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>
85 #include <sys/protosw.h>
86 #include <sys/socket.h>
87 #include <sys/socketvar.h>
88 #include <sys/socketops.h>
89 #include <sys/resourcevar.h>
90 #include <sys/signalvar.h>
91 #include <sys/sysctl.h>
94 #include <vm/vm_zone.h>
97 #include <sys/thread2.h>
98 #include <sys/socketvar2.h>
100 #include <machine/limits.h>
103 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
106 static void filt_sordetach(struct knote *kn);
107 static int filt_soread(struct knote *kn, long hint);
108 static void filt_sowdetach(struct knote *kn);
109 static int filt_sowrite(struct knote *kn, long hint);
110 static int filt_solisten(struct knote *kn, long hint);
112 static struct filterops solisten_filtops =
113 { FILTEROP_ISFD, NULL, filt_sordetach, filt_solisten };
114 static struct filterops soread_filtops =
115 { FILTEROP_ISFD, NULL, filt_sordetach, filt_soread };
116 static struct filterops sowrite_filtops =
117 { FILTEROP_ISFD, NULL, filt_sowdetach, filt_sowrite };
118 static struct filterops soexcept_filtops =
119 { FILTEROP_ISFD, NULL, filt_sordetach, filt_soread };
121 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
122 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
123 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
126 static int somaxconn = SOMAXCONN;
127 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
128 &somaxconn, 0, "Maximum pending socket connection queue size");
131 * Socket operation routines.
132 * These routines are called by the routines in
133 * sys_socket.c or from a system process, and
134 * implement the semantics of socket operations by
135 * switching out to the protocol specific routines.
139 * Get a socket structure, and initialize it.
140 * Note that it would probably be better to allocate socket
141 * and PCB at the same time, but I'm not convinced that all
142 * the protocols can be easily modified to do this.
150 waitmask = waitok ? M_WAITOK : M_NOWAIT;
151 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
153 /* XXX race condition for reentrant kernel */
154 TAILQ_INIT(&so->so_aiojobq);
155 TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist);
156 TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist);
157 lwkt_token_init(&so->so_rcv.ssb_token, 1, "rcvtok");
158 lwkt_token_init(&so->so_snd.ssb_token, 1, "rcvtok");
159 so->so_state = SS_NOFDREF;
166 socreate(int dom, struct socket **aso, int type,
167 int proto, struct thread *td)
169 struct proc *p = td->td_proc;
172 struct pru_attach_info ai;
176 prp = pffindproto(dom, proto, type);
178 prp = pffindtype(dom, type);
180 if (prp == 0 || prp->pr_usrreqs->pru_attach == 0)
181 return (EPROTONOSUPPORT);
183 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
184 prp->pr_domain->dom_family != PF_LOCAL &&
185 prp->pr_domain->dom_family != PF_INET &&
186 prp->pr_domain->dom_family != PF_INET6 &&
187 prp->pr_domain->dom_family != PF_ROUTE) {
188 return (EPROTONOSUPPORT);
191 if (prp->pr_type != type)
193 so = soalloc(p != 0);
198 * Callers of socreate() presumably will connect up a descriptor
199 * and call soclose() if they cannot. This represents our so_refs
200 * (which should be 1) from soalloc().
202 soclrstate(so, SS_NOFDREF);
205 * Set a default port for protocol processing. No action will occur
206 * on the socket on this port until an inpcb is attached to it and
207 * is able to match incoming packets, or until the socket becomes
208 * available to userland.
210 so->so_port = cpu0_soport(so, NULL, NULL);
212 TAILQ_INIT(&so->so_incomp);
213 TAILQ_INIT(&so->so_comp);
215 so->so_cred = crhold(p->p_ucred);
217 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
218 ai.p_ucred = p->p_ucred;
219 ai.fd_rdir = p->p_fd->fd_rdir;
222 * Auto-sizing of socket buffers is managed by the protocols and
223 * the appropriate flags must be set in the pru_attach function.
225 error = so_pru_attach(so, proto, &ai);
227 sosetstate(so, SS_NOFDREF);
228 sofree(so); /* from soalloc */
233 * NOTE: Returns referenced socket.
240 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
244 error = so_pru_bind(so, nam, td);
249 sodealloc(struct socket *so)
251 if (so->so_rcv.ssb_hiwat)
252 (void)chgsbsize(so->so_cred->cr_uidinfo,
253 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
254 if (so->so_snd.ssb_hiwat)
255 (void)chgsbsize(so->so_cred->cr_uidinfo,
256 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
258 /* remove accept filter if present */
259 if (so->so_accf != NULL)
260 do_setopt_accept_filter(so, NULL);
267 solisten(struct socket *so, int backlog, struct thread *td)
271 short oldopt, oldqlimit;
274 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
278 oldopt = so->so_options;
279 oldqlimit = so->so_qlimit;
282 lwkt_gettoken(&so->so_rcv.ssb_token);
283 if (TAILQ_EMPTY(&so->so_comp))
284 so->so_options |= SO_ACCEPTCONN;
285 lwkt_reltoken(&so->so_rcv.ssb_token);
286 if (backlog < 0 || backlog > somaxconn)
288 so->so_qlimit = backlog;
289 /* SCTP needs to look at tweak both the inbound backlog parameter AND
290 * the so_options (UDP model both connect's and gets inbound
291 * connections .. implicitly).
293 error = so_pru_listen(so, td);
296 /* Restore the params */
297 so->so_options = oldopt;
298 so->so_qlimit = oldqlimit;
306 * Destroy a disconnected socket. This routine is a NOP if entities
307 * still have a reference on the socket:
309 * so_pcb - The protocol stack still has a reference
310 * SS_NOFDREF - There is no longer a file pointer reference
313 sofree(struct socket *so)
315 struct socket *head = so->so_head;
318 * Arbitrage the last free.
320 KKASSERT(so->so_refs > 0);
321 if (atomic_fetchadd_int(&so->so_refs, -1) != 1)
324 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
327 * We're done, clean up
330 lwkt_gettoken(&head->so_rcv.ssb_token);
331 if (so->so_state & SS_INCOMP) {
332 TAILQ_REMOVE(&head->so_incomp, so, so_list);
334 } else if (so->so_state & SS_COMP) {
336 * We must not decommission a socket that's
337 * on the accept(2) queue. If we do, then
338 * accept(2) may hang after select(2) indicated
339 * that the listening socket was ready.
341 lwkt_reltoken(&head->so_rcv.ssb_token);
344 panic("sofree: not queued");
346 soclrstate(so, SS_INCOMP);
348 lwkt_reltoken(&head->so_rcv.ssb_token);
350 ssb_release(&so->so_snd, so);
356 * Close a socket on last file table reference removal.
357 * Initiate disconnect if connected.
358 * Free socket when disconnect complete.
361 soclose(struct socket *so, int fflag)
365 funsetown(so->so_sigio);
366 if (so->so_pcb == NULL)
368 if (so->so_state & SS_ISCONNECTED) {
369 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
370 error = sodisconnect(so);
374 if (so->so_options & SO_LINGER) {
375 if ((so->so_state & SS_ISDISCONNECTING) &&
378 while (so->so_state & SS_ISCONNECTED) {
379 error = tsleep(&so->so_timeo, PCATCH,
380 "soclos", so->so_linger * hz);
390 error2 = so_pru_detach(so);
395 lwkt_gettoken(&so->so_rcv.ssb_token);
396 if (so->so_options & SO_ACCEPTCONN) {
399 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
400 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
401 soclrstate(sp, SS_INCOMP);
406 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
407 TAILQ_REMOVE(&so->so_comp, sp, so_list);
408 soclrstate(sp, SS_COMP);
414 lwkt_reltoken(&so->so_rcv.ssb_token);
415 if (so->so_state & SS_NOFDREF)
416 panic("soclose: NOFDREF");
417 sosetstate(so, SS_NOFDREF); /* take ref */
418 sofree(so); /* dispose of ref */
423 * Abort and destroy a socket. Only one abort can be in progress
424 * at any given moment.
427 soabort(struct socket *so)
434 soaborta(struct socket *so)
441 soabort_oncpu(struct socket *so)
444 so_pru_abort_oncpu(so);
448 soaccept(struct socket *so, struct sockaddr **nam)
452 if ((so->so_state & SS_NOFDREF) == 0)
453 panic("soaccept: !NOFDREF");
454 soreference(so); /* create ref */
455 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */
456 error = so_pru_accept(so, nam);
461 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
465 if (so->so_options & SO_ACCEPTCONN)
468 * If protocol is connection-based, can only connect once.
469 * Otherwise, if connected, try to disconnect first.
470 * This allows user to disconnect by connecting to, e.g.,
473 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
474 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
475 (error = sodisconnect(so)))) {
479 * Prevent accumulated error from previous connection
483 error = so_pru_connect(so, nam, td);
489 soconnect2(struct socket *so1, struct socket *so2)
493 error = so_pru_connect2(so1, so2);
498 sodisconnect(struct socket *so)
502 if ((so->so_state & SS_ISCONNECTED) == 0) {
506 if (so->so_state & SS_ISDISCONNECTING) {
510 error = so_pru_disconnect(so);
515 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
518 * If send must go all at once and message is larger than
519 * send buffering, then hard error.
520 * Lock against other senders.
521 * If must go all at once and not enough room now, then
522 * inform user that this would block and do nothing.
523 * Otherwise, if nonblocking, send as much as possible.
524 * The data to be sent is described by "uio" if nonzero,
525 * otherwise by the mbuf chain "top" (which must be null
526 * if uio is not). Data provided in mbuf chain must be small
527 * enough to send all at once.
529 * Returns nonzero on error, timeout or signal; callers
530 * must check for short counts if EINTR/ERESTART are returned.
531 * Data and control buffers are freed on return.
534 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
535 struct mbuf *top, struct mbuf *control, int flags,
542 int clen = 0, error, dontroute, mlen;
543 int atomic = sosendallatonce(so) || top;
547 resid = uio->uio_resid;
549 resid = (size_t)top->m_pkthdr.len;
552 for (m = top; m; m = m->m_next)
554 KKASSERT(top->m_pkthdr.len == len);
559 * WARNING! resid is unsigned, space and len are signed. space
560 * can wind up negative if the sockbuf is overcommitted.
562 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
563 * type sockets since that's an error.
565 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
571 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
572 (so->so_proto->pr_flags & PR_ATOMIC);
573 if (td->td_lwp != NULL)
574 td->td_lwp->lwp_ru.ru_msgsnd++;
576 clen = control->m_len;
577 #define gotoerr(errcode) { error = errcode; goto release; }
580 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
585 if (so->so_state & SS_CANTSENDMORE)
588 error = so->so_error;
592 if ((so->so_state & SS_ISCONNECTED) == 0) {
594 * `sendto' and `sendmsg' is allowed on a connection-
595 * based socket if it supports implied connect.
596 * Return ENOTCONN if not connected and no address is
599 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
600 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
601 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
602 !(resid == 0 && clen != 0))
604 } else if (addr == 0)
605 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
606 ENOTCONN : EDESTADDRREQ);
608 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
609 clen > so->so_snd.ssb_hiwat) {
612 space = ssb_space(&so->so_snd);
615 if ((space < 0 || (size_t)space < resid + clen) && uio &&
616 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
617 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
618 gotoerr(EWOULDBLOCK);
619 ssb_unlock(&so->so_snd);
620 error = ssb_wait(&so->so_snd);
630 * Data is prepackaged in "top".
634 top->m_flags |= M_EOR;
638 m = m_getl((int)resid, MB_WAIT, MT_DATA,
639 top == NULL ? M_PKTHDR : 0, &mlen);
642 m->m_pkthdr.rcvif = NULL;
644 len = imin((int)szmin(mlen, resid), space);
645 if (resid < MINCLSIZE) {
647 * For datagram protocols, leave room
648 * for protocol headers in first mbuf.
650 if (atomic && top == 0 && len < mlen)
654 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
655 resid = uio->uio_resid;
658 top->m_pkthdr.len += len;
664 top->m_flags |= M_EOR;
667 } while (space > 0 && atomic);
669 so->so_options |= SO_DONTROUTE;
670 if (flags & MSG_OOB) {
671 pru_flags = PRUS_OOB;
672 } else if ((flags & MSG_EOF) &&
673 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
676 * If the user set MSG_EOF, the protocol
677 * understands this flag and nothing left to
678 * send then use PRU_SEND_EOF instead of PRU_SEND.
680 pru_flags = PRUS_EOF;
681 } else if (resid > 0 && space > 0) {
682 /* If there is more to send, set PRUS_MORETOCOME */
683 pru_flags = PRUS_MORETOCOME;
688 * XXX all the SS_CANTSENDMORE checks previously
689 * done could be out of date. We could have recieved
690 * a reset packet in an interrupt or maybe we slept
691 * while doing page faults in uiomove() etc. We could
692 * probably recheck again inside the splnet() protection
693 * here, but there are probably other places that this
694 * also happens. We must rethink this.
696 error = so_pru_send(so, pru_flags, top, addr, control, td);
698 so->so_options &= ~SO_DONTROUTE;
705 } while (resid && space > 0);
709 ssb_unlock(&so->so_snd);
719 * A specialization of sosend() for UDP based on protocol-specific knowledge:
720 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
721 * sosendallatonce() returns true,
722 * the "atomic" variable is true,
723 * and sosendudp() blocks until space is available for the entire send.
724 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
725 * PR_IMPLOPCL flags set.
726 * UDP has no out-of-band data.
727 * UDP has no control data.
728 * UDP does not support MSG_EOR.
731 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
732 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
734 boolean_t dontroute; /* temporary SO_DONTROUTE setting */
739 if (td->td_lwp != NULL)
740 td->td_lwp->lwp_ru.ru_msgsnd++;
744 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
745 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
748 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
752 if (so->so_state & SS_CANTSENDMORE)
755 error = so->so_error;
759 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
760 gotoerr(EDESTADDRREQ);
761 if (resid > so->so_snd.ssb_hiwat)
763 space = ssb_space(&so->so_snd);
764 if (uio && (space < 0 || (size_t)space < resid)) {
765 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
766 gotoerr(EWOULDBLOCK);
767 ssb_unlock(&so->so_snd);
768 error = ssb_wait(&so->so_snd);
775 top = m_uiomove(uio);
780 dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
782 so->so_options |= SO_DONTROUTE;
784 error = so_pru_send(so, 0, top, addr, NULL, td);
785 top = NULL; /* sent or freed in lower layer */
788 so->so_options &= ~SO_DONTROUTE;
791 ssb_unlock(&so->so_snd);
799 * Implement receive operations on a socket.
801 * We depend on the way that records are added to the signalsockbuf
802 * by sbappend*. In particular, each record (mbufs linked through m_next)
803 * must begin with an address if the protocol so specifies,
804 * followed by an optional mbuf or mbufs containing ancillary data,
805 * and then zero or more mbufs of data.
807 * Although the signalsockbuf is locked, new data may still be appended.
808 * A token inside the ssb_lock deals with MP issues and still allows
809 * the network to access the socket if we block in a uio.
811 * The caller may receive the data as a single mbuf chain by supplying
812 * an mbuf **mp0 for use in returning the chain. The uio is then used
813 * only for the count in uio_resid.
816 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
817 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
820 struct mbuf *free_chain = NULL;
821 int flags, len, error, offset;
822 struct protosw *pr = so->so_proto;
824 size_t resid, orig_resid;
827 resid = uio->uio_resid;
829 resid = (size_t)(sio->sb_climit - sio->sb_cc);
837 flags = *flagsp &~ MSG_EOR;
840 if (flags & MSG_OOB) {
841 m = m_get(MB_WAIT, MT_DATA);
844 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
850 KKASSERT(resid >= (size_t)m->m_len);
851 resid -= (size_t)m->m_len;
852 } while (resid > 0 && m);
855 uio->uio_resid = resid;
856 error = uiomove(mtod(m, caddr_t),
857 (int)szmin(resid, m->m_len),
859 resid = uio->uio_resid;
861 } while (uio->uio_resid && error == 0 && m);
868 if ((so->so_state & SS_ISCONFIRMING) && resid)
872 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
876 m = so->so_rcv.ssb_mb;
878 * If we have less data than requested, block awaiting more
879 * (subject to any timeout) if:
880 * 1. the current count is less than the low water mark, or
881 * 2. MSG_WAITALL is set, and it is possible to do the entire
882 * receive operation at once if we block (resid <= hiwat).
883 * 3. MSG_DONTWAIT is not set
884 * If MSG_WAITALL is set but resid is larger than the receive buffer,
885 * we have to do the receive in sections, and thus risk returning
886 * a short count if a timeout or signal occurs after we start.
888 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
889 (size_t)so->so_rcv.ssb_cc < resid) &&
890 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
891 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
892 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
893 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
897 error = so->so_error;
898 if ((flags & MSG_PEEK) == 0)
902 if (so->so_state & SS_CANTRCVMORE) {
908 for (; m; m = m->m_next) {
909 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
910 m = so->so_rcv.ssb_mb;
914 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
915 (pr->pr_flags & PR_CONNREQUIRED)) {
921 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
925 ssb_unlock(&so->so_rcv);
926 error = ssb_wait(&so->so_rcv);
932 if (uio && uio->uio_td && uio->uio_td->td_proc)
933 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
936 * note: m should be == sb_mb here. Cache the next record while
937 * cleaning up. Note that calling m_free*() will break out critical
940 KKASSERT(m == so->so_rcv.ssb_mb);
943 * Skip any address mbufs prepending the record.
945 if (pr->pr_flags & PR_ADDR) {
946 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
949 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
950 if (flags & MSG_PEEK)
953 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
957 * Skip any control mbufs prepending the record.
960 if (pr->pr_flags & PR_ADDR_OPT) {
962 * For SCTP we may be getting a
963 * whole message OR a partial delivery.
965 if (m && m->m_type == MT_SONAME) {
968 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
969 if (flags & MSG_PEEK)
972 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
976 while (m && m->m_type == MT_CONTROL && error == 0) {
977 if (flags & MSG_PEEK) {
979 *controlp = m_copy(m, 0, m->m_len);
980 m = m->m_next; /* XXX race */
983 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
984 if (pr->pr_domain->dom_externalize &&
985 mtod(m, struct cmsghdr *)->cmsg_type ==
987 error = (*pr->pr_domain->dom_externalize)(m);
991 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
994 if (controlp && *controlp) {
996 controlp = &(*controlp)->m_next;
1005 if (type == MT_OOBDATA)
1010 * Copy to the UIO or mbuf return chain (*mp).
1014 while (m && resid > 0 && error == 0) {
1015 if (m->m_type == MT_OOBDATA) {
1016 if (type != MT_OOBDATA)
1018 } else if (type == MT_OOBDATA)
1021 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1023 soclrstate(so, SS_RCVATMARK);
1024 len = (resid > INT_MAX) ? INT_MAX : resid;
1025 if (so->so_oobmark && len > so->so_oobmark - offset)
1026 len = so->so_oobmark - offset;
1027 if (len > m->m_len - moff)
1028 len = m->m_len - moff;
1031 * Copy out to the UIO or pass the mbufs back to the SIO.
1032 * The SIO is dealt with when we eat the mbuf, but deal
1033 * with the resid here either way.
1036 uio->uio_resid = resid;
1037 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1038 resid = uio->uio_resid;
1042 resid -= (size_t)len;
1046 * Eat the entire mbuf or just a piece of it
1048 if (len == m->m_len - moff) {
1049 if (m->m_flags & M_EOR)
1052 if (m->m_flags & M_NOTIFICATION)
1053 flags |= MSG_NOTIFICATION;
1055 if (flags & MSG_PEEK) {
1060 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1064 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1068 if (flags & MSG_PEEK) {
1072 n = m_copym(m, 0, len, MB_WAIT);
1078 so->so_rcv.ssb_cc -= len;
1081 if (so->so_oobmark) {
1082 if ((flags & MSG_PEEK) == 0) {
1083 so->so_oobmark -= len;
1084 if (so->so_oobmark == 0) {
1085 sosetstate(so, SS_RCVATMARK);
1090 if (offset == so->so_oobmark)
1094 if (flags & MSG_EOR)
1097 * If the MSG_WAITALL flag is set (for non-atomic socket),
1098 * we must not quit until resid == 0 or an error
1099 * termination. If a signal/timeout occurs, return
1100 * with a short count but without error.
1101 * Keep signalsockbuf locked against other readers.
1103 while ((flags & MSG_WAITALL) && m == NULL &&
1104 resid > 0 && !sosendallatonce(so) &&
1105 so->so_rcv.ssb_mb == NULL) {
1106 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1109 * The window might have closed to zero, make
1110 * sure we send an ack now that we've drained
1111 * the buffer or we might end up blocking until
1112 * the idle takes over (5 seconds).
1114 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1115 so_pru_rcvd(so, flags);
1116 error = ssb_wait(&so->so_rcv);
1118 ssb_unlock(&so->so_rcv);
1122 m = so->so_rcv.ssb_mb;
1127 * If an atomic read was requested but unread data still remains
1128 * in the record, set MSG_TRUNC.
1130 if (m && pr->pr_flags & PR_ATOMIC)
1134 * Cleanup. If an atomic read was requested drop any unread data.
1136 if ((flags & MSG_PEEK) == 0) {
1137 if (m && (pr->pr_flags & PR_ATOMIC))
1138 sbdroprecord(&so->so_rcv.sb);
1139 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1140 so_pru_rcvd(so, flags);
1143 if (orig_resid == resid && orig_resid &&
1144 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1145 ssb_unlock(&so->so_rcv);
1152 ssb_unlock(&so->so_rcv);
1155 m_freem(free_chain);
1160 soshutdown(struct socket *so, int how)
1162 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1168 return (so_pru_shutdown(so));
1173 sorflush(struct socket *so)
1175 struct signalsockbuf *ssb = &so->so_rcv;
1176 struct protosw *pr = so->so_proto;
1177 struct signalsockbuf asb;
1179 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1181 ssb_lock(ssb, M_WAITOK);
1186 * Can't just blow up the ssb structure here
1189 ssb->ssb_unused01 = 0;
1193 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1197 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1198 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1199 ssb_release(&asb, so);
1204 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1206 struct accept_filter_arg *afap = NULL;
1207 struct accept_filter *afp;
1208 struct so_accf *af = so->so_accf;
1211 /* do not set/remove accept filters on non listen sockets */
1212 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1217 /* removing the filter */
1220 if (af->so_accept_filter != NULL &&
1221 af->so_accept_filter->accf_destroy != NULL) {
1222 af->so_accept_filter->accf_destroy(so);
1224 if (af->so_accept_filter_str != NULL) {
1225 FREE(af->so_accept_filter_str, M_ACCF);
1230 so->so_options &= ~SO_ACCEPTFILTER;
1233 /* adding a filter */
1234 /* must remove previous filter first */
1239 /* don't put large objects on the kernel stack */
1240 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1241 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1242 afap->af_name[sizeof(afap->af_name)-1] = '\0';
1243 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1246 afp = accept_filt_get(afap->af_name);
1251 MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1252 if (afp->accf_create != NULL) {
1253 if (afap->af_name[0] != '\0') {
1254 int len = strlen(afap->af_name) + 1;
1256 MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1257 strcpy(af->so_accept_filter_str, afap->af_name);
1259 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1260 if (af->so_accept_filter_arg == NULL) {
1261 FREE(af->so_accept_filter_str, M_ACCF);
1268 af->so_accept_filter = afp;
1270 so->so_options |= SO_ACCEPTFILTER;
1279 * Perhaps this routine, and sooptcopyout(), below, ought to come in
1280 * an additional variant to handle the case where the option value needs
1281 * to be some kind of integer, but not a specific size.
1282 * In addition to their use here, these functions are also called by the
1283 * protocol-level pr_ctloutput() routines.
1286 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1288 return soopt_to_kbuf(sopt, buf, len, minlen);
1292 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1296 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1297 KKASSERT(kva_p(buf));
1300 * If the user gives us more than we wanted, we ignore it,
1301 * but if we don't get the minimum length the caller
1302 * wants, we return EINVAL. On success, sopt->sopt_valsize
1303 * is set to however much we actually retrieved.
1305 if ((valsize = sopt->sopt_valsize) < minlen)
1308 sopt->sopt_valsize = valsize = len;
1310 bcopy(sopt->sopt_val, buf, valsize);
1316 sosetopt(struct socket *so, struct sockopt *sopt)
1322 struct signalsockbuf *sotmp;
1325 sopt->sopt_dir = SOPT_SET;
1326 if (sopt->sopt_level != SOL_SOCKET) {
1327 if (so->so_proto && so->so_proto->pr_ctloutput) {
1328 return (so_pru_ctloutput(so, sopt));
1330 error = ENOPROTOOPT;
1332 switch (sopt->sopt_name) {
1334 case SO_ACCEPTFILTER:
1335 error = do_setopt_accept_filter(so, sopt);
1341 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1345 so->so_linger = l.l_linger;
1347 so->so_options |= SO_LINGER;
1349 so->so_options &= ~SO_LINGER;
1355 case SO_USELOOPBACK:
1361 error = sooptcopyin(sopt, &optval, sizeof optval,
1366 so->so_options |= sopt->sopt_name;
1368 so->so_options &= ~sopt->sopt_name;
1375 error = sooptcopyin(sopt, &optval, sizeof optval,
1381 * Values < 1 make no sense for any of these
1382 * options, so disallow them.
1389 switch (sopt->sopt_name) {
1392 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1393 &so->so_snd : &so->so_rcv, (u_long)optval,
1395 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1399 sotmp = (sopt->sopt_name == SO_SNDBUF) ?
1400 &so->so_snd : &so->so_rcv;
1401 atomic_clear_int(&sotmp->ssb_flags,
1406 * Make sure the low-water is never greater than
1410 so->so_snd.ssb_lowat =
1411 (optval > so->so_snd.ssb_hiwat) ?
1412 so->so_snd.ssb_hiwat : optval;
1413 atomic_clear_int(&so->so_snd.ssb_flags,
1417 so->so_rcv.ssb_lowat =
1418 (optval > so->so_rcv.ssb_hiwat) ?
1419 so->so_rcv.ssb_hiwat : optval;
1420 atomic_clear_int(&so->so_rcv.ssb_flags,
1428 error = sooptcopyin(sopt, &tv, sizeof tv,
1433 /* assert(hz > 0); */
1434 if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz ||
1435 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1439 /* assert(tick > 0); */
1440 /* assert(ULONG_MAX - SHRT_MAX >= 1000000); */
1441 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
1442 if (val > SHRT_MAX) {
1446 if (val == 0 && tv.tv_usec != 0)
1449 switch (sopt->sopt_name) {
1451 so->so_snd.ssb_timeo = val;
1454 so->so_rcv.ssb_timeo = val;
1459 error = ENOPROTOOPT;
1462 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1463 (void) so_pru_ctloutput(so, sopt);
1470 /* Helper routine for getsockopt */
1472 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1474 soopt_from_kbuf(sopt, buf, len);
1479 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1484 sopt->sopt_valsize = 0;
1488 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1489 KKASSERT(kva_p(buf));
1492 * Documented get behavior is that we always return a value,
1493 * possibly truncated to fit in the user's buffer.
1494 * Traditional behavior is that we always tell the user
1495 * precisely how much we copied, rather than something useful
1496 * like the total amount we had available for her.
1497 * Note that this interface is not idempotent; the entire answer must
1498 * generated ahead of time.
1500 valsize = szmin(len, sopt->sopt_valsize);
1501 sopt->sopt_valsize = valsize;
1502 if (sopt->sopt_val != 0) {
1503 bcopy(buf, sopt->sopt_val, valsize);
1508 sogetopt(struct socket *so, struct sockopt *sopt)
1514 struct accept_filter_arg *afap;
1518 sopt->sopt_dir = SOPT_GET;
1519 if (sopt->sopt_level != SOL_SOCKET) {
1520 if (so->so_proto && so->so_proto->pr_ctloutput) {
1521 return (so_pru_ctloutput(so, sopt));
1523 return (ENOPROTOOPT);
1525 switch (sopt->sopt_name) {
1527 case SO_ACCEPTFILTER:
1528 if ((so->so_options & SO_ACCEPTCONN) == 0)
1530 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1531 M_TEMP, M_WAITOK | M_ZERO);
1532 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1533 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1534 if (so->so_accf->so_accept_filter_str != NULL)
1535 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1537 error = sooptcopyout(sopt, afap, sizeof(*afap));
1543 l.l_onoff = so->so_options & SO_LINGER;
1544 l.l_linger = so->so_linger;
1545 error = sooptcopyout(sopt, &l, sizeof l);
1548 case SO_USELOOPBACK:
1557 optval = so->so_options & sopt->sopt_name;
1559 error = sooptcopyout(sopt, &optval, sizeof optval);
1563 optval = so->so_type;
1567 optval = so->so_error;
1572 optval = so->so_snd.ssb_hiwat;
1576 optval = so->so_rcv.ssb_hiwat;
1580 optval = so->so_snd.ssb_lowat;
1584 optval = so->so_rcv.ssb_lowat;
1589 optval = (sopt->sopt_name == SO_SNDTIMEO ?
1590 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1592 tv.tv_sec = optval / hz;
1593 tv.tv_usec = (optval % hz) * ustick;
1594 error = sooptcopyout(sopt, &tv, sizeof tv);
1598 error = ENOPROTOOPT;
1605 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1607 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1609 struct mbuf *m, *m_prev;
1610 int sopt_size = sopt->sopt_valsize, msize;
1612 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1616 m->m_len = min(msize, sopt_size);
1617 sopt_size -= m->m_len;
1621 while (sopt_size > 0) {
1622 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1623 MT_DATA, 0, &msize);
1628 m->m_len = min(msize, sopt_size);
1629 sopt_size -= m->m_len;
1636 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1638 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1640 soopt_to_mbuf(sopt, m);
1645 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
1650 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1652 if (sopt->sopt_val == NULL)
1654 val = sopt->sopt_val;
1655 valsize = sopt->sopt_valsize;
1656 while (m != NULL && valsize >= m->m_len) {
1657 bcopy(val, mtod(m, char *), m->m_len);
1658 valsize -= m->m_len;
1659 val = (caddr_t)val + m->m_len;
1662 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1663 panic("ip6_sooptmcopyin");
1666 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1668 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1670 return soopt_from_mbuf(sopt, m);
1674 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
1676 struct mbuf *m0 = m;
1681 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1683 if (sopt->sopt_val == NULL)
1685 val = sopt->sopt_val;
1686 maxsize = sopt->sopt_valsize;
1687 while (m != NULL && maxsize >= m->m_len) {
1688 bcopy(mtod(m, char *), val, m->m_len);
1689 maxsize -= m->m_len;
1690 val = (caddr_t)val + m->m_len;
1691 valsize += m->m_len;
1695 /* enough soopt buffer should be given from user-land */
1699 sopt->sopt_valsize = valsize;
1704 sohasoutofband(struct socket *so)
1706 if (so->so_sigio != NULL)
1707 pgsigio(so->so_sigio, SIGURG, 0);
1708 KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
1712 sokqfilter(struct file *fp, struct knote *kn)
1714 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1715 struct signalsockbuf *ssb;
1717 switch (kn->kn_filter) {
1719 if (so->so_options & SO_ACCEPTCONN)
1720 kn->kn_fop = &solisten_filtops;
1722 kn->kn_fop = &soread_filtops;
1726 kn->kn_fop = &sowrite_filtops;
1730 kn->kn_fop = &soexcept_filtops;
1734 return (EOPNOTSUPP);
1737 knote_insert(&ssb->ssb_kq.ki_note, kn);
1738 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
1743 filt_sordetach(struct knote *kn)
1745 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1747 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
1748 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
1749 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
1754 filt_soread(struct knote *kn, long hint)
1756 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1758 if (kn->kn_sfflags & NOTE_OOB) {
1759 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
1760 kn->kn_fflags |= NOTE_OOB;
1765 kn->kn_data = so->so_rcv.ssb_cc;
1768 * Only set EOF if all data has been exhausted.
1770 if ((so->so_state & SS_CANTRCVMORE) && kn->kn_data == 0) {
1771 kn->kn_flags |= EV_EOF;
1772 kn->kn_fflags = so->so_error;
1775 if (so->so_error) /* temporary udp error */
1777 if (kn->kn_sfflags & NOTE_LOWAT)
1778 return (kn->kn_data >= kn->kn_sdata);
1779 return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
1780 !TAILQ_EMPTY(&so->so_comp));
1784 filt_sowdetach(struct knote *kn)
1786 struct socket *so = (struct socket *)kn->kn_fp->f_data;
1788 knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
1789 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
1790 atomic_clear_int(&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));