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/proc.h>
132 #include <sys/protosw.h>
133 #include <sys/sbuf.h>
134 #include <sys/socket.h>
135 #include <sys/socketvar.h>
136 #include <sys/resourcevar.h>
137 #include <net/route.h>
138 #include <sys/signalvar.h>
139 #include <sys/stat.h>
141 #include <sys/sysctl.h>
142 #include <sys/taskqueue.h>
145 #include <sys/unpcb.h>
146 #include <sys/jail.h>
147 #include <sys/syslog.h>
148 #include <netinet/in.h>
149 #include <netinet/in_pcb.h>
150 #include <netinet/tcp.h>
152 #include <net/vnet.h>
154 #include <security/mac/mac_framework.h>
158 #ifdef COMPAT_FREEBSD32
159 #include <sys/mount.h>
160 #include <sys/sysent.h>
161 #include <compat/freebsd32/freebsd32.h>
164 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
166 static void so_rdknl_lock(void *);
167 static void so_rdknl_unlock(void *);
168 static void so_rdknl_assert_lock(void *, int);
169 static void so_wrknl_lock(void *);
170 static void so_wrknl_unlock(void *);
171 static void so_wrknl_assert_lock(void *, int);
173 static void filt_sordetach(struct knote *kn);
174 static int filt_soread(struct knote *kn, long hint);
175 static void filt_sowdetach(struct knote *kn);
176 static int filt_sowrite(struct knote *kn, long hint);
177 static int filt_soempty(struct knote *kn, long hint);
178 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
179 fo_kqfilter_t soo_kqfilter;
181 static struct filterops soread_filtops = {
183 .f_detach = filt_sordetach,
184 .f_event = filt_soread,
186 static struct filterops sowrite_filtops = {
188 .f_detach = filt_sowdetach,
189 .f_event = filt_sowrite,
191 static struct filterops soempty_filtops = {
193 .f_detach = filt_sowdetach,
194 .f_event = filt_soempty,
197 so_gen_t so_gencnt; /* generation count for sockets */
199 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
200 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
202 #define VNET_SO_ASSERT(so) \
203 VNET_ASSERT(curvnet != NULL, \
204 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
206 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
207 #define V_socket_hhh VNET(socket_hhh)
210 * Limit on the number of connections in the listen queue waiting
212 * NB: The original sysctl somaxconn is still available but hidden
213 * to prevent confusion about the actual purpose of this number.
215 static u_int somaxconn = SOMAXCONN;
218 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
224 error = sysctl_handle_int(oidp, &val, 0, req);
225 if (error || !req->newptr )
229 * The purpose of the UINT_MAX / 3 limit, is so that the formula
231 * below, will not overflow.
234 if (val < 1 || val > UINT_MAX / 3)
240 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
241 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
242 sysctl_somaxconn, "I",
243 "Maximum listen socket pending connection accept queue size");
244 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
245 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, 0,
246 sizeof(int), sysctl_somaxconn, "I",
247 "Maximum listen socket pending connection accept queue size (compat)");
249 static int numopensockets;
250 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
251 &numopensockets, 0, "Number of open sockets");
254 * accept_mtx locks down per-socket fields relating to accept queues. See
255 * socketvar.h for an annotation of the protected fields of struct socket.
257 struct mtx accept_mtx;
258 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
261 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
264 static struct mtx so_global_mtx;
265 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
268 * General IPC sysctl name space, used by sockets and a variety of other IPC
271 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
275 * Initialize the socket subsystem and set up the socket
278 static uma_zone_t socket_zone;
282 socket_zone_change(void *tag)
285 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
289 socket_hhook_register(int subtype)
292 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
293 &V_socket_hhh[subtype],
294 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
295 printf("%s: WARNING: unable to register hook\n", __func__);
299 socket_hhook_deregister(int subtype)
302 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
303 printf("%s: WARNING: unable to deregister hook\n", __func__);
307 socket_init(void *tag)
310 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
311 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
312 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
313 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
314 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
315 EVENTHANDLER_PRI_FIRST);
317 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
320 socket_vnet_init(const void *unused __unused)
324 /* We expect a contiguous range */
325 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
326 socket_hhook_register(i);
328 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
329 socket_vnet_init, NULL);
332 socket_vnet_uninit(const void *unused __unused)
336 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
337 socket_hhook_deregister(i);
339 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
340 socket_vnet_uninit, NULL);
343 * Initialise maxsockets. This SYSINIT must be run after
347 init_maxsockets(void *ignored)
350 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
351 maxsockets = imax(maxsockets, maxfiles);
353 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
356 * Sysctl to get and set the maximum global sockets limit. Notify protocols
357 * of the change so that they can update their dependent limits as required.
360 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
362 int error, newmaxsockets;
364 newmaxsockets = maxsockets;
365 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
366 if (error == 0 && req->newptr && newmaxsockets != maxsockets) {
367 if (newmaxsockets > maxsockets &&
368 newmaxsockets <= maxfiles) {
369 maxsockets = newmaxsockets;
370 EVENTHANDLER_INVOKE(maxsockets_change);
376 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
377 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &maxsockets, 0,
378 sysctl_maxsockets, "IU",
379 "Maximum number of sockets available");
382 * Socket operation routines. These routines are called by the routines in
383 * sys_socket.c or from a system process, and implement the semantics of
384 * socket operations by switching out to the protocol specific routines.
388 * Get a socket structure from our zone, and initialize it. Note that it
389 * would probably be better to allocate socket and PCB at the same time, but
390 * I'm not convinced that all the protocols can be easily modified to do
393 * soalloc() returns a socket with a ref count of 0.
395 static struct socket *
396 soalloc(struct vnet *vnet)
400 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
404 if (mac_socket_init(so, M_NOWAIT) != 0) {
405 uma_zfree(socket_zone, so);
409 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
410 uma_zfree(socket_zone, so);
415 * The socket locking protocol allows to lock 2 sockets at a time,
416 * however, the first one must be a listening socket. WITNESS lacks
417 * a feature to change class of an existing lock, so we use DUPOK.
419 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
420 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
421 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
422 so->so_rcv.sb_sel = &so->so_rdsel;
423 so->so_snd.sb_sel = &so->so_wrsel;
424 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
425 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
426 TAILQ_INIT(&so->so_snd.sb_aiojobq);
427 TAILQ_INIT(&so->so_rcv.sb_aiojobq);
428 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
429 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
431 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
432 __func__, __LINE__, so));
435 /* We shouldn't need the so_global_mtx */
436 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
437 /* Do we need more comprehensive error returns? */
438 uma_zfree(socket_zone, so);
441 mtx_lock(&so_global_mtx);
442 so->so_gencnt = ++so_gencnt;
445 vnet->vnet_sockcnt++;
447 mtx_unlock(&so_global_mtx);
453 * Free the storage associated with a socket at the socket layer, tear down
454 * locks, labels, etc. All protocol state is assumed already to have been
455 * torn down (and possibly never set up) by the caller.
458 sodealloc(struct socket *so)
461 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
462 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
464 mtx_lock(&so_global_mtx);
465 so->so_gencnt = ++so_gencnt;
466 --numopensockets; /* Could be below, but faster here. */
468 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
469 __func__, __LINE__, so));
470 so->so_vnet->vnet_sockcnt--;
472 mtx_unlock(&so_global_mtx);
474 mac_socket_destroy(so);
476 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
478 khelp_destroy_osd(&so->osd);
479 if (SOLISTENING(so)) {
480 if (so->sol_accept_filter != NULL)
481 accept_filt_setopt(so, NULL);
483 if (so->so_rcv.sb_hiwat)
484 (void)chgsbsize(so->so_cred->cr_uidinfo,
485 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
486 if (so->so_snd.sb_hiwat)
487 (void)chgsbsize(so->so_cred->cr_uidinfo,
488 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
489 sx_destroy(&so->so_snd.sb_sx);
490 sx_destroy(&so->so_rcv.sb_sx);
491 SOCKBUF_LOCK_DESTROY(&so->so_snd);
492 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
495 mtx_destroy(&so->so_lock);
496 uma_zfree(socket_zone, so);
500 * socreate returns a socket with a ref count of 1. The socket should be
501 * closed with soclose().
504 socreate(int dom, struct socket **aso, int type, int proto,
505 struct ucred *cred, struct thread *td)
512 prp = pffindproto(dom, proto, type);
514 prp = pffindtype(dom, type);
517 /* No support for domain. */
518 if (pffinddomain(dom) == NULL)
519 return (EAFNOSUPPORT);
520 /* No support for socket type. */
521 if (proto == 0 && type != 0)
523 return (EPROTONOSUPPORT);
525 if (prp->pr_usrreqs->pru_attach == NULL ||
526 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
527 return (EPROTONOSUPPORT);
529 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
530 return (EPROTONOSUPPORT);
532 if (prp->pr_type != type)
534 so = soalloc(CRED_TO_VNET(cred));
539 so->so_cred = crhold(cred);
540 if ((prp->pr_domain->dom_family == PF_INET) ||
541 (prp->pr_domain->dom_family == PF_INET6) ||
542 (prp->pr_domain->dom_family == PF_ROUTE))
543 so->so_fibnum = td->td_proc->p_fibnum;
548 mac_socket_create(cred, so);
550 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
551 so_rdknl_assert_lock);
552 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
553 so_wrknl_assert_lock);
555 * Auto-sizing of socket buffers is managed by the protocols and
556 * the appropriate flags must be set in the pru_attach function.
558 CURVNET_SET(so->so_vnet);
559 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
571 static int regression_sonewconn_earlytest = 1;
572 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
573 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
576 static struct timeval overinterval = { 60, 0 };
577 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
579 "Delay in seconds between warnings for listen socket overflows");
582 * When an attempt at a new connection is noted on a socket which accepts
583 * connections, sonewconn is called. If the connection is possible (subject
584 * to space constraints, etc.) then we allocate a new structure, properly
585 * linked into the data structure of the original socket, and return this.
586 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
588 * Note: the ref count on the socket is 0 on return.
591 sonewconn(struct socket *head, int connstatus)
597 const char localprefix[] = "local:";
598 char descrbuf[SUNPATHLEN + sizeof(localprefix)];
600 char addrbuf[INET6_ADDRSTRLEN];
602 char addrbuf[INET_ADDRSTRLEN];
607 over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
609 if (regression_sonewconn_earlytest && over) {
613 head->sol_overcount++;
614 dolog = !!ratecheck(&head->sol_lastover, &overinterval);
617 * If we're going to log, copy the overflow count and queue
618 * length from the listen socket before dropping the lock.
619 * Also, reset the overflow count.
622 overcount = head->sol_overcount;
623 head->sol_overcount = 0;
624 qlen = head->sol_qlen;
626 SOLISTEN_UNLOCK(head);
630 * Try to print something descriptive about the
631 * socket for the error message.
633 sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
635 switch (head->so_proto->pr_domain->dom_family) {
636 #if defined(INET) || defined(INET6)
642 if (head->so_proto->pr_domain->dom_family ==
644 (sotoinpcb(head)->inp_inc.inc_flags &
647 &sotoinpcb(head)->inp_inc.inc6_laddr);
648 sbuf_printf(&descrsb, "[%s]", addrbuf);
654 sotoinpcb(head)->inp_inc.inc_laddr,
656 sbuf_cat(&descrsb, addrbuf);
659 sbuf_printf(&descrsb, ":%hu (proto %u)",
660 ntohs(sotoinpcb(head)->inp_inc.inc_lport),
661 head->so_proto->pr_protocol);
663 #endif /* INET || INET6 */
665 sbuf_cat(&descrsb, localprefix);
666 if (sotounpcb(head)->unp_addr != NULL)
668 sotounpcb(head)->unp_addr->sun_len -
669 offsetof(struct sockaddr_un,
675 sotounpcb(head)->unp_addr->sun_path,
678 sbuf_cat(&descrsb, "(unknown)");
683 * If we can't print something more specific, at least
684 * print the domain name.
686 if (sbuf_finish(&descrsb) != 0 ||
687 sbuf_len(&descrsb) <= 0) {
688 sbuf_clear(&descrsb);
690 head->so_proto->pr_domain->dom_name ?:
692 sbuf_finish(&descrsb);
694 KASSERT(sbuf_len(&descrsb) > 0,
695 ("%s: sbuf creation failed", __func__));
697 "%s: pcb %p (%s): Listen queue overflow: "
698 "%i already in queue awaiting acceptance "
699 "(%d occurrences)\n",
700 __func__, head->so_pcb, sbuf_data(&descrsb),
702 sbuf_delete(&descrsb);
709 SOLISTEN_UNLOCK(head);
710 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
712 so = soalloc(head->so_vnet);
714 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
715 "limit reached or out of memory\n",
716 __func__, head->so_pcb);
719 so->so_listen = head;
720 so->so_type = head->so_type;
721 so->so_options = head->so_options & ~SO_ACCEPTCONN;
722 so->so_linger = head->so_linger;
723 so->so_state = head->so_state | SS_NOFDREF;
724 so->so_fibnum = head->so_fibnum;
725 so->so_proto = head->so_proto;
726 so->so_cred = crhold(head->so_cred);
728 mac_socket_newconn(head, so);
730 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
731 so_rdknl_assert_lock);
732 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
733 so_wrknl_assert_lock);
734 VNET_SO_ASSERT(head);
735 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
737 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
738 __func__, head->so_pcb);
741 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
743 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
744 __func__, head->so_pcb);
747 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
748 so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
749 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
750 so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
751 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
752 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
755 if (head->sol_accept_filter != NULL)
757 so->so_state |= connstatus;
758 soref(head); /* A socket on (in)complete queue refs head. */
760 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
761 so->so_qstate = SQ_COMP;
763 solisten_wakeup(head); /* unlocks */
766 * Keep removing sockets from the head until there's room for
767 * us to insert on the tail. In pre-locking revisions, this
768 * was a simple if(), but as we could be racing with other
769 * threads and soabort() requires dropping locks, we must
770 * loop waiting for the condition to be true.
772 while (head->sol_incqlen > head->sol_qlimit) {
775 sp = TAILQ_FIRST(&head->sol_incomp);
776 TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
779 sp->so_qstate = SQ_NONE;
780 sp->so_listen = NULL;
782 sorele(head); /* does SOLISTEN_UNLOCK, head stays */
786 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
787 so->so_qstate = SQ_INCOMP;
789 SOLISTEN_UNLOCK(head);
794 #if defined(SCTP) || defined(SCTP_SUPPORT)
796 * Socket part of sctp_peeloff(). Detach a new socket from an
797 * association. The new socket is returned with a reference.
800 sopeeloff(struct socket *head)
804 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
805 __func__, __LINE__, head));
806 so = soalloc(head->so_vnet);
808 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
809 "limit reached or out of memory\n",
810 __func__, head->so_pcb);
813 so->so_type = head->so_type;
814 so->so_options = head->so_options;
815 so->so_linger = head->so_linger;
816 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
817 so->so_fibnum = head->so_fibnum;
818 so->so_proto = head->so_proto;
819 so->so_cred = crhold(head->so_cred);
821 mac_socket_newconn(head, so);
823 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
824 so_rdknl_assert_lock);
825 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
826 so_wrknl_assert_lock);
827 VNET_SO_ASSERT(head);
828 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
830 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
831 __func__, head->so_pcb);
834 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
836 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
837 __func__, head->so_pcb);
840 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
841 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
842 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
843 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
844 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
845 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
854 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
858 CURVNET_SET(so->so_vnet);
859 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
865 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
869 CURVNET_SET(so->so_vnet);
870 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
876 * solisten() transitions a socket from a non-listening state to a listening
877 * state, but can also be used to update the listen queue depth on an
878 * existing listen socket. The protocol will call back into the sockets
879 * layer using solisten_proto_check() and solisten_proto() to check and set
880 * socket-layer listen state. Call backs are used so that the protocol can
881 * acquire both protocol and socket layer locks in whatever order is required
884 * Protocol implementors are advised to hold the socket lock across the
885 * socket-layer test and set to avoid races at the socket layer.
888 solisten(struct socket *so, int backlog, struct thread *td)
892 CURVNET_SET(so->so_vnet);
893 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
899 solisten_proto_check(struct socket *so)
902 SOCK_LOCK_ASSERT(so);
904 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
911 solisten_proto(struct socket *so, int backlog)
913 int sbrcv_lowat, sbsnd_lowat;
914 u_int sbrcv_hiwat, sbsnd_hiwat;
915 short sbrcv_flags, sbsnd_flags;
916 sbintime_t sbrcv_timeo, sbsnd_timeo;
918 SOCK_LOCK_ASSERT(so);
924 * Change this socket to listening state.
926 sbrcv_lowat = so->so_rcv.sb_lowat;
927 sbsnd_lowat = so->so_snd.sb_lowat;
928 sbrcv_hiwat = so->so_rcv.sb_hiwat;
929 sbsnd_hiwat = so->so_snd.sb_hiwat;
930 sbrcv_flags = so->so_rcv.sb_flags;
931 sbsnd_flags = so->so_snd.sb_flags;
932 sbrcv_timeo = so->so_rcv.sb_timeo;
933 sbsnd_timeo = so->so_snd.sb_timeo;
935 sbdestroy(&so->so_snd, so);
936 sbdestroy(&so->so_rcv, so);
937 sx_destroy(&so->so_snd.sb_sx);
938 sx_destroy(&so->so_rcv.sb_sx);
939 SOCKBUF_LOCK_DESTROY(&so->so_snd);
940 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
944 sizeof(struct socket) - offsetof(struct socket, so_rcv));
947 so->sol_sbrcv_lowat = sbrcv_lowat;
948 so->sol_sbsnd_lowat = sbsnd_lowat;
949 so->sol_sbrcv_hiwat = sbrcv_hiwat;
950 so->sol_sbsnd_hiwat = sbsnd_hiwat;
951 so->sol_sbrcv_flags = sbrcv_flags;
952 so->sol_sbsnd_flags = sbsnd_flags;
953 so->sol_sbrcv_timeo = sbrcv_timeo;
954 so->sol_sbsnd_timeo = sbsnd_timeo;
956 so->sol_qlen = so->sol_incqlen = 0;
957 TAILQ_INIT(&so->sol_incomp);
958 TAILQ_INIT(&so->sol_comp);
960 so->sol_accept_filter = NULL;
961 so->sol_accept_filter_arg = NULL;
962 so->sol_accept_filter_str = NULL;
964 so->sol_upcall = NULL;
965 so->sol_upcallarg = NULL;
967 so->so_options |= SO_ACCEPTCONN;
970 if (backlog < 0 || backlog > somaxconn)
972 so->sol_qlimit = backlog;
976 * Wakeup listeners/subsystems once we have a complete connection.
977 * Enters with lock, returns unlocked.
980 solisten_wakeup(struct socket *sol)
983 if (sol->sol_upcall != NULL)
984 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
986 selwakeuppri(&sol->so_rdsel, PSOCK);
987 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
989 SOLISTEN_UNLOCK(sol);
990 wakeup_one(&sol->sol_comp);
991 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
992 pgsigio(&sol->so_sigio, SIGIO, 0);
996 * Return single connection off a listening socket queue. Main consumer of
997 * the function is kern_accept4(). Some modules, that do their own accept
998 * management also use the function.
1000 * Listening socket must be locked on entry and is returned unlocked on
1002 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1005 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1010 SOLISTEN_LOCK_ASSERT(head);
1012 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1013 head->so_error == 0) {
1014 error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH,
1017 SOLISTEN_UNLOCK(head);
1021 if (head->so_error) {
1022 error = head->so_error;
1024 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1025 error = EWOULDBLOCK;
1029 SOLISTEN_UNLOCK(head);
1032 so = TAILQ_FIRST(&head->sol_comp);
1034 KASSERT(so->so_qstate == SQ_COMP,
1035 ("%s: so %p not SQ_COMP", __func__, so));
1038 so->so_qstate = SQ_NONE;
1039 so->so_listen = NULL;
1040 TAILQ_REMOVE(&head->sol_comp, so, so_list);
1041 if (flags & ACCEPT4_INHERIT)
1042 so->so_state |= (head->so_state & SS_NBIO);
1044 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1053 * Evaluate the reference count and named references on a socket; if no
1054 * references remain, free it. This should be called whenever a reference is
1055 * released, such as in sorele(), but also when named reference flags are
1056 * cleared in socket or protocol code.
1058 * sofree() will free the socket if:
1060 * - There are no outstanding file descriptor references or related consumers
1063 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
1065 * - The protocol does not have an outstanding strong reference on the socket
1068 * - The socket is not in a completed connection queue, so a process has been
1069 * notified that it is present. If it is removed, the user process may
1070 * block in accept() despite select() saying the socket was ready.
1073 sofree(struct socket *so)
1075 struct protosw *pr = so->so_proto;
1076 bool last __diagused;
1078 SOCK_LOCK_ASSERT(so);
1080 if ((so->so_state & (SS_NOFDREF | SS_PROTOREF)) != SS_NOFDREF ||
1081 refcount_load(&so->so_count) != 0 || so->so_qstate == SQ_COMP) {
1086 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
1089 sol = so->so_listen;
1090 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
1093 * To solve race between close of a listening socket and
1094 * a socket on its incomplete queue, we need to lock both.
1095 * The order is first listening socket, then regular.
1096 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
1097 * function and the listening socket are the only pointers
1098 * to so. To preserve so and sol, we reference both and then
1100 * After relock the socket may not move to so_comp since it
1101 * doesn't have PCB already, but it may be removed from
1102 * so_incomp. If that happens, we share responsiblity on
1103 * freeing the socket, but soclose() has already removed
1111 if (so->so_qstate == SQ_INCOMP) {
1112 KASSERT(so->so_listen == sol,
1113 ("%s: so %p migrated out of sol %p",
1114 __func__, so, sol));
1115 TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1117 last = refcount_release(&sol->so_count);
1118 KASSERT(!last, ("%s: released last reference for %p",
1120 so->so_qstate = SQ_NONE;
1121 so->so_listen = NULL;
1123 KASSERT(so->so_listen == NULL,
1124 ("%s: so %p not on (in)comp with so_listen",
1127 KASSERT(refcount_load(&so->so_count) == 1,
1128 ("%s: so %p count %u", __func__, so, so->so_count));
1131 if (SOLISTENING(so))
1132 so->so_error = ECONNABORTED;
1135 if (so->so_dtor != NULL)
1139 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1140 (*pr->pr_domain->dom_dispose)(so);
1141 if (pr->pr_usrreqs->pru_detach != NULL)
1142 (*pr->pr_usrreqs->pru_detach)(so);
1145 * From this point on, we assume that no other references to this
1146 * socket exist anywhere else in the stack. Therefore, no locks need
1147 * to be acquired or held.
1149 * We used to do a lot of socket buffer and socket locking here, as
1150 * well as invoke sorflush() and perform wakeups. The direct call to
1151 * dom_dispose() and sbdestroy() are an inlining of what was
1152 * necessary from sorflush().
1154 * Notice that the socket buffer and kqueue state are torn down
1155 * before calling pru_detach. This means that protocols shold not
1156 * assume they can perform socket wakeups, etc, in their detach code.
1158 if (!SOLISTENING(so)) {
1159 sbdestroy(&so->so_snd, so);
1160 sbdestroy(&so->so_rcv, so);
1162 seldrain(&so->so_rdsel);
1163 seldrain(&so->so_wrsel);
1164 knlist_destroy(&so->so_rdsel.si_note);
1165 knlist_destroy(&so->so_wrsel.si_note);
1170 * Close a socket on last file table reference removal. Initiate disconnect
1171 * if connected. Free socket when disconnect complete.
1173 * This function will sorele() the socket. Note that soclose() may be called
1174 * prior to the ref count reaching zero. The actual socket structure will
1175 * not be freed until the ref count reaches zero.
1178 soclose(struct socket *so)
1180 struct accept_queue lqueue;
1181 struct socket *sp, *tsp;
1183 bool last __diagused;
1185 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1187 CURVNET_SET(so->so_vnet);
1188 funsetown(&so->so_sigio);
1189 if (so->so_state & SS_ISCONNECTED) {
1190 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1191 error = sodisconnect(so);
1193 if (error == ENOTCONN)
1199 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1200 if ((so->so_state & SS_ISDISCONNECTING) &&
1201 (so->so_state & SS_NBIO))
1203 while (so->so_state & SS_ISCONNECTED) {
1204 error = tsleep(&so->so_timeo,
1205 PSOCK | PCATCH, "soclos",
1206 so->so_linger * hz);
1214 if (so->so_proto->pr_usrreqs->pru_close != NULL)
1215 (*so->so_proto->pr_usrreqs->pru_close)(so);
1217 TAILQ_INIT(&lqueue);
1219 if (SOLISTENING(so)) {
1220 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1221 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1223 so->sol_qlen = so->sol_incqlen = 0;
1225 TAILQ_FOREACH(sp, &lqueue, so_list) {
1227 sp->so_qstate = SQ_NONE;
1228 sp->so_listen = NULL;
1230 last = refcount_release(&so->so_count);
1231 KASSERT(!last, ("%s: released last reference for %p",
1235 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1236 so->so_state |= SS_NOFDREF;
1238 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1240 if (refcount_load(&sp->so_count) == 0) {
1244 /* See the handling of queued sockets in sofree(). */
1253 * soabort() is used to abruptly tear down a connection, such as when a
1254 * resource limit is reached (listen queue depth exceeded), or if a listen
1255 * socket is closed while there are sockets waiting to be accepted.
1257 * This interface is tricky, because it is called on an unreferenced socket,
1258 * and must be called only by a thread that has actually removed the socket
1259 * from the listen queue it was on, or races with other threads are risked.
1261 * This interface will call into the protocol code, so must not be called
1262 * with any socket locks held. Protocols do call it while holding their own
1263 * recursible protocol mutexes, but this is something that should be subject
1264 * to review in the future.
1267 soabort(struct socket *so)
1271 * In as much as is possible, assert that no references to this
1272 * socket are held. This is not quite the same as asserting that the
1273 * current thread is responsible for arranging for no references, but
1274 * is as close as we can get for now.
1276 KASSERT(so->so_count == 0, ("soabort: so_count"));
1277 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1278 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1281 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1282 (*so->so_proto->pr_usrreqs->pru_abort)(so);
1288 soaccept(struct socket *so, struct sockaddr **nam)
1293 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1294 so->so_state &= ~SS_NOFDREF;
1297 CURVNET_SET(so->so_vnet);
1298 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1304 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1307 return (soconnectat(AT_FDCWD, so, nam, td));
1311 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1316 if (SOLISTENING(so))
1317 return (EOPNOTSUPP);
1319 CURVNET_SET(so->so_vnet);
1321 * If protocol is connection-based, can only connect once.
1322 * Otherwise, if connected, try to disconnect first. This allows
1323 * user to disconnect by connecting to, e.g., a null address.
1325 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1326 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1327 (error = sodisconnect(so)))) {
1331 * Prevent accumulated error from previous connection from
1335 if (fd == AT_FDCWD) {
1336 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1339 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1349 soconnect2(struct socket *so1, struct socket *so2)
1353 CURVNET_SET(so1->so_vnet);
1354 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1360 sodisconnect(struct socket *so)
1364 if ((so->so_state & SS_ISCONNECTED) == 0)
1366 if (so->so_state & SS_ISDISCONNECTING)
1369 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1374 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1375 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1379 int clen = 0, error, dontroute;
1381 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1382 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1383 ("sosend_dgram: !PR_ATOMIC"));
1386 resid = uio->uio_resid;
1388 resid = top->m_pkthdr.len;
1390 * In theory resid should be unsigned. However, space must be
1391 * signed, as it might be less than 0 if we over-committed, and we
1392 * must use a signed comparison of space and resid. On the other
1393 * hand, a negative resid causes us to loop sending 0-length
1394 * segments to the protocol.
1402 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1404 td->td_ru.ru_msgsnd++;
1405 if (control != NULL)
1406 clen = control->m_len;
1408 SOCKBUF_LOCK(&so->so_snd);
1409 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1410 SOCKBUF_UNLOCK(&so->so_snd);
1415 error = so->so_error;
1417 SOCKBUF_UNLOCK(&so->so_snd);
1420 if ((so->so_state & SS_ISCONNECTED) == 0) {
1422 * `sendto' and `sendmsg' is allowed on a connection-based
1423 * socket if it supports implied connect. Return ENOTCONN if
1424 * not connected and no address is supplied.
1426 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1427 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1428 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1429 !(resid == 0 && clen != 0)) {
1430 SOCKBUF_UNLOCK(&so->so_snd);
1434 } else if (addr == NULL) {
1435 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1438 error = EDESTADDRREQ;
1439 SOCKBUF_UNLOCK(&so->so_snd);
1445 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1446 * problem and need fixing.
1448 space = sbspace(&so->so_snd);
1449 if (flags & MSG_OOB)
1452 SOCKBUF_UNLOCK(&so->so_snd);
1453 if (resid > space) {
1459 if (flags & MSG_EOR)
1460 top->m_flags |= M_EOR;
1463 * Copy the data from userland into a mbuf chain.
1464 * If no data is to be copied in, a single empty mbuf
1467 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1468 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1470 error = EFAULT; /* only possible error */
1473 space -= resid - uio->uio_resid;
1474 resid = uio->uio_resid;
1476 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1478 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1483 so->so_options |= SO_DONTROUTE;
1487 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1488 * of date. We could have received a reset packet in an interrupt or
1489 * maybe we slept while doing page faults in uiomove() etc. We could
1490 * probably recheck again inside the locking protection here, but
1491 * there are probably other places that this also happens. We must
1495 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1496 (flags & MSG_OOB) ? PRUS_OOB :
1498 * If the user set MSG_EOF, the protocol understands this flag and
1499 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1501 ((flags & MSG_EOF) &&
1502 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1505 /* If there is more to send set PRUS_MORETOCOME */
1506 (flags & MSG_MORETOCOME) ||
1507 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1508 top, addr, control, td);
1511 so->so_options &= ~SO_DONTROUTE;
1520 if (control != NULL)
1526 * Send on a socket. If send must go all at once and message is larger than
1527 * send buffering, then hard error. Lock against other senders. If must go
1528 * all at once and not enough room now, then inform user that this would
1529 * block and do nothing. Otherwise, if nonblocking, send as much as
1530 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1531 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1532 * in mbuf chain must be small enough to send all at once.
1534 * Returns nonzero on error, timeout or signal; callers must check for short
1535 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1539 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1540 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1544 int clen = 0, error, dontroute;
1545 int atomic = sosendallatonce(so) || top;
1548 struct ktls_session *tls;
1549 int tls_enq_cnt, tls_pruflag;
1553 tls_rtype = TLS_RLTYPE_APP;
1556 resid = uio->uio_resid;
1557 else if ((top->m_flags & M_PKTHDR) != 0)
1558 resid = top->m_pkthdr.len;
1560 resid = m_length(top, NULL);
1562 * In theory resid should be unsigned. However, space must be
1563 * signed, as it might be less than 0 if we over-committed, and we
1564 * must use a signed comparison of space and resid. On the other
1565 * hand, a negative resid causes us to loop sending 0-length
1566 * segments to the protocol.
1568 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1569 * type sockets since that's an error.
1571 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1577 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1578 (so->so_proto->pr_flags & PR_ATOMIC);
1580 td->td_ru.ru_msgsnd++;
1581 if (control != NULL)
1582 clen = control->m_len;
1584 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1590 tls = ktls_hold(so->so_snd.sb_tls_info);
1592 if (tls->mode == TCP_TLS_MODE_SW)
1593 tls_pruflag = PRUS_NOTREADY;
1595 if (control != NULL) {
1596 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1598 if (clen >= sizeof(*cm) &&
1599 cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1600 tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1608 if (resid == 0 && !ktls_permit_empty_frames(tls)) {
1617 SOCKBUF_LOCK(&so->so_snd);
1618 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1619 SOCKBUF_UNLOCK(&so->so_snd);
1624 error = so->so_error;
1626 SOCKBUF_UNLOCK(&so->so_snd);
1629 if ((so->so_state & SS_ISCONNECTED) == 0) {
1631 * `sendto' and `sendmsg' is allowed on a connection-
1632 * based socket if it supports implied connect.
1633 * Return ENOTCONN if not connected and no address is
1636 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1637 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1638 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1639 !(resid == 0 && clen != 0)) {
1640 SOCKBUF_UNLOCK(&so->so_snd);
1644 } else if (addr == NULL) {
1645 SOCKBUF_UNLOCK(&so->so_snd);
1646 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1649 error = EDESTADDRREQ;
1653 space = sbspace(&so->so_snd);
1654 if (flags & MSG_OOB)
1656 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1657 clen > so->so_snd.sb_hiwat) {
1658 SOCKBUF_UNLOCK(&so->so_snd);
1662 if (space < resid + clen &&
1663 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1664 if ((so->so_state & SS_NBIO) ||
1665 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1666 SOCKBUF_UNLOCK(&so->so_snd);
1667 error = EWOULDBLOCK;
1670 error = sbwait(&so->so_snd);
1671 SOCKBUF_UNLOCK(&so->so_snd);
1676 SOCKBUF_UNLOCK(&so->so_snd);
1681 if (flags & MSG_EOR)
1682 top->m_flags |= M_EOR;
1685 ktls_frame(top, tls, &tls_enq_cnt,
1687 tls_rtype = TLS_RLTYPE_APP;
1692 * Copy the data from userland into a mbuf
1693 * chain. If resid is 0, which can happen
1694 * only if we have control to send, then
1695 * a single empty mbuf is returned. This
1696 * is a workaround to prevent protocol send
1701 top = m_uiotombuf(uio, M_WAITOK, space,
1702 tls->params.max_frame_len,
1704 ((flags & MSG_EOR) ? M_EOR : 0));
1706 ktls_frame(top, tls,
1707 &tls_enq_cnt, tls_rtype);
1709 tls_rtype = TLS_RLTYPE_APP;
1712 top = m_uiotombuf(uio, M_WAITOK, space,
1713 (atomic ? max_hdr : 0),
1714 (atomic ? M_PKTHDR : 0) |
1715 ((flags & MSG_EOR) ? M_EOR : 0));
1717 error = EFAULT; /* only possible error */
1720 space -= resid - uio->uio_resid;
1721 resid = uio->uio_resid;
1725 so->so_options |= SO_DONTROUTE;
1729 * XXX all the SBS_CANTSENDMORE checks previously
1730 * done could be out of date. We could have received
1731 * a reset packet in an interrupt or maybe we slept
1732 * while doing page faults in uiomove() etc. We
1733 * could probably recheck again inside the locking
1734 * protection here, but there are probably other
1735 * places that this also happens. We must rethink
1740 pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1742 * If the user set MSG_EOF, the protocol understands
1743 * this flag and nothing left to send then use
1744 * PRU_SEND_EOF instead of PRU_SEND.
1746 ((flags & MSG_EOF) &&
1747 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1750 /* If there is more to send set PRUS_MORETOCOME. */
1751 (flags & MSG_MORETOCOME) ||
1752 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1755 pru_flag |= tls_pruflag;
1758 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1759 pru_flag, top, addr, control, td);
1763 so->so_options &= ~SO_DONTROUTE;
1768 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1774 ktls_enqueue(top, so, tls_enq_cnt);
1783 } while (resid && space > 0);
1787 SOCK_IO_SEND_UNLOCK(so);
1795 if (control != NULL)
1801 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1802 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1806 CURVNET_SET(so->so_vnet);
1807 if (!SOLISTENING(so))
1808 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1809 top, control, flags, td);
1820 * The part of soreceive() that implements reading non-inline out-of-band
1821 * data from a socket. For more complete comments, see soreceive(), from
1822 * which this code originated.
1824 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1825 * unable to return an mbuf chain to the caller.
1828 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1830 struct protosw *pr = so->so_proto;
1834 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1837 m = m_get(M_WAITOK, MT_DATA);
1838 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1842 error = uiomove(mtod(m, void *),
1843 (int) min(uio->uio_resid, m->m_len), uio);
1845 } while (uio->uio_resid && error == 0 && m);
1853 * Following replacement or removal of the first mbuf on the first mbuf chain
1854 * of a socket buffer, push necessary state changes back into the socket
1855 * buffer so that other consumers see the values consistently. 'nextrecord'
1856 * is the callers locally stored value of the original value of
1857 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1858 * NOTE: 'nextrecord' may be NULL.
1860 static __inline void
1861 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1864 SOCKBUF_LOCK_ASSERT(sb);
1866 * First, update for the new value of nextrecord. If necessary, make
1867 * it the first record.
1869 if (sb->sb_mb != NULL)
1870 sb->sb_mb->m_nextpkt = nextrecord;
1872 sb->sb_mb = nextrecord;
1875 * Now update any dependent socket buffer fields to reflect the new
1876 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1877 * addition of a second clause that takes care of the case where
1878 * sb_mb has been updated, but remains the last record.
1880 if (sb->sb_mb == NULL) {
1881 sb->sb_mbtail = NULL;
1882 sb->sb_lastrecord = NULL;
1883 } else if (sb->sb_mb->m_nextpkt == NULL)
1884 sb->sb_lastrecord = sb->sb_mb;
1888 * Implement receive operations on a socket. We depend on the way that
1889 * records are added to the sockbuf by sbappend. In particular, each record
1890 * (mbufs linked through m_next) must begin with an address if the protocol
1891 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1892 * data, and then zero or more mbufs of data. In order to allow parallelism
1893 * between network receive and copying to user space, as well as avoid
1894 * sleeping with a mutex held, we release the socket buffer mutex during the
1895 * user space copy. Although the sockbuf is locked, new data may still be
1896 * appended, and thus we must maintain consistency of the sockbuf during that
1899 * The caller may receive the data as a single mbuf chain by supplying an
1900 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1901 * the count in uio_resid.
1904 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1905 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1907 struct mbuf *m, **mp;
1908 int flags, error, offset;
1910 struct protosw *pr = so->so_proto;
1911 struct mbuf *nextrecord;
1913 ssize_t orig_resid = uio->uio_resid;
1918 if (controlp != NULL)
1921 flags = *flagsp &~ MSG_EOR;
1924 if (flags & MSG_OOB)
1925 return (soreceive_rcvoob(so, uio, flags));
1928 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1929 && uio->uio_resid) {
1931 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1934 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1939 SOCKBUF_LOCK(&so->so_rcv);
1940 m = so->so_rcv.sb_mb;
1942 * If we have less data than requested, block awaiting more (subject
1943 * to any timeout) if:
1944 * 1. the current count is less than the low water mark, or
1945 * 2. MSG_DONTWAIT is not set
1947 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1948 sbavail(&so->so_rcv) < uio->uio_resid) &&
1949 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1950 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1951 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1952 ("receive: m == %p sbavail == %u",
1953 m, sbavail(&so->so_rcv)));
1954 if (so->so_error || so->so_rerror) {
1958 error = so->so_error;
1960 error = so->so_rerror;
1961 if ((flags & MSG_PEEK) == 0) {
1967 SOCKBUF_UNLOCK(&so->so_rcv);
1970 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1971 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1975 else if (so->so_rcv.sb_tlsdcc == 0 &&
1976 so->so_rcv.sb_tlscc == 0) {
1980 SOCKBUF_UNLOCK(&so->so_rcv);
1984 for (; m != NULL; m = m->m_next)
1985 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1986 m = so->so_rcv.sb_mb;
1989 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
1990 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
1991 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1992 SOCKBUF_UNLOCK(&so->so_rcv);
1996 if (uio->uio_resid == 0) {
1997 SOCKBUF_UNLOCK(&so->so_rcv);
2000 if ((so->so_state & SS_NBIO) ||
2001 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2002 SOCKBUF_UNLOCK(&so->so_rcv);
2003 error = EWOULDBLOCK;
2006 SBLASTRECORDCHK(&so->so_rcv);
2007 SBLASTMBUFCHK(&so->so_rcv);
2008 error = sbwait(&so->so_rcv);
2009 SOCKBUF_UNLOCK(&so->so_rcv);
2016 * From this point onward, we maintain 'nextrecord' as a cache of the
2017 * pointer to the next record in the socket buffer. We must keep the
2018 * various socket buffer pointers and local stack versions of the
2019 * pointers in sync, pushing out modifications before dropping the
2020 * socket buffer mutex, and re-reading them when picking it up.
2022 * Otherwise, we will race with the network stack appending new data
2023 * or records onto the socket buffer by using inconsistent/stale
2024 * versions of the field, possibly resulting in socket buffer
2027 * By holding the high-level sblock(), we prevent simultaneous
2028 * readers from pulling off the front of the socket buffer.
2030 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2032 uio->uio_td->td_ru.ru_msgrcv++;
2033 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2034 SBLASTRECORDCHK(&so->so_rcv);
2035 SBLASTMBUFCHK(&so->so_rcv);
2036 nextrecord = m->m_nextpkt;
2037 if (pr->pr_flags & PR_ADDR) {
2038 KASSERT(m->m_type == MT_SONAME,
2039 ("m->m_type == %d", m->m_type));
2042 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2044 if (flags & MSG_PEEK) {
2047 sbfree(&so->so_rcv, m);
2048 so->so_rcv.sb_mb = m_free(m);
2049 m = so->so_rcv.sb_mb;
2050 sockbuf_pushsync(&so->so_rcv, nextrecord);
2055 * Process one or more MT_CONTROL mbufs present before any data mbufs
2056 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
2057 * just copy the data; if !MSG_PEEK, we call into the protocol to
2058 * perform externalization (or freeing if controlp == NULL).
2060 if (m != NULL && m->m_type == MT_CONTROL) {
2061 struct mbuf *cm = NULL, *cmn;
2062 struct mbuf **cme = &cm;
2064 struct cmsghdr *cmsg;
2065 struct tls_get_record tgr;
2068 * For MSG_TLSAPPDATA, check for an alert record.
2069 * If found, return ENXIO without removing
2070 * it from the receive queue. This allows a subsequent
2071 * call without MSG_TLSAPPDATA to receive it.
2072 * Note that, for TLS, there should only be a single
2073 * control mbuf with the TLS_GET_RECORD message in it.
2075 if (flags & MSG_TLSAPPDATA) {
2076 cmsg = mtod(m, struct cmsghdr *);
2077 if (cmsg->cmsg_type == TLS_GET_RECORD &&
2078 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2079 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2080 if (__predict_false(tgr.tls_type ==
2081 TLS_RLTYPE_ALERT)) {
2082 SOCKBUF_UNLOCK(&so->so_rcv);
2091 if (flags & MSG_PEEK) {
2092 if (controlp != NULL) {
2093 *controlp = m_copym(m, 0, m->m_len,
2095 controlp = &(*controlp)->m_next;
2099 sbfree(&so->so_rcv, m);
2100 so->so_rcv.sb_mb = m->m_next;
2103 cme = &(*cme)->m_next;
2104 m = so->so_rcv.sb_mb;
2106 } while (m != NULL && m->m_type == MT_CONTROL);
2107 if ((flags & MSG_PEEK) == 0)
2108 sockbuf_pushsync(&so->so_rcv, nextrecord);
2109 while (cm != NULL) {
2112 if (pr->pr_domain->dom_externalize != NULL) {
2113 SOCKBUF_UNLOCK(&so->so_rcv);
2115 error = (*pr->pr_domain->dom_externalize)
2116 (cm, controlp, flags);
2117 SOCKBUF_LOCK(&so->so_rcv);
2118 } else if (controlp != NULL)
2122 if (controlp != NULL) {
2123 while (*controlp != NULL)
2124 controlp = &(*controlp)->m_next;
2129 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2131 nextrecord = so->so_rcv.sb_mb;
2135 if ((flags & MSG_PEEK) == 0) {
2136 KASSERT(m->m_nextpkt == nextrecord,
2137 ("soreceive: post-control, nextrecord !sync"));
2138 if (nextrecord == NULL) {
2139 KASSERT(so->so_rcv.sb_mb == m,
2140 ("soreceive: post-control, sb_mb!=m"));
2141 KASSERT(so->so_rcv.sb_lastrecord == m,
2142 ("soreceive: post-control, lastrecord!=m"));
2146 if (type == MT_OOBDATA)
2149 if ((flags & MSG_PEEK) == 0) {
2150 KASSERT(so->so_rcv.sb_mb == nextrecord,
2151 ("soreceive: sb_mb != nextrecord"));
2152 if (so->so_rcv.sb_mb == NULL) {
2153 KASSERT(so->so_rcv.sb_lastrecord == NULL,
2154 ("soreceive: sb_lastercord != NULL"));
2158 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2159 SBLASTRECORDCHK(&so->so_rcv);
2160 SBLASTMBUFCHK(&so->so_rcv);
2163 * Now continue to read any data mbufs off of the head of the socket
2164 * buffer until the read request is satisfied. Note that 'type' is
2165 * used to store the type of any mbuf reads that have happened so far
2166 * such that soreceive() can stop reading if the type changes, which
2167 * causes soreceive() to return only one of regular data and inline
2168 * out-of-band data in a single socket receive operation.
2172 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2175 * If the type of mbuf has changed since the last mbuf
2176 * examined ('type'), end the receive operation.
2178 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2179 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2180 if (type != m->m_type)
2182 } else if (type == MT_OOBDATA)
2185 KASSERT(m->m_type == MT_DATA,
2186 ("m->m_type == %d", m->m_type));
2187 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2188 len = uio->uio_resid;
2189 if (so->so_oobmark && len > so->so_oobmark - offset)
2190 len = so->so_oobmark - offset;
2191 if (len > m->m_len - moff)
2192 len = m->m_len - moff;
2194 * If mp is set, just pass back the mbufs. Otherwise copy
2195 * them out via the uio, then free. Sockbuf must be
2196 * consistent here (points to current mbuf, it points to next
2197 * record) when we drop priority; we must note any additions
2198 * to the sockbuf when we block interrupts again.
2201 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2202 SBLASTRECORDCHK(&so->so_rcv);
2203 SBLASTMBUFCHK(&so->so_rcv);
2204 SOCKBUF_UNLOCK(&so->so_rcv);
2205 if ((m->m_flags & M_EXTPG) != 0)
2206 error = m_unmapped_uiomove(m, moff, uio,
2209 error = uiomove(mtod(m, char *) + moff,
2211 SOCKBUF_LOCK(&so->so_rcv);
2214 * The MT_SONAME mbuf has already been removed
2215 * from the record, so it is necessary to
2216 * remove the data mbufs, if any, to preserve
2217 * the invariant in the case of PR_ADDR that
2218 * requires MT_SONAME mbufs at the head of
2221 if (pr->pr_flags & PR_ATOMIC &&
2222 ((flags & MSG_PEEK) == 0))
2223 (void)sbdroprecord_locked(&so->so_rcv);
2224 SOCKBUF_UNLOCK(&so->so_rcv);
2228 uio->uio_resid -= len;
2229 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2230 if (len == m->m_len - moff) {
2231 if (m->m_flags & M_EOR)
2233 if (flags & MSG_PEEK) {
2237 nextrecord = m->m_nextpkt;
2238 sbfree(&so->so_rcv, m);
2240 m->m_nextpkt = NULL;
2243 so->so_rcv.sb_mb = m = m->m_next;
2246 so->so_rcv.sb_mb = m_free(m);
2247 m = so->so_rcv.sb_mb;
2249 sockbuf_pushsync(&so->so_rcv, nextrecord);
2250 SBLASTRECORDCHK(&so->so_rcv);
2251 SBLASTMBUFCHK(&so->so_rcv);
2254 if (flags & MSG_PEEK)
2258 if (flags & MSG_DONTWAIT) {
2259 *mp = m_copym(m, 0, len,
2263 * m_copym() couldn't
2265 * Adjust uio_resid back
2267 * down by len bytes,
2268 * which we didn't end
2269 * up "copying" over).
2271 uio->uio_resid += len;
2275 SOCKBUF_UNLOCK(&so->so_rcv);
2276 *mp = m_copym(m, 0, len,
2278 SOCKBUF_LOCK(&so->so_rcv);
2281 sbcut_locked(&so->so_rcv, len);
2284 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2285 if (so->so_oobmark) {
2286 if ((flags & MSG_PEEK) == 0) {
2287 so->so_oobmark -= len;
2288 if (so->so_oobmark == 0) {
2289 so->so_rcv.sb_state |= SBS_RCVATMARK;
2294 if (offset == so->so_oobmark)
2298 if (flags & MSG_EOR)
2301 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2302 * must not quit until "uio->uio_resid == 0" or an error
2303 * termination. If a signal/timeout occurs, return with a
2304 * short count but without error. Keep sockbuf locked
2305 * against other readers.
2307 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2308 !sosendallatonce(so) && nextrecord == NULL) {
2309 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2310 if (so->so_error || so->so_rerror ||
2311 so->so_rcv.sb_state & SBS_CANTRCVMORE)
2314 * Notify the protocol that some data has been
2315 * drained before blocking.
2317 if (pr->pr_flags & PR_WANTRCVD) {
2318 SOCKBUF_UNLOCK(&so->so_rcv);
2320 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2321 SOCKBUF_LOCK(&so->so_rcv);
2323 SBLASTRECORDCHK(&so->so_rcv);
2324 SBLASTMBUFCHK(&so->so_rcv);
2326 * We could receive some data while was notifying
2327 * the protocol. Skip blocking in this case.
2329 if (so->so_rcv.sb_mb == NULL) {
2330 error = sbwait(&so->so_rcv);
2332 SOCKBUF_UNLOCK(&so->so_rcv);
2336 m = so->so_rcv.sb_mb;
2338 nextrecord = m->m_nextpkt;
2342 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2343 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2345 if ((flags & MSG_PEEK) == 0)
2346 (void) sbdroprecord_locked(&so->so_rcv);
2348 if ((flags & MSG_PEEK) == 0) {
2351 * First part is an inline SB_EMPTY_FIXUP(). Second
2352 * part makes sure sb_lastrecord is up-to-date if
2353 * there is still data in the socket buffer.
2355 so->so_rcv.sb_mb = nextrecord;
2356 if (so->so_rcv.sb_mb == NULL) {
2357 so->so_rcv.sb_mbtail = NULL;
2358 so->so_rcv.sb_lastrecord = NULL;
2359 } else if (nextrecord->m_nextpkt == NULL)
2360 so->so_rcv.sb_lastrecord = nextrecord;
2362 SBLASTRECORDCHK(&so->so_rcv);
2363 SBLASTMBUFCHK(&so->so_rcv);
2365 * If soreceive() is being done from the socket callback,
2366 * then don't need to generate ACK to peer to update window,
2367 * since ACK will be generated on return to TCP.
2369 if (!(flags & MSG_SOCALLBCK) &&
2370 (pr->pr_flags & PR_WANTRCVD)) {
2371 SOCKBUF_UNLOCK(&so->so_rcv);
2373 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2374 SOCKBUF_LOCK(&so->so_rcv);
2377 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2378 if (orig_resid == uio->uio_resid && orig_resid &&
2379 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2380 SOCKBUF_UNLOCK(&so->so_rcv);
2383 SOCKBUF_UNLOCK(&so->so_rcv);
2388 SOCK_IO_RECV_UNLOCK(so);
2393 * Optimized version of soreceive() for stream (TCP) sockets.
2396 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2397 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2399 int len = 0, error = 0, flags, oresid;
2401 struct mbuf *m, *n = NULL;
2403 /* We only do stream sockets. */
2404 if (so->so_type != SOCK_STREAM)
2409 flags = *flagsp &~ MSG_EOR;
2412 if (controlp != NULL)
2414 if (flags & MSG_OOB)
2415 return (soreceive_rcvoob(so, uio, flags));
2423 * KTLS store TLS records as records with a control message to
2424 * describe the framing.
2426 * We check once here before acquiring locks to optimize the
2429 if (sb->sb_tls_info != NULL)
2430 return (soreceive_generic(so, psa, uio, mp0, controlp,
2434 /* Prevent other readers from entering the socket. */
2435 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2441 if (sb->sb_tls_info != NULL) {
2443 SOCK_IO_RECV_UNLOCK(so);
2444 return (soreceive_generic(so, psa, uio, mp0, controlp,
2449 /* Easy one, no space to copyout anything. */
2450 if (uio->uio_resid == 0) {
2454 oresid = uio->uio_resid;
2456 /* We will never ever get anything unless we are or were connected. */
2457 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2463 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2465 /* Abort if socket has reported problems. */
2467 if (sbavail(sb) > 0)
2469 if (oresid > uio->uio_resid)
2471 error = so->so_error;
2472 if (!(flags & MSG_PEEK))
2477 /* Door is closed. Deliver what is left, if any. */
2478 if (sb->sb_state & SBS_CANTRCVMORE) {
2479 if (sbavail(sb) > 0)
2485 /* Socket buffer is empty and we shall not block. */
2486 if (sbavail(sb) == 0 &&
2487 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2492 /* Socket buffer got some data that we shall deliver now. */
2493 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2494 ((so->so_state & SS_NBIO) ||
2495 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2496 sbavail(sb) >= sb->sb_lowat ||
2497 sbavail(sb) >= uio->uio_resid ||
2498 sbavail(sb) >= sb->sb_hiwat) ) {
2502 /* On MSG_WAITALL we must wait until all data or error arrives. */
2503 if ((flags & MSG_WAITALL) &&
2504 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2508 * Wait and block until (more) data comes in.
2509 * NB: Drops the sockbuf lock during wait.
2517 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2518 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2519 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2523 uio->uio_td->td_ru.ru_msgrcv++;
2525 /* Fill uio until full or current end of socket buffer is reached. */
2526 len = min(uio->uio_resid, sbavail(sb));
2528 /* Dequeue as many mbufs as possible. */
2529 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2533 m_cat(*mp0, sb->sb_mb);
2535 m != NULL && m->m_len <= len;
2537 KASSERT(!(m->m_flags & M_NOTAVAIL),
2538 ("%s: m %p not available", __func__, m));
2540 uio->uio_resid -= m->m_len;
2546 sb->sb_lastrecord = sb->sb_mb;
2547 if (sb->sb_mb == NULL)
2550 /* Copy the remainder. */
2552 KASSERT(sb->sb_mb != NULL,
2553 ("%s: len > 0 && sb->sb_mb empty", __func__));
2555 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2557 len = 0; /* Don't flush data from sockbuf. */
2559 uio->uio_resid -= len;
2570 /* NB: Must unlock socket buffer as uiomove may sleep. */
2572 error = m_mbuftouio(uio, sb->sb_mb, len);
2577 SBLASTRECORDCHK(sb);
2581 * Remove the delivered data from the socket buffer unless we
2582 * were only peeking.
2584 if (!(flags & MSG_PEEK)) {
2586 sbdrop_locked(sb, len);
2588 /* Notify protocol that we drained some data. */
2589 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2590 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2591 !(flags & MSG_SOCALLBCK))) {
2594 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2600 * For MSG_WAITALL we may have to loop again and wait for
2601 * more data to come in.
2603 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2606 SBLASTRECORDCHK(sb);
2609 SOCK_IO_RECV_UNLOCK(so);
2614 * Optimized version of soreceive() for simple datagram cases from userspace.
2615 * Unlike in the stream case, we're able to drop a datagram if copyout()
2616 * fails, and because we handle datagrams atomically, we don't need to use a
2617 * sleep lock to prevent I/O interlacing.
2620 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2621 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2623 struct mbuf *m, *m2;
2626 struct protosw *pr = so->so_proto;
2627 struct mbuf *nextrecord;
2631 if (controlp != NULL)
2634 flags = *flagsp &~ MSG_EOR;
2639 * For any complicated cases, fall back to the full
2640 * soreceive_generic().
2642 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2643 return (soreceive_generic(so, psa, uio, mp0, controlp,
2647 * Enforce restrictions on use.
2649 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2650 ("soreceive_dgram: wantrcvd"));
2651 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2652 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2653 ("soreceive_dgram: SBS_RCVATMARK"));
2654 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2655 ("soreceive_dgram: P_CONNREQUIRED"));
2658 * Loop blocking while waiting for a datagram.
2660 SOCKBUF_LOCK(&so->so_rcv);
2661 while ((m = so->so_rcv.sb_mb) == NULL) {
2662 KASSERT(sbavail(&so->so_rcv) == 0,
2663 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2664 sbavail(&so->so_rcv)));
2666 error = so->so_error;
2668 SOCKBUF_UNLOCK(&so->so_rcv);
2671 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2672 uio->uio_resid == 0) {
2673 SOCKBUF_UNLOCK(&so->so_rcv);
2676 if ((so->so_state & SS_NBIO) ||
2677 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2678 SOCKBUF_UNLOCK(&so->so_rcv);
2679 return (EWOULDBLOCK);
2681 SBLASTRECORDCHK(&so->so_rcv);
2682 SBLASTMBUFCHK(&so->so_rcv);
2683 error = sbwait(&so->so_rcv);
2685 SOCKBUF_UNLOCK(&so->so_rcv);
2689 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2692 uio->uio_td->td_ru.ru_msgrcv++;
2693 SBLASTRECORDCHK(&so->so_rcv);
2694 SBLASTMBUFCHK(&so->so_rcv);
2695 nextrecord = m->m_nextpkt;
2696 if (nextrecord == NULL) {
2697 KASSERT(so->so_rcv.sb_lastrecord == m,
2698 ("soreceive_dgram: lastrecord != m"));
2701 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2702 ("soreceive_dgram: m_nextpkt != nextrecord"));
2705 * Pull 'm' and its chain off the front of the packet queue.
2707 so->so_rcv.sb_mb = NULL;
2708 sockbuf_pushsync(&so->so_rcv, nextrecord);
2711 * Walk 'm's chain and free that many bytes from the socket buffer.
2713 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2714 sbfree(&so->so_rcv, m2);
2717 * Do a few last checks before we let go of the lock.
2719 SBLASTRECORDCHK(&so->so_rcv);
2720 SBLASTMBUFCHK(&so->so_rcv);
2721 SOCKBUF_UNLOCK(&so->so_rcv);
2723 if (pr->pr_flags & PR_ADDR) {
2724 KASSERT(m->m_type == MT_SONAME,
2725 ("m->m_type == %d", m->m_type));
2727 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2732 /* XXXRW: Can this happen? */
2737 * Packet to copyout() is now in 'm' and it is disconnected from the
2740 * Process one or more MT_CONTROL mbufs present before any data mbufs
2741 * in the first mbuf chain on the socket buffer. We call into the
2742 * protocol to perform externalization (or freeing if controlp ==
2743 * NULL). In some cases there can be only MT_CONTROL mbufs without
2746 if (m->m_type == MT_CONTROL) {
2747 struct mbuf *cm = NULL, *cmn;
2748 struct mbuf **cme = &cm;
2754 cme = &(*cme)->m_next;
2756 } while (m != NULL && m->m_type == MT_CONTROL);
2757 while (cm != NULL) {
2760 if (pr->pr_domain->dom_externalize != NULL) {
2761 error = (*pr->pr_domain->dom_externalize)
2762 (cm, controlp, flags);
2763 } else if (controlp != NULL)
2767 if (controlp != NULL) {
2768 while (*controlp != NULL)
2769 controlp = &(*controlp)->m_next;
2774 KASSERT(m == NULL || m->m_type == MT_DATA,
2775 ("soreceive_dgram: !data"));
2776 while (m != NULL && uio->uio_resid > 0) {
2777 len = uio->uio_resid;
2780 error = uiomove(mtod(m, char *), (int)len, uio);
2785 if (len == m->m_len)
2802 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2803 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2807 CURVNET_SET(so->so_vnet);
2808 if (!SOLISTENING(so))
2809 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2810 mp0, controlp, flagsp));
2818 soshutdown(struct socket *so, int how)
2820 struct protosw *pr = so->so_proto;
2821 int error, soerror_enotconn;
2823 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2826 soerror_enotconn = 0;
2828 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2830 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2831 * invoked on a datagram sockets, however historically we would
2832 * actually tear socket down. This is known to be leveraged by
2833 * some applications to unblock process waiting in recvXXX(2)
2834 * by other process that it shares that socket with. Try to meet
2835 * both backward-compatibility and POSIX requirements by forcing
2836 * ENOTCONN but still asking protocol to perform pru_shutdown().
2838 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so))
2840 soerror_enotconn = 1;
2843 if (SOLISTENING(so)) {
2844 if (how != SHUT_WR) {
2846 so->so_error = ECONNABORTED;
2847 solisten_wakeup(so); /* unlocks so */
2852 CURVNET_SET(so->so_vnet);
2853 if (pr->pr_usrreqs->pru_flush != NULL)
2854 (*pr->pr_usrreqs->pru_flush)(so, how);
2857 if (how != SHUT_RD) {
2858 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2859 wakeup(&so->so_timeo);
2861 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2863 wakeup(&so->so_timeo);
2867 return (soerror_enotconn ? ENOTCONN : 0);
2871 sorflush(struct socket *so)
2873 struct sockbuf *sb = &so->so_rcv;
2874 struct protosw *pr = so->so_proto;
2881 * In order to avoid calling dom_dispose with the socket buffer mutex
2882 * held, and in order to generally avoid holding the lock for a long
2883 * time, we make a copy of the socket buffer and clear the original
2884 * (except locks, state). The new socket buffer copy won't have
2885 * initialized locks so we can only call routines that won't use or
2886 * assert those locks.
2888 * Dislodge threads currently blocked in receive and wait to acquire
2889 * a lock against other simultaneous readers before clearing the
2890 * socket buffer. Don't let our acquire be interrupted by a signal
2891 * despite any existing socket disposition on interruptable waiting.
2894 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
2895 KASSERT(error == 0, ("%s: cannot lock sock %p recv buffer",
2899 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2900 * and mutex data unchanged.
2903 bzero(&aso, sizeof(aso));
2904 aso.so_pcb = so->so_pcb;
2905 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2906 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2907 bzero(&sb->sb_startzero,
2908 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2910 SOCK_IO_RECV_UNLOCK(so);
2913 * Dispose of special rights and flush the copied socket. Don't call
2914 * any unsafe routines (that rely on locks being initialized) on aso.
2916 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2917 (*pr->pr_domain->dom_dispose)(&aso);
2918 sbrelease_internal(&aso.so_rcv, so);
2922 * Wrapper for Socket established helper hook.
2923 * Parameters: socket, context of the hook point, hook id.
2926 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2928 struct socket_hhook_data hhook_data = {
2935 CURVNET_SET(so->so_vnet);
2936 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2939 /* Ugly but needed, since hhooks return void for now */
2940 return (hhook_data.status);
2944 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2945 * additional variant to handle the case where the option value needs to be
2946 * some kind of integer, but not a specific size. In addition to their use
2947 * here, these functions are also called by the protocol-level pr_ctloutput()
2951 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2956 * If the user gives us more than we wanted, we ignore it, but if we
2957 * don't get the minimum length the caller wants, we return EINVAL.
2958 * On success, sopt->sopt_valsize is set to however much we actually
2961 if ((valsize = sopt->sopt_valsize) < minlen)
2964 sopt->sopt_valsize = valsize = len;
2966 if (sopt->sopt_td != NULL)
2967 return (copyin(sopt->sopt_val, buf, valsize));
2969 bcopy(sopt->sopt_val, buf, valsize);
2974 * Kernel version of setsockopt(2).
2976 * XXX: optlen is size_t, not socklen_t
2979 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2982 struct sockopt sopt;
2984 sopt.sopt_level = level;
2985 sopt.sopt_name = optname;
2986 sopt.sopt_dir = SOPT_SET;
2987 sopt.sopt_val = optval;
2988 sopt.sopt_valsize = optlen;
2989 sopt.sopt_td = NULL;
2990 return (sosetopt(so, &sopt));
2994 sosetopt(struct socket *so, struct sockopt *sopt)
2999 sbintime_t val, *valp;
3005 CURVNET_SET(so->so_vnet);
3007 if (sopt->sopt_level != SOL_SOCKET) {
3008 if (so->so_proto->pr_ctloutput != NULL)
3009 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3011 error = ENOPROTOOPT;
3013 switch (sopt->sopt_name) {
3014 case SO_ACCEPTFILTER:
3015 error = accept_filt_setopt(so, sopt);
3021 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3024 if (l.l_linger < 0 ||
3025 l.l_linger > USHRT_MAX ||
3026 l.l_linger > (INT_MAX / hz)) {
3031 so->so_linger = l.l_linger;
3033 so->so_options |= SO_LINGER;
3035 so->so_options &= ~SO_LINGER;
3042 case SO_USELOOPBACK:
3046 case SO_REUSEPORT_LB:
3054 error = sooptcopyin(sopt, &optval, sizeof optval,
3060 so->so_options |= sopt->sopt_name;
3062 so->so_options &= ~sopt->sopt_name;
3067 error = sooptcopyin(sopt, &optval, sizeof optval,
3072 if (optval < 0 || optval >= rt_numfibs) {
3076 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3077 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3078 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3079 so->so_fibnum = optval;
3084 case SO_USER_COOKIE:
3085 error = sooptcopyin(sopt, &val32, sizeof val32,
3089 so->so_user_cookie = val32;
3096 error = sooptcopyin(sopt, &optval, sizeof optval,
3102 * Values < 1 make no sense for any of these options,
3110 error = sbsetopt(so, sopt->sopt_name, optval);
3115 #ifdef COMPAT_FREEBSD32
3116 if (SV_CURPROC_FLAG(SV_ILP32)) {
3117 struct timeval32 tv32;
3119 error = sooptcopyin(sopt, &tv32, sizeof tv32,
3121 CP(tv32, tv, tv_sec);
3122 CP(tv32, tv, tv_usec);
3125 error = sooptcopyin(sopt, &tv, sizeof tv,
3129 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3130 tv.tv_usec >= 1000000) {
3134 if (tv.tv_sec > INT32_MAX)
3139 valp = sopt->sopt_name == SO_SNDTIMEO ?
3140 (SOLISTENING(so) ? &so->sol_sbsnd_timeo :
3141 &so->so_snd.sb_timeo) :
3142 (SOLISTENING(so) ? &so->sol_sbrcv_timeo :
3143 &so->so_rcv.sb_timeo);
3150 error = sooptcopyin(sopt, &extmac, sizeof extmac,
3154 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3162 error = sooptcopyin(sopt, &optval, sizeof optval,
3166 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3170 so->so_ts_clock = optval;
3173 case SO_MAX_PACING_RATE:
3174 error = sooptcopyin(sopt, &val32, sizeof(val32),
3178 so->so_max_pacing_rate = val32;
3182 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3183 error = hhook_run_socket(so, sopt,
3186 error = ENOPROTOOPT;
3189 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3190 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
3198 * Helper routine for getsockopt.
3201 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3209 * Documented get behavior is that we always return a value, possibly
3210 * truncated to fit in the user's buffer. Traditional behavior is
3211 * that we always tell the user precisely how much we copied, rather
3212 * than something useful like the total amount we had available for
3213 * her. Note that this interface is not idempotent; the entire
3214 * answer must be generated ahead of time.
3216 valsize = min(len, sopt->sopt_valsize);
3217 sopt->sopt_valsize = valsize;
3218 if (sopt->sopt_val != NULL) {
3219 if (sopt->sopt_td != NULL)
3220 error = copyout(buf, sopt->sopt_val, valsize);
3222 bcopy(buf, sopt->sopt_val, valsize);
3228 sogetopt(struct socket *so, struct sockopt *sopt)
3237 CURVNET_SET(so->so_vnet);
3239 if (sopt->sopt_level != SOL_SOCKET) {
3240 if (so->so_proto->pr_ctloutput != NULL)
3241 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3243 error = ENOPROTOOPT;
3247 switch (sopt->sopt_name) {
3248 case SO_ACCEPTFILTER:
3249 error = accept_filt_getopt(so, sopt);
3254 l.l_onoff = so->so_options & SO_LINGER;
3255 l.l_linger = so->so_linger;
3257 error = sooptcopyout(sopt, &l, sizeof l);
3260 case SO_USELOOPBACK:
3266 case SO_REUSEPORT_LB:
3276 optval = so->so_options & sopt->sopt_name;
3278 error = sooptcopyout(sopt, &optval, sizeof optval);
3282 optval = so->so_proto->pr_domain->dom_family;
3286 optval = so->so_type;
3290 optval = so->so_proto->pr_protocol;
3296 optval = so->so_error;
3299 optval = so->so_rerror;
3306 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3307 so->so_snd.sb_hiwat;
3311 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3312 so->so_rcv.sb_hiwat;
3316 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3317 so->so_snd.sb_lowat;
3321 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3322 so->so_rcv.sb_lowat;
3328 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3329 (SOLISTENING(so) ? so->sol_sbsnd_timeo :
3330 so->so_snd.sb_timeo) :
3331 (SOLISTENING(so) ? so->sol_sbrcv_timeo :
3332 so->so_rcv.sb_timeo));
3334 #ifdef COMPAT_FREEBSD32
3335 if (SV_CURPROC_FLAG(SV_ILP32)) {
3336 struct timeval32 tv32;
3338 CP(tv, tv32, tv_sec);
3339 CP(tv, tv32, tv_usec);
3340 error = sooptcopyout(sopt, &tv32, sizeof tv32);
3343 error = sooptcopyout(sopt, &tv, sizeof tv);
3348 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3352 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3356 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3364 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3368 error = mac_getsockopt_peerlabel(
3369 sopt->sopt_td->td_ucred, so, &extmac);
3372 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3378 case SO_LISTENQLIMIT:
3379 optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3383 optval = SOLISTENING(so) ? so->sol_qlen : 0;
3386 case SO_LISTENINCQLEN:
3387 optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3391 optval = so->so_ts_clock;
3394 case SO_MAX_PACING_RATE:
3395 optval = so->so_max_pacing_rate;
3399 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3400 error = hhook_run_socket(so, sopt,
3403 error = ENOPROTOOPT;
3415 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3417 struct mbuf *m, *m_prev;
3418 int sopt_size = sopt->sopt_valsize;
3420 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3423 if (sopt_size > MLEN) {
3424 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3425 if ((m->m_flags & M_EXT) == 0) {
3429 m->m_len = min(MCLBYTES, sopt_size);
3431 m->m_len = min(MLEN, sopt_size);
3433 sopt_size -= m->m_len;
3438 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3443 if (sopt_size > MLEN) {
3444 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3446 if ((m->m_flags & M_EXT) == 0) {
3451 m->m_len = min(MCLBYTES, sopt_size);
3453 m->m_len = min(MLEN, sopt_size);
3455 sopt_size -= m->m_len;
3463 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3465 struct mbuf *m0 = m;
3467 if (sopt->sopt_val == NULL)
3469 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3470 if (sopt->sopt_td != NULL) {
3473 error = copyin(sopt->sopt_val, mtod(m, char *),
3480 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3481 sopt->sopt_valsize -= m->m_len;
3482 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3485 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3486 panic("ip6_sooptmcopyin");
3491 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3493 struct mbuf *m0 = m;
3496 if (sopt->sopt_val == NULL)
3498 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3499 if (sopt->sopt_td != NULL) {
3502 error = copyout(mtod(m, char *), sopt->sopt_val,
3509 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3510 sopt->sopt_valsize -= m->m_len;
3511 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3512 valsize += m->m_len;
3516 /* enough soopt buffer should be given from user-land */
3520 sopt->sopt_valsize = valsize;
3525 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3526 * out-of-band data, which will then notify socket consumers.
3529 sohasoutofband(struct socket *so)
3532 if (so->so_sigio != NULL)
3533 pgsigio(&so->so_sigio, SIGURG, 0);
3534 selwakeuppri(&so->so_rdsel, PSOCK);
3538 sopoll(struct socket *so, int events, struct ucred *active_cred,
3543 * We do not need to set or assert curvnet as long as everyone uses
3546 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3551 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3557 if (SOLISTENING(so)) {
3558 if (!(events & (POLLIN | POLLRDNORM)))
3560 else if (!TAILQ_EMPTY(&so->sol_comp))
3561 revents = events & (POLLIN | POLLRDNORM);
3562 else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3563 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3565 selrecord(td, &so->so_rdsel);
3570 SOCKBUF_LOCK(&so->so_snd);
3571 SOCKBUF_LOCK(&so->so_rcv);
3572 if (events & (POLLIN | POLLRDNORM))
3573 if (soreadabledata(so))
3574 revents |= events & (POLLIN | POLLRDNORM);
3575 if (events & (POLLOUT | POLLWRNORM))
3576 if (sowriteable(so))
3577 revents |= events & (POLLOUT | POLLWRNORM);
3578 if (events & (POLLPRI | POLLRDBAND))
3579 if (so->so_oobmark ||
3580 (so->so_rcv.sb_state & SBS_RCVATMARK))
3581 revents |= events & (POLLPRI | POLLRDBAND);
3582 if ((events & POLLINIGNEOF) == 0) {
3583 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3584 revents |= events & (POLLIN | POLLRDNORM);
3585 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3589 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3590 revents |= events & POLLRDHUP;
3593 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
3594 selrecord(td, &so->so_rdsel);
3595 so->so_rcv.sb_flags |= SB_SEL;
3597 if (events & (POLLOUT | POLLWRNORM)) {
3598 selrecord(td, &so->so_wrsel);
3599 so->so_snd.sb_flags |= SB_SEL;
3602 SOCKBUF_UNLOCK(&so->so_rcv);
3603 SOCKBUF_UNLOCK(&so->so_snd);
3610 soo_kqfilter(struct file *fp, struct knote *kn)
3612 struct socket *so = kn->kn_fp->f_data;
3616 switch (kn->kn_filter) {
3618 kn->kn_fop = &soread_filtops;
3619 knl = &so->so_rdsel.si_note;
3623 kn->kn_fop = &sowrite_filtops;
3624 knl = &so->so_wrsel.si_note;
3628 kn->kn_fop = &soempty_filtops;
3629 knl = &so->so_wrsel.si_note;
3637 if (SOLISTENING(so)) {
3638 knlist_add(knl, kn, 1);
3641 knlist_add(knl, kn, 1);
3642 sb->sb_flags |= SB_KNOTE;
3650 * Some routines that return EOPNOTSUPP for entry points that are not
3651 * supported by a protocol. Fill in as needed.
3654 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3661 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3668 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3675 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3682 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3690 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3697 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3705 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3712 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3713 struct ifnet *ifp, struct thread *td)
3720 pru_disconnect_notsupp(struct socket *so)
3727 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3734 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3741 pru_rcvd_notsupp(struct socket *so, int flags)
3748 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3755 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3756 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3759 if (control != NULL)
3761 if ((flags & PRUS_NOTREADY) == 0)
3763 return (EOPNOTSUPP);
3767 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3770 return (EOPNOTSUPP);
3774 * This isn't really a ``null'' operation, but it's the default one and
3775 * doesn't do anything destructive.
3778 pru_sense_null(struct socket *so, struct stat *sb)
3781 sb->st_blksize = so->so_snd.sb_hiwat;
3786 pru_shutdown_notsupp(struct socket *so)
3793 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3800 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3801 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3808 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3809 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3816 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3824 filt_sordetach(struct knote *kn)
3826 struct socket *so = kn->kn_fp->f_data;
3829 knlist_remove(&so->so_rdsel.si_note, kn, 1);
3830 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3831 so->so_rcv.sb_flags &= ~SB_KNOTE;
3832 so_rdknl_unlock(so);
3837 filt_soread(struct knote *kn, long hint)
3841 so = kn->kn_fp->f_data;
3843 if (SOLISTENING(so)) {
3844 SOCK_LOCK_ASSERT(so);
3845 kn->kn_data = so->sol_qlen;
3847 kn->kn_flags |= EV_EOF;
3848 kn->kn_fflags = so->so_error;
3851 return (!TAILQ_EMPTY(&so->sol_comp));
3854 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3856 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3857 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3858 kn->kn_flags |= EV_EOF;
3859 kn->kn_fflags = so->so_error;
3861 } else if (so->so_error || so->so_rerror)
3864 if (kn->kn_sfflags & NOTE_LOWAT) {
3865 if (kn->kn_data >= kn->kn_sdata)
3867 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3870 /* This hook returning non-zero indicates an event, not error */
3871 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3875 filt_sowdetach(struct knote *kn)
3877 struct socket *so = kn->kn_fp->f_data;
3880 knlist_remove(&so->so_wrsel.si_note, kn, 1);
3881 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3882 so->so_snd.sb_flags &= ~SB_KNOTE;
3883 so_wrknl_unlock(so);
3888 filt_sowrite(struct knote *kn, long hint)
3892 so = kn->kn_fp->f_data;
3894 if (SOLISTENING(so))
3897 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3898 kn->kn_data = sbspace(&so->so_snd);
3900 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3902 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3903 kn->kn_flags |= EV_EOF;
3904 kn->kn_fflags = so->so_error;
3906 } else if (so->so_error) /* temporary udp error */
3908 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3909 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3911 else if (kn->kn_sfflags & NOTE_LOWAT)
3912 return (kn->kn_data >= kn->kn_sdata);
3914 return (kn->kn_data >= so->so_snd.sb_lowat);
3918 filt_soempty(struct knote *kn, long hint)
3922 so = kn->kn_fp->f_data;
3924 if (SOLISTENING(so))
3927 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3928 kn->kn_data = sbused(&so->so_snd);
3930 if (kn->kn_data == 0)
3937 socheckuid(struct socket *so, uid_t uid)
3942 if (so->so_cred->cr_uid != uid)
3948 * These functions are used by protocols to notify the socket layer (and its
3949 * consumers) of state changes in the sockets driven by protocol-side events.
3953 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3955 * Normal sequence from the active (originating) side is that
3956 * soisconnecting() is called during processing of connect() call, resulting
3957 * in an eventual call to soisconnected() if/when the connection is
3958 * established. When the connection is torn down soisdisconnecting() is
3959 * called during processing of disconnect() call, and soisdisconnected() is
3960 * called when the connection to the peer is totally severed. The semantics
3961 * of these routines are such that connectionless protocols can call
3962 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3963 * calls when setting up a ``connection'' takes no time.
3965 * From the passive side, a socket is created with two queues of sockets:
3966 * so_incomp for connections in progress and so_comp for connections already
3967 * made and awaiting user acceptance. As a protocol is preparing incoming
3968 * connections, it creates a socket structure queued on so_incomp by calling
3969 * sonewconn(). When the connection is established, soisconnected() is
3970 * called, and transfers the socket structure to so_comp, making it available
3973 * If a socket is closed with sockets on either so_incomp or so_comp, these
3974 * sockets are dropped.
3976 * If higher-level protocols are implemented in the kernel, the wakeups done
3977 * here will sometimes cause software-interrupt process scheduling.
3980 soisconnecting(struct socket *so)
3984 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3985 so->so_state |= SS_ISCONNECTING;
3990 soisconnected(struct socket *so)
3992 bool last __diagused;
3995 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3996 so->so_state |= SS_ISCONNECTED;
3998 if (so->so_qstate == SQ_INCOMP) {
3999 struct socket *head = so->so_listen;
4002 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
4004 * Promoting a socket from incomplete queue to complete, we
4005 * need to go through reverse order of locking. We first do
4006 * trylock, and if that doesn't succeed, we go the hard way
4007 * leaving a reference and rechecking consistency after proper
4010 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
4013 SOLISTEN_LOCK(head);
4015 if (__predict_false(head != so->so_listen)) {
4017 * The socket went off the listen queue,
4018 * should be lost race to close(2) of sol.
4019 * The socket is about to soabort().
4025 last = refcount_release(&head->so_count);
4026 KASSERT(!last, ("%s: released last reference for %p",
4030 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4031 TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4032 head->sol_incqlen--;
4033 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4035 so->so_qstate = SQ_COMP;
4037 solisten_wakeup(head); /* unlocks */
4039 SOCKBUF_LOCK(&so->so_rcv);
4040 soupcall_set(so, SO_RCV,
4041 head->sol_accept_filter->accf_callback,
4042 head->sol_accept_filter_arg);
4043 so->so_options &= ~SO_ACCEPTFILTER;
4044 ret = head->sol_accept_filter->accf_callback(so,
4045 head->sol_accept_filter_arg, M_NOWAIT);
4046 if (ret == SU_ISCONNECTED) {
4047 soupcall_clear(so, SO_RCV);
4048 SOCKBUF_UNLOCK(&so->so_rcv);
4051 SOCKBUF_UNLOCK(&so->so_rcv);
4053 SOLISTEN_UNLOCK(head);
4058 wakeup(&so->so_timeo);
4064 soisdisconnecting(struct socket *so)
4068 so->so_state &= ~SS_ISCONNECTING;
4069 so->so_state |= SS_ISDISCONNECTING;
4071 if (!SOLISTENING(so)) {
4072 SOCKBUF_LOCK(&so->so_rcv);
4073 socantrcvmore_locked(so);
4074 SOCKBUF_LOCK(&so->so_snd);
4075 socantsendmore_locked(so);
4078 wakeup(&so->so_timeo);
4082 soisdisconnected(struct socket *so)
4088 * There is at least one reader of so_state that does not
4089 * acquire socket lock, namely soreceive_generic(). Ensure
4090 * that it never sees all flags that track connection status
4091 * cleared, by ordering the update with a barrier semantic of
4092 * our release thread fence.
4094 so->so_state |= SS_ISDISCONNECTED;
4095 atomic_thread_fence_rel();
4096 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4098 if (!SOLISTENING(so)) {
4100 SOCKBUF_LOCK(&so->so_rcv);
4101 socantrcvmore_locked(so);
4102 SOCKBUF_LOCK(&so->so_snd);
4103 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4104 socantsendmore_locked(so);
4107 wakeup(&so->so_timeo);
4111 soiolock(struct socket *so, struct sx *sx, int flags)
4115 KASSERT((flags & SBL_VALID) == flags,
4116 ("soiolock: invalid flags %#x", flags));
4118 if ((flags & SBL_WAIT) != 0) {
4119 if ((flags & SBL_NOINTR) != 0) {
4122 error = sx_xlock_sig(sx);
4126 } else if (!sx_try_xlock(sx)) {
4127 return (EWOULDBLOCK);
4130 if (__predict_false(SOLISTENING(so))) {
4138 soiounlock(struct sx *sx)
4144 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4147 sodupsockaddr(const struct sockaddr *sa, int mflags)
4149 struct sockaddr *sa2;
4151 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4153 bcopy(sa, sa2, sa->sa_len);
4158 * Register per-socket destructor.
4161 sodtor_set(struct socket *so, so_dtor_t *func)
4164 SOCK_LOCK_ASSERT(so);
4169 * Register per-socket buffer upcalls.
4172 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4176 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4186 panic("soupcall_set: bad which");
4188 SOCKBUF_LOCK_ASSERT(sb);
4189 sb->sb_upcall = func;
4190 sb->sb_upcallarg = arg;
4191 sb->sb_flags |= SB_UPCALL;
4195 soupcall_clear(struct socket *so, int which)
4199 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4209 panic("soupcall_clear: bad which");
4211 SOCKBUF_LOCK_ASSERT(sb);
4212 KASSERT(sb->sb_upcall != NULL,
4213 ("%s: so %p no upcall to clear", __func__, so));
4214 sb->sb_upcall = NULL;
4215 sb->sb_upcallarg = NULL;
4216 sb->sb_flags &= ~SB_UPCALL;
4220 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4223 SOLISTEN_LOCK_ASSERT(so);
4224 so->sol_upcall = func;
4225 so->sol_upcallarg = arg;
4229 so_rdknl_lock(void *arg)
4231 struct socket *so = arg;
4233 if (SOLISTENING(so))
4236 SOCKBUF_LOCK(&so->so_rcv);
4240 so_rdknl_unlock(void *arg)
4242 struct socket *so = arg;
4244 if (SOLISTENING(so))
4247 SOCKBUF_UNLOCK(&so->so_rcv);
4251 so_rdknl_assert_lock(void *arg, int what)
4253 struct socket *so = arg;
4255 if (what == LA_LOCKED) {
4256 if (SOLISTENING(so))
4257 SOCK_LOCK_ASSERT(so);
4259 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4261 if (SOLISTENING(so))
4262 SOCK_UNLOCK_ASSERT(so);
4264 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4269 so_wrknl_lock(void *arg)
4271 struct socket *so = arg;
4273 if (SOLISTENING(so))
4276 SOCKBUF_LOCK(&so->so_snd);
4280 so_wrknl_unlock(void *arg)
4282 struct socket *so = arg;
4284 if (SOLISTENING(so))
4287 SOCKBUF_UNLOCK(&so->so_snd);
4291 so_wrknl_assert_lock(void *arg, int what)
4293 struct socket *so = arg;
4295 if (what == LA_LOCKED) {
4296 if (SOLISTENING(so))
4297 SOCK_LOCK_ASSERT(so);
4299 SOCKBUF_LOCK_ASSERT(&so->so_snd);
4301 if (SOLISTENING(so))
4302 SOCK_UNLOCK_ASSERT(so);
4304 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4309 * Create an external-format (``xsocket'') structure using the information in
4310 * the kernel-format socket structure pointed to by so. This is done to
4311 * reduce the spew of irrelevant information over this interface, to isolate
4312 * user code from changes in the kernel structure, and potentially to provide
4313 * information-hiding if we decide that some of this information should be
4314 * hidden from users.
4317 sotoxsocket(struct socket *so, struct xsocket *xso)
4320 bzero(xso, sizeof(*xso));
4321 xso->xso_len = sizeof *xso;
4322 xso->xso_so = (uintptr_t)so;
4323 xso->so_type = so->so_type;
4324 xso->so_options = so->so_options;
4325 xso->so_linger = so->so_linger;
4326 xso->so_state = so->so_state;
4327 xso->so_pcb = (uintptr_t)so->so_pcb;
4328 xso->xso_protocol = so->so_proto->pr_protocol;
4329 xso->xso_family = so->so_proto->pr_domain->dom_family;
4330 xso->so_timeo = so->so_timeo;
4331 xso->so_error = so->so_error;
4332 xso->so_uid = so->so_cred->cr_uid;
4333 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4334 if (SOLISTENING(so)) {
4335 xso->so_qlen = so->sol_qlen;
4336 xso->so_incqlen = so->sol_incqlen;
4337 xso->so_qlimit = so->sol_qlimit;
4338 xso->so_oobmark = 0;
4340 xso->so_state |= so->so_qstate;
4341 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4342 xso->so_oobmark = so->so_oobmark;
4343 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4344 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4349 so_sockbuf_rcv(struct socket *so)
4352 return (&so->so_rcv);
4356 so_sockbuf_snd(struct socket *so)
4359 return (&so->so_snd);
4363 so_state_get(const struct socket *so)
4366 return (so->so_state);
4370 so_state_set(struct socket *so, int val)
4377 so_options_get(const struct socket *so)
4380 return (so->so_options);
4384 so_options_set(struct socket *so, int val)
4387 so->so_options = val;
4391 so_error_get(const struct socket *so)
4394 return (so->so_error);
4398 so_error_set(struct socket *so, int val)
4405 so_linger_get(const struct socket *so)
4408 return (so->so_linger);
4412 so_linger_set(struct socket *so, int val)
4415 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4416 ("%s: val %d out of range", __func__, val));
4418 so->so_linger = val;
4422 so_protosw_get(const struct socket *so)
4425 return (so->so_proto);
4429 so_protosw_set(struct socket *so, struct protosw *val)
4436 so_sorwakeup(struct socket *so)
4443 so_sowwakeup(struct socket *so)
4450 so_sorwakeup_locked(struct socket *so)
4453 sorwakeup_locked(so);
4457 so_sowwakeup_locked(struct socket *so)
4460 sowwakeup_locked(so);
4464 so_lock(struct socket *so)
4471 so_unlock(struct socket *so)