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;
200 error = so_pru_attach(so, proto, &ai);
202 so->so_state |= SS_NOFDREF;
211 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
216 error = so_pru_bind(so, nam, td);
222 sodealloc(struct socket *so)
224 if (so->so_rcv.ssb_hiwat)
225 (void)chgsbsize(so->so_cred->cr_uidinfo,
226 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
227 if (so->so_snd.ssb_hiwat)
228 (void)chgsbsize(so->so_cred->cr_uidinfo,
229 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
231 /* remove accept filter if present */
232 if (so->so_accf != NULL)
233 do_setopt_accept_filter(so, NULL);
240 solisten(struct socket *so, int backlog, struct thread *td)
244 short oldopt, oldqlimit;
248 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) {
254 oldopt = so->so_options;
255 oldqlimit = so->so_qlimit;
258 if (TAILQ_EMPTY(&so->so_comp))
259 so->so_options |= SO_ACCEPTCONN;
260 if (backlog < 0 || backlog > somaxconn)
262 so->so_qlimit = backlog;
263 /* SCTP needs to look at tweak both the inbound backlog parameter AND
264 * the so_options (UDP model both connect's and gets inbound
265 * connections .. implicitly).
267 error = so_pru_listen(so, td);
270 /* Restore the params */
271 so->so_options = oldopt;
272 so->so_qlimit = oldqlimit;
282 * Destroy a disconnected socket. This routine is a NOP if entities
283 * still have a reference on the socket:
285 * so_pcb - The protocol stack still has a reference
286 * SS_NOFDREF - There is no longer a file pointer reference
287 * SS_ABORTING - An abort netmsg is in-flight
290 sofree(struct socket *so)
292 struct socket *head = so->so_head;
294 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
296 if (so->so_state & SS_ABORTING)
299 if (so->so_state & SS_INCOMP) {
300 TAILQ_REMOVE(&head->so_incomp, so, so_list);
302 } else if (so->so_state & SS_COMP) {
304 * We must not decommission a socket that's
305 * on the accept(2) queue. If we do, then
306 * accept(2) may hang after select(2) indicated
307 * that the listening socket was ready.
311 panic("sofree: not queued");
313 so->so_state &= ~SS_INCOMP;
316 ssb_release(&so->so_snd, so);
322 * Close a socket on last file table reference removal.
323 * Initiate disconnect if connected.
324 * Free socket when disconnect complete.
327 soclose(struct socket *so, int fflag)
332 funsetown(so->so_sigio);
333 if (so->so_pcb == NULL)
335 if (so->so_state & SS_ISCONNECTED) {
336 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
337 error = sodisconnect(so);
341 if (so->so_options & SO_LINGER) {
342 if ((so->so_state & SS_ISDISCONNECTING) &&
345 while (so->so_state & SS_ISCONNECTED) {
346 error = tsleep((caddr_t)&so->so_timeo,
347 PCATCH, "soclos", so->so_linger * hz);
357 error2 = so_pru_detach(so);
362 if (so->so_options & SO_ACCEPTCONN) {
365 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
366 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
367 sp->so_state &= ~SS_INCOMP;
372 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
373 TAILQ_REMOVE(&so->so_comp, sp, so_list);
374 sp->so_state &= ~SS_COMP;
380 if (so->so_state & SS_NOFDREF)
381 panic("soclose: NOFDREF");
382 so->so_state |= SS_NOFDREF;
389 * Abort and destroy a socket. Only one abort can be in progress
390 * at any given moment.
393 soabort(struct socket *so)
395 if ((so->so_state & SS_ABORTING) == 0) {
396 so->so_state |= SS_ABORTING;
402 soaborta(struct socket *so)
404 if ((so->so_state & SS_ABORTING) == 0) {
405 so->so_state |= SS_ABORTING;
411 soaccept(struct socket *so, struct sockaddr **nam)
416 if ((so->so_state & SS_NOFDREF) == 0)
417 panic("soaccept: !NOFDREF");
418 so->so_state &= ~SS_NOFDREF;
419 error = so_pru_accept(so, nam);
425 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
429 if (so->so_options & SO_ACCEPTCONN)
433 * If protocol is connection-based, can only connect once.
434 * Otherwise, if connected, try to disconnect first.
435 * This allows user to disconnect by connecting to, e.g.,
438 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
439 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
440 (error = sodisconnect(so)))) {
444 * Prevent accumulated error from previous connection
448 error = so_pru_connect(so, nam, td);
455 soconnect2(struct socket *so1, struct socket *so2)
460 error = so_pru_connect2(so1, so2);
466 sodisconnect(struct socket *so)
471 if ((so->so_state & SS_ISCONNECTED) == 0) {
475 if (so->so_state & SS_ISDISCONNECTING) {
479 error = so_pru_disconnect(so);
485 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
488 * If send must go all at once and message is larger than
489 * send buffering, then hard error.
490 * Lock against other senders.
491 * If must go all at once and not enough room now, then
492 * inform user that this would block and do nothing.
493 * Otherwise, if nonblocking, send as much as possible.
494 * The data to be sent is described by "uio" if nonzero,
495 * otherwise by the mbuf chain "top" (which must be null
496 * if uio is not). Data provided in mbuf chain must be small
497 * enough to send all at once.
499 * Returns nonzero on error, timeout or signal; callers
500 * must check for short counts if EINTR/ERESTART are returned.
501 * Data and control buffers are freed on return.
504 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
505 struct mbuf *top, struct mbuf *control, int flags,
510 long space, len, resid;
511 int clen = 0, error, dontroute, mlen;
512 int atomic = sosendallatonce(so) || top;
516 resid = uio->uio_resid;
518 resid = top->m_pkthdr.len;
520 * In theory resid should be unsigned.
521 * However, space must be signed, as it might be less than 0
522 * if we over-committed, and we must use a signed comparison
523 * of space and resid. On the other hand, a negative resid
524 * causes us to loop sending 0-length segments to the protocol.
526 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
527 * type sockets since that's an error.
529 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
535 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
536 (so->so_proto->pr_flags & PR_ATOMIC);
537 if (td->td_lwp != NULL)
538 td->td_lwp->lwp_ru.ru_msgsnd++;
540 clen = control->m_len;
541 #define gotoerr(errcode) { error = errcode; crit_exit(); goto release; }
544 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
549 if (so->so_state & SS_CANTSENDMORE)
552 error = so->so_error;
557 if ((so->so_state & SS_ISCONNECTED) == 0) {
559 * `sendto' and `sendmsg' is allowed on a connection-
560 * based socket if it supports implied connect.
561 * Return ENOTCONN if not connected and no address is
564 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
565 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
566 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
567 !(resid == 0 && clen != 0))
569 } else if (addr == 0)
570 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
571 ENOTCONN : EDESTADDRREQ);
573 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
574 clen > so->so_snd.ssb_hiwat) {
577 space = ssb_space(&so->so_snd);
580 if (space < resid + clen && uio &&
581 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
582 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
583 gotoerr(EWOULDBLOCK);
584 ssb_unlock(&so->so_snd);
585 error = ssb_wait(&so->so_snd);
597 * Data is prepackaged in "top".
601 top->m_flags |= M_EOR;
603 m = m_getl(resid, MB_WAIT, MT_DATA,
604 top == NULL ? M_PKTHDR : 0, &mlen);
607 m->m_pkthdr.rcvif = (struct ifnet *)0;
609 len = min(min(mlen, resid), space);
610 if (resid < MINCLSIZE) {
612 * For datagram protocols, leave room
613 * for protocol headers in first mbuf.
615 if (atomic && top == 0 && len < mlen)
619 error = uiomove(mtod(m, caddr_t), (int)len, uio);
620 resid = uio->uio_resid;
623 top->m_pkthdr.len += len;
629 top->m_flags |= M_EOR;
632 } while (space > 0 && atomic);
634 so->so_options |= SO_DONTROUTE;
635 if (flags & MSG_OOB) {
636 pru_flags = PRUS_OOB;
637 } else if ((flags & MSG_EOF) &&
638 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
641 * If the user set MSG_EOF, the protocol
642 * understands this flag and nothing left to
643 * send then use PRU_SEND_EOF instead of PRU_SEND.
645 pru_flags = PRUS_EOF;
646 } else if (resid > 0 && space > 0) {
647 /* If there is more to send, set PRUS_MORETOCOME */
648 pru_flags = PRUS_MORETOCOME;
654 * XXX all the SS_CANTSENDMORE checks previously
655 * done could be out of date. We could have recieved
656 * a reset packet in an interrupt or maybe we slept
657 * while doing page faults in uiomove() etc. We could
658 * probably recheck again inside the splnet() protection
659 * here, but there are probably other places that this
660 * also happens. We must rethink this.
662 error = so_pru_send(so, pru_flags, top, addr, control, td);
665 so->so_options &= ~SO_DONTROUTE;
672 } while (resid && space > 0);
676 ssb_unlock(&so->so_snd);
686 * A specialization of sosend() for UDP based on protocol-specific knowledge:
687 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
688 * sosendallatonce() returns true,
689 * the "atomic" variable is true,
690 * and sosendudp() blocks until space is available for the entire send.
691 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
692 * PR_IMPLOPCL flags set.
693 * UDP has no out-of-band data.
694 * UDP has no control data.
695 * UDP does not support MSG_EOR.
698 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
699 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
702 boolean_t dontroute; /* temporary SO_DONTROUTE setting */
704 if (td->td_lwp != NULL)
705 td->td_lwp->lwp_ru.ru_msgsnd++;
709 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
710 resid = uio ? uio->uio_resid : top->m_pkthdr.len;
713 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
718 if (so->so_state & SS_CANTSENDMORE)
721 error = so->so_error;
726 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
727 gotoerr(EDESTADDRREQ);
728 if (resid > so->so_snd.ssb_hiwat)
730 if (uio && ssb_space(&so->so_snd) < resid) {
731 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
732 gotoerr(EWOULDBLOCK);
733 ssb_unlock(&so->so_snd);
734 error = ssb_wait(&so->so_snd);
743 top = m_uiomove(uio);
748 dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
750 so->so_options |= SO_DONTROUTE;
752 error = so_pru_send(so, 0, top, addr, NULL, td);
753 top = NULL; /* sent or freed in lower layer */
756 so->so_options &= ~SO_DONTROUTE;
759 ssb_unlock(&so->so_snd);
767 * Implement receive operations on a socket.
768 * We depend on the way that records are added to the signalsockbuf
769 * by sbappend*. In particular, each record (mbufs linked through m_next)
770 * must begin with an address if the protocol so specifies,
771 * followed by an optional mbuf or mbufs containing ancillary data,
772 * and then zero or more mbufs of data.
773 * In order to avoid blocking network interrupts for the entire time here,
774 * we exit the critical section while doing the actual copy to user space.
775 * Although the signalsockbuf is locked, new data may still be appended,
776 * and thus we must maintain consistency of the signalsockbuf during that time.
778 * The caller may receive the data as a single mbuf chain by supplying
779 * an mbuf **mp0 for use in returning the chain. The uio is then used
780 * only for the count in uio_resid.
783 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
784 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
787 struct mbuf *free_chain = NULL;
788 int flags, len, error, offset;
789 struct protosw *pr = so->so_proto;
791 int resid, orig_resid;
794 resid = uio->uio_resid;
796 resid = (int)(sio->sb_climit - sio->sb_cc);
804 flags = *flagsp &~ MSG_EOR;
807 if (flags & MSG_OOB) {
808 m = m_get(MB_WAIT, MT_DATA);
811 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
818 } while (resid > 0 && m);
821 uio->uio_resid = resid;
822 error = uiomove(mtod(m, caddr_t),
823 (int)min(resid, m->m_len), uio);
824 resid = uio->uio_resid;
826 } while (uio->uio_resid && error == 0 && m);
833 if (so->so_state & SS_ISCONFIRMING && resid)
838 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
842 m = so->so_rcv.ssb_mb;
844 * If we have less data than requested, block awaiting more
845 * (subject to any timeout) if:
846 * 1. the current count is less than the low water mark, or
847 * 2. MSG_WAITALL is set, and it is possible to do the entire
848 * receive operation at once if we block (resid <= hiwat).
849 * 3. MSG_DONTWAIT is not set
850 * If MSG_WAITALL is set but resid is larger than the receive buffer,
851 * we have to do the receive in sections, and thus risk returning
852 * a short count if a timeout or signal occurs after we start.
854 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
855 so->so_rcv.ssb_cc < resid) &&
856 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
857 ((flags & MSG_WAITALL) && resid <= so->so_rcv.ssb_hiwat)) &&
858 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
859 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
863 error = so->so_error;
864 if ((flags & MSG_PEEK) == 0)
868 if (so->so_state & SS_CANTRCVMORE) {
874 for (; m; m = m->m_next) {
875 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
876 m = so->so_rcv.ssb_mb;
880 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
881 (pr->pr_flags & PR_CONNREQUIRED)) {
887 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
891 ssb_unlock(&so->so_rcv);
892 error = ssb_wait(&so->so_rcv);
899 if (uio && uio->uio_td && uio->uio_td->td_proc)
900 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
903 * note: m should be == sb_mb here. Cache the next record while
904 * cleaning up. Note that calling m_free*() will break out critical
907 KKASSERT(m == so->so_rcv.ssb_mb);
910 * Skip any address mbufs prepending the record.
912 if (pr->pr_flags & PR_ADDR) {
913 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
916 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
917 if (flags & MSG_PEEK)
920 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
924 * Skip any control mbufs prepending the record.
927 if (pr->pr_flags & PR_ADDR_OPT) {
929 * For SCTP we may be getting a
930 * whole message OR a partial delivery.
932 if (m && m->m_type == MT_SONAME) {
935 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
936 if (flags & MSG_PEEK)
939 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
943 while (m && m->m_type == MT_CONTROL && error == 0) {
944 if (flags & MSG_PEEK) {
946 *controlp = m_copy(m, 0, m->m_len);
947 m = m->m_next; /* XXX race */
950 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
951 if (pr->pr_domain->dom_externalize &&
952 mtod(m, struct cmsghdr *)->cmsg_type ==
954 error = (*pr->pr_domain->dom_externalize)(m);
958 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
961 if (controlp && *controlp) {
963 controlp = &(*controlp)->m_next;
972 if (type == MT_OOBDATA)
977 * Copy to the UIO or mbuf return chain (*mp).
981 while (m && resid > 0 && error == 0) {
982 if (m->m_type == MT_OOBDATA) {
983 if (type != MT_OOBDATA)
985 } else if (type == MT_OOBDATA)
988 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
990 so->so_state &= ~SS_RCVATMARK;
992 if (so->so_oobmark && len > so->so_oobmark - offset)
993 len = so->so_oobmark - offset;
994 if (len > m->m_len - moff)
995 len = m->m_len - moff;
998 * Copy out to the UIO or pass the mbufs back to the SIO.
999 * The SIO is dealt with when we eat the mbuf, but deal
1000 * with the resid here either way.
1004 uio->uio_resid = resid;
1005 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1006 resid = uio->uio_resid;
1015 * Eat the entire mbuf or just a piece of it
1017 if (len == m->m_len - moff) {
1018 if (m->m_flags & M_EOR)
1021 if (m->m_flags & M_NOTIFICATION)
1022 flags |= MSG_NOTIFICATION;
1024 if (flags & MSG_PEEK) {
1029 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1033 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1037 if (flags & MSG_PEEK) {
1041 n = m_copym(m, 0, len, MB_WAIT);
1047 so->so_rcv.ssb_cc -= len;
1050 if (so->so_oobmark) {
1051 if ((flags & MSG_PEEK) == 0) {
1052 so->so_oobmark -= len;
1053 if (so->so_oobmark == 0) {
1054 so->so_state |= SS_RCVATMARK;
1059 if (offset == so->so_oobmark)
1063 if (flags & MSG_EOR)
1066 * If the MSG_WAITALL flag is set (for non-atomic socket),
1067 * we must not quit until resid == 0 or an error
1068 * termination. If a signal/timeout occurs, return
1069 * with a short count but without error.
1070 * Keep signalsockbuf locked against other readers.
1072 while ((flags & MSG_WAITALL) && m == NULL &&
1073 resid > 0 && !sosendallatonce(so) &&
1074 so->so_rcv.ssb_mb == NULL) {
1075 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1078 * The window might have closed to zero, make
1079 * sure we send an ack now that we've drained
1080 * the buffer or we might end up blocking until
1081 * the idle takes over (5 seconds).
1083 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1084 so_pru_rcvd(so, flags);
1085 error = ssb_wait(&so->so_rcv);
1087 ssb_unlock(&so->so_rcv);
1091 m = so->so_rcv.ssb_mb;
1096 * If an atomic read was requested but unread data still remains
1097 * in the record, set MSG_TRUNC.
1099 if (m && pr->pr_flags & PR_ATOMIC)
1103 * Cleanup. If an atomic read was requested drop any unread data.
1105 if ((flags & MSG_PEEK) == 0) {
1106 if (m && (pr->pr_flags & PR_ATOMIC))
1107 sbdroprecord(&so->so_rcv.sb);
1108 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1109 so_pru_rcvd(so, flags);
1112 if (orig_resid == resid && orig_resid &&
1113 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1114 ssb_unlock(&so->so_rcv);
1122 ssb_unlock(&so->so_rcv);
1126 m_freem(free_chain);
1131 soshutdown(struct socket *so, int how)
1133 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1139 return (so_pru_shutdown(so));
1144 sorflush(struct socket *so)
1146 struct signalsockbuf *ssb = &so->so_rcv;
1147 struct protosw *pr = so->so_proto;
1148 struct signalsockbuf asb;
1150 ssb->ssb_flags |= SSB_NOINTR;
1151 (void) ssb_lock(ssb, M_WAITOK);
1157 bzero((caddr_t)ssb, sizeof (*ssb));
1158 if (asb.ssb_flags & SSB_KNOTE) {
1159 ssb->ssb_sel.si_note = asb.ssb_sel.si_note;
1160 ssb->ssb_flags = SSB_KNOTE;
1164 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1165 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1166 ssb_release(&asb, so);
1171 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1173 struct accept_filter_arg *afap = NULL;
1174 struct accept_filter *afp;
1175 struct so_accf *af = so->so_accf;
1178 /* do not set/remove accept filters on non listen sockets */
1179 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1184 /* removing the filter */
1187 if (af->so_accept_filter != NULL &&
1188 af->so_accept_filter->accf_destroy != NULL) {
1189 af->so_accept_filter->accf_destroy(so);
1191 if (af->so_accept_filter_str != NULL) {
1192 FREE(af->so_accept_filter_str, M_ACCF);
1197 so->so_options &= ~SO_ACCEPTFILTER;
1200 /* adding a filter */
1201 /* must remove previous filter first */
1206 /* don't put large objects on the kernel stack */
1207 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1208 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1209 afap->af_name[sizeof(afap->af_name)-1] = '\0';
1210 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1213 afp = accept_filt_get(afap->af_name);
1218 MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1219 if (afp->accf_create != NULL) {
1220 if (afap->af_name[0] != '\0') {
1221 int len = strlen(afap->af_name) + 1;
1223 MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1224 strcpy(af->so_accept_filter_str, afap->af_name);
1226 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1227 if (af->so_accept_filter_arg == NULL) {
1228 FREE(af->so_accept_filter_str, M_ACCF);
1235 af->so_accept_filter = afp;
1237 so->so_options |= SO_ACCEPTFILTER;
1246 * Perhaps this routine, and sooptcopyout(), below, ought to come in
1247 * an additional variant to handle the case where the option value needs
1248 * to be some kind of integer, but not a specific size.
1249 * In addition to their use here, these functions are also called by the
1250 * protocol-level pr_ctloutput() routines.
1253 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1255 return soopt_to_kbuf(sopt, buf, len, minlen);
1259 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1263 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1264 KKASSERT(kva_p(buf));
1267 * If the user gives us more than we wanted, we ignore it,
1268 * but if we don't get the minimum length the caller
1269 * wants, we return EINVAL. On success, sopt->sopt_valsize
1270 * is set to however much we actually retrieved.
1272 if ((valsize = sopt->sopt_valsize) < minlen)
1275 sopt->sopt_valsize = valsize = len;
1277 bcopy(sopt->sopt_val, buf, valsize);
1283 sosetopt(struct socket *so, struct sockopt *sopt)
1291 sopt->sopt_dir = SOPT_SET;
1292 if (sopt->sopt_level != SOL_SOCKET) {
1293 if (so->so_proto && so->so_proto->pr_ctloutput) {
1294 return (so_pru_ctloutput(so, sopt));
1296 error = ENOPROTOOPT;
1298 switch (sopt->sopt_name) {
1300 case SO_ACCEPTFILTER:
1301 error = do_setopt_accept_filter(so, sopt);
1307 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1311 so->so_linger = l.l_linger;
1313 so->so_options |= SO_LINGER;
1315 so->so_options &= ~SO_LINGER;
1321 case SO_USELOOPBACK:
1327 error = sooptcopyin(sopt, &optval, sizeof optval,
1332 so->so_options |= sopt->sopt_name;
1334 so->so_options &= ~sopt->sopt_name;
1341 error = sooptcopyin(sopt, &optval, sizeof optval,
1347 * Values < 1 make no sense for any of these
1348 * options, so disallow them.
1355 switch (sopt->sopt_name) {
1358 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1359 &so->so_snd : &so->so_rcv, (u_long)optval,
1361 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1368 * Make sure the low-water is never greater than
1372 so->so_snd.ssb_lowat =
1373 (optval > so->so_snd.ssb_hiwat) ?
1374 so->so_snd.ssb_hiwat : optval;
1377 so->so_rcv.ssb_lowat =
1378 (optval > so->so_rcv.ssb_hiwat) ?
1379 so->so_rcv.ssb_hiwat : optval;
1386 error = sooptcopyin(sopt, &tv, sizeof tv,
1391 /* assert(hz > 0); */
1392 if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz ||
1393 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1397 /* assert(tick > 0); */
1398 /* assert(ULONG_MAX - SHRT_MAX >= 1000000); */
1399 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
1400 if (val > SHRT_MAX) {
1404 if (val == 0 && tv.tv_usec != 0)
1407 switch (sopt->sopt_name) {
1409 so->so_snd.ssb_timeo = val;
1412 so->so_rcv.ssb_timeo = val;
1417 error = ENOPROTOOPT;
1420 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1421 (void) so_pru_ctloutput(so, sopt);
1428 /* Helper routine for getsockopt */
1430 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1432 soopt_from_kbuf(sopt, buf, len);
1437 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1441 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1442 KKASSERT(kva_p(buf));
1445 * Documented get behavior is that we always return a value,
1446 * possibly truncated to fit in the user's buffer.
1447 * Traditional behavior is that we always tell the user
1448 * precisely how much we copied, rather than something useful
1449 * like the total amount we had available for her.
1450 * Note that this interface is not idempotent; the entire answer must
1451 * generated ahead of time.
1453 valsize = min(len, sopt->sopt_valsize);
1454 sopt->sopt_valsize = valsize;
1455 if (sopt->sopt_val != 0) {
1456 bcopy(buf, sopt->sopt_val, valsize);
1461 sogetopt(struct socket *so, struct sockopt *sopt)
1467 struct accept_filter_arg *afap;
1471 sopt->sopt_dir = SOPT_GET;
1472 if (sopt->sopt_level != SOL_SOCKET) {
1473 if (so->so_proto && so->so_proto->pr_ctloutput) {
1474 return (so_pru_ctloutput(so, sopt));
1476 return (ENOPROTOOPT);
1478 switch (sopt->sopt_name) {
1480 case SO_ACCEPTFILTER:
1481 if ((so->so_options & SO_ACCEPTCONN) == 0)
1483 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1484 M_TEMP, M_WAITOK | M_ZERO);
1485 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1486 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1487 if (so->so_accf->so_accept_filter_str != NULL)
1488 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1490 error = sooptcopyout(sopt, afap, sizeof(*afap));
1496 l.l_onoff = so->so_options & SO_LINGER;
1497 l.l_linger = so->so_linger;
1498 error = sooptcopyout(sopt, &l, sizeof l);
1501 case SO_USELOOPBACK:
1510 optval = so->so_options & sopt->sopt_name;
1512 error = sooptcopyout(sopt, &optval, sizeof optval);
1516 optval = so->so_type;
1520 optval = so->so_error;
1525 optval = so->so_snd.ssb_hiwat;
1529 optval = so->so_rcv.ssb_hiwat;
1533 optval = so->so_snd.ssb_lowat;
1537 optval = so->so_rcv.ssb_lowat;
1542 optval = (sopt->sopt_name == SO_SNDTIMEO ?
1543 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1545 tv.tv_sec = optval / hz;
1546 tv.tv_usec = (optval % hz) * tick;
1547 error = sooptcopyout(sopt, &tv, sizeof tv);
1551 error = ENOPROTOOPT;
1558 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1560 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1562 struct mbuf *m, *m_prev;
1563 int sopt_size = sopt->sopt_valsize, msize;
1565 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1569 m->m_len = min(msize, sopt_size);
1570 sopt_size -= m->m_len;
1574 while (sopt_size > 0) {
1575 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1576 MT_DATA, 0, &msize);
1581 m->m_len = min(msize, sopt_size);
1582 sopt_size -= m->m_len;
1589 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1591 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1593 soopt_to_mbuf(sopt, m);
1598 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
1603 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1605 if (sopt->sopt_val == NULL)
1607 val = sopt->sopt_val;
1608 valsize = sopt->sopt_valsize;
1609 while (m != NULL && valsize >= m->m_len) {
1610 bcopy(val, mtod(m, char *), m->m_len);
1611 valsize -= m->m_len;
1612 val = (caddr_t)val + m->m_len;
1615 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1616 panic("ip6_sooptmcopyin");
1619 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1621 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1623 return soopt_from_mbuf(sopt, m);
1627 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
1629 struct mbuf *m0 = m;
1634 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1636 if (sopt->sopt_val == NULL)
1638 val = sopt->sopt_val;
1639 maxsize = sopt->sopt_valsize;
1640 while (m != NULL && maxsize >= m->m_len) {
1641 bcopy(mtod(m, char *), val, m->m_len);
1642 maxsize -= m->m_len;
1643 val = (caddr_t)val + m->m_len;
1644 valsize += m->m_len;
1648 /* enough soopt buffer should be given from user-land */
1652 sopt->sopt_valsize = valsize;
1657 sohasoutofband(struct socket *so)
1659 if (so->so_sigio != NULL)
1660 pgsigio(so->so_sigio, SIGURG, 0);
1661 selwakeup(&so->so_rcv.ssb_sel);
1665 sopoll(struct socket *so, int events, struct ucred *cred, struct thread *td)
1671 if (events & (POLLIN | POLLRDNORM))
1673 revents |= events & (POLLIN | POLLRDNORM);
1675 if (events & POLLINIGNEOF)
1676 if (so->so_rcv.ssb_cc >= so->so_rcv.ssb_lowat ||
1677 !TAILQ_EMPTY(&so->so_comp) || so->so_error)
1678 revents |= POLLINIGNEOF;
1680 if (events & (POLLOUT | POLLWRNORM))
1681 if (sowriteable(so))
1682 revents |= events & (POLLOUT | POLLWRNORM);
1684 if (events & (POLLPRI | POLLRDBAND))
1685 if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
1686 revents |= events & (POLLPRI | POLLRDBAND);
1690 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
1692 selrecord(td, &so->so_rcv.ssb_sel);
1693 so->so_rcv.ssb_flags |= SSB_SEL;
1696 if (events & (POLLOUT | POLLWRNORM)) {
1697 selrecord(td, &so->so_snd.ssb_sel);
1698 so->so_snd.ssb_flags |= SSB_SEL;
1707 sokqfilter(struct file *fp, struct knote *kn)
1709 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1710 struct signalsockbuf *ssb;
1712 switch (kn->kn_filter) {
1714 if (so->so_options & SO_ACCEPTCONN)
1715 kn->kn_fop = &solisten_filtops;
1717 kn->kn_fop = &soread_filtops;
1721 kn->kn_fop = &sowrite_filtops;
1729 SLIST_INSERT_HEAD(&ssb->ssb_sel.si_note, kn, kn_selnext);
1730 ssb->ssb_flags |= SSB_KNOTE;
1736 filt_sordetach(struct knote *kn)
1738 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1741 SLIST_REMOVE(&so->so_rcv.ssb_sel.si_note, kn, knote, kn_selnext);
1742 if (SLIST_EMPTY(&so->so_rcv.ssb_sel.si_note))
1743 so->so_rcv.ssb_flags &= ~SSB_KNOTE;
1749 filt_soread(struct knote *kn, long hint)
1751 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1753 kn->kn_data = so->so_rcv.ssb_cc;
1754 if (so->so_state & SS_CANTRCVMORE) {
1755 kn->kn_flags |= EV_EOF;
1756 kn->kn_fflags = so->so_error;
1759 if (so->so_error) /* temporary udp error */
1761 if (kn->kn_sfflags & NOTE_LOWAT)
1762 return (kn->kn_data >= kn->kn_sdata);
1763 return (kn->kn_data >= so->so_rcv.ssb_lowat);
1767 filt_sowdetach(struct knote *kn)
1769 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1772 SLIST_REMOVE(&so->so_snd.ssb_sel.si_note, kn, knote, kn_selnext);
1773 if (SLIST_EMPTY(&so->so_snd.ssb_sel.si_note))
1774 so->so_snd.ssb_flags &= ~SSB_KNOTE;
1780 filt_sowrite(struct knote *kn, long hint)
1782 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1784 kn->kn_data = ssb_space(&so->so_snd);
1785 if (so->so_state & SS_CANTSENDMORE) {
1786 kn->kn_flags |= EV_EOF;
1787 kn->kn_fflags = so->so_error;
1790 if (so->so_error) /* temporary udp error */
1792 if (((so->so_state & SS_ISCONNECTED) == 0) &&
1793 (so->so_proto->pr_flags & PR_CONNREQUIRED))
1795 if (kn->kn_sfflags & NOTE_LOWAT)
1796 return (kn->kn_data >= kn->kn_sdata);
1797 return (kn->kn_data >= so->so_snd.ssb_lowat);
1802 filt_solisten(struct knote *kn, long hint)
1804 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1806 kn->kn_data = so->so_qlen;
1807 return (! TAILQ_EMPTY(&so->so_comp));