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/capsicum.h>
116 #include <sys/fcntl.h>
117 #include <sys/limits.h>
118 #include <sys/lock.h>
120 #include <sys/malloc.h>
121 #include <sys/mbuf.h>
122 #include <sys/mutex.h>
123 #include <sys/domain.h>
124 #include <sys/file.h> /* for struct knote */
125 #include <sys/hhook.h>
126 #include <sys/kernel.h>
127 #include <sys/khelp.h>
128 #include <sys/ktls.h>
129 #include <sys/event.h>
130 #include <sys/eventhandler.h>
131 #include <sys/poll.h>
132 #include <sys/proc.h>
133 #include <sys/protosw.h>
134 #include <sys/sbuf.h>
135 #include <sys/socket.h>
136 #include <sys/socketvar.h>
137 #include <sys/resourcevar.h>
138 #include <net/route.h>
139 #include <sys/signalvar.h>
140 #include <sys/stat.h>
142 #include <sys/sysctl.h>
143 #include <sys/taskqueue.h>
146 #include <sys/unpcb.h>
147 #include <sys/jail.h>
148 #include <sys/syslog.h>
149 #include <netinet/in.h>
150 #include <netinet/in_pcb.h>
151 #include <netinet/tcp.h>
153 #include <net/vnet.h>
155 #include <security/mac/mac_framework.h>
159 #ifdef COMPAT_FREEBSD32
160 #include <sys/mount.h>
161 #include <sys/sysent.h>
162 #include <compat/freebsd32/freebsd32.h>
165 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
167 static void so_rdknl_lock(void *);
168 static void so_rdknl_unlock(void *);
169 static void so_rdknl_assert_lock(void *, int);
170 static void so_wrknl_lock(void *);
171 static void so_wrknl_unlock(void *);
172 static void so_wrknl_assert_lock(void *, int);
174 static void filt_sordetach(struct knote *kn);
175 static int filt_soread(struct knote *kn, long hint);
176 static void filt_sowdetach(struct knote *kn);
177 static int filt_sowrite(struct knote *kn, long hint);
178 static int filt_soempty(struct knote *kn, long hint);
179 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
180 fo_kqfilter_t soo_kqfilter;
182 static struct filterops soread_filtops = {
184 .f_detach = filt_sordetach,
185 .f_event = filt_soread,
187 static struct filterops sowrite_filtops = {
189 .f_detach = filt_sowdetach,
190 .f_event = filt_sowrite,
192 static struct filterops soempty_filtops = {
194 .f_detach = filt_sowdetach,
195 .f_event = filt_soempty,
198 so_gen_t so_gencnt; /* generation count for sockets */
200 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
201 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
203 #define VNET_SO_ASSERT(so) \
204 VNET_ASSERT(curvnet != NULL, \
205 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
207 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
208 #define V_socket_hhh VNET(socket_hhh)
211 * Limit on the number of connections in the listen queue waiting
213 * NB: The original sysctl somaxconn is still available but hidden
214 * to prevent confusion about the actual purpose of this number.
216 static u_int somaxconn = SOMAXCONN;
219 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
225 error = sysctl_handle_int(oidp, &val, 0, req);
226 if (error || !req->newptr )
230 * The purpose of the UINT_MAX / 3 limit, is so that the formula
232 * below, will not overflow.
235 if (val < 1 || val > UINT_MAX / 3)
241 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
242 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
243 sysctl_somaxconn, "I",
244 "Maximum listen socket pending connection accept queue size");
245 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
246 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, 0,
247 sizeof(int), sysctl_somaxconn, "I",
248 "Maximum listen socket pending connection accept queue size (compat)");
250 static int numopensockets;
251 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
252 &numopensockets, 0, "Number of open sockets");
255 * accept_mtx locks down per-socket fields relating to accept queues. See
256 * socketvar.h for an annotation of the protected fields of struct socket.
258 struct mtx accept_mtx;
259 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
262 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
265 static struct mtx so_global_mtx;
266 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
269 * General IPC sysctl name space, used by sockets and a variety of other IPC
272 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
276 * Initialize the socket subsystem and set up the socket
279 static uma_zone_t socket_zone;
283 socket_zone_change(void *tag)
286 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
290 socket_hhook_register(int subtype)
293 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
294 &V_socket_hhh[subtype],
295 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
296 printf("%s: WARNING: unable to register hook\n", __func__);
300 socket_hhook_deregister(int subtype)
303 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
304 printf("%s: WARNING: unable to deregister hook\n", __func__);
308 socket_init(void *tag)
311 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
312 NULL, NULL, UMA_ALIGN_PTR, 0);
313 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
314 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
315 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
316 EVENTHANDLER_PRI_FIRST);
318 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
321 socket_vnet_init(const void *unused __unused)
325 /* We expect a contiguous range */
326 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
327 socket_hhook_register(i);
329 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
330 socket_vnet_init, NULL);
333 socket_vnet_uninit(const void *unused __unused)
337 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
338 socket_hhook_deregister(i);
340 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
341 socket_vnet_uninit, NULL);
344 * Initialise maxsockets. This SYSINIT must be run after
348 init_maxsockets(void *ignored)
351 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
352 maxsockets = imax(maxsockets, maxfiles);
354 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
357 * Sysctl to get and set the maximum global sockets limit. Notify protocols
358 * of the change so that they can update their dependent limits as required.
361 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
363 int error, newmaxsockets;
365 newmaxsockets = maxsockets;
366 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
367 if (error == 0 && req->newptr && newmaxsockets != maxsockets) {
368 if (newmaxsockets > maxsockets &&
369 newmaxsockets <= maxfiles) {
370 maxsockets = newmaxsockets;
371 EVENTHANDLER_INVOKE(maxsockets_change);
377 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
378 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &maxsockets, 0,
379 sysctl_maxsockets, "IU",
380 "Maximum number of sockets available");
383 * Socket operation routines. These routines are called by the routines in
384 * sys_socket.c or from a system process, and implement the semantics of
385 * socket operations by switching out to the protocol specific routines.
389 * Get a socket structure from our zone, and initialize it. Note that it
390 * would probably be better to allocate socket and PCB at the same time, but
391 * I'm not convinced that all the protocols can be easily modified to do
394 * soalloc() returns a socket with a ref count of 0.
396 static struct socket *
397 soalloc(struct vnet *vnet)
401 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
405 if (mac_socket_init(so, M_NOWAIT) != 0) {
406 uma_zfree(socket_zone, so);
410 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
411 uma_zfree(socket_zone, so);
416 * The socket locking protocol allows to lock 2 sockets at a time,
417 * however, the first one must be a listening socket. WITNESS lacks
418 * a feature to change class of an existing lock, so we use DUPOK.
420 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
421 mtx_init(&so->so_snd_mtx, "so_snd", NULL, MTX_DEF);
422 mtx_init(&so->so_rcv_mtx, "so_rcv", NULL, MTX_DEF);
423 so->so_rcv.sb_sel = &so->so_rdsel;
424 so->so_snd.sb_sel = &so->so_wrsel;
425 sx_init(&so->so_snd_sx, "so_snd_sx");
426 sx_init(&so->so_rcv_sx, "so_rcv_sx");
427 TAILQ_INIT(&so->so_snd.sb_aiojobq);
428 TAILQ_INIT(&so->so_rcv.sb_aiojobq);
429 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
430 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
432 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
433 __func__, __LINE__, so));
436 /* We shouldn't need the so_global_mtx */
437 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
438 /* Do we need more comprehensive error returns? */
439 uma_zfree(socket_zone, so);
442 mtx_lock(&so_global_mtx);
443 so->so_gencnt = ++so_gencnt;
446 vnet->vnet_sockcnt++;
448 mtx_unlock(&so_global_mtx);
454 * Free the storage associated with a socket at the socket layer, tear down
455 * locks, labels, etc. All protocol state is assumed already to have been
456 * torn down (and possibly never set up) by the caller.
459 sodealloc(struct socket *so)
462 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
463 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
465 mtx_lock(&so_global_mtx);
466 so->so_gencnt = ++so_gencnt;
467 --numopensockets; /* Could be below, but faster here. */
469 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
470 __func__, __LINE__, so));
471 so->so_vnet->vnet_sockcnt--;
473 mtx_unlock(&so_global_mtx);
475 mac_socket_destroy(so);
477 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
479 khelp_destroy_osd(&so->osd);
480 if (SOLISTENING(so)) {
481 if (so->sol_accept_filter != NULL)
482 accept_filt_setopt(so, NULL);
484 if (so->so_rcv.sb_hiwat)
485 (void)chgsbsize(so->so_cred->cr_uidinfo,
486 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
487 if (so->so_snd.sb_hiwat)
488 (void)chgsbsize(so->so_cred->cr_uidinfo,
489 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
490 sx_destroy(&so->so_snd_sx);
491 sx_destroy(&so->so_rcv_sx);
492 mtx_destroy(&so->so_snd_mtx);
493 mtx_destroy(&so->so_rcv_mtx);
496 mtx_destroy(&so->so_lock);
497 uma_zfree(socket_zone, so);
501 * socreate returns a socket with a ref count of 1 and a file descriptor
502 * reference. The socket should be closed with soclose().
505 socreate(int dom, struct socket **aso, int type, int proto,
506 struct ucred *cred, struct thread *td)
513 prp = pffindproto(dom, proto, type);
515 prp = pffindtype(dom, type);
518 /* No support for domain. */
519 if (pffinddomain(dom) == NULL)
520 return (EAFNOSUPPORT);
521 /* No support for socket type. */
522 if (proto == 0 && type != 0)
524 return (EPROTONOSUPPORT);
526 if (prp->pr_usrreqs->pru_attach == NULL ||
527 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
528 return (EPROTONOSUPPORT);
530 if (IN_CAPABILITY_MODE(td) && (prp->pr_flags & PR_CAPATTACH) == 0)
533 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
534 return (EPROTONOSUPPORT);
536 if (prp->pr_type != type)
538 so = soalloc(CRED_TO_VNET(cred));
543 so->so_cred = crhold(cred);
544 if ((prp->pr_domain->dom_family == PF_INET) ||
545 (prp->pr_domain->dom_family == PF_INET6) ||
546 (prp->pr_domain->dom_family == PF_ROUTE))
547 so->so_fibnum = td->td_proc->p_fibnum;
552 mac_socket_create(cred, so);
554 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
555 so_rdknl_assert_lock);
556 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
557 so_wrknl_assert_lock);
558 if ((prp->pr_flags & PR_SOCKBUF) == 0) {
559 so->so_snd.sb_mtx = &so->so_snd_mtx;
560 so->so_rcv.sb_mtx = &so->so_rcv_mtx;
563 * Auto-sizing of socket buffers is managed by the protocols and
564 * the appropriate flags must be set in the pru_attach function.
566 CURVNET_SET(so->so_vnet);
567 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
579 static int regression_sonewconn_earlytest = 1;
580 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
581 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
584 static struct timeval overinterval = { 60, 0 };
585 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
587 "Delay in seconds between warnings for listen socket overflows");
590 * When an attempt at a new connection is noted on a socket which accepts
591 * connections, sonewconn is called. If the connection is possible (subject
592 * to space constraints, etc.) then we allocate a new structure, properly
593 * linked into the data structure of the original socket, and return this.
594 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
596 * Note: the ref count on the socket is 0 on return.
599 sonewconn(struct socket *head, int connstatus)
605 const char localprefix[] = "local:";
606 char descrbuf[SUNPATHLEN + sizeof(localprefix)];
608 char addrbuf[INET6_ADDRSTRLEN];
610 char addrbuf[INET_ADDRSTRLEN];
615 over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
617 if (regression_sonewconn_earlytest && over) {
621 head->sol_overcount++;
622 dolog = !!ratecheck(&head->sol_lastover, &overinterval);
625 * If we're going to log, copy the overflow count and queue
626 * length from the listen socket before dropping the lock.
627 * Also, reset the overflow count.
630 overcount = head->sol_overcount;
631 head->sol_overcount = 0;
632 qlen = head->sol_qlen;
634 SOLISTEN_UNLOCK(head);
638 * Try to print something descriptive about the
639 * socket for the error message.
641 sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
643 switch (head->so_proto->pr_domain->dom_family) {
644 #if defined(INET) || defined(INET6)
650 if (head->so_proto->pr_domain->dom_family ==
652 (sotoinpcb(head)->inp_inc.inc_flags &
655 &sotoinpcb(head)->inp_inc.inc6_laddr);
656 sbuf_printf(&descrsb, "[%s]", addrbuf);
662 sotoinpcb(head)->inp_inc.inc_laddr,
664 sbuf_cat(&descrsb, addrbuf);
667 sbuf_printf(&descrsb, ":%hu (proto %u)",
668 ntohs(sotoinpcb(head)->inp_inc.inc_lport),
669 head->so_proto->pr_protocol);
671 #endif /* INET || INET6 */
673 sbuf_cat(&descrsb, localprefix);
674 if (sotounpcb(head)->unp_addr != NULL)
676 sotounpcb(head)->unp_addr->sun_len -
677 offsetof(struct sockaddr_un,
683 sotounpcb(head)->unp_addr->sun_path,
686 sbuf_cat(&descrsb, "(unknown)");
691 * If we can't print something more specific, at least
692 * print the domain name.
694 if (sbuf_finish(&descrsb) != 0 ||
695 sbuf_len(&descrsb) <= 0) {
696 sbuf_clear(&descrsb);
698 head->so_proto->pr_domain->dom_name ?:
700 sbuf_finish(&descrsb);
702 KASSERT(sbuf_len(&descrsb) > 0,
703 ("%s: sbuf creation failed", __func__));
704 if (head->so_cred == 0) {
706 "%s: pcb %p (%s): Listen queue overflow: "
707 "%i already in queue awaiting acceptance "
708 "(%d occurrences)\n",
709 __func__, head->so_pcb, sbuf_data(&descrsb),
712 log(LOG_DEBUG, "%s: pcb %p (%s): Listen queue overflow: "
713 "%i already in queue awaiting acceptance "
714 "(%d occurrences), euid %d, rgid %d, jail %s\n",
715 __func__, head->so_pcb, sbuf_data(&descrsb),
717 head->so_cred->cr_uid, head->so_cred->cr_rgid,
718 head->so_cred->cr_prison ?
719 head->so_cred->cr_prison->pr_name :
722 sbuf_delete(&descrsb);
729 SOLISTEN_UNLOCK(head);
730 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
732 so = soalloc(head->so_vnet);
734 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
735 "limit reached or out of memory\n",
736 __func__, head->so_pcb);
739 so->so_listen = head;
740 so->so_type = head->so_type;
741 so->so_options = head->so_options & ~SO_ACCEPTCONN;
742 so->so_linger = head->so_linger;
743 so->so_state = head->so_state;
744 so->so_fibnum = head->so_fibnum;
745 so->so_proto = head->so_proto;
746 so->so_cred = crhold(head->so_cred);
748 mac_socket_newconn(head, so);
750 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
751 so_rdknl_assert_lock);
752 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
753 so_wrknl_assert_lock);
754 VNET_SO_ASSERT(head);
755 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
757 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
758 __func__, head->so_pcb);
761 if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) {
762 so->so_snd.sb_mtx = &so->so_snd_mtx;
763 so->so_rcv.sb_mtx = &so->so_rcv_mtx;
765 /* Socket starts with a reference from the listen queue. */
766 refcount_init(&so->so_count, 1);
767 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
768 MPASS(refcount_release(&so->so_count));
770 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
771 __func__, head->so_pcb);
774 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
775 so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
776 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
777 so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
778 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
779 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
782 if (head->sol_accept_filter != NULL)
784 so->so_state |= connstatus;
785 soref(head); /* A socket on (in)complete queue refs head. */
787 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
788 so->so_qstate = SQ_COMP;
790 solisten_wakeup(head); /* unlocks */
793 * Keep removing sockets from the head until there's room for
794 * us to insert on the tail. In pre-locking revisions, this
795 * was a simple if(), but as we could be racing with other
796 * threads and soabort() requires dropping locks, we must
797 * loop waiting for the condition to be true.
799 while (head->sol_incqlen > head->sol_qlimit) {
802 sp = TAILQ_FIRST(&head->sol_incomp);
803 TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
806 sp->so_qstate = SQ_NONE;
807 sp->so_listen = NULL;
809 sorele_locked(head); /* does SOLISTEN_UNLOCK, head stays */
813 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
814 so->so_qstate = SQ_INCOMP;
816 SOLISTEN_UNLOCK(head);
821 #if defined(SCTP) || defined(SCTP_SUPPORT)
823 * Socket part of sctp_peeloff(). Detach a new socket from an
824 * association. The new socket is returned with a reference.
827 sopeeloff(struct socket *head)
831 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
832 __func__, __LINE__, head));
833 so = soalloc(head->so_vnet);
835 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
836 "limit reached or out of memory\n",
837 __func__, head->so_pcb);
840 so->so_type = head->so_type;
841 so->so_options = head->so_options;
842 so->so_linger = head->so_linger;
843 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
844 so->so_fibnum = head->so_fibnum;
845 so->so_proto = head->so_proto;
846 so->so_cred = crhold(head->so_cred);
848 mac_socket_newconn(head, so);
850 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
851 so_rdknl_assert_lock);
852 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
853 so_wrknl_assert_lock);
854 VNET_SO_ASSERT(head);
855 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
857 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
858 __func__, head->so_pcb);
861 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
863 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
864 __func__, head->so_pcb);
867 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
868 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
869 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
870 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
871 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
872 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
881 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
885 CURVNET_SET(so->so_vnet);
886 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
892 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
896 CURVNET_SET(so->so_vnet);
897 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
903 * solisten() transitions a socket from a non-listening state to a listening
904 * state, but can also be used to update the listen queue depth on an
905 * existing listen socket. The protocol will call back into the sockets
906 * layer using solisten_proto_check() and solisten_proto() to check and set
907 * socket-layer listen state. Call backs are used so that the protocol can
908 * acquire both protocol and socket layer locks in whatever order is required
911 * Protocol implementors are advised to hold the socket lock across the
912 * socket-layer test and set to avoid races at the socket layer.
915 solisten(struct socket *so, int backlog, struct thread *td)
919 CURVNET_SET(so->so_vnet);
920 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
926 * Prepare for a call to solisten_proto(). Acquire all socket buffer locks in
927 * order to interlock with socket I/O.
930 solisten_proto_check(struct socket *so)
932 SOCK_LOCK_ASSERT(so);
934 if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
935 SS_ISDISCONNECTING)) != 0)
939 * Sleeping is not permitted here, so simply fail if userspace is
940 * attempting to transmit or receive on the socket. This kind of
941 * transient failure is not ideal, but it should occur only if userspace
942 * is misusing the socket interfaces.
944 if (!sx_try_xlock(&so->so_snd_sx))
946 if (!sx_try_xlock(&so->so_rcv_sx)) {
947 sx_xunlock(&so->so_snd_sx);
950 mtx_lock(&so->so_snd_mtx);
951 mtx_lock(&so->so_rcv_mtx);
953 /* Interlock with soo_aio_queue(). */
954 if (!SOLISTENING(so) &&
955 ((so->so_snd.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0 ||
956 (so->so_rcv.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0)) {
957 solisten_proto_abort(so);
964 * Undo the setup done by solisten_proto_check().
967 solisten_proto_abort(struct socket *so)
969 mtx_unlock(&so->so_snd_mtx);
970 mtx_unlock(&so->so_rcv_mtx);
971 sx_xunlock(&so->so_snd_sx);
972 sx_xunlock(&so->so_rcv_sx);
976 solisten_proto(struct socket *so, int backlog)
978 int sbrcv_lowat, sbsnd_lowat;
979 u_int sbrcv_hiwat, sbsnd_hiwat;
980 short sbrcv_flags, sbsnd_flags;
981 sbintime_t sbrcv_timeo, sbsnd_timeo;
983 SOCK_LOCK_ASSERT(so);
984 KASSERT((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
985 SS_ISDISCONNECTING)) == 0,
986 ("%s: bad socket state %p", __func__, so));
992 * Change this socket to listening state.
994 sbrcv_lowat = so->so_rcv.sb_lowat;
995 sbsnd_lowat = so->so_snd.sb_lowat;
996 sbrcv_hiwat = so->so_rcv.sb_hiwat;
997 sbsnd_hiwat = so->so_snd.sb_hiwat;
998 sbrcv_flags = so->so_rcv.sb_flags;
999 sbsnd_flags = so->so_snd.sb_flags;
1000 sbrcv_timeo = so->so_rcv.sb_timeo;
1001 sbsnd_timeo = so->so_snd.sb_timeo;
1003 sbdestroy(so, SO_SND);
1004 sbdestroy(so, SO_RCV);
1008 sizeof(struct socket) - offsetof(struct socket, so_rcv));
1011 so->sol_sbrcv_lowat = sbrcv_lowat;
1012 so->sol_sbsnd_lowat = sbsnd_lowat;
1013 so->sol_sbrcv_hiwat = sbrcv_hiwat;
1014 so->sol_sbsnd_hiwat = sbsnd_hiwat;
1015 so->sol_sbrcv_flags = sbrcv_flags;
1016 so->sol_sbsnd_flags = sbsnd_flags;
1017 so->sol_sbrcv_timeo = sbrcv_timeo;
1018 so->sol_sbsnd_timeo = sbsnd_timeo;
1020 so->sol_qlen = so->sol_incqlen = 0;
1021 TAILQ_INIT(&so->sol_incomp);
1022 TAILQ_INIT(&so->sol_comp);
1024 so->sol_accept_filter = NULL;
1025 so->sol_accept_filter_arg = NULL;
1026 so->sol_accept_filter_str = NULL;
1028 so->sol_upcall = NULL;
1029 so->sol_upcallarg = NULL;
1031 so->so_options |= SO_ACCEPTCONN;
1034 if (backlog < 0 || backlog > somaxconn)
1035 backlog = somaxconn;
1036 so->sol_qlimit = backlog;
1038 mtx_unlock(&so->so_snd_mtx);
1039 mtx_unlock(&so->so_rcv_mtx);
1040 sx_xunlock(&so->so_snd_sx);
1041 sx_xunlock(&so->so_rcv_sx);
1045 * Wakeup listeners/subsystems once we have a complete connection.
1046 * Enters with lock, returns unlocked.
1049 solisten_wakeup(struct socket *sol)
1052 if (sol->sol_upcall != NULL)
1053 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
1055 selwakeuppri(&sol->so_rdsel, PSOCK);
1056 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
1058 SOLISTEN_UNLOCK(sol);
1059 wakeup_one(&sol->sol_comp);
1060 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
1061 pgsigio(&sol->so_sigio, SIGIO, 0);
1065 * Return single connection off a listening socket queue. Main consumer of
1066 * the function is kern_accept4(). Some modules, that do their own accept
1067 * management also use the function. The socket reference held by the
1068 * listen queue is handed to the caller.
1070 * Listening socket must be locked on entry and is returned unlocked on
1072 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1075 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1080 SOLISTEN_LOCK_ASSERT(head);
1082 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1083 head->so_error == 0) {
1084 error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH,
1087 SOLISTEN_UNLOCK(head);
1091 if (head->so_error) {
1092 error = head->so_error;
1094 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1095 error = EWOULDBLOCK;
1099 SOLISTEN_UNLOCK(head);
1102 so = TAILQ_FIRST(&head->sol_comp);
1104 KASSERT(so->so_qstate == SQ_COMP,
1105 ("%s: so %p not SQ_COMP", __func__, so));
1107 so->so_qstate = SQ_NONE;
1108 so->so_listen = NULL;
1109 TAILQ_REMOVE(&head->sol_comp, so, so_list);
1110 if (flags & ACCEPT4_INHERIT)
1111 so->so_state |= (head->so_state & SS_NBIO);
1113 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1115 sorele_locked(head);
1122 * Free socket upon release of the very last reference.
1125 sofree(struct socket *so)
1127 struct protosw *pr = so->so_proto;
1129 SOCK_LOCK_ASSERT(so);
1130 KASSERT(refcount_load(&so->so_count) == 0,
1131 ("%s: so %p has references", __func__, so));
1132 KASSERT(SOLISTENING(so) || so->so_qstate == SQ_NONE,
1133 ("%s: so %p is on listen queue", __func__, so));
1137 if (so->so_dtor != NULL)
1141 if ((pr->pr_flags & PR_RIGHTS) && !SOLISTENING(so)) {
1142 MPASS(pr->pr_domain->dom_dispose != NULL);
1143 (*pr->pr_domain->dom_dispose)(so);
1145 if (pr->pr_usrreqs->pru_detach != NULL)
1146 (*pr->pr_usrreqs->pru_detach)(so);
1149 * From this point on, we assume that no other references to this
1150 * socket exist anywhere else in the stack. Therefore, no locks need
1151 * to be acquired or held.
1153 if (!(pr->pr_flags & PR_SOCKBUF) && !SOLISTENING(so)) {
1154 sbdestroy(so, SO_SND);
1155 sbdestroy(so, SO_RCV);
1157 seldrain(&so->so_rdsel);
1158 seldrain(&so->so_wrsel);
1159 knlist_destroy(&so->so_rdsel.si_note);
1160 knlist_destroy(&so->so_wrsel.si_note);
1165 * Release a reference on a socket while holding the socket lock.
1166 * Unlocks the socket lock before returning.
1169 sorele_locked(struct socket *so)
1171 SOCK_LOCK_ASSERT(so);
1172 if (refcount_release(&so->so_count))
1179 * Close a socket on last file table reference removal. Initiate disconnect
1180 * if connected. Free socket when disconnect complete.
1182 * This function will sorele() the socket. Note that soclose() may be called
1183 * prior to the ref count reaching zero. The actual socket structure will
1184 * not be freed until the ref count reaches zero.
1187 soclose(struct socket *so)
1189 struct accept_queue lqueue;
1191 bool listening, last __diagused;
1193 CURVNET_SET(so->so_vnet);
1194 funsetown(&so->so_sigio);
1195 if (so->so_state & SS_ISCONNECTED) {
1196 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1197 error = sodisconnect(so);
1199 if (error == ENOTCONN)
1205 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1206 if ((so->so_state & SS_ISDISCONNECTING) &&
1207 (so->so_state & SS_NBIO))
1209 while (so->so_state & SS_ISCONNECTED) {
1210 error = tsleep(&so->so_timeo,
1211 PSOCK | PCATCH, "soclos",
1212 so->so_linger * hz);
1220 if (so->so_proto->pr_usrreqs->pru_close != NULL)
1221 (*so->so_proto->pr_usrreqs->pru_close)(so);
1224 if ((listening = SOLISTENING(so))) {
1227 TAILQ_INIT(&lqueue);
1228 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1229 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1231 so->sol_qlen = so->sol_incqlen = 0;
1233 TAILQ_FOREACH(sp, &lqueue, so_list) {
1235 sp->so_qstate = SQ_NONE;
1236 sp->so_listen = NULL;
1238 last = refcount_release(&so->so_count);
1239 KASSERT(!last, ("%s: released last reference for %p",
1245 struct socket *sp, *tsp;
1247 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp)
1255 * soabort() is used to abruptly tear down a connection, such as when a
1256 * resource limit is reached (listen queue depth exceeded), or if a listen
1257 * socket is closed while there are sockets waiting to be accepted.
1259 * This interface is tricky, because it is called on an unreferenced socket,
1260 * and must be called only by a thread that has actually removed the socket
1261 * from the listen queue it was on. Likely this thread holds the last
1262 * reference on the socket and soabort() will proceed with sofree(). But
1263 * it might be not the last, as the sockets on the listen queues are seen
1264 * from the protocol side.
1266 * This interface will call into the protocol code, so must not be called
1267 * with any socket locks held. Protocols do call it while holding their own
1268 * recursible protocol mutexes, but this is something that should be subject
1269 * to review in the future.
1271 * Usually socket should have a single reference left, but this is not a
1272 * requirement. In the past, when we have had named references for file
1273 * descriptor and protocol, we asserted that none of them are being held.
1276 soabort(struct socket *so)
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)
1292 CURVNET_SET(so->so_vnet);
1293 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1299 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1302 return (soconnectat(AT_FDCWD, so, nam, td));
1306 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1310 CURVNET_SET(so->so_vnet);
1312 * If protocol is connection-based, can only connect once.
1313 * Otherwise, if connected, try to disconnect first. This allows
1314 * user to disconnect by connecting to, e.g., a null address.
1316 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1317 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1318 (error = sodisconnect(so)))) {
1322 * Prevent accumulated error from previous connection from
1326 if (fd == AT_FDCWD) {
1327 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1330 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1340 soconnect2(struct socket *so1, struct socket *so2)
1344 CURVNET_SET(so1->so_vnet);
1345 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1351 sodisconnect(struct socket *so)
1355 if ((so->so_state & SS_ISCONNECTED) == 0)
1357 if (so->so_state & SS_ISDISCONNECTING)
1360 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1365 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1366 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1370 int clen = 0, error, dontroute;
1372 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1373 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1374 ("sosend_dgram: !PR_ATOMIC"));
1377 resid = uio->uio_resid;
1379 resid = top->m_pkthdr.len;
1381 * In theory resid should be unsigned. However, space must be
1382 * signed, as it might be less than 0 if we over-committed, and we
1383 * must use a signed comparison of space and resid. On the other
1384 * hand, a negative resid causes us to loop sending 0-length
1385 * segments to the protocol.
1393 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1395 td->td_ru.ru_msgsnd++;
1396 if (control != NULL)
1397 clen = control->m_len;
1399 SOCKBUF_LOCK(&so->so_snd);
1400 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1401 SOCKBUF_UNLOCK(&so->so_snd);
1406 error = so->so_error;
1408 SOCKBUF_UNLOCK(&so->so_snd);
1411 if ((so->so_state & SS_ISCONNECTED) == 0) {
1413 * `sendto' and `sendmsg' is allowed on a connection-based
1414 * socket if it supports implied connect. Return ENOTCONN if
1415 * not connected and no address is supplied.
1417 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1418 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1419 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1420 !(resid == 0 && clen != 0)) {
1421 SOCKBUF_UNLOCK(&so->so_snd);
1425 } else if (addr == NULL) {
1426 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1429 error = EDESTADDRREQ;
1430 SOCKBUF_UNLOCK(&so->so_snd);
1436 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1437 * problem and need fixing.
1439 space = sbspace(&so->so_snd);
1440 if (flags & MSG_OOB)
1443 SOCKBUF_UNLOCK(&so->so_snd);
1444 if (resid > space) {
1450 if (flags & MSG_EOR)
1451 top->m_flags |= M_EOR;
1454 * Copy the data from userland into a mbuf chain.
1455 * If no data is to be copied in, a single empty mbuf
1458 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1459 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1461 error = EFAULT; /* only possible error */
1464 space -= resid - uio->uio_resid;
1465 resid = uio->uio_resid;
1467 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1469 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1474 so->so_options |= SO_DONTROUTE;
1478 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1479 * of date. We could have received a reset packet in an interrupt or
1480 * maybe we slept while doing page faults in uiomove() etc. We could
1481 * probably recheck again inside the locking protection here, but
1482 * there are probably other places that this also happens. We must
1486 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1487 (flags & MSG_OOB) ? PRUS_OOB :
1489 * If the user set MSG_EOF, the protocol understands this flag and
1490 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1492 ((flags & MSG_EOF) &&
1493 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1496 /* If there is more to send set PRUS_MORETOCOME */
1497 (flags & MSG_MORETOCOME) ||
1498 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1499 top, addr, control, td);
1502 so->so_options &= ~SO_DONTROUTE;
1511 if (control != NULL)
1517 * Send on a socket. If send must go all at once and message is larger than
1518 * send buffering, then hard error. Lock against other senders. If must go
1519 * all at once and not enough room now, then inform user that this would
1520 * block and do nothing. Otherwise, if nonblocking, send as much as
1521 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1522 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1523 * in mbuf chain must be small enough to send all at once.
1525 * Returns nonzero on error, timeout or signal; callers must check for short
1526 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1530 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1531 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1535 int clen = 0, error, dontroute;
1536 int atomic = sosendallatonce(so) || top;
1539 struct ktls_session *tls;
1540 int tls_enq_cnt, tls_pruflag;
1544 tls_rtype = TLS_RLTYPE_APP;
1547 resid = uio->uio_resid;
1548 else if ((top->m_flags & M_PKTHDR) != 0)
1549 resid = top->m_pkthdr.len;
1551 resid = m_length(top, NULL);
1553 * In theory resid should be unsigned. However, space must be
1554 * signed, as it might be less than 0 if we over-committed, and we
1555 * must use a signed comparison of space and resid. On the other
1556 * hand, a negative resid causes us to loop sending 0-length
1557 * segments to the protocol.
1559 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1560 * type sockets since that's an error.
1562 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1568 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1569 (so->so_proto->pr_flags & PR_ATOMIC);
1571 td->td_ru.ru_msgsnd++;
1572 if (control != NULL)
1573 clen = control->m_len;
1575 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1581 tls = ktls_hold(so->so_snd.sb_tls_info);
1583 if (tls->mode == TCP_TLS_MODE_SW)
1584 tls_pruflag = PRUS_NOTREADY;
1586 if (control != NULL) {
1587 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1589 if (clen >= sizeof(*cm) &&
1590 cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1591 tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1599 if (resid == 0 && !ktls_permit_empty_frames(tls)) {
1608 SOCKBUF_LOCK(&so->so_snd);
1609 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1610 SOCKBUF_UNLOCK(&so->so_snd);
1615 error = so->so_error;
1617 SOCKBUF_UNLOCK(&so->so_snd);
1620 if ((so->so_state & SS_ISCONNECTED) == 0) {
1622 * `sendto' and `sendmsg' is allowed on a connection-
1623 * based socket if it supports implied connect.
1624 * Return ENOTCONN if not connected and no address is
1627 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1628 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1629 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1630 !(resid == 0 && clen != 0)) {
1631 SOCKBUF_UNLOCK(&so->so_snd);
1635 } else if (addr == NULL) {
1636 SOCKBUF_UNLOCK(&so->so_snd);
1637 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1640 error = EDESTADDRREQ;
1644 space = sbspace(&so->so_snd);
1645 if (flags & MSG_OOB)
1647 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1648 clen > so->so_snd.sb_hiwat) {
1649 SOCKBUF_UNLOCK(&so->so_snd);
1653 if (space < resid + clen &&
1654 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1655 if ((so->so_state & SS_NBIO) ||
1656 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1657 SOCKBUF_UNLOCK(&so->so_snd);
1658 error = EWOULDBLOCK;
1661 error = sbwait(so, SO_SND);
1662 SOCKBUF_UNLOCK(&so->so_snd);
1667 SOCKBUF_UNLOCK(&so->so_snd);
1672 if (flags & MSG_EOR)
1673 top->m_flags |= M_EOR;
1676 ktls_frame(top, tls, &tls_enq_cnt,
1678 tls_rtype = TLS_RLTYPE_APP;
1683 * Copy the data from userland into a mbuf
1684 * chain. If resid is 0, which can happen
1685 * only if we have control to send, then
1686 * a single empty mbuf is returned. This
1687 * is a workaround to prevent protocol send
1692 top = m_uiotombuf(uio, M_WAITOK, space,
1693 tls->params.max_frame_len,
1695 ((flags & MSG_EOR) ? M_EOR : 0));
1697 ktls_frame(top, tls,
1698 &tls_enq_cnt, tls_rtype);
1700 tls_rtype = TLS_RLTYPE_APP;
1703 top = m_uiotombuf(uio, M_WAITOK, space,
1704 (atomic ? max_hdr : 0),
1705 (atomic ? M_PKTHDR : 0) |
1706 ((flags & MSG_EOR) ? M_EOR : 0));
1708 error = EFAULT; /* only possible error */
1711 space -= resid - uio->uio_resid;
1712 resid = uio->uio_resid;
1716 so->so_options |= SO_DONTROUTE;
1720 * XXX all the SBS_CANTSENDMORE checks previously
1721 * done could be out of date. We could have received
1722 * a reset packet in an interrupt or maybe we slept
1723 * while doing page faults in uiomove() etc. We
1724 * could probably recheck again inside the locking
1725 * protection here, but there are probably other
1726 * places that this also happens. We must rethink
1731 pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1733 * If the user set MSG_EOF, the protocol understands
1734 * this flag and nothing left to send then use
1735 * PRU_SEND_EOF instead of PRU_SEND.
1737 ((flags & MSG_EOF) &&
1738 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1741 /* If there is more to send set PRUS_MORETOCOME. */
1742 (flags & MSG_MORETOCOME) ||
1743 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1746 pru_flag |= tls_pruflag;
1749 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1750 pru_flag, top, addr, control, td);
1754 so->so_options &= ~SO_DONTROUTE;
1759 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1765 ktls_enqueue(top, so, tls_enq_cnt);
1774 } while (resid && space > 0);
1778 SOCK_IO_SEND_UNLOCK(so);
1786 if (control != NULL)
1792 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1793 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1797 CURVNET_SET(so->so_vnet);
1798 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1799 top, control, flags, td);
1805 * The part of soreceive() that implements reading non-inline out-of-band
1806 * data from a socket. For more complete comments, see soreceive(), from
1807 * which this code originated.
1809 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1810 * unable to return an mbuf chain to the caller.
1813 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1815 struct protosw *pr = so->so_proto;
1819 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1822 m = m_get(M_WAITOK, MT_DATA);
1823 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1827 error = uiomove(mtod(m, void *),
1828 (int) min(uio->uio_resid, m->m_len), uio);
1830 } while (uio->uio_resid && error == 0 && m);
1838 * Following replacement or removal of the first mbuf on the first mbuf chain
1839 * of a socket buffer, push necessary state changes back into the socket
1840 * buffer so that other consumers see the values consistently. 'nextrecord'
1841 * is the callers locally stored value of the original value of
1842 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1843 * NOTE: 'nextrecord' may be NULL.
1845 static __inline void
1846 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1849 SOCKBUF_LOCK_ASSERT(sb);
1851 * First, update for the new value of nextrecord. If necessary, make
1852 * it the first record.
1854 if (sb->sb_mb != NULL)
1855 sb->sb_mb->m_nextpkt = nextrecord;
1857 sb->sb_mb = nextrecord;
1860 * Now update any dependent socket buffer fields to reflect the new
1861 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1862 * addition of a second clause that takes care of the case where
1863 * sb_mb has been updated, but remains the last record.
1865 if (sb->sb_mb == NULL) {
1866 sb->sb_mbtail = NULL;
1867 sb->sb_lastrecord = NULL;
1868 } else if (sb->sb_mb->m_nextpkt == NULL)
1869 sb->sb_lastrecord = sb->sb_mb;
1873 * Implement receive operations on a socket. We depend on the way that
1874 * records are added to the sockbuf by sbappend. In particular, each record
1875 * (mbufs linked through m_next) must begin with an address if the protocol
1876 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1877 * data, and then zero or more mbufs of data. In order to allow parallelism
1878 * between network receive and copying to user space, as well as avoid
1879 * sleeping with a mutex held, we release the socket buffer mutex during the
1880 * user space copy. Although the sockbuf is locked, new data may still be
1881 * appended, and thus we must maintain consistency of the sockbuf during that
1884 * The caller may receive the data as a single mbuf chain by supplying an
1885 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1886 * the count in uio_resid.
1889 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1890 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1892 struct mbuf *m, **mp;
1893 int flags, error, offset;
1895 struct protosw *pr = so->so_proto;
1896 struct mbuf *nextrecord;
1898 ssize_t orig_resid = uio->uio_resid;
1899 bool report_real_len = false;
1904 if (controlp != NULL)
1906 if (flagsp != NULL) {
1907 report_real_len = *flagsp & MSG_TRUNC;
1908 *flagsp &= ~MSG_TRUNC;
1909 flags = *flagsp &~ MSG_EOR;
1912 if (flags & MSG_OOB)
1913 return (soreceive_rcvoob(so, uio, flags));
1916 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1917 && uio->uio_resid) {
1919 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1922 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1927 SOCKBUF_LOCK(&so->so_rcv);
1928 m = so->so_rcv.sb_mb;
1930 * If we have less data than requested, block awaiting more (subject
1931 * to any timeout) if:
1932 * 1. the current count is less than the low water mark, or
1933 * 2. MSG_DONTWAIT is not set
1935 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1936 sbavail(&so->so_rcv) < uio->uio_resid) &&
1937 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1938 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1939 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1940 ("receive: m == %p sbavail == %u",
1941 m, sbavail(&so->so_rcv)));
1942 if (so->so_error || so->so_rerror) {
1946 error = so->so_error;
1948 error = so->so_rerror;
1949 if ((flags & MSG_PEEK) == 0) {
1955 SOCKBUF_UNLOCK(&so->so_rcv);
1958 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1959 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1963 else if (so->so_rcv.sb_tlsdcc == 0 &&
1964 so->so_rcv.sb_tlscc == 0) {
1968 SOCKBUF_UNLOCK(&so->so_rcv);
1972 for (; m != NULL; m = m->m_next)
1973 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1974 m = so->so_rcv.sb_mb;
1977 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
1978 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
1979 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1980 SOCKBUF_UNLOCK(&so->so_rcv);
1984 if (uio->uio_resid == 0 && !report_real_len) {
1985 SOCKBUF_UNLOCK(&so->so_rcv);
1988 if ((so->so_state & SS_NBIO) ||
1989 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1990 SOCKBUF_UNLOCK(&so->so_rcv);
1991 error = EWOULDBLOCK;
1994 SBLASTRECORDCHK(&so->so_rcv);
1995 SBLASTMBUFCHK(&so->so_rcv);
1996 error = sbwait(so, SO_RCV);
1997 SOCKBUF_UNLOCK(&so->so_rcv);
2004 * From this point onward, we maintain 'nextrecord' as a cache of the
2005 * pointer to the next record in the socket buffer. We must keep the
2006 * various socket buffer pointers and local stack versions of the
2007 * pointers in sync, pushing out modifications before dropping the
2008 * socket buffer mutex, and re-reading them when picking it up.
2010 * Otherwise, we will race with the network stack appending new data
2011 * or records onto the socket buffer by using inconsistent/stale
2012 * versions of the field, possibly resulting in socket buffer
2015 * By holding the high-level sblock(), we prevent simultaneous
2016 * readers from pulling off the front of the socket buffer.
2018 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2020 uio->uio_td->td_ru.ru_msgrcv++;
2021 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2022 SBLASTRECORDCHK(&so->so_rcv);
2023 SBLASTMBUFCHK(&so->so_rcv);
2024 nextrecord = m->m_nextpkt;
2025 if (pr->pr_flags & PR_ADDR) {
2026 KASSERT(m->m_type == MT_SONAME,
2027 ("m->m_type == %d", m->m_type));
2030 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2032 if (flags & MSG_PEEK) {
2035 sbfree(&so->so_rcv, m);
2036 so->so_rcv.sb_mb = m_free(m);
2037 m = so->so_rcv.sb_mb;
2038 sockbuf_pushsync(&so->so_rcv, nextrecord);
2043 * Process one or more MT_CONTROL mbufs present before any data mbufs
2044 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
2045 * just copy the data; if !MSG_PEEK, we call into the protocol to
2046 * perform externalization (or freeing if controlp == NULL).
2048 if (m != NULL && m->m_type == MT_CONTROL) {
2049 struct mbuf *cm = NULL, *cmn;
2050 struct mbuf **cme = &cm;
2052 struct cmsghdr *cmsg;
2053 struct tls_get_record tgr;
2056 * For MSG_TLSAPPDATA, check for an alert record.
2057 * If found, return ENXIO without removing
2058 * it from the receive queue. This allows a subsequent
2059 * call without MSG_TLSAPPDATA to receive it.
2060 * Note that, for TLS, there should only be a single
2061 * control mbuf with the TLS_GET_RECORD message in it.
2063 if (flags & MSG_TLSAPPDATA) {
2064 cmsg = mtod(m, struct cmsghdr *);
2065 if (cmsg->cmsg_type == TLS_GET_RECORD &&
2066 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2067 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2068 if (__predict_false(tgr.tls_type ==
2069 TLS_RLTYPE_ALERT)) {
2070 SOCKBUF_UNLOCK(&so->so_rcv);
2079 if (flags & MSG_PEEK) {
2080 if (controlp != NULL) {
2081 *controlp = m_copym(m, 0, m->m_len,
2083 controlp = &(*controlp)->m_next;
2087 sbfree(&so->so_rcv, m);
2088 so->so_rcv.sb_mb = m->m_next;
2091 cme = &(*cme)->m_next;
2092 m = so->so_rcv.sb_mb;
2094 } while (m != NULL && m->m_type == MT_CONTROL);
2095 if ((flags & MSG_PEEK) == 0)
2096 sockbuf_pushsync(&so->so_rcv, nextrecord);
2097 while (cm != NULL) {
2100 if (pr->pr_domain->dom_externalize != NULL) {
2101 SOCKBUF_UNLOCK(&so->so_rcv);
2103 error = (*pr->pr_domain->dom_externalize)
2104 (cm, controlp, flags);
2105 SOCKBUF_LOCK(&so->so_rcv);
2106 } else if (controlp != NULL)
2110 if (controlp != NULL) {
2111 while (*controlp != NULL)
2112 controlp = &(*controlp)->m_next;
2117 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2119 nextrecord = so->so_rcv.sb_mb;
2123 if ((flags & MSG_PEEK) == 0) {
2124 KASSERT(m->m_nextpkt == nextrecord,
2125 ("soreceive: post-control, nextrecord !sync"));
2126 if (nextrecord == NULL) {
2127 KASSERT(so->so_rcv.sb_mb == m,
2128 ("soreceive: post-control, sb_mb!=m"));
2129 KASSERT(so->so_rcv.sb_lastrecord == m,
2130 ("soreceive: post-control, lastrecord!=m"));
2134 if (type == MT_OOBDATA)
2137 if ((flags & MSG_PEEK) == 0) {
2138 KASSERT(so->so_rcv.sb_mb == nextrecord,
2139 ("soreceive: sb_mb != nextrecord"));
2140 if (so->so_rcv.sb_mb == NULL) {
2141 KASSERT(so->so_rcv.sb_lastrecord == NULL,
2142 ("soreceive: sb_lastercord != NULL"));
2146 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2147 SBLASTRECORDCHK(&so->so_rcv);
2148 SBLASTMBUFCHK(&so->so_rcv);
2151 * Now continue to read any data mbufs off of the head of the socket
2152 * buffer until the read request is satisfied. Note that 'type' is
2153 * used to store the type of any mbuf reads that have happened so far
2154 * such that soreceive() can stop reading if the type changes, which
2155 * causes soreceive() to return only one of regular data and inline
2156 * out-of-band data in a single socket receive operation.
2160 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2163 * If the type of mbuf has changed since the last mbuf
2164 * examined ('type'), end the receive operation.
2166 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2167 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2168 if (type != m->m_type)
2170 } else if (type == MT_OOBDATA)
2173 KASSERT(m->m_type == MT_DATA,
2174 ("m->m_type == %d", m->m_type));
2175 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2176 len = uio->uio_resid;
2177 if (so->so_oobmark && len > so->so_oobmark - offset)
2178 len = so->so_oobmark - offset;
2179 if (len > m->m_len - moff)
2180 len = m->m_len - moff;
2182 * If mp is set, just pass back the mbufs. Otherwise copy
2183 * them out via the uio, then free. Sockbuf must be
2184 * consistent here (points to current mbuf, it points to next
2185 * record) when we drop priority; we must note any additions
2186 * to the sockbuf when we block interrupts again.
2189 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2190 SBLASTRECORDCHK(&so->so_rcv);
2191 SBLASTMBUFCHK(&so->so_rcv);
2192 SOCKBUF_UNLOCK(&so->so_rcv);
2193 if ((m->m_flags & M_EXTPG) != 0)
2194 error = m_unmapped_uiomove(m, moff, uio,
2197 error = uiomove(mtod(m, char *) + moff,
2199 SOCKBUF_LOCK(&so->so_rcv);
2202 * The MT_SONAME mbuf has already been removed
2203 * from the record, so it is necessary to
2204 * remove the data mbufs, if any, to preserve
2205 * the invariant in the case of PR_ADDR that
2206 * requires MT_SONAME mbufs at the head of
2209 if (pr->pr_flags & PR_ATOMIC &&
2210 ((flags & MSG_PEEK) == 0))
2211 (void)sbdroprecord_locked(&so->so_rcv);
2212 SOCKBUF_UNLOCK(&so->so_rcv);
2216 uio->uio_resid -= len;
2217 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2218 if (len == m->m_len - moff) {
2219 if (m->m_flags & M_EOR)
2221 if (flags & MSG_PEEK) {
2225 nextrecord = m->m_nextpkt;
2226 sbfree(&so->so_rcv, m);
2228 m->m_nextpkt = NULL;
2231 so->so_rcv.sb_mb = m = m->m_next;
2234 so->so_rcv.sb_mb = m_free(m);
2235 m = so->so_rcv.sb_mb;
2237 sockbuf_pushsync(&so->so_rcv, nextrecord);
2238 SBLASTRECORDCHK(&so->so_rcv);
2239 SBLASTMBUFCHK(&so->so_rcv);
2242 if (flags & MSG_PEEK)
2246 if (flags & MSG_DONTWAIT) {
2247 *mp = m_copym(m, 0, len,
2251 * m_copym() couldn't
2253 * Adjust uio_resid back
2255 * down by len bytes,
2256 * which we didn't end
2257 * up "copying" over).
2259 uio->uio_resid += len;
2263 SOCKBUF_UNLOCK(&so->so_rcv);
2264 *mp = m_copym(m, 0, len,
2266 SOCKBUF_LOCK(&so->so_rcv);
2269 sbcut_locked(&so->so_rcv, len);
2272 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2273 if (so->so_oobmark) {
2274 if ((flags & MSG_PEEK) == 0) {
2275 so->so_oobmark -= len;
2276 if (so->so_oobmark == 0) {
2277 so->so_rcv.sb_state |= SBS_RCVATMARK;
2282 if (offset == so->so_oobmark)
2286 if (flags & MSG_EOR)
2289 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2290 * must not quit until "uio->uio_resid == 0" or an error
2291 * termination. If a signal/timeout occurs, return with a
2292 * short count but without error. Keep sockbuf locked
2293 * against other readers.
2295 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2296 !sosendallatonce(so) && nextrecord == NULL) {
2297 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2298 if (so->so_error || so->so_rerror ||
2299 so->so_rcv.sb_state & SBS_CANTRCVMORE)
2302 * Notify the protocol that some data has been
2303 * drained before blocking.
2305 if (pr->pr_flags & PR_WANTRCVD) {
2306 SOCKBUF_UNLOCK(&so->so_rcv);
2308 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2309 SOCKBUF_LOCK(&so->so_rcv);
2311 SBLASTRECORDCHK(&so->so_rcv);
2312 SBLASTMBUFCHK(&so->so_rcv);
2314 * We could receive some data while was notifying
2315 * the protocol. Skip blocking in this case.
2317 if (so->so_rcv.sb_mb == NULL) {
2318 error = sbwait(so, SO_RCV);
2320 SOCKBUF_UNLOCK(&so->so_rcv);
2324 m = so->so_rcv.sb_mb;
2326 nextrecord = m->m_nextpkt;
2330 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2331 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2332 if (report_real_len)
2333 uio->uio_resid -= m_length(m, NULL) - moff;
2335 if ((flags & MSG_PEEK) == 0)
2336 (void) sbdroprecord_locked(&so->so_rcv);
2338 if ((flags & MSG_PEEK) == 0) {
2341 * First part is an inline SB_EMPTY_FIXUP(). Second
2342 * part makes sure sb_lastrecord is up-to-date if
2343 * there is still data in the socket buffer.
2345 so->so_rcv.sb_mb = nextrecord;
2346 if (so->so_rcv.sb_mb == NULL) {
2347 so->so_rcv.sb_mbtail = NULL;
2348 so->so_rcv.sb_lastrecord = NULL;
2349 } else if (nextrecord->m_nextpkt == NULL)
2350 so->so_rcv.sb_lastrecord = nextrecord;
2352 SBLASTRECORDCHK(&so->so_rcv);
2353 SBLASTMBUFCHK(&so->so_rcv);
2355 * If soreceive() is being done from the socket callback,
2356 * then don't need to generate ACK to peer to update window,
2357 * since ACK will be generated on return to TCP.
2359 if (!(flags & MSG_SOCALLBCK) &&
2360 (pr->pr_flags & PR_WANTRCVD)) {
2361 SOCKBUF_UNLOCK(&so->so_rcv);
2363 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2364 SOCKBUF_LOCK(&so->so_rcv);
2367 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2368 if (orig_resid == uio->uio_resid && orig_resid &&
2369 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2370 SOCKBUF_UNLOCK(&so->so_rcv);
2373 SOCKBUF_UNLOCK(&so->so_rcv);
2378 SOCK_IO_RECV_UNLOCK(so);
2383 * Optimized version of soreceive() for stream (TCP) sockets.
2386 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2387 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2389 int len = 0, error = 0, flags, oresid;
2391 struct mbuf *m, *n = NULL;
2393 /* We only do stream sockets. */
2394 if (so->so_type != SOCK_STREAM)
2399 flags = *flagsp &~ MSG_EOR;
2402 if (controlp != NULL)
2404 if (flags & MSG_OOB)
2405 return (soreceive_rcvoob(so, uio, flags));
2413 * KTLS store TLS records as records with a control message to
2414 * describe the framing.
2416 * We check once here before acquiring locks to optimize the
2419 if (sb->sb_tls_info != NULL)
2420 return (soreceive_generic(so, psa, uio, mp0, controlp,
2424 /* Prevent other readers from entering the socket. */
2425 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2431 if (sb->sb_tls_info != NULL) {
2433 SOCK_IO_RECV_UNLOCK(so);
2434 return (soreceive_generic(so, psa, uio, mp0, controlp,
2439 /* Easy one, no space to copyout anything. */
2440 if (uio->uio_resid == 0) {
2444 oresid = uio->uio_resid;
2446 /* We will never ever get anything unless we are or were connected. */
2447 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2453 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2455 /* Abort if socket has reported problems. */
2457 if (sbavail(sb) > 0)
2459 if (oresid > uio->uio_resid)
2461 error = so->so_error;
2462 if (!(flags & MSG_PEEK))
2467 /* Door is closed. Deliver what is left, if any. */
2468 if (sb->sb_state & SBS_CANTRCVMORE) {
2469 if (sbavail(sb) > 0)
2475 /* Socket buffer is empty and we shall not block. */
2476 if (sbavail(sb) == 0 &&
2477 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2482 /* Socket buffer got some data that we shall deliver now. */
2483 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2484 ((so->so_state & SS_NBIO) ||
2485 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2486 sbavail(sb) >= sb->sb_lowat ||
2487 sbavail(sb) >= uio->uio_resid ||
2488 sbavail(sb) >= sb->sb_hiwat) ) {
2492 /* On MSG_WAITALL we must wait until all data or error arrives. */
2493 if ((flags & MSG_WAITALL) &&
2494 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2498 * Wait and block until (more) data comes in.
2499 * NB: Drops the sockbuf lock during wait.
2501 error = sbwait(so, SO_RCV);
2507 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2508 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2509 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2513 uio->uio_td->td_ru.ru_msgrcv++;
2515 /* Fill uio until full or current end of socket buffer is reached. */
2516 len = min(uio->uio_resid, sbavail(sb));
2518 /* Dequeue as many mbufs as possible. */
2519 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2523 m_cat(*mp0, sb->sb_mb);
2525 m != NULL && m->m_len <= len;
2527 KASSERT(!(m->m_flags & M_NOTAVAIL),
2528 ("%s: m %p not available", __func__, m));
2530 uio->uio_resid -= m->m_len;
2536 sb->sb_lastrecord = sb->sb_mb;
2537 if (sb->sb_mb == NULL)
2540 /* Copy the remainder. */
2542 KASSERT(sb->sb_mb != NULL,
2543 ("%s: len > 0 && sb->sb_mb empty", __func__));
2545 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2547 len = 0; /* Don't flush data from sockbuf. */
2549 uio->uio_resid -= len;
2560 /* NB: Must unlock socket buffer as uiomove may sleep. */
2562 error = m_mbuftouio(uio, sb->sb_mb, len);
2567 SBLASTRECORDCHK(sb);
2571 * Remove the delivered data from the socket buffer unless we
2572 * were only peeking.
2574 if (!(flags & MSG_PEEK)) {
2576 sbdrop_locked(sb, len);
2578 /* Notify protocol that we drained some data. */
2579 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2580 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2581 !(flags & MSG_SOCALLBCK))) {
2584 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2590 * For MSG_WAITALL we may have to loop again and wait for
2591 * more data to come in.
2593 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2596 SBLASTRECORDCHK(sb);
2599 SOCK_IO_RECV_UNLOCK(so);
2604 * Optimized version of soreceive() for simple datagram cases from userspace.
2605 * Unlike in the stream case, we're able to drop a datagram if copyout()
2606 * fails, and because we handle datagrams atomically, we don't need to use a
2607 * sleep lock to prevent I/O interlacing.
2610 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2611 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2613 struct mbuf *m, *m2;
2616 struct protosw *pr = so->so_proto;
2617 struct mbuf *nextrecord;
2621 if (controlp != NULL)
2624 flags = *flagsp &~ MSG_EOR;
2629 * For any complicated cases, fall back to the full
2630 * soreceive_generic().
2632 if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC)))
2633 return (soreceive_generic(so, psa, uio, mp0, controlp,
2637 * Enforce restrictions on use.
2639 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2640 ("soreceive_dgram: wantrcvd"));
2641 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2642 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2643 ("soreceive_dgram: SBS_RCVATMARK"));
2644 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2645 ("soreceive_dgram: P_CONNREQUIRED"));
2648 * Loop blocking while waiting for a datagram.
2650 SOCKBUF_LOCK(&so->so_rcv);
2651 while ((m = so->so_rcv.sb_mb) == NULL) {
2652 KASSERT(sbavail(&so->so_rcv) == 0,
2653 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2654 sbavail(&so->so_rcv)));
2656 error = so->so_error;
2658 SOCKBUF_UNLOCK(&so->so_rcv);
2661 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2662 uio->uio_resid == 0) {
2663 SOCKBUF_UNLOCK(&so->so_rcv);
2666 if ((so->so_state & SS_NBIO) ||
2667 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2668 SOCKBUF_UNLOCK(&so->so_rcv);
2669 return (EWOULDBLOCK);
2671 SBLASTRECORDCHK(&so->so_rcv);
2672 SBLASTMBUFCHK(&so->so_rcv);
2673 error = sbwait(so, SO_RCV);
2675 SOCKBUF_UNLOCK(&so->so_rcv);
2679 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2682 uio->uio_td->td_ru.ru_msgrcv++;
2683 SBLASTRECORDCHK(&so->so_rcv);
2684 SBLASTMBUFCHK(&so->so_rcv);
2685 nextrecord = m->m_nextpkt;
2686 if (nextrecord == NULL) {
2687 KASSERT(so->so_rcv.sb_lastrecord == m,
2688 ("soreceive_dgram: lastrecord != m"));
2691 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2692 ("soreceive_dgram: m_nextpkt != nextrecord"));
2695 * Pull 'm' and its chain off the front of the packet queue.
2697 so->so_rcv.sb_mb = NULL;
2698 sockbuf_pushsync(&so->so_rcv, nextrecord);
2701 * Walk 'm's chain and free that many bytes from the socket buffer.
2703 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2704 sbfree(&so->so_rcv, m2);
2707 * Do a few last checks before we let go of the lock.
2709 SBLASTRECORDCHK(&so->so_rcv);
2710 SBLASTMBUFCHK(&so->so_rcv);
2711 SOCKBUF_UNLOCK(&so->so_rcv);
2713 if (pr->pr_flags & PR_ADDR) {
2714 KASSERT(m->m_type == MT_SONAME,
2715 ("m->m_type == %d", m->m_type));
2717 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2722 /* XXXRW: Can this happen? */
2727 * Packet to copyout() is now in 'm' and it is disconnected from the
2730 * Process one or more MT_CONTROL mbufs present before any data mbufs
2731 * in the first mbuf chain on the socket buffer. We call into the
2732 * protocol to perform externalization (or freeing if controlp ==
2733 * NULL). In some cases there can be only MT_CONTROL mbufs without
2736 if (m->m_type == MT_CONTROL) {
2737 struct mbuf *cm = NULL, *cmn;
2738 struct mbuf **cme = &cm;
2744 cme = &(*cme)->m_next;
2746 } while (m != NULL && m->m_type == MT_CONTROL);
2747 while (cm != NULL) {
2750 if (pr->pr_domain->dom_externalize != NULL) {
2751 error = (*pr->pr_domain->dom_externalize)
2752 (cm, controlp, flags);
2753 } else if (controlp != NULL)
2757 if (controlp != NULL) {
2758 while (*controlp != NULL)
2759 controlp = &(*controlp)->m_next;
2764 KASSERT(m == NULL || m->m_type == MT_DATA,
2765 ("soreceive_dgram: !data"));
2766 while (m != NULL && uio->uio_resid > 0) {
2767 len = uio->uio_resid;
2770 error = uiomove(mtod(m, char *), (int)len, uio);
2775 if (len == m->m_len)
2792 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2793 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2797 CURVNET_SET(so->so_vnet);
2798 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2799 mp0, controlp, flagsp));
2805 soshutdown(struct socket *so, int how)
2808 int error, soerror_enotconn;
2810 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2813 soerror_enotconn = 0;
2816 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2818 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2819 * invoked on a datagram sockets, however historically we would
2820 * actually tear socket down. This is known to be leveraged by
2821 * some applications to unblock process waiting in recvXXX(2)
2822 * by other process that it shares that socket with. Try to meet
2823 * both backward-compatibility and POSIX requirements by forcing
2824 * ENOTCONN but still asking protocol to perform pru_shutdown().
2826 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) {
2830 soerror_enotconn = 1;
2833 if (SOLISTENING(so)) {
2834 if (how != SHUT_WR) {
2835 so->so_error = ECONNABORTED;
2836 solisten_wakeup(so); /* unlocks so */
2844 CURVNET_SET(so->so_vnet);
2846 if (pr->pr_usrreqs->pru_flush != NULL)
2847 (*pr->pr_usrreqs->pru_flush)(so, how);
2850 if (how != SHUT_RD) {
2851 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2852 wakeup(&so->so_timeo);
2854 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2856 wakeup(&so->so_timeo);
2860 return (soerror_enotconn ? ENOTCONN : 0);
2864 sorflush(struct socket *so)
2872 * Dislodge threads currently blocked in receive and wait to acquire
2873 * a lock against other simultaneous readers before clearing the
2874 * socket buffer. Don't let our acquire be interrupted by a signal
2875 * despite any existing socket disposition on interruptable waiting.
2879 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
2881 KASSERT(SOLISTENING(so),
2882 ("%s: soiolock(%p) failed", __func__, so));
2887 if (pr->pr_flags & PR_RIGHTS) {
2888 MPASS(pr->pr_domain->dom_dispose != NULL);
2889 (*pr->pr_domain->dom_dispose)(so);
2891 sbrelease(so, SO_RCV);
2892 SOCK_IO_RECV_UNLOCK(so);
2898 * Wrapper for Socket established helper hook.
2899 * Parameters: socket, context of the hook point, hook id.
2902 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2904 struct socket_hhook_data hhook_data = {
2911 CURVNET_SET(so->so_vnet);
2912 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2915 /* Ugly but needed, since hhooks return void for now */
2916 return (hhook_data.status);
2920 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2921 * additional variant to handle the case where the option value needs to be
2922 * some kind of integer, but not a specific size. In addition to their use
2923 * here, these functions are also called by the protocol-level pr_ctloutput()
2927 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2932 * If the user gives us more than we wanted, we ignore it, but if we
2933 * don't get the minimum length the caller wants, we return EINVAL.
2934 * On success, sopt->sopt_valsize is set to however much we actually
2937 if ((valsize = sopt->sopt_valsize) < minlen)
2940 sopt->sopt_valsize = valsize = len;
2942 if (sopt->sopt_td != NULL)
2943 return (copyin(sopt->sopt_val, buf, valsize));
2945 bcopy(sopt->sopt_val, buf, valsize);
2950 * Kernel version of setsockopt(2).
2952 * XXX: optlen is size_t, not socklen_t
2955 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2958 struct sockopt sopt;
2960 sopt.sopt_level = level;
2961 sopt.sopt_name = optname;
2962 sopt.sopt_dir = SOPT_SET;
2963 sopt.sopt_val = optval;
2964 sopt.sopt_valsize = optlen;
2965 sopt.sopt_td = NULL;
2966 return (sosetopt(so, &sopt));
2970 sosetopt(struct socket *so, struct sockopt *sopt)
2975 sbintime_t val, *valp;
2981 CURVNET_SET(so->so_vnet);
2983 if (sopt->sopt_level != SOL_SOCKET) {
2984 if (so->so_proto->pr_ctloutput != NULL)
2985 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2987 error = ENOPROTOOPT;
2989 switch (sopt->sopt_name) {
2990 case SO_ACCEPTFILTER:
2991 error = accept_filt_setopt(so, sopt);
2997 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3000 if (l.l_linger < 0 ||
3001 l.l_linger > USHRT_MAX ||
3002 l.l_linger > (INT_MAX / hz)) {
3007 so->so_linger = l.l_linger;
3009 so->so_options |= SO_LINGER;
3011 so->so_options &= ~SO_LINGER;
3018 case SO_USELOOPBACK:
3022 case SO_REUSEPORT_LB:
3030 error = sooptcopyin(sopt, &optval, sizeof optval,
3036 so->so_options |= sopt->sopt_name;
3038 so->so_options &= ~sopt->sopt_name;
3043 error = sooptcopyin(sopt, &optval, sizeof optval,
3048 if (optval < 0 || optval >= rt_numfibs) {
3052 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3053 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3054 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3055 so->so_fibnum = optval;
3060 case SO_USER_COOKIE:
3061 error = sooptcopyin(sopt, &val32, sizeof val32,
3065 so->so_user_cookie = val32;
3072 error = sooptcopyin(sopt, &optval, sizeof optval,
3078 * Values < 1 make no sense for any of these options,
3086 error = sbsetopt(so, sopt->sopt_name, optval);
3091 #ifdef COMPAT_FREEBSD32
3092 if (SV_CURPROC_FLAG(SV_ILP32)) {
3093 struct timeval32 tv32;
3095 error = sooptcopyin(sopt, &tv32, sizeof tv32,
3097 CP(tv32, tv, tv_sec);
3098 CP(tv32, tv, tv_usec);
3101 error = sooptcopyin(sopt, &tv, sizeof tv,
3105 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3106 tv.tv_usec >= 1000000) {
3110 if (tv.tv_sec > INT32_MAX)
3115 valp = sopt->sopt_name == SO_SNDTIMEO ?
3116 (SOLISTENING(so) ? &so->sol_sbsnd_timeo :
3117 &so->so_snd.sb_timeo) :
3118 (SOLISTENING(so) ? &so->sol_sbrcv_timeo :
3119 &so->so_rcv.sb_timeo);
3126 error = sooptcopyin(sopt, &extmac, sizeof extmac,
3130 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3138 error = sooptcopyin(sopt, &optval, sizeof optval,
3142 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3146 so->so_ts_clock = optval;
3149 case SO_MAX_PACING_RATE:
3150 error = sooptcopyin(sopt, &val32, sizeof(val32),
3154 so->so_max_pacing_rate = val32;
3158 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3159 error = hhook_run_socket(so, sopt,
3162 error = ENOPROTOOPT;
3165 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3166 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
3174 * Helper routine for getsockopt.
3177 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3185 * Documented get behavior is that we always return a value, possibly
3186 * truncated to fit in the user's buffer. Traditional behavior is
3187 * that we always tell the user precisely how much we copied, rather
3188 * than something useful like the total amount we had available for
3189 * her. Note that this interface is not idempotent; the entire
3190 * answer must be generated ahead of time.
3192 valsize = min(len, sopt->sopt_valsize);
3193 sopt->sopt_valsize = valsize;
3194 if (sopt->sopt_val != NULL) {
3195 if (sopt->sopt_td != NULL)
3196 error = copyout(buf, sopt->sopt_val, valsize);
3198 bcopy(buf, sopt->sopt_val, valsize);
3204 sogetopt(struct socket *so, struct sockopt *sopt)
3213 CURVNET_SET(so->so_vnet);
3215 if (sopt->sopt_level != SOL_SOCKET) {
3216 if (so->so_proto->pr_ctloutput != NULL)
3217 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3219 error = ENOPROTOOPT;
3223 switch (sopt->sopt_name) {
3224 case SO_ACCEPTFILTER:
3225 error = accept_filt_getopt(so, sopt);
3230 l.l_onoff = so->so_options & SO_LINGER;
3231 l.l_linger = so->so_linger;
3233 error = sooptcopyout(sopt, &l, sizeof l);
3236 case SO_USELOOPBACK:
3242 case SO_REUSEPORT_LB:
3252 optval = so->so_options & sopt->sopt_name;
3254 error = sooptcopyout(sopt, &optval, sizeof optval);
3258 optval = so->so_proto->pr_domain->dom_family;
3262 optval = so->so_type;
3266 optval = so->so_proto->pr_protocol;
3272 optval = so->so_error;
3275 optval = so->so_rerror;
3282 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3283 so->so_snd.sb_hiwat;
3287 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3288 so->so_rcv.sb_hiwat;
3292 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3293 so->so_snd.sb_lowat;
3297 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3298 so->so_rcv.sb_lowat;
3304 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3305 (SOLISTENING(so) ? so->sol_sbsnd_timeo :
3306 so->so_snd.sb_timeo) :
3307 (SOLISTENING(so) ? so->sol_sbrcv_timeo :
3308 so->so_rcv.sb_timeo));
3310 #ifdef COMPAT_FREEBSD32
3311 if (SV_CURPROC_FLAG(SV_ILP32)) {
3312 struct timeval32 tv32;
3314 CP(tv, tv32, tv_sec);
3315 CP(tv, tv32, tv_usec);
3316 error = sooptcopyout(sopt, &tv32, sizeof tv32);
3319 error = sooptcopyout(sopt, &tv, sizeof tv);
3324 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3328 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3332 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3340 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3344 error = mac_getsockopt_peerlabel(
3345 sopt->sopt_td->td_ucred, so, &extmac);
3348 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3354 case SO_LISTENQLIMIT:
3355 optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3359 optval = SOLISTENING(so) ? so->sol_qlen : 0;
3362 case SO_LISTENINCQLEN:
3363 optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3367 optval = so->so_ts_clock;
3370 case SO_MAX_PACING_RATE:
3371 optval = so->so_max_pacing_rate;
3375 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3376 error = hhook_run_socket(so, sopt,
3379 error = ENOPROTOOPT;
3391 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3393 struct mbuf *m, *m_prev;
3394 int sopt_size = sopt->sopt_valsize;
3396 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3399 if (sopt_size > MLEN) {
3400 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3401 if ((m->m_flags & M_EXT) == 0) {
3405 m->m_len = min(MCLBYTES, sopt_size);
3407 m->m_len = min(MLEN, sopt_size);
3409 sopt_size -= m->m_len;
3414 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3419 if (sopt_size > MLEN) {
3420 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3422 if ((m->m_flags & M_EXT) == 0) {
3427 m->m_len = min(MCLBYTES, sopt_size);
3429 m->m_len = min(MLEN, sopt_size);
3431 sopt_size -= m->m_len;
3439 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3441 struct mbuf *m0 = m;
3443 if (sopt->sopt_val == NULL)
3445 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3446 if (sopt->sopt_td != NULL) {
3449 error = copyin(sopt->sopt_val, mtod(m, char *),
3456 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3457 sopt->sopt_valsize -= m->m_len;
3458 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3461 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3462 panic("ip6_sooptmcopyin");
3467 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3469 struct mbuf *m0 = m;
3472 if (sopt->sopt_val == NULL)
3474 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3475 if (sopt->sopt_td != NULL) {
3478 error = copyout(mtod(m, char *), sopt->sopt_val,
3485 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3486 sopt->sopt_valsize -= m->m_len;
3487 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3488 valsize += m->m_len;
3492 /* enough soopt buffer should be given from user-land */
3496 sopt->sopt_valsize = valsize;
3501 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3502 * out-of-band data, which will then notify socket consumers.
3505 sohasoutofband(struct socket *so)
3508 if (so->so_sigio != NULL)
3509 pgsigio(&so->so_sigio, SIGURG, 0);
3510 selwakeuppri(&so->so_rdsel, PSOCK);
3514 sopoll(struct socket *so, int events, struct ucred *active_cred,
3519 * We do not need to set or assert curvnet as long as everyone uses
3522 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3527 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3533 if (SOLISTENING(so)) {
3534 if (!(events & (POLLIN | POLLRDNORM)))
3536 else if (!TAILQ_EMPTY(&so->sol_comp))
3537 revents = events & (POLLIN | POLLRDNORM);
3538 else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3539 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3541 selrecord(td, &so->so_rdsel);
3546 SOCK_SENDBUF_LOCK(so);
3547 SOCK_RECVBUF_LOCK(so);
3548 if (events & (POLLIN | POLLRDNORM))
3549 if (soreadabledata(so))
3550 revents |= events & (POLLIN | POLLRDNORM);
3551 if (events & (POLLOUT | POLLWRNORM))
3552 if (sowriteable(so))
3553 revents |= events & (POLLOUT | POLLWRNORM);
3554 if (events & (POLLPRI | POLLRDBAND))
3555 if (so->so_oobmark ||
3556 (so->so_rcv.sb_state & SBS_RCVATMARK))
3557 revents |= events & (POLLPRI | POLLRDBAND);
3558 if ((events & POLLINIGNEOF) == 0) {
3559 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3560 revents |= events & (POLLIN | POLLRDNORM);
3561 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3565 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3566 revents |= events & POLLRDHUP;
3569 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
3570 selrecord(td, &so->so_rdsel);
3571 so->so_rcv.sb_flags |= SB_SEL;
3573 if (events & (POLLOUT | POLLWRNORM)) {
3574 selrecord(td, &so->so_wrsel);
3575 so->so_snd.sb_flags |= SB_SEL;
3578 SOCK_RECVBUF_UNLOCK(so);
3579 SOCK_SENDBUF_UNLOCK(so);
3586 soo_kqfilter(struct file *fp, struct knote *kn)
3588 struct socket *so = kn->kn_fp->f_data;
3593 switch (kn->kn_filter) {
3595 kn->kn_fop = &soread_filtops;
3596 knl = &so->so_rdsel.si_note;
3601 kn->kn_fop = &sowrite_filtops;
3602 knl = &so->so_wrsel.si_note;
3607 kn->kn_fop = &soempty_filtops;
3608 knl = &so->so_wrsel.si_note;
3617 if (SOLISTENING(so)) {
3618 knlist_add(knl, kn, 1);
3620 SOCK_BUF_LOCK(so, which);
3621 knlist_add(knl, kn, 1);
3622 sb->sb_flags |= SB_KNOTE;
3623 SOCK_BUF_UNLOCK(so, which);
3630 * Some routines that return EOPNOTSUPP for entry points that are not
3631 * supported by a protocol. Fill in as needed.
3634 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3641 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3648 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3655 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3662 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3670 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3677 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3685 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3692 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3693 struct ifnet *ifp, struct thread *td)
3700 pru_disconnect_notsupp(struct socket *so)
3707 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3714 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3721 pru_rcvd_notsupp(struct socket *so, int flags)
3728 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3735 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3736 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3739 if (control != NULL)
3741 if ((flags & PRUS_NOTREADY) == 0)
3743 return (EOPNOTSUPP);
3747 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3750 return (EOPNOTSUPP);
3754 * This isn't really a ``null'' operation, but it's the default one and
3755 * doesn't do anything destructive.
3758 pru_sense_null(struct socket *so, struct stat *sb)
3761 sb->st_blksize = so->so_snd.sb_hiwat;
3766 pru_shutdown_notsupp(struct socket *so)
3773 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3780 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3781 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3788 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3789 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3796 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3804 filt_sordetach(struct knote *kn)
3806 struct socket *so = kn->kn_fp->f_data;
3809 knlist_remove(&so->so_rdsel.si_note, kn, 1);
3810 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3811 so->so_rcv.sb_flags &= ~SB_KNOTE;
3812 so_rdknl_unlock(so);
3817 filt_soread(struct knote *kn, long hint)
3821 so = kn->kn_fp->f_data;
3823 if (SOLISTENING(so)) {
3824 SOCK_LOCK_ASSERT(so);
3825 kn->kn_data = so->sol_qlen;
3827 kn->kn_flags |= EV_EOF;
3828 kn->kn_fflags = so->so_error;
3831 return (!TAILQ_EMPTY(&so->sol_comp));
3834 SOCK_RECVBUF_LOCK_ASSERT(so);
3836 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3837 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3838 kn->kn_flags |= EV_EOF;
3839 kn->kn_fflags = so->so_error;
3841 } else if (so->so_error || so->so_rerror)
3844 if (kn->kn_sfflags & NOTE_LOWAT) {
3845 if (kn->kn_data >= kn->kn_sdata)
3847 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3850 /* This hook returning non-zero indicates an event, not error */
3851 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3855 filt_sowdetach(struct knote *kn)
3857 struct socket *so = kn->kn_fp->f_data;
3860 knlist_remove(&so->so_wrsel.si_note, kn, 1);
3861 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3862 so->so_snd.sb_flags &= ~SB_KNOTE;
3863 so_wrknl_unlock(so);
3868 filt_sowrite(struct knote *kn, long hint)
3872 so = kn->kn_fp->f_data;
3874 if (SOLISTENING(so))
3877 SOCK_SENDBUF_LOCK_ASSERT(so);
3878 kn->kn_data = sbspace(&so->so_snd);
3880 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3882 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3883 kn->kn_flags |= EV_EOF;
3884 kn->kn_fflags = so->so_error;
3886 } else if (so->so_error) /* temporary udp error */
3888 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3889 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3891 else if (kn->kn_sfflags & NOTE_LOWAT)
3892 return (kn->kn_data >= kn->kn_sdata);
3894 return (kn->kn_data >= so->so_snd.sb_lowat);
3898 filt_soempty(struct knote *kn, long hint)
3902 so = kn->kn_fp->f_data;
3904 if (SOLISTENING(so))
3907 SOCK_SENDBUF_LOCK_ASSERT(so);
3908 kn->kn_data = sbused(&so->so_snd);
3910 if (kn->kn_data == 0)
3917 socheckuid(struct socket *so, uid_t uid)
3922 if (so->so_cred->cr_uid != uid)
3928 * These functions are used by protocols to notify the socket layer (and its
3929 * consumers) of state changes in the sockets driven by protocol-side events.
3933 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3935 * Normal sequence from the active (originating) side is that
3936 * soisconnecting() is called during processing of connect() call, resulting
3937 * in an eventual call to soisconnected() if/when the connection is
3938 * established. When the connection is torn down soisdisconnecting() is
3939 * called during processing of disconnect() call, and soisdisconnected() is
3940 * called when the connection to the peer is totally severed. The semantics
3941 * of these routines are such that connectionless protocols can call
3942 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3943 * calls when setting up a ``connection'' takes no time.
3945 * From the passive side, a socket is created with two queues of sockets:
3946 * so_incomp for connections in progress and so_comp for connections already
3947 * made and awaiting user acceptance. As a protocol is preparing incoming
3948 * connections, it creates a socket structure queued on so_incomp by calling
3949 * sonewconn(). When the connection is established, soisconnected() is
3950 * called, and transfers the socket structure to so_comp, making it available
3953 * If a socket is closed with sockets on either so_incomp or so_comp, these
3954 * sockets are dropped.
3956 * If higher-level protocols are implemented in the kernel, the wakeups done
3957 * here will sometimes cause software-interrupt process scheduling.
3960 soisconnecting(struct socket *so)
3964 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3965 so->so_state |= SS_ISCONNECTING;
3970 soisconnected(struct socket *so)
3972 bool last __diagused;
3975 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3976 so->so_state |= SS_ISCONNECTED;
3978 if (so->so_qstate == SQ_INCOMP) {
3979 struct socket *head = so->so_listen;
3982 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3984 * Promoting a socket from incomplete queue to complete, we
3985 * need to go through reverse order of locking. We first do
3986 * trylock, and if that doesn't succeed, we go the hard way
3987 * leaving a reference and rechecking consistency after proper
3990 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3993 SOLISTEN_LOCK(head);
3995 if (__predict_false(head != so->so_listen)) {
3997 * The socket went off the listen queue,
3998 * should be lost race to close(2) of sol.
3999 * The socket is about to soabort().
4002 sorele_locked(head);
4005 last = refcount_release(&head->so_count);
4006 KASSERT(!last, ("%s: released last reference for %p",
4010 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4011 TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4012 head->sol_incqlen--;
4013 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4015 so->so_qstate = SQ_COMP;
4017 solisten_wakeup(head); /* unlocks */
4019 SOCK_RECVBUF_LOCK(so);
4020 soupcall_set(so, SO_RCV,
4021 head->sol_accept_filter->accf_callback,
4022 head->sol_accept_filter_arg);
4023 so->so_options &= ~SO_ACCEPTFILTER;
4024 ret = head->sol_accept_filter->accf_callback(so,
4025 head->sol_accept_filter_arg, M_NOWAIT);
4026 if (ret == SU_ISCONNECTED) {
4027 soupcall_clear(so, SO_RCV);
4028 SOCK_RECVBUF_UNLOCK(so);
4031 SOCK_RECVBUF_UNLOCK(so);
4033 SOLISTEN_UNLOCK(head);
4038 wakeup(&so->so_timeo);
4044 soisdisconnecting(struct socket *so)
4048 so->so_state &= ~SS_ISCONNECTING;
4049 so->so_state |= SS_ISDISCONNECTING;
4051 if (!SOLISTENING(so)) {
4052 SOCK_RECVBUF_LOCK(so);
4053 socantrcvmore_locked(so);
4054 SOCK_SENDBUF_LOCK(so);
4055 socantsendmore_locked(so);
4058 wakeup(&so->so_timeo);
4062 soisdisconnected(struct socket *so)
4068 * There is at least one reader of so_state that does not
4069 * acquire socket lock, namely soreceive_generic(). Ensure
4070 * that it never sees all flags that track connection status
4071 * cleared, by ordering the update with a barrier semantic of
4072 * our release thread fence.
4074 so->so_state |= SS_ISDISCONNECTED;
4075 atomic_thread_fence_rel();
4076 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4078 if (!SOLISTENING(so)) {
4080 SOCK_RECVBUF_LOCK(so);
4081 socantrcvmore_locked(so);
4082 SOCK_SENDBUF_LOCK(so);
4083 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4084 socantsendmore_locked(so);
4087 wakeup(&so->so_timeo);
4091 soiolock(struct socket *so, struct sx *sx, int flags)
4095 KASSERT((flags & SBL_VALID) == flags,
4096 ("soiolock: invalid flags %#x", flags));
4098 if ((flags & SBL_WAIT) != 0) {
4099 if ((flags & SBL_NOINTR) != 0) {
4102 error = sx_xlock_sig(sx);
4106 } else if (!sx_try_xlock(sx)) {
4107 return (EWOULDBLOCK);
4110 if (__predict_false(SOLISTENING(so))) {
4118 soiounlock(struct sx *sx)
4124 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4127 sodupsockaddr(const struct sockaddr *sa, int mflags)
4129 struct sockaddr *sa2;
4131 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4133 bcopy(sa, sa2, sa->sa_len);
4138 * Register per-socket destructor.
4141 sodtor_set(struct socket *so, so_dtor_t *func)
4144 SOCK_LOCK_ASSERT(so);
4149 * Register per-socket buffer upcalls.
4152 soupcall_set(struct socket *so, sb_which which, so_upcall_t func, void *arg)
4156 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4166 SOCK_BUF_LOCK_ASSERT(so, which);
4167 sb->sb_upcall = func;
4168 sb->sb_upcallarg = arg;
4169 sb->sb_flags |= SB_UPCALL;
4173 soupcall_clear(struct socket *so, sb_which which)
4177 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4187 SOCK_BUF_LOCK_ASSERT(so, which);
4188 KASSERT(sb->sb_upcall != NULL,
4189 ("%s: so %p no upcall to clear", __func__, so));
4190 sb->sb_upcall = NULL;
4191 sb->sb_upcallarg = NULL;
4192 sb->sb_flags &= ~SB_UPCALL;
4196 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4199 SOLISTEN_LOCK_ASSERT(so);
4200 so->sol_upcall = func;
4201 so->sol_upcallarg = arg;
4205 so_rdknl_lock(void *arg)
4207 struct socket *so = arg;
4210 if (SOLISTENING(so)) {
4213 SOCK_RECVBUF_LOCK(so);
4214 if (__predict_false(SOLISTENING(so))) {
4215 SOCK_RECVBUF_UNLOCK(so);
4222 so_rdknl_unlock(void *arg)
4224 struct socket *so = arg;
4226 if (SOLISTENING(so))
4227 SOLISTEN_UNLOCK(so);
4229 SOCK_RECVBUF_UNLOCK(so);
4233 so_rdknl_assert_lock(void *arg, int what)
4235 struct socket *so = arg;
4237 if (what == LA_LOCKED) {
4238 if (SOLISTENING(so))
4239 SOLISTEN_LOCK_ASSERT(so);
4241 SOCK_RECVBUF_LOCK_ASSERT(so);
4243 if (SOLISTENING(so))
4244 SOLISTEN_UNLOCK_ASSERT(so);
4246 SOCK_RECVBUF_UNLOCK_ASSERT(so);
4251 so_wrknl_lock(void *arg)
4253 struct socket *so = arg;
4256 if (SOLISTENING(so)) {
4259 SOCK_SENDBUF_LOCK(so);
4260 if (__predict_false(SOLISTENING(so))) {
4261 SOCK_SENDBUF_UNLOCK(so);
4268 so_wrknl_unlock(void *arg)
4270 struct socket *so = arg;
4272 if (SOLISTENING(so))
4273 SOLISTEN_UNLOCK(so);
4275 SOCK_SENDBUF_UNLOCK(so);
4279 so_wrknl_assert_lock(void *arg, int what)
4281 struct socket *so = arg;
4283 if (what == LA_LOCKED) {
4284 if (SOLISTENING(so))
4285 SOLISTEN_LOCK_ASSERT(so);
4287 SOCK_SENDBUF_LOCK_ASSERT(so);
4289 if (SOLISTENING(so))
4290 SOLISTEN_UNLOCK_ASSERT(so);
4292 SOCK_SENDBUF_UNLOCK_ASSERT(so);
4297 * Create an external-format (``xsocket'') structure using the information in
4298 * the kernel-format socket structure pointed to by so. This is done to
4299 * reduce the spew of irrelevant information over this interface, to isolate
4300 * user code from changes in the kernel structure, and potentially to provide
4301 * information-hiding if we decide that some of this information should be
4302 * hidden from users.
4305 sotoxsocket(struct socket *so, struct xsocket *xso)
4308 bzero(xso, sizeof(*xso));
4309 xso->xso_len = sizeof *xso;
4310 xso->xso_so = (uintptr_t)so;
4311 xso->so_type = so->so_type;
4312 xso->so_options = so->so_options;
4313 xso->so_linger = so->so_linger;
4314 xso->so_state = so->so_state;
4315 xso->so_pcb = (uintptr_t)so->so_pcb;
4316 xso->xso_protocol = so->so_proto->pr_protocol;
4317 xso->xso_family = so->so_proto->pr_domain->dom_family;
4318 xso->so_timeo = so->so_timeo;
4319 xso->so_error = so->so_error;
4320 xso->so_uid = so->so_cred->cr_uid;
4321 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4322 if (SOLISTENING(so)) {
4323 xso->so_qlen = so->sol_qlen;
4324 xso->so_incqlen = so->sol_incqlen;
4325 xso->so_qlimit = so->sol_qlimit;
4326 xso->so_oobmark = 0;
4328 xso->so_state |= so->so_qstate;
4329 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4330 xso->so_oobmark = so->so_oobmark;
4331 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4332 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4337 so_sockbuf_rcv(struct socket *so)
4340 return (&so->so_rcv);
4344 so_sockbuf_snd(struct socket *so)
4347 return (&so->so_snd);
4351 so_state_get(const struct socket *so)
4354 return (so->so_state);
4358 so_state_set(struct socket *so, int val)
4365 so_options_get(const struct socket *so)
4368 return (so->so_options);
4372 so_options_set(struct socket *so, int val)
4375 so->so_options = val;
4379 so_error_get(const struct socket *so)
4382 return (so->so_error);
4386 so_error_set(struct socket *so, int val)
4393 so_linger_get(const struct socket *so)
4396 return (so->so_linger);
4400 so_linger_set(struct socket *so, int val)
4403 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4404 ("%s: val %d out of range", __func__, val));
4406 so->so_linger = val;
4410 so_protosw_get(const struct socket *so)
4413 return (so->so_proto);
4417 so_protosw_set(struct socket *so, struct protosw *val)
4424 so_sorwakeup(struct socket *so)
4431 so_sowwakeup(struct socket *so)
4438 so_sorwakeup_locked(struct socket *so)
4441 sorwakeup_locked(so);
4445 so_sowwakeup_locked(struct socket *so)
4448 sowwakeup_locked(so);
4452 so_lock(struct socket *so)
4459 so_unlock(struct socket *so)