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>
98 #include <sys/spinlock2.h>
100 #include <machine/limits.h>
102 extern int tcp_sosend_agglim;
103 extern int tcp_sosend_async;
104 extern int udp_sosend_async;
107 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
110 static void filt_sordetach(struct knote *kn);
111 static int filt_soread(struct knote *kn, long hint);
112 static void filt_sowdetach(struct knote *kn);
113 static int filt_sowrite(struct knote *kn, long hint);
114 static int filt_solisten(struct knote *kn, long hint);
116 static void sodiscard(struct socket *so);
117 static int soclose_sync(struct socket *so, int fflag);
118 static void soclose_fast(struct socket *so);
120 static struct filterops solisten_filtops =
121 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten };
122 static struct filterops soread_filtops =
123 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
124 static struct filterops sowrite_filtops =
125 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite };
126 static struct filterops soexcept_filtops =
127 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
129 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
130 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
131 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
134 static int somaxconn = SOMAXCONN;
135 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
136 &somaxconn, 0, "Maximum pending socket connection queue size");
138 static int use_soclose_fast = 1;
139 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW,
140 &use_soclose_fast, 0, "Fast socket close");
142 int use_soaccept_pred_fast = 1;
143 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW,
144 &use_soaccept_pred_fast, 0, "Fast socket accept predication");
146 int use_sendfile_async = 1;
147 SYSCTL_INT(_kern_ipc, OID_AUTO, sendfile_async, CTLFLAG_RW,
148 &use_sendfile_async, 0, "sendfile uses asynchronized pru_send");
151 * Socket operation routines.
152 * These routines are called by the routines in
153 * sys_socket.c or from a system process, and
154 * implement the semantics of socket operations by
155 * switching out to the protocol specific routines.
159 * Get a socket structure, and initialize it.
160 * Note that it would probably be better to allocate socket
161 * and PCB at the same time, but I'm not convinced that all
162 * the protocols can be easily modified to do this.
165 soalloc(int waitok, struct protosw *pr)
170 waitmask = waitok ? M_WAITOK : M_NOWAIT;
171 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
173 /* XXX race condition for reentrant kernel */
175 TAILQ_INIT(&so->so_aiojobq);
176 TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist);
177 TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist);
178 lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
179 lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
180 spin_init(&so->so_rcvd_spin);
181 netmsg_init(&so->so_rcvd_msg.base, so, &netisr_adone_rport,
182 MSGF_DROPABLE, so->so_proto->pr_usrreqs->pru_rcvd);
183 so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC;
184 so->so_state = SS_NOFDREF;
191 socreate(int dom, struct socket **aso, int type,
192 int proto, struct thread *td)
194 struct proc *p = td->td_proc;
197 struct pru_attach_info ai;
201 prp = pffindproto(dom, proto, type);
203 prp = pffindtype(dom, type);
205 if (prp == NULL || prp->pr_usrreqs->pru_attach == 0)
206 return (EPROTONOSUPPORT);
208 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
209 prp->pr_domain->dom_family != PF_LOCAL &&
210 prp->pr_domain->dom_family != PF_INET &&
211 prp->pr_domain->dom_family != PF_INET6 &&
212 prp->pr_domain->dom_family != PF_ROUTE) {
213 return (EPROTONOSUPPORT);
216 if (prp->pr_type != type)
218 so = soalloc(p != NULL, prp);
223 * Callers of socreate() presumably will connect up a descriptor
224 * and call soclose() if they cannot. This represents our so_refs
225 * (which should be 1) from soalloc().
227 soclrstate(so, SS_NOFDREF);
230 * Set a default port for protocol processing. No action will occur
231 * on the socket on this port until an inpcb is attached to it and
232 * is able to match incoming packets, or until the socket becomes
233 * available to userland.
235 * We normally default the socket to the protocol thread on cpu 0.
236 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
237 * thread and all pr_*()/pru_*() calls are executed synchronously.
239 if (prp->pr_flags & PR_SYNC_PORT)
240 so->so_port = &netisr_sync_port;
242 so->so_port = netisr_portfn(0);
244 TAILQ_INIT(&so->so_incomp);
245 TAILQ_INIT(&so->so_comp);
247 so->so_cred = crhold(p->p_ucred);
248 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
249 ai.p_ucred = p->p_ucred;
250 ai.fd_rdir = p->p_fd->fd_rdir;
253 * Auto-sizing of socket buffers is managed by the protocols and
254 * the appropriate flags must be set in the pru_attach function.
256 error = so_pru_attach(so, proto, &ai);
258 sosetstate(so, SS_NOFDREF);
259 sofree(so); /* from soalloc */
264 * NOTE: Returns referenced socket.
271 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
275 error = so_pru_bind(so, nam, td);
280 sodealloc(struct socket *so)
282 if (so->so_rcv.ssb_hiwat)
283 (void)chgsbsize(so->so_cred->cr_uidinfo,
284 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
285 if (so->so_snd.ssb_hiwat)
286 (void)chgsbsize(so->so_cred->cr_uidinfo,
287 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
289 /* remove accept filter if present */
290 if (so->so_accf != NULL)
291 do_setopt_accept_filter(so, NULL);
294 if (so->so_faddr != NULL)
295 kfree(so->so_faddr, M_SONAME);
300 solisten(struct socket *so, int backlog, struct thread *td)
304 short oldopt, oldqlimit;
307 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
311 oldopt = so->so_options;
312 oldqlimit = so->so_qlimit;
315 lwkt_gettoken(&so->so_rcv.ssb_token);
316 if (TAILQ_EMPTY(&so->so_comp))
317 so->so_options |= SO_ACCEPTCONN;
318 lwkt_reltoken(&so->so_rcv.ssb_token);
319 if (backlog < 0 || backlog > somaxconn)
321 so->so_qlimit = backlog;
322 /* SCTP needs to look at tweak both the inbound backlog parameter AND
323 * the so_options (UDP model both connect's and gets inbound
324 * connections .. implicitly).
326 error = so_pru_listen(so, td);
329 /* Restore the params */
330 so->so_options = oldopt;
331 so->so_qlimit = oldqlimit;
339 * Destroy a disconnected socket. This routine is a NOP if entities
340 * still have a reference on the socket:
342 * so_pcb - The protocol stack still has a reference
343 * SS_NOFDREF - There is no longer a file pointer reference
346 sofree(struct socket *so)
351 * This is a bit hackish at the moment. We need to interlock
352 * any accept queue we are on before we potentially lose the
353 * last reference to avoid races against a re-reference from
354 * someone operating on the queue.
356 while ((head = so->so_head) != NULL) {
357 lwkt_getpooltoken(head);
358 if (so->so_head == head)
360 lwkt_relpooltoken(head);
364 * Arbitrage the last free.
366 KKASSERT(so->so_refs > 0);
367 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
369 lwkt_relpooltoken(head);
373 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
374 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
377 * We're done, remove ourselves from the accept queue we are
378 * on, if we are on one.
381 if (so->so_state & SS_INCOMP) {
382 TAILQ_REMOVE(&head->so_incomp, so, so_list);
384 } else if (so->so_state & SS_COMP) {
386 * We must not decommission a socket that's
387 * on the accept(2) queue. If we do, then
388 * accept(2) may hang after select(2) indicated
389 * that the listening socket was ready.
391 lwkt_relpooltoken(head);
394 panic("sofree: not queued");
396 soclrstate(so, SS_INCOMP);
398 lwkt_relpooltoken(head);
400 ssb_release(&so->so_snd, so);
406 * Close a socket on last file table reference removal.
407 * Initiate disconnect if connected.
408 * Free socket when disconnect complete.
411 soclose(struct socket *so, int fflag)
415 funsetown(&so->so_sigio);
416 if (!use_soclose_fast ||
417 (so->so_proto->pr_flags & PR_SYNC_PORT) ||
418 (so->so_options & SO_LINGER)) {
419 error = soclose_sync(so, fflag);
428 sodiscard(struct socket *so)
430 lwkt_getpooltoken(so);
431 if (so->so_options & SO_ACCEPTCONN) {
434 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
435 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
436 soclrstate(sp, SS_INCOMP);
441 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
442 TAILQ_REMOVE(&so->so_comp, sp, so_list);
443 soclrstate(sp, SS_COMP);
449 lwkt_relpooltoken(so);
451 if (so->so_state & SS_NOFDREF)
452 panic("soclose: NOFDREF");
453 sosetstate(so, SS_NOFDREF); /* take ref */
457 soclose_sync(struct socket *so, int fflag)
461 if (so->so_pcb == NULL)
463 if (so->so_state & SS_ISCONNECTED) {
464 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
465 error = sodisconnect(so);
469 if (so->so_options & SO_LINGER) {
470 if ((so->so_state & SS_ISDISCONNECTING) &&
473 while (so->so_state & SS_ISCONNECTED) {
474 error = tsleep(&so->so_timeo, PCATCH,
475 "soclos", so->so_linger * hz);
485 error2 = so_pru_detach(so);
491 so_pru_sync(so); /* unpend async sending */
492 sofree(so); /* dispose of ref */
498 soclose_sofree_async_handler(netmsg_t msg)
500 sofree(msg->base.nm_so);
504 soclose_sofree_async(struct socket *so)
506 struct netmsg_base *base = &so->so_clomsg;
508 netmsg_init(base, so, &netisr_apanic_rport, 0,
509 soclose_sofree_async_handler);
510 lwkt_sendmsg(so->so_port, &base->lmsg);
514 soclose_disconn_async_handler(netmsg_t msg)
516 struct socket *so = msg->base.nm_so;
518 if ((so->so_state & SS_ISCONNECTED) &&
519 (so->so_state & SS_ISDISCONNECTING) == 0)
520 so_pru_disconnect_direct(so);
523 so_pru_detach_direct(so);
530 soclose_disconn_async(struct socket *so)
532 struct netmsg_base *base = &so->so_clomsg;
534 netmsg_init(base, so, &netisr_apanic_rport, 0,
535 soclose_disconn_async_handler);
536 lwkt_sendmsg(so->so_port, &base->lmsg);
540 soclose_detach_async_handler(netmsg_t msg)
542 struct socket *so = msg->base.nm_so;
545 so_pru_detach_direct(so);
552 soclose_detach_async(struct socket *so)
554 struct netmsg_base *base = &so->so_clomsg;
556 netmsg_init(base, so, &netisr_apanic_rport, 0,
557 soclose_detach_async_handler);
558 lwkt_sendmsg(so->so_port, &base->lmsg);
562 soclose_fast(struct socket *so)
564 if (so->so_pcb == NULL)
567 if ((so->so_state & SS_ISCONNECTED) &&
568 (so->so_state & SS_ISDISCONNECTING) == 0) {
569 soclose_disconn_async(so);
574 soclose_detach_async(so);
580 soclose_sofree_async(so);
584 * Abort and destroy a socket. Only one abort can be in progress
585 * at any given moment.
588 soabort(struct socket *so)
595 soaborta(struct socket *so)
602 soabort_oncpu(struct socket *so)
605 so_pru_abort_oncpu(so);
609 * so is passed in ref'd, which becomes owned by
610 * the cleared SS_NOFDREF flag.
613 soaccept_generic(struct socket *so)
615 if ((so->so_state & SS_NOFDREF) == 0)
616 panic("soaccept: !NOFDREF");
617 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */
621 soaccept(struct socket *so, struct sockaddr **nam)
625 soaccept_generic(so);
626 error = so_pru_accept(so, nam);
631 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
635 if (so->so_options & SO_ACCEPTCONN)
638 * If protocol is connection-based, can only connect once.
639 * Otherwise, if connected, try to disconnect first.
640 * This allows user to disconnect by connecting to, e.g.,
643 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
644 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
645 (error = sodisconnect(so)))) {
649 * Prevent accumulated error from previous connection
653 error = so_pru_connect(so, nam, td);
659 soconnect2(struct socket *so1, struct socket *so2)
663 error = so_pru_connect2(so1, so2);
668 sodisconnect(struct socket *so)
672 if ((so->so_state & SS_ISCONNECTED) == 0) {
676 if (so->so_state & SS_ISDISCONNECTING) {
680 error = so_pru_disconnect(so);
685 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
688 * If send must go all at once and message is larger than
689 * send buffering, then hard error.
690 * Lock against other senders.
691 * If must go all at once and not enough room now, then
692 * inform user that this would block and do nothing.
693 * Otherwise, if nonblocking, send as much as possible.
694 * The data to be sent is described by "uio" if nonzero,
695 * otherwise by the mbuf chain "top" (which must be null
696 * if uio is not). Data provided in mbuf chain must be small
697 * enough to send all at once.
699 * Returns nonzero on error, timeout or signal; callers
700 * must check for short counts if EINTR/ERESTART are returned.
701 * Data and control buffers are freed on return.
704 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
705 struct mbuf *top, struct mbuf *control, int flags,
712 int clen = 0, error, dontroute, mlen;
713 int atomic = sosendallatonce(so) || top;
717 resid = uio->uio_resid;
719 resid = (size_t)top->m_pkthdr.len;
722 for (m = top; m; m = m->m_next)
724 KKASSERT(top->m_pkthdr.len == len);
729 * WARNING! resid is unsigned, space and len are signed. space
730 * can wind up negative if the sockbuf is overcommitted.
732 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
733 * type sockets since that's an error.
735 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
741 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
742 (so->so_proto->pr_flags & PR_ATOMIC);
743 if (td->td_lwp != NULL)
744 td->td_lwp->lwp_ru.ru_msgsnd++;
746 clen = control->m_len;
747 #define gotoerr(errcode) { error = errcode; goto release; }
750 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
755 if (so->so_state & SS_CANTSENDMORE)
758 error = so->so_error;
762 if ((so->so_state & SS_ISCONNECTED) == 0) {
764 * `sendto' and `sendmsg' is allowed on a connection-
765 * based socket if it supports implied connect.
766 * Return ENOTCONN if not connected and no address is
769 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
770 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
771 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
772 !(resid == 0 && clen != 0))
774 } else if (addr == NULL)
775 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
776 ENOTCONN : EDESTADDRREQ);
778 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
779 clen > so->so_snd.ssb_hiwat) {
782 space = ssb_space(&so->so_snd);
785 if ((space < 0 || (size_t)space < resid + clen) && uio &&
786 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
787 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
788 gotoerr(EWOULDBLOCK);
789 ssb_unlock(&so->so_snd);
790 error = ssb_wait(&so->so_snd);
800 * Data is prepackaged in "top".
804 top->m_flags |= M_EOR;
808 m = m_getl((int)resid, MB_WAIT, MT_DATA,
809 top == NULL ? M_PKTHDR : 0, &mlen);
812 m->m_pkthdr.rcvif = NULL;
814 len = imin((int)szmin(mlen, resid), space);
815 if (resid < MINCLSIZE) {
817 * For datagram protocols, leave room
818 * for protocol headers in first mbuf.
820 if (atomic && top == NULL && len < mlen)
824 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
825 resid = uio->uio_resid;
828 top->m_pkthdr.len += len;
834 top->m_flags |= M_EOR;
837 } while (space > 0 && atomic);
839 so->so_options |= SO_DONTROUTE;
840 if (flags & MSG_OOB) {
841 pru_flags = PRUS_OOB;
842 } else if ((flags & MSG_EOF) &&
843 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
846 * If the user set MSG_EOF, the protocol
847 * understands this flag and nothing left to
848 * send then use PRU_SEND_EOF instead of PRU_SEND.
850 pru_flags = PRUS_EOF;
851 } else if (resid > 0 && space > 0) {
852 /* If there is more to send, set PRUS_MORETOCOME */
853 pru_flags = PRUS_MORETOCOME;
858 * XXX all the SS_CANTSENDMORE checks previously
859 * done could be out of date. We could have recieved
860 * a reset packet in an interrupt or maybe we slept
861 * while doing page faults in uiomove() etc. We could
862 * probably recheck again inside the splnet() protection
863 * here, but there are probably other places that this
864 * also happens. We must rethink this.
866 error = so_pru_send(so, pru_flags, top, addr, control, td);
868 so->so_options &= ~SO_DONTROUTE;
875 } while (resid && space > 0);
879 ssb_unlock(&so->so_snd);
889 * A specialization of sosend() for UDP based on protocol-specific knowledge:
890 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
891 * sosendallatonce() returns true,
892 * the "atomic" variable is true,
893 * and sosendudp() blocks until space is available for the entire send.
894 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
895 * PR_IMPLOPCL flags set.
896 * UDP has no out-of-band data.
897 * UDP has no control data.
898 * UDP does not support MSG_EOR.
901 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
902 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
905 int error, pru_flags = 0;
908 if (td->td_lwp != NULL)
909 td->td_lwp->lwp_ru.ru_msgsnd++;
913 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
914 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
917 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
921 if (so->so_state & SS_CANTSENDMORE)
924 error = so->so_error;
928 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
929 gotoerr(EDESTADDRREQ);
930 if (resid > so->so_snd.ssb_hiwat)
932 space = ssb_space(&so->so_snd);
933 if (uio && (space < 0 || (size_t)space < resid)) {
934 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
935 gotoerr(EWOULDBLOCK);
936 ssb_unlock(&so->so_snd);
937 error = ssb_wait(&so->so_snd);
944 top = m_uiomove(uio);
949 if (flags & MSG_DONTROUTE)
950 pru_flags |= PRUS_DONTROUTE;
952 if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
953 so_pru_send_async(so, pru_flags, top, addr, NULL, td);
956 error = so_pru_send(so, pru_flags, top, addr, NULL, td);
958 top = NULL; /* sent or freed in lower layer */
961 ssb_unlock(&so->so_snd);
969 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
970 struct mbuf *top, struct mbuf *control, int flags,
982 KKASSERT(top == NULL);
984 resid = uio->uio_resid;
987 resid = (size_t)top->m_pkthdr.len;
990 for (m = top; m; m = m->m_next)
992 KKASSERT(top->m_pkthdr.len == len);
997 * WARNING! resid is unsigned, space and len are signed. space
998 * can wind up negative if the sockbuf is overcommitted.
1000 * Also check to make sure that MSG_EOR isn't used on TCP
1002 if (flags & MSG_EOR) {
1008 /* TCP doesn't do control messages (rights, creds, etc) */
1009 if (control->m_len) {
1013 m_freem(control); /* empty control, just free it */
1017 if (td->td_lwp != NULL)
1018 td->td_lwp->lwp_ru.ru_msgsnd++;
1020 #define gotoerr(errcode) { error = errcode; goto release; }
1023 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1028 if (so->so_state & SS_CANTSENDMORE)
1031 error = so->so_error;
1035 if ((so->so_state & SS_ISCONNECTED) == 0 &&
1036 (so->so_state & SS_ISCONFIRMING) == 0)
1038 if (allatonce && resid > so->so_snd.ssb_hiwat)
1041 space = ssb_space_prealloc(&so->so_snd);
1042 if (flags & MSG_OOB)
1044 if ((space < 0 || (size_t)space < resid) && !allatonce &&
1045 space < so->so_snd.ssb_lowat) {
1046 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1047 gotoerr(EWOULDBLOCK);
1048 ssb_unlock(&so->so_snd);
1049 error = ssb_wait(&so->so_snd);
1056 int cnt = 0, async = 0;
1060 * Data is prepackaged in "top".
1064 if (resid > INT_MAX)
1066 m = m_getl((int)resid, MB_WAIT, MT_DATA,
1067 top == NULL ? M_PKTHDR : 0, &mlen);
1069 m->m_pkthdr.len = 0;
1070 m->m_pkthdr.rcvif = NULL;
1072 len = imin((int)szmin(mlen, resid), space);
1074 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1075 resid = uio->uio_resid;
1078 top->m_pkthdr.len += len;
1085 } while (space > 0 && cnt < tcp_sosend_agglim);
1087 if (tcp_sosend_async)
1090 if (flags & MSG_OOB) {
1091 pru_flags = PRUS_OOB;
1093 } else if ((flags & MSG_EOF) && resid == 0) {
1094 pru_flags = PRUS_EOF;
1095 } else if (resid > 0 && space > 0) {
1096 /* If there is more to send, set PRUS_MORETOCOME */
1097 pru_flags = PRUS_MORETOCOME;
1103 if (flags & MSG_SYNC)
1107 * XXX all the SS_CANTSENDMORE checks previously
1108 * done could be out of date. We could have recieved
1109 * a reset packet in an interrupt or maybe we slept
1110 * while doing page faults in uiomove() etc. We could
1111 * probably recheck again inside the splnet() protection
1112 * here, but there are probably other places that this
1113 * also happens. We must rethink this.
1115 for (m = top; m; m = m->m_next)
1116 ssb_preallocstream(&so->so_snd, m);
1118 error = so_pru_send(so, pru_flags, top,
1121 so_pru_send_async(so, pru_flags, top,
1130 } while (resid && space > 0);
1134 ssb_unlock(&so->so_snd);
1144 * Implement receive operations on a socket.
1146 * We depend on the way that records are added to the signalsockbuf
1147 * by sbappend*. In particular, each record (mbufs linked through m_next)
1148 * must begin with an address if the protocol so specifies,
1149 * followed by an optional mbuf or mbufs containing ancillary data,
1150 * and then zero or more mbufs of data.
1152 * Although the signalsockbuf is locked, new data may still be appended.
1153 * A token inside the ssb_lock deals with MP issues and still allows
1154 * the network to access the socket if we block in a uio.
1156 * The caller may receive the data as a single mbuf chain by supplying
1157 * an mbuf **mp0 for use in returning the chain. The uio is then used
1158 * only for the count in uio_resid.
1161 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1162 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1165 struct mbuf *free_chain = NULL;
1166 int flags, len, error, offset;
1167 struct protosw *pr = so->so_proto;
1169 size_t resid, orig_resid;
1172 resid = uio->uio_resid;
1174 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1182 flags = *flagsp &~ MSG_EOR;
1185 if (flags & MSG_OOB) {
1186 m = m_get(MB_WAIT, MT_DATA);
1189 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1195 KKASSERT(resid >= (size_t)m->m_len);
1196 resid -= (size_t)m->m_len;
1197 } while (resid > 0 && m);
1200 uio->uio_resid = resid;
1201 error = uiomove(mtod(m, caddr_t),
1202 (int)szmin(resid, m->m_len),
1204 resid = uio->uio_resid;
1206 } while (uio->uio_resid && error == 0 && m);
1213 if ((so->so_state & SS_ISCONFIRMING) && resid)
1217 * The token interlocks against the protocol thread while
1218 * ssb_lock is a blocking lock against other userland entities.
1220 lwkt_gettoken(&so->so_rcv.ssb_token);
1222 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1226 m = so->so_rcv.ssb_mb;
1228 * If we have less data than requested, block awaiting more
1229 * (subject to any timeout) if:
1230 * 1. the current count is less than the low water mark, or
1231 * 2. MSG_WAITALL is set, and it is possible to do the entire
1232 * receive operation at once if we block (resid <= hiwat).
1233 * 3. MSG_DONTWAIT is not set
1234 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1235 * we have to do the receive in sections, and thus risk returning
1236 * a short count if a timeout or signal occurs after we start.
1238 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1239 (size_t)so->so_rcv.ssb_cc < resid) &&
1240 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1241 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1242 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1243 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1247 error = so->so_error;
1248 if ((flags & MSG_PEEK) == 0)
1252 if (so->so_state & SS_CANTRCVMORE) {
1258 for (; m; m = m->m_next) {
1259 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1260 m = so->so_rcv.ssb_mb;
1264 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1265 (pr->pr_flags & PR_CONNREQUIRED)) {
1271 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1272 error = EWOULDBLOCK;
1275 ssb_unlock(&so->so_rcv);
1276 error = ssb_wait(&so->so_rcv);
1282 if (uio && uio->uio_td && uio->uio_td->td_proc)
1283 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1286 * note: m should be == sb_mb here. Cache the next record while
1287 * cleaning up. Note that calling m_free*() will break out critical
1290 KKASSERT(m == so->so_rcv.ssb_mb);
1293 * Skip any address mbufs prepending the record.
1295 if (pr->pr_flags & PR_ADDR) {
1296 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1299 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1300 if (flags & MSG_PEEK)
1303 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1307 * Skip any control mbufs prepending the record.
1310 if (pr->pr_flags & PR_ADDR_OPT) {
1312 * For SCTP we may be getting a
1313 * whole message OR a partial delivery.
1315 if (m && m->m_type == MT_SONAME) {
1318 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1319 if (flags & MSG_PEEK)
1322 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1326 while (m && m->m_type == MT_CONTROL && error == 0) {
1327 if (flags & MSG_PEEK) {
1329 *controlp = m_copy(m, 0, m->m_len);
1330 m = m->m_next; /* XXX race */
1333 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1334 if (pr->pr_domain->dom_externalize &&
1335 mtod(m, struct cmsghdr *)->cmsg_type ==
1337 error = (*pr->pr_domain->dom_externalize)(m);
1341 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1344 if (controlp && *controlp) {
1346 controlp = &(*controlp)->m_next;
1355 if (type == MT_OOBDATA)
1360 * Copy to the UIO or mbuf return chain (*mp).
1364 while (m && resid > 0 && error == 0) {
1365 if (m->m_type == MT_OOBDATA) {
1366 if (type != MT_OOBDATA)
1368 } else if (type == MT_OOBDATA)
1371 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1373 soclrstate(so, SS_RCVATMARK);
1374 len = (resid > INT_MAX) ? INT_MAX : resid;
1375 if (so->so_oobmark && len > so->so_oobmark - offset)
1376 len = so->so_oobmark - offset;
1377 if (len > m->m_len - moff)
1378 len = m->m_len - moff;
1381 * Copy out to the UIO or pass the mbufs back to the SIO.
1382 * The SIO is dealt with when we eat the mbuf, but deal
1383 * with the resid here either way.
1386 uio->uio_resid = resid;
1387 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1388 resid = uio->uio_resid;
1392 resid -= (size_t)len;
1396 * Eat the entire mbuf or just a piece of it
1398 if (len == m->m_len - moff) {
1399 if (m->m_flags & M_EOR)
1402 if (m->m_flags & M_NOTIFICATION)
1403 flags |= MSG_NOTIFICATION;
1405 if (flags & MSG_PEEK) {
1410 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1414 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1418 if (flags & MSG_PEEK) {
1422 n = m_copym(m, 0, len, MB_WAIT);
1428 so->so_rcv.ssb_cc -= len;
1431 if (so->so_oobmark) {
1432 if ((flags & MSG_PEEK) == 0) {
1433 so->so_oobmark -= len;
1434 if (so->so_oobmark == 0) {
1435 sosetstate(so, SS_RCVATMARK);
1440 if (offset == so->so_oobmark)
1444 if (flags & MSG_EOR)
1447 * If the MSG_WAITALL flag is set (for non-atomic socket),
1448 * we must not quit until resid == 0 or an error
1449 * termination. If a signal/timeout occurs, return
1450 * with a short count but without error.
1451 * Keep signalsockbuf locked against other readers.
1453 while ((flags & MSG_WAITALL) && m == NULL &&
1454 resid > 0 && !sosendallatonce(so) &&
1455 so->so_rcv.ssb_mb == NULL) {
1456 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1459 * The window might have closed to zero, make
1460 * sure we send an ack now that we've drained
1461 * the buffer or we might end up blocking until
1462 * the idle takes over (5 seconds).
1464 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1465 so_pru_rcvd(so, flags);
1466 error = ssb_wait(&so->so_rcv);
1468 ssb_unlock(&so->so_rcv);
1472 m = so->so_rcv.ssb_mb;
1477 * If an atomic read was requested but unread data still remains
1478 * in the record, set MSG_TRUNC.
1480 if (m && pr->pr_flags & PR_ATOMIC)
1484 * Cleanup. If an atomic read was requested drop any unread data.
1486 if ((flags & MSG_PEEK) == 0) {
1487 if (m && (pr->pr_flags & PR_ATOMIC))
1488 sbdroprecord(&so->so_rcv.sb);
1489 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1490 so_pru_rcvd(so, flags);
1493 if (orig_resid == resid && orig_resid &&
1494 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1495 ssb_unlock(&so->so_rcv);
1502 ssb_unlock(&so->so_rcv);
1504 lwkt_reltoken(&so->so_rcv.ssb_token);
1506 m_freem(free_chain);
1511 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1512 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1515 struct mbuf *free_chain = NULL;
1516 int flags, len, error, offset;
1517 struct protosw *pr = so->so_proto;
1519 size_t resid, orig_resid;
1522 resid = uio->uio_resid;
1524 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1532 flags = *flagsp &~ MSG_EOR;
1535 if (flags & MSG_OOB) {
1536 m = m_get(MB_WAIT, MT_DATA);
1539 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1545 KKASSERT(resid >= (size_t)m->m_len);
1546 resid -= (size_t)m->m_len;
1547 } while (resid > 0 && m);
1550 uio->uio_resid = resid;
1551 error = uiomove(mtod(m, caddr_t),
1552 (int)szmin(resid, m->m_len),
1554 resid = uio->uio_resid;
1556 } while (uio->uio_resid && error == 0 && m);
1565 * The token interlocks against the protocol thread while
1566 * ssb_lock is a blocking lock against other userland entities.
1568 lwkt_gettoken(&so->so_rcv.ssb_token);
1570 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1574 m = so->so_rcv.ssb_mb;
1576 * If we have less data than requested, block awaiting more
1577 * (subject to any timeout) if:
1578 * 1. the current count is less than the low water mark, or
1579 * 2. MSG_WAITALL is set, and it is possible to do the entire
1580 * receive operation at once if we block (resid <= hiwat).
1581 * 3. MSG_DONTWAIT is not set
1582 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1583 * we have to do the receive in sections, and thus risk returning
1584 * a short count if a timeout or signal occurs after we start.
1586 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1587 (size_t)so->so_rcv.ssb_cc < resid) &&
1588 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1589 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1590 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1594 error = so->so_error;
1595 if ((flags & MSG_PEEK) == 0)
1599 if (so->so_state & SS_CANTRCVMORE) {
1605 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1606 (pr->pr_flags & PR_CONNREQUIRED)) {
1612 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1613 error = EWOULDBLOCK;
1616 ssb_unlock(&so->so_rcv);
1617 error = ssb_wait(&so->so_rcv);
1623 if (uio && uio->uio_td && uio->uio_td->td_proc)
1624 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1627 * note: m should be == sb_mb here. Cache the next record while
1628 * cleaning up. Note that calling m_free*() will break out critical
1631 KKASSERT(m == so->so_rcv.ssb_mb);
1634 * Copy to the UIO or mbuf return chain (*mp).
1638 while (m && resid > 0 && error == 0) {
1639 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1642 soclrstate(so, SS_RCVATMARK);
1643 len = (resid > INT_MAX) ? INT_MAX : resid;
1644 if (so->so_oobmark && len > so->so_oobmark - offset)
1645 len = so->so_oobmark - offset;
1646 if (len > m->m_len - moff)
1647 len = m->m_len - moff;
1650 * Copy out to the UIO or pass the mbufs back to the SIO.
1651 * The SIO is dealt with when we eat the mbuf, but deal
1652 * with the resid here either way.
1655 uio->uio_resid = resid;
1656 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1657 resid = uio->uio_resid;
1661 resid -= (size_t)len;
1665 * Eat the entire mbuf or just a piece of it
1667 if (len == m->m_len - moff) {
1668 if (flags & MSG_PEEK) {
1673 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1677 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1681 if (flags & MSG_PEEK) {
1685 n = m_copym(m, 0, len, MB_WAIT);
1691 so->so_rcv.ssb_cc -= len;
1694 if (so->so_oobmark) {
1695 if ((flags & MSG_PEEK) == 0) {
1696 so->so_oobmark -= len;
1697 if (so->so_oobmark == 0) {
1698 sosetstate(so, SS_RCVATMARK);
1703 if (offset == so->so_oobmark)
1708 * If the MSG_WAITALL flag is set (for non-atomic socket),
1709 * we must not quit until resid == 0 or an error
1710 * termination. If a signal/timeout occurs, return
1711 * with a short count but without error.
1712 * Keep signalsockbuf locked against other readers.
1714 while ((flags & MSG_WAITALL) && m == NULL &&
1715 resid > 0 && !sosendallatonce(so) &&
1716 so->so_rcv.ssb_mb == NULL) {
1717 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1720 * The window might have closed to zero, make
1721 * sure we send an ack now that we've drained
1722 * the buffer or we might end up blocking until
1723 * the idle takes over (5 seconds).
1726 so_pru_rcvd_async(so);
1727 error = ssb_wait(&so->so_rcv);
1729 ssb_unlock(&so->so_rcv);
1733 m = so->so_rcv.ssb_mb;
1738 * Cleanup. If an atomic read was requested drop any unread data.
1740 if ((flags & MSG_PEEK) == 0) {
1742 so_pru_rcvd_async(so);
1745 if (orig_resid == resid && orig_resid &&
1746 (so->so_state & SS_CANTRCVMORE) == 0) {
1747 ssb_unlock(&so->so_rcv);
1754 ssb_unlock(&so->so_rcv);
1756 lwkt_reltoken(&so->so_rcv.ssb_token);
1758 m_freem(free_chain);
1763 * Shut a socket down. Note that we do not get a frontend lock as we
1764 * want to be able to shut the socket down even if another thread is
1765 * blocked in a read(), thus waking it up.
1768 soshutdown(struct socket *so, int how)
1770 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1773 if (how != SHUT_WR) {
1774 /*ssb_lock(&so->so_rcv, M_WAITOK);*/
1776 /*ssb_unlock(&so->so_rcv);*/
1779 return (so_pru_shutdown(so));
1784 sorflush(struct socket *so)
1786 struct signalsockbuf *ssb = &so->so_rcv;
1787 struct protosw *pr = so->so_proto;
1788 struct signalsockbuf asb;
1790 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1792 lwkt_gettoken(&ssb->ssb_token);
1797 * Can't just blow up the ssb structure here
1799 bzero(&ssb->sb, sizeof(ssb->sb));
1804 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1806 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1807 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1808 ssb_release(&asb, so);
1810 lwkt_reltoken(&ssb->ssb_token);
1815 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1817 struct accept_filter_arg *afap = NULL;
1818 struct accept_filter *afp;
1819 struct so_accf *af = so->so_accf;
1822 /* do not set/remove accept filters on non listen sockets */
1823 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1828 /* removing the filter */
1831 if (af->so_accept_filter != NULL &&
1832 af->so_accept_filter->accf_destroy != NULL) {
1833 af->so_accept_filter->accf_destroy(so);
1835 if (af->so_accept_filter_str != NULL) {
1836 kfree(af->so_accept_filter_str, M_ACCF);
1841 so->so_options &= ~SO_ACCEPTFILTER;
1844 /* adding a filter */
1845 /* must remove previous filter first */
1850 /* don't put large objects on the kernel stack */
1851 afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
1852 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1853 afap->af_name[sizeof(afap->af_name)-1] = '\0';
1854 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1857 afp = accept_filt_get(afap->af_name);
1862 af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1863 if (afp->accf_create != NULL) {
1864 if (afap->af_name[0] != '\0') {
1865 int len = strlen(afap->af_name) + 1;
1867 af->so_accept_filter_str = kmalloc(len, M_ACCF,
1869 strcpy(af->so_accept_filter_str, afap->af_name);
1871 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1872 if (af->so_accept_filter_arg == NULL) {
1873 kfree(af->so_accept_filter_str, M_ACCF);
1880 af->so_accept_filter = afp;
1882 so->so_options |= SO_ACCEPTFILTER;
1885 kfree(afap, M_TEMP);
1891 * Perhaps this routine, and sooptcopyout(), below, ought to come in
1892 * an additional variant to handle the case where the option value needs
1893 * to be some kind of integer, but not a specific size.
1894 * In addition to their use here, these functions are also called by the
1895 * protocol-level pr_ctloutput() routines.
1898 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1900 return soopt_to_kbuf(sopt, buf, len, minlen);
1904 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1908 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1909 KKASSERT(kva_p(buf));
1912 * If the user gives us more than we wanted, we ignore it,
1913 * but if we don't get the minimum length the caller
1914 * wants, we return EINVAL. On success, sopt->sopt_valsize
1915 * is set to however much we actually retrieved.
1917 if ((valsize = sopt->sopt_valsize) < minlen)
1920 sopt->sopt_valsize = valsize = len;
1922 bcopy(sopt->sopt_val, buf, valsize);
1928 sosetopt(struct socket *so, struct sockopt *sopt)
1934 struct signalsockbuf *sotmp;
1937 sopt->sopt_dir = SOPT_SET;
1938 if (sopt->sopt_level != SOL_SOCKET) {
1939 if (so->so_proto && so->so_proto->pr_ctloutput) {
1940 return (so_pr_ctloutput(so, sopt));
1942 error = ENOPROTOOPT;
1944 switch (sopt->sopt_name) {
1946 case SO_ACCEPTFILTER:
1947 error = do_setopt_accept_filter(so, sopt);
1953 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1957 so->so_linger = l.l_linger;
1959 so->so_options |= SO_LINGER;
1961 so->so_options &= ~SO_LINGER;
1967 case SO_USELOOPBACK:
1973 error = sooptcopyin(sopt, &optval, sizeof optval,
1978 so->so_options |= sopt->sopt_name;
1980 so->so_options &= ~sopt->sopt_name;
1987 error = sooptcopyin(sopt, &optval, sizeof optval,
1993 * Values < 1 make no sense for any of these
1994 * options, so disallow them.
2001 switch (sopt->sopt_name) {
2004 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2005 &so->so_snd : &so->so_rcv, (u_long)optval,
2007 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2011 sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2012 &so->so_snd : &so->so_rcv;
2013 atomic_clear_int(&sotmp->ssb_flags,
2018 * Make sure the low-water is never greater than
2022 so->so_snd.ssb_lowat =
2023 (optval > so->so_snd.ssb_hiwat) ?
2024 so->so_snd.ssb_hiwat : optval;
2025 atomic_clear_int(&so->so_snd.ssb_flags,
2029 so->so_rcv.ssb_lowat =
2030 (optval > so->so_rcv.ssb_hiwat) ?
2031 so->so_rcv.ssb_hiwat : optval;
2032 atomic_clear_int(&so->so_rcv.ssb_flags,
2040 error = sooptcopyin(sopt, &tv, sizeof tv,
2045 /* assert(hz > 0); */
2046 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2047 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2051 /* assert(tick > 0); */
2052 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2053 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2054 if (val > INT_MAX) {
2058 if (val == 0 && tv.tv_usec != 0)
2061 switch (sopt->sopt_name) {
2063 so->so_snd.ssb_timeo = val;
2066 so->so_rcv.ssb_timeo = val;
2071 error = ENOPROTOOPT;
2074 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2075 (void) so_pr_ctloutput(so, sopt);
2082 /* Helper routine for getsockopt */
2084 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2086 soopt_from_kbuf(sopt, buf, len);
2091 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2096 sopt->sopt_valsize = 0;
2100 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2101 KKASSERT(kva_p(buf));
2104 * Documented get behavior is that we always return a value,
2105 * possibly truncated to fit in the user's buffer.
2106 * Traditional behavior is that we always tell the user
2107 * precisely how much we copied, rather than something useful
2108 * like the total amount we had available for her.
2109 * Note that this interface is not idempotent; the entire answer must
2110 * generated ahead of time.
2112 valsize = szmin(len, sopt->sopt_valsize);
2113 sopt->sopt_valsize = valsize;
2114 if (sopt->sopt_val != 0) {
2115 bcopy(buf, sopt->sopt_val, valsize);
2120 sogetopt(struct socket *so, struct sockopt *sopt)
2127 struct accept_filter_arg *afap;
2131 sopt->sopt_dir = SOPT_GET;
2132 if (sopt->sopt_level != SOL_SOCKET) {
2133 if (so->so_proto && so->so_proto->pr_ctloutput) {
2134 return (so_pr_ctloutput(so, sopt));
2136 return (ENOPROTOOPT);
2138 switch (sopt->sopt_name) {
2140 case SO_ACCEPTFILTER:
2141 if ((so->so_options & SO_ACCEPTCONN) == 0)
2143 afap = kmalloc(sizeof(*afap), M_TEMP,
2145 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2146 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2147 if (so->so_accf->so_accept_filter_str != NULL)
2148 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2150 error = sooptcopyout(sopt, afap, sizeof(*afap));
2151 kfree(afap, M_TEMP);
2156 l.l_onoff = so->so_options & SO_LINGER;
2157 l.l_linger = so->so_linger;
2158 error = sooptcopyout(sopt, &l, sizeof l);
2161 case SO_USELOOPBACK:
2170 optval = so->so_options & sopt->sopt_name;
2172 error = sooptcopyout(sopt, &optval, sizeof optval);
2176 optval = so->so_type;
2180 optval = so->so_error;
2185 optval = so->so_snd.ssb_hiwat;
2189 optval = so->so_rcv.ssb_hiwat;
2193 optval = so->so_snd.ssb_lowat;
2197 optval = so->so_rcv.ssb_lowat;
2202 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2203 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2205 tv.tv_sec = optval / hz;
2206 tv.tv_usec = (optval % hz) * ustick;
2207 error = sooptcopyout(sopt, &tv, sizeof tv);
2211 optval_l = ssb_space(&so->so_snd);
2212 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2216 error = ENOPROTOOPT;
2223 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2225 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2227 struct mbuf *m, *m_prev;
2228 int sopt_size = sopt->sopt_valsize, msize;
2230 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
2234 m->m_len = min(msize, sopt_size);
2235 sopt_size -= m->m_len;
2239 while (sopt_size > 0) {
2240 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
2241 MT_DATA, 0, &msize);
2246 m->m_len = min(msize, sopt_size);
2247 sopt_size -= m->m_len;
2254 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2256 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2258 soopt_to_mbuf(sopt, m);
2263 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2268 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2270 if (sopt->sopt_val == NULL)
2272 val = sopt->sopt_val;
2273 valsize = sopt->sopt_valsize;
2274 while (m != NULL && valsize >= m->m_len) {
2275 bcopy(val, mtod(m, char *), m->m_len);
2276 valsize -= m->m_len;
2277 val = (caddr_t)val + m->m_len;
2280 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2281 panic("ip6_sooptmcopyin");
2284 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2286 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2288 return soopt_from_mbuf(sopt, m);
2292 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2294 struct mbuf *m0 = m;
2299 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2301 if (sopt->sopt_val == NULL)
2303 val = sopt->sopt_val;
2304 maxsize = sopt->sopt_valsize;
2305 while (m != NULL && maxsize >= m->m_len) {
2306 bcopy(mtod(m, char *), val, m->m_len);
2307 maxsize -= m->m_len;
2308 val = (caddr_t)val + m->m_len;
2309 valsize += m->m_len;
2313 /* enough soopt buffer should be given from user-land */
2317 sopt->sopt_valsize = valsize;
2322 sohasoutofband(struct socket *so)
2324 if (so->so_sigio != NULL)
2325 pgsigio(so->so_sigio, SIGURG, 0);
2326 KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
2330 sokqfilter(struct file *fp, struct knote *kn)
2332 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2333 struct signalsockbuf *ssb;
2335 switch (kn->kn_filter) {
2337 if (so->so_options & SO_ACCEPTCONN)
2338 kn->kn_fop = &solisten_filtops;
2340 kn->kn_fop = &soread_filtops;
2344 kn->kn_fop = &sowrite_filtops;
2348 kn->kn_fop = &soexcept_filtops;
2352 return (EOPNOTSUPP);
2355 knote_insert(&ssb->ssb_kq.ki_note, kn);
2356 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2361 filt_sordetach(struct knote *kn)
2363 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2365 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2366 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2367 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2372 filt_soread(struct knote *kn, long hint)
2374 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2376 if (kn->kn_sfflags & NOTE_OOB) {
2377 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2378 kn->kn_fflags |= NOTE_OOB;
2383 kn->kn_data = so->so_rcv.ssb_cc;
2385 if (so->so_state & SS_CANTRCVMORE) {
2387 * Only set NODATA if all data has been exhausted.
2389 if (kn->kn_data == 0)
2390 kn->kn_flags |= EV_NODATA;
2391 kn->kn_flags |= EV_EOF;
2392 kn->kn_fflags = so->so_error;
2395 if (so->so_error) /* temporary udp error */
2397 if (kn->kn_sfflags & NOTE_LOWAT)
2398 return (kn->kn_data >= kn->kn_sdata);
2399 return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2400 !TAILQ_EMPTY(&so->so_comp));
2404 filt_sowdetach(struct knote *kn)
2406 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2408 knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2409 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2410 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2415 filt_sowrite(struct knote *kn, long hint)
2417 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2419 kn->kn_data = ssb_space(&so->so_snd);
2420 if (so->so_state & SS_CANTSENDMORE) {
2421 kn->kn_flags |= (EV_EOF | EV_NODATA);
2422 kn->kn_fflags = so->so_error;
2425 if (so->so_error) /* temporary udp error */
2427 if (((so->so_state & SS_ISCONNECTED) == 0) &&
2428 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2430 if (kn->kn_sfflags & NOTE_LOWAT)
2431 return (kn->kn_data >= kn->kn_sdata);
2432 return (kn->kn_data >= so->so_snd.ssb_lowat);
2437 filt_solisten(struct knote *kn, long hint)
2439 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2441 kn->kn_data = so->so_qlen;
2442 return (! TAILQ_EMPTY(&so->so_comp));