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 __FBSDID("$FreeBSD$");
108 #include "opt_inet.h"
109 #include "opt_inet6.h"
110 #include "opt_kern_tls.h"
111 #include "opt_sctp.h"
113 #include <sys/param.h>
114 #include <sys/systm.h>
115 #include <sys/fcntl.h>
116 #include <sys/limits.h>
117 #include <sys/lock.h>
119 #include <sys/malloc.h>
120 #include <sys/mbuf.h>
121 #include <sys/mutex.h>
122 #include <sys/domain.h>
123 #include <sys/file.h> /* for struct knote */
124 #include <sys/hhook.h>
125 #include <sys/kernel.h>
126 #include <sys/khelp.h>
127 #include <sys/ktls.h>
128 #include <sys/event.h>
129 #include <sys/eventhandler.h>
130 #include <sys/poll.h>
131 #include <sys/priv.h>
132 #include <sys/proc.h>
133 #include <sys/protosw.h>
134 #include <sys/sbuf.h>
135 #include <sys/socket.h>
136 #include <sys/socketvar.h>
137 #include <sys/resourcevar.h>
138 #include <net/route.h>
139 #include <sys/signalvar.h>
140 #include <sys/stat.h>
142 #include <sys/sysctl.h>
143 #include <sys/taskqueue.h>
146 #include <sys/unpcb.h>
147 #include <sys/jail.h>
148 #include <sys/syslog.h>
149 #include <netinet/in.h>
150 #include <netinet/in_pcb.h>
151 #include <netinet/tcp.h>
153 #include <net/vnet.h>
155 #include <security/mac/mac_framework.h>
159 #ifdef COMPAT_FREEBSD32
160 #include <sys/mount.h>
161 #include <sys/sysent.h>
162 #include <compat/freebsd32/freebsd32.h>
165 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
167 static void so_rdknl_lock(void *);
168 static void so_rdknl_unlock(void *);
169 static void so_rdknl_assert_lock(void *, int);
170 static void so_wrknl_lock(void *);
171 static void so_wrknl_unlock(void *);
172 static void so_wrknl_assert_lock(void *, int);
174 static void filt_sordetach(struct knote *kn);
175 static int filt_soread(struct knote *kn, long hint);
176 static void filt_sowdetach(struct knote *kn);
177 static int filt_sowrite(struct knote *kn, long hint);
178 static int filt_soempty(struct knote *kn, long hint);
179 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
180 fo_kqfilter_t soo_kqfilter;
182 static struct filterops soread_filtops = {
184 .f_detach = filt_sordetach,
185 .f_event = filt_soread,
187 static struct filterops sowrite_filtops = {
189 .f_detach = filt_sowdetach,
190 .f_event = filt_sowrite,
192 static struct filterops soempty_filtops = {
194 .f_detach = filt_sowdetach,
195 .f_event = filt_soempty,
198 so_gen_t so_gencnt; /* generation count for sockets */
200 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
201 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
203 #define VNET_SO_ASSERT(so) \
204 VNET_ASSERT(curvnet != NULL, \
205 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
207 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
208 #define V_socket_hhh VNET(socket_hhh)
211 * Limit on the number of connections in the listen queue waiting
213 * NB: The original sysctl somaxconn is still available but hidden
214 * to prevent confusion about the actual purpose of this number.
216 static u_int somaxconn = SOMAXCONN;
219 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
225 error = sysctl_handle_int(oidp, &val, 0, req);
226 if (error || !req->newptr )
230 * The purpose of the UINT_MAX / 3 limit, is so that the formula
232 * below, will not overflow.
235 if (val < 1 || val > UINT_MAX / 3)
241 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
242 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
243 sysctl_somaxconn, "I",
244 "Maximum listen socket pending connection accept queue size");
245 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
246 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, 0,
247 sizeof(int), sysctl_somaxconn, "I",
248 "Maximum listen socket pending connection accept queue size (compat)");
250 static int numopensockets;
251 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
252 &numopensockets, 0, "Number of open sockets");
255 * accept_mtx locks down per-socket fields relating to accept queues. See
256 * socketvar.h for an annotation of the protected fields of struct socket.
258 struct mtx accept_mtx;
259 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
262 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
265 static struct mtx so_global_mtx;
266 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
269 * General IPC sysctl name space, used by sockets and a variety of other IPC
272 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
276 * Initialize the socket subsystem and set up the socket
279 static uma_zone_t socket_zone;
283 socket_zone_change(void *tag)
286 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
290 socket_hhook_register(int subtype)
293 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
294 &V_socket_hhh[subtype],
295 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
296 printf("%s: WARNING: unable to register hook\n", __func__);
300 socket_hhook_deregister(int subtype)
303 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
304 printf("%s: WARNING: unable to deregister hook\n", __func__);
308 socket_init(void *tag)
311 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
312 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
313 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
314 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
315 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
316 EVENTHANDLER_PRI_FIRST);
318 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
321 socket_vnet_init(const void *unused __unused)
325 /* We expect a contiguous range */
326 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
327 socket_hhook_register(i);
329 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
330 socket_vnet_init, NULL);
333 socket_vnet_uninit(const void *unused __unused)
337 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
338 socket_hhook_deregister(i);
340 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
341 socket_vnet_uninit, NULL);
344 * Initialise maxsockets. This SYSINIT must be run after
348 init_maxsockets(void *ignored)
351 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
352 maxsockets = imax(maxsockets, maxfiles);
354 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
357 * Sysctl to get and set the maximum global sockets limit. Notify protocols
358 * of the change so that they can update their dependent limits as required.
361 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
363 int error, newmaxsockets;
365 newmaxsockets = maxsockets;
366 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
367 if (error == 0 && req->newptr && newmaxsockets != maxsockets) {
368 if (newmaxsockets > maxsockets &&
369 newmaxsockets <= maxfiles) {
370 maxsockets = newmaxsockets;
371 EVENTHANDLER_INVOKE(maxsockets_change);
377 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
378 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &maxsockets, 0,
379 sysctl_maxsockets, "IU",
380 "Maximum number of sockets available");
383 * Socket operation routines. These routines are called by the routines in
384 * sys_socket.c or from a system process, and implement the semantics of
385 * socket operations by switching out to the protocol specific routines.
389 * Get a socket structure from our zone, and initialize it. Note that it
390 * would probably be better to allocate socket and PCB at the same time, but
391 * I'm not convinced that all the protocols can be easily modified to do
394 * soalloc() returns a socket with a ref count of 0.
396 static struct socket *
397 soalloc(struct vnet *vnet)
401 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
405 if (mac_socket_init(so, M_NOWAIT) != 0) {
406 uma_zfree(socket_zone, so);
410 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
411 uma_zfree(socket_zone, so);
416 * The socket locking protocol allows to lock 2 sockets at a time,
417 * however, the first one must be a listening socket. WITNESS lacks
418 * a feature to change class of an existing lock, so we use DUPOK.
420 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
421 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
422 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
423 so->so_rcv.sb_sel = &so->so_rdsel;
424 so->so_snd.sb_sel = &so->so_wrsel;
425 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
426 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
427 TAILQ_INIT(&so->so_snd.sb_aiojobq);
428 TAILQ_INIT(&so->so_rcv.sb_aiojobq);
429 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
430 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
432 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
433 __func__, __LINE__, so));
436 /* We shouldn't need the so_global_mtx */
437 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
438 /* Do we need more comprehensive error returns? */
439 uma_zfree(socket_zone, so);
442 mtx_lock(&so_global_mtx);
443 so->so_gencnt = ++so_gencnt;
446 vnet->vnet_sockcnt++;
448 mtx_unlock(&so_global_mtx);
454 * Free the storage associated with a socket at the socket layer, tear down
455 * locks, labels, etc. All protocol state is assumed already to have been
456 * torn down (and possibly never set up) by the caller.
459 sodealloc(struct socket *so)
462 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
463 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
465 mtx_lock(&so_global_mtx);
466 so->so_gencnt = ++so_gencnt;
467 --numopensockets; /* Could be below, but faster here. */
469 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
470 __func__, __LINE__, so));
471 so->so_vnet->vnet_sockcnt--;
473 mtx_unlock(&so_global_mtx);
475 mac_socket_destroy(so);
477 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
479 khelp_destroy_osd(&so->osd);
480 if (SOLISTENING(so)) {
481 if (so->sol_accept_filter != NULL)
482 accept_filt_setopt(so, NULL);
484 if (so->so_rcv.sb_hiwat)
485 (void)chgsbsize(so->so_cred->cr_uidinfo,
486 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
487 if (so->so_snd.sb_hiwat)
488 (void)chgsbsize(so->so_cred->cr_uidinfo,
489 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
490 sx_destroy(&so->so_snd.sb_sx);
491 sx_destroy(&so->so_rcv.sb_sx);
492 SOCKBUF_LOCK_DESTROY(&so->so_snd);
493 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
496 mtx_destroy(&so->so_lock);
497 uma_zfree(socket_zone, so);
501 * socreate returns a socket with a ref count of 1. The socket should be
502 * closed with soclose().
505 socreate(int dom, struct socket **aso, int type, int proto,
506 struct ucred *cred, struct thread *td)
513 prp = pffindproto(dom, proto, type);
515 prp = pffindtype(dom, type);
518 /* No support for domain. */
519 if (pffinddomain(dom) == NULL)
520 return (EAFNOSUPPORT);
521 /* No support for socket type. */
522 if (proto == 0 && type != 0)
524 return (EPROTONOSUPPORT);
526 if (prp->pr_usrreqs->pru_attach == NULL ||
527 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
528 return (EPROTONOSUPPORT);
530 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
531 return (EPROTONOSUPPORT);
533 if (prp->pr_type != type)
535 so = soalloc(CRED_TO_VNET(cred));
540 so->so_cred = crhold(cred);
541 if ((prp->pr_domain->dom_family == PF_INET) ||
542 (prp->pr_domain->dom_family == PF_INET6) ||
543 (prp->pr_domain->dom_family == PF_ROUTE))
544 so->so_fibnum = td->td_proc->p_fibnum;
549 mac_socket_create(cred, so);
551 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
552 so_rdknl_assert_lock);
553 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
554 so_wrknl_assert_lock);
556 * Auto-sizing of socket buffers is managed by the protocols and
557 * the appropriate flags must be set in the pru_attach function.
559 CURVNET_SET(so->so_vnet);
560 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
572 static int regression_sonewconn_earlytest = 1;
573 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
574 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
577 static int sooverprio = LOG_DEBUG;
578 SYSCTL_INT(_kern_ipc, OID_AUTO, sooverprio, CTLFLAG_RW,
579 &sooverprio, 0, "Log priority for listen socket overflows: 0..7 or -1 to disable");
581 static struct timeval overinterval = { 60, 0 };
582 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
584 "Delay in seconds between warnings for listen socket overflows");
587 * When an attempt at a new connection is noted on a socket which accepts
588 * connections, sonewconn is called. If the connection is possible (subject
589 * to space constraints, etc.) then we allocate a new structure, properly
590 * linked into the data structure of the original socket, and return this.
591 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
593 * Note: the ref count on the socket is 0 on return.
596 sonewconn(struct socket *head, int connstatus)
602 const char localprefix[] = "local:";
603 char descrbuf[SUNPATHLEN + sizeof(localprefix)];
605 char addrbuf[INET6_ADDRSTRLEN];
607 char addrbuf[INET_ADDRSTRLEN];
612 over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
614 if (regression_sonewconn_earlytest && over) {
618 head->sol_overcount++;
619 dolog = (sooverprio >= 0) &&
620 !!ratecheck(&head->sol_lastover, &overinterval);
623 * If we're going to log, copy the overflow count and queue
624 * length from the listen socket before dropping the lock.
625 * Also, reset the overflow count.
628 overcount = head->sol_overcount;
629 head->sol_overcount = 0;
630 qlen = head->sol_qlen;
632 SOLISTEN_UNLOCK(head);
636 * Try to print something descriptive about the
637 * socket for the error message.
639 sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
641 switch (head->so_proto->pr_domain->dom_family) {
642 #if defined(INET) || defined(INET6)
648 if (head->so_proto->pr_domain->dom_family ==
650 (sotoinpcb(head)->inp_inc.inc_flags &
653 &sotoinpcb(head)->inp_inc.inc6_laddr);
654 sbuf_printf(&descrsb, "[%s]", addrbuf);
660 sotoinpcb(head)->inp_inc.inc_laddr,
662 sbuf_cat(&descrsb, addrbuf);
665 sbuf_printf(&descrsb, ":%hu (proto %u)",
666 ntohs(sotoinpcb(head)->inp_inc.inc_lport),
667 head->so_proto->pr_protocol);
669 #endif /* INET || INET6 */
671 sbuf_cat(&descrsb, localprefix);
672 if (sotounpcb(head)->unp_addr != NULL)
674 sotounpcb(head)->unp_addr->sun_len -
675 offsetof(struct sockaddr_un,
681 sotounpcb(head)->unp_addr->sun_path,
684 sbuf_cat(&descrsb, "(unknown)");
689 * If we can't print something more specific, at least
690 * print the domain name.
692 if (sbuf_finish(&descrsb) != 0 ||
693 sbuf_len(&descrsb) <= 0) {
694 sbuf_clear(&descrsb);
696 head->so_proto->pr_domain->dom_name ?:
698 sbuf_finish(&descrsb);
700 KASSERT(sbuf_len(&descrsb) > 0,
701 ("%s: sbuf creation failed", __func__));
702 log(LOG_PRI(sooverprio),
703 "%s: pcb %p (%s): Listen queue overflow: "
704 "%i already in queue awaiting acceptance "
705 "(%d occurrences)\n",
706 __func__, head->so_pcb, sbuf_data(&descrsb),
708 sbuf_delete(&descrsb);
715 SOLISTEN_UNLOCK(head);
716 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
718 so = soalloc(head->so_vnet);
720 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
721 "limit reached or out of memory\n",
722 __func__, head->so_pcb);
725 so->so_listen = head;
726 so->so_type = head->so_type;
727 so->so_options = head->so_options & ~SO_ACCEPTCONN;
728 so->so_linger = head->so_linger;
729 so->so_state = head->so_state | SS_NOFDREF;
730 so->so_fibnum = head->so_fibnum;
731 so->so_proto = head->so_proto;
732 so->so_cred = crhold(head->so_cred);
734 mac_socket_newconn(head, so);
736 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
737 so_rdknl_assert_lock);
738 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
739 so_wrknl_assert_lock);
740 VNET_SO_ASSERT(head);
741 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
743 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
744 __func__, head->so_pcb);
747 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
749 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
750 __func__, head->so_pcb);
753 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
754 so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
755 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
756 so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
757 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
758 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
761 if (head->sol_accept_filter != NULL)
763 so->so_state |= connstatus;
764 soref(head); /* A socket on (in)complete queue refs head. */
766 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
767 so->so_qstate = SQ_COMP;
769 solisten_wakeup(head); /* unlocks */
772 * Keep removing sockets from the head until there's room for
773 * us to insert on the tail. In pre-locking revisions, this
774 * was a simple if(), but as we could be racing with other
775 * threads and soabort() requires dropping locks, we must
776 * loop waiting for the condition to be true.
778 while (head->sol_incqlen > head->sol_qlimit) {
781 sp = TAILQ_FIRST(&head->sol_incomp);
782 TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
785 sp->so_qstate = SQ_NONE;
786 sp->so_listen = NULL;
788 sorele(head); /* does SOLISTEN_UNLOCK, head stays */
792 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
793 so->so_qstate = SQ_INCOMP;
795 SOLISTEN_UNLOCK(head);
800 #if defined(SCTP) || defined(SCTP_SUPPORT)
802 * Socket part of sctp_peeloff(). Detach a new socket from an
803 * association. The new socket is returned with a reference.
806 sopeeloff(struct socket *head)
810 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
811 __func__, __LINE__, head));
812 so = soalloc(head->so_vnet);
814 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
815 "limit reached or out of memory\n",
816 __func__, head->so_pcb);
819 so->so_type = head->so_type;
820 so->so_options = head->so_options;
821 so->so_linger = head->so_linger;
822 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
823 so->so_fibnum = head->so_fibnum;
824 so->so_proto = head->so_proto;
825 so->so_cred = crhold(head->so_cred);
827 mac_socket_newconn(head, so);
829 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
830 so_rdknl_assert_lock);
831 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
832 so_wrknl_assert_lock);
833 VNET_SO_ASSERT(head);
834 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
836 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
837 __func__, head->so_pcb);
840 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
842 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
843 __func__, head->so_pcb);
846 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
847 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
848 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
849 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
850 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
851 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
860 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
864 CURVNET_SET(so->so_vnet);
865 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
871 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
875 CURVNET_SET(so->so_vnet);
876 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
882 * solisten() transitions a socket from a non-listening state to a listening
883 * state, but can also be used to update the listen queue depth on an
884 * existing listen socket. The protocol will call back into the sockets
885 * layer using solisten_proto_check() and solisten_proto() to check and set
886 * socket-layer listen state. Call backs are used so that the protocol can
887 * acquire both protocol and socket layer locks in whatever order is required
890 * Protocol implementors are advised to hold the socket lock across the
891 * socket-layer test and set to avoid races at the socket layer.
894 solisten(struct socket *so, int backlog, struct thread *td)
898 CURVNET_SET(so->so_vnet);
899 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
905 solisten_proto_check(struct socket *so)
908 SOCK_LOCK_ASSERT(so);
910 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
917 solisten_proto(struct socket *so, int backlog)
919 int sbrcv_lowat, sbsnd_lowat;
920 u_int sbrcv_hiwat, sbsnd_hiwat;
921 short sbrcv_flags, sbsnd_flags;
922 sbintime_t sbrcv_timeo, sbsnd_timeo;
924 SOCK_LOCK_ASSERT(so);
930 * Change this socket to listening state.
932 sbrcv_lowat = so->so_rcv.sb_lowat;
933 sbsnd_lowat = so->so_snd.sb_lowat;
934 sbrcv_hiwat = so->so_rcv.sb_hiwat;
935 sbsnd_hiwat = so->so_snd.sb_hiwat;
936 sbrcv_flags = so->so_rcv.sb_flags;
937 sbsnd_flags = so->so_snd.sb_flags;
938 sbrcv_timeo = so->so_rcv.sb_timeo;
939 sbsnd_timeo = so->so_snd.sb_timeo;
941 sbdestroy(&so->so_snd, so);
942 sbdestroy(&so->so_rcv, so);
943 sx_destroy(&so->so_snd.sb_sx);
944 sx_destroy(&so->so_rcv.sb_sx);
945 SOCKBUF_LOCK_DESTROY(&so->so_snd);
946 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
950 sizeof(struct socket) - offsetof(struct socket, so_rcv));
953 so->sol_sbrcv_lowat = sbrcv_lowat;
954 so->sol_sbsnd_lowat = sbsnd_lowat;
955 so->sol_sbrcv_hiwat = sbrcv_hiwat;
956 so->sol_sbsnd_hiwat = sbsnd_hiwat;
957 so->sol_sbrcv_flags = sbrcv_flags;
958 so->sol_sbsnd_flags = sbsnd_flags;
959 so->sol_sbrcv_timeo = sbrcv_timeo;
960 so->sol_sbsnd_timeo = sbsnd_timeo;
962 so->sol_qlen = so->sol_incqlen = 0;
963 TAILQ_INIT(&so->sol_incomp);
964 TAILQ_INIT(&so->sol_comp);
966 so->sol_accept_filter = NULL;
967 so->sol_accept_filter_arg = NULL;
968 so->sol_accept_filter_str = NULL;
970 so->sol_upcall = NULL;
971 so->sol_upcallarg = NULL;
973 so->so_options |= SO_ACCEPTCONN;
976 if (backlog < 0 || backlog > somaxconn)
978 so->sol_qlimit = backlog;
982 * Wakeup listeners/subsystems once we have a complete connection.
983 * Enters with lock, returns unlocked.
986 solisten_wakeup(struct socket *sol)
989 if (sol->sol_upcall != NULL)
990 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
992 selwakeuppri(&sol->so_rdsel, PSOCK);
993 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
995 SOLISTEN_UNLOCK(sol);
996 wakeup_one(&sol->sol_comp);
997 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
998 pgsigio(&sol->so_sigio, SIGIO, 0);
1002 * Return single connection off a listening socket queue. Main consumer of
1003 * the function is kern_accept4(). Some modules, that do their own accept
1004 * management also use the function.
1006 * Listening socket must be locked on entry and is returned unlocked on
1008 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1011 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1016 SOLISTEN_LOCK_ASSERT(head);
1018 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1019 head->so_error == 0) {
1020 error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH,
1023 SOLISTEN_UNLOCK(head);
1027 if (head->so_error) {
1028 error = head->so_error;
1030 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1031 error = EWOULDBLOCK;
1035 SOLISTEN_UNLOCK(head);
1038 so = TAILQ_FIRST(&head->sol_comp);
1040 KASSERT(so->so_qstate == SQ_COMP,
1041 ("%s: so %p not SQ_COMP", __func__, so));
1044 so->so_qstate = SQ_NONE;
1045 so->so_listen = NULL;
1046 TAILQ_REMOVE(&head->sol_comp, so, so_list);
1047 if (flags & ACCEPT4_INHERIT)
1048 so->so_state |= (head->so_state & SS_NBIO);
1050 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1059 * Evaluate the reference count and named references on a socket; if no
1060 * references remain, free it. This should be called whenever a reference is
1061 * released, such as in sorele(), but also when named reference flags are
1062 * cleared in socket or protocol code.
1064 * sofree() will free the socket if:
1066 * - There are no outstanding file descriptor references or related consumers
1069 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
1071 * - The protocol does not have an outstanding strong reference on the socket
1074 * - The socket is not in a completed connection queue, so a process has been
1075 * notified that it is present. If it is removed, the user process may
1076 * block in accept() despite select() saying the socket was ready.
1079 sofree(struct socket *so)
1081 struct protosw *pr = so->so_proto;
1082 bool last __diagused;
1084 SOCK_LOCK_ASSERT(so);
1086 if ((so->so_state & (SS_NOFDREF | SS_PROTOREF)) != SS_NOFDREF ||
1087 refcount_load(&so->so_count) != 0 || so->so_qstate == SQ_COMP) {
1092 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
1095 sol = so->so_listen;
1096 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
1099 * To solve race between close of a listening socket and
1100 * a socket on its incomplete queue, we need to lock both.
1101 * The order is first listening socket, then regular.
1102 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
1103 * function and the listening socket are the only pointers
1104 * to so. To preserve so and sol, we reference both and then
1106 * After relock the socket may not move to so_comp since it
1107 * doesn't have PCB already, but it may be removed from
1108 * so_incomp. If that happens, we share responsiblity on
1109 * freeing the socket, but soclose() has already removed
1117 if (so->so_qstate == SQ_INCOMP) {
1118 KASSERT(so->so_listen == sol,
1119 ("%s: so %p migrated out of sol %p",
1120 __func__, so, sol));
1121 TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1123 last = refcount_release(&sol->so_count);
1124 KASSERT(!last, ("%s: released last reference for %p",
1126 so->so_qstate = SQ_NONE;
1127 so->so_listen = NULL;
1129 KASSERT(so->so_listen == NULL,
1130 ("%s: so %p not on (in)comp with so_listen",
1133 KASSERT(refcount_load(&so->so_count) == 1,
1134 ("%s: so %p count %u", __func__, so, so->so_count));
1137 if (SOLISTENING(so))
1138 so->so_error = ECONNABORTED;
1141 if (so->so_dtor != NULL)
1145 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1146 (*pr->pr_domain->dom_dispose)(so);
1147 if (pr->pr_usrreqs->pru_detach != NULL)
1148 (*pr->pr_usrreqs->pru_detach)(so);
1151 * From this point on, we assume that no other references to this
1152 * socket exist anywhere else in the stack. Therefore, no locks need
1153 * to be acquired or held.
1155 * We used to do a lot of socket buffer and socket locking here, as
1156 * well as invoke sorflush() and perform wakeups. The direct call to
1157 * dom_dispose() and sbdestroy() are an inlining of what was
1158 * necessary from sorflush().
1160 * Notice that the socket buffer and kqueue state are torn down
1161 * before calling pru_detach. This means that protocols shold not
1162 * assume they can perform socket wakeups, etc, in their detach code.
1164 if (!SOLISTENING(so)) {
1165 sbdestroy(&so->so_snd, so);
1166 sbdestroy(&so->so_rcv, so);
1168 seldrain(&so->so_rdsel);
1169 seldrain(&so->so_wrsel);
1170 knlist_destroy(&so->so_rdsel.si_note);
1171 knlist_destroy(&so->so_wrsel.si_note);
1176 * Close a socket on last file table reference removal. Initiate disconnect
1177 * if connected. Free socket when disconnect complete.
1179 * This function will sorele() the socket. Note that soclose() may be called
1180 * prior to the ref count reaching zero. The actual socket structure will
1181 * not be freed until the ref count reaches zero.
1184 soclose(struct socket *so)
1186 struct accept_queue lqueue;
1187 struct socket *sp, *tsp;
1189 bool last __diagused;
1191 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1193 CURVNET_SET(so->so_vnet);
1194 funsetown(&so->so_sigio);
1195 if (so->so_state & SS_ISCONNECTED) {
1196 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1197 error = sodisconnect(so);
1199 if (error == ENOTCONN)
1205 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1206 if ((so->so_state & SS_ISDISCONNECTING) &&
1207 (so->so_state & SS_NBIO))
1209 while (so->so_state & SS_ISCONNECTED) {
1210 error = tsleep(&so->so_timeo,
1211 PSOCK | PCATCH, "soclos",
1212 so->so_linger * hz);
1220 if (so->so_proto->pr_usrreqs->pru_close != NULL)
1221 (*so->so_proto->pr_usrreqs->pru_close)(so);
1223 TAILQ_INIT(&lqueue);
1225 if (SOLISTENING(so)) {
1226 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1227 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1229 so->sol_qlen = so->sol_incqlen = 0;
1231 TAILQ_FOREACH(sp, &lqueue, so_list) {
1233 sp->so_qstate = SQ_NONE;
1234 sp->so_listen = NULL;
1236 last = refcount_release(&so->so_count);
1237 KASSERT(!last, ("%s: released last reference for %p",
1241 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1242 so->so_state |= SS_NOFDREF;
1244 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1246 if (refcount_load(&sp->so_count) == 0) {
1250 /* See the handling of queued sockets in sofree(). */
1259 * soabort() is used to abruptly tear down a connection, such as when a
1260 * resource limit is reached (listen queue depth exceeded), or if a listen
1261 * socket is closed while there are sockets waiting to be accepted.
1263 * This interface is tricky, because it is called on an unreferenced socket,
1264 * and must be called only by a thread that has actually removed the socket
1265 * from the listen queue it was on, or races with other threads are risked.
1267 * This interface will call into the protocol code, so must not be called
1268 * with any socket locks held. Protocols do call it while holding their own
1269 * recursible protocol mutexes, but this is something that should be subject
1270 * to review in the future.
1273 soabort(struct socket *so)
1277 * In as much as is possible, assert that no references to this
1278 * socket are held. This is not quite the same as asserting that the
1279 * current thread is responsible for arranging for no references, but
1280 * is as close as we can get for now.
1282 KASSERT(so->so_count == 0, ("soabort: so_count"));
1283 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1284 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1287 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1288 (*so->so_proto->pr_usrreqs->pru_abort)(so);
1294 soaccept(struct socket *so, struct sockaddr **nam)
1299 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1300 so->so_state &= ~SS_NOFDREF;
1303 CURVNET_SET(so->so_vnet);
1304 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1310 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1313 return (soconnectat(AT_FDCWD, so, nam, td));
1317 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1322 if (SOLISTENING(so))
1323 return (EOPNOTSUPP);
1325 CURVNET_SET(so->so_vnet);
1327 * If protocol is connection-based, can only connect once.
1328 * Otherwise, if connected, try to disconnect first. This allows
1329 * user to disconnect by connecting to, e.g., a null address.
1331 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1332 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1333 (error = sodisconnect(so)))) {
1337 * Prevent accumulated error from previous connection from
1341 if (fd == AT_FDCWD) {
1342 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1345 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1355 soconnect2(struct socket *so1, struct socket *so2)
1359 CURVNET_SET(so1->so_vnet);
1360 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1366 sodisconnect(struct socket *so)
1370 if ((so->so_state & SS_ISCONNECTED) == 0)
1372 if (so->so_state & SS_ISDISCONNECTING)
1375 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1380 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1381 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1385 int clen = 0, error, dontroute;
1387 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1388 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1389 ("sosend_dgram: !PR_ATOMIC"));
1392 resid = uio->uio_resid;
1394 resid = top->m_pkthdr.len;
1396 * In theory resid should be unsigned. However, space must be
1397 * signed, as it might be less than 0 if we over-committed, and we
1398 * must use a signed comparison of space and resid. On the other
1399 * hand, a negative resid causes us to loop sending 0-length
1400 * segments to the protocol.
1408 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1410 td->td_ru.ru_msgsnd++;
1411 if (control != NULL)
1412 clen = control->m_len;
1414 SOCKBUF_LOCK(&so->so_snd);
1415 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1416 SOCKBUF_UNLOCK(&so->so_snd);
1421 error = so->so_error;
1423 SOCKBUF_UNLOCK(&so->so_snd);
1426 if ((so->so_state & SS_ISCONNECTED) == 0) {
1428 * `sendto' and `sendmsg' is allowed on a connection-based
1429 * socket if it supports implied connect. Return ENOTCONN if
1430 * not connected and no address is supplied.
1432 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1433 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1434 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1435 !(resid == 0 && clen != 0)) {
1436 SOCKBUF_UNLOCK(&so->so_snd);
1440 } else if (addr == NULL) {
1441 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1444 error = EDESTADDRREQ;
1445 SOCKBUF_UNLOCK(&so->so_snd);
1451 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1452 * problem and need fixing.
1454 space = sbspace(&so->so_snd);
1455 if (flags & MSG_OOB)
1458 SOCKBUF_UNLOCK(&so->so_snd);
1459 if (resid > space) {
1465 if (flags & MSG_EOR)
1466 top->m_flags |= M_EOR;
1469 * Copy the data from userland into a mbuf chain.
1470 * If no data is to be copied in, a single empty mbuf
1473 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1474 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1476 error = EFAULT; /* only possible error */
1479 space -= resid - uio->uio_resid;
1480 resid = uio->uio_resid;
1482 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1484 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1489 so->so_options |= SO_DONTROUTE;
1493 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1494 * of date. We could have received a reset packet in an interrupt or
1495 * maybe we slept while doing page faults in uiomove() etc. We could
1496 * probably recheck again inside the locking protection here, but
1497 * there are probably other places that this also happens. We must
1501 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1502 (flags & MSG_OOB) ? PRUS_OOB :
1504 * If the user set MSG_EOF, the protocol understands this flag and
1505 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1507 ((flags & MSG_EOF) &&
1508 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1511 /* If there is more to send set PRUS_MORETOCOME */
1512 (flags & MSG_MORETOCOME) ||
1513 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1514 top, addr, control, td);
1517 so->so_options &= ~SO_DONTROUTE;
1526 if (control != NULL)
1532 * Send on a socket. If send must go all at once and message is larger than
1533 * send buffering, then hard error. Lock against other senders. If must go
1534 * all at once and not enough room now, then inform user that this would
1535 * block and do nothing. Otherwise, if nonblocking, send as much as
1536 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1537 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1538 * in mbuf chain must be small enough to send all at once.
1540 * Returns nonzero on error, timeout or signal; callers must check for short
1541 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1545 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1546 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1550 int clen = 0, error, dontroute;
1551 int atomic = sosendallatonce(so) || top;
1554 struct ktls_session *tls;
1555 int tls_enq_cnt, tls_pruflag;
1559 tls_rtype = TLS_RLTYPE_APP;
1562 resid = uio->uio_resid;
1563 else if ((top->m_flags & M_PKTHDR) != 0)
1564 resid = top->m_pkthdr.len;
1566 resid = m_length(top, NULL);
1568 * In theory resid should be unsigned. However, space must be
1569 * signed, as it might be less than 0 if we over-committed, and we
1570 * must use a signed comparison of space and resid. On the other
1571 * hand, a negative resid causes us to loop sending 0-length
1572 * segments to the protocol.
1574 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1575 * type sockets since that's an error.
1577 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1583 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1584 (so->so_proto->pr_flags & PR_ATOMIC);
1586 td->td_ru.ru_msgsnd++;
1587 if (control != NULL)
1588 clen = control->m_len;
1590 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1596 tls = ktls_hold(so->so_snd.sb_tls_info);
1598 if (tls->mode == TCP_TLS_MODE_SW)
1599 tls_pruflag = PRUS_NOTREADY;
1601 if (control != NULL) {
1602 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1604 if (clen >= sizeof(*cm) &&
1605 cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1606 tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1614 if (resid == 0 && !ktls_permit_empty_frames(tls)) {
1623 SOCKBUF_LOCK(&so->so_snd);
1624 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1625 SOCKBUF_UNLOCK(&so->so_snd);
1630 error = so->so_error;
1632 SOCKBUF_UNLOCK(&so->so_snd);
1635 if ((so->so_state & SS_ISCONNECTED) == 0) {
1637 * `sendto' and `sendmsg' is allowed on a connection-
1638 * based socket if it supports implied connect.
1639 * Return ENOTCONN if not connected and no address is
1642 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1643 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1644 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1645 !(resid == 0 && clen != 0)) {
1646 SOCKBUF_UNLOCK(&so->so_snd);
1650 } else if (addr == NULL) {
1651 SOCKBUF_UNLOCK(&so->so_snd);
1652 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1655 error = EDESTADDRREQ;
1659 space = sbspace(&so->so_snd);
1660 if (flags & MSG_OOB)
1662 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1663 clen > so->so_snd.sb_hiwat) {
1664 SOCKBUF_UNLOCK(&so->so_snd);
1668 if (space < resid + clen &&
1669 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1670 if ((so->so_state & SS_NBIO) ||
1671 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1672 SOCKBUF_UNLOCK(&so->so_snd);
1673 error = EWOULDBLOCK;
1676 error = sbwait(&so->so_snd);
1677 SOCKBUF_UNLOCK(&so->so_snd);
1682 SOCKBUF_UNLOCK(&so->so_snd);
1687 if (flags & MSG_EOR)
1688 top->m_flags |= M_EOR;
1691 ktls_frame(top, tls, &tls_enq_cnt,
1693 tls_rtype = TLS_RLTYPE_APP;
1698 * Copy the data from userland into a mbuf
1699 * chain. If resid is 0, which can happen
1700 * only if we have control to send, then
1701 * a single empty mbuf is returned. This
1702 * is a workaround to prevent protocol send
1707 top = m_uiotombuf(uio, M_WAITOK, space,
1708 tls->params.max_frame_len,
1710 ((flags & MSG_EOR) ? M_EOR : 0));
1712 ktls_frame(top, tls,
1713 &tls_enq_cnt, tls_rtype);
1715 tls_rtype = TLS_RLTYPE_APP;
1718 top = m_uiotombuf(uio, M_WAITOK, space,
1719 (atomic ? max_hdr : 0),
1720 (atomic ? M_PKTHDR : 0) |
1721 ((flags & MSG_EOR) ? M_EOR : 0));
1723 error = EFAULT; /* only possible error */
1726 space -= resid - uio->uio_resid;
1727 resid = uio->uio_resid;
1731 so->so_options |= SO_DONTROUTE;
1735 * XXX all the SBS_CANTSENDMORE checks previously
1736 * done could be out of date. We could have received
1737 * a reset packet in an interrupt or maybe we slept
1738 * while doing page faults in uiomove() etc. We
1739 * could probably recheck again inside the locking
1740 * protection here, but there are probably other
1741 * places that this also happens. We must rethink
1746 pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1748 * If the user set MSG_EOF, the protocol understands
1749 * this flag and nothing left to send then use
1750 * PRU_SEND_EOF instead of PRU_SEND.
1752 ((flags & MSG_EOF) &&
1753 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1756 /* If there is more to send set PRUS_MORETOCOME. */
1757 (flags & MSG_MORETOCOME) ||
1758 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1761 pru_flag |= tls_pruflag;
1764 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1765 pru_flag, top, addr, control, td);
1769 so->so_options &= ~SO_DONTROUTE;
1774 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1780 ktls_enqueue(top, so, tls_enq_cnt);
1789 } while (resid && space > 0);
1793 SOCK_IO_SEND_UNLOCK(so);
1801 if (control != NULL)
1807 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1808 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1812 CURVNET_SET(so->so_vnet);
1813 if (!SOLISTENING(so))
1814 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1815 top, control, flags, td);
1826 * The part of soreceive() that implements reading non-inline out-of-band
1827 * data from a socket. For more complete comments, see soreceive(), from
1828 * which this code originated.
1830 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1831 * unable to return an mbuf chain to the caller.
1834 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1836 struct protosw *pr = so->so_proto;
1840 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1843 m = m_get(M_WAITOK, MT_DATA);
1844 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1848 error = uiomove(mtod(m, void *),
1849 (int) min(uio->uio_resid, m->m_len), uio);
1851 } while (uio->uio_resid && error == 0 && m);
1859 * Following replacement or removal of the first mbuf on the first mbuf chain
1860 * of a socket buffer, push necessary state changes back into the socket
1861 * buffer so that other consumers see the values consistently. 'nextrecord'
1862 * is the callers locally stored value of the original value of
1863 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1864 * NOTE: 'nextrecord' may be NULL.
1866 static __inline void
1867 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1870 SOCKBUF_LOCK_ASSERT(sb);
1872 * First, update for the new value of nextrecord. If necessary, make
1873 * it the first record.
1875 if (sb->sb_mb != NULL)
1876 sb->sb_mb->m_nextpkt = nextrecord;
1878 sb->sb_mb = nextrecord;
1881 * Now update any dependent socket buffer fields to reflect the new
1882 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1883 * addition of a second clause that takes care of the case where
1884 * sb_mb has been updated, but remains the last record.
1886 if (sb->sb_mb == NULL) {
1887 sb->sb_mbtail = NULL;
1888 sb->sb_lastrecord = NULL;
1889 } else if (sb->sb_mb->m_nextpkt == NULL)
1890 sb->sb_lastrecord = sb->sb_mb;
1894 * Implement receive operations on a socket. We depend on the way that
1895 * records are added to the sockbuf by sbappend. In particular, each record
1896 * (mbufs linked through m_next) must begin with an address if the protocol
1897 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1898 * data, and then zero or more mbufs of data. In order to allow parallelism
1899 * between network receive and copying to user space, as well as avoid
1900 * sleeping with a mutex held, we release the socket buffer mutex during the
1901 * user space copy. Although the sockbuf is locked, new data may still be
1902 * appended, and thus we must maintain consistency of the sockbuf during that
1905 * The caller may receive the data as a single mbuf chain by supplying an
1906 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1907 * the count in uio_resid.
1910 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1911 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1913 struct mbuf *m, **mp;
1914 int flags, error, offset;
1916 struct protosw *pr = so->so_proto;
1917 struct mbuf *nextrecord;
1919 ssize_t orig_resid = uio->uio_resid;
1920 bool report_real_len = false;
1925 if (controlp != NULL)
1927 if (flagsp != NULL) {
1928 report_real_len = *flagsp & MSG_TRUNC;
1929 *flagsp &= ~MSG_TRUNC;
1930 flags = *flagsp &~ MSG_EOR;
1933 if (flags & MSG_OOB)
1934 return (soreceive_rcvoob(so, uio, flags));
1937 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1938 && uio->uio_resid) {
1940 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1943 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1948 SOCKBUF_LOCK(&so->so_rcv);
1949 m = so->so_rcv.sb_mb;
1951 * If we have less data than requested, block awaiting more (subject
1952 * to any timeout) if:
1953 * 1. the current count is less than the low water mark, or
1954 * 2. MSG_DONTWAIT is not set
1956 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1957 sbavail(&so->so_rcv) < uio->uio_resid) &&
1958 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1959 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1960 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1961 ("receive: m == %p sbavail == %u",
1962 m, sbavail(&so->so_rcv)));
1963 if (so->so_error || so->so_rerror) {
1967 error = so->so_error;
1969 error = so->so_rerror;
1970 if ((flags & MSG_PEEK) == 0) {
1976 SOCKBUF_UNLOCK(&so->so_rcv);
1979 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1980 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1984 else if (so->so_rcv.sb_tlsdcc == 0 &&
1985 so->so_rcv.sb_tlscc == 0) {
1989 SOCKBUF_UNLOCK(&so->so_rcv);
1993 for (; m != NULL; m = m->m_next)
1994 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1995 m = so->so_rcv.sb_mb;
1998 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
1999 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
2000 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
2001 SOCKBUF_UNLOCK(&so->so_rcv);
2005 if (uio->uio_resid == 0 && !report_real_len) {
2006 SOCKBUF_UNLOCK(&so->so_rcv);
2009 if ((so->so_state & SS_NBIO) ||
2010 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2011 SOCKBUF_UNLOCK(&so->so_rcv);
2012 error = EWOULDBLOCK;
2015 SBLASTRECORDCHK(&so->so_rcv);
2016 SBLASTMBUFCHK(&so->so_rcv);
2017 error = sbwait(&so->so_rcv);
2018 SOCKBUF_UNLOCK(&so->so_rcv);
2025 * From this point onward, we maintain 'nextrecord' as a cache of the
2026 * pointer to the next record in the socket buffer. We must keep the
2027 * various socket buffer pointers and local stack versions of the
2028 * pointers in sync, pushing out modifications before dropping the
2029 * socket buffer mutex, and re-reading them when picking it up.
2031 * Otherwise, we will race with the network stack appending new data
2032 * or records onto the socket buffer by using inconsistent/stale
2033 * versions of the field, possibly resulting in socket buffer
2036 * By holding the high-level sblock(), we prevent simultaneous
2037 * readers from pulling off the front of the socket buffer.
2039 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2041 uio->uio_td->td_ru.ru_msgrcv++;
2042 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2043 SBLASTRECORDCHK(&so->so_rcv);
2044 SBLASTMBUFCHK(&so->so_rcv);
2045 nextrecord = m->m_nextpkt;
2046 if (pr->pr_flags & PR_ADDR) {
2047 KASSERT(m->m_type == MT_SONAME,
2048 ("m->m_type == %d", m->m_type));
2051 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2053 if (flags & MSG_PEEK) {
2056 sbfree(&so->so_rcv, m);
2057 so->so_rcv.sb_mb = m_free(m);
2058 m = so->so_rcv.sb_mb;
2059 sockbuf_pushsync(&so->so_rcv, nextrecord);
2064 * Process one or more MT_CONTROL mbufs present before any data mbufs
2065 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
2066 * just copy the data; if !MSG_PEEK, we call into the protocol to
2067 * perform externalization (or freeing if controlp == NULL).
2069 if (m != NULL && m->m_type == MT_CONTROL) {
2070 struct mbuf *cm = NULL, *cmn;
2071 struct mbuf **cme = &cm;
2073 struct cmsghdr *cmsg;
2074 struct tls_get_record tgr;
2077 * For MSG_TLSAPPDATA, check for an alert record.
2078 * If found, return ENXIO without removing
2079 * it from the receive queue. This allows a subsequent
2080 * call without MSG_TLSAPPDATA to receive it.
2081 * Note that, for TLS, there should only be a single
2082 * control mbuf with the TLS_GET_RECORD message in it.
2084 if (flags & MSG_TLSAPPDATA) {
2085 cmsg = mtod(m, struct cmsghdr *);
2086 if (cmsg->cmsg_type == TLS_GET_RECORD &&
2087 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2088 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2089 if (__predict_false(tgr.tls_type ==
2090 TLS_RLTYPE_ALERT)) {
2091 SOCKBUF_UNLOCK(&so->so_rcv);
2100 if (flags & MSG_PEEK) {
2101 if (controlp != NULL) {
2102 *controlp = m_copym(m, 0, m->m_len,
2104 controlp = &(*controlp)->m_next;
2108 sbfree(&so->so_rcv, m);
2109 so->so_rcv.sb_mb = m->m_next;
2112 cme = &(*cme)->m_next;
2113 m = so->so_rcv.sb_mb;
2115 } while (m != NULL && m->m_type == MT_CONTROL);
2116 if ((flags & MSG_PEEK) == 0)
2117 sockbuf_pushsync(&so->so_rcv, nextrecord);
2118 while (cm != NULL) {
2121 if (pr->pr_domain->dom_externalize != NULL) {
2122 SOCKBUF_UNLOCK(&so->so_rcv);
2124 error = (*pr->pr_domain->dom_externalize)
2125 (cm, controlp, flags);
2126 SOCKBUF_LOCK(&so->so_rcv);
2127 } else if (controlp != NULL)
2131 if (controlp != NULL) {
2132 while (*controlp != NULL)
2133 controlp = &(*controlp)->m_next;
2138 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2140 nextrecord = so->so_rcv.sb_mb;
2144 if ((flags & MSG_PEEK) == 0) {
2145 KASSERT(m->m_nextpkt == nextrecord,
2146 ("soreceive: post-control, nextrecord !sync"));
2147 if (nextrecord == NULL) {
2148 KASSERT(so->so_rcv.sb_mb == m,
2149 ("soreceive: post-control, sb_mb!=m"));
2150 KASSERT(so->so_rcv.sb_lastrecord == m,
2151 ("soreceive: post-control, lastrecord!=m"));
2155 if (type == MT_OOBDATA)
2158 if ((flags & MSG_PEEK) == 0) {
2159 KASSERT(so->so_rcv.sb_mb == nextrecord,
2160 ("soreceive: sb_mb != nextrecord"));
2161 if (so->so_rcv.sb_mb == NULL) {
2162 KASSERT(so->so_rcv.sb_lastrecord == NULL,
2163 ("soreceive: sb_lastercord != NULL"));
2167 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2168 SBLASTRECORDCHK(&so->so_rcv);
2169 SBLASTMBUFCHK(&so->so_rcv);
2172 * Now continue to read any data mbufs off of the head of the socket
2173 * buffer until the read request is satisfied. Note that 'type' is
2174 * used to store the type of any mbuf reads that have happened so far
2175 * such that soreceive() can stop reading if the type changes, which
2176 * causes soreceive() to return only one of regular data and inline
2177 * out-of-band data in a single socket receive operation.
2181 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2184 * If the type of mbuf has changed since the last mbuf
2185 * examined ('type'), end the receive operation.
2187 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2188 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2189 if (type != m->m_type)
2191 } else if (type == MT_OOBDATA)
2194 KASSERT(m->m_type == MT_DATA,
2195 ("m->m_type == %d", m->m_type));
2196 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2197 len = uio->uio_resid;
2198 if (so->so_oobmark && len > so->so_oobmark - offset)
2199 len = so->so_oobmark - offset;
2200 if (len > m->m_len - moff)
2201 len = m->m_len - moff;
2203 * If mp is set, just pass back the mbufs. Otherwise copy
2204 * them out via the uio, then free. Sockbuf must be
2205 * consistent here (points to current mbuf, it points to next
2206 * record) when we drop priority; we must note any additions
2207 * to the sockbuf when we block interrupts again.
2210 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2211 SBLASTRECORDCHK(&so->so_rcv);
2212 SBLASTMBUFCHK(&so->so_rcv);
2213 SOCKBUF_UNLOCK(&so->so_rcv);
2214 if ((m->m_flags & M_EXTPG) != 0)
2215 error = m_unmapped_uiomove(m, moff, uio,
2218 error = uiomove(mtod(m, char *) + moff,
2220 SOCKBUF_LOCK(&so->so_rcv);
2223 * The MT_SONAME mbuf has already been removed
2224 * from the record, so it is necessary to
2225 * remove the data mbufs, if any, to preserve
2226 * the invariant in the case of PR_ADDR that
2227 * requires MT_SONAME mbufs at the head of
2230 if (pr->pr_flags & PR_ATOMIC &&
2231 ((flags & MSG_PEEK) == 0))
2232 (void)sbdroprecord_locked(&so->so_rcv);
2233 SOCKBUF_UNLOCK(&so->so_rcv);
2237 uio->uio_resid -= len;
2238 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2239 if (len == m->m_len - moff) {
2240 if (m->m_flags & M_EOR)
2242 if (flags & MSG_PEEK) {
2246 nextrecord = m->m_nextpkt;
2247 sbfree(&so->so_rcv, m);
2249 m->m_nextpkt = NULL;
2252 so->so_rcv.sb_mb = m = m->m_next;
2255 so->so_rcv.sb_mb = m_free(m);
2256 m = so->so_rcv.sb_mb;
2258 sockbuf_pushsync(&so->so_rcv, nextrecord);
2259 SBLASTRECORDCHK(&so->so_rcv);
2260 SBLASTMBUFCHK(&so->so_rcv);
2263 if (flags & MSG_PEEK)
2267 if (flags & MSG_DONTWAIT) {
2268 *mp = m_copym(m, 0, len,
2272 * m_copym() couldn't
2274 * Adjust uio_resid back
2276 * down by len bytes,
2277 * which we didn't end
2278 * up "copying" over).
2280 uio->uio_resid += len;
2284 SOCKBUF_UNLOCK(&so->so_rcv);
2285 *mp = m_copym(m, 0, len,
2287 SOCKBUF_LOCK(&so->so_rcv);
2290 sbcut_locked(&so->so_rcv, len);
2293 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2294 if (so->so_oobmark) {
2295 if ((flags & MSG_PEEK) == 0) {
2296 so->so_oobmark -= len;
2297 if (so->so_oobmark == 0) {
2298 so->so_rcv.sb_state |= SBS_RCVATMARK;
2303 if (offset == so->so_oobmark)
2307 if (flags & MSG_EOR)
2310 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2311 * must not quit until "uio->uio_resid == 0" or an error
2312 * termination. If a signal/timeout occurs, return with a
2313 * short count but without error. Keep sockbuf locked
2314 * against other readers.
2316 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2317 !sosendallatonce(so) && nextrecord == NULL) {
2318 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2319 if (so->so_error || so->so_rerror ||
2320 so->so_rcv.sb_state & SBS_CANTRCVMORE)
2323 * Notify the protocol that some data has been
2324 * drained before blocking.
2326 if (pr->pr_flags & PR_WANTRCVD) {
2327 SOCKBUF_UNLOCK(&so->so_rcv);
2329 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2330 SOCKBUF_LOCK(&so->so_rcv);
2332 SBLASTRECORDCHK(&so->so_rcv);
2333 SBLASTMBUFCHK(&so->so_rcv);
2335 * We could receive some data while was notifying
2336 * the protocol. Skip blocking in this case.
2338 if (so->so_rcv.sb_mb == NULL) {
2339 error = sbwait(&so->so_rcv);
2341 SOCKBUF_UNLOCK(&so->so_rcv);
2345 m = so->so_rcv.sb_mb;
2347 nextrecord = m->m_nextpkt;
2351 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2352 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2353 if (report_real_len)
2354 uio->uio_resid -= m_length(m, NULL) - moff;
2356 if ((flags & MSG_PEEK) == 0)
2357 (void) sbdroprecord_locked(&so->so_rcv);
2359 if ((flags & MSG_PEEK) == 0) {
2362 * First part is an inline SB_EMPTY_FIXUP(). Second
2363 * part makes sure sb_lastrecord is up-to-date if
2364 * there is still data in the socket buffer.
2366 so->so_rcv.sb_mb = nextrecord;
2367 if (so->so_rcv.sb_mb == NULL) {
2368 so->so_rcv.sb_mbtail = NULL;
2369 so->so_rcv.sb_lastrecord = NULL;
2370 } else if (nextrecord->m_nextpkt == NULL)
2371 so->so_rcv.sb_lastrecord = nextrecord;
2373 SBLASTRECORDCHK(&so->so_rcv);
2374 SBLASTMBUFCHK(&so->so_rcv);
2376 * If soreceive() is being done from the socket callback,
2377 * then don't need to generate ACK to peer to update window,
2378 * since ACK will be generated on return to TCP.
2380 if (!(flags & MSG_SOCALLBCK) &&
2381 (pr->pr_flags & PR_WANTRCVD)) {
2382 SOCKBUF_UNLOCK(&so->so_rcv);
2384 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2385 SOCKBUF_LOCK(&so->so_rcv);
2388 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2389 if (orig_resid == uio->uio_resid && orig_resid &&
2390 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2391 SOCKBUF_UNLOCK(&so->so_rcv);
2394 SOCKBUF_UNLOCK(&so->so_rcv);
2399 SOCK_IO_RECV_UNLOCK(so);
2404 * Optimized version of soreceive() for stream (TCP) sockets.
2407 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2408 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2410 int len = 0, error = 0, flags, oresid;
2412 struct mbuf *m, *n = NULL;
2414 /* We only do stream sockets. */
2415 if (so->so_type != SOCK_STREAM)
2420 flags = *flagsp &~ MSG_EOR;
2423 if (controlp != NULL)
2425 if (flags & MSG_OOB)
2426 return (soreceive_rcvoob(so, uio, flags));
2434 * KTLS store TLS records as records with a control message to
2435 * describe the framing.
2437 * We check once here before acquiring locks to optimize the
2440 if (sb->sb_tls_info != NULL)
2441 return (soreceive_generic(so, psa, uio, mp0, controlp,
2445 /* Prevent other readers from entering the socket. */
2446 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2452 if (sb->sb_tls_info != NULL) {
2454 SOCK_IO_RECV_UNLOCK(so);
2455 return (soreceive_generic(so, psa, uio, mp0, controlp,
2460 /* Easy one, no space to copyout anything. */
2461 if (uio->uio_resid == 0) {
2465 oresid = uio->uio_resid;
2467 /* We will never ever get anything unless we are or were connected. */
2468 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2474 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2476 /* Abort if socket has reported problems. */
2478 if (sbavail(sb) > 0)
2480 if (oresid > uio->uio_resid)
2482 error = so->so_error;
2483 if (!(flags & MSG_PEEK))
2488 /* Door is closed. Deliver what is left, if any. */
2489 if (sb->sb_state & SBS_CANTRCVMORE) {
2490 if (sbavail(sb) > 0)
2496 /* Socket buffer is empty and we shall not block. */
2497 if (sbavail(sb) == 0 &&
2498 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2503 /* Socket buffer got some data that we shall deliver now. */
2504 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2505 ((so->so_state & SS_NBIO) ||
2506 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2507 sbavail(sb) >= sb->sb_lowat ||
2508 sbavail(sb) >= uio->uio_resid ||
2509 sbavail(sb) >= sb->sb_hiwat) ) {
2513 /* On MSG_WAITALL we must wait until all data or error arrives. */
2514 if ((flags & MSG_WAITALL) &&
2515 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2519 * Wait and block until (more) data comes in.
2520 * NB: Drops the sockbuf lock during wait.
2528 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2529 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2530 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2534 uio->uio_td->td_ru.ru_msgrcv++;
2536 /* Fill uio until full or current end of socket buffer is reached. */
2537 len = min(uio->uio_resid, sbavail(sb));
2539 /* Dequeue as many mbufs as possible. */
2540 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2544 m_cat(*mp0, sb->sb_mb);
2546 m != NULL && m->m_len <= len;
2548 KASSERT(!(m->m_flags & M_NOTAVAIL),
2549 ("%s: m %p not available", __func__, m));
2551 uio->uio_resid -= m->m_len;
2557 sb->sb_lastrecord = sb->sb_mb;
2558 if (sb->sb_mb == NULL)
2561 /* Copy the remainder. */
2563 KASSERT(sb->sb_mb != NULL,
2564 ("%s: len > 0 && sb->sb_mb empty", __func__));
2566 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2568 len = 0; /* Don't flush data from sockbuf. */
2570 uio->uio_resid -= len;
2581 /* NB: Must unlock socket buffer as uiomove may sleep. */
2583 error = m_mbuftouio(uio, sb->sb_mb, len);
2588 SBLASTRECORDCHK(sb);
2592 * Remove the delivered data from the socket buffer unless we
2593 * were only peeking.
2595 if (!(flags & MSG_PEEK)) {
2597 sbdrop_locked(sb, len);
2599 /* Notify protocol that we drained some data. */
2600 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2601 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2602 !(flags & MSG_SOCALLBCK))) {
2605 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2611 * For MSG_WAITALL we may have to loop again and wait for
2612 * more data to come in.
2614 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2617 SBLASTRECORDCHK(sb);
2620 SOCK_IO_RECV_UNLOCK(so);
2625 * Optimized version of soreceive() for simple datagram cases from userspace.
2626 * Unlike in the stream case, we're able to drop a datagram if copyout()
2627 * fails, and because we handle datagrams atomically, we don't need to use a
2628 * sleep lock to prevent I/O interlacing.
2631 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2632 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2634 struct mbuf *m, *m2;
2637 struct protosw *pr = so->so_proto;
2638 struct mbuf *nextrecord;
2642 if (controlp != NULL)
2645 flags = *flagsp &~ MSG_EOR;
2650 * For any complicated cases, fall back to the full
2651 * soreceive_generic().
2653 if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC)))
2654 return (soreceive_generic(so, psa, uio, mp0, controlp,
2658 * Enforce restrictions on use.
2660 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2661 ("soreceive_dgram: wantrcvd"));
2662 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2663 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2664 ("soreceive_dgram: SBS_RCVATMARK"));
2665 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2666 ("soreceive_dgram: P_CONNREQUIRED"));
2669 * Loop blocking while waiting for a datagram.
2671 SOCKBUF_LOCK(&so->so_rcv);
2672 while ((m = so->so_rcv.sb_mb) == NULL) {
2673 KASSERT(sbavail(&so->so_rcv) == 0,
2674 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2675 sbavail(&so->so_rcv)));
2677 error = so->so_error;
2679 SOCKBUF_UNLOCK(&so->so_rcv);
2682 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2683 uio->uio_resid == 0) {
2684 SOCKBUF_UNLOCK(&so->so_rcv);
2687 if ((so->so_state & SS_NBIO) ||
2688 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2689 SOCKBUF_UNLOCK(&so->so_rcv);
2690 return (EWOULDBLOCK);
2692 SBLASTRECORDCHK(&so->so_rcv);
2693 SBLASTMBUFCHK(&so->so_rcv);
2694 error = sbwait(&so->so_rcv);
2696 SOCKBUF_UNLOCK(&so->so_rcv);
2700 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2703 uio->uio_td->td_ru.ru_msgrcv++;
2704 SBLASTRECORDCHK(&so->so_rcv);
2705 SBLASTMBUFCHK(&so->so_rcv);
2706 nextrecord = m->m_nextpkt;
2707 if (nextrecord == NULL) {
2708 KASSERT(so->so_rcv.sb_lastrecord == m,
2709 ("soreceive_dgram: lastrecord != m"));
2712 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2713 ("soreceive_dgram: m_nextpkt != nextrecord"));
2716 * Pull 'm' and its chain off the front of the packet queue.
2718 so->so_rcv.sb_mb = NULL;
2719 sockbuf_pushsync(&so->so_rcv, nextrecord);
2722 * Walk 'm's chain and free that many bytes from the socket buffer.
2724 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2725 sbfree(&so->so_rcv, m2);
2728 * Do a few last checks before we let go of the lock.
2730 SBLASTRECORDCHK(&so->so_rcv);
2731 SBLASTMBUFCHK(&so->so_rcv);
2732 SOCKBUF_UNLOCK(&so->so_rcv);
2734 if (pr->pr_flags & PR_ADDR) {
2735 KASSERT(m->m_type == MT_SONAME,
2736 ("m->m_type == %d", m->m_type));
2738 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2743 /* XXXRW: Can this happen? */
2748 * Packet to copyout() is now in 'm' and it is disconnected from the
2751 * Process one or more MT_CONTROL mbufs present before any data mbufs
2752 * in the first mbuf chain on the socket buffer. We call into the
2753 * protocol to perform externalization (or freeing if controlp ==
2754 * NULL). In some cases there can be only MT_CONTROL mbufs without
2757 if (m->m_type == MT_CONTROL) {
2758 struct mbuf *cm = NULL, *cmn;
2759 struct mbuf **cme = &cm;
2765 cme = &(*cme)->m_next;
2767 } while (m != NULL && m->m_type == MT_CONTROL);
2768 while (cm != NULL) {
2771 if (pr->pr_domain->dom_externalize != NULL) {
2772 error = (*pr->pr_domain->dom_externalize)
2773 (cm, controlp, flags);
2774 } else if (controlp != NULL)
2778 if (controlp != NULL) {
2779 while (*controlp != NULL)
2780 controlp = &(*controlp)->m_next;
2785 KASSERT(m == NULL || m->m_type == MT_DATA,
2786 ("soreceive_dgram: !data"));
2787 while (m != NULL && uio->uio_resid > 0) {
2788 len = uio->uio_resid;
2791 error = uiomove(mtod(m, char *), (int)len, uio);
2796 if (len == m->m_len)
2813 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2814 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2818 CURVNET_SET(so->so_vnet);
2819 if (!SOLISTENING(so))
2820 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2821 mp0, controlp, flagsp));
2829 soshutdown(struct socket *so, int how)
2831 struct protosw *pr = so->so_proto;
2832 int error, soerror_enotconn;
2834 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2837 soerror_enotconn = 0;
2839 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2841 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2842 * invoked on a datagram sockets, however historically we would
2843 * actually tear socket down. This is known to be leveraged by
2844 * some applications to unblock process waiting in recvXXX(2)
2845 * by other process that it shares that socket with. Try to meet
2846 * both backward-compatibility and POSIX requirements by forcing
2847 * ENOTCONN but still asking protocol to perform pru_shutdown().
2849 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so))
2851 soerror_enotconn = 1;
2854 if (SOLISTENING(so)) {
2855 if (how != SHUT_WR) {
2857 so->so_error = ECONNABORTED;
2858 solisten_wakeup(so); /* unlocks so */
2863 CURVNET_SET(so->so_vnet);
2864 if (pr->pr_usrreqs->pru_flush != NULL)
2865 (*pr->pr_usrreqs->pru_flush)(so, how);
2868 if (how != SHUT_RD) {
2869 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2870 wakeup(&so->so_timeo);
2872 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2874 wakeup(&so->so_timeo);
2878 return (soerror_enotconn ? ENOTCONN : 0);
2882 sorflush(struct socket *so)
2884 struct sockbuf *sb = &so->so_rcv;
2885 struct protosw *pr = so->so_proto;
2892 * In order to avoid calling dom_dispose with the socket buffer mutex
2893 * held, and in order to generally avoid holding the lock for a long
2894 * time, we make a copy of the socket buffer and clear the original
2895 * (except locks, state). The new socket buffer copy won't have
2896 * initialized locks so we can only call routines that won't use or
2897 * assert those locks.
2899 * Dislodge threads currently blocked in receive and wait to acquire
2900 * a lock against other simultaneous readers before clearing the
2901 * socket buffer. Don't let our acquire be interrupted by a signal
2902 * despite any existing socket disposition on interruptable waiting.
2905 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
2906 KASSERT(error == 0, ("%s: cannot lock sock %p recv buffer",
2910 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2911 * and mutex data unchanged.
2914 bzero(&aso, sizeof(aso));
2915 aso.so_pcb = so->so_pcb;
2916 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2917 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2918 bzero(&sb->sb_startzero,
2919 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2921 SOCK_IO_RECV_UNLOCK(so);
2924 * Dispose of special rights and flush the copied socket. Don't call
2925 * any unsafe routines (that rely on locks being initialized) on aso.
2927 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2928 (*pr->pr_domain->dom_dispose)(&aso);
2929 sbrelease_internal(&aso.so_rcv, so);
2933 * Wrapper for Socket established helper hook.
2934 * Parameters: socket, context of the hook point, hook id.
2937 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2939 struct socket_hhook_data hhook_data = {
2946 CURVNET_SET(so->so_vnet);
2947 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2950 /* Ugly but needed, since hhooks return void for now */
2951 return (hhook_data.status);
2955 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2956 * additional variant to handle the case where the option value needs to be
2957 * some kind of integer, but not a specific size. In addition to their use
2958 * here, these functions are also called by the protocol-level pr_ctloutput()
2962 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2967 * If the user gives us more than we wanted, we ignore it, but if we
2968 * don't get the minimum length the caller wants, we return EINVAL.
2969 * On success, sopt->sopt_valsize is set to however much we actually
2972 if ((valsize = sopt->sopt_valsize) < minlen)
2975 sopt->sopt_valsize = valsize = len;
2977 if (sopt->sopt_td != NULL)
2978 return (copyin(sopt->sopt_val, buf, valsize));
2980 bcopy(sopt->sopt_val, buf, valsize);
2984 u_long nl_maxsockbuf = 512 * 1024 * 1024; /* 512M, XXX: init based on physmem */
2987 sogetmaxbuf(struct socket *so)
2989 if (so->so_proto->pr_domain->dom_family != PF_NETLINK)
2991 return ((priv_check(curthread, PRIV_NET_ROUTE) == 0) ? nl_maxsockbuf : sb_max);
2995 * Kernel version of setsockopt(2).
2997 * XXX: optlen is size_t, not socklen_t
3000 so_setsockopt(struct socket *so, int level, int optname, void *optval,
3003 struct sockopt sopt;
3005 sopt.sopt_level = level;
3006 sopt.sopt_name = optname;
3007 sopt.sopt_dir = SOPT_SET;
3008 sopt.sopt_val = optval;
3009 sopt.sopt_valsize = optlen;
3010 sopt.sopt_td = NULL;
3011 return (sosetopt(so, &sopt));
3015 sosetopt(struct socket *so, struct sockopt *sopt)
3020 sbintime_t val, *valp;
3026 CURVNET_SET(so->so_vnet);
3028 if (sopt->sopt_level != SOL_SOCKET) {
3029 if (so->so_proto->pr_ctloutput != NULL)
3030 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3032 error = ENOPROTOOPT;
3034 switch (sopt->sopt_name) {
3035 case SO_ACCEPTFILTER:
3036 error = accept_filt_setopt(so, sopt);
3042 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3045 if (l.l_linger < 0 ||
3046 l.l_linger > USHRT_MAX ||
3047 l.l_linger > (INT_MAX / hz)) {
3052 so->so_linger = l.l_linger;
3054 so->so_options |= SO_LINGER;
3056 so->so_options &= ~SO_LINGER;
3063 case SO_USELOOPBACK:
3067 case SO_REUSEPORT_LB:
3075 error = sooptcopyin(sopt, &optval, sizeof optval,
3081 so->so_options |= sopt->sopt_name;
3083 so->so_options &= ~sopt->sopt_name;
3088 error = sooptcopyin(sopt, &optval, sizeof optval,
3093 if (optval < 0 || optval >= rt_numfibs) {
3097 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3098 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3099 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3100 so->so_fibnum = optval;
3105 case SO_USER_COOKIE:
3106 error = sooptcopyin(sopt, &val32, sizeof val32,
3110 so->so_user_cookie = val32;
3117 error = sooptcopyin(sopt, &optval, sizeof optval,
3123 * Values < 1 make no sense for any of these options,
3131 error = sbsetopt(so, sopt->sopt_name, optval);
3136 #ifdef COMPAT_FREEBSD32
3137 if (SV_CURPROC_FLAG(SV_ILP32)) {
3138 struct timeval32 tv32;
3140 error = sooptcopyin(sopt, &tv32, sizeof tv32,
3142 CP(tv32, tv, tv_sec);
3143 CP(tv32, tv, tv_usec);
3146 error = sooptcopyin(sopt, &tv, sizeof tv,
3150 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3151 tv.tv_usec >= 1000000) {
3155 if (tv.tv_sec > INT32_MAX)
3160 valp = sopt->sopt_name == SO_SNDTIMEO ?
3161 (SOLISTENING(so) ? &so->sol_sbsnd_timeo :
3162 &so->so_snd.sb_timeo) :
3163 (SOLISTENING(so) ? &so->sol_sbrcv_timeo :
3164 &so->so_rcv.sb_timeo);
3171 error = sooptcopyin(sopt, &extmac, sizeof extmac,
3175 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3183 error = sooptcopyin(sopt, &optval, sizeof optval,
3187 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3191 so->so_ts_clock = optval;
3194 case SO_MAX_PACING_RATE:
3195 error = sooptcopyin(sopt, &val32, sizeof(val32),
3199 so->so_max_pacing_rate = val32;
3203 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3204 error = hhook_run_socket(so, sopt,
3207 error = ENOPROTOOPT;
3210 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3211 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
3219 * Helper routine for getsockopt.
3222 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3230 * Documented get behavior is that we always return a value, possibly
3231 * truncated to fit in the user's buffer. Traditional behavior is
3232 * that we always tell the user precisely how much we copied, rather
3233 * than something useful like the total amount we had available for
3234 * her. Note that this interface is not idempotent; the entire
3235 * answer must be generated ahead of time.
3237 valsize = min(len, sopt->sopt_valsize);
3238 sopt->sopt_valsize = valsize;
3239 if (sopt->sopt_val != NULL) {
3240 if (sopt->sopt_td != NULL)
3241 error = copyout(buf, sopt->sopt_val, valsize);
3243 bcopy(buf, sopt->sopt_val, valsize);
3249 sogetopt(struct socket *so, struct sockopt *sopt)
3258 CURVNET_SET(so->so_vnet);
3260 if (sopt->sopt_level != SOL_SOCKET) {
3261 if (so->so_proto->pr_ctloutput != NULL)
3262 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3264 error = ENOPROTOOPT;
3268 switch (sopt->sopt_name) {
3269 case SO_ACCEPTFILTER:
3270 error = accept_filt_getopt(so, sopt);
3275 l.l_onoff = so->so_options & SO_LINGER;
3276 l.l_linger = so->so_linger;
3278 error = sooptcopyout(sopt, &l, sizeof l);
3281 case SO_USELOOPBACK:
3287 case SO_REUSEPORT_LB:
3297 optval = so->so_options & sopt->sopt_name;
3299 error = sooptcopyout(sopt, &optval, sizeof optval);
3303 optval = so->so_proto->pr_domain->dom_family;
3307 optval = so->so_type;
3311 optval = so->so_proto->pr_protocol;
3317 optval = so->so_error;
3320 optval = so->so_rerror;
3327 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3328 so->so_snd.sb_hiwat;
3332 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3333 so->so_rcv.sb_hiwat;
3337 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3338 so->so_snd.sb_lowat;
3342 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3343 so->so_rcv.sb_lowat;
3349 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3350 (SOLISTENING(so) ? so->sol_sbsnd_timeo :
3351 so->so_snd.sb_timeo) :
3352 (SOLISTENING(so) ? so->sol_sbrcv_timeo :
3353 so->so_rcv.sb_timeo));
3355 #ifdef COMPAT_FREEBSD32
3356 if (SV_CURPROC_FLAG(SV_ILP32)) {
3357 struct timeval32 tv32;
3359 CP(tv, tv32, tv_sec);
3360 CP(tv, tv32, tv_usec);
3361 error = sooptcopyout(sopt, &tv32, sizeof tv32);
3364 error = sooptcopyout(sopt, &tv, sizeof tv);
3369 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3373 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3377 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3385 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3389 error = mac_getsockopt_peerlabel(
3390 sopt->sopt_td->td_ucred, so, &extmac);
3393 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3399 case SO_LISTENQLIMIT:
3400 optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3404 optval = SOLISTENING(so) ? so->sol_qlen : 0;
3407 case SO_LISTENINCQLEN:
3408 optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3412 optval = so->so_ts_clock;
3415 case SO_MAX_PACING_RATE:
3416 optval = so->so_max_pacing_rate;
3420 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3421 error = hhook_run_socket(so, sopt,
3424 error = ENOPROTOOPT;
3436 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3438 struct mbuf *m, *m_prev;
3439 int sopt_size = sopt->sopt_valsize;
3441 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3444 if (sopt_size > MLEN) {
3445 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3446 if ((m->m_flags & M_EXT) == 0) {
3450 m->m_len = min(MCLBYTES, sopt_size);
3452 m->m_len = min(MLEN, sopt_size);
3454 sopt_size -= m->m_len;
3459 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3464 if (sopt_size > MLEN) {
3465 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3467 if ((m->m_flags & M_EXT) == 0) {
3472 m->m_len = min(MCLBYTES, sopt_size);
3474 m->m_len = min(MLEN, sopt_size);
3476 sopt_size -= m->m_len;
3484 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3486 struct mbuf *m0 = m;
3488 if (sopt->sopt_val == NULL)
3490 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3491 if (sopt->sopt_td != NULL) {
3494 error = copyin(sopt->sopt_val, mtod(m, char *),
3501 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3502 sopt->sopt_valsize -= m->m_len;
3503 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3506 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3507 panic("ip6_sooptmcopyin");
3512 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3514 struct mbuf *m0 = m;
3517 if (sopt->sopt_val == NULL)
3519 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3520 if (sopt->sopt_td != NULL) {
3523 error = copyout(mtod(m, char *), sopt->sopt_val,
3530 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3531 sopt->sopt_valsize -= m->m_len;
3532 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3533 valsize += m->m_len;
3537 /* enough soopt buffer should be given from user-land */
3541 sopt->sopt_valsize = valsize;
3546 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3547 * out-of-band data, which will then notify socket consumers.
3550 sohasoutofband(struct socket *so)
3553 if (so->so_sigio != NULL)
3554 pgsigio(&so->so_sigio, SIGURG, 0);
3555 selwakeuppri(&so->so_rdsel, PSOCK);
3559 sopoll(struct socket *so, int events, struct ucred *active_cred,
3564 * We do not need to set or assert curvnet as long as everyone uses
3567 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3572 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3578 if (SOLISTENING(so)) {
3579 if (!(events & (POLLIN | POLLRDNORM)))
3581 else if (!TAILQ_EMPTY(&so->sol_comp))
3582 revents = events & (POLLIN | POLLRDNORM);
3583 else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3584 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3586 selrecord(td, &so->so_rdsel);
3591 SOCKBUF_LOCK(&so->so_snd);
3592 SOCKBUF_LOCK(&so->so_rcv);
3593 if (events & (POLLIN | POLLRDNORM))
3594 if (soreadabledata(so))
3595 revents |= events & (POLLIN | POLLRDNORM);
3596 if (events & (POLLOUT | POLLWRNORM))
3597 if (sowriteable(so))
3598 revents |= events & (POLLOUT | POLLWRNORM);
3599 if (events & (POLLPRI | POLLRDBAND))
3600 if (so->so_oobmark ||
3601 (so->so_rcv.sb_state & SBS_RCVATMARK))
3602 revents |= events & (POLLPRI | POLLRDBAND);
3603 if ((events & POLLINIGNEOF) == 0) {
3604 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3605 revents |= events & (POLLIN | POLLRDNORM);
3606 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3610 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3611 revents |= events & POLLRDHUP;
3614 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
3615 selrecord(td, &so->so_rdsel);
3616 so->so_rcv.sb_flags |= SB_SEL;
3618 if (events & (POLLOUT | POLLWRNORM)) {
3619 selrecord(td, &so->so_wrsel);
3620 so->so_snd.sb_flags |= SB_SEL;
3623 SOCKBUF_UNLOCK(&so->so_rcv);
3624 SOCKBUF_UNLOCK(&so->so_snd);
3631 soo_kqfilter(struct file *fp, struct knote *kn)
3633 struct socket *so = kn->kn_fp->f_data;
3637 switch (kn->kn_filter) {
3639 kn->kn_fop = &soread_filtops;
3640 knl = &so->so_rdsel.si_note;
3644 kn->kn_fop = &sowrite_filtops;
3645 knl = &so->so_wrsel.si_note;
3649 kn->kn_fop = &soempty_filtops;
3650 knl = &so->so_wrsel.si_note;
3658 if (SOLISTENING(so)) {
3659 knlist_add(knl, kn, 1);
3662 knlist_add(knl, kn, 1);
3663 sb->sb_flags |= SB_KNOTE;
3671 * Some routines that return EOPNOTSUPP for entry points that are not
3672 * supported by a protocol. Fill in as needed.
3675 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3682 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3689 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3696 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3703 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3711 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3718 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3726 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3733 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3734 struct ifnet *ifp, struct thread *td)
3741 pru_disconnect_notsupp(struct socket *so)
3748 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3755 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3762 pru_rcvd_notsupp(struct socket *so, int flags)
3769 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3776 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3777 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3780 if (control != NULL)
3782 if ((flags & PRUS_NOTREADY) == 0)
3784 return (EOPNOTSUPP);
3788 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3791 return (EOPNOTSUPP);
3795 * This isn't really a ``null'' operation, but it's the default one and
3796 * doesn't do anything destructive.
3799 pru_sense_null(struct socket *so, struct stat *sb)
3802 sb->st_blksize = so->so_snd.sb_hiwat;
3807 pru_shutdown_notsupp(struct socket *so)
3814 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3821 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3822 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3829 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3830 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3837 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3845 filt_sordetach(struct knote *kn)
3847 struct socket *so = kn->kn_fp->f_data;
3850 knlist_remove(&so->so_rdsel.si_note, kn, 1);
3851 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3852 so->so_rcv.sb_flags &= ~SB_KNOTE;
3853 so_rdknl_unlock(so);
3858 filt_soread(struct knote *kn, long hint)
3862 so = kn->kn_fp->f_data;
3864 if (SOLISTENING(so)) {
3865 SOCK_LOCK_ASSERT(so);
3866 kn->kn_data = so->sol_qlen;
3868 kn->kn_flags |= EV_EOF;
3869 kn->kn_fflags = so->so_error;
3872 return (!TAILQ_EMPTY(&so->sol_comp));
3875 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3877 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3878 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3879 kn->kn_flags |= EV_EOF;
3880 kn->kn_fflags = so->so_error;
3882 } else if (so->so_error || so->so_rerror)
3885 if (kn->kn_sfflags & NOTE_LOWAT) {
3886 if (kn->kn_data >= kn->kn_sdata)
3888 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3891 /* This hook returning non-zero indicates an event, not error */
3892 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3896 filt_sowdetach(struct knote *kn)
3898 struct socket *so = kn->kn_fp->f_data;
3901 knlist_remove(&so->so_wrsel.si_note, kn, 1);
3902 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3903 so->so_snd.sb_flags &= ~SB_KNOTE;
3904 so_wrknl_unlock(so);
3909 filt_sowrite(struct knote *kn, long hint)
3913 so = kn->kn_fp->f_data;
3915 if (SOLISTENING(so))
3918 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3919 kn->kn_data = sbspace(&so->so_snd);
3921 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3923 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3924 kn->kn_flags |= EV_EOF;
3925 kn->kn_fflags = so->so_error;
3927 } else if (so->so_error) /* temporary udp error */
3929 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3930 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3932 else if (kn->kn_sfflags & NOTE_LOWAT)
3933 return (kn->kn_data >= kn->kn_sdata);
3935 return (kn->kn_data >= so->so_snd.sb_lowat);
3939 filt_soempty(struct knote *kn, long hint)
3943 so = kn->kn_fp->f_data;
3945 if (SOLISTENING(so))
3948 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3949 kn->kn_data = sbused(&so->so_snd);
3951 if (kn->kn_data == 0)
3958 socheckuid(struct socket *so, uid_t uid)
3963 if (so->so_cred->cr_uid != uid)
3969 * These functions are used by protocols to notify the socket layer (and its
3970 * consumers) of state changes in the sockets driven by protocol-side events.
3974 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3976 * Normal sequence from the active (originating) side is that
3977 * soisconnecting() is called during processing of connect() call, resulting
3978 * in an eventual call to soisconnected() if/when the connection is
3979 * established. When the connection is torn down soisdisconnecting() is
3980 * called during processing of disconnect() call, and soisdisconnected() is
3981 * called when the connection to the peer is totally severed. The semantics
3982 * of these routines are such that connectionless protocols can call
3983 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3984 * calls when setting up a ``connection'' takes no time.
3986 * From the passive side, a socket is created with two queues of sockets:
3987 * so_incomp for connections in progress and so_comp for connections already
3988 * made and awaiting user acceptance. As a protocol is preparing incoming
3989 * connections, it creates a socket structure queued on so_incomp by calling
3990 * sonewconn(). When the connection is established, soisconnected() is
3991 * called, and transfers the socket structure to so_comp, making it available
3994 * If a socket is closed with sockets on either so_incomp or so_comp, these
3995 * sockets are dropped.
3997 * If higher-level protocols are implemented in the kernel, the wakeups done
3998 * here will sometimes cause software-interrupt process scheduling.
4001 soisconnecting(struct socket *so)
4005 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
4006 so->so_state |= SS_ISCONNECTING;
4011 soisconnected(struct socket *so)
4013 bool last __diagused;
4016 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
4017 so->so_state |= SS_ISCONNECTED;
4019 if (so->so_qstate == SQ_INCOMP) {
4020 struct socket *head = so->so_listen;
4023 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
4025 * Promoting a socket from incomplete queue to complete, we
4026 * need to go through reverse order of locking. We first do
4027 * trylock, and if that doesn't succeed, we go the hard way
4028 * leaving a reference and rechecking consistency after proper
4031 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
4034 SOLISTEN_LOCK(head);
4036 if (__predict_false(head != so->so_listen)) {
4038 * The socket went off the listen queue,
4039 * should be lost race to close(2) of sol.
4040 * The socket is about to soabort().
4046 last = refcount_release(&head->so_count);
4047 KASSERT(!last, ("%s: released last reference for %p",
4051 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4052 TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4053 head->sol_incqlen--;
4054 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4056 so->so_qstate = SQ_COMP;
4058 solisten_wakeup(head); /* unlocks */
4060 SOCKBUF_LOCK(&so->so_rcv);
4061 soupcall_set(so, SO_RCV,
4062 head->sol_accept_filter->accf_callback,
4063 head->sol_accept_filter_arg);
4064 so->so_options &= ~SO_ACCEPTFILTER;
4065 ret = head->sol_accept_filter->accf_callback(so,
4066 head->sol_accept_filter_arg, M_NOWAIT);
4067 if (ret == SU_ISCONNECTED) {
4068 soupcall_clear(so, SO_RCV);
4069 SOCKBUF_UNLOCK(&so->so_rcv);
4072 SOCKBUF_UNLOCK(&so->so_rcv);
4074 SOLISTEN_UNLOCK(head);
4079 wakeup(&so->so_timeo);
4085 soisdisconnecting(struct socket *so)
4089 so->so_state &= ~SS_ISCONNECTING;
4090 so->so_state |= SS_ISDISCONNECTING;
4092 if (!SOLISTENING(so)) {
4093 SOCKBUF_LOCK(&so->so_rcv);
4094 socantrcvmore_locked(so);
4095 SOCKBUF_LOCK(&so->so_snd);
4096 socantsendmore_locked(so);
4099 wakeup(&so->so_timeo);
4103 soisdisconnected(struct socket *so)
4109 * There is at least one reader of so_state that does not
4110 * acquire socket lock, namely soreceive_generic(). Ensure
4111 * that it never sees all flags that track connection status
4112 * cleared, by ordering the update with a barrier semantic of
4113 * our release thread fence.
4115 so->so_state |= SS_ISDISCONNECTED;
4116 atomic_thread_fence_rel();
4117 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4119 if (!SOLISTENING(so)) {
4121 SOCKBUF_LOCK(&so->so_rcv);
4122 socantrcvmore_locked(so);
4123 SOCKBUF_LOCK(&so->so_snd);
4124 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4125 socantsendmore_locked(so);
4128 wakeup(&so->so_timeo);
4132 soiolock(struct socket *so, struct sx *sx, int flags)
4136 KASSERT((flags & SBL_VALID) == flags,
4137 ("soiolock: invalid flags %#x", flags));
4139 if ((flags & SBL_WAIT) != 0) {
4140 if ((flags & SBL_NOINTR) != 0) {
4143 error = sx_xlock_sig(sx);
4147 } else if (!sx_try_xlock(sx)) {
4148 return (EWOULDBLOCK);
4151 if (__predict_false(SOLISTENING(so))) {
4159 soiounlock(struct sx *sx)
4165 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4168 sodupsockaddr(const struct sockaddr *sa, int mflags)
4170 struct sockaddr *sa2;
4172 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4174 bcopy(sa, sa2, sa->sa_len);
4179 * Register per-socket destructor.
4182 sodtor_set(struct socket *so, so_dtor_t *func)
4185 SOCK_LOCK_ASSERT(so);
4190 * Register per-socket buffer upcalls.
4193 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4197 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4207 panic("soupcall_set: bad which");
4209 SOCKBUF_LOCK_ASSERT(sb);
4210 sb->sb_upcall = func;
4211 sb->sb_upcallarg = arg;
4212 sb->sb_flags |= SB_UPCALL;
4216 soupcall_clear(struct socket *so, int which)
4220 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4230 panic("soupcall_clear: bad which");
4232 SOCKBUF_LOCK_ASSERT(sb);
4233 KASSERT(sb->sb_upcall != NULL,
4234 ("%s: so %p no upcall to clear", __func__, so));
4235 sb->sb_upcall = NULL;
4236 sb->sb_upcallarg = NULL;
4237 sb->sb_flags &= ~SB_UPCALL;
4241 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4244 SOLISTEN_LOCK_ASSERT(so);
4245 so->sol_upcall = func;
4246 so->sol_upcallarg = arg;
4250 so_rdknl_lock(void *arg)
4252 struct socket *so = arg;
4254 if (SOLISTENING(so))
4257 SOCKBUF_LOCK(&so->so_rcv);
4261 so_rdknl_unlock(void *arg)
4263 struct socket *so = arg;
4265 if (SOLISTENING(so))
4268 SOCKBUF_UNLOCK(&so->so_rcv);
4272 so_rdknl_assert_lock(void *arg, int what)
4274 struct socket *so = arg;
4276 if (what == LA_LOCKED) {
4277 if (SOLISTENING(so))
4278 SOCK_LOCK_ASSERT(so);
4280 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4282 if (SOLISTENING(so))
4283 SOCK_UNLOCK_ASSERT(so);
4285 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4290 so_wrknl_lock(void *arg)
4292 struct socket *so = arg;
4294 if (SOLISTENING(so))
4297 SOCKBUF_LOCK(&so->so_snd);
4301 so_wrknl_unlock(void *arg)
4303 struct socket *so = arg;
4305 if (SOLISTENING(so))
4308 SOCKBUF_UNLOCK(&so->so_snd);
4312 so_wrknl_assert_lock(void *arg, int what)
4314 struct socket *so = arg;
4316 if (what == LA_LOCKED) {
4317 if (SOLISTENING(so))
4318 SOCK_LOCK_ASSERT(so);
4320 SOCKBUF_LOCK_ASSERT(&so->so_snd);
4322 if (SOLISTENING(so))
4323 SOCK_UNLOCK_ASSERT(so);
4325 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4330 * Create an external-format (``xsocket'') structure using the information in
4331 * the kernel-format socket structure pointed to by so. This is done to
4332 * reduce the spew of irrelevant information over this interface, to isolate
4333 * user code from changes in the kernel structure, and potentially to provide
4334 * information-hiding if we decide that some of this information should be
4335 * hidden from users.
4338 sotoxsocket(struct socket *so, struct xsocket *xso)
4341 bzero(xso, sizeof(*xso));
4342 xso->xso_len = sizeof *xso;
4343 xso->xso_so = (uintptr_t)so;
4344 xso->so_type = so->so_type;
4345 xso->so_options = so->so_options;
4346 xso->so_linger = so->so_linger;
4347 xso->so_state = so->so_state;
4348 xso->so_pcb = (uintptr_t)so->so_pcb;
4349 xso->xso_protocol = so->so_proto->pr_protocol;
4350 xso->xso_family = so->so_proto->pr_domain->dom_family;
4351 xso->so_timeo = so->so_timeo;
4352 xso->so_error = so->so_error;
4353 xso->so_uid = so->so_cred->cr_uid;
4354 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4355 if (SOLISTENING(so)) {
4356 xso->so_qlen = so->sol_qlen;
4357 xso->so_incqlen = so->sol_incqlen;
4358 xso->so_qlimit = so->sol_qlimit;
4359 xso->so_oobmark = 0;
4361 xso->so_state |= so->so_qstate;
4362 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4363 xso->so_oobmark = so->so_oobmark;
4364 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4365 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4370 so_sockbuf_rcv(struct socket *so)
4373 return (&so->so_rcv);
4377 so_sockbuf_snd(struct socket *so)
4380 return (&so->so_snd);
4384 so_state_get(const struct socket *so)
4387 return (so->so_state);
4391 so_state_set(struct socket *so, int val)
4398 so_options_get(const struct socket *so)
4401 return (so->so_options);
4405 so_options_set(struct socket *so, int val)
4408 so->so_options = val;
4412 so_error_get(const struct socket *so)
4415 return (so->so_error);
4419 so_error_set(struct socket *so, int val)
4426 so_linger_get(const struct socket *so)
4429 return (so->so_linger);
4433 so_linger_set(struct socket *so, int val)
4436 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4437 ("%s: val %d out of range", __func__, val));
4439 so->so_linger = val;
4443 so_protosw_get(const struct socket *so)
4446 return (so->so_proto);
4450 so_protosw_set(struct socket *so, struct protosw *val)
4457 so_sorwakeup(struct socket *so)
4464 so_sowwakeup(struct socket *so)
4471 so_sorwakeup_locked(struct socket *so)
4474 sorwakeup_locked(so);
4478 so_sowwakeup_locked(struct socket *so)
4481 sowwakeup_locked(so);
4485 so_lock(struct socket *so)
4492 so_unlock(struct socket *so)