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>
138 #include <sys/jail.h>
139 #include <sys/syslog.h>
140 #include <netinet/in.h>
142 #include <net/vnet.h>
144 #include <security/mac/mac_framework.h>
148 #ifdef COMPAT_FREEBSD32
149 #include <sys/mount.h>
150 #include <sys/sysent.h>
151 #include <compat/freebsd32/freebsd32.h>
154 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
157 static void filt_sordetach(struct knote *kn);
158 static int filt_soread(struct knote *kn, long hint);
159 static void filt_sowdetach(struct knote *kn);
160 static int filt_sowrite(struct knote *kn, long hint);
161 static int filt_solisten(struct knote *kn, long hint);
162 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
164 static struct filterops solisten_filtops = {
166 .f_detach = filt_sordetach,
167 .f_event = filt_solisten,
169 static struct filterops soread_filtops = {
171 .f_detach = filt_sordetach,
172 .f_event = filt_soread,
174 static struct filterops sowrite_filtops = {
176 .f_detach = filt_sowdetach,
177 .f_event = filt_sowrite,
180 so_gen_t so_gencnt; /* generation count for sockets */
182 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
183 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
185 #define VNET_SO_ASSERT(so) \
186 VNET_ASSERT(curvnet != NULL, \
187 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
189 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
190 #define V_socket_hhh VNET(socket_hhh)
193 * Limit on the number of connections in the listen queue waiting
195 * NB: The orginal sysctl somaxconn is still available but hidden
196 * to prevent confusion about the actual purpose of this number.
198 static int somaxconn = SOMAXCONN;
201 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
207 error = sysctl_handle_int(oidp, &val, 0, req);
208 if (error || !req->newptr )
211 if (val < 1 || val > USHRT_MAX)
217 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW,
218 0, sizeof(int), sysctl_somaxconn, "I",
219 "Maximum listen socket pending connection accept queue size");
220 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
221 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP,
222 0, sizeof(int), sysctl_somaxconn, "I",
223 "Maximum listen socket pending connection accept queue size (compat)");
225 static int numopensockets;
226 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
227 &numopensockets, 0, "Number of open sockets");
230 * accept_mtx locks down per-socket fields relating to accept queues. See
231 * socketvar.h for an annotation of the protected fields of struct socket.
233 struct mtx accept_mtx;
234 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
237 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
240 static struct mtx so_global_mtx;
241 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
244 * General IPC sysctl name space, used by sockets and a variety of other IPC
247 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
250 * Initialize the socket subsystem and set up the socket
253 static uma_zone_t socket_zone;
257 socket_zone_change(void *tag)
260 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
264 socket_hhook_register(int subtype)
267 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
268 &V_socket_hhh[subtype],
269 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
270 printf("%s: WARNING: unable to register hook\n", __func__);
274 socket_hhook_deregister(int subtype)
277 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
278 printf("%s: WARNING: unable to deregister hook\n", __func__);
282 socket_init(void *tag)
285 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
286 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
287 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
288 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
289 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
290 EVENTHANDLER_PRI_FIRST);
292 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
295 socket_vnet_init(const void *unused __unused)
299 /* We expect a contiguous range */
300 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
301 socket_hhook_register(i);
303 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
304 socket_vnet_init, NULL);
307 socket_vnet_uninit(const void *unused __unused)
311 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
312 socket_hhook_deregister(i);
314 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
315 socket_vnet_uninit, NULL);
318 * Initialise maxsockets. This SYSINIT must be run after
322 init_maxsockets(void *ignored)
325 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
326 maxsockets = imax(maxsockets, maxfiles);
328 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
331 * Sysctl to get and set the maximum global sockets limit. Notify protocols
332 * of the change so that they can update their dependent limits as required.
335 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
337 int error, newmaxsockets;
339 newmaxsockets = maxsockets;
340 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
341 if (error == 0 && req->newptr) {
342 if (newmaxsockets > maxsockets &&
343 newmaxsockets <= maxfiles) {
344 maxsockets = newmaxsockets;
345 EVENTHANDLER_INVOKE(maxsockets_change);
351 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
352 &maxsockets, 0, sysctl_maxsockets, "IU",
353 "Maximum number of sockets avaliable");
356 * Socket operation routines. These routines are called by the routines in
357 * sys_socket.c or from a system process, and implement the semantics of
358 * socket operations by switching out to the protocol specific routines.
362 * Get a socket structure from our zone, and initialize it. Note that it
363 * would probably be better to allocate socket and PCB at the same time, but
364 * I'm not convinced that all the protocols can be easily modified to do
367 * soalloc() returns a socket with a ref count of 0.
369 static struct socket *
370 soalloc(struct vnet *vnet)
374 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
378 if (mac_socket_init(so, M_NOWAIT) != 0) {
379 uma_zfree(socket_zone, so);
383 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
384 uma_zfree(socket_zone, so);
388 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
389 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
390 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
391 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
392 TAILQ_INIT(&so->so_aiojobq);
393 mtx_lock(&so_global_mtx);
394 so->so_gencnt = ++so_gencnt;
397 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
398 __func__, __LINE__, so));
399 vnet->vnet_sockcnt++;
402 mtx_unlock(&so_global_mtx);
405 /* We shouldn't need the so_global_mtx */
406 if (V_socket_hhh[HHOOK_SOCKET_CREATE]->hhh_nhooks > 0) {
407 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE))
408 /* Do we need more comprehensive error returns? */
417 * Free the storage associated with a socket at the socket layer, tear down
418 * locks, labels, etc. All protocol state is assumed already to have been
419 * torn down (and possibly never set up) by the caller.
422 sodealloc(struct socket *so)
425 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
426 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
428 mtx_lock(&so_global_mtx);
429 so->so_gencnt = ++so_gencnt;
430 --numopensockets; /* Could be below, but faster here. */
432 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
433 __func__, __LINE__, so));
434 so->so_vnet->vnet_sockcnt--;
436 mtx_unlock(&so_global_mtx);
437 if (so->so_rcv.sb_hiwat)
438 (void)chgsbsize(so->so_cred->cr_uidinfo,
439 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
440 if (so->so_snd.sb_hiwat)
441 (void)chgsbsize(so->so_cred->cr_uidinfo,
442 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
443 /* remove acccept filter if one is present. */
444 if (so->so_accf != NULL)
445 do_setopt_accept_filter(so, NULL);
447 mac_socket_destroy(so);
449 CURVNET_SET(so->so_vnet);
450 if (V_socket_hhh[HHOOK_SOCKET_CLOSE]->hhh_nhooks > 0)
451 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
455 khelp_destroy_osd(&so->osd);
456 sx_destroy(&so->so_snd.sb_sx);
457 sx_destroy(&so->so_rcv.sb_sx);
458 SOCKBUF_LOCK_DESTROY(&so->so_snd);
459 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
460 uma_zfree(socket_zone, so);
464 * socreate returns a socket with a ref count of 1. The socket should be
465 * closed with soclose().
468 socreate(int dom, struct socket **aso, int type, int proto,
469 struct ucred *cred, struct thread *td)
476 prp = pffindproto(dom, proto, type);
478 prp = pffindtype(dom, type);
481 /* No support for domain. */
482 if (pffinddomain(dom) == NULL)
483 return (EAFNOSUPPORT);
484 /* No support for socket type. */
485 if (proto == 0 && type != 0)
487 return (EPROTONOSUPPORT);
489 if (prp->pr_usrreqs->pru_attach == NULL ||
490 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
491 return (EPROTONOSUPPORT);
493 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
494 return (EPROTONOSUPPORT);
496 if (prp->pr_type != type)
498 so = soalloc(CRED_TO_VNET(cred));
502 TAILQ_INIT(&so->so_incomp);
503 TAILQ_INIT(&so->so_comp);
505 so->so_cred = crhold(cred);
506 if ((prp->pr_domain->dom_family == PF_INET) ||
507 (prp->pr_domain->dom_family == PF_INET6) ||
508 (prp->pr_domain->dom_family == PF_ROUTE))
509 so->so_fibnum = td->td_proc->p_fibnum;
514 mac_socket_create(cred, so);
516 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
517 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
520 * Auto-sizing of socket buffers is managed by the protocols and
521 * the appropriate flags must be set in the pru_attach function.
523 CURVNET_SET(so->so_vnet);
524 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
527 KASSERT(so->so_count == 1, ("socreate: so_count %d",
538 static int regression_sonewconn_earlytest = 1;
539 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
540 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
544 * When an attempt at a new connection is noted on a socket which accepts
545 * connections, sonewconn is called. If the connection is possible (subject
546 * to space constraints, etc.) then we allocate a new structure, propoerly
547 * linked into the data structure of the original socket, and return this.
548 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
550 * Note: the ref count on the socket is 0 on return.
553 sonewconn(struct socket *head, int connstatus)
555 static struct timeval lastover;
556 static struct timeval overinterval = { 60, 0 };
557 static int overcount;
563 over = (head->so_qlen > 3 * head->so_qlimit / 2);
566 if (regression_sonewconn_earlytest && over) {
572 if (ratecheck(&lastover, &overinterval)) {
573 log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
574 "%i already in queue awaiting acceptance "
575 "(%d occurrences)\n",
576 __func__, head->so_pcb, head->so_qlen, overcount);
583 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
584 __func__, __LINE__, head));
585 so = soalloc(head->so_vnet);
587 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
588 "limit reached or out of memory\n",
589 __func__, head->so_pcb);
592 if ((head->so_options & SO_ACCEPTFILTER) != 0)
595 so->so_type = head->so_type;
596 so->so_options = head->so_options &~ SO_ACCEPTCONN;
597 so->so_linger = head->so_linger;
598 so->so_state = head->so_state | SS_NOFDREF;
599 so->so_fibnum = head->so_fibnum;
600 so->so_proto = head->so_proto;
601 so->so_cred = crhold(head->so_cred);
603 mac_socket_newconn(head, so);
605 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
606 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
607 VNET_SO_ASSERT(head);
608 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
610 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
611 __func__, head->so_pcb);
614 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
616 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
617 __func__, head->so_pcb);
620 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
621 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
622 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
623 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
624 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
625 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
626 so->so_state |= connstatus;
629 * The accept socket may be tearing down but we just
630 * won a race on the ACCEPT_LOCK.
631 * However, if sctp_peeloff() is called on a 1-to-many
632 * style socket, the SO_ACCEPTCONN doesn't need to be set.
634 if (!(head->so_options & SO_ACCEPTCONN) &&
635 ((head->so_proto->pr_protocol != IPPROTO_SCTP) ||
636 (head->so_type != SOCK_SEQPACKET))) {
639 sofree(so); /* NB: returns ACCEPT_UNLOCK'ed. */
643 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
644 so->so_qstate |= SQ_COMP;
648 * Keep removing sockets from the head until there's room for
649 * us to insert on the tail. In pre-locking revisions, this
650 * was a simple if(), but as we could be racing with other
651 * threads and soabort() requires dropping locks, we must
652 * loop waiting for the condition to be true.
654 while (head->so_incqlen > head->so_qlimit) {
656 sp = TAILQ_FIRST(&head->so_incomp);
657 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
659 sp->so_qstate &= ~SQ_INCOMP;
665 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
666 so->so_qstate |= SQ_INCOMP;
672 wakeup_one(&head->so_timeo);
678 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
682 CURVNET_SET(so->so_vnet);
683 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
689 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
693 CURVNET_SET(so->so_vnet);
694 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
700 * solisten() transitions a socket from a non-listening state to a listening
701 * state, but can also be used to update the listen queue depth on an
702 * existing listen socket. The protocol will call back into the sockets
703 * layer using solisten_proto_check() and solisten_proto() to check and set
704 * socket-layer listen state. Call backs are used so that the protocol can
705 * acquire both protocol and socket layer locks in whatever order is required
708 * Protocol implementors are advised to hold the socket lock across the
709 * socket-layer test and set to avoid races at the socket layer.
712 solisten(struct socket *so, int backlog, struct thread *td)
716 CURVNET_SET(so->so_vnet);
717 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
723 solisten_proto_check(struct socket *so)
726 SOCK_LOCK_ASSERT(so);
728 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
735 solisten_proto(struct socket *so, int backlog)
738 SOCK_LOCK_ASSERT(so);
740 if (backlog < 0 || backlog > somaxconn)
742 so->so_qlimit = backlog;
743 so->so_options |= SO_ACCEPTCONN;
747 * Evaluate the reference count and named references on a socket; if no
748 * references remain, free it. This should be called whenever a reference is
749 * released, such as in sorele(), but also when named reference flags are
750 * cleared in socket or protocol code.
752 * sofree() will free the socket if:
754 * - There are no outstanding file descriptor references or related consumers
757 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
759 * - The protocol does not have an outstanding strong reference on the socket
762 * - The socket is not in a completed connection queue, so a process has been
763 * notified that it is present. If it is removed, the user process may
764 * block in accept() despite select() saying the socket was ready.
767 sofree(struct socket *so)
769 struct protosw *pr = so->so_proto;
772 ACCEPT_LOCK_ASSERT();
773 SOCK_LOCK_ASSERT(so);
775 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
776 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
784 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
785 (so->so_qstate & SQ_INCOMP) != 0,
786 ("sofree: so_head != NULL, but neither SQ_COMP nor "
788 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
789 (so->so_qstate & SQ_INCOMP) == 0,
790 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
791 TAILQ_REMOVE(&head->so_incomp, so, so_list);
793 so->so_qstate &= ~SQ_INCOMP;
796 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
797 (so->so_qstate & SQ_INCOMP) == 0,
798 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
799 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
800 if (so->so_options & SO_ACCEPTCONN) {
801 KASSERT((TAILQ_EMPTY(&so->so_comp)),
802 ("sofree: so_comp populated"));
803 KASSERT((TAILQ_EMPTY(&so->so_incomp)),
804 ("sofree: so_incomp populated"));
810 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
811 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
812 if (pr->pr_usrreqs->pru_detach != NULL)
813 (*pr->pr_usrreqs->pru_detach)(so);
816 * From this point on, we assume that no other references to this
817 * socket exist anywhere else in the stack. Therefore, no locks need
818 * to be acquired or held.
820 * We used to do a lot of socket buffer and socket locking here, as
821 * well as invoke sorflush() and perform wakeups. The direct call to
822 * dom_dispose() and sbrelease_internal() are an inlining of what was
823 * necessary from sorflush().
825 * Notice that the socket buffer and kqueue state are torn down
826 * before calling pru_detach. This means that protocols shold not
827 * assume they can perform socket wakeups, etc, in their detach code.
829 sbdestroy(&so->so_snd, so);
830 sbdestroy(&so->so_rcv, so);
831 seldrain(&so->so_snd.sb_sel);
832 seldrain(&so->so_rcv.sb_sel);
833 knlist_destroy(&so->so_rcv.sb_sel.si_note);
834 knlist_destroy(&so->so_snd.sb_sel.si_note);
839 * Close a socket on last file table reference removal. Initiate disconnect
840 * if connected. Free socket when disconnect complete.
842 * This function will sorele() the socket. Note that soclose() may be called
843 * prior to the ref count reaching zero. The actual socket structure will
844 * not be freed until the ref count reaches zero.
847 soclose(struct socket *so)
851 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
853 CURVNET_SET(so->so_vnet);
854 funsetown(&so->so_sigio);
855 if (so->so_state & SS_ISCONNECTED) {
856 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
857 error = sodisconnect(so);
859 if (error == ENOTCONN)
864 if (so->so_options & SO_LINGER) {
865 if ((so->so_state & SS_ISDISCONNECTING) &&
866 (so->so_state & SS_NBIO))
868 while (so->so_state & SS_ISCONNECTED) {
869 error = tsleep(&so->so_timeo,
870 PSOCK | PCATCH, "soclos",
879 if (so->so_proto->pr_usrreqs->pru_close != NULL)
880 (*so->so_proto->pr_usrreqs->pru_close)(so);
882 if (so->so_options & SO_ACCEPTCONN) {
885 * Prevent new additions to the accept queues due
886 * to ACCEPT_LOCK races while we are draining them.
888 so->so_options &= ~SO_ACCEPTCONN;
889 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
890 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
892 sp->so_qstate &= ~SQ_INCOMP;
898 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
899 TAILQ_REMOVE(&so->so_comp, sp, so_list);
901 sp->so_qstate &= ~SQ_COMP;
907 KASSERT((TAILQ_EMPTY(&so->so_comp)),
908 ("%s: so_comp populated", __func__));
909 KASSERT((TAILQ_EMPTY(&so->so_incomp)),
910 ("%s: so_incomp populated", __func__));
913 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
914 so->so_state |= SS_NOFDREF;
915 sorele(so); /* NB: Returns with ACCEPT_UNLOCK(). */
921 * soabort() is used to abruptly tear down a connection, such as when a
922 * resource limit is reached (listen queue depth exceeded), or if a listen
923 * socket is closed while there are sockets waiting to be accepted.
925 * This interface is tricky, because it is called on an unreferenced socket,
926 * and must be called only by a thread that has actually removed the socket
927 * from the listen queue it was on, or races with other threads are risked.
929 * This interface will call into the protocol code, so must not be called
930 * with any socket locks held. Protocols do call it while holding their own
931 * recursible protocol mutexes, but this is something that should be subject
932 * to review in the future.
935 soabort(struct socket *so)
939 * In as much as is possible, assert that no references to this
940 * socket are held. This is not quite the same as asserting that the
941 * current thread is responsible for arranging for no references, but
942 * is as close as we can get for now.
944 KASSERT(so->so_count == 0, ("soabort: so_count"));
945 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
946 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
947 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
948 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
951 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
952 (*so->so_proto->pr_usrreqs->pru_abort)(so);
959 soaccept(struct socket *so, struct sockaddr **nam)
964 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
965 so->so_state &= ~SS_NOFDREF;
968 CURVNET_SET(so->so_vnet);
969 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
975 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
978 return (soconnectat(AT_FDCWD, so, nam, td));
982 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
986 if (so->so_options & SO_ACCEPTCONN)
989 CURVNET_SET(so->so_vnet);
991 * If protocol is connection-based, can only connect once.
992 * Otherwise, if connected, try to disconnect first. This allows
993 * user to disconnect by connecting to, e.g., a null address.
995 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
996 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
997 (error = sodisconnect(so)))) {
1001 * Prevent accumulated error from previous connection from
1005 if (fd == AT_FDCWD) {
1006 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1009 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1019 soconnect2(struct socket *so1, struct socket *so2)
1023 CURVNET_SET(so1->so_vnet);
1024 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1030 sodisconnect(struct socket *so)
1034 if ((so->so_state & SS_ISCONNECTED) == 0)
1036 if (so->so_state & SS_ISDISCONNECTING)
1039 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1043 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1046 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1047 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1051 int clen = 0, error, dontroute;
1053 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1054 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1055 ("sosend_dgram: !PR_ATOMIC"));
1058 resid = uio->uio_resid;
1060 resid = top->m_pkthdr.len;
1062 * In theory resid should be unsigned. However, space must be
1063 * signed, as it might be less than 0 if we over-committed, and we
1064 * must use a signed comparison of space and resid. On the other
1065 * hand, a negative resid causes us to loop sending 0-length
1066 * segments to the protocol.
1074 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1076 td->td_ru.ru_msgsnd++;
1077 if (control != NULL)
1078 clen = control->m_len;
1080 SOCKBUF_LOCK(&so->so_snd);
1081 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1082 SOCKBUF_UNLOCK(&so->so_snd);
1087 error = so->so_error;
1089 SOCKBUF_UNLOCK(&so->so_snd);
1092 if ((so->so_state & SS_ISCONNECTED) == 0) {
1094 * `sendto' and `sendmsg' is allowed on a connection-based
1095 * socket if it supports implied connect. Return ENOTCONN if
1096 * not connected and no address is supplied.
1098 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1099 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1100 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1101 !(resid == 0 && clen != 0)) {
1102 SOCKBUF_UNLOCK(&so->so_snd);
1106 } else if (addr == NULL) {
1107 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1110 error = EDESTADDRREQ;
1111 SOCKBUF_UNLOCK(&so->so_snd);
1117 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1118 * problem and need fixing.
1120 space = sbspace(&so->so_snd);
1121 if (flags & MSG_OOB)
1124 SOCKBUF_UNLOCK(&so->so_snd);
1125 if (resid > space) {
1131 if (flags & MSG_EOR)
1132 top->m_flags |= M_EOR;
1135 * Copy the data from userland into a mbuf chain.
1136 * If no data is to be copied in, a single empty mbuf
1139 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1140 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1142 error = EFAULT; /* only possible error */
1145 space -= resid - uio->uio_resid;
1146 resid = uio->uio_resid;
1148 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1150 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1155 so->so_options |= SO_DONTROUTE;
1159 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1160 * of date. We could have recieved a reset packet in an interrupt or
1161 * maybe we slept while doing page faults in uiomove() etc. We could
1162 * probably recheck again inside the locking protection here, but
1163 * there are probably other places that this also happens. We must
1167 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1168 (flags & MSG_OOB) ? PRUS_OOB :
1170 * If the user set MSG_EOF, the protocol understands this flag and
1171 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1173 ((flags & MSG_EOF) &&
1174 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1177 /* If there is more to send set PRUS_MORETOCOME */
1178 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1179 top, addr, control, td);
1182 so->so_options &= ~SO_DONTROUTE;
1191 if (control != NULL)
1197 * Send on a socket. If send must go all at once and message is larger than
1198 * send buffering, then hard error. Lock against other senders. If must go
1199 * all at once and not enough room now, then inform user that this would
1200 * block and do nothing. Otherwise, if nonblocking, send as much as
1201 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1202 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1203 * in mbuf chain must be small enough to send all at once.
1205 * Returns nonzero on error, timeout or signal; callers must check for short
1206 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1210 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1211 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1215 int clen = 0, error, dontroute;
1216 int atomic = sosendallatonce(so) || top;
1219 resid = uio->uio_resid;
1221 resid = top->m_pkthdr.len;
1223 * In theory resid should be unsigned. However, space must be
1224 * signed, as it might be less than 0 if we over-committed, and we
1225 * must use a signed comparison of space and resid. On the other
1226 * hand, a negative resid causes us to loop sending 0-length
1227 * segments to the protocol.
1229 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1230 * type sockets since that's an error.
1232 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1238 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1239 (so->so_proto->pr_flags & PR_ATOMIC);
1241 td->td_ru.ru_msgsnd++;
1242 if (control != NULL)
1243 clen = control->m_len;
1245 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1251 SOCKBUF_LOCK(&so->so_snd);
1252 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1253 SOCKBUF_UNLOCK(&so->so_snd);
1258 error = so->so_error;
1260 SOCKBUF_UNLOCK(&so->so_snd);
1263 if ((so->so_state & SS_ISCONNECTED) == 0) {
1265 * `sendto' and `sendmsg' is allowed on a connection-
1266 * based socket if it supports implied connect.
1267 * Return ENOTCONN if not connected and no address is
1270 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1271 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1272 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1273 !(resid == 0 && clen != 0)) {
1274 SOCKBUF_UNLOCK(&so->so_snd);
1278 } else if (addr == NULL) {
1279 SOCKBUF_UNLOCK(&so->so_snd);
1280 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1283 error = EDESTADDRREQ;
1287 space = sbspace(&so->so_snd);
1288 if (flags & MSG_OOB)
1290 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1291 clen > so->so_snd.sb_hiwat) {
1292 SOCKBUF_UNLOCK(&so->so_snd);
1296 if (space < resid + clen &&
1297 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1298 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1299 SOCKBUF_UNLOCK(&so->so_snd);
1300 error = EWOULDBLOCK;
1303 error = sbwait(&so->so_snd);
1304 SOCKBUF_UNLOCK(&so->so_snd);
1309 SOCKBUF_UNLOCK(&so->so_snd);
1314 if (flags & MSG_EOR)
1315 top->m_flags |= M_EOR;
1318 * Copy the data from userland into a mbuf
1319 * chain. If no data is to be copied in,
1320 * a single empty mbuf is returned.
1322 top = m_uiotombuf(uio, M_WAITOK, space,
1323 (atomic ? max_hdr : 0),
1324 (atomic ? M_PKTHDR : 0) |
1325 ((flags & MSG_EOR) ? M_EOR : 0));
1327 error = EFAULT; /* only possible error */
1330 space -= resid - uio->uio_resid;
1331 resid = uio->uio_resid;
1335 so->so_options |= SO_DONTROUTE;
1339 * XXX all the SBS_CANTSENDMORE checks previously
1340 * done could be out of date. We could have recieved
1341 * a reset packet in an interrupt or maybe we slept
1342 * while doing page faults in uiomove() etc. We
1343 * could probably recheck again inside the locking
1344 * protection here, but there are probably other
1345 * places that this also happens. We must rethink
1349 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1350 (flags & MSG_OOB) ? PRUS_OOB :
1352 * If the user set MSG_EOF, the protocol understands
1353 * this flag and nothing left to send then use
1354 * PRU_SEND_EOF instead of PRU_SEND.
1356 ((flags & MSG_EOF) &&
1357 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1360 /* If there is more to send set PRUS_MORETOCOME. */
1361 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1362 top, addr, control, td);
1365 so->so_options &= ~SO_DONTROUTE;
1373 } while (resid && space > 0);
1377 sbunlock(&so->so_snd);
1381 if (control != NULL)
1387 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1388 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1392 CURVNET_SET(so->so_vnet);
1393 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1394 control, flags, td);
1400 * The part of soreceive() that implements reading non-inline out-of-band
1401 * data from a socket. For more complete comments, see soreceive(), from
1402 * which this code originated.
1404 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1405 * unable to return an mbuf chain to the caller.
1408 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1410 struct protosw *pr = so->so_proto;
1414 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1417 m = m_get(M_WAITOK, MT_DATA);
1418 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1422 error = uiomove(mtod(m, void *),
1423 (int) min(uio->uio_resid, m->m_len), uio);
1425 } while (uio->uio_resid && error == 0 && m);
1433 * Following replacement or removal of the first mbuf on the first mbuf chain
1434 * of a socket buffer, push necessary state changes back into the socket
1435 * buffer so that other consumers see the values consistently. 'nextrecord'
1436 * is the callers locally stored value of the original value of
1437 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1438 * NOTE: 'nextrecord' may be NULL.
1440 static __inline void
1441 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1444 SOCKBUF_LOCK_ASSERT(sb);
1446 * First, update for the new value of nextrecord. If necessary, make
1447 * it the first record.
1449 if (sb->sb_mb != NULL)
1450 sb->sb_mb->m_nextpkt = nextrecord;
1452 sb->sb_mb = nextrecord;
1455 * Now update any dependent socket buffer fields to reflect the new
1456 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1457 * addition of a second clause that takes care of the case where
1458 * sb_mb has been updated, but remains the last record.
1460 if (sb->sb_mb == NULL) {
1461 sb->sb_mbtail = NULL;
1462 sb->sb_lastrecord = NULL;
1463 } else if (sb->sb_mb->m_nextpkt == NULL)
1464 sb->sb_lastrecord = sb->sb_mb;
1468 * Implement receive operations on a socket. We depend on the way that
1469 * records are added to the sockbuf by sbappend. In particular, each record
1470 * (mbufs linked through m_next) must begin with an address if the protocol
1471 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1472 * data, and then zero or more mbufs of data. In order to allow parallelism
1473 * between network receive and copying to user space, as well as avoid
1474 * sleeping with a mutex held, we release the socket buffer mutex during the
1475 * user space copy. Although the sockbuf is locked, new data may still be
1476 * appended, and thus we must maintain consistency of the sockbuf during that
1479 * The caller may receive the data as a single mbuf chain by supplying an
1480 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1481 * the count in uio_resid.
1484 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1485 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1487 struct mbuf *m, **mp;
1488 int flags, error, offset;
1490 struct protosw *pr = so->so_proto;
1491 struct mbuf *nextrecord;
1493 ssize_t orig_resid = uio->uio_resid;
1498 if (controlp != NULL)
1501 flags = *flagsp &~ MSG_EOR;
1504 if (flags & MSG_OOB)
1505 return (soreceive_rcvoob(so, uio, flags));
1508 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1509 && uio->uio_resid) {
1511 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1514 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1519 SOCKBUF_LOCK(&so->so_rcv);
1520 m = so->so_rcv.sb_mb;
1522 * If we have less data than requested, block awaiting more (subject
1523 * to any timeout) if:
1524 * 1. the current count is less than the low water mark, or
1525 * 2. MSG_DONTWAIT is not set
1527 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1528 so->so_rcv.sb_cc < uio->uio_resid) &&
1529 so->so_rcv.sb_cc < so->so_rcv.sb_lowat &&
1530 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1531 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1532 ("receive: m == %p so->so_rcv.sb_cc == %u",
1533 m, so->so_rcv.sb_cc));
1537 error = so->so_error;
1538 if ((flags & MSG_PEEK) == 0)
1540 SOCKBUF_UNLOCK(&so->so_rcv);
1543 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1544 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1546 SOCKBUF_UNLOCK(&so->so_rcv);
1551 for (; m != NULL; m = m->m_next)
1552 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1553 m = so->so_rcv.sb_mb;
1556 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1557 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1558 SOCKBUF_UNLOCK(&so->so_rcv);
1562 if (uio->uio_resid == 0) {
1563 SOCKBUF_UNLOCK(&so->so_rcv);
1566 if ((so->so_state & SS_NBIO) ||
1567 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1568 SOCKBUF_UNLOCK(&so->so_rcv);
1569 error = EWOULDBLOCK;
1572 SBLASTRECORDCHK(&so->so_rcv);
1573 SBLASTMBUFCHK(&so->so_rcv);
1574 error = sbwait(&so->so_rcv);
1575 SOCKBUF_UNLOCK(&so->so_rcv);
1582 * From this point onward, we maintain 'nextrecord' as a cache of the
1583 * pointer to the next record in the socket buffer. We must keep the
1584 * various socket buffer pointers and local stack versions of the
1585 * pointers in sync, pushing out modifications before dropping the
1586 * socket buffer mutex, and re-reading them when picking it up.
1588 * Otherwise, we will race with the network stack appending new data
1589 * or records onto the socket buffer by using inconsistent/stale
1590 * versions of the field, possibly resulting in socket buffer
1593 * By holding the high-level sblock(), we prevent simultaneous
1594 * readers from pulling off the front of the socket buffer.
1596 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1598 uio->uio_td->td_ru.ru_msgrcv++;
1599 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1600 SBLASTRECORDCHK(&so->so_rcv);
1601 SBLASTMBUFCHK(&so->so_rcv);
1602 nextrecord = m->m_nextpkt;
1603 if (pr->pr_flags & PR_ADDR) {
1604 KASSERT(m->m_type == MT_SONAME,
1605 ("m->m_type == %d", m->m_type));
1608 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1610 if (flags & MSG_PEEK) {
1613 sbfree(&so->so_rcv, m);
1614 so->so_rcv.sb_mb = m_free(m);
1615 m = so->so_rcv.sb_mb;
1616 sockbuf_pushsync(&so->so_rcv, nextrecord);
1621 * Process one or more MT_CONTROL mbufs present before any data mbufs
1622 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1623 * just copy the data; if !MSG_PEEK, we call into the protocol to
1624 * perform externalization (or freeing if controlp == NULL).
1626 if (m != NULL && m->m_type == MT_CONTROL) {
1627 struct mbuf *cm = NULL, *cmn;
1628 struct mbuf **cme = &cm;
1631 if (flags & MSG_PEEK) {
1632 if (controlp != NULL) {
1633 *controlp = m_copy(m, 0, m->m_len);
1634 controlp = &(*controlp)->m_next;
1638 sbfree(&so->so_rcv, m);
1639 so->so_rcv.sb_mb = m->m_next;
1642 cme = &(*cme)->m_next;
1643 m = so->so_rcv.sb_mb;
1645 } while (m != NULL && m->m_type == MT_CONTROL);
1646 if ((flags & MSG_PEEK) == 0)
1647 sockbuf_pushsync(&so->so_rcv, nextrecord);
1648 while (cm != NULL) {
1651 if (pr->pr_domain->dom_externalize != NULL) {
1652 SOCKBUF_UNLOCK(&so->so_rcv);
1654 error = (*pr->pr_domain->dom_externalize)
1655 (cm, controlp, flags);
1656 SOCKBUF_LOCK(&so->so_rcv);
1657 } else if (controlp != NULL)
1661 if (controlp != NULL) {
1663 while (*controlp != NULL)
1664 controlp = &(*controlp)->m_next;
1669 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1671 nextrecord = so->so_rcv.sb_mb;
1675 if ((flags & MSG_PEEK) == 0) {
1676 KASSERT(m->m_nextpkt == nextrecord,
1677 ("soreceive: post-control, nextrecord !sync"));
1678 if (nextrecord == NULL) {
1679 KASSERT(so->so_rcv.sb_mb == m,
1680 ("soreceive: post-control, sb_mb!=m"));
1681 KASSERT(so->so_rcv.sb_lastrecord == m,
1682 ("soreceive: post-control, lastrecord!=m"));
1686 if (type == MT_OOBDATA)
1689 if ((flags & MSG_PEEK) == 0) {
1690 KASSERT(so->so_rcv.sb_mb == nextrecord,
1691 ("soreceive: sb_mb != nextrecord"));
1692 if (so->so_rcv.sb_mb == NULL) {
1693 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1694 ("soreceive: sb_lastercord != NULL"));
1698 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1699 SBLASTRECORDCHK(&so->so_rcv);
1700 SBLASTMBUFCHK(&so->so_rcv);
1703 * Now continue to read any data mbufs off of the head of the socket
1704 * buffer until the read request is satisfied. Note that 'type' is
1705 * used to store the type of any mbuf reads that have happened so far
1706 * such that soreceive() can stop reading if the type changes, which
1707 * causes soreceive() to return only one of regular data and inline
1708 * out-of-band data in a single socket receive operation.
1712 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1714 * If the type of mbuf has changed since the last mbuf
1715 * examined ('type'), end the receive operation.
1717 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1718 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1719 if (type != m->m_type)
1721 } else if (type == MT_OOBDATA)
1724 KASSERT(m->m_type == MT_DATA,
1725 ("m->m_type == %d", m->m_type));
1726 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1727 len = uio->uio_resid;
1728 if (so->so_oobmark && len > so->so_oobmark - offset)
1729 len = so->so_oobmark - offset;
1730 if (len > m->m_len - moff)
1731 len = m->m_len - moff;
1733 * If mp is set, just pass back the mbufs. Otherwise copy
1734 * them out via the uio, then free. Sockbuf must be
1735 * consistent here (points to current mbuf, it points to next
1736 * record) when we drop priority; we must note any additions
1737 * to the sockbuf when we block interrupts again.
1740 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1741 SBLASTRECORDCHK(&so->so_rcv);
1742 SBLASTMBUFCHK(&so->so_rcv);
1743 SOCKBUF_UNLOCK(&so->so_rcv);
1744 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1745 SOCKBUF_LOCK(&so->so_rcv);
1748 * The MT_SONAME mbuf has already been removed
1749 * from the record, so it is necessary to
1750 * remove the data mbufs, if any, to preserve
1751 * the invariant in the case of PR_ADDR that
1752 * requires MT_SONAME mbufs at the head of
1755 if (m && pr->pr_flags & PR_ATOMIC &&
1756 ((flags & MSG_PEEK) == 0))
1757 (void)sbdroprecord_locked(&so->so_rcv);
1758 SOCKBUF_UNLOCK(&so->so_rcv);
1762 uio->uio_resid -= len;
1763 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1764 if (len == m->m_len - moff) {
1765 if (m->m_flags & M_EOR)
1767 if (flags & MSG_PEEK) {
1771 nextrecord = m->m_nextpkt;
1772 sbfree(&so->so_rcv, m);
1774 m->m_nextpkt = NULL;
1777 so->so_rcv.sb_mb = m = m->m_next;
1780 so->so_rcv.sb_mb = m_free(m);
1781 m = so->so_rcv.sb_mb;
1783 sockbuf_pushsync(&so->so_rcv, nextrecord);
1784 SBLASTRECORDCHK(&so->so_rcv);
1785 SBLASTMBUFCHK(&so->so_rcv);
1788 if (flags & MSG_PEEK)
1792 if (flags & MSG_DONTWAIT) {
1793 *mp = m_copym(m, 0, len,
1797 * m_copym() couldn't
1799 * Adjust uio_resid back
1801 * down by len bytes,
1802 * which we didn't end
1803 * up "copying" over).
1805 uio->uio_resid += len;
1809 SOCKBUF_UNLOCK(&so->so_rcv);
1810 *mp = m_copym(m, 0, len,
1812 SOCKBUF_LOCK(&so->so_rcv);
1817 so->so_rcv.sb_cc -= len;
1820 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1821 if (so->so_oobmark) {
1822 if ((flags & MSG_PEEK) == 0) {
1823 so->so_oobmark -= len;
1824 if (so->so_oobmark == 0) {
1825 so->so_rcv.sb_state |= SBS_RCVATMARK;
1830 if (offset == so->so_oobmark)
1834 if (flags & MSG_EOR)
1837 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1838 * must not quit until "uio->uio_resid == 0" or an error
1839 * termination. If a signal/timeout occurs, return with a
1840 * short count but without error. Keep sockbuf locked
1841 * against other readers.
1843 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1844 !sosendallatonce(so) && nextrecord == NULL) {
1845 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1847 so->so_rcv.sb_state & SBS_CANTRCVMORE)
1850 * Notify the protocol that some data has been
1851 * drained before blocking.
1853 if (pr->pr_flags & PR_WANTRCVD) {
1854 SOCKBUF_UNLOCK(&so->so_rcv);
1856 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1857 SOCKBUF_LOCK(&so->so_rcv);
1859 SBLASTRECORDCHK(&so->so_rcv);
1860 SBLASTMBUFCHK(&so->so_rcv);
1862 * We could receive some data while was notifying
1863 * the protocol. Skip blocking in this case.
1865 if (so->so_rcv.sb_mb == NULL) {
1866 error = sbwait(&so->so_rcv);
1868 SOCKBUF_UNLOCK(&so->so_rcv);
1872 m = so->so_rcv.sb_mb;
1874 nextrecord = m->m_nextpkt;
1878 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1879 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1881 if ((flags & MSG_PEEK) == 0)
1882 (void) sbdroprecord_locked(&so->so_rcv);
1884 if ((flags & MSG_PEEK) == 0) {
1887 * First part is an inline SB_EMPTY_FIXUP(). Second
1888 * part makes sure sb_lastrecord is up-to-date if
1889 * there is still data in the socket buffer.
1891 so->so_rcv.sb_mb = nextrecord;
1892 if (so->so_rcv.sb_mb == NULL) {
1893 so->so_rcv.sb_mbtail = NULL;
1894 so->so_rcv.sb_lastrecord = NULL;
1895 } else if (nextrecord->m_nextpkt == NULL)
1896 so->so_rcv.sb_lastrecord = nextrecord;
1898 SBLASTRECORDCHK(&so->so_rcv);
1899 SBLASTMBUFCHK(&so->so_rcv);
1901 * If soreceive() is being done from the socket callback,
1902 * then don't need to generate ACK to peer to update window,
1903 * since ACK will be generated on return to TCP.
1905 if (!(flags & MSG_SOCALLBCK) &&
1906 (pr->pr_flags & PR_WANTRCVD)) {
1907 SOCKBUF_UNLOCK(&so->so_rcv);
1909 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1910 SOCKBUF_LOCK(&so->so_rcv);
1913 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1914 if (orig_resid == uio->uio_resid && orig_resid &&
1915 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1916 SOCKBUF_UNLOCK(&so->so_rcv);
1919 SOCKBUF_UNLOCK(&so->so_rcv);
1924 sbunlock(&so->so_rcv);
1929 * Optimized version of soreceive() for stream (TCP) sockets.
1930 * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled.
1933 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1934 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1936 int len = 0, error = 0, flags, oresid;
1938 struct mbuf *m, *n = NULL;
1940 /* We only do stream sockets. */
1941 if (so->so_type != SOCK_STREAM)
1945 if (controlp != NULL)
1948 flags = *flagsp &~ MSG_EOR;
1951 if (flags & MSG_OOB)
1952 return (soreceive_rcvoob(so, uio, flags));
1958 /* Prevent other readers from entering the socket. */
1959 error = sblock(sb, SBLOCKWAIT(flags));
1964 /* Easy one, no space to copyout anything. */
1965 if (uio->uio_resid == 0) {
1969 oresid = uio->uio_resid;
1971 /* We will never ever get anything unless we are or were connected. */
1972 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1978 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1980 /* Abort if socket has reported problems. */
1984 if (oresid > uio->uio_resid)
1986 error = so->so_error;
1987 if (!(flags & MSG_PEEK))
1992 /* Door is closed. Deliver what is left, if any. */
1993 if (sb->sb_state & SBS_CANTRCVMORE) {
2000 /* Socket buffer is empty and we shall not block. */
2001 if (sb->sb_cc == 0 &&
2002 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2007 /* Socket buffer got some data that we shall deliver now. */
2008 if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
2009 ((sb->sb_flags & SS_NBIO) ||
2010 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2011 sb->sb_cc >= sb->sb_lowat ||
2012 sb->sb_cc >= uio->uio_resid ||
2013 sb->sb_cc >= sb->sb_hiwat) ) {
2017 /* On MSG_WAITALL we must wait until all data or error arrives. */
2018 if ((flags & MSG_WAITALL) &&
2019 (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_hiwat))
2023 * Wait and block until (more) data comes in.
2024 * NB: Drops the sockbuf lock during wait.
2032 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2033 KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
2034 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2038 uio->uio_td->td_ru.ru_msgrcv++;
2040 /* Fill uio until full or current end of socket buffer is reached. */
2041 len = min(uio->uio_resid, sb->sb_cc);
2043 /* Dequeue as many mbufs as possible. */
2044 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2048 m_cat(*mp0, sb->sb_mb);
2050 m != NULL && m->m_len <= len;
2053 uio->uio_resid -= m->m_len;
2059 sb->sb_lastrecord = sb->sb_mb;
2060 if (sb->sb_mb == NULL)
2063 /* Copy the remainder. */
2065 KASSERT(sb->sb_mb != NULL,
2066 ("%s: len > 0 && sb->sb_mb empty", __func__));
2068 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2070 len = 0; /* Don't flush data from sockbuf. */
2072 uio->uio_resid -= len;
2083 /* NB: Must unlock socket buffer as uiomove may sleep. */
2085 error = m_mbuftouio(uio, sb->sb_mb, len);
2090 SBLASTRECORDCHK(sb);
2094 * Remove the delivered data from the socket buffer unless we
2095 * were only peeking.
2097 if (!(flags & MSG_PEEK)) {
2099 sbdrop_locked(sb, len);
2101 /* Notify protocol that we drained some data. */
2102 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2103 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2104 !(flags & MSG_SOCALLBCK))) {
2107 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2113 * For MSG_WAITALL we may have to loop again and wait for
2114 * more data to come in.
2116 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2119 SOCKBUF_LOCK_ASSERT(sb);
2120 SBLASTRECORDCHK(sb);
2128 * Optimized version of soreceive() for simple datagram cases from userspace.
2129 * Unlike in the stream case, we're able to drop a datagram if copyout()
2130 * fails, and because we handle datagrams atomically, we don't need to use a
2131 * sleep lock to prevent I/O interlacing.
2134 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2135 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2137 struct mbuf *m, *m2;
2140 struct protosw *pr = so->so_proto;
2141 struct mbuf *nextrecord;
2145 if (controlp != NULL)
2148 flags = *flagsp &~ MSG_EOR;
2153 * For any complicated cases, fall back to the full
2154 * soreceive_generic().
2156 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2157 return (soreceive_generic(so, psa, uio, mp0, controlp,
2161 * Enforce restrictions on use.
2163 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2164 ("soreceive_dgram: wantrcvd"));
2165 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2166 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2167 ("soreceive_dgram: SBS_RCVATMARK"));
2168 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2169 ("soreceive_dgram: P_CONNREQUIRED"));
2172 * Loop blocking while waiting for a datagram.
2174 SOCKBUF_LOCK(&so->so_rcv);
2175 while ((m = so->so_rcv.sb_mb) == NULL) {
2176 KASSERT(so->so_rcv.sb_cc == 0,
2177 ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2180 error = so->so_error;
2182 SOCKBUF_UNLOCK(&so->so_rcv);
2185 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2186 uio->uio_resid == 0) {
2187 SOCKBUF_UNLOCK(&so->so_rcv);
2190 if ((so->so_state & SS_NBIO) ||
2191 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2192 SOCKBUF_UNLOCK(&so->so_rcv);
2193 return (EWOULDBLOCK);
2195 SBLASTRECORDCHK(&so->so_rcv);
2196 SBLASTMBUFCHK(&so->so_rcv);
2197 error = sbwait(&so->so_rcv);
2199 SOCKBUF_UNLOCK(&so->so_rcv);
2203 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2206 uio->uio_td->td_ru.ru_msgrcv++;
2207 SBLASTRECORDCHK(&so->so_rcv);
2208 SBLASTMBUFCHK(&so->so_rcv);
2209 nextrecord = m->m_nextpkt;
2210 if (nextrecord == NULL) {
2211 KASSERT(so->so_rcv.sb_lastrecord == m,
2212 ("soreceive_dgram: lastrecord != m"));
2215 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2216 ("soreceive_dgram: m_nextpkt != nextrecord"));
2219 * Pull 'm' and its chain off the front of the packet queue.
2221 so->so_rcv.sb_mb = NULL;
2222 sockbuf_pushsync(&so->so_rcv, nextrecord);
2225 * Walk 'm's chain and free that many bytes from the socket buffer.
2227 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2228 sbfree(&so->so_rcv, m2);
2231 * Do a few last checks before we let go of the lock.
2233 SBLASTRECORDCHK(&so->so_rcv);
2234 SBLASTMBUFCHK(&so->so_rcv);
2235 SOCKBUF_UNLOCK(&so->so_rcv);
2237 if (pr->pr_flags & PR_ADDR) {
2238 KASSERT(m->m_type == MT_SONAME,
2239 ("m->m_type == %d", m->m_type));
2241 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2246 /* XXXRW: Can this happen? */
2251 * Packet to copyout() is now in 'm' and it is disconnected from the
2254 * Process one or more MT_CONTROL mbufs present before any data mbufs
2255 * in the first mbuf chain on the socket buffer. We call into the
2256 * protocol to perform externalization (or freeing if controlp ==
2259 if (m->m_type == MT_CONTROL) {
2260 struct mbuf *cm = NULL, *cmn;
2261 struct mbuf **cme = &cm;
2267 cme = &(*cme)->m_next;
2269 } while (m != NULL && m->m_type == MT_CONTROL);
2270 while (cm != NULL) {
2273 if (pr->pr_domain->dom_externalize != NULL) {
2274 error = (*pr->pr_domain->dom_externalize)
2275 (cm, controlp, flags);
2276 } else if (controlp != NULL)
2280 if (controlp != NULL) {
2281 while (*controlp != NULL)
2282 controlp = &(*controlp)->m_next;
2287 KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2289 while (m != NULL && uio->uio_resid > 0) {
2290 len = uio->uio_resid;
2293 error = uiomove(mtod(m, char *), (int)len, uio);
2298 if (len == m->m_len)
2314 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2315 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2319 CURVNET_SET(so->so_vnet);
2320 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2327 soshutdown(struct socket *so, int how)
2329 struct protosw *pr = so->so_proto;
2332 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2335 CURVNET_SET(so->so_vnet);
2336 if (pr->pr_usrreqs->pru_flush != NULL)
2337 (*pr->pr_usrreqs->pru_flush)(so, how);
2340 if (how != SHUT_RD) {
2341 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2342 wakeup(&so->so_timeo);
2346 wakeup(&so->so_timeo);
2352 sorflush(struct socket *so)
2354 struct sockbuf *sb = &so->so_rcv;
2355 struct protosw *pr = so->so_proto;
2361 * In order to avoid calling dom_dispose with the socket buffer mutex
2362 * held, and in order to generally avoid holding the lock for a long
2363 * time, we make a copy of the socket buffer and clear the original
2364 * (except locks, state). The new socket buffer copy won't have
2365 * initialized locks so we can only call routines that won't use or
2366 * assert those locks.
2368 * Dislodge threads currently blocked in receive and wait to acquire
2369 * a lock against other simultaneous readers before clearing the
2370 * socket buffer. Don't let our acquire be interrupted by a signal
2371 * despite any existing socket disposition on interruptable waiting.
2374 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2377 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2378 * and mutex data unchanged.
2381 bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2382 bcopy(&sb->sb_startzero, &asb.sb_startzero,
2383 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2384 bzero(&sb->sb_startzero,
2385 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2390 * Dispose of special rights and flush the socket buffer. Don't call
2391 * any unsafe routines (that rely on locks being initialized) on asb.
2393 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2394 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
2395 sbrelease_internal(&asb, so);
2399 * Wrapper for Socket established helper hook.
2400 * Parameters: socket, context of the hook point, hook id.
2403 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2405 struct socket_hhook_data hhook_data = {
2411 hhook_run_hooks(V_socket_hhh[h_id], &hhook_data, &so->osd);
2413 /* Ugly but needed, since hhooks return void for now */
2414 return (hhook_data.status);
2418 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2419 * additional variant to handle the case where the option value needs to be
2420 * some kind of integer, but not a specific size. In addition to their use
2421 * here, these functions are also called by the protocol-level pr_ctloutput()
2425 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2430 * If the user gives us more than we wanted, we ignore it, but if we
2431 * don't get the minimum length the caller wants, we return EINVAL.
2432 * On success, sopt->sopt_valsize is set to however much we actually
2435 if ((valsize = sopt->sopt_valsize) < minlen)
2438 sopt->sopt_valsize = valsize = len;
2440 if (sopt->sopt_td != NULL)
2441 return (copyin(sopt->sopt_val, buf, valsize));
2443 bcopy(sopt->sopt_val, buf, valsize);
2448 * Kernel version of setsockopt(2).
2450 * XXX: optlen is size_t, not socklen_t
2453 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2456 struct sockopt sopt;
2458 sopt.sopt_level = level;
2459 sopt.sopt_name = optname;
2460 sopt.sopt_dir = SOPT_SET;
2461 sopt.sopt_val = optval;
2462 sopt.sopt_valsize = optlen;
2463 sopt.sopt_td = NULL;
2464 return (sosetopt(so, &sopt));
2468 sosetopt(struct socket *so, struct sockopt *sopt)
2479 CURVNET_SET(so->so_vnet);
2481 if (sopt->sopt_level != SOL_SOCKET) {
2482 if (so->so_proto->pr_ctloutput != NULL) {
2483 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2487 error = ENOPROTOOPT;
2489 switch (sopt->sopt_name) {
2490 case SO_ACCEPTFILTER:
2491 error = do_setopt_accept_filter(so, sopt);
2497 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2502 so->so_linger = l.l_linger;
2504 so->so_options |= SO_LINGER;
2506 so->so_options &= ~SO_LINGER;
2513 case SO_USELOOPBACK:
2523 error = sooptcopyin(sopt, &optval, sizeof optval,
2529 so->so_options |= sopt->sopt_name;
2531 so->so_options &= ~sopt->sopt_name;
2536 error = sooptcopyin(sopt, &optval, sizeof optval,
2541 if (optval < 0 || optval >= rt_numfibs) {
2545 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2546 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2547 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2548 so->so_fibnum = optval;
2553 case SO_USER_COOKIE:
2554 error = sooptcopyin(sopt, &val32, sizeof val32,
2558 so->so_user_cookie = val32;
2565 error = sooptcopyin(sopt, &optval, sizeof optval,
2571 * Values < 1 make no sense for any of these options,
2579 switch (sopt->sopt_name) {
2582 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2583 &so->so_snd : &so->so_rcv, (u_long)optval,
2584 so, curthread) == 0) {
2588 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2589 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2593 * Make sure the low-water is never greater than the
2597 SOCKBUF_LOCK(&so->so_snd);
2598 so->so_snd.sb_lowat =
2599 (optval > so->so_snd.sb_hiwat) ?
2600 so->so_snd.sb_hiwat : optval;
2601 SOCKBUF_UNLOCK(&so->so_snd);
2604 SOCKBUF_LOCK(&so->so_rcv);
2605 so->so_rcv.sb_lowat =
2606 (optval > so->so_rcv.sb_hiwat) ?
2607 so->so_rcv.sb_hiwat : optval;
2608 SOCKBUF_UNLOCK(&so->so_rcv);
2615 #ifdef COMPAT_FREEBSD32
2616 if (SV_CURPROC_FLAG(SV_ILP32)) {
2617 struct timeval32 tv32;
2619 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2621 CP(tv32, tv, tv_sec);
2622 CP(tv32, tv, tv_usec);
2625 error = sooptcopyin(sopt, &tv, sizeof tv,
2629 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2630 tv.tv_usec >= 1000000) {
2634 if (tv.tv_sec > INT32_MAX)
2638 switch (sopt->sopt_name) {
2640 so->so_snd.sb_timeo = val;
2643 so->so_rcv.sb_timeo = val;
2650 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2654 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2662 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2663 error = hhook_run_socket(so, sopt,
2666 error = ENOPROTOOPT;
2669 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2670 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
2678 * Helper routine for getsockopt.
2681 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2689 * Documented get behavior is that we always return a value, possibly
2690 * truncated to fit in the user's buffer. Traditional behavior is
2691 * that we always tell the user precisely how much we copied, rather
2692 * than something useful like the total amount we had available for
2693 * her. Note that this interface is not idempotent; the entire
2694 * answer must generated ahead of time.
2696 valsize = min(len, sopt->sopt_valsize);
2697 sopt->sopt_valsize = valsize;
2698 if (sopt->sopt_val != NULL) {
2699 if (sopt->sopt_td != NULL)
2700 error = copyout(buf, sopt->sopt_val, valsize);
2702 bcopy(buf, sopt->sopt_val, valsize);
2708 sogetopt(struct socket *so, struct sockopt *sopt)
2717 CURVNET_SET(so->so_vnet);
2719 if (sopt->sopt_level != SOL_SOCKET) {
2720 if (so->so_proto->pr_ctloutput != NULL)
2721 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2723 error = ENOPROTOOPT;
2727 switch (sopt->sopt_name) {
2728 case SO_ACCEPTFILTER:
2729 error = do_getopt_accept_filter(so, sopt);
2734 l.l_onoff = so->so_options & SO_LINGER;
2735 l.l_linger = so->so_linger;
2737 error = sooptcopyout(sopt, &l, sizeof l);
2740 case SO_USELOOPBACK:
2752 optval = so->so_options & sopt->sopt_name;
2754 error = sooptcopyout(sopt, &optval, sizeof optval);
2758 optval = so->so_type;
2762 optval = so->so_proto->pr_protocol;
2767 optval = so->so_error;
2773 optval = so->so_snd.sb_hiwat;
2777 optval = so->so_rcv.sb_hiwat;
2781 optval = so->so_snd.sb_lowat;
2785 optval = so->so_rcv.sb_lowat;
2790 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
2791 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2792 #ifdef COMPAT_FREEBSD32
2793 if (SV_CURPROC_FLAG(SV_ILP32)) {
2794 struct timeval32 tv32;
2796 CP(tv, tv32, tv_sec);
2797 CP(tv, tv32, tv_usec);
2798 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2801 error = sooptcopyout(sopt, &tv, sizeof tv);
2806 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2810 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2814 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2822 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2826 error = mac_getsockopt_peerlabel(
2827 sopt->sopt_td->td_ucred, so, &extmac);
2830 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2836 case SO_LISTENQLIMIT:
2837 optval = so->so_qlimit;
2841 optval = so->so_qlen;
2844 case SO_LISTENINCQLEN:
2845 optval = so->so_incqlen;
2849 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2850 error = hhook_run_socket(so, sopt,
2853 error = ENOPROTOOPT;
2865 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2867 struct mbuf *m, *m_prev;
2868 int sopt_size = sopt->sopt_valsize;
2870 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2873 if (sopt_size > MLEN) {
2874 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
2875 if ((m->m_flags & M_EXT) == 0) {
2879 m->m_len = min(MCLBYTES, sopt_size);
2881 m->m_len = min(MLEN, sopt_size);
2883 sopt_size -= m->m_len;
2888 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2893 if (sopt_size > MLEN) {
2894 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
2896 if ((m->m_flags & M_EXT) == 0) {
2901 m->m_len = min(MCLBYTES, sopt_size);
2903 m->m_len = min(MLEN, sopt_size);
2905 sopt_size -= m->m_len;
2913 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2915 struct mbuf *m0 = m;
2917 if (sopt->sopt_val == NULL)
2919 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2920 if (sopt->sopt_td != NULL) {
2923 error = copyin(sopt->sopt_val, mtod(m, char *),
2930 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2931 sopt->sopt_valsize -= m->m_len;
2932 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2935 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2936 panic("ip6_sooptmcopyin");
2941 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2943 struct mbuf *m0 = m;
2946 if (sopt->sopt_val == NULL)
2948 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2949 if (sopt->sopt_td != NULL) {
2952 error = copyout(mtod(m, char *), sopt->sopt_val,
2959 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2960 sopt->sopt_valsize -= m->m_len;
2961 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2962 valsize += m->m_len;
2966 /* enough soopt buffer should be given from user-land */
2970 sopt->sopt_valsize = valsize;
2975 * sohasoutofband(): protocol notifies socket layer of the arrival of new
2976 * out-of-band data, which will then notify socket consumers.
2979 sohasoutofband(struct socket *so)
2982 if (so->so_sigio != NULL)
2983 pgsigio(&so->so_sigio, SIGURG, 0);
2984 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2988 sopoll(struct socket *so, int events, struct ucred *active_cred,
2993 * We do not need to set or assert curvnet as long as everyone uses
2996 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3001 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3006 SOCKBUF_LOCK(&so->so_snd);
3007 SOCKBUF_LOCK(&so->so_rcv);
3008 if (events & (POLLIN | POLLRDNORM))
3009 if (soreadabledata(so))
3010 revents |= events & (POLLIN | POLLRDNORM);
3012 if (events & (POLLOUT | POLLWRNORM))
3013 if (sowriteable(so))
3014 revents |= events & (POLLOUT | POLLWRNORM);
3016 if (events & (POLLPRI | POLLRDBAND))
3017 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
3018 revents |= events & (POLLPRI | POLLRDBAND);
3020 if ((events & POLLINIGNEOF) == 0) {
3021 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3022 revents |= events & (POLLIN | POLLRDNORM);
3023 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3029 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3030 selrecord(td, &so->so_rcv.sb_sel);
3031 so->so_rcv.sb_flags |= SB_SEL;
3034 if (events & (POLLOUT | POLLWRNORM)) {
3035 selrecord(td, &so->so_snd.sb_sel);
3036 so->so_snd.sb_flags |= SB_SEL;
3040 SOCKBUF_UNLOCK(&so->so_rcv);
3041 SOCKBUF_UNLOCK(&so->so_snd);
3046 soo_kqfilter(struct file *fp, struct knote *kn)
3048 struct socket *so = kn->kn_fp->f_data;
3051 switch (kn->kn_filter) {
3053 if (so->so_options & SO_ACCEPTCONN)
3054 kn->kn_fop = &solisten_filtops;
3056 kn->kn_fop = &soread_filtops;
3060 kn->kn_fop = &sowrite_filtops;
3068 knlist_add(&sb->sb_sel.si_note, kn, 1);
3069 sb->sb_flags |= SB_KNOTE;
3075 * Some routines that return EOPNOTSUPP for entry points that are not
3076 * supported by a protocol. Fill in as needed.
3079 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3086 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3093 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3100 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3108 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3115 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3123 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3130 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3131 struct ifnet *ifp, struct thread *td)
3138 pru_disconnect_notsupp(struct socket *so)
3145 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3152 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3159 pru_rcvd_notsupp(struct socket *so, int flags)
3166 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3173 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3174 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3181 * This isn't really a ``null'' operation, but it's the default one and
3182 * doesn't do anything destructive.
3185 pru_sense_null(struct socket *so, struct stat *sb)
3188 sb->st_blksize = so->so_snd.sb_hiwat;
3193 pru_shutdown_notsupp(struct socket *so)
3200 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3207 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3208 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3215 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3216 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3223 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3231 filt_sordetach(struct knote *kn)
3233 struct socket *so = kn->kn_fp->f_data;
3235 SOCKBUF_LOCK(&so->so_rcv);
3236 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3237 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3238 so->so_rcv.sb_flags &= ~SB_KNOTE;
3239 SOCKBUF_UNLOCK(&so->so_rcv);
3244 filt_soread(struct knote *kn, long hint)
3248 so = kn->kn_fp->f_data;
3249 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3251 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3252 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3253 kn->kn_flags |= EV_EOF;
3254 kn->kn_fflags = so->so_error;
3256 } else if (so->so_error) /* temporary udp error */
3259 if (kn->kn_sfflags & NOTE_LOWAT) {
3260 if (kn->kn_data >= kn->kn_sdata)
3263 if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat)
3267 if (V_socket_hhh[HHOOK_FILT_SOREAD]->hhh_nhooks > 0)
3268 /* This hook returning non-zero indicates an event, not error */
3269 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3275 filt_sowdetach(struct knote *kn)
3277 struct socket *so = kn->kn_fp->f_data;
3279 SOCKBUF_LOCK(&so->so_snd);
3280 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3281 if (knlist_empty(&so->so_snd.sb_sel.si_note))
3282 so->so_snd.sb_flags &= ~SB_KNOTE;
3283 SOCKBUF_UNLOCK(&so->so_snd);
3288 filt_sowrite(struct knote *kn, long hint)
3292 so = kn->kn_fp->f_data;
3293 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3294 kn->kn_data = sbspace(&so->so_snd);
3296 if (V_socket_hhh[HHOOK_FILT_SOWRITE]->hhh_nhooks > 0)
3297 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3299 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3300 kn->kn_flags |= EV_EOF;
3301 kn->kn_fflags = so->so_error;
3303 } else if (so->so_error) /* temporary udp error */
3305 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3306 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3308 else if (kn->kn_sfflags & NOTE_LOWAT)
3309 return (kn->kn_data >= kn->kn_sdata);
3311 return (kn->kn_data >= so->so_snd.sb_lowat);
3316 filt_solisten(struct knote *kn, long hint)
3318 struct socket *so = kn->kn_fp->f_data;
3320 kn->kn_data = so->so_qlen;
3321 return (!TAILQ_EMPTY(&so->so_comp));
3325 socheckuid(struct socket *so, uid_t uid)
3330 if (so->so_cred->cr_uid != uid)
3336 * These functions are used by protocols to notify the socket layer (and its
3337 * consumers) of state changes in the sockets driven by protocol-side events.
3341 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3343 * Normal sequence from the active (originating) side is that
3344 * soisconnecting() is called during processing of connect() call, resulting
3345 * in an eventual call to soisconnected() if/when the connection is
3346 * established. When the connection is torn down soisdisconnecting() is
3347 * called during processing of disconnect() call, and soisdisconnected() is
3348 * called when the connection to the peer is totally severed. The semantics
3349 * of these routines are such that connectionless protocols can call
3350 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3351 * calls when setting up a ``connection'' takes no time.
3353 * From the passive side, a socket is created with two queues of sockets:
3354 * so_incomp for connections in progress and so_comp for connections already
3355 * made and awaiting user acceptance. As a protocol is preparing incoming
3356 * connections, it creates a socket structure queued on so_incomp by calling
3357 * sonewconn(). When the connection is established, soisconnected() is
3358 * called, and transfers the socket structure to so_comp, making it available
3361 * If a socket is closed with sockets on either so_incomp or so_comp, these
3362 * sockets are dropped.
3364 * If higher-level protocols are implemented in the kernel, the wakeups done
3365 * here will sometimes cause software-interrupt process scheduling.
3368 soisconnecting(struct socket *so)
3372 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3373 so->so_state |= SS_ISCONNECTING;
3378 soisconnected(struct socket *so)
3380 struct socket *head;
3386 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3387 so->so_state |= SS_ISCONNECTED;
3389 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3390 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3392 TAILQ_REMOVE(&head->so_incomp, so, so_list);
3394 so->so_qstate &= ~SQ_INCOMP;
3395 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3397 so->so_qstate |= SQ_COMP;
3400 wakeup_one(&head->so_timeo);
3403 soupcall_set(so, SO_RCV,
3404 head->so_accf->so_accept_filter->accf_callback,
3405 head->so_accf->so_accept_filter_arg);
3406 so->so_options &= ~SO_ACCEPTFILTER;
3407 ret = head->so_accf->so_accept_filter->accf_callback(so,
3408 head->so_accf->so_accept_filter_arg, M_NOWAIT);
3409 if (ret == SU_ISCONNECTED)
3410 soupcall_clear(so, SO_RCV);
3412 if (ret == SU_ISCONNECTED)
3419 wakeup(&so->so_timeo);
3425 soisdisconnecting(struct socket *so)
3429 * Note: This code assumes that SOCK_LOCK(so) and
3430 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3432 SOCKBUF_LOCK(&so->so_rcv);
3433 so->so_state &= ~SS_ISCONNECTING;
3434 so->so_state |= SS_ISDISCONNECTING;
3435 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3436 sorwakeup_locked(so);
3437 SOCKBUF_LOCK(&so->so_snd);
3438 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3439 sowwakeup_locked(so);
3440 wakeup(&so->so_timeo);
3444 soisdisconnected(struct socket *so)
3448 * Note: This code assumes that SOCK_LOCK(so) and
3449 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3451 SOCKBUF_LOCK(&so->so_rcv);
3452 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3453 so->so_state |= SS_ISDISCONNECTED;
3454 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3455 sorwakeup_locked(so);
3456 SOCKBUF_LOCK(&so->so_snd);
3457 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3458 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3459 sowwakeup_locked(so);
3460 wakeup(&so->so_timeo);
3464 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3467 sodupsockaddr(const struct sockaddr *sa, int mflags)
3469 struct sockaddr *sa2;
3471 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3473 bcopy(sa, sa2, sa->sa_len);
3478 * Register per-socket buffer upcalls.
3481 soupcall_set(struct socket *so, int which,
3482 int (*func)(struct socket *, void *, int), void *arg)
3494 panic("soupcall_set: bad which");
3496 SOCKBUF_LOCK_ASSERT(sb);
3498 /* XXX: accf_http actually wants to do this on purpose. */
3499 KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3501 sb->sb_upcall = func;
3502 sb->sb_upcallarg = arg;
3503 sb->sb_flags |= SB_UPCALL;
3507 soupcall_clear(struct socket *so, int which)
3519 panic("soupcall_clear: bad which");
3521 SOCKBUF_LOCK_ASSERT(sb);
3522 KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3523 sb->sb_upcall = NULL;
3524 sb->sb_upcallarg = NULL;
3525 sb->sb_flags &= ~SB_UPCALL;
3529 * Create an external-format (``xsocket'') structure using the information in
3530 * the kernel-format socket structure pointed to by so. This is done to
3531 * reduce the spew of irrelevant information over this interface, to isolate
3532 * user code from changes in the kernel structure, and potentially to provide
3533 * information-hiding if we decide that some of this information should be
3534 * hidden from users.
3537 sotoxsocket(struct socket *so, struct xsocket *xso)
3540 xso->xso_len = sizeof *xso;
3542 xso->so_type = so->so_type;
3543 xso->so_options = so->so_options;
3544 xso->so_linger = so->so_linger;
3545 xso->so_state = so->so_state;
3546 xso->so_pcb = so->so_pcb;
3547 xso->xso_protocol = so->so_proto->pr_protocol;
3548 xso->xso_family = so->so_proto->pr_domain->dom_family;
3549 xso->so_qlen = so->so_qlen;
3550 xso->so_incqlen = so->so_incqlen;
3551 xso->so_qlimit = so->so_qlimit;
3552 xso->so_timeo = so->so_timeo;
3553 xso->so_error = so->so_error;
3554 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3555 xso->so_oobmark = so->so_oobmark;
3556 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3557 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3558 xso->so_uid = so->so_cred->cr_uid;
3563 * Socket accessor functions to provide external consumers with
3564 * a safe interface to socket state
3569 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *),
3573 TAILQ_FOREACH(so, &so->so_comp, so_list)
3578 so_sockbuf_rcv(struct socket *so)
3581 return (&so->so_rcv);
3585 so_sockbuf_snd(struct socket *so)
3588 return (&so->so_snd);
3592 so_state_get(const struct socket *so)
3595 return (so->so_state);
3599 so_state_set(struct socket *so, int val)
3606 so_options_get(const struct socket *so)
3609 return (so->so_options);
3613 so_options_set(struct socket *so, int val)
3616 so->so_options = val;
3620 so_error_get(const struct socket *so)
3623 return (so->so_error);
3627 so_error_set(struct socket *so, int val)
3634 so_linger_get(const struct socket *so)
3637 return (so->so_linger);
3641 so_linger_set(struct socket *so, int val)
3644 so->so_linger = val;
3648 so_protosw_get(const struct socket *so)
3651 return (so->so_proto);
3655 so_protosw_set(struct socket *so, struct protosw *val)
3662 so_sorwakeup(struct socket *so)
3669 so_sowwakeup(struct socket *so)
3676 so_sorwakeup_locked(struct socket *so)
3679 sorwakeup_locked(so);
3683 so_sowwakeup_locked(struct socket *so)
3686 sowwakeup_locked(so);
3690 so_lock(struct socket *so)
3697 so_unlock(struct socket *so)