2 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 1982, 1986, 1988, 1990, 1993
36 * The Regents of the University of California. All rights reserved.
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
67 * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.24 2003/11/11 17:18:18 silby Exp $
68 * $DragonFly: src/sys/kern/uipc_socket.c,v 1.55 2008/09/02 16:17:52 dillon Exp $
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/fcntl.h>
77 #include <sys/malloc.h>
79 #include <sys/domain.h>
80 #include <sys/file.h> /* for struct knote */
81 #include <sys/kernel.h>
82 #include <sys/malloc.h>
83 #include <sys/event.h>
86 #include <sys/protosw.h>
87 #include <sys/socket.h>
88 #include <sys/socketvar.h>
89 #include <sys/socketops.h>
90 #include <sys/resourcevar.h>
91 #include <sys/signalvar.h>
92 #include <sys/sysctl.h>
95 #include <vm/vm_zone.h>
98 #include <sys/thread2.h>
99 #include <sys/socketvar2.h>
101 #include <machine/limits.h>
104 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
107 static void filt_sordetach(struct knote *kn);
108 static int filt_soread(struct knote *kn, long hint);
109 static void filt_sowdetach(struct knote *kn);
110 static int filt_sowrite(struct knote *kn, long hint);
111 static int filt_solisten(struct knote *kn, long hint);
113 static struct filterops solisten_filtops =
114 { 1, NULL, filt_sordetach, filt_solisten };
115 static struct filterops soread_filtops =
116 { 1, NULL, filt_sordetach, filt_soread };
117 static struct filterops sowrite_filtops =
118 { 1, NULL, filt_sowdetach, filt_sowrite };
120 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
121 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
122 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
125 static int somaxconn = SOMAXCONN;
126 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
127 &somaxconn, 0, "Maximum pending socket connection queue size");
130 * Socket operation routines.
131 * These routines are called by the routines in
132 * sys_socket.c or from a system process, and
133 * implement the semantics of socket operations by
134 * switching out to the protocol specific routines.
138 * Get a socket structure, and initialize it.
139 * Note that it would probably be better to allocate socket
140 * and PCB at the same time, but I'm not convinced that all
141 * the protocols can be easily modified to do this.
149 waitmask = waitok ? M_WAITOK : M_NOWAIT;
150 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
152 /* XXX race condition for reentrant kernel */
153 TAILQ_INIT(&so->so_aiojobq);
154 TAILQ_INIT(&so->so_rcv.ssb_sel.si_mlist);
155 TAILQ_INIT(&so->so_snd.ssb_sel.si_mlist);
161 socreate(int dom, struct socket **aso, int type,
162 int proto, struct thread *td)
164 struct proc *p = td->td_proc;
167 struct pru_attach_info ai;
171 prp = pffindproto(dom, proto, type);
173 prp = pffindtype(dom, type);
175 if (prp == 0 || prp->pr_usrreqs->pru_attach == 0)
176 return (EPROTONOSUPPORT);
178 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
179 prp->pr_domain->dom_family != PF_LOCAL &&
180 prp->pr_domain->dom_family != PF_INET &&
181 prp->pr_domain->dom_family != PF_INET6 &&
182 prp->pr_domain->dom_family != PF_ROUTE) {
183 return (EPROTONOSUPPORT);
186 if (prp->pr_type != type)
188 so = soalloc(p != 0);
193 * Set a default port for protocol processing. No action will occur
194 * on the socket on this port until an inpcb is attached to it and
195 * is able to match incoming packets, or until the socket becomes
196 * available to userland.
198 so->so_port = cpu0_soport(so, NULL, NULL);
200 TAILQ_INIT(&so->so_incomp);
201 TAILQ_INIT(&so->so_comp);
203 so->so_cred = crhold(p->p_ucred);
205 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
206 ai.p_ucred = p->p_ucred;
207 ai.fd_rdir = p->p_fd->fd_rdir;
210 * Auto-sizing of socket buffers is managed by the protocols and
211 * the appropriate flags must be set in the pru_attach function.
213 error = so_pru_attach(so, proto, &ai);
215 so->so_state |= SS_NOFDREF;
225 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
230 error = so_pru_bind(so, nam, td);
236 sodealloc(struct socket *so)
238 if (so->so_rcv.ssb_hiwat)
239 (void)chgsbsize(so->so_cred->cr_uidinfo,
240 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
241 if (so->so_snd.ssb_hiwat)
242 (void)chgsbsize(so->so_cred->cr_uidinfo,
243 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
245 /* remove accept filter if present */
246 if (so->so_accf != NULL)
247 do_setopt_accept_filter(so, NULL);
254 solisten(struct socket *so, int backlog, struct thread *td)
258 short oldopt, oldqlimit;
262 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) {
268 oldopt = so->so_options;
269 oldqlimit = so->so_qlimit;
272 if (TAILQ_EMPTY(&so->so_comp))
273 so->so_options |= SO_ACCEPTCONN;
274 if (backlog < 0 || backlog > somaxconn)
276 so->so_qlimit = backlog;
277 /* SCTP needs to look at tweak both the inbound backlog parameter AND
278 * the so_options (UDP model both connect's and gets inbound
279 * connections .. implicitly).
281 error = so_pru_listen(so, td);
284 /* Restore the params */
285 so->so_options = oldopt;
286 so->so_qlimit = oldqlimit;
296 * Destroy a disconnected socket. This routine is a NOP if entities
297 * still have a reference on the socket:
299 * so_pcb - The protocol stack still has a reference
300 * SS_NOFDREF - There is no longer a file pointer reference
301 * SS_ABORTING - An abort netmsg is in-flight
304 sofree(struct socket *so)
306 struct socket *head = so->so_head;
308 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
310 if (so->so_state & SS_ABORTING)
313 if (so->so_state & SS_INCOMP) {
314 TAILQ_REMOVE(&head->so_incomp, so, so_list);
316 } else if (so->so_state & SS_COMP) {
318 * We must not decommission a socket that's
319 * on the accept(2) queue. If we do, then
320 * accept(2) may hang after select(2) indicated
321 * that the listening socket was ready.
325 panic("sofree: not queued");
327 so->so_state &= ~SS_INCOMP;
330 ssb_release(&so->so_snd, so);
336 * Close a socket on last file table reference removal.
337 * Initiate disconnect if connected.
338 * Free socket when disconnect complete.
341 soclose(struct socket *so, int fflag)
346 funsetown(so->so_sigio);
347 if (so->so_pcb == NULL)
349 if (so->so_state & SS_ISCONNECTED) {
350 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
351 error = sodisconnect(so);
355 if (so->so_options & SO_LINGER) {
356 if ((so->so_state & SS_ISDISCONNECTING) &&
359 while (so->so_state & SS_ISCONNECTED) {
360 error = tsleep((caddr_t)&so->so_timeo,
361 PCATCH, "soclos", so->so_linger * hz);
371 error2 = so_pru_detach(so);
376 if (so->so_options & SO_ACCEPTCONN) {
379 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
380 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
381 sp->so_state &= ~SS_INCOMP;
386 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
387 TAILQ_REMOVE(&so->so_comp, sp, so_list);
388 sp->so_state &= ~SS_COMP;
394 if (so->so_state & SS_NOFDREF)
395 panic("soclose: NOFDREF");
396 so->so_state |= SS_NOFDREF;
403 * Abort and destroy a socket. Only one abort can be in progress
404 * at any given moment.
407 soabort(struct socket *so)
409 if ((so->so_state & SS_ABORTING) == 0) {
410 so->so_state |= SS_ABORTING;
416 soaborta(struct socket *so)
418 if ((so->so_state & SS_ABORTING) == 0) {
419 so->so_state |= SS_ABORTING;
425 soabort_oncpu(struct socket *so)
427 if ((so->so_state & SS_ABORTING) == 0) {
428 so->so_state |= SS_ABORTING;
429 so_pru_abort_oncpu(so);
434 soaccept(struct socket *so, struct sockaddr **nam)
439 if ((so->so_state & SS_NOFDREF) == 0)
440 panic("soaccept: !NOFDREF");
441 so->so_state &= ~SS_NOFDREF;
442 error = so_pru_accept(so, nam);
448 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
452 if (so->so_options & SO_ACCEPTCONN)
456 * If protocol is connection-based, can only connect once.
457 * Otherwise, if connected, try to disconnect first.
458 * This allows user to disconnect by connecting to, e.g.,
461 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
462 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
463 (error = sodisconnect(so)))) {
467 * Prevent accumulated error from previous connection
471 error = so_pru_connect(so, nam, td);
478 soconnect2(struct socket *so1, struct socket *so2)
483 error = so_pru_connect2(so1, so2);
489 sodisconnect(struct socket *so)
494 if ((so->so_state & SS_ISCONNECTED) == 0) {
498 if (so->so_state & SS_ISDISCONNECTING) {
502 error = so_pru_disconnect(so);
508 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
511 * If send must go all at once and message is larger than
512 * send buffering, then hard error.
513 * Lock against other senders.
514 * If must go all at once and not enough room now, then
515 * inform user that this would block and do nothing.
516 * Otherwise, if nonblocking, send as much as possible.
517 * The data to be sent is described by "uio" if nonzero,
518 * otherwise by the mbuf chain "top" (which must be null
519 * if uio is not). Data provided in mbuf chain must be small
520 * enough to send all at once.
522 * Returns nonzero on error, timeout or signal; callers
523 * must check for short counts if EINTR/ERESTART are returned.
524 * Data and control buffers are freed on return.
527 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
528 struct mbuf *top, struct mbuf *control, int flags,
535 int clen = 0, error, dontroute, mlen;
536 int atomic = sosendallatonce(so) || top;
540 resid = uio->uio_resid;
542 resid = (size_t)top->m_pkthdr.len;
545 * WARNING! resid is unsigned, space and len are signed. space
546 * can wind up negative if the sockbuf is overcommitted.
548 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
549 * type sockets since that's an error.
551 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
557 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
558 (so->so_proto->pr_flags & PR_ATOMIC);
559 if (td->td_lwp != NULL)
560 td->td_lwp->lwp_ru.ru_msgsnd++;
562 clen = control->m_len;
563 #define gotoerr(errcode) { error = errcode; crit_exit(); goto release; }
566 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
572 if (so->so_state & SS_CANTSENDMORE)
575 error = so->so_error;
580 if ((so->so_state & SS_ISCONNECTED) == 0) {
582 * `sendto' and `sendmsg' is allowed on a connection-
583 * based socket if it supports implied connect.
584 * Return ENOTCONN if not connected and no address is
587 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
588 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
589 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
590 !(resid == 0 && clen != 0))
592 } else if (addr == 0)
593 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
594 ENOTCONN : EDESTADDRREQ);
596 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
597 clen > so->so_snd.ssb_hiwat) {
600 space = ssb_space(&so->so_snd);
603 if ((space < 0 || (size_t)space < resid + clen) && uio &&
604 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
605 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
606 gotoerr(EWOULDBLOCK);
607 ssb_unlock(&so->so_snd);
608 error = ssb_wait(&so->so_snd);
620 * Data is prepackaged in "top".
624 top->m_flags |= M_EOR;
628 m = m_getl((int)resid, MB_WAIT, MT_DATA,
629 top == NULL ? M_PKTHDR : 0, &mlen);
632 m->m_pkthdr.rcvif = NULL;
634 len = imin((int)szmin(mlen, resid), space);
635 if (resid < MINCLSIZE) {
637 * For datagram protocols, leave room
638 * for protocol headers in first mbuf.
640 if (atomic && top == 0 && len < mlen)
644 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
645 resid = uio->uio_resid;
648 top->m_pkthdr.len += len;
654 top->m_flags |= M_EOR;
657 } while (space > 0 && atomic);
659 so->so_options |= SO_DONTROUTE;
660 if (flags & MSG_OOB) {
661 pru_flags = PRUS_OOB;
662 } else if ((flags & MSG_EOF) &&
663 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
666 * If the user set MSG_EOF, the protocol
667 * understands this flag and nothing left to
668 * send then use PRU_SEND_EOF instead of PRU_SEND.
670 pru_flags = PRUS_EOF;
671 } else if (resid > 0 && space > 0) {
672 /* If there is more to send, set PRUS_MORETOCOME */
673 pru_flags = PRUS_MORETOCOME;
679 * XXX all the SS_CANTSENDMORE checks previously
680 * done could be out of date. We could have recieved
681 * a reset packet in an interrupt or maybe we slept
682 * while doing page faults in uiomove() etc. We could
683 * probably recheck again inside the splnet() protection
684 * here, but there are probably other places that this
685 * also happens. We must rethink this.
687 error = so_pru_send(so, pru_flags, top, addr, control, td);
690 so->so_options &= ~SO_DONTROUTE;
697 } while (resid && space > 0);
701 ssb_unlock(&so->so_snd);
711 * A specialization of sosend() for UDP based on protocol-specific knowledge:
712 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
713 * sosendallatonce() returns true,
714 * the "atomic" variable is true,
715 * and sosendudp() blocks until space is available for the entire send.
716 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
717 * PR_IMPLOPCL flags set.
718 * UDP has no out-of-band data.
719 * UDP has no control data.
720 * UDP does not support MSG_EOR.
723 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
724 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
726 boolean_t dontroute; /* temporary SO_DONTROUTE setting */
731 if (td->td_lwp != NULL)
732 td->td_lwp->lwp_ru.ru_msgsnd++;
736 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
737 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
740 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
745 if (so->so_state & SS_CANTSENDMORE)
748 error = so->so_error;
753 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
754 gotoerr(EDESTADDRREQ);
755 if (resid > so->so_snd.ssb_hiwat)
757 space = ssb_space(&so->so_snd);
758 if (uio && (space < 0 || (size_t)space < resid)) {
759 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
760 gotoerr(EWOULDBLOCK);
761 ssb_unlock(&so->so_snd);
762 error = ssb_wait(&so->so_snd);
771 top = m_uiomove(uio);
776 dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
778 so->so_options |= SO_DONTROUTE;
780 error = so_pru_send(so, 0, top, addr, NULL, td);
781 top = NULL; /* sent or freed in lower layer */
784 so->so_options &= ~SO_DONTROUTE;
787 ssb_unlock(&so->so_snd);
795 * Implement receive operations on a socket.
796 * We depend on the way that records are added to the signalsockbuf
797 * by sbappend*. In particular, each record (mbufs linked through m_next)
798 * must begin with an address if the protocol so specifies,
799 * followed by an optional mbuf or mbufs containing ancillary data,
800 * and then zero or more mbufs of data.
801 * In order to avoid blocking network interrupts for the entire time here,
802 * we exit the critical section while doing the actual copy to user space.
803 * Although the signalsockbuf is locked, new data may still be appended,
804 * and thus we must maintain consistency of the signalsockbuf during that time.
806 * The caller may receive the data as a single mbuf chain by supplying
807 * an mbuf **mp0 for use in returning the chain. The uio is then used
808 * only for the count in uio_resid.
811 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
812 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
815 struct mbuf *free_chain = NULL;
816 int flags, len, error, offset;
817 struct protosw *pr = so->so_proto;
819 size_t resid, orig_resid;
822 resid = uio->uio_resid;
824 resid = (size_t)(sio->sb_climit - sio->sb_cc);
832 flags = *flagsp &~ MSG_EOR;
835 if (flags & MSG_OOB) {
836 m = m_get(MB_WAIT, MT_DATA);
839 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
845 KKASSERT(resid >= (size_t)m->m_len);
846 resid -= (size_t)m->m_len;
847 } while (resid > 0 && m);
850 uio->uio_resid = resid;
851 error = uiomove(mtod(m, caddr_t),
852 (int)szmin(resid, m->m_len),
854 resid = uio->uio_resid;
856 } while (uio->uio_resid && error == 0 && m);
863 if ((so->so_state & SS_ISCONFIRMING) && resid)
868 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
872 m = so->so_rcv.ssb_mb;
874 * If we have less data than requested, block awaiting more
875 * (subject to any timeout) if:
876 * 1. the current count is less than the low water mark, or
877 * 2. MSG_WAITALL is set, and it is possible to do the entire
878 * receive operation at once if we block (resid <= hiwat).
879 * 3. MSG_DONTWAIT is not set
880 * If MSG_WAITALL is set but resid is larger than the receive buffer,
881 * we have to do the receive in sections, and thus risk returning
882 * a short count if a timeout or signal occurs after we start.
884 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
885 (size_t)so->so_rcv.ssb_cc < resid) &&
886 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
887 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
888 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
889 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
893 error = so->so_error;
894 if ((flags & MSG_PEEK) == 0)
898 if (so->so_state & SS_CANTRCVMORE) {
904 for (; m; m = m->m_next) {
905 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
906 m = so->so_rcv.ssb_mb;
910 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
911 (pr->pr_flags & PR_CONNREQUIRED)) {
917 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
921 ssb_unlock(&so->so_rcv);
922 error = ssb_wait(&so->so_rcv);
929 if (uio && uio->uio_td && uio->uio_td->td_proc)
930 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
933 * note: m should be == sb_mb here. Cache the next record while
934 * cleaning up. Note that calling m_free*() will break out critical
937 KKASSERT(m == so->so_rcv.ssb_mb);
940 * Skip any address mbufs prepending the record.
942 if (pr->pr_flags & PR_ADDR) {
943 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
946 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
947 if (flags & MSG_PEEK)
950 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
954 * Skip any control mbufs prepending the record.
957 if (pr->pr_flags & PR_ADDR_OPT) {
959 * For SCTP we may be getting a
960 * whole message OR a partial delivery.
962 if (m && m->m_type == MT_SONAME) {
965 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
966 if (flags & MSG_PEEK)
969 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
973 while (m && m->m_type == MT_CONTROL && error == 0) {
974 if (flags & MSG_PEEK) {
976 *controlp = m_copy(m, 0, m->m_len);
977 m = m->m_next; /* XXX race */
980 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
981 if (pr->pr_domain->dom_externalize &&
982 mtod(m, struct cmsghdr *)->cmsg_type ==
984 error = (*pr->pr_domain->dom_externalize)(m);
988 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
991 if (controlp && *controlp) {
993 controlp = &(*controlp)->m_next;
1002 if (type == MT_OOBDATA)
1007 * Copy to the UIO or mbuf return chain (*mp).
1011 while (m && resid > 0 && error == 0) {
1012 if (m->m_type == MT_OOBDATA) {
1013 if (type != MT_OOBDATA)
1015 } else if (type == MT_OOBDATA)
1018 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1020 so->so_state &= ~SS_RCVATMARK;
1021 len = (resid > INT_MAX) ? INT_MAX : resid;
1022 if (so->so_oobmark && len > so->so_oobmark - offset)
1023 len = so->so_oobmark - offset;
1024 if (len > m->m_len - moff)
1025 len = m->m_len - moff;
1028 * Copy out to the UIO or pass the mbufs back to the SIO.
1029 * The SIO is dealt with when we eat the mbuf, but deal
1030 * with the resid here either way.
1034 uio->uio_resid = resid;
1035 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1036 resid = uio->uio_resid;
1041 resid -= (size_t)len;
1045 * Eat the entire mbuf or just a piece of it
1047 if (len == m->m_len - moff) {
1048 if (m->m_flags & M_EOR)
1051 if (m->m_flags & M_NOTIFICATION)
1052 flags |= MSG_NOTIFICATION;
1054 if (flags & MSG_PEEK) {
1059 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1063 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1067 if (flags & MSG_PEEK) {
1071 n = m_copym(m, 0, len, MB_WAIT);
1077 so->so_rcv.ssb_cc -= len;
1080 if (so->so_oobmark) {
1081 if ((flags & MSG_PEEK) == 0) {
1082 so->so_oobmark -= len;
1083 if (so->so_oobmark == 0) {
1084 so->so_state |= SS_RCVATMARK;
1089 if (offset == so->so_oobmark)
1093 if (flags & MSG_EOR)
1096 * If the MSG_WAITALL flag is set (for non-atomic socket),
1097 * we must not quit until resid == 0 or an error
1098 * termination. If a signal/timeout occurs, return
1099 * with a short count but without error.
1100 * Keep signalsockbuf locked against other readers.
1102 while ((flags & MSG_WAITALL) && m == NULL &&
1103 resid > 0 && !sosendallatonce(so) &&
1104 so->so_rcv.ssb_mb == NULL) {
1105 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1108 * The window might have closed to zero, make
1109 * sure we send an ack now that we've drained
1110 * the buffer or we might end up blocking until
1111 * the idle takes over (5 seconds).
1113 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1114 so_pru_rcvd(so, flags);
1115 error = ssb_wait(&so->so_rcv);
1117 ssb_unlock(&so->so_rcv);
1121 m = so->so_rcv.ssb_mb;
1126 * If an atomic read was requested but unread data still remains
1127 * in the record, set MSG_TRUNC.
1129 if (m && pr->pr_flags & PR_ATOMIC)
1133 * Cleanup. If an atomic read was requested drop any unread data.
1135 if ((flags & MSG_PEEK) == 0) {
1136 if (m && (pr->pr_flags & PR_ATOMIC))
1137 sbdroprecord(&so->so_rcv.sb);
1138 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1139 so_pru_rcvd(so, flags);
1142 if (orig_resid == resid && orig_resid &&
1143 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1144 ssb_unlock(&so->so_rcv);
1152 ssb_unlock(&so->so_rcv);
1156 m_freem(free_chain);
1161 soshutdown(struct socket *so, int how)
1163 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1169 return (so_pru_shutdown(so));
1174 sorflush(struct socket *so)
1176 struct signalsockbuf *ssb = &so->so_rcv;
1177 struct protosw *pr = so->so_proto;
1178 struct signalsockbuf asb;
1180 ssb->ssb_flags |= SSB_NOINTR;
1181 (void) ssb_lock(ssb, M_WAITOK);
1187 bzero((caddr_t)ssb, sizeof (*ssb));
1188 if (asb.ssb_flags & SSB_KNOTE) {
1189 ssb->ssb_sel.si_note = asb.ssb_sel.si_note;
1190 ssb->ssb_flags = SSB_KNOTE;
1194 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1195 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1196 ssb_release(&asb, so);
1201 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1203 struct accept_filter_arg *afap = NULL;
1204 struct accept_filter *afp;
1205 struct so_accf *af = so->so_accf;
1208 /* do not set/remove accept filters on non listen sockets */
1209 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1214 /* removing the filter */
1217 if (af->so_accept_filter != NULL &&
1218 af->so_accept_filter->accf_destroy != NULL) {
1219 af->so_accept_filter->accf_destroy(so);
1221 if (af->so_accept_filter_str != NULL) {
1222 FREE(af->so_accept_filter_str, M_ACCF);
1227 so->so_options &= ~SO_ACCEPTFILTER;
1230 /* adding a filter */
1231 /* must remove previous filter first */
1236 /* don't put large objects on the kernel stack */
1237 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1238 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1239 afap->af_name[sizeof(afap->af_name)-1] = '\0';
1240 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1243 afp = accept_filt_get(afap->af_name);
1248 MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1249 if (afp->accf_create != NULL) {
1250 if (afap->af_name[0] != '\0') {
1251 int len = strlen(afap->af_name) + 1;
1253 MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1254 strcpy(af->so_accept_filter_str, afap->af_name);
1256 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1257 if (af->so_accept_filter_arg == NULL) {
1258 FREE(af->so_accept_filter_str, M_ACCF);
1265 af->so_accept_filter = afp;
1267 so->so_options |= SO_ACCEPTFILTER;
1276 * Perhaps this routine, and sooptcopyout(), below, ought to come in
1277 * an additional variant to handle the case where the option value needs
1278 * to be some kind of integer, but not a specific size.
1279 * In addition to their use here, these functions are also called by the
1280 * protocol-level pr_ctloutput() routines.
1283 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1285 return soopt_to_kbuf(sopt, buf, len, minlen);
1289 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1293 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1294 KKASSERT(kva_p(buf));
1297 * If the user gives us more than we wanted, we ignore it,
1298 * but if we don't get the minimum length the caller
1299 * wants, we return EINVAL. On success, sopt->sopt_valsize
1300 * is set to however much we actually retrieved.
1302 if ((valsize = sopt->sopt_valsize) < minlen)
1305 sopt->sopt_valsize = valsize = len;
1307 bcopy(sopt->sopt_val, buf, valsize);
1313 sosetopt(struct socket *so, struct sockopt *sopt)
1321 sopt->sopt_dir = SOPT_SET;
1322 if (sopt->sopt_level != SOL_SOCKET) {
1323 if (so->so_proto && so->so_proto->pr_ctloutput) {
1324 return (so_pru_ctloutput(so, sopt));
1326 error = ENOPROTOOPT;
1328 switch (sopt->sopt_name) {
1330 case SO_ACCEPTFILTER:
1331 error = do_setopt_accept_filter(so, sopt);
1337 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1341 so->so_linger = l.l_linger;
1343 so->so_options |= SO_LINGER;
1345 so->so_options &= ~SO_LINGER;
1351 case SO_USELOOPBACK:
1357 error = sooptcopyin(sopt, &optval, sizeof optval,
1362 so->so_options |= sopt->sopt_name;
1364 so->so_options &= ~sopt->sopt_name;
1371 error = sooptcopyin(sopt, &optval, sizeof optval,
1377 * Values < 1 make no sense for any of these
1378 * options, so disallow them.
1385 switch (sopt->sopt_name) {
1388 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1389 &so->so_snd : &so->so_rcv, (u_long)optval,
1391 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1395 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
1396 &so->so_rcv)->ssb_flags &= ~SSB_AUTOSIZE;
1400 * Make sure the low-water is never greater than
1404 so->so_snd.ssb_lowat =
1405 (optval > so->so_snd.ssb_hiwat) ?
1406 so->so_snd.ssb_hiwat : optval;
1407 so->so_snd.ssb_flags &= ~SSB_AUTOLOWAT;
1410 so->so_rcv.ssb_lowat =
1411 (optval > so->so_rcv.ssb_hiwat) ?
1412 so->so_rcv.ssb_hiwat : optval;
1413 so->so_rcv.ssb_flags &= ~SSB_AUTOLOWAT;
1420 error = sooptcopyin(sopt, &tv, sizeof tv,
1425 /* assert(hz > 0); */
1426 if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz ||
1427 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1431 /* assert(tick > 0); */
1432 /* assert(ULONG_MAX - SHRT_MAX >= 1000000); */
1433 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
1434 if (val > SHRT_MAX) {
1438 if (val == 0 && tv.tv_usec != 0)
1441 switch (sopt->sopt_name) {
1443 so->so_snd.ssb_timeo = val;
1446 so->so_rcv.ssb_timeo = val;
1451 error = ENOPROTOOPT;
1454 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1455 (void) so_pru_ctloutput(so, sopt);
1462 /* Helper routine for getsockopt */
1464 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1466 soopt_from_kbuf(sopt, buf, len);
1471 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1476 sopt->sopt_valsize = 0;
1480 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1481 KKASSERT(kva_p(buf));
1484 * Documented get behavior is that we always return a value,
1485 * possibly truncated to fit in the user's buffer.
1486 * Traditional behavior is that we always tell the user
1487 * precisely how much we copied, rather than something useful
1488 * like the total amount we had available for her.
1489 * Note that this interface is not idempotent; the entire answer must
1490 * generated ahead of time.
1492 valsize = szmin(len, sopt->sopt_valsize);
1493 sopt->sopt_valsize = valsize;
1494 if (sopt->sopt_val != 0) {
1495 bcopy(buf, sopt->sopt_val, valsize);
1500 sogetopt(struct socket *so, struct sockopt *sopt)
1506 struct accept_filter_arg *afap;
1510 sopt->sopt_dir = SOPT_GET;
1511 if (sopt->sopt_level != SOL_SOCKET) {
1512 if (so->so_proto && so->so_proto->pr_ctloutput) {
1513 return (so_pru_ctloutput(so, sopt));
1515 return (ENOPROTOOPT);
1517 switch (sopt->sopt_name) {
1519 case SO_ACCEPTFILTER:
1520 if ((so->so_options & SO_ACCEPTCONN) == 0)
1522 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1523 M_TEMP, M_WAITOK | M_ZERO);
1524 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1525 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1526 if (so->so_accf->so_accept_filter_str != NULL)
1527 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1529 error = sooptcopyout(sopt, afap, sizeof(*afap));
1535 l.l_onoff = so->so_options & SO_LINGER;
1536 l.l_linger = so->so_linger;
1537 error = sooptcopyout(sopt, &l, sizeof l);
1540 case SO_USELOOPBACK:
1549 optval = so->so_options & sopt->sopt_name;
1551 error = sooptcopyout(sopt, &optval, sizeof optval);
1555 optval = so->so_type;
1559 optval = so->so_error;
1564 optval = so->so_snd.ssb_hiwat;
1568 optval = so->so_rcv.ssb_hiwat;
1572 optval = so->so_snd.ssb_lowat;
1576 optval = so->so_rcv.ssb_lowat;
1581 optval = (sopt->sopt_name == SO_SNDTIMEO ?
1582 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1584 tv.tv_sec = optval / hz;
1585 tv.tv_usec = (optval % hz) * ustick;
1586 error = sooptcopyout(sopt, &tv, sizeof tv);
1590 error = ENOPROTOOPT;
1597 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1599 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1601 struct mbuf *m, *m_prev;
1602 int sopt_size = sopt->sopt_valsize, msize;
1604 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1608 m->m_len = min(msize, sopt_size);
1609 sopt_size -= m->m_len;
1613 while (sopt_size > 0) {
1614 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1615 MT_DATA, 0, &msize);
1620 m->m_len = min(msize, sopt_size);
1621 sopt_size -= m->m_len;
1628 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1630 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1632 soopt_to_mbuf(sopt, m);
1637 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
1642 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1644 if (sopt->sopt_val == NULL)
1646 val = sopt->sopt_val;
1647 valsize = sopt->sopt_valsize;
1648 while (m != NULL && valsize >= m->m_len) {
1649 bcopy(val, mtod(m, char *), m->m_len);
1650 valsize -= m->m_len;
1651 val = (caddr_t)val + m->m_len;
1654 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1655 panic("ip6_sooptmcopyin");
1658 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1660 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1662 return soopt_from_mbuf(sopt, m);
1666 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
1668 struct mbuf *m0 = m;
1673 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1675 if (sopt->sopt_val == NULL)
1677 val = sopt->sopt_val;
1678 maxsize = sopt->sopt_valsize;
1679 while (m != NULL && maxsize >= m->m_len) {
1680 bcopy(mtod(m, char *), val, m->m_len);
1681 maxsize -= m->m_len;
1682 val = (caddr_t)val + m->m_len;
1683 valsize += m->m_len;
1687 /* enough soopt buffer should be given from user-land */
1691 sopt->sopt_valsize = valsize;
1696 sohasoutofband(struct socket *so)
1698 if (so->so_sigio != NULL)
1699 pgsigio(so->so_sigio, SIGURG, 0);
1700 selwakeup(&so->so_rcv.ssb_sel);
1704 sopoll(struct socket *so, int events, struct ucred *cred, struct thread *td)
1710 if (events & (POLLIN | POLLRDNORM))
1712 revents |= events & (POLLIN | POLLRDNORM);
1714 if (events & POLLINIGNEOF)
1715 if (so->so_rcv.ssb_cc >= so->so_rcv.ssb_lowat ||
1716 !TAILQ_EMPTY(&so->so_comp) || so->so_error)
1717 revents |= POLLINIGNEOF;
1719 if (events & (POLLOUT | POLLWRNORM))
1720 if (sowriteable(so))
1721 revents |= events & (POLLOUT | POLLWRNORM);
1723 if (events & (POLLPRI | POLLRDBAND))
1724 if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
1725 revents |= events & (POLLPRI | POLLRDBAND);
1729 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
1731 selrecord(td, &so->so_rcv.ssb_sel);
1732 so->so_rcv.ssb_flags |= SSB_SEL;
1735 if (events & (POLLOUT | POLLWRNORM)) {
1736 selrecord(td, &so->so_snd.ssb_sel);
1737 so->so_snd.ssb_flags |= SSB_SEL;
1746 sokqfilter(struct file *fp, struct knote *kn)
1748 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1749 struct signalsockbuf *ssb;
1751 switch (kn->kn_filter) {
1753 if (so->so_options & SO_ACCEPTCONN)
1754 kn->kn_fop = &solisten_filtops;
1756 kn->kn_fop = &soread_filtops;
1760 kn->kn_fop = &sowrite_filtops;
1768 SLIST_INSERT_HEAD(&ssb->ssb_sel.si_note, kn, kn_selnext);
1769 ssb->ssb_flags |= SSB_KNOTE;
1775 filt_sordetach(struct knote *kn)
1777 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1780 SLIST_REMOVE(&so->so_rcv.ssb_sel.si_note, kn, knote, kn_selnext);
1781 if (SLIST_EMPTY(&so->so_rcv.ssb_sel.si_note))
1782 so->so_rcv.ssb_flags &= ~SSB_KNOTE;
1788 filt_soread(struct knote *kn, long hint)
1790 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1792 kn->kn_data = so->so_rcv.ssb_cc;
1793 if (so->so_state & SS_CANTRCVMORE) {
1794 kn->kn_flags |= EV_EOF;
1795 kn->kn_fflags = so->so_error;
1798 if (so->so_error) /* temporary udp error */
1800 if (kn->kn_sfflags & NOTE_LOWAT)
1801 return (kn->kn_data >= kn->kn_sdata);
1802 return (kn->kn_data >= so->so_rcv.ssb_lowat);
1806 filt_sowdetach(struct knote *kn)
1808 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1811 SLIST_REMOVE(&so->so_snd.ssb_sel.si_note, kn, knote, kn_selnext);
1812 if (SLIST_EMPTY(&so->so_snd.ssb_sel.si_note))
1813 so->so_snd.ssb_flags &= ~SSB_KNOTE;
1819 filt_sowrite(struct knote *kn, long hint)
1821 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1823 kn->kn_data = ssb_space(&so->so_snd);
1824 if (so->so_state & SS_CANTSENDMORE) {
1825 kn->kn_flags |= EV_EOF;
1826 kn->kn_fflags = so->so_error;
1829 if (so->so_error) /* temporary udp error */
1831 if (((so->so_state & SS_ISCONNECTED) == 0) &&
1832 (so->so_proto->pr_flags & PR_CONNREQUIRED))
1834 if (kn->kn_sfflags & NOTE_LOWAT)
1835 return (kn->kn_data >= kn->kn_sdata);
1836 return (kn->kn_data >= so->so_snd.ssb_lowat);
1841 filt_solisten(struct knote *kn, long hint)
1843 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1845 kn->kn_data = so->so_qlen;
1846 return (! TAILQ_EMPTY(&so->so_comp));