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)
1196 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1197 if ((so->so_state & SS_ISDISCONNECTING) &&
1198 (so->so_state & SS_NBIO))
1200 while (so->so_state & SS_ISCONNECTED) {
1201 error = tsleep(&so->so_timeo,
1202 PSOCK | PCATCH, "soclos",
1203 so->so_linger * hz);
1211 if (so->so_proto->pr_usrreqs->pru_close != NULL)
1212 (*so->so_proto->pr_usrreqs->pru_close)(so);
1215 if ((listening = (so->so_options & SO_ACCEPTCONN))) {
1218 TAILQ_INIT(&lqueue);
1219 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1220 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1222 so->sol_qlen = so->sol_incqlen = 0;
1224 TAILQ_FOREACH(sp, &lqueue, so_list) {
1226 sp->so_qstate = SQ_NONE;
1227 sp->so_listen = NULL;
1229 /* Guaranteed not to be the last. */
1230 refcount_release(&so->so_count);
1233 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1234 so->so_state |= SS_NOFDREF;
1237 struct socket *sp, *tsp;
1239 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1241 if (sp->so_count == 0) {
1245 /* sp is now in sofree() */
1254 * soabort() is used to abruptly tear down a connection, such as when a
1255 * resource limit is reached (listen queue depth exceeded), or if a listen
1256 * socket is closed while there are sockets waiting to be accepted.
1258 * This interface is tricky, because it is called on an unreferenced socket,
1259 * and must be called only by a thread that has actually removed the socket
1260 * from the listen queue it was on, or races with other threads are risked.
1262 * This interface will call into the protocol code, so must not be called
1263 * with any socket locks held. Protocols do call it while holding their own
1264 * recursible protocol mutexes, but this is something that should be subject
1265 * to review in the future.
1268 soabort(struct socket *so)
1272 * In as much as is possible, assert that no references to this
1273 * socket are held. This is not quite the same as asserting that the
1274 * current thread is responsible for arranging for no references, but
1275 * is as close as we can get for now.
1277 KASSERT(so->so_count == 0, ("soabort: so_count"));
1278 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1279 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1282 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1283 (*so->so_proto->pr_usrreqs->pru_abort)(so);
1289 soaccept(struct socket *so, struct sockaddr **nam)
1294 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1295 so->so_state &= ~SS_NOFDREF;
1298 CURVNET_SET(so->so_vnet);
1299 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1305 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1308 return (soconnectat(AT_FDCWD, so, nam, td));
1312 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1316 if (so->so_options & SO_ACCEPTCONN)
1317 return (EOPNOTSUPP);
1319 CURVNET_SET(so->so_vnet);
1321 * If protocol is connection-based, can only connect once.
1322 * Otherwise, if connected, try to disconnect first. This allows
1323 * user to disconnect by connecting to, e.g., a null address.
1325 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1326 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1327 (error = sodisconnect(so)))) {
1331 * Prevent accumulated error from previous connection from
1335 if (fd == AT_FDCWD) {
1336 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1339 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1349 soconnect2(struct socket *so1, struct socket *so2)
1353 CURVNET_SET(so1->so_vnet);
1354 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1360 sodisconnect(struct socket *so)
1364 if ((so->so_state & SS_ISCONNECTED) == 0)
1366 if (so->so_state & SS_ISDISCONNECTING)
1369 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1373 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1376 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1377 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1381 int clen = 0, error, dontroute;
1383 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1384 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1385 ("sosend_dgram: !PR_ATOMIC"));
1388 resid = uio->uio_resid;
1390 resid = top->m_pkthdr.len;
1392 * In theory resid should be unsigned. However, space must be
1393 * signed, as it might be less than 0 if we over-committed, and we
1394 * must use a signed comparison of space and resid. On the other
1395 * hand, a negative resid causes us to loop sending 0-length
1396 * segments to the protocol.
1404 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1406 td->td_ru.ru_msgsnd++;
1407 if (control != NULL)
1408 clen = control->m_len;
1410 SOCKBUF_LOCK(&so->so_snd);
1411 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1412 SOCKBUF_UNLOCK(&so->so_snd);
1417 error = so->so_error;
1419 SOCKBUF_UNLOCK(&so->so_snd);
1422 if ((so->so_state & SS_ISCONNECTED) == 0) {
1424 * `sendto' and `sendmsg' is allowed on a connection-based
1425 * socket if it supports implied connect. Return ENOTCONN if
1426 * not connected and no address is supplied.
1428 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1429 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1430 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1431 !(resid == 0 && clen != 0)) {
1432 SOCKBUF_UNLOCK(&so->so_snd);
1436 } else if (addr == NULL) {
1437 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1440 error = EDESTADDRREQ;
1441 SOCKBUF_UNLOCK(&so->so_snd);
1447 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1448 * problem and need fixing.
1450 space = sbspace(&so->so_snd);
1451 if (flags & MSG_OOB)
1454 SOCKBUF_UNLOCK(&so->so_snd);
1455 if (resid > space) {
1461 if (flags & MSG_EOR)
1462 top->m_flags |= M_EOR;
1465 * Copy the data from userland into a mbuf chain.
1466 * If no data is to be copied in, a single empty mbuf
1469 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1470 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1472 error = EFAULT; /* only possible error */
1475 space -= resid - uio->uio_resid;
1476 resid = uio->uio_resid;
1478 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1480 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1485 so->so_options |= SO_DONTROUTE;
1489 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1490 * of date. We could have received a reset packet in an interrupt or
1491 * maybe we slept while doing page faults in uiomove() etc. We could
1492 * probably recheck again inside the locking protection here, but
1493 * there are probably other places that this also happens. We must
1497 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1498 (flags & MSG_OOB) ? PRUS_OOB :
1500 * If the user set MSG_EOF, the protocol understands this flag and
1501 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1503 ((flags & MSG_EOF) &&
1504 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1507 /* If there is more to send set PRUS_MORETOCOME */
1508 (flags & MSG_MORETOCOME) ||
1509 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1510 top, addr, control, td);
1513 so->so_options &= ~SO_DONTROUTE;
1522 if (control != NULL)
1528 * Send on a socket. If send must go all at once and message is larger than
1529 * send buffering, then hard error. Lock against other senders. If must go
1530 * all at once and not enough room now, then inform user that this would
1531 * block and do nothing. Otherwise, if nonblocking, send as much as
1532 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1533 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1534 * in mbuf chain must be small enough to send all at once.
1536 * Returns nonzero on error, timeout or signal; callers must check for short
1537 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1541 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1542 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1546 int clen = 0, error, dontroute;
1547 int atomic = sosendallatonce(so) || top;
1550 struct ktls_session *tls;
1551 int tls_enq_cnt, tls_pruflag;
1555 tls_rtype = TLS_RLTYPE_APP;
1558 resid = uio->uio_resid;
1559 else if ((top->m_flags & M_PKTHDR) != 0)
1560 resid = top->m_pkthdr.len;
1562 resid = m_length(top, NULL);
1564 * In theory resid should be unsigned. However, space must be
1565 * signed, as it might be less than 0 if we over-committed, and we
1566 * must use a signed comparison of space and resid. On the other
1567 * hand, a negative resid causes us to loop sending 0-length
1568 * segments to the protocol.
1570 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1571 * type sockets since that's an error.
1573 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1579 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1580 (so->so_proto->pr_flags & PR_ATOMIC);
1582 td->td_ru.ru_msgsnd++;
1583 if (control != NULL)
1584 clen = control->m_len;
1586 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1592 tls = ktls_hold(so->so_snd.sb_tls_info);
1594 if (tls->mode == TCP_TLS_MODE_SW)
1595 tls_pruflag = PRUS_NOTREADY;
1597 if (control != NULL) {
1598 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1600 if (clen >= sizeof(*cm) &&
1601 cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1602 tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1614 SOCKBUF_LOCK(&so->so_snd);
1615 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1616 SOCKBUF_UNLOCK(&so->so_snd);
1621 error = so->so_error;
1623 SOCKBUF_UNLOCK(&so->so_snd);
1626 if ((so->so_state & SS_ISCONNECTED) == 0) {
1628 * `sendto' and `sendmsg' is allowed on a connection-
1629 * based socket if it supports implied connect.
1630 * Return ENOTCONN if not connected and no address is
1633 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1634 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1635 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1636 !(resid == 0 && clen != 0)) {
1637 SOCKBUF_UNLOCK(&so->so_snd);
1641 } else if (addr == NULL) {
1642 SOCKBUF_UNLOCK(&so->so_snd);
1643 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1646 error = EDESTADDRREQ;
1650 space = sbspace(&so->so_snd);
1651 if (flags & MSG_OOB)
1653 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1654 clen > so->so_snd.sb_hiwat) {
1655 SOCKBUF_UNLOCK(&so->so_snd);
1659 if (space < resid + clen &&
1660 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1661 if ((so->so_state & SS_NBIO) ||
1662 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1663 SOCKBUF_UNLOCK(&so->so_snd);
1664 error = EWOULDBLOCK;
1667 error = sbwait(&so->so_snd);
1668 SOCKBUF_UNLOCK(&so->so_snd);
1673 SOCKBUF_UNLOCK(&so->so_snd);
1678 if (flags & MSG_EOR)
1679 top->m_flags |= M_EOR;
1682 ktls_frame(top, tls, &tls_enq_cnt,
1684 tls_rtype = TLS_RLTYPE_APP;
1689 * Copy the data from userland into a mbuf
1690 * chain. If resid is 0, which can happen
1691 * only if we have control to send, then
1692 * a single empty mbuf is returned. This
1693 * is a workaround to prevent protocol send
1698 top = m_uiotombuf(uio, M_WAITOK, space,
1699 tls->params.max_frame_len,
1701 ((flags & MSG_EOR) ? M_EOR : 0));
1703 ktls_frame(top, tls,
1704 &tls_enq_cnt, tls_rtype);
1706 tls_rtype = TLS_RLTYPE_APP;
1709 top = m_uiotombuf(uio, M_WAITOK, space,
1710 (atomic ? max_hdr : 0),
1711 (atomic ? M_PKTHDR : 0) |
1712 ((flags & MSG_EOR) ? M_EOR : 0));
1714 error = EFAULT; /* only possible error */
1717 space -= resid - uio->uio_resid;
1718 resid = uio->uio_resid;
1722 so->so_options |= SO_DONTROUTE;
1726 * XXX all the SBS_CANTSENDMORE checks previously
1727 * done could be out of date. We could have received
1728 * a reset packet in an interrupt or maybe we slept
1729 * while doing page faults in uiomove() etc. We
1730 * could probably recheck again inside the locking
1731 * protection here, but there are probably other
1732 * places that this also happens. We must rethink
1737 pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1739 * If the user set MSG_EOF, the protocol understands
1740 * this flag and nothing left to send then use
1741 * PRU_SEND_EOF instead of PRU_SEND.
1743 ((flags & MSG_EOF) &&
1744 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1747 /* If there is more to send set PRUS_MORETOCOME. */
1748 (flags & MSG_MORETOCOME) ||
1749 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1752 pru_flag |= tls_pruflag;
1755 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1756 pru_flag, top, addr, control, td);
1760 so->so_options &= ~SO_DONTROUTE;
1765 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1767 * Note that error is intentionally
1770 * Like sendfile(), we rely on the
1771 * completion routine (pru_ready())
1772 * to free the mbufs in the event that
1773 * pru_send() encountered an error and
1774 * did not append them to the sockbuf.
1777 ktls_enqueue(top, so, tls_enq_cnt);
1785 } while (resid && space > 0);
1789 sbunlock(&so->so_snd);
1797 if (control != NULL)
1803 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1804 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1808 CURVNET_SET(so->so_vnet);
1809 if (!SOLISTENING(so))
1810 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1811 top, control, flags, td);
1822 * The part of soreceive() that implements reading non-inline out-of-band
1823 * data from a socket. For more complete comments, see soreceive(), from
1824 * which this code originated.
1826 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1827 * unable to return an mbuf chain to the caller.
1830 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1832 struct protosw *pr = so->so_proto;
1836 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1839 m = m_get(M_WAITOK, MT_DATA);
1840 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1844 error = uiomove(mtod(m, void *),
1845 (int) min(uio->uio_resid, m->m_len), uio);
1847 } while (uio->uio_resid && error == 0 && m);
1855 * Following replacement or removal of the first mbuf on the first mbuf chain
1856 * of a socket buffer, push necessary state changes back into the socket
1857 * buffer so that other consumers see the values consistently. 'nextrecord'
1858 * is the callers locally stored value of the original value of
1859 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1860 * NOTE: 'nextrecord' may be NULL.
1862 static __inline void
1863 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1866 SOCKBUF_LOCK_ASSERT(sb);
1868 * First, update for the new value of nextrecord. If necessary, make
1869 * it the first record.
1871 if (sb->sb_mb != NULL)
1872 sb->sb_mb->m_nextpkt = nextrecord;
1874 sb->sb_mb = nextrecord;
1877 * Now update any dependent socket buffer fields to reflect the new
1878 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1879 * addition of a second clause that takes care of the case where
1880 * sb_mb has been updated, but remains the last record.
1882 if (sb->sb_mb == NULL) {
1883 sb->sb_mbtail = NULL;
1884 sb->sb_lastrecord = NULL;
1885 } else if (sb->sb_mb->m_nextpkt == NULL)
1886 sb->sb_lastrecord = sb->sb_mb;
1890 * Implement receive operations on a socket. We depend on the way that
1891 * records are added to the sockbuf by sbappend. In particular, each record
1892 * (mbufs linked through m_next) must begin with an address if the protocol
1893 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1894 * data, and then zero or more mbufs of data. In order to allow parallelism
1895 * between network receive and copying to user space, as well as avoid
1896 * sleeping with a mutex held, we release the socket buffer mutex during the
1897 * user space copy. Although the sockbuf is locked, new data may still be
1898 * appended, and thus we must maintain consistency of the sockbuf during that
1901 * The caller may receive the data as a single mbuf chain by supplying an
1902 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1903 * the count in uio_resid.
1906 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1907 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1909 struct mbuf *m, **mp;
1910 int flags, error, offset;
1912 struct protosw *pr = so->so_proto;
1913 struct mbuf *nextrecord;
1915 ssize_t orig_resid = uio->uio_resid;
1920 if (controlp != NULL)
1923 flags = *flagsp &~ MSG_EOR;
1926 if (flags & MSG_OOB)
1927 return (soreceive_rcvoob(so, uio, flags));
1930 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1931 && uio->uio_resid) {
1933 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1936 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1941 SOCKBUF_LOCK(&so->so_rcv);
1942 m = so->so_rcv.sb_mb;
1944 * If we have less data than requested, block awaiting more (subject
1945 * to any timeout) if:
1946 * 1. the current count is less than the low water mark, or
1947 * 2. MSG_DONTWAIT is not set
1949 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1950 sbavail(&so->so_rcv) < uio->uio_resid) &&
1951 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1952 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1953 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1954 ("receive: m == %p sbavail == %u",
1955 m, sbavail(&so->so_rcv)));
1959 error = so->so_error;
1960 if ((flags & MSG_PEEK) == 0)
1962 SOCKBUF_UNLOCK(&so->so_rcv);
1965 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1966 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1970 else if (so->so_rcv.sb_tlsdcc == 0 &&
1971 so->so_rcv.sb_tlscc == 0) {
1975 SOCKBUF_UNLOCK(&so->so_rcv);
1979 for (; m != NULL; m = m->m_next)
1980 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1981 m = so->so_rcv.sb_mb;
1984 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
1985 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
1986 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1987 SOCKBUF_UNLOCK(&so->so_rcv);
1991 if (uio->uio_resid == 0) {
1992 SOCKBUF_UNLOCK(&so->so_rcv);
1995 if ((so->so_state & SS_NBIO) ||
1996 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1997 SOCKBUF_UNLOCK(&so->so_rcv);
1998 error = EWOULDBLOCK;
2001 SBLASTRECORDCHK(&so->so_rcv);
2002 SBLASTMBUFCHK(&so->so_rcv);
2003 error = sbwait(&so->so_rcv);
2004 SOCKBUF_UNLOCK(&so->so_rcv);
2011 * From this point onward, we maintain 'nextrecord' as a cache of the
2012 * pointer to the next record in the socket buffer. We must keep the
2013 * various socket buffer pointers and local stack versions of the
2014 * pointers in sync, pushing out modifications before dropping the
2015 * socket buffer mutex, and re-reading them when picking it up.
2017 * Otherwise, we will race with the network stack appending new data
2018 * or records onto the socket buffer by using inconsistent/stale
2019 * versions of the field, possibly resulting in socket buffer
2022 * By holding the high-level sblock(), we prevent simultaneous
2023 * readers from pulling off the front of the socket buffer.
2025 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2027 uio->uio_td->td_ru.ru_msgrcv++;
2028 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2029 SBLASTRECORDCHK(&so->so_rcv);
2030 SBLASTMBUFCHK(&so->so_rcv);
2031 nextrecord = m->m_nextpkt;
2032 if (pr->pr_flags & PR_ADDR) {
2033 KASSERT(m->m_type == MT_SONAME,
2034 ("m->m_type == %d", m->m_type));
2037 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2039 if (flags & MSG_PEEK) {
2042 sbfree(&so->so_rcv, m);
2043 so->so_rcv.sb_mb = m_free(m);
2044 m = so->so_rcv.sb_mb;
2045 sockbuf_pushsync(&so->so_rcv, nextrecord);
2050 * Process one or more MT_CONTROL mbufs present before any data mbufs
2051 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
2052 * just copy the data; if !MSG_PEEK, we call into the protocol to
2053 * perform externalization (or freeing if controlp == NULL).
2055 if (m != NULL && m->m_type == MT_CONTROL) {
2056 struct mbuf *cm = NULL, *cmn;
2057 struct mbuf **cme = &cm;
2059 struct cmsghdr *cmsg;
2060 struct tls_get_record tgr;
2063 * For MSG_TLSAPPDATA, check for a non-application data
2064 * record. If found, return ENXIO without removing
2065 * it from the receive queue. This allows a subsequent
2066 * call without MSG_TLSAPPDATA to receive it.
2067 * Note that, for TLS, there should only be a single
2068 * control mbuf with the TLS_GET_RECORD message in it.
2070 if (flags & MSG_TLSAPPDATA) {
2071 cmsg = mtod(m, struct cmsghdr *);
2072 if (cmsg->cmsg_type == TLS_GET_RECORD &&
2073 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2074 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2075 /* This will need to change for TLS 1.3. */
2076 if (tgr.tls_type != TLS_RLTYPE_APP) {
2077 SOCKBUF_UNLOCK(&so->so_rcv);
2086 if (flags & MSG_PEEK) {
2087 if (controlp != NULL) {
2088 *controlp = m_copym(m, 0, m->m_len,
2090 controlp = &(*controlp)->m_next;
2094 sbfree(&so->so_rcv, m);
2095 so->so_rcv.sb_mb = m->m_next;
2098 cme = &(*cme)->m_next;
2099 m = so->so_rcv.sb_mb;
2101 } while (m != NULL && m->m_type == MT_CONTROL);
2102 if ((flags & MSG_PEEK) == 0)
2103 sockbuf_pushsync(&so->so_rcv, nextrecord);
2104 while (cm != NULL) {
2107 if (pr->pr_domain->dom_externalize != NULL) {
2108 SOCKBUF_UNLOCK(&so->so_rcv);
2110 error = (*pr->pr_domain->dom_externalize)
2111 (cm, controlp, flags);
2112 SOCKBUF_LOCK(&so->so_rcv);
2113 } else if (controlp != NULL)
2117 if (controlp != NULL) {
2119 while (*controlp != NULL)
2120 controlp = &(*controlp)->m_next;
2125 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2127 nextrecord = so->so_rcv.sb_mb;
2131 if ((flags & MSG_PEEK) == 0) {
2132 KASSERT(m->m_nextpkt == nextrecord,
2133 ("soreceive: post-control, nextrecord !sync"));
2134 if (nextrecord == NULL) {
2135 KASSERT(so->so_rcv.sb_mb == m,
2136 ("soreceive: post-control, sb_mb!=m"));
2137 KASSERT(so->so_rcv.sb_lastrecord == m,
2138 ("soreceive: post-control, lastrecord!=m"));
2142 if (type == MT_OOBDATA)
2145 if ((flags & MSG_PEEK) == 0) {
2146 KASSERT(so->so_rcv.sb_mb == nextrecord,
2147 ("soreceive: sb_mb != nextrecord"));
2148 if (so->so_rcv.sb_mb == NULL) {
2149 KASSERT(so->so_rcv.sb_lastrecord == NULL,
2150 ("soreceive: sb_lastercord != NULL"));
2154 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2155 SBLASTRECORDCHK(&so->so_rcv);
2156 SBLASTMBUFCHK(&so->so_rcv);
2159 * Now continue to read any data mbufs off of the head of the socket
2160 * buffer until the read request is satisfied. Note that 'type' is
2161 * used to store the type of any mbuf reads that have happened so far
2162 * such that soreceive() can stop reading if the type changes, which
2163 * causes soreceive() to return only one of regular data and inline
2164 * out-of-band data in a single socket receive operation.
2168 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2171 * If the type of mbuf has changed since the last mbuf
2172 * examined ('type'), end the receive operation.
2174 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2175 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2176 if (type != m->m_type)
2178 } else if (type == MT_OOBDATA)
2181 KASSERT(m->m_type == MT_DATA,
2182 ("m->m_type == %d", m->m_type));
2183 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2184 len = uio->uio_resid;
2185 if (so->so_oobmark && len > so->so_oobmark - offset)
2186 len = so->so_oobmark - offset;
2187 if (len > m->m_len - moff)
2188 len = m->m_len - moff;
2190 * If mp is set, just pass back the mbufs. Otherwise copy
2191 * them out via the uio, then free. Sockbuf must be
2192 * consistent here (points to current mbuf, it points to next
2193 * record) when we drop priority; we must note any additions
2194 * to the sockbuf when we block interrupts again.
2197 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2198 SBLASTRECORDCHK(&so->so_rcv);
2199 SBLASTMBUFCHK(&so->so_rcv);
2200 SOCKBUF_UNLOCK(&so->so_rcv);
2201 if ((m->m_flags & M_EXTPG) != 0)
2202 error = m_unmappedtouio(m, moff, uio, (int)len);
2204 error = uiomove(mtod(m, char *) + moff,
2206 SOCKBUF_LOCK(&so->so_rcv);
2209 * The MT_SONAME mbuf has already been removed
2210 * from the record, so it is necessary to
2211 * remove the data mbufs, if any, to preserve
2212 * the invariant in the case of PR_ADDR that
2213 * requires MT_SONAME mbufs at the head of
2216 if (pr->pr_flags & PR_ATOMIC &&
2217 ((flags & MSG_PEEK) == 0))
2218 (void)sbdroprecord_locked(&so->so_rcv);
2219 SOCKBUF_UNLOCK(&so->so_rcv);
2223 uio->uio_resid -= len;
2224 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2225 if (len == m->m_len - moff) {
2226 if (m->m_flags & M_EOR)
2228 if (flags & MSG_PEEK) {
2232 nextrecord = m->m_nextpkt;
2233 sbfree(&so->so_rcv, m);
2235 m->m_nextpkt = NULL;
2238 so->so_rcv.sb_mb = m = m->m_next;
2241 so->so_rcv.sb_mb = m_free(m);
2242 m = so->so_rcv.sb_mb;
2244 sockbuf_pushsync(&so->so_rcv, nextrecord);
2245 SBLASTRECORDCHK(&so->so_rcv);
2246 SBLASTMBUFCHK(&so->so_rcv);
2249 if (flags & MSG_PEEK)
2253 if (flags & MSG_DONTWAIT) {
2254 *mp = m_copym(m, 0, len,
2258 * m_copym() couldn't
2260 * Adjust uio_resid back
2262 * down by len bytes,
2263 * which we didn't end
2264 * up "copying" over).
2266 uio->uio_resid += len;
2270 SOCKBUF_UNLOCK(&so->so_rcv);
2271 *mp = m_copym(m, 0, len,
2273 SOCKBUF_LOCK(&so->so_rcv);
2276 sbcut_locked(&so->so_rcv, len);
2279 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2280 if (so->so_oobmark) {
2281 if ((flags & MSG_PEEK) == 0) {
2282 so->so_oobmark -= len;
2283 if (so->so_oobmark == 0) {
2284 so->so_rcv.sb_state |= SBS_RCVATMARK;
2289 if (offset == so->so_oobmark)
2293 if (flags & MSG_EOR)
2296 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2297 * must not quit until "uio->uio_resid == 0" or an error
2298 * termination. If a signal/timeout occurs, return with a
2299 * short count but without error. Keep sockbuf locked
2300 * against other readers.
2302 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2303 !sosendallatonce(so) && nextrecord == NULL) {
2304 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2306 so->so_rcv.sb_state & SBS_CANTRCVMORE)
2309 * Notify the protocol that some data has been
2310 * drained before blocking.
2312 if (pr->pr_flags & PR_WANTRCVD) {
2313 SOCKBUF_UNLOCK(&so->so_rcv);
2315 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2316 SOCKBUF_LOCK(&so->so_rcv);
2318 SBLASTRECORDCHK(&so->so_rcv);
2319 SBLASTMBUFCHK(&so->so_rcv);
2321 * We could receive some data while was notifying
2322 * the protocol. Skip blocking in this case.
2324 if (so->so_rcv.sb_mb == NULL) {
2325 error = sbwait(&so->so_rcv);
2327 SOCKBUF_UNLOCK(&so->so_rcv);
2331 m = so->so_rcv.sb_mb;
2333 nextrecord = m->m_nextpkt;
2337 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2338 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2340 if ((flags & MSG_PEEK) == 0)
2341 (void) sbdroprecord_locked(&so->so_rcv);
2343 if ((flags & MSG_PEEK) == 0) {
2346 * First part is an inline SB_EMPTY_FIXUP(). Second
2347 * part makes sure sb_lastrecord is up-to-date if
2348 * there is still data in the socket buffer.
2350 so->so_rcv.sb_mb = nextrecord;
2351 if (so->so_rcv.sb_mb == NULL) {
2352 so->so_rcv.sb_mbtail = NULL;
2353 so->so_rcv.sb_lastrecord = NULL;
2354 } else if (nextrecord->m_nextpkt == NULL)
2355 so->so_rcv.sb_lastrecord = nextrecord;
2357 SBLASTRECORDCHK(&so->so_rcv);
2358 SBLASTMBUFCHK(&so->so_rcv);
2360 * If soreceive() is being done from the socket callback,
2361 * then don't need to generate ACK to peer to update window,
2362 * since ACK will be generated on return to TCP.
2364 if (!(flags & MSG_SOCALLBCK) &&
2365 (pr->pr_flags & PR_WANTRCVD)) {
2366 SOCKBUF_UNLOCK(&so->so_rcv);
2368 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2369 SOCKBUF_LOCK(&so->so_rcv);
2372 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2373 if (orig_resid == uio->uio_resid && orig_resid &&
2374 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2375 SOCKBUF_UNLOCK(&so->so_rcv);
2378 SOCKBUF_UNLOCK(&so->so_rcv);
2383 sbunlock(&so->so_rcv);
2388 * Optimized version of soreceive() for stream (TCP) sockets.
2391 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2392 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2394 int len = 0, error = 0, flags, oresid;
2396 struct mbuf *m, *n = NULL;
2398 /* We only do stream sockets. */
2399 if (so->so_type != SOCK_STREAM)
2404 flags = *flagsp &~ MSG_EOR;
2407 if (controlp != NULL)
2409 if (flags & MSG_OOB)
2410 return (soreceive_rcvoob(so, uio, flags));
2418 * KTLS store TLS records as records with a control message to
2419 * describe the framing.
2421 * We check once here before acquiring locks to optimize the
2424 if (sb->sb_tls_info != NULL)
2425 return (soreceive_generic(so, psa, uio, mp0, controlp,
2429 /* Prevent other readers from entering the socket. */
2430 error = sblock(sb, SBLOCKWAIT(flags));
2436 if (sb->sb_tls_info != NULL) {
2439 return (soreceive_generic(so, psa, uio, mp0, controlp,
2444 /* Easy one, no space to copyout anything. */
2445 if (uio->uio_resid == 0) {
2449 oresid = uio->uio_resid;
2451 /* We will never ever get anything unless we are or were connected. */
2452 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2458 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2460 /* Abort if socket has reported problems. */
2462 if (sbavail(sb) > 0)
2464 if (oresid > uio->uio_resid)
2466 error = so->so_error;
2467 if (!(flags & MSG_PEEK))
2472 /* Door is closed. Deliver what is left, if any. */
2473 if (sb->sb_state & SBS_CANTRCVMORE) {
2474 if (sbavail(sb) > 0)
2480 /* Socket buffer is empty and we shall not block. */
2481 if (sbavail(sb) == 0 &&
2482 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2487 /* Socket buffer got some data that we shall deliver now. */
2488 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2489 ((so->so_state & SS_NBIO) ||
2490 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2491 sbavail(sb) >= sb->sb_lowat ||
2492 sbavail(sb) >= uio->uio_resid ||
2493 sbavail(sb) >= sb->sb_hiwat) ) {
2497 /* On MSG_WAITALL we must wait until all data or error arrives. */
2498 if ((flags & MSG_WAITALL) &&
2499 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2503 * Wait and block until (more) data comes in.
2504 * NB: Drops the sockbuf lock during wait.
2512 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2513 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2514 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2518 uio->uio_td->td_ru.ru_msgrcv++;
2520 /* Fill uio until full or current end of socket buffer is reached. */
2521 len = min(uio->uio_resid, sbavail(sb));
2523 /* Dequeue as many mbufs as possible. */
2524 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2528 m_cat(*mp0, sb->sb_mb);
2530 m != NULL && m->m_len <= len;
2532 KASSERT(!(m->m_flags & M_NOTAVAIL),
2533 ("%s: m %p not available", __func__, m));
2535 uio->uio_resid -= m->m_len;
2541 sb->sb_lastrecord = sb->sb_mb;
2542 if (sb->sb_mb == NULL)
2545 /* Copy the remainder. */
2547 KASSERT(sb->sb_mb != NULL,
2548 ("%s: len > 0 && sb->sb_mb empty", __func__));
2550 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2552 len = 0; /* Don't flush data from sockbuf. */
2554 uio->uio_resid -= len;
2565 /* NB: Must unlock socket buffer as uiomove may sleep. */
2567 error = m_mbuftouio(uio, sb->sb_mb, len);
2572 SBLASTRECORDCHK(sb);
2576 * Remove the delivered data from the socket buffer unless we
2577 * were only peeking.
2579 if (!(flags & MSG_PEEK)) {
2581 sbdrop_locked(sb, len);
2583 /* Notify protocol that we drained some data. */
2584 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2585 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2586 !(flags & MSG_SOCALLBCK))) {
2589 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2595 * For MSG_WAITALL we may have to loop again and wait for
2596 * more data to come in.
2598 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2601 SOCKBUF_LOCK_ASSERT(sb);
2602 SBLASTRECORDCHK(sb);
2610 * Optimized version of soreceive() for simple datagram cases from userspace.
2611 * Unlike in the stream case, we're able to drop a datagram if copyout()
2612 * fails, and because we handle datagrams atomically, we don't need to use a
2613 * sleep lock to prevent I/O interlacing.
2616 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2617 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2619 struct mbuf *m, *m2;
2622 struct protosw *pr = so->so_proto;
2623 struct mbuf *nextrecord;
2627 if (controlp != NULL)
2630 flags = *flagsp &~ MSG_EOR;
2635 * For any complicated cases, fall back to the full
2636 * soreceive_generic().
2638 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2639 return (soreceive_generic(so, psa, uio, mp0, controlp,
2643 * Enforce restrictions on use.
2645 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2646 ("soreceive_dgram: wantrcvd"));
2647 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2648 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2649 ("soreceive_dgram: SBS_RCVATMARK"));
2650 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2651 ("soreceive_dgram: P_CONNREQUIRED"));
2654 * Loop blocking while waiting for a datagram.
2656 SOCKBUF_LOCK(&so->so_rcv);
2657 while ((m = so->so_rcv.sb_mb) == NULL) {
2658 KASSERT(sbavail(&so->so_rcv) == 0,
2659 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2660 sbavail(&so->so_rcv)));
2662 error = so->so_error;
2664 SOCKBUF_UNLOCK(&so->so_rcv);
2667 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2668 uio->uio_resid == 0) {
2669 SOCKBUF_UNLOCK(&so->so_rcv);
2672 if ((so->so_state & SS_NBIO) ||
2673 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2674 SOCKBUF_UNLOCK(&so->so_rcv);
2675 return (EWOULDBLOCK);
2677 SBLASTRECORDCHK(&so->so_rcv);
2678 SBLASTMBUFCHK(&so->so_rcv);
2679 error = sbwait(&so->so_rcv);
2681 SOCKBUF_UNLOCK(&so->so_rcv);
2685 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2688 uio->uio_td->td_ru.ru_msgrcv++;
2689 SBLASTRECORDCHK(&so->so_rcv);
2690 SBLASTMBUFCHK(&so->so_rcv);
2691 nextrecord = m->m_nextpkt;
2692 if (nextrecord == NULL) {
2693 KASSERT(so->so_rcv.sb_lastrecord == m,
2694 ("soreceive_dgram: lastrecord != m"));
2697 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2698 ("soreceive_dgram: m_nextpkt != nextrecord"));
2701 * Pull 'm' and its chain off the front of the packet queue.
2703 so->so_rcv.sb_mb = NULL;
2704 sockbuf_pushsync(&so->so_rcv, nextrecord);
2707 * Walk 'm's chain and free that many bytes from the socket buffer.
2709 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2710 sbfree(&so->so_rcv, m2);
2713 * Do a few last checks before we let go of the lock.
2715 SBLASTRECORDCHK(&so->so_rcv);
2716 SBLASTMBUFCHK(&so->so_rcv);
2717 SOCKBUF_UNLOCK(&so->so_rcv);
2719 if (pr->pr_flags & PR_ADDR) {
2720 KASSERT(m->m_type == MT_SONAME,
2721 ("m->m_type == %d", m->m_type));
2723 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2728 /* XXXRW: Can this happen? */
2733 * Packet to copyout() is now in 'm' and it is disconnected from the
2736 * Process one or more MT_CONTROL mbufs present before any data mbufs
2737 * in the first mbuf chain on the socket buffer. We call into the
2738 * protocol to perform externalization (or freeing if controlp ==
2739 * NULL). In some cases there can be only MT_CONTROL mbufs without
2742 if (m->m_type == MT_CONTROL) {
2743 struct mbuf *cm = NULL, *cmn;
2744 struct mbuf **cme = &cm;
2750 cme = &(*cme)->m_next;
2752 } while (m != NULL && m->m_type == MT_CONTROL);
2753 while (cm != NULL) {
2756 if (pr->pr_domain->dom_externalize != NULL) {
2757 error = (*pr->pr_domain->dom_externalize)
2758 (cm, controlp, flags);
2759 } else if (controlp != NULL)
2763 if (controlp != NULL) {
2764 while (*controlp != NULL)
2765 controlp = &(*controlp)->m_next;
2770 KASSERT(m == NULL || m->m_type == MT_DATA,
2771 ("soreceive_dgram: !data"));
2772 while (m != NULL && uio->uio_resid > 0) {
2773 len = uio->uio_resid;
2776 error = uiomove(mtod(m, char *), (int)len, uio);
2781 if (len == m->m_len)
2798 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2799 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2803 CURVNET_SET(so->so_vnet);
2804 if (!SOLISTENING(so))
2805 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2806 mp0, controlp, flagsp));
2814 soshutdown(struct socket *so, int how)
2816 struct protosw *pr = so->so_proto;
2817 int error, soerror_enotconn;
2819 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2822 soerror_enotconn = 0;
2824 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2826 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2827 * invoked on a datagram sockets, however historically we would
2828 * actually tear socket down. This is known to be leveraged by
2829 * some applications to unblock process waiting in recvXXX(2)
2830 * by other process that it shares that socket with. Try to meet
2831 * both backward-compatibility and POSIX requirements by forcing
2832 * ENOTCONN but still asking protocol to perform pru_shutdown().
2834 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so))
2836 soerror_enotconn = 1;
2839 if (SOLISTENING(so)) {
2840 if (how != SHUT_WR) {
2842 so->so_error = ECONNABORTED;
2843 solisten_wakeup(so); /* unlocks so */
2848 CURVNET_SET(so->so_vnet);
2849 if (pr->pr_usrreqs->pru_flush != NULL)
2850 (*pr->pr_usrreqs->pru_flush)(so, how);
2853 if (how != SHUT_RD) {
2854 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2855 wakeup(&so->so_timeo);
2857 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2859 wakeup(&so->so_timeo);
2863 return (soerror_enotconn ? ENOTCONN : 0);
2867 sorflush(struct socket *so)
2869 struct sockbuf *sb = &so->so_rcv;
2870 struct protosw *pr = so->so_proto;
2876 * In order to avoid calling dom_dispose with the socket buffer mutex
2877 * held, and in order to generally avoid holding the lock for a long
2878 * time, we make a copy of the socket buffer and clear the original
2879 * (except locks, state). The new socket buffer copy won't have
2880 * initialized locks so we can only call routines that won't use or
2881 * assert those locks.
2883 * Dislodge threads currently blocked in receive and wait to acquire
2884 * a lock against other simultaneous readers before clearing the
2885 * socket buffer. Don't let our acquire be interrupted by a signal
2886 * despite any existing socket disposition on interruptable waiting.
2889 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2892 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2893 * and mutex data unchanged.
2896 bzero(&aso, sizeof(aso));
2897 aso.so_pcb = so->so_pcb;
2898 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2899 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2900 bzero(&sb->sb_startzero,
2901 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2906 * Dispose of special rights and flush the copied socket. Don't call
2907 * any unsafe routines (that rely on locks being initialized) on aso.
2909 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2910 (*pr->pr_domain->dom_dispose)(&aso);
2911 sbrelease_internal(&aso.so_rcv, so);
2915 * Wrapper for Socket established helper hook.
2916 * Parameters: socket, context of the hook point, hook id.
2919 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2921 struct socket_hhook_data hhook_data = {
2928 CURVNET_SET(so->so_vnet);
2929 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2932 /* Ugly but needed, since hhooks return void for now */
2933 return (hhook_data.status);
2937 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2938 * additional variant to handle the case where the option value needs to be
2939 * some kind of integer, but not a specific size. In addition to their use
2940 * here, these functions are also called by the protocol-level pr_ctloutput()
2944 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2949 * If the user gives us more than we wanted, we ignore it, but if we
2950 * don't get the minimum length the caller wants, we return EINVAL.
2951 * On success, sopt->sopt_valsize is set to however much we actually
2954 if ((valsize = sopt->sopt_valsize) < minlen)
2957 sopt->sopt_valsize = valsize = len;
2959 if (sopt->sopt_td != NULL)
2960 return (copyin(sopt->sopt_val, buf, valsize));
2962 bcopy(sopt->sopt_val, buf, valsize);
2967 * Kernel version of setsockopt(2).
2969 * XXX: optlen is size_t, not socklen_t
2972 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2975 struct sockopt sopt;
2977 sopt.sopt_level = level;
2978 sopt.sopt_name = optname;
2979 sopt.sopt_dir = SOPT_SET;
2980 sopt.sopt_val = optval;
2981 sopt.sopt_valsize = optlen;
2982 sopt.sopt_td = NULL;
2983 return (sosetopt(so, &sopt));
2987 sosetopt(struct socket *so, struct sockopt *sopt)
2998 CURVNET_SET(so->so_vnet);
3000 if (sopt->sopt_level != SOL_SOCKET) {
3001 if (so->so_proto->pr_ctloutput != NULL)
3002 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3004 error = ENOPROTOOPT;
3006 switch (sopt->sopt_name) {
3007 case SO_ACCEPTFILTER:
3008 error = accept_filt_setopt(so, sopt);
3014 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3017 if (l.l_linger < 0 ||
3018 l.l_linger > USHRT_MAX ||
3019 l.l_linger > (INT_MAX / hz)) {
3024 so->so_linger = l.l_linger;
3026 so->so_options |= SO_LINGER;
3028 so->so_options &= ~SO_LINGER;
3035 case SO_USELOOPBACK:
3039 case SO_REUSEPORT_LB:
3046 error = sooptcopyin(sopt, &optval, sizeof optval,
3052 so->so_options |= sopt->sopt_name;
3054 so->so_options &= ~sopt->sopt_name;
3059 error = sooptcopyin(sopt, &optval, sizeof optval,
3064 if (optval < 0 || optval >= rt_numfibs) {
3068 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3069 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3070 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3071 so->so_fibnum = optval;
3076 case SO_USER_COOKIE:
3077 error = sooptcopyin(sopt, &val32, sizeof val32,
3081 so->so_user_cookie = val32;
3088 error = sooptcopyin(sopt, &optval, sizeof optval,
3094 * Values < 1 make no sense for any of these options,
3102 error = sbsetopt(so, sopt->sopt_name, optval);
3107 #ifdef COMPAT_FREEBSD32
3108 if (SV_CURPROC_FLAG(SV_ILP32)) {
3109 struct timeval32 tv32;
3111 error = sooptcopyin(sopt, &tv32, sizeof tv32,
3113 CP(tv32, tv, tv_sec);
3114 CP(tv32, tv, tv_usec);
3117 error = sooptcopyin(sopt, &tv, sizeof tv,
3121 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3122 tv.tv_usec >= 1000000) {
3126 if (tv.tv_sec > INT32_MAX)
3130 switch (sopt->sopt_name) {
3132 so->so_snd.sb_timeo = val;
3135 so->so_rcv.sb_timeo = val;
3142 error = sooptcopyin(sopt, &extmac, sizeof extmac,
3146 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3154 error = sooptcopyin(sopt, &optval, sizeof optval,
3158 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3162 so->so_ts_clock = optval;
3165 case SO_MAX_PACING_RATE:
3166 error = sooptcopyin(sopt, &val32, sizeof(val32),
3170 so->so_max_pacing_rate = val32;
3174 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3175 error = hhook_run_socket(so, sopt,
3178 error = ENOPROTOOPT;
3181 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3182 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
3190 * Helper routine for getsockopt.
3193 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3201 * Documented get behavior is that we always return a value, possibly
3202 * truncated to fit in the user's buffer. Traditional behavior is
3203 * that we always tell the user precisely how much we copied, rather
3204 * than something useful like the total amount we had available for
3205 * her. Note that this interface is not idempotent; the entire
3206 * answer must be generated ahead of time.
3208 valsize = min(len, sopt->sopt_valsize);
3209 sopt->sopt_valsize = valsize;
3210 if (sopt->sopt_val != NULL) {
3211 if (sopt->sopt_td != NULL)
3212 error = copyout(buf, sopt->sopt_val, valsize);
3214 bcopy(buf, sopt->sopt_val, valsize);
3220 sogetopt(struct socket *so, struct sockopt *sopt)
3229 CURVNET_SET(so->so_vnet);
3231 if (sopt->sopt_level != SOL_SOCKET) {
3232 if (so->so_proto->pr_ctloutput != NULL)
3233 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3235 error = ENOPROTOOPT;
3239 switch (sopt->sopt_name) {
3240 case SO_ACCEPTFILTER:
3241 error = accept_filt_getopt(so, sopt);
3246 l.l_onoff = so->so_options & SO_LINGER;
3247 l.l_linger = so->so_linger;
3249 error = sooptcopyout(sopt, &l, sizeof l);
3252 case SO_USELOOPBACK:
3258 case SO_REUSEPORT_LB:
3267 optval = so->so_options & sopt->sopt_name;
3269 error = sooptcopyout(sopt, &optval, sizeof optval);
3273 optval = so->so_proto->pr_domain->dom_family;
3277 optval = so->so_type;
3281 optval = so->so_proto->pr_protocol;
3286 optval = so->so_error;
3292 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3293 so->so_snd.sb_hiwat;
3297 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3298 so->so_rcv.sb_hiwat;
3302 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3303 so->so_snd.sb_lowat;
3307 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3308 so->so_rcv.sb_lowat;
3313 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3314 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3315 #ifdef COMPAT_FREEBSD32
3316 if (SV_CURPROC_FLAG(SV_ILP32)) {
3317 struct timeval32 tv32;
3319 CP(tv, tv32, tv_sec);
3320 CP(tv, tv32, tv_usec);
3321 error = sooptcopyout(sopt, &tv32, sizeof tv32);
3324 error = sooptcopyout(sopt, &tv, sizeof tv);
3329 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3333 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3337 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3345 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3349 error = mac_getsockopt_peerlabel(
3350 sopt->sopt_td->td_ucred, so, &extmac);
3353 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3359 case SO_LISTENQLIMIT:
3360 optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3364 optval = SOLISTENING(so) ? so->sol_qlen : 0;
3367 case SO_LISTENINCQLEN:
3368 optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3372 optval = so->so_ts_clock;
3375 case SO_MAX_PACING_RATE:
3376 optval = so->so_max_pacing_rate;
3380 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3381 error = hhook_run_socket(so, sopt,
3384 error = ENOPROTOOPT;
3396 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3398 struct mbuf *m, *m_prev;
3399 int sopt_size = sopt->sopt_valsize;
3401 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3404 if (sopt_size > MLEN) {
3405 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3406 if ((m->m_flags & M_EXT) == 0) {
3410 m->m_len = min(MCLBYTES, sopt_size);
3412 m->m_len = min(MLEN, sopt_size);
3414 sopt_size -= m->m_len;
3419 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3424 if (sopt_size > MLEN) {
3425 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3427 if ((m->m_flags & M_EXT) == 0) {
3432 m->m_len = min(MCLBYTES, sopt_size);
3434 m->m_len = min(MLEN, sopt_size);
3436 sopt_size -= m->m_len;
3444 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3446 struct mbuf *m0 = m;
3448 if (sopt->sopt_val == NULL)
3450 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3451 if (sopt->sopt_td != NULL) {
3454 error = copyin(sopt->sopt_val, mtod(m, char *),
3461 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3462 sopt->sopt_valsize -= m->m_len;
3463 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3466 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3467 panic("ip6_sooptmcopyin");
3472 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3474 struct mbuf *m0 = m;
3477 if (sopt->sopt_val == NULL)
3479 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3480 if (sopt->sopt_td != NULL) {
3483 error = copyout(mtod(m, char *), sopt->sopt_val,
3490 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3491 sopt->sopt_valsize -= m->m_len;
3492 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3493 valsize += m->m_len;
3497 /* enough soopt buffer should be given from user-land */
3501 sopt->sopt_valsize = valsize;
3506 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3507 * out-of-band data, which will then notify socket consumers.
3510 sohasoutofband(struct socket *so)
3513 if (so->so_sigio != NULL)
3514 pgsigio(&so->so_sigio, SIGURG, 0);
3515 selwakeuppri(&so->so_rdsel, PSOCK);
3519 sopoll(struct socket *so, int events, struct ucred *active_cred,
3524 * We do not need to set or assert curvnet as long as everyone uses
3527 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3532 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3538 if (SOLISTENING(so)) {
3539 if (!(events & (POLLIN | POLLRDNORM)))
3541 else if (!TAILQ_EMPTY(&so->sol_comp))
3542 revents = events & (POLLIN | POLLRDNORM);
3543 else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3544 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3546 selrecord(td, &so->so_rdsel);
3551 SOCKBUF_LOCK(&so->so_snd);
3552 SOCKBUF_LOCK(&so->so_rcv);
3553 if (events & (POLLIN | POLLRDNORM))
3554 if (soreadabledata(so))
3555 revents |= events & (POLLIN | POLLRDNORM);
3556 if (events & (POLLOUT | POLLWRNORM))
3557 if (sowriteable(so))
3558 revents |= events & (POLLOUT | POLLWRNORM);
3559 if (events & (POLLPRI | POLLRDBAND))
3560 if (so->so_oobmark ||
3561 (so->so_rcv.sb_state & SBS_RCVATMARK))
3562 revents |= events & (POLLPRI | POLLRDBAND);
3563 if ((events & POLLINIGNEOF) == 0) {
3564 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3565 revents |= events & (POLLIN | POLLRDNORM);
3566 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3572 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3573 selrecord(td, &so->so_rdsel);
3574 so->so_rcv.sb_flags |= SB_SEL;
3576 if (events & (POLLOUT | POLLWRNORM)) {
3577 selrecord(td, &so->so_wrsel);
3578 so->so_snd.sb_flags |= SB_SEL;
3581 SOCKBUF_UNLOCK(&so->so_rcv);
3582 SOCKBUF_UNLOCK(&so->so_snd);
3589 soo_kqfilter(struct file *fp, struct knote *kn)
3591 struct socket *so = kn->kn_fp->f_data;
3595 switch (kn->kn_filter) {
3597 kn->kn_fop = &soread_filtops;
3598 knl = &so->so_rdsel.si_note;
3602 kn->kn_fop = &sowrite_filtops;
3603 knl = &so->so_wrsel.si_note;
3607 kn->kn_fop = &soempty_filtops;
3608 knl = &so->so_wrsel.si_note;
3616 if (SOLISTENING(so)) {
3617 knlist_add(knl, kn, 1);
3620 knlist_add(knl, kn, 1);
3621 sb->sb_flags |= SB_KNOTE;
3629 * Some routines that return EOPNOTSUPP for entry points that are not
3630 * supported by a protocol. Fill in as needed.
3633 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3640 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3647 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3654 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3661 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3669 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3676 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3684 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3691 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3692 struct ifnet *ifp, struct thread *td)
3699 pru_disconnect_notsupp(struct socket *so)
3706 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3713 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3720 pru_rcvd_notsupp(struct socket *so, int flags)
3727 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3734 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3735 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3742 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3745 return (EOPNOTSUPP);
3749 * This isn't really a ``null'' operation, but it's the default one and
3750 * doesn't do anything destructive.
3753 pru_sense_null(struct socket *so, struct stat *sb)
3756 sb->st_blksize = so->so_snd.sb_hiwat;
3761 pru_shutdown_notsupp(struct socket *so)
3768 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3775 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3776 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3783 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3784 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3791 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3799 filt_sordetach(struct knote *kn)
3801 struct socket *so = kn->kn_fp->f_data;
3804 knlist_remove(&so->so_rdsel.si_note, kn, 1);
3805 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3806 so->so_rcv.sb_flags &= ~SB_KNOTE;
3807 so_rdknl_unlock(so);
3812 filt_soread(struct knote *kn, long hint)
3816 so = kn->kn_fp->f_data;
3818 if (SOLISTENING(so)) {
3819 SOCK_LOCK_ASSERT(so);
3820 kn->kn_data = so->sol_qlen;
3822 kn->kn_flags |= EV_EOF;
3823 kn->kn_fflags = so->so_error;
3826 return (!TAILQ_EMPTY(&so->sol_comp));
3829 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3831 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3832 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3833 kn->kn_flags |= EV_EOF;
3834 kn->kn_fflags = so->so_error;
3836 } else if (so->so_error) /* temporary udp error */
3839 if (kn->kn_sfflags & NOTE_LOWAT) {
3840 if (kn->kn_data >= kn->kn_sdata)
3842 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3845 /* This hook returning non-zero indicates an event, not error */
3846 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3850 filt_sowdetach(struct knote *kn)
3852 struct socket *so = kn->kn_fp->f_data;
3855 knlist_remove(&so->so_wrsel.si_note, kn, 1);
3856 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3857 so->so_snd.sb_flags &= ~SB_KNOTE;
3858 so_wrknl_unlock(so);
3863 filt_sowrite(struct knote *kn, long hint)
3867 so = kn->kn_fp->f_data;
3869 if (SOLISTENING(so))
3872 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3873 kn->kn_data = sbspace(&so->so_snd);
3875 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3877 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3878 kn->kn_flags |= EV_EOF;
3879 kn->kn_fflags = so->so_error;
3881 } else if (so->so_error) /* temporary udp error */
3883 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3884 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3886 else if (kn->kn_sfflags & NOTE_LOWAT)
3887 return (kn->kn_data >= kn->kn_sdata);
3889 return (kn->kn_data >= so->so_snd.sb_lowat);
3893 filt_soempty(struct knote *kn, long hint)
3897 so = kn->kn_fp->f_data;
3899 if (SOLISTENING(so))
3902 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3903 kn->kn_data = sbused(&so->so_snd);
3905 if (kn->kn_data == 0)
3912 socheckuid(struct socket *so, uid_t uid)
3917 if (so->so_cred->cr_uid != uid)
3923 * These functions are used by protocols to notify the socket layer (and its
3924 * consumers) of state changes in the sockets driven by protocol-side events.
3928 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3930 * Normal sequence from the active (originating) side is that
3931 * soisconnecting() is called during processing of connect() call, resulting
3932 * in an eventual call to soisconnected() if/when the connection is
3933 * established. When the connection is torn down soisdisconnecting() is
3934 * called during processing of disconnect() call, and soisdisconnected() is
3935 * called when the connection to the peer is totally severed. The semantics
3936 * of these routines are such that connectionless protocols can call
3937 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3938 * calls when setting up a ``connection'' takes no time.
3940 * From the passive side, a socket is created with two queues of sockets:
3941 * so_incomp for connections in progress and so_comp for connections already
3942 * made and awaiting user acceptance. As a protocol is preparing incoming
3943 * connections, it creates a socket structure queued on so_incomp by calling
3944 * sonewconn(). When the connection is established, soisconnected() is
3945 * called, and transfers the socket structure to so_comp, making it available
3948 * If a socket is closed with sockets on either so_incomp or so_comp, these
3949 * sockets are dropped.
3951 * If higher-level protocols are implemented in the kernel, the wakeups done
3952 * here will sometimes cause software-interrupt process scheduling.
3955 soisconnecting(struct socket *so)
3959 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3960 so->so_state |= SS_ISCONNECTING;
3965 soisconnected(struct socket *so)
3969 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3970 so->so_state |= SS_ISCONNECTED;
3972 if (so->so_qstate == SQ_INCOMP) {
3973 struct socket *head = so->so_listen;
3976 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3978 * Promoting a socket from incomplete queue to complete, we
3979 * need to go through reverse order of locking. We first do
3980 * trylock, and if that doesn't succeed, we go the hard way
3981 * leaving a reference and rechecking consistency after proper
3984 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3987 SOLISTEN_LOCK(head);
3989 if (__predict_false(head != so->so_listen)) {
3991 * The socket went off the listen queue,
3992 * should be lost race to close(2) of sol.
3993 * The socket is about to soabort().
3999 /* Not the last one, as so holds a ref. */
4000 refcount_release(&head->so_count);
4003 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4004 TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4005 head->sol_incqlen--;
4006 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4008 so->so_qstate = SQ_COMP;
4010 solisten_wakeup(head); /* unlocks */
4012 SOCKBUF_LOCK(&so->so_rcv);
4013 soupcall_set(so, SO_RCV,
4014 head->sol_accept_filter->accf_callback,
4015 head->sol_accept_filter_arg);
4016 so->so_options &= ~SO_ACCEPTFILTER;
4017 ret = head->sol_accept_filter->accf_callback(so,
4018 head->sol_accept_filter_arg, M_NOWAIT);
4019 if (ret == SU_ISCONNECTED) {
4020 soupcall_clear(so, SO_RCV);
4021 SOCKBUF_UNLOCK(&so->so_rcv);
4024 SOCKBUF_UNLOCK(&so->so_rcv);
4026 SOLISTEN_UNLOCK(head);
4031 wakeup(&so->so_timeo);
4037 soisdisconnecting(struct socket *so)
4041 so->so_state &= ~SS_ISCONNECTING;
4042 so->so_state |= SS_ISDISCONNECTING;
4044 if (!SOLISTENING(so)) {
4045 SOCKBUF_LOCK(&so->so_rcv);
4046 socantrcvmore_locked(so);
4047 SOCKBUF_LOCK(&so->so_snd);
4048 socantsendmore_locked(so);
4051 wakeup(&so->so_timeo);
4055 soisdisconnected(struct socket *so)
4061 * There is at least one reader of so_state that does not
4062 * acquire socket lock, namely soreceive_generic(). Ensure
4063 * that it never sees all flags that track connection status
4064 * cleared, by ordering the update with a barrier semantic of
4065 * our release thread fence.
4067 so->so_state |= SS_ISDISCONNECTED;
4068 atomic_thread_fence_rel();
4069 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4071 if (!SOLISTENING(so)) {
4073 SOCKBUF_LOCK(&so->so_rcv);
4074 socantrcvmore_locked(so);
4075 SOCKBUF_LOCK(&so->so_snd);
4076 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4077 socantsendmore_locked(so);
4080 wakeup(&so->so_timeo);
4084 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4087 sodupsockaddr(const struct sockaddr *sa, int mflags)
4089 struct sockaddr *sa2;
4091 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4093 bcopy(sa, sa2, sa->sa_len);
4098 * Register per-socket destructor.
4101 sodtor_set(struct socket *so, so_dtor_t *func)
4104 SOCK_LOCK_ASSERT(so);
4109 * Register per-socket buffer upcalls.
4112 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4116 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4126 panic("soupcall_set: bad which");
4128 SOCKBUF_LOCK_ASSERT(sb);
4129 sb->sb_upcall = func;
4130 sb->sb_upcallarg = arg;
4131 sb->sb_flags |= SB_UPCALL;
4135 soupcall_clear(struct socket *so, int which)
4139 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4149 panic("soupcall_clear: bad which");
4151 SOCKBUF_LOCK_ASSERT(sb);
4152 KASSERT(sb->sb_upcall != NULL,
4153 ("%s: so %p no upcall to clear", __func__, so));
4154 sb->sb_upcall = NULL;
4155 sb->sb_upcallarg = NULL;
4156 sb->sb_flags &= ~SB_UPCALL;
4160 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4163 SOLISTEN_LOCK_ASSERT(so);
4164 so->sol_upcall = func;
4165 so->sol_upcallarg = arg;
4169 so_rdknl_lock(void *arg)
4171 struct socket *so = arg;
4173 if (SOLISTENING(so))
4176 SOCKBUF_LOCK(&so->so_rcv);
4180 so_rdknl_unlock(void *arg)
4182 struct socket *so = arg;
4184 if (SOLISTENING(so))
4187 SOCKBUF_UNLOCK(&so->so_rcv);
4191 so_rdknl_assert_lock(void *arg, int what)
4193 struct socket *so = arg;
4195 if (what == LA_LOCKED) {
4196 if (SOLISTENING(so))
4197 SOCK_LOCK_ASSERT(so);
4199 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4201 if (SOLISTENING(so))
4202 SOCK_UNLOCK_ASSERT(so);
4204 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4209 so_wrknl_lock(void *arg)
4211 struct socket *so = arg;
4213 if (SOLISTENING(so))
4216 SOCKBUF_LOCK(&so->so_snd);
4220 so_wrknl_unlock(void *arg)
4222 struct socket *so = arg;
4224 if (SOLISTENING(so))
4227 SOCKBUF_UNLOCK(&so->so_snd);
4231 so_wrknl_assert_lock(void *arg, int what)
4233 struct socket *so = arg;
4235 if (what == LA_LOCKED) {
4236 if (SOLISTENING(so))
4237 SOCK_LOCK_ASSERT(so);
4239 SOCKBUF_LOCK_ASSERT(&so->so_snd);
4241 if (SOLISTENING(so))
4242 SOCK_UNLOCK_ASSERT(so);
4244 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4249 * Create an external-format (``xsocket'') structure using the information in
4250 * the kernel-format socket structure pointed to by so. This is done to
4251 * reduce the spew of irrelevant information over this interface, to isolate
4252 * user code from changes in the kernel structure, and potentially to provide
4253 * information-hiding if we decide that some of this information should be
4254 * hidden from users.
4257 sotoxsocket(struct socket *so, struct xsocket *xso)
4260 bzero(xso, sizeof(*xso));
4261 xso->xso_len = sizeof *xso;
4262 xso->xso_so = (uintptr_t)so;
4263 xso->so_type = so->so_type;
4264 xso->so_options = so->so_options;
4265 xso->so_linger = so->so_linger;
4266 xso->so_state = so->so_state;
4267 xso->so_pcb = (uintptr_t)so->so_pcb;
4268 xso->xso_protocol = so->so_proto->pr_protocol;
4269 xso->xso_family = so->so_proto->pr_domain->dom_family;
4270 xso->so_timeo = so->so_timeo;
4271 xso->so_error = so->so_error;
4272 xso->so_uid = so->so_cred->cr_uid;
4273 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4274 if (SOLISTENING(so)) {
4275 xso->so_qlen = so->sol_qlen;
4276 xso->so_incqlen = so->sol_incqlen;
4277 xso->so_qlimit = so->sol_qlimit;
4278 xso->so_oobmark = 0;
4280 xso->so_state |= so->so_qstate;
4281 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4282 xso->so_oobmark = so->so_oobmark;
4283 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4284 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4289 so_sockbuf_rcv(struct socket *so)
4292 return (&so->so_rcv);
4296 so_sockbuf_snd(struct socket *so)
4299 return (&so->so_snd);
4303 so_state_get(const struct socket *so)
4306 return (so->so_state);
4310 so_state_set(struct socket *so, int val)
4317 so_options_get(const struct socket *so)
4320 return (so->so_options);
4324 so_options_set(struct socket *so, int val)
4327 so->so_options = val;
4331 so_error_get(const struct socket *so)
4334 return (so->so_error);
4338 so_error_set(struct socket *so, int val)
4345 so_linger_get(const struct socket *so)
4348 return (so->so_linger);
4352 so_linger_set(struct socket *so, int val)
4355 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4356 ("%s: val %d out of range", __func__, val));
4358 so->so_linger = val;
4362 so_protosw_get(const struct socket *so)
4365 return (so->so_proto);
4369 so_protosw_set(struct socket *so, struct protosw *val)
4376 so_sorwakeup(struct socket *so)
4383 so_sowwakeup(struct socket *so)
4390 so_sorwakeup_locked(struct socket *so)
4393 sorwakeup_locked(so);
4397 so_sowwakeup_locked(struct socket *so)
4400 sowwakeup_locked(so);
4404 so_lock(struct socket *so)
4411 so_unlock(struct socket *so)