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 1105 SOCK_LOCK(so); 1106 if ((listening = (so->so_options & SO_ACCEPTCONN))) { 1107 struct socket *sp; 1108 1109 TAILQ_INIT(&lqueue); 1110 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list); 1111 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list); 1112 1113 so->sol_qlen = so->sol_incqlen = 0; 1114 1115 TAILQ_FOREACH(sp, &lqueue, so_list) { 1116 SOCK_LOCK(sp); 1117 sp->so_qstate = SQ_NONE; 1118 sp->so_listen = NULL; 1119 SOCK_UNLOCK(sp); 1120 /* Guaranteed not to be the last. */ 1121 refcount_release(&so->so_count); 1122 } 1123 } 1124 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF")); 1125 so->so_state |= SS_NOFDREF; 1126 sorele(so); 1127 if (listening) { 1128 struct socket *sp; 1129 1130 TAILQ_FOREACH(sp, &lqueue, so_list) { 1131 SOCK_LOCK(sp); 1132 if (sp->so_count == 0) { 1133 SOCK_UNLOCK(sp); 1134 soabort(sp); 1135 } else 1136 /* sp is now in sofree() */ 1137 SOCK_UNLOCK(sp); 1138 } 1139 } 1140 CURVNET_RESTORE(); 1141 return (error); 1142 } 1143 1144 /* 1145 * soabort() is used to abruptly tear down a connection, such as when a 1146 * resource limit is reached (listen queue depth exceeded), or if a listen 1147 * socket is closed while there are sockets waiting to be accepted. 1148 * 1149 * This interface is tricky, because it is called on an unreferenced socket, 1150 * and must be called only by a thread that has actually removed the socket 1151 * from the listen queue it was on, or races with other threads are risked. 1152 * 1153 * This interface will call into the protocol code, so must not be called 1154 * with any socket locks held. Protocols do call it while holding their own 1155 * recursible protocol mutexes, but this is something that should be subject 1156 * to review in the future. 1157 */ 1158 void 1159 soabort(struct socket *so) 1160 { 1161 1162 /* 1163 * In as much as is possible, assert that no references to this 1164 * socket are held. This is not quite the same as asserting that the 1165 * current thread is responsible for arranging for no references, but 1166 * is as close as we can get for now. 1167 */ 1168 KASSERT(so->so_count == 0, ("soabort: so_count")); 1169 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF")); 1170 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF")); 1171 KASSERT(so->so_qstate == SQ_NONE, ("soabort: !SQ_NONE")); 1172 VNET_SO_ASSERT(so); 1173 1174 if (so->so_proto->pr_usrreqs->pru_abort != NULL) 1175 (*so->so_proto->pr_usrreqs->pru_abort)(so); 1176 SOCK_LOCK(so); 1177 sofree(so); 1178 } 1179 1180 int 1181 soaccept(struct socket *so, struct sockaddr **nam) 1182 { 1183 int error; 1184 1185 SOCK_LOCK(so); 1186 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF")); 1187 so->so_state &= ~SS_NOFDREF; 1188 SOCK_UNLOCK(so); 1189 1190 CURVNET_SET(so->so_vnet); 1191 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); 1192 CURVNET_RESTORE(); 1193 return (error); 1194 } 1195 1196 int 1197 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 1198 { 1199 1200 return (soconnectat(AT_FDCWD, so, nam, td)); 1201 } 1202 1203 int 1204 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 1205 { 1206 int error; 1207 1208 if (so->so_options & SO_ACCEPTCONN) 1209 return (EOPNOTSUPP); 1210 1211 CURVNET_SET(so->so_vnet); 1212 /* 1213 * If protocol is connection-based, can only connect once. 1214 * Otherwise, if connected, try to disconnect first. This allows 1215 * user to disconnect by connecting to, e.g., a null address. 1216 */ 1217 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 1218 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 1219 (error = sodisconnect(so)))) { 1220 error = EISCONN; 1221 } else { 1222 /* 1223 * Prevent accumulated error from previous connection from 1224 * biting us. 1225 */ 1226 so->so_error = 0; 1227 if (fd == AT_FDCWD) { 1228 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, 1229 nam, td); 1230 } else { 1231 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd, 1232 so, nam, td); 1233 } 1234 } 1235 CURVNET_RESTORE(); 1236 1237 return (error); 1238 } 1239 1240 int 1241 soconnect2(struct socket *so1, struct socket *so2) 1242 { 1243 int error; 1244 1245 CURVNET_SET(so1->so_vnet); 1246 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); 1247 CURVNET_RESTORE(); 1248 return (error); 1249 } 1250 1251 int 1252 sodisconnect(struct socket *so) 1253 { 1254 int error; 1255 1256 if ((so->so_state & SS_ISCONNECTED) == 0) 1257 return (ENOTCONN); 1258 if (so->so_state & SS_ISDISCONNECTING) 1259 return (EALREADY); 1260 VNET_SO_ASSERT(so); 1261 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 1262 return (error); 1263 } 1264 1265 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) 1266 1267 int 1268 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1269 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1270 { 1271 long space; 1272 ssize_t resid; 1273 int clen = 0, error, dontroute; 1274 1275 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 1276 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 1277 ("sosend_dgram: !PR_ATOMIC")); 1278 1279 if (uio != NULL) 1280 resid = uio->uio_resid; 1281 else 1282 resid = top->m_pkthdr.len; 1283 /* 1284 * In theory resid should be unsigned. However, space must be 1285 * signed, as it might be less than 0 if we over-committed, and we 1286 * must use a signed comparison of space and resid. On the other 1287 * hand, a negative resid causes us to loop sending 0-length 1288 * segments to the protocol. 1289 */ 1290 if (resid < 0) { 1291 error = EINVAL; 1292 goto out; 1293 } 1294 1295 dontroute = 1296 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 1297 if (td != NULL) 1298 td->td_ru.ru_msgsnd++; 1299 if (control != NULL) 1300 clen = control->m_len; 1301 1302 SOCKBUF_LOCK(&so->so_snd); 1303 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1304 SOCKBUF_UNLOCK(&so->so_snd); 1305 error = EPIPE; 1306 goto out; 1307 } 1308 if (so->so_error) { 1309 error = so->so_error; 1310 so->so_error = 0; 1311 SOCKBUF_UNLOCK(&so->so_snd); 1312 goto out; 1313 } 1314 if ((so->so_state & SS_ISCONNECTED) == 0) { 1315 /* 1316 * `sendto' and `sendmsg' is allowed on a connection-based 1317 * socket if it supports implied connect. Return ENOTCONN if 1318 * not connected and no address is supplied. 1319 */ 1320 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1321 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1322 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1323 !(resid == 0 && clen != 0)) { 1324 SOCKBUF_UNLOCK(&so->so_snd); 1325 error = ENOTCONN; 1326 goto out; 1327 } 1328 } else if (addr == NULL) { 1329 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1330 error = ENOTCONN; 1331 else 1332 error = EDESTADDRREQ; 1333 SOCKBUF_UNLOCK(&so->so_snd); 1334 goto out; 1335 } 1336 } 1337 1338 /* 1339 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1340 * problem and need fixing. 1341 */ 1342 space = sbspace(&so->so_snd); 1343 if (flags & MSG_OOB) 1344 space += 1024; 1345 space -= clen; 1346 SOCKBUF_UNLOCK(&so->so_snd); 1347 if (resid > space) { 1348 error = EMSGSIZE; 1349 goto out; 1350 } 1351 if (uio == NULL) { 1352 resid = 0; 1353 if (flags & MSG_EOR) 1354 top->m_flags |= M_EOR; 1355 } else { 1356 /* 1357 * Copy the data from userland into a mbuf chain. 1358 * If no data is to be copied in, a single empty mbuf 1359 * is returned. 1360 */ 1361 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1362 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1363 if (top == NULL) { 1364 error = EFAULT; /* only possible error */ 1365 goto out; 1366 } 1367 space -= resid - uio->uio_resid; 1368 resid = uio->uio_resid; 1369 } 1370 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1371 /* 1372 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1373 * than with. 1374 */ 1375 if (dontroute) { 1376 SOCK_LOCK(so); 1377 so->so_options |= SO_DONTROUTE; 1378 SOCK_UNLOCK(so); 1379 } 1380 /* 1381 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1382 * of date. We could have received a reset packet in an interrupt or 1383 * maybe we slept while doing page faults in uiomove() etc. We could 1384 * probably recheck again inside the locking protection here, but 1385 * there are probably other places that this also happens. We must 1386 * rethink this. 1387 */ 1388 VNET_SO_ASSERT(so); 1389 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1390 (flags & MSG_OOB) ? PRUS_OOB : 1391 /* 1392 * If the user set MSG_EOF, the protocol understands this flag and 1393 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1394 */ 1395 ((flags & MSG_EOF) && 1396 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1397 (resid <= 0)) ? 1398 PRUS_EOF : 1399 /* If there is more to send set PRUS_MORETOCOME */ 1400 (flags & MSG_MORETOCOME) || 1401 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1402 top, addr, control, td); 1403 if (dontroute) { 1404 SOCK_LOCK(so); 1405 so->so_options &= ~SO_DONTROUTE; 1406 SOCK_UNLOCK(so); 1407 } 1408 clen = 0; 1409 control = NULL; 1410 top = NULL; 1411 out: 1412 if (top != NULL) 1413 m_freem(top); 1414 if (control != NULL) 1415 m_freem(control); 1416 return (error); 1417 } 1418 1419 /* 1420 * Send on a socket. If send must go all at once and message is larger than 1421 * send buffering, then hard error. Lock against other senders. If must go 1422 * all at once and not enough room now, then inform user that this would 1423 * block and do nothing. Otherwise, if nonblocking, send as much as 1424 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1425 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1426 * in mbuf chain must be small enough to send all at once. 1427 * 1428 * Returns nonzero on error, timeout or signal; callers must check for short 1429 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1430 * on return. 1431 */ 1432 int 1433 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1434 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1435 { 1436 long space; 1437 ssize_t resid; 1438 int clen = 0, error, dontroute; 1439 int atomic = sosendallatonce(so) || top; 1440 1441 if (uio != NULL) 1442 resid = uio->uio_resid; 1443 else 1444 resid = top->m_pkthdr.len; 1445 /* 1446 * In theory resid should be unsigned. However, space must be 1447 * signed, as it might be less than 0 if we over-committed, and we 1448 * must use a signed comparison of space and resid. On the other 1449 * hand, a negative resid causes us to loop sending 0-length 1450 * segments to the protocol. 1451 * 1452 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1453 * type sockets since that's an error. 1454 */ 1455 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1456 error = EINVAL; 1457 goto out; 1458 } 1459 1460 dontroute = 1461 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1462 (so->so_proto->pr_flags & PR_ATOMIC); 1463 if (td != NULL) 1464 td->td_ru.ru_msgsnd++; 1465 if (control != NULL) 1466 clen = control->m_len; 1467 1468 error = sblock(&so->so_snd, SBLOCKWAIT(flags)); 1469 if (error) 1470 goto out; 1471 1472 restart: 1473 do { 1474 SOCKBUF_LOCK(&so->so_snd); 1475 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1476 SOCKBUF_UNLOCK(&so->so_snd); 1477 error = EPIPE; 1478 goto release; 1479 } 1480 if (so->so_error) { 1481 error = so->so_error; 1482 so->so_error = 0; 1483 SOCKBUF_UNLOCK(&so->so_snd); 1484 goto release; 1485 } 1486 if ((so->so_state & SS_ISCONNECTED) == 0) { 1487 /* 1488 * `sendto' and `sendmsg' is allowed on a connection- 1489 * based socket if it supports implied connect. 1490 * Return ENOTCONN if not connected and no address is 1491 * supplied. 1492 */ 1493 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1494 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1495 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1496 !(resid == 0 && clen != 0)) { 1497 SOCKBUF_UNLOCK(&so->so_snd); 1498 error = ENOTCONN; 1499 goto release; 1500 } 1501 } else if (addr == NULL) { 1502 SOCKBUF_UNLOCK(&so->so_snd); 1503 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1504 error = ENOTCONN; 1505 else 1506 error = EDESTADDRREQ; 1507 goto release; 1508 } 1509 } 1510 space = sbspace(&so->so_snd); 1511 if (flags & MSG_OOB) 1512 space += 1024; 1513 if ((atomic && resid > so->so_snd.sb_hiwat) || 1514 clen > so->so_snd.sb_hiwat) { 1515 SOCKBUF_UNLOCK(&so->so_snd); 1516 error = EMSGSIZE; 1517 goto release; 1518 } 1519 if (space < resid + clen && 1520 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1521 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) { 1522 SOCKBUF_UNLOCK(&so->so_snd); 1523 error = EWOULDBLOCK; 1524 goto release; 1525 } 1526 error = sbwait(&so->so_snd); 1527 SOCKBUF_UNLOCK(&so->so_snd); 1528 if (error) 1529 goto release; 1530 goto restart; 1531 } 1532 SOCKBUF_UNLOCK(&so->so_snd); 1533 space -= clen; 1534 do { 1535 if (uio == NULL) { 1536 resid = 0; 1537 if (flags & MSG_EOR) 1538 top->m_flags |= M_EOR; 1539 } else { 1540 /* 1541 * Copy the data from userland into a mbuf 1542 * chain. If resid is 0, which can happen 1543 * only if we have control to send, then 1544 * a single empty mbuf is returned. This 1545 * is a workaround to prevent protocol send 1546 * methods to panic. 1547 */ 1548 top = m_uiotombuf(uio, M_WAITOK, space, 1549 (atomic ? max_hdr : 0), 1550 (atomic ? M_PKTHDR : 0) | 1551 ((flags & MSG_EOR) ? M_EOR : 0)); 1552 if (top == NULL) { 1553 error = EFAULT; /* only possible error */ 1554 goto release; 1555 } 1556 space -= resid - uio->uio_resid; 1557 resid = uio->uio_resid; 1558 } 1559 if (dontroute) { 1560 SOCK_LOCK(so); 1561 so->so_options |= SO_DONTROUTE; 1562 SOCK_UNLOCK(so); 1563 } 1564 /* 1565 * XXX all the SBS_CANTSENDMORE checks previously 1566 * done could be out of date. We could have received 1567 * a reset packet in an interrupt or maybe we slept 1568 * while doing page faults in uiomove() etc. We 1569 * could probably recheck again inside the locking 1570 * protection here, but there are probably other 1571 * places that this also happens. We must rethink 1572 * this. 1573 */ 1574 VNET_SO_ASSERT(so); 1575 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1576 (flags & MSG_OOB) ? PRUS_OOB : 1577 /* 1578 * If the user set MSG_EOF, the protocol understands 1579 * this flag and nothing left to send then use 1580 * PRU_SEND_EOF instead of PRU_SEND. 1581 */ 1582 ((flags & MSG_EOF) && 1583 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1584 (resid <= 0)) ? 1585 PRUS_EOF : 1586 /* If there is more to send set PRUS_MORETOCOME. */ 1587 (flags & MSG_MORETOCOME) || 1588 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1589 top, addr, control, td); 1590 if (dontroute) { 1591 SOCK_LOCK(so); 1592 so->so_options &= ~SO_DONTROUTE; 1593 SOCK_UNLOCK(so); 1594 } 1595 clen = 0; 1596 control = NULL; 1597 top = NULL; 1598 if (error) 1599 goto release; 1600 } while (resid && space > 0); 1601 } while (resid); 1602 1603 release: 1604 sbunlock(&so->so_snd); 1605 out: 1606 if (top != NULL) 1607 m_freem(top); 1608 if (control != NULL) 1609 m_freem(control); 1610 return (error); 1611 } 1612 1613 int 1614 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1615 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1616 { 1617 int error; 1618 1619 CURVNET_SET(so->so_vnet); 1620 if (!SOLISTENING(so)) 1621 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, 1622 top, control, flags, td); 1623 else { 1624 m_freem(top); 1625 m_freem(control); 1626 error = ENOTCONN; 1627 } 1628 CURVNET_RESTORE(); 1629 return (error); 1630 } 1631 1632 /* 1633 * The part of soreceive() that implements reading non-inline out-of-band 1634 * data from a socket. For more complete comments, see soreceive(), from 1635 * which this code originated. 1636 * 1637 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1638 * unable to return an mbuf chain to the caller. 1639 */ 1640 static int 1641 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1642 { 1643 struct protosw *pr = so->so_proto; 1644 struct mbuf *m; 1645 int error; 1646 1647 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1648 VNET_SO_ASSERT(so); 1649 1650 m = m_get(M_WAITOK, MT_DATA); 1651 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 1652 if (error) 1653 goto bad; 1654 do { 1655 error = uiomove(mtod(m, void *), 1656 (int) min(uio->uio_resid, m->m_len), uio); 1657 m = m_free(m); 1658 } while (uio->uio_resid && error == 0 && m); 1659 bad: 1660 if (m != NULL) 1661 m_freem(m); 1662 return (error); 1663 } 1664 1665 /* 1666 * Following replacement or removal of the first mbuf on the first mbuf chain 1667 * of a socket buffer, push necessary state changes back into the socket 1668 * buffer so that other consumers see the values consistently. 'nextrecord' 1669 * is the callers locally stored value of the original value of 1670 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1671 * NOTE: 'nextrecord' may be NULL. 1672 */ 1673 static __inline void 1674 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1675 { 1676 1677 SOCKBUF_LOCK_ASSERT(sb); 1678 /* 1679 * First, update for the new value of nextrecord. If necessary, make 1680 * it the first record. 1681 */ 1682 if (sb->sb_mb != NULL) 1683 sb->sb_mb->m_nextpkt = nextrecord; 1684 else 1685 sb->sb_mb = nextrecord; 1686 1687 /* 1688 * Now update any dependent socket buffer fields to reflect the new 1689 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1690 * addition of a second clause that takes care of the case where 1691 * sb_mb has been updated, but remains the last record. 1692 */ 1693 if (sb->sb_mb == NULL) { 1694 sb->sb_mbtail = NULL; 1695 sb->sb_lastrecord = NULL; 1696 } else if (sb->sb_mb->m_nextpkt == NULL) 1697 sb->sb_lastrecord = sb->sb_mb; 1698 } 1699 1700 /* 1701 * Implement receive operations on a socket. We depend on the way that 1702 * records are added to the sockbuf by sbappend. In particular, each record 1703 * (mbufs linked through m_next) must begin with an address if the protocol 1704 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1705 * data, and then zero or more mbufs of data. In order to allow parallelism 1706 * between network receive and copying to user space, as well as avoid 1707 * sleeping with a mutex held, we release the socket buffer mutex during the 1708 * user space copy. Although the sockbuf is locked, new data may still be 1709 * appended, and thus we must maintain consistency of the sockbuf during that 1710 * time. 1711 * 1712 * The caller may receive the data as a single mbuf chain by supplying an 1713 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1714 * the count in uio_resid. 1715 */ 1716 int 1717 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1718 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1719 { 1720 struct mbuf *m, **mp; 1721 int flags, error, offset; 1722 ssize_t len; 1723 struct protosw *pr = so->so_proto; 1724 struct mbuf *nextrecord; 1725 int moff, type = 0; 1726 ssize_t orig_resid = uio->uio_resid; 1727 1728 mp = mp0; 1729 if (psa != NULL) 1730 *psa = NULL; 1731 if (controlp != NULL) 1732 *controlp = NULL; 1733 if (flagsp != NULL) 1734 flags = *flagsp &~ MSG_EOR; 1735 else 1736 flags = 0; 1737 if (flags & MSG_OOB) 1738 return (soreceive_rcvoob(so, uio, flags)); 1739 if (mp != NULL) 1740 *mp = NULL; 1741 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 1742 && uio->uio_resid) { 1743 VNET_SO_ASSERT(so); 1744 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 1745 } 1746 1747 error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); 1748 if (error) 1749 return (error); 1750 1751 restart: 1752 SOCKBUF_LOCK(&so->so_rcv); 1753 m = so->so_rcv.sb_mb; 1754 /* 1755 * If we have less data than requested, block awaiting more (subject 1756 * to any timeout) if: 1757 * 1. the current count is less than the low water mark, or 1758 * 2. MSG_DONTWAIT is not set 1759 */ 1760 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1761 sbavail(&so->so_rcv) < uio->uio_resid) && 1762 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && 1763 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 1764 KASSERT(m != NULL || !sbavail(&so->so_rcv), 1765 ("receive: m == %p sbavail == %u", 1766 m, sbavail(&so->so_rcv))); 1767 if (so->so_error) { 1768 if (m != NULL) 1769 goto dontblock; 1770 error = so->so_error; 1771 if ((flags & MSG_PEEK) == 0) 1772 so->so_error = 0; 1773 SOCKBUF_UNLOCK(&so->so_rcv); 1774 goto release; 1775 } 1776 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1777 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1778 if (m == NULL) { 1779 SOCKBUF_UNLOCK(&so->so_rcv); 1780 goto release; 1781 } else 1782 goto dontblock; 1783 } 1784 for (; m != NULL; m = m->m_next) 1785 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1786 m = so->so_rcv.sb_mb; 1787 goto dontblock; 1788 } 1789 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1790 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1791 SOCKBUF_UNLOCK(&so->so_rcv); 1792 error = ENOTCONN; 1793 goto release; 1794 } 1795 if (uio->uio_resid == 0) { 1796 SOCKBUF_UNLOCK(&so->so_rcv); 1797 goto release; 1798 } 1799 if ((so->so_state & SS_NBIO) || 1800 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1801 SOCKBUF_UNLOCK(&so->so_rcv); 1802 error = EWOULDBLOCK; 1803 goto release; 1804 } 1805 SBLASTRECORDCHK(&so->so_rcv); 1806 SBLASTMBUFCHK(&so->so_rcv); 1807 error = sbwait(&so->so_rcv); 1808 SOCKBUF_UNLOCK(&so->so_rcv); 1809 if (error) 1810 goto release; 1811 goto restart; 1812 } 1813 dontblock: 1814 /* 1815 * From this point onward, we maintain 'nextrecord' as a cache of the 1816 * pointer to the next record in the socket buffer. We must keep the 1817 * various socket buffer pointers and local stack versions of the 1818 * pointers in sync, pushing out modifications before dropping the 1819 * socket buffer mutex, and re-reading them when picking it up. 1820 * 1821 * Otherwise, we will race with the network stack appending new data 1822 * or records onto the socket buffer by using inconsistent/stale 1823 * versions of the field, possibly resulting in socket buffer 1824 * corruption. 1825 * 1826 * By holding the high-level sblock(), we prevent simultaneous 1827 * readers from pulling off the front of the socket buffer. 1828 */ 1829 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1830 if (uio->uio_td) 1831 uio->uio_td->td_ru.ru_msgrcv++; 1832 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 1833 SBLASTRECORDCHK(&so->so_rcv); 1834 SBLASTMBUFCHK(&so->so_rcv); 1835 nextrecord = m->m_nextpkt; 1836 if (pr->pr_flags & PR_ADDR) { 1837 KASSERT(m->m_type == MT_SONAME, 1838 ("m->m_type == %d", m->m_type)); 1839 orig_resid = 0; 1840 if (psa != NULL) 1841 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 1842 M_NOWAIT); 1843 if (flags & MSG_PEEK) { 1844 m = m->m_next; 1845 } else { 1846 sbfree(&so->so_rcv, m); 1847 so->so_rcv.sb_mb = m_free(m); 1848 m = so->so_rcv.sb_mb; 1849 sockbuf_pushsync(&so->so_rcv, nextrecord); 1850 } 1851 } 1852 1853 /* 1854 * Process one or more MT_CONTROL mbufs present before any data mbufs 1855 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 1856 * just copy the data; if !MSG_PEEK, we call into the protocol to 1857 * perform externalization (or freeing if controlp == NULL). 1858 */ 1859 if (m != NULL && m->m_type == MT_CONTROL) { 1860 struct mbuf *cm = NULL, *cmn; 1861 struct mbuf **cme = &cm; 1862 1863 do { 1864 if (flags & MSG_PEEK) { 1865 if (controlp != NULL) { 1866 *controlp = m_copym(m, 0, m->m_len, 1867 M_NOWAIT); 1868 controlp = &(*controlp)->m_next; 1869 } 1870 m = m->m_next; 1871 } else { 1872 sbfree(&so->so_rcv, m); 1873 so->so_rcv.sb_mb = m->m_next; 1874 m->m_next = NULL; 1875 *cme = m; 1876 cme = &(*cme)->m_next; 1877 m = so->so_rcv.sb_mb; 1878 } 1879 } while (m != NULL && m->m_type == MT_CONTROL); 1880 if ((flags & MSG_PEEK) == 0) 1881 sockbuf_pushsync(&so->so_rcv, nextrecord); 1882 while (cm != NULL) { 1883 cmn = cm->m_next; 1884 cm->m_next = NULL; 1885 if (pr->pr_domain->dom_externalize != NULL) { 1886 SOCKBUF_UNLOCK(&so->so_rcv); 1887 VNET_SO_ASSERT(so); 1888 error = (*pr->pr_domain->dom_externalize) 1889 (cm, controlp, flags); 1890 SOCKBUF_LOCK(&so->so_rcv); 1891 } else if (controlp != NULL) 1892 *controlp = cm; 1893 else 1894 m_freem(cm); 1895 if (controlp != NULL) { 1896 orig_resid = 0; 1897 while (*controlp != NULL) 1898 controlp = &(*controlp)->m_next; 1899 } 1900 cm = cmn; 1901 } 1902 if (m != NULL) 1903 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 1904 else 1905 nextrecord = so->so_rcv.sb_mb; 1906 orig_resid = 0; 1907 } 1908 if (m != NULL) { 1909 if ((flags & MSG_PEEK) == 0) { 1910 KASSERT(m->m_nextpkt == nextrecord, 1911 ("soreceive: post-control, nextrecord !sync")); 1912 if (nextrecord == NULL) { 1913 KASSERT(so->so_rcv.sb_mb == m, 1914 ("soreceive: post-control, sb_mb!=m")); 1915 KASSERT(so->so_rcv.sb_lastrecord == m, 1916 ("soreceive: post-control, lastrecord!=m")); 1917 } 1918 } 1919 type = m->m_type; 1920 if (type == MT_OOBDATA) 1921 flags |= MSG_OOB; 1922 } else { 1923 if ((flags & MSG_PEEK) == 0) { 1924 KASSERT(so->so_rcv.sb_mb == nextrecord, 1925 ("soreceive: sb_mb != nextrecord")); 1926 if (so->so_rcv.sb_mb == NULL) { 1927 KASSERT(so->so_rcv.sb_lastrecord == NULL, 1928 ("soreceive: sb_lastercord != NULL")); 1929 } 1930 } 1931 } 1932 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1933 SBLASTRECORDCHK(&so->so_rcv); 1934 SBLASTMBUFCHK(&so->so_rcv); 1935 1936 /* 1937 * Now continue to read any data mbufs off of the head of the socket 1938 * buffer until the read request is satisfied. Note that 'type' is 1939 * used to store the type of any mbuf reads that have happened so far 1940 * such that soreceive() can stop reading if the type changes, which 1941 * causes soreceive() to return only one of regular data and inline 1942 * out-of-band data in a single socket receive operation. 1943 */ 1944 moff = 0; 1945 offset = 0; 1946 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 1947 && error == 0) { 1948 /* 1949 * If the type of mbuf has changed since the last mbuf 1950 * examined ('type'), end the receive operation. 1951 */ 1952 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1953 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 1954 if (type != m->m_type) 1955 break; 1956 } else if (type == MT_OOBDATA) 1957 break; 1958 else 1959 KASSERT(m->m_type == MT_DATA, 1960 ("m->m_type == %d", m->m_type)); 1961 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 1962 len = uio->uio_resid; 1963 if (so->so_oobmark && len > so->so_oobmark - offset) 1964 len = so->so_oobmark - offset; 1965 if (len > m->m_len - moff) 1966 len = m->m_len - moff; 1967 /* 1968 * If mp is set, just pass back the mbufs. Otherwise copy 1969 * them out via the uio, then free. Sockbuf must be 1970 * consistent here (points to current mbuf, it points to next 1971 * record) when we drop priority; we must note any additions 1972 * to the sockbuf when we block interrupts again. 1973 */ 1974 if (mp == NULL) { 1975 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1976 SBLASTRECORDCHK(&so->so_rcv); 1977 SBLASTMBUFCHK(&so->so_rcv); 1978 SOCKBUF_UNLOCK(&so->so_rcv); 1979 error = uiomove(mtod(m, char *) + moff, (int)len, uio); 1980 SOCKBUF_LOCK(&so->so_rcv); 1981 if (error) { 1982 /* 1983 * The MT_SONAME mbuf has already been removed 1984 * from the record, so it is necessary to 1985 * remove the data mbufs, if any, to preserve 1986 * the invariant in the case of PR_ADDR that 1987 * requires MT_SONAME mbufs at the head of 1988 * each record. 1989 */ 1990 if (pr->pr_flags & PR_ATOMIC && 1991 ((flags & MSG_PEEK) == 0)) 1992 (void)sbdroprecord_locked(&so->so_rcv); 1993 SOCKBUF_UNLOCK(&so->so_rcv); 1994 goto release; 1995 } 1996 } else 1997 uio->uio_resid -= len; 1998 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1999 if (len == m->m_len - moff) { 2000 if (m->m_flags & M_EOR) 2001 flags |= MSG_EOR; 2002 if (flags & MSG_PEEK) { 2003 m = m->m_next; 2004 moff = 0; 2005 } else { 2006 nextrecord = m->m_nextpkt; 2007 sbfree(&so->so_rcv, m); 2008 if (mp != NULL) { 2009 m->m_nextpkt = NULL; 2010 *mp = m; 2011 mp = &m->m_next; 2012 so->so_rcv.sb_mb = m = m->m_next; 2013 *mp = NULL; 2014 } else { 2015 so->so_rcv.sb_mb = m_free(m); 2016 m = so->so_rcv.sb_mb; 2017 } 2018 sockbuf_pushsync(&so->so_rcv, nextrecord); 2019 SBLASTRECORDCHK(&so->so_rcv); 2020 SBLASTMBUFCHK(&so->so_rcv); 2021 } 2022 } else { 2023 if (flags & MSG_PEEK) 2024 moff += len; 2025 else { 2026 if (mp != NULL) { 2027 if (flags & MSG_DONTWAIT) { 2028 *mp = m_copym(m, 0, len, 2029 M_NOWAIT); 2030 if (*mp == NULL) { 2031 /* 2032 * m_copym() couldn't 2033 * allocate an mbuf. 2034 * Adjust uio_resid back 2035 * (it was adjusted 2036 * down by len bytes, 2037 * which we didn't end 2038 * up "copying" over). 2039 */ 2040 uio->uio_resid += len; 2041 break; 2042 } 2043 } else { 2044 SOCKBUF_UNLOCK(&so->so_rcv); 2045 *mp = m_copym(m, 0, len, 2046 M_WAITOK); 2047 SOCKBUF_LOCK(&so->so_rcv); 2048 } 2049 } 2050 sbcut_locked(&so->so_rcv, len); 2051 } 2052 } 2053 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2054 if (so->so_oobmark) { 2055 if ((flags & MSG_PEEK) == 0) { 2056 so->so_oobmark -= len; 2057 if (so->so_oobmark == 0) { 2058 so->so_rcv.sb_state |= SBS_RCVATMARK; 2059 break; 2060 } 2061 } else { 2062 offset += len; 2063 if (offset == so->so_oobmark) 2064 break; 2065 } 2066 } 2067 if (flags & MSG_EOR) 2068 break; 2069 /* 2070 * If the MSG_WAITALL flag is set (for non-atomic socket), we 2071 * must not quit until "uio->uio_resid == 0" or an error 2072 * termination. If a signal/timeout occurs, return with a 2073 * short count but without error. Keep sockbuf locked 2074 * against other readers. 2075 */ 2076 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 2077 !sosendallatonce(so) && nextrecord == NULL) { 2078 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2079 if (so->so_error || 2080 so->so_rcv.sb_state & SBS_CANTRCVMORE) 2081 break; 2082 /* 2083 * Notify the protocol that some data has been 2084 * drained before blocking. 2085 */ 2086 if (pr->pr_flags & PR_WANTRCVD) { 2087 SOCKBUF_UNLOCK(&so->so_rcv); 2088 VNET_SO_ASSERT(so); 2089 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2090 SOCKBUF_LOCK(&so->so_rcv); 2091 } 2092 SBLASTRECORDCHK(&so->so_rcv); 2093 SBLASTMBUFCHK(&so->so_rcv); 2094 /* 2095 * We could receive some data while was notifying 2096 * the protocol. Skip blocking in this case. 2097 */ 2098 if (so->so_rcv.sb_mb == NULL) { 2099 error = sbwait(&so->so_rcv); 2100 if (error) { 2101 SOCKBUF_UNLOCK(&so->so_rcv); 2102 goto release; 2103 } 2104 } 2105 m = so->so_rcv.sb_mb; 2106 if (m != NULL) 2107 nextrecord = m->m_nextpkt; 2108 } 2109 } 2110 2111 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2112 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 2113 flags |= MSG_TRUNC; 2114 if ((flags & MSG_PEEK) == 0) 2115 (void) sbdroprecord_locked(&so->so_rcv); 2116 } 2117 if ((flags & MSG_PEEK) == 0) { 2118 if (m == NULL) { 2119 /* 2120 * First part is an inline SB_EMPTY_FIXUP(). Second 2121 * part makes sure sb_lastrecord is up-to-date if 2122 * there is still data in the socket buffer. 2123 */ 2124 so->so_rcv.sb_mb = nextrecord; 2125 if (so->so_rcv.sb_mb == NULL) { 2126 so->so_rcv.sb_mbtail = NULL; 2127 so->so_rcv.sb_lastrecord = NULL; 2128 } else if (nextrecord->m_nextpkt == NULL) 2129 so->so_rcv.sb_lastrecord = nextrecord; 2130 } 2131 SBLASTRECORDCHK(&so->so_rcv); 2132 SBLASTMBUFCHK(&so->so_rcv); 2133 /* 2134 * If soreceive() is being done from the socket callback, 2135 * then don't need to generate ACK to peer to update window, 2136 * since ACK will be generated on return to TCP. 2137 */ 2138 if (!(flags & MSG_SOCALLBCK) && 2139 (pr->pr_flags & PR_WANTRCVD)) { 2140 SOCKBUF_UNLOCK(&so->so_rcv); 2141 VNET_SO_ASSERT(so); 2142 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2143 SOCKBUF_LOCK(&so->so_rcv); 2144 } 2145 } 2146 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2147 if (orig_resid == uio->uio_resid && orig_resid && 2148 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 2149 SOCKBUF_UNLOCK(&so->so_rcv); 2150 goto restart; 2151 } 2152 SOCKBUF_UNLOCK(&so->so_rcv); 2153 2154 if (flagsp != NULL) 2155 *flagsp |= flags; 2156 release: 2157 sbunlock(&so->so_rcv); 2158 return (error); 2159 } 2160 2161 /* 2162 * Optimized version of soreceive() for stream (TCP) sockets. 2163 * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled. 2164 */ 2165 int 2166 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 2167 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2168 { 2169 int len = 0, error = 0, flags, oresid; 2170 struct sockbuf *sb; 2171 struct mbuf *m, *n = NULL; 2172 2173 /* We only do stream sockets. */ 2174 if (so->so_type != SOCK_STREAM) 2175 return (EINVAL); 2176 if (psa != NULL) 2177 *psa = NULL; 2178 if (controlp != NULL) 2179 return (EINVAL); 2180 if (flagsp != NULL) 2181 flags = *flagsp &~ MSG_EOR; 2182 else 2183 flags = 0; 2184 if (flags & MSG_OOB) 2185 return (soreceive_rcvoob(so, uio, flags)); 2186 if (mp0 != NULL) 2187 *mp0 = NULL; 2188 2189 sb = &so->so_rcv; 2190 2191 /* Prevent other readers from entering the socket. */ 2192 error = sblock(sb, SBLOCKWAIT(flags)); 2193 if (error) 2194 goto out; 2195 SOCKBUF_LOCK(sb); 2196 2197 /* Easy one, no space to copyout anything. */ 2198 if (uio->uio_resid == 0) { 2199 error = EINVAL; 2200 goto out; 2201 } 2202 oresid = uio->uio_resid; 2203 2204 /* We will never ever get anything unless we are or were connected. */ 2205 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 2206 error = ENOTCONN; 2207 goto out; 2208 } 2209 2210 restart: 2211 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2212 2213 /* Abort if socket has reported problems. */ 2214 if (so->so_error) { 2215 if (sbavail(sb) > 0) 2216 goto deliver; 2217 if (oresid > uio->uio_resid) 2218 goto out; 2219 error = so->so_error; 2220 if (!(flags & MSG_PEEK)) 2221 so->so_error = 0; 2222 goto out; 2223 } 2224 2225 /* Door is closed. Deliver what is left, if any. */ 2226 if (sb->sb_state & SBS_CANTRCVMORE) { 2227 if (sbavail(sb) > 0) 2228 goto deliver; 2229 else 2230 goto out; 2231 } 2232 2233 /* Socket buffer is empty and we shall not block. */ 2234 if (sbavail(sb) == 0 && 2235 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2236 error = EAGAIN; 2237 goto out; 2238 } 2239 2240 /* Socket buffer got some data that we shall deliver now. */ 2241 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 2242 ((so->so_state & SS_NBIO) || 2243 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2244 sbavail(sb) >= sb->sb_lowat || 2245 sbavail(sb) >= uio->uio_resid || 2246 sbavail(sb) >= sb->sb_hiwat) ) { 2247 goto deliver; 2248 } 2249 2250 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2251 if ((flags & MSG_WAITALL) && 2252 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 2253 goto deliver; 2254 2255 /* 2256 * Wait and block until (more) data comes in. 2257 * NB: Drops the sockbuf lock during wait. 2258 */ 2259 error = sbwait(sb); 2260 if (error) 2261 goto out; 2262 goto restart; 2263 2264 deliver: 2265 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2266 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 2267 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2268 2269 /* Statistics. */ 2270 if (uio->uio_td) 2271 uio->uio_td->td_ru.ru_msgrcv++; 2272 2273 /* Fill uio until full or current end of socket buffer is reached. */ 2274 len = min(uio->uio_resid, sbavail(sb)); 2275 if (mp0 != NULL) { 2276 /* Dequeue as many mbufs as possible. */ 2277 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2278 if (*mp0 == NULL) 2279 *mp0 = sb->sb_mb; 2280 else 2281 m_cat(*mp0, sb->sb_mb); 2282 for (m = sb->sb_mb; 2283 m != NULL && m->m_len <= len; 2284 m = m->m_next) { 2285 KASSERT(!(m->m_flags & M_NOTAVAIL), 2286 ("%s: m %p not available", __func__, m)); 2287 len -= m->m_len; 2288 uio->uio_resid -= m->m_len; 2289 sbfree(sb, m); 2290 n = m; 2291 } 2292 n->m_next = NULL; 2293 sb->sb_mb = m; 2294 sb->sb_lastrecord = sb->sb_mb; 2295 if (sb->sb_mb == NULL) 2296 SB_EMPTY_FIXUP(sb); 2297 } 2298 /* Copy the remainder. */ 2299 if (len > 0) { 2300 KASSERT(sb->sb_mb != NULL, 2301 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2302 2303 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2304 if (m == NULL) 2305 len = 0; /* Don't flush data from sockbuf. */ 2306 else 2307 uio->uio_resid -= len; 2308 if (*mp0 != NULL) 2309 m_cat(*mp0, m); 2310 else 2311 *mp0 = m; 2312 if (*mp0 == NULL) { 2313 error = ENOBUFS; 2314 goto out; 2315 } 2316 } 2317 } else { 2318 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2319 SOCKBUF_UNLOCK(sb); 2320 error = m_mbuftouio(uio, sb->sb_mb, len); 2321 SOCKBUF_LOCK(sb); 2322 if (error) 2323 goto out; 2324 } 2325 SBLASTRECORDCHK(sb); 2326 SBLASTMBUFCHK(sb); 2327 2328 /* 2329 * Remove the delivered data from the socket buffer unless we 2330 * were only peeking. 2331 */ 2332 if (!(flags & MSG_PEEK)) { 2333 if (len > 0) 2334 sbdrop_locked(sb, len); 2335 2336 /* Notify protocol that we drained some data. */ 2337 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2338 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2339 !(flags & MSG_SOCALLBCK))) { 2340 SOCKBUF_UNLOCK(sb); 2341 VNET_SO_ASSERT(so); 2342 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags); 2343 SOCKBUF_LOCK(sb); 2344 } 2345 } 2346 2347 /* 2348 * For MSG_WAITALL we may have to loop again and wait for 2349 * more data to come in. 2350 */ 2351 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2352 goto restart; 2353 out: 2354 SOCKBUF_LOCK_ASSERT(sb); 2355 SBLASTRECORDCHK(sb); 2356 SBLASTMBUFCHK(sb); 2357 SOCKBUF_UNLOCK(sb); 2358 sbunlock(sb); 2359 return (error); 2360 } 2361 2362 /* 2363 * Optimized version of soreceive() for simple datagram cases from userspace. 2364 * Unlike in the stream case, we're able to drop a datagram if copyout() 2365 * fails, and because we handle datagrams atomically, we don't need to use a 2366 * sleep lock to prevent I/O interlacing. 2367 */ 2368 int 2369 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2370 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2371 { 2372 struct mbuf *m, *m2; 2373 int flags, error; 2374 ssize_t len; 2375 struct protosw *pr = so->so_proto; 2376 struct mbuf *nextrecord; 2377 2378 if (psa != NULL) 2379 *psa = NULL; 2380 if (controlp != NULL) 2381 *controlp = NULL; 2382 if (flagsp != NULL) 2383 flags = *flagsp &~ MSG_EOR; 2384 else 2385 flags = 0; 2386 2387 /* 2388 * For any complicated cases, fall back to the full 2389 * soreceive_generic(). 2390 */ 2391 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB)) 2392 return (soreceive_generic(so, psa, uio, mp0, controlp, 2393 flagsp)); 2394 2395 /* 2396 * Enforce restrictions on use. 2397 */ 2398 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2399 ("soreceive_dgram: wantrcvd")); 2400 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2401 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2402 ("soreceive_dgram: SBS_RCVATMARK")); 2403 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2404 ("soreceive_dgram: P_CONNREQUIRED")); 2405 2406 /* 2407 * Loop blocking while waiting for a datagram. 2408 */ 2409 SOCKBUF_LOCK(&so->so_rcv); 2410 while ((m = so->so_rcv.sb_mb) == NULL) { 2411 KASSERT(sbavail(&so->so_rcv) == 0, 2412 ("soreceive_dgram: sb_mb NULL but sbavail %u", 2413 sbavail(&so->so_rcv))); 2414 if (so->so_error) { 2415 error = so->so_error; 2416 so->so_error = 0; 2417 SOCKBUF_UNLOCK(&so->so_rcv); 2418 return (error); 2419 } 2420 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2421 uio->uio_resid == 0) { 2422 SOCKBUF_UNLOCK(&so->so_rcv); 2423 return (0); 2424 } 2425 if ((so->so_state & SS_NBIO) || 2426 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2427 SOCKBUF_UNLOCK(&so->so_rcv); 2428 return (EWOULDBLOCK); 2429 } 2430 SBLASTRECORDCHK(&so->so_rcv); 2431 SBLASTMBUFCHK(&so->so_rcv); 2432 error = sbwait(&so->so_rcv); 2433 if (error) { 2434 SOCKBUF_UNLOCK(&so->so_rcv); 2435 return (error); 2436 } 2437 } 2438 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2439 2440 if (uio->uio_td) 2441 uio->uio_td->td_ru.ru_msgrcv++; 2442 SBLASTRECORDCHK(&so->so_rcv); 2443 SBLASTMBUFCHK(&so->so_rcv); 2444 nextrecord = m->m_nextpkt; 2445 if (nextrecord == NULL) { 2446 KASSERT(so->so_rcv.sb_lastrecord == m, 2447 ("soreceive_dgram: lastrecord != m")); 2448 } 2449 2450 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2451 ("soreceive_dgram: m_nextpkt != nextrecord")); 2452 2453 /* 2454 * Pull 'm' and its chain off the front of the packet queue. 2455 */ 2456 so->so_rcv.sb_mb = NULL; 2457 sockbuf_pushsync(&so->so_rcv, nextrecord); 2458 2459 /* 2460 * Walk 'm's chain and free that many bytes from the socket buffer. 2461 */ 2462 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2463 sbfree(&so->so_rcv, m2); 2464 2465 /* 2466 * Do a few last checks before we let go of the lock. 2467 */ 2468 SBLASTRECORDCHK(&so->so_rcv); 2469 SBLASTMBUFCHK(&so->so_rcv); 2470 SOCKBUF_UNLOCK(&so->so_rcv); 2471 2472 if (pr->pr_flags & PR_ADDR) { 2473 KASSERT(m->m_type == MT_SONAME, 2474 ("m->m_type == %d", m->m_type)); 2475 if (psa != NULL) 2476 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2477 M_NOWAIT); 2478 m = m_free(m); 2479 } 2480 if (m == NULL) { 2481 /* XXXRW: Can this happen? */ 2482 return (0); 2483 } 2484 2485 /* 2486 * Packet to copyout() is now in 'm' and it is disconnected from the 2487 * queue. 2488 * 2489 * Process one or more MT_CONTROL mbufs present before any data mbufs 2490 * in the first mbuf chain on the socket buffer. We call into the 2491 * protocol to perform externalization (or freeing if controlp == 2492 * NULL). In some cases there can be only MT_CONTROL mbufs without 2493 * MT_DATA mbufs. 2494 */ 2495 if (m->m_type == MT_CONTROL) { 2496 struct mbuf *cm = NULL, *cmn; 2497 struct mbuf **cme = &cm; 2498 2499 do { 2500 m2 = m->m_next; 2501 m->m_next = NULL; 2502 *cme = m; 2503 cme = &(*cme)->m_next; 2504 m = m2; 2505 } while (m != NULL && m->m_type == MT_CONTROL); 2506 while (cm != NULL) { 2507 cmn = cm->m_next; 2508 cm->m_next = NULL; 2509 if (pr->pr_domain->dom_externalize != NULL) { 2510 error = (*pr->pr_domain->dom_externalize) 2511 (cm, controlp, flags); 2512 } else if (controlp != NULL) 2513 *controlp = cm; 2514 else 2515 m_freem(cm); 2516 if (controlp != NULL) { 2517 while (*controlp != NULL) 2518 controlp = &(*controlp)->m_next; 2519 } 2520 cm = cmn; 2521 } 2522 } 2523 KASSERT(m == NULL || m->m_type == MT_DATA, 2524 ("soreceive_dgram: !data")); 2525 while (m != NULL && uio->uio_resid > 0) { 2526 len = uio->uio_resid; 2527 if (len > m->m_len) 2528 len = m->m_len; 2529 error = uiomove(mtod(m, char *), (int)len, uio); 2530 if (error) { 2531 m_freem(m); 2532 return (error); 2533 } 2534 if (len == m->m_len) 2535 m = m_free(m); 2536 else { 2537 m->m_data += len; 2538 m->m_len -= len; 2539 } 2540 } 2541 if (m != NULL) { 2542 flags |= MSG_TRUNC; 2543 m_freem(m); 2544 } 2545 if (flagsp != NULL) 2546 *flagsp |= flags; 2547 return (0); 2548 } 2549 2550 int 2551 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2552 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2553 { 2554 int error; 2555 2556 CURVNET_SET(so->so_vnet); 2557 if (!SOLISTENING(so)) 2558 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, 2559 mp0, controlp, flagsp)); 2560 else 2561 error = ENOTCONN; 2562 CURVNET_RESTORE(); 2563 return (error); 2564 } 2565 2566 int 2567 soshutdown(struct socket *so, int how) 2568 { 2569 struct protosw *pr = so->so_proto; 2570 int error, soerror_enotconn; 2571 2572 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2573 return (EINVAL); 2574 2575 soerror_enotconn = 0; 2576 if ((so->so_state & 2577 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 2578 /* 2579 * POSIX mandates us to return ENOTCONN when shutdown(2) is 2580 * invoked on a datagram sockets, however historically we would 2581 * actually tear socket down. This is known to be leveraged by 2582 * some applications to unblock process waiting in recvXXX(2) 2583 * by other process that it shares that socket with. Try to meet 2584 * both backward-compatibility and POSIX requirements by forcing 2585 * ENOTCONN but still asking protocol to perform pru_shutdown(). 2586 */ 2587 if (so->so_type != SOCK_DGRAM) 2588 return (ENOTCONN); 2589 soerror_enotconn = 1; 2590 } 2591 2592 CURVNET_SET(so->so_vnet); 2593 if (pr->pr_usrreqs->pru_flush != NULL) 2594 (*pr->pr_usrreqs->pru_flush)(so, how); 2595 if (how != SHUT_WR) 2596 sorflush(so); 2597 if (how != SHUT_RD) { 2598 error = (*pr->pr_usrreqs->pru_shutdown)(so); 2599 wakeup(&so->so_timeo); 2600 CURVNET_RESTORE(); 2601 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error); 2602 } 2603 wakeup(&so->so_timeo); 2604 CURVNET_RESTORE(); 2605 2606 return (soerror_enotconn ? ENOTCONN : 0); 2607 } 2608 2609 void 2610 sorflush(struct socket *so) 2611 { 2612 struct sockbuf *sb = &so->so_rcv; 2613 struct protosw *pr = so->so_proto; 2614 struct socket aso; 2615 2616 VNET_SO_ASSERT(so); 2617 2618 /* 2619 * In order to avoid calling dom_dispose with the socket buffer mutex 2620 * held, and in order to generally avoid holding the lock for a long 2621 * time, we make a copy of the socket buffer and clear the original 2622 * (except locks, state). The new socket buffer copy won't have 2623 * initialized locks so we can only call routines that won't use or 2624 * assert those locks. 2625 * 2626 * Dislodge threads currently blocked in receive and wait to acquire 2627 * a lock against other simultaneous readers before clearing the 2628 * socket buffer. Don't let our acquire be interrupted by a signal 2629 * despite any existing socket disposition on interruptable waiting. 2630 */ 2631 socantrcvmore(so); 2632 (void) sblock(sb, SBL_WAIT | SBL_NOINTR); 2633 2634 /* 2635 * Invalidate/clear most of the sockbuf structure, but leave selinfo 2636 * and mutex data unchanged. 2637 */ 2638 SOCKBUF_LOCK(sb); 2639 bzero(&aso, sizeof(aso)); 2640 aso.so_pcb = so->so_pcb; 2641 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero, 2642 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2643 bzero(&sb->sb_startzero, 2644 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2645 SOCKBUF_UNLOCK(sb); 2646 sbunlock(sb); 2647 2648 /* 2649 * Dispose of special rights and flush the copied socket. Don't call 2650 * any unsafe routines (that rely on locks being initialized) on aso. 2651 */ 2652 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 2653 (*pr->pr_domain->dom_dispose)(&aso); 2654 sbrelease_internal(&aso.so_rcv, so); 2655 } 2656 2657 /* 2658 * Wrapper for Socket established helper hook. 2659 * Parameters: socket, context of the hook point, hook id. 2660 */ 2661 static int inline 2662 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 2663 { 2664 struct socket_hhook_data hhook_data = { 2665 .so = so, 2666 .hctx = hctx, 2667 .m = NULL, 2668 .status = 0 2669 }; 2670 2671 CURVNET_SET(so->so_vnet); 2672 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 2673 CURVNET_RESTORE(); 2674 2675 /* Ugly but needed, since hhooks return void for now */ 2676 return (hhook_data.status); 2677 } 2678 2679 /* 2680 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 2681 * additional variant to handle the case where the option value needs to be 2682 * some kind of integer, but not a specific size. In addition to their use 2683 * here, these functions are also called by the protocol-level pr_ctloutput() 2684 * routines. 2685 */ 2686 int 2687 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2688 { 2689 size_t valsize; 2690 2691 /* 2692 * If the user gives us more than we wanted, we ignore it, but if we 2693 * don't get the minimum length the caller wants, we return EINVAL. 2694 * On success, sopt->sopt_valsize is set to however much we actually 2695 * retrieved. 2696 */ 2697 if ((valsize = sopt->sopt_valsize) < minlen) 2698 return EINVAL; 2699 if (valsize > len) 2700 sopt->sopt_valsize = valsize = len; 2701 2702 if (sopt->sopt_td != NULL) 2703 return (copyin(sopt->sopt_val, buf, valsize)); 2704 2705 bcopy(sopt->sopt_val, buf, valsize); 2706 return (0); 2707 } 2708 2709 /* 2710 * Kernel version of setsockopt(2). 2711 * 2712 * XXX: optlen is size_t, not socklen_t 2713 */ 2714 int 2715 so_setsockopt(struct socket *so, int level, int optname, void *optval, 2716 size_t optlen) 2717 { 2718 struct sockopt sopt; 2719 2720 sopt.sopt_level = level; 2721 sopt.sopt_name = optname; 2722 sopt.sopt_dir = SOPT_SET; 2723 sopt.sopt_val = optval; 2724 sopt.sopt_valsize = optlen; 2725 sopt.sopt_td = NULL; 2726 return (sosetopt(so, &sopt)); 2727 } 2728 2729 int 2730 sosetopt(struct socket *so, struct sockopt *sopt) 2731 { 2732 int error, optval; 2733 struct linger l; 2734 struct timeval tv; 2735 sbintime_t val; 2736 uint32_t val32; 2737 #ifdef MAC 2738 struct mac extmac; 2739 #endif 2740 2741 CURVNET_SET(so->so_vnet); 2742 error = 0; 2743 if (sopt->sopt_level != SOL_SOCKET) { 2744 if (so->so_proto->pr_ctloutput != NULL) { 2745 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2746 CURVNET_RESTORE(); 2747 return (error); 2748 } 2749 error = ENOPROTOOPT; 2750 } else { 2751 switch (sopt->sopt_name) { 2752 case SO_ACCEPTFILTER: 2753 error = accept_filt_setopt(so, sopt); 2754 if (error) 2755 goto bad; 2756 break; 2757 2758 case SO_LINGER: 2759 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 2760 if (error) 2761 goto bad; 2762 2763 SOCK_LOCK(so); 2764 so->so_linger = l.l_linger; 2765 if (l.l_onoff) 2766 so->so_options |= SO_LINGER; 2767 else 2768 so->so_options &= ~SO_LINGER; 2769 SOCK_UNLOCK(so); 2770 break; 2771 2772 case SO_DEBUG: 2773 case SO_KEEPALIVE: 2774 case SO_DONTROUTE: 2775 case SO_USELOOPBACK: 2776 case SO_BROADCAST: 2777 case SO_REUSEADDR: 2778 case SO_REUSEPORT: 2779 case SO_OOBINLINE: 2780 case SO_TIMESTAMP: 2781 case SO_BINTIME: 2782 case SO_NOSIGPIPE: 2783 case SO_NO_DDP: 2784 case SO_NO_OFFLOAD: 2785 error = sooptcopyin(sopt, &optval, sizeof optval, 2786 sizeof optval); 2787 if (error) 2788 goto bad; 2789 SOCK_LOCK(so); 2790 if (optval) 2791 so->so_options |= sopt->sopt_name; 2792 else 2793 so->so_options &= ~sopt->sopt_name; 2794 SOCK_UNLOCK(so); 2795 break; 2796 2797 case SO_SETFIB: 2798 error = sooptcopyin(sopt, &optval, sizeof optval, 2799 sizeof optval); 2800 if (error) 2801 goto bad; 2802 2803 if (optval < 0 || optval >= rt_numfibs) { 2804 error = EINVAL; 2805 goto bad; 2806 } 2807 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 2808 (so->so_proto->pr_domain->dom_family == PF_INET6) || 2809 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 2810 so->so_fibnum = optval; 2811 else 2812 so->so_fibnum = 0; 2813 break; 2814 2815 case SO_USER_COOKIE: 2816 error = sooptcopyin(sopt, &val32, sizeof val32, 2817 sizeof val32); 2818 if (error) 2819 goto bad; 2820 so->so_user_cookie = val32; 2821 break; 2822 2823 case SO_SNDBUF: 2824 case SO_RCVBUF: 2825 case SO_SNDLOWAT: 2826 case SO_RCVLOWAT: 2827 error = sooptcopyin(sopt, &optval, sizeof optval, 2828 sizeof optval); 2829 if (error) 2830 goto bad; 2831 2832 /* 2833 * Values < 1 make no sense for any of these options, 2834 * so disallow them. 2835 */ 2836 if (optval < 1) { 2837 error = EINVAL; 2838 goto bad; 2839 } 2840 2841 error = sbsetopt(so, sopt->sopt_name, optval); 2842 break; 2843 2844 case SO_SNDTIMEO: 2845 case SO_RCVTIMEO: 2846 #ifdef COMPAT_FREEBSD32 2847 if (SV_CURPROC_FLAG(SV_ILP32)) { 2848 struct timeval32 tv32; 2849 2850 error = sooptcopyin(sopt, &tv32, sizeof tv32, 2851 sizeof tv32); 2852 CP(tv32, tv, tv_sec); 2853 CP(tv32, tv, tv_usec); 2854 } else 2855 #endif 2856 error = sooptcopyin(sopt, &tv, sizeof tv, 2857 sizeof tv); 2858 if (error) 2859 goto bad; 2860 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 2861 tv.tv_usec >= 1000000) { 2862 error = EDOM; 2863 goto bad; 2864 } 2865 if (tv.tv_sec > INT32_MAX) 2866 val = SBT_MAX; 2867 else 2868 val = tvtosbt(tv); 2869 switch (sopt->sopt_name) { 2870 case SO_SNDTIMEO: 2871 so->so_snd.sb_timeo = val; 2872 break; 2873 case SO_RCVTIMEO: 2874 so->so_rcv.sb_timeo = val; 2875 break; 2876 } 2877 break; 2878 2879 case SO_LABEL: 2880 #ifdef MAC 2881 error = sooptcopyin(sopt, &extmac, sizeof extmac, 2882 sizeof extmac); 2883 if (error) 2884 goto bad; 2885 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 2886 so, &extmac); 2887 #else 2888 error = EOPNOTSUPP; 2889 #endif 2890 break; 2891 2892 case SO_TS_CLOCK: 2893 error = sooptcopyin(sopt, &optval, sizeof optval, 2894 sizeof optval); 2895 if (error) 2896 goto bad; 2897 if (optval < 0 || optval > SO_TS_CLOCK_MAX) { 2898 error = EINVAL; 2899 goto bad; 2900 } 2901 so->so_ts_clock = optval; 2902 break; 2903 2904 case SO_MAX_PACING_RATE: 2905 error = sooptcopyin(sopt, &val32, sizeof(val32), 2906 sizeof(val32)); 2907 if (error) 2908 goto bad; 2909 so->so_max_pacing_rate = val32; 2910 break; 2911 2912 default: 2913 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 2914 error = hhook_run_socket(so, sopt, 2915 HHOOK_SOCKET_OPT); 2916 else 2917 error = ENOPROTOOPT; 2918 break; 2919 } 2920 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 2921 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 2922 } 2923 bad: 2924 CURVNET_RESTORE(); 2925 return (error); 2926 } 2927 2928 /* 2929 * Helper routine for getsockopt. 2930 */ 2931 int 2932 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 2933 { 2934 int error; 2935 size_t valsize; 2936 2937 error = 0; 2938 2939 /* 2940 * Documented get behavior is that we always return a value, possibly 2941 * truncated to fit in the user's buffer. Traditional behavior is 2942 * that we always tell the user precisely how much we copied, rather 2943 * than something useful like the total amount we had available for 2944 * her. Note that this interface is not idempotent; the entire 2945 * answer must be generated ahead of time. 2946 */ 2947 valsize = min(len, sopt->sopt_valsize); 2948 sopt->sopt_valsize = valsize; 2949 if (sopt->sopt_val != NULL) { 2950 if (sopt->sopt_td != NULL) 2951 error = copyout(buf, sopt->sopt_val, valsize); 2952 else 2953 bcopy(buf, sopt->sopt_val, valsize); 2954 } 2955 return (error); 2956 } 2957 2958 int 2959 sogetopt(struct socket *so, struct sockopt *sopt) 2960 { 2961 int error, optval; 2962 struct linger l; 2963 struct timeval tv; 2964 #ifdef MAC 2965 struct mac extmac; 2966 #endif 2967 2968 CURVNET_SET(so->so_vnet); 2969 error = 0; 2970 if (sopt->sopt_level != SOL_SOCKET) { 2971 if (so->so_proto->pr_ctloutput != NULL) 2972 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2973 else 2974 error = ENOPROTOOPT; 2975 CURVNET_RESTORE(); 2976 return (error); 2977 } else { 2978 switch (sopt->sopt_name) { 2979 case SO_ACCEPTFILTER: 2980 error = accept_filt_getopt(so, sopt); 2981 break; 2982 2983 case SO_LINGER: 2984 SOCK_LOCK(so); 2985 l.l_onoff = so->so_options & SO_LINGER; 2986 l.l_linger = so->so_linger; 2987 SOCK_UNLOCK(so); 2988 error = sooptcopyout(sopt, &l, sizeof l); 2989 break; 2990 2991 case SO_USELOOPBACK: 2992 case SO_DONTROUTE: 2993 case SO_DEBUG: 2994 case SO_KEEPALIVE: 2995 case SO_REUSEADDR: 2996 case SO_REUSEPORT: 2997 case SO_BROADCAST: 2998 case SO_OOBINLINE: 2999 case SO_ACCEPTCONN: 3000 case SO_TIMESTAMP: 3001 case SO_BINTIME: 3002 case SO_NOSIGPIPE: 3003 optval = so->so_options & sopt->sopt_name; 3004 integer: 3005 error = sooptcopyout(sopt, &optval, sizeof optval); 3006 break; 3007 3008 case SO_TYPE: 3009 optval = so->so_type; 3010 goto integer; 3011 3012 case SO_PROTOCOL: 3013 optval = so->so_proto->pr_protocol; 3014 goto integer; 3015 3016 case SO_ERROR: 3017 SOCK_LOCK(so); 3018 optval = so->so_error; 3019 so->so_error = 0; 3020 SOCK_UNLOCK(so); 3021 goto integer; 3022 3023 case SO_SNDBUF: 3024 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : 3025 so->so_snd.sb_hiwat; 3026 goto integer; 3027 3028 case SO_RCVBUF: 3029 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : 3030 so->so_rcv.sb_hiwat; 3031 goto integer; 3032 3033 case SO_SNDLOWAT: 3034 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : 3035 so->so_snd.sb_lowat; 3036 goto integer; 3037 3038 case SO_RCVLOWAT: 3039 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : 3040 so->so_rcv.sb_lowat; 3041 goto integer; 3042 3043 case SO_SNDTIMEO: 3044 case SO_RCVTIMEO: 3045 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 3046 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 3047 #ifdef COMPAT_FREEBSD32 3048 if (SV_CURPROC_FLAG(SV_ILP32)) { 3049 struct timeval32 tv32; 3050 3051 CP(tv, tv32, tv_sec); 3052 CP(tv, tv32, tv_usec); 3053 error = sooptcopyout(sopt, &tv32, sizeof tv32); 3054 } else 3055 #endif 3056 error = sooptcopyout(sopt, &tv, sizeof tv); 3057 break; 3058 3059 case SO_LABEL: 3060 #ifdef MAC 3061 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3062 sizeof(extmac)); 3063 if (error) 3064 goto bad; 3065 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 3066 so, &extmac); 3067 if (error) 3068 goto bad; 3069 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3070 #else 3071 error = EOPNOTSUPP; 3072 #endif 3073 break; 3074 3075 case SO_PEERLABEL: 3076 #ifdef MAC 3077 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3078 sizeof(extmac)); 3079 if (error) 3080 goto bad; 3081 error = mac_getsockopt_peerlabel( 3082 sopt->sopt_td->td_ucred, so, &extmac); 3083 if (error) 3084 goto bad; 3085 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3086 #else 3087 error = EOPNOTSUPP; 3088 #endif 3089 break; 3090 3091 case SO_LISTENQLIMIT: 3092 optval = SOLISTENING(so) ? so->sol_qlimit : 0; 3093 goto integer; 3094 3095 case SO_LISTENQLEN: 3096 optval = SOLISTENING(so) ? so->sol_qlen : 0; 3097 goto integer; 3098 3099 case SO_LISTENINCQLEN: 3100 optval = SOLISTENING(so) ? so->sol_incqlen : 0; 3101 goto integer; 3102 3103 case SO_TS_CLOCK: 3104 optval = so->so_ts_clock; 3105 goto integer; 3106 3107 case SO_MAX_PACING_RATE: 3108 optval = so->so_max_pacing_rate; 3109 goto integer; 3110 3111 default: 3112 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3113 error = hhook_run_socket(so, sopt, 3114 HHOOK_SOCKET_OPT); 3115 else 3116 error = ENOPROTOOPT; 3117 break; 3118 } 3119 } 3120 #ifdef MAC 3121 bad: 3122 #endif 3123 CURVNET_RESTORE(); 3124 return (error); 3125 } 3126 3127 int 3128 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 3129 { 3130 struct mbuf *m, *m_prev; 3131 int sopt_size = sopt->sopt_valsize; 3132 3133 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3134 if (m == NULL) 3135 return ENOBUFS; 3136 if (sopt_size > MLEN) { 3137 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 3138 if ((m->m_flags & M_EXT) == 0) { 3139 m_free(m); 3140 return ENOBUFS; 3141 } 3142 m->m_len = min(MCLBYTES, sopt_size); 3143 } else { 3144 m->m_len = min(MLEN, sopt_size); 3145 } 3146 sopt_size -= m->m_len; 3147 *mp = m; 3148 m_prev = m; 3149 3150 while (sopt_size) { 3151 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3152 if (m == NULL) { 3153 m_freem(*mp); 3154 return ENOBUFS; 3155 } 3156 if (sopt_size > MLEN) { 3157 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 3158 M_NOWAIT); 3159 if ((m->m_flags & M_EXT) == 0) { 3160 m_freem(m); 3161 m_freem(*mp); 3162 return ENOBUFS; 3163 } 3164 m->m_len = min(MCLBYTES, sopt_size); 3165 } else { 3166 m->m_len = min(MLEN, sopt_size); 3167 } 3168 sopt_size -= m->m_len; 3169 m_prev->m_next = m; 3170 m_prev = m; 3171 } 3172 return (0); 3173 } 3174 3175 int 3176 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 3177 { 3178 struct mbuf *m0 = m; 3179 3180 if (sopt->sopt_val == NULL) 3181 return (0); 3182 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3183 if (sopt->sopt_td != NULL) { 3184 int error; 3185 3186 error = copyin(sopt->sopt_val, mtod(m, char *), 3187 m->m_len); 3188 if (error != 0) { 3189 m_freem(m0); 3190 return(error); 3191 } 3192 } else 3193 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 3194 sopt->sopt_valsize -= m->m_len; 3195 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3196 m = m->m_next; 3197 } 3198 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 3199 panic("ip6_sooptmcopyin"); 3200 return (0); 3201 } 3202 3203 int 3204 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 3205 { 3206 struct mbuf *m0 = m; 3207 size_t valsize = 0; 3208 3209 if (sopt->sopt_val == NULL) 3210 return (0); 3211 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3212 if (sopt->sopt_td != NULL) { 3213 int error; 3214 3215 error = copyout(mtod(m, char *), sopt->sopt_val, 3216 m->m_len); 3217 if (error != 0) { 3218 m_freem(m0); 3219 return(error); 3220 } 3221 } else 3222 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 3223 sopt->sopt_valsize -= m->m_len; 3224 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3225 valsize += m->m_len; 3226 m = m->m_next; 3227 } 3228 if (m != NULL) { 3229 /* enough soopt buffer should be given from user-land */ 3230 m_freem(m0); 3231 return(EINVAL); 3232 } 3233 sopt->sopt_valsize = valsize; 3234 return (0); 3235 } 3236 3237 /* 3238 * sohasoutofband(): protocol notifies socket layer of the arrival of new 3239 * out-of-band data, which will then notify socket consumers. 3240 */ 3241 void 3242 sohasoutofband(struct socket *so) 3243 { 3244 3245 if (so->so_sigio != NULL) 3246 pgsigio(&so->so_sigio, SIGURG, 0); 3247 selwakeuppri(&so->so_rdsel, PSOCK); 3248 } 3249 3250 int 3251 sopoll(struct socket *so, int events, struct ucred *active_cred, 3252 struct thread *td) 3253 { 3254 3255 /* 3256 * We do not need to set or assert curvnet as long as everyone uses 3257 * sopoll_generic(). 3258 */ 3259 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, 3260 td)); 3261 } 3262 3263 int 3264 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3265 struct thread *td) 3266 { 3267 int revents; 3268 3269 SOCK_LOCK(so); 3270 if (SOLISTENING(so)) { 3271 if (!(events & (POLLIN | POLLRDNORM))) 3272 revents = 0; 3273 else if (!TAILQ_EMPTY(&so->sol_comp)) 3274 revents = events & (POLLIN | POLLRDNORM); 3275 else { 3276 selrecord(td, &so->so_rdsel); 3277 revents = 0; 3278 } 3279 } else { 3280 revents = 0; 3281 SOCKBUF_LOCK(&so->so_snd); 3282 SOCKBUF_LOCK(&so->so_rcv); 3283 if (events & (POLLIN | POLLRDNORM)) 3284 if (soreadabledata(so)) 3285 revents |= events & (POLLIN | POLLRDNORM); 3286 if (events & (POLLOUT | POLLWRNORM)) 3287 if (sowriteable(so)) 3288 revents |= events & (POLLOUT | POLLWRNORM); 3289 if (events & (POLLPRI | POLLRDBAND)) 3290 if (so->so_oobmark || 3291 (so->so_rcv.sb_state & SBS_RCVATMARK)) 3292 revents |= events & (POLLPRI | POLLRDBAND); 3293 if ((events & POLLINIGNEOF) == 0) { 3294 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3295 revents |= events & (POLLIN | POLLRDNORM); 3296 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3297 revents |= POLLHUP; 3298 } 3299 } 3300 if (revents == 0) { 3301 if (events & 3302 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 3303 selrecord(td, &so->so_rdsel); 3304 so->so_rcv.sb_flags |= SB_SEL; 3305 } 3306 if (events & (POLLOUT | POLLWRNORM)) { 3307 selrecord(td, &so->so_wrsel); 3308 so->so_snd.sb_flags |= SB_SEL; 3309 } 3310 } 3311 SOCKBUF_UNLOCK(&so->so_rcv); 3312 SOCKBUF_UNLOCK(&so->so_snd); 3313 } 3314 SOCK_UNLOCK(so); 3315 return (revents); 3316 } 3317 3318 int 3319 soo_kqfilter(struct file *fp, struct knote *kn) 3320 { 3321 struct socket *so = kn->kn_fp->f_data; 3322 struct sockbuf *sb; 3323 struct knlist *knl; 3324 3325 switch (kn->kn_filter) { 3326 case EVFILT_READ: 3327 kn->kn_fop = &soread_filtops; 3328 knl = &so->so_rdsel.si_note; 3329 sb = &so->so_rcv; 3330 break; 3331 case EVFILT_WRITE: 3332 kn->kn_fop = &sowrite_filtops; 3333 knl = &so->so_wrsel.si_note; 3334 sb = &so->so_snd; 3335 break; 3336 case EVFILT_EMPTY: 3337 kn->kn_fop = &soempty_filtops; 3338 knl = &so->so_wrsel.si_note; 3339 sb = &so->so_snd; 3340 break; 3341 default: 3342 return (EINVAL); 3343 } 3344 3345 SOCK_LOCK(so); 3346 if (SOLISTENING(so)) { 3347 knlist_add(knl, kn, 1); 3348 } else { 3349 SOCKBUF_LOCK(sb); 3350 knlist_add(knl, kn, 1); 3351 sb->sb_flags |= SB_KNOTE; 3352 SOCKBUF_UNLOCK(sb); 3353 } 3354 SOCK_UNLOCK(so); 3355 return (0); 3356 } 3357 3358 /* 3359 * Some routines that return EOPNOTSUPP for entry points that are not 3360 * supported by a protocol. Fill in as needed. 3361 */ 3362 int 3363 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 3364 { 3365 3366 return EOPNOTSUPP; 3367 } 3368 3369 int 3370 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job) 3371 { 3372 3373 return EOPNOTSUPP; 3374 } 3375 3376 int 3377 pru_attach_notsupp(struct socket *so, int proto, struct thread *td) 3378 { 3379 3380 return EOPNOTSUPP; 3381 } 3382 3383 int 3384 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3385 { 3386 3387 return EOPNOTSUPP; 3388 } 3389 3390 int 3391 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3392 struct thread *td) 3393 { 3394 3395 return EOPNOTSUPP; 3396 } 3397 3398 int 3399 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3400 { 3401 3402 return EOPNOTSUPP; 3403 } 3404 3405 int 3406 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3407 struct thread *td) 3408 { 3409 3410 return EOPNOTSUPP; 3411 } 3412 3413 int 3414 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 3415 { 3416 3417 return EOPNOTSUPP; 3418 } 3419 3420 int 3421 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 3422 struct ifnet *ifp, struct thread *td) 3423 { 3424 3425 return EOPNOTSUPP; 3426 } 3427 3428 int 3429 pru_disconnect_notsupp(struct socket *so) 3430 { 3431 3432 return EOPNOTSUPP; 3433 } 3434 3435 int 3436 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) 3437 { 3438 3439 return EOPNOTSUPP; 3440 } 3441 3442 int 3443 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) 3444 { 3445 3446 return EOPNOTSUPP; 3447 } 3448 3449 int 3450 pru_rcvd_notsupp(struct socket *so, int flags) 3451 { 3452 3453 return EOPNOTSUPP; 3454 } 3455 3456 int 3457 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 3458 { 3459 3460 return EOPNOTSUPP; 3461 } 3462 3463 int 3464 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, 3465 struct sockaddr *addr, struct mbuf *control, struct thread *td) 3466 { 3467 3468 return EOPNOTSUPP; 3469 } 3470 3471 int 3472 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count) 3473 { 3474 3475 return (EOPNOTSUPP); 3476 } 3477 3478 /* 3479 * This isn't really a ``null'' operation, but it's the default one and 3480 * doesn't do anything destructive. 3481 */ 3482 int 3483 pru_sense_null(struct socket *so, struct stat *sb) 3484 { 3485 3486 sb->st_blksize = so->so_snd.sb_hiwat; 3487 return 0; 3488 } 3489 3490 int 3491 pru_shutdown_notsupp(struct socket *so) 3492 { 3493 3494 return EOPNOTSUPP; 3495 } 3496 3497 int 3498 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) 3499 { 3500 3501 return EOPNOTSUPP; 3502 } 3503 3504 int 3505 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 3506 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 3507 { 3508 3509 return EOPNOTSUPP; 3510 } 3511 3512 int 3513 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 3514 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 3515 { 3516 3517 return EOPNOTSUPP; 3518 } 3519 3520 int 3521 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, 3522 struct thread *td) 3523 { 3524 3525 return EOPNOTSUPP; 3526 } 3527 3528 static void 3529 filt_sordetach(struct knote *kn) 3530 { 3531 struct socket *so = kn->kn_fp->f_data; 3532 3533 so_rdknl_lock(so); 3534 knlist_remove(&so->so_rdsel.si_note, kn, 1); 3535 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note)) 3536 so->so_rcv.sb_flags &= ~SB_KNOTE; 3537 so_rdknl_unlock(so); 3538 } 3539 3540 /*ARGSUSED*/ 3541 static int 3542 filt_soread(struct knote *kn, long hint) 3543 { 3544 struct socket *so; 3545 3546 so = kn->kn_fp->f_data; 3547 3548 if (SOLISTENING(so)) { 3549 SOCK_LOCK_ASSERT(so); 3550 kn->kn_data = so->sol_qlen; 3551 return (!TAILQ_EMPTY(&so->sol_comp)); 3552 } 3553 3554 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3555 3556 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 3557 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3558 kn->kn_flags |= EV_EOF; 3559 kn->kn_fflags = so->so_error; 3560 return (1); 3561 } else if (so->so_error) /* temporary udp error */ 3562 return (1); 3563 3564 if (kn->kn_sfflags & NOTE_LOWAT) { 3565 if (kn->kn_data >= kn->kn_sdata) 3566 return (1); 3567 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) 3568 return (1); 3569 3570 /* This hook returning non-zero indicates an event, not error */ 3571 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); 3572 } 3573 3574 static void 3575 filt_sowdetach(struct knote *kn) 3576 { 3577 struct socket *so = kn->kn_fp->f_data; 3578 3579 so_wrknl_lock(so); 3580 knlist_remove(&so->so_wrsel.si_note, kn, 1); 3581 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note)) 3582 so->so_snd.sb_flags &= ~SB_KNOTE; 3583 so_wrknl_unlock(so); 3584 } 3585 3586 /*ARGSUSED*/ 3587 static int 3588 filt_sowrite(struct knote *kn, long hint) 3589 { 3590 struct socket *so; 3591 3592 so = kn->kn_fp->f_data; 3593 3594 if (SOLISTENING(so)) 3595 return (0); 3596 3597 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3598 kn->kn_data = sbspace(&so->so_snd); 3599 3600 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); 3601 3602 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3603 kn->kn_flags |= EV_EOF; 3604 kn->kn_fflags = so->so_error; 3605 return (1); 3606 } else if (so->so_error) /* temporary udp error */ 3607 return (1); 3608 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3609 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3610 return (0); 3611 else if (kn->kn_sfflags & NOTE_LOWAT) 3612 return (kn->kn_data >= kn->kn_sdata); 3613 else 3614 return (kn->kn_data >= so->so_snd.sb_lowat); 3615 } 3616 3617 static int 3618 filt_soempty(struct knote *kn, long hint) 3619 { 3620 struct socket *so; 3621 3622 so = kn->kn_fp->f_data; 3623 3624 if (SOLISTENING(so)) 3625 return (1); 3626 3627 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3628 kn->kn_data = sbused(&so->so_snd); 3629 3630 if (kn->kn_data == 0) 3631 return (1); 3632 else 3633 return (0); 3634 } 3635 3636 int 3637 socheckuid(struct socket *so, uid_t uid) 3638 { 3639 3640 if (so == NULL) 3641 return (EPERM); 3642 if (so->so_cred->cr_uid != uid) 3643 return (EPERM); 3644 return (0); 3645 } 3646 3647 /* 3648 * These functions are used by protocols to notify the socket layer (and its 3649 * consumers) of state changes in the sockets driven by protocol-side events. 3650 */ 3651 3652 /* 3653 * Procedures to manipulate state flags of socket and do appropriate wakeups. 3654 * 3655 * Normal sequence from the active (originating) side is that 3656 * soisconnecting() is called during processing of connect() call, resulting 3657 * in an eventual call to soisconnected() if/when the connection is 3658 * established. When the connection is torn down soisdisconnecting() is 3659 * called during processing of disconnect() call, and soisdisconnected() is 3660 * called when the connection to the peer is totally severed. The semantics 3661 * of these routines are such that connectionless protocols can call 3662 * soisconnected() and soisdisconnected() only, bypassing the in-progress 3663 * calls when setting up a ``connection'' takes no time. 3664 * 3665 * From the passive side, a socket is created with two queues of sockets: 3666 * so_incomp for connections in progress and so_comp for connections already 3667 * made and awaiting user acceptance. As a protocol is preparing incoming 3668 * connections, it creates a socket structure queued on so_incomp by calling 3669 * sonewconn(). When the connection is established, soisconnected() is 3670 * called, and transfers the socket structure to so_comp, making it available 3671 * to accept(). 3672 * 3673 * If a socket is closed with sockets on either so_incomp or so_comp, these 3674 * sockets are dropped. 3675 * 3676 * If higher-level protocols are implemented in the kernel, the wakeups done 3677 * here will sometimes cause software-interrupt process scheduling. 3678 */ 3679 void 3680 soisconnecting(struct socket *so) 3681 { 3682 3683 SOCK_LOCK(so); 3684 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 3685 so->so_state |= SS_ISCONNECTING; 3686 SOCK_UNLOCK(so); 3687 } 3688 3689 void 3690 soisconnected(struct socket *so) 3691 { 3692 3693 SOCK_LOCK(so); 3694 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 3695 so->so_state |= SS_ISCONNECTED; 3696 3697 if (so->so_qstate == SQ_INCOMP) { 3698 struct socket *head = so->so_listen; 3699 int ret; 3700 3701 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so)); 3702 /* 3703 * Promoting a socket from incomplete queue to complete, we 3704 * need to go through reverse order of locking. We first do 3705 * trylock, and if that doesn't succeed, we go the hard way 3706 * leaving a reference and rechecking consistency after proper 3707 * locking. 3708 */ 3709 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) { 3710 soref(head); 3711 SOCK_UNLOCK(so); 3712 SOLISTEN_LOCK(head); 3713 SOCK_LOCK(so); 3714 if (__predict_false(head != so->so_listen)) { 3715 /* 3716 * The socket went off the listen queue, 3717 * should be lost race to close(2) of sol. 3718 * The socket is about to soabort(). 3719 */ 3720 SOCK_UNLOCK(so); 3721 sorele(head); 3722 return; 3723 } 3724 /* Not the last one, as so holds a ref. */ 3725 refcount_release(&head->so_count); 3726 } 3727 again: 3728 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 3729 TAILQ_REMOVE(&head->sol_incomp, so, so_list); 3730 head->sol_incqlen--; 3731 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 3732 head->sol_qlen++; 3733 so->so_qstate = SQ_COMP; 3734 SOCK_UNLOCK(so); 3735 solisten_wakeup(head); /* unlocks */ 3736 } else { 3737 SOCKBUF_LOCK(&so->so_rcv); 3738 soupcall_set(so, SO_RCV, 3739 head->sol_accept_filter->accf_callback, 3740 head->sol_accept_filter_arg); 3741 so->so_options &= ~SO_ACCEPTFILTER; 3742 ret = head->sol_accept_filter->accf_callback(so, 3743 head->sol_accept_filter_arg, M_NOWAIT); 3744 if (ret == SU_ISCONNECTED) { 3745 soupcall_clear(so, SO_RCV); 3746 SOCKBUF_UNLOCK(&so->so_rcv); 3747 goto again; 3748 } 3749 SOCKBUF_UNLOCK(&so->so_rcv); 3750 SOCK_UNLOCK(so); 3751 SOLISTEN_UNLOCK(head); 3752 } 3753 return; 3754 } 3755 SOCK_UNLOCK(so); 3756 wakeup(&so->so_timeo); 3757 sorwakeup(so); 3758 sowwakeup(so); 3759 } 3760 3761 void 3762 soisdisconnecting(struct socket *so) 3763 { 3764 3765 SOCK_LOCK(so); 3766 so->so_state &= ~SS_ISCONNECTING; 3767 so->so_state |= SS_ISDISCONNECTING; 3768 3769 if (!SOLISTENING(so)) { 3770 SOCKBUF_LOCK(&so->so_rcv); 3771 socantrcvmore_locked(so); 3772 SOCKBUF_LOCK(&so->so_snd); 3773 socantsendmore_locked(so); 3774 } 3775 SOCK_UNLOCK(so); 3776 wakeup(&so->so_timeo); 3777 } 3778 3779 void 3780 soisdisconnected(struct socket *so) 3781 { 3782 3783 SOCK_LOCK(so); 3784 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 3785 so->so_state |= SS_ISDISCONNECTED; 3786 3787 if (!SOLISTENING(so)) { 3788 SOCK_UNLOCK(so); 3789 SOCKBUF_LOCK(&so->so_rcv); 3790 socantrcvmore_locked(so); 3791 SOCKBUF_LOCK(&so->so_snd); 3792 sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); 3793 socantsendmore_locked(so); 3794 } else 3795 SOCK_UNLOCK(so); 3796 wakeup(&so->so_timeo); 3797 } 3798 3799 /* 3800 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 3801 */ 3802 struct sockaddr * 3803 sodupsockaddr(const struct sockaddr *sa, int mflags) 3804 { 3805 struct sockaddr *sa2; 3806 3807 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 3808 if (sa2) 3809 bcopy(sa, sa2, sa->sa_len); 3810 return sa2; 3811 } 3812 3813 /* 3814 * Register per-socket buffer upcalls. 3815 */ 3816 void 3817 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg) 3818 { 3819 struct sockbuf *sb; 3820 3821 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 3822 3823 switch (which) { 3824 case SO_RCV: 3825 sb = &so->so_rcv; 3826 break; 3827 case SO_SND: 3828 sb = &so->so_snd; 3829 break; 3830 default: 3831 panic("soupcall_set: bad which"); 3832 } 3833 SOCKBUF_LOCK_ASSERT(sb); 3834 sb->sb_upcall = func; 3835 sb->sb_upcallarg = arg; 3836 sb->sb_flags |= SB_UPCALL; 3837 } 3838 3839 void 3840 soupcall_clear(struct socket *so, int which) 3841 { 3842 struct sockbuf *sb; 3843 3844 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 3845 3846 switch (which) { 3847 case SO_RCV: 3848 sb = &so->so_rcv; 3849 break; 3850 case SO_SND: 3851 sb = &so->so_snd; 3852 break; 3853 default: 3854 panic("soupcall_clear: bad which"); 3855 } 3856 SOCKBUF_LOCK_ASSERT(sb); 3857 KASSERT(sb->sb_upcall != NULL, 3858 ("%s: so %p no upcall to clear", __func__, so)); 3859 sb->sb_upcall = NULL; 3860 sb->sb_upcallarg = NULL; 3861 sb->sb_flags &= ~SB_UPCALL; 3862 } 3863 3864 void 3865 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg) 3866 { 3867 3868 SOLISTEN_LOCK_ASSERT(so); 3869 so->sol_upcall = func; 3870 so->sol_upcallarg = arg; 3871 } 3872 3873 static void 3874 so_rdknl_lock(void *arg) 3875 { 3876 struct socket *so = arg; 3877 3878 if (SOLISTENING(so)) 3879 SOCK_LOCK(so); 3880 else 3881 SOCKBUF_LOCK(&so->so_rcv); 3882 } 3883 3884 static void 3885 so_rdknl_unlock(void *arg) 3886 { 3887 struct socket *so = arg; 3888 3889 if (SOLISTENING(so)) 3890 SOCK_UNLOCK(so); 3891 else 3892 SOCKBUF_UNLOCK(&so->so_rcv); 3893 } 3894 3895 static void 3896 so_rdknl_assert_locked(void *arg) 3897 { 3898 struct socket *so = arg; 3899 3900 if (SOLISTENING(so)) 3901 SOCK_LOCK_ASSERT(so); 3902 else 3903 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3904 } 3905 3906 static void 3907 so_rdknl_assert_unlocked(void *arg) 3908 { 3909 struct socket *so = arg; 3910 3911 if (SOLISTENING(so)) 3912 SOCK_UNLOCK_ASSERT(so); 3913 else 3914 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); 3915 } 3916 3917 static void 3918 so_wrknl_lock(void *arg) 3919 { 3920 struct socket *so = arg; 3921 3922 if (SOLISTENING(so)) 3923 SOCK_LOCK(so); 3924 else 3925 SOCKBUF_LOCK(&so->so_snd); 3926 } 3927 3928 static void 3929 so_wrknl_unlock(void *arg) 3930 { 3931 struct socket *so = arg; 3932 3933 if (SOLISTENING(so)) 3934 SOCK_UNLOCK(so); 3935 else 3936 SOCKBUF_UNLOCK(&so->so_snd); 3937 } 3938 3939 static void 3940 so_wrknl_assert_locked(void *arg) 3941 { 3942 struct socket *so = arg; 3943 3944 if (SOLISTENING(so)) 3945 SOCK_LOCK_ASSERT(so); 3946 else 3947 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3948 } 3949 3950 static void 3951 so_wrknl_assert_unlocked(void *arg) 3952 { 3953 struct socket *so = arg; 3954 3955 if (SOLISTENING(so)) 3956 SOCK_UNLOCK_ASSERT(so); 3957 else 3958 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 3959 } 3960 3961 /* 3962 * Create an external-format (``xsocket'') structure using the information in 3963 * the kernel-format socket structure pointed to by so. This is done to 3964 * reduce the spew of irrelevant information over this interface, to isolate 3965 * user code from changes in the kernel structure, and potentially to provide 3966 * information-hiding if we decide that some of this information should be 3967 * hidden from users. 3968 */ 3969 void 3970 sotoxsocket(struct socket *so, struct xsocket *xso) 3971 { 3972 3973 xso->xso_len = sizeof *xso; 3974 xso->xso_so = so; 3975 xso->so_type = so->so_type; 3976 xso->so_options = so->so_options; 3977 xso->so_linger = so->so_linger; 3978 xso->so_state = so->so_state; 3979 xso->so_pcb = so->so_pcb; 3980 xso->xso_protocol = so->so_proto->pr_protocol; 3981 xso->xso_family = so->so_proto->pr_domain->dom_family; 3982 xso->so_timeo = so->so_timeo; 3983 xso->so_error = so->so_error; 3984 xso->so_uid = so->so_cred->cr_uid; 3985 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 3986 if (SOLISTENING(so)) { 3987 xso->so_qlen = so->sol_qlen; 3988 xso->so_incqlen = so->sol_incqlen; 3989 xso->so_qlimit = so->sol_qlimit; 3990 xso->so_oobmark = 0; 3991 bzero(&xso->so_snd, sizeof(xso->so_snd)); 3992 bzero(&xso->so_rcv, sizeof(xso->so_rcv)); 3993 } else { 3994 xso->so_state |= so->so_qstate; 3995 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0; 3996 xso->so_oobmark = so->so_oobmark; 3997 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 3998 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 3999 } 4000 } 4001 4002 struct sockbuf * 4003 so_sockbuf_rcv(struct socket *so) 4004 { 4005 4006 return (&so->so_rcv); 4007 } 4008 4009 struct sockbuf * 4010 so_sockbuf_snd(struct socket *so) 4011 { 4012 4013 return (&so->so_snd); 4014 } 4015 4016 int 4017 so_state_get(const struct socket *so) 4018 { 4019 4020 return (so->so_state); 4021 } 4022 4023 void 4024 so_state_set(struct socket *so, int val) 4025 { 4026 4027 so->so_state = val; 4028 } 4029 4030 int 4031 so_options_get(const struct socket *so) 4032 { 4033 4034 return (so->so_options); 4035 } 4036 4037 void 4038 so_options_set(struct socket *so, int val) 4039 { 4040 4041 so->so_options = val; 4042 } 4043 4044 int 4045 so_error_get(const struct socket *so) 4046 { 4047 4048 return (so->so_error); 4049 } 4050 4051 void 4052 so_error_set(struct socket *so, int val) 4053 { 4054 4055 so->so_error = val; 4056 } 4057 4058 int 4059 so_linger_get(const struct socket *so) 4060 { 4061 4062 return (so->so_linger); 4063 } 4064 4065 void 4066 so_linger_set(struct socket *so, int val) 4067 { 4068 4069 so->so_linger = val; 4070 } 4071 4072 struct protosw * 4073 so_protosw_get(const struct socket *so) 4074 { 4075 4076 return (so->so_proto); 4077 } 4078 4079 void 4080 so_protosw_set(struct socket *so, struct protosw *val) 4081 { 4082 4083 so->so_proto = val; 4084 } 4085 4086 void 4087 so_sorwakeup(struct socket *so) 4088 { 4089 4090 sorwakeup(so); 4091 } 4092 4093 void 4094 so_sowwakeup(struct socket *so) 4095 { 4096 4097 sowwakeup(so); 4098 } 4099 4100 void 4101 so_sorwakeup_locked(struct socket *so) 4102 { 4103 4104 sorwakeup_locked(so); 4105 } 4106 4107 void 4108 so_sowwakeup_locked(struct socket *so) 4109 { 4110 4111 sowwakeup_locked(so); 4112 } 4113 4114 void 4115 so_lock(struct socket *so) 4116 { 4117 4118 SOCK_LOCK(so); 4119 } 4120 4121 void 4122 so_unlock(struct socket *so) 4123 { 4124 4125 SOCK_UNLOCK(so); 4126 } 4127