2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993
5 * The Regents of the University of California.
6 * Copyright (c) 2004 The FreeBSD Foundation
7 * Copyright (c) 2004-2008 Robert N. M. Watson
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
38 * Comments on the socket life cycle:
40 * soalloc() sets of socket layer state for a socket, called only by
41 * socreate() and sonewconn(). Socket layer private.
43 * sodealloc() tears down socket layer state for a socket, called only by
44 * sofree() and sonewconn(). Socket layer private.
46 * pru_attach() associates protocol layer state with an allocated socket;
47 * called only once, may fail, aborting socket allocation. This is called
48 * from socreate() and sonewconn(). Socket layer private.
50 * pru_detach() disassociates protocol layer state from an attached socket,
51 * and will be called exactly once for sockets in which pru_attach() has
52 * been successfully called. If pru_attach() returned an error,
53 * pru_detach() will not be called. Socket layer private.
55 * pru_abort() and pru_close() notify the protocol layer that the last
56 * consumer of a socket is starting to tear down the socket, and that the
57 * protocol should terminate the connection. Historically, pru_abort() also
58 * detached protocol state from the socket state, but this is no longer the
61 * socreate() creates a socket and attaches protocol state. This is a public
62 * interface that may be used by socket layer consumers to create new
65 * sonewconn() creates a socket and attaches protocol state. This is a
66 * public interface that may be used by protocols to create new sockets when
67 * a new connection is received and will be available for accept() on a
70 * soclose() destroys a socket after possibly waiting for it to disconnect.
71 * This is a public interface that socket consumers should use to close and
72 * release a socket when done with it.
74 * soabort() destroys a socket without waiting for it to disconnect (used
75 * only for incoming connections that are already partially or fully
76 * connected). This is used internally by the socket layer when clearing
77 * listen socket queues (due to overflow or close on the listen socket), but
78 * is also a public interface protocols may use to abort connections in
79 * their incomplete listen queues should they no longer be required. Sockets
80 * placed in completed connection listen queues should not be aborted for
81 * reasons described in the comment above the soclose() implementation. This
82 * is not a general purpose close routine, and except in the specific
83 * circumstances described here, should not be used.
85 * sofree() will free a socket and its protocol state if all references on
86 * the socket have been released, and is the public interface to attempt to
87 * free a socket when a reference is removed. This is a socket layer private
90 * NOTE: In addition to socreate() and soclose(), which provide a single
91 * socket reference to the consumer to be managed as required, there are two
92 * calls to explicitly manage socket references, soref(), and sorele().
93 * Currently, these are generally required only when transitioning a socket
94 * from a listen queue to a file descriptor, in order to prevent garbage
95 * collection of the socket at an untimely moment. For a number of reasons,
96 * these interfaces are not preferred, and should be avoided.
98 * NOTE: With regard to VNETs the general rule is that callers do not set
99 * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
100 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
101 * and sorflush(), which are usually called from a pre-set VNET context.
102 * sopoll() currently does not need a VNET context to be set.
105 #include <sys/cdefs.h>
106 #include "opt_inet.h"
107 #include "opt_inet6.h"
108 #include "opt_kern_tls.h"
109 #include "opt_sctp.h"
111 #include <sys/param.h>
112 #include <sys/systm.h>
113 #include <sys/fcntl.h>
114 #include <sys/limits.h>
115 #include <sys/lock.h>
117 #include <sys/malloc.h>
118 #include <sys/mbuf.h>
119 #include <sys/mutex.h>
120 #include <sys/domain.h>
121 #include <sys/file.h> /* for struct knote */
122 #include <sys/hhook.h>
123 #include <sys/kernel.h>
124 #include <sys/khelp.h>
125 #include <sys/ktls.h>
126 #include <sys/event.h>
127 #include <sys/eventhandler.h>
128 #include <sys/poll.h>
129 #include <sys/priv.h>
130 #include <sys/proc.h>
131 #include <sys/protosw.h>
132 #include <sys/sbuf.h>
133 #include <sys/socket.h>
134 #include <sys/socketvar.h>
135 #include <sys/resourcevar.h>
136 #include <net/route.h>
137 #include <sys/signalvar.h>
138 #include <sys/stat.h>
140 #include <sys/sysctl.h>
141 #include <sys/taskqueue.h>
144 #include <sys/unpcb.h>
145 #include <sys/jail.h>
146 #include <sys/syslog.h>
147 #include <netinet/in.h>
148 #include <netinet/in_pcb.h>
149 #include <netinet/tcp.h>
151 #include <net/vnet.h>
153 #include <security/mac/mac_framework.h>
157 #ifdef COMPAT_FREEBSD32
158 #include <sys/mount.h>
159 #include <sys/sysent.h>
160 #include <compat/freebsd32/freebsd32.h>
163 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
165 static void so_rdknl_lock(void *);
166 static void so_rdknl_unlock(void *);
167 static void so_rdknl_assert_lock(void *, int);
168 static void so_wrknl_lock(void *);
169 static void so_wrknl_unlock(void *);
170 static void so_wrknl_assert_lock(void *, int);
172 static void filt_sordetach(struct knote *kn);
173 static int filt_soread(struct knote *kn, long hint);
174 static void filt_sowdetach(struct knote *kn);
175 static int filt_sowrite(struct knote *kn, long hint);
176 static int filt_soempty(struct knote *kn, long hint);
177 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
178 fo_kqfilter_t soo_kqfilter;
180 static struct filterops soread_filtops = {
182 .f_detach = filt_sordetach,
183 .f_event = filt_soread,
185 static struct filterops sowrite_filtops = {
187 .f_detach = filt_sowdetach,
188 .f_event = filt_sowrite,
190 static struct filterops soempty_filtops = {
192 .f_detach = filt_sowdetach,
193 .f_event = filt_soempty,
196 so_gen_t so_gencnt; /* generation count for sockets */
198 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
199 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
201 #define VNET_SO_ASSERT(so) \
202 VNET_ASSERT(curvnet != NULL, \
203 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
205 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
206 #define V_socket_hhh VNET(socket_hhh)
209 * Limit on the number of connections in the listen queue waiting
211 * NB: The original sysctl somaxconn is still available but hidden
212 * to prevent confusion about the actual purpose of this number.
214 static u_int somaxconn = SOMAXCONN;
217 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
223 error = sysctl_handle_int(oidp, &val, 0, req);
224 if (error || !req->newptr )
228 * The purpose of the UINT_MAX / 3 limit, is so that the formula
230 * below, will not overflow.
233 if (val < 1 || val > UINT_MAX / 3)
239 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
240 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
241 sysctl_somaxconn, "I",
242 "Maximum listen socket pending connection accept queue size");
243 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
244 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, 0,
245 sizeof(int), sysctl_somaxconn, "I",
246 "Maximum listen socket pending connection accept queue size (compat)");
248 static int numopensockets;
249 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
250 &numopensockets, 0, "Number of open sockets");
253 * accept_mtx locks down per-socket fields relating to accept queues. See
254 * socketvar.h for an annotation of the protected fields of struct socket.
256 struct mtx accept_mtx;
257 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
260 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
263 static struct mtx so_global_mtx;
264 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
267 * General IPC sysctl name space, used by sockets and a variety of other IPC
270 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
274 * Initialize the socket subsystem and set up the socket
277 static uma_zone_t socket_zone;
281 socket_zone_change(void *tag)
284 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
288 socket_hhook_register(int subtype)
291 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
292 &V_socket_hhh[subtype],
293 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
294 printf("%s: WARNING: unable to register hook\n", __func__);
298 socket_hhook_deregister(int subtype)
301 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
302 printf("%s: WARNING: unable to deregister hook\n", __func__);
306 socket_init(void *tag)
309 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
310 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
311 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
312 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
313 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
314 EVENTHANDLER_PRI_FIRST);
316 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
319 socket_vnet_init(const void *unused __unused)
323 /* We expect a contiguous range */
324 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
325 socket_hhook_register(i);
327 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
328 socket_vnet_init, NULL);
331 socket_vnet_uninit(const void *unused __unused)
335 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
336 socket_hhook_deregister(i);
338 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
339 socket_vnet_uninit, NULL);
342 * Initialise maxsockets. This SYSINIT must be run after
346 init_maxsockets(void *ignored)
349 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
350 maxsockets = imax(maxsockets, maxfiles);
352 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
355 * Sysctl to get and set the maximum global sockets limit. Notify protocols
356 * of the change so that they can update their dependent limits as required.
359 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
361 int error, newmaxsockets;
363 newmaxsockets = maxsockets;
364 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
365 if (error == 0 && req->newptr && newmaxsockets != maxsockets) {
366 if (newmaxsockets > maxsockets &&
367 newmaxsockets <= maxfiles) {
368 maxsockets = newmaxsockets;
369 EVENTHANDLER_INVOKE(maxsockets_change);
375 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
376 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &maxsockets, 0,
377 sysctl_maxsockets, "IU",
378 "Maximum number of sockets available");
381 * Socket operation routines. These routines are called by the routines in
382 * sys_socket.c or from a system process, and implement the semantics of
383 * socket operations by switching out to the protocol specific routines.
387 * Get a socket structure from our zone, and initialize it. Note that it
388 * would probably be better to allocate socket and PCB at the same time, but
389 * I'm not convinced that all the protocols can be easily modified to do
392 * soalloc() returns a socket with a ref count of 0.
394 static struct socket *
395 soalloc(struct vnet *vnet)
399 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
403 if (mac_socket_init(so, M_NOWAIT) != 0) {
404 uma_zfree(socket_zone, so);
408 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
409 uma_zfree(socket_zone, so);
414 * The socket locking protocol allows to lock 2 sockets at a time,
415 * however, the first one must be a listening socket. WITNESS lacks
416 * a feature to change class of an existing lock, so we use DUPOK.
418 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
419 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
420 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
421 so->so_rcv.sb_sel = &so->so_rdsel;
422 so->so_snd.sb_sel = &so->so_wrsel;
423 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
424 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
425 TAILQ_INIT(&so->so_snd.sb_aiojobq);
426 TAILQ_INIT(&so->so_rcv.sb_aiojobq);
427 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
428 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
430 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
431 __func__, __LINE__, so));
434 /* We shouldn't need the so_global_mtx */
435 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
436 /* Do we need more comprehensive error returns? */
437 uma_zfree(socket_zone, so);
440 mtx_lock(&so_global_mtx);
441 so->so_gencnt = ++so_gencnt;
444 vnet->vnet_sockcnt++;
446 mtx_unlock(&so_global_mtx);
452 * Free the storage associated with a socket at the socket layer, tear down
453 * locks, labels, etc. All protocol state is assumed already to have been
454 * torn down (and possibly never set up) by the caller.
457 sodealloc(struct socket *so)
460 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
461 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
463 mtx_lock(&so_global_mtx);
464 so->so_gencnt = ++so_gencnt;
465 --numopensockets; /* Could be below, but faster here. */
467 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
468 __func__, __LINE__, so));
469 so->so_vnet->vnet_sockcnt--;
471 mtx_unlock(&so_global_mtx);
473 mac_socket_destroy(so);
475 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
477 khelp_destroy_osd(&so->osd);
478 if (SOLISTENING(so)) {
479 if (so->sol_accept_filter != NULL)
480 accept_filt_setopt(so, NULL);
482 if (so->so_rcv.sb_hiwat)
483 (void)chgsbsize(so->so_cred->cr_uidinfo,
484 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
485 if (so->so_snd.sb_hiwat)
486 (void)chgsbsize(so->so_cred->cr_uidinfo,
487 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
488 sx_destroy(&so->so_snd.sb_sx);
489 sx_destroy(&so->so_rcv.sb_sx);
490 SOCKBUF_LOCK_DESTROY(&so->so_snd);
491 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
494 mtx_destroy(&so->so_lock);
495 uma_zfree(socket_zone, so);
499 * socreate returns a socket with a ref count of 1. The socket should be
500 * closed with soclose().
503 socreate(int dom, struct socket **aso, int type, int proto,
504 struct ucred *cred, struct thread *td)
511 prp = pffindproto(dom, proto, type);
513 prp = pffindtype(dom, type);
516 /* No support for domain. */
517 if (pffinddomain(dom) == NULL)
518 return (EAFNOSUPPORT);
519 /* No support for socket type. */
520 if (proto == 0 && type != 0)
522 return (EPROTONOSUPPORT);
524 if (prp->pr_usrreqs->pru_attach == NULL ||
525 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
526 return (EPROTONOSUPPORT);
528 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
529 return (EPROTONOSUPPORT);
531 if (prp->pr_type != type)
533 so = soalloc(CRED_TO_VNET(cred));
538 so->so_cred = crhold(cred);
539 if ((prp->pr_domain->dom_family == PF_INET) ||
540 (prp->pr_domain->dom_family == PF_INET6) ||
541 (prp->pr_domain->dom_family == PF_ROUTE))
542 so->so_fibnum = td->td_proc->p_fibnum;
547 mac_socket_create(cred, so);
549 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
550 so_rdknl_assert_lock);
551 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
552 so_wrknl_assert_lock);
554 * Auto-sizing of socket buffers is managed by the protocols and
555 * the appropriate flags must be set in the pru_attach function.
557 CURVNET_SET(so->so_vnet);
558 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
570 static int regression_sonewconn_earlytest = 1;
571 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
572 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
575 static int sooverprio = LOG_DEBUG;
576 SYSCTL_INT(_kern_ipc, OID_AUTO, sooverprio, CTLFLAG_RW,
577 &sooverprio, 0, "Log priority for listen socket overflows: 0..7 or -1 to disable");
579 static struct timeval overinterval = { 60, 0 };
580 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
582 "Delay in seconds between warnings for listen socket overflows");
585 * When an attempt at a new connection is noted on a socket which accepts
586 * connections, sonewconn is called. If the connection is possible (subject
587 * to space constraints, etc.) then we allocate a new structure, properly
588 * linked into the data structure of the original socket, and return this.
589 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
591 * Note: the ref count on the socket is 0 on return.
594 sonewconn(struct socket *head, int connstatus)
600 const char localprefix[] = "local:";
601 char descrbuf[SUNPATHLEN + sizeof(localprefix)];
603 char addrbuf[INET6_ADDRSTRLEN];
605 char addrbuf[INET_ADDRSTRLEN];
610 over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
612 if (regression_sonewconn_earlytest && over) {
616 head->sol_overcount++;
617 dolog = (sooverprio >= 0) &&
618 !!ratecheck(&head->sol_lastover, &overinterval);
621 * If we're going to log, copy the overflow count and queue
622 * length from the listen socket before dropping the lock.
623 * Also, reset the overflow count.
626 overcount = head->sol_overcount;
627 head->sol_overcount = 0;
628 qlen = head->sol_qlen;
630 SOLISTEN_UNLOCK(head);
634 * Try to print something descriptive about the
635 * socket for the error message.
637 sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
639 switch (head->so_proto->pr_domain->dom_family) {
640 #if defined(INET) || defined(INET6)
646 if (head->so_proto->pr_domain->dom_family ==
648 (sotoinpcb(head)->inp_inc.inc_flags &
651 &sotoinpcb(head)->inp_inc.inc6_laddr);
652 sbuf_printf(&descrsb, "[%s]", addrbuf);
658 sotoinpcb(head)->inp_inc.inc_laddr,
660 sbuf_cat(&descrsb, addrbuf);
663 sbuf_printf(&descrsb, ":%hu (proto %u)",
664 ntohs(sotoinpcb(head)->inp_inc.inc_lport),
665 head->so_proto->pr_protocol);
667 #endif /* INET || INET6 */
669 sbuf_cat(&descrsb, localprefix);
670 if (sotounpcb(head)->unp_addr != NULL)
672 sotounpcb(head)->unp_addr->sun_len -
673 offsetof(struct sockaddr_un,
679 sotounpcb(head)->unp_addr->sun_path,
682 sbuf_cat(&descrsb, "(unknown)");
687 * If we can't print something more specific, at least
688 * print the domain name.
690 if (sbuf_finish(&descrsb) != 0 ||
691 sbuf_len(&descrsb) <= 0) {
692 sbuf_clear(&descrsb);
694 head->so_proto->pr_domain->dom_name ?:
696 sbuf_finish(&descrsb);
698 KASSERT(sbuf_len(&descrsb) > 0,
699 ("%s: sbuf creation failed", __func__));
700 log(LOG_PRI(sooverprio),
701 "%s: pcb %p (%s): Listen queue overflow: "
702 "%i already in queue awaiting acceptance "
703 "(%d occurrences)\n",
704 __func__, head->so_pcb, sbuf_data(&descrsb),
706 sbuf_delete(&descrsb);
713 SOLISTEN_UNLOCK(head);
714 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
716 so = soalloc(head->so_vnet);
718 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
719 "limit reached or out of memory\n",
720 __func__, head->so_pcb);
723 so->so_listen = head;
724 so->so_type = head->so_type;
725 so->so_options = head->so_options & ~SO_ACCEPTCONN;
726 so->so_linger = head->so_linger;
727 so->so_state = head->so_state | SS_NOFDREF;
728 so->so_fibnum = head->so_fibnum;
729 so->so_proto = head->so_proto;
730 so->so_cred = crhold(head->so_cred);
732 mac_socket_newconn(head, so);
734 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
735 so_rdknl_assert_lock);
736 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
737 so_wrknl_assert_lock);
738 VNET_SO_ASSERT(head);
739 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
741 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
742 __func__, head->so_pcb);
745 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
747 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
748 __func__, head->so_pcb);
751 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
752 so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
753 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
754 so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
755 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
756 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
759 if (head->sol_accept_filter != NULL)
761 so->so_state |= connstatus;
762 soref(head); /* A socket on (in)complete queue refs head. */
764 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
765 so->so_qstate = SQ_COMP;
767 solisten_wakeup(head); /* unlocks */
770 * Keep removing sockets from the head until there's room for
771 * us to insert on the tail. In pre-locking revisions, this
772 * was a simple if(), but as we could be racing with other
773 * threads and soabort() requires dropping locks, we must
774 * loop waiting for the condition to be true.
776 while (head->sol_incqlen > head->sol_qlimit) {
779 sp = TAILQ_FIRST(&head->sol_incomp);
780 TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
783 sp->so_qstate = SQ_NONE;
784 sp->so_listen = NULL;
786 sorele(head); /* does SOLISTEN_UNLOCK, head stays */
790 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
791 so->so_qstate = SQ_INCOMP;
793 SOLISTEN_UNLOCK(head);
798 #if defined(SCTP) || defined(SCTP_SUPPORT)
800 * Socket part of sctp_peeloff(). Detach a new socket from an
801 * association. The new socket is returned with a reference.
804 sopeeloff(struct socket *head)
808 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
809 __func__, __LINE__, head));
810 so = soalloc(head->so_vnet);
812 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
813 "limit reached or out of memory\n",
814 __func__, head->so_pcb);
817 so->so_type = head->so_type;
818 so->so_options = head->so_options;
819 so->so_linger = head->so_linger;
820 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
821 so->so_fibnum = head->so_fibnum;
822 so->so_proto = head->so_proto;
823 so->so_cred = crhold(head->so_cred);
825 mac_socket_newconn(head, so);
827 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
828 so_rdknl_assert_lock);
829 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
830 so_wrknl_assert_lock);
831 VNET_SO_ASSERT(head);
832 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
834 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
835 __func__, head->so_pcb);
838 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
840 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
841 __func__, head->so_pcb);
844 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
845 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
846 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
847 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
848 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
849 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
858 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
862 CURVNET_SET(so->so_vnet);
863 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
869 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
873 CURVNET_SET(so->so_vnet);
874 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
880 * solisten() transitions a socket from a non-listening state to a listening
881 * state, but can also be used to update the listen queue depth on an
882 * existing listen socket. The protocol will call back into the sockets
883 * layer using solisten_proto_check() and solisten_proto() to check and set
884 * socket-layer listen state. Call backs are used so that the protocol can
885 * acquire both protocol and socket layer locks in whatever order is required
888 * Protocol implementors are advised to hold the socket lock across the
889 * socket-layer test and set to avoid races at the socket layer.
892 solisten(struct socket *so, int backlog, struct thread *td)
896 CURVNET_SET(so->so_vnet);
897 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
903 solisten_proto_check(struct socket *so)
906 SOCK_LOCK_ASSERT(so);
908 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
915 solisten_proto(struct socket *so, int backlog)
917 int sbrcv_lowat, sbsnd_lowat;
918 u_int sbrcv_hiwat, sbsnd_hiwat;
919 short sbrcv_flags, sbsnd_flags;
920 sbintime_t sbrcv_timeo, sbsnd_timeo;
922 SOCK_LOCK_ASSERT(so);
928 * Change this socket to listening state.
930 sbrcv_lowat = so->so_rcv.sb_lowat;
931 sbsnd_lowat = so->so_snd.sb_lowat;
932 sbrcv_hiwat = so->so_rcv.sb_hiwat;
933 sbsnd_hiwat = so->so_snd.sb_hiwat;
934 sbrcv_flags = so->so_rcv.sb_flags;
935 sbsnd_flags = so->so_snd.sb_flags;
936 sbrcv_timeo = so->so_rcv.sb_timeo;
937 sbsnd_timeo = so->so_snd.sb_timeo;
939 sbdestroy(&so->so_snd, so);
940 sbdestroy(&so->so_rcv, so);
941 sx_destroy(&so->so_snd.sb_sx);
942 sx_destroy(&so->so_rcv.sb_sx);
943 SOCKBUF_LOCK_DESTROY(&so->so_snd);
944 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
948 sizeof(struct socket) - offsetof(struct socket, so_rcv));
951 so->sol_sbrcv_lowat = sbrcv_lowat;
952 so->sol_sbsnd_lowat = sbsnd_lowat;
953 so->sol_sbrcv_hiwat = sbrcv_hiwat;
954 so->sol_sbsnd_hiwat = sbsnd_hiwat;
955 so->sol_sbrcv_flags = sbrcv_flags;
956 so->sol_sbsnd_flags = sbsnd_flags;
957 so->sol_sbrcv_timeo = sbrcv_timeo;
958 so->sol_sbsnd_timeo = sbsnd_timeo;
960 so->sol_qlen = so->sol_incqlen = 0;
961 TAILQ_INIT(&so->sol_incomp);
962 TAILQ_INIT(&so->sol_comp);
964 so->sol_accept_filter = NULL;
965 so->sol_accept_filter_arg = NULL;
966 so->sol_accept_filter_str = NULL;
968 so->sol_upcall = NULL;
969 so->sol_upcallarg = NULL;
971 so->so_options |= SO_ACCEPTCONN;
974 if (backlog < 0 || backlog > somaxconn)
976 so->sol_qlimit = backlog;
980 * Wakeup listeners/subsystems once we have a complete connection.
981 * Enters with lock, returns unlocked.
984 solisten_wakeup(struct socket *sol)
987 if (sol->sol_upcall != NULL)
988 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
990 selwakeuppri(&sol->so_rdsel, PSOCK);
991 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
993 SOLISTEN_UNLOCK(sol);
994 wakeup_one(&sol->sol_comp);
995 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
996 pgsigio(&sol->so_sigio, SIGIO, 0);
1000 * Return single connection off a listening socket queue. Main consumer of
1001 * the function is kern_accept4(). Some modules, that do their own accept
1002 * management also use the function.
1004 * Listening socket must be locked on entry and is returned unlocked on
1006 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1009 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1014 SOLISTEN_LOCK_ASSERT(head);
1016 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1017 head->so_error == 0) {
1018 error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH,
1021 SOLISTEN_UNLOCK(head);
1025 if (head->so_error) {
1026 error = head->so_error;
1028 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1029 error = EWOULDBLOCK;
1033 SOLISTEN_UNLOCK(head);
1036 so = TAILQ_FIRST(&head->sol_comp);
1038 KASSERT(so->so_qstate == SQ_COMP,
1039 ("%s: so %p not SQ_COMP", __func__, so));
1042 so->so_qstate = SQ_NONE;
1043 so->so_listen = NULL;
1044 TAILQ_REMOVE(&head->sol_comp, so, so_list);
1045 if (flags & ACCEPT4_INHERIT)
1046 so->so_state |= (head->so_state & SS_NBIO);
1048 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1057 * Evaluate the reference count and named references on a socket; if no
1058 * references remain, free it. This should be called whenever a reference is
1059 * released, such as in sorele(), but also when named reference flags are
1060 * cleared in socket or protocol code.
1062 * sofree() will free the socket if:
1064 * - There are no outstanding file descriptor references or related consumers
1067 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
1069 * - The protocol does not have an outstanding strong reference on the socket
1072 * - The socket is not in a completed connection queue, so a process has been
1073 * notified that it is present. If it is removed, the user process may
1074 * block in accept() despite select() saying the socket was ready.
1077 sofree(struct socket *so)
1079 struct protosw *pr = so->so_proto;
1080 bool last __diagused;
1082 SOCK_LOCK_ASSERT(so);
1084 if ((so->so_state & (SS_NOFDREF | SS_PROTOREF)) != SS_NOFDREF ||
1085 refcount_load(&so->so_count) != 0 || so->so_qstate == SQ_COMP) {
1090 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
1093 sol = so->so_listen;
1094 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
1097 * To solve race between close of a listening socket and
1098 * a socket on its incomplete queue, we need to lock both.
1099 * The order is first listening socket, then regular.
1100 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
1101 * function and the listening socket are the only pointers
1102 * to so. To preserve so and sol, we reference both and then
1104 * After relock the socket may not move to so_comp since it
1105 * doesn't have PCB already, but it may be removed from
1106 * so_incomp. If that happens, we share responsiblity on
1107 * freeing the socket, but soclose() has already removed
1115 if (so->so_qstate == SQ_INCOMP) {
1116 KASSERT(so->so_listen == sol,
1117 ("%s: so %p migrated out of sol %p",
1118 __func__, so, sol));
1119 TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1121 last = refcount_release(&sol->so_count);
1122 KASSERT(!last, ("%s: released last reference for %p",
1124 so->so_qstate = SQ_NONE;
1125 so->so_listen = NULL;
1127 KASSERT(so->so_listen == NULL,
1128 ("%s: so %p not on (in)comp with so_listen",
1131 KASSERT(refcount_load(&so->so_count) == 1,
1132 ("%s: so %p count %u", __func__, so, so->so_count));
1135 if (SOLISTENING(so))
1136 so->so_error = ECONNABORTED;
1139 if (so->so_dtor != NULL)
1143 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1144 (*pr->pr_domain->dom_dispose)(so);
1145 if (pr->pr_usrreqs->pru_detach != NULL)
1146 (*pr->pr_usrreqs->pru_detach)(so);
1149 * From this point on, we assume that no other references to this
1150 * socket exist anywhere else in the stack. Therefore, no locks need
1151 * to be acquired or held.
1153 * We used to do a lot of socket buffer and socket locking here, as
1154 * well as invoke sorflush() and perform wakeups. The direct call to
1155 * dom_dispose() and sbdestroy() are an inlining of what was
1156 * necessary from sorflush().
1158 * Notice that the socket buffer and kqueue state are torn down
1159 * before calling pru_detach. This means that protocols shold not
1160 * assume they can perform socket wakeups, etc, in their detach code.
1162 if (!SOLISTENING(so)) {
1163 sbdestroy(&so->so_snd, so);
1164 sbdestroy(&so->so_rcv, so);
1166 seldrain(&so->so_rdsel);
1167 seldrain(&so->so_wrsel);
1168 knlist_destroy(&so->so_rdsel.si_note);
1169 knlist_destroy(&so->so_wrsel.si_note);
1174 * Close a socket on last file table reference removal. Initiate disconnect
1175 * if connected. Free socket when disconnect complete.
1177 * This function will sorele() the socket. Note that soclose() may be called
1178 * prior to the ref count reaching zero. The actual socket structure will
1179 * not be freed until the ref count reaches zero.
1182 soclose(struct socket *so)
1184 struct accept_queue lqueue;
1185 struct socket *sp, *tsp;
1187 bool last __diagused;
1189 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1191 CURVNET_SET(so->so_vnet);
1192 funsetown(&so->so_sigio);
1193 if (so->so_state & SS_ISCONNECTED) {
1194 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1195 error = sodisconnect(so);
1197 if (error == ENOTCONN)
1203 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1204 if ((so->so_state & SS_ISDISCONNECTING) &&
1205 (so->so_state & SS_NBIO))
1207 while (so->so_state & SS_ISCONNECTED) {
1208 error = tsleep(&so->so_timeo,
1209 PSOCK | PCATCH, "soclos",
1210 so->so_linger * hz);
1218 if (so->so_proto->pr_usrreqs->pru_close != NULL)
1219 (*so->so_proto->pr_usrreqs->pru_close)(so);
1221 TAILQ_INIT(&lqueue);
1223 if (SOLISTENING(so)) {
1224 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1225 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1227 so->sol_qlen = so->sol_incqlen = 0;
1229 TAILQ_FOREACH(sp, &lqueue, so_list) {
1231 sp->so_qstate = SQ_NONE;
1232 sp->so_listen = NULL;
1234 last = refcount_release(&so->so_count);
1235 KASSERT(!last, ("%s: released last reference for %p",
1239 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1240 so->so_state |= SS_NOFDREF;
1242 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1244 if (refcount_load(&sp->so_count) == 0) {
1248 /* See the handling of queued sockets in sofree(). */
1257 * soabort() is used to abruptly tear down a connection, such as when a
1258 * resource limit is reached (listen queue depth exceeded), or if a listen
1259 * socket is closed while there are sockets waiting to be accepted.
1261 * This interface is tricky, because it is called on an unreferenced socket,
1262 * and must be called only by a thread that has actually removed the socket
1263 * from the listen queue it was on, or races with other threads are risked.
1265 * This interface will call into the protocol code, so must not be called
1266 * with any socket locks held. Protocols do call it while holding their own
1267 * recursible protocol mutexes, but this is something that should be subject
1268 * to review in the future.
1271 soabort(struct socket *so)
1275 * In as much as is possible, assert that no references to this
1276 * socket are held. This is not quite the same as asserting that the
1277 * current thread is responsible for arranging for no references, but
1278 * is as close as we can get for now.
1280 KASSERT(so->so_count == 0, ("soabort: so_count"));
1281 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1282 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1285 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1286 (*so->so_proto->pr_usrreqs->pru_abort)(so);
1292 soaccept(struct socket *so, struct sockaddr **nam)
1297 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1298 so->so_state &= ~SS_NOFDREF;
1301 CURVNET_SET(so->so_vnet);
1302 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1308 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1311 return (soconnectat(AT_FDCWD, so, nam, td));
1315 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1320 if (SOLISTENING(so))
1321 return (EOPNOTSUPP);
1323 CURVNET_SET(so->so_vnet);
1325 * If protocol is connection-based, can only connect once.
1326 * Otherwise, if connected, try to disconnect first. This allows
1327 * user to disconnect by connecting to, e.g., a null address.
1329 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1330 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1331 (error = sodisconnect(so)))) {
1335 * Prevent accumulated error from previous connection from
1339 if (fd == AT_FDCWD) {
1340 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1343 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1353 soconnect2(struct socket *so1, struct socket *so2)
1357 CURVNET_SET(so1->so_vnet);
1358 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1364 sodisconnect(struct socket *so)
1368 if ((so->so_state & SS_ISCONNECTED) == 0)
1370 if (so->so_state & SS_ISDISCONNECTING)
1373 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1378 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1379 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1383 int clen = 0, error, dontroute;
1385 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1386 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1387 ("sosend_dgram: !PR_ATOMIC"));
1390 resid = uio->uio_resid;
1392 resid = top->m_pkthdr.len;
1394 * In theory resid should be unsigned. However, space must be
1395 * signed, as it might be less than 0 if we over-committed, and we
1396 * must use a signed comparison of space and resid. On the other
1397 * hand, a negative resid causes us to loop sending 0-length
1398 * segments to the protocol.
1406 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1408 td->td_ru.ru_msgsnd++;
1409 if (control != NULL)
1410 clen = control->m_len;
1412 SOCKBUF_LOCK(&so->so_snd);
1413 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1414 SOCKBUF_UNLOCK(&so->so_snd);
1419 error = so->so_error;
1421 SOCKBUF_UNLOCK(&so->so_snd);
1424 if ((so->so_state & SS_ISCONNECTED) == 0) {
1426 * `sendto' and `sendmsg' is allowed on a connection-based
1427 * socket if it supports implied connect. Return ENOTCONN if
1428 * not connected and no address is supplied.
1430 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1431 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1432 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1433 !(resid == 0 && clen != 0)) {
1434 SOCKBUF_UNLOCK(&so->so_snd);
1438 } else if (addr == NULL) {
1439 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1442 error = EDESTADDRREQ;
1443 SOCKBUF_UNLOCK(&so->so_snd);
1449 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1450 * problem and need fixing.
1452 space = sbspace(&so->so_snd);
1453 if (flags & MSG_OOB)
1456 SOCKBUF_UNLOCK(&so->so_snd);
1457 if (resid > space) {
1463 if (flags & MSG_EOR)
1464 top->m_flags |= M_EOR;
1467 * Copy the data from userland into a mbuf chain.
1468 * If no data is to be copied in, a single empty mbuf
1471 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1472 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1474 error = EFAULT; /* only possible error */
1477 space -= resid - uio->uio_resid;
1478 resid = uio->uio_resid;
1480 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1482 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1487 so->so_options |= SO_DONTROUTE;
1491 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1492 * of date. We could have received a reset packet in an interrupt or
1493 * maybe we slept while doing page faults in uiomove() etc. We could
1494 * probably recheck again inside the locking protection here, but
1495 * there are probably other places that this also happens. We must
1499 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1500 (flags & MSG_OOB) ? PRUS_OOB :
1502 * If the user set MSG_EOF, the protocol understands this flag and
1503 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1505 ((flags & MSG_EOF) &&
1506 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1509 /* If there is more to send set PRUS_MORETOCOME */
1510 (flags & MSG_MORETOCOME) ||
1511 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1512 top, addr, control, td);
1515 so->so_options &= ~SO_DONTROUTE;
1524 if (control != NULL)
1530 * Send on a socket. If send must go all at once and message is larger than
1531 * send buffering, then hard error. Lock against other senders. If must go
1532 * all at once and not enough room now, then inform user that this would
1533 * block and do nothing. Otherwise, if nonblocking, send as much as
1534 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1535 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1536 * in mbuf chain must be small enough to send all at once.
1538 * Returns nonzero on error, timeout or signal; callers must check for short
1539 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1543 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1544 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1548 int clen = 0, error, dontroute;
1549 int atomic = sosendallatonce(so) || top;
1552 struct ktls_session *tls;
1553 int tls_enq_cnt, tls_pruflag;
1557 tls_rtype = TLS_RLTYPE_APP;
1560 resid = uio->uio_resid;
1561 else if ((top->m_flags & M_PKTHDR) != 0)
1562 resid = top->m_pkthdr.len;
1564 resid = m_length(top, NULL);
1566 * In theory resid should be unsigned. However, space must be
1567 * signed, as it might be less than 0 if we over-committed, and we
1568 * must use a signed comparison of space and resid. On the other
1569 * hand, a negative resid causes us to loop sending 0-length
1570 * segments to the protocol.
1572 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1573 * type sockets since that's an error.
1575 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1581 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1582 (so->so_proto->pr_flags & PR_ATOMIC);
1584 td->td_ru.ru_msgsnd++;
1585 if (control != NULL)
1586 clen = control->m_len;
1588 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1594 tls = ktls_hold(so->so_snd.sb_tls_info);
1596 if (tls->mode == TCP_TLS_MODE_SW)
1597 tls_pruflag = PRUS_NOTREADY;
1599 if (control != NULL) {
1600 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1602 if (clen >= sizeof(*cm) &&
1603 cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1604 tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1612 if (resid == 0 && !ktls_permit_empty_frames(tls)) {
1621 SOCKBUF_LOCK(&so->so_snd);
1622 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1623 SOCKBUF_UNLOCK(&so->so_snd);
1628 error = so->so_error;
1630 SOCKBUF_UNLOCK(&so->so_snd);
1633 if ((so->so_state & SS_ISCONNECTED) == 0) {
1635 * `sendto' and `sendmsg' is allowed on a connection-
1636 * based socket if it supports implied connect.
1637 * Return ENOTCONN if not connected and no address is
1640 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1641 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1642 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1643 !(resid == 0 && clen != 0)) {
1644 SOCKBUF_UNLOCK(&so->so_snd);
1648 } else if (addr == NULL) {
1649 SOCKBUF_UNLOCK(&so->so_snd);
1650 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1653 error = EDESTADDRREQ;
1657 space = sbspace(&so->so_snd);
1658 if (flags & MSG_OOB)
1660 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1661 clen > so->so_snd.sb_hiwat) {
1662 SOCKBUF_UNLOCK(&so->so_snd);
1666 if (space < resid + clen &&
1667 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1668 if ((so->so_state & SS_NBIO) ||
1669 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1670 SOCKBUF_UNLOCK(&so->so_snd);
1671 error = EWOULDBLOCK;
1674 error = sbwait(&so->so_snd);
1675 SOCKBUF_UNLOCK(&so->so_snd);
1680 SOCKBUF_UNLOCK(&so->so_snd);
1685 if (flags & MSG_EOR)
1686 top->m_flags |= M_EOR;
1689 ktls_frame(top, tls, &tls_enq_cnt,
1691 tls_rtype = TLS_RLTYPE_APP;
1696 * Copy the data from userland into a mbuf
1697 * chain. If resid is 0, which can happen
1698 * only if we have control to send, then
1699 * a single empty mbuf is returned. This
1700 * is a workaround to prevent protocol send
1705 top = m_uiotombuf(uio, M_WAITOK, space,
1706 tls->params.max_frame_len,
1708 ((flags & MSG_EOR) ? M_EOR : 0));
1710 ktls_frame(top, tls,
1711 &tls_enq_cnt, tls_rtype);
1713 tls_rtype = TLS_RLTYPE_APP;
1716 top = m_uiotombuf(uio, M_WAITOK, space,
1717 (atomic ? max_hdr : 0),
1718 (atomic ? M_PKTHDR : 0) |
1719 ((flags & MSG_EOR) ? M_EOR : 0));
1721 error = EFAULT; /* only possible error */
1724 space -= resid - uio->uio_resid;
1725 resid = uio->uio_resid;
1729 so->so_options |= SO_DONTROUTE;
1733 * XXX all the SBS_CANTSENDMORE checks previously
1734 * done could be out of date. We could have received
1735 * a reset packet in an interrupt or maybe we slept
1736 * while doing page faults in uiomove() etc. We
1737 * could probably recheck again inside the locking
1738 * protection here, but there are probably other
1739 * places that this also happens. We must rethink
1744 pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1746 * If the user set MSG_EOF, the protocol understands
1747 * this flag and nothing left to send then use
1748 * PRU_SEND_EOF instead of PRU_SEND.
1750 ((flags & MSG_EOF) &&
1751 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1754 /* If there is more to send set PRUS_MORETOCOME. */
1755 (flags & MSG_MORETOCOME) ||
1756 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1759 pru_flag |= tls_pruflag;
1762 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1763 pru_flag, top, addr, control, td);
1767 so->so_options &= ~SO_DONTROUTE;
1772 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1778 ktls_enqueue(top, so, tls_enq_cnt);
1787 } while (resid && space > 0);
1791 SOCK_IO_SEND_UNLOCK(so);
1799 if (control != NULL)
1805 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1806 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1810 CURVNET_SET(so->so_vnet);
1811 if (!SOLISTENING(so))
1812 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1813 top, control, flags, td);
1824 * The part of soreceive() that implements reading non-inline out-of-band
1825 * data from a socket. For more complete comments, see soreceive(), from
1826 * which this code originated.
1828 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1829 * unable to return an mbuf chain to the caller.
1832 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1834 struct protosw *pr = so->so_proto;
1838 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1841 m = m_get(M_WAITOK, MT_DATA);
1842 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1846 error = uiomove(mtod(m, void *),
1847 (int) min(uio->uio_resid, m->m_len), uio);
1849 } while (uio->uio_resid && error == 0 && m);
1857 * Following replacement or removal of the first mbuf on the first mbuf chain
1858 * of a socket buffer, push necessary state changes back into the socket
1859 * buffer so that other consumers see the values consistently. 'nextrecord'
1860 * is the callers locally stored value of the original value of
1861 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1862 * NOTE: 'nextrecord' may be NULL.
1864 static __inline void
1865 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1868 SOCKBUF_LOCK_ASSERT(sb);
1870 * First, update for the new value of nextrecord. If necessary, make
1871 * it the first record.
1873 if (sb->sb_mb != NULL)
1874 sb->sb_mb->m_nextpkt = nextrecord;
1876 sb->sb_mb = nextrecord;
1879 * Now update any dependent socket buffer fields to reflect the new
1880 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1881 * addition of a second clause that takes care of the case where
1882 * sb_mb has been updated, but remains the last record.
1884 if (sb->sb_mb == NULL) {
1885 sb->sb_mbtail = NULL;
1886 sb->sb_lastrecord = NULL;
1887 } else if (sb->sb_mb->m_nextpkt == NULL)
1888 sb->sb_lastrecord = sb->sb_mb;
1892 * Implement receive operations on a socket. We depend on the way that
1893 * records are added to the sockbuf by sbappend. In particular, each record
1894 * (mbufs linked through m_next) must begin with an address if the protocol
1895 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1896 * data, and then zero or more mbufs of data. In order to allow parallelism
1897 * between network receive and copying to user space, as well as avoid
1898 * sleeping with a mutex held, we release the socket buffer mutex during the
1899 * user space copy. Although the sockbuf is locked, new data may still be
1900 * appended, and thus we must maintain consistency of the sockbuf during that
1903 * The caller may receive the data as a single mbuf chain by supplying an
1904 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1905 * the count in uio_resid.
1908 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1909 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1911 struct mbuf *m, **mp;
1912 int flags, error, offset;
1914 struct protosw *pr = so->so_proto;
1915 struct mbuf *nextrecord;
1917 ssize_t orig_resid = uio->uio_resid;
1918 bool report_real_len = false;
1923 if (controlp != NULL)
1925 if (flagsp != NULL) {
1926 report_real_len = *flagsp & MSG_TRUNC;
1927 *flagsp &= ~MSG_TRUNC;
1928 flags = *flagsp &~ MSG_EOR;
1931 if (flags & MSG_OOB)
1932 return (soreceive_rcvoob(so, uio, flags));
1935 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1936 && uio->uio_resid) {
1938 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1941 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1946 SOCKBUF_LOCK(&so->so_rcv);
1947 m = so->so_rcv.sb_mb;
1949 * If we have less data than requested, block awaiting more (subject
1950 * to any timeout) if:
1951 * 1. the current count is less than the low water mark, or
1952 * 2. MSG_DONTWAIT is not set
1954 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1955 sbavail(&so->so_rcv) < uio->uio_resid) &&
1956 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1957 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1958 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1959 ("receive: m == %p sbavail == %u",
1960 m, sbavail(&so->so_rcv)));
1961 if (so->so_error || so->so_rerror) {
1965 error = so->so_error;
1967 error = so->so_rerror;
1968 if ((flags & MSG_PEEK) == 0) {
1974 SOCKBUF_UNLOCK(&so->so_rcv);
1977 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1978 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1982 else if (so->so_rcv.sb_tlsdcc == 0 &&
1983 so->so_rcv.sb_tlscc == 0) {
1987 SOCKBUF_UNLOCK(&so->so_rcv);
1991 for (; m != NULL; m = m->m_next)
1992 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1993 m = so->so_rcv.sb_mb;
1996 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
1997 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
1998 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1999 SOCKBUF_UNLOCK(&so->so_rcv);
2003 if (uio->uio_resid == 0 && !report_real_len) {
2004 SOCKBUF_UNLOCK(&so->so_rcv);
2007 if ((so->so_state & SS_NBIO) ||
2008 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2009 SOCKBUF_UNLOCK(&so->so_rcv);
2010 error = EWOULDBLOCK;
2013 SBLASTRECORDCHK(&so->so_rcv);
2014 SBLASTMBUFCHK(&so->so_rcv);
2015 error = sbwait(&so->so_rcv);
2016 SOCKBUF_UNLOCK(&so->so_rcv);
2023 * From this point onward, we maintain 'nextrecord' as a cache of the
2024 * pointer to the next record in the socket buffer. We must keep the
2025 * various socket buffer pointers and local stack versions of the
2026 * pointers in sync, pushing out modifications before dropping the
2027 * socket buffer mutex, and re-reading them when picking it up.
2029 * Otherwise, we will race with the network stack appending new data
2030 * or records onto the socket buffer by using inconsistent/stale
2031 * versions of the field, possibly resulting in socket buffer
2034 * By holding the high-level sblock(), we prevent simultaneous
2035 * readers from pulling off the front of the socket buffer.
2037 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2039 uio->uio_td->td_ru.ru_msgrcv++;
2040 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2041 SBLASTRECORDCHK(&so->so_rcv);
2042 SBLASTMBUFCHK(&so->so_rcv);
2043 nextrecord = m->m_nextpkt;
2044 if (pr->pr_flags & PR_ADDR) {
2045 KASSERT(m->m_type == MT_SONAME,
2046 ("m->m_type == %d", m->m_type));
2049 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2051 if (flags & MSG_PEEK) {
2054 sbfree(&so->so_rcv, m);
2055 so->so_rcv.sb_mb = m_free(m);
2056 m = so->so_rcv.sb_mb;
2057 sockbuf_pushsync(&so->so_rcv, nextrecord);
2062 * Process one or more MT_CONTROL mbufs present before any data mbufs
2063 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
2064 * just copy the data; if !MSG_PEEK, we call into the protocol to
2065 * perform externalization (or freeing if controlp == NULL).
2067 if (m != NULL && m->m_type == MT_CONTROL) {
2068 struct mbuf *cm = NULL, *cmn;
2069 struct mbuf **cme = &cm;
2071 struct cmsghdr *cmsg;
2072 struct tls_get_record tgr;
2075 * For MSG_TLSAPPDATA, check for an alert record.
2076 * If found, return ENXIO without removing
2077 * it from the receive queue. This allows a subsequent
2078 * call without MSG_TLSAPPDATA to receive it.
2079 * Note that, for TLS, there should only be a single
2080 * control mbuf with the TLS_GET_RECORD message in it.
2082 if (flags & MSG_TLSAPPDATA) {
2083 cmsg = mtod(m, struct cmsghdr *);
2084 if (cmsg->cmsg_type == TLS_GET_RECORD &&
2085 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2086 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2087 if (__predict_false(tgr.tls_type ==
2088 TLS_RLTYPE_ALERT)) {
2089 SOCKBUF_UNLOCK(&so->so_rcv);
2098 if (flags & MSG_PEEK) {
2099 if (controlp != NULL) {
2100 *controlp = m_copym(m, 0, m->m_len,
2102 controlp = &(*controlp)->m_next;
2106 sbfree(&so->so_rcv, m);
2107 so->so_rcv.sb_mb = m->m_next;
2110 cme = &(*cme)->m_next;
2111 m = so->so_rcv.sb_mb;
2113 } while (m != NULL && m->m_type == MT_CONTROL);
2114 if ((flags & MSG_PEEK) == 0)
2115 sockbuf_pushsync(&so->so_rcv, nextrecord);
2116 while (cm != NULL) {
2119 if (pr->pr_domain->dom_externalize != NULL) {
2120 SOCKBUF_UNLOCK(&so->so_rcv);
2122 error = (*pr->pr_domain->dom_externalize)
2123 (cm, controlp, flags);
2124 SOCKBUF_LOCK(&so->so_rcv);
2125 } else if (controlp != NULL)
2129 if (controlp != NULL) {
2130 while (*controlp != NULL)
2131 controlp = &(*controlp)->m_next;
2136 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2138 nextrecord = so->so_rcv.sb_mb;
2142 if ((flags & MSG_PEEK) == 0) {
2143 KASSERT(m->m_nextpkt == nextrecord,
2144 ("soreceive: post-control, nextrecord !sync"));
2145 if (nextrecord == NULL) {
2146 KASSERT(so->so_rcv.sb_mb == m,
2147 ("soreceive: post-control, sb_mb!=m"));
2148 KASSERT(so->so_rcv.sb_lastrecord == m,
2149 ("soreceive: post-control, lastrecord!=m"));
2153 if (type == MT_OOBDATA)
2156 if ((flags & MSG_PEEK) == 0) {
2157 KASSERT(so->so_rcv.sb_mb == nextrecord,
2158 ("soreceive: sb_mb != nextrecord"));
2159 if (so->so_rcv.sb_mb == NULL) {
2160 KASSERT(so->so_rcv.sb_lastrecord == NULL,
2161 ("soreceive: sb_lastercord != NULL"));
2165 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2166 SBLASTRECORDCHK(&so->so_rcv);
2167 SBLASTMBUFCHK(&so->so_rcv);
2170 * Now continue to read any data mbufs off of the head of the socket
2171 * buffer until the read request is satisfied. Note that 'type' is
2172 * used to store the type of any mbuf reads that have happened so far
2173 * such that soreceive() can stop reading if the type changes, which
2174 * causes soreceive() to return only one of regular data and inline
2175 * out-of-band data in a single socket receive operation.
2179 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2182 * If the type of mbuf has changed since the last mbuf
2183 * examined ('type'), end the receive operation.
2185 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2186 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2187 if (type != m->m_type)
2189 } else if (type == MT_OOBDATA)
2192 KASSERT(m->m_type == MT_DATA,
2193 ("m->m_type == %d", m->m_type));
2194 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2195 len = uio->uio_resid;
2196 if (so->so_oobmark && len > so->so_oobmark - offset)
2197 len = so->so_oobmark - offset;
2198 if (len > m->m_len - moff)
2199 len = m->m_len - moff;
2201 * If mp is set, just pass back the mbufs. Otherwise copy
2202 * them out via the uio, then free. Sockbuf must be
2203 * consistent here (points to current mbuf, it points to next
2204 * record) when we drop priority; we must note any additions
2205 * to the sockbuf when we block interrupts again.
2208 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2209 SBLASTRECORDCHK(&so->so_rcv);
2210 SBLASTMBUFCHK(&so->so_rcv);
2211 SOCKBUF_UNLOCK(&so->so_rcv);
2212 if ((m->m_flags & M_EXTPG) != 0)
2213 error = m_unmapped_uiomove(m, moff, uio,
2216 error = uiomove(mtod(m, char *) + moff,
2218 SOCKBUF_LOCK(&so->so_rcv);
2221 * The MT_SONAME mbuf has already been removed
2222 * from the record, so it is necessary to
2223 * remove the data mbufs, if any, to preserve
2224 * the invariant in the case of PR_ADDR that
2225 * requires MT_SONAME mbufs at the head of
2228 if (pr->pr_flags & PR_ATOMIC &&
2229 ((flags & MSG_PEEK) == 0))
2230 (void)sbdroprecord_locked(&so->so_rcv);
2231 SOCKBUF_UNLOCK(&so->so_rcv);
2235 uio->uio_resid -= len;
2236 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2237 if (len == m->m_len - moff) {
2238 if (m->m_flags & M_EOR)
2240 if (flags & MSG_PEEK) {
2244 nextrecord = m->m_nextpkt;
2245 sbfree(&so->so_rcv, m);
2247 m->m_nextpkt = NULL;
2250 so->so_rcv.sb_mb = m = m->m_next;
2253 so->so_rcv.sb_mb = m_free(m);
2254 m = so->so_rcv.sb_mb;
2256 sockbuf_pushsync(&so->so_rcv, nextrecord);
2257 SBLASTRECORDCHK(&so->so_rcv);
2258 SBLASTMBUFCHK(&so->so_rcv);
2261 if (flags & MSG_PEEK)
2265 if (flags & MSG_DONTWAIT) {
2266 *mp = m_copym(m, 0, len,
2270 * m_copym() couldn't
2272 * Adjust uio_resid back
2274 * down by len bytes,
2275 * which we didn't end
2276 * up "copying" over).
2278 uio->uio_resid += len;
2282 SOCKBUF_UNLOCK(&so->so_rcv);
2283 *mp = m_copym(m, 0, len,
2285 SOCKBUF_LOCK(&so->so_rcv);
2288 sbcut_locked(&so->so_rcv, len);
2291 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2292 if (so->so_oobmark) {
2293 if ((flags & MSG_PEEK) == 0) {
2294 so->so_oobmark -= len;
2295 if (so->so_oobmark == 0) {
2296 so->so_rcv.sb_state |= SBS_RCVATMARK;
2301 if (offset == so->so_oobmark)
2305 if (flags & MSG_EOR)
2308 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2309 * must not quit until "uio->uio_resid == 0" or an error
2310 * termination. If a signal/timeout occurs, return with a
2311 * short count but without error. Keep sockbuf locked
2312 * against other readers.
2314 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2315 !sosendallatonce(so) && nextrecord == NULL) {
2316 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2317 if (so->so_error || so->so_rerror ||
2318 so->so_rcv.sb_state & SBS_CANTRCVMORE)
2321 * Notify the protocol that some data has been
2322 * drained before blocking.
2324 if (pr->pr_flags & PR_WANTRCVD) {
2325 SOCKBUF_UNLOCK(&so->so_rcv);
2327 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2328 SOCKBUF_LOCK(&so->so_rcv);
2330 SBLASTRECORDCHK(&so->so_rcv);
2331 SBLASTMBUFCHK(&so->so_rcv);
2333 * We could receive some data while was notifying
2334 * the protocol. Skip blocking in this case.
2336 if (so->so_rcv.sb_mb == NULL) {
2337 error = sbwait(&so->so_rcv);
2339 SOCKBUF_UNLOCK(&so->so_rcv);
2343 m = so->so_rcv.sb_mb;
2345 nextrecord = m->m_nextpkt;
2349 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2350 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2351 if (report_real_len)
2352 uio->uio_resid -= m_length(m, NULL) - moff;
2354 if ((flags & MSG_PEEK) == 0)
2355 (void) sbdroprecord_locked(&so->so_rcv);
2357 if ((flags & MSG_PEEK) == 0) {
2360 * First part is an inline SB_EMPTY_FIXUP(). Second
2361 * part makes sure sb_lastrecord is up-to-date if
2362 * there is still data in the socket buffer.
2364 so->so_rcv.sb_mb = nextrecord;
2365 if (so->so_rcv.sb_mb == NULL) {
2366 so->so_rcv.sb_mbtail = NULL;
2367 so->so_rcv.sb_lastrecord = NULL;
2368 } else if (nextrecord->m_nextpkt == NULL)
2369 so->so_rcv.sb_lastrecord = nextrecord;
2371 SBLASTRECORDCHK(&so->so_rcv);
2372 SBLASTMBUFCHK(&so->so_rcv);
2374 * If soreceive() is being done from the socket callback,
2375 * then don't need to generate ACK to peer to update window,
2376 * since ACK will be generated on return to TCP.
2378 if (!(flags & MSG_SOCALLBCK) &&
2379 (pr->pr_flags & PR_WANTRCVD)) {
2380 SOCKBUF_UNLOCK(&so->so_rcv);
2382 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2383 SOCKBUF_LOCK(&so->so_rcv);
2386 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2387 if (orig_resid == uio->uio_resid && orig_resid &&
2388 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2389 SOCKBUF_UNLOCK(&so->so_rcv);
2392 SOCKBUF_UNLOCK(&so->so_rcv);
2397 SOCK_IO_RECV_UNLOCK(so);
2402 * Optimized version of soreceive() for stream (TCP) sockets.
2405 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2406 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2408 int len = 0, error = 0, flags, oresid;
2410 struct mbuf *m, *n = NULL;
2412 /* We only do stream sockets. */
2413 if (so->so_type != SOCK_STREAM)
2418 flags = *flagsp &~ MSG_EOR;
2421 if (controlp != NULL)
2423 if (flags & MSG_OOB)
2424 return (soreceive_rcvoob(so, uio, flags));
2432 * KTLS store TLS records as records with a control message to
2433 * describe the framing.
2435 * We check once here before acquiring locks to optimize the
2438 if (sb->sb_tls_info != NULL)
2439 return (soreceive_generic(so, psa, uio, mp0, controlp,
2443 /* Prevent other readers from entering the socket. */
2444 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2450 if (sb->sb_tls_info != NULL) {
2452 SOCK_IO_RECV_UNLOCK(so);
2453 return (soreceive_generic(so, psa, uio, mp0, controlp,
2458 /* Easy one, no space to copyout anything. */
2459 if (uio->uio_resid == 0) {
2463 oresid = uio->uio_resid;
2465 /* We will never ever get anything unless we are or were connected. */
2466 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2472 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2474 /* Abort if socket has reported problems. */
2476 if (sbavail(sb) > 0)
2478 if (oresid > uio->uio_resid)
2480 error = so->so_error;
2481 if (!(flags & MSG_PEEK))
2486 /* Door is closed. Deliver what is left, if any. */
2487 if (sb->sb_state & SBS_CANTRCVMORE) {
2488 if (sbavail(sb) > 0)
2494 /* Socket buffer is empty and we shall not block. */
2495 if (sbavail(sb) == 0 &&
2496 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2501 /* Socket buffer got some data that we shall deliver now. */
2502 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2503 ((so->so_state & SS_NBIO) ||
2504 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2505 sbavail(sb) >= sb->sb_lowat ||
2506 sbavail(sb) >= uio->uio_resid ||
2507 sbavail(sb) >= sb->sb_hiwat) ) {
2511 /* On MSG_WAITALL we must wait until all data or error arrives. */
2512 if ((flags & MSG_WAITALL) &&
2513 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2517 * Wait and block until (more) data comes in.
2518 * NB: Drops the sockbuf lock during wait.
2526 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2527 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2528 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2532 uio->uio_td->td_ru.ru_msgrcv++;
2534 /* Fill uio until full or current end of socket buffer is reached. */
2535 len = min(uio->uio_resid, sbavail(sb));
2537 /* Dequeue as many mbufs as possible. */
2538 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2542 m_cat(*mp0, sb->sb_mb);
2544 m != NULL && m->m_len <= len;
2546 KASSERT(!(m->m_flags & M_NOTAVAIL),
2547 ("%s: m %p not available", __func__, m));
2549 uio->uio_resid -= m->m_len;
2555 sb->sb_lastrecord = sb->sb_mb;
2556 if (sb->sb_mb == NULL)
2559 /* Copy the remainder. */
2561 KASSERT(sb->sb_mb != NULL,
2562 ("%s: len > 0 && sb->sb_mb empty", __func__));
2564 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2566 len = 0; /* Don't flush data from sockbuf. */
2568 uio->uio_resid -= len;
2579 /* NB: Must unlock socket buffer as uiomove may sleep. */
2581 error = m_mbuftouio(uio, sb->sb_mb, len);
2586 SBLASTRECORDCHK(sb);
2590 * Remove the delivered data from the socket buffer unless we
2591 * were only peeking.
2593 if (!(flags & MSG_PEEK)) {
2595 sbdrop_locked(sb, len);
2597 /* Notify protocol that we drained some data. */
2598 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2599 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2600 !(flags & MSG_SOCALLBCK))) {
2603 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2609 * For MSG_WAITALL we may have to loop again and wait for
2610 * more data to come in.
2612 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2615 SBLASTRECORDCHK(sb);
2618 SOCK_IO_RECV_UNLOCK(so);
2623 * Optimized version of soreceive() for simple datagram cases from userspace.
2624 * Unlike in the stream case, we're able to drop a datagram if copyout()
2625 * fails, and because we handle datagrams atomically, we don't need to use a
2626 * sleep lock to prevent I/O interlacing.
2629 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2630 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2632 struct mbuf *m, *m2;
2635 struct protosw *pr = so->so_proto;
2636 struct mbuf *nextrecord;
2640 if (controlp != NULL)
2643 flags = *flagsp &~ MSG_EOR;
2648 * For any complicated cases, fall back to the full
2649 * soreceive_generic().
2651 if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC)))
2652 return (soreceive_generic(so, psa, uio, mp0, controlp,
2656 * Enforce restrictions on use.
2658 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2659 ("soreceive_dgram: wantrcvd"));
2660 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2661 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2662 ("soreceive_dgram: SBS_RCVATMARK"));
2663 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2664 ("soreceive_dgram: P_CONNREQUIRED"));
2667 * Loop blocking while waiting for a datagram.
2669 SOCKBUF_LOCK(&so->so_rcv);
2670 while ((m = so->so_rcv.sb_mb) == NULL) {
2671 KASSERT(sbavail(&so->so_rcv) == 0,
2672 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2673 sbavail(&so->so_rcv)));
2675 error = so->so_error;
2677 SOCKBUF_UNLOCK(&so->so_rcv);
2680 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2681 uio->uio_resid == 0) {
2682 SOCKBUF_UNLOCK(&so->so_rcv);
2685 if ((so->so_state & SS_NBIO) ||
2686 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2687 SOCKBUF_UNLOCK(&so->so_rcv);
2688 return (EWOULDBLOCK);
2690 SBLASTRECORDCHK(&so->so_rcv);
2691 SBLASTMBUFCHK(&so->so_rcv);
2692 error = sbwait(&so->so_rcv);
2694 SOCKBUF_UNLOCK(&so->so_rcv);
2698 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2701 uio->uio_td->td_ru.ru_msgrcv++;
2702 SBLASTRECORDCHK(&so->so_rcv);
2703 SBLASTMBUFCHK(&so->so_rcv);
2704 nextrecord = m->m_nextpkt;
2705 if (nextrecord == NULL) {
2706 KASSERT(so->so_rcv.sb_lastrecord == m,
2707 ("soreceive_dgram: lastrecord != m"));
2710 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2711 ("soreceive_dgram: m_nextpkt != nextrecord"));
2714 * Pull 'm' and its chain off the front of the packet queue.
2716 so->so_rcv.sb_mb = NULL;
2717 sockbuf_pushsync(&so->so_rcv, nextrecord);
2720 * Walk 'm's chain and free that many bytes from the socket buffer.
2722 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2723 sbfree(&so->so_rcv, m2);
2726 * Do a few last checks before we let go of the lock.
2728 SBLASTRECORDCHK(&so->so_rcv);
2729 SBLASTMBUFCHK(&so->so_rcv);
2730 SOCKBUF_UNLOCK(&so->so_rcv);
2732 if (pr->pr_flags & PR_ADDR) {
2733 KASSERT(m->m_type == MT_SONAME,
2734 ("m->m_type == %d", m->m_type));
2736 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2741 /* XXXRW: Can this happen? */
2746 * Packet to copyout() is now in 'm' and it is disconnected from the
2749 * Process one or more MT_CONTROL mbufs present before any data mbufs
2750 * in the first mbuf chain on the socket buffer. We call into the
2751 * protocol to perform externalization (or freeing if controlp ==
2752 * NULL). In some cases there can be only MT_CONTROL mbufs without
2755 if (m->m_type == MT_CONTROL) {
2756 struct mbuf *cm = NULL, *cmn;
2757 struct mbuf **cme = &cm;
2763 cme = &(*cme)->m_next;
2765 } while (m != NULL && m->m_type == MT_CONTROL);
2766 while (cm != NULL) {
2769 if (pr->pr_domain->dom_externalize != NULL) {
2770 error = (*pr->pr_domain->dom_externalize)
2771 (cm, controlp, flags);
2772 } else if (controlp != NULL)
2776 if (controlp != NULL) {
2777 while (*controlp != NULL)
2778 controlp = &(*controlp)->m_next;
2783 KASSERT(m == NULL || m->m_type == MT_DATA,
2784 ("soreceive_dgram: !data"));
2785 while (m != NULL && uio->uio_resid > 0) {
2786 len = uio->uio_resid;
2789 error = uiomove(mtod(m, char *), (int)len, uio);
2794 if (len == m->m_len)
2811 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2812 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2816 CURVNET_SET(so->so_vnet);
2817 if (!SOLISTENING(so))
2818 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2819 mp0, controlp, flagsp));
2827 soshutdown(struct socket *so, int how)
2829 struct protosw *pr = so->so_proto;
2830 int error, soerror_enotconn;
2832 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2835 soerror_enotconn = 0;
2837 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2839 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2840 * invoked on a datagram sockets, however historically we would
2841 * actually tear socket down. This is known to be leveraged by
2842 * some applications to unblock process waiting in recvXXX(2)
2843 * by other process that it shares that socket with. Try to meet
2844 * both backward-compatibility and POSIX requirements by forcing
2845 * ENOTCONN but still asking protocol to perform pru_shutdown().
2847 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so))
2849 soerror_enotconn = 1;
2852 if (SOLISTENING(so)) {
2853 if (how != SHUT_WR) {
2855 so->so_error = ECONNABORTED;
2856 solisten_wakeup(so); /* unlocks so */
2861 CURVNET_SET(so->so_vnet);
2862 if (pr->pr_usrreqs->pru_flush != NULL)
2863 (*pr->pr_usrreqs->pru_flush)(so, how);
2866 if (how != SHUT_RD) {
2867 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2868 wakeup(&so->so_timeo);
2870 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2872 wakeup(&so->so_timeo);
2876 return (soerror_enotconn ? ENOTCONN : 0);
2880 sorflush(struct socket *so)
2882 struct sockbuf *sb = &so->so_rcv;
2883 struct protosw *pr = so->so_proto;
2890 * In order to avoid calling dom_dispose with the socket buffer mutex
2891 * held, and in order to generally avoid holding the lock for a long
2892 * time, we make a copy of the socket buffer and clear the original
2893 * (except locks, state). The new socket buffer copy won't have
2894 * initialized locks so we can only call routines that won't use or
2895 * assert those locks.
2897 * Dislodge threads currently blocked in receive and wait to acquire
2898 * a lock against other simultaneous readers before clearing the
2899 * socket buffer. Don't let our acquire be interrupted by a signal
2900 * despite any existing socket disposition on interruptable waiting.
2903 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
2904 KASSERT(error == 0, ("%s: cannot lock sock %p recv buffer",
2908 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2909 * and mutex data unchanged.
2912 bzero(&aso, sizeof(aso));
2913 aso.so_pcb = so->so_pcb;
2914 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2915 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2916 bzero(&sb->sb_startzero,
2917 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2919 SOCK_IO_RECV_UNLOCK(so);
2922 * Dispose of special rights and flush the copied socket. Don't call
2923 * any unsafe routines (that rely on locks being initialized) on aso.
2925 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2926 (*pr->pr_domain->dom_dispose)(&aso);
2927 sbrelease_internal(&aso.so_rcv, so);
2931 * Wrapper for Socket established helper hook.
2932 * Parameters: socket, context of the hook point, hook id.
2935 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2937 struct socket_hhook_data hhook_data = {
2944 CURVNET_SET(so->so_vnet);
2945 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2948 /* Ugly but needed, since hhooks return void for now */
2949 return (hhook_data.status);
2953 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2954 * additional variant to handle the case where the option value needs to be
2955 * some kind of integer, but not a specific size. In addition to their use
2956 * here, these functions are also called by the protocol-level pr_ctloutput()
2960 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2965 * If the user gives us more than we wanted, we ignore it, but if we
2966 * don't get the minimum length the caller wants, we return EINVAL.
2967 * On success, sopt->sopt_valsize is set to however much we actually
2970 if ((valsize = sopt->sopt_valsize) < minlen)
2973 sopt->sopt_valsize = valsize = len;
2975 if (sopt->sopt_td != NULL)
2976 return (copyin(sopt->sopt_val, buf, valsize));
2978 bcopy(sopt->sopt_val, buf, valsize);
2982 u_long nl_maxsockbuf = 512 * 1024 * 1024; /* 512M, XXX: init based on physmem */
2985 sogetmaxbuf(struct socket *so)
2987 if (so->so_proto->pr_domain->dom_family != PF_NETLINK)
2989 return ((priv_check(curthread, PRIV_NET_ROUTE) == 0) ? nl_maxsockbuf : sb_max);
2993 * Kernel version of setsockopt(2).
2995 * XXX: optlen is size_t, not socklen_t
2998 so_setsockopt(struct socket *so, int level, int optname, void *optval,
3001 struct sockopt sopt;
3003 sopt.sopt_level = level;
3004 sopt.sopt_name = optname;
3005 sopt.sopt_dir = SOPT_SET;
3006 sopt.sopt_val = optval;
3007 sopt.sopt_valsize = optlen;
3008 sopt.sopt_td = NULL;
3009 return (sosetopt(so, &sopt));
3013 sosetopt(struct socket *so, struct sockopt *sopt)
3018 sbintime_t val, *valp;
3024 CURVNET_SET(so->so_vnet);
3026 if (sopt->sopt_level != SOL_SOCKET) {
3027 if (so->so_proto->pr_ctloutput != NULL)
3028 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3030 error = ENOPROTOOPT;
3032 switch (sopt->sopt_name) {
3033 case SO_ACCEPTFILTER:
3034 error = accept_filt_setopt(so, sopt);
3040 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3043 if (l.l_linger < 0 ||
3044 l.l_linger > USHRT_MAX ||
3045 l.l_linger > (INT_MAX / hz)) {
3050 so->so_linger = l.l_linger;
3052 so->so_options |= SO_LINGER;
3054 so->so_options &= ~SO_LINGER;
3061 case SO_USELOOPBACK:
3065 case SO_REUSEPORT_LB:
3073 error = sooptcopyin(sopt, &optval, sizeof optval,
3079 so->so_options |= sopt->sopt_name;
3081 so->so_options &= ~sopt->sopt_name;
3086 error = sooptcopyin(sopt, &optval, sizeof optval,
3091 if (optval < 0 || optval >= rt_numfibs) {
3095 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3096 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3097 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3098 so->so_fibnum = optval;
3103 case SO_USER_COOKIE:
3104 error = sooptcopyin(sopt, &val32, sizeof val32,
3108 so->so_user_cookie = val32;
3115 error = sooptcopyin(sopt, &optval, sizeof optval,
3121 * Values < 1 make no sense for any of these options,
3129 error = sbsetopt(so, sopt->sopt_name, optval);
3134 #ifdef COMPAT_FREEBSD32
3135 if (SV_CURPROC_FLAG(SV_ILP32)) {
3136 struct timeval32 tv32;
3138 error = sooptcopyin(sopt, &tv32, sizeof tv32,
3140 CP(tv32, tv, tv_sec);
3141 CP(tv32, tv, tv_usec);
3144 error = sooptcopyin(sopt, &tv, sizeof tv,
3148 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3149 tv.tv_usec >= 1000000) {
3153 if (tv.tv_sec > INT32_MAX)
3158 valp = sopt->sopt_name == SO_SNDTIMEO ?
3159 (SOLISTENING(so) ? &so->sol_sbsnd_timeo :
3160 &so->so_snd.sb_timeo) :
3161 (SOLISTENING(so) ? &so->sol_sbrcv_timeo :
3162 &so->so_rcv.sb_timeo);
3169 error = sooptcopyin(sopt, &extmac, sizeof extmac,
3173 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3181 error = sooptcopyin(sopt, &optval, sizeof optval,
3185 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3189 so->so_ts_clock = optval;
3192 case SO_MAX_PACING_RATE:
3193 error = sooptcopyin(sopt, &val32, sizeof(val32),
3197 so->so_max_pacing_rate = val32;
3201 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3202 error = hhook_run_socket(so, sopt,
3205 error = ENOPROTOOPT;
3208 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3209 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
3217 * Helper routine for getsockopt.
3220 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3228 * Documented get behavior is that we always return a value, possibly
3229 * truncated to fit in the user's buffer. Traditional behavior is
3230 * that we always tell the user precisely how much we copied, rather
3231 * than something useful like the total amount we had available for
3232 * her. Note that this interface is not idempotent; the entire
3233 * answer must be generated ahead of time.
3235 valsize = min(len, sopt->sopt_valsize);
3236 sopt->sopt_valsize = valsize;
3237 if (sopt->sopt_val != NULL) {
3238 if (sopt->sopt_td != NULL)
3239 error = copyout(buf, sopt->sopt_val, valsize);
3241 bcopy(buf, sopt->sopt_val, valsize);
3247 sogetopt(struct socket *so, struct sockopt *sopt)
3256 CURVNET_SET(so->so_vnet);
3258 if (sopt->sopt_level != SOL_SOCKET) {
3259 if (so->so_proto->pr_ctloutput != NULL)
3260 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3262 error = ENOPROTOOPT;
3266 switch (sopt->sopt_name) {
3267 case SO_ACCEPTFILTER:
3268 error = accept_filt_getopt(so, sopt);
3273 l.l_onoff = so->so_options & SO_LINGER;
3274 l.l_linger = so->so_linger;
3276 error = sooptcopyout(sopt, &l, sizeof l);
3279 case SO_USELOOPBACK:
3285 case SO_REUSEPORT_LB:
3295 optval = so->so_options & sopt->sopt_name;
3297 error = sooptcopyout(sopt, &optval, sizeof optval);
3301 optval = so->so_proto->pr_domain->dom_family;
3305 optval = so->so_type;
3309 optval = so->so_proto->pr_protocol;
3315 optval = so->so_error;
3318 optval = so->so_rerror;
3325 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3326 so->so_snd.sb_hiwat;
3330 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3331 so->so_rcv.sb_hiwat;
3335 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3336 so->so_snd.sb_lowat;
3340 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3341 so->so_rcv.sb_lowat;
3347 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3348 (SOLISTENING(so) ? so->sol_sbsnd_timeo :
3349 so->so_snd.sb_timeo) :
3350 (SOLISTENING(so) ? so->sol_sbrcv_timeo :
3351 so->so_rcv.sb_timeo));
3353 #ifdef COMPAT_FREEBSD32
3354 if (SV_CURPROC_FLAG(SV_ILP32)) {
3355 struct timeval32 tv32;
3357 CP(tv, tv32, tv_sec);
3358 CP(tv, tv32, tv_usec);
3359 error = sooptcopyout(sopt, &tv32, sizeof tv32);
3362 error = sooptcopyout(sopt, &tv, sizeof tv);
3367 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3371 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3375 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3383 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3387 error = mac_getsockopt_peerlabel(
3388 sopt->sopt_td->td_ucred, so, &extmac);
3391 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3397 case SO_LISTENQLIMIT:
3398 optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3402 optval = SOLISTENING(so) ? so->sol_qlen : 0;
3405 case SO_LISTENINCQLEN:
3406 optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3410 optval = so->so_ts_clock;
3413 case SO_MAX_PACING_RATE:
3414 optval = so->so_max_pacing_rate;
3418 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3419 error = hhook_run_socket(so, sopt,
3422 error = ENOPROTOOPT;
3434 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3436 struct mbuf *m, *m_prev;
3437 int sopt_size = sopt->sopt_valsize;
3439 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3442 if (sopt_size > MLEN) {
3443 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3444 if ((m->m_flags & M_EXT) == 0) {
3448 m->m_len = min(MCLBYTES, sopt_size);
3450 m->m_len = min(MLEN, sopt_size);
3452 sopt_size -= m->m_len;
3457 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3462 if (sopt_size > MLEN) {
3463 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3465 if ((m->m_flags & M_EXT) == 0) {
3470 m->m_len = min(MCLBYTES, sopt_size);
3472 m->m_len = min(MLEN, sopt_size);
3474 sopt_size -= m->m_len;
3482 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3484 struct mbuf *m0 = m;
3486 if (sopt->sopt_val == NULL)
3488 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3489 if (sopt->sopt_td != NULL) {
3492 error = copyin(sopt->sopt_val, mtod(m, char *),
3499 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3500 sopt->sopt_valsize -= m->m_len;
3501 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3504 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3505 panic("ip6_sooptmcopyin");
3510 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3512 struct mbuf *m0 = m;
3515 if (sopt->sopt_val == NULL)
3517 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3518 if (sopt->sopt_td != NULL) {
3521 error = copyout(mtod(m, char *), sopt->sopt_val,
3528 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3529 sopt->sopt_valsize -= m->m_len;
3530 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3531 valsize += m->m_len;
3535 /* enough soopt buffer should be given from user-land */
3539 sopt->sopt_valsize = valsize;
3544 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3545 * out-of-band data, which will then notify socket consumers.
3548 sohasoutofband(struct socket *so)
3551 if (so->so_sigio != NULL)
3552 pgsigio(&so->so_sigio, SIGURG, 0);
3553 selwakeuppri(&so->so_rdsel, PSOCK);
3557 sopoll(struct socket *so, int events, struct ucred *active_cred,
3562 * We do not need to set or assert curvnet as long as everyone uses
3565 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3570 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3576 if (SOLISTENING(so)) {
3577 if (!(events & (POLLIN | POLLRDNORM)))
3579 else if (!TAILQ_EMPTY(&so->sol_comp))
3580 revents = events & (POLLIN | POLLRDNORM);
3581 else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3582 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3584 selrecord(td, &so->so_rdsel);
3589 SOCKBUF_LOCK(&so->so_snd);
3590 SOCKBUF_LOCK(&so->so_rcv);
3591 if (events & (POLLIN | POLLRDNORM))
3592 if (soreadabledata(so))
3593 revents |= events & (POLLIN | POLLRDNORM);
3594 if (events & (POLLOUT | POLLWRNORM))
3595 if (sowriteable(so))
3596 revents |= events & (POLLOUT | POLLWRNORM);
3597 if (events & (POLLPRI | POLLRDBAND))
3598 if (so->so_oobmark ||
3599 (so->so_rcv.sb_state & SBS_RCVATMARK))
3600 revents |= events & (POLLPRI | POLLRDBAND);
3601 if ((events & POLLINIGNEOF) == 0) {
3602 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3603 revents |= events & (POLLIN | POLLRDNORM);
3604 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3608 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3609 revents |= events & POLLRDHUP;
3612 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
3613 selrecord(td, &so->so_rdsel);
3614 so->so_rcv.sb_flags |= SB_SEL;
3616 if (events & (POLLOUT | POLLWRNORM)) {
3617 selrecord(td, &so->so_wrsel);
3618 so->so_snd.sb_flags |= SB_SEL;
3621 SOCKBUF_UNLOCK(&so->so_rcv);
3622 SOCKBUF_UNLOCK(&so->so_snd);
3629 soo_kqfilter(struct file *fp, struct knote *kn)
3631 struct socket *so = kn->kn_fp->f_data;
3635 switch (kn->kn_filter) {
3637 kn->kn_fop = &soread_filtops;
3638 knl = &so->so_rdsel.si_note;
3642 kn->kn_fop = &sowrite_filtops;
3643 knl = &so->so_wrsel.si_note;
3647 kn->kn_fop = &soempty_filtops;
3648 knl = &so->so_wrsel.si_note;
3656 if (SOLISTENING(so)) {
3657 knlist_add(knl, kn, 1);
3660 knlist_add(knl, kn, 1);
3661 sb->sb_flags |= SB_KNOTE;
3669 * Some routines that return EOPNOTSUPP for entry points that are not
3670 * supported by a protocol. Fill in as needed.
3673 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3680 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3687 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3694 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3701 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3709 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3716 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3724 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3731 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3732 struct ifnet *ifp, struct thread *td)
3739 pru_disconnect_notsupp(struct socket *so)
3746 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3753 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3760 pru_rcvd_notsupp(struct socket *so, int flags)
3767 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3774 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3775 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3778 if (control != NULL)
3780 if ((flags & PRUS_NOTREADY) == 0)
3782 return (EOPNOTSUPP);
3786 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3789 return (EOPNOTSUPP);
3793 * This isn't really a ``null'' operation, but it's the default one and
3794 * doesn't do anything destructive.
3797 pru_sense_null(struct socket *so, struct stat *sb)
3800 sb->st_blksize = so->so_snd.sb_hiwat;
3805 pru_shutdown_notsupp(struct socket *so)
3812 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3819 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3820 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3827 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3828 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3835 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3843 filt_sordetach(struct knote *kn)
3845 struct socket *so = kn->kn_fp->f_data;
3848 knlist_remove(&so->so_rdsel.si_note, kn, 1);
3849 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3850 so->so_rcv.sb_flags &= ~SB_KNOTE;
3851 so_rdknl_unlock(so);
3856 filt_soread(struct knote *kn, long hint)
3860 so = kn->kn_fp->f_data;
3862 if (SOLISTENING(so)) {
3863 SOCK_LOCK_ASSERT(so);
3864 kn->kn_data = so->sol_qlen;
3866 kn->kn_flags |= EV_EOF;
3867 kn->kn_fflags = so->so_error;
3870 return (!TAILQ_EMPTY(&so->sol_comp));
3873 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3875 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3876 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3877 kn->kn_flags |= EV_EOF;
3878 kn->kn_fflags = so->so_error;
3880 } else if (so->so_error || so->so_rerror)
3883 if (kn->kn_sfflags & NOTE_LOWAT) {
3884 if (kn->kn_data >= kn->kn_sdata)
3886 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3889 /* This hook returning non-zero indicates an event, not error */
3890 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3894 filt_sowdetach(struct knote *kn)
3896 struct socket *so = kn->kn_fp->f_data;
3899 knlist_remove(&so->so_wrsel.si_note, kn, 1);
3900 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3901 so->so_snd.sb_flags &= ~SB_KNOTE;
3902 so_wrknl_unlock(so);
3907 filt_sowrite(struct knote *kn, long hint)
3911 so = kn->kn_fp->f_data;
3913 if (SOLISTENING(so))
3916 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3917 kn->kn_data = sbspace(&so->so_snd);
3919 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3921 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3922 kn->kn_flags |= EV_EOF;
3923 kn->kn_fflags = so->so_error;
3925 } else if (so->so_error) /* temporary udp error */
3927 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3928 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3930 else if (kn->kn_sfflags & NOTE_LOWAT)
3931 return (kn->kn_data >= kn->kn_sdata);
3933 return (kn->kn_data >= so->so_snd.sb_lowat);
3937 filt_soempty(struct knote *kn, long hint)
3941 so = kn->kn_fp->f_data;
3943 if (SOLISTENING(so))
3946 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3947 kn->kn_data = sbused(&so->so_snd);
3949 if (kn->kn_data == 0)
3956 socheckuid(struct socket *so, uid_t uid)
3961 if (so->so_cred->cr_uid != uid)
3967 * These functions are used by protocols to notify the socket layer (and its
3968 * consumers) of state changes in the sockets driven by protocol-side events.
3972 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3974 * Normal sequence from the active (originating) side is that
3975 * soisconnecting() is called during processing of connect() call, resulting
3976 * in an eventual call to soisconnected() if/when the connection is
3977 * established. When the connection is torn down soisdisconnecting() is
3978 * called during processing of disconnect() call, and soisdisconnected() is
3979 * called when the connection to the peer is totally severed. The semantics
3980 * of these routines are such that connectionless protocols can call
3981 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3982 * calls when setting up a ``connection'' takes no time.
3984 * From the passive side, a socket is created with two queues of sockets:
3985 * so_incomp for connections in progress and so_comp for connections already
3986 * made and awaiting user acceptance. As a protocol is preparing incoming
3987 * connections, it creates a socket structure queued on so_incomp by calling
3988 * sonewconn(). When the connection is established, soisconnected() is
3989 * called, and transfers the socket structure to so_comp, making it available
3992 * If a socket is closed with sockets on either so_incomp or so_comp, these
3993 * sockets are dropped.
3995 * If higher-level protocols are implemented in the kernel, the wakeups done
3996 * here will sometimes cause software-interrupt process scheduling.
3999 soisconnecting(struct socket *so)
4003 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
4004 so->so_state |= SS_ISCONNECTING;
4009 soisconnected(struct socket *so)
4011 bool last __diagused;
4014 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
4015 so->so_state |= SS_ISCONNECTED;
4017 if (so->so_qstate == SQ_INCOMP) {
4018 struct socket *head = so->so_listen;
4021 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
4023 * Promoting a socket from incomplete queue to complete, we
4024 * need to go through reverse order of locking. We first do
4025 * trylock, and if that doesn't succeed, we go the hard way
4026 * leaving a reference and rechecking consistency after proper
4029 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
4032 SOLISTEN_LOCK(head);
4034 if (__predict_false(head != so->so_listen)) {
4036 * The socket went off the listen queue,
4037 * should be lost race to close(2) of sol.
4038 * The socket is about to soabort().
4044 last = refcount_release(&head->so_count);
4045 KASSERT(!last, ("%s: released last reference for %p",
4049 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4050 TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4051 head->sol_incqlen--;
4052 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4054 so->so_qstate = SQ_COMP;
4056 solisten_wakeup(head); /* unlocks */
4058 SOCKBUF_LOCK(&so->so_rcv);
4059 soupcall_set(so, SO_RCV,
4060 head->sol_accept_filter->accf_callback,
4061 head->sol_accept_filter_arg);
4062 so->so_options &= ~SO_ACCEPTFILTER;
4063 ret = head->sol_accept_filter->accf_callback(so,
4064 head->sol_accept_filter_arg, M_NOWAIT);
4065 if (ret == SU_ISCONNECTED) {
4066 soupcall_clear(so, SO_RCV);
4067 SOCKBUF_UNLOCK(&so->so_rcv);
4070 SOCKBUF_UNLOCK(&so->so_rcv);
4072 SOLISTEN_UNLOCK(head);
4077 wakeup(&so->so_timeo);
4083 soisdisconnecting(struct socket *so)
4087 so->so_state &= ~SS_ISCONNECTING;
4088 so->so_state |= SS_ISDISCONNECTING;
4090 if (!SOLISTENING(so)) {
4091 SOCKBUF_LOCK(&so->so_rcv);
4092 socantrcvmore_locked(so);
4093 SOCKBUF_LOCK(&so->so_snd);
4094 socantsendmore_locked(so);
4097 wakeup(&so->so_timeo);
4101 soisdisconnected(struct socket *so)
4107 * There is at least one reader of so_state that does not
4108 * acquire socket lock, namely soreceive_generic(). Ensure
4109 * that it never sees all flags that track connection status
4110 * cleared, by ordering the update with a barrier semantic of
4111 * our release thread fence.
4113 so->so_state |= SS_ISDISCONNECTED;
4114 atomic_thread_fence_rel();
4115 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4117 if (!SOLISTENING(so)) {
4119 SOCKBUF_LOCK(&so->so_rcv);
4120 socantrcvmore_locked(so);
4121 SOCKBUF_LOCK(&so->so_snd);
4122 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4123 socantsendmore_locked(so);
4126 wakeup(&so->so_timeo);
4130 soiolock(struct socket *so, struct sx *sx, int flags)
4134 KASSERT((flags & SBL_VALID) == flags,
4135 ("soiolock: invalid flags %#x", flags));
4137 if ((flags & SBL_WAIT) != 0) {
4138 if ((flags & SBL_NOINTR) != 0) {
4141 error = sx_xlock_sig(sx);
4145 } else if (!sx_try_xlock(sx)) {
4146 return (EWOULDBLOCK);
4149 if (__predict_false(SOLISTENING(so))) {
4157 soiounlock(struct sx *sx)
4163 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4166 sodupsockaddr(const struct sockaddr *sa, int mflags)
4168 struct sockaddr *sa2;
4170 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4172 bcopy(sa, sa2, sa->sa_len);
4177 * Register per-socket destructor.
4180 sodtor_set(struct socket *so, so_dtor_t *func)
4183 SOCK_LOCK_ASSERT(so);
4188 * Register per-socket buffer upcalls.
4191 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4195 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4205 panic("soupcall_set: bad which");
4207 SOCKBUF_LOCK_ASSERT(sb);
4208 sb->sb_upcall = func;
4209 sb->sb_upcallarg = arg;
4210 sb->sb_flags |= SB_UPCALL;
4214 soupcall_clear(struct socket *so, int which)
4218 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4228 panic("soupcall_clear: bad which");
4230 SOCKBUF_LOCK_ASSERT(sb);
4231 KASSERT(sb->sb_upcall != NULL,
4232 ("%s: so %p no upcall to clear", __func__, so));
4233 sb->sb_upcall = NULL;
4234 sb->sb_upcallarg = NULL;
4235 sb->sb_flags &= ~SB_UPCALL;
4239 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4242 SOLISTEN_LOCK_ASSERT(so);
4243 so->sol_upcall = func;
4244 so->sol_upcallarg = arg;
4248 so_rdknl_lock(void *arg)
4250 struct socket *so = arg;
4252 if (SOLISTENING(so))
4255 SOCKBUF_LOCK(&so->so_rcv);
4259 so_rdknl_unlock(void *arg)
4261 struct socket *so = arg;
4263 if (SOLISTENING(so))
4266 SOCKBUF_UNLOCK(&so->so_rcv);
4270 so_rdknl_assert_lock(void *arg, int what)
4272 struct socket *so = arg;
4274 if (what == LA_LOCKED) {
4275 if (SOLISTENING(so))
4276 SOCK_LOCK_ASSERT(so);
4278 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4280 if (SOLISTENING(so))
4281 SOCK_UNLOCK_ASSERT(so);
4283 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4288 so_wrknl_lock(void *arg)
4290 struct socket *so = arg;
4292 if (SOLISTENING(so))
4295 SOCKBUF_LOCK(&so->so_snd);
4299 so_wrknl_unlock(void *arg)
4301 struct socket *so = arg;
4303 if (SOLISTENING(so))
4306 SOCKBUF_UNLOCK(&so->so_snd);
4310 so_wrknl_assert_lock(void *arg, int what)
4312 struct socket *so = arg;
4314 if (what == LA_LOCKED) {
4315 if (SOLISTENING(so))
4316 SOCK_LOCK_ASSERT(so);
4318 SOCKBUF_LOCK_ASSERT(&so->so_snd);
4320 if (SOLISTENING(so))
4321 SOCK_UNLOCK_ASSERT(so);
4323 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4328 * Create an external-format (``xsocket'') structure using the information in
4329 * the kernel-format socket structure pointed to by so. This is done to
4330 * reduce the spew of irrelevant information over this interface, to isolate
4331 * user code from changes in the kernel structure, and potentially to provide
4332 * information-hiding if we decide that some of this information should be
4333 * hidden from users.
4336 sotoxsocket(struct socket *so, struct xsocket *xso)
4339 bzero(xso, sizeof(*xso));
4340 xso->xso_len = sizeof *xso;
4341 xso->xso_so = (uintptr_t)so;
4342 xso->so_type = so->so_type;
4343 xso->so_options = so->so_options;
4344 xso->so_linger = so->so_linger;
4345 xso->so_state = so->so_state;
4346 xso->so_pcb = (uintptr_t)so->so_pcb;
4347 xso->xso_protocol = so->so_proto->pr_protocol;
4348 xso->xso_family = so->so_proto->pr_domain->dom_family;
4349 xso->so_timeo = so->so_timeo;
4350 xso->so_error = so->so_error;
4351 xso->so_uid = so->so_cred->cr_uid;
4352 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4353 if (SOLISTENING(so)) {
4354 xso->so_qlen = so->sol_qlen;
4355 xso->so_incqlen = so->sol_incqlen;
4356 xso->so_qlimit = so->sol_qlimit;
4357 xso->so_oobmark = 0;
4359 xso->so_state |= so->so_qstate;
4360 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4361 xso->so_oobmark = so->so_oobmark;
4362 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4363 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4368 so_sockbuf_rcv(struct socket *so)
4371 return (&so->so_rcv);
4375 so_sockbuf_snd(struct socket *so)
4378 return (&so->so_snd);
4382 so_state_get(const struct socket *so)
4385 return (so->so_state);
4389 so_state_set(struct socket *so, int val)
4396 so_options_get(const struct socket *so)
4399 return (so->so_options);
4403 so_options_set(struct socket *so, int val)
4406 so->so_options = val;
4410 so_error_get(const struct socket *so)
4413 return (so->so_error);
4417 so_error_set(struct socket *so, int val)
4424 so_linger_get(const struct socket *so)
4427 return (so->so_linger);
4431 so_linger_set(struct socket *so, int val)
4434 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4435 ("%s: val %d out of range", __func__, val));
4437 so->so_linger = val;
4441 so_protosw_get(const struct socket *so)
4444 return (so->so_proto);
4448 so_protosw_set(struct socket *so, struct protosw *val)
4455 so_sorwakeup(struct socket *so)
4462 so_sowwakeup(struct socket *so)
4469 so_sorwakeup_locked(struct socket *so)
4472 sorwakeup_locked(so);
4476 so_sowwakeup_locked(struct socket *so)
4479 sowwakeup_locked(so);
4483 so_lock(struct socket *so)
4490 so_unlock(struct socket *so)