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);
192 TAILQ_INIT(&so->so_incomp);
193 TAILQ_INIT(&so->so_comp);
195 so->so_cred = crhold(p->p_ucred);
197 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
198 ai.p_ucred = p->p_ucred;
199 ai.fd_rdir = p->p_fd->fd_rdir;
201 * Auto-sizing of socket buffers is managed by the protocols and
202 * the appropriate flags must be set in the pru_attach function.
204 error = so_pru_attach(so, proto, &ai);
206 so->so_state |= SS_NOFDREF;
215 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
220 error = so_pru_bind(so, nam, td);
226 sodealloc(struct socket *so)
228 if (so->so_rcv.ssb_hiwat)
229 (void)chgsbsize(so->so_cred->cr_uidinfo,
230 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
231 if (so->so_snd.ssb_hiwat)
232 (void)chgsbsize(so->so_cred->cr_uidinfo,
233 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
235 /* remove accept filter if present */
236 if (so->so_accf != NULL)
237 do_setopt_accept_filter(so, NULL);
244 solisten(struct socket *so, int backlog, struct thread *td)
248 short oldopt, oldqlimit;
252 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) {
258 oldopt = so->so_options;
259 oldqlimit = so->so_qlimit;
262 if (TAILQ_EMPTY(&so->so_comp))
263 so->so_options |= SO_ACCEPTCONN;
264 if (backlog < 0 || backlog > somaxconn)
266 so->so_qlimit = backlog;
267 /* SCTP needs to look at tweak both the inbound backlog parameter AND
268 * the so_options (UDP model both connect's and gets inbound
269 * connections .. implicitly).
271 error = so_pru_listen(so, td);
274 /* Restore the params */
275 so->so_options = oldopt;
276 so->so_qlimit = oldqlimit;
286 * Destroy a disconnected socket. This routine is a NOP if entities
287 * still have a reference on the socket:
289 * so_pcb - The protocol stack still has a reference
290 * SS_NOFDREF - There is no longer a file pointer reference
291 * SS_ABORTING - An abort netmsg is in-flight
294 sofree(struct socket *so)
296 struct socket *head = so->so_head;
298 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
300 if (so->so_state & SS_ABORTING)
303 if (so->so_state & SS_INCOMP) {
304 TAILQ_REMOVE(&head->so_incomp, so, so_list);
306 } else if (so->so_state & SS_COMP) {
308 * We must not decommission a socket that's
309 * on the accept(2) queue. If we do, then
310 * accept(2) may hang after select(2) indicated
311 * that the listening socket was ready.
315 panic("sofree: not queued");
317 so->so_state &= ~SS_INCOMP;
320 ssb_release(&so->so_snd, so);
326 * Close a socket on last file table reference removal.
327 * Initiate disconnect if connected.
328 * Free socket when disconnect complete.
331 soclose(struct socket *so, int fflag)
336 funsetown(so->so_sigio);
337 if (so->so_pcb == NULL)
339 if (so->so_state & SS_ISCONNECTED) {
340 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
341 error = sodisconnect(so);
345 if (so->so_options & SO_LINGER) {
346 if ((so->so_state & SS_ISDISCONNECTING) &&
349 while (so->so_state & SS_ISCONNECTED) {
350 error = tsleep((caddr_t)&so->so_timeo,
351 PCATCH, "soclos", so->so_linger * hz);
361 error2 = so_pru_detach(so);
366 if (so->so_options & SO_ACCEPTCONN) {
369 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
370 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
371 sp->so_state &= ~SS_INCOMP;
376 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
377 TAILQ_REMOVE(&so->so_comp, sp, so_list);
378 sp->so_state &= ~SS_COMP;
384 if (so->so_state & SS_NOFDREF)
385 panic("soclose: NOFDREF");
386 so->so_state |= SS_NOFDREF;
393 * Abort and destroy a socket. Only one abort can be in progress
394 * at any given moment.
397 soabort(struct socket *so)
399 if ((so->so_state & SS_ABORTING) == 0) {
400 so->so_state |= SS_ABORTING;
406 soaborta(struct socket *so)
408 if ((so->so_state & SS_ABORTING) == 0) {
409 so->so_state |= SS_ABORTING;
415 soabort_oncpu(struct socket *so)
417 if ((so->so_state & SS_ABORTING) == 0) {
418 so->so_state |= SS_ABORTING;
419 so_pru_abort_oncpu(so);
424 soaccept(struct socket *so, struct sockaddr **nam)
429 if ((so->so_state & SS_NOFDREF) == 0)
430 panic("soaccept: !NOFDREF");
431 so->so_state &= ~SS_NOFDREF;
432 error = so_pru_accept(so, nam);
438 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
442 if (so->so_options & SO_ACCEPTCONN)
446 * If protocol is connection-based, can only connect once.
447 * Otherwise, if connected, try to disconnect first.
448 * This allows user to disconnect by connecting to, e.g.,
451 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
452 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
453 (error = sodisconnect(so)))) {
457 * Prevent accumulated error from previous connection
461 error = so_pru_connect(so, nam, td);
468 soconnect2(struct socket *so1, struct socket *so2)
473 error = so_pru_connect2(so1, so2);
479 sodisconnect(struct socket *so)
484 if ((so->so_state & SS_ISCONNECTED) == 0) {
488 if (so->so_state & SS_ISDISCONNECTING) {
492 error = so_pru_disconnect(so);
498 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
501 * If send must go all at once and message is larger than
502 * send buffering, then hard error.
503 * Lock against other senders.
504 * If must go all at once and not enough room now, then
505 * inform user that this would block and do nothing.
506 * Otherwise, if nonblocking, send as much as possible.
507 * The data to be sent is described by "uio" if nonzero,
508 * otherwise by the mbuf chain "top" (which must be null
509 * if uio is not). Data provided in mbuf chain must be small
510 * enough to send all at once.
512 * Returns nonzero on error, timeout or signal; callers
513 * must check for short counts if EINTR/ERESTART are returned.
514 * Data and control buffers are freed on return.
517 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
518 struct mbuf *top, struct mbuf *control, int flags,
523 long space, len, resid;
524 int clen = 0, error, dontroute, mlen;
525 int atomic = sosendallatonce(so) || top;
529 resid = uio->uio_resid;
531 resid = top->m_pkthdr.len;
533 * In theory resid should be unsigned.
534 * However, space must be signed, as it might be less than 0
535 * if we over-committed, and we must use a signed comparison
536 * of space and resid. On the other hand, a negative resid
537 * causes us to loop sending 0-length segments to the protocol.
539 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
540 * type sockets since that's an error.
542 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
548 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
549 (so->so_proto->pr_flags & PR_ATOMIC);
550 if (td->td_lwp != NULL)
551 td->td_lwp->lwp_ru.ru_msgsnd++;
553 clen = control->m_len;
554 #define gotoerr(errcode) { error = errcode; crit_exit(); goto release; }
557 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
562 if (so->so_state & SS_CANTSENDMORE)
565 error = so->so_error;
570 if ((so->so_state & SS_ISCONNECTED) == 0) {
572 * `sendto' and `sendmsg' is allowed on a connection-
573 * based socket if it supports implied connect.
574 * Return ENOTCONN if not connected and no address is
577 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
578 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
579 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
580 !(resid == 0 && clen != 0))
582 } else if (addr == 0)
583 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
584 ENOTCONN : EDESTADDRREQ);
586 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
587 clen > so->so_snd.ssb_hiwat) {
590 space = ssb_space(&so->so_snd);
593 if (space < resid + clen && uio &&
594 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
595 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
596 gotoerr(EWOULDBLOCK);
597 ssb_unlock(&so->so_snd);
598 error = ssb_wait(&so->so_snd);
610 * Data is prepackaged in "top".
614 top->m_flags |= M_EOR;
616 m = m_getl(resid, MB_WAIT, MT_DATA,
617 top == NULL ? M_PKTHDR : 0, &mlen);
620 m->m_pkthdr.rcvif = NULL;
622 len = min(min(mlen, resid), space);
623 if (resid < MINCLSIZE) {
625 * For datagram protocols, leave room
626 * for protocol headers in first mbuf.
628 if (atomic && top == 0 && len < mlen)
632 error = uiomove(mtod(m, caddr_t), (int)len, uio);
633 resid = uio->uio_resid;
636 top->m_pkthdr.len += len;
642 top->m_flags |= M_EOR;
645 } while (space > 0 && atomic);
647 so->so_options |= SO_DONTROUTE;
648 if (flags & MSG_OOB) {
649 pru_flags = PRUS_OOB;
650 } else if ((flags & MSG_EOF) &&
651 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
654 * If the user set MSG_EOF, the protocol
655 * understands this flag and nothing left to
656 * send then use PRU_SEND_EOF instead of PRU_SEND.
658 pru_flags = PRUS_EOF;
659 } else if (resid > 0 && space > 0) {
660 /* If there is more to send, set PRUS_MORETOCOME */
661 pru_flags = PRUS_MORETOCOME;
667 * XXX all the SS_CANTSENDMORE checks previously
668 * done could be out of date. We could have recieved
669 * a reset packet in an interrupt or maybe we slept
670 * while doing page faults in uiomove() etc. We could
671 * probably recheck again inside the splnet() protection
672 * here, but there are probably other places that this
673 * also happens. We must rethink this.
675 error = so_pru_send(so, pru_flags, top, addr, control, td);
678 so->so_options &= ~SO_DONTROUTE;
685 } while (resid && space > 0);
689 ssb_unlock(&so->so_snd);
699 * A specialization of sosend() for UDP based on protocol-specific knowledge:
700 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
701 * sosendallatonce() returns true,
702 * the "atomic" variable is true,
703 * and sosendudp() blocks until space is available for the entire send.
704 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
705 * PR_IMPLOPCL flags set.
706 * UDP has no out-of-band data.
707 * UDP has no control data.
708 * UDP does not support MSG_EOR.
711 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
712 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
715 boolean_t dontroute; /* temporary SO_DONTROUTE setting */
717 if (td->td_lwp != NULL)
718 td->td_lwp->lwp_ru.ru_msgsnd++;
722 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
723 resid = uio ? uio->uio_resid : top->m_pkthdr.len;
726 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
731 if (so->so_state & SS_CANTSENDMORE)
734 error = so->so_error;
739 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
740 gotoerr(EDESTADDRREQ);
741 if (resid > so->so_snd.ssb_hiwat)
743 if (uio && ssb_space(&so->so_snd) < resid) {
744 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
745 gotoerr(EWOULDBLOCK);
746 ssb_unlock(&so->so_snd);
747 error = ssb_wait(&so->so_snd);
756 top = m_uiomove(uio);
761 dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
763 so->so_options |= SO_DONTROUTE;
765 error = so_pru_send(so, 0, top, addr, NULL, td);
766 top = NULL; /* sent or freed in lower layer */
769 so->so_options &= ~SO_DONTROUTE;
772 ssb_unlock(&so->so_snd);
780 * Implement receive operations on a socket.
781 * We depend on the way that records are added to the signalsockbuf
782 * by sbappend*. In particular, each record (mbufs linked through m_next)
783 * must begin with an address if the protocol so specifies,
784 * followed by an optional mbuf or mbufs containing ancillary data,
785 * and then zero or more mbufs of data.
786 * In order to avoid blocking network interrupts for the entire time here,
787 * we exit the critical section while doing the actual copy to user space.
788 * Although the signalsockbuf is locked, new data may still be appended,
789 * and thus we must maintain consistency of the signalsockbuf during that time.
791 * The caller may receive the data as a single mbuf chain by supplying
792 * an mbuf **mp0 for use in returning the chain. The uio is then used
793 * only for the count in uio_resid.
796 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
797 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
800 struct mbuf *free_chain = NULL;
801 int flags, len, error, offset;
802 struct protosw *pr = so->so_proto;
804 int resid, orig_resid;
807 resid = uio->uio_resid;
809 resid = (int)(sio->sb_climit - sio->sb_cc);
817 flags = *flagsp &~ MSG_EOR;
820 if (flags & MSG_OOB) {
821 m = m_get(MB_WAIT, MT_DATA);
824 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
831 } while (resid > 0 && m);
834 uio->uio_resid = resid;
835 error = uiomove(mtod(m, caddr_t),
836 (int)min(resid, m->m_len), uio);
837 resid = uio->uio_resid;
839 } while (uio->uio_resid && error == 0 && m);
846 if (so->so_state & SS_ISCONFIRMING && resid)
851 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
855 m = so->so_rcv.ssb_mb;
857 * If we have less data than requested, block awaiting more
858 * (subject to any timeout) if:
859 * 1. the current count is less than the low water mark, or
860 * 2. MSG_WAITALL is set, and it is possible to do the entire
861 * receive operation at once if we block (resid <= hiwat).
862 * 3. MSG_DONTWAIT is not set
863 * If MSG_WAITALL is set but resid is larger than the receive buffer,
864 * we have to do the receive in sections, and thus risk returning
865 * a short count if a timeout or signal occurs after we start.
867 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
868 so->so_rcv.ssb_cc < resid) &&
869 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
870 ((flags & MSG_WAITALL) && resid <= so->so_rcv.ssb_hiwat)) &&
871 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
872 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
876 error = so->so_error;
877 if ((flags & MSG_PEEK) == 0)
881 if (so->so_state & SS_CANTRCVMORE) {
887 for (; m; m = m->m_next) {
888 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
889 m = so->so_rcv.ssb_mb;
893 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
894 (pr->pr_flags & PR_CONNREQUIRED)) {
900 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
904 ssb_unlock(&so->so_rcv);
905 error = ssb_wait(&so->so_rcv);
912 if (uio && uio->uio_td && uio->uio_td->td_proc)
913 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
916 * note: m should be == sb_mb here. Cache the next record while
917 * cleaning up. Note that calling m_free*() will break out critical
920 KKASSERT(m == so->so_rcv.ssb_mb);
923 * Skip any address mbufs prepending the record.
925 if (pr->pr_flags & PR_ADDR) {
926 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
929 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
930 if (flags & MSG_PEEK)
933 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
937 * Skip any control mbufs prepending the record.
940 if (pr->pr_flags & PR_ADDR_OPT) {
942 * For SCTP we may be getting a
943 * whole message OR a partial delivery.
945 if (m && m->m_type == MT_SONAME) {
948 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
949 if (flags & MSG_PEEK)
952 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
956 while (m && m->m_type == MT_CONTROL && error == 0) {
957 if (flags & MSG_PEEK) {
959 *controlp = m_copy(m, 0, m->m_len);
960 m = m->m_next; /* XXX race */
963 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
964 if (pr->pr_domain->dom_externalize &&
965 mtod(m, struct cmsghdr *)->cmsg_type ==
967 error = (*pr->pr_domain->dom_externalize)(m);
971 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
974 if (controlp && *controlp) {
976 controlp = &(*controlp)->m_next;
985 if (type == MT_OOBDATA)
990 * Copy to the UIO or mbuf return chain (*mp).
994 while (m && resid > 0 && error == 0) {
995 if (m->m_type == MT_OOBDATA) {
996 if (type != MT_OOBDATA)
998 } else if (type == MT_OOBDATA)
1001 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1003 so->so_state &= ~SS_RCVATMARK;
1005 if (so->so_oobmark && len > so->so_oobmark - offset)
1006 len = so->so_oobmark - offset;
1007 if (len > m->m_len - moff)
1008 len = m->m_len - moff;
1011 * Copy out to the UIO or pass the mbufs back to the SIO.
1012 * The SIO is dealt with when we eat the mbuf, but deal
1013 * with the resid here either way.
1017 uio->uio_resid = resid;
1018 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1019 resid = uio->uio_resid;
1028 * Eat the entire mbuf or just a piece of it
1030 if (len == m->m_len - moff) {
1031 if (m->m_flags & M_EOR)
1034 if (m->m_flags & M_NOTIFICATION)
1035 flags |= MSG_NOTIFICATION;
1037 if (flags & MSG_PEEK) {
1042 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1046 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1050 if (flags & MSG_PEEK) {
1054 n = m_copym(m, 0, len, MB_WAIT);
1060 so->so_rcv.ssb_cc -= len;
1063 if (so->so_oobmark) {
1064 if ((flags & MSG_PEEK) == 0) {
1065 so->so_oobmark -= len;
1066 if (so->so_oobmark == 0) {
1067 so->so_state |= SS_RCVATMARK;
1072 if (offset == so->so_oobmark)
1076 if (flags & MSG_EOR)
1079 * If the MSG_WAITALL flag is set (for non-atomic socket),
1080 * we must not quit until resid == 0 or an error
1081 * termination. If a signal/timeout occurs, return
1082 * with a short count but without error.
1083 * Keep signalsockbuf locked against other readers.
1085 while ((flags & MSG_WAITALL) && m == NULL &&
1086 resid > 0 && !sosendallatonce(so) &&
1087 so->so_rcv.ssb_mb == NULL) {
1088 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1091 * The window might have closed to zero, make
1092 * sure we send an ack now that we've drained
1093 * the buffer or we might end up blocking until
1094 * the idle takes over (5 seconds).
1096 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1097 so_pru_rcvd(so, flags);
1098 error = ssb_wait(&so->so_rcv);
1100 ssb_unlock(&so->so_rcv);
1104 m = so->so_rcv.ssb_mb;
1109 * If an atomic read was requested but unread data still remains
1110 * in the record, set MSG_TRUNC.
1112 if (m && pr->pr_flags & PR_ATOMIC)
1116 * Cleanup. If an atomic read was requested drop any unread data.
1118 if ((flags & MSG_PEEK) == 0) {
1119 if (m && (pr->pr_flags & PR_ATOMIC))
1120 sbdroprecord(&so->so_rcv.sb);
1121 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1122 so_pru_rcvd(so, flags);
1125 if (orig_resid == resid && orig_resid &&
1126 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1127 ssb_unlock(&so->so_rcv);
1135 ssb_unlock(&so->so_rcv);
1139 m_freem(free_chain);
1144 soshutdown(struct socket *so, int how)
1146 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1152 return (so_pru_shutdown(so));
1157 sorflush(struct socket *so)
1159 struct signalsockbuf *ssb = &so->so_rcv;
1160 struct protosw *pr = so->so_proto;
1161 struct signalsockbuf asb;
1163 ssb->ssb_flags |= SSB_NOINTR;
1164 (void) ssb_lock(ssb, M_WAITOK);
1170 bzero((caddr_t)ssb, sizeof (*ssb));
1171 if (asb.ssb_flags & SSB_KNOTE) {
1172 ssb->ssb_sel.si_note = asb.ssb_sel.si_note;
1173 ssb->ssb_flags = SSB_KNOTE;
1177 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1178 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1179 ssb_release(&asb, so);
1184 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1186 struct accept_filter_arg *afap = NULL;
1187 struct accept_filter *afp;
1188 struct so_accf *af = so->so_accf;
1191 /* do not set/remove accept filters on non listen sockets */
1192 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1197 /* removing the filter */
1200 if (af->so_accept_filter != NULL &&
1201 af->so_accept_filter->accf_destroy != NULL) {
1202 af->so_accept_filter->accf_destroy(so);
1204 if (af->so_accept_filter_str != NULL) {
1205 FREE(af->so_accept_filter_str, M_ACCF);
1210 so->so_options &= ~SO_ACCEPTFILTER;
1213 /* adding a filter */
1214 /* must remove previous filter first */
1219 /* don't put large objects on the kernel stack */
1220 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1221 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1222 afap->af_name[sizeof(afap->af_name)-1] = '\0';
1223 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1226 afp = accept_filt_get(afap->af_name);
1231 MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1232 if (afp->accf_create != NULL) {
1233 if (afap->af_name[0] != '\0') {
1234 int len = strlen(afap->af_name) + 1;
1236 MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1237 strcpy(af->so_accept_filter_str, afap->af_name);
1239 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1240 if (af->so_accept_filter_arg == NULL) {
1241 FREE(af->so_accept_filter_str, M_ACCF);
1248 af->so_accept_filter = afp;
1250 so->so_options |= SO_ACCEPTFILTER;
1259 * Perhaps this routine, and sooptcopyout(), below, ought to come in
1260 * an additional variant to handle the case where the option value needs
1261 * to be some kind of integer, but not a specific size.
1262 * In addition to their use here, these functions are also called by the
1263 * protocol-level pr_ctloutput() routines.
1266 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1268 return soopt_to_kbuf(sopt, buf, len, minlen);
1272 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1276 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1277 KKASSERT(kva_p(buf));
1280 * If the user gives us more than we wanted, we ignore it,
1281 * but if we don't get the minimum length the caller
1282 * wants, we return EINVAL. On success, sopt->sopt_valsize
1283 * is set to however much we actually retrieved.
1285 if ((valsize = sopt->sopt_valsize) < minlen)
1288 sopt->sopt_valsize = valsize = len;
1290 bcopy(sopt->sopt_val, buf, valsize);
1296 sosetopt(struct socket *so, struct sockopt *sopt)
1304 sopt->sopt_dir = SOPT_SET;
1305 if (sopt->sopt_level != SOL_SOCKET) {
1306 if (so->so_proto && so->so_proto->pr_ctloutput) {
1307 return (so_pru_ctloutput(so, sopt));
1309 error = ENOPROTOOPT;
1311 switch (sopt->sopt_name) {
1313 case SO_ACCEPTFILTER:
1314 error = do_setopt_accept_filter(so, sopt);
1320 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1324 so->so_linger = l.l_linger;
1326 so->so_options |= SO_LINGER;
1328 so->so_options &= ~SO_LINGER;
1334 case SO_USELOOPBACK:
1340 error = sooptcopyin(sopt, &optval, sizeof optval,
1345 so->so_options |= sopt->sopt_name;
1347 so->so_options &= ~sopt->sopt_name;
1354 error = sooptcopyin(sopt, &optval, sizeof optval,
1360 * Values < 1 make no sense for any of these
1361 * options, so disallow them.
1368 switch (sopt->sopt_name) {
1371 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1372 &so->so_snd : &so->so_rcv, (u_long)optval,
1374 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1378 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
1379 &so->so_rcv)->ssb_flags &= ~SSB_AUTOSIZE;
1383 * Make sure the low-water is never greater than
1387 so->so_snd.ssb_lowat =
1388 (optval > so->so_snd.ssb_hiwat) ?
1389 so->so_snd.ssb_hiwat : optval;
1392 so->so_rcv.ssb_lowat =
1393 (optval > so->so_rcv.ssb_hiwat) ?
1394 so->so_rcv.ssb_hiwat : optval;
1401 error = sooptcopyin(sopt, &tv, sizeof tv,
1406 /* assert(hz > 0); */
1407 if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz ||
1408 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1412 /* assert(tick > 0); */
1413 /* assert(ULONG_MAX - SHRT_MAX >= 1000000); */
1414 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
1415 if (val > SHRT_MAX) {
1419 if (val == 0 && tv.tv_usec != 0)
1422 switch (sopt->sopt_name) {
1424 so->so_snd.ssb_timeo = val;
1427 so->so_rcv.ssb_timeo = val;
1432 error = ENOPROTOOPT;
1435 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1436 (void) so_pru_ctloutput(so, sopt);
1443 /* Helper routine for getsockopt */
1445 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1447 soopt_from_kbuf(sopt, buf, len);
1452 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1456 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1457 KKASSERT(kva_p(buf));
1460 * Documented get behavior is that we always return a value,
1461 * possibly truncated to fit in the user's buffer.
1462 * Traditional behavior is that we always tell the user
1463 * precisely how much we copied, rather than something useful
1464 * like the total amount we had available for her.
1465 * Note that this interface is not idempotent; the entire answer must
1466 * generated ahead of time.
1468 valsize = min(len, sopt->sopt_valsize);
1469 sopt->sopt_valsize = valsize;
1470 if (sopt->sopt_val != 0) {
1471 bcopy(buf, sopt->sopt_val, valsize);
1476 sogetopt(struct socket *so, struct sockopt *sopt)
1482 struct accept_filter_arg *afap;
1486 sopt->sopt_dir = SOPT_GET;
1487 if (sopt->sopt_level != SOL_SOCKET) {
1488 if (so->so_proto && so->so_proto->pr_ctloutput) {
1489 return (so_pru_ctloutput(so, sopt));
1491 return (ENOPROTOOPT);
1493 switch (sopt->sopt_name) {
1495 case SO_ACCEPTFILTER:
1496 if ((so->so_options & SO_ACCEPTCONN) == 0)
1498 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1499 M_TEMP, M_WAITOK | M_ZERO);
1500 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1501 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1502 if (so->so_accf->so_accept_filter_str != NULL)
1503 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1505 error = sooptcopyout(sopt, afap, sizeof(*afap));
1511 l.l_onoff = so->so_options & SO_LINGER;
1512 l.l_linger = so->so_linger;
1513 error = sooptcopyout(sopt, &l, sizeof l);
1516 case SO_USELOOPBACK:
1525 optval = so->so_options & sopt->sopt_name;
1527 error = sooptcopyout(sopt, &optval, sizeof optval);
1531 optval = so->so_type;
1535 optval = so->so_error;
1540 optval = so->so_snd.ssb_hiwat;
1544 optval = so->so_rcv.ssb_hiwat;
1548 optval = so->so_snd.ssb_lowat;
1552 optval = so->so_rcv.ssb_lowat;
1557 optval = (sopt->sopt_name == SO_SNDTIMEO ?
1558 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1560 tv.tv_sec = optval / hz;
1561 tv.tv_usec = (optval % hz) * tick;
1562 error = sooptcopyout(sopt, &tv, sizeof tv);
1566 error = ENOPROTOOPT;
1573 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1575 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1577 struct mbuf *m, *m_prev;
1578 int sopt_size = sopt->sopt_valsize, msize;
1580 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1584 m->m_len = min(msize, sopt_size);
1585 sopt_size -= m->m_len;
1589 while (sopt_size > 0) {
1590 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1591 MT_DATA, 0, &msize);
1596 m->m_len = min(msize, sopt_size);
1597 sopt_size -= m->m_len;
1604 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1606 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1608 soopt_to_mbuf(sopt, m);
1613 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
1618 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1620 if (sopt->sopt_val == NULL)
1622 val = sopt->sopt_val;
1623 valsize = sopt->sopt_valsize;
1624 while (m != NULL && valsize >= m->m_len) {
1625 bcopy(val, mtod(m, char *), m->m_len);
1626 valsize -= m->m_len;
1627 val = (caddr_t)val + m->m_len;
1630 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1631 panic("ip6_sooptmcopyin");
1634 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1636 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1638 return soopt_from_mbuf(sopt, m);
1642 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
1644 struct mbuf *m0 = m;
1649 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1651 if (sopt->sopt_val == NULL)
1653 val = sopt->sopt_val;
1654 maxsize = sopt->sopt_valsize;
1655 while (m != NULL && maxsize >= m->m_len) {
1656 bcopy(mtod(m, char *), val, m->m_len);
1657 maxsize -= m->m_len;
1658 val = (caddr_t)val + m->m_len;
1659 valsize += m->m_len;
1663 /* enough soopt buffer should be given from user-land */
1667 sopt->sopt_valsize = valsize;
1672 sohasoutofband(struct socket *so)
1674 if (so->so_sigio != NULL)
1675 pgsigio(so->so_sigio, SIGURG, 0);
1676 selwakeup(&so->so_rcv.ssb_sel);
1680 sopoll(struct socket *so, int events, struct ucred *cred, struct thread *td)
1686 if (events & (POLLIN | POLLRDNORM))
1688 revents |= events & (POLLIN | POLLRDNORM);
1690 if (events & POLLINIGNEOF)
1691 if (so->so_rcv.ssb_cc >= so->so_rcv.ssb_lowat ||
1692 !TAILQ_EMPTY(&so->so_comp) || so->so_error)
1693 revents |= POLLINIGNEOF;
1695 if (events & (POLLOUT | POLLWRNORM))
1696 if (sowriteable(so))
1697 revents |= events & (POLLOUT | POLLWRNORM);
1699 if (events & (POLLPRI | POLLRDBAND))
1700 if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
1701 revents |= events & (POLLPRI | POLLRDBAND);
1705 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
1707 selrecord(td, &so->so_rcv.ssb_sel);
1708 so->so_rcv.ssb_flags |= SSB_SEL;
1711 if (events & (POLLOUT | POLLWRNORM)) {
1712 selrecord(td, &so->so_snd.ssb_sel);
1713 so->so_snd.ssb_flags |= SSB_SEL;
1722 sokqfilter(struct file *fp, struct knote *kn)
1724 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1725 struct signalsockbuf *ssb;
1727 switch (kn->kn_filter) {
1729 if (so->so_options & SO_ACCEPTCONN)
1730 kn->kn_fop = &solisten_filtops;
1732 kn->kn_fop = &soread_filtops;
1736 kn->kn_fop = &sowrite_filtops;
1744 SLIST_INSERT_HEAD(&ssb->ssb_sel.si_note, kn, kn_selnext);
1745 ssb->ssb_flags |= SSB_KNOTE;
1751 filt_sordetach(struct knote *kn)
1753 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1756 SLIST_REMOVE(&so->so_rcv.ssb_sel.si_note, kn, knote, kn_selnext);
1757 if (SLIST_EMPTY(&so->so_rcv.ssb_sel.si_note))
1758 so->so_rcv.ssb_flags &= ~SSB_KNOTE;
1764 filt_soread(struct knote *kn, long hint)
1766 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1768 kn->kn_data = so->so_rcv.ssb_cc;
1769 if (so->so_state & SS_CANTRCVMORE) {
1770 kn->kn_flags |= EV_EOF;
1771 kn->kn_fflags = so->so_error;
1774 if (so->so_error) /* temporary udp error */
1776 if (kn->kn_sfflags & NOTE_LOWAT)
1777 return (kn->kn_data >= kn->kn_sdata);
1778 return (kn->kn_data >= so->so_rcv.ssb_lowat);
1782 filt_sowdetach(struct knote *kn)
1784 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1787 SLIST_REMOVE(&so->so_snd.ssb_sel.si_note, kn, knote, kn_selnext);
1788 if (SLIST_EMPTY(&so->so_snd.ssb_sel.si_note))
1789 so->so_snd.ssb_flags &= ~SSB_KNOTE;
1795 filt_sowrite(struct knote *kn, long hint)
1797 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1799 kn->kn_data = ssb_space(&so->so_snd);
1800 if (so->so_state & SS_CANTSENDMORE) {
1801 kn->kn_flags |= EV_EOF;
1802 kn->kn_fflags = so->so_error;
1805 if (so->so_error) /* temporary udp error */
1807 if (((so->so_state & SS_ISCONNECTED) == 0) &&
1808 (so->so_proto->pr_flags & PR_CONNREQUIRED))
1810 if (kn->kn_sfflags & NOTE_LOWAT)
1811 return (kn->kn_data >= kn->kn_sdata);
1812 return (kn->kn_data >= so->so_snd.ssb_lowat);
1817 filt_solisten(struct knote *kn, long hint)
1819 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1821 kn->kn_data = so->so_qlen;
1822 return (! TAILQ_EMPTY(&so->so_comp));