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 crfree(so->so_cred); 479 khelp_destroy_osd(&so->osd); 480 if (SOLISTENING(so)) { 481 if (so->sol_accept_filter != NULL) 482 accept_filt_setopt(so, NULL); 483 } else { 484 if (so->so_rcv.sb_hiwat) 485 (void)chgsbsize(so->so_cred->cr_uidinfo, 486 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 487 if (so->so_snd.sb_hiwat) 488 (void)chgsbsize(so->so_cred->cr_uidinfo, 489 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 490 sx_destroy(&so->so_snd.sb_sx); 491 sx_destroy(&so->so_rcv.sb_sx); 492 SOCKBUF_LOCK_DESTROY(&so->so_snd); 493 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 494 } 495 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_linger = head->so_linger; 722 so->so_state = head->so_state | SS_NOFDREF; 723 so->so_fibnum = head->so_fibnum; 724 so->so_proto = head->so_proto; 725 so->so_cred = crhold(head->so_cred); 726 #ifdef MAC 727 mac_socket_newconn(head, so); 728 #endif 729 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 730 so_rdknl_assert_lock); 731 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 732 so_wrknl_assert_lock); 733 VNET_SO_ASSERT(head); 734 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) { 735 sodealloc(so); 736 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 737 __func__, head->so_pcb); 738 return (NULL); 739 } 740 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 741 sodealloc(so); 742 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 743 __func__, head->so_pcb); 744 return (NULL); 745 } 746 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat; 747 so->so_snd.sb_lowat = head->sol_sbsnd_lowat; 748 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo; 749 so->so_snd.sb_timeo = head->sol_sbsnd_timeo; 750 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE; 751 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE; 752 753 SOLISTEN_LOCK(head); 754 if (head->sol_accept_filter != NULL) 755 connstatus = 0; 756 so->so_state |= connstatus; 757 so->so_options = head->so_options & ~SO_ACCEPTCONN; 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, &head->so_lock, 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 bool listening; 1180 int error = 0; 1181 1182 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter")); 1183 1184 CURVNET_SET(so->so_vnet); 1185 funsetown(&so->so_sigio); 1186 if (so->so_state & SS_ISCONNECTED) { 1187 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 1188 error = sodisconnect(so); 1189 if (error) { 1190 if (error == ENOTCONN) 1191 error = 0; 1192 goto drop; 1193 } 1194 } 1195 if (so->so_options & SO_LINGER) { 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 ((listening = (so->so_options & SO_ACCEPTCONN))) { 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 (listening) { 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 if (so->so_options & SO_ACCEPTCONN) 1316 return (EOPNOTSUPP); 1317 1318 CURVNET_SET(so->so_vnet); 1319 /* 1320 * If protocol is connection-based, can only connect once. 1321 * Otherwise, if connected, try to disconnect first. This allows 1322 * user to disconnect by connecting to, e.g., a null address. 1323 */ 1324 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 1325 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 1326 (error = sodisconnect(so)))) { 1327 error = EISCONN; 1328 } else { 1329 /* 1330 * Prevent accumulated error from previous connection from 1331 * biting us. 1332 */ 1333 so->so_error = 0; 1334 if (fd == AT_FDCWD) { 1335 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, 1336 nam, td); 1337 } else { 1338 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd, 1339 so, nam, td); 1340 } 1341 } 1342 CURVNET_RESTORE(); 1343 1344 return (error); 1345 } 1346 1347 int 1348 soconnect2(struct socket *so1, struct socket *so2) 1349 { 1350 int error; 1351 1352 CURVNET_SET(so1->so_vnet); 1353 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); 1354 CURVNET_RESTORE(); 1355 return (error); 1356 } 1357 1358 int 1359 sodisconnect(struct socket *so) 1360 { 1361 int error; 1362 1363 if ((so->so_state & SS_ISCONNECTED) == 0) 1364 return (ENOTCONN); 1365 if (so->so_state & SS_ISDISCONNECTING) 1366 return (EALREADY); 1367 VNET_SO_ASSERT(so); 1368 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 1369 return (error); 1370 } 1371 1372 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) 1373 1374 int 1375 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1376 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1377 { 1378 long space; 1379 ssize_t resid; 1380 int clen = 0, error, dontroute; 1381 1382 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 1383 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 1384 ("sosend_dgram: !PR_ATOMIC")); 1385 1386 if (uio != NULL) 1387 resid = uio->uio_resid; 1388 else 1389 resid = top->m_pkthdr.len; 1390 /* 1391 * In theory resid should be unsigned. However, space must be 1392 * signed, as it might be less than 0 if we over-committed, and we 1393 * must use a signed comparison of space and resid. On the other 1394 * hand, a negative resid causes us to loop sending 0-length 1395 * segments to the protocol. 1396 */ 1397 if (resid < 0) { 1398 error = EINVAL; 1399 goto out; 1400 } 1401 1402 dontroute = 1403 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 1404 if (td != NULL) 1405 td->td_ru.ru_msgsnd++; 1406 if (control != NULL) 1407 clen = control->m_len; 1408 1409 SOCKBUF_LOCK(&so->so_snd); 1410 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1411 SOCKBUF_UNLOCK(&so->so_snd); 1412 error = EPIPE; 1413 goto out; 1414 } 1415 if (so->so_error) { 1416 error = so->so_error; 1417 so->so_error = 0; 1418 SOCKBUF_UNLOCK(&so->so_snd); 1419 goto out; 1420 } 1421 if ((so->so_state & SS_ISCONNECTED) == 0) { 1422 /* 1423 * `sendto' and `sendmsg' is allowed on a connection-based 1424 * socket if it supports implied connect. Return ENOTCONN if 1425 * not connected and no address is supplied. 1426 */ 1427 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1428 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1429 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1430 !(resid == 0 && clen != 0)) { 1431 SOCKBUF_UNLOCK(&so->so_snd); 1432 error = ENOTCONN; 1433 goto out; 1434 } 1435 } else if (addr == NULL) { 1436 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1437 error = ENOTCONN; 1438 else 1439 error = EDESTADDRREQ; 1440 SOCKBUF_UNLOCK(&so->so_snd); 1441 goto out; 1442 } 1443 } 1444 1445 /* 1446 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1447 * problem and need fixing. 1448 */ 1449 space = sbspace(&so->so_snd); 1450 if (flags & MSG_OOB) 1451 space += 1024; 1452 space -= clen; 1453 SOCKBUF_UNLOCK(&so->so_snd); 1454 if (resid > space) { 1455 error = EMSGSIZE; 1456 goto out; 1457 } 1458 if (uio == NULL) { 1459 resid = 0; 1460 if (flags & MSG_EOR) 1461 top->m_flags |= M_EOR; 1462 } else { 1463 /* 1464 * Copy the data from userland into a mbuf chain. 1465 * If no data is to be copied in, a single empty mbuf 1466 * is returned. 1467 */ 1468 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1469 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1470 if (top == NULL) { 1471 error = EFAULT; /* only possible error */ 1472 goto out; 1473 } 1474 space -= resid - uio->uio_resid; 1475 resid = uio->uio_resid; 1476 } 1477 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1478 /* 1479 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1480 * than with. 1481 */ 1482 if (dontroute) { 1483 SOCK_LOCK(so); 1484 so->so_options |= SO_DONTROUTE; 1485 SOCK_UNLOCK(so); 1486 } 1487 /* 1488 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1489 * of date. We could have received a reset packet in an interrupt or 1490 * maybe we slept while doing page faults in uiomove() etc. We could 1491 * probably recheck again inside the locking protection here, but 1492 * there are probably other places that this also happens. We must 1493 * rethink this. 1494 */ 1495 VNET_SO_ASSERT(so); 1496 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1497 (flags & MSG_OOB) ? PRUS_OOB : 1498 /* 1499 * If the user set MSG_EOF, the protocol understands this flag and 1500 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1501 */ 1502 ((flags & MSG_EOF) && 1503 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1504 (resid <= 0)) ? 1505 PRUS_EOF : 1506 /* If there is more to send set PRUS_MORETOCOME */ 1507 (flags & MSG_MORETOCOME) || 1508 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1509 top, addr, control, td); 1510 if (dontroute) { 1511 SOCK_LOCK(so); 1512 so->so_options &= ~SO_DONTROUTE; 1513 SOCK_UNLOCK(so); 1514 } 1515 clen = 0; 1516 control = NULL; 1517 top = NULL; 1518 out: 1519 if (top != NULL) 1520 m_freem(top); 1521 if (control != NULL) 1522 m_freem(control); 1523 return (error); 1524 } 1525 1526 /* 1527 * Send on a socket. If send must go all at once and message is larger than 1528 * send buffering, then hard error. Lock against other senders. If must go 1529 * all at once and not enough room now, then inform user that this would 1530 * block and do nothing. Otherwise, if nonblocking, send as much as 1531 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1532 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1533 * in mbuf chain must be small enough to send all at once. 1534 * 1535 * Returns nonzero on error, timeout or signal; callers must check for short 1536 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1537 * on return. 1538 */ 1539 int 1540 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1541 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1542 { 1543 long space; 1544 ssize_t resid; 1545 int clen = 0, error, dontroute; 1546 int atomic = sosendallatonce(so) || top; 1547 int pru_flag; 1548 #ifdef KERN_TLS 1549 struct ktls_session *tls; 1550 int tls_enq_cnt, tls_pruflag; 1551 uint8_t tls_rtype; 1552 1553 tls = NULL; 1554 tls_rtype = TLS_RLTYPE_APP; 1555 #endif 1556 if (uio != NULL) 1557 resid = uio->uio_resid; 1558 else if ((top->m_flags & M_PKTHDR) != 0) 1559 resid = top->m_pkthdr.len; 1560 else 1561 resid = m_length(top, NULL); 1562 /* 1563 * In theory resid should be unsigned. However, space must be 1564 * signed, as it might be less than 0 if we over-committed, and we 1565 * must use a signed comparison of space and resid. On the other 1566 * hand, a negative resid causes us to loop sending 0-length 1567 * segments to the protocol. 1568 * 1569 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1570 * type sockets since that's an error. 1571 */ 1572 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1573 error = EINVAL; 1574 goto out; 1575 } 1576 1577 dontroute = 1578 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1579 (so->so_proto->pr_flags & PR_ATOMIC); 1580 if (td != NULL) 1581 td->td_ru.ru_msgsnd++; 1582 if (control != NULL) 1583 clen = control->m_len; 1584 1585 error = sblock(&so->so_snd, SBLOCKWAIT(flags)); 1586 if (error) 1587 goto out; 1588 1589 #ifdef KERN_TLS 1590 tls_pruflag = 0; 1591 tls = ktls_hold(so->so_snd.sb_tls_info); 1592 if (tls != NULL) { 1593 if (tls->mode == TCP_TLS_MODE_SW) 1594 tls_pruflag = PRUS_NOTREADY; 1595 1596 if (control != NULL) { 1597 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1598 1599 if (clen >= sizeof(*cm) && 1600 cm->cmsg_type == TLS_SET_RECORD_TYPE) { 1601 tls_rtype = *((uint8_t *)CMSG_DATA(cm)); 1602 clen = 0; 1603 m_freem(control); 1604 control = NULL; 1605 atomic = 1; 1606 } 1607 } 1608 } 1609 #endif 1610 1611 restart: 1612 do { 1613 SOCKBUF_LOCK(&so->so_snd); 1614 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1615 SOCKBUF_UNLOCK(&so->so_snd); 1616 error = EPIPE; 1617 goto release; 1618 } 1619 if (so->so_error) { 1620 error = so->so_error; 1621 so->so_error = 0; 1622 SOCKBUF_UNLOCK(&so->so_snd); 1623 goto release; 1624 } 1625 if ((so->so_state & SS_ISCONNECTED) == 0) { 1626 /* 1627 * `sendto' and `sendmsg' is allowed on a connection- 1628 * based socket if it supports implied connect. 1629 * Return ENOTCONN if not connected and no address is 1630 * supplied. 1631 */ 1632 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1633 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1634 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1635 !(resid == 0 && clen != 0)) { 1636 SOCKBUF_UNLOCK(&so->so_snd); 1637 error = ENOTCONN; 1638 goto release; 1639 } 1640 } else if (addr == NULL) { 1641 SOCKBUF_UNLOCK(&so->so_snd); 1642 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1643 error = ENOTCONN; 1644 else 1645 error = EDESTADDRREQ; 1646 goto release; 1647 } 1648 } 1649 space = sbspace(&so->so_snd); 1650 if (flags & MSG_OOB) 1651 space += 1024; 1652 if ((atomic && resid > so->so_snd.sb_hiwat) || 1653 clen > so->so_snd.sb_hiwat) { 1654 SOCKBUF_UNLOCK(&so->so_snd); 1655 error = EMSGSIZE; 1656 goto release; 1657 } 1658 if (space < resid + clen && 1659 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1660 if ((so->so_state & SS_NBIO) || 1661 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) { 1662 SOCKBUF_UNLOCK(&so->so_snd); 1663 error = EWOULDBLOCK; 1664 goto release; 1665 } 1666 error = sbwait(&so->so_snd); 1667 SOCKBUF_UNLOCK(&so->so_snd); 1668 if (error) 1669 goto release; 1670 goto restart; 1671 } 1672 SOCKBUF_UNLOCK(&so->so_snd); 1673 space -= clen; 1674 do { 1675 if (uio == NULL) { 1676 resid = 0; 1677 if (flags & MSG_EOR) 1678 top->m_flags |= M_EOR; 1679 #ifdef KERN_TLS 1680 if (tls != NULL) { 1681 ktls_frame(top, tls, &tls_enq_cnt, 1682 tls_rtype); 1683 tls_rtype = TLS_RLTYPE_APP; 1684 } 1685 #endif 1686 } else { 1687 /* 1688 * Copy the data from userland into a mbuf 1689 * chain. If resid is 0, which can happen 1690 * only if we have control to send, then 1691 * a single empty mbuf is returned. This 1692 * is a workaround to prevent protocol send 1693 * methods to panic. 1694 */ 1695 #ifdef KERN_TLS 1696 if (tls != NULL) { 1697 top = m_uiotombuf(uio, M_WAITOK, space, 1698 tls->params.max_frame_len, 1699 M_EXTPG | 1700 ((flags & MSG_EOR) ? M_EOR : 0)); 1701 if (top != NULL) { 1702 ktls_frame(top, tls, 1703 &tls_enq_cnt, tls_rtype); 1704 } 1705 tls_rtype = TLS_RLTYPE_APP; 1706 } else 1707 #endif 1708 top = m_uiotombuf(uio, M_WAITOK, space, 1709 (atomic ? max_hdr : 0), 1710 (atomic ? M_PKTHDR : 0) | 1711 ((flags & MSG_EOR) ? M_EOR : 0)); 1712 if (top == NULL) { 1713 error = EFAULT; /* only possible error */ 1714 goto release; 1715 } 1716 space -= resid - uio->uio_resid; 1717 resid = uio->uio_resid; 1718 } 1719 if (dontroute) { 1720 SOCK_LOCK(so); 1721 so->so_options |= SO_DONTROUTE; 1722 SOCK_UNLOCK(so); 1723 } 1724 /* 1725 * XXX all the SBS_CANTSENDMORE checks previously 1726 * done could be out of date. We could have received 1727 * a reset packet in an interrupt or maybe we slept 1728 * while doing page faults in uiomove() etc. We 1729 * could probably recheck again inside the locking 1730 * protection here, but there are probably other 1731 * places that this also happens. We must rethink 1732 * this. 1733 */ 1734 VNET_SO_ASSERT(so); 1735 1736 pru_flag = (flags & MSG_OOB) ? PRUS_OOB : 1737 /* 1738 * If the user set MSG_EOF, the protocol understands 1739 * this flag and nothing left to send then use 1740 * PRU_SEND_EOF instead of PRU_SEND. 1741 */ 1742 ((flags & MSG_EOF) && 1743 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1744 (resid <= 0)) ? 1745 PRUS_EOF : 1746 /* If there is more to send set PRUS_MORETOCOME. */ 1747 (flags & MSG_MORETOCOME) || 1748 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; 1749 1750 #ifdef KERN_TLS 1751 pru_flag |= tls_pruflag; 1752 #endif 1753 1754 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1755 pru_flag, top, addr, control, td); 1756 1757 if (dontroute) { 1758 SOCK_LOCK(so); 1759 so->so_options &= ~SO_DONTROUTE; 1760 SOCK_UNLOCK(so); 1761 } 1762 1763 #ifdef KERN_TLS 1764 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) { 1765 /* 1766 * Note that error is intentionally 1767 * ignored. 1768 * 1769 * Like sendfile(), we rely on the 1770 * completion routine (pru_ready()) 1771 * to free the mbufs in the event that 1772 * pru_send() encountered an error and 1773 * did not append them to the sockbuf. 1774 */ 1775 soref(so); 1776 ktls_enqueue(top, so, tls_enq_cnt); 1777 } 1778 #endif 1779 clen = 0; 1780 control = NULL; 1781 top = NULL; 1782 if (error) 1783 goto release; 1784 } while (resid && space > 0); 1785 } while (resid); 1786 1787 release: 1788 sbunlock(&so->so_snd); 1789 out: 1790 #ifdef KERN_TLS 1791 if (tls != NULL) 1792 ktls_free(tls); 1793 #endif 1794 if (top != NULL) 1795 m_freem(top); 1796 if (control != NULL) 1797 m_freem(control); 1798 return (error); 1799 } 1800 1801 int 1802 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1803 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1804 { 1805 int error; 1806 1807 CURVNET_SET(so->so_vnet); 1808 if (!SOLISTENING(so)) 1809 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, 1810 top, control, flags, td); 1811 else { 1812 m_freem(top); 1813 m_freem(control); 1814 error = ENOTCONN; 1815 } 1816 CURVNET_RESTORE(); 1817 return (error); 1818 } 1819 1820 /* 1821 * The part of soreceive() that implements reading non-inline out-of-band 1822 * data from a socket. For more complete comments, see soreceive(), from 1823 * which this code originated. 1824 * 1825 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1826 * unable to return an mbuf chain to the caller. 1827 */ 1828 static int 1829 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1830 { 1831 struct protosw *pr = so->so_proto; 1832 struct mbuf *m; 1833 int error; 1834 1835 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1836 VNET_SO_ASSERT(so); 1837 1838 m = m_get(M_WAITOK, MT_DATA); 1839 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 1840 if (error) 1841 goto bad; 1842 do { 1843 error = uiomove(mtod(m, void *), 1844 (int) min(uio->uio_resid, m->m_len), uio); 1845 m = m_free(m); 1846 } while (uio->uio_resid && error == 0 && m); 1847 bad: 1848 if (m != NULL) 1849 m_freem(m); 1850 return (error); 1851 } 1852 1853 /* 1854 * Following replacement or removal of the first mbuf on the first mbuf chain 1855 * of a socket buffer, push necessary state changes back into the socket 1856 * buffer so that other consumers see the values consistently. 'nextrecord' 1857 * is the callers locally stored value of the original value of 1858 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1859 * NOTE: 'nextrecord' may be NULL. 1860 */ 1861 static __inline void 1862 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1863 { 1864 1865 SOCKBUF_LOCK_ASSERT(sb); 1866 /* 1867 * First, update for the new value of nextrecord. If necessary, make 1868 * it the first record. 1869 */ 1870 if (sb->sb_mb != NULL) 1871 sb->sb_mb->m_nextpkt = nextrecord; 1872 else 1873 sb->sb_mb = nextrecord; 1874 1875 /* 1876 * Now update any dependent socket buffer fields to reflect the new 1877 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1878 * addition of a second clause that takes care of the case where 1879 * sb_mb has been updated, but remains the last record. 1880 */ 1881 if (sb->sb_mb == NULL) { 1882 sb->sb_mbtail = NULL; 1883 sb->sb_lastrecord = NULL; 1884 } else if (sb->sb_mb->m_nextpkt == NULL) 1885 sb->sb_lastrecord = sb->sb_mb; 1886 } 1887 1888 /* 1889 * Implement receive operations on a socket. We depend on the way that 1890 * records are added to the sockbuf by sbappend. In particular, each record 1891 * (mbufs linked through m_next) must begin with an address if the protocol 1892 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1893 * data, and then zero or more mbufs of data. In order to allow parallelism 1894 * between network receive and copying to user space, as well as avoid 1895 * sleeping with a mutex held, we release the socket buffer mutex during the 1896 * user space copy. Although the sockbuf is locked, new data may still be 1897 * appended, and thus we must maintain consistency of the sockbuf during that 1898 * time. 1899 * 1900 * The caller may receive the data as a single mbuf chain by supplying an 1901 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1902 * the count in uio_resid. 1903 */ 1904 int 1905 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1906 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1907 { 1908 struct mbuf *m, **mp; 1909 int flags, error, offset; 1910 ssize_t len; 1911 struct protosw *pr = so->so_proto; 1912 struct mbuf *nextrecord; 1913 int moff, type = 0; 1914 ssize_t orig_resid = uio->uio_resid; 1915 1916 mp = mp0; 1917 if (psa != NULL) 1918 *psa = NULL; 1919 if (controlp != NULL) 1920 *controlp = NULL; 1921 if (flagsp != NULL) 1922 flags = *flagsp &~ MSG_EOR; 1923 else 1924 flags = 0; 1925 if (flags & MSG_OOB) 1926 return (soreceive_rcvoob(so, uio, flags)); 1927 if (mp != NULL) 1928 *mp = NULL; 1929 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 1930 && uio->uio_resid) { 1931 VNET_SO_ASSERT(so); 1932 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 1933 } 1934 1935 error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); 1936 if (error) 1937 return (error); 1938 1939 restart: 1940 SOCKBUF_LOCK(&so->so_rcv); 1941 m = so->so_rcv.sb_mb; 1942 /* 1943 * If we have less data than requested, block awaiting more (subject 1944 * to any timeout) if: 1945 * 1. the current count is less than the low water mark, or 1946 * 2. MSG_DONTWAIT is not set 1947 */ 1948 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1949 sbavail(&so->so_rcv) < uio->uio_resid) && 1950 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && 1951 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 1952 KASSERT(m != NULL || !sbavail(&so->so_rcv), 1953 ("receive: m == %p sbavail == %u", 1954 m, sbavail(&so->so_rcv))); 1955 if (so->so_error) { 1956 if (m != NULL) 1957 goto dontblock; 1958 error = so->so_error; 1959 if ((flags & MSG_PEEK) == 0) 1960 so->so_error = 0; 1961 SOCKBUF_UNLOCK(&so->so_rcv); 1962 goto release; 1963 } 1964 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1965 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1966 if (m != NULL) 1967 goto dontblock; 1968 #ifdef KERN_TLS 1969 else if (so->so_rcv.sb_tlsdcc == 0 && 1970 so->so_rcv.sb_tlscc == 0) { 1971 #else 1972 else { 1973 #endif 1974 SOCKBUF_UNLOCK(&so->so_rcv); 1975 goto release; 1976 } 1977 } 1978 for (; m != NULL; m = m->m_next) 1979 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1980 m = so->so_rcv.sb_mb; 1981 goto dontblock; 1982 } 1983 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED | 1984 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 && 1985 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 1986 SOCKBUF_UNLOCK(&so->so_rcv); 1987 error = ENOTCONN; 1988 goto release; 1989 } 1990 if (uio->uio_resid == 0) { 1991 SOCKBUF_UNLOCK(&so->so_rcv); 1992 goto release; 1993 } 1994 if ((so->so_state & SS_NBIO) || 1995 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1996 SOCKBUF_UNLOCK(&so->so_rcv); 1997 error = EWOULDBLOCK; 1998 goto release; 1999 } 2000 SBLASTRECORDCHK(&so->so_rcv); 2001 SBLASTMBUFCHK(&so->so_rcv); 2002 error = sbwait(&so->so_rcv); 2003 SOCKBUF_UNLOCK(&so->so_rcv); 2004 if (error) 2005 goto release; 2006 goto restart; 2007 } 2008 dontblock: 2009 /* 2010 * From this point onward, we maintain 'nextrecord' as a cache of the 2011 * pointer to the next record in the socket buffer. We must keep the 2012 * various socket buffer pointers and local stack versions of the 2013 * pointers in sync, pushing out modifications before dropping the 2014 * socket buffer mutex, and re-reading them when picking it up. 2015 * 2016 * Otherwise, we will race with the network stack appending new data 2017 * or records onto the socket buffer by using inconsistent/stale 2018 * versions of the field, possibly resulting in socket buffer 2019 * corruption. 2020 * 2021 * By holding the high-level sblock(), we prevent simultaneous 2022 * readers from pulling off the front of the socket buffer. 2023 */ 2024 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2025 if (uio->uio_td) 2026 uio->uio_td->td_ru.ru_msgrcv++; 2027 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 2028 SBLASTRECORDCHK(&so->so_rcv); 2029 SBLASTMBUFCHK(&so->so_rcv); 2030 nextrecord = m->m_nextpkt; 2031 if (pr->pr_flags & PR_ADDR) { 2032 KASSERT(m->m_type == MT_SONAME, 2033 ("m->m_type == %d", m->m_type)); 2034 orig_resid = 0; 2035 if (psa != NULL) 2036 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2037 M_NOWAIT); 2038 if (flags & MSG_PEEK) { 2039 m = m->m_next; 2040 } else { 2041 sbfree(&so->so_rcv, m); 2042 so->so_rcv.sb_mb = m_free(m); 2043 m = so->so_rcv.sb_mb; 2044 sockbuf_pushsync(&so->so_rcv, nextrecord); 2045 } 2046 } 2047 2048 /* 2049 * Process one or more MT_CONTROL mbufs present before any data mbufs 2050 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 2051 * just copy the data; if !MSG_PEEK, we call into the protocol to 2052 * perform externalization (or freeing if controlp == NULL). 2053 */ 2054 if (m != NULL && m->m_type == MT_CONTROL) { 2055 struct mbuf *cm = NULL, *cmn; 2056 struct mbuf **cme = &cm; 2057 #ifdef KERN_TLS 2058 struct cmsghdr *cmsg; 2059 struct tls_get_record tgr; 2060 2061 /* 2062 * For MSG_TLSAPPDATA, check for a non-application data 2063 * record. If found, return ENXIO without removing 2064 * it from the receive queue. This allows a subsequent 2065 * call without MSG_TLSAPPDATA to receive it. 2066 * Note that, for TLS, there should only be a single 2067 * control mbuf with the TLS_GET_RECORD message in it. 2068 */ 2069 if (flags & MSG_TLSAPPDATA) { 2070 cmsg = mtod(m, struct cmsghdr *); 2071 if (cmsg->cmsg_type == TLS_GET_RECORD && 2072 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) { 2073 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr)); 2074 /* This will need to change for TLS 1.3. */ 2075 if (tgr.tls_type != TLS_RLTYPE_APP) { 2076 SOCKBUF_UNLOCK(&so->so_rcv); 2077 error = ENXIO; 2078 goto release; 2079 } 2080 } 2081 } 2082 #endif 2083 2084 do { 2085 if (flags & MSG_PEEK) { 2086 if (controlp != NULL) { 2087 *controlp = m_copym(m, 0, m->m_len, 2088 M_NOWAIT); 2089 controlp = &(*controlp)->m_next; 2090 } 2091 m = m->m_next; 2092 } else { 2093 sbfree(&so->so_rcv, m); 2094 so->so_rcv.sb_mb = m->m_next; 2095 m->m_next = NULL; 2096 *cme = m; 2097 cme = &(*cme)->m_next; 2098 m = so->so_rcv.sb_mb; 2099 } 2100 } while (m != NULL && m->m_type == MT_CONTROL); 2101 if ((flags & MSG_PEEK) == 0) 2102 sockbuf_pushsync(&so->so_rcv, nextrecord); 2103 while (cm != NULL) { 2104 cmn = cm->m_next; 2105 cm->m_next = NULL; 2106 if (pr->pr_domain->dom_externalize != NULL) { 2107 SOCKBUF_UNLOCK(&so->so_rcv); 2108 VNET_SO_ASSERT(so); 2109 error = (*pr->pr_domain->dom_externalize) 2110 (cm, controlp, flags); 2111 SOCKBUF_LOCK(&so->so_rcv); 2112 } else if (controlp != NULL) 2113 *controlp = cm; 2114 else 2115 m_freem(cm); 2116 if (controlp != NULL) { 2117 orig_resid = 0; 2118 while (*controlp != NULL) 2119 controlp = &(*controlp)->m_next; 2120 } 2121 cm = cmn; 2122 } 2123 if (m != NULL) 2124 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 2125 else 2126 nextrecord = so->so_rcv.sb_mb; 2127 orig_resid = 0; 2128 } 2129 if (m != NULL) { 2130 if ((flags & MSG_PEEK) == 0) { 2131 KASSERT(m->m_nextpkt == nextrecord, 2132 ("soreceive: post-control, nextrecord !sync")); 2133 if (nextrecord == NULL) { 2134 KASSERT(so->so_rcv.sb_mb == m, 2135 ("soreceive: post-control, sb_mb!=m")); 2136 KASSERT(so->so_rcv.sb_lastrecord == m, 2137 ("soreceive: post-control, lastrecord!=m")); 2138 } 2139 } 2140 type = m->m_type; 2141 if (type == MT_OOBDATA) 2142 flags |= MSG_OOB; 2143 } else { 2144 if ((flags & MSG_PEEK) == 0) { 2145 KASSERT(so->so_rcv.sb_mb == nextrecord, 2146 ("soreceive: sb_mb != nextrecord")); 2147 if (so->so_rcv.sb_mb == NULL) { 2148 KASSERT(so->so_rcv.sb_lastrecord == NULL, 2149 ("soreceive: sb_lastercord != NULL")); 2150 } 2151 } 2152 } 2153 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2154 SBLASTRECORDCHK(&so->so_rcv); 2155 SBLASTMBUFCHK(&so->so_rcv); 2156 2157 /* 2158 * Now continue to read any data mbufs off of the head of the socket 2159 * buffer until the read request is satisfied. Note that 'type' is 2160 * used to store the type of any mbuf reads that have happened so far 2161 * such that soreceive() can stop reading if the type changes, which 2162 * causes soreceive() to return only one of regular data and inline 2163 * out-of-band data in a single socket receive operation. 2164 */ 2165 moff = 0; 2166 offset = 0; 2167 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 2168 && error == 0) { 2169 /* 2170 * If the type of mbuf has changed since the last mbuf 2171 * examined ('type'), end the receive operation. 2172 */ 2173 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2174 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 2175 if (type != m->m_type) 2176 break; 2177 } else if (type == MT_OOBDATA) 2178 break; 2179 else 2180 KASSERT(m->m_type == MT_DATA, 2181 ("m->m_type == %d", m->m_type)); 2182 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 2183 len = uio->uio_resid; 2184 if (so->so_oobmark && len > so->so_oobmark - offset) 2185 len = so->so_oobmark - offset; 2186 if (len > m->m_len - moff) 2187 len = m->m_len - moff; 2188 /* 2189 * If mp is set, just pass back the mbufs. Otherwise copy 2190 * them out via the uio, then free. Sockbuf must be 2191 * consistent here (points to current mbuf, it points to next 2192 * record) when we drop priority; we must note any additions 2193 * to the sockbuf when we block interrupts again. 2194 */ 2195 if (mp == NULL) { 2196 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2197 SBLASTRECORDCHK(&so->so_rcv); 2198 SBLASTMBUFCHK(&so->so_rcv); 2199 SOCKBUF_UNLOCK(&so->so_rcv); 2200 if ((m->m_flags & M_EXTPG) != 0) 2201 error = m_unmappedtouio(m, moff, uio, (int)len); 2202 else 2203 error = uiomove(mtod(m, char *) + moff, 2204 (int)len, uio); 2205 SOCKBUF_LOCK(&so->so_rcv); 2206 if (error) { 2207 /* 2208 * The MT_SONAME mbuf has already been removed 2209 * from the record, so it is necessary to 2210 * remove the data mbufs, if any, to preserve 2211 * the invariant in the case of PR_ADDR that 2212 * requires MT_SONAME mbufs at the head of 2213 * each record. 2214 */ 2215 if (pr->pr_flags & PR_ATOMIC && 2216 ((flags & MSG_PEEK) == 0)) 2217 (void)sbdroprecord_locked(&so->so_rcv); 2218 SOCKBUF_UNLOCK(&so->so_rcv); 2219 goto release; 2220 } 2221 } else 2222 uio->uio_resid -= len; 2223 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2224 if (len == m->m_len - moff) { 2225 if (m->m_flags & M_EOR) 2226 flags |= MSG_EOR; 2227 if (flags & MSG_PEEK) { 2228 m = m->m_next; 2229 moff = 0; 2230 } else { 2231 nextrecord = m->m_nextpkt; 2232 sbfree(&so->so_rcv, m); 2233 if (mp != NULL) { 2234 m->m_nextpkt = NULL; 2235 *mp = m; 2236 mp = &m->m_next; 2237 so->so_rcv.sb_mb = m = m->m_next; 2238 *mp = NULL; 2239 } else { 2240 so->so_rcv.sb_mb = m_free(m); 2241 m = so->so_rcv.sb_mb; 2242 } 2243 sockbuf_pushsync(&so->so_rcv, nextrecord); 2244 SBLASTRECORDCHK(&so->so_rcv); 2245 SBLASTMBUFCHK(&so->so_rcv); 2246 } 2247 } else { 2248 if (flags & MSG_PEEK) 2249 moff += len; 2250 else { 2251 if (mp != NULL) { 2252 if (flags & MSG_DONTWAIT) { 2253 *mp = m_copym(m, 0, len, 2254 M_NOWAIT); 2255 if (*mp == NULL) { 2256 /* 2257 * m_copym() couldn't 2258 * allocate an mbuf. 2259 * Adjust uio_resid back 2260 * (it was adjusted 2261 * down by len bytes, 2262 * which we didn't end 2263 * up "copying" over). 2264 */ 2265 uio->uio_resid += len; 2266 break; 2267 } 2268 } else { 2269 SOCKBUF_UNLOCK(&so->so_rcv); 2270 *mp = m_copym(m, 0, len, 2271 M_WAITOK); 2272 SOCKBUF_LOCK(&so->so_rcv); 2273 } 2274 } 2275 sbcut_locked(&so->so_rcv, len); 2276 } 2277 } 2278 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2279 if (so->so_oobmark) { 2280 if ((flags & MSG_PEEK) == 0) { 2281 so->so_oobmark -= len; 2282 if (so->so_oobmark == 0) { 2283 so->so_rcv.sb_state |= SBS_RCVATMARK; 2284 break; 2285 } 2286 } else { 2287 offset += len; 2288 if (offset == so->so_oobmark) 2289 break; 2290 } 2291 } 2292 if (flags & MSG_EOR) 2293 break; 2294 /* 2295 * If the MSG_WAITALL flag is set (for non-atomic socket), we 2296 * must not quit until "uio->uio_resid == 0" or an error 2297 * termination. If a signal/timeout occurs, return with a 2298 * short count but without error. Keep sockbuf locked 2299 * against other readers. 2300 */ 2301 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 2302 !sosendallatonce(so) && nextrecord == NULL) { 2303 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2304 if (so->so_error || 2305 so->so_rcv.sb_state & SBS_CANTRCVMORE) 2306 break; 2307 /* 2308 * Notify the protocol that some data has been 2309 * drained before blocking. 2310 */ 2311 if (pr->pr_flags & PR_WANTRCVD) { 2312 SOCKBUF_UNLOCK(&so->so_rcv); 2313 VNET_SO_ASSERT(so); 2314 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2315 SOCKBUF_LOCK(&so->so_rcv); 2316 } 2317 SBLASTRECORDCHK(&so->so_rcv); 2318 SBLASTMBUFCHK(&so->so_rcv); 2319 /* 2320 * We could receive some data while was notifying 2321 * the protocol. Skip blocking in this case. 2322 */ 2323 if (so->so_rcv.sb_mb == NULL) { 2324 error = sbwait(&so->so_rcv); 2325 if (error) { 2326 SOCKBUF_UNLOCK(&so->so_rcv); 2327 goto release; 2328 } 2329 } 2330 m = so->so_rcv.sb_mb; 2331 if (m != NULL) 2332 nextrecord = m->m_nextpkt; 2333 } 2334 } 2335 2336 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2337 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 2338 flags |= MSG_TRUNC; 2339 if ((flags & MSG_PEEK) == 0) 2340 (void) sbdroprecord_locked(&so->so_rcv); 2341 } 2342 if ((flags & MSG_PEEK) == 0) { 2343 if (m == NULL) { 2344 /* 2345 * First part is an inline SB_EMPTY_FIXUP(). Second 2346 * part makes sure sb_lastrecord is up-to-date if 2347 * there is still data in the socket buffer. 2348 */ 2349 so->so_rcv.sb_mb = nextrecord; 2350 if (so->so_rcv.sb_mb == NULL) { 2351 so->so_rcv.sb_mbtail = NULL; 2352 so->so_rcv.sb_lastrecord = NULL; 2353 } else if (nextrecord->m_nextpkt == NULL) 2354 so->so_rcv.sb_lastrecord = nextrecord; 2355 } 2356 SBLASTRECORDCHK(&so->so_rcv); 2357 SBLASTMBUFCHK(&so->so_rcv); 2358 /* 2359 * If soreceive() is being done from the socket callback, 2360 * then don't need to generate ACK to peer to update window, 2361 * since ACK will be generated on return to TCP. 2362 */ 2363 if (!(flags & MSG_SOCALLBCK) && 2364 (pr->pr_flags & PR_WANTRCVD)) { 2365 SOCKBUF_UNLOCK(&so->so_rcv); 2366 VNET_SO_ASSERT(so); 2367 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2368 SOCKBUF_LOCK(&so->so_rcv); 2369 } 2370 } 2371 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2372 if (orig_resid == uio->uio_resid && orig_resid && 2373 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 2374 SOCKBUF_UNLOCK(&so->so_rcv); 2375 goto restart; 2376 } 2377 SOCKBUF_UNLOCK(&so->so_rcv); 2378 2379 if (flagsp != NULL) 2380 *flagsp |= flags; 2381 release: 2382 sbunlock(&so->so_rcv); 2383 return (error); 2384 } 2385 2386 /* 2387 * Optimized version of soreceive() for stream (TCP) sockets. 2388 */ 2389 int 2390 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 2391 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2392 { 2393 int len = 0, error = 0, flags, oresid; 2394 struct sockbuf *sb; 2395 struct mbuf *m, *n = NULL; 2396 2397 /* We only do stream sockets. */ 2398 if (so->so_type != SOCK_STREAM) 2399 return (EINVAL); 2400 if (psa != NULL) 2401 *psa = NULL; 2402 if (flagsp != NULL) 2403 flags = *flagsp &~ MSG_EOR; 2404 else 2405 flags = 0; 2406 if (controlp != NULL) 2407 *controlp = NULL; 2408 if (flags & MSG_OOB) 2409 return (soreceive_rcvoob(so, uio, flags)); 2410 if (mp0 != NULL) 2411 *mp0 = NULL; 2412 2413 sb = &so->so_rcv; 2414 2415 #ifdef KERN_TLS 2416 /* 2417 * KTLS store TLS records as records with a control message to 2418 * describe the framing. 2419 * 2420 * We check once here before acquiring locks to optimize the 2421 * common case. 2422 */ 2423 if (sb->sb_tls_info != NULL) 2424 return (soreceive_generic(so, psa, uio, mp0, controlp, 2425 flagsp)); 2426 #endif 2427 2428 /* Prevent other readers from entering the socket. */ 2429 error = sblock(sb, SBLOCKWAIT(flags)); 2430 if (error) 2431 return (error); 2432 SOCKBUF_LOCK(sb); 2433 2434 #ifdef KERN_TLS 2435 if (sb->sb_tls_info != NULL) { 2436 SOCKBUF_UNLOCK(sb); 2437 sbunlock(sb); 2438 return (soreceive_generic(so, psa, uio, mp0, controlp, 2439 flagsp)); 2440 } 2441 #endif 2442 2443 /* Easy one, no space to copyout anything. */ 2444 if (uio->uio_resid == 0) { 2445 error = EINVAL; 2446 goto out; 2447 } 2448 oresid = uio->uio_resid; 2449 2450 /* We will never ever get anything unless we are or were connected. */ 2451 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 2452 error = ENOTCONN; 2453 goto out; 2454 } 2455 2456 restart: 2457 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2458 2459 /* Abort if socket has reported problems. */ 2460 if (so->so_error) { 2461 if (sbavail(sb) > 0) 2462 goto deliver; 2463 if (oresid > uio->uio_resid) 2464 goto out; 2465 error = so->so_error; 2466 if (!(flags & MSG_PEEK)) 2467 so->so_error = 0; 2468 goto out; 2469 } 2470 2471 /* Door is closed. Deliver what is left, if any. */ 2472 if (sb->sb_state & SBS_CANTRCVMORE) { 2473 if (sbavail(sb) > 0) 2474 goto deliver; 2475 else 2476 goto out; 2477 } 2478 2479 /* Socket buffer is empty and we shall not block. */ 2480 if (sbavail(sb) == 0 && 2481 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2482 error = EAGAIN; 2483 goto out; 2484 } 2485 2486 /* Socket buffer got some data that we shall deliver now. */ 2487 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 2488 ((so->so_state & SS_NBIO) || 2489 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2490 sbavail(sb) >= sb->sb_lowat || 2491 sbavail(sb) >= uio->uio_resid || 2492 sbavail(sb) >= sb->sb_hiwat) ) { 2493 goto deliver; 2494 } 2495 2496 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2497 if ((flags & MSG_WAITALL) && 2498 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 2499 goto deliver; 2500 2501 /* 2502 * Wait and block until (more) data comes in. 2503 * NB: Drops the sockbuf lock during wait. 2504 */ 2505 error = sbwait(sb); 2506 if (error) 2507 goto out; 2508 goto restart; 2509 2510 deliver: 2511 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2512 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 2513 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2514 2515 /* Statistics. */ 2516 if (uio->uio_td) 2517 uio->uio_td->td_ru.ru_msgrcv++; 2518 2519 /* Fill uio until full or current end of socket buffer is reached. */ 2520 len = min(uio->uio_resid, sbavail(sb)); 2521 if (mp0 != NULL) { 2522 /* Dequeue as many mbufs as possible. */ 2523 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2524 if (*mp0 == NULL) 2525 *mp0 = sb->sb_mb; 2526 else 2527 m_cat(*mp0, sb->sb_mb); 2528 for (m = sb->sb_mb; 2529 m != NULL && m->m_len <= len; 2530 m = m->m_next) { 2531 KASSERT(!(m->m_flags & M_NOTAVAIL), 2532 ("%s: m %p not available", __func__, m)); 2533 len -= m->m_len; 2534 uio->uio_resid -= m->m_len; 2535 sbfree(sb, m); 2536 n = m; 2537 } 2538 n->m_next = NULL; 2539 sb->sb_mb = m; 2540 sb->sb_lastrecord = sb->sb_mb; 2541 if (sb->sb_mb == NULL) 2542 SB_EMPTY_FIXUP(sb); 2543 } 2544 /* Copy the remainder. */ 2545 if (len > 0) { 2546 KASSERT(sb->sb_mb != NULL, 2547 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2548 2549 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2550 if (m == NULL) 2551 len = 0; /* Don't flush data from sockbuf. */ 2552 else 2553 uio->uio_resid -= len; 2554 if (*mp0 != NULL) 2555 m_cat(*mp0, m); 2556 else 2557 *mp0 = m; 2558 if (*mp0 == NULL) { 2559 error = ENOBUFS; 2560 goto out; 2561 } 2562 } 2563 } else { 2564 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2565 SOCKBUF_UNLOCK(sb); 2566 error = m_mbuftouio(uio, sb->sb_mb, len); 2567 SOCKBUF_LOCK(sb); 2568 if (error) 2569 goto out; 2570 } 2571 SBLASTRECORDCHK(sb); 2572 SBLASTMBUFCHK(sb); 2573 2574 /* 2575 * Remove the delivered data from the socket buffer unless we 2576 * were only peeking. 2577 */ 2578 if (!(flags & MSG_PEEK)) { 2579 if (len > 0) 2580 sbdrop_locked(sb, len); 2581 2582 /* Notify protocol that we drained some data. */ 2583 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2584 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2585 !(flags & MSG_SOCALLBCK))) { 2586 SOCKBUF_UNLOCK(sb); 2587 VNET_SO_ASSERT(so); 2588 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags); 2589 SOCKBUF_LOCK(sb); 2590 } 2591 } 2592 2593 /* 2594 * For MSG_WAITALL we may have to loop again and wait for 2595 * more data to come in. 2596 */ 2597 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2598 goto restart; 2599 out: 2600 SOCKBUF_LOCK_ASSERT(sb); 2601 SBLASTRECORDCHK(sb); 2602 SBLASTMBUFCHK(sb); 2603 SOCKBUF_UNLOCK(sb); 2604 sbunlock(sb); 2605 return (error); 2606 } 2607 2608 /* 2609 * Optimized version of soreceive() for simple datagram cases from userspace. 2610 * Unlike in the stream case, we're able to drop a datagram if copyout() 2611 * fails, and because we handle datagrams atomically, we don't need to use a 2612 * sleep lock to prevent I/O interlacing. 2613 */ 2614 int 2615 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2616 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2617 { 2618 struct mbuf *m, *m2; 2619 int flags, error; 2620 ssize_t len; 2621 struct protosw *pr = so->so_proto; 2622 struct mbuf *nextrecord; 2623 2624 if (psa != NULL) 2625 *psa = NULL; 2626 if (controlp != NULL) 2627 *controlp = NULL; 2628 if (flagsp != NULL) 2629 flags = *flagsp &~ MSG_EOR; 2630 else 2631 flags = 0; 2632 2633 /* 2634 * For any complicated cases, fall back to the full 2635 * soreceive_generic(). 2636 */ 2637 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB)) 2638 return (soreceive_generic(so, psa, uio, mp0, controlp, 2639 flagsp)); 2640 2641 /* 2642 * Enforce restrictions on use. 2643 */ 2644 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2645 ("soreceive_dgram: wantrcvd")); 2646 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2647 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2648 ("soreceive_dgram: SBS_RCVATMARK")); 2649 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2650 ("soreceive_dgram: P_CONNREQUIRED")); 2651 2652 /* 2653 * Loop blocking while waiting for a datagram. 2654 */ 2655 SOCKBUF_LOCK(&so->so_rcv); 2656 while ((m = so->so_rcv.sb_mb) == NULL) { 2657 KASSERT(sbavail(&so->so_rcv) == 0, 2658 ("soreceive_dgram: sb_mb NULL but sbavail %u", 2659 sbavail(&so->so_rcv))); 2660 if (so->so_error) { 2661 error = so->so_error; 2662 so->so_error = 0; 2663 SOCKBUF_UNLOCK(&so->so_rcv); 2664 return (error); 2665 } 2666 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2667 uio->uio_resid == 0) { 2668 SOCKBUF_UNLOCK(&so->so_rcv); 2669 return (0); 2670 } 2671 if ((so->so_state & SS_NBIO) || 2672 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2673 SOCKBUF_UNLOCK(&so->so_rcv); 2674 return (EWOULDBLOCK); 2675 } 2676 SBLASTRECORDCHK(&so->so_rcv); 2677 SBLASTMBUFCHK(&so->so_rcv); 2678 error = sbwait(&so->so_rcv); 2679 if (error) { 2680 SOCKBUF_UNLOCK(&so->so_rcv); 2681 return (error); 2682 } 2683 } 2684 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2685 2686 if (uio->uio_td) 2687 uio->uio_td->td_ru.ru_msgrcv++; 2688 SBLASTRECORDCHK(&so->so_rcv); 2689 SBLASTMBUFCHK(&so->so_rcv); 2690 nextrecord = m->m_nextpkt; 2691 if (nextrecord == NULL) { 2692 KASSERT(so->so_rcv.sb_lastrecord == m, 2693 ("soreceive_dgram: lastrecord != m")); 2694 } 2695 2696 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2697 ("soreceive_dgram: m_nextpkt != nextrecord")); 2698 2699 /* 2700 * Pull 'm' and its chain off the front of the packet queue. 2701 */ 2702 so->so_rcv.sb_mb = NULL; 2703 sockbuf_pushsync(&so->so_rcv, nextrecord); 2704 2705 /* 2706 * Walk 'm's chain and free that many bytes from the socket buffer. 2707 */ 2708 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2709 sbfree(&so->so_rcv, m2); 2710 2711 /* 2712 * Do a few last checks before we let go of the lock. 2713 */ 2714 SBLASTRECORDCHK(&so->so_rcv); 2715 SBLASTMBUFCHK(&so->so_rcv); 2716 SOCKBUF_UNLOCK(&so->so_rcv); 2717 2718 if (pr->pr_flags & PR_ADDR) { 2719 KASSERT(m->m_type == MT_SONAME, 2720 ("m->m_type == %d", m->m_type)); 2721 if (psa != NULL) 2722 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2723 M_NOWAIT); 2724 m = m_free(m); 2725 } 2726 if (m == NULL) { 2727 /* XXXRW: Can this happen? */ 2728 return (0); 2729 } 2730 2731 /* 2732 * Packet to copyout() is now in 'm' and it is disconnected from the 2733 * queue. 2734 * 2735 * Process one or more MT_CONTROL mbufs present before any data mbufs 2736 * in the first mbuf chain on the socket buffer. We call into the 2737 * protocol to perform externalization (or freeing if controlp == 2738 * NULL). In some cases there can be only MT_CONTROL mbufs without 2739 * MT_DATA mbufs. 2740 */ 2741 if (m->m_type == MT_CONTROL) { 2742 struct mbuf *cm = NULL, *cmn; 2743 struct mbuf **cme = &cm; 2744 2745 do { 2746 m2 = m->m_next; 2747 m->m_next = NULL; 2748 *cme = m; 2749 cme = &(*cme)->m_next; 2750 m = m2; 2751 } while (m != NULL && m->m_type == MT_CONTROL); 2752 while (cm != NULL) { 2753 cmn = cm->m_next; 2754 cm->m_next = NULL; 2755 if (pr->pr_domain->dom_externalize != NULL) { 2756 error = (*pr->pr_domain->dom_externalize) 2757 (cm, controlp, flags); 2758 } else if (controlp != NULL) 2759 *controlp = cm; 2760 else 2761 m_freem(cm); 2762 if (controlp != NULL) { 2763 while (*controlp != NULL) 2764 controlp = &(*controlp)->m_next; 2765 } 2766 cm = cmn; 2767 } 2768 } 2769 KASSERT(m == NULL || m->m_type == MT_DATA, 2770 ("soreceive_dgram: !data")); 2771 while (m != NULL && uio->uio_resid > 0) { 2772 len = uio->uio_resid; 2773 if (len > m->m_len) 2774 len = m->m_len; 2775 error = uiomove(mtod(m, char *), (int)len, uio); 2776 if (error) { 2777 m_freem(m); 2778 return (error); 2779 } 2780 if (len == m->m_len) 2781 m = m_free(m); 2782 else { 2783 m->m_data += len; 2784 m->m_len -= len; 2785 } 2786 } 2787 if (m != NULL) { 2788 flags |= MSG_TRUNC; 2789 m_freem(m); 2790 } 2791 if (flagsp != NULL) 2792 *flagsp |= flags; 2793 return (0); 2794 } 2795 2796 int 2797 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2798 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2799 { 2800 int error; 2801 2802 CURVNET_SET(so->so_vnet); 2803 if (!SOLISTENING(so)) 2804 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, 2805 mp0, controlp, flagsp)); 2806 else 2807 error = ENOTCONN; 2808 CURVNET_RESTORE(); 2809 return (error); 2810 } 2811 2812 int 2813 soshutdown(struct socket *so, int how) 2814 { 2815 struct protosw *pr = so->so_proto; 2816 int error, soerror_enotconn; 2817 2818 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2819 return (EINVAL); 2820 2821 soerror_enotconn = 0; 2822 if ((so->so_state & 2823 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 2824 /* 2825 * POSIX mandates us to return ENOTCONN when shutdown(2) is 2826 * invoked on a datagram sockets, however historically we would 2827 * actually tear socket down. This is known to be leveraged by 2828 * some applications to unblock process waiting in recvXXX(2) 2829 * by other process that it shares that socket with. Try to meet 2830 * both backward-compatibility and POSIX requirements by forcing 2831 * ENOTCONN but still asking protocol to perform pru_shutdown(). 2832 */ 2833 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) 2834 return (ENOTCONN); 2835 soerror_enotconn = 1; 2836 } 2837 2838 if (SOLISTENING(so)) { 2839 if (how != SHUT_WR) { 2840 SOLISTEN_LOCK(so); 2841 so->so_error = ECONNABORTED; 2842 solisten_wakeup(so); /* unlocks so */ 2843 } 2844 goto done; 2845 } 2846 2847 CURVNET_SET(so->so_vnet); 2848 if (pr->pr_usrreqs->pru_flush != NULL) 2849 (*pr->pr_usrreqs->pru_flush)(so, how); 2850 if (how != SHUT_WR) 2851 sorflush(so); 2852 if (how != SHUT_RD) { 2853 error = (*pr->pr_usrreqs->pru_shutdown)(so); 2854 wakeup(&so->so_timeo); 2855 CURVNET_RESTORE(); 2856 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error); 2857 } 2858 wakeup(&so->so_timeo); 2859 CURVNET_RESTORE(); 2860 2861 done: 2862 return (soerror_enotconn ? ENOTCONN : 0); 2863 } 2864 2865 void 2866 sorflush(struct socket *so) 2867 { 2868 struct sockbuf *sb = &so->so_rcv; 2869 struct protosw *pr = so->so_proto; 2870 struct socket aso; 2871 2872 VNET_SO_ASSERT(so); 2873 2874 /* 2875 * In order to avoid calling dom_dispose with the socket buffer mutex 2876 * held, and in order to generally avoid holding the lock for a long 2877 * time, we make a copy of the socket buffer and clear the original 2878 * (except locks, state). The new socket buffer copy won't have 2879 * initialized locks so we can only call routines that won't use or 2880 * assert those locks. 2881 * 2882 * Dislodge threads currently blocked in receive and wait to acquire 2883 * a lock against other simultaneous readers before clearing the 2884 * socket buffer. Don't let our acquire be interrupted by a signal 2885 * despite any existing socket disposition on interruptable waiting. 2886 */ 2887 socantrcvmore(so); 2888 (void) sblock(sb, SBL_WAIT | SBL_NOINTR); 2889 2890 /* 2891 * Invalidate/clear most of the sockbuf structure, but leave selinfo 2892 * and mutex data unchanged. 2893 */ 2894 SOCKBUF_LOCK(sb); 2895 bzero(&aso, sizeof(aso)); 2896 aso.so_pcb = so->so_pcb; 2897 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero, 2898 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2899 bzero(&sb->sb_startzero, 2900 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2901 SOCKBUF_UNLOCK(sb); 2902 sbunlock(sb); 2903 2904 /* 2905 * Dispose of special rights and flush the copied socket. Don't call 2906 * any unsafe routines (that rely on locks being initialized) on aso. 2907 */ 2908 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 2909 (*pr->pr_domain->dom_dispose)(&aso); 2910 sbrelease_internal(&aso.so_rcv, so); 2911 } 2912 2913 /* 2914 * Wrapper for Socket established helper hook. 2915 * Parameters: socket, context of the hook point, hook id. 2916 */ 2917 static int inline 2918 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 2919 { 2920 struct socket_hhook_data hhook_data = { 2921 .so = so, 2922 .hctx = hctx, 2923 .m = NULL, 2924 .status = 0 2925 }; 2926 2927 CURVNET_SET(so->so_vnet); 2928 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 2929 CURVNET_RESTORE(); 2930 2931 /* Ugly but needed, since hhooks return void for now */ 2932 return (hhook_data.status); 2933 } 2934 2935 /* 2936 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 2937 * additional variant to handle the case where the option value needs to be 2938 * some kind of integer, but not a specific size. In addition to their use 2939 * here, these functions are also called by the protocol-level pr_ctloutput() 2940 * routines. 2941 */ 2942 int 2943 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2944 { 2945 size_t valsize; 2946 2947 /* 2948 * If the user gives us more than we wanted, we ignore it, but if we 2949 * don't get the minimum length the caller wants, we return EINVAL. 2950 * On success, sopt->sopt_valsize is set to however much we actually 2951 * retrieved. 2952 */ 2953 if ((valsize = sopt->sopt_valsize) < minlen) 2954 return EINVAL; 2955 if (valsize > len) 2956 sopt->sopt_valsize = valsize = len; 2957 2958 if (sopt->sopt_td != NULL) 2959 return (copyin(sopt->sopt_val, buf, valsize)); 2960 2961 bcopy(sopt->sopt_val, buf, valsize); 2962 return (0); 2963 } 2964 2965 /* 2966 * Kernel version of setsockopt(2). 2967 * 2968 * XXX: optlen is size_t, not socklen_t 2969 */ 2970 int 2971 so_setsockopt(struct socket *so, int level, int optname, void *optval, 2972 size_t optlen) 2973 { 2974 struct sockopt sopt; 2975 2976 sopt.sopt_level = level; 2977 sopt.sopt_name = optname; 2978 sopt.sopt_dir = SOPT_SET; 2979 sopt.sopt_val = optval; 2980 sopt.sopt_valsize = optlen; 2981 sopt.sopt_td = NULL; 2982 return (sosetopt(so, &sopt)); 2983 } 2984 2985 int 2986 sosetopt(struct socket *so, struct sockopt *sopt) 2987 { 2988 int error, optval; 2989 struct linger l; 2990 struct timeval tv; 2991 sbintime_t val; 2992 uint32_t val32; 2993 #ifdef MAC 2994 struct mac extmac; 2995 #endif 2996 2997 CURVNET_SET(so->so_vnet); 2998 error = 0; 2999 if (sopt->sopt_level != SOL_SOCKET) { 3000 if (so->so_proto->pr_ctloutput != NULL) 3001 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3002 else 3003 error = ENOPROTOOPT; 3004 } else { 3005 switch (sopt->sopt_name) { 3006 case SO_ACCEPTFILTER: 3007 error = accept_filt_setopt(so, sopt); 3008 if (error) 3009 goto bad; 3010 break; 3011 3012 case SO_LINGER: 3013 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 3014 if (error) 3015 goto bad; 3016 if (l.l_linger < 0 || 3017 l.l_linger > USHRT_MAX || 3018 l.l_linger > (INT_MAX / hz)) { 3019 error = EDOM; 3020 goto bad; 3021 } 3022 SOCK_LOCK(so); 3023 so->so_linger = l.l_linger; 3024 if (l.l_onoff) 3025 so->so_options |= SO_LINGER; 3026 else 3027 so->so_options &= ~SO_LINGER; 3028 SOCK_UNLOCK(so); 3029 break; 3030 3031 case SO_DEBUG: 3032 case SO_KEEPALIVE: 3033 case SO_DONTROUTE: 3034 case SO_USELOOPBACK: 3035 case SO_BROADCAST: 3036 case SO_REUSEADDR: 3037 case SO_REUSEPORT: 3038 case SO_REUSEPORT_LB: 3039 case SO_OOBINLINE: 3040 case SO_TIMESTAMP: 3041 case SO_BINTIME: 3042 case SO_NOSIGPIPE: 3043 case SO_NO_DDP: 3044 case SO_NO_OFFLOAD: 3045 error = sooptcopyin(sopt, &optval, sizeof optval, 3046 sizeof optval); 3047 if (error) 3048 goto bad; 3049 SOCK_LOCK(so); 3050 if (optval) 3051 so->so_options |= sopt->sopt_name; 3052 else 3053 so->so_options &= ~sopt->sopt_name; 3054 SOCK_UNLOCK(so); 3055 break; 3056 3057 case SO_SETFIB: 3058 error = sooptcopyin(sopt, &optval, sizeof optval, 3059 sizeof optval); 3060 if (error) 3061 goto bad; 3062 3063 if (optval < 0 || optval >= rt_numfibs) { 3064 error = EINVAL; 3065 goto bad; 3066 } 3067 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 3068 (so->so_proto->pr_domain->dom_family == PF_INET6) || 3069 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 3070 so->so_fibnum = optval; 3071 else 3072 so->so_fibnum = 0; 3073 break; 3074 3075 case SO_USER_COOKIE: 3076 error = sooptcopyin(sopt, &val32, sizeof val32, 3077 sizeof val32); 3078 if (error) 3079 goto bad; 3080 so->so_user_cookie = val32; 3081 break; 3082 3083 case SO_SNDBUF: 3084 case SO_RCVBUF: 3085 case SO_SNDLOWAT: 3086 case SO_RCVLOWAT: 3087 error = sooptcopyin(sopt, &optval, sizeof optval, 3088 sizeof optval); 3089 if (error) 3090 goto bad; 3091 3092 /* 3093 * Values < 1 make no sense for any of these options, 3094 * so disallow them. 3095 */ 3096 if (optval < 1) { 3097 error = EINVAL; 3098 goto bad; 3099 } 3100 3101 error = sbsetopt(so, sopt->sopt_name, optval); 3102 break; 3103 3104 case SO_SNDTIMEO: 3105 case SO_RCVTIMEO: 3106 #ifdef COMPAT_FREEBSD32 3107 if (SV_CURPROC_FLAG(SV_ILP32)) { 3108 struct timeval32 tv32; 3109 3110 error = sooptcopyin(sopt, &tv32, sizeof tv32, 3111 sizeof tv32); 3112 CP(tv32, tv, tv_sec); 3113 CP(tv32, tv, tv_usec); 3114 } else 3115 #endif 3116 error = sooptcopyin(sopt, &tv, sizeof tv, 3117 sizeof tv); 3118 if (error) 3119 goto bad; 3120 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 3121 tv.tv_usec >= 1000000) { 3122 error = EDOM; 3123 goto bad; 3124 } 3125 if (tv.tv_sec > INT32_MAX) 3126 val = SBT_MAX; 3127 else 3128 val = tvtosbt(tv); 3129 switch (sopt->sopt_name) { 3130 case SO_SNDTIMEO: 3131 so->so_snd.sb_timeo = val; 3132 break; 3133 case SO_RCVTIMEO: 3134 so->so_rcv.sb_timeo = val; 3135 break; 3136 } 3137 break; 3138 3139 case SO_LABEL: 3140 #ifdef MAC 3141 error = sooptcopyin(sopt, &extmac, sizeof extmac, 3142 sizeof extmac); 3143 if (error) 3144 goto bad; 3145 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 3146 so, &extmac); 3147 #else 3148 error = EOPNOTSUPP; 3149 #endif 3150 break; 3151 3152 case SO_TS_CLOCK: 3153 error = sooptcopyin(sopt, &optval, sizeof optval, 3154 sizeof optval); 3155 if (error) 3156 goto bad; 3157 if (optval < 0 || optval > SO_TS_CLOCK_MAX) { 3158 error = EINVAL; 3159 goto bad; 3160 } 3161 so->so_ts_clock = optval; 3162 break; 3163 3164 case SO_MAX_PACING_RATE: 3165 error = sooptcopyin(sopt, &val32, sizeof(val32), 3166 sizeof(val32)); 3167 if (error) 3168 goto bad; 3169 so->so_max_pacing_rate = val32; 3170 break; 3171 3172 default: 3173 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3174 error = hhook_run_socket(so, sopt, 3175 HHOOK_SOCKET_OPT); 3176 else 3177 error = ENOPROTOOPT; 3178 break; 3179 } 3180 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 3181 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 3182 } 3183 bad: 3184 CURVNET_RESTORE(); 3185 return (error); 3186 } 3187 3188 /* 3189 * Helper routine for getsockopt. 3190 */ 3191 int 3192 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 3193 { 3194 int error; 3195 size_t valsize; 3196 3197 error = 0; 3198 3199 /* 3200 * Documented get behavior is that we always return a value, possibly 3201 * truncated to fit in the user's buffer. Traditional behavior is 3202 * that we always tell the user precisely how much we copied, rather 3203 * than something useful like the total amount we had available for 3204 * her. Note that this interface is not idempotent; the entire 3205 * answer must be generated ahead of time. 3206 */ 3207 valsize = min(len, sopt->sopt_valsize); 3208 sopt->sopt_valsize = valsize; 3209 if (sopt->sopt_val != NULL) { 3210 if (sopt->sopt_td != NULL) 3211 error = copyout(buf, sopt->sopt_val, valsize); 3212 else 3213 bcopy(buf, sopt->sopt_val, valsize); 3214 } 3215 return (error); 3216 } 3217 3218 int 3219 sogetopt(struct socket *so, struct sockopt *sopt) 3220 { 3221 int error, optval; 3222 struct linger l; 3223 struct timeval tv; 3224 #ifdef MAC 3225 struct mac extmac; 3226 #endif 3227 3228 CURVNET_SET(so->so_vnet); 3229 error = 0; 3230 if (sopt->sopt_level != SOL_SOCKET) { 3231 if (so->so_proto->pr_ctloutput != NULL) 3232 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3233 else 3234 error = ENOPROTOOPT; 3235 CURVNET_RESTORE(); 3236 return (error); 3237 } else { 3238 switch (sopt->sopt_name) { 3239 case SO_ACCEPTFILTER: 3240 error = accept_filt_getopt(so, sopt); 3241 break; 3242 3243 case SO_LINGER: 3244 SOCK_LOCK(so); 3245 l.l_onoff = so->so_options & SO_LINGER; 3246 l.l_linger = so->so_linger; 3247 SOCK_UNLOCK(so); 3248 error = sooptcopyout(sopt, &l, sizeof l); 3249 break; 3250 3251 case SO_USELOOPBACK: 3252 case SO_DONTROUTE: 3253 case SO_DEBUG: 3254 case SO_KEEPALIVE: 3255 case SO_REUSEADDR: 3256 case SO_REUSEPORT: 3257 case SO_REUSEPORT_LB: 3258 case SO_BROADCAST: 3259 case SO_OOBINLINE: 3260 case SO_ACCEPTCONN: 3261 case SO_TIMESTAMP: 3262 case SO_BINTIME: 3263 case SO_NOSIGPIPE: 3264 case SO_NO_DDP: 3265 case SO_NO_OFFLOAD: 3266 optval = so->so_options & sopt->sopt_name; 3267 integer: 3268 error = sooptcopyout(sopt, &optval, sizeof optval); 3269 break; 3270 3271 case SO_DOMAIN: 3272 optval = so->so_proto->pr_domain->dom_family; 3273 goto integer; 3274 3275 case SO_TYPE: 3276 optval = so->so_type; 3277 goto integer; 3278 3279 case SO_PROTOCOL: 3280 optval = so->so_proto->pr_protocol; 3281 goto integer; 3282 3283 case SO_ERROR: 3284 SOCK_LOCK(so); 3285 optval = so->so_error; 3286 so->so_error = 0; 3287 SOCK_UNLOCK(so); 3288 goto integer; 3289 3290 case SO_SNDBUF: 3291 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : 3292 so->so_snd.sb_hiwat; 3293 goto integer; 3294 3295 case SO_RCVBUF: 3296 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : 3297 so->so_rcv.sb_hiwat; 3298 goto integer; 3299 3300 case SO_SNDLOWAT: 3301 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : 3302 so->so_snd.sb_lowat; 3303 goto integer; 3304 3305 case SO_RCVLOWAT: 3306 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : 3307 so->so_rcv.sb_lowat; 3308 goto integer; 3309 3310 case SO_SNDTIMEO: 3311 case SO_RCVTIMEO: 3312 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 3313 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 3314 #ifdef COMPAT_FREEBSD32 3315 if (SV_CURPROC_FLAG(SV_ILP32)) { 3316 struct timeval32 tv32; 3317 3318 CP(tv, tv32, tv_sec); 3319 CP(tv, tv32, tv_usec); 3320 error = sooptcopyout(sopt, &tv32, sizeof tv32); 3321 } else 3322 #endif 3323 error = sooptcopyout(sopt, &tv, sizeof tv); 3324 break; 3325 3326 case SO_LABEL: 3327 #ifdef MAC 3328 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3329 sizeof(extmac)); 3330 if (error) 3331 goto bad; 3332 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 3333 so, &extmac); 3334 if (error) 3335 goto bad; 3336 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3337 #else 3338 error = EOPNOTSUPP; 3339 #endif 3340 break; 3341 3342 case SO_PEERLABEL: 3343 #ifdef MAC 3344 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3345 sizeof(extmac)); 3346 if (error) 3347 goto bad; 3348 error = mac_getsockopt_peerlabel( 3349 sopt->sopt_td->td_ucred, so, &extmac); 3350 if (error) 3351 goto bad; 3352 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3353 #else 3354 error = EOPNOTSUPP; 3355 #endif 3356 break; 3357 3358 case SO_LISTENQLIMIT: 3359 optval = SOLISTENING(so) ? so->sol_qlimit : 0; 3360 goto integer; 3361 3362 case SO_LISTENQLEN: 3363 optval = SOLISTENING(so) ? so->sol_qlen : 0; 3364 goto integer; 3365 3366 case SO_LISTENINCQLEN: 3367 optval = SOLISTENING(so) ? so->sol_incqlen : 0; 3368 goto integer; 3369 3370 case SO_TS_CLOCK: 3371 optval = so->so_ts_clock; 3372 goto integer; 3373 3374 case SO_MAX_PACING_RATE: 3375 optval = so->so_max_pacing_rate; 3376 goto integer; 3377 3378 default: 3379 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3380 error = hhook_run_socket(so, sopt, 3381 HHOOK_SOCKET_OPT); 3382 else 3383 error = ENOPROTOOPT; 3384 break; 3385 } 3386 } 3387 #ifdef MAC 3388 bad: 3389 #endif 3390 CURVNET_RESTORE(); 3391 return (error); 3392 } 3393 3394 int 3395 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 3396 { 3397 struct mbuf *m, *m_prev; 3398 int sopt_size = sopt->sopt_valsize; 3399 3400 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3401 if (m == NULL) 3402 return ENOBUFS; 3403 if (sopt_size > MLEN) { 3404 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 3405 if ((m->m_flags & M_EXT) == 0) { 3406 m_free(m); 3407 return ENOBUFS; 3408 } 3409 m->m_len = min(MCLBYTES, sopt_size); 3410 } else { 3411 m->m_len = min(MLEN, sopt_size); 3412 } 3413 sopt_size -= m->m_len; 3414 *mp = m; 3415 m_prev = m; 3416 3417 while (sopt_size) { 3418 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3419 if (m == NULL) { 3420 m_freem(*mp); 3421 return ENOBUFS; 3422 } 3423 if (sopt_size > MLEN) { 3424 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 3425 M_NOWAIT); 3426 if ((m->m_flags & M_EXT) == 0) { 3427 m_freem(m); 3428 m_freem(*mp); 3429 return ENOBUFS; 3430 } 3431 m->m_len = min(MCLBYTES, sopt_size); 3432 } else { 3433 m->m_len = min(MLEN, sopt_size); 3434 } 3435 sopt_size -= m->m_len; 3436 m_prev->m_next = m; 3437 m_prev = m; 3438 } 3439 return (0); 3440 } 3441 3442 int 3443 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 3444 { 3445 struct mbuf *m0 = m; 3446 3447 if (sopt->sopt_val == NULL) 3448 return (0); 3449 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3450 if (sopt->sopt_td != NULL) { 3451 int error; 3452 3453 error = copyin(sopt->sopt_val, mtod(m, char *), 3454 m->m_len); 3455 if (error != 0) { 3456 m_freem(m0); 3457 return(error); 3458 } 3459 } else 3460 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 3461 sopt->sopt_valsize -= m->m_len; 3462 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3463 m = m->m_next; 3464 } 3465 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 3466 panic("ip6_sooptmcopyin"); 3467 return (0); 3468 } 3469 3470 int 3471 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 3472 { 3473 struct mbuf *m0 = m; 3474 size_t valsize = 0; 3475 3476 if (sopt->sopt_val == NULL) 3477 return (0); 3478 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3479 if (sopt->sopt_td != NULL) { 3480 int error; 3481 3482 error = copyout(mtod(m, char *), sopt->sopt_val, 3483 m->m_len); 3484 if (error != 0) { 3485 m_freem(m0); 3486 return(error); 3487 } 3488 } else 3489 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 3490 sopt->sopt_valsize -= m->m_len; 3491 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3492 valsize += m->m_len; 3493 m = m->m_next; 3494 } 3495 if (m != NULL) { 3496 /* enough soopt buffer should be given from user-land */ 3497 m_freem(m0); 3498 return(EINVAL); 3499 } 3500 sopt->sopt_valsize = valsize; 3501 return (0); 3502 } 3503 3504 /* 3505 * sohasoutofband(): protocol notifies socket layer of the arrival of new 3506 * out-of-band data, which will then notify socket consumers. 3507 */ 3508 void 3509 sohasoutofband(struct socket *so) 3510 { 3511 3512 if (so->so_sigio != NULL) 3513 pgsigio(&so->so_sigio, SIGURG, 0); 3514 selwakeuppri(&so->so_rdsel, PSOCK); 3515 } 3516 3517 int 3518 sopoll(struct socket *so, int events, struct ucred *active_cred, 3519 struct thread *td) 3520 { 3521 3522 /* 3523 * We do not need to set or assert curvnet as long as everyone uses 3524 * sopoll_generic(). 3525 */ 3526 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, 3527 td)); 3528 } 3529 3530 int 3531 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3532 struct thread *td) 3533 { 3534 int revents; 3535 3536 SOCK_LOCK(so); 3537 if (SOLISTENING(so)) { 3538 if (!(events & (POLLIN | POLLRDNORM))) 3539 revents = 0; 3540 else if (!TAILQ_EMPTY(&so->sol_comp)) 3541 revents = events & (POLLIN | POLLRDNORM); 3542 else if ((events & POLLINIGNEOF) == 0 && so->so_error) 3543 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP; 3544 else { 3545 selrecord(td, &so->so_rdsel); 3546 revents = 0; 3547 } 3548 } else { 3549 revents = 0; 3550 SOCKBUF_LOCK(&so->so_snd); 3551 SOCKBUF_LOCK(&so->so_rcv); 3552 if (events & (POLLIN | POLLRDNORM)) 3553 if (soreadabledata(so)) 3554 revents |= events & (POLLIN | POLLRDNORM); 3555 if (events & (POLLOUT | POLLWRNORM)) 3556 if (sowriteable(so)) 3557 revents |= events & (POLLOUT | POLLWRNORM); 3558 if (events & (POLLPRI | POLLRDBAND)) 3559 if (so->so_oobmark || 3560 (so->so_rcv.sb_state & SBS_RCVATMARK)) 3561 revents |= events & (POLLPRI | POLLRDBAND); 3562 if ((events & POLLINIGNEOF) == 0) { 3563 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3564 revents |= events & (POLLIN | POLLRDNORM); 3565 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3566 revents |= POLLHUP; 3567 } 3568 } 3569 if (revents == 0) { 3570 if (events & 3571 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 3572 selrecord(td, &so->so_rdsel); 3573 so->so_rcv.sb_flags |= SB_SEL; 3574 } 3575 if (events & (POLLOUT | POLLWRNORM)) { 3576 selrecord(td, &so->so_wrsel); 3577 so->so_snd.sb_flags |= SB_SEL; 3578 } 3579 } 3580 SOCKBUF_UNLOCK(&so->so_rcv); 3581 SOCKBUF_UNLOCK(&so->so_snd); 3582 } 3583 SOCK_UNLOCK(so); 3584 return (revents); 3585 } 3586 3587 int 3588 soo_kqfilter(struct file *fp, struct knote *kn) 3589 { 3590 struct socket *so = kn->kn_fp->f_data; 3591 struct sockbuf *sb; 3592 struct knlist *knl; 3593 3594 switch (kn->kn_filter) { 3595 case EVFILT_READ: 3596 kn->kn_fop = &soread_filtops; 3597 knl = &so->so_rdsel.si_note; 3598 sb = &so->so_rcv; 3599 break; 3600 case EVFILT_WRITE: 3601 kn->kn_fop = &sowrite_filtops; 3602 knl = &so->so_wrsel.si_note; 3603 sb = &so->so_snd; 3604 break; 3605 case EVFILT_EMPTY: 3606 kn->kn_fop = &soempty_filtops; 3607 knl = &so->so_wrsel.si_note; 3608 sb = &so->so_snd; 3609 break; 3610 default: 3611 return (EINVAL); 3612 } 3613 3614 SOCK_LOCK(so); 3615 if (SOLISTENING(so)) { 3616 knlist_add(knl, kn, 1); 3617 } else { 3618 SOCKBUF_LOCK(sb); 3619 knlist_add(knl, kn, 1); 3620 sb->sb_flags |= SB_KNOTE; 3621 SOCKBUF_UNLOCK(sb); 3622 } 3623 SOCK_UNLOCK(so); 3624 return (0); 3625 } 3626 3627 /* 3628 * Some routines that return EOPNOTSUPP for entry points that are not 3629 * supported by a protocol. Fill in as needed. 3630 */ 3631 int 3632 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 3633 { 3634 3635 return EOPNOTSUPP; 3636 } 3637 3638 int 3639 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job) 3640 { 3641 3642 return EOPNOTSUPP; 3643 } 3644 3645 int 3646 pru_attach_notsupp(struct socket *so, int proto, struct thread *td) 3647 { 3648 3649 return EOPNOTSUPP; 3650 } 3651 3652 int 3653 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3654 { 3655 3656 return EOPNOTSUPP; 3657 } 3658 3659 int 3660 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3661 struct thread *td) 3662 { 3663 3664 return EOPNOTSUPP; 3665 } 3666 3667 int 3668 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3669 { 3670 3671 return EOPNOTSUPP; 3672 } 3673 3674 int 3675 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3676 struct thread *td) 3677 { 3678 3679 return EOPNOTSUPP; 3680 } 3681 3682 int 3683 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 3684 { 3685 3686 return EOPNOTSUPP; 3687 } 3688 3689 int 3690 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 3691 struct ifnet *ifp, struct thread *td) 3692 { 3693 3694 return EOPNOTSUPP; 3695 } 3696 3697 int 3698 pru_disconnect_notsupp(struct socket *so) 3699 { 3700 3701 return EOPNOTSUPP; 3702 } 3703 3704 int 3705 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) 3706 { 3707 3708 return EOPNOTSUPP; 3709 } 3710 3711 int 3712 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) 3713 { 3714 3715 return EOPNOTSUPP; 3716 } 3717 3718 int 3719 pru_rcvd_notsupp(struct socket *so, int flags) 3720 { 3721 3722 return EOPNOTSUPP; 3723 } 3724 3725 int 3726 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 3727 { 3728 3729 return EOPNOTSUPP; 3730 } 3731 3732 int 3733 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, 3734 struct sockaddr *addr, struct mbuf *control, struct thread *td) 3735 { 3736 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