1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993 5 * The Regents of the University of California. 6 * Copyright (c) 2004 The FreeBSD Foundation 7 * Copyright (c) 2004-2008 Robert N. M. Watson 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 35 */ 36 37 /* 38 * Comments on the socket life cycle: 39 * 40 * soalloc() sets of socket layer state for a socket, called only by 41 * socreate() and sonewconn(). Socket layer private. 42 * 43 * sodealloc() tears down socket layer state for a socket, called only by 44 * sofree() and sonewconn(). Socket layer private. 45 * 46 * pru_attach() associates protocol layer state with an allocated socket; 47 * called only once, may fail, aborting socket allocation. This is called 48 * from socreate() and sonewconn(). Socket layer private. 49 * 50 * pru_detach() disassociates protocol layer state from an attached socket, 51 * and will be called exactly once for sockets in which pru_attach() has 52 * been successfully called. If pru_attach() returned an error, 53 * pru_detach() will not be called. Socket layer private. 54 * 55 * pru_abort() and pru_close() notify the protocol layer that the last 56 * consumer of a socket is starting to tear down the socket, and that the 57 * protocol should terminate the connection. Historically, pru_abort() also 58 * detached protocol state from the socket state, but this is no longer the 59 * case. 60 * 61 * socreate() creates a socket and attaches protocol state. This is a public 62 * interface that may be used by socket layer consumers to create new 63 * sockets. 64 * 65 * sonewconn() creates a socket and attaches protocol state. This is a 66 * public interface that may be used by protocols to create new sockets when 67 * a new connection is received and will be available for accept() on a 68 * listen socket. 69 * 70 * soclose() destroys a socket after possibly waiting for it to disconnect. 71 * This is a public interface that socket consumers should use to close and 72 * release a socket when done with it. 73 * 74 * soabort() destroys a socket without waiting for it to disconnect (used 75 * only for incoming connections that are already partially or fully 76 * connected). This is used internally by the socket layer when clearing 77 * listen socket queues (due to overflow or close on the listen socket), but 78 * is also a public interface protocols may use to abort connections in 79 * their incomplete listen queues should they no longer be required. Sockets 80 * placed in completed connection listen queues should not be aborted for 81 * reasons described in the comment above the soclose() implementation. This 82 * is not a general purpose close routine, and except in the specific 83 * circumstances described here, should not be used. 84 * 85 * sofree() will free a socket and its protocol state if all references on 86 * the socket have been released, and is the public interface to attempt to 87 * free a socket when a reference is removed. This is a socket layer private 88 * interface. 89 * 90 * NOTE: In addition to socreate() and soclose(), which provide a single 91 * socket reference to the consumer to be managed as required, there are two 92 * calls to explicitly manage socket references, soref(), and sorele(). 93 * Currently, these are generally required only when transitioning a socket 94 * from a listen queue to a file descriptor, in order to prevent garbage 95 * collection of the socket at an untimely moment. For a number of reasons, 96 * these interfaces are not preferred, and should be avoided. 97 * 98 * NOTE: With regard to VNETs the general rule is that callers do not set 99 * curvnet. Exceptions to this rule include soabort(), sodisconnect(), 100 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn() 101 * and sorflush(), which are usually called from a pre-set VNET context. 102 * sopoll() currently does not need a VNET context to be set. 103 */ 104 105 #include <sys/cdefs.h> 106 __FBSDID("$FreeBSD$"); 107 108 #include "opt_inet.h" 109 #include "opt_inet6.h" 110 #include "opt_sctp.h" 111 112 #include <sys/param.h> 113 #include <sys/systm.h> 114 #include <sys/fcntl.h> 115 #include <sys/limits.h> 116 #include <sys/lock.h> 117 #include <sys/mac.h> 118 #include <sys/malloc.h> 119 #include <sys/mbuf.h> 120 #include <sys/mutex.h> 121 #include <sys/domain.h> 122 #include <sys/file.h> /* for struct knote */ 123 #include <sys/hhook.h> 124 #include <sys/kernel.h> 125 #include <sys/khelp.h> 126 #include <sys/event.h> 127 #include <sys/eventhandler.h> 128 #include <sys/poll.h> 129 #include <sys/proc.h> 130 #include <sys/protosw.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/jail.h> 142 #include <sys/syslog.h> 143 #include <netinet/in.h> 144 145 #include <net/vnet.h> 146 147 #include <security/mac/mac_framework.h> 148 149 #include <vm/uma.h> 150 151 #ifdef COMPAT_FREEBSD32 152 #include <sys/mount.h> 153 #include <sys/sysent.h> 154 #include <compat/freebsd32/freebsd32.h> 155 #endif 156 157 static int soreceive_rcvoob(struct socket *so, struct uio *uio, 158 int flags); 159 static void so_rdknl_lock(void *); 160 static void so_rdknl_unlock(void *); 161 static void so_rdknl_assert_locked(void *); 162 static void so_rdknl_assert_unlocked(void *); 163 static void so_wrknl_lock(void *); 164 static void so_wrknl_unlock(void *); 165 static void so_wrknl_assert_locked(void *); 166 static void so_wrknl_assert_unlocked(void *); 167 168 static void filt_sordetach(struct knote *kn); 169 static int filt_soread(struct knote *kn, long hint); 170 static void filt_sowdetach(struct knote *kn); 171 static int filt_sowrite(struct knote *kn, long hint); 172 static int filt_soempty(struct knote *kn, long hint); 173 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id); 174 fo_kqfilter_t soo_kqfilter; 175 176 static struct filterops soread_filtops = { 177 .f_isfd = 1, 178 .f_detach = filt_sordetach, 179 .f_event = filt_soread, 180 }; 181 static struct filterops sowrite_filtops = { 182 .f_isfd = 1, 183 .f_detach = filt_sowdetach, 184 .f_event = filt_sowrite, 185 }; 186 static struct filterops soempty_filtops = { 187 .f_isfd = 1, 188 .f_detach = filt_sowdetach, 189 .f_event = filt_soempty, 190 }; 191 192 so_gen_t so_gencnt; /* generation count for sockets */ 193 194 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 195 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 196 197 #define VNET_SO_ASSERT(so) \ 198 VNET_ASSERT(curvnet != NULL, \ 199 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so))); 200 201 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]); 202 #define V_socket_hhh VNET(socket_hhh) 203 204 /* 205 * Limit on the number of connections in the listen queue waiting 206 * for accept(2). 207 * NB: The original sysctl somaxconn is still available but hidden 208 * to prevent confusion about the actual purpose of this number. 209 */ 210 static u_int somaxconn = SOMAXCONN; 211 212 static int 213 sysctl_somaxconn(SYSCTL_HANDLER_ARGS) 214 { 215 int error; 216 int val; 217 218 val = somaxconn; 219 error = sysctl_handle_int(oidp, &val, 0, req); 220 if (error || !req->newptr ) 221 return (error); 222 223 /* 224 * The purpose of the UINT_MAX / 3 limit, is so that the formula 225 * 3 * so_qlimit / 2 226 * below, will not overflow. 227 */ 228 229 if (val < 1 || val > UINT_MAX / 3) 230 return (EINVAL); 231 232 somaxconn = val; 233 return (0); 234 } 235 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW, 236 0, sizeof(int), sysctl_somaxconn, "I", 237 "Maximum listen socket pending connection accept queue size"); 238 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, 239 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP, 240 0, sizeof(int), sysctl_somaxconn, "I", 241 "Maximum listen socket pending connection accept queue size (compat)"); 242 243 static int numopensockets; 244 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD, 245 &numopensockets, 0, "Number of open sockets"); 246 247 /* 248 * accept_mtx locks down per-socket fields relating to accept queues. See 249 * socketvar.h for an annotation of the protected fields of struct socket. 250 */ 251 struct mtx accept_mtx; 252 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF); 253 254 /* 255 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket 256 * so_gencnt field. 257 */ 258 static struct mtx so_global_mtx; 259 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF); 260 261 /* 262 * General IPC sysctl name space, used by sockets and a variety of other IPC 263 * types. 264 */ 265 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); 266 267 /* 268 * Initialize the socket subsystem and set up the socket 269 * memory allocator. 270 */ 271 static uma_zone_t socket_zone; 272 int maxsockets; 273 274 static void 275 socket_zone_change(void *tag) 276 { 277 278 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 279 } 280 281 static void 282 socket_hhook_register(int subtype) 283 { 284 285 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype, 286 &V_socket_hhh[subtype], 287 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 288 printf("%s: WARNING: unable to register hook\n", __func__); 289 } 290 291 static void 292 socket_hhook_deregister(int subtype) 293 { 294 295 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0) 296 printf("%s: WARNING: unable to deregister hook\n", __func__); 297 } 298 299 static void 300 socket_init(void *tag) 301 { 302 303 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL, 304 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 305 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 306 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached"); 307 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL, 308 EVENTHANDLER_PRI_FIRST); 309 } 310 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL); 311 312 static void 313 socket_vnet_init(const void *unused __unused) 314 { 315 int i; 316 317 /* We expect a contiguous range */ 318 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 319 socket_hhook_register(i); 320 } 321 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 322 socket_vnet_init, NULL); 323 324 static void 325 socket_vnet_uninit(const void *unused __unused) 326 { 327 int i; 328 329 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 330 socket_hhook_deregister(i); 331 } 332 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 333 socket_vnet_uninit, NULL); 334 335 /* 336 * Initialise maxsockets. This SYSINIT must be run after 337 * tunable_mbinit(). 338 */ 339 static void 340 init_maxsockets(void *ignored) 341 { 342 343 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 344 maxsockets = imax(maxsockets, maxfiles); 345 } 346 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); 347 348 /* 349 * Sysctl to get and set the maximum global sockets limit. Notify protocols 350 * of the change so that they can update their dependent limits as required. 351 */ 352 static int 353 sysctl_maxsockets(SYSCTL_HANDLER_ARGS) 354 { 355 int error, newmaxsockets; 356 357 newmaxsockets = maxsockets; 358 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req); 359 if (error == 0 && req->newptr) { 360 if (newmaxsockets > maxsockets && 361 newmaxsockets <= maxfiles) { 362 maxsockets = newmaxsockets; 363 EVENTHANDLER_INVOKE(maxsockets_change); 364 } else 365 error = EINVAL; 366 } 367 return (error); 368 } 369 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW, 370 &maxsockets, 0, sysctl_maxsockets, "IU", 371 "Maximum number of sockets available"); 372 373 /* 374 * Socket operation routines. These routines are called by the routines in 375 * sys_socket.c or from a system process, and implement the semantics of 376 * socket operations by switching out to the protocol specific routines. 377 */ 378 379 /* 380 * Get a socket structure from our zone, and initialize it. Note that it 381 * would probably be better to allocate socket and PCB at the same time, but 382 * I'm not convinced that all the protocols can be easily modified to do 383 * this. 384 * 385 * soalloc() returns a socket with a ref count of 0. 386 */ 387 static struct socket * 388 soalloc(struct vnet *vnet) 389 { 390 struct socket *so; 391 392 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO); 393 if (so == NULL) 394 return (NULL); 395 #ifdef MAC 396 if (mac_socket_init(so, M_NOWAIT) != 0) { 397 uma_zfree(socket_zone, so); 398 return (NULL); 399 } 400 #endif 401 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) { 402 uma_zfree(socket_zone, so); 403 return (NULL); 404 } 405 406 /* 407 * The socket locking protocol allows to lock 2 sockets at a time, 408 * however, the first one must be a listening socket. WITNESS lacks 409 * a feature to change class of an existing lock, so we use DUPOK. 410 */ 411 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK); 412 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd"); 413 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv"); 414 so->so_rcv.sb_sel = &so->so_rdsel; 415 so->so_snd.sb_sel = &so->so_wrsel; 416 sx_init(&so->so_snd.sb_sx, "so_snd_sx"); 417 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx"); 418 TAILQ_INIT(&so->so_snd.sb_aiojobq); 419 TAILQ_INIT(&so->so_rcv.sb_aiojobq); 420 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so); 421 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so); 422 #ifdef VIMAGE 423 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p", 424 __func__, __LINE__, so)); 425 so->so_vnet = vnet; 426 #endif 427 /* We shouldn't need the so_global_mtx */ 428 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) { 429 /* Do we need more comprehensive error returns? */ 430 uma_zfree(socket_zone, so); 431 return (NULL); 432 } 433 mtx_lock(&so_global_mtx); 434 so->so_gencnt = ++so_gencnt; 435 ++numopensockets; 436 #ifdef VIMAGE 437 vnet->vnet_sockcnt++; 438 #endif 439 mtx_unlock(&so_global_mtx); 440 441 return (so); 442 } 443 444 /* 445 * Free the storage associated with a socket at the socket layer, tear down 446 * locks, labels, etc. All protocol state is assumed already to have been 447 * torn down (and possibly never set up) by the caller. 448 */ 449 static void 450 sodealloc(struct socket *so) 451 { 452 453 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); 454 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL")); 455 456 mtx_lock(&so_global_mtx); 457 so->so_gencnt = ++so_gencnt; 458 --numopensockets; /* Could be below, but faster here. */ 459 #ifdef VIMAGE 460 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p", 461 __func__, __LINE__, so)); 462 so->so_vnet->vnet_sockcnt--; 463 #endif 464 mtx_unlock(&so_global_mtx); 465 #ifdef MAC 466 mac_socket_destroy(so); 467 #endif 468 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE); 469 470 crfree(so->so_cred); 471 khelp_destroy_osd(&so->osd); 472 if (SOLISTENING(so)) { 473 if (so->sol_accept_filter != NULL) 474 accept_filt_setopt(so, NULL); 475 } else { 476 if (so->so_rcv.sb_hiwat) 477 (void)chgsbsize(so->so_cred->cr_uidinfo, 478 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 479 if (so->so_snd.sb_hiwat) 480 (void)chgsbsize(so->so_cred->cr_uidinfo, 481 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 482 sx_destroy(&so->so_snd.sb_sx); 483 sx_destroy(&so->so_rcv.sb_sx); 484 SOCKBUF_LOCK_DESTROY(&so->so_snd); 485 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 486 } 487 mtx_destroy(&so->so_lock); 488 uma_zfree(socket_zone, so); 489 } 490 491 /* 492 * socreate returns a socket with a ref count of 1. The socket should be 493 * closed with soclose(). 494 */ 495 int 496 socreate(int dom, struct socket **aso, int type, int proto, 497 struct ucred *cred, struct thread *td) 498 { 499 struct protosw *prp; 500 struct socket *so; 501 int error; 502 503 if (proto) 504 prp = pffindproto(dom, proto, type); 505 else 506 prp = pffindtype(dom, type); 507 508 if (prp == NULL) { 509 /* No support for domain. */ 510 if (pffinddomain(dom) == NULL) 511 return (EAFNOSUPPORT); 512 /* No support for socket type. */ 513 if (proto == 0 && type != 0) 514 return (EPROTOTYPE); 515 return (EPROTONOSUPPORT); 516 } 517 if (prp->pr_usrreqs->pru_attach == NULL || 518 prp->pr_usrreqs->pru_attach == pru_attach_notsupp) 519 return (EPROTONOSUPPORT); 520 521 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0) 522 return (EPROTONOSUPPORT); 523 524 if (prp->pr_type != type) 525 return (EPROTOTYPE); 526 so = soalloc(CRED_TO_VNET(cred)); 527 if (so == NULL) 528 return (ENOBUFS); 529 530 so->so_type = type; 531 so->so_cred = crhold(cred); 532 if ((prp->pr_domain->dom_family == PF_INET) || 533 (prp->pr_domain->dom_family == PF_INET6) || 534 (prp->pr_domain->dom_family == PF_ROUTE)) 535 so->so_fibnum = td->td_proc->p_fibnum; 536 else 537 so->so_fibnum = 0; 538 so->so_proto = prp; 539 #ifdef MAC 540 mac_socket_create(cred, so); 541 #endif 542 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 543 so_rdknl_assert_locked, so_rdknl_assert_unlocked); 544 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 545 so_wrknl_assert_locked, so_wrknl_assert_unlocked); 546 /* 547 * Auto-sizing of socket buffers is managed by the protocols and 548 * the appropriate flags must be set in the pru_attach function. 549 */ 550 CURVNET_SET(so->so_vnet); 551 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); 552 CURVNET_RESTORE(); 553 if (error) { 554 sodealloc(so); 555 return (error); 556 } 557 soref(so); 558 *aso = so; 559 return (0); 560 } 561 562 #ifdef REGRESSION 563 static int regression_sonewconn_earlytest = 1; 564 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW, 565 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test"); 566 #endif 567 568 /* 569 * When an attempt at a new connection is noted on a socket which accepts 570 * connections, sonewconn is called. If the connection is possible (subject 571 * to space constraints, etc.) then we allocate a new structure, properly 572 * linked into the data structure of the original socket, and return this. 573 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED. 574 * 575 * Note: the ref count on the socket is 0 on return. 576 */ 577 struct socket * 578 sonewconn(struct socket *head, int connstatus) 579 { 580 static struct timeval lastover; 581 static struct timeval overinterval = { 60, 0 }; 582 static int overcount; 583 584 struct socket *so; 585 u_int over; 586 587 SOLISTEN_LOCK(head); 588 over = (head->sol_qlen > 3 * head->sol_qlimit / 2); 589 SOLISTEN_UNLOCK(head); 590 #ifdef REGRESSION 591 if (regression_sonewconn_earlytest && over) { 592 #else 593 if (over) { 594 #endif 595 overcount++; 596 597 if (ratecheck(&lastover, &overinterval)) { 598 log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: " 599 "%i already in queue awaiting acceptance " 600 "(%d occurrences)\n", 601 __func__, head->so_pcb, head->sol_qlen, overcount); 602 603 overcount = 0; 604 } 605 606 return (NULL); 607 } 608 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL", 609 __func__, head)); 610 so = soalloc(head->so_vnet); 611 if (so == NULL) { 612 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 613 "limit reached or out of memory\n", 614 __func__, head->so_pcb); 615 return (NULL); 616 } 617 so->so_listen = head; 618 so->so_type = head->so_type; 619 so->so_linger = head->so_linger; 620 so->so_state = head->so_state | SS_NOFDREF; 621 so->so_fibnum = head->so_fibnum; 622 so->so_proto = head->so_proto; 623 so->so_cred = crhold(head->so_cred); 624 #ifdef MAC 625 mac_socket_newconn(head, so); 626 #endif 627 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 628 so_rdknl_assert_locked, so_rdknl_assert_unlocked); 629 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 630 so_wrknl_assert_locked, so_wrknl_assert_unlocked); 631 VNET_SO_ASSERT(head); 632 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) { 633 sodealloc(so); 634 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 635 __func__, head->so_pcb); 636 return (NULL); 637 } 638 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 639 sodealloc(so); 640 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 641 __func__, head->so_pcb); 642 return (NULL); 643 } 644 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat; 645 so->so_snd.sb_lowat = head->sol_sbsnd_lowat; 646 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo; 647 so->so_snd.sb_timeo = head->sol_sbsnd_timeo; 648 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE; 649 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE; 650 651 SOLISTEN_LOCK(head); 652 if (head->sol_accept_filter != NULL) 653 connstatus = 0; 654 so->so_state |= connstatus; 655 so->so_options = head->so_options & ~SO_ACCEPTCONN; 656 soref(head); /* A socket on (in)complete queue refs head. */ 657 if (connstatus) { 658 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 659 so->so_qstate = SQ_COMP; 660 head->sol_qlen++; 661 solisten_wakeup(head); /* unlocks */ 662 } else { 663 /* 664 * Keep removing sockets from the head until there's room for 665 * us to insert on the tail. In pre-locking revisions, this 666 * was a simple if(), but as we could be racing with other 667 * threads and soabort() requires dropping locks, we must 668 * loop waiting for the condition to be true. 669 */ 670 while (head->sol_incqlen > head->sol_qlimit) { 671 struct socket *sp; 672 673 sp = TAILQ_FIRST(&head->sol_incomp); 674 TAILQ_REMOVE(&head->sol_incomp, sp, so_list); 675 head->sol_incqlen--; 676 SOCK_LOCK(sp); 677 sp->so_qstate = SQ_NONE; 678 sp->so_listen = NULL; 679 SOCK_UNLOCK(sp); 680 sorele(head); /* does SOLISTEN_UNLOCK, head stays */ 681 soabort(sp); 682 SOLISTEN_LOCK(head); 683 } 684 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list); 685 so->so_qstate = SQ_INCOMP; 686 head->sol_incqlen++; 687 SOLISTEN_UNLOCK(head); 688 } 689 return (so); 690 } 691 692 #ifdef SCTP 693 /* 694 * Socket part of sctp_peeloff(). Detach a new socket from an 695 * association. The new socket is returned with a reference. 696 */ 697 struct socket * 698 sopeeloff(struct socket *head) 699 { 700 struct socket *so; 701 702 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p", 703 __func__, __LINE__, head)); 704 so = soalloc(head->so_vnet); 705 if (so == NULL) { 706 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 707 "limit reached or out of memory\n", 708 __func__, head->so_pcb); 709 return (NULL); 710 } 711 so->so_type = head->so_type; 712 so->so_options = head->so_options; 713 so->so_linger = head->so_linger; 714 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED; 715 so->so_fibnum = head->so_fibnum; 716 so->so_proto = head->so_proto; 717 so->so_cred = crhold(head->so_cred); 718 #ifdef MAC 719 mac_socket_newconn(head, so); 720 #endif 721 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 722 so_rdknl_assert_locked, so_rdknl_assert_unlocked); 723 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 724 so_wrknl_assert_locked, so_wrknl_assert_unlocked); 725 VNET_SO_ASSERT(head); 726 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { 727 sodealloc(so); 728 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 729 __func__, head->so_pcb); 730 return (NULL); 731 } 732 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 733 sodealloc(so); 734 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 735 __func__, head->so_pcb); 736 return (NULL); 737 } 738 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 739 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 740 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 741 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 742 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; 743 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; 744 745 soref(so); 746 747 return (so); 748 } 749 #endif /* SCTP */ 750 751 int 752 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 753 { 754 int error; 755 756 CURVNET_SET(so->so_vnet); 757 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td); 758 CURVNET_RESTORE(); 759 return (error); 760 } 761 762 int 763 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 764 { 765 int error; 766 767 CURVNET_SET(so->so_vnet); 768 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td); 769 CURVNET_RESTORE(); 770 return (error); 771 } 772 773 /* 774 * solisten() transitions a socket from a non-listening state to a listening 775 * state, but can also be used to update the listen queue depth on an 776 * existing listen socket. The protocol will call back into the sockets 777 * layer using solisten_proto_check() and solisten_proto() to check and set 778 * socket-layer listen state. Call backs are used so that the protocol can 779 * acquire both protocol and socket layer locks in whatever order is required 780 * by the protocol. 781 * 782 * Protocol implementors are advised to hold the socket lock across the 783 * socket-layer test and set to avoid races at the socket layer. 784 */ 785 int 786 solisten(struct socket *so, int backlog, struct thread *td) 787 { 788 int error; 789 790 CURVNET_SET(so->so_vnet); 791 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td); 792 CURVNET_RESTORE(); 793 return (error); 794 } 795 796 int 797 solisten_proto_check(struct socket *so) 798 { 799 800 SOCK_LOCK_ASSERT(so); 801 802 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 803 SS_ISDISCONNECTING)) 804 return (EINVAL); 805 return (0); 806 } 807 808 void 809 solisten_proto(struct socket *so, int backlog) 810 { 811 int sbrcv_lowat, sbsnd_lowat; 812 u_int sbrcv_hiwat, sbsnd_hiwat; 813 short sbrcv_flags, sbsnd_flags; 814 sbintime_t sbrcv_timeo, sbsnd_timeo; 815 816 SOCK_LOCK_ASSERT(so); 817 818 if (SOLISTENING(so)) 819 goto listening; 820 821 /* 822 * Change this socket to listening state. 823 */ 824 sbrcv_lowat = so->so_rcv.sb_lowat; 825 sbsnd_lowat = so->so_snd.sb_lowat; 826 sbrcv_hiwat = so->so_rcv.sb_hiwat; 827 sbsnd_hiwat = so->so_snd.sb_hiwat; 828 sbrcv_flags = so->so_rcv.sb_flags; 829 sbsnd_flags = so->so_snd.sb_flags; 830 sbrcv_timeo = so->so_rcv.sb_timeo; 831 sbsnd_timeo = so->so_snd.sb_timeo; 832 833 sbdestroy(&so->so_snd, so); 834 sbdestroy(&so->so_rcv, so); 835 sx_destroy(&so->so_snd.sb_sx); 836 sx_destroy(&so->so_rcv.sb_sx); 837 SOCKBUF_LOCK_DESTROY(&so->so_snd); 838 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 839 840 #ifdef INVARIANTS 841 bzero(&so->so_rcv, 842 sizeof(struct socket) - offsetof(struct socket, so_rcv)); 843 #endif 844 845 so->sol_sbrcv_lowat = sbrcv_lowat; 846 so->sol_sbsnd_lowat = sbsnd_lowat; 847 so->sol_sbrcv_hiwat = sbrcv_hiwat; 848 so->sol_sbsnd_hiwat = sbsnd_hiwat; 849 so->sol_sbrcv_flags = sbrcv_flags; 850 so->sol_sbsnd_flags = sbsnd_flags; 851 so->sol_sbrcv_timeo = sbrcv_timeo; 852 so->sol_sbsnd_timeo = sbsnd_timeo; 853 854 so->sol_qlen = so->sol_incqlen = 0; 855 TAILQ_INIT(&so->sol_incomp); 856 TAILQ_INIT(&so->sol_comp); 857 858 so->sol_accept_filter = NULL; 859 so->sol_accept_filter_arg = NULL; 860 so->sol_accept_filter_str = NULL; 861 862 so->sol_upcall = NULL; 863 so->sol_upcallarg = NULL; 864 865 so->so_options |= SO_ACCEPTCONN; 866 867 listening: 868 if (backlog < 0 || backlog > somaxconn) 869 backlog = somaxconn; 870 so->sol_qlimit = backlog; 871 } 872 873 /* 874 * Wakeup listeners/subsystems once we have a complete connection. 875 * Enters with lock, returns unlocked. 876 */ 877 void 878 solisten_wakeup(struct socket *sol) 879 { 880 881 if (sol->sol_upcall != NULL) 882 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT); 883 else { 884 selwakeuppri(&sol->so_rdsel, PSOCK); 885 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0); 886 } 887 SOLISTEN_UNLOCK(sol); 888 wakeup_one(&sol->sol_comp); 889 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL) 890 pgsigio(&sol->so_sigio, SIGIO, 0); 891 } 892 893 /* 894 * Return single connection off a listening socket queue. Main consumer of 895 * the function is kern_accept4(). Some modules, that do their own accept 896 * management also use the function. 897 * 898 * Listening socket must be locked on entry and is returned unlocked on 899 * return. 900 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT. 901 */ 902 int 903 solisten_dequeue(struct socket *head, struct socket **ret, int flags) 904 { 905 struct socket *so; 906 int error; 907 908 SOLISTEN_LOCK_ASSERT(head); 909 910 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) && 911 head->so_error == 0) { 912 error = msleep(&head->sol_comp, &head->so_lock, PSOCK | PCATCH, 913 "accept", 0); 914 if (error != 0) { 915 SOLISTEN_UNLOCK(head); 916 return (error); 917 } 918 } 919 if (head->so_error) { 920 error = head->so_error; 921 head->so_error = 0; 922 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp)) 923 error = EWOULDBLOCK; 924 else 925 error = 0; 926 if (error) { 927 SOLISTEN_UNLOCK(head); 928 return (error); 929 } 930 so = TAILQ_FIRST(&head->sol_comp); 931 SOCK_LOCK(so); 932 KASSERT(so->so_qstate == SQ_COMP, 933 ("%s: so %p not SQ_COMP", __func__, so)); 934 soref(so); 935 head->sol_qlen--; 936 so->so_qstate = SQ_NONE; 937 so->so_listen = NULL; 938 TAILQ_REMOVE(&head->sol_comp, so, so_list); 939 if (flags & ACCEPT4_INHERIT) 940 so->so_state |= (head->so_state & SS_NBIO); 941 else 942 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0; 943 SOCK_UNLOCK(so); 944 sorele(head); 945 946 *ret = so; 947 return (0); 948 } 949 950 /* 951 * Evaluate the reference count and named references on a socket; if no 952 * references remain, free it. This should be called whenever a reference is 953 * released, such as in sorele(), but also when named reference flags are 954 * cleared in socket or protocol code. 955 * 956 * sofree() will free the socket if: 957 * 958 * - There are no outstanding file descriptor references or related consumers 959 * (so_count == 0). 960 * 961 * - The socket has been closed by user space, if ever open (SS_NOFDREF). 962 * 963 * - The protocol does not have an outstanding strong reference on the socket 964 * (SS_PROTOREF). 965 * 966 * - The socket is not in a completed connection queue, so a process has been 967 * notified that it is present. If it is removed, the user process may 968 * block in accept() despite select() saying the socket was ready. 969 */ 970 void 971 sofree(struct socket *so) 972 { 973 struct protosw *pr = so->so_proto; 974 975 SOCK_LOCK_ASSERT(so); 976 977 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 || 978 (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) { 979 SOCK_UNLOCK(so); 980 return; 981 } 982 983 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) { 984 struct socket *sol; 985 986 sol = so->so_listen; 987 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so)); 988 989 /* 990 * To solve race between close of a listening socket and 991 * a socket on its incomplete queue, we need to lock both. 992 * The order is first listening socket, then regular. 993 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this 994 * function and the listening socket are the only pointers 995 * to so. To preserve so and sol, we reference both and then 996 * relock. 997 * After relock the socket may not move to so_comp since it 998 * doesn't have PCB already, but it may be removed from 999 * so_incomp. If that happens, we share responsiblity on 1000 * freeing the socket, but soclose() has already removed 1001 * it from queue. 1002 */ 1003 soref(sol); 1004 soref(so); 1005 SOCK_UNLOCK(so); 1006 SOLISTEN_LOCK(sol); 1007 SOCK_LOCK(so); 1008 if (so->so_qstate == SQ_INCOMP) { 1009 KASSERT(so->so_listen == sol, 1010 ("%s: so %p migrated out of sol %p", 1011 __func__, so, sol)); 1012 TAILQ_REMOVE(&sol->sol_incomp, so, so_list); 1013 sol->sol_incqlen--; 1014 /* This is guarenteed not to be the last. */ 1015 refcount_release(&sol->so_count); 1016 so->so_qstate = SQ_NONE; 1017 so->so_listen = NULL; 1018 } else 1019 KASSERT(so->so_listen == NULL, 1020 ("%s: so %p not on (in)comp with so_listen", 1021 __func__, so)); 1022 sorele(sol); 1023 KASSERT(so->so_count == 1, 1024 ("%s: so %p count %u", __func__, so, so->so_count)); 1025 so->so_count = 0; 1026 } 1027 if (SOLISTENING(so)) 1028 so->so_error = ECONNABORTED; 1029 SOCK_UNLOCK(so); 1030 1031 if (so->so_dtor != NULL) 1032 so->so_dtor(so); 1033 1034 VNET_SO_ASSERT(so); 1035 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 1036 (*pr->pr_domain->dom_dispose)(so); 1037 if (pr->pr_usrreqs->pru_detach != NULL) 1038 (*pr->pr_usrreqs->pru_detach)(so); 1039 1040 /* 1041 * From this point on, we assume that no other references to this 1042 * socket exist anywhere else in the stack. Therefore, no locks need 1043 * to be acquired or held. 1044 * 1045 * We used to do a lot of socket buffer and socket locking here, as 1046 * well as invoke sorflush() and perform wakeups. The direct call to 1047 * dom_dispose() and sbdestroy() are an inlining of what was 1048 * necessary from sorflush(). 1049 * 1050 * Notice that the socket buffer and kqueue state are torn down 1051 * before calling pru_detach. This means that protocols shold not 1052 * assume they can perform socket wakeups, etc, in their detach code. 1053 */ 1054 if (!SOLISTENING(so)) { 1055 sbdestroy(&so->so_snd, so); 1056 sbdestroy(&so->so_rcv, so); 1057 } 1058 seldrain(&so->so_rdsel); 1059 seldrain(&so->so_wrsel); 1060 knlist_destroy(&so->so_rdsel.si_note); 1061 knlist_destroy(&so->so_wrsel.si_note); 1062 sodealloc(so); 1063 } 1064 1065 /* 1066 * Close a socket on last file table reference removal. Initiate disconnect 1067 * if connected. Free socket when disconnect complete. 1068 * 1069 * This function will sorele() the socket. Note that soclose() may be called 1070 * prior to the ref count reaching zero. The actual socket structure will 1071 * not be freed until the ref count reaches zero. 1072 */ 1073 int 1074 soclose(struct socket *so) 1075 { 1076 struct accept_queue lqueue; 1077 bool listening; 1078 int error = 0; 1079 1080 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter")); 1081 1082 CURVNET_SET(so->so_vnet); 1083 funsetown(&so->so_sigio); 1084 if (so->so_state & SS_ISCONNECTED) { 1085 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 1086 error = sodisconnect(so); 1087 if (error) { 1088 if (error == ENOTCONN) 1089 error = 0; 1090 goto drop; 1091 } 1092 } 1093 if (so->so_options & SO_LINGER) { 1094 if ((so->so_state & SS_ISDISCONNECTING) && 1095 (so->so_state & SS_NBIO)) 1096 goto drop; 1097 while (so->so_state & SS_ISCONNECTED) { 1098 error = tsleep(&so->so_timeo, 1099 PSOCK | PCATCH, "soclos", 1100 so->so_linger * hz); 1101 if (error) 1102 break; 1103 } 1104 } 1105 } 1106 1107 drop: 1108 if (so->so_proto->pr_usrreqs->pru_close != NULL) 1109 (*so->so_proto->pr_usrreqs->pru_close)(so); 1110 1111 SOCK_LOCK(so); 1112 if ((listening = (so->so_options & SO_ACCEPTCONN))) { 1113 struct socket *sp; 1114 1115 TAILQ_INIT(&lqueue); 1116 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list); 1117 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list); 1118 1119 so->sol_qlen = so->sol_incqlen = 0; 1120 1121 TAILQ_FOREACH(sp, &lqueue, so_list) { 1122 SOCK_LOCK(sp); 1123 sp->so_qstate = SQ_NONE; 1124 sp->so_listen = NULL; 1125 SOCK_UNLOCK(sp); 1126 /* Guaranteed not to be the last. */ 1127 refcount_release(&so->so_count); 1128 } 1129 } 1130 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF")); 1131 so->so_state |= SS_NOFDREF; 1132 sorele(so); 1133 if (listening) { 1134 struct socket *sp; 1135 1136 TAILQ_FOREACH(sp, &lqueue, so_list) { 1137 SOCK_LOCK(sp); 1138 if (sp->so_count == 0) { 1139 SOCK_UNLOCK(sp); 1140 soabort(sp); 1141 } else 1142 /* sp is now in sofree() */ 1143 SOCK_UNLOCK(sp); 1144 } 1145 } 1146 CURVNET_RESTORE(); 1147 return (error); 1148 } 1149 1150 /* 1151 * soabort() is used to abruptly tear down a connection, such as when a 1152 * resource limit is reached (listen queue depth exceeded), or if a listen 1153 * socket is closed while there are sockets waiting to be accepted. 1154 * 1155 * This interface is tricky, because it is called on an unreferenced socket, 1156 * and must be called only by a thread that has actually removed the socket 1157 * from the listen queue it was on, or races with other threads are risked. 1158 * 1159 * This interface will call into the protocol code, so must not be called 1160 * with any socket locks held. Protocols do call it while holding their own 1161 * recursible protocol mutexes, but this is something that should be subject 1162 * to review in the future. 1163 */ 1164 void 1165 soabort(struct socket *so) 1166 { 1167 1168 /* 1169 * In as much as is possible, assert that no references to this 1170 * socket are held. This is not quite the same as asserting that the 1171 * current thread is responsible for arranging for no references, but 1172 * is as close as we can get for now. 1173 */ 1174 KASSERT(so->so_count == 0, ("soabort: so_count")); 1175 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF")); 1176 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF")); 1177 VNET_SO_ASSERT(so); 1178 1179 if (so->so_proto->pr_usrreqs->pru_abort != NULL) 1180 (*so->so_proto->pr_usrreqs->pru_abort)(so); 1181 SOCK_LOCK(so); 1182 sofree(so); 1183 } 1184 1185 int 1186 soaccept(struct socket *so, struct sockaddr **nam) 1187 { 1188 int error; 1189 1190 SOCK_LOCK(so); 1191 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF")); 1192 so->so_state &= ~SS_NOFDREF; 1193 SOCK_UNLOCK(so); 1194 1195 CURVNET_SET(so->so_vnet); 1196 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); 1197 CURVNET_RESTORE(); 1198 return (error); 1199 } 1200 1201 int 1202 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 1203 { 1204 1205 return (soconnectat(AT_FDCWD, so, nam, td)); 1206 } 1207 1208 int 1209 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 1210 { 1211 int error; 1212 1213 if (so->so_options & SO_ACCEPTCONN) 1214 return (EOPNOTSUPP); 1215 1216 CURVNET_SET(so->so_vnet); 1217 /* 1218 * If protocol is connection-based, can only connect once. 1219 * Otherwise, if connected, try to disconnect first. This allows 1220 * user to disconnect by connecting to, e.g., a null address. 1221 */ 1222 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 1223 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 1224 (error = sodisconnect(so)))) { 1225 error = EISCONN; 1226 } else { 1227 /* 1228 * Prevent accumulated error from previous connection from 1229 * biting us. 1230 */ 1231 so->so_error = 0; 1232 if (fd == AT_FDCWD) { 1233 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, 1234 nam, td); 1235 } else { 1236 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd, 1237 so, nam, td); 1238 } 1239 } 1240 CURVNET_RESTORE(); 1241 1242 return (error); 1243 } 1244 1245 int 1246 soconnect2(struct socket *so1, struct socket *so2) 1247 { 1248 int error; 1249 1250 CURVNET_SET(so1->so_vnet); 1251 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); 1252 CURVNET_RESTORE(); 1253 return (error); 1254 } 1255 1256 int 1257 sodisconnect(struct socket *so) 1258 { 1259 int error; 1260 1261 if ((so->so_state & SS_ISCONNECTED) == 0) 1262 return (ENOTCONN); 1263 if (so->so_state & SS_ISDISCONNECTING) 1264 return (EALREADY); 1265 VNET_SO_ASSERT(so); 1266 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 1267 return (error); 1268 } 1269 1270 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) 1271 1272 int 1273 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1274 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1275 { 1276 long space; 1277 ssize_t resid; 1278 int clen = 0, error, dontroute; 1279 1280 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 1281 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 1282 ("sosend_dgram: !PR_ATOMIC")); 1283 1284 if (uio != NULL) 1285 resid = uio->uio_resid; 1286 else 1287 resid = top->m_pkthdr.len; 1288 /* 1289 * In theory resid should be unsigned. However, space must be 1290 * signed, as it might be less than 0 if we over-committed, and we 1291 * must use a signed comparison of space and resid. On the other 1292 * hand, a negative resid causes us to loop sending 0-length 1293 * segments to the protocol. 1294 */ 1295 if (resid < 0) { 1296 error = EINVAL; 1297 goto out; 1298 } 1299 1300 dontroute = 1301 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 1302 if (td != NULL) 1303 td->td_ru.ru_msgsnd++; 1304 if (control != NULL) 1305 clen = control->m_len; 1306 1307 SOCKBUF_LOCK(&so->so_snd); 1308 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1309 SOCKBUF_UNLOCK(&so->so_snd); 1310 error = EPIPE; 1311 goto out; 1312 } 1313 if (so->so_error) { 1314 error = so->so_error; 1315 so->so_error = 0; 1316 SOCKBUF_UNLOCK(&so->so_snd); 1317 goto out; 1318 } 1319 if ((so->so_state & SS_ISCONNECTED) == 0) { 1320 /* 1321 * `sendto' and `sendmsg' is allowed on a connection-based 1322 * socket if it supports implied connect. Return ENOTCONN if 1323 * not connected and no address is supplied. 1324 */ 1325 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1326 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1327 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1328 !(resid == 0 && clen != 0)) { 1329 SOCKBUF_UNLOCK(&so->so_snd); 1330 error = ENOTCONN; 1331 goto out; 1332 } 1333 } else if (addr == NULL) { 1334 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1335 error = ENOTCONN; 1336 else 1337 error = EDESTADDRREQ; 1338 SOCKBUF_UNLOCK(&so->so_snd); 1339 goto out; 1340 } 1341 } 1342 1343 /* 1344 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1345 * problem and need fixing. 1346 */ 1347 space = sbspace(&so->so_snd); 1348 if (flags & MSG_OOB) 1349 space += 1024; 1350 space -= clen; 1351 SOCKBUF_UNLOCK(&so->so_snd); 1352 if (resid > space) { 1353 error = EMSGSIZE; 1354 goto out; 1355 } 1356 if (uio == NULL) { 1357 resid = 0; 1358 if (flags & MSG_EOR) 1359 top->m_flags |= M_EOR; 1360 } else { 1361 /* 1362 * Copy the data from userland into a mbuf chain. 1363 * If no data is to be copied in, a single empty mbuf 1364 * is returned. 1365 */ 1366 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1367 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1368 if (top == NULL) { 1369 error = EFAULT; /* only possible error */ 1370 goto out; 1371 } 1372 space -= resid - uio->uio_resid; 1373 resid = uio->uio_resid; 1374 } 1375 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1376 /* 1377 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1378 * than with. 1379 */ 1380 if (dontroute) { 1381 SOCK_LOCK(so); 1382 so->so_options |= SO_DONTROUTE; 1383 SOCK_UNLOCK(so); 1384 } 1385 /* 1386 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1387 * of date. We could have received a reset packet in an interrupt or 1388 * maybe we slept while doing page faults in uiomove() etc. We could 1389 * probably recheck again inside the locking protection here, but 1390 * there are probably other places that this also happens. We must 1391 * rethink this. 1392 */ 1393 VNET_SO_ASSERT(so); 1394 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1395 (flags & MSG_OOB) ? PRUS_OOB : 1396 /* 1397 * If the user set MSG_EOF, the protocol understands this flag and 1398 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1399 */ 1400 ((flags & MSG_EOF) && 1401 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1402 (resid <= 0)) ? 1403 PRUS_EOF : 1404 /* If there is more to send set PRUS_MORETOCOME */ 1405 (flags & MSG_MORETOCOME) || 1406 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1407 top, addr, control, td); 1408 if (dontroute) { 1409 SOCK_LOCK(so); 1410 so->so_options &= ~SO_DONTROUTE; 1411 SOCK_UNLOCK(so); 1412 } 1413 clen = 0; 1414 control = NULL; 1415 top = NULL; 1416 out: 1417 if (top != NULL) 1418 m_freem(top); 1419 if (control != NULL) 1420 m_freem(control); 1421 return (error); 1422 } 1423 1424 /* 1425 * Send on a socket. If send must go all at once and message is larger than 1426 * send buffering, then hard error. Lock against other senders. If must go 1427 * all at once and not enough room now, then inform user that this would 1428 * block and do nothing. Otherwise, if nonblocking, send as much as 1429 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1430 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1431 * in mbuf chain must be small enough to send all at once. 1432 * 1433 * Returns nonzero on error, timeout or signal; callers must check for short 1434 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1435 * on return. 1436 */ 1437 int 1438 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1439 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1440 { 1441 long space; 1442 ssize_t resid; 1443 int clen = 0, error, dontroute; 1444 int atomic = sosendallatonce(so) || top; 1445 1446 if (uio != NULL) 1447 resid = uio->uio_resid; 1448 else 1449 resid = top->m_pkthdr.len; 1450 /* 1451 * In theory resid should be unsigned. However, space must be 1452 * signed, as it might be less than 0 if we over-committed, and we 1453 * must use a signed comparison of space and resid. On the other 1454 * hand, a negative resid causes us to loop sending 0-length 1455 * segments to the protocol. 1456 * 1457 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1458 * type sockets since that's an error. 1459 */ 1460 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1461 error = EINVAL; 1462 goto out; 1463 } 1464 1465 dontroute = 1466 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1467 (so->so_proto->pr_flags & PR_ATOMIC); 1468 if (td != NULL) 1469 td->td_ru.ru_msgsnd++; 1470 if (control != NULL) 1471 clen = control->m_len; 1472 1473 error = sblock(&so->so_snd, SBLOCKWAIT(flags)); 1474 if (error) 1475 goto out; 1476 1477 restart: 1478 do { 1479 SOCKBUF_LOCK(&so->so_snd); 1480 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1481 SOCKBUF_UNLOCK(&so->so_snd); 1482 error = EPIPE; 1483 goto release; 1484 } 1485 if (so->so_error) { 1486 error = so->so_error; 1487 so->so_error = 0; 1488 SOCKBUF_UNLOCK(&so->so_snd); 1489 goto release; 1490 } 1491 if ((so->so_state & SS_ISCONNECTED) == 0) { 1492 /* 1493 * `sendto' and `sendmsg' is allowed on a connection- 1494 * based socket if it supports implied connect. 1495 * Return ENOTCONN if not connected and no address is 1496 * supplied. 1497 */ 1498 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1499 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1500 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1501 !(resid == 0 && clen != 0)) { 1502 SOCKBUF_UNLOCK(&so->so_snd); 1503 error = ENOTCONN; 1504 goto release; 1505 } 1506 } else if (addr == NULL) { 1507 SOCKBUF_UNLOCK(&so->so_snd); 1508 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1509 error = ENOTCONN; 1510 else 1511 error = EDESTADDRREQ; 1512 goto release; 1513 } 1514 } 1515 space = sbspace(&so->so_snd); 1516 if (flags & MSG_OOB) 1517 space += 1024; 1518 if ((atomic && resid > so->so_snd.sb_hiwat) || 1519 clen > so->so_snd.sb_hiwat) { 1520 SOCKBUF_UNLOCK(&so->so_snd); 1521 error = EMSGSIZE; 1522 goto release; 1523 } 1524 if (space < resid + clen && 1525 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1526 if ((so->so_state & SS_NBIO) || 1527 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) { 1528 SOCKBUF_UNLOCK(&so->so_snd); 1529 error = EWOULDBLOCK; 1530 goto release; 1531 } 1532 error = sbwait(&so->so_snd); 1533 SOCKBUF_UNLOCK(&so->so_snd); 1534 if (error) 1535 goto release; 1536 goto restart; 1537 } 1538 SOCKBUF_UNLOCK(&so->so_snd); 1539 space -= clen; 1540 do { 1541 if (uio == NULL) { 1542 resid = 0; 1543 if (flags & MSG_EOR) 1544 top->m_flags |= M_EOR; 1545 } else { 1546 /* 1547 * Copy the data from userland into a mbuf 1548 * chain. If resid is 0, which can happen 1549 * only if we have control to send, then 1550 * a single empty mbuf is returned. This 1551 * is a workaround to prevent protocol send 1552 * methods to panic. 1553 */ 1554 top = m_uiotombuf(uio, M_WAITOK, space, 1555 (atomic ? max_hdr : 0), 1556 (atomic ? M_PKTHDR : 0) | 1557 ((flags & MSG_EOR) ? M_EOR : 0)); 1558 if (top == NULL) { 1559 error = EFAULT; /* only possible error */ 1560 goto release; 1561 } 1562 space -= resid - uio->uio_resid; 1563 resid = uio->uio_resid; 1564 } 1565 if (dontroute) { 1566 SOCK_LOCK(so); 1567 so->so_options |= SO_DONTROUTE; 1568 SOCK_UNLOCK(so); 1569 } 1570 /* 1571 * XXX all the SBS_CANTSENDMORE checks previously 1572 * done could be out of date. We could have received 1573 * a reset packet in an interrupt or maybe we slept 1574 * while doing page faults in uiomove() etc. We 1575 * could probably recheck again inside the locking 1576 * protection here, but there are probably other 1577 * places that this also happens. We must rethink 1578 * this. 1579 */ 1580 VNET_SO_ASSERT(so); 1581 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1582 (flags & MSG_OOB) ? PRUS_OOB : 1583 /* 1584 * If the user set MSG_EOF, the protocol understands 1585 * this flag and nothing left to send then use 1586 * PRU_SEND_EOF instead of PRU_SEND. 1587 */ 1588 ((flags & MSG_EOF) && 1589 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1590 (resid <= 0)) ? 1591 PRUS_EOF : 1592 /* If there is more to send set PRUS_MORETOCOME. */ 1593 (flags & MSG_MORETOCOME) || 1594 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1595 top, addr, control, td); 1596 if (dontroute) { 1597 SOCK_LOCK(so); 1598 so->so_options &= ~SO_DONTROUTE; 1599 SOCK_UNLOCK(so); 1600 } 1601 clen = 0; 1602 control = NULL; 1603 top = NULL; 1604 if (error) 1605 goto release; 1606 } while (resid && space > 0); 1607 } while (resid); 1608 1609 release: 1610 sbunlock(&so->so_snd); 1611 out: 1612 if (top != NULL) 1613 m_freem(top); 1614 if (control != NULL) 1615 m_freem(control); 1616 return (error); 1617 } 1618 1619 int 1620 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1621 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1622 { 1623 int error; 1624 1625 CURVNET_SET(so->so_vnet); 1626 if (!SOLISTENING(so)) 1627 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, 1628 top, control, flags, td); 1629 else { 1630 m_freem(top); 1631 m_freem(control); 1632 error = ENOTCONN; 1633 } 1634 CURVNET_RESTORE(); 1635 return (error); 1636 } 1637 1638 /* 1639 * The part of soreceive() that implements reading non-inline out-of-band 1640 * data from a socket. For more complete comments, see soreceive(), from 1641 * which this code originated. 1642 * 1643 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1644 * unable to return an mbuf chain to the caller. 1645 */ 1646 static int 1647 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1648 { 1649 struct protosw *pr = so->so_proto; 1650 struct mbuf *m; 1651 int error; 1652 1653 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1654 VNET_SO_ASSERT(so); 1655 1656 m = m_get(M_WAITOK, MT_DATA); 1657 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 1658 if (error) 1659 goto bad; 1660 do { 1661 error = uiomove(mtod(m, void *), 1662 (int) min(uio->uio_resid, m->m_len), uio); 1663 m = m_free(m); 1664 } while (uio->uio_resid && error == 0 && m); 1665 bad: 1666 if (m != NULL) 1667 m_freem(m); 1668 return (error); 1669 } 1670 1671 /* 1672 * Following replacement or removal of the first mbuf on the first mbuf chain 1673 * of a socket buffer, push necessary state changes back into the socket 1674 * buffer so that other consumers see the values consistently. 'nextrecord' 1675 * is the callers locally stored value of the original value of 1676 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1677 * NOTE: 'nextrecord' may be NULL. 1678 */ 1679 static __inline void 1680 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1681 { 1682 1683 SOCKBUF_LOCK_ASSERT(sb); 1684 /* 1685 * First, update for the new value of nextrecord. If necessary, make 1686 * it the first record. 1687 */ 1688 if (sb->sb_mb != NULL) 1689 sb->sb_mb->m_nextpkt = nextrecord; 1690 else 1691 sb->sb_mb = nextrecord; 1692 1693 /* 1694 * Now update any dependent socket buffer fields to reflect the new 1695 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1696 * addition of a second clause that takes care of the case where 1697 * sb_mb has been updated, but remains the last record. 1698 */ 1699 if (sb->sb_mb == NULL) { 1700 sb->sb_mbtail = NULL; 1701 sb->sb_lastrecord = NULL; 1702 } else if (sb->sb_mb->m_nextpkt == NULL) 1703 sb->sb_lastrecord = sb->sb_mb; 1704 } 1705 1706 /* 1707 * Implement receive operations on a socket. We depend on the way that 1708 * records are added to the sockbuf by sbappend. In particular, each record 1709 * (mbufs linked through m_next) must begin with an address if the protocol 1710 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1711 * data, and then zero or more mbufs of data. In order to allow parallelism 1712 * between network receive and copying to user space, as well as avoid 1713 * sleeping with a mutex held, we release the socket buffer mutex during the 1714 * user space copy. Although the sockbuf is locked, new data may still be 1715 * appended, and thus we must maintain consistency of the sockbuf during that 1716 * time. 1717 * 1718 * The caller may receive the data as a single mbuf chain by supplying an 1719 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1720 * the count in uio_resid. 1721 */ 1722 int 1723 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1724 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1725 { 1726 struct mbuf *m, **mp; 1727 int flags, error, offset; 1728 ssize_t len; 1729 struct protosw *pr = so->so_proto; 1730 struct mbuf *nextrecord; 1731 int moff, type = 0; 1732 ssize_t orig_resid = uio->uio_resid; 1733 1734 mp = mp0; 1735 if (psa != NULL) 1736 *psa = NULL; 1737 if (controlp != NULL) 1738 *controlp = NULL; 1739 if (flagsp != NULL) 1740 flags = *flagsp &~ MSG_EOR; 1741 else 1742 flags = 0; 1743 if (flags & MSG_OOB) 1744 return (soreceive_rcvoob(so, uio, flags)); 1745 if (mp != NULL) 1746 *mp = NULL; 1747 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 1748 && uio->uio_resid) { 1749 VNET_SO_ASSERT(so); 1750 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 1751 } 1752 1753 error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); 1754 if (error) 1755 return (error); 1756 1757 restart: 1758 SOCKBUF_LOCK(&so->so_rcv); 1759 m = so->so_rcv.sb_mb; 1760 /* 1761 * If we have less data than requested, block awaiting more (subject 1762 * to any timeout) if: 1763 * 1. the current count is less than the low water mark, or 1764 * 2. MSG_DONTWAIT is not set 1765 */ 1766 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1767 sbavail(&so->so_rcv) < uio->uio_resid) && 1768 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && 1769 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 1770 KASSERT(m != NULL || !sbavail(&so->so_rcv), 1771 ("receive: m == %p sbavail == %u", 1772 m, sbavail(&so->so_rcv))); 1773 if (so->so_error) { 1774 if (m != NULL) 1775 goto dontblock; 1776 error = so->so_error; 1777 if ((flags & MSG_PEEK) == 0) 1778 so->so_error = 0; 1779 SOCKBUF_UNLOCK(&so->so_rcv); 1780 goto release; 1781 } 1782 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1783 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1784 if (m == NULL) { 1785 SOCKBUF_UNLOCK(&so->so_rcv); 1786 goto release; 1787 } else 1788 goto dontblock; 1789 } 1790 for (; m != NULL; m = m->m_next) 1791 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1792 m = so->so_rcv.sb_mb; 1793 goto dontblock; 1794 } 1795 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1796 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1797 SOCKBUF_UNLOCK(&so->so_rcv); 1798 error = ENOTCONN; 1799 goto release; 1800 } 1801 if (uio->uio_resid == 0) { 1802 SOCKBUF_UNLOCK(&so->so_rcv); 1803 goto release; 1804 } 1805 if ((so->so_state & SS_NBIO) || 1806 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1807 SOCKBUF_UNLOCK(&so->so_rcv); 1808 error = EWOULDBLOCK; 1809 goto release; 1810 } 1811 SBLASTRECORDCHK(&so->so_rcv); 1812 SBLASTMBUFCHK(&so->so_rcv); 1813 error = sbwait(&so->so_rcv); 1814 SOCKBUF_UNLOCK(&so->so_rcv); 1815 if (error) 1816 goto release; 1817 goto restart; 1818 } 1819 dontblock: 1820 /* 1821 * From this point onward, we maintain 'nextrecord' as a cache of the 1822 * pointer to the next record in the socket buffer. We must keep the 1823 * various socket buffer pointers and local stack versions of the 1824 * pointers in sync, pushing out modifications before dropping the 1825 * socket buffer mutex, and re-reading them when picking it up. 1826 * 1827 * Otherwise, we will race with the network stack appending new data 1828 * or records onto the socket buffer by using inconsistent/stale 1829 * versions of the field, possibly resulting in socket buffer 1830 * corruption. 1831 * 1832 * By holding the high-level sblock(), we prevent simultaneous 1833 * readers from pulling off the front of the socket buffer. 1834 */ 1835 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1836 if (uio->uio_td) 1837 uio->uio_td->td_ru.ru_msgrcv++; 1838 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 1839 SBLASTRECORDCHK(&so->so_rcv); 1840 SBLASTMBUFCHK(&so->so_rcv); 1841 nextrecord = m->m_nextpkt; 1842 if (pr->pr_flags & PR_ADDR) { 1843 KASSERT(m->m_type == MT_SONAME, 1844 ("m->m_type == %d", m->m_type)); 1845 orig_resid = 0; 1846 if (psa != NULL) 1847 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 1848 M_NOWAIT); 1849 if (flags & MSG_PEEK) { 1850 m = m->m_next; 1851 } else { 1852 sbfree(&so->so_rcv, m); 1853 so->so_rcv.sb_mb = m_free(m); 1854 m = so->so_rcv.sb_mb; 1855 sockbuf_pushsync(&so->so_rcv, nextrecord); 1856 } 1857 } 1858 1859 /* 1860 * Process one or more MT_CONTROL mbufs present before any data mbufs 1861 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 1862 * just copy the data; if !MSG_PEEK, we call into the protocol to 1863 * perform externalization (or freeing if controlp == NULL). 1864 */ 1865 if (m != NULL && m->m_type == MT_CONTROL) { 1866 struct mbuf *cm = NULL, *cmn; 1867 struct mbuf **cme = &cm; 1868 1869 do { 1870 if (flags & MSG_PEEK) { 1871 if (controlp != NULL) { 1872 *controlp = m_copym(m, 0, m->m_len, 1873 M_NOWAIT); 1874 controlp = &(*controlp)->m_next; 1875 } 1876 m = m->m_next; 1877 } else { 1878 sbfree(&so->so_rcv, m); 1879 so->so_rcv.sb_mb = m->m_next; 1880 m->m_next = NULL; 1881 *cme = m; 1882 cme = &(*cme)->m_next; 1883 m = so->so_rcv.sb_mb; 1884 } 1885 } while (m != NULL && m->m_type == MT_CONTROL); 1886 if ((flags & MSG_PEEK) == 0) 1887 sockbuf_pushsync(&so->so_rcv, nextrecord); 1888 while (cm != NULL) { 1889 cmn = cm->m_next; 1890 cm->m_next = NULL; 1891 if (pr->pr_domain->dom_externalize != NULL) { 1892 SOCKBUF_UNLOCK(&so->so_rcv); 1893 VNET_SO_ASSERT(so); 1894 error = (*pr->pr_domain->dom_externalize) 1895 (cm, controlp, flags); 1896 SOCKBUF_LOCK(&so->so_rcv); 1897 } else if (controlp != NULL) 1898 *controlp = cm; 1899 else 1900 m_freem(cm); 1901 if (controlp != NULL) { 1902 orig_resid = 0; 1903 while (*controlp != NULL) 1904 controlp = &(*controlp)->m_next; 1905 } 1906 cm = cmn; 1907 } 1908 if (m != NULL) 1909 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 1910 else 1911 nextrecord = so->so_rcv.sb_mb; 1912 orig_resid = 0; 1913 } 1914 if (m != NULL) { 1915 if ((flags & MSG_PEEK) == 0) { 1916 KASSERT(m->m_nextpkt == nextrecord, 1917 ("soreceive: post-control, nextrecord !sync")); 1918 if (nextrecord == NULL) { 1919 KASSERT(so->so_rcv.sb_mb == m, 1920 ("soreceive: post-control, sb_mb!=m")); 1921 KASSERT(so->so_rcv.sb_lastrecord == m, 1922 ("soreceive: post-control, lastrecord!=m")); 1923 } 1924 } 1925 type = m->m_type; 1926 if (type == MT_OOBDATA) 1927 flags |= MSG_OOB; 1928 } else { 1929 if ((flags & MSG_PEEK) == 0) { 1930 KASSERT(so->so_rcv.sb_mb == nextrecord, 1931 ("soreceive: sb_mb != nextrecord")); 1932 if (so->so_rcv.sb_mb == NULL) { 1933 KASSERT(so->so_rcv.sb_lastrecord == NULL, 1934 ("soreceive: sb_lastercord != NULL")); 1935 } 1936 } 1937 } 1938 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1939 SBLASTRECORDCHK(&so->so_rcv); 1940 SBLASTMBUFCHK(&so->so_rcv); 1941 1942 /* 1943 * Now continue to read any data mbufs off of the head of the socket 1944 * buffer until the read request is satisfied. Note that 'type' is 1945 * used to store the type of any mbuf reads that have happened so far 1946 * such that soreceive() can stop reading if the type changes, which 1947 * causes soreceive() to return only one of regular data and inline 1948 * out-of-band data in a single socket receive operation. 1949 */ 1950 moff = 0; 1951 offset = 0; 1952 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 1953 && error == 0) { 1954 /* 1955 * If the type of mbuf has changed since the last mbuf 1956 * examined ('type'), end the receive operation. 1957 */ 1958 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1959 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 1960 if (type != m->m_type) 1961 break; 1962 } else if (type == MT_OOBDATA) 1963 break; 1964 else 1965 KASSERT(m->m_type == MT_DATA, 1966 ("m->m_type == %d", m->m_type)); 1967 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 1968 len = uio->uio_resid; 1969 if (so->so_oobmark && len > so->so_oobmark - offset) 1970 len = so->so_oobmark - offset; 1971 if (len > m->m_len - moff) 1972 len = m->m_len - moff; 1973 /* 1974 * If mp is set, just pass back the mbufs. Otherwise copy 1975 * them out via the uio, then free. Sockbuf must be 1976 * consistent here (points to current mbuf, it points to next 1977 * record) when we drop priority; we must note any additions 1978 * to the sockbuf when we block interrupts again. 1979 */ 1980 if (mp == NULL) { 1981 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1982 SBLASTRECORDCHK(&so->so_rcv); 1983 SBLASTMBUFCHK(&so->so_rcv); 1984 SOCKBUF_UNLOCK(&so->so_rcv); 1985 if ((m->m_flags & M_NOMAP) != 0) 1986 error = m_unmappedtouio(m, moff, uio, (int)len); 1987 else 1988 error = uiomove(mtod(m, char *) + moff, 1989 (int)len, uio); 1990 SOCKBUF_LOCK(&so->so_rcv); 1991 if (error) { 1992 /* 1993 * The MT_SONAME mbuf has already been removed 1994 * from the record, so it is necessary to 1995 * remove the data mbufs, if any, to preserve 1996 * the invariant in the case of PR_ADDR that 1997 * requires MT_SONAME mbufs at the head of 1998 * each record. 1999 */ 2000 if (pr->pr_flags & PR_ATOMIC && 2001 ((flags & MSG_PEEK) == 0)) 2002 (void)sbdroprecord_locked(&so->so_rcv); 2003 SOCKBUF_UNLOCK(&so->so_rcv); 2004 goto release; 2005 } 2006 } else 2007 uio->uio_resid -= len; 2008 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2009 if (len == m->m_len - moff) { 2010 if (m->m_flags & M_EOR) 2011 flags |= MSG_EOR; 2012 if (flags & MSG_PEEK) { 2013 m = m->m_next; 2014 moff = 0; 2015 } else { 2016 nextrecord = m->m_nextpkt; 2017 sbfree(&so->so_rcv, m); 2018 if (mp != NULL) { 2019 m->m_nextpkt = NULL; 2020 *mp = m; 2021 mp = &m->m_next; 2022 so->so_rcv.sb_mb = m = m->m_next; 2023 *mp = NULL; 2024 } else { 2025 so->so_rcv.sb_mb = m_free(m); 2026 m = so->so_rcv.sb_mb; 2027 } 2028 sockbuf_pushsync(&so->so_rcv, nextrecord); 2029 SBLASTRECORDCHK(&so->so_rcv); 2030 SBLASTMBUFCHK(&so->so_rcv); 2031 } 2032 } else { 2033 if (flags & MSG_PEEK) 2034 moff += len; 2035 else { 2036 if (mp != NULL) { 2037 if (flags & MSG_DONTWAIT) { 2038 *mp = m_copym(m, 0, len, 2039 M_NOWAIT); 2040 if (*mp == NULL) { 2041 /* 2042 * m_copym() couldn't 2043 * allocate an mbuf. 2044 * Adjust uio_resid back 2045 * (it was adjusted 2046 * down by len bytes, 2047 * which we didn't end 2048 * up "copying" over). 2049 */ 2050 uio->uio_resid += len; 2051 break; 2052 } 2053 } else { 2054 SOCKBUF_UNLOCK(&so->so_rcv); 2055 *mp = m_copym(m, 0, len, 2056 M_WAITOK); 2057 SOCKBUF_LOCK(&so->so_rcv); 2058 } 2059 } 2060 sbcut_locked(&so->so_rcv, len); 2061 } 2062 } 2063 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2064 if (so->so_oobmark) { 2065 if ((flags & MSG_PEEK) == 0) { 2066 so->so_oobmark -= len; 2067 if (so->so_oobmark == 0) { 2068 so->so_rcv.sb_state |= SBS_RCVATMARK; 2069 break; 2070 } 2071 } else { 2072 offset += len; 2073 if (offset == so->so_oobmark) 2074 break; 2075 } 2076 } 2077 if (flags & MSG_EOR) 2078 break; 2079 /* 2080 * If the MSG_WAITALL flag is set (for non-atomic socket), we 2081 * must not quit until "uio->uio_resid == 0" or an error 2082 * termination. If a signal/timeout occurs, return with a 2083 * short count but without error. Keep sockbuf locked 2084 * against other readers. 2085 */ 2086 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 2087 !sosendallatonce(so) && nextrecord == NULL) { 2088 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2089 if (so->so_error || 2090 so->so_rcv.sb_state & SBS_CANTRCVMORE) 2091 break; 2092 /* 2093 * Notify the protocol that some data has been 2094 * drained before blocking. 2095 */ 2096 if (pr->pr_flags & PR_WANTRCVD) { 2097 SOCKBUF_UNLOCK(&so->so_rcv); 2098 VNET_SO_ASSERT(so); 2099 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2100 SOCKBUF_LOCK(&so->so_rcv); 2101 } 2102 SBLASTRECORDCHK(&so->so_rcv); 2103 SBLASTMBUFCHK(&so->so_rcv); 2104 /* 2105 * We could receive some data while was notifying 2106 * the protocol. Skip blocking in this case. 2107 */ 2108 if (so->so_rcv.sb_mb == NULL) { 2109 error = sbwait(&so->so_rcv); 2110 if (error) { 2111 SOCKBUF_UNLOCK(&so->so_rcv); 2112 goto release; 2113 } 2114 } 2115 m = so->so_rcv.sb_mb; 2116 if (m != NULL) 2117 nextrecord = m->m_nextpkt; 2118 } 2119 } 2120 2121 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2122 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 2123 flags |= MSG_TRUNC; 2124 if ((flags & MSG_PEEK) == 0) 2125 (void) sbdroprecord_locked(&so->so_rcv); 2126 } 2127 if ((flags & MSG_PEEK) == 0) { 2128 if (m == NULL) { 2129 /* 2130 * First part is an inline SB_EMPTY_FIXUP(). Second 2131 * part makes sure sb_lastrecord is up-to-date if 2132 * there is still data in the socket buffer. 2133 */ 2134 so->so_rcv.sb_mb = nextrecord; 2135 if (so->so_rcv.sb_mb == NULL) { 2136 so->so_rcv.sb_mbtail = NULL; 2137 so->so_rcv.sb_lastrecord = NULL; 2138 } else if (nextrecord->m_nextpkt == NULL) 2139 so->so_rcv.sb_lastrecord = nextrecord; 2140 } 2141 SBLASTRECORDCHK(&so->so_rcv); 2142 SBLASTMBUFCHK(&so->so_rcv); 2143 /* 2144 * If soreceive() is being done from the socket callback, 2145 * then don't need to generate ACK to peer to update window, 2146 * since ACK will be generated on return to TCP. 2147 */ 2148 if (!(flags & MSG_SOCALLBCK) && 2149 (pr->pr_flags & PR_WANTRCVD)) { 2150 SOCKBUF_UNLOCK(&so->so_rcv); 2151 VNET_SO_ASSERT(so); 2152 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2153 SOCKBUF_LOCK(&so->so_rcv); 2154 } 2155 } 2156 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2157 if (orig_resid == uio->uio_resid && orig_resid && 2158 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 2159 SOCKBUF_UNLOCK(&so->so_rcv); 2160 goto restart; 2161 } 2162 SOCKBUF_UNLOCK(&so->so_rcv); 2163 2164 if (flagsp != NULL) 2165 *flagsp |= flags; 2166 release: 2167 sbunlock(&so->so_rcv); 2168 return (error); 2169 } 2170 2171 /* 2172 * Optimized version of soreceive() for stream (TCP) sockets. 2173 */ 2174 int 2175 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 2176 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2177 { 2178 int len = 0, error = 0, flags, oresid; 2179 struct sockbuf *sb; 2180 struct mbuf *m, *n = NULL; 2181 2182 /* We only do stream sockets. */ 2183 if (so->so_type != SOCK_STREAM) 2184 return (EINVAL); 2185 if (psa != NULL) 2186 *psa = NULL; 2187 if (flagsp != NULL) 2188 flags = *flagsp &~ MSG_EOR; 2189 else 2190 flags = 0; 2191 if (controlp != NULL) 2192 *controlp = NULL; 2193 if (flags & MSG_OOB) 2194 return (soreceive_rcvoob(so, uio, flags)); 2195 if (mp0 != NULL) 2196 *mp0 = NULL; 2197 2198 sb = &so->so_rcv; 2199 2200 /* Prevent other readers from entering the socket. */ 2201 error = sblock(sb, SBLOCKWAIT(flags)); 2202 if (error) 2203 return (error); 2204 SOCKBUF_LOCK(sb); 2205 2206 /* Easy one, no space to copyout anything. */ 2207 if (uio->uio_resid == 0) { 2208 error = EINVAL; 2209 goto out; 2210 } 2211 oresid = uio->uio_resid; 2212 2213 /* We will never ever get anything unless we are or were connected. */ 2214 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 2215 error = ENOTCONN; 2216 goto out; 2217 } 2218 2219 restart: 2220 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2221 2222 /* Abort if socket has reported problems. */ 2223 if (so->so_error) { 2224 if (sbavail(sb) > 0) 2225 goto deliver; 2226 if (oresid > uio->uio_resid) 2227 goto out; 2228 error = so->so_error; 2229 if (!(flags & MSG_PEEK)) 2230 so->so_error = 0; 2231 goto out; 2232 } 2233 2234 /* Door is closed. Deliver what is left, if any. */ 2235 if (sb->sb_state & SBS_CANTRCVMORE) { 2236 if (sbavail(sb) > 0) 2237 goto deliver; 2238 else 2239 goto out; 2240 } 2241 2242 /* Socket buffer is empty and we shall not block. */ 2243 if (sbavail(sb) == 0 && 2244 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2245 error = EAGAIN; 2246 goto out; 2247 } 2248 2249 /* Socket buffer got some data that we shall deliver now. */ 2250 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 2251 ((so->so_state & SS_NBIO) || 2252 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2253 sbavail(sb) >= sb->sb_lowat || 2254 sbavail(sb) >= uio->uio_resid || 2255 sbavail(sb) >= sb->sb_hiwat) ) { 2256 goto deliver; 2257 } 2258 2259 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2260 if ((flags & MSG_WAITALL) && 2261 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 2262 goto deliver; 2263 2264 /* 2265 * Wait and block until (more) data comes in. 2266 * NB: Drops the sockbuf lock during wait. 2267 */ 2268 error = sbwait(sb); 2269 if (error) 2270 goto out; 2271 goto restart; 2272 2273 deliver: 2274 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2275 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 2276 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2277 2278 /* Statistics. */ 2279 if (uio->uio_td) 2280 uio->uio_td->td_ru.ru_msgrcv++; 2281 2282 /* Fill uio until full or current end of socket buffer is reached. */ 2283 len = min(uio->uio_resid, sbavail(sb)); 2284 if (mp0 != NULL) { 2285 /* Dequeue as many mbufs as possible. */ 2286 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2287 if (*mp0 == NULL) 2288 *mp0 = sb->sb_mb; 2289 else 2290 m_cat(*mp0, sb->sb_mb); 2291 for (m = sb->sb_mb; 2292 m != NULL && m->m_len <= len; 2293 m = m->m_next) { 2294 KASSERT(!(m->m_flags & M_NOTAVAIL), 2295 ("%s: m %p not available", __func__, m)); 2296 len -= m->m_len; 2297 uio->uio_resid -= m->m_len; 2298 sbfree(sb, m); 2299 n = m; 2300 } 2301 n->m_next = NULL; 2302 sb->sb_mb = m; 2303 sb->sb_lastrecord = sb->sb_mb; 2304 if (sb->sb_mb == NULL) 2305 SB_EMPTY_FIXUP(sb); 2306 } 2307 /* Copy the remainder. */ 2308 if (len > 0) { 2309 KASSERT(sb->sb_mb != NULL, 2310 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2311 2312 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2313 if (m == NULL) 2314 len = 0; /* Don't flush data from sockbuf. */ 2315 else 2316 uio->uio_resid -= len; 2317 if (*mp0 != NULL) 2318 m_cat(*mp0, m); 2319 else 2320 *mp0 = m; 2321 if (*mp0 == NULL) { 2322 error = ENOBUFS; 2323 goto out; 2324 } 2325 } 2326 } else { 2327 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2328 SOCKBUF_UNLOCK(sb); 2329 error = m_mbuftouio(uio, sb->sb_mb, len); 2330 SOCKBUF_LOCK(sb); 2331 if (error) 2332 goto out; 2333 } 2334 SBLASTRECORDCHK(sb); 2335 SBLASTMBUFCHK(sb); 2336 2337 /* 2338 * Remove the delivered data from the socket buffer unless we 2339 * were only peeking. 2340 */ 2341 if (!(flags & MSG_PEEK)) { 2342 if (len > 0) 2343 sbdrop_locked(sb, len); 2344 2345 /* Notify protocol that we drained some data. */ 2346 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2347 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2348 !(flags & MSG_SOCALLBCK))) { 2349 SOCKBUF_UNLOCK(sb); 2350 VNET_SO_ASSERT(so); 2351 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags); 2352 SOCKBUF_LOCK(sb); 2353 } 2354 } 2355 2356 /* 2357 * For MSG_WAITALL we may have to loop again and wait for 2358 * more data to come in. 2359 */ 2360 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2361 goto restart; 2362 out: 2363 SOCKBUF_LOCK_ASSERT(sb); 2364 SBLASTRECORDCHK(sb); 2365 SBLASTMBUFCHK(sb); 2366 SOCKBUF_UNLOCK(sb); 2367 sbunlock(sb); 2368 return (error); 2369 } 2370 2371 /* 2372 * Optimized version of soreceive() for simple datagram cases from userspace. 2373 * Unlike in the stream case, we're able to drop a datagram if copyout() 2374 * fails, and because we handle datagrams atomically, we don't need to use a 2375 * sleep lock to prevent I/O interlacing. 2376 */ 2377 int 2378 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2379 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2380 { 2381 struct mbuf *m, *m2; 2382 int flags, error; 2383 ssize_t len; 2384 struct protosw *pr = so->so_proto; 2385 struct mbuf *nextrecord; 2386 2387 if (psa != NULL) 2388 *psa = NULL; 2389 if (controlp != NULL) 2390 *controlp = NULL; 2391 if (flagsp != NULL) 2392 flags = *flagsp &~ MSG_EOR; 2393 else 2394 flags = 0; 2395 2396 /* 2397 * For any complicated cases, fall back to the full 2398 * soreceive_generic(). 2399 */ 2400 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB)) 2401 return (soreceive_generic(so, psa, uio, mp0, controlp, 2402 flagsp)); 2403 2404 /* 2405 * Enforce restrictions on use. 2406 */ 2407 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2408 ("soreceive_dgram: wantrcvd")); 2409 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2410 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2411 ("soreceive_dgram: SBS_RCVATMARK")); 2412 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2413 ("soreceive_dgram: P_CONNREQUIRED")); 2414 2415 /* 2416 * Loop blocking while waiting for a datagram. 2417 */ 2418 SOCKBUF_LOCK(&so->so_rcv); 2419 while ((m = so->so_rcv.sb_mb) == NULL) { 2420 KASSERT(sbavail(&so->so_rcv) == 0, 2421 ("soreceive_dgram: sb_mb NULL but sbavail %u", 2422 sbavail(&so->so_rcv))); 2423 if (so->so_error) { 2424 error = so->so_error; 2425 so->so_error = 0; 2426 SOCKBUF_UNLOCK(&so->so_rcv); 2427 return (error); 2428 } 2429 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2430 uio->uio_resid == 0) { 2431 SOCKBUF_UNLOCK(&so->so_rcv); 2432 return (0); 2433 } 2434 if ((so->so_state & SS_NBIO) || 2435 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2436 SOCKBUF_UNLOCK(&so->so_rcv); 2437 return (EWOULDBLOCK); 2438 } 2439 SBLASTRECORDCHK(&so->so_rcv); 2440 SBLASTMBUFCHK(&so->so_rcv); 2441 error = sbwait(&so->so_rcv); 2442 if (error) { 2443 SOCKBUF_UNLOCK(&so->so_rcv); 2444 return (error); 2445 } 2446 } 2447 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2448 2449 if (uio->uio_td) 2450 uio->uio_td->td_ru.ru_msgrcv++; 2451 SBLASTRECORDCHK(&so->so_rcv); 2452 SBLASTMBUFCHK(&so->so_rcv); 2453 nextrecord = m->m_nextpkt; 2454 if (nextrecord == NULL) { 2455 KASSERT(so->so_rcv.sb_lastrecord == m, 2456 ("soreceive_dgram: lastrecord != m")); 2457 } 2458 2459 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2460 ("soreceive_dgram: m_nextpkt != nextrecord")); 2461 2462 /* 2463 * Pull 'm' and its chain off the front of the packet queue. 2464 */ 2465 so->so_rcv.sb_mb = NULL; 2466 sockbuf_pushsync(&so->so_rcv, nextrecord); 2467 2468 /* 2469 * Walk 'm's chain and free that many bytes from the socket buffer. 2470 */ 2471 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2472 sbfree(&so->so_rcv, m2); 2473 2474 /* 2475 * Do a few last checks before we let go of the lock. 2476 */ 2477 SBLASTRECORDCHK(&so->so_rcv); 2478 SBLASTMBUFCHK(&so->so_rcv); 2479 SOCKBUF_UNLOCK(&so->so_rcv); 2480 2481 if (pr->pr_flags & PR_ADDR) { 2482 KASSERT(m->m_type == MT_SONAME, 2483 ("m->m_type == %d", m->m_type)); 2484 if (psa != NULL) 2485 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2486 M_NOWAIT); 2487 m = m_free(m); 2488 } 2489 if (m == NULL) { 2490 /* XXXRW: Can this happen? */ 2491 return (0); 2492 } 2493 2494 /* 2495 * Packet to copyout() is now in 'm' and it is disconnected from the 2496 * queue. 2497 * 2498 * Process one or more MT_CONTROL mbufs present before any data mbufs 2499 * in the first mbuf chain on the socket buffer. We call into the 2500 * protocol to perform externalization (or freeing if controlp == 2501 * NULL). In some cases there can be only MT_CONTROL mbufs without 2502 * MT_DATA mbufs. 2503 */ 2504 if (m->m_type == MT_CONTROL) { 2505 struct mbuf *cm = NULL, *cmn; 2506 struct mbuf **cme = &cm; 2507 2508 do { 2509 m2 = m->m_next; 2510 m->m_next = NULL; 2511 *cme = m; 2512 cme = &(*cme)->m_next; 2513 m = m2; 2514 } while (m != NULL && m->m_type == MT_CONTROL); 2515 while (cm != NULL) { 2516 cmn = cm->m_next; 2517 cm->m_next = NULL; 2518 if (pr->pr_domain->dom_externalize != NULL) { 2519 error = (*pr->pr_domain->dom_externalize) 2520 (cm, controlp, flags); 2521 } else if (controlp != NULL) 2522 *controlp = cm; 2523 else 2524 m_freem(cm); 2525 if (controlp != NULL) { 2526 while (*controlp != NULL) 2527 controlp = &(*controlp)->m_next; 2528 } 2529 cm = cmn; 2530 } 2531 } 2532 KASSERT(m == NULL || m->m_type == MT_DATA, 2533 ("soreceive_dgram: !data")); 2534 while (m != NULL && uio->uio_resid > 0) { 2535 len = uio->uio_resid; 2536 if (len > m->m_len) 2537 len = m->m_len; 2538 error = uiomove(mtod(m, char *), (int)len, uio); 2539 if (error) { 2540 m_freem(m); 2541 return (error); 2542 } 2543 if (len == m->m_len) 2544 m = m_free(m); 2545 else { 2546 m->m_data += len; 2547 m->m_len -= len; 2548 } 2549 } 2550 if (m != NULL) { 2551 flags |= MSG_TRUNC; 2552 m_freem(m); 2553 } 2554 if (flagsp != NULL) 2555 *flagsp |= flags; 2556 return (0); 2557 } 2558 2559 int 2560 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2561 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2562 { 2563 int error; 2564 2565 CURVNET_SET(so->so_vnet); 2566 if (!SOLISTENING(so)) 2567 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, 2568 mp0, controlp, flagsp)); 2569 else 2570 error = ENOTCONN; 2571 CURVNET_RESTORE(); 2572 return (error); 2573 } 2574 2575 int 2576 soshutdown(struct socket *so, int how) 2577 { 2578 struct protosw *pr = so->so_proto; 2579 int error, soerror_enotconn; 2580 2581 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2582 return (EINVAL); 2583 2584 soerror_enotconn = 0; 2585 if ((so->so_state & 2586 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 2587 /* 2588 * POSIX mandates us to return ENOTCONN when shutdown(2) is 2589 * invoked on a datagram sockets, however historically we would 2590 * actually tear socket down. This is known to be leveraged by 2591 * some applications to unblock process waiting in recvXXX(2) 2592 * by other process that it shares that socket with. Try to meet 2593 * both backward-compatibility and POSIX requirements by forcing 2594 * ENOTCONN but still asking protocol to perform pru_shutdown(). 2595 */ 2596 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) 2597 return (ENOTCONN); 2598 soerror_enotconn = 1; 2599 } 2600 2601 if (SOLISTENING(so)) { 2602 if (how != SHUT_WR) { 2603 SOLISTEN_LOCK(so); 2604 so->so_error = ECONNABORTED; 2605 solisten_wakeup(so); /* unlocks so */ 2606 } 2607 goto done; 2608 } 2609 2610 CURVNET_SET(so->so_vnet); 2611 if (pr->pr_usrreqs->pru_flush != NULL) 2612 (*pr->pr_usrreqs->pru_flush)(so, how); 2613 if (how != SHUT_WR) 2614 sorflush(so); 2615 if (how != SHUT_RD) { 2616 error = (*pr->pr_usrreqs->pru_shutdown)(so); 2617 wakeup(&so->so_timeo); 2618 CURVNET_RESTORE(); 2619 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error); 2620 } 2621 wakeup(&so->so_timeo); 2622 CURVNET_RESTORE(); 2623 2624 done: 2625 return (soerror_enotconn ? ENOTCONN : 0); 2626 } 2627 2628 void 2629 sorflush(struct socket *so) 2630 { 2631 struct sockbuf *sb = &so->so_rcv; 2632 struct protosw *pr = so->so_proto; 2633 struct socket aso; 2634 2635 VNET_SO_ASSERT(so); 2636 2637 /* 2638 * In order to avoid calling dom_dispose with the socket buffer mutex 2639 * held, and in order to generally avoid holding the lock for a long 2640 * time, we make a copy of the socket buffer and clear the original 2641 * (except locks, state). The new socket buffer copy won't have 2642 * initialized locks so we can only call routines that won't use or 2643 * assert those locks. 2644 * 2645 * Dislodge threads currently blocked in receive and wait to acquire 2646 * a lock against other simultaneous readers before clearing the 2647 * socket buffer. Don't let our acquire be interrupted by a signal 2648 * despite any existing socket disposition on interruptable waiting. 2649 */ 2650 socantrcvmore(so); 2651 (void) sblock(sb, SBL_WAIT | SBL_NOINTR); 2652 2653 /* 2654 * Invalidate/clear most of the sockbuf structure, but leave selinfo 2655 * and mutex data unchanged. 2656 */ 2657 SOCKBUF_LOCK(sb); 2658 bzero(&aso, sizeof(aso)); 2659 aso.so_pcb = so->so_pcb; 2660 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero, 2661 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2662 bzero(&sb->sb_startzero, 2663 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2664 SOCKBUF_UNLOCK(sb); 2665 sbunlock(sb); 2666 2667 /* 2668 * Dispose of special rights and flush the copied socket. Don't call 2669 * any unsafe routines (that rely on locks being initialized) on aso. 2670 */ 2671 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 2672 (*pr->pr_domain->dom_dispose)(&aso); 2673 sbrelease_internal(&aso.so_rcv, so); 2674 } 2675 2676 /* 2677 * Wrapper for Socket established helper hook. 2678 * Parameters: socket, context of the hook point, hook id. 2679 */ 2680 static int inline 2681 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 2682 { 2683 struct socket_hhook_data hhook_data = { 2684 .so = so, 2685 .hctx = hctx, 2686 .m = NULL, 2687 .status = 0 2688 }; 2689 2690 CURVNET_SET(so->so_vnet); 2691 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 2692 CURVNET_RESTORE(); 2693 2694 /* Ugly but needed, since hhooks return void for now */ 2695 return (hhook_data.status); 2696 } 2697 2698 /* 2699 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 2700 * additional variant to handle the case where the option value needs to be 2701 * some kind of integer, but not a specific size. In addition to their use 2702 * here, these functions are also called by the protocol-level pr_ctloutput() 2703 * routines. 2704 */ 2705 int 2706 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2707 { 2708 size_t valsize; 2709 2710 /* 2711 * If the user gives us more than we wanted, we ignore it, but if we 2712 * don't get the minimum length the caller wants, we return EINVAL. 2713 * On success, sopt->sopt_valsize is set to however much we actually 2714 * retrieved. 2715 */ 2716 if ((valsize = sopt->sopt_valsize) < minlen) 2717 return EINVAL; 2718 if (valsize > len) 2719 sopt->sopt_valsize = valsize = len; 2720 2721 if (sopt->sopt_td != NULL) 2722 return (copyin(sopt->sopt_val, buf, valsize)); 2723 2724 bcopy(sopt->sopt_val, buf, valsize); 2725 return (0); 2726 } 2727 2728 /* 2729 * Kernel version of setsockopt(2). 2730 * 2731 * XXX: optlen is size_t, not socklen_t 2732 */ 2733 int 2734 so_setsockopt(struct socket *so, int level, int optname, void *optval, 2735 size_t optlen) 2736 { 2737 struct sockopt sopt; 2738 2739 sopt.sopt_level = level; 2740 sopt.sopt_name = optname; 2741 sopt.sopt_dir = SOPT_SET; 2742 sopt.sopt_val = optval; 2743 sopt.sopt_valsize = optlen; 2744 sopt.sopt_td = NULL; 2745 return (sosetopt(so, &sopt)); 2746 } 2747 2748 int 2749 sosetopt(struct socket *so, struct sockopt *sopt) 2750 { 2751 int error, optval; 2752 struct linger l; 2753 struct timeval tv; 2754 sbintime_t val; 2755 uint32_t val32; 2756 #ifdef MAC 2757 struct mac extmac; 2758 #endif 2759 2760 CURVNET_SET(so->so_vnet); 2761 error = 0; 2762 if (sopt->sopt_level != SOL_SOCKET) { 2763 if (so->so_proto->pr_ctloutput != NULL) 2764 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2765 else 2766 error = ENOPROTOOPT; 2767 } else { 2768 switch (sopt->sopt_name) { 2769 case SO_ACCEPTFILTER: 2770 error = accept_filt_setopt(so, sopt); 2771 if (error) 2772 goto bad; 2773 break; 2774 2775 case SO_LINGER: 2776 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 2777 if (error) 2778 goto bad; 2779 if (l.l_linger < 0 || 2780 l.l_linger > USHRT_MAX || 2781 l.l_linger > (INT_MAX / hz)) { 2782 error = EDOM; 2783 goto bad; 2784 } 2785 SOCK_LOCK(so); 2786 so->so_linger = l.l_linger; 2787 if (l.l_onoff) 2788 so->so_options |= SO_LINGER; 2789 else 2790 so->so_options &= ~SO_LINGER; 2791 SOCK_UNLOCK(so); 2792 break; 2793 2794 case SO_DEBUG: 2795 case SO_KEEPALIVE: 2796 case SO_DONTROUTE: 2797 case SO_USELOOPBACK: 2798 case SO_BROADCAST: 2799 case SO_REUSEADDR: 2800 case SO_REUSEPORT: 2801 case SO_REUSEPORT_LB: 2802 case SO_OOBINLINE: 2803 case SO_TIMESTAMP: 2804 case SO_BINTIME: 2805 case SO_NOSIGPIPE: 2806 case SO_NO_DDP: 2807 case SO_NO_OFFLOAD: 2808 error = sooptcopyin(sopt, &optval, sizeof optval, 2809 sizeof optval); 2810 if (error) 2811 goto bad; 2812 SOCK_LOCK(so); 2813 if (optval) 2814 so->so_options |= sopt->sopt_name; 2815 else 2816 so->so_options &= ~sopt->sopt_name; 2817 SOCK_UNLOCK(so); 2818 break; 2819 2820 case SO_SETFIB: 2821 error = sooptcopyin(sopt, &optval, sizeof optval, 2822 sizeof optval); 2823 if (error) 2824 goto bad; 2825 2826 if (optval < 0 || optval >= rt_numfibs) { 2827 error = EINVAL; 2828 goto bad; 2829 } 2830 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 2831 (so->so_proto->pr_domain->dom_family == PF_INET6) || 2832 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 2833 so->so_fibnum = optval; 2834 else 2835 so->so_fibnum = 0; 2836 break; 2837 2838 case SO_USER_COOKIE: 2839 error = sooptcopyin(sopt, &val32, sizeof val32, 2840 sizeof val32); 2841 if (error) 2842 goto bad; 2843 so->so_user_cookie = val32; 2844 break; 2845 2846 case SO_SNDBUF: 2847 case SO_RCVBUF: 2848 case SO_SNDLOWAT: 2849 case SO_RCVLOWAT: 2850 error = sooptcopyin(sopt, &optval, sizeof optval, 2851 sizeof optval); 2852 if (error) 2853 goto bad; 2854 2855 /* 2856 * Values < 1 make no sense for any of these options, 2857 * so disallow them. 2858 */ 2859 if (optval < 1) { 2860 error = EINVAL; 2861 goto bad; 2862 } 2863 2864 error = sbsetopt(so, sopt->sopt_name, optval); 2865 break; 2866 2867 case SO_SNDTIMEO: 2868 case SO_RCVTIMEO: 2869 #ifdef COMPAT_FREEBSD32 2870 if (SV_CURPROC_FLAG(SV_ILP32)) { 2871 struct timeval32 tv32; 2872 2873 error = sooptcopyin(sopt, &tv32, sizeof tv32, 2874 sizeof tv32); 2875 CP(tv32, tv, tv_sec); 2876 CP(tv32, tv, tv_usec); 2877 } else 2878 #endif 2879 error = sooptcopyin(sopt, &tv, sizeof tv, 2880 sizeof tv); 2881 if (error) 2882 goto bad; 2883 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 2884 tv.tv_usec >= 1000000) { 2885 error = EDOM; 2886 goto bad; 2887 } 2888 if (tv.tv_sec > INT32_MAX) 2889 val = SBT_MAX; 2890 else 2891 val = tvtosbt(tv); 2892 switch (sopt->sopt_name) { 2893 case SO_SNDTIMEO: 2894 so->so_snd.sb_timeo = val; 2895 break; 2896 case SO_RCVTIMEO: 2897 so->so_rcv.sb_timeo = val; 2898 break; 2899 } 2900 break; 2901 2902 case SO_LABEL: 2903 #ifdef MAC 2904 error = sooptcopyin(sopt, &extmac, sizeof extmac, 2905 sizeof extmac); 2906 if (error) 2907 goto bad; 2908 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 2909 so, &extmac); 2910 #else 2911 error = EOPNOTSUPP; 2912 #endif 2913 break; 2914 2915 case SO_TS_CLOCK: 2916 error = sooptcopyin(sopt, &optval, sizeof optval, 2917 sizeof optval); 2918 if (error) 2919 goto bad; 2920 if (optval < 0 || optval > SO_TS_CLOCK_MAX) { 2921 error = EINVAL; 2922 goto bad; 2923 } 2924 so->so_ts_clock = optval; 2925 break; 2926 2927 case SO_MAX_PACING_RATE: 2928 error = sooptcopyin(sopt, &val32, sizeof(val32), 2929 sizeof(val32)); 2930 if (error) 2931 goto bad; 2932 so->so_max_pacing_rate = val32; 2933 break; 2934 2935 default: 2936 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 2937 error = hhook_run_socket(so, sopt, 2938 HHOOK_SOCKET_OPT); 2939 else 2940 error = ENOPROTOOPT; 2941 break; 2942 } 2943 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 2944 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 2945 } 2946 bad: 2947 CURVNET_RESTORE(); 2948 return (error); 2949 } 2950 2951 /* 2952 * Helper routine for getsockopt. 2953 */ 2954 int 2955 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 2956 { 2957 int error; 2958 size_t valsize; 2959 2960 error = 0; 2961 2962 /* 2963 * Documented get behavior is that we always return a value, possibly 2964 * truncated to fit in the user's buffer. Traditional behavior is 2965 * that we always tell the user precisely how much we copied, rather 2966 * than something useful like the total amount we had available for 2967 * her. Note that this interface is not idempotent; the entire 2968 * answer must be generated ahead of time. 2969 */ 2970 valsize = min(len, sopt->sopt_valsize); 2971 sopt->sopt_valsize = valsize; 2972 if (sopt->sopt_val != NULL) { 2973 if (sopt->sopt_td != NULL) 2974 error = copyout(buf, sopt->sopt_val, valsize); 2975 else 2976 bcopy(buf, sopt->sopt_val, valsize); 2977 } 2978 return (error); 2979 } 2980 2981 int 2982 sogetopt(struct socket *so, struct sockopt *sopt) 2983 { 2984 int error, optval; 2985 struct linger l; 2986 struct timeval tv; 2987 #ifdef MAC 2988 struct mac extmac; 2989 #endif 2990 2991 CURVNET_SET(so->so_vnet); 2992 error = 0; 2993 if (sopt->sopt_level != SOL_SOCKET) { 2994 if (so->so_proto->pr_ctloutput != NULL) 2995 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2996 else 2997 error = ENOPROTOOPT; 2998 CURVNET_RESTORE(); 2999 return (error); 3000 } else { 3001 switch (sopt->sopt_name) { 3002 case SO_ACCEPTFILTER: 3003 error = accept_filt_getopt(so, sopt); 3004 break; 3005 3006 case SO_LINGER: 3007 SOCK_LOCK(so); 3008 l.l_onoff = so->so_options & SO_LINGER; 3009 l.l_linger = so->so_linger; 3010 SOCK_UNLOCK(so); 3011 error = sooptcopyout(sopt, &l, sizeof l); 3012 break; 3013 3014 case SO_USELOOPBACK: 3015 case SO_DONTROUTE: 3016 case SO_DEBUG: 3017 case SO_KEEPALIVE: 3018 case SO_REUSEADDR: 3019 case SO_REUSEPORT: 3020 case SO_REUSEPORT_LB: 3021 case SO_BROADCAST: 3022 case SO_OOBINLINE: 3023 case SO_ACCEPTCONN: 3024 case SO_TIMESTAMP: 3025 case SO_BINTIME: 3026 case SO_NOSIGPIPE: 3027 optval = so->so_options & sopt->sopt_name; 3028 integer: 3029 error = sooptcopyout(sopt, &optval, sizeof optval); 3030 break; 3031 3032 case SO_DOMAIN: 3033 optval = so->so_proto->pr_domain->dom_family; 3034 goto integer; 3035 3036 case SO_TYPE: 3037 optval = so->so_type; 3038 goto integer; 3039 3040 case SO_PROTOCOL: 3041 optval = so->so_proto->pr_protocol; 3042 goto integer; 3043 3044 case SO_ERROR: 3045 SOCK_LOCK(so); 3046 optval = so->so_error; 3047 so->so_error = 0; 3048 SOCK_UNLOCK(so); 3049 goto integer; 3050 3051 case SO_SNDBUF: 3052 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : 3053 so->so_snd.sb_hiwat; 3054 goto integer; 3055 3056 case SO_RCVBUF: 3057 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : 3058 so->so_rcv.sb_hiwat; 3059 goto integer; 3060 3061 case SO_SNDLOWAT: 3062 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : 3063 so->so_snd.sb_lowat; 3064 goto integer; 3065 3066 case SO_RCVLOWAT: 3067 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : 3068 so->so_rcv.sb_lowat; 3069 goto integer; 3070 3071 case SO_SNDTIMEO: 3072 case SO_RCVTIMEO: 3073 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 3074 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 3075 #ifdef COMPAT_FREEBSD32 3076 if (SV_CURPROC_FLAG(SV_ILP32)) { 3077 struct timeval32 tv32; 3078 3079 CP(tv, tv32, tv_sec); 3080 CP(tv, tv32, tv_usec); 3081 error = sooptcopyout(sopt, &tv32, sizeof tv32); 3082 } else 3083 #endif 3084 error = sooptcopyout(sopt, &tv, sizeof tv); 3085 break; 3086 3087 case SO_LABEL: 3088 #ifdef MAC 3089 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3090 sizeof(extmac)); 3091 if (error) 3092 goto bad; 3093 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 3094 so, &extmac); 3095 if (error) 3096 goto bad; 3097 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3098 #else 3099 error = EOPNOTSUPP; 3100 #endif 3101 break; 3102 3103 case SO_PEERLABEL: 3104 #ifdef MAC 3105 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3106 sizeof(extmac)); 3107 if (error) 3108 goto bad; 3109 error = mac_getsockopt_peerlabel( 3110 sopt->sopt_td->td_ucred, so, &extmac); 3111 if (error) 3112 goto bad; 3113 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3114 #else 3115 error = EOPNOTSUPP; 3116 #endif 3117 break; 3118 3119 case SO_LISTENQLIMIT: 3120 optval = SOLISTENING(so) ? so->sol_qlimit : 0; 3121 goto integer; 3122 3123 case SO_LISTENQLEN: 3124 optval = SOLISTENING(so) ? so->sol_qlen : 0; 3125 goto integer; 3126 3127 case SO_LISTENINCQLEN: 3128 optval = SOLISTENING(so) ? so->sol_incqlen : 0; 3129 goto integer; 3130 3131 case SO_TS_CLOCK: 3132 optval = so->so_ts_clock; 3133 goto integer; 3134 3135 case SO_MAX_PACING_RATE: 3136 optval = so->so_max_pacing_rate; 3137 goto integer; 3138 3139 default: 3140 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3141 error = hhook_run_socket(so, sopt, 3142 HHOOK_SOCKET_OPT); 3143 else 3144 error = ENOPROTOOPT; 3145 break; 3146 } 3147 } 3148 #ifdef MAC 3149 bad: 3150 #endif 3151 CURVNET_RESTORE(); 3152 return (error); 3153 } 3154 3155 int 3156 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 3157 { 3158 struct mbuf *m, *m_prev; 3159 int sopt_size = sopt->sopt_valsize; 3160 3161 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3162 if (m == NULL) 3163 return ENOBUFS; 3164 if (sopt_size > MLEN) { 3165 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 3166 if ((m->m_flags & M_EXT) == 0) { 3167 m_free(m); 3168 return ENOBUFS; 3169 } 3170 m->m_len = min(MCLBYTES, sopt_size); 3171 } else { 3172 m->m_len = min(MLEN, sopt_size); 3173 } 3174 sopt_size -= m->m_len; 3175 *mp = m; 3176 m_prev = m; 3177 3178 while (sopt_size) { 3179 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3180 if (m == NULL) { 3181 m_freem(*mp); 3182 return ENOBUFS; 3183 } 3184 if (sopt_size > MLEN) { 3185 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 3186 M_NOWAIT); 3187 if ((m->m_flags & M_EXT) == 0) { 3188 m_freem(m); 3189 m_freem(*mp); 3190 return ENOBUFS; 3191 } 3192 m->m_len = min(MCLBYTES, sopt_size); 3193 } else { 3194 m->m_len = min(MLEN, sopt_size); 3195 } 3196 sopt_size -= m->m_len; 3197 m_prev->m_next = m; 3198 m_prev = m; 3199 } 3200 return (0); 3201 } 3202 3203 int 3204 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 3205 { 3206 struct mbuf *m0 = m; 3207 3208 if (sopt->sopt_val == NULL) 3209 return (0); 3210 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3211 if (sopt->sopt_td != NULL) { 3212 int error; 3213 3214 error = copyin(sopt->sopt_val, mtod(m, char *), 3215 m->m_len); 3216 if (error != 0) { 3217 m_freem(m0); 3218 return(error); 3219 } 3220 } else 3221 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 3222 sopt->sopt_valsize -= m->m_len; 3223 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3224 m = m->m_next; 3225 } 3226 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 3227 panic("ip6_sooptmcopyin"); 3228 return (0); 3229 } 3230 3231 int 3232 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 3233 { 3234 struct mbuf *m0 = m; 3235 size_t valsize = 0; 3236 3237 if (sopt->sopt_val == NULL) 3238 return (0); 3239 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3240 if (sopt->sopt_td != NULL) { 3241 int error; 3242 3243 error = copyout(mtod(m, char *), sopt->sopt_val, 3244 m->m_len); 3245 if (error != 0) { 3246 m_freem(m0); 3247 return(error); 3248 } 3249 } else 3250 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 3251 sopt->sopt_valsize -= m->m_len; 3252 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3253 valsize += m->m_len; 3254 m = m->m_next; 3255 } 3256 if (m != NULL) { 3257 /* enough soopt buffer should be given from user-land */ 3258 m_freem(m0); 3259 return(EINVAL); 3260 } 3261 sopt->sopt_valsize = valsize; 3262 return (0); 3263 } 3264 3265 /* 3266 * sohasoutofband(): protocol notifies socket layer of the arrival of new 3267 * out-of-band data, which will then notify socket consumers. 3268 */ 3269 void 3270 sohasoutofband(struct socket *so) 3271 { 3272 3273 if (so->so_sigio != NULL) 3274 pgsigio(&so->so_sigio, SIGURG, 0); 3275 selwakeuppri(&so->so_rdsel, PSOCK); 3276 } 3277 3278 int 3279 sopoll(struct socket *so, int events, struct ucred *active_cred, 3280 struct thread *td) 3281 { 3282 3283 /* 3284 * We do not need to set or assert curvnet as long as everyone uses 3285 * sopoll_generic(). 3286 */ 3287 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, 3288 td)); 3289 } 3290 3291 int 3292 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3293 struct thread *td) 3294 { 3295 int revents; 3296 3297 SOCK_LOCK(so); 3298 if (SOLISTENING(so)) { 3299 if (!(events & (POLLIN | POLLRDNORM))) 3300 revents = 0; 3301 else if (!TAILQ_EMPTY(&so->sol_comp)) 3302 revents = events & (POLLIN | POLLRDNORM); 3303 else if ((events & POLLINIGNEOF) == 0 && so->so_error) 3304 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP; 3305 else { 3306 selrecord(td, &so->so_rdsel); 3307 revents = 0; 3308 } 3309 } else { 3310 revents = 0; 3311 SOCKBUF_LOCK(&so->so_snd); 3312 SOCKBUF_LOCK(&so->so_rcv); 3313 if (events & (POLLIN | POLLRDNORM)) 3314 if (soreadabledata(so)) 3315 revents |= events & (POLLIN | POLLRDNORM); 3316 if (events & (POLLOUT | POLLWRNORM)) 3317 if (sowriteable(so)) 3318 revents |= events & (POLLOUT | POLLWRNORM); 3319 if (events & (POLLPRI | POLLRDBAND)) 3320 if (so->so_oobmark || 3321 (so->so_rcv.sb_state & SBS_RCVATMARK)) 3322 revents |= events & (POLLPRI | POLLRDBAND); 3323 if ((events & POLLINIGNEOF) == 0) { 3324 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3325 revents |= events & (POLLIN | POLLRDNORM); 3326 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3327 revents |= POLLHUP; 3328 } 3329 } 3330 if (revents == 0) { 3331 if (events & 3332 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 3333 selrecord(td, &so->so_rdsel); 3334 so->so_rcv.sb_flags |= SB_SEL; 3335 } 3336 if (events & (POLLOUT | POLLWRNORM)) { 3337 selrecord(td, &so->so_wrsel); 3338 so->so_snd.sb_flags |= SB_SEL; 3339 } 3340 } 3341 SOCKBUF_UNLOCK(&so->so_rcv); 3342 SOCKBUF_UNLOCK(&so->so_snd); 3343 } 3344 SOCK_UNLOCK(so); 3345 return (revents); 3346 } 3347 3348 int 3349 soo_kqfilter(struct file *fp, struct knote *kn) 3350 { 3351 struct socket *so = kn->kn_fp->f_data; 3352 struct sockbuf *sb; 3353 struct knlist *knl; 3354 3355 switch (kn->kn_filter) { 3356 case EVFILT_READ: 3357 kn->kn_fop = &soread_filtops; 3358 knl = &so->so_rdsel.si_note; 3359 sb = &so->so_rcv; 3360 break; 3361 case EVFILT_WRITE: 3362 kn->kn_fop = &sowrite_filtops; 3363 knl = &so->so_wrsel.si_note; 3364 sb = &so->so_snd; 3365 break; 3366 case EVFILT_EMPTY: 3367 kn->kn_fop = &soempty_filtops; 3368 knl = &so->so_wrsel.si_note; 3369 sb = &so->so_snd; 3370 break; 3371 default: 3372 return (EINVAL); 3373 } 3374 3375 SOCK_LOCK(so); 3376 if (SOLISTENING(so)) { 3377 knlist_add(knl, kn, 1); 3378 } else { 3379 SOCKBUF_LOCK(sb); 3380 knlist_add(knl, kn, 1); 3381 sb->sb_flags |= SB_KNOTE; 3382 SOCKBUF_UNLOCK(sb); 3383 } 3384 SOCK_UNLOCK(so); 3385 return (0); 3386 } 3387 3388 /* 3389 * Some routines that return EOPNOTSUPP for entry points that are not 3390 * supported by a protocol. Fill in as needed. 3391 */ 3392 int 3393 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 3394 { 3395 3396 return EOPNOTSUPP; 3397 } 3398 3399 int 3400 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job) 3401 { 3402 3403 return EOPNOTSUPP; 3404 } 3405 3406 int 3407 pru_attach_notsupp(struct socket *so, int proto, struct thread *td) 3408 { 3409 3410 return EOPNOTSUPP; 3411 } 3412 3413 int 3414 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3415 { 3416 3417 return EOPNOTSUPP; 3418 } 3419 3420 int 3421 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3422 struct thread *td) 3423 { 3424 3425 return EOPNOTSUPP; 3426 } 3427 3428 int 3429 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3430 { 3431 3432 return EOPNOTSUPP; 3433 } 3434 3435 int 3436 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3437 struct thread *td) 3438 { 3439 3440 return EOPNOTSUPP; 3441 } 3442 3443 int 3444 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 3445 { 3446 3447 return EOPNOTSUPP; 3448 } 3449 3450 int 3451 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 3452 struct ifnet *ifp, struct thread *td) 3453 { 3454 3455 return EOPNOTSUPP; 3456 } 3457 3458 int 3459 pru_disconnect_notsupp(struct socket *so) 3460 { 3461 3462 return EOPNOTSUPP; 3463 } 3464 3465 int 3466 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) 3467 { 3468 3469 return EOPNOTSUPP; 3470 } 3471 3472 int 3473 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) 3474 { 3475 3476 return EOPNOTSUPP; 3477 } 3478 3479 int 3480 pru_rcvd_notsupp(struct socket *so, int flags) 3481 { 3482 3483 return EOPNOTSUPP; 3484 } 3485 3486 int 3487 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 3488 { 3489 3490 return EOPNOTSUPP; 3491 } 3492 3493 int 3494 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, 3495 struct sockaddr *addr, struct mbuf *control, struct thread *td) 3496 { 3497 3498 return EOPNOTSUPP; 3499 } 3500 3501 int 3502 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count) 3503 { 3504 3505 return (EOPNOTSUPP); 3506 } 3507 3508 /* 3509 * This isn't really a ``null'' operation, but it's the default one and 3510 * doesn't do anything destructive. 3511 */ 3512 int 3513 pru_sense_null(struct socket *so, struct stat *sb) 3514 { 3515 3516 sb->st_blksize = so->so_snd.sb_hiwat; 3517 return 0; 3518 } 3519 3520 int 3521 pru_shutdown_notsupp(struct socket *so) 3522 { 3523 3524 return EOPNOTSUPP; 3525 } 3526 3527 int 3528 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) 3529 { 3530 3531 return EOPNOTSUPP; 3532 } 3533 3534 int 3535 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 3536 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 3537 { 3538 3539 return EOPNOTSUPP; 3540 } 3541 3542 int 3543 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 3544 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 3545 { 3546 3547 return EOPNOTSUPP; 3548 } 3549 3550 int 3551 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, 3552 struct thread *td) 3553 { 3554 3555 return EOPNOTSUPP; 3556 } 3557 3558 static void 3559 filt_sordetach(struct knote *kn) 3560 { 3561 struct socket *so = kn->kn_fp->f_data; 3562 3563 so_rdknl_lock(so); 3564 knlist_remove(&so->so_rdsel.si_note, kn, 1); 3565 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note)) 3566 so->so_rcv.sb_flags &= ~SB_KNOTE; 3567 so_rdknl_unlock(so); 3568 } 3569 3570 /*ARGSUSED*/ 3571 static int 3572 filt_soread(struct knote *kn, long hint) 3573 { 3574 struct socket *so; 3575 3576 so = kn->kn_fp->f_data; 3577 3578 if (SOLISTENING(so)) { 3579 SOCK_LOCK_ASSERT(so); 3580 kn->kn_data = so->sol_qlen; 3581 if (so->so_error) { 3582 kn->kn_flags |= EV_EOF; 3583 kn->kn_fflags = so->so_error; 3584 return (1); 3585 } 3586 return (!TAILQ_EMPTY(&so->sol_comp)); 3587 } 3588 3589 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3590 3591 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 3592 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3593 kn->kn_flags |= EV_EOF; 3594 kn->kn_fflags = so->so_error; 3595 return (1); 3596 } else if (so->so_error) /* temporary udp error */ 3597 return (1); 3598 3599 if (kn->kn_sfflags & NOTE_LOWAT) { 3600 if (kn->kn_data >= kn->kn_sdata) 3601 return (1); 3602 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) 3603 return (1); 3604 3605 /* This hook returning non-zero indicates an event, not error */ 3606 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); 3607 } 3608 3609 static void 3610 filt_sowdetach(struct knote *kn) 3611 { 3612 struct socket *so = kn->kn_fp->f_data; 3613 3614 so_wrknl_lock(so); 3615 knlist_remove(&so->so_wrsel.si_note, kn, 1); 3616 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note)) 3617 so->so_snd.sb_flags &= ~SB_KNOTE; 3618 so_wrknl_unlock(so); 3619 } 3620 3621 /*ARGSUSED*/ 3622 static int 3623 filt_sowrite(struct knote *kn, long hint) 3624 { 3625 struct socket *so; 3626 3627 so = kn->kn_fp->f_data; 3628 3629 if (SOLISTENING(so)) 3630 return (0); 3631 3632 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3633 kn->kn_data = sbspace(&so->so_snd); 3634 3635 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); 3636 3637 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3638 kn->kn_flags |= EV_EOF; 3639 kn->kn_fflags = so->so_error; 3640 return (1); 3641 } else if (so->so_error) /* temporary udp error */ 3642 return (1); 3643 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3644 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3645 return (0); 3646 else if (kn->kn_sfflags & NOTE_LOWAT) 3647 return (kn->kn_data >= kn->kn_sdata); 3648 else 3649 return (kn->kn_data >= so->so_snd.sb_lowat); 3650 } 3651 3652 static int 3653 filt_soempty(struct knote *kn, long hint) 3654 { 3655 struct socket *so; 3656 3657 so = kn->kn_fp->f_data; 3658 3659 if (SOLISTENING(so)) 3660 return (1); 3661 3662 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3663 kn->kn_data = sbused(&so->so_snd); 3664 3665 if (kn->kn_data == 0) 3666 return (1); 3667 else 3668 return (0); 3669 } 3670 3671 int 3672 socheckuid(struct socket *so, uid_t uid) 3673 { 3674 3675 if (so == NULL) 3676 return (EPERM); 3677 if (so->so_cred->cr_uid != uid) 3678 return (EPERM); 3679 return (0); 3680 } 3681 3682 /* 3683 * These functions are used by protocols to notify the socket layer (and its 3684 * consumers) of state changes in the sockets driven by protocol-side events. 3685 */ 3686 3687 /* 3688 * Procedures to manipulate state flags of socket and do appropriate wakeups. 3689 * 3690 * Normal sequence from the active (originating) side is that 3691 * soisconnecting() is called during processing of connect() call, resulting 3692 * in an eventual call to soisconnected() if/when the connection is 3693 * established. When the connection is torn down soisdisconnecting() is 3694 * called during processing of disconnect() call, and soisdisconnected() is 3695 * called when the connection to the peer is totally severed. The semantics 3696 * of these routines are such that connectionless protocols can call 3697 * soisconnected() and soisdisconnected() only, bypassing the in-progress 3698 * calls when setting up a ``connection'' takes no time. 3699 * 3700 * From the passive side, a socket is created with two queues of sockets: 3701 * so_incomp for connections in progress and so_comp for connections already 3702 * made and awaiting user acceptance. As a protocol is preparing incoming 3703 * connections, it creates a socket structure queued on so_incomp by calling 3704 * sonewconn(). When the connection is established, soisconnected() is 3705 * called, and transfers the socket structure to so_comp, making it available 3706 * to accept(). 3707 * 3708 * If a socket is closed with sockets on either so_incomp or so_comp, these 3709 * sockets are dropped. 3710 * 3711 * If higher-level protocols are implemented in the kernel, the wakeups done 3712 * here will sometimes cause software-interrupt process scheduling. 3713 */ 3714 void 3715 soisconnecting(struct socket *so) 3716 { 3717 3718 SOCK_LOCK(so); 3719 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 3720 so->so_state |= SS_ISCONNECTING; 3721 SOCK_UNLOCK(so); 3722 } 3723 3724 void 3725 soisconnected(struct socket *so) 3726 { 3727 3728 SOCK_LOCK(so); 3729 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 3730 so->so_state |= SS_ISCONNECTED; 3731 3732 if (so->so_qstate == SQ_INCOMP) { 3733 struct socket *head = so->so_listen; 3734 int ret; 3735 3736 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so)); 3737 /* 3738 * Promoting a socket from incomplete queue to complete, we 3739 * need to go through reverse order of locking. We first do 3740 * trylock, and if that doesn't succeed, we go the hard way 3741 * leaving a reference and rechecking consistency after proper 3742 * locking. 3743 */ 3744 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) { 3745 soref(head); 3746 SOCK_UNLOCK(so); 3747 SOLISTEN_LOCK(head); 3748 SOCK_LOCK(so); 3749 if (__predict_false(head != so->so_listen)) { 3750 /* 3751 * The socket went off the listen queue, 3752 * should be lost race to close(2) of sol. 3753 * The socket is about to soabort(). 3754 */ 3755 SOCK_UNLOCK(so); 3756 sorele(head); 3757 return; 3758 } 3759 /* Not the last one, as so holds a ref. */ 3760 refcount_release(&head->so_count); 3761 } 3762 again: 3763 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 3764 TAILQ_REMOVE(&head->sol_incomp, so, so_list); 3765 head->sol_incqlen--; 3766 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 3767 head->sol_qlen++; 3768 so->so_qstate = SQ_COMP; 3769 SOCK_UNLOCK(so); 3770 solisten_wakeup(head); /* unlocks */ 3771 } else { 3772 SOCKBUF_LOCK(&so->so_rcv); 3773 soupcall_set(so, SO_RCV, 3774 head->sol_accept_filter->accf_callback, 3775 head->sol_accept_filter_arg); 3776 so->so_options &= ~SO_ACCEPTFILTER; 3777 ret = head->sol_accept_filter->accf_callback(so, 3778 head->sol_accept_filter_arg, M_NOWAIT); 3779 if (ret == SU_ISCONNECTED) { 3780 soupcall_clear(so, SO_RCV); 3781 SOCKBUF_UNLOCK(&so->so_rcv); 3782 goto again; 3783 } 3784 SOCKBUF_UNLOCK(&so->so_rcv); 3785 SOCK_UNLOCK(so); 3786 SOLISTEN_UNLOCK(head); 3787 } 3788 return; 3789 } 3790 SOCK_UNLOCK(so); 3791 wakeup(&so->so_timeo); 3792 sorwakeup(so); 3793 sowwakeup(so); 3794 } 3795 3796 void 3797 soisdisconnecting(struct socket *so) 3798 { 3799 3800 SOCK_LOCK(so); 3801 so->so_state &= ~SS_ISCONNECTING; 3802 so->so_state |= SS_ISDISCONNECTING; 3803 3804 if (!SOLISTENING(so)) { 3805 SOCKBUF_LOCK(&so->so_rcv); 3806 socantrcvmore_locked(so); 3807 SOCKBUF_LOCK(&so->so_snd); 3808 socantsendmore_locked(so); 3809 } 3810 SOCK_UNLOCK(so); 3811 wakeup(&so->so_timeo); 3812 } 3813 3814 void 3815 soisdisconnected(struct socket *so) 3816 { 3817 3818 SOCK_LOCK(so); 3819 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 3820 so->so_state |= SS_ISDISCONNECTED; 3821 3822 if (!SOLISTENING(so)) { 3823 SOCK_UNLOCK(so); 3824 SOCKBUF_LOCK(&so->so_rcv); 3825 socantrcvmore_locked(so); 3826 SOCKBUF_LOCK(&so->so_snd); 3827 sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); 3828 socantsendmore_locked(so); 3829 } else 3830 SOCK_UNLOCK(so); 3831 wakeup(&so->so_timeo); 3832 } 3833 3834 /* 3835 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 3836 */ 3837 struct sockaddr * 3838 sodupsockaddr(const struct sockaddr *sa, int mflags) 3839 { 3840 struct sockaddr *sa2; 3841 3842 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 3843 if (sa2) 3844 bcopy(sa, sa2, sa->sa_len); 3845 return sa2; 3846 } 3847 3848 /* 3849 * Register per-socket destructor. 3850 */ 3851 void 3852 sodtor_set(struct socket *so, so_dtor_t *func) 3853 { 3854 3855 SOCK_LOCK_ASSERT(so); 3856 so->so_dtor = func; 3857 } 3858 3859 /* 3860 * Register per-socket buffer upcalls. 3861 */ 3862 void 3863 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg) 3864 { 3865 struct sockbuf *sb; 3866 3867 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 3868 3869 switch (which) { 3870 case SO_RCV: 3871 sb = &so->so_rcv; 3872 break; 3873 case SO_SND: 3874 sb = &so->so_snd; 3875 break; 3876 default: 3877 panic("soupcall_set: bad which"); 3878 } 3879 SOCKBUF_LOCK_ASSERT(sb); 3880 sb->sb_upcall = func; 3881 sb->sb_upcallarg = arg; 3882 sb->sb_flags |= SB_UPCALL; 3883 } 3884 3885 void 3886 soupcall_clear(struct socket *so, int which) 3887 { 3888 struct sockbuf *sb; 3889 3890 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 3891 3892 switch (which) { 3893 case SO_RCV: 3894 sb = &so->so_rcv; 3895 break; 3896 case SO_SND: 3897 sb = &so->so_snd; 3898 break; 3899 default: 3900 panic("soupcall_clear: bad which"); 3901 } 3902 SOCKBUF_LOCK_ASSERT(sb); 3903 KASSERT(sb->sb_upcall != NULL, 3904 ("%s: so %p no upcall to clear", __func__, so)); 3905 sb->sb_upcall = NULL; 3906 sb->sb_upcallarg = NULL; 3907 sb->sb_flags &= ~SB_UPCALL; 3908 } 3909 3910 void 3911 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg) 3912 { 3913 3914 SOLISTEN_LOCK_ASSERT(so); 3915 so->sol_upcall = func; 3916 so->sol_upcallarg = arg; 3917 } 3918 3919 static void 3920 so_rdknl_lock(void *arg) 3921 { 3922 struct socket *so = arg; 3923 3924 if (SOLISTENING(so)) 3925 SOCK_LOCK(so); 3926 else 3927 SOCKBUF_LOCK(&so->so_rcv); 3928 } 3929 3930 static void 3931 so_rdknl_unlock(void *arg) 3932 { 3933 struct socket *so = arg; 3934 3935 if (SOLISTENING(so)) 3936 SOCK_UNLOCK(so); 3937 else 3938 SOCKBUF_UNLOCK(&so->so_rcv); 3939 } 3940 3941 static void 3942 so_rdknl_assert_locked(void *arg) 3943 { 3944 struct socket *so = arg; 3945 3946 if (SOLISTENING(so)) 3947 SOCK_LOCK_ASSERT(so); 3948 else 3949 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3950 } 3951 3952 static void 3953 so_rdknl_assert_unlocked(void *arg) 3954 { 3955 struct socket *so = arg; 3956 3957 if (SOLISTENING(so)) 3958 SOCK_UNLOCK_ASSERT(so); 3959 else 3960 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); 3961 } 3962 3963 static void 3964 so_wrknl_lock(void *arg) 3965 { 3966 struct socket *so = arg; 3967 3968 if (SOLISTENING(so)) 3969 SOCK_LOCK(so); 3970 else 3971 SOCKBUF_LOCK(&so->so_snd); 3972 } 3973 3974 static void 3975 so_wrknl_unlock(void *arg) 3976 { 3977 struct socket *so = arg; 3978 3979 if (SOLISTENING(so)) 3980 SOCK_UNLOCK(so); 3981 else 3982 SOCKBUF_UNLOCK(&so->so_snd); 3983 } 3984 3985 static void 3986 so_wrknl_assert_locked(void *arg) 3987 { 3988 struct socket *so = arg; 3989 3990 if (SOLISTENING(so)) 3991 SOCK_LOCK_ASSERT(so); 3992 else 3993 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3994 } 3995 3996 static void 3997 so_wrknl_assert_unlocked(void *arg) 3998 { 3999 struct socket *so = arg; 4000 4001 if (SOLISTENING(so)) 4002 SOCK_UNLOCK_ASSERT(so); 4003 else 4004 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 4005 } 4006 4007 /* 4008 * Create an external-format (``xsocket'') structure using the information in 4009 * the kernel-format socket structure pointed to by so. This is done to 4010 * reduce the spew of irrelevant information over this interface, to isolate 4011 * user code from changes in the kernel structure, and potentially to provide 4012 * information-hiding if we decide that some of this information should be 4013 * hidden from users. 4014 */ 4015 void 4016 sotoxsocket(struct socket *so, struct xsocket *xso) 4017 { 4018 4019 bzero(xso, sizeof(*xso)); 4020 xso->xso_len = sizeof *xso; 4021 xso->xso_so = (uintptr_t)so; 4022 xso->so_type = so->so_type; 4023 xso->so_options = so->so_options; 4024 xso->so_linger = so->so_linger; 4025 xso->so_state = so->so_state; 4026 xso->so_pcb = (uintptr_t)so->so_pcb; 4027 xso->xso_protocol = so->so_proto->pr_protocol; 4028 xso->xso_family = so->so_proto->pr_domain->dom_family; 4029 xso->so_timeo = so->so_timeo; 4030 xso->so_error = so->so_error; 4031 xso->so_uid = so->so_cred->cr_uid; 4032 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 4033 if (SOLISTENING(so)) { 4034 xso->so_qlen = so->sol_qlen; 4035 xso->so_incqlen = so->sol_incqlen; 4036 xso->so_qlimit = so->sol_qlimit; 4037 xso->so_oobmark = 0; 4038 } else { 4039 xso->so_state |= so->so_qstate; 4040 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0; 4041 xso->so_oobmark = so->so_oobmark; 4042 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 4043 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 4044 } 4045 } 4046 4047 struct sockbuf * 4048 so_sockbuf_rcv(struct socket *so) 4049 { 4050 4051 return (&so->so_rcv); 4052 } 4053 4054 struct sockbuf * 4055 so_sockbuf_snd(struct socket *so) 4056 { 4057 4058 return (&so->so_snd); 4059 } 4060 4061 int 4062 so_state_get(const struct socket *so) 4063 { 4064 4065 return (so->so_state); 4066 } 4067 4068 void 4069 so_state_set(struct socket *so, int val) 4070 { 4071 4072 so->so_state = val; 4073 } 4074 4075 int 4076 so_options_get(const struct socket *so) 4077 { 4078 4079 return (so->so_options); 4080 } 4081 4082 void 4083 so_options_set(struct socket *so, int val) 4084 { 4085 4086 so->so_options = val; 4087 } 4088 4089 int 4090 so_error_get(const struct socket *so) 4091 { 4092 4093 return (so->so_error); 4094 } 4095 4096 void 4097 so_error_set(struct socket *so, int val) 4098 { 4099 4100 so->so_error = val; 4101 } 4102 4103 int 4104 so_linger_get(const struct socket *so) 4105 { 4106 4107 return (so->so_linger); 4108 } 4109 4110 void 4111 so_linger_set(struct socket *so, int val) 4112 { 4113 4114 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz), 4115 ("%s: val %d out of range", __func__, val)); 4116 4117 so->so_linger = val; 4118 } 4119 4120 struct protosw * 4121 so_protosw_get(const struct socket *so) 4122 { 4123 4124 return (so->so_proto); 4125 } 4126 4127 void 4128 so_protosw_set(struct socket *so, struct protosw *val) 4129 { 4130 4131 so->so_proto = val; 4132 } 4133 4134 void 4135 so_sorwakeup(struct socket *so) 4136 { 4137 4138 sorwakeup(so); 4139 } 4140 4141 void 4142 so_sowwakeup(struct socket *so) 4143 { 4144 4145 sowwakeup(so); 4146 } 4147 4148 void 4149 so_sorwakeup_locked(struct socket *so) 4150 { 4151 4152 sorwakeup_locked(so); 4153 } 4154 4155 void 4156 so_sowwakeup_locked(struct socket *so) 4157 { 4158 4159 sowwakeup_locked(so); 4160 } 4161 4162 void 4163 so_lock(struct socket *so) 4164 { 4165 4166 SOCK_LOCK(so); 4167 } 4168 4169 void 4170 so_unlock(struct socket *so) 4171 { 4172 4173 SOCK_UNLOCK(so); 4174 } 4175