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"
109 #include "opt_sctp.h"
111 #include <sys/param.h>
112 #include <sys/systm.h>
113 #include <sys/fcntl.h>
114 #include <sys/limits.h>
115 #include <sys/lock.h>
117 #include <sys/malloc.h>
118 #include <sys/mbuf.h>
119 #include <sys/mutex.h>
120 #include <sys/domain.h>
121 #include <sys/file.h> /* for struct knote */
122 #include <sys/hhook.h>
123 #include <sys/kernel.h>
124 #include <sys/khelp.h>
125 #include <sys/event.h>
126 #include <sys/eventhandler.h>
127 #include <sys/poll.h>
128 #include <sys/proc.h>
129 #include <sys/protosw.h>
130 #include <sys/socket.h>
131 #include <sys/socketvar.h>
132 #include <sys/resourcevar.h>
133 #include <net/route.h>
134 #include <sys/signalvar.h>
135 #include <sys/stat.h>
137 #include <sys/sysctl.h>
138 #include <sys/taskqueue.h>
140 #include <sys/jail.h>
141 #include <sys/syslog.h>
142 #include <netinet/in.h>
144 #include <net/vnet.h>
146 #include <security/mac/mac_framework.h>
150 #ifdef COMPAT_FREEBSD32
151 #include <sys/mount.h>
152 #include <sys/sysent.h>
153 #include <compat/freebsd32/freebsd32.h>
156 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
158 static void so_rdknl_lock(void *);
159 static void so_rdknl_unlock(void *);
160 static void so_rdknl_assert_locked(void *);
161 static void so_rdknl_assert_unlocked(void *);
162 static void so_wrknl_lock(void *);
163 static void so_wrknl_unlock(void *);
164 static void so_wrknl_assert_locked(void *);
165 static void so_wrknl_assert_unlocked(void *);
167 static void filt_sordetach(struct knote *kn);
168 static int filt_soread(struct knote *kn, long hint);
169 static void filt_sowdetach(struct knote *kn);
170 static int filt_sowrite(struct knote *kn, long hint);
171 static int filt_soempty(struct knote *kn, long hint);
172 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
173 fo_kqfilter_t soo_kqfilter;
175 static struct filterops soread_filtops = {
177 .f_detach = filt_sordetach,
178 .f_event = filt_soread,
180 static struct filterops sowrite_filtops = {
182 .f_detach = filt_sowdetach,
183 .f_event = filt_sowrite,
185 static struct filterops soempty_filtops = {
187 .f_detach = filt_sowdetach,
188 .f_event = filt_soempty,
191 so_gen_t so_gencnt; /* generation count for sockets */
193 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
194 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
196 #define VNET_SO_ASSERT(so) \
197 VNET_ASSERT(curvnet != NULL, \
198 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
200 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
201 #define V_socket_hhh VNET(socket_hhh)
204 * Limit on the number of connections in the listen queue waiting
206 * NB: The original sysctl somaxconn is still available but hidden
207 * to prevent confusion about the actual purpose of this number.
209 static u_int somaxconn = SOMAXCONN;
212 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
218 error = sysctl_handle_int(oidp, &val, 0, req);
219 if (error || !req->newptr )
223 * The purpose of the UINT_MAX / 3 limit, is so that the formula
225 * below, will not overflow.
228 if (val < 1 || val > UINT_MAX / 3)
234 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW,
235 0, sizeof(int), sysctl_somaxconn, "I",
236 "Maximum listen socket pending connection accept queue size");
237 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
238 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP,
239 0, sizeof(int), sysctl_somaxconn, "I",
240 "Maximum listen socket pending connection accept queue size (compat)");
242 static int numopensockets;
243 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
244 &numopensockets, 0, "Number of open sockets");
247 * accept_mtx locks down per-socket fields relating to accept queues. See
248 * socketvar.h for an annotation of the protected fields of struct socket.
250 struct mtx accept_mtx;
251 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
254 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
257 static struct mtx so_global_mtx;
258 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
261 * General IPC sysctl name space, used by sockets and a variety of other IPC
264 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
267 * Initialize the socket subsystem and set up the socket
270 static uma_zone_t socket_zone;
274 socket_zone_change(void *tag)
277 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
281 socket_hhook_register(int subtype)
284 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
285 &V_socket_hhh[subtype],
286 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
287 printf("%s: WARNING: unable to register hook\n", __func__);
291 socket_hhook_deregister(int subtype)
294 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
295 printf("%s: WARNING: unable to deregister hook\n", __func__);
299 socket_init(void *tag)
302 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
303 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
304 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
305 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
306 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
307 EVENTHANDLER_PRI_FIRST);
309 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
312 socket_vnet_init(const void *unused __unused)
316 /* We expect a contiguous range */
317 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
318 socket_hhook_register(i);
320 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
321 socket_vnet_init, NULL);
324 socket_vnet_uninit(const void *unused __unused)
328 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
329 socket_hhook_deregister(i);
331 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
332 socket_vnet_uninit, NULL);
335 * Initialise maxsockets. This SYSINIT must be run after
339 init_maxsockets(void *ignored)
342 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
343 maxsockets = imax(maxsockets, maxfiles);
345 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
348 * Sysctl to get and set the maximum global sockets limit. Notify protocols
349 * of the change so that they can update their dependent limits as required.
352 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
354 int error, newmaxsockets;
356 newmaxsockets = maxsockets;
357 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
358 if (error == 0 && req->newptr) {
359 if (newmaxsockets > maxsockets &&
360 newmaxsockets <= maxfiles) {
361 maxsockets = newmaxsockets;
362 EVENTHANDLER_INVOKE(maxsockets_change);
368 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
369 &maxsockets, 0, sysctl_maxsockets, "IU",
370 "Maximum number of sockets available");
373 * Socket operation routines. These routines are called by the routines in
374 * sys_socket.c or from a system process, and implement the semantics of
375 * socket operations by switching out to the protocol specific routines.
379 * Get a socket structure from our zone, and initialize it. Note that it
380 * would probably be better to allocate socket and PCB at the same time, but
381 * I'm not convinced that all the protocols can be easily modified to do
384 * soalloc() returns a socket with a ref count of 0.
386 static struct socket *
387 soalloc(struct vnet *vnet)
391 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
395 if (mac_socket_init(so, M_NOWAIT) != 0) {
396 uma_zfree(socket_zone, so);
400 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
401 uma_zfree(socket_zone, so);
406 * The socket locking protocol allows to lock 2 sockets at a time,
407 * however, the first one must be a listening socket. WITNESS lacks
408 * a feature to change class of an existing lock, so we use DUPOK.
410 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
411 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
412 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
413 so->so_rcv.sb_sel = &so->so_rdsel;
414 so->so_snd.sb_sel = &so->so_wrsel;
415 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
416 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
417 TAILQ_INIT(&so->so_snd.sb_aiojobq);
418 TAILQ_INIT(&so->so_rcv.sb_aiojobq);
419 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
420 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
422 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
423 __func__, __LINE__, so));
426 /* We shouldn't need the so_global_mtx */
427 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
428 /* Do we need more comprehensive error returns? */
429 uma_zfree(socket_zone, so);
432 mtx_lock(&so_global_mtx);
433 so->so_gencnt = ++so_gencnt;
436 vnet->vnet_sockcnt++;
438 mtx_unlock(&so_global_mtx);
444 * Free the storage associated with a socket at the socket layer, tear down
445 * locks, labels, etc. All protocol state is assumed already to have been
446 * torn down (and possibly never set up) by the caller.
449 sodealloc(struct socket *so)
452 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
453 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
455 mtx_lock(&so_global_mtx);
456 so->so_gencnt = ++so_gencnt;
457 --numopensockets; /* Could be below, but faster here. */
459 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
460 __func__, __LINE__, so));
461 so->so_vnet->vnet_sockcnt--;
463 mtx_unlock(&so_global_mtx);
464 if (so->so_rcv.sb_hiwat)
465 (void)chgsbsize(so->so_cred->cr_uidinfo,
466 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
467 if (so->so_snd.sb_hiwat)
468 (void)chgsbsize(so->so_cred->cr_uidinfo,
469 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
471 mac_socket_destroy(so);
473 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
476 khelp_destroy_osd(&so->osd);
477 if (SOLISTENING(so)) {
478 if (so->sol_accept_filter != NULL)
479 accept_filt_setopt(so, NULL);
481 sx_destroy(&so->so_snd.sb_sx);
482 sx_destroy(&so->so_rcv.sb_sx);
483 SOCKBUF_LOCK_DESTROY(&so->so_snd);
484 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
486 mtx_destroy(&so->so_lock);
487 uma_zfree(socket_zone, so);
491 * socreate returns a socket with a ref count of 1. The socket should be
492 * closed with soclose().
495 socreate(int dom, struct socket **aso, int type, int proto,
496 struct ucred *cred, struct thread *td)
503 prp = pffindproto(dom, proto, type);
505 prp = pffindtype(dom, type);
508 /* No support for domain. */
509 if (pffinddomain(dom) == NULL)
510 return (EAFNOSUPPORT);
511 /* No support for socket type. */
512 if (proto == 0 && type != 0)
514 return (EPROTONOSUPPORT);
516 if (prp->pr_usrreqs->pru_attach == NULL ||
517 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
518 return (EPROTONOSUPPORT);
520 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
521 return (EPROTONOSUPPORT);
523 if (prp->pr_type != type)
525 so = soalloc(CRED_TO_VNET(cred));
530 so->so_cred = crhold(cred);
531 if ((prp->pr_domain->dom_family == PF_INET) ||
532 (prp->pr_domain->dom_family == PF_INET6) ||
533 (prp->pr_domain->dom_family == PF_ROUTE))
534 so->so_fibnum = td->td_proc->p_fibnum;
539 mac_socket_create(cred, so);
541 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
542 so_rdknl_assert_locked, so_rdknl_assert_unlocked);
543 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
544 so_wrknl_assert_locked, so_wrknl_assert_unlocked);
546 * Auto-sizing of socket buffers is managed by the protocols and
547 * the appropriate flags must be set in the pru_attach function.
549 CURVNET_SET(so->so_vnet);
550 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
562 static int regression_sonewconn_earlytest = 1;
563 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
564 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
568 * When an attempt at a new connection is noted on a socket which accepts
569 * connections, sonewconn is called. If the connection is possible (subject
570 * to space constraints, etc.) then we allocate a new structure, properly
571 * linked into the data structure of the original socket, and return this.
572 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
574 * Note: the ref count on the socket is 0 on return.
577 sonewconn(struct socket *head, int connstatus)
579 static struct timeval lastover;
580 static struct timeval overinterval = { 60, 0 };
581 static int overcount;
587 over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
588 SOLISTEN_UNLOCK(head);
590 if (regression_sonewconn_earlytest && over) {
596 if (ratecheck(&lastover, &overinterval)) {
597 log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
598 "%i already in queue awaiting acceptance "
599 "(%d occurrences)\n",
600 __func__, head->so_pcb, head->sol_qlen, overcount);
607 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
609 so = soalloc(head->so_vnet);
611 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
612 "limit reached or out of memory\n",
613 __func__, head->so_pcb);
616 so->so_listen = head;
617 so->so_type = head->so_type;
618 so->so_linger = head->so_linger;
619 so->so_state = head->so_state | SS_NOFDREF;
620 so->so_fibnum = head->so_fibnum;
621 so->so_proto = head->so_proto;
622 so->so_cred = crhold(head->so_cred);
624 mac_socket_newconn(head, so);
626 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
627 so_rdknl_assert_locked, so_rdknl_assert_unlocked);
628 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
629 so_wrknl_assert_locked, so_wrknl_assert_unlocked);
630 VNET_SO_ASSERT(head);
631 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
633 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
634 __func__, head->so_pcb);
637 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
639 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
640 __func__, head->so_pcb);
643 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
644 so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
645 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
646 so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
647 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
648 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
651 if (head->sol_accept_filter != NULL)
653 so->so_state |= connstatus;
654 so->so_options = head->so_options & ~SO_ACCEPTCONN;
655 soref(head); /* A socket on (in)complete queue refs head. */
657 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
658 so->so_qstate = SQ_COMP;
660 solisten_wakeup(head); /* unlocks */
663 * Keep removing sockets from the head until there's room for
664 * us to insert on the tail. In pre-locking revisions, this
665 * was a simple if(), but as we could be racing with other
666 * threads and soabort() requires dropping locks, we must
667 * loop waiting for the condition to be true.
669 while (head->sol_incqlen > head->sol_qlimit) {
672 sp = TAILQ_FIRST(&head->sol_incomp);
673 TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
676 sp->so_qstate = SQ_NONE;
677 sp->so_listen = NULL;
679 sorele(head); /* does SOLISTEN_UNLOCK, head stays */
683 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
684 so->so_qstate = SQ_INCOMP;
686 SOLISTEN_UNLOCK(head);
693 * Socket part of sctp_peeloff(). Detach a new socket from an
694 * association. The new socket is returned with a reference.
697 sopeeloff(struct socket *head)
701 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
702 __func__, __LINE__, head));
703 so = soalloc(head->so_vnet);
705 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
706 "limit reached or out of memory\n",
707 __func__, head->so_pcb);
710 so->so_type = head->so_type;
711 so->so_options = head->so_options;
712 so->so_linger = head->so_linger;
713 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
714 so->so_fibnum = head->so_fibnum;
715 so->so_proto = head->so_proto;
716 so->so_cred = crhold(head->so_cred);
718 mac_socket_newconn(head, so);
720 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
721 so_rdknl_assert_locked, so_rdknl_assert_unlocked);
722 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
723 so_wrknl_assert_locked, so_wrknl_assert_unlocked);
724 VNET_SO_ASSERT(head);
725 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
727 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
728 __func__, head->so_pcb);
731 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
733 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
734 __func__, head->so_pcb);
737 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
738 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
739 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
740 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
741 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
742 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
751 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
755 CURVNET_SET(so->so_vnet);
756 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
762 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
766 CURVNET_SET(so->so_vnet);
767 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
773 * solisten() transitions a socket from a non-listening state to a listening
774 * state, but can also be used to update the listen queue depth on an
775 * existing listen socket. The protocol will call back into the sockets
776 * layer using solisten_proto_check() and solisten_proto() to check and set
777 * socket-layer listen state. Call backs are used so that the protocol can
778 * acquire both protocol and socket layer locks in whatever order is required
781 * Protocol implementors are advised to hold the socket lock across the
782 * socket-layer test and set to avoid races at the socket layer.
785 solisten(struct socket *so, int backlog, struct thread *td)
789 CURVNET_SET(so->so_vnet);
790 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
796 solisten_proto_check(struct socket *so)
799 SOCK_LOCK_ASSERT(so);
801 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
808 solisten_proto(struct socket *so, int backlog)
810 int sbrcv_lowat, sbsnd_lowat;
811 u_int sbrcv_hiwat, sbsnd_hiwat;
812 short sbrcv_flags, sbsnd_flags;
813 sbintime_t sbrcv_timeo, sbsnd_timeo;
815 SOCK_LOCK_ASSERT(so);
821 * Change this socket to listening state.
823 sbrcv_lowat = so->so_rcv.sb_lowat;
824 sbsnd_lowat = so->so_snd.sb_lowat;
825 sbrcv_hiwat = so->so_rcv.sb_hiwat;
826 sbsnd_hiwat = so->so_snd.sb_hiwat;
827 sbrcv_flags = so->so_rcv.sb_flags;
828 sbsnd_flags = so->so_snd.sb_flags;
829 sbrcv_timeo = so->so_rcv.sb_timeo;
830 sbsnd_timeo = so->so_snd.sb_timeo;
832 sbdestroy(&so->so_snd, so);
833 sbdestroy(&so->so_rcv, so);
834 sx_destroy(&so->so_snd.sb_sx);
835 sx_destroy(&so->so_rcv.sb_sx);
836 SOCKBUF_LOCK_DESTROY(&so->so_snd);
837 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
841 sizeof(struct socket) - offsetof(struct socket, so_rcv));
844 so->sol_sbrcv_lowat = sbrcv_lowat;
845 so->sol_sbsnd_lowat = sbsnd_lowat;
846 so->sol_sbrcv_hiwat = sbrcv_hiwat;
847 so->sol_sbsnd_hiwat = sbsnd_hiwat;
848 so->sol_sbrcv_flags = sbrcv_flags;
849 so->sol_sbsnd_flags = sbsnd_flags;
850 so->sol_sbrcv_timeo = sbrcv_timeo;
851 so->sol_sbsnd_timeo = sbsnd_timeo;
853 so->sol_qlen = so->sol_incqlen = 0;
854 TAILQ_INIT(&so->sol_incomp);
855 TAILQ_INIT(&so->sol_comp);
857 so->sol_accept_filter = NULL;
858 so->sol_accept_filter_arg = NULL;
859 so->sol_accept_filter_str = NULL;
861 so->so_options |= SO_ACCEPTCONN;
864 if (backlog < 0 || backlog > somaxconn)
866 so->sol_qlimit = backlog;
870 * Wakeup listeners/subsystems once we have a complete connection.
871 * Enters with lock, returns unlocked.
874 solisten_wakeup(struct socket *sol)
877 if (sol->sol_upcall != NULL)
878 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
880 selwakeuppri(&sol->so_rdsel, PSOCK);
881 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
883 SOLISTEN_UNLOCK(sol);
884 wakeup_one(&sol->sol_comp);
888 * Return single connection off a listening socket queue. Main consumer of
889 * the function is kern_accept4(). Some modules, that do their own accept
890 * management also use the function.
892 * Listening socket must be locked on entry and is returned unlocked on
894 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
897 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
902 SOLISTEN_LOCK_ASSERT(head);
904 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
905 head->so_error == 0) {
906 error = msleep(&head->sol_comp, &head->so_lock, PSOCK | PCATCH,
909 SOLISTEN_UNLOCK(head);
913 if (head->so_error) {
914 error = head->so_error;
916 SOLISTEN_UNLOCK(head);
919 if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp)) {
920 SOLISTEN_UNLOCK(head);
921 return (EWOULDBLOCK);
923 so = TAILQ_FIRST(&head->sol_comp);
925 KASSERT(so->so_qstate == SQ_COMP,
926 ("%s: so %p not SQ_COMP", __func__, so));
929 so->so_qstate = SQ_NONE;
930 so->so_listen = NULL;
931 TAILQ_REMOVE(&head->sol_comp, so, so_list);
932 if (flags & ACCEPT4_INHERIT)
933 so->so_state |= (head->so_state & SS_NBIO);
935 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
944 * Evaluate the reference count and named references on a socket; if no
945 * references remain, free it. This should be called whenever a reference is
946 * released, such as in sorele(), but also when named reference flags are
947 * cleared in socket or protocol code.
949 * sofree() will free the socket if:
951 * - There are no outstanding file descriptor references or related consumers
954 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
956 * - The protocol does not have an outstanding strong reference on the socket
959 * - The socket is not in a completed connection queue, so a process has been
960 * notified that it is present. If it is removed, the user process may
961 * block in accept() despite select() saying the socket was ready.
964 sofree(struct socket *so)
966 struct protosw *pr = so->so_proto;
968 SOCK_LOCK_ASSERT(so);
970 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
971 (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) {
976 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
980 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
983 * To solve race between close of a listening socket and
984 * a socket on its incomplete queue, we need to lock both.
985 * The order is first listening socket, then regular.
986 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
987 * function and the listening socket are the only pointers
988 * to so. To preserve so and sol, we reference both and then
990 * After relock the socket may not move to so_comp since it
991 * doesn't have PCB already, but it may be removed from
992 * so_incomp. If that happens, we share responsiblity on
993 * freeing the socket, but soclose() has already removed
1001 if (so->so_qstate == SQ_INCOMP) {
1002 KASSERT(so->so_listen == sol,
1003 ("%s: so %p migrated out of sol %p",
1004 __func__, so, sol));
1005 TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1007 /* This is guarenteed not to be the last. */
1008 refcount_release(&sol->so_count);
1009 so->so_qstate = SQ_NONE;
1010 so->so_listen = NULL;
1012 KASSERT(so->so_listen == NULL,
1013 ("%s: so %p not on (in)comp with so_listen",
1016 KASSERT(so->so_count == 1,
1017 ("%s: so %p count %u", __func__, so, so->so_count));
1020 if (SOLISTENING(so))
1021 so->so_error = ECONNABORTED;
1025 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1026 (*pr->pr_domain->dom_dispose)(so);
1027 if (pr->pr_usrreqs->pru_detach != NULL)
1028 (*pr->pr_usrreqs->pru_detach)(so);
1031 * From this point on, we assume that no other references to this
1032 * socket exist anywhere else in the stack. Therefore, no locks need
1033 * to be acquired or held.
1035 * We used to do a lot of socket buffer and socket locking here, as
1036 * well as invoke sorflush() and perform wakeups. The direct call to
1037 * dom_dispose() and sbrelease_internal() are an inlining of what was
1038 * necessary from sorflush().
1040 * Notice that the socket buffer and kqueue state are torn down
1041 * before calling pru_detach. This means that protocols shold not
1042 * assume they can perform socket wakeups, etc, in their detach code.
1044 if (!SOLISTENING(so)) {
1045 sbdestroy(&so->so_snd, so);
1046 sbdestroy(&so->so_rcv, so);
1048 seldrain(&so->so_rdsel);
1049 seldrain(&so->so_wrsel);
1050 knlist_destroy(&so->so_rdsel.si_note);
1051 knlist_destroy(&so->so_wrsel.si_note);
1056 * Close a socket on last file table reference removal. Initiate disconnect
1057 * if connected. Free socket when disconnect complete.
1059 * This function will sorele() the socket. Note that soclose() may be called
1060 * prior to the ref count reaching zero. The actual socket structure will
1061 * not be freed until the ref count reaches zero.
1064 soclose(struct socket *so)
1066 struct accept_queue lqueue;
1070 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1072 CURVNET_SET(so->so_vnet);
1073 funsetown(&so->so_sigio);
1074 if (so->so_state & SS_ISCONNECTED) {
1075 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1076 error = sodisconnect(so);
1078 if (error == ENOTCONN)
1083 if (so->so_options & SO_LINGER) {
1084 if ((so->so_state & SS_ISDISCONNECTING) &&
1085 (so->so_state & SS_NBIO))
1087 while (so->so_state & SS_ISCONNECTED) {
1088 error = tsleep(&so->so_timeo,
1089 PSOCK | PCATCH, "soclos",
1090 so->so_linger * hz);
1098 if (so->so_proto->pr_usrreqs->pru_close != NULL)
1099 (*so->so_proto->pr_usrreqs->pru_close)(so);
1102 if ((listening = (so->so_options & SO_ACCEPTCONN))) {
1105 TAILQ_INIT(&lqueue);
1106 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1107 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1109 so->sol_qlen = so->sol_incqlen = 0;
1111 TAILQ_FOREACH(sp, &lqueue, so_list) {
1113 sp->so_qstate = SQ_NONE;
1114 sp->so_listen = NULL;
1116 /* Guaranteed not to be the last. */
1117 refcount_release(&so->so_count);
1120 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1121 so->so_state |= SS_NOFDREF;
1126 TAILQ_FOREACH(sp, &lqueue, so_list) {
1128 if (sp->so_count == 0) {
1132 /* sp is now in sofree() */
1141 * soabort() is used to abruptly tear down a connection, such as when a
1142 * resource limit is reached (listen queue depth exceeded), or if a listen
1143 * socket is closed while there are sockets waiting to be accepted.
1145 * This interface is tricky, because it is called on an unreferenced socket,
1146 * and must be called only by a thread that has actually removed the socket
1147 * from the listen queue it was on, or races with other threads are risked.
1149 * This interface will call into the protocol code, so must not be called
1150 * with any socket locks held. Protocols do call it while holding their own
1151 * recursible protocol mutexes, but this is something that should be subject
1152 * to review in the future.
1155 soabort(struct socket *so)
1159 * In as much as is possible, assert that no references to this
1160 * socket are held. This is not quite the same as asserting that the
1161 * current thread is responsible for arranging for no references, but
1162 * is as close as we can get for now.
1164 KASSERT(so->so_count == 0, ("soabort: so_count"));
1165 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1166 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1167 KASSERT(so->so_qstate == SQ_NONE, ("soabort: !SQ_NONE"));
1170 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1171 (*so->so_proto->pr_usrreqs->pru_abort)(so);
1177 soaccept(struct socket *so, struct sockaddr **nam)
1182 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1183 so->so_state &= ~SS_NOFDREF;
1186 CURVNET_SET(so->so_vnet);
1187 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1193 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1196 return (soconnectat(AT_FDCWD, so, nam, td));
1200 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1204 if (so->so_options & SO_ACCEPTCONN)
1205 return (EOPNOTSUPP);
1207 CURVNET_SET(so->so_vnet);
1209 * If protocol is connection-based, can only connect once.
1210 * Otherwise, if connected, try to disconnect first. This allows
1211 * user to disconnect by connecting to, e.g., a null address.
1213 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1214 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1215 (error = sodisconnect(so)))) {
1219 * Prevent accumulated error from previous connection from
1223 if (fd == AT_FDCWD) {
1224 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1227 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1237 soconnect2(struct socket *so1, struct socket *so2)
1241 CURVNET_SET(so1->so_vnet);
1242 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1248 sodisconnect(struct socket *so)
1252 if ((so->so_state & SS_ISCONNECTED) == 0)
1254 if (so->so_state & SS_ISDISCONNECTING)
1257 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1261 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1264 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1265 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1269 int clen = 0, error, dontroute;
1271 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1272 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1273 ("sosend_dgram: !PR_ATOMIC"));
1276 resid = uio->uio_resid;
1278 resid = top->m_pkthdr.len;
1280 * In theory resid should be unsigned. However, space must be
1281 * signed, as it might be less than 0 if we over-committed, and we
1282 * must use a signed comparison of space and resid. On the other
1283 * hand, a negative resid causes us to loop sending 0-length
1284 * segments to the protocol.
1292 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1294 td->td_ru.ru_msgsnd++;
1295 if (control != NULL)
1296 clen = control->m_len;
1298 SOCKBUF_LOCK(&so->so_snd);
1299 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1300 SOCKBUF_UNLOCK(&so->so_snd);
1305 error = so->so_error;
1307 SOCKBUF_UNLOCK(&so->so_snd);
1310 if ((so->so_state & SS_ISCONNECTED) == 0) {
1312 * `sendto' and `sendmsg' is allowed on a connection-based
1313 * socket if it supports implied connect. Return ENOTCONN if
1314 * not connected and no address is supplied.
1316 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1317 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1318 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1319 !(resid == 0 && clen != 0)) {
1320 SOCKBUF_UNLOCK(&so->so_snd);
1324 } else if (addr == NULL) {
1325 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1328 error = EDESTADDRREQ;
1329 SOCKBUF_UNLOCK(&so->so_snd);
1335 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1336 * problem and need fixing.
1338 space = sbspace(&so->so_snd);
1339 if (flags & MSG_OOB)
1342 SOCKBUF_UNLOCK(&so->so_snd);
1343 if (resid > space) {
1349 if (flags & MSG_EOR)
1350 top->m_flags |= M_EOR;
1353 * Copy the data from userland into a mbuf chain.
1354 * If no data is to be copied in, a single empty mbuf
1357 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1358 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1360 error = EFAULT; /* only possible error */
1363 space -= resid - uio->uio_resid;
1364 resid = uio->uio_resid;
1366 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1368 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1373 so->so_options |= SO_DONTROUTE;
1377 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1378 * of date. We could have received a reset packet in an interrupt or
1379 * maybe we slept while doing page faults in uiomove() etc. We could
1380 * probably recheck again inside the locking protection here, but
1381 * there are probably other places that this also happens. We must
1385 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1386 (flags & MSG_OOB) ? PRUS_OOB :
1388 * If the user set MSG_EOF, the protocol understands this flag and
1389 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1391 ((flags & MSG_EOF) &&
1392 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1395 /* If there is more to send set PRUS_MORETOCOME */
1396 (flags & MSG_MORETOCOME) ||
1397 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1398 top, addr, control, td);
1401 so->so_options &= ~SO_DONTROUTE;
1410 if (control != NULL)
1416 * Send on a socket. If send must go all at once and message is larger than
1417 * send buffering, then hard error. Lock against other senders. If must go
1418 * all at once and not enough room now, then inform user that this would
1419 * block and do nothing. Otherwise, if nonblocking, send as much as
1420 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1421 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1422 * in mbuf chain must be small enough to send all at once.
1424 * Returns nonzero on error, timeout or signal; callers must check for short
1425 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1429 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1430 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1434 int clen = 0, error, dontroute;
1435 int atomic = sosendallatonce(so) || top;
1438 resid = uio->uio_resid;
1440 resid = top->m_pkthdr.len;
1442 * In theory resid should be unsigned. However, space must be
1443 * signed, as it might be less than 0 if we over-committed, and we
1444 * must use a signed comparison of space and resid. On the other
1445 * hand, a negative resid causes us to loop sending 0-length
1446 * segments to the protocol.
1448 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1449 * type sockets since that's an error.
1451 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1457 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1458 (so->so_proto->pr_flags & PR_ATOMIC);
1460 td->td_ru.ru_msgsnd++;
1461 if (control != NULL)
1462 clen = control->m_len;
1464 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1470 SOCKBUF_LOCK(&so->so_snd);
1471 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1472 SOCKBUF_UNLOCK(&so->so_snd);
1477 error = so->so_error;
1479 SOCKBUF_UNLOCK(&so->so_snd);
1482 if ((so->so_state & SS_ISCONNECTED) == 0) {
1484 * `sendto' and `sendmsg' is allowed on a connection-
1485 * based socket if it supports implied connect.
1486 * Return ENOTCONN if not connected and no address is
1489 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1490 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1491 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1492 !(resid == 0 && clen != 0)) {
1493 SOCKBUF_UNLOCK(&so->so_snd);
1497 } else if (addr == NULL) {
1498 SOCKBUF_UNLOCK(&so->so_snd);
1499 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1502 error = EDESTADDRREQ;
1506 space = sbspace(&so->so_snd);
1507 if (flags & MSG_OOB)
1509 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1510 clen > so->so_snd.sb_hiwat) {
1511 SOCKBUF_UNLOCK(&so->so_snd);
1515 if (space < resid + clen &&
1516 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1517 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1518 SOCKBUF_UNLOCK(&so->so_snd);
1519 error = EWOULDBLOCK;
1522 error = sbwait(&so->so_snd);
1523 SOCKBUF_UNLOCK(&so->so_snd);
1528 SOCKBUF_UNLOCK(&so->so_snd);
1533 if (flags & MSG_EOR)
1534 top->m_flags |= M_EOR;
1537 * Copy the data from userland into a mbuf
1538 * chain. If resid is 0, which can happen
1539 * only if we have control to send, then
1540 * a single empty mbuf is returned. This
1541 * is a workaround to prevent protocol send
1544 top = m_uiotombuf(uio, M_WAITOK, space,
1545 (atomic ? max_hdr : 0),
1546 (atomic ? M_PKTHDR : 0) |
1547 ((flags & MSG_EOR) ? M_EOR : 0));
1549 error = EFAULT; /* only possible error */
1552 space -= resid - uio->uio_resid;
1553 resid = uio->uio_resid;
1557 so->so_options |= SO_DONTROUTE;
1561 * XXX all the SBS_CANTSENDMORE checks previously
1562 * done could be out of date. We could have received
1563 * a reset packet in an interrupt or maybe we slept
1564 * while doing page faults in uiomove() etc. We
1565 * could probably recheck again inside the locking
1566 * protection here, but there are probably other
1567 * places that this also happens. We must rethink
1571 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1572 (flags & MSG_OOB) ? PRUS_OOB :
1574 * If the user set MSG_EOF, the protocol understands
1575 * this flag and nothing left to send then use
1576 * PRU_SEND_EOF instead of PRU_SEND.
1578 ((flags & MSG_EOF) &&
1579 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1582 /* If there is more to send set PRUS_MORETOCOME. */
1583 (flags & MSG_MORETOCOME) ||
1584 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1585 top, addr, control, td);
1588 so->so_options &= ~SO_DONTROUTE;
1596 } while (resid && space > 0);
1600 sbunlock(&so->so_snd);
1604 if (control != NULL)
1610 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1611 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1615 CURVNET_SET(so->so_vnet);
1616 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1617 control, flags, td);
1623 * The part of soreceive() that implements reading non-inline out-of-band
1624 * data from a socket. For more complete comments, see soreceive(), from
1625 * which this code originated.
1627 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1628 * unable to return an mbuf chain to the caller.
1631 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1633 struct protosw *pr = so->so_proto;
1637 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1640 m = m_get(M_WAITOK, MT_DATA);
1641 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1645 error = uiomove(mtod(m, void *),
1646 (int) min(uio->uio_resid, m->m_len), uio);
1648 } while (uio->uio_resid && error == 0 && m);
1656 * Following replacement or removal of the first mbuf on the first mbuf chain
1657 * of a socket buffer, push necessary state changes back into the socket
1658 * buffer so that other consumers see the values consistently. 'nextrecord'
1659 * is the callers locally stored value of the original value of
1660 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1661 * NOTE: 'nextrecord' may be NULL.
1663 static __inline void
1664 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1667 SOCKBUF_LOCK_ASSERT(sb);
1669 * First, update for the new value of nextrecord. If necessary, make
1670 * it the first record.
1672 if (sb->sb_mb != NULL)
1673 sb->sb_mb->m_nextpkt = nextrecord;
1675 sb->sb_mb = nextrecord;
1678 * Now update any dependent socket buffer fields to reflect the new
1679 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1680 * addition of a second clause that takes care of the case where
1681 * sb_mb has been updated, but remains the last record.
1683 if (sb->sb_mb == NULL) {
1684 sb->sb_mbtail = NULL;
1685 sb->sb_lastrecord = NULL;
1686 } else if (sb->sb_mb->m_nextpkt == NULL)
1687 sb->sb_lastrecord = sb->sb_mb;
1691 * Implement receive operations on a socket. We depend on the way that
1692 * records are added to the sockbuf by sbappend. In particular, each record
1693 * (mbufs linked through m_next) must begin with an address if the protocol
1694 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1695 * data, and then zero or more mbufs of data. In order to allow parallelism
1696 * between network receive and copying to user space, as well as avoid
1697 * sleeping with a mutex held, we release the socket buffer mutex during the
1698 * user space copy. Although the sockbuf is locked, new data may still be
1699 * appended, and thus we must maintain consistency of the sockbuf during that
1702 * The caller may receive the data as a single mbuf chain by supplying an
1703 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1704 * the count in uio_resid.
1707 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1708 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1710 struct mbuf *m, **mp;
1711 int flags, error, offset;
1713 struct protosw *pr = so->so_proto;
1714 struct mbuf *nextrecord;
1716 ssize_t orig_resid = uio->uio_resid;
1721 if (controlp != NULL)
1724 flags = *flagsp &~ MSG_EOR;
1727 if (flags & MSG_OOB)
1728 return (soreceive_rcvoob(so, uio, flags));
1731 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1732 && uio->uio_resid) {
1734 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1737 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1742 SOCKBUF_LOCK(&so->so_rcv);
1743 m = so->so_rcv.sb_mb;
1745 * If we have less data than requested, block awaiting more (subject
1746 * to any timeout) if:
1747 * 1. the current count is less than the low water mark, or
1748 * 2. MSG_DONTWAIT is not set
1750 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1751 sbavail(&so->so_rcv) < uio->uio_resid) &&
1752 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1753 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1754 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1755 ("receive: m == %p sbavail == %u",
1756 m, sbavail(&so->so_rcv)));
1760 error = so->so_error;
1761 if ((flags & MSG_PEEK) == 0)
1763 SOCKBUF_UNLOCK(&so->so_rcv);
1766 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1767 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1769 SOCKBUF_UNLOCK(&so->so_rcv);
1774 for (; m != NULL; m = m->m_next)
1775 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1776 m = so->so_rcv.sb_mb;
1779 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1780 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1781 SOCKBUF_UNLOCK(&so->so_rcv);
1785 if (uio->uio_resid == 0) {
1786 SOCKBUF_UNLOCK(&so->so_rcv);
1789 if ((so->so_state & SS_NBIO) ||
1790 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1791 SOCKBUF_UNLOCK(&so->so_rcv);
1792 error = EWOULDBLOCK;
1795 SBLASTRECORDCHK(&so->so_rcv);
1796 SBLASTMBUFCHK(&so->so_rcv);
1797 error = sbwait(&so->so_rcv);
1798 SOCKBUF_UNLOCK(&so->so_rcv);
1805 * From this point onward, we maintain 'nextrecord' as a cache of the
1806 * pointer to the next record in the socket buffer. We must keep the
1807 * various socket buffer pointers and local stack versions of the
1808 * pointers in sync, pushing out modifications before dropping the
1809 * socket buffer mutex, and re-reading them when picking it up.
1811 * Otherwise, we will race with the network stack appending new data
1812 * or records onto the socket buffer by using inconsistent/stale
1813 * versions of the field, possibly resulting in socket buffer
1816 * By holding the high-level sblock(), we prevent simultaneous
1817 * readers from pulling off the front of the socket buffer.
1819 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1821 uio->uio_td->td_ru.ru_msgrcv++;
1822 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1823 SBLASTRECORDCHK(&so->so_rcv);
1824 SBLASTMBUFCHK(&so->so_rcv);
1825 nextrecord = m->m_nextpkt;
1826 if (pr->pr_flags & PR_ADDR) {
1827 KASSERT(m->m_type == MT_SONAME,
1828 ("m->m_type == %d", m->m_type));
1831 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1833 if (flags & MSG_PEEK) {
1836 sbfree(&so->so_rcv, m);
1837 so->so_rcv.sb_mb = m_free(m);
1838 m = so->so_rcv.sb_mb;
1839 sockbuf_pushsync(&so->so_rcv, nextrecord);
1844 * Process one or more MT_CONTROL mbufs present before any data mbufs
1845 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1846 * just copy the data; if !MSG_PEEK, we call into the protocol to
1847 * perform externalization (or freeing if controlp == NULL).
1849 if (m != NULL && m->m_type == MT_CONTROL) {
1850 struct mbuf *cm = NULL, *cmn;
1851 struct mbuf **cme = &cm;
1854 if (flags & MSG_PEEK) {
1855 if (controlp != NULL) {
1856 *controlp = m_copym(m, 0, m->m_len,
1858 controlp = &(*controlp)->m_next;
1862 sbfree(&so->so_rcv, m);
1863 so->so_rcv.sb_mb = m->m_next;
1866 cme = &(*cme)->m_next;
1867 m = so->so_rcv.sb_mb;
1869 } while (m != NULL && m->m_type == MT_CONTROL);
1870 if ((flags & MSG_PEEK) == 0)
1871 sockbuf_pushsync(&so->so_rcv, nextrecord);
1872 while (cm != NULL) {
1875 if (pr->pr_domain->dom_externalize != NULL) {
1876 SOCKBUF_UNLOCK(&so->so_rcv);
1878 error = (*pr->pr_domain->dom_externalize)
1879 (cm, controlp, flags);
1880 SOCKBUF_LOCK(&so->so_rcv);
1881 } else if (controlp != NULL)
1885 if (controlp != NULL) {
1887 while (*controlp != NULL)
1888 controlp = &(*controlp)->m_next;
1893 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1895 nextrecord = so->so_rcv.sb_mb;
1899 if ((flags & MSG_PEEK) == 0) {
1900 KASSERT(m->m_nextpkt == nextrecord,
1901 ("soreceive: post-control, nextrecord !sync"));
1902 if (nextrecord == NULL) {
1903 KASSERT(so->so_rcv.sb_mb == m,
1904 ("soreceive: post-control, sb_mb!=m"));
1905 KASSERT(so->so_rcv.sb_lastrecord == m,
1906 ("soreceive: post-control, lastrecord!=m"));
1910 if (type == MT_OOBDATA)
1913 if ((flags & MSG_PEEK) == 0) {
1914 KASSERT(so->so_rcv.sb_mb == nextrecord,
1915 ("soreceive: sb_mb != nextrecord"));
1916 if (so->so_rcv.sb_mb == NULL) {
1917 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1918 ("soreceive: sb_lastercord != NULL"));
1922 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1923 SBLASTRECORDCHK(&so->so_rcv);
1924 SBLASTMBUFCHK(&so->so_rcv);
1927 * Now continue to read any data mbufs off of the head of the socket
1928 * buffer until the read request is satisfied. Note that 'type' is
1929 * used to store the type of any mbuf reads that have happened so far
1930 * such that soreceive() can stop reading if the type changes, which
1931 * causes soreceive() to return only one of regular data and inline
1932 * out-of-band data in a single socket receive operation.
1936 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
1939 * If the type of mbuf has changed since the last mbuf
1940 * examined ('type'), end the receive operation.
1942 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1943 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1944 if (type != m->m_type)
1946 } else if (type == MT_OOBDATA)
1949 KASSERT(m->m_type == MT_DATA,
1950 ("m->m_type == %d", m->m_type));
1951 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1952 len = uio->uio_resid;
1953 if (so->so_oobmark && len > so->so_oobmark - offset)
1954 len = so->so_oobmark - offset;
1955 if (len > m->m_len - moff)
1956 len = m->m_len - moff;
1958 * If mp is set, just pass back the mbufs. Otherwise copy
1959 * them out via the uio, then free. Sockbuf must be
1960 * consistent here (points to current mbuf, it points to next
1961 * record) when we drop priority; we must note any additions
1962 * to the sockbuf when we block interrupts again.
1965 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1966 SBLASTRECORDCHK(&so->so_rcv);
1967 SBLASTMBUFCHK(&so->so_rcv);
1968 SOCKBUF_UNLOCK(&so->so_rcv);
1969 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1970 SOCKBUF_LOCK(&so->so_rcv);
1973 * The MT_SONAME mbuf has already been removed
1974 * from the record, so it is necessary to
1975 * remove the data mbufs, if any, to preserve
1976 * the invariant in the case of PR_ADDR that
1977 * requires MT_SONAME mbufs at the head of
1980 if (pr->pr_flags & PR_ATOMIC &&
1981 ((flags & MSG_PEEK) == 0))
1982 (void)sbdroprecord_locked(&so->so_rcv);
1983 SOCKBUF_UNLOCK(&so->so_rcv);
1987 uio->uio_resid -= len;
1988 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1989 if (len == m->m_len - moff) {
1990 if (m->m_flags & M_EOR)
1992 if (flags & MSG_PEEK) {
1996 nextrecord = m->m_nextpkt;
1997 sbfree(&so->so_rcv, m);
1999 m->m_nextpkt = NULL;
2002 so->so_rcv.sb_mb = m = m->m_next;
2005 so->so_rcv.sb_mb = m_free(m);
2006 m = so->so_rcv.sb_mb;
2008 sockbuf_pushsync(&so->so_rcv, nextrecord);
2009 SBLASTRECORDCHK(&so->so_rcv);
2010 SBLASTMBUFCHK(&so->so_rcv);
2013 if (flags & MSG_PEEK)
2017 if (flags & MSG_DONTWAIT) {
2018 *mp = m_copym(m, 0, len,
2022 * m_copym() couldn't
2024 * Adjust uio_resid back
2026 * down by len bytes,
2027 * which we didn't end
2028 * up "copying" over).
2030 uio->uio_resid += len;
2034 SOCKBUF_UNLOCK(&so->so_rcv);
2035 *mp = m_copym(m, 0, len,
2037 SOCKBUF_LOCK(&so->so_rcv);
2040 sbcut_locked(&so->so_rcv, len);
2043 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2044 if (so->so_oobmark) {
2045 if ((flags & MSG_PEEK) == 0) {
2046 so->so_oobmark -= len;
2047 if (so->so_oobmark == 0) {
2048 so->so_rcv.sb_state |= SBS_RCVATMARK;
2053 if (offset == so->so_oobmark)
2057 if (flags & MSG_EOR)
2060 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2061 * must not quit until "uio->uio_resid == 0" or an error
2062 * termination. If a signal/timeout occurs, return with a
2063 * short count but without error. Keep sockbuf locked
2064 * against other readers.
2066 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2067 !sosendallatonce(so) && nextrecord == NULL) {
2068 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2070 so->so_rcv.sb_state & SBS_CANTRCVMORE)
2073 * Notify the protocol that some data has been
2074 * drained before blocking.
2076 if (pr->pr_flags & PR_WANTRCVD) {
2077 SOCKBUF_UNLOCK(&so->so_rcv);
2079 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2080 SOCKBUF_LOCK(&so->so_rcv);
2082 SBLASTRECORDCHK(&so->so_rcv);
2083 SBLASTMBUFCHK(&so->so_rcv);
2085 * We could receive some data while was notifying
2086 * the protocol. Skip blocking in this case.
2088 if (so->so_rcv.sb_mb == NULL) {
2089 error = sbwait(&so->so_rcv);
2091 SOCKBUF_UNLOCK(&so->so_rcv);
2095 m = so->so_rcv.sb_mb;
2097 nextrecord = m->m_nextpkt;
2101 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2102 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2104 if ((flags & MSG_PEEK) == 0)
2105 (void) sbdroprecord_locked(&so->so_rcv);
2107 if ((flags & MSG_PEEK) == 0) {
2110 * First part is an inline SB_EMPTY_FIXUP(). Second
2111 * part makes sure sb_lastrecord is up-to-date if
2112 * there is still data in the socket buffer.
2114 so->so_rcv.sb_mb = nextrecord;
2115 if (so->so_rcv.sb_mb == NULL) {
2116 so->so_rcv.sb_mbtail = NULL;
2117 so->so_rcv.sb_lastrecord = NULL;
2118 } else if (nextrecord->m_nextpkt == NULL)
2119 so->so_rcv.sb_lastrecord = nextrecord;
2121 SBLASTRECORDCHK(&so->so_rcv);
2122 SBLASTMBUFCHK(&so->so_rcv);
2124 * If soreceive() is being done from the socket callback,
2125 * then don't need to generate ACK to peer to update window,
2126 * since ACK will be generated on return to TCP.
2128 if (!(flags & MSG_SOCALLBCK) &&
2129 (pr->pr_flags & PR_WANTRCVD)) {
2130 SOCKBUF_UNLOCK(&so->so_rcv);
2132 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2133 SOCKBUF_LOCK(&so->so_rcv);
2136 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2137 if (orig_resid == uio->uio_resid && orig_resid &&
2138 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2139 SOCKBUF_UNLOCK(&so->so_rcv);
2142 SOCKBUF_UNLOCK(&so->so_rcv);
2147 sbunlock(&so->so_rcv);
2152 * Optimized version of soreceive() for stream (TCP) sockets.
2153 * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled.
2156 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2157 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2159 int len = 0, error = 0, flags, oresid;
2161 struct mbuf *m, *n = NULL;
2163 /* We only do stream sockets. */
2164 if (so->so_type != SOCK_STREAM)
2168 if (controlp != NULL)
2171 flags = *flagsp &~ MSG_EOR;
2174 if (flags & MSG_OOB)
2175 return (soreceive_rcvoob(so, uio, flags));
2181 /* Prevent other readers from entering the socket. */
2182 error = sblock(sb, SBLOCKWAIT(flags));
2187 /* Easy one, no space to copyout anything. */
2188 if (uio->uio_resid == 0) {
2192 oresid = uio->uio_resid;
2194 /* We will never ever get anything unless we are or were connected. */
2195 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2201 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2203 /* Abort if socket has reported problems. */
2205 if (sbavail(sb) > 0)
2207 if (oresid > uio->uio_resid)
2209 error = so->so_error;
2210 if (!(flags & MSG_PEEK))
2215 /* Door is closed. Deliver what is left, if any. */
2216 if (sb->sb_state & SBS_CANTRCVMORE) {
2217 if (sbavail(sb) > 0)
2223 /* Socket buffer is empty and we shall not block. */
2224 if (sbavail(sb) == 0 &&
2225 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2230 /* Socket buffer got some data that we shall deliver now. */
2231 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2232 ((so->so_state & SS_NBIO) ||
2233 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2234 sbavail(sb) >= sb->sb_lowat ||
2235 sbavail(sb) >= uio->uio_resid ||
2236 sbavail(sb) >= sb->sb_hiwat) ) {
2240 /* On MSG_WAITALL we must wait until all data or error arrives. */
2241 if ((flags & MSG_WAITALL) &&
2242 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2246 * Wait and block until (more) data comes in.
2247 * NB: Drops the sockbuf lock during wait.
2255 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2256 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2257 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2261 uio->uio_td->td_ru.ru_msgrcv++;
2263 /* Fill uio until full or current end of socket buffer is reached. */
2264 len = min(uio->uio_resid, sbavail(sb));
2266 /* Dequeue as many mbufs as possible. */
2267 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2271 m_cat(*mp0, sb->sb_mb);
2273 m != NULL && m->m_len <= len;
2275 KASSERT(!(m->m_flags & M_NOTAVAIL),
2276 ("%s: m %p not available", __func__, m));
2278 uio->uio_resid -= m->m_len;
2284 sb->sb_lastrecord = sb->sb_mb;
2285 if (sb->sb_mb == NULL)
2288 /* Copy the remainder. */
2290 KASSERT(sb->sb_mb != NULL,
2291 ("%s: len > 0 && sb->sb_mb empty", __func__));
2293 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2295 len = 0; /* Don't flush data from sockbuf. */
2297 uio->uio_resid -= len;
2308 /* NB: Must unlock socket buffer as uiomove may sleep. */
2310 error = m_mbuftouio(uio, sb->sb_mb, len);
2315 SBLASTRECORDCHK(sb);
2319 * Remove the delivered data from the socket buffer unless we
2320 * were only peeking.
2322 if (!(flags & MSG_PEEK)) {
2324 sbdrop_locked(sb, len);
2326 /* Notify protocol that we drained some data. */
2327 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2328 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2329 !(flags & MSG_SOCALLBCK))) {
2332 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2338 * For MSG_WAITALL we may have to loop again and wait for
2339 * more data to come in.
2341 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2344 SOCKBUF_LOCK_ASSERT(sb);
2345 SBLASTRECORDCHK(sb);
2353 * Optimized version of soreceive() for simple datagram cases from userspace.
2354 * Unlike in the stream case, we're able to drop a datagram if copyout()
2355 * fails, and because we handle datagrams atomically, we don't need to use a
2356 * sleep lock to prevent I/O interlacing.
2359 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2360 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2362 struct mbuf *m, *m2;
2365 struct protosw *pr = so->so_proto;
2366 struct mbuf *nextrecord;
2370 if (controlp != NULL)
2373 flags = *flagsp &~ MSG_EOR;
2378 * For any complicated cases, fall back to the full
2379 * soreceive_generic().
2381 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2382 return (soreceive_generic(so, psa, uio, mp0, controlp,
2386 * Enforce restrictions on use.
2388 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2389 ("soreceive_dgram: wantrcvd"));
2390 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2391 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2392 ("soreceive_dgram: SBS_RCVATMARK"));
2393 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2394 ("soreceive_dgram: P_CONNREQUIRED"));
2397 * Loop blocking while waiting for a datagram.
2399 SOCKBUF_LOCK(&so->so_rcv);
2400 while ((m = so->so_rcv.sb_mb) == NULL) {
2401 KASSERT(sbavail(&so->so_rcv) == 0,
2402 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2403 sbavail(&so->so_rcv)));
2405 error = so->so_error;
2407 SOCKBUF_UNLOCK(&so->so_rcv);
2410 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2411 uio->uio_resid == 0) {
2412 SOCKBUF_UNLOCK(&so->so_rcv);
2415 if ((so->so_state & SS_NBIO) ||
2416 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2417 SOCKBUF_UNLOCK(&so->so_rcv);
2418 return (EWOULDBLOCK);
2420 SBLASTRECORDCHK(&so->so_rcv);
2421 SBLASTMBUFCHK(&so->so_rcv);
2422 error = sbwait(&so->so_rcv);
2424 SOCKBUF_UNLOCK(&so->so_rcv);
2428 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2431 uio->uio_td->td_ru.ru_msgrcv++;
2432 SBLASTRECORDCHK(&so->so_rcv);
2433 SBLASTMBUFCHK(&so->so_rcv);
2434 nextrecord = m->m_nextpkt;
2435 if (nextrecord == NULL) {
2436 KASSERT(so->so_rcv.sb_lastrecord == m,
2437 ("soreceive_dgram: lastrecord != m"));
2440 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2441 ("soreceive_dgram: m_nextpkt != nextrecord"));
2444 * Pull 'm' and its chain off the front of the packet queue.
2446 so->so_rcv.sb_mb = NULL;
2447 sockbuf_pushsync(&so->so_rcv, nextrecord);
2450 * Walk 'm's chain and free that many bytes from the socket buffer.
2452 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2453 sbfree(&so->so_rcv, m2);
2456 * Do a few last checks before we let go of the lock.
2458 SBLASTRECORDCHK(&so->so_rcv);
2459 SBLASTMBUFCHK(&so->so_rcv);
2460 SOCKBUF_UNLOCK(&so->so_rcv);
2462 if (pr->pr_flags & PR_ADDR) {
2463 KASSERT(m->m_type == MT_SONAME,
2464 ("m->m_type == %d", m->m_type));
2466 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2471 /* XXXRW: Can this happen? */
2476 * Packet to copyout() is now in 'm' and it is disconnected from the
2479 * Process one or more MT_CONTROL mbufs present before any data mbufs
2480 * in the first mbuf chain on the socket buffer. We call into the
2481 * protocol to perform externalization (or freeing if controlp ==
2482 * NULL). In some cases there can be only MT_CONTROL mbufs without
2485 if (m->m_type == MT_CONTROL) {
2486 struct mbuf *cm = NULL, *cmn;
2487 struct mbuf **cme = &cm;
2493 cme = &(*cme)->m_next;
2495 } while (m != NULL && m->m_type == MT_CONTROL);
2496 while (cm != NULL) {
2499 if (pr->pr_domain->dom_externalize != NULL) {
2500 error = (*pr->pr_domain->dom_externalize)
2501 (cm, controlp, flags);
2502 } else if (controlp != NULL)
2506 if (controlp != NULL) {
2507 while (*controlp != NULL)
2508 controlp = &(*controlp)->m_next;
2513 KASSERT(m == NULL || m->m_type == MT_DATA,
2514 ("soreceive_dgram: !data"));
2515 while (m != NULL && uio->uio_resid > 0) {
2516 len = uio->uio_resid;
2519 error = uiomove(mtod(m, char *), (int)len, uio);
2524 if (len == m->m_len)
2541 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2542 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2546 CURVNET_SET(so->so_vnet);
2547 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2554 soshutdown(struct socket *so, int how)
2556 struct protosw *pr = so->so_proto;
2557 int error, soerror_enotconn;
2559 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2562 soerror_enotconn = 0;
2564 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2566 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2567 * invoked on a datagram sockets, however historically we would
2568 * actually tear socket down. This is known to be leveraged by
2569 * some applications to unblock process waiting in recvXXX(2)
2570 * by other process that it shares that socket with. Try to meet
2571 * both backward-compatibility and POSIX requirements by forcing
2572 * ENOTCONN but still asking protocol to perform pru_shutdown().
2574 if (so->so_type != SOCK_DGRAM)
2576 soerror_enotconn = 1;
2579 CURVNET_SET(so->so_vnet);
2580 if (pr->pr_usrreqs->pru_flush != NULL)
2581 (*pr->pr_usrreqs->pru_flush)(so, how);
2584 if (how != SHUT_RD) {
2585 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2586 wakeup(&so->so_timeo);
2588 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2590 wakeup(&so->so_timeo);
2593 return (soerror_enotconn ? ENOTCONN : 0);
2597 sorflush(struct socket *so)
2599 struct sockbuf *sb = &so->so_rcv;
2600 struct protosw *pr = so->so_proto;
2606 * In order to avoid calling dom_dispose with the socket buffer mutex
2607 * held, and in order to generally avoid holding the lock for a long
2608 * time, we make a copy of the socket buffer and clear the original
2609 * (except locks, state). The new socket buffer copy won't have
2610 * initialized locks so we can only call routines that won't use or
2611 * assert those locks.
2613 * Dislodge threads currently blocked in receive and wait to acquire
2614 * a lock against other simultaneous readers before clearing the
2615 * socket buffer. Don't let our acquire be interrupted by a signal
2616 * despite any existing socket disposition on interruptable waiting.
2619 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2622 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2623 * and mutex data unchanged.
2626 bzero(&aso, sizeof(aso));
2627 aso.so_pcb = so->so_pcb;
2628 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2629 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2630 bzero(&sb->sb_startzero,
2631 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2636 * Dispose of special rights and flush the copied socket. Don't call
2637 * any unsafe routines (that rely on locks being initialized) on aso.
2639 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2640 (*pr->pr_domain->dom_dispose)(&aso);
2641 sbrelease_internal(&aso.so_rcv, so);
2645 * Wrapper for Socket established helper hook.
2646 * Parameters: socket, context of the hook point, hook id.
2649 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2651 struct socket_hhook_data hhook_data = {
2658 CURVNET_SET(so->so_vnet);
2659 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2662 /* Ugly but needed, since hhooks return void for now */
2663 return (hhook_data.status);
2667 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2668 * additional variant to handle the case where the option value needs to be
2669 * some kind of integer, but not a specific size. In addition to their use
2670 * here, these functions are also called by the protocol-level pr_ctloutput()
2674 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2679 * If the user gives us more than we wanted, we ignore it, but if we
2680 * don't get the minimum length the caller wants, we return EINVAL.
2681 * On success, sopt->sopt_valsize is set to however much we actually
2684 if ((valsize = sopt->sopt_valsize) < minlen)
2687 sopt->sopt_valsize = valsize = len;
2689 if (sopt->sopt_td != NULL)
2690 return (copyin(sopt->sopt_val, buf, valsize));
2692 bcopy(sopt->sopt_val, buf, valsize);
2697 * Kernel version of setsockopt(2).
2699 * XXX: optlen is size_t, not socklen_t
2702 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2705 struct sockopt sopt;
2707 sopt.sopt_level = level;
2708 sopt.sopt_name = optname;
2709 sopt.sopt_dir = SOPT_SET;
2710 sopt.sopt_val = optval;
2711 sopt.sopt_valsize = optlen;
2712 sopt.sopt_td = NULL;
2713 return (sosetopt(so, &sopt));
2717 sosetopt(struct socket *so, struct sockopt *sopt)
2728 CURVNET_SET(so->so_vnet);
2730 if (sopt->sopt_level != SOL_SOCKET) {
2731 if (so->so_proto->pr_ctloutput != NULL) {
2732 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2736 error = ENOPROTOOPT;
2738 switch (sopt->sopt_name) {
2739 case SO_ACCEPTFILTER:
2740 error = accept_filt_setopt(so, sopt);
2746 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2751 so->so_linger = l.l_linger;
2753 so->so_options |= SO_LINGER;
2755 so->so_options &= ~SO_LINGER;
2762 case SO_USELOOPBACK:
2772 error = sooptcopyin(sopt, &optval, sizeof optval,
2778 so->so_options |= sopt->sopt_name;
2780 so->so_options &= ~sopt->sopt_name;
2785 error = sooptcopyin(sopt, &optval, sizeof optval,
2790 if (optval < 0 || optval >= rt_numfibs) {
2794 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2795 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2796 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2797 so->so_fibnum = optval;
2802 case SO_USER_COOKIE:
2803 error = sooptcopyin(sopt, &val32, sizeof val32,
2807 so->so_user_cookie = val32;
2814 error = sooptcopyin(sopt, &optval, sizeof optval,
2820 * Values < 1 make no sense for any of these options,
2828 switch (sopt->sopt_name) {
2831 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2832 &so->so_snd : &so->so_rcv, (u_long)optval,
2833 so, curthread) == 0) {
2837 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2838 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2842 * Make sure the low-water is never greater than the
2846 SOCKBUF_LOCK(&so->so_snd);
2847 so->so_snd.sb_lowat =
2848 (optval > so->so_snd.sb_hiwat) ?
2849 so->so_snd.sb_hiwat : optval;
2850 SOCKBUF_UNLOCK(&so->so_snd);
2853 SOCKBUF_LOCK(&so->so_rcv);
2854 so->so_rcv.sb_lowat =
2855 (optval > so->so_rcv.sb_hiwat) ?
2856 so->so_rcv.sb_hiwat : optval;
2857 SOCKBUF_UNLOCK(&so->so_rcv);
2864 #ifdef COMPAT_FREEBSD32
2865 if (SV_CURPROC_FLAG(SV_ILP32)) {
2866 struct timeval32 tv32;
2868 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2870 CP(tv32, tv, tv_sec);
2871 CP(tv32, tv, tv_usec);
2874 error = sooptcopyin(sopt, &tv, sizeof tv,
2878 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2879 tv.tv_usec >= 1000000) {
2883 if (tv.tv_sec > INT32_MAX)
2887 switch (sopt->sopt_name) {
2889 so->so_snd.sb_timeo = val;
2892 so->so_rcv.sb_timeo = val;
2899 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2903 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2911 error = sooptcopyin(sopt, &optval, sizeof optval,
2915 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
2919 so->so_ts_clock = optval;
2922 case SO_MAX_PACING_RATE:
2923 error = sooptcopyin(sopt, &val32, sizeof(val32),
2927 so->so_max_pacing_rate = val32;
2931 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2932 error = hhook_run_socket(so, sopt,
2935 error = ENOPROTOOPT;
2938 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2939 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
2947 * Helper routine for getsockopt.
2950 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2958 * Documented get behavior is that we always return a value, possibly
2959 * truncated to fit in the user's buffer. Traditional behavior is
2960 * that we always tell the user precisely how much we copied, rather
2961 * than something useful like the total amount we had available for
2962 * her. Note that this interface is not idempotent; the entire
2963 * answer must be generated ahead of time.
2965 valsize = min(len, sopt->sopt_valsize);
2966 sopt->sopt_valsize = valsize;
2967 if (sopt->sopt_val != NULL) {
2968 if (sopt->sopt_td != NULL)
2969 error = copyout(buf, sopt->sopt_val, valsize);
2971 bcopy(buf, sopt->sopt_val, valsize);
2977 sogetopt(struct socket *so, struct sockopt *sopt)
2986 CURVNET_SET(so->so_vnet);
2988 if (sopt->sopt_level != SOL_SOCKET) {
2989 if (so->so_proto->pr_ctloutput != NULL)
2990 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2992 error = ENOPROTOOPT;
2996 switch (sopt->sopt_name) {
2997 case SO_ACCEPTFILTER:
2998 error = accept_filt_getopt(so, sopt);
3003 l.l_onoff = so->so_options & SO_LINGER;
3004 l.l_linger = so->so_linger;
3006 error = sooptcopyout(sopt, &l, sizeof l);
3009 case SO_USELOOPBACK:
3021 optval = so->so_options & sopt->sopt_name;
3023 error = sooptcopyout(sopt, &optval, sizeof optval);
3027 optval = so->so_type;
3031 optval = so->so_proto->pr_protocol;
3036 optval = so->so_error;
3042 optval = so->so_snd.sb_hiwat;
3046 optval = so->so_rcv.sb_hiwat;
3050 optval = so->so_snd.sb_lowat;
3054 optval = so->so_rcv.sb_lowat;
3059 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3060 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3061 #ifdef COMPAT_FREEBSD32
3062 if (SV_CURPROC_FLAG(SV_ILP32)) {
3063 struct timeval32 tv32;
3065 CP(tv, tv32, tv_sec);
3066 CP(tv, tv32, tv_usec);
3067 error = sooptcopyout(sopt, &tv32, sizeof tv32);
3070 error = sooptcopyout(sopt, &tv, sizeof tv);
3075 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3079 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3083 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3091 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3095 error = mac_getsockopt_peerlabel(
3096 sopt->sopt_td->td_ucred, so, &extmac);
3099 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3105 case SO_LISTENQLIMIT:
3106 optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3110 optval = SOLISTENING(so) ? so->sol_qlen : 0;
3113 case SO_LISTENINCQLEN:
3114 optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3118 optval = so->so_ts_clock;
3121 case SO_MAX_PACING_RATE:
3122 optval = so->so_max_pacing_rate;
3126 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3127 error = hhook_run_socket(so, sopt,
3130 error = ENOPROTOOPT;
3142 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3144 struct mbuf *m, *m_prev;
3145 int sopt_size = sopt->sopt_valsize;
3147 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3150 if (sopt_size > MLEN) {
3151 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3152 if ((m->m_flags & M_EXT) == 0) {
3156 m->m_len = min(MCLBYTES, sopt_size);
3158 m->m_len = min(MLEN, sopt_size);
3160 sopt_size -= m->m_len;
3165 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3170 if (sopt_size > MLEN) {
3171 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3173 if ((m->m_flags & M_EXT) == 0) {
3178 m->m_len = min(MCLBYTES, sopt_size);
3180 m->m_len = min(MLEN, sopt_size);
3182 sopt_size -= m->m_len;
3190 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3192 struct mbuf *m0 = m;
3194 if (sopt->sopt_val == NULL)
3196 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3197 if (sopt->sopt_td != NULL) {
3200 error = copyin(sopt->sopt_val, mtod(m, char *),
3207 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3208 sopt->sopt_valsize -= m->m_len;
3209 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3212 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3213 panic("ip6_sooptmcopyin");
3218 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3220 struct mbuf *m0 = m;
3223 if (sopt->sopt_val == NULL)
3225 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3226 if (sopt->sopt_td != NULL) {
3229 error = copyout(mtod(m, char *), sopt->sopt_val,
3236 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3237 sopt->sopt_valsize -= m->m_len;
3238 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3239 valsize += m->m_len;
3243 /* enough soopt buffer should be given from user-land */
3247 sopt->sopt_valsize = valsize;
3252 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3253 * out-of-band data, which will then notify socket consumers.
3256 sohasoutofband(struct socket *so)
3259 if (so->so_sigio != NULL)
3260 pgsigio(&so->so_sigio, SIGURG, 0);
3261 selwakeuppri(&so->so_rdsel, PSOCK);
3265 sopoll(struct socket *so, int events, struct ucred *active_cred,
3270 * We do not need to set or assert curvnet as long as everyone uses
3273 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3278 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3284 if (SOLISTENING(so)) {
3285 if (!(events & (POLLIN | POLLRDNORM)))
3287 else if (!TAILQ_EMPTY(&so->sol_comp))
3288 revents = events & (POLLIN | POLLRDNORM);
3290 selrecord(td, &so->so_rdsel);
3295 SOCKBUF_LOCK(&so->so_snd);
3296 SOCKBUF_LOCK(&so->so_rcv);
3297 if (events & (POLLIN | POLLRDNORM))
3298 if (soreadabledata(so))
3299 revents |= events & (POLLIN | POLLRDNORM);
3300 if (events & (POLLOUT | POLLWRNORM))
3301 if (sowriteable(so))
3302 revents |= events & (POLLOUT | POLLWRNORM);
3303 if (events & (POLLPRI | POLLRDBAND))
3304 if (so->so_oobmark ||
3305 (so->so_rcv.sb_state & SBS_RCVATMARK))
3306 revents |= events & (POLLPRI | POLLRDBAND);
3307 if ((events & POLLINIGNEOF) == 0) {
3308 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3309 revents |= events & (POLLIN | POLLRDNORM);
3310 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3316 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3317 selrecord(td, &so->so_rdsel);
3318 so->so_rcv.sb_flags |= SB_SEL;
3320 if (events & (POLLOUT | POLLWRNORM)) {
3321 selrecord(td, &so->so_wrsel);
3322 so->so_snd.sb_flags |= SB_SEL;
3325 SOCKBUF_UNLOCK(&so->so_rcv);
3326 SOCKBUF_UNLOCK(&so->so_snd);
3333 soo_kqfilter(struct file *fp, struct knote *kn)
3335 struct socket *so = kn->kn_fp->f_data;
3339 switch (kn->kn_filter) {
3341 kn->kn_fop = &soread_filtops;
3342 knl = &so->so_rdsel.si_note;
3346 kn->kn_fop = &sowrite_filtops;
3347 knl = &so->so_wrsel.si_note;
3351 kn->kn_fop = &soempty_filtops;
3352 knl = &so->so_wrsel.si_note;
3360 if (SOLISTENING(so)) {
3361 knlist_add(knl, kn, 1);
3364 knlist_add(knl, kn, 1);
3365 sb->sb_flags |= SB_KNOTE;
3373 * Some routines that return EOPNOTSUPP for entry points that are not
3374 * supported by a protocol. Fill in as needed.
3377 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3384 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3391 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3398 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3405 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3413 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3420 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3428 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3435 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3436 struct ifnet *ifp, struct thread *td)
3443 pru_disconnect_notsupp(struct socket *so)
3450 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3457 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3464 pru_rcvd_notsupp(struct socket *so, int flags)
3471 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3478 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3479 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3486 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3489 return (EOPNOTSUPP);
3493 * This isn't really a ``null'' operation, but it's the default one and
3494 * doesn't do anything destructive.
3497 pru_sense_null(struct socket *so, struct stat *sb)
3500 sb->st_blksize = so->so_snd.sb_hiwat;
3505 pru_shutdown_notsupp(struct socket *so)
3512 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3519 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3520 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3527 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3528 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3535 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3543 filt_sordetach(struct knote *kn)
3545 struct socket *so = kn->kn_fp->f_data;
3548 knlist_remove(&so->so_rdsel.si_note, kn, 1);
3549 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3550 so->so_rcv.sb_flags &= ~SB_KNOTE;
3551 so_rdknl_unlock(so);
3556 filt_soread(struct knote *kn, long hint)
3560 so = kn->kn_fp->f_data;
3562 if (SOLISTENING(so)) {
3563 SOCK_LOCK_ASSERT(so);
3564 kn->kn_data = so->sol_qlen;
3565 return (!TAILQ_EMPTY(&so->sol_comp));
3568 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3570 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3571 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3572 kn->kn_flags |= EV_EOF;
3573 kn->kn_fflags = so->so_error;
3575 } else if (so->so_error) /* temporary udp error */
3578 if (kn->kn_sfflags & NOTE_LOWAT) {
3579 if (kn->kn_data >= kn->kn_sdata)
3581 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3584 /* This hook returning non-zero indicates an event, not error */
3585 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3589 filt_sowdetach(struct knote *kn)
3591 struct socket *so = kn->kn_fp->f_data;
3594 knlist_remove(&so->so_wrsel.si_note, kn, 1);
3595 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3596 so->so_snd.sb_flags &= ~SB_KNOTE;
3597 so_wrknl_unlock(so);
3602 filt_sowrite(struct knote *kn, long hint)
3606 so = kn->kn_fp->f_data;
3608 if (SOLISTENING(so))
3611 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3612 kn->kn_data = sbspace(&so->so_snd);
3614 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3616 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3617 kn->kn_flags |= EV_EOF;
3618 kn->kn_fflags = so->so_error;
3620 } else if (so->so_error) /* temporary udp error */
3622 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3623 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3625 else if (kn->kn_sfflags & NOTE_LOWAT)
3626 return (kn->kn_data >= kn->kn_sdata);
3628 return (kn->kn_data >= so->so_snd.sb_lowat);
3632 filt_soempty(struct knote *kn, long hint)
3636 so = kn->kn_fp->f_data;
3638 if (SOLISTENING(so))
3641 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3642 kn->kn_data = sbused(&so->so_snd);
3644 if (kn->kn_data == 0)
3651 socheckuid(struct socket *so, uid_t uid)
3656 if (so->so_cred->cr_uid != uid)
3662 * These functions are used by protocols to notify the socket layer (and its
3663 * consumers) of state changes in the sockets driven by protocol-side events.
3667 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3669 * Normal sequence from the active (originating) side is that
3670 * soisconnecting() is called during processing of connect() call, resulting
3671 * in an eventual call to soisconnected() if/when the connection is
3672 * established. When the connection is torn down soisdisconnecting() is
3673 * called during processing of disconnect() call, and soisdisconnected() is
3674 * called when the connection to the peer is totally severed. The semantics
3675 * of these routines are such that connectionless protocols can call
3676 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3677 * calls when setting up a ``connection'' takes no time.
3679 * From the passive side, a socket is created with two queues of sockets:
3680 * so_incomp for connections in progress and so_comp for connections already
3681 * made and awaiting user acceptance. As a protocol is preparing incoming
3682 * connections, it creates a socket structure queued on so_incomp by calling
3683 * sonewconn(). When the connection is established, soisconnected() is
3684 * called, and transfers the socket structure to so_comp, making it available
3687 * If a socket is closed with sockets on either so_incomp or so_comp, these
3688 * sockets are dropped.
3690 * If higher-level protocols are implemented in the kernel, the wakeups done
3691 * here will sometimes cause software-interrupt process scheduling.
3694 soisconnecting(struct socket *so)
3698 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3699 so->so_state |= SS_ISCONNECTING;
3704 soisconnected(struct socket *so)
3706 struct socket *head;
3710 * XXXGL: this is the only place where we acquire socket locks
3711 * in reverse order: first child, then listening socket. To
3712 * avoid possible LOR, use try semantics.
3716 if ((head = so->so_listen) != NULL &&
3717 __predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3721 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3722 so->so_state |= SS_ISCONNECTED;
3723 if (head != NULL && (so->so_qstate == SQ_INCOMP)) {
3725 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3726 TAILQ_REMOVE(&head->sol_incomp, so, so_list);
3727 head->sol_incqlen--;
3728 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
3730 so->so_qstate = SQ_COMP;
3732 solisten_wakeup(head); /* unlocks */
3734 SOCKBUF_LOCK(&so->so_rcv);
3735 soupcall_set(so, SO_RCV,
3736 head->sol_accept_filter->accf_callback,
3737 head->sol_accept_filter_arg);
3738 so->so_options &= ~SO_ACCEPTFILTER;
3739 ret = head->sol_accept_filter->accf_callback(so,
3740 head->sol_accept_filter_arg, M_NOWAIT);
3741 if (ret == SU_ISCONNECTED) {
3742 soupcall_clear(so, SO_RCV);
3743 SOCKBUF_UNLOCK(&so->so_rcv);
3746 SOCKBUF_UNLOCK(&so->so_rcv);
3748 SOLISTEN_UNLOCK(head);
3753 SOLISTEN_UNLOCK(head);
3755 wakeup(&so->so_timeo);
3761 soisdisconnecting(struct socket *so)
3765 so->so_state &= ~SS_ISCONNECTING;
3766 so->so_state |= SS_ISDISCONNECTING;
3768 if (!SOLISTENING(so)) {
3769 SOCKBUF_LOCK(&so->so_rcv);
3770 socantrcvmore_locked(so);
3771 SOCKBUF_LOCK(&so->so_snd);
3772 socantsendmore_locked(so);
3775 wakeup(&so->so_timeo);
3779 soisdisconnected(struct socket *so)
3783 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3784 so->so_state |= SS_ISDISCONNECTED;
3786 if (!SOLISTENING(so)) {
3787 SOCKBUF_LOCK(&so->so_rcv);
3788 socantrcvmore_locked(so);
3789 SOCKBUF_LOCK(&so->so_snd);
3790 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
3791 socantsendmore_locked(so);
3794 wakeup(&so->so_timeo);
3798 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3801 sodupsockaddr(const struct sockaddr *sa, int mflags)
3803 struct sockaddr *sa2;
3805 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3807 bcopy(sa, sa2, sa->sa_len);
3812 * Register per-socket buffer upcalls.
3815 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
3819 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
3829 panic("soupcall_set: bad which");
3831 SOCKBUF_LOCK_ASSERT(sb);
3832 sb->sb_upcall = func;
3833 sb->sb_upcallarg = arg;
3834 sb->sb_flags |= SB_UPCALL;
3838 soupcall_clear(struct socket *so, int which)
3842 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
3852 panic("soupcall_clear: bad which");
3854 SOCKBUF_LOCK_ASSERT(sb);
3855 KASSERT(sb->sb_upcall != NULL,
3856 ("%s: so %p no upcall to clear", __func__, so));
3857 sb->sb_upcall = NULL;
3858 sb->sb_upcallarg = NULL;
3859 sb->sb_flags &= ~SB_UPCALL;
3863 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
3866 SOLISTEN_LOCK_ASSERT(so);
3867 so->sol_upcall = func;
3868 so->sol_upcallarg = arg;
3872 so_rdknl_lock(void *arg)
3874 struct socket *so = arg;
3876 if (SOLISTENING(so))
3879 SOCKBUF_LOCK(&so->so_rcv);
3883 so_rdknl_unlock(void *arg)
3885 struct socket *so = arg;
3887 if (SOLISTENING(so))
3890 SOCKBUF_UNLOCK(&so->so_rcv);
3894 so_rdknl_assert_locked(void *arg)
3896 struct socket *so = arg;
3898 if (SOLISTENING(so))
3899 SOCK_LOCK_ASSERT(so);
3901 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3905 so_rdknl_assert_unlocked(void *arg)
3907 struct socket *so = arg;
3909 if (SOLISTENING(so))
3910 SOCK_UNLOCK_ASSERT(so);
3912 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
3916 so_wrknl_lock(void *arg)
3918 struct socket *so = arg;
3920 if (SOLISTENING(so))
3923 SOCKBUF_LOCK(&so->so_snd);
3927 so_wrknl_unlock(void *arg)
3929 struct socket *so = arg;
3931 if (SOLISTENING(so))
3934 SOCKBUF_UNLOCK(&so->so_snd);
3938 so_wrknl_assert_locked(void *arg)
3940 struct socket *so = arg;
3942 if (SOLISTENING(so))
3943 SOCK_LOCK_ASSERT(so);
3945 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3949 so_wrknl_assert_unlocked(void *arg)
3951 struct socket *so = arg;
3953 if (SOLISTENING(so))
3954 SOCK_UNLOCK_ASSERT(so);
3956 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
3960 * Create an external-format (``xsocket'') structure using the information in
3961 * the kernel-format socket structure pointed to by so. This is done to
3962 * reduce the spew of irrelevant information over this interface, to isolate
3963 * user code from changes in the kernel structure, and potentially to provide
3964 * information-hiding if we decide that some of this information should be
3965 * hidden from users.
3968 sotoxsocket(struct socket *so, struct xsocket *xso)
3971 xso->xso_len = sizeof *xso;
3973 xso->so_type = so->so_type;
3974 xso->so_options = so->so_options;
3975 xso->so_linger = so->so_linger;
3976 xso->so_state = so->so_state;
3977 xso->so_pcb = so->so_pcb;
3978 xso->xso_protocol = so->so_proto->pr_protocol;
3979 xso->xso_family = so->so_proto->pr_domain->dom_family;
3980 xso->so_timeo = so->so_timeo;
3981 xso->so_error = so->so_error;
3982 xso->so_uid = so->so_cred->cr_uid;
3983 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3984 if (SOLISTENING(so)) {
3985 xso->so_qlen = so->sol_qlen;
3986 xso->so_incqlen = so->sol_incqlen;
3987 xso->so_qlimit = so->sol_qlimit;
3988 xso->so_oobmark = 0;
3989 bzero(&xso->so_snd, sizeof(xso->so_snd));
3990 bzero(&xso->so_rcv, sizeof(xso->so_rcv));
3992 xso->so_state |= so->so_qstate;
3993 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
3994 xso->so_oobmark = so->so_oobmark;
3995 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3996 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4001 so_sockbuf_rcv(struct socket *so)
4004 return (&so->so_rcv);
4008 so_sockbuf_snd(struct socket *so)
4011 return (&so->so_snd);
4015 so_state_get(const struct socket *so)
4018 return (so->so_state);
4022 so_state_set(struct socket *so, int val)
4029 so_options_get(const struct socket *so)
4032 return (so->so_options);
4036 so_options_set(struct socket *so, int val)
4039 so->so_options = val;
4043 so_error_get(const struct socket *so)
4046 return (so->so_error);
4050 so_error_set(struct socket *so, int val)
4057 so_linger_get(const struct socket *so)
4060 return (so->so_linger);
4064 so_linger_set(struct socket *so, int val)
4067 so->so_linger = val;
4071 so_protosw_get(const struct socket *so)
4074 return (so->so_proto);
4078 so_protosw_set(struct socket *so, struct protosw *val)
4085 so_sorwakeup(struct socket *so)
4092 so_sowwakeup(struct socket *so)
4099 so_sorwakeup_locked(struct socket *so)
4102 sorwakeup_locked(so);
4106 so_sowwakeup_locked(struct socket *so)
4109 sowwakeup_locked(so);
4113 so_lock(struct socket *so)
4120 so_unlock(struct socket *so)