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;
103 extern int udp_sosnd_async;
106 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
109 static void filt_sordetach(struct knote *kn);
110 static int filt_soread(struct knote *kn, long hint);
111 static void filt_sowdetach(struct knote *kn);
112 static int filt_sowrite(struct knote *kn, long hint);
113 static int filt_solisten(struct knote *kn, long hint);
115 static void sodiscard(struct socket *so);
116 static int soclose_sync(struct socket *so, int fflag);
117 static void soclose_fast(struct socket *so);
119 static struct filterops solisten_filtops =
120 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten };
121 static struct filterops soread_filtops =
122 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
123 static struct filterops sowrite_filtops =
124 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite };
125 static struct filterops soexcept_filtops =
126 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
128 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
129 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
130 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
133 static int somaxconn = SOMAXCONN;
134 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
135 &somaxconn, 0, "Maximum pending socket connection queue size");
137 static int use_soclose_fast = 1;
138 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW,
139 &use_soclose_fast, 0, "Fast socket close");
141 int use_soaccept_pred_fast = 1;
142 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW,
143 &use_soaccept_pred_fast, 0, "Fast socket accept predication");
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, and initialize it.
155 * Note that it would probably be better to allocate socket
156 * and PCB at the same time, but I'm not convinced that all
157 * the protocols can be easily modified to do this.
165 waitmask = waitok ? M_WAITOK : M_NOWAIT;
166 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
168 /* XXX race condition for reentrant kernel */
169 TAILQ_INIT(&so->so_aiojobq);
170 TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist);
171 TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist);
172 lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
173 lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
174 so->so_state = SS_NOFDREF;
181 socreate(int dom, struct socket **aso, int type,
182 int proto, struct thread *td)
184 struct proc *p = td->td_proc;
187 struct pru_attach_info ai;
191 prp = pffindproto(dom, proto, type);
193 prp = pffindtype(dom, type);
195 if (prp == NULL || prp->pr_usrreqs->pru_attach == 0)
196 return (EPROTONOSUPPORT);
198 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
199 prp->pr_domain->dom_family != PF_LOCAL &&
200 prp->pr_domain->dom_family != PF_INET &&
201 prp->pr_domain->dom_family != PF_INET6 &&
202 prp->pr_domain->dom_family != PF_ROUTE) {
203 return (EPROTONOSUPPORT);
206 if (prp->pr_type != type)
208 so = soalloc(p != NULL);
213 * Callers of socreate() presumably will connect up a descriptor
214 * and call soclose() if they cannot. This represents our so_refs
215 * (which should be 1) from soalloc().
217 soclrstate(so, SS_NOFDREF);
220 * Set a default port for protocol processing. No action will occur
221 * on the socket on this port until an inpcb is attached to it and
222 * is able to match incoming packets, or until the socket becomes
223 * available to userland.
225 * We normally default the socket to the protocol thread on cpu 0.
226 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
227 * thread and all pr_*()/pru_*() calls are executed synchronously.
229 if (prp->pr_flags & PR_SYNC_PORT)
230 so->so_port = &netisr_sync_port;
232 so->so_port = cpu_portfn(0);
234 TAILQ_INIT(&so->so_incomp);
235 TAILQ_INIT(&so->so_comp);
237 so->so_cred = crhold(p->p_ucred);
239 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
240 ai.p_ucred = p->p_ucred;
241 ai.fd_rdir = p->p_fd->fd_rdir;
244 * Auto-sizing of socket buffers is managed by the protocols and
245 * the appropriate flags must be set in the pru_attach function.
247 error = so_pru_attach(so, proto, &ai);
249 sosetstate(so, SS_NOFDREF);
250 sofree(so); /* from soalloc */
255 * NOTE: Returns referenced socket.
262 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
266 error = so_pru_bind(so, nam, td);
271 sodealloc(struct socket *so)
273 if (so->so_rcv.ssb_hiwat)
274 (void)chgsbsize(so->so_cred->cr_uidinfo,
275 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
276 if (so->so_snd.ssb_hiwat)
277 (void)chgsbsize(so->so_cred->cr_uidinfo,
278 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
280 /* remove accept filter if present */
281 if (so->so_accf != NULL)
282 do_setopt_accept_filter(so, NULL);
285 if (so->so_faddr != NULL)
286 kfree(so->so_faddr, M_SONAME);
291 solisten(struct socket *so, int backlog, struct thread *td)
295 short oldopt, oldqlimit;
298 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
302 oldopt = so->so_options;
303 oldqlimit = so->so_qlimit;
306 lwkt_gettoken(&so->so_rcv.ssb_token);
307 if (TAILQ_EMPTY(&so->so_comp))
308 so->so_options |= SO_ACCEPTCONN;
309 lwkt_reltoken(&so->so_rcv.ssb_token);
310 if (backlog < 0 || backlog > somaxconn)
312 so->so_qlimit = backlog;
313 /* SCTP needs to look at tweak both the inbound backlog parameter AND
314 * the so_options (UDP model both connect's and gets inbound
315 * connections .. implicitly).
317 error = so_pru_listen(so, td);
320 /* Restore the params */
321 so->so_options = oldopt;
322 so->so_qlimit = oldqlimit;
330 * Destroy a disconnected socket. This routine is a NOP if entities
331 * still have a reference on the socket:
333 * so_pcb - The protocol stack still has a reference
334 * SS_NOFDREF - There is no longer a file pointer reference
337 sofree(struct socket *so)
342 * This is a bit hackish at the moment. We need to interlock
343 * any accept queue we are on before we potentially lose the
344 * last reference to avoid races against a re-reference from
345 * someone operating on the queue.
347 while ((head = so->so_head) != NULL) {
348 lwkt_getpooltoken(head);
349 if (so->so_head == head)
351 lwkt_relpooltoken(head);
355 * Arbitrage the last free.
357 KKASSERT(so->so_refs > 0);
358 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
360 lwkt_relpooltoken(head);
364 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
365 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
368 * We're done, remove ourselves from the accept queue we are
369 * on, if we are on one.
372 if (so->so_state & SS_INCOMP) {
373 TAILQ_REMOVE(&head->so_incomp, so, so_list);
375 } else if (so->so_state & SS_COMP) {
377 * We must not decommission a socket that's
378 * on the accept(2) queue. If we do, then
379 * accept(2) may hang after select(2) indicated
380 * that the listening socket was ready.
382 lwkt_relpooltoken(head);
385 panic("sofree: not queued");
387 soclrstate(so, SS_INCOMP);
389 lwkt_relpooltoken(head);
391 ssb_release(&so->so_snd, so);
397 * Close a socket on last file table reference removal.
398 * Initiate disconnect if connected.
399 * Free socket when disconnect complete.
402 soclose(struct socket *so, int fflag)
406 funsetown(&so->so_sigio);
407 if (!use_soclose_fast ||
408 (so->so_proto->pr_flags & PR_SYNC_PORT) ||
409 (so->so_options & SO_LINGER)) {
410 error = soclose_sync(so, fflag);
419 sodiscard(struct socket *so)
421 lwkt_getpooltoken(so);
422 if (so->so_options & SO_ACCEPTCONN) {
425 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
426 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
427 soclrstate(sp, SS_INCOMP);
432 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
433 TAILQ_REMOVE(&so->so_comp, sp, so_list);
434 soclrstate(sp, SS_COMP);
440 lwkt_relpooltoken(so);
442 if (so->so_state & SS_NOFDREF)
443 panic("soclose: NOFDREF");
444 sosetstate(so, SS_NOFDREF); /* take ref */
448 soclose_sync(struct socket *so, int fflag)
452 if (so->so_pcb == NULL)
454 if (so->so_state & SS_ISCONNECTED) {
455 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
456 error = sodisconnect(so);
460 if (so->so_options & SO_LINGER) {
461 if ((so->so_state & SS_ISDISCONNECTING) &&
464 while (so->so_state & SS_ISCONNECTED) {
465 error = tsleep(&so->so_timeo, PCATCH,
466 "soclos", so->so_linger * hz);
476 error2 = so_pru_detach(so);
482 so_pru_sync(so); /* unpend async sending */
483 sofree(so); /* dispose of ref */
489 soclose_sofree_async_handler(netmsg_t msg)
491 sofree(msg->base.nm_so);
495 soclose_sofree_async(struct socket *so)
497 struct netmsg_base *base = &so->so_clomsg;
499 netmsg_init(base, so, &netisr_apanic_rport, 0,
500 soclose_sofree_async_handler);
501 lwkt_sendmsg(so->so_port, &base->lmsg);
505 soclose_disconn_async_handler(netmsg_t msg)
507 struct socket *so = msg->base.nm_so;
509 if ((so->so_state & SS_ISCONNECTED) &&
510 (so->so_state & SS_ISDISCONNECTING) == 0)
511 so_pru_disconnect_direct(so);
514 so_pru_detach_direct(so);
521 soclose_disconn_async(struct socket *so)
523 struct netmsg_base *base = &so->so_clomsg;
525 netmsg_init(base, so, &netisr_apanic_rport, 0,
526 soclose_disconn_async_handler);
527 lwkt_sendmsg(so->so_port, &base->lmsg);
531 soclose_detach_async_handler(netmsg_t msg)
533 struct socket *so = msg->base.nm_so;
536 so_pru_detach_direct(so);
543 soclose_detach_async(struct socket *so)
545 struct netmsg_base *base = &so->so_clomsg;
547 netmsg_init(base, so, &netisr_apanic_rport, 0,
548 soclose_detach_async_handler);
549 lwkt_sendmsg(so->so_port, &base->lmsg);
553 soclose_fast(struct socket *so)
555 if (so->so_pcb == NULL)
558 if ((so->so_state & SS_ISCONNECTED) &&
559 (so->so_state & SS_ISDISCONNECTING) == 0) {
560 soclose_disconn_async(so);
565 soclose_detach_async(so);
571 soclose_sofree_async(so);
575 * Abort and destroy a socket. Only one abort can be in progress
576 * at any given moment.
579 soabort(struct socket *so)
586 soaborta(struct socket *so)
593 soabort_oncpu(struct socket *so)
596 so_pru_abort_oncpu(so);
600 * so is passed in ref'd, which becomes owned by
601 * the cleared SS_NOFDREF flag.
604 soaccept_generic(struct socket *so)
606 if ((so->so_state & SS_NOFDREF) == 0)
607 panic("soaccept: !NOFDREF");
608 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */
612 soaccept(struct socket *so, struct sockaddr **nam)
616 soaccept_generic(so);
617 error = so_pru_accept(so, nam);
622 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
626 if (so->so_options & SO_ACCEPTCONN)
629 * If protocol is connection-based, can only connect once.
630 * Otherwise, if connected, try to disconnect first.
631 * This allows user to disconnect by connecting to, e.g.,
634 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
635 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
636 (error = sodisconnect(so)))) {
640 * Prevent accumulated error from previous connection
644 error = so_pru_connect(so, nam, td);
650 soconnect2(struct socket *so1, struct socket *so2)
654 error = so_pru_connect2(so1, so2);
659 sodisconnect(struct socket *so)
663 if ((so->so_state & SS_ISCONNECTED) == 0) {
667 if (so->so_state & SS_ISDISCONNECTING) {
671 error = so_pru_disconnect(so);
676 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
679 * If send must go all at once and message is larger than
680 * send buffering, then hard error.
681 * Lock against other senders.
682 * If must go all at once and not enough room now, then
683 * inform user that this would block and do nothing.
684 * Otherwise, if nonblocking, send as much as possible.
685 * The data to be sent is described by "uio" if nonzero,
686 * otherwise by the mbuf chain "top" (which must be null
687 * if uio is not). Data provided in mbuf chain must be small
688 * enough to send all at once.
690 * Returns nonzero on error, timeout or signal; callers
691 * must check for short counts if EINTR/ERESTART are returned.
692 * Data and control buffers are freed on return.
695 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
696 struct mbuf *top, struct mbuf *control, int flags,
703 int clen = 0, error, dontroute, mlen;
704 int atomic = sosendallatonce(so) || top;
708 resid = uio->uio_resid;
710 resid = (size_t)top->m_pkthdr.len;
713 for (m = top; m; m = m->m_next)
715 KKASSERT(top->m_pkthdr.len == len);
720 * WARNING! resid is unsigned, space and len are signed. space
721 * can wind up negative if the sockbuf is overcommitted.
723 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
724 * type sockets since that's an error.
726 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
732 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
733 (so->so_proto->pr_flags & PR_ATOMIC);
734 if (td->td_lwp != NULL)
735 td->td_lwp->lwp_ru.ru_msgsnd++;
737 clen = control->m_len;
738 #define gotoerr(errcode) { error = errcode; goto release; }
741 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
746 if (so->so_state & SS_CANTSENDMORE)
749 error = so->so_error;
753 if ((so->so_state & SS_ISCONNECTED) == 0) {
755 * `sendto' and `sendmsg' is allowed on a connection-
756 * based socket if it supports implied connect.
757 * Return ENOTCONN if not connected and no address is
760 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
761 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
762 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
763 !(resid == 0 && clen != 0))
765 } else if (addr == NULL)
766 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
767 ENOTCONN : EDESTADDRREQ);
769 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
770 clen > so->so_snd.ssb_hiwat) {
773 space = ssb_space(&so->so_snd);
776 if ((space < 0 || (size_t)space < resid + clen) && uio &&
777 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
778 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
779 gotoerr(EWOULDBLOCK);
780 ssb_unlock(&so->so_snd);
781 error = ssb_wait(&so->so_snd);
791 * Data is prepackaged in "top".
795 top->m_flags |= M_EOR;
799 m = m_getl((int)resid, MB_WAIT, MT_DATA,
800 top == NULL ? M_PKTHDR : 0, &mlen);
803 m->m_pkthdr.rcvif = NULL;
805 len = imin((int)szmin(mlen, resid), space);
806 if (resid < MINCLSIZE) {
808 * For datagram protocols, leave room
809 * for protocol headers in first mbuf.
811 if (atomic && top == NULL && len < mlen)
815 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
816 resid = uio->uio_resid;
819 top->m_pkthdr.len += len;
825 top->m_flags |= M_EOR;
828 } while (space > 0 && atomic);
830 so->so_options |= SO_DONTROUTE;
831 if (flags & MSG_OOB) {
832 pru_flags = PRUS_OOB;
833 } else if ((flags & MSG_EOF) &&
834 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
837 * If the user set MSG_EOF, the protocol
838 * understands this flag and nothing left to
839 * send then use PRU_SEND_EOF instead of PRU_SEND.
841 pru_flags = PRUS_EOF;
842 } else if (resid > 0 && space > 0) {
843 /* If there is more to send, set PRUS_MORETOCOME */
844 pru_flags = PRUS_MORETOCOME;
849 * XXX all the SS_CANTSENDMORE checks previously
850 * done could be out of date. We could have recieved
851 * a reset packet in an interrupt or maybe we slept
852 * while doing page faults in uiomove() etc. We could
853 * probably recheck again inside the splnet() protection
854 * here, but there are probably other places that this
855 * also happens. We must rethink this.
857 error = so_pru_send(so, pru_flags, top, addr, control, td);
859 so->so_options &= ~SO_DONTROUTE;
866 } while (resid && space > 0);
870 ssb_unlock(&so->so_snd);
880 * A specialization of sosend() for UDP based on protocol-specific knowledge:
881 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
882 * sosendallatonce() returns true,
883 * the "atomic" variable is true,
884 * and sosendudp() blocks until space is available for the entire send.
885 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
886 * PR_IMPLOPCL flags set.
887 * UDP has no out-of-band data.
888 * UDP has no control data.
889 * UDP does not support MSG_EOR.
892 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
893 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
896 int error, pru_flags = 0;
899 if (td->td_lwp != NULL)
900 td->td_lwp->lwp_ru.ru_msgsnd++;
904 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
905 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
908 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
912 if (so->so_state & SS_CANTSENDMORE)
915 error = so->so_error;
919 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
920 gotoerr(EDESTADDRREQ);
921 if (resid > so->so_snd.ssb_hiwat)
923 space = ssb_space(&so->so_snd);
924 if (uio && (space < 0 || (size_t)space < resid)) {
925 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
926 gotoerr(EWOULDBLOCK);
927 ssb_unlock(&so->so_snd);
928 error = ssb_wait(&so->so_snd);
935 top = m_uiomove(uio);
940 if (flags & MSG_DONTROUTE)
941 pru_flags |= PRUS_DONTROUTE;
943 if (udp_sosnd_async && (flags & MSG_SYNC) == 0) {
944 so_pru_send_async(so, pru_flags, top, addr, NULL, td);
947 error = so_pru_send(so, pru_flags, top, addr, NULL, td);
949 top = NULL; /* sent or freed in lower layer */
952 ssb_unlock(&so->so_snd);
960 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
961 struct mbuf *top, struct mbuf *control, int flags,
973 KKASSERT(top == NULL);
975 resid = uio->uio_resid;
978 resid = (size_t)top->m_pkthdr.len;
981 for (m = top; m; m = m->m_next)
983 KKASSERT(top->m_pkthdr.len == len);
988 * WARNING! resid is unsigned, space and len are signed. space
989 * can wind up negative if the sockbuf is overcommitted.
991 * Also check to make sure that MSG_EOR isn't used on TCP
993 if (flags & MSG_EOR) {
999 /* TCP doesn't do control messages (rights, creds, etc) */
1000 if (control->m_len) {
1004 m_freem(control); /* empty control, just free it */
1008 if (td->td_lwp != NULL)
1009 td->td_lwp->lwp_ru.ru_msgsnd++;
1011 #define gotoerr(errcode) { error = errcode; goto release; }
1014 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1019 if (so->so_state & SS_CANTSENDMORE)
1022 error = so->so_error;
1026 if ((so->so_state & SS_ISCONNECTED) == 0 &&
1027 (so->so_state & SS_ISCONFIRMING) == 0)
1029 if (allatonce && resid > so->so_snd.ssb_hiwat)
1032 space = ssb_space(&so->so_snd);
1033 if (flags & MSG_OOB)
1035 if ((space < 0 || (size_t)space < resid) && !allatonce &&
1036 space < so->so_snd.ssb_lowat) {
1037 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1038 gotoerr(EWOULDBLOCK);
1039 ssb_unlock(&so->so_snd);
1040 error = ssb_wait(&so->so_snd);
1047 int cnt = 0, async = 0;
1051 * Data is prepackaged in "top".
1055 if (resid > INT_MAX)
1057 m = m_getl((int)resid, MB_WAIT, MT_DATA,
1058 top == NULL ? M_PKTHDR : 0, &mlen);
1060 m->m_pkthdr.len = 0;
1061 m->m_pkthdr.rcvif = NULL;
1063 len = imin((int)szmin(mlen, resid), space);
1065 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1066 resid = uio->uio_resid;
1069 top->m_pkthdr.len += len;
1076 } while (space > 0 && cnt < tcp_sosnd_agglim);
1078 if (tcp_sosnd_async)
1081 if (flags & MSG_OOB) {
1082 pru_flags = PRUS_OOB;
1084 } else if ((flags & MSG_EOF) && resid == 0) {
1085 pru_flags = PRUS_EOF;
1086 } else if (resid > 0 && space > 0) {
1087 /* If there is more to send, set PRUS_MORETOCOME */
1088 pru_flags = PRUS_MORETOCOME;
1094 if (flags & MSG_SYNC)
1098 * XXX all the SS_CANTSENDMORE checks previously
1099 * done could be out of date. We could have recieved
1100 * a reset packet in an interrupt or maybe we slept
1101 * while doing page faults in uiomove() etc. We could
1102 * probably recheck again inside the splnet() protection
1103 * here, but there are probably other places that this
1104 * also happens. We must rethink this.
1107 error = so_pru_send(so, pru_flags, top,
1110 so_pru_send_async(so, pru_flags, top,
1119 } while (resid && space > 0);
1123 ssb_unlock(&so->so_snd);
1133 * Implement receive operations on a socket.
1135 * We depend on the way that records are added to the signalsockbuf
1136 * by sbappend*. In particular, each record (mbufs linked through m_next)
1137 * must begin with an address if the protocol so specifies,
1138 * followed by an optional mbuf or mbufs containing ancillary data,
1139 * and then zero or more mbufs of data.
1141 * Although the signalsockbuf is locked, new data may still be appended.
1142 * A token inside the ssb_lock deals with MP issues and still allows
1143 * the network to access the socket if we block in a uio.
1145 * The caller may receive the data as a single mbuf chain by supplying
1146 * an mbuf **mp0 for use in returning the chain. The uio is then used
1147 * only for the count in uio_resid.
1150 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1151 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1154 struct mbuf *free_chain = NULL;
1155 int flags, len, error, offset;
1156 struct protosw *pr = so->so_proto;
1158 size_t resid, orig_resid;
1161 resid = uio->uio_resid;
1163 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1171 flags = *flagsp &~ MSG_EOR;
1174 if (flags & MSG_OOB) {
1175 m = m_get(MB_WAIT, MT_DATA);
1178 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1184 KKASSERT(resid >= (size_t)m->m_len);
1185 resid -= (size_t)m->m_len;
1186 } while (resid > 0 && m);
1189 uio->uio_resid = resid;
1190 error = uiomove(mtod(m, caddr_t),
1191 (int)szmin(resid, m->m_len),
1193 resid = uio->uio_resid;
1195 } while (uio->uio_resid && error == 0 && m);
1202 if ((so->so_state & SS_ISCONFIRMING) && resid)
1206 * The token interlocks against the protocol thread while
1207 * ssb_lock is a blocking lock against other userland entities.
1209 lwkt_gettoken(&so->so_rcv.ssb_token);
1211 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1215 m = so->so_rcv.ssb_mb;
1217 * If we have less data than requested, block awaiting more
1218 * (subject to any timeout) if:
1219 * 1. the current count is less than the low water mark, or
1220 * 2. MSG_WAITALL is set, and it is possible to do the entire
1221 * receive operation at once if we block (resid <= hiwat).
1222 * 3. MSG_DONTWAIT is not set
1223 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1224 * we have to do the receive in sections, and thus risk returning
1225 * a short count if a timeout or signal occurs after we start.
1227 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1228 (size_t)so->so_rcv.ssb_cc < resid) &&
1229 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1230 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1231 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1232 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1236 error = so->so_error;
1237 if ((flags & MSG_PEEK) == 0)
1241 if (so->so_state & SS_CANTRCVMORE) {
1247 for (; m; m = m->m_next) {
1248 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1249 m = so->so_rcv.ssb_mb;
1253 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1254 (pr->pr_flags & PR_CONNREQUIRED)) {
1260 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1261 error = EWOULDBLOCK;
1264 ssb_unlock(&so->so_rcv);
1265 error = ssb_wait(&so->so_rcv);
1271 if (uio && uio->uio_td && uio->uio_td->td_proc)
1272 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1275 * note: m should be == sb_mb here. Cache the next record while
1276 * cleaning up. Note that calling m_free*() will break out critical
1279 KKASSERT(m == so->so_rcv.ssb_mb);
1282 * Skip any address mbufs prepending the record.
1284 if (pr->pr_flags & PR_ADDR) {
1285 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1288 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1289 if (flags & MSG_PEEK)
1292 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1296 * Skip any control mbufs prepending the record.
1299 if (pr->pr_flags & PR_ADDR_OPT) {
1301 * For SCTP we may be getting a
1302 * whole message OR a partial delivery.
1304 if (m && m->m_type == MT_SONAME) {
1307 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1308 if (flags & MSG_PEEK)
1311 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1315 while (m && m->m_type == MT_CONTROL && error == 0) {
1316 if (flags & MSG_PEEK) {
1318 *controlp = m_copy(m, 0, m->m_len);
1319 m = m->m_next; /* XXX race */
1322 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1323 if (pr->pr_domain->dom_externalize &&
1324 mtod(m, struct cmsghdr *)->cmsg_type ==
1326 error = (*pr->pr_domain->dom_externalize)(m);
1330 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1333 if (controlp && *controlp) {
1335 controlp = &(*controlp)->m_next;
1344 if (type == MT_OOBDATA)
1349 * Copy to the UIO or mbuf return chain (*mp).
1353 while (m && resid > 0 && error == 0) {
1354 if (m->m_type == MT_OOBDATA) {
1355 if (type != MT_OOBDATA)
1357 } else if (type == MT_OOBDATA)
1360 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1362 soclrstate(so, SS_RCVATMARK);
1363 len = (resid > INT_MAX) ? INT_MAX : resid;
1364 if (so->so_oobmark && len > so->so_oobmark - offset)
1365 len = so->so_oobmark - offset;
1366 if (len > m->m_len - moff)
1367 len = m->m_len - moff;
1370 * Copy out to the UIO or pass the mbufs back to the SIO.
1371 * The SIO is dealt with when we eat the mbuf, but deal
1372 * with the resid here either way.
1375 uio->uio_resid = resid;
1376 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1377 resid = uio->uio_resid;
1381 resid -= (size_t)len;
1385 * Eat the entire mbuf or just a piece of it
1387 if (len == m->m_len - moff) {
1388 if (m->m_flags & M_EOR)
1391 if (m->m_flags & M_NOTIFICATION)
1392 flags |= MSG_NOTIFICATION;
1394 if (flags & MSG_PEEK) {
1399 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1403 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1407 if (flags & MSG_PEEK) {
1411 n = m_copym(m, 0, len, MB_WAIT);
1417 so->so_rcv.ssb_cc -= len;
1420 if (so->so_oobmark) {
1421 if ((flags & MSG_PEEK) == 0) {
1422 so->so_oobmark -= len;
1423 if (so->so_oobmark == 0) {
1424 sosetstate(so, SS_RCVATMARK);
1429 if (offset == so->so_oobmark)
1433 if (flags & MSG_EOR)
1436 * If the MSG_WAITALL flag is set (for non-atomic socket),
1437 * we must not quit until resid == 0 or an error
1438 * termination. If a signal/timeout occurs, return
1439 * with a short count but without error.
1440 * Keep signalsockbuf locked against other readers.
1442 while ((flags & MSG_WAITALL) && m == NULL &&
1443 resid > 0 && !sosendallatonce(so) &&
1444 so->so_rcv.ssb_mb == NULL) {
1445 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1448 * The window might have closed to zero, make
1449 * sure we send an ack now that we've drained
1450 * the buffer or we might end up blocking until
1451 * the idle takes over (5 seconds).
1453 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1454 so_pru_rcvd(so, flags);
1455 error = ssb_wait(&so->so_rcv);
1457 ssb_unlock(&so->so_rcv);
1461 m = so->so_rcv.ssb_mb;
1466 * If an atomic read was requested but unread data still remains
1467 * in the record, set MSG_TRUNC.
1469 if (m && pr->pr_flags & PR_ATOMIC)
1473 * Cleanup. If an atomic read was requested drop any unread data.
1475 if ((flags & MSG_PEEK) == 0) {
1476 if (m && (pr->pr_flags & PR_ATOMIC))
1477 sbdroprecord(&so->so_rcv.sb);
1478 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1479 so_pru_rcvd(so, flags);
1482 if (orig_resid == resid && orig_resid &&
1483 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1484 ssb_unlock(&so->so_rcv);
1491 ssb_unlock(&so->so_rcv);
1493 lwkt_reltoken(&so->so_rcv.ssb_token);
1495 m_freem(free_chain);
1500 * Shut a socket down. Note that we do not get a frontend lock as we
1501 * want to be able to shut the socket down even if another thread is
1502 * blocked in a read(), thus waking it up.
1505 soshutdown(struct socket *so, int how)
1507 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1510 if (how != SHUT_WR) {
1511 /*ssb_lock(&so->so_rcv, M_WAITOK);*/
1513 /*ssb_unlock(&so->so_rcv);*/
1516 return (so_pru_shutdown(so));
1521 sorflush(struct socket *so)
1523 struct signalsockbuf *ssb = &so->so_rcv;
1524 struct protosw *pr = so->so_proto;
1525 struct signalsockbuf asb;
1527 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1529 lwkt_gettoken(&ssb->ssb_token);
1534 * Can't just blow up the ssb structure here
1536 bzero(&ssb->sb, sizeof(ssb->sb));
1541 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1543 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1544 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1545 ssb_release(&asb, so);
1547 lwkt_reltoken(&ssb->ssb_token);
1552 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1554 struct accept_filter_arg *afap = NULL;
1555 struct accept_filter *afp;
1556 struct so_accf *af = so->so_accf;
1559 /* do not set/remove accept filters on non listen sockets */
1560 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1565 /* removing the filter */
1568 if (af->so_accept_filter != NULL &&
1569 af->so_accept_filter->accf_destroy != NULL) {
1570 af->so_accept_filter->accf_destroy(so);
1572 if (af->so_accept_filter_str != NULL) {
1573 kfree(af->so_accept_filter_str, M_ACCF);
1578 so->so_options &= ~SO_ACCEPTFILTER;
1581 /* adding a filter */
1582 /* must remove previous filter first */
1587 /* don't put large objects on the kernel stack */
1588 afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
1589 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1590 afap->af_name[sizeof(afap->af_name)-1] = '\0';
1591 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1594 afp = accept_filt_get(afap->af_name);
1599 af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1600 if (afp->accf_create != NULL) {
1601 if (afap->af_name[0] != '\0') {
1602 int len = strlen(afap->af_name) + 1;
1604 af->so_accept_filter_str = kmalloc(len, M_ACCF,
1606 strcpy(af->so_accept_filter_str, afap->af_name);
1608 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1609 if (af->so_accept_filter_arg == NULL) {
1610 kfree(af->so_accept_filter_str, M_ACCF);
1617 af->so_accept_filter = afp;
1619 so->so_options |= SO_ACCEPTFILTER;
1622 kfree(afap, M_TEMP);
1628 * Perhaps this routine, and sooptcopyout(), below, ought to come in
1629 * an additional variant to handle the case where the option value needs
1630 * to be some kind of integer, but not a specific size.
1631 * In addition to their use here, these functions are also called by the
1632 * protocol-level pr_ctloutput() routines.
1635 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1637 return soopt_to_kbuf(sopt, buf, len, minlen);
1641 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1645 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1646 KKASSERT(kva_p(buf));
1649 * If the user gives us more than we wanted, we ignore it,
1650 * but if we don't get the minimum length the caller
1651 * wants, we return EINVAL. On success, sopt->sopt_valsize
1652 * is set to however much we actually retrieved.
1654 if ((valsize = sopt->sopt_valsize) < minlen)
1657 sopt->sopt_valsize = valsize = len;
1659 bcopy(sopt->sopt_val, buf, valsize);
1665 sosetopt(struct socket *so, struct sockopt *sopt)
1671 struct signalsockbuf *sotmp;
1674 sopt->sopt_dir = SOPT_SET;
1675 if (sopt->sopt_level != SOL_SOCKET) {
1676 if (so->so_proto && so->so_proto->pr_ctloutput) {
1677 return (so_pr_ctloutput(so, sopt));
1679 error = ENOPROTOOPT;
1681 switch (sopt->sopt_name) {
1683 case SO_ACCEPTFILTER:
1684 error = do_setopt_accept_filter(so, sopt);
1690 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1694 so->so_linger = l.l_linger;
1696 so->so_options |= SO_LINGER;
1698 so->so_options &= ~SO_LINGER;
1704 case SO_USELOOPBACK:
1710 error = sooptcopyin(sopt, &optval, sizeof optval,
1715 so->so_options |= sopt->sopt_name;
1717 so->so_options &= ~sopt->sopt_name;
1724 error = sooptcopyin(sopt, &optval, sizeof optval,
1730 * Values < 1 make no sense for any of these
1731 * options, so disallow them.
1738 switch (sopt->sopt_name) {
1741 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1742 &so->so_snd : &so->so_rcv, (u_long)optval,
1744 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1748 sotmp = (sopt->sopt_name == SO_SNDBUF) ?
1749 &so->so_snd : &so->so_rcv;
1750 atomic_clear_int(&sotmp->ssb_flags,
1755 * Make sure the low-water is never greater than
1759 so->so_snd.ssb_lowat =
1760 (optval > so->so_snd.ssb_hiwat) ?
1761 so->so_snd.ssb_hiwat : optval;
1762 atomic_clear_int(&so->so_snd.ssb_flags,
1766 so->so_rcv.ssb_lowat =
1767 (optval > so->so_rcv.ssb_hiwat) ?
1768 so->so_rcv.ssb_hiwat : optval;
1769 atomic_clear_int(&so->so_rcv.ssb_flags,
1777 error = sooptcopyin(sopt, &tv, sizeof tv,
1782 /* assert(hz > 0); */
1783 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
1784 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1788 /* assert(tick > 0); */
1789 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
1790 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
1791 if (val > INT_MAX) {
1795 if (val == 0 && tv.tv_usec != 0)
1798 switch (sopt->sopt_name) {
1800 so->so_snd.ssb_timeo = val;
1803 so->so_rcv.ssb_timeo = val;
1808 error = ENOPROTOOPT;
1811 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1812 (void) so_pr_ctloutput(so, sopt);
1819 /* Helper routine for getsockopt */
1821 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1823 soopt_from_kbuf(sopt, buf, len);
1828 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1833 sopt->sopt_valsize = 0;
1837 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1838 KKASSERT(kva_p(buf));
1841 * Documented get behavior is that we always return a value,
1842 * possibly truncated to fit in the user's buffer.
1843 * Traditional behavior is that we always tell the user
1844 * precisely how much we copied, rather than something useful
1845 * like the total amount we had available for her.
1846 * Note that this interface is not idempotent; the entire answer must
1847 * generated ahead of time.
1849 valsize = szmin(len, sopt->sopt_valsize);
1850 sopt->sopt_valsize = valsize;
1851 if (sopt->sopt_val != 0) {
1852 bcopy(buf, sopt->sopt_val, valsize);
1857 sogetopt(struct socket *so, struct sockopt *sopt)
1864 struct accept_filter_arg *afap;
1868 sopt->sopt_dir = SOPT_GET;
1869 if (sopt->sopt_level != SOL_SOCKET) {
1870 if (so->so_proto && so->so_proto->pr_ctloutput) {
1871 return (so_pr_ctloutput(so, sopt));
1873 return (ENOPROTOOPT);
1875 switch (sopt->sopt_name) {
1877 case SO_ACCEPTFILTER:
1878 if ((so->so_options & SO_ACCEPTCONN) == 0)
1880 afap = kmalloc(sizeof(*afap), M_TEMP,
1882 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1883 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1884 if (so->so_accf->so_accept_filter_str != NULL)
1885 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1887 error = sooptcopyout(sopt, afap, sizeof(*afap));
1888 kfree(afap, M_TEMP);
1893 l.l_onoff = so->so_options & SO_LINGER;
1894 l.l_linger = so->so_linger;
1895 error = sooptcopyout(sopt, &l, sizeof l);
1898 case SO_USELOOPBACK:
1907 optval = so->so_options & sopt->sopt_name;
1909 error = sooptcopyout(sopt, &optval, sizeof optval);
1913 optval = so->so_type;
1917 optval = so->so_error;
1922 optval = so->so_snd.ssb_hiwat;
1926 optval = so->so_rcv.ssb_hiwat;
1930 optval = so->so_snd.ssb_lowat;
1934 optval = so->so_rcv.ssb_lowat;
1939 optval = (sopt->sopt_name == SO_SNDTIMEO ?
1940 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1942 tv.tv_sec = optval / hz;
1943 tv.tv_usec = (optval % hz) * ustick;
1944 error = sooptcopyout(sopt, &tv, sizeof tv);
1948 optval_l = ssb_space(&so->so_snd);
1949 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
1953 error = ENOPROTOOPT;
1960 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1962 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1964 struct mbuf *m, *m_prev;
1965 int sopt_size = sopt->sopt_valsize, msize;
1967 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1971 m->m_len = min(msize, sopt_size);
1972 sopt_size -= m->m_len;
1976 while (sopt_size > 0) {
1977 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1978 MT_DATA, 0, &msize);
1983 m->m_len = min(msize, sopt_size);
1984 sopt_size -= m->m_len;
1991 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1993 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1995 soopt_to_mbuf(sopt, m);
2000 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2005 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2007 if (sopt->sopt_val == NULL)
2009 val = sopt->sopt_val;
2010 valsize = sopt->sopt_valsize;
2011 while (m != NULL && valsize >= m->m_len) {
2012 bcopy(val, mtod(m, char *), m->m_len);
2013 valsize -= m->m_len;
2014 val = (caddr_t)val + m->m_len;
2017 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2018 panic("ip6_sooptmcopyin");
2021 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2023 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2025 return soopt_from_mbuf(sopt, m);
2029 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2031 struct mbuf *m0 = m;
2036 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2038 if (sopt->sopt_val == NULL)
2040 val = sopt->sopt_val;
2041 maxsize = sopt->sopt_valsize;
2042 while (m != NULL && maxsize >= m->m_len) {
2043 bcopy(mtod(m, char *), val, m->m_len);
2044 maxsize -= m->m_len;
2045 val = (caddr_t)val + m->m_len;
2046 valsize += m->m_len;
2050 /* enough soopt buffer should be given from user-land */
2054 sopt->sopt_valsize = valsize;
2059 sohasoutofband(struct socket *so)
2061 if (so->so_sigio != NULL)
2062 pgsigio(so->so_sigio, SIGURG, 0);
2063 KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
2067 sokqfilter(struct file *fp, struct knote *kn)
2069 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2070 struct signalsockbuf *ssb;
2072 switch (kn->kn_filter) {
2074 if (so->so_options & SO_ACCEPTCONN)
2075 kn->kn_fop = &solisten_filtops;
2077 kn->kn_fop = &soread_filtops;
2081 kn->kn_fop = &sowrite_filtops;
2085 kn->kn_fop = &soexcept_filtops;
2089 return (EOPNOTSUPP);
2092 knote_insert(&ssb->ssb_kq.ki_note, kn);
2093 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2098 filt_sordetach(struct knote *kn)
2100 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2102 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2103 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2104 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2109 filt_soread(struct knote *kn, long hint)
2111 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2113 if (kn->kn_sfflags & NOTE_OOB) {
2114 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2115 kn->kn_fflags |= NOTE_OOB;
2120 kn->kn_data = so->so_rcv.ssb_cc;
2122 if (so->so_state & SS_CANTRCVMORE) {
2124 * Only set NODATA if all data has been exhausted.
2126 if (kn->kn_data == 0)
2127 kn->kn_flags |= EV_NODATA;
2128 kn->kn_flags |= EV_EOF;
2129 kn->kn_fflags = so->so_error;
2132 if (so->so_error) /* temporary udp error */
2134 if (kn->kn_sfflags & NOTE_LOWAT)
2135 return (kn->kn_data >= kn->kn_sdata);
2136 return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2137 !TAILQ_EMPTY(&so->so_comp));
2141 filt_sowdetach(struct knote *kn)
2143 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2145 knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2146 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2147 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2152 filt_sowrite(struct knote *kn, long hint)
2154 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2156 kn->kn_data = ssb_space(&so->so_snd);
2157 if (so->so_state & SS_CANTSENDMORE) {
2158 kn->kn_flags |= (EV_EOF | EV_NODATA);
2159 kn->kn_fflags = so->so_error;
2162 if (so->so_error) /* temporary udp error */
2164 if (((so->so_state & SS_ISCONNECTED) == 0) &&
2165 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2167 if (kn->kn_sfflags & NOTE_LOWAT)
2168 return (kn->kn_data >= kn->kn_sdata);
2169 return (kn->kn_data >= so->so_snd.ssb_lowat);
2174 filt_solisten(struct knote *kn, long hint)
2176 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2178 kn->kn_data = so->so_qlen;
2179 return (! TAILQ_EMPTY(&so->so_comp));