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 (controlp != NULL) 2184 *controlp = NULL; 2185 if (flags & MSG_OOB) 2186 return (soreceive_rcvoob(so, uio, flags)); 2187 if (mp0 != NULL) 2188 *mp0 = NULL; 2189 2190 sb = &so->so_rcv; 2191 2192 /* Prevent other readers from entering the socket. */ 2193 error = sblock(sb, SBLOCKWAIT(flags)); 2194 if (error) 2195 goto out; 2196 SOCKBUF_LOCK(sb); 2197 2198 /* Easy one, no space to copyout anything. */ 2199 if (uio->uio_resid == 0) { 2200 error = EINVAL; 2201 goto out; 2202 } 2203 oresid = uio->uio_resid; 2204 2205 /* We will never ever get anything unless we are or were connected. */ 2206 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 2207 error = ENOTCONN; 2208 goto out; 2209 } 2210 2211 restart: 2212 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2213 2214 /* Abort if socket has reported problems. */ 2215 if (so->so_error) { 2216 if (sbavail(sb) > 0) 2217 goto deliver; 2218 if (oresid > uio->uio_resid) 2219 goto out; 2220 error = so->so_error; 2221 if (!(flags & MSG_PEEK)) 2222 so->so_error = 0; 2223 goto out; 2224 } 2225 2226 /* Door is closed. Deliver what is left, if any. */ 2227 if (sb->sb_state & SBS_CANTRCVMORE) { 2228 if (sbavail(sb) > 0) 2229 goto deliver; 2230 else 2231 goto out; 2232 } 2233 2234 /* Socket buffer is empty and we shall not block. */ 2235 if (sbavail(sb) == 0 && 2236 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2237 error = EAGAIN; 2238 goto out; 2239 } 2240 2241 /* Socket buffer got some data that we shall deliver now. */ 2242 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 2243 ((so->so_state & SS_NBIO) || 2244 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2245 sbavail(sb) >= sb->sb_lowat || 2246 sbavail(sb) >= uio->uio_resid || 2247 sbavail(sb) >= sb->sb_hiwat) ) { 2248 goto deliver; 2249 } 2250 2251 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2252 if ((flags & MSG_WAITALL) && 2253 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 2254 goto deliver; 2255 2256 /* 2257 * Wait and block until (more) data comes in. 2258 * NB: Drops the sockbuf lock during wait. 2259 */ 2260 error = sbwait(sb); 2261 if (error) 2262 goto out; 2263 goto restart; 2264 2265 deliver: 2266 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2267 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 2268 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2269 2270 /* Statistics. */ 2271 if (uio->uio_td) 2272 uio->uio_td->td_ru.ru_msgrcv++; 2273 2274 /* Fill uio until full or current end of socket buffer is reached. */ 2275 len = min(uio->uio_resid, sbavail(sb)); 2276 if (mp0 != NULL) { 2277 /* Dequeue as many mbufs as possible. */ 2278 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2279 if (*mp0 == NULL) 2280 *mp0 = sb->sb_mb; 2281 else 2282 m_cat(*mp0, sb->sb_mb); 2283 for (m = sb->sb_mb; 2284 m != NULL && m->m_len <= len; 2285 m = m->m_next) { 2286 KASSERT(!(m->m_flags & M_NOTAVAIL), 2287 ("%s: m %p not available", __func__, m)); 2288 len -= m->m_len; 2289 uio->uio_resid -= m->m_len; 2290 sbfree(sb, m); 2291 n = m; 2292 } 2293 n->m_next = NULL; 2294 sb->sb_mb = m; 2295 sb->sb_lastrecord = sb->sb_mb; 2296 if (sb->sb_mb == NULL) 2297 SB_EMPTY_FIXUP(sb); 2298 } 2299 /* Copy the remainder. */ 2300 if (len > 0) { 2301 KASSERT(sb->sb_mb != NULL, 2302 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2303 2304 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2305 if (m == NULL) 2306 len = 0; /* Don't flush data from sockbuf. */ 2307 else 2308 uio->uio_resid -= len; 2309 if (*mp0 != NULL) 2310 m_cat(*mp0, m); 2311 else 2312 *mp0 = m; 2313 if (*mp0 == NULL) { 2314 error = ENOBUFS; 2315 goto out; 2316 } 2317 } 2318 } else { 2319 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2320 SOCKBUF_UNLOCK(sb); 2321 error = m_mbuftouio(uio, sb->sb_mb, len); 2322 SOCKBUF_LOCK(sb); 2323 if (error) 2324 goto out; 2325 } 2326 SBLASTRECORDCHK(sb); 2327 SBLASTMBUFCHK(sb); 2328 2329 /* 2330 * Remove the delivered data from the socket buffer unless we 2331 * were only peeking. 2332 */ 2333 if (!(flags & MSG_PEEK)) { 2334 if (len > 0) 2335 sbdrop_locked(sb, len); 2336 2337 /* Notify protocol that we drained some data. */ 2338 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2339 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2340 !(flags & MSG_SOCALLBCK))) { 2341 SOCKBUF_UNLOCK(sb); 2342 VNET_SO_ASSERT(so); 2343 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags); 2344 SOCKBUF_LOCK(sb); 2345 } 2346 } 2347 2348 /* 2349 * For MSG_WAITALL we may have to loop again and wait for 2350 * more data to come in. 2351 */ 2352 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2353 goto restart; 2354 out: 2355 SOCKBUF_LOCK_ASSERT(sb); 2356 SBLASTRECORDCHK(sb); 2357 SBLASTMBUFCHK(sb); 2358 SOCKBUF_UNLOCK(sb); 2359 sbunlock(sb); 2360 return (error); 2361 } 2362 2363 /* 2364 * Optimized version of soreceive() for simple datagram cases from userspace. 2365 * Unlike in the stream case, we're able to drop a datagram if copyout() 2366 * fails, and because we handle datagrams atomically, we don't need to use a 2367 * sleep lock to prevent I/O interlacing. 2368 */ 2369 int 2370 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2371 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2372 { 2373 struct mbuf *m, *m2; 2374 int flags, error; 2375 ssize_t len; 2376 struct protosw *pr = so->so_proto; 2377 struct mbuf *nextrecord; 2378 2379 if (psa != NULL) 2380 *psa = NULL; 2381 if (controlp != NULL) 2382 *controlp = NULL; 2383 if (flagsp != NULL) 2384 flags = *flagsp &~ MSG_EOR; 2385 else 2386 flags = 0; 2387 2388 /* 2389 * For any complicated cases, fall back to the full 2390 * soreceive_generic(). 2391 */ 2392 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB)) 2393 return (soreceive_generic(so, psa, uio, mp0, controlp, 2394 flagsp)); 2395 2396 /* 2397 * Enforce restrictions on use. 2398 */ 2399 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2400 ("soreceive_dgram: wantrcvd")); 2401 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2402 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2403 ("soreceive_dgram: SBS_RCVATMARK")); 2404 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2405 ("soreceive_dgram: P_CONNREQUIRED")); 2406 2407 /* 2408 * Loop blocking while waiting for a datagram. 2409 */ 2410 SOCKBUF_LOCK(&so->so_rcv); 2411 while ((m = so->so_rcv.sb_mb) == NULL) { 2412 KASSERT(sbavail(&so->so_rcv) == 0, 2413 ("soreceive_dgram: sb_mb NULL but sbavail %u", 2414 sbavail(&so->so_rcv))); 2415 if (so->so_error) { 2416 error = so->so_error; 2417 so->so_error = 0; 2418 SOCKBUF_UNLOCK(&so->so_rcv); 2419 return (error); 2420 } 2421 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2422 uio->uio_resid == 0) { 2423 SOCKBUF_UNLOCK(&so->so_rcv); 2424 return (0); 2425 } 2426 if ((so->so_state & SS_NBIO) || 2427 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2428 SOCKBUF_UNLOCK(&so->so_rcv); 2429 return (EWOULDBLOCK); 2430 } 2431 SBLASTRECORDCHK(&so->so_rcv); 2432 SBLASTMBUFCHK(&so->so_rcv); 2433 error = sbwait(&so->so_rcv); 2434 if (error) { 2435 SOCKBUF_UNLOCK(&so->so_rcv); 2436 return (error); 2437 } 2438 } 2439 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2440 2441 if (uio->uio_td) 2442 uio->uio_td->td_ru.ru_msgrcv++; 2443 SBLASTRECORDCHK(&so->so_rcv); 2444 SBLASTMBUFCHK(&so->so_rcv); 2445 nextrecord = m->m_nextpkt; 2446 if (nextrecord == NULL) { 2447 KASSERT(so->so_rcv.sb_lastrecord == m, 2448 ("soreceive_dgram: lastrecord != m")); 2449 } 2450 2451 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2452 ("soreceive_dgram: m_nextpkt != nextrecord")); 2453 2454 /* 2455 * Pull 'm' and its chain off the front of the packet queue. 2456 */ 2457 so->so_rcv.sb_mb = NULL; 2458 sockbuf_pushsync(&so->so_rcv, nextrecord); 2459 2460 /* 2461 * Walk 'm's chain and free that many bytes from the socket buffer. 2462 */ 2463 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2464 sbfree(&so->so_rcv, m2); 2465 2466 /* 2467 * Do a few last checks before we let go of the lock. 2468 */ 2469 SBLASTRECORDCHK(&so->so_rcv); 2470 SBLASTMBUFCHK(&so->so_rcv); 2471 SOCKBUF_UNLOCK(&so->so_rcv); 2472 2473 if (pr->pr_flags & PR_ADDR) { 2474 KASSERT(m->m_type == MT_SONAME, 2475 ("m->m_type == %d", m->m_type)); 2476 if (psa != NULL) 2477 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2478 M_NOWAIT); 2479 m = m_free(m); 2480 } 2481 if (m == NULL) { 2482 /* XXXRW: Can this happen? */ 2483 return (0); 2484 } 2485 2486 /* 2487 * Packet to copyout() is now in 'm' and it is disconnected from the 2488 * queue. 2489 * 2490 * Process one or more MT_CONTROL mbufs present before any data mbufs 2491 * in the first mbuf chain on the socket buffer. We call into the 2492 * protocol to perform externalization (or freeing if controlp == 2493 * NULL). In some cases there can be only MT_CONTROL mbufs without 2494 * MT_DATA mbufs. 2495 */ 2496 if (m->m_type == MT_CONTROL) { 2497 struct mbuf *cm = NULL, *cmn; 2498 struct mbuf **cme = &cm; 2499 2500 do { 2501 m2 = m->m_next; 2502 m->m_next = NULL; 2503 *cme = m; 2504 cme = &(*cme)->m_next; 2505 m = m2; 2506 } while (m != NULL && m->m_type == MT_CONTROL); 2507 while (cm != NULL) { 2508 cmn = cm->m_next; 2509 cm->m_next = NULL; 2510 if (pr->pr_domain->dom_externalize != NULL) { 2511 error = (*pr->pr_domain->dom_externalize) 2512 (cm, controlp, flags); 2513 } else if (controlp != NULL) 2514 *controlp = cm; 2515 else 2516 m_freem(cm); 2517 if (controlp != NULL) { 2518 while (*controlp != NULL) 2519 controlp = &(*controlp)->m_next; 2520 } 2521 cm = cmn; 2522 } 2523 } 2524 KASSERT(m == NULL || m->m_type == MT_DATA, 2525 ("soreceive_dgram: !data")); 2526 while (m != NULL && uio->uio_resid > 0) { 2527 len = uio->uio_resid; 2528 if (len > m->m_len) 2529 len = m->m_len; 2530 error = uiomove(mtod(m, char *), (int)len, uio); 2531 if (error) { 2532 m_freem(m); 2533 return (error); 2534 } 2535 if (len == m->m_len) 2536 m = m_free(m); 2537 else { 2538 m->m_data += len; 2539 m->m_len -= len; 2540 } 2541 } 2542 if (m != NULL) { 2543 flags |= MSG_TRUNC; 2544 m_freem(m); 2545 } 2546 if (flagsp != NULL) 2547 *flagsp |= flags; 2548 return (0); 2549 } 2550 2551 int 2552 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2553 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2554 { 2555 int error; 2556 2557 CURVNET_SET(so->so_vnet); 2558 if (!SOLISTENING(so)) 2559 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, 2560 mp0, controlp, flagsp)); 2561 else 2562 error = ENOTCONN; 2563 CURVNET_RESTORE(); 2564 return (error); 2565 } 2566 2567 int 2568 soshutdown(struct socket *so, int how) 2569 { 2570 struct protosw *pr = so->so_proto; 2571 int error, soerror_enotconn; 2572 2573 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2574 return (EINVAL); 2575 2576 soerror_enotconn = 0; 2577 if ((so->so_state & 2578 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 2579 /* 2580 * POSIX mandates us to return ENOTCONN when shutdown(2) is 2581 * invoked on a datagram sockets, however historically we would 2582 * actually tear socket down. This is known to be leveraged by 2583 * some applications to unblock process waiting in recvXXX(2) 2584 * by other process that it shares that socket with. Try to meet 2585 * both backward-compatibility and POSIX requirements by forcing 2586 * ENOTCONN but still asking protocol to perform pru_shutdown(). 2587 */ 2588 if (so->so_type != SOCK_DGRAM) 2589 return (ENOTCONN); 2590 soerror_enotconn = 1; 2591 } 2592 2593 CURVNET_SET(so->so_vnet); 2594 if (pr->pr_usrreqs->pru_flush != NULL) 2595 (*pr->pr_usrreqs->pru_flush)(so, how); 2596 if (how != SHUT_WR) 2597 sorflush(so); 2598 if (how != SHUT_RD) { 2599 error = (*pr->pr_usrreqs->pru_shutdown)(so); 2600 wakeup(&so->so_timeo); 2601 CURVNET_RESTORE(); 2602 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error); 2603 } 2604 wakeup(&so->so_timeo); 2605 CURVNET_RESTORE(); 2606 2607 return (soerror_enotconn ? ENOTCONN : 0); 2608 } 2609 2610 void 2611 sorflush(struct socket *so) 2612 { 2613 struct sockbuf *sb = &so->so_rcv; 2614 struct protosw *pr = so->so_proto; 2615 struct socket aso; 2616 2617 VNET_SO_ASSERT(so); 2618 2619 /* 2620 * In order to avoid calling dom_dispose with the socket buffer mutex 2621 * held, and in order to generally avoid holding the lock for a long 2622 * time, we make a copy of the socket buffer and clear the original 2623 * (except locks, state). The new socket buffer copy won't have 2624 * initialized locks so we can only call routines that won't use or 2625 * assert those locks. 2626 * 2627 * Dislodge threads currently blocked in receive and wait to acquire 2628 * a lock against other simultaneous readers before clearing the 2629 * socket buffer. Don't let our acquire be interrupted by a signal 2630 * despite any existing socket disposition on interruptable waiting. 2631 */ 2632 socantrcvmore(so); 2633 (void) sblock(sb, SBL_WAIT | SBL_NOINTR); 2634 2635 /* 2636 * Invalidate/clear most of the sockbuf structure, but leave selinfo 2637 * and mutex data unchanged. 2638 */ 2639 SOCKBUF_LOCK(sb); 2640 bzero(&aso, sizeof(aso)); 2641 aso.so_pcb = so->so_pcb; 2642 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero, 2643 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2644 bzero(&sb->sb_startzero, 2645 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2646 SOCKBUF_UNLOCK(sb); 2647 sbunlock(sb); 2648 2649 /* 2650 * Dispose of special rights and flush the copied socket. Don't call 2651 * any unsafe routines (that rely on locks being initialized) on aso. 2652 */ 2653 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 2654 (*pr->pr_domain->dom_dispose)(&aso); 2655 sbrelease_internal(&aso.so_rcv, so); 2656 } 2657 2658 /* 2659 * Wrapper for Socket established helper hook. 2660 * Parameters: socket, context of the hook point, hook id. 2661 */ 2662 static int inline 2663 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 2664 { 2665 struct socket_hhook_data hhook_data = { 2666 .so = so, 2667 .hctx = hctx, 2668 .m = NULL, 2669 .status = 0 2670 }; 2671 2672 CURVNET_SET(so->so_vnet); 2673 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 2674 CURVNET_RESTORE(); 2675 2676 /* Ugly but needed, since hhooks return void for now */ 2677 return (hhook_data.status); 2678 } 2679 2680 /* 2681 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 2682 * additional variant to handle the case where the option value needs to be 2683 * some kind of integer, but not a specific size. In addition to their use 2684 * here, these functions are also called by the protocol-level pr_ctloutput() 2685 * routines. 2686 */ 2687 int 2688 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2689 { 2690 size_t valsize; 2691 2692 /* 2693 * If the user gives us more than we wanted, we ignore it, but if we 2694 * don't get the minimum length the caller wants, we return EINVAL. 2695 * On success, sopt->sopt_valsize is set to however much we actually 2696 * retrieved. 2697 */ 2698 if ((valsize = sopt->sopt_valsize) < minlen) 2699 return EINVAL; 2700 if (valsize > len) 2701 sopt->sopt_valsize = valsize = len; 2702 2703 if (sopt->sopt_td != NULL) 2704 return (copyin(sopt->sopt_val, buf, valsize)); 2705 2706 bcopy(sopt->sopt_val, buf, valsize); 2707 return (0); 2708 } 2709 2710 /* 2711 * Kernel version of setsockopt(2). 2712 * 2713 * XXX: optlen is size_t, not socklen_t 2714 */ 2715 int 2716 so_setsockopt(struct socket *so, int level, int optname, void *optval, 2717 size_t optlen) 2718 { 2719 struct sockopt sopt; 2720 2721 sopt.sopt_level = level; 2722 sopt.sopt_name = optname; 2723 sopt.sopt_dir = SOPT_SET; 2724 sopt.sopt_val = optval; 2725 sopt.sopt_valsize = optlen; 2726 sopt.sopt_td = NULL; 2727 return (sosetopt(so, &sopt)); 2728 } 2729 2730 int 2731 sosetopt(struct socket *so, struct sockopt *sopt) 2732 { 2733 int error, optval; 2734 struct linger l; 2735 struct timeval tv; 2736 sbintime_t val; 2737 uint32_t val32; 2738 #ifdef MAC 2739 struct mac extmac; 2740 #endif 2741 2742 CURVNET_SET(so->so_vnet); 2743 error = 0; 2744 if (sopt->sopt_level != SOL_SOCKET) { 2745 if (so->so_proto->pr_ctloutput != NULL) { 2746 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2747 CURVNET_RESTORE(); 2748 return (error); 2749 } 2750 error = ENOPROTOOPT; 2751 } else { 2752 switch (sopt->sopt_name) { 2753 case SO_ACCEPTFILTER: 2754 error = accept_filt_setopt(so, sopt); 2755 if (error) 2756 goto bad; 2757 break; 2758 2759 case SO_LINGER: 2760 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 2761 if (error) 2762 goto bad; 2763 2764 SOCK_LOCK(so); 2765 so->so_linger = l.l_linger; 2766 if (l.l_onoff) 2767 so->so_options |= SO_LINGER; 2768 else 2769 so->so_options &= ~SO_LINGER; 2770 SOCK_UNLOCK(so); 2771 break; 2772 2773 case SO_DEBUG: 2774 case SO_KEEPALIVE: 2775 case SO_DONTROUTE: 2776 case SO_USELOOPBACK: 2777 case SO_BROADCAST: 2778 case SO_REUSEADDR: 2779 case SO_REUSEPORT: 2780 case SO_REUSEPORT_LB: 2781 case SO_OOBINLINE: 2782 case SO_TIMESTAMP: 2783 case SO_BINTIME: 2784 case SO_NOSIGPIPE: 2785 case SO_NO_DDP: 2786 case SO_NO_OFFLOAD: 2787 error = sooptcopyin(sopt, &optval, sizeof optval, 2788 sizeof optval); 2789 if (error) 2790 goto bad; 2791 SOCK_LOCK(so); 2792 if (optval) 2793 so->so_options |= sopt->sopt_name; 2794 else 2795 so->so_options &= ~sopt->sopt_name; 2796 SOCK_UNLOCK(so); 2797 break; 2798 2799 case SO_SETFIB: 2800 error = sooptcopyin(sopt, &optval, sizeof optval, 2801 sizeof optval); 2802 if (error) 2803 goto bad; 2804 2805 if (optval < 0 || optval >= rt_numfibs) { 2806 error = EINVAL; 2807 goto bad; 2808 } 2809 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 2810 (so->so_proto->pr_domain->dom_family == PF_INET6) || 2811 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 2812 so->so_fibnum = optval; 2813 else 2814 so->so_fibnum = 0; 2815 break; 2816 2817 case SO_USER_COOKIE: 2818 error = sooptcopyin(sopt, &val32, sizeof val32, 2819 sizeof val32); 2820 if (error) 2821 goto bad; 2822 so->so_user_cookie = val32; 2823 break; 2824 2825 case SO_SNDBUF: 2826 case SO_RCVBUF: 2827 case SO_SNDLOWAT: 2828 case SO_RCVLOWAT: 2829 error = sooptcopyin(sopt, &optval, sizeof optval, 2830 sizeof optval); 2831 if (error) 2832 goto bad; 2833 2834 /* 2835 * Values < 1 make no sense for any of these options, 2836 * so disallow them. 2837 */ 2838 if (optval < 1) { 2839 error = EINVAL; 2840 goto bad; 2841 } 2842 2843 error = sbsetopt(so, sopt->sopt_name, optval); 2844 break; 2845 2846 case SO_SNDTIMEO: 2847 case SO_RCVTIMEO: 2848 #ifdef COMPAT_FREEBSD32 2849 if (SV_CURPROC_FLAG(SV_ILP32)) { 2850 struct timeval32 tv32; 2851 2852 error = sooptcopyin(sopt, &tv32, sizeof tv32, 2853 sizeof tv32); 2854 CP(tv32, tv, tv_sec); 2855 CP(tv32, tv, tv_usec); 2856 } else 2857 #endif 2858 error = sooptcopyin(sopt, &tv, sizeof tv, 2859 sizeof tv); 2860 if (error) 2861 goto bad; 2862 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 2863 tv.tv_usec >= 1000000) { 2864 error = EDOM; 2865 goto bad; 2866 } 2867 if (tv.tv_sec > INT32_MAX) 2868 val = SBT_MAX; 2869 else 2870 val = tvtosbt(tv); 2871 switch (sopt->sopt_name) { 2872 case SO_SNDTIMEO: 2873 so->so_snd.sb_timeo = val; 2874 break; 2875 case SO_RCVTIMEO: 2876 so->so_rcv.sb_timeo = val; 2877 break; 2878 } 2879 break; 2880 2881 case SO_LABEL: 2882 #ifdef MAC 2883 error = sooptcopyin(sopt, &extmac, sizeof extmac, 2884 sizeof extmac); 2885 if (error) 2886 goto bad; 2887 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 2888 so, &extmac); 2889 #else 2890 error = EOPNOTSUPP; 2891 #endif 2892 break; 2893 2894 case SO_TS_CLOCK: 2895 error = sooptcopyin(sopt, &optval, sizeof optval, 2896 sizeof optval); 2897 if (error) 2898 goto bad; 2899 if (optval < 0 || optval > SO_TS_CLOCK_MAX) { 2900 error = EINVAL; 2901 goto bad; 2902 } 2903 so->so_ts_clock = optval; 2904 break; 2905 2906 case SO_MAX_PACING_RATE: 2907 error = sooptcopyin(sopt, &val32, sizeof(val32), 2908 sizeof(val32)); 2909 if (error) 2910 goto bad; 2911 so->so_max_pacing_rate = val32; 2912 break; 2913 2914 default: 2915 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 2916 error = hhook_run_socket(so, sopt, 2917 HHOOK_SOCKET_OPT); 2918 else 2919 error = ENOPROTOOPT; 2920 break; 2921 } 2922 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 2923 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 2924 } 2925 bad: 2926 CURVNET_RESTORE(); 2927 return (error); 2928 } 2929 2930 /* 2931 * Helper routine for getsockopt. 2932 */ 2933 int 2934 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 2935 { 2936 int error; 2937 size_t valsize; 2938 2939 error = 0; 2940 2941 /* 2942 * Documented get behavior is that we always return a value, possibly 2943 * truncated to fit in the user's buffer. Traditional behavior is 2944 * that we always tell the user precisely how much we copied, rather 2945 * than something useful like the total amount we had available for 2946 * her. Note that this interface is not idempotent; the entire 2947 * answer must be generated ahead of time. 2948 */ 2949 valsize = min(len, sopt->sopt_valsize); 2950 sopt->sopt_valsize = valsize; 2951 if (sopt->sopt_val != NULL) { 2952 if (sopt->sopt_td != NULL) 2953 error = copyout(buf, sopt->sopt_val, valsize); 2954 else 2955 bcopy(buf, sopt->sopt_val, valsize); 2956 } 2957 return (error); 2958 } 2959 2960 int 2961 sogetopt(struct socket *so, struct sockopt *sopt) 2962 { 2963 int error, optval; 2964 struct linger l; 2965 struct timeval tv; 2966 #ifdef MAC 2967 struct mac extmac; 2968 #endif 2969 2970 CURVNET_SET(so->so_vnet); 2971 error = 0; 2972 if (sopt->sopt_level != SOL_SOCKET) { 2973 if (so->so_proto->pr_ctloutput != NULL) 2974 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2975 else 2976 error = ENOPROTOOPT; 2977 CURVNET_RESTORE(); 2978 return (error); 2979 } else { 2980 switch (sopt->sopt_name) { 2981 case SO_ACCEPTFILTER: 2982 error = accept_filt_getopt(so, sopt); 2983 break; 2984 2985 case SO_LINGER: 2986 SOCK_LOCK(so); 2987 l.l_onoff = so->so_options & SO_LINGER; 2988 l.l_linger = so->so_linger; 2989 SOCK_UNLOCK(so); 2990 error = sooptcopyout(sopt, &l, sizeof l); 2991 break; 2992 2993 case SO_USELOOPBACK: 2994 case SO_DONTROUTE: 2995 case SO_DEBUG: 2996 case SO_KEEPALIVE: 2997 case SO_REUSEADDR: 2998 case SO_REUSEPORT: 2999 case SO_REUSEPORT_LB: 3000 case SO_BROADCAST: 3001 case SO_OOBINLINE: 3002 case SO_ACCEPTCONN: 3003 case SO_TIMESTAMP: 3004 case SO_BINTIME: 3005 case SO_NOSIGPIPE: 3006 optval = so->so_options & sopt->sopt_name; 3007 integer: 3008 error = sooptcopyout(sopt, &optval, sizeof optval); 3009 break; 3010 3011 case SO_DOMAIN: 3012 optval = so->so_proto->pr_domain->dom_family; 3013 goto integer; 3014 3015 case SO_TYPE: 3016 optval = so->so_type; 3017 goto integer; 3018 3019 case SO_PROTOCOL: 3020 optval = so->so_proto->pr_protocol; 3021 goto integer; 3022 3023 case SO_ERROR: 3024 SOCK_LOCK(so); 3025 optval = so->so_error; 3026 so->so_error = 0; 3027 SOCK_UNLOCK(so); 3028 goto integer; 3029 3030 case SO_SNDBUF: 3031 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : 3032 so->so_snd.sb_hiwat; 3033 goto integer; 3034 3035 case SO_RCVBUF: 3036 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : 3037 so->so_rcv.sb_hiwat; 3038 goto integer; 3039 3040 case SO_SNDLOWAT: 3041 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : 3042 so->so_snd.sb_lowat; 3043 goto integer; 3044 3045 case SO_RCVLOWAT: 3046 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : 3047 so->so_rcv.sb_lowat; 3048 goto integer; 3049 3050 case SO_SNDTIMEO: 3051 case SO_RCVTIMEO: 3052 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 3053 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 3054 #ifdef COMPAT_FREEBSD32 3055 if (SV_CURPROC_FLAG(SV_ILP32)) { 3056 struct timeval32 tv32; 3057 3058 CP(tv, tv32, tv_sec); 3059 CP(tv, tv32, tv_usec); 3060 error = sooptcopyout(sopt, &tv32, sizeof tv32); 3061 } else 3062 #endif 3063 error = sooptcopyout(sopt, &tv, sizeof tv); 3064 break; 3065 3066 case SO_LABEL: 3067 #ifdef MAC 3068 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3069 sizeof(extmac)); 3070 if (error) 3071 goto bad; 3072 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 3073 so, &extmac); 3074 if (error) 3075 goto bad; 3076 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3077 #else 3078 error = EOPNOTSUPP; 3079 #endif 3080 break; 3081 3082 case SO_PEERLABEL: 3083 #ifdef MAC 3084 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3085 sizeof(extmac)); 3086 if (error) 3087 goto bad; 3088 error = mac_getsockopt_peerlabel( 3089 sopt->sopt_td->td_ucred, so, &extmac); 3090 if (error) 3091 goto bad; 3092 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3093 #else 3094 error = EOPNOTSUPP; 3095 #endif 3096 break; 3097 3098 case SO_LISTENQLIMIT: 3099 optval = SOLISTENING(so) ? so->sol_qlimit : 0; 3100 goto integer; 3101 3102 case SO_LISTENQLEN: 3103 optval = SOLISTENING(so) ? so->sol_qlen : 0; 3104 goto integer; 3105 3106 case SO_LISTENINCQLEN: 3107 optval = SOLISTENING(so) ? so->sol_incqlen : 0; 3108 goto integer; 3109 3110 case SO_TS_CLOCK: 3111 optval = so->so_ts_clock; 3112 goto integer; 3113 3114 case SO_MAX_PACING_RATE: 3115 optval = so->so_max_pacing_rate; 3116 goto integer; 3117 3118 default: 3119 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3120 error = hhook_run_socket(so, sopt, 3121 HHOOK_SOCKET_OPT); 3122 else 3123 error = ENOPROTOOPT; 3124 break; 3125 } 3126 } 3127 #ifdef MAC 3128 bad: 3129 #endif 3130 CURVNET_RESTORE(); 3131 return (error); 3132 } 3133 3134 int 3135 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 3136 { 3137 struct mbuf *m, *m_prev; 3138 int sopt_size = sopt->sopt_valsize; 3139 3140 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3141 if (m == NULL) 3142 return ENOBUFS; 3143 if (sopt_size > MLEN) { 3144 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 3145 if ((m->m_flags & M_EXT) == 0) { 3146 m_free(m); 3147 return ENOBUFS; 3148 } 3149 m->m_len = min(MCLBYTES, sopt_size); 3150 } else { 3151 m->m_len = min(MLEN, sopt_size); 3152 } 3153 sopt_size -= m->m_len; 3154 *mp = m; 3155 m_prev = m; 3156 3157 while (sopt_size) { 3158 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3159 if (m == NULL) { 3160 m_freem(*mp); 3161 return ENOBUFS; 3162 } 3163 if (sopt_size > MLEN) { 3164 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 3165 M_NOWAIT); 3166 if ((m->m_flags & M_EXT) == 0) { 3167 m_freem(m); 3168 m_freem(*mp); 3169 return ENOBUFS; 3170 } 3171 m->m_len = min(MCLBYTES, sopt_size); 3172 } else { 3173 m->m_len = min(MLEN, sopt_size); 3174 } 3175 sopt_size -= m->m_len; 3176 m_prev->m_next = m; 3177 m_prev = m; 3178 } 3179 return (0); 3180 } 3181 3182 int 3183 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 3184 { 3185 struct mbuf *m0 = m; 3186 3187 if (sopt->sopt_val == NULL) 3188 return (0); 3189 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3190 if (sopt->sopt_td != NULL) { 3191 int error; 3192 3193 error = copyin(sopt->sopt_val, mtod(m, char *), 3194 m->m_len); 3195 if (error != 0) { 3196 m_freem(m0); 3197 return(error); 3198 } 3199 } else 3200 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 3201 sopt->sopt_valsize -= m->m_len; 3202 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3203 m = m->m_next; 3204 } 3205 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 3206 panic("ip6_sooptmcopyin"); 3207 return (0); 3208 } 3209 3210 int 3211 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 3212 { 3213 struct mbuf *m0 = m; 3214 size_t valsize = 0; 3215 3216 if (sopt->sopt_val == NULL) 3217 return (0); 3218 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3219 if (sopt->sopt_td != NULL) { 3220 int error; 3221 3222 error = copyout(mtod(m, char *), sopt->sopt_val, 3223 m->m_len); 3224 if (error != 0) { 3225 m_freem(m0); 3226 return(error); 3227 } 3228 } else 3229 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 3230 sopt->sopt_valsize -= m->m_len; 3231 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3232 valsize += m->m_len; 3233 m = m->m_next; 3234 } 3235 if (m != NULL) { 3236 /* enough soopt buffer should be given from user-land */ 3237 m_freem(m0); 3238 return(EINVAL); 3239 } 3240 sopt->sopt_valsize = valsize; 3241 return (0); 3242 } 3243 3244 /* 3245 * sohasoutofband(): protocol notifies socket layer of the arrival of new 3246 * out-of-band data, which will then notify socket consumers. 3247 */ 3248 void 3249 sohasoutofband(struct socket *so) 3250 { 3251 3252 if (so->so_sigio != NULL) 3253 pgsigio(&so->so_sigio, SIGURG, 0); 3254 selwakeuppri(&so->so_rdsel, PSOCK); 3255 } 3256 3257 int 3258 sopoll(struct socket *so, int events, struct ucred *active_cred, 3259 struct thread *td) 3260 { 3261 3262 /* 3263 * We do not need to set or assert curvnet as long as everyone uses 3264 * sopoll_generic(). 3265 */ 3266 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, 3267 td)); 3268 } 3269 3270 int 3271 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3272 struct thread *td) 3273 { 3274 int revents; 3275 3276 SOCK_LOCK(so); 3277 if (SOLISTENING(so)) { 3278 if (!(events & (POLLIN | POLLRDNORM))) 3279 revents = 0; 3280 else if (!TAILQ_EMPTY(&so->sol_comp)) 3281 revents = events & (POLLIN | POLLRDNORM); 3282 else { 3283 selrecord(td, &so->so_rdsel); 3284 revents = 0; 3285 } 3286 } else { 3287 revents = 0; 3288 SOCKBUF_LOCK(&so->so_snd); 3289 SOCKBUF_LOCK(&so->so_rcv); 3290 if (events & (POLLIN | POLLRDNORM)) 3291 if (soreadabledata(so)) 3292 revents |= events & (POLLIN | POLLRDNORM); 3293 if (events & (POLLOUT | POLLWRNORM)) 3294 if (sowriteable(so)) 3295 revents |= events & (POLLOUT | POLLWRNORM); 3296 if (events & (POLLPRI | POLLRDBAND)) 3297 if (so->so_oobmark || 3298 (so->so_rcv.sb_state & SBS_RCVATMARK)) 3299 revents |= events & (POLLPRI | POLLRDBAND); 3300 if ((events & POLLINIGNEOF) == 0) { 3301 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3302 revents |= events & (POLLIN | POLLRDNORM); 3303 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3304 revents |= POLLHUP; 3305 } 3306 } 3307 if (revents == 0) { 3308 if (events & 3309 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 3310 selrecord(td, &so->so_rdsel); 3311 so->so_rcv.sb_flags |= SB_SEL; 3312 } 3313 if (events & (POLLOUT | POLLWRNORM)) { 3314 selrecord(td, &so->so_wrsel); 3315 so->so_snd.sb_flags |= SB_SEL; 3316 } 3317 } 3318 SOCKBUF_UNLOCK(&so->so_rcv); 3319 SOCKBUF_UNLOCK(&so->so_snd); 3320 } 3321 SOCK_UNLOCK(so); 3322 return (revents); 3323 } 3324 3325 int 3326 soo_kqfilter(struct file *fp, struct knote *kn) 3327 { 3328 struct socket *so = kn->kn_fp->f_data; 3329 struct sockbuf *sb; 3330 struct knlist *knl; 3331 3332 switch (kn->kn_filter) { 3333 case EVFILT_READ: 3334 kn->kn_fop = &soread_filtops; 3335 knl = &so->so_rdsel.si_note; 3336 sb = &so->so_rcv; 3337 break; 3338 case EVFILT_WRITE: 3339 kn->kn_fop = &sowrite_filtops; 3340 knl = &so->so_wrsel.si_note; 3341 sb = &so->so_snd; 3342 break; 3343 case EVFILT_EMPTY: 3344 kn->kn_fop = &soempty_filtops; 3345 knl = &so->so_wrsel.si_note; 3346 sb = &so->so_snd; 3347 break; 3348 default: 3349 return (EINVAL); 3350 } 3351 3352 SOCK_LOCK(so); 3353 if (SOLISTENING(so)) { 3354 knlist_add(knl, kn, 1); 3355 } else { 3356 SOCKBUF_LOCK(sb); 3357 knlist_add(knl, kn, 1); 3358 sb->sb_flags |= SB_KNOTE; 3359 SOCKBUF_UNLOCK(sb); 3360 } 3361 SOCK_UNLOCK(so); 3362 return (0); 3363 } 3364 3365 /* 3366 * Some routines that return EOPNOTSUPP for entry points that are not 3367 * supported by a protocol. Fill in as needed. 3368 */ 3369 int 3370 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 3371 { 3372 3373 return EOPNOTSUPP; 3374 } 3375 3376 int 3377 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job) 3378 { 3379 3380 return EOPNOTSUPP; 3381 } 3382 3383 int 3384 pru_attach_notsupp(struct socket *so, int proto, struct thread *td) 3385 { 3386 3387 return EOPNOTSUPP; 3388 } 3389 3390 int 3391 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3392 { 3393 3394 return EOPNOTSUPP; 3395 } 3396 3397 int 3398 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3399 struct thread *td) 3400 { 3401 3402 return EOPNOTSUPP; 3403 } 3404 3405 int 3406 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3407 { 3408 3409 return EOPNOTSUPP; 3410 } 3411 3412 int 3413 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3414 struct thread *td) 3415 { 3416 3417 return EOPNOTSUPP; 3418 } 3419 3420 int 3421 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 3422 { 3423 3424 return EOPNOTSUPP; 3425 } 3426 3427 int 3428 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 3429 struct ifnet *ifp, struct thread *td) 3430 { 3431 3432 return EOPNOTSUPP; 3433 } 3434 3435 int 3436 pru_disconnect_notsupp(struct socket *so) 3437 { 3438 3439 return EOPNOTSUPP; 3440 } 3441 3442 int 3443 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) 3444 { 3445 3446 return EOPNOTSUPP; 3447 } 3448 3449 int 3450 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) 3451 { 3452 3453 return EOPNOTSUPP; 3454 } 3455 3456 int 3457 pru_rcvd_notsupp(struct socket *so, int flags) 3458 { 3459 3460 return EOPNOTSUPP; 3461 } 3462 3463 int 3464 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 3465 { 3466 3467 return EOPNOTSUPP; 3468 } 3469 3470 int 3471 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, 3472 struct sockaddr *addr, struct mbuf *control, struct thread *td) 3473 { 3474 3475 return EOPNOTSUPP; 3476 } 3477 3478 int 3479 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count) 3480 { 3481 3482 return (EOPNOTSUPP); 3483 } 3484 3485 /* 3486 * This isn't really a ``null'' operation, but it's the default one and 3487 * doesn't do anything destructive. 3488 */ 3489 int 3490 pru_sense_null(struct socket *so, struct stat *sb) 3491 { 3492 3493 sb->st_blksize = so->so_snd.sb_hiwat; 3494 return 0; 3495 } 3496 3497 int 3498 pru_shutdown_notsupp(struct socket *so) 3499 { 3500 3501 return EOPNOTSUPP; 3502 } 3503 3504 int 3505 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) 3506 { 3507 3508 return EOPNOTSUPP; 3509 } 3510 3511 int 3512 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 3513 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 3514 { 3515 3516 return EOPNOTSUPP; 3517 } 3518 3519 int 3520 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 3521 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 3522 { 3523 3524 return EOPNOTSUPP; 3525 } 3526 3527 int 3528 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, 3529 struct thread *td) 3530 { 3531 3532 return EOPNOTSUPP; 3533 } 3534 3535 static void 3536 filt_sordetach(struct knote *kn) 3537 { 3538 struct socket *so = kn->kn_fp->f_data; 3539 3540 so_rdknl_lock(so); 3541 knlist_remove(&so->so_rdsel.si_note, kn, 1); 3542 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note)) 3543 so->so_rcv.sb_flags &= ~SB_KNOTE; 3544 so_rdknl_unlock(so); 3545 } 3546 3547 /*ARGSUSED*/ 3548 static int 3549 filt_soread(struct knote *kn, long hint) 3550 { 3551 struct socket *so; 3552 3553 so = kn->kn_fp->f_data; 3554 3555 if (SOLISTENING(so)) { 3556 SOCK_LOCK_ASSERT(so); 3557 kn->kn_data = so->sol_qlen; 3558 return (!TAILQ_EMPTY(&so->sol_comp)); 3559 } 3560 3561 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3562 3563 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 3564 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3565 kn->kn_flags |= EV_EOF; 3566 kn->kn_fflags = so->so_error; 3567 return (1); 3568 } else if (so->so_error) /* temporary udp error */ 3569 return (1); 3570 3571 if (kn->kn_sfflags & NOTE_LOWAT) { 3572 if (kn->kn_data >= kn->kn_sdata) 3573 return (1); 3574 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) 3575 return (1); 3576 3577 /* This hook returning non-zero indicates an event, not error */ 3578 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); 3579 } 3580 3581 static void 3582 filt_sowdetach(struct knote *kn) 3583 { 3584 struct socket *so = kn->kn_fp->f_data; 3585 3586 so_wrknl_lock(so); 3587 knlist_remove(&so->so_wrsel.si_note, kn, 1); 3588 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note)) 3589 so->so_snd.sb_flags &= ~SB_KNOTE; 3590 so_wrknl_unlock(so); 3591 } 3592 3593 /*ARGSUSED*/ 3594 static int 3595 filt_sowrite(struct knote *kn, long hint) 3596 { 3597 struct socket *so; 3598 3599 so = kn->kn_fp->f_data; 3600 3601 if (SOLISTENING(so)) 3602 return (0); 3603 3604 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3605 kn->kn_data = sbspace(&so->so_snd); 3606 3607 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); 3608 3609 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3610 kn->kn_flags |= EV_EOF; 3611 kn->kn_fflags = so->so_error; 3612 return (1); 3613 } else if (so->so_error) /* temporary udp error */ 3614 return (1); 3615 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3616 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3617 return (0); 3618 else if (kn->kn_sfflags & NOTE_LOWAT) 3619 return (kn->kn_data >= kn->kn_sdata); 3620 else 3621 return (kn->kn_data >= so->so_snd.sb_lowat); 3622 } 3623 3624 static int 3625 filt_soempty(struct knote *kn, long hint) 3626 { 3627 struct socket *so; 3628 3629 so = kn->kn_fp->f_data; 3630 3631 if (SOLISTENING(so)) 3632 return (1); 3633 3634 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3635 kn->kn_data = sbused(&so->so_snd); 3636 3637 if (kn->kn_data == 0) 3638 return (1); 3639 else 3640 return (0); 3641 } 3642 3643 int 3644 socheckuid(struct socket *so, uid_t uid) 3645 { 3646 3647 if (so == NULL) 3648 return (EPERM); 3649 if (so->so_cred->cr_uid != uid) 3650 return (EPERM); 3651 return (0); 3652 } 3653 3654 /* 3655 * These functions are used by protocols to notify the socket layer (and its 3656 * consumers) of state changes in the sockets driven by protocol-side events. 3657 */ 3658 3659 /* 3660 * Procedures to manipulate state flags of socket and do appropriate wakeups. 3661 * 3662 * Normal sequence from the active (originating) side is that 3663 * soisconnecting() is called during processing of connect() call, resulting 3664 * in an eventual call to soisconnected() if/when the connection is 3665 * established. When the connection is torn down soisdisconnecting() is 3666 * called during processing of disconnect() call, and soisdisconnected() is 3667 * called when the connection to the peer is totally severed. The semantics 3668 * of these routines are such that connectionless protocols can call 3669 * soisconnected() and soisdisconnected() only, bypassing the in-progress 3670 * calls when setting up a ``connection'' takes no time. 3671 * 3672 * From the passive side, a socket is created with two queues of sockets: 3673 * so_incomp for connections in progress and so_comp for connections already 3674 * made and awaiting user acceptance. As a protocol is preparing incoming 3675 * connections, it creates a socket structure queued on so_incomp by calling 3676 * sonewconn(). When the connection is established, soisconnected() is 3677 * called, and transfers the socket structure to so_comp, making it available 3678 * to accept(). 3679 * 3680 * If a socket is closed with sockets on either so_incomp or so_comp, these 3681 * sockets are dropped. 3682 * 3683 * If higher-level protocols are implemented in the kernel, the wakeups done 3684 * here will sometimes cause software-interrupt process scheduling. 3685 */ 3686 void 3687 soisconnecting(struct socket *so) 3688 { 3689 3690 SOCK_LOCK(so); 3691 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 3692 so->so_state |= SS_ISCONNECTING; 3693 SOCK_UNLOCK(so); 3694 } 3695 3696 void 3697 soisconnected(struct socket *so) 3698 { 3699 3700 SOCK_LOCK(so); 3701 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 3702 so->so_state |= SS_ISCONNECTED; 3703 3704 if (so->so_qstate == SQ_INCOMP) { 3705 struct socket *head = so->so_listen; 3706 int ret; 3707 3708 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so)); 3709 /* 3710 * Promoting a socket from incomplete queue to complete, we 3711 * need to go through reverse order of locking. We first do 3712 * trylock, and if that doesn't succeed, we go the hard way 3713 * leaving a reference and rechecking consistency after proper 3714 * locking. 3715 */ 3716 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) { 3717 soref(head); 3718 SOCK_UNLOCK(so); 3719 SOLISTEN_LOCK(head); 3720 SOCK_LOCK(so); 3721 if (__predict_false(head != so->so_listen)) { 3722 /* 3723 * The socket went off the listen queue, 3724 * should be lost race to close(2) of sol. 3725 * The socket is about to soabort(). 3726 */ 3727 SOCK_UNLOCK(so); 3728 sorele(head); 3729 return; 3730 } 3731 /* Not the last one, as so holds a ref. */ 3732 refcount_release(&head->so_count); 3733 } 3734 again: 3735 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 3736 TAILQ_REMOVE(&head->sol_incomp, so, so_list); 3737 head->sol_incqlen--; 3738 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 3739 head->sol_qlen++; 3740 so->so_qstate = SQ_COMP; 3741 SOCK_UNLOCK(so); 3742 solisten_wakeup(head); /* unlocks */ 3743 } else { 3744 SOCKBUF_LOCK(&so->so_rcv); 3745 soupcall_set(so, SO_RCV, 3746 head->sol_accept_filter->accf_callback, 3747 head->sol_accept_filter_arg); 3748 so->so_options &= ~SO_ACCEPTFILTER; 3749 ret = head->sol_accept_filter->accf_callback(so, 3750 head->sol_accept_filter_arg, M_NOWAIT); 3751 if (ret == SU_ISCONNECTED) { 3752 soupcall_clear(so, SO_RCV); 3753 SOCKBUF_UNLOCK(&so->so_rcv); 3754 goto again; 3755 } 3756 SOCKBUF_UNLOCK(&so->so_rcv); 3757 SOCK_UNLOCK(so); 3758 SOLISTEN_UNLOCK(head); 3759 } 3760 return; 3761 } 3762 SOCK_UNLOCK(so); 3763 wakeup(&so->so_timeo); 3764 sorwakeup(so); 3765 sowwakeup(so); 3766 } 3767 3768 void 3769 soisdisconnecting(struct socket *so) 3770 { 3771 3772 SOCK_LOCK(so); 3773 so->so_state &= ~SS_ISCONNECTING; 3774 so->so_state |= SS_ISDISCONNECTING; 3775 3776 if (!SOLISTENING(so)) { 3777 SOCKBUF_LOCK(&so->so_rcv); 3778 socantrcvmore_locked(so); 3779 SOCKBUF_LOCK(&so->so_snd); 3780 socantsendmore_locked(so); 3781 } 3782 SOCK_UNLOCK(so); 3783 wakeup(&so->so_timeo); 3784 } 3785 3786 void 3787 soisdisconnected(struct socket *so) 3788 { 3789 3790 SOCK_LOCK(so); 3791 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 3792 so->so_state |= SS_ISDISCONNECTED; 3793 3794 if (!SOLISTENING(so)) { 3795 SOCK_UNLOCK(so); 3796 SOCKBUF_LOCK(&so->so_rcv); 3797 socantrcvmore_locked(so); 3798 SOCKBUF_LOCK(&so->so_snd); 3799 sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); 3800 socantsendmore_locked(so); 3801 } else 3802 SOCK_UNLOCK(so); 3803 wakeup(&so->so_timeo); 3804 } 3805 3806 /* 3807 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 3808 */ 3809 struct sockaddr * 3810 sodupsockaddr(const struct sockaddr *sa, int mflags) 3811 { 3812 struct sockaddr *sa2; 3813 3814 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 3815 if (sa2) 3816 bcopy(sa, sa2, sa->sa_len); 3817 return sa2; 3818 } 3819 3820 /* 3821 * Register per-socket destructor. 3822 */ 3823 void 3824 sodtor_set(struct socket *so, so_dtor_t *func) 3825 { 3826 3827 SOCK_LOCK_ASSERT(so); 3828 so->so_dtor = func; 3829 } 3830 3831 /* 3832 * Register per-socket buffer upcalls. 3833 */ 3834 void 3835 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg) 3836 { 3837 struct sockbuf *sb; 3838 3839 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 3840 3841 switch (which) { 3842 case SO_RCV: 3843 sb = &so->so_rcv; 3844 break; 3845 case SO_SND: 3846 sb = &so->so_snd; 3847 break; 3848 default: 3849 panic("soupcall_set: bad which"); 3850 } 3851 SOCKBUF_LOCK_ASSERT(sb); 3852 sb->sb_upcall = func; 3853 sb->sb_upcallarg = arg; 3854 sb->sb_flags |= SB_UPCALL; 3855 } 3856 3857 void 3858 soupcall_clear(struct socket *so, int which) 3859 { 3860 struct sockbuf *sb; 3861 3862 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 3863 3864 switch (which) { 3865 case SO_RCV: 3866 sb = &so->so_rcv; 3867 break; 3868 case SO_SND: 3869 sb = &so->so_snd; 3870 break; 3871 default: 3872 panic("soupcall_clear: bad which"); 3873 } 3874 SOCKBUF_LOCK_ASSERT(sb); 3875 KASSERT(sb->sb_upcall != NULL, 3876 ("%s: so %p no upcall to clear", __func__, so)); 3877 sb->sb_upcall = NULL; 3878 sb->sb_upcallarg = NULL; 3879 sb->sb_flags &= ~SB_UPCALL; 3880 } 3881 3882 void 3883 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg) 3884 { 3885 3886 SOLISTEN_LOCK_ASSERT(so); 3887 so->sol_upcall = func; 3888 so->sol_upcallarg = arg; 3889 } 3890 3891 static void 3892 so_rdknl_lock(void *arg) 3893 { 3894 struct socket *so = arg; 3895 3896 if (SOLISTENING(so)) 3897 SOCK_LOCK(so); 3898 else 3899 SOCKBUF_LOCK(&so->so_rcv); 3900 } 3901 3902 static void 3903 so_rdknl_unlock(void *arg) 3904 { 3905 struct socket *so = arg; 3906 3907 if (SOLISTENING(so)) 3908 SOCK_UNLOCK(so); 3909 else 3910 SOCKBUF_UNLOCK(&so->so_rcv); 3911 } 3912 3913 static void 3914 so_rdknl_assert_locked(void *arg) 3915 { 3916 struct socket *so = arg; 3917 3918 if (SOLISTENING(so)) 3919 SOCK_LOCK_ASSERT(so); 3920 else 3921 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3922 } 3923 3924 static void 3925 so_rdknl_assert_unlocked(void *arg) 3926 { 3927 struct socket *so = arg; 3928 3929 if (SOLISTENING(so)) 3930 SOCK_UNLOCK_ASSERT(so); 3931 else 3932 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); 3933 } 3934 3935 static void 3936 so_wrknl_lock(void *arg) 3937 { 3938 struct socket *so = arg; 3939 3940 if (SOLISTENING(so)) 3941 SOCK_LOCK(so); 3942 else 3943 SOCKBUF_LOCK(&so->so_snd); 3944 } 3945 3946 static void 3947 so_wrknl_unlock(void *arg) 3948 { 3949 struct socket *so = arg; 3950 3951 if (SOLISTENING(so)) 3952 SOCK_UNLOCK(so); 3953 else 3954 SOCKBUF_UNLOCK(&so->so_snd); 3955 } 3956 3957 static void 3958 so_wrknl_assert_locked(void *arg) 3959 { 3960 struct socket *so = arg; 3961 3962 if (SOLISTENING(so)) 3963 SOCK_LOCK_ASSERT(so); 3964 else 3965 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3966 } 3967 3968 static void 3969 so_wrknl_assert_unlocked(void *arg) 3970 { 3971 struct socket *so = arg; 3972 3973 if (SOLISTENING(so)) 3974 SOCK_UNLOCK_ASSERT(so); 3975 else 3976 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 3977 } 3978 3979 /* 3980 * Create an external-format (``xsocket'') structure using the information in 3981 * the kernel-format socket structure pointed to by so. This is done to 3982 * reduce the spew of irrelevant information over this interface, to isolate 3983 * user code from changes in the kernel structure, and potentially to provide 3984 * information-hiding if we decide that some of this information should be 3985 * hidden from users. 3986 */ 3987 void 3988 sotoxsocket(struct socket *so, struct xsocket *xso) 3989 { 3990 3991 xso->xso_len = sizeof *xso; 3992 xso->xso_so = (uintptr_t)so; 3993 xso->so_type = so->so_type; 3994 xso->so_options = so->so_options; 3995 xso->so_linger = so->so_linger; 3996 xso->so_state = so->so_state; 3997 xso->so_pcb = (uintptr_t)so->so_pcb; 3998 xso->xso_protocol = so->so_proto->pr_protocol; 3999 xso->xso_family = so->so_proto->pr_domain->dom_family; 4000 xso->so_timeo = so->so_timeo; 4001 xso->so_error = so->so_error; 4002 xso->so_uid = so->so_cred->cr_uid; 4003 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 4004 if (SOLISTENING(so)) { 4005 xso->so_qlen = so->sol_qlen; 4006 xso->so_incqlen = so->sol_incqlen; 4007 xso->so_qlimit = so->sol_qlimit; 4008 xso->so_oobmark = 0; 4009 bzero(&xso->so_snd, sizeof(xso->so_snd)); 4010 bzero(&xso->so_rcv, sizeof(xso->so_rcv)); 4011 } else { 4012 xso->so_state |= so->so_qstate; 4013 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0; 4014 xso->so_oobmark = so->so_oobmark; 4015 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 4016 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 4017 } 4018 } 4019 4020 struct sockbuf * 4021 so_sockbuf_rcv(struct socket *so) 4022 { 4023 4024 return (&so->so_rcv); 4025 } 4026 4027 struct sockbuf * 4028 so_sockbuf_snd(struct socket *so) 4029 { 4030 4031 return (&so->so_snd); 4032 } 4033 4034 int 4035 so_state_get(const struct socket *so) 4036 { 4037 4038 return (so->so_state); 4039 } 4040 4041 void 4042 so_state_set(struct socket *so, int val) 4043 { 4044 4045 so->so_state = val; 4046 } 4047 4048 int 4049 so_options_get(const struct socket *so) 4050 { 4051 4052 return (so->so_options); 4053 } 4054 4055 void 4056 so_options_set(struct socket *so, int val) 4057 { 4058 4059 so->so_options = val; 4060 } 4061 4062 int 4063 so_error_get(const struct socket *so) 4064 { 4065 4066 return (so->so_error); 4067 } 4068 4069 void 4070 so_error_set(struct socket *so, int val) 4071 { 4072 4073 so->so_error = val; 4074 } 4075 4076 int 4077 so_linger_get(const struct socket *so) 4078 { 4079 4080 return (so->so_linger); 4081 } 4082 4083 void 4084 so_linger_set(struct socket *so, int val) 4085 { 4086 4087 so->so_linger = val; 4088 } 4089 4090 struct protosw * 4091 so_protosw_get(const struct socket *so) 4092 { 4093 4094 return (so->so_proto); 4095 } 4096 4097 void 4098 so_protosw_set(struct socket *so, struct protosw *val) 4099 { 4100 4101 so->so_proto = val; 4102 } 4103 4104 void 4105 so_sorwakeup(struct socket *so) 4106 { 4107 4108 sorwakeup(so); 4109 } 4110 4111 void 4112 so_sowwakeup(struct socket *so) 4113 { 4114 4115 sowwakeup(so); 4116 } 4117 4118 void 4119 so_sorwakeup_locked(struct socket *so) 4120 { 4121 4122 sorwakeup_locked(so); 4123 } 4124 4125 void 4126 so_sowwakeup_locked(struct socket *so) 4127 { 4128 4129 sowwakeup_locked(so); 4130 } 4131 4132 void 4133 so_lock(struct socket *so) 4134 { 4135 4136 SOCK_LOCK(so); 4137 } 4138 4139 void 4140 so_unlock(struct socket *so) 4141 { 4142 4143 SOCK_UNLOCK(so); 4144 } 4145