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 } 890 891 /* 892 * Return single connection off a listening socket queue. Main consumer of 893 * the function is kern_accept4(). Some modules, that do their own accept 894 * management also use the function. 895 * 896 * Listening socket must be locked on entry and is returned unlocked on 897 * return. 898 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT. 899 */ 900 int 901 solisten_dequeue(struct socket *head, struct socket **ret, int flags) 902 { 903 struct socket *so; 904 int error; 905 906 SOLISTEN_LOCK_ASSERT(head); 907 908 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) && 909 head->so_error == 0) { 910 error = msleep(&head->sol_comp, &head->so_lock, PSOCK | PCATCH, 911 "accept", 0); 912 if (error != 0) { 913 SOLISTEN_UNLOCK(head); 914 return (error); 915 } 916 } 917 if (head->so_error) { 918 error = head->so_error; 919 head->so_error = 0; 920 SOLISTEN_UNLOCK(head); 921 return (error); 922 } 923 if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp)) { 924 SOLISTEN_UNLOCK(head); 925 return (EWOULDBLOCK); 926 } 927 so = TAILQ_FIRST(&head->sol_comp); 928 SOCK_LOCK(so); 929 KASSERT(so->so_qstate == SQ_COMP, 930 ("%s: so %p not SQ_COMP", __func__, so)); 931 soref(so); 932 head->sol_qlen--; 933 so->so_qstate = SQ_NONE; 934 so->so_listen = NULL; 935 TAILQ_REMOVE(&head->sol_comp, so, so_list); 936 if (flags & ACCEPT4_INHERIT) 937 so->so_state |= (head->so_state & SS_NBIO); 938 else 939 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0; 940 SOCK_UNLOCK(so); 941 sorele(head); 942 943 *ret = so; 944 return (0); 945 } 946 947 /* 948 * Evaluate the reference count and named references on a socket; if no 949 * references remain, free it. This should be called whenever a reference is 950 * released, such as in sorele(), but also when named reference flags are 951 * cleared in socket or protocol code. 952 * 953 * sofree() will free the socket if: 954 * 955 * - There are no outstanding file descriptor references or related consumers 956 * (so_count == 0). 957 * 958 * - The socket has been closed by user space, if ever open (SS_NOFDREF). 959 * 960 * - The protocol does not have an outstanding strong reference on the socket 961 * (SS_PROTOREF). 962 * 963 * - The socket is not in a completed connection queue, so a process has been 964 * notified that it is present. If it is removed, the user process may 965 * block in accept() despite select() saying the socket was ready. 966 */ 967 void 968 sofree(struct socket *so) 969 { 970 struct protosw *pr = so->so_proto; 971 972 SOCK_LOCK_ASSERT(so); 973 974 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 || 975 (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) { 976 SOCK_UNLOCK(so); 977 return; 978 } 979 980 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) { 981 struct socket *sol; 982 983 sol = so->so_listen; 984 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so)); 985 986 /* 987 * To solve race between close of a listening socket and 988 * a socket on its incomplete queue, we need to lock both. 989 * The order is first listening socket, then regular. 990 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this 991 * function and the listening socket are the only pointers 992 * to so. To preserve so and sol, we reference both and then 993 * relock. 994 * After relock the socket may not move to so_comp since it 995 * doesn't have PCB already, but it may be removed from 996 * so_incomp. If that happens, we share responsiblity on 997 * freeing the socket, but soclose() has already removed 998 * it from queue. 999 */ 1000 soref(sol); 1001 soref(so); 1002 SOCK_UNLOCK(so); 1003 SOLISTEN_LOCK(sol); 1004 SOCK_LOCK(so); 1005 if (so->so_qstate == SQ_INCOMP) { 1006 KASSERT(so->so_listen == sol, 1007 ("%s: so %p migrated out of sol %p", 1008 __func__, so, sol)); 1009 TAILQ_REMOVE(&sol->sol_incomp, so, so_list); 1010 sol->sol_incqlen--; 1011 /* This is guarenteed not to be the last. */ 1012 refcount_release(&sol->so_count); 1013 so->so_qstate = SQ_NONE; 1014 so->so_listen = NULL; 1015 } else 1016 KASSERT(so->so_listen == NULL, 1017 ("%s: so %p not on (in)comp with so_listen", 1018 __func__, so)); 1019 sorele(sol); 1020 KASSERT(so->so_count == 1, 1021 ("%s: so %p count %u", __func__, so, so->so_count)); 1022 so->so_count = 0; 1023 } 1024 if (SOLISTENING(so)) 1025 so->so_error = ECONNABORTED; 1026 SOCK_UNLOCK(so); 1027 1028 VNET_SO_ASSERT(so); 1029 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 1030 (*pr->pr_domain->dom_dispose)(so); 1031 if (pr->pr_usrreqs->pru_detach != NULL) 1032 (*pr->pr_usrreqs->pru_detach)(so); 1033 1034 /* 1035 * From this point on, we assume that no other references to this 1036 * socket exist anywhere else in the stack. Therefore, no locks need 1037 * to be acquired or held. 1038 * 1039 * We used to do a lot of socket buffer and socket locking here, as 1040 * well as invoke sorflush() and perform wakeups. The direct call to 1041 * dom_dispose() and sbrelease_internal() are an inlining of what was 1042 * necessary from sorflush(). 1043 * 1044 * Notice that the socket buffer and kqueue state are torn down 1045 * before calling pru_detach. This means that protocols shold not 1046 * assume they can perform socket wakeups, etc, in their detach code. 1047 */ 1048 if (!SOLISTENING(so)) { 1049 sbdestroy(&so->so_snd, so); 1050 sbdestroy(&so->so_rcv, so); 1051 } 1052 seldrain(&so->so_rdsel); 1053 seldrain(&so->so_wrsel); 1054 knlist_destroy(&so->so_rdsel.si_note); 1055 knlist_destroy(&so->so_wrsel.si_note); 1056 sodealloc(so); 1057 } 1058 1059 /* 1060 * Close a socket on last file table reference removal. Initiate disconnect 1061 * if connected. Free socket when disconnect complete. 1062 * 1063 * This function will sorele() the socket. Note that soclose() may be called 1064 * prior to the ref count reaching zero. The actual socket structure will 1065 * not be freed until the ref count reaches zero. 1066 */ 1067 int 1068 soclose(struct socket *so) 1069 { 1070 struct accept_queue lqueue; 1071 bool listening; 1072 int error = 0; 1073 1074 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter")); 1075 1076 CURVNET_SET(so->so_vnet); 1077 funsetown(&so->so_sigio); 1078 if (so->so_state & SS_ISCONNECTED) { 1079 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 1080 error = sodisconnect(so); 1081 if (error) { 1082 if (error == ENOTCONN) 1083 error = 0; 1084 goto drop; 1085 } 1086 } 1087 if (so->so_options & SO_LINGER) { 1088 if ((so->so_state & SS_ISDISCONNECTING) && 1089 (so->so_state & SS_NBIO)) 1090 goto drop; 1091 while (so->so_state & SS_ISCONNECTED) { 1092 error = tsleep(&so->so_timeo, 1093 PSOCK | PCATCH, "soclos", 1094 so->so_linger * hz); 1095 if (error) 1096 break; 1097 } 1098 } 1099 } 1100 1101 drop: 1102 if (so->so_proto->pr_usrreqs->pru_close != NULL) 1103 (*so->so_proto->pr_usrreqs->pru_close)(so); 1104 if (so->so_dtor != NULL) 1105 so->so_dtor(so); 1106 1107 SOCK_LOCK(so); 1108 if ((listening = (so->so_options & SO_ACCEPTCONN))) { 1109 struct socket *sp; 1110 1111 TAILQ_INIT(&lqueue); 1112 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list); 1113 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list); 1114 1115 so->sol_qlen = so->sol_incqlen = 0; 1116 1117 TAILQ_FOREACH(sp, &lqueue, so_list) { 1118 SOCK_LOCK(sp); 1119 sp->so_qstate = SQ_NONE; 1120 sp->so_listen = NULL; 1121 SOCK_UNLOCK(sp); 1122 /* Guaranteed not to be the last. */ 1123 refcount_release(&so->so_count); 1124 } 1125 } 1126 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF")); 1127 so->so_state |= SS_NOFDREF; 1128 sorele(so); 1129 if (listening) { 1130 struct socket *sp; 1131 1132 TAILQ_FOREACH(sp, &lqueue, so_list) { 1133 SOCK_LOCK(sp); 1134 if (sp->so_count == 0) { 1135 SOCK_UNLOCK(sp); 1136 soabort(sp); 1137 } else 1138 /* sp is now in sofree() */ 1139 SOCK_UNLOCK(sp); 1140 } 1141 } 1142 CURVNET_RESTORE(); 1143 return (error); 1144 } 1145 1146 /* 1147 * soabort() is used to abruptly tear down a connection, such as when a 1148 * resource limit is reached (listen queue depth exceeded), or if a listen 1149 * socket is closed while there are sockets waiting to be accepted. 1150 * 1151 * This interface is tricky, because it is called on an unreferenced socket, 1152 * and must be called only by a thread that has actually removed the socket 1153 * from the listen queue it was on, or races with other threads are risked. 1154 * 1155 * This interface will call into the protocol code, so must not be called 1156 * with any socket locks held. Protocols do call it while holding their own 1157 * recursible protocol mutexes, but this is something that should be subject 1158 * to review in the future. 1159 */ 1160 void 1161 soabort(struct socket *so) 1162 { 1163 1164 /* 1165 * In as much as is possible, assert that no references to this 1166 * socket are held. This is not quite the same as asserting that the 1167 * current thread is responsible for arranging for no references, but 1168 * is as close as we can get for now. 1169 */ 1170 KASSERT(so->so_count == 0, ("soabort: so_count")); 1171 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF")); 1172 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF")); 1173 KASSERT(so->so_qstate == SQ_NONE, ("soabort: !SQ_NONE")); 1174 VNET_SO_ASSERT(so); 1175 1176 if (so->so_proto->pr_usrreqs->pru_abort != NULL) 1177 (*so->so_proto->pr_usrreqs->pru_abort)(so); 1178 SOCK_LOCK(so); 1179 sofree(so); 1180 } 1181 1182 int 1183 soaccept(struct socket *so, struct sockaddr **nam) 1184 { 1185 int error; 1186 1187 SOCK_LOCK(so); 1188 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF")); 1189 so->so_state &= ~SS_NOFDREF; 1190 SOCK_UNLOCK(so); 1191 1192 CURVNET_SET(so->so_vnet); 1193 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); 1194 CURVNET_RESTORE(); 1195 return (error); 1196 } 1197 1198 int 1199 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 1200 { 1201 1202 return (soconnectat(AT_FDCWD, so, nam, td)); 1203 } 1204 1205 int 1206 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 1207 { 1208 int error; 1209 1210 if (so->so_options & SO_ACCEPTCONN) 1211 return (EOPNOTSUPP); 1212 1213 CURVNET_SET(so->so_vnet); 1214 /* 1215 * If protocol is connection-based, can only connect once. 1216 * Otherwise, if connected, try to disconnect first. This allows 1217 * user to disconnect by connecting to, e.g., a null address. 1218 */ 1219 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 1220 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 1221 (error = sodisconnect(so)))) { 1222 error = EISCONN; 1223 } else { 1224 /* 1225 * Prevent accumulated error from previous connection from 1226 * biting us. 1227 */ 1228 so->so_error = 0; 1229 if (fd == AT_FDCWD) { 1230 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, 1231 nam, td); 1232 } else { 1233 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd, 1234 so, nam, td); 1235 } 1236 } 1237 CURVNET_RESTORE(); 1238 1239 return (error); 1240 } 1241 1242 int 1243 soconnect2(struct socket *so1, struct socket *so2) 1244 { 1245 int error; 1246 1247 CURVNET_SET(so1->so_vnet); 1248 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); 1249 CURVNET_RESTORE(); 1250 return (error); 1251 } 1252 1253 int 1254 sodisconnect(struct socket *so) 1255 { 1256 int error; 1257 1258 if ((so->so_state & SS_ISCONNECTED) == 0) 1259 return (ENOTCONN); 1260 if (so->so_state & SS_ISDISCONNECTING) 1261 return (EALREADY); 1262 VNET_SO_ASSERT(so); 1263 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 1264 return (error); 1265 } 1266 1267 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) 1268 1269 int 1270 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1271 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1272 { 1273 long space; 1274 ssize_t resid; 1275 int clen = 0, error, dontroute; 1276 1277 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 1278 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 1279 ("sosend_dgram: !PR_ATOMIC")); 1280 1281 if (uio != NULL) 1282 resid = uio->uio_resid; 1283 else 1284 resid = top->m_pkthdr.len; 1285 /* 1286 * In theory resid should be unsigned. However, space must be 1287 * signed, as it might be less than 0 if we over-committed, and we 1288 * must use a signed comparison of space and resid. On the other 1289 * hand, a negative resid causes us to loop sending 0-length 1290 * segments to the protocol. 1291 */ 1292 if (resid < 0) { 1293 error = EINVAL; 1294 goto out; 1295 } 1296 1297 dontroute = 1298 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 1299 if (td != NULL) 1300 td->td_ru.ru_msgsnd++; 1301 if (control != NULL) 1302 clen = control->m_len; 1303 1304 SOCKBUF_LOCK(&so->so_snd); 1305 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1306 SOCKBUF_UNLOCK(&so->so_snd); 1307 error = EPIPE; 1308 goto out; 1309 } 1310 if (so->so_error) { 1311 error = so->so_error; 1312 so->so_error = 0; 1313 SOCKBUF_UNLOCK(&so->so_snd); 1314 goto out; 1315 } 1316 if ((so->so_state & SS_ISCONNECTED) == 0) { 1317 /* 1318 * `sendto' and `sendmsg' is allowed on a connection-based 1319 * socket if it supports implied connect. Return ENOTCONN if 1320 * not connected and no address is supplied. 1321 */ 1322 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1323 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1324 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1325 !(resid == 0 && clen != 0)) { 1326 SOCKBUF_UNLOCK(&so->so_snd); 1327 error = ENOTCONN; 1328 goto out; 1329 } 1330 } else if (addr == NULL) { 1331 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1332 error = ENOTCONN; 1333 else 1334 error = EDESTADDRREQ; 1335 SOCKBUF_UNLOCK(&so->so_snd); 1336 goto out; 1337 } 1338 } 1339 1340 /* 1341 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1342 * problem and need fixing. 1343 */ 1344 space = sbspace(&so->so_snd); 1345 if (flags & MSG_OOB) 1346 space += 1024; 1347 space -= clen; 1348 SOCKBUF_UNLOCK(&so->so_snd); 1349 if (resid > space) { 1350 error = EMSGSIZE; 1351 goto out; 1352 } 1353 if (uio == NULL) { 1354 resid = 0; 1355 if (flags & MSG_EOR) 1356 top->m_flags |= M_EOR; 1357 } else { 1358 /* 1359 * Copy the data from userland into a mbuf chain. 1360 * If no data is to be copied in, a single empty mbuf 1361 * is returned. 1362 */ 1363 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1364 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1365 if (top == NULL) { 1366 error = EFAULT; /* only possible error */ 1367 goto out; 1368 } 1369 space -= resid - uio->uio_resid; 1370 resid = uio->uio_resid; 1371 } 1372 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1373 /* 1374 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1375 * than with. 1376 */ 1377 if (dontroute) { 1378 SOCK_LOCK(so); 1379 so->so_options |= SO_DONTROUTE; 1380 SOCK_UNLOCK(so); 1381 } 1382 /* 1383 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1384 * of date. We could have received a reset packet in an interrupt or 1385 * maybe we slept while doing page faults in uiomove() etc. We could 1386 * probably recheck again inside the locking protection here, but 1387 * there are probably other places that this also happens. We must 1388 * rethink this. 1389 */ 1390 VNET_SO_ASSERT(so); 1391 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1392 (flags & MSG_OOB) ? PRUS_OOB : 1393 /* 1394 * If the user set MSG_EOF, the protocol understands this flag and 1395 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1396 */ 1397 ((flags & MSG_EOF) && 1398 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1399 (resid <= 0)) ? 1400 PRUS_EOF : 1401 /* If there is more to send set PRUS_MORETOCOME */ 1402 (flags & MSG_MORETOCOME) || 1403 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1404 top, addr, control, td); 1405 if (dontroute) { 1406 SOCK_LOCK(so); 1407 so->so_options &= ~SO_DONTROUTE; 1408 SOCK_UNLOCK(so); 1409 } 1410 clen = 0; 1411 control = NULL; 1412 top = NULL; 1413 out: 1414 if (top != NULL) 1415 m_freem(top); 1416 if (control != NULL) 1417 m_freem(control); 1418 return (error); 1419 } 1420 1421 /* 1422 * Send on a socket. If send must go all at once and message is larger than 1423 * send buffering, then hard error. Lock against other senders. If must go 1424 * all at once and not enough room now, then inform user that this would 1425 * block and do nothing. Otherwise, if nonblocking, send as much as 1426 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1427 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1428 * in mbuf chain must be small enough to send all at once. 1429 * 1430 * Returns nonzero on error, timeout or signal; callers must check for short 1431 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1432 * on return. 1433 */ 1434 int 1435 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1436 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1437 { 1438 long space; 1439 ssize_t resid; 1440 int clen = 0, error, dontroute; 1441 int atomic = sosendallatonce(so) || top; 1442 1443 if (uio != NULL) 1444 resid = uio->uio_resid; 1445 else 1446 resid = top->m_pkthdr.len; 1447 /* 1448 * In theory resid should be unsigned. However, space must be 1449 * signed, as it might be less than 0 if we over-committed, and we 1450 * must use a signed comparison of space and resid. On the other 1451 * hand, a negative resid causes us to loop sending 0-length 1452 * segments to the protocol. 1453 * 1454 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1455 * type sockets since that's an error. 1456 */ 1457 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1458 error = EINVAL; 1459 goto out; 1460 } 1461 1462 dontroute = 1463 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1464 (so->so_proto->pr_flags & PR_ATOMIC); 1465 if (td != NULL) 1466 td->td_ru.ru_msgsnd++; 1467 if (control != NULL) 1468 clen = control->m_len; 1469 1470 error = sblock(&so->so_snd, SBLOCKWAIT(flags)); 1471 if (error) 1472 goto out; 1473 1474 restart: 1475 do { 1476 SOCKBUF_LOCK(&so->so_snd); 1477 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1478 SOCKBUF_UNLOCK(&so->so_snd); 1479 error = EPIPE; 1480 goto release; 1481 } 1482 if (so->so_error) { 1483 error = so->so_error; 1484 so->so_error = 0; 1485 SOCKBUF_UNLOCK(&so->so_snd); 1486 goto release; 1487 } 1488 if ((so->so_state & SS_ISCONNECTED) == 0) { 1489 /* 1490 * `sendto' and `sendmsg' is allowed on a connection- 1491 * based socket if it supports implied connect. 1492 * Return ENOTCONN if not connected and no address is 1493 * supplied. 1494 */ 1495 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1496 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1497 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1498 !(resid == 0 && clen != 0)) { 1499 SOCKBUF_UNLOCK(&so->so_snd); 1500 error = ENOTCONN; 1501 goto release; 1502 } 1503 } else if (addr == NULL) { 1504 SOCKBUF_UNLOCK(&so->so_snd); 1505 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1506 error = ENOTCONN; 1507 else 1508 error = EDESTADDRREQ; 1509 goto release; 1510 } 1511 } 1512 space = sbspace(&so->so_snd); 1513 if (flags & MSG_OOB) 1514 space += 1024; 1515 if ((atomic && resid > so->so_snd.sb_hiwat) || 1516 clen > so->so_snd.sb_hiwat) { 1517 SOCKBUF_UNLOCK(&so->so_snd); 1518 error = EMSGSIZE; 1519 goto release; 1520 } 1521 if (space < resid + clen && 1522 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1523 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) { 1524 SOCKBUF_UNLOCK(&so->so_snd); 1525 error = EWOULDBLOCK; 1526 goto release; 1527 } 1528 error = sbwait(&so->so_snd); 1529 SOCKBUF_UNLOCK(&so->so_snd); 1530 if (error) 1531 goto release; 1532 goto restart; 1533 } 1534 SOCKBUF_UNLOCK(&so->so_snd); 1535 space -= clen; 1536 do { 1537 if (uio == NULL) { 1538 resid = 0; 1539 if (flags & MSG_EOR) 1540 top->m_flags |= M_EOR; 1541 } else { 1542 /* 1543 * Copy the data from userland into a mbuf 1544 * chain. If resid is 0, which can happen 1545 * only if we have control to send, then 1546 * a single empty mbuf is returned. This 1547 * is a workaround to prevent protocol send 1548 * methods to panic. 1549 */ 1550 top = m_uiotombuf(uio, M_WAITOK, space, 1551 (atomic ? max_hdr : 0), 1552 (atomic ? M_PKTHDR : 0) | 1553 ((flags & MSG_EOR) ? M_EOR : 0)); 1554 if (top == NULL) { 1555 error = EFAULT; /* only possible error */ 1556 goto release; 1557 } 1558 space -= resid - uio->uio_resid; 1559 resid = uio->uio_resid; 1560 } 1561 if (dontroute) { 1562 SOCK_LOCK(so); 1563 so->so_options |= SO_DONTROUTE; 1564 SOCK_UNLOCK(so); 1565 } 1566 /* 1567 * XXX all the SBS_CANTSENDMORE checks previously 1568 * done could be out of date. We could have received 1569 * a reset packet in an interrupt or maybe we slept 1570 * while doing page faults in uiomove() etc. We 1571 * could probably recheck again inside the locking 1572 * protection here, but there are probably other 1573 * places that this also happens. We must rethink 1574 * this. 1575 */ 1576 VNET_SO_ASSERT(so); 1577 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1578 (flags & MSG_OOB) ? PRUS_OOB : 1579 /* 1580 * If the user set MSG_EOF, the protocol understands 1581 * this flag and nothing left to send then use 1582 * PRU_SEND_EOF instead of PRU_SEND. 1583 */ 1584 ((flags & MSG_EOF) && 1585 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1586 (resid <= 0)) ? 1587 PRUS_EOF : 1588 /* If there is more to send set PRUS_MORETOCOME. */ 1589 (flags & MSG_MORETOCOME) || 1590 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1591 top, addr, control, td); 1592 if (dontroute) { 1593 SOCK_LOCK(so); 1594 so->so_options &= ~SO_DONTROUTE; 1595 SOCK_UNLOCK(so); 1596 } 1597 clen = 0; 1598 control = NULL; 1599 top = NULL; 1600 if (error) 1601 goto release; 1602 } while (resid && space > 0); 1603 } while (resid); 1604 1605 release: 1606 sbunlock(&so->so_snd); 1607 out: 1608 if (top != NULL) 1609 m_freem(top); 1610 if (control != NULL) 1611 m_freem(control); 1612 return (error); 1613 } 1614 1615 int 1616 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1617 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1618 { 1619 int error; 1620 1621 CURVNET_SET(so->so_vnet); 1622 if (!SOLISTENING(so)) 1623 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, 1624 top, control, flags, td); 1625 else { 1626 m_freem(top); 1627 m_freem(control); 1628 error = ENOTCONN; 1629 } 1630 CURVNET_RESTORE(); 1631 return (error); 1632 } 1633 1634 /* 1635 * The part of soreceive() that implements reading non-inline out-of-band 1636 * data from a socket. For more complete comments, see soreceive(), from 1637 * which this code originated. 1638 * 1639 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1640 * unable to return an mbuf chain to the caller. 1641 */ 1642 static int 1643 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1644 { 1645 struct protosw *pr = so->so_proto; 1646 struct mbuf *m; 1647 int error; 1648 1649 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1650 VNET_SO_ASSERT(so); 1651 1652 m = m_get(M_WAITOK, MT_DATA); 1653 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 1654 if (error) 1655 goto bad; 1656 do { 1657 error = uiomove(mtod(m, void *), 1658 (int) min(uio->uio_resid, m->m_len), uio); 1659 m = m_free(m); 1660 } while (uio->uio_resid && error == 0 && m); 1661 bad: 1662 if (m != NULL) 1663 m_freem(m); 1664 return (error); 1665 } 1666 1667 /* 1668 * Following replacement or removal of the first mbuf on the first mbuf chain 1669 * of a socket buffer, push necessary state changes back into the socket 1670 * buffer so that other consumers see the values consistently. 'nextrecord' 1671 * is the callers locally stored value of the original value of 1672 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1673 * NOTE: 'nextrecord' may be NULL. 1674 */ 1675 static __inline void 1676 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1677 { 1678 1679 SOCKBUF_LOCK_ASSERT(sb); 1680 /* 1681 * First, update for the new value of nextrecord. If necessary, make 1682 * it the first record. 1683 */ 1684 if (sb->sb_mb != NULL) 1685 sb->sb_mb->m_nextpkt = nextrecord; 1686 else 1687 sb->sb_mb = nextrecord; 1688 1689 /* 1690 * Now update any dependent socket buffer fields to reflect the new 1691 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1692 * addition of a second clause that takes care of the case where 1693 * sb_mb has been updated, but remains the last record. 1694 */ 1695 if (sb->sb_mb == NULL) { 1696 sb->sb_mbtail = NULL; 1697 sb->sb_lastrecord = NULL; 1698 } else if (sb->sb_mb->m_nextpkt == NULL) 1699 sb->sb_lastrecord = sb->sb_mb; 1700 } 1701 1702 /* 1703 * Implement receive operations on a socket. We depend on the way that 1704 * records are added to the sockbuf by sbappend. In particular, each record 1705 * (mbufs linked through m_next) must begin with an address if the protocol 1706 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1707 * data, and then zero or more mbufs of data. In order to allow parallelism 1708 * between network receive and copying to user space, as well as avoid 1709 * sleeping with a mutex held, we release the socket buffer mutex during the 1710 * user space copy. Although the sockbuf is locked, new data may still be 1711 * appended, and thus we must maintain consistency of the sockbuf during that 1712 * time. 1713 * 1714 * The caller may receive the data as a single mbuf chain by supplying an 1715 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1716 * the count in uio_resid. 1717 */ 1718 int 1719 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1720 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1721 { 1722 struct mbuf *m, **mp; 1723 int flags, error, offset; 1724 ssize_t len; 1725 struct protosw *pr = so->so_proto; 1726 struct mbuf *nextrecord; 1727 int moff, type = 0; 1728 ssize_t orig_resid = uio->uio_resid; 1729 1730 mp = mp0; 1731 if (psa != NULL) 1732 *psa = NULL; 1733 if (controlp != NULL) 1734 *controlp = NULL; 1735 if (flagsp != NULL) 1736 flags = *flagsp &~ MSG_EOR; 1737 else 1738 flags = 0; 1739 if (flags & MSG_OOB) 1740 return (soreceive_rcvoob(so, uio, flags)); 1741 if (mp != NULL) 1742 *mp = NULL; 1743 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 1744 && uio->uio_resid) { 1745 VNET_SO_ASSERT(so); 1746 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 1747 } 1748 1749 error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); 1750 if (error) 1751 return (error); 1752 1753 restart: 1754 SOCKBUF_LOCK(&so->so_rcv); 1755 m = so->so_rcv.sb_mb; 1756 /* 1757 * If we have less data than requested, block awaiting more (subject 1758 * to any timeout) if: 1759 * 1. the current count is less than the low water mark, or 1760 * 2. MSG_DONTWAIT is not set 1761 */ 1762 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1763 sbavail(&so->so_rcv) < uio->uio_resid) && 1764 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && 1765 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 1766 KASSERT(m != NULL || !sbavail(&so->so_rcv), 1767 ("receive: m == %p sbavail == %u", 1768 m, sbavail(&so->so_rcv))); 1769 if (so->so_error) { 1770 if (m != NULL) 1771 goto dontblock; 1772 error = so->so_error; 1773 if ((flags & MSG_PEEK) == 0) 1774 so->so_error = 0; 1775 SOCKBUF_UNLOCK(&so->so_rcv); 1776 goto release; 1777 } 1778 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1779 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1780 if (m == NULL) { 1781 SOCKBUF_UNLOCK(&so->so_rcv); 1782 goto release; 1783 } else 1784 goto dontblock; 1785 } 1786 for (; m != NULL; m = m->m_next) 1787 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1788 m = so->so_rcv.sb_mb; 1789 goto dontblock; 1790 } 1791 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1792 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1793 SOCKBUF_UNLOCK(&so->so_rcv); 1794 error = ENOTCONN; 1795 goto release; 1796 } 1797 if (uio->uio_resid == 0) { 1798 SOCKBUF_UNLOCK(&so->so_rcv); 1799 goto release; 1800 } 1801 if ((so->so_state & SS_NBIO) || 1802 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1803 SOCKBUF_UNLOCK(&so->so_rcv); 1804 error = EWOULDBLOCK; 1805 goto release; 1806 } 1807 SBLASTRECORDCHK(&so->so_rcv); 1808 SBLASTMBUFCHK(&so->so_rcv); 1809 error = sbwait(&so->so_rcv); 1810 SOCKBUF_UNLOCK(&so->so_rcv); 1811 if (error) 1812 goto release; 1813 goto restart; 1814 } 1815 dontblock: 1816 /* 1817 * From this point onward, we maintain 'nextrecord' as a cache of the 1818 * pointer to the next record in the socket buffer. We must keep the 1819 * various socket buffer pointers and local stack versions of the 1820 * pointers in sync, pushing out modifications before dropping the 1821 * socket buffer mutex, and re-reading them when picking it up. 1822 * 1823 * Otherwise, we will race with the network stack appending new data 1824 * or records onto the socket buffer by using inconsistent/stale 1825 * versions of the field, possibly resulting in socket buffer 1826 * corruption. 1827 * 1828 * By holding the high-level sblock(), we prevent simultaneous 1829 * readers from pulling off the front of the socket buffer. 1830 */ 1831 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1832 if (uio->uio_td) 1833 uio->uio_td->td_ru.ru_msgrcv++; 1834 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 1835 SBLASTRECORDCHK(&so->so_rcv); 1836 SBLASTMBUFCHK(&so->so_rcv); 1837 nextrecord = m->m_nextpkt; 1838 if (pr->pr_flags & PR_ADDR) { 1839 KASSERT(m->m_type == MT_SONAME, 1840 ("m->m_type == %d", m->m_type)); 1841 orig_resid = 0; 1842 if (psa != NULL) 1843 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 1844 M_NOWAIT); 1845 if (flags & MSG_PEEK) { 1846 m = m->m_next; 1847 } else { 1848 sbfree(&so->so_rcv, m); 1849 so->so_rcv.sb_mb = m_free(m); 1850 m = so->so_rcv.sb_mb; 1851 sockbuf_pushsync(&so->so_rcv, nextrecord); 1852 } 1853 } 1854 1855 /* 1856 * Process one or more MT_CONTROL mbufs present before any data mbufs 1857 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 1858 * just copy the data; if !MSG_PEEK, we call into the protocol to 1859 * perform externalization (or freeing if controlp == NULL). 1860 */ 1861 if (m != NULL && m->m_type == MT_CONTROL) { 1862 struct mbuf *cm = NULL, *cmn; 1863 struct mbuf **cme = &cm; 1864 1865 do { 1866 if (flags & MSG_PEEK) { 1867 if (controlp != NULL) { 1868 *controlp = m_copym(m, 0, m->m_len, 1869 M_NOWAIT); 1870 controlp = &(*controlp)->m_next; 1871 } 1872 m = m->m_next; 1873 } else { 1874 sbfree(&so->so_rcv, m); 1875 so->so_rcv.sb_mb = m->m_next; 1876 m->m_next = NULL; 1877 *cme = m; 1878 cme = &(*cme)->m_next; 1879 m = so->so_rcv.sb_mb; 1880 } 1881 } while (m != NULL && m->m_type == MT_CONTROL); 1882 if ((flags & MSG_PEEK) == 0) 1883 sockbuf_pushsync(&so->so_rcv, nextrecord); 1884 while (cm != NULL) { 1885 cmn = cm->m_next; 1886 cm->m_next = NULL; 1887 if (pr->pr_domain->dom_externalize != NULL) { 1888 SOCKBUF_UNLOCK(&so->so_rcv); 1889 VNET_SO_ASSERT(so); 1890 error = (*pr->pr_domain->dom_externalize) 1891 (cm, controlp, flags); 1892 SOCKBUF_LOCK(&so->so_rcv); 1893 } else if (controlp != NULL) 1894 *controlp = cm; 1895 else 1896 m_freem(cm); 1897 if (controlp != NULL) { 1898 orig_resid = 0; 1899 while (*controlp != NULL) 1900 controlp = &(*controlp)->m_next; 1901 } 1902 cm = cmn; 1903 } 1904 if (m != NULL) 1905 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 1906 else 1907 nextrecord = so->so_rcv.sb_mb; 1908 orig_resid = 0; 1909 } 1910 if (m != NULL) { 1911 if ((flags & MSG_PEEK) == 0) { 1912 KASSERT(m->m_nextpkt == nextrecord, 1913 ("soreceive: post-control, nextrecord !sync")); 1914 if (nextrecord == NULL) { 1915 KASSERT(so->so_rcv.sb_mb == m, 1916 ("soreceive: post-control, sb_mb!=m")); 1917 KASSERT(so->so_rcv.sb_lastrecord == m, 1918 ("soreceive: post-control, lastrecord!=m")); 1919 } 1920 } 1921 type = m->m_type; 1922 if (type == MT_OOBDATA) 1923 flags |= MSG_OOB; 1924 } else { 1925 if ((flags & MSG_PEEK) == 0) { 1926 KASSERT(so->so_rcv.sb_mb == nextrecord, 1927 ("soreceive: sb_mb != nextrecord")); 1928 if (so->so_rcv.sb_mb == NULL) { 1929 KASSERT(so->so_rcv.sb_lastrecord == NULL, 1930 ("soreceive: sb_lastercord != NULL")); 1931 } 1932 } 1933 } 1934 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1935 SBLASTRECORDCHK(&so->so_rcv); 1936 SBLASTMBUFCHK(&so->so_rcv); 1937 1938 /* 1939 * Now continue to read any data mbufs off of the head of the socket 1940 * buffer until the read request is satisfied. Note that 'type' is 1941 * used to store the type of any mbuf reads that have happened so far 1942 * such that soreceive() can stop reading if the type changes, which 1943 * causes soreceive() to return only one of regular data and inline 1944 * out-of-band data in a single socket receive operation. 1945 */ 1946 moff = 0; 1947 offset = 0; 1948 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 1949 && error == 0) { 1950 /* 1951 * If the type of mbuf has changed since the last mbuf 1952 * examined ('type'), end the receive operation. 1953 */ 1954 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1955 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 1956 if (type != m->m_type) 1957 break; 1958 } else if (type == MT_OOBDATA) 1959 break; 1960 else 1961 KASSERT(m->m_type == MT_DATA, 1962 ("m->m_type == %d", m->m_type)); 1963 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 1964 len = uio->uio_resid; 1965 if (so->so_oobmark && len > so->so_oobmark - offset) 1966 len = so->so_oobmark - offset; 1967 if (len > m->m_len - moff) 1968 len = m->m_len - moff; 1969 /* 1970 * If mp is set, just pass back the mbufs. Otherwise copy 1971 * them out via the uio, then free. Sockbuf must be 1972 * consistent here (points to current mbuf, it points to next 1973 * record) when we drop priority; we must note any additions 1974 * to the sockbuf when we block interrupts again. 1975 */ 1976 if (mp == NULL) { 1977 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1978 SBLASTRECORDCHK(&so->so_rcv); 1979 SBLASTMBUFCHK(&so->so_rcv); 1980 SOCKBUF_UNLOCK(&so->so_rcv); 1981 error = uiomove(mtod(m, char *) + moff, (int)len, uio); 1982 SOCKBUF_LOCK(&so->so_rcv); 1983 if (error) { 1984 /* 1985 * The MT_SONAME mbuf has already been removed 1986 * from the record, so it is necessary to 1987 * remove the data mbufs, if any, to preserve 1988 * the invariant in the case of PR_ADDR that 1989 * requires MT_SONAME mbufs at the head of 1990 * each record. 1991 */ 1992 if (pr->pr_flags & PR_ATOMIC && 1993 ((flags & MSG_PEEK) == 0)) 1994 (void)sbdroprecord_locked(&so->so_rcv); 1995 SOCKBUF_UNLOCK(&so->so_rcv); 1996 goto release; 1997 } 1998 } else 1999 uio->uio_resid -= len; 2000 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2001 if (len == m->m_len - moff) { 2002 if (m->m_flags & M_EOR) 2003 flags |= MSG_EOR; 2004 if (flags & MSG_PEEK) { 2005 m = m->m_next; 2006 moff = 0; 2007 } else { 2008 nextrecord = m->m_nextpkt; 2009 sbfree(&so->so_rcv, m); 2010 if (mp != NULL) { 2011 m->m_nextpkt = NULL; 2012 *mp = m; 2013 mp = &m->m_next; 2014 so->so_rcv.sb_mb = m = m->m_next; 2015 *mp = NULL; 2016 } else { 2017 so->so_rcv.sb_mb = m_free(m); 2018 m = so->so_rcv.sb_mb; 2019 } 2020 sockbuf_pushsync(&so->so_rcv, nextrecord); 2021 SBLASTRECORDCHK(&so->so_rcv); 2022 SBLASTMBUFCHK(&so->so_rcv); 2023 } 2024 } else { 2025 if (flags & MSG_PEEK) 2026 moff += len; 2027 else { 2028 if (mp != NULL) { 2029 if (flags & MSG_DONTWAIT) { 2030 *mp = m_copym(m, 0, len, 2031 M_NOWAIT); 2032 if (*mp == NULL) { 2033 /* 2034 * m_copym() couldn't 2035 * allocate an mbuf. 2036 * Adjust uio_resid back 2037 * (it was adjusted 2038 * down by len bytes, 2039 * which we didn't end 2040 * up "copying" over). 2041 */ 2042 uio->uio_resid += len; 2043 break; 2044 } 2045 } else { 2046 SOCKBUF_UNLOCK(&so->so_rcv); 2047 *mp = m_copym(m, 0, len, 2048 M_WAITOK); 2049 SOCKBUF_LOCK(&so->so_rcv); 2050 } 2051 } 2052 sbcut_locked(&so->so_rcv, len); 2053 } 2054 } 2055 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2056 if (so->so_oobmark) { 2057 if ((flags & MSG_PEEK) == 0) { 2058 so->so_oobmark -= len; 2059 if (so->so_oobmark == 0) { 2060 so->so_rcv.sb_state |= SBS_RCVATMARK; 2061 break; 2062 } 2063 } else { 2064 offset += len; 2065 if (offset == so->so_oobmark) 2066 break; 2067 } 2068 } 2069 if (flags & MSG_EOR) 2070 break; 2071 /* 2072 * If the MSG_WAITALL flag is set (for non-atomic socket), we 2073 * must not quit until "uio->uio_resid == 0" or an error 2074 * termination. If a signal/timeout occurs, return with a 2075 * short count but without error. Keep sockbuf locked 2076 * against other readers. 2077 */ 2078 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 2079 !sosendallatonce(so) && nextrecord == NULL) { 2080 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2081 if (so->so_error || 2082 so->so_rcv.sb_state & SBS_CANTRCVMORE) 2083 break; 2084 /* 2085 * Notify the protocol that some data has been 2086 * drained before blocking. 2087 */ 2088 if (pr->pr_flags & PR_WANTRCVD) { 2089 SOCKBUF_UNLOCK(&so->so_rcv); 2090 VNET_SO_ASSERT(so); 2091 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2092 SOCKBUF_LOCK(&so->so_rcv); 2093 } 2094 SBLASTRECORDCHK(&so->so_rcv); 2095 SBLASTMBUFCHK(&so->so_rcv); 2096 /* 2097 * We could receive some data while was notifying 2098 * the protocol. Skip blocking in this case. 2099 */ 2100 if (so->so_rcv.sb_mb == NULL) { 2101 error = sbwait(&so->so_rcv); 2102 if (error) { 2103 SOCKBUF_UNLOCK(&so->so_rcv); 2104 goto release; 2105 } 2106 } 2107 m = so->so_rcv.sb_mb; 2108 if (m != NULL) 2109 nextrecord = m->m_nextpkt; 2110 } 2111 } 2112 2113 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2114 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 2115 flags |= MSG_TRUNC; 2116 if ((flags & MSG_PEEK) == 0) 2117 (void) sbdroprecord_locked(&so->so_rcv); 2118 } 2119 if ((flags & MSG_PEEK) == 0) { 2120 if (m == NULL) { 2121 /* 2122 * First part is an inline SB_EMPTY_FIXUP(). Second 2123 * part makes sure sb_lastrecord is up-to-date if 2124 * there is still data in the socket buffer. 2125 */ 2126 so->so_rcv.sb_mb = nextrecord; 2127 if (so->so_rcv.sb_mb == NULL) { 2128 so->so_rcv.sb_mbtail = NULL; 2129 so->so_rcv.sb_lastrecord = NULL; 2130 } else if (nextrecord->m_nextpkt == NULL) 2131 so->so_rcv.sb_lastrecord = nextrecord; 2132 } 2133 SBLASTRECORDCHK(&so->so_rcv); 2134 SBLASTMBUFCHK(&so->so_rcv); 2135 /* 2136 * If soreceive() is being done from the socket callback, 2137 * then don't need to generate ACK to peer to update window, 2138 * since ACK will be generated on return to TCP. 2139 */ 2140 if (!(flags & MSG_SOCALLBCK) && 2141 (pr->pr_flags & PR_WANTRCVD)) { 2142 SOCKBUF_UNLOCK(&so->so_rcv); 2143 VNET_SO_ASSERT(so); 2144 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2145 SOCKBUF_LOCK(&so->so_rcv); 2146 } 2147 } 2148 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2149 if (orig_resid == uio->uio_resid && orig_resid && 2150 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 2151 SOCKBUF_UNLOCK(&so->so_rcv); 2152 goto restart; 2153 } 2154 SOCKBUF_UNLOCK(&so->so_rcv); 2155 2156 if (flagsp != NULL) 2157 *flagsp |= flags; 2158 release: 2159 sbunlock(&so->so_rcv); 2160 return (error); 2161 } 2162 2163 /* 2164 * Optimized version of soreceive() for stream (TCP) sockets. 2165 */ 2166 int 2167 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 2168 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2169 { 2170 int len = 0, error = 0, flags, oresid; 2171 struct sockbuf *sb; 2172 struct mbuf *m, *n = NULL; 2173 2174 /* We only do stream sockets. */ 2175 if (so->so_type != SOCK_STREAM) 2176 return (EINVAL); 2177 if (psa != NULL) 2178 *psa = NULL; 2179 if (flagsp != NULL) 2180 flags = *flagsp &~ MSG_EOR; 2181 else 2182 flags = 0; 2183 if (flags & (MSG_PEEK|MSG_WAITALL)) 2184 return (soreceive_generic(so, psa, uio, mp0, controlp, flagsp)); 2185 if (controlp != NULL) 2186 *controlp = NULL; 2187 if (flags & MSG_OOB) 2188 return (soreceive_rcvoob(so, uio, flags)); 2189 if (mp0 != NULL) 2190 *mp0 = NULL; 2191 2192 sb = &so->so_rcv; 2193 2194 /* Prevent other readers from entering the socket. */ 2195 error = sblock(sb, SBLOCKWAIT(flags)); 2196 if (error) 2197 goto out; 2198 SOCKBUF_LOCK(sb); 2199 2200 /* Easy one, no space to copyout anything. */ 2201 if (uio->uio_resid == 0) { 2202 error = EINVAL; 2203 goto out; 2204 } 2205 oresid = uio->uio_resid; 2206 2207 /* We will never ever get anything unless we are or were connected. */ 2208 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 2209 error = ENOTCONN; 2210 goto out; 2211 } 2212 2213 restart: 2214 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2215 2216 /* Abort if socket has reported problems. */ 2217 if (so->so_error) { 2218 if (sbavail(sb) > 0) 2219 goto deliver; 2220 if (oresid > uio->uio_resid) 2221 goto out; 2222 error = so->so_error; 2223 if (!(flags & MSG_PEEK)) 2224 so->so_error = 0; 2225 goto out; 2226 } 2227 2228 /* Door is closed. Deliver what is left, if any. */ 2229 if (sb->sb_state & SBS_CANTRCVMORE) { 2230 if (sbavail(sb) > 0) 2231 goto deliver; 2232 else 2233 goto out; 2234 } 2235 2236 /* Socket buffer is empty and we shall not block. */ 2237 if (sbavail(sb) == 0 && 2238 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2239 error = EAGAIN; 2240 goto out; 2241 } 2242 2243 /* Socket buffer got some data that we shall deliver now. */ 2244 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 2245 ((so->so_state & SS_NBIO) || 2246 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2247 sbavail(sb) >= sb->sb_lowat || 2248 sbavail(sb) >= uio->uio_resid || 2249 sbavail(sb) >= sb->sb_hiwat) ) { 2250 goto deliver; 2251 } 2252 2253 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2254 if ((flags & MSG_WAITALL) && 2255 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 2256 goto deliver; 2257 2258 /* 2259 * Wait and block until (more) data comes in. 2260 * NB: Drops the sockbuf lock during wait. 2261 */ 2262 error = sbwait(sb); 2263 if (error) 2264 goto out; 2265 goto restart; 2266 2267 deliver: 2268 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2269 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 2270 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2271 2272 /* Statistics. */ 2273 if (uio->uio_td) 2274 uio->uio_td->td_ru.ru_msgrcv++; 2275 2276 /* Fill uio until full or current end of socket buffer is reached. */ 2277 len = min(uio->uio_resid, sbavail(sb)); 2278 if (mp0 != NULL) { 2279 /* Dequeue as many mbufs as possible. */ 2280 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2281 if (*mp0 == NULL) 2282 *mp0 = sb->sb_mb; 2283 else 2284 m_cat(*mp0, sb->sb_mb); 2285 for (m = sb->sb_mb; 2286 m != NULL && m->m_len <= len; 2287 m = m->m_next) { 2288 KASSERT(!(m->m_flags & M_NOTAVAIL), 2289 ("%s: m %p not available", __func__, m)); 2290 len -= m->m_len; 2291 uio->uio_resid -= m->m_len; 2292 sbfree(sb, m); 2293 n = m; 2294 } 2295 n->m_next = NULL; 2296 sb->sb_mb = m; 2297 sb->sb_lastrecord = sb->sb_mb; 2298 if (sb->sb_mb == NULL) 2299 SB_EMPTY_FIXUP(sb); 2300 } 2301 /* Copy the remainder. */ 2302 if (len > 0) { 2303 KASSERT(sb->sb_mb != NULL, 2304 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2305 2306 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2307 if (m == NULL) 2308 len = 0; /* Don't flush data from sockbuf. */ 2309 else 2310 uio->uio_resid -= len; 2311 if (*mp0 != NULL) 2312 m_cat(*mp0, m); 2313 else 2314 *mp0 = m; 2315 if (*mp0 == NULL) { 2316 error = ENOBUFS; 2317 goto out; 2318 } 2319 } 2320 } else { 2321 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2322 SOCKBUF_UNLOCK(sb); 2323 error = m_mbuftouio(uio, sb->sb_mb, len); 2324 SOCKBUF_LOCK(sb); 2325 if (error) 2326 goto out; 2327 } 2328 SBLASTRECORDCHK(sb); 2329 SBLASTMBUFCHK(sb); 2330 2331 /* 2332 * Remove the delivered data from the socket buffer unless we 2333 * were only peeking. 2334 */ 2335 if (!(flags & MSG_PEEK)) { 2336 if (len > 0) 2337 sbdrop_locked(sb, len); 2338 2339 /* Notify protocol that we drained some data. */ 2340 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2341 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2342 !(flags & MSG_SOCALLBCK))) { 2343 SOCKBUF_UNLOCK(sb); 2344 VNET_SO_ASSERT(so); 2345 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags); 2346 SOCKBUF_LOCK(sb); 2347 } 2348 } 2349 2350 /* 2351 * For MSG_WAITALL we may have to loop again and wait for 2352 * more data to come in. 2353 */ 2354 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2355 goto restart; 2356 out: 2357 SOCKBUF_LOCK_ASSERT(sb); 2358 SBLASTRECORDCHK(sb); 2359 SBLASTMBUFCHK(sb); 2360 SOCKBUF_UNLOCK(sb); 2361 sbunlock(sb); 2362 return (error); 2363 } 2364 2365 /* 2366 * Optimized version of soreceive() for simple datagram cases from userspace. 2367 * Unlike in the stream case, we're able to drop a datagram if copyout() 2368 * fails, and because we handle datagrams atomically, we don't need to use a 2369 * sleep lock to prevent I/O interlacing. 2370 */ 2371 int 2372 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2373 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2374 { 2375 struct mbuf *m, *m2; 2376 int flags, error; 2377 ssize_t len; 2378 struct protosw *pr = so->so_proto; 2379 struct mbuf *nextrecord; 2380 2381 if (psa != NULL) 2382 *psa = NULL; 2383 if (controlp != NULL) 2384 *controlp = NULL; 2385 if (flagsp != NULL) 2386 flags = *flagsp &~ MSG_EOR; 2387 else 2388 flags = 0; 2389 2390 /* 2391 * For any complicated cases, fall back to the full 2392 * soreceive_generic(). 2393 */ 2394 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB)) 2395 return (soreceive_generic(so, psa, uio, mp0, controlp, 2396 flagsp)); 2397 2398 /* 2399 * Enforce restrictions on use. 2400 */ 2401 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2402 ("soreceive_dgram: wantrcvd")); 2403 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2404 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2405 ("soreceive_dgram: SBS_RCVATMARK")); 2406 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2407 ("soreceive_dgram: P_CONNREQUIRED")); 2408 2409 /* 2410 * Loop blocking while waiting for a datagram. 2411 */ 2412 SOCKBUF_LOCK(&so->so_rcv); 2413 while ((m = so->so_rcv.sb_mb) == NULL) { 2414 KASSERT(sbavail(&so->so_rcv) == 0, 2415 ("soreceive_dgram: sb_mb NULL but sbavail %u", 2416 sbavail(&so->so_rcv))); 2417 if (so->so_error) { 2418 error = so->so_error; 2419 so->so_error = 0; 2420 SOCKBUF_UNLOCK(&so->so_rcv); 2421 return (error); 2422 } 2423 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2424 uio->uio_resid == 0) { 2425 SOCKBUF_UNLOCK(&so->so_rcv); 2426 return (0); 2427 } 2428 if ((so->so_state & SS_NBIO) || 2429 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2430 SOCKBUF_UNLOCK(&so->so_rcv); 2431 return (EWOULDBLOCK); 2432 } 2433 SBLASTRECORDCHK(&so->so_rcv); 2434 SBLASTMBUFCHK(&so->so_rcv); 2435 error = sbwait(&so->so_rcv); 2436 if (error) { 2437 SOCKBUF_UNLOCK(&so->so_rcv); 2438 return (error); 2439 } 2440 } 2441 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2442 2443 if (uio->uio_td) 2444 uio->uio_td->td_ru.ru_msgrcv++; 2445 SBLASTRECORDCHK(&so->so_rcv); 2446 SBLASTMBUFCHK(&so->so_rcv); 2447 nextrecord = m->m_nextpkt; 2448 if (nextrecord == NULL) { 2449 KASSERT(so->so_rcv.sb_lastrecord == m, 2450 ("soreceive_dgram: lastrecord != m")); 2451 } 2452 2453 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2454 ("soreceive_dgram: m_nextpkt != nextrecord")); 2455 2456 /* 2457 * Pull 'm' and its chain off the front of the packet queue. 2458 */ 2459 so->so_rcv.sb_mb = NULL; 2460 sockbuf_pushsync(&so->so_rcv, nextrecord); 2461 2462 /* 2463 * Walk 'm's chain and free that many bytes from the socket buffer. 2464 */ 2465 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2466 sbfree(&so->so_rcv, m2); 2467 2468 /* 2469 * Do a few last checks before we let go of the lock. 2470 */ 2471 SBLASTRECORDCHK(&so->so_rcv); 2472 SBLASTMBUFCHK(&so->so_rcv); 2473 SOCKBUF_UNLOCK(&so->so_rcv); 2474 2475 if (pr->pr_flags & PR_ADDR) { 2476 KASSERT(m->m_type == MT_SONAME, 2477 ("m->m_type == %d", m->m_type)); 2478 if (psa != NULL) 2479 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2480 M_NOWAIT); 2481 m = m_free(m); 2482 } 2483 if (m == NULL) { 2484 /* XXXRW: Can this happen? */ 2485 return (0); 2486 } 2487 2488 /* 2489 * Packet to copyout() is now in 'm' and it is disconnected from the 2490 * queue. 2491 * 2492 * Process one or more MT_CONTROL mbufs present before any data mbufs 2493 * in the first mbuf chain on the socket buffer. We call into the 2494 * protocol to perform externalization (or freeing if controlp == 2495 * NULL). In some cases there can be only MT_CONTROL mbufs without 2496 * MT_DATA mbufs. 2497 */ 2498 if (m->m_type == MT_CONTROL) { 2499 struct mbuf *cm = NULL, *cmn; 2500 struct mbuf **cme = &cm; 2501 2502 do { 2503 m2 = m->m_next; 2504 m->m_next = NULL; 2505 *cme = m; 2506 cme = &(*cme)->m_next; 2507 m = m2; 2508 } while (m != NULL && m->m_type == MT_CONTROL); 2509 while (cm != NULL) { 2510 cmn = cm->m_next; 2511 cm->m_next = NULL; 2512 if (pr->pr_domain->dom_externalize != NULL) { 2513 error = (*pr->pr_domain->dom_externalize) 2514 (cm, controlp, flags); 2515 } else if (controlp != NULL) 2516 *controlp = cm; 2517 else 2518 m_freem(cm); 2519 if (controlp != NULL) { 2520 while (*controlp != NULL) 2521 controlp = &(*controlp)->m_next; 2522 } 2523 cm = cmn; 2524 } 2525 } 2526 KASSERT(m == NULL || m->m_type == MT_DATA, 2527 ("soreceive_dgram: !data")); 2528 while (m != NULL && uio->uio_resid > 0) { 2529 len = uio->uio_resid; 2530 if (len > m->m_len) 2531 len = m->m_len; 2532 error = uiomove(mtod(m, char *), (int)len, uio); 2533 if (error) { 2534 m_freem(m); 2535 return (error); 2536 } 2537 if (len == m->m_len) 2538 m = m_free(m); 2539 else { 2540 m->m_data += len; 2541 m->m_len -= len; 2542 } 2543 } 2544 if (m != NULL) { 2545 flags |= MSG_TRUNC; 2546 m_freem(m); 2547 } 2548 if (flagsp != NULL) 2549 *flagsp |= flags; 2550 return (0); 2551 } 2552 2553 int 2554 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2555 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2556 { 2557 int error; 2558 2559 CURVNET_SET(so->so_vnet); 2560 if (!SOLISTENING(so)) 2561 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, 2562 mp0, controlp, flagsp)); 2563 else 2564 error = ENOTCONN; 2565 CURVNET_RESTORE(); 2566 return (error); 2567 } 2568 2569 int 2570 soshutdown(struct socket *so, int how) 2571 { 2572 struct protosw *pr = so->so_proto; 2573 int error, soerror_enotconn; 2574 2575 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2576 return (EINVAL); 2577 2578 soerror_enotconn = 0; 2579 if ((so->so_state & 2580 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 2581 /* 2582 * POSIX mandates us to return ENOTCONN when shutdown(2) is 2583 * invoked on a datagram sockets, however historically we would 2584 * actually tear socket down. This is known to be leveraged by 2585 * some applications to unblock process waiting in recvXXX(2) 2586 * by other process that it shares that socket with. Try to meet 2587 * both backward-compatibility and POSIX requirements by forcing 2588 * ENOTCONN but still asking protocol to perform pru_shutdown(). 2589 */ 2590 if (so->so_type != SOCK_DGRAM) 2591 return (ENOTCONN); 2592 soerror_enotconn = 1; 2593 } 2594 2595 CURVNET_SET(so->so_vnet); 2596 if (pr->pr_usrreqs->pru_flush != NULL) 2597 (*pr->pr_usrreqs->pru_flush)(so, how); 2598 if (how != SHUT_WR) 2599 sorflush(so); 2600 if (how != SHUT_RD) { 2601 error = (*pr->pr_usrreqs->pru_shutdown)(so); 2602 wakeup(&so->so_timeo); 2603 CURVNET_RESTORE(); 2604 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error); 2605 } 2606 wakeup(&so->so_timeo); 2607 CURVNET_RESTORE(); 2608 2609 return (soerror_enotconn ? ENOTCONN : 0); 2610 } 2611 2612 void 2613 sorflush(struct socket *so) 2614 { 2615 struct sockbuf *sb = &so->so_rcv; 2616 struct protosw *pr = so->so_proto; 2617 struct socket aso; 2618 2619 VNET_SO_ASSERT(so); 2620 2621 /* 2622 * In order to avoid calling dom_dispose with the socket buffer mutex 2623 * held, and in order to generally avoid holding the lock for a long 2624 * time, we make a copy of the socket buffer and clear the original 2625 * (except locks, state). The new socket buffer copy won't have 2626 * initialized locks so we can only call routines that won't use or 2627 * assert those locks. 2628 * 2629 * Dislodge threads currently blocked in receive and wait to acquire 2630 * a lock against other simultaneous readers before clearing the 2631 * socket buffer. Don't let our acquire be interrupted by a signal 2632 * despite any existing socket disposition on interruptable waiting. 2633 */ 2634 socantrcvmore(so); 2635 (void) sblock(sb, SBL_WAIT | SBL_NOINTR); 2636 2637 /* 2638 * Invalidate/clear most of the sockbuf structure, but leave selinfo 2639 * and mutex data unchanged. 2640 */ 2641 SOCKBUF_LOCK(sb); 2642 bzero(&aso, sizeof(aso)); 2643 aso.so_pcb = so->so_pcb; 2644 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero, 2645 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2646 bzero(&sb->sb_startzero, 2647 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2648 SOCKBUF_UNLOCK(sb); 2649 sbunlock(sb); 2650 2651 /* 2652 * Dispose of special rights and flush the copied socket. Don't call 2653 * any unsafe routines (that rely on locks being initialized) on aso. 2654 */ 2655 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 2656 (*pr->pr_domain->dom_dispose)(&aso); 2657 sbrelease_internal(&aso.so_rcv, so); 2658 } 2659 2660 /* 2661 * Wrapper for Socket established helper hook. 2662 * Parameters: socket, context of the hook point, hook id. 2663 */ 2664 static int inline 2665 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 2666 { 2667 struct socket_hhook_data hhook_data = { 2668 .so = so, 2669 .hctx = hctx, 2670 .m = NULL, 2671 .status = 0 2672 }; 2673 2674 CURVNET_SET(so->so_vnet); 2675 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 2676 CURVNET_RESTORE(); 2677 2678 /* Ugly but needed, since hhooks return void for now */ 2679 return (hhook_data.status); 2680 } 2681 2682 /* 2683 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 2684 * additional variant to handle the case where the option value needs to be 2685 * some kind of integer, but not a specific size. In addition to their use 2686 * here, these functions are also called by the protocol-level pr_ctloutput() 2687 * routines. 2688 */ 2689 int 2690 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2691 { 2692 size_t valsize; 2693 2694 /* 2695 * If the user gives us more than we wanted, we ignore it, but if we 2696 * don't get the minimum length the caller wants, we return EINVAL. 2697 * On success, sopt->sopt_valsize is set to however much we actually 2698 * retrieved. 2699 */ 2700 if ((valsize = sopt->sopt_valsize) < minlen) 2701 return EINVAL; 2702 if (valsize > len) 2703 sopt->sopt_valsize = valsize = len; 2704 2705 if (sopt->sopt_td != NULL) 2706 return (copyin(sopt->sopt_val, buf, valsize)); 2707 2708 bcopy(sopt->sopt_val, buf, valsize); 2709 return (0); 2710 } 2711 2712 /* 2713 * Kernel version of setsockopt(2). 2714 * 2715 * XXX: optlen is size_t, not socklen_t 2716 */ 2717 int 2718 so_setsockopt(struct socket *so, int level, int optname, void *optval, 2719 size_t optlen) 2720 { 2721 struct sockopt sopt; 2722 2723 sopt.sopt_level = level; 2724 sopt.sopt_name = optname; 2725 sopt.sopt_dir = SOPT_SET; 2726 sopt.sopt_val = optval; 2727 sopt.sopt_valsize = optlen; 2728 sopt.sopt_td = NULL; 2729 return (sosetopt(so, &sopt)); 2730 } 2731 2732 int 2733 sosetopt(struct socket *so, struct sockopt *sopt) 2734 { 2735 int error, optval; 2736 struct linger l; 2737 struct timeval tv; 2738 sbintime_t val; 2739 uint32_t val32; 2740 #ifdef MAC 2741 struct mac extmac; 2742 #endif 2743 2744 CURVNET_SET(so->so_vnet); 2745 error = 0; 2746 if (sopt->sopt_level != SOL_SOCKET) { 2747 if (so->so_proto->pr_ctloutput != NULL) { 2748 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2749 CURVNET_RESTORE(); 2750 return (error); 2751 } 2752 error = ENOPROTOOPT; 2753 } else { 2754 switch (sopt->sopt_name) { 2755 case SO_ACCEPTFILTER: 2756 error = accept_filt_setopt(so, sopt); 2757 if (error) 2758 goto bad; 2759 break; 2760 2761 case SO_LINGER: 2762 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 2763 if (error) 2764 goto bad; 2765 2766 SOCK_LOCK(so); 2767 so->so_linger = l.l_linger; 2768 if (l.l_onoff) 2769 so->so_options |= SO_LINGER; 2770 else 2771 so->so_options &= ~SO_LINGER; 2772 SOCK_UNLOCK(so); 2773 break; 2774 2775 case SO_DEBUG: 2776 case SO_KEEPALIVE: 2777 case SO_DONTROUTE: 2778 case SO_USELOOPBACK: 2779 case SO_BROADCAST: 2780 case SO_REUSEADDR: 2781 case SO_REUSEPORT: 2782 case SO_REUSEPORT_LB: 2783 case SO_OOBINLINE: 2784 case SO_TIMESTAMP: 2785 case SO_BINTIME: 2786 case SO_NOSIGPIPE: 2787 case SO_NO_DDP: 2788 case SO_NO_OFFLOAD: 2789 error = sooptcopyin(sopt, &optval, sizeof optval, 2790 sizeof optval); 2791 if (error) 2792 goto bad; 2793 SOCK_LOCK(so); 2794 if (optval) 2795 so->so_options |= sopt->sopt_name; 2796 else 2797 so->so_options &= ~sopt->sopt_name; 2798 SOCK_UNLOCK(so); 2799 break; 2800 2801 case SO_SETFIB: 2802 error = sooptcopyin(sopt, &optval, sizeof optval, 2803 sizeof optval); 2804 if (error) 2805 goto bad; 2806 2807 if (optval < 0 || optval >= rt_numfibs) { 2808 error = EINVAL; 2809 goto bad; 2810 } 2811 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 2812 (so->so_proto->pr_domain->dom_family == PF_INET6) || 2813 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 2814 so->so_fibnum = optval; 2815 else 2816 so->so_fibnum = 0; 2817 break; 2818 2819 case SO_USER_COOKIE: 2820 error = sooptcopyin(sopt, &val32, sizeof val32, 2821 sizeof val32); 2822 if (error) 2823 goto bad; 2824 so->so_user_cookie = val32; 2825 break; 2826 2827 case SO_SNDBUF: 2828 case SO_RCVBUF: 2829 case SO_SNDLOWAT: 2830 case SO_RCVLOWAT: 2831 error = sooptcopyin(sopt, &optval, sizeof optval, 2832 sizeof optval); 2833 if (error) 2834 goto bad; 2835 2836 /* 2837 * Values < 1 make no sense for any of these options, 2838 * so disallow them. 2839 */ 2840 if (optval < 1) { 2841 error = EINVAL; 2842 goto bad; 2843 } 2844 2845 error = sbsetopt(so, sopt->sopt_name, optval); 2846 break; 2847 2848 case SO_SNDTIMEO: 2849 case SO_RCVTIMEO: 2850 #ifdef COMPAT_FREEBSD32 2851 if (SV_CURPROC_FLAG(SV_ILP32)) { 2852 struct timeval32 tv32; 2853 2854 error = sooptcopyin(sopt, &tv32, sizeof tv32, 2855 sizeof tv32); 2856 CP(tv32, tv, tv_sec); 2857 CP(tv32, tv, tv_usec); 2858 } else 2859 #endif 2860 error = sooptcopyin(sopt, &tv, sizeof tv, 2861 sizeof tv); 2862 if (error) 2863 goto bad; 2864 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 2865 tv.tv_usec >= 1000000) { 2866 error = EDOM; 2867 goto bad; 2868 } 2869 if (tv.tv_sec > INT32_MAX) 2870 val = SBT_MAX; 2871 else 2872 val = tvtosbt(tv); 2873 switch (sopt->sopt_name) { 2874 case SO_SNDTIMEO: 2875 so->so_snd.sb_timeo = val; 2876 break; 2877 case SO_RCVTIMEO: 2878 so->so_rcv.sb_timeo = val; 2879 break; 2880 } 2881 break; 2882 2883 case SO_LABEL: 2884 #ifdef MAC 2885 error = sooptcopyin(sopt, &extmac, sizeof extmac, 2886 sizeof extmac); 2887 if (error) 2888 goto bad; 2889 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 2890 so, &extmac); 2891 #else 2892 error = EOPNOTSUPP; 2893 #endif 2894 break; 2895 2896 case SO_TS_CLOCK: 2897 error = sooptcopyin(sopt, &optval, sizeof optval, 2898 sizeof optval); 2899 if (error) 2900 goto bad; 2901 if (optval < 0 || optval > SO_TS_CLOCK_MAX) { 2902 error = EINVAL; 2903 goto bad; 2904 } 2905 so->so_ts_clock = optval; 2906 break; 2907 2908 case SO_MAX_PACING_RATE: 2909 error = sooptcopyin(sopt, &val32, sizeof(val32), 2910 sizeof(val32)); 2911 if (error) 2912 goto bad; 2913 so->so_max_pacing_rate = val32; 2914 break; 2915 2916 default: 2917 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 2918 error = hhook_run_socket(so, sopt, 2919 HHOOK_SOCKET_OPT); 2920 else 2921 error = ENOPROTOOPT; 2922 break; 2923 } 2924 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 2925 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 2926 } 2927 bad: 2928 CURVNET_RESTORE(); 2929 return (error); 2930 } 2931 2932 /* 2933 * Helper routine for getsockopt. 2934 */ 2935 int 2936 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 2937 { 2938 int error; 2939 size_t valsize; 2940 2941 error = 0; 2942 2943 /* 2944 * Documented get behavior is that we always return a value, possibly 2945 * truncated to fit in the user's buffer. Traditional behavior is 2946 * that we always tell the user precisely how much we copied, rather 2947 * than something useful like the total amount we had available for 2948 * her. Note that this interface is not idempotent; the entire 2949 * answer must be generated ahead of time. 2950 */ 2951 valsize = min(len, sopt->sopt_valsize); 2952 sopt->sopt_valsize = valsize; 2953 if (sopt->sopt_val != NULL) { 2954 if (sopt->sopt_td != NULL) 2955 error = copyout(buf, sopt->sopt_val, valsize); 2956 else 2957 bcopy(buf, sopt->sopt_val, valsize); 2958 } 2959 return (error); 2960 } 2961 2962 int 2963 sogetopt(struct socket *so, struct sockopt *sopt) 2964 { 2965 int error, optval; 2966 struct linger l; 2967 struct timeval tv; 2968 #ifdef MAC 2969 struct mac extmac; 2970 #endif 2971 2972 CURVNET_SET(so->so_vnet); 2973 error = 0; 2974 if (sopt->sopt_level != SOL_SOCKET) { 2975 if (so->so_proto->pr_ctloutput != NULL) 2976 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2977 else 2978 error = ENOPROTOOPT; 2979 CURVNET_RESTORE(); 2980 return (error); 2981 } else { 2982 switch (sopt->sopt_name) { 2983 case SO_ACCEPTFILTER: 2984 error = accept_filt_getopt(so, sopt); 2985 break; 2986 2987 case SO_LINGER: 2988 SOCK_LOCK(so); 2989 l.l_onoff = so->so_options & SO_LINGER; 2990 l.l_linger = so->so_linger; 2991 SOCK_UNLOCK(so); 2992 error = sooptcopyout(sopt, &l, sizeof l); 2993 break; 2994 2995 case SO_USELOOPBACK: 2996 case SO_DONTROUTE: 2997 case SO_DEBUG: 2998 case SO_KEEPALIVE: 2999 case SO_REUSEADDR: 3000 case SO_REUSEPORT: 3001 case SO_REUSEPORT_LB: 3002 case SO_BROADCAST: 3003 case SO_OOBINLINE: 3004 case SO_ACCEPTCONN: 3005 case SO_TIMESTAMP: 3006 case SO_BINTIME: 3007 case SO_NOSIGPIPE: 3008 optval = so->so_options & sopt->sopt_name; 3009 integer: 3010 error = sooptcopyout(sopt, &optval, sizeof optval); 3011 break; 3012 3013 case SO_TYPE: 3014 optval = so->so_type; 3015 goto integer; 3016 3017 case SO_PROTOCOL: 3018 optval = so->so_proto->pr_protocol; 3019 goto integer; 3020 3021 case SO_ERROR: 3022 SOCK_LOCK(so); 3023 optval = so->so_error; 3024 so->so_error = 0; 3025 SOCK_UNLOCK(so); 3026 goto integer; 3027 3028 case SO_SNDBUF: 3029 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : 3030 so->so_snd.sb_hiwat; 3031 goto integer; 3032 3033 case SO_RCVBUF: 3034 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : 3035 so->so_rcv.sb_hiwat; 3036 goto integer; 3037 3038 case SO_SNDLOWAT: 3039 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : 3040 so->so_snd.sb_lowat; 3041 goto integer; 3042 3043 case SO_RCVLOWAT: 3044 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : 3045 so->so_rcv.sb_lowat; 3046 goto integer; 3047 3048 case SO_SNDTIMEO: 3049 case SO_RCVTIMEO: 3050 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 3051 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 3052 #ifdef COMPAT_FREEBSD32 3053 if (SV_CURPROC_FLAG(SV_ILP32)) { 3054 struct timeval32 tv32; 3055 3056 CP(tv, tv32, tv_sec); 3057 CP(tv, tv32, tv_usec); 3058 error = sooptcopyout(sopt, &tv32, sizeof tv32); 3059 } else 3060 #endif 3061 error = sooptcopyout(sopt, &tv, sizeof tv); 3062 break; 3063 3064 case SO_LABEL: 3065 #ifdef MAC 3066 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3067 sizeof(extmac)); 3068 if (error) 3069 goto bad; 3070 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 3071 so, &extmac); 3072 if (error) 3073 goto bad; 3074 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3075 #else 3076 error = EOPNOTSUPP; 3077 #endif 3078 break; 3079 3080 case SO_PEERLABEL: 3081 #ifdef MAC 3082 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3083 sizeof(extmac)); 3084 if (error) 3085 goto bad; 3086 error = mac_getsockopt_peerlabel( 3087 sopt->sopt_td->td_ucred, so, &extmac); 3088 if (error) 3089 goto bad; 3090 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3091 #else 3092 error = EOPNOTSUPP; 3093 #endif 3094 break; 3095 3096 case SO_LISTENQLIMIT: 3097 optval = SOLISTENING(so) ? so->sol_qlimit : 0; 3098 goto integer; 3099 3100 case SO_LISTENQLEN: 3101 optval = SOLISTENING(so) ? so->sol_qlen : 0; 3102 goto integer; 3103 3104 case SO_LISTENINCQLEN: 3105 optval = SOLISTENING(so) ? so->sol_incqlen : 0; 3106 goto integer; 3107 3108 case SO_TS_CLOCK: 3109 optval = so->so_ts_clock; 3110 goto integer; 3111 3112 case SO_MAX_PACING_RATE: 3113 optval = so->so_max_pacing_rate; 3114 goto integer; 3115 3116 default: 3117 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3118 error = hhook_run_socket(so, sopt, 3119 HHOOK_SOCKET_OPT); 3120 else 3121 error = ENOPROTOOPT; 3122 break; 3123 } 3124 } 3125 #ifdef MAC 3126 bad: 3127 #endif 3128 CURVNET_RESTORE(); 3129 return (error); 3130 } 3131 3132 int 3133 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 3134 { 3135 struct mbuf *m, *m_prev; 3136 int sopt_size = sopt->sopt_valsize; 3137 3138 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3139 if (m == NULL) 3140 return ENOBUFS; 3141 if (sopt_size > MLEN) { 3142 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 3143 if ((m->m_flags & M_EXT) == 0) { 3144 m_free(m); 3145 return ENOBUFS; 3146 } 3147 m->m_len = min(MCLBYTES, sopt_size); 3148 } else { 3149 m->m_len = min(MLEN, sopt_size); 3150 } 3151 sopt_size -= m->m_len; 3152 *mp = m; 3153 m_prev = m; 3154 3155 while (sopt_size) { 3156 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3157 if (m == NULL) { 3158 m_freem(*mp); 3159 return ENOBUFS; 3160 } 3161 if (sopt_size > MLEN) { 3162 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 3163 M_NOWAIT); 3164 if ((m->m_flags & M_EXT) == 0) { 3165 m_freem(m); 3166 m_freem(*mp); 3167 return ENOBUFS; 3168 } 3169 m->m_len = min(MCLBYTES, sopt_size); 3170 } else { 3171 m->m_len = min(MLEN, sopt_size); 3172 } 3173 sopt_size -= m->m_len; 3174 m_prev->m_next = m; 3175 m_prev = m; 3176 } 3177 return (0); 3178 } 3179 3180 int 3181 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 3182 { 3183 struct mbuf *m0 = m; 3184 3185 if (sopt->sopt_val == NULL) 3186 return (0); 3187 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3188 if (sopt->sopt_td != NULL) { 3189 int error; 3190 3191 error = copyin(sopt->sopt_val, mtod(m, char *), 3192 m->m_len); 3193 if (error != 0) { 3194 m_freem(m0); 3195 return(error); 3196 } 3197 } else 3198 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 3199 sopt->sopt_valsize -= m->m_len; 3200 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3201 m = m->m_next; 3202 } 3203 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 3204 panic("ip6_sooptmcopyin"); 3205 return (0); 3206 } 3207 3208 int 3209 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 3210 { 3211 struct mbuf *m0 = m; 3212 size_t valsize = 0; 3213 3214 if (sopt->sopt_val == NULL) 3215 return (0); 3216 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3217 if (sopt->sopt_td != NULL) { 3218 int error; 3219 3220 error = copyout(mtod(m, char *), sopt->sopt_val, 3221 m->m_len); 3222 if (error != 0) { 3223 m_freem(m0); 3224 return(error); 3225 } 3226 } else 3227 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 3228 sopt->sopt_valsize -= m->m_len; 3229 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3230 valsize += m->m_len; 3231 m = m->m_next; 3232 } 3233 if (m != NULL) { 3234 /* enough soopt buffer should be given from user-land */ 3235 m_freem(m0); 3236 return(EINVAL); 3237 } 3238 sopt->sopt_valsize = valsize; 3239 return (0); 3240 } 3241 3242 /* 3243 * sohasoutofband(): protocol notifies socket layer of the arrival of new 3244 * out-of-band data, which will then notify socket consumers. 3245 */ 3246 void 3247 sohasoutofband(struct socket *so) 3248 { 3249 3250 if (so->so_sigio != NULL) 3251 pgsigio(&so->so_sigio, SIGURG, 0); 3252 selwakeuppri(&so->so_rdsel, PSOCK); 3253 } 3254 3255 int 3256 sopoll(struct socket *so, int events, struct ucred *active_cred, 3257 struct thread *td) 3258 { 3259 3260 /* 3261 * We do not need to set or assert curvnet as long as everyone uses 3262 * sopoll_generic(). 3263 */ 3264 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, 3265 td)); 3266 } 3267 3268 int 3269 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3270 struct thread *td) 3271 { 3272 int revents; 3273 3274 SOCK_LOCK(so); 3275 if (SOLISTENING(so)) { 3276 if (!(events & (POLLIN | POLLRDNORM))) 3277 revents = 0; 3278 else if (!TAILQ_EMPTY(&so->sol_comp)) 3279 revents = events & (POLLIN | POLLRDNORM); 3280 else { 3281 selrecord(td, &so->so_rdsel); 3282 revents = 0; 3283 } 3284 } else { 3285 revents = 0; 3286 SOCKBUF_LOCK(&so->so_snd); 3287 SOCKBUF_LOCK(&so->so_rcv); 3288 if (events & (POLLIN | POLLRDNORM)) 3289 if (soreadabledata(so)) 3290 revents |= events & (POLLIN | POLLRDNORM); 3291 if (events & (POLLOUT | POLLWRNORM)) 3292 if (sowriteable(so)) 3293 revents |= events & (POLLOUT | POLLWRNORM); 3294 if (events & (POLLPRI | POLLRDBAND)) 3295 if (so->so_oobmark || 3296 (so->so_rcv.sb_state & SBS_RCVATMARK)) 3297 revents |= events & (POLLPRI | POLLRDBAND); 3298 if ((events & POLLINIGNEOF) == 0) { 3299 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3300 revents |= events & (POLLIN | POLLRDNORM); 3301 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3302 revents |= POLLHUP; 3303 } 3304 } 3305 if (revents == 0) { 3306 if (events & 3307 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 3308 selrecord(td, &so->so_rdsel); 3309 so->so_rcv.sb_flags |= SB_SEL; 3310 } 3311 if (events & (POLLOUT | POLLWRNORM)) { 3312 selrecord(td, &so->so_wrsel); 3313 so->so_snd.sb_flags |= SB_SEL; 3314 } 3315 } 3316 SOCKBUF_UNLOCK(&so->so_rcv); 3317 SOCKBUF_UNLOCK(&so->so_snd); 3318 } 3319 SOCK_UNLOCK(so); 3320 return (revents); 3321 } 3322 3323 int 3324 soo_kqfilter(struct file *fp, struct knote *kn) 3325 { 3326 struct socket *so = kn->kn_fp->f_data; 3327 struct sockbuf *sb; 3328 struct knlist *knl; 3329 3330 switch (kn->kn_filter) { 3331 case EVFILT_READ: 3332 kn->kn_fop = &soread_filtops; 3333 knl = &so->so_rdsel.si_note; 3334 sb = &so->so_rcv; 3335 break; 3336 case EVFILT_WRITE: 3337 kn->kn_fop = &sowrite_filtops; 3338 knl = &so->so_wrsel.si_note; 3339 sb = &so->so_snd; 3340 break; 3341 case EVFILT_EMPTY: 3342 kn->kn_fop = &soempty_filtops; 3343 knl = &so->so_wrsel.si_note; 3344 sb = &so->so_snd; 3345 break; 3346 default: 3347 return (EINVAL); 3348 } 3349 3350 SOCK_LOCK(so); 3351 if (SOLISTENING(so)) { 3352 knlist_add(knl, kn, 1); 3353 } else { 3354 SOCKBUF_LOCK(sb); 3355 knlist_add(knl, kn, 1); 3356 sb->sb_flags |= SB_KNOTE; 3357 SOCKBUF_UNLOCK(sb); 3358 } 3359 SOCK_UNLOCK(so); 3360 return (0); 3361 } 3362 3363 /* 3364 * Some routines that return EOPNOTSUPP for entry points that are not 3365 * supported by a protocol. Fill in as needed. 3366 */ 3367 int 3368 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 3369 { 3370 3371 return EOPNOTSUPP; 3372 } 3373 3374 int 3375 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job) 3376 { 3377 3378 return EOPNOTSUPP; 3379 } 3380 3381 int 3382 pru_attach_notsupp(struct socket *so, int proto, struct thread *td) 3383 { 3384 3385 return EOPNOTSUPP; 3386 } 3387 3388 int 3389 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3390 { 3391 3392 return EOPNOTSUPP; 3393 } 3394 3395 int 3396 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3397 struct thread *td) 3398 { 3399 3400 return EOPNOTSUPP; 3401 } 3402 3403 int 3404 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3405 { 3406 3407 return EOPNOTSUPP; 3408 } 3409 3410 int 3411 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3412 struct thread *td) 3413 { 3414 3415 return EOPNOTSUPP; 3416 } 3417 3418 int 3419 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 3420 { 3421 3422 return EOPNOTSUPP; 3423 } 3424 3425 int 3426 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 3427 struct ifnet *ifp, struct thread *td) 3428 { 3429 3430 return EOPNOTSUPP; 3431 } 3432 3433 int 3434 pru_disconnect_notsupp(struct socket *so) 3435 { 3436 3437 return EOPNOTSUPP; 3438 } 3439 3440 int 3441 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) 3442 { 3443 3444 return EOPNOTSUPP; 3445 } 3446 3447 int 3448 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) 3449 { 3450 3451 return EOPNOTSUPP; 3452 } 3453 3454 int 3455 pru_rcvd_notsupp(struct socket *so, int flags) 3456 { 3457 3458 return EOPNOTSUPP; 3459 } 3460 3461 int 3462 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 3463 { 3464 3465 return EOPNOTSUPP; 3466 } 3467 3468 int 3469 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, 3470 struct sockaddr *addr, struct mbuf *control, struct thread *td) 3471 { 3472 3473 return EOPNOTSUPP; 3474 } 3475 3476 int 3477 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count) 3478 { 3479 3480 return (EOPNOTSUPP); 3481 } 3482 3483 /* 3484 * This isn't really a ``null'' operation, but it's the default one and 3485 * doesn't do anything destructive. 3486 */ 3487 int 3488 pru_sense_null(struct socket *so, struct stat *sb) 3489 { 3490 3491 sb->st_blksize = so->so_snd.sb_hiwat; 3492 return 0; 3493 } 3494 3495 int 3496 pru_shutdown_notsupp(struct socket *so) 3497 { 3498 3499 return EOPNOTSUPP; 3500 } 3501 3502 int 3503 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) 3504 { 3505 3506 return EOPNOTSUPP; 3507 } 3508 3509 int 3510 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 3511 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 3512 { 3513 3514 return EOPNOTSUPP; 3515 } 3516 3517 int 3518 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 3519 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 3520 { 3521 3522 return EOPNOTSUPP; 3523 } 3524 3525 int 3526 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, 3527 struct thread *td) 3528 { 3529 3530 return EOPNOTSUPP; 3531 } 3532 3533 static void 3534 filt_sordetach(struct knote *kn) 3535 { 3536 struct socket *so = kn->kn_fp->f_data; 3537 3538 so_rdknl_lock(so); 3539 knlist_remove(&so->so_rdsel.si_note, kn, 1); 3540 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note)) 3541 so->so_rcv.sb_flags &= ~SB_KNOTE; 3542 so_rdknl_unlock(so); 3543 } 3544 3545 /*ARGSUSED*/ 3546 static int 3547 filt_soread(struct knote *kn, long hint) 3548 { 3549 struct socket *so; 3550 3551 so = kn->kn_fp->f_data; 3552 3553 if (SOLISTENING(so)) { 3554 SOCK_LOCK_ASSERT(so); 3555 kn->kn_data = so->sol_qlen; 3556 return (!TAILQ_EMPTY(&so->sol_comp)); 3557 } 3558 3559 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3560 3561 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 3562 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3563 kn->kn_flags |= EV_EOF; 3564 kn->kn_fflags = so->so_error; 3565 return (1); 3566 } else if (so->so_error) /* temporary udp error */ 3567 return (1); 3568 3569 if (kn->kn_sfflags & NOTE_LOWAT) { 3570 if (kn->kn_data >= kn->kn_sdata) 3571 return (1); 3572 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) 3573 return (1); 3574 3575 /* This hook returning non-zero indicates an event, not error */ 3576 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); 3577 } 3578 3579 static void 3580 filt_sowdetach(struct knote *kn) 3581 { 3582 struct socket *so = kn->kn_fp->f_data; 3583 3584 so_wrknl_lock(so); 3585 knlist_remove(&so->so_wrsel.si_note, kn, 1); 3586 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note)) 3587 so->so_snd.sb_flags &= ~SB_KNOTE; 3588 so_wrknl_unlock(so); 3589 } 3590 3591 /*ARGSUSED*/ 3592 static int 3593 filt_sowrite(struct knote *kn, long hint) 3594 { 3595 struct socket *so; 3596 3597 so = kn->kn_fp->f_data; 3598 3599 if (SOLISTENING(so)) 3600 return (0); 3601 3602 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3603 kn->kn_data = sbspace(&so->so_snd); 3604 3605 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); 3606 3607 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3608 kn->kn_flags |= EV_EOF; 3609 kn->kn_fflags = so->so_error; 3610 return (1); 3611 } else if (so->so_error) /* temporary udp error */ 3612 return (1); 3613 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3614 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3615 return (0); 3616 else if (kn->kn_sfflags & NOTE_LOWAT) 3617 return (kn->kn_data >= kn->kn_sdata); 3618 else 3619 return (kn->kn_data >= so->so_snd.sb_lowat); 3620 } 3621 3622 static int 3623 filt_soempty(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 (1); 3631 3632 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3633 kn->kn_data = sbused(&so->so_snd); 3634 3635 if (kn->kn_data == 0) 3636 return (1); 3637 else 3638 return (0); 3639 } 3640 3641 int 3642 socheckuid(struct socket *so, uid_t uid) 3643 { 3644 3645 if (so == NULL) 3646 return (EPERM); 3647 if (so->so_cred->cr_uid != uid) 3648 return (EPERM); 3649 return (0); 3650 } 3651 3652 /* 3653 * These functions are used by protocols to notify the socket layer (and its 3654 * consumers) of state changes in the sockets driven by protocol-side events. 3655 */ 3656 3657 /* 3658 * Procedures to manipulate state flags of socket and do appropriate wakeups. 3659 * 3660 * Normal sequence from the active (originating) side is that 3661 * soisconnecting() is called during processing of connect() call, resulting 3662 * in an eventual call to soisconnected() if/when the connection is 3663 * established. When the connection is torn down soisdisconnecting() is 3664 * called during processing of disconnect() call, and soisdisconnected() is 3665 * called when the connection to the peer is totally severed. The semantics 3666 * of these routines are such that connectionless protocols can call 3667 * soisconnected() and soisdisconnected() only, bypassing the in-progress 3668 * calls when setting up a ``connection'' takes no time. 3669 * 3670 * From the passive side, a socket is created with two queues of sockets: 3671 * so_incomp for connections in progress and so_comp for connections already 3672 * made and awaiting user acceptance. As a protocol is preparing incoming 3673 * connections, it creates a socket structure queued on so_incomp by calling 3674 * sonewconn(). When the connection is established, soisconnected() is 3675 * called, and transfers the socket structure to so_comp, making it available 3676 * to accept(). 3677 * 3678 * If a socket is closed with sockets on either so_incomp or so_comp, these 3679 * sockets are dropped. 3680 * 3681 * If higher-level protocols are implemented in the kernel, the wakeups done 3682 * here will sometimes cause software-interrupt process scheduling. 3683 */ 3684 void 3685 soisconnecting(struct socket *so) 3686 { 3687 3688 SOCK_LOCK(so); 3689 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 3690 so->so_state |= SS_ISCONNECTING; 3691 SOCK_UNLOCK(so); 3692 } 3693 3694 void 3695 soisconnected(struct socket *so) 3696 { 3697 3698 SOCK_LOCK(so); 3699 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 3700 so->so_state |= SS_ISCONNECTED; 3701 3702 if (so->so_qstate == SQ_INCOMP) { 3703 struct socket *head = so->so_listen; 3704 int ret; 3705 3706 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so)); 3707 /* 3708 * Promoting a socket from incomplete queue to complete, we 3709 * need to go through reverse order of locking. We first do 3710 * trylock, and if that doesn't succeed, we go the hard way 3711 * leaving a reference and rechecking consistency after proper 3712 * locking. 3713 */ 3714 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) { 3715 soref(head); 3716 SOCK_UNLOCK(so); 3717 SOLISTEN_LOCK(head); 3718 SOCK_LOCK(so); 3719 if (__predict_false(head != so->so_listen)) { 3720 /* 3721 * The socket went off the listen queue, 3722 * should be lost race to close(2) of sol. 3723 * The socket is about to soabort(). 3724 */ 3725 SOCK_UNLOCK(so); 3726 sorele(head); 3727 return; 3728 } 3729 /* Not the last one, as so holds a ref. */ 3730 refcount_release(&head->so_count); 3731 } 3732 again: 3733 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 3734 TAILQ_REMOVE(&head->sol_incomp, so, so_list); 3735 head->sol_incqlen--; 3736 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 3737 head->sol_qlen++; 3738 so->so_qstate = SQ_COMP; 3739 SOCK_UNLOCK(so); 3740 solisten_wakeup(head); /* unlocks */ 3741 } else { 3742 SOCKBUF_LOCK(&so->so_rcv); 3743 soupcall_set(so, SO_RCV, 3744 head->sol_accept_filter->accf_callback, 3745 head->sol_accept_filter_arg); 3746 so->so_options &= ~SO_ACCEPTFILTER; 3747 ret = head->sol_accept_filter->accf_callback(so, 3748 head->sol_accept_filter_arg, M_NOWAIT); 3749 if (ret == SU_ISCONNECTED) { 3750 soupcall_clear(so, SO_RCV); 3751 SOCKBUF_UNLOCK(&so->so_rcv); 3752 goto again; 3753 } 3754 SOCKBUF_UNLOCK(&so->so_rcv); 3755 SOCK_UNLOCK(so); 3756 SOLISTEN_UNLOCK(head); 3757 } 3758 return; 3759 } 3760 SOCK_UNLOCK(so); 3761 wakeup(&so->so_timeo); 3762 sorwakeup(so); 3763 sowwakeup(so); 3764 } 3765 3766 void 3767 soisdisconnecting(struct socket *so) 3768 { 3769 3770 SOCK_LOCK(so); 3771 so->so_state &= ~SS_ISCONNECTING; 3772 so->so_state |= SS_ISDISCONNECTING; 3773 3774 if (!SOLISTENING(so)) { 3775 SOCKBUF_LOCK(&so->so_rcv); 3776 socantrcvmore_locked(so); 3777 SOCKBUF_LOCK(&so->so_snd); 3778 socantsendmore_locked(so); 3779 } 3780 SOCK_UNLOCK(so); 3781 wakeup(&so->so_timeo); 3782 } 3783 3784 void 3785 soisdisconnected(struct socket *so) 3786 { 3787 3788 SOCK_LOCK(so); 3789 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 3790 so->so_state |= SS_ISDISCONNECTED; 3791 3792 if (!SOLISTENING(so)) { 3793 SOCK_UNLOCK(so); 3794 SOCKBUF_LOCK(&so->so_rcv); 3795 socantrcvmore_locked(so); 3796 SOCKBUF_LOCK(&so->so_snd); 3797 sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); 3798 socantsendmore_locked(so); 3799 } else 3800 SOCK_UNLOCK(so); 3801 wakeup(&so->so_timeo); 3802 } 3803 3804 /* 3805 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 3806 */ 3807 struct sockaddr * 3808 sodupsockaddr(const struct sockaddr *sa, int mflags) 3809 { 3810 struct sockaddr *sa2; 3811 3812 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 3813 if (sa2) 3814 bcopy(sa, sa2, sa->sa_len); 3815 return sa2; 3816 } 3817 3818 /* 3819 * Register per-socket destructor. 3820 */ 3821 void 3822 sodtor_set(struct socket *so, so_dtor_t *func) 3823 { 3824 3825 SOCK_LOCK_ASSERT(so); 3826 so->so_dtor = func; 3827 } 3828 3829 /* 3830 * Register per-socket buffer upcalls. 3831 */ 3832 void 3833 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg) 3834 { 3835 struct sockbuf *sb; 3836 3837 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 3838 3839 switch (which) { 3840 case SO_RCV: 3841 sb = &so->so_rcv; 3842 break; 3843 case SO_SND: 3844 sb = &so->so_snd; 3845 break; 3846 default: 3847 panic("soupcall_set: bad which"); 3848 } 3849 SOCKBUF_LOCK_ASSERT(sb); 3850 sb->sb_upcall = func; 3851 sb->sb_upcallarg = arg; 3852 sb->sb_flags |= SB_UPCALL; 3853 } 3854 3855 void 3856 soupcall_clear(struct socket *so, int which) 3857 { 3858 struct sockbuf *sb; 3859 3860 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 3861 3862 switch (which) { 3863 case SO_RCV: 3864 sb = &so->so_rcv; 3865 break; 3866 case SO_SND: 3867 sb = &so->so_snd; 3868 break; 3869 default: 3870 panic("soupcall_clear: bad which"); 3871 } 3872 SOCKBUF_LOCK_ASSERT(sb); 3873 KASSERT(sb->sb_upcall != NULL, 3874 ("%s: so %p no upcall to clear", __func__, so)); 3875 sb->sb_upcall = NULL; 3876 sb->sb_upcallarg = NULL; 3877 sb->sb_flags &= ~SB_UPCALL; 3878 } 3879 3880 void 3881 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg) 3882 { 3883 3884 SOLISTEN_LOCK_ASSERT(so); 3885 so->sol_upcall = func; 3886 so->sol_upcallarg = arg; 3887 } 3888 3889 static void 3890 so_rdknl_lock(void *arg) 3891 { 3892 struct socket *so = arg; 3893 3894 if (SOLISTENING(so)) 3895 SOCK_LOCK(so); 3896 else 3897 SOCKBUF_LOCK(&so->so_rcv); 3898 } 3899 3900 static void 3901 so_rdknl_unlock(void *arg) 3902 { 3903 struct socket *so = arg; 3904 3905 if (SOLISTENING(so)) 3906 SOCK_UNLOCK(so); 3907 else 3908 SOCKBUF_UNLOCK(&so->so_rcv); 3909 } 3910 3911 static void 3912 so_rdknl_assert_locked(void *arg) 3913 { 3914 struct socket *so = arg; 3915 3916 if (SOLISTENING(so)) 3917 SOCK_LOCK_ASSERT(so); 3918 else 3919 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3920 } 3921 3922 static void 3923 so_rdknl_assert_unlocked(void *arg) 3924 { 3925 struct socket *so = arg; 3926 3927 if (SOLISTENING(so)) 3928 SOCK_UNLOCK_ASSERT(so); 3929 else 3930 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); 3931 } 3932 3933 static void 3934 so_wrknl_lock(void *arg) 3935 { 3936 struct socket *so = arg; 3937 3938 if (SOLISTENING(so)) 3939 SOCK_LOCK(so); 3940 else 3941 SOCKBUF_LOCK(&so->so_snd); 3942 } 3943 3944 static void 3945 so_wrknl_unlock(void *arg) 3946 { 3947 struct socket *so = arg; 3948 3949 if (SOLISTENING(so)) 3950 SOCK_UNLOCK(so); 3951 else 3952 SOCKBUF_UNLOCK(&so->so_snd); 3953 } 3954 3955 static void 3956 so_wrknl_assert_locked(void *arg) 3957 { 3958 struct socket *so = arg; 3959 3960 if (SOLISTENING(so)) 3961 SOCK_LOCK_ASSERT(so); 3962 else 3963 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3964 } 3965 3966 static void 3967 so_wrknl_assert_unlocked(void *arg) 3968 { 3969 struct socket *so = arg; 3970 3971 if (SOLISTENING(so)) 3972 SOCK_UNLOCK_ASSERT(so); 3973 else 3974 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 3975 } 3976 3977 /* 3978 * Create an external-format (``xsocket'') structure using the information in 3979 * the kernel-format socket structure pointed to by so. This is done to 3980 * reduce the spew of irrelevant information over this interface, to isolate 3981 * user code from changes in the kernel structure, and potentially to provide 3982 * information-hiding if we decide that some of this information should be 3983 * hidden from users. 3984 */ 3985 void 3986 sotoxsocket(struct socket *so, struct xsocket *xso) 3987 { 3988 3989 xso->xso_len = sizeof *xso; 3990 xso->xso_so = so; 3991 xso->so_type = so->so_type; 3992 xso->so_options = so->so_options; 3993 xso->so_linger = so->so_linger; 3994 xso->so_state = so->so_state; 3995 xso->so_pcb = so->so_pcb; 3996 xso->xso_protocol = so->so_proto->pr_protocol; 3997 xso->xso_family = so->so_proto->pr_domain->dom_family; 3998 xso->so_timeo = so->so_timeo; 3999 xso->so_error = so->so_error; 4000 xso->so_uid = so->so_cred->cr_uid; 4001 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 4002 if (SOLISTENING(so)) { 4003 xso->so_qlen = so->sol_qlen; 4004 xso->so_incqlen = so->sol_incqlen; 4005 xso->so_qlimit = so->sol_qlimit; 4006 xso->so_oobmark = 0; 4007 bzero(&xso->so_snd, sizeof(xso->so_snd)); 4008 bzero(&xso->so_rcv, sizeof(xso->so_rcv)); 4009 } else { 4010 xso->so_state |= so->so_qstate; 4011 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0; 4012 xso->so_oobmark = so->so_oobmark; 4013 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 4014 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 4015 } 4016 } 4017 4018 struct sockbuf * 4019 so_sockbuf_rcv(struct socket *so) 4020 { 4021 4022 return (&so->so_rcv); 4023 } 4024 4025 struct sockbuf * 4026 so_sockbuf_snd(struct socket *so) 4027 { 4028 4029 return (&so->so_snd); 4030 } 4031 4032 int 4033 so_state_get(const struct socket *so) 4034 { 4035 4036 return (so->so_state); 4037 } 4038 4039 void 4040 so_state_set(struct socket *so, int val) 4041 { 4042 4043 so->so_state = val; 4044 } 4045 4046 int 4047 so_options_get(const struct socket *so) 4048 { 4049 4050 return (so->so_options); 4051 } 4052 4053 void 4054 so_options_set(struct socket *so, int val) 4055 { 4056 4057 so->so_options = val; 4058 } 4059 4060 int 4061 so_error_get(const struct socket *so) 4062 { 4063 4064 return (so->so_error); 4065 } 4066 4067 void 4068 so_error_set(struct socket *so, int val) 4069 { 4070 4071 so->so_error = val; 4072 } 4073 4074 int 4075 so_linger_get(const struct socket *so) 4076 { 4077 4078 return (so->so_linger); 4079 } 4080 4081 void 4082 so_linger_set(struct socket *so, int val) 4083 { 4084 4085 so->so_linger = val; 4086 } 4087 4088 struct protosw * 4089 so_protosw_get(const struct socket *so) 4090 { 4091 4092 return (so->so_proto); 4093 } 4094 4095 void 4096 so_protosw_set(struct socket *so, struct protosw *val) 4097 { 4098 4099 so->so_proto = val; 4100 } 4101 4102 void 4103 so_sorwakeup(struct socket *so) 4104 { 4105 4106 sorwakeup(so); 4107 } 4108 4109 void 4110 so_sowwakeup(struct socket *so) 4111 { 4112 4113 sowwakeup(so); 4114 } 4115 4116 void 4117 so_sorwakeup_locked(struct socket *so) 4118 { 4119 4120 sorwakeup_locked(so); 4121 } 4122 4123 void 4124 so_sowwakeup_locked(struct socket *so) 4125 { 4126 4127 sowwakeup_locked(so); 4128 } 4129 4130 void 4131 so_lock(struct socket *so) 4132 { 4133 4134 SOCK_LOCK(so); 4135 } 4136 4137 void 4138 so_unlock(struct socket *so) 4139 { 4140 4141 SOCK_UNLOCK(so); 4142 } 4143