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_MPSAFE, 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_MPSAFE, 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 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
258 static struct mtx so_global_mtx;
259 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
262 * General IPC sysctl name space, used by sockets and a variety of other IPC
265 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
269 * Initialize the socket subsystem and set up the socket
272 static uma_zone_t socket_zone;
276 socket_zone_change(void *tag)
279 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
283 socket_hhook_register(int subtype)
286 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
287 &V_socket_hhh[subtype],
288 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
289 printf("%s: WARNING: unable to register hook\n", __func__);
293 socket_hhook_deregister(int subtype)
296 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
297 printf("%s: WARNING: unable to deregister hook\n", __func__);
301 socket_init(void *tag)
304 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
305 NULL, NULL, UMA_ALIGN_PTR, 0);
306 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
307 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
308 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
309 EVENTHANDLER_PRI_FIRST);
311 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
314 socket_vnet_init(const void *unused __unused)
318 /* We expect a contiguous range */
319 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
320 socket_hhook_register(i);
322 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
323 socket_vnet_init, NULL);
326 socket_vnet_uninit(const void *unused __unused)
330 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
331 socket_hhook_deregister(i);
333 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
334 socket_vnet_uninit, NULL);
337 * Initialise maxsockets. This SYSINIT must be run after
341 init_maxsockets(void *ignored)
344 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
345 maxsockets = imax(maxsockets, maxfiles);
347 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
350 * Sysctl to get and set the maximum global sockets limit. Notify protocols
351 * of the change so that they can update their dependent limits as required.
354 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
356 int error, newmaxsockets;
358 newmaxsockets = maxsockets;
359 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
360 if (error == 0 && req->newptr && newmaxsockets != maxsockets) {
361 if (newmaxsockets > maxsockets &&
362 newmaxsockets <= maxfiles) {
363 maxsockets = newmaxsockets;
364 EVENTHANDLER_INVOKE(maxsockets_change);
370 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
371 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &maxsockets, 0,
372 sysctl_maxsockets, "IU",
373 "Maximum number of sockets available");
376 * Socket operation routines. These routines are called by the routines in
377 * sys_socket.c or from a system process, and implement the semantics of
378 * socket operations by switching out to the protocol specific routines.
382 * Get a socket structure from our zone, and initialize it. Note that it
383 * would probably be better to allocate socket and PCB at the same time, but
384 * I'm not convinced that all the protocols can be easily modified to do
387 * soalloc() returns a socket with a ref count of 0.
389 static struct socket *
390 soalloc(struct vnet *vnet)
394 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
398 if (mac_socket_init(so, M_NOWAIT) != 0) {
399 uma_zfree(socket_zone, so);
403 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
404 uma_zfree(socket_zone, so);
409 * The socket locking protocol allows to lock 2 sockets at a time,
410 * however, the first one must be a listening socket. WITNESS lacks
411 * a feature to change class of an existing lock, so we use DUPOK.
413 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
414 mtx_init(&so->so_snd_mtx, "so_snd", NULL, MTX_DEF);
415 mtx_init(&so->so_rcv_mtx, "so_rcv", NULL, MTX_DEF);
416 so->so_rcv.sb_sel = &so->so_rdsel;
417 so->so_snd.sb_sel = &so->so_wrsel;
418 sx_init(&so->so_snd_sx, "so_snd_sx");
419 sx_init(&so->so_rcv_sx, "so_rcv_sx");
420 TAILQ_INIT(&so->so_snd.sb_aiojobq);
421 TAILQ_INIT(&so->so_rcv.sb_aiojobq);
422 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
423 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
425 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
426 __func__, __LINE__, so));
429 /* We shouldn't need the so_global_mtx */
430 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
431 /* Do we need more comprehensive error returns? */
432 uma_zfree(socket_zone, so);
435 mtx_lock(&so_global_mtx);
436 so->so_gencnt = ++so_gencnt;
439 vnet->vnet_sockcnt++;
441 mtx_unlock(&so_global_mtx);
447 * Free the storage associated with a socket at the socket layer, tear down
448 * locks, labels, etc. All protocol state is assumed already to have been
449 * torn down (and possibly never set up) by the caller.
452 sodealloc(struct socket *so)
455 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
456 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
458 mtx_lock(&so_global_mtx);
459 so->so_gencnt = ++so_gencnt;
460 --numopensockets; /* Could be below, but faster here. */
462 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
463 __func__, __LINE__, so));
464 so->so_vnet->vnet_sockcnt--;
466 mtx_unlock(&so_global_mtx);
468 mac_socket_destroy(so);
470 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
472 khelp_destroy_osd(&so->osd);
473 if (SOLISTENING(so)) {
474 if (so->sol_accept_filter != NULL)
475 accept_filt_setopt(so, NULL);
477 if (so->so_rcv.sb_hiwat)
478 (void)chgsbsize(so->so_cred->cr_uidinfo,
479 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
480 if (so->so_snd.sb_hiwat)
481 (void)chgsbsize(so->so_cred->cr_uidinfo,
482 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
483 sx_destroy(&so->so_snd_sx);
484 sx_destroy(&so->so_rcv_sx);
485 mtx_destroy(&so->so_snd_mtx);
486 mtx_destroy(&so->so_rcv_mtx);
489 mtx_destroy(&so->so_lock);
490 uma_zfree(socket_zone, so);
494 * socreate returns a socket with a ref count of 1 and a file descriptor
495 * reference. The socket should be closed with soclose().
498 socreate(int dom, struct socket **aso, int type, int proto,
499 struct ucred *cred, struct thread *td)
506 * XXX: divert(4) historically abused PF_INET. Keep this compatibility
507 * shim until all applications have been updated.
509 if (__predict_false(dom == PF_INET && type == SOCK_RAW &&
510 proto == IPPROTO_DIVERT)) {
512 printf("%s uses obsolete way to create divert(4) socket\n",
513 td->td_proc->p_comm);
516 prp = pffindproto(dom, type, proto);
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);
527 MPASS(prp->pr_attach);
529 if (IN_CAPABILITY_MODE(td) && (prp->pr_flags & PR_CAPATTACH) == 0)
532 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
533 return (EPROTONOSUPPORT);
535 so = soalloc(CRED_TO_VNET(cred));
540 so->so_cred = crhold(cred);
541 if ((prp->pr_domain->dom_family == PF_INET) ||
542 (prp->pr_domain->dom_family == PF_INET6) ||
543 (prp->pr_domain->dom_family == PF_ROUTE))
544 so->so_fibnum = td->td_proc->p_fibnum;
549 mac_socket_create(cred, so);
551 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
552 so_rdknl_assert_lock);
553 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
554 so_wrknl_assert_lock);
555 if ((prp->pr_flags & PR_SOCKBUF) == 0) {
556 so->so_snd.sb_mtx = &so->so_snd_mtx;
557 so->so_rcv.sb_mtx = &so->so_rcv_mtx;
560 * Auto-sizing of socket buffers is managed by the protocols and
561 * the appropriate flags must be set in the pru_attach function.
563 CURVNET_SET(so->so_vnet);
564 error = prp->pr_attach(so, proto, td);
576 static int regression_sonewconn_earlytest = 1;
577 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
578 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
581 static struct timeval overinterval = { 60, 0 };
582 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
584 "Delay in seconds between warnings for listen socket overflows");
587 * When an attempt at a new connection is noted on a socket which supports
588 * accept(2), the protocol has two options:
589 * 1) Call legacy sonewconn() function, which would call protocol attach
590 * method, same as used for socket(2).
591 * 2) Call solisten_clone(), do attach that is specific to a cloned connection,
592 * and then call solisten_enqueue().
594 * Note: the ref count on the socket is 0 on return.
597 solisten_clone(struct socket *head)
603 const char localprefix[] = "local:";
604 char descrbuf[SUNPATHLEN + sizeof(localprefix)];
606 char addrbuf[INET6_ADDRSTRLEN];
608 char addrbuf[INET_ADDRSTRLEN];
613 over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
615 if (regression_sonewconn_earlytest && over) {
619 head->sol_overcount++;
620 dolog = !!ratecheck(&head->sol_lastover, &overinterval);
623 * If we're going to log, copy the overflow count and queue
624 * length from the listen socket before dropping the lock.
625 * Also, reset the overflow count.
628 overcount = head->sol_overcount;
629 head->sol_overcount = 0;
630 qlen = head->sol_qlen;
632 SOLISTEN_UNLOCK(head);
636 * Try to print something descriptive about the
637 * socket for the error message.
639 sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
641 switch (head->so_proto->pr_domain->dom_family) {
642 #if defined(INET) || defined(INET6)
648 if (head->so_proto->pr_domain->dom_family ==
650 (sotoinpcb(head)->inp_inc.inc_flags &
653 &sotoinpcb(head)->inp_inc.inc6_laddr);
654 sbuf_printf(&descrsb, "[%s]", addrbuf);
660 sotoinpcb(head)->inp_inc.inc_laddr,
662 sbuf_cat(&descrsb, addrbuf);
665 sbuf_printf(&descrsb, ":%hu (proto %u)",
666 ntohs(sotoinpcb(head)->inp_inc.inc_lport),
667 head->so_proto->pr_protocol);
669 #endif /* INET || INET6 */
671 sbuf_cat(&descrsb, localprefix);
672 if (sotounpcb(head)->unp_addr != NULL)
674 sotounpcb(head)->unp_addr->sun_len -
675 offsetof(struct sockaddr_un,
681 sotounpcb(head)->unp_addr->sun_path,
684 sbuf_cat(&descrsb, "(unknown)");
689 * If we can't print something more specific, at least
690 * print the domain name.
692 if (sbuf_finish(&descrsb) != 0 ||
693 sbuf_len(&descrsb) <= 0) {
694 sbuf_clear(&descrsb);
696 head->so_proto->pr_domain->dom_name ?:
698 sbuf_finish(&descrsb);
700 KASSERT(sbuf_len(&descrsb) > 0,
701 ("%s: sbuf creation failed", __func__));
703 * Preserve the historic listen queue overflow log
704 * message, that starts with "sonewconn:". It has
705 * been known to sysadmins for years and also test
706 * sys/kern/sonewconn_overflow checks for it.
708 if (head->so_cred == 0) {
709 log(LOG_DEBUG, "sonewconn: pcb %p (%s): "
710 "Listen queue overflow: %i already in "
711 "queue awaiting acceptance (%d "
712 "occurrences)\n", head->so_pcb,
716 log(LOG_DEBUG, "sonewconn: pcb %p (%s): "
717 "Listen queue overflow: "
718 "%i already in queue awaiting acceptance "
719 "(%d occurrences), euid %d, rgid %d, jail %s\n",
720 head->so_pcb, sbuf_data(&descrsb), qlen,
721 overcount, head->so_cred->cr_uid,
722 head->so_cred->cr_rgid,
723 head->so_cred->cr_prison ?
724 head->so_cred->cr_prison->pr_name :
727 sbuf_delete(&descrsb);
734 SOLISTEN_UNLOCK(head);
735 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
737 so = soalloc(head->so_vnet);
739 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
740 "limit reached or out of memory\n",
741 __func__, head->so_pcb);
744 so->so_listen = head;
745 so->so_type = head->so_type;
746 so->so_options = head->so_options & ~SO_ACCEPTCONN;
747 so->so_linger = head->so_linger;
748 so->so_state = head->so_state;
749 so->so_fibnum = head->so_fibnum;
750 so->so_proto = head->so_proto;
751 so->so_cred = crhold(head->so_cred);
753 mac_socket_newconn(head, so);
755 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
756 so_rdknl_assert_lock);
757 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
758 so_wrknl_assert_lock);
759 VNET_SO_ASSERT(head);
760 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
762 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
763 __func__, head->so_pcb);
766 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
767 so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
768 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
769 so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
770 so->so_rcv.sb_flags = head->sol_sbrcv_flags & SB_AUTOSIZE;
771 so->so_snd.sb_flags = head->sol_sbsnd_flags & SB_AUTOSIZE;
772 if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) {
773 so->so_snd.sb_mtx = &so->so_snd_mtx;
774 so->so_rcv.sb_mtx = &so->so_rcv_mtx;
780 /* Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED. */
782 sonewconn(struct socket *head, int connstatus)
786 if ((so = solisten_clone(head)) == NULL)
789 if (so->so_proto->pr_attach(so, 0, NULL) != 0) {
791 log(LOG_DEBUG, "%s: pcb %p: pr_attach() failed\n",
792 __func__, head->so_pcb);
796 (void)solisten_enqueue(so, connstatus);
802 * Enqueue socket cloned by solisten_clone() to the listen queue of the
803 * listener it has been cloned from.
805 * Return 'true' if socket landed on complete queue, otherwise 'false'.
808 solisten_enqueue(struct socket *so, int connstatus)
810 struct socket *head = so->so_listen;
812 MPASS(refcount_load(&so->so_count) == 0);
813 refcount_init(&so->so_count, 1);
816 if (head->sol_accept_filter != NULL)
818 so->so_state |= connstatus;
819 soref(head); /* A socket on (in)complete queue refs head. */
821 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
822 so->so_qstate = SQ_COMP;
824 solisten_wakeup(head); /* unlocks */
828 * Keep removing sockets from the head until there's room for
829 * us to insert on the tail. In pre-locking revisions, this
830 * was a simple if(), but as we could be racing with other
831 * threads and soabort() requires dropping locks, we must
832 * loop waiting for the condition to be true.
834 while (head->sol_incqlen > head->sol_qlimit) {
837 sp = TAILQ_FIRST(&head->sol_incomp);
838 TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
841 sp->so_qstate = SQ_NONE;
842 sp->so_listen = NULL;
844 sorele_locked(head); /* does SOLISTEN_UNLOCK, head stays */
848 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
849 so->so_qstate = SQ_INCOMP;
851 SOLISTEN_UNLOCK(head);
856 #if defined(SCTP) || defined(SCTP_SUPPORT)
858 * Socket part of sctp_peeloff(). Detach a new socket from an
859 * association. The new socket is returned with a reference.
861 * XXXGL: reduce copy-paste with solisten_clone().
864 sopeeloff(struct socket *head)
868 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
869 __func__, __LINE__, head));
870 so = soalloc(head->so_vnet);
872 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
873 "limit reached or out of memory\n",
874 __func__, head->so_pcb);
877 so->so_type = head->so_type;
878 so->so_options = head->so_options;
879 so->so_linger = head->so_linger;
880 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
881 so->so_fibnum = head->so_fibnum;
882 so->so_proto = head->so_proto;
883 so->so_cred = crhold(head->so_cred);
885 mac_socket_newconn(head, so);
887 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
888 so_rdknl_assert_lock);
889 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
890 so_wrknl_assert_lock);
891 VNET_SO_ASSERT(head);
892 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
894 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
895 __func__, head->so_pcb);
898 if ((*so->so_proto->pr_attach)(so, 0, NULL)) {
900 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
901 __func__, head->so_pcb);
904 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
905 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
906 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
907 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
908 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
909 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
918 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
922 CURVNET_SET(so->so_vnet);
923 error = so->so_proto->pr_bind(so, nam, td);
929 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
933 CURVNET_SET(so->so_vnet);
934 error = so->so_proto->pr_bindat(fd, so, nam, td);
940 * solisten() transitions a socket from a non-listening state to a listening
941 * state, but can also be used to update the listen queue depth on an
942 * existing listen socket. The protocol will call back into the sockets
943 * layer using solisten_proto_check() and solisten_proto() to check and set
944 * socket-layer listen state. Call backs are used so that the protocol can
945 * acquire both protocol and socket layer locks in whatever order is required
948 * Protocol implementors are advised to hold the socket lock across the
949 * socket-layer test and set to avoid races at the socket layer.
952 solisten(struct socket *so, int backlog, struct thread *td)
956 CURVNET_SET(so->so_vnet);
957 error = so->so_proto->pr_listen(so, backlog, td);
963 * Prepare for a call to solisten_proto(). Acquire all socket buffer locks in
964 * order to interlock with socket I/O.
967 solisten_proto_check(struct socket *so)
969 SOCK_LOCK_ASSERT(so);
971 if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
972 SS_ISDISCONNECTING)) != 0)
976 * Sleeping is not permitted here, so simply fail if userspace is
977 * attempting to transmit or receive on the socket. This kind of
978 * transient failure is not ideal, but it should occur only if userspace
979 * is misusing the socket interfaces.
981 if (!sx_try_xlock(&so->so_snd_sx))
983 if (!sx_try_xlock(&so->so_rcv_sx)) {
984 sx_xunlock(&so->so_snd_sx);
987 mtx_lock(&so->so_snd_mtx);
988 mtx_lock(&so->so_rcv_mtx);
990 /* Interlock with soo_aio_queue(). */
991 if (!SOLISTENING(so) &&
992 ((so->so_snd.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0 ||
993 (so->so_rcv.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0)) {
994 solisten_proto_abort(so);
1001 * Undo the setup done by solisten_proto_check().
1004 solisten_proto_abort(struct socket *so)
1006 mtx_unlock(&so->so_snd_mtx);
1007 mtx_unlock(&so->so_rcv_mtx);
1008 sx_xunlock(&so->so_snd_sx);
1009 sx_xunlock(&so->so_rcv_sx);
1013 solisten_proto(struct socket *so, int backlog)
1015 int sbrcv_lowat, sbsnd_lowat;
1016 u_int sbrcv_hiwat, sbsnd_hiwat;
1017 short sbrcv_flags, sbsnd_flags;
1018 sbintime_t sbrcv_timeo, sbsnd_timeo;
1020 SOCK_LOCK_ASSERT(so);
1021 KASSERT((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
1022 SS_ISDISCONNECTING)) == 0,
1023 ("%s: bad socket state %p", __func__, so));
1025 if (SOLISTENING(so))
1029 * Change this socket to listening state.
1031 sbrcv_lowat = so->so_rcv.sb_lowat;
1032 sbsnd_lowat = so->so_snd.sb_lowat;
1033 sbrcv_hiwat = so->so_rcv.sb_hiwat;
1034 sbsnd_hiwat = so->so_snd.sb_hiwat;
1035 sbrcv_flags = so->so_rcv.sb_flags;
1036 sbsnd_flags = so->so_snd.sb_flags;
1037 sbrcv_timeo = so->so_rcv.sb_timeo;
1038 sbsnd_timeo = so->so_snd.sb_timeo;
1040 sbdestroy(so, SO_SND);
1041 sbdestroy(so, SO_RCV);
1045 sizeof(struct socket) - offsetof(struct socket, so_rcv));
1048 so->sol_sbrcv_lowat = sbrcv_lowat;
1049 so->sol_sbsnd_lowat = sbsnd_lowat;
1050 so->sol_sbrcv_hiwat = sbrcv_hiwat;
1051 so->sol_sbsnd_hiwat = sbsnd_hiwat;
1052 so->sol_sbrcv_flags = sbrcv_flags;
1053 so->sol_sbsnd_flags = sbsnd_flags;
1054 so->sol_sbrcv_timeo = sbrcv_timeo;
1055 so->sol_sbsnd_timeo = sbsnd_timeo;
1057 so->sol_qlen = so->sol_incqlen = 0;
1058 TAILQ_INIT(&so->sol_incomp);
1059 TAILQ_INIT(&so->sol_comp);
1061 so->sol_accept_filter = NULL;
1062 so->sol_accept_filter_arg = NULL;
1063 so->sol_accept_filter_str = NULL;
1065 so->sol_upcall = NULL;
1066 so->sol_upcallarg = NULL;
1068 so->so_options |= SO_ACCEPTCONN;
1071 if (backlog < 0 || backlog > somaxconn)
1072 backlog = somaxconn;
1073 so->sol_qlimit = backlog;
1075 mtx_unlock(&so->so_snd_mtx);
1076 mtx_unlock(&so->so_rcv_mtx);
1077 sx_xunlock(&so->so_snd_sx);
1078 sx_xunlock(&so->so_rcv_sx);
1082 * Wakeup listeners/subsystems once we have a complete connection.
1083 * Enters with lock, returns unlocked.
1086 solisten_wakeup(struct socket *sol)
1089 if (sol->sol_upcall != NULL)
1090 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
1092 selwakeuppri(&sol->so_rdsel, PSOCK);
1093 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
1095 SOLISTEN_UNLOCK(sol);
1096 wakeup_one(&sol->sol_comp);
1097 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
1098 pgsigio(&sol->so_sigio, SIGIO, 0);
1102 * Return single connection off a listening socket queue. Main consumer of
1103 * the function is kern_accept4(). Some modules, that do their own accept
1104 * management also use the function. The socket reference held by the
1105 * listen queue is handed to the caller.
1107 * Listening socket must be locked on entry and is returned unlocked on
1109 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1112 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1117 SOLISTEN_LOCK_ASSERT(head);
1119 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1120 head->so_error == 0) {
1121 error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH,
1124 SOLISTEN_UNLOCK(head);
1128 if (head->so_error) {
1129 error = head->so_error;
1131 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1132 error = EWOULDBLOCK;
1136 SOLISTEN_UNLOCK(head);
1139 so = TAILQ_FIRST(&head->sol_comp);
1141 KASSERT(so->so_qstate == SQ_COMP,
1142 ("%s: so %p not SQ_COMP", __func__, so));
1144 so->so_qstate = SQ_NONE;
1145 so->so_listen = NULL;
1146 TAILQ_REMOVE(&head->sol_comp, so, so_list);
1147 if (flags & ACCEPT4_INHERIT)
1148 so->so_state |= (head->so_state & SS_NBIO);
1150 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1152 sorele_locked(head);
1159 * Free socket upon release of the very last reference.
1162 sofree(struct socket *so)
1164 struct protosw *pr = so->so_proto;
1166 SOCK_LOCK_ASSERT(so);
1167 KASSERT(refcount_load(&so->so_count) == 0,
1168 ("%s: so %p has references", __func__, so));
1169 KASSERT(SOLISTENING(so) || so->so_qstate == SQ_NONE,
1170 ("%s: so %p is on listen queue", __func__, so));
1174 if (so->so_dtor != NULL)
1178 if ((pr->pr_flags & PR_RIGHTS) && !SOLISTENING(so)) {
1179 MPASS(pr->pr_domain->dom_dispose != NULL);
1180 (*pr->pr_domain->dom_dispose)(so);
1182 if (pr->pr_detach != NULL)
1186 * From this point on, we assume that no other references to this
1187 * socket exist anywhere else in the stack. Therefore, no locks need
1188 * to be acquired or held.
1190 if (!(pr->pr_flags & PR_SOCKBUF) && !SOLISTENING(so)) {
1191 sbdestroy(so, SO_SND);
1192 sbdestroy(so, SO_RCV);
1194 seldrain(&so->so_rdsel);
1195 seldrain(&so->so_wrsel);
1196 knlist_destroy(&so->so_rdsel.si_note);
1197 knlist_destroy(&so->so_wrsel.si_note);
1202 * Release a reference on a socket while holding the socket lock.
1203 * Unlocks the socket lock before returning.
1206 sorele_locked(struct socket *so)
1208 SOCK_LOCK_ASSERT(so);
1209 if (refcount_release(&so->so_count))
1216 * Close a socket on last file table reference removal. Initiate disconnect
1217 * if connected. Free socket when disconnect complete.
1219 * This function will sorele() the socket. Note that soclose() may be called
1220 * prior to the ref count reaching zero. The actual socket structure will
1221 * not be freed until the ref count reaches zero.
1224 soclose(struct socket *so)
1226 struct accept_queue lqueue;
1228 bool listening, last __diagused;
1230 CURVNET_SET(so->so_vnet);
1231 funsetown(&so->so_sigio);
1232 if (so->so_state & SS_ISCONNECTED) {
1233 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1234 error = sodisconnect(so);
1236 if (error == ENOTCONN)
1242 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1243 if ((so->so_state & SS_ISDISCONNECTING) &&
1244 (so->so_state & SS_NBIO))
1246 while (so->so_state & SS_ISCONNECTED) {
1247 error = tsleep(&so->so_timeo,
1248 PSOCK | PCATCH, "soclos",
1249 so->so_linger * hz);
1257 if (so->so_proto->pr_close != NULL)
1258 so->so_proto->pr_close(so);
1261 if ((listening = SOLISTENING(so))) {
1264 TAILQ_INIT(&lqueue);
1265 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1266 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1268 so->sol_qlen = so->sol_incqlen = 0;
1270 TAILQ_FOREACH(sp, &lqueue, so_list) {
1272 sp->so_qstate = SQ_NONE;
1273 sp->so_listen = NULL;
1275 last = refcount_release(&so->so_count);
1276 KASSERT(!last, ("%s: released last reference for %p",
1282 struct socket *sp, *tsp;
1284 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp)
1292 * soabort() is used to abruptly tear down a connection, such as when a
1293 * resource limit is reached (listen queue depth exceeded), or if a listen
1294 * socket is closed while there are sockets waiting to be accepted.
1296 * This interface is tricky, because it is called on an unreferenced socket,
1297 * and must be called only by a thread that has actually removed the socket
1298 * from the listen queue it was on. Likely this thread holds the last
1299 * reference on the socket and soabort() will proceed with sofree(). But
1300 * it might be not the last, as the sockets on the listen queues are seen
1301 * from the protocol side.
1303 * This interface will call into the protocol code, so must not be called
1304 * with any socket locks held. Protocols do call it while holding their own
1305 * recursible protocol mutexes, but this is something that should be subject
1306 * to review in the future.
1308 * Usually socket should have a single reference left, but this is not a
1309 * requirement. In the past, when we have had named references for file
1310 * descriptor and protocol, we asserted that none of them are being held.
1313 soabort(struct socket *so)
1318 if (so->so_proto->pr_abort != NULL)
1319 so->so_proto->pr_abort(so);
1325 soaccept(struct socket *so, struct sockaddr **nam)
1329 CURVNET_SET(so->so_vnet);
1330 error = so->so_proto->pr_accept(so, nam);
1336 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1339 return (soconnectat(AT_FDCWD, so, nam, td));
1343 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1347 CURVNET_SET(so->so_vnet);
1350 * If protocol is connection-based, can only connect once.
1351 * Otherwise, if connected, try to disconnect first. This allows
1352 * user to disconnect by connecting to, e.g., a null address.
1354 * Note, this check is racy and may need to be re-evaluated at the
1357 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1358 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1359 (error = sodisconnect(so)))) {
1363 * Prevent accumulated error from previous connection from
1367 if (fd == AT_FDCWD) {
1368 error = so->so_proto->pr_connect(so, nam, td);
1370 error = so->so_proto->pr_connectat(fd, so, nam, td);
1379 soconnect2(struct socket *so1, struct socket *so2)
1383 CURVNET_SET(so1->so_vnet);
1384 error = so1->so_proto->pr_connect2(so1, so2);
1390 sodisconnect(struct socket *so)
1394 if ((so->so_state & SS_ISCONNECTED) == 0)
1396 if (so->so_state & SS_ISDISCONNECTING)
1399 error = so->so_proto->pr_disconnect(so);
1404 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1405 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1409 int clen = 0, error, dontroute;
1411 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1412 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1413 ("sosend_dgram: !PR_ATOMIC"));
1416 resid = uio->uio_resid;
1418 resid = top->m_pkthdr.len;
1420 * In theory resid should be unsigned. However, space must be
1421 * signed, as it might be less than 0 if we over-committed, and we
1422 * must use a signed comparison of space and resid. On the other
1423 * hand, a negative resid causes us to loop sending 0-length
1424 * segments to the protocol.
1432 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1434 td->td_ru.ru_msgsnd++;
1435 if (control != NULL)
1436 clen = control->m_len;
1438 SOCKBUF_LOCK(&so->so_snd);
1439 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1440 SOCKBUF_UNLOCK(&so->so_snd);
1445 error = so->so_error;
1447 SOCKBUF_UNLOCK(&so->so_snd);
1450 if ((so->so_state & SS_ISCONNECTED) == 0) {
1452 * `sendto' and `sendmsg' is allowed on a connection-based
1453 * socket if it supports implied connect. Return ENOTCONN if
1454 * not connected and no address is supplied.
1456 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1457 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1458 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1459 !(resid == 0 && clen != 0)) {
1460 SOCKBUF_UNLOCK(&so->so_snd);
1464 } else if (addr == NULL) {
1465 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1468 error = EDESTADDRREQ;
1469 SOCKBUF_UNLOCK(&so->so_snd);
1475 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1476 * problem and need fixing.
1478 space = sbspace(&so->so_snd);
1479 if (flags & MSG_OOB)
1482 SOCKBUF_UNLOCK(&so->so_snd);
1483 if (resid > space) {
1489 if (flags & MSG_EOR)
1490 top->m_flags |= M_EOR;
1493 * Copy the data from userland into a mbuf chain.
1494 * If no data is to be copied in, a single empty mbuf
1497 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1498 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1500 error = EFAULT; /* only possible error */
1503 space -= resid - uio->uio_resid;
1504 resid = uio->uio_resid;
1506 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1508 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1513 so->so_options |= SO_DONTROUTE;
1517 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1518 * of date. We could have received a reset packet in an interrupt or
1519 * maybe we slept while doing page faults in uiomove() etc. We could
1520 * probably recheck again inside the locking protection here, but
1521 * there are probably other places that this also happens. We must
1525 error = so->so_proto->pr_send(so, (flags & MSG_OOB) ? PRUS_OOB :
1527 * If the user set MSG_EOF, the protocol understands this flag and
1528 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1530 ((flags & MSG_EOF) &&
1531 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1534 /* If there is more to send set PRUS_MORETOCOME */
1535 (flags & MSG_MORETOCOME) ||
1536 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1537 top, addr, control, td);
1540 so->so_options &= ~SO_DONTROUTE;
1549 if (control != NULL)
1555 * Send on a socket. If send must go all at once and message is larger than
1556 * send buffering, then hard error. Lock against other senders. If must go
1557 * all at once and not enough room now, then inform user that this would
1558 * block and do nothing. Otherwise, if nonblocking, send as much as
1559 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1560 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1561 * in mbuf chain must be small enough to send all at once.
1563 * Returns nonzero on error, timeout or signal; callers must check for short
1564 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1568 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1569 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1573 int clen = 0, error, dontroute;
1574 int atomic = sosendallatonce(so) || top;
1577 struct ktls_session *tls;
1578 int tls_enq_cnt, tls_send_flag;
1582 tls_rtype = TLS_RLTYPE_APP;
1585 resid = uio->uio_resid;
1586 else if ((top->m_flags & M_PKTHDR) != 0)
1587 resid = top->m_pkthdr.len;
1589 resid = m_length(top, NULL);
1591 * In theory resid should be unsigned. However, space must be
1592 * signed, as it might be less than 0 if we over-committed, and we
1593 * must use a signed comparison of space and resid. On the other
1594 * hand, a negative resid causes us to loop sending 0-length
1595 * segments to the protocol.
1597 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1598 * type sockets since that's an error.
1600 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1606 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1607 (so->so_proto->pr_flags & PR_ATOMIC);
1609 td->td_ru.ru_msgsnd++;
1610 if (control != NULL)
1611 clen = control->m_len;
1613 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1619 tls = ktls_hold(so->so_snd.sb_tls_info);
1621 if (tls->mode == TCP_TLS_MODE_SW)
1622 tls_send_flag = PRUS_NOTREADY;
1624 if (control != NULL) {
1625 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1627 if (clen >= sizeof(*cm) &&
1628 cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1629 tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1637 if (resid == 0 && !ktls_permit_empty_frames(tls)) {
1646 SOCKBUF_LOCK(&so->so_snd);
1647 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1648 SOCKBUF_UNLOCK(&so->so_snd);
1653 error = so->so_error;
1655 SOCKBUF_UNLOCK(&so->so_snd);
1658 if ((so->so_state & SS_ISCONNECTED) == 0) {
1660 * `sendto' and `sendmsg' is allowed on a connection-
1661 * based socket if it supports implied connect.
1662 * Return ENOTCONN if not connected and no address is
1665 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1666 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1667 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1668 !(resid == 0 && clen != 0)) {
1669 SOCKBUF_UNLOCK(&so->so_snd);
1673 } else if (addr == NULL) {
1674 SOCKBUF_UNLOCK(&so->so_snd);
1675 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1678 error = EDESTADDRREQ;
1682 space = sbspace(&so->so_snd);
1683 if (flags & MSG_OOB)
1685 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1686 clen > so->so_snd.sb_hiwat) {
1687 SOCKBUF_UNLOCK(&so->so_snd);
1691 if (space < resid + clen &&
1692 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1693 if ((so->so_state & SS_NBIO) ||
1694 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1695 SOCKBUF_UNLOCK(&so->so_snd);
1696 error = EWOULDBLOCK;
1699 error = sbwait(so, SO_SND);
1700 SOCKBUF_UNLOCK(&so->so_snd);
1705 SOCKBUF_UNLOCK(&so->so_snd);
1710 if (flags & MSG_EOR)
1711 top->m_flags |= M_EOR;
1714 ktls_frame(top, tls, &tls_enq_cnt,
1716 tls_rtype = TLS_RLTYPE_APP;
1721 * Copy the data from userland into a mbuf
1722 * chain. If resid is 0, which can happen
1723 * only if we have control to send, then
1724 * a single empty mbuf is returned. This
1725 * is a workaround to prevent protocol send
1730 top = m_uiotombuf(uio, M_WAITOK, space,
1731 tls->params.max_frame_len,
1733 ((flags & MSG_EOR) ? M_EOR : 0));
1735 ktls_frame(top, tls,
1736 &tls_enq_cnt, tls_rtype);
1738 tls_rtype = TLS_RLTYPE_APP;
1741 top = m_uiotombuf(uio, M_WAITOK, space,
1742 (atomic ? max_hdr : 0),
1743 (atomic ? M_PKTHDR : 0) |
1744 ((flags & MSG_EOR) ? M_EOR : 0));
1746 error = EFAULT; /* only possible error */
1749 space -= resid - uio->uio_resid;
1750 resid = uio->uio_resid;
1754 so->so_options |= SO_DONTROUTE;
1758 * XXX all the SBS_CANTSENDMORE checks previously
1759 * done could be out of date. We could have received
1760 * a reset packet in an interrupt or maybe we slept
1761 * while doing page faults in uiomove() etc. We
1762 * could probably recheck again inside the locking
1763 * protection here, but there are probably other
1764 * places that this also happens. We must rethink
1769 pr_send_flag = (flags & MSG_OOB) ? PRUS_OOB :
1771 * If the user set MSG_EOF, the protocol understands
1772 * this flag and nothing left to send then use
1773 * PRU_SEND_EOF instead of PRU_SEND.
1775 ((flags & MSG_EOF) &&
1776 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1779 /* If there is more to send set PRUS_MORETOCOME. */
1780 (flags & MSG_MORETOCOME) ||
1781 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1784 pr_send_flag |= tls_send_flag;
1787 error = so->so_proto->pr_send(so, pr_send_flag, top,
1792 so->so_options &= ~SO_DONTROUTE;
1797 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1803 ktls_enqueue(top, so, tls_enq_cnt);
1812 } while (resid && space > 0);
1816 SOCK_IO_SEND_UNLOCK(so);
1824 if (control != NULL)
1830 * Send to a socket from a kernel thread.
1832 * XXXGL: in almost all cases uio is NULL and the mbuf is supplied.
1833 * Exception is nfs/bootp_subr.c. It is arguable that the VNET context needs
1834 * to be set at all. This function should just boil down to a static inline
1835 * calling the protocol method.
1838 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1839 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1843 CURVNET_SET(so->so_vnet);
1844 error = so->so_proto->pr_sosend(so, addr, uio,
1845 top, control, flags, td);
1851 * send(2), write(2) or aio_write(2) on a socket.
1854 sousrsend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1855 struct mbuf *control, int flags, struct proc *userproc)
1862 len = uio->uio_resid;
1863 CURVNET_SET(so->so_vnet);
1864 error = so->so_proto->pr_sosend(so, addr, uio, NULL, control, flags,
1869 * Clear transient errors for stream protocols if they made
1870 * some progress. Make exclusion for aio(4) that would
1871 * schedule a new write in case of EWOULDBLOCK and clear
1872 * error itself. See soaio_process_job().
1874 if (uio->uio_resid != len &&
1875 (so->so_proto->pr_flags & PR_ATOMIC) == 0 &&
1877 (error == ERESTART || error == EINTR ||
1878 error == EWOULDBLOCK))
1880 /* Generation of SIGPIPE can be controlled per socket. */
1881 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0 &&
1882 (flags & MSG_NOSIGNAL) == 0) {
1883 if (userproc != NULL) {
1885 PROC_LOCK(userproc);
1886 kern_psignal(userproc, SIGPIPE);
1887 PROC_UNLOCK(userproc);
1889 PROC_LOCK(td->td_proc);
1890 tdsignal(td, SIGPIPE);
1891 PROC_UNLOCK(td->td_proc);
1899 * The part of soreceive() that implements reading non-inline out-of-band
1900 * data from a socket. For more complete comments, see soreceive(), from
1901 * which this code originated.
1903 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1904 * unable to return an mbuf chain to the caller.
1907 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1909 struct protosw *pr = so->so_proto;
1913 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1916 m = m_get(M_WAITOK, MT_DATA);
1917 error = pr->pr_rcvoob(so, m, flags & MSG_PEEK);
1921 error = uiomove(mtod(m, void *),
1922 (int) min(uio->uio_resid, m->m_len), uio);
1924 } while (uio->uio_resid && error == 0 && m);
1932 * Following replacement or removal of the first mbuf on the first mbuf chain
1933 * of a socket buffer, push necessary state changes back into the socket
1934 * buffer so that other consumers see the values consistently. 'nextrecord'
1935 * is the callers locally stored value of the original value of
1936 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1937 * NOTE: 'nextrecord' may be NULL.
1939 static __inline void
1940 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1943 SOCKBUF_LOCK_ASSERT(sb);
1945 * First, update for the new value of nextrecord. If necessary, make
1946 * it the first record.
1948 if (sb->sb_mb != NULL)
1949 sb->sb_mb->m_nextpkt = nextrecord;
1951 sb->sb_mb = nextrecord;
1954 * Now update any dependent socket buffer fields to reflect the new
1955 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1956 * addition of a second clause that takes care of the case where
1957 * sb_mb has been updated, but remains the last record.
1959 if (sb->sb_mb == NULL) {
1960 sb->sb_mbtail = NULL;
1961 sb->sb_lastrecord = NULL;
1962 } else if (sb->sb_mb->m_nextpkt == NULL)
1963 sb->sb_lastrecord = sb->sb_mb;
1967 * Implement receive operations on a socket. We depend on the way that
1968 * records are added to the sockbuf by sbappend. In particular, each record
1969 * (mbufs linked through m_next) must begin with an address if the protocol
1970 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1971 * data, and then zero or more mbufs of data. In order to allow parallelism
1972 * between network receive and copying to user space, as well as avoid
1973 * sleeping with a mutex held, we release the socket buffer mutex during the
1974 * user space copy. Although the sockbuf is locked, new data may still be
1975 * appended, and thus we must maintain consistency of the sockbuf during that
1978 * The caller may receive the data as a single mbuf chain by supplying an
1979 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1980 * the count in uio_resid.
1983 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1984 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1986 struct mbuf *m, **mp;
1987 int flags, error, offset;
1989 struct protosw *pr = so->so_proto;
1990 struct mbuf *nextrecord;
1992 ssize_t orig_resid = uio->uio_resid;
1993 bool report_real_len = false;
1998 if (controlp != NULL)
2000 if (flagsp != NULL) {
2001 report_real_len = *flagsp & MSG_TRUNC;
2002 *flagsp &= ~MSG_TRUNC;
2003 flags = *flagsp &~ MSG_EOR;
2006 if (flags & MSG_OOB)
2007 return (soreceive_rcvoob(so, uio, flags));
2010 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
2011 && uio->uio_resid) {
2016 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2021 SOCKBUF_LOCK(&so->so_rcv);
2022 m = so->so_rcv.sb_mb;
2024 * If we have less data than requested, block awaiting more (subject
2025 * to any timeout) if:
2026 * 1. the current count is less than the low water mark, or
2027 * 2. MSG_DONTWAIT is not set
2029 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
2030 sbavail(&so->so_rcv) < uio->uio_resid) &&
2031 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
2032 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
2033 KASSERT(m != NULL || !sbavail(&so->so_rcv),
2034 ("receive: m == %p sbavail == %u",
2035 m, sbavail(&so->so_rcv)));
2036 if (so->so_error || so->so_rerror) {
2040 error = so->so_error;
2042 error = so->so_rerror;
2043 if ((flags & MSG_PEEK) == 0) {
2049 SOCKBUF_UNLOCK(&so->so_rcv);
2052 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2053 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2057 else if (so->so_rcv.sb_tlsdcc == 0 &&
2058 so->so_rcv.sb_tlscc == 0) {
2062 SOCKBUF_UNLOCK(&so->so_rcv);
2066 for (; m != NULL; m = m->m_next)
2067 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
2068 m = so->so_rcv.sb_mb;
2071 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
2072 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
2073 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
2074 SOCKBUF_UNLOCK(&so->so_rcv);
2078 if (uio->uio_resid == 0 && !report_real_len) {
2079 SOCKBUF_UNLOCK(&so->so_rcv);
2082 if ((so->so_state & SS_NBIO) ||
2083 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2084 SOCKBUF_UNLOCK(&so->so_rcv);
2085 error = EWOULDBLOCK;
2088 SBLASTRECORDCHK(&so->so_rcv);
2089 SBLASTMBUFCHK(&so->so_rcv);
2090 error = sbwait(so, SO_RCV);
2091 SOCKBUF_UNLOCK(&so->so_rcv);
2098 * From this point onward, we maintain 'nextrecord' as a cache of the
2099 * pointer to the next record in the socket buffer. We must keep the
2100 * various socket buffer pointers and local stack versions of the
2101 * pointers in sync, pushing out modifications before dropping the
2102 * socket buffer mutex, and re-reading them when picking it up.
2104 * Otherwise, we will race with the network stack appending new data
2105 * or records onto the socket buffer by using inconsistent/stale
2106 * versions of the field, possibly resulting in socket buffer
2109 * By holding the high-level sblock(), we prevent simultaneous
2110 * readers from pulling off the front of the socket buffer.
2112 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2114 uio->uio_td->td_ru.ru_msgrcv++;
2115 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2116 SBLASTRECORDCHK(&so->so_rcv);
2117 SBLASTMBUFCHK(&so->so_rcv);
2118 nextrecord = m->m_nextpkt;
2119 if (pr->pr_flags & PR_ADDR) {
2120 KASSERT(m->m_type == MT_SONAME,
2121 ("m->m_type == %d", m->m_type));
2124 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2126 if (flags & MSG_PEEK) {
2129 sbfree(&so->so_rcv, m);
2130 so->so_rcv.sb_mb = m_free(m);
2131 m = so->so_rcv.sb_mb;
2132 sockbuf_pushsync(&so->so_rcv, nextrecord);
2137 * Process one or more MT_CONTROL mbufs present before any data mbufs
2138 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
2139 * just copy the data; if !MSG_PEEK, we call into the protocol to
2140 * perform externalization (or freeing if controlp == NULL).
2142 if (m != NULL && m->m_type == MT_CONTROL) {
2143 struct mbuf *cm = NULL, *cmn;
2144 struct mbuf **cme = &cm;
2146 struct cmsghdr *cmsg;
2147 struct tls_get_record tgr;
2150 * For MSG_TLSAPPDATA, check for an alert record.
2151 * If found, return ENXIO without removing
2152 * it from the receive queue. This allows a subsequent
2153 * call without MSG_TLSAPPDATA to receive it.
2154 * Note that, for TLS, there should only be a single
2155 * control mbuf with the TLS_GET_RECORD message in it.
2157 if (flags & MSG_TLSAPPDATA) {
2158 cmsg = mtod(m, struct cmsghdr *);
2159 if (cmsg->cmsg_type == TLS_GET_RECORD &&
2160 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2161 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2162 if (__predict_false(tgr.tls_type ==
2163 TLS_RLTYPE_ALERT)) {
2164 SOCKBUF_UNLOCK(&so->so_rcv);
2173 if (flags & MSG_PEEK) {
2174 if (controlp != NULL) {
2175 *controlp = m_copym(m, 0, m->m_len,
2177 controlp = &(*controlp)->m_next;
2181 sbfree(&so->so_rcv, m);
2182 so->so_rcv.sb_mb = m->m_next;
2185 cme = &(*cme)->m_next;
2186 m = so->so_rcv.sb_mb;
2188 } while (m != NULL && m->m_type == MT_CONTROL);
2189 if ((flags & MSG_PEEK) == 0)
2190 sockbuf_pushsync(&so->so_rcv, nextrecord);
2191 while (cm != NULL) {
2194 if (pr->pr_domain->dom_externalize != NULL) {
2195 SOCKBUF_UNLOCK(&so->so_rcv);
2197 error = (*pr->pr_domain->dom_externalize)
2198 (cm, controlp, flags);
2199 SOCKBUF_LOCK(&so->so_rcv);
2200 } else if (controlp != NULL)
2204 if (controlp != NULL) {
2205 while (*controlp != NULL)
2206 controlp = &(*controlp)->m_next;
2211 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2213 nextrecord = so->so_rcv.sb_mb;
2217 if ((flags & MSG_PEEK) == 0) {
2218 KASSERT(m->m_nextpkt == nextrecord,
2219 ("soreceive: post-control, nextrecord !sync"));
2220 if (nextrecord == NULL) {
2221 KASSERT(so->so_rcv.sb_mb == m,
2222 ("soreceive: post-control, sb_mb!=m"));
2223 KASSERT(so->so_rcv.sb_lastrecord == m,
2224 ("soreceive: post-control, lastrecord!=m"));
2228 if (type == MT_OOBDATA)
2231 if ((flags & MSG_PEEK) == 0) {
2232 KASSERT(so->so_rcv.sb_mb == nextrecord,
2233 ("soreceive: sb_mb != nextrecord"));
2234 if (so->so_rcv.sb_mb == NULL) {
2235 KASSERT(so->so_rcv.sb_lastrecord == NULL,
2236 ("soreceive: sb_lastercord != NULL"));
2240 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2241 SBLASTRECORDCHK(&so->so_rcv);
2242 SBLASTMBUFCHK(&so->so_rcv);
2245 * Now continue to read any data mbufs off of the head of the socket
2246 * buffer until the read request is satisfied. Note that 'type' is
2247 * used to store the type of any mbuf reads that have happened so far
2248 * such that soreceive() can stop reading if the type changes, which
2249 * causes soreceive() to return only one of regular data and inline
2250 * out-of-band data in a single socket receive operation.
2254 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2257 * If the type of mbuf has changed since the last mbuf
2258 * examined ('type'), end the receive operation.
2260 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2261 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2262 if (type != m->m_type)
2264 } else if (type == MT_OOBDATA)
2267 KASSERT(m->m_type == MT_DATA,
2268 ("m->m_type == %d", m->m_type));
2269 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2270 len = uio->uio_resid;
2271 if (so->so_oobmark && len > so->so_oobmark - offset)
2272 len = so->so_oobmark - offset;
2273 if (len > m->m_len - moff)
2274 len = m->m_len - moff;
2276 * If mp is set, just pass back the mbufs. Otherwise copy
2277 * them out via the uio, then free. Sockbuf must be
2278 * consistent here (points to current mbuf, it points to next
2279 * record) when we drop priority; we must note any additions
2280 * to the sockbuf when we block interrupts again.
2283 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2284 SBLASTRECORDCHK(&so->so_rcv);
2285 SBLASTMBUFCHK(&so->so_rcv);
2286 SOCKBUF_UNLOCK(&so->so_rcv);
2287 if ((m->m_flags & M_EXTPG) != 0)
2288 error = m_unmapped_uiomove(m, moff, uio,
2291 error = uiomove(mtod(m, char *) + moff,
2293 SOCKBUF_LOCK(&so->so_rcv);
2296 * The MT_SONAME mbuf has already been removed
2297 * from the record, so it is necessary to
2298 * remove the data mbufs, if any, to preserve
2299 * the invariant in the case of PR_ADDR that
2300 * requires MT_SONAME mbufs at the head of
2303 if (pr->pr_flags & PR_ATOMIC &&
2304 ((flags & MSG_PEEK) == 0))
2305 (void)sbdroprecord_locked(&so->so_rcv);
2306 SOCKBUF_UNLOCK(&so->so_rcv);
2310 uio->uio_resid -= len;
2311 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2312 if (len == m->m_len - moff) {
2313 if (m->m_flags & M_EOR)
2315 if (flags & MSG_PEEK) {
2319 nextrecord = m->m_nextpkt;
2320 sbfree(&so->so_rcv, m);
2322 m->m_nextpkt = NULL;
2325 so->so_rcv.sb_mb = m = m->m_next;
2328 so->so_rcv.sb_mb = m_free(m);
2329 m = so->so_rcv.sb_mb;
2331 sockbuf_pushsync(&so->so_rcv, nextrecord);
2332 SBLASTRECORDCHK(&so->so_rcv);
2333 SBLASTMBUFCHK(&so->so_rcv);
2336 if (flags & MSG_PEEK)
2340 if (flags & MSG_DONTWAIT) {
2341 *mp = m_copym(m, 0, len,
2345 * m_copym() couldn't
2347 * Adjust uio_resid back
2349 * down by len bytes,
2350 * which we didn't end
2351 * up "copying" over).
2353 uio->uio_resid += len;
2357 SOCKBUF_UNLOCK(&so->so_rcv);
2358 *mp = m_copym(m, 0, len,
2360 SOCKBUF_LOCK(&so->so_rcv);
2363 sbcut_locked(&so->so_rcv, len);
2366 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2367 if (so->so_oobmark) {
2368 if ((flags & MSG_PEEK) == 0) {
2369 so->so_oobmark -= len;
2370 if (so->so_oobmark == 0) {
2371 so->so_rcv.sb_state |= SBS_RCVATMARK;
2376 if (offset == so->so_oobmark)
2380 if (flags & MSG_EOR)
2383 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2384 * must not quit until "uio->uio_resid == 0" or an error
2385 * termination. If a signal/timeout occurs, return with a
2386 * short count but without error. Keep sockbuf locked
2387 * against other readers.
2389 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2390 !sosendallatonce(so) && nextrecord == NULL) {
2391 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2392 if (so->so_error || so->so_rerror ||
2393 so->so_rcv.sb_state & SBS_CANTRCVMORE)
2396 * Notify the protocol that some data has been
2397 * drained before blocking.
2399 if (pr->pr_flags & PR_WANTRCVD) {
2400 SOCKBUF_UNLOCK(&so->so_rcv);
2402 pr->pr_rcvd(so, flags);
2403 SOCKBUF_LOCK(&so->so_rcv);
2405 SBLASTRECORDCHK(&so->so_rcv);
2406 SBLASTMBUFCHK(&so->so_rcv);
2408 * We could receive some data while was notifying
2409 * the protocol. Skip blocking in this case.
2411 if (so->so_rcv.sb_mb == NULL) {
2412 error = sbwait(so, SO_RCV);
2414 SOCKBUF_UNLOCK(&so->so_rcv);
2418 m = so->so_rcv.sb_mb;
2420 nextrecord = m->m_nextpkt;
2424 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2425 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2426 if (report_real_len)
2427 uio->uio_resid -= m_length(m, NULL) - moff;
2429 if ((flags & MSG_PEEK) == 0)
2430 (void) sbdroprecord_locked(&so->so_rcv);
2432 if ((flags & MSG_PEEK) == 0) {
2435 * First part is an inline SB_EMPTY_FIXUP(). Second
2436 * part makes sure sb_lastrecord is up-to-date if
2437 * there is still data in the socket buffer.
2439 so->so_rcv.sb_mb = nextrecord;
2440 if (so->so_rcv.sb_mb == NULL) {
2441 so->so_rcv.sb_mbtail = NULL;
2442 so->so_rcv.sb_lastrecord = NULL;
2443 } else if (nextrecord->m_nextpkt == NULL)
2444 so->so_rcv.sb_lastrecord = nextrecord;
2446 SBLASTRECORDCHK(&so->so_rcv);
2447 SBLASTMBUFCHK(&so->so_rcv);
2449 * If soreceive() is being done from the socket callback,
2450 * then don't need to generate ACK to peer to update window,
2451 * since ACK will be generated on return to TCP.
2453 if (!(flags & MSG_SOCALLBCK) &&
2454 (pr->pr_flags & PR_WANTRCVD)) {
2455 SOCKBUF_UNLOCK(&so->so_rcv);
2457 pr->pr_rcvd(so, flags);
2458 SOCKBUF_LOCK(&so->so_rcv);
2461 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2462 if (orig_resid == uio->uio_resid && orig_resid &&
2463 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2464 SOCKBUF_UNLOCK(&so->so_rcv);
2467 SOCKBUF_UNLOCK(&so->so_rcv);
2472 SOCK_IO_RECV_UNLOCK(so);
2477 * Optimized version of soreceive() for stream (TCP) sockets.
2480 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2481 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2483 int len = 0, error = 0, flags, oresid;
2485 struct mbuf *m, *n = NULL;
2487 /* We only do stream sockets. */
2488 if (so->so_type != SOCK_STREAM)
2493 flags = *flagsp &~ MSG_EOR;
2496 if (controlp != NULL)
2498 if (flags & MSG_OOB)
2499 return (soreceive_rcvoob(so, uio, flags));
2507 * KTLS store TLS records as records with a control message to
2508 * describe the framing.
2510 * We check once here before acquiring locks to optimize the
2513 if (sb->sb_tls_info != NULL)
2514 return (soreceive_generic(so, psa, uio, mp0, controlp,
2518 /* Prevent other readers from entering the socket. */
2519 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2525 if (sb->sb_tls_info != NULL) {
2527 SOCK_IO_RECV_UNLOCK(so);
2528 return (soreceive_generic(so, psa, uio, mp0, controlp,
2533 /* Easy one, no space to copyout anything. */
2534 if (uio->uio_resid == 0) {
2538 oresid = uio->uio_resid;
2540 /* We will never ever get anything unless we are or were connected. */
2541 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2547 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2549 /* Abort if socket has reported problems. */
2551 if (sbavail(sb) > 0)
2553 if (oresid > uio->uio_resid)
2555 error = so->so_error;
2556 if (!(flags & MSG_PEEK))
2561 /* Door is closed. Deliver what is left, if any. */
2562 if (sb->sb_state & SBS_CANTRCVMORE) {
2563 if (sbavail(sb) > 0)
2569 /* Socket buffer is empty and we shall not block. */
2570 if (sbavail(sb) == 0 &&
2571 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2576 /* Socket buffer got some data that we shall deliver now. */
2577 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2578 ((so->so_state & SS_NBIO) ||
2579 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2580 sbavail(sb) >= sb->sb_lowat ||
2581 sbavail(sb) >= uio->uio_resid ||
2582 sbavail(sb) >= sb->sb_hiwat) ) {
2586 /* On MSG_WAITALL we must wait until all data or error arrives. */
2587 if ((flags & MSG_WAITALL) &&
2588 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2592 * Wait and block until (more) data comes in.
2593 * NB: Drops the sockbuf lock during wait.
2595 error = sbwait(so, SO_RCV);
2601 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2602 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2603 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2607 uio->uio_td->td_ru.ru_msgrcv++;
2609 /* Fill uio until full or current end of socket buffer is reached. */
2610 len = min(uio->uio_resid, sbavail(sb));
2612 /* Dequeue as many mbufs as possible. */
2613 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2617 m_cat(*mp0, sb->sb_mb);
2619 m != NULL && m->m_len <= len;
2621 KASSERT(!(m->m_flags & M_NOTAVAIL),
2622 ("%s: m %p not available", __func__, m));
2624 uio->uio_resid -= m->m_len;
2630 sb->sb_lastrecord = sb->sb_mb;
2631 if (sb->sb_mb == NULL)
2634 /* Copy the remainder. */
2636 KASSERT(sb->sb_mb != NULL,
2637 ("%s: len > 0 && sb->sb_mb empty", __func__));
2639 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2641 len = 0; /* Don't flush data from sockbuf. */
2643 uio->uio_resid -= len;
2654 /* NB: Must unlock socket buffer as uiomove may sleep. */
2656 error = m_mbuftouio(uio, sb->sb_mb, len);
2661 SBLASTRECORDCHK(sb);
2665 * Remove the delivered data from the socket buffer unless we
2666 * were only peeking.
2668 if (!(flags & MSG_PEEK)) {
2670 sbdrop_locked(sb, len);
2672 /* Notify protocol that we drained some data. */
2673 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2674 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2675 !(flags & MSG_SOCALLBCK))) {
2678 so->so_proto->pr_rcvd(so, flags);
2684 * For MSG_WAITALL we may have to loop again and wait for
2685 * more data to come in.
2687 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2690 SBLASTRECORDCHK(sb);
2693 SOCK_IO_RECV_UNLOCK(so);
2698 * Optimized version of soreceive() for simple datagram cases from userspace.
2699 * Unlike in the stream case, we're able to drop a datagram if copyout()
2700 * fails, and because we handle datagrams atomically, we don't need to use a
2701 * sleep lock to prevent I/O interlacing.
2704 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2705 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2707 struct mbuf *m, *m2;
2710 struct protosw *pr = so->so_proto;
2711 struct mbuf *nextrecord;
2715 if (controlp != NULL)
2718 flags = *flagsp &~ MSG_EOR;
2723 * For any complicated cases, fall back to the full
2724 * soreceive_generic().
2726 if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC)))
2727 return (soreceive_generic(so, psa, uio, mp0, controlp,
2731 * Enforce restrictions on use.
2733 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2734 ("soreceive_dgram: wantrcvd"));
2735 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2736 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2737 ("soreceive_dgram: SBS_RCVATMARK"));
2738 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2739 ("soreceive_dgram: P_CONNREQUIRED"));
2742 * Loop blocking while waiting for a datagram.
2744 SOCKBUF_LOCK(&so->so_rcv);
2745 while ((m = so->so_rcv.sb_mb) == NULL) {
2746 KASSERT(sbavail(&so->so_rcv) == 0,
2747 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2748 sbavail(&so->so_rcv)));
2750 error = so->so_error;
2752 SOCKBUF_UNLOCK(&so->so_rcv);
2755 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2756 uio->uio_resid == 0) {
2757 SOCKBUF_UNLOCK(&so->so_rcv);
2760 if ((so->so_state & SS_NBIO) ||
2761 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2762 SOCKBUF_UNLOCK(&so->so_rcv);
2763 return (EWOULDBLOCK);
2765 SBLASTRECORDCHK(&so->so_rcv);
2766 SBLASTMBUFCHK(&so->so_rcv);
2767 error = sbwait(so, SO_RCV);
2769 SOCKBUF_UNLOCK(&so->so_rcv);
2773 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2776 uio->uio_td->td_ru.ru_msgrcv++;
2777 SBLASTRECORDCHK(&so->so_rcv);
2778 SBLASTMBUFCHK(&so->so_rcv);
2779 nextrecord = m->m_nextpkt;
2780 if (nextrecord == NULL) {
2781 KASSERT(so->so_rcv.sb_lastrecord == m,
2782 ("soreceive_dgram: lastrecord != m"));
2785 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2786 ("soreceive_dgram: m_nextpkt != nextrecord"));
2789 * Pull 'm' and its chain off the front of the packet queue.
2791 so->so_rcv.sb_mb = NULL;
2792 sockbuf_pushsync(&so->so_rcv, nextrecord);
2795 * Walk 'm's chain and free that many bytes from the socket buffer.
2797 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2798 sbfree(&so->so_rcv, m2);
2801 * Do a few last checks before we let go of the lock.
2803 SBLASTRECORDCHK(&so->so_rcv);
2804 SBLASTMBUFCHK(&so->so_rcv);
2805 SOCKBUF_UNLOCK(&so->so_rcv);
2807 if (pr->pr_flags & PR_ADDR) {
2808 KASSERT(m->m_type == MT_SONAME,
2809 ("m->m_type == %d", m->m_type));
2811 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2816 /* XXXRW: Can this happen? */
2821 * Packet to copyout() is now in 'm' and it is disconnected from the
2824 * Process one or more MT_CONTROL mbufs present before any data mbufs
2825 * in the first mbuf chain on the socket buffer. We call into the
2826 * protocol to perform externalization (or freeing if controlp ==
2827 * NULL). In some cases there can be only MT_CONTROL mbufs without
2830 if (m->m_type == MT_CONTROL) {
2831 struct mbuf *cm = NULL, *cmn;
2832 struct mbuf **cme = &cm;
2838 cme = &(*cme)->m_next;
2840 } while (m != NULL && m->m_type == MT_CONTROL);
2841 while (cm != NULL) {
2844 if (pr->pr_domain->dom_externalize != NULL) {
2845 error = (*pr->pr_domain->dom_externalize)
2846 (cm, controlp, flags);
2847 } else if (controlp != NULL)
2851 if (controlp != NULL) {
2852 while (*controlp != NULL)
2853 controlp = &(*controlp)->m_next;
2858 KASSERT(m == NULL || m->m_type == MT_DATA,
2859 ("soreceive_dgram: !data"));
2860 while (m != NULL && uio->uio_resid > 0) {
2861 len = uio->uio_resid;
2864 error = uiomove(mtod(m, char *), (int)len, uio);
2869 if (len == m->m_len)
2886 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2887 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2891 CURVNET_SET(so->so_vnet);
2892 error = so->so_proto->pr_soreceive(so, psa, uio, mp0, controlp, flagsp);
2898 soshutdown(struct socket *so, int how)
2901 int error, soerror_enotconn;
2903 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2906 soerror_enotconn = 0;
2909 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2911 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2912 * invoked on a datagram sockets, however historically we would
2913 * actually tear socket down. This is known to be leveraged by
2914 * some applications to unblock process waiting in recvXXX(2)
2915 * by other process that it shares that socket with. Try to meet
2916 * both backward-compatibility and POSIX requirements by forcing
2917 * ENOTCONN but still asking protocol to perform pru_shutdown().
2919 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) {
2923 soerror_enotconn = 1;
2926 if (SOLISTENING(so)) {
2927 if (how != SHUT_WR) {
2928 so->so_error = ECONNABORTED;
2929 solisten_wakeup(so); /* unlocks so */
2937 CURVNET_SET(so->so_vnet);
2939 if (pr->pr_flush != NULL)
2940 pr->pr_flush(so, how);
2943 if (how != SHUT_RD) {
2944 error = pr->pr_shutdown(so);
2945 wakeup(&so->so_timeo);
2947 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2949 wakeup(&so->so_timeo);
2953 return (soerror_enotconn ? ENOTCONN : 0);
2957 sorflush(struct socket *so)
2965 * Dislodge threads currently blocked in receive and wait to acquire
2966 * a lock against other simultaneous readers before clearing the
2967 * socket buffer. Don't let our acquire be interrupted by a signal
2968 * despite any existing socket disposition on interruptable waiting.
2972 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
2974 KASSERT(SOLISTENING(so),
2975 ("%s: soiolock(%p) failed", __func__, so));
2980 if (pr->pr_flags & PR_RIGHTS) {
2981 MPASS(pr->pr_domain->dom_dispose != NULL);
2982 (*pr->pr_domain->dom_dispose)(so);
2984 sbrelease(so, SO_RCV);
2985 SOCK_IO_RECV_UNLOCK(so);
2991 * Wrapper for Socket established helper hook.
2992 * Parameters: socket, context of the hook point, hook id.
2995 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2997 struct socket_hhook_data hhook_data = {
3004 CURVNET_SET(so->so_vnet);
3005 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
3008 /* Ugly but needed, since hhooks return void for now */
3009 return (hhook_data.status);
3013 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
3014 * additional variant to handle the case where the option value needs to be
3015 * some kind of integer, but not a specific size. In addition to their use
3016 * here, these functions are also called by the protocol-level pr_ctloutput()
3020 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
3025 * If the user gives us more than we wanted, we ignore it, but if we
3026 * don't get the minimum length the caller wants, we return EINVAL.
3027 * On success, sopt->sopt_valsize is set to however much we actually
3030 if ((valsize = sopt->sopt_valsize) < minlen)
3033 sopt->sopt_valsize = valsize = len;
3035 if (sopt->sopt_td != NULL)
3036 return (copyin(sopt->sopt_val, buf, valsize));
3038 bcopy(sopt->sopt_val, buf, valsize);
3043 * Kernel version of setsockopt(2).
3045 * XXX: optlen is size_t, not socklen_t
3048 so_setsockopt(struct socket *so, int level, int optname, void *optval,
3051 struct sockopt sopt;
3053 sopt.sopt_level = level;
3054 sopt.sopt_name = optname;
3055 sopt.sopt_dir = SOPT_SET;
3056 sopt.sopt_val = optval;
3057 sopt.sopt_valsize = optlen;
3058 sopt.sopt_td = NULL;
3059 return (sosetopt(so, &sopt));
3063 sosetopt(struct socket *so, struct sockopt *sopt)
3068 sbintime_t val, *valp;
3074 CURVNET_SET(so->so_vnet);
3076 if (sopt->sopt_level != SOL_SOCKET) {
3077 if (so->so_proto->pr_ctloutput != NULL)
3078 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3080 error = ENOPROTOOPT;
3082 switch (sopt->sopt_name) {
3083 case SO_ACCEPTFILTER:
3084 error = accept_filt_setopt(so, sopt);
3090 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3093 if (l.l_linger < 0 ||
3094 l.l_linger > USHRT_MAX ||
3095 l.l_linger > (INT_MAX / hz)) {
3100 so->so_linger = l.l_linger;
3102 so->so_options |= SO_LINGER;
3104 so->so_options &= ~SO_LINGER;
3111 case SO_USELOOPBACK:
3115 case SO_REUSEPORT_LB:
3123 error = sooptcopyin(sopt, &optval, sizeof optval,
3129 so->so_options |= sopt->sopt_name;
3131 so->so_options &= ~sopt->sopt_name;
3136 error = sooptcopyin(sopt, &optval, sizeof optval,
3141 if (optval < 0 || optval >= rt_numfibs) {
3145 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3146 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3147 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3148 so->so_fibnum = optval;
3153 case SO_USER_COOKIE:
3154 error = sooptcopyin(sopt, &val32, sizeof val32,
3158 so->so_user_cookie = val32;
3165 error = so->so_proto->pr_setsbopt(so, sopt);
3172 #ifdef COMPAT_FREEBSD32
3173 if (SV_CURPROC_FLAG(SV_ILP32)) {
3174 struct timeval32 tv32;
3176 error = sooptcopyin(sopt, &tv32, sizeof tv32,
3178 CP(tv32, tv, tv_sec);
3179 CP(tv32, tv, tv_usec);
3182 error = sooptcopyin(sopt, &tv, sizeof tv,
3186 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3187 tv.tv_usec >= 1000000) {
3191 if (tv.tv_sec > INT32_MAX)
3196 valp = sopt->sopt_name == SO_SNDTIMEO ?
3197 (SOLISTENING(so) ? &so->sol_sbsnd_timeo :
3198 &so->so_snd.sb_timeo) :
3199 (SOLISTENING(so) ? &so->sol_sbrcv_timeo :
3200 &so->so_rcv.sb_timeo);
3207 error = sooptcopyin(sopt, &extmac, sizeof extmac,
3211 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3219 error = sooptcopyin(sopt, &optval, sizeof optval,
3223 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3227 so->so_ts_clock = optval;
3230 case SO_MAX_PACING_RATE:
3231 error = sooptcopyin(sopt, &val32, sizeof(val32),
3235 so->so_max_pacing_rate = val32;
3239 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3240 error = hhook_run_socket(so, sopt,
3243 error = ENOPROTOOPT;
3246 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3247 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
3255 * Helper routine for getsockopt.
3258 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3266 * Documented get behavior is that we always return a value, possibly
3267 * truncated to fit in the user's buffer. Traditional behavior is
3268 * that we always tell the user precisely how much we copied, rather
3269 * than something useful like the total amount we had available for
3270 * her. Note that this interface is not idempotent; the entire
3271 * answer must be generated ahead of time.
3273 valsize = min(len, sopt->sopt_valsize);
3274 sopt->sopt_valsize = valsize;
3275 if (sopt->sopt_val != NULL) {
3276 if (sopt->sopt_td != NULL)
3277 error = copyout(buf, sopt->sopt_val, valsize);
3279 bcopy(buf, sopt->sopt_val, valsize);
3285 sogetopt(struct socket *so, struct sockopt *sopt)
3294 CURVNET_SET(so->so_vnet);
3296 if (sopt->sopt_level != SOL_SOCKET) {
3297 if (so->so_proto->pr_ctloutput != NULL)
3298 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3300 error = ENOPROTOOPT;
3304 switch (sopt->sopt_name) {
3305 case SO_ACCEPTFILTER:
3306 error = accept_filt_getopt(so, sopt);
3311 l.l_onoff = so->so_options & SO_LINGER;
3312 l.l_linger = so->so_linger;
3314 error = sooptcopyout(sopt, &l, sizeof l);
3317 case SO_USELOOPBACK:
3323 case SO_REUSEPORT_LB:
3333 optval = so->so_options & sopt->sopt_name;
3335 error = sooptcopyout(sopt, &optval, sizeof optval);
3339 optval = so->so_proto->pr_domain->dom_family;
3343 optval = so->so_type;
3347 optval = so->so_proto->pr_protocol;
3353 optval = so->so_error;
3356 optval = so->so_rerror;
3363 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3364 so->so_snd.sb_hiwat;
3368 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3369 so->so_rcv.sb_hiwat;
3373 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3374 so->so_snd.sb_lowat;
3378 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3379 so->so_rcv.sb_lowat;
3385 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3386 (SOLISTENING(so) ? so->sol_sbsnd_timeo :
3387 so->so_snd.sb_timeo) :
3388 (SOLISTENING(so) ? so->sol_sbrcv_timeo :
3389 so->so_rcv.sb_timeo));
3391 #ifdef COMPAT_FREEBSD32
3392 if (SV_CURPROC_FLAG(SV_ILP32)) {
3393 struct timeval32 tv32;
3395 CP(tv, tv32, tv_sec);
3396 CP(tv, tv32, tv_usec);
3397 error = sooptcopyout(sopt, &tv32, sizeof tv32);
3400 error = sooptcopyout(sopt, &tv, sizeof tv);
3405 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3409 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3413 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3421 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3425 error = mac_getsockopt_peerlabel(
3426 sopt->sopt_td->td_ucred, so, &extmac);
3429 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3435 case SO_LISTENQLIMIT:
3436 optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3440 optval = SOLISTENING(so) ? so->sol_qlen : 0;
3443 case SO_LISTENINCQLEN:
3444 optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3448 optval = so->so_ts_clock;
3451 case SO_MAX_PACING_RATE:
3452 optval = so->so_max_pacing_rate;
3456 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3457 error = hhook_run_socket(so, sopt,
3460 error = ENOPROTOOPT;
3472 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3474 struct mbuf *m, *m_prev;
3475 int sopt_size = sopt->sopt_valsize;
3477 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3480 if (sopt_size > MLEN) {
3481 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3482 if ((m->m_flags & M_EXT) == 0) {
3486 m->m_len = min(MCLBYTES, sopt_size);
3488 m->m_len = min(MLEN, sopt_size);
3490 sopt_size -= m->m_len;
3495 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3500 if (sopt_size > MLEN) {
3501 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3503 if ((m->m_flags & M_EXT) == 0) {
3508 m->m_len = min(MCLBYTES, sopt_size);
3510 m->m_len = min(MLEN, sopt_size);
3512 sopt_size -= m->m_len;
3520 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3522 struct mbuf *m0 = m;
3524 if (sopt->sopt_val == NULL)
3526 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3527 if (sopt->sopt_td != NULL) {
3530 error = copyin(sopt->sopt_val, mtod(m, char *),
3537 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3538 sopt->sopt_valsize -= m->m_len;
3539 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3542 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3543 panic("ip6_sooptmcopyin");
3548 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3550 struct mbuf *m0 = m;
3553 if (sopt->sopt_val == NULL)
3555 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3556 if (sopt->sopt_td != NULL) {
3559 error = copyout(mtod(m, char *), sopt->sopt_val,
3566 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3567 sopt->sopt_valsize -= m->m_len;
3568 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3569 valsize += m->m_len;
3573 /* enough soopt buffer should be given from user-land */
3577 sopt->sopt_valsize = valsize;
3582 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3583 * out-of-band data, which will then notify socket consumers.
3586 sohasoutofband(struct socket *so)
3589 if (so->so_sigio != NULL)
3590 pgsigio(&so->so_sigio, SIGURG, 0);
3591 selwakeuppri(&so->so_rdsel, PSOCK);
3595 sopoll(struct socket *so, int events, struct ucred *active_cred,
3600 * We do not need to set or assert curvnet as long as everyone uses
3603 return (so->so_proto->pr_sopoll(so, events, active_cred, td));
3607 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3613 if (SOLISTENING(so)) {
3614 if (!(events & (POLLIN | POLLRDNORM)))
3616 else if (!TAILQ_EMPTY(&so->sol_comp))
3617 revents = events & (POLLIN | POLLRDNORM);
3618 else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3619 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3621 selrecord(td, &so->so_rdsel);
3626 SOCK_SENDBUF_LOCK(so);
3627 SOCK_RECVBUF_LOCK(so);
3628 if (events & (POLLIN | POLLRDNORM))
3629 if (soreadabledata(so))
3630 revents |= events & (POLLIN | POLLRDNORM);
3631 if (events & (POLLOUT | POLLWRNORM))
3632 if (sowriteable(so))
3633 revents |= events & (POLLOUT | POLLWRNORM);
3634 if (events & (POLLPRI | POLLRDBAND))
3635 if (so->so_oobmark ||
3636 (so->so_rcv.sb_state & SBS_RCVATMARK))
3637 revents |= events & (POLLPRI | POLLRDBAND);
3638 if ((events & POLLINIGNEOF) == 0) {
3639 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3640 revents |= events & (POLLIN | POLLRDNORM);
3641 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3645 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3646 revents |= events & POLLRDHUP;
3649 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
3650 selrecord(td, &so->so_rdsel);
3651 so->so_rcv.sb_flags |= SB_SEL;
3653 if (events & (POLLOUT | POLLWRNORM)) {
3654 selrecord(td, &so->so_wrsel);
3655 so->so_snd.sb_flags |= SB_SEL;
3658 SOCK_RECVBUF_UNLOCK(so);
3659 SOCK_SENDBUF_UNLOCK(so);
3666 soo_kqfilter(struct file *fp, struct knote *kn)
3668 struct socket *so = kn->kn_fp->f_data;
3673 switch (kn->kn_filter) {
3675 kn->kn_fop = &soread_filtops;
3676 knl = &so->so_rdsel.si_note;
3681 kn->kn_fop = &sowrite_filtops;
3682 knl = &so->so_wrsel.si_note;
3687 kn->kn_fop = &soempty_filtops;
3688 knl = &so->so_wrsel.si_note;
3697 if (SOLISTENING(so)) {
3698 knlist_add(knl, kn, 1);
3700 SOCK_BUF_LOCK(so, which);
3701 knlist_add(knl, kn, 1);
3702 sb->sb_flags |= SB_KNOTE;
3703 SOCK_BUF_UNLOCK(so, which);
3710 filt_sordetach(struct knote *kn)
3712 struct socket *so = kn->kn_fp->f_data;
3715 knlist_remove(&so->so_rdsel.si_note, kn, 1);
3716 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3717 so->so_rcv.sb_flags &= ~SB_KNOTE;
3718 so_rdknl_unlock(so);
3723 filt_soread(struct knote *kn, long hint)
3727 so = kn->kn_fp->f_data;
3729 if (SOLISTENING(so)) {
3730 SOCK_LOCK_ASSERT(so);
3731 kn->kn_data = so->sol_qlen;
3733 kn->kn_flags |= EV_EOF;
3734 kn->kn_fflags = so->so_error;
3737 return (!TAILQ_EMPTY(&so->sol_comp));
3740 SOCK_RECVBUF_LOCK_ASSERT(so);
3742 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3743 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3744 kn->kn_flags |= EV_EOF;
3745 kn->kn_fflags = so->so_error;
3747 } else if (so->so_error || so->so_rerror)
3750 if (kn->kn_sfflags & NOTE_LOWAT) {
3751 if (kn->kn_data >= kn->kn_sdata)
3753 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3756 /* This hook returning non-zero indicates an event, not error */
3757 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3761 filt_sowdetach(struct knote *kn)
3763 struct socket *so = kn->kn_fp->f_data;
3766 knlist_remove(&so->so_wrsel.si_note, kn, 1);
3767 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3768 so->so_snd.sb_flags &= ~SB_KNOTE;
3769 so_wrknl_unlock(so);
3774 filt_sowrite(struct knote *kn, long hint)
3778 so = kn->kn_fp->f_data;
3780 if (SOLISTENING(so))
3783 SOCK_SENDBUF_LOCK_ASSERT(so);
3784 kn->kn_data = sbspace(&so->so_snd);
3786 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3788 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3789 kn->kn_flags |= EV_EOF;
3790 kn->kn_fflags = so->so_error;
3792 } else if (so->so_error) /* temporary udp error */
3794 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3795 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3797 else if (kn->kn_sfflags & NOTE_LOWAT)
3798 return (kn->kn_data >= kn->kn_sdata);
3800 return (kn->kn_data >= so->so_snd.sb_lowat);
3804 filt_soempty(struct knote *kn, long hint)
3808 so = kn->kn_fp->f_data;
3810 if (SOLISTENING(so))
3813 SOCK_SENDBUF_LOCK_ASSERT(so);
3814 kn->kn_data = sbused(&so->so_snd);
3816 if (kn->kn_data == 0)
3823 socheckuid(struct socket *so, uid_t uid)
3828 if (so->so_cred->cr_uid != uid)
3834 * These functions are used by protocols to notify the socket layer (and its
3835 * consumers) of state changes in the sockets driven by protocol-side events.
3839 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3841 * Normal sequence from the active (originating) side is that
3842 * soisconnecting() is called during processing of connect() call, resulting
3843 * in an eventual call to soisconnected() if/when the connection is
3844 * established. When the connection is torn down soisdisconnecting() is
3845 * called during processing of disconnect() call, and soisdisconnected() is
3846 * called when the connection to the peer is totally severed. The semantics
3847 * of these routines are such that connectionless protocols can call
3848 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3849 * calls when setting up a ``connection'' takes no time.
3851 * From the passive side, a socket is created with two queues of sockets:
3852 * so_incomp for connections in progress and so_comp for connections already
3853 * made and awaiting user acceptance. As a protocol is preparing incoming
3854 * connections, it creates a socket structure queued on so_incomp by calling
3855 * sonewconn(). When the connection is established, soisconnected() is
3856 * called, and transfers the socket structure to so_comp, making it available
3859 * If a socket is closed with sockets on either so_incomp or so_comp, these
3860 * sockets are dropped.
3862 * If higher-level protocols are implemented in the kernel, the wakeups done
3863 * here will sometimes cause software-interrupt process scheduling.
3866 soisconnecting(struct socket *so)
3870 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3871 so->so_state |= SS_ISCONNECTING;
3876 soisconnected(struct socket *so)
3878 bool last __diagused;
3881 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3882 so->so_state |= SS_ISCONNECTED;
3884 if (so->so_qstate == SQ_INCOMP) {
3885 struct socket *head = so->so_listen;
3888 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3890 * Promoting a socket from incomplete queue to complete, we
3891 * need to go through reverse order of locking. We first do
3892 * trylock, and if that doesn't succeed, we go the hard way
3893 * leaving a reference and rechecking consistency after proper
3896 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3899 SOLISTEN_LOCK(head);
3901 if (__predict_false(head != so->so_listen)) {
3903 * The socket went off the listen queue,
3904 * should be lost race to close(2) of sol.
3905 * The socket is about to soabort().
3908 sorele_locked(head);
3911 last = refcount_release(&head->so_count);
3912 KASSERT(!last, ("%s: released last reference for %p",
3916 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3917 TAILQ_REMOVE(&head->sol_incomp, so, so_list);
3918 head->sol_incqlen--;
3919 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
3921 so->so_qstate = SQ_COMP;
3923 solisten_wakeup(head); /* unlocks */
3925 SOCK_RECVBUF_LOCK(so);
3926 soupcall_set(so, SO_RCV,
3927 head->sol_accept_filter->accf_callback,
3928 head->sol_accept_filter_arg);
3929 so->so_options &= ~SO_ACCEPTFILTER;
3930 ret = head->sol_accept_filter->accf_callback(so,
3931 head->sol_accept_filter_arg, M_NOWAIT);
3932 if (ret == SU_ISCONNECTED) {
3933 soupcall_clear(so, SO_RCV);
3934 SOCK_RECVBUF_UNLOCK(so);
3937 SOCK_RECVBUF_UNLOCK(so);
3939 SOLISTEN_UNLOCK(head);
3944 wakeup(&so->so_timeo);
3950 soisdisconnecting(struct socket *so)
3954 so->so_state &= ~SS_ISCONNECTING;
3955 so->so_state |= SS_ISDISCONNECTING;
3957 if (!SOLISTENING(so)) {
3958 SOCK_RECVBUF_LOCK(so);
3959 socantrcvmore_locked(so);
3960 SOCK_SENDBUF_LOCK(so);
3961 socantsendmore_locked(so);
3964 wakeup(&so->so_timeo);
3968 soisdisconnected(struct socket *so)
3974 * There is at least one reader of so_state that does not
3975 * acquire socket lock, namely soreceive_generic(). Ensure
3976 * that it never sees all flags that track connection status
3977 * cleared, by ordering the update with a barrier semantic of
3978 * our release thread fence.
3980 so->so_state |= SS_ISDISCONNECTED;
3981 atomic_thread_fence_rel();
3982 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3984 if (!SOLISTENING(so)) {
3986 SOCK_RECVBUF_LOCK(so);
3987 socantrcvmore_locked(so);
3988 SOCK_SENDBUF_LOCK(so);
3989 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
3990 socantsendmore_locked(so);
3993 wakeup(&so->so_timeo);
3997 soiolock(struct socket *so, struct sx *sx, int flags)
4001 KASSERT((flags & SBL_VALID) == flags,
4002 ("soiolock: invalid flags %#x", flags));
4004 if ((flags & SBL_WAIT) != 0) {
4005 if ((flags & SBL_NOINTR) != 0) {
4008 error = sx_xlock_sig(sx);
4012 } else if (!sx_try_xlock(sx)) {
4013 return (EWOULDBLOCK);
4016 if (__predict_false(SOLISTENING(so))) {
4024 soiounlock(struct sx *sx)
4030 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4033 sodupsockaddr(const struct sockaddr *sa, int mflags)
4035 struct sockaddr *sa2;
4037 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4039 bcopy(sa, sa2, sa->sa_len);
4044 * Register per-socket destructor.
4047 sodtor_set(struct socket *so, so_dtor_t *func)
4050 SOCK_LOCK_ASSERT(so);
4055 * Register per-socket buffer upcalls.
4058 soupcall_set(struct socket *so, sb_which which, so_upcall_t func, void *arg)
4062 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4072 SOCK_BUF_LOCK_ASSERT(so, which);
4073 sb->sb_upcall = func;
4074 sb->sb_upcallarg = arg;
4075 sb->sb_flags |= SB_UPCALL;
4079 soupcall_clear(struct socket *so, sb_which which)
4083 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4093 SOCK_BUF_LOCK_ASSERT(so, which);
4094 KASSERT(sb->sb_upcall != NULL,
4095 ("%s: so %p no upcall to clear", __func__, so));
4096 sb->sb_upcall = NULL;
4097 sb->sb_upcallarg = NULL;
4098 sb->sb_flags &= ~SB_UPCALL;
4102 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4105 SOLISTEN_LOCK_ASSERT(so);
4106 so->sol_upcall = func;
4107 so->sol_upcallarg = arg;
4111 so_rdknl_lock(void *arg)
4113 struct socket *so = arg;
4116 if (SOLISTENING(so)) {
4119 SOCK_RECVBUF_LOCK(so);
4120 if (__predict_false(SOLISTENING(so))) {
4121 SOCK_RECVBUF_UNLOCK(so);
4128 so_rdknl_unlock(void *arg)
4130 struct socket *so = arg;
4132 if (SOLISTENING(so))
4133 SOLISTEN_UNLOCK(so);
4135 SOCK_RECVBUF_UNLOCK(so);
4139 so_rdknl_assert_lock(void *arg, int what)
4141 struct socket *so = arg;
4143 if (what == LA_LOCKED) {
4144 if (SOLISTENING(so))
4145 SOLISTEN_LOCK_ASSERT(so);
4147 SOCK_RECVBUF_LOCK_ASSERT(so);
4149 if (SOLISTENING(so))
4150 SOLISTEN_UNLOCK_ASSERT(so);
4152 SOCK_RECVBUF_UNLOCK_ASSERT(so);
4157 so_wrknl_lock(void *arg)
4159 struct socket *so = arg;
4162 if (SOLISTENING(so)) {
4165 SOCK_SENDBUF_LOCK(so);
4166 if (__predict_false(SOLISTENING(so))) {
4167 SOCK_SENDBUF_UNLOCK(so);
4174 so_wrknl_unlock(void *arg)
4176 struct socket *so = arg;
4178 if (SOLISTENING(so))
4179 SOLISTEN_UNLOCK(so);
4181 SOCK_SENDBUF_UNLOCK(so);
4185 so_wrknl_assert_lock(void *arg, int what)
4187 struct socket *so = arg;
4189 if (what == LA_LOCKED) {
4190 if (SOLISTENING(so))
4191 SOLISTEN_LOCK_ASSERT(so);
4193 SOCK_SENDBUF_LOCK_ASSERT(so);
4195 if (SOLISTENING(so))
4196 SOLISTEN_UNLOCK_ASSERT(so);
4198 SOCK_SENDBUF_UNLOCK_ASSERT(so);
4203 * Create an external-format (``xsocket'') structure using the information in
4204 * the kernel-format socket structure pointed to by so. This is done to
4205 * reduce the spew of irrelevant information over this interface, to isolate
4206 * user code from changes in the kernel structure, and potentially to provide
4207 * information-hiding if we decide that some of this information should be
4208 * hidden from users.
4211 sotoxsocket(struct socket *so, struct xsocket *xso)
4214 bzero(xso, sizeof(*xso));
4215 xso->xso_len = sizeof *xso;
4216 xso->xso_so = (uintptr_t)so;
4217 xso->so_type = so->so_type;
4218 xso->so_options = so->so_options;
4219 xso->so_linger = so->so_linger;
4220 xso->so_state = so->so_state;
4221 xso->so_pcb = (uintptr_t)so->so_pcb;
4222 xso->xso_protocol = so->so_proto->pr_protocol;
4223 xso->xso_family = so->so_proto->pr_domain->dom_family;
4224 xso->so_timeo = so->so_timeo;
4225 xso->so_error = so->so_error;
4226 xso->so_uid = so->so_cred->cr_uid;
4227 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4228 if (SOLISTENING(so)) {
4229 xso->so_qlen = so->sol_qlen;
4230 xso->so_incqlen = so->sol_incqlen;
4231 xso->so_qlimit = so->sol_qlimit;
4232 xso->so_oobmark = 0;
4234 xso->so_state |= so->so_qstate;
4235 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4236 xso->so_oobmark = so->so_oobmark;
4237 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4238 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4243 so_sockbuf_rcv(struct socket *so)
4246 return (&so->so_rcv);
4250 so_sockbuf_snd(struct socket *so)
4253 return (&so->so_snd);
4257 so_state_get(const struct socket *so)
4260 return (so->so_state);
4264 so_state_set(struct socket *so, int val)
4271 so_options_get(const struct socket *so)
4274 return (so->so_options);
4278 so_options_set(struct socket *so, int val)
4281 so->so_options = val;
4285 so_error_get(const struct socket *so)
4288 return (so->so_error);
4292 so_error_set(struct socket *so, int val)
4299 so_linger_get(const struct socket *so)
4302 return (so->so_linger);
4306 so_linger_set(struct socket *so, int val)
4309 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4310 ("%s: val %d out of range", __func__, val));
4312 so->so_linger = val;
4316 so_protosw_get(const struct socket *so)
4319 return (so->so_proto);
4323 so_protosw_set(struct socket *so, struct protosw *val)
4330 so_sorwakeup(struct socket *so)
4337 so_sowwakeup(struct socket *so)
4344 so_sorwakeup_locked(struct socket *so)
4347 sorwakeup_locked(so);
4351 so_sowwakeup_locked(struct socket *so)
4354 sowwakeup_locked(so);
4358 so_lock(struct socket *so)
4365 so_unlock(struct socket *so)