uipc_socket.c revision 218559
129759Swollman/*- 229759Swollman * Copyright (c) 1982, 1986, 1988, 1990, 1993 329759Swollman * The Regents of the University of California. 429759Swollman * Copyright (c) 2004 The FreeBSD Foundation 529759Swollman * Copyright (c) 2004-2008 Robert N. M. Watson 629759Swollman * All rights reserved. 729759Swollman * 829759Swollman * Redistribution and use in source and binary forms, with or without 929759Swollman * modification, are permitted provided that the following conditions 1029759Swollman * are met: 1129759Swollman * 1. Redistributions of source code must retain the above copyright 1229759Swollman * notice, this list of conditions and the following disclaimer. 1329759Swollman * 2. Redistributions in binary form must reproduce the above copyright 1429759Swollman * notice, this list of conditions and the following disclaimer in the 1529759Swollman * documentation and/or other materials provided with the distribution. 1629759Swollman * 4. Neither the name of the University nor the names of its contributors 1729759Swollman * may be used to endorse or promote products derived from this software 1829759Swollman * without specific prior written permission. 1929759Swollman * 2029759Swollman * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 2129759Swollman * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 2229759Swollman * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 2329759Swollman * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 2429759Swollman * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2529759Swollman * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2629759Swollman * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2729759Swollman * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2829759Swollman * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2929759Swollman * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3029759Swollman * SUCH DAMAGE. 3129759Swollman * 3229759Swollman * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94 3329759Swollman */ 3487715Smarkm 3587715Smarkm/* 3687715Smarkm * Comments on the socket life cycle: 3787715Smarkm * 3887715Smarkm * soalloc() sets of socket layer state for a socket, called only by 3987715Smarkm * socreate() and sonewconn(). Socket layer private. 4087715Smarkm * 4187715Smarkm * sodealloc() tears down socket layer state for a socket, called only by 4229759Swollman * sofree() and sonewconn(). Socket layer private. 4387715Smarkm * 4429759Swollman * pru_attach() associates protocol layer state with an allocated socket; 4529759Swollman * called only once, may fail, aborting socket allocation. This is called 4629759Swollman * from socreate() and sonewconn(). Socket layer private. 4729759Swollman * 4829759Swollman * pru_detach() disassociates protocol layer state from an attached socket, 4929759Swollman * and will be called exactly once for sockets in which pru_attach() has 5029759Swollman * been successfully called. If pru_attach() returned an error, 5129759Swollman * pru_detach() will not be called. Socket layer private. 5229759Swollman * 5329759Swollman * pru_abort() and pru_close() notify the protocol layer that the last 5429759Swollman * consumer of a socket is starting to tear down the socket, and that the 5529759Swollman * protocol should terminate the connection. Historically, pru_abort() also 5629759Swollman * detached protocol state from the socket state, but this is no longer the 5729759Swollman * case. 5829759Swollman * 5929759Swollman * socreate() creates a socket and attaches protocol state. This is a public 6029759Swollman * interface that may be used by socket layer consumers to create new 6129759Swollman * sockets. 6229759Swollman * 6329759Swollman * sonewconn() creates a socket and attaches protocol state. This is a 6429759Swollman * public interface that may be used by protocols to create new sockets when 6529759Swollman * a new connection is received and will be available for accept() on a 6629759Swollman * listen socket. 6729759Swollman * 6829759Swollman * soclose() destroys a socket after possibly waiting for it to disconnect. 69158160Sbde * This is a public interface that socket consumers should use to close and 70158160Sbde * release a socket when done with it. 71158160Sbde * 72158160Sbde * soabort() destroys a socket without waiting for it to disconnect (used 73158160Sbde * only for incoming connections that are already partially or fully 74158160Sbde * connected). This is used internally by the socket layer when clearing 75158160Sbde * listen socket queues (due to overflow or close on the listen socket), but 76158160Sbde * is also a public interface protocols may use to abort connections in 77158160Sbde * their incomplete listen queues should they no longer be required. Sockets 78158160Sbde * placed in completed connection listen queues should not be aborted for 79158160Sbde * reasons described in the comment above the soclose() implementation. This 80158160Sbde * is not a general purpose close routine, and except in the specific 81158160Sbde * circumstances described here, should not be used. 82158160Sbde * 83158160Sbde * sofree() will free a socket and its protocol state if all references on 84158160Sbde * the socket have been released, and is the public interface to attempt to 85158160Sbde * free a socket when a reference is removed. This is a socket layer private 86158160Sbde * interface. 8729759Swollman * 8829759Swollman * NOTE: In addition to socreate() and soclose(), which provide a single 8929759Swollman * socket reference to the consumer to be managed as required, there are two 9029759Swollman * calls to explicitly manage socket references, soref(), and sorele(). 9129759Swollman * Currently, these are generally required only when transitioning a socket 9229759Swollman * from a listen queue to a file descriptor, in order to prevent garbage 9329759Swollman * collection of the socket at an untimely moment. For a number of reasons, 9429759Swollman * these interfaces are not preferred, and should be avoided. 95158160Sbde */ 9629759Swollman 9729759Swollman#include <sys/cdefs.h> 9829759Swollman__FBSDID("$FreeBSD: head/sys/kern/uipc_socket.c 218559 2011-02-11 13:27:00Z bz $"); 99175387Sdelphij 10029759Swollman#include "opt_inet.h" 10129759Swollman#include "opt_inet6.h" 10229759Swollman#include "opt_zero.h" 10329759Swollman#include "opt_compat.h" 10429759Swollman 10529759Swollman#include <sys/param.h> 10629759Swollman#include <sys/systm.h> 10729759Swollman#include <sys/fcntl.h> 10829759Swollman#include <sys/limits.h> 10929759Swollman#include <sys/lock.h> 11029759Swollman#include <sys/mac.h> 11129759Swollman#include <sys/malloc.h> 11229759Swollman#include <sys/mbuf.h> 11329759Swollman#include <sys/mutex.h> 114158160Sbde#include <sys/domain.h> 115158160Sbde#include <sys/file.h> /* for struct knote */ 116158160Sbde#include <sys/kernel.h> 117158160Sbde#include <sys/event.h> 118158160Sbde#include <sys/eventhandler.h> 119158160Sbde#include <sys/poll.h> 120158160Sbde#include <sys/proc.h> 121158160Sbde#include <sys/protosw.h> 122158160Sbde#include <sys/socket.h> 12329759Swollman#include <sys/socketvar.h> 124158160Sbde#include <sys/resourcevar.h> 125158160Sbde#include <net/route.h> 126158160Sbde#include <sys/signalvar.h> 127158160Sbde#include <sys/stat.h> 128158160Sbde#include <sys/sx.h> 129158160Sbde#include <sys/sysctl.h> 130158160Sbde#include <sys/uio.h> 13129759Swollman#include <sys/jail.h> 13229759Swollman 13329759Swollman#include <net/vnet.h> 13429759Swollman 13529759Swollman#include <security/mac/mac_framework.h> 13629759Swollman 13729759Swollman#include <vm/uma.h> 13829759Swollman 13929759Swollman#ifdef COMPAT_FREEBSD32 14029759Swollman#include <sys/mount.h> 14129759Swollman#include <sys/sysent.h> 14229759Swollman#include <compat/freebsd32/freebsd32.h> 14329759Swollman#endif 14429759Swollman 14529759Swollmanstatic int soreceive_rcvoob(struct socket *so, struct uio *uio, 14629759Swollman int flags); 14729759Swollman 14829759Swollmanstatic void filt_sordetach(struct knote *kn); 14929759Swollmanstatic int filt_soread(struct knote *kn, long hint); 15029759Swollmanstatic void filt_sowdetach(struct knote *kn); 15129759Swollmanstatic int filt_sowrite(struct knote *kn, long hint); 15229759Swollmanstatic int filt_solisten(struct knote *kn, long hint); 15329759Swollman 15429759Swollmanstatic struct filterops solisten_filtops = { 15529759Swollman .f_isfd = 1, 15629759Swollman .f_detach = filt_sordetach, 15729759Swollman .f_event = filt_solisten, 15829759Swollman}; 15929759Swollmanstatic struct filterops soread_filtops = { 16029759Swollman .f_isfd = 1, 16129759Swollman .f_detach = filt_sordetach, 16229759Swollman .f_event = filt_soread, 16329759Swollman}; 16429759Swollmanstatic struct filterops sowrite_filtops = { 16529759Swollman .f_isfd = 1, 16629759Swollman .f_detach = filt_sowdetach, 16729759Swollman .f_event = filt_sowrite, 16829759Swollman}; 16929759Swollman 17029759Swollmanuma_zone_t socket_zone; 17129759Swollmanso_gen_t so_gencnt; /* generation count for sockets */ 17229759Swollman 17329759Swollmanint maxsockets; 17429759Swollman 17529759SwollmanMALLOC_DEFINE(M_SONAME, "soname", "socket name"); 176158161SbdeMALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 17729759Swollman 17829759Swollmanstatic int somaxconn = SOMAXCONN; 17929759Swollmanstatic int sysctl_somaxconn(SYSCTL_HANDLER_ARGS); 18029759Swollman/* XXX: we dont have SYSCTL_USHORT */ 18129759SwollmanSYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW, 18229759Swollman 0, sizeof(int), sysctl_somaxconn, "I", "Maximum pending socket connection " 18329759Swollman "queue size"); 18429759Swollmanstatic int numopensockets; 18529759SwollmanSYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD, 18629759Swollman &numopensockets, 0, "Number of open sockets"); 18729759Swollman#ifdef ZERO_COPY_SOCKETS 18829759Swollman/* These aren't static because they're used in other files. */ 18929759Swollmanint so_zero_copy_send = 1; 190158161Sbdeint so_zero_copy_receive = 1; 19129759SwollmanSYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0, 192158160Sbde "Zero copy controls"); 193158160SbdeSYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW, 19429759Swollman &so_zero_copy_receive, 0, "Enable zero copy receive"); 19529759SwollmanSYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW, 19629759Swollman &so_zero_copy_send, 0, "Enable zero copy send"); 19729759Swollman#endif /* ZERO_COPY_SOCKETS */ 198158160Sbde 199158160Sbde/* 200158160Sbde * accept_mtx locks down per-socket fields relating to accept queues. See 201158160Sbde * socketvar.h for an annotation of the protected fields of struct socket. 202158160Sbde */ 203158160Sbdestruct mtx accept_mtx; 204158160SbdeMTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF); 205158160Sbde 206158160Sbde/* 207158160Sbde * so_global_mtx protects so_gencnt, numopensockets, and the per-socket 20829759Swollman * so_gencnt field. 20929759Swollman */ 210158160Sbdestatic struct mtx so_global_mtx; 211158160SbdeMTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF); 212158160Sbde 213158160Sbde/* 214158160Sbde * General IPC sysctl name space, used by sockets and a variety of other IPC 215158160Sbde * types. 216158160Sbde */ 217158160SbdeSYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); 21829759Swollman 21929759Swollman/* 22029759Swollman * Sysctl to get and set the maximum global sockets limit. Notify protocols 22129759Swollman * of the change so that they can update their dependent limits as required. 22229759Swollman */ 22329759Swollmanstatic int 22429759Swollmansysctl_maxsockets(SYSCTL_HANDLER_ARGS) 22529759Swollman{ 22629759Swollman int error, newmaxsockets; 22729759Swollman 22829759Swollman newmaxsockets = maxsockets; 22929759Swollman error = sysctl_handle_int(oidp, &newmaxsockets, 0, req); 23029759Swollman if (error == 0 && req->newptr) { 23129759Swollman if (newmaxsockets > maxsockets) { 23229759Swollman maxsockets = newmaxsockets; 23329759Swollman if (maxsockets > ((maxfiles / 4) * 3)) { 23429759Swollman maxfiles = (maxsockets * 5) / 4; 23529759Swollman maxfilesperproc = (maxfiles * 9) / 10; 23629759Swollman } 23729759Swollman EVENTHANDLER_INVOKE(maxsockets_change); 23829759Swollman } else 23929759Swollman error = EINVAL; 24029759Swollman } 24129759Swollman return (error); 24229759Swollman} 24329759Swollman 24429759SwollmanSYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW, 24529759Swollman &maxsockets, 0, sysctl_maxsockets, "IU", 24629759Swollman "Maximum number of sockets avaliable"); 24729759Swollman 24829759Swollman/* 24929759Swollman * Initialise maxsockets. This SYSINIT must be run after 25029759Swollman * tunable_mbinit(). 25129759Swollman */ 25229759Swollmanstatic void 25329759Swollmaninit_maxsockets(void *ignored) 25429759Swollman{ 25529759Swollman 25629759Swollman TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 25729759Swollman maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters)); 25829759Swollman} 25929759SwollmanSYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); 26029759Swollman 26129759Swollman/* 26229759Swollman * Socket operation routines. These routines are called by the routines in 26329759Swollman * sys_socket.c or from a system process, and implement the semantics of 26429759Swollman * socket operations by switching out to the protocol specific routines. 26529759Swollman */ 26629759Swollman 26729759Swollman/* 26829759Swollman * Get a socket structure from our zone, and initialize it. Note that it 26929759Swollman * would probably be better to allocate socket and PCB at the same time, but 27029759Swollman * I'm not convinced that all the protocols can be easily modified to do 27129759Swollman * this. 27229759Swollman * 27329759Swollman * soalloc() returns a socket with a ref count of 0. 27429759Swollman */ 27529759Swollmanstatic struct socket * 27629759Swollmansoalloc(struct vnet *vnet) 27729759Swollman{ 27829759Swollman struct socket *so; 27929759Swollman 28029759Swollman so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO); 28129759Swollman if (so == NULL) 28229759Swollman return (NULL); 28329759Swollman#ifdef MAC 28429759Swollman if (mac_socket_init(so, M_NOWAIT) != 0) { 285 uma_zfree(socket_zone, so); 286 return (NULL); 287 } 288#endif 289 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd"); 290 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv"); 291 sx_init(&so->so_snd.sb_sx, "so_snd_sx"); 292 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx"); 293 TAILQ_INIT(&so->so_aiojobq); 294 mtx_lock(&so_global_mtx); 295 so->so_gencnt = ++so_gencnt; 296 ++numopensockets; 297#ifdef VIMAGE 298 vnet->vnet_sockcnt++; 299 so->so_vnet = vnet; 300#endif 301 mtx_unlock(&so_global_mtx); 302 return (so); 303} 304 305/* 306 * Free the storage associated with a socket at the socket layer, tear down 307 * locks, labels, etc. All protocol state is assumed already to have been 308 * torn down (and possibly never set up) by the caller. 309 */ 310static void 311sodealloc(struct socket *so) 312{ 313 314 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); 315 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL")); 316 317 mtx_lock(&so_global_mtx); 318 so->so_gencnt = ++so_gencnt; 319 --numopensockets; /* Could be below, but faster here. */ 320#ifdef VIMAGE 321 so->so_vnet->vnet_sockcnt--; 322#endif 323 mtx_unlock(&so_global_mtx); 324 if (so->so_rcv.sb_hiwat) 325 (void)chgsbsize(so->so_cred->cr_uidinfo, 326 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 327 if (so->so_snd.sb_hiwat) 328 (void)chgsbsize(so->so_cred->cr_uidinfo, 329 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 330#ifdef INET 331 /* remove acccept filter if one is present. */ 332 if (so->so_accf != NULL) 333 do_setopt_accept_filter(so, NULL); 334#endif 335#ifdef MAC 336 mac_socket_destroy(so); 337#endif 338 crfree(so->so_cred); 339 sx_destroy(&so->so_snd.sb_sx); 340 sx_destroy(&so->so_rcv.sb_sx); 341 SOCKBUF_LOCK_DESTROY(&so->so_snd); 342 SOCKBUF_LOCK_DESTROY(&so->so_rcv); 343 uma_zfree(socket_zone, so); 344} 345 346/* 347 * socreate returns a socket with a ref count of 1. The socket should be 348 * closed with soclose(). 349 */ 350int 351socreate(int dom, struct socket **aso, int type, int proto, 352 struct ucred *cred, struct thread *td) 353{ 354 struct protosw *prp; 355 struct socket *so; 356 int error; 357 358 if (proto) 359 prp = pffindproto(dom, proto, type); 360 else 361 prp = pffindtype(dom, type); 362 363 if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL || 364 prp->pr_usrreqs->pru_attach == pru_attach_notsupp) 365 return (EPROTONOSUPPORT); 366 367 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0) 368 return (EPROTONOSUPPORT); 369 370 if (prp->pr_type != type) 371 return (EPROTOTYPE); 372 so = soalloc(CRED_TO_VNET(cred)); 373 if (so == NULL) 374 return (ENOBUFS); 375 376 TAILQ_INIT(&so->so_incomp); 377 TAILQ_INIT(&so->so_comp); 378 so->so_type = type; 379 so->so_cred = crhold(cred); 380 if ((prp->pr_domain->dom_family == PF_INET) || 381 (prp->pr_domain->dom_family == PF_ROUTE)) 382 so->so_fibnum = td->td_proc->p_fibnum; 383 else 384 so->so_fibnum = 0; 385 so->so_proto = prp; 386#ifdef MAC 387 mac_socket_create(cred, so); 388#endif 389 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv)); 390 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd)); 391 so->so_count = 1; 392 /* 393 * Auto-sizing of socket buffers is managed by the protocols and 394 * the appropriate flags must be set in the pru_attach function. 395 */ 396 CURVNET_SET(so->so_vnet); 397 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); 398 CURVNET_RESTORE(); 399 if (error) { 400 KASSERT(so->so_count == 1, ("socreate: so_count %d", 401 so->so_count)); 402 so->so_count = 0; 403 sodealloc(so); 404 return (error); 405 } 406 *aso = so; 407 return (0); 408} 409 410#ifdef REGRESSION 411static int regression_sonewconn_earlytest = 1; 412SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW, 413 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test"); 414#endif 415 416/* 417 * When an attempt at a new connection is noted on a socket which accepts 418 * connections, sonewconn is called. If the connection is possible (subject 419 * to space constraints, etc.) then we allocate a new structure, propoerly 420 * linked into the data structure of the original socket, and return this. 421 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. 422 * 423 * Note: the ref count on the socket is 0 on return. 424 */ 425struct socket * 426sonewconn(struct socket *head, int connstatus) 427{ 428 struct socket *so; 429 int over; 430 431 ACCEPT_LOCK(); 432 over = (head->so_qlen > 3 * head->so_qlimit / 2); 433 ACCEPT_UNLOCK(); 434#ifdef REGRESSION 435 if (regression_sonewconn_earlytest && over) 436#else 437 if (over) 438#endif 439 return (NULL); 440 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p", 441 __func__, __LINE__, head)); 442 so = soalloc(head->so_vnet); 443 if (so == NULL) 444 return (NULL); 445 if ((head->so_options & SO_ACCEPTFILTER) != 0) 446 connstatus = 0; 447 so->so_head = head; 448 so->so_type = head->so_type; 449 so->so_options = head->so_options &~ SO_ACCEPTCONN; 450 so->so_linger = head->so_linger; 451 so->so_state = head->so_state | SS_NOFDREF; 452 so->so_fibnum = head->so_fibnum; 453 so->so_proto = head->so_proto; 454 so->so_cred = crhold(head->so_cred); 455#ifdef MAC 456 mac_socket_newconn(head, so); 457#endif 458 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv)); 459 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd)); 460 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) || 461 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { 462 sodealloc(so); 463 return (NULL); 464 } 465 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 466 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 467 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 468 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 469 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; 470 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; 471 so->so_state |= connstatus; 472 ACCEPT_LOCK(); 473 if (connstatus) { 474 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 475 so->so_qstate |= SQ_COMP; 476 head->so_qlen++; 477 } else { 478 /* 479 * Keep removing sockets from the head until there's room for 480 * us to insert on the tail. In pre-locking revisions, this 481 * was a simple if(), but as we could be racing with other 482 * threads and soabort() requires dropping locks, we must 483 * loop waiting for the condition to be true. 484 */ 485 while (head->so_incqlen > head->so_qlimit) { 486 struct socket *sp; 487 sp = TAILQ_FIRST(&head->so_incomp); 488 TAILQ_REMOVE(&head->so_incomp, sp, so_list); 489 head->so_incqlen--; 490 sp->so_qstate &= ~SQ_INCOMP; 491 sp->so_head = NULL; 492 ACCEPT_UNLOCK(); 493 soabort(sp); 494 ACCEPT_LOCK(); 495 } 496 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); 497 so->so_qstate |= SQ_INCOMP; 498 head->so_incqlen++; 499 } 500 ACCEPT_UNLOCK(); 501 if (connstatus) { 502 sorwakeup(head); 503 wakeup_one(&head->so_timeo); 504 } 505 return (so); 506} 507 508int 509sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 510{ 511 int error; 512 513 CURVNET_SET(so->so_vnet); 514 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td); 515 CURVNET_RESTORE(); 516 return error; 517} 518 519/* 520 * solisten() transitions a socket from a non-listening state to a listening 521 * state, but can also be used to update the listen queue depth on an 522 * existing listen socket. The protocol will call back into the sockets 523 * layer using solisten_proto_check() and solisten_proto() to check and set 524 * socket-layer listen state. Call backs are used so that the protocol can 525 * acquire both protocol and socket layer locks in whatever order is required 526 * by the protocol. 527 * 528 * Protocol implementors are advised to hold the socket lock across the 529 * socket-layer test and set to avoid races at the socket layer. 530 */ 531int 532solisten(struct socket *so, int backlog, struct thread *td) 533{ 534 535 return ((*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td)); 536} 537 538int 539solisten_proto_check(struct socket *so) 540{ 541 542 SOCK_LOCK_ASSERT(so); 543 544 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 545 SS_ISDISCONNECTING)) 546 return (EINVAL); 547 return (0); 548} 549 550void 551solisten_proto(struct socket *so, int backlog) 552{ 553 554 SOCK_LOCK_ASSERT(so); 555 556 if (backlog < 0 || backlog > somaxconn) 557 backlog = somaxconn; 558 so->so_qlimit = backlog; 559 so->so_options |= SO_ACCEPTCONN; 560} 561 562/* 563 * Evaluate the reference count and named references on a socket; if no 564 * references remain, free it. This should be called whenever a reference is 565 * released, such as in sorele(), but also when named reference flags are 566 * cleared in socket or protocol code. 567 * 568 * sofree() will free the socket if: 569 * 570 * - There are no outstanding file descriptor references or related consumers 571 * (so_count == 0). 572 * 573 * - The socket has been closed by user space, if ever open (SS_NOFDREF). 574 * 575 * - The protocol does not have an outstanding strong reference on the socket 576 * (SS_PROTOREF). 577 * 578 * - The socket is not in a completed connection queue, so a process has been 579 * notified that it is present. If it is removed, the user process may 580 * block in accept() despite select() saying the socket was ready. 581 */ 582void 583sofree(struct socket *so) 584{ 585 struct protosw *pr = so->so_proto; 586 struct socket *head; 587 588 ACCEPT_LOCK_ASSERT(); 589 SOCK_LOCK_ASSERT(so); 590 591 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 || 592 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) { 593 SOCK_UNLOCK(so); 594 ACCEPT_UNLOCK(); 595 return; 596 } 597 598 head = so->so_head; 599 if (head != NULL) { 600 KASSERT((so->so_qstate & SQ_COMP) != 0 || 601 (so->so_qstate & SQ_INCOMP) != 0, 602 ("sofree: so_head != NULL, but neither SQ_COMP nor " 603 "SQ_INCOMP")); 604 KASSERT((so->so_qstate & SQ_COMP) == 0 || 605 (so->so_qstate & SQ_INCOMP) == 0, 606 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP")); 607 TAILQ_REMOVE(&head->so_incomp, so, so_list); 608 head->so_incqlen--; 609 so->so_qstate &= ~SQ_INCOMP; 610 so->so_head = NULL; 611 } 612 KASSERT((so->so_qstate & SQ_COMP) == 0 && 613 (so->so_qstate & SQ_INCOMP) == 0, 614 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)", 615 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP)); 616 if (so->so_options & SO_ACCEPTCONN) { 617 KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated")); 618 KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_comp populated")); 619 } 620 SOCK_UNLOCK(so); 621 ACCEPT_UNLOCK(); 622 623 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 624 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb); 625 if (pr->pr_usrreqs->pru_detach != NULL) 626 (*pr->pr_usrreqs->pru_detach)(so); 627 628 /* 629 * From this point on, we assume that no other references to this 630 * socket exist anywhere else in the stack. Therefore, no locks need 631 * to be acquired or held. 632 * 633 * We used to do a lot of socket buffer and socket locking here, as 634 * well as invoke sorflush() and perform wakeups. The direct call to 635 * dom_dispose() and sbrelease_internal() are an inlining of what was 636 * necessary from sorflush(). 637 * 638 * Notice that the socket buffer and kqueue state are torn down 639 * before calling pru_detach. This means that protocols shold not 640 * assume they can perform socket wakeups, etc, in their detach code. 641 */ 642 sbdestroy(&so->so_snd, so); 643 sbdestroy(&so->so_rcv, so); 644 knlist_destroy(&so->so_rcv.sb_sel.si_note); 645 knlist_destroy(&so->so_snd.sb_sel.si_note); 646 sodealloc(so); 647} 648 649/* 650 * Close a socket on last file table reference removal. Initiate disconnect 651 * if connected. Free socket when disconnect complete. 652 * 653 * This function will sorele() the socket. Note that soclose() may be called 654 * prior to the ref count reaching zero. The actual socket structure will 655 * not be freed until the ref count reaches zero. 656 */ 657int 658soclose(struct socket *so) 659{ 660 int error = 0; 661 662 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter")); 663 664 CURVNET_SET(so->so_vnet); 665 funsetown(&so->so_sigio); 666 if (so->so_state & SS_ISCONNECTED) { 667 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 668 error = sodisconnect(so); 669 if (error) { 670 if (error == ENOTCONN) 671 error = 0; 672 goto drop; 673 } 674 } 675 if (so->so_options & SO_LINGER) { 676 if ((so->so_state & SS_ISDISCONNECTING) && 677 (so->so_state & SS_NBIO)) 678 goto drop; 679 while (so->so_state & SS_ISCONNECTED) { 680 error = tsleep(&so->so_timeo, 681 PSOCK | PCATCH, "soclos", so->so_linger * hz); 682 if (error) 683 break; 684 } 685 } 686 } 687 688drop: 689 if (so->so_proto->pr_usrreqs->pru_close != NULL) 690 (*so->so_proto->pr_usrreqs->pru_close)(so); 691 if (so->so_options & SO_ACCEPTCONN) { 692 struct socket *sp; 693 ACCEPT_LOCK(); 694 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) { 695 TAILQ_REMOVE(&so->so_incomp, sp, so_list); 696 so->so_incqlen--; 697 sp->so_qstate &= ~SQ_INCOMP; 698 sp->so_head = NULL; 699 ACCEPT_UNLOCK(); 700 soabort(sp); 701 ACCEPT_LOCK(); 702 } 703 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) { 704 TAILQ_REMOVE(&so->so_comp, sp, so_list); 705 so->so_qlen--; 706 sp->so_qstate &= ~SQ_COMP; 707 sp->so_head = NULL; 708 ACCEPT_UNLOCK(); 709 soabort(sp); 710 ACCEPT_LOCK(); 711 } 712 ACCEPT_UNLOCK(); 713 } 714 ACCEPT_LOCK(); 715 SOCK_LOCK(so); 716 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF")); 717 so->so_state |= SS_NOFDREF; 718 sorele(so); 719 CURVNET_RESTORE(); 720 return (error); 721} 722 723/* 724 * soabort() is used to abruptly tear down a connection, such as when a 725 * resource limit is reached (listen queue depth exceeded), or if a listen 726 * socket is closed while there are sockets waiting to be accepted. 727 * 728 * This interface is tricky, because it is called on an unreferenced socket, 729 * and must be called only by a thread that has actually removed the socket 730 * from the listen queue it was on, or races with other threads are risked. 731 * 732 * This interface will call into the protocol code, so must not be called 733 * with any socket locks held. Protocols do call it while holding their own 734 * recursible protocol mutexes, but this is something that should be subject 735 * to review in the future. 736 */ 737void 738soabort(struct socket *so) 739{ 740 741 /* 742 * In as much as is possible, assert that no references to this 743 * socket are held. This is not quite the same as asserting that the 744 * current thread is responsible for arranging for no references, but 745 * is as close as we can get for now. 746 */ 747 KASSERT(so->so_count == 0, ("soabort: so_count")); 748 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF")); 749 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF")); 750 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP")); 751 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP")); 752 753 if (so->so_proto->pr_usrreqs->pru_abort != NULL) 754 (*so->so_proto->pr_usrreqs->pru_abort)(so); 755 ACCEPT_LOCK(); 756 SOCK_LOCK(so); 757 sofree(so); 758} 759 760int 761soaccept(struct socket *so, struct sockaddr **nam) 762{ 763 int error; 764 765 SOCK_LOCK(so); 766 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF")); 767 so->so_state &= ~SS_NOFDREF; 768 SOCK_UNLOCK(so); 769 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); 770 return (error); 771} 772 773int 774soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 775{ 776 int error; 777 778 if (so->so_options & SO_ACCEPTCONN) 779 return (EOPNOTSUPP); 780 781 CURVNET_SET(so->so_vnet); 782 /* 783 * If protocol is connection-based, can only connect once. 784 * Otherwise, if connected, try to disconnect first. This allows 785 * user to disconnect by connecting to, e.g., a null address. 786 */ 787 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 788 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 789 (error = sodisconnect(so)))) { 790 error = EISCONN; 791 } else { 792 /* 793 * Prevent accumulated error from previous connection from 794 * biting us. 795 */ 796 so->so_error = 0; 797 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td); 798 } 799 CURVNET_RESTORE(); 800 801 return (error); 802} 803 804int 805soconnect2(struct socket *so1, struct socket *so2) 806{ 807 808 return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2)); 809} 810 811int 812sodisconnect(struct socket *so) 813{ 814 int error; 815 816 if ((so->so_state & SS_ISCONNECTED) == 0) 817 return (ENOTCONN); 818 if (so->so_state & SS_ISDISCONNECTING) 819 return (EALREADY); 820 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); 821 return (error); 822} 823 824#ifdef ZERO_COPY_SOCKETS 825struct so_zerocopy_stats{ 826 int size_ok; 827 int align_ok; 828 int found_ifp; 829}; 830struct so_zerocopy_stats so_zerocp_stats = {0,0,0}; 831#include <netinet/in.h> 832#include <net/route.h> 833#include <netinet/in_pcb.h> 834#include <vm/vm.h> 835#include <vm/vm_page.h> 836#include <vm/vm_object.h> 837 838/* 839 * sosend_copyin() is only used if zero copy sockets are enabled. Otherwise 840 * sosend_dgram() and sosend_generic() use m_uiotombuf(). 841 * 842 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or 843 * all of the data referenced by the uio. If desired, it uses zero-copy. 844 * *space will be updated to reflect data copied in. 845 * 846 * NB: If atomic I/O is requested, the caller must already have checked that 847 * space can hold resid bytes. 848 * 849 * NB: In the event of an error, the caller may need to free the partial 850 * chain pointed to by *mpp. The contents of both *uio and *space may be 851 * modified even in the case of an error. 852 */ 853static int 854sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space, 855 int flags) 856{ 857 struct mbuf *m, **mp, *top; 858 long len, resid; 859 int error; 860#ifdef ZERO_COPY_SOCKETS 861 int cow_send; 862#endif 863 864 *retmp = top = NULL; 865 mp = ⊤ 866 len = 0; 867 resid = uio->uio_resid; 868 error = 0; 869 do { 870#ifdef ZERO_COPY_SOCKETS 871 cow_send = 0; 872#endif /* ZERO_COPY_SOCKETS */ 873 if (resid >= MINCLSIZE) { 874#ifdef ZERO_COPY_SOCKETS 875 if (top == NULL) { 876 m = m_gethdr(M_WAITOK, MT_DATA); 877 m->m_pkthdr.len = 0; 878 m->m_pkthdr.rcvif = NULL; 879 } else 880 m = m_get(M_WAITOK, MT_DATA); 881 if (so_zero_copy_send && 882 resid>=PAGE_SIZE && 883 *space>=PAGE_SIZE && 884 uio->uio_iov->iov_len>=PAGE_SIZE) { 885 so_zerocp_stats.size_ok++; 886 so_zerocp_stats.align_ok++; 887 cow_send = socow_setup(m, uio); 888 len = cow_send; 889 } 890 if (!cow_send) { 891 m_clget(m, M_WAITOK); 892 len = min(min(MCLBYTES, resid), *space); 893 } 894#else /* ZERO_COPY_SOCKETS */ 895 if (top == NULL) { 896 m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR); 897 m->m_pkthdr.len = 0; 898 m->m_pkthdr.rcvif = NULL; 899 } else 900 m = m_getcl(M_WAIT, MT_DATA, 0); 901 len = min(min(MCLBYTES, resid), *space); 902#endif /* ZERO_COPY_SOCKETS */ 903 } else { 904 if (top == NULL) { 905 m = m_gethdr(M_WAIT, MT_DATA); 906 m->m_pkthdr.len = 0; 907 m->m_pkthdr.rcvif = NULL; 908 909 len = min(min(MHLEN, resid), *space); 910 /* 911 * For datagram protocols, leave room 912 * for protocol headers in first mbuf. 913 */ 914 if (atomic && m && len < MHLEN) 915 MH_ALIGN(m, len); 916 } else { 917 m = m_get(M_WAIT, MT_DATA); 918 len = min(min(MLEN, resid), *space); 919 } 920 } 921 if (m == NULL) { 922 error = ENOBUFS; 923 goto out; 924 } 925 926 *space -= len; 927#ifdef ZERO_COPY_SOCKETS 928 if (cow_send) 929 error = 0; 930 else 931#endif /* ZERO_COPY_SOCKETS */ 932 error = uiomove(mtod(m, void *), (int)len, uio); 933 resid = uio->uio_resid; 934 m->m_len = len; 935 *mp = m; 936 top->m_pkthdr.len += len; 937 if (error) 938 goto out; 939 mp = &m->m_next; 940 if (resid <= 0) { 941 if (flags & MSG_EOR) 942 top->m_flags |= M_EOR; 943 break; 944 } 945 } while (*space > 0 && atomic); 946out: 947 *retmp = top; 948 return (error); 949} 950#endif /*ZERO_COPY_SOCKETS*/ 951 952#define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) 953 954int 955sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 956 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 957{ 958 long space, resid; 959 int clen = 0, error, dontroute; 960#ifdef ZERO_COPY_SOCKETS 961 int atomic = sosendallatonce(so) || top; 962#endif 963 964 KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM")); 965 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 966 ("sodgram_send: !PR_ATOMIC")); 967 968 if (uio != NULL) 969 resid = uio->uio_resid; 970 else 971 resid = top->m_pkthdr.len; 972 /* 973 * In theory resid should be unsigned. However, space must be 974 * signed, as it might be less than 0 if we over-committed, and we 975 * must use a signed comparison of space and resid. On the other 976 * hand, a negative resid causes us to loop sending 0-length 977 * segments to the protocol. 978 */ 979 if (resid < 0) { 980 error = EINVAL; 981 goto out; 982 } 983 984 dontroute = 985 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 986 if (td != NULL) 987 td->td_ru.ru_msgsnd++; 988 if (control != NULL) 989 clen = control->m_len; 990 991 SOCKBUF_LOCK(&so->so_snd); 992 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 993 SOCKBUF_UNLOCK(&so->so_snd); 994 error = EPIPE; 995 goto out; 996 } 997 if (so->so_error) { 998 error = so->so_error; 999 so->so_error = 0; 1000 SOCKBUF_UNLOCK(&so->so_snd); 1001 goto out; 1002 } 1003 if ((so->so_state & SS_ISCONNECTED) == 0) { 1004 /* 1005 * `sendto' and `sendmsg' is allowed on a connection-based 1006 * socket if it supports implied connect. Return ENOTCONN if 1007 * not connected and no address is supplied. 1008 */ 1009 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1010 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1011 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1012 !(resid == 0 && clen != 0)) { 1013 SOCKBUF_UNLOCK(&so->so_snd); 1014 error = ENOTCONN; 1015 goto out; 1016 } 1017 } else if (addr == NULL) { 1018 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1019 error = ENOTCONN; 1020 else 1021 error = EDESTADDRREQ; 1022 SOCKBUF_UNLOCK(&so->so_snd); 1023 goto out; 1024 } 1025 } 1026 1027 /* 1028 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 1029 * problem and need fixing. 1030 */ 1031 space = sbspace(&so->so_snd); 1032 if (flags & MSG_OOB) 1033 space += 1024; 1034 space -= clen; 1035 SOCKBUF_UNLOCK(&so->so_snd); 1036 if (resid > space) { 1037 error = EMSGSIZE; 1038 goto out; 1039 } 1040 if (uio == NULL) { 1041 resid = 0; 1042 if (flags & MSG_EOR) 1043 top->m_flags |= M_EOR; 1044 } else { 1045#ifdef ZERO_COPY_SOCKETS 1046 error = sosend_copyin(uio, &top, atomic, &space, flags); 1047 if (error) 1048 goto out; 1049#else 1050 /* 1051 * Copy the data from userland into a mbuf chain. 1052 * If no data is to be copied in, a single empty mbuf 1053 * is returned. 1054 */ 1055 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 1056 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 1057 if (top == NULL) { 1058 error = EFAULT; /* only possible error */ 1059 goto out; 1060 } 1061 space -= resid - uio->uio_resid; 1062#endif 1063 resid = uio->uio_resid; 1064 } 1065 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 1066 /* 1067 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 1068 * than with. 1069 */ 1070 if (dontroute) { 1071 SOCK_LOCK(so); 1072 so->so_options |= SO_DONTROUTE; 1073 SOCK_UNLOCK(so); 1074 } 1075 /* 1076 * XXX all the SBS_CANTSENDMORE checks previously done could be out 1077 * of date. We could have recieved a reset packet in an interrupt or 1078 * maybe we slept while doing page faults in uiomove() etc. We could 1079 * probably recheck again inside the locking protection here, but 1080 * there are probably other places that this also happens. We must 1081 * rethink this. 1082 */ 1083 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1084 (flags & MSG_OOB) ? PRUS_OOB : 1085 /* 1086 * If the user set MSG_EOF, the protocol understands this flag and 1087 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 1088 */ 1089 ((flags & MSG_EOF) && 1090 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1091 (resid <= 0)) ? 1092 PRUS_EOF : 1093 /* If there is more to send set PRUS_MORETOCOME */ 1094 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1095 top, addr, control, td); 1096 if (dontroute) { 1097 SOCK_LOCK(so); 1098 so->so_options &= ~SO_DONTROUTE; 1099 SOCK_UNLOCK(so); 1100 } 1101 clen = 0; 1102 control = NULL; 1103 top = NULL; 1104out: 1105 if (top != NULL) 1106 m_freem(top); 1107 if (control != NULL) 1108 m_freem(control); 1109 return (error); 1110} 1111 1112/* 1113 * Send on a socket. If send must go all at once and message is larger than 1114 * send buffering, then hard error. Lock against other senders. If must go 1115 * all at once and not enough room now, then inform user that this would 1116 * block and do nothing. Otherwise, if nonblocking, send as much as 1117 * possible. The data to be sent is described by "uio" if nonzero, otherwise 1118 * by the mbuf chain "top" (which must be null if uio is not). Data provided 1119 * in mbuf chain must be small enough to send all at once. 1120 * 1121 * Returns nonzero on error, timeout or signal; callers must check for short 1122 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 1123 * on return. 1124 */ 1125int 1126sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 1127 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1128{ 1129 long space, resid; 1130 int clen = 0, error, dontroute; 1131 int atomic = sosendallatonce(so) || top; 1132 1133 if (uio != NULL) 1134 resid = uio->uio_resid; 1135 else 1136 resid = top->m_pkthdr.len; 1137 /* 1138 * In theory resid should be unsigned. However, space must be 1139 * signed, as it might be less than 0 if we over-committed, and we 1140 * must use a signed comparison of space and resid. On the other 1141 * hand, a negative resid causes us to loop sending 0-length 1142 * segments to the protocol. 1143 * 1144 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 1145 * type sockets since that's an error. 1146 */ 1147 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 1148 error = EINVAL; 1149 goto out; 1150 } 1151 1152 dontroute = 1153 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 1154 (so->so_proto->pr_flags & PR_ATOMIC); 1155 if (td != NULL) 1156 td->td_ru.ru_msgsnd++; 1157 if (control != NULL) 1158 clen = control->m_len; 1159 1160 error = sblock(&so->so_snd, SBLOCKWAIT(flags)); 1161 if (error) 1162 goto out; 1163 1164restart: 1165 do { 1166 SOCKBUF_LOCK(&so->so_snd); 1167 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1168 SOCKBUF_UNLOCK(&so->so_snd); 1169 error = EPIPE; 1170 goto release; 1171 } 1172 if (so->so_error) { 1173 error = so->so_error; 1174 so->so_error = 0; 1175 SOCKBUF_UNLOCK(&so->so_snd); 1176 goto release; 1177 } 1178 if ((so->so_state & SS_ISCONNECTED) == 0) { 1179 /* 1180 * `sendto' and `sendmsg' is allowed on a connection- 1181 * based socket if it supports implied connect. 1182 * Return ENOTCONN if not connected and no address is 1183 * supplied. 1184 */ 1185 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 1186 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 1187 if ((so->so_state & SS_ISCONFIRMING) == 0 && 1188 !(resid == 0 && clen != 0)) { 1189 SOCKBUF_UNLOCK(&so->so_snd); 1190 error = ENOTCONN; 1191 goto release; 1192 } 1193 } else if (addr == NULL) { 1194 SOCKBUF_UNLOCK(&so->so_snd); 1195 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 1196 error = ENOTCONN; 1197 else 1198 error = EDESTADDRREQ; 1199 goto release; 1200 } 1201 } 1202 space = sbspace(&so->so_snd); 1203 if (flags & MSG_OOB) 1204 space += 1024; 1205 if ((atomic && resid > so->so_snd.sb_hiwat) || 1206 clen > so->so_snd.sb_hiwat) { 1207 SOCKBUF_UNLOCK(&so->so_snd); 1208 error = EMSGSIZE; 1209 goto release; 1210 } 1211 if (space < resid + clen && 1212 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 1213 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) { 1214 SOCKBUF_UNLOCK(&so->so_snd); 1215 error = EWOULDBLOCK; 1216 goto release; 1217 } 1218 error = sbwait(&so->so_snd); 1219 SOCKBUF_UNLOCK(&so->so_snd); 1220 if (error) 1221 goto release; 1222 goto restart; 1223 } 1224 SOCKBUF_UNLOCK(&so->so_snd); 1225 space -= clen; 1226 do { 1227 if (uio == NULL) { 1228 resid = 0; 1229 if (flags & MSG_EOR) 1230 top->m_flags |= M_EOR; 1231 } else { 1232#ifdef ZERO_COPY_SOCKETS 1233 error = sosend_copyin(uio, &top, atomic, 1234 &space, flags); 1235 if (error != 0) 1236 goto release; 1237#else 1238 /* 1239 * Copy the data from userland into a mbuf 1240 * chain. If no data is to be copied in, 1241 * a single empty mbuf is returned. 1242 */ 1243 top = m_uiotombuf(uio, M_WAITOK, space, 1244 (atomic ? max_hdr : 0), 1245 (atomic ? M_PKTHDR : 0) | 1246 ((flags & MSG_EOR) ? M_EOR : 0)); 1247 if (top == NULL) { 1248 error = EFAULT; /* only possible error */ 1249 goto release; 1250 } 1251 space -= resid - uio->uio_resid; 1252#endif 1253 resid = uio->uio_resid; 1254 } 1255 if (dontroute) { 1256 SOCK_LOCK(so); 1257 so->so_options |= SO_DONTROUTE; 1258 SOCK_UNLOCK(so); 1259 } 1260 /* 1261 * XXX all the SBS_CANTSENDMORE checks previously 1262 * done could be out of date. We could have recieved 1263 * a reset packet in an interrupt or maybe we slept 1264 * while doing page faults in uiomove() etc. We 1265 * could probably recheck again inside the locking 1266 * protection here, but there are probably other 1267 * places that this also happens. We must rethink 1268 * this. 1269 */ 1270 error = (*so->so_proto->pr_usrreqs->pru_send)(so, 1271 (flags & MSG_OOB) ? PRUS_OOB : 1272 /* 1273 * If the user set MSG_EOF, the protocol understands 1274 * this flag and nothing left to send then use 1275 * PRU_SEND_EOF instead of PRU_SEND. 1276 */ 1277 ((flags & MSG_EOF) && 1278 (so->so_proto->pr_flags & PR_IMPLOPCL) && 1279 (resid <= 0)) ? 1280 PRUS_EOF : 1281 /* If there is more to send set PRUS_MORETOCOME. */ 1282 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 1283 top, addr, control, td); 1284 if (dontroute) { 1285 SOCK_LOCK(so); 1286 so->so_options &= ~SO_DONTROUTE; 1287 SOCK_UNLOCK(so); 1288 } 1289 clen = 0; 1290 control = NULL; 1291 top = NULL; 1292 if (error) 1293 goto release; 1294 } while (resid && space > 0); 1295 } while (resid); 1296 1297release: 1298 sbunlock(&so->so_snd); 1299out: 1300 if (top != NULL) 1301 m_freem(top); 1302 if (control != NULL) 1303 m_freem(control); 1304 return (error); 1305} 1306 1307int 1308sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 1309 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 1310{ 1311 int error; 1312 1313 CURVNET_SET(so->so_vnet); 1314 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top, 1315 control, flags, td); 1316 CURVNET_RESTORE(); 1317 return (error); 1318} 1319 1320/* 1321 * The part of soreceive() that implements reading non-inline out-of-band 1322 * data from a socket. For more complete comments, see soreceive(), from 1323 * which this code originated. 1324 * 1325 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 1326 * unable to return an mbuf chain to the caller. 1327 */ 1328static int 1329soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 1330{ 1331 struct protosw *pr = so->so_proto; 1332 struct mbuf *m; 1333 int error; 1334 1335 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 1336 1337 m = m_get(M_WAIT, MT_DATA); 1338 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK); 1339 if (error) 1340 goto bad; 1341 do { 1342#ifdef ZERO_COPY_SOCKETS 1343 if (so_zero_copy_receive) { 1344 int disposable; 1345 1346 if ((m->m_flags & M_EXT) 1347 && (m->m_ext.ext_type == EXT_DISPOSABLE)) 1348 disposable = 1; 1349 else 1350 disposable = 0; 1351 1352 error = uiomoveco(mtod(m, void *), 1353 min(uio->uio_resid, m->m_len), 1354 uio, disposable); 1355 } else 1356#endif /* ZERO_COPY_SOCKETS */ 1357 error = uiomove(mtod(m, void *), 1358 (int) min(uio->uio_resid, m->m_len), uio); 1359 m = m_free(m); 1360 } while (uio->uio_resid && error == 0 && m); 1361bad: 1362 if (m != NULL) 1363 m_freem(m); 1364 return (error); 1365} 1366 1367/* 1368 * Following replacement or removal of the first mbuf on the first mbuf chain 1369 * of a socket buffer, push necessary state changes back into the socket 1370 * buffer so that other consumers see the values consistently. 'nextrecord' 1371 * is the callers locally stored value of the original value of 1372 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 1373 * NOTE: 'nextrecord' may be NULL. 1374 */ 1375static __inline void 1376sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 1377{ 1378 1379 SOCKBUF_LOCK_ASSERT(sb); 1380 /* 1381 * First, update for the new value of nextrecord. If necessary, make 1382 * it the first record. 1383 */ 1384 if (sb->sb_mb != NULL) 1385 sb->sb_mb->m_nextpkt = nextrecord; 1386 else 1387 sb->sb_mb = nextrecord; 1388 1389 /* 1390 * Now update any dependent socket buffer fields to reflect the new 1391 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 1392 * addition of a second clause that takes care of the case where 1393 * sb_mb has been updated, but remains the last record. 1394 */ 1395 if (sb->sb_mb == NULL) { 1396 sb->sb_mbtail = NULL; 1397 sb->sb_lastrecord = NULL; 1398 } else if (sb->sb_mb->m_nextpkt == NULL) 1399 sb->sb_lastrecord = sb->sb_mb; 1400} 1401 1402 1403/* 1404 * Implement receive operations on a socket. We depend on the way that 1405 * records are added to the sockbuf by sbappend. In particular, each record 1406 * (mbufs linked through m_next) must begin with an address if the protocol 1407 * so specifies, followed by an optional mbuf or mbufs containing ancillary 1408 * data, and then zero or more mbufs of data. In order to allow parallelism 1409 * between network receive and copying to user space, as well as avoid 1410 * sleeping with a mutex held, we release the socket buffer mutex during the 1411 * user space copy. Although the sockbuf is locked, new data may still be 1412 * appended, and thus we must maintain consistency of the sockbuf during that 1413 * time. 1414 * 1415 * The caller may receive the data as a single mbuf chain by supplying an 1416 * mbuf **mp0 for use in returning the chain. The uio is then used only for 1417 * the count in uio_resid. 1418 */ 1419int 1420soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 1421 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1422{ 1423 struct mbuf *m, **mp; 1424 int flags, len, error, offset; 1425 struct protosw *pr = so->so_proto; 1426 struct mbuf *nextrecord; 1427 int moff, type = 0; 1428 int orig_resid = uio->uio_resid; 1429 1430 mp = mp0; 1431 if (psa != NULL) 1432 *psa = NULL; 1433 if (controlp != NULL) 1434 *controlp = NULL; 1435 if (flagsp != NULL) 1436 flags = *flagsp &~ MSG_EOR; 1437 else 1438 flags = 0; 1439 if (flags & MSG_OOB) 1440 return (soreceive_rcvoob(so, uio, flags)); 1441 if (mp != NULL) 1442 *mp = NULL; 1443 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING) 1444 && uio->uio_resid) 1445 (*pr->pr_usrreqs->pru_rcvd)(so, 0); 1446 1447 error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); 1448 if (error) 1449 return (error); 1450 1451restart: 1452 SOCKBUF_LOCK(&so->so_rcv); 1453 m = so->so_rcv.sb_mb; 1454 /* 1455 * If we have less data than requested, block awaiting more (subject 1456 * to any timeout) if: 1457 * 1. the current count is less than the low water mark, or 1458 * 2. MSG_WAITALL is set, and it is possible to do the entire 1459 * receive operation at once if we block (resid <= hiwat). 1460 * 3. MSG_DONTWAIT is not set 1461 * If MSG_WAITALL is set but resid is larger than the receive buffer, 1462 * we have to do the receive in sections, and thus risk returning a 1463 * short count if a timeout or signal occurs after we start. 1464 */ 1465 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 1466 so->so_rcv.sb_cc < uio->uio_resid) && 1467 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || 1468 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && 1469 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 1470 KASSERT(m != NULL || !so->so_rcv.sb_cc, 1471 ("receive: m == %p so->so_rcv.sb_cc == %u", 1472 m, so->so_rcv.sb_cc)); 1473 if (so->so_error) { 1474 if (m != NULL) 1475 goto dontblock; 1476 error = so->so_error; 1477 if ((flags & MSG_PEEK) == 0) 1478 so->so_error = 0; 1479 SOCKBUF_UNLOCK(&so->so_rcv); 1480 goto release; 1481 } 1482 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1483 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1484 if (m == NULL) { 1485 SOCKBUF_UNLOCK(&so->so_rcv); 1486 goto release; 1487 } else 1488 goto dontblock; 1489 } 1490 for (; m != NULL; m = m->m_next) 1491 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 1492 m = so->so_rcv.sb_mb; 1493 goto dontblock; 1494 } 1495 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 && 1496 (so->so_proto->pr_flags & PR_CONNREQUIRED)) { 1497 SOCKBUF_UNLOCK(&so->so_rcv); 1498 error = ENOTCONN; 1499 goto release; 1500 } 1501 if (uio->uio_resid == 0) { 1502 SOCKBUF_UNLOCK(&so->so_rcv); 1503 goto release; 1504 } 1505 if ((so->so_state & SS_NBIO) || 1506 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 1507 SOCKBUF_UNLOCK(&so->so_rcv); 1508 error = EWOULDBLOCK; 1509 goto release; 1510 } 1511 SBLASTRECORDCHK(&so->so_rcv); 1512 SBLASTMBUFCHK(&so->so_rcv); 1513 error = sbwait(&so->so_rcv); 1514 SOCKBUF_UNLOCK(&so->so_rcv); 1515 if (error) 1516 goto release; 1517 goto restart; 1518 } 1519dontblock: 1520 /* 1521 * From this point onward, we maintain 'nextrecord' as a cache of the 1522 * pointer to the next record in the socket buffer. We must keep the 1523 * various socket buffer pointers and local stack versions of the 1524 * pointers in sync, pushing out modifications before dropping the 1525 * socket buffer mutex, and re-reading them when picking it up. 1526 * 1527 * Otherwise, we will race with the network stack appending new data 1528 * or records onto the socket buffer by using inconsistent/stale 1529 * versions of the field, possibly resulting in socket buffer 1530 * corruption. 1531 * 1532 * By holding the high-level sblock(), we prevent simultaneous 1533 * readers from pulling off the front of the socket buffer. 1534 */ 1535 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1536 if (uio->uio_td) 1537 uio->uio_td->td_ru.ru_msgrcv++; 1538 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 1539 SBLASTRECORDCHK(&so->so_rcv); 1540 SBLASTMBUFCHK(&so->so_rcv); 1541 nextrecord = m->m_nextpkt; 1542 if (pr->pr_flags & PR_ADDR) { 1543 KASSERT(m->m_type == MT_SONAME, 1544 ("m->m_type == %d", m->m_type)); 1545 orig_resid = 0; 1546 if (psa != NULL) 1547 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 1548 M_NOWAIT); 1549 if (flags & MSG_PEEK) { 1550 m = m->m_next; 1551 } else { 1552 sbfree(&so->so_rcv, m); 1553 so->so_rcv.sb_mb = m_free(m); 1554 m = so->so_rcv.sb_mb; 1555 sockbuf_pushsync(&so->so_rcv, nextrecord); 1556 } 1557 } 1558 1559 /* 1560 * Process one or more MT_CONTROL mbufs present before any data mbufs 1561 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 1562 * just copy the data; if !MSG_PEEK, we call into the protocol to 1563 * perform externalization (or freeing if controlp == NULL). 1564 */ 1565 if (m != NULL && m->m_type == MT_CONTROL) { 1566 struct mbuf *cm = NULL, *cmn; 1567 struct mbuf **cme = &cm; 1568 1569 do { 1570 if (flags & MSG_PEEK) { 1571 if (controlp != NULL) { 1572 *controlp = m_copy(m, 0, m->m_len); 1573 controlp = &(*controlp)->m_next; 1574 } 1575 m = m->m_next; 1576 } else { 1577 sbfree(&so->so_rcv, m); 1578 so->so_rcv.sb_mb = m->m_next; 1579 m->m_next = NULL; 1580 *cme = m; 1581 cme = &(*cme)->m_next; 1582 m = so->so_rcv.sb_mb; 1583 } 1584 } while (m != NULL && m->m_type == MT_CONTROL); 1585 if ((flags & MSG_PEEK) == 0) 1586 sockbuf_pushsync(&so->so_rcv, nextrecord); 1587 while (cm != NULL) { 1588 cmn = cm->m_next; 1589 cm->m_next = NULL; 1590 if (pr->pr_domain->dom_externalize != NULL) { 1591 SOCKBUF_UNLOCK(&so->so_rcv); 1592 error = (*pr->pr_domain->dom_externalize) 1593 (cm, controlp); 1594 SOCKBUF_LOCK(&so->so_rcv); 1595 } else if (controlp != NULL) 1596 *controlp = cm; 1597 else 1598 m_freem(cm); 1599 if (controlp != NULL) { 1600 orig_resid = 0; 1601 while (*controlp != NULL) 1602 controlp = &(*controlp)->m_next; 1603 } 1604 cm = cmn; 1605 } 1606 if (m != NULL) 1607 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 1608 else 1609 nextrecord = so->so_rcv.sb_mb; 1610 orig_resid = 0; 1611 } 1612 if (m != NULL) { 1613 if ((flags & MSG_PEEK) == 0) { 1614 KASSERT(m->m_nextpkt == nextrecord, 1615 ("soreceive: post-control, nextrecord !sync")); 1616 if (nextrecord == NULL) { 1617 KASSERT(so->so_rcv.sb_mb == m, 1618 ("soreceive: post-control, sb_mb!=m")); 1619 KASSERT(so->so_rcv.sb_lastrecord == m, 1620 ("soreceive: post-control, lastrecord!=m")); 1621 } 1622 } 1623 type = m->m_type; 1624 if (type == MT_OOBDATA) 1625 flags |= MSG_OOB; 1626 } else { 1627 if ((flags & MSG_PEEK) == 0) { 1628 KASSERT(so->so_rcv.sb_mb == nextrecord, 1629 ("soreceive: sb_mb != nextrecord")); 1630 if (so->so_rcv.sb_mb == NULL) { 1631 KASSERT(so->so_rcv.sb_lastrecord == NULL, 1632 ("soreceive: sb_lastercord != NULL")); 1633 } 1634 } 1635 } 1636 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1637 SBLASTRECORDCHK(&so->so_rcv); 1638 SBLASTMBUFCHK(&so->so_rcv); 1639 1640 /* 1641 * Now continue to read any data mbufs off of the head of the socket 1642 * buffer until the read request is satisfied. Note that 'type' is 1643 * used to store the type of any mbuf reads that have happened so far 1644 * such that soreceive() can stop reading if the type changes, which 1645 * causes soreceive() to return only one of regular data and inline 1646 * out-of-band data in a single socket receive operation. 1647 */ 1648 moff = 0; 1649 offset = 0; 1650 while (m != NULL && uio->uio_resid > 0 && error == 0) { 1651 /* 1652 * If the type of mbuf has changed since the last mbuf 1653 * examined ('type'), end the receive operation. 1654 */ 1655 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1656 if (m->m_type == MT_OOBDATA) { 1657 if (type != MT_OOBDATA) 1658 break; 1659 } else if (type == MT_OOBDATA) 1660 break; 1661 else 1662 KASSERT(m->m_type == MT_DATA, 1663 ("m->m_type == %d", m->m_type)); 1664 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 1665 len = uio->uio_resid; 1666 if (so->so_oobmark && len > so->so_oobmark - offset) 1667 len = so->so_oobmark - offset; 1668 if (len > m->m_len - moff) 1669 len = m->m_len - moff; 1670 /* 1671 * If mp is set, just pass back the mbufs. Otherwise copy 1672 * them out via the uio, then free. Sockbuf must be 1673 * consistent here (points to current mbuf, it points to next 1674 * record) when we drop priority; we must note any additions 1675 * to the sockbuf when we block interrupts again. 1676 */ 1677 if (mp == NULL) { 1678 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1679 SBLASTRECORDCHK(&so->so_rcv); 1680 SBLASTMBUFCHK(&so->so_rcv); 1681 SOCKBUF_UNLOCK(&so->so_rcv); 1682#ifdef ZERO_COPY_SOCKETS 1683 if (so_zero_copy_receive) { 1684 int disposable; 1685 1686 if ((m->m_flags & M_EXT) 1687 && (m->m_ext.ext_type == EXT_DISPOSABLE)) 1688 disposable = 1; 1689 else 1690 disposable = 0; 1691 1692 error = uiomoveco(mtod(m, char *) + moff, 1693 (int)len, uio, 1694 disposable); 1695 } else 1696#endif /* ZERO_COPY_SOCKETS */ 1697 error = uiomove(mtod(m, char *) + moff, (int)len, uio); 1698 SOCKBUF_LOCK(&so->so_rcv); 1699 if (error) { 1700 /* 1701 * The MT_SONAME mbuf has already been removed 1702 * from the record, so it is necessary to 1703 * remove the data mbufs, if any, to preserve 1704 * the invariant in the case of PR_ADDR that 1705 * requires MT_SONAME mbufs at the head of 1706 * each record. 1707 */ 1708 if (m && pr->pr_flags & PR_ATOMIC && 1709 ((flags & MSG_PEEK) == 0)) 1710 (void)sbdroprecord_locked(&so->so_rcv); 1711 SOCKBUF_UNLOCK(&so->so_rcv); 1712 goto release; 1713 } 1714 } else 1715 uio->uio_resid -= len; 1716 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1717 if (len == m->m_len - moff) { 1718 if (m->m_flags & M_EOR) 1719 flags |= MSG_EOR; 1720 if (flags & MSG_PEEK) { 1721 m = m->m_next; 1722 moff = 0; 1723 } else { 1724 nextrecord = m->m_nextpkt; 1725 sbfree(&so->so_rcv, m); 1726 if (mp != NULL) { 1727 *mp = m; 1728 mp = &m->m_next; 1729 so->so_rcv.sb_mb = m = m->m_next; 1730 *mp = NULL; 1731 } else { 1732 so->so_rcv.sb_mb = m_free(m); 1733 m = so->so_rcv.sb_mb; 1734 } 1735 sockbuf_pushsync(&so->so_rcv, nextrecord); 1736 SBLASTRECORDCHK(&so->so_rcv); 1737 SBLASTMBUFCHK(&so->so_rcv); 1738 } 1739 } else { 1740 if (flags & MSG_PEEK) 1741 moff += len; 1742 else { 1743 if (mp != NULL) { 1744 int copy_flag; 1745 1746 if (flags & MSG_DONTWAIT) 1747 copy_flag = M_DONTWAIT; 1748 else 1749 copy_flag = M_WAIT; 1750 if (copy_flag == M_WAIT) 1751 SOCKBUF_UNLOCK(&so->so_rcv); 1752 *mp = m_copym(m, 0, len, copy_flag); 1753 if (copy_flag == M_WAIT) 1754 SOCKBUF_LOCK(&so->so_rcv); 1755 if (*mp == NULL) { 1756 /* 1757 * m_copym() couldn't 1758 * allocate an mbuf. Adjust 1759 * uio_resid back (it was 1760 * adjusted down by len 1761 * bytes, which we didn't end 1762 * up "copying" over). 1763 */ 1764 uio->uio_resid += len; 1765 break; 1766 } 1767 } 1768 m->m_data += len; 1769 m->m_len -= len; 1770 so->so_rcv.sb_cc -= len; 1771 } 1772 } 1773 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1774 if (so->so_oobmark) { 1775 if ((flags & MSG_PEEK) == 0) { 1776 so->so_oobmark -= len; 1777 if (so->so_oobmark == 0) { 1778 so->so_rcv.sb_state |= SBS_RCVATMARK; 1779 break; 1780 } 1781 } else { 1782 offset += len; 1783 if (offset == so->so_oobmark) 1784 break; 1785 } 1786 } 1787 if (flags & MSG_EOR) 1788 break; 1789 /* 1790 * If the MSG_WAITALL flag is set (for non-atomic socket), we 1791 * must not quit until "uio->uio_resid == 0" or an error 1792 * termination. If a signal/timeout occurs, return with a 1793 * short count but without error. Keep sockbuf locked 1794 * against other readers. 1795 */ 1796 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 1797 !sosendallatonce(so) && nextrecord == NULL) { 1798 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1799 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE) 1800 break; 1801 /* 1802 * Notify the protocol that some data has been 1803 * drained before blocking. 1804 */ 1805 if (pr->pr_flags & PR_WANTRCVD) { 1806 SOCKBUF_UNLOCK(&so->so_rcv); 1807 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 1808 SOCKBUF_LOCK(&so->so_rcv); 1809 } 1810 SBLASTRECORDCHK(&so->so_rcv); 1811 SBLASTMBUFCHK(&so->so_rcv); 1812 error = sbwait(&so->so_rcv); 1813 if (error) { 1814 SOCKBUF_UNLOCK(&so->so_rcv); 1815 goto release; 1816 } 1817 m = so->so_rcv.sb_mb; 1818 if (m != NULL) 1819 nextrecord = m->m_nextpkt; 1820 } 1821 } 1822 1823 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1824 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 1825 flags |= MSG_TRUNC; 1826 if ((flags & MSG_PEEK) == 0) 1827 (void) sbdroprecord_locked(&so->so_rcv); 1828 } 1829 if ((flags & MSG_PEEK) == 0) { 1830 if (m == NULL) { 1831 /* 1832 * First part is an inline SB_EMPTY_FIXUP(). Second 1833 * part makes sure sb_lastrecord is up-to-date if 1834 * there is still data in the socket buffer. 1835 */ 1836 so->so_rcv.sb_mb = nextrecord; 1837 if (so->so_rcv.sb_mb == NULL) { 1838 so->so_rcv.sb_mbtail = NULL; 1839 so->so_rcv.sb_lastrecord = NULL; 1840 } else if (nextrecord->m_nextpkt == NULL) 1841 so->so_rcv.sb_lastrecord = nextrecord; 1842 } 1843 SBLASTRECORDCHK(&so->so_rcv); 1844 SBLASTMBUFCHK(&so->so_rcv); 1845 /* 1846 * If soreceive() is being done from the socket callback, 1847 * then don't need to generate ACK to peer to update window, 1848 * since ACK will be generated on return to TCP. 1849 */ 1850 if (!(flags & MSG_SOCALLBCK) && 1851 (pr->pr_flags & PR_WANTRCVD)) { 1852 SOCKBUF_UNLOCK(&so->so_rcv); 1853 (*pr->pr_usrreqs->pru_rcvd)(so, flags); 1854 SOCKBUF_LOCK(&so->so_rcv); 1855 } 1856 } 1857 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1858 if (orig_resid == uio->uio_resid && orig_resid && 1859 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 1860 SOCKBUF_UNLOCK(&so->so_rcv); 1861 goto restart; 1862 } 1863 SOCKBUF_UNLOCK(&so->so_rcv); 1864 1865 if (flagsp != NULL) 1866 *flagsp |= flags; 1867release: 1868 sbunlock(&so->so_rcv); 1869 return (error); 1870} 1871 1872/* 1873 * Optimized version of soreceive() for stream (TCP) sockets. 1874 */ 1875#ifdef TCP_SORECEIVE_STREAM 1876int 1877soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 1878 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1879{ 1880 int len = 0, error = 0, flags, oresid; 1881 struct sockbuf *sb; 1882 struct mbuf *m, *n = NULL; 1883 1884 /* We only do stream sockets. */ 1885 if (so->so_type != SOCK_STREAM) 1886 return (EINVAL); 1887 if (psa != NULL) 1888 *psa = NULL; 1889 if (controlp != NULL) 1890 return (EINVAL); 1891 if (flagsp != NULL) 1892 flags = *flagsp &~ MSG_EOR; 1893 else 1894 flags = 0; 1895 if (flags & MSG_OOB) 1896 return (soreceive_rcvoob(so, uio, flags)); 1897 if (mp0 != NULL) 1898 *mp0 = NULL; 1899 1900 sb = &so->so_rcv; 1901 1902 /* Prevent other readers from entering the socket. */ 1903 error = sblock(sb, SBLOCKWAIT(flags)); 1904 if (error) 1905 goto out; 1906 SOCKBUF_LOCK(sb); 1907 1908 /* Easy one, no space to copyout anything. */ 1909 if (uio->uio_resid == 0) { 1910 error = EINVAL; 1911 goto out; 1912 } 1913 oresid = uio->uio_resid; 1914 1915 /* We will never ever get anything unless we are connected. */ 1916 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 1917 /* When disconnecting there may be still some data left. */ 1918 if (sb->sb_cc > 0) 1919 goto deliver; 1920 if (!(so->so_state & SS_ISDISCONNECTED)) 1921 error = ENOTCONN; 1922 goto out; 1923 } 1924 1925 /* Socket buffer is empty and we shall not block. */ 1926 if (sb->sb_cc == 0 && 1927 ((sb->sb_flags & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 1928 error = EAGAIN; 1929 goto out; 1930 } 1931 1932restart: 1933 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1934 1935 /* Abort if socket has reported problems. */ 1936 if (so->so_error) { 1937 if (sb->sb_cc > 0) 1938 goto deliver; 1939 if (oresid > uio->uio_resid) 1940 goto out; 1941 error = so->so_error; 1942 if (!(flags & MSG_PEEK)) 1943 so->so_error = 0; 1944 goto out; 1945 } 1946 1947 /* Door is closed. Deliver what is left, if any. */ 1948 if (sb->sb_state & SBS_CANTRCVMORE) { 1949 if (sb->sb_cc > 0) 1950 goto deliver; 1951 else 1952 goto out; 1953 } 1954 1955 /* Socket buffer got some data that we shall deliver now. */ 1956 if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) && 1957 ((sb->sb_flags & SS_NBIO) || 1958 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 1959 sb->sb_cc >= sb->sb_lowat || 1960 sb->sb_cc >= uio->uio_resid || 1961 sb->sb_cc >= sb->sb_hiwat) ) { 1962 goto deliver; 1963 } 1964 1965 /* On MSG_WAITALL we must wait until all data or error arrives. */ 1966 if ((flags & MSG_WAITALL) && 1967 (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_lowat)) 1968 goto deliver; 1969 1970 /* 1971 * Wait and block until (more) data comes in. 1972 * NB: Drops the sockbuf lock during wait. 1973 */ 1974 error = sbwait(sb); 1975 if (error) 1976 goto out; 1977 goto restart; 1978 1979deliver: 1980 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1981 KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__)); 1982 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 1983 1984 /* Statistics. */ 1985 if (uio->uio_td) 1986 uio->uio_td->td_ru.ru_msgrcv++; 1987 1988 /* Fill uio until full or current end of socket buffer is reached. */ 1989 len = min(uio->uio_resid, sb->sb_cc); 1990 if (mp0 != NULL) { 1991 /* Dequeue as many mbufs as possible. */ 1992 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 1993 for (*mp0 = m = sb->sb_mb; 1994 m != NULL && m->m_len <= len; 1995 m = m->m_next) { 1996 len -= m->m_len; 1997 uio->uio_resid -= m->m_len; 1998 sbfree(sb, m); 1999 n = m; 2000 } 2001 sb->sb_mb = m; 2002 if (sb->sb_mb == NULL) 2003 SB_EMPTY_FIXUP(sb); 2004 n->m_next = NULL; 2005 } 2006 /* Copy the remainder. */ 2007 if (len > 0) { 2008 KASSERT(sb->sb_mb != NULL, 2009 ("%s: len > 0 && sb->sb_mb empty", __func__)); 2010 2011 m = m_copym(sb->sb_mb, 0, len, M_DONTWAIT); 2012 if (m == NULL) 2013 len = 0; /* Don't flush data from sockbuf. */ 2014 else 2015 uio->uio_resid -= m->m_len; 2016 if (*mp0 != NULL) 2017 n->m_next = m; 2018 else 2019 *mp0 = m; 2020 if (*mp0 == NULL) { 2021 error = ENOBUFS; 2022 goto out; 2023 } 2024 } 2025 } else { 2026 /* NB: Must unlock socket buffer as uiomove may sleep. */ 2027 SOCKBUF_UNLOCK(sb); 2028 error = m_mbuftouio(uio, sb->sb_mb, len); 2029 SOCKBUF_LOCK(sb); 2030 if (error) 2031 goto out; 2032 } 2033 SBLASTRECORDCHK(sb); 2034 SBLASTMBUFCHK(sb); 2035 2036 /* 2037 * Remove the delivered data from the socket buffer unless we 2038 * were only peeking. 2039 */ 2040 if (!(flags & MSG_PEEK)) { 2041 if (len > 0) 2042 sbdrop_locked(sb, len); 2043 2044 /* Notify protocol that we drained some data. */ 2045 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 2046 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 2047 !(flags & MSG_SOCALLBCK))) { 2048 SOCKBUF_UNLOCK(sb); 2049 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags); 2050 SOCKBUF_LOCK(sb); 2051 } 2052 } 2053 2054 /* 2055 * For MSG_WAITALL we may have to loop again and wait for 2056 * more data to come in. 2057 */ 2058 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 2059 goto restart; 2060out: 2061 SOCKBUF_LOCK_ASSERT(sb); 2062 SBLASTRECORDCHK(sb); 2063 SBLASTMBUFCHK(sb); 2064 SOCKBUF_UNLOCK(sb); 2065 sbunlock(sb); 2066 return (error); 2067} 2068#endif /* TCP_SORECEIVE_STREAM */ 2069 2070/* 2071 * Optimized version of soreceive() for simple datagram cases from userspace. 2072 * Unlike in the stream case, we're able to drop a datagram if copyout() 2073 * fails, and because we handle datagrams atomically, we don't need to use a 2074 * sleep lock to prevent I/O interlacing. 2075 */ 2076int 2077soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2078 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2079{ 2080 struct mbuf *m, *m2; 2081 int flags, len, error; 2082 struct protosw *pr = so->so_proto; 2083 struct mbuf *nextrecord; 2084 2085 if (psa != NULL) 2086 *psa = NULL; 2087 if (controlp != NULL) 2088 *controlp = NULL; 2089 if (flagsp != NULL) 2090 flags = *flagsp &~ MSG_EOR; 2091 else 2092 flags = 0; 2093 2094 /* 2095 * For any complicated cases, fall back to the full 2096 * soreceive_generic(). 2097 */ 2098 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB)) 2099 return (soreceive_generic(so, psa, uio, mp0, controlp, 2100 flagsp)); 2101 2102 /* 2103 * Enforce restrictions on use. 2104 */ 2105 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 2106 ("soreceive_dgram: wantrcvd")); 2107 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 2108 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 2109 ("soreceive_dgram: SBS_RCVATMARK")); 2110 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 2111 ("soreceive_dgram: P_CONNREQUIRED")); 2112 2113 /* 2114 * Loop blocking while waiting for a datagram. 2115 */ 2116 SOCKBUF_LOCK(&so->so_rcv); 2117 while ((m = so->so_rcv.sb_mb) == NULL) { 2118 KASSERT(so->so_rcv.sb_cc == 0, 2119 ("soreceive_dgram: sb_mb NULL but sb_cc %u", 2120 so->so_rcv.sb_cc)); 2121 if (so->so_error) { 2122 error = so->so_error; 2123 so->so_error = 0; 2124 SOCKBUF_UNLOCK(&so->so_rcv); 2125 return (error); 2126 } 2127 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2128 uio->uio_resid == 0) { 2129 SOCKBUF_UNLOCK(&so->so_rcv); 2130 return (0); 2131 } 2132 if ((so->so_state & SS_NBIO) || 2133 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2134 SOCKBUF_UNLOCK(&so->so_rcv); 2135 return (EWOULDBLOCK); 2136 } 2137 SBLASTRECORDCHK(&so->so_rcv); 2138 SBLASTMBUFCHK(&so->so_rcv); 2139 error = sbwait(&so->so_rcv); 2140 if (error) { 2141 SOCKBUF_UNLOCK(&so->so_rcv); 2142 return (error); 2143 } 2144 } 2145 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2146 2147 if (uio->uio_td) 2148 uio->uio_td->td_ru.ru_msgrcv++; 2149 SBLASTRECORDCHK(&so->so_rcv); 2150 SBLASTMBUFCHK(&so->so_rcv); 2151 nextrecord = m->m_nextpkt; 2152 if (nextrecord == NULL) { 2153 KASSERT(so->so_rcv.sb_lastrecord == m, 2154 ("soreceive_dgram: lastrecord != m")); 2155 } 2156 2157 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 2158 ("soreceive_dgram: m_nextpkt != nextrecord")); 2159 2160 /* 2161 * Pull 'm' and its chain off the front of the packet queue. 2162 */ 2163 so->so_rcv.sb_mb = NULL; 2164 sockbuf_pushsync(&so->so_rcv, nextrecord); 2165 2166 /* 2167 * Walk 'm's chain and free that many bytes from the socket buffer. 2168 */ 2169 for (m2 = m; m2 != NULL; m2 = m2->m_next) 2170 sbfree(&so->so_rcv, m2); 2171 2172 /* 2173 * Do a few last checks before we let go of the lock. 2174 */ 2175 SBLASTRECORDCHK(&so->so_rcv); 2176 SBLASTMBUFCHK(&so->so_rcv); 2177 SOCKBUF_UNLOCK(&so->so_rcv); 2178 2179 if (pr->pr_flags & PR_ADDR) { 2180 KASSERT(m->m_type == MT_SONAME, 2181 ("m->m_type == %d", m->m_type)); 2182 if (psa != NULL) 2183 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2184 M_NOWAIT); 2185 m = m_free(m); 2186 } 2187 if (m == NULL) { 2188 /* XXXRW: Can this happen? */ 2189 return (0); 2190 } 2191 2192 /* 2193 * Packet to copyout() is now in 'm' and it is disconnected from the 2194 * queue. 2195 * 2196 * Process one or more MT_CONTROL mbufs present before any data mbufs 2197 * in the first mbuf chain on the socket buffer. We call into the 2198 * protocol to perform externalization (or freeing if controlp == 2199 * NULL). 2200 */ 2201 if (m->m_type == MT_CONTROL) { 2202 struct mbuf *cm = NULL, *cmn; 2203 struct mbuf **cme = &cm; 2204 2205 do { 2206 m2 = m->m_next; 2207 m->m_next = NULL; 2208 *cme = m; 2209 cme = &(*cme)->m_next; 2210 m = m2; 2211 } while (m != NULL && m->m_type == MT_CONTROL); 2212 while (cm != NULL) { 2213 cmn = cm->m_next; 2214 cm->m_next = NULL; 2215 if (pr->pr_domain->dom_externalize != NULL) { 2216 error = (*pr->pr_domain->dom_externalize) 2217 (cm, controlp); 2218 } else if (controlp != NULL) 2219 *controlp = cm; 2220 else 2221 m_freem(cm); 2222 if (controlp != NULL) { 2223 while (*controlp != NULL) 2224 controlp = &(*controlp)->m_next; 2225 } 2226 cm = cmn; 2227 } 2228 } 2229 KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data")); 2230 2231 while (m != NULL && uio->uio_resid > 0) { 2232 len = uio->uio_resid; 2233 if (len > m->m_len) 2234 len = m->m_len; 2235 error = uiomove(mtod(m, char *), (int)len, uio); 2236 if (error) { 2237 m_freem(m); 2238 return (error); 2239 } 2240 if (len == m->m_len) 2241 m = m_free(m); 2242 else { 2243 m->m_data += len; 2244 m->m_len -= len; 2245 } 2246 } 2247 if (m != NULL) 2248 flags |= MSG_TRUNC; 2249 m_freem(m); 2250 if (flagsp != NULL) 2251 *flagsp |= flags; 2252 return (0); 2253} 2254 2255int 2256soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 2257 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2258{ 2259 2260 return (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0, 2261 controlp, flagsp)); 2262} 2263 2264int 2265soshutdown(struct socket *so, int how) 2266{ 2267 struct protosw *pr = so->so_proto; 2268 int error; 2269 2270 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR)) 2271 return (EINVAL); 2272 if (pr->pr_usrreqs->pru_flush != NULL) { 2273 (*pr->pr_usrreqs->pru_flush)(so, how); 2274 } 2275 if (how != SHUT_WR) 2276 sorflush(so); 2277 if (how != SHUT_RD) { 2278 CURVNET_SET(so->so_vnet); 2279 error = (*pr->pr_usrreqs->pru_shutdown)(so); 2280 CURVNET_RESTORE(); 2281 return (error); 2282 } 2283 return (0); 2284} 2285 2286void 2287sorflush(struct socket *so) 2288{ 2289 struct sockbuf *sb = &so->so_rcv; 2290 struct protosw *pr = so->so_proto; 2291 struct sockbuf asb; 2292 2293 /* 2294 * In order to avoid calling dom_dispose with the socket buffer mutex 2295 * held, and in order to generally avoid holding the lock for a long 2296 * time, we make a copy of the socket buffer and clear the original 2297 * (except locks, state). The new socket buffer copy won't have 2298 * initialized locks so we can only call routines that won't use or 2299 * assert those locks. 2300 * 2301 * Dislodge threads currently blocked in receive and wait to acquire 2302 * a lock against other simultaneous readers before clearing the 2303 * socket buffer. Don't let our acquire be interrupted by a signal 2304 * despite any existing socket disposition on interruptable waiting. 2305 */ 2306 CURVNET_SET(so->so_vnet); 2307 socantrcvmore(so); 2308 (void) sblock(sb, SBL_WAIT | SBL_NOINTR); 2309 2310 /* 2311 * Invalidate/clear most of the sockbuf structure, but leave selinfo 2312 * and mutex data unchanged. 2313 */ 2314 SOCKBUF_LOCK(sb); 2315 bzero(&asb, offsetof(struct sockbuf, sb_startzero)); 2316 bcopy(&sb->sb_startzero, &asb.sb_startzero, 2317 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2318 bzero(&sb->sb_startzero, 2319 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); 2320 SOCKBUF_UNLOCK(sb); 2321 sbunlock(sb); 2322 2323 /* 2324 * Dispose of special rights and flush the socket buffer. Don't call 2325 * any unsafe routines (that rely on locks being initialized) on asb. 2326 */ 2327 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) 2328 (*pr->pr_domain->dom_dispose)(asb.sb_mb); 2329 sbrelease_internal(&asb, so); 2330 CURVNET_RESTORE(); 2331} 2332 2333/* 2334 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 2335 * additional variant to handle the case where the option value needs to be 2336 * some kind of integer, but not a specific size. In addition to their use 2337 * here, these functions are also called by the protocol-level pr_ctloutput() 2338 * routines. 2339 */ 2340int 2341sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 2342{ 2343 size_t valsize; 2344 2345 /* 2346 * If the user gives us more than we wanted, we ignore it, but if we 2347 * don't get the minimum length the caller wants, we return EINVAL. 2348 * On success, sopt->sopt_valsize is set to however much we actually 2349 * retrieved. 2350 */ 2351 if ((valsize = sopt->sopt_valsize) < minlen) 2352 return EINVAL; 2353 if (valsize > len) 2354 sopt->sopt_valsize = valsize = len; 2355 2356 if (sopt->sopt_td != NULL) 2357 return (copyin(sopt->sopt_val, buf, valsize)); 2358 2359 bcopy(sopt->sopt_val, buf, valsize); 2360 return (0); 2361} 2362 2363/* 2364 * Kernel version of setsockopt(2). 2365 * 2366 * XXX: optlen is size_t, not socklen_t 2367 */ 2368int 2369so_setsockopt(struct socket *so, int level, int optname, void *optval, 2370 size_t optlen) 2371{ 2372 struct sockopt sopt; 2373 2374 sopt.sopt_level = level; 2375 sopt.sopt_name = optname; 2376 sopt.sopt_dir = SOPT_SET; 2377 sopt.sopt_val = optval; 2378 sopt.sopt_valsize = optlen; 2379 sopt.sopt_td = NULL; 2380 return (sosetopt(so, &sopt)); 2381} 2382 2383int 2384sosetopt(struct socket *so, struct sockopt *sopt) 2385{ 2386 int error, optval; 2387 struct linger l; 2388 struct timeval tv; 2389 u_long val; 2390 uint32_t val32; 2391#ifdef MAC 2392 struct mac extmac; 2393#endif 2394 2395 error = 0; 2396 if (sopt->sopt_level != SOL_SOCKET) { 2397 if (so->so_proto && so->so_proto->pr_ctloutput) 2398 return ((*so->so_proto->pr_ctloutput) 2399 (so, sopt)); 2400 error = ENOPROTOOPT; 2401 } else { 2402 switch (sopt->sopt_name) { 2403#ifdef INET 2404 case SO_ACCEPTFILTER: 2405 error = do_setopt_accept_filter(so, sopt); 2406 if (error) 2407 goto bad; 2408 break; 2409#endif 2410 case SO_LINGER: 2411 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 2412 if (error) 2413 goto bad; 2414 2415 SOCK_LOCK(so); 2416 so->so_linger = l.l_linger; 2417 if (l.l_onoff) 2418 so->so_options |= SO_LINGER; 2419 else 2420 so->so_options &= ~SO_LINGER; 2421 SOCK_UNLOCK(so); 2422 break; 2423 2424 case SO_DEBUG: 2425 case SO_KEEPALIVE: 2426 case SO_DONTROUTE: 2427 case SO_USELOOPBACK: 2428 case SO_BROADCAST: 2429 case SO_REUSEADDR: 2430 case SO_REUSEPORT: 2431 case SO_OOBINLINE: 2432 case SO_TIMESTAMP: 2433 case SO_BINTIME: 2434 case SO_NOSIGPIPE: 2435 case SO_NO_DDP: 2436 case SO_NO_OFFLOAD: 2437 error = sooptcopyin(sopt, &optval, sizeof optval, 2438 sizeof optval); 2439 if (error) 2440 goto bad; 2441 SOCK_LOCK(so); 2442 if (optval) 2443 so->so_options |= sopt->sopt_name; 2444 else 2445 so->so_options &= ~sopt->sopt_name; 2446 SOCK_UNLOCK(so); 2447 break; 2448 2449 case SO_SETFIB: 2450 error = sooptcopyin(sopt, &optval, sizeof optval, 2451 sizeof optval); 2452 if (optval < 1 || optval > rt_numfibs) { 2453 error = EINVAL; 2454 goto bad; 2455 } 2456 if ((so->so_proto->pr_domain->dom_family == PF_INET) || 2457 (so->so_proto->pr_domain->dom_family == PF_ROUTE)) { 2458 so->so_fibnum = optval; 2459 /* Note: ignore error */ 2460 if (so->so_proto && so->so_proto->pr_ctloutput) 2461 (*so->so_proto->pr_ctloutput)(so, sopt); 2462 } else { 2463 so->so_fibnum = 0; 2464 } 2465 break; 2466 2467 case SO_USER_COOKIE: 2468 error = sooptcopyin(sopt, &val32, sizeof val32, 2469 sizeof val32); 2470 if (error) 2471 goto bad; 2472 so->so_user_cookie = val32; 2473 break; 2474 2475 case SO_SNDBUF: 2476 case SO_RCVBUF: 2477 case SO_SNDLOWAT: 2478 case SO_RCVLOWAT: 2479 error = sooptcopyin(sopt, &optval, sizeof optval, 2480 sizeof optval); 2481 if (error) 2482 goto bad; 2483 2484 /* 2485 * Values < 1 make no sense for any of these options, 2486 * so disallow them. 2487 */ 2488 if (optval < 1) { 2489 error = EINVAL; 2490 goto bad; 2491 } 2492 2493 switch (sopt->sopt_name) { 2494 case SO_SNDBUF: 2495 case SO_RCVBUF: 2496 if (sbreserve(sopt->sopt_name == SO_SNDBUF ? 2497 &so->so_snd : &so->so_rcv, (u_long)optval, 2498 so, curthread) == 0) { 2499 error = ENOBUFS; 2500 goto bad; 2501 } 2502 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd : 2503 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE; 2504 break; 2505 2506 /* 2507 * Make sure the low-water is never greater than the 2508 * high-water. 2509 */ 2510 case SO_SNDLOWAT: 2511 SOCKBUF_LOCK(&so->so_snd); 2512 so->so_snd.sb_lowat = 2513 (optval > so->so_snd.sb_hiwat) ? 2514 so->so_snd.sb_hiwat : optval; 2515 SOCKBUF_UNLOCK(&so->so_snd); 2516 break; 2517 case SO_RCVLOWAT: 2518 SOCKBUF_LOCK(&so->so_rcv); 2519 so->so_rcv.sb_lowat = 2520 (optval > so->so_rcv.sb_hiwat) ? 2521 so->so_rcv.sb_hiwat : optval; 2522 SOCKBUF_UNLOCK(&so->so_rcv); 2523 break; 2524 } 2525 break; 2526 2527 case SO_SNDTIMEO: 2528 case SO_RCVTIMEO: 2529#ifdef COMPAT_FREEBSD32 2530 if (SV_CURPROC_FLAG(SV_ILP32)) { 2531 struct timeval32 tv32; 2532 2533 error = sooptcopyin(sopt, &tv32, sizeof tv32, 2534 sizeof tv32); 2535 CP(tv32, tv, tv_sec); 2536 CP(tv32, tv, tv_usec); 2537 } else 2538#endif 2539 error = sooptcopyin(sopt, &tv, sizeof tv, 2540 sizeof tv); 2541 if (error) 2542 goto bad; 2543 2544 /* assert(hz > 0); */ 2545 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz || 2546 tv.tv_usec < 0 || tv.tv_usec >= 1000000) { 2547 error = EDOM; 2548 goto bad; 2549 } 2550 /* assert(tick > 0); */ 2551 /* assert(ULONG_MAX - INT_MAX >= 1000000); */ 2552 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick; 2553 if (val > INT_MAX) { 2554 error = EDOM; 2555 goto bad; 2556 } 2557 if (val == 0 && tv.tv_usec != 0) 2558 val = 1; 2559 2560 switch (sopt->sopt_name) { 2561 case SO_SNDTIMEO: 2562 so->so_snd.sb_timeo = val; 2563 break; 2564 case SO_RCVTIMEO: 2565 so->so_rcv.sb_timeo = val; 2566 break; 2567 } 2568 break; 2569 2570 case SO_LABEL: 2571#ifdef MAC 2572 error = sooptcopyin(sopt, &extmac, sizeof extmac, 2573 sizeof extmac); 2574 if (error) 2575 goto bad; 2576 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 2577 so, &extmac); 2578#else 2579 error = EOPNOTSUPP; 2580#endif 2581 break; 2582 2583 default: 2584 error = ENOPROTOOPT; 2585 break; 2586 } 2587 if (error == 0 && so->so_proto != NULL && 2588 so->so_proto->pr_ctloutput != NULL) { 2589 (void) ((*so->so_proto->pr_ctloutput) 2590 (so, sopt)); 2591 } 2592 } 2593bad: 2594 return (error); 2595} 2596 2597/* 2598 * Helper routine for getsockopt. 2599 */ 2600int 2601sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 2602{ 2603 int error; 2604 size_t valsize; 2605 2606 error = 0; 2607 2608 /* 2609 * Documented get behavior is that we always return a value, possibly 2610 * truncated to fit in the user's buffer. Traditional behavior is 2611 * that we always tell the user precisely how much we copied, rather 2612 * than something useful like the total amount we had available for 2613 * her. Note that this interface is not idempotent; the entire 2614 * answer must generated ahead of time. 2615 */ 2616 valsize = min(len, sopt->sopt_valsize); 2617 sopt->sopt_valsize = valsize; 2618 if (sopt->sopt_val != NULL) { 2619 if (sopt->sopt_td != NULL) 2620 error = copyout(buf, sopt->sopt_val, valsize); 2621 else 2622 bcopy(buf, sopt->sopt_val, valsize); 2623 } 2624 return (error); 2625} 2626 2627int 2628sogetopt(struct socket *so, struct sockopt *sopt) 2629{ 2630 int error, optval; 2631 struct linger l; 2632 struct timeval tv; 2633#ifdef MAC 2634 struct mac extmac; 2635#endif 2636 2637 error = 0; 2638 if (sopt->sopt_level != SOL_SOCKET) { 2639 if (so->so_proto && so->so_proto->pr_ctloutput) { 2640 return ((*so->so_proto->pr_ctloutput) 2641 (so, sopt)); 2642 } else 2643 return (ENOPROTOOPT); 2644 } else { 2645 switch (sopt->sopt_name) { 2646#ifdef INET 2647 case SO_ACCEPTFILTER: 2648 error = do_getopt_accept_filter(so, sopt); 2649 break; 2650#endif 2651 case SO_LINGER: 2652 SOCK_LOCK(so); 2653 l.l_onoff = so->so_options & SO_LINGER; 2654 l.l_linger = so->so_linger; 2655 SOCK_UNLOCK(so); 2656 error = sooptcopyout(sopt, &l, sizeof l); 2657 break; 2658 2659 case SO_USELOOPBACK: 2660 case SO_DONTROUTE: 2661 case SO_DEBUG: 2662 case SO_KEEPALIVE: 2663 case SO_REUSEADDR: 2664 case SO_REUSEPORT: 2665 case SO_BROADCAST: 2666 case SO_OOBINLINE: 2667 case SO_ACCEPTCONN: 2668 case SO_TIMESTAMP: 2669 case SO_BINTIME: 2670 case SO_NOSIGPIPE: 2671 optval = so->so_options & sopt->sopt_name; 2672integer: 2673 error = sooptcopyout(sopt, &optval, sizeof optval); 2674 break; 2675 2676 case SO_TYPE: 2677 optval = so->so_type; 2678 goto integer; 2679 2680 case SO_ERROR: 2681 SOCK_LOCK(so); 2682 optval = so->so_error; 2683 so->so_error = 0; 2684 SOCK_UNLOCK(so); 2685 goto integer; 2686 2687 case SO_SNDBUF: 2688 optval = so->so_snd.sb_hiwat; 2689 goto integer; 2690 2691 case SO_RCVBUF: 2692 optval = so->so_rcv.sb_hiwat; 2693 goto integer; 2694 2695 case SO_SNDLOWAT: 2696 optval = so->so_snd.sb_lowat; 2697 goto integer; 2698 2699 case SO_RCVLOWAT: 2700 optval = so->so_rcv.sb_lowat; 2701 goto integer; 2702 2703 case SO_SNDTIMEO: 2704 case SO_RCVTIMEO: 2705 optval = (sopt->sopt_name == SO_SNDTIMEO ? 2706 so->so_snd.sb_timeo : so->so_rcv.sb_timeo); 2707 2708 tv.tv_sec = optval / hz; 2709 tv.tv_usec = (optval % hz) * tick; 2710#ifdef COMPAT_FREEBSD32 2711 if (SV_CURPROC_FLAG(SV_ILP32)) { 2712 struct timeval32 tv32; 2713 2714 CP(tv, tv32, tv_sec); 2715 CP(tv, tv32, tv_usec); 2716 error = sooptcopyout(sopt, &tv32, sizeof tv32); 2717 } else 2718#endif 2719 error = sooptcopyout(sopt, &tv, sizeof tv); 2720 break; 2721 2722 case SO_LABEL: 2723#ifdef MAC 2724 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 2725 sizeof(extmac)); 2726 if (error) 2727 return (error); 2728 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 2729 so, &extmac); 2730 if (error) 2731 return (error); 2732 error = sooptcopyout(sopt, &extmac, sizeof extmac); 2733#else 2734 error = EOPNOTSUPP; 2735#endif 2736 break; 2737 2738 case SO_PEERLABEL: 2739#ifdef MAC 2740 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 2741 sizeof(extmac)); 2742 if (error) 2743 return (error); 2744 error = mac_getsockopt_peerlabel( 2745 sopt->sopt_td->td_ucred, so, &extmac); 2746 if (error) 2747 return (error); 2748 error = sooptcopyout(sopt, &extmac, sizeof extmac); 2749#else 2750 error = EOPNOTSUPP; 2751#endif 2752 break; 2753 2754 case SO_LISTENQLIMIT: 2755 optval = so->so_qlimit; 2756 goto integer; 2757 2758 case SO_LISTENQLEN: 2759 optval = so->so_qlen; 2760 goto integer; 2761 2762 case SO_LISTENINCQLEN: 2763 optval = so->so_incqlen; 2764 goto integer; 2765 2766 default: 2767 error = ENOPROTOOPT; 2768 break; 2769 } 2770 return (error); 2771 } 2772} 2773 2774/* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */ 2775int 2776soopt_getm(struct sockopt *sopt, struct mbuf **mp) 2777{ 2778 struct mbuf *m, *m_prev; 2779 int sopt_size = sopt->sopt_valsize; 2780 2781 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA); 2782 if (m == NULL) 2783 return ENOBUFS; 2784 if (sopt_size > MLEN) { 2785 MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT); 2786 if ((m->m_flags & M_EXT) == 0) { 2787 m_free(m); 2788 return ENOBUFS; 2789 } 2790 m->m_len = min(MCLBYTES, sopt_size); 2791 } else { 2792 m->m_len = min(MLEN, sopt_size); 2793 } 2794 sopt_size -= m->m_len; 2795 *mp = m; 2796 m_prev = m; 2797 2798 while (sopt_size) { 2799 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA); 2800 if (m == NULL) { 2801 m_freem(*mp); 2802 return ENOBUFS; 2803 } 2804 if (sopt_size > MLEN) { 2805 MCLGET(m, sopt->sopt_td != NULL ? M_WAIT : 2806 M_DONTWAIT); 2807 if ((m->m_flags & M_EXT) == 0) { 2808 m_freem(m); 2809 m_freem(*mp); 2810 return ENOBUFS; 2811 } 2812 m->m_len = min(MCLBYTES, sopt_size); 2813 } else { 2814 m->m_len = min(MLEN, sopt_size); 2815 } 2816 sopt_size -= m->m_len; 2817 m_prev->m_next = m; 2818 m_prev = m; 2819 } 2820 return (0); 2821} 2822 2823/* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */ 2824int 2825soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 2826{ 2827 struct mbuf *m0 = m; 2828 2829 if (sopt->sopt_val == NULL) 2830 return (0); 2831 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 2832 if (sopt->sopt_td != NULL) { 2833 int error; 2834 2835 error = copyin(sopt->sopt_val, mtod(m, char *), 2836 m->m_len); 2837 if (error != 0) { 2838 m_freem(m0); 2839 return(error); 2840 } 2841 } else 2842 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 2843 sopt->sopt_valsize -= m->m_len; 2844 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 2845 m = m->m_next; 2846 } 2847 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 2848 panic("ip6_sooptmcopyin"); 2849 return (0); 2850} 2851 2852/* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */ 2853int 2854soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 2855{ 2856 struct mbuf *m0 = m; 2857 size_t valsize = 0; 2858 2859 if (sopt->sopt_val == NULL) 2860 return (0); 2861 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 2862 if (sopt->sopt_td != NULL) { 2863 int error; 2864 2865 error = copyout(mtod(m, char *), sopt->sopt_val, 2866 m->m_len); 2867 if (error != 0) { 2868 m_freem(m0); 2869 return(error); 2870 } 2871 } else 2872 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 2873 sopt->sopt_valsize -= m->m_len; 2874 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 2875 valsize += m->m_len; 2876 m = m->m_next; 2877 } 2878 if (m != NULL) { 2879 /* enough soopt buffer should be given from user-land */ 2880 m_freem(m0); 2881 return(EINVAL); 2882 } 2883 sopt->sopt_valsize = valsize; 2884 return (0); 2885} 2886 2887/* 2888 * sohasoutofband(): protocol notifies socket layer of the arrival of new 2889 * out-of-band data, which will then notify socket consumers. 2890 */ 2891void 2892sohasoutofband(struct socket *so) 2893{ 2894 2895 if (so->so_sigio != NULL) 2896 pgsigio(&so->so_sigio, SIGURG, 0); 2897 selwakeuppri(&so->so_rcv.sb_sel, PSOCK); 2898} 2899 2900int 2901sopoll(struct socket *so, int events, struct ucred *active_cred, 2902 struct thread *td) 2903{ 2904 2905 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred, 2906 td)); 2907} 2908 2909int 2910sopoll_generic(struct socket *so, int events, struct ucred *active_cred, 2911 struct thread *td) 2912{ 2913 int revents = 0; 2914 2915 SOCKBUF_LOCK(&so->so_snd); 2916 SOCKBUF_LOCK(&so->so_rcv); 2917 if (events & (POLLIN | POLLRDNORM)) 2918 if (soreadabledata(so)) 2919 revents |= events & (POLLIN | POLLRDNORM); 2920 2921 if (events & (POLLOUT | POLLWRNORM)) 2922 if (sowriteable(so)) 2923 revents |= events & (POLLOUT | POLLWRNORM); 2924 2925 if (events & (POLLPRI | POLLRDBAND)) 2926 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK)) 2927 revents |= events & (POLLPRI | POLLRDBAND); 2928 2929 if ((events & POLLINIGNEOF) == 0) { 2930 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2931 revents |= events & (POLLIN | POLLRDNORM); 2932 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 2933 revents |= POLLHUP; 2934 } 2935 } 2936 2937 if (revents == 0) { 2938 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) { 2939 selrecord(td, &so->so_rcv.sb_sel); 2940 so->so_rcv.sb_flags |= SB_SEL; 2941 } 2942 2943 if (events & (POLLOUT | POLLWRNORM)) { 2944 selrecord(td, &so->so_snd.sb_sel); 2945 so->so_snd.sb_flags |= SB_SEL; 2946 } 2947 } 2948 2949 SOCKBUF_UNLOCK(&so->so_rcv); 2950 SOCKBUF_UNLOCK(&so->so_snd); 2951 return (revents); 2952} 2953 2954int 2955soo_kqfilter(struct file *fp, struct knote *kn) 2956{ 2957 struct socket *so = kn->kn_fp->f_data; 2958 struct sockbuf *sb; 2959 2960 switch (kn->kn_filter) { 2961 case EVFILT_READ: 2962 if (so->so_options & SO_ACCEPTCONN) 2963 kn->kn_fop = &solisten_filtops; 2964 else 2965 kn->kn_fop = &soread_filtops; 2966 sb = &so->so_rcv; 2967 break; 2968 case EVFILT_WRITE: 2969 kn->kn_fop = &sowrite_filtops; 2970 sb = &so->so_snd; 2971 break; 2972 default: 2973 return (EINVAL); 2974 } 2975 2976 SOCKBUF_LOCK(sb); 2977 knlist_add(&sb->sb_sel.si_note, kn, 1); 2978 sb->sb_flags |= SB_KNOTE; 2979 SOCKBUF_UNLOCK(sb); 2980 return (0); 2981} 2982 2983/* 2984 * Some routines that return EOPNOTSUPP for entry points that are not 2985 * supported by a protocol. Fill in as needed. 2986 */ 2987int 2988pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 2989{ 2990 2991 return EOPNOTSUPP; 2992} 2993 2994int 2995pru_attach_notsupp(struct socket *so, int proto, struct thread *td) 2996{ 2997 2998 return EOPNOTSUPP; 2999} 3000 3001int 3002pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3003{ 3004 3005 return EOPNOTSUPP; 3006} 3007 3008int 3009pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 3010{ 3011 3012 return EOPNOTSUPP; 3013} 3014 3015int 3016pru_connect2_notsupp(struct socket *so1, struct socket *so2) 3017{ 3018 3019 return EOPNOTSUPP; 3020} 3021 3022int 3023pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 3024 struct ifnet *ifp, struct thread *td) 3025{ 3026 3027 return EOPNOTSUPP; 3028} 3029 3030int 3031pru_disconnect_notsupp(struct socket *so) 3032{ 3033 3034 return EOPNOTSUPP; 3035} 3036 3037int 3038pru_listen_notsupp(struct socket *so, int backlog, struct thread *td) 3039{ 3040 3041 return EOPNOTSUPP; 3042} 3043 3044int 3045pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam) 3046{ 3047 3048 return EOPNOTSUPP; 3049} 3050 3051int 3052pru_rcvd_notsupp(struct socket *so, int flags) 3053{ 3054 3055 return EOPNOTSUPP; 3056} 3057 3058int 3059pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 3060{ 3061 3062 return EOPNOTSUPP; 3063} 3064 3065int 3066pru_send_notsupp(struct socket *so, int flags, struct mbuf *m, 3067 struct sockaddr *addr, struct mbuf *control, struct thread *td) 3068{ 3069 3070 return EOPNOTSUPP; 3071} 3072 3073/* 3074 * This isn't really a ``null'' operation, but it's the default one and 3075 * doesn't do anything destructive. 3076 */ 3077int 3078pru_sense_null(struct socket *so, struct stat *sb) 3079{ 3080 3081 sb->st_blksize = so->so_snd.sb_hiwat; 3082 return 0; 3083} 3084 3085int 3086pru_shutdown_notsupp(struct socket *so) 3087{ 3088 3089 return EOPNOTSUPP; 3090} 3091 3092int 3093pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam) 3094{ 3095 3096 return EOPNOTSUPP; 3097} 3098 3099int 3100pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 3101 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 3102{ 3103 3104 return EOPNOTSUPP; 3105} 3106 3107int 3108pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 3109 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 3110{ 3111 3112 return EOPNOTSUPP; 3113} 3114 3115int 3116pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred, 3117 struct thread *td) 3118{ 3119 3120 return EOPNOTSUPP; 3121} 3122 3123static void 3124filt_sordetach(struct knote *kn) 3125{ 3126 struct socket *so = kn->kn_fp->f_data; 3127 3128 SOCKBUF_LOCK(&so->so_rcv); 3129 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1); 3130 if (knlist_empty(&so->so_rcv.sb_sel.si_note)) 3131 so->so_rcv.sb_flags &= ~SB_KNOTE; 3132 SOCKBUF_UNLOCK(&so->so_rcv); 3133} 3134 3135/*ARGSUSED*/ 3136static int 3137filt_soread(struct knote *kn, long hint) 3138{ 3139 struct socket *so; 3140 3141 so = kn->kn_fp->f_data; 3142 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3143 3144 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl; 3145 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3146 kn->kn_flags |= EV_EOF; 3147 kn->kn_fflags = so->so_error; 3148 return (1); 3149 } else if (so->so_error) /* temporary udp error */ 3150 return (1); 3151 else if (kn->kn_sfflags & NOTE_LOWAT) 3152 return (kn->kn_data >= kn->kn_sdata); 3153 else 3154 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat); 3155} 3156 3157static void 3158filt_sowdetach(struct knote *kn) 3159{ 3160 struct socket *so = kn->kn_fp->f_data; 3161 3162 SOCKBUF_LOCK(&so->so_snd); 3163 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1); 3164 if (knlist_empty(&so->so_snd.sb_sel.si_note)) 3165 so->so_snd.sb_flags &= ~SB_KNOTE; 3166 SOCKBUF_UNLOCK(&so->so_snd); 3167} 3168 3169/*ARGSUSED*/ 3170static int 3171filt_sowrite(struct knote *kn, long hint) 3172{ 3173 struct socket *so; 3174 3175 so = kn->kn_fp->f_data; 3176 SOCKBUF_LOCK_ASSERT(&so->so_snd); 3177 kn->kn_data = sbspace(&so->so_snd); 3178 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 3179 kn->kn_flags |= EV_EOF; 3180 kn->kn_fflags = so->so_error; 3181 return (1); 3182 } else if (so->so_error) /* temporary udp error */ 3183 return (1); 3184 else if (((so->so_state & SS_ISCONNECTED) == 0) && 3185 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 3186 return (0); 3187 else if (kn->kn_sfflags & NOTE_LOWAT) 3188 return (kn->kn_data >= kn->kn_sdata); 3189 else 3190 return (kn->kn_data >= so->so_snd.sb_lowat); 3191} 3192 3193/*ARGSUSED*/ 3194static int 3195filt_solisten(struct knote *kn, long hint) 3196{ 3197 struct socket *so = kn->kn_fp->f_data; 3198 3199 kn->kn_data = so->so_qlen; 3200 return (! TAILQ_EMPTY(&so->so_comp)); 3201} 3202 3203int 3204socheckuid(struct socket *so, uid_t uid) 3205{ 3206 3207 if (so == NULL) 3208 return (EPERM); 3209 if (so->so_cred->cr_uid != uid) 3210 return (EPERM); 3211 return (0); 3212} 3213 3214static int 3215sysctl_somaxconn(SYSCTL_HANDLER_ARGS) 3216{ 3217 int error; 3218 int val; 3219 3220 val = somaxconn; 3221 error = sysctl_handle_int(oidp, &val, 0, req); 3222 if (error || !req->newptr ) 3223 return (error); 3224 3225 if (val < 1 || val > USHRT_MAX) 3226 return (EINVAL); 3227 3228 somaxconn = val; 3229 return (0); 3230} 3231 3232/* 3233 * These functions are used by protocols to notify the socket layer (and its 3234 * consumers) of state changes in the sockets driven by protocol-side events. 3235 */ 3236 3237/* 3238 * Procedures to manipulate state flags of socket and do appropriate wakeups. 3239 * 3240 * Normal sequence from the active (originating) side is that 3241 * soisconnecting() is called during processing of connect() call, resulting 3242 * in an eventual call to soisconnected() if/when the connection is 3243 * established. When the connection is torn down soisdisconnecting() is 3244 * called during processing of disconnect() call, and soisdisconnected() is 3245 * called when the connection to the peer is totally severed. The semantics 3246 * of these routines are such that connectionless protocols can call 3247 * soisconnected() and soisdisconnected() only, bypassing the in-progress 3248 * calls when setting up a ``connection'' takes no time. 3249 * 3250 * From the passive side, a socket is created with two queues of sockets: 3251 * so_incomp for connections in progress and so_comp for connections already 3252 * made and awaiting user acceptance. As a protocol is preparing incoming 3253 * connections, it creates a socket structure queued on so_incomp by calling 3254 * sonewconn(). When the connection is established, soisconnected() is 3255 * called, and transfers the socket structure to so_comp, making it available 3256 * to accept(). 3257 * 3258 * If a socket is closed with sockets on either so_incomp or so_comp, these 3259 * sockets are dropped. 3260 * 3261 * If higher-level protocols are implemented in the kernel, the wakeups done 3262 * here will sometimes cause software-interrupt process scheduling. 3263 */ 3264void 3265soisconnecting(struct socket *so) 3266{ 3267 3268 SOCK_LOCK(so); 3269 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 3270 so->so_state |= SS_ISCONNECTING; 3271 SOCK_UNLOCK(so); 3272} 3273 3274void 3275soisconnected(struct socket *so) 3276{ 3277 struct socket *head; 3278 int ret; 3279 3280restart: 3281 ACCEPT_LOCK(); 3282 SOCK_LOCK(so); 3283 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 3284 so->so_state |= SS_ISCONNECTED; 3285 head = so->so_head; 3286 if (head != NULL && (so->so_qstate & SQ_INCOMP)) { 3287 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 3288 SOCK_UNLOCK(so); 3289 TAILQ_REMOVE(&head->so_incomp, so, so_list); 3290 head->so_incqlen--; 3291 so->so_qstate &= ~SQ_INCOMP; 3292 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 3293 head->so_qlen++; 3294 so->so_qstate |= SQ_COMP; 3295 ACCEPT_UNLOCK(); 3296 sorwakeup(head); 3297 wakeup_one(&head->so_timeo); 3298 } else { 3299 ACCEPT_UNLOCK(); 3300 soupcall_set(so, SO_RCV, 3301 head->so_accf->so_accept_filter->accf_callback, 3302 head->so_accf->so_accept_filter_arg); 3303 so->so_options &= ~SO_ACCEPTFILTER; 3304 ret = head->so_accf->so_accept_filter->accf_callback(so, 3305 head->so_accf->so_accept_filter_arg, M_DONTWAIT); 3306 if (ret == SU_ISCONNECTED) 3307 soupcall_clear(so, SO_RCV); 3308 SOCK_UNLOCK(so); 3309 if (ret == SU_ISCONNECTED) 3310 goto restart; 3311 } 3312 return; 3313 } 3314 SOCK_UNLOCK(so); 3315 ACCEPT_UNLOCK(); 3316 wakeup(&so->so_timeo); 3317 sorwakeup(so); 3318 sowwakeup(so); 3319} 3320 3321void 3322soisdisconnecting(struct socket *so) 3323{ 3324 3325 /* 3326 * Note: This code assumes that SOCK_LOCK(so) and 3327 * SOCKBUF_LOCK(&so->so_rcv) are the same. 3328 */ 3329 SOCKBUF_LOCK(&so->so_rcv); 3330 so->so_state &= ~SS_ISCONNECTING; 3331 so->so_state |= SS_ISDISCONNECTING; 3332 so->so_rcv.sb_state |= SBS_CANTRCVMORE; 3333 sorwakeup_locked(so); 3334 SOCKBUF_LOCK(&so->so_snd); 3335 so->so_snd.sb_state |= SBS_CANTSENDMORE; 3336 sowwakeup_locked(so); 3337 wakeup(&so->so_timeo); 3338} 3339 3340void 3341soisdisconnected(struct socket *so) 3342{ 3343 3344 /* 3345 * Note: This code assumes that SOCK_LOCK(so) and 3346 * SOCKBUF_LOCK(&so->so_rcv) are the same. 3347 */ 3348 SOCKBUF_LOCK(&so->so_rcv); 3349 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 3350 so->so_state |= SS_ISDISCONNECTED; 3351 so->so_rcv.sb_state |= SBS_CANTRCVMORE; 3352 sorwakeup_locked(so); 3353 SOCKBUF_LOCK(&so->so_snd); 3354 so->so_snd.sb_state |= SBS_CANTSENDMORE; 3355 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc); 3356 sowwakeup_locked(so); 3357 wakeup(&so->so_timeo); 3358} 3359 3360/* 3361 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 3362 */ 3363struct sockaddr * 3364sodupsockaddr(const struct sockaddr *sa, int mflags) 3365{ 3366 struct sockaddr *sa2; 3367 3368 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 3369 if (sa2) 3370 bcopy(sa, sa2, sa->sa_len); 3371 return sa2; 3372} 3373 3374/* 3375 * Register per-socket buffer upcalls. 3376 */ 3377void 3378soupcall_set(struct socket *so, int which, 3379 int (*func)(struct socket *, void *, int), void *arg) 3380{ 3381 struct sockbuf *sb; 3382 3383 switch (which) { 3384 case SO_RCV: 3385 sb = &so->so_rcv; 3386 break; 3387 case SO_SND: 3388 sb = &so->so_snd; 3389 break; 3390 default: 3391 panic("soupcall_set: bad which"); 3392 } 3393 SOCKBUF_LOCK_ASSERT(sb); 3394#if 0 3395 /* XXX: accf_http actually wants to do this on purpose. */ 3396 KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall")); 3397#endif 3398 sb->sb_upcall = func; 3399 sb->sb_upcallarg = arg; 3400 sb->sb_flags |= SB_UPCALL; 3401} 3402 3403void 3404soupcall_clear(struct socket *so, int which) 3405{ 3406 struct sockbuf *sb; 3407 3408 switch (which) { 3409 case SO_RCV: 3410 sb = &so->so_rcv; 3411 break; 3412 case SO_SND: 3413 sb = &so->so_snd; 3414 break; 3415 default: 3416 panic("soupcall_clear: bad which"); 3417 } 3418 SOCKBUF_LOCK_ASSERT(sb); 3419 KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear")); 3420 sb->sb_upcall = NULL; 3421 sb->sb_upcallarg = NULL; 3422 sb->sb_flags &= ~SB_UPCALL; 3423} 3424 3425/* 3426 * Create an external-format (``xsocket'') structure using the information in 3427 * the kernel-format socket structure pointed to by so. This is done to 3428 * reduce the spew of irrelevant information over this interface, to isolate 3429 * user code from changes in the kernel structure, and potentially to provide 3430 * information-hiding if we decide that some of this information should be 3431 * hidden from users. 3432 */ 3433void 3434sotoxsocket(struct socket *so, struct xsocket *xso) 3435{ 3436 3437 xso->xso_len = sizeof *xso; 3438 xso->xso_so = so; 3439 xso->so_type = so->so_type; 3440 xso->so_options = so->so_options; 3441 xso->so_linger = so->so_linger; 3442 xso->so_state = so->so_state; 3443 xso->so_pcb = so->so_pcb; 3444 xso->xso_protocol = so->so_proto->pr_protocol; 3445 xso->xso_family = so->so_proto->pr_domain->dom_family; 3446 xso->so_qlen = so->so_qlen; 3447 xso->so_incqlen = so->so_incqlen; 3448 xso->so_qlimit = so->so_qlimit; 3449 xso->so_timeo = so->so_timeo; 3450 xso->so_error = so->so_error; 3451 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 3452 xso->so_oobmark = so->so_oobmark; 3453 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 3454 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 3455 xso->so_uid = so->so_cred->cr_uid; 3456} 3457 3458 3459/* 3460 * Socket accessor functions to provide external consumers with 3461 * a safe interface to socket state 3462 * 3463 */ 3464 3465void 3466so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), void *arg) 3467{ 3468 3469 TAILQ_FOREACH(so, &so->so_comp, so_list) 3470 func(so, arg); 3471} 3472 3473struct sockbuf * 3474so_sockbuf_rcv(struct socket *so) 3475{ 3476 3477 return (&so->so_rcv); 3478} 3479 3480struct sockbuf * 3481so_sockbuf_snd(struct socket *so) 3482{ 3483 3484 return (&so->so_snd); 3485} 3486 3487int 3488so_state_get(const struct socket *so) 3489{ 3490 3491 return (so->so_state); 3492} 3493 3494void 3495so_state_set(struct socket *so, int val) 3496{ 3497 3498 so->so_state = val; 3499} 3500 3501int 3502so_options_get(const struct socket *so) 3503{ 3504 3505 return (so->so_options); 3506} 3507 3508void 3509so_options_set(struct socket *so, int val) 3510{ 3511 3512 so->so_options = val; 3513} 3514 3515int 3516so_error_get(const struct socket *so) 3517{ 3518 3519 return (so->so_error); 3520} 3521 3522void 3523so_error_set(struct socket *so, int val) 3524{ 3525 3526 so->so_error = val; 3527} 3528 3529int 3530so_linger_get(const struct socket *so) 3531{ 3532 3533 return (so->so_linger); 3534} 3535 3536void 3537so_linger_set(struct socket *so, int val) 3538{ 3539 3540 so->so_linger = val; 3541} 3542 3543struct protosw * 3544so_protosw_get(const struct socket *so) 3545{ 3546 3547 return (so->so_proto); 3548} 3549 3550void 3551so_protosw_set(struct socket *so, struct protosw *val) 3552{ 3553 3554 so->so_proto = val; 3555} 3556 3557void 3558so_sorwakeup(struct socket *so) 3559{ 3560 3561 sorwakeup(so); 3562} 3563 3564void 3565so_sowwakeup(struct socket *so) 3566{ 3567 3568 sowwakeup(so); 3569} 3570 3571void 3572so_sorwakeup_locked(struct socket *so) 3573{ 3574 3575 sorwakeup_locked(so); 3576} 3577 3578void 3579so_sowwakeup_locked(struct socket *so) 3580{ 3581 3582 sowwakeup_locked(so); 3583} 3584 3585void 3586so_lock(struct socket *so) 3587{ 3588 SOCK_LOCK(so); 3589} 3590 3591void 3592so_unlock(struct socket *so) 3593{ 3594 SOCK_UNLOCK(so); 3595} 3596