PageRenderTime 121ms CodeModel.GetById 4ms app.highlight 101ms RepoModel.GetById 1ms app.codeStats 1ms

/sys/kern/uipc_socket.c

https://github.com/brauceunix/libuinet
C | 3902 lines | 2830 code | 357 blank | 715 comment | 768 complexity | 567e0eb2b3bb65b01dd0a33ff9a26120 MD5 | raw file

Large files files are truncated, but you can click here to view the full file

   1/*-
   2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
   3 *	The Regents of the University of California.
   4 * Copyright (c) 2004 The FreeBSD Foundation
   5 * Copyright (c) 2004-2008 Robert N. M. Watson
   6 * All rights reserved.
   7 *
   8 * Redistribution and use in source and binary forms, with or without
   9 * modification, are permitted provided that the following conditions
  10 * are met:
  11 * 1. Redistributions of source code must retain the above copyright
  12 *    notice, this list of conditions and the following disclaimer.
  13 * 2. Redistributions in binary form must reproduce the above copyright
  14 *    notice, this list of conditions and the following disclaimer in the
  15 *    documentation and/or other materials provided with the distribution.
  16 * 4. Neither the name of the University nor the names of its contributors
  17 *    may be used to endorse or promote products derived from this software
  18 *    without specific prior written permission.
  19 *
  20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30 * SUCH DAMAGE.
  31 *
  32 *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
  33 */
  34
  35/*
  36 * Comments on the socket life cycle:
  37 *
  38 * soalloc() sets of socket layer state for a socket, called only by
  39 * socreate() and sonewconn().  Socket layer private.
  40 *
  41 * sodealloc() tears down socket layer state for a socket, called only by
  42 * sofree() and sonewconn().  Socket layer private.
  43 *
  44 * pru_attach() associates protocol layer state with an allocated socket;
  45 * called only once, may fail, aborting socket allocation.  This is called
  46 * from socreate() and sonewconn().  Socket layer private.
  47 *
  48 * pru_detach() disassociates protocol layer state from an attached socket,
  49 * and will be called exactly once for sockets in which pru_attach() has
  50 * been successfully called.  If pru_attach() returned an error,
  51 * pru_detach() will not be called.  Socket layer private.
  52 *
  53 * pru_abort() and pru_close() notify the protocol layer that the last
  54 * consumer of a socket is starting to tear down the socket, and that the
  55 * protocol should terminate the connection.  Historically, pru_abort() also
  56 * detached protocol state from the socket state, but this is no longer the
  57 * case.
  58 *
  59 * socreate() creates a socket and attaches protocol state.  This is a public
  60 * interface that may be used by socket layer consumers to create new
  61 * sockets.
  62 *
  63 * sonewconn() creates a socket and attaches protocol state.  This is a
  64 * public interface  that may be used by protocols to create new sockets when
  65 * a new connection is received and will be available for accept() on a
  66 * listen socket.
  67 *
  68 * soclose() destroys a socket after possibly waiting for it to disconnect.
  69 * This is a public interface that socket consumers should use to close and
  70 * release a socket when done with it.
  71 *
  72 * soabort() destroys a socket without waiting for it to disconnect (used
  73 * only for incoming connections that are already partially or fully
  74 * connected).  This is used internally by the socket layer when clearing
  75 * listen socket queues (due to overflow or close on the listen socket), but
  76 * is also a public interface protocols may use to abort connections in
  77 * their incomplete listen queues should they no longer be required.  Sockets
  78 * placed in completed connection listen queues should not be aborted for
  79 * reasons described in the comment above the soclose() implementation.  This
  80 * is not a general purpose close routine, and except in the specific
  81 * circumstances described here, should not be used.
  82 *
  83 * sofree() will free a socket and its protocol state if all references on
  84 * the socket have been released, and is the public interface to attempt to
  85 * free a socket when a reference is removed.  This is a socket layer private
  86 * interface.
  87 *
  88 * NOTE: In addition to socreate() and soclose(), which provide a single
  89 * socket reference to the consumer to be managed as required, there are two
  90 * calls to explicitly manage socket references, soref(), and sorele().
  91 * Currently, these are generally required only when transitioning a socket
  92 * from a listen queue to a file descriptor, in order to prevent garbage
  93 * collection of the socket at an untimely moment.  For a number of reasons,
  94 * these interfaces are not preferred, and should be avoided.
  95 * 
  96 * NOTE: With regard to VNETs the general rule is that callers do not set
  97 * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
  98 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
  99 * and sorflush(), which are usually called from a pre-set VNET context.
 100 * sopoll() currently does not need a VNET context to be set.
 101 */
 102
 103#include <sys/cdefs.h>
 104__FBSDID("$FreeBSD: release/9.1.0/sys/kern/uipc_socket.c 233353 2012-03-23 11:26:54Z kib $");
 105
 106#include "opt_inet.h"
 107#include "opt_inet6.h"
 108#include "opt_zero.h"
 109#include "opt_compat.h"
 110#include "opt_passiveinet.h"
 111#include "opt_promiscinet.h"
 112
 113#include <sys/param.h>
 114#include <sys/systm.h>
 115#include <sys/fcntl.h>
 116#include <sys/limits.h>
 117#include <sys/lock.h>
 118#include <sys/mac.h>
 119#include <sys/malloc.h>
 120#include <sys/mbuf.h>
 121#include <sys/mutex.h>
 122#include <sys/domain.h>
 123#include <sys/file.h>			/* for struct knote */
 124#include <sys/kernel.h>
 125#include <sys/event.h>
 126#include <sys/eventhandler.h>
 127#include <sys/poll.h>
 128#include <sys/proc.h>
 129#include <sys/protosw.h>
 130#include <sys/socket.h>
 131#include <sys/socketvar.h>
 132#include <sys/resourcevar.h>
 133#include <net/route.h>
 134#include <sys/signalvar.h>
 135#include <sys/stat.h>
 136#include <sys/sx.h>
 137#include <sys/sysctl.h>
 138#include <sys/uio.h>
 139#include <sys/jail.h>
 140
 141#include <net/vnet.h>
 142
 143#ifdef MAC
 144#include <security/mac/mac_framework.h>
 145#endif /* MAC */
 146
 147#ifdef PROMISCUOUS_INET
 148#include <netinet/in_promisc.h>
 149#endif
 150
 151#include <vm/uma.h>
 152
 153#ifdef COMPAT_FREEBSD32
 154#include <sys/mount.h>
 155#include <sys/sysent.h>
 156#include <compat/freebsd32/freebsd32.h>
 157#endif
 158
 159static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
 160		    int flags);
 161
 162static void	filt_sordetach(struct knote *kn);
 163static int	filt_soread(struct knote *kn, long hint);
 164static void	filt_sowdetach(struct knote *kn);
 165static int	filt_sowrite(struct knote *kn, long hint);
 166static int	filt_solisten(struct knote *kn, long hint);
 167
 168static struct filterops solisten_filtops = {
 169	.f_isfd = 1,
 170	.f_detach = filt_sordetach,
 171	.f_event = filt_solisten,
 172};
 173static struct filterops soread_filtops = {
 174	.f_isfd = 1,
 175	.f_detach = filt_sordetach,
 176	.f_event = filt_soread,
 177};
 178static struct filterops sowrite_filtops = {
 179	.f_isfd = 1,
 180	.f_detach = filt_sowdetach,
 181	.f_event = filt_sowrite,
 182};
 183
 184uma_zone_t socket_zone;
 185so_gen_t	so_gencnt;	/* generation count for sockets */
 186
 187int	maxsockets;
 188
 189MALLOC_DEFINE(M_SONAME, "soname", "socket name");
 190MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
 191
 192#define	VNET_SO_ASSERT(so)						\
 193	VNET_ASSERT(curvnet != NULL,					\
 194	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
 195
 196static int somaxconn = SOMAXCONN;
 197static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS);
 198/* XXX: we dont have SYSCTL_USHORT */
 199SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
 200    0, sizeof(int), sysctl_somaxconn, "I", "Maximum pending socket connection "
 201    "queue size");
 202static int numopensockets;
 203SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
 204    &numopensockets, 0, "Number of open sockets");
 205#ifdef ZERO_COPY_SOCKETS
 206/* These aren't static because they're used in other files. */
 207int so_zero_copy_send = 1;
 208int so_zero_copy_receive = 1;
 209SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
 210    "Zero copy controls");
 211SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
 212    &so_zero_copy_receive, 0, "Enable zero copy receive");
 213SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
 214    &so_zero_copy_send, 0, "Enable zero copy send");
 215#endif /* ZERO_COPY_SOCKETS */
 216
 217/*
 218 * accept_mtx locks down per-socket fields relating to accept queues.  See
 219 * socketvar.h for an annotation of the protected fields of struct socket.
 220 */
 221struct mtx accept_mtx;
 222MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
 223
 224/*
 225 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
 226 * so_gencnt field.
 227 */
 228static struct mtx so_global_mtx;
 229MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
 230
 231/*
 232 * General IPC sysctl name space, used by sockets and a variety of other IPC
 233 * types.
 234 */
 235SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
 236
 237/*
 238 * Sysctl to get and set the maximum global sockets limit.  Notify protocols
 239 * of the change so that they can update their dependent limits as required.
 240 */
 241static int
 242sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
 243{
 244	int error, newmaxsockets;
 245
 246	newmaxsockets = maxsockets;
 247	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
 248	if (error == 0 && req->newptr) {
 249		if (newmaxsockets > maxsockets) {
 250			maxsockets = newmaxsockets;
 251			if (maxsockets > ((maxfiles / 4) * 3)) {
 252				maxfiles = (maxsockets * 5) / 4;
 253				maxfilesperproc = (maxfiles * 9) / 10;
 254			}
 255			EVENTHANDLER_INVOKE(maxsockets_change);
 256		} else
 257			error = EINVAL;
 258	}
 259	return (error);
 260}
 261
 262SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
 263    &maxsockets, 0, sysctl_maxsockets, "IU",
 264    "Maximum number of sockets avaliable");
 265
 266/*
 267 * Initialise maxsockets.  This SYSINIT must be run after
 268 * tunable_mbinit().
 269 */
 270static void
 271init_maxsockets(void *ignored)
 272{
 273
 274	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
 275	maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
 276}
 277SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
 278
 279/*
 280 * Socket operation routines.  These routines are called by the routines in
 281 * sys_socket.c or from a system process, and implement the semantics of
 282 * socket operations by switching out to the protocol specific routines.
 283 */
 284
 285/*
 286 * Get a socket structure from our zone, and initialize it.  Note that it
 287 * would probably be better to allocate socket and PCB at the same time, but
 288 * I'm not convinced that all the protocols can be easily modified to do
 289 * this.
 290 *
 291 * soalloc() returns a socket with a ref count of 0.
 292 */
 293static struct socket *
 294soalloc(struct vnet *vnet)
 295{
 296	struct socket *so;
 297
 298	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
 299	if (so == NULL)
 300		return (NULL);
 301#ifdef MAC
 302	if (mac_socket_init(so, M_NOWAIT) != 0) {
 303		uma_zfree(socket_zone, so);
 304		return (NULL);
 305	}
 306#endif
 307#ifdef PROMISCUOUS_INET
 308	if (in_promisc_socket_init(so, M_NOWAIT) != 0) {
 309#ifdef MAC
 310		mac_socket_destroy(so);
 311#endif
 312		uma_zfree(socket_zone, so);
 313		return (NULL);
 314	}
 315#endif /* PROMISCUOUS_INET */
 316	SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
 317	SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
 318	sx_init(&so->so_snd.sb_sx, "so_snd_sx");
 319	sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
 320	TAILQ_INIT(&so->so_aiojobq);
 321	mtx_lock(&so_global_mtx);
 322	so->so_gencnt = ++so_gencnt;
 323	++numopensockets;
 324#ifdef VIMAGE
 325	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
 326	    __func__, __LINE__, so));
 327	vnet->vnet_sockcnt++;
 328	so->so_vnet = vnet;
 329#endif
 330	mtx_unlock(&so_global_mtx);
 331	return (so);
 332}
 333
 334/*
 335 * Free the storage associated with a socket at the socket layer, tear down
 336 * locks, labels, etc.  All protocol state is assumed already to have been
 337 * torn down (and possibly never set up) by the caller.
 338 */
 339static void
 340sodealloc(struct socket *so)
 341{
 342
 343	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
 344	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
 345
 346	mtx_lock(&so_global_mtx);
 347	so->so_gencnt = ++so_gencnt;
 348	--numopensockets;	/* Could be below, but faster here. */
 349#ifdef VIMAGE
 350	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
 351	    __func__, __LINE__, so));
 352	so->so_vnet->vnet_sockcnt--;
 353#endif
 354	mtx_unlock(&so_global_mtx);
 355	if (so->so_rcv.sb_hiwat)
 356		(void)chgsbsize(so->so_cred->cr_uidinfo,
 357		    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
 358	if (so->so_snd.sb_hiwat)
 359		(void)chgsbsize(so->so_cred->cr_uidinfo,
 360		    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
 361#ifdef INET
 362	/* remove acccept filter if one is present. */
 363	if (so->so_accf != NULL)
 364		do_setopt_accept_filter(so, NULL);
 365#endif
 366#ifdef PROMISCUOUS_INET
 367	in_promisc_socket_destroy(so);
 368#endif
 369#ifdef MAC
 370	mac_socket_destroy(so);
 371#endif
 372	crfree(so->so_cred);
 373	sx_destroy(&so->so_snd.sb_sx);
 374	sx_destroy(&so->so_rcv.sb_sx);
 375	SOCKBUF_LOCK_DESTROY(&so->so_snd);
 376	SOCKBUF_LOCK_DESTROY(&so->so_rcv);
 377	uma_zfree(socket_zone, so);
 378}
 379
 380/*
 381 * socreate returns a socket with a ref count of 1.  The socket should be
 382 * closed with soclose().
 383 */
 384int
 385socreate(int dom, struct socket **aso, int type, int proto,
 386    struct ucred *cred, struct thread *td)
 387{
 388	struct protosw *prp;
 389	struct socket *so;
 390	int error;
 391
 392	if (proto)
 393		prp = pffindproto(dom, proto, type);
 394	else
 395		prp = pffindtype(dom, type);
 396
 397	if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
 398	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
 399		return (EPROTONOSUPPORT);
 400
 401	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
 402		return (EPROTONOSUPPORT);
 403
 404	if (prp->pr_type != type)
 405		return (EPROTOTYPE);
 406	so = soalloc(CRED_TO_VNET(cred));
 407	if (so == NULL)
 408		return (ENOBUFS);
 409
 410	TAILQ_INIT(&so->so_incomp);
 411	TAILQ_INIT(&so->so_comp);
 412	so->so_type = type;
 413	so->so_cred = crhold(cred);
 414	if ((prp->pr_domain->dom_family == PF_INET) ||
 415	    (prp->pr_domain->dom_family == PF_INET6) ||
 416	    (prp->pr_domain->dom_family == PF_ROUTE))
 417		so->so_fibnum = td->td_proc->p_fibnum;
 418	else
 419		so->so_fibnum = 0;
 420	so->so_proto = prp;
 421#ifdef MAC
 422	mac_socket_create(cred, so);
 423#endif
 424	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
 425	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
 426	so->so_count = 1;
 427	/*
 428	 * Auto-sizing of socket buffers is managed by the protocols and
 429	 * the appropriate flags must be set in the pru_attach function.
 430	 */
 431	CURVNET_SET(so->so_vnet);
 432	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
 433	CURVNET_RESTORE();
 434	if (error) {
 435		KASSERT(so->so_count == 1, ("socreate: so_count %d",
 436		    so->so_count));
 437		so->so_count = 0;
 438		sodealloc(so);
 439		return (error);
 440	}
 441	*aso = so;
 442	return (0);
 443}
 444
 445#ifdef REGRESSION
 446static int regression_sonewconn_earlytest = 1;
 447SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
 448    &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
 449#endif
 450
 451/*
 452 * When an attempt at a new connection is noted on a socket which accepts
 453 * connections, sonewconn is called.  If the connection is possible (subject
 454 * to space constraints, etc.) then we allocate a new structure, properly
 455 * linked into the data structure of the original socket, and return this.
 456 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
 457 *
 458 * Note: the ref count on the socket is 0 on return.
 459 */
 460struct socket *
 461sonewconn(struct socket *head, int connstatus)
 462{
 463	struct socket *so;
 464	int over;
 465
 466	ACCEPT_LOCK();
 467	over = (head->so_qlen > 3 * head->so_qlimit / 2);
 468	ACCEPT_UNLOCK();
 469#ifdef REGRESSION
 470	if (regression_sonewconn_earlytest && over)
 471#else
 472	if (over)
 473#endif
 474		return (NULL);
 475	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
 476	    __func__, __LINE__, head));
 477	so = soalloc(head->so_vnet);
 478	if (so == NULL)
 479		return (NULL);
 480	if ((head->so_options & SO_ACCEPTFILTER) != 0)
 481		connstatus = 0;
 482	so->so_head = head;
 483	so->so_type = head->so_type;
 484	so->so_options = head->so_options &~ SO_ACCEPTCONN;
 485	so->so_linger = head->so_linger;
 486	so->so_state = head->so_state | SS_NOFDREF;
 487	so->so_fibnum = head->so_fibnum;
 488	so->so_proto = head->so_proto;
 489	so->so_cred = crhold(head->so_cred);
 490#ifdef MAC
 491	mac_socket_newconn(head, so);
 492#endif
 493#ifdef PROMISCUOUS_INET
 494	in_promisc_socket_newconn(head, so);
 495#endif
 496	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
 497	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
 498	VNET_SO_ASSERT(head);
 499	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
 500	    (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
 501		sodealloc(so);
 502		return (NULL);
 503	}
 504	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
 505	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
 506	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
 507	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
 508	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
 509	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
 510	so->so_state |= connstatus;
 511	ACCEPT_LOCK();
 512	if (connstatus) {
 513		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
 514		so->so_qstate |= SQ_COMP;
 515		head->so_qlen++;
 516	} else {
 517		/*
 518		 * Keep removing sockets from the head until there's room for
 519		 * us to insert on the tail.  In pre-locking revisions, this
 520		 * was a simple if(), but as we could be racing with other
 521		 * threads and soabort() requires dropping locks, we must
 522		 * loop waiting for the condition to be true.
 523		 */
 524		while (head->so_incqlen > head->so_qlimit) {
 525			struct socket *sp;
 526			sp = TAILQ_FIRST(&head->so_incomp);
 527			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
 528			head->so_incqlen--;
 529			sp->so_qstate &= ~SQ_INCOMP;
 530			sp->so_head = NULL;
 531			ACCEPT_UNLOCK();
 532			soabort(sp);
 533			ACCEPT_LOCK();
 534		}
 535		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
 536		so->so_qstate |= SQ_INCOMP;
 537		head->so_incqlen++;
 538	}
 539	ACCEPT_UNLOCK();
 540	if (connstatus) {
 541		sorwakeup(head);
 542		wakeup_one(&head->so_timeo);
 543	}
 544	return (so);
 545}
 546
 547#ifdef PASSIVE_INET
 548/*
 549 * When a new connection is completed on a listening socket that is
 550 * configured for passive reassembly, sonewconn_passive_client is called to
 551 * create a socket representing the client side of the reassembled
 552 * connection.  We allocate a new structure, inherit configuration from the
 553 * listening socket, and return this.
 554 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTING.
 555 *
 556 * Note: the ref count on the socket is 1 on return.
 557 */
 558struct socket *
 559sonewconn_passive_client(struct socket *head, int connstatus)
 560{
 561	struct socket *so;
 562
 563	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
 564	    __func__, __LINE__, head));
 565	so = soalloc(head->so_vnet);
 566	if (so == NULL)
 567		return (NULL);
 568	so->so_head = NULL; /* just inheriting from head, not otherwise associating */
 569	so->so_type = head->so_type;
 570	so->so_options = head->so_options &~ SO_ACCEPTCONN;
 571	so->so_linger = head->so_linger;
 572	so->so_state = head->so_state | SS_NOFDREF;
 573	so->so_fibnum = head->so_fibnum;
 574	so->so_proto = head->so_proto;
 575	so->so_cred = crhold(head->so_cred);
 576#ifdef MAC
 577	mac_socket_newconn(head, so);
 578#endif
 579#ifdef PROMISCUOUS_INET
 580	in_promisc_socket_newconn(head, so);
 581#endif
 582	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
 583	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
 584	VNET_SO_ASSERT(head);
 585	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
 586	    (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
 587		sodealloc(so);
 588		return (NULL);
 589	}
 590	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
 591	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
 592	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
 593	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
 594	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
 595	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
 596	so->so_state |= connstatus;
 597
 598	return (so);
 599}
 600#endif /* PASSIVE_INET */
 601
 602int
 603sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
 604{
 605	int error;
 606
 607	CURVNET_SET(so->so_vnet);
 608	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
 609	CURVNET_RESTORE();
 610	return error;
 611}
 612
 613/*
 614 * solisten() transitions a socket from a non-listening state to a listening
 615 * state, but can also be used to update the listen queue depth on an
 616 * existing listen socket.  The protocol will call back into the sockets
 617 * layer using solisten_proto_check() and solisten_proto() to check and set
 618 * socket-layer listen state.  Call backs are used so that the protocol can
 619 * acquire both protocol and socket layer locks in whatever order is required
 620 * by the protocol.
 621 *
 622 * Protocol implementors are advised to hold the socket lock across the
 623 * socket-layer test and set to avoid races at the socket layer.
 624 */
 625int
 626solisten(struct socket *so, int backlog, struct thread *td)
 627{
 628	int error;
 629
 630	CURVNET_SET(so->so_vnet);
 631	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
 632	CURVNET_RESTORE();
 633	return error;
 634}
 635
 636int
 637solisten_proto_check(struct socket *so)
 638{
 639
 640	SOCK_LOCK_ASSERT(so);
 641
 642	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
 643	    SS_ISDISCONNECTING))
 644		return (EINVAL);
 645	return (0);
 646}
 647
 648void
 649solisten_proto(struct socket *so, int backlog)
 650{
 651
 652	SOCK_LOCK_ASSERT(so);
 653
 654	if (backlog < 0 || backlog > somaxconn)
 655		backlog = somaxconn;
 656	so->so_qlimit = backlog;
 657	so->so_options |= SO_ACCEPTCONN;
 658}
 659
 660/*
 661 * Evaluate the reference count and named references on a socket; if no
 662 * references remain, free it.  This should be called whenever a reference is
 663 * released, such as in sorele(), but also when named reference flags are
 664 * cleared in socket or protocol code.
 665 *
 666 * sofree() will free the socket if:
 667 *
 668 * - There are no outstanding file descriptor references or related consumers
 669 *   (so_count == 0).
 670 *
 671 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
 672 *
 673 * - The protocol does not have an outstanding strong reference on the socket
 674 *   (SS_PROTOREF).
 675 *
 676 * - The socket is not in a completed connection queue, so a process has been
 677 *   notified that it is present.  If it is removed, the user process may
 678 *   block in accept() despite select() saying the socket was ready.
 679 */
 680void
 681sofree(struct socket *so)
 682{
 683	struct protosw *pr = so->so_proto;
 684	struct socket *head;
 685
 686	ACCEPT_LOCK_ASSERT();
 687	SOCK_LOCK_ASSERT(so);
 688
 689	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
 690	    (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
 691		SOCK_UNLOCK(so);
 692		ACCEPT_UNLOCK();
 693		return;
 694	}
 695
 696	head = so->so_head;
 697	if (head != NULL) {
 698		KASSERT((so->so_qstate & SQ_COMP) != 0 ||
 699		    (so->so_qstate & SQ_INCOMP) != 0,
 700		    ("sofree: so_head != NULL, but neither SQ_COMP nor "
 701		    "SQ_INCOMP"));
 702		KASSERT((so->so_qstate & SQ_COMP) == 0 ||
 703		    (so->so_qstate & SQ_INCOMP) == 0,
 704		    ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
 705		TAILQ_REMOVE(&head->so_incomp, so, so_list);
 706		head->so_incqlen--;
 707		so->so_qstate &= ~SQ_INCOMP;
 708		so->so_head = NULL;
 709	}
 710	KASSERT((so->so_qstate & SQ_COMP) == 0 &&
 711	    (so->so_qstate & SQ_INCOMP) == 0,
 712	    ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
 713	    so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
 714	if (so->so_options & SO_ACCEPTCONN) {
 715		KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated"));
 716		KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_comp populated"));
 717	}
 718	SOCK_UNLOCK(so);
 719	ACCEPT_UNLOCK();
 720
 721	VNET_SO_ASSERT(so);
 722	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
 723		(*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
 724	if (pr->pr_usrreqs->pru_detach != NULL)
 725		(*pr->pr_usrreqs->pru_detach)(so);
 726
 727	/*
 728	 * From this point on, we assume that no other references to this
 729	 * socket exist anywhere else in the stack.  Therefore, no locks need
 730	 * to be acquired or held.
 731	 *
 732	 * We used to do a lot of socket buffer and socket locking here, as
 733	 * well as invoke sorflush() and perform wakeups.  The direct call to
 734	 * dom_dispose() and sbrelease_internal() are an inlining of what was
 735	 * necessary from sorflush().
 736	 *
 737	 * Notice that the socket buffer and kqueue state are torn down
 738	 * before calling pru_detach.  This means that protocols shold not
 739	 * assume they can perform socket wakeups, etc, in their detach code.
 740	 */
 741	sbdestroy(&so->so_snd, so);
 742	sbdestroy(&so->so_rcv, so);
 743	seldrain(&so->so_snd.sb_sel);
 744	seldrain(&so->so_rcv.sb_sel);
 745	knlist_destroy(&so->so_rcv.sb_sel.si_note);
 746	knlist_destroy(&so->so_snd.sb_sel.si_note);
 747	sodealloc(so);
 748}
 749
 750/*
 751 * Close a socket on last file table reference removal.  Initiate disconnect
 752 * if connected.  Free socket when disconnect complete.
 753 *
 754 * This function will sorele() the socket.  Note that soclose() may be called
 755 * prior to the ref count reaching zero.  The actual socket structure will
 756 * not be freed until the ref count reaches zero.
 757 */
 758int
 759soclose(struct socket *so)
 760{
 761	int error = 0;
 762
 763	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
 764
 765	CURVNET_SET(so->so_vnet);
 766	funsetown(&so->so_sigio);
 767	if (so->so_state & SS_ISCONNECTED) {
 768		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
 769			error = sodisconnect(so);
 770			if (error) {
 771				if (error == ENOTCONN)
 772					error = 0;
 773				goto drop;
 774			}
 775		}
 776		if (so->so_options & SO_LINGER) {
 777			if ((so->so_state & SS_ISDISCONNECTING) &&
 778			    (so->so_state & SS_NBIO))
 779				goto drop;
 780			while (so->so_state & SS_ISCONNECTED) {
 781				error = tsleep(&so->so_timeo,
 782				    PSOCK | PCATCH, "soclos", so->so_linger * hz);
 783				if (error)
 784					break;
 785			}
 786		}
 787	}
 788
 789drop:
 790	if (so->so_proto->pr_usrreqs->pru_close != NULL)
 791		(*so->so_proto->pr_usrreqs->pru_close)(so);
 792	if (so->so_options & SO_ACCEPTCONN) {
 793		struct socket *sp;
 794		ACCEPT_LOCK();
 795		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
 796			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
 797			so->so_incqlen--;
 798			sp->so_qstate &= ~SQ_INCOMP;
 799			sp->so_head = NULL;
 800			ACCEPT_UNLOCK();
 801			soabort(sp);
 802			ACCEPT_LOCK();
 803		}
 804		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
 805			TAILQ_REMOVE(&so->so_comp, sp, so_list);
 806			so->so_qlen--;
 807			sp->so_qstate &= ~SQ_COMP;
 808			sp->so_head = NULL;
 809			ACCEPT_UNLOCK();
 810			soabort(sp);
 811			ACCEPT_LOCK();
 812		}
 813		ACCEPT_UNLOCK();
 814	}
 815	ACCEPT_LOCK();
 816	SOCK_LOCK(so);
 817	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
 818	so->so_state |= SS_NOFDREF;
 819	sorele(so);
 820	CURVNET_RESTORE();
 821	return (error);
 822}
 823
 824/*
 825 * soabort() is used to abruptly tear down a connection, such as when a
 826 * resource limit is reached (listen queue depth exceeded), or if a listen
 827 * socket is closed while there are sockets waiting to be accepted.
 828 *
 829 * This interface is tricky, because it is called on an unreferenced socket,
 830 * and must be called only by a thread that has actually removed the socket
 831 * from the listen queue it was on, or races with other threads are risked.
 832 *
 833 * This interface will call into the protocol code, so must not be called
 834 * with any socket locks held.  Protocols do call it while holding their own
 835 * recursible protocol mutexes, but this is something that should be subject
 836 * to review in the future.
 837 */
 838void
 839soabort(struct socket *so)
 840{
 841
 842	/*
 843	 * In as much as is possible, assert that no references to this
 844	 * socket are held.  This is not quite the same as asserting that the
 845	 * current thread is responsible for arranging for no references, but
 846	 * is as close as we can get for now.
 847	 */
 848	KASSERT(so->so_count == 0, ("soabort: so_count"));
 849	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
 850	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
 851	KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
 852	KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
 853	VNET_SO_ASSERT(so);
 854
 855	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
 856		(*so->so_proto->pr_usrreqs->pru_abort)(so);
 857	ACCEPT_LOCK();
 858	SOCK_LOCK(so);
 859	sofree(so);
 860}
 861
 862int
 863soaccept(struct socket *so, struct sockaddr **nam)
 864{
 865	int error;
 866
 867	SOCK_LOCK(so);
 868	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
 869	so->so_state &= ~SS_NOFDREF;
 870	SOCK_UNLOCK(so);
 871
 872	CURVNET_SET(so->so_vnet);
 873	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
 874	CURVNET_RESTORE();
 875	return (error);
 876}
 877
 878int
 879soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
 880{
 881	int error;
 882
 883	if (so->so_options & SO_ACCEPTCONN)
 884		return (EOPNOTSUPP);
 885
 886	CURVNET_SET(so->so_vnet);
 887	/*
 888	 * If protocol is connection-based, can only connect once.
 889	 * Otherwise, if connected, try to disconnect first.  This allows
 890	 * user to disconnect by connecting to, e.g., a null address.
 891	 */
 892	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
 893	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
 894	    (error = sodisconnect(so)))) {
 895		error = EISCONN;
 896	} else {
 897		/*
 898		 * Prevent accumulated error from previous connection from
 899		 * biting us.
 900		 */
 901		so->so_error = 0;
 902		error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
 903	}
 904	CURVNET_RESTORE();
 905
 906	return (error);
 907}
 908
 909int
 910soconnect2(struct socket *so1, struct socket *so2)
 911{
 912	int error;
 913
 914	CURVNET_SET(so1->so_vnet);
 915	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
 916	CURVNET_RESTORE();
 917	return (error);
 918}
 919
 920int
 921sodisconnect(struct socket *so)
 922{
 923	int error;
 924
 925	if ((so->so_state & SS_ISCONNECTED) == 0)
 926		return (ENOTCONN);
 927	if (so->so_state & SS_ISDISCONNECTING)
 928		return (EALREADY);
 929	VNET_SO_ASSERT(so);
 930	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
 931	return (error);
 932}
 933
 934#ifdef ZERO_COPY_SOCKETS
 935struct so_zerocopy_stats{
 936	int size_ok;
 937	int align_ok;
 938	int found_ifp;
 939};
 940struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
 941#include <netinet/in.h>
 942#include <net/route.h>
 943#include <netinet/in_pcb.h>
 944#include <vm/vm.h>
 945#include <vm/vm_page.h>
 946#include <vm/vm_object.h>
 947
 948/*
 949 * sosend_copyin() is only used if zero copy sockets are enabled.  Otherwise
 950 * sosend_dgram() and sosend_generic() use m_uiotombuf().
 951 * 
 952 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
 953 * all of the data referenced by the uio.  If desired, it uses zero-copy.
 954 * *space will be updated to reflect data copied in.
 955 *
 956 * NB: If atomic I/O is requested, the caller must already have checked that
 957 * space can hold resid bytes.
 958 *
 959 * NB: In the event of an error, the caller may need to free the partial
 960 * chain pointed to by *mpp.  The contents of both *uio and *space may be
 961 * modified even in the case of an error.
 962 */
 963static int
 964sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
 965    int flags)
 966{
 967	struct mbuf *m, **mp, *top;
 968	long len;
 969	ssize_t resid;
 970	int error;
 971#ifdef ZERO_COPY_SOCKETS
 972	int cow_send;
 973#endif
 974
 975	*retmp = top = NULL;
 976	mp = &top;
 977	len = 0;
 978	resid = uio->uio_resid;
 979	error = 0;
 980	do {
 981#ifdef ZERO_COPY_SOCKETS
 982		cow_send = 0;
 983#endif /* ZERO_COPY_SOCKETS */
 984		if (resid >= MINCLSIZE) {
 985#ifdef ZERO_COPY_SOCKETS
 986			if (top == NULL) {
 987				m = m_gethdr(M_WAITOK, MT_DATA);
 988				m->m_pkthdr.len = 0;
 989				m->m_pkthdr.rcvif = NULL;
 990			} else
 991				m = m_get(M_WAITOK, MT_DATA);
 992			if (so_zero_copy_send &&
 993			    resid>=PAGE_SIZE &&
 994			    *space>=PAGE_SIZE &&
 995			    uio->uio_iov->iov_len>=PAGE_SIZE) {
 996				so_zerocp_stats.size_ok++;
 997				so_zerocp_stats.align_ok++;
 998				cow_send = socow_setup(m, uio);
 999				len = cow_send;
1000			}
1001			if (!cow_send) {
1002				m_clget(m, M_WAITOK);
1003				len = min(min(MCLBYTES, resid), *space);
1004			}
1005#else /* ZERO_COPY_SOCKETS */
1006			if (top == NULL) {
1007				m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
1008				m->m_pkthdr.len = 0;
1009				m->m_pkthdr.rcvif = NULL;
1010			} else
1011				m = m_getcl(M_WAIT, MT_DATA, 0);
1012			len = min(min(MCLBYTES, resid), *space);
1013#endif /* ZERO_COPY_SOCKETS */
1014		} else {
1015			if (top == NULL) {
1016				m = m_gethdr(M_WAIT, MT_DATA);
1017				m->m_pkthdr.len = 0;
1018				m->m_pkthdr.rcvif = NULL;
1019
1020				len = min(min(MHLEN, resid), *space);
1021				/*
1022				 * For datagram protocols, leave room
1023				 * for protocol headers in first mbuf.
1024				 */
1025				if (atomic && m && len < MHLEN)
1026					MH_ALIGN(m, len);
1027			} else {
1028				m = m_get(M_WAIT, MT_DATA);
1029				len = min(min(MLEN, resid), *space);
1030			}
1031		}
1032		if (m == NULL) {
1033			error = ENOBUFS;
1034			goto out;
1035		}
1036
1037		*space -= len;
1038#ifdef ZERO_COPY_SOCKETS
1039		if (cow_send)
1040			error = 0;
1041		else
1042#endif /* ZERO_COPY_SOCKETS */
1043		error = uiomove(mtod(m, void *), (int)len, uio);
1044		resid = uio->uio_resid;
1045		m->m_len = len;
1046		*mp = m;
1047		top->m_pkthdr.len += len;
1048		if (error)
1049			goto out;
1050		mp = &m->m_next;
1051		if (resid <= 0) {
1052			if (flags & MSG_EOR)
1053				top->m_flags |= M_EOR;
1054			break;
1055		}
1056	} while (*space > 0 && atomic);
1057out:
1058	*retmp = top;
1059	return (error);
1060}
1061#endif /*ZERO_COPY_SOCKETS*/
1062
1063#define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1064
1065int
1066sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1067    struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1068{
1069	long space;
1070	ssize_t resid;
1071	int clen = 0, error, dontroute;
1072#ifdef ZERO_COPY_SOCKETS
1073	int atomic = sosendallatonce(so) || top;
1074#endif
1075
1076	KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
1077	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1078	    ("sodgram_send: !PR_ATOMIC"));
1079
1080	if (uio != NULL)
1081		resid = uio->uio_resid;
1082	else
1083		resid = top->m_pkthdr.len;
1084	/*
1085	 * In theory resid should be unsigned.  However, space must be
1086	 * signed, as it might be less than 0 if we over-committed, and we
1087	 * must use a signed comparison of space and resid.  On the other
1088	 * hand, a negative resid causes us to loop sending 0-length
1089	 * segments to the protocol.
1090	 */
1091	if (resid < 0) {
1092		error = EINVAL;
1093		goto out;
1094	}
1095
1096	dontroute =
1097	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1098	if (td != NULL)
1099		td->td_ru.ru_msgsnd++;
1100	if (control != NULL)
1101		clen = control->m_len;
1102
1103	SOCKBUF_LOCK(&so->so_snd);
1104	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1105		SOCKBUF_UNLOCK(&so->so_snd);
1106		error = EPIPE;
1107		goto out;
1108	}
1109	if (so->so_error) {
1110		error = so->so_error;
1111		so->so_error = 0;
1112		SOCKBUF_UNLOCK(&so->so_snd);
1113		goto out;
1114	}
1115	if ((so->so_state & SS_ISCONNECTED) == 0) {
1116		/*
1117		 * `sendto' and `sendmsg' is allowed on a connection-based
1118		 * socket if it supports implied connect.  Return ENOTCONN if
1119		 * not connected and no address is supplied.
1120		 */
1121		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1122		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1123			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1124			    !(resid == 0 && clen != 0)) {
1125				SOCKBUF_UNLOCK(&so->so_snd);
1126				error = ENOTCONN;
1127				goto out;
1128			}
1129		} else if (addr == NULL) {
1130			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1131				error = ENOTCONN;
1132			else
1133				error = EDESTADDRREQ;
1134			SOCKBUF_UNLOCK(&so->so_snd);
1135			goto out;
1136		}
1137	}
1138
1139	/*
1140	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1141	 * problem and need fixing.
1142	 */
1143	space = sbspace(&so->so_snd);
1144	if (flags & MSG_OOB)
1145		space += 1024;
1146	space -= clen;
1147	SOCKBUF_UNLOCK(&so->so_snd);
1148	if (resid > space) {
1149		error = EMSGSIZE;
1150		goto out;
1151	}
1152	if (uio == NULL) {
1153		resid = 0;
1154		if (flags & MSG_EOR)
1155			top->m_flags |= M_EOR;
1156	} else {
1157#ifdef ZERO_COPY_SOCKETS
1158		error = sosend_copyin(uio, &top, atomic, &space, flags);
1159		if (error)
1160			goto out;
1161#else
1162		/*
1163		 * Copy the data from userland into a mbuf chain.
1164		 * If no data is to be copied in, a single empty mbuf
1165		 * is returned.
1166		 */
1167		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1168		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1169		if (top == NULL) {
1170			error = EFAULT;	/* only possible error */
1171			goto out;
1172		}
1173		space -= resid - uio->uio_resid;
1174#endif
1175		resid = uio->uio_resid;
1176	}
1177	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1178	/*
1179	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1180	 * than with.
1181	 */
1182	if (dontroute) {
1183		SOCK_LOCK(so);
1184		so->so_options |= SO_DONTROUTE;
1185		SOCK_UNLOCK(so);
1186	}
1187	/*
1188	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1189	 * of date.  We could have recieved a reset packet in an interrupt or
1190	 * maybe we slept while doing page faults in uiomove() etc.  We could
1191	 * probably recheck again inside the locking protection here, but
1192	 * there are probably other places that this also happens.  We must
1193	 * rethink this.
1194	 */
1195	VNET_SO_ASSERT(so);
1196	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1197	    (flags & MSG_OOB) ? PRUS_OOB :
1198	/*
1199	 * If the user set MSG_EOF, the protocol understands this flag and
1200	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1201	 */
1202	    ((flags & MSG_EOF) &&
1203	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1204	     (resid <= 0)) ?
1205		PRUS_EOF :
1206		/* If there is more to send set PRUS_MORETOCOME */
1207		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1208		top, addr, control, td);
1209	if (dontroute) {
1210		SOCK_LOCK(so);
1211		so->so_options &= ~SO_DONTROUTE;
1212		SOCK_UNLOCK(so);
1213	}
1214	clen = 0;
1215	control = NULL;
1216	top = NULL;
1217out:
1218	if (top != NULL)
1219		m_freem(top);
1220	if (control != NULL)
1221		m_freem(control);
1222	return (error);
1223}
1224
1225/*
1226 * Send on a socket.  If send must go all at once and message is larger than
1227 * send buffering, then hard error.  Lock against other senders.  If must go
1228 * all at once and not enough room now, then inform user that this would
1229 * block and do nothing.  Otherwise, if nonblocking, send as much as
1230 * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1231 * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1232 * in mbuf chain must be small enough to send all at once.
1233 *
1234 * Returns nonzero on error, timeout or signal; callers must check for short
1235 * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1236 * on return.
1237 */
1238int
1239sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1240    struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1241{
1242	long space;
1243	ssize_t resid;
1244	int clen = 0, error, dontroute;
1245	int atomic = sosendallatonce(so) || top;
1246
1247	if (uio != NULL)
1248		resid = uio->uio_resid;
1249	else
1250		resid = top->m_pkthdr.len;
1251	/*
1252	 * In theory resid should be unsigned.  However, space must be
1253	 * signed, as it might be less than 0 if we over-committed, and we
1254	 * must use a signed comparison of space and resid.  On the other
1255	 * hand, a negative resid causes us to loop sending 0-length
1256	 * segments to the protocol.
1257	 *
1258	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1259	 * type sockets since that's an error.
1260	 */
1261	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1262		error = EINVAL;
1263		goto out;
1264	}
1265
1266	dontroute =
1267	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1268	    (so->so_proto->pr_flags & PR_ATOMIC);
1269	if (td != NULL)
1270		td->td_ru.ru_msgsnd++;
1271	if (control != NULL)
1272		clen = control->m_len;
1273
1274	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1275	if (error)
1276		goto out;
1277
1278restart:
1279	do {
1280		SOCKBUF_LOCK(&so->so_snd);
1281		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1282			SOCKBUF_UNLOCK(&so->so_snd);
1283			error = EPIPE;
1284			goto release;
1285		}
1286		if (so->so_error) {
1287			error = so->so_error;
1288			so->so_error = 0;
1289			SOCKBUF_UNLOCK(&so->so_snd);
1290			goto release;
1291		}
1292		if ((so->so_state & SS_ISCONNECTED) == 0) {
1293			/*
1294			 * `sendto' and `sendmsg' is allowed on a connection-
1295			 * based socket if it supports implied connect.
1296			 * Return ENOTCONN if not connected and no address is
1297			 * supplied.
1298			 */
1299			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1300			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1301				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1302				    !(resid == 0 && clen != 0)) {
1303					SOCKBUF_UNLOCK(&so->so_snd);
1304					error = ENOTCONN;
1305					goto release;
1306				}
1307			} else if (addr == NULL) {
1308				SOCKBUF_UNLOCK(&so->so_snd);
1309				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1310					error = ENOTCONN;
1311				else
1312					error = EDESTADDRREQ;
1313				goto release;
1314			}
1315		}
1316		space = sbspace(&so->so_snd);
1317		if (flags & MSG_OOB)
1318			space += 1024;
1319		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1320		    clen > so->so_snd.sb_hiwat) {
1321			SOCKBUF_UNLOCK(&so->so_snd);
1322			error = EMSGSIZE;
1323			goto release;
1324		}
1325		if (space < resid + clen &&
1326		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1327			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1328				if (so->so_upcallprep.soup_send) {
1329					so->so_upcallprep.soup_send(so,
1330						so->so_upcallprep.soup_send_arg,
1331						resid);
1332				}
1333				SOCKBUF_UNLOCK(&so->so_snd);
1334				error = EWOULDBLOCK;
1335				goto release;
1336			}
1337			error = sbwait(&so->so_snd);
1338			SOCKBUF_UNLOCK(&so->so_snd);
1339			if (error)
1340				goto release;
1341			goto restart;
1342		}
1343		SOCKBUF_UNLOCK(&so->so_snd);
1344		space -= clen;
1345		do {
1346			if (uio == NULL) {
1347				resid = 0;
1348				if (flags & MSG_EOR)
1349					top->m_flags |= M_EOR;
1350			} else {
1351#ifdef ZERO_COPY_SOCKETS
1352				error = sosend_copyin(uio, &top, atomic,
1353				    &space, flags);
1354				if (error != 0)
1355					goto release;
1356#else
1357				/*
1358				 * Copy the data from userland into a mbuf
1359				 * chain.  If no data is to be copied in,
1360				 * a single empty mbuf is returned.
1361				 */
1362				top = m_uiotombuf(uio, M_WAITOK, space,
1363				    (atomic ? max_hdr : 0),
1364				    (atomic ? M_PKTHDR : 0) |
1365				    ((flags & MSG_EOR) ? M_EOR : 0));
1366				if (top == NULL) {
1367					error = EFAULT; /* only possible error */
1368					goto release;
1369				}
1370				space -= resid - uio->uio_resid;
1371#endif
1372				resid = uio->uio_resid;
1373			}
1374			if (dontroute) {
1375				SOCK_LOCK(so);
1376				so->so_options |= SO_DONTROUTE;
1377				SOCK_UNLOCK(so);
1378			}
1379			/*
1380			 * XXX all the SBS_CANTSENDMORE checks previously
1381			 * done could be out of date.  We could have recieved
1382			 * a reset packet in an interrupt or maybe we slept
1383			 * while doing page faults in uiomove() etc.  We
1384			 * could probably recheck again inside the locking
1385			 * protection here, but there are probably other
1386			 * places that this also happens.  We must rethink
1387			 * this.
1388			 */
1389			VNET_SO_ASSERT(so);
1390			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1391			    (flags & MSG_OOB) ? PRUS_OOB :
1392			/*
1393			 * If the user set MSG_EOF, the protocol understands
1394			 * this flag and nothing left to send then use
1395			 * PRU_SEND_EOF instead of PRU_SEND.
1396			 */
1397			    ((flags & MSG_EOF) &&
1398			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1399			     (resid <= 0)) ?
1400				PRUS_EOF :
1401			/* If there is more to send set PRUS_MORETOCOME. */
1402			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1403			    top, addr, control, td);
1404			if (dontroute) {
1405				SOCK_LOCK(so);
1406				so->so_options &= ~SO_DONTROUTE;
1407				SOCK_UNLOCK(so);
1408			}
1409			clen = 0;
1410			control = NULL;
1411			top = NULL;
1412			if (error)
1413				goto release;
1414		} while (resid && space > 0);
1415	} while (resid);
1416
1417release:
1418	sbunlock(&so->so_snd);
1419out:
1420	if (top != NULL)
1421		m_freem(top);
1422	if (control != NULL)
1423		m_freem(control);
1424	return (error);
1425}
1426
1427int
1428sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1429    struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1430{
1431	int error;
1432
1433	CURVNET_SET(so->so_vnet);
1434	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1435	    control, flags, td);
1436	CURVNET_RESTORE();
1437	return (error);
1438}
1439
1440/*
1441 * The part of soreceive() that implements reading non-inline out-of-band
1442 * data from a socket.  For more complete comments, see soreceive(), from
1443 * which this code originated.
1444 *
1445 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1446 * unable to return an mbuf chain to the caller.
1447 */
1448static int
1449soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1450{
1451	struct protosw *pr = so->so_proto;
1452	struct mbuf *m;
1453	int error;
1454
1455	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1456	VNET_SO_ASSERT(so);
1457
1458	m = m_get(M_WAIT, MT_DATA);
1459	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1460	if (error)
1461		goto bad;
1462	do {
1463#ifdef ZERO_COPY_SOCKETS
1464		if (so_zero_copy_receive) {
1465			int disposable;
1466
1467			if ((m->m_flags & M_EXT)
1468			 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1469				disposable = 1;
1470			else
1471				disposable = 0;
1472
1473			error = uiomoveco(mtod(m, void *),
1474					  min(uio->uio_resid, m->m_len),
1475					  uio, disposable);
1476		} else
1477#endif /* ZERO_COPY_SOCKETS */
1478		error = uiomove(mtod(m, void *),
1479		    (int) min(uio->uio_resid, m->m_len), uio);
1480		m = m_free(m);
1481	} while (uio->uio_resid && error == 0 && m);
1482bad:
1483	if (m != NULL)
1484		m_freem(m);
1485	return (error);
1486}
1487
1488/*
1489 * Following replacement or removal of the first mbuf on the first mbuf chain
1490 * of a socket buffer, push necessary state changes back into the socket
1491 * buffer so that other consumers see the values consistently.  'nextrecord'
1492 * is the callers locally stored value of the original value of
1493 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1494 * NOTE: 'nextrecord' may be NULL.
1495 */
1496static __inline void
1497sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1498{
1499
1500	SOCKBUF_LOCK_ASSERT(sb);
1501	/*
1502	 * First, update for the new value of nextrecord.  If necessary, make
1503	 * it the first record.
1504	 */
1505	if (sb->sb_mb != NULL)
1506		sb->sb_mb->m_nextpkt = nextrecord;
1507	else
1508		sb->sb_mb = nextrecord;
1509
1510        /*
1511         * Now update any dependent socket buffer fields to reflect the new
1512         * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1513	 * addition of a second clause that takes care of the case where
1514	 * sb_mb has been updated, but remains the last record.
1515         */
1516        if (sb->sb_mb == NULL) {
1517                sb->sb_mbtail = NULL;
1518                sb->sb_lastrecord = NULL;
1519        } else if (sb->sb_mb->m_nextpkt == NULL)
1520                sb->sb_lastrecord = sb->sb_mb;
1521}
1522
1523
1524/*
1525 * Implement receive operations on a socket.  We depend on the way that
1526 * records are added to the sockbuf by sbappend.  In particular, each record
1527 * (mbufs linked through m_next) must begin with an address if the protocol
1528 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1529 * data, and then zero or more mbufs of data.  In order to allow parallelism
1530 * between network receive and copying to user space, as well as avoid
1531 * sleeping with a mutex held, we release the socket buffer mutex during the
1532 * user space copy.  Although the sockbuf is locked, new data may still be
1533 * appended, and thus we must maintain consistency of the sockbuf during that
1534 * time.
1535 *
1536 * The caller may receive the data as a single mbuf chain by supplying an
1537 * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1538 * the count in uio_resid.
1539 */
1540int
1541soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1542    struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1543{
1544	struct mbuf *m, **mp;
1545	int flags, error, offset;
1546	ssize_t len;
1547	struct protosw *pr = so->so_proto;
1548	struct mbuf *nextrecord;
1549	int moff, type = 0, last_m_flags, hole_break = 0;
1550	ssize_t orig_resid = uio->uio_resid;
1551
1552	mp = mp0;
1553	if (psa != NULL)
1554		*psa = NULL;
1555	if (controlp != NULL)
1556		*controlp = NULL;
1557	if (flagsp != NULL) {
1558		hole_break = *flagsp & MSG_HOLE_BREAK;
1559		*flagsp &= ~MSG_HOLE_BREAK;
1560		flags = *flagsp &~ MSG_EOR;
1561	} else
1562		flags = 0;
1563	if (flags & MSG_OOB)
1564		return (soreceive_rcvoob(so, uio, flags));
1565	if (mp != NULL)
1566		*mp = NULL;
1567	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1568	    && uio->uio_resid) {
1569		VNET_SO_ASSERT(so);
1570		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1571	}
1572
1573	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1574	if (error)
1575		return (error);
1576
1577restart:
1578	SOCKBUF_LOCK(&so->so_rcv);
1579	m = so->so_rcv.sb_mb;
1580	/*
1581	 * If we have less data than requested, block awaiting more (subject
1582	 * to any timeout) if:
1583	 *   1. the current count is less than the low water mark, or
1584	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1585	 *	receive operation at once if we block (resid <= hiwat).
1586	 *   3. MSG_DONTWAIT is not set
1587	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1588	 * we have to do the receive in sections, and thus risk returning a
1589	 * short count if a timeout or signal occurs after we start.
1590	 */
1591	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1592	    so->so_rcv.sb_cc < uio->uio_resid) &&
1593	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1594	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1595	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1596		KASSERT(m != NULL || !so->so_rcv.sb_cc,
1597		    ("receive: m == %p so->so_rcv.sb_cc == %u",
1598		    m, so->so_rcv.sb_cc));
1599		if (so->so_error) {
1600			if (m != NULL)
1601				goto dontblock;
1602			error = so->so_error;
1603			if ((flags & MSG_PEEK) == 0)
1604				so->so_error = 0;
1605			SOCKBUF_UNLOCK(&so->so_rcv);
1606			goto release;
1607		}
1608		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1609		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1610			if (m == NULL) {
1611				SOCKBUF_UNLOCK(&so->so_rcv);
1612				goto release;
1613			} else
1614				goto dontblock;
1615		}
1616		for (; m != NULL; m = m->m_next)
1617			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1618				m = so->so_rcv.sb_mb;
1619				goto dontblock;
1620			}
1621		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1622		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1623			SOCKBUF_UNLOCK(&so->so_rcv);
1624			error = ENOTCONN;
1625			goto release;
1626		}
1627		if (uio->uio_resid == 0) {
1628			SOCKBUF_UNLOCK(&so->so_rcv);
1629			goto release;
1630		}
1631		if ((so->so_state & SS_NBIO) ||
1632		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1633			if (so->so_upcallprep.soup_receive != NULL) {
1634				so->so_upcallprep.soup_receive(so,
1635					so->so_upcallprep.soup_receive_arg,
1636					orig_resid - uio->uio_resid, uio->uio_resid);
1637			}
1638			SOCKBUF_UNLOCK(&so->so_rcv);
1639			error = EWOULDBLOCK;
1640			goto release;
1641		}
1642		SBLASTRECORDCHK(&so->so_rcv);
1643		SBLASTMBUFCHK(&so->so_rcv);
1644		error = sbwait(&so->so_rcv);
1645		SOCKBUF_UNLOCK(&so->so_rcv);
1646		if (error)
1647			goto release;
1648		goto restart;
1649	}
1650dontblock:
1651	/*
1652	 * From this point onward, we maintain 'nextrecord' as a cache of the
1653	 * pointer to the next record in the socket buffer.  We must keep the
1654	 * various socket buffer pointers and local stack versions of the
1655	 * pointers in sync, pushing out modifications before dropping the
1656	 * socket buffer mutex, and re-reading them when picking it up.
1657	 *
1658	 * Otherwise, we will race with the network stack appending new data
1659	 * or records onto the socket buffer by using inconsistent/stale
1660	 * versions of the field, possibly resulting in socket buffer
1661	 * corruption.
1662	 *
1663	 * By holding the high-level sblock(), we prevent simultaneous
1664	 * readers from pulling off the front of the socket buffer.
1665	 */
1666	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1667	if (uio->u

Large files files are truncated, but you can click here to view the full file