2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2004 The FreeBSD Foundation
5 * Copyright (c) 2004-2008 Robert N. M. Watson
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
36 * Comments on the socket life cycle:
38 * soalloc() sets of socket layer state for a socket, called only by
39 * socreate() and sonewconn(). Socket layer private.
41 * sodealloc() tears down socket layer state for a socket, called only by
42 * sofree() and sonewconn(). Socket layer private.
44 * pru_attach() associates protocol layer state with an allocated socket;
45 * called only once, may fail, aborting socket allocation. This is called
46 * from socreate() and sonewconn(). Socket layer private.
48 * pru_detach() disassociates protocol layer state from an attached socket,
49 * and will be called exactly once for sockets in which pru_attach() has
50 * been successfully called. If pru_attach() returned an error,
51 * pru_detach() will not be called. Socket layer private.
53 * pru_abort() and pru_close() notify the protocol layer that the last
54 * consumer of a socket is starting to tear down the socket, and that the
55 * protocol should terminate the connection. Historically, pru_abort() also
56 * detached protocol state from the socket state, but this is no longer the
59 * socreate() creates a socket and attaches protocol state. This is a public
60 * interface that may be used by socket layer consumers to create new
63 * sonewconn() creates a socket and attaches protocol state. This is a
64 * public interface that may be used by protocols to create new sockets when
65 * a new connection is received and will be available for accept() on a
68 * soclose() destroys a socket after possibly waiting for it to disconnect.
69 * This is a public interface that socket consumers should use to close and
70 * release a socket when done with it.
72 * soabort() destroys a socket without waiting for it to disconnect (used
73 * only for incoming connections that are already partially or fully
74 * connected). This is used internally by the socket layer when clearing
75 * listen socket queues (due to overflow or close on the listen socket), but
76 * is also a public interface protocols may use to abort connections in
77 * their incomplete listen queues should they no longer be required. Sockets
78 * placed in completed connection listen queues should not be aborted for
79 * reasons described in the comment above the soclose() implementation. This
80 * is not a general purpose close routine, and except in the specific
81 * circumstances described here, should not be used.
83 * sofree() will free a socket and its protocol state if all references on
84 * the socket have been released, and is the public interface to attempt to
85 * free a socket when a reference is removed. This is a socket layer private
88 * NOTE: In addition to socreate() and soclose(), which provide a single
89 * socket reference to the consumer to be managed as required, there are two
90 * calls to explicitly manage socket references, soref(), and sorele().
91 * Currently, these are generally required only when transitioning a socket
92 * from a listen queue to a file descriptor, in order to prevent garbage
93 * collection of the socket at an untimely moment. For a number of reasons,
94 * these interfaces are not preferred, and should be avoided.
96 * NOTE: With regard to VNETs the general rule is that callers do not set
97 * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
98 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
99 * and sorflush(), which are usually called from a pre-set VNET context.
100 * sopoll() currently does not need a VNET context to be set.
103 #include <sys/cdefs.h>
104 __FBSDID("$FreeBSD$");
106 #include "opt_inet.h"
107 #include "opt_inet6.h"
108 #include "opt_compat.h"
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/fcntl.h>
113 #include <sys/limits.h>
114 #include <sys/lock.h>
116 #include <sys/malloc.h>
117 #include <sys/mbuf.h>
118 #include <sys/mutex.h>
119 #include <sys/domain.h>
120 #include <sys/file.h> /* for struct knote */
121 #include <sys/hhook.h>
122 #include <sys/kernel.h>
123 #include <sys/khelp.h>
124 #include <sys/event.h>
125 #include <sys/eventhandler.h>
126 #include <sys/poll.h>
127 #include <sys/proc.h>
128 #include <sys/protosw.h>
129 #include <sys/socket.h>
130 #include <sys/socketvar.h>
131 #include <sys/resourcevar.h>
132 #include <net/route.h>
133 #include <sys/signalvar.h>
134 #include <sys/stat.h>
136 #include <sys/sysctl.h>
137 #include <sys/taskqueue.h>
139 #include <sys/jail.h>
140 #include <sys/syslog.h>
141 #include <netinet/in.h>
143 #include <net/vnet.h>
145 #include <security/mac/mac_framework.h>
149 #ifdef COMPAT_FREEBSD32
150 #include <sys/mount.h>
151 #include <sys/sysent.h>
152 #include <compat/freebsd32/freebsd32.h>
155 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
158 static void filt_sordetach(struct knote *kn);
159 static int filt_soread(struct knote *kn, long hint);
160 static void filt_sowdetach(struct knote *kn);
161 static int filt_sowrite(struct knote *kn, long hint);
162 static int filt_solisten(struct knote *kn, long hint);
163 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
164 static int filt_soempty(struct knote *kn, long hint);
165 fo_kqfilter_t soo_kqfilter;
167 static struct filterops solisten_filtops = {
169 .f_detach = filt_sordetach,
170 .f_event = filt_solisten,
172 static struct filterops soread_filtops = {
174 .f_detach = filt_sordetach,
175 .f_event = filt_soread,
177 static struct filterops sowrite_filtops = {
179 .f_detach = filt_sowdetach,
180 .f_event = filt_sowrite,
182 static struct filterops soempty_filtops = {
184 .f_detach = filt_sowdetach,
185 .f_event = filt_soempty,
188 so_gen_t so_gencnt; /* generation count for sockets */
190 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
191 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
193 #define VNET_SO_ASSERT(so) \
194 VNET_ASSERT(curvnet != NULL, \
195 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
197 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
198 #define V_socket_hhh VNET(socket_hhh)
201 * Limit on the number of connections in the listen queue waiting
203 * NB: The original sysctl somaxconn is still available but hidden
204 * to prevent confusion about the actual purpose of this number.
206 static u_int somaxconn = SOMAXCONN;
209 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
215 error = sysctl_handle_int(oidp, &val, 0, req);
216 if (error || !req->newptr )
220 * The purpose of the UINT_MAX / 3 limit, is so that the formula
222 * below, will not overflow.
225 if (val < 1 || val > UINT_MAX / 3)
231 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW,
232 0, sizeof(int), sysctl_somaxconn, "I",
233 "Maximum listen socket pending connection accept queue size");
234 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
235 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP,
236 0, sizeof(int), sysctl_somaxconn, "I",
237 "Maximum listen socket pending connection accept queue size (compat)");
239 static int numopensockets;
240 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
241 &numopensockets, 0, "Number of open sockets");
244 * accept_mtx locks down per-socket fields relating to accept queues. See
245 * socketvar.h for an annotation of the protected fields of struct socket.
247 struct mtx accept_mtx;
248 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
251 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
254 static struct mtx so_global_mtx;
255 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
258 * General IPC sysctl name space, used by sockets and a variety of other IPC
261 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
264 * Initialize the socket subsystem and set up the socket
267 static uma_zone_t socket_zone;
271 socket_zone_change(void *tag)
274 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
278 socket_hhook_register(int subtype)
281 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
282 &V_socket_hhh[subtype],
283 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
284 printf("%s: WARNING: unable to register hook\n", __func__);
288 socket_hhook_deregister(int subtype)
291 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
292 printf("%s: WARNING: unable to deregister hook\n", __func__);
296 socket_init(void *tag)
299 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
300 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
301 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
302 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
303 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
304 EVENTHANDLER_PRI_FIRST);
306 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
309 socket_vnet_init(const void *unused __unused)
313 /* We expect a contiguous range */
314 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
315 socket_hhook_register(i);
317 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
318 socket_vnet_init, NULL);
321 socket_vnet_uninit(const void *unused __unused)
325 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
326 socket_hhook_deregister(i);
328 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
329 socket_vnet_uninit, NULL);
332 * Initialise maxsockets. This SYSINIT must be run after
336 init_maxsockets(void *ignored)
339 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
340 maxsockets = imax(maxsockets, maxfiles);
342 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
345 * Sysctl to get and set the maximum global sockets limit. Notify protocols
346 * of the change so that they can update their dependent limits as required.
349 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
351 int error, newmaxsockets;
353 newmaxsockets = maxsockets;
354 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
355 if (error == 0 && req->newptr) {
356 if (newmaxsockets > maxsockets &&
357 newmaxsockets <= maxfiles) {
358 maxsockets = newmaxsockets;
359 EVENTHANDLER_INVOKE(maxsockets_change);
365 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
366 &maxsockets, 0, sysctl_maxsockets, "IU",
367 "Maximum number of sockets available");
370 * Socket operation routines. These routines are called by the routines in
371 * sys_socket.c or from a system process, and implement the semantics of
372 * socket operations by switching out to the protocol specific routines.
376 * Get a socket structure from our zone, and initialize it. Note that it
377 * would probably be better to allocate socket and PCB at the same time, but
378 * I'm not convinced that all the protocols can be easily modified to do
381 * soalloc() returns a socket with a ref count of 0.
383 static struct socket *
384 soalloc(struct vnet *vnet)
388 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
392 if (mac_socket_init(so, M_NOWAIT) != 0) {
393 uma_zfree(socket_zone, so);
397 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
398 uma_zfree(socket_zone, so);
402 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
403 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
404 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
405 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
406 TAILQ_INIT(&so->so_snd.sb_aiojobq);
407 TAILQ_INIT(&so->so_rcv.sb_aiojobq);
408 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
409 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
411 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
412 __func__, __LINE__, so));
415 /* We shouldn't need the so_global_mtx */
416 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
417 /* Do we need more comprehensive error returns? */
418 uma_zfree(socket_zone, so);
421 mtx_lock(&so_global_mtx);
422 so->so_gencnt = ++so_gencnt;
425 vnet->vnet_sockcnt++;
427 mtx_unlock(&so_global_mtx);
433 * Free the storage associated with a socket at the socket layer, tear down
434 * locks, labels, etc. All protocol state is assumed already to have been
435 * torn down (and possibly never set up) by the caller.
438 sodealloc(struct socket *so)
441 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
442 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
444 mtx_lock(&so_global_mtx);
445 so->so_gencnt = ++so_gencnt;
446 --numopensockets; /* Could be below, but faster here. */
448 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
449 __func__, __LINE__, so));
450 so->so_vnet->vnet_sockcnt--;
452 mtx_unlock(&so_global_mtx);
453 if (so->so_rcv.sb_hiwat)
454 (void)chgsbsize(so->so_cred->cr_uidinfo,
455 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
456 if (so->so_snd.sb_hiwat)
457 (void)chgsbsize(so->so_cred->cr_uidinfo,
458 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
459 /* remove accept filter if one is present. */
460 if (so->so_accf != NULL)
461 do_setopt_accept_filter(so, NULL);
463 mac_socket_destroy(so);
465 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
468 khelp_destroy_osd(&so->osd);
469 sx_destroy(&so->so_snd.sb_sx);
470 sx_destroy(&so->so_rcv.sb_sx);
471 SOCKBUF_LOCK_DESTROY(&so->so_snd);
472 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
473 uma_zfree(socket_zone, so);
477 * socreate returns a socket with a ref count of 1. The socket should be
478 * closed with soclose().
481 socreate(int dom, struct socket **aso, int type, int proto,
482 struct ucred *cred, struct thread *td)
489 prp = pffindproto(dom, proto, type);
491 prp = pffindtype(dom, type);
494 /* No support for domain. */
495 if (pffinddomain(dom) == NULL)
496 return (EAFNOSUPPORT);
497 /* No support for socket type. */
498 if (proto == 0 && type != 0)
500 return (EPROTONOSUPPORT);
502 if (prp->pr_usrreqs->pru_attach == NULL ||
503 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
504 return (EPROTONOSUPPORT);
506 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
507 return (EPROTONOSUPPORT);
509 if (prp->pr_type != type)
511 so = soalloc(CRED_TO_VNET(cred));
515 TAILQ_INIT(&so->so_incomp);
516 TAILQ_INIT(&so->so_comp);
518 so->so_cred = crhold(cred);
519 if ((prp->pr_domain->dom_family == PF_INET) ||
520 (prp->pr_domain->dom_family == PF_INET6) ||
521 (prp->pr_domain->dom_family == PF_ROUTE))
522 so->so_fibnum = td->td_proc->p_fibnum;
527 mac_socket_create(cred, so);
529 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
530 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
533 * Auto-sizing of socket buffers is managed by the protocols and
534 * the appropriate flags must be set in the pru_attach function.
536 CURVNET_SET(so->so_vnet);
537 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
540 KASSERT(so->so_count == 1, ("socreate: so_count %d",
551 static int regression_sonewconn_earlytest = 1;
552 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
553 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
557 * When an attempt at a new connection is noted on a socket which accepts
558 * connections, sonewconn is called. If the connection is possible (subject
559 * to space constraints, etc.) then we allocate a new structure, properly
560 * linked into the data structure of the original socket, and return this.
561 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
563 * Note: the ref count on the socket is 0 on return.
566 sonewconn(struct socket *head, int connstatus)
568 static struct timeval lastover;
569 static struct timeval overinterval = { 60, 0 };
570 static int overcount;
576 over = (head->so_qlen > 3 * head->so_qlimit / 2);
579 if (regression_sonewconn_earlytest && over) {
585 if (ratecheck(&lastover, &overinterval)) {
586 log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
587 "%i already in queue awaiting acceptance "
588 "(%d occurrences)\n",
589 __func__, head->so_pcb, head->so_qlen, overcount);
596 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
597 __func__, __LINE__, head));
598 so = soalloc(head->so_vnet);
600 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
601 "limit reached or out of memory\n",
602 __func__, head->so_pcb);
605 if ((head->so_options & SO_ACCEPTFILTER) != 0)
608 so->so_type = head->so_type;
609 so->so_options = head->so_options &~ SO_ACCEPTCONN;
610 so->so_linger = head->so_linger;
611 so->so_state = head->so_state | SS_NOFDREF;
612 so->so_fibnum = head->so_fibnum;
613 so->so_proto = head->so_proto;
614 so->so_cred = crhold(head->so_cred);
616 mac_socket_newconn(head, so);
618 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
619 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
620 VNET_SO_ASSERT(head);
621 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
623 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
624 __func__, head->so_pcb);
627 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
629 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
630 __func__, head->so_pcb);
633 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
634 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
635 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
636 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
637 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
638 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
639 so->so_state |= connstatus;
642 * The accept socket may be tearing down but we just
643 * won a race on the ACCEPT_LOCK.
644 * However, if sctp_peeloff() is called on a 1-to-many
645 * style socket, the SO_ACCEPTCONN doesn't need to be set.
647 if (!(head->so_options & SO_ACCEPTCONN) &&
648 ((head->so_proto->pr_protocol != IPPROTO_SCTP) ||
649 (head->so_type != SOCK_SEQPACKET))) {
652 sofree(so); /* NB: returns ACCEPT_UNLOCK'ed. */
656 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
657 so->so_qstate |= SQ_COMP;
661 * Keep removing sockets from the head until there's room for
662 * us to insert on the tail. In pre-locking revisions, this
663 * was a simple if(), but as we could be racing with other
664 * threads and soabort() requires dropping locks, we must
665 * loop waiting for the condition to be true.
667 while (head->so_incqlen > head->so_qlimit) {
669 sp = TAILQ_FIRST(&head->so_incomp);
670 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
672 sp->so_qstate &= ~SQ_INCOMP;
678 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
679 so->so_qstate |= SQ_INCOMP;
685 wakeup_one(&head->so_timeo);
691 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
695 CURVNET_SET(so->so_vnet);
696 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
702 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
706 CURVNET_SET(so->so_vnet);
707 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
713 * solisten() transitions a socket from a non-listening state to a listening
714 * state, but can also be used to update the listen queue depth on an
715 * existing listen socket. The protocol will call back into the sockets
716 * layer using solisten_proto_check() and solisten_proto() to check and set
717 * socket-layer listen state. Call backs are used so that the protocol can
718 * acquire both protocol and socket layer locks in whatever order is required
721 * Protocol implementors are advised to hold the socket lock across the
722 * socket-layer test and set to avoid races at the socket layer.
725 solisten(struct socket *so, int backlog, struct thread *td)
729 CURVNET_SET(so->so_vnet);
730 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
736 solisten_proto_check(struct socket *so)
739 SOCK_LOCK_ASSERT(so);
741 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
748 solisten_proto(struct socket *so, int backlog)
751 SOCK_LOCK_ASSERT(so);
753 if (backlog < 0 || backlog > somaxconn)
755 so->so_qlimit = backlog;
756 so->so_options |= SO_ACCEPTCONN;
760 * Evaluate the reference count and named references on a socket; if no
761 * references remain, free it. This should be called whenever a reference is
762 * released, such as in sorele(), but also when named reference flags are
763 * cleared in socket or protocol code.
765 * sofree() will free the socket if:
767 * - There are no outstanding file descriptor references or related consumers
770 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
772 * - The protocol does not have an outstanding strong reference on the socket
775 * - The socket is not in a completed connection queue, so a process has been
776 * notified that it is present. If it is removed, the user process may
777 * block in accept() despite select() saying the socket was ready.
780 sofree(struct socket *so)
782 struct protosw *pr = so->so_proto;
785 ACCEPT_LOCK_ASSERT();
786 SOCK_LOCK_ASSERT(so);
788 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
789 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
797 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
798 (so->so_qstate & SQ_INCOMP) != 0,
799 ("sofree: so_head != NULL, but neither SQ_COMP nor "
801 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
802 (so->so_qstate & SQ_INCOMP) == 0,
803 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
804 TAILQ_REMOVE(&head->so_incomp, so, so_list);
806 so->so_qstate &= ~SQ_INCOMP;
809 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
810 (so->so_qstate & SQ_INCOMP) == 0,
811 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
812 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
813 if (so->so_options & SO_ACCEPTCONN) {
814 KASSERT((TAILQ_EMPTY(&so->so_comp)),
815 ("sofree: so_comp populated"));
816 KASSERT((TAILQ_EMPTY(&so->so_incomp)),
817 ("sofree: so_incomp populated"));
823 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
824 (*pr->pr_domain->dom_dispose)(so);
825 if (pr->pr_usrreqs->pru_detach != NULL)
826 (*pr->pr_usrreqs->pru_detach)(so);
829 * From this point on, we assume that no other references to this
830 * socket exist anywhere else in the stack. Therefore, no locks need
831 * to be acquired or held.
833 * We used to do a lot of socket buffer and socket locking here, as
834 * well as invoke sorflush() and perform wakeups. The direct call to
835 * dom_dispose() and sbrelease_internal() are an inlining of what was
836 * necessary from sorflush().
838 * Notice that the socket buffer and kqueue state are torn down
839 * before calling pru_detach. This means that protocols shold not
840 * assume they can perform socket wakeups, etc, in their detach code.
842 sbdestroy(&so->so_snd, so);
843 sbdestroy(&so->so_rcv, so);
844 seldrain(&so->so_snd.sb_sel);
845 seldrain(&so->so_rcv.sb_sel);
846 knlist_destroy(&so->so_rcv.sb_sel.si_note);
847 knlist_destroy(&so->so_snd.sb_sel.si_note);
852 * Close a socket on last file table reference removal. Initiate disconnect
853 * if connected. Free socket when disconnect complete.
855 * This function will sorele() the socket. Note that soclose() may be called
856 * prior to the ref count reaching zero. The actual socket structure will
857 * not be freed until the ref count reaches zero.
860 soclose(struct socket *so)
864 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
866 CURVNET_SET(so->so_vnet);
867 funsetown(&so->so_sigio);
868 if (so->so_state & SS_ISCONNECTED) {
869 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
870 error = sodisconnect(so);
872 if (error == ENOTCONN)
877 if (so->so_options & SO_LINGER) {
878 if ((so->so_state & SS_ISDISCONNECTING) &&
879 (so->so_state & SS_NBIO))
881 while (so->so_state & SS_ISCONNECTED) {
882 error = tsleep(&so->so_timeo,
883 PSOCK | PCATCH, "soclos",
892 if (so->so_proto->pr_usrreqs->pru_close != NULL)
893 (*so->so_proto->pr_usrreqs->pru_close)(so);
895 if (so->so_options & SO_ACCEPTCONN) {
898 * Prevent new additions to the accept queues due
899 * to ACCEPT_LOCK races while we are draining them.
901 so->so_options &= ~SO_ACCEPTCONN;
902 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
903 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
905 sp->so_qstate &= ~SQ_INCOMP;
911 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
912 TAILQ_REMOVE(&so->so_comp, sp, so_list);
914 sp->so_qstate &= ~SQ_COMP;
920 KASSERT((TAILQ_EMPTY(&so->so_comp)),
921 ("%s: so_comp populated", __func__));
922 KASSERT((TAILQ_EMPTY(&so->so_incomp)),
923 ("%s: so_incomp populated", __func__));
926 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
927 so->so_state |= SS_NOFDREF;
928 sorele(so); /* NB: Returns with ACCEPT_UNLOCK(). */
934 * soabort() is used to abruptly tear down a connection, such as when a
935 * resource limit is reached (listen queue depth exceeded), or if a listen
936 * socket is closed while there are sockets waiting to be accepted.
938 * This interface is tricky, because it is called on an unreferenced socket,
939 * and must be called only by a thread that has actually removed the socket
940 * from the listen queue it was on, or races with other threads are risked.
942 * This interface will call into the protocol code, so must not be called
943 * with any socket locks held. Protocols do call it while holding their own
944 * recursible protocol mutexes, but this is something that should be subject
945 * to review in the future.
948 soabort(struct socket *so)
952 * In as much as is possible, assert that no references to this
953 * socket are held. This is not quite the same as asserting that the
954 * current thread is responsible for arranging for no references, but
955 * is as close as we can get for now.
957 KASSERT(so->so_count == 0, ("soabort: so_count"));
958 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
959 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
960 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
961 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
964 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
965 (*so->so_proto->pr_usrreqs->pru_abort)(so);
972 soaccept(struct socket *so, struct sockaddr **nam)
977 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
978 so->so_state &= ~SS_NOFDREF;
981 CURVNET_SET(so->so_vnet);
982 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
988 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
991 return (soconnectat(AT_FDCWD, so, nam, td));
995 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
999 if (so->so_options & SO_ACCEPTCONN)
1000 return (EOPNOTSUPP);
1002 CURVNET_SET(so->so_vnet);
1004 * If protocol is connection-based, can only connect once.
1005 * Otherwise, if connected, try to disconnect first. This allows
1006 * user to disconnect by connecting to, e.g., a null address.
1008 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1009 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1010 (error = sodisconnect(so)))) {
1014 * Prevent accumulated error from previous connection from
1018 if (fd == AT_FDCWD) {
1019 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1022 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1032 soconnect2(struct socket *so1, struct socket *so2)
1036 CURVNET_SET(so1->so_vnet);
1037 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1043 sodisconnect(struct socket *so)
1047 if ((so->so_state & SS_ISCONNECTED) == 0)
1049 if (so->so_state & SS_ISDISCONNECTING)
1052 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1056 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1059 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1060 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1064 int clen = 0, error, dontroute;
1066 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1067 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1068 ("sosend_dgram: !PR_ATOMIC"));
1071 resid = uio->uio_resid;
1073 resid = top->m_pkthdr.len;
1075 * In theory resid should be unsigned. However, space must be
1076 * signed, as it might be less than 0 if we over-committed, and we
1077 * must use a signed comparison of space and resid. On the other
1078 * hand, a negative resid causes us to loop sending 0-length
1079 * segments to the protocol.
1087 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1089 td->td_ru.ru_msgsnd++;
1090 if (control != NULL)
1091 clen = control->m_len;
1093 SOCKBUF_LOCK(&so->so_snd);
1094 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1095 SOCKBUF_UNLOCK(&so->so_snd);
1100 error = so->so_error;
1102 SOCKBUF_UNLOCK(&so->so_snd);
1105 if ((so->so_state & SS_ISCONNECTED) == 0) {
1107 * `sendto' and `sendmsg' is allowed on a connection-based
1108 * socket if it supports implied connect. Return ENOTCONN if
1109 * not connected and no address is supplied.
1111 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1112 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1113 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1114 !(resid == 0 && clen != 0)) {
1115 SOCKBUF_UNLOCK(&so->so_snd);
1119 } else if (addr == NULL) {
1120 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1123 error = EDESTADDRREQ;
1124 SOCKBUF_UNLOCK(&so->so_snd);
1130 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1131 * problem and need fixing.
1133 space = sbspace(&so->so_snd);
1134 if (flags & MSG_OOB)
1137 SOCKBUF_UNLOCK(&so->so_snd);
1138 if (resid > space) {
1144 if (flags & MSG_EOR)
1145 top->m_flags |= M_EOR;
1148 * Copy the data from userland into a mbuf chain.
1149 * If no data is to be copied in, a single empty mbuf
1152 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1153 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1155 error = EFAULT; /* only possible error */
1158 space -= resid - uio->uio_resid;
1159 resid = uio->uio_resid;
1161 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1163 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1168 so->so_options |= SO_DONTROUTE;
1172 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1173 * of date. We could have received a reset packet in an interrupt or
1174 * maybe we slept while doing page faults in uiomove() etc. We could
1175 * probably recheck again inside the locking protection here, but
1176 * there are probably other places that this also happens. We must
1180 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1181 (flags & MSG_OOB) ? PRUS_OOB :
1183 * If the user set MSG_EOF, the protocol understands this flag and
1184 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1186 ((flags & MSG_EOF) &&
1187 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1190 /* If there is more to send set PRUS_MORETOCOME */
1191 (flags & MSG_MORETOCOME) ||
1192 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1193 top, addr, control, td);
1196 so->so_options &= ~SO_DONTROUTE;
1205 if (control != NULL)
1211 * Send on a socket. If send must go all at once and message is larger than
1212 * send buffering, then hard error. Lock against other senders. If must go
1213 * all at once and not enough room now, then inform user that this would
1214 * block and do nothing. Otherwise, if nonblocking, send as much as
1215 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1216 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1217 * in mbuf chain must be small enough to send all at once.
1219 * Returns nonzero on error, timeout or signal; callers must check for short
1220 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1224 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1225 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1229 int clen = 0, error, dontroute;
1230 int atomic = sosendallatonce(so) || top;
1233 resid = uio->uio_resid;
1235 resid = top->m_pkthdr.len;
1237 * In theory resid should be unsigned. However, space must be
1238 * signed, as it might be less than 0 if we over-committed, and we
1239 * must use a signed comparison of space and resid. On the other
1240 * hand, a negative resid causes us to loop sending 0-length
1241 * segments to the protocol.
1243 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1244 * type sockets since that's an error.
1246 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1252 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1253 (so->so_proto->pr_flags & PR_ATOMIC);
1255 td->td_ru.ru_msgsnd++;
1256 if (control != NULL)
1257 clen = control->m_len;
1259 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1265 SOCKBUF_LOCK(&so->so_snd);
1266 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1267 SOCKBUF_UNLOCK(&so->so_snd);
1272 error = so->so_error;
1274 SOCKBUF_UNLOCK(&so->so_snd);
1277 if ((so->so_state & SS_ISCONNECTED) == 0) {
1279 * `sendto' and `sendmsg' is allowed on a connection-
1280 * based socket if it supports implied connect.
1281 * Return ENOTCONN if not connected and no address is
1284 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1285 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1286 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1287 !(resid == 0 && clen != 0)) {
1288 SOCKBUF_UNLOCK(&so->so_snd);
1292 } else if (addr == NULL) {
1293 SOCKBUF_UNLOCK(&so->so_snd);
1294 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1297 error = EDESTADDRREQ;
1301 space = sbspace(&so->so_snd);
1302 if (flags & MSG_OOB)
1304 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1305 clen > so->so_snd.sb_hiwat) {
1306 SOCKBUF_UNLOCK(&so->so_snd);
1310 if (space < resid + clen &&
1311 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1312 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1313 SOCKBUF_UNLOCK(&so->so_snd);
1314 error = EWOULDBLOCK;
1317 error = sbwait(&so->so_snd);
1318 SOCKBUF_UNLOCK(&so->so_snd);
1323 SOCKBUF_UNLOCK(&so->so_snd);
1328 if (flags & MSG_EOR)
1329 top->m_flags |= M_EOR;
1332 * Copy the data from userland into a mbuf
1333 * chain. If resid is 0, which can happen
1334 * only if we have control to send, then
1335 * a single empty mbuf is returned. This
1336 * is a workaround to prevent protocol send
1339 top = m_uiotombuf(uio, M_WAITOK, space,
1340 (atomic ? max_hdr : 0),
1341 (atomic ? M_PKTHDR : 0) |
1342 ((flags & MSG_EOR) ? M_EOR : 0));
1344 error = EFAULT; /* only possible error */
1347 space -= resid - uio->uio_resid;
1348 resid = uio->uio_resid;
1352 so->so_options |= SO_DONTROUTE;
1356 * XXX all the SBS_CANTSENDMORE checks previously
1357 * done could be out of date. We could have received
1358 * a reset packet in an interrupt or maybe we slept
1359 * while doing page faults in uiomove() etc. We
1360 * could probably recheck again inside the locking
1361 * protection here, but there are probably other
1362 * places that this also happens. We must rethink
1366 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1367 (flags & MSG_OOB) ? PRUS_OOB :
1369 * If the user set MSG_EOF, the protocol understands
1370 * this flag and nothing left to send then use
1371 * PRU_SEND_EOF instead of PRU_SEND.
1373 ((flags & MSG_EOF) &&
1374 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1377 /* If there is more to send set PRUS_MORETOCOME. */
1378 (flags & MSG_MORETOCOME) ||
1379 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1380 top, addr, control, td);
1383 so->so_options &= ~SO_DONTROUTE;
1391 } while (resid && space > 0);
1395 sbunlock(&so->so_snd);
1399 if (control != NULL)
1405 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1406 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1410 CURVNET_SET(so->so_vnet);
1411 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1412 control, flags, td);
1418 * The part of soreceive() that implements reading non-inline out-of-band
1419 * data from a socket. For more complete comments, see soreceive(), from
1420 * which this code originated.
1422 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1423 * unable to return an mbuf chain to the caller.
1426 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1428 struct protosw *pr = so->so_proto;
1432 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1435 m = m_get(M_WAITOK, MT_DATA);
1436 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1440 error = uiomove(mtod(m, void *),
1441 (int) min(uio->uio_resid, m->m_len), uio);
1443 } while (uio->uio_resid && error == 0 && m);
1451 * Following replacement or removal of the first mbuf on the first mbuf chain
1452 * of a socket buffer, push necessary state changes back into the socket
1453 * buffer so that other consumers see the values consistently. 'nextrecord'
1454 * is the callers locally stored value of the original value of
1455 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1456 * NOTE: 'nextrecord' may be NULL.
1458 static __inline void
1459 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1462 SOCKBUF_LOCK_ASSERT(sb);
1464 * First, update for the new value of nextrecord. If necessary, make
1465 * it the first record.
1467 if (sb->sb_mb != NULL)
1468 sb->sb_mb->m_nextpkt = nextrecord;
1470 sb->sb_mb = nextrecord;
1473 * Now update any dependent socket buffer fields to reflect the new
1474 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1475 * addition of a second clause that takes care of the case where
1476 * sb_mb has been updated, but remains the last record.
1478 if (sb->sb_mb == NULL) {
1479 sb->sb_mbtail = NULL;
1480 sb->sb_lastrecord = NULL;
1481 } else if (sb->sb_mb->m_nextpkt == NULL)
1482 sb->sb_lastrecord = sb->sb_mb;
1486 * Implement receive operations on a socket. We depend on the way that
1487 * records are added to the sockbuf by sbappend. In particular, each record
1488 * (mbufs linked through m_next) must begin with an address if the protocol
1489 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1490 * data, and then zero or more mbufs of data. In order to allow parallelism
1491 * between network receive and copying to user space, as well as avoid
1492 * sleeping with a mutex held, we release the socket buffer mutex during the
1493 * user space copy. Although the sockbuf is locked, new data may still be
1494 * appended, and thus we must maintain consistency of the sockbuf during that
1497 * The caller may receive the data as a single mbuf chain by supplying an
1498 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1499 * the count in uio_resid.
1502 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1503 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1505 struct mbuf *m, **mp;
1506 int flags, error, offset;
1508 struct protosw *pr = so->so_proto;
1509 struct mbuf *nextrecord;
1511 ssize_t orig_resid = uio->uio_resid;
1516 if (controlp != NULL)
1519 flags = *flagsp &~ MSG_EOR;
1522 if (flags & MSG_OOB)
1523 return (soreceive_rcvoob(so, uio, flags));
1526 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1527 && uio->uio_resid) {
1529 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1532 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1537 SOCKBUF_LOCK(&so->so_rcv);
1538 m = so->so_rcv.sb_mb;
1540 * If we have less data than requested, block awaiting more (subject
1541 * to any timeout) if:
1542 * 1. the current count is less than the low water mark, or
1543 * 2. MSG_DONTWAIT is not set
1545 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1546 sbavail(&so->so_rcv) < uio->uio_resid) &&
1547 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1548 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1549 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1550 ("receive: m == %p sbavail == %u",
1551 m, sbavail(&so->so_rcv)));
1555 error = so->so_error;
1556 if ((flags & MSG_PEEK) == 0)
1558 SOCKBUF_UNLOCK(&so->so_rcv);
1561 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1562 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1564 SOCKBUF_UNLOCK(&so->so_rcv);
1569 for (; m != NULL; m = m->m_next)
1570 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1571 m = so->so_rcv.sb_mb;
1574 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1575 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1576 SOCKBUF_UNLOCK(&so->so_rcv);
1580 if (uio->uio_resid == 0) {
1581 SOCKBUF_UNLOCK(&so->so_rcv);
1584 if ((so->so_state & SS_NBIO) ||
1585 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1586 SOCKBUF_UNLOCK(&so->so_rcv);
1587 error = EWOULDBLOCK;
1590 SBLASTRECORDCHK(&so->so_rcv);
1591 SBLASTMBUFCHK(&so->so_rcv);
1592 error = sbwait(&so->so_rcv);
1593 SOCKBUF_UNLOCK(&so->so_rcv);
1600 * From this point onward, we maintain 'nextrecord' as a cache of the
1601 * pointer to the next record in the socket buffer. We must keep the
1602 * various socket buffer pointers and local stack versions of the
1603 * pointers in sync, pushing out modifications before dropping the
1604 * socket buffer mutex, and re-reading them when picking it up.
1606 * Otherwise, we will race with the network stack appending new data
1607 * or records onto the socket buffer by using inconsistent/stale
1608 * versions of the field, possibly resulting in socket buffer
1611 * By holding the high-level sblock(), we prevent simultaneous
1612 * readers from pulling off the front of the socket buffer.
1614 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1616 uio->uio_td->td_ru.ru_msgrcv++;
1617 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1618 SBLASTRECORDCHK(&so->so_rcv);
1619 SBLASTMBUFCHK(&so->so_rcv);
1620 nextrecord = m->m_nextpkt;
1621 if (pr->pr_flags & PR_ADDR) {
1622 KASSERT(m->m_type == MT_SONAME,
1623 ("m->m_type == %d", m->m_type));
1626 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1628 if (flags & MSG_PEEK) {
1631 sbfree(&so->so_rcv, m);
1632 so->so_rcv.sb_mb = m_free(m);
1633 m = so->so_rcv.sb_mb;
1634 sockbuf_pushsync(&so->so_rcv, nextrecord);
1639 * Process one or more MT_CONTROL mbufs present before any data mbufs
1640 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1641 * just copy the data; if !MSG_PEEK, we call into the protocol to
1642 * perform externalization (or freeing if controlp == NULL).
1644 if (m != NULL && m->m_type == MT_CONTROL) {
1645 struct mbuf *cm = NULL, *cmn;
1646 struct mbuf **cme = &cm;
1649 if (flags & MSG_PEEK) {
1650 if (controlp != NULL) {
1651 *controlp = m_copym(m, 0, m->m_len,
1653 controlp = &(*controlp)->m_next;
1657 sbfree(&so->so_rcv, m);
1658 so->so_rcv.sb_mb = m->m_next;
1661 cme = &(*cme)->m_next;
1662 m = so->so_rcv.sb_mb;
1664 } while (m != NULL && m->m_type == MT_CONTROL);
1665 if ((flags & MSG_PEEK) == 0)
1666 sockbuf_pushsync(&so->so_rcv, nextrecord);
1667 while (cm != NULL) {
1670 if (pr->pr_domain->dom_externalize != NULL) {
1671 SOCKBUF_UNLOCK(&so->so_rcv);
1673 error = (*pr->pr_domain->dom_externalize)
1674 (cm, controlp, flags);
1675 SOCKBUF_LOCK(&so->so_rcv);
1676 } else if (controlp != NULL)
1680 if (controlp != NULL) {
1682 while (*controlp != NULL)
1683 controlp = &(*controlp)->m_next;
1688 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1690 nextrecord = so->so_rcv.sb_mb;
1694 if ((flags & MSG_PEEK) == 0) {
1695 KASSERT(m->m_nextpkt == nextrecord,
1696 ("soreceive: post-control, nextrecord !sync"));
1697 if (nextrecord == NULL) {
1698 KASSERT(so->so_rcv.sb_mb == m,
1699 ("soreceive: post-control, sb_mb!=m"));
1700 KASSERT(so->so_rcv.sb_lastrecord == m,
1701 ("soreceive: post-control, lastrecord!=m"));
1705 if (type == MT_OOBDATA)
1708 if ((flags & MSG_PEEK) == 0) {
1709 KASSERT(so->so_rcv.sb_mb == nextrecord,
1710 ("soreceive: sb_mb != nextrecord"));
1711 if (so->so_rcv.sb_mb == NULL) {
1712 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1713 ("soreceive: sb_lastercord != NULL"));
1717 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1718 SBLASTRECORDCHK(&so->so_rcv);
1719 SBLASTMBUFCHK(&so->so_rcv);
1722 * Now continue to read any data mbufs off of the head of the socket
1723 * buffer until the read request is satisfied. Note that 'type' is
1724 * used to store the type of any mbuf reads that have happened so far
1725 * such that soreceive() can stop reading if the type changes, which
1726 * causes soreceive() to return only one of regular data and inline
1727 * out-of-band data in a single socket receive operation.
1731 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
1734 * If the type of mbuf has changed since the last mbuf
1735 * examined ('type'), end the receive operation.
1737 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1738 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1739 if (type != m->m_type)
1741 } else if (type == MT_OOBDATA)
1744 KASSERT(m->m_type == MT_DATA,
1745 ("m->m_type == %d", m->m_type));
1746 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1747 len = uio->uio_resid;
1748 if (so->so_oobmark && len > so->so_oobmark - offset)
1749 len = so->so_oobmark - offset;
1750 if (len > m->m_len - moff)
1751 len = m->m_len - moff;
1753 * If mp is set, just pass back the mbufs. Otherwise copy
1754 * them out via the uio, then free. Sockbuf must be
1755 * consistent here (points to current mbuf, it points to next
1756 * record) when we drop priority; we must note any additions
1757 * to the sockbuf when we block interrupts again.
1760 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1761 SBLASTRECORDCHK(&so->so_rcv);
1762 SBLASTMBUFCHK(&so->so_rcv);
1763 SOCKBUF_UNLOCK(&so->so_rcv);
1764 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1765 SOCKBUF_LOCK(&so->so_rcv);
1768 * The MT_SONAME mbuf has already been removed
1769 * from the record, so it is necessary to
1770 * remove the data mbufs, if any, to preserve
1771 * the invariant in the case of PR_ADDR that
1772 * requires MT_SONAME mbufs at the head of
1775 if (m && pr->pr_flags & PR_ATOMIC &&
1776 ((flags & MSG_PEEK) == 0))
1777 (void)sbdroprecord_locked(&so->so_rcv);
1778 SOCKBUF_UNLOCK(&so->so_rcv);
1782 uio->uio_resid -= len;
1783 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1784 if (len == m->m_len - moff) {
1785 if (m->m_flags & M_EOR)
1787 if (flags & MSG_PEEK) {
1791 nextrecord = m->m_nextpkt;
1792 sbfree(&so->so_rcv, m);
1794 m->m_nextpkt = NULL;
1797 so->so_rcv.sb_mb = m = m->m_next;
1800 so->so_rcv.sb_mb = m_free(m);
1801 m = so->so_rcv.sb_mb;
1803 sockbuf_pushsync(&so->so_rcv, nextrecord);
1804 SBLASTRECORDCHK(&so->so_rcv);
1805 SBLASTMBUFCHK(&so->so_rcv);
1808 if (flags & MSG_PEEK)
1812 if (flags & MSG_DONTWAIT) {
1813 *mp = m_copym(m, 0, len,
1817 * m_copym() couldn't
1819 * Adjust uio_resid back
1821 * down by len bytes,
1822 * which we didn't end
1823 * up "copying" over).
1825 uio->uio_resid += len;
1829 SOCKBUF_UNLOCK(&so->so_rcv);
1830 *mp = m_copym(m, 0, len,
1832 SOCKBUF_LOCK(&so->so_rcv);
1835 sbcut_locked(&so->so_rcv, len);
1838 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1839 if (so->so_oobmark) {
1840 if ((flags & MSG_PEEK) == 0) {
1841 so->so_oobmark -= len;
1842 if (so->so_oobmark == 0) {
1843 so->so_rcv.sb_state |= SBS_RCVATMARK;
1848 if (offset == so->so_oobmark)
1852 if (flags & MSG_EOR)
1855 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1856 * must not quit until "uio->uio_resid == 0" or an error
1857 * termination. If a signal/timeout occurs, return with a
1858 * short count but without error. Keep sockbuf locked
1859 * against other readers.
1861 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1862 !sosendallatonce(so) && nextrecord == NULL) {
1863 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1865 so->so_rcv.sb_state & SBS_CANTRCVMORE)
1868 * Notify the protocol that some data has been
1869 * drained before blocking.
1871 if (pr->pr_flags & PR_WANTRCVD) {
1872 SOCKBUF_UNLOCK(&so->so_rcv);
1874 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1875 SOCKBUF_LOCK(&so->so_rcv);
1877 SBLASTRECORDCHK(&so->so_rcv);
1878 SBLASTMBUFCHK(&so->so_rcv);
1880 * We could receive some data while was notifying
1881 * the protocol. Skip blocking in this case.
1883 if (so->so_rcv.sb_mb == NULL) {
1884 error = sbwait(&so->so_rcv);
1886 SOCKBUF_UNLOCK(&so->so_rcv);
1890 m = so->so_rcv.sb_mb;
1892 nextrecord = m->m_nextpkt;
1896 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1897 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1899 if ((flags & MSG_PEEK) == 0)
1900 (void) sbdroprecord_locked(&so->so_rcv);
1902 if ((flags & MSG_PEEK) == 0) {
1905 * First part is an inline SB_EMPTY_FIXUP(). Second
1906 * part makes sure sb_lastrecord is up-to-date if
1907 * there is still data in the socket buffer.
1909 so->so_rcv.sb_mb = nextrecord;
1910 if (so->so_rcv.sb_mb == NULL) {
1911 so->so_rcv.sb_mbtail = NULL;
1912 so->so_rcv.sb_lastrecord = NULL;
1913 } else if (nextrecord->m_nextpkt == NULL)
1914 so->so_rcv.sb_lastrecord = nextrecord;
1916 SBLASTRECORDCHK(&so->so_rcv);
1917 SBLASTMBUFCHK(&so->so_rcv);
1919 * If soreceive() is being done from the socket callback,
1920 * then don't need to generate ACK to peer to update window,
1921 * since ACK will be generated on return to TCP.
1923 if (!(flags & MSG_SOCALLBCK) &&
1924 (pr->pr_flags & PR_WANTRCVD)) {
1925 SOCKBUF_UNLOCK(&so->so_rcv);
1927 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1928 SOCKBUF_LOCK(&so->so_rcv);
1931 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1932 if (orig_resid == uio->uio_resid && orig_resid &&
1933 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1934 SOCKBUF_UNLOCK(&so->so_rcv);
1937 SOCKBUF_UNLOCK(&so->so_rcv);
1942 sbunlock(&so->so_rcv);
1947 * Optimized version of soreceive() for stream (TCP) sockets.
1948 * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled.
1951 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1952 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1954 int len = 0, error = 0, flags, oresid;
1956 struct mbuf *m, *n = NULL;
1958 /* We only do stream sockets. */
1959 if (so->so_type != SOCK_STREAM)
1963 if (controlp != NULL)
1966 flags = *flagsp &~ MSG_EOR;
1969 if (flags & MSG_OOB)
1970 return (soreceive_rcvoob(so, uio, flags));
1976 /* Prevent other readers from entering the socket. */
1977 error = sblock(sb, SBLOCKWAIT(flags));
1982 /* Easy one, no space to copyout anything. */
1983 if (uio->uio_resid == 0) {
1987 oresid = uio->uio_resid;
1989 /* We will never ever get anything unless we are or were connected. */
1990 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1996 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1998 /* Abort if socket has reported problems. */
2000 if (sbavail(sb) > 0)
2002 if (oresid > uio->uio_resid)
2004 error = so->so_error;
2005 if (!(flags & MSG_PEEK))
2010 /* Door is closed. Deliver what is left, if any. */
2011 if (sb->sb_state & SBS_CANTRCVMORE) {
2012 if (sbavail(sb) > 0)
2018 /* Socket buffer is empty and we shall not block. */
2019 if (sbavail(sb) == 0 &&
2020 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2025 /* Socket buffer got some data that we shall deliver now. */
2026 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2027 ((so->so_state & SS_NBIO) ||
2028 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2029 sbavail(sb) >= sb->sb_lowat ||
2030 sbavail(sb) >= uio->uio_resid ||
2031 sbavail(sb) >= sb->sb_hiwat) ) {
2035 /* On MSG_WAITALL we must wait until all data or error arrives. */
2036 if ((flags & MSG_WAITALL) &&
2037 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2041 * Wait and block until (more) data comes in.
2042 * NB: Drops the sockbuf lock during wait.
2050 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2051 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2052 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2056 uio->uio_td->td_ru.ru_msgrcv++;
2058 /* Fill uio until full or current end of socket buffer is reached. */
2059 len = min(uio->uio_resid, sbavail(sb));
2061 /* Dequeue as many mbufs as possible. */
2062 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2066 m_cat(*mp0, sb->sb_mb);
2068 m != NULL && m->m_len <= len;
2070 KASSERT(!(m->m_flags & M_NOTAVAIL),
2071 ("%s: m %p not available", __func__, m));
2073 uio->uio_resid -= m->m_len;
2079 sb->sb_lastrecord = sb->sb_mb;
2080 if (sb->sb_mb == NULL)
2083 /* Copy the remainder. */
2085 KASSERT(sb->sb_mb != NULL,
2086 ("%s: len > 0 && sb->sb_mb empty", __func__));
2088 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2090 len = 0; /* Don't flush data from sockbuf. */
2092 uio->uio_resid -= len;
2103 /* NB: Must unlock socket buffer as uiomove may sleep. */
2105 error = m_mbuftouio(uio, sb->sb_mb, len);
2110 SBLASTRECORDCHK(sb);
2114 * Remove the delivered data from the socket buffer unless we
2115 * were only peeking.
2117 if (!(flags & MSG_PEEK)) {
2119 sbdrop_locked(sb, len);
2121 /* Notify protocol that we drained some data. */
2122 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2123 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2124 !(flags & MSG_SOCALLBCK))) {
2127 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2133 * For MSG_WAITALL we may have to loop again and wait for
2134 * more data to come in.
2136 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2139 SOCKBUF_LOCK_ASSERT(sb);
2140 SBLASTRECORDCHK(sb);
2148 * Optimized version of soreceive() for simple datagram cases from userspace.
2149 * Unlike in the stream case, we're able to drop a datagram if copyout()
2150 * fails, and because we handle datagrams atomically, we don't need to use a
2151 * sleep lock to prevent I/O interlacing.
2154 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2155 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2157 struct mbuf *m, *m2;
2160 struct protosw *pr = so->so_proto;
2161 struct mbuf *nextrecord;
2165 if (controlp != NULL)
2168 flags = *flagsp &~ MSG_EOR;
2173 * For any complicated cases, fall back to the full
2174 * soreceive_generic().
2176 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2177 return (soreceive_generic(so, psa, uio, mp0, controlp,
2181 * Enforce restrictions on use.
2183 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2184 ("soreceive_dgram: wantrcvd"));
2185 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2186 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2187 ("soreceive_dgram: SBS_RCVATMARK"));
2188 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2189 ("soreceive_dgram: P_CONNREQUIRED"));
2192 * Loop blocking while waiting for a datagram.
2194 SOCKBUF_LOCK(&so->so_rcv);
2195 while ((m = so->so_rcv.sb_mb) == NULL) {
2196 KASSERT(sbavail(&so->so_rcv) == 0,
2197 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2198 sbavail(&so->so_rcv)));
2200 error = so->so_error;
2202 SOCKBUF_UNLOCK(&so->so_rcv);
2205 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2206 uio->uio_resid == 0) {
2207 SOCKBUF_UNLOCK(&so->so_rcv);
2210 if ((so->so_state & SS_NBIO) ||
2211 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2212 SOCKBUF_UNLOCK(&so->so_rcv);
2213 return (EWOULDBLOCK);
2215 SBLASTRECORDCHK(&so->so_rcv);
2216 SBLASTMBUFCHK(&so->so_rcv);
2217 error = sbwait(&so->so_rcv);
2219 SOCKBUF_UNLOCK(&so->so_rcv);
2223 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2226 uio->uio_td->td_ru.ru_msgrcv++;
2227 SBLASTRECORDCHK(&so->so_rcv);
2228 SBLASTMBUFCHK(&so->so_rcv);
2229 nextrecord = m->m_nextpkt;
2230 if (nextrecord == NULL) {
2231 KASSERT(so->so_rcv.sb_lastrecord == m,
2232 ("soreceive_dgram: lastrecord != m"));
2235 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2236 ("soreceive_dgram: m_nextpkt != nextrecord"));
2239 * Pull 'm' and its chain off the front of the packet queue.
2241 so->so_rcv.sb_mb = NULL;
2242 sockbuf_pushsync(&so->so_rcv, nextrecord);
2245 * Walk 'm's chain and free that many bytes from the socket buffer.
2247 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2248 sbfree(&so->so_rcv, m2);
2251 * Do a few last checks before we let go of the lock.
2253 SBLASTRECORDCHK(&so->so_rcv);
2254 SBLASTMBUFCHK(&so->so_rcv);
2255 SOCKBUF_UNLOCK(&so->so_rcv);
2257 if (pr->pr_flags & PR_ADDR) {
2258 KASSERT(m->m_type == MT_SONAME,
2259 ("m->m_type == %d", m->m_type));
2261 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2266 /* XXXRW: Can this happen? */
2271 * Packet to copyout() is now in 'm' and it is disconnected from the
2274 * Process one or more MT_CONTROL mbufs present before any data mbufs
2275 * in the first mbuf chain on the socket buffer. We call into the
2276 * protocol to perform externalization (or freeing if controlp ==
2277 * NULL). In some cases there can be only MT_CONTROL mbufs without
2280 if (m->m_type == MT_CONTROL) {
2281 struct mbuf *cm = NULL, *cmn;
2282 struct mbuf **cme = &cm;
2288 cme = &(*cme)->m_next;
2290 } while (m != NULL && m->m_type == MT_CONTROL);
2291 while (cm != NULL) {
2294 if (pr->pr_domain->dom_externalize != NULL) {
2295 error = (*pr->pr_domain->dom_externalize)
2296 (cm, controlp, flags);
2297 } else if (controlp != NULL)
2301 if (controlp != NULL) {
2302 while (*controlp != NULL)
2303 controlp = &(*controlp)->m_next;
2308 KASSERT(m == NULL || m->m_type == MT_DATA,
2309 ("soreceive_dgram: !data"));
2310 while (m != NULL && uio->uio_resid > 0) {
2311 len = uio->uio_resid;
2314 error = uiomove(mtod(m, char *), (int)len, uio);
2319 if (len == m->m_len)
2336 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2337 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2341 CURVNET_SET(so->so_vnet);
2342 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2349 soshutdown(struct socket *so, int how)
2351 struct protosw *pr = so->so_proto;
2354 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2357 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0)
2360 CURVNET_SET(so->so_vnet);
2361 if (pr->pr_usrreqs->pru_flush != NULL)
2362 (*pr->pr_usrreqs->pru_flush)(so, how);
2365 if (how != SHUT_RD) {
2366 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2367 wakeup(&so->so_timeo);
2371 wakeup(&so->so_timeo);
2377 sorflush(struct socket *so)
2379 struct sockbuf *sb = &so->so_rcv;
2380 struct protosw *pr = so->so_proto;
2386 * In order to avoid calling dom_dispose with the socket buffer mutex
2387 * held, and in order to generally avoid holding the lock for a long
2388 * time, we make a copy of the socket buffer and clear the original
2389 * (except locks, state). The new socket buffer copy won't have
2390 * initialized locks so we can only call routines that won't use or
2391 * assert those locks.
2393 * Dislodge threads currently blocked in receive and wait to acquire
2394 * a lock against other simultaneous readers before clearing the
2395 * socket buffer. Don't let our acquire be interrupted by a signal
2396 * despite any existing socket disposition on interruptable waiting.
2399 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2402 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2403 * and mutex data unchanged.
2406 bzero(&aso, sizeof(aso));
2407 aso.so_pcb = so->so_pcb;
2408 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2409 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2410 bzero(&sb->sb_startzero,
2411 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2416 * Dispose of special rights and flush the copied socket. Don't call
2417 * any unsafe routines (that rely on locks being initialized) on aso.
2419 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2420 (*pr->pr_domain->dom_dispose)(&aso);
2421 sbrelease_internal(&aso.so_rcv, so);
2425 * Wrapper for Socket established helper hook.
2426 * Parameters: socket, context of the hook point, hook id.
2429 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2431 struct socket_hhook_data hhook_data = {
2438 CURVNET_SET(so->so_vnet);
2439 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2442 /* Ugly but needed, since hhooks return void for now */
2443 return (hhook_data.status);
2447 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2448 * additional variant to handle the case where the option value needs to be
2449 * some kind of integer, but not a specific size. In addition to their use
2450 * here, these functions are also called by the protocol-level pr_ctloutput()
2454 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2459 * If the user gives us more than we wanted, we ignore it, but if we
2460 * don't get the minimum length the caller wants, we return EINVAL.
2461 * On success, sopt->sopt_valsize is set to however much we actually
2464 if ((valsize = sopt->sopt_valsize) < minlen)
2467 sopt->sopt_valsize = valsize = len;
2469 if (sopt->sopt_td != NULL)
2470 return (copyin(sopt->sopt_val, buf, valsize));
2472 bcopy(sopt->sopt_val, buf, valsize);
2477 * Kernel version of setsockopt(2).
2479 * XXX: optlen is size_t, not socklen_t
2482 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2485 struct sockopt sopt;
2487 sopt.sopt_level = level;
2488 sopt.sopt_name = optname;
2489 sopt.sopt_dir = SOPT_SET;
2490 sopt.sopt_val = optval;
2491 sopt.sopt_valsize = optlen;
2492 sopt.sopt_td = NULL;
2493 return (sosetopt(so, &sopt));
2497 sosetopt(struct socket *so, struct sockopt *sopt)
2508 CURVNET_SET(so->so_vnet);
2510 if (sopt->sopt_level != SOL_SOCKET) {
2511 if (so->so_proto->pr_ctloutput != NULL) {
2512 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2516 error = ENOPROTOOPT;
2518 switch (sopt->sopt_name) {
2519 case SO_ACCEPTFILTER:
2520 error = do_setopt_accept_filter(so, sopt);
2526 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2531 so->so_linger = l.l_linger;
2533 so->so_options |= SO_LINGER;
2535 so->so_options &= ~SO_LINGER;
2542 case SO_USELOOPBACK:
2552 error = sooptcopyin(sopt, &optval, sizeof optval,
2558 so->so_options |= sopt->sopt_name;
2560 so->so_options &= ~sopt->sopt_name;
2565 error = sooptcopyin(sopt, &optval, sizeof optval,
2570 if (optval < 0 || optval >= rt_numfibs) {
2574 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2575 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2576 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2577 so->so_fibnum = optval;
2582 case SO_USER_COOKIE:
2583 error = sooptcopyin(sopt, &val32, sizeof val32,
2587 so->so_user_cookie = val32;
2594 error = sooptcopyin(sopt, &optval, sizeof optval,
2600 * Values < 1 make no sense for any of these options,
2608 switch (sopt->sopt_name) {
2611 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2612 &so->so_snd : &so->so_rcv, (u_long)optval,
2613 so, curthread) == 0) {
2617 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2618 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2622 * Make sure the low-water is never greater than the
2626 SOCKBUF_LOCK(&so->so_snd);
2627 so->so_snd.sb_lowat =
2628 (optval > so->so_snd.sb_hiwat) ?
2629 so->so_snd.sb_hiwat : optval;
2630 SOCKBUF_UNLOCK(&so->so_snd);
2633 SOCKBUF_LOCK(&so->so_rcv);
2634 so->so_rcv.sb_lowat =
2635 (optval > so->so_rcv.sb_hiwat) ?
2636 so->so_rcv.sb_hiwat : optval;
2637 SOCKBUF_UNLOCK(&so->so_rcv);
2644 #ifdef COMPAT_FREEBSD32
2645 if (SV_CURPROC_FLAG(SV_ILP32)) {
2646 struct timeval32 tv32;
2648 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2650 CP(tv32, tv, tv_sec);
2651 CP(tv32, tv, tv_usec);
2654 error = sooptcopyin(sopt, &tv, sizeof tv,
2658 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2659 tv.tv_usec >= 1000000) {
2663 if (tv.tv_sec > INT32_MAX)
2667 switch (sopt->sopt_name) {
2669 so->so_snd.sb_timeo = val;
2672 so->so_rcv.sb_timeo = val;
2679 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2683 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2691 error = sooptcopyin(sopt, &optval, sizeof optval,
2695 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
2699 so->so_ts_clock = optval;
2702 case SO_MAX_PACING_RATE:
2703 error = sooptcopyin(sopt, &val32, sizeof(val32),
2707 so->so_max_pacing_rate = val32;
2711 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2712 error = hhook_run_socket(so, sopt,
2715 error = ENOPROTOOPT;
2718 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2719 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
2727 * Helper routine for getsockopt.
2730 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2738 * Documented get behavior is that we always return a value, possibly
2739 * truncated to fit in the user's buffer. Traditional behavior is
2740 * that we always tell the user precisely how much we copied, rather
2741 * than something useful like the total amount we had available for
2742 * her. Note that this interface is not idempotent; the entire
2743 * answer must be generated ahead of time.
2745 valsize = min(len, sopt->sopt_valsize);
2746 sopt->sopt_valsize = valsize;
2747 if (sopt->sopt_val != NULL) {
2748 if (sopt->sopt_td != NULL)
2749 error = copyout(buf, sopt->sopt_val, valsize);
2751 bcopy(buf, sopt->sopt_val, valsize);
2757 sogetopt(struct socket *so, struct sockopt *sopt)
2766 CURVNET_SET(so->so_vnet);
2768 if (sopt->sopt_level != SOL_SOCKET) {
2769 if (so->so_proto->pr_ctloutput != NULL)
2770 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2772 error = ENOPROTOOPT;
2776 switch (sopt->sopt_name) {
2777 case SO_ACCEPTFILTER:
2778 error = do_getopt_accept_filter(so, sopt);
2783 l.l_onoff = so->so_options & SO_LINGER;
2784 l.l_linger = so->so_linger;
2786 error = sooptcopyout(sopt, &l, sizeof l);
2789 case SO_USELOOPBACK:
2801 optval = so->so_options & sopt->sopt_name;
2803 error = sooptcopyout(sopt, &optval, sizeof optval);
2807 optval = so->so_type;
2811 optval = so->so_proto->pr_protocol;
2816 optval = so->so_error;
2822 optval = so->so_snd.sb_hiwat;
2826 optval = so->so_rcv.sb_hiwat;
2830 optval = so->so_snd.sb_lowat;
2834 optval = so->so_rcv.sb_lowat;
2839 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
2840 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2841 #ifdef COMPAT_FREEBSD32
2842 if (SV_CURPROC_FLAG(SV_ILP32)) {
2843 struct timeval32 tv32;
2845 CP(tv, tv32, tv_sec);
2846 CP(tv, tv32, tv_usec);
2847 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2850 error = sooptcopyout(sopt, &tv, sizeof tv);
2855 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2859 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2863 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2871 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2875 error = mac_getsockopt_peerlabel(
2876 sopt->sopt_td->td_ucred, so, &extmac);
2879 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2885 case SO_LISTENQLIMIT:
2886 optval = so->so_qlimit;
2890 optval = so->so_qlen;
2893 case SO_LISTENINCQLEN:
2894 optval = so->so_incqlen;
2898 optval = so->so_ts_clock;
2901 case SO_MAX_PACING_RATE:
2902 optval = so->so_max_pacing_rate;
2906 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2907 error = hhook_run_socket(so, sopt,
2910 error = ENOPROTOOPT;
2922 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2924 struct mbuf *m, *m_prev;
2925 int sopt_size = sopt->sopt_valsize;
2927 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2930 if (sopt_size > MLEN) {
2931 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
2932 if ((m->m_flags & M_EXT) == 0) {
2936 m->m_len = min(MCLBYTES, sopt_size);
2938 m->m_len = min(MLEN, sopt_size);
2940 sopt_size -= m->m_len;
2945 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2950 if (sopt_size > MLEN) {
2951 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
2953 if ((m->m_flags & M_EXT) == 0) {
2958 m->m_len = min(MCLBYTES, sopt_size);
2960 m->m_len = min(MLEN, sopt_size);
2962 sopt_size -= m->m_len;
2970 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2972 struct mbuf *m0 = m;
2974 if (sopt->sopt_val == NULL)
2976 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2977 if (sopt->sopt_td != NULL) {
2980 error = copyin(sopt->sopt_val, mtod(m, char *),
2987 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2988 sopt->sopt_valsize -= m->m_len;
2989 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2992 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2993 panic("ip6_sooptmcopyin");
2998 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3000 struct mbuf *m0 = m;
3003 if (sopt->sopt_val == NULL)
3005 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3006 if (sopt->sopt_td != NULL) {
3009 error = copyout(mtod(m, char *), sopt->sopt_val,
3016 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3017 sopt->sopt_valsize -= m->m_len;
3018 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3019 valsize += m->m_len;
3023 /* enough soopt buffer should be given from user-land */
3027 sopt->sopt_valsize = valsize;
3032 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3033 * out-of-band data, which will then notify socket consumers.
3036 sohasoutofband(struct socket *so)
3039 if (so->so_sigio != NULL)
3040 pgsigio(&so->so_sigio, SIGURG, 0);
3041 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
3045 sopoll(struct socket *so, int events, struct ucred *active_cred,
3050 * We do not need to set or assert curvnet as long as everyone uses
3053 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3058 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3063 SOCKBUF_LOCK(&so->so_snd);
3064 SOCKBUF_LOCK(&so->so_rcv);
3065 if (events & (POLLIN | POLLRDNORM))
3066 if (soreadabledata(so))
3067 revents |= events & (POLLIN | POLLRDNORM);
3069 if (events & (POLLOUT | POLLWRNORM))
3070 if (sowriteable(so))
3071 revents |= events & (POLLOUT | POLLWRNORM);
3073 if (events & (POLLPRI | POLLRDBAND))
3074 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
3075 revents |= events & (POLLPRI | POLLRDBAND);
3077 if ((events & POLLINIGNEOF) == 0) {
3078 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3079 revents |= events & (POLLIN | POLLRDNORM);
3080 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3086 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3087 selrecord(td, &so->so_rcv.sb_sel);
3088 so->so_rcv.sb_flags |= SB_SEL;
3091 if (events & (POLLOUT | POLLWRNORM)) {
3092 selrecord(td, &so->so_snd.sb_sel);
3093 so->so_snd.sb_flags |= SB_SEL;
3097 SOCKBUF_UNLOCK(&so->so_rcv);
3098 SOCKBUF_UNLOCK(&so->so_snd);
3103 soo_kqfilter(struct file *fp, struct knote *kn)
3105 struct socket *so = kn->kn_fp->f_data;
3108 switch (kn->kn_filter) {
3110 if (so->so_options & SO_ACCEPTCONN)
3111 kn->kn_fop = &solisten_filtops;
3113 kn->kn_fop = &soread_filtops;
3117 kn->kn_fop = &sowrite_filtops;
3121 kn->kn_fop = &soempty_filtops;
3129 knlist_add(&sb->sb_sel.si_note, kn, 1);
3130 sb->sb_flags |= SB_KNOTE;
3136 * Some routines that return EOPNOTSUPP for entry points that are not
3137 * supported by a protocol. Fill in as needed.
3140 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3147 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3154 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3161 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3168 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3176 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3183 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3191 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3198 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3199 struct ifnet *ifp, struct thread *td)
3206 pru_disconnect_notsupp(struct socket *so)
3213 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3220 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3227 pru_rcvd_notsupp(struct socket *so, int flags)
3234 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3241 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3242 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3249 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3252 return (EOPNOTSUPP);
3256 * This isn't really a ``null'' operation, but it's the default one and
3257 * doesn't do anything destructive.
3260 pru_sense_null(struct socket *so, struct stat *sb)
3263 sb->st_blksize = so->so_snd.sb_hiwat;
3268 pru_shutdown_notsupp(struct socket *so)
3275 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3282 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3283 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3290 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3291 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3298 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3306 filt_sordetach(struct knote *kn)
3308 struct socket *so = kn->kn_fp->f_data;
3310 SOCKBUF_LOCK(&so->so_rcv);
3311 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3312 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3313 so->so_rcv.sb_flags &= ~SB_KNOTE;
3314 SOCKBUF_UNLOCK(&so->so_rcv);
3319 filt_soread(struct knote *kn, long hint)
3323 so = kn->kn_fp->f_data;
3324 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3326 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3327 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3328 kn->kn_flags |= EV_EOF;
3329 kn->kn_fflags = so->so_error;
3331 } else if (so->so_error) /* temporary udp error */
3334 if (kn->kn_sfflags & NOTE_LOWAT) {
3335 if (kn->kn_data >= kn->kn_sdata)
3338 if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3342 /* This hook returning non-zero indicates an event, not error */
3343 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3347 filt_sowdetach(struct knote *kn)
3349 struct socket *so = kn->kn_fp->f_data;
3351 SOCKBUF_LOCK(&so->so_snd);
3352 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3353 if (knlist_empty(&so->so_snd.sb_sel.si_note))
3354 so->so_snd.sb_flags &= ~SB_KNOTE;
3355 SOCKBUF_UNLOCK(&so->so_snd);
3360 filt_sowrite(struct knote *kn, long hint)
3364 so = kn->kn_fp->f_data;
3365 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3366 kn->kn_data = sbspace(&so->so_snd);
3368 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3370 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3371 kn->kn_flags |= EV_EOF;
3372 kn->kn_fflags = so->so_error;
3374 } else if (so->so_error) /* temporary udp error */
3376 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3377 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3379 else if (kn->kn_sfflags & NOTE_LOWAT)
3380 return (kn->kn_data >= kn->kn_sdata);
3382 return (kn->kn_data >= so->so_snd.sb_lowat);
3386 filt_soempty(struct knote *kn, long hint)
3390 so = kn->kn_fp->f_data;
3391 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3392 kn->kn_data = sbused(&so->so_snd);
3394 if (kn->kn_data == 0)
3402 filt_solisten(struct knote *kn, long hint)
3404 struct socket *so = kn->kn_fp->f_data;
3406 kn->kn_data = so->so_qlen;
3407 return (!TAILQ_EMPTY(&so->so_comp));
3411 socheckuid(struct socket *so, uid_t uid)
3416 if (so->so_cred->cr_uid != uid)
3422 * These functions are used by protocols to notify the socket layer (and its
3423 * consumers) of state changes in the sockets driven by protocol-side events.
3427 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3429 * Normal sequence from the active (originating) side is that
3430 * soisconnecting() is called during processing of connect() call, resulting
3431 * in an eventual call to soisconnected() if/when the connection is
3432 * established. When the connection is torn down soisdisconnecting() is
3433 * called during processing of disconnect() call, and soisdisconnected() is
3434 * called when the connection to the peer is totally severed. The semantics
3435 * of these routines are such that connectionless protocols can call
3436 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3437 * calls when setting up a ``connection'' takes no time.
3439 * From the passive side, a socket is created with two queues of sockets:
3440 * so_incomp for connections in progress and so_comp for connections already
3441 * made and awaiting user acceptance. As a protocol is preparing incoming
3442 * connections, it creates a socket structure queued on so_incomp by calling
3443 * sonewconn(). When the connection is established, soisconnected() is
3444 * called, and transfers the socket structure to so_comp, making it available
3447 * If a socket is closed with sockets on either so_incomp or so_comp, these
3448 * sockets are dropped.
3450 * If higher-level protocols are implemented in the kernel, the wakeups done
3451 * here will sometimes cause software-interrupt process scheduling.
3454 soisconnecting(struct socket *so)
3458 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3459 so->so_state |= SS_ISCONNECTING;
3464 soisconnected(struct socket *so)
3466 struct socket *head;
3472 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3473 so->so_state |= SS_ISCONNECTED;
3475 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3476 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3478 TAILQ_REMOVE(&head->so_incomp, so, so_list);
3480 so->so_qstate &= ~SQ_INCOMP;
3481 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3483 so->so_qstate |= SQ_COMP;
3486 wakeup_one(&head->so_timeo);
3489 soupcall_set(so, SO_RCV,
3490 head->so_accf->so_accept_filter->accf_callback,
3491 head->so_accf->so_accept_filter_arg);
3492 so->so_options &= ~SO_ACCEPTFILTER;
3493 ret = head->so_accf->so_accept_filter->accf_callback(so,
3494 head->so_accf->so_accept_filter_arg, M_NOWAIT);
3495 if (ret == SU_ISCONNECTED)
3496 soupcall_clear(so, SO_RCV);
3498 if (ret == SU_ISCONNECTED)
3505 wakeup(&so->so_timeo);
3511 soisdisconnecting(struct socket *so)
3515 * Note: This code assumes that SOCK_LOCK(so) and
3516 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3518 SOCKBUF_LOCK(&so->so_rcv);
3519 so->so_state &= ~SS_ISCONNECTING;
3520 so->so_state |= SS_ISDISCONNECTING;
3521 socantrcvmore_locked(so);
3522 SOCKBUF_LOCK(&so->so_snd);
3523 socantsendmore_locked(so);
3524 wakeup(&so->so_timeo);
3528 soisdisconnected(struct socket *so)
3532 * Note: This code assumes that SOCK_LOCK(so) and
3533 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3535 SOCKBUF_LOCK(&so->so_rcv);
3536 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3537 so->so_state |= SS_ISDISCONNECTED;
3538 socantrcvmore_locked(so);
3539 SOCKBUF_LOCK(&so->so_snd);
3540 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
3541 socantsendmore_locked(so);
3542 wakeup(&so->so_timeo);
3546 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3549 sodupsockaddr(const struct sockaddr *sa, int mflags)
3551 struct sockaddr *sa2;
3553 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3555 bcopy(sa, sa2, sa->sa_len);
3560 * Register per-socket buffer upcalls.
3563 soupcall_set(struct socket *so, int which,
3564 int (*func)(struct socket *, void *, int), void *arg)
3576 panic("soupcall_set: bad which");
3578 SOCKBUF_LOCK_ASSERT(sb);
3580 /* XXX: accf_http actually wants to do this on purpose. */
3581 KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3583 sb->sb_upcall = func;
3584 sb->sb_upcallarg = arg;
3585 sb->sb_flags |= SB_UPCALL;
3589 soupcall_clear(struct socket *so, int which)
3601 panic("soupcall_clear: bad which");
3603 SOCKBUF_LOCK_ASSERT(sb);
3604 KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3605 sb->sb_upcall = NULL;
3606 sb->sb_upcallarg = NULL;
3607 sb->sb_flags &= ~SB_UPCALL;
3611 * Create an external-format (``xsocket'') structure using the information in
3612 * the kernel-format socket structure pointed to by so. This is done to
3613 * reduce the spew of irrelevant information over this interface, to isolate
3614 * user code from changes in the kernel structure, and potentially to provide
3615 * information-hiding if we decide that some of this information should be
3616 * hidden from users.
3619 sotoxsocket(struct socket *so, struct xsocket *xso)
3622 xso->xso_len = sizeof *xso;
3624 xso->so_type = so->so_type;
3625 xso->so_options = so->so_options;
3626 xso->so_linger = so->so_linger;
3627 xso->so_state = so->so_state;
3628 xso->so_pcb = so->so_pcb;
3629 xso->xso_protocol = so->so_proto->pr_protocol;
3630 xso->xso_family = so->so_proto->pr_domain->dom_family;
3631 xso->so_qlen = so->so_qlen;
3632 xso->so_incqlen = so->so_incqlen;
3633 xso->so_qlimit = so->so_qlimit;
3634 xso->so_timeo = so->so_timeo;
3635 xso->so_error = so->so_error;
3636 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3637 xso->so_oobmark = so->so_oobmark;
3638 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3639 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3640 xso->so_uid = so->so_cred->cr_uid;
3645 * Socket accessor functions to provide external consumers with
3646 * a safe interface to socket state
3651 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *),
3655 TAILQ_FOREACH(so, &so->so_comp, so_list)
3660 so_sockbuf_rcv(struct socket *so)
3663 return (&so->so_rcv);
3667 so_sockbuf_snd(struct socket *so)
3670 return (&so->so_snd);
3674 so_state_get(const struct socket *so)
3677 return (so->so_state);
3681 so_state_set(struct socket *so, int val)
3688 so_options_get(const struct socket *so)
3691 return (so->so_options);
3695 so_options_set(struct socket *so, int val)
3698 so->so_options = val;
3702 so_error_get(const struct socket *so)
3705 return (so->so_error);
3709 so_error_set(struct socket *so, int val)
3716 so_linger_get(const struct socket *so)
3719 return (so->so_linger);
3723 so_linger_set(struct socket *so, int val)
3726 so->so_linger = val;
3730 so_protosw_get(const struct socket *so)
3733 return (so->so_proto);
3737 so_protosw_set(struct socket *so, struct protosw *val)
3744 so_sorwakeup(struct socket *so)
3751 so_sowwakeup(struct socket *so)
3758 so_sorwakeup_locked(struct socket *so)
3761 sorwakeup_locked(so);
3765 so_sowwakeup_locked(struct socket *so)
3768 sowwakeup_locked(so);
3772 so_lock(struct socket *so)
3779 so_unlock(struct socket *so)