2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993
5 * The Regents of the University of California.
6 * Copyright (c) 2004 The FreeBSD Foundation
7 * Copyright (c) 2004-2008 Robert N. M. Watson
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
38 * Comments on the socket life cycle:
40 * soalloc() sets of socket layer state for a socket, called only by
41 * socreate() and sonewconn(). Socket layer private.
43 * sodealloc() tears down socket layer state for a socket, called only by
44 * sofree() and sonewconn(). Socket layer private.
46 * pru_attach() associates protocol layer state with an allocated socket;
47 * called only once, may fail, aborting socket allocation. This is called
48 * from socreate() and sonewconn(). Socket layer private.
50 * pru_detach() disassociates protocol layer state from an attached socket,
51 * and will be called exactly once for sockets in which pru_attach() has
52 * been successfully called. If pru_attach() returned an error,
53 * pru_detach() will not be called. Socket layer private.
55 * pru_abort() and pru_close() notify the protocol layer that the last
56 * consumer of a socket is starting to tear down the socket, and that the
57 * protocol should terminate the connection. Historically, pru_abort() also
58 * detached protocol state from the socket state, but this is no longer the
61 * socreate() creates a socket and attaches protocol state. This is a public
62 * interface that may be used by socket layer consumers to create new
65 * sonewconn() creates a socket and attaches protocol state. This is a
66 * public interface that may be used by protocols to create new sockets when
67 * a new connection is received and will be available for accept() on a
70 * soclose() destroys a socket after possibly waiting for it to disconnect.
71 * This is a public interface that socket consumers should use to close and
72 * release a socket when done with it.
74 * soabort() destroys a socket without waiting for it to disconnect (used
75 * only for incoming connections that are already partially or fully
76 * connected). This is used internally by the socket layer when clearing
77 * listen socket queues (due to overflow or close on the listen socket), but
78 * is also a public interface protocols may use to abort connections in
79 * their incomplete listen queues should they no longer be required. Sockets
80 * placed in completed connection listen queues should not be aborted for
81 * reasons described in the comment above the soclose() implementation. This
82 * is not a general purpose close routine, and except in the specific
83 * circumstances described here, should not be used.
85 * sofree() will free a socket and its protocol state if all references on
86 * the socket have been released, and is the public interface to attempt to
87 * free a socket when a reference is removed. This is a socket layer private
90 * NOTE: In addition to socreate() and soclose(), which provide a single
91 * socket reference to the consumer to be managed as required, there are two
92 * calls to explicitly manage socket references, soref(), and sorele().
93 * Currently, these are generally required only when transitioning a socket
94 * from a listen queue to a file descriptor, in order to prevent garbage
95 * collection of the socket at an untimely moment. For a number of reasons,
96 * these interfaces are not preferred, and should be avoided.
98 * NOTE: With regard to VNETs the general rule is that callers do not set
99 * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
100 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
101 * and sorflush(), which are usually called from a pre-set VNET context.
102 * sopoll() currently does not need a VNET context to be set.
105 #include <sys/cdefs.h>
106 __FBSDID("$FreeBSD$");
108 #include "opt_inet.h"
109 #include "opt_inet6.h"
110 #include "opt_kern_tls.h"
111 #include "opt_sctp.h"
113 #include <sys/param.h>
114 #include <sys/systm.h>
115 #include <sys/fcntl.h>
116 #include <sys/limits.h>
117 #include <sys/lock.h>
119 #include <sys/malloc.h>
120 #include <sys/mbuf.h>
121 #include <sys/mutex.h>
122 #include <sys/domain.h>
123 #include <sys/file.h> /* for struct knote */
124 #include <sys/hhook.h>
125 #include <sys/kernel.h>
126 #include <sys/khelp.h>
127 #include <sys/ktls.h>
128 #include <sys/event.h>
129 #include <sys/eventhandler.h>
130 #include <sys/poll.h>
131 #include <sys/proc.h>
132 #include <sys/protosw.h>
133 #include <sys/sbuf.h>
134 #include <sys/socket.h>
135 #include <sys/socketvar.h>
136 #include <sys/resourcevar.h>
137 #include <net/route.h>
138 #include <sys/signalvar.h>
139 #include <sys/stat.h>
141 #include <sys/sysctl.h>
142 #include <sys/taskqueue.h>
145 #include <sys/unpcb.h>
146 #include <sys/jail.h>
147 #include <sys/syslog.h>
148 #include <netinet/in.h>
149 #include <netinet/in_pcb.h>
150 #include <netinet/tcp.h>
152 #include <net/vnet.h>
154 #include <security/mac/mac_framework.h>
158 #ifdef COMPAT_FREEBSD32
159 #include <sys/mount.h>
160 #include <sys/sysent.h>
161 #include <compat/freebsd32/freebsd32.h>
164 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
166 static void so_rdknl_lock(void *);
167 static void so_rdknl_unlock(void *);
168 static void so_rdknl_assert_lock(void *, int);
169 static void so_wrknl_lock(void *);
170 static void so_wrknl_unlock(void *);
171 static void so_wrknl_assert_lock(void *, int);
173 static void filt_sordetach(struct knote *kn);
174 static int filt_soread(struct knote *kn, long hint);
175 static void filt_sowdetach(struct knote *kn);
176 static int filt_sowrite(struct knote *kn, long hint);
177 static int filt_soempty(struct knote *kn, long hint);
178 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
179 fo_kqfilter_t soo_kqfilter;
181 static struct filterops soread_filtops = {
183 .f_detach = filt_sordetach,
184 .f_event = filt_soread,
186 static struct filterops sowrite_filtops = {
188 .f_detach = filt_sowdetach,
189 .f_event = filt_sowrite,
191 static struct filterops soempty_filtops = {
193 .f_detach = filt_sowdetach,
194 .f_event = filt_soempty,
197 so_gen_t so_gencnt; /* generation count for sockets */
199 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
200 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
202 #define VNET_SO_ASSERT(so) \
203 VNET_ASSERT(curvnet != NULL, \
204 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
206 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
207 #define V_socket_hhh VNET(socket_hhh)
210 * Limit on the number of connections in the listen queue waiting
212 * NB: The original sysctl somaxconn is still available but hidden
213 * to prevent confusion about the actual purpose of this number.
215 static u_int somaxconn = SOMAXCONN;
218 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
224 error = sysctl_handle_int(oidp, &val, 0, req);
225 if (error || !req->newptr )
229 * The purpose of the UINT_MAX / 3 limit, is so that the formula
231 * below, will not overflow.
234 if (val < 1 || val > UINT_MAX / 3)
240 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
241 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, sizeof(int),
242 sysctl_somaxconn, "I",
243 "Maximum listen socket pending connection accept queue size");
244 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
245 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, 0,
246 sizeof(int), sysctl_somaxconn, "I",
247 "Maximum listen socket pending connection accept queue size (compat)");
249 static int numopensockets;
250 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
251 &numopensockets, 0, "Number of open sockets");
254 * accept_mtx locks down per-socket fields relating to accept queues. See
255 * socketvar.h for an annotation of the protected fields of struct socket.
257 struct mtx accept_mtx;
258 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
261 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
264 static struct mtx so_global_mtx;
265 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
268 * General IPC sysctl name space, used by sockets and a variety of other IPC
271 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
275 * Initialize the socket subsystem and set up the socket
278 static uma_zone_t socket_zone;
282 socket_zone_change(void *tag)
285 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
289 socket_hhook_register(int subtype)
292 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
293 &V_socket_hhh[subtype],
294 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
295 printf("%s: WARNING: unable to register hook\n", __func__);
299 socket_hhook_deregister(int subtype)
302 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
303 printf("%s: WARNING: unable to deregister hook\n", __func__);
307 socket_init(void *tag)
310 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
311 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
312 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
313 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
314 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
315 EVENTHANDLER_PRI_FIRST);
317 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
320 socket_vnet_init(const void *unused __unused)
324 /* We expect a contiguous range */
325 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
326 socket_hhook_register(i);
328 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
329 socket_vnet_init, NULL);
332 socket_vnet_uninit(const void *unused __unused)
336 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
337 socket_hhook_deregister(i);
339 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
340 socket_vnet_uninit, NULL);
343 * Initialise maxsockets. This SYSINIT must be run after
347 init_maxsockets(void *ignored)
350 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
351 maxsockets = imax(maxsockets, maxfiles);
353 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
356 * Sysctl to get and set the maximum global sockets limit. Notify protocols
357 * of the change so that they can update their dependent limits as required.
360 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
362 int error, newmaxsockets;
364 newmaxsockets = maxsockets;
365 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
366 if (error == 0 && req->newptr) {
367 if (newmaxsockets > maxsockets &&
368 newmaxsockets <= maxfiles) {
369 maxsockets = newmaxsockets;
370 EVENTHANDLER_INVOKE(maxsockets_change);
376 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
377 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &maxsockets, 0,
378 sysctl_maxsockets, "IU",
379 "Maximum number of sockets available");
382 * Socket operation routines. These routines are called by the routines in
383 * sys_socket.c or from a system process, and implement the semantics of
384 * socket operations by switching out to the protocol specific routines.
388 * Get a socket structure from our zone, and initialize it. Note that it
389 * would probably be better to allocate socket and PCB at the same time, but
390 * I'm not convinced that all the protocols can be easily modified to do
393 * soalloc() returns a socket with a ref count of 0.
395 static struct socket *
396 soalloc(struct vnet *vnet)
400 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
404 if (mac_socket_init(so, M_NOWAIT) != 0) {
405 uma_zfree(socket_zone, so);
409 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
410 uma_zfree(socket_zone, so);
415 * The socket locking protocol allows to lock 2 sockets at a time,
416 * however, the first one must be a listening socket. WITNESS lacks
417 * a feature to change class of an existing lock, so we use DUPOK.
419 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
420 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
421 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
422 so->so_rcv.sb_sel = &so->so_rdsel;
423 so->so_snd.sb_sel = &so->so_wrsel;
424 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
425 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
426 TAILQ_INIT(&so->so_snd.sb_aiojobq);
427 TAILQ_INIT(&so->so_rcv.sb_aiojobq);
428 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
429 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
431 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
432 __func__, __LINE__, so));
435 /* We shouldn't need the so_global_mtx */
436 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
437 /* Do we need more comprehensive error returns? */
438 uma_zfree(socket_zone, so);
441 mtx_lock(&so_global_mtx);
442 so->so_gencnt = ++so_gencnt;
445 vnet->vnet_sockcnt++;
447 mtx_unlock(&so_global_mtx);
453 * Free the storage associated with a socket at the socket layer, tear down
454 * locks, labels, etc. All protocol state is assumed already to have been
455 * torn down (and possibly never set up) by the caller.
458 sodealloc(struct socket *so)
461 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
462 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
464 mtx_lock(&so_global_mtx);
465 so->so_gencnt = ++so_gencnt;
466 --numopensockets; /* Could be below, but faster here. */
468 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
469 __func__, __LINE__, so));
470 so->so_vnet->vnet_sockcnt--;
472 mtx_unlock(&so_global_mtx);
474 mac_socket_destroy(so);
476 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
479 khelp_destroy_osd(&so->osd);
480 if (SOLISTENING(so)) {
481 if (so->sol_accept_filter != NULL)
482 accept_filt_setopt(so, NULL);
484 if (so->so_rcv.sb_hiwat)
485 (void)chgsbsize(so->so_cred->cr_uidinfo,
486 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
487 if (so->so_snd.sb_hiwat)
488 (void)chgsbsize(so->so_cred->cr_uidinfo,
489 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
490 sx_destroy(&so->so_snd.sb_sx);
491 sx_destroy(&so->so_rcv.sb_sx);
492 SOCKBUF_LOCK_DESTROY(&so->so_snd);
493 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
495 mtx_destroy(&so->so_lock);
496 uma_zfree(socket_zone, so);
500 * socreate returns a socket with a ref count of 1. The socket should be
501 * closed with soclose().
504 socreate(int dom, struct socket **aso, int type, int proto,
505 struct ucred *cred, struct thread *td)
512 prp = pffindproto(dom, proto, type);
514 prp = pffindtype(dom, type);
517 /* No support for domain. */
518 if (pffinddomain(dom) == NULL)
519 return (EAFNOSUPPORT);
520 /* No support for socket type. */
521 if (proto == 0 && type != 0)
523 return (EPROTONOSUPPORT);
525 if (prp->pr_usrreqs->pru_attach == NULL ||
526 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
527 return (EPROTONOSUPPORT);
529 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
530 return (EPROTONOSUPPORT);
532 if (prp->pr_type != type)
534 so = soalloc(CRED_TO_VNET(cred));
539 so->so_cred = crhold(cred);
540 if ((prp->pr_domain->dom_family == PF_INET) ||
541 (prp->pr_domain->dom_family == PF_INET6) ||
542 (prp->pr_domain->dom_family == PF_ROUTE))
543 so->so_fibnum = td->td_proc->p_fibnum;
548 mac_socket_create(cred, so);
550 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
551 so_rdknl_assert_lock);
552 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
553 so_wrknl_assert_lock);
555 * Auto-sizing of socket buffers is managed by the protocols and
556 * the appropriate flags must be set in the pru_attach function.
558 CURVNET_SET(so->so_vnet);
559 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
571 static int regression_sonewconn_earlytest = 1;
572 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
573 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
576 static struct timeval overinterval = { 60, 0 };
577 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
579 "Delay in seconds between warnings for listen socket overflows");
582 * When an attempt at a new connection is noted on a socket which accepts
583 * connections, sonewconn is called. If the connection is possible (subject
584 * to space constraints, etc.) then we allocate a new structure, properly
585 * linked into the data structure of the original socket, and return this.
586 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
588 * Note: the ref count on the socket is 0 on return.
591 sonewconn(struct socket *head, int connstatus)
597 const char localprefix[] = "local:";
598 char descrbuf[SUNPATHLEN + sizeof(localprefix)];
600 char addrbuf[INET6_ADDRSTRLEN];
602 char addrbuf[INET_ADDRSTRLEN];
607 over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
609 if (regression_sonewconn_earlytest && over) {
613 head->sol_overcount++;
614 dolog = !!ratecheck(&head->sol_lastover, &overinterval);
617 * If we're going to log, copy the overflow count and queue
618 * length from the listen socket before dropping the lock.
619 * Also, reset the overflow count.
622 overcount = head->sol_overcount;
623 head->sol_overcount = 0;
624 qlen = head->sol_qlen;
626 SOLISTEN_UNLOCK(head);
630 * Try to print something descriptive about the
631 * socket for the error message.
633 sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
635 switch (head->so_proto->pr_domain->dom_family) {
636 #if defined(INET) || defined(INET6)
642 if (head->so_proto->pr_domain->dom_family ==
644 (sotoinpcb(head)->inp_inc.inc_flags &
647 &sotoinpcb(head)->inp_inc.inc6_laddr);
648 sbuf_printf(&descrsb, "[%s]", addrbuf);
654 sotoinpcb(head)->inp_inc.inc_laddr,
656 sbuf_cat(&descrsb, addrbuf);
659 sbuf_printf(&descrsb, ":%hu (proto %u)",
660 ntohs(sotoinpcb(head)->inp_inc.inc_lport),
661 head->so_proto->pr_protocol);
663 #endif /* INET || INET6 */
665 sbuf_cat(&descrsb, localprefix);
666 if (sotounpcb(head)->unp_addr != NULL)
668 sotounpcb(head)->unp_addr->sun_len -
669 offsetof(struct sockaddr_un,
675 sotounpcb(head)->unp_addr->sun_path,
678 sbuf_cat(&descrsb, "(unknown)");
683 * If we can't print something more specific, at least
684 * print the domain name.
686 if (sbuf_finish(&descrsb) != 0 ||
687 sbuf_len(&descrsb) <= 0) {
688 sbuf_clear(&descrsb);
690 head->so_proto->pr_domain->dom_name ?:
692 sbuf_finish(&descrsb);
694 KASSERT(sbuf_len(&descrsb) > 0,
695 ("%s: sbuf creation failed", __func__));
697 "%s: pcb %p (%s): Listen queue overflow: "
698 "%i already in queue awaiting acceptance "
699 "(%d occurrences)\n",
700 __func__, head->so_pcb, sbuf_data(&descrsb),
702 sbuf_delete(&descrsb);
709 SOLISTEN_UNLOCK(head);
710 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
712 so = soalloc(head->so_vnet);
714 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
715 "limit reached or out of memory\n",
716 __func__, head->so_pcb);
719 so->so_listen = head;
720 so->so_type = head->so_type;
721 so->so_linger = head->so_linger;
722 so->so_state = head->so_state | SS_NOFDREF;
723 so->so_fibnum = head->so_fibnum;
724 so->so_proto = head->so_proto;
725 so->so_cred = crhold(head->so_cred);
727 mac_socket_newconn(head, so);
729 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
730 so_rdknl_assert_lock);
731 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
732 so_wrknl_assert_lock);
733 VNET_SO_ASSERT(head);
734 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
736 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
737 __func__, head->so_pcb);
740 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
742 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
743 __func__, head->so_pcb);
746 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
747 so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
748 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
749 so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
750 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
751 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
754 if (head->sol_accept_filter != NULL)
756 so->so_state |= connstatus;
757 so->so_options = head->so_options & ~SO_ACCEPTCONN;
758 soref(head); /* A socket on (in)complete queue refs head. */
760 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
761 so->so_qstate = SQ_COMP;
763 solisten_wakeup(head); /* unlocks */
766 * Keep removing sockets from the head until there's room for
767 * us to insert on the tail. In pre-locking revisions, this
768 * was a simple if(), but as we could be racing with other
769 * threads and soabort() requires dropping locks, we must
770 * loop waiting for the condition to be true.
772 while (head->sol_incqlen > head->sol_qlimit) {
775 sp = TAILQ_FIRST(&head->sol_incomp);
776 TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
779 sp->so_qstate = SQ_NONE;
780 sp->so_listen = NULL;
782 sorele(head); /* does SOLISTEN_UNLOCK, head stays */
786 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
787 so->so_qstate = SQ_INCOMP;
789 SOLISTEN_UNLOCK(head);
794 #if defined(SCTP) || defined(SCTP_SUPPORT)
796 * Socket part of sctp_peeloff(). Detach a new socket from an
797 * association. The new socket is returned with a reference.
800 sopeeloff(struct socket *head)
804 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
805 __func__, __LINE__, head));
806 so = soalloc(head->so_vnet);
808 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
809 "limit reached or out of memory\n",
810 __func__, head->so_pcb);
813 so->so_type = head->so_type;
814 so->so_options = head->so_options;
815 so->so_linger = head->so_linger;
816 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
817 so->so_fibnum = head->so_fibnum;
818 so->so_proto = head->so_proto;
819 so->so_cred = crhold(head->so_cred);
821 mac_socket_newconn(head, so);
823 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
824 so_rdknl_assert_lock);
825 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
826 so_wrknl_assert_lock);
827 VNET_SO_ASSERT(head);
828 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
830 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
831 __func__, head->so_pcb);
834 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
836 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
837 __func__, head->so_pcb);
840 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
841 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
842 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
843 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
844 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
845 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
854 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
858 CURVNET_SET(so->so_vnet);
859 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
865 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
869 CURVNET_SET(so->so_vnet);
870 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
876 * solisten() transitions a socket from a non-listening state to a listening
877 * state, but can also be used to update the listen queue depth on an
878 * existing listen socket. The protocol will call back into the sockets
879 * layer using solisten_proto_check() and solisten_proto() to check and set
880 * socket-layer listen state. Call backs are used so that the protocol can
881 * acquire both protocol and socket layer locks in whatever order is required
884 * Protocol implementors are advised to hold the socket lock across the
885 * socket-layer test and set to avoid races at the socket layer.
888 solisten(struct socket *so, int backlog, struct thread *td)
892 CURVNET_SET(so->so_vnet);
893 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
899 solisten_proto_check(struct socket *so)
902 SOCK_LOCK_ASSERT(so);
904 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
911 solisten_proto(struct socket *so, int backlog)
913 int sbrcv_lowat, sbsnd_lowat;
914 u_int sbrcv_hiwat, sbsnd_hiwat;
915 short sbrcv_flags, sbsnd_flags;
916 sbintime_t sbrcv_timeo, sbsnd_timeo;
918 SOCK_LOCK_ASSERT(so);
924 * Change this socket to listening state.
926 sbrcv_lowat = so->so_rcv.sb_lowat;
927 sbsnd_lowat = so->so_snd.sb_lowat;
928 sbrcv_hiwat = so->so_rcv.sb_hiwat;
929 sbsnd_hiwat = so->so_snd.sb_hiwat;
930 sbrcv_flags = so->so_rcv.sb_flags;
931 sbsnd_flags = so->so_snd.sb_flags;
932 sbrcv_timeo = so->so_rcv.sb_timeo;
933 sbsnd_timeo = so->so_snd.sb_timeo;
935 sbdestroy(&so->so_snd, so);
936 sbdestroy(&so->so_rcv, so);
937 sx_destroy(&so->so_snd.sb_sx);
938 sx_destroy(&so->so_rcv.sb_sx);
939 SOCKBUF_LOCK_DESTROY(&so->so_snd);
940 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
944 sizeof(struct socket) - offsetof(struct socket, so_rcv));
947 so->sol_sbrcv_lowat = sbrcv_lowat;
948 so->sol_sbsnd_lowat = sbsnd_lowat;
949 so->sol_sbrcv_hiwat = sbrcv_hiwat;
950 so->sol_sbsnd_hiwat = sbsnd_hiwat;
951 so->sol_sbrcv_flags = sbrcv_flags;
952 so->sol_sbsnd_flags = sbsnd_flags;
953 so->sol_sbrcv_timeo = sbrcv_timeo;
954 so->sol_sbsnd_timeo = sbsnd_timeo;
956 so->sol_qlen = so->sol_incqlen = 0;
957 TAILQ_INIT(&so->sol_incomp);
958 TAILQ_INIT(&so->sol_comp);
960 so->sol_accept_filter = NULL;
961 so->sol_accept_filter_arg = NULL;
962 so->sol_accept_filter_str = NULL;
964 so->sol_upcall = NULL;
965 so->sol_upcallarg = NULL;
967 so->so_options |= SO_ACCEPTCONN;
970 if (backlog < 0 || backlog > somaxconn)
972 so->sol_qlimit = backlog;
976 * Wakeup listeners/subsystems once we have a complete connection.
977 * Enters with lock, returns unlocked.
980 solisten_wakeup(struct socket *sol)
983 if (sol->sol_upcall != NULL)
984 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
986 selwakeuppri(&sol->so_rdsel, PSOCK);
987 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
989 SOLISTEN_UNLOCK(sol);
990 wakeup_one(&sol->sol_comp);
991 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
992 pgsigio(&sol->so_sigio, SIGIO, 0);
996 * Return single connection off a listening socket queue. Main consumer of
997 * the function is kern_accept4(). Some modules, that do their own accept
998 * management also use the function.
1000 * Listening socket must be locked on entry and is returned unlocked on
1002 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1005 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1010 SOLISTEN_LOCK_ASSERT(head);
1012 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1013 head->so_error == 0) {
1014 error = msleep(&head->sol_comp, &head->so_lock, PSOCK | PCATCH,
1017 SOLISTEN_UNLOCK(head);
1021 if (head->so_error) {
1022 error = head->so_error;
1024 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1025 error = EWOULDBLOCK;
1029 SOLISTEN_UNLOCK(head);
1032 so = TAILQ_FIRST(&head->sol_comp);
1034 KASSERT(so->so_qstate == SQ_COMP,
1035 ("%s: so %p not SQ_COMP", __func__, so));
1038 so->so_qstate = SQ_NONE;
1039 so->so_listen = NULL;
1040 TAILQ_REMOVE(&head->sol_comp, so, so_list);
1041 if (flags & ACCEPT4_INHERIT)
1042 so->so_state |= (head->so_state & SS_NBIO);
1044 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1053 * Evaluate the reference count and named references on a socket; if no
1054 * references remain, free it. This should be called whenever a reference is
1055 * released, such as in sorele(), but also when named reference flags are
1056 * cleared in socket or protocol code.
1058 * sofree() will free the socket if:
1060 * - There are no outstanding file descriptor references or related consumers
1063 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
1065 * - The protocol does not have an outstanding strong reference on the socket
1068 * - The socket is not in a completed connection queue, so a process has been
1069 * notified that it is present. If it is removed, the user process may
1070 * block in accept() despite select() saying the socket was ready.
1073 sofree(struct socket *so)
1075 struct protosw *pr = so->so_proto;
1077 SOCK_LOCK_ASSERT(so);
1079 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
1080 (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) {
1085 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
1088 sol = so->so_listen;
1089 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
1092 * To solve race between close of a listening socket and
1093 * a socket on its incomplete queue, we need to lock both.
1094 * The order is first listening socket, then regular.
1095 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
1096 * function and the listening socket are the only pointers
1097 * to so. To preserve so and sol, we reference both and then
1099 * After relock the socket may not move to so_comp since it
1100 * doesn't have PCB already, but it may be removed from
1101 * so_incomp. If that happens, we share responsiblity on
1102 * freeing the socket, but soclose() has already removed
1110 if (so->so_qstate == SQ_INCOMP) {
1111 KASSERT(so->so_listen == sol,
1112 ("%s: so %p migrated out of sol %p",
1113 __func__, so, sol));
1114 TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1116 /* This is guarenteed not to be the last. */
1117 refcount_release(&sol->so_count);
1118 so->so_qstate = SQ_NONE;
1119 so->so_listen = NULL;
1121 KASSERT(so->so_listen == NULL,
1122 ("%s: so %p not on (in)comp with so_listen",
1125 KASSERT(so->so_count == 1,
1126 ("%s: so %p count %u", __func__, so, so->so_count));
1129 if (SOLISTENING(so))
1130 so->so_error = ECONNABORTED;
1133 if (so->so_dtor != NULL)
1137 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1138 (*pr->pr_domain->dom_dispose)(so);
1139 if (pr->pr_usrreqs->pru_detach != NULL)
1140 (*pr->pr_usrreqs->pru_detach)(so);
1143 * From this point on, we assume that no other references to this
1144 * socket exist anywhere else in the stack. Therefore, no locks need
1145 * to be acquired or held.
1147 * We used to do a lot of socket buffer and socket locking here, as
1148 * well as invoke sorflush() and perform wakeups. The direct call to
1149 * dom_dispose() and sbdestroy() are an inlining of what was
1150 * necessary from sorflush().
1152 * Notice that the socket buffer and kqueue state are torn down
1153 * before calling pru_detach. This means that protocols shold not
1154 * assume they can perform socket wakeups, etc, in their detach code.
1156 if (!SOLISTENING(so)) {
1157 sbdestroy(&so->so_snd, so);
1158 sbdestroy(&so->so_rcv, so);
1160 seldrain(&so->so_rdsel);
1161 seldrain(&so->so_wrsel);
1162 knlist_destroy(&so->so_rdsel.si_note);
1163 knlist_destroy(&so->so_wrsel.si_note);
1168 * Close a socket on last file table reference removal. Initiate disconnect
1169 * if connected. Free socket when disconnect complete.
1171 * This function will sorele() the socket. Note that soclose() may be called
1172 * prior to the ref count reaching zero. The actual socket structure will
1173 * not be freed until the ref count reaches zero.
1176 soclose(struct socket *so)
1178 struct accept_queue lqueue;
1182 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1184 CURVNET_SET(so->so_vnet);
1185 funsetown(&so->so_sigio);
1186 if (so->so_state & SS_ISCONNECTED) {
1187 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1188 error = sodisconnect(so);
1190 if (error == ENOTCONN)
1195 if (so->so_options & SO_LINGER) {
1196 if ((so->so_state & SS_ISDISCONNECTING) &&
1197 (so->so_state & SS_NBIO))
1199 while (so->so_state & SS_ISCONNECTED) {
1200 error = tsleep(&so->so_timeo,
1201 PSOCK | PCATCH, "soclos",
1202 so->so_linger * hz);
1210 if (so->so_proto->pr_usrreqs->pru_close != NULL)
1211 (*so->so_proto->pr_usrreqs->pru_close)(so);
1214 if ((listening = (so->so_options & SO_ACCEPTCONN))) {
1217 TAILQ_INIT(&lqueue);
1218 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1219 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1221 so->sol_qlen = so->sol_incqlen = 0;
1223 TAILQ_FOREACH(sp, &lqueue, so_list) {
1225 sp->so_qstate = SQ_NONE;
1226 sp->so_listen = NULL;
1228 /* Guaranteed not to be the last. */
1229 refcount_release(&so->so_count);
1232 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1233 so->so_state |= SS_NOFDREF;
1236 struct socket *sp, *tsp;
1238 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1240 if (sp->so_count == 0) {
1244 /* sp is now in sofree() */
1253 * soabort() is used to abruptly tear down a connection, such as when a
1254 * resource limit is reached (listen queue depth exceeded), or if a listen
1255 * socket is closed while there are sockets waiting to be accepted.
1257 * This interface is tricky, because it is called on an unreferenced socket,
1258 * and must be called only by a thread that has actually removed the socket
1259 * from the listen queue it was on, or races with other threads are risked.
1261 * This interface will call into the protocol code, so must not be called
1262 * with any socket locks held. Protocols do call it while holding their own
1263 * recursible protocol mutexes, but this is something that should be subject
1264 * to review in the future.
1267 soabort(struct socket *so)
1271 * In as much as is possible, assert that no references to this
1272 * socket are held. This is not quite the same as asserting that the
1273 * current thread is responsible for arranging for no references, but
1274 * is as close as we can get for now.
1276 KASSERT(so->so_count == 0, ("soabort: so_count"));
1277 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1278 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1281 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1282 (*so->so_proto->pr_usrreqs->pru_abort)(so);
1288 soaccept(struct socket *so, struct sockaddr **nam)
1293 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1294 so->so_state &= ~SS_NOFDREF;
1297 CURVNET_SET(so->so_vnet);
1298 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1304 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1307 return (soconnectat(AT_FDCWD, so, nam, td));
1311 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1315 if (so->so_options & SO_ACCEPTCONN)
1316 return (EOPNOTSUPP);
1318 CURVNET_SET(so->so_vnet);
1320 * If protocol is connection-based, can only connect once.
1321 * Otherwise, if connected, try to disconnect first. This allows
1322 * user to disconnect by connecting to, e.g., a null address.
1324 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1325 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1326 (error = sodisconnect(so)))) {
1330 * Prevent accumulated error from previous connection from
1334 if (fd == AT_FDCWD) {
1335 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1338 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1348 soconnect2(struct socket *so1, struct socket *so2)
1352 CURVNET_SET(so1->so_vnet);
1353 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1359 sodisconnect(struct socket *so)
1363 if ((so->so_state & SS_ISCONNECTED) == 0)
1365 if (so->so_state & SS_ISDISCONNECTING)
1368 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1372 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1375 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1376 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1380 int clen = 0, error, dontroute;
1382 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1383 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1384 ("sosend_dgram: !PR_ATOMIC"));
1387 resid = uio->uio_resid;
1389 resid = top->m_pkthdr.len;
1391 * In theory resid should be unsigned. However, space must be
1392 * signed, as it might be less than 0 if we over-committed, and we
1393 * must use a signed comparison of space and resid. On the other
1394 * hand, a negative resid causes us to loop sending 0-length
1395 * segments to the protocol.
1403 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1405 td->td_ru.ru_msgsnd++;
1406 if (control != NULL)
1407 clen = control->m_len;
1409 SOCKBUF_LOCK(&so->so_snd);
1410 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1411 SOCKBUF_UNLOCK(&so->so_snd);
1416 error = so->so_error;
1418 SOCKBUF_UNLOCK(&so->so_snd);
1421 if ((so->so_state & SS_ISCONNECTED) == 0) {
1423 * `sendto' and `sendmsg' is allowed on a connection-based
1424 * socket if it supports implied connect. Return ENOTCONN if
1425 * not connected and no address is supplied.
1427 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1428 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1429 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1430 !(resid == 0 && clen != 0)) {
1431 SOCKBUF_UNLOCK(&so->so_snd);
1435 } else if (addr == NULL) {
1436 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1439 error = EDESTADDRREQ;
1440 SOCKBUF_UNLOCK(&so->so_snd);
1446 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1447 * problem and need fixing.
1449 space = sbspace(&so->so_snd);
1450 if (flags & MSG_OOB)
1453 SOCKBUF_UNLOCK(&so->so_snd);
1454 if (resid > space) {
1460 if (flags & MSG_EOR)
1461 top->m_flags |= M_EOR;
1464 * Copy the data from userland into a mbuf chain.
1465 * If no data is to be copied in, a single empty mbuf
1468 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1469 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1471 error = EFAULT; /* only possible error */
1474 space -= resid - uio->uio_resid;
1475 resid = uio->uio_resid;
1477 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1479 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1484 so->so_options |= SO_DONTROUTE;
1488 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1489 * of date. We could have received a reset packet in an interrupt or
1490 * maybe we slept while doing page faults in uiomove() etc. We could
1491 * probably recheck again inside the locking protection here, but
1492 * there are probably other places that this also happens. We must
1496 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1497 (flags & MSG_OOB) ? PRUS_OOB :
1499 * If the user set MSG_EOF, the protocol understands this flag and
1500 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1502 ((flags & MSG_EOF) &&
1503 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1506 /* If there is more to send set PRUS_MORETOCOME */
1507 (flags & MSG_MORETOCOME) ||
1508 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1509 top, addr, control, td);
1512 so->so_options &= ~SO_DONTROUTE;
1521 if (control != NULL)
1527 * Send on a socket. If send must go all at once and message is larger than
1528 * send buffering, then hard error. Lock against other senders. If must go
1529 * all at once and not enough room now, then inform user that this would
1530 * block and do nothing. Otherwise, if nonblocking, send as much as
1531 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1532 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1533 * in mbuf chain must be small enough to send all at once.
1535 * Returns nonzero on error, timeout or signal; callers must check for short
1536 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1540 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1541 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1545 int clen = 0, error, dontroute;
1546 int atomic = sosendallatonce(so) || top;
1549 struct ktls_session *tls;
1550 int tls_enq_cnt, tls_pruflag;
1554 tls_rtype = TLS_RLTYPE_APP;
1557 resid = uio->uio_resid;
1558 else if ((top->m_flags & M_PKTHDR) != 0)
1559 resid = top->m_pkthdr.len;
1561 resid = m_length(top, NULL);
1563 * In theory resid should be unsigned. However, space must be
1564 * signed, as it might be less than 0 if we over-committed, and we
1565 * must use a signed comparison of space and resid. On the other
1566 * hand, a negative resid causes us to loop sending 0-length
1567 * segments to the protocol.
1569 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1570 * type sockets since that's an error.
1572 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1578 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1579 (so->so_proto->pr_flags & PR_ATOMIC);
1581 td->td_ru.ru_msgsnd++;
1582 if (control != NULL)
1583 clen = control->m_len;
1585 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1591 tls = ktls_hold(so->so_snd.sb_tls_info);
1593 if (tls->mode == TCP_TLS_MODE_SW)
1594 tls_pruflag = PRUS_NOTREADY;
1596 if (control != NULL) {
1597 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1599 if (clen >= sizeof(*cm) &&
1600 cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1601 tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1613 SOCKBUF_LOCK(&so->so_snd);
1614 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1615 SOCKBUF_UNLOCK(&so->so_snd);
1620 error = so->so_error;
1622 SOCKBUF_UNLOCK(&so->so_snd);
1625 if ((so->so_state & SS_ISCONNECTED) == 0) {
1627 * `sendto' and `sendmsg' is allowed on a connection-
1628 * based socket if it supports implied connect.
1629 * Return ENOTCONN if not connected and no address is
1632 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1633 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1634 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1635 !(resid == 0 && clen != 0)) {
1636 SOCKBUF_UNLOCK(&so->so_snd);
1640 } else if (addr == NULL) {
1641 SOCKBUF_UNLOCK(&so->so_snd);
1642 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1645 error = EDESTADDRREQ;
1649 space = sbspace(&so->so_snd);
1650 if (flags & MSG_OOB)
1652 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1653 clen > so->so_snd.sb_hiwat) {
1654 SOCKBUF_UNLOCK(&so->so_snd);
1658 if (space < resid + clen &&
1659 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1660 if ((so->so_state & SS_NBIO) ||
1661 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1662 SOCKBUF_UNLOCK(&so->so_snd);
1663 error = EWOULDBLOCK;
1666 error = sbwait(&so->so_snd);
1667 SOCKBUF_UNLOCK(&so->so_snd);
1672 SOCKBUF_UNLOCK(&so->so_snd);
1677 if (flags & MSG_EOR)
1678 top->m_flags |= M_EOR;
1681 ktls_frame(top, tls, &tls_enq_cnt,
1683 tls_rtype = TLS_RLTYPE_APP;
1688 * Copy the data from userland into a mbuf
1689 * chain. If resid is 0, which can happen
1690 * only if we have control to send, then
1691 * a single empty mbuf is returned. This
1692 * is a workaround to prevent protocol send
1697 top = m_uiotombuf(uio, M_WAITOK, space,
1698 tls->params.max_frame_len,
1700 ((flags & MSG_EOR) ? M_EOR : 0));
1702 ktls_frame(top, tls,
1703 &tls_enq_cnt, tls_rtype);
1705 tls_rtype = TLS_RLTYPE_APP;
1708 top = m_uiotombuf(uio, M_WAITOK, space,
1709 (atomic ? max_hdr : 0),
1710 (atomic ? M_PKTHDR : 0) |
1711 ((flags & MSG_EOR) ? M_EOR : 0));
1713 error = EFAULT; /* only possible error */
1716 space -= resid - uio->uio_resid;
1717 resid = uio->uio_resid;
1721 so->so_options |= SO_DONTROUTE;
1725 * XXX all the SBS_CANTSENDMORE checks previously
1726 * done could be out of date. We could have received
1727 * a reset packet in an interrupt or maybe we slept
1728 * while doing page faults in uiomove() etc. We
1729 * could probably recheck again inside the locking
1730 * protection here, but there are probably other
1731 * places that this also happens. We must rethink
1736 pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1738 * If the user set MSG_EOF, the protocol understands
1739 * this flag and nothing left to send then use
1740 * PRU_SEND_EOF instead of PRU_SEND.
1742 ((flags & MSG_EOF) &&
1743 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1746 /* If there is more to send set PRUS_MORETOCOME. */
1747 (flags & MSG_MORETOCOME) ||
1748 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1751 pru_flag |= tls_pruflag;
1754 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1755 pru_flag, top, addr, control, td);
1759 so->so_options &= ~SO_DONTROUTE;
1764 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1766 * Note that error is intentionally
1769 * Like sendfile(), we rely on the
1770 * completion routine (pru_ready())
1771 * to free the mbufs in the event that
1772 * pru_send() encountered an error and
1773 * did not append them to the sockbuf.
1776 ktls_enqueue(top, so, tls_enq_cnt);
1784 } while (resid && space > 0);
1788 sbunlock(&so->so_snd);
1796 if (control != NULL)
1802 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1803 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1807 CURVNET_SET(so->so_vnet);
1808 if (!SOLISTENING(so))
1809 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1810 top, control, flags, td);
1821 * The part of soreceive() that implements reading non-inline out-of-band
1822 * data from a socket. For more complete comments, see soreceive(), from
1823 * which this code originated.
1825 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1826 * unable to return an mbuf chain to the caller.
1829 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1831 struct protosw *pr = so->so_proto;
1835 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1838 m = m_get(M_WAITOK, MT_DATA);
1839 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1843 error = uiomove(mtod(m, void *),
1844 (int) min(uio->uio_resid, m->m_len), uio);
1846 } while (uio->uio_resid && error == 0 && m);
1854 * Following replacement or removal of the first mbuf on the first mbuf chain
1855 * of a socket buffer, push necessary state changes back into the socket
1856 * buffer so that other consumers see the values consistently. 'nextrecord'
1857 * is the callers locally stored value of the original value of
1858 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1859 * NOTE: 'nextrecord' may be NULL.
1861 static __inline void
1862 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1865 SOCKBUF_LOCK_ASSERT(sb);
1867 * First, update for the new value of nextrecord. If necessary, make
1868 * it the first record.
1870 if (sb->sb_mb != NULL)
1871 sb->sb_mb->m_nextpkt = nextrecord;
1873 sb->sb_mb = nextrecord;
1876 * Now update any dependent socket buffer fields to reflect the new
1877 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1878 * addition of a second clause that takes care of the case where
1879 * sb_mb has been updated, but remains the last record.
1881 if (sb->sb_mb == NULL) {
1882 sb->sb_mbtail = NULL;
1883 sb->sb_lastrecord = NULL;
1884 } else if (sb->sb_mb->m_nextpkt == NULL)
1885 sb->sb_lastrecord = sb->sb_mb;
1889 * Implement receive operations on a socket. We depend on the way that
1890 * records are added to the sockbuf by sbappend. In particular, each record
1891 * (mbufs linked through m_next) must begin with an address if the protocol
1892 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1893 * data, and then zero or more mbufs of data. In order to allow parallelism
1894 * between network receive and copying to user space, as well as avoid
1895 * sleeping with a mutex held, we release the socket buffer mutex during the
1896 * user space copy. Although the sockbuf is locked, new data may still be
1897 * appended, and thus we must maintain consistency of the sockbuf during that
1900 * The caller may receive the data as a single mbuf chain by supplying an
1901 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1902 * the count in uio_resid.
1905 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1906 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1908 struct mbuf *m, **mp;
1909 int flags, error, offset;
1911 struct protosw *pr = so->so_proto;
1912 struct mbuf *nextrecord;
1914 ssize_t orig_resid = uio->uio_resid;
1919 if (controlp != NULL)
1922 flags = *flagsp &~ MSG_EOR;
1925 if (flags & MSG_OOB)
1926 return (soreceive_rcvoob(so, uio, flags));
1929 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1930 && uio->uio_resid) {
1932 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1935 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1940 SOCKBUF_LOCK(&so->so_rcv);
1941 m = so->so_rcv.sb_mb;
1943 * If we have less data than requested, block awaiting more (subject
1944 * to any timeout) if:
1945 * 1. the current count is less than the low water mark, or
1946 * 2. MSG_DONTWAIT is not set
1948 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1949 sbavail(&so->so_rcv) < uio->uio_resid) &&
1950 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1951 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1952 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1953 ("receive: m == %p sbavail == %u",
1954 m, sbavail(&so->so_rcv)));
1958 error = so->so_error;
1959 if ((flags & MSG_PEEK) == 0)
1961 SOCKBUF_UNLOCK(&so->so_rcv);
1964 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1965 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1969 else if (so->so_rcv.sb_tlsdcc == 0 &&
1970 so->so_rcv.sb_tlscc == 0) {
1974 SOCKBUF_UNLOCK(&so->so_rcv);
1978 for (; m != NULL; m = m->m_next)
1979 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1980 m = so->so_rcv.sb_mb;
1983 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
1984 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
1985 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1986 SOCKBUF_UNLOCK(&so->so_rcv);
1990 if (uio->uio_resid == 0) {
1991 SOCKBUF_UNLOCK(&so->so_rcv);
1994 if ((so->so_state & SS_NBIO) ||
1995 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1996 SOCKBUF_UNLOCK(&so->so_rcv);
1997 error = EWOULDBLOCK;
2000 SBLASTRECORDCHK(&so->so_rcv);
2001 SBLASTMBUFCHK(&so->so_rcv);
2002 error = sbwait(&so->so_rcv);
2003 SOCKBUF_UNLOCK(&so->so_rcv);
2010 * From this point onward, we maintain 'nextrecord' as a cache of the
2011 * pointer to the next record in the socket buffer. We must keep the
2012 * various socket buffer pointers and local stack versions of the
2013 * pointers in sync, pushing out modifications before dropping the
2014 * socket buffer mutex, and re-reading them when picking it up.
2016 * Otherwise, we will race with the network stack appending new data
2017 * or records onto the socket buffer by using inconsistent/stale
2018 * versions of the field, possibly resulting in socket buffer
2021 * By holding the high-level sblock(), we prevent simultaneous
2022 * readers from pulling off the front of the socket buffer.
2024 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2026 uio->uio_td->td_ru.ru_msgrcv++;
2027 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2028 SBLASTRECORDCHK(&so->so_rcv);
2029 SBLASTMBUFCHK(&so->so_rcv);
2030 nextrecord = m->m_nextpkt;
2031 if (pr->pr_flags & PR_ADDR) {
2032 KASSERT(m->m_type == MT_SONAME,
2033 ("m->m_type == %d", m->m_type));
2036 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2038 if (flags & MSG_PEEK) {
2041 sbfree(&so->so_rcv, m);
2042 so->so_rcv.sb_mb = m_free(m);
2043 m = so->so_rcv.sb_mb;
2044 sockbuf_pushsync(&so->so_rcv, nextrecord);
2049 * Process one or more MT_CONTROL mbufs present before any data mbufs
2050 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
2051 * just copy the data; if !MSG_PEEK, we call into the protocol to
2052 * perform externalization (or freeing if controlp == NULL).
2054 if (m != NULL && m->m_type == MT_CONTROL) {
2055 struct mbuf *cm = NULL, *cmn;
2056 struct mbuf **cme = &cm;
2058 struct cmsghdr *cmsg;
2059 struct tls_get_record tgr;
2062 * For MSG_TLSAPPDATA, check for a non-application data
2063 * record. If found, return ENXIO without removing
2064 * it from the receive queue. This allows a subsequent
2065 * call without MSG_TLSAPPDATA to receive it.
2066 * Note that, for TLS, there should only be a single
2067 * control mbuf with the TLS_GET_RECORD message in it.
2069 if (flags & MSG_TLSAPPDATA) {
2070 cmsg = mtod(m, struct cmsghdr *);
2071 if (cmsg->cmsg_type == TLS_GET_RECORD &&
2072 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2073 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2074 /* This will need to change for TLS 1.3. */
2075 if (tgr.tls_type != TLS_RLTYPE_APP) {
2076 SOCKBUF_UNLOCK(&so->so_rcv);
2085 if (flags & MSG_PEEK) {
2086 if (controlp != NULL) {
2087 *controlp = m_copym(m, 0, m->m_len,
2089 controlp = &(*controlp)->m_next;
2093 sbfree(&so->so_rcv, m);
2094 so->so_rcv.sb_mb = m->m_next;
2097 cme = &(*cme)->m_next;
2098 m = so->so_rcv.sb_mb;
2100 } while (m != NULL && m->m_type == MT_CONTROL);
2101 if ((flags & MSG_PEEK) == 0)
2102 sockbuf_pushsync(&so->so_rcv, nextrecord);
2103 while (cm != NULL) {
2106 if (pr->pr_domain->dom_externalize != NULL) {
2107 SOCKBUF_UNLOCK(&so->so_rcv);
2109 error = (*pr->pr_domain->dom_externalize)
2110 (cm, controlp, flags);
2111 SOCKBUF_LOCK(&so->so_rcv);
2112 } else if (controlp != NULL)
2116 if (controlp != NULL) {
2118 while (*controlp != NULL)
2119 controlp = &(*controlp)->m_next;
2124 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2126 nextrecord = so->so_rcv.sb_mb;
2130 if ((flags & MSG_PEEK) == 0) {
2131 KASSERT(m->m_nextpkt == nextrecord,
2132 ("soreceive: post-control, nextrecord !sync"));
2133 if (nextrecord == NULL) {
2134 KASSERT(so->so_rcv.sb_mb == m,
2135 ("soreceive: post-control, sb_mb!=m"));
2136 KASSERT(so->so_rcv.sb_lastrecord == m,
2137 ("soreceive: post-control, lastrecord!=m"));
2141 if (type == MT_OOBDATA)
2144 if ((flags & MSG_PEEK) == 0) {
2145 KASSERT(so->so_rcv.sb_mb == nextrecord,
2146 ("soreceive: sb_mb != nextrecord"));
2147 if (so->so_rcv.sb_mb == NULL) {
2148 KASSERT(so->so_rcv.sb_lastrecord == NULL,
2149 ("soreceive: sb_lastercord != NULL"));
2153 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2154 SBLASTRECORDCHK(&so->so_rcv);
2155 SBLASTMBUFCHK(&so->so_rcv);
2158 * Now continue to read any data mbufs off of the head of the socket
2159 * buffer until the read request is satisfied. Note that 'type' is
2160 * used to store the type of any mbuf reads that have happened so far
2161 * such that soreceive() can stop reading if the type changes, which
2162 * causes soreceive() to return only one of regular data and inline
2163 * out-of-band data in a single socket receive operation.
2167 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2170 * If the type of mbuf has changed since the last mbuf
2171 * examined ('type'), end the receive operation.
2173 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2174 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2175 if (type != m->m_type)
2177 } else if (type == MT_OOBDATA)
2180 KASSERT(m->m_type == MT_DATA,
2181 ("m->m_type == %d", m->m_type));
2182 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2183 len = uio->uio_resid;
2184 if (so->so_oobmark && len > so->so_oobmark - offset)
2185 len = so->so_oobmark - offset;
2186 if (len > m->m_len - moff)
2187 len = m->m_len - moff;
2189 * If mp is set, just pass back the mbufs. Otherwise copy
2190 * them out via the uio, then free. Sockbuf must be
2191 * consistent here (points to current mbuf, it points to next
2192 * record) when we drop priority; we must note any additions
2193 * to the sockbuf when we block interrupts again.
2196 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2197 SBLASTRECORDCHK(&so->so_rcv);
2198 SBLASTMBUFCHK(&so->so_rcv);
2199 SOCKBUF_UNLOCK(&so->so_rcv);
2200 if ((m->m_flags & M_EXTPG) != 0)
2201 error = m_unmappedtouio(m, moff, uio, (int)len);
2203 error = uiomove(mtod(m, char *) + moff,
2205 SOCKBUF_LOCK(&so->so_rcv);
2208 * The MT_SONAME mbuf has already been removed
2209 * from the record, so it is necessary to
2210 * remove the data mbufs, if any, to preserve
2211 * the invariant in the case of PR_ADDR that
2212 * requires MT_SONAME mbufs at the head of
2215 if (pr->pr_flags & PR_ATOMIC &&
2216 ((flags & MSG_PEEK) == 0))
2217 (void)sbdroprecord_locked(&so->so_rcv);
2218 SOCKBUF_UNLOCK(&so->so_rcv);
2222 uio->uio_resid -= len;
2223 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2224 if (len == m->m_len - moff) {
2225 if (m->m_flags & M_EOR)
2227 if (flags & MSG_PEEK) {
2231 nextrecord = m->m_nextpkt;
2232 sbfree(&so->so_rcv, m);
2234 m->m_nextpkt = NULL;
2237 so->so_rcv.sb_mb = m = m->m_next;
2240 so->so_rcv.sb_mb = m_free(m);
2241 m = so->so_rcv.sb_mb;
2243 sockbuf_pushsync(&so->so_rcv, nextrecord);
2244 SBLASTRECORDCHK(&so->so_rcv);
2245 SBLASTMBUFCHK(&so->so_rcv);
2248 if (flags & MSG_PEEK)
2252 if (flags & MSG_DONTWAIT) {
2253 *mp = m_copym(m, 0, len,
2257 * m_copym() couldn't
2259 * Adjust uio_resid back
2261 * down by len bytes,
2262 * which we didn't end
2263 * up "copying" over).
2265 uio->uio_resid += len;
2269 SOCKBUF_UNLOCK(&so->so_rcv);
2270 *mp = m_copym(m, 0, len,
2272 SOCKBUF_LOCK(&so->so_rcv);
2275 sbcut_locked(&so->so_rcv, len);
2278 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2279 if (so->so_oobmark) {
2280 if ((flags & MSG_PEEK) == 0) {
2281 so->so_oobmark -= len;
2282 if (so->so_oobmark == 0) {
2283 so->so_rcv.sb_state |= SBS_RCVATMARK;
2288 if (offset == so->so_oobmark)
2292 if (flags & MSG_EOR)
2295 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2296 * must not quit until "uio->uio_resid == 0" or an error
2297 * termination. If a signal/timeout occurs, return with a
2298 * short count but without error. Keep sockbuf locked
2299 * against other readers.
2301 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2302 !sosendallatonce(so) && nextrecord == NULL) {
2303 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2305 so->so_rcv.sb_state & SBS_CANTRCVMORE)
2308 * Notify the protocol that some data has been
2309 * drained before blocking.
2311 if (pr->pr_flags & PR_WANTRCVD) {
2312 SOCKBUF_UNLOCK(&so->so_rcv);
2314 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2315 SOCKBUF_LOCK(&so->so_rcv);
2317 SBLASTRECORDCHK(&so->so_rcv);
2318 SBLASTMBUFCHK(&so->so_rcv);
2320 * We could receive some data while was notifying
2321 * the protocol. Skip blocking in this case.
2323 if (so->so_rcv.sb_mb == NULL) {
2324 error = sbwait(&so->so_rcv);
2326 SOCKBUF_UNLOCK(&so->so_rcv);
2330 m = so->so_rcv.sb_mb;
2332 nextrecord = m->m_nextpkt;
2336 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2337 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2339 if ((flags & MSG_PEEK) == 0)
2340 (void) sbdroprecord_locked(&so->so_rcv);
2342 if ((flags & MSG_PEEK) == 0) {
2345 * First part is an inline SB_EMPTY_FIXUP(). Second
2346 * part makes sure sb_lastrecord is up-to-date if
2347 * there is still data in the socket buffer.
2349 so->so_rcv.sb_mb = nextrecord;
2350 if (so->so_rcv.sb_mb == NULL) {
2351 so->so_rcv.sb_mbtail = NULL;
2352 so->so_rcv.sb_lastrecord = NULL;
2353 } else if (nextrecord->m_nextpkt == NULL)
2354 so->so_rcv.sb_lastrecord = nextrecord;
2356 SBLASTRECORDCHK(&so->so_rcv);
2357 SBLASTMBUFCHK(&so->so_rcv);
2359 * If soreceive() is being done from the socket callback,
2360 * then don't need to generate ACK to peer to update window,
2361 * since ACK will be generated on return to TCP.
2363 if (!(flags & MSG_SOCALLBCK) &&
2364 (pr->pr_flags & PR_WANTRCVD)) {
2365 SOCKBUF_UNLOCK(&so->so_rcv);
2367 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2368 SOCKBUF_LOCK(&so->so_rcv);
2371 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2372 if (orig_resid == uio->uio_resid && orig_resid &&
2373 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2374 SOCKBUF_UNLOCK(&so->so_rcv);
2377 SOCKBUF_UNLOCK(&so->so_rcv);
2382 sbunlock(&so->so_rcv);
2387 * Optimized version of soreceive() for stream (TCP) sockets.
2390 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2391 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2393 int len = 0, error = 0, flags, oresid;
2395 struct mbuf *m, *n = NULL;
2397 /* We only do stream sockets. */
2398 if (so->so_type != SOCK_STREAM)
2403 flags = *flagsp &~ MSG_EOR;
2406 if (controlp != NULL)
2408 if (flags & MSG_OOB)
2409 return (soreceive_rcvoob(so, uio, flags));
2417 * KTLS store TLS records as records with a control message to
2418 * describe the framing.
2420 * We check once here before acquiring locks to optimize the
2423 if (sb->sb_tls_info != NULL)
2424 return (soreceive_generic(so, psa, uio, mp0, controlp,
2428 /* Prevent other readers from entering the socket. */
2429 error = sblock(sb, SBLOCKWAIT(flags));
2435 if (sb->sb_tls_info != NULL) {
2438 return (soreceive_generic(so, psa, uio, mp0, controlp,
2443 /* Easy one, no space to copyout anything. */
2444 if (uio->uio_resid == 0) {
2448 oresid = uio->uio_resid;
2450 /* We will never ever get anything unless we are or were connected. */
2451 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2457 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2459 /* Abort if socket has reported problems. */
2461 if (sbavail(sb) > 0)
2463 if (oresid > uio->uio_resid)
2465 error = so->so_error;
2466 if (!(flags & MSG_PEEK))
2471 /* Door is closed. Deliver what is left, if any. */
2472 if (sb->sb_state & SBS_CANTRCVMORE) {
2473 if (sbavail(sb) > 0)
2479 /* Socket buffer is empty and we shall not block. */
2480 if (sbavail(sb) == 0 &&
2481 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2486 /* Socket buffer got some data that we shall deliver now. */
2487 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2488 ((so->so_state & SS_NBIO) ||
2489 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2490 sbavail(sb) >= sb->sb_lowat ||
2491 sbavail(sb) >= uio->uio_resid ||
2492 sbavail(sb) >= sb->sb_hiwat) ) {
2496 /* On MSG_WAITALL we must wait until all data or error arrives. */
2497 if ((flags & MSG_WAITALL) &&
2498 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2502 * Wait and block until (more) data comes in.
2503 * NB: Drops the sockbuf lock during wait.
2511 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2512 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2513 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2517 uio->uio_td->td_ru.ru_msgrcv++;
2519 /* Fill uio until full or current end of socket buffer is reached. */
2520 len = min(uio->uio_resid, sbavail(sb));
2522 /* Dequeue as many mbufs as possible. */
2523 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2527 m_cat(*mp0, sb->sb_mb);
2529 m != NULL && m->m_len <= len;
2531 KASSERT(!(m->m_flags & M_NOTAVAIL),
2532 ("%s: m %p not available", __func__, m));
2534 uio->uio_resid -= m->m_len;
2540 sb->sb_lastrecord = sb->sb_mb;
2541 if (sb->sb_mb == NULL)
2544 /* Copy the remainder. */
2546 KASSERT(sb->sb_mb != NULL,
2547 ("%s: len > 0 && sb->sb_mb empty", __func__));
2549 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2551 len = 0; /* Don't flush data from sockbuf. */
2553 uio->uio_resid -= len;
2564 /* NB: Must unlock socket buffer as uiomove may sleep. */
2566 error = m_mbuftouio(uio, sb->sb_mb, len);
2571 SBLASTRECORDCHK(sb);
2575 * Remove the delivered data from the socket buffer unless we
2576 * were only peeking.
2578 if (!(flags & MSG_PEEK)) {
2580 sbdrop_locked(sb, len);
2582 /* Notify protocol that we drained some data. */
2583 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2584 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2585 !(flags & MSG_SOCALLBCK))) {
2588 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2594 * For MSG_WAITALL we may have to loop again and wait for
2595 * more data to come in.
2597 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2600 SOCKBUF_LOCK_ASSERT(sb);
2601 SBLASTRECORDCHK(sb);
2609 * Optimized version of soreceive() for simple datagram cases from userspace.
2610 * Unlike in the stream case, we're able to drop a datagram if copyout()
2611 * fails, and because we handle datagrams atomically, we don't need to use a
2612 * sleep lock to prevent I/O interlacing.
2615 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2616 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2618 struct mbuf *m, *m2;
2621 struct protosw *pr = so->so_proto;
2622 struct mbuf *nextrecord;
2626 if (controlp != NULL)
2629 flags = *flagsp &~ MSG_EOR;
2634 * For any complicated cases, fall back to the full
2635 * soreceive_generic().
2637 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2638 return (soreceive_generic(so, psa, uio, mp0, controlp,
2642 * Enforce restrictions on use.
2644 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2645 ("soreceive_dgram: wantrcvd"));
2646 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2647 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2648 ("soreceive_dgram: SBS_RCVATMARK"));
2649 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2650 ("soreceive_dgram: P_CONNREQUIRED"));
2653 * Loop blocking while waiting for a datagram.
2655 SOCKBUF_LOCK(&so->so_rcv);
2656 while ((m = so->so_rcv.sb_mb) == NULL) {
2657 KASSERT(sbavail(&so->so_rcv) == 0,
2658 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2659 sbavail(&so->so_rcv)));
2661 error = so->so_error;
2663 SOCKBUF_UNLOCK(&so->so_rcv);
2666 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2667 uio->uio_resid == 0) {
2668 SOCKBUF_UNLOCK(&so->so_rcv);
2671 if ((so->so_state & SS_NBIO) ||
2672 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2673 SOCKBUF_UNLOCK(&so->so_rcv);
2674 return (EWOULDBLOCK);
2676 SBLASTRECORDCHK(&so->so_rcv);
2677 SBLASTMBUFCHK(&so->so_rcv);
2678 error = sbwait(&so->so_rcv);
2680 SOCKBUF_UNLOCK(&so->so_rcv);
2684 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2687 uio->uio_td->td_ru.ru_msgrcv++;
2688 SBLASTRECORDCHK(&so->so_rcv);
2689 SBLASTMBUFCHK(&so->so_rcv);
2690 nextrecord = m->m_nextpkt;
2691 if (nextrecord == NULL) {
2692 KASSERT(so->so_rcv.sb_lastrecord == m,
2693 ("soreceive_dgram: lastrecord != m"));
2696 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2697 ("soreceive_dgram: m_nextpkt != nextrecord"));
2700 * Pull 'm' and its chain off the front of the packet queue.
2702 so->so_rcv.sb_mb = NULL;
2703 sockbuf_pushsync(&so->so_rcv, nextrecord);
2706 * Walk 'm's chain and free that many bytes from the socket buffer.
2708 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2709 sbfree(&so->so_rcv, m2);
2712 * Do a few last checks before we let go of the lock.
2714 SBLASTRECORDCHK(&so->so_rcv);
2715 SBLASTMBUFCHK(&so->so_rcv);
2716 SOCKBUF_UNLOCK(&so->so_rcv);
2718 if (pr->pr_flags & PR_ADDR) {
2719 KASSERT(m->m_type == MT_SONAME,
2720 ("m->m_type == %d", m->m_type));
2722 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2727 /* XXXRW: Can this happen? */
2732 * Packet to copyout() is now in 'm' and it is disconnected from the
2735 * Process one or more MT_CONTROL mbufs present before any data mbufs
2736 * in the first mbuf chain on the socket buffer. We call into the
2737 * protocol to perform externalization (or freeing if controlp ==
2738 * NULL). In some cases there can be only MT_CONTROL mbufs without
2741 if (m->m_type == MT_CONTROL) {
2742 struct mbuf *cm = NULL, *cmn;
2743 struct mbuf **cme = &cm;
2749 cme = &(*cme)->m_next;
2751 } while (m != NULL && m->m_type == MT_CONTROL);
2752 while (cm != NULL) {
2755 if (pr->pr_domain->dom_externalize != NULL) {
2756 error = (*pr->pr_domain->dom_externalize)
2757 (cm, controlp, flags);
2758 } else if (controlp != NULL)
2762 if (controlp != NULL) {
2763 while (*controlp != NULL)
2764 controlp = &(*controlp)->m_next;
2769 KASSERT(m == NULL || m->m_type == MT_DATA,
2770 ("soreceive_dgram: !data"));
2771 while (m != NULL && uio->uio_resid > 0) {
2772 len = uio->uio_resid;
2775 error = uiomove(mtod(m, char *), (int)len, uio);
2780 if (len == m->m_len)
2797 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2798 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2802 CURVNET_SET(so->so_vnet);
2803 if (!SOLISTENING(so))
2804 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2805 mp0, controlp, flagsp));
2813 soshutdown(struct socket *so, int how)
2815 struct protosw *pr = so->so_proto;
2816 int error, soerror_enotconn;
2818 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2821 soerror_enotconn = 0;
2823 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2825 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2826 * invoked on a datagram sockets, however historically we would
2827 * actually tear socket down. This is known to be leveraged by
2828 * some applications to unblock process waiting in recvXXX(2)
2829 * by other process that it shares that socket with. Try to meet
2830 * both backward-compatibility and POSIX requirements by forcing
2831 * ENOTCONN but still asking protocol to perform pru_shutdown().
2833 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so))
2835 soerror_enotconn = 1;
2838 if (SOLISTENING(so)) {
2839 if (how != SHUT_WR) {
2841 so->so_error = ECONNABORTED;
2842 solisten_wakeup(so); /* unlocks so */
2847 CURVNET_SET(so->so_vnet);
2848 if (pr->pr_usrreqs->pru_flush != NULL)
2849 (*pr->pr_usrreqs->pru_flush)(so, how);
2852 if (how != SHUT_RD) {
2853 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2854 wakeup(&so->so_timeo);
2856 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2858 wakeup(&so->so_timeo);
2862 return (soerror_enotconn ? ENOTCONN : 0);
2866 sorflush(struct socket *so)
2868 struct sockbuf *sb = &so->so_rcv;
2869 struct protosw *pr = so->so_proto;
2875 * In order to avoid calling dom_dispose with the socket buffer mutex
2876 * held, and in order to generally avoid holding the lock for a long
2877 * time, we make a copy of the socket buffer and clear the original
2878 * (except locks, state). The new socket buffer copy won't have
2879 * initialized locks so we can only call routines that won't use or
2880 * assert those locks.
2882 * Dislodge threads currently blocked in receive and wait to acquire
2883 * a lock against other simultaneous readers before clearing the
2884 * socket buffer. Don't let our acquire be interrupted by a signal
2885 * despite any existing socket disposition on interruptable waiting.
2888 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2891 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2892 * and mutex data unchanged.
2895 bzero(&aso, sizeof(aso));
2896 aso.so_pcb = so->so_pcb;
2897 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2898 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2899 bzero(&sb->sb_startzero,
2900 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2905 * Dispose of special rights and flush the copied socket. Don't call
2906 * any unsafe routines (that rely on locks being initialized) on aso.
2908 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2909 (*pr->pr_domain->dom_dispose)(&aso);
2910 sbrelease_internal(&aso.so_rcv, so);
2914 * Wrapper for Socket established helper hook.
2915 * Parameters: socket, context of the hook point, hook id.
2918 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2920 struct socket_hhook_data hhook_data = {
2927 CURVNET_SET(so->so_vnet);
2928 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2931 /* Ugly but needed, since hhooks return void for now */
2932 return (hhook_data.status);
2936 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2937 * additional variant to handle the case where the option value needs to be
2938 * some kind of integer, but not a specific size. In addition to their use
2939 * here, these functions are also called by the protocol-level pr_ctloutput()
2943 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2948 * If the user gives us more than we wanted, we ignore it, but if we
2949 * don't get the minimum length the caller wants, we return EINVAL.
2950 * On success, sopt->sopt_valsize is set to however much we actually
2953 if ((valsize = sopt->sopt_valsize) < minlen)
2956 sopt->sopt_valsize = valsize = len;
2958 if (sopt->sopt_td != NULL)
2959 return (copyin(sopt->sopt_val, buf, valsize));
2961 bcopy(sopt->sopt_val, buf, valsize);
2966 * Kernel version of setsockopt(2).
2968 * XXX: optlen is size_t, not socklen_t
2971 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2974 struct sockopt sopt;
2976 sopt.sopt_level = level;
2977 sopt.sopt_name = optname;
2978 sopt.sopt_dir = SOPT_SET;
2979 sopt.sopt_val = optval;
2980 sopt.sopt_valsize = optlen;
2981 sopt.sopt_td = NULL;
2982 return (sosetopt(so, &sopt));
2986 sosetopt(struct socket *so, struct sockopt *sopt)
2997 CURVNET_SET(so->so_vnet);
2999 if (sopt->sopt_level != SOL_SOCKET) {
3000 if (so->so_proto->pr_ctloutput != NULL)
3001 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3003 error = ENOPROTOOPT;
3005 switch (sopt->sopt_name) {
3006 case SO_ACCEPTFILTER:
3007 error = accept_filt_setopt(so, sopt);
3013 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3016 if (l.l_linger < 0 ||
3017 l.l_linger > USHRT_MAX ||
3018 l.l_linger > (INT_MAX / hz)) {
3023 so->so_linger = l.l_linger;
3025 so->so_options |= SO_LINGER;
3027 so->so_options &= ~SO_LINGER;
3034 case SO_USELOOPBACK:
3038 case SO_REUSEPORT_LB:
3045 error = sooptcopyin(sopt, &optval, sizeof optval,
3051 so->so_options |= sopt->sopt_name;
3053 so->so_options &= ~sopt->sopt_name;
3058 error = sooptcopyin(sopt, &optval, sizeof optval,
3063 if (optval < 0 || optval >= rt_numfibs) {
3067 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3068 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3069 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3070 so->so_fibnum = optval;
3075 case SO_USER_COOKIE:
3076 error = sooptcopyin(sopt, &val32, sizeof val32,
3080 so->so_user_cookie = val32;
3087 error = sooptcopyin(sopt, &optval, sizeof optval,
3093 * Values < 1 make no sense for any of these options,
3101 error = sbsetopt(so, sopt->sopt_name, optval);
3106 #ifdef COMPAT_FREEBSD32
3107 if (SV_CURPROC_FLAG(SV_ILP32)) {
3108 struct timeval32 tv32;
3110 error = sooptcopyin(sopt, &tv32, sizeof tv32,
3112 CP(tv32, tv, tv_sec);
3113 CP(tv32, tv, tv_usec);
3116 error = sooptcopyin(sopt, &tv, sizeof tv,
3120 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3121 tv.tv_usec >= 1000000) {
3125 if (tv.tv_sec > INT32_MAX)
3129 switch (sopt->sopt_name) {
3131 so->so_snd.sb_timeo = val;
3134 so->so_rcv.sb_timeo = val;
3141 error = sooptcopyin(sopt, &extmac, sizeof extmac,
3145 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3153 error = sooptcopyin(sopt, &optval, sizeof optval,
3157 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3161 so->so_ts_clock = optval;
3164 case SO_MAX_PACING_RATE:
3165 error = sooptcopyin(sopt, &val32, sizeof(val32),
3169 so->so_max_pacing_rate = val32;
3173 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3174 error = hhook_run_socket(so, sopt,
3177 error = ENOPROTOOPT;
3180 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3181 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
3189 * Helper routine for getsockopt.
3192 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3200 * Documented get behavior is that we always return a value, possibly
3201 * truncated to fit in the user's buffer. Traditional behavior is
3202 * that we always tell the user precisely how much we copied, rather
3203 * than something useful like the total amount we had available for
3204 * her. Note that this interface is not idempotent; the entire
3205 * answer must be generated ahead of time.
3207 valsize = min(len, sopt->sopt_valsize);
3208 sopt->sopt_valsize = valsize;
3209 if (sopt->sopt_val != NULL) {
3210 if (sopt->sopt_td != NULL)
3211 error = copyout(buf, sopt->sopt_val, valsize);
3213 bcopy(buf, sopt->sopt_val, valsize);
3219 sogetopt(struct socket *so, struct sockopt *sopt)
3228 CURVNET_SET(so->so_vnet);
3230 if (sopt->sopt_level != SOL_SOCKET) {
3231 if (so->so_proto->pr_ctloutput != NULL)
3232 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3234 error = ENOPROTOOPT;
3238 switch (sopt->sopt_name) {
3239 case SO_ACCEPTFILTER:
3240 error = accept_filt_getopt(so, sopt);
3245 l.l_onoff = so->so_options & SO_LINGER;
3246 l.l_linger = so->so_linger;
3248 error = sooptcopyout(sopt, &l, sizeof l);
3251 case SO_USELOOPBACK:
3257 case SO_REUSEPORT_LB:
3266 optval = so->so_options & sopt->sopt_name;
3268 error = sooptcopyout(sopt, &optval, sizeof optval);
3272 optval = so->so_proto->pr_domain->dom_family;
3276 optval = so->so_type;
3280 optval = so->so_proto->pr_protocol;
3285 optval = so->so_error;
3291 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3292 so->so_snd.sb_hiwat;
3296 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3297 so->so_rcv.sb_hiwat;
3301 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3302 so->so_snd.sb_lowat;
3306 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3307 so->so_rcv.sb_lowat;
3312 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3313 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3314 #ifdef COMPAT_FREEBSD32
3315 if (SV_CURPROC_FLAG(SV_ILP32)) {
3316 struct timeval32 tv32;
3318 CP(tv, tv32, tv_sec);
3319 CP(tv, tv32, tv_usec);
3320 error = sooptcopyout(sopt, &tv32, sizeof tv32);
3323 error = sooptcopyout(sopt, &tv, sizeof tv);
3328 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3332 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3336 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3344 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3348 error = mac_getsockopt_peerlabel(
3349 sopt->sopt_td->td_ucred, so, &extmac);
3352 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3358 case SO_LISTENQLIMIT:
3359 optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3363 optval = SOLISTENING(so) ? so->sol_qlen : 0;
3366 case SO_LISTENINCQLEN:
3367 optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3371 optval = so->so_ts_clock;
3374 case SO_MAX_PACING_RATE:
3375 optval = so->so_max_pacing_rate;
3379 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3380 error = hhook_run_socket(so, sopt,
3383 error = ENOPROTOOPT;
3395 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3397 struct mbuf *m, *m_prev;
3398 int sopt_size = sopt->sopt_valsize;
3400 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3403 if (sopt_size > MLEN) {
3404 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3405 if ((m->m_flags & M_EXT) == 0) {
3409 m->m_len = min(MCLBYTES, sopt_size);
3411 m->m_len = min(MLEN, sopt_size);
3413 sopt_size -= m->m_len;
3418 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3423 if (sopt_size > MLEN) {
3424 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3426 if ((m->m_flags & M_EXT) == 0) {
3431 m->m_len = min(MCLBYTES, sopt_size);
3433 m->m_len = min(MLEN, sopt_size);
3435 sopt_size -= m->m_len;
3443 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3445 struct mbuf *m0 = m;
3447 if (sopt->sopt_val == NULL)
3449 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3450 if (sopt->sopt_td != NULL) {
3453 error = copyin(sopt->sopt_val, mtod(m, char *),
3460 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3461 sopt->sopt_valsize -= m->m_len;
3462 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3465 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3466 panic("ip6_sooptmcopyin");
3471 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3473 struct mbuf *m0 = m;
3476 if (sopt->sopt_val == NULL)
3478 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3479 if (sopt->sopt_td != NULL) {
3482 error = copyout(mtod(m, char *), sopt->sopt_val,
3489 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3490 sopt->sopt_valsize -= m->m_len;
3491 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3492 valsize += m->m_len;
3496 /* enough soopt buffer should be given from user-land */
3500 sopt->sopt_valsize = valsize;
3505 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3506 * out-of-band data, which will then notify socket consumers.
3509 sohasoutofband(struct socket *so)
3512 if (so->so_sigio != NULL)
3513 pgsigio(&so->so_sigio, SIGURG, 0);
3514 selwakeuppri(&so->so_rdsel, PSOCK);
3518 sopoll(struct socket *so, int events, struct ucred *active_cred,
3523 * We do not need to set or assert curvnet as long as everyone uses
3526 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3531 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3537 if (SOLISTENING(so)) {
3538 if (!(events & (POLLIN | POLLRDNORM)))
3540 else if (!TAILQ_EMPTY(&so->sol_comp))
3541 revents = events & (POLLIN | POLLRDNORM);
3542 else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3543 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3545 selrecord(td, &so->so_rdsel);
3550 SOCKBUF_LOCK(&so->so_snd);
3551 SOCKBUF_LOCK(&so->so_rcv);
3552 if (events & (POLLIN | POLLRDNORM))
3553 if (soreadabledata(so))
3554 revents |= events & (POLLIN | POLLRDNORM);
3555 if (events & (POLLOUT | POLLWRNORM))
3556 if (sowriteable(so))
3557 revents |= events & (POLLOUT | POLLWRNORM);
3558 if (events & (POLLPRI | POLLRDBAND))
3559 if (so->so_oobmark ||
3560 (so->so_rcv.sb_state & SBS_RCVATMARK))
3561 revents |= events & (POLLPRI | POLLRDBAND);
3562 if ((events & POLLINIGNEOF) == 0) {
3563 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3564 revents |= events & (POLLIN | POLLRDNORM);
3565 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3571 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3572 selrecord(td, &so->so_rdsel);
3573 so->so_rcv.sb_flags |= SB_SEL;
3575 if (events & (POLLOUT | POLLWRNORM)) {
3576 selrecord(td, &so->so_wrsel);
3577 so->so_snd.sb_flags |= SB_SEL;
3580 SOCKBUF_UNLOCK(&so->so_rcv);
3581 SOCKBUF_UNLOCK(&so->so_snd);
3588 soo_kqfilter(struct file *fp, struct knote *kn)
3590 struct socket *so = kn->kn_fp->f_data;
3594 switch (kn->kn_filter) {
3596 kn->kn_fop = &soread_filtops;
3597 knl = &so->so_rdsel.si_note;
3601 kn->kn_fop = &sowrite_filtops;
3602 knl = &so->so_wrsel.si_note;
3606 kn->kn_fop = &soempty_filtops;
3607 knl = &so->so_wrsel.si_note;
3615 if (SOLISTENING(so)) {
3616 knlist_add(knl, kn, 1);
3619 knlist_add(knl, kn, 1);
3620 sb->sb_flags |= SB_KNOTE;
3628 * Some routines that return EOPNOTSUPP for entry points that are not
3629 * supported by a protocol. Fill in as needed.
3632 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3639 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3646 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3653 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3660 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3668 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3675 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3683 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3690 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3691 struct ifnet *ifp, struct thread *td)
3698 pru_disconnect_notsupp(struct socket *so)
3705 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3712 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3719 pru_rcvd_notsupp(struct socket *so, int flags)
3726 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3733 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3734 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3741 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3744 return (EOPNOTSUPP);
3748 * This isn't really a ``null'' operation, but it's the default one and
3749 * doesn't do anything destructive.
3752 pru_sense_null(struct socket *so, struct stat *sb)
3755 sb->st_blksize = so->so_snd.sb_hiwat;
3760 pru_shutdown_notsupp(struct socket *so)
3767 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3774 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3775 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3782 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3783 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3790 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3798 filt_sordetach(struct knote *kn)
3800 struct socket *so = kn->kn_fp->f_data;
3803 knlist_remove(&so->so_rdsel.si_note, kn, 1);
3804 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3805 so->so_rcv.sb_flags &= ~SB_KNOTE;
3806 so_rdknl_unlock(so);
3811 filt_soread(struct knote *kn, long hint)
3815 so = kn->kn_fp->f_data;
3817 if (SOLISTENING(so)) {
3818 SOCK_LOCK_ASSERT(so);
3819 kn->kn_data = so->sol_qlen;
3821 kn->kn_flags |= EV_EOF;
3822 kn->kn_fflags = so->so_error;
3825 return (!TAILQ_EMPTY(&so->sol_comp));
3828 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3830 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3831 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3832 kn->kn_flags |= EV_EOF;
3833 kn->kn_fflags = so->so_error;
3835 } else if (so->so_error) /* temporary udp error */
3838 if (kn->kn_sfflags & NOTE_LOWAT) {
3839 if (kn->kn_data >= kn->kn_sdata)
3841 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3844 /* This hook returning non-zero indicates an event, not error */
3845 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3849 filt_sowdetach(struct knote *kn)
3851 struct socket *so = kn->kn_fp->f_data;
3854 knlist_remove(&so->so_wrsel.si_note, kn, 1);
3855 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3856 so->so_snd.sb_flags &= ~SB_KNOTE;
3857 so_wrknl_unlock(so);
3862 filt_sowrite(struct knote *kn, long hint)
3866 so = kn->kn_fp->f_data;
3868 if (SOLISTENING(so))
3871 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3872 kn->kn_data = sbspace(&so->so_snd);
3874 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3876 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3877 kn->kn_flags |= EV_EOF;
3878 kn->kn_fflags = so->so_error;
3880 } else if (so->so_error) /* temporary udp error */
3882 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3883 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3885 else if (kn->kn_sfflags & NOTE_LOWAT)
3886 return (kn->kn_data >= kn->kn_sdata);
3888 return (kn->kn_data >= so->so_snd.sb_lowat);
3892 filt_soempty(struct knote *kn, long hint)
3896 so = kn->kn_fp->f_data;
3898 if (SOLISTENING(so))
3901 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3902 kn->kn_data = sbused(&so->so_snd);
3904 if (kn->kn_data == 0)
3911 socheckuid(struct socket *so, uid_t uid)
3916 if (so->so_cred->cr_uid != uid)
3922 * These functions are used by protocols to notify the socket layer (and its
3923 * consumers) of state changes in the sockets driven by protocol-side events.
3927 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3929 * Normal sequence from the active (originating) side is that
3930 * soisconnecting() is called during processing of connect() call, resulting
3931 * in an eventual call to soisconnected() if/when the connection is
3932 * established. When the connection is torn down soisdisconnecting() is
3933 * called during processing of disconnect() call, and soisdisconnected() is
3934 * called when the connection to the peer is totally severed. The semantics
3935 * of these routines are such that connectionless protocols can call
3936 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3937 * calls when setting up a ``connection'' takes no time.
3939 * From the passive side, a socket is created with two queues of sockets:
3940 * so_incomp for connections in progress and so_comp for connections already
3941 * made and awaiting user acceptance. As a protocol is preparing incoming
3942 * connections, it creates a socket structure queued on so_incomp by calling
3943 * sonewconn(). When the connection is established, soisconnected() is
3944 * called, and transfers the socket structure to so_comp, making it available
3947 * If a socket is closed with sockets on either so_incomp or so_comp, these
3948 * sockets are dropped.
3950 * If higher-level protocols are implemented in the kernel, the wakeups done
3951 * here will sometimes cause software-interrupt process scheduling.
3954 soisconnecting(struct socket *so)
3958 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3959 so->so_state |= SS_ISCONNECTING;
3964 soisconnected(struct socket *so)
3968 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3969 so->so_state |= SS_ISCONNECTED;
3971 if (so->so_qstate == SQ_INCOMP) {
3972 struct socket *head = so->so_listen;
3975 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3977 * Promoting a socket from incomplete queue to complete, we
3978 * need to go through reverse order of locking. We first do
3979 * trylock, and if that doesn't succeed, we go the hard way
3980 * leaving a reference and rechecking consistency after proper
3983 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3986 SOLISTEN_LOCK(head);
3988 if (__predict_false(head != so->so_listen)) {
3990 * The socket went off the listen queue,
3991 * should be lost race to close(2) of sol.
3992 * The socket is about to soabort().
3998 /* Not the last one, as so holds a ref. */
3999 refcount_release(&head->so_count);
4002 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4003 TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4004 head->sol_incqlen--;
4005 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4007 so->so_qstate = SQ_COMP;
4009 solisten_wakeup(head); /* unlocks */
4011 SOCKBUF_LOCK(&so->so_rcv);
4012 soupcall_set(so, SO_RCV,
4013 head->sol_accept_filter->accf_callback,
4014 head->sol_accept_filter_arg);
4015 so->so_options &= ~SO_ACCEPTFILTER;
4016 ret = head->sol_accept_filter->accf_callback(so,
4017 head->sol_accept_filter_arg, M_NOWAIT);
4018 if (ret == SU_ISCONNECTED) {
4019 soupcall_clear(so, SO_RCV);
4020 SOCKBUF_UNLOCK(&so->so_rcv);
4023 SOCKBUF_UNLOCK(&so->so_rcv);
4025 SOLISTEN_UNLOCK(head);
4030 wakeup(&so->so_timeo);
4036 soisdisconnecting(struct socket *so)
4040 so->so_state &= ~SS_ISCONNECTING;
4041 so->so_state |= SS_ISDISCONNECTING;
4043 if (!SOLISTENING(so)) {
4044 SOCKBUF_LOCK(&so->so_rcv);
4045 socantrcvmore_locked(so);
4046 SOCKBUF_LOCK(&so->so_snd);
4047 socantsendmore_locked(so);
4050 wakeup(&so->so_timeo);
4054 soisdisconnected(struct socket *so)
4060 * There is at least one reader of so_state that does not
4061 * acquire socket lock, namely soreceive_generic(). Ensure
4062 * that it never sees all flags that track connection status
4063 * cleared, by ordering the update with a barrier semantic of
4064 * our release thread fence.
4066 so->so_state |= SS_ISDISCONNECTED;
4067 atomic_thread_fence_rel();
4068 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4070 if (!SOLISTENING(so)) {
4072 SOCKBUF_LOCK(&so->so_rcv);
4073 socantrcvmore_locked(so);
4074 SOCKBUF_LOCK(&so->so_snd);
4075 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4076 socantsendmore_locked(so);
4079 wakeup(&so->so_timeo);
4083 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4086 sodupsockaddr(const struct sockaddr *sa, int mflags)
4088 struct sockaddr *sa2;
4090 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4092 bcopy(sa, sa2, sa->sa_len);
4097 * Register per-socket destructor.
4100 sodtor_set(struct socket *so, so_dtor_t *func)
4103 SOCK_LOCK_ASSERT(so);
4108 * Register per-socket buffer upcalls.
4111 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4115 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4125 panic("soupcall_set: bad which");
4127 SOCKBUF_LOCK_ASSERT(sb);
4128 sb->sb_upcall = func;
4129 sb->sb_upcallarg = arg;
4130 sb->sb_flags |= SB_UPCALL;
4134 soupcall_clear(struct socket *so, int which)
4138 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4148 panic("soupcall_clear: bad which");
4150 SOCKBUF_LOCK_ASSERT(sb);
4151 KASSERT(sb->sb_upcall != NULL,
4152 ("%s: so %p no upcall to clear", __func__, so));
4153 sb->sb_upcall = NULL;
4154 sb->sb_upcallarg = NULL;
4155 sb->sb_flags &= ~SB_UPCALL;
4159 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4162 SOLISTEN_LOCK_ASSERT(so);
4163 so->sol_upcall = func;
4164 so->sol_upcallarg = arg;
4168 so_rdknl_lock(void *arg)
4170 struct socket *so = arg;
4172 if (SOLISTENING(so))
4175 SOCKBUF_LOCK(&so->so_rcv);
4179 so_rdknl_unlock(void *arg)
4181 struct socket *so = arg;
4183 if (SOLISTENING(so))
4186 SOCKBUF_UNLOCK(&so->so_rcv);
4190 so_rdknl_assert_lock(void *arg, int what)
4192 struct socket *so = arg;
4194 if (what == LA_LOCKED) {
4195 if (SOLISTENING(so))
4196 SOCK_LOCK_ASSERT(so);
4198 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4200 if (SOLISTENING(so))
4201 SOCK_UNLOCK_ASSERT(so);
4203 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4208 so_wrknl_lock(void *arg)
4210 struct socket *so = arg;
4212 if (SOLISTENING(so))
4215 SOCKBUF_LOCK(&so->so_snd);
4219 so_wrknl_unlock(void *arg)
4221 struct socket *so = arg;
4223 if (SOLISTENING(so))
4226 SOCKBUF_UNLOCK(&so->so_snd);
4230 so_wrknl_assert_lock(void *arg, int what)
4232 struct socket *so = arg;
4234 if (what == LA_LOCKED) {
4235 if (SOLISTENING(so))
4236 SOCK_LOCK_ASSERT(so);
4238 SOCKBUF_LOCK_ASSERT(&so->so_snd);
4240 if (SOLISTENING(so))
4241 SOCK_UNLOCK_ASSERT(so);
4243 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4248 * Create an external-format (``xsocket'') structure using the information in
4249 * the kernel-format socket structure pointed to by so. This is done to
4250 * reduce the spew of irrelevant information over this interface, to isolate
4251 * user code from changes in the kernel structure, and potentially to provide
4252 * information-hiding if we decide that some of this information should be
4253 * hidden from users.
4256 sotoxsocket(struct socket *so, struct xsocket *xso)
4259 bzero(xso, sizeof(*xso));
4260 xso->xso_len = sizeof *xso;
4261 xso->xso_so = (uintptr_t)so;
4262 xso->so_type = so->so_type;
4263 xso->so_options = so->so_options;
4264 xso->so_linger = so->so_linger;
4265 xso->so_state = so->so_state;
4266 xso->so_pcb = (uintptr_t)so->so_pcb;
4267 xso->xso_protocol = so->so_proto->pr_protocol;
4268 xso->xso_family = so->so_proto->pr_domain->dom_family;
4269 xso->so_timeo = so->so_timeo;
4270 xso->so_error = so->so_error;
4271 xso->so_uid = so->so_cred->cr_uid;
4272 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4273 if (SOLISTENING(so)) {
4274 xso->so_qlen = so->sol_qlen;
4275 xso->so_incqlen = so->sol_incqlen;
4276 xso->so_qlimit = so->sol_qlimit;
4277 xso->so_oobmark = 0;
4279 xso->so_state |= so->so_qstate;
4280 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4281 xso->so_oobmark = so->so_oobmark;
4282 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4283 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4288 so_sockbuf_rcv(struct socket *so)
4291 return (&so->so_rcv);
4295 so_sockbuf_snd(struct socket *so)
4298 return (&so->so_snd);
4302 so_state_get(const struct socket *so)
4305 return (so->so_state);
4309 so_state_set(struct socket *so, int val)
4316 so_options_get(const struct socket *so)
4319 return (so->so_options);
4323 so_options_set(struct socket *so, int val)
4326 so->so_options = val;
4330 so_error_get(const struct socket *so)
4333 return (so->so_error);
4337 so_error_set(struct socket *so, int val)
4344 so_linger_get(const struct socket *so)
4347 return (so->so_linger);
4351 so_linger_set(struct socket *so, int val)
4354 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4355 ("%s: val %d out of range", __func__, val));
4357 so->so_linger = val;
4361 so_protosw_get(const struct socket *so)
4364 return (so->so_proto);
4368 so_protosw_set(struct socket *so, struct protosw *val)
4375 so_sorwakeup(struct socket *so)
4382 so_sowwakeup(struct socket *so)
4389 so_sorwakeup_locked(struct socket *so)
4392 sorwakeup_locked(so);
4396 so_sowwakeup_locked(struct socket *so)
4399 sowwakeup_locked(so);
4403 so_lock(struct socket *so)
4410 so_unlock(struct socket *so)