2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993
5 * The Regents of the University of California.
6 * Copyright (c) 2004 The FreeBSD Foundation
7 * Copyright (c) 2004-2008 Robert N. M. Watson
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
38 * Comments on the socket life cycle:
40 * soalloc() sets of socket layer state for a socket, called only by
41 * socreate() and sonewconn(). Socket layer private.
43 * sodealloc() tears down socket layer state for a socket, called only by
44 * sofree() and sonewconn(). Socket layer private.
46 * pru_attach() associates protocol layer state with an allocated socket;
47 * called only once, may fail, aborting socket allocation. This is called
48 * from socreate() and sonewconn(). Socket layer private.
50 * pru_detach() disassociates protocol layer state from an attached socket,
51 * and will be called exactly once for sockets in which pru_attach() has
52 * been successfully called. If pru_attach() returned an error,
53 * pru_detach() will not be called. Socket layer private.
55 * pru_abort() and pru_close() notify the protocol layer that the last
56 * consumer of a socket is starting to tear down the socket, and that the
57 * protocol should terminate the connection. Historically, pru_abort() also
58 * detached protocol state from the socket state, but this is no longer the
61 * socreate() creates a socket and attaches protocol state. This is a public
62 * interface that may be used by socket layer consumers to create new
65 * sonewconn() creates a socket and attaches protocol state. This is a
66 * public interface that may be used by protocols to create new sockets when
67 * a new connection is received and will be available for accept() on a
70 * soclose() destroys a socket after possibly waiting for it to disconnect.
71 * This is a public interface that socket consumers should use to close and
72 * release a socket when done with it.
74 * soabort() destroys a socket without waiting for it to disconnect (used
75 * only for incoming connections that are already partially or fully
76 * connected). This is used internally by the socket layer when clearing
77 * listen socket queues (due to overflow or close on the listen socket), but
78 * is also a public interface protocols may use to abort connections in
79 * their incomplete listen queues should they no longer be required. Sockets
80 * placed in completed connection listen queues should not be aborted for
81 * reasons described in the comment above the soclose() implementation. This
82 * is not a general purpose close routine, and except in the specific
83 * circumstances described here, should not be used.
85 * sofree() will free a socket and its protocol state if all references on
86 * the socket have been released, and is the public interface to attempt to
87 * free a socket when a reference is removed. This is a socket layer private
90 * NOTE: In addition to socreate() and soclose(), which provide a single
91 * socket reference to the consumer to be managed as required, there are two
92 * calls to explicitly manage socket references, soref(), and sorele().
93 * Currently, these are generally required only when transitioning a socket
94 * from a listen queue to a file descriptor, in order to prevent garbage
95 * collection of the socket at an untimely moment. For a number of reasons,
96 * these interfaces are not preferred, and should be avoided.
98 * NOTE: With regard to VNETs the general rule is that callers do not set
99 * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
100 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
101 * and sorflush(), which are usually called from a pre-set VNET context.
102 * sopoll() currently does not need a VNET context to be set.
105 #include <sys/cdefs.h>
106 __FBSDID("$FreeBSD$");
108 #include "opt_inet.h"
109 #include "opt_inet6.h"
110 #include "opt_sctp.h"
112 #include <sys/param.h>
113 #include <sys/systm.h>
114 #include <sys/fcntl.h>
115 #include <sys/limits.h>
116 #include <sys/lock.h>
118 #include <sys/malloc.h>
119 #include <sys/mbuf.h>
120 #include <sys/mutex.h>
121 #include <sys/domain.h>
122 #include <sys/file.h> /* for struct knote */
123 #include <sys/hhook.h>
124 #include <sys/kernel.h>
125 #include <sys/khelp.h>
126 #include <sys/event.h>
127 #include <sys/eventhandler.h>
128 #include <sys/poll.h>
129 #include <sys/proc.h>
130 #include <sys/protosw.h>
131 #include <sys/socket.h>
132 #include <sys/socketvar.h>
133 #include <sys/resourcevar.h>
134 #include <net/route.h>
135 #include <sys/signalvar.h>
136 #include <sys/stat.h>
138 #include <sys/sysctl.h>
139 #include <sys/taskqueue.h>
141 #include <sys/jail.h>
142 #include <sys/syslog.h>
143 #include <netinet/in.h>
145 #include <net/vnet.h>
147 #include <security/mac/mac_framework.h>
151 #ifdef COMPAT_FREEBSD32
152 #include <sys/mount.h>
153 #include <sys/sysent.h>
154 #include <compat/freebsd32/freebsd32.h>
157 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
159 static void so_rdknl_lock(void *);
160 static void so_rdknl_unlock(void *);
161 static void so_rdknl_assert_locked(void *);
162 static void so_rdknl_assert_unlocked(void *);
163 static void so_wrknl_lock(void *);
164 static void so_wrknl_unlock(void *);
165 static void so_wrknl_assert_locked(void *);
166 static void so_wrknl_assert_unlocked(void *);
168 static void filt_sordetach(struct knote *kn);
169 static int filt_soread(struct knote *kn, long hint);
170 static void filt_sowdetach(struct knote *kn);
171 static int filt_sowrite(struct knote *kn, long hint);
172 static int filt_soempty(struct knote *kn, long hint);
173 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
174 fo_kqfilter_t soo_kqfilter;
176 static struct filterops soread_filtops = {
178 .f_detach = filt_sordetach,
179 .f_event = filt_soread,
181 static struct filterops sowrite_filtops = {
183 .f_detach = filt_sowdetach,
184 .f_event = filt_sowrite,
186 static struct filterops soempty_filtops = {
188 .f_detach = filt_sowdetach,
189 .f_event = filt_soempty,
192 so_gen_t so_gencnt; /* generation count for sockets */
194 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
195 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
197 #define VNET_SO_ASSERT(so) \
198 VNET_ASSERT(curvnet != NULL, \
199 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
201 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
202 #define V_socket_hhh VNET(socket_hhh)
205 * Limit on the number of connections in the listen queue waiting
207 * NB: The original sysctl somaxconn is still available but hidden
208 * to prevent confusion about the actual purpose of this number.
210 static u_int somaxconn = SOMAXCONN;
213 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
219 error = sysctl_handle_int(oidp, &val, 0, req);
220 if (error || !req->newptr )
224 * The purpose of the UINT_MAX / 3 limit, is so that the formula
226 * below, will not overflow.
229 if (val < 1 || val > UINT_MAX / 3)
235 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW,
236 0, sizeof(int), sysctl_somaxconn, "I",
237 "Maximum listen socket pending connection accept queue size");
238 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
239 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP,
240 0, sizeof(int), sysctl_somaxconn, "I",
241 "Maximum listen socket pending connection accept queue size (compat)");
243 static int numopensockets;
244 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
245 &numopensockets, 0, "Number of open sockets");
248 * accept_mtx locks down per-socket fields relating to accept queues. See
249 * socketvar.h for an annotation of the protected fields of struct socket.
251 struct mtx accept_mtx;
252 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
255 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
258 static struct mtx so_global_mtx;
259 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
262 * General IPC sysctl name space, used by sockets and a variety of other IPC
265 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
268 * Initialize the socket subsystem and set up the socket
271 static uma_zone_t socket_zone;
275 socket_zone_change(void *tag)
278 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
282 socket_hhook_register(int subtype)
285 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
286 &V_socket_hhh[subtype],
287 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
288 printf("%s: WARNING: unable to register hook\n", __func__);
292 socket_hhook_deregister(int subtype)
295 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
296 printf("%s: WARNING: unable to deregister hook\n", __func__);
300 socket_init(void *tag)
303 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
304 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
305 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
306 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
307 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
308 EVENTHANDLER_PRI_FIRST);
310 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
313 socket_vnet_init(const void *unused __unused)
317 /* We expect a contiguous range */
318 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
319 socket_hhook_register(i);
321 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
322 socket_vnet_init, NULL);
325 socket_vnet_uninit(const void *unused __unused)
329 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
330 socket_hhook_deregister(i);
332 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
333 socket_vnet_uninit, NULL);
336 * Initialise maxsockets. This SYSINIT must be run after
340 init_maxsockets(void *ignored)
343 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
344 maxsockets = imax(maxsockets, maxfiles);
346 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
349 * Sysctl to get and set the maximum global sockets limit. Notify protocols
350 * of the change so that they can update their dependent limits as required.
353 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
355 int error, newmaxsockets;
357 newmaxsockets = maxsockets;
358 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
359 if (error == 0 && req->newptr) {
360 if (newmaxsockets > maxsockets &&
361 newmaxsockets <= maxfiles) {
362 maxsockets = newmaxsockets;
363 EVENTHANDLER_INVOKE(maxsockets_change);
369 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
370 &maxsockets, 0, sysctl_maxsockets, "IU",
371 "Maximum number of sockets available");
374 * Socket operation routines. These routines are called by the routines in
375 * sys_socket.c or from a system process, and implement the semantics of
376 * socket operations by switching out to the protocol specific routines.
380 * Get a socket structure from our zone, and initialize it. Note that it
381 * would probably be better to allocate socket and PCB at the same time, but
382 * I'm not convinced that all the protocols can be easily modified to do
385 * soalloc() returns a socket with a ref count of 0.
387 static struct socket *
388 soalloc(struct vnet *vnet)
392 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
396 if (mac_socket_init(so, M_NOWAIT) != 0) {
397 uma_zfree(socket_zone, so);
401 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
402 uma_zfree(socket_zone, so);
407 * The socket locking protocol allows to lock 2 sockets at a time,
408 * however, the first one must be a listening socket. WITNESS lacks
409 * a feature to change class of an existing lock, so we use DUPOK.
411 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
412 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
413 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
414 so->so_rcv.sb_sel = &so->so_rdsel;
415 so->so_snd.sb_sel = &so->so_wrsel;
416 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
417 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
418 TAILQ_INIT(&so->so_snd.sb_aiojobq);
419 TAILQ_INIT(&so->so_rcv.sb_aiojobq);
420 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
421 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
423 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
424 __func__, __LINE__, so));
427 /* We shouldn't need the so_global_mtx */
428 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
429 /* Do we need more comprehensive error returns? */
430 uma_zfree(socket_zone, so);
433 mtx_lock(&so_global_mtx);
434 so->so_gencnt = ++so_gencnt;
437 vnet->vnet_sockcnt++;
439 mtx_unlock(&so_global_mtx);
445 * Free the storage associated with a socket at the socket layer, tear down
446 * locks, labels, etc. All protocol state is assumed already to have been
447 * torn down (and possibly never set up) by the caller.
450 sodealloc(struct socket *so)
453 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
454 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
456 mtx_lock(&so_global_mtx);
457 so->so_gencnt = ++so_gencnt;
458 --numopensockets; /* Could be below, but faster here. */
460 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
461 __func__, __LINE__, so));
462 so->so_vnet->vnet_sockcnt--;
464 mtx_unlock(&so_global_mtx);
466 mac_socket_destroy(so);
468 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
471 khelp_destroy_osd(&so->osd);
472 if (SOLISTENING(so)) {
473 if (so->sol_accept_filter != NULL)
474 accept_filt_setopt(so, NULL);
476 if (so->so_rcv.sb_hiwat)
477 (void)chgsbsize(so->so_cred->cr_uidinfo,
478 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
479 if (so->so_snd.sb_hiwat)
480 (void)chgsbsize(so->so_cred->cr_uidinfo,
481 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
482 sx_destroy(&so->so_snd.sb_sx);
483 sx_destroy(&so->so_rcv.sb_sx);
484 SOCKBUF_LOCK_DESTROY(&so->so_snd);
485 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
487 mtx_destroy(&so->so_lock);
488 uma_zfree(socket_zone, so);
492 * socreate returns a socket with a ref count of 1. The socket should be
493 * closed with soclose().
496 socreate(int dom, struct socket **aso, int type, int proto,
497 struct ucred *cred, struct thread *td)
504 prp = pffindproto(dom, proto, type);
506 prp = pffindtype(dom, type);
509 /* No support for domain. */
510 if (pffinddomain(dom) == NULL)
511 return (EAFNOSUPPORT);
512 /* No support for socket type. */
513 if (proto == 0 && type != 0)
515 return (EPROTONOSUPPORT);
517 if (prp->pr_usrreqs->pru_attach == NULL ||
518 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
519 return (EPROTONOSUPPORT);
521 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
522 return (EPROTONOSUPPORT);
524 if (prp->pr_type != type)
526 so = soalloc(CRED_TO_VNET(cred));
531 so->so_cred = crhold(cred);
532 if ((prp->pr_domain->dom_family == PF_INET) ||
533 (prp->pr_domain->dom_family == PF_INET6) ||
534 (prp->pr_domain->dom_family == PF_ROUTE))
535 so->so_fibnum = td->td_proc->p_fibnum;
540 mac_socket_create(cred, so);
542 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
543 so_rdknl_assert_locked, so_rdknl_assert_unlocked);
544 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
545 so_wrknl_assert_locked, so_wrknl_assert_unlocked);
547 * Auto-sizing of socket buffers is managed by the protocols and
548 * the appropriate flags must be set in the pru_attach function.
550 CURVNET_SET(so->so_vnet);
551 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
563 static int regression_sonewconn_earlytest = 1;
564 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
565 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
569 * When an attempt at a new connection is noted on a socket which accepts
570 * connections, sonewconn is called. If the connection is possible (subject
571 * to space constraints, etc.) then we allocate a new structure, properly
572 * linked into the data structure of the original socket, and return this.
573 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
575 * Note: the ref count on the socket is 0 on return.
578 sonewconn(struct socket *head, int connstatus)
580 static struct timeval lastover;
581 static struct timeval overinterval = { 60, 0 };
582 static int overcount;
588 over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
589 SOLISTEN_UNLOCK(head);
591 if (regression_sonewconn_earlytest && over) {
597 if (ratecheck(&lastover, &overinterval)) {
598 log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
599 "%i already in queue awaiting acceptance "
600 "(%d occurrences)\n",
601 __func__, head->so_pcb, head->sol_qlen, overcount);
608 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
610 so = soalloc(head->so_vnet);
612 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
613 "limit reached or out of memory\n",
614 __func__, head->so_pcb);
617 so->so_listen = head;
618 so->so_type = head->so_type;
619 so->so_linger = head->so_linger;
620 so->so_state = head->so_state | SS_NOFDREF;
621 so->so_fibnum = head->so_fibnum;
622 so->so_proto = head->so_proto;
623 so->so_cred = crhold(head->so_cred);
625 mac_socket_newconn(head, so);
627 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
628 so_rdknl_assert_locked, so_rdknl_assert_unlocked);
629 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
630 so_wrknl_assert_locked, so_wrknl_assert_unlocked);
631 VNET_SO_ASSERT(head);
632 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
634 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
635 __func__, head->so_pcb);
638 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
640 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
641 __func__, head->so_pcb);
644 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
645 so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
646 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
647 so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
648 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
649 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
652 if (head->sol_accept_filter != NULL)
654 so->so_state |= connstatus;
655 so->so_options = head->so_options & ~SO_ACCEPTCONN;
656 soref(head); /* A socket on (in)complete queue refs head. */
658 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
659 so->so_qstate = SQ_COMP;
661 solisten_wakeup(head); /* unlocks */
664 * Keep removing sockets from the head until there's room for
665 * us to insert on the tail. In pre-locking revisions, this
666 * was a simple if(), but as we could be racing with other
667 * threads and soabort() requires dropping locks, we must
668 * loop waiting for the condition to be true.
670 while (head->sol_incqlen > head->sol_qlimit) {
673 sp = TAILQ_FIRST(&head->sol_incomp);
674 TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
677 sp->so_qstate = SQ_NONE;
678 sp->so_listen = NULL;
680 sorele(head); /* does SOLISTEN_UNLOCK, head stays */
684 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
685 so->so_qstate = SQ_INCOMP;
687 SOLISTEN_UNLOCK(head);
692 #if defined(SCTP) || defined(SCTP_SUPPORT)
694 * Socket part of sctp_peeloff(). Detach a new socket from an
695 * association. The new socket is returned with a reference.
698 sopeeloff(struct socket *head)
702 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
703 __func__, __LINE__, head));
704 so = soalloc(head->so_vnet);
706 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
707 "limit reached or out of memory\n",
708 __func__, head->so_pcb);
711 so->so_type = head->so_type;
712 so->so_options = head->so_options;
713 so->so_linger = head->so_linger;
714 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
715 so->so_fibnum = head->so_fibnum;
716 so->so_proto = head->so_proto;
717 so->so_cred = crhold(head->so_cred);
719 mac_socket_newconn(head, so);
721 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
722 so_rdknl_assert_locked, so_rdknl_assert_unlocked);
723 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
724 so_wrknl_assert_locked, so_wrknl_assert_unlocked);
725 VNET_SO_ASSERT(head);
726 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
728 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
729 __func__, head->so_pcb);
732 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
734 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
735 __func__, head->so_pcb);
738 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
739 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
740 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
741 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
742 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
743 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
752 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
756 CURVNET_SET(so->so_vnet);
757 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
763 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
767 CURVNET_SET(so->so_vnet);
768 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
774 * solisten() transitions a socket from a non-listening state to a listening
775 * state, but can also be used to update the listen queue depth on an
776 * existing listen socket. The protocol will call back into the sockets
777 * layer using solisten_proto_check() and solisten_proto() to check and set
778 * socket-layer listen state. Call backs are used so that the protocol can
779 * acquire both protocol and socket layer locks in whatever order is required
782 * Protocol implementors are advised to hold the socket lock across the
783 * socket-layer test and set to avoid races at the socket layer.
786 solisten(struct socket *so, int backlog, struct thread *td)
790 CURVNET_SET(so->so_vnet);
791 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
797 solisten_proto_check(struct socket *so)
800 SOCK_LOCK_ASSERT(so);
802 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
809 solisten_proto(struct socket *so, int backlog)
811 int sbrcv_lowat, sbsnd_lowat;
812 u_int sbrcv_hiwat, sbsnd_hiwat;
813 short sbrcv_flags, sbsnd_flags;
814 sbintime_t sbrcv_timeo, sbsnd_timeo;
816 SOCK_LOCK_ASSERT(so);
822 * Change this socket to listening state.
824 sbrcv_lowat = so->so_rcv.sb_lowat;
825 sbsnd_lowat = so->so_snd.sb_lowat;
826 sbrcv_hiwat = so->so_rcv.sb_hiwat;
827 sbsnd_hiwat = so->so_snd.sb_hiwat;
828 sbrcv_flags = so->so_rcv.sb_flags;
829 sbsnd_flags = so->so_snd.sb_flags;
830 sbrcv_timeo = so->so_rcv.sb_timeo;
831 sbsnd_timeo = so->so_snd.sb_timeo;
833 sbdestroy(&so->so_snd, so);
834 sbdestroy(&so->so_rcv, so);
835 sx_destroy(&so->so_snd.sb_sx);
836 sx_destroy(&so->so_rcv.sb_sx);
837 SOCKBUF_LOCK_DESTROY(&so->so_snd);
838 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
842 sizeof(struct socket) - offsetof(struct socket, so_rcv));
845 so->sol_sbrcv_lowat = sbrcv_lowat;
846 so->sol_sbsnd_lowat = sbsnd_lowat;
847 so->sol_sbrcv_hiwat = sbrcv_hiwat;
848 so->sol_sbsnd_hiwat = sbsnd_hiwat;
849 so->sol_sbrcv_flags = sbrcv_flags;
850 so->sol_sbsnd_flags = sbsnd_flags;
851 so->sol_sbrcv_timeo = sbrcv_timeo;
852 so->sol_sbsnd_timeo = sbsnd_timeo;
854 so->sol_qlen = so->sol_incqlen = 0;
855 TAILQ_INIT(&so->sol_incomp);
856 TAILQ_INIT(&so->sol_comp);
858 so->sol_accept_filter = NULL;
859 so->sol_accept_filter_arg = NULL;
860 so->sol_accept_filter_str = NULL;
862 so->sol_upcall = NULL;
863 so->sol_upcallarg = NULL;
865 so->so_options |= SO_ACCEPTCONN;
868 if (backlog < 0 || backlog > somaxconn)
870 so->sol_qlimit = backlog;
874 * Wakeup listeners/subsystems once we have a complete connection.
875 * Enters with lock, returns unlocked.
878 solisten_wakeup(struct socket *sol)
881 if (sol->sol_upcall != NULL)
882 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
884 selwakeuppri(&sol->so_rdsel, PSOCK);
885 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
887 SOLISTEN_UNLOCK(sol);
888 wakeup_one(&sol->sol_comp);
889 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
890 pgsigio(&sol->so_sigio, SIGIO, 0);
894 * Return single connection off a listening socket queue. Main consumer of
895 * the function is kern_accept4(). Some modules, that do their own accept
896 * management also use the function.
898 * Listening socket must be locked on entry and is returned unlocked on
900 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
903 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
908 SOLISTEN_LOCK_ASSERT(head);
910 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
911 head->so_error == 0) {
912 error = msleep(&head->sol_comp, &head->so_lock, PSOCK | PCATCH,
915 SOLISTEN_UNLOCK(head);
919 if (head->so_error) {
920 error = head->so_error;
922 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
927 SOLISTEN_UNLOCK(head);
930 so = TAILQ_FIRST(&head->sol_comp);
932 KASSERT(so->so_qstate == SQ_COMP,
933 ("%s: so %p not SQ_COMP", __func__, so));
936 so->so_qstate = SQ_NONE;
937 so->so_listen = NULL;
938 TAILQ_REMOVE(&head->sol_comp, so, so_list);
939 if (flags & ACCEPT4_INHERIT)
940 so->so_state |= (head->so_state & SS_NBIO);
942 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
951 * Evaluate the reference count and named references on a socket; if no
952 * references remain, free it. This should be called whenever a reference is
953 * released, such as in sorele(), but also when named reference flags are
954 * cleared in socket or protocol code.
956 * sofree() will free the socket if:
958 * - There are no outstanding file descriptor references or related consumers
961 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
963 * - The protocol does not have an outstanding strong reference on the socket
966 * - The socket is not in a completed connection queue, so a process has been
967 * notified that it is present. If it is removed, the user process may
968 * block in accept() despite select() saying the socket was ready.
971 sofree(struct socket *so)
973 struct protosw *pr = so->so_proto;
975 SOCK_LOCK_ASSERT(so);
977 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
978 (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) {
983 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
987 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
990 * To solve race between close of a listening socket and
991 * a socket on its incomplete queue, we need to lock both.
992 * The order is first listening socket, then regular.
993 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
994 * function and the listening socket are the only pointers
995 * to so. To preserve so and sol, we reference both and then
997 * After relock the socket may not move to so_comp since it
998 * doesn't have PCB already, but it may be removed from
999 * so_incomp. If that happens, we share responsiblity on
1000 * freeing the socket, but soclose() has already removed
1008 if (so->so_qstate == SQ_INCOMP) {
1009 KASSERT(so->so_listen == sol,
1010 ("%s: so %p migrated out of sol %p",
1011 __func__, so, sol));
1012 TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1014 /* This is guarenteed not to be the last. */
1015 refcount_release(&sol->so_count);
1016 so->so_qstate = SQ_NONE;
1017 so->so_listen = NULL;
1019 KASSERT(so->so_listen == NULL,
1020 ("%s: so %p not on (in)comp with so_listen",
1023 KASSERT(so->so_count == 1,
1024 ("%s: so %p count %u", __func__, so, so->so_count));
1027 if (SOLISTENING(so))
1028 so->so_error = ECONNABORTED;
1031 if (so->so_dtor != NULL)
1035 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1036 (*pr->pr_domain->dom_dispose)(so);
1037 if (pr->pr_usrreqs->pru_detach != NULL)
1038 (*pr->pr_usrreqs->pru_detach)(so);
1041 * From this point on, we assume that no other references to this
1042 * socket exist anywhere else in the stack. Therefore, no locks need
1043 * to be acquired or held.
1045 * We used to do a lot of socket buffer and socket locking here, as
1046 * well as invoke sorflush() and perform wakeups. The direct call to
1047 * dom_dispose() and sbrelease_internal() are an inlining of what was
1048 * necessary from sorflush().
1050 * Notice that the socket buffer and kqueue state are torn down
1051 * before calling pru_detach. This means that protocols shold not
1052 * assume they can perform socket wakeups, etc, in their detach code.
1054 if (!SOLISTENING(so)) {
1055 sbdestroy(&so->so_snd, so);
1056 sbdestroy(&so->so_rcv, so);
1058 seldrain(&so->so_rdsel);
1059 seldrain(&so->so_wrsel);
1060 knlist_destroy(&so->so_rdsel.si_note);
1061 knlist_destroy(&so->so_wrsel.si_note);
1066 * Close a socket on last file table reference removal. Initiate disconnect
1067 * if connected. Free socket when disconnect complete.
1069 * This function will sorele() the socket. Note that soclose() may be called
1070 * prior to the ref count reaching zero. The actual socket structure will
1071 * not be freed until the ref count reaches zero.
1074 soclose(struct socket *so)
1076 struct accept_queue lqueue;
1080 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1082 CURVNET_SET(so->so_vnet);
1083 funsetown(&so->so_sigio);
1084 if (so->so_state & SS_ISCONNECTED) {
1085 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1086 error = sodisconnect(so);
1088 if (error == ENOTCONN)
1094 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1095 if ((so->so_state & SS_ISDISCONNECTING) &&
1096 (so->so_state & SS_NBIO))
1098 while (so->so_state & SS_ISCONNECTED) {
1099 error = tsleep(&so->so_timeo,
1100 PSOCK | PCATCH, "soclos",
1101 so->so_linger * hz);
1109 if (so->so_proto->pr_usrreqs->pru_close != NULL)
1110 (*so->so_proto->pr_usrreqs->pru_close)(so);
1113 if ((listening = (so->so_options & SO_ACCEPTCONN))) {
1116 TAILQ_INIT(&lqueue);
1117 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1118 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1120 so->sol_qlen = so->sol_incqlen = 0;
1122 TAILQ_FOREACH(sp, &lqueue, so_list) {
1124 sp->so_qstate = SQ_NONE;
1125 sp->so_listen = NULL;
1127 /* Guaranteed not to be the last. */
1128 refcount_release(&so->so_count);
1131 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1132 so->so_state |= SS_NOFDREF;
1135 struct socket *sp, *tsp;
1137 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1139 if (sp->so_count == 0) {
1143 /* sp is now in sofree() */
1152 * soabort() is used to abruptly tear down a connection, such as when a
1153 * resource limit is reached (listen queue depth exceeded), or if a listen
1154 * socket is closed while there are sockets waiting to be accepted.
1156 * This interface is tricky, because it is called on an unreferenced socket,
1157 * and must be called only by a thread that has actually removed the socket
1158 * from the listen queue it was on, or races with other threads are risked.
1160 * This interface will call into the protocol code, so must not be called
1161 * with any socket locks held. Protocols do call it while holding their own
1162 * recursible protocol mutexes, but this is something that should be subject
1163 * to review in the future.
1166 soabort(struct socket *so)
1170 * In as much as is possible, assert that no references to this
1171 * socket are held. This is not quite the same as asserting that the
1172 * current thread is responsible for arranging for no references, but
1173 * is as close as we can get for now.
1175 KASSERT(so->so_count == 0, ("soabort: so_count"));
1176 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1177 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1180 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1181 (*so->so_proto->pr_usrreqs->pru_abort)(so);
1187 soaccept(struct socket *so, struct sockaddr **nam)
1192 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1193 so->so_state &= ~SS_NOFDREF;
1196 CURVNET_SET(so->so_vnet);
1197 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1203 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1206 return (soconnectat(AT_FDCWD, so, nam, td));
1210 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1214 if (so->so_options & SO_ACCEPTCONN)
1215 return (EOPNOTSUPP);
1217 CURVNET_SET(so->so_vnet);
1219 * If protocol is connection-based, can only connect once.
1220 * Otherwise, if connected, try to disconnect first. This allows
1221 * user to disconnect by connecting to, e.g., a null address.
1223 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1224 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1225 (error = sodisconnect(so)))) {
1229 * Prevent accumulated error from previous connection from
1233 if (fd == AT_FDCWD) {
1234 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1237 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1247 soconnect2(struct socket *so1, struct socket *so2)
1251 CURVNET_SET(so1->so_vnet);
1252 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1258 sodisconnect(struct socket *so)
1262 if ((so->so_state & SS_ISCONNECTED) == 0)
1264 if (so->so_state & SS_ISDISCONNECTING)
1267 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1271 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1274 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1275 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1279 int clen = 0, error, dontroute;
1281 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1282 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1283 ("sosend_dgram: !PR_ATOMIC"));
1286 resid = uio->uio_resid;
1288 resid = top->m_pkthdr.len;
1290 * In theory resid should be unsigned. However, space must be
1291 * signed, as it might be less than 0 if we over-committed, and we
1292 * must use a signed comparison of space and resid. On the other
1293 * hand, a negative resid causes us to loop sending 0-length
1294 * segments to the protocol.
1302 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1304 td->td_ru.ru_msgsnd++;
1305 if (control != NULL)
1306 clen = control->m_len;
1308 SOCKBUF_LOCK(&so->so_snd);
1309 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1310 SOCKBUF_UNLOCK(&so->so_snd);
1315 error = so->so_error;
1317 SOCKBUF_UNLOCK(&so->so_snd);
1320 if ((so->so_state & SS_ISCONNECTED) == 0) {
1322 * `sendto' and `sendmsg' is allowed on a connection-based
1323 * socket if it supports implied connect. Return ENOTCONN if
1324 * not connected and no address is supplied.
1326 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1327 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1328 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1329 !(resid == 0 && clen != 0)) {
1330 SOCKBUF_UNLOCK(&so->so_snd);
1334 } else if (addr == NULL) {
1335 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1338 error = EDESTADDRREQ;
1339 SOCKBUF_UNLOCK(&so->so_snd);
1345 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1346 * problem and need fixing.
1348 space = sbspace(&so->so_snd);
1349 if (flags & MSG_OOB)
1352 SOCKBUF_UNLOCK(&so->so_snd);
1353 if (resid > space) {
1359 if (flags & MSG_EOR)
1360 top->m_flags |= M_EOR;
1363 * Copy the data from userland into a mbuf chain.
1364 * If no data is to be copied in, a single empty mbuf
1367 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1368 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1370 error = EFAULT; /* only possible error */
1373 space -= resid - uio->uio_resid;
1374 resid = uio->uio_resid;
1376 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1378 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1383 so->so_options |= SO_DONTROUTE;
1387 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1388 * of date. We could have received a reset packet in an interrupt or
1389 * maybe we slept while doing page faults in uiomove() etc. We could
1390 * probably recheck again inside the locking protection here, but
1391 * there are probably other places that this also happens. We must
1395 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1396 (flags & MSG_OOB) ? PRUS_OOB :
1398 * If the user set MSG_EOF, the protocol understands this flag and
1399 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1401 ((flags & MSG_EOF) &&
1402 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1405 /* If there is more to send set PRUS_MORETOCOME */
1406 (flags & MSG_MORETOCOME) ||
1407 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1408 top, addr, control, td);
1411 so->so_options &= ~SO_DONTROUTE;
1420 if (control != NULL)
1426 * Send on a socket. If send must go all at once and message is larger than
1427 * send buffering, then hard error. Lock against other senders. If must go
1428 * all at once and not enough room now, then inform user that this would
1429 * block and do nothing. Otherwise, if nonblocking, send as much as
1430 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1431 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1432 * in mbuf chain must be small enough to send all at once.
1434 * Returns nonzero on error, timeout or signal; callers must check for short
1435 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1439 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1440 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1444 int clen = 0, error, dontroute;
1445 int atomic = sosendallatonce(so) || top;
1448 resid = uio->uio_resid;
1450 resid = top->m_pkthdr.len;
1452 * In theory resid should be unsigned. However, space must be
1453 * signed, as it might be less than 0 if we over-committed, and we
1454 * must use a signed comparison of space and resid. On the other
1455 * hand, a negative resid causes us to loop sending 0-length
1456 * segments to the protocol.
1458 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1459 * type sockets since that's an error.
1461 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1467 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1468 (so->so_proto->pr_flags & PR_ATOMIC);
1470 td->td_ru.ru_msgsnd++;
1471 if (control != NULL)
1472 clen = control->m_len;
1474 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1480 SOCKBUF_LOCK(&so->so_snd);
1481 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1482 SOCKBUF_UNLOCK(&so->so_snd);
1487 error = so->so_error;
1489 SOCKBUF_UNLOCK(&so->so_snd);
1492 if ((so->so_state & SS_ISCONNECTED) == 0) {
1494 * `sendto' and `sendmsg' is allowed on a connection-
1495 * based socket if it supports implied connect.
1496 * Return ENOTCONN if not connected and no address is
1499 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1500 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1501 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1502 !(resid == 0 && clen != 0)) {
1503 SOCKBUF_UNLOCK(&so->so_snd);
1507 } else if (addr == NULL) {
1508 SOCKBUF_UNLOCK(&so->so_snd);
1509 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1512 error = EDESTADDRREQ;
1516 space = sbspace(&so->so_snd);
1517 if (flags & MSG_OOB)
1519 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1520 clen > so->so_snd.sb_hiwat) {
1521 SOCKBUF_UNLOCK(&so->so_snd);
1525 if (space < resid + clen &&
1526 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1527 if ((so->so_state & SS_NBIO) ||
1528 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1529 SOCKBUF_UNLOCK(&so->so_snd);
1530 error = EWOULDBLOCK;
1533 error = sbwait(&so->so_snd);
1534 SOCKBUF_UNLOCK(&so->so_snd);
1539 SOCKBUF_UNLOCK(&so->so_snd);
1544 if (flags & MSG_EOR)
1545 top->m_flags |= M_EOR;
1548 * Copy the data from userland into a mbuf
1549 * chain. If resid is 0, which can happen
1550 * only if we have control to send, then
1551 * a single empty mbuf is returned. This
1552 * is a workaround to prevent protocol send
1555 top = m_uiotombuf(uio, M_WAITOK, space,
1556 (atomic ? max_hdr : 0),
1557 (atomic ? M_PKTHDR : 0) |
1558 ((flags & MSG_EOR) ? M_EOR : 0));
1560 error = EFAULT; /* only possible error */
1563 space -= resid - uio->uio_resid;
1564 resid = uio->uio_resid;
1568 so->so_options |= SO_DONTROUTE;
1572 * XXX all the SBS_CANTSENDMORE checks previously
1573 * done could be out of date. We could have received
1574 * a reset packet in an interrupt or maybe we slept
1575 * while doing page faults in uiomove() etc. We
1576 * could probably recheck again inside the locking
1577 * protection here, but there are probably other
1578 * places that this also happens. We must rethink
1582 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1583 (flags & MSG_OOB) ? PRUS_OOB :
1585 * If the user set MSG_EOF, the protocol understands
1586 * this flag and nothing left to send then use
1587 * PRU_SEND_EOF instead of PRU_SEND.
1589 ((flags & MSG_EOF) &&
1590 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1593 /* If there is more to send set PRUS_MORETOCOME. */
1594 (flags & MSG_MORETOCOME) ||
1595 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1596 top, addr, control, td);
1599 so->so_options &= ~SO_DONTROUTE;
1607 } while (resid && space > 0);
1611 sbunlock(&so->so_snd);
1615 if (control != NULL)
1621 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1622 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1626 CURVNET_SET(so->so_vnet);
1627 if (!SOLISTENING(so))
1628 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1629 top, control, flags, td);
1640 * The part of soreceive() that implements reading non-inline out-of-band
1641 * data from a socket. For more complete comments, see soreceive(), from
1642 * which this code originated.
1644 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1645 * unable to return an mbuf chain to the caller.
1648 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1650 struct protosw *pr = so->so_proto;
1654 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1657 m = m_get(M_WAITOK, MT_DATA);
1658 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1662 error = uiomove(mtod(m, void *),
1663 (int) min(uio->uio_resid, m->m_len), uio);
1665 } while (uio->uio_resid && error == 0 && m);
1673 * Following replacement or removal of the first mbuf on the first mbuf chain
1674 * of a socket buffer, push necessary state changes back into the socket
1675 * buffer so that other consumers see the values consistently. 'nextrecord'
1676 * is the callers locally stored value of the original value of
1677 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1678 * NOTE: 'nextrecord' may be NULL.
1680 static __inline void
1681 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1684 SOCKBUF_LOCK_ASSERT(sb);
1686 * First, update for the new value of nextrecord. If necessary, make
1687 * it the first record.
1689 if (sb->sb_mb != NULL)
1690 sb->sb_mb->m_nextpkt = nextrecord;
1692 sb->sb_mb = nextrecord;
1695 * Now update any dependent socket buffer fields to reflect the new
1696 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1697 * addition of a second clause that takes care of the case where
1698 * sb_mb has been updated, but remains the last record.
1700 if (sb->sb_mb == NULL) {
1701 sb->sb_mbtail = NULL;
1702 sb->sb_lastrecord = NULL;
1703 } else if (sb->sb_mb->m_nextpkt == NULL)
1704 sb->sb_lastrecord = sb->sb_mb;
1708 * Implement receive operations on a socket. We depend on the way that
1709 * records are added to the sockbuf by sbappend. In particular, each record
1710 * (mbufs linked through m_next) must begin with an address if the protocol
1711 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1712 * data, and then zero or more mbufs of data. In order to allow parallelism
1713 * between network receive and copying to user space, as well as avoid
1714 * sleeping with a mutex held, we release the socket buffer mutex during the
1715 * user space copy. Although the sockbuf is locked, new data may still be
1716 * appended, and thus we must maintain consistency of the sockbuf during that
1719 * The caller may receive the data as a single mbuf chain by supplying an
1720 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1721 * the count in uio_resid.
1724 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1725 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1727 struct mbuf *m, **mp;
1728 int flags, error, offset;
1730 struct protosw *pr = so->so_proto;
1731 struct mbuf *nextrecord;
1733 ssize_t orig_resid = uio->uio_resid;
1738 if (controlp != NULL)
1741 flags = *flagsp &~ MSG_EOR;
1744 if (flags & MSG_OOB)
1745 return (soreceive_rcvoob(so, uio, flags));
1748 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1749 && uio->uio_resid) {
1751 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1754 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1759 SOCKBUF_LOCK(&so->so_rcv);
1760 m = so->so_rcv.sb_mb;
1762 * If we have less data than requested, block awaiting more (subject
1763 * to any timeout) if:
1764 * 1. the current count is less than the low water mark, or
1765 * 2. MSG_DONTWAIT is not set
1767 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1768 sbavail(&so->so_rcv) < uio->uio_resid) &&
1769 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1770 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1771 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1772 ("receive: m == %p sbavail == %u",
1773 m, sbavail(&so->so_rcv)));
1777 error = so->so_error;
1778 if ((flags & MSG_PEEK) == 0)
1780 SOCKBUF_UNLOCK(&so->so_rcv);
1783 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1784 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1786 SOCKBUF_UNLOCK(&so->so_rcv);
1791 for (; m != NULL; m = m->m_next)
1792 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1793 m = so->so_rcv.sb_mb;
1796 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
1797 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
1798 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1799 SOCKBUF_UNLOCK(&so->so_rcv);
1803 if (uio->uio_resid == 0) {
1804 SOCKBUF_UNLOCK(&so->so_rcv);
1807 if ((so->so_state & SS_NBIO) ||
1808 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1809 SOCKBUF_UNLOCK(&so->so_rcv);
1810 error = EWOULDBLOCK;
1813 SBLASTRECORDCHK(&so->so_rcv);
1814 SBLASTMBUFCHK(&so->so_rcv);
1815 error = sbwait(&so->so_rcv);
1816 SOCKBUF_UNLOCK(&so->so_rcv);
1823 * From this point onward, we maintain 'nextrecord' as a cache of the
1824 * pointer to the next record in the socket buffer. We must keep the
1825 * various socket buffer pointers and local stack versions of the
1826 * pointers in sync, pushing out modifications before dropping the
1827 * socket buffer mutex, and re-reading them when picking it up.
1829 * Otherwise, we will race with the network stack appending new data
1830 * or records onto the socket buffer by using inconsistent/stale
1831 * versions of the field, possibly resulting in socket buffer
1834 * By holding the high-level sblock(), we prevent simultaneous
1835 * readers from pulling off the front of the socket buffer.
1837 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1839 uio->uio_td->td_ru.ru_msgrcv++;
1840 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1841 SBLASTRECORDCHK(&so->so_rcv);
1842 SBLASTMBUFCHK(&so->so_rcv);
1843 nextrecord = m->m_nextpkt;
1844 if (pr->pr_flags & PR_ADDR) {
1845 KASSERT(m->m_type == MT_SONAME,
1846 ("m->m_type == %d", m->m_type));
1849 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1851 if (flags & MSG_PEEK) {
1854 sbfree(&so->so_rcv, m);
1855 so->so_rcv.sb_mb = m_free(m);
1856 m = so->so_rcv.sb_mb;
1857 sockbuf_pushsync(&so->so_rcv, nextrecord);
1862 * Process one or more MT_CONTROL mbufs present before any data mbufs
1863 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1864 * just copy the data; if !MSG_PEEK, we call into the protocol to
1865 * perform externalization (or freeing if controlp == NULL).
1867 if (m != NULL && m->m_type == MT_CONTROL) {
1868 struct mbuf *cm = NULL, *cmn;
1869 struct mbuf **cme = &cm;
1872 if (flags & MSG_PEEK) {
1873 if (controlp != NULL) {
1874 *controlp = m_copym(m, 0, m->m_len,
1876 controlp = &(*controlp)->m_next;
1880 sbfree(&so->so_rcv, m);
1881 so->so_rcv.sb_mb = m->m_next;
1884 cme = &(*cme)->m_next;
1885 m = so->so_rcv.sb_mb;
1887 } while (m != NULL && m->m_type == MT_CONTROL);
1888 if ((flags & MSG_PEEK) == 0)
1889 sockbuf_pushsync(&so->so_rcv, nextrecord);
1890 while (cm != NULL) {
1893 if (pr->pr_domain->dom_externalize != NULL) {
1894 SOCKBUF_UNLOCK(&so->so_rcv);
1896 error = (*pr->pr_domain->dom_externalize)
1897 (cm, controlp, flags);
1898 SOCKBUF_LOCK(&so->so_rcv);
1899 } else if (controlp != NULL)
1903 if (controlp != NULL) {
1905 while (*controlp != NULL)
1906 controlp = &(*controlp)->m_next;
1911 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1913 nextrecord = so->so_rcv.sb_mb;
1917 if ((flags & MSG_PEEK) == 0) {
1918 KASSERT(m->m_nextpkt == nextrecord,
1919 ("soreceive: post-control, nextrecord !sync"));
1920 if (nextrecord == NULL) {
1921 KASSERT(so->so_rcv.sb_mb == m,
1922 ("soreceive: post-control, sb_mb!=m"));
1923 KASSERT(so->so_rcv.sb_lastrecord == m,
1924 ("soreceive: post-control, lastrecord!=m"));
1928 if (type == MT_OOBDATA)
1931 if ((flags & MSG_PEEK) == 0) {
1932 KASSERT(so->so_rcv.sb_mb == nextrecord,
1933 ("soreceive: sb_mb != nextrecord"));
1934 if (so->so_rcv.sb_mb == NULL) {
1935 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1936 ("soreceive: sb_lastercord != NULL"));
1940 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1941 SBLASTRECORDCHK(&so->so_rcv);
1942 SBLASTMBUFCHK(&so->so_rcv);
1945 * Now continue to read any data mbufs off of the head of the socket
1946 * buffer until the read request is satisfied. Note that 'type' is
1947 * used to store the type of any mbuf reads that have happened so far
1948 * such that soreceive() can stop reading if the type changes, which
1949 * causes soreceive() to return only one of regular data and inline
1950 * out-of-band data in a single socket receive operation.
1954 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
1957 * If the type of mbuf has changed since the last mbuf
1958 * examined ('type'), end the receive operation.
1960 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1961 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1962 if (type != m->m_type)
1964 } else if (type == MT_OOBDATA)
1967 KASSERT(m->m_type == MT_DATA,
1968 ("m->m_type == %d", m->m_type));
1969 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1970 len = uio->uio_resid;
1971 if (so->so_oobmark && len > so->so_oobmark - offset)
1972 len = so->so_oobmark - offset;
1973 if (len > m->m_len - moff)
1974 len = m->m_len - moff;
1976 * If mp is set, just pass back the mbufs. Otherwise copy
1977 * them out via the uio, then free. Sockbuf must be
1978 * consistent here (points to current mbuf, it points to next
1979 * record) when we drop priority; we must note any additions
1980 * to the sockbuf when we block interrupts again.
1983 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1984 SBLASTRECORDCHK(&so->so_rcv);
1985 SBLASTMBUFCHK(&so->so_rcv);
1986 SOCKBUF_UNLOCK(&so->so_rcv);
1987 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1988 SOCKBUF_LOCK(&so->so_rcv);
1991 * The MT_SONAME mbuf has already been removed
1992 * from the record, so it is necessary to
1993 * remove the data mbufs, if any, to preserve
1994 * the invariant in the case of PR_ADDR that
1995 * requires MT_SONAME mbufs at the head of
1998 if (pr->pr_flags & PR_ATOMIC &&
1999 ((flags & MSG_PEEK) == 0))
2000 (void)sbdroprecord_locked(&so->so_rcv);
2001 SOCKBUF_UNLOCK(&so->so_rcv);
2005 uio->uio_resid -= len;
2006 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2007 if (len == m->m_len - moff) {
2008 if (m->m_flags & M_EOR)
2010 if (flags & MSG_PEEK) {
2014 nextrecord = m->m_nextpkt;
2015 sbfree(&so->so_rcv, m);
2017 m->m_nextpkt = NULL;
2020 so->so_rcv.sb_mb = m = m->m_next;
2023 so->so_rcv.sb_mb = m_free(m);
2024 m = so->so_rcv.sb_mb;
2026 sockbuf_pushsync(&so->so_rcv, nextrecord);
2027 SBLASTRECORDCHK(&so->so_rcv);
2028 SBLASTMBUFCHK(&so->so_rcv);
2031 if (flags & MSG_PEEK)
2035 if (flags & MSG_DONTWAIT) {
2036 *mp = m_copym(m, 0, len,
2040 * m_copym() couldn't
2042 * Adjust uio_resid back
2044 * down by len bytes,
2045 * which we didn't end
2046 * up "copying" over).
2048 uio->uio_resid += len;
2052 SOCKBUF_UNLOCK(&so->so_rcv);
2053 *mp = m_copym(m, 0, len,
2055 SOCKBUF_LOCK(&so->so_rcv);
2058 sbcut_locked(&so->so_rcv, len);
2061 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2062 if (so->so_oobmark) {
2063 if ((flags & MSG_PEEK) == 0) {
2064 so->so_oobmark -= len;
2065 if (so->so_oobmark == 0) {
2066 so->so_rcv.sb_state |= SBS_RCVATMARK;
2071 if (offset == so->so_oobmark)
2075 if (flags & MSG_EOR)
2078 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2079 * must not quit until "uio->uio_resid == 0" or an error
2080 * termination. If a signal/timeout occurs, return with a
2081 * short count but without error. Keep sockbuf locked
2082 * against other readers.
2084 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2085 !sosendallatonce(so) && nextrecord == NULL) {
2086 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2088 so->so_rcv.sb_state & SBS_CANTRCVMORE)
2091 * Notify the protocol that some data has been
2092 * drained before blocking.
2094 if (pr->pr_flags & PR_WANTRCVD) {
2095 SOCKBUF_UNLOCK(&so->so_rcv);
2097 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2098 SOCKBUF_LOCK(&so->so_rcv);
2100 SBLASTRECORDCHK(&so->so_rcv);
2101 SBLASTMBUFCHK(&so->so_rcv);
2103 * We could receive some data while was notifying
2104 * the protocol. Skip blocking in this case.
2106 if (so->so_rcv.sb_mb == NULL) {
2107 error = sbwait(&so->so_rcv);
2109 SOCKBUF_UNLOCK(&so->so_rcv);
2113 m = so->so_rcv.sb_mb;
2115 nextrecord = m->m_nextpkt;
2119 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2120 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2122 if ((flags & MSG_PEEK) == 0)
2123 (void) sbdroprecord_locked(&so->so_rcv);
2125 if ((flags & MSG_PEEK) == 0) {
2128 * First part is an inline SB_EMPTY_FIXUP(). Second
2129 * part makes sure sb_lastrecord is up-to-date if
2130 * there is still data in the socket buffer.
2132 so->so_rcv.sb_mb = nextrecord;
2133 if (so->so_rcv.sb_mb == NULL) {
2134 so->so_rcv.sb_mbtail = NULL;
2135 so->so_rcv.sb_lastrecord = NULL;
2136 } else if (nextrecord->m_nextpkt == NULL)
2137 so->so_rcv.sb_lastrecord = nextrecord;
2139 SBLASTRECORDCHK(&so->so_rcv);
2140 SBLASTMBUFCHK(&so->so_rcv);
2142 * If soreceive() is being done from the socket callback,
2143 * then don't need to generate ACK to peer to update window,
2144 * since ACK will be generated on return to TCP.
2146 if (!(flags & MSG_SOCALLBCK) &&
2147 (pr->pr_flags & PR_WANTRCVD)) {
2148 SOCKBUF_UNLOCK(&so->so_rcv);
2150 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2151 SOCKBUF_LOCK(&so->so_rcv);
2154 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2155 if (orig_resid == uio->uio_resid && orig_resid &&
2156 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2157 SOCKBUF_UNLOCK(&so->so_rcv);
2160 SOCKBUF_UNLOCK(&so->so_rcv);
2165 sbunlock(&so->so_rcv);
2170 * Optimized version of soreceive() for stream (TCP) sockets.
2173 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2174 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2176 int len = 0, error = 0, flags, oresid;
2178 struct mbuf *m, *n = NULL;
2180 /* We only do stream sockets. */
2181 if (so->so_type != SOCK_STREAM)
2186 flags = *flagsp &~ MSG_EOR;
2189 if (controlp != NULL)
2191 if (flags & MSG_OOB)
2192 return (soreceive_rcvoob(so, uio, flags));
2198 /* Prevent other readers from entering the socket. */
2199 error = sblock(sb, SBLOCKWAIT(flags));
2204 /* Easy one, no space to copyout anything. */
2205 if (uio->uio_resid == 0) {
2209 oresid = uio->uio_resid;
2211 /* We will never ever get anything unless we are or were connected. */
2212 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2218 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2220 /* Abort if socket has reported problems. */
2222 if (sbavail(sb) > 0)
2224 if (oresid > uio->uio_resid)
2226 error = so->so_error;
2227 if (!(flags & MSG_PEEK))
2232 /* Door is closed. Deliver what is left, if any. */
2233 if (sb->sb_state & SBS_CANTRCVMORE) {
2234 if (sbavail(sb) > 0)
2240 /* Socket buffer is empty and we shall not block. */
2241 if (sbavail(sb) == 0 &&
2242 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2247 /* Socket buffer got some data that we shall deliver now. */
2248 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2249 ((so->so_state & SS_NBIO) ||
2250 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2251 sbavail(sb) >= sb->sb_lowat ||
2252 sbavail(sb) >= uio->uio_resid ||
2253 sbavail(sb) >= sb->sb_hiwat) ) {
2257 /* On MSG_WAITALL we must wait until all data or error arrives. */
2258 if ((flags & MSG_WAITALL) &&
2259 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2263 * Wait and block until (more) data comes in.
2264 * NB: Drops the sockbuf lock during wait.
2272 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2273 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2274 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2278 uio->uio_td->td_ru.ru_msgrcv++;
2280 /* Fill uio until full or current end of socket buffer is reached. */
2281 len = min(uio->uio_resid, sbavail(sb));
2283 /* Dequeue as many mbufs as possible. */
2284 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2288 m_cat(*mp0, sb->sb_mb);
2290 m != NULL && m->m_len <= len;
2292 KASSERT(!(m->m_flags & M_NOTAVAIL),
2293 ("%s: m %p not available", __func__, m));
2295 uio->uio_resid -= m->m_len;
2301 sb->sb_lastrecord = sb->sb_mb;
2302 if (sb->sb_mb == NULL)
2305 /* Copy the remainder. */
2307 KASSERT(sb->sb_mb != NULL,
2308 ("%s: len > 0 && sb->sb_mb empty", __func__));
2310 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2312 len = 0; /* Don't flush data from sockbuf. */
2314 uio->uio_resid -= len;
2325 /* NB: Must unlock socket buffer as uiomove may sleep. */
2327 error = m_mbuftouio(uio, sb->sb_mb, len);
2332 SBLASTRECORDCHK(sb);
2336 * Remove the delivered data from the socket buffer unless we
2337 * were only peeking.
2339 if (!(flags & MSG_PEEK)) {
2341 sbdrop_locked(sb, len);
2343 /* Notify protocol that we drained some data. */
2344 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2345 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2346 !(flags & MSG_SOCALLBCK))) {
2349 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2355 * For MSG_WAITALL we may have to loop again and wait for
2356 * more data to come in.
2358 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2361 SOCKBUF_LOCK_ASSERT(sb);
2362 SBLASTRECORDCHK(sb);
2370 * Optimized version of soreceive() for simple datagram cases from userspace.
2371 * Unlike in the stream case, we're able to drop a datagram if copyout()
2372 * fails, and because we handle datagrams atomically, we don't need to use a
2373 * sleep lock to prevent I/O interlacing.
2376 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2377 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2379 struct mbuf *m, *m2;
2382 struct protosw *pr = so->so_proto;
2383 struct mbuf *nextrecord;
2387 if (controlp != NULL)
2390 flags = *flagsp &~ MSG_EOR;
2395 * For any complicated cases, fall back to the full
2396 * soreceive_generic().
2398 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2399 return (soreceive_generic(so, psa, uio, mp0, controlp,
2403 * Enforce restrictions on use.
2405 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2406 ("soreceive_dgram: wantrcvd"));
2407 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2408 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2409 ("soreceive_dgram: SBS_RCVATMARK"));
2410 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2411 ("soreceive_dgram: P_CONNREQUIRED"));
2414 * Loop blocking while waiting for a datagram.
2416 SOCKBUF_LOCK(&so->so_rcv);
2417 while ((m = so->so_rcv.sb_mb) == NULL) {
2418 KASSERT(sbavail(&so->so_rcv) == 0,
2419 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2420 sbavail(&so->so_rcv)));
2422 error = so->so_error;
2424 SOCKBUF_UNLOCK(&so->so_rcv);
2427 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2428 uio->uio_resid == 0) {
2429 SOCKBUF_UNLOCK(&so->so_rcv);
2432 if ((so->so_state & SS_NBIO) ||
2433 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2434 SOCKBUF_UNLOCK(&so->so_rcv);
2435 return (EWOULDBLOCK);
2437 SBLASTRECORDCHK(&so->so_rcv);
2438 SBLASTMBUFCHK(&so->so_rcv);
2439 error = sbwait(&so->so_rcv);
2441 SOCKBUF_UNLOCK(&so->so_rcv);
2445 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2448 uio->uio_td->td_ru.ru_msgrcv++;
2449 SBLASTRECORDCHK(&so->so_rcv);
2450 SBLASTMBUFCHK(&so->so_rcv);
2451 nextrecord = m->m_nextpkt;
2452 if (nextrecord == NULL) {
2453 KASSERT(so->so_rcv.sb_lastrecord == m,
2454 ("soreceive_dgram: lastrecord != m"));
2457 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2458 ("soreceive_dgram: m_nextpkt != nextrecord"));
2461 * Pull 'm' and its chain off the front of the packet queue.
2463 so->so_rcv.sb_mb = NULL;
2464 sockbuf_pushsync(&so->so_rcv, nextrecord);
2467 * Walk 'm's chain and free that many bytes from the socket buffer.
2469 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2470 sbfree(&so->so_rcv, m2);
2473 * Do a few last checks before we let go of the lock.
2475 SBLASTRECORDCHK(&so->so_rcv);
2476 SBLASTMBUFCHK(&so->so_rcv);
2477 SOCKBUF_UNLOCK(&so->so_rcv);
2479 if (pr->pr_flags & PR_ADDR) {
2480 KASSERT(m->m_type == MT_SONAME,
2481 ("m->m_type == %d", m->m_type));
2483 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2488 /* XXXRW: Can this happen? */
2493 * Packet to copyout() is now in 'm' and it is disconnected from the
2496 * Process one or more MT_CONTROL mbufs present before any data mbufs
2497 * in the first mbuf chain on the socket buffer. We call into the
2498 * protocol to perform externalization (or freeing if controlp ==
2499 * NULL). In some cases there can be only MT_CONTROL mbufs without
2502 if (m->m_type == MT_CONTROL) {
2503 struct mbuf *cm = NULL, *cmn;
2504 struct mbuf **cme = &cm;
2510 cme = &(*cme)->m_next;
2512 } while (m != NULL && m->m_type == MT_CONTROL);
2513 while (cm != NULL) {
2516 if (pr->pr_domain->dom_externalize != NULL) {
2517 error = (*pr->pr_domain->dom_externalize)
2518 (cm, controlp, flags);
2519 } else if (controlp != NULL)
2523 if (controlp != NULL) {
2524 while (*controlp != NULL)
2525 controlp = &(*controlp)->m_next;
2530 KASSERT(m == NULL || m->m_type == MT_DATA,
2531 ("soreceive_dgram: !data"));
2532 while (m != NULL && uio->uio_resid > 0) {
2533 len = uio->uio_resid;
2536 error = uiomove(mtod(m, char *), (int)len, uio);
2541 if (len == m->m_len)
2558 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2559 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2563 CURVNET_SET(so->so_vnet);
2564 if (!SOLISTENING(so))
2565 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2566 mp0, controlp, flagsp));
2574 soshutdown(struct socket *so, int how)
2576 struct protosw *pr = so->so_proto;
2577 int error, soerror_enotconn;
2579 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2582 soerror_enotconn = 0;
2584 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2586 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2587 * invoked on a datagram sockets, however historically we would
2588 * actually tear socket down. This is known to be leveraged by
2589 * some applications to unblock process waiting in recvXXX(2)
2590 * by other process that it shares that socket with. Try to meet
2591 * both backward-compatibility and POSIX requirements by forcing
2592 * ENOTCONN but still asking protocol to perform pru_shutdown().
2594 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so))
2596 soerror_enotconn = 1;
2599 if (SOLISTENING(so)) {
2600 if (how != SHUT_WR) {
2602 so->so_error = ECONNABORTED;
2603 solisten_wakeup(so); /* unlocks so */
2608 CURVNET_SET(so->so_vnet);
2609 if (pr->pr_usrreqs->pru_flush != NULL)
2610 (*pr->pr_usrreqs->pru_flush)(so, how);
2613 if (how != SHUT_RD) {
2614 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2615 wakeup(&so->so_timeo);
2617 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2619 wakeup(&so->so_timeo);
2623 return (soerror_enotconn ? ENOTCONN : 0);
2627 sorflush(struct socket *so)
2629 struct sockbuf *sb = &so->so_rcv;
2630 struct protosw *pr = so->so_proto;
2636 * In order to avoid calling dom_dispose with the socket buffer mutex
2637 * held, and in order to generally avoid holding the lock for a long
2638 * time, we make a copy of the socket buffer and clear the original
2639 * (except locks, state). The new socket buffer copy won't have
2640 * initialized locks so we can only call routines that won't use or
2641 * assert those locks.
2643 * Dislodge threads currently blocked in receive and wait to acquire
2644 * a lock against other simultaneous readers before clearing the
2645 * socket buffer. Don't let our acquire be interrupted by a signal
2646 * despite any existing socket disposition on interruptable waiting.
2649 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2652 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2653 * and mutex data unchanged.
2656 bzero(&aso, sizeof(aso));
2657 aso.so_pcb = so->so_pcb;
2658 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2659 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2660 bzero(&sb->sb_startzero,
2661 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2666 * Dispose of special rights and flush the copied socket. Don't call
2667 * any unsafe routines (that rely on locks being initialized) on aso.
2669 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2670 (*pr->pr_domain->dom_dispose)(&aso);
2671 sbrelease_internal(&aso.so_rcv, so);
2675 * Wrapper for Socket established helper hook.
2676 * Parameters: socket, context of the hook point, hook id.
2679 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2681 struct socket_hhook_data hhook_data = {
2688 CURVNET_SET(so->so_vnet);
2689 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2692 /* Ugly but needed, since hhooks return void for now */
2693 return (hhook_data.status);
2697 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2698 * additional variant to handle the case where the option value needs to be
2699 * some kind of integer, but not a specific size. In addition to their use
2700 * here, these functions are also called by the protocol-level pr_ctloutput()
2704 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2709 * If the user gives us more than we wanted, we ignore it, but if we
2710 * don't get the minimum length the caller wants, we return EINVAL.
2711 * On success, sopt->sopt_valsize is set to however much we actually
2714 if ((valsize = sopt->sopt_valsize) < minlen)
2717 sopt->sopt_valsize = valsize = len;
2719 if (sopt->sopt_td != NULL)
2720 return (copyin(sopt->sopt_val, buf, valsize));
2722 bcopy(sopt->sopt_val, buf, valsize);
2727 * Kernel version of setsockopt(2).
2729 * XXX: optlen is size_t, not socklen_t
2732 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2735 struct sockopt sopt;
2737 sopt.sopt_level = level;
2738 sopt.sopt_name = optname;
2739 sopt.sopt_dir = SOPT_SET;
2740 sopt.sopt_val = optval;
2741 sopt.sopt_valsize = optlen;
2742 sopt.sopt_td = NULL;
2743 return (sosetopt(so, &sopt));
2747 sosetopt(struct socket *so, struct sockopt *sopt)
2758 CURVNET_SET(so->so_vnet);
2760 if (sopt->sopt_level != SOL_SOCKET) {
2761 if (so->so_proto->pr_ctloutput != NULL) {
2762 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2766 error = ENOPROTOOPT;
2768 switch (sopt->sopt_name) {
2769 case SO_ACCEPTFILTER:
2770 error = accept_filt_setopt(so, sopt);
2776 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2779 if (l.l_linger < 0 ||
2780 l.l_linger > USHRT_MAX ||
2781 l.l_linger > (INT_MAX / hz)) {
2786 so->so_linger = l.l_linger;
2788 so->so_options |= SO_LINGER;
2790 so->so_options &= ~SO_LINGER;
2797 case SO_USELOOPBACK:
2801 case SO_REUSEPORT_LB:
2808 error = sooptcopyin(sopt, &optval, sizeof optval,
2814 so->so_options |= sopt->sopt_name;
2816 so->so_options &= ~sopt->sopt_name;
2821 error = sooptcopyin(sopt, &optval, sizeof optval,
2826 if (optval < 0 || optval >= rt_numfibs) {
2830 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2831 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2832 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2833 so->so_fibnum = optval;
2838 case SO_USER_COOKIE:
2839 error = sooptcopyin(sopt, &val32, sizeof val32,
2843 so->so_user_cookie = val32;
2850 error = sooptcopyin(sopt, &optval, sizeof optval,
2856 * Values < 1 make no sense for any of these options,
2864 error = sbsetopt(so, sopt->sopt_name, optval);
2869 #ifdef COMPAT_FREEBSD32
2870 if (SV_CURPROC_FLAG(SV_ILP32)) {
2871 struct timeval32 tv32;
2873 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2875 CP(tv32, tv, tv_sec);
2876 CP(tv32, tv, tv_usec);
2879 error = sooptcopyin(sopt, &tv, sizeof tv,
2883 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2884 tv.tv_usec >= 1000000) {
2888 if (tv.tv_sec > INT32_MAX)
2892 switch (sopt->sopt_name) {
2894 so->so_snd.sb_timeo = val;
2897 so->so_rcv.sb_timeo = val;
2904 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2908 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2916 error = sooptcopyin(sopt, &optval, sizeof optval,
2920 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
2924 so->so_ts_clock = optval;
2927 case SO_MAX_PACING_RATE:
2928 error = sooptcopyin(sopt, &val32, sizeof(val32),
2932 so->so_max_pacing_rate = val32;
2936 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2937 error = hhook_run_socket(so, sopt,
2940 error = ENOPROTOOPT;
2943 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2944 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
2952 * Helper routine for getsockopt.
2955 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2963 * Documented get behavior is that we always return a value, possibly
2964 * truncated to fit in the user's buffer. Traditional behavior is
2965 * that we always tell the user precisely how much we copied, rather
2966 * than something useful like the total amount we had available for
2967 * her. Note that this interface is not idempotent; the entire
2968 * answer must be generated ahead of time.
2970 valsize = min(len, sopt->sopt_valsize);
2971 sopt->sopt_valsize = valsize;
2972 if (sopt->sopt_val != NULL) {
2973 if (sopt->sopt_td != NULL)
2974 error = copyout(buf, sopt->sopt_val, valsize);
2976 bcopy(buf, sopt->sopt_val, valsize);
2982 sogetopt(struct socket *so, struct sockopt *sopt)
2991 CURVNET_SET(so->so_vnet);
2993 if (sopt->sopt_level != SOL_SOCKET) {
2994 if (so->so_proto->pr_ctloutput != NULL)
2995 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2997 error = ENOPROTOOPT;
3001 switch (sopt->sopt_name) {
3002 case SO_ACCEPTFILTER:
3003 error = accept_filt_getopt(so, sopt);
3008 l.l_onoff = so->so_options & SO_LINGER;
3009 l.l_linger = so->so_linger;
3011 error = sooptcopyout(sopt, &l, sizeof l);
3014 case SO_USELOOPBACK:
3020 case SO_REUSEPORT_LB:
3029 optval = so->so_options & sopt->sopt_name;
3031 error = sooptcopyout(sopt, &optval, sizeof optval);
3035 optval = so->so_proto->pr_domain->dom_family;
3039 optval = so->so_type;
3043 optval = so->so_proto->pr_protocol;
3048 optval = so->so_error;
3054 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3055 so->so_snd.sb_hiwat;
3059 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3060 so->so_rcv.sb_hiwat;
3064 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3065 so->so_snd.sb_lowat;
3069 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3070 so->so_rcv.sb_lowat;
3075 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3076 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3077 #ifdef COMPAT_FREEBSD32
3078 if (SV_CURPROC_FLAG(SV_ILP32)) {
3079 struct timeval32 tv32;
3081 CP(tv, tv32, tv_sec);
3082 CP(tv, tv32, tv_usec);
3083 error = sooptcopyout(sopt, &tv32, sizeof tv32);
3086 error = sooptcopyout(sopt, &tv, sizeof tv);
3091 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3095 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3099 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3107 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3111 error = mac_getsockopt_peerlabel(
3112 sopt->sopt_td->td_ucred, so, &extmac);
3115 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3121 case SO_LISTENQLIMIT:
3122 optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3126 optval = SOLISTENING(so) ? so->sol_qlen : 0;
3129 case SO_LISTENINCQLEN:
3130 optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3134 optval = so->so_ts_clock;
3137 case SO_MAX_PACING_RATE:
3138 optval = so->so_max_pacing_rate;
3142 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3143 error = hhook_run_socket(so, sopt,
3146 error = ENOPROTOOPT;
3158 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3160 struct mbuf *m, *m_prev;
3161 int sopt_size = sopt->sopt_valsize;
3163 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3166 if (sopt_size > MLEN) {
3167 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3168 if ((m->m_flags & M_EXT) == 0) {
3172 m->m_len = min(MCLBYTES, sopt_size);
3174 m->m_len = min(MLEN, sopt_size);
3176 sopt_size -= m->m_len;
3181 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3186 if (sopt_size > MLEN) {
3187 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3189 if ((m->m_flags & M_EXT) == 0) {
3194 m->m_len = min(MCLBYTES, sopt_size);
3196 m->m_len = min(MLEN, sopt_size);
3198 sopt_size -= m->m_len;
3206 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3208 struct mbuf *m0 = m;
3210 if (sopt->sopt_val == NULL)
3212 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3213 if (sopt->sopt_td != NULL) {
3216 error = copyin(sopt->sopt_val, mtod(m, char *),
3223 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3224 sopt->sopt_valsize -= m->m_len;
3225 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3228 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3229 panic("ip6_sooptmcopyin");
3234 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3236 struct mbuf *m0 = m;
3239 if (sopt->sopt_val == NULL)
3241 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3242 if (sopt->sopt_td != NULL) {
3245 error = copyout(mtod(m, char *), sopt->sopt_val,
3252 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3253 sopt->sopt_valsize -= m->m_len;
3254 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3255 valsize += m->m_len;
3259 /* enough soopt buffer should be given from user-land */
3263 sopt->sopt_valsize = valsize;
3268 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3269 * out-of-band data, which will then notify socket consumers.
3272 sohasoutofband(struct socket *so)
3275 if (so->so_sigio != NULL)
3276 pgsigio(&so->so_sigio, SIGURG, 0);
3277 selwakeuppri(&so->so_rdsel, PSOCK);
3281 sopoll(struct socket *so, int events, struct ucred *active_cred,
3286 * We do not need to set or assert curvnet as long as everyone uses
3289 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3294 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3300 if (SOLISTENING(so)) {
3301 if (!(events & (POLLIN | POLLRDNORM)))
3303 else if (!TAILQ_EMPTY(&so->sol_comp))
3304 revents = events & (POLLIN | POLLRDNORM);
3305 else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3306 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3308 selrecord(td, &so->so_rdsel);
3313 SOCKBUF_LOCK(&so->so_snd);
3314 SOCKBUF_LOCK(&so->so_rcv);
3315 if (events & (POLLIN | POLLRDNORM))
3316 if (soreadabledata(so))
3317 revents |= events & (POLLIN | POLLRDNORM);
3318 if (events & (POLLOUT | POLLWRNORM))
3319 if (sowriteable(so))
3320 revents |= events & (POLLOUT | POLLWRNORM);
3321 if (events & (POLLPRI | POLLRDBAND))
3322 if (so->so_oobmark ||
3323 (so->so_rcv.sb_state & SBS_RCVATMARK))
3324 revents |= events & (POLLPRI | POLLRDBAND);
3325 if ((events & POLLINIGNEOF) == 0) {
3326 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3327 revents |= events & (POLLIN | POLLRDNORM);
3328 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3334 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3335 selrecord(td, &so->so_rdsel);
3336 so->so_rcv.sb_flags |= SB_SEL;
3338 if (events & (POLLOUT | POLLWRNORM)) {
3339 selrecord(td, &so->so_wrsel);
3340 so->so_snd.sb_flags |= SB_SEL;
3343 SOCKBUF_UNLOCK(&so->so_rcv);
3344 SOCKBUF_UNLOCK(&so->so_snd);
3351 soo_kqfilter(struct file *fp, struct knote *kn)
3353 struct socket *so = kn->kn_fp->f_data;
3357 switch (kn->kn_filter) {
3359 kn->kn_fop = &soread_filtops;
3360 knl = &so->so_rdsel.si_note;
3364 kn->kn_fop = &sowrite_filtops;
3365 knl = &so->so_wrsel.si_note;
3369 kn->kn_fop = &soempty_filtops;
3370 knl = &so->so_wrsel.si_note;
3378 if (SOLISTENING(so)) {
3379 knlist_add(knl, kn, 1);
3382 knlist_add(knl, kn, 1);
3383 sb->sb_flags |= SB_KNOTE;
3391 * Some routines that return EOPNOTSUPP for entry points that are not
3392 * supported by a protocol. Fill in as needed.
3395 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3402 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3409 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3416 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3423 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3431 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3438 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3446 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3453 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3454 struct ifnet *ifp, struct thread *td)
3461 pru_disconnect_notsupp(struct socket *so)
3468 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3475 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3482 pru_rcvd_notsupp(struct socket *so, int flags)
3489 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3496 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3497 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3504 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3507 return (EOPNOTSUPP);
3511 * This isn't really a ``null'' operation, but it's the default one and
3512 * doesn't do anything destructive.
3515 pru_sense_null(struct socket *so, struct stat *sb)
3518 sb->st_blksize = so->so_snd.sb_hiwat;
3523 pru_shutdown_notsupp(struct socket *so)
3530 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3537 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3538 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3545 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3546 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3553 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3561 filt_sordetach(struct knote *kn)
3563 struct socket *so = kn->kn_fp->f_data;
3566 knlist_remove(&so->so_rdsel.si_note, kn, 1);
3567 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3568 so->so_rcv.sb_flags &= ~SB_KNOTE;
3569 so_rdknl_unlock(so);
3574 filt_soread(struct knote *kn, long hint)
3578 so = kn->kn_fp->f_data;
3580 if (SOLISTENING(so)) {
3581 SOCK_LOCK_ASSERT(so);
3582 kn->kn_data = so->sol_qlen;
3584 kn->kn_flags |= EV_EOF;
3585 kn->kn_fflags = so->so_error;
3588 return (!TAILQ_EMPTY(&so->sol_comp));
3591 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3593 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3594 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3595 kn->kn_flags |= EV_EOF;
3596 kn->kn_fflags = so->so_error;
3598 } else if (so->so_error) /* temporary udp error */
3601 if (kn->kn_sfflags & NOTE_LOWAT) {
3602 if (kn->kn_data >= kn->kn_sdata)
3604 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3607 /* This hook returning non-zero indicates an event, not error */
3608 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3612 filt_sowdetach(struct knote *kn)
3614 struct socket *so = kn->kn_fp->f_data;
3617 knlist_remove(&so->so_wrsel.si_note, kn, 1);
3618 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3619 so->so_snd.sb_flags &= ~SB_KNOTE;
3620 so_wrknl_unlock(so);
3625 filt_sowrite(struct knote *kn, long hint)
3629 so = kn->kn_fp->f_data;
3631 if (SOLISTENING(so))
3634 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3635 kn->kn_data = sbspace(&so->so_snd);
3637 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3639 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3640 kn->kn_flags |= EV_EOF;
3641 kn->kn_fflags = so->so_error;
3643 } else if (so->so_error) /* temporary udp error */
3645 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3646 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3648 else if (kn->kn_sfflags & NOTE_LOWAT)
3649 return (kn->kn_data >= kn->kn_sdata);
3651 return (kn->kn_data >= so->so_snd.sb_lowat);
3655 filt_soempty(struct knote *kn, long hint)
3659 so = kn->kn_fp->f_data;
3661 if (SOLISTENING(so))
3664 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3665 kn->kn_data = sbused(&so->so_snd);
3667 if (kn->kn_data == 0)
3674 socheckuid(struct socket *so, uid_t uid)
3679 if (so->so_cred->cr_uid != uid)
3685 * These functions are used by protocols to notify the socket layer (and its
3686 * consumers) of state changes in the sockets driven by protocol-side events.
3690 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3692 * Normal sequence from the active (originating) side is that
3693 * soisconnecting() is called during processing of connect() call, resulting
3694 * in an eventual call to soisconnected() if/when the connection is
3695 * established. When the connection is torn down soisdisconnecting() is
3696 * called during processing of disconnect() call, and soisdisconnected() is
3697 * called when the connection to the peer is totally severed. The semantics
3698 * of these routines are such that connectionless protocols can call
3699 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3700 * calls when setting up a ``connection'' takes no time.
3702 * From the passive side, a socket is created with two queues of sockets:
3703 * so_incomp for connections in progress and so_comp for connections already
3704 * made and awaiting user acceptance. As a protocol is preparing incoming
3705 * connections, it creates a socket structure queued on so_incomp by calling
3706 * sonewconn(). When the connection is established, soisconnected() is
3707 * called, and transfers the socket structure to so_comp, making it available
3710 * If a socket is closed with sockets on either so_incomp or so_comp, these
3711 * sockets are dropped.
3713 * If higher-level protocols are implemented in the kernel, the wakeups done
3714 * here will sometimes cause software-interrupt process scheduling.
3717 soisconnecting(struct socket *so)
3721 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3722 so->so_state |= SS_ISCONNECTING;
3727 soisconnected(struct socket *so)
3731 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3732 so->so_state |= SS_ISCONNECTED;
3734 if (so->so_qstate == SQ_INCOMP) {
3735 struct socket *head = so->so_listen;
3738 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3740 * Promoting a socket from incomplete queue to complete, we
3741 * need to go through reverse order of locking. We first do
3742 * trylock, and if that doesn't succeed, we go the hard way
3743 * leaving a reference and rechecking consistency after proper
3746 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3749 SOLISTEN_LOCK(head);
3751 if (__predict_false(head != so->so_listen)) {
3753 * The socket went off the listen queue,
3754 * should be lost race to close(2) of sol.
3755 * The socket is about to soabort().
3761 /* Not the last one, as so holds a ref. */
3762 refcount_release(&head->so_count);
3765 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3766 TAILQ_REMOVE(&head->sol_incomp, so, so_list);
3767 head->sol_incqlen--;
3768 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
3770 so->so_qstate = SQ_COMP;
3772 solisten_wakeup(head); /* unlocks */
3774 SOCKBUF_LOCK(&so->so_rcv);
3775 soupcall_set(so, SO_RCV,
3776 head->sol_accept_filter->accf_callback,
3777 head->sol_accept_filter_arg);
3778 so->so_options &= ~SO_ACCEPTFILTER;
3779 ret = head->sol_accept_filter->accf_callback(so,
3780 head->sol_accept_filter_arg, M_NOWAIT);
3781 if (ret == SU_ISCONNECTED) {
3782 soupcall_clear(so, SO_RCV);
3783 SOCKBUF_UNLOCK(&so->so_rcv);
3786 SOCKBUF_UNLOCK(&so->so_rcv);
3788 SOLISTEN_UNLOCK(head);
3793 wakeup(&so->so_timeo);
3799 soisdisconnecting(struct socket *so)
3803 so->so_state &= ~SS_ISCONNECTING;
3804 so->so_state |= SS_ISDISCONNECTING;
3806 if (!SOLISTENING(so)) {
3807 SOCKBUF_LOCK(&so->so_rcv);
3808 socantrcvmore_locked(so);
3809 SOCKBUF_LOCK(&so->so_snd);
3810 socantsendmore_locked(so);
3813 wakeup(&so->so_timeo);
3817 soisdisconnected(struct socket *so)
3823 * There is at least one reader of so_state that does not
3824 * acquire socket lock, namely soreceive_generic(). Ensure
3825 * that it never sees all flags that track connection status
3826 * cleared, by ordering the update with a barrier semantic of
3827 * our release thread fence.
3829 so->so_state |= SS_ISDISCONNECTED;
3830 atomic_thread_fence_rel();
3831 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3833 if (!SOLISTENING(so)) {
3835 SOCKBUF_LOCK(&so->so_rcv);
3836 socantrcvmore_locked(so);
3837 SOCKBUF_LOCK(&so->so_snd);
3838 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
3839 socantsendmore_locked(so);
3842 wakeup(&so->so_timeo);
3846 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3849 sodupsockaddr(const struct sockaddr *sa, int mflags)
3851 struct sockaddr *sa2;
3853 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3855 bcopy(sa, sa2, sa->sa_len);
3860 * Register per-socket destructor.
3863 sodtor_set(struct socket *so, so_dtor_t *func)
3866 SOCK_LOCK_ASSERT(so);
3871 * Register per-socket buffer upcalls.
3874 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
3878 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
3888 panic("soupcall_set: bad which");
3890 SOCKBUF_LOCK_ASSERT(sb);
3891 sb->sb_upcall = func;
3892 sb->sb_upcallarg = arg;
3893 sb->sb_flags |= SB_UPCALL;
3897 soupcall_clear(struct socket *so, int which)
3901 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
3911 panic("soupcall_clear: bad which");
3913 SOCKBUF_LOCK_ASSERT(sb);
3914 KASSERT(sb->sb_upcall != NULL,
3915 ("%s: so %p no upcall to clear", __func__, so));
3916 sb->sb_upcall = NULL;
3917 sb->sb_upcallarg = NULL;
3918 sb->sb_flags &= ~SB_UPCALL;
3922 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
3925 SOLISTEN_LOCK_ASSERT(so);
3926 so->sol_upcall = func;
3927 so->sol_upcallarg = arg;
3931 so_rdknl_lock(void *arg)
3933 struct socket *so = arg;
3935 if (SOLISTENING(so))
3938 SOCKBUF_LOCK(&so->so_rcv);
3942 so_rdknl_unlock(void *arg)
3944 struct socket *so = arg;
3946 if (SOLISTENING(so))
3949 SOCKBUF_UNLOCK(&so->so_rcv);
3953 so_rdknl_assert_locked(void *arg)
3955 struct socket *so = arg;
3957 if (SOLISTENING(so))
3958 SOCK_LOCK_ASSERT(so);
3960 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3964 so_rdknl_assert_unlocked(void *arg)
3966 struct socket *so = arg;
3968 if (SOLISTENING(so))
3969 SOCK_UNLOCK_ASSERT(so);
3971 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
3975 so_wrknl_lock(void *arg)
3977 struct socket *so = arg;
3979 if (SOLISTENING(so))
3982 SOCKBUF_LOCK(&so->so_snd);
3986 so_wrknl_unlock(void *arg)
3988 struct socket *so = arg;
3990 if (SOLISTENING(so))
3993 SOCKBUF_UNLOCK(&so->so_snd);
3997 so_wrknl_assert_locked(void *arg)
3999 struct socket *so = arg;
4001 if (SOLISTENING(so))
4002 SOCK_LOCK_ASSERT(so);
4004 SOCKBUF_LOCK_ASSERT(&so->so_snd);
4008 so_wrknl_assert_unlocked(void *arg)
4010 struct socket *so = arg;
4012 if (SOLISTENING(so))
4013 SOCK_UNLOCK_ASSERT(so);
4015 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4019 * Create an external-format (``xsocket'') structure using the information in
4020 * the kernel-format socket structure pointed to by so. This is done to
4021 * reduce the spew of irrelevant information over this interface, to isolate
4022 * user code from changes in the kernel structure, and potentially to provide
4023 * information-hiding if we decide that some of this information should be
4024 * hidden from users.
4027 sotoxsocket(struct socket *so, struct xsocket *xso)
4030 bzero(xso, sizeof(*xso));
4031 xso->xso_len = sizeof *xso;
4032 xso->xso_so = (uintptr_t)so;
4033 xso->so_type = so->so_type;
4034 xso->so_options = so->so_options;
4035 xso->so_linger = so->so_linger;
4036 xso->so_state = so->so_state;
4037 xso->so_pcb = (uintptr_t)so->so_pcb;
4038 xso->xso_protocol = so->so_proto->pr_protocol;
4039 xso->xso_family = so->so_proto->pr_domain->dom_family;
4040 xso->so_timeo = so->so_timeo;
4041 xso->so_error = so->so_error;
4042 xso->so_uid = so->so_cred->cr_uid;
4043 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4044 if (SOLISTENING(so)) {
4045 xso->so_qlen = so->sol_qlen;
4046 xso->so_incqlen = so->sol_incqlen;
4047 xso->so_qlimit = so->sol_qlimit;
4048 xso->so_oobmark = 0;
4050 xso->so_state |= so->so_qstate;
4051 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4052 xso->so_oobmark = so->so_oobmark;
4053 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4054 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4059 so_sockbuf_rcv(struct socket *so)
4062 return (&so->so_rcv);
4066 so_sockbuf_snd(struct socket *so)
4069 return (&so->so_snd);
4073 so_state_get(const struct socket *so)
4076 return (so->so_state);
4080 so_state_set(struct socket *so, int val)
4087 so_options_get(const struct socket *so)
4090 return (so->so_options);
4094 so_options_set(struct socket *so, int val)
4097 so->so_options = val;
4101 so_error_get(const struct socket *so)
4104 return (so->so_error);
4108 so_error_set(struct socket *so, int val)
4115 so_linger_get(const struct socket *so)
4118 return (so->so_linger);
4122 so_linger_set(struct socket *so, int val)
4125 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4126 ("%s: val %d out of range", __func__, val));
4128 so->so_linger = val;
4132 so_protosw_get(const struct socket *so)
4135 return (so->so_proto);
4139 so_protosw_set(struct socket *so, struct protosw *val)
4146 so_sorwakeup(struct socket *so)
4153 so_sowwakeup(struct socket *so)
4160 so_sorwakeup_locked(struct socket *so)
4163 sorwakeup_locked(so);
4167 so_sowwakeup_locked(struct socket *so)
4170 sowwakeup_locked(so);
4174 so_lock(struct socket *so)
4181 so_unlock(struct socket *so)