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 $
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/fcntl.h>
76 #include <sys/malloc.h>
78 #include <sys/domain.h>
79 #include <sys/file.h> /* for struct knote */
80 #include <sys/kernel.h>
81 #include <sys/event.h>
83 #include <sys/protosw.h>
84 #include <sys/socket.h>
85 #include <sys/socketvar.h>
86 #include <sys/socketops.h>
87 #include <sys/resourcevar.h>
88 #include <sys/signalvar.h>
89 #include <sys/sysctl.h>
92 #include <vm/vm_zone.h>
94 #include <net/netmsg2.h>
96 #include <sys/thread2.h>
97 #include <sys/socketvar2.h>
99 #include <machine/limits.h>
101 extern int tcp_sosnd_agglim;
102 extern int tcp_sosnd_async;
105 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
108 static void filt_sordetach(struct knote *kn);
109 static int filt_soread(struct knote *kn, long hint);
110 static void filt_sowdetach(struct knote *kn);
111 static int filt_sowrite(struct knote *kn, long hint);
112 static int filt_solisten(struct knote *kn, long hint);
114 static void sodiscard(struct socket *so);
115 static int soclose_sync(struct socket *so, int fflag);
116 static void soclose_fast(struct socket *so);
118 static struct filterops solisten_filtops =
119 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten };
120 static struct filterops soread_filtops =
121 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
122 static struct filterops sowrite_filtops =
123 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite };
124 static struct filterops soexcept_filtops =
125 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
127 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
128 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
129 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
132 static int somaxconn = SOMAXCONN;
133 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
134 &somaxconn, 0, "Maximum pending socket connection queue size");
136 static int use_soclose_fast = 1;
137 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW,
138 &use_soclose_fast, 0, "Fast socket close");
140 int use_soaccept_pred_fast = 1;
141 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW,
142 &use_soaccept_pred_fast, 0, "Fast socket accept predication");
145 * Socket operation routines.
146 * These routines are called by the routines in
147 * sys_socket.c or from a system process, and
148 * implement the semantics of socket operations by
149 * switching out to the protocol specific routines.
153 * Get a socket structure, and initialize it.
154 * Note that it would probably be better to allocate socket
155 * and PCB at the same time, but I'm not convinced that all
156 * the protocols can be easily modified to do this.
164 waitmask = waitok ? M_WAITOK : M_NOWAIT;
165 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
167 /* XXX race condition for reentrant kernel */
168 TAILQ_INIT(&so->so_aiojobq);
169 TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist);
170 TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist);
171 lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
172 lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
173 so->so_state = SS_NOFDREF;
180 socreate(int dom, struct socket **aso, int type,
181 int proto, struct thread *td)
183 struct proc *p = td->td_proc;
186 struct pru_attach_info ai;
190 prp = pffindproto(dom, proto, type);
192 prp = pffindtype(dom, type);
194 if (prp == 0 || prp->pr_usrreqs->pru_attach == 0)
195 return (EPROTONOSUPPORT);
197 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
198 prp->pr_domain->dom_family != PF_LOCAL &&
199 prp->pr_domain->dom_family != PF_INET &&
200 prp->pr_domain->dom_family != PF_INET6 &&
201 prp->pr_domain->dom_family != PF_ROUTE) {
202 return (EPROTONOSUPPORT);
205 if (prp->pr_type != type)
207 so = soalloc(p != 0);
212 * Callers of socreate() presumably will connect up a descriptor
213 * and call soclose() if they cannot. This represents our so_refs
214 * (which should be 1) from soalloc().
216 soclrstate(so, SS_NOFDREF);
219 * Set a default port for protocol processing. No action will occur
220 * on the socket on this port until an inpcb is attached to it and
221 * is able to match incoming packets, or until the socket becomes
222 * available to userland.
224 * We normally default the socket to the protocol thread on cpu 0.
225 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
226 * thread and all pr_*()/pru_*() calls are executed synchronously.
228 if (prp->pr_flags & PR_SYNC_PORT)
229 so->so_port = &netisr_sync_port;
231 so->so_port = cpu_portfn(0);
233 TAILQ_INIT(&so->so_incomp);
234 TAILQ_INIT(&so->so_comp);
236 so->so_cred = crhold(p->p_ucred);
238 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
239 ai.p_ucred = p->p_ucred;
240 ai.fd_rdir = p->p_fd->fd_rdir;
243 * Auto-sizing of socket buffers is managed by the protocols and
244 * the appropriate flags must be set in the pru_attach function.
246 error = so_pru_attach(so, proto, &ai);
248 sosetstate(so, SS_NOFDREF);
249 sofree(so); /* from soalloc */
254 * NOTE: Returns referenced socket.
261 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
265 error = so_pru_bind(so, nam, td);
270 sodealloc(struct socket *so)
272 if (so->so_rcv.ssb_hiwat)
273 (void)chgsbsize(so->so_cred->cr_uidinfo,
274 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
275 if (so->so_snd.ssb_hiwat)
276 (void)chgsbsize(so->so_cred->cr_uidinfo,
277 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
279 /* remove accept filter if present */
280 if (so->so_accf != NULL)
281 do_setopt_accept_filter(so, NULL);
288 solisten(struct socket *so, int backlog, struct thread *td)
292 short oldopt, oldqlimit;
295 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
299 oldopt = so->so_options;
300 oldqlimit = so->so_qlimit;
303 lwkt_gettoken(&so->so_rcv.ssb_token);
304 if (TAILQ_EMPTY(&so->so_comp))
305 so->so_options |= SO_ACCEPTCONN;
306 lwkt_reltoken(&so->so_rcv.ssb_token);
307 if (backlog < 0 || backlog > somaxconn)
309 so->so_qlimit = backlog;
310 /* SCTP needs to look at tweak both the inbound backlog parameter AND
311 * the so_options (UDP model both connect's and gets inbound
312 * connections .. implicitly).
314 error = so_pru_listen(so, td);
317 /* Restore the params */
318 so->so_options = oldopt;
319 so->so_qlimit = oldqlimit;
327 * Destroy a disconnected socket. This routine is a NOP if entities
328 * still have a reference on the socket:
330 * so_pcb - The protocol stack still has a reference
331 * SS_NOFDREF - There is no longer a file pointer reference
334 sofree(struct socket *so)
339 * This is a bit hackish at the moment. We need to interlock
340 * any accept queue we are on before we potentially lose the
341 * last reference to avoid races against a re-reference from
342 * someone operating on the queue.
344 while ((head = so->so_head) != NULL) {
345 lwkt_getpooltoken(head);
346 if (so->so_head == head)
348 lwkt_relpooltoken(head);
352 * Arbitrage the last free.
354 KKASSERT(so->so_refs > 0);
355 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
357 lwkt_relpooltoken(head);
361 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
362 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
365 * We're done, remove ourselves from the accept queue we are
366 * on, if we are on one.
369 if (so->so_state & SS_INCOMP) {
370 TAILQ_REMOVE(&head->so_incomp, so, so_list);
372 } else if (so->so_state & SS_COMP) {
374 * We must not decommission a socket that's
375 * on the accept(2) queue. If we do, then
376 * accept(2) may hang after select(2) indicated
377 * that the listening socket was ready.
379 lwkt_relpooltoken(head);
382 panic("sofree: not queued");
384 soclrstate(so, SS_INCOMP);
386 lwkt_relpooltoken(head);
388 ssb_release(&so->so_snd, so);
394 * Close a socket on last file table reference removal.
395 * Initiate disconnect if connected.
396 * Free socket when disconnect complete.
399 soclose(struct socket *so, int fflag)
403 funsetown(&so->so_sigio);
404 if (!use_soclose_fast ||
405 (so->so_proto->pr_flags & PR_SYNC_PORT) ||
406 (so->so_options & SO_LINGER)) {
407 error = soclose_sync(so, fflag);
416 sodiscard(struct socket *so)
418 lwkt_getpooltoken(so);
419 if (so->so_options & SO_ACCEPTCONN) {
422 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
423 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
424 soclrstate(sp, SS_INCOMP);
429 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
430 TAILQ_REMOVE(&so->so_comp, sp, so_list);
431 soclrstate(sp, SS_COMP);
437 lwkt_relpooltoken(so);
439 if (so->so_state & SS_NOFDREF)
440 panic("soclose: NOFDREF");
441 sosetstate(so, SS_NOFDREF); /* take ref */
445 soclose_sync(struct socket *so, int fflag)
449 if (so->so_pcb == NULL)
451 if (so->so_state & SS_ISCONNECTED) {
452 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
453 error = sodisconnect(so);
457 if (so->so_options & SO_LINGER) {
458 if ((so->so_state & SS_ISDISCONNECTING) &&
461 while (so->so_state & SS_ISCONNECTED) {
462 error = tsleep(&so->so_timeo, PCATCH,
463 "soclos", so->so_linger * hz);
473 error2 = so_pru_detach(so);
479 so_pru_sync(so); /* unpend async sending */
480 sofree(so); /* dispose of ref */
486 soclose_sofree_async_handler(netmsg_t msg)
488 sofree(msg->base.nm_so);
492 soclose_sofree_async(struct socket *so)
494 struct netmsg_base *base = &so->so_clomsg;
496 netmsg_init(base, so, &netisr_apanic_rport, 0,
497 soclose_sofree_async_handler);
498 lwkt_sendmsg(so->so_port, &base->lmsg);
502 soclose_disconn_async_handler(netmsg_t msg)
504 struct socket *so = msg->base.nm_so;
506 if ((so->so_state & SS_ISCONNECTED) &&
507 (so->so_state & SS_ISDISCONNECTING) == 0)
508 so_pru_disconnect_direct(so);
511 so_pru_detach_direct(so);
518 soclose_disconn_async(struct socket *so)
520 struct netmsg_base *base = &so->so_clomsg;
522 netmsg_init(base, so, &netisr_apanic_rport, 0,
523 soclose_disconn_async_handler);
524 lwkt_sendmsg(so->so_port, &base->lmsg);
528 soclose_detach_async_handler(netmsg_t msg)
530 struct socket *so = msg->base.nm_so;
533 so_pru_detach_direct(so);
540 soclose_detach_async(struct socket *so)
542 struct netmsg_base *base = &so->so_clomsg;
544 netmsg_init(base, so, &netisr_apanic_rport, 0,
545 soclose_detach_async_handler);
546 lwkt_sendmsg(so->so_port, &base->lmsg);
550 soclose_fast(struct socket *so)
552 if (so->so_pcb == NULL)
555 if ((so->so_state & SS_ISCONNECTED) &&
556 (so->so_state & SS_ISDISCONNECTING) == 0) {
557 soclose_disconn_async(so);
562 soclose_detach_async(so);
568 soclose_sofree_async(so);
572 * Abort and destroy a socket. Only one abort can be in progress
573 * at any given moment.
576 soabort(struct socket *so)
583 soaborta(struct socket *so)
590 soabort_oncpu(struct socket *so)
593 so_pru_abort_oncpu(so);
597 * so is passed in ref'd, which becomes owned by
598 * the cleared SS_NOFDREF flag.
601 soaccept(struct socket *so, struct sockaddr **nam)
605 if ((so->so_state & SS_NOFDREF) == 0)
606 panic("soaccept: !NOFDREF");
607 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */
608 error = so_pru_accept_direct(so, nam);
613 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
617 if (so->so_options & SO_ACCEPTCONN)
620 * If protocol is connection-based, can only connect once.
621 * Otherwise, if connected, try to disconnect first.
622 * This allows user to disconnect by connecting to, e.g.,
625 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
626 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
627 (error = sodisconnect(so)))) {
631 * Prevent accumulated error from previous connection
635 error = so_pru_connect(so, nam, td);
641 soconnect2(struct socket *so1, struct socket *so2)
645 error = so_pru_connect2(so1, so2);
650 sodisconnect(struct socket *so)
654 if ((so->so_state & SS_ISCONNECTED) == 0) {
658 if (so->so_state & SS_ISDISCONNECTING) {
662 error = so_pru_disconnect(so);
667 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
670 * If send must go all at once and message is larger than
671 * send buffering, then hard error.
672 * Lock against other senders.
673 * If must go all at once and not enough room now, then
674 * inform user that this would block and do nothing.
675 * Otherwise, if nonblocking, send as much as possible.
676 * The data to be sent is described by "uio" if nonzero,
677 * otherwise by the mbuf chain "top" (which must be null
678 * if uio is not). Data provided in mbuf chain must be small
679 * enough to send all at once.
681 * Returns nonzero on error, timeout or signal; callers
682 * must check for short counts if EINTR/ERESTART are returned.
683 * Data and control buffers are freed on return.
686 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
687 struct mbuf *top, struct mbuf *control, int flags,
694 int clen = 0, error, dontroute, mlen;
695 int atomic = sosendallatonce(so) || top;
699 resid = uio->uio_resid;
701 resid = (size_t)top->m_pkthdr.len;
704 for (m = top; m; m = m->m_next)
706 KKASSERT(top->m_pkthdr.len == len);
711 * WARNING! resid is unsigned, space and len are signed. space
712 * can wind up negative if the sockbuf is overcommitted.
714 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
715 * type sockets since that's an error.
717 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
723 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
724 (so->so_proto->pr_flags & PR_ATOMIC);
725 if (td->td_lwp != NULL)
726 td->td_lwp->lwp_ru.ru_msgsnd++;
728 clen = control->m_len;
729 #define gotoerr(errcode) { error = errcode; goto release; }
732 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
737 if (so->so_state & SS_CANTSENDMORE)
740 error = so->so_error;
744 if ((so->so_state & SS_ISCONNECTED) == 0) {
746 * `sendto' and `sendmsg' is allowed on a connection-
747 * based socket if it supports implied connect.
748 * Return ENOTCONN if not connected and no address is
751 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
752 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
753 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
754 !(resid == 0 && clen != 0))
756 } else if (addr == 0)
757 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
758 ENOTCONN : EDESTADDRREQ);
760 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
761 clen > so->so_snd.ssb_hiwat) {
764 space = ssb_space(&so->so_snd);
767 if ((space < 0 || (size_t)space < resid + clen) && uio &&
768 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
769 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
770 gotoerr(EWOULDBLOCK);
771 ssb_unlock(&so->so_snd);
772 error = ssb_wait(&so->so_snd);
782 * Data is prepackaged in "top".
786 top->m_flags |= M_EOR;
790 m = m_getl((int)resid, MB_WAIT, MT_DATA,
791 top == NULL ? M_PKTHDR : 0, &mlen);
794 m->m_pkthdr.rcvif = NULL;
796 len = imin((int)szmin(mlen, resid), space);
797 if (resid < MINCLSIZE) {
799 * For datagram protocols, leave room
800 * for protocol headers in first mbuf.
802 if (atomic && top == 0 && len < mlen)
806 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
807 resid = uio->uio_resid;
810 top->m_pkthdr.len += len;
816 top->m_flags |= M_EOR;
819 } while (space > 0 && atomic);
821 so->so_options |= SO_DONTROUTE;
822 if (flags & MSG_OOB) {
823 pru_flags = PRUS_OOB;
824 } else if ((flags & MSG_EOF) &&
825 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
828 * If the user set MSG_EOF, the protocol
829 * understands this flag and nothing left to
830 * send then use PRU_SEND_EOF instead of PRU_SEND.
832 pru_flags = PRUS_EOF;
833 } else if (resid > 0 && space > 0) {
834 /* If there is more to send, set PRUS_MORETOCOME */
835 pru_flags = PRUS_MORETOCOME;
840 * XXX all the SS_CANTSENDMORE checks previously
841 * done could be out of date. We could have recieved
842 * a reset packet in an interrupt or maybe we slept
843 * while doing page faults in uiomove() etc. We could
844 * probably recheck again inside the splnet() protection
845 * here, but there are probably other places that this
846 * also happens. We must rethink this.
848 error = so_pru_send(so, pru_flags, top, addr, control, td);
850 so->so_options &= ~SO_DONTROUTE;
857 } while (resid && space > 0);
861 ssb_unlock(&so->so_snd);
871 * A specialization of sosend() for UDP based on protocol-specific knowledge:
872 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
873 * sosendallatonce() returns true,
874 * the "atomic" variable is true,
875 * and sosendudp() blocks until space is available for the entire send.
876 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
877 * PR_IMPLOPCL flags set.
878 * UDP has no out-of-band data.
879 * UDP has no control data.
880 * UDP does not support MSG_EOR.
883 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
884 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
886 boolean_t dontroute; /* temporary SO_DONTROUTE setting */
891 if (td->td_lwp != NULL)
892 td->td_lwp->lwp_ru.ru_msgsnd++;
896 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
897 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
900 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
904 if (so->so_state & SS_CANTSENDMORE)
907 error = so->so_error;
911 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
912 gotoerr(EDESTADDRREQ);
913 if (resid > so->so_snd.ssb_hiwat)
915 space = ssb_space(&so->so_snd);
916 if (uio && (space < 0 || (size_t)space < resid)) {
917 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
918 gotoerr(EWOULDBLOCK);
919 ssb_unlock(&so->so_snd);
920 error = ssb_wait(&so->so_snd);
927 top = m_uiomove(uio);
932 dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
934 so->so_options |= SO_DONTROUTE;
936 error = so_pru_send(so, 0, top, addr, NULL, td);
937 top = NULL; /* sent or freed in lower layer */
940 so->so_options &= ~SO_DONTROUTE;
943 ssb_unlock(&so->so_snd);
951 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
952 struct mbuf *top, struct mbuf *control, int flags,
964 KKASSERT(top == NULL);
966 resid = uio->uio_resid;
969 resid = (size_t)top->m_pkthdr.len;
972 for (m = top; m; m = m->m_next)
974 KKASSERT(top->m_pkthdr.len == len);
979 * WARNING! resid is unsigned, space and len are signed. space
980 * can wind up negative if the sockbuf is overcommitted.
982 * Also check to make sure that MSG_EOR isn't used on TCP
984 if (flags & MSG_EOR) {
990 /* TCP doesn't do control messages (rights, creds, etc) */
991 if (control->m_len) {
995 m_freem(control); /* empty control, just free it */
999 if (td->td_lwp != NULL)
1000 td->td_lwp->lwp_ru.ru_msgsnd++;
1002 #define gotoerr(errcode) { error = errcode; goto release; }
1005 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1010 if (so->so_state & SS_CANTSENDMORE)
1013 error = so->so_error;
1017 if ((so->so_state & SS_ISCONNECTED) == 0 &&
1018 (so->so_state & SS_ISCONFIRMING) == 0)
1020 if (allatonce && resid > so->so_snd.ssb_hiwat)
1023 space = ssb_space(&so->so_snd);
1024 if (flags & MSG_OOB)
1026 if ((space < 0 || (size_t)space < resid) && !allatonce &&
1027 space < so->so_snd.ssb_lowat) {
1028 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1029 gotoerr(EWOULDBLOCK);
1030 ssb_unlock(&so->so_snd);
1031 error = ssb_wait(&so->so_snd);
1038 int cnt = 0, async = 0;
1042 * Data is prepackaged in "top".
1046 if (resid > INT_MAX)
1048 m = m_getl((int)resid, MB_WAIT, MT_DATA,
1049 top == NULL ? M_PKTHDR : 0, &mlen);
1051 m->m_pkthdr.len = 0;
1052 m->m_pkthdr.rcvif = NULL;
1054 len = imin((int)szmin(mlen, resid), space);
1056 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1057 resid = uio->uio_resid;
1060 top->m_pkthdr.len += len;
1067 } while (space > 0 && cnt < tcp_sosnd_agglim);
1069 if (tcp_sosnd_async)
1072 if (flags & MSG_OOB) {
1073 pru_flags = PRUS_OOB;
1075 } else if ((flags & MSG_EOF) && resid == 0) {
1076 pru_flags = PRUS_EOF;
1077 } else if (resid > 0 && space > 0) {
1078 /* If there is more to send, set PRUS_MORETOCOME */
1079 pru_flags = PRUS_MORETOCOME;
1085 if (flags & MSG_SYNC)
1089 * XXX all the SS_CANTSENDMORE checks previously
1090 * done could be out of date. We could have recieved
1091 * a reset packet in an interrupt or maybe we slept
1092 * while doing page faults in uiomove() etc. We could
1093 * probably recheck again inside the splnet() protection
1094 * here, but there are probably other places that this
1095 * also happens. We must rethink this.
1098 error = so_pru_send(so, pru_flags, top,
1101 so_pru_send_async(so, pru_flags, top,
1110 } while (resid && space > 0);
1114 ssb_unlock(&so->so_snd);
1124 * Implement receive operations on a socket.
1126 * We depend on the way that records are added to the signalsockbuf
1127 * by sbappend*. In particular, each record (mbufs linked through m_next)
1128 * must begin with an address if the protocol so specifies,
1129 * followed by an optional mbuf or mbufs containing ancillary data,
1130 * and then zero or more mbufs of data.
1132 * Although the signalsockbuf is locked, new data may still be appended.
1133 * A token inside the ssb_lock deals with MP issues and still allows
1134 * the network to access the socket if we block in a uio.
1136 * The caller may receive the data as a single mbuf chain by supplying
1137 * an mbuf **mp0 for use in returning the chain. The uio is then used
1138 * only for the count in uio_resid.
1141 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1142 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1145 struct mbuf *free_chain = NULL;
1146 int flags, len, error, offset;
1147 struct protosw *pr = so->so_proto;
1149 size_t resid, orig_resid;
1152 resid = uio->uio_resid;
1154 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1162 flags = *flagsp &~ MSG_EOR;
1165 if (flags & MSG_OOB) {
1166 m = m_get(MB_WAIT, MT_DATA);
1169 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1175 KKASSERT(resid >= (size_t)m->m_len);
1176 resid -= (size_t)m->m_len;
1177 } while (resid > 0 && m);
1180 uio->uio_resid = resid;
1181 error = uiomove(mtod(m, caddr_t),
1182 (int)szmin(resid, m->m_len),
1184 resid = uio->uio_resid;
1186 } while (uio->uio_resid && error == 0 && m);
1193 if ((so->so_state & SS_ISCONFIRMING) && resid)
1197 * The token interlocks against the protocol thread while
1198 * ssb_lock is a blocking lock against other userland entities.
1200 lwkt_gettoken(&so->so_rcv.ssb_token);
1202 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1206 m = so->so_rcv.ssb_mb;
1208 * If we have less data than requested, block awaiting more
1209 * (subject to any timeout) if:
1210 * 1. the current count is less than the low water mark, or
1211 * 2. MSG_WAITALL is set, and it is possible to do the entire
1212 * receive operation at once if we block (resid <= hiwat).
1213 * 3. MSG_DONTWAIT is not set
1214 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1215 * we have to do the receive in sections, and thus risk returning
1216 * a short count if a timeout or signal occurs after we start.
1218 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1219 (size_t)so->so_rcv.ssb_cc < resid) &&
1220 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1221 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1222 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1223 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1227 error = so->so_error;
1228 if ((flags & MSG_PEEK) == 0)
1232 if (so->so_state & SS_CANTRCVMORE) {
1238 for (; m; m = m->m_next) {
1239 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1240 m = so->so_rcv.ssb_mb;
1244 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1245 (pr->pr_flags & PR_CONNREQUIRED)) {
1251 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1252 error = EWOULDBLOCK;
1255 ssb_unlock(&so->so_rcv);
1256 error = ssb_wait(&so->so_rcv);
1262 if (uio && uio->uio_td && uio->uio_td->td_proc)
1263 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1266 * note: m should be == sb_mb here. Cache the next record while
1267 * cleaning up. Note that calling m_free*() will break out critical
1270 KKASSERT(m == so->so_rcv.ssb_mb);
1273 * Skip any address mbufs prepending the record.
1275 if (pr->pr_flags & PR_ADDR) {
1276 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1279 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1280 if (flags & MSG_PEEK)
1283 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1287 * Skip any control mbufs prepending the record.
1290 if (pr->pr_flags & PR_ADDR_OPT) {
1292 * For SCTP we may be getting a
1293 * whole message OR a partial delivery.
1295 if (m && m->m_type == MT_SONAME) {
1298 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1299 if (flags & MSG_PEEK)
1302 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1306 while (m && m->m_type == MT_CONTROL && error == 0) {
1307 if (flags & MSG_PEEK) {
1309 *controlp = m_copy(m, 0, m->m_len);
1310 m = m->m_next; /* XXX race */
1313 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1314 if (pr->pr_domain->dom_externalize &&
1315 mtod(m, struct cmsghdr *)->cmsg_type ==
1317 error = (*pr->pr_domain->dom_externalize)(m);
1321 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1324 if (controlp && *controlp) {
1326 controlp = &(*controlp)->m_next;
1335 if (type == MT_OOBDATA)
1340 * Copy to the UIO or mbuf return chain (*mp).
1344 while (m && resid > 0 && error == 0) {
1345 if (m->m_type == MT_OOBDATA) {
1346 if (type != MT_OOBDATA)
1348 } else if (type == MT_OOBDATA)
1351 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1353 soclrstate(so, SS_RCVATMARK);
1354 len = (resid > INT_MAX) ? INT_MAX : resid;
1355 if (so->so_oobmark && len > so->so_oobmark - offset)
1356 len = so->so_oobmark - offset;
1357 if (len > m->m_len - moff)
1358 len = m->m_len - moff;
1361 * Copy out to the UIO or pass the mbufs back to the SIO.
1362 * The SIO is dealt with when we eat the mbuf, but deal
1363 * with the resid here either way.
1366 uio->uio_resid = resid;
1367 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1368 resid = uio->uio_resid;
1372 resid -= (size_t)len;
1376 * Eat the entire mbuf or just a piece of it
1378 if (len == m->m_len - moff) {
1379 if (m->m_flags & M_EOR)
1382 if (m->m_flags & M_NOTIFICATION)
1383 flags |= MSG_NOTIFICATION;
1385 if (flags & MSG_PEEK) {
1390 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1394 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1398 if (flags & MSG_PEEK) {
1402 n = m_copym(m, 0, len, MB_WAIT);
1408 so->so_rcv.ssb_cc -= len;
1411 if (so->so_oobmark) {
1412 if ((flags & MSG_PEEK) == 0) {
1413 so->so_oobmark -= len;
1414 if (so->so_oobmark == 0) {
1415 sosetstate(so, SS_RCVATMARK);
1420 if (offset == so->so_oobmark)
1424 if (flags & MSG_EOR)
1427 * If the MSG_WAITALL flag is set (for non-atomic socket),
1428 * we must not quit until resid == 0 or an error
1429 * termination. If a signal/timeout occurs, return
1430 * with a short count but without error.
1431 * Keep signalsockbuf locked against other readers.
1433 while ((flags & MSG_WAITALL) && m == NULL &&
1434 resid > 0 && !sosendallatonce(so) &&
1435 so->so_rcv.ssb_mb == NULL) {
1436 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1439 * The window might have closed to zero, make
1440 * sure we send an ack now that we've drained
1441 * the buffer or we might end up blocking until
1442 * the idle takes over (5 seconds).
1444 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1445 so_pru_rcvd(so, flags);
1446 error = ssb_wait(&so->so_rcv);
1448 ssb_unlock(&so->so_rcv);
1452 m = so->so_rcv.ssb_mb;
1457 * If an atomic read was requested but unread data still remains
1458 * in the record, set MSG_TRUNC.
1460 if (m && pr->pr_flags & PR_ATOMIC)
1464 * Cleanup. If an atomic read was requested drop any unread data.
1466 if ((flags & MSG_PEEK) == 0) {
1467 if (m && (pr->pr_flags & PR_ATOMIC))
1468 sbdroprecord(&so->so_rcv.sb);
1469 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1470 so_pru_rcvd(so, flags);
1473 if (orig_resid == resid && orig_resid &&
1474 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1475 ssb_unlock(&so->so_rcv);
1482 ssb_unlock(&so->so_rcv);
1484 lwkt_reltoken(&so->so_rcv.ssb_token);
1486 m_freem(free_chain);
1491 * Shut a socket down. Note that we do not get a frontend lock as we
1492 * want to be able to shut the socket down even if another thread is
1493 * blocked in a read(), thus waking it up.
1496 soshutdown(struct socket *so, int how)
1498 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1501 if (how != SHUT_WR) {
1502 /*ssb_lock(&so->so_rcv, M_WAITOK);*/
1504 /*ssb_unlock(&so->so_rcv);*/
1507 return (so_pru_shutdown(so));
1512 sorflush(struct socket *so)
1514 struct signalsockbuf *ssb = &so->so_rcv;
1515 struct protosw *pr = so->so_proto;
1516 struct signalsockbuf asb;
1518 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1520 lwkt_gettoken(&ssb->ssb_token);
1525 * Can't just blow up the ssb structure here
1527 bzero(&ssb->sb, sizeof(ssb->sb));
1532 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1534 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1535 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1536 ssb_release(&asb, so);
1538 lwkt_reltoken(&ssb->ssb_token);
1543 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1545 struct accept_filter_arg *afap = NULL;
1546 struct accept_filter *afp;
1547 struct so_accf *af = so->so_accf;
1550 /* do not set/remove accept filters on non listen sockets */
1551 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1556 /* removing the filter */
1559 if (af->so_accept_filter != NULL &&
1560 af->so_accept_filter->accf_destroy != NULL) {
1561 af->so_accept_filter->accf_destroy(so);
1563 if (af->so_accept_filter_str != NULL) {
1564 FREE(af->so_accept_filter_str, M_ACCF);
1569 so->so_options &= ~SO_ACCEPTFILTER;
1572 /* adding a filter */
1573 /* must remove previous filter first */
1578 /* don't put large objects on the kernel stack */
1579 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1580 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1581 afap->af_name[sizeof(afap->af_name)-1] = '\0';
1582 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1585 afp = accept_filt_get(afap->af_name);
1590 MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1591 if (afp->accf_create != NULL) {
1592 if (afap->af_name[0] != '\0') {
1593 int len = strlen(afap->af_name) + 1;
1595 MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1596 strcpy(af->so_accept_filter_str, afap->af_name);
1598 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1599 if (af->so_accept_filter_arg == NULL) {
1600 FREE(af->so_accept_filter_str, M_ACCF);
1607 af->so_accept_filter = afp;
1609 so->so_options |= SO_ACCEPTFILTER;
1618 * Perhaps this routine, and sooptcopyout(), below, ought to come in
1619 * an additional variant to handle the case where the option value needs
1620 * to be some kind of integer, but not a specific size.
1621 * In addition to their use here, these functions are also called by the
1622 * protocol-level pr_ctloutput() routines.
1625 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1627 return soopt_to_kbuf(sopt, buf, len, minlen);
1631 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1635 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1636 KKASSERT(kva_p(buf));
1639 * If the user gives us more than we wanted, we ignore it,
1640 * but if we don't get the minimum length the caller
1641 * wants, we return EINVAL. On success, sopt->sopt_valsize
1642 * is set to however much we actually retrieved.
1644 if ((valsize = sopt->sopt_valsize) < minlen)
1647 sopt->sopt_valsize = valsize = len;
1649 bcopy(sopt->sopt_val, buf, valsize);
1655 sosetopt(struct socket *so, struct sockopt *sopt)
1661 struct signalsockbuf *sotmp;
1664 sopt->sopt_dir = SOPT_SET;
1665 if (sopt->sopt_level != SOL_SOCKET) {
1666 if (so->so_proto && so->so_proto->pr_ctloutput) {
1667 return (so_pr_ctloutput(so, sopt));
1669 error = ENOPROTOOPT;
1671 switch (sopt->sopt_name) {
1673 case SO_ACCEPTFILTER:
1674 error = do_setopt_accept_filter(so, sopt);
1680 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1684 so->so_linger = l.l_linger;
1686 so->so_options |= SO_LINGER;
1688 so->so_options &= ~SO_LINGER;
1694 case SO_USELOOPBACK:
1700 error = sooptcopyin(sopt, &optval, sizeof optval,
1705 so->so_options |= sopt->sopt_name;
1707 so->so_options &= ~sopt->sopt_name;
1714 error = sooptcopyin(sopt, &optval, sizeof optval,
1720 * Values < 1 make no sense for any of these
1721 * options, so disallow them.
1728 switch (sopt->sopt_name) {
1731 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1732 &so->so_snd : &so->so_rcv, (u_long)optval,
1734 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1738 sotmp = (sopt->sopt_name == SO_SNDBUF) ?
1739 &so->so_snd : &so->so_rcv;
1740 atomic_clear_int(&sotmp->ssb_flags,
1745 * Make sure the low-water is never greater than
1749 so->so_snd.ssb_lowat =
1750 (optval > so->so_snd.ssb_hiwat) ?
1751 so->so_snd.ssb_hiwat : optval;
1752 atomic_clear_int(&so->so_snd.ssb_flags,
1756 so->so_rcv.ssb_lowat =
1757 (optval > so->so_rcv.ssb_hiwat) ?
1758 so->so_rcv.ssb_hiwat : optval;
1759 atomic_clear_int(&so->so_rcv.ssb_flags,
1767 error = sooptcopyin(sopt, &tv, sizeof tv,
1772 /* assert(hz > 0); */
1773 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
1774 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1778 /* assert(tick > 0); */
1779 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
1780 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
1781 if (val > INT_MAX) {
1785 if (val == 0 && tv.tv_usec != 0)
1788 switch (sopt->sopt_name) {
1790 so->so_snd.ssb_timeo = val;
1793 so->so_rcv.ssb_timeo = val;
1798 error = ENOPROTOOPT;
1801 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1802 (void) so_pr_ctloutput(so, sopt);
1809 /* Helper routine for getsockopt */
1811 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1813 soopt_from_kbuf(sopt, buf, len);
1818 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1823 sopt->sopt_valsize = 0;
1827 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1828 KKASSERT(kva_p(buf));
1831 * Documented get behavior is that we always return a value,
1832 * possibly truncated to fit in the user's buffer.
1833 * Traditional behavior is that we always tell the user
1834 * precisely how much we copied, rather than something useful
1835 * like the total amount we had available for her.
1836 * Note that this interface is not idempotent; the entire answer must
1837 * generated ahead of time.
1839 valsize = szmin(len, sopt->sopt_valsize);
1840 sopt->sopt_valsize = valsize;
1841 if (sopt->sopt_val != 0) {
1842 bcopy(buf, sopt->sopt_val, valsize);
1847 sogetopt(struct socket *so, struct sockopt *sopt)
1854 struct accept_filter_arg *afap;
1858 sopt->sopt_dir = SOPT_GET;
1859 if (sopt->sopt_level != SOL_SOCKET) {
1860 if (so->so_proto && so->so_proto->pr_ctloutput) {
1861 return (so_pr_ctloutput(so, sopt));
1863 return (ENOPROTOOPT);
1865 switch (sopt->sopt_name) {
1867 case SO_ACCEPTFILTER:
1868 if ((so->so_options & SO_ACCEPTCONN) == 0)
1870 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1871 M_TEMP, M_WAITOK | M_ZERO);
1872 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1873 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1874 if (so->so_accf->so_accept_filter_str != NULL)
1875 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1877 error = sooptcopyout(sopt, afap, sizeof(*afap));
1883 l.l_onoff = so->so_options & SO_LINGER;
1884 l.l_linger = so->so_linger;
1885 error = sooptcopyout(sopt, &l, sizeof l);
1888 case SO_USELOOPBACK:
1897 optval = so->so_options & sopt->sopt_name;
1899 error = sooptcopyout(sopt, &optval, sizeof optval);
1903 optval = so->so_type;
1907 optval = so->so_error;
1912 optval = so->so_snd.ssb_hiwat;
1916 optval = so->so_rcv.ssb_hiwat;
1920 optval = so->so_snd.ssb_lowat;
1924 optval = so->so_rcv.ssb_lowat;
1929 optval = (sopt->sopt_name == SO_SNDTIMEO ?
1930 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1932 tv.tv_sec = optval / hz;
1933 tv.tv_usec = (optval % hz) * ustick;
1934 error = sooptcopyout(sopt, &tv, sizeof tv);
1938 optval_l = ssb_space(&so->so_snd);
1939 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
1943 error = ENOPROTOOPT;
1950 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1952 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1954 struct mbuf *m, *m_prev;
1955 int sopt_size = sopt->sopt_valsize, msize;
1957 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1961 m->m_len = min(msize, sopt_size);
1962 sopt_size -= m->m_len;
1966 while (sopt_size > 0) {
1967 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1968 MT_DATA, 0, &msize);
1973 m->m_len = min(msize, sopt_size);
1974 sopt_size -= m->m_len;
1981 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1983 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1985 soopt_to_mbuf(sopt, m);
1990 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
1995 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1997 if (sopt->sopt_val == NULL)
1999 val = sopt->sopt_val;
2000 valsize = sopt->sopt_valsize;
2001 while (m != NULL && valsize >= m->m_len) {
2002 bcopy(val, mtod(m, char *), m->m_len);
2003 valsize -= m->m_len;
2004 val = (caddr_t)val + m->m_len;
2007 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2008 panic("ip6_sooptmcopyin");
2011 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2013 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2015 return soopt_from_mbuf(sopt, m);
2019 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2021 struct mbuf *m0 = m;
2026 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2028 if (sopt->sopt_val == NULL)
2030 val = sopt->sopt_val;
2031 maxsize = sopt->sopt_valsize;
2032 while (m != NULL && maxsize >= m->m_len) {
2033 bcopy(mtod(m, char *), val, m->m_len);
2034 maxsize -= m->m_len;
2035 val = (caddr_t)val + m->m_len;
2036 valsize += m->m_len;
2040 /* enough soopt buffer should be given from user-land */
2044 sopt->sopt_valsize = valsize;
2049 sohasoutofband(struct socket *so)
2051 if (so->so_sigio != NULL)
2052 pgsigio(so->so_sigio, SIGURG, 0);
2053 KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
2057 sokqfilter(struct file *fp, struct knote *kn)
2059 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2060 struct signalsockbuf *ssb;
2062 switch (kn->kn_filter) {
2064 if (so->so_options & SO_ACCEPTCONN)
2065 kn->kn_fop = &solisten_filtops;
2067 kn->kn_fop = &soread_filtops;
2071 kn->kn_fop = &sowrite_filtops;
2075 kn->kn_fop = &soexcept_filtops;
2079 return (EOPNOTSUPP);
2082 knote_insert(&ssb->ssb_kq.ki_note, kn);
2083 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2088 filt_sordetach(struct knote *kn)
2090 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2092 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2093 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2094 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2099 filt_soread(struct knote *kn, long hint)
2101 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2103 if (kn->kn_sfflags & NOTE_OOB) {
2104 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2105 kn->kn_fflags |= NOTE_OOB;
2110 kn->kn_data = so->so_rcv.ssb_cc;
2112 if (so->so_state & SS_CANTRCVMORE) {
2114 * Only set NODATA if all data has been exhausted.
2116 if (kn->kn_data == 0)
2117 kn->kn_flags |= EV_NODATA;
2118 kn->kn_flags |= EV_EOF;
2119 kn->kn_fflags = so->so_error;
2122 if (so->so_error) /* temporary udp error */
2124 if (kn->kn_sfflags & NOTE_LOWAT)
2125 return (kn->kn_data >= kn->kn_sdata);
2126 return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2127 !TAILQ_EMPTY(&so->so_comp));
2131 filt_sowdetach(struct knote *kn)
2133 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2135 knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2136 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2137 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2142 filt_sowrite(struct knote *kn, long hint)
2144 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2146 kn->kn_data = ssb_space(&so->so_snd);
2147 if (so->so_state & SS_CANTSENDMORE) {
2148 kn->kn_flags |= (EV_EOF | EV_NODATA);
2149 kn->kn_fflags = so->so_error;
2152 if (so->so_error) /* temporary udp error */
2154 if (((so->so_state & SS_ISCONNECTED) == 0) &&
2155 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2157 if (kn->kn_sfflags & NOTE_LOWAT)
2158 return (kn->kn_data >= kn->kn_sdata);
2159 return (kn->kn_data >= so->so_snd.ssb_lowat);
2164 filt_solisten(struct knote *kn, long hint)
2166 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2168 kn->kn_data = so->so_qlen;
2169 return (! TAILQ_EMPTY(&so->so_comp));