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_locked(void *);
169 static void so_rdknl_assert_unlocked(void *);
170 static void so_wrknl_lock(void *);
171 static void so_wrknl_unlock(void *);
172 static void so_wrknl_assert_locked(void *);
173 static void so_wrknl_assert_unlocked(void *);
175 static void filt_sordetach(struct knote *kn);
176 static int filt_soread(struct knote *kn, long hint);
177 static void filt_sowdetach(struct knote *kn);
178 static int filt_sowrite(struct knote *kn, long hint);
179 static int filt_soempty(struct knote *kn, long hint);
180 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
181 fo_kqfilter_t soo_kqfilter;
183 static struct filterops soread_filtops = {
185 .f_detach = filt_sordetach,
186 .f_event = filt_soread,
188 static struct filterops sowrite_filtops = {
190 .f_detach = filt_sowdetach,
191 .f_event = filt_sowrite,
193 static struct filterops soempty_filtops = {
195 .f_detach = filt_sowdetach,
196 .f_event = filt_soempty,
199 so_gen_t so_gencnt; /* generation count for sockets */
201 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
202 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
204 #define VNET_SO_ASSERT(so) \
205 VNET_ASSERT(curvnet != NULL, \
206 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
208 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
209 #define V_socket_hhh VNET(socket_hhh)
212 * Limit on the number of connections in the listen queue waiting
214 * NB: The original sysctl somaxconn is still available but hidden
215 * to prevent confusion about the actual purpose of this number.
217 static u_int somaxconn = SOMAXCONN;
220 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
226 error = sysctl_handle_int(oidp, &val, 0, req);
227 if (error || !req->newptr )
231 * The purpose of the UINT_MAX / 3 limit, is so that the formula
233 * below, will not overflow.
236 if (val < 1 || val > UINT_MAX / 3)
242 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
243 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, sizeof(int),
244 sysctl_somaxconn, "I",
245 "Maximum listen socket pending connection accept queue size");
246 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
247 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, 0,
248 sizeof(int), sysctl_somaxconn, "I",
249 "Maximum listen socket pending connection accept queue size (compat)");
251 static int numopensockets;
252 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
253 &numopensockets, 0, "Number of open sockets");
256 * accept_mtx locks down per-socket fields relating to accept queues. See
257 * socketvar.h for an annotation of the protected fields of struct socket.
259 struct mtx accept_mtx;
260 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
263 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
266 static struct mtx so_global_mtx;
267 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
270 * General IPC sysctl name space, used by sockets and a variety of other IPC
273 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
277 * Initialize the socket subsystem and set up the socket
280 static uma_zone_t socket_zone;
284 socket_zone_change(void *tag)
287 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
291 socket_hhook_register(int subtype)
294 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
295 &V_socket_hhh[subtype],
296 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
297 printf("%s: WARNING: unable to register hook\n", __func__);
301 socket_hhook_deregister(int subtype)
304 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
305 printf("%s: WARNING: unable to deregister hook\n", __func__);
309 socket_init(void *tag)
312 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
313 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
314 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
315 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
316 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
317 EVENTHANDLER_PRI_FIRST);
319 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
322 socket_vnet_init(const void *unused __unused)
326 /* We expect a contiguous range */
327 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
328 socket_hhook_register(i);
330 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
331 socket_vnet_init, NULL);
334 socket_vnet_uninit(const void *unused __unused)
338 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
339 socket_hhook_deregister(i);
341 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
342 socket_vnet_uninit, NULL);
345 * Initialise maxsockets. This SYSINIT must be run after
349 init_maxsockets(void *ignored)
352 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
353 maxsockets = imax(maxsockets, maxfiles);
355 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
358 * Sysctl to get and set the maximum global sockets limit. Notify protocols
359 * of the change so that they can update their dependent limits as required.
362 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
364 int error, newmaxsockets;
366 newmaxsockets = maxsockets;
367 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
368 if (error == 0 && req->newptr) {
369 if (newmaxsockets > maxsockets &&
370 newmaxsockets <= maxfiles) {
371 maxsockets = newmaxsockets;
372 EVENTHANDLER_INVOKE(maxsockets_change);
378 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
379 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &maxsockets, 0,
380 sysctl_maxsockets, "IU",
381 "Maximum number of sockets available");
384 * Socket operation routines. These routines are called by the routines in
385 * sys_socket.c or from a system process, and implement the semantics of
386 * socket operations by switching out to the protocol specific routines.
390 * Get a socket structure from our zone, and initialize it. Note that it
391 * would probably be better to allocate socket and PCB at the same time, but
392 * I'm not convinced that all the protocols can be easily modified to do
395 * soalloc() returns a socket with a ref count of 0.
397 static struct socket *
398 soalloc(struct vnet *vnet)
402 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
406 if (mac_socket_init(so, M_NOWAIT) != 0) {
407 uma_zfree(socket_zone, so);
411 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
412 uma_zfree(socket_zone, so);
417 * The socket locking protocol allows to lock 2 sockets at a time,
418 * however, the first one must be a listening socket. WITNESS lacks
419 * a feature to change class of an existing lock, so we use DUPOK.
421 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
422 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
423 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
424 so->so_rcv.sb_sel = &so->so_rdsel;
425 so->so_snd.sb_sel = &so->so_wrsel;
426 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
427 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
428 TAILQ_INIT(&so->so_snd.sb_aiojobq);
429 TAILQ_INIT(&so->so_rcv.sb_aiojobq);
430 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
431 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
433 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
434 __func__, __LINE__, so));
437 /* We shouldn't need the so_global_mtx */
438 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
439 /* Do we need more comprehensive error returns? */
440 uma_zfree(socket_zone, so);
443 mtx_lock(&so_global_mtx);
444 so->so_gencnt = ++so_gencnt;
447 vnet->vnet_sockcnt++;
449 mtx_unlock(&so_global_mtx);
455 * Free the storage associated with a socket at the socket layer, tear down
456 * locks, labels, etc. All protocol state is assumed already to have been
457 * torn down (and possibly never set up) by the caller.
460 sodealloc(struct socket *so)
463 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
464 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
466 mtx_lock(&so_global_mtx);
467 so->so_gencnt = ++so_gencnt;
468 --numopensockets; /* Could be below, but faster here. */
470 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
471 __func__, __LINE__, so));
472 so->so_vnet->vnet_sockcnt--;
474 mtx_unlock(&so_global_mtx);
476 mac_socket_destroy(so);
478 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
481 khelp_destroy_osd(&so->osd);
482 if (SOLISTENING(so)) {
483 if (so->sol_accept_filter != NULL)
484 accept_filt_setopt(so, NULL);
486 if (so->so_rcv.sb_hiwat)
487 (void)chgsbsize(so->so_cred->cr_uidinfo,
488 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
489 if (so->so_snd.sb_hiwat)
490 (void)chgsbsize(so->so_cred->cr_uidinfo,
491 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
492 sx_destroy(&so->so_snd.sb_sx);
493 sx_destroy(&so->so_rcv.sb_sx);
494 SOCKBUF_LOCK_DESTROY(&so->so_snd);
495 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
497 mtx_destroy(&so->so_lock);
498 uma_zfree(socket_zone, so);
502 * socreate returns a socket with a ref count of 1. The socket should be
503 * closed with soclose().
506 socreate(int dom, struct socket **aso, int type, int proto,
507 struct ucred *cred, struct thread *td)
514 prp = pffindproto(dom, proto, type);
516 prp = pffindtype(dom, type);
519 /* No support for domain. */
520 if (pffinddomain(dom) == NULL)
521 return (EAFNOSUPPORT);
522 /* No support for socket type. */
523 if (proto == 0 && type != 0)
525 return (EPROTONOSUPPORT);
527 if (prp->pr_usrreqs->pru_attach == NULL ||
528 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
529 return (EPROTONOSUPPORT);
531 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
532 return (EPROTONOSUPPORT);
534 if (prp->pr_type != type)
536 so = soalloc(CRED_TO_VNET(cred));
541 so->so_cred = crhold(cred);
542 if ((prp->pr_domain->dom_family == PF_INET) ||
543 (prp->pr_domain->dom_family == PF_INET6) ||
544 (prp->pr_domain->dom_family == PF_ROUTE))
545 so->so_fibnum = td->td_proc->p_fibnum;
550 mac_socket_create(cred, so);
552 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
553 so_rdknl_assert_locked, so_rdknl_assert_unlocked);
554 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
555 so_wrknl_assert_locked, so_wrknl_assert_unlocked);
557 * Auto-sizing of socket buffers is managed by the protocols and
558 * the appropriate flags must be set in the pru_attach function.
560 CURVNET_SET(so->so_vnet);
561 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
573 static int regression_sonewconn_earlytest = 1;
574 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
575 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
578 static struct timeval overinterval = { 60, 0 };
579 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
581 "Delay in seconds between warnings for listen socket overflows");
584 * When an attempt at a new connection is noted on a socket which accepts
585 * connections, sonewconn is called. If the connection is possible (subject
586 * to space constraints, etc.) then we allocate a new structure, properly
587 * linked into the data structure of the original socket, and return this.
588 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
590 * Note: the ref count on the socket is 0 on return.
593 sonewconn(struct socket *head, int connstatus)
599 const char localprefix[] = "local:";
600 char descrbuf[SUNPATHLEN + sizeof(localprefix)];
602 char addrbuf[INET6_ADDRSTRLEN];
604 char addrbuf[INET_ADDRSTRLEN];
609 over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
611 if (regression_sonewconn_earlytest && over) {
615 head->sol_overcount++;
616 dolog = !!ratecheck(&head->sol_lastover, &overinterval);
619 * If we're going to log, copy the overflow count and queue
620 * length from the listen socket before dropping the lock.
621 * Also, reset the overflow count.
624 overcount = head->sol_overcount;
625 head->sol_overcount = 0;
626 qlen = head->sol_qlen;
628 SOLISTEN_UNLOCK(head);
632 * Try to print something descriptive about the
633 * socket for the error message.
635 sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
637 switch (head->so_proto->pr_domain->dom_family) {
638 #if defined(INET) || defined(INET6)
644 if (head->so_proto->pr_domain->dom_family ==
646 (sotoinpcb(head)->inp_inc.inc_flags &
649 &sotoinpcb(head)->inp_inc.inc6_laddr);
650 sbuf_printf(&descrsb, "[%s]", addrbuf);
656 sotoinpcb(head)->inp_inc.inc_laddr,
658 sbuf_cat(&descrsb, addrbuf);
661 sbuf_printf(&descrsb, ":%hu (proto %u)",
662 ntohs(sotoinpcb(head)->inp_inc.inc_lport),
663 head->so_proto->pr_protocol);
665 #endif /* INET || INET6 */
667 sbuf_cat(&descrsb, localprefix);
668 if (sotounpcb(head)->unp_addr != NULL)
670 sotounpcb(head)->unp_addr->sun_len -
671 offsetof(struct sockaddr_un,
677 sotounpcb(head)->unp_addr->sun_path,
680 sbuf_cat(&descrsb, "(unknown)");
685 * If we can't print something more specific, at least
686 * print the domain name.
688 if (sbuf_finish(&descrsb) != 0 ||
689 sbuf_len(&descrsb) <= 0) {
690 sbuf_clear(&descrsb);
692 head->so_proto->pr_domain->dom_name ?:
694 sbuf_finish(&descrsb);
696 KASSERT(sbuf_len(&descrsb) > 0,
697 ("%s: sbuf creation failed", __func__));
699 "%s: pcb %p (%s): Listen queue overflow: "
700 "%i already in queue awaiting acceptance "
701 "(%d occurrences)\n",
702 __func__, head->so_pcb, sbuf_data(&descrsb),
704 sbuf_delete(&descrsb);
711 SOLISTEN_UNLOCK(head);
712 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
714 so = soalloc(head->so_vnet);
716 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
717 "limit reached or out of memory\n",
718 __func__, head->so_pcb);
721 so->so_listen = head;
722 so->so_type = head->so_type;
723 so->so_linger = head->so_linger;
724 so->so_state = head->so_state | SS_NOFDREF;
725 so->so_fibnum = head->so_fibnum;
726 so->so_proto = head->so_proto;
727 so->so_cred = crhold(head->so_cred);
729 mac_socket_newconn(head, so);
731 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
732 so_rdknl_assert_locked, so_rdknl_assert_unlocked);
733 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
734 so_wrknl_assert_locked, so_wrknl_assert_unlocked);
735 VNET_SO_ASSERT(head);
736 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
738 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
739 __func__, head->so_pcb);
742 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
744 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
745 __func__, head->so_pcb);
748 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
749 so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
750 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
751 so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
752 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
753 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
756 if (head->sol_accept_filter != NULL)
758 so->so_state |= connstatus;
759 so->so_options = head->so_options & ~SO_ACCEPTCONN;
760 soref(head); /* A socket on (in)complete queue refs head. */
762 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
763 so->so_qstate = SQ_COMP;
765 solisten_wakeup(head); /* unlocks */
768 * Keep removing sockets from the head until there's room for
769 * us to insert on the tail. In pre-locking revisions, this
770 * was a simple if(), but as we could be racing with other
771 * threads and soabort() requires dropping locks, we must
772 * loop waiting for the condition to be true.
774 while (head->sol_incqlen > head->sol_qlimit) {
777 sp = TAILQ_FIRST(&head->sol_incomp);
778 TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
781 sp->so_qstate = SQ_NONE;
782 sp->so_listen = NULL;
784 sorele(head); /* does SOLISTEN_UNLOCK, head stays */
788 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
789 so->so_qstate = SQ_INCOMP;
791 SOLISTEN_UNLOCK(head);
798 * Socket part of sctp_peeloff(). Detach a new socket from an
799 * association. The new socket is returned with a reference.
802 sopeeloff(struct socket *head)
806 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
807 __func__, __LINE__, head));
808 so = soalloc(head->so_vnet);
810 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
811 "limit reached or out of memory\n",
812 __func__, head->so_pcb);
815 so->so_type = head->so_type;
816 so->so_options = head->so_options;
817 so->so_linger = head->so_linger;
818 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
819 so->so_fibnum = head->so_fibnum;
820 so->so_proto = head->so_proto;
821 so->so_cred = crhold(head->so_cred);
823 mac_socket_newconn(head, so);
825 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
826 so_rdknl_assert_locked, so_rdknl_assert_unlocked);
827 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
828 so_wrknl_assert_locked, so_wrknl_assert_unlocked);
829 VNET_SO_ASSERT(head);
830 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
832 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
833 __func__, head->so_pcb);
836 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
838 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
839 __func__, head->so_pcb);
842 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
843 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
844 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
845 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
846 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
847 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
856 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
860 CURVNET_SET(so->so_vnet);
861 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
867 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
871 CURVNET_SET(so->so_vnet);
872 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
878 * solisten() transitions a socket from a non-listening state to a listening
879 * state, but can also be used to update the listen queue depth on an
880 * existing listen socket. The protocol will call back into the sockets
881 * layer using solisten_proto_check() and solisten_proto() to check and set
882 * socket-layer listen state. Call backs are used so that the protocol can
883 * acquire both protocol and socket layer locks in whatever order is required
886 * Protocol implementors are advised to hold the socket lock across the
887 * socket-layer test and set to avoid races at the socket layer.
890 solisten(struct socket *so, int backlog, struct thread *td)
894 CURVNET_SET(so->so_vnet);
895 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
901 solisten_proto_check(struct socket *so)
904 SOCK_LOCK_ASSERT(so);
906 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
913 solisten_proto(struct socket *so, int backlog)
915 int sbrcv_lowat, sbsnd_lowat;
916 u_int sbrcv_hiwat, sbsnd_hiwat;
917 short sbrcv_flags, sbsnd_flags;
918 sbintime_t sbrcv_timeo, sbsnd_timeo;
920 SOCK_LOCK_ASSERT(so);
926 * Change this socket to listening state.
928 sbrcv_lowat = so->so_rcv.sb_lowat;
929 sbsnd_lowat = so->so_snd.sb_lowat;
930 sbrcv_hiwat = so->so_rcv.sb_hiwat;
931 sbsnd_hiwat = so->so_snd.sb_hiwat;
932 sbrcv_flags = so->so_rcv.sb_flags;
933 sbsnd_flags = so->so_snd.sb_flags;
934 sbrcv_timeo = so->so_rcv.sb_timeo;
935 sbsnd_timeo = so->so_snd.sb_timeo;
937 sbdestroy(&so->so_snd, so);
938 sbdestroy(&so->so_rcv, so);
939 sx_destroy(&so->so_snd.sb_sx);
940 sx_destroy(&so->so_rcv.sb_sx);
941 SOCKBUF_LOCK_DESTROY(&so->so_snd);
942 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
946 sizeof(struct socket) - offsetof(struct socket, so_rcv));
949 so->sol_sbrcv_lowat = sbrcv_lowat;
950 so->sol_sbsnd_lowat = sbsnd_lowat;
951 so->sol_sbrcv_hiwat = sbrcv_hiwat;
952 so->sol_sbsnd_hiwat = sbsnd_hiwat;
953 so->sol_sbrcv_flags = sbrcv_flags;
954 so->sol_sbsnd_flags = sbsnd_flags;
955 so->sol_sbrcv_timeo = sbrcv_timeo;
956 so->sol_sbsnd_timeo = sbsnd_timeo;
958 so->sol_qlen = so->sol_incqlen = 0;
959 TAILQ_INIT(&so->sol_incomp);
960 TAILQ_INIT(&so->sol_comp);
962 so->sol_accept_filter = NULL;
963 so->sol_accept_filter_arg = NULL;
964 so->sol_accept_filter_str = NULL;
966 so->sol_upcall = NULL;
967 so->sol_upcallarg = NULL;
969 so->so_options |= SO_ACCEPTCONN;
972 if (backlog < 0 || backlog > somaxconn)
974 so->sol_qlimit = backlog;
978 * Wakeup listeners/subsystems once we have a complete connection.
979 * Enters with lock, returns unlocked.
982 solisten_wakeup(struct socket *sol)
985 if (sol->sol_upcall != NULL)
986 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
988 selwakeuppri(&sol->so_rdsel, PSOCK);
989 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
991 SOLISTEN_UNLOCK(sol);
992 wakeup_one(&sol->sol_comp);
993 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
994 pgsigio(&sol->so_sigio, SIGIO, 0);
998 * Return single connection off a listening socket queue. Main consumer of
999 * the function is kern_accept4(). Some modules, that do their own accept
1000 * management also use the function.
1002 * Listening socket must be locked on entry and is returned unlocked on
1004 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1007 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1012 SOLISTEN_LOCK_ASSERT(head);
1014 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1015 head->so_error == 0) {
1016 error = msleep(&head->sol_comp, &head->so_lock, PSOCK | PCATCH,
1019 SOLISTEN_UNLOCK(head);
1023 if (head->so_error) {
1024 error = head->so_error;
1026 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1027 error = EWOULDBLOCK;
1031 SOLISTEN_UNLOCK(head);
1034 so = TAILQ_FIRST(&head->sol_comp);
1036 KASSERT(so->so_qstate == SQ_COMP,
1037 ("%s: so %p not SQ_COMP", __func__, so));
1040 so->so_qstate = SQ_NONE;
1041 so->so_listen = NULL;
1042 TAILQ_REMOVE(&head->sol_comp, so, so_list);
1043 if (flags & ACCEPT4_INHERIT)
1044 so->so_state |= (head->so_state & SS_NBIO);
1046 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1055 * Evaluate the reference count and named references on a socket; if no
1056 * references remain, free it. This should be called whenever a reference is
1057 * released, such as in sorele(), but also when named reference flags are
1058 * cleared in socket or protocol code.
1060 * sofree() will free the socket if:
1062 * - There are no outstanding file descriptor references or related consumers
1065 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
1067 * - The protocol does not have an outstanding strong reference on the socket
1070 * - The socket is not in a completed connection queue, so a process has been
1071 * notified that it is present. If it is removed, the user process may
1072 * block in accept() despite select() saying the socket was ready.
1075 sofree(struct socket *so)
1077 struct protosw *pr = so->so_proto;
1079 SOCK_LOCK_ASSERT(so);
1081 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
1082 (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) {
1087 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
1090 sol = so->so_listen;
1091 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
1094 * To solve race between close of a listening socket and
1095 * a socket on its incomplete queue, we need to lock both.
1096 * The order is first listening socket, then regular.
1097 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
1098 * function and the listening socket are the only pointers
1099 * to so. To preserve so and sol, we reference both and then
1101 * After relock the socket may not move to so_comp since it
1102 * doesn't have PCB already, but it may be removed from
1103 * so_incomp. If that happens, we share responsiblity on
1104 * freeing the socket, but soclose() has already removed
1112 if (so->so_qstate == SQ_INCOMP) {
1113 KASSERT(so->so_listen == sol,
1114 ("%s: so %p migrated out of sol %p",
1115 __func__, so, sol));
1116 TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1118 /* This is guarenteed not to be the last. */
1119 refcount_release(&sol->so_count);
1120 so->so_qstate = SQ_NONE;
1121 so->so_listen = NULL;
1123 KASSERT(so->so_listen == NULL,
1124 ("%s: so %p not on (in)comp with so_listen",
1127 KASSERT(so->so_count == 1,
1128 ("%s: so %p count %u", __func__, so, so->so_count));
1131 if (SOLISTENING(so))
1132 so->so_error = ECONNABORTED;
1135 if (so->so_dtor != NULL)
1139 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1140 (*pr->pr_domain->dom_dispose)(so);
1141 if (pr->pr_usrreqs->pru_detach != NULL)
1142 (*pr->pr_usrreqs->pru_detach)(so);
1145 * From this point on, we assume that no other references to this
1146 * socket exist anywhere else in the stack. Therefore, no locks need
1147 * to be acquired or held.
1149 * We used to do a lot of socket buffer and socket locking here, as
1150 * well as invoke sorflush() and perform wakeups. The direct call to
1151 * dom_dispose() and sbdestroy() are an inlining of what was
1152 * necessary from sorflush().
1154 * Notice that the socket buffer and kqueue state are torn down
1155 * before calling pru_detach. This means that protocols shold not
1156 * assume they can perform socket wakeups, etc, in their detach code.
1158 if (!SOLISTENING(so)) {
1159 sbdestroy(&so->so_snd, so);
1160 sbdestroy(&so->so_rcv, so);
1162 seldrain(&so->so_rdsel);
1163 seldrain(&so->so_wrsel);
1164 knlist_destroy(&so->so_rdsel.si_note);
1165 knlist_destroy(&so->so_wrsel.si_note);
1170 * Close a socket on last file table reference removal. Initiate disconnect
1171 * if connected. Free socket when disconnect complete.
1173 * This function will sorele() the socket. Note that soclose() may be called
1174 * prior to the ref count reaching zero. The actual socket structure will
1175 * not be freed until the ref count reaches zero.
1178 soclose(struct socket *so)
1180 struct accept_queue lqueue;
1184 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1186 CURVNET_SET(so->so_vnet);
1187 funsetown(&so->so_sigio);
1188 if (so->so_state & SS_ISCONNECTED) {
1189 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1190 error = sodisconnect(so);
1192 if (error == ENOTCONN)
1197 if (so->so_options & SO_LINGER) {
1198 if ((so->so_state & SS_ISDISCONNECTING) &&
1199 (so->so_state & SS_NBIO))
1201 while (so->so_state & SS_ISCONNECTED) {
1202 error = tsleep(&so->so_timeo,
1203 PSOCK | PCATCH, "soclos",
1204 so->so_linger * hz);
1212 if (so->so_proto->pr_usrreqs->pru_close != NULL)
1213 (*so->so_proto->pr_usrreqs->pru_close)(so);
1216 if ((listening = (so->so_options & SO_ACCEPTCONN))) {
1219 TAILQ_INIT(&lqueue);
1220 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1221 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1223 so->sol_qlen = so->sol_incqlen = 0;
1225 TAILQ_FOREACH(sp, &lqueue, so_list) {
1227 sp->so_qstate = SQ_NONE;
1228 sp->so_listen = NULL;
1230 /* Guaranteed not to be the last. */
1231 refcount_release(&so->so_count);
1234 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1235 so->so_state |= SS_NOFDREF;
1238 struct socket *sp, *tsp;
1240 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1242 if (sp->so_count == 0) {
1246 /* sp is now in sofree() */
1255 * soabort() is used to abruptly tear down a connection, such as when a
1256 * resource limit is reached (listen queue depth exceeded), or if a listen
1257 * socket is closed while there are sockets waiting to be accepted.
1259 * This interface is tricky, because it is called on an unreferenced socket,
1260 * and must be called only by a thread that has actually removed the socket
1261 * from the listen queue it was on, or races with other threads are risked.
1263 * This interface will call into the protocol code, so must not be called
1264 * with any socket locks held. Protocols do call it while holding their own
1265 * recursible protocol mutexes, but this is something that should be subject
1266 * to review in the future.
1269 soabort(struct socket *so)
1273 * In as much as is possible, assert that no references to this
1274 * socket are held. This is not quite the same as asserting that the
1275 * current thread is responsible for arranging for no references, but
1276 * is as close as we can get for now.
1278 KASSERT(so->so_count == 0, ("soabort: so_count"));
1279 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1280 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1283 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1284 (*so->so_proto->pr_usrreqs->pru_abort)(so);
1290 soaccept(struct socket *so, struct sockaddr **nam)
1295 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1296 so->so_state &= ~SS_NOFDREF;
1299 CURVNET_SET(so->so_vnet);
1300 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1306 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1309 return (soconnectat(AT_FDCWD, so, nam, td));
1313 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1317 if (so->so_options & SO_ACCEPTCONN)
1318 return (EOPNOTSUPP);
1320 CURVNET_SET(so->so_vnet);
1322 * If protocol is connection-based, can only connect once.
1323 * Otherwise, if connected, try to disconnect first. This allows
1324 * user to disconnect by connecting to, e.g., a null address.
1326 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1327 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1328 (error = sodisconnect(so)))) {
1332 * Prevent accumulated error from previous connection from
1336 if (fd == AT_FDCWD) {
1337 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1340 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1350 soconnect2(struct socket *so1, struct socket *so2)
1354 CURVNET_SET(so1->so_vnet);
1355 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1361 sodisconnect(struct socket *so)
1365 if ((so->so_state & SS_ISCONNECTED) == 0)
1367 if (so->so_state & SS_ISDISCONNECTING)
1370 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1374 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1377 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1378 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1382 int clen = 0, error, dontroute;
1384 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1385 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1386 ("sosend_dgram: !PR_ATOMIC"));
1389 resid = uio->uio_resid;
1391 resid = top->m_pkthdr.len;
1393 * In theory resid should be unsigned. However, space must be
1394 * signed, as it might be less than 0 if we over-committed, and we
1395 * must use a signed comparison of space and resid. On the other
1396 * hand, a negative resid causes us to loop sending 0-length
1397 * segments to the protocol.
1405 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1407 td->td_ru.ru_msgsnd++;
1408 if (control != NULL)
1409 clen = control->m_len;
1411 SOCKBUF_LOCK(&so->so_snd);
1412 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1413 SOCKBUF_UNLOCK(&so->so_snd);
1418 error = so->so_error;
1420 SOCKBUF_UNLOCK(&so->so_snd);
1423 if ((so->so_state & SS_ISCONNECTED) == 0) {
1425 * `sendto' and `sendmsg' is allowed on a connection-based
1426 * socket if it supports implied connect. Return ENOTCONN if
1427 * not connected and no address is supplied.
1429 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1430 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1431 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1432 !(resid == 0 && clen != 0)) {
1433 SOCKBUF_UNLOCK(&so->so_snd);
1437 } else if (addr == NULL) {
1438 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1441 error = EDESTADDRREQ;
1442 SOCKBUF_UNLOCK(&so->so_snd);
1448 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1449 * problem and need fixing.
1451 space = sbspace(&so->so_snd);
1452 if (flags & MSG_OOB)
1455 SOCKBUF_UNLOCK(&so->so_snd);
1456 if (resid > space) {
1462 if (flags & MSG_EOR)
1463 top->m_flags |= M_EOR;
1466 * Copy the data from userland into a mbuf chain.
1467 * If no data is to be copied in, a single empty mbuf
1470 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1471 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1473 error = EFAULT; /* only possible error */
1476 space -= resid - uio->uio_resid;
1477 resid = uio->uio_resid;
1479 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1481 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1486 so->so_options |= SO_DONTROUTE;
1490 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1491 * of date. We could have received a reset packet in an interrupt or
1492 * maybe we slept while doing page faults in uiomove() etc. We could
1493 * probably recheck again inside the locking protection here, but
1494 * there are probably other places that this also happens. We must
1498 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1499 (flags & MSG_OOB) ? PRUS_OOB :
1501 * If the user set MSG_EOF, the protocol understands this flag and
1502 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1504 ((flags & MSG_EOF) &&
1505 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1508 /* If there is more to send set PRUS_MORETOCOME */
1509 (flags & MSG_MORETOCOME) ||
1510 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1511 top, addr, control, td);
1514 so->so_options &= ~SO_DONTROUTE;
1523 if (control != NULL)
1529 * Send on a socket. If send must go all at once and message is larger than
1530 * send buffering, then hard error. Lock against other senders. If must go
1531 * all at once and not enough room now, then inform user that this would
1532 * block and do nothing. Otherwise, if nonblocking, send as much as
1533 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1534 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1535 * in mbuf chain must be small enough to send all at once.
1537 * Returns nonzero on error, timeout or signal; callers must check for short
1538 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1542 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1543 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1547 int clen = 0, error, dontroute;
1548 int atomic = sosendallatonce(so) || top;
1551 struct ktls_session *tls;
1552 int tls_enq_cnt, tls_pruflag;
1556 tls_rtype = TLS_RLTYPE_APP;
1559 resid = uio->uio_resid;
1560 else if ((top->m_flags & M_PKTHDR) != 0)
1561 resid = top->m_pkthdr.len;
1563 resid = m_length(top, NULL);
1565 * In theory resid should be unsigned. However, space must be
1566 * signed, as it might be less than 0 if we over-committed, and we
1567 * must use a signed comparison of space and resid. On the other
1568 * hand, a negative resid causes us to loop sending 0-length
1569 * segments to the protocol.
1571 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1572 * type sockets since that's an error.
1574 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1580 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1581 (so->so_proto->pr_flags & PR_ATOMIC);
1583 td->td_ru.ru_msgsnd++;
1584 if (control != NULL)
1585 clen = control->m_len;
1587 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1593 tls = ktls_hold(so->so_snd.sb_tls_info);
1595 if (tls->mode == TCP_TLS_MODE_SW)
1596 tls_pruflag = PRUS_NOTREADY;
1598 if (control != NULL) {
1599 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1601 if (clen >= sizeof(*cm) &&
1602 cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1603 tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1615 SOCKBUF_LOCK(&so->so_snd);
1616 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1617 SOCKBUF_UNLOCK(&so->so_snd);
1622 error = so->so_error;
1624 SOCKBUF_UNLOCK(&so->so_snd);
1627 if ((so->so_state & SS_ISCONNECTED) == 0) {
1629 * `sendto' and `sendmsg' is allowed on a connection-
1630 * based socket if it supports implied connect.
1631 * Return ENOTCONN if not connected and no address is
1634 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1635 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1636 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1637 !(resid == 0 && clen != 0)) {
1638 SOCKBUF_UNLOCK(&so->so_snd);
1642 } else if (addr == NULL) {
1643 SOCKBUF_UNLOCK(&so->so_snd);
1644 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1647 error = EDESTADDRREQ;
1651 space = sbspace(&so->so_snd);
1652 if (flags & MSG_OOB)
1654 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1655 clen > so->so_snd.sb_hiwat) {
1656 SOCKBUF_UNLOCK(&so->so_snd);
1660 if (space < resid + clen &&
1661 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1662 if ((so->so_state & SS_NBIO) ||
1663 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1664 SOCKBUF_UNLOCK(&so->so_snd);
1665 error = EWOULDBLOCK;
1668 error = sbwait(&so->so_snd);
1669 SOCKBUF_UNLOCK(&so->so_snd);
1674 SOCKBUF_UNLOCK(&so->so_snd);
1679 if (flags & MSG_EOR)
1680 top->m_flags |= M_EOR;
1683 * Copy the data from userland into a mbuf
1684 * chain. If resid is 0, which can happen
1685 * only if we have control to send, then
1686 * a single empty mbuf is returned. This
1687 * is a workaround to prevent protocol send
1692 top = m_uiotombuf(uio, M_WAITOK, space,
1693 tls->params.max_frame_len,
1695 ((flags & MSG_EOR) ? M_EOR : 0));
1697 ktls_frame(top, tls,
1698 &tls_enq_cnt, tls_rtype);
1700 tls_rtype = TLS_RLTYPE_APP;
1703 top = m_uiotombuf(uio, M_WAITOK, space,
1704 (atomic ? max_hdr : 0),
1705 (atomic ? M_PKTHDR : 0) |
1706 ((flags & MSG_EOR) ? M_EOR : 0));
1708 error = EFAULT; /* only possible error */
1711 space -= resid - uio->uio_resid;
1712 resid = uio->uio_resid;
1716 so->so_options |= SO_DONTROUTE;
1720 * XXX all the SBS_CANTSENDMORE checks previously
1721 * done could be out of date. We could have received
1722 * a reset packet in an interrupt or maybe we slept
1723 * while doing page faults in uiomove() etc. We
1724 * could probably recheck again inside the locking
1725 * protection here, but there are probably other
1726 * places that this also happens. We must rethink
1731 pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1733 * If the user set MSG_EOF, the protocol understands
1734 * this flag and nothing left to send then use
1735 * PRU_SEND_EOF instead of PRU_SEND.
1737 ((flags & MSG_EOF) &&
1738 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1741 /* If there is more to send set PRUS_MORETOCOME. */
1742 (flags & MSG_MORETOCOME) ||
1743 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1746 pru_flag |= tls_pruflag;
1749 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1750 pru_flag, top, addr, control, td);
1754 so->so_options &= ~SO_DONTROUTE;
1759 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1761 * Note that error is intentionally
1764 * Like sendfile(), we rely on the
1765 * completion routine (pru_ready())
1766 * to free the mbufs in the event that
1767 * pru_send() encountered an error and
1768 * did not append them to the sockbuf.
1771 ktls_enqueue(top, so, tls_enq_cnt);
1779 } while (resid && space > 0);
1783 sbunlock(&so->so_snd);
1791 if (control != NULL)
1797 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1798 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1802 CURVNET_SET(so->so_vnet);
1803 if (!SOLISTENING(so))
1804 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1805 top, control, flags, td);
1816 * The part of soreceive() that implements reading non-inline out-of-band
1817 * data from a socket. For more complete comments, see soreceive(), from
1818 * which this code originated.
1820 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1821 * unable to return an mbuf chain to the caller.
1824 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1826 struct protosw *pr = so->so_proto;
1830 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1833 m = m_get(M_WAITOK, MT_DATA);
1834 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1838 error = uiomove(mtod(m, void *),
1839 (int) min(uio->uio_resid, m->m_len), uio);
1841 } while (uio->uio_resid && error == 0 && m);
1849 * Following replacement or removal of the first mbuf on the first mbuf chain
1850 * of a socket buffer, push necessary state changes back into the socket
1851 * buffer so that other consumers see the values consistently. 'nextrecord'
1852 * is the callers locally stored value of the original value of
1853 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1854 * NOTE: 'nextrecord' may be NULL.
1856 static __inline void
1857 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1860 SOCKBUF_LOCK_ASSERT(sb);
1862 * First, update for the new value of nextrecord. If necessary, make
1863 * it the first record.
1865 if (sb->sb_mb != NULL)
1866 sb->sb_mb->m_nextpkt = nextrecord;
1868 sb->sb_mb = nextrecord;
1871 * Now update any dependent socket buffer fields to reflect the new
1872 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1873 * addition of a second clause that takes care of the case where
1874 * sb_mb has been updated, but remains the last record.
1876 if (sb->sb_mb == NULL) {
1877 sb->sb_mbtail = NULL;
1878 sb->sb_lastrecord = NULL;
1879 } else if (sb->sb_mb->m_nextpkt == NULL)
1880 sb->sb_lastrecord = sb->sb_mb;
1884 * Implement receive operations on a socket. We depend on the way that
1885 * records are added to the sockbuf by sbappend. In particular, each record
1886 * (mbufs linked through m_next) must begin with an address if the protocol
1887 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1888 * data, and then zero or more mbufs of data. In order to allow parallelism
1889 * between network receive and copying to user space, as well as avoid
1890 * sleeping with a mutex held, we release the socket buffer mutex during the
1891 * user space copy. Although the sockbuf is locked, new data may still be
1892 * appended, and thus we must maintain consistency of the sockbuf during that
1895 * The caller may receive the data as a single mbuf chain by supplying an
1896 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1897 * the count in uio_resid.
1900 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1901 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1903 struct mbuf *m, **mp;
1904 int flags, error, offset;
1906 struct protosw *pr = so->so_proto;
1907 struct mbuf *nextrecord;
1909 ssize_t orig_resid = uio->uio_resid;
1914 if (controlp != NULL)
1917 flags = *flagsp &~ MSG_EOR;
1920 if (flags & MSG_OOB)
1921 return (soreceive_rcvoob(so, uio, flags));
1924 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1925 && uio->uio_resid) {
1927 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1930 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1935 SOCKBUF_LOCK(&so->so_rcv);
1936 m = so->so_rcv.sb_mb;
1938 * If we have less data than requested, block awaiting more (subject
1939 * to any timeout) if:
1940 * 1. the current count is less than the low water mark, or
1941 * 2. MSG_DONTWAIT is not set
1943 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1944 sbavail(&so->so_rcv) < uio->uio_resid) &&
1945 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1946 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1947 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1948 ("receive: m == %p sbavail == %u",
1949 m, sbavail(&so->so_rcv)));
1953 error = so->so_error;
1954 if ((flags & MSG_PEEK) == 0)
1956 SOCKBUF_UNLOCK(&so->so_rcv);
1959 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1960 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1962 SOCKBUF_UNLOCK(&so->so_rcv);
1967 for (; m != NULL; m = m->m_next)
1968 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1969 m = so->so_rcv.sb_mb;
1972 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1973 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1974 SOCKBUF_UNLOCK(&so->so_rcv);
1978 if (uio->uio_resid == 0) {
1979 SOCKBUF_UNLOCK(&so->so_rcv);
1982 if ((so->so_state & SS_NBIO) ||
1983 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1984 SOCKBUF_UNLOCK(&so->so_rcv);
1985 error = EWOULDBLOCK;
1988 SBLASTRECORDCHK(&so->so_rcv);
1989 SBLASTMBUFCHK(&so->so_rcv);
1990 error = sbwait(&so->so_rcv);
1991 SOCKBUF_UNLOCK(&so->so_rcv);
1998 * From this point onward, we maintain 'nextrecord' as a cache of the
1999 * pointer to the next record in the socket buffer. We must keep the
2000 * various socket buffer pointers and local stack versions of the
2001 * pointers in sync, pushing out modifications before dropping the
2002 * socket buffer mutex, and re-reading them when picking it up.
2004 * Otherwise, we will race with the network stack appending new data
2005 * or records onto the socket buffer by using inconsistent/stale
2006 * versions of the field, possibly resulting in socket buffer
2009 * By holding the high-level sblock(), we prevent simultaneous
2010 * readers from pulling off the front of the socket buffer.
2012 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2014 uio->uio_td->td_ru.ru_msgrcv++;
2015 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2016 SBLASTRECORDCHK(&so->so_rcv);
2017 SBLASTMBUFCHK(&so->so_rcv);
2018 nextrecord = m->m_nextpkt;
2019 if (pr->pr_flags & PR_ADDR) {
2020 KASSERT(m->m_type == MT_SONAME,
2021 ("m->m_type == %d", m->m_type));
2024 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2026 if (flags & MSG_PEEK) {
2029 sbfree(&so->so_rcv, m);
2030 so->so_rcv.sb_mb = m_free(m);
2031 m = so->so_rcv.sb_mb;
2032 sockbuf_pushsync(&so->so_rcv, nextrecord);
2037 * Process one or more MT_CONTROL mbufs present before any data mbufs
2038 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
2039 * just copy the data; if !MSG_PEEK, we call into the protocol to
2040 * perform externalization (or freeing if controlp == NULL).
2042 if (m != NULL && m->m_type == MT_CONTROL) {
2043 struct mbuf *cm = NULL, *cmn;
2044 struct mbuf **cme = &cm;
2047 if (flags & MSG_PEEK) {
2048 if (controlp != NULL) {
2049 *controlp = m_copym(m, 0, m->m_len,
2051 controlp = &(*controlp)->m_next;
2055 sbfree(&so->so_rcv, m);
2056 so->so_rcv.sb_mb = m->m_next;
2059 cme = &(*cme)->m_next;
2060 m = so->so_rcv.sb_mb;
2062 } while (m != NULL && m->m_type == MT_CONTROL);
2063 if ((flags & MSG_PEEK) == 0)
2064 sockbuf_pushsync(&so->so_rcv, nextrecord);
2065 while (cm != NULL) {
2068 if (pr->pr_domain->dom_externalize != NULL) {
2069 SOCKBUF_UNLOCK(&so->so_rcv);
2071 error = (*pr->pr_domain->dom_externalize)
2072 (cm, controlp, flags);
2073 SOCKBUF_LOCK(&so->so_rcv);
2074 } else if (controlp != NULL)
2078 if (controlp != NULL) {
2080 while (*controlp != NULL)
2081 controlp = &(*controlp)->m_next;
2086 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2088 nextrecord = so->so_rcv.sb_mb;
2092 if ((flags & MSG_PEEK) == 0) {
2093 KASSERT(m->m_nextpkt == nextrecord,
2094 ("soreceive: post-control, nextrecord !sync"));
2095 if (nextrecord == NULL) {
2096 KASSERT(so->so_rcv.sb_mb == m,
2097 ("soreceive: post-control, sb_mb!=m"));
2098 KASSERT(so->so_rcv.sb_lastrecord == m,
2099 ("soreceive: post-control, lastrecord!=m"));
2103 if (type == MT_OOBDATA)
2106 if ((flags & MSG_PEEK) == 0) {
2107 KASSERT(so->so_rcv.sb_mb == nextrecord,
2108 ("soreceive: sb_mb != nextrecord"));
2109 if (so->so_rcv.sb_mb == NULL) {
2110 KASSERT(so->so_rcv.sb_lastrecord == NULL,
2111 ("soreceive: sb_lastercord != NULL"));
2115 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2116 SBLASTRECORDCHK(&so->so_rcv);
2117 SBLASTMBUFCHK(&so->so_rcv);
2120 * Now continue to read any data mbufs off of the head of the socket
2121 * buffer until the read request is satisfied. Note that 'type' is
2122 * used to store the type of any mbuf reads that have happened so far
2123 * such that soreceive() can stop reading if the type changes, which
2124 * causes soreceive() to return only one of regular data and inline
2125 * out-of-band data in a single socket receive operation.
2129 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2132 * If the type of mbuf has changed since the last mbuf
2133 * examined ('type'), end the receive operation.
2135 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2136 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2137 if (type != m->m_type)
2139 } else if (type == MT_OOBDATA)
2142 KASSERT(m->m_type == MT_DATA,
2143 ("m->m_type == %d", m->m_type));
2144 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2145 len = uio->uio_resid;
2146 if (so->so_oobmark && len > so->so_oobmark - offset)
2147 len = so->so_oobmark - offset;
2148 if (len > m->m_len - moff)
2149 len = m->m_len - moff;
2151 * If mp is set, just pass back the mbufs. Otherwise copy
2152 * them out via the uio, then free. Sockbuf must be
2153 * consistent here (points to current mbuf, it points to next
2154 * record) when we drop priority; we must note any additions
2155 * to the sockbuf when we block interrupts again.
2158 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2159 SBLASTRECORDCHK(&so->so_rcv);
2160 SBLASTMBUFCHK(&so->so_rcv);
2161 SOCKBUF_UNLOCK(&so->so_rcv);
2162 if ((m->m_flags & M_NOMAP) != 0)
2163 error = m_unmappedtouio(m, moff, uio, (int)len);
2165 error = uiomove(mtod(m, char *) + moff,
2167 SOCKBUF_LOCK(&so->so_rcv);
2170 * The MT_SONAME mbuf has already been removed
2171 * from the record, so it is necessary to
2172 * remove the data mbufs, if any, to preserve
2173 * the invariant in the case of PR_ADDR that
2174 * requires MT_SONAME mbufs at the head of
2177 if (pr->pr_flags & PR_ATOMIC &&
2178 ((flags & MSG_PEEK) == 0))
2179 (void)sbdroprecord_locked(&so->so_rcv);
2180 SOCKBUF_UNLOCK(&so->so_rcv);
2184 uio->uio_resid -= len;
2185 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2186 if (len == m->m_len - moff) {
2187 if (m->m_flags & M_EOR)
2189 if (flags & MSG_PEEK) {
2193 nextrecord = m->m_nextpkt;
2194 sbfree(&so->so_rcv, m);
2196 m->m_nextpkt = NULL;
2199 so->so_rcv.sb_mb = m = m->m_next;
2202 so->so_rcv.sb_mb = m_free(m);
2203 m = so->so_rcv.sb_mb;
2205 sockbuf_pushsync(&so->so_rcv, nextrecord);
2206 SBLASTRECORDCHK(&so->so_rcv);
2207 SBLASTMBUFCHK(&so->so_rcv);
2210 if (flags & MSG_PEEK)
2214 if (flags & MSG_DONTWAIT) {
2215 *mp = m_copym(m, 0, len,
2219 * m_copym() couldn't
2221 * Adjust uio_resid back
2223 * down by len bytes,
2224 * which we didn't end
2225 * up "copying" over).
2227 uio->uio_resid += len;
2231 SOCKBUF_UNLOCK(&so->so_rcv);
2232 *mp = m_copym(m, 0, len,
2234 SOCKBUF_LOCK(&so->so_rcv);
2237 sbcut_locked(&so->so_rcv, len);
2240 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2241 if (so->so_oobmark) {
2242 if ((flags & MSG_PEEK) == 0) {
2243 so->so_oobmark -= len;
2244 if (so->so_oobmark == 0) {
2245 so->so_rcv.sb_state |= SBS_RCVATMARK;
2250 if (offset == so->so_oobmark)
2254 if (flags & MSG_EOR)
2257 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2258 * must not quit until "uio->uio_resid == 0" or an error
2259 * termination. If a signal/timeout occurs, return with a
2260 * short count but without error. Keep sockbuf locked
2261 * against other readers.
2263 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2264 !sosendallatonce(so) && nextrecord == NULL) {
2265 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2267 so->so_rcv.sb_state & SBS_CANTRCVMORE)
2270 * Notify the protocol that some data has been
2271 * drained before blocking.
2273 if (pr->pr_flags & PR_WANTRCVD) {
2274 SOCKBUF_UNLOCK(&so->so_rcv);
2276 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2277 SOCKBUF_LOCK(&so->so_rcv);
2279 SBLASTRECORDCHK(&so->so_rcv);
2280 SBLASTMBUFCHK(&so->so_rcv);
2282 * We could receive some data while was notifying
2283 * the protocol. Skip blocking in this case.
2285 if (so->so_rcv.sb_mb == NULL) {
2286 error = sbwait(&so->so_rcv);
2288 SOCKBUF_UNLOCK(&so->so_rcv);
2292 m = so->so_rcv.sb_mb;
2294 nextrecord = m->m_nextpkt;
2298 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2299 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2301 if ((flags & MSG_PEEK) == 0)
2302 (void) sbdroprecord_locked(&so->so_rcv);
2304 if ((flags & MSG_PEEK) == 0) {
2307 * First part is an inline SB_EMPTY_FIXUP(). Second
2308 * part makes sure sb_lastrecord is up-to-date if
2309 * there is still data in the socket buffer.
2311 so->so_rcv.sb_mb = nextrecord;
2312 if (so->so_rcv.sb_mb == NULL) {
2313 so->so_rcv.sb_mbtail = NULL;
2314 so->so_rcv.sb_lastrecord = NULL;
2315 } else if (nextrecord->m_nextpkt == NULL)
2316 so->so_rcv.sb_lastrecord = nextrecord;
2318 SBLASTRECORDCHK(&so->so_rcv);
2319 SBLASTMBUFCHK(&so->so_rcv);
2321 * If soreceive() is being done from the socket callback,
2322 * then don't need to generate ACK to peer to update window,
2323 * since ACK will be generated on return to TCP.
2325 if (!(flags & MSG_SOCALLBCK) &&
2326 (pr->pr_flags & PR_WANTRCVD)) {
2327 SOCKBUF_UNLOCK(&so->so_rcv);
2329 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2330 SOCKBUF_LOCK(&so->so_rcv);
2333 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2334 if (orig_resid == uio->uio_resid && orig_resid &&
2335 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2336 SOCKBUF_UNLOCK(&so->so_rcv);
2339 SOCKBUF_UNLOCK(&so->so_rcv);
2344 sbunlock(&so->so_rcv);
2349 * Optimized version of soreceive() for stream (TCP) sockets.
2352 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2353 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2355 int len = 0, error = 0, flags, oresid;
2357 struct mbuf *m, *n = NULL;
2359 /* We only do stream sockets. */
2360 if (so->so_type != SOCK_STREAM)
2365 flags = *flagsp &~ MSG_EOR;
2368 if (controlp != NULL)
2370 if (flags & MSG_OOB)
2371 return (soreceive_rcvoob(so, uio, flags));
2379 * KTLS store TLS records as records with a control message to
2380 * describe the framing.
2382 * We check once here before acquiring locks to optimize the
2385 if (sb->sb_tls_info != NULL)
2386 return (soreceive_generic(so, psa, uio, mp0, controlp,
2390 /* Prevent other readers from entering the socket. */
2391 error = sblock(sb, SBLOCKWAIT(flags));
2397 if (sb->sb_tls_info != NULL) {
2400 return (soreceive_generic(so, psa, uio, mp0, controlp,
2405 /* Easy one, no space to copyout anything. */
2406 if (uio->uio_resid == 0) {
2410 oresid = uio->uio_resid;
2412 /* We will never ever get anything unless we are or were connected. */
2413 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2419 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2421 /* Abort if socket has reported problems. */
2423 if (sbavail(sb) > 0)
2425 if (oresid > uio->uio_resid)
2427 error = so->so_error;
2428 if (!(flags & MSG_PEEK))
2433 /* Door is closed. Deliver what is left, if any. */
2434 if (sb->sb_state & SBS_CANTRCVMORE) {
2435 if (sbavail(sb) > 0)
2441 /* Socket buffer is empty and we shall not block. */
2442 if (sbavail(sb) == 0 &&
2443 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2448 /* Socket buffer got some data that we shall deliver now. */
2449 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2450 ((so->so_state & SS_NBIO) ||
2451 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2452 sbavail(sb) >= sb->sb_lowat ||
2453 sbavail(sb) >= uio->uio_resid ||
2454 sbavail(sb) >= sb->sb_hiwat) ) {
2458 /* On MSG_WAITALL we must wait until all data or error arrives. */
2459 if ((flags & MSG_WAITALL) &&
2460 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2464 * Wait and block until (more) data comes in.
2465 * NB: Drops the sockbuf lock during wait.
2473 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2474 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2475 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2479 uio->uio_td->td_ru.ru_msgrcv++;
2481 /* Fill uio until full or current end of socket buffer is reached. */
2482 len = min(uio->uio_resid, sbavail(sb));
2484 /* Dequeue as many mbufs as possible. */
2485 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2489 m_cat(*mp0, sb->sb_mb);
2491 m != NULL && m->m_len <= len;
2493 KASSERT(!(m->m_flags & M_NOTAVAIL),
2494 ("%s: m %p not available", __func__, m));
2496 uio->uio_resid -= m->m_len;
2502 sb->sb_lastrecord = sb->sb_mb;
2503 if (sb->sb_mb == NULL)
2506 /* Copy the remainder. */
2508 KASSERT(sb->sb_mb != NULL,
2509 ("%s: len > 0 && sb->sb_mb empty", __func__));
2511 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2513 len = 0; /* Don't flush data from sockbuf. */
2515 uio->uio_resid -= len;
2526 /* NB: Must unlock socket buffer as uiomove may sleep. */
2528 error = m_mbuftouio(uio, sb->sb_mb, len);
2533 SBLASTRECORDCHK(sb);
2537 * Remove the delivered data from the socket buffer unless we
2538 * were only peeking.
2540 if (!(flags & MSG_PEEK)) {
2542 sbdrop_locked(sb, len);
2544 /* Notify protocol that we drained some data. */
2545 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2546 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2547 !(flags & MSG_SOCALLBCK))) {
2550 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2556 * For MSG_WAITALL we may have to loop again and wait for
2557 * more data to come in.
2559 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2562 SOCKBUF_LOCK_ASSERT(sb);
2563 SBLASTRECORDCHK(sb);
2571 * Optimized version of soreceive() for simple datagram cases from userspace.
2572 * Unlike in the stream case, we're able to drop a datagram if copyout()
2573 * fails, and because we handle datagrams atomically, we don't need to use a
2574 * sleep lock to prevent I/O interlacing.
2577 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2578 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2580 struct mbuf *m, *m2;
2583 struct protosw *pr = so->so_proto;
2584 struct mbuf *nextrecord;
2588 if (controlp != NULL)
2591 flags = *flagsp &~ MSG_EOR;
2596 * For any complicated cases, fall back to the full
2597 * soreceive_generic().
2599 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2600 return (soreceive_generic(so, psa, uio, mp0, controlp,
2604 * Enforce restrictions on use.
2606 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2607 ("soreceive_dgram: wantrcvd"));
2608 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2609 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2610 ("soreceive_dgram: SBS_RCVATMARK"));
2611 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2612 ("soreceive_dgram: P_CONNREQUIRED"));
2615 * Loop blocking while waiting for a datagram.
2617 SOCKBUF_LOCK(&so->so_rcv);
2618 while ((m = so->so_rcv.sb_mb) == NULL) {
2619 KASSERT(sbavail(&so->so_rcv) == 0,
2620 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2621 sbavail(&so->so_rcv)));
2623 error = so->so_error;
2625 SOCKBUF_UNLOCK(&so->so_rcv);
2628 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2629 uio->uio_resid == 0) {
2630 SOCKBUF_UNLOCK(&so->so_rcv);
2633 if ((so->so_state & SS_NBIO) ||
2634 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2635 SOCKBUF_UNLOCK(&so->so_rcv);
2636 return (EWOULDBLOCK);
2638 SBLASTRECORDCHK(&so->so_rcv);
2639 SBLASTMBUFCHK(&so->so_rcv);
2640 error = sbwait(&so->so_rcv);
2642 SOCKBUF_UNLOCK(&so->so_rcv);
2646 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2649 uio->uio_td->td_ru.ru_msgrcv++;
2650 SBLASTRECORDCHK(&so->so_rcv);
2651 SBLASTMBUFCHK(&so->so_rcv);
2652 nextrecord = m->m_nextpkt;
2653 if (nextrecord == NULL) {
2654 KASSERT(so->so_rcv.sb_lastrecord == m,
2655 ("soreceive_dgram: lastrecord != m"));
2658 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2659 ("soreceive_dgram: m_nextpkt != nextrecord"));
2662 * Pull 'm' and its chain off the front of the packet queue.
2664 so->so_rcv.sb_mb = NULL;
2665 sockbuf_pushsync(&so->so_rcv, nextrecord);
2668 * Walk 'm's chain and free that many bytes from the socket buffer.
2670 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2671 sbfree(&so->so_rcv, m2);
2674 * Do a few last checks before we let go of the lock.
2676 SBLASTRECORDCHK(&so->so_rcv);
2677 SBLASTMBUFCHK(&so->so_rcv);
2678 SOCKBUF_UNLOCK(&so->so_rcv);
2680 if (pr->pr_flags & PR_ADDR) {
2681 KASSERT(m->m_type == MT_SONAME,
2682 ("m->m_type == %d", m->m_type));
2684 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2689 /* XXXRW: Can this happen? */
2694 * Packet to copyout() is now in 'm' and it is disconnected from the
2697 * Process one or more MT_CONTROL mbufs present before any data mbufs
2698 * in the first mbuf chain on the socket buffer. We call into the
2699 * protocol to perform externalization (or freeing if controlp ==
2700 * NULL). In some cases there can be only MT_CONTROL mbufs without
2703 if (m->m_type == MT_CONTROL) {
2704 struct mbuf *cm = NULL, *cmn;
2705 struct mbuf **cme = &cm;
2711 cme = &(*cme)->m_next;
2713 } while (m != NULL && m->m_type == MT_CONTROL);
2714 while (cm != NULL) {
2717 if (pr->pr_domain->dom_externalize != NULL) {
2718 error = (*pr->pr_domain->dom_externalize)
2719 (cm, controlp, flags);
2720 } else if (controlp != NULL)
2724 if (controlp != NULL) {
2725 while (*controlp != NULL)
2726 controlp = &(*controlp)->m_next;
2731 KASSERT(m == NULL || m->m_type == MT_DATA,
2732 ("soreceive_dgram: !data"));
2733 while (m != NULL && uio->uio_resid > 0) {
2734 len = uio->uio_resid;
2737 error = uiomove(mtod(m, char *), (int)len, uio);
2742 if (len == m->m_len)
2759 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2760 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2764 CURVNET_SET(so->so_vnet);
2765 if (!SOLISTENING(so))
2766 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2767 mp0, controlp, flagsp));
2775 soshutdown(struct socket *so, int how)
2777 struct protosw *pr = so->so_proto;
2778 int error, soerror_enotconn;
2780 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2783 soerror_enotconn = 0;
2785 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2787 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2788 * invoked on a datagram sockets, however historically we would
2789 * actually tear socket down. This is known to be leveraged by
2790 * some applications to unblock process waiting in recvXXX(2)
2791 * by other process that it shares that socket with. Try to meet
2792 * both backward-compatibility and POSIX requirements by forcing
2793 * ENOTCONN but still asking protocol to perform pru_shutdown().
2795 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so))
2797 soerror_enotconn = 1;
2800 if (SOLISTENING(so)) {
2801 if (how != SHUT_WR) {
2803 so->so_error = ECONNABORTED;
2804 solisten_wakeup(so); /* unlocks so */
2809 CURVNET_SET(so->so_vnet);
2810 if (pr->pr_usrreqs->pru_flush != NULL)
2811 (*pr->pr_usrreqs->pru_flush)(so, how);
2814 if (how != SHUT_RD) {
2815 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2816 wakeup(&so->so_timeo);
2818 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2820 wakeup(&so->so_timeo);
2824 return (soerror_enotconn ? ENOTCONN : 0);
2828 sorflush(struct socket *so)
2830 struct sockbuf *sb = &so->so_rcv;
2831 struct protosw *pr = so->so_proto;
2837 * In order to avoid calling dom_dispose with the socket buffer mutex
2838 * held, and in order to generally avoid holding the lock for a long
2839 * time, we make a copy of the socket buffer and clear the original
2840 * (except locks, state). The new socket buffer copy won't have
2841 * initialized locks so we can only call routines that won't use or
2842 * assert those locks.
2844 * Dislodge threads currently blocked in receive and wait to acquire
2845 * a lock against other simultaneous readers before clearing the
2846 * socket buffer. Don't let our acquire be interrupted by a signal
2847 * despite any existing socket disposition on interruptable waiting.
2850 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2853 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2854 * and mutex data unchanged.
2857 bzero(&aso, sizeof(aso));
2858 aso.so_pcb = so->so_pcb;
2859 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2860 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2861 bzero(&sb->sb_startzero,
2862 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2867 * Dispose of special rights and flush the copied socket. Don't call
2868 * any unsafe routines (that rely on locks being initialized) on aso.
2870 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2871 (*pr->pr_domain->dom_dispose)(&aso);
2872 sbrelease_internal(&aso.so_rcv, so);
2876 * Wrapper for Socket established helper hook.
2877 * Parameters: socket, context of the hook point, hook id.
2880 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2882 struct socket_hhook_data hhook_data = {
2889 CURVNET_SET(so->so_vnet);
2890 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2893 /* Ugly but needed, since hhooks return void for now */
2894 return (hhook_data.status);
2898 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2899 * additional variant to handle the case where the option value needs to be
2900 * some kind of integer, but not a specific size. In addition to their use
2901 * here, these functions are also called by the protocol-level pr_ctloutput()
2905 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2910 * If the user gives us more than we wanted, we ignore it, but if we
2911 * don't get the minimum length the caller wants, we return EINVAL.
2912 * On success, sopt->sopt_valsize is set to however much we actually
2915 if ((valsize = sopt->sopt_valsize) < minlen)
2918 sopt->sopt_valsize = valsize = len;
2920 if (sopt->sopt_td != NULL)
2921 return (copyin(sopt->sopt_val, buf, valsize));
2923 bcopy(sopt->sopt_val, buf, valsize);
2928 * Kernel version of setsockopt(2).
2930 * XXX: optlen is size_t, not socklen_t
2933 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2936 struct sockopt sopt;
2938 sopt.sopt_level = level;
2939 sopt.sopt_name = optname;
2940 sopt.sopt_dir = SOPT_SET;
2941 sopt.sopt_val = optval;
2942 sopt.sopt_valsize = optlen;
2943 sopt.sopt_td = NULL;
2944 return (sosetopt(so, &sopt));
2948 sosetopt(struct socket *so, struct sockopt *sopt)
2959 CURVNET_SET(so->so_vnet);
2961 if (sopt->sopt_level != SOL_SOCKET) {
2962 if (so->so_proto->pr_ctloutput != NULL)
2963 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2965 error = ENOPROTOOPT;
2967 switch (sopt->sopt_name) {
2968 case SO_ACCEPTFILTER:
2969 error = accept_filt_setopt(so, sopt);
2975 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2978 if (l.l_linger < 0 ||
2979 l.l_linger > USHRT_MAX ||
2980 l.l_linger > (INT_MAX / hz)) {
2985 so->so_linger = l.l_linger;
2987 so->so_options |= SO_LINGER;
2989 so->so_options &= ~SO_LINGER;
2996 case SO_USELOOPBACK:
3000 case SO_REUSEPORT_LB:
3007 error = sooptcopyin(sopt, &optval, sizeof optval,
3013 so->so_options |= sopt->sopt_name;
3015 so->so_options &= ~sopt->sopt_name;
3020 error = sooptcopyin(sopt, &optval, sizeof optval,
3025 if (optval < 0 || optval >= rt_numfibs) {
3029 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3030 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3031 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3032 so->so_fibnum = optval;
3037 case SO_USER_COOKIE:
3038 error = sooptcopyin(sopt, &val32, sizeof val32,
3042 so->so_user_cookie = val32;
3049 error = sooptcopyin(sopt, &optval, sizeof optval,
3055 * Values < 1 make no sense for any of these options,
3063 error = sbsetopt(so, sopt->sopt_name, optval);
3068 #ifdef COMPAT_FREEBSD32
3069 if (SV_CURPROC_FLAG(SV_ILP32)) {
3070 struct timeval32 tv32;
3072 error = sooptcopyin(sopt, &tv32, sizeof tv32,
3074 CP(tv32, tv, tv_sec);
3075 CP(tv32, tv, tv_usec);
3078 error = sooptcopyin(sopt, &tv, sizeof tv,
3082 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3083 tv.tv_usec >= 1000000) {
3087 if (tv.tv_sec > INT32_MAX)
3091 switch (sopt->sopt_name) {
3093 so->so_snd.sb_timeo = val;
3096 so->so_rcv.sb_timeo = val;
3103 error = sooptcopyin(sopt, &extmac, sizeof extmac,
3107 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3115 error = sooptcopyin(sopt, &optval, sizeof optval,
3119 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3123 so->so_ts_clock = optval;
3126 case SO_MAX_PACING_RATE:
3127 error = sooptcopyin(sopt, &val32, sizeof(val32),
3131 so->so_max_pacing_rate = val32;
3135 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3136 error = hhook_run_socket(so, sopt,
3139 error = ENOPROTOOPT;
3142 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3143 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
3151 * Helper routine for getsockopt.
3154 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3162 * Documented get behavior is that we always return a value, possibly
3163 * truncated to fit in the user's buffer. Traditional behavior is
3164 * that we always tell the user precisely how much we copied, rather
3165 * than something useful like the total amount we had available for
3166 * her. Note that this interface is not idempotent; the entire
3167 * answer must be generated ahead of time.
3169 valsize = min(len, sopt->sopt_valsize);
3170 sopt->sopt_valsize = valsize;
3171 if (sopt->sopt_val != NULL) {
3172 if (sopt->sopt_td != NULL)
3173 error = copyout(buf, sopt->sopt_val, valsize);
3175 bcopy(buf, sopt->sopt_val, valsize);
3181 sogetopt(struct socket *so, struct sockopt *sopt)
3190 CURVNET_SET(so->so_vnet);
3192 if (sopt->sopt_level != SOL_SOCKET) {
3193 if (so->so_proto->pr_ctloutput != NULL)
3194 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3196 error = ENOPROTOOPT;
3200 switch (sopt->sopt_name) {
3201 case SO_ACCEPTFILTER:
3202 error = accept_filt_getopt(so, sopt);
3207 l.l_onoff = so->so_options & SO_LINGER;
3208 l.l_linger = so->so_linger;
3210 error = sooptcopyout(sopt, &l, sizeof l);
3213 case SO_USELOOPBACK:
3219 case SO_REUSEPORT_LB:
3226 optval = so->so_options & sopt->sopt_name;
3228 error = sooptcopyout(sopt, &optval, sizeof optval);
3232 optval = so->so_proto->pr_domain->dom_family;
3236 optval = so->so_type;
3240 optval = so->so_proto->pr_protocol;
3245 optval = so->so_error;
3251 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3252 so->so_snd.sb_hiwat;
3256 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3257 so->so_rcv.sb_hiwat;
3261 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3262 so->so_snd.sb_lowat;
3266 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3267 so->so_rcv.sb_lowat;
3272 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3273 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3274 #ifdef COMPAT_FREEBSD32
3275 if (SV_CURPROC_FLAG(SV_ILP32)) {
3276 struct timeval32 tv32;
3278 CP(tv, tv32, tv_sec);
3279 CP(tv, tv32, tv_usec);
3280 error = sooptcopyout(sopt, &tv32, sizeof tv32);
3283 error = sooptcopyout(sopt, &tv, sizeof tv);
3288 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3292 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3296 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3304 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3308 error = mac_getsockopt_peerlabel(
3309 sopt->sopt_td->td_ucred, so, &extmac);
3312 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3318 case SO_LISTENQLIMIT:
3319 optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3323 optval = SOLISTENING(so) ? so->sol_qlen : 0;
3326 case SO_LISTENINCQLEN:
3327 optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3331 optval = so->so_ts_clock;
3334 case SO_MAX_PACING_RATE:
3335 optval = so->so_max_pacing_rate;
3339 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3340 error = hhook_run_socket(so, sopt,
3343 error = ENOPROTOOPT;
3355 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3357 struct mbuf *m, *m_prev;
3358 int sopt_size = sopt->sopt_valsize;
3360 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3363 if (sopt_size > MLEN) {
3364 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3365 if ((m->m_flags & M_EXT) == 0) {
3369 m->m_len = min(MCLBYTES, sopt_size);
3371 m->m_len = min(MLEN, sopt_size);
3373 sopt_size -= m->m_len;
3378 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3383 if (sopt_size > MLEN) {
3384 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3386 if ((m->m_flags & M_EXT) == 0) {
3391 m->m_len = min(MCLBYTES, sopt_size);
3393 m->m_len = min(MLEN, sopt_size);
3395 sopt_size -= m->m_len;
3403 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3405 struct mbuf *m0 = m;
3407 if (sopt->sopt_val == NULL)
3409 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3410 if (sopt->sopt_td != NULL) {
3413 error = copyin(sopt->sopt_val, mtod(m, char *),
3420 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3421 sopt->sopt_valsize -= m->m_len;
3422 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3425 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3426 panic("ip6_sooptmcopyin");
3431 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3433 struct mbuf *m0 = m;
3436 if (sopt->sopt_val == NULL)
3438 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3439 if (sopt->sopt_td != NULL) {
3442 error = copyout(mtod(m, char *), sopt->sopt_val,
3449 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3450 sopt->sopt_valsize -= m->m_len;
3451 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3452 valsize += m->m_len;
3456 /* enough soopt buffer should be given from user-land */
3460 sopt->sopt_valsize = valsize;
3465 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3466 * out-of-band data, which will then notify socket consumers.
3469 sohasoutofband(struct socket *so)
3472 if (so->so_sigio != NULL)
3473 pgsigio(&so->so_sigio, SIGURG, 0);
3474 selwakeuppri(&so->so_rdsel, PSOCK);
3478 sopoll(struct socket *so, int events, struct ucred *active_cred,
3483 * We do not need to set or assert curvnet as long as everyone uses
3486 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3491 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3497 if (SOLISTENING(so)) {
3498 if (!(events & (POLLIN | POLLRDNORM)))
3500 else if (!TAILQ_EMPTY(&so->sol_comp))
3501 revents = events & (POLLIN | POLLRDNORM);
3502 else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3503 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3505 selrecord(td, &so->so_rdsel);
3510 SOCKBUF_LOCK(&so->so_snd);
3511 SOCKBUF_LOCK(&so->so_rcv);
3512 if (events & (POLLIN | POLLRDNORM))
3513 if (soreadabledata(so))
3514 revents |= events & (POLLIN | POLLRDNORM);
3515 if (events & (POLLOUT | POLLWRNORM))
3516 if (sowriteable(so))
3517 revents |= events & (POLLOUT | POLLWRNORM);
3518 if (events & (POLLPRI | POLLRDBAND))
3519 if (so->so_oobmark ||
3520 (so->so_rcv.sb_state & SBS_RCVATMARK))
3521 revents |= events & (POLLPRI | POLLRDBAND);
3522 if ((events & POLLINIGNEOF) == 0) {
3523 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3524 revents |= events & (POLLIN | POLLRDNORM);
3525 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3531 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3532 selrecord(td, &so->so_rdsel);
3533 so->so_rcv.sb_flags |= SB_SEL;
3535 if (events & (POLLOUT | POLLWRNORM)) {
3536 selrecord(td, &so->so_wrsel);
3537 so->so_snd.sb_flags |= SB_SEL;
3540 SOCKBUF_UNLOCK(&so->so_rcv);
3541 SOCKBUF_UNLOCK(&so->so_snd);
3548 soo_kqfilter(struct file *fp, struct knote *kn)
3550 struct socket *so = kn->kn_fp->f_data;
3554 switch (kn->kn_filter) {
3556 kn->kn_fop = &soread_filtops;
3557 knl = &so->so_rdsel.si_note;
3561 kn->kn_fop = &sowrite_filtops;
3562 knl = &so->so_wrsel.si_note;
3566 kn->kn_fop = &soempty_filtops;
3567 knl = &so->so_wrsel.si_note;
3575 if (SOLISTENING(so)) {
3576 knlist_add(knl, kn, 1);
3579 knlist_add(knl, kn, 1);
3580 sb->sb_flags |= SB_KNOTE;
3588 * Some routines that return EOPNOTSUPP for entry points that are not
3589 * supported by a protocol. Fill in as needed.
3592 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3599 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3606 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3613 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3620 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3628 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3635 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3643 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3650 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3651 struct ifnet *ifp, struct thread *td)
3658 pru_disconnect_notsupp(struct socket *so)
3665 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3672 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3679 pru_rcvd_notsupp(struct socket *so, int flags)
3686 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3693 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3694 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3701 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3704 return (EOPNOTSUPP);
3708 * This isn't really a ``null'' operation, but it's the default one and
3709 * doesn't do anything destructive.
3712 pru_sense_null(struct socket *so, struct stat *sb)
3715 sb->st_blksize = so->so_snd.sb_hiwat;
3720 pru_shutdown_notsupp(struct socket *so)
3727 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3734 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3735 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3742 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3743 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3750 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3758 filt_sordetach(struct knote *kn)
3760 struct socket *so = kn->kn_fp->f_data;
3763 knlist_remove(&so->so_rdsel.si_note, kn, 1);
3764 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3765 so->so_rcv.sb_flags &= ~SB_KNOTE;
3766 so_rdknl_unlock(so);
3771 filt_soread(struct knote *kn, long hint)
3775 so = kn->kn_fp->f_data;
3777 if (SOLISTENING(so)) {
3778 SOCK_LOCK_ASSERT(so);
3779 kn->kn_data = so->sol_qlen;
3781 kn->kn_flags |= EV_EOF;
3782 kn->kn_fflags = so->so_error;
3785 return (!TAILQ_EMPTY(&so->sol_comp));
3788 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3790 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3791 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3792 kn->kn_flags |= EV_EOF;
3793 kn->kn_fflags = so->so_error;
3795 } else if (so->so_error) /* temporary udp error */
3798 if (kn->kn_sfflags & NOTE_LOWAT) {
3799 if (kn->kn_data >= kn->kn_sdata)
3801 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3804 /* This hook returning non-zero indicates an event, not error */
3805 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3809 filt_sowdetach(struct knote *kn)
3811 struct socket *so = kn->kn_fp->f_data;
3814 knlist_remove(&so->so_wrsel.si_note, kn, 1);
3815 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3816 so->so_snd.sb_flags &= ~SB_KNOTE;
3817 so_wrknl_unlock(so);
3822 filt_sowrite(struct knote *kn, long hint)
3826 so = kn->kn_fp->f_data;
3828 if (SOLISTENING(so))
3831 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3832 kn->kn_data = sbspace(&so->so_snd);
3834 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3836 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3837 kn->kn_flags |= EV_EOF;
3838 kn->kn_fflags = so->so_error;
3840 } else if (so->so_error) /* temporary udp error */
3842 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3843 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3845 else if (kn->kn_sfflags & NOTE_LOWAT)
3846 return (kn->kn_data >= kn->kn_sdata);
3848 return (kn->kn_data >= so->so_snd.sb_lowat);
3852 filt_soempty(struct knote *kn, long hint)
3856 so = kn->kn_fp->f_data;
3858 if (SOLISTENING(so))
3861 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3862 kn->kn_data = sbused(&so->so_snd);
3864 if (kn->kn_data == 0)
3871 socheckuid(struct socket *so, uid_t uid)
3876 if (so->so_cred->cr_uid != uid)
3882 * These functions are used by protocols to notify the socket layer (and its
3883 * consumers) of state changes in the sockets driven by protocol-side events.
3887 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3889 * Normal sequence from the active (originating) side is that
3890 * soisconnecting() is called during processing of connect() call, resulting
3891 * in an eventual call to soisconnected() if/when the connection is
3892 * established. When the connection is torn down soisdisconnecting() is
3893 * called during processing of disconnect() call, and soisdisconnected() is
3894 * called when the connection to the peer is totally severed. The semantics
3895 * of these routines are such that connectionless protocols can call
3896 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3897 * calls when setting up a ``connection'' takes no time.
3899 * From the passive side, a socket is created with two queues of sockets:
3900 * so_incomp for connections in progress and so_comp for connections already
3901 * made and awaiting user acceptance. As a protocol is preparing incoming
3902 * connections, it creates a socket structure queued on so_incomp by calling
3903 * sonewconn(). When the connection is established, soisconnected() is
3904 * called, and transfers the socket structure to so_comp, making it available
3907 * If a socket is closed with sockets on either so_incomp or so_comp, these
3908 * sockets are dropped.
3910 * If higher-level protocols are implemented in the kernel, the wakeups done
3911 * here will sometimes cause software-interrupt process scheduling.
3914 soisconnecting(struct socket *so)
3918 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3919 so->so_state |= SS_ISCONNECTING;
3924 soisconnected(struct socket *so)
3928 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3929 so->so_state |= SS_ISCONNECTED;
3931 if (so->so_qstate == SQ_INCOMP) {
3932 struct socket *head = so->so_listen;
3935 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3937 * Promoting a socket from incomplete queue to complete, we
3938 * need to go through reverse order of locking. We first do
3939 * trylock, and if that doesn't succeed, we go the hard way
3940 * leaving a reference and rechecking consistency after proper
3943 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3946 SOLISTEN_LOCK(head);
3948 if (__predict_false(head != so->so_listen)) {
3950 * The socket went off the listen queue,
3951 * should be lost race to close(2) of sol.
3952 * The socket is about to soabort().
3958 /* Not the last one, as so holds a ref. */
3959 refcount_release(&head->so_count);
3962 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3963 TAILQ_REMOVE(&head->sol_incomp, so, so_list);
3964 head->sol_incqlen--;
3965 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
3967 so->so_qstate = SQ_COMP;
3969 solisten_wakeup(head); /* unlocks */
3971 SOCKBUF_LOCK(&so->so_rcv);
3972 soupcall_set(so, SO_RCV,
3973 head->sol_accept_filter->accf_callback,
3974 head->sol_accept_filter_arg);
3975 so->so_options &= ~SO_ACCEPTFILTER;
3976 ret = head->sol_accept_filter->accf_callback(so,
3977 head->sol_accept_filter_arg, M_NOWAIT);
3978 if (ret == SU_ISCONNECTED) {
3979 soupcall_clear(so, SO_RCV);
3980 SOCKBUF_UNLOCK(&so->so_rcv);
3983 SOCKBUF_UNLOCK(&so->so_rcv);
3985 SOLISTEN_UNLOCK(head);
3990 wakeup(&so->so_timeo);
3996 soisdisconnecting(struct socket *so)
4000 so->so_state &= ~SS_ISCONNECTING;
4001 so->so_state |= SS_ISDISCONNECTING;
4003 if (!SOLISTENING(so)) {
4004 SOCKBUF_LOCK(&so->so_rcv);
4005 socantrcvmore_locked(so);
4006 SOCKBUF_LOCK(&so->so_snd);
4007 socantsendmore_locked(so);
4010 wakeup(&so->so_timeo);
4014 soisdisconnected(struct socket *so)
4018 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4019 so->so_state |= SS_ISDISCONNECTED;
4021 if (!SOLISTENING(so)) {
4023 SOCKBUF_LOCK(&so->so_rcv);
4024 socantrcvmore_locked(so);
4025 SOCKBUF_LOCK(&so->so_snd);
4026 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4027 socantsendmore_locked(so);
4030 wakeup(&so->so_timeo);
4034 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4037 sodupsockaddr(const struct sockaddr *sa, int mflags)
4039 struct sockaddr *sa2;
4041 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4043 bcopy(sa, sa2, sa->sa_len);
4048 * Register per-socket destructor.
4051 sodtor_set(struct socket *so, so_dtor_t *func)
4054 SOCK_LOCK_ASSERT(so);
4059 * Register per-socket buffer upcalls.
4062 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4066 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4076 panic("soupcall_set: bad which");
4078 SOCKBUF_LOCK_ASSERT(sb);
4079 sb->sb_upcall = func;
4080 sb->sb_upcallarg = arg;
4081 sb->sb_flags |= SB_UPCALL;
4085 soupcall_clear(struct socket *so, int which)
4089 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4099 panic("soupcall_clear: bad which");
4101 SOCKBUF_LOCK_ASSERT(sb);
4102 KASSERT(sb->sb_upcall != NULL,
4103 ("%s: so %p no upcall to clear", __func__, so));
4104 sb->sb_upcall = NULL;
4105 sb->sb_upcallarg = NULL;
4106 sb->sb_flags &= ~SB_UPCALL;
4110 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4113 SOLISTEN_LOCK_ASSERT(so);
4114 so->sol_upcall = func;
4115 so->sol_upcallarg = arg;
4119 so_rdknl_lock(void *arg)
4121 struct socket *so = arg;
4123 if (SOLISTENING(so))
4126 SOCKBUF_LOCK(&so->so_rcv);
4130 so_rdknl_unlock(void *arg)
4132 struct socket *so = arg;
4134 if (SOLISTENING(so))
4137 SOCKBUF_UNLOCK(&so->so_rcv);
4141 so_rdknl_assert_locked(void *arg)
4143 struct socket *so = arg;
4145 if (SOLISTENING(so))
4146 SOCK_LOCK_ASSERT(so);
4148 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4152 so_rdknl_assert_unlocked(void *arg)
4154 struct socket *so = arg;
4156 if (SOLISTENING(so))
4157 SOCK_UNLOCK_ASSERT(so);
4159 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4163 so_wrknl_lock(void *arg)
4165 struct socket *so = arg;
4167 if (SOLISTENING(so))
4170 SOCKBUF_LOCK(&so->so_snd);
4174 so_wrknl_unlock(void *arg)
4176 struct socket *so = arg;
4178 if (SOLISTENING(so))
4181 SOCKBUF_UNLOCK(&so->so_snd);
4185 so_wrknl_assert_locked(void *arg)
4187 struct socket *so = arg;
4189 if (SOLISTENING(so))
4190 SOCK_LOCK_ASSERT(so);
4192 SOCKBUF_LOCK_ASSERT(&so->so_snd);
4196 so_wrknl_assert_unlocked(void *arg)
4198 struct socket *so = arg;
4200 if (SOLISTENING(so))
4201 SOCK_UNLOCK_ASSERT(so);
4203 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4207 * Create an external-format (``xsocket'') structure using the information in
4208 * the kernel-format socket structure pointed to by so. This is done to
4209 * reduce the spew of irrelevant information over this interface, to isolate
4210 * user code from changes in the kernel structure, and potentially to provide
4211 * information-hiding if we decide that some of this information should be
4212 * hidden from users.
4215 sotoxsocket(struct socket *so, struct xsocket *xso)
4218 bzero(xso, sizeof(*xso));
4219 xso->xso_len = sizeof *xso;
4220 xso->xso_so = (uintptr_t)so;
4221 xso->so_type = so->so_type;
4222 xso->so_options = so->so_options;
4223 xso->so_linger = so->so_linger;
4224 xso->so_state = so->so_state;
4225 xso->so_pcb = (uintptr_t)so->so_pcb;
4226 xso->xso_protocol = so->so_proto->pr_protocol;
4227 xso->xso_family = so->so_proto->pr_domain->dom_family;
4228 xso->so_timeo = so->so_timeo;
4229 xso->so_error = so->so_error;
4230 xso->so_uid = so->so_cred->cr_uid;
4231 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4232 if (SOLISTENING(so)) {
4233 xso->so_qlen = so->sol_qlen;
4234 xso->so_incqlen = so->sol_incqlen;
4235 xso->so_qlimit = so->sol_qlimit;
4236 xso->so_oobmark = 0;
4238 xso->so_state |= so->so_qstate;
4239 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4240 xso->so_oobmark = so->so_oobmark;
4241 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4242 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4247 so_sockbuf_rcv(struct socket *so)
4250 return (&so->so_rcv);
4254 so_sockbuf_snd(struct socket *so)
4257 return (&so->so_snd);
4261 so_state_get(const struct socket *so)
4264 return (so->so_state);
4268 so_state_set(struct socket *so, int val)
4275 so_options_get(const struct socket *so)
4278 return (so->so_options);
4282 so_options_set(struct socket *so, int val)
4285 so->so_options = val;
4289 so_error_get(const struct socket *so)
4292 return (so->so_error);
4296 so_error_set(struct socket *so, int val)
4303 so_linger_get(const struct socket *so)
4306 return (so->so_linger);
4310 so_linger_set(struct socket *so, int val)
4313 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4314 ("%s: val %d out of range", __func__, val));
4316 so->so_linger = val;
4320 so_protosw_get(const struct socket *so)
4323 return (so->so_proto);
4327 so_protosw_set(struct socket *so, struct protosw *val)
4334 so_sorwakeup(struct socket *so)
4341 so_sowwakeup(struct socket *so)
4348 so_sorwakeup_locked(struct socket *so)
4351 sorwakeup_locked(so);
4355 so_sowwakeup_locked(struct socket *so)
4358 sowwakeup_locked(so);
4362 so_lock(struct socket *so)
4369 so_unlock(struct socket *so)