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
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
32 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
36 * Comments on the socket life cycle:
38 * soalloc() sets of socket layer state for a socket, called only by
39 * socreate() and sonewconn(). Socket layer private.
41 * sodealloc() tears down socket layer state for a socket, called only by
42 * sofree() and sonewconn(). Socket layer private.
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.
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.
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
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
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
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.
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.
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
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.
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.
103 #include <sys/cdefs.h>
104 __FBSDID("$FreeBSD$");
106 #include "opt_inet.h"
107 #include "opt_inet6.h"
108 #include "opt_compat.h"
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/fcntl.h>
113 #include <sys/limits.h>
114 #include <sys/lock.h>
116 #include <sys/malloc.h>
117 #include <sys/mbuf.h>
118 #include <sys/mutex.h>
119 #include <sys/domain.h>
120 #include <sys/file.h> /* for struct knote */
121 #include <sys/hhook.h>
122 #include <sys/kernel.h>
123 #include <sys/khelp.h>
124 #include <sys/event.h>
125 #include <sys/eventhandler.h>
126 #include <sys/poll.h>
127 #include <sys/proc.h>
128 #include <sys/protosw.h>
129 #include <sys/socket.h>
130 #include <sys/socketvar.h>
131 #include <sys/resourcevar.h>
132 #include <net/route.h>
133 #include <sys/signalvar.h>
134 #include <sys/stat.h>
136 #include <sys/sysctl.h>
137 #include <sys/taskqueue.h>
139 #include <sys/jail.h>
140 #include <sys/syslog.h>
141 #include <netinet/in.h>
143 #include <net/vnet.h>
145 #include <security/mac/mac_framework.h>
149 #ifdef COMPAT_FREEBSD32
150 #include <sys/mount.h>
151 #include <sys/sysent.h>
152 #include <compat/freebsd32/freebsd32.h>
155 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
158 static void filt_sordetach(struct knote *kn);
159 static int filt_soread(struct knote *kn, long hint);
160 static void filt_sowdetach(struct knote *kn);
161 static int filt_sowrite(struct knote *kn, long hint);
162 static int filt_solisten(struct knote *kn, long hint);
163 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
164 fo_kqfilter_t soo_kqfilter;
166 static struct filterops solisten_filtops = {
168 .f_detach = filt_sordetach,
169 .f_event = filt_solisten,
171 static struct filterops soread_filtops = {
173 .f_detach = filt_sordetach,
174 .f_event = filt_soread,
176 static struct filterops sowrite_filtops = {
178 .f_detach = filt_sowdetach,
179 .f_event = filt_sowrite,
182 so_gen_t so_gencnt; /* generation count for sockets */
184 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
185 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
187 #define VNET_SO_ASSERT(so) \
188 VNET_ASSERT(curvnet != NULL, \
189 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
191 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
192 #define V_socket_hhh VNET(socket_hhh)
195 * Limit on the number of connections in the listen queue waiting
197 * NB: The original sysctl somaxconn is still available but hidden
198 * to prevent confusion about the actual purpose of this number.
200 static u_int somaxconn = SOMAXCONN;
203 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
209 error = sysctl_handle_int(oidp, &val, 0, req);
210 if (error || !req->newptr )
214 * The purpose of the UINT_MAX / 3 limit, is so that the formula
216 * below, will not overflow.
219 if (val < 1 || val > UINT_MAX / 3)
225 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW,
226 0, sizeof(int), sysctl_somaxconn, "I",
227 "Maximum listen socket pending connection accept queue size");
228 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
229 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP,
230 0, sizeof(int), sysctl_somaxconn, "I",
231 "Maximum listen socket pending connection accept queue size (compat)");
233 static int numopensockets;
234 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
235 &numopensockets, 0, "Number of open sockets");
238 * accept_mtx locks down per-socket fields relating to accept queues. See
239 * socketvar.h for an annotation of the protected fields of struct socket.
241 struct mtx accept_mtx;
242 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
245 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
248 static struct mtx so_global_mtx;
249 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
252 * General IPC sysctl name space, used by sockets and a variety of other IPC
255 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
258 * Initialize the socket subsystem and set up the socket
261 static uma_zone_t socket_zone;
265 socket_zone_change(void *tag)
268 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
272 socket_hhook_register(int subtype)
275 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
276 &V_socket_hhh[subtype],
277 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
278 printf("%s: WARNING: unable to register hook\n", __func__);
282 socket_hhook_deregister(int subtype)
285 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
286 printf("%s: WARNING: unable to deregister hook\n", __func__);
290 socket_init(void *tag)
293 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
294 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
295 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
296 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
297 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
298 EVENTHANDLER_PRI_FIRST);
300 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
303 socket_vnet_init(const void *unused __unused)
307 /* We expect a contiguous range */
308 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
309 socket_hhook_register(i);
311 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
312 socket_vnet_init, NULL);
315 socket_vnet_uninit(const void *unused __unused)
319 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
320 socket_hhook_deregister(i);
322 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
323 socket_vnet_uninit, NULL);
326 * Initialise maxsockets. This SYSINIT must be run after
330 init_maxsockets(void *ignored)
333 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
334 maxsockets = imax(maxsockets, maxfiles);
336 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
339 * Sysctl to get and set the maximum global sockets limit. Notify protocols
340 * of the change so that they can update their dependent limits as required.
343 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
345 int error, newmaxsockets;
347 newmaxsockets = maxsockets;
348 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
349 if (error == 0 && req->newptr) {
350 if (newmaxsockets > maxsockets &&
351 newmaxsockets <= maxfiles) {
352 maxsockets = newmaxsockets;
353 EVENTHANDLER_INVOKE(maxsockets_change);
359 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
360 &maxsockets, 0, sysctl_maxsockets, "IU",
361 "Maximum number of sockets available");
364 * Socket operation routines. These routines are called by the routines in
365 * sys_socket.c or from a system process, and implement the semantics of
366 * socket operations by switching out to the protocol specific routines.
370 * Get a socket structure from our zone, and initialize it. Note that it
371 * would probably be better to allocate socket and PCB at the same time, but
372 * I'm not convinced that all the protocols can be easily modified to do
375 * soalloc() returns a socket with a ref count of 0.
377 static struct socket *
378 soalloc(struct vnet *vnet)
382 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
386 if (mac_socket_init(so, M_NOWAIT) != 0) {
387 uma_zfree(socket_zone, so);
391 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
392 uma_zfree(socket_zone, so);
396 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
397 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
398 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
399 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
400 TAILQ_INIT(&so->so_snd.sb_aiojobq);
401 TAILQ_INIT(&so->so_rcv.sb_aiojobq);
402 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
403 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
405 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
406 __func__, __LINE__, so));
409 /* We shouldn't need the so_global_mtx */
410 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
411 /* Do we need more comprehensive error returns? */
412 uma_zfree(socket_zone, so);
415 mtx_lock(&so_global_mtx);
416 so->so_gencnt = ++so_gencnt;
419 vnet->vnet_sockcnt++;
421 mtx_unlock(&so_global_mtx);
427 * Free the storage associated with a socket at the socket layer, tear down
428 * locks, labels, etc. All protocol state is assumed already to have been
429 * torn down (and possibly never set up) by the caller.
432 sodealloc(struct socket *so)
435 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
436 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
438 mtx_lock(&so_global_mtx);
439 so->so_gencnt = ++so_gencnt;
440 --numopensockets; /* Could be below, but faster here. */
442 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
443 __func__, __LINE__, so));
444 so->so_vnet->vnet_sockcnt--;
446 mtx_unlock(&so_global_mtx);
447 if (so->so_rcv.sb_hiwat)
448 (void)chgsbsize(so->so_cred->cr_uidinfo,
449 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
450 if (so->so_snd.sb_hiwat)
451 (void)chgsbsize(so->so_cred->cr_uidinfo,
452 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
453 /* remove accept filter if one is present. */
454 if (so->so_accf != NULL)
455 do_setopt_accept_filter(so, NULL);
457 mac_socket_destroy(so);
459 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
462 khelp_destroy_osd(&so->osd);
463 sx_destroy(&so->so_snd.sb_sx);
464 sx_destroy(&so->so_rcv.sb_sx);
465 SOCKBUF_LOCK_DESTROY(&so->so_snd);
466 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
467 uma_zfree(socket_zone, so);
471 * socreate returns a socket with a ref count of 1. The socket should be
472 * closed with soclose().
475 socreate(int dom, struct socket **aso, int type, int proto,
476 struct ucred *cred, struct thread *td)
483 prp = pffindproto(dom, proto, type);
485 prp = pffindtype(dom, type);
488 /* No support for domain. */
489 if (pffinddomain(dom) == NULL)
490 return (EAFNOSUPPORT);
491 /* No support for socket type. */
492 if (proto == 0 && type != 0)
494 return (EPROTONOSUPPORT);
496 if (prp->pr_usrreqs->pru_attach == NULL ||
497 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
498 return (EPROTONOSUPPORT);
500 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
501 return (EPROTONOSUPPORT);
503 if (prp->pr_type != type)
505 so = soalloc(CRED_TO_VNET(cred));
509 TAILQ_INIT(&so->so_incomp);
510 TAILQ_INIT(&so->so_comp);
512 so->so_cred = crhold(cred);
513 if ((prp->pr_domain->dom_family == PF_INET) ||
514 (prp->pr_domain->dom_family == PF_INET6) ||
515 (prp->pr_domain->dom_family == PF_ROUTE))
516 so->so_fibnum = td->td_proc->p_fibnum;
521 mac_socket_create(cred, so);
523 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
524 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
527 * Auto-sizing of socket buffers is managed by the protocols and
528 * the appropriate flags must be set in the pru_attach function.
530 CURVNET_SET(so->so_vnet);
531 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
534 KASSERT(so->so_count == 1, ("socreate: so_count %d",
545 static int regression_sonewconn_earlytest = 1;
546 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
547 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
551 * When an attempt at a new connection is noted on a socket which accepts
552 * connections, sonewconn is called. If the connection is possible (subject
553 * to space constraints, etc.) then we allocate a new structure, properly
554 * linked into the data structure of the original socket, and return this.
555 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
557 * Note: the ref count on the socket is 0 on return.
560 sonewconn(struct socket *head, int connstatus)
562 static struct timeval lastover;
563 static struct timeval overinterval = { 60, 0 };
564 static int overcount;
570 over = (head->so_qlen > 3 * head->so_qlimit / 2);
573 if (regression_sonewconn_earlytest && over) {
579 if (ratecheck(&lastover, &overinterval)) {
580 log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
581 "%i already in queue awaiting acceptance "
582 "(%d occurrences)\n",
583 __func__, head->so_pcb, head->so_qlen, overcount);
590 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
591 __func__, __LINE__, head));
592 so = soalloc(head->so_vnet);
594 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
595 "limit reached or out of memory\n",
596 __func__, head->so_pcb);
599 if ((head->so_options & SO_ACCEPTFILTER) != 0)
602 so->so_type = head->so_type;
603 so->so_options = head->so_options &~ SO_ACCEPTCONN;
604 so->so_linger = head->so_linger;
605 so->so_state = head->so_state | SS_NOFDREF;
606 so->so_fibnum = head->so_fibnum;
607 so->so_proto = head->so_proto;
608 so->so_cred = crhold(head->so_cred);
610 mac_socket_newconn(head, so);
612 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
613 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
614 VNET_SO_ASSERT(head);
615 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
617 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
618 __func__, head->so_pcb);
621 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
623 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
624 __func__, head->so_pcb);
627 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
628 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
629 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
630 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
631 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
632 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
633 so->so_state |= connstatus;
636 * The accept socket may be tearing down but we just
637 * won a race on the ACCEPT_LOCK.
638 * However, if sctp_peeloff() is called on a 1-to-many
639 * style socket, the SO_ACCEPTCONN doesn't need to be set.
641 if (!(head->so_options & SO_ACCEPTCONN) &&
642 ((head->so_proto->pr_protocol != IPPROTO_SCTP) ||
643 (head->so_type != SOCK_SEQPACKET))) {
646 sofree(so); /* NB: returns ACCEPT_UNLOCK'ed. */
650 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
651 so->so_qstate |= SQ_COMP;
655 * Keep removing sockets from the head until there's room for
656 * us to insert on the tail. In pre-locking revisions, this
657 * was a simple if(), but as we could be racing with other
658 * threads and soabort() requires dropping locks, we must
659 * loop waiting for the condition to be true.
661 while (head->so_incqlen > head->so_qlimit) {
663 sp = TAILQ_FIRST(&head->so_incomp);
664 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
666 sp->so_qstate &= ~SQ_INCOMP;
672 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
673 so->so_qstate |= SQ_INCOMP;
679 wakeup_one(&head->so_timeo);
685 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
689 CURVNET_SET(so->so_vnet);
690 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
696 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
700 CURVNET_SET(so->so_vnet);
701 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
707 * solisten() transitions a socket from a non-listening state to a listening
708 * state, but can also be used to update the listen queue depth on an
709 * existing listen socket. The protocol will call back into the sockets
710 * layer using solisten_proto_check() and solisten_proto() to check and set
711 * socket-layer listen state. Call backs are used so that the protocol can
712 * acquire both protocol and socket layer locks in whatever order is required
715 * Protocol implementors are advised to hold the socket lock across the
716 * socket-layer test and set to avoid races at the socket layer.
719 solisten(struct socket *so, int backlog, struct thread *td)
723 CURVNET_SET(so->so_vnet);
724 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
730 solisten_proto_check(struct socket *so)
733 SOCK_LOCK_ASSERT(so);
735 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
742 solisten_proto(struct socket *so, int backlog)
745 SOCK_LOCK_ASSERT(so);
747 if (backlog < 0 || backlog > somaxconn)
749 so->so_qlimit = backlog;
750 so->so_options |= SO_ACCEPTCONN;
754 * Evaluate the reference count and named references on a socket; if no
755 * references remain, free it. This should be called whenever a reference is
756 * released, such as in sorele(), but also when named reference flags are
757 * cleared in socket or protocol code.
759 * sofree() will free the socket if:
761 * - There are no outstanding file descriptor references or related consumers
764 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
766 * - The protocol does not have an outstanding strong reference on the socket
769 * - The socket is not in a completed connection queue, so a process has been
770 * notified that it is present. If it is removed, the user process may
771 * block in accept() despite select() saying the socket was ready.
774 sofree(struct socket *so)
776 struct protosw *pr = so->so_proto;
779 ACCEPT_LOCK_ASSERT();
780 SOCK_LOCK_ASSERT(so);
782 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
783 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
791 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
792 (so->so_qstate & SQ_INCOMP) != 0,
793 ("sofree: so_head != NULL, but neither SQ_COMP nor "
795 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
796 (so->so_qstate & SQ_INCOMP) == 0,
797 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
798 TAILQ_REMOVE(&head->so_incomp, so, so_list);
800 so->so_qstate &= ~SQ_INCOMP;
803 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
804 (so->so_qstate & SQ_INCOMP) == 0,
805 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
806 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
807 if (so->so_options & SO_ACCEPTCONN) {
808 KASSERT((TAILQ_EMPTY(&so->so_comp)),
809 ("sofree: so_comp populated"));
810 KASSERT((TAILQ_EMPTY(&so->so_incomp)),
811 ("sofree: so_incomp populated"));
817 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
818 (*pr->pr_domain->dom_dispose)(so);
819 if (pr->pr_usrreqs->pru_detach != NULL)
820 (*pr->pr_usrreqs->pru_detach)(so);
823 * From this point on, we assume that no other references to this
824 * socket exist anywhere else in the stack. Therefore, no locks need
825 * to be acquired or held.
827 * We used to do a lot of socket buffer and socket locking here, as
828 * well as invoke sorflush() and perform wakeups. The direct call to
829 * dom_dispose() and sbrelease_internal() are an inlining of what was
830 * necessary from sorflush().
832 * Notice that the socket buffer and kqueue state are torn down
833 * before calling pru_detach. This means that protocols shold not
834 * assume they can perform socket wakeups, etc, in their detach code.
836 sbdestroy(&so->so_snd, so);
837 sbdestroy(&so->so_rcv, so);
838 seldrain(&so->so_snd.sb_sel);
839 seldrain(&so->so_rcv.sb_sel);
840 knlist_destroy(&so->so_rcv.sb_sel.si_note);
841 knlist_destroy(&so->so_snd.sb_sel.si_note);
846 * Close a socket on last file table reference removal. Initiate disconnect
847 * if connected. Free socket when disconnect complete.
849 * This function will sorele() the socket. Note that soclose() may be called
850 * prior to the ref count reaching zero. The actual socket structure will
851 * not be freed until the ref count reaches zero.
854 soclose(struct socket *so)
858 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
860 CURVNET_SET(so->so_vnet);
861 funsetown(&so->so_sigio);
862 if (so->so_state & SS_ISCONNECTED) {
863 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
864 error = sodisconnect(so);
866 if (error == ENOTCONN)
871 if (so->so_options & SO_LINGER) {
872 if ((so->so_state & SS_ISDISCONNECTING) &&
873 (so->so_state & SS_NBIO))
875 while (so->so_state & SS_ISCONNECTED) {
876 error = tsleep(&so->so_timeo,
877 PSOCK | PCATCH, "soclos",
886 if (so->so_proto->pr_usrreqs->pru_close != NULL)
887 (*so->so_proto->pr_usrreqs->pru_close)(so);
889 if (so->so_options & SO_ACCEPTCONN) {
892 * Prevent new additions to the accept queues due
893 * to ACCEPT_LOCK races while we are draining them.
895 so->so_options &= ~SO_ACCEPTCONN;
896 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
897 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
899 sp->so_qstate &= ~SQ_INCOMP;
905 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
906 TAILQ_REMOVE(&so->so_comp, sp, so_list);
908 sp->so_qstate &= ~SQ_COMP;
914 KASSERT((TAILQ_EMPTY(&so->so_comp)),
915 ("%s: so_comp populated", __func__));
916 KASSERT((TAILQ_EMPTY(&so->so_incomp)),
917 ("%s: so_incomp populated", __func__));
920 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
921 so->so_state |= SS_NOFDREF;
922 sorele(so); /* NB: Returns with ACCEPT_UNLOCK(). */
928 * soabort() is used to abruptly tear down a connection, such as when a
929 * resource limit is reached (listen queue depth exceeded), or if a listen
930 * socket is closed while there are sockets waiting to be accepted.
932 * This interface is tricky, because it is called on an unreferenced socket,
933 * and must be called only by a thread that has actually removed the socket
934 * from the listen queue it was on, or races with other threads are risked.
936 * This interface will call into the protocol code, so must not be called
937 * with any socket locks held. Protocols do call it while holding their own
938 * recursible protocol mutexes, but this is something that should be subject
939 * to review in the future.
942 soabort(struct socket *so)
946 * In as much as is possible, assert that no references to this
947 * socket are held. This is not quite the same as asserting that the
948 * current thread is responsible for arranging for no references, but
949 * is as close as we can get for now.
951 KASSERT(so->so_count == 0, ("soabort: so_count"));
952 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
953 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
954 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
955 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
958 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
959 (*so->so_proto->pr_usrreqs->pru_abort)(so);
966 soaccept(struct socket *so, struct sockaddr **nam)
971 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
972 so->so_state &= ~SS_NOFDREF;
975 CURVNET_SET(so->so_vnet);
976 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
982 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
985 return (soconnectat(AT_FDCWD, so, nam, td));
989 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
993 if (so->so_options & SO_ACCEPTCONN)
996 CURVNET_SET(so->so_vnet);
998 * If protocol is connection-based, can only connect once.
999 * Otherwise, if connected, try to disconnect first. This allows
1000 * user to disconnect by connecting to, e.g., a null address.
1002 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1003 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1004 (error = sodisconnect(so)))) {
1008 * Prevent accumulated error from previous connection from
1012 if (fd == AT_FDCWD) {
1013 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1016 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1026 soconnect2(struct socket *so1, struct socket *so2)
1030 CURVNET_SET(so1->so_vnet);
1031 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1037 sodisconnect(struct socket *so)
1041 if ((so->so_state & SS_ISCONNECTED) == 0)
1043 if (so->so_state & SS_ISDISCONNECTING)
1046 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1050 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1053 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1054 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1058 int clen = 0, error, dontroute;
1060 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1061 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1062 ("sosend_dgram: !PR_ATOMIC"));
1065 resid = uio->uio_resid;
1067 resid = top->m_pkthdr.len;
1069 * In theory resid should be unsigned. However, space must be
1070 * signed, as it might be less than 0 if we over-committed, and we
1071 * must use a signed comparison of space and resid. On the other
1072 * hand, a negative resid causes us to loop sending 0-length
1073 * segments to the protocol.
1081 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1083 td->td_ru.ru_msgsnd++;
1084 if (control != NULL)
1085 clen = control->m_len;
1087 SOCKBUF_LOCK(&so->so_snd);
1088 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1089 SOCKBUF_UNLOCK(&so->so_snd);
1094 error = so->so_error;
1096 SOCKBUF_UNLOCK(&so->so_snd);
1099 if ((so->so_state & SS_ISCONNECTED) == 0) {
1101 * `sendto' and `sendmsg' is allowed on a connection-based
1102 * socket if it supports implied connect. Return ENOTCONN if
1103 * not connected and no address is supplied.
1105 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1106 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1107 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1108 !(resid == 0 && clen != 0)) {
1109 SOCKBUF_UNLOCK(&so->so_snd);
1113 } else if (addr == NULL) {
1114 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1117 error = EDESTADDRREQ;
1118 SOCKBUF_UNLOCK(&so->so_snd);
1124 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1125 * problem and need fixing.
1127 space = sbspace(&so->so_snd);
1128 if (flags & MSG_OOB)
1131 SOCKBUF_UNLOCK(&so->so_snd);
1132 if (resid > space) {
1138 if (flags & MSG_EOR)
1139 top->m_flags |= M_EOR;
1142 * Copy the data from userland into a mbuf chain.
1143 * If no data is to be copied in, a single empty mbuf
1146 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1147 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1149 error = EFAULT; /* only possible error */
1152 space -= resid - uio->uio_resid;
1153 resid = uio->uio_resid;
1155 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1157 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1162 so->so_options |= SO_DONTROUTE;
1166 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1167 * of date. We could have received a reset packet in an interrupt or
1168 * maybe we slept while doing page faults in uiomove() etc. We could
1169 * probably recheck again inside the locking protection here, but
1170 * there are probably other places that this also happens. We must
1174 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1175 (flags & MSG_OOB) ? PRUS_OOB :
1177 * If the user set MSG_EOF, the protocol understands this flag and
1178 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1180 ((flags & MSG_EOF) &&
1181 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1184 /* If there is more to send set PRUS_MORETOCOME */
1185 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1186 top, addr, control, td);
1189 so->so_options &= ~SO_DONTROUTE;
1198 if (control != NULL)
1204 * Send on a socket. If send must go all at once and message is larger than
1205 * send buffering, then hard error. Lock against other senders. If must go
1206 * all at once and not enough room now, then inform user that this would
1207 * block and do nothing. Otherwise, if nonblocking, send as much as
1208 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1209 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1210 * in mbuf chain must be small enough to send all at once.
1212 * Returns nonzero on error, timeout or signal; callers must check for short
1213 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1217 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1218 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1222 int clen = 0, error, dontroute;
1223 int atomic = sosendallatonce(so) || top;
1226 resid = uio->uio_resid;
1228 resid = top->m_pkthdr.len;
1230 * In theory resid should be unsigned. However, space must be
1231 * signed, as it might be less than 0 if we over-committed, and we
1232 * must use a signed comparison of space and resid. On the other
1233 * hand, a negative resid causes us to loop sending 0-length
1234 * segments to the protocol.
1236 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1237 * type sockets since that's an error.
1239 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1245 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1246 (so->so_proto->pr_flags & PR_ATOMIC);
1248 td->td_ru.ru_msgsnd++;
1249 if (control != NULL)
1250 clen = control->m_len;
1252 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1258 SOCKBUF_LOCK(&so->so_snd);
1259 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1260 SOCKBUF_UNLOCK(&so->so_snd);
1265 error = so->so_error;
1267 SOCKBUF_UNLOCK(&so->so_snd);
1270 if ((so->so_state & SS_ISCONNECTED) == 0) {
1272 * `sendto' and `sendmsg' is allowed on a connection-
1273 * based socket if it supports implied connect.
1274 * Return ENOTCONN if not connected and no address is
1277 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1278 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1279 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1280 !(resid == 0 && clen != 0)) {
1281 SOCKBUF_UNLOCK(&so->so_snd);
1285 } else if (addr == NULL) {
1286 SOCKBUF_UNLOCK(&so->so_snd);
1287 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1290 error = EDESTADDRREQ;
1294 space = sbspace(&so->so_snd);
1295 if (flags & MSG_OOB)
1297 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1298 clen > so->so_snd.sb_hiwat) {
1299 SOCKBUF_UNLOCK(&so->so_snd);
1303 if (space < resid + clen &&
1304 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1305 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1306 SOCKBUF_UNLOCK(&so->so_snd);
1307 error = EWOULDBLOCK;
1310 error = sbwait(&so->so_snd);
1311 SOCKBUF_UNLOCK(&so->so_snd);
1316 SOCKBUF_UNLOCK(&so->so_snd);
1321 if (flags & MSG_EOR)
1322 top->m_flags |= M_EOR;
1325 * Copy the data from userland into a mbuf
1326 * chain. If resid is 0, which can happen
1327 * only if we have control to send, then
1328 * a single empty mbuf is returned. This
1329 * is a workaround to prevent protocol send
1332 top = m_uiotombuf(uio, M_WAITOK, space,
1333 (atomic ? max_hdr : 0),
1334 (atomic ? M_PKTHDR : 0) |
1335 ((flags & MSG_EOR) ? M_EOR : 0));
1337 error = EFAULT; /* only possible error */
1340 space -= resid - uio->uio_resid;
1341 resid = uio->uio_resid;
1345 so->so_options |= SO_DONTROUTE;
1349 * XXX all the SBS_CANTSENDMORE checks previously
1350 * done could be out of date. We could have received
1351 * a reset packet in an interrupt or maybe we slept
1352 * while doing page faults in uiomove() etc. We
1353 * could probably recheck again inside the locking
1354 * protection here, but there are probably other
1355 * places that this also happens. We must rethink
1359 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1360 (flags & MSG_OOB) ? PRUS_OOB :
1362 * If the user set MSG_EOF, the protocol understands
1363 * this flag and nothing left to send then use
1364 * PRU_SEND_EOF instead of PRU_SEND.
1366 ((flags & MSG_EOF) &&
1367 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1370 /* If there is more to send set PRUS_MORETOCOME. */
1371 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1372 top, addr, control, td);
1375 so->so_options &= ~SO_DONTROUTE;
1383 } while (resid && space > 0);
1387 sbunlock(&so->so_snd);
1391 if (control != NULL)
1397 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1398 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1402 CURVNET_SET(so->so_vnet);
1403 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1404 control, flags, td);
1410 * The part of soreceive() that implements reading non-inline out-of-band
1411 * data from a socket. For more complete comments, see soreceive(), from
1412 * which this code originated.
1414 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1415 * unable to return an mbuf chain to the caller.
1418 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1420 struct protosw *pr = so->so_proto;
1424 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1427 m = m_get(M_WAITOK, MT_DATA);
1428 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1432 error = uiomove(mtod(m, void *),
1433 (int) min(uio->uio_resid, m->m_len), uio);
1435 } while (uio->uio_resid && error == 0 && m);
1443 * Following replacement or removal of the first mbuf on the first mbuf chain
1444 * of a socket buffer, push necessary state changes back into the socket
1445 * buffer so that other consumers see the values consistently. 'nextrecord'
1446 * is the callers locally stored value of the original value of
1447 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1448 * NOTE: 'nextrecord' may be NULL.
1450 static __inline void
1451 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1454 SOCKBUF_LOCK_ASSERT(sb);
1456 * First, update for the new value of nextrecord. If necessary, make
1457 * it the first record.
1459 if (sb->sb_mb != NULL)
1460 sb->sb_mb->m_nextpkt = nextrecord;
1462 sb->sb_mb = nextrecord;
1465 * Now update any dependent socket buffer fields to reflect the new
1466 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1467 * addition of a second clause that takes care of the case where
1468 * sb_mb has been updated, but remains the last record.
1470 if (sb->sb_mb == NULL) {
1471 sb->sb_mbtail = NULL;
1472 sb->sb_lastrecord = NULL;
1473 } else if (sb->sb_mb->m_nextpkt == NULL)
1474 sb->sb_lastrecord = sb->sb_mb;
1478 * Implement receive operations on a socket. We depend on the way that
1479 * records are added to the sockbuf by sbappend. In particular, each record
1480 * (mbufs linked through m_next) must begin with an address if the protocol
1481 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1482 * data, and then zero or more mbufs of data. In order to allow parallelism
1483 * between network receive and copying to user space, as well as avoid
1484 * sleeping with a mutex held, we release the socket buffer mutex during the
1485 * user space copy. Although the sockbuf is locked, new data may still be
1486 * appended, and thus we must maintain consistency of the sockbuf during that
1489 * The caller may receive the data as a single mbuf chain by supplying an
1490 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1491 * the count in uio_resid.
1494 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1495 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1497 struct mbuf *m, **mp;
1498 int flags, error, offset;
1500 struct protosw *pr = so->so_proto;
1501 struct mbuf *nextrecord;
1503 ssize_t orig_resid = uio->uio_resid;
1508 if (controlp != NULL)
1511 flags = *flagsp &~ MSG_EOR;
1514 if (flags & MSG_OOB)
1515 return (soreceive_rcvoob(so, uio, flags));
1518 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1519 && uio->uio_resid) {
1521 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1524 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1529 SOCKBUF_LOCK(&so->so_rcv);
1530 m = so->so_rcv.sb_mb;
1532 * If we have less data than requested, block awaiting more (subject
1533 * to any timeout) if:
1534 * 1. the current count is less than the low water mark, or
1535 * 2. MSG_DONTWAIT is not set
1537 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1538 sbavail(&so->so_rcv) < uio->uio_resid) &&
1539 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1540 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1541 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1542 ("receive: m == %p sbavail == %u",
1543 m, sbavail(&so->so_rcv)));
1547 error = so->so_error;
1548 if ((flags & MSG_PEEK) == 0)
1550 SOCKBUF_UNLOCK(&so->so_rcv);
1553 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1554 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1556 SOCKBUF_UNLOCK(&so->so_rcv);
1561 for (; m != NULL; m = m->m_next)
1562 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1563 m = so->so_rcv.sb_mb;
1566 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1567 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1568 SOCKBUF_UNLOCK(&so->so_rcv);
1572 if (uio->uio_resid == 0) {
1573 SOCKBUF_UNLOCK(&so->so_rcv);
1576 if ((so->so_state & SS_NBIO) ||
1577 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1578 SOCKBUF_UNLOCK(&so->so_rcv);
1579 error = EWOULDBLOCK;
1582 SBLASTRECORDCHK(&so->so_rcv);
1583 SBLASTMBUFCHK(&so->so_rcv);
1584 error = sbwait(&so->so_rcv);
1585 SOCKBUF_UNLOCK(&so->so_rcv);
1592 * From this point onward, we maintain 'nextrecord' as a cache of the
1593 * pointer to the next record in the socket buffer. We must keep the
1594 * various socket buffer pointers and local stack versions of the
1595 * pointers in sync, pushing out modifications before dropping the
1596 * socket buffer mutex, and re-reading them when picking it up.
1598 * Otherwise, we will race with the network stack appending new data
1599 * or records onto the socket buffer by using inconsistent/stale
1600 * versions of the field, possibly resulting in socket buffer
1603 * By holding the high-level sblock(), we prevent simultaneous
1604 * readers from pulling off the front of the socket buffer.
1606 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1608 uio->uio_td->td_ru.ru_msgrcv++;
1609 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1610 SBLASTRECORDCHK(&so->so_rcv);
1611 SBLASTMBUFCHK(&so->so_rcv);
1612 nextrecord = m->m_nextpkt;
1613 if (pr->pr_flags & PR_ADDR) {
1614 KASSERT(m->m_type == MT_SONAME,
1615 ("m->m_type == %d", m->m_type));
1618 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1620 if (flags & MSG_PEEK) {
1623 sbfree(&so->so_rcv, m);
1624 so->so_rcv.sb_mb = m_free(m);
1625 m = so->so_rcv.sb_mb;
1626 sockbuf_pushsync(&so->so_rcv, nextrecord);
1631 * Process one or more MT_CONTROL mbufs present before any data mbufs
1632 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1633 * just copy the data; if !MSG_PEEK, we call into the protocol to
1634 * perform externalization (or freeing if controlp == NULL).
1636 if (m != NULL && m->m_type == MT_CONTROL) {
1637 struct mbuf *cm = NULL, *cmn;
1638 struct mbuf **cme = &cm;
1641 if (flags & MSG_PEEK) {
1642 if (controlp != NULL) {
1643 *controlp = m_copy(m, 0, m->m_len);
1644 controlp = &(*controlp)->m_next;
1648 sbfree(&so->so_rcv, m);
1649 so->so_rcv.sb_mb = m->m_next;
1652 cme = &(*cme)->m_next;
1653 m = so->so_rcv.sb_mb;
1655 } while (m != NULL && m->m_type == MT_CONTROL);
1656 if ((flags & MSG_PEEK) == 0)
1657 sockbuf_pushsync(&so->so_rcv, nextrecord);
1658 while (cm != NULL) {
1661 if (pr->pr_domain->dom_externalize != NULL) {
1662 SOCKBUF_UNLOCK(&so->so_rcv);
1664 error = (*pr->pr_domain->dom_externalize)
1665 (cm, controlp, flags);
1666 SOCKBUF_LOCK(&so->so_rcv);
1667 } else if (controlp != NULL)
1671 if (controlp != NULL) {
1673 while (*controlp != NULL)
1674 controlp = &(*controlp)->m_next;
1679 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1681 nextrecord = so->so_rcv.sb_mb;
1685 if ((flags & MSG_PEEK) == 0) {
1686 KASSERT(m->m_nextpkt == nextrecord,
1687 ("soreceive: post-control, nextrecord !sync"));
1688 if (nextrecord == NULL) {
1689 KASSERT(so->so_rcv.sb_mb == m,
1690 ("soreceive: post-control, sb_mb!=m"));
1691 KASSERT(so->so_rcv.sb_lastrecord == m,
1692 ("soreceive: post-control, lastrecord!=m"));
1696 if (type == MT_OOBDATA)
1699 if ((flags & MSG_PEEK) == 0) {
1700 KASSERT(so->so_rcv.sb_mb == nextrecord,
1701 ("soreceive: sb_mb != nextrecord"));
1702 if (so->so_rcv.sb_mb == NULL) {
1703 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1704 ("soreceive: sb_lastercord != NULL"));
1708 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1709 SBLASTRECORDCHK(&so->so_rcv);
1710 SBLASTMBUFCHK(&so->so_rcv);
1713 * Now continue to read any data mbufs off of the head of the socket
1714 * buffer until the read request is satisfied. Note that 'type' is
1715 * used to store the type of any mbuf reads that have happened so far
1716 * such that soreceive() can stop reading if the type changes, which
1717 * causes soreceive() to return only one of regular data and inline
1718 * out-of-band data in a single socket receive operation.
1722 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
1725 * If the type of mbuf has changed since the last mbuf
1726 * examined ('type'), end the receive operation.
1728 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1729 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1730 if (type != m->m_type)
1732 } else if (type == MT_OOBDATA)
1735 KASSERT(m->m_type == MT_DATA,
1736 ("m->m_type == %d", m->m_type));
1737 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1738 len = uio->uio_resid;
1739 if (so->so_oobmark && len > so->so_oobmark - offset)
1740 len = so->so_oobmark - offset;
1741 if (len > m->m_len - moff)
1742 len = m->m_len - moff;
1744 * If mp is set, just pass back the mbufs. Otherwise copy
1745 * them out via the uio, then free. Sockbuf must be
1746 * consistent here (points to current mbuf, it points to next
1747 * record) when we drop priority; we must note any additions
1748 * to the sockbuf when we block interrupts again.
1751 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1752 SBLASTRECORDCHK(&so->so_rcv);
1753 SBLASTMBUFCHK(&so->so_rcv);
1754 SOCKBUF_UNLOCK(&so->so_rcv);
1755 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1756 SOCKBUF_LOCK(&so->so_rcv);
1759 * The MT_SONAME mbuf has already been removed
1760 * from the record, so it is necessary to
1761 * remove the data mbufs, if any, to preserve
1762 * the invariant in the case of PR_ADDR that
1763 * requires MT_SONAME mbufs at the head of
1766 if (m && pr->pr_flags & PR_ATOMIC &&
1767 ((flags & MSG_PEEK) == 0))
1768 (void)sbdroprecord_locked(&so->so_rcv);
1769 SOCKBUF_UNLOCK(&so->so_rcv);
1773 uio->uio_resid -= len;
1774 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1775 if (len == m->m_len - moff) {
1776 if (m->m_flags & M_EOR)
1778 if (flags & MSG_PEEK) {
1782 nextrecord = m->m_nextpkt;
1783 sbfree(&so->so_rcv, m);
1785 m->m_nextpkt = NULL;
1788 so->so_rcv.sb_mb = m = m->m_next;
1791 so->so_rcv.sb_mb = m_free(m);
1792 m = so->so_rcv.sb_mb;
1794 sockbuf_pushsync(&so->so_rcv, nextrecord);
1795 SBLASTRECORDCHK(&so->so_rcv);
1796 SBLASTMBUFCHK(&so->so_rcv);
1799 if (flags & MSG_PEEK)
1803 if (flags & MSG_DONTWAIT) {
1804 *mp = m_copym(m, 0, len,
1808 * m_copym() couldn't
1810 * Adjust uio_resid back
1812 * down by len bytes,
1813 * which we didn't end
1814 * up "copying" over).
1816 uio->uio_resid += len;
1820 SOCKBUF_UNLOCK(&so->so_rcv);
1821 *mp = m_copym(m, 0, len,
1823 SOCKBUF_LOCK(&so->so_rcv);
1826 sbcut_locked(&so->so_rcv, len);
1829 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1830 if (so->so_oobmark) {
1831 if ((flags & MSG_PEEK) == 0) {
1832 so->so_oobmark -= len;
1833 if (so->so_oobmark == 0) {
1834 so->so_rcv.sb_state |= SBS_RCVATMARK;
1839 if (offset == so->so_oobmark)
1843 if (flags & MSG_EOR)
1846 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1847 * must not quit until "uio->uio_resid == 0" or an error
1848 * termination. If a signal/timeout occurs, return with a
1849 * short count but without error. Keep sockbuf locked
1850 * against other readers.
1852 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1853 !sosendallatonce(so) && nextrecord == NULL) {
1854 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1856 so->so_rcv.sb_state & SBS_CANTRCVMORE)
1859 * Notify the protocol that some data has been
1860 * drained before blocking.
1862 if (pr->pr_flags & PR_WANTRCVD) {
1863 SOCKBUF_UNLOCK(&so->so_rcv);
1865 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1866 SOCKBUF_LOCK(&so->so_rcv);
1868 SBLASTRECORDCHK(&so->so_rcv);
1869 SBLASTMBUFCHK(&so->so_rcv);
1871 * We could receive some data while was notifying
1872 * the protocol. Skip blocking in this case.
1874 if (so->so_rcv.sb_mb == NULL) {
1875 error = sbwait(&so->so_rcv);
1877 SOCKBUF_UNLOCK(&so->so_rcv);
1881 m = so->so_rcv.sb_mb;
1883 nextrecord = m->m_nextpkt;
1887 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1888 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1890 if ((flags & MSG_PEEK) == 0)
1891 (void) sbdroprecord_locked(&so->so_rcv);
1893 if ((flags & MSG_PEEK) == 0) {
1896 * First part is an inline SB_EMPTY_FIXUP(). Second
1897 * part makes sure sb_lastrecord is up-to-date if
1898 * there is still data in the socket buffer.
1900 so->so_rcv.sb_mb = nextrecord;
1901 if (so->so_rcv.sb_mb == NULL) {
1902 so->so_rcv.sb_mbtail = NULL;
1903 so->so_rcv.sb_lastrecord = NULL;
1904 } else if (nextrecord->m_nextpkt == NULL)
1905 so->so_rcv.sb_lastrecord = nextrecord;
1907 SBLASTRECORDCHK(&so->so_rcv);
1908 SBLASTMBUFCHK(&so->so_rcv);
1910 * If soreceive() is being done from the socket callback,
1911 * then don't need to generate ACK to peer to update window,
1912 * since ACK will be generated on return to TCP.
1914 if (!(flags & MSG_SOCALLBCK) &&
1915 (pr->pr_flags & PR_WANTRCVD)) {
1916 SOCKBUF_UNLOCK(&so->so_rcv);
1918 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1919 SOCKBUF_LOCK(&so->so_rcv);
1922 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1923 if (orig_resid == uio->uio_resid && orig_resid &&
1924 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1925 SOCKBUF_UNLOCK(&so->so_rcv);
1928 SOCKBUF_UNLOCK(&so->so_rcv);
1933 sbunlock(&so->so_rcv);
1938 * Optimized version of soreceive() for stream (TCP) sockets.
1939 * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled.
1942 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1943 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1945 int len = 0, error = 0, flags, oresid;
1947 struct mbuf *m, *n = NULL;
1949 /* We only do stream sockets. */
1950 if (so->so_type != SOCK_STREAM)
1954 if (controlp != NULL)
1957 flags = *flagsp &~ MSG_EOR;
1960 if (flags & MSG_OOB)
1961 return (soreceive_rcvoob(so, uio, flags));
1967 /* Prevent other readers from entering the socket. */
1968 error = sblock(sb, SBLOCKWAIT(flags));
1973 /* Easy one, no space to copyout anything. */
1974 if (uio->uio_resid == 0) {
1978 oresid = uio->uio_resid;
1980 /* We will never ever get anything unless we are or were connected. */
1981 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1987 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1989 /* Abort if socket has reported problems. */
1991 if (sbavail(sb) > 0)
1993 if (oresid > uio->uio_resid)
1995 error = so->so_error;
1996 if (!(flags & MSG_PEEK))
2001 /* Door is closed. Deliver what is left, if any. */
2002 if (sb->sb_state & SBS_CANTRCVMORE) {
2003 if (sbavail(sb) > 0)
2009 /* Socket buffer is empty and we shall not block. */
2010 if (sbavail(sb) == 0 &&
2011 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2016 /* Socket buffer got some data that we shall deliver now. */
2017 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2018 ((so->so_state & SS_NBIO) ||
2019 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2020 sbavail(sb) >= sb->sb_lowat ||
2021 sbavail(sb) >= uio->uio_resid ||
2022 sbavail(sb) >= sb->sb_hiwat) ) {
2026 /* On MSG_WAITALL we must wait until all data or error arrives. */
2027 if ((flags & MSG_WAITALL) &&
2028 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2032 * Wait and block until (more) data comes in.
2033 * NB: Drops the sockbuf lock during wait.
2041 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2042 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2043 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2047 uio->uio_td->td_ru.ru_msgrcv++;
2049 /* Fill uio until full or current end of socket buffer is reached. */
2050 len = min(uio->uio_resid, sbavail(sb));
2052 /* Dequeue as many mbufs as possible. */
2053 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2057 m_cat(*mp0, sb->sb_mb);
2059 m != NULL && m->m_len <= len;
2061 KASSERT(!(m->m_flags & M_NOTAVAIL),
2062 ("%s: m %p not available", __func__, m));
2064 uio->uio_resid -= m->m_len;
2070 sb->sb_lastrecord = sb->sb_mb;
2071 if (sb->sb_mb == NULL)
2074 /* Copy the remainder. */
2076 KASSERT(sb->sb_mb != NULL,
2077 ("%s: len > 0 && sb->sb_mb empty", __func__));
2079 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2081 len = 0; /* Don't flush data from sockbuf. */
2083 uio->uio_resid -= len;
2094 /* NB: Must unlock socket buffer as uiomove may sleep. */
2096 error = m_mbuftouio(uio, sb->sb_mb, len);
2101 SBLASTRECORDCHK(sb);
2105 * Remove the delivered data from the socket buffer unless we
2106 * were only peeking.
2108 if (!(flags & MSG_PEEK)) {
2110 sbdrop_locked(sb, len);
2112 /* Notify protocol that we drained some data. */
2113 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2114 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2115 !(flags & MSG_SOCALLBCK))) {
2118 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2124 * For MSG_WAITALL we may have to loop again and wait for
2125 * more data to come in.
2127 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2130 SOCKBUF_LOCK_ASSERT(sb);
2131 SBLASTRECORDCHK(sb);
2139 * Optimized version of soreceive() for simple datagram cases from userspace.
2140 * Unlike in the stream case, we're able to drop a datagram if copyout()
2141 * fails, and because we handle datagrams atomically, we don't need to use a
2142 * sleep lock to prevent I/O interlacing.
2145 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2146 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2148 struct mbuf *m, *m2;
2151 struct protosw *pr = so->so_proto;
2152 struct mbuf *nextrecord;
2156 if (controlp != NULL)
2159 flags = *flagsp &~ MSG_EOR;
2164 * For any complicated cases, fall back to the full
2165 * soreceive_generic().
2167 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2168 return (soreceive_generic(so, psa, uio, mp0, controlp,
2172 * Enforce restrictions on use.
2174 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2175 ("soreceive_dgram: wantrcvd"));
2176 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2177 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2178 ("soreceive_dgram: SBS_RCVATMARK"));
2179 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2180 ("soreceive_dgram: P_CONNREQUIRED"));
2183 * Loop blocking while waiting for a datagram.
2185 SOCKBUF_LOCK(&so->so_rcv);
2186 while ((m = so->so_rcv.sb_mb) == NULL) {
2187 KASSERT(sbavail(&so->so_rcv) == 0,
2188 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2189 sbavail(&so->so_rcv)));
2191 error = so->so_error;
2193 SOCKBUF_UNLOCK(&so->so_rcv);
2196 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2197 uio->uio_resid == 0) {
2198 SOCKBUF_UNLOCK(&so->so_rcv);
2201 if ((so->so_state & SS_NBIO) ||
2202 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2203 SOCKBUF_UNLOCK(&so->so_rcv);
2204 return (EWOULDBLOCK);
2206 SBLASTRECORDCHK(&so->so_rcv);
2207 SBLASTMBUFCHK(&so->so_rcv);
2208 error = sbwait(&so->so_rcv);
2210 SOCKBUF_UNLOCK(&so->so_rcv);
2214 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2217 uio->uio_td->td_ru.ru_msgrcv++;
2218 SBLASTRECORDCHK(&so->so_rcv);
2219 SBLASTMBUFCHK(&so->so_rcv);
2220 nextrecord = m->m_nextpkt;
2221 if (nextrecord == NULL) {
2222 KASSERT(so->so_rcv.sb_lastrecord == m,
2223 ("soreceive_dgram: lastrecord != m"));
2226 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2227 ("soreceive_dgram: m_nextpkt != nextrecord"));
2230 * Pull 'm' and its chain off the front of the packet queue.
2232 so->so_rcv.sb_mb = NULL;
2233 sockbuf_pushsync(&so->so_rcv, nextrecord);
2236 * Walk 'm's chain and free that many bytes from the socket buffer.
2238 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2239 sbfree(&so->so_rcv, m2);
2242 * Do a few last checks before we let go of the lock.
2244 SBLASTRECORDCHK(&so->so_rcv);
2245 SBLASTMBUFCHK(&so->so_rcv);
2246 SOCKBUF_UNLOCK(&so->so_rcv);
2248 if (pr->pr_flags & PR_ADDR) {
2249 KASSERT(m->m_type == MT_SONAME,
2250 ("m->m_type == %d", m->m_type));
2252 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2257 /* XXXRW: Can this happen? */
2262 * Packet to copyout() is now in 'm' and it is disconnected from the
2265 * Process one or more MT_CONTROL mbufs present before any data mbufs
2266 * in the first mbuf chain on the socket buffer. We call into the
2267 * protocol to perform externalization (or freeing if controlp ==
2268 * NULL). In some cases there can be only MT_CONTROL mbufs without
2271 if (m->m_type == MT_CONTROL) {
2272 struct mbuf *cm = NULL, *cmn;
2273 struct mbuf **cme = &cm;
2279 cme = &(*cme)->m_next;
2281 } while (m != NULL && m->m_type == MT_CONTROL);
2282 while (cm != NULL) {
2285 if (pr->pr_domain->dom_externalize != NULL) {
2286 error = (*pr->pr_domain->dom_externalize)
2287 (cm, controlp, flags);
2288 } else if (controlp != NULL)
2292 if (controlp != NULL) {
2293 while (*controlp != NULL)
2294 controlp = &(*controlp)->m_next;
2299 KASSERT(m == NULL || m->m_type == MT_DATA,
2300 ("soreceive_dgram: !data"));
2301 while (m != NULL && uio->uio_resid > 0) {
2302 len = uio->uio_resid;
2305 error = uiomove(mtod(m, char *), (int)len, uio);
2310 if (len == m->m_len)
2327 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2328 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2332 CURVNET_SET(so->so_vnet);
2333 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2340 soshutdown(struct socket *so, int how)
2342 struct protosw *pr = so->so_proto;
2345 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2348 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0)
2351 CURVNET_SET(so->so_vnet);
2352 if (pr->pr_usrreqs->pru_flush != NULL)
2353 (*pr->pr_usrreqs->pru_flush)(so, how);
2356 if (how != SHUT_RD) {
2357 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2358 wakeup(&so->so_timeo);
2362 wakeup(&so->so_timeo);
2368 sorflush(struct socket *so)
2370 struct sockbuf *sb = &so->so_rcv;
2371 struct protosw *pr = so->so_proto;
2377 * In order to avoid calling dom_dispose with the socket buffer mutex
2378 * held, and in order to generally avoid holding the lock for a long
2379 * time, we make a copy of the socket buffer and clear the original
2380 * (except locks, state). The new socket buffer copy won't have
2381 * initialized locks so we can only call routines that won't use or
2382 * assert those locks.
2384 * Dislodge threads currently blocked in receive and wait to acquire
2385 * a lock against other simultaneous readers before clearing the
2386 * socket buffer. Don't let our acquire be interrupted by a signal
2387 * despite any existing socket disposition on interruptable waiting.
2390 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2393 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2394 * and mutex data unchanged.
2397 bzero(&aso, sizeof(aso));
2398 aso.so_pcb = so->so_pcb;
2399 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2400 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2401 bzero(&sb->sb_startzero,
2402 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2407 * Dispose of special rights and flush the copied socket. Don't call
2408 * any unsafe routines (that rely on locks being initialized) on aso.
2410 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2411 (*pr->pr_domain->dom_dispose)(&aso);
2412 sbrelease_internal(&aso.so_rcv, so);
2416 * Wrapper for Socket established helper hook.
2417 * Parameters: socket, context of the hook point, hook id.
2420 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2422 struct socket_hhook_data hhook_data = {
2429 CURVNET_SET(so->so_vnet);
2430 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2433 /* Ugly but needed, since hhooks return void for now */
2434 return (hhook_data.status);
2438 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2439 * additional variant to handle the case where the option value needs to be
2440 * some kind of integer, but not a specific size. In addition to their use
2441 * here, these functions are also called by the protocol-level pr_ctloutput()
2445 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2450 * If the user gives us more than we wanted, we ignore it, but if we
2451 * don't get the minimum length the caller wants, we return EINVAL.
2452 * On success, sopt->sopt_valsize is set to however much we actually
2455 if ((valsize = sopt->sopt_valsize) < minlen)
2458 sopt->sopt_valsize = valsize = len;
2460 if (sopt->sopt_td != NULL)
2461 return (copyin(sopt->sopt_val, buf, valsize));
2463 bcopy(sopt->sopt_val, buf, valsize);
2468 * Kernel version of setsockopt(2).
2470 * XXX: optlen is size_t, not socklen_t
2473 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2476 struct sockopt sopt;
2478 sopt.sopt_level = level;
2479 sopt.sopt_name = optname;
2480 sopt.sopt_dir = SOPT_SET;
2481 sopt.sopt_val = optval;
2482 sopt.sopt_valsize = optlen;
2483 sopt.sopt_td = NULL;
2484 return (sosetopt(so, &sopt));
2488 sosetopt(struct socket *so, struct sockopt *sopt)
2499 CURVNET_SET(so->so_vnet);
2501 if (sopt->sopt_level != SOL_SOCKET) {
2502 if (so->so_proto->pr_ctloutput != NULL) {
2503 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2507 error = ENOPROTOOPT;
2509 switch (sopt->sopt_name) {
2510 case SO_ACCEPTFILTER:
2511 error = do_setopt_accept_filter(so, sopt);
2517 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2522 so->so_linger = l.l_linger;
2524 so->so_options |= SO_LINGER;
2526 so->so_options &= ~SO_LINGER;
2533 case SO_USELOOPBACK:
2543 error = sooptcopyin(sopt, &optval, sizeof optval,
2549 so->so_options |= sopt->sopt_name;
2551 so->so_options &= ~sopt->sopt_name;
2556 error = sooptcopyin(sopt, &optval, sizeof optval,
2561 if (optval < 0 || optval >= rt_numfibs) {
2565 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2566 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2567 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2568 so->so_fibnum = optval;
2573 case SO_USER_COOKIE:
2574 error = sooptcopyin(sopt, &val32, sizeof val32,
2578 so->so_user_cookie = val32;
2585 error = sooptcopyin(sopt, &optval, sizeof optval,
2591 * Values < 1 make no sense for any of these options,
2599 switch (sopt->sopt_name) {
2602 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2603 &so->so_snd : &so->so_rcv, (u_long)optval,
2604 so, curthread) == 0) {
2608 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2609 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2613 * Make sure the low-water is never greater than the
2617 SOCKBUF_LOCK(&so->so_snd);
2618 so->so_snd.sb_lowat =
2619 (optval > so->so_snd.sb_hiwat) ?
2620 so->so_snd.sb_hiwat : optval;
2621 SOCKBUF_UNLOCK(&so->so_snd);
2624 SOCKBUF_LOCK(&so->so_rcv);
2625 so->so_rcv.sb_lowat =
2626 (optval > so->so_rcv.sb_hiwat) ?
2627 so->so_rcv.sb_hiwat : optval;
2628 SOCKBUF_UNLOCK(&so->so_rcv);
2635 #ifdef COMPAT_FREEBSD32
2636 if (SV_CURPROC_FLAG(SV_ILP32)) {
2637 struct timeval32 tv32;
2639 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2641 CP(tv32, tv, tv_sec);
2642 CP(tv32, tv, tv_usec);
2645 error = sooptcopyin(sopt, &tv, sizeof tv,
2649 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2650 tv.tv_usec >= 1000000) {
2654 if (tv.tv_sec > INT32_MAX)
2658 switch (sopt->sopt_name) {
2660 so->so_snd.sb_timeo = val;
2663 so->so_rcv.sb_timeo = val;
2670 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2674 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2682 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2683 error = hhook_run_socket(so, sopt,
2686 error = ENOPROTOOPT;
2689 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2690 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
2698 * Helper routine for getsockopt.
2701 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2709 * Documented get behavior is that we always return a value, possibly
2710 * truncated to fit in the user's buffer. Traditional behavior is
2711 * that we always tell the user precisely how much we copied, rather
2712 * than something useful like the total amount we had available for
2713 * her. Note that this interface is not idempotent; the entire
2714 * answer must be generated ahead of time.
2716 valsize = min(len, sopt->sopt_valsize);
2717 sopt->sopt_valsize = valsize;
2718 if (sopt->sopt_val != NULL) {
2719 if (sopt->sopt_td != NULL)
2720 error = copyout(buf, sopt->sopt_val, valsize);
2722 bcopy(buf, sopt->sopt_val, valsize);
2728 sogetopt(struct socket *so, struct sockopt *sopt)
2737 CURVNET_SET(so->so_vnet);
2739 if (sopt->sopt_level != SOL_SOCKET) {
2740 if (so->so_proto->pr_ctloutput != NULL)
2741 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2743 error = ENOPROTOOPT;
2747 switch (sopt->sopt_name) {
2748 case SO_ACCEPTFILTER:
2749 error = do_getopt_accept_filter(so, sopt);
2754 l.l_onoff = so->so_options & SO_LINGER;
2755 l.l_linger = so->so_linger;
2757 error = sooptcopyout(sopt, &l, sizeof l);
2760 case SO_USELOOPBACK:
2772 optval = so->so_options & sopt->sopt_name;
2774 error = sooptcopyout(sopt, &optval, sizeof optval);
2778 optval = so->so_type;
2782 optval = so->so_proto->pr_protocol;
2787 optval = so->so_error;
2793 optval = so->so_snd.sb_hiwat;
2797 optval = so->so_rcv.sb_hiwat;
2801 optval = so->so_snd.sb_lowat;
2805 optval = so->so_rcv.sb_lowat;
2810 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
2811 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2812 #ifdef COMPAT_FREEBSD32
2813 if (SV_CURPROC_FLAG(SV_ILP32)) {
2814 struct timeval32 tv32;
2816 CP(tv, tv32, tv_sec);
2817 CP(tv, tv32, tv_usec);
2818 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2821 error = sooptcopyout(sopt, &tv, sizeof tv);
2826 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2830 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2834 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2842 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2846 error = mac_getsockopt_peerlabel(
2847 sopt->sopt_td->td_ucred, so, &extmac);
2850 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2856 case SO_LISTENQLIMIT:
2857 optval = so->so_qlimit;
2861 optval = so->so_qlen;
2864 case SO_LISTENINCQLEN:
2865 optval = so->so_incqlen;
2869 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2870 error = hhook_run_socket(so, sopt,
2873 error = ENOPROTOOPT;
2885 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2887 struct mbuf *m, *m_prev;
2888 int sopt_size = sopt->sopt_valsize;
2890 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2893 if (sopt_size > MLEN) {
2894 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
2895 if ((m->m_flags & M_EXT) == 0) {
2899 m->m_len = min(MCLBYTES, sopt_size);
2901 m->m_len = min(MLEN, sopt_size);
2903 sopt_size -= m->m_len;
2908 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2913 if (sopt_size > MLEN) {
2914 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
2916 if ((m->m_flags & M_EXT) == 0) {
2921 m->m_len = min(MCLBYTES, sopt_size);
2923 m->m_len = min(MLEN, sopt_size);
2925 sopt_size -= m->m_len;
2933 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2935 struct mbuf *m0 = m;
2937 if (sopt->sopt_val == NULL)
2939 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2940 if (sopt->sopt_td != NULL) {
2943 error = copyin(sopt->sopt_val, mtod(m, char *),
2950 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2951 sopt->sopt_valsize -= m->m_len;
2952 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2955 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2956 panic("ip6_sooptmcopyin");
2961 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2963 struct mbuf *m0 = m;
2966 if (sopt->sopt_val == NULL)
2968 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2969 if (sopt->sopt_td != NULL) {
2972 error = copyout(mtod(m, char *), sopt->sopt_val,
2979 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2980 sopt->sopt_valsize -= m->m_len;
2981 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2982 valsize += m->m_len;
2986 /* enough soopt buffer should be given from user-land */
2990 sopt->sopt_valsize = valsize;
2995 * sohasoutofband(): protocol notifies socket layer of the arrival of new
2996 * out-of-band data, which will then notify socket consumers.
2999 sohasoutofband(struct socket *so)
3002 if (so->so_sigio != NULL)
3003 pgsigio(&so->so_sigio, SIGURG, 0);
3004 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
3008 sopoll(struct socket *so, int events, struct ucred *active_cred,
3013 * We do not need to set or assert curvnet as long as everyone uses
3016 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3021 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3026 SOCKBUF_LOCK(&so->so_snd);
3027 SOCKBUF_LOCK(&so->so_rcv);
3028 if (events & (POLLIN | POLLRDNORM))
3029 if (soreadabledata(so))
3030 revents |= events & (POLLIN | POLLRDNORM);
3032 if (events & (POLLOUT | POLLWRNORM))
3033 if (sowriteable(so))
3034 revents |= events & (POLLOUT | POLLWRNORM);
3036 if (events & (POLLPRI | POLLRDBAND))
3037 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
3038 revents |= events & (POLLPRI | POLLRDBAND);
3040 if ((events & POLLINIGNEOF) == 0) {
3041 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3042 revents |= events & (POLLIN | POLLRDNORM);
3043 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3049 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3050 selrecord(td, &so->so_rcv.sb_sel);
3051 so->so_rcv.sb_flags |= SB_SEL;
3054 if (events & (POLLOUT | POLLWRNORM)) {
3055 selrecord(td, &so->so_snd.sb_sel);
3056 so->so_snd.sb_flags |= SB_SEL;
3060 SOCKBUF_UNLOCK(&so->so_rcv);
3061 SOCKBUF_UNLOCK(&so->so_snd);
3066 soo_kqfilter(struct file *fp, struct knote *kn)
3068 struct socket *so = kn->kn_fp->f_data;
3071 switch (kn->kn_filter) {
3073 if (so->so_options & SO_ACCEPTCONN)
3074 kn->kn_fop = &solisten_filtops;
3076 kn->kn_fop = &soread_filtops;
3080 kn->kn_fop = &sowrite_filtops;
3088 knlist_add(&sb->sb_sel.si_note, kn, 1);
3089 sb->sb_flags |= SB_KNOTE;
3095 * Some routines that return EOPNOTSUPP for entry points that are not
3096 * supported by a protocol. Fill in as needed.
3099 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3106 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3113 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3120 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3127 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3135 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3142 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3150 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3157 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3158 struct ifnet *ifp, struct thread *td)
3165 pru_disconnect_notsupp(struct socket *so)
3172 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3179 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3186 pru_rcvd_notsupp(struct socket *so, int flags)
3193 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3200 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3201 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3208 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3211 return (EOPNOTSUPP);
3215 * This isn't really a ``null'' operation, but it's the default one and
3216 * doesn't do anything destructive.
3219 pru_sense_null(struct socket *so, struct stat *sb)
3222 sb->st_blksize = so->so_snd.sb_hiwat;
3227 pru_shutdown_notsupp(struct socket *so)
3234 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3241 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3242 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3249 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3250 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3257 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3265 filt_sordetach(struct knote *kn)
3267 struct socket *so = kn->kn_fp->f_data;
3269 SOCKBUF_LOCK(&so->so_rcv);
3270 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3271 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3272 so->so_rcv.sb_flags &= ~SB_KNOTE;
3273 SOCKBUF_UNLOCK(&so->so_rcv);
3278 filt_soread(struct knote *kn, long hint)
3282 so = kn->kn_fp->f_data;
3283 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3285 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3286 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3287 kn->kn_flags |= EV_EOF;
3288 kn->kn_fflags = so->so_error;
3290 } else if (so->so_error) /* temporary udp error */
3293 if (kn->kn_sfflags & NOTE_LOWAT) {
3294 if (kn->kn_data >= kn->kn_sdata)
3297 if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3301 /* This hook returning non-zero indicates an event, not error */
3302 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3306 filt_sowdetach(struct knote *kn)
3308 struct socket *so = kn->kn_fp->f_data;
3310 SOCKBUF_LOCK(&so->so_snd);
3311 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3312 if (knlist_empty(&so->so_snd.sb_sel.si_note))
3313 so->so_snd.sb_flags &= ~SB_KNOTE;
3314 SOCKBUF_UNLOCK(&so->so_snd);
3319 filt_sowrite(struct knote *kn, long hint)
3323 so = kn->kn_fp->f_data;
3324 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3325 kn->kn_data = sbspace(&so->so_snd);
3327 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3329 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3330 kn->kn_flags |= EV_EOF;
3331 kn->kn_fflags = so->so_error;
3333 } else if (so->so_error) /* temporary udp error */
3335 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3336 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3338 else if (kn->kn_sfflags & NOTE_LOWAT)
3339 return (kn->kn_data >= kn->kn_sdata);
3341 return (kn->kn_data >= so->so_snd.sb_lowat);
3346 filt_solisten(struct knote *kn, long hint)
3348 struct socket *so = kn->kn_fp->f_data;
3350 kn->kn_data = so->so_qlen;
3351 return (!TAILQ_EMPTY(&so->so_comp));
3355 socheckuid(struct socket *so, uid_t uid)
3360 if (so->so_cred->cr_uid != uid)
3366 * These functions are used by protocols to notify the socket layer (and its
3367 * consumers) of state changes in the sockets driven by protocol-side events.
3371 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3373 * Normal sequence from the active (originating) side is that
3374 * soisconnecting() is called during processing of connect() call, resulting
3375 * in an eventual call to soisconnected() if/when the connection is
3376 * established. When the connection is torn down soisdisconnecting() is
3377 * called during processing of disconnect() call, and soisdisconnected() is
3378 * called when the connection to the peer is totally severed. The semantics
3379 * of these routines are such that connectionless protocols can call
3380 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3381 * calls when setting up a ``connection'' takes no time.
3383 * From the passive side, a socket is created with two queues of sockets:
3384 * so_incomp for connections in progress and so_comp for connections already
3385 * made and awaiting user acceptance. As a protocol is preparing incoming
3386 * connections, it creates a socket structure queued on so_incomp by calling
3387 * sonewconn(). When the connection is established, soisconnected() is
3388 * called, and transfers the socket structure to so_comp, making it available
3391 * If a socket is closed with sockets on either so_incomp or so_comp, these
3392 * sockets are dropped.
3394 * If higher-level protocols are implemented in the kernel, the wakeups done
3395 * here will sometimes cause software-interrupt process scheduling.
3398 soisconnecting(struct socket *so)
3402 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3403 so->so_state |= SS_ISCONNECTING;
3408 soisconnected(struct socket *so)
3410 struct socket *head;
3416 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3417 so->so_state |= SS_ISCONNECTED;
3419 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3420 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3422 TAILQ_REMOVE(&head->so_incomp, so, so_list);
3424 so->so_qstate &= ~SQ_INCOMP;
3425 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3427 so->so_qstate |= SQ_COMP;
3430 wakeup_one(&head->so_timeo);
3433 soupcall_set(so, SO_RCV,
3434 head->so_accf->so_accept_filter->accf_callback,
3435 head->so_accf->so_accept_filter_arg);
3436 so->so_options &= ~SO_ACCEPTFILTER;
3437 ret = head->so_accf->so_accept_filter->accf_callback(so,
3438 head->so_accf->so_accept_filter_arg, M_NOWAIT);
3439 if (ret == SU_ISCONNECTED)
3440 soupcall_clear(so, SO_RCV);
3442 if (ret == SU_ISCONNECTED)
3449 wakeup(&so->so_timeo);
3455 soisdisconnecting(struct socket *so)
3459 * Note: This code assumes that SOCK_LOCK(so) and
3460 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3462 SOCKBUF_LOCK(&so->so_rcv);
3463 so->so_state &= ~SS_ISCONNECTING;
3464 so->so_state |= SS_ISDISCONNECTING;
3465 socantrcvmore_locked(so);
3466 SOCKBUF_LOCK(&so->so_snd);
3467 socantsendmore_locked(so);
3468 wakeup(&so->so_timeo);
3472 soisdisconnected(struct socket *so)
3476 * Note: This code assumes that SOCK_LOCK(so) and
3477 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3479 SOCKBUF_LOCK(&so->so_rcv);
3480 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3481 so->so_state |= SS_ISDISCONNECTED;
3482 socantrcvmore_locked(so);
3483 SOCKBUF_LOCK(&so->so_snd);
3484 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
3485 socantsendmore_locked(so);
3486 wakeup(&so->so_timeo);
3490 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3493 sodupsockaddr(const struct sockaddr *sa, int mflags)
3495 struct sockaddr *sa2;
3497 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3499 bcopy(sa, sa2, sa->sa_len);
3504 * Register per-socket buffer upcalls.
3507 soupcall_set(struct socket *so, int which,
3508 int (*func)(struct socket *, void *, int), void *arg)
3520 panic("soupcall_set: bad which");
3522 SOCKBUF_LOCK_ASSERT(sb);
3524 /* XXX: accf_http actually wants to do this on purpose. */
3525 KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3527 sb->sb_upcall = func;
3528 sb->sb_upcallarg = arg;
3529 sb->sb_flags |= SB_UPCALL;
3533 soupcall_clear(struct socket *so, int which)
3545 panic("soupcall_clear: bad which");
3547 SOCKBUF_LOCK_ASSERT(sb);
3548 KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3549 sb->sb_upcall = NULL;
3550 sb->sb_upcallarg = NULL;
3551 sb->sb_flags &= ~SB_UPCALL;
3555 * Create an external-format (``xsocket'') structure using the information in
3556 * the kernel-format socket structure pointed to by so. This is done to
3557 * reduce the spew of irrelevant information over this interface, to isolate
3558 * user code from changes in the kernel structure, and potentially to provide
3559 * information-hiding if we decide that some of this information should be
3560 * hidden from users.
3563 sotoxsocket(struct socket *so, struct xsocket *xso)
3566 xso->xso_len = sizeof *xso;
3568 xso->so_type = so->so_type;
3569 xso->so_options = so->so_options;
3570 xso->so_linger = so->so_linger;
3571 xso->so_state = so->so_state;
3572 xso->so_pcb = so->so_pcb;
3573 xso->xso_protocol = so->so_proto->pr_protocol;
3574 xso->xso_family = so->so_proto->pr_domain->dom_family;
3575 xso->so_qlen = so->so_qlen;
3576 xso->so_incqlen = so->so_incqlen;
3577 xso->so_qlimit = so->so_qlimit;
3578 xso->so_timeo = so->so_timeo;
3579 xso->so_error = so->so_error;
3580 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3581 xso->so_oobmark = so->so_oobmark;
3582 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3583 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3584 xso->so_uid = so->so_cred->cr_uid;
3589 * Socket accessor functions to provide external consumers with
3590 * a safe interface to socket state
3595 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *),
3599 TAILQ_FOREACH(so, &so->so_comp, so_list)
3604 so_sockbuf_rcv(struct socket *so)
3607 return (&so->so_rcv);
3611 so_sockbuf_snd(struct socket *so)
3614 return (&so->so_snd);
3618 so_state_get(const struct socket *so)
3621 return (so->so_state);
3625 so_state_set(struct socket *so, int val)
3632 so_options_get(const struct socket *so)
3635 return (so->so_options);
3639 so_options_set(struct socket *so, int val)
3642 so->so_options = val;
3646 so_error_get(const struct socket *so)
3649 return (so->so_error);
3653 so_error_set(struct socket *so, int val)
3660 so_linger_get(const struct socket *so)
3663 return (so->so_linger);
3667 so_linger_set(struct socket *so, int val)
3670 so->so_linger = val;
3674 so_protosw_get(const struct socket *so)
3677 return (so->so_proto);
3681 so_protosw_set(struct socket *so, struct protosw *val)
3688 so_sorwakeup(struct socket *so)
3695 so_sowwakeup(struct socket *so)
3702 so_sorwakeup_locked(struct socket *so)
3705 sorwakeup_locked(so);
3709 so_sowwakeup_locked(struct socket *so)
3712 sowwakeup_locked(so);
3716 so_lock(struct socket *so)
3723 so_unlock(struct socket *so)