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