1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993 3 * The Regents of the University of California. 4 * Copyright (c) 2004 The FreeBSD Foundation 5 * Copyright (c) 2004-2008 Robert N. M. Watson 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 33 */ 34 35 /* 36 * Comments on the socket life cycle: 37 * 38 * soalloc() sets of socket layer state for a socket, called only by 39 * socreate() and sonewconn(). Socket layer private. 40 * 41 * sodealloc() tears down socket layer state for a socket, called only by 42 * sofree() and sonewconn(). Socket layer private. 43 * 44 * pru_attach() associates protocol layer state with an allocated socket; 45 * called only once, may fail, aborting socket allocation. This is called 46 * from socreate() and sonewconn(). Socket layer private. 47 * 48 * pru_detach() disassociates protocol layer state from an attached socket, 49 * and will be called exactly once for sockets in which pru_attach() has 50 * been successfully called. If pru_attach() returned an error, 51 * pru_detach() will not be called. Socket layer private. 52 * 53 * pru_abort() and pru_close() notify the protocol layer that the last 54 * consumer of a socket is starting to tear down the socket, and that the 55 * protocol should terminate the connection. Historically, pru_abort() also 56 * detached protocol state from the socket state, but this is no longer the 57 * case. 58 * 59 * socreate() creates a socket and attaches protocol state. This is a public 60 * interface that may be used by socket layer consumers to create new 61 * sockets. 62 * 63 * sonewconn() creates a socket and attaches protocol state. This is a 64 * public interface that may be used by protocols to create new sockets when 65 * a new connection is received and will be available for accept() on a 66 * listen socket. 67 * 68 * soclose() destroys a socket after possibly waiting for it to disconnect. 69 * This is a public interface that socket consumers should use to close and 70 * release a socket when done with it. 71 * 72 * soabort() destroys a socket without waiting for it to disconnect (used 73 * only for incoming connections that are already partially or fully 74 * connected). This is used internally by the socket layer when clearing 75 * listen socket queues (due to overflow or close on the listen socket), but 76 * is also a public interface protocols may use to abort connections in 77 * their incomplete listen queues should they no longer be required. Sockets 78 * placed in completed connection listen queues should not be aborted for 79 * reasons described in the comment above the soclose() implementation. This 80 * is not a general purpose close routine, and except in the specific 81 * circumstances described here, should not be used. 82 * 83 * sofree() will free a socket and its protocol state if all references on 84 * the socket have been released, and is the public interface to attempt to 85 * free a socket when a reference is removed. This is a socket layer private 86 * interface. 87 * 88 * NOTE: In addition to socreate() and soclose(), which provide a single 89 * socket reference to the consumer to be managed as required, there are two 90 * calls to explicitly manage socket references, soref(), and sorele(). 91 * Currently, these are generally required only when transitioning a socket 92 * from a listen queue to a file descriptor, in order to prevent garbage 93 * collection of the socket at an untimely moment. For a number of reasons, 94 * these interfaces are not preferred, and should be avoided. 95 * 96 * NOTE: With regard to VNETs the general rule is that callers do not set 97 * curvnet. Exceptions to this rule include soabort(), sodisconnect(), 98 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn() 99 * and sorflush(), which are usually called from a pre-set VNET context. 100 * sopoll() currently does not need a VNET context to be set. 101 */ 102 103 #include <sys/cdefs.h> 104 __FBSDID("$FreeBSD$"); 105 106 #include "opt_inet.h" 107 #include "opt_inet6.h" 108 #include "opt_compat.h" 109 #include "opt_sctp.h" 110 111 #include <sys/param.h> 112 #include <sys/systm.h> 113 #include <sys/fcntl.h> 114 #include <sys/limits.h> 115 #include <sys/lock.h> 116 #include <sys/mac.h> 117 #include <sys/malloc.h> 118 #include <sys/mbuf.h> 119 #include <sys/mutex.h> 120 #include <sys/domain.h> 121 #include <sys/file.h> /* for struct knote */ 122 #include <sys/hhook.h> 123 #include <sys/kernel.h> 124 #include <sys/khelp.h> 125 #include <sys/event.h> 126 #include <sys/eventhandler.h> 127 #include <sys/poll.h> 128 #include <sys/proc.h> 129 #include <sys/protosw.h> 130 #include <sys/socket.h> 131 #include <sys/socketvar.h> 132 #include <sys/resourcevar.h> 133 #include <net/route.h> 134 #include <sys/signalvar.h> 135 #include <sys/stat.h> 136 #include <sys/sx.h> 137 #include <sys/sysctl.h> 138 #include <sys/taskqueue.h> 139 #include <sys/uio.h> 140 #include <sys/jail.h> 141 #include <sys/syslog.h> 142 #include <netinet/in.h> 143 144 #include <net/vnet.h> 145 146 #include <security/mac/mac_framework.h> 147 148 #include <vm/uma.h> 149 150 #ifdef COMPAT_FREEBSD32 151 #include <sys/mount.h> 152 #include <sys/sysent.h> 153 #include <compat/freebsd32/freebsd32.h> 154 #endif 155 156 static int soreceive_rcvoob(struct socket *so, struct uio *uio, 157 int flags); 158 static void so_rdknl_lock(void *); 159 static void so_rdknl_unlock(void *); 160 static void so_rdknl_assert_locked(void *); 161 static void so_rdknl_assert_unlocked(void *); 162 static void so_wrknl_lock(void *); 163 static void so_wrknl_unlock(void *); 164 static void so_wrknl_assert_locked(void *); 165 static void so_wrknl_assert_unlocked(void *); 166 167 static void filt_sordetach(struct knote *kn); 168 static int filt_soread(struct knote *kn, long hint); 169 static void filt_sowdetach(struct knote *kn); 170 static int filt_sowrite(struct knote *kn, long hint); 171 static int filt_soempty(struct knote *kn, long hint); 172 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id); 173 fo_kqfilter_t soo_kqfilter; 174 175 static struct filterops soread_filtops = { 176 .f_isfd = 1, 177 .f_detach = filt_sordetach, 178 .f_event = filt_soread, 179 }; 180 static struct filterops sowrite_filtops = { 181 .f_isfd = 1, 182 .f_detach = filt_sowdetach, 183 .f_event = filt_sowrite, 184 }; 185 static struct filterops soempty_filtops = { 186 .f_isfd = 1, 187 .f_detach = filt_sowdetach, 188 .f_event = filt_soempty, 189 }; 190 191 so_gen_t so_gencnt; /* generation count for sockets */ 192 193 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 194 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 195 196 #define VNET_SO_ASSERT(so) \ 197 VNET_ASSERT(curvnet != NULL, \ 198 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so))); 199 200 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]); 201 #define V_socket_hhh VNET(socket_hhh) 202 203 /* 204 * Limit on the number of connections in the listen queue waiting 205 * for accept(2). 206 * NB: The original sysctl somaxconn is still available but hidden 207 * to prevent confusion about the actual purpose of this number. 208 */ 209 static u_int somaxconn = SOMAXCONN; 210 211 static int 212 sysctl_somaxconn(SYSCTL_HANDLER_ARGS) 213 { 214 int error; 215 int val; 216 217 val = somaxconn; 218 error = sysctl_handle_int(oidp, &val, 0, req); 219 if (error || !req->newptr ) 220 return (error); 221 222 /* 223 * The purpose of the UINT_MAX / 3 limit, is so that the formula 224 * 3 * so_qlimit / 2 225 * below, will not overflow. 226 */ 227 228 if (val < 1 || val > UINT_MAX / 3) 229 return (EINVAL); 230 231 somaxconn = val; 232 return (0); 233 } 234 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW, 235 0, sizeof(int), sysctl_somaxconn, "I", 236 "Maximum listen socket pending connection accept queue size"); 237 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, 238 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP, 239 0, sizeof(int), sysctl_somaxconn, "I", 240 "Maximum listen socket pending connection accept queue size (compat)"); 241 242 static int numopensockets; 243 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD, 244 &numopensockets, 0, "Number of open sockets"); 245 246 /* 247 * accept_mtx locks down per-socket fields relating to accept queues. See 248 * socketvar.h for an annotation of the protected fields of struct socket. 249 */ 250 struct mtx accept_mtx; 251 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF); 252 253 /* 254 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket 255 * so_gencnt field. 256 */ 257 static struct mtx so_global_mtx; 258 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF); 259 260 /* 261 * General IPC sysctl name space, used by sockets and a variety of other IPC 262 * types. 263 */ 264 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); 265 266 /* 267 * Initialize the socket subsystem and set up the socket 268 * memory allocator. 269 */ 270 static uma_zone_t socket_zone; 271 int maxsockets; 272 273 static void 274 socket_zone_change(void *tag) 275 { 276 277 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 278 } 279 280 static void 281 socket_hhook_register(int subtype) 282 { 283 284 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype, 285 &V_socket_hhh[subtype], 286 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 287 printf("%s: WARNING: unable to register hook\n", __func__); 288 } 289 290 static void 291 socket_hhook_deregister(int subtype) 292 { 293 294 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0) 295 printf("%s: WARNING: unable to deregister hook\n", __func__); 296 } 297 298 static void 299 socket_init(void *tag) 300 { 301 302 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL, 303 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 304 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 305 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached"); 306 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL, 307 EVENTHANDLER_PRI_FIRST); 308 } 309 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL); 310 311 static void 312 socket_vnet_init(const void *unused __unused) 313 { 314 int i; 315 316 /* We expect a contiguous range */ 317 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 318 socket_hhook_register(i); 319 } 320 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 321 socket_vnet_init, NULL); 322 323 static void 324 socket_vnet_uninit(const void *unused __unused) 325 { 326 int i; 327 328 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 329 socket_hhook_deregister(i); 330 } 331 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 332 socket_vnet_uninit, NULL); 333 334 /* 335 * Initialise maxsockets. This SYSINIT must be run after 336 * tunable_mbinit(). 337 */ 338 static void 339 init_maxsockets(void *ignored) 340 { 341 342 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 343 maxsockets = imax(maxsockets, maxfiles); 344 } 345 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); 346 347 /* 348 * Sysctl to get and set the maximum global sockets limit. Notify protocols 349 * of the change so that they can update their dependent limits as required. 350 */ 351 static int 352 sysctl_maxsockets(SYSCTL_HANDLER_ARGS) 353 { 354 int error, newmaxsockets; 355 356 newmaxsockets = maxsockets; 357 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req); 358 if (error == 0 && req->newptr) { 359 if (newmaxsockets > maxsockets && 360 newmaxsockets <= maxfiles) { 361 maxsockets = newmaxsockets; 362 EVENTHANDLER_INVOKE(maxsockets_change); 363 } else 364 error = EINVAL; 365 } 366 return (error); 367 } 368 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW, 369 &maxsockets, 0, sysctl_maxsockets, "IU", 370 "Maximum number of sockets available"); 371 372 /* 373 * Socket operation routines. These routines are called by the routines in 374 * sys_socket.c or from a system process, and implement the semantics of 375 * socket operations by switching out to the protocol specific routines. 376 */ 377 378 /* 379 * Get a socket structure from our zone, and initialize it. Note that it 380 * would probably be better to allocate socket and PCB at the same time, but 381 * I'm not convinced that all the protocols can be easily modified to do 382 * this. 383 * 384 * soalloc() returns a socket with a ref count of 0. 385 */ 386 static struct socket * 387 soalloc(struct vnet *vnet) 388 { 389 struct socket *so; 390 391 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO); 392 if (so == NULL) 393 return (NULL); 394 #ifdef MAC 395 if (mac_socket_init(so, M_NOWAIT) != 0) { 396 uma_zfree(socket_zone, so); 397 return (NULL); 398 } 399 #endif 400 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) { 401 uma_zfree(socket_zone, so); 402 return (NULL); 403 } 404 405 /* 406 * The socket locking protocol allows to lock 2 sockets at a time, 407 * however, the first one must be a listening socket. WITNESS lacks 408 * a feature to change class of an existing lock, so we use DUPOK. 409 */ 410 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK); 411 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd"); 412 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv"); 413 so->so_rcv.sb_sel = &so->so_rdsel; 414 so->so_snd.sb_sel = &so->so_wrsel; 415 sx_init(&so->so_snd.sb_sx, "so_snd_sx"); 416 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx"); 417 TAILQ_INIT(&so->so_snd.sb_aiojobq); 418 TAILQ_INIT(&so->so_rcv.sb_aiojobq); 419 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so); 420 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so); 421 #ifdef VIMAGE 422 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p", 423 __func__, __LINE__, so)); 424 so->so_vnet = vnet; 425 #endif 426 /* We shouldn't need the so_global_mtx */ 427 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) { 428 /* Do we need more comprehensive error returns? */ 429 uma_zfree(socket_zone, so); 430 return (NULL); 431 } 432 mtx_lock(&so_global_mtx); 433 so->so_gencnt = ++so_gencnt; 434 ++numopensockets; 435 #ifdef VIMAGE 436 vnet->vnet_sockcnt++; 437 #endif 438 mtx_unlock(&so_global_mtx); 439 440 return (so); 441 } 442 443 /* 444 * Free the storage associated with a socket at the socket layer, tear down 445 * locks, labels, etc. All protocol state is assumed already to have been 446 * torn down (and possibly never set up) by the caller. 447 */ 448 static void 449 sodealloc(struct socket *so) 450 { 451 452 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); 453 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL")); 454 455 mtx_lock(&so_global_mtx); 456 so->so_gencnt = ++so_gencnt; 457 --numopensockets; /* Could be below, but faster here. */ 458 #ifdef VIMAGE 459 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p", 460 __func__, __LINE__, so)); 461 so->so_vnet->vnet_sockcnt--; 462 #endif 463 mtx_unlock(&so_global_mtx); 464 #ifdef MAC 465 mac_socket_destroy(so); 466 #endif 467 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE); 468 469 crfree(so->so_cred); 470 khelp_destroy_osd(&so->osd); 471 if (SOLISTENING(so)) { 472 if (so->sol_accept_filter != NULL) 473 accept_filt_setopt(so, NULL); 474 } else { 475 if (so->so_rcv.sb_hiwat) 476 (void)chgsbsize(so->so_cred->cr_uidinfo, 477 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 478 if (so->so_snd.sb_hiwat) 479 (void)chgsbsize(so->so_cred->cr_uidinfo, 480 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 481 sx_destroy(&so->so_snd.sb_sx); 482 sx_destroy(&so->so_rcv.sb_sx); 483 SOCKBUF_LOCK_DESTROY(&so->so_snd); 484 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 485 } 486 mtx_destroy(&so->so_lock); 487 uma_zfree(socket_zone, so); 488 } 489 490 /* 491 * socreate returns a socket with a ref count of 1. The socket should be 492 * closed with soclose(). 493 */ 494 int 495 socreate(int dom, struct socket **aso, int type, int proto, 496 struct ucred *cred, struct thread *td) 497 { 498 struct protosw *prp; 499 struct socket *so; 500 int error; 501 502 if (proto) 503 prp = pffindproto(dom, proto, type); 504 else 505 prp = pffindtype(dom, type); 506 507 if (prp == NULL) { 508 /* No support for domain. */ 509 if (pffinddomain(dom) == NULL) 510 return (EAFNOSUPPORT); 511 /* No support for socket type. */ 512 if (proto == 0 && type != 0) 513 return (EPROTOTYPE); 514 return (EPROTONOSUPPORT); 515 } 516 if (prp->pr_usrreqs->pru_attach == NULL || 517 prp->pr_usrreqs->pru_attach == pru_attach_notsupp) 518 return (EPROTONOSUPPORT); 519 520 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0) 521 return (EPROTONOSUPPORT); 522 523 if (prp->pr_type != type) 524 return (EPROTOTYPE); 525 so = soalloc(CRED_TO_VNET(cred)); 526 if (so == NULL) 527 return (ENOBUFS); 528 529 so->so_type = type; 530 so->so_cred = crhold(cred); 531 if ((prp->pr_domain->dom_family == PF_INET) || 532 (prp->pr_domain->dom_family == PF_INET6) || 533 (prp->pr_domain->dom_family == PF_ROUTE)) 534 so->so_fibnum = td->td_proc->p_fibnum; 535 else 536 so->so_fibnum = 0; 537 so->so_proto = prp; 538 #ifdef MAC 539 mac_socket_create(cred, so); 540 #endif 541 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 542 so_rdknl_assert_locked, so_rdknl_assert_unlocked); 543 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 544 so_wrknl_assert_locked, so_wrknl_assert_unlocked); 545 /* 546 * Auto-sizing of socket buffers is managed by the protocols and 547 * the appropriate flags must be set in the pru_attach function. 548 */ 549 CURVNET_SET(so->so_vnet); 550 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); 551 CURVNET_RESTORE(); 552 if (error) { 553 sodealloc(so); 554 return (error); 555 } 556 soref(so); 557 *aso = so; 558 return (0); 559 } 560 561 #ifdef REGRESSION 562 static int regression_sonewconn_earlytest = 1; 563 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW, 564 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test"); 565 #endif 566 567 /* 568 * When an attempt at a new connection is noted on a socket which accepts 569 * connections, sonewconn is called. If the connection is possible (subject 570 * to space constraints, etc.) then we allocate a new structure, properly 571 * linked into the data structure of the original socket, and return this. 572 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED. 573 * 574 * Note: the ref count on the socket is 0 on return. 575 */ 576 struct socket * 577 sonewconn(struct socket *head, int connstatus) 578 { 579 static struct timeval lastover; 580 static struct timeval overinterval = { 60, 0 }; 581 static int overcount; 582 583 struct socket *so; 584 u_int over; 585 586 SOLISTEN_LOCK(head); 587 over = (head->sol_qlen > 3 * head->sol_qlimit / 2); 588 SOLISTEN_UNLOCK(head); 589 #ifdef REGRESSION 590 if (regression_sonewconn_earlytest && over) { 591 #else 592 if (over) { 593 #endif 594 overcount++; 595 596 if (ratecheck(&lastover, &overinterval)) { 597 log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: " 598 "%i already in queue awaiting acceptance " 599 "(%d occurrences)\n", 600 __func__, head->so_pcb, head->sol_qlen, overcount); 601 602 overcount = 0; 603 } 604 605 return (NULL); 606 } 607 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL", 608 __func__, head)); 609 so = soalloc(head->so_vnet); 610 if (so == NULL) { 611 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 612 "limit reached or out of memory\n", 613 __func__, head->so_pcb); 614 return (NULL); 615 } 616 so->so_listen = head; 617 so->so_type = head->so_type; 618 so->so_linger = head->so_linger; 619 so->so_state = head->so_state | SS_NOFDREF; 620 so->so_fibnum = head->so_fibnum; 621 so->so_proto = head->so_proto; 622 so->so_cred = crhold(head->so_cred); 623 #ifdef MAC 624 mac_socket_newconn(head, so); 625 #endif 626 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 627 so_rdknl_assert_locked, so_rdknl_assert_unlocked); 628 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 629 so_wrknl_assert_locked, so_wrknl_assert_unlocked); 630 VNET_SO_ASSERT(head); 631 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) { 632 sodealloc(so); 633 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 634 __func__, head->so_pcb); 635 return (NULL); 636 } 637 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 638 sodealloc(so); 639 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 640 __func__, head->so_pcb); 641 return (NULL); 642 } 643 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat; 644 so->so_snd.sb_lowat = head->sol_sbsnd_lowat; 645 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo; 646 so->so_snd.sb_timeo = head->sol_sbsnd_timeo; 647 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE; 648 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE; 649 650 SOLISTEN_LOCK(head); 651 if (head->sol_accept_filter != NULL) 652 connstatus = 0; 653 so->so_state |= connstatus; 654 so->so_options = head->so_options & ~SO_ACCEPTCONN; 655 soref(head); /* A socket on (in)complete queue refs head. */ 656 if (connstatus) { 657 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 658 so->so_qstate = SQ_COMP; 659 head->sol_qlen++; 660 solisten_wakeup(head); /* unlocks */ 661 } else { 662 /* 663 * Keep removing sockets from the head until there's room for 664 * us to insert on the tail. In pre-locking revisions, this 665 * was a simple if(), but as we could be racing with other 666 * threads and soabort() requires dropping locks, we must 667 * loop waiting for the condition to be true. 668 */ 669 while (head->sol_incqlen > head->sol_qlimit) { 670 struct socket *sp; 671 672 sp = TAILQ_FIRST(&head->sol_incomp); 673 TAILQ_REMOVE(&head->sol_incomp, sp, so_list); 674 head->sol_incqlen--; 675 SOCK_LOCK(sp); 676 sp->so_qstate = SQ_NONE; 677 sp->so_listen = NULL; 678 SOCK_UNLOCK(sp); 679 sorele(head); /* does SOLISTEN_UNLOCK, head stays */ 680 soabort(sp); 681 SOLISTEN_LOCK(head); 682 } 683 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list); 684 so->so_qstate = SQ_INCOMP; 685 head->sol_incqlen++; 686 SOLISTEN_UNLOCK(head); 687 } 688 return (so); 689 } 690 691 #ifdef SCTP 692 /* 693 * Socket part of sctp_peeloff(). Detach a new socket from an 694 * association. The new socket is returned with a reference. 695 */ 696 struct socket * 697 sopeeloff(struct socket *head) 698 { 699 struct socket *so; 700 701 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p", 702 __func__, __LINE__, head)); 703 so = soalloc(head->so_vnet); 704 if (so == NULL) { 705 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 706 "limit reached or out of memory\n", 707 __func__, head->so_pcb); 708 return (NULL); 709 } 710 so->so_type = head->so_type; 711 so->so_options = head->so_options; 712 so->so_linger = head->so_linger; 713 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED; 714 so->so_fibnum = head->so_fibnum; 715 so->so_proto = head->so_proto; 716 so->so_cred = crhold(head->so_cred); 717 #ifdef MAC 718 mac_socket_newconn(head, so); 719 #endif 720 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 721 so_rdknl_assert_locked, so_rdknl_assert_unlocked); 722 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 723 so_wrknl_assert_locked, so_wrknl_assert_unlocked); 724 VNET_SO_ASSERT(head); 725 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { 726 sodealloc(so); 727 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 728 __func__, head->so_pcb); 729 return (NULL); 730 } 731 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 732 sodealloc(so); 733 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 734 __func__, head->so_pcb); 735 return (NULL); 736 } 737 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 738 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 739 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 740 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 741 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; 742 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; 743 744 soref(so); 745 746 return (so); 747 } 748 #endif /* SCTP */ 749 750 int 751 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 752 { 753 int error; 754 755 CURVNET_SET(so->so_vnet); 756 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td); 757 CURVNET_RESTORE(); 758 return (error); 759 } 760 761 int 762 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 763 { 764 int error; 765 766 CURVNET_SET(so->so_vnet); 767 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td); 768 CURVNET_RESTORE(); 769 return (error); 770 } 771 772 /* 773 * solisten() transitions a socket from a non-listening state to a listening 774 * state, but can also be used to update the listen queue depth on an 775 * existing listen socket. The protocol will call back into the sockets 776 * layer using solisten_proto_check() and solisten_proto() to check and set 777 * socket-layer listen state. Call backs are used so that the protocol can 778 * acquire both protocol and socket layer locks in whatever order is required 779 * by the protocol. 780 * 781 * Protocol implementors are advised to hold the socket lock across the 782 * socket-layer test and set to avoid races at the socket layer. 783 */ 784 int 785 solisten(struct socket *so, int backlog, struct thread *td) 786 { 787 int error; 788 789 CURVNET_SET(so->so_vnet); 790 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td); 791 CURVNET_RESTORE(); 792 return (error); 793 } 794 795 int 796 solisten_proto_check(struct socket *so) 797 { 798 799 SOCK_LOCK_ASSERT(so); 800 801 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 802 SS_ISDISCONNECTING)) 803 return (EINVAL); 804 return (0); 805 } 806 807 void 808 solisten_proto(struct socket *so, int backlog) 809 { 810 int sbrcv_lowat, sbsnd_lowat; 811 u_int sbrcv_hiwat, sbsnd_hiwat; 812 short sbrcv_flags, sbsnd_flags; 813 sbintime_t sbrcv_timeo, sbsnd_timeo; 814 815 SOCK_LOCK_ASSERT(so); 816 817 if (SOLISTENING(so)) 818 goto listening; 819 820 /* 821 * Change this socket to listening state. 822 */ 823 sbrcv_lowat = so->so_rcv.sb_lowat; 824 sbsnd_lowat = so->so_snd.sb_lowat; 825 sbrcv_hiwat = so->so_rcv.sb_hiwat; 826 sbsnd_hiwat = so->so_snd.sb_hiwat; 827 sbrcv_flags = so->so_rcv.sb_flags; 828 sbsnd_flags = so->so_snd.sb_flags; 829 sbrcv_timeo = so->so_rcv.sb_timeo; 830 sbsnd_timeo = so->so_snd.sb_timeo; 831 832 sbdestroy(&so->so_snd, so); 833 sbdestroy(&so->so_rcv, so); 834 sx_destroy(&so->so_snd.sb_sx); 835 sx_destroy(&so->so_rcv.sb_sx); 836 SOCKBUF_LOCK_DESTROY(&so->so_snd); 837 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 838 839 #ifdef INVARIANTS 840 bzero(&so->so_rcv, 841 sizeof(struct socket) - offsetof(struct socket, so_rcv)); 842 #endif 843 844 so->sol_sbrcv_lowat = sbrcv_lowat; 845 so->sol_sbsnd_lowat = sbsnd_lowat; 846 so->sol_sbrcv_hiwat = sbrcv_hiwat; 847 so->sol_sbsnd_hiwat = sbsnd_hiwat; 848 so->sol_sbrcv_flags = sbrcv_flags; 849 so->sol_sbsnd_flags = sbsnd_flags; 850 so->sol_sbrcv_timeo = sbrcv_timeo; 851 so->sol_sbsnd_timeo = sbsnd_timeo; 852 853 so->sol_qlen = so->sol_incqlen = 0; 854 TAILQ_INIT(&so->sol_incomp); 855 TAILQ_INIT(&so->sol_comp); 856 857 so->sol_accept_filter = NULL; 858 so->sol_accept_filter_arg = NULL; 859 so->sol_accept_filter_str = NULL; 860 861 so->sol_upcall = NULL; 862 so->sol_upcallarg = NULL; 863 864 so->so_options |= SO_ACCEPTCONN; 865 866 listening: 867 if (backlog < 0 || backlog > somaxconn) 868 backlog = somaxconn; 869 so->sol_qlimit = backlog; 870 } 871 872 /* 873 * Wakeup listeners/subsystems once we have a complete connection. 874 * Enters with lock, returns unlocked. 875 */ 876 void 877 solisten_wakeup(struct socket *sol) 878 { 879 880 if (sol->sol_upcall != NULL) 881 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT); 882 else { 883 selwakeuppri(&sol->so_rdsel, PSOCK); 884 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0); 885 } 886 SOLISTEN_UNLOCK(sol); 887 wakeup_one(&sol->sol_comp); 888 } 889 890 /* 891 * Return single connection off a listening socket queue. Main consumer of 892 * the function is kern_accept4(). Some modules, that do their own accept 893 * management also use the function. 894 * 895 * Listening socket must be locked on entry and is returned unlocked on 896 * return. 897 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT. 898 */ 899 int 900 solisten_dequeue(struct socket *head, struct socket **ret, int flags) 901 { 902 struct socket *so; 903 int error; 904 905 SOLISTEN_LOCK_ASSERT(head); 906 907 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) && 908 head->so_error == 0) { 909 error = msleep(&head->sol_comp, &head->so_lock, PSOCK | PCATCH, 910 "accept", 0); 911 if (error != 0) { 912 SOLISTEN_UNLOCK(head); 913 return (error); 914 } 915 } 916 if (head->so_error) { 917 error = head->so_error; 918 head->so_error = 0; 919 SOLISTEN_UNLOCK(head); 920 return (error); 921 } 922 if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp)) { 923 SOLISTEN_UNLOCK(head); 924 return (EWOULDBLOCK); 925 } 926 so = TAILQ_FIRST(&head->sol_comp); 927 SOCK_LOCK(so); 928 KASSERT(so->so_qstate == SQ_COMP, 929 ("%s: so %p not SQ_COMP", __func__, so)); 930 soref(so); 931 head->sol_qlen--; 932 so->so_qstate = SQ_NONE; 933 so->so_listen = NULL; 934 TAILQ_REMOVE(&head->sol_comp, so, so_list); 935 if (flags & ACCEPT4_INHERIT) 936 so->so_state |= (head->so_state & SS_NBIO); 937 else 938 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0; 939 SOCK_UNLOCK(so); 940 sorele(head); 941 942 *ret = so; 943 return (0); 944 } 945 946 /* 947 * Evaluate the reference count and named references on a socket; if no 948 * references remain, free it. This should be called whenever a reference is 949 * released, such as in sorele(), but also when named reference flags are 950 * cleared in socket or protocol code. 951 * 952 * sofree() will free the socket if: 953 * 954 * - There are no outstanding file descriptor references or related consumers 955 * (so_count == 0). 956 * 957 * - The socket has been closed by user space, if ever open (SS_NOFDREF). 958 * 959 * - The protocol does not have an outstanding strong reference on the socket 960 * (SS_PROTOREF). 961 * 962 * - The socket is not in a completed connection queue, so a process has been 963 * notified that it is present. If it is removed, the user process may 964 * block in accept() despite select() saying the socket was ready. 965 */ 966 void 967 sofree(struct socket *so) 968 { 969 struct protosw *pr = so->so_proto; 970 971 SOCK_LOCK_ASSERT(so); 972 973 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 || 974 (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) { 975 SOCK_UNLOCK(so); 976 return; 977 } 978 979 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) { 980 struct socket *sol; 981 982 sol = so->so_listen; 983 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so)); 984 985 /* 986 * To solve race between close of a listening socket and 987 * a socket on its incomplete queue, we need to lock both. 988 * The order is first listening socket, then regular. 989 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this 990 * function and the listening socket are the only pointers 991 * to so. To preserve so and sol, we reference both and then 992 * relock. 993 * After relock the socket may not move to so_comp since it 994 * doesn't have PCB already, but it may be removed from 995 * so_incomp. If that happens, we share responsiblity on 996 * freeing the socket, but soclose() has already removed 997 * it from queue. 998 */ 999 soref(sol); 1000 soref(so); 1001 SOCK_UNLOCK(so); 1002 SOLISTEN_LOCK(sol); 1003 SOCK_LOCK(so); 1004 if (so->so_qstate == SQ_INCOMP) { 1005 KASSERT(so->so_listen == sol, 1006 ("%s: so %p migrated out of sol %p", 1007 __func__, so, sol)); 1008 TAILQ_REMOVE(&sol->sol_incomp, so, so_list); 1009 sol->sol_incqlen--; 1010 /* This is guarenteed not to be the last. */ 1011 refcount_release(&sol->so_count); 1012 so->so_qstate = SQ_NONE; 1013 so->so_listen = NULL; 1014 } else 1015 KASSERT(so->so_listen == NULL, 1016 ("%s: so %p not on (in)comp with so_listen", 1017 __func__, so)); 1018 sorele(sol); 1019 KASSERT(so->so_count == 1, 1020 ("%s: so %p count %u", __func__, so, so->so_count)); 1021 so->so_count = 0; 1022 } 1023 if (SOLISTENING(so)) 1024 so->so_error = ECONNABORTED; 1025 SOCK_UNLOCK(so); 1026 1027 VNET_SO_ASSERT(so); 1028 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 1029 (*pr->pr_domain->dom_dispose)(so); 1030 if (pr->pr_usrreqs->pru_detach != NULL) 1031 (*pr->pr_usrreqs->pru_detach)(so); 1032 1033 /* 1034 * From this point on, we assume that no other references to this 1035 * socket exist anywhere else in the stack. Therefore, no locks need 1036 * to be acquired or held. 1037 * 1038 * We used to do a lot of socket buffer and socket locking here, as 1039 * well as invoke sorflush() and perform wakeups. The direct call to 1040 * dom_dispose() and sbrelease_internal() are an inlining of what was 1041 * necessary from sorflush(). 1042 * 1043 * Notice that the socket buffer and kqueue state are torn down 1044 * before calling pru_detach. This means that protocols shold not 1045 * assume they can perform socket wakeups, etc, in their detach code. 1046 */ 1047 if (!SOLISTENING(so)) { 1048 sbdestroy(&so->so_snd, so); 1049 sbdestroy(&so->so_rcv, so); 1050 } 1051 seldrain(&so->so_rdsel); 1052 seldrain(&so->so_wrsel); 1053 knlist_destroy(&so->so_rdsel.si_note); 1054 knlist_destroy(&so->so_wrsel.si_note); 1055 sodealloc(so); 1056 } 1057 1058 /* 1059 * Close a socket on last file table reference removal. Initiate disconnect 1060 * if connected. Free socket when disconnect complete. 1061 * 1062 * This function will sorele() the socket. Note that soclose() may be called 1063 * prior to the ref count reaching zero. The actual socket structure will 1064 * not be freed until the ref count reaches zero. 1065 */ 1066 int 1067 soclose(struct socket *so) 1068 { 1069 struct accept_queue lqueue; 1070 bool listening; 1071 int error = 0; 1072 1073 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter")); 1074 1075 CURVNET_SET(so->so_vnet); 1076 funsetown(&so->so_sigio); 1077 if (so->so_state & SS_ISCONNECTED) { 1078 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 1079 error = sodisconnect(so); 1080 if (error) { 1081 if (error == ENOTCONN) 1082 error = 0; 1083 goto drop; 1084 } 1085 } 1086 if (so->so_options & SO_LINGER) { 1087 if ((so->so_state & SS_ISDISCONNECTING) && 1088 (so->so_state & SS_NBIO)) 1089 goto drop; 1090 while (so->so_state & SS_ISCONNECTED) { 1091 error = tsleep(&so->so_timeo, 1092 PSOCK | PCATCH, "soclos", 1093 so->so_linger * hz); 1094 if (error) 1095 break; 1096 } 1097 } 1098 } 1099 1100 drop: 1101 if (so->so_proto->pr_usrreqs->pru_close != NULL) 1102 (*so->so_proto->pr_usrreqs->pru_close)(so); 1103 1104 SOCK_LOCK(so); 1105 if ((listening = (so->so_options & SO_ACCEPTCONN))) { 1106 struct socket *sp; 1107 1108 TAILQ_INIT(&lqueue); 1109 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list); 1110 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list); 1111 1112 so->sol_qlen = so->sol_incqlen = 0; 1113 1114 TAILQ_FOREACH(sp, &lqueue, so_list) { 1115 SOCK_LOCK(sp); 1116 sp->so_qstate = SQ_NONE; 1117 sp->so_listen = NULL; 1118 SOCK_UNLOCK(sp); 1119 /* Guaranteed not to be the last. */ 1120 refcount_release(&so->so_count); 1121 } 1122 } 1123 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF")); 1124 so->so_state |= SS_NOFDREF; 1125 sorele(so); 1126 if (listening) { 1127 struct socket *sp; 1128 1129 TAILQ_FOREACH(sp, &lqueue, so_list) { 1130 SOCK_LOCK(sp); 1131 if (sp->so_count == 0) { 1132 SOCK_UNLOCK(sp); 1133 soabort(sp); 1134 } else 1135 /* sp is now in sofree() */ 1136 SOCK_UNLOCK(sp); 1137 } 1138 } 1139 CURVNET_RESTORE(); 1140 return (error); 1141 } 1142 1143 /* 1144 * soabort() is used to abruptly tear down a connection, such as when a 1145 * resource limit is reached (listen queue depth exceeded), or if a listen 1146 * socket is closed while there are sockets waiting to be accepted. 1147 * 1148 * This interface is tricky, because it is called on an unreferenced socket, 1149 * and must be called only by a thread that has actually removed the socket 1150 * from the listen queue it was on, or races with other threads are risked. 1151 * 1152 * This interface will call into the protocol code, so must not be called 1153 * with any socket locks held. Protocols do call it while holding their own 1154 * recursible protocol mutexes, but this is something that should be subject 1155 * to review in the future. 1156 */ 1157 void 1158 soabort(struct socket *so) 1159 { 1160 1161 /* 1162 * In as much as is possible, assert that no references to this 1163 * socket are held. This is not quite the same as asserting that the 1164 * current thread is responsible for arranging for no references, but 1165 * is as close as we can get for now. 1166 */ 1167 KASSERT(so->so_count == 0, ("soabort: so_count")); 1168 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF")); 1169 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF")); 1170 KASSERT(so->so_qstate == SQ_NONE, ("soabort: !SQ_NONE")); 1171 VNET_SO_ASSERT(so); 1172 1173 if (so->so_proto->pr_usrreqs->pru_abort != NULL) 1174 (*so->so_proto->pr_usrreqs->pru_abort)(so); 1175 SOCK_LOCK(so); 1176 sofree(so); 1177 } 1178 1179 int 1180 soaccept(struct socket *so, struct sockaddr **nam) 1181 { 1182 int error; 1183 1184 SOCK_LOCK(so); 1185 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF")); 1186 so->so_state &= ~SS_NOFDREF; 1187 SOCK_UNLOCK(so); 1188 1189 CURVNET_SET(so->so_vnet); 1190 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); 1191 CURVNET_RESTORE(); 1192 return (error); 1193 } 1194 1195 int 1196 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 1197 { 1198 1199 return (soconnectat(AT_FDCWD, so, nam, td)); 1200 } 1201 1202 int 1203 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 1204 { 1205 int error; 1206 1207 if (so->so_options & SO_ACCEPTCONN) 1208 return (EOPNOTSUPP); 1209 1210 CURVNET_SET(so->so_vnet); 1211 /* 1212 * If protocol is connection-based, can only connect once. 1213 * Otherwise, if connected, try to disconnect first. This allows 1214 * user to disconnect by connecting to, e.g., a null address. 1215 */ 1216 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 1217 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 1218 (error = sodisconnect(so)))) { 1219 error = EISCONN; 1220 } else { 1221 /* 1222 * Prevent accumulated error from previous connection from 1223 * biting us. 1224 */ 1225 so->so_error = 0; 1226 if (fd == AT_FDCWD) { 1227 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, 1228 nam, td); 1229 } else { 1230 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd, 1231 so, nam, td); 1232 } 1233 } 1234 CURVNET_RESTORE(); 1235 1236 return (error); 1237 } 1238 1239 int 1240 soconnect2(struct socket *so1, struct socket *so2) 1241 { 1242 int error; 1243 1244 CURVNET_SET(so1->so_vnet); 1245 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); 1246 CURVNET_RESTORE(); 1247 return (error); 1248 } 1249 1250 int 1251 sodisconnect(struct socket *so) 1252 { 1253 int error; 1254 1255 if ((so->so_state & SS_ISCONNECTED) == 0) 1256 return (ENOTCONN); 1257 if (so->so_state & SS_ISDISCONNECTING) 1258 return (EALREADY); 1259 VNET_SO_ASSERT(so); 1260 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 1261 return (error); 1262 } 1263 1264 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) 1265 1266 int 1267 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1268 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1269 { 1270 long space; 1271 ssize_t resid; 1272 int clen = 0, error, dontroute; 1273 1274 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 1275 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 1276 ("sosend_dgram: !PR_ATOMIC")); 1277 1278 if (uio != NULL) 1279 resid = uio->uio_resid; 1280 else 1281 resid = top->m_pkthdr.len; 1282 /* 1283 * In theory resid should be unsigned. However, space must be 1284 * signed, as it might be less than 0 if we over-committed, and we 1285 * must use a signed comparison of space and resid. On the other 1286 * hand, a negative resid causes us to loop sending 0-length 1287 * segments to the protocol. 1288 */ 1289 if (resid < 0) { 1290 error = EINVAL; 1291 goto out; 1292 } 1293 1294 dontroute = 1295 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 1296 if (td != NULL) 1297 td->td_ru.ru_msgsnd++; 1298 if (control != NULL) 1299 clen = control->m_len; 1300 1301 SOCKBUF_LOCK(&so->so_snd); 1302 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1303 SOCKBUF_UNLOCK(&so->so_snd); 1304 error = EPIPE; 1305 goto out; 1306 } 1307 if (so->so_error) { 1308 error = so->so_error; 1309 so->so_error = 0; 1310 SOCKBUF_UNLOCK(&so->so_snd); 1311 goto out; 1312 } 1313 if ((so->so_state & SS_ISCONNECTED) == 0) { 1314 /* 1315 * `sendto' and `sendmsg' is allowed on a connection-based 1316 * socket if it supports implied connect. Return ENOTCONN if 1317 * not connected and no address is supplied. 1318 */ 1319 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1320 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1321 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1322 !(resid == 0 && clen != 0)) { 1323 SOCKBUF_UNLOCK(&so->so_snd); 1324 error = ENOTCONN; 1325 goto out; 1326 } 1327 } else if (addr == NULL) { 1328 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1329 error = ENOTCONN; 1330 else 1331 error = EDESTADDRREQ; 1332 SOCKBUF_UNLOCK(&so->so_snd); 1333 goto out; 1334 } 1335 } 1336 1337 /* 1338 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1339 * problem and need fixing. 1340 */ 1341 space = sbspace(&so->so_snd); 1342 if (flags & MSG_OOB) 1343 space += 1024; 1344 space -= clen; 1345 SOCKBUF_UNLOCK(&so->so_snd); 1346 if (resid > space) { 1347 error = EMSGSIZE; 1348 goto out; 1349 } 1350 if (uio == NULL) { 1351 resid = 0; 1352 if (flags & MSG_EOR) 1353 top->m_flags |= M_EOR; 1354 } else { 1355 /* 1356 * Copy the data from userland into a mbuf chain. 1357 * If no data is to be copied in, a single empty mbuf 1358 * is returned. 1359 */ 1360 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1361 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1362 if (top == NULL) { 1363 error = EFAULT; /* only possible error */ 1364 goto out; 1365 } 1366 space -= resid - uio->uio_resid; 1367 resid = uio->uio_resid; 1368 } 1369 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1370 /* 1371 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1372 * than with. 1373 */ 1374 if (dontroute) { 1375 SOCK_LOCK(so); 1376 so->so_options |= SO_DONTROUTE; 1377 SOCK_UNLOCK(so); 1378 } 1379 /* 1380 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1381 * of date. We could have received a reset packet in an interrupt or 1382 * maybe we slept while doing page faults in uiomove() etc. We could 1383 * probably recheck again inside the locking protection here, but 1384 * there are probably other places that this also happens. We must 1385 * rethink this. 1386 */ 1387 VNET_SO_ASSERT(so); 1388 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1389 (flags & MSG_OOB) ? PRUS_OOB : 1390 /* 1391 * If the user set MSG_EOF, the protocol understands this flag and 1392 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1393 */ 1394 ((flags & MSG_EOF) && 1395 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1396 (resid <= 0)) ? 1397 PRUS_EOF : 1398 /* If there is more to send set PRUS_MORETOCOME */ 1399 (flags & MSG_MORETOCOME) || 1400 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1401 top, addr, control, td); 1402 if (dontroute) { 1403 SOCK_LOCK(so); 1404 so->so_options &= ~SO_DONTROUTE; 1405 SOCK_UNLOCK(so); 1406 } 1407 clen = 0; 1408 control = NULL; 1409 top = NULL; 1410 out: 1411 if (top != NULL) 1412 m_freem(top); 1413 if (control != NULL) 1414 m_freem(control); 1415 return (error); 1416 } 1417 1418 /* 1419 * Send on a socket. If send must go all at once and message is larger than 1420 * send buffering, then hard error. Lock against other senders. If must go 1421 * all at once and not enough room now, then inform user that this would 1422 * block and do nothing. Otherwise, if nonblocking, send as much as 1423 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1424 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1425 * in mbuf chain must be small enough to send all at once. 1426 * 1427 * Returns nonzero on error, timeout or signal; callers must check for short 1428 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1429 * on return. 1430 */ 1431 int 1432 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1433 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1434 { 1435 long space; 1436 ssize_t resid; 1437 int clen = 0, error, dontroute; 1438 int atomic = sosendallatonce(so) || top; 1439 1440 if (uio != NULL) 1441 resid = uio->uio_resid; 1442 else 1443 resid = top->m_pkthdr.len; 1444 /* 1445 * In theory resid should be unsigned. However, space must be 1446 * signed, as it might be less than 0 if we over-committed, and we 1447 * must use a signed comparison of space and resid. On the other 1448 * hand, a negative resid causes us to loop sending 0-length 1449 * segments to the protocol. 1450 * 1451 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1452 * type sockets since that's an error. 1453 */ 1454 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1455 error = EINVAL; 1456 goto out; 1457 } 1458 1459 dontroute = 1460 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1461 (so->so_proto->pr_flags & PR_ATOMIC); 1462 if (td != NULL) 1463 td->td_ru.ru_msgsnd++; 1464 if (control != NULL) 1465 clen = control->m_len; 1466 1467 error = sblock(&so->so_snd, SBLOCKWAIT(flags)); 1468 if (error) 1469 goto out; 1470 1471 restart: 1472 do { 1473 SOCKBUF_LOCK(&so->so_snd); 1474 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1475 SOCKBUF_UNLOCK(&so->so_snd); 1476 error = EPIPE; 1477 goto release; 1478 } 1479 if (so->so_error) { 1480 error = so->so_error; 1481 so->so_error = 0; 1482 SOCKBUF_UNLOCK(&so->so_snd); 1483 goto release; 1484 } 1485 if ((so->so_state & SS_ISCONNECTED) == 0) { 1486 /* 1487 * `sendto' and `sendmsg' is allowed on a connection- 1488 * based socket if it supports implied connect. 1489 * Return ENOTCONN if not connected and no address is 1490 * supplied. 1491 */ 1492 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1493 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1494 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1495 !(resid == 0 && clen != 0)) { 1496 SOCKBUF_UNLOCK(&so->so_snd); 1497 error = ENOTCONN; 1498 goto release; 1499 } 1500 } else if (addr == NULL) { 1501 SOCKBUF_UNLOCK(&so->so_snd); 1502 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1503 error = ENOTCONN; 1504 else 1505 error = EDESTADDRREQ; 1506 goto release; 1507 } 1508 } 1509 space = sbspace(&so->so_snd); 1510 if (flags & MSG_OOB) 1511 space += 1024; 1512 if ((atomic && resid > so->so_snd.sb_hiwat) || 1513 clen > so->so_snd.sb_hiwat) { 1514 SOCKBUF_UNLOCK(&so->so_snd); 1515 error = EMSGSIZE; 1516 goto release; 1517 } 1518 if (space < resid + clen && 1519 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1520 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) { 1521 SOCKBUF_UNLOCK(&so->so_snd); 1522 error = EWOULDBLOCK; 1523 goto release; 1524 } 1525 error = sbwait(&so->so_snd); 1526 SOCKBUF_UNLOCK(&so->so_snd); 1527 if (error) 1528 goto release; 1529 goto restart; 1530 } 1531 SOCKBUF_UNLOCK(&so->so_snd); 1532 space -= clen; 1533 do { 1534 if (uio == NULL) { 1535 resid = 0; 1536 if (flags & MSG_EOR) 1537 top->m_flags |= M_EOR; 1538 } else { 1539 /* 1540 * Copy the data from userland into a mbuf 1541 * chain. If resid is 0, which can happen 1542 * only if we have control to send, then 1543 * a single empty mbuf is returned. This 1544 * is a workaround to prevent protocol send 1545 * methods to panic. 1546 */ 1547 top = m_uiotombuf(uio, M_WAITOK, space, 1548 (atomic ? max_hdr : 0), 1549 (atomic ? M_PKTHDR : 0) | 1550 ((flags & MSG_EOR) ? M_EOR : 0)); 1551 if (top == NULL) { 1552 error = EFAULT; /* only possible error */ 1553 goto release; 1554 } 1555 space -= resid - uio->uio_resid; 1556 resid = uio->uio_resid; 1557 } 1558 if (dontroute) { 1559 SOCK_LOCK(so); 1560 so->so_options |= SO_DONTROUTE; 1561 SOCK_UNLOCK(so); 1562 } 1563 /* 1564 * XXX all the SBS_CANTSENDMORE checks previously 1565 * done could be out of date. We could have received 1566 * a reset packet in an interrupt or maybe we slept 1567 * while doing page faults in uiomove() etc. We 1568 * could probably recheck again inside the locking 1569 * protection here, but there are probably other 1570 * places that this also happens. We must rethink 1571 * this. 1572 */ 1573 VNET_SO_ASSERT(so); 1574 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1575 (flags & MSG_OOB) ? PRUS_OOB : 1576 /* 1577 * If the user set MSG_EOF, the protocol understands 1578 * this flag and nothing left to send then use 1579 * PRU_SEND_EOF instead of PRU_SEND. 1580 */ 1581 ((flags & MSG_EOF) && 1582 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1583 (resid <= 0)) ? 1584 PRUS_EOF : 1585 /* If there is more to send set PRUS_MORETOCOME. */ 1586 (flags & MSG_MORETOCOME) || 1587 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1588 top, addr, control, td); 1589 if (dontroute) { 1590 SOCK_LOCK(so); 1591 so->so_options &= ~SO_DONTROUTE; 1592 SOCK_UNLOCK(so); 1593 } 1594 clen = 0; 1595 control = NULL; 1596 top = NULL; 1597 if (error) 1598 goto release; 1599 } while (resid && space > 0); 1600 } while (resid); 1601 1602 release: 1603 sbunlock(&so->so_snd); 1604 out: 1605 if (top != NULL) 1606 m_freem(top); 1607 if (control != NULL) 1608 m_freem(control); 1609 return (error); 1610 } 1611 1612 int 1613 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1614 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1615 { 1616 int error; 1617 1618 CURVNET_SET(so->so_vnet); 1619 if (!SOLISTENING(so)) 1620 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, 1621 top, control, flags, td); 1622 else { 1623 m_freem(top); 1624 m_freem(control); 1625 error = ENOTCONN; 1626 } 1627 CURVNET_RESTORE(); 1628 return (error); 1629 } 1630 1631 /* 1632 * The part of soreceive() that implements reading non-inline out-of-band 1633 * data from a socket. For more complete comments, see soreceive(), from 1634 * which this code originated. 1635 * 1636 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1637 * unable to return an mbuf chain to the caller. 1638 */ 1639 static int 1640 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1641 { 1642 struct protosw *pr = so->so_proto; 1643 struct mbuf *m; 1644 int error; 1645 1646 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1647 VNET_SO_ASSERT(so); 1648 1649 m = m_get(M_WAITOK, MT_DATA); 1650 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 1651 if (error) 1652 goto bad; 1653 do { 1654 error = uiomove(mtod(m, void *), 1655 (int) min(uio->uio_resid, m->m_len), uio); 1656 m = m_free(m); 1657 } while (uio->uio_resid && error == 0 && m); 1658 bad: 1659 if (m != NULL) 1660 m_freem(m); 1661 return (error); 1662 } 1663 1664 /* 1665 * Following replacement or removal of the first mbuf on the first mbuf chain 1666 * of a socket buffer, push necessary state changes back into the socket 1667 * buffer so that other consumers see the values consistently. 'nextrecord' 1668 * is the callers locally stored value of the original value of 1669 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1670 * NOTE: 'nextrecord' may be NULL. 1671 */ 1672 static __inline void 1673 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1674 { 1675 1676 SOCKBUF_LOCK_ASSERT(sb); 1677 /* 1678 * First, update for the new value of nextrecord. If necessary, make 1679 * it the first record. 1680 */ 1681 if (sb->sb_mb != NULL) 1682 sb->sb_mb->m_nextpkt = nextrecord; 1683 else 1684 sb->sb_mb = nextrecord; 1685 1686 /* 1687 * Now update any dependent socket buffer fields to reflect the new 1688 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1689 * addition of a second clause that takes care of the case where 1690 * sb_mb has been updated, but remains the last record. 1691 */ 1692 if (sb->sb_mb == NULL) { 1693 sb->sb_mbtail = NULL; 1694 sb->sb_lastrecord = NULL; 1695 } else if (sb->sb_mb->m_nextpkt == NULL) 1696 sb->sb_lastrecord = sb->sb_mb; 1697 } 1698 1699 /* 1700 * Implement receive operations on a socket. We depend on the way that 1701 * records are added to the sockbuf by sbappend. In particular, each record 1702 * (mbufs linked through m_next) must begin with an address if the protocol 1703 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1704 * data, and then zero or more mbufs of data. In order to allow parallelism 1705 * between network receive and copying to user space, as well as avoid 1706 * sleeping with a mutex held, we release the socket buffer mutex during the 1707 * user space copy. Although the sockbuf is locked, new data may still be 1708 * appended, and thus we must maintain consistency of the sockbuf during that 1709 * time. 1710 * 1711 * The caller may receive the data as a single mbuf chain by supplying an 1712 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1713 * the count in uio_resid. 1714 */ 1715 int 1716 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1717 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1718 { 1719 struct mbuf *m, **mp; 1720 int flags, error, offset; 1721 ssize_t len; 1722 struct protosw *pr = so->so_proto; 1723 struct mbuf *nextrecord; 1724 int moff, type = 0; 1725 ssize_t orig_resid = uio->uio_resid; 1726 1727 mp = mp0; 1728 if (psa != NULL) 1729 *psa = NULL; 1730 if (controlp != NULL) 1731 *controlp = NULL; 1732 if (flagsp != NULL) 1733 flags = *flagsp &~ MSG_EOR; 1734 else 1735 flags = 0; 1736 if (flags & MSG_OOB) 1737 return (soreceive_rcvoob(so, uio, flags)); 1738 if (mp != NULL) 1739 *mp = NULL; 1740 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 1741 && uio->uio_resid) { 1742 VNET_SO_ASSERT(so); 1743 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 1744 } 1745 1746 error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); 1747 if (error) 1748 return (error); 1749 1750 restart: 1751 SOCKBUF_LOCK(&so->so_rcv); 1752 m = so->so_rcv.sb_mb; 1753 /* 1754 * If we have less data than requested, block awaiting more (subject 1755 * to any timeout) if: 1756 * 1. the current count is less than the low water mark, or 1757 * 2. MSG_DONTWAIT is not set 1758 */ 1759 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1760 sbavail(&so->so_rcv) < uio->uio_resid) && 1761 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && 1762 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 1763 KASSERT(m != NULL || !sbavail(&so->so_rcv), 1764 ("receive: m == %p sbavail == %u", 1765 m, sbavail(&so->so_rcv))); 1766 if (so->so_error) { 1767 if (m != NULL) 1768 goto dontblock; 1769 error = so->so_error; 1770 if ((flags & MSG_PEEK) == 0) 1771 so->so_error = 0; 1772 SOCKBUF_UNLOCK(&so->so_rcv); 1773 goto release; 1774 } 1775 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1776 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1777 if (m == NULL) { 1778 SOCKBUF_UNLOCK(&so->so_rcv); 1779 goto release; 1780 } else 1781 goto dontblock; 1782 } 1783 for (; m != NULL; m = m->m_next) 1784 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1785 m = so->so_rcv.sb_mb; 1786 goto dontblock; 1787 } 1788 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1789 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1790 SOCKBUF_UNLOCK(&so->so_rcv); 1791 error = ENOTCONN; 1792 goto release; 1793 } 1794 if (uio->uio_resid == 0) { 1795 SOCKBUF_UNLOCK(&so->so_rcv); 1796 goto release; 1797 } 1798 if ((so->so_state & SS_NBIO) || 1799 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1800 SOCKBUF_UNLOCK(&so->so_rcv); 1801 error = EWOULDBLOCK; 1802 goto release; 1803 } 1804 SBLASTRECORDCHK(&so->so_rcv); 1805 SBLASTMBUFCHK(&so->so_rcv); 1806 error = sbwait(&so->so_rcv); 1807 SOCKBUF_UNLOCK(&so->so_rcv); 1808 if (error) 1809 goto release; 1810 goto restart; 1811 } 1812 dontblock: 1813 /* 1814 * From this point onward, we maintain 'nextrecord' as a cache of the 1815 * pointer to the next record in the socket buffer. We must keep the 1816 * various socket buffer pointers and local stack versions of the 1817 * pointers in sync, pushing out modifications before dropping the 1818 * socket buffer mutex, and re-reading them when picking it up. 1819 * 1820 * Otherwise, we will race with the network stack appending new data 1821 * or records onto the socket buffer by using inconsistent/stale 1822 * versions of the field, possibly resulting in socket buffer 1823 * corruption. 1824 * 1825 * By holding the high-level sblock(), we prevent simultaneous 1826 * readers from pulling off the front of the socket buffer. 1827 */ 1828 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1829 if (uio->uio_td) 1830 uio->uio_td->td_ru.ru_msgrcv++; 1831 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 1832 SBLASTRECORDCHK(&so->so_rcv); 1833 SBLASTMBUFCHK(&so->so_rcv); 1834 nextrecord = m->m_nextpkt; 1835 if (pr->pr_flags & PR_ADDR) { 1836 KASSERT(m->m_type == MT_SONAME, 1837 ("m->m_type == %d", m->m_type)); 1838 orig_resid = 0; 1839 if (psa != NULL) 1840 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 1841 M_NOWAIT); 1842 if (flags & MSG_PEEK) { 1843 m = m->m_next; 1844 } else { 1845 sbfree(&so->so_rcv, m); 1846 so->so_rcv.sb_mb = m_free(m); 1847 m = so->so_rcv.sb_mb; 1848 sockbuf_pushsync(&so->so_rcv, nextrecord); 1849 } 1850 } 1851 1852 /* 1853 * Process one or more MT_CONTROL mbufs present before any data mbufs 1854 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 1855 * just copy the data; if !MSG_PEEK, we call into the protocol to 1856 * perform externalization (or freeing if controlp == NULL). 1857 */ 1858 if (m != NULL && m->m_type == MT_CONTROL) { 1859 struct mbuf *cm = NULL, *cmn; 1860 struct mbuf **cme = &cm; 1861 1862 do { 1863 if (flags & MSG_PEEK) { 1864 if (controlp != NULL) { 1865 *controlp = m_copym(m, 0, m->m_len, 1866 M_NOWAIT); 1867 controlp = &(*controlp)->m_next; 1868 } 1869 m = m->m_next; 1870 } else { 1871 sbfree(&so->so_rcv, m); 1872 so->so_rcv.sb_mb = m->m_next; 1873 m->m_next = NULL; 1874 *cme = m; 1875 cme = &(*cme)->m_next; 1876 m = so->so_rcv.sb_mb; 1877 } 1878 } while (m != NULL && m->m_type == MT_CONTROL); 1879 if ((flags & MSG_PEEK) == 0) 1880 sockbuf_pushsync(&so->so_rcv, nextrecord); 1881 while (cm != NULL) { 1882 cmn = cm->m_next; 1883 cm->m_next = NULL; 1884 if (pr->pr_domain->dom_externalize != NULL) { 1885 SOCKBUF_UNLOCK(&so->so_rcv); 1886 VNET_SO_ASSERT(so); 1887 error = (*pr->pr_domain->dom_externalize) 1888 (cm, controlp, flags); 1889 SOCKBUF_LOCK(&so->so_rcv); 1890 } else if (controlp != NULL) 1891 *controlp = cm; 1892 else 1893 m_freem(cm); 1894 if (controlp != NULL) { 1895 orig_resid = 0; 1896 while (*controlp != NULL) 1897 controlp = &(*controlp)->m_next; 1898 } 1899 cm = cmn; 1900 } 1901 if (m != NULL) 1902 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 1903 else 1904 nextrecord = so->so_rcv.sb_mb; 1905 orig_resid = 0; 1906 } 1907 if (m != NULL) { 1908 if ((flags & MSG_PEEK) == 0) { 1909 KASSERT(m->m_nextpkt == nextrecord, 1910 ("soreceive: post-control, nextrecord !sync")); 1911 if (nextrecord == NULL) { 1912 KASSERT(so->so_rcv.sb_mb == m, 1913 ("soreceive: post-control, sb_mb!=m")); 1914 KASSERT(so->so_rcv.sb_lastrecord == m, 1915 ("soreceive: post-control, lastrecord!=m")); 1916 } 1917 } 1918 type = m->m_type; 1919 if (type == MT_OOBDATA) 1920 flags |= MSG_OOB; 1921 } else { 1922 if ((flags & MSG_PEEK) == 0) { 1923 KASSERT(so->so_rcv.sb_mb == nextrecord, 1924 ("soreceive: sb_mb != nextrecord")); 1925 if (so->so_rcv.sb_mb == NULL) { 1926 KASSERT(so->so_rcv.sb_lastrecord == NULL, 1927 ("soreceive: sb_lastercord != NULL")); 1928 } 1929 } 1930 } 1931 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1932 SBLASTRECORDCHK(&so->so_rcv); 1933 SBLASTMBUFCHK(&so->so_rcv); 1934 1935 /* 1936 * Now continue to read any data mbufs off of the head of the socket 1937 * buffer until the read request is satisfied. Note that 'type' is 1938 * used to store the type of any mbuf reads that have happened so far 1939 * such that soreceive() can stop reading if the type changes, which 1940 * causes soreceive() to return only one of regular data and inline 1941 * out-of-band data in a single socket receive operation. 1942 */ 1943 moff = 0; 1944 offset = 0; 1945 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 1946 && error == 0) { 1947 /* 1948 * If the type of mbuf has changed since the last mbuf 1949 * examined ('type'), end the receive operation. 1950 */ 1951 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1952 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 1953 if (type != m->m_type) 1954 break; 1955 } else if (type == MT_OOBDATA) 1956 break; 1957 else 1958 KASSERT(m->m_type == MT_DATA, 1959 ("m->m_type == %d", m->m_type)); 1960 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 1961 len = uio->uio_resid; 1962 if (so->so_oobmark && len > so->so_oobmark - offset) 1963 len = so->so_oobmark - offset; 1964 if (len > m->m_len - moff) 1965 len = m->m_len - moff; 1966 /* 1967 * If mp is set, just pass back the mbufs. Otherwise copy 1968 * them out via the uio, then free. Sockbuf must be 1969 * consistent here (points to current mbuf, it points to next 1970 * record) when we drop priority; we must note any additions 1971 * to the sockbuf when we block interrupts again. 1972 */ 1973 if (mp == NULL) { 1974 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1975 SBLASTRECORDCHK(&so->so_rcv); 1976 SBLASTMBUFCHK(&so->so_rcv); 1977 SOCKBUF_UNLOCK(&so->so_rcv); 1978 error = uiomove(mtod(m, char *) + moff, (int)len, uio); 1979 SOCKBUF_LOCK(&so->so_rcv); 1980 if (error) { 1981 /* 1982 * The MT_SONAME mbuf has already been removed 1983 * from the record, so it is necessary to 1984 * remove the data mbufs, if any, to preserve 1985 * the invariant in the case of PR_ADDR that 1986 * requires MT_SONAME mbufs at the head of 1987 * each record. 1988 */ 1989 if (pr->pr_flags & PR_ATOMIC && 1990 ((flags & MSG_PEEK) == 0)) 1991 (void)sbdroprecord_locked(&so->so_rcv); 1992 SOCKBUF_UNLOCK(&so->so_rcv); 1993 goto release; 1994 } 1995 } else 1996 uio->uio_resid -= len; 1997 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1998 if (len == m->m_len - moff) { 1999 if (m->m_flags & M_EOR) 2000 flags |= MSG_EOR; 2001 if (flags & MSG_PEEK) { 2002 m = m->m_next; 2003 moff = 0; 2004 } else { 2005 nextrecord = m->m_nextpkt; 2006 sbfree(&so->so_rcv, m); 2007 if (mp != NULL) { 2008 m->m_nextpkt = NULL; 2009 *mp = m; 2010 mp = &m->m_next; 2011 so->so_rcv.sb_mb = m = m->m_next; 2012 *mp = NULL; 2013 } else { 2014 so->so_rcv.sb_mb = m_free(m); 2015 m = so->so_rcv.sb_mb; 2016 } 2017 sockbuf_pushsync(&so->so_rcv, nextrecord); 2018 SBLASTRECORDCHK(&so->so_rcv); 2019 SBLASTMBUFCHK(&so->so_rcv); 2020 } 2021 } else { 2022 if (flags & MSG_PEEK) 2023 moff += len; 2024 else { 2025 if (mp != NULL) { 2026 if (flags & MSG_DONTWAIT) { 2027 *mp = m_copym(m, 0, len, 2028 M_NOWAIT); 2029 if (*mp == NULL) { 2030 /* 2031 * m_copym() couldn't 2032 * allocate an mbuf. 2033 * Adjust uio_resid back 2034 * (it was adjusted 2035 * down by len bytes, 2036 * which we didn't end 2037 * up "copying" over). 2038 */ 2039 uio->uio_resid += len; 2040 break; 2041 } 2042 } else { 2043 SOCKBUF_UNLOCK(&so->so_rcv); 2044 *mp = m_copym(m, 0, len, 2045 M_WAITOK); 2046 SOCKBUF_LOCK(&so->so_rcv); 2047 } 2048 } 2049 sbcut_locked(&so->so_rcv, len); 2050 } 2051 } 2052 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2053 if (so->so_oobmark) { 2054 if ((flags & MSG_PEEK) == 0) { 2055 so->so_oobmark -= len; 2056 if (so->so_oobmark == 0) { 2057 so->so_rcv.sb_state |= SBS_RCVATMARK; 2058 break; 2059 } 2060 } else { 2061 offset += len; 2062 if (offset == so->so_oobmark) 2063 break; 2064 } 2065 } 2066 if (flags & MSG_EOR) 2067 break; 2068 /* 2069 * If the MSG_WAITALL flag is set (for non-atomic socket), we 2070 * must not quit until "uio->uio_resid == 0" or an error 2071 * termination. If a signal/timeout occurs, return with a 2072 * short count but without error. Keep sockbuf locked 2073 * against other readers. 2074 */ 2075 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 2076 !sosendallatonce(so) && nextrecord == NULL) { 2077 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2078 if (so->so_error || 2079 so->so_rcv.sb_state & SBS_CANTRCVMORE) 2080 break; 2081 /* 2082 * Notify the protocol that some data has been 2083 * drained before blocking. 2084 */ 2085 if (pr->pr_flags & PR_WANTRCVD) { 2086 SOCKBUF_UNLOCK(&so->so_rcv); 2087 VNET_SO_ASSERT(so); 2088 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2089 SOCKBUF_LOCK(&so->so_rcv); 2090 } 2091 SBLASTRECORDCHK(&so->so_rcv); 2092 SBLASTMBUFCHK(&so->so_rcv); 2093 /* 2094 * We could receive some data while was notifying 2095 * the protocol. Skip blocking in this case. 2096 */ 2097 if (so->so_rcv.sb_mb == NULL) { 2098 error = sbwait(&so->so_rcv); 2099 if (error) { 2100 SOCKBUF_UNLOCK(&so->so_rcv); 2101 goto release; 2102 } 2103 } 2104 m = so->so_rcv.sb_mb; 2105 if (m != NULL) 2106 nextrecord = m->m_nextpkt; 2107 } 2108 } 2109 2110 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2111 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 2112 flags |= MSG_TRUNC; 2113 if ((flags & MSG_PEEK) == 0) 2114 (void) sbdroprecord_locked(&so->so_rcv); 2115 } 2116 if ((flags & MSG_PEEK) == 0) { 2117 if (m == NULL) { 2118 /* 2119 * First part is an inline SB_EMPTY_FIXUP(). Second 2120 * part makes sure sb_lastrecord is up-to-date if 2121 * there is still data in the socket buffer. 2122 */ 2123 so->so_rcv.sb_mb = nextrecord; 2124 if (so->so_rcv.sb_mb == NULL) { 2125 so->so_rcv.sb_mbtail = NULL; 2126 so->so_rcv.sb_lastrecord = NULL; 2127 } else if (nextrecord->m_nextpkt == NULL) 2128 so->so_rcv.sb_lastrecord = nextrecord; 2129 } 2130 SBLASTRECORDCHK(&so->so_rcv); 2131 SBLASTMBUFCHK(&so->so_rcv); 2132 /* 2133 * If soreceive() is being done from the socket callback, 2134 * then don't need to generate ACK to peer to update window, 2135 * since ACK will be generated on return to TCP. 2136 */ 2137 if (!(flags & MSG_SOCALLBCK) && 2138 (pr->pr_flags & PR_WANTRCVD)) { 2139 SOCKBUF_UNLOCK(&so->so_rcv); 2140 VNET_SO_ASSERT(so); 2141 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 2142 SOCKBUF_LOCK(&so->so_rcv); 2143 } 2144 } 2145 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2146 if (orig_resid == uio->uio_resid && orig_resid && 2147 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 2148 SOCKBUF_UNLOCK(&so->so_rcv); 2149 goto restart; 2150 } 2151 SOCKBUF_UNLOCK(&so->so_rcv); 2152 2153 if (flagsp != NULL) 2154 *flagsp |= flags; 2155 release: 2156 sbunlock(&so->so_rcv); 2157 return (error); 2158 } 2159 2160 /* 2161 * Optimized version of soreceive() for stream (TCP) sockets. 2162 * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled. 2163 */ 2164 int 2165 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 2166 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2167 { 2168 int len = 0, error = 0, flags, oresid; 2169 struct sockbuf *sb; 2170 struct mbuf *m, *n = NULL; 2171 2172 /* We only do stream sockets. */ 2173 if (so->so_type != SOCK_STREAM) 2174 return (EINVAL); 2175 if (psa != NULL) 2176 *psa = NULL; 2177 if (controlp != NULL) 2178 return (EINVAL); 2179 if (flagsp != NULL) 2180 flags = *flagsp &~ MSG_EOR; 2181 else 2182 flags = 0; 2183 if (flags & MSG_OOB) 2184 return (soreceive_rcvoob(so, uio, flags)); 2185 if (mp0 != NULL) 2186 *mp0 = NULL; 2187 2188 sb = &so->so_rcv; 2189 2190 /* Prevent other readers from entering the socket. */ 2191 error = sblock(sb, SBLOCKWAIT(flags)); 2192 if (error) 2193 goto out; 2194 SOCKBUF_LOCK(sb); 2195 2196 /* Easy one, no space to copyout anything. */ 2197 if (uio->uio_resid == 0) { 2198 error = EINVAL; 2199 goto out; 2200 } 2201 oresid = uio->uio_resid; 2202 2203 /* We will never ever get anything unless we are or were connected. */ 2204 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 2205 error = ENOTCONN; 2206 goto out; 2207 } 2208 2209 restart: 2210 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2211 2212 /* Abort if socket has reported problems. */ 2213 if (so->so_error) { 2214 if (sbavail(sb) > 0) 2215 goto deliver; 2216 if (oresid > uio->uio_resid) 2217 goto out; 2218 error = so->so_error; 2219 if (!(flags & MSG_PEEK)) 2220 so->so_error = 0; 2221 goto out; 2222 } 2223 2224 /* Door is closed. Deliver what is left, if any. */ 2225 if (sb->sb_state & SBS_CANTRCVMORE) { 2226 if (sbavail(sb) > 0) 2227 goto deliver; 2228 else 2229 goto out; 2230 } 2231 2232 /* Socket buffer is empty and we shall not block. */ 2233 if (sbavail(sb) == 0 && 2234 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 2235 error = EAGAIN; 2236 goto out; 2237 } 2238 2239 /* Socket buffer got some data that we shall deliver now. */ 2240 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 2241 ((so->so_state & SS_NBIO) || 2242 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 2243 sbavail(sb) >= sb->sb_lowat || 2244 sbavail(sb) >= uio->uio_resid || 2245 sbavail(sb) >= sb->sb_hiwat) ) { 2246 goto deliver; 2247 } 2248 2249 /* On MSG_WAITALL we must wait until all data or error arrives. */ 2250 if ((flags & MSG_WAITALL) && 2251 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 2252 goto deliver; 2253 2254 /* 2255 * Wait and block until (more) data comes in. 2256 * NB: Drops the sockbuf lock during wait. 2257 */ 2258 error = sbwait(sb); 2259 if (error) 2260 goto out; 2261 goto restart; 2262 2263 deliver: 2264 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2265 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 2266 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 2267 2268 /* Statistics. */ 2269 if (uio->uio_td) 2270 uio->uio_td->td_ru.ru_msgrcv++; 2271 2272 /* Fill uio until full or current end of socket buffer is reached. */ 2273 len = min(uio->uio_resid, sbavail(sb)); 2274 if (mp0 != NULL) { 2275 /* Dequeue as many mbufs as possible. */ 2276 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 2277 if (*mp0 == NULL) 2278 *mp0 = sb->sb_mb; 2279 else 2280 m_cat(*mp0, sb->sb_mb); 2281 for (m = sb->sb_mb; 2282 m != NULL && m->m_len <= len; 2283 m = m->m_next) { 2284 KASSERT(!(m->m_flags & M_NOTAVAIL), 2285 ("%s: m %p not available", __func__, m)); 2286 len -= m->m_len; 2287 uio->uio_resid -= m->m_len; 2288 sbfree(sb, m); 2289 n = m; 2290 } 2291 n->m_next = NULL; 2292 sb->sb_mb = m; 2293 sb->sb_lastrecord = sb->sb_mb; 2294 if (sb->sb_mb == NULL) 2295 SB_EMPTY_FIXUP(sb); 2296 } 2297 /* Copy the remainder. */ 2298 if (len > 0) { 2299 KASSERT(sb->sb_mb != NULL, 2300 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2301 2302 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 2303 if (m == NULL) 2304 len = 0; /* Don't flush data from sockbuf. */ 2305 else 2306 uio->uio_resid -= len; 2307 if (*mp0 != NULL) 2308 m_cat(*mp0, m); 2309 else 2310 *mp0 = m; 2311 if (*mp0 == NULL) { 2312 error = ENOBUFS; 2313 goto out; 2314 } 2315 } 2316 } else { 2317 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2318 SOCKBUF_UNLOCK(sb); 2319 error = m_mbuftouio(uio, sb->sb_mb, len); 2320 SOCKBUF_LOCK(sb); 2321 if (error) 2322 goto out; 2323 } 2324 SBLASTRECORDCHK(sb); 2325 SBLASTMBUFCHK(sb); 2326 2327 /* 2328 * Remove the delivered data from the socket buffer unless we 2329 * were only peeking. 2330 */ 2331 if (!(flags & MSG_PEEK)) { 2332 if (len > 0) 2333 sbdrop_locked(sb, len); 2334 2335 /* Notify protocol that we drained some data. */ 2336 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2337 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2338 !(flags & MSG_SOCALLBCK))) { 2339 SOCKBUF_UNLOCK(sb); 2340 VNET_SO_ASSERT(so); 2341 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags); 2342 SOCKBUF_LOCK(sb); 2343 } 2344 } 2345 2346 /* 2347 * For MSG_WAITALL we may have to loop again and wait for 2348 * more data to come in. 2349 */ 2350 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2351 goto restart; 2352 out: 2353 SOCKBUF_LOCK_ASSERT(sb); 2354 SBLASTRECORDCHK(sb); 2355 SBLASTMBUFCHK(sb); 2356 SOCKBUF_UNLOCK(sb); 2357 sbunlock(sb); 2358 return (error); 2359 } 2360 2361 /* 2362 * Optimized version of soreceive() for simple datagram cases from userspace. 2363 * Unlike in the stream case, we're able to drop a datagram if copyout() 2364 * fails, and because we handle datagrams atomically, we don't need to use a 2365 * sleep lock to prevent I/O interlacing. 2366 */ 2367 int 2368 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2369 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2370 { 2371 struct mbuf *m, *m2; 2372 int flags, error; 2373 ssize_t len; 2374 struct protosw *pr = so->so_proto; 2375 struct mbuf *nextrecord; 2376 2377 if (psa != NULL) 2378 *psa = NULL; 2379 if (controlp != NULL) 2380 *controlp = NULL; 2381 if (flagsp != NULL) 2382 flags = *flagsp &~ MSG_EOR; 2383 else 2384 flags = 0; 2385 2386 /* 2387 * For any complicated cases, fall back to the full 2388 * soreceive_generic(). 2389 */ 2390 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB)) 2391 return (soreceive_generic(so, psa, uio, mp0, controlp, 2392 flagsp)); 2393 2394 /* 2395 * Enforce restrictions on use. 2396 */ 2397 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2398 ("soreceive_dgram: wantrcvd")); 2399 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2400 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2401 ("soreceive_dgram: SBS_RCVATMARK")); 2402 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2403 ("soreceive_dgram: P_CONNREQUIRED")); 2404 2405 /* 2406 * Loop blocking while waiting for a datagram. 2407 */ 2408 SOCKBUF_LOCK(&so->so_rcv); 2409 while ((m = so->so_rcv.sb_mb) == NULL) { 2410 KASSERT(sbavail(&so->so_rcv) == 0, 2411 ("soreceive_dgram: sb_mb NULL but sbavail %u", 2412 sbavail(&so->so_rcv))); 2413 if (so->so_error) { 2414 error = so->so_error; 2415 so->so_error = 0; 2416 SOCKBUF_UNLOCK(&so->so_rcv); 2417 return (error); 2418 } 2419 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2420 uio->uio_resid == 0) { 2421 SOCKBUF_UNLOCK(&so->so_rcv); 2422 return (0); 2423 } 2424 if ((so->so_state & SS_NBIO) || 2425 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2426 SOCKBUF_UNLOCK(&so->so_rcv); 2427 return (EWOULDBLOCK); 2428 } 2429 SBLASTRECORDCHK(&so->so_rcv); 2430 SBLASTMBUFCHK(&so->so_rcv); 2431 error = sbwait(&so->so_rcv); 2432 if (error) { 2433 SOCKBUF_UNLOCK(&so->so_rcv); 2434 return (error); 2435 } 2436 } 2437 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2438 2439 if (uio->uio_td) 2440 uio->uio_td->td_ru.ru_msgrcv++; 2441 SBLASTRECORDCHK(&so->so_rcv); 2442 SBLASTMBUFCHK(&so->so_rcv); 2443 nextrecord = m->m_nextpkt; 2444 if (nextrecord == NULL) { 2445 KASSERT(so->so_rcv.sb_lastrecord == m, 2446 ("soreceive_dgram: lastrecord != m")); 2447 } 2448 2449 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2450 ("soreceive_dgram: m_nextpkt != nextrecord")); 2451 2452 /* 2453 * Pull 'm' and its chain off the front of the packet queue. 2454 */ 2455 so->so_rcv.sb_mb = NULL; 2456 sockbuf_pushsync(&so->so_rcv, nextrecord); 2457 2458 /* 2459 * Walk 'm's chain and free that many bytes from the socket buffer. 2460 */ 2461 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2462 sbfree(&so->so_rcv, m2); 2463 2464 /* 2465 * Do a few last checks before we let go of the lock. 2466 */ 2467 SBLASTRECORDCHK(&so->so_rcv); 2468 SBLASTMBUFCHK(&so->so_rcv); 2469 SOCKBUF_UNLOCK(&so->so_rcv); 2470 2471 if (pr->pr_flags & PR_ADDR) { 2472 KASSERT(m->m_type == MT_SONAME, 2473 ("m->m_type == %d", m->m_type)); 2474 if (psa != NULL) 2475 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2476 M_NOWAIT); 2477 m = m_free(m); 2478 } 2479 if (m == NULL) { 2480 /* XXXRW: Can this happen? */ 2481 return (0); 2482 } 2483 2484 /* 2485 * Packet to copyout() is now in 'm' and it is disconnected from the 2486 * queue. 2487 * 2488 * Process one or more MT_CONTROL mbufs present before any data mbufs 2489 * in the first mbuf chain on the socket buffer. We call into the 2490 * protocol to perform externalization (or freeing if controlp == 2491 * NULL). In some cases there can be only MT_CONTROL mbufs without 2492 * MT_DATA mbufs. 2493 */ 2494 if (m->m_type == MT_CONTROL) { 2495 struct mbuf *cm = NULL, *cmn; 2496 struct mbuf **cme = &cm; 2497 2498 do { 2499 m2 = m->m_next; 2500 m->m_next = NULL; 2501 *cme = m; 2502 cme = &(*cme)->m_next; 2503 m = m2; 2504 } while (m != NULL && m->m_type == MT_CONTROL); 2505 while (cm != NULL) { 2506 cmn = cm->m_next; 2507 cm->m_next = NULL; 2508 if (pr->pr_domain->dom_externalize != NULL) { 2509 error = (*pr->pr_domain->dom_externalize) 2510 (cm, controlp, flags); 2511 } else if (controlp != NULL) 2512 *controlp = cm; 2513 else 2514 m_freem(cm); 2515 if (controlp != NULL) { 2516 while (*controlp != NULL) 2517 controlp = &(*controlp)->m_next; 2518 } 2519 cm = cmn; 2520 } 2521 } 2522 KASSERT(m == NULL || m->m_type == MT_DATA, 2523 ("soreceive_dgram: !data")); 2524 while (m != NULL && uio->uio_resid > 0) { 2525 len = uio->uio_resid; 2526 if (len > m->m_len) 2527 len = m->m_len; 2528 error = uiomove(mtod(m, char *), (int)len, uio); 2529 if (error) { 2530 m_freem(m); 2531 return (error); 2532 } 2533 if (len == m->m_len) 2534 m = m_free(m); 2535 else { 2536 m->m_data += len; 2537 m->m_len -= len; 2538 } 2539 } 2540 if (m != NULL) { 2541 flags |= MSG_TRUNC; 2542 m_freem(m); 2543 } 2544 if (flagsp != NULL) 2545 *flagsp |= flags; 2546 return (0); 2547 } 2548 2549 int 2550 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2551 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2552 { 2553 int error; 2554 2555 CURVNET_SET(so->so_vnet); 2556 if (!SOLISTENING(so)) 2557 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, 2558 mp0, controlp, flagsp)); 2559 else 2560 error = ENOTCONN; 2561 CURVNET_RESTORE(); 2562 return (error); 2563 } 2564 2565 int 2566 soshutdown(struct socket *so, int how) 2567 { 2568 struct protosw *pr = so->so_proto; 2569 int error, soerror_enotconn; 2570 2571 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2572 return (EINVAL); 2573 2574 soerror_enotconn = 0; 2575 if ((so->so_state & 2576 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 2577 /* 2578 * POSIX mandates us to return ENOTCONN when shutdown(2) is 2579 * invoked on a datagram sockets, however historically we would 2580 * actually tear socket down. This is known to be leveraged by 2581 * some applications to unblock process waiting in recvXXX(2) 2582 * by other process that it shares that socket with. Try to meet 2583 * both backward-compatibility and POSIX requirements by forcing 2584 * ENOTCONN but still asking protocol to perform pru_shutdown(). 2585 */ 2586 if (so->so_type != SOCK_DGRAM) 2587 return (ENOTCONN); 2588 soerror_enotconn = 1; 2589 } 2590 2591 CURVNET_SET(so->so_vnet); 2592 if (pr->pr_usrreqs->pru_flush != NULL) 2593 (*pr->pr_usrreqs->pru_flush)(so, how); 2594 if (how != SHUT_WR) 2595 sorflush(so); 2596 if (how != SHUT_RD) { 2597 error = (*pr->pr_usrreqs->pru_shutdown)(so); 2598 wakeup(&so->so_timeo); 2599 CURVNET_RESTORE(); 2600 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error); 2601 } 2602 wakeup(&so->so_timeo); 2603 CURVNET_RESTORE(); 2604 2605 return (soerror_enotconn ? ENOTCONN : 0); 2606 } 2607 2608 void 2609 sorflush(struct socket *so) 2610 { 2611 struct sockbuf *sb = &so->so_rcv; 2612 struct protosw *pr = so->so_proto; 2613 struct socket aso; 2614 2615 VNET_SO_ASSERT(so); 2616 2617 /* 2618 * In order to avoid calling dom_dispose with the socket buffer mutex 2619 * held, and in order to generally avoid holding the lock for a long 2620 * time, we make a copy of the socket buffer and clear the original 2621 * (except locks, state). The new socket buffer copy won't have 2622 * initialized locks so we can only call routines that won't use or 2623 * assert those locks. 2624 * 2625 * Dislodge threads currently blocked in receive and wait to acquire 2626 * a lock against other simultaneous readers before clearing the 2627 * socket buffer. Don't let our acquire be interrupted by a signal 2628 * despite any existing socket disposition on interruptable waiting. 2629 */ 2630 socantrcvmore(so); 2631 (void) sblock(sb, SBL_WAIT | SBL_NOINTR); 2632 2633 /* 2634 * Invalidate/clear most of the sockbuf structure, but leave selinfo 2635 * and mutex data unchanged. 2636 */ 2637 SOCKBUF_LOCK(sb); 2638 bzero(&aso, sizeof(aso)); 2639 aso.so_pcb = so->so_pcb; 2640 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero, 2641 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2642 bzero(&sb->sb_startzero, 2643 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2644 SOCKBUF_UNLOCK(sb); 2645 sbunlock(sb); 2646 2647 /* 2648 * Dispose of special rights and flush the copied socket. Don't call 2649 * any unsafe routines (that rely on locks being initialized) on aso. 2650 */ 2651 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 2652 (*pr->pr_domain->dom_dispose)(&aso); 2653 sbrelease_internal(&aso.so_rcv, so); 2654 } 2655 2656 /* 2657 * Wrapper for Socket established helper hook. 2658 * Parameters: socket, context of the hook point, hook id. 2659 */ 2660 static int inline 2661 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 2662 { 2663 struct socket_hhook_data hhook_data = { 2664 .so = so, 2665 .hctx = hctx, 2666 .m = NULL, 2667 .status = 0 2668 }; 2669 2670 CURVNET_SET(so->so_vnet); 2671 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 2672 CURVNET_RESTORE(); 2673 2674 /* Ugly but needed, since hhooks return void for now */ 2675 return (hhook_data.status); 2676 } 2677 2678 /* 2679 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 2680 * additional variant to handle the case where the option value needs to be 2681 * some kind of integer, but not a specific size. In addition to their use 2682 * here, these functions are also called by the protocol-level pr_ctloutput() 2683 * routines. 2684 */ 2685 int 2686 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2687 { 2688 size_t valsize; 2689 2690 /* 2691 * If the user gives us more than we wanted, we ignore it, but if we 2692 * don't get the minimum length the caller wants, we return EINVAL. 2693 * On success, sopt->sopt_valsize is set to however much we actually 2694 * retrieved. 2695 */ 2696 if ((valsize = sopt->sopt_valsize) < minlen) 2697 return EINVAL; 2698 if (valsize > len) 2699 sopt->sopt_valsize = valsize = len; 2700 2701 if (sopt->sopt_td != NULL) 2702 return (copyin(sopt->sopt_val, buf, valsize)); 2703 2704 bcopy(sopt->sopt_val, buf, valsize); 2705 return (0); 2706 } 2707 2708 /* 2709 * Kernel version of setsockopt(2). 2710 * 2711 * XXX: optlen is size_t, not socklen_t 2712 */ 2713 int 2714 so_setsockopt(struct socket *so, int level, int optname, void *optval, 2715 size_t optlen) 2716 { 2717 struct sockopt sopt; 2718 2719 sopt.sopt_level = level; 2720 sopt.sopt_name = optname; 2721 sopt.sopt_dir = SOPT_SET; 2722 sopt.sopt_val = optval; 2723 sopt.sopt_valsize = optlen; 2724 sopt.sopt_td = NULL; 2725 return (sosetopt(so, &sopt)); 2726 } 2727 2728 int 2729 sosetopt(struct socket *so, struct sockopt *sopt) 2730 { 2731 int error, optval; 2732 struct linger l; 2733 struct timeval tv; 2734 sbintime_t val; 2735 uint32_t val32; 2736 #ifdef MAC 2737 struct mac extmac; 2738 #endif 2739 2740 CURVNET_SET(so->so_vnet); 2741 error = 0; 2742 if (sopt->sopt_level != SOL_SOCKET) { 2743 if (so->so_proto->pr_ctloutput != NULL) { 2744 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2745 CURVNET_RESTORE(); 2746 return (error); 2747 } 2748 error = ENOPROTOOPT; 2749 } else { 2750 switch (sopt->sopt_name) { 2751 case SO_ACCEPTFILTER: 2752 error = accept_filt_setopt(so, sopt); 2753 if (error) 2754 goto bad; 2755 break; 2756 2757 case SO_LINGER: 2758 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 2759 if (error) 2760 goto bad; 2761 2762 SOCK_LOCK(so); 2763 so->so_linger = l.l_linger; 2764 if (l.l_onoff) 2765 so->so_options |= SO_LINGER; 2766 else 2767 so->so_options &= ~SO_LINGER; 2768 SOCK_UNLOCK(so); 2769 break; 2770 2771 case SO_DEBUG: 2772 case SO_KEEPALIVE: 2773 case SO_DONTROUTE: 2774 case SO_USELOOPBACK: 2775 case SO_BROADCAST: 2776 case SO_REUSEADDR: 2777 case SO_REUSEPORT: 2778 case SO_OOBINLINE: 2779 case SO_TIMESTAMP: 2780 case SO_BINTIME: 2781 case SO_NOSIGPIPE: 2782 case SO_NO_DDP: 2783 case SO_NO_OFFLOAD: 2784 error = sooptcopyin(sopt, &optval, sizeof optval, 2785 sizeof optval); 2786 if (error) 2787 goto bad; 2788 SOCK_LOCK(so); 2789 if (optval) 2790 so->so_options |= sopt->sopt_name; 2791 else 2792 so->so_options &= ~sopt->sopt_name; 2793 SOCK_UNLOCK(so); 2794 break; 2795 2796 case SO_SETFIB: 2797 error = sooptcopyin(sopt, &optval, sizeof optval, 2798 sizeof optval); 2799 if (error) 2800 goto bad; 2801 2802 if (optval < 0 || optval >= rt_numfibs) { 2803 error = EINVAL; 2804 goto bad; 2805 } 2806 if (((so->so_proto->pr_domain->dom_family == PF_INET) || 2807 (so->so_proto->pr_domain->dom_family == PF_INET6) || 2808 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) 2809 so->so_fibnum = optval; 2810 else 2811 so->so_fibnum = 0; 2812 break; 2813 2814 case SO_USER_COOKIE: 2815 error = sooptcopyin(sopt, &val32, sizeof val32, 2816 sizeof val32); 2817 if (error) 2818 goto bad; 2819 so->so_user_cookie = val32; 2820 break; 2821 2822 case SO_SNDBUF: 2823 case SO_RCVBUF: 2824 case SO_SNDLOWAT: 2825 case SO_RCVLOWAT: 2826 error = sooptcopyin(sopt, &optval, sizeof optval, 2827 sizeof optval); 2828 if (error) 2829 goto bad; 2830 2831 /* 2832 * Values < 1 make no sense for any of these options, 2833 * so disallow them. 2834 */ 2835 if (optval < 1) { 2836 error = EINVAL; 2837 goto bad; 2838 } 2839 2840 error = sbsetopt(so, sopt->sopt_name, optval); 2841 break; 2842 2843 case SO_SNDTIMEO: 2844 case SO_RCVTIMEO: 2845 #ifdef COMPAT_FREEBSD32 2846 if (SV_CURPROC_FLAG(SV_ILP32)) { 2847 struct timeval32 tv32; 2848 2849 error = sooptcopyin(sopt, &tv32, sizeof tv32, 2850 sizeof tv32); 2851 CP(tv32, tv, tv_sec); 2852 CP(tv32, tv, tv_usec); 2853 } else 2854 #endif 2855 error = sooptcopyin(sopt, &tv, sizeof tv, 2856 sizeof tv); 2857 if (error) 2858 goto bad; 2859 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 2860 tv.tv_usec >= 1000000) { 2861 error = EDOM; 2862 goto bad; 2863 } 2864 if (tv.tv_sec > INT32_MAX) 2865 val = SBT_MAX; 2866 else 2867 val = tvtosbt(tv); 2868 switch (sopt->sopt_name) { 2869 case SO_SNDTIMEO: 2870 so->so_snd.sb_timeo = val; 2871 break; 2872 case SO_RCVTIMEO: 2873 so->so_rcv.sb_timeo = val; 2874 break; 2875 } 2876 break; 2877 2878 case SO_LABEL: 2879 #ifdef MAC 2880 error = sooptcopyin(sopt, &extmac, sizeof extmac, 2881 sizeof extmac); 2882 if (error) 2883 goto bad; 2884 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 2885 so, &extmac); 2886 #else 2887 error = EOPNOTSUPP; 2888 #endif 2889 break; 2890 2891 case SO_TS_CLOCK: 2892 error = sooptcopyin(sopt, &optval, sizeof optval, 2893 sizeof optval); 2894 if (error) 2895 goto bad; 2896 if (optval < 0 || optval > SO_TS_CLOCK_MAX) { 2897 error = EINVAL; 2898 goto bad; 2899 } 2900 so->so_ts_clock = optval; 2901 break; 2902 2903 case SO_MAX_PACING_RATE: 2904 error = sooptcopyin(sopt, &val32, sizeof(val32), 2905 sizeof(val32)); 2906 if (error) 2907 goto bad; 2908 so->so_max_pacing_rate = val32; 2909 break; 2910 2911 default: 2912 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 2913 error = hhook_run_socket(so, sopt, 2914 HHOOK_SOCKET_OPT); 2915 else 2916 error = ENOPROTOOPT; 2917 break; 2918 } 2919 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 2920 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 2921 } 2922 bad: 2923 CURVNET_RESTORE(); 2924 return (error); 2925 } 2926 2927 /* 2928 * Helper routine for getsockopt. 2929 */ 2930 int 2931 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 2932 { 2933 int error; 2934 size_t valsize; 2935 2936 error = 0; 2937 2938 /* 2939 * Documented get behavior is that we always return a value, possibly 2940 * truncated to fit in the user's buffer. Traditional behavior is 2941 * that we always tell the user precisely how much we copied, rather 2942 * than something useful like the total amount we had available for 2943 * her. Note that this interface is not idempotent; the entire 2944 * answer must be generated ahead of time. 2945 */ 2946 valsize = min(len, sopt->sopt_valsize); 2947 sopt->sopt_valsize = valsize; 2948 if (sopt->sopt_val != NULL) { 2949 if (sopt->sopt_td != NULL) 2950 error = copyout(buf, sopt->sopt_val, valsize); 2951 else 2952 bcopy(buf, sopt->sopt_val, valsize); 2953 } 2954 return (error); 2955 } 2956 2957 int 2958 sogetopt(struct socket *so, struct sockopt *sopt) 2959 { 2960 int error, optval; 2961 struct linger l; 2962 struct timeval tv; 2963 #ifdef MAC 2964 struct mac extmac; 2965 #endif 2966 2967 CURVNET_SET(so->so_vnet); 2968 error = 0; 2969 if (sopt->sopt_level != SOL_SOCKET) { 2970 if (so->so_proto->pr_ctloutput != NULL) 2971 error = (*so->so_proto->pr_ctloutput)(so, sopt); 2972 else 2973 error = ENOPROTOOPT; 2974 CURVNET_RESTORE(); 2975 return (error); 2976 } else { 2977 switch (sopt->sopt_name) { 2978 case SO_ACCEPTFILTER: 2979 error = accept_filt_getopt(so, sopt); 2980 break; 2981 2982 case SO_LINGER: 2983 SOCK_LOCK(so); 2984 l.l_onoff = so->so_options & SO_LINGER; 2985 l.l_linger = so->so_linger; 2986 SOCK_UNLOCK(so); 2987 error = sooptcopyout(sopt, &l, sizeof l); 2988 break; 2989 2990 case SO_USELOOPBACK: 2991 case SO_DONTROUTE: 2992 case SO_DEBUG: 2993 case SO_KEEPALIVE: 2994 case SO_REUSEADDR: 2995 case SO_REUSEPORT: 2996 case SO_BROADCAST: 2997 case SO_OOBINLINE: 2998 case SO_ACCEPTCONN: 2999 case SO_TIMESTAMP: 3000 case SO_BINTIME: 3001 case SO_NOSIGPIPE: 3002 optval = so->so_options & sopt->sopt_name; 3003 integer: 3004 error = sooptcopyout(sopt, &optval, sizeof optval); 3005 break; 3006 3007 case SO_TYPE: 3008 optval = so->so_type; 3009 goto integer; 3010 3011 case SO_PROTOCOL: 3012 optval = so->so_proto->pr_protocol; 3013 goto integer; 3014 3015 case SO_ERROR: 3016 SOCK_LOCK(so); 3017 optval = so->so_error; 3018 so->so_error = 0; 3019 SOCK_UNLOCK(so); 3020 goto integer; 3021 3022 case SO_SNDBUF: 3023 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : 3024 so->so_snd.sb_hiwat; 3025 goto integer; 3026 3027 case SO_RCVBUF: 3028 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : 3029 so->so_rcv.sb_hiwat; 3030 goto integer; 3031 3032 case SO_SNDLOWAT: 3033 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : 3034 so->so_snd.sb_lowat; 3035 goto integer; 3036 3037 case SO_RCVLOWAT: 3038 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : 3039 so->so_rcv.sb_lowat; 3040 goto integer; 3041 3042 case SO_SNDTIMEO: 3043 case SO_RCVTIMEO: 3044 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 3045 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 3046 #ifdef COMPAT_FREEBSD32 3047 if (SV_CURPROC_FLAG(SV_ILP32)) { 3048 struct timeval32 tv32; 3049 3050 CP(tv, tv32, tv_sec); 3051 CP(tv, tv32, tv_usec); 3052 error = sooptcopyout(sopt, &tv32, sizeof tv32); 3053 } else 3054 #endif 3055 error = sooptcopyout(sopt, &tv, sizeof tv); 3056 break; 3057 3058 case SO_LABEL: 3059 #ifdef MAC 3060 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3061 sizeof(extmac)); 3062 if (error) 3063 goto bad; 3064 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 3065 so, &extmac); 3066 if (error) 3067 goto bad; 3068 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3069 #else 3070 error = EOPNOTSUPP; 3071 #endif 3072 break; 3073 3074 case SO_PEERLABEL: 3075 #ifdef MAC 3076 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 3077 sizeof(extmac)); 3078 if (error) 3079 goto bad; 3080 error = mac_getsockopt_peerlabel( 3081 sopt->sopt_td->td_ucred, so, &extmac); 3082 if (error) 3083 goto bad; 3084 error = sooptcopyout(sopt, &extmac, sizeof extmac); 3085 #else 3086 error = EOPNOTSUPP; 3087 #endif 3088 break; 3089 3090 case SO_LISTENQLIMIT: 3091 optval = SOLISTENING(so) ? so->sol_qlimit : 0; 3092 goto integer; 3093 3094 case SO_LISTENQLEN: 3095 optval = SOLISTENING(so) ? so->sol_qlen : 0; 3096 goto integer; 3097 3098 case SO_LISTENINCQLEN: 3099 optval = SOLISTENING(so) ? so->sol_incqlen : 0; 3100 goto integer; 3101 3102 case SO_TS_CLOCK: 3103 optval = so->so_ts_clock; 3104 goto integer; 3105 3106 case SO_MAX_PACING_RATE: 3107 optval = so->so_max_pacing_rate; 3108 goto integer; 3109 3110 default: 3111 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 3112 error = hhook_run_socket(so, sopt, 3113 HHOOK_SOCKET_OPT); 3114 else 3115 error = ENOPROTOOPT; 3116 break; 3117 } 3118 } 3119 #ifdef MAC 3120 bad: 3121 #endif 3122 CURVNET_RESTORE(); 3123 return (error); 3124 } 3125 3126 int 3127 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 3128 { 3129 struct mbuf *m, *m_prev; 3130 int sopt_size = sopt->sopt_valsize; 3131 3132 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3133 if (m == NULL) 3134 return ENOBUFS; 3135 if (sopt_size > MLEN) { 3136 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 3137 if ((m->m_flags & M_EXT) == 0) { 3138 m_free(m); 3139 return ENOBUFS; 3140 } 3141 m->m_len = min(MCLBYTES, sopt_size); 3142 } else { 3143 m->m_len = min(MLEN, sopt_size); 3144 } 3145 sopt_size -= m->m_len; 3146 *mp = m; 3147 m_prev = m; 3148 3149 while (sopt_size) { 3150 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 3151 if (m == NULL) { 3152 m_freem(*mp); 3153 return ENOBUFS; 3154 } 3155 if (sopt_size > MLEN) { 3156 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 3157 M_NOWAIT); 3158 if ((m->m_flags & M_EXT) == 0) { 3159 m_freem(m); 3160 m_freem(*mp); 3161 return ENOBUFS; 3162 } 3163 m->m_len = min(MCLBYTES, sopt_size); 3164 } else { 3165 m->m_len = min(MLEN, sopt_size); 3166 } 3167 sopt_size -= m->m_len; 3168 m_prev->m_next = m; 3169 m_prev = m; 3170 } 3171 return (0); 3172 } 3173 3174 int 3175 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 3176 { 3177 struct mbuf *m0 = m; 3178 3179 if (sopt->sopt_val == NULL) 3180 return (0); 3181 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3182 if (sopt->sopt_td != NULL) { 3183 int error; 3184 3185 error = copyin(sopt->sopt_val, mtod(m, char *), 3186 m->m_len); 3187 if (error != 0) { 3188 m_freem(m0); 3189 return(error); 3190 } 3191 } else 3192 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 3193 sopt->sopt_valsize -= m->m_len; 3194 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3195 m = m->m_next; 3196 } 3197 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 3198 panic("ip6_sooptmcopyin"); 3199 return (0); 3200 } 3201 3202 int 3203 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 3204 { 3205 struct mbuf *m0 = m; 3206 size_t valsize = 0; 3207 3208 if (sopt->sopt_val == NULL) 3209 return (0); 3210 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 3211 if (sopt->sopt_td != NULL) { 3212 int error; 3213 3214 error = copyout(mtod(m, char *), sopt->sopt_val, 3215 m->m_len); 3216 if (error != 0) { 3217 m_freem(m0); 3218 return(error); 3219 } 3220 } else 3221 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 3222 sopt->sopt_valsize -= m->m_len; 3223 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 3224 valsize += m->m_len; 3225 m = m->m_next; 3226 } 3227 if (m != NULL) { 3228 /* enough soopt buffer should be given from user-land */ 3229 m_freem(m0); 3230 return(EINVAL); 3231 } 3232 sopt->sopt_valsize = valsize; 3233 return (0); 3234 } 3235 3236 /* 3237 * sohasoutofband(): protocol notifies socket layer of the arrival of new 3238 * out-of-band data, which will then notify socket consumers. 3239 */ 3240 void 3241 sohasoutofband(struct socket *so) 3242 { 3243 3244 if (so->so_sigio != NULL) 3245 pgsigio(&so->so_sigio, SIGURG, 0); 3246 selwakeuppri(&so->so_rdsel, PSOCK); 3247 } 3248 3249 int 3250 sopoll(struct socket *so, int events, struct ucred *active_cred, 3251 struct thread *td) 3252 { 3253 3254 /* 3255 * We do not need to set or assert curvnet as long as everyone uses 3256 * sopoll_generic(). 3257 */ 3258 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, 3259 td)); 3260 } 3261 3262 int 3263 sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 3264 struct thread *td) 3265 { 3266 int revents; 3267 3268 SOCK_LOCK(so); 3269 if (SOLISTENING(so)) { 3270 if (!(events & (POLLIN | POLLRDNORM))) 3271 revents = 0; 3272 else if (!TAILQ_EMPTY(&so->sol_comp)) 3273 revents = events & (POLLIN | POLLRDNORM); 3274 else { 3275 selrecord(td, &so->so_rdsel); 3276 revents = 0; 3277 } 3278 } else { 3279 revents = 0; 3280 SOCKBUF_LOCK(&so->so_snd); 3281 SOCKBUF_LOCK(&so->so_rcv); 3282 if (events & (POLLIN | POLLRDNORM)) 3283 if (soreadabledata(so)) 3284 revents |= events & (POLLIN | POLLRDNORM); 3285 if (events & (POLLOUT | POLLWRNORM)) 3286 if (sowriteable(so)) 3287 revents |= events & (POLLOUT | POLLWRNORM); 3288 if (events & (POLLPRI | POLLRDBAND)) 3289 if (so->so_oobmark || 3290 (so->so_rcv.sb_state & SBS_RCVATMARK)) 3291 revents |= events & (POLLPRI | POLLRDBAND); 3292 if ((events & POLLINIGNEOF) == 0) { 3293 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3294 revents |= events & (POLLIN | POLLRDNORM); 3295 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 3296 revents |= POLLHUP; 3297 } 3298 } 3299 if (revents == 0) { 3300 if (events & 3301 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 3302 selrecord(td, &so->so_rdsel); 3303 so->so_rcv.sb_flags |= SB_SEL; 3304 } 3305 if (events & (POLLOUT | POLLWRNORM)) { 3306 selrecord(td, &so->so_wrsel); 3307 so->so_snd.sb_flags |= SB_SEL; 3308 } 3309 } 3310 SOCKBUF_UNLOCK(&so->so_rcv); 3311 SOCKBUF_UNLOCK(&so->so_snd); 3312 } 3313 SOCK_UNLOCK(so); 3314 return (revents); 3315 } 3316 3317 int 3318 soo_kqfilter(struct file *fp, struct knote *kn) 3319 { 3320 struct socket *so = kn->kn_fp->f_data; 3321 struct sockbuf *sb; 3322 struct knlist *knl; 3323 3324 switch (kn->kn_filter) { 3325 case EVFILT_READ: 3326 kn->kn_fop = &soread_filtops; 3327 knl = &so->so_rdsel.si_note; 3328 sb = &so->so_rcv; 3329 break; 3330 case EVFILT_WRITE: 3331 kn->kn_fop = &sowrite_filtops; 3332 knl = &so->so_wrsel.si_note; 3333 sb = &so->so_snd; 3334 break; 3335 case EVFILT_EMPTY: 3336 kn->kn_fop = &soempty_filtops; 3337 knl = &so->so_wrsel.si_note; 3338 sb = &so->so_snd; 3339 break; 3340 default: 3341 return (EINVAL); 3342 } 3343 3344 SOCK_LOCK(so); 3345 if (SOLISTENING(so)) { 3346 knlist_add(knl, kn, 1); 3347 } else { 3348 SOCKBUF_LOCK(sb); 3349 knlist_add(knl, kn, 1); 3350 sb->sb_flags |= SB_KNOTE; 3351 SOCKBUF_UNLOCK(sb); 3352 } 3353 SOCK_UNLOCK(so); 3354 return (0); 3355 } 3356 3357 /* 3358 * Some routines that return EOPNOTSUPP for entry points that are not 3359 * supported by a protocol. Fill in as needed. 3360 */ 3361 int 3362 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 3363 { 3364 3365 return EOPNOTSUPP; 3366 } 3367 3368 int 3369 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job) 3370 { 3371 3372 return EOPNOTSUPP; 3373 } 3374 3375 int 3376 pru_attach_notsupp(struct socket *so, int proto, struct thread *td) 3377 { 3378 3379 return EOPNOTSUPP; 3380 } 3381 3382 int 3383 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3384 { 3385 3386 return EOPNOTSUPP; 3387 } 3388 3389 int 3390 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3391 struct thread *td) 3392 { 3393 3394 return EOPNOTSUPP; 3395 } 3396 3397 int 3398 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3399 { 3400 3401 return EOPNOTSUPP; 3402 } 3403 3404 int 3405 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam, 3406 struct thread *td) 3407 { 3408 3409 return EOPNOTSUPP; 3410 } 3411 3412 int 3413 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 3414 { 3415 3416 return EOPNOTSUPP; 3417 } 3418 3419 int 3420 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 3421 struct ifnet *ifp, struct thread *td) 3422 { 3423 3424 return EOPNOTSUPP; 3425 } 3426 3427 int 3428 pru_disconnect_notsupp(struct socket *so) 3429 { 3430 3431 return EOPNOTSUPP; 3432 } 3433 3434 int 3435 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) 3436 { 3437 3438 return EOPNOTSUPP; 3439 } 3440 3441 int 3442 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) 3443 { 3444 3445 return EOPNOTSUPP; 3446 } 3447 3448 int 3449 pru_rcvd_notsupp(struct socket *so, int flags) 3450 { 3451 3452 return EOPNOTSUPP; 3453 } 3454 3455 int 3456 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 3457 { 3458 3459 return EOPNOTSUPP; 3460 } 3461 3462 int 3463 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, 3464 struct sockaddr *addr, struct mbuf *control, struct thread *td) 3465 { 3466 3467 return EOPNOTSUPP; 3468 } 3469 3470 int 3471 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count) 3472 { 3473 3474 return (EOPNOTSUPP); 3475 } 3476 3477 /* 3478 * This isn't really a ``null'' operation, but it's the default one and 3479 * doesn't do anything destructive. 3480 */ 3481 int 3482 pru_sense_null(struct socket *so, struct stat *sb) 3483 { 3484 3485 sb->st_blksize = so->so_snd.sb_hiwat; 3486 return 0; 3487 } 3488 3489 int 3490 pru_shutdown_notsupp(struct socket *so) 3491 { 3492 3493 return EOPNOTSUPP; 3494 } 3495 3496 int 3497 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) 3498 { 3499 3500 return EOPNOTSUPP; 3501 } 3502 3503 int 3504 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 3505 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 3506 { 3507 3508 return EOPNOTSUPP; 3509 } 3510 3511 int 3512 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 3513 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 3514 { 3515 3516 return EOPNOTSUPP; 3517 } 3518 3519 int 3520 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, 3521 struct thread *td) 3522 { 3523 3524 return EOPNOTSUPP; 3525 } 3526 3527 static void 3528 filt_sordetach(struct knote *kn) 3529 { 3530 struct socket *so = kn->kn_fp->f_data; 3531 3532 so_rdknl_lock(so); 3533 knlist_remove(&so->so_rdsel.si_note, kn, 1); 3534 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note)) 3535 so->so_rcv.sb_flags &= ~SB_KNOTE; 3536 so_rdknl_unlock(so); 3537 } 3538 3539 /*ARGSUSED*/ 3540 static int 3541 filt_soread(struct knote *kn, long hint) 3542 { 3543 struct socket *so; 3544 3545 so = kn->kn_fp->f_data; 3546 3547 if (SOLISTENING(so)) { 3548 SOCK_LOCK_ASSERT(so); 3549 kn->kn_data = so->sol_qlen; 3550 return (!TAILQ_EMPTY(&so->sol_comp)); 3551 } 3552 3553 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3554 3555 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 3556 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3557 kn->kn_flags |= EV_EOF; 3558 kn->kn_fflags = so->so_error; 3559 return (1); 3560 } else if (so->so_error) /* temporary udp error */ 3561 return (1); 3562 3563 if (kn->kn_sfflags & NOTE_LOWAT) { 3564 if (kn->kn_data >= kn->kn_sdata) 3565 return (1); 3566 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) 3567 return (1); 3568 3569 /* This hook returning non-zero indicates an event, not error */ 3570 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); 3571 } 3572 3573 static void 3574 filt_sowdetach(struct knote *kn) 3575 { 3576 struct socket *so = kn->kn_fp->f_data; 3577 3578 so_wrknl_lock(so); 3579 knlist_remove(&so->so_wrsel.si_note, kn, 1); 3580 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note)) 3581 so->so_snd.sb_flags &= ~SB_KNOTE; 3582 so_wrknl_unlock(so); 3583 } 3584 3585 /*ARGSUSED*/ 3586 static int 3587 filt_sowrite(struct knote *kn, long hint) 3588 { 3589 struct socket *so; 3590 3591 so = kn->kn_fp->f_data; 3592 3593 if (SOLISTENING(so)) 3594 return (0); 3595 3596 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3597 kn->kn_data = sbspace(&so->so_snd); 3598 3599 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); 3600 3601 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3602 kn->kn_flags |= EV_EOF; 3603 kn->kn_fflags = so->so_error; 3604 return (1); 3605 } else if (so->so_error) /* temporary udp error */ 3606 return (1); 3607 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3608 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3609 return (0); 3610 else if (kn->kn_sfflags & NOTE_LOWAT) 3611 return (kn->kn_data >= kn->kn_sdata); 3612 else 3613 return (kn->kn_data >= so->so_snd.sb_lowat); 3614 } 3615 3616 static int 3617 filt_soempty(struct knote *kn, long hint) 3618 { 3619 struct socket *so; 3620 3621 so = kn->kn_fp->f_data; 3622 3623 if (SOLISTENING(so)) 3624 return (1); 3625 3626 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3627 kn->kn_data = sbused(&so->so_snd); 3628 3629 if (kn->kn_data == 0) 3630 return (1); 3631 else 3632 return (0); 3633 } 3634 3635 int 3636 socheckuid(struct socket *so, uid_t uid) 3637 { 3638 3639 if (so == NULL) 3640 return (EPERM); 3641 if (so->so_cred->cr_uid != uid) 3642 return (EPERM); 3643 return (0); 3644 } 3645 3646 /* 3647 * These functions are used by protocols to notify the socket layer (and its 3648 * consumers) of state changes in the sockets driven by protocol-side events. 3649 */ 3650 3651 /* 3652 * Procedures to manipulate state flags of socket and do appropriate wakeups. 3653 * 3654 * Normal sequence from the active (originating) side is that 3655 * soisconnecting() is called during processing of connect() call, resulting 3656 * in an eventual call to soisconnected() if/when the connection is 3657 * established. When the connection is torn down soisdisconnecting() is 3658 * called during processing of disconnect() call, and soisdisconnected() is 3659 * called when the connection to the peer is totally severed. The semantics 3660 * of these routines are such that connectionless protocols can call 3661 * soisconnected() and soisdisconnected() only, bypassing the in-progress 3662 * calls when setting up a ``connection'' takes no time. 3663 * 3664 * From the passive side, a socket is created with two queues of sockets: 3665 * so_incomp for connections in progress and so_comp for connections already 3666 * made and awaiting user acceptance. As a protocol is preparing incoming 3667 * connections, it creates a socket structure queued on so_incomp by calling 3668 * sonewconn(). When the connection is established, soisconnected() is 3669 * called, and transfers the socket structure to so_comp, making it available 3670 * to accept(). 3671 * 3672 * If a socket is closed with sockets on either so_incomp or so_comp, these 3673 * sockets are dropped. 3674 * 3675 * If higher-level protocols are implemented in the kernel, the wakeups done 3676 * here will sometimes cause software-interrupt process scheduling. 3677 */ 3678 void 3679 soisconnecting(struct socket *so) 3680 { 3681 3682 SOCK_LOCK(so); 3683 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 3684 so->so_state |= SS_ISCONNECTING; 3685 SOCK_UNLOCK(so); 3686 } 3687 3688 void 3689 soisconnected(struct socket *so) 3690 { 3691 struct socket *head; 3692 int ret; 3693 3694 /* 3695 * XXXGL: this is the only place where we acquire socket locks 3696 * in reverse order: first child, then listening socket. To 3697 * avoid possible LOR, use try semantics. 3698 */ 3699 restart: 3700 SOCK_LOCK(so); 3701 if ((head = so->so_listen) != NULL && 3702 __predict_false(SOLISTEN_TRYLOCK(head) == 0)) { 3703 SOCK_UNLOCK(so); 3704 goto restart; 3705 } 3706 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 3707 so->so_state |= SS_ISCONNECTED; 3708 if (head != NULL && (so->so_qstate == SQ_INCOMP)) { 3709 again: 3710 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 3711 TAILQ_REMOVE(&head->sol_incomp, so, so_list); 3712 head->sol_incqlen--; 3713 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 3714 head->sol_qlen++; 3715 so->so_qstate = SQ_COMP; 3716 SOCK_UNLOCK(so); 3717 solisten_wakeup(head); /* unlocks */ 3718 } else { 3719 SOCKBUF_LOCK(&so->so_rcv); 3720 soupcall_set(so, SO_RCV, 3721 head->sol_accept_filter->accf_callback, 3722 head->sol_accept_filter_arg); 3723 so->so_options &= ~SO_ACCEPTFILTER; 3724 ret = head->sol_accept_filter->accf_callback(so, 3725 head->sol_accept_filter_arg, M_NOWAIT); 3726 if (ret == SU_ISCONNECTED) { 3727 soupcall_clear(so, SO_RCV); 3728 SOCKBUF_UNLOCK(&so->so_rcv); 3729 goto again; 3730 } 3731 SOCKBUF_UNLOCK(&so->so_rcv); 3732 SOCK_UNLOCK(so); 3733 SOLISTEN_UNLOCK(head); 3734 } 3735 return; 3736 } 3737 if (head != NULL) 3738 SOLISTEN_UNLOCK(head); 3739 SOCK_UNLOCK(so); 3740 wakeup(&so->so_timeo); 3741 sorwakeup(so); 3742 sowwakeup(so); 3743 } 3744 3745 void 3746 soisdisconnecting(struct socket *so) 3747 { 3748 3749 SOCK_LOCK(so); 3750 so->so_state &= ~SS_ISCONNECTING; 3751 so->so_state |= SS_ISDISCONNECTING; 3752 3753 if (!SOLISTENING(so)) { 3754 SOCKBUF_LOCK(&so->so_rcv); 3755 socantrcvmore_locked(so); 3756 SOCKBUF_LOCK(&so->so_snd); 3757 socantsendmore_locked(so); 3758 } 3759 SOCK_UNLOCK(so); 3760 wakeup(&so->so_timeo); 3761 } 3762 3763 void 3764 soisdisconnected(struct socket *so) 3765 { 3766 3767 SOCK_LOCK(so); 3768 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 3769 so->so_state |= SS_ISDISCONNECTED; 3770 3771 if (!SOLISTENING(so)) { 3772 SOCK_UNLOCK(so); 3773 SOCKBUF_LOCK(&so->so_rcv); 3774 socantrcvmore_locked(so); 3775 SOCKBUF_LOCK(&so->so_snd); 3776 sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); 3777 socantsendmore_locked(so); 3778 } else 3779 SOCK_UNLOCK(so); 3780 wakeup(&so->so_timeo); 3781 } 3782 3783 /* 3784 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 3785 */ 3786 struct sockaddr * 3787 sodupsockaddr(const struct sockaddr *sa, int mflags) 3788 { 3789 struct sockaddr *sa2; 3790 3791 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 3792 if (sa2) 3793 bcopy(sa, sa2, sa->sa_len); 3794 return sa2; 3795 } 3796 3797 /* 3798 * Register per-socket buffer upcalls. 3799 */ 3800 void 3801 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg) 3802 { 3803 struct sockbuf *sb; 3804 3805 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 3806 3807 switch (which) { 3808 case SO_RCV: 3809 sb = &so->so_rcv; 3810 break; 3811 case SO_SND: 3812 sb = &so->so_snd; 3813 break; 3814 default: 3815 panic("soupcall_set: bad which"); 3816 } 3817 SOCKBUF_LOCK_ASSERT(sb); 3818 sb->sb_upcall = func; 3819 sb->sb_upcallarg = arg; 3820 sb->sb_flags |= SB_UPCALL; 3821 } 3822 3823 void 3824 soupcall_clear(struct socket *so, int which) 3825 { 3826 struct sockbuf *sb; 3827 3828 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 3829 3830 switch (which) { 3831 case SO_RCV: 3832 sb = &so->so_rcv; 3833 break; 3834 case SO_SND: 3835 sb = &so->so_snd; 3836 break; 3837 default: 3838 panic("soupcall_clear: bad which"); 3839 } 3840 SOCKBUF_LOCK_ASSERT(sb); 3841 KASSERT(sb->sb_upcall != NULL, 3842 ("%s: so %p no upcall to clear", __func__, so)); 3843 sb->sb_upcall = NULL; 3844 sb->sb_upcallarg = NULL; 3845 sb->sb_flags &= ~SB_UPCALL; 3846 } 3847 3848 void 3849 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg) 3850 { 3851 3852 SOLISTEN_LOCK_ASSERT(so); 3853 so->sol_upcall = func; 3854 so->sol_upcallarg = arg; 3855 } 3856 3857 static void 3858 so_rdknl_lock(void *arg) 3859 { 3860 struct socket *so = arg; 3861 3862 if (SOLISTENING(so)) 3863 SOCK_LOCK(so); 3864 else 3865 SOCKBUF_LOCK(&so->so_rcv); 3866 } 3867 3868 static void 3869 so_rdknl_unlock(void *arg) 3870 { 3871 struct socket *so = arg; 3872 3873 if (SOLISTENING(so)) 3874 SOCK_UNLOCK(so); 3875 else 3876 SOCKBUF_UNLOCK(&so->so_rcv); 3877 } 3878 3879 static void 3880 so_rdknl_assert_locked(void *arg) 3881 { 3882 struct socket *so = arg; 3883 3884 if (SOLISTENING(so)) 3885 SOCK_LOCK_ASSERT(so); 3886 else 3887 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3888 } 3889 3890 static void 3891 so_rdknl_assert_unlocked(void *arg) 3892 { 3893 struct socket *so = arg; 3894 3895 if (SOLISTENING(so)) 3896 SOCK_UNLOCK_ASSERT(so); 3897 else 3898 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); 3899 } 3900 3901 static void 3902 so_wrknl_lock(void *arg) 3903 { 3904 struct socket *so = arg; 3905 3906 if (SOLISTENING(so)) 3907 SOCK_LOCK(so); 3908 else 3909 SOCKBUF_LOCK(&so->so_snd); 3910 } 3911 3912 static void 3913 so_wrknl_unlock(void *arg) 3914 { 3915 struct socket *so = arg; 3916 3917 if (SOLISTENING(so)) 3918 SOCK_UNLOCK(so); 3919 else 3920 SOCKBUF_UNLOCK(&so->so_snd); 3921 } 3922 3923 static void 3924 so_wrknl_assert_locked(void *arg) 3925 { 3926 struct socket *so = arg; 3927 3928 if (SOLISTENING(so)) 3929 SOCK_LOCK_ASSERT(so); 3930 else 3931 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3932 } 3933 3934 static void 3935 so_wrknl_assert_unlocked(void *arg) 3936 { 3937 struct socket *so = arg; 3938 3939 if (SOLISTENING(so)) 3940 SOCK_UNLOCK_ASSERT(so); 3941 else 3942 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 3943 } 3944 3945 /* 3946 * Create an external-format (``xsocket'') structure using the information in 3947 * the kernel-format socket structure pointed to by so. This is done to 3948 * reduce the spew of irrelevant information over this interface, to isolate 3949 * user code from changes in the kernel structure, and potentially to provide 3950 * information-hiding if we decide that some of this information should be 3951 * hidden from users. 3952 */ 3953 void 3954 sotoxsocket(struct socket *so, struct xsocket *xso) 3955 { 3956 3957 xso->xso_len = sizeof *xso; 3958 xso->xso_so = so; 3959 xso->so_type = so->so_type; 3960 xso->so_options = so->so_options; 3961 xso->so_linger = so->so_linger; 3962 xso->so_state = so->so_state; 3963 xso->so_pcb = so->so_pcb; 3964 xso->xso_protocol = so->so_proto->pr_protocol; 3965 xso->xso_family = so->so_proto->pr_domain->dom_family; 3966 xso->so_timeo = so->so_timeo; 3967 xso->so_error = so->so_error; 3968 xso->so_uid = so->so_cred->cr_uid; 3969 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 3970 if (SOLISTENING(so)) { 3971 xso->so_qlen = so->sol_qlen; 3972 xso->so_incqlen = so->sol_incqlen; 3973 xso->so_qlimit = so->sol_qlimit; 3974 xso->so_oobmark = 0; 3975 bzero(&xso->so_snd, sizeof(xso->so_snd)); 3976 bzero(&xso->so_rcv, sizeof(xso->so_rcv)); 3977 } else { 3978 xso->so_state |= so->so_qstate; 3979 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0; 3980 xso->so_oobmark = so->so_oobmark; 3981 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 3982 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 3983 } 3984 } 3985 3986 struct sockbuf * 3987 so_sockbuf_rcv(struct socket *so) 3988 { 3989 3990 return (&so->so_rcv); 3991 } 3992 3993 struct sockbuf * 3994 so_sockbuf_snd(struct socket *so) 3995 { 3996 3997 return (&so->so_snd); 3998 } 3999 4000 int 4001 so_state_get(const struct socket *so) 4002 { 4003 4004 return (so->so_state); 4005 } 4006 4007 void 4008 so_state_set(struct socket *so, int val) 4009 { 4010 4011 so->so_state = val; 4012 } 4013 4014 int 4015 so_options_get(const struct socket *so) 4016 { 4017 4018 return (so->so_options); 4019 } 4020 4021 void 4022 so_options_set(struct socket *so, int val) 4023 { 4024 4025 so->so_options = val; 4026 } 4027 4028 int 4029 so_error_get(const struct socket *so) 4030 { 4031 4032 return (so->so_error); 4033 } 4034 4035 void 4036 so_error_set(struct socket *so, int val) 4037 { 4038 4039 so->so_error = val; 4040 } 4041 4042 int 4043 so_linger_get(const struct socket *so) 4044 { 4045 4046 return (so->so_linger); 4047 } 4048 4049 void 4050 so_linger_set(struct socket *so, int val) 4051 { 4052 4053 so->so_linger = val; 4054 } 4055 4056 struct protosw * 4057 so_protosw_get(const struct socket *so) 4058 { 4059 4060 return (so->so_proto); 4061 } 4062 4063 void 4064 so_protosw_set(struct socket *so, struct protosw *val) 4065 { 4066 4067 so->so_proto = val; 4068 } 4069 4070 void 4071 so_sorwakeup(struct socket *so) 4072 { 4073 4074 sorwakeup(so); 4075 } 4076 4077 void 4078 so_sowwakeup(struct socket *so) 4079 { 4080 4081 sowwakeup(so); 4082 } 4083 4084 void 4085 so_sorwakeup_locked(struct socket *so) 4086 { 4087 4088 sorwakeup_locked(so); 4089 } 4090 4091 void 4092 so_sowwakeup_locked(struct socket *so) 4093 { 4094 4095 sowwakeup_locked(so); 4096 } 4097 4098 void 4099 so_lock(struct socket *so) 4100 { 4101 4102 SOCK_LOCK(so); 4103 } 4104 4105 void 4106 so_unlock(struct socket *so) 4107 { 4108 4109 SOCK_UNLOCK(so); 4110 } 4111