2 * Copyright (c) 2004 The FreeBSD Foundation
3 * Copyright (c) 2004-2005 Robert N. M. Watson
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 4. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
40 #include "opt_compat.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/fcntl.h>
45 #include <sys/limits.h>
48 #include <sys/malloc.h>
50 #include <sys/mutex.h>
51 #include <sys/domain.h>
52 #include <sys/file.h> /* for struct knote */
53 #include <sys/kernel.h>
54 #include <sys/event.h>
57 #include <sys/protosw.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/resourcevar.h>
61 #include <sys/signalvar.h>
62 #include <sys/sysctl.h>
69 #include <sys/mount.h>
70 #include <compat/freebsd32/freebsd32.h>
72 extern struct sysentvec ia32_freebsd_sysvec;
75 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
78 static void filt_sordetach(struct knote *kn);
79 static int filt_soread(struct knote *kn, long hint);
80 static void filt_sowdetach(struct knote *kn);
81 static int filt_sowrite(struct knote *kn, long hint);
82 static int filt_solisten(struct knote *kn, long hint);
84 static struct filterops solisten_filtops =
85 { 1, NULL, filt_sordetach, filt_solisten };
86 static struct filterops soread_filtops =
87 { 1, NULL, filt_sordetach, filt_soread };
88 static struct filterops sowrite_filtops =
89 { 1, NULL, filt_sowdetach, filt_sowrite };
91 uma_zone_t socket_zone;
92 so_gen_t so_gencnt; /* generation count for sockets */
94 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
95 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
97 SYSCTL_DECL(_kern_ipc);
99 static int somaxconn = SOMAXCONN;
100 static int somaxconn_sysctl(SYSCTL_HANDLER_ARGS);
101 /* XXX: we dont have SYSCTL_USHORT */
102 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
103 0, sizeof(int), somaxconn_sysctl, "I", "Maximum pending socket connection "
105 static int numopensockets;
106 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
107 &numopensockets, 0, "Number of open sockets");
108 #ifdef ZERO_COPY_SOCKETS
109 /* These aren't static because they're used in other files. */
110 int so_zero_copy_send = 1;
111 int so_zero_copy_receive = 1;
112 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
113 "Zero copy controls");
114 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
115 &so_zero_copy_receive, 0, "Enable zero copy receive");
116 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
117 &so_zero_copy_send, 0, "Enable zero copy send");
118 #endif /* ZERO_COPY_SOCKETS */
121 * accept_mtx locks down per-socket fields relating to accept queues. See
122 * socketvar.h for an annotation of the protected fields of struct socket.
124 struct mtx accept_mtx;
125 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
128 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
131 static struct mtx so_global_mtx;
132 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
135 * Socket operation routines.
136 * These routines are called by the routines in
137 * sys_socket.c or from a system process, and
138 * implement the semantics of socket operations by
139 * switching out to the protocol specific routines.
143 * Get a socket structure from our zone, and initialize it.
144 * Note that it would probably be better to allocate socket
145 * and PCB at the same time, but I'm not convinced that all
146 * the protocols can be easily modified to do this.
148 * soalloc() returns a socket with a ref count of 0.
155 so = uma_zalloc(socket_zone, mflags | M_ZERO);
158 if (mac_init_socket(so, mflags) != 0) {
159 uma_zfree(socket_zone, so);
163 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
164 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
165 TAILQ_INIT(&so->so_aiojobq);
166 mtx_lock(&so_global_mtx);
167 so->so_gencnt = ++so_gencnt;
169 mtx_unlock(&so_global_mtx);
175 * socreate returns a socket with a ref count of 1. The socket should be
176 * closed with soclose().
179 socreate(dom, aso, type, proto, cred, td)
192 prp = pffindproto(dom, proto, type);
194 prp = pffindtype(dom, type);
196 if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
197 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
198 return (EPROTONOSUPPORT);
200 if (jailed(cred) && jail_socket_unixiproute_only &&
201 prp->pr_domain->dom_family != PF_LOCAL &&
202 prp->pr_domain->dom_family != PF_INET &&
203 prp->pr_domain->dom_family != PF_ROUTE) {
204 return (EPROTONOSUPPORT);
207 if (prp->pr_type != type)
209 so = soalloc(M_WAITOK);
213 TAILQ_INIT(&so->so_incomp);
214 TAILQ_INIT(&so->so_comp);
216 so->so_cred = crhold(cred);
219 mac_create_socket(cred, so);
221 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
223 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
226 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
230 so->so_state |= SS_NOFDREF;
241 struct sockaddr *nam;
245 return ((*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td));
249 sodealloc(struct socket *so)
252 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
253 mtx_lock(&so_global_mtx);
254 so->so_gencnt = ++so_gencnt;
255 mtx_unlock(&so_global_mtx);
256 if (so->so_rcv.sb_hiwat)
257 (void)chgsbsize(so->so_cred->cr_uidinfo,
258 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
259 if (so->so_snd.sb_hiwat)
260 (void)chgsbsize(so->so_cred->cr_uidinfo,
261 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
263 /* remove acccept filter if one is present. */
264 if (so->so_accf != NULL)
265 do_setopt_accept_filter(so, NULL);
268 mac_destroy_socket(so);
271 SOCKBUF_LOCK_DESTROY(&so->so_snd);
272 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
273 uma_zfree(socket_zone, so);
274 mtx_lock(&so_global_mtx);
276 mtx_unlock(&so_global_mtx);
280 * solisten() transitions a socket from a non-listening state to a listening
281 * state, but can also be used to update the listen queue depth on an
282 * existing listen socket. The protocol will call back into the sockets
283 * layer using solisten_proto_check() and solisten_proto() to check and set
284 * socket-layer listen state. Call backs are used so that the protocol can
285 * acquire both protocol and socket layer locks in whatever order is required
288 * Protocol implementors are advised to hold the socket lock across the
289 * socket-layer test and set to avoid races at the socket layer.
292 solisten(so, backlog, td)
298 return ((*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td));
302 solisten_proto_check(so)
306 SOCK_LOCK_ASSERT(so);
308 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
315 solisten_proto(so, backlog)
320 SOCK_LOCK_ASSERT(so);
322 if (backlog < 0 || backlog > somaxconn)
324 so->so_qlimit = backlog;
325 so->so_options |= SO_ACCEPTCONN;
329 * Attempt to free a socket. This should really be sotryfree().
331 * We free the socket if the protocol is no longer interested in the socket,
332 * there's no file descriptor reference, and the refcount is 0. While the
333 * calling macro sotryfree() tests the refcount, sofree() has to test it
334 * again as it's possible to race with an accept()ing thread if the socket is
335 * in an listen queue of a listen socket, as being in the listen queue
336 * doesn't elevate the reference count. sofree() acquires the accept mutex
337 * early for this test in order to avoid that race.
345 ACCEPT_LOCK_ASSERT();
346 SOCK_LOCK_ASSERT(so);
348 if (so->so_pcb != NULL || (so->so_state & SS_NOFDREF) == 0 ||
357 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
358 (so->so_qstate & SQ_INCOMP) != 0,
359 ("sofree: so_head != NULL, but neither SQ_COMP nor "
361 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
362 (so->so_qstate & SQ_INCOMP) == 0,
363 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
365 * accept(2) is responsible draining the completed
366 * connection queue and freeing those sockets, so
367 * we just return here if this socket is currently
368 * on the completed connection queue. Otherwise,
369 * accept(2) may hang after select(2) has indicating
370 * that a listening socket was ready. If it's an
371 * incomplete connection, we remove it from the queue
372 * and free it; otherwise, it won't be released until
373 * the listening socket is closed.
375 if ((so->so_qstate & SQ_COMP) != 0) {
380 TAILQ_REMOVE(&head->so_incomp, so, so_list);
382 so->so_qstate &= ~SQ_INCOMP;
385 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
386 (so->so_qstate & SQ_INCOMP) == 0,
387 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
388 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
391 SOCKBUF_LOCK(&so->so_snd);
392 so->so_snd.sb_flags |= SB_NOINTR;
393 (void)sblock(&so->so_snd, M_WAITOK);
395 * socantsendmore_locked() drops the socket buffer mutex so that it
396 * can safely perform wakeups. Re-acquire the mutex before
399 socantsendmore_locked(so);
400 SOCKBUF_LOCK(&so->so_snd);
401 sbunlock(&so->so_snd);
402 sbrelease_locked(&so->so_snd, so);
403 SOCKBUF_UNLOCK(&so->so_snd);
405 knlist_destroy(&so->so_rcv.sb_sel.si_note);
406 knlist_destroy(&so->so_snd.sb_sel.si_note);
411 * Close a socket on last file table reference removal.
412 * Initiate disconnect if connected.
413 * Free socket when disconnect complete.
415 * This function will sorele() the socket. Note that soclose() may be
416 * called prior to the ref count reaching zero. The actual socket
417 * structure will not be freed until the ref count reaches zero.
425 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
427 funsetown(&so->so_sigio);
428 if (so->so_options & SO_ACCEPTCONN) {
431 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
432 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
434 sp->so_qstate &= ~SQ_INCOMP;
440 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
441 TAILQ_REMOVE(&so->so_comp, sp, so_list);
443 sp->so_qstate &= ~SQ_COMP;
451 if (so->so_pcb == NULL)
453 if (so->so_state & SS_ISCONNECTED) {
454 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
455 error = sodisconnect(so);
459 if (so->so_options & SO_LINGER) {
460 if ((so->so_state & SS_ISDISCONNECTING) &&
461 (so->so_state & SS_NBIO))
463 while (so->so_state & SS_ISCONNECTED) {
464 error = tsleep(&so->so_timeo,
465 PSOCK | PCATCH, "soclos", so->so_linger * hz);
472 if (so->so_pcb != NULL) {
473 int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so);
480 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
481 so->so_state |= SS_NOFDREF;
487 * soabort() must not be called with any socket locks held, as it calls
488 * into the protocol, which will call back into the socket code causing
489 * it to acquire additional socket locks that may cause recursion or lock
498 error = (*so->so_proto->pr_usrreqs->pru_abort)(so);
502 sotryfree(so); /* note: does not decrement the ref count */
511 struct sockaddr **nam;
516 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
517 so->so_state &= ~SS_NOFDREF;
519 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
524 soconnect(so, nam, td)
526 struct sockaddr *nam;
531 if (so->so_options & SO_ACCEPTCONN)
534 * If protocol is connection-based, can only connect once.
535 * Otherwise, if connected, try to disconnect first.
536 * This allows user to disconnect by connecting to, e.g.,
539 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
540 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
541 (error = sodisconnect(so)))) {
545 * Prevent accumulated error from previous connection
549 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
561 return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
570 if ((so->so_state & SS_ISCONNECTED) == 0)
572 if (so->so_state & SS_ISDISCONNECTING)
574 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
578 #ifdef ZERO_COPY_SOCKETS
579 struct so_zerocopy_stats{
584 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
585 #include <netinet/in.h>
586 #include <net/route.h>
587 #include <netinet/in_pcb.h>
589 #include <vm/vm_page.h>
590 #include <vm/vm_object.h>
591 #endif /*ZERO_COPY_SOCKETS*/
594 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
595 * all of the data referenced by the uio. If desired, it uses zero-copy.
596 * *space will be updated to reflect data copied in.
598 * NB: If atomic I/O is requested, the caller must already have checked that
599 * space can hold resid bytes.
601 * NB: In the event of an error, the caller may need to free the partial
602 * chain pointed to by *mpp. The contents of both *uio and *space may be
603 * modified even in the case of an error.
606 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
609 struct mbuf *m, **mp, *top;
612 #ifdef ZERO_COPY_SOCKETS
619 resid = uio->uio_resid;
622 #ifdef ZERO_COPY_SOCKETS
624 #endif /* ZERO_COPY_SOCKETS */
625 if (resid >= MINCLSIZE) {
626 #ifdef ZERO_COPY_SOCKETS
628 MGETHDR(m, M_TRYWAIT, MT_DATA);
634 m->m_pkthdr.rcvif = NULL;
636 MGET(m, M_TRYWAIT, MT_DATA);
642 if (so_zero_copy_send &&
645 uio->uio_iov->iov_len>=PAGE_SIZE) {
646 so_zerocp_stats.size_ok++;
647 so_zerocp_stats.align_ok++;
648 cow_send = socow_setup(m, uio);
652 MCLGET(m, M_TRYWAIT);
653 if ((m->m_flags & M_EXT) == 0) {
657 len = min(min(MCLBYTES, resid),
661 #else /* ZERO_COPY_SOCKETS */
663 m = m_getcl(M_TRYWAIT, MT_DATA, M_PKTHDR);
665 m->m_pkthdr.rcvif = NULL;
667 m = m_getcl(M_TRYWAIT, MT_DATA, 0);
668 len = min(min(MCLBYTES, resid), *space);
669 #endif /* ZERO_COPY_SOCKETS */
672 m = m_gethdr(M_TRYWAIT, MT_DATA);
674 m->m_pkthdr.rcvif = NULL;
676 len = min(min(MHLEN, resid), *space);
678 * For datagram protocols, leave room
679 * for protocol headers in first mbuf.
681 if (atomic && m && len < MHLEN)
684 m = m_get(M_TRYWAIT, MT_DATA);
685 len = min(min(MLEN, resid), *space);
694 #ifdef ZERO_COPY_SOCKETS
698 #endif /* ZERO_COPY_SOCKETS */
699 error = uiomove(mtod(m, void *), (int)len, uio);
700 resid = uio->uio_resid;
703 top->m_pkthdr.len += len;
709 top->m_flags |= M_EOR;
712 } while (*space > 0 && atomic);
718 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
721 sosend_dgram(so, addr, uio, top, control, flags, td)
723 struct sockaddr *addr;
726 struct mbuf *control;
731 int clen = 0, error, dontroute;
732 int atomic = sosendallatonce(so) || top;
734 KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
735 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
736 ("sodgram_send: !PR_ATOMIC"));
739 resid = uio->uio_resid;
741 resid = top->m_pkthdr.len;
743 * In theory resid should be unsigned.
744 * However, space must be signed, as it might be less than 0
745 * if we over-committed, and we must use a signed comparison
746 * of space and resid. On the other hand, a negative resid
747 * causes us to loop sending 0-length segments to the protocol.
749 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
750 * type sockets since that's an error.
758 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
760 td->td_proc->p_stats->p_ru.ru_msgsnd++;
762 clen = control->m_len;
764 SOCKBUF_LOCK(&so->so_snd);
765 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
766 SOCKBUF_UNLOCK(&so->so_snd);
771 error = so->so_error;
773 SOCKBUF_UNLOCK(&so->so_snd);
776 if ((so->so_state & SS_ISCONNECTED) == 0) {
778 * `sendto' and `sendmsg' is allowed on a connection-
779 * based socket if it supports implied connect.
780 * Return ENOTCONN if not connected and no address is
783 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
784 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
785 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
786 !(resid == 0 && clen != 0)) {
787 SOCKBUF_UNLOCK(&so->so_snd);
791 } else if (addr == NULL) {
792 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
795 error = EDESTADDRREQ;
796 SOCKBUF_UNLOCK(&so->so_snd);
802 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
803 * problem and need fixing.
805 space = sbspace(&so->so_snd);
813 SOCKBUF_UNLOCK(&so->so_snd);
817 top->m_flags |= M_EOR;
819 error = sosend_copyin(uio, &top, atomic, &space, flags);
822 resid = uio->uio_resid;
824 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
826 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
831 so->so_options |= SO_DONTROUTE;
835 * XXX all the SBS_CANTSENDMORE checks previously
836 * done could be out of date. We could have recieved
837 * a reset packet in an interrupt or maybe we slept
838 * while doing page faults in uiomove() etc. We could
839 * probably recheck again inside the locking protection
840 * here, but there are probably other places that this
841 * also happens. We must rethink this.
843 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
844 (flags & MSG_OOB) ? PRUS_OOB :
846 * If the user set MSG_EOF, the protocol
847 * understands this flag and nothing left to
848 * send then use PRU_SEND_EOF instead of PRU_SEND.
850 ((flags & MSG_EOF) &&
851 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
854 /* If there is more to send set PRUS_MORETOCOME */
855 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
856 top, addr, control, td);
859 so->so_options &= ~SO_DONTROUTE;
875 * If send must go all at once and message is larger than
876 * send buffering, then hard error.
877 * Lock against other senders.
878 * If must go all at once and not enough room now, then
879 * inform user that this would block and do nothing.
880 * Otherwise, if nonblocking, send as much as possible.
881 * The data to be sent is described by "uio" if nonzero,
882 * otherwise by the mbuf chain "top" (which must be null
883 * if uio is not). Data provided in mbuf chain must be small
884 * enough to send all at once.
886 * Returns nonzero on error, timeout or signal; callers
887 * must check for short counts if EINTR/ERESTART are returned.
888 * Data and control buffers are freed on return.
890 #define snderr(errno) { error = (errno); goto release; }
892 sosend(so, addr, uio, top, control, flags, td)
894 struct sockaddr *addr;
897 struct mbuf *control;
902 int clen = 0, error, dontroute;
903 int atomic = sosendallatonce(so) || top;
906 resid = uio->uio_resid;
908 resid = top->m_pkthdr.len;
910 * In theory resid should be unsigned.
911 * However, space must be signed, as it might be less than 0
912 * if we over-committed, and we must use a signed comparison
913 * of space and resid. On the other hand, a negative resid
914 * causes us to loop sending 0-length segments to the protocol.
916 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
917 * type sockets since that's an error.
919 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
925 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
926 (so->so_proto->pr_flags & PR_ATOMIC);
928 td->td_proc->p_stats->p_ru.ru_msgsnd++;
930 clen = control->m_len;
932 SOCKBUF_LOCK(&so->so_snd);
934 SOCKBUF_LOCK_ASSERT(&so->so_snd);
935 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
939 SOCKBUF_LOCK_ASSERT(&so->so_snd);
940 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
943 error = so->so_error;
947 if ((so->so_state & SS_ISCONNECTED) == 0) {
949 * `sendto' and `sendmsg' is allowed on a connection-
950 * based socket if it supports implied connect.
951 * Return ENOTCONN if not connected and no address is
954 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
955 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
956 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
957 !(resid == 0 && clen != 0))
959 } else if (addr == NULL)
960 snderr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
961 ENOTCONN : EDESTADDRREQ);
963 space = sbspace(&so->so_snd);
966 if ((atomic && resid > so->so_snd.sb_hiwat) ||
967 clen > so->so_snd.sb_hiwat)
969 if (space < resid + clen &&
970 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
971 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO))
973 sbunlock(&so->so_snd);
974 error = sbwait(&so->so_snd);
979 SOCKBUF_UNLOCK(&so->so_snd);
985 top->m_flags |= M_EOR;
987 error = sosend_copyin(uio, &top, atomic,
990 SOCKBUF_LOCK(&so->so_snd);
993 resid = uio->uio_resid;
997 so->so_options |= SO_DONTROUTE;
1001 * XXX all the SBS_CANTSENDMORE checks previously
1002 * done could be out of date. We could have recieved
1003 * a reset packet in an interrupt or maybe we slept
1004 * while doing page faults in uiomove() etc. We could
1005 * probably recheck again inside the locking protection
1006 * here, but there are probably other places that this
1007 * also happens. We must rethink this.
1009 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1010 (flags & MSG_OOB) ? PRUS_OOB :
1012 * If the user set MSG_EOF, the protocol
1013 * understands this flag and nothing left to
1014 * send then use PRU_SEND_EOF instead of PRU_SEND.
1016 ((flags & MSG_EOF) &&
1017 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1020 /* If there is more to send set PRUS_MORETOCOME */
1021 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1022 top, addr, control, td);
1025 so->so_options &= ~SO_DONTROUTE;
1032 SOCKBUF_LOCK(&so->so_snd);
1035 } while (resid && space > 0);
1036 SOCKBUF_LOCK(&so->so_snd);
1040 SOCKBUF_LOCK_ASSERT(&so->so_snd);
1041 sbunlock(&so->so_snd);
1043 SOCKBUF_LOCK_ASSERT(&so->so_snd);
1044 SOCKBUF_UNLOCK(&so->so_snd);
1048 if (control != NULL)
1055 * The part of soreceive() that implements reading non-inline out-of-band
1056 * data from a socket. For more complete comments, see soreceive(), from
1057 * which this code originated.
1059 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1060 * unable to return an mbuf chain to the caller.
1063 soreceive_rcvoob(so, uio, flags)
1068 struct protosw *pr = so->so_proto;
1072 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1074 m = m_get(M_TRYWAIT, MT_DATA);
1077 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1081 #ifdef ZERO_COPY_SOCKETS
1082 if (so_zero_copy_receive) {
1085 if ((m->m_flags & M_EXT)
1086 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1091 error = uiomoveco(mtod(m, void *),
1092 min(uio->uio_resid, m->m_len),
1095 #endif /* ZERO_COPY_SOCKETS */
1096 error = uiomove(mtod(m, void *),
1097 (int) min(uio->uio_resid, m->m_len), uio);
1099 } while (uio->uio_resid && error == 0 && m);
1107 * Following replacement or removal of the first mbuf on the first mbuf chain
1108 * of a socket buffer, push necessary state changes back into the socket
1109 * buffer so that other consumers see the values consistently. 'nextrecord'
1110 * is the callers locally stored value of the original value of
1111 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1112 * NOTE: 'nextrecord' may be NULL.
1114 static __inline void
1115 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1118 SOCKBUF_LOCK_ASSERT(sb);
1120 * First, update for the new value of nextrecord. If necessary, make
1121 * it the first record.
1123 if (sb->sb_mb != NULL)
1124 sb->sb_mb->m_nextpkt = nextrecord;
1126 sb->sb_mb = nextrecord;
1129 * Now update any dependent socket buffer fields to reflect the new
1130 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1131 * addition of a second clause that takes care of the case where
1132 * sb_mb has been updated, but remains the last record.
1134 if (sb->sb_mb == NULL) {
1135 sb->sb_mbtail = NULL;
1136 sb->sb_lastrecord = NULL;
1137 } else if (sb->sb_mb->m_nextpkt == NULL)
1138 sb->sb_lastrecord = sb->sb_mb;
1143 * Implement receive operations on a socket.
1144 * We depend on the way that records are added to the sockbuf
1145 * by sbappend*. In particular, each record (mbufs linked through m_next)
1146 * must begin with an address if the protocol so specifies,
1147 * followed by an optional mbuf or mbufs containing ancillary data,
1148 * and then zero or more mbufs of data.
1149 * In order to avoid blocking network interrupts for the entire time here,
1150 * we splx() while doing the actual copy to user space.
1151 * Although the sockbuf is locked, new data may still be appended,
1152 * and thus we must maintain consistency of the sockbuf during that time.
1154 * The caller may receive the data as a single mbuf chain by supplying
1155 * an mbuf **mp0 for use in returning the chain. The uio is then used
1156 * only for the count in uio_resid.
1159 soreceive(so, psa, uio, mp0, controlp, flagsp)
1161 struct sockaddr **psa;
1164 struct mbuf **controlp;
1167 struct mbuf *m, **mp;
1168 int flags, len, error, offset;
1169 struct protosw *pr = so->so_proto;
1170 struct mbuf *nextrecord;
1172 int orig_resid = uio->uio_resid;
1177 if (controlp != NULL)
1180 flags = *flagsp &~ MSG_EOR;
1183 if (flags & MSG_OOB)
1184 return (soreceive_rcvoob(so, uio, flags));
1187 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1189 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1191 SOCKBUF_LOCK(&so->so_rcv);
1193 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1194 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1198 m = so->so_rcv.sb_mb;
1200 * If we have less data than requested, block awaiting more
1201 * (subject to any timeout) if:
1202 * 1. the current count is less than the low water mark, or
1203 * 2. MSG_WAITALL is set, and it is possible to do the entire
1204 * receive operation at once if we block (resid <= hiwat).
1205 * 3. MSG_DONTWAIT is not set
1206 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1207 * we have to do the receive in sections, and thus risk returning
1208 * a short count if a timeout or signal occurs after we start.
1210 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1211 so->so_rcv.sb_cc < uio->uio_resid) &&
1212 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1213 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1214 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1215 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1216 ("receive: m == %p so->so_rcv.sb_cc == %u",
1217 m, so->so_rcv.sb_cc));
1221 error = so->so_error;
1222 if ((flags & MSG_PEEK) == 0)
1226 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1227 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1233 for (; m != NULL; m = m->m_next)
1234 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1235 m = so->so_rcv.sb_mb;
1238 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1239 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1243 if (uio->uio_resid == 0)
1245 if ((so->so_state & SS_NBIO) ||
1246 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1247 error = EWOULDBLOCK;
1250 SBLASTRECORDCHK(&so->so_rcv);
1251 SBLASTMBUFCHK(&so->so_rcv);
1252 sbunlock(&so->so_rcv);
1253 error = sbwait(&so->so_rcv);
1260 * From this point onward, we maintain 'nextrecord' as a cache of the
1261 * pointer to the next record in the socket buffer. We must keep the
1262 * various socket buffer pointers and local stack versions of the
1263 * pointers in sync, pushing out modifications before dropping the
1264 * socket buffer mutex, and re-reading them when picking it up.
1266 * Otherwise, we will race with the network stack appending new data
1267 * or records onto the socket buffer by using inconsistent/stale
1268 * versions of the field, possibly resulting in socket buffer
1271 * By holding the high-level sblock(), we prevent simultaneous
1272 * readers from pulling off the front of the socket buffer.
1274 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1276 uio->uio_td->td_proc->p_stats->p_ru.ru_msgrcv++;
1277 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1278 SBLASTRECORDCHK(&so->so_rcv);
1279 SBLASTMBUFCHK(&so->so_rcv);
1280 nextrecord = m->m_nextpkt;
1281 if (pr->pr_flags & PR_ADDR) {
1282 KASSERT(m->m_type == MT_SONAME,
1283 ("m->m_type == %d", m->m_type));
1286 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1288 if (flags & MSG_PEEK) {
1291 sbfree(&so->so_rcv, m);
1292 so->so_rcv.sb_mb = m_free(m);
1293 m = so->so_rcv.sb_mb;
1294 sockbuf_pushsync(&so->so_rcv, nextrecord);
1299 * Process one or more MT_CONTROL mbufs present before any data mbufs
1300 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1301 * just copy the data; if !MSG_PEEK, we call into the protocol to
1302 * perform externalization (or freeing if controlp == NULL).
1304 if (m != NULL && m->m_type == MT_CONTROL) {
1305 struct mbuf *cm = NULL, *cmn;
1306 struct mbuf **cme = &cm;
1309 if (flags & MSG_PEEK) {
1310 if (controlp != NULL) {
1311 *controlp = m_copy(m, 0, m->m_len);
1312 controlp = &(*controlp)->m_next;
1316 sbfree(&so->so_rcv, m);
1317 so->so_rcv.sb_mb = m->m_next;
1320 cme = &(*cme)->m_next;
1321 m = so->so_rcv.sb_mb;
1323 } while (m != NULL && m->m_type == MT_CONTROL);
1324 if ((flags & MSG_PEEK) == 0)
1325 sockbuf_pushsync(&so->so_rcv, nextrecord);
1326 while (cm != NULL) {
1329 if (pr->pr_domain->dom_externalize != NULL) {
1330 SOCKBUF_UNLOCK(&so->so_rcv);
1331 error = (*pr->pr_domain->dom_externalize)
1333 SOCKBUF_LOCK(&so->so_rcv);
1334 } else if (controlp != NULL)
1338 if (controlp != NULL) {
1340 while (*controlp != NULL)
1341 controlp = &(*controlp)->m_next;
1345 if (so->so_rcv.sb_mb)
1346 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1352 if ((flags & MSG_PEEK) == 0) {
1353 KASSERT(m->m_nextpkt == nextrecord,
1354 ("soreceive: post-control, nextrecord !sync"));
1355 if (nextrecord == NULL) {
1356 KASSERT(so->so_rcv.sb_mb == m,
1357 ("soreceive: post-control, sb_mb!=m"));
1358 KASSERT(so->so_rcv.sb_lastrecord == m,
1359 ("soreceive: post-control, lastrecord!=m"));
1363 if (type == MT_OOBDATA)
1366 if ((flags & MSG_PEEK) == 0) {
1367 KASSERT(so->so_rcv.sb_mb == nextrecord,
1368 ("soreceive: sb_mb != nextrecord"));
1369 if (so->so_rcv.sb_mb == NULL) {
1370 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1371 ("soreceive: sb_lastercord != NULL"));
1375 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1376 SBLASTRECORDCHK(&so->so_rcv);
1377 SBLASTMBUFCHK(&so->so_rcv);
1380 * Now continue to read any data mbufs off of the head of the socket
1381 * buffer until the read request is satisfied. Note that 'type' is
1382 * used to store the type of any mbuf reads that have happened so far
1383 * such that soreceive() can stop reading if the type changes, which
1384 * causes soreceive() to return only one of regular data and inline
1385 * out-of-band data in a single socket receive operation.
1389 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1391 * If the type of mbuf has changed since the last mbuf
1392 * examined ('type'), end the receive operation.
1394 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1395 if (m->m_type == MT_OOBDATA) {
1396 if (type != MT_OOBDATA)
1398 } else if (type == MT_OOBDATA)
1401 KASSERT(m->m_type == MT_DATA,
1402 ("m->m_type == %d", m->m_type));
1403 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1404 len = uio->uio_resid;
1405 if (so->so_oobmark && len > so->so_oobmark - offset)
1406 len = so->so_oobmark - offset;
1407 if (len > m->m_len - moff)
1408 len = m->m_len - moff;
1410 * If mp is set, just pass back the mbufs.
1411 * Otherwise copy them out via the uio, then free.
1412 * Sockbuf must be consistent here (points to current mbuf,
1413 * it points to next record) when we drop priority;
1414 * we must note any additions to the sockbuf when we
1415 * block interrupts again.
1418 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1419 SBLASTRECORDCHK(&so->so_rcv);
1420 SBLASTMBUFCHK(&so->so_rcv);
1421 SOCKBUF_UNLOCK(&so->so_rcv);
1422 #ifdef ZERO_COPY_SOCKETS
1423 if (so_zero_copy_receive) {
1426 if ((m->m_flags & M_EXT)
1427 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1432 error = uiomoveco(mtod(m, char *) + moff,
1436 #endif /* ZERO_COPY_SOCKETS */
1437 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1438 SOCKBUF_LOCK(&so->so_rcv);
1442 uio->uio_resid -= len;
1443 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1444 if (len == m->m_len - moff) {
1445 if (m->m_flags & M_EOR)
1447 if (flags & MSG_PEEK) {
1451 nextrecord = m->m_nextpkt;
1452 sbfree(&so->so_rcv, m);
1456 so->so_rcv.sb_mb = m = m->m_next;
1459 so->so_rcv.sb_mb = m_free(m);
1460 m = so->so_rcv.sb_mb;
1462 sockbuf_pushsync(&so->so_rcv, nextrecord);
1463 SBLASTRECORDCHK(&so->so_rcv);
1464 SBLASTMBUFCHK(&so->so_rcv);
1467 if (flags & MSG_PEEK)
1473 if (flags & MSG_DONTWAIT)
1474 copy_flag = M_DONTWAIT;
1476 copy_flag = M_TRYWAIT;
1477 if (copy_flag == M_TRYWAIT)
1478 SOCKBUF_UNLOCK(&so->so_rcv);
1479 *mp = m_copym(m, 0, len, copy_flag);
1480 if (copy_flag == M_TRYWAIT)
1481 SOCKBUF_LOCK(&so->so_rcv);
1484 * m_copym() couldn't allocate an mbuf.
1485 * Adjust uio_resid back (it was adjusted
1486 * down by len bytes, which we didn't end
1487 * up "copying" over).
1489 uio->uio_resid += len;
1495 so->so_rcv.sb_cc -= len;
1498 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1499 if (so->so_oobmark) {
1500 if ((flags & MSG_PEEK) == 0) {
1501 so->so_oobmark -= len;
1502 if (so->so_oobmark == 0) {
1503 so->so_rcv.sb_state |= SBS_RCVATMARK;
1508 if (offset == so->so_oobmark)
1512 if (flags & MSG_EOR)
1515 * If the MSG_WAITALL flag is set (for non-atomic socket),
1516 * we must not quit until "uio->uio_resid == 0" or an error
1517 * termination. If a signal/timeout occurs, return
1518 * with a short count but without error.
1519 * Keep sockbuf locked against other readers.
1521 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1522 !sosendallatonce(so) && nextrecord == NULL) {
1523 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1524 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1527 * Notify the protocol that some data has been
1528 * drained before blocking.
1530 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb != NULL) {
1531 SOCKBUF_UNLOCK(&so->so_rcv);
1532 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1533 SOCKBUF_LOCK(&so->so_rcv);
1535 SBLASTRECORDCHK(&so->so_rcv);
1536 SBLASTMBUFCHK(&so->so_rcv);
1537 error = sbwait(&so->so_rcv);
1540 m = so->so_rcv.sb_mb;
1542 nextrecord = m->m_nextpkt;
1546 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1547 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1549 if ((flags & MSG_PEEK) == 0)
1550 (void) sbdroprecord_locked(&so->so_rcv);
1552 if ((flags & MSG_PEEK) == 0) {
1555 * First part is an inline SB_EMPTY_FIXUP(). Second
1556 * part makes sure sb_lastrecord is up-to-date if
1557 * there is still data in the socket buffer.
1559 so->so_rcv.sb_mb = nextrecord;
1560 if (so->so_rcv.sb_mb == NULL) {
1561 so->so_rcv.sb_mbtail = NULL;
1562 so->so_rcv.sb_lastrecord = NULL;
1563 } else if (nextrecord->m_nextpkt == NULL)
1564 so->so_rcv.sb_lastrecord = nextrecord;
1566 SBLASTRECORDCHK(&so->so_rcv);
1567 SBLASTMBUFCHK(&so->so_rcv);
1569 * If soreceive() is being done from the socket callback, then
1570 * don't need to generate ACK to peer to update window, since
1571 * ACK will be generated on return to TCP.
1573 if (!(flags & MSG_SOCALLBCK) &&
1574 (pr->pr_flags & PR_WANTRCVD) && so->so_pcb) {
1575 SOCKBUF_UNLOCK(&so->so_rcv);
1576 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1577 SOCKBUF_LOCK(&so->so_rcv);
1580 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1581 if (orig_resid == uio->uio_resid && orig_resid &&
1582 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1583 sbunlock(&so->so_rcv);
1590 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1591 sbunlock(&so->so_rcv);
1593 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1594 SOCKBUF_UNLOCK(&so->so_rcv);
1603 struct protosw *pr = so->so_proto;
1605 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1611 return ((*pr->pr_usrreqs->pru_shutdown)(so));
1619 struct sockbuf *sb = &so->so_rcv;
1620 struct protosw *pr = so->so_proto;
1624 * XXXRW: This is quite ugly. Previously, this code made a copy of
1625 * the socket buffer, then zero'd the original to clear the buffer
1626 * fields. However, with mutexes in the socket buffer, this causes
1627 * problems. We only clear the zeroable bits of the original;
1628 * however, we have to initialize and destroy the mutex in the copy
1629 * so that dom_dispose() and sbrelease() can lock t as needed.
1632 sb->sb_flags |= SB_NOINTR;
1633 (void) sblock(sb, M_WAITOK);
1635 * socantrcvmore_locked() drops the socket buffer mutex so that it
1636 * can safely perform wakeups. Re-acquire the mutex before
1639 socantrcvmore_locked(so);
1643 * Invalidate/clear most of the sockbuf structure, but leave
1644 * selinfo and mutex data unchanged.
1646 bzero(&asb, offsetof(struct sockbuf, sb_startzero));
1647 bcopy(&sb->sb_startzero, &asb.sb_startzero,
1648 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1649 bzero(&sb->sb_startzero,
1650 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1653 SOCKBUF_LOCK_INIT(&asb, "so_rcv");
1654 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1655 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
1656 sbrelease(&asb, so);
1657 SOCKBUF_LOCK_DESTROY(&asb);
1661 * Perhaps this routine, and sooptcopyout(), below, ought to come in
1662 * an additional variant to handle the case where the option value needs
1663 * to be some kind of integer, but not a specific size.
1664 * In addition to their use here, these functions are also called by the
1665 * protocol-level pr_ctloutput() routines.
1668 sooptcopyin(sopt, buf, len, minlen)
1669 struct sockopt *sopt;
1677 * If the user gives us more than we wanted, we ignore it,
1678 * but if we don't get the minimum length the caller
1679 * wants, we return EINVAL. On success, sopt->sopt_valsize
1680 * is set to however much we actually retrieved.
1682 if ((valsize = sopt->sopt_valsize) < minlen)
1685 sopt->sopt_valsize = valsize = len;
1687 if (sopt->sopt_td != NULL)
1688 return (copyin(sopt->sopt_val, buf, valsize));
1690 bcopy(sopt->sopt_val, buf, valsize);
1695 * Kernel version of setsockopt(2)/
1696 * XXX: optlen is size_t, not socklen_t
1699 so_setsockopt(struct socket *so, int level, int optname, void *optval,
1702 struct sockopt sopt;
1704 sopt.sopt_level = level;
1705 sopt.sopt_name = optname;
1706 sopt.sopt_dir = SOPT_SET;
1707 sopt.sopt_val = optval;
1708 sopt.sopt_valsize = optlen;
1709 sopt.sopt_td = NULL;
1710 return (sosetopt(so, &sopt));
1716 struct sockopt *sopt;
1727 if (sopt->sopt_level != SOL_SOCKET) {
1728 if (so->so_proto && so->so_proto->pr_ctloutput)
1729 return ((*so->so_proto->pr_ctloutput)
1731 error = ENOPROTOOPT;
1733 switch (sopt->sopt_name) {
1735 case SO_ACCEPTFILTER:
1736 error = do_setopt_accept_filter(so, sopt);
1742 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1747 so->so_linger = l.l_linger;
1749 so->so_options |= SO_LINGER;
1751 so->so_options &= ~SO_LINGER;
1758 case SO_USELOOPBACK:
1766 error = sooptcopyin(sopt, &optval, sizeof optval,
1772 so->so_options |= sopt->sopt_name;
1774 so->so_options &= ~sopt->sopt_name;
1782 error = sooptcopyin(sopt, &optval, sizeof optval,
1788 * Values < 1 make no sense for any of these
1789 * options, so disallow them.
1796 switch (sopt->sopt_name) {
1799 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
1800 &so->so_snd : &so->so_rcv, (u_long)optval,
1801 so, curthread) == 0) {
1808 * Make sure the low-water is never greater than
1812 SOCKBUF_LOCK(&so->so_snd);
1813 so->so_snd.sb_lowat =
1814 (optval > so->so_snd.sb_hiwat) ?
1815 so->so_snd.sb_hiwat : optval;
1816 SOCKBUF_UNLOCK(&so->so_snd);
1819 SOCKBUF_LOCK(&so->so_rcv);
1820 so->so_rcv.sb_lowat =
1821 (optval > so->so_rcv.sb_hiwat) ?
1822 so->so_rcv.sb_hiwat : optval;
1823 SOCKBUF_UNLOCK(&so->so_rcv);
1831 if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
1832 struct timeval32 tv32;
1834 error = sooptcopyin(sopt, &tv32, sizeof tv32,
1836 CP(tv32, tv, tv_sec);
1837 CP(tv32, tv, tv_usec);
1840 error = sooptcopyin(sopt, &tv, sizeof tv,
1845 /* assert(hz > 0); */
1846 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
1847 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1851 /* assert(tick > 0); */
1852 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
1853 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
1854 if (val > INT_MAX) {
1858 if (val == 0 && tv.tv_usec != 0)
1861 switch (sopt->sopt_name) {
1863 so->so_snd.sb_timeo = val;
1866 so->so_rcv.sb_timeo = val;
1873 error = sooptcopyin(sopt, &extmac, sizeof extmac,
1877 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
1885 error = ENOPROTOOPT;
1888 if (error == 0 && so->so_proto != NULL &&
1889 so->so_proto->pr_ctloutput != NULL) {
1890 (void) ((*so->so_proto->pr_ctloutput)
1898 /* Helper routine for getsockopt */
1900 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1908 * Documented get behavior is that we always return a value,
1909 * possibly truncated to fit in the user's buffer.
1910 * Traditional behavior is that we always tell the user
1911 * precisely how much we copied, rather than something useful
1912 * like the total amount we had available for her.
1913 * Note that this interface is not idempotent; the entire answer must
1914 * generated ahead of time.
1916 valsize = min(len, sopt->sopt_valsize);
1917 sopt->sopt_valsize = valsize;
1918 if (sopt->sopt_val != NULL) {
1919 if (sopt->sopt_td != NULL)
1920 error = copyout(buf, sopt->sopt_val, valsize);
1922 bcopy(buf, sopt->sopt_val, valsize);
1930 struct sockopt *sopt;
1940 if (sopt->sopt_level != SOL_SOCKET) {
1941 if (so->so_proto && so->so_proto->pr_ctloutput) {
1942 return ((*so->so_proto->pr_ctloutput)
1945 return (ENOPROTOOPT);
1947 switch (sopt->sopt_name) {
1949 case SO_ACCEPTFILTER:
1950 error = do_getopt_accept_filter(so, sopt);
1955 l.l_onoff = so->so_options & SO_LINGER;
1956 l.l_linger = so->so_linger;
1958 error = sooptcopyout(sopt, &l, sizeof l);
1961 case SO_USELOOPBACK:
1973 optval = so->so_options & sopt->sopt_name;
1975 error = sooptcopyout(sopt, &optval, sizeof optval);
1979 optval = so->so_type;
1983 optval = so->so_error;
1988 optval = so->so_snd.sb_hiwat;
1992 optval = so->so_rcv.sb_hiwat;
1996 optval = so->so_snd.sb_lowat;
2000 optval = so->so_rcv.sb_lowat;
2005 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2006 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2008 tv.tv_sec = optval / hz;
2009 tv.tv_usec = (optval % hz) * tick;
2011 if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
2012 struct timeval32 tv32;
2014 CP(tv, tv32, tv_sec);
2015 CP(tv, tv32, tv_usec);
2016 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2019 error = sooptcopyout(sopt, &tv, sizeof tv);
2024 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2028 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2032 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2040 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2044 error = mac_getsockopt_peerlabel(
2045 sopt->sopt_td->td_ucred, so, &extmac);
2048 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2054 case SO_LISTENQLIMIT:
2055 optval = so->so_qlimit;
2059 optval = so->so_qlen;
2062 case SO_LISTENINCQLEN:
2063 optval = so->so_incqlen;
2067 error = ENOPROTOOPT;
2074 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2076 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2078 struct mbuf *m, *m_prev;
2079 int sopt_size = sopt->sopt_valsize;
2081 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
2084 if (sopt_size > MLEN) {
2085 MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT);
2086 if ((m->m_flags & M_EXT) == 0) {
2090 m->m_len = min(MCLBYTES, sopt_size);
2092 m->m_len = min(MLEN, sopt_size);
2094 sopt_size -= m->m_len;
2099 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
2104 if (sopt_size > MLEN) {
2105 MCLGET(m, sopt->sopt_td != NULL ? M_TRYWAIT :
2107 if ((m->m_flags & M_EXT) == 0) {
2112 m->m_len = min(MCLBYTES, sopt_size);
2114 m->m_len = min(MLEN, sopt_size);
2116 sopt_size -= m->m_len;
2123 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2125 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2127 struct mbuf *m0 = m;
2129 if (sopt->sopt_val == NULL)
2131 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2132 if (sopt->sopt_td != NULL) {
2135 error = copyin(sopt->sopt_val, mtod(m, char *),
2142 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2143 sopt->sopt_valsize -= m->m_len;
2144 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2147 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2148 panic("ip6_sooptmcopyin");
2152 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2154 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2156 struct mbuf *m0 = m;
2159 if (sopt->sopt_val == NULL)
2161 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2162 if (sopt->sopt_td != NULL) {
2165 error = copyout(mtod(m, char *), sopt->sopt_val,
2172 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2173 sopt->sopt_valsize -= m->m_len;
2174 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2175 valsize += m->m_len;
2179 /* enough soopt buffer should be given from user-land */
2183 sopt->sopt_valsize = valsize;
2191 if (so->so_sigio != NULL)
2192 pgsigio(&so->so_sigio, SIGURG, 0);
2193 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2197 sopoll(struct socket *so, int events, struct ucred *active_cred,
2202 SOCKBUF_LOCK(&so->so_snd);
2203 SOCKBUF_LOCK(&so->so_rcv);
2204 if (events & (POLLIN | POLLRDNORM))
2206 revents |= events & (POLLIN | POLLRDNORM);
2208 if (events & POLLINIGNEOF)
2209 if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
2210 !TAILQ_EMPTY(&so->so_comp) || so->so_error)
2211 revents |= POLLINIGNEOF;
2213 if (events & (POLLOUT | POLLWRNORM))
2214 if (sowriteable(so))
2215 revents |= events & (POLLOUT | POLLWRNORM);
2217 if (events & (POLLPRI | POLLRDBAND))
2218 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2219 revents |= events & (POLLPRI | POLLRDBAND);
2223 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
2225 selrecord(td, &so->so_rcv.sb_sel);
2226 so->so_rcv.sb_flags |= SB_SEL;
2229 if (events & (POLLOUT | POLLWRNORM)) {
2230 selrecord(td, &so->so_snd.sb_sel);
2231 so->so_snd.sb_flags |= SB_SEL;
2235 SOCKBUF_UNLOCK(&so->so_rcv);
2236 SOCKBUF_UNLOCK(&so->so_snd);
2241 soo_kqfilter(struct file *fp, struct knote *kn)
2243 struct socket *so = kn->kn_fp->f_data;
2246 switch (kn->kn_filter) {
2248 if (so->so_options & SO_ACCEPTCONN)
2249 kn->kn_fop = &solisten_filtops;
2251 kn->kn_fop = &soread_filtops;
2255 kn->kn_fop = &sowrite_filtops;
2263 knlist_add(&sb->sb_sel.si_note, kn, 1);
2264 sb->sb_flags |= SB_KNOTE;
2270 filt_sordetach(struct knote *kn)
2272 struct socket *so = kn->kn_fp->f_data;
2274 SOCKBUF_LOCK(&so->so_rcv);
2275 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
2276 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
2277 so->so_rcv.sb_flags &= ~SB_KNOTE;
2278 SOCKBUF_UNLOCK(&so->so_rcv);
2283 filt_soread(struct knote *kn, long hint)
2287 so = kn->kn_fp->f_data;
2288 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2290 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
2291 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2292 kn->kn_flags |= EV_EOF;
2293 kn->kn_fflags = so->so_error;
2295 } else if (so->so_error) /* temporary udp error */
2297 else if (kn->kn_sfflags & NOTE_LOWAT)
2298 return (kn->kn_data >= kn->kn_sdata);
2300 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
2304 filt_sowdetach(struct knote *kn)
2306 struct socket *so = kn->kn_fp->f_data;
2308 SOCKBUF_LOCK(&so->so_snd);
2309 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
2310 if (knlist_empty(&so->so_snd.sb_sel.si_note))
2311 so->so_snd.sb_flags &= ~SB_KNOTE;
2312 SOCKBUF_UNLOCK(&so->so_snd);
2317 filt_sowrite(struct knote *kn, long hint)
2321 so = kn->kn_fp->f_data;
2322 SOCKBUF_LOCK_ASSERT(&so->so_snd);
2323 kn->kn_data = sbspace(&so->so_snd);
2324 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2325 kn->kn_flags |= EV_EOF;
2326 kn->kn_fflags = so->so_error;
2328 } else if (so->so_error) /* temporary udp error */
2330 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2331 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2333 else if (kn->kn_sfflags & NOTE_LOWAT)
2334 return (kn->kn_data >= kn->kn_sdata);
2336 return (kn->kn_data >= so->so_snd.sb_lowat);
2341 filt_solisten(struct knote *kn, long hint)
2343 struct socket *so = kn->kn_fp->f_data;
2345 kn->kn_data = so->so_qlen;
2346 return (! TAILQ_EMPTY(&so->so_comp));
2350 socheckuid(struct socket *so, uid_t uid)
2355 if (so->so_cred->cr_uid != uid)
2361 somaxconn_sysctl(SYSCTL_HANDLER_ARGS)
2367 error = sysctl_handle_int(oidp, &val, sizeof(int), req);
2368 if (error || !req->newptr )
2371 if (val < 1 || val > USHRT_MAX)