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) 2004 Jeffrey M. Hsu. All rights reserved.
37 * License terms: all terms for the DragonFly license above plus the following:
39 * 4. All advertising materials mentioning features or use of this software
40 * must display the following acknowledgement:
42 * This product includes software developed by Jeffrey M. Hsu
43 * for the DragonFly Project.
45 * This requirement may be waived with permission from Jeffrey Hsu.
46 * This requirement will sunset and may be removed on July 8 2005,
47 * after which the standard DragonFly license (as shown above) will
52 * Copyright (c) 1982, 1986, 1988, 1990, 1993
53 * The Regents of the University of California. All rights reserved.
55 * Redistribution and use in source and binary forms, with or without
56 * modification, are permitted provided that the following conditions
58 * 1. Redistributions of source code must retain the above copyright
59 * notice, this list of conditions and the following disclaimer.
60 * 2. Redistributions in binary form must reproduce the above copyright
61 * notice, this list of conditions and the following disclaimer in the
62 * documentation and/or other materials provided with the distribution.
63 * 3. All advertising materials mentioning features or use of this software
64 * must display the following acknowledgement:
65 * This product includes software developed by the University of
66 * California, Berkeley and its contributors.
67 * 4. Neither the name of the University nor the names of its contributors
68 * may be used to endorse or promote products derived from this software
69 * without specific prior written permission.
71 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
72 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
73 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
74 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
75 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
76 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
77 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
78 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
79 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
80 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
83 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
84 * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.24 2003/11/11 17:18:18 silby Exp $
85 * $DragonFly: src/sys/kern/uipc_socket.c,v 1.40 2006/12/23 23:47:54 swildner Exp $
91 #include <sys/param.h>
92 #include <sys/systm.h>
93 #include <sys/fcntl.h>
94 #include <sys/malloc.h>
96 #include <sys/domain.h>
97 #include <sys/file.h> /* for struct knote */
98 #include <sys/kernel.h>
99 #include <sys/malloc.h>
100 #include <sys/event.h>
101 #include <sys/poll.h>
102 #include <sys/proc.h>
103 #include <sys/protosw.h>
104 #include <sys/socket.h>
105 #include <sys/socketvar.h>
106 #include <sys/socketops.h>
107 #include <sys/resourcevar.h>
108 #include <sys/signalvar.h>
109 #include <sys/sysctl.h>
111 #include <sys/jail.h>
112 #include <vm/vm_zone.h>
114 #include <sys/thread2.h>
116 #include <machine/limits.h>
119 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
122 static void filt_sordetach(struct knote *kn);
123 static int filt_soread(struct knote *kn, long hint);
124 static void filt_sowdetach(struct knote *kn);
125 static int filt_sowrite(struct knote *kn, long hint);
126 static int filt_solisten(struct knote *kn, long hint);
128 static struct filterops solisten_filtops =
129 { 1, NULL, filt_sordetach, filt_solisten };
130 static struct filterops soread_filtops =
131 { 1, NULL, filt_sordetach, filt_soread };
132 static struct filterops sowrite_filtops =
133 { 1, NULL, filt_sowdetach, filt_sowrite };
135 struct vm_zone *socket_zone;
137 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
138 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
141 static int somaxconn = SOMAXCONN;
142 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
143 &somaxconn, 0, "Maximum pending socket connection queue size");
146 * Socket operation routines.
147 * These routines are called by the routines in
148 * sys_socket.c or from a system process, and
149 * implement the semantics of socket operations by
150 * switching out to the protocol specific routines.
154 * Get a socket structure from our zone, and initialize it.
155 * We don't implement `waitok' yet (see comments in uipc_domain.c).
156 * Note that it would probably be better to allocate socket
157 * and PCB at the same time, but I'm not convinced that all
158 * the protocols can be easily modified to do this.
165 so = zalloc(socket_zone);
167 /* XXX race condition for reentrant kernel */
168 bzero(so, sizeof *so);
169 TAILQ_INIT(&so->so_aiojobq);
170 TAILQ_INIT(&so->so_rcv.sb_sel.si_mlist);
171 TAILQ_INIT(&so->so_snd.sb_sel.si_mlist);
177 socreate(int dom, struct socket **aso, int type,
178 int proto, struct thread *td)
180 struct proc *p = td->td_proc;
183 struct pru_attach_info ai;
187 prp = pffindproto(dom, proto, type);
189 prp = pffindtype(dom, type);
191 if (prp == 0 || prp->pr_usrreqs->pru_attach == 0)
192 return (EPROTONOSUPPORT);
194 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
195 prp->pr_domain->dom_family != PF_LOCAL &&
196 prp->pr_domain->dom_family != PF_INET &&
197 prp->pr_domain->dom_family != PF_ROUTE) {
198 return (EPROTONOSUPPORT);
201 if (prp->pr_type != type)
203 so = soalloc(p != 0);
207 TAILQ_INIT(&so->so_incomp);
208 TAILQ_INIT(&so->so_comp);
210 so->so_cred = crhold(p->p_ucred);
212 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
213 ai.p_ucred = p->p_ucred;
214 ai.fd_rdir = p->p_fd->fd_rdir;
215 error = so_pru_attach(so, proto, &ai);
217 so->so_state |= SS_NOFDREF;
226 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
231 error = so_pru_bind(so, nam, td);
237 sodealloc(struct socket *so)
239 if (so->so_rcv.sb_hiwat)
240 (void)chgsbsize(so->so_cred->cr_uidinfo,
241 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
242 if (so->so_snd.sb_hiwat)
243 (void)chgsbsize(so->so_cred->cr_uidinfo,
244 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
246 /* remove accept filter if present */
247 if (so->so_accf != NULL)
248 do_setopt_accept_filter(so, NULL);
251 zfree(socket_zone, so);
255 solisten(struct socket *so, int backlog, struct thread *td)
259 short oldopt, oldqlimit;
263 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) {
269 oldopt = so->so_options;
270 oldqlimit = so->so_qlimit;
273 if (TAILQ_EMPTY(&so->so_comp))
274 so->so_options |= SO_ACCEPTCONN;
275 if (backlog < 0 || backlog > somaxconn)
277 so->so_qlimit = backlog;
278 /* SCTP needs to look at tweak both the inbound backlog parameter AND
279 * the so_options (UDP model both connect's and gets inbound
280 * connections .. implicitly).
282 error = so_pru_listen(so, td);
285 /* Restore the params */
286 so->so_options = oldopt;
287 so->so_qlimit = oldqlimit;
297 sofree(struct socket *so)
299 struct socket *head = so->so_head;
301 if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
304 if (so->so_state & SS_INCOMP) {
305 TAILQ_REMOVE(&head->so_incomp, so, so_list);
307 } else if (so->so_state & SS_COMP) {
309 * We must not decommission a socket that's
310 * on the accept(2) queue. If we do, then
311 * accept(2) may hang after select(2) indicated
312 * that the listening socket was ready.
316 panic("sofree: not queued");
318 so->so_state &= ~SS_INCOMP;
321 sbrelease(&so->so_snd, so);
327 * Close a socket on last file table reference removal.
328 * Initiate disconnect if connected.
329 * Free socket when disconnect complete.
332 soclose(struct socket *so, int fflag)
337 funsetown(so->so_sigio);
338 if (so->so_pcb == NULL)
340 if (so->so_state & SS_ISCONNECTED) {
341 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
342 error = sodisconnect(so);
346 if (so->so_options & SO_LINGER) {
347 if ((so->so_state & SS_ISDISCONNECTING) &&
350 while (so->so_state & SS_ISCONNECTED) {
351 error = tsleep((caddr_t)&so->so_timeo,
352 PCATCH, "soclos", so->so_linger * hz);
362 error2 = so_pru_detach(so);
367 if (so->so_options & SO_ACCEPTCONN) {
368 struct socket *sp, *sonext;
370 sp = TAILQ_FIRST(&so->so_incomp);
371 for (; sp != NULL; sp = sonext) {
372 sonext = TAILQ_NEXT(sp, so_list);
375 for (sp = TAILQ_FIRST(&so->so_comp); sp != NULL; sp = sonext) {
376 sonext = TAILQ_NEXT(sp, so_list);
377 /* Dequeue from so_comp since sofree() won't do it */
378 TAILQ_REMOVE(&so->so_comp, sp, so_list);
380 sp->so_state &= ~SS_COMP;
385 if (so->so_state & SS_NOFDREF)
386 panic("soclose: NOFDREF");
387 so->so_state |= SS_NOFDREF;
394 * Must be called from a critical section.
397 soabort(struct socket *so)
401 error = so_pru_abort(so);
410 soaccept(struct socket *so, struct sockaddr **nam)
415 if ((so->so_state & SS_NOFDREF) == 0)
416 panic("soaccept: !NOFDREF");
417 so->so_state &= ~SS_NOFDREF;
418 error = so_pru_accept(so, nam);
424 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
428 if (so->so_options & SO_ACCEPTCONN)
432 * If protocol is connection-based, can only connect once.
433 * Otherwise, if connected, try to disconnect first.
434 * This allows user to disconnect by connecting to, e.g.,
437 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
438 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
439 (error = sodisconnect(so)))) {
443 * Prevent accumulated error from previous connection
447 error = so_pru_connect(so, nam, td);
454 soconnect2(struct socket *so1, struct socket *so2)
459 error = so_pru_connect2(so1, so2);
465 sodisconnect(struct socket *so)
470 if ((so->so_state & SS_ISCONNECTED) == 0) {
474 if (so->so_state & SS_ISDISCONNECTING) {
478 error = so_pru_disconnect(so);
484 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
487 * If send must go all at once and message is larger than
488 * send buffering, then hard error.
489 * Lock against other senders.
490 * If must go all at once and not enough room now, then
491 * inform user that this would block and do nothing.
492 * Otherwise, if nonblocking, send as much as possible.
493 * The data to be sent is described by "uio" if nonzero,
494 * otherwise by the mbuf chain "top" (which must be null
495 * if uio is not). Data provided in mbuf chain must be small
496 * enough to send all at once.
498 * Returns nonzero on error, timeout or signal; callers
499 * must check for short counts if EINTR/ERESTART are returned.
500 * Data and control buffers are freed on return.
503 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
504 struct mbuf *top, struct mbuf *control, int flags,
509 long space, len, resid;
510 int clen = 0, error, dontroute, mlen;
511 int atomic = sosendallatonce(so) || top;
515 resid = uio->uio_resid;
517 resid = top->m_pkthdr.len;
519 * In theory resid should be unsigned.
520 * However, space must be signed, as it might be less than 0
521 * if we over-committed, and we must use a signed comparison
522 * of space and resid. On the other hand, a negative resid
523 * causes us to loop sending 0-length segments to the protocol.
525 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
526 * type sockets since that's an error.
528 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
534 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
535 (so->so_proto->pr_flags & PR_ATOMIC);
536 if (td->td_proc && td->td_proc->p_stats)
537 td->td_proc->p_stats->p_ru.ru_msgsnd++;
539 clen = control->m_len;
540 #define gotoerr(errcode) { error = errcode; crit_exit(); goto release; }
543 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
548 if (so->so_state & SS_CANTSENDMORE)
551 error = so->so_error;
556 if ((so->so_state & SS_ISCONNECTED) == 0) {
558 * `sendto' and `sendmsg' is allowed on a connection-
559 * based socket if it supports implied connect.
560 * Return ENOTCONN if not connected and no address is
563 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
564 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
565 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
566 !(resid == 0 && clen != 0))
568 } else if (addr == 0)
569 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
570 ENOTCONN : EDESTADDRREQ);
572 space = sbspace(&so->so_snd);
575 if ((atomic && resid > so->so_snd.sb_hiwat) ||
576 clen > so->so_snd.sb_hiwat)
578 if (space < resid + clen && uio &&
579 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
580 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
581 gotoerr(EWOULDBLOCK);
582 sbunlock(&so->so_snd);
583 error = sbwait(&so->so_snd);
595 * Data is prepackaged in "top".
599 top->m_flags |= M_EOR;
601 m = m_getl(resid, MB_WAIT, MT_DATA,
602 top == NULL ? M_PKTHDR : 0, &mlen);
605 m->m_pkthdr.rcvif = (struct ifnet *)0;
607 len = min(min(mlen, resid), space);
608 if (resid < MINCLSIZE) {
610 * For datagram protocols, leave room
611 * for protocol headers in first mbuf.
613 if (atomic && top == 0 && len < mlen)
617 error = uiomove(mtod(m, caddr_t), (int)len, uio);
618 resid = uio->uio_resid;
621 top->m_pkthdr.len += len;
627 top->m_flags |= M_EOR;
630 } while (space > 0 && atomic);
632 so->so_options |= SO_DONTROUTE;
633 if (flags & MSG_OOB) {
634 pru_flags = PRUS_OOB;
635 } else if ((flags & MSG_EOF) &&
636 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
639 * If the user set MSG_EOF, the protocol
640 * understands this flag and nothing left to
641 * send then use PRU_SEND_EOF instead of PRU_SEND.
643 pru_flags = PRUS_EOF;
644 } else if (resid > 0 && space > 0) {
645 /* If there is more to send, set PRUS_MORETOCOME */
646 pru_flags = PRUS_MORETOCOME;
652 * XXX all the SS_CANTSENDMORE checks previously
653 * done could be out of date. We could have recieved
654 * a reset packet in an interrupt or maybe we slept
655 * while doing page faults in uiomove() etc. We could
656 * probably recheck again inside the splnet() protection
657 * here, but there are probably other places that this
658 * also happens. We must rethink this.
660 error = so_pru_send(so, pru_flags, top, addr, control, td);
663 so->so_options &= ~SO_DONTROUTE;
670 } while (resid && space > 0);
674 sbunlock(&so->so_snd);
684 * A specialization of sosend() for UDP based on protocol-specific knowledge:
685 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
686 * sosendallatonce() returns true,
687 * the "atomic" variable is true,
688 * and sosendudp() blocks until space is available for the entire send.
689 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
690 * PR_IMPLOPCL flags set.
691 * UDP has no out-of-band data.
692 * UDP has no control data.
693 * UDP does not support MSG_EOR.
696 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
697 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
700 boolean_t dontroute; /* temporary SO_DONTROUTE setting */
702 if (td->td_proc && td->td_proc->p_stats)
703 td->td_proc->p_stats->p_ru.ru_msgsnd++;
707 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
708 resid = uio ? uio->uio_resid : top->m_pkthdr.len;
711 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
716 if (so->so_state & SS_CANTSENDMORE)
719 error = so->so_error;
724 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
725 gotoerr(EDESTADDRREQ);
726 if (resid > so->so_snd.sb_hiwat)
728 if (uio && sbspace(&so->so_snd) < resid) {
729 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
730 gotoerr(EWOULDBLOCK);
731 sbunlock(&so->so_snd);
732 error = sbwait(&so->so_snd);
741 top = m_uiomove(uio);
746 dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
748 so->so_options |= SO_DONTROUTE;
750 error = so_pru_send(so, 0, top, addr, NULL, td);
751 top = NULL; /* sent or freed in lower layer */
754 so->so_options &= ~SO_DONTROUTE;
757 sbunlock(&so->so_snd);
765 * Implement receive operations on a socket.
766 * We depend on the way that records are added to the sockbuf
767 * by sbappend*. In particular, each record (mbufs linked through m_next)
768 * must begin with an address if the protocol so specifies,
769 * followed by an optional mbuf or mbufs containing ancillary data,
770 * and then zero or more mbufs of data.
771 * In order to avoid blocking network interrupts for the entire time here,
772 * we exit the critical section while doing the actual copy to user space.
773 * Although the sockbuf is locked, new data may still be appended,
774 * and thus we must maintain consistency of the sockbuf during that time.
776 * The caller may receive the data as a single mbuf chain by supplying
777 * an mbuf **mp0 for use in returning the chain. The uio is then used
778 * only for the count in uio_resid.
781 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
782 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
784 struct mbuf *m, *n, **mp;
785 struct mbuf *free_chain = NULL;
786 int flags, len, error, offset;
787 struct protosw *pr = so->so_proto;
789 int orig_resid = uio->uio_resid;
797 flags = *flagsp &~ MSG_EOR;
800 if (flags & MSG_OOB) {
801 m = m_get(MB_WAIT, MT_DATA);
804 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
808 error = uiomove(mtod(m, caddr_t),
809 (int) min(uio->uio_resid, m->m_len), uio);
811 } while (uio->uio_resid && error == 0 && m);
819 if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
824 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
828 m = so->so_rcv.sb_mb;
830 * If we have less data than requested, block awaiting more
831 * (subject to any timeout) if:
832 * 1. the current count is less than the low water mark, or
833 * 2. MSG_WAITALL is set, and it is possible to do the entire
834 * receive operation at once if we block (resid <= hiwat).
835 * 3. MSG_DONTWAIT is not set
836 * If MSG_WAITALL is set but resid is larger than the receive buffer,
837 * we have to do the receive in sections, and thus risk returning
838 * a short count if a timeout or signal occurs after we start.
840 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
841 so->so_rcv.sb_cc < uio->uio_resid) &&
842 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
843 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
844 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
845 KASSERT(m != NULL || !so->so_rcv.sb_cc, ("receive 1"));
849 error = so->so_error;
850 if ((flags & MSG_PEEK) == 0)
854 if (so->so_state & SS_CANTRCVMORE) {
860 for (; m; m = m->m_next) {
861 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
862 m = so->so_rcv.sb_mb;
866 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
867 (pr->pr_flags & PR_CONNREQUIRED)) {
871 if (uio->uio_resid == 0)
873 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
877 sbunlock(&so->so_rcv);
878 error = sbwait(&so->so_rcv);
885 if (uio->uio_td && uio->uio_td->td_proc)
886 uio->uio_td->td_proc->p_stats->p_ru.ru_msgrcv++;
889 * note: m should be == sb_mb here. Cache the next record while
890 * cleaning up. Note that calling m_free*() will break out critical
893 KKASSERT(m == so->so_rcv.sb_mb);
896 * Skip any address mbufs prepending the record.
898 if (pr->pr_flags & PR_ADDR) {
899 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
902 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
903 if (flags & MSG_PEEK)
906 m = sbunlinkmbuf(&so->so_rcv, m, &free_chain);
910 * Skip any control mbufs prepending the record.
913 if (pr->pr_flags & PR_ADDR_OPT) {
915 * For SCTP we may be getting a
916 * whole message OR a partial delivery.
918 if (m && m->m_type == MT_SONAME) {
921 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
922 if (flags & MSG_PEEK)
925 m = sbunlinkmbuf(&so->so_rcv, m, &free_chain);
929 while (m && m->m_type == MT_CONTROL && error == 0) {
930 if (flags & MSG_PEEK) {
932 *controlp = m_copy(m, 0, m->m_len);
933 m = m->m_next; /* XXX race */
936 n = sbunlinkmbuf(&so->so_rcv, m, NULL);
937 if (pr->pr_domain->dom_externalize &&
938 mtod(m, struct cmsghdr *)->cmsg_type ==
940 error = (*pr->pr_domain->dom_externalize)(m);
944 m = sbunlinkmbuf(&so->so_rcv, m, &free_chain);
947 if (controlp && *controlp) {
949 controlp = &(*controlp)->m_next;
958 if (type == MT_OOBDATA)
963 * Copy to the UIO or mbuf return chain (*mp).
967 while (m && uio->uio_resid > 0 && error == 0) {
968 if (m->m_type == MT_OOBDATA) {
969 if (type != MT_OOBDATA)
971 } else if (type == MT_OOBDATA)
974 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
976 so->so_state &= ~SS_RCVATMARK;
977 len = uio->uio_resid;
978 if (so->so_oobmark && len > so->so_oobmark - offset)
979 len = so->so_oobmark - offset;
980 if (len > m->m_len - moff)
981 len = m->m_len - moff;
983 * If mp is set, just pass back the mbufs.
984 * Otherwise copy them out via the uio, then free.
985 * Sockbuf must be consistent here (points to current mbuf,
986 * it points to next record) when we drop priority;
987 * we must note any additions to the sockbuf when we
988 * block interrupts again.
992 error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio);
997 uio->uio_resid -= len;
1001 * Eat the entire mbuf or just a piece of it
1003 if (len == m->m_len - moff) {
1004 if (m->m_flags & M_EOR)
1007 if (m->m_flags & M_NOTIFICATION)
1008 flags |= MSG_NOTIFICATION;
1010 if (flags & MSG_PEEK) {
1015 n = sbunlinkmbuf(&so->so_rcv, m, NULL);
1020 m = sbunlinkmbuf(&so->so_rcv, m, &free_chain);
1024 if (flags & MSG_PEEK) {
1028 *mp = m_copym(m, 0, len, MB_WAIT);
1031 so->so_rcv.sb_cc -= len;
1034 if (so->so_oobmark) {
1035 if ((flags & MSG_PEEK) == 0) {
1036 so->so_oobmark -= len;
1037 if (so->so_oobmark == 0) {
1038 so->so_state |= SS_RCVATMARK;
1043 if (offset == so->so_oobmark)
1047 if (flags & MSG_EOR)
1050 * If the MSG_WAITALL flag is set (for non-atomic socket),
1051 * we must not quit until "uio->uio_resid == 0" or an error
1052 * termination. If a signal/timeout occurs, return
1053 * with a short count but without error.
1054 * Keep sockbuf locked against other readers.
1056 while (flags & MSG_WAITALL && m == NULL &&
1057 uio->uio_resid > 0 && !sosendallatonce(so) &&
1058 so->so_rcv.sb_mb == NULL) {
1059 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1062 * The window might have closed to zero, make
1063 * sure we send an ack now that we've drained
1064 * the buffer or we might end up blocking until
1065 * the idle takes over (5 seconds).
1067 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1068 so_pru_rcvd(so, flags);
1069 error = sbwait(&so->so_rcv);
1071 sbunlock(&so->so_rcv);
1075 m = so->so_rcv.sb_mb;
1080 * If an atomic read was requested but unread data still remains
1081 * in the record, set MSG_TRUNC.
1083 if (m && pr->pr_flags & PR_ATOMIC)
1087 * Cleanup. If an atomic read was requested drop any unread data.
1089 if ((flags & MSG_PEEK) == 0) {
1090 if (m && (pr->pr_flags & PR_ATOMIC))
1091 sbdroprecord(&so->so_rcv);
1092 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1093 so_pru_rcvd(so, flags);
1096 if (orig_resid == uio->uio_resid && orig_resid &&
1097 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1098 sbunlock(&so->so_rcv);
1106 sbunlock(&so->so_rcv);
1110 m_freem(free_chain);
1115 soshutdown(struct socket *so, int how)
1117 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1123 return (so_pru_shutdown(so));
1128 sorflush(struct socket *so)
1130 struct sockbuf *sb = &so->so_rcv;
1131 struct protosw *pr = so->so_proto;
1134 sb->sb_flags |= SB_NOINTR;
1135 (void) sblock(sb, M_WAITOK);
1141 bzero((caddr_t)sb, sizeof (*sb));
1142 if (asb.sb_flags & SB_KNOTE) {
1143 sb->sb_sel.si_note = asb.sb_sel.si_note;
1144 sb->sb_flags = SB_KNOTE;
1148 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1149 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
1150 sbrelease(&asb, so);
1155 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1157 struct accept_filter_arg *afap = NULL;
1158 struct accept_filter *afp;
1159 struct so_accf *af = so->so_accf;
1162 /* do not set/remove accept filters on non listen sockets */
1163 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1168 /* removing the filter */
1171 if (af->so_accept_filter != NULL &&
1172 af->so_accept_filter->accf_destroy != NULL) {
1173 af->so_accept_filter->accf_destroy(so);
1175 if (af->so_accept_filter_str != NULL) {
1176 FREE(af->so_accept_filter_str, M_ACCF);
1181 so->so_options &= ~SO_ACCEPTFILTER;
1184 /* adding a filter */
1185 /* must remove previous filter first */
1190 /* don't put large objects on the kernel stack */
1191 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1192 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1193 afap->af_name[sizeof(afap->af_name)-1] = '\0';
1194 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1197 afp = accept_filt_get(afap->af_name);
1202 MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK);
1203 bzero(af, sizeof(*af));
1204 if (afp->accf_create != NULL) {
1205 if (afap->af_name[0] != '\0') {
1206 int len = strlen(afap->af_name) + 1;
1208 MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1209 strcpy(af->so_accept_filter_str, afap->af_name);
1211 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1212 if (af->so_accept_filter_arg == NULL) {
1213 FREE(af->so_accept_filter_str, M_ACCF);
1220 af->so_accept_filter = afp;
1222 so->so_options |= SO_ACCEPTFILTER;
1231 * Perhaps this routine, and sooptcopyout(), below, ought to come in
1232 * an additional variant to handle the case where the option value needs
1233 * to be some kind of integer, but not a specific size.
1234 * In addition to their use here, these functions are also called by the
1235 * protocol-level pr_ctloutput() routines.
1238 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1243 * If the user gives us more than we wanted, we ignore it,
1244 * but if we don't get the minimum length the caller
1245 * wants, we return EINVAL. On success, sopt->sopt_valsize
1246 * is set to however much we actually retrieved.
1248 if ((valsize = sopt->sopt_valsize) < minlen)
1251 sopt->sopt_valsize = valsize = len;
1253 if (sopt->sopt_td != NULL)
1254 return (copyin(sopt->sopt_val, buf, valsize));
1256 bcopy(sopt->sopt_val, buf, valsize);
1261 sosetopt(struct socket *so, struct sockopt *sopt)
1269 sopt->sopt_dir = SOPT_SET;
1270 if (sopt->sopt_level != SOL_SOCKET) {
1271 if (so->so_proto && so->so_proto->pr_ctloutput) {
1272 return (so_pr_ctloutput(so, sopt));
1274 error = ENOPROTOOPT;
1276 switch (sopt->sopt_name) {
1278 case SO_ACCEPTFILTER:
1279 error = do_setopt_accept_filter(so, sopt);
1285 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1289 so->so_linger = l.l_linger;
1291 so->so_options |= SO_LINGER;
1293 so->so_options &= ~SO_LINGER;
1299 case SO_USELOOPBACK:
1305 error = sooptcopyin(sopt, &optval, sizeof optval,
1310 so->so_options |= sopt->sopt_name;
1312 so->so_options &= ~sopt->sopt_name;
1319 error = sooptcopyin(sopt, &optval, sizeof optval,
1325 * Values < 1 make no sense for any of these
1326 * options, so disallow them.
1333 switch (sopt->sopt_name) {
1336 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
1337 &so->so_snd : &so->so_rcv, (u_long)optval,
1339 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1346 * Make sure the low-water is never greater than
1350 so->so_snd.sb_lowat =
1351 (optval > so->so_snd.sb_hiwat) ?
1352 so->so_snd.sb_hiwat : optval;
1355 so->so_rcv.sb_lowat =
1356 (optval > so->so_rcv.sb_hiwat) ?
1357 so->so_rcv.sb_hiwat : optval;
1364 error = sooptcopyin(sopt, &tv, sizeof tv,
1369 /* assert(hz > 0); */
1370 if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz ||
1371 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1375 /* assert(tick > 0); */
1376 /* assert(ULONG_MAX - SHRT_MAX >= 1000000); */
1377 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
1378 if (val > SHRT_MAX) {
1382 if (val == 0 && tv.tv_usec != 0)
1385 switch (sopt->sopt_name) {
1387 so->so_snd.sb_timeo = val;
1390 so->so_rcv.sb_timeo = val;
1395 error = ENOPROTOOPT;
1398 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1399 (void) so_pr_ctloutput(so, sopt);
1406 /* Helper routine for getsockopt */
1408 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1416 * Documented get behavior is that we always return a value,
1417 * possibly truncated to fit in the user's buffer.
1418 * Traditional behavior is that we always tell the user
1419 * precisely how much we copied, rather than something useful
1420 * like the total amount we had available for her.
1421 * Note that this interface is not idempotent; the entire answer must
1422 * generated ahead of time.
1424 valsize = min(len, sopt->sopt_valsize);
1425 sopt->sopt_valsize = valsize;
1426 if (sopt->sopt_val != 0) {
1427 if (sopt->sopt_td != NULL)
1428 error = copyout(buf, sopt->sopt_val, valsize);
1430 bcopy(buf, sopt->sopt_val, valsize);
1436 sogetopt(struct socket *so, struct sockopt *sopt)
1442 struct accept_filter_arg *afap;
1446 sopt->sopt_dir = SOPT_GET;
1447 if (sopt->sopt_level != SOL_SOCKET) {
1448 if (so->so_proto && so->so_proto->pr_ctloutput) {
1449 return (so_pr_ctloutput(so, sopt));
1451 return (ENOPROTOOPT);
1453 switch (sopt->sopt_name) {
1455 case SO_ACCEPTFILTER:
1456 if ((so->so_options & SO_ACCEPTCONN) == 0)
1458 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1460 bzero(afap, sizeof(*afap));
1461 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1462 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1463 if (so->so_accf->so_accept_filter_str != NULL)
1464 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1466 error = sooptcopyout(sopt, afap, sizeof(*afap));
1472 l.l_onoff = so->so_options & SO_LINGER;
1473 l.l_linger = so->so_linger;
1474 error = sooptcopyout(sopt, &l, sizeof l);
1477 case SO_USELOOPBACK:
1486 optval = so->so_options & sopt->sopt_name;
1488 error = sooptcopyout(sopt, &optval, sizeof optval);
1492 optval = so->so_type;
1496 optval = so->so_error;
1501 optval = so->so_snd.sb_hiwat;
1505 optval = so->so_rcv.sb_hiwat;
1509 optval = so->so_snd.sb_lowat;
1513 optval = so->so_rcv.sb_lowat;
1518 optval = (sopt->sopt_name == SO_SNDTIMEO ?
1519 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1521 tv.tv_sec = optval / hz;
1522 tv.tv_usec = (optval % hz) * tick;
1523 error = sooptcopyout(sopt, &tv, sizeof tv);
1527 error = ENOPROTOOPT;
1534 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1536 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1538 struct mbuf *m, *m_prev;
1539 int sopt_size = sopt->sopt_valsize, msize;
1541 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1545 m->m_len = min(msize, sopt_size);
1546 sopt_size -= m->m_len;
1550 while (sopt_size > 0) {
1551 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1552 MT_DATA, 0, &msize);
1557 m->m_len = min(msize, sopt_size);
1558 sopt_size -= m->m_len;
1565 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1567 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1569 struct mbuf *m0 = m;
1571 if (sopt->sopt_val == NULL)
1573 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
1574 if (sopt->sopt_td != NULL) {
1577 error = copyin(sopt->sopt_val, mtod(m, char *),
1584 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
1585 sopt->sopt_valsize -= m->m_len;
1586 sopt->sopt_val = (caddr_t)sopt->sopt_val + m->m_len;
1589 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1590 panic("ip6_sooptmcopyin");
1594 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1596 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1598 struct mbuf *m0 = m;
1601 if (sopt->sopt_val == NULL)
1603 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
1604 if (sopt->sopt_td != NULL) {
1607 error = copyout(mtod(m, char *), sopt->sopt_val,
1614 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
1615 sopt->sopt_valsize -= m->m_len;
1616 sopt->sopt_val = (caddr_t)sopt->sopt_val + m->m_len;
1617 valsize += m->m_len;
1621 /* enough soopt buffer should be given from user-land */
1625 sopt->sopt_valsize = valsize;
1630 sohasoutofband(struct socket *so)
1632 if (so->so_sigio != NULL)
1633 pgsigio(so->so_sigio, SIGURG, 0);
1634 selwakeup(&so->so_rcv.sb_sel);
1638 sopoll(struct socket *so, int events, struct ucred *cred, struct thread *td)
1644 if (events & (POLLIN | POLLRDNORM))
1646 revents |= events & (POLLIN | POLLRDNORM);
1648 if (events & POLLINIGNEOF)
1649 if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
1650 !TAILQ_EMPTY(&so->so_comp) || so->so_error)
1651 revents |= POLLINIGNEOF;
1653 if (events & (POLLOUT | POLLWRNORM))
1654 if (sowriteable(so))
1655 revents |= events & (POLLOUT | POLLWRNORM);
1657 if (events & (POLLPRI | POLLRDBAND))
1658 if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
1659 revents |= events & (POLLPRI | POLLRDBAND);
1663 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
1665 selrecord(td, &so->so_rcv.sb_sel);
1666 so->so_rcv.sb_flags |= SB_SEL;
1669 if (events & (POLLOUT | POLLWRNORM)) {
1670 selrecord(td, &so->so_snd.sb_sel);
1671 so->so_snd.sb_flags |= SB_SEL;
1680 sokqfilter(struct file *fp, struct knote *kn)
1682 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1685 switch (kn->kn_filter) {
1687 if (so->so_options & SO_ACCEPTCONN)
1688 kn->kn_fop = &solisten_filtops;
1690 kn->kn_fop = &soread_filtops;
1694 kn->kn_fop = &sowrite_filtops;
1702 SLIST_INSERT_HEAD(&sb->sb_sel.si_note, kn, kn_selnext);
1703 sb->sb_flags |= SB_KNOTE;
1709 filt_sordetach(struct knote *kn)
1711 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1714 SLIST_REMOVE(&so->so_rcv.sb_sel.si_note, kn, knote, kn_selnext);
1715 if (SLIST_EMPTY(&so->so_rcv.sb_sel.si_note))
1716 so->so_rcv.sb_flags &= ~SB_KNOTE;
1722 filt_soread(struct knote *kn, long hint)
1724 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1726 kn->kn_data = so->so_rcv.sb_cc;
1727 if (so->so_state & SS_CANTRCVMORE) {
1728 kn->kn_flags |= EV_EOF;
1729 kn->kn_fflags = so->so_error;
1732 if (so->so_error) /* temporary udp error */
1734 if (kn->kn_sfflags & NOTE_LOWAT)
1735 return (kn->kn_data >= kn->kn_sdata);
1736 return (kn->kn_data >= so->so_rcv.sb_lowat);
1740 filt_sowdetach(struct knote *kn)
1742 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1745 SLIST_REMOVE(&so->so_snd.sb_sel.si_note, kn, knote, kn_selnext);
1746 if (SLIST_EMPTY(&so->so_snd.sb_sel.si_note))
1747 so->so_snd.sb_flags &= ~SB_KNOTE;
1753 filt_sowrite(struct knote *kn, long hint)
1755 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1757 kn->kn_data = sbspace(&so->so_snd);
1758 if (so->so_state & SS_CANTSENDMORE) {
1759 kn->kn_flags |= EV_EOF;
1760 kn->kn_fflags = so->so_error;
1763 if (so->so_error) /* temporary udp error */
1765 if (((so->so_state & SS_ISCONNECTED) == 0) &&
1766 (so->so_proto->pr_flags & PR_CONNREQUIRED))
1768 if (kn->kn_sfflags & NOTE_LOWAT)
1769 return (kn->kn_data >= kn->kn_sdata);
1770 return (kn->kn_data >= so->so_snd.sb_lowat);
1775 filt_solisten(struct knote *kn, long hint)
1777 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1779 kn->kn_data = so->so_qlen;
1780 return (! TAILQ_EMPTY(&so->so_comp));