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat; 748 so->so_snd.sb_lowat = head->sol_sbsnd_lowat; 749 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo; 750 so->so_snd.sb_timeo = head->sol_sbsnd_timeo; 751 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE; 752 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE; 753 754 SOLISTEN_LOCK(head); 755 if (head->sol_accept_filter != NULL) 756 connstatus = 0; 757 so->so_state |= connstatus; 758 soref(head); /* A socket on (in)complete queue refs head. */ 759 if (connstatus) { 760 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 761 so->so_qstate = SQ_COMP; 762 head->sol_qlen++; 763 solisten_wakeup(head); /* unlocks */ 764 } else { 765 /* 766 * Keep removing sockets from the head until there's room for 767 * us to insert on the tail. In pre-locking revisions, this 768 * was a simple if(), but as we could be racing with other 769 * threads and soabort() requires dropping locks, we must 770 * loop waiting for the condition to be true. 771 */ 772 while (head->sol_incqlen > head->sol_qlimit) { 773 struct socket *sp; 774 775 sp = TAILQ_FIRST(&head->sol_incomp); 776 TAILQ_REMOVE(&head->sol_incomp, sp, so_list); 777 head->sol_incqlen--; 778 SOCK_LOCK(sp); 779 sp->so_qstate = SQ_NONE; 780 sp->so_listen = NULL; 781 SOCK_UNLOCK(sp); 782 sorele(head); /* does SOLISTEN_UNLOCK, head stays */ 783 soabort(sp); 784 SOLISTEN_LOCK(head); 785 } 786 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list); 787 so->so_qstate = SQ_INCOMP; 788 head->sol_incqlen++; 789 SOLISTEN_UNLOCK(head); 790 } 791 return (so); 792 } 793 794 #if defined(SCTP) || defined(SCTP_SUPPORT) 795 /* 796 * Socket part of sctp_peeloff(). Detach a new socket from an 797 * association. The new socket is returned with a reference. 798 */ 799 struct socket * 800 sopeeloff(struct socket *head) 801 { 802 struct socket *so; 803 804 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p", 805 __func__, __LINE__, head)); 806 so = soalloc(head->so_vnet); 807 if (so == NULL) { 808 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 809 "limit reached or out of memory\n", 810 __func__, head->so_pcb); 811 return (NULL); 812 } 813 so->so_type = head->so_type; 814 so->so_options = head->so_options; 815 so->so_linger = head->so_linger; 816 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED; 817 so->so_fibnum = head->so_fibnum; 818 so->so_proto = head->so_proto; 819 so->so_cred = crhold(head->so_cred); 820 #ifdef MAC 821 mac_socket_newconn(head, so); 822 #endif 823 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 824 so_rdknl_assert_lock); 825 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 826 so_wrknl_assert_lock); 827 VNET_SO_ASSERT(head); 828 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { 829 sodealloc(so); 830 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 831 __func__, head->so_pcb); 832 return (NULL); 833 } 834 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 835 sodealloc(so); 836 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 837 __func__, head->so_pcb); 838 return (NULL); 839 } 840 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 841 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 842 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 843 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 844 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; 845 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; 846 847 soref(so); 848 849 return (so); 850 } 851 #endif /* SCTP */ 852 853 int 854 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 855 { 856 int error; 857 858 CURVNET_SET(so->so_vnet); 859 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td); 860 CURVNET_RESTORE(); 861 return (error); 862 } 863 864 int 865 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 866 { 867 int error; 868 869 CURVNET_SET(so->so_vnet); 870 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td); 871 CURVNET_RESTORE(); 872 return (error); 873 } 874 875 /* 876 * solisten() transitions a socket from a non-listening state to a listening 877 * state, but can also be used to update the listen queue depth on an 878 * existing listen socket. The protocol will call back into the sockets 879 * layer using solisten_proto_check() and solisten_proto() to check and set 880 * socket-layer listen state. Call backs are used so that the protocol can 881 * acquire both protocol and socket layer locks in whatever order is required 882 * by the protocol. 883 * 884 * Protocol implementors are advised to hold the socket lock across the 885 * socket-layer test and set to avoid races at the socket layer. 886 */ 887 int 888 solisten(struct socket *so, int backlog, struct thread *td) 889 { 890 int error; 891 892 CURVNET_SET(so->so_vnet); 893 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td); 894 CURVNET_RESTORE(); 895 return (error); 896 } 897 898 int 899 solisten_proto_check(struct socket *so) 900 { 901 902 SOCK_LOCK_ASSERT(so); 903 904 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 905 SS_ISDISCONNECTING)) 906 return (EINVAL); 907 return (0); 908 } 909 910 void 911 solisten_proto(struct socket *so, int backlog) 912 { 913 int sbrcv_lowat, sbsnd_lowat; 914 u_int sbrcv_hiwat, sbsnd_hiwat; 915 short sbrcv_flags, sbsnd_flags; 916 sbintime_t sbrcv_timeo, sbsnd_timeo; 917 918 SOCK_LOCK_ASSERT(so); 919 920 if (SOLISTENING(so)) 921 goto listening; 922 923 /* 924 * Change this socket to listening state. 925 */ 926 sbrcv_lowat = so->so_rcv.sb_lowat; 927 sbsnd_lowat = so->so_snd.sb_lowat; 928 sbrcv_hiwat = so->so_rcv.sb_hiwat; 929 sbsnd_hiwat = so->so_snd.sb_hiwat; 930 sbrcv_flags = so->so_rcv.sb_flags; 931 sbsnd_flags = so->so_snd.sb_flags; 932 sbrcv_timeo = so->so_rcv.sb_timeo; 933 sbsnd_timeo = so->so_snd.sb_timeo; 934 935 sbdestroy(&so->so_snd, so); 936 sbdestroy(&so->so_rcv, so); 937 sx_destroy(&so->so_snd.sb_sx); 938 sx_destroy(&so->so_rcv.sb_sx); 939 SOCKBUF_LOCK_DESTROY(&so->so_snd); 940 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 941 942 #ifdef INVARIANTS 943 bzero(&so->so_rcv, 944 sizeof(struct socket) - offsetof(struct socket, so_rcv)); 945 #endif 946 947 so->sol_sbrcv_lowat = sbrcv_lowat; 948 so->sol_sbsnd_lowat = sbsnd_lowat; 949 so->sol_sbrcv_hiwat = sbrcv_hiwat; 950 so->sol_sbsnd_hiwat = sbsnd_hiwat; 951 so->sol_sbrcv_flags = sbrcv_flags; 952 so->sol_sbsnd_flags = sbsnd_flags; 953 so->sol_sbrcv_timeo = sbrcv_timeo; 954 so->sol_sbsnd_timeo = sbsnd_timeo; 955 956 so->sol_qlen = so->sol_incqlen = 0; 957 TAILQ_INIT(&so->sol_incomp); 958 TAILQ_INIT(&so->sol_comp); 959 960 so->sol_accept_filter = NULL; 961 so->sol_accept_filter_arg = NULL; 962 so->sol_accept_filter_str = NULL; 963 964 so->sol_upcall = NULL; 965 so->sol_upcallarg = NULL; 966 967 so->so_options |= SO_ACCEPTCONN; 968 969 listening: 970 if (backlog < 0 || backlog > somaxconn) 971 backlog = somaxconn; 972 so->sol_qlimit = backlog; 973 } 974 975 /* 976 * Wakeup listeners/subsystems once we have a complete connection. 977 * Enters with lock, returns unlocked. 978 */ 979 void 980 solisten_wakeup(struct socket *sol) 981 { 982 983 if (sol->sol_upcall != NULL) 984 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT); 985 else { 986 selwakeuppri(&sol->so_rdsel, PSOCK); 987 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0); 988 } 989 SOLISTEN_UNLOCK(sol); 990 wakeup_one(&sol->sol_comp); 991 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL) 992 pgsigio(&sol->so_sigio, SIGIO, 0); 993 } 994 995 /* 996 * Return single connection off a listening socket queue. Main consumer of 997 * the function is kern_accept4(). Some modules, that do their own accept 998 * management also use the function. 999 * 1000 * Listening socket must be locked on entry and is returned unlocked on 1001 * return. 1002 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT. 1003 */ 1004 int 1005 solisten_dequeue(struct socket *head, struct socket **ret, int flags) 1006 { 1007 struct socket *so; 1008 int error; 1009 1010 SOLISTEN_LOCK_ASSERT(head); 1011 1012 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) && 1013 head->so_error == 0) { 1014 error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH, 1015 "accept", 0); 1016 if (error != 0) { 1017 SOLISTEN_UNLOCK(head); 1018 return (error); 1019 } 1020 } 1021 if (head->so_error) { 1022 error = head->so_error; 1023 head->so_error = 0; 1024 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp)) 1025 error = EWOULDBLOCK; 1026 else 1027 error = 0; 1028 if (error) { 1029 SOLISTEN_UNLOCK(head); 1030 return (error); 1031 } 1032 so = TAILQ_FIRST(&head->sol_comp); 1033 SOCK_LOCK(so); 1034 KASSERT(so->so_qstate == SQ_COMP, 1035 ("%s: so %p not SQ_COMP", __func__, so)); 1036 soref(so); 1037 head->sol_qlen--; 1038 so->so_qstate = SQ_NONE; 1039 so->so_listen = NULL; 1040 TAILQ_REMOVE(&head->sol_comp, so, so_list); 1041 if (flags & ACCEPT4_INHERIT) 1042 so->so_state |= (head->so_state & SS_NBIO); 1043 else 1044 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0; 1045 SOCK_UNLOCK(so); 1046 sorele(head); 1047 1048 *ret = so; 1049 return (0); 1050 } 1051 1052 /* 1053 * Evaluate the reference count and named references on a socket; if no 1054 * references remain, free it. This should be called whenever a reference is 1055 * released, such as in sorele(), but also when named reference flags are 1056 * cleared in socket or protocol code. 1057 * 1058 * sofree() will free the socket if: 1059 * 1060 * - There are no outstanding file descriptor references or related consumers 1061 * (so_count == 0). 1062 * 1063 * - The socket has been closed by user space, if ever open (SS_NOFDREF). 1064 * 1065 * - The protocol does not have an outstanding strong reference on the socket 1066 * (SS_PROTOREF). 1067 * 1068 * - The socket is not in a completed connection queue, so a process has been 1069 * notified that it is present. If it is removed, the user process may 1070 * block in accept() despite select() saying the socket was ready. 1071 */ 1072 void 1073 sofree(struct socket *so) 1074 { 1075 struct protosw *pr = so->so_proto; 1076 1077 SOCK_LOCK_ASSERT(so); 1078 1079 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 || 1080 (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) { 1081 SOCK_UNLOCK(so); 1082 return; 1083 } 1084 1085 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) { 1086 struct socket *sol; 1087 1088 sol = so->so_listen; 1089 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so)); 1090 1091 /* 1092 * To solve race between close of a listening socket and 1093 * a socket on its incomplete queue, we need to lock both. 1094 * The order is first listening socket, then regular. 1095 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this 1096 * function and the listening socket are the only pointers 1097 * to so. To preserve so and sol, we reference both and then 1098 * relock. 1099 * After relock the socket may not move to so_comp since it 1100 * doesn't have PCB already, but it may be removed from 1101 * so_incomp. If that happens, we share responsiblity on 1102 * freeing the socket, but soclose() has already removed 1103 * it from queue. 1104 */ 1105 soref(sol); 1106 soref(so); 1107 SOCK_UNLOCK(so); 1108 SOLISTEN_LOCK(sol); 1109 SOCK_LOCK(so); 1110 if (so->so_qstate == SQ_INCOMP) { 1111 KASSERT(so->so_listen == sol, 1112 ("%s: so %p migrated out of sol %p", 1113 __func__, so, sol)); 1114 TAILQ_REMOVE(&sol->sol_incomp, so, so_list); 1115 sol->sol_incqlen--; 1116 /* This is guarenteed not to be the last. */ 1117 refcount_release(&sol->so_count); 1118 so->so_qstate = SQ_NONE; 1119 so->so_listen = NULL; 1120 } else 1121 KASSERT(so->so_listen == NULL, 1122 ("%s: so %p not on (in)comp with so_listen", 1123 __func__, so)); 1124 sorele(sol); 1125 KASSERT(so->so_count == 1, 1126 ("%s: so %p count %u", __func__, so, so->so_count)); 1127 so->so_count = 0; 1128 } 1129 if (SOLISTENING(so)) 1130 so->so_error = ECONNABORTED; 1131 SOCK_UNLOCK(so); 1132 1133 if (so->so_dtor != NULL) 1134 so->so_dtor(so); 1135 1136 VNET_SO_ASSERT(so); 1137 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 1138 (*pr->pr_domain->dom_dispose)(so); 1139 if (pr->pr_usrreqs->pru_detach != NULL) 1140 (*pr->pr_usrreqs->pru_detach)(so); 1141 1142 /* 1143 * From this point on, we assume that no other references to this 1144 * socket exist anywhere else in the stack. Therefore, no locks need 1145 * to be acquired or held. 1146 * 1147 * We used to do a lot of socket buffer and socket locking here, as 1148 * well as invoke sorflush() and perform wakeups. The direct call to 1149 * dom_dispose() and sbdestroy() are an inlining of what was 1150 * necessary from sorflush(). 1151 * 1152 * Notice that the socket buffer and kqueue state are torn down 1153 * before calling pru_detach. This means that protocols shold not 1154 * assume they can perform socket wakeups, etc, in their detach code. 1155 */ 1156 if (!SOLISTENING(so)) { 1157 sbdestroy(&so->so_snd, so); 1158 sbdestroy(&so->so_rcv, so); 1159 } 1160 seldrain(&so->so_rdsel); 1161 seldrain(&so->so_wrsel); 1162 knlist_destroy(&so->so_rdsel.si_note); 1163 knlist_destroy(&so->so_wrsel.si_note); 1164 sodealloc(so); 1165 } 1166 1167 /* 1168 * Close a socket on last file table reference removal. Initiate disconnect 1169 * if connected. Free socket when disconnect complete. 1170 * 1171 * This function will sorele() the socket. Note that soclose() may be called 1172 * prior to the ref count reaching zero. The actual socket structure will 1173 * not be freed until the ref count reaches zero. 1174 */ 1175 int 1176 soclose(struct socket *so) 1177 { 1178 struct accept_queue lqueue; 1179 int error = 0; 1180 1181 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter")); 1182 1183 CURVNET_SET(so->so_vnet); 1184 funsetown(&so->so_sigio); 1185 if (so->so_state & SS_ISCONNECTED) { 1186 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 1187 error = sodisconnect(so); 1188 if (error) { 1189 if (error == ENOTCONN) 1190 error = 0; 1191 goto drop; 1192 } 1193 } 1194 1195 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) { 1196 if ((so->so_state & SS_ISDISCONNECTING) && 1197 (so->so_state & SS_NBIO)) 1198 goto drop; 1199 while (so->so_state & SS_ISCONNECTED) { 1200 error = tsleep(&so->so_timeo, 1201 PSOCK | PCATCH, "soclos", 1202 so->so_linger * hz); 1203 if (error) 1204 break; 1205 } 1206 } 1207 } 1208 1209 drop: 1210 if (so->so_proto->pr_usrreqs->pru_close != NULL) 1211 (*so->so_proto->pr_usrreqs->pru_close)(so); 1212 1213 SOCK_LOCK(so); 1214 if (SOLISTENING(so)) { 1215 struct socket *sp; 1216 1217 TAILQ_INIT(&lqueue); 1218 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list); 1219 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list); 1220 1221 so->sol_qlen = so->sol_incqlen = 0; 1222 1223 TAILQ_FOREACH(sp, &lqueue, so_list) { 1224 SOCK_LOCK(sp); 1225 sp->so_qstate = SQ_NONE; 1226 sp->so_listen = NULL; 1227 SOCK_UNLOCK(sp); 1228 /* Guaranteed not to be the last. */ 1229 refcount_release(&so->so_count); 1230 } 1231 } 1232 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF")); 1233 so->so_state |= SS_NOFDREF; 1234 sorele(so); 1235 if (SOLISTENING(so)) { 1236 struct socket *sp, *tsp; 1237 1238 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) { 1239 SOCK_LOCK(sp); 1240 if (sp->so_count == 0) { 1241 SOCK_UNLOCK(sp); 1242 soabort(sp); 1243 } else 1244 /* sp is now in sofree() */ 1245 SOCK_UNLOCK(sp); 1246 } 1247 } 1248 CURVNET_RESTORE(); 1249 return (error); 1250 } 1251 1252 /* 1253 * soabort() is used to abruptly tear down a connection, such as when a 1254 * resource limit is reached (listen queue depth exceeded), or if a listen 1255 * socket is closed while there are sockets waiting to be accepted. 1256 * 1257 * This interface is tricky, because it is called on an unreferenced socket, 1258 * and must be called only by a thread that has actually removed the socket 1259 * from the listen queue it was on, or races with other threads are risked. 1260 * 1261 * This interface will call into the protocol code, so must not be called 1262 * with any socket locks held. Protocols do call it while holding their own 1263 * recursible protocol mutexes, but this is something that should be subject 1264 * to review in the future. 1265 */ 1266 void 1267 soabort(struct socket *so) 1268 { 1269 1270 /* 1271 * In as much as is possible, assert that no references to this 1272 * socket are held. This is not quite the same as asserting that the 1273 * current thread is responsible for arranging for no references, but 1274 * is as close as we can get for now. 1275 */ 1276 KASSERT(so->so_count == 0, ("soabort: so_count")); 1277 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF")); 1278 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF")); 1279 VNET_SO_ASSERT(so); 1280 1281 if (so->so_proto->pr_usrreqs->pru_abort != NULL) 1282 (*so->so_proto->pr_usrreqs->pru_abort)(so); 1283 SOCK_LOCK(so); 1284 sofree(so); 1285 } 1286 1287 int 1288 soaccept(struct socket *so, struct sockaddr **nam) 1289 { 1290 int error; 1291 1292 SOCK_LOCK(so); 1293 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF")); 1294 so->so_state &= ~SS_NOFDREF; 1295 SOCK_UNLOCK(so); 1296 1297 CURVNET_SET(so->so_vnet); 1298 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); 1299 CURVNET_RESTORE(); 1300 return (error); 1301 } 1302 1303 int 1304 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 1305 { 1306 1307 return (soconnectat(AT_FDCWD, so, nam, td)); 1308 } 1309 1310 int 1311 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 1312 { 1313 int error; 1314 1315 /* XXXMJ racy */ 1316 if (SOLISTENING(so)) 1317 return (EOPNOTSUPP); 1318 1319 CURVNET_SET(so->so_vnet); 1320 /* 1321 * If protocol is connection-based, can only connect once. 1322 * Otherwise, if connected, try to disconnect first. This allows 1323 * user to disconnect by connecting to, e.g., a null address. 1324 */ 1325 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 1326 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 1327 (error = sodisconnect(so)))) { 1328 error = EISCONN; 1329 } else { 1330 /* 1331 * Prevent accumulated error from previous connection from 1332 * biting us. 1333 */ 1334 so->so_error = 0; 1335 if (fd == AT_FDCWD) { 1336 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, 1337 nam, td); 1338 } else { 1339 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd, 1340 so, nam, td); 1341 } 1342 } 1343 CURVNET_RESTORE(); 1344 1345 return (error); 1346 } 1347 1348 int 1349 soconnect2(struct socket *so1, struct socket *so2) 1350 { 1351 int error; 1352 1353 CURVNET_SET(so1->so_vnet); 1354 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); 1355 CURVNET_RESTORE(); 1356 return (error); 1357 } 1358 1359 int 1360 sodisconnect(struct socket *so) 1361 { 1362 int error; 1363 1364 if ((so->so_state & SS_ISCONNECTED) == 0) 1365 return (ENOTCONN); 1366 if (so->so_state & SS_ISDISCONNECTING) 1367 return (EALREADY); 1368 VNET_SO_ASSERT(so); 1369 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 1370 return (error); 1371 } 1372 1373 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) 1374 1375 int 1376 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1377 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1378 { 1379 long space; 1380 ssize_t resid; 1381 int clen = 0, error, dontroute; 1382 1383 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 1384 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 1385 ("sosend_dgram: !PR_ATOMIC")); 1386 1387 if (uio != NULL) 1388 resid = uio->uio_resid; 1389 else 1390 resid = top->m_pkthdr.len; 1391 /* 1392 * In theory resid should be unsigned. However, space must be 1393 * signed, as it might be less than 0 if we over-committed, and we 1394 * must use a signed comparison of space and resid. On the other 1395 * hand, a negative resid causes us to loop sending 0-length 1396 * segments to the protocol. 1397 */ 1398 if (resid < 0) { 1399 error = EINVAL; 1400 goto out; 1401 } 1402 1403 dontroute = 1404 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 1405 if (td != NULL) 1406 td->td_ru.ru_msgsnd++; 1407 if (control != NULL) 1408 clen = control->m_len; 1409 1410 SOCKBUF_LOCK(&so->so_snd); 1411 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1412 SOCKBUF_UNLOCK(&so->so_snd); 1413 error = EPIPE; 1414 goto out; 1415 } 1416 if (so->so_error) { 1417 error = so->so_error; 1418 so->so_error = 0; 1419 SOCKBUF_UNLOCK(&so->so_snd); 1420 goto out; 1421 } 1422 if ((so->so_state & SS_ISCONNECTED) == 0) { 1423 /* 1424 * `sendto' and `sendmsg' is allowed on a connection-based 1425 * socket if it supports implied connect. Return ENOTCONN if 1426 * not connected and no address is supplied. 1427 */ 1428 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1429 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1430 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1431 !(resid == 0 && clen != 0)) { 1432 SOCKBUF_UNLOCK(&so->so_snd); 1433 error = ENOTCONN; 1434 goto out; 1435 } 1436 } else if (addr == NULL) { 1437 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1438 error = ENOTCONN; 1439 else 1440 error = EDESTADDRREQ; 1441 SOCKBUF_UNLOCK(&so->so_snd); 1442 goto out; 1443 } 1444 } 1445 1446 /* 1447 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1448 * problem and need fixing. 1449 */ 1450 space = sbspace(&so->so_snd); 1451 if (flags & MSG_OOB) 1452 space += 1024; 1453 space -= clen; 1454 SOCKBUF_UNLOCK(&so->so_snd); 1455 if (resid > space) { 1456 error = EMSGSIZE; 1457 goto out; 1458 } 1459 if (uio == NULL) { 1460 resid = 0; 1461 if (flags & MSG_EOR) 1462 top->m_flags |= M_EOR; 1463 } else { 1464 /* 1465 * Copy the data from userland into a mbuf chain. 1466 * If no data is to be copied in, a single empty mbuf 1467 * is returned. 1468 */ 1469 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1470 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1471 if (top == NULL) { 1472 error = EFAULT; /* only possible error */ 1473 goto out; 1474 } 1475 space -= resid - uio->uio_resid; 1476 resid = uio->uio_resid; 1477 } 1478 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1479 /* 1480 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1481 * than with. 1482 */ 1483 if (dontroute) { 1484 SOCK_LOCK(so); 1485 so->so_options |= SO_DONTROUTE; 1486 SOCK_UNLOCK(so); 1487 } 1488 /* 1489 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1490 * of date. We could have received a reset packet in an interrupt or 1491 * maybe we slept while doing page faults in uiomove() etc. We could 1492 * probably recheck again inside the locking protection here, but 1493 * there are probably other places that this also happens. We must 1494 * rethink this. 1495 */ 1496 VNET_SO_ASSERT(so); 1497 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1498 (flags & MSG_OOB) ? PRUS_OOB : 1499 /* 1500 * If the user set MSG_EOF, the protocol understands this flag and 1501 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1502 */ 1503 ((flags & MSG_EOF) && 1504 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1505 (resid <= 0)) ? 1506 PRUS_EOF : 1507 /* If there is more to send set PRUS_MORETOCOME */ 1508 (flags & MSG_MORETOCOME) || 1509 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1510 top, addr, control, td); 1511 if (dontroute) { 1512 SOCK_LOCK(so); 1513 so->so_options &= ~SO_DONTROUTE; 1514 SOCK_UNLOCK(so); 1515 } 1516 clen = 0; 1517 control = NULL; 1518 top = NULL; 1519 out: 1520 if (top != NULL) 1521 m_freem(top); 1522 if (control != NULL) 1523 m_freem(control); 1524 return (error); 1525 } 1526 1527 /* 1528 * Send on a socket. If send must go all at once and message is larger than 1529 * send buffering, then hard error. Lock against other senders. If must go 1530 * all at once and not enough room now, then inform user that this would 1531 * block and do nothing. Otherwise, if nonblocking, send as much as 1532 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1533 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1534 * in mbuf chain must be small enough to send all at once. 1535 * 1536 * Returns nonzero on error, timeout or signal; callers must check for short 1537 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1538 * on return. 1539 */ 1540 int 1541 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1542 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1543 { 1544 long space; 1545 ssize_t resid; 1546 int clen = 0, error, dontroute; 1547 int atomic = sosendallatonce(so) || top; 1548 int pru_flag; 1549 #ifdef KERN_TLS 1550 struct ktls_session *tls; 1551 int tls_enq_cnt, tls_pruflag; 1552 uint8_t tls_rtype; 1553 1554 tls = NULL; 1555 tls_rtype = TLS_RLTYPE_APP; 1556 #endif 1557 if (uio != NULL) 1558 resid = uio->uio_resid; 1559 else if ((top->m_flags & M_PKTHDR) != 0) 1560 resid = top->m_pkthdr.len; 1561 else 1562 resid = m_length(top, NULL); 1563 /* 1564 * In theory resid should be unsigned. However, space must be 1565 * signed, as it might be less than 0 if we over-committed, and we 1566 * must use a signed comparison of space and resid. On the other 1567 * hand, a negative resid causes us to loop sending 0-length 1568 * segments to the protocol. 1569 * 1570 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1571 * type sockets since that's an error. 1572 */ 1573 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1574 error = EINVAL; 1575 goto out; 1576 } 1577 1578 dontroute = 1579 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1580 (so->so_proto->pr_flags & PR_ATOMIC); 1581 if (td != NULL) 1582 td->td_ru.ru_msgsnd++; 1583 if (control != NULL) 1584 clen = control->m_len; 1585 1586 error = sblock(&so->so_snd, SBLOCKWAIT(flags)); 1587 if (error) 1588 goto out; 1589 1590 #ifdef KERN_TLS 1591 tls_pruflag = 0; 1592 tls = ktls_hold(so->so_snd.sb_tls_info); 1593 if (tls != NULL) { 1594 if (tls->mode == TCP_TLS_MODE_SW) 1595 tls_pruflag = PRUS_NOTREADY; 1596 1597 if (control != NULL) { 1598 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1599 1600 if (clen >= sizeof(*cm) && 1601 cm->cmsg_type == TLS_SET_RECORD_TYPE) { 1602 tls_rtype = *((uint8_t *)CMSG_DATA(cm)); 1603 clen = 0; 1604 m_freem(control); 1605 control = NULL; 1606 atomic = 1; 1607 } 1608 } 1609 } 1610 #endif 1611 1612 restart: 1613 do { 1614 SOCKBUF_LOCK(&so->so_snd); 1615 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1616 SOCKBUF_UNLOCK(&so->so_snd); 1617 error = EPIPE; 1618 goto release; 1619 } 1620 if (so->so_error) { 1621 error = so->so_error; 1622 so->so_error = 0; 1623 SOCKBUF_UNLOCK(&so->so_snd); 1624 goto release; 1625 } 1626 if ((so->so_state & SS_ISCONNECTED) == 0) { 1627 /* 1628 * `sendto' and `sendmsg' is allowed on a connection- 1629 * based socket if it supports implied connect. 1630 * Return ENOTCONN if not connected and no address is 1631 * supplied. 1632 */ 1633 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1634 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1635 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1636 !(resid == 0 && clen != 0)) { 1637 SOCKBUF_UNLOCK(&so->so_snd); 1638 error = ENOTCONN; 1639 goto release; 1640 } 1641 } else if (addr == NULL) { 1642 SOCKBUF_UNLOCK(&so->so_snd); 1643 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1644 error = ENOTCONN; 1645 else 1646 error = EDESTADDRREQ; 1647 goto release; 1648 } 1649 } 1650 space = sbspace(&so->so_snd); 1651 if (flags & MSG_OOB) 1652 space += 1024; 1653 if ((atomic && resid > so->so_snd.sb_hiwat) || 1654 clen > so->so_snd.sb_hiwat) { 1655 SOCKBUF_UNLOCK(&so->so_snd); 1656 error = EMSGSIZE; 1657 goto release; 1658 } 1659 if (space < resid + clen && 1660 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1661 if ((so->so_state & SS_NBIO) || 1662 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) { 1663 SOCKBUF_UNLOCK(&so->so_snd); 1664 error = EWOULDBLOCK; 1665 goto release; 1666 } 1667 error = sbwait(&so->so_snd); 1668 SOCKBUF_UNLOCK(&so->so_snd); 1669 if (error) 1670 goto release; 1671 goto restart; 1672 } 1673 SOCKBUF_UNLOCK(&so->so_snd); 1674 space -= clen; 1675 do { 1676 if (uio == NULL) { 1677 resid = 0; 1678 if (flags & MSG_EOR) 1679 top->m_flags |= M_EOR; 1680 #ifdef KERN_TLS 1681 if (tls != NULL) { 1682 ktls_frame(top, tls, &tls_enq_cnt, 1683 tls_rtype); 1684 tls_rtype = TLS_RLTYPE_APP; 1685 } 1686 #endif 1687 } else { 1688 /* 1689 * Copy the data from userland into a mbuf 1690 * chain. If resid is 0, which can happen 1691 * only if we have control to send, then 1692 * a single empty mbuf is returned. This 1693 * is a workaround to prevent protocol send 1694 * methods to panic. 1695 */ 1696 #ifdef KERN_TLS 1697 if (tls != NULL) { 1698 top = m_uiotombuf(uio, M_WAITOK, space, 1699 tls->params.max_frame_len, 1700 M_EXTPG | 1701 ((flags & MSG_EOR) ? M_EOR : 0)); 1702 if (top != NULL) { 1703 ktls_frame(top, tls, 1704 &tls_enq_cnt, tls_rtype); 1705 } 1706 tls_rtype = TLS_RLTYPE_APP; 1707 } else 1708 #endif 1709 top = m_uiotombuf(uio, M_WAITOK, space, 1710 (atomic ? max_hdr : 0), 1711 (atomic ? M_PKTHDR : 0) | 1712 ((flags & MSG_EOR) ? M_EOR : 0)); 1713 if (top == NULL) { 1714 error = EFAULT; /* only possible error */ 1715 goto release; 1716 } 1717 space -= resid - uio->uio_resid; 1718 resid = uio->uio_resid; 1719 } 1720 if (dontroute) { 1721 SOCK_LOCK(so); 1722 so->so_options |= SO_DONTROUTE; 1723 SOCK_UNLOCK(so); 1724 } 1725 /* 1726 * XXX all the SBS_CANTSENDMORE checks previously 1727 * done could be out of date. We could have received 1728 * a reset packet in an interrupt or maybe we slept 1729 * while doing page faults in uiomove() etc. We 1730 * could probably recheck again inside the locking 1731 * protection here, but there are probably other 1732 * places that this also happens. We must rethink 1733 * this. 1734 */ 1735 VNET_SO_ASSERT(so); 1736 1737 pru_flag = (flags & MSG_OOB) ? PRUS_OOB : 1738 /* 1739 * If the user set MSG_EOF, the protocol understands 1740 * this flag and nothing left to send then use 1741 * PRU_SEND_EOF instead of PRU_SEND. 1742 */ 1743 ((flags & MSG_EOF) && 1744 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1745 (resid <= 0)) ? 1746 PRUS_EOF : 1747 /* If there is more to send set PRUS_MORETOCOME. */ 1748 (flags & MSG_MORETOCOME) || 1749 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; 1750 1751 #ifdef KERN_TLS 1752 pru_flag |= tls_pruflag; 1753 #endif 1754 1755 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1756 pru_flag, top, addr, control, td); 1757 1758 if (dontroute) { 1759 SOCK_LOCK(so); 1760 so->so_options &= ~SO_DONTROUTE; 1761 SOCK_UNLOCK(so); 1762 } 1763 1764 #ifdef KERN_TLS 1765 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) { 1766 if (error != 0) { 1767 m_freem(top); 1768 top = NULL; 1769 } else { 1770 soref(so); 1771 ktls_enqueue(top, so, tls_enq_cnt); 1772 } 1773 } 1774 #endif 1775 clen = 0; 1776 control = NULL; 1777 top = NULL; 1778 if (error) 1779 goto release; 1780 } while (resid && space > 0); 1781 } while (resid); 1782 1783 release: 1784 sbunlock(&so->so_snd); 1785 out: 1786 #ifdef KERN_TLS 1787 if (tls != NULL) 1788 ktls_free(tls); 1789 #endif 1790 if (top != NULL) 1791 m_freem(top); 1792 if (control != NULL) 1793 m_freem(control); 1794 return (error); 1795 } 1796 1797 int 1798 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1799 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1800 { 1801 int error; 1802 1803 CURVNET_SET(so->so_vnet); 1804 if (!SOLISTENING(so)) 1805 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, 1806 top, control, flags, td); 1807 else { 1808 m_freem(top); 1809 m_freem(control); 1810 error = ENOTCONN; 1811 } 1812 CURVNET_RESTORE(); 1813 return (error); 1814 } 1815 1816 /* 1817 * The part of soreceive() that implements reading non-inline out-of-band 1818 * data from a socket. For more complete comments, see soreceive(), from 1819 * which this code originated. 1820 * 1821 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1822 * unable to return an mbuf chain to the caller. 1823 */ 1824 static int 1825 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1826 { 1827 struct protosw *pr = so->so_proto; 1828 struct mbuf *m; 1829 int error; 1830 1831 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1832 VNET_SO_ASSERT(so); 1833 1834 m = m_get(M_WAITOK, MT_DATA); 1835 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 1836 if (error) 1837 goto bad; 1838 do { 1839 error = uiomove(mtod(m, void *), 1840 (int) min(uio->uio_resid, m->m_len), uio); 1841 m = m_free(m); 1842 } while (uio->uio_resid && error == 0 && m); 1843 bad: 1844 if (m != NULL) 1845 m_freem(m); 1846 return (error); 1847 } 1848 1849 /* 1850 * Following replacement or removal of the first mbuf on the first mbuf chain 1851 * of a socket buffer, push necessary state changes back into the socket 1852 * buffer so that other consumers see the values consistently. 'nextrecord' 1853 * is the callers locally stored value of the original value of 1854 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1855 * NOTE: 'nextrecord' may be NULL. 1856 */ 1857 static __inline void 1858 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1859 { 1860 1861 SOCKBUF_LOCK_ASSERT(sb); 1862 /* 1863 * First, update for the new value of nextrecord. If necessary, make 1864 * it the first record. 1865 */ 1866 if (sb->sb_mb != NULL) 1867 sb->sb_mb->m_nextpkt = nextrecord; 1868 else 1869 sb->sb_mb = nextrecord; 1870 1871 /* 1872 * Now update any dependent socket buffer fields to reflect the new 1873 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1874 * addition of a second clause that takes care of the case where 1875 * sb_mb has been updated, but remains the last record. 1876 */ 1877 if (sb->sb_mb == NULL) { 1878 sb->sb_mbtail = NULL; 1879 sb->sb_lastrecord = NULL; 1880 } else if (sb->sb_mb->m_nextpkt == NULL) 1881 sb->sb_lastrecord = sb->sb_mb; 1882 } 1883 1884 /* 1885 * Implement receive operations on a socket. We depend on the way that 1886 * records are added to the sockbuf by sbappend. In particular, each record 1887 * (mbufs linked through m_next) must begin with an address if the protocol 1888 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1889 * data, and then zero or more mbufs of data. In order to allow parallelism 1890 * between network receive and copying to user space, as well as avoid 1891 * sleeping with a mutex held, we release the socket buffer mutex during the 1892 * user space copy. Although the sockbuf is locked, new data may still be 1893 * appended, and thus we must maintain consistency of the sockbuf during that 1894 * time. 1895 * 1896 * The caller may receive the data as a single mbuf chain by supplying an 1897 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1898 * the count in uio_resid. 1899 */ 1900 int 1901 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1902 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1903 { 1904 struct mbuf *m, **mp; 1905 int flags, error, offset; 1906 ssize_t len; 1907 struct protosw *pr = so->so_proto; 1908 struct mbuf *nextrecord; 1909 int moff, type = 0; 1910 ssize_t orig_resid = uio->uio_resid; 1911 1912 mp = mp0; 1913 if (psa != NULL) 1914 *psa = NULL; 1915 if (controlp != NULL) 1916 *controlp = NULL; 1917 if (flagsp != NULL) 1918 flags = *flagsp &~ MSG_EOR; 1919 else 1920 flags = 0; 1921 if (flags & MSG_OOB) 1922 return (soreceive_rcvoob(so, uio, flags)); 1923 if (mp != NULL) 1924 *mp = NULL; 1925 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 1926 && uio->uio_resid) { 1927 VNET_SO_ASSERT(so); 1928 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 1929 } 1930 1931 error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); 1932 if (error) 1933 return (error); 1934 1935 restart: 1936 SOCKBUF_LOCK(&so->so_rcv); 1937 m = so->so_rcv.sb_mb; 1938 /* 1939 * If we have less data than requested, block awaiting more (subject 1940 * to any timeout) if: 1941 * 1. the current count is less than the low water mark, or 1942 * 2. MSG_DONTWAIT is not set 1943 */ 1944 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1945 sbavail(&so->so_rcv) < uio->uio_resid) && 1946 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && 1947 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 1948 KASSERT(m != NULL || !sbavail(&so->so_rcv), 1949 ("receive: m == %p sbavail == %u", 1950 m, sbavail(&so->so_rcv))); 1951 if (so->so_error) { 1952 if (m != NULL) 1953 goto dontblock; 1954 error = so->so_error; 1955 if ((flags & MSG_PEEK) == 0) 1956 so->so_error = 0; 1957 SOCKBUF_UNLOCK(&so->so_rcv); 1958 goto release; 1959 } 1960 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1961 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1962 if (m != NULL) 1963 goto dontblock; 1964 #ifdef KERN_TLS 1965 else if (so->so_rcv.sb_tlsdcc == 0 && 1966 so->so_rcv.sb_tlscc == 0) { 1967 #else 1968 else { 1969 #endif 1970 SOCKBUF_UNLOCK(&so->so_rcv); 1971 goto release; 1972 } 1973 } 1974 for (; m != NULL; m = m->m_next) 1975 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1976 m = so->so_rcv.sb_mb; 1977 goto dontblock; 1978 } 1979 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED | 1980 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 && 1981 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 1982 SOCKBUF_UNLOCK(&so->so_rcv); 1983 error = ENOTCONN; 1984 goto release; 1985 } 1986 if (uio->uio_resid == 0) { 1987 SOCKBUF_UNLOCK(&so->so_rcv); 1988 goto release; 1989 } 1990 if ((so->so_state & SS_NBIO) || 1991 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1992 SOCKBUF_UNLOCK(&so->so_rcv); 1993 error = EWOULDBLOCK; 1994 goto release; 1995 } 1996 SBLASTRECORDCHK(&so->so_rcv); 1997 SBLASTMBUFCHK(&so->so_rcv); 1998 error = sbwait(&so->so_rcv); 1999 SOCKBUF_UNLOCK(&so->so_rcv); 2000 if (error) 2001 goto release; 2002 goto restart; 2003 } 2004 dontblock: 2005 /* 2006 * From this point onward, we maintain 'nextrecord' as a cache of the 2007 * pointer to the next record in the socket buffer. We must keep the 2008 * various socket buffer pointers and local stack versions of the 2009 * pointers in sync, pushing out modifications before dropping the 2010 * socket buffer mutex, and re-reading them when picking it up. 2011 * 2012 * Otherwise, we will race with the network stack appending new data 2013 * or records onto the socket buffer by using inconsistent/stale 2014 * versions of the field, possibly resulting in socket buffer 2015 * corruption. 2016 * 2017 * By holding the high-level sblock(), we prevent simultaneous 2018 * readers from pulling off the front of the socket buffer. 2019 */ 2020 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2021 if (uio->uio_td) 2022 uio->uio_td->td_ru.ru_msgrcv++; 2023 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 2024 SBLASTRECORDCHK(&so->so_rcv); 2025 SBLASTMBUFCHK(&so->so_rcv); 2026 nextrecord = m->m_nextpkt; 2027 if (pr->pr_flags & PR_ADDR) { 2028 KASSERT(m->m_type == MT_SONAME, 2029 ("m->m_type == %d", m->m_type)); 2030 orig_resid = 0; 2031 if (psa != NULL) 2032 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2033 M_NOWAIT); 2034 if (flags & MSG_PEEK) { 2035 m = m->m_next; 2036 } else { 2037 sbfree(&so->so_rcv, m); 2038 so->so_rcv.sb_mb = m_free(m); 2039 m = so->so_rcv.sb_mb; 2040 sockbuf_pushsync(&so->so_rcv, nextrecord); 2041 } 2042 } 2043 2044 /* 2045 * Process one or more MT_CONTROL mbufs present before any data mbufs 2046 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 2047 * just copy the data; if !MSG_PEEK, we call into the protocol to 2048 * perform externalization (or freeing if controlp == NULL). 2049 */ 2050 if (m != NULL && m->m_type == MT_CONTROL) { 2051 struct mbuf *cm = NULL, *cmn; 2052 struct mbuf **cme = &cm; 2053 #ifdef KERN_TLS 2054 struct cmsghdr *cmsg; 2055 struct tls_get_record tgr; 2056 2057 /* 2058 * For MSG_TLSAPPDATA, check for a non-application data 2059 * record. If found, return ENXIO without removing 2060 * it from the receive queue. This allows a subsequent 2061 * call without MSG_TLSAPPDATA to receive it. 2062 * Note that, for TLS, there should only be a single 2063 * control mbuf with the TLS_GET_RECORD message in it. 2064 */ 2065 if (flags & MSG_TLSAPPDATA) { 2066 cmsg = mtod(m, struct cmsghdr *); 2067 if (cmsg->cmsg_type == TLS_GET_RECORD && 2068 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) { 2069 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr)); 2070 /* This will need to change for TLS 1.3. */ 2071 if (tgr.tls_type != TLS_RLTYPE_APP) { 2072 SOCKBUF_UNLOCK(&so->so_rcv); 2073 error = ENXIO; 2074 goto release; 2075 } 2076 } 2077 } 2078 #endif 2079 2080 do { 2081 if (flags & MSG_PEEK) { 2082 if (controlp != NULL) { 2083 *controlp = m_copym(m, 0, m->m_len, 2084 M_NOWAIT); 2085 controlp = &(*controlp)->m_next; 2086 } 2087 m = m->m_next; 2088 } else { 2089 sbfree(&so->so_rcv, m); 2090 so->so_rcv.sb_mb = m->m_next; 2091 m->m_next = NULL; 2092 *cme = m; 2093 cme = &(*cme)->m_next; 2094 m = so->so_rcv.sb_mb; 2095 } 2096 } while (m != NULL && m->m_type == MT_CONTROL); 2097 if ((flags & MSG_PEEK) == 0) 2098 sockbuf_pushsync(&so->so_rcv, nextrecord); 2099 while (cm != NULL) { 2100 cmn = cm->m_next; 2101 cm->m_next = NULL; 2102 if (pr->pr_domain->dom_externalize != NULL) { 2103 SOCKBUF_UNLOCK(&so->so_rcv); 2104 VNET_SO_ASSERT(so); 2105 error = (*pr->pr_domain->dom_externalize) 2106 (cm, controlp, flags); 2107 SOCKBUF_LOCK(&so->so_rcv); 2108 } else if (controlp != NULL) 2109 *controlp = cm; 2110 else 2111 m_freem(cm); 2112 if (controlp != NULL) { 2113 orig_resid = 0; 2114 while (*controlp != NULL) 2115 controlp = &(*controlp)->m_next; 2116 } 2117 cm = cmn; 2118 } 2119 if (m != NULL) 2120 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 2121 else 2122 nextrecord = so->so_rcv.sb_mb; 2123 orig_resid = 0; 2124 } 2125 if (m != NULL) { 2126 if ((flags & MSG_PEEK) == 0) { 2127 KASSERT(m->m_nextpkt == nextrecord, 2128 ("soreceive: post-control, nextrecord !sync")); 2129 if (nextrecord == NULL) { 2130 KASSERT(so->so_rcv.sb_mb == m, 2131 ("soreceive: post-control, sb_mb!=m")); 2132 KASSERT(so->so_rcv.sb_lastrecord == m, 2133 ("soreceive: post-control, lastrecord!=m")); 2134 } 2135 } 2136 type = m->m_type; 2137 if (type == MT_OOBDATA) 2138 flags |= MSG_OOB; 2139 } else { 2140 if ((flags & MSG_PEEK) == 0) { 2141 KASSERT(so->so_rcv.sb_mb == nextrecord, 2142 ("soreceive: sb_mb != nextrecord")); 2143 if (so->so_rcv.sb_mb == NULL) { 2144 KASSERT(so->so_rcv.sb_lastrecord == NULL, 2145 ("soreceive: sb_lastercord != NULL")); 2146 } 2147 } 2148 } 2149 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2150 SBLASTRECORDCHK(&so->so_rcv); 2151 SBLASTMBUFCHK(&so->so_rcv); 2152 2153 /* 2154 * Now continue to read any data mbufs off of the head of the socket 2155 * buffer until the read request is satisfied. Note that 'type' is 2156 * used to store the type of any mbuf reads that have happened so far 2157 * such that soreceive() can stop reading if the type changes, which 2158 * causes soreceive() to return only one of regular data and inline 2159 * out-of-band data in a single socket receive operation. 2160 */ 2161 moff = 0; 2162 offset = 0; 2163 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 2164 && error == 0) { 2165 /* 2166 * If the type of mbuf has changed since the last mbuf 2167 * examined ('type'), end the receive operation. 2168 */ 2169 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2170 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 2171 if (type != m->m_type) 2172 break; 2173 } else if (type == MT_OOBDATA) 2174 break; 2175 else 2176 KASSERT(m->m_type == MT_DATA, 2177 ("m->m_type == %d", m->m_type)); 2178 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 2179 len = uio->uio_resid; 2180 if (so->so_oobmark && len > so->so_oobmark - offset) 2181 len = so->so_oobmark - offset; 2182 if (len > m->m_len - moff) 2183 len = m->m_len - moff; 2184 /* 2185 * If mp is set, just pass back the mbufs. Otherwise copy 2186 * them out via the uio, then free. Sockbuf must be 2187 * consistent here (points to current mbuf, it points to next 2188 * record) when we drop priority; we must note any additions 2189 * to the sockbuf when we block interrupts again. 2190 */ 2191 if (mp == NULL) { 2192 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2193 SBLASTRECORDCHK(&so->so_rcv); 2194 SBLASTMBUFCHK(&so->so_rcv); 2195 SOCKBUF_UNLOCK(&so->so_rcv); 2196 if ((m->m_flags & M_EXTPG) != 0) 2197 error = m_unmappedtouio(m, moff, uio, (int)len); 2198 else 2199 error = uiomove(mtod(m, char *) + moff, 2200 (int)len, uio); 2201 SOCKBUF_LOCK(&so->so_rcv); 2202 if (error) { 2203 /* 2204 * The MT_SONAME mbuf has already been removed 2205 * from the record, so it is necessary to 2206 * remove the data mbufs, if any, to preserve 2207 * the invariant in the case of PR_ADDR that 2208 * requires MT_SONAME mbufs at the head of 2209 * each record. 2210 */ 2211 if (pr->pr_flags & PR_ATOMIC && 2212 ((flags & MSG_PEEK) == 0)) 2213 (void)sbdroprecord_locked(&so->so_rcv); 2214 SOCKBUF_UNLOCK(&so->so_rcv); 2215 goto release; 2216 } 2217 } else 2218 uio->uio_resid -= len; 2219 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2220 if (len == m->m_len - moff) { 2221 if (m->m_flags & M_EOR) 2222 flags |= MSG_EOR; 2223 if (flags & MSG_PEEK) { 2224 m = m->m_next; 2225 moff = 0; 2226 } else { 2227 nextrecord = m->m_nextpkt; 2228 sbfree(&so->so_rcv, m); 2229 if (mp != NULL) { 2230 m->m_nextpkt = NULL; 2231 *mp = m; 2232 mp = &m->m_next; 2233 so->so_rcv.sb_mb = m = m->m_next; 2234 *mp = NULL; 2235 } else { 2236 so->so_rcv.sb_mb = m_free(m); 2237 m = so->so_rcv.sb_mb; 2238 } 2239 sockbuf_pushsync(&so->so_rcv, nextrecord); 2240 SBLASTRECORDCHK(&so->so_rcv); 2241 SBLASTMBUFCHK(&so->so_rcv); 2242 } 2243 } else { 2244 if (flags & MSG_PEEK) 2245 moff += len; 2246 else { 2247 if (mp != NULL) { 2248 if (flags & MSG_DONTWAIT) { 2249 *mp = m_copym(m, 0, len, 2250 M_NOWAIT); 2251 if (*mp == NULL) { 2252 /* 2253 * m_copym() couldn't 2254 * allocate an mbuf. 2255 * Adjust uio_resid back 2256 * (it was adjusted 2257 * down by len bytes, 2258 * which we didn't end 2259 * up "copying" over). 2260 */ 2261 uio->uio_resid += len; 2262 break; 2263 } 2264 } else { 2265 SOCKBUF_UNLOCK(&so->so_rcv); 2266 *mp = m_copym(m, 0, len, 2267 M_WAITOK); 2268 SOCKBUF_LOCK(&so->so_rcv); 2269 } 2270 } 2271 sbcut_locked(&so->so_rcv, len); 2272 } 2273 } 2274 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2275 if (so->so_oobmark) { 2276 if ((flags & MSG_PEEK) == 0) { 2277 so->so_oobmark -= len; 2278 if (so->so_oobmark == 0) { 2279 so->so_rcv.sb_state |= SBS_RCVATMARK; 2280 break; 2281 } 2282 } else { 2283 offset += len; 2284 if (offset == so->so_oobmark) 2285 break; 2286 } 2287 } 2288 if (flags & MSG_EOR) 2289 break; 2290 /* 2291 * If the MSG_WAITALL flag is set (for non-atomic socket), we 2292 * must not quit until "uio->uio_resid == 0" or an error 2293 * termination. If a signal/timeout occurs, return with a 2294 * short count but without error. Keep sockbuf locked 2295 * against other readers. 2296 */ 2297 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 2298 !sosendallatonce(so) && nextrecord == NULL) { 2299 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2300 if (so->so_error || 2301 so->so_rcv.sb_state & SBS_CANTRCVMORE) 2302 break; 2303 /* 2304 * Notify the protocol that some data has been 2305 * drained before blocking. 2306 */ 2307 if (pr->pr_flags & PR_WANTRCVD) { 2308 SOCKBUF_UNLOCK(&so->so_rcv); 2309 VNET_SO_ASSERT(so); 2310 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2311 SOCKBUF_LOCK(&so->so_rcv); 2312 } 2313 SBLASTRECORDCHK(&so->so_rcv); 2314 SBLASTMBUFCHK(&so->so_rcv); 2315 /* 2316 * We could receive some data while was notifying 2317 * the protocol. Skip blocking in this case. 2318 */ 2319 if (so->so_rcv.sb_mb == NULL) { 2320 error = sbwait(&so->so_rcv); 2321 if (error) { 2322 SOCKBUF_UNLOCK(&so->so_rcv); 2323 goto release; 2324 } 2325 } 2326 m = so->so_rcv.sb_mb; 2327 if (m != NULL) 2328 nextrecord = m->m_nextpkt; 2329 } 2330 } 2331 2332 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2333 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 2334 flags |= MSG_TRUNC; 2335 if ((flags & MSG_PEEK) == 0) 2336 (void) sbdroprecord_locked(&so->so_rcv); 2337 } 2338 if ((flags & MSG_PEEK) == 0) { 2339 if (m == NULL) { 2340 /* 2341 * First part is an inline SB_EMPTY_FIXUP(). Second 2342 * part makes sure sb_lastrecord is up-to-date if 2343 * there is still data in the socket buffer. 2344 */ 2345 so->so_rcv.sb_mb = nextrecord; 2346 if (so->so_rcv.sb_mb == NULL) { 2347 so->so_rcv.sb_mbtail = NULL; 2348 so->so_rcv.sb_lastrecord = NULL; 2349 } else if (nextrecord->m_nextpkt == NULL) 2350 so->so_rcv.sb_lastrecord = nextrecord; 2351 } 2352 SBLASTRECORDCHK(&so->so_rcv); 2353 SBLASTMBUFCHK(&so->so_rcv); 2354 /* 2355 * If soreceive() is being done from the socket callback, 2356 * then don't need to generate ACK to peer to update window, 2357 * since ACK will be generated on return to TCP. 2358 */ 2359 if (!(flags & MSG_SOCALLBCK) && 2360 (pr->pr_flags & PR_WANTRCVD)) { 2361 SOCKBUF_UNLOCK(&so->so_rcv); 2362 VNET_SO_ASSERT(so); 2363 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2364 SOCKBUF_LOCK(&so->so_rcv); 2365 } 2366 } 2367 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2368 if (orig_resid == uio->uio_resid && orig_resid && 2369 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 2370 SOCKBUF_UNLOCK(&so->so_rcv); 2371 goto restart; 2372 } 2373 SOCKBUF_UNLOCK(&so->so_rcv); 2374 2375 if (flagsp != NULL) 2376 *flagsp |= flags; 2377 release: 2378 sbunlock(&so->so_rcv); 2379 return (error); 2380 } 2381 2382 /* 2383 * Optimized version of soreceive() for stream (TCP) sockets. 2384 */ 2385 int 2386 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 2387 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2388 { 2389 int len = 0, error = 0, flags, oresid; 2390 struct sockbuf *sb; 2391 struct mbuf *m, *n = NULL; 2392 2393 /* We only do stream sockets. */ 2394 if (so->so_type != SOCK_STREAM) 2395 return (EINVAL); 2396 if (psa != NULL) 2397 *psa = NULL; 2398 if (flagsp != NULL) 2399 flags = *flagsp &~ MSG_EOR; 2400 else 2401 flags = 0; 2402 if (controlp != NULL) 2403 *controlp = NULL; 2404 if (flags & MSG_OOB) 2405 return (soreceive_rcvoob(so, uio, flags)); 2406 if (mp0 != NULL) 2407 *mp0 = NULL; 2408 2409 sb = &so->so_rcv; 2410 2411 #ifdef KERN_TLS 2412 /* 2413 * KTLS store TLS records as records with a control message to 2414 * describe the framing. 2415 * 2416 * We check once here before acquiring locks to optimize the 2417 * common case. 2418 */ 2419 if (sb->sb_tls_info != NULL) 2420 return (soreceive_generic(so, psa, uio, mp0, controlp, 2421 flagsp)); 2422 #endif 2423 2424 /* Prevent other readers from entering the socket. */ 2425 error = sblock(sb, SBLOCKWAIT(flags)); 2426 if (error) 2427 return (error); 2428 SOCKBUF_LOCK(sb); 2429 2430 #ifdef KERN_TLS 2431 if (sb->sb_tls_info != NULL) { 2432 SOCKBUF_UNLOCK(sb); 2433 sbunlock(sb); 2434 return (soreceive_generic(so, psa, uio, mp0, controlp, 2435 flagsp)); 2436 } 2437 #endif 2438 2439 /* Easy one, no space to copyout anything. */ 2440 if (uio->uio_resid == 0) { 2441 error = EINVAL; 2442 goto out; 2443 } 2444 oresid = uio->uio_resid; 2445 2446 /* We will never ever get anything unless we are or were connected. */ 2447 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 2448 error = ENOTCONN; 2449 goto out; 2450 } 2451 2452 restart: 2453 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2454 2455 /* Abort if socket has reported problems. */ 2456 if (so->so_error) { 2457 if (sbavail(sb) > 0) 2458 goto deliver; 2459 if (oresid > uio->uio_resid) 2460 goto out; 2461 error = so->so_error; 2462 if (!(flags & MSG_PEEK)) 2463 so->so_error = 0; 2464 goto out; 2465 } 2466 2467 /* Door is closed. Deliver what is left, if any. */ 2468 if (sb->sb_state & SBS_CANTRCVMORE) { 2469 if (sbavail(sb) > 0) 2470 goto deliver; 2471 else 2472 goto out; 2473 } 2474 2475 /* Socket buffer is empty and we shall not block. */ 2476 if (sbavail(sb) == 0 && 2477 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2478 error = EAGAIN; 2479 goto out; 2480 } 2481 2482 /* Socket buffer got some data that we shall deliver now. */ 2483 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 2484 ((so->so_state & SS_NBIO) || 2485 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2486 sbavail(sb) >= sb->sb_lowat || 2487 sbavail(sb) >= uio->uio_resid || 2488 sbavail(sb) >= sb->sb_hiwat) ) { 2489 goto deliver; 2490 } 2491 2492 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2493 if ((flags & MSG_WAITALL) && 2494 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 2495 goto deliver; 2496 2497 /* 2498 * Wait and block until (more) data comes in. 2499 * NB: Drops the sockbuf lock during wait. 2500 */ 2501 error = sbwait(sb); 2502 if (error) 2503 goto out; 2504 goto restart; 2505 2506 deliver: 2507 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2508 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 2509 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2510 2511 /* Statistics. */ 2512 if (uio->uio_td) 2513 uio->uio_td->td_ru.ru_msgrcv++; 2514 2515 /* Fill uio until full or current end of socket buffer is reached. */ 2516 len = min(uio->uio_resid, sbavail(sb)); 2517 if (mp0 != NULL) { 2518 /* Dequeue as many mbufs as possible. */ 2519 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2520 if (*mp0 == NULL) 2521 *mp0 = sb->sb_mb; 2522 else 2523 m_cat(*mp0, sb->sb_mb); 2524 for (m = sb->sb_mb; 2525 m != NULL && m->m_len <= len; 2526 m = m->m_next) { 2527 KASSERT(!(m->m_flags & M_NOTAVAIL), 2528 ("%s: m %p not available", __func__, m)); 2529 len -= m->m_len; 2530 uio->uio_resid -= m->m_len; 2531 sbfree(sb, m); 2532 n = m; 2533 } 2534 n->m_next = NULL; 2535 sb->sb_mb = m; 2536 sb->sb_lastrecord = sb->sb_mb; 2537 if (sb->sb_mb == NULL) 2538 SB_EMPTY_FIXUP(sb); 2539 } 2540 /* Copy the remainder. */ 2541 if (len > 0) { 2542 KASSERT(sb->sb_mb != NULL, 2543 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2544 2545 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2546 if (m == NULL) 2547 len = 0; /* Don't flush data from sockbuf. */ 2548 else 2549 uio->uio_resid -= len; 2550 if (*mp0 != NULL) 2551 m_cat(*mp0, m); 2552 else 2553 *mp0 = m; 2554 if (*mp0 == NULL) { 2555 error = ENOBUFS; 2556 goto out; 2557 } 2558 } 2559 } else { 2560 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2561 SOCKBUF_UNLOCK(sb); 2562 error = m_mbuftouio(uio, sb->sb_mb, len); 2563 SOCKBUF_LOCK(sb); 2564 if (error) 2565 goto out; 2566 } 2567 SBLASTRECORDCHK(sb); 2568 SBLASTMBUFCHK(sb); 2569 2570 /* 2571 * Remove the delivered data from the socket buffer unless we 2572 * were only peeking. 2573 */ 2574 if (!(flags & MSG_PEEK)) { 2575 if (len > 0) 2576 sbdrop_locked(sb, len); 2577 2578 /* Notify protocol that we drained some data. */ 2579 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2580 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2581 !(flags & MSG_SOCALLBCK))) { 2582 SOCKBUF_UNLOCK(sb); 2583 VNET_SO_ASSERT(so); 2584 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags); 2585 SOCKBUF_LOCK(sb); 2586 } 2587 } 2588 2589 /* 2590 * For MSG_WAITALL we may have to loop again and wait for 2591 * more data to come in. 2592 */ 2593 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2594 goto restart; 2595 out: 2596 SOCKBUF_LOCK_ASSERT(sb); 2597 SBLASTRECORDCHK(sb); 2598 SBLASTMBUFCHK(sb); 2599 SOCKBUF_UNLOCK(sb); 2600 sbunlock(sb); 2601 return (error); 2602 } 2603 2604 /* 2605 * Optimized version of soreceive() for simple datagram cases from userspace. 2606 * Unlike in the stream case, we're able to drop a datagram if copyout() 2607 * fails, and because we handle datagrams atomically, we don't need to use a 2608 * sleep lock to prevent I/O interlacing. 2609 */ 2610 int 2611 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2612 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2613 { 2614 struct mbuf *m, *m2; 2615 int flags, error; 2616 ssize_t len; 2617 struct protosw *pr = so->so_proto; 2618 struct mbuf *nextrecord; 2619 2620 if (psa != NULL) 2621 *psa = NULL; 2622 if (controlp != NULL) 2623 *controlp = NULL; 2624 if (flagsp != NULL) 2625 flags = *flagsp &~ MSG_EOR; 2626 else 2627 flags = 0; 2628 2629 /* 2630 * For any complicated cases, fall back to the full 2631 * soreceive_generic(). 2632 */ 2633 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB)) 2634 return (soreceive_generic(so, psa, uio, mp0, controlp, 2635 flagsp)); 2636 2637 /* 2638 * Enforce restrictions on use. 2639 */ 2640 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2641 ("soreceive_dgram: wantrcvd")); 2642 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2643 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2644 ("soreceive_dgram: SBS_RCVATMARK")); 2645 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2646 ("soreceive_dgram: P_CONNREQUIRED")); 2647 2648 /* 2649 * Loop blocking while waiting for a datagram. 2650 */ 2651 SOCKBUF_LOCK(&so->so_rcv); 2652 while ((m = so->so_rcv.sb_mb) == NULL) { 2653 KASSERT(sbavail(&so->so_rcv) == 0, 2654 ("soreceive_dgram: sb_mb NULL but sbavail %u", 2655 sbavail(&so->so_rcv))); 2656 if (so->so_error) { 2657 error = so->so_error; 2658 so->so_error = 0; 2659 SOCKBUF_UNLOCK(&so->so_rcv); 2660 return (error); 2661 } 2662 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2663 uio->uio_resid == 0) { 2664 SOCKBUF_UNLOCK(&so->so_rcv); 2665 return (0); 2666 } 2667 if ((so->so_state & SS_NBIO) || 2668 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2669 SOCKBUF_UNLOCK(&so->so_rcv); 2670 return (EWOULDBLOCK); 2671 } 2672 SBLASTRECORDCHK(&so->so_rcv); 2673 SBLASTMBUFCHK(&so->so_rcv); 2674 error = sbwait(&so->so_rcv); 2675 if (error) { 2676 SOCKBUF_UNLOCK(&so->so_rcv); 2677 return (error); 2678 } 2679 } 2680 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2681 2682 if (uio->uio_td) 2683 uio->uio_td->td_ru.ru_msgrcv++; 2684 SBLASTRECORDCHK(&so->so_rcv); 2685 SBLASTMBUFCHK(&so->so_rcv); 2686 nextrecord = m->m_nextpkt; 2687 if (nextrecord == NULL) { 2688 KASSERT(so->so_rcv.sb_lastrecord == m, 2689 ("soreceive_dgram: lastrecord != m")); 2690 } 2691 2692 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2693 ("soreceive_dgram: m_nextpkt != nextrecord")); 2694 2695 /* 2696 * Pull 'm' and its chain off the front of the packet queue. 2697 */ 2698 so->so_rcv.sb_mb = NULL; 2699 sockbuf_pushsync(&so->so_rcv, nextrecord); 2700 2701 /* 2702 * Walk 'm's chain and free that many bytes from the socket buffer. 2703 */ 2704 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2705 sbfree(&so->so_rcv, m2); 2706 2707 /* 2708 * Do a few last checks before we let go of the lock. 2709 */ 2710 SBLASTRECORDCHK(&so->so_rcv); 2711 SBLASTMBUFCHK(&so->so_rcv); 2712 SOCKBUF_UNLOCK(&so->so_rcv); 2713 2714 if (pr->pr_flags & PR_ADDR) { 2715 KASSERT(m->m_type == MT_SONAME, 2716 ("m->m_type == %d", m->m_type)); 2717 if (psa != NULL) 2718 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2719 M_NOWAIT); 2720 m = m_free(m); 2721 } 2722 if (m == NULL) { 2723 /* XXXRW: Can this happen? */ 2724 return (0); 2725 } 2726 2727 /* 2728 * Packet to copyout() is now in 'm' and it is disconnected from the 2729 * queue. 2730 * 2731 * Process one or more MT_CONTROL mbufs present before any data mbufs 2732 * in the first mbuf chain on the socket buffer. We call into the 2733 * protocol to perform externalization (or freeing if controlp == 2734 * NULL). In some cases there can be only MT_CONTROL mbufs without 2735 * MT_DATA mbufs. 2736 */ 2737 if (m->m_type == MT_CONTROL) { 2738 struct mbuf *cm = NULL, *cmn; 2739 struct mbuf **cme = &cm; 2740 2741 do { 2742 m2 = m->m_next; 2743 m->m_next = NULL; 2744 *cme = m; 2745 cme = &(*cme)->m_next; 2746 m = m2; 2747 } while (m != NULL && m->m_type == MT_CONTROL); 2748 while (cm != NULL) { 2749 cmn = cm->m_next; 2750 cm->m_next = NULL; 2751 if (pr->pr_domain->dom_externalize != NULL) { 2752 error = (*pr->pr_domain->dom_externalize) 2753 (cm, controlp, flags); 2754 } else if (controlp != NULL) 2755 *controlp = cm; 2756 else 2757 m_freem(cm); 2758 if (controlp != NULL) { 2759 while (*controlp != NULL) 2760 controlp = &(*controlp)->m_next; 2761 } 2762 cm = cmn; 2763 } 2764 } 2765 KASSERT(m == NULL || m->m_type == MT_DATA, 2766 ("soreceive_dgram: !data")); 2767 while (m != NULL && uio->uio_resid > 0) { 2768 len = uio->uio_resid; 2769 if (len > m->m_len) 2770 len = m->m_len; 2771 error = uiomove(mtod(m, char *), (int)len, uio); 2772 if (error) { 2773 m_freem(m); 2774 return (error); 2775 } 2776 if (len == m->m_len) 2777 m = m_free(m); 2778 else { 2779 m->m_data += len; 2780 m->m_len -= len; 2781 } 2782 } 2783 if (m != NULL) { 2784 flags |= MSG_TRUNC; 2785 m_freem(m); 2786 } 2787 if (flagsp != NULL) 2788 *flagsp |= flags; 2789 return (0); 2790 } 2791 2792 int 2793 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2794 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2795 { 2796 int error; 2797 2798 CURVNET_SET(so->so_vnet); 2799 if (!SOLISTENING(so)) 2800 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, 2801 mp0, controlp, flagsp)); 2802 else 2803 error = ENOTCONN; 2804 CURVNET_RESTORE(); 2805 return (error); 2806 } 2807 2808 int 2809 soshutdown(struct socket *so, int how) 2810 { 2811 struct protosw *pr = so->so_proto; 2812 int error, soerror_enotconn; 2813 2814 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2815 return (EINVAL); 2816 2817 soerror_enotconn = 0; 2818 if ((so->so_state & 2819 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 2820 /* 2821 * POSIX mandates us to return ENOTCONN when shutdown(2) is 2822 * invoked on a datagram sockets, however historically we would 2823 * actually tear socket down. This is known to be leveraged by 2824 * some applications to unblock process waiting in recvXXX(2) 2825 * by other process that it shares that socket with. Try to meet 2826 * both backward-compatibility and POSIX requirements by forcing 2827 * ENOTCONN but still asking protocol to perform pru_shutdown(). 2828 */ 2829 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) 2830 return (ENOTCONN); 2831 soerror_enotconn = 1; 2832 } 2833 2834 if (SOLISTENING(so)) { 2835 if (how != SHUT_WR) { 2836 SOLISTEN_LOCK(so); 2837 so->so_error = ECONNABORTED; 2838 solisten_wakeup(so); /* unlocks so */ 2839 } 2840 goto done; 2841 } 2842 2843 CURVNET_SET(so->so_vnet); 2844 if (pr->pr_usrreqs->pru_flush != NULL) 2845 (*pr->pr_usrreqs->pru_flush)(so, how); 2846 if (how != SHUT_WR) 2847 sorflush(so); 2848 if (how != SHUT_RD) { 2849 error = (*pr->pr_usrreqs->pru_shutdown)(so); 2850 wakeup(&so->so_timeo); 2851 CURVNET_RESTORE(); 2852 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error); 2853 } 2854 wakeup(&so->so_timeo); 2855 CURVNET_RESTORE(); 2856 2857 done: 2858 return (soerror_enotconn ? ENOTCONN : 0); 2859 } 2860 2861 void 2862 sorflush(struct socket *so) 2863 { 2864 struct sockbuf *sb = &so->so_rcv; 2865 struct protosw *pr = so->so_proto; 2866 struct socket aso; 2867 2868 VNET_SO_ASSERT(so); 2869 2870 /* 2871 * In order to avoid calling dom_dispose with the socket buffer mutex 2872 * held, and in order to generally avoid holding the lock for a long 2873 * time, we make a copy of the socket buffer and clear the original 2874 * (except locks, state). The new socket buffer copy won't have 2875 * initialized locks so we can only call routines that won't use or 2876 * assert those locks. 2877 * 2878 * Dislodge threads currently blocked in receive and wait to acquire 2879 * a lock against other simultaneous readers before clearing the 2880 * socket buffer. Don't let our acquire be interrupted by a signal 2881 * despite any existing socket disposition on interruptable waiting. 2882 */ 2883 socantrcvmore(so); 2884 (void) sblock(sb, SBL_WAIT | SBL_NOINTR); 2885 2886 /* 2887 * Invalidate/clear most of the sockbuf structure, but leave selinfo 2888 * and mutex data unchanged. 2889 */ 2890 SOCKBUF_LOCK(sb); 2891 bzero(&aso, sizeof(aso)); 2892 aso.so_pcb = so->so_pcb; 2893 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero, 2894 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2895 bzero(&sb->sb_startzero, 2896 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2897 SOCKBUF_UNLOCK(sb); 2898 sbunlock(sb); 2899 2900 /* 2901 * Dispose of special rights and flush the copied socket. Don't call 2902 * any unsafe routines (that rely on locks being initialized) on aso. 2903 */ 2904 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 2905 (*pr->pr_domain->dom_dispose)(&aso); 2906 sbrelease_internal(&aso.so_rcv, so); 2907 } 2908 2909 /* 2910 * Wrapper for Socket established helper hook. 2911 * Parameters: socket, context of the hook point, hook id. 2912 */ 2913 static int inline 2914 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 2915 { 2916 struct socket_hhook_data hhook_data = { 2917 .so = so, 2918 .hctx = hctx, 2919 .m = NULL, 2920 .status = 0 2921 }; 2922 2923 CURVNET_SET(so->so_vnet); 2924 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 2925 CURVNET_RESTORE(); 2926 2927 /* Ugly but needed, since hhooks return void for now */ 2928 return (hhook_data.status); 2929 } 2930 2931 /* 2932 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 2933 * additional variant to handle the case where the option value needs to be 2934 * some kind of integer, but not a specific size. In addition to their use 2935 * here, these functions are also called by the protocol-level pr_ctloutput() 2936 * routines. 2937 */ 2938 int 2939 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2940 { 2941 size_t valsize; 2942 2943 /* 2944 * If the user gives us more than we wanted, we ignore it, but if we 2945 * don't get the minimum length the caller wants, we return EINVAL. 2946 * On success, sopt->sopt_valsize is set to however much we actually 2947 * retrieved. 2948 */ 2949 if ((valsize = sopt->sopt_valsize) < minlen) 2950 return EINVAL; 2951 if (valsize > len) 2952 sopt->sopt_valsize = valsize = len; 2953 2954 if (sopt->sopt_td != NULL) 2955 return (copyin(sopt->sopt_val, buf, valsize)); 2956 2957 bcopy(sopt->sopt_val, buf, valsize); 2958 return (0); 2959 } 2960 2961 /* 2962 * Kernel version of setsockopt(2). 2963 * 2964 * XXX: optlen is size_t, not socklen_t 2965 */ 2966 int 2967 so_setsockopt(struct socket *so, int level, int optname, void *optval, 2968 size_t optlen) 2969 { 2970 struct sockopt sopt; 2971 2972 sopt.sopt_level = level; 2973 sopt.sopt_name = optname; 2974 sopt.sopt_dir = SOPT_SET; 2975 sopt.sopt_val = optval; 2976 sopt.sopt_valsize = optlen; 2977 sopt.sopt_td = NULL; 2978 return (sosetopt(so, &sopt)); 2979 } 2980 2981 int 2982 sosetopt(struct socket *so, struct sockopt *sopt) 2983 { 2984 int error, optval; 2985 struct linger l; 2986 struct timeval tv; 2987 sbintime_t val; 2988 uint32_t val32; 2989 #ifdef MAC 2990 struct mac extmac; 2991 #endif 2992 2993 CURVNET_SET(so->so_vnet); 2994 error = 0; 2995 if (sopt->sopt_level != SOL_SOCKET) { 2996 if (so->so_proto->pr_ctloutput != NULL) 2997 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2998 else 2999 error = ENOPROTOOPT; 3000 } else { 3001 switch (sopt->sopt_name) { 3002 case SO_ACCEPTFILTER: 3003 error = accept_filt_setopt(so, sopt); 3004 if (error) 3005 goto bad; 3006 break; 3007 3008 case SO_LINGER: 3009 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 3010 if (error) 3011 goto bad; 3012 if (l.l_linger < 0 || 3013 l.l_linger > USHRT_MAX || 3014 l.l_linger > (INT_MAX / hz)) { 3015 error = EDOM; 3016 goto bad; 3017 } 3018 SOCK_LOCK(so); 3019 so->so_linger = l.l_linger; 3020 if (l.l_onoff) 3021 so->so_options |= SO_LINGER; 3022 else 3023 so->so_options &= ~SO_LINGER; 3024 SOCK_UNLOCK(so); 3025 break; 3026 3027 case SO_DEBUG: 3028 case SO_KEEPALIVE: 3029 case SO_DONTROUTE: 3030 case SO_USELOOPBACK: 3031 case SO_BROADCAST: 3032 case SO_REUSEADDR: 3033 case SO_REUSEPORT: 3034 case SO_REUSEPORT_LB: 3035 case SO_OOBINLINE: 3036 case SO_TIMESTAMP: 3037 case SO_BINTIME: 3038 case SO_NOSIGPIPE: 3039 case SO_NO_DDP: 3040 case SO_NO_OFFLOAD: 3041 error = sooptcopyin(sopt, &optval, sizeof optval, 3042 sizeof optval); 3043 if (error) 3044 goto bad; 3045 SOCK_LOCK(so); 3046 if (optval) 3047 so->so_options |= sopt->sopt_name; 3048 else 3049 so->so_options &= ~sopt->sopt_name; 3050 SOCK_UNLOCK(so); 3051 break; 3052 3053 case SO_SETFIB: 3054 error = sooptcopyin(sopt, &optval, sizeof optval, 3055 sizeof optval); 3056 if (error) 3057 goto bad; 3058 3059 if (optval < 0 || optval >= rt_numfibs) { 3060 error = EINVAL; 3061 goto bad; 3062 } 3063 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 3064 (so->so_proto->pr_domain->dom_family == PF_INET6) || 3065 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 3066 so->so_fibnum = optval; 3067 else 3068 so->so_fibnum = 0; 3069 break; 3070 3071 case SO_USER_COOKIE: 3072 error = sooptcopyin(sopt, &val32, sizeof val32, 3073 sizeof val32); 3074 if (error) 3075 goto bad; 3076 so->so_user_cookie = val32; 3077 break; 3078 3079 case SO_SNDBUF: 3080 case SO_RCVBUF: 3081 case SO_SNDLOWAT: 3082 case SO_RCVLOWAT: 3083 error = sooptcopyin(sopt, &optval, sizeof optval, 3084 sizeof optval); 3085 if (error) 3086 goto bad; 3087 3088 /* 3089 * Values < 1 make no sense for any of these options, 3090 * so disallow them. 3091 */ 3092 if (optval < 1) { 3093 error = EINVAL; 3094 goto bad; 3095 } 3096 3097 error = sbsetopt(so, sopt->sopt_name, optval); 3098 break; 3099 3100 case SO_SNDTIMEO: 3101 case SO_RCVTIMEO: 3102 #ifdef COMPAT_FREEBSD32 3103 if (SV_CURPROC_FLAG(SV_ILP32)) { 3104 struct timeval32 tv32; 3105 3106 error = sooptcopyin(sopt, &tv32, sizeof tv32, 3107 sizeof tv32); 3108 CP(tv32, tv, tv_sec); 3109 CP(tv32, tv, tv_usec); 3110 } else 3111 #endif 3112 error = sooptcopyin(sopt, &tv, sizeof tv, 3113 sizeof tv); 3114 if (error) 3115 goto bad; 3116 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 3117 tv.tv_usec >= 1000000) { 3118 error = EDOM; 3119 goto bad; 3120 } 3121 if (tv.tv_sec > INT32_MAX) 3122 val = SBT_MAX; 3123 else 3124 val = tvtosbt(tv); 3125 switch (sopt->sopt_name) { 3126 case SO_SNDTIMEO: 3127 so->so_snd.sb_timeo = val; 3128 break; 3129 case SO_RCVTIMEO: 3130 so->so_rcv.sb_timeo = val; 3131 break; 3132 } 3133 break; 3134 3135 case SO_LABEL: 3136 #ifdef MAC 3137 error = sooptcopyin(sopt, &extmac, sizeof extmac, 3138 sizeof extmac); 3139 if (error) 3140 goto bad; 3141 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 3142 so, &extmac); 3143 #else 3144 error = EOPNOTSUPP; 3145 #endif 3146 break; 3147 3148 case SO_TS_CLOCK: 3149 error = sooptcopyin(sopt, &optval, sizeof optval, 3150 sizeof optval); 3151 if (error) 3152 goto bad; 3153 if (optval < 0 || optval > SO_TS_CLOCK_MAX) { 3154 error = EINVAL; 3155 goto bad; 3156 } 3157 so->so_ts_clock = optval; 3158 break; 3159 3160 case SO_MAX_PACING_RATE: 3161 error = sooptcopyin(sopt, &val32, sizeof(val32), 3162 sizeof(val32)); 3163 if (error) 3164 goto bad; 3165 so->so_max_pacing_rate = val32; 3166 break; 3167 3168 default: 3169 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3170 error = hhook_run_socket(so, sopt, 3171 HHOOK_SOCKET_OPT); 3172 else 3173 error = ENOPROTOOPT; 3174 break; 3175 } 3176 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 3177 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 3178 } 3179 bad: 3180 CURVNET_RESTORE(); 3181 return (error); 3182 } 3183 3184 /* 3185 * Helper routine for getsockopt. 3186 */ 3187 int 3188 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 3189 { 3190 int error; 3191 size_t valsize; 3192 3193 error = 0; 3194 3195 /* 3196 * Documented get behavior is that we always return a value, possibly 3197 * truncated to fit in the user's buffer. Traditional behavior is 3198 * that we always tell the user precisely how much we copied, rather 3199 * than something useful like the total amount we had available for 3200 * her. Note that this interface is not idempotent; the entire 3201 * answer must be generated ahead of time. 3202 */ 3203 valsize = min(len, sopt->sopt_valsize); 3204 sopt->sopt_valsize = valsize; 3205 if (sopt->sopt_val != NULL) { 3206 if (sopt->sopt_td != NULL) 3207 error = copyout(buf, sopt->sopt_val, valsize); 3208 else 3209 bcopy(buf, sopt->sopt_val, valsize); 3210 } 3211 return (error); 3212 } 3213 3214 int 3215 sogetopt(struct socket *so, struct sockopt *sopt) 3216 { 3217 int error, optval; 3218 struct linger l; 3219 struct timeval tv; 3220 #ifdef MAC 3221 struct mac extmac; 3222 #endif 3223 3224 CURVNET_SET(so->so_vnet); 3225 error = 0; 3226 if (sopt->sopt_level != SOL_SOCKET) { 3227 if (so->so_proto->pr_ctloutput != NULL) 3228 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3229 else 3230 error = ENOPROTOOPT; 3231 CURVNET_RESTORE(); 3232 return (error); 3233 } else { 3234 switch (sopt->sopt_name) { 3235 case SO_ACCEPTFILTER: 3236 error = accept_filt_getopt(so, sopt); 3237 break; 3238 3239 case SO_LINGER: 3240 SOCK_LOCK(so); 3241 l.l_onoff = so->so_options & SO_LINGER; 3242 l.l_linger = so->so_linger; 3243 SOCK_UNLOCK(so); 3244 error = sooptcopyout(sopt, &l, sizeof l); 3245 break; 3246 3247 case SO_USELOOPBACK: 3248 case SO_DONTROUTE: 3249 case SO_DEBUG: 3250 case SO_KEEPALIVE: 3251 case SO_REUSEADDR: 3252 case SO_REUSEPORT: 3253 case SO_REUSEPORT_LB: 3254 case SO_BROADCAST: 3255 case SO_OOBINLINE: 3256 case SO_ACCEPTCONN: 3257 case SO_TIMESTAMP: 3258 case SO_BINTIME: 3259 case SO_NOSIGPIPE: 3260 case SO_NO_DDP: 3261 case SO_NO_OFFLOAD: 3262 optval = so->so_options & sopt->sopt_name; 3263 integer: 3264 error = sooptcopyout(sopt, &optval, sizeof optval); 3265 break; 3266 3267 case SO_DOMAIN: 3268 optval = so->so_proto->pr_domain->dom_family; 3269 goto integer; 3270 3271 case SO_TYPE: 3272 optval = so->so_type; 3273 goto integer; 3274 3275 case SO_PROTOCOL: 3276 optval = so->so_proto->pr_protocol; 3277 goto integer; 3278 3279 case SO_ERROR: 3280 SOCK_LOCK(so); 3281 optval = so->so_error; 3282 so->so_error = 0; 3283 SOCK_UNLOCK(so); 3284 goto integer; 3285 3286 case SO_SNDBUF: 3287 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : 3288 so->so_snd.sb_hiwat; 3289 goto integer; 3290 3291 case SO_RCVBUF: 3292 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : 3293 so->so_rcv.sb_hiwat; 3294 goto integer; 3295 3296 case SO_SNDLOWAT: 3297 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : 3298 so->so_snd.sb_lowat; 3299 goto integer; 3300 3301 case SO_RCVLOWAT: 3302 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : 3303 so->so_rcv.sb_lowat; 3304 goto integer; 3305 3306 case SO_SNDTIMEO: 3307 case SO_RCVTIMEO: 3308 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 3309 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 3310 #ifdef COMPAT_FREEBSD32 3311 if (SV_CURPROC_FLAG(SV_ILP32)) { 3312 struct timeval32 tv32; 3313 3314 CP(tv, tv32, tv_sec); 3315 CP(tv, tv32, tv_usec); 3316 error = sooptcopyout(sopt, &tv32, sizeof tv32); 3317 } else 3318 #endif 3319 error = sooptcopyout(sopt, &tv, sizeof tv); 3320 break; 3321 3322 case SO_LABEL: 3323 #ifdef MAC 3324 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3325 sizeof(extmac)); 3326 if (error) 3327 goto bad; 3328 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 3329 so, &extmac); 3330 if (error) 3331 goto bad; 3332 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3333 #else 3334 error = EOPNOTSUPP; 3335 #endif 3336 break; 3337 3338 case SO_PEERLABEL: 3339 #ifdef MAC 3340 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3341 sizeof(extmac)); 3342 if (error) 3343 goto bad; 3344 error = mac_getsockopt_peerlabel( 3345 sopt->sopt_td->td_ucred, so, &extmac); 3346 if (error) 3347 goto bad; 3348 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3349 #else 3350 error = EOPNOTSUPP; 3351 #endif 3352 break; 3353 3354 case SO_LISTENQLIMIT: 3355 optval = SOLISTENING(so) ? so->sol_qlimit : 0; 3356 goto integer; 3357 3358 case SO_LISTENQLEN: 3359 optval = SOLISTENING(so) ? so->sol_qlen : 0; 3360 goto integer; 3361 3362 case SO_LISTENINCQLEN: 3363 optval = SOLISTENING(so) ? so->sol_incqlen : 0; 3364 goto integer; 3365 3366 case SO_TS_CLOCK: 3367 optval = so->so_ts_clock; 3368 goto integer; 3369 3370 case SO_MAX_PACING_RATE: 3371 optval = so->so_max_pacing_rate; 3372 goto integer; 3373 3374 default: 3375 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3376 error = hhook_run_socket(so, sopt, 3377 HHOOK_SOCKET_OPT); 3378 else 3379 error = ENOPROTOOPT; 3380 break; 3381 } 3382 } 3383 #ifdef MAC 3384 bad: 3385 #endif 3386 CURVNET_RESTORE(); 3387 return (error); 3388 } 3389 3390 int 3391 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 3392 { 3393 struct mbuf *m, *m_prev; 3394 int sopt_size = sopt->sopt_valsize; 3395 3396 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3397 if (m == NULL) 3398 return ENOBUFS; 3399 if (sopt_size > MLEN) { 3400 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 3401 if ((m->m_flags & M_EXT) == 0) { 3402 m_free(m); 3403 return ENOBUFS; 3404 } 3405 m->m_len = min(MCLBYTES, sopt_size); 3406 } else { 3407 m->m_len = min(MLEN, sopt_size); 3408 } 3409 sopt_size -= m->m_len; 3410 *mp = m; 3411 m_prev = m; 3412 3413 while (sopt_size) { 3414 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3415 if (m == NULL) { 3416 m_freem(*mp); 3417 return ENOBUFS; 3418 } 3419 if (sopt_size > MLEN) { 3420 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 3421 M_NOWAIT); 3422 if ((m->m_flags & M_EXT) == 0) { 3423 m_freem(m); 3424 m_freem(*mp); 3425 return ENOBUFS; 3426 } 3427 m->m_len = min(MCLBYTES, sopt_size); 3428 } else { 3429 m->m_len = min(MLEN, sopt_size); 3430 } 3431 sopt_size -= m->m_len; 3432 m_prev->m_next = m; 3433 m_prev = m; 3434 } 3435 return (0); 3436 } 3437 3438 int 3439 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 3440 { 3441 struct mbuf *m0 = m; 3442 3443 if (sopt->sopt_val == NULL) 3444 return (0); 3445 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3446 if (sopt->sopt_td != NULL) { 3447 int error; 3448 3449 error = copyin(sopt->sopt_val, mtod(m, char *), 3450 m->m_len); 3451 if (error != 0) { 3452 m_freem(m0); 3453 return(error); 3454 } 3455 } else 3456 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 3457 sopt->sopt_valsize -= m->m_len; 3458 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3459 m = m->m_next; 3460 } 3461 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 3462 panic("ip6_sooptmcopyin"); 3463 return (0); 3464 } 3465 3466 int 3467 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 3468 { 3469 struct mbuf *m0 = m; 3470 size_t valsize = 0; 3471 3472 if (sopt->sopt_val == NULL) 3473 return (0); 3474 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3475 if (sopt->sopt_td != NULL) { 3476 int error; 3477 3478 error = copyout(mtod(m, char *), sopt->sopt_val, 3479 m->m_len); 3480 if (error != 0) { 3481 m_freem(m0); 3482 return(error); 3483 } 3484 } else 3485 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 3486 sopt->sopt_valsize -= m->m_len; 3487 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3488 valsize += m->m_len; 3489 m = m->m_next; 3490 } 3491 if (m != NULL) { 3492 /* enough soopt buffer should be given from user-land */ 3493 m_freem(m0); 3494 return(EINVAL); 3495 } 3496 sopt->sopt_valsize = valsize; 3497 return (0); 3498 } 3499 3500 /* 3501 * sohasoutofband(): protocol notifies socket layer of the arrival of new 3502 * out-of-band data, which will then notify socket consumers. 3503 */ 3504 void 3505 sohasoutofband(struct socket *so) 3506 { 3507 3508 if (so->so_sigio != NULL) 3509 pgsigio(&so->so_sigio, SIGURG, 0); 3510 selwakeuppri(&so->so_rdsel, PSOCK); 3511 } 3512 3513 int 3514 sopoll(struct socket *so, int events, struct ucred *active_cred, 3515 struct thread *td) 3516 { 3517 3518 /* 3519 * We do not need to set or assert curvnet as long as everyone uses 3520 * sopoll_generic(). 3521 */ 3522 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, 3523 td)); 3524 } 3525 3526 int 3527 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3528 struct thread *td) 3529 { 3530 int revents; 3531 3532 SOCK_LOCK(so); 3533 if (SOLISTENING(so)) { 3534 if (!(events & (POLLIN | POLLRDNORM))) 3535 revents = 0; 3536 else if (!TAILQ_EMPTY(&so->sol_comp)) 3537 revents = events & (POLLIN | POLLRDNORM); 3538 else if ((events & POLLINIGNEOF) == 0 && so->so_error) 3539 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP; 3540 else { 3541 selrecord(td, &so->so_rdsel); 3542 revents = 0; 3543 } 3544 } else { 3545 revents = 0; 3546 SOCKBUF_LOCK(&so->so_snd); 3547 SOCKBUF_LOCK(&so->so_rcv); 3548 if (events & (POLLIN | POLLRDNORM)) 3549 if (soreadabledata(so)) 3550 revents |= events & (POLLIN | POLLRDNORM); 3551 if (events & (POLLOUT | POLLWRNORM)) 3552 if (sowriteable(so)) 3553 revents |= events & (POLLOUT | POLLWRNORM); 3554 if (events & (POLLPRI | POLLRDBAND)) 3555 if (so->so_oobmark || 3556 (so->so_rcv.sb_state & SBS_RCVATMARK)) 3557 revents |= events & (POLLPRI | POLLRDBAND); 3558 if ((events & POLLINIGNEOF) == 0) { 3559 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3560 revents |= events & (POLLIN | POLLRDNORM); 3561 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3562 revents |= POLLHUP; 3563 } 3564 } 3565 if (revents == 0) { 3566 if (events & 3567 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 3568 selrecord(td, &so->so_rdsel); 3569 so->so_rcv.sb_flags |= SB_SEL; 3570 } 3571 if (events & (POLLOUT | POLLWRNORM)) { 3572 selrecord(td, &so->so_wrsel); 3573 so->so_snd.sb_flags |= SB_SEL; 3574 } 3575 } 3576 SOCKBUF_UNLOCK(&so->so_rcv); 3577 SOCKBUF_UNLOCK(&so->so_snd); 3578 } 3579 SOCK_UNLOCK(so); 3580 return (revents); 3581 } 3582 3583 int 3584 soo_kqfilter(struct file *fp, struct knote *kn) 3585 { 3586 struct socket *so = kn->kn_fp->f_data; 3587 struct sockbuf *sb; 3588 struct knlist *knl; 3589 3590 switch (kn->kn_filter) { 3591 case EVFILT_READ: 3592 kn->kn_fop = &soread_filtops; 3593 knl = &so->so_rdsel.si_note; 3594 sb = &so->so_rcv; 3595 break; 3596 case EVFILT_WRITE: 3597 kn->kn_fop = &sowrite_filtops; 3598 knl = &so->so_wrsel.si_note; 3599 sb = &so->so_snd; 3600 break; 3601 case EVFILT_EMPTY: 3602 kn->kn_fop = &soempty_filtops; 3603 knl = &so->so_wrsel.si_note; 3604 sb = &so->so_snd; 3605 break; 3606 default: 3607 return (EINVAL); 3608 } 3609 3610 SOCK_LOCK(so); 3611 if (SOLISTENING(so)) { 3612 knlist_add(knl, kn, 1); 3613 } else { 3614 SOCKBUF_LOCK(sb); 3615 knlist_add(knl, kn, 1); 3616 sb->sb_flags |= SB_KNOTE; 3617 SOCKBUF_UNLOCK(sb); 3618 } 3619 SOCK_UNLOCK(so); 3620 return (0); 3621 } 3622 3623 /* 3624 * Some routines that return EOPNOTSUPP for entry points that are not 3625 * supported by a protocol. Fill in as needed. 3626 */ 3627 int 3628 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 3629 { 3630 3631 return EOPNOTSUPP; 3632 } 3633 3634 int 3635 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job) 3636 { 3637 3638 return EOPNOTSUPP; 3639 } 3640 3641 int 3642 pru_attach_notsupp(struct socket *so, int proto, struct thread *td) 3643 { 3644 3645 return EOPNOTSUPP; 3646 } 3647 3648 int 3649 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3650 { 3651 3652 return EOPNOTSUPP; 3653 } 3654 3655 int 3656 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3657 struct thread *td) 3658 { 3659 3660 return EOPNOTSUPP; 3661 } 3662 3663 int 3664 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3665 { 3666 3667 return EOPNOTSUPP; 3668 } 3669 3670 int 3671 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3672 struct thread *td) 3673 { 3674 3675 return EOPNOTSUPP; 3676 } 3677 3678 int 3679 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 3680 { 3681 3682 return EOPNOTSUPP; 3683 } 3684 3685 int 3686 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 3687 struct ifnet *ifp, struct thread *td) 3688 { 3689 3690 return EOPNOTSUPP; 3691 } 3692 3693 int 3694 pru_disconnect_notsupp(struct socket *so) 3695 { 3696 3697 return EOPNOTSUPP; 3698 } 3699 3700 int 3701 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) 3702 { 3703 3704 return EOPNOTSUPP; 3705 } 3706 3707 int 3708 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) 3709 { 3710 3711 return EOPNOTSUPP; 3712 } 3713 3714 int 3715 pru_rcvd_notsupp(struct socket *so, int flags) 3716 { 3717 3718 return EOPNOTSUPP; 3719 } 3720 3721 int 3722 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 3723 { 3724 3725 return EOPNOTSUPP; 3726 } 3727 3728 int 3729 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, 3730 struct sockaddr *addr, struct mbuf *control, struct thread *td) 3731 { 3732 3733 if (control != NULL) 3734 m_freem(control); 3735 if ((flags & PRUS_NOTREADY) == 0) 3736 m_freem(m); 3737 return (EOPNOTSUPP); 3738 } 3739 3740 int 3741 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count) 3742 { 3743 3744 return (EOPNOTSUPP); 3745 } 3746 3747 /* 3748 * This isn't really a ``null'' operation, but it's the default one and 3749 * doesn't do anything destructive. 3750 */ 3751 int 3752 pru_sense_null(struct socket *so, struct stat *sb) 3753 { 3754 3755 sb->st_blksize = so->so_snd.sb_hiwat; 3756 return 0; 3757 } 3758 3759 int 3760 pru_shutdown_notsupp(struct socket *so) 3761 { 3762 3763 return EOPNOTSUPP; 3764 } 3765 3766 int 3767 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) 3768 { 3769 3770 return EOPNOTSUPP; 3771 } 3772 3773 int 3774 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 3775 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 3776 { 3777 3778 return EOPNOTSUPP; 3779 } 3780 3781 int 3782 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 3783 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 3784 { 3785 3786 return EOPNOTSUPP; 3787 } 3788 3789 int 3790 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, 3791 struct thread *td) 3792 { 3793 3794 return EOPNOTSUPP; 3795 } 3796 3797 static void 3798 filt_sordetach(struct knote *kn) 3799 { 3800 struct socket *so = kn->kn_fp->f_data; 3801 3802 so_rdknl_lock(so); 3803 knlist_remove(&so->so_rdsel.si_note, kn, 1); 3804 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note)) 3805 so->so_rcv.sb_flags &= ~SB_KNOTE; 3806 so_rdknl_unlock(so); 3807 } 3808 3809 /*ARGSUSED*/ 3810 static int 3811 filt_soread(struct knote *kn, long hint) 3812 { 3813 struct socket *so; 3814 3815 so = kn->kn_fp->f_data; 3816 3817 if (SOLISTENING(so)) { 3818 SOCK_LOCK_ASSERT(so); 3819 kn->kn_data = so->sol_qlen; 3820 if (so->so_error) { 3821 kn->kn_flags |= EV_EOF; 3822 kn->kn_fflags = so->so_error; 3823 return (1); 3824 } 3825 return (!TAILQ_EMPTY(&so->sol_comp)); 3826 } 3827 3828 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3829 3830 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 3831 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3832 kn->kn_flags |= EV_EOF; 3833 kn->kn_fflags = so->so_error; 3834 return (1); 3835 } else if (so->so_error) /* temporary udp error */ 3836 return (1); 3837 3838 if (kn->kn_sfflags & NOTE_LOWAT) { 3839 if (kn->kn_data >= kn->kn_sdata) 3840 return (1); 3841 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) 3842 return (1); 3843 3844 /* This hook returning non-zero indicates an event, not error */ 3845 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); 3846 } 3847 3848 static void 3849 filt_sowdetach(struct knote *kn) 3850 { 3851 struct socket *so = kn->kn_fp->f_data; 3852 3853 so_wrknl_lock(so); 3854 knlist_remove(&so->so_wrsel.si_note, kn, 1); 3855 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note)) 3856 so->so_snd.sb_flags &= ~SB_KNOTE; 3857 so_wrknl_unlock(so); 3858 } 3859 3860 /*ARGSUSED*/ 3861 static int 3862 filt_sowrite(struct knote *kn, long hint) 3863 { 3864 struct socket *so; 3865 3866 so = kn->kn_fp->f_data; 3867 3868 if (SOLISTENING(so)) 3869 return (0); 3870 3871 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3872 kn->kn_data = sbspace(&so->so_snd); 3873 3874 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); 3875 3876 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3877 kn->kn_flags |= EV_EOF; 3878 kn->kn_fflags = so->so_error; 3879 return (1); 3880 } else if (so->so_error) /* temporary udp error */ 3881 return (1); 3882 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3883 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3884 return (0); 3885 else if (kn->kn_sfflags & NOTE_LOWAT) 3886 return (kn->kn_data >= kn->kn_sdata); 3887 else 3888 return (kn->kn_data >= so->so_snd.sb_lowat); 3889 } 3890 3891 static int 3892 filt_soempty(struct knote *kn, long hint) 3893 { 3894 struct socket *so; 3895 3896 so = kn->kn_fp->f_data; 3897 3898 if (SOLISTENING(so)) 3899 return (1); 3900 3901 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3902 kn->kn_data = sbused(&so->so_snd); 3903 3904 if (kn->kn_data == 0) 3905 return (1); 3906 else 3907 return (0); 3908 } 3909 3910 int 3911 socheckuid(struct socket *so, uid_t uid) 3912 { 3913 3914 if (so == NULL) 3915 return (EPERM); 3916 if (so->so_cred->cr_uid != uid) 3917 return (EPERM); 3918 return (0); 3919 } 3920 3921 /* 3922 * These functions are used by protocols to notify the socket layer (and its 3923 * consumers) of state changes in the sockets driven by protocol-side events. 3924 */ 3925 3926 /* 3927 * Procedures to manipulate state flags of socket and do appropriate wakeups. 3928 * 3929 * Normal sequence from the active (originating) side is that 3930 * soisconnecting() is called during processing of connect() call, resulting 3931 * in an eventual call to soisconnected() if/when the connection is 3932 * established. When the connection is torn down soisdisconnecting() is 3933 * called during processing of disconnect() call, and soisdisconnected() is 3934 * called when the connection to the peer is totally severed. The semantics 3935 * of these routines are such that connectionless protocols can call 3936 * soisconnected() and soisdisconnected() only, bypassing the in-progress 3937 * calls when setting up a ``connection'' takes no time. 3938 * 3939 * From the passive side, a socket is created with two queues of sockets: 3940 * so_incomp for connections in progress and so_comp for connections already 3941 * made and awaiting user acceptance. As a protocol is preparing incoming 3942 * connections, it creates a socket structure queued on so_incomp by calling 3943 * sonewconn(). When the connection is established, soisconnected() is 3944 * called, and transfers the socket structure to so_comp, making it available 3945 * to accept(). 3946 * 3947 * If a socket is closed with sockets on either so_incomp or so_comp, these 3948 * sockets are dropped. 3949 * 3950 * If higher-level protocols are implemented in the kernel, the wakeups done 3951 * here will sometimes cause software-interrupt process scheduling. 3952 */ 3953 void 3954 soisconnecting(struct socket *so) 3955 { 3956 3957 SOCK_LOCK(so); 3958 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 3959 so->so_state |= SS_ISCONNECTING; 3960 SOCK_UNLOCK(so); 3961 } 3962 3963 void 3964 soisconnected(struct socket *so) 3965 { 3966 3967 SOCK_LOCK(so); 3968 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 3969 so->so_state |= SS_ISCONNECTED; 3970 3971 if (so->so_qstate == SQ_INCOMP) { 3972 struct socket *head = so->so_listen; 3973 int ret; 3974 3975 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so)); 3976 /* 3977 * Promoting a socket from incomplete queue to complete, we 3978 * need to go through reverse order of locking. We first do 3979 * trylock, and if that doesn't succeed, we go the hard way 3980 * leaving a reference and rechecking consistency after proper 3981 * locking. 3982 */ 3983 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) { 3984 soref(head); 3985 SOCK_UNLOCK(so); 3986 SOLISTEN_LOCK(head); 3987 SOCK_LOCK(so); 3988 if (__predict_false(head != so->so_listen)) { 3989 /* 3990 * The socket went off the listen queue, 3991 * should be lost race to close(2) of sol. 3992 * The socket is about to soabort(). 3993 */ 3994 SOCK_UNLOCK(so); 3995 sorele(head); 3996 return; 3997 } 3998 /* Not the last one, as so holds a ref. */ 3999 refcount_release(&head->so_count); 4000 } 4001 again: 4002 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 4003 TAILQ_REMOVE(&head->sol_incomp, so, so_list); 4004 head->sol_incqlen--; 4005 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 4006 head->sol_qlen++; 4007 so->so_qstate = SQ_COMP; 4008 SOCK_UNLOCK(so); 4009 solisten_wakeup(head); /* unlocks */ 4010 } else { 4011 SOCKBUF_LOCK(&so->so_rcv); 4012 soupcall_set(so, SO_RCV, 4013 head->sol_accept_filter->accf_callback, 4014 head->sol_accept_filter_arg); 4015 so->so_options &= ~SO_ACCEPTFILTER; 4016 ret = head->sol_accept_filter->accf_callback(so, 4017 head->sol_accept_filter_arg, M_NOWAIT); 4018 if (ret == SU_ISCONNECTED) { 4019 soupcall_clear(so, SO_RCV); 4020 SOCKBUF_UNLOCK(&so->so_rcv); 4021 goto again; 4022 } 4023 SOCKBUF_UNLOCK(&so->so_rcv); 4024 SOCK_UNLOCK(so); 4025 SOLISTEN_UNLOCK(head); 4026 } 4027 return; 4028 } 4029 SOCK_UNLOCK(so); 4030 wakeup(&so->so_timeo); 4031 sorwakeup(so); 4032 sowwakeup(so); 4033 } 4034 4035 void 4036 soisdisconnecting(struct socket *so) 4037 { 4038 4039 SOCK_LOCK(so); 4040 so->so_state &= ~SS_ISCONNECTING; 4041 so->so_state |= SS_ISDISCONNECTING; 4042 4043 if (!SOLISTENING(so)) { 4044 SOCKBUF_LOCK(&so->so_rcv); 4045 socantrcvmore_locked(so); 4046 SOCKBUF_LOCK(&so->so_snd); 4047 socantsendmore_locked(so); 4048 } 4049 SOCK_UNLOCK(so); 4050 wakeup(&so->so_timeo); 4051 } 4052 4053 void 4054 soisdisconnected(struct socket *so) 4055 { 4056 4057 SOCK_LOCK(so); 4058 4059 /* 4060 * There is at least one reader of so_state that does not 4061 * acquire socket lock, namely soreceive_generic(). Ensure 4062 * that it never sees all flags that track connection status 4063 * cleared, by ordering the update with a barrier semantic of 4064 * our release thread fence. 4065 */ 4066 so->so_state |= SS_ISDISCONNECTED; 4067 atomic_thread_fence_rel(); 4068 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 4069 4070 if (!SOLISTENING(so)) { 4071 SOCK_UNLOCK(so); 4072 SOCKBUF_LOCK(&so->so_rcv); 4073 socantrcvmore_locked(so); 4074 SOCKBUF_LOCK(&so->so_snd); 4075 sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); 4076 socantsendmore_locked(so); 4077 } else 4078 SOCK_UNLOCK(so); 4079 wakeup(&so->so_timeo); 4080 } 4081 4082 /* 4083 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 4084 */ 4085 struct sockaddr * 4086 sodupsockaddr(const struct sockaddr *sa, int mflags) 4087 { 4088 struct sockaddr *sa2; 4089 4090 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 4091 if (sa2) 4092 bcopy(sa, sa2, sa->sa_len); 4093 return sa2; 4094 } 4095 4096 /* 4097 * Register per-socket destructor. 4098 */ 4099 void 4100 sodtor_set(struct socket *so, so_dtor_t *func) 4101 { 4102 4103 SOCK_LOCK_ASSERT(so); 4104 so->so_dtor = func; 4105 } 4106 4107 /* 4108 * Register per-socket buffer upcalls. 4109 */ 4110 void 4111 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg) 4112 { 4113 struct sockbuf *sb; 4114 4115 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4116 4117 switch (which) { 4118 case SO_RCV: 4119 sb = &so->so_rcv; 4120 break; 4121 case SO_SND: 4122 sb = &so->so_snd; 4123 break; 4124 default: 4125 panic("soupcall_set: bad which"); 4126 } 4127 SOCKBUF_LOCK_ASSERT(sb); 4128 sb->sb_upcall = func; 4129 sb->sb_upcallarg = arg; 4130 sb->sb_flags |= SB_UPCALL; 4131 } 4132 4133 void 4134 soupcall_clear(struct socket *so, int which) 4135 { 4136 struct sockbuf *sb; 4137 4138 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4139 4140 switch (which) { 4141 case SO_RCV: 4142 sb = &so->so_rcv; 4143 break; 4144 case SO_SND: 4145 sb = &so->so_snd; 4146 break; 4147 default: 4148 panic("soupcall_clear: bad which"); 4149 } 4150 SOCKBUF_LOCK_ASSERT(sb); 4151 KASSERT(sb->sb_upcall != NULL, 4152 ("%s: so %p no upcall to clear", __func__, so)); 4153 sb->sb_upcall = NULL; 4154 sb->sb_upcallarg = NULL; 4155 sb->sb_flags &= ~SB_UPCALL; 4156 } 4157 4158 void 4159 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg) 4160 { 4161 4162 SOLISTEN_LOCK_ASSERT(so); 4163 so->sol_upcall = func; 4164 so->sol_upcallarg = arg; 4165 } 4166 4167 static void 4168 so_rdknl_lock(void *arg) 4169 { 4170 struct socket *so = arg; 4171 4172 if (SOLISTENING(so)) 4173 SOCK_LOCK(so); 4174 else 4175 SOCKBUF_LOCK(&so->so_rcv); 4176 } 4177 4178 static void 4179 so_rdknl_unlock(void *arg) 4180 { 4181 struct socket *so = arg; 4182 4183 if (SOLISTENING(so)) 4184 SOCK_UNLOCK(so); 4185 else 4186 SOCKBUF_UNLOCK(&so->so_rcv); 4187 } 4188 4189 static void 4190 so_rdknl_assert_lock(void *arg, int what) 4191 { 4192 struct socket *so = arg; 4193 4194 if (what == LA_LOCKED) { 4195 if (SOLISTENING(so)) 4196 SOCK_LOCK_ASSERT(so); 4197 else 4198 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 4199 } else { 4200 if (SOLISTENING(so)) 4201 SOCK_UNLOCK_ASSERT(so); 4202 else 4203 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); 4204 } 4205 } 4206 4207 static void 4208 so_wrknl_lock(void *arg) 4209 { 4210 struct socket *so = arg; 4211 4212 if (SOLISTENING(so)) 4213 SOCK_LOCK(so); 4214 else 4215 SOCKBUF_LOCK(&so->so_snd); 4216 } 4217 4218 static void 4219 so_wrknl_unlock(void *arg) 4220 { 4221 struct socket *so = arg; 4222 4223 if (SOLISTENING(so)) 4224 SOCK_UNLOCK(so); 4225 else 4226 SOCKBUF_UNLOCK(&so->so_snd); 4227 } 4228 4229 static void 4230 so_wrknl_assert_lock(void *arg, int what) 4231 { 4232 struct socket *so = arg; 4233 4234 if (what == LA_LOCKED) { 4235 if (SOLISTENING(so)) 4236 SOCK_LOCK_ASSERT(so); 4237 else 4238 SOCKBUF_LOCK_ASSERT(&so->so_snd); 4239 } else { 4240 if (SOLISTENING(so)) 4241 SOCK_UNLOCK_ASSERT(so); 4242 else 4243 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 4244 } 4245 } 4246 4247 /* 4248 * Create an external-format (``xsocket'') structure using the information in 4249 * the kernel-format socket structure pointed to by so. This is done to 4250 * reduce the spew of irrelevant information over this interface, to isolate 4251 * user code from changes in the kernel structure, and potentially to provide 4252 * information-hiding if we decide that some of this information should be 4253 * hidden from users. 4254 */ 4255 void 4256 sotoxsocket(struct socket *so, struct xsocket *xso) 4257 { 4258 4259 bzero(xso, sizeof(*xso)); 4260 xso->xso_len = sizeof *xso; 4261 xso->xso_so = (uintptr_t)so; 4262 xso->so_type = so->so_type; 4263 xso->so_options = so->so_options; 4264 xso->so_linger = so->so_linger; 4265 xso->so_state = so->so_state; 4266 xso->so_pcb = (uintptr_t)so->so_pcb; 4267 xso->xso_protocol = so->so_proto->pr_protocol; 4268 xso->xso_family = so->so_proto->pr_domain->dom_family; 4269 xso->so_timeo = so->so_timeo; 4270 xso->so_error = so->so_error; 4271 xso->so_uid = so->so_cred->cr_uid; 4272 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 4273 if (SOLISTENING(so)) { 4274 xso->so_qlen = so->sol_qlen; 4275 xso->so_incqlen = so->sol_incqlen; 4276 xso->so_qlimit = so->sol_qlimit; 4277 xso->so_oobmark = 0; 4278 } else { 4279 xso->so_state |= so->so_qstate; 4280 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0; 4281 xso->so_oobmark = so->so_oobmark; 4282 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 4283 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 4284 } 4285 } 4286 4287 struct sockbuf * 4288 so_sockbuf_rcv(struct socket *so) 4289 { 4290 4291 return (&so->so_rcv); 4292 } 4293 4294 struct sockbuf * 4295 so_sockbuf_snd(struct socket *so) 4296 { 4297 4298 return (&so->so_snd); 4299 } 4300 4301 int 4302 so_state_get(const struct socket *so) 4303 { 4304 4305 return (so->so_state); 4306 } 4307 4308 void 4309 so_state_set(struct socket *so, int val) 4310 { 4311 4312 so->so_state = val; 4313 } 4314 4315 int 4316 so_options_get(const struct socket *so) 4317 { 4318 4319 return (so->so_options); 4320 } 4321 4322 void 4323 so_options_set(struct socket *so, int val) 4324 { 4325 4326 so->so_options = val; 4327 } 4328 4329 int 4330 so_error_get(const struct socket *so) 4331 { 4332 4333 return (so->so_error); 4334 } 4335 4336 void 4337 so_error_set(struct socket *so, int val) 4338 { 4339 4340 so->so_error = val; 4341 } 4342 4343 int 4344 so_linger_get(const struct socket *so) 4345 { 4346 4347 return (so->so_linger); 4348 } 4349 4350 void 4351 so_linger_set(struct socket *so, int val) 4352 { 4353 4354 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz), 4355 ("%s: val %d out of range", __func__, val)); 4356 4357 so->so_linger = val; 4358 } 4359 4360 struct protosw * 4361 so_protosw_get(const struct socket *so) 4362 { 4363 4364 return (so->so_proto); 4365 } 4366 4367 void 4368 so_protosw_set(struct socket *so, struct protosw *val) 4369 { 4370 4371 so->so_proto = val; 4372 } 4373 4374 void 4375 so_sorwakeup(struct socket *so) 4376 { 4377 4378 sorwakeup(so); 4379 } 4380 4381 void 4382 so_sowwakeup(struct socket *so) 4383 { 4384 4385 sowwakeup(so); 4386 } 4387 4388 void 4389 so_sorwakeup_locked(struct socket *so) 4390 { 4391 4392 sorwakeup_locked(so); 4393 } 4394 4395 void 4396 so_sowwakeup_locked(struct socket *so) 4397 { 4398 4399 sowwakeup_locked(so); 4400 } 4401 4402 void 4403 so_lock(struct socket *so) 4404 { 4405 4406 SOCK_LOCK(so); 4407 } 4408 4409 void 4410 so_unlock(struct socket *so) 4411 { 4412 4413 SOCK_UNLOCK(so); 4414 } 4415