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 35 /* 36 * Comments on the socket life cycle: 37 * 38 * soalloc() sets of socket layer state for a socket, called only by 39 * socreate() and sonewconn(). Socket layer private. 40 * 41 * sodealloc() tears down socket layer state for a socket, called only by 42 * sofree() and sonewconn(). Socket layer private. 43 * 44 * pru_attach() associates protocol layer state with an allocated socket; 45 * called only once, may fail, aborting socket allocation. This is called 46 * from socreate() and sonewconn(). Socket layer private. 47 * 48 * pru_detach() disassociates protocol layer state from an attached socket, 49 * and will be called exactly once for sockets in which pru_attach() has 50 * been successfully called. If pru_attach() returned an error, 51 * pru_detach() will not be called. Socket layer private. 52 * 53 * pru_abort() and pru_close() notify the protocol layer that the last 54 * consumer of a socket is starting to tear down the socket, and that the 55 * protocol should terminate the connection. Historically, pru_abort() also 56 * detached protocol state from the socket state, but this is no longer the 57 * case. 58 * 59 * socreate() creates a socket and attaches protocol state. This is a public 60 * interface that may be used by socket layer consumers to create new 61 * sockets. 62 * 63 * sonewconn() creates a socket and attaches protocol state. This is a 64 * public interface that may be used by protocols to create new sockets when 65 * a new connection is received and will be available for accept() on a 66 * listen socket. 67 * 68 * soclose() destroys a socket after possibly waiting for it to disconnect. 69 * This is a public interface that socket consumers should use to close and 70 * release a socket when done with it. 71 * 72 * soabort() destroys a socket without waiting for it to disconnect (used 73 * only for incoming connections that are already partially or fully 74 * connected). This is used internally by the socket layer when clearing 75 * listen socket queues (due to overflow or close on the listen socket), but 76 * is also a public interface protocols may use to abort connections in 77 * their incomplete listen queues should they no longer be required. Sockets 78 * placed in completed connection listen queues should not be aborted for 79 * reasons described in the comment above the soclose() implementation. This 80 * is not a general purpose close routine, and except in the specific 81 * circumstances described here, should not be used. 82 * 83 * sofree() will free a socket and its protocol state if all references on 84 * the socket have been released, and is the public interface to attempt to 85 * free a socket when a reference is removed. This is a socket layer private 86 * interface. 87 * 88 * NOTE: In addition to socreate() and soclose(), which provide a single 89 * socket reference to the consumer to be managed as required, there are two 90 * calls to explicitly manage socket references, soref(), and sorele(). 91 * Currently, these are generally required only when transitioning a socket 92 * from a listen queue to a file descriptor, in order to prevent garbage 93 * collection of the socket at an untimely moment. For a number of reasons, 94 * these interfaces are not preferred, and should be avoided. 95 * 96 * NOTE: With regard to VNETs the general rule is that callers do not set 97 * curvnet. Exceptions to this rule include soabort(), sodisconnect(), 98 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn() 99 * and sorflush(), which are usually called from a pre-set VNET context. 100 * sopoll() currently does not need a VNET context to be set. 101 */ 102 103 #include <sys/cdefs.h> 104 #include "opt_inet.h" 105 #include "opt_inet6.h" 106 #include "opt_kern_tls.h" 107 #include "opt_sctp.h" 108 109 #include <sys/param.h> 110 #include <sys/systm.h> 111 #include <sys/capsicum.h> 112 #include <sys/fcntl.h> 113 #include <sys/limits.h> 114 #include <sys/lock.h> 115 #include <sys/mac.h> 116 #include <sys/malloc.h> 117 #include <sys/mbuf.h> 118 #include <sys/mutex.h> 119 #include <sys/domain.h> 120 #include <sys/file.h> /* for struct knote */ 121 #include <sys/hhook.h> 122 #include <sys/kernel.h> 123 #include <sys/khelp.h> 124 #include <sys/ktls.h> 125 #include <sys/event.h> 126 #include <sys/eventhandler.h> 127 #include <sys/poll.h> 128 #include <sys/proc.h> 129 #include <sys/protosw.h> 130 #include <sys/sbuf.h> 131 #include <sys/socket.h> 132 #include <sys/socketvar.h> 133 #include <sys/resourcevar.h> 134 #include <net/route.h> 135 #include <sys/signalvar.h> 136 #include <sys/stat.h> 137 #include <sys/sx.h> 138 #include <sys/sysctl.h> 139 #include <sys/taskqueue.h> 140 #include <sys/uio.h> 141 #include <sys/un.h> 142 #include <sys/unpcb.h> 143 #include <sys/jail.h> 144 #include <sys/syslog.h> 145 #include <netinet/in.h> 146 #include <netinet/in_pcb.h> 147 #include <netinet/tcp.h> 148 149 #include <net/vnet.h> 150 151 #include <security/mac/mac_framework.h> 152 153 #include <vm/uma.h> 154 155 #ifdef COMPAT_FREEBSD32 156 #include <sys/mount.h> 157 #include <sys/sysent.h> 158 #include <compat/freebsd32/freebsd32.h> 159 #endif 160 161 static int soreceive_rcvoob(struct socket *so, struct uio *uio, 162 int flags); 163 static void so_rdknl_lock(void *); 164 static void so_rdknl_unlock(void *); 165 static void so_rdknl_assert_lock(void *, int); 166 static void so_wrknl_lock(void *); 167 static void so_wrknl_unlock(void *); 168 static void so_wrknl_assert_lock(void *, int); 169 170 static void filt_sordetach(struct knote *kn); 171 static int filt_soread(struct knote *kn, long hint); 172 static void filt_sowdetach(struct knote *kn); 173 static int filt_sowrite(struct knote *kn, long hint); 174 static int filt_soempty(struct knote *kn, long hint); 175 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id); 176 fo_kqfilter_t soo_kqfilter; 177 178 static struct filterops soread_filtops = { 179 .f_isfd = 1, 180 .f_detach = filt_sordetach, 181 .f_event = filt_soread, 182 }; 183 static struct filterops sowrite_filtops = { 184 .f_isfd = 1, 185 .f_detach = filt_sowdetach, 186 .f_event = filt_sowrite, 187 }; 188 static struct filterops soempty_filtops = { 189 .f_isfd = 1, 190 .f_detach = filt_sowdetach, 191 .f_event = filt_soempty, 192 }; 193 194 so_gen_t so_gencnt; /* generation count for sockets */ 195 196 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 197 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 198 199 #define VNET_SO_ASSERT(so) \ 200 VNET_ASSERT(curvnet != NULL, \ 201 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so))); 202 203 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]); 204 #define V_socket_hhh VNET(socket_hhh) 205 206 /* 207 * Limit on the number of connections in the listen queue waiting 208 * for accept(2). 209 * NB: The original sysctl somaxconn is still available but hidden 210 * to prevent confusion about the actual purpose of this number. 211 */ 212 static u_int somaxconn = SOMAXCONN; 213 214 static int 215 sysctl_somaxconn(SYSCTL_HANDLER_ARGS) 216 { 217 int error; 218 int val; 219 220 val = somaxconn; 221 error = sysctl_handle_int(oidp, &val, 0, req); 222 if (error || !req->newptr ) 223 return (error); 224 225 /* 226 * The purpose of the UINT_MAX / 3 limit, is so that the formula 227 * 3 * so_qlimit / 2 228 * below, will not overflow. 229 */ 230 231 if (val < 1 || val > UINT_MAX / 3) 232 return (EINVAL); 233 234 somaxconn = val; 235 return (0); 236 } 237 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, 238 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int), 239 sysctl_somaxconn, "I", 240 "Maximum listen socket pending connection accept queue size"); 241 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, 242 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, 0, 243 sizeof(int), sysctl_somaxconn, "I", 244 "Maximum listen socket pending connection accept queue size (compat)"); 245 246 static int numopensockets; 247 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD, 248 &numopensockets, 0, "Number of open sockets"); 249 250 /* 251 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket 252 * so_gencnt field. 253 */ 254 static struct mtx so_global_mtx; 255 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF); 256 257 /* 258 * General IPC sysctl name space, used by sockets and a variety of other IPC 259 * types. 260 */ 261 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 262 "IPC"); 263 264 /* 265 * Initialize the socket subsystem and set up the socket 266 * memory allocator. 267 */ 268 static uma_zone_t socket_zone; 269 int maxsockets; 270 271 static void 272 socket_zone_change(void *tag) 273 { 274 275 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 276 } 277 278 static void 279 socket_hhook_register(int subtype) 280 { 281 282 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype, 283 &V_socket_hhh[subtype], 284 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 285 printf("%s: WARNING: unable to register hook\n", __func__); 286 } 287 288 static void 289 socket_hhook_deregister(int subtype) 290 { 291 292 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0) 293 printf("%s: WARNING: unable to deregister hook\n", __func__); 294 } 295 296 static void 297 socket_init(void *tag) 298 { 299 300 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL, 301 NULL, NULL, UMA_ALIGN_PTR, 0); 302 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 303 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached"); 304 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL, 305 EVENTHANDLER_PRI_FIRST); 306 } 307 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL); 308 309 static void 310 socket_vnet_init(const void *unused __unused) 311 { 312 int i; 313 314 /* We expect a contiguous range */ 315 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 316 socket_hhook_register(i); 317 } 318 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 319 socket_vnet_init, NULL); 320 321 static void 322 socket_vnet_uninit(const void *unused __unused) 323 { 324 int i; 325 326 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 327 socket_hhook_deregister(i); 328 } 329 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 330 socket_vnet_uninit, NULL); 331 332 /* 333 * Initialise maxsockets. This SYSINIT must be run after 334 * tunable_mbinit(). 335 */ 336 static void 337 init_maxsockets(void *ignored) 338 { 339 340 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 341 maxsockets = imax(maxsockets, maxfiles); 342 } 343 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); 344 345 /* 346 * Sysctl to get and set the maximum global sockets limit. Notify protocols 347 * of the change so that they can update their dependent limits as required. 348 */ 349 static int 350 sysctl_maxsockets(SYSCTL_HANDLER_ARGS) 351 { 352 int error, newmaxsockets; 353 354 newmaxsockets = maxsockets; 355 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req); 356 if (error == 0 && req->newptr && newmaxsockets != maxsockets) { 357 if (newmaxsockets > maxsockets && 358 newmaxsockets <= maxfiles) { 359 maxsockets = newmaxsockets; 360 EVENTHANDLER_INVOKE(maxsockets_change); 361 } else 362 error = EINVAL; 363 } 364 return (error); 365 } 366 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, 367 CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NOFETCH | CTLFLAG_MPSAFE, 368 &maxsockets, 0, sysctl_maxsockets, "IU", 369 "Maximum number of sockets available"); 370 371 /* 372 * Socket operation routines. These routines are called by the routines in 373 * sys_socket.c or from a system process, and implement the semantics of 374 * socket operations by switching out to the protocol specific routines. 375 */ 376 377 /* 378 * Get a socket structure from our zone, and initialize it. Note that it 379 * would probably be better to allocate socket and PCB at the same time, but 380 * I'm not convinced that all the protocols can be easily modified to do 381 * this. 382 * 383 * soalloc() returns a socket with a ref count of 0. 384 */ 385 static struct socket * 386 soalloc(struct vnet *vnet) 387 { 388 struct socket *so; 389 390 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO); 391 if (so == NULL) 392 return (NULL); 393 #ifdef MAC 394 if (mac_socket_init(so, M_NOWAIT) != 0) { 395 uma_zfree(socket_zone, so); 396 return (NULL); 397 } 398 #endif 399 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) { 400 uma_zfree(socket_zone, so); 401 return (NULL); 402 } 403 404 /* 405 * The socket locking protocol allows to lock 2 sockets at a time, 406 * however, the first one must be a listening socket. WITNESS lacks 407 * a feature to change class of an existing lock, so we use DUPOK. 408 */ 409 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK); 410 mtx_init(&so->so_snd_mtx, "so_snd", NULL, MTX_DEF); 411 mtx_init(&so->so_rcv_mtx, "so_rcv", NULL, MTX_DEF); 412 so->so_rcv.sb_sel = &so->so_rdsel; 413 so->so_snd.sb_sel = &so->so_wrsel; 414 sx_init(&so->so_snd_sx, "so_snd_sx"); 415 sx_init(&so->so_rcv_sx, "so_rcv_sx"); 416 TAILQ_INIT(&so->so_snd.sb_aiojobq); 417 TAILQ_INIT(&so->so_rcv.sb_aiojobq); 418 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so); 419 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so); 420 #ifdef VIMAGE 421 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p", 422 __func__, __LINE__, so)); 423 so->so_vnet = vnet; 424 #endif 425 /* We shouldn't need the so_global_mtx */ 426 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) { 427 /* Do we need more comprehensive error returns? */ 428 uma_zfree(socket_zone, so); 429 return (NULL); 430 } 431 mtx_lock(&so_global_mtx); 432 so->so_gencnt = ++so_gencnt; 433 ++numopensockets; 434 #ifdef VIMAGE 435 vnet->vnet_sockcnt++; 436 #endif 437 mtx_unlock(&so_global_mtx); 438 439 return (so); 440 } 441 442 /* 443 * Free the storage associated with a socket at the socket layer, tear down 444 * locks, labels, etc. All protocol state is assumed already to have been 445 * torn down (and possibly never set up) by the caller. 446 */ 447 void 448 sodealloc(struct socket *so) 449 { 450 451 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); 452 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL")); 453 454 mtx_lock(&so_global_mtx); 455 so->so_gencnt = ++so_gencnt; 456 --numopensockets; /* Could be below, but faster here. */ 457 #ifdef VIMAGE 458 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p", 459 __func__, __LINE__, so)); 460 so->so_vnet->vnet_sockcnt--; 461 #endif 462 mtx_unlock(&so_global_mtx); 463 #ifdef MAC 464 mac_socket_destroy(so); 465 #endif 466 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE); 467 468 khelp_destroy_osd(&so->osd); 469 if (SOLISTENING(so)) { 470 if (so->sol_accept_filter != NULL) 471 accept_filt_setopt(so, NULL); 472 } else { 473 if (so->so_rcv.sb_hiwat) 474 (void)chgsbsize(so->so_cred->cr_uidinfo, 475 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 476 if (so->so_snd.sb_hiwat) 477 (void)chgsbsize(so->so_cred->cr_uidinfo, 478 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 479 sx_destroy(&so->so_snd_sx); 480 sx_destroy(&so->so_rcv_sx); 481 mtx_destroy(&so->so_snd_mtx); 482 mtx_destroy(&so->so_rcv_mtx); 483 } 484 crfree(so->so_cred); 485 mtx_destroy(&so->so_lock); 486 uma_zfree(socket_zone, so); 487 } 488 489 /* 490 * socreate returns a socket with a ref count of 1 and a file descriptor 491 * reference. The socket should be closed with soclose(). 492 */ 493 int 494 socreate(int dom, struct socket **aso, int type, int proto, 495 struct ucred *cred, struct thread *td) 496 { 497 struct protosw *prp; 498 struct socket *so; 499 int error; 500 501 /* 502 * XXX: divert(4) historically abused PF_INET. Keep this compatibility 503 * shim until all applications have been updated. 504 */ 505 if (__predict_false(dom == PF_INET && type == SOCK_RAW && 506 proto == IPPROTO_DIVERT)) { 507 dom = PF_DIVERT; 508 printf("%s uses obsolete way to create divert(4) socket\n", 509 td->td_proc->p_comm); 510 } 511 512 prp = pffindproto(dom, type, proto); 513 if (prp == NULL) { 514 /* No support for domain. */ 515 if (pffinddomain(dom) == NULL) 516 return (EAFNOSUPPORT); 517 /* No support for socket type. */ 518 if (proto == 0 && type != 0) 519 return (EPROTOTYPE); 520 return (EPROTONOSUPPORT); 521 } 522 523 MPASS(prp->pr_attach); 524 525 if (IN_CAPABILITY_MODE(td) && (prp->pr_flags & PR_CAPATTACH) == 0) 526 return (ECAPMODE); 527 528 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0) 529 return (EPROTONOSUPPORT); 530 531 so = soalloc(CRED_TO_VNET(cred)); 532 if (so == NULL) 533 return (ENOBUFS); 534 535 so->so_type = type; 536 so->so_cred = crhold(cred); 537 if ((prp->pr_domain->dom_family == PF_INET) || 538 (prp->pr_domain->dom_family == PF_INET6) || 539 (prp->pr_domain->dom_family == PF_ROUTE)) 540 so->so_fibnum = td->td_proc->p_fibnum; 541 else 542 so->so_fibnum = 0; 543 so->so_proto = prp; 544 #ifdef MAC 545 mac_socket_create(cred, so); 546 #endif 547 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 548 so_rdknl_assert_lock); 549 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 550 so_wrknl_assert_lock); 551 if ((prp->pr_flags & PR_SOCKBUF) == 0) { 552 so->so_snd.sb_mtx = &so->so_snd_mtx; 553 so->so_rcv.sb_mtx = &so->so_rcv_mtx; 554 } 555 /* 556 * Auto-sizing of socket buffers is managed by the protocols and 557 * the appropriate flags must be set in the pru_attach function. 558 */ 559 CURVNET_SET(so->so_vnet); 560 error = prp->pr_attach(so, proto, td); 561 CURVNET_RESTORE(); 562 if (error) { 563 sodealloc(so); 564 return (error); 565 } 566 soref(so); 567 *aso = so; 568 return (0); 569 } 570 571 #ifdef REGRESSION 572 static int regression_sonewconn_earlytest = 1; 573 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW, 574 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test"); 575 #endif 576 577 static int sooverprio = LOG_DEBUG; 578 SYSCTL_INT(_kern_ipc, OID_AUTO, sooverprio, CTLFLAG_RW, 579 &sooverprio, 0, "Log priority for listen socket overflows: 0..7 or -1 to disable"); 580 581 static struct timeval overinterval = { 60, 0 }; 582 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW, 583 &overinterval, 584 "Delay in seconds between warnings for listen socket overflows"); 585 586 /* 587 * When an attempt at a new connection is noted on a socket which supports 588 * accept(2), the protocol has two options: 589 * 1) Call legacy sonewconn() function, which would call protocol attach 590 * method, same as used for socket(2). 591 * 2) Call solisten_clone(), do attach that is specific to a cloned connection, 592 * and then call solisten_enqueue(). 593 * 594 * Note: the ref count on the socket is 0 on return. 595 */ 596 struct socket * 597 solisten_clone(struct socket *head) 598 { 599 struct sbuf descrsb; 600 struct socket *so; 601 int len, overcount; 602 u_int qlen; 603 const char localprefix[] = "local:"; 604 char descrbuf[SUNPATHLEN + sizeof(localprefix)]; 605 #if defined(INET6) 606 char addrbuf[INET6_ADDRSTRLEN]; 607 #elif defined(INET) 608 char addrbuf[INET_ADDRSTRLEN]; 609 #endif 610 bool dolog, over; 611 612 SOLISTEN_LOCK(head); 613 over = (head->sol_qlen > 3 * head->sol_qlimit / 2); 614 #ifdef REGRESSION 615 if (regression_sonewconn_earlytest && over) { 616 #else 617 if (over) { 618 #endif 619 head->sol_overcount++; 620 dolog = (sooverprio >= 0) && 621 !!ratecheck(&head->sol_lastover, &overinterval); 622 623 /* 624 * If we're going to log, copy the overflow count and queue 625 * length from the listen socket before dropping the lock. 626 * Also, reset the overflow count. 627 */ 628 if (dolog) { 629 overcount = head->sol_overcount; 630 head->sol_overcount = 0; 631 qlen = head->sol_qlen; 632 } 633 SOLISTEN_UNLOCK(head); 634 635 if (dolog) { 636 /* 637 * Try to print something descriptive about the 638 * socket for the error message. 639 */ 640 sbuf_new(&descrsb, descrbuf, sizeof(descrbuf), 641 SBUF_FIXEDLEN); 642 switch (head->so_proto->pr_domain->dom_family) { 643 #if defined(INET) || defined(INET6) 644 #ifdef INET 645 case AF_INET: 646 #endif 647 #ifdef INET6 648 case AF_INET6: 649 if (head->so_proto->pr_domain->dom_family == 650 AF_INET6 || 651 (sotoinpcb(head)->inp_inc.inc_flags & 652 INC_ISIPV6)) { 653 ip6_sprintf(addrbuf, 654 &sotoinpcb(head)->inp_inc.inc6_laddr); 655 sbuf_printf(&descrsb, "[%s]", addrbuf); 656 } else 657 #endif 658 { 659 #ifdef INET 660 inet_ntoa_r( 661 sotoinpcb(head)->inp_inc.inc_laddr, 662 addrbuf); 663 sbuf_cat(&descrsb, addrbuf); 664 #endif 665 } 666 sbuf_printf(&descrsb, ":%hu (proto %u)", 667 ntohs(sotoinpcb(head)->inp_inc.inc_lport), 668 head->so_proto->pr_protocol); 669 break; 670 #endif /* INET || INET6 */ 671 case AF_UNIX: 672 sbuf_cat(&descrsb, localprefix); 673 if (sotounpcb(head)->unp_addr != NULL) 674 len = 675 sotounpcb(head)->unp_addr->sun_len - 676 offsetof(struct sockaddr_un, 677 sun_path); 678 else 679 len = 0; 680 if (len > 0) 681 sbuf_bcat(&descrsb, 682 sotounpcb(head)->unp_addr->sun_path, 683 len); 684 else 685 sbuf_cat(&descrsb, "(unknown)"); 686 break; 687 } 688 689 /* 690 * If we can't print something more specific, at least 691 * print the domain name. 692 */ 693 if (sbuf_finish(&descrsb) != 0 || 694 sbuf_len(&descrsb) <= 0) { 695 sbuf_clear(&descrsb); 696 sbuf_cat(&descrsb, 697 head->so_proto->pr_domain->dom_name ?: 698 "unknown"); 699 sbuf_finish(&descrsb); 700 } 701 KASSERT(sbuf_len(&descrsb) > 0, 702 ("%s: sbuf creation failed", __func__)); 703 /* 704 * Preserve the historic listen queue overflow log 705 * message, that starts with "sonewconn:". It has 706 * been known to sysadmins for years and also test 707 * sys/kern/sonewconn_overflow checks for it. 708 */ 709 if (head->so_cred == 0) { 710 log(LOG_PRI(sooverprio), 711 "sonewconn: pcb %p (%s): " 712 "Listen queue overflow: %i already in " 713 "queue awaiting acceptance (%d " 714 "occurrences)\n", head->so_pcb, 715 sbuf_data(&descrsb), 716 qlen, overcount); 717 } else { 718 log(LOG_PRI(sooverprio), 719 "sonewconn: pcb %p (%s): " 720 "Listen queue overflow: " 721 "%i already in queue awaiting acceptance " 722 "(%d occurrences), euid %d, rgid %d, jail %s\n", 723 head->so_pcb, sbuf_data(&descrsb), qlen, 724 overcount, head->so_cred->cr_uid, 725 head->so_cred->cr_rgid, 726 head->so_cred->cr_prison ? 727 head->so_cred->cr_prison->pr_name : 728 "not_jailed"); 729 } 730 sbuf_delete(&descrsb); 731 732 overcount = 0; 733 } 734 735 return (NULL); 736 } 737 SOLISTEN_UNLOCK(head); 738 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL", 739 __func__, head)); 740 so = soalloc(head->so_vnet); 741 if (so == NULL) { 742 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 743 "limit reached or out of memory\n", 744 __func__, head->so_pcb); 745 return (NULL); 746 } 747 so->so_listen = head; 748 so->so_type = head->so_type; 749 /* 750 * POSIX is ambiguous on what options an accept(2)ed socket should 751 * inherit from the listener. Words "create a new socket" may be 752 * interpreted as not inheriting anything. Best programming practice 753 * for application developers is to not rely on such inheritance. 754 * FreeBSD had historically inherited all so_options excluding 755 * SO_ACCEPTCONN, which virtually means all SOL_SOCKET level options, 756 * including those completely irrelevant to a new born socket. For 757 * compatibility with older versions we will inherit a list of 758 * meaningful options. 759 */ 760 so->so_options = head->so_options & (SO_KEEPALIVE | SO_DONTROUTE | 761 SO_LINGER | SO_OOBINLINE | SO_NOSIGPIPE); 762 so->so_linger = head->so_linger; 763 so->so_state = head->so_state; 764 so->so_fibnum = head->so_fibnum; 765 so->so_proto = head->so_proto; 766 so->so_cred = crhold(head->so_cred); 767 #ifdef MAC 768 mac_socket_newconn(head, so); 769 #endif 770 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 771 so_rdknl_assert_lock); 772 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 773 so_wrknl_assert_lock); 774 VNET_SO_ASSERT(head); 775 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) { 776 sodealloc(so); 777 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 778 __func__, head->so_pcb); 779 return (NULL); 780 } 781 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat; 782 so->so_snd.sb_lowat = head->sol_sbsnd_lowat; 783 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo; 784 so->so_snd.sb_timeo = head->sol_sbsnd_timeo; 785 so->so_rcv.sb_flags = head->sol_sbrcv_flags & SB_AUTOSIZE; 786 so->so_snd.sb_flags = head->sol_sbsnd_flags & SB_AUTOSIZE; 787 if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) { 788 so->so_snd.sb_mtx = &so->so_snd_mtx; 789 so->so_rcv.sb_mtx = &so->so_rcv_mtx; 790 } 791 792 return (so); 793 } 794 795 /* Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED. */ 796 struct socket * 797 sonewconn(struct socket *head, int connstatus) 798 { 799 struct socket *so; 800 801 if ((so = solisten_clone(head)) == NULL) 802 return (NULL); 803 804 if (so->so_proto->pr_attach(so, 0, NULL) != 0) { 805 sodealloc(so); 806 log(LOG_DEBUG, "%s: pcb %p: pr_attach() failed\n", 807 __func__, head->so_pcb); 808 return (NULL); 809 } 810 811 (void)solisten_enqueue(so, connstatus); 812 813 return (so); 814 } 815 816 /* 817 * Enqueue socket cloned by solisten_clone() to the listen queue of the 818 * listener it has been cloned from. 819 * 820 * Return 'true' if socket landed on complete queue, otherwise 'false'. 821 */ 822 bool 823 solisten_enqueue(struct socket *so, int connstatus) 824 { 825 struct socket *head = so->so_listen; 826 827 MPASS(refcount_load(&so->so_count) == 0); 828 refcount_init(&so->so_count, 1); 829 830 SOLISTEN_LOCK(head); 831 if (head->sol_accept_filter != NULL) 832 connstatus = 0; 833 so->so_state |= connstatus; 834 soref(head); /* A socket on (in)complete queue refs head. */ 835 if (connstatus) { 836 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 837 so->so_qstate = SQ_COMP; 838 head->sol_qlen++; 839 solisten_wakeup(head); /* unlocks */ 840 return (true); 841 } else { 842 /* 843 * Keep removing sockets from the head until there's room for 844 * us to insert on the tail. In pre-locking revisions, this 845 * was a simple if(), but as we could be racing with other 846 * threads and soabort() requires dropping locks, we must 847 * loop waiting for the condition to be true. 848 */ 849 while (head->sol_incqlen > head->sol_qlimit) { 850 struct socket *sp; 851 852 sp = TAILQ_FIRST(&head->sol_incomp); 853 TAILQ_REMOVE(&head->sol_incomp, sp, so_list); 854 head->sol_incqlen--; 855 SOCK_LOCK(sp); 856 sp->so_qstate = SQ_NONE; 857 sp->so_listen = NULL; 858 SOCK_UNLOCK(sp); 859 sorele_locked(head); /* does SOLISTEN_UNLOCK, head stays */ 860 soabort(sp); 861 SOLISTEN_LOCK(head); 862 } 863 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list); 864 so->so_qstate = SQ_INCOMP; 865 head->sol_incqlen++; 866 SOLISTEN_UNLOCK(head); 867 return (false); 868 } 869 } 870 871 #if defined(SCTP) || defined(SCTP_SUPPORT) 872 /* 873 * Socket part of sctp_peeloff(). Detach a new socket from an 874 * association. The new socket is returned with a reference. 875 * 876 * XXXGL: reduce copy-paste with solisten_clone(). 877 */ 878 struct socket * 879 sopeeloff(struct socket *head) 880 { 881 struct socket *so; 882 883 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p", 884 __func__, __LINE__, head)); 885 so = soalloc(head->so_vnet); 886 if (so == NULL) { 887 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 888 "limit reached or out of memory\n", 889 __func__, head->so_pcb); 890 return (NULL); 891 } 892 so->so_type = head->so_type; 893 so->so_options = head->so_options; 894 so->so_linger = head->so_linger; 895 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED; 896 so->so_fibnum = head->so_fibnum; 897 so->so_proto = head->so_proto; 898 so->so_cred = crhold(head->so_cred); 899 #ifdef MAC 900 mac_socket_newconn(head, so); 901 #endif 902 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 903 so_rdknl_assert_lock); 904 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 905 so_wrknl_assert_lock); 906 VNET_SO_ASSERT(head); 907 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { 908 sodealloc(so); 909 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 910 __func__, head->so_pcb); 911 return (NULL); 912 } 913 if ((*so->so_proto->pr_attach)(so, 0, NULL)) { 914 sodealloc(so); 915 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 916 __func__, head->so_pcb); 917 return (NULL); 918 } 919 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 920 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 921 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 922 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 923 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; 924 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; 925 926 soref(so); 927 928 return (so); 929 } 930 #endif /* SCTP */ 931 932 int 933 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 934 { 935 int error; 936 937 CURVNET_SET(so->so_vnet); 938 error = so->so_proto->pr_bind(so, nam, td); 939 CURVNET_RESTORE(); 940 return (error); 941 } 942 943 int 944 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 945 { 946 int error; 947 948 CURVNET_SET(so->so_vnet); 949 error = so->so_proto->pr_bindat(fd, so, nam, td); 950 CURVNET_RESTORE(); 951 return (error); 952 } 953 954 /* 955 * solisten() transitions a socket from a non-listening state to a listening 956 * state, but can also be used to update the listen queue depth on an 957 * existing listen socket. The protocol will call back into the sockets 958 * layer using solisten_proto_check() and solisten_proto() to check and set 959 * socket-layer listen state. Call backs are used so that the protocol can 960 * acquire both protocol and socket layer locks in whatever order is required 961 * by the protocol. 962 * 963 * Protocol implementors are advised to hold the socket lock across the 964 * socket-layer test and set to avoid races at the socket layer. 965 */ 966 int 967 solisten(struct socket *so, int backlog, struct thread *td) 968 { 969 int error; 970 971 CURVNET_SET(so->so_vnet); 972 error = so->so_proto->pr_listen(so, backlog, td); 973 CURVNET_RESTORE(); 974 return (error); 975 } 976 977 /* 978 * Prepare for a call to solisten_proto(). Acquire all socket buffer locks in 979 * order to interlock with socket I/O. 980 */ 981 int 982 solisten_proto_check(struct socket *so) 983 { 984 SOCK_LOCK_ASSERT(so); 985 986 if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 987 SS_ISDISCONNECTING)) != 0) 988 return (EINVAL); 989 990 /* 991 * Sleeping is not permitted here, so simply fail if userspace is 992 * attempting to transmit or receive on the socket. This kind of 993 * transient failure is not ideal, but it should occur only if userspace 994 * is misusing the socket interfaces. 995 */ 996 if (!sx_try_xlock(&so->so_snd_sx)) 997 return (EAGAIN); 998 if (!sx_try_xlock(&so->so_rcv_sx)) { 999 sx_xunlock(&so->so_snd_sx); 1000 return (EAGAIN); 1001 } 1002 mtx_lock(&so->so_snd_mtx); 1003 mtx_lock(&so->so_rcv_mtx); 1004 1005 /* Interlock with soo_aio_queue() and KTLS. */ 1006 if (!SOLISTENING(so)) { 1007 bool ktls; 1008 1009 #ifdef KERN_TLS 1010 ktls = so->so_snd.sb_tls_info != NULL || 1011 so->so_rcv.sb_tls_info != NULL; 1012 #else 1013 ktls = false; 1014 #endif 1015 if (ktls || 1016 (so->so_snd.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0 || 1017 (so->so_rcv.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0) { 1018 solisten_proto_abort(so); 1019 return (EINVAL); 1020 } 1021 } 1022 1023 return (0); 1024 } 1025 1026 /* 1027 * Undo the setup done by solisten_proto_check(). 1028 */ 1029 void 1030 solisten_proto_abort(struct socket *so) 1031 { 1032 mtx_unlock(&so->so_snd_mtx); 1033 mtx_unlock(&so->so_rcv_mtx); 1034 sx_xunlock(&so->so_snd_sx); 1035 sx_xunlock(&so->so_rcv_sx); 1036 } 1037 1038 void 1039 solisten_proto(struct socket *so, int backlog) 1040 { 1041 int sbrcv_lowat, sbsnd_lowat; 1042 u_int sbrcv_hiwat, sbsnd_hiwat; 1043 short sbrcv_flags, sbsnd_flags; 1044 sbintime_t sbrcv_timeo, sbsnd_timeo; 1045 1046 SOCK_LOCK_ASSERT(so); 1047 KASSERT((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 1048 SS_ISDISCONNECTING)) == 0, 1049 ("%s: bad socket state %p", __func__, so)); 1050 1051 if (SOLISTENING(so)) 1052 goto listening; 1053 1054 /* 1055 * Change this socket to listening state. 1056 */ 1057 sbrcv_lowat = so->so_rcv.sb_lowat; 1058 sbsnd_lowat = so->so_snd.sb_lowat; 1059 sbrcv_hiwat = so->so_rcv.sb_hiwat; 1060 sbsnd_hiwat = so->so_snd.sb_hiwat; 1061 sbrcv_flags = so->so_rcv.sb_flags; 1062 sbsnd_flags = so->so_snd.sb_flags; 1063 sbrcv_timeo = so->so_rcv.sb_timeo; 1064 sbsnd_timeo = so->so_snd.sb_timeo; 1065 1066 sbdestroy(so, SO_SND); 1067 sbdestroy(so, SO_RCV); 1068 1069 #ifdef INVARIANTS 1070 bzero(&so->so_rcv, 1071 sizeof(struct socket) - offsetof(struct socket, so_rcv)); 1072 #endif 1073 1074 so->sol_sbrcv_lowat = sbrcv_lowat; 1075 so->sol_sbsnd_lowat = sbsnd_lowat; 1076 so->sol_sbrcv_hiwat = sbrcv_hiwat; 1077 so->sol_sbsnd_hiwat = sbsnd_hiwat; 1078 so->sol_sbrcv_flags = sbrcv_flags; 1079 so->sol_sbsnd_flags = sbsnd_flags; 1080 so->sol_sbrcv_timeo = sbrcv_timeo; 1081 so->sol_sbsnd_timeo = sbsnd_timeo; 1082 1083 so->sol_qlen = so->sol_incqlen = 0; 1084 TAILQ_INIT(&so->sol_incomp); 1085 TAILQ_INIT(&so->sol_comp); 1086 1087 so->sol_accept_filter = NULL; 1088 so->sol_accept_filter_arg = NULL; 1089 so->sol_accept_filter_str = NULL; 1090 1091 so->sol_upcall = NULL; 1092 so->sol_upcallarg = NULL; 1093 1094 so->so_options |= SO_ACCEPTCONN; 1095 1096 listening: 1097 if (backlog < 0 || backlog > somaxconn) 1098 backlog = somaxconn; 1099 so->sol_qlimit = backlog; 1100 1101 mtx_unlock(&so->so_snd_mtx); 1102 mtx_unlock(&so->so_rcv_mtx); 1103 sx_xunlock(&so->so_snd_sx); 1104 sx_xunlock(&so->so_rcv_sx); 1105 } 1106 1107 /* 1108 * Wakeup listeners/subsystems once we have a complete connection. 1109 * Enters with lock, returns unlocked. 1110 */ 1111 void 1112 solisten_wakeup(struct socket *sol) 1113 { 1114 1115 if (sol->sol_upcall != NULL) 1116 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT); 1117 else { 1118 selwakeuppri(&sol->so_rdsel, PSOCK); 1119 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0); 1120 } 1121 SOLISTEN_UNLOCK(sol); 1122 wakeup_one(&sol->sol_comp); 1123 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL) 1124 pgsigio(&sol->so_sigio, SIGIO, 0); 1125 } 1126 1127 /* 1128 * Return single connection off a listening socket queue. Main consumer of 1129 * the function is kern_accept4(). Some modules, that do their own accept 1130 * management also use the function. The socket reference held by the 1131 * listen queue is handed to the caller. 1132 * 1133 * Listening socket must be locked on entry and is returned unlocked on 1134 * return. 1135 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT. 1136 */ 1137 int 1138 solisten_dequeue(struct socket *head, struct socket **ret, int flags) 1139 { 1140 struct socket *so; 1141 int error; 1142 1143 SOLISTEN_LOCK_ASSERT(head); 1144 1145 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) && 1146 head->so_error == 0) { 1147 error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH, 1148 "accept", 0); 1149 if (error != 0) { 1150 SOLISTEN_UNLOCK(head); 1151 return (error); 1152 } 1153 } 1154 if (head->so_error) { 1155 error = head->so_error; 1156 head->so_error = 0; 1157 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp)) 1158 error = EWOULDBLOCK; 1159 else 1160 error = 0; 1161 if (error) { 1162 SOLISTEN_UNLOCK(head); 1163 return (error); 1164 } 1165 so = TAILQ_FIRST(&head->sol_comp); 1166 SOCK_LOCK(so); 1167 KASSERT(so->so_qstate == SQ_COMP, 1168 ("%s: so %p not SQ_COMP", __func__, so)); 1169 head->sol_qlen--; 1170 so->so_qstate = SQ_NONE; 1171 so->so_listen = NULL; 1172 TAILQ_REMOVE(&head->sol_comp, so, so_list); 1173 if (flags & ACCEPT4_INHERIT) 1174 so->so_state |= (head->so_state & SS_NBIO); 1175 else 1176 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0; 1177 SOCK_UNLOCK(so); 1178 sorele_locked(head); 1179 1180 *ret = so; 1181 return (0); 1182 } 1183 1184 /* 1185 * Free socket upon release of the very last reference. 1186 */ 1187 static void 1188 sofree(struct socket *so) 1189 { 1190 struct protosw *pr = so->so_proto; 1191 1192 SOCK_LOCK_ASSERT(so); 1193 KASSERT(refcount_load(&so->so_count) == 0, 1194 ("%s: so %p has references", __func__, so)); 1195 KASSERT(SOLISTENING(so) || so->so_qstate == SQ_NONE, 1196 ("%s: so %p is on listen queue", __func__, so)); 1197 1198 SOCK_UNLOCK(so); 1199 1200 if (so->so_dtor != NULL) 1201 so->so_dtor(so); 1202 1203 VNET_SO_ASSERT(so); 1204 if ((pr->pr_flags & PR_RIGHTS) && !SOLISTENING(so)) { 1205 MPASS(pr->pr_domain->dom_dispose != NULL); 1206 (*pr->pr_domain->dom_dispose)(so); 1207 } 1208 if (pr->pr_detach != NULL) 1209 pr->pr_detach(so); 1210 1211 /* 1212 * From this point on, we assume that no other references to this 1213 * socket exist anywhere else in the stack. Therefore, no locks need 1214 * to be acquired or held. 1215 */ 1216 if (!(pr->pr_flags & PR_SOCKBUF) && !SOLISTENING(so)) { 1217 sbdestroy(so, SO_SND); 1218 sbdestroy(so, SO_RCV); 1219 } 1220 seldrain(&so->so_rdsel); 1221 seldrain(&so->so_wrsel); 1222 knlist_destroy(&so->so_rdsel.si_note); 1223 knlist_destroy(&so->so_wrsel.si_note); 1224 sodealloc(so); 1225 } 1226 1227 /* 1228 * Release a reference on a socket while holding the socket lock. 1229 * Unlocks the socket lock before returning. 1230 */ 1231 void 1232 sorele_locked(struct socket *so) 1233 { 1234 SOCK_LOCK_ASSERT(so); 1235 if (refcount_release(&so->so_count)) 1236 sofree(so); 1237 else 1238 SOCK_UNLOCK(so); 1239 } 1240 1241 /* 1242 * Close a socket on last file table reference removal. Initiate disconnect 1243 * if connected. Free socket when disconnect complete. 1244 * 1245 * This function will sorele() the socket. Note that soclose() may be called 1246 * prior to the ref count reaching zero. The actual socket structure will 1247 * not be freed until the ref count reaches zero. 1248 */ 1249 int 1250 soclose(struct socket *so) 1251 { 1252 struct accept_queue lqueue; 1253 int error = 0; 1254 bool listening, last __diagused; 1255 1256 CURVNET_SET(so->so_vnet); 1257 funsetown(&so->so_sigio); 1258 if (so->so_state & SS_ISCONNECTED) { 1259 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 1260 error = sodisconnect(so); 1261 if (error) { 1262 if (error == ENOTCONN) 1263 error = 0; 1264 goto drop; 1265 } 1266 } 1267 1268 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) { 1269 if ((so->so_state & SS_ISDISCONNECTING) && 1270 (so->so_state & SS_NBIO)) 1271 goto drop; 1272 while (so->so_state & SS_ISCONNECTED) { 1273 error = tsleep(&so->so_timeo, 1274 PSOCK | PCATCH, "soclos", 1275 so->so_linger * hz); 1276 if (error) 1277 break; 1278 } 1279 } 1280 } 1281 1282 drop: 1283 if (so->so_proto->pr_close != NULL) 1284 so->so_proto->pr_close(so); 1285 1286 SOCK_LOCK(so); 1287 if ((listening = SOLISTENING(so))) { 1288 struct socket *sp; 1289 1290 TAILQ_INIT(&lqueue); 1291 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list); 1292 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list); 1293 1294 so->sol_qlen = so->sol_incqlen = 0; 1295 1296 TAILQ_FOREACH(sp, &lqueue, so_list) { 1297 SOCK_LOCK(sp); 1298 sp->so_qstate = SQ_NONE; 1299 sp->so_listen = NULL; 1300 SOCK_UNLOCK(sp); 1301 last = refcount_release(&so->so_count); 1302 KASSERT(!last, ("%s: released last reference for %p", 1303 __func__, so)); 1304 } 1305 } 1306 sorele_locked(so); 1307 if (listening) { 1308 struct socket *sp, *tsp; 1309 1310 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) 1311 soabort(sp); 1312 } 1313 CURVNET_RESTORE(); 1314 return (error); 1315 } 1316 1317 /* 1318 * soabort() is used to abruptly tear down a connection, such as when a 1319 * resource limit is reached (listen queue depth exceeded), or if a listen 1320 * socket is closed while there are sockets waiting to be accepted. 1321 * 1322 * This interface is tricky, because it is called on an unreferenced socket, 1323 * and must be called only by a thread that has actually removed the socket 1324 * from the listen queue it was on. Likely this thread holds the last 1325 * reference on the socket and soabort() will proceed with sofree(). But 1326 * it might be not the last, as the sockets on the listen queues are seen 1327 * from the protocol side. 1328 * 1329 * This interface will call into the protocol code, so must not be called 1330 * with any socket locks held. Protocols do call it while holding their own 1331 * recursible protocol mutexes, but this is something that should be subject 1332 * to review in the future. 1333 * 1334 * Usually socket should have a single reference left, but this is not a 1335 * requirement. In the past, when we have had named references for file 1336 * descriptor and protocol, we asserted that none of them are being held. 1337 */ 1338 void 1339 soabort(struct socket *so) 1340 { 1341 1342 VNET_SO_ASSERT(so); 1343 1344 if (so->so_proto->pr_abort != NULL) 1345 so->so_proto->pr_abort(so); 1346 SOCK_LOCK(so); 1347 sorele_locked(so); 1348 } 1349 1350 int 1351 soaccept(struct socket *so, struct sockaddr **nam) 1352 { 1353 int error; 1354 1355 CURVNET_SET(so->so_vnet); 1356 error = so->so_proto->pr_accept(so, nam); 1357 CURVNET_RESTORE(); 1358 return (error); 1359 } 1360 1361 int 1362 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 1363 { 1364 1365 return (soconnectat(AT_FDCWD, so, nam, td)); 1366 } 1367 1368 int 1369 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 1370 { 1371 int error; 1372 1373 CURVNET_SET(so->so_vnet); 1374 1375 /* 1376 * If protocol is connection-based, can only connect once. 1377 * Otherwise, if connected, try to disconnect first. This allows 1378 * user to disconnect by connecting to, e.g., a null address. 1379 * 1380 * Note, this check is racy and may need to be re-evaluated at the 1381 * protocol layer. 1382 */ 1383 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 1384 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 1385 (error = sodisconnect(so)))) { 1386 error = EISCONN; 1387 } else { 1388 /* 1389 * Prevent accumulated error from previous connection from 1390 * biting us. 1391 */ 1392 so->so_error = 0; 1393 if (fd == AT_FDCWD) { 1394 error = so->so_proto->pr_connect(so, nam, td); 1395 } else { 1396 error = so->so_proto->pr_connectat(fd, so, nam, td); 1397 } 1398 } 1399 CURVNET_RESTORE(); 1400 1401 return (error); 1402 } 1403 1404 int 1405 soconnect2(struct socket *so1, struct socket *so2) 1406 { 1407 int error; 1408 1409 CURVNET_SET(so1->so_vnet); 1410 error = so1->so_proto->pr_connect2(so1, so2); 1411 CURVNET_RESTORE(); 1412 return (error); 1413 } 1414 1415 int 1416 sodisconnect(struct socket *so) 1417 { 1418 int error; 1419 1420 if ((so->so_state & SS_ISCONNECTED) == 0) 1421 return (ENOTCONN); 1422 if (so->so_state & SS_ISDISCONNECTING) 1423 return (EALREADY); 1424 VNET_SO_ASSERT(so); 1425 error = so->so_proto->pr_disconnect(so); 1426 return (error); 1427 } 1428 1429 int 1430 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1431 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1432 { 1433 long space; 1434 ssize_t resid; 1435 int clen = 0, error, dontroute; 1436 1437 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 1438 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 1439 ("sosend_dgram: !PR_ATOMIC")); 1440 1441 if (uio != NULL) 1442 resid = uio->uio_resid; 1443 else 1444 resid = top->m_pkthdr.len; 1445 /* 1446 * In theory resid should be unsigned. However, space must be 1447 * signed, as it might be less than 0 if we over-committed, and we 1448 * must use a signed comparison of space and resid. On the other 1449 * hand, a negative resid causes us to loop sending 0-length 1450 * segments to the protocol. 1451 */ 1452 if (resid < 0) { 1453 error = EINVAL; 1454 goto out; 1455 } 1456 1457 dontroute = 1458 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 1459 if (td != NULL) 1460 td->td_ru.ru_msgsnd++; 1461 if (control != NULL) 1462 clen = control->m_len; 1463 1464 SOCKBUF_LOCK(&so->so_snd); 1465 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1466 SOCKBUF_UNLOCK(&so->so_snd); 1467 error = EPIPE; 1468 goto out; 1469 } 1470 if (so->so_error) { 1471 error = so->so_error; 1472 so->so_error = 0; 1473 SOCKBUF_UNLOCK(&so->so_snd); 1474 goto out; 1475 } 1476 if ((so->so_state & SS_ISCONNECTED) == 0) { 1477 /* 1478 * `sendto' and `sendmsg' is allowed on a connection-based 1479 * socket if it supports implied connect. Return ENOTCONN if 1480 * not connected and no address is supplied. 1481 */ 1482 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1483 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1484 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1485 !(resid == 0 && clen != 0)) { 1486 SOCKBUF_UNLOCK(&so->so_snd); 1487 error = ENOTCONN; 1488 goto out; 1489 } 1490 } else if (addr == NULL) { 1491 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1492 error = ENOTCONN; 1493 else 1494 error = EDESTADDRREQ; 1495 SOCKBUF_UNLOCK(&so->so_snd); 1496 goto out; 1497 } 1498 } 1499 1500 /* 1501 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1502 * problem and need fixing. 1503 */ 1504 space = sbspace(&so->so_snd); 1505 if (flags & MSG_OOB) 1506 space += 1024; 1507 space -= clen; 1508 SOCKBUF_UNLOCK(&so->so_snd); 1509 if (resid > space) { 1510 error = EMSGSIZE; 1511 goto out; 1512 } 1513 if (uio == NULL) { 1514 resid = 0; 1515 if (flags & MSG_EOR) 1516 top->m_flags |= M_EOR; 1517 } else { 1518 /* 1519 * Copy the data from userland into a mbuf chain. 1520 * If no data is to be copied in, a single empty mbuf 1521 * is returned. 1522 */ 1523 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1524 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1525 if (top == NULL) { 1526 error = EFAULT; /* only possible error */ 1527 goto out; 1528 } 1529 space -= resid - uio->uio_resid; 1530 resid = uio->uio_resid; 1531 } 1532 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1533 /* 1534 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1535 * than with. 1536 */ 1537 if (dontroute) { 1538 SOCK_LOCK(so); 1539 so->so_options |= SO_DONTROUTE; 1540 SOCK_UNLOCK(so); 1541 } 1542 /* 1543 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1544 * of date. We could have received a reset packet in an interrupt or 1545 * maybe we slept while doing page faults in uiomove() etc. We could 1546 * probably recheck again inside the locking protection here, but 1547 * there are probably other places that this also happens. We must 1548 * rethink this. 1549 */ 1550 VNET_SO_ASSERT(so); 1551 error = so->so_proto->pr_send(so, (flags & MSG_OOB) ? PRUS_OOB : 1552 /* 1553 * If the user set MSG_EOF, the protocol understands this flag and 1554 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1555 */ 1556 ((flags & MSG_EOF) && 1557 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1558 (resid <= 0)) ? 1559 PRUS_EOF : 1560 /* If there is more to send set PRUS_MORETOCOME */ 1561 (flags & MSG_MORETOCOME) || 1562 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1563 top, addr, control, td); 1564 if (dontroute) { 1565 SOCK_LOCK(so); 1566 so->so_options &= ~SO_DONTROUTE; 1567 SOCK_UNLOCK(so); 1568 } 1569 clen = 0; 1570 control = NULL; 1571 top = NULL; 1572 out: 1573 if (top != NULL) 1574 m_freem(top); 1575 if (control != NULL) 1576 m_freem(control); 1577 return (error); 1578 } 1579 1580 /* 1581 * Send on a socket. If send must go all at once and message is larger than 1582 * send buffering, then hard error. Lock against other senders. If must go 1583 * all at once and not enough room now, then inform user that this would 1584 * block and do nothing. Otherwise, if nonblocking, send as much as 1585 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1586 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1587 * in mbuf chain must be small enough to send all at once. 1588 * 1589 * Returns nonzero on error, timeout or signal; callers must check for short 1590 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1591 * on return. 1592 */ 1593 int 1594 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1595 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1596 { 1597 long space; 1598 ssize_t resid; 1599 int clen = 0, error, dontroute; 1600 int atomic = sosendallatonce(so) || top; 1601 int pr_send_flag; 1602 #ifdef KERN_TLS 1603 struct ktls_session *tls; 1604 int tls_enq_cnt, tls_send_flag; 1605 uint8_t tls_rtype; 1606 1607 tls = NULL; 1608 tls_rtype = TLS_RLTYPE_APP; 1609 #endif 1610 if (uio != NULL) 1611 resid = uio->uio_resid; 1612 else if ((top->m_flags & M_PKTHDR) != 0) 1613 resid = top->m_pkthdr.len; 1614 else 1615 resid = m_length(top, NULL); 1616 /* 1617 * In theory resid should be unsigned. However, space must be 1618 * signed, as it might be less than 0 if we over-committed, and we 1619 * must use a signed comparison of space and resid. On the other 1620 * hand, a negative resid causes us to loop sending 0-length 1621 * segments to the protocol. 1622 * 1623 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1624 * type sockets since that's an error. 1625 */ 1626 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1627 error = EINVAL; 1628 goto out; 1629 } 1630 1631 dontroute = 1632 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1633 (so->so_proto->pr_flags & PR_ATOMIC); 1634 if (td != NULL) 1635 td->td_ru.ru_msgsnd++; 1636 if (control != NULL) 1637 clen = control->m_len; 1638 1639 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 1640 if (error) 1641 goto out; 1642 1643 #ifdef KERN_TLS 1644 tls_send_flag = 0; 1645 tls = ktls_hold(so->so_snd.sb_tls_info); 1646 if (tls != NULL) { 1647 if (tls->mode == TCP_TLS_MODE_SW) 1648 tls_send_flag = PRUS_NOTREADY; 1649 1650 if (control != NULL) { 1651 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1652 1653 if (clen >= sizeof(*cm) && 1654 cm->cmsg_type == TLS_SET_RECORD_TYPE) { 1655 tls_rtype = *((uint8_t *)CMSG_DATA(cm)); 1656 clen = 0; 1657 m_freem(control); 1658 control = NULL; 1659 atomic = 1; 1660 } 1661 } 1662 1663 if (resid == 0 && !ktls_permit_empty_frames(tls)) { 1664 error = EINVAL; 1665 goto release; 1666 } 1667 } 1668 #endif 1669 1670 restart: 1671 do { 1672 SOCKBUF_LOCK(&so->so_snd); 1673 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1674 SOCKBUF_UNLOCK(&so->so_snd); 1675 error = EPIPE; 1676 goto release; 1677 } 1678 if (so->so_error) { 1679 error = so->so_error; 1680 so->so_error = 0; 1681 SOCKBUF_UNLOCK(&so->so_snd); 1682 goto release; 1683 } 1684 if ((so->so_state & SS_ISCONNECTED) == 0) { 1685 /* 1686 * `sendto' and `sendmsg' is allowed on a connection- 1687 * based socket if it supports implied connect. 1688 * Return ENOTCONN if not connected and no address is 1689 * supplied. 1690 */ 1691 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1692 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1693 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1694 !(resid == 0 && clen != 0)) { 1695 SOCKBUF_UNLOCK(&so->so_snd); 1696 error = ENOTCONN; 1697 goto release; 1698 } 1699 } else if (addr == NULL) { 1700 SOCKBUF_UNLOCK(&so->so_snd); 1701 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1702 error = ENOTCONN; 1703 else 1704 error = EDESTADDRREQ; 1705 goto release; 1706 } 1707 } 1708 space = sbspace(&so->so_snd); 1709 if (flags & MSG_OOB) 1710 space += 1024; 1711 if ((atomic && resid > so->so_snd.sb_hiwat) || 1712 clen > so->so_snd.sb_hiwat) { 1713 SOCKBUF_UNLOCK(&so->so_snd); 1714 error = EMSGSIZE; 1715 goto release; 1716 } 1717 if (space < resid + clen && 1718 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1719 if ((so->so_state & SS_NBIO) || 1720 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) { 1721 SOCKBUF_UNLOCK(&so->so_snd); 1722 error = EWOULDBLOCK; 1723 goto release; 1724 } 1725 error = sbwait(so, SO_SND); 1726 SOCKBUF_UNLOCK(&so->so_snd); 1727 if (error) 1728 goto release; 1729 goto restart; 1730 } 1731 SOCKBUF_UNLOCK(&so->so_snd); 1732 space -= clen; 1733 do { 1734 if (uio == NULL) { 1735 resid = 0; 1736 if (flags & MSG_EOR) 1737 top->m_flags |= M_EOR; 1738 #ifdef KERN_TLS 1739 if (tls != NULL) { 1740 ktls_frame(top, tls, &tls_enq_cnt, 1741 tls_rtype); 1742 tls_rtype = TLS_RLTYPE_APP; 1743 } 1744 #endif 1745 } else { 1746 /* 1747 * Copy the data from userland into a mbuf 1748 * chain. If resid is 0, which can happen 1749 * only if we have control to send, then 1750 * a single empty mbuf is returned. This 1751 * is a workaround to prevent protocol send 1752 * methods to panic. 1753 */ 1754 #ifdef KERN_TLS 1755 if (tls != NULL) { 1756 top = m_uiotombuf(uio, M_WAITOK, space, 1757 tls->params.max_frame_len, 1758 M_EXTPG | 1759 ((flags & MSG_EOR) ? M_EOR : 0)); 1760 if (top != NULL) { 1761 ktls_frame(top, tls, 1762 &tls_enq_cnt, tls_rtype); 1763 } 1764 tls_rtype = TLS_RLTYPE_APP; 1765 } else 1766 #endif 1767 top = m_uiotombuf(uio, M_WAITOK, space, 1768 (atomic ? max_hdr : 0), 1769 (atomic ? M_PKTHDR : 0) | 1770 ((flags & MSG_EOR) ? M_EOR : 0)); 1771 if (top == NULL) { 1772 error = EFAULT; /* only possible error */ 1773 goto release; 1774 } 1775 space -= resid - uio->uio_resid; 1776 resid = uio->uio_resid; 1777 } 1778 if (dontroute) { 1779 SOCK_LOCK(so); 1780 so->so_options |= SO_DONTROUTE; 1781 SOCK_UNLOCK(so); 1782 } 1783 /* 1784 * XXX all the SBS_CANTSENDMORE checks previously 1785 * done could be out of date. We could have received 1786 * a reset packet in an interrupt or maybe we slept 1787 * while doing page faults in uiomove() etc. We 1788 * could probably recheck again inside the locking 1789 * protection here, but there are probably other 1790 * places that this also happens. We must rethink 1791 * this. 1792 */ 1793 VNET_SO_ASSERT(so); 1794 1795 pr_send_flag = (flags & MSG_OOB) ? PRUS_OOB : 1796 /* 1797 * If the user set MSG_EOF, the protocol understands 1798 * this flag and nothing left to send then use 1799 * PRU_SEND_EOF instead of PRU_SEND. 1800 */ 1801 ((flags & MSG_EOF) && 1802 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1803 (resid <= 0)) ? 1804 PRUS_EOF : 1805 /* If there is more to send set PRUS_MORETOCOME. */ 1806 (flags & MSG_MORETOCOME) || 1807 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; 1808 1809 #ifdef KERN_TLS 1810 pr_send_flag |= tls_send_flag; 1811 #endif 1812 1813 error = so->so_proto->pr_send(so, pr_send_flag, top, 1814 addr, control, td); 1815 1816 if (dontroute) { 1817 SOCK_LOCK(so); 1818 so->so_options &= ~SO_DONTROUTE; 1819 SOCK_UNLOCK(so); 1820 } 1821 1822 #ifdef KERN_TLS 1823 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) { 1824 if (error != 0) { 1825 m_freem(top); 1826 top = NULL; 1827 } else { 1828 soref(so); 1829 ktls_enqueue(top, so, tls_enq_cnt); 1830 } 1831 } 1832 #endif 1833 clen = 0; 1834 control = NULL; 1835 top = NULL; 1836 if (error) 1837 goto release; 1838 } while (resid && space > 0); 1839 } while (resid); 1840 1841 release: 1842 SOCK_IO_SEND_UNLOCK(so); 1843 out: 1844 #ifdef KERN_TLS 1845 if (tls != NULL) 1846 ktls_free(tls); 1847 #endif 1848 if (top != NULL) 1849 m_freem(top); 1850 if (control != NULL) 1851 m_freem(control); 1852 return (error); 1853 } 1854 1855 /* 1856 * Send to a socket from a kernel thread. 1857 * 1858 * XXXGL: in almost all cases uio is NULL and the mbuf is supplied. 1859 * Exception is nfs/bootp_subr.c. It is arguable that the VNET context needs 1860 * to be set at all. This function should just boil down to a static inline 1861 * calling the protocol method. 1862 */ 1863 int 1864 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1865 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1866 { 1867 int error; 1868 1869 CURVNET_SET(so->so_vnet); 1870 error = so->so_proto->pr_sosend(so, addr, uio, 1871 top, control, flags, td); 1872 CURVNET_RESTORE(); 1873 return (error); 1874 } 1875 1876 /* 1877 * send(2), write(2) or aio_write(2) on a socket. 1878 */ 1879 int 1880 sousrsend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1881 struct mbuf *control, int flags, struct proc *userproc) 1882 { 1883 struct thread *td; 1884 ssize_t len; 1885 int error; 1886 1887 td = uio->uio_td; 1888 len = uio->uio_resid; 1889 CURVNET_SET(so->so_vnet); 1890 error = so->so_proto->pr_sosend(so, addr, uio, NULL, control, flags, 1891 td); 1892 CURVNET_RESTORE(); 1893 if (error != 0) { 1894 /* 1895 * Clear transient errors for stream protocols if they made 1896 * some progress. Make exclusion for aio(4) that would 1897 * schedule a new write in case of EWOULDBLOCK and clear 1898 * error itself. See soaio_process_job(). 1899 */ 1900 if (uio->uio_resid != len && 1901 (so->so_proto->pr_flags & PR_ATOMIC) == 0 && 1902 userproc == NULL && 1903 (error == ERESTART || error == EINTR || 1904 error == EWOULDBLOCK)) 1905 error = 0; 1906 /* Generation of SIGPIPE can be controlled per socket. */ 1907 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0 && 1908 (flags & MSG_NOSIGNAL) == 0) { 1909 if (userproc != NULL) { 1910 /* aio(4) job */ 1911 PROC_LOCK(userproc); 1912 kern_psignal(userproc, SIGPIPE); 1913 PROC_UNLOCK(userproc); 1914 } else { 1915 PROC_LOCK(td->td_proc); 1916 tdsignal(td, SIGPIPE); 1917 PROC_UNLOCK(td->td_proc); 1918 } 1919 } 1920 } 1921 return (error); 1922 } 1923 1924 /* 1925 * The part of soreceive() that implements reading non-inline out-of-band 1926 * data from a socket. For more complete comments, see soreceive(), from 1927 * which this code originated. 1928 * 1929 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1930 * unable to return an mbuf chain to the caller. 1931 */ 1932 static int 1933 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1934 { 1935 struct protosw *pr = so->so_proto; 1936 struct mbuf *m; 1937 int error; 1938 1939 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1940 VNET_SO_ASSERT(so); 1941 1942 m = m_get(M_WAITOK, MT_DATA); 1943 error = pr->pr_rcvoob(so, m, flags & MSG_PEEK); 1944 if (error) 1945 goto bad; 1946 do { 1947 error = uiomove(mtod(m, void *), 1948 (int) min(uio->uio_resid, m->m_len), uio); 1949 m = m_free(m); 1950 } while (uio->uio_resid && error == 0 && m); 1951 bad: 1952 if (m != NULL) 1953 m_freem(m); 1954 return (error); 1955 } 1956 1957 /* 1958 * Following replacement or removal of the first mbuf on the first mbuf chain 1959 * of a socket buffer, push necessary state changes back into the socket 1960 * buffer so that other consumers see the values consistently. 'nextrecord' 1961 * is the callers locally stored value of the original value of 1962 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1963 * NOTE: 'nextrecord' may be NULL. 1964 */ 1965 static __inline void 1966 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1967 { 1968 1969 SOCKBUF_LOCK_ASSERT(sb); 1970 /* 1971 * First, update for the new value of nextrecord. If necessary, make 1972 * it the first record. 1973 */ 1974 if (sb->sb_mb != NULL) 1975 sb->sb_mb->m_nextpkt = nextrecord; 1976 else 1977 sb->sb_mb = nextrecord; 1978 1979 /* 1980 * Now update any dependent socket buffer fields to reflect the new 1981 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1982 * addition of a second clause that takes care of the case where 1983 * sb_mb has been updated, but remains the last record. 1984 */ 1985 if (sb->sb_mb == NULL) { 1986 sb->sb_mbtail = NULL; 1987 sb->sb_lastrecord = NULL; 1988 } else if (sb->sb_mb->m_nextpkt == NULL) 1989 sb->sb_lastrecord = sb->sb_mb; 1990 } 1991 1992 /* 1993 * Implement receive operations on a socket. We depend on the way that 1994 * records are added to the sockbuf by sbappend. In particular, each record 1995 * (mbufs linked through m_next) must begin with an address if the protocol 1996 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1997 * data, and then zero or more mbufs of data. In order to allow parallelism 1998 * between network receive and copying to user space, as well as avoid 1999 * sleeping with a mutex held, we release the socket buffer mutex during the 2000 * user space copy. Although the sockbuf is locked, new data may still be 2001 * appended, and thus we must maintain consistency of the sockbuf during that 2002 * time. 2003 * 2004 * The caller may receive the data as a single mbuf chain by supplying an 2005 * mbuf **mp0 for use in returning the chain. The uio is then used only for 2006 * the count in uio_resid. 2007 */ 2008 int 2009 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 2010 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2011 { 2012 struct mbuf *m, **mp; 2013 int flags, error, offset; 2014 ssize_t len; 2015 struct protosw *pr = so->so_proto; 2016 struct mbuf *nextrecord; 2017 int moff, type = 0; 2018 ssize_t orig_resid = uio->uio_resid; 2019 bool report_real_len = false; 2020 2021 mp = mp0; 2022 if (psa != NULL) 2023 *psa = NULL; 2024 if (controlp != NULL) 2025 *controlp = NULL; 2026 if (flagsp != NULL) { 2027 report_real_len = *flagsp & MSG_TRUNC; 2028 *flagsp &= ~MSG_TRUNC; 2029 flags = *flagsp &~ MSG_EOR; 2030 } else 2031 flags = 0; 2032 if (flags & MSG_OOB) 2033 return (soreceive_rcvoob(so, uio, flags)); 2034 if (mp != NULL) 2035 *mp = NULL; 2036 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 2037 && uio->uio_resid) { 2038 VNET_SO_ASSERT(so); 2039 pr->pr_rcvd(so, 0); 2040 } 2041 2042 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 2043 if (error) 2044 return (error); 2045 2046 restart: 2047 SOCKBUF_LOCK(&so->so_rcv); 2048 m = so->so_rcv.sb_mb; 2049 /* 2050 * If we have less data than requested, block awaiting more (subject 2051 * to any timeout) if: 2052 * 1. the current count is less than the low water mark, or 2053 * 2. MSG_DONTWAIT is not set 2054 */ 2055 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 2056 sbavail(&so->so_rcv) < uio->uio_resid) && 2057 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && 2058 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 2059 KASSERT(m != NULL || !sbavail(&so->so_rcv), 2060 ("receive: m == %p sbavail == %u", 2061 m, sbavail(&so->so_rcv))); 2062 if (so->so_error || so->so_rerror) { 2063 if (m != NULL) 2064 goto dontblock; 2065 if (so->so_error) 2066 error = so->so_error; 2067 else 2068 error = so->so_rerror; 2069 if ((flags & MSG_PEEK) == 0) { 2070 if (so->so_error) 2071 so->so_error = 0; 2072 else 2073 so->so_rerror = 0; 2074 } 2075 SOCKBUF_UNLOCK(&so->so_rcv); 2076 goto release; 2077 } 2078 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2079 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2080 if (m != NULL) 2081 goto dontblock; 2082 #ifdef KERN_TLS 2083 else if (so->so_rcv.sb_tlsdcc == 0 && 2084 so->so_rcv.sb_tlscc == 0) { 2085 #else 2086 else { 2087 #endif 2088 SOCKBUF_UNLOCK(&so->so_rcv); 2089 goto release; 2090 } 2091 } 2092 for (; m != NULL; m = m->m_next) 2093 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 2094 m = so->so_rcv.sb_mb; 2095 goto dontblock; 2096 } 2097 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED | 2098 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 && 2099 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 2100 SOCKBUF_UNLOCK(&so->so_rcv); 2101 error = ENOTCONN; 2102 goto release; 2103 } 2104 if (uio->uio_resid == 0 && !report_real_len) { 2105 SOCKBUF_UNLOCK(&so->so_rcv); 2106 goto release; 2107 } 2108 if ((so->so_state & SS_NBIO) || 2109 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2110 SOCKBUF_UNLOCK(&so->so_rcv); 2111 error = EWOULDBLOCK; 2112 goto release; 2113 } 2114 SBLASTRECORDCHK(&so->so_rcv); 2115 SBLASTMBUFCHK(&so->so_rcv); 2116 error = sbwait(so, SO_RCV); 2117 SOCKBUF_UNLOCK(&so->so_rcv); 2118 if (error) 2119 goto release; 2120 goto restart; 2121 } 2122 dontblock: 2123 /* 2124 * From this point onward, we maintain 'nextrecord' as a cache of the 2125 * pointer to the next record in the socket buffer. We must keep the 2126 * various socket buffer pointers and local stack versions of the 2127 * pointers in sync, pushing out modifications before dropping the 2128 * socket buffer mutex, and re-reading them when picking it up. 2129 * 2130 * Otherwise, we will race with the network stack appending new data 2131 * or records onto the socket buffer by using inconsistent/stale 2132 * versions of the field, possibly resulting in socket buffer 2133 * corruption. 2134 * 2135 * By holding the high-level sblock(), we prevent simultaneous 2136 * readers from pulling off the front of the socket buffer. 2137 */ 2138 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2139 if (uio->uio_td) 2140 uio->uio_td->td_ru.ru_msgrcv++; 2141 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 2142 SBLASTRECORDCHK(&so->so_rcv); 2143 SBLASTMBUFCHK(&so->so_rcv); 2144 nextrecord = m->m_nextpkt; 2145 if (pr->pr_flags & PR_ADDR) { 2146 KASSERT(m->m_type == MT_SONAME, 2147 ("m->m_type == %d", m->m_type)); 2148 orig_resid = 0; 2149 if (psa != NULL) 2150 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2151 M_NOWAIT); 2152 if (flags & MSG_PEEK) { 2153 m = m->m_next; 2154 } else { 2155 sbfree(&so->so_rcv, m); 2156 so->so_rcv.sb_mb = m_free(m); 2157 m = so->so_rcv.sb_mb; 2158 sockbuf_pushsync(&so->so_rcv, nextrecord); 2159 } 2160 } 2161 2162 /* 2163 * Process one or more MT_CONTROL mbufs present before any data mbufs 2164 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 2165 * just copy the data; if !MSG_PEEK, we call into the protocol to 2166 * perform externalization (or freeing if controlp == NULL). 2167 */ 2168 if (m != NULL && m->m_type == MT_CONTROL) { 2169 struct mbuf *cm = NULL, *cmn; 2170 struct mbuf **cme = &cm; 2171 #ifdef KERN_TLS 2172 struct cmsghdr *cmsg; 2173 struct tls_get_record tgr; 2174 2175 /* 2176 * For MSG_TLSAPPDATA, check for an alert record. 2177 * If found, return ENXIO without removing 2178 * it from the receive queue. This allows a subsequent 2179 * call without MSG_TLSAPPDATA to receive it. 2180 * Note that, for TLS, there should only be a single 2181 * control mbuf with the TLS_GET_RECORD message in it. 2182 */ 2183 if (flags & MSG_TLSAPPDATA) { 2184 cmsg = mtod(m, struct cmsghdr *); 2185 if (cmsg->cmsg_type == TLS_GET_RECORD && 2186 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) { 2187 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr)); 2188 if (__predict_false(tgr.tls_type == 2189 TLS_RLTYPE_ALERT)) { 2190 SOCKBUF_UNLOCK(&so->so_rcv); 2191 error = ENXIO; 2192 goto release; 2193 } 2194 } 2195 } 2196 #endif 2197 2198 do { 2199 if (flags & MSG_PEEK) { 2200 if (controlp != NULL) { 2201 *controlp = m_copym(m, 0, m->m_len, 2202 M_NOWAIT); 2203 controlp = &(*controlp)->m_next; 2204 } 2205 m = m->m_next; 2206 } else { 2207 sbfree(&so->so_rcv, m); 2208 so->so_rcv.sb_mb = m->m_next; 2209 m->m_next = NULL; 2210 *cme = m; 2211 cme = &(*cme)->m_next; 2212 m = so->so_rcv.sb_mb; 2213 } 2214 } while (m != NULL && m->m_type == MT_CONTROL); 2215 if ((flags & MSG_PEEK) == 0) 2216 sockbuf_pushsync(&so->so_rcv, nextrecord); 2217 while (cm != NULL) { 2218 cmn = cm->m_next; 2219 cm->m_next = NULL; 2220 if (pr->pr_domain->dom_externalize != NULL) { 2221 SOCKBUF_UNLOCK(&so->so_rcv); 2222 VNET_SO_ASSERT(so); 2223 error = (*pr->pr_domain->dom_externalize) 2224 (cm, controlp, flags); 2225 SOCKBUF_LOCK(&so->so_rcv); 2226 } else if (controlp != NULL) 2227 *controlp = cm; 2228 else 2229 m_freem(cm); 2230 if (controlp != NULL) { 2231 while (*controlp != NULL) 2232 controlp = &(*controlp)->m_next; 2233 } 2234 cm = cmn; 2235 } 2236 if (m != NULL) 2237 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 2238 else 2239 nextrecord = so->so_rcv.sb_mb; 2240 orig_resid = 0; 2241 } 2242 if (m != NULL) { 2243 if ((flags & MSG_PEEK) == 0) { 2244 KASSERT(m->m_nextpkt == nextrecord, 2245 ("soreceive: post-control, nextrecord !sync")); 2246 if (nextrecord == NULL) { 2247 KASSERT(so->so_rcv.sb_mb == m, 2248 ("soreceive: post-control, sb_mb!=m")); 2249 KASSERT(so->so_rcv.sb_lastrecord == m, 2250 ("soreceive: post-control, lastrecord!=m")); 2251 } 2252 } 2253 type = m->m_type; 2254 if (type == MT_OOBDATA) 2255 flags |= MSG_OOB; 2256 } else { 2257 if ((flags & MSG_PEEK) == 0) { 2258 KASSERT(so->so_rcv.sb_mb == nextrecord, 2259 ("soreceive: sb_mb != nextrecord")); 2260 if (so->so_rcv.sb_mb == NULL) { 2261 KASSERT(so->so_rcv.sb_lastrecord == NULL, 2262 ("soreceive: sb_lastercord != NULL")); 2263 } 2264 } 2265 } 2266 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2267 SBLASTRECORDCHK(&so->so_rcv); 2268 SBLASTMBUFCHK(&so->so_rcv); 2269 2270 /* 2271 * Now continue to read any data mbufs off of the head of the socket 2272 * buffer until the read request is satisfied. Note that 'type' is 2273 * used to store the type of any mbuf reads that have happened so far 2274 * such that soreceive() can stop reading if the type changes, which 2275 * causes soreceive() to return only one of regular data and inline 2276 * out-of-band data in a single socket receive operation. 2277 */ 2278 moff = 0; 2279 offset = 0; 2280 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 2281 && error == 0) { 2282 /* 2283 * If the type of mbuf has changed since the last mbuf 2284 * examined ('type'), end the receive operation. 2285 */ 2286 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2287 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 2288 if (type != m->m_type) 2289 break; 2290 } else if (type == MT_OOBDATA) 2291 break; 2292 else 2293 KASSERT(m->m_type == MT_DATA, 2294 ("m->m_type == %d", m->m_type)); 2295 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 2296 len = uio->uio_resid; 2297 if (so->so_oobmark && len > so->so_oobmark - offset) 2298 len = so->so_oobmark - offset; 2299 if (len > m->m_len - moff) 2300 len = m->m_len - moff; 2301 /* 2302 * If mp is set, just pass back the mbufs. Otherwise copy 2303 * them out via the uio, then free. Sockbuf must be 2304 * consistent here (points to current mbuf, it points to next 2305 * record) when we drop priority; we must note any additions 2306 * to the sockbuf when we block interrupts again. 2307 */ 2308 if (mp == NULL) { 2309 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2310 SBLASTRECORDCHK(&so->so_rcv); 2311 SBLASTMBUFCHK(&so->so_rcv); 2312 SOCKBUF_UNLOCK(&so->so_rcv); 2313 if ((m->m_flags & M_EXTPG) != 0) 2314 error = m_unmapped_uiomove(m, moff, uio, 2315 (int)len); 2316 else 2317 error = uiomove(mtod(m, char *) + moff, 2318 (int)len, uio); 2319 SOCKBUF_LOCK(&so->so_rcv); 2320 if (error) { 2321 /* 2322 * The MT_SONAME mbuf has already been removed 2323 * from the record, so it is necessary to 2324 * remove the data mbufs, if any, to preserve 2325 * the invariant in the case of PR_ADDR that 2326 * requires MT_SONAME mbufs at the head of 2327 * each record. 2328 */ 2329 if (pr->pr_flags & PR_ATOMIC && 2330 ((flags & MSG_PEEK) == 0)) 2331 (void)sbdroprecord_locked(&so->so_rcv); 2332 SOCKBUF_UNLOCK(&so->so_rcv); 2333 goto release; 2334 } 2335 } else 2336 uio->uio_resid -= len; 2337 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2338 if (len == m->m_len - moff) { 2339 if (m->m_flags & M_EOR) 2340 flags |= MSG_EOR; 2341 if (flags & MSG_PEEK) { 2342 m = m->m_next; 2343 moff = 0; 2344 } else { 2345 nextrecord = m->m_nextpkt; 2346 sbfree(&so->so_rcv, m); 2347 if (mp != NULL) { 2348 m->m_nextpkt = NULL; 2349 *mp = m; 2350 mp = &m->m_next; 2351 so->so_rcv.sb_mb = m = m->m_next; 2352 *mp = NULL; 2353 } else { 2354 so->so_rcv.sb_mb = m_free(m); 2355 m = so->so_rcv.sb_mb; 2356 } 2357 sockbuf_pushsync(&so->so_rcv, nextrecord); 2358 SBLASTRECORDCHK(&so->so_rcv); 2359 SBLASTMBUFCHK(&so->so_rcv); 2360 } 2361 } else { 2362 if (flags & MSG_PEEK) 2363 moff += len; 2364 else { 2365 if (mp != NULL) { 2366 if (flags & MSG_DONTWAIT) { 2367 *mp = m_copym(m, 0, len, 2368 M_NOWAIT); 2369 if (*mp == NULL) { 2370 /* 2371 * m_copym() couldn't 2372 * allocate an mbuf. 2373 * Adjust uio_resid back 2374 * (it was adjusted 2375 * down by len bytes, 2376 * which we didn't end 2377 * up "copying" over). 2378 */ 2379 uio->uio_resid += len; 2380 break; 2381 } 2382 } else { 2383 SOCKBUF_UNLOCK(&so->so_rcv); 2384 *mp = m_copym(m, 0, len, 2385 M_WAITOK); 2386 SOCKBUF_LOCK(&so->so_rcv); 2387 } 2388 } 2389 sbcut_locked(&so->so_rcv, len); 2390 } 2391 } 2392 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2393 if (so->so_oobmark) { 2394 if ((flags & MSG_PEEK) == 0) { 2395 so->so_oobmark -= len; 2396 if (so->so_oobmark == 0) { 2397 so->so_rcv.sb_state |= SBS_RCVATMARK; 2398 break; 2399 } 2400 } else { 2401 offset += len; 2402 if (offset == so->so_oobmark) 2403 break; 2404 } 2405 } 2406 if (flags & MSG_EOR) 2407 break; 2408 /* 2409 * If the MSG_WAITALL flag is set (for non-atomic socket), we 2410 * must not quit until "uio->uio_resid == 0" or an error 2411 * termination. If a signal/timeout occurs, return with a 2412 * short count but without error. Keep sockbuf locked 2413 * against other readers. 2414 */ 2415 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 2416 !sosendallatonce(so) && nextrecord == NULL) { 2417 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2418 if (so->so_error || so->so_rerror || 2419 so->so_rcv.sb_state & SBS_CANTRCVMORE) 2420 break; 2421 /* 2422 * Notify the protocol that some data has been 2423 * drained before blocking. 2424 */ 2425 if (pr->pr_flags & PR_WANTRCVD) { 2426 SOCKBUF_UNLOCK(&so->so_rcv); 2427 VNET_SO_ASSERT(so); 2428 pr->pr_rcvd(so, flags); 2429 SOCKBUF_LOCK(&so->so_rcv); 2430 if (__predict_false(so->so_rcv.sb_mb == NULL && 2431 (so->so_error || so->so_rerror || 2432 so->so_rcv.sb_state & SBS_CANTRCVMORE))) 2433 break; 2434 } 2435 SBLASTRECORDCHK(&so->so_rcv); 2436 SBLASTMBUFCHK(&so->so_rcv); 2437 /* 2438 * We could receive some data while was notifying 2439 * the protocol. Skip blocking in this case. 2440 */ 2441 if (so->so_rcv.sb_mb == NULL) { 2442 error = sbwait(so, SO_RCV); 2443 if (error) { 2444 SOCKBUF_UNLOCK(&so->so_rcv); 2445 goto release; 2446 } 2447 } 2448 m = so->so_rcv.sb_mb; 2449 if (m != NULL) 2450 nextrecord = m->m_nextpkt; 2451 } 2452 } 2453 2454 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2455 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 2456 if (report_real_len) 2457 uio->uio_resid -= m_length(m, NULL) - moff; 2458 flags |= MSG_TRUNC; 2459 if ((flags & MSG_PEEK) == 0) 2460 (void) sbdroprecord_locked(&so->so_rcv); 2461 } 2462 if ((flags & MSG_PEEK) == 0) { 2463 if (m == NULL) { 2464 /* 2465 * First part is an inline SB_EMPTY_FIXUP(). Second 2466 * part makes sure sb_lastrecord is up-to-date if 2467 * there is still data in the socket buffer. 2468 */ 2469 so->so_rcv.sb_mb = nextrecord; 2470 if (so->so_rcv.sb_mb == NULL) { 2471 so->so_rcv.sb_mbtail = NULL; 2472 so->so_rcv.sb_lastrecord = NULL; 2473 } else if (nextrecord->m_nextpkt == NULL) 2474 so->so_rcv.sb_lastrecord = nextrecord; 2475 } 2476 SBLASTRECORDCHK(&so->so_rcv); 2477 SBLASTMBUFCHK(&so->so_rcv); 2478 /* 2479 * If soreceive() is being done from the socket callback, 2480 * then don't need to generate ACK to peer to update window, 2481 * since ACK will be generated on return to TCP. 2482 */ 2483 if (!(flags & MSG_SOCALLBCK) && 2484 (pr->pr_flags & PR_WANTRCVD)) { 2485 SOCKBUF_UNLOCK(&so->so_rcv); 2486 VNET_SO_ASSERT(so); 2487 pr->pr_rcvd(so, flags); 2488 SOCKBUF_LOCK(&so->so_rcv); 2489 } 2490 } 2491 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2492 if (orig_resid == uio->uio_resid && orig_resid && 2493 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 2494 SOCKBUF_UNLOCK(&so->so_rcv); 2495 goto restart; 2496 } 2497 SOCKBUF_UNLOCK(&so->so_rcv); 2498 2499 if (flagsp != NULL) 2500 *flagsp |= flags; 2501 release: 2502 SOCK_IO_RECV_UNLOCK(so); 2503 return (error); 2504 } 2505 2506 /* 2507 * Optimized version of soreceive() for stream (TCP) sockets. 2508 */ 2509 int 2510 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 2511 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2512 { 2513 int len = 0, error = 0, flags, oresid; 2514 struct sockbuf *sb; 2515 struct mbuf *m, *n = NULL; 2516 2517 /* We only do stream sockets. */ 2518 if (so->so_type != SOCK_STREAM) 2519 return (EINVAL); 2520 if (psa != NULL) 2521 *psa = NULL; 2522 if (flagsp != NULL) 2523 flags = *flagsp &~ MSG_EOR; 2524 else 2525 flags = 0; 2526 if (controlp != NULL) 2527 *controlp = NULL; 2528 if (flags & MSG_OOB) 2529 return (soreceive_rcvoob(so, uio, flags)); 2530 if (mp0 != NULL) 2531 *mp0 = NULL; 2532 2533 sb = &so->so_rcv; 2534 2535 #ifdef KERN_TLS 2536 /* 2537 * KTLS store TLS records as records with a control message to 2538 * describe the framing. 2539 * 2540 * We check once here before acquiring locks to optimize the 2541 * common case. 2542 */ 2543 if (sb->sb_tls_info != NULL) 2544 return (soreceive_generic(so, psa, uio, mp0, controlp, 2545 flagsp)); 2546 #endif 2547 2548 /* Prevent other readers from entering the socket. */ 2549 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 2550 if (error) 2551 return (error); 2552 SOCKBUF_LOCK(sb); 2553 2554 #ifdef KERN_TLS 2555 if (sb->sb_tls_info != NULL) { 2556 SOCKBUF_UNLOCK(sb); 2557 SOCK_IO_RECV_UNLOCK(so); 2558 return (soreceive_generic(so, psa, uio, mp0, controlp, 2559 flagsp)); 2560 } 2561 #endif 2562 2563 /* Easy one, no space to copyout anything. */ 2564 if (uio->uio_resid == 0) { 2565 error = EINVAL; 2566 goto out; 2567 } 2568 oresid = uio->uio_resid; 2569 2570 /* We will never ever get anything unless we are or were connected. */ 2571 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 2572 error = ENOTCONN; 2573 goto out; 2574 } 2575 2576 restart: 2577 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2578 2579 /* Abort if socket has reported problems. */ 2580 if (so->so_error) { 2581 if (sbavail(sb) > 0) 2582 goto deliver; 2583 if (oresid > uio->uio_resid) 2584 goto out; 2585 error = so->so_error; 2586 if (!(flags & MSG_PEEK)) 2587 so->so_error = 0; 2588 goto out; 2589 } 2590 2591 /* Door is closed. Deliver what is left, if any. */ 2592 if (sb->sb_state & SBS_CANTRCVMORE) { 2593 if (sbavail(sb) > 0) 2594 goto deliver; 2595 else 2596 goto out; 2597 } 2598 2599 /* Socket buffer is empty and we shall not block. */ 2600 if (sbavail(sb) == 0 && 2601 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2602 error = EAGAIN; 2603 goto out; 2604 } 2605 2606 /* Socket buffer got some data that we shall deliver now. */ 2607 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 2608 ((so->so_state & SS_NBIO) || 2609 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2610 sbavail(sb) >= sb->sb_lowat || 2611 sbavail(sb) >= uio->uio_resid || 2612 sbavail(sb) >= sb->sb_hiwat) ) { 2613 goto deliver; 2614 } 2615 2616 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2617 if ((flags & MSG_WAITALL) && 2618 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 2619 goto deliver; 2620 2621 /* 2622 * Wait and block until (more) data comes in. 2623 * NB: Drops the sockbuf lock during wait. 2624 */ 2625 error = sbwait(so, SO_RCV); 2626 if (error) 2627 goto out; 2628 goto restart; 2629 2630 deliver: 2631 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2632 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 2633 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2634 2635 /* Statistics. */ 2636 if (uio->uio_td) 2637 uio->uio_td->td_ru.ru_msgrcv++; 2638 2639 /* Fill uio until full or current end of socket buffer is reached. */ 2640 len = min(uio->uio_resid, sbavail(sb)); 2641 if (mp0 != NULL) { 2642 /* Dequeue as many mbufs as possible. */ 2643 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2644 if (*mp0 == NULL) 2645 *mp0 = sb->sb_mb; 2646 else 2647 m_cat(*mp0, sb->sb_mb); 2648 for (m = sb->sb_mb; 2649 m != NULL && m->m_len <= len; 2650 m = m->m_next) { 2651 KASSERT(!(m->m_flags & M_NOTAVAIL), 2652 ("%s: m %p not available", __func__, m)); 2653 len -= m->m_len; 2654 uio->uio_resid -= m->m_len; 2655 sbfree(sb, m); 2656 n = m; 2657 } 2658 n->m_next = NULL; 2659 sb->sb_mb = m; 2660 sb->sb_lastrecord = sb->sb_mb; 2661 if (sb->sb_mb == NULL) 2662 SB_EMPTY_FIXUP(sb); 2663 } 2664 /* Copy the remainder. */ 2665 if (len > 0) { 2666 KASSERT(sb->sb_mb != NULL, 2667 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2668 2669 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2670 if (m == NULL) 2671 len = 0; /* Don't flush data from sockbuf. */ 2672 else 2673 uio->uio_resid -= len; 2674 if (*mp0 != NULL) 2675 m_cat(*mp0, m); 2676 else 2677 *mp0 = m; 2678 if (*mp0 == NULL) { 2679 error = ENOBUFS; 2680 goto out; 2681 } 2682 } 2683 } else { 2684 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2685 SOCKBUF_UNLOCK(sb); 2686 error = m_mbuftouio(uio, sb->sb_mb, len); 2687 SOCKBUF_LOCK(sb); 2688 if (error) 2689 goto out; 2690 } 2691 SBLASTRECORDCHK(sb); 2692 SBLASTMBUFCHK(sb); 2693 2694 /* 2695 * Remove the delivered data from the socket buffer unless we 2696 * were only peeking. 2697 */ 2698 if (!(flags & MSG_PEEK)) { 2699 if (len > 0) 2700 sbdrop_locked(sb, len); 2701 2702 /* Notify protocol that we drained some data. */ 2703 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2704 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2705 !(flags & MSG_SOCALLBCK))) { 2706 SOCKBUF_UNLOCK(sb); 2707 VNET_SO_ASSERT(so); 2708 so->so_proto->pr_rcvd(so, flags); 2709 SOCKBUF_LOCK(sb); 2710 } 2711 } 2712 2713 /* 2714 * For MSG_WAITALL we may have to loop again and wait for 2715 * more data to come in. 2716 */ 2717 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2718 goto restart; 2719 out: 2720 SBLASTRECORDCHK(sb); 2721 SBLASTMBUFCHK(sb); 2722 SOCKBUF_UNLOCK(sb); 2723 SOCK_IO_RECV_UNLOCK(so); 2724 return (error); 2725 } 2726 2727 /* 2728 * Optimized version of soreceive() for simple datagram cases from userspace. 2729 * Unlike in the stream case, we're able to drop a datagram if copyout() 2730 * fails, and because we handle datagrams atomically, we don't need to use a 2731 * sleep lock to prevent I/O interlacing. 2732 */ 2733 int 2734 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2735 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2736 { 2737 struct mbuf *m, *m2; 2738 int flags, error; 2739 ssize_t len; 2740 struct protosw *pr = so->so_proto; 2741 struct mbuf *nextrecord; 2742 2743 if (psa != NULL) 2744 *psa = NULL; 2745 if (controlp != NULL) 2746 *controlp = NULL; 2747 if (flagsp != NULL) 2748 flags = *flagsp &~ MSG_EOR; 2749 else 2750 flags = 0; 2751 2752 /* 2753 * For any complicated cases, fall back to the full 2754 * soreceive_generic(). 2755 */ 2756 if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC))) 2757 return (soreceive_generic(so, psa, uio, mp0, controlp, 2758 flagsp)); 2759 2760 /* 2761 * Enforce restrictions on use. 2762 */ 2763 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2764 ("soreceive_dgram: wantrcvd")); 2765 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2766 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2767 ("soreceive_dgram: SBS_RCVATMARK")); 2768 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2769 ("soreceive_dgram: P_CONNREQUIRED")); 2770 2771 /* 2772 * Loop blocking while waiting for a datagram. 2773 */ 2774 SOCKBUF_LOCK(&so->so_rcv); 2775 while ((m = so->so_rcv.sb_mb) == NULL) { 2776 KASSERT(sbavail(&so->so_rcv) == 0, 2777 ("soreceive_dgram: sb_mb NULL but sbavail %u", 2778 sbavail(&so->so_rcv))); 2779 if (so->so_error) { 2780 error = so->so_error; 2781 so->so_error = 0; 2782 SOCKBUF_UNLOCK(&so->so_rcv); 2783 return (error); 2784 } 2785 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2786 uio->uio_resid == 0) { 2787 SOCKBUF_UNLOCK(&so->so_rcv); 2788 return (0); 2789 } 2790 if ((so->so_state & SS_NBIO) || 2791 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2792 SOCKBUF_UNLOCK(&so->so_rcv); 2793 return (EWOULDBLOCK); 2794 } 2795 SBLASTRECORDCHK(&so->so_rcv); 2796 SBLASTMBUFCHK(&so->so_rcv); 2797 error = sbwait(so, SO_RCV); 2798 if (error) { 2799 SOCKBUF_UNLOCK(&so->so_rcv); 2800 return (error); 2801 } 2802 } 2803 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2804 2805 if (uio->uio_td) 2806 uio->uio_td->td_ru.ru_msgrcv++; 2807 SBLASTRECORDCHK(&so->so_rcv); 2808 SBLASTMBUFCHK(&so->so_rcv); 2809 nextrecord = m->m_nextpkt; 2810 if (nextrecord == NULL) { 2811 KASSERT(so->so_rcv.sb_lastrecord == m, 2812 ("soreceive_dgram: lastrecord != m")); 2813 } 2814 2815 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2816 ("soreceive_dgram: m_nextpkt != nextrecord")); 2817 2818 /* 2819 * Pull 'm' and its chain off the front of the packet queue. 2820 */ 2821 so->so_rcv.sb_mb = NULL; 2822 sockbuf_pushsync(&so->so_rcv, nextrecord); 2823 2824 /* 2825 * Walk 'm's chain and free that many bytes from the socket buffer. 2826 */ 2827 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2828 sbfree(&so->so_rcv, m2); 2829 2830 /* 2831 * Do a few last checks before we let go of the lock. 2832 */ 2833 SBLASTRECORDCHK(&so->so_rcv); 2834 SBLASTMBUFCHK(&so->so_rcv); 2835 SOCKBUF_UNLOCK(&so->so_rcv); 2836 2837 if (pr->pr_flags & PR_ADDR) { 2838 KASSERT(m->m_type == MT_SONAME, 2839 ("m->m_type == %d", m->m_type)); 2840 if (psa != NULL) 2841 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2842 M_NOWAIT); 2843 m = m_free(m); 2844 } 2845 if (m == NULL) { 2846 /* XXXRW: Can this happen? */ 2847 return (0); 2848 } 2849 2850 /* 2851 * Packet to copyout() is now in 'm' and it is disconnected from the 2852 * queue. 2853 * 2854 * Process one or more MT_CONTROL mbufs present before any data mbufs 2855 * in the first mbuf chain on the socket buffer. We call into the 2856 * protocol to perform externalization (or freeing if controlp == 2857 * NULL). In some cases there can be only MT_CONTROL mbufs without 2858 * MT_DATA mbufs. 2859 */ 2860 if (m->m_type == MT_CONTROL) { 2861 struct mbuf *cm = NULL, *cmn; 2862 struct mbuf **cme = &cm; 2863 2864 do { 2865 m2 = m->m_next; 2866 m->m_next = NULL; 2867 *cme = m; 2868 cme = &(*cme)->m_next; 2869 m = m2; 2870 } while (m != NULL && m->m_type == MT_CONTROL); 2871 while (cm != NULL) { 2872 cmn = cm->m_next; 2873 cm->m_next = NULL; 2874 if (pr->pr_domain->dom_externalize != NULL) { 2875 error = (*pr->pr_domain->dom_externalize) 2876 (cm, controlp, flags); 2877 } else if (controlp != NULL) 2878 *controlp = cm; 2879 else 2880 m_freem(cm); 2881 if (controlp != NULL) { 2882 while (*controlp != NULL) 2883 controlp = &(*controlp)->m_next; 2884 } 2885 cm = cmn; 2886 } 2887 } 2888 KASSERT(m == NULL || m->m_type == MT_DATA, 2889 ("soreceive_dgram: !data")); 2890 while (m != NULL && uio->uio_resid > 0) { 2891 len = uio->uio_resid; 2892 if (len > m->m_len) 2893 len = m->m_len; 2894 error = uiomove(mtod(m, char *), (int)len, uio); 2895 if (error) { 2896 m_freem(m); 2897 return (error); 2898 } 2899 if (len == m->m_len) 2900 m = m_free(m); 2901 else { 2902 m->m_data += len; 2903 m->m_len -= len; 2904 } 2905 } 2906 if (m != NULL) { 2907 flags |= MSG_TRUNC; 2908 m_freem(m); 2909 } 2910 if (flagsp != NULL) 2911 *flagsp |= flags; 2912 return (0); 2913 } 2914 2915 int 2916 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2917 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2918 { 2919 int error; 2920 2921 CURVNET_SET(so->so_vnet); 2922 error = so->so_proto->pr_soreceive(so, psa, uio, mp0, controlp, flagsp); 2923 CURVNET_RESTORE(); 2924 return (error); 2925 } 2926 2927 int 2928 soshutdown(struct socket *so, int how) 2929 { 2930 struct protosw *pr; 2931 int error, soerror_enotconn; 2932 2933 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2934 return (EINVAL); 2935 2936 soerror_enotconn = 0; 2937 SOCK_LOCK(so); 2938 if ((so->so_state & 2939 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 2940 /* 2941 * POSIX mandates us to return ENOTCONN when shutdown(2) is 2942 * invoked on a datagram sockets, however historically we would 2943 * actually tear socket down. This is known to be leveraged by 2944 * some applications to unblock process waiting in recvXXX(2) 2945 * by other process that it shares that socket with. Try to meet 2946 * both backward-compatibility and POSIX requirements by forcing 2947 * ENOTCONN but still asking protocol to perform pru_shutdown(). 2948 */ 2949 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) { 2950 SOCK_UNLOCK(so); 2951 return (ENOTCONN); 2952 } 2953 soerror_enotconn = 1; 2954 } 2955 2956 if (SOLISTENING(so)) { 2957 if (how != SHUT_WR) { 2958 so->so_error = ECONNABORTED; 2959 solisten_wakeup(so); /* unlocks so */ 2960 } else { 2961 SOCK_UNLOCK(so); 2962 } 2963 goto done; 2964 } 2965 SOCK_UNLOCK(so); 2966 2967 CURVNET_SET(so->so_vnet); 2968 pr = so->so_proto; 2969 if (pr->pr_flush != NULL) 2970 pr->pr_flush(so, how); 2971 if (how != SHUT_WR) 2972 sorflush(so); 2973 if (how != SHUT_RD) { 2974 error = pr->pr_shutdown(so); 2975 wakeup(&so->so_timeo); 2976 CURVNET_RESTORE(); 2977 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error); 2978 } 2979 wakeup(&so->so_timeo); 2980 CURVNET_RESTORE(); 2981 2982 done: 2983 return (soerror_enotconn ? ENOTCONN : 0); 2984 } 2985 2986 void 2987 sorflush(struct socket *so) 2988 { 2989 struct protosw *pr; 2990 int error; 2991 2992 VNET_SO_ASSERT(so); 2993 2994 /* 2995 * Dislodge threads currently blocked in receive and wait to acquire 2996 * a lock against other simultaneous readers before clearing the 2997 * socket buffer. Don't let our acquire be interrupted by a signal 2998 * despite any existing socket disposition on interruptable waiting. 2999 */ 3000 socantrcvmore(so); 3001 3002 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR); 3003 if (error != 0) { 3004 KASSERT(SOLISTENING(so), 3005 ("%s: soiolock(%p) failed", __func__, so)); 3006 return; 3007 } 3008 3009 pr = so->so_proto; 3010 if (pr->pr_flags & PR_RIGHTS) { 3011 MPASS(pr->pr_domain->dom_dispose != NULL); 3012 (*pr->pr_domain->dom_dispose)(so); 3013 } else { 3014 sbrelease(so, SO_RCV); 3015 SOCK_IO_RECV_UNLOCK(so); 3016 } 3017 3018 } 3019 3020 /* 3021 * Wrapper for Socket established helper hook. 3022 * Parameters: socket, context of the hook point, hook id. 3023 */ 3024 static int inline 3025 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 3026 { 3027 struct socket_hhook_data hhook_data = { 3028 .so = so, 3029 .hctx = hctx, 3030 .m = NULL, 3031 .status = 0 3032 }; 3033 3034 CURVNET_SET(so->so_vnet); 3035 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 3036 CURVNET_RESTORE(); 3037 3038 /* Ugly but needed, since hhooks return void for now */ 3039 return (hhook_data.status); 3040 } 3041 3042 /* 3043 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 3044 * additional variant to handle the case where the option value needs to be 3045 * some kind of integer, but not a specific size. In addition to their use 3046 * here, these functions are also called by the protocol-level pr_ctloutput() 3047 * routines. 3048 */ 3049 int 3050 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 3051 { 3052 size_t valsize; 3053 3054 /* 3055 * If the user gives us more than we wanted, we ignore it, but if we 3056 * don't get the minimum length the caller wants, we return EINVAL. 3057 * On success, sopt->sopt_valsize is set to however much we actually 3058 * retrieved. 3059 */ 3060 if ((valsize = sopt->sopt_valsize) < minlen) 3061 return EINVAL; 3062 if (valsize > len) 3063 sopt->sopt_valsize = valsize = len; 3064 3065 if (sopt->sopt_td != NULL) 3066 return (copyin(sopt->sopt_val, buf, valsize)); 3067 3068 bcopy(sopt->sopt_val, buf, valsize); 3069 return (0); 3070 } 3071 3072 /* 3073 * Kernel version of setsockopt(2). 3074 * 3075 * XXX: optlen is size_t, not socklen_t 3076 */ 3077 int 3078 so_setsockopt(struct socket *so, int level, int optname, void *optval, 3079 size_t optlen) 3080 { 3081 struct sockopt sopt; 3082 3083 sopt.sopt_level = level; 3084 sopt.sopt_name = optname; 3085 sopt.sopt_dir = SOPT_SET; 3086 sopt.sopt_val = optval; 3087 sopt.sopt_valsize = optlen; 3088 sopt.sopt_td = NULL; 3089 return (sosetopt(so, &sopt)); 3090 } 3091 3092 int 3093 sosetopt(struct socket *so, struct sockopt *sopt) 3094 { 3095 int error, optval; 3096 struct linger l; 3097 struct timeval tv; 3098 sbintime_t val, *valp; 3099 uint32_t val32; 3100 #ifdef MAC 3101 struct mac extmac; 3102 #endif 3103 3104 CURVNET_SET(so->so_vnet); 3105 error = 0; 3106 if (sopt->sopt_level != SOL_SOCKET) { 3107 if (so->so_proto->pr_ctloutput != NULL) 3108 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3109 else 3110 error = ENOPROTOOPT; 3111 } else { 3112 switch (sopt->sopt_name) { 3113 case SO_ACCEPTFILTER: 3114 error = accept_filt_setopt(so, sopt); 3115 if (error) 3116 goto bad; 3117 break; 3118 3119 case SO_LINGER: 3120 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 3121 if (error) 3122 goto bad; 3123 if (l.l_linger < 0 || 3124 l.l_linger > USHRT_MAX || 3125 l.l_linger > (INT_MAX / hz)) { 3126 error = EDOM; 3127 goto bad; 3128 } 3129 SOCK_LOCK(so); 3130 so->so_linger = l.l_linger; 3131 if (l.l_onoff) 3132 so->so_options |= SO_LINGER; 3133 else 3134 so->so_options &= ~SO_LINGER; 3135 SOCK_UNLOCK(so); 3136 break; 3137 3138 case SO_DEBUG: 3139 case SO_KEEPALIVE: 3140 case SO_DONTROUTE: 3141 case SO_USELOOPBACK: 3142 case SO_BROADCAST: 3143 case SO_REUSEADDR: 3144 case SO_REUSEPORT: 3145 case SO_REUSEPORT_LB: 3146 case SO_OOBINLINE: 3147 case SO_TIMESTAMP: 3148 case SO_BINTIME: 3149 case SO_NOSIGPIPE: 3150 case SO_NO_DDP: 3151 case SO_NO_OFFLOAD: 3152 case SO_RERROR: 3153 error = sooptcopyin(sopt, &optval, sizeof optval, 3154 sizeof optval); 3155 if (error) 3156 goto bad; 3157 SOCK_LOCK(so); 3158 if (optval) 3159 so->so_options |= sopt->sopt_name; 3160 else 3161 so->so_options &= ~sopt->sopt_name; 3162 SOCK_UNLOCK(so); 3163 break; 3164 3165 case SO_SETFIB: 3166 error = sooptcopyin(sopt, &optval, sizeof optval, 3167 sizeof optval); 3168 if (error) 3169 goto bad; 3170 3171 if (optval < 0 || optval >= rt_numfibs) { 3172 error = EINVAL; 3173 goto bad; 3174 } 3175 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 3176 (so->so_proto->pr_domain->dom_family == PF_INET6) || 3177 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 3178 so->so_fibnum = optval; 3179 else 3180 so->so_fibnum = 0; 3181 break; 3182 3183 case SO_USER_COOKIE: 3184 error = sooptcopyin(sopt, &val32, sizeof val32, 3185 sizeof val32); 3186 if (error) 3187 goto bad; 3188 so->so_user_cookie = val32; 3189 break; 3190 3191 case SO_SNDBUF: 3192 case SO_RCVBUF: 3193 case SO_SNDLOWAT: 3194 case SO_RCVLOWAT: 3195 error = so->so_proto->pr_setsbopt(so, sopt); 3196 if (error) 3197 goto bad; 3198 break; 3199 3200 case SO_SNDTIMEO: 3201 case SO_RCVTIMEO: 3202 #ifdef COMPAT_FREEBSD32 3203 if (SV_CURPROC_FLAG(SV_ILP32)) { 3204 struct timeval32 tv32; 3205 3206 error = sooptcopyin(sopt, &tv32, sizeof tv32, 3207 sizeof tv32); 3208 CP(tv32, tv, tv_sec); 3209 CP(tv32, tv, tv_usec); 3210 } else 3211 #endif 3212 error = sooptcopyin(sopt, &tv, sizeof tv, 3213 sizeof tv); 3214 if (error) 3215 goto bad; 3216 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 3217 tv.tv_usec >= 1000000) { 3218 error = EDOM; 3219 goto bad; 3220 } 3221 if (tv.tv_sec > INT32_MAX) 3222 val = SBT_MAX; 3223 else 3224 val = tvtosbt(tv); 3225 SOCK_LOCK(so); 3226 valp = sopt->sopt_name == SO_SNDTIMEO ? 3227 (SOLISTENING(so) ? &so->sol_sbsnd_timeo : 3228 &so->so_snd.sb_timeo) : 3229 (SOLISTENING(so) ? &so->sol_sbrcv_timeo : 3230 &so->so_rcv.sb_timeo); 3231 *valp = val; 3232 SOCK_UNLOCK(so); 3233 break; 3234 3235 case SO_LABEL: 3236 #ifdef MAC 3237 error = sooptcopyin(sopt, &extmac, sizeof extmac, 3238 sizeof extmac); 3239 if (error) 3240 goto bad; 3241 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 3242 so, &extmac); 3243 #else 3244 error = EOPNOTSUPP; 3245 #endif 3246 break; 3247 3248 case SO_TS_CLOCK: 3249 error = sooptcopyin(sopt, &optval, sizeof optval, 3250 sizeof optval); 3251 if (error) 3252 goto bad; 3253 if (optval < 0 || optval > SO_TS_CLOCK_MAX) { 3254 error = EINVAL; 3255 goto bad; 3256 } 3257 so->so_ts_clock = optval; 3258 break; 3259 3260 case SO_MAX_PACING_RATE: 3261 error = sooptcopyin(sopt, &val32, sizeof(val32), 3262 sizeof(val32)); 3263 if (error) 3264 goto bad; 3265 so->so_max_pacing_rate = val32; 3266 break; 3267 3268 default: 3269 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3270 error = hhook_run_socket(so, sopt, 3271 HHOOK_SOCKET_OPT); 3272 else 3273 error = ENOPROTOOPT; 3274 break; 3275 } 3276 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 3277 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 3278 } 3279 bad: 3280 CURVNET_RESTORE(); 3281 return (error); 3282 } 3283 3284 /* 3285 * Helper routine for getsockopt. 3286 */ 3287 int 3288 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 3289 { 3290 int error; 3291 size_t valsize; 3292 3293 error = 0; 3294 3295 /* 3296 * Documented get behavior is that we always return a value, possibly 3297 * truncated to fit in the user's buffer. Traditional behavior is 3298 * that we always tell the user precisely how much we copied, rather 3299 * than something useful like the total amount we had available for 3300 * her. Note that this interface is not idempotent; the entire 3301 * answer must be generated ahead of time. 3302 */ 3303 valsize = min(len, sopt->sopt_valsize); 3304 sopt->sopt_valsize = valsize; 3305 if (sopt->sopt_val != NULL) { 3306 if (sopt->sopt_td != NULL) 3307 error = copyout(buf, sopt->sopt_val, valsize); 3308 else 3309 bcopy(buf, sopt->sopt_val, valsize); 3310 } 3311 return (error); 3312 } 3313 3314 int 3315 sogetopt(struct socket *so, struct sockopt *sopt) 3316 { 3317 int error, optval; 3318 struct linger l; 3319 struct timeval tv; 3320 #ifdef MAC 3321 struct mac extmac; 3322 #endif 3323 3324 CURVNET_SET(so->so_vnet); 3325 error = 0; 3326 if (sopt->sopt_level != SOL_SOCKET) { 3327 if (so->so_proto->pr_ctloutput != NULL) 3328 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3329 else 3330 error = ENOPROTOOPT; 3331 CURVNET_RESTORE(); 3332 return (error); 3333 } else { 3334 switch (sopt->sopt_name) { 3335 case SO_ACCEPTFILTER: 3336 error = accept_filt_getopt(so, sopt); 3337 break; 3338 3339 case SO_LINGER: 3340 SOCK_LOCK(so); 3341 l.l_onoff = so->so_options & SO_LINGER; 3342 l.l_linger = so->so_linger; 3343 SOCK_UNLOCK(so); 3344 error = sooptcopyout(sopt, &l, sizeof l); 3345 break; 3346 3347 case SO_USELOOPBACK: 3348 case SO_DONTROUTE: 3349 case SO_DEBUG: 3350 case SO_KEEPALIVE: 3351 case SO_REUSEADDR: 3352 case SO_REUSEPORT: 3353 case SO_REUSEPORT_LB: 3354 case SO_BROADCAST: 3355 case SO_OOBINLINE: 3356 case SO_ACCEPTCONN: 3357 case SO_TIMESTAMP: 3358 case SO_BINTIME: 3359 case SO_NOSIGPIPE: 3360 case SO_NO_DDP: 3361 case SO_NO_OFFLOAD: 3362 case SO_RERROR: 3363 optval = so->so_options & sopt->sopt_name; 3364 integer: 3365 error = sooptcopyout(sopt, &optval, sizeof optval); 3366 break; 3367 3368 case SO_DOMAIN: 3369 optval = so->so_proto->pr_domain->dom_family; 3370 goto integer; 3371 3372 case SO_TYPE: 3373 optval = so->so_type; 3374 goto integer; 3375 3376 case SO_PROTOCOL: 3377 optval = so->so_proto->pr_protocol; 3378 goto integer; 3379 3380 case SO_ERROR: 3381 SOCK_LOCK(so); 3382 if (so->so_error) { 3383 optval = so->so_error; 3384 so->so_error = 0; 3385 } else { 3386 optval = so->so_rerror; 3387 so->so_rerror = 0; 3388 } 3389 SOCK_UNLOCK(so); 3390 goto integer; 3391 3392 case SO_SNDBUF: 3393 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : 3394 so->so_snd.sb_hiwat; 3395 goto integer; 3396 3397 case SO_RCVBUF: 3398 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : 3399 so->so_rcv.sb_hiwat; 3400 goto integer; 3401 3402 case SO_SNDLOWAT: 3403 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : 3404 so->so_snd.sb_lowat; 3405 goto integer; 3406 3407 case SO_RCVLOWAT: 3408 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : 3409 so->so_rcv.sb_lowat; 3410 goto integer; 3411 3412 case SO_SNDTIMEO: 3413 case SO_RCVTIMEO: 3414 SOCK_LOCK(so); 3415 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 3416 (SOLISTENING(so) ? so->sol_sbsnd_timeo : 3417 so->so_snd.sb_timeo) : 3418 (SOLISTENING(so) ? so->sol_sbrcv_timeo : 3419 so->so_rcv.sb_timeo)); 3420 SOCK_UNLOCK(so); 3421 #ifdef COMPAT_FREEBSD32 3422 if (SV_CURPROC_FLAG(SV_ILP32)) { 3423 struct timeval32 tv32; 3424 3425 CP(tv, tv32, tv_sec); 3426 CP(tv, tv32, tv_usec); 3427 error = sooptcopyout(sopt, &tv32, sizeof tv32); 3428 } else 3429 #endif 3430 error = sooptcopyout(sopt, &tv, sizeof tv); 3431 break; 3432 3433 case SO_LABEL: 3434 #ifdef MAC 3435 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3436 sizeof(extmac)); 3437 if (error) 3438 goto bad; 3439 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 3440 so, &extmac); 3441 if (error) 3442 goto bad; 3443 /* Don't copy out extmac, it is unchanged. */ 3444 #else 3445 error = EOPNOTSUPP; 3446 #endif 3447 break; 3448 3449 case SO_PEERLABEL: 3450 #ifdef MAC 3451 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3452 sizeof(extmac)); 3453 if (error) 3454 goto bad; 3455 error = mac_getsockopt_peerlabel( 3456 sopt->sopt_td->td_ucred, so, &extmac); 3457 if (error) 3458 goto bad; 3459 /* Don't copy out extmac, it is unchanged. */ 3460 #else 3461 error = EOPNOTSUPP; 3462 #endif 3463 break; 3464 3465 case SO_LISTENQLIMIT: 3466 optval = SOLISTENING(so) ? so->sol_qlimit : 0; 3467 goto integer; 3468 3469 case SO_LISTENQLEN: 3470 optval = SOLISTENING(so) ? so->sol_qlen : 0; 3471 goto integer; 3472 3473 case SO_LISTENINCQLEN: 3474 optval = SOLISTENING(so) ? so->sol_incqlen : 0; 3475 goto integer; 3476 3477 case SO_TS_CLOCK: 3478 optval = so->so_ts_clock; 3479 goto integer; 3480 3481 case SO_MAX_PACING_RATE: 3482 optval = so->so_max_pacing_rate; 3483 goto integer; 3484 3485 default: 3486 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3487 error = hhook_run_socket(so, sopt, 3488 HHOOK_SOCKET_OPT); 3489 else 3490 error = ENOPROTOOPT; 3491 break; 3492 } 3493 } 3494 #ifdef MAC 3495 bad: 3496 #endif 3497 CURVNET_RESTORE(); 3498 return (error); 3499 } 3500 3501 int 3502 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 3503 { 3504 struct mbuf *m, *m_prev; 3505 int sopt_size = sopt->sopt_valsize; 3506 3507 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3508 if (m == NULL) 3509 return ENOBUFS; 3510 if (sopt_size > MLEN) { 3511 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 3512 if ((m->m_flags & M_EXT) == 0) { 3513 m_free(m); 3514 return ENOBUFS; 3515 } 3516 m->m_len = min(MCLBYTES, sopt_size); 3517 } else { 3518 m->m_len = min(MLEN, sopt_size); 3519 } 3520 sopt_size -= m->m_len; 3521 *mp = m; 3522 m_prev = m; 3523 3524 while (sopt_size) { 3525 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3526 if (m == NULL) { 3527 m_freem(*mp); 3528 return ENOBUFS; 3529 } 3530 if (sopt_size > MLEN) { 3531 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 3532 M_NOWAIT); 3533 if ((m->m_flags & M_EXT) == 0) { 3534 m_freem(m); 3535 m_freem(*mp); 3536 return ENOBUFS; 3537 } 3538 m->m_len = min(MCLBYTES, sopt_size); 3539 } else { 3540 m->m_len = min(MLEN, sopt_size); 3541 } 3542 sopt_size -= m->m_len; 3543 m_prev->m_next = m; 3544 m_prev = m; 3545 } 3546 return (0); 3547 } 3548 3549 int 3550 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 3551 { 3552 struct mbuf *m0 = m; 3553 3554 if (sopt->sopt_val == NULL) 3555 return (0); 3556 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3557 if (sopt->sopt_td != NULL) { 3558 int error; 3559 3560 error = copyin(sopt->sopt_val, mtod(m, char *), 3561 m->m_len); 3562 if (error != 0) { 3563 m_freem(m0); 3564 return(error); 3565 } 3566 } else 3567 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 3568 sopt->sopt_valsize -= m->m_len; 3569 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3570 m = m->m_next; 3571 } 3572 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 3573 panic("ip6_sooptmcopyin"); 3574 return (0); 3575 } 3576 3577 int 3578 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 3579 { 3580 struct mbuf *m0 = m; 3581 size_t valsize = 0; 3582 3583 if (sopt->sopt_val == NULL) 3584 return (0); 3585 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3586 if (sopt->sopt_td != NULL) { 3587 int error; 3588 3589 error = copyout(mtod(m, char *), sopt->sopt_val, 3590 m->m_len); 3591 if (error != 0) { 3592 m_freem(m0); 3593 return(error); 3594 } 3595 } else 3596 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 3597 sopt->sopt_valsize -= m->m_len; 3598 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3599 valsize += m->m_len; 3600 m = m->m_next; 3601 } 3602 if (m != NULL) { 3603 /* enough soopt buffer should be given from user-land */ 3604 m_freem(m0); 3605 return(EINVAL); 3606 } 3607 sopt->sopt_valsize = valsize; 3608 return (0); 3609 } 3610 3611 /* 3612 * sohasoutofband(): protocol notifies socket layer of the arrival of new 3613 * out-of-band data, which will then notify socket consumers. 3614 */ 3615 void 3616 sohasoutofband(struct socket *so) 3617 { 3618 3619 if (so->so_sigio != NULL) 3620 pgsigio(&so->so_sigio, SIGURG, 0); 3621 selwakeuppri(&so->so_rdsel, PSOCK); 3622 } 3623 3624 int 3625 sopoll(struct socket *so, int events, struct ucred *active_cred, 3626 struct thread *td) 3627 { 3628 3629 /* 3630 * We do not need to set or assert curvnet as long as everyone uses 3631 * sopoll_generic(). 3632 */ 3633 return (so->so_proto->pr_sopoll(so, events, active_cred, td)); 3634 } 3635 3636 int 3637 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3638 struct thread *td) 3639 { 3640 int revents; 3641 3642 SOCK_LOCK(so); 3643 if (SOLISTENING(so)) { 3644 if (!(events & (POLLIN | POLLRDNORM))) 3645 revents = 0; 3646 else if (!TAILQ_EMPTY(&so->sol_comp)) 3647 revents = events & (POLLIN | POLLRDNORM); 3648 else if ((events & POLLINIGNEOF) == 0 && so->so_error) 3649 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP; 3650 else { 3651 selrecord(td, &so->so_rdsel); 3652 revents = 0; 3653 } 3654 } else { 3655 revents = 0; 3656 SOCK_SENDBUF_LOCK(so); 3657 SOCK_RECVBUF_LOCK(so); 3658 if (events & (POLLIN | POLLRDNORM)) 3659 if (soreadabledata(so)) 3660 revents |= events & (POLLIN | POLLRDNORM); 3661 if (events & (POLLOUT | POLLWRNORM)) 3662 if (sowriteable(so)) 3663 revents |= events & (POLLOUT | POLLWRNORM); 3664 if (events & (POLLPRI | POLLRDBAND)) 3665 if (so->so_oobmark || 3666 (so->so_rcv.sb_state & SBS_RCVATMARK)) 3667 revents |= events & (POLLPRI | POLLRDBAND); 3668 if ((events & POLLINIGNEOF) == 0) { 3669 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3670 revents |= events & (POLLIN | POLLRDNORM); 3671 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3672 revents |= POLLHUP; 3673 } 3674 } 3675 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3676 revents |= events & POLLRDHUP; 3677 if (revents == 0) { 3678 if (events & 3679 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) { 3680 selrecord(td, &so->so_rdsel); 3681 so->so_rcv.sb_flags |= SB_SEL; 3682 } 3683 if (events & (POLLOUT | POLLWRNORM)) { 3684 selrecord(td, &so->so_wrsel); 3685 so->so_snd.sb_flags |= SB_SEL; 3686 } 3687 } 3688 SOCK_RECVBUF_UNLOCK(so); 3689 SOCK_SENDBUF_UNLOCK(so); 3690 } 3691 SOCK_UNLOCK(so); 3692 return (revents); 3693 } 3694 3695 int 3696 soo_kqfilter(struct file *fp, struct knote *kn) 3697 { 3698 struct socket *so = kn->kn_fp->f_data; 3699 struct sockbuf *sb; 3700 sb_which which; 3701 struct knlist *knl; 3702 3703 switch (kn->kn_filter) { 3704 case EVFILT_READ: 3705 kn->kn_fop = &soread_filtops; 3706 knl = &so->so_rdsel.si_note; 3707 sb = &so->so_rcv; 3708 which = SO_RCV; 3709 break; 3710 case EVFILT_WRITE: 3711 kn->kn_fop = &sowrite_filtops; 3712 knl = &so->so_wrsel.si_note; 3713 sb = &so->so_snd; 3714 which = SO_SND; 3715 break; 3716 case EVFILT_EMPTY: 3717 kn->kn_fop = &soempty_filtops; 3718 knl = &so->so_wrsel.si_note; 3719 sb = &so->so_snd; 3720 which = SO_SND; 3721 break; 3722 default: 3723 return (EINVAL); 3724 } 3725 3726 SOCK_LOCK(so); 3727 if (SOLISTENING(so)) { 3728 knlist_add(knl, kn, 1); 3729 } else { 3730 SOCK_BUF_LOCK(so, which); 3731 knlist_add(knl, kn, 1); 3732 sb->sb_flags |= SB_KNOTE; 3733 SOCK_BUF_UNLOCK(so, which); 3734 } 3735 SOCK_UNLOCK(so); 3736 return (0); 3737 } 3738 3739 static void 3740 filt_sordetach(struct knote *kn) 3741 { 3742 struct socket *so = kn->kn_fp->f_data; 3743 3744 so_rdknl_lock(so); 3745 knlist_remove(&so->so_rdsel.si_note, kn, 1); 3746 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note)) 3747 so->so_rcv.sb_flags &= ~SB_KNOTE; 3748 so_rdknl_unlock(so); 3749 } 3750 3751 /*ARGSUSED*/ 3752 static int 3753 filt_soread(struct knote *kn, long hint) 3754 { 3755 struct socket *so; 3756 3757 so = kn->kn_fp->f_data; 3758 3759 if (SOLISTENING(so)) { 3760 SOCK_LOCK_ASSERT(so); 3761 kn->kn_data = so->sol_qlen; 3762 if (so->so_error) { 3763 kn->kn_flags |= EV_EOF; 3764 kn->kn_fflags = so->so_error; 3765 return (1); 3766 } 3767 return (!TAILQ_EMPTY(&so->sol_comp)); 3768 } 3769 3770 SOCK_RECVBUF_LOCK_ASSERT(so); 3771 3772 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 3773 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3774 kn->kn_flags |= EV_EOF; 3775 kn->kn_fflags = so->so_error; 3776 return (1); 3777 } else if (so->so_error || so->so_rerror) 3778 return (1); 3779 3780 if (kn->kn_sfflags & NOTE_LOWAT) { 3781 if (kn->kn_data >= kn->kn_sdata) 3782 return (1); 3783 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) 3784 return (1); 3785 3786 /* This hook returning non-zero indicates an event, not error */ 3787 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); 3788 } 3789 3790 static void 3791 filt_sowdetach(struct knote *kn) 3792 { 3793 struct socket *so = kn->kn_fp->f_data; 3794 3795 so_wrknl_lock(so); 3796 knlist_remove(&so->so_wrsel.si_note, kn, 1); 3797 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note)) 3798 so->so_snd.sb_flags &= ~SB_KNOTE; 3799 so_wrknl_unlock(so); 3800 } 3801 3802 /*ARGSUSED*/ 3803 static int 3804 filt_sowrite(struct knote *kn, long hint) 3805 { 3806 struct socket *so; 3807 3808 so = kn->kn_fp->f_data; 3809 3810 if (SOLISTENING(so)) 3811 return (0); 3812 3813 SOCK_SENDBUF_LOCK_ASSERT(so); 3814 kn->kn_data = sbspace(&so->so_snd); 3815 3816 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); 3817 3818 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3819 kn->kn_flags |= EV_EOF; 3820 kn->kn_fflags = so->so_error; 3821 return (1); 3822 } else if (so->so_error) /* temporary udp error */ 3823 return (1); 3824 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3825 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3826 return (0); 3827 else if (kn->kn_sfflags & NOTE_LOWAT) 3828 return (kn->kn_data >= kn->kn_sdata); 3829 else 3830 return (kn->kn_data >= so->so_snd.sb_lowat); 3831 } 3832 3833 static int 3834 filt_soempty(struct knote *kn, long hint) 3835 { 3836 struct socket *so; 3837 3838 so = kn->kn_fp->f_data; 3839 3840 if (SOLISTENING(so)) 3841 return (1); 3842 3843 SOCK_SENDBUF_LOCK_ASSERT(so); 3844 kn->kn_data = sbused(&so->so_snd); 3845 3846 if (kn->kn_data == 0) 3847 return (1); 3848 else 3849 return (0); 3850 } 3851 3852 int 3853 socheckuid(struct socket *so, uid_t uid) 3854 { 3855 3856 if (so == NULL) 3857 return (EPERM); 3858 if (so->so_cred->cr_uid != uid) 3859 return (EPERM); 3860 return (0); 3861 } 3862 3863 /* 3864 * These functions are used by protocols to notify the socket layer (and its 3865 * consumers) of state changes in the sockets driven by protocol-side events. 3866 */ 3867 3868 /* 3869 * Procedures to manipulate state flags of socket and do appropriate wakeups. 3870 * 3871 * Normal sequence from the active (originating) side is that 3872 * soisconnecting() is called during processing of connect() call, resulting 3873 * in an eventual call to soisconnected() if/when the connection is 3874 * established. When the connection is torn down soisdisconnecting() is 3875 * called during processing of disconnect() call, and soisdisconnected() is 3876 * called when the connection to the peer is totally severed. The semantics 3877 * of these routines are such that connectionless protocols can call 3878 * soisconnected() and soisdisconnected() only, bypassing the in-progress 3879 * calls when setting up a ``connection'' takes no time. 3880 * 3881 * From the passive side, a socket is created with two queues of sockets: 3882 * so_incomp for connections in progress and so_comp for connections already 3883 * made and awaiting user acceptance. As a protocol is preparing incoming 3884 * connections, it creates a socket structure queued on so_incomp by calling 3885 * sonewconn(). When the connection is established, soisconnected() is 3886 * called, and transfers the socket structure to so_comp, making it available 3887 * to accept(). 3888 * 3889 * If a socket is closed with sockets on either so_incomp or so_comp, these 3890 * sockets are dropped. 3891 * 3892 * If higher-level protocols are implemented in the kernel, the wakeups done 3893 * here will sometimes cause software-interrupt process scheduling. 3894 */ 3895 void 3896 soisconnecting(struct socket *so) 3897 { 3898 3899 SOCK_LOCK(so); 3900 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 3901 so->so_state |= SS_ISCONNECTING; 3902 SOCK_UNLOCK(so); 3903 } 3904 3905 void 3906 soisconnected(struct socket *so) 3907 { 3908 bool last __diagused; 3909 3910 SOCK_LOCK(so); 3911 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 3912 so->so_state |= SS_ISCONNECTED; 3913 3914 if (so->so_qstate == SQ_INCOMP) { 3915 struct socket *head = so->so_listen; 3916 int ret; 3917 3918 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so)); 3919 /* 3920 * Promoting a socket from incomplete queue to complete, we 3921 * need to go through reverse order of locking. We first do 3922 * trylock, and if that doesn't succeed, we go the hard way 3923 * leaving a reference and rechecking consistency after proper 3924 * locking. 3925 */ 3926 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) { 3927 soref(head); 3928 SOCK_UNLOCK(so); 3929 SOLISTEN_LOCK(head); 3930 SOCK_LOCK(so); 3931 if (__predict_false(head != so->so_listen)) { 3932 /* 3933 * The socket went off the listen queue, 3934 * should be lost race to close(2) of sol. 3935 * The socket is about to soabort(). 3936 */ 3937 SOCK_UNLOCK(so); 3938 sorele_locked(head); 3939 return; 3940 } 3941 last = refcount_release(&head->so_count); 3942 KASSERT(!last, ("%s: released last reference for %p", 3943 __func__, head)); 3944 } 3945 again: 3946 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 3947 TAILQ_REMOVE(&head->sol_incomp, so, so_list); 3948 head->sol_incqlen--; 3949 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 3950 head->sol_qlen++; 3951 so->so_qstate = SQ_COMP; 3952 SOCK_UNLOCK(so); 3953 solisten_wakeup(head); /* unlocks */ 3954 } else { 3955 SOCK_RECVBUF_LOCK(so); 3956 soupcall_set(so, SO_RCV, 3957 head->sol_accept_filter->accf_callback, 3958 head->sol_accept_filter_arg); 3959 so->so_options &= ~SO_ACCEPTFILTER; 3960 ret = head->sol_accept_filter->accf_callback(so, 3961 head->sol_accept_filter_arg, M_NOWAIT); 3962 if (ret == SU_ISCONNECTED) { 3963 soupcall_clear(so, SO_RCV); 3964 SOCK_RECVBUF_UNLOCK(so); 3965 goto again; 3966 } 3967 SOCK_RECVBUF_UNLOCK(so); 3968 SOCK_UNLOCK(so); 3969 SOLISTEN_UNLOCK(head); 3970 } 3971 return; 3972 } 3973 SOCK_UNLOCK(so); 3974 wakeup(&so->so_timeo); 3975 sorwakeup(so); 3976 sowwakeup(so); 3977 } 3978 3979 void 3980 soisdisconnecting(struct socket *so) 3981 { 3982 3983 SOCK_LOCK(so); 3984 so->so_state &= ~SS_ISCONNECTING; 3985 so->so_state |= SS_ISDISCONNECTING; 3986 3987 if (!SOLISTENING(so)) { 3988 SOCK_RECVBUF_LOCK(so); 3989 socantrcvmore_locked(so); 3990 SOCK_SENDBUF_LOCK(so); 3991 socantsendmore_locked(so); 3992 } 3993 SOCK_UNLOCK(so); 3994 wakeup(&so->so_timeo); 3995 } 3996 3997 void 3998 soisdisconnected(struct socket *so) 3999 { 4000 4001 SOCK_LOCK(so); 4002 4003 /* 4004 * There is at least one reader of so_state that does not 4005 * acquire socket lock, namely soreceive_generic(). Ensure 4006 * that it never sees all flags that track connection status 4007 * cleared, by ordering the update with a barrier semantic of 4008 * our release thread fence. 4009 */ 4010 so->so_state |= SS_ISDISCONNECTED; 4011 atomic_thread_fence_rel(); 4012 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 4013 4014 if (!SOLISTENING(so)) { 4015 SOCK_UNLOCK(so); 4016 SOCK_RECVBUF_LOCK(so); 4017 socantrcvmore_locked(so); 4018 SOCK_SENDBUF_LOCK(so); 4019 sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); 4020 socantsendmore_locked(so); 4021 } else 4022 SOCK_UNLOCK(so); 4023 wakeup(&so->so_timeo); 4024 } 4025 4026 int 4027 soiolock(struct socket *so, struct sx *sx, int flags) 4028 { 4029 int error; 4030 4031 KASSERT((flags & SBL_VALID) == flags, 4032 ("soiolock: invalid flags %#x", flags)); 4033 4034 if ((flags & SBL_WAIT) != 0) { 4035 if ((flags & SBL_NOINTR) != 0) { 4036 sx_xlock(sx); 4037 } else { 4038 error = sx_xlock_sig(sx); 4039 if (error != 0) 4040 return (error); 4041 } 4042 } else if (!sx_try_xlock(sx)) { 4043 return (EWOULDBLOCK); 4044 } 4045 4046 if (__predict_false(SOLISTENING(so))) { 4047 sx_xunlock(sx); 4048 return (ENOTCONN); 4049 } 4050 return (0); 4051 } 4052 4053 void 4054 soiounlock(struct sx *sx) 4055 { 4056 sx_xunlock(sx); 4057 } 4058 4059 /* 4060 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 4061 */ 4062 struct sockaddr * 4063 sodupsockaddr(const struct sockaddr *sa, int mflags) 4064 { 4065 struct sockaddr *sa2; 4066 4067 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 4068 if (sa2) 4069 bcopy(sa, sa2, sa->sa_len); 4070 return sa2; 4071 } 4072 4073 /* 4074 * Register per-socket destructor. 4075 */ 4076 void 4077 sodtor_set(struct socket *so, so_dtor_t *func) 4078 { 4079 4080 SOCK_LOCK_ASSERT(so); 4081 so->so_dtor = func; 4082 } 4083 4084 /* 4085 * Register per-socket buffer upcalls. 4086 */ 4087 void 4088 soupcall_set(struct socket *so, sb_which which, so_upcall_t func, void *arg) 4089 { 4090 struct sockbuf *sb; 4091 4092 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4093 4094 switch (which) { 4095 case SO_RCV: 4096 sb = &so->so_rcv; 4097 break; 4098 case SO_SND: 4099 sb = &so->so_snd; 4100 break; 4101 } 4102 SOCK_BUF_LOCK_ASSERT(so, which); 4103 sb->sb_upcall = func; 4104 sb->sb_upcallarg = arg; 4105 sb->sb_flags |= SB_UPCALL; 4106 } 4107 4108 void 4109 soupcall_clear(struct socket *so, sb_which which) 4110 { 4111 struct sockbuf *sb; 4112 4113 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4114 4115 switch (which) { 4116 case SO_RCV: 4117 sb = &so->so_rcv; 4118 break; 4119 case SO_SND: 4120 sb = &so->so_snd; 4121 break; 4122 } 4123 SOCK_BUF_LOCK_ASSERT(so, which); 4124 KASSERT(sb->sb_upcall != NULL, 4125 ("%s: so %p no upcall to clear", __func__, so)); 4126 sb->sb_upcall = NULL; 4127 sb->sb_upcallarg = NULL; 4128 sb->sb_flags &= ~SB_UPCALL; 4129 } 4130 4131 void 4132 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg) 4133 { 4134 4135 SOLISTEN_LOCK_ASSERT(so); 4136 so->sol_upcall = func; 4137 so->sol_upcallarg = arg; 4138 } 4139 4140 static void 4141 so_rdknl_lock(void *arg) 4142 { 4143 struct socket *so = arg; 4144 4145 retry: 4146 if (SOLISTENING(so)) { 4147 SOLISTEN_LOCK(so); 4148 } else { 4149 SOCK_RECVBUF_LOCK(so); 4150 if (__predict_false(SOLISTENING(so))) { 4151 SOCK_RECVBUF_UNLOCK(so); 4152 goto retry; 4153 } 4154 } 4155 } 4156 4157 static void 4158 so_rdknl_unlock(void *arg) 4159 { 4160 struct socket *so = arg; 4161 4162 if (SOLISTENING(so)) 4163 SOLISTEN_UNLOCK(so); 4164 else 4165 SOCK_RECVBUF_UNLOCK(so); 4166 } 4167 4168 static void 4169 so_rdknl_assert_lock(void *arg, int what) 4170 { 4171 struct socket *so = arg; 4172 4173 if (what == LA_LOCKED) { 4174 if (SOLISTENING(so)) 4175 SOLISTEN_LOCK_ASSERT(so); 4176 else 4177 SOCK_RECVBUF_LOCK_ASSERT(so); 4178 } else { 4179 if (SOLISTENING(so)) 4180 SOLISTEN_UNLOCK_ASSERT(so); 4181 else 4182 SOCK_RECVBUF_UNLOCK_ASSERT(so); 4183 } 4184 } 4185 4186 static void 4187 so_wrknl_lock(void *arg) 4188 { 4189 struct socket *so = arg; 4190 4191 retry: 4192 if (SOLISTENING(so)) { 4193 SOLISTEN_LOCK(so); 4194 } else { 4195 SOCK_SENDBUF_LOCK(so); 4196 if (__predict_false(SOLISTENING(so))) { 4197 SOCK_SENDBUF_UNLOCK(so); 4198 goto retry; 4199 } 4200 } 4201 } 4202 4203 static void 4204 so_wrknl_unlock(void *arg) 4205 { 4206 struct socket *so = arg; 4207 4208 if (SOLISTENING(so)) 4209 SOLISTEN_UNLOCK(so); 4210 else 4211 SOCK_SENDBUF_UNLOCK(so); 4212 } 4213 4214 static void 4215 so_wrknl_assert_lock(void *arg, int what) 4216 { 4217 struct socket *so = arg; 4218 4219 if (what == LA_LOCKED) { 4220 if (SOLISTENING(so)) 4221 SOLISTEN_LOCK_ASSERT(so); 4222 else 4223 SOCK_SENDBUF_LOCK_ASSERT(so); 4224 } else { 4225 if (SOLISTENING(so)) 4226 SOLISTEN_UNLOCK_ASSERT(so); 4227 else 4228 SOCK_SENDBUF_UNLOCK_ASSERT(so); 4229 } 4230 } 4231 4232 /* 4233 * Create an external-format (``xsocket'') structure using the information in 4234 * the kernel-format socket structure pointed to by so. This is done to 4235 * reduce the spew of irrelevant information over this interface, to isolate 4236 * user code from changes in the kernel structure, and potentially to provide 4237 * information-hiding if we decide that some of this information should be 4238 * hidden from users. 4239 */ 4240 void 4241 sotoxsocket(struct socket *so, struct xsocket *xso) 4242 { 4243 4244 bzero(xso, sizeof(*xso)); 4245 xso->xso_len = sizeof *xso; 4246 xso->xso_so = (uintptr_t)so; 4247 xso->so_type = so->so_type; 4248 xso->so_options = so->so_options; 4249 xso->so_linger = so->so_linger; 4250 xso->so_state = so->so_state; 4251 xso->so_pcb = (uintptr_t)so->so_pcb; 4252 xso->xso_protocol = so->so_proto->pr_protocol; 4253 xso->xso_family = so->so_proto->pr_domain->dom_family; 4254 xso->so_timeo = so->so_timeo; 4255 xso->so_error = so->so_error; 4256 xso->so_uid = so->so_cred->cr_uid; 4257 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 4258 if (SOLISTENING(so)) { 4259 xso->so_qlen = so->sol_qlen; 4260 xso->so_incqlen = so->sol_incqlen; 4261 xso->so_qlimit = so->sol_qlimit; 4262 xso->so_oobmark = 0; 4263 } else { 4264 xso->so_state |= so->so_qstate; 4265 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0; 4266 xso->so_oobmark = so->so_oobmark; 4267 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 4268 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 4269 } 4270 } 4271 4272 struct sockbuf * 4273 so_sockbuf_rcv(struct socket *so) 4274 { 4275 4276 return (&so->so_rcv); 4277 } 4278 4279 struct sockbuf * 4280 so_sockbuf_snd(struct socket *so) 4281 { 4282 4283 return (&so->so_snd); 4284 } 4285 4286 int 4287 so_state_get(const struct socket *so) 4288 { 4289 4290 return (so->so_state); 4291 } 4292 4293 void 4294 so_state_set(struct socket *so, int val) 4295 { 4296 4297 so->so_state = val; 4298 } 4299 4300 int 4301 so_options_get(const struct socket *so) 4302 { 4303 4304 return (so->so_options); 4305 } 4306 4307 void 4308 so_options_set(struct socket *so, int val) 4309 { 4310 4311 so->so_options = val; 4312 } 4313 4314 int 4315 so_error_get(const struct socket *so) 4316 { 4317 4318 return (so->so_error); 4319 } 4320 4321 void 4322 so_error_set(struct socket *so, int val) 4323 { 4324 4325 so->so_error = val; 4326 } 4327 4328 int 4329 so_linger_get(const struct socket *so) 4330 { 4331 4332 return (so->so_linger); 4333 } 4334 4335 void 4336 so_linger_set(struct socket *so, int val) 4337 { 4338 4339 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz), 4340 ("%s: val %d out of range", __func__, val)); 4341 4342 so->so_linger = val; 4343 } 4344 4345 struct protosw * 4346 so_protosw_get(const struct socket *so) 4347 { 4348 4349 return (so->so_proto); 4350 } 4351 4352 void 4353 so_protosw_set(struct socket *so, struct protosw *val) 4354 { 4355 4356 so->so_proto = val; 4357 } 4358 4359 void 4360 so_sorwakeup(struct socket *so) 4361 { 4362 4363 sorwakeup(so); 4364 } 4365 4366 void 4367 so_sowwakeup(struct socket *so) 4368 { 4369 4370 sowwakeup(so); 4371 } 4372 4373 void 4374 so_sorwakeup_locked(struct socket *so) 4375 { 4376 4377 sorwakeup_locked(so); 4378 } 4379 4380 void 4381 so_sowwakeup_locked(struct socket *so) 4382 { 4383 4384 sowwakeup_locked(so); 4385 } 4386 4387 void 4388 so_lock(struct socket *so) 4389 { 4390 4391 SOCK_LOCK(so); 4392 } 4393 4394 void 4395 so_unlock(struct socket *so) 4396 { 4397 4398 SOCK_UNLOCK(so); 4399 } 4400