2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993
5 * The Regents of the University of California.
6 * Copyright (c) 2004 The FreeBSD Foundation
7 * Copyright (c) 2004-2008 Robert N. M. Watson
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
38 * Comments on the socket life cycle:
40 * soalloc() sets of socket layer state for a socket, called only by
41 * socreate() and sonewconn(). Socket layer private.
43 * sodealloc() tears down socket layer state for a socket, called only by
44 * sofree() and sonewconn(). Socket layer private.
46 * pru_attach() associates protocol layer state with an allocated socket;
47 * called only once, may fail, aborting socket allocation. This is called
48 * from socreate() and sonewconn(). Socket layer private.
50 * pru_detach() disassociates protocol layer state from an attached socket,
51 * and will be called exactly once for sockets in which pru_attach() has
52 * been successfully called. If pru_attach() returned an error,
53 * pru_detach() will not be called. Socket layer private.
55 * pru_abort() and pru_close() notify the protocol layer that the last
56 * consumer of a socket is starting to tear down the socket, and that the
57 * protocol should terminate the connection. Historically, pru_abort() also
58 * detached protocol state from the socket state, but this is no longer the
61 * socreate() creates a socket and attaches protocol state. This is a public
62 * interface that may be used by socket layer consumers to create new
65 * sonewconn() creates a socket and attaches protocol state. This is a
66 * public interface that may be used by protocols to create new sockets when
67 * a new connection is received and will be available for accept() on a
70 * soclose() destroys a socket after possibly waiting for it to disconnect.
71 * This is a public interface that socket consumers should use to close and
72 * release a socket when done with it.
74 * soabort() destroys a socket without waiting for it to disconnect (used
75 * only for incoming connections that are already partially or fully
76 * connected). This is used internally by the socket layer when clearing
77 * listen socket queues (due to overflow or close on the listen socket), but
78 * is also a public interface protocols may use to abort connections in
79 * their incomplete listen queues should they no longer be required. Sockets
80 * placed in completed connection listen queues should not be aborted for
81 * reasons described in the comment above the soclose() implementation. This
82 * is not a general purpose close routine, and except in the specific
83 * circumstances described here, should not be used.
85 * sofree() will free a socket and its protocol state if all references on
86 * the socket have been released, and is the public interface to attempt to
87 * free a socket when a reference is removed. This is a socket layer private
90 * NOTE: In addition to socreate() and soclose(), which provide a single
91 * socket reference to the consumer to be managed as required, there are two
92 * calls to explicitly manage socket references, soref(), and sorele().
93 * Currently, these are generally required only when transitioning a socket
94 * from a listen queue to a file descriptor, in order to prevent garbage
95 * collection of the socket at an untimely moment. For a number of reasons,
96 * these interfaces are not preferred, and should be avoided.
98 * NOTE: With regard to VNETs the general rule is that callers do not set
99 * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
100 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
101 * and sorflush(), which are usually called from a pre-set VNET context.
102 * sopoll() currently does not need a VNET context to be set.
105 #include <sys/cdefs.h>
106 __FBSDID("$FreeBSD$");
108 #include "opt_inet.h"
109 #include "opt_inet6.h"
110 #include "opt_kern_tls.h"
111 #include "opt_sctp.h"
113 #include <sys/param.h>
114 #include <sys/systm.h>
115 #include <sys/capsicum.h>
116 #include <sys/fcntl.h>
117 #include <sys/limits.h>
118 #include <sys/lock.h>
120 #include <sys/malloc.h>
121 #include <sys/mbuf.h>
122 #include <sys/mutex.h>
123 #include <sys/domain.h>
124 #include <sys/file.h> /* for struct knote */
125 #include <sys/hhook.h>
126 #include <sys/kernel.h>
127 #include <sys/khelp.h>
128 #include <sys/ktls.h>
129 #include <sys/event.h>
130 #include <sys/eventhandler.h>
131 #include <sys/poll.h>
132 #include <sys/proc.h>
133 #include <sys/protosw.h>
134 #include <sys/sbuf.h>
135 #include <sys/socket.h>
136 #include <sys/socketvar.h>
137 #include <sys/resourcevar.h>
138 #include <net/route.h>
139 #include <sys/signalvar.h>
140 #include <sys/stat.h>
142 #include <sys/sysctl.h>
143 #include <sys/taskqueue.h>
146 #include <sys/unpcb.h>
147 #include <sys/jail.h>
148 #include <sys/syslog.h>
149 #include <netinet/in.h>
150 #include <netinet/in_pcb.h>
151 #include <netinet/tcp.h>
153 #include <net/vnet.h>
155 #include <security/mac/mac_framework.h>
159 #ifdef COMPAT_FREEBSD32
160 #include <sys/mount.h>
161 #include <sys/sysent.h>
162 #include <compat/freebsd32/freebsd32.h>
165 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
167 static void so_rdknl_lock(void *);
168 static void so_rdknl_unlock(void *);
169 static void so_rdknl_assert_lock(void *, int);
170 static void so_wrknl_lock(void *);
171 static void so_wrknl_unlock(void *);
172 static void so_wrknl_assert_lock(void *, int);
174 static void filt_sordetach(struct knote *kn);
175 static int filt_soread(struct knote *kn, long hint);
176 static void filt_sowdetach(struct knote *kn);
177 static int filt_sowrite(struct knote *kn, long hint);
178 static int filt_soempty(struct knote *kn, long hint);
179 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
180 fo_kqfilter_t soo_kqfilter;
182 static struct filterops soread_filtops = {
184 .f_detach = filt_sordetach,
185 .f_event = filt_soread,
187 static struct filterops sowrite_filtops = {
189 .f_detach = filt_sowdetach,
190 .f_event = filt_sowrite,
192 static struct filterops soempty_filtops = {
194 .f_detach = filt_sowdetach,
195 .f_event = filt_soempty,
198 so_gen_t so_gencnt; /* generation count for sockets */
200 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
201 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
203 #define VNET_SO_ASSERT(so) \
204 VNET_ASSERT(curvnet != NULL, \
205 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
207 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
208 #define V_socket_hhh VNET(socket_hhh)
211 * Limit on the number of connections in the listen queue waiting
213 * NB: The original sysctl somaxconn is still available but hidden
214 * to prevent confusion about the actual purpose of this number.
216 static u_int somaxconn = SOMAXCONN;
219 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
225 error = sysctl_handle_int(oidp, &val, 0, req);
226 if (error || !req->newptr )
230 * The purpose of the UINT_MAX / 3 limit, is so that the formula
232 * below, will not overflow.
235 if (val < 1 || val > UINT_MAX / 3)
241 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
242 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, sizeof(int),
243 sysctl_somaxconn, "I",
244 "Maximum listen socket pending connection accept queue size");
245 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
246 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, 0,
247 sizeof(int), sysctl_somaxconn, "I",
248 "Maximum listen socket pending connection accept queue size (compat)");
250 static int numopensockets;
251 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
252 &numopensockets, 0, "Number of open sockets");
255 * accept_mtx locks down per-socket fields relating to accept queues. See
256 * socketvar.h for an annotation of the protected fields of struct socket.
258 struct mtx accept_mtx;
259 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
262 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
265 static struct mtx so_global_mtx;
266 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
269 * General IPC sysctl name space, used by sockets and a variety of other IPC
272 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
276 * Initialize the socket subsystem and set up the socket
279 static uma_zone_t socket_zone;
283 socket_zone_change(void *tag)
286 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
290 socket_hhook_register(int subtype)
293 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
294 &V_socket_hhh[subtype],
295 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
296 printf("%s: WARNING: unable to register hook\n", __func__);
300 socket_hhook_deregister(int subtype)
303 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
304 printf("%s: WARNING: unable to deregister hook\n", __func__);
308 socket_init(void *tag)
311 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
312 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
313 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
314 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
315 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
316 EVENTHANDLER_PRI_FIRST);
318 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
321 socket_vnet_init(const void *unused __unused)
325 /* We expect a contiguous range */
326 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
327 socket_hhook_register(i);
329 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
330 socket_vnet_init, NULL);
333 socket_vnet_uninit(const void *unused __unused)
337 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
338 socket_hhook_deregister(i);
340 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
341 socket_vnet_uninit, NULL);
344 * Initialise maxsockets. This SYSINIT must be run after
348 init_maxsockets(void *ignored)
351 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
352 maxsockets = imax(maxsockets, maxfiles);
354 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
357 * Sysctl to get and set the maximum global sockets limit. Notify protocols
358 * of the change so that they can update their dependent limits as required.
361 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
363 int error, newmaxsockets;
365 newmaxsockets = maxsockets;
366 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
367 if (error == 0 && req->newptr) {
368 if (newmaxsockets > maxsockets &&
369 newmaxsockets <= maxfiles) {
370 maxsockets = newmaxsockets;
371 EVENTHANDLER_INVOKE(maxsockets_change);
377 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
378 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &maxsockets, 0,
379 sysctl_maxsockets, "IU",
380 "Maximum number of sockets available");
383 * Socket operation routines. These routines are called by the routines in
384 * sys_socket.c or from a system process, and implement the semantics of
385 * socket operations by switching out to the protocol specific routines.
389 * Get a socket structure from our zone, and initialize it. Note that it
390 * would probably be better to allocate socket and PCB at the same time, but
391 * I'm not convinced that all the protocols can be easily modified to do
394 * soalloc() returns a socket with a ref count of 0.
396 static struct socket *
397 soalloc(struct vnet *vnet)
401 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
405 if (mac_socket_init(so, M_NOWAIT) != 0) {
406 uma_zfree(socket_zone, so);
410 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
411 uma_zfree(socket_zone, so);
416 * The socket locking protocol allows to lock 2 sockets at a time,
417 * however, the first one must be a listening socket. WITNESS lacks
418 * a feature to change class of an existing lock, so we use DUPOK.
420 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
421 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
422 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
423 so->so_rcv.sb_sel = &so->so_rdsel;
424 so->so_snd.sb_sel = &so->so_wrsel;
425 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
426 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
427 TAILQ_INIT(&so->so_snd.sb_aiojobq);
428 TAILQ_INIT(&so->so_rcv.sb_aiojobq);
429 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
430 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
432 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
433 __func__, __LINE__, so));
436 /* We shouldn't need the so_global_mtx */
437 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
438 /* Do we need more comprehensive error returns? */
439 uma_zfree(socket_zone, so);
442 mtx_lock(&so_global_mtx);
443 so->so_gencnt = ++so_gencnt;
446 vnet->vnet_sockcnt++;
448 mtx_unlock(&so_global_mtx);
454 * Free the storage associated with a socket at the socket layer, tear down
455 * locks, labels, etc. All protocol state is assumed already to have been
456 * torn down (and possibly never set up) by the caller.
459 sodealloc(struct socket *so)
462 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
463 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
465 mtx_lock(&so_global_mtx);
466 so->so_gencnt = ++so_gencnt;
467 --numopensockets; /* Could be below, but faster here. */
469 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
470 __func__, __LINE__, so));
471 so->so_vnet->vnet_sockcnt--;
473 mtx_unlock(&so_global_mtx);
475 mac_socket_destroy(so);
477 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
479 khelp_destroy_osd(&so->osd);
480 if (SOLISTENING(so)) {
481 if (so->sol_accept_filter != NULL)
482 accept_filt_setopt(so, NULL);
484 if (so->so_rcv.sb_hiwat)
485 (void)chgsbsize(so->so_cred->cr_uidinfo,
486 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
487 if (so->so_snd.sb_hiwat)
488 (void)chgsbsize(so->so_cred->cr_uidinfo,
489 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
490 sx_destroy(&so->so_snd.sb_sx);
491 sx_destroy(&so->so_rcv.sb_sx);
492 SOCKBUF_LOCK_DESTROY(&so->so_snd);
493 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
496 mtx_destroy(&so->so_lock);
497 uma_zfree(socket_zone, so);
501 * socreate returns a socket with a ref count of 1. The socket should be
502 * closed with soclose().
505 socreate(int dom, struct socket **aso, int type, int proto,
506 struct ucred *cred, struct thread *td)
513 prp = pffindproto(dom, proto, type);
515 prp = pffindtype(dom, type);
518 /* No support for domain. */
519 if (pffinddomain(dom) == NULL)
520 return (EAFNOSUPPORT);
521 /* No support for socket type. */
522 if (proto == 0 && type != 0)
524 return (EPROTONOSUPPORT);
526 if (prp->pr_usrreqs->pru_attach == NULL ||
527 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
528 return (EPROTONOSUPPORT);
530 if (IN_CAPABILITY_MODE(td) && (prp->pr_flags & PR_CAPATTACH) == 0)
533 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
534 return (EPROTONOSUPPORT);
536 if (prp->pr_type != type)
538 so = soalloc(CRED_TO_VNET(cred));
543 so->so_cred = crhold(cred);
544 if ((prp->pr_domain->dom_family == PF_INET) ||
545 (prp->pr_domain->dom_family == PF_INET6) ||
546 (prp->pr_domain->dom_family == PF_ROUTE))
547 so->so_fibnum = td->td_proc->p_fibnum;
552 mac_socket_create(cred, so);
554 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
555 so_rdknl_assert_lock);
556 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
557 so_wrknl_assert_lock);
559 * Auto-sizing of socket buffers is managed by the protocols and
560 * the appropriate flags must be set in the pru_attach function.
562 CURVNET_SET(so->so_vnet);
563 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
575 static int regression_sonewconn_earlytest = 1;
576 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
577 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
580 static struct timeval overinterval = { 60, 0 };
581 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
583 "Delay in seconds between warnings for listen socket overflows");
586 * When an attempt at a new connection is noted on a socket which accepts
587 * connections, sonewconn is called. If the connection is possible (subject
588 * to space constraints, etc.) then we allocate a new structure, properly
589 * linked into the data structure of the original socket, and return this.
590 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
592 * Note: the ref count on the socket is 0 on return.
595 sonewconn(struct socket *head, int connstatus)
601 const char localprefix[] = "local:";
602 char descrbuf[SUNPATHLEN + sizeof(localprefix)];
604 char addrbuf[INET6_ADDRSTRLEN];
606 char addrbuf[INET_ADDRSTRLEN];
611 over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
613 if (regression_sonewconn_earlytest && over) {
617 head->sol_overcount++;
618 dolog = !!ratecheck(&head->sol_lastover, &overinterval);
621 * If we're going to log, copy the overflow count and queue
622 * length from the listen socket before dropping the lock.
623 * Also, reset the overflow count.
626 overcount = head->sol_overcount;
627 head->sol_overcount = 0;
628 qlen = head->sol_qlen;
630 SOLISTEN_UNLOCK(head);
634 * Try to print something descriptive about the
635 * socket for the error message.
637 sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
639 switch (head->so_proto->pr_domain->dom_family) {
640 #if defined(INET) || defined(INET6)
646 if (head->so_proto->pr_domain->dom_family ==
648 (sotoinpcb(head)->inp_inc.inc_flags &
651 &sotoinpcb(head)->inp_inc.inc6_laddr);
652 sbuf_printf(&descrsb, "[%s]", addrbuf);
658 sotoinpcb(head)->inp_inc.inc_laddr,
660 sbuf_cat(&descrsb, addrbuf);
663 sbuf_printf(&descrsb, ":%hu (proto %u)",
664 ntohs(sotoinpcb(head)->inp_inc.inc_lport),
665 head->so_proto->pr_protocol);
667 #endif /* INET || INET6 */
669 sbuf_cat(&descrsb, localprefix);
670 if (sotounpcb(head)->unp_addr != NULL)
672 sotounpcb(head)->unp_addr->sun_len -
673 offsetof(struct sockaddr_un,
679 sotounpcb(head)->unp_addr->sun_path,
682 sbuf_cat(&descrsb, "(unknown)");
687 * If we can't print something more specific, at least
688 * print the domain name.
690 if (sbuf_finish(&descrsb) != 0 ||
691 sbuf_len(&descrsb) <= 0) {
692 sbuf_clear(&descrsb);
694 head->so_proto->pr_domain->dom_name ?:
696 sbuf_finish(&descrsb);
698 KASSERT(sbuf_len(&descrsb) > 0,
699 ("%s: sbuf creation failed", __func__));
701 "%s: pcb %p (%s): Listen queue overflow: "
702 "%i already in queue awaiting acceptance "
703 "(%d occurrences)\n",
704 __func__, head->so_pcb, sbuf_data(&descrsb),
706 sbuf_delete(&descrsb);
713 SOLISTEN_UNLOCK(head);
714 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
716 so = soalloc(head->so_vnet);
718 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
719 "limit reached or out of memory\n",
720 __func__, head->so_pcb);
723 so->so_listen = head;
724 so->so_type = head->so_type;
725 so->so_options = head->so_options & ~SO_ACCEPTCONN;
726 so->so_linger = head->so_linger;
727 so->so_state = head->so_state | SS_NOFDREF;
728 so->so_fibnum = head->so_fibnum;
729 so->so_proto = head->so_proto;
730 so->so_cred = crhold(head->so_cred);
732 mac_socket_newconn(head, so);
734 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
735 so_rdknl_assert_lock);
736 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
737 so_wrknl_assert_lock);
738 VNET_SO_ASSERT(head);
739 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
741 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
742 __func__, head->so_pcb);
745 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
747 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
748 __func__, head->so_pcb);
751 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
752 so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
753 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
754 so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
755 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
756 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
759 if (head->sol_accept_filter != NULL)
761 so->so_state |= connstatus;
762 soref(head); /* A socket on (in)complete queue refs head. */
764 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
765 so->so_qstate = SQ_COMP;
767 solisten_wakeup(head); /* unlocks */
770 * Keep removing sockets from the head until there's room for
771 * us to insert on the tail. In pre-locking revisions, this
772 * was a simple if(), but as we could be racing with other
773 * threads and soabort() requires dropping locks, we must
774 * loop waiting for the condition to be true.
776 while (head->sol_incqlen > head->sol_qlimit) {
779 sp = TAILQ_FIRST(&head->sol_incomp);
780 TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
783 sp->so_qstate = SQ_NONE;
784 sp->so_listen = NULL;
786 sorele(head); /* does SOLISTEN_UNLOCK, head stays */
790 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
791 so->so_qstate = SQ_INCOMP;
793 SOLISTEN_UNLOCK(head);
798 #if defined(SCTP) || defined(SCTP_SUPPORT)
800 * Socket part of sctp_peeloff(). Detach a new socket from an
801 * association. The new socket is returned with a reference.
804 sopeeloff(struct socket *head)
808 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
809 __func__, __LINE__, head));
810 so = soalloc(head->so_vnet);
812 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
813 "limit reached or out of memory\n",
814 __func__, head->so_pcb);
817 so->so_type = head->so_type;
818 so->so_options = head->so_options;
819 so->so_linger = head->so_linger;
820 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
821 so->so_fibnum = head->so_fibnum;
822 so->so_proto = head->so_proto;
823 so->so_cred = crhold(head->so_cred);
825 mac_socket_newconn(head, so);
827 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
828 so_rdknl_assert_lock);
829 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
830 so_wrknl_assert_lock);
831 VNET_SO_ASSERT(head);
832 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
834 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
835 __func__, head->so_pcb);
838 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
840 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
841 __func__, head->so_pcb);
844 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
845 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
846 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
847 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
848 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
849 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
858 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
862 CURVNET_SET(so->so_vnet);
863 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
869 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
873 CURVNET_SET(so->so_vnet);
874 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
880 * solisten() transitions a socket from a non-listening state to a listening
881 * state, but can also be used to update the listen queue depth on an
882 * existing listen socket. The protocol will call back into the sockets
883 * layer using solisten_proto_check() and solisten_proto() to check and set
884 * socket-layer listen state. Call backs are used so that the protocol can
885 * acquire both protocol and socket layer locks in whatever order is required
888 * Protocol implementors are advised to hold the socket lock across the
889 * socket-layer test and set to avoid races at the socket layer.
892 solisten(struct socket *so, int backlog, struct thread *td)
896 CURVNET_SET(so->so_vnet);
897 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
903 solisten_proto_check(struct socket *so)
906 SOCK_LOCK_ASSERT(so);
908 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
915 solisten_proto(struct socket *so, int backlog)
917 int sbrcv_lowat, sbsnd_lowat;
918 u_int sbrcv_hiwat, sbsnd_hiwat;
919 short sbrcv_flags, sbsnd_flags;
920 sbintime_t sbrcv_timeo, sbsnd_timeo;
922 SOCK_LOCK_ASSERT(so);
928 * Change this socket to listening state.
930 sbrcv_lowat = so->so_rcv.sb_lowat;
931 sbsnd_lowat = so->so_snd.sb_lowat;
932 sbrcv_hiwat = so->so_rcv.sb_hiwat;
933 sbsnd_hiwat = so->so_snd.sb_hiwat;
934 sbrcv_flags = so->so_rcv.sb_flags;
935 sbsnd_flags = so->so_snd.sb_flags;
936 sbrcv_timeo = so->so_rcv.sb_timeo;
937 sbsnd_timeo = so->so_snd.sb_timeo;
939 sbdestroy(&so->so_snd, so);
940 sbdestroy(&so->so_rcv, so);
941 sx_destroy(&so->so_snd.sb_sx);
942 sx_destroy(&so->so_rcv.sb_sx);
943 SOCKBUF_LOCK_DESTROY(&so->so_snd);
944 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
948 sizeof(struct socket) - offsetof(struct socket, so_rcv));
951 so->sol_sbrcv_lowat = sbrcv_lowat;
952 so->sol_sbsnd_lowat = sbsnd_lowat;
953 so->sol_sbrcv_hiwat = sbrcv_hiwat;
954 so->sol_sbsnd_hiwat = sbsnd_hiwat;
955 so->sol_sbrcv_flags = sbrcv_flags;
956 so->sol_sbsnd_flags = sbsnd_flags;
957 so->sol_sbrcv_timeo = sbrcv_timeo;
958 so->sol_sbsnd_timeo = sbsnd_timeo;
960 so->sol_qlen = so->sol_incqlen = 0;
961 TAILQ_INIT(&so->sol_incomp);
962 TAILQ_INIT(&so->sol_comp);
964 so->sol_accept_filter = NULL;
965 so->sol_accept_filter_arg = NULL;
966 so->sol_accept_filter_str = NULL;
968 so->sol_upcall = NULL;
969 so->sol_upcallarg = NULL;
971 so->so_options |= SO_ACCEPTCONN;
974 if (backlog < 0 || backlog > somaxconn)
976 so->sol_qlimit = backlog;
980 * Wakeup listeners/subsystems once we have a complete connection.
981 * Enters with lock, returns unlocked.
984 solisten_wakeup(struct socket *sol)
987 if (sol->sol_upcall != NULL)
988 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
990 selwakeuppri(&sol->so_rdsel, PSOCK);
991 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
993 SOLISTEN_UNLOCK(sol);
994 wakeup_one(&sol->sol_comp);
995 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
996 pgsigio(&sol->so_sigio, SIGIO, 0);
1000 * Return single connection off a listening socket queue. Main consumer of
1001 * the function is kern_accept4(). Some modules, that do their own accept
1002 * management also use the function.
1004 * Listening socket must be locked on entry and is returned unlocked on
1006 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1009 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1014 SOLISTEN_LOCK_ASSERT(head);
1016 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1017 head->so_error == 0) {
1018 error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH,
1021 SOLISTEN_UNLOCK(head);
1025 if (head->so_error) {
1026 error = head->so_error;
1028 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1029 error = EWOULDBLOCK;
1033 SOLISTEN_UNLOCK(head);
1036 so = TAILQ_FIRST(&head->sol_comp);
1038 KASSERT(so->so_qstate == SQ_COMP,
1039 ("%s: so %p not SQ_COMP", __func__, so));
1042 so->so_qstate = SQ_NONE;
1043 so->so_listen = NULL;
1044 TAILQ_REMOVE(&head->sol_comp, so, so_list);
1045 if (flags & ACCEPT4_INHERIT)
1046 so->so_state |= (head->so_state & SS_NBIO);
1048 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1057 * Evaluate the reference count and named references on a socket; if no
1058 * references remain, free it. This should be called whenever a reference is
1059 * released, such as in sorele(), but also when named reference flags are
1060 * cleared in socket or protocol code.
1062 * sofree() will free the socket if:
1064 * - There are no outstanding file descriptor references or related consumers
1067 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
1069 * - The protocol does not have an outstanding strong reference on the socket
1072 * - The socket is not in a completed connection queue, so a process has been
1073 * notified that it is present. If it is removed, the user process may
1074 * block in accept() despite select() saying the socket was ready.
1077 sofree(struct socket *so)
1079 struct protosw *pr = so->so_proto;
1081 SOCK_LOCK_ASSERT(so);
1083 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
1084 (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) {
1089 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
1092 sol = so->so_listen;
1093 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
1096 * To solve race between close of a listening socket and
1097 * a socket on its incomplete queue, we need to lock both.
1098 * The order is first listening socket, then regular.
1099 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
1100 * function and the listening socket are the only pointers
1101 * to so. To preserve so and sol, we reference both and then
1103 * After relock the socket may not move to so_comp since it
1104 * doesn't have PCB already, but it may be removed from
1105 * so_incomp. If that happens, we share responsiblity on
1106 * freeing the socket, but soclose() has already removed
1114 if (so->so_qstate == SQ_INCOMP) {
1115 KASSERT(so->so_listen == sol,
1116 ("%s: so %p migrated out of sol %p",
1117 __func__, so, sol));
1118 TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1120 /* This is guarenteed not to be the last. */
1121 refcount_release(&sol->so_count);
1122 so->so_qstate = SQ_NONE;
1123 so->so_listen = NULL;
1125 KASSERT(so->so_listen == NULL,
1126 ("%s: so %p not on (in)comp with so_listen",
1129 KASSERT(so->so_count == 1,
1130 ("%s: so %p count %u", __func__, so, so->so_count));
1133 if (SOLISTENING(so))
1134 so->so_error = ECONNABORTED;
1137 if (so->so_dtor != NULL)
1141 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1142 (*pr->pr_domain->dom_dispose)(so);
1143 if (pr->pr_usrreqs->pru_detach != NULL)
1144 (*pr->pr_usrreqs->pru_detach)(so);
1147 * From this point on, we assume that no other references to this
1148 * socket exist anywhere else in the stack. Therefore, no locks need
1149 * to be acquired or held.
1151 * We used to do a lot of socket buffer and socket locking here, as
1152 * well as invoke sorflush() and perform wakeups. The direct call to
1153 * dom_dispose() and sbdestroy() are an inlining of what was
1154 * necessary from sorflush().
1156 * Notice that the socket buffer and kqueue state are torn down
1157 * before calling pru_detach. This means that protocols shold not
1158 * assume they can perform socket wakeups, etc, in their detach code.
1160 if (!SOLISTENING(so)) {
1161 sbdestroy(&so->so_snd, so);
1162 sbdestroy(&so->so_rcv, so);
1164 seldrain(&so->so_rdsel);
1165 seldrain(&so->so_wrsel);
1166 knlist_destroy(&so->so_rdsel.si_note);
1167 knlist_destroy(&so->so_wrsel.si_note);
1172 * Close a socket on last file table reference removal. Initiate disconnect
1173 * if connected. Free socket when disconnect complete.
1175 * This function will sorele() the socket. Note that soclose() may be called
1176 * prior to the ref count reaching zero. The actual socket structure will
1177 * not be freed until the ref count reaches zero.
1180 soclose(struct socket *so)
1182 struct accept_queue lqueue;
1185 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1187 CURVNET_SET(so->so_vnet);
1188 funsetown(&so->so_sigio);
1189 if (so->so_state & SS_ISCONNECTED) {
1190 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1191 error = sodisconnect(so);
1193 if (error == ENOTCONN)
1199 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1200 if ((so->so_state & SS_ISDISCONNECTING) &&
1201 (so->so_state & SS_NBIO))
1203 while (so->so_state & SS_ISCONNECTED) {
1204 error = tsleep(&so->so_timeo,
1205 PSOCK | PCATCH, "soclos",
1206 so->so_linger * hz);
1214 if (so->so_proto->pr_usrreqs->pru_close != NULL)
1215 (*so->so_proto->pr_usrreqs->pru_close)(so);
1218 if (SOLISTENING(so)) {
1221 TAILQ_INIT(&lqueue);
1222 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1223 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1225 so->sol_qlen = so->sol_incqlen = 0;
1227 TAILQ_FOREACH(sp, &lqueue, so_list) {
1229 sp->so_qstate = SQ_NONE;
1230 sp->so_listen = NULL;
1232 /* Guaranteed not to be the last. */
1233 refcount_release(&so->so_count);
1236 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1237 so->so_state |= SS_NOFDREF;
1239 if (SOLISTENING(so)) {
1240 struct socket *sp, *tsp;
1242 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1244 if (sp->so_count == 0) {
1248 /* sp is now in sofree() */
1257 * soabort() is used to abruptly tear down a connection, such as when a
1258 * resource limit is reached (listen queue depth exceeded), or if a listen
1259 * socket is closed while there are sockets waiting to be accepted.
1261 * This interface is tricky, because it is called on an unreferenced socket,
1262 * and must be called only by a thread that has actually removed the socket
1263 * from the listen queue it was on, or races with other threads are risked.
1265 * This interface will call into the protocol code, so must not be called
1266 * with any socket locks held. Protocols do call it while holding their own
1267 * recursible protocol mutexes, but this is something that should be subject
1268 * to review in the future.
1271 soabort(struct socket *so)
1275 * In as much as is possible, assert that no references to this
1276 * socket are held. This is not quite the same as asserting that the
1277 * current thread is responsible for arranging for no references, but
1278 * is as close as we can get for now.
1280 KASSERT(so->so_count == 0, ("soabort: so_count"));
1281 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1282 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1285 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1286 (*so->so_proto->pr_usrreqs->pru_abort)(so);
1292 soaccept(struct socket *so, struct sockaddr **nam)
1297 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1298 so->so_state &= ~SS_NOFDREF;
1301 CURVNET_SET(so->so_vnet);
1302 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1308 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1311 return (soconnectat(AT_FDCWD, so, nam, td));
1315 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1320 if (SOLISTENING(so))
1321 return (EOPNOTSUPP);
1323 CURVNET_SET(so->so_vnet);
1325 * If protocol is connection-based, can only connect once.
1326 * Otherwise, if connected, try to disconnect first. This allows
1327 * user to disconnect by connecting to, e.g., a null address.
1329 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1330 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1331 (error = sodisconnect(so)))) {
1335 * Prevent accumulated error from previous connection from
1339 if (fd == AT_FDCWD) {
1340 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1343 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1353 soconnect2(struct socket *so1, struct socket *so2)
1357 CURVNET_SET(so1->so_vnet);
1358 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1364 sodisconnect(struct socket *so)
1368 if ((so->so_state & SS_ISCONNECTED) == 0)
1370 if (so->so_state & SS_ISDISCONNECTING)
1373 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1377 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1380 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1381 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1385 int clen = 0, error, dontroute;
1387 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1388 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1389 ("sosend_dgram: !PR_ATOMIC"));
1392 resid = uio->uio_resid;
1394 resid = top->m_pkthdr.len;
1396 * In theory resid should be unsigned. However, space must be
1397 * signed, as it might be less than 0 if we over-committed, and we
1398 * must use a signed comparison of space and resid. On the other
1399 * hand, a negative resid causes us to loop sending 0-length
1400 * segments to the protocol.
1408 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1410 td->td_ru.ru_msgsnd++;
1411 if (control != NULL)
1412 clen = control->m_len;
1414 SOCKBUF_LOCK(&so->so_snd);
1415 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1416 SOCKBUF_UNLOCK(&so->so_snd);
1421 error = so->so_error;
1423 SOCKBUF_UNLOCK(&so->so_snd);
1426 if ((so->so_state & SS_ISCONNECTED) == 0) {
1428 * `sendto' and `sendmsg' is allowed on a connection-based
1429 * socket if it supports implied connect. Return ENOTCONN if
1430 * not connected and no address is supplied.
1432 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1433 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1434 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1435 !(resid == 0 && clen != 0)) {
1436 SOCKBUF_UNLOCK(&so->so_snd);
1440 } else if (addr == NULL) {
1441 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1444 error = EDESTADDRREQ;
1445 SOCKBUF_UNLOCK(&so->so_snd);
1451 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1452 * problem and need fixing.
1454 space = sbspace(&so->so_snd);
1455 if (flags & MSG_OOB)
1458 SOCKBUF_UNLOCK(&so->so_snd);
1459 if (resid > space) {
1465 if (flags & MSG_EOR)
1466 top->m_flags |= M_EOR;
1469 * Copy the data from userland into a mbuf chain.
1470 * If no data is to be copied in, a single empty mbuf
1473 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1474 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1476 error = EFAULT; /* only possible error */
1479 space -= resid - uio->uio_resid;
1480 resid = uio->uio_resid;
1482 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1484 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1489 so->so_options |= SO_DONTROUTE;
1493 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1494 * of date. We could have received a reset packet in an interrupt or
1495 * maybe we slept while doing page faults in uiomove() etc. We could
1496 * probably recheck again inside the locking protection here, but
1497 * there are probably other places that this also happens. We must
1501 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1502 (flags & MSG_OOB) ? PRUS_OOB :
1504 * If the user set MSG_EOF, the protocol understands this flag and
1505 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1507 ((flags & MSG_EOF) &&
1508 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1511 /* If there is more to send set PRUS_MORETOCOME */
1512 (flags & MSG_MORETOCOME) ||
1513 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1514 top, addr, control, td);
1517 so->so_options &= ~SO_DONTROUTE;
1526 if (control != NULL)
1532 * Send on a socket. If send must go all at once and message is larger than
1533 * send buffering, then hard error. Lock against other senders. If must go
1534 * all at once and not enough room now, then inform user that this would
1535 * block and do nothing. Otherwise, if nonblocking, send as much as
1536 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1537 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1538 * in mbuf chain must be small enough to send all at once.
1540 * Returns nonzero on error, timeout or signal; callers must check for short
1541 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1545 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1546 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1550 int clen = 0, error, dontroute;
1551 int atomic = sosendallatonce(so) || top;
1554 struct ktls_session *tls;
1555 int tls_enq_cnt, tls_pruflag;
1559 tls_rtype = TLS_RLTYPE_APP;
1562 resid = uio->uio_resid;
1563 else if ((top->m_flags & M_PKTHDR) != 0)
1564 resid = top->m_pkthdr.len;
1566 resid = m_length(top, NULL);
1568 * In theory resid should be unsigned. However, space must be
1569 * signed, as it might be less than 0 if we over-committed, and we
1570 * must use a signed comparison of space and resid. On the other
1571 * hand, a negative resid causes us to loop sending 0-length
1572 * segments to the protocol.
1574 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1575 * type sockets since that's an error.
1577 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1583 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1584 (so->so_proto->pr_flags & PR_ATOMIC);
1586 td->td_ru.ru_msgsnd++;
1587 if (control != NULL)
1588 clen = control->m_len;
1590 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1596 tls = ktls_hold(so->so_snd.sb_tls_info);
1598 if (tls->mode == TCP_TLS_MODE_SW)
1599 tls_pruflag = PRUS_NOTREADY;
1601 if (control != NULL) {
1602 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1604 if (clen >= sizeof(*cm) &&
1605 cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1606 tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1618 SOCKBUF_LOCK(&so->so_snd);
1619 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1620 SOCKBUF_UNLOCK(&so->so_snd);
1625 error = so->so_error;
1627 SOCKBUF_UNLOCK(&so->so_snd);
1630 if ((so->so_state & SS_ISCONNECTED) == 0) {
1632 * `sendto' and `sendmsg' is allowed on a connection-
1633 * based socket if it supports implied connect.
1634 * Return ENOTCONN if not connected and no address is
1637 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1638 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1639 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1640 !(resid == 0 && clen != 0)) {
1641 SOCKBUF_UNLOCK(&so->so_snd);
1645 } else if (addr == NULL) {
1646 SOCKBUF_UNLOCK(&so->so_snd);
1647 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1650 error = EDESTADDRREQ;
1654 space = sbspace(&so->so_snd);
1655 if (flags & MSG_OOB)
1657 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1658 clen > so->so_snd.sb_hiwat) {
1659 SOCKBUF_UNLOCK(&so->so_snd);
1663 if (space < resid + clen &&
1664 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1665 if ((so->so_state & SS_NBIO) ||
1666 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1667 SOCKBUF_UNLOCK(&so->so_snd);
1668 error = EWOULDBLOCK;
1671 error = sbwait(&so->so_snd);
1672 SOCKBUF_UNLOCK(&so->so_snd);
1677 SOCKBUF_UNLOCK(&so->so_snd);
1682 if (flags & MSG_EOR)
1683 top->m_flags |= M_EOR;
1686 ktls_frame(top, tls, &tls_enq_cnt,
1688 tls_rtype = TLS_RLTYPE_APP;
1693 * Copy the data from userland into a mbuf
1694 * chain. If resid is 0, which can happen
1695 * only if we have control to send, then
1696 * a single empty mbuf is returned. This
1697 * is a workaround to prevent protocol send
1702 top = m_uiotombuf(uio, M_WAITOK, space,
1703 tls->params.max_frame_len,
1705 ((flags & MSG_EOR) ? M_EOR : 0));
1707 ktls_frame(top, tls,
1708 &tls_enq_cnt, tls_rtype);
1710 tls_rtype = TLS_RLTYPE_APP;
1713 top = m_uiotombuf(uio, M_WAITOK, space,
1714 (atomic ? max_hdr : 0),
1715 (atomic ? M_PKTHDR : 0) |
1716 ((flags & MSG_EOR) ? M_EOR : 0));
1718 error = EFAULT; /* only possible error */
1721 space -= resid - uio->uio_resid;
1722 resid = uio->uio_resid;
1726 so->so_options |= SO_DONTROUTE;
1730 * XXX all the SBS_CANTSENDMORE checks previously
1731 * done could be out of date. We could have received
1732 * a reset packet in an interrupt or maybe we slept
1733 * while doing page faults in uiomove() etc. We
1734 * could probably recheck again inside the locking
1735 * protection here, but there are probably other
1736 * places that this also happens. We must rethink
1741 pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1743 * If the user set MSG_EOF, the protocol understands
1744 * this flag and nothing left to send then use
1745 * PRU_SEND_EOF instead of PRU_SEND.
1747 ((flags & MSG_EOF) &&
1748 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1751 /* If there is more to send set PRUS_MORETOCOME. */
1752 (flags & MSG_MORETOCOME) ||
1753 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1756 pru_flag |= tls_pruflag;
1759 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1760 pru_flag, top, addr, control, td);
1764 so->so_options &= ~SO_DONTROUTE;
1769 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1775 ktls_enqueue(top, so, tls_enq_cnt);
1784 } while (resid && space > 0);
1788 sbunlock(&so->so_snd);
1796 if (control != NULL)
1802 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1803 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1807 CURVNET_SET(so->so_vnet);
1808 if (!SOLISTENING(so))
1809 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1810 top, control, flags, td);
1821 * The part of soreceive() that implements reading non-inline out-of-band
1822 * data from a socket. For more complete comments, see soreceive(), from
1823 * which this code originated.
1825 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1826 * unable to return an mbuf chain to the caller.
1829 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1831 struct protosw *pr = so->so_proto;
1835 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1838 m = m_get(M_WAITOK, MT_DATA);
1839 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1843 error = uiomove(mtod(m, void *),
1844 (int) min(uio->uio_resid, m->m_len), uio);
1846 } while (uio->uio_resid && error == 0 && m);
1854 * Following replacement or removal of the first mbuf on the first mbuf chain
1855 * of a socket buffer, push necessary state changes back into the socket
1856 * buffer so that other consumers see the values consistently. 'nextrecord'
1857 * is the callers locally stored value of the original value of
1858 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1859 * NOTE: 'nextrecord' may be NULL.
1861 static __inline void
1862 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1865 SOCKBUF_LOCK_ASSERT(sb);
1867 * First, update for the new value of nextrecord. If necessary, make
1868 * it the first record.
1870 if (sb->sb_mb != NULL)
1871 sb->sb_mb->m_nextpkt = nextrecord;
1873 sb->sb_mb = nextrecord;
1876 * Now update any dependent socket buffer fields to reflect the new
1877 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1878 * addition of a second clause that takes care of the case where
1879 * sb_mb has been updated, but remains the last record.
1881 if (sb->sb_mb == NULL) {
1882 sb->sb_mbtail = NULL;
1883 sb->sb_lastrecord = NULL;
1884 } else if (sb->sb_mb->m_nextpkt == NULL)
1885 sb->sb_lastrecord = sb->sb_mb;
1889 * Implement receive operations on a socket. We depend on the way that
1890 * records are added to the sockbuf by sbappend. In particular, each record
1891 * (mbufs linked through m_next) must begin with an address if the protocol
1892 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1893 * data, and then zero or more mbufs of data. In order to allow parallelism
1894 * between network receive and copying to user space, as well as avoid
1895 * sleeping with a mutex held, we release the socket buffer mutex during the
1896 * user space copy. Although the sockbuf is locked, new data may still be
1897 * appended, and thus we must maintain consistency of the sockbuf during that
1900 * The caller may receive the data as a single mbuf chain by supplying an
1901 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1902 * the count in uio_resid.
1905 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1906 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1908 struct mbuf *m, **mp;
1909 int flags, error, offset;
1911 struct protosw *pr = so->so_proto;
1912 struct mbuf *nextrecord;
1914 ssize_t orig_resid = uio->uio_resid;
1919 if (controlp != NULL)
1922 flags = *flagsp &~ MSG_EOR;
1925 if (flags & MSG_OOB)
1926 return (soreceive_rcvoob(so, uio, flags));
1929 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1930 && uio->uio_resid) {
1932 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1935 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1940 SOCKBUF_LOCK(&so->so_rcv);
1941 m = so->so_rcv.sb_mb;
1943 * If we have less data than requested, block awaiting more (subject
1944 * to any timeout) if:
1945 * 1. the current count is less than the low water mark, or
1946 * 2. MSG_DONTWAIT is not set
1948 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1949 sbavail(&so->so_rcv) < uio->uio_resid) &&
1950 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1951 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1952 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1953 ("receive: m == %p sbavail == %u",
1954 m, sbavail(&so->so_rcv)));
1955 if (so->so_error || so->so_rerror) {
1959 error = so->so_error;
1961 error = so->so_rerror;
1962 if ((flags & MSG_PEEK) == 0) {
1968 SOCKBUF_UNLOCK(&so->so_rcv);
1971 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1972 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1976 else if (so->so_rcv.sb_tlsdcc == 0 &&
1977 so->so_rcv.sb_tlscc == 0) {
1981 SOCKBUF_UNLOCK(&so->so_rcv);
1985 for (; m != NULL; m = m->m_next)
1986 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1987 m = so->so_rcv.sb_mb;
1990 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
1991 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
1992 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1993 SOCKBUF_UNLOCK(&so->so_rcv);
1997 if (uio->uio_resid == 0) {
1998 SOCKBUF_UNLOCK(&so->so_rcv);
2001 if ((so->so_state & SS_NBIO) ||
2002 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2003 SOCKBUF_UNLOCK(&so->so_rcv);
2004 error = EWOULDBLOCK;
2007 SBLASTRECORDCHK(&so->so_rcv);
2008 SBLASTMBUFCHK(&so->so_rcv);
2009 error = sbwait(&so->so_rcv);
2010 SOCKBUF_UNLOCK(&so->so_rcv);
2017 * From this point onward, we maintain 'nextrecord' as a cache of the
2018 * pointer to the next record in the socket buffer. We must keep the
2019 * various socket buffer pointers and local stack versions of the
2020 * pointers in sync, pushing out modifications before dropping the
2021 * socket buffer mutex, and re-reading them when picking it up.
2023 * Otherwise, we will race with the network stack appending new data
2024 * or records onto the socket buffer by using inconsistent/stale
2025 * versions of the field, possibly resulting in socket buffer
2028 * By holding the high-level sblock(), we prevent simultaneous
2029 * readers from pulling off the front of the socket buffer.
2031 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2033 uio->uio_td->td_ru.ru_msgrcv++;
2034 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2035 SBLASTRECORDCHK(&so->so_rcv);
2036 SBLASTMBUFCHK(&so->so_rcv);
2037 nextrecord = m->m_nextpkt;
2038 if (pr->pr_flags & PR_ADDR) {
2039 KASSERT(m->m_type == MT_SONAME,
2040 ("m->m_type == %d", m->m_type));
2043 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2045 if (flags & MSG_PEEK) {
2048 sbfree(&so->so_rcv, m);
2049 so->so_rcv.sb_mb = m_free(m);
2050 m = so->so_rcv.sb_mb;
2051 sockbuf_pushsync(&so->so_rcv, nextrecord);
2056 * Process one or more MT_CONTROL mbufs present before any data mbufs
2057 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
2058 * just copy the data; if !MSG_PEEK, we call into the protocol to
2059 * perform externalization (or freeing if controlp == NULL).
2061 if (m != NULL && m->m_type == MT_CONTROL) {
2062 struct mbuf *cm = NULL, *cmn;
2063 struct mbuf **cme = &cm;
2065 struct cmsghdr *cmsg;
2066 struct tls_get_record tgr;
2069 * For MSG_TLSAPPDATA, check for a non-application data
2070 * record. If found, return ENXIO without removing
2071 * it from the receive queue. This allows a subsequent
2072 * call without MSG_TLSAPPDATA to receive it.
2073 * Note that, for TLS, there should only be a single
2074 * control mbuf with the TLS_GET_RECORD message in it.
2076 if (flags & MSG_TLSAPPDATA) {
2077 cmsg = mtod(m, struct cmsghdr *);
2078 if (cmsg->cmsg_type == TLS_GET_RECORD &&
2079 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2080 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2081 /* This will need to change for TLS 1.3. */
2082 if (tgr.tls_type != TLS_RLTYPE_APP) {
2083 SOCKBUF_UNLOCK(&so->so_rcv);
2092 if (flags & MSG_PEEK) {
2093 if (controlp != NULL) {
2094 *controlp = m_copym(m, 0, m->m_len,
2096 controlp = &(*controlp)->m_next;
2100 sbfree(&so->so_rcv, m);
2101 so->so_rcv.sb_mb = m->m_next;
2104 cme = &(*cme)->m_next;
2105 m = so->so_rcv.sb_mb;
2107 } while (m != NULL && m->m_type == MT_CONTROL);
2108 if ((flags & MSG_PEEK) == 0)
2109 sockbuf_pushsync(&so->so_rcv, nextrecord);
2110 while (cm != NULL) {
2113 if (pr->pr_domain->dom_externalize != NULL) {
2114 SOCKBUF_UNLOCK(&so->so_rcv);
2116 error = (*pr->pr_domain->dom_externalize)
2117 (cm, controlp, flags);
2118 SOCKBUF_LOCK(&so->so_rcv);
2119 } else if (controlp != NULL)
2123 if (controlp != NULL) {
2125 while (*controlp != NULL)
2126 controlp = &(*controlp)->m_next;
2131 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2133 nextrecord = so->so_rcv.sb_mb;
2137 if ((flags & MSG_PEEK) == 0) {
2138 KASSERT(m->m_nextpkt == nextrecord,
2139 ("soreceive: post-control, nextrecord !sync"));
2140 if (nextrecord == NULL) {
2141 KASSERT(so->so_rcv.sb_mb == m,
2142 ("soreceive: post-control, sb_mb!=m"));
2143 KASSERT(so->so_rcv.sb_lastrecord == m,
2144 ("soreceive: post-control, lastrecord!=m"));
2148 if (type == MT_OOBDATA)
2151 if ((flags & MSG_PEEK) == 0) {
2152 KASSERT(so->so_rcv.sb_mb == nextrecord,
2153 ("soreceive: sb_mb != nextrecord"));
2154 if (so->so_rcv.sb_mb == NULL) {
2155 KASSERT(so->so_rcv.sb_lastrecord == NULL,
2156 ("soreceive: sb_lastercord != NULL"));
2160 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2161 SBLASTRECORDCHK(&so->so_rcv);
2162 SBLASTMBUFCHK(&so->so_rcv);
2165 * Now continue to read any data mbufs off of the head of the socket
2166 * buffer until the read request is satisfied. Note that 'type' is
2167 * used to store the type of any mbuf reads that have happened so far
2168 * such that soreceive() can stop reading if the type changes, which
2169 * causes soreceive() to return only one of regular data and inline
2170 * out-of-band data in a single socket receive operation.
2174 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2177 * If the type of mbuf has changed since the last mbuf
2178 * examined ('type'), end the receive operation.
2180 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2181 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2182 if (type != m->m_type)
2184 } else if (type == MT_OOBDATA)
2187 KASSERT(m->m_type == MT_DATA,
2188 ("m->m_type == %d", m->m_type));
2189 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2190 len = uio->uio_resid;
2191 if (so->so_oobmark && len > so->so_oobmark - offset)
2192 len = so->so_oobmark - offset;
2193 if (len > m->m_len - moff)
2194 len = m->m_len - moff;
2196 * If mp is set, just pass back the mbufs. Otherwise copy
2197 * them out via the uio, then free. Sockbuf must be
2198 * consistent here (points to current mbuf, it points to next
2199 * record) when we drop priority; we must note any additions
2200 * to the sockbuf when we block interrupts again.
2203 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2204 SBLASTRECORDCHK(&so->so_rcv);
2205 SBLASTMBUFCHK(&so->so_rcv);
2206 SOCKBUF_UNLOCK(&so->so_rcv);
2207 if ((m->m_flags & M_EXTPG) != 0)
2208 error = m_unmapped_uiomove(m, moff, uio,
2211 error = uiomove(mtod(m, char *) + moff,
2213 SOCKBUF_LOCK(&so->so_rcv);
2216 * The MT_SONAME mbuf has already been removed
2217 * from the record, so it is necessary to
2218 * remove the data mbufs, if any, to preserve
2219 * the invariant in the case of PR_ADDR that
2220 * requires MT_SONAME mbufs at the head of
2223 if (pr->pr_flags & PR_ATOMIC &&
2224 ((flags & MSG_PEEK) == 0))
2225 (void)sbdroprecord_locked(&so->so_rcv);
2226 SOCKBUF_UNLOCK(&so->so_rcv);
2230 uio->uio_resid -= len;
2231 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2232 if (len == m->m_len - moff) {
2233 if (m->m_flags & M_EOR)
2235 if (flags & MSG_PEEK) {
2239 nextrecord = m->m_nextpkt;
2240 sbfree(&so->so_rcv, m);
2242 m->m_nextpkt = NULL;
2245 so->so_rcv.sb_mb = m = m->m_next;
2248 so->so_rcv.sb_mb = m_free(m);
2249 m = so->so_rcv.sb_mb;
2251 sockbuf_pushsync(&so->so_rcv, nextrecord);
2252 SBLASTRECORDCHK(&so->so_rcv);
2253 SBLASTMBUFCHK(&so->so_rcv);
2256 if (flags & MSG_PEEK)
2260 if (flags & MSG_DONTWAIT) {
2261 *mp = m_copym(m, 0, len,
2265 * m_copym() couldn't
2267 * Adjust uio_resid back
2269 * down by len bytes,
2270 * which we didn't end
2271 * up "copying" over).
2273 uio->uio_resid += len;
2277 SOCKBUF_UNLOCK(&so->so_rcv);
2278 *mp = m_copym(m, 0, len,
2280 SOCKBUF_LOCK(&so->so_rcv);
2283 sbcut_locked(&so->so_rcv, len);
2286 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2287 if (so->so_oobmark) {
2288 if ((flags & MSG_PEEK) == 0) {
2289 so->so_oobmark -= len;
2290 if (so->so_oobmark == 0) {
2291 so->so_rcv.sb_state |= SBS_RCVATMARK;
2296 if (offset == so->so_oobmark)
2300 if (flags & MSG_EOR)
2303 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2304 * must not quit until "uio->uio_resid == 0" or an error
2305 * termination. If a signal/timeout occurs, return with a
2306 * short count but without error. Keep sockbuf locked
2307 * against other readers.
2309 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2310 !sosendallatonce(so) && nextrecord == NULL) {
2311 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2312 if (so->so_error || so->so_rerror ||
2313 so->so_rcv.sb_state & SBS_CANTRCVMORE)
2316 * Notify the protocol that some data has been
2317 * drained before blocking.
2319 if (pr->pr_flags & PR_WANTRCVD) {
2320 SOCKBUF_UNLOCK(&so->so_rcv);
2322 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2323 SOCKBUF_LOCK(&so->so_rcv);
2325 SBLASTRECORDCHK(&so->so_rcv);
2326 SBLASTMBUFCHK(&so->so_rcv);
2328 * We could receive some data while was notifying
2329 * the protocol. Skip blocking in this case.
2331 if (so->so_rcv.sb_mb == NULL) {
2332 error = sbwait(&so->so_rcv);
2334 SOCKBUF_UNLOCK(&so->so_rcv);
2338 m = so->so_rcv.sb_mb;
2340 nextrecord = m->m_nextpkt;
2344 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2345 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2347 if ((flags & MSG_PEEK) == 0)
2348 (void) sbdroprecord_locked(&so->so_rcv);
2350 if ((flags & MSG_PEEK) == 0) {
2353 * First part is an inline SB_EMPTY_FIXUP(). Second
2354 * part makes sure sb_lastrecord is up-to-date if
2355 * there is still data in the socket buffer.
2357 so->so_rcv.sb_mb = nextrecord;
2358 if (so->so_rcv.sb_mb == NULL) {
2359 so->so_rcv.sb_mbtail = NULL;
2360 so->so_rcv.sb_lastrecord = NULL;
2361 } else if (nextrecord->m_nextpkt == NULL)
2362 so->so_rcv.sb_lastrecord = nextrecord;
2364 SBLASTRECORDCHK(&so->so_rcv);
2365 SBLASTMBUFCHK(&so->so_rcv);
2367 * If soreceive() is being done from the socket callback,
2368 * then don't need to generate ACK to peer to update window,
2369 * since ACK will be generated on return to TCP.
2371 if (!(flags & MSG_SOCALLBCK) &&
2372 (pr->pr_flags & PR_WANTRCVD)) {
2373 SOCKBUF_UNLOCK(&so->so_rcv);
2375 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2376 SOCKBUF_LOCK(&so->so_rcv);
2379 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2380 if (orig_resid == uio->uio_resid && orig_resid &&
2381 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2382 SOCKBUF_UNLOCK(&so->so_rcv);
2385 SOCKBUF_UNLOCK(&so->so_rcv);
2390 sbunlock(&so->so_rcv);
2395 * Optimized version of soreceive() for stream (TCP) sockets.
2398 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2399 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2401 int len = 0, error = 0, flags, oresid;
2403 struct mbuf *m, *n = NULL;
2405 /* We only do stream sockets. */
2406 if (so->so_type != SOCK_STREAM)
2411 flags = *flagsp &~ MSG_EOR;
2414 if (controlp != NULL)
2416 if (flags & MSG_OOB)
2417 return (soreceive_rcvoob(so, uio, flags));
2425 * KTLS store TLS records as records with a control message to
2426 * describe the framing.
2428 * We check once here before acquiring locks to optimize the
2431 if (sb->sb_tls_info != NULL)
2432 return (soreceive_generic(so, psa, uio, mp0, controlp,
2436 /* Prevent other readers from entering the socket. */
2437 error = sblock(sb, SBLOCKWAIT(flags));
2443 if (sb->sb_tls_info != NULL) {
2446 return (soreceive_generic(so, psa, uio, mp0, controlp,
2451 /* Easy one, no space to copyout anything. */
2452 if (uio->uio_resid == 0) {
2456 oresid = uio->uio_resid;
2458 /* We will never ever get anything unless we are or were connected. */
2459 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2465 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2467 /* Abort if socket has reported problems. */
2469 if (sbavail(sb) > 0)
2471 if (oresid > uio->uio_resid)
2473 error = so->so_error;
2474 if (!(flags & MSG_PEEK))
2479 /* Door is closed. Deliver what is left, if any. */
2480 if (sb->sb_state & SBS_CANTRCVMORE) {
2481 if (sbavail(sb) > 0)
2487 /* Socket buffer is empty and we shall not block. */
2488 if (sbavail(sb) == 0 &&
2489 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2494 /* Socket buffer got some data that we shall deliver now. */
2495 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2496 ((so->so_state & SS_NBIO) ||
2497 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2498 sbavail(sb) >= sb->sb_lowat ||
2499 sbavail(sb) >= uio->uio_resid ||
2500 sbavail(sb) >= sb->sb_hiwat) ) {
2504 /* On MSG_WAITALL we must wait until all data or error arrives. */
2505 if ((flags & MSG_WAITALL) &&
2506 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2510 * Wait and block until (more) data comes in.
2511 * NB: Drops the sockbuf lock during wait.
2519 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2520 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2521 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2525 uio->uio_td->td_ru.ru_msgrcv++;
2527 /* Fill uio until full or current end of socket buffer is reached. */
2528 len = min(uio->uio_resid, sbavail(sb));
2530 /* Dequeue as many mbufs as possible. */
2531 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2535 m_cat(*mp0, sb->sb_mb);
2537 m != NULL && m->m_len <= len;
2539 KASSERT(!(m->m_flags & M_NOTAVAIL),
2540 ("%s: m %p not available", __func__, m));
2542 uio->uio_resid -= m->m_len;
2548 sb->sb_lastrecord = sb->sb_mb;
2549 if (sb->sb_mb == NULL)
2552 /* Copy the remainder. */
2554 KASSERT(sb->sb_mb != NULL,
2555 ("%s: len > 0 && sb->sb_mb empty", __func__));
2557 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2559 len = 0; /* Don't flush data from sockbuf. */
2561 uio->uio_resid -= len;
2572 /* NB: Must unlock socket buffer as uiomove may sleep. */
2574 error = m_mbuftouio(uio, sb->sb_mb, len);
2579 SBLASTRECORDCHK(sb);
2583 * Remove the delivered data from the socket buffer unless we
2584 * were only peeking.
2586 if (!(flags & MSG_PEEK)) {
2588 sbdrop_locked(sb, len);
2590 /* Notify protocol that we drained some data. */
2591 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2592 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2593 !(flags & MSG_SOCALLBCK))) {
2596 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2602 * For MSG_WAITALL we may have to loop again and wait for
2603 * more data to come in.
2605 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2608 SOCKBUF_LOCK_ASSERT(sb);
2609 SBLASTRECORDCHK(sb);
2617 * Optimized version of soreceive() for simple datagram cases from userspace.
2618 * Unlike in the stream case, we're able to drop a datagram if copyout()
2619 * fails, and because we handle datagrams atomically, we don't need to use a
2620 * sleep lock to prevent I/O interlacing.
2623 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2624 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2626 struct mbuf *m, *m2;
2629 struct protosw *pr = so->so_proto;
2630 struct mbuf *nextrecord;
2634 if (controlp != NULL)
2637 flags = *flagsp &~ MSG_EOR;
2642 * For any complicated cases, fall back to the full
2643 * soreceive_generic().
2645 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2646 return (soreceive_generic(so, psa, uio, mp0, controlp,
2650 * Enforce restrictions on use.
2652 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2653 ("soreceive_dgram: wantrcvd"));
2654 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2655 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2656 ("soreceive_dgram: SBS_RCVATMARK"));
2657 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2658 ("soreceive_dgram: P_CONNREQUIRED"));
2661 * Loop blocking while waiting for a datagram.
2663 SOCKBUF_LOCK(&so->so_rcv);
2664 while ((m = so->so_rcv.sb_mb) == NULL) {
2665 KASSERT(sbavail(&so->so_rcv) == 0,
2666 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2667 sbavail(&so->so_rcv)));
2669 error = so->so_error;
2671 SOCKBUF_UNLOCK(&so->so_rcv);
2674 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2675 uio->uio_resid == 0) {
2676 SOCKBUF_UNLOCK(&so->so_rcv);
2679 if ((so->so_state & SS_NBIO) ||
2680 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2681 SOCKBUF_UNLOCK(&so->so_rcv);
2682 return (EWOULDBLOCK);
2684 SBLASTRECORDCHK(&so->so_rcv);
2685 SBLASTMBUFCHK(&so->so_rcv);
2686 error = sbwait(&so->so_rcv);
2688 SOCKBUF_UNLOCK(&so->so_rcv);
2692 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2695 uio->uio_td->td_ru.ru_msgrcv++;
2696 SBLASTRECORDCHK(&so->so_rcv);
2697 SBLASTMBUFCHK(&so->so_rcv);
2698 nextrecord = m->m_nextpkt;
2699 if (nextrecord == NULL) {
2700 KASSERT(so->so_rcv.sb_lastrecord == m,
2701 ("soreceive_dgram: lastrecord != m"));
2704 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2705 ("soreceive_dgram: m_nextpkt != nextrecord"));
2708 * Pull 'm' and its chain off the front of the packet queue.
2710 so->so_rcv.sb_mb = NULL;
2711 sockbuf_pushsync(&so->so_rcv, nextrecord);
2714 * Walk 'm's chain and free that many bytes from the socket buffer.
2716 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2717 sbfree(&so->so_rcv, m2);
2720 * Do a few last checks before we let go of the lock.
2722 SBLASTRECORDCHK(&so->so_rcv);
2723 SBLASTMBUFCHK(&so->so_rcv);
2724 SOCKBUF_UNLOCK(&so->so_rcv);
2726 if (pr->pr_flags & PR_ADDR) {
2727 KASSERT(m->m_type == MT_SONAME,
2728 ("m->m_type == %d", m->m_type));
2730 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2735 /* XXXRW: Can this happen? */
2740 * Packet to copyout() is now in 'm' and it is disconnected from the
2743 * Process one or more MT_CONTROL mbufs present before any data mbufs
2744 * in the first mbuf chain on the socket buffer. We call into the
2745 * protocol to perform externalization (or freeing if controlp ==
2746 * NULL). In some cases there can be only MT_CONTROL mbufs without
2749 if (m->m_type == MT_CONTROL) {
2750 struct mbuf *cm = NULL, *cmn;
2751 struct mbuf **cme = &cm;
2757 cme = &(*cme)->m_next;
2759 } while (m != NULL && m->m_type == MT_CONTROL);
2760 while (cm != NULL) {
2763 if (pr->pr_domain->dom_externalize != NULL) {
2764 error = (*pr->pr_domain->dom_externalize)
2765 (cm, controlp, flags);
2766 } else if (controlp != NULL)
2770 if (controlp != NULL) {
2771 while (*controlp != NULL)
2772 controlp = &(*controlp)->m_next;
2777 KASSERT(m == NULL || m->m_type == MT_DATA,
2778 ("soreceive_dgram: !data"));
2779 while (m != NULL && uio->uio_resid > 0) {
2780 len = uio->uio_resid;
2783 error = uiomove(mtod(m, char *), (int)len, uio);
2788 if (len == m->m_len)
2805 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2806 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2810 CURVNET_SET(so->so_vnet);
2811 if (!SOLISTENING(so))
2812 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2813 mp0, controlp, flagsp));
2821 soshutdown(struct socket *so, int how)
2823 struct protosw *pr = so->so_proto;
2824 int error, soerror_enotconn;
2826 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2829 soerror_enotconn = 0;
2831 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2833 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2834 * invoked on a datagram sockets, however historically we would
2835 * actually tear socket down. This is known to be leveraged by
2836 * some applications to unblock process waiting in recvXXX(2)
2837 * by other process that it shares that socket with. Try to meet
2838 * both backward-compatibility and POSIX requirements by forcing
2839 * ENOTCONN but still asking protocol to perform pru_shutdown().
2841 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so))
2843 soerror_enotconn = 1;
2846 if (SOLISTENING(so)) {
2847 if (how != SHUT_WR) {
2849 so->so_error = ECONNABORTED;
2850 solisten_wakeup(so); /* unlocks so */
2855 CURVNET_SET(so->so_vnet);
2856 if (pr->pr_usrreqs->pru_flush != NULL)
2857 (*pr->pr_usrreqs->pru_flush)(so, how);
2860 if (how != SHUT_RD) {
2861 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2862 wakeup(&so->so_timeo);
2864 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2866 wakeup(&so->so_timeo);
2870 return (soerror_enotconn ? ENOTCONN : 0);
2874 sorflush(struct socket *so)
2876 struct sockbuf *sb = &so->so_rcv;
2877 struct protosw *pr = so->so_proto;
2883 * In order to avoid calling dom_dispose with the socket buffer mutex
2884 * held, and in order to generally avoid holding the lock for a long
2885 * time, we make a copy of the socket buffer and clear the original
2886 * (except locks, state). The new socket buffer copy won't have
2887 * initialized locks so we can only call routines that won't use or
2888 * assert those locks.
2890 * Dislodge threads currently blocked in receive and wait to acquire
2891 * a lock against other simultaneous readers before clearing the
2892 * socket buffer. Don't let our acquire be interrupted by a signal
2893 * despite any existing socket disposition on interruptable waiting.
2896 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2899 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2900 * and mutex data unchanged.
2903 bzero(&aso, sizeof(aso));
2904 aso.so_pcb = so->so_pcb;
2905 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2906 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2907 bzero(&sb->sb_startzero,
2908 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2913 * Dispose of special rights and flush the copied socket. Don't call
2914 * any unsafe routines (that rely on locks being initialized) on aso.
2916 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2917 (*pr->pr_domain->dom_dispose)(&aso);
2918 sbrelease_internal(&aso.so_rcv, so);
2922 * Wrapper for Socket established helper hook.
2923 * Parameters: socket, context of the hook point, hook id.
2926 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2928 struct socket_hhook_data hhook_data = {
2935 CURVNET_SET(so->so_vnet);
2936 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2939 /* Ugly but needed, since hhooks return void for now */
2940 return (hhook_data.status);
2944 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2945 * additional variant to handle the case where the option value needs to be
2946 * some kind of integer, but not a specific size. In addition to their use
2947 * here, these functions are also called by the protocol-level pr_ctloutput()
2951 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2956 * If the user gives us more than we wanted, we ignore it, but if we
2957 * don't get the minimum length the caller wants, we return EINVAL.
2958 * On success, sopt->sopt_valsize is set to however much we actually
2961 if ((valsize = sopt->sopt_valsize) < minlen)
2964 sopt->sopt_valsize = valsize = len;
2966 if (sopt->sopt_td != NULL)
2967 return (copyin(sopt->sopt_val, buf, valsize));
2969 bcopy(sopt->sopt_val, buf, valsize);
2974 * Kernel version of setsockopt(2).
2976 * XXX: optlen is size_t, not socklen_t
2979 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2982 struct sockopt sopt;
2984 sopt.sopt_level = level;
2985 sopt.sopt_name = optname;
2986 sopt.sopt_dir = SOPT_SET;
2987 sopt.sopt_val = optval;
2988 sopt.sopt_valsize = optlen;
2989 sopt.sopt_td = NULL;
2990 return (sosetopt(so, &sopt));
2994 sosetopt(struct socket *so, struct sockopt *sopt)
3005 CURVNET_SET(so->so_vnet);
3007 if (sopt->sopt_level != SOL_SOCKET) {
3008 if (so->so_proto->pr_ctloutput != NULL)
3009 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3011 error = ENOPROTOOPT;
3013 switch (sopt->sopt_name) {
3014 case SO_ACCEPTFILTER:
3015 error = accept_filt_setopt(so, sopt);
3021 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3024 if (l.l_linger < 0 ||
3025 l.l_linger > USHRT_MAX ||
3026 l.l_linger > (INT_MAX / hz)) {
3031 so->so_linger = l.l_linger;
3033 so->so_options |= SO_LINGER;
3035 so->so_options &= ~SO_LINGER;
3042 case SO_USELOOPBACK:
3046 case SO_REUSEPORT_LB:
3054 error = sooptcopyin(sopt, &optval, sizeof optval,
3060 so->so_options |= sopt->sopt_name;
3062 so->so_options &= ~sopt->sopt_name;
3067 error = sooptcopyin(sopt, &optval, sizeof optval,
3072 if (optval < 0 || optval >= rt_numfibs) {
3076 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3077 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3078 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3079 so->so_fibnum = optval;
3084 case SO_USER_COOKIE:
3085 error = sooptcopyin(sopt, &val32, sizeof val32,
3089 so->so_user_cookie = val32;
3096 error = sooptcopyin(sopt, &optval, sizeof optval,
3102 * Values < 1 make no sense for any of these options,
3110 error = sbsetopt(so, sopt->sopt_name, optval);
3115 #ifdef COMPAT_FREEBSD32
3116 if (SV_CURPROC_FLAG(SV_ILP32)) {
3117 struct timeval32 tv32;
3119 error = sooptcopyin(sopt, &tv32, sizeof tv32,
3121 CP(tv32, tv, tv_sec);
3122 CP(tv32, tv, tv_usec);
3125 error = sooptcopyin(sopt, &tv, sizeof tv,
3129 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3130 tv.tv_usec >= 1000000) {
3134 if (tv.tv_sec > INT32_MAX)
3138 switch (sopt->sopt_name) {
3140 so->so_snd.sb_timeo = val;
3143 so->so_rcv.sb_timeo = val;
3150 error = sooptcopyin(sopt, &extmac, sizeof extmac,
3154 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3162 error = sooptcopyin(sopt, &optval, sizeof optval,
3166 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3170 so->so_ts_clock = optval;
3173 case SO_MAX_PACING_RATE:
3174 error = sooptcopyin(sopt, &val32, sizeof(val32),
3178 so->so_max_pacing_rate = val32;
3182 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3183 error = hhook_run_socket(so, sopt,
3186 error = ENOPROTOOPT;
3189 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3190 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
3198 * Helper routine for getsockopt.
3201 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3209 * Documented get behavior is that we always return a value, possibly
3210 * truncated to fit in the user's buffer. Traditional behavior is
3211 * that we always tell the user precisely how much we copied, rather
3212 * than something useful like the total amount we had available for
3213 * her. Note that this interface is not idempotent; the entire
3214 * answer must be generated ahead of time.
3216 valsize = min(len, sopt->sopt_valsize);
3217 sopt->sopt_valsize = valsize;
3218 if (sopt->sopt_val != NULL) {
3219 if (sopt->sopt_td != NULL)
3220 error = copyout(buf, sopt->sopt_val, valsize);
3222 bcopy(buf, sopt->sopt_val, valsize);
3228 sogetopt(struct socket *so, struct sockopt *sopt)
3237 CURVNET_SET(so->so_vnet);
3239 if (sopt->sopt_level != SOL_SOCKET) {
3240 if (so->so_proto->pr_ctloutput != NULL)
3241 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3243 error = ENOPROTOOPT;
3247 switch (sopt->sopt_name) {
3248 case SO_ACCEPTFILTER:
3249 error = accept_filt_getopt(so, sopt);
3254 l.l_onoff = so->so_options & SO_LINGER;
3255 l.l_linger = so->so_linger;
3257 error = sooptcopyout(sopt, &l, sizeof l);
3260 case SO_USELOOPBACK:
3266 case SO_REUSEPORT_LB:
3276 optval = so->so_options & sopt->sopt_name;
3278 error = sooptcopyout(sopt, &optval, sizeof optval);
3282 optval = so->so_proto->pr_domain->dom_family;
3286 optval = so->so_type;
3290 optval = so->so_proto->pr_protocol;
3296 optval = so->so_error;
3299 optval = so->so_rerror;
3306 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3307 so->so_snd.sb_hiwat;
3311 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3312 so->so_rcv.sb_hiwat;
3316 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3317 so->so_snd.sb_lowat;
3321 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3322 so->so_rcv.sb_lowat;
3327 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3328 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3329 #ifdef COMPAT_FREEBSD32
3330 if (SV_CURPROC_FLAG(SV_ILP32)) {
3331 struct timeval32 tv32;
3333 CP(tv, tv32, tv_sec);
3334 CP(tv, tv32, tv_usec);
3335 error = sooptcopyout(sopt, &tv32, sizeof tv32);
3338 error = sooptcopyout(sopt, &tv, sizeof tv);
3343 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3347 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3351 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3359 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3363 error = mac_getsockopt_peerlabel(
3364 sopt->sopt_td->td_ucred, so, &extmac);
3367 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3373 case SO_LISTENQLIMIT:
3374 optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3378 optval = SOLISTENING(so) ? so->sol_qlen : 0;
3381 case SO_LISTENINCQLEN:
3382 optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3386 optval = so->so_ts_clock;
3389 case SO_MAX_PACING_RATE:
3390 optval = so->so_max_pacing_rate;
3394 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3395 error = hhook_run_socket(so, sopt,
3398 error = ENOPROTOOPT;
3410 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3412 struct mbuf *m, *m_prev;
3413 int sopt_size = sopt->sopt_valsize;
3415 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3418 if (sopt_size > MLEN) {
3419 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3420 if ((m->m_flags & M_EXT) == 0) {
3424 m->m_len = min(MCLBYTES, sopt_size);
3426 m->m_len = min(MLEN, sopt_size);
3428 sopt_size -= m->m_len;
3433 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3438 if (sopt_size > MLEN) {
3439 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3441 if ((m->m_flags & M_EXT) == 0) {
3446 m->m_len = min(MCLBYTES, sopt_size);
3448 m->m_len = min(MLEN, sopt_size);
3450 sopt_size -= m->m_len;
3458 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3460 struct mbuf *m0 = m;
3462 if (sopt->sopt_val == NULL)
3464 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3465 if (sopt->sopt_td != NULL) {
3468 error = copyin(sopt->sopt_val, mtod(m, char *),
3475 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3476 sopt->sopt_valsize -= m->m_len;
3477 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3480 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3481 panic("ip6_sooptmcopyin");
3486 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3488 struct mbuf *m0 = m;
3491 if (sopt->sopt_val == NULL)
3493 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3494 if (sopt->sopt_td != NULL) {
3497 error = copyout(mtod(m, char *), sopt->sopt_val,
3504 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3505 sopt->sopt_valsize -= m->m_len;
3506 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3507 valsize += m->m_len;
3511 /* enough soopt buffer should be given from user-land */
3515 sopt->sopt_valsize = valsize;
3520 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3521 * out-of-band data, which will then notify socket consumers.
3524 sohasoutofband(struct socket *so)
3527 if (so->so_sigio != NULL)
3528 pgsigio(&so->so_sigio, SIGURG, 0);
3529 selwakeuppri(&so->so_rdsel, PSOCK);
3533 sopoll(struct socket *so, int events, struct ucred *active_cred,
3538 * We do not need to set or assert curvnet as long as everyone uses
3541 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3546 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3552 if (SOLISTENING(so)) {
3553 if (!(events & (POLLIN | POLLRDNORM)))
3555 else if (!TAILQ_EMPTY(&so->sol_comp))
3556 revents = events & (POLLIN | POLLRDNORM);
3557 else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3558 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3560 selrecord(td, &so->so_rdsel);
3565 SOCKBUF_LOCK(&so->so_snd);
3566 SOCKBUF_LOCK(&so->so_rcv);
3567 if (events & (POLLIN | POLLRDNORM))
3568 if (soreadabledata(so))
3569 revents |= events & (POLLIN | POLLRDNORM);
3570 if (events & (POLLOUT | POLLWRNORM))
3571 if (sowriteable(so))
3572 revents |= events & (POLLOUT | POLLWRNORM);
3573 if (events & (POLLPRI | POLLRDBAND))
3574 if (so->so_oobmark ||
3575 (so->so_rcv.sb_state & SBS_RCVATMARK))
3576 revents |= events & (POLLPRI | POLLRDBAND);
3577 if ((events & POLLINIGNEOF) == 0) {
3578 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3579 revents |= events & (POLLIN | POLLRDNORM);
3580 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3584 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3585 revents |= events & POLLRDHUP;
3588 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
3589 selrecord(td, &so->so_rdsel);
3590 so->so_rcv.sb_flags |= SB_SEL;
3592 if (events & (POLLOUT | POLLWRNORM)) {
3593 selrecord(td, &so->so_wrsel);
3594 so->so_snd.sb_flags |= SB_SEL;
3597 SOCKBUF_UNLOCK(&so->so_rcv);
3598 SOCKBUF_UNLOCK(&so->so_snd);
3605 soo_kqfilter(struct file *fp, struct knote *kn)
3607 struct socket *so = kn->kn_fp->f_data;
3611 switch (kn->kn_filter) {
3613 kn->kn_fop = &soread_filtops;
3614 knl = &so->so_rdsel.si_note;
3618 kn->kn_fop = &sowrite_filtops;
3619 knl = &so->so_wrsel.si_note;
3623 kn->kn_fop = &soempty_filtops;
3624 knl = &so->so_wrsel.si_note;
3632 if (SOLISTENING(so)) {
3633 knlist_add(knl, kn, 1);
3636 knlist_add(knl, kn, 1);
3637 sb->sb_flags |= SB_KNOTE;
3645 * Some routines that return EOPNOTSUPP for entry points that are not
3646 * supported by a protocol. Fill in as needed.
3649 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3656 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3663 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3670 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3677 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3685 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3692 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3700 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3707 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3708 struct ifnet *ifp, struct thread *td)
3715 pru_disconnect_notsupp(struct socket *so)
3722 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3729 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3736 pru_rcvd_notsupp(struct socket *so, int flags)
3743 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3750 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3751 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3754 if (control != NULL)
3756 if ((flags & PRUS_NOTREADY) == 0)
3758 return (EOPNOTSUPP);
3762 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3765 return (EOPNOTSUPP);
3769 * This isn't really a ``null'' operation, but it's the default one and
3770 * doesn't do anything destructive.
3773 pru_sense_null(struct socket *so, struct stat *sb)
3776 sb->st_blksize = so->so_snd.sb_hiwat;
3781 pru_shutdown_notsupp(struct socket *so)
3788 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3795 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3796 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3803 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3804 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3811 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3819 filt_sordetach(struct knote *kn)
3821 struct socket *so = kn->kn_fp->f_data;
3824 knlist_remove(&so->so_rdsel.si_note, kn, 1);
3825 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3826 so->so_rcv.sb_flags &= ~SB_KNOTE;
3827 so_rdknl_unlock(so);
3832 filt_soread(struct knote *kn, long hint)
3836 so = kn->kn_fp->f_data;
3838 if (SOLISTENING(so)) {
3839 SOCK_LOCK_ASSERT(so);
3840 kn->kn_data = so->sol_qlen;
3842 kn->kn_flags |= EV_EOF;
3843 kn->kn_fflags = so->so_error;
3846 return (!TAILQ_EMPTY(&so->sol_comp));
3849 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3851 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3852 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3853 kn->kn_flags |= EV_EOF;
3854 kn->kn_fflags = so->so_error;
3856 } else if (so->so_error || so->so_rerror)
3859 if (kn->kn_sfflags & NOTE_LOWAT) {
3860 if (kn->kn_data >= kn->kn_sdata)
3862 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3865 /* This hook returning non-zero indicates an event, not error */
3866 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3870 filt_sowdetach(struct knote *kn)
3872 struct socket *so = kn->kn_fp->f_data;
3875 knlist_remove(&so->so_wrsel.si_note, kn, 1);
3876 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3877 so->so_snd.sb_flags &= ~SB_KNOTE;
3878 so_wrknl_unlock(so);
3883 filt_sowrite(struct knote *kn, long hint)
3887 so = kn->kn_fp->f_data;
3889 if (SOLISTENING(so))
3892 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3893 kn->kn_data = sbspace(&so->so_snd);
3895 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3897 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3898 kn->kn_flags |= EV_EOF;
3899 kn->kn_fflags = so->so_error;
3901 } else if (so->so_error) /* temporary udp error */
3903 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3904 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3906 else if (kn->kn_sfflags & NOTE_LOWAT)
3907 return (kn->kn_data >= kn->kn_sdata);
3909 return (kn->kn_data >= so->so_snd.sb_lowat);
3913 filt_soempty(struct knote *kn, long hint)
3917 so = kn->kn_fp->f_data;
3919 if (SOLISTENING(so))
3922 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3923 kn->kn_data = sbused(&so->so_snd);
3925 if (kn->kn_data == 0)
3932 socheckuid(struct socket *so, uid_t uid)
3937 if (so->so_cred->cr_uid != uid)
3943 * These functions are used by protocols to notify the socket layer (and its
3944 * consumers) of state changes in the sockets driven by protocol-side events.
3948 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3950 * Normal sequence from the active (originating) side is that
3951 * soisconnecting() is called during processing of connect() call, resulting
3952 * in an eventual call to soisconnected() if/when the connection is
3953 * established. When the connection is torn down soisdisconnecting() is
3954 * called during processing of disconnect() call, and soisdisconnected() is
3955 * called when the connection to the peer is totally severed. The semantics
3956 * of these routines are such that connectionless protocols can call
3957 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3958 * calls when setting up a ``connection'' takes no time.
3960 * From the passive side, a socket is created with two queues of sockets:
3961 * so_incomp for connections in progress and so_comp for connections already
3962 * made and awaiting user acceptance. As a protocol is preparing incoming
3963 * connections, it creates a socket structure queued on so_incomp by calling
3964 * sonewconn(). When the connection is established, soisconnected() is
3965 * called, and transfers the socket structure to so_comp, making it available
3968 * If a socket is closed with sockets on either so_incomp or so_comp, these
3969 * sockets are dropped.
3971 * If higher-level protocols are implemented in the kernel, the wakeups done
3972 * here will sometimes cause software-interrupt process scheduling.
3975 soisconnecting(struct socket *so)
3979 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3980 so->so_state |= SS_ISCONNECTING;
3985 soisconnected(struct socket *so)
3989 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3990 so->so_state |= SS_ISCONNECTED;
3992 if (so->so_qstate == SQ_INCOMP) {
3993 struct socket *head = so->so_listen;
3996 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3998 * Promoting a socket from incomplete queue to complete, we
3999 * need to go through reverse order of locking. We first do
4000 * trylock, and if that doesn't succeed, we go the hard way
4001 * leaving a reference and rechecking consistency after proper
4004 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
4007 SOLISTEN_LOCK(head);
4009 if (__predict_false(head != so->so_listen)) {
4011 * The socket went off the listen queue,
4012 * should be lost race to close(2) of sol.
4013 * The socket is about to soabort().
4019 /* Not the last one, as so holds a ref. */
4020 refcount_release(&head->so_count);
4023 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4024 TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4025 head->sol_incqlen--;
4026 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4028 so->so_qstate = SQ_COMP;
4030 solisten_wakeup(head); /* unlocks */
4032 SOCKBUF_LOCK(&so->so_rcv);
4033 soupcall_set(so, SO_RCV,
4034 head->sol_accept_filter->accf_callback,
4035 head->sol_accept_filter_arg);
4036 so->so_options &= ~SO_ACCEPTFILTER;
4037 ret = head->sol_accept_filter->accf_callback(so,
4038 head->sol_accept_filter_arg, M_NOWAIT);
4039 if (ret == SU_ISCONNECTED) {
4040 soupcall_clear(so, SO_RCV);
4041 SOCKBUF_UNLOCK(&so->so_rcv);
4044 SOCKBUF_UNLOCK(&so->so_rcv);
4046 SOLISTEN_UNLOCK(head);
4051 wakeup(&so->so_timeo);
4057 soisdisconnecting(struct socket *so)
4061 so->so_state &= ~SS_ISCONNECTING;
4062 so->so_state |= SS_ISDISCONNECTING;
4064 if (!SOLISTENING(so)) {
4065 SOCKBUF_LOCK(&so->so_rcv);
4066 socantrcvmore_locked(so);
4067 SOCKBUF_LOCK(&so->so_snd);
4068 socantsendmore_locked(so);
4071 wakeup(&so->so_timeo);
4075 soisdisconnected(struct socket *so)
4081 * There is at least one reader of so_state that does not
4082 * acquire socket lock, namely soreceive_generic(). Ensure
4083 * that it never sees all flags that track connection status
4084 * cleared, by ordering the update with a barrier semantic of
4085 * our release thread fence.
4087 so->so_state |= SS_ISDISCONNECTED;
4088 atomic_thread_fence_rel();
4089 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4091 if (!SOLISTENING(so)) {
4093 SOCKBUF_LOCK(&so->so_rcv);
4094 socantrcvmore_locked(so);
4095 SOCKBUF_LOCK(&so->so_snd);
4096 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4097 socantsendmore_locked(so);
4100 wakeup(&so->so_timeo);
4104 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4107 sodupsockaddr(const struct sockaddr *sa, int mflags)
4109 struct sockaddr *sa2;
4111 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4113 bcopy(sa, sa2, sa->sa_len);
4118 * Register per-socket destructor.
4121 sodtor_set(struct socket *so, so_dtor_t *func)
4124 SOCK_LOCK_ASSERT(so);
4129 * Register per-socket buffer upcalls.
4132 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4136 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4146 panic("soupcall_set: bad which");
4148 SOCKBUF_LOCK_ASSERT(sb);
4149 sb->sb_upcall = func;
4150 sb->sb_upcallarg = arg;
4151 sb->sb_flags |= SB_UPCALL;
4155 soupcall_clear(struct socket *so, int which)
4159 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4169 panic("soupcall_clear: bad which");
4171 SOCKBUF_LOCK_ASSERT(sb);
4172 KASSERT(sb->sb_upcall != NULL,
4173 ("%s: so %p no upcall to clear", __func__, so));
4174 sb->sb_upcall = NULL;
4175 sb->sb_upcallarg = NULL;
4176 sb->sb_flags &= ~SB_UPCALL;
4180 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4183 SOLISTEN_LOCK_ASSERT(so);
4184 so->sol_upcall = func;
4185 so->sol_upcallarg = arg;
4189 so_rdknl_lock(void *arg)
4191 struct socket *so = arg;
4193 if (SOLISTENING(so))
4196 SOCKBUF_LOCK(&so->so_rcv);
4200 so_rdknl_unlock(void *arg)
4202 struct socket *so = arg;
4204 if (SOLISTENING(so))
4207 SOCKBUF_UNLOCK(&so->so_rcv);
4211 so_rdknl_assert_lock(void *arg, int what)
4213 struct socket *so = arg;
4215 if (what == LA_LOCKED) {
4216 if (SOLISTENING(so))
4217 SOCK_LOCK_ASSERT(so);
4219 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4221 if (SOLISTENING(so))
4222 SOCK_UNLOCK_ASSERT(so);
4224 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4229 so_wrknl_lock(void *arg)
4231 struct socket *so = arg;
4233 if (SOLISTENING(so))
4236 SOCKBUF_LOCK(&so->so_snd);
4240 so_wrknl_unlock(void *arg)
4242 struct socket *so = arg;
4244 if (SOLISTENING(so))
4247 SOCKBUF_UNLOCK(&so->so_snd);
4251 so_wrknl_assert_lock(void *arg, int what)
4253 struct socket *so = arg;
4255 if (what == LA_LOCKED) {
4256 if (SOLISTENING(so))
4257 SOCK_LOCK_ASSERT(so);
4259 SOCKBUF_LOCK_ASSERT(&so->so_snd);
4261 if (SOLISTENING(so))
4262 SOCK_UNLOCK_ASSERT(so);
4264 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4269 * Create an external-format (``xsocket'') structure using the information in
4270 * the kernel-format socket structure pointed to by so. This is done to
4271 * reduce the spew of irrelevant information over this interface, to isolate
4272 * user code from changes in the kernel structure, and potentially to provide
4273 * information-hiding if we decide that some of this information should be
4274 * hidden from users.
4277 sotoxsocket(struct socket *so, struct xsocket *xso)
4280 bzero(xso, sizeof(*xso));
4281 xso->xso_len = sizeof *xso;
4282 xso->xso_so = (uintptr_t)so;
4283 xso->so_type = so->so_type;
4284 xso->so_options = so->so_options;
4285 xso->so_linger = so->so_linger;
4286 xso->so_state = so->so_state;
4287 xso->so_pcb = (uintptr_t)so->so_pcb;
4288 xso->xso_protocol = so->so_proto->pr_protocol;
4289 xso->xso_family = so->so_proto->pr_domain->dom_family;
4290 xso->so_timeo = so->so_timeo;
4291 xso->so_error = so->so_error;
4292 xso->so_uid = so->so_cred->cr_uid;
4293 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4294 if (SOLISTENING(so)) {
4295 xso->so_qlen = so->sol_qlen;
4296 xso->so_incqlen = so->sol_incqlen;
4297 xso->so_qlimit = so->sol_qlimit;
4298 xso->so_oobmark = 0;
4300 xso->so_state |= so->so_qstate;
4301 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4302 xso->so_oobmark = so->so_oobmark;
4303 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4304 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4309 so_sockbuf_rcv(struct socket *so)
4312 return (&so->so_rcv);
4316 so_sockbuf_snd(struct socket *so)
4319 return (&so->so_snd);
4323 so_state_get(const struct socket *so)
4326 return (so->so_state);
4330 so_state_set(struct socket *so, int val)
4337 so_options_get(const struct socket *so)
4340 return (so->so_options);
4344 so_options_set(struct socket *so, int val)
4347 so->so_options = val;
4351 so_error_get(const struct socket *so)
4354 return (so->so_error);
4358 so_error_set(struct socket *so, int val)
4365 so_linger_get(const struct socket *so)
4368 return (so->so_linger);
4372 so_linger_set(struct socket *so, int val)
4375 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4376 ("%s: val %d out of range", __func__, val));
4378 so->so_linger = val;
4382 so_protosw_get(const struct socket *so)
4385 return (so->so_proto);
4389 so_protosw_set(struct socket *so, struct protosw *val)
4396 so_sorwakeup(struct socket *so)
4403 so_sowwakeup(struct socket *so)
4410 so_sorwakeup_locked(struct socket *so)
4413 sorwakeup_locked(so);
4417 so_sowwakeup_locked(struct socket *so)
4420 sowwakeup_locked(so);
4424 so_lock(struct socket *so)
4431 so_unlock(struct socket *so)