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>
95 #include <net/netisr2.h>
97 #include <sys/thread2.h>
98 #include <sys/socketvar2.h>
99 #include <sys/spinlock2.h>
101 #include <machine/limits.h>
104 extern int tcp_sosend_agglim;
105 extern int tcp_sosend_async;
106 extern int udp_sosend_async;
107 extern int udp_sosend_prepend;
109 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
112 static void filt_sordetach(struct knote *kn);
113 static int filt_soread(struct knote *kn, long hint);
114 static void filt_sowdetach(struct knote *kn);
115 static int filt_sowrite(struct knote *kn, long hint);
116 static int filt_solisten(struct knote *kn, long hint);
118 static void sodiscard(struct socket *so);
119 static int soclose_sync(struct socket *so, int fflag);
120 static void soclose_fast(struct socket *so);
122 static struct filterops solisten_filtops =
123 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten };
124 static struct filterops soread_filtops =
125 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
126 static struct filterops sowrite_filtops =
127 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite };
128 static struct filterops soexcept_filtops =
129 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
131 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
132 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
133 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
136 static int somaxconn = SOMAXCONN;
137 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
138 &somaxconn, 0, "Maximum pending socket connection queue size");
140 static int use_soclose_fast = 1;
141 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW,
142 &use_soclose_fast, 0, "Fast socket close");
144 int use_soaccept_pred_fast = 1;
145 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW,
146 &use_soaccept_pred_fast, 0, "Fast socket accept predication");
148 int use_sendfile_async = 1;
149 SYSCTL_INT(_kern_ipc, OID_AUTO, sendfile_async, CTLFLAG_RW,
150 &use_sendfile_async, 0, "sendfile uses asynchronized pru_send");
153 * Socket operation routines.
154 * These routines are called by the routines in
155 * sys_socket.c or from a system process, and
156 * implement the semantics of socket operations by
157 * switching out to the protocol specific routines.
161 * Get a socket structure, and initialize it.
162 * Note that it would probably be better to allocate socket
163 * and PCB at the same time, but I'm not convinced that all
164 * the protocols can be easily modified to do this.
167 soalloc(int waitok, struct protosw *pr)
172 waitmask = waitok ? M_WAITOK : M_NOWAIT;
173 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
175 /* XXX race condition for reentrant kernel */
177 TAILQ_INIT(&so->so_aiojobq);
178 TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist);
179 TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist);
180 lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
181 lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
182 spin_init(&so->so_rcvd_spin);
183 netmsg_init(&so->so_rcvd_msg.base, so, &netisr_adone_rport,
184 MSGF_DROPABLE, so->so_proto->pr_usrreqs->pru_rcvd);
185 so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC;
186 so->so_state = SS_NOFDREF;
193 socreate(int dom, struct socket **aso, int type,
194 int proto, struct thread *td)
196 struct proc *p = td->td_proc;
199 struct pru_attach_info ai;
203 prp = pffindproto(dom, proto, type);
205 prp = pffindtype(dom, type);
207 if (prp == NULL || prp->pr_usrreqs->pru_attach == 0)
208 return (EPROTONOSUPPORT);
210 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
211 prp->pr_domain->dom_family != PF_LOCAL &&
212 prp->pr_domain->dom_family != PF_INET &&
213 prp->pr_domain->dom_family != PF_INET6 &&
214 prp->pr_domain->dom_family != PF_ROUTE) {
215 return (EPROTONOSUPPORT);
218 if (prp->pr_type != type)
220 so = soalloc(p != NULL, prp);
225 * Callers of socreate() presumably will connect up a descriptor
226 * and call soclose() if they cannot. This represents our so_refs
227 * (which should be 1) from soalloc().
229 soclrstate(so, SS_NOFDREF);
232 * Set a default port for protocol processing. No action will occur
233 * on the socket on this port until an inpcb is attached to it and
234 * is able to match incoming packets, or until the socket becomes
235 * available to userland.
237 * We normally default the socket to the protocol thread on cpu 0.
238 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
239 * thread and all pr_*()/pru_*() calls are executed synchronously.
241 if (prp->pr_flags & PR_SYNC_PORT)
242 so->so_port = &netisr_sync_port;
244 so->so_port = netisr_cpuport(0);
246 TAILQ_INIT(&so->so_incomp);
247 TAILQ_INIT(&so->so_comp);
249 so->so_cred = crhold(p->p_ucred);
250 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
251 ai.p_ucred = p->p_ucred;
252 ai.fd_rdir = p->p_fd->fd_rdir;
255 * Auto-sizing of socket buffers is managed by the protocols and
256 * the appropriate flags must be set in the pru_attach function.
258 error = so_pru_attach(so, proto, &ai);
260 sosetstate(so, SS_NOFDREF);
261 sofree(so); /* from soalloc */
266 * NOTE: Returns referenced socket.
273 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
277 error = so_pru_bind(so, nam, td);
282 sodealloc(struct socket *so)
284 if (so->so_rcv.ssb_hiwat)
285 (void)chgsbsize(so->so_cred->cr_uidinfo,
286 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
287 if (so->so_snd.ssb_hiwat)
288 (void)chgsbsize(so->so_cred->cr_uidinfo,
289 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
291 /* remove accept filter if present */
292 if (so->so_accf != NULL)
293 do_setopt_accept_filter(so, NULL);
296 if (so->so_faddr != NULL)
297 kfree(so->so_faddr, M_SONAME);
302 solisten(struct socket *so, int backlog, struct thread *td)
306 short oldopt, oldqlimit;
309 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
313 oldopt = so->so_options;
314 oldqlimit = so->so_qlimit;
317 lwkt_gettoken(&so->so_rcv.ssb_token);
318 if (TAILQ_EMPTY(&so->so_comp))
319 so->so_options |= SO_ACCEPTCONN;
320 lwkt_reltoken(&so->so_rcv.ssb_token);
321 if (backlog < 0 || backlog > somaxconn)
323 so->so_qlimit = backlog;
324 /* SCTP needs to look at tweak both the inbound backlog parameter AND
325 * the so_options (UDP model both connect's and gets inbound
326 * connections .. implicitly).
328 error = so_pru_listen(so, td);
331 /* Restore the params */
332 so->so_options = oldopt;
333 so->so_qlimit = oldqlimit;
341 * Destroy a disconnected socket. This routine is a NOP if entities
342 * still have a reference on the socket:
344 * so_pcb - The protocol stack still has a reference
345 * SS_NOFDREF - There is no longer a file pointer reference
348 sofree(struct socket *so)
353 * This is a bit hackish at the moment. We need to interlock
354 * any accept queue we are on before we potentially lose the
355 * last reference to avoid races against a re-reference from
356 * someone operating on the queue.
358 while ((head = so->so_head) != NULL) {
359 lwkt_getpooltoken(head);
360 if (so->so_head == head)
362 lwkt_relpooltoken(head);
366 * Arbitrage the last free.
368 KKASSERT(so->so_refs > 0);
369 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
371 lwkt_relpooltoken(head);
375 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
376 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
379 * We're done, remove ourselves from the accept queue we are
380 * on, if we are on one.
383 if (so->so_state & SS_INCOMP) {
384 TAILQ_REMOVE(&head->so_incomp, so, so_list);
386 } else if (so->so_state & SS_COMP) {
388 * We must not decommission a socket that's
389 * on the accept(2) queue. If we do, then
390 * accept(2) may hang after select(2) indicated
391 * that the listening socket was ready.
393 lwkt_relpooltoken(head);
396 panic("sofree: not queued");
398 soclrstate(so, SS_INCOMP);
400 lwkt_relpooltoken(head);
402 ssb_release(&so->so_snd, so);
408 * Close a socket on last file table reference removal.
409 * Initiate disconnect if connected.
410 * Free socket when disconnect complete.
413 soclose(struct socket *so, int fflag)
417 funsetown(&so->so_sigio);
418 if (!use_soclose_fast ||
419 (so->so_proto->pr_flags & PR_SYNC_PORT) ||
420 (so->so_options & SO_LINGER)) {
421 error = soclose_sync(so, fflag);
430 sodiscard(struct socket *so)
432 lwkt_getpooltoken(so);
433 if (so->so_options & SO_ACCEPTCONN) {
436 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
437 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
438 soclrstate(sp, SS_INCOMP);
443 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
444 TAILQ_REMOVE(&so->so_comp, sp, so_list);
445 soclrstate(sp, SS_COMP);
451 lwkt_relpooltoken(so);
453 if (so->so_state & SS_NOFDREF)
454 panic("soclose: NOFDREF");
455 sosetstate(so, SS_NOFDREF); /* take ref */
459 soclose_sync(struct socket *so, int fflag)
463 if (so->so_pcb == NULL)
465 if (so->so_state & SS_ISCONNECTED) {
466 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
467 error = sodisconnect(so);
471 if (so->so_options & SO_LINGER) {
472 if ((so->so_state & SS_ISDISCONNECTING) &&
475 while (so->so_state & SS_ISCONNECTED) {
476 error = tsleep(&so->so_timeo, PCATCH,
477 "soclos", so->so_linger * hz);
487 error2 = so_pru_detach(so);
493 so_pru_sync(so); /* unpend async sending */
494 sofree(so); /* dispose of ref */
500 soclose_sofree_async_handler(netmsg_t msg)
502 sofree(msg->base.nm_so);
506 soclose_sofree_async(struct socket *so)
508 struct netmsg_base *base = &so->so_clomsg;
510 netmsg_init(base, so, &netisr_apanic_rport, 0,
511 soclose_sofree_async_handler);
512 lwkt_sendmsg(so->so_port, &base->lmsg);
516 soclose_disconn_async_handler(netmsg_t msg)
518 struct socket *so = msg->base.nm_so;
520 if ((so->so_state & SS_ISCONNECTED) &&
521 (so->so_state & SS_ISDISCONNECTING) == 0)
522 so_pru_disconnect_direct(so);
525 so_pru_detach_direct(so);
532 soclose_disconn_async(struct socket *so)
534 struct netmsg_base *base = &so->so_clomsg;
536 netmsg_init(base, so, &netisr_apanic_rport, 0,
537 soclose_disconn_async_handler);
538 lwkt_sendmsg(so->so_port, &base->lmsg);
542 soclose_detach_async_handler(netmsg_t msg)
544 struct socket *so = msg->base.nm_so;
547 so_pru_detach_direct(so);
554 soclose_detach_async(struct socket *so)
556 struct netmsg_base *base = &so->so_clomsg;
558 netmsg_init(base, so, &netisr_apanic_rport, 0,
559 soclose_detach_async_handler);
560 lwkt_sendmsg(so->so_port, &base->lmsg);
564 soclose_fast(struct socket *so)
566 if (so->so_pcb == NULL)
569 if ((so->so_state & SS_ISCONNECTED) &&
570 (so->so_state & SS_ISDISCONNECTING) == 0) {
571 soclose_disconn_async(so);
576 soclose_detach_async(so);
582 soclose_sofree_async(so);
586 * Abort and destroy a socket. Only one abort can be in progress
587 * at any given moment.
590 soabort(struct socket *so)
597 soaborta(struct socket *so)
604 soabort_oncpu(struct socket *so)
607 so_pru_abort_oncpu(so);
611 * so is passed in ref'd, which becomes owned by
612 * the cleared SS_NOFDREF flag.
615 soaccept_generic(struct socket *so)
617 if ((so->so_state & SS_NOFDREF) == 0)
618 panic("soaccept: !NOFDREF");
619 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */
623 soaccept(struct socket *so, struct sockaddr **nam)
627 soaccept_generic(so);
628 error = so_pru_accept(so, nam);
633 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
637 if (so->so_options & SO_ACCEPTCONN)
640 * If protocol is connection-based, can only connect once.
641 * Otherwise, if connected, try to disconnect first.
642 * This allows user to disconnect by connecting to, e.g.,
645 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
646 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
647 (error = sodisconnect(so)))) {
651 * Prevent accumulated error from previous connection
655 error = so_pru_connect(so, nam, td);
661 soconnect2(struct socket *so1, struct socket *so2)
665 error = so_pru_connect2(so1, so2);
670 sodisconnect(struct socket *so)
674 if ((so->so_state & SS_ISCONNECTED) == 0) {
678 if (so->so_state & SS_ISDISCONNECTING) {
682 error = so_pru_disconnect(so);
687 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
690 * If send must go all at once and message is larger than
691 * send buffering, then hard error.
692 * Lock against other senders.
693 * If must go all at once and not enough room now, then
694 * inform user that this would block and do nothing.
695 * Otherwise, if nonblocking, send as much as possible.
696 * The data to be sent is described by "uio" if nonzero,
697 * otherwise by the mbuf chain "top" (which must be null
698 * if uio is not). Data provided in mbuf chain must be small
699 * enough to send all at once.
701 * Returns nonzero on error, timeout or signal; callers
702 * must check for short counts if EINTR/ERESTART are returned.
703 * Data and control buffers are freed on return.
706 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
707 struct mbuf *top, struct mbuf *control, int flags,
714 int clen = 0, error, dontroute, mlen;
715 int atomic = sosendallatonce(so) || top;
719 resid = uio->uio_resid;
721 resid = (size_t)top->m_pkthdr.len;
724 for (m = top; m; m = m->m_next)
726 KKASSERT(top->m_pkthdr.len == len);
731 * WARNING! resid is unsigned, space and len are signed. space
732 * can wind up negative if the sockbuf is overcommitted.
734 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
735 * type sockets since that's an error.
737 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
743 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
744 (so->so_proto->pr_flags & PR_ATOMIC);
745 if (td->td_lwp != NULL)
746 td->td_lwp->lwp_ru.ru_msgsnd++;
748 clen = control->m_len;
749 #define gotoerr(errcode) { error = errcode; goto release; }
752 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
757 if (so->so_state & SS_CANTSENDMORE)
760 error = so->so_error;
764 if ((so->so_state & SS_ISCONNECTED) == 0) {
766 * `sendto' and `sendmsg' is allowed on a connection-
767 * based socket if it supports implied connect.
768 * Return ENOTCONN if not connected and no address is
771 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
772 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
773 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
774 !(resid == 0 && clen != 0))
776 } else if (addr == NULL)
777 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
778 ENOTCONN : EDESTADDRREQ);
780 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
781 clen > so->so_snd.ssb_hiwat) {
784 space = ssb_space(&so->so_snd);
787 if ((space < 0 || (size_t)space < resid + clen) && uio &&
788 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
789 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
790 gotoerr(EWOULDBLOCK);
791 ssb_unlock(&so->so_snd);
792 error = ssb_wait(&so->so_snd);
802 * Data is prepackaged in "top".
806 top->m_flags |= M_EOR;
810 m = m_getl((int)resid, MB_WAIT, MT_DATA,
811 top == NULL ? M_PKTHDR : 0, &mlen);
814 m->m_pkthdr.rcvif = NULL;
816 len = imin((int)szmin(mlen, resid), space);
817 if (resid < MINCLSIZE) {
819 * For datagram protocols, leave room
820 * for protocol headers in first mbuf.
822 if (atomic && top == NULL && len < mlen)
826 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
827 resid = uio->uio_resid;
830 top->m_pkthdr.len += len;
836 top->m_flags |= M_EOR;
839 } while (space > 0 && atomic);
841 so->so_options |= SO_DONTROUTE;
842 if (flags & MSG_OOB) {
843 pru_flags = PRUS_OOB;
844 } else if ((flags & MSG_EOF) &&
845 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
848 * If the user set MSG_EOF, the protocol
849 * understands this flag and nothing left to
850 * send then use PRU_SEND_EOF instead of PRU_SEND.
852 pru_flags = PRUS_EOF;
853 } else if (resid > 0 && space > 0) {
854 /* If there is more to send, set PRUS_MORETOCOME */
855 pru_flags = PRUS_MORETOCOME;
860 * XXX all the SS_CANTSENDMORE checks previously
861 * done could be out of date. We could have recieved
862 * a reset packet in an interrupt or maybe we slept
863 * while doing page faults in uiomove() etc. We could
864 * probably recheck again inside the splnet() protection
865 * here, but there are probably other places that this
866 * also happens. We must rethink this.
868 error = so_pru_send(so, pru_flags, top, addr, control, td);
870 so->so_options &= ~SO_DONTROUTE;
877 } while (resid && space > 0);
881 ssb_unlock(&so->so_snd);
892 * A specialization of sosend() for UDP based on protocol-specific knowledge:
893 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
894 * sosendallatonce() returns true,
895 * the "atomic" variable is true,
896 * and sosendudp() blocks until space is available for the entire send.
897 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
898 * PR_IMPLOPCL flags set.
899 * UDP has no out-of-band data.
900 * UDP has no control data.
901 * UDP does not support MSG_EOR.
904 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
905 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
908 int error, pru_flags = 0;
911 if (td->td_lwp != NULL)
912 td->td_lwp->lwp_ru.ru_msgsnd++;
916 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
917 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
920 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
924 if (so->so_state & SS_CANTSENDMORE)
927 error = so->so_error;
931 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
932 gotoerr(EDESTADDRREQ);
933 if (resid > so->so_snd.ssb_hiwat)
935 space = ssb_space(&so->so_snd);
936 if (uio && (space < 0 || (size_t)space < resid)) {
937 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
938 gotoerr(EWOULDBLOCK);
939 ssb_unlock(&so->so_snd);
940 error = ssb_wait(&so->so_snd);
947 int hdrlen = max_hdr;
950 * We try to optimize out the additional mbuf
951 * allocations in M_PREPEND() on output path, e.g.
952 * - udp_output(), when it tries to prepend protocol
954 * - Link layer output function, when it tries to
955 * prepend link layer header.
957 * This probably will not benefit any data that will
958 * be fragmented, so this optimization is only performed
959 * when the size of data and max size of protocol+link
960 * headers fit into one mbuf cluster.
962 if (uio->uio_resid > MCLBYTES - hdrlen ||
963 !udp_sosend_prepend) {
964 top = m_uiomove(uio);
970 top = m_getl(uio->uio_resid + hdrlen, MB_WAIT,
971 MT_DATA, M_PKTHDR, &nsize);
972 KASSERT(nsize >= uio->uio_resid + hdrlen,
973 ("sosendudp invalid nsize %d, "
974 "resid %zu, hdrlen %d",
975 nsize, uio->uio_resid, hdrlen));
977 top->m_len = uio->uio_resid;
978 top->m_pkthdr.len = uio->uio_resid;
979 top->m_data += hdrlen;
981 error = uiomove(mtod(top, caddr_t), top->m_len, uio);
987 if (flags & MSG_DONTROUTE)
988 pru_flags |= PRUS_DONTROUTE;
990 if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
991 so_pru_send_async(so, pru_flags, top, addr, NULL, td);
994 error = so_pru_send(so, pru_flags, top, addr, NULL, td);
996 top = NULL; /* sent or freed in lower layer */
999 ssb_unlock(&so->so_snd);
1007 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1008 struct mbuf *top, struct mbuf *control, int flags,
1020 KKASSERT(top == NULL);
1022 resid = uio->uio_resid;
1025 resid = (size_t)top->m_pkthdr.len;
1028 for (m = top; m; m = m->m_next)
1030 KKASSERT(top->m_pkthdr.len == len);
1035 * WARNING! resid is unsigned, space and len are signed. space
1036 * can wind up negative if the sockbuf is overcommitted.
1038 * Also check to make sure that MSG_EOR isn't used on TCP
1040 if (flags & MSG_EOR) {
1046 /* TCP doesn't do control messages (rights, creds, etc) */
1047 if (control->m_len) {
1051 m_freem(control); /* empty control, just free it */
1055 if (td->td_lwp != NULL)
1056 td->td_lwp->lwp_ru.ru_msgsnd++;
1058 #define gotoerr(errcode) { error = errcode; goto release; }
1061 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1066 if (so->so_state & SS_CANTSENDMORE)
1069 error = so->so_error;
1073 if ((so->so_state & SS_ISCONNECTED) == 0 &&
1074 (so->so_state & SS_ISCONFIRMING) == 0)
1076 if (allatonce && resid > so->so_snd.ssb_hiwat)
1079 space = ssb_space_prealloc(&so->so_snd);
1080 if (flags & MSG_OOB)
1082 if ((space < 0 || (size_t)space < resid) && !allatonce &&
1083 space < so->so_snd.ssb_lowat) {
1084 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1085 gotoerr(EWOULDBLOCK);
1086 ssb_unlock(&so->so_snd);
1087 error = ssb_wait(&so->so_snd);
1094 int cnt = 0, async = 0;
1098 * Data is prepackaged in "top".
1102 if (resid > INT_MAX)
1104 m = m_getl((int)resid, MB_WAIT, MT_DATA,
1105 top == NULL ? M_PKTHDR : 0, &mlen);
1107 m->m_pkthdr.len = 0;
1108 m->m_pkthdr.rcvif = NULL;
1110 len = imin((int)szmin(mlen, resid), space);
1112 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1113 resid = uio->uio_resid;
1116 top->m_pkthdr.len += len;
1123 } while (space > 0 && cnt < tcp_sosend_agglim);
1125 if (tcp_sosend_async)
1128 if (flags & MSG_OOB) {
1129 pru_flags = PRUS_OOB;
1131 } else if ((flags & MSG_EOF) && resid == 0) {
1132 pru_flags = PRUS_EOF;
1133 } else if (resid > 0 && space > 0) {
1134 /* If there is more to send, set PRUS_MORETOCOME */
1135 pru_flags = PRUS_MORETOCOME;
1141 if (flags & MSG_SYNC)
1145 * XXX all the SS_CANTSENDMORE checks previously
1146 * done could be out of date. We could have recieved
1147 * a reset packet in an interrupt or maybe we slept
1148 * while doing page faults in uiomove() etc. We could
1149 * probably recheck again inside the splnet() protection
1150 * here, but there are probably other places that this
1151 * also happens. We must rethink this.
1153 for (m = top; m; m = m->m_next)
1154 ssb_preallocstream(&so->so_snd, m);
1156 error = so_pru_send(so, pru_flags, top,
1159 so_pru_send_async(so, pru_flags, top,
1168 } while (resid && space > 0);
1172 ssb_unlock(&so->so_snd);
1183 * Implement receive operations on a socket.
1185 * We depend on the way that records are added to the signalsockbuf
1186 * by sbappend*. In particular, each record (mbufs linked through m_next)
1187 * must begin with an address if the protocol so specifies,
1188 * followed by an optional mbuf or mbufs containing ancillary data,
1189 * and then zero or more mbufs of data.
1191 * Although the signalsockbuf is locked, new data may still be appended.
1192 * A token inside the ssb_lock deals with MP issues and still allows
1193 * the network to access the socket if we block in a uio.
1195 * The caller may receive the data as a single mbuf chain by supplying
1196 * an mbuf **mp0 for use in returning the chain. The uio is then used
1197 * only for the count in uio_resid.
1200 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1201 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1204 struct mbuf *free_chain = NULL;
1205 int flags, len, error, offset;
1206 struct protosw *pr = so->so_proto;
1208 size_t resid, orig_resid;
1211 resid = uio->uio_resid;
1213 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1221 flags = *flagsp &~ MSG_EOR;
1224 if (flags & MSG_OOB) {
1225 m = m_get(MB_WAIT, MT_DATA);
1228 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1234 KKASSERT(resid >= (size_t)m->m_len);
1235 resid -= (size_t)m->m_len;
1236 } while (resid > 0 && m);
1239 uio->uio_resid = resid;
1240 error = uiomove(mtod(m, caddr_t),
1241 (int)szmin(resid, m->m_len),
1243 resid = uio->uio_resid;
1245 } while (uio->uio_resid && error == 0 && m);
1252 if ((so->so_state & SS_ISCONFIRMING) && resid)
1256 * The token interlocks against the protocol thread while
1257 * ssb_lock is a blocking lock against other userland entities.
1259 lwkt_gettoken(&so->so_rcv.ssb_token);
1261 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1265 m = so->so_rcv.ssb_mb;
1267 * If we have less data than requested, block awaiting more
1268 * (subject to any timeout) if:
1269 * 1. the current count is less than the low water mark, or
1270 * 2. MSG_WAITALL is set, and it is possible to do the entire
1271 * receive operation at once if we block (resid <= hiwat).
1272 * 3. MSG_DONTWAIT is not set
1273 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1274 * we have to do the receive in sections, and thus risk returning
1275 * a short count if a timeout or signal occurs after we start.
1277 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1278 (size_t)so->so_rcv.ssb_cc < resid) &&
1279 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1280 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1281 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1282 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1286 error = so->so_error;
1287 if ((flags & MSG_PEEK) == 0)
1291 if (so->so_state & SS_CANTRCVMORE) {
1297 for (; m; m = m->m_next) {
1298 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1299 m = so->so_rcv.ssb_mb;
1303 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1304 (pr->pr_flags & PR_CONNREQUIRED)) {
1310 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1311 error = EWOULDBLOCK;
1314 ssb_unlock(&so->so_rcv);
1315 error = ssb_wait(&so->so_rcv);
1321 if (uio && uio->uio_td && uio->uio_td->td_proc)
1322 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1325 * note: m should be == sb_mb here. Cache the next record while
1326 * cleaning up. Note that calling m_free*() will break out critical
1329 KKASSERT(m == so->so_rcv.ssb_mb);
1332 * Skip any address mbufs prepending the record.
1334 if (pr->pr_flags & PR_ADDR) {
1335 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1338 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1339 if (flags & MSG_PEEK)
1342 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1346 * Skip any control mbufs prepending the record.
1349 if (pr->pr_flags & PR_ADDR_OPT) {
1351 * For SCTP we may be getting a
1352 * whole message OR a partial delivery.
1354 if (m && m->m_type == MT_SONAME) {
1357 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1358 if (flags & MSG_PEEK)
1361 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1365 while (m && m->m_type == MT_CONTROL && error == 0) {
1366 if (flags & MSG_PEEK) {
1368 *controlp = m_copy(m, 0, m->m_len);
1369 m = m->m_next; /* XXX race */
1372 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1373 if (pr->pr_domain->dom_externalize &&
1374 mtod(m, struct cmsghdr *)->cmsg_type ==
1376 error = (*pr->pr_domain->dom_externalize)(m);
1380 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1383 if (controlp && *controlp) {
1385 controlp = &(*controlp)->m_next;
1394 if (type == MT_OOBDATA)
1399 * Copy to the UIO or mbuf return chain (*mp).
1403 while (m && resid > 0 && error == 0) {
1404 if (m->m_type == MT_OOBDATA) {
1405 if (type != MT_OOBDATA)
1407 } else if (type == MT_OOBDATA)
1410 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1412 soclrstate(so, SS_RCVATMARK);
1413 len = (resid > INT_MAX) ? INT_MAX : resid;
1414 if (so->so_oobmark && len > so->so_oobmark - offset)
1415 len = so->so_oobmark - offset;
1416 if (len > m->m_len - moff)
1417 len = m->m_len - moff;
1420 * Copy out to the UIO or pass the mbufs back to the SIO.
1421 * The SIO is dealt with when we eat the mbuf, but deal
1422 * with the resid here either way.
1425 uio->uio_resid = resid;
1426 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1427 resid = uio->uio_resid;
1431 resid -= (size_t)len;
1435 * Eat the entire mbuf or just a piece of it
1437 if (len == m->m_len - moff) {
1438 if (m->m_flags & M_EOR)
1441 if (m->m_flags & M_NOTIFICATION)
1442 flags |= MSG_NOTIFICATION;
1444 if (flags & MSG_PEEK) {
1449 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1453 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1457 if (flags & MSG_PEEK) {
1461 n = m_copym(m, 0, len, MB_WAIT);
1467 so->so_rcv.ssb_cc -= len;
1470 if (so->so_oobmark) {
1471 if ((flags & MSG_PEEK) == 0) {
1472 so->so_oobmark -= len;
1473 if (so->so_oobmark == 0) {
1474 sosetstate(so, SS_RCVATMARK);
1479 if (offset == so->so_oobmark)
1483 if (flags & MSG_EOR)
1486 * If the MSG_WAITALL flag is set (for non-atomic socket),
1487 * we must not quit until resid == 0 or an error
1488 * termination. If a signal/timeout occurs, return
1489 * with a short count but without error.
1490 * Keep signalsockbuf locked against other readers.
1492 while ((flags & MSG_WAITALL) && m == NULL &&
1493 resid > 0 && !sosendallatonce(so) &&
1494 so->so_rcv.ssb_mb == NULL) {
1495 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1498 * The window might have closed to zero, make
1499 * sure we send an ack now that we've drained
1500 * the buffer or we might end up blocking until
1501 * the idle takes over (5 seconds).
1503 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1504 so_pru_rcvd(so, flags);
1505 error = ssb_wait(&so->so_rcv);
1507 ssb_unlock(&so->so_rcv);
1511 m = so->so_rcv.ssb_mb;
1516 * If an atomic read was requested but unread data still remains
1517 * in the record, set MSG_TRUNC.
1519 if (m && pr->pr_flags & PR_ATOMIC)
1523 * Cleanup. If an atomic read was requested drop any unread data.
1525 if ((flags & MSG_PEEK) == 0) {
1526 if (m && (pr->pr_flags & PR_ATOMIC))
1527 sbdroprecord(&so->so_rcv.sb);
1528 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1529 so_pru_rcvd(so, flags);
1532 if (orig_resid == resid && orig_resid &&
1533 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1534 ssb_unlock(&so->so_rcv);
1541 ssb_unlock(&so->so_rcv);
1543 lwkt_reltoken(&so->so_rcv.ssb_token);
1545 m_freem(free_chain);
1550 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1551 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1554 struct mbuf *free_chain = NULL;
1555 int flags, len, error, offset;
1556 struct protosw *pr = so->so_proto;
1558 size_t resid, orig_resid;
1561 resid = uio->uio_resid;
1563 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1571 flags = *flagsp &~ MSG_EOR;
1574 if (flags & MSG_OOB) {
1575 m = m_get(MB_WAIT, MT_DATA);
1578 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1584 KKASSERT(resid >= (size_t)m->m_len);
1585 resid -= (size_t)m->m_len;
1586 } while (resid > 0 && m);
1589 uio->uio_resid = resid;
1590 error = uiomove(mtod(m, caddr_t),
1591 (int)szmin(resid, m->m_len),
1593 resid = uio->uio_resid;
1595 } while (uio->uio_resid && error == 0 && m);
1604 * The token interlocks against the protocol thread while
1605 * ssb_lock is a blocking lock against other userland entities.
1607 lwkt_gettoken(&so->so_rcv.ssb_token);
1609 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1613 m = so->so_rcv.ssb_mb;
1615 * If we have less data than requested, block awaiting more
1616 * (subject to any timeout) if:
1617 * 1. the current count is less than the low water mark, or
1618 * 2. MSG_WAITALL is set, and it is possible to do the entire
1619 * receive operation at once if we block (resid <= hiwat).
1620 * 3. MSG_DONTWAIT is not set
1621 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1622 * we have to do the receive in sections, and thus risk returning
1623 * a short count if a timeout or signal occurs after we start.
1625 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1626 (size_t)so->so_rcv.ssb_cc < resid) &&
1627 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1628 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1629 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1633 error = so->so_error;
1634 if ((flags & MSG_PEEK) == 0)
1638 if (so->so_state & SS_CANTRCVMORE) {
1644 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1645 (pr->pr_flags & PR_CONNREQUIRED)) {
1651 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1652 error = EWOULDBLOCK;
1655 ssb_unlock(&so->so_rcv);
1656 error = ssb_wait(&so->so_rcv);
1662 if (uio && uio->uio_td && uio->uio_td->td_proc)
1663 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1666 * note: m should be == sb_mb here. Cache the next record while
1667 * cleaning up. Note that calling m_free*() will break out critical
1670 KKASSERT(m == so->so_rcv.ssb_mb);
1673 * Copy to the UIO or mbuf return chain (*mp).
1677 while (m && resid > 0 && error == 0) {
1678 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1681 soclrstate(so, SS_RCVATMARK);
1682 len = (resid > INT_MAX) ? INT_MAX : resid;
1683 if (so->so_oobmark && len > so->so_oobmark - offset)
1684 len = so->so_oobmark - offset;
1685 if (len > m->m_len - moff)
1686 len = m->m_len - moff;
1689 * Copy out to the UIO or pass the mbufs back to the SIO.
1690 * The SIO is dealt with when we eat the mbuf, but deal
1691 * with the resid here either way.
1694 uio->uio_resid = resid;
1695 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1696 resid = uio->uio_resid;
1700 resid -= (size_t)len;
1704 * Eat the entire mbuf or just a piece of it
1706 if (len == m->m_len - moff) {
1707 if (flags & MSG_PEEK) {
1712 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1716 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1720 if (flags & MSG_PEEK) {
1724 n = m_copym(m, 0, len, MB_WAIT);
1730 so->so_rcv.ssb_cc -= len;
1733 if (so->so_oobmark) {
1734 if ((flags & MSG_PEEK) == 0) {
1735 so->so_oobmark -= len;
1736 if (so->so_oobmark == 0) {
1737 sosetstate(so, SS_RCVATMARK);
1742 if (offset == so->so_oobmark)
1747 * If the MSG_WAITALL flag is set (for non-atomic socket),
1748 * we must not quit until resid == 0 or an error
1749 * termination. If a signal/timeout occurs, return
1750 * with a short count but without error.
1751 * Keep signalsockbuf locked against other readers.
1753 while ((flags & MSG_WAITALL) && m == NULL &&
1754 resid > 0 && !sosendallatonce(so) &&
1755 so->so_rcv.ssb_mb == NULL) {
1756 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1759 * The window might have closed to zero, make
1760 * sure we send an ack now that we've drained
1761 * the buffer or we might end up blocking until
1762 * the idle takes over (5 seconds).
1765 so_pru_rcvd_async(so);
1766 error = ssb_wait(&so->so_rcv);
1768 ssb_unlock(&so->so_rcv);
1772 m = so->so_rcv.ssb_mb;
1777 * Cleanup. If an atomic read was requested drop any unread data.
1779 if ((flags & MSG_PEEK) == 0) {
1781 so_pru_rcvd_async(so);
1784 if (orig_resid == resid && orig_resid &&
1785 (so->so_state & SS_CANTRCVMORE) == 0) {
1786 ssb_unlock(&so->so_rcv);
1793 ssb_unlock(&so->so_rcv);
1795 lwkt_reltoken(&so->so_rcv.ssb_token);
1797 m_freem(free_chain);
1802 * Shut a socket down. Note that we do not get a frontend lock as we
1803 * want to be able to shut the socket down even if another thread is
1804 * blocked in a read(), thus waking it up.
1807 soshutdown(struct socket *so, int how)
1809 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1812 if (how != SHUT_WR) {
1813 /*ssb_lock(&so->so_rcv, M_WAITOK);*/
1815 /*ssb_unlock(&so->so_rcv);*/
1818 return (so_pru_shutdown(so));
1823 sorflush(struct socket *so)
1825 struct signalsockbuf *ssb = &so->so_rcv;
1826 struct protosw *pr = so->so_proto;
1827 struct signalsockbuf asb;
1829 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1831 lwkt_gettoken(&ssb->ssb_token);
1836 * Can't just blow up the ssb structure here
1838 bzero(&ssb->sb, sizeof(ssb->sb));
1843 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1845 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1846 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1847 ssb_release(&asb, so);
1849 lwkt_reltoken(&ssb->ssb_token);
1854 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1856 struct accept_filter_arg *afap = NULL;
1857 struct accept_filter *afp;
1858 struct so_accf *af = so->so_accf;
1861 /* do not set/remove accept filters on non listen sockets */
1862 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1867 /* removing the filter */
1870 if (af->so_accept_filter != NULL &&
1871 af->so_accept_filter->accf_destroy != NULL) {
1872 af->so_accept_filter->accf_destroy(so);
1874 if (af->so_accept_filter_str != NULL) {
1875 kfree(af->so_accept_filter_str, M_ACCF);
1880 so->so_options &= ~SO_ACCEPTFILTER;
1883 /* adding a filter */
1884 /* must remove previous filter first */
1889 /* don't put large objects on the kernel stack */
1890 afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
1891 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1892 afap->af_name[sizeof(afap->af_name)-1] = '\0';
1893 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1896 afp = accept_filt_get(afap->af_name);
1901 af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1902 if (afp->accf_create != NULL) {
1903 if (afap->af_name[0] != '\0') {
1904 int len = strlen(afap->af_name) + 1;
1906 af->so_accept_filter_str = kmalloc(len, M_ACCF,
1908 strcpy(af->so_accept_filter_str, afap->af_name);
1910 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1911 if (af->so_accept_filter_arg == NULL) {
1912 kfree(af->so_accept_filter_str, M_ACCF);
1919 af->so_accept_filter = afp;
1921 so->so_options |= SO_ACCEPTFILTER;
1924 kfree(afap, M_TEMP);
1930 * Perhaps this routine, and sooptcopyout(), below, ought to come in
1931 * an additional variant to handle the case where the option value needs
1932 * to be some kind of integer, but not a specific size.
1933 * In addition to their use here, these functions are also called by the
1934 * protocol-level pr_ctloutput() routines.
1937 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1939 return soopt_to_kbuf(sopt, buf, len, minlen);
1943 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1947 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1948 KKASSERT(kva_p(buf));
1951 * If the user gives us more than we wanted, we ignore it,
1952 * but if we don't get the minimum length the caller
1953 * wants, we return EINVAL. On success, sopt->sopt_valsize
1954 * is set to however much we actually retrieved.
1956 if ((valsize = sopt->sopt_valsize) < minlen)
1959 sopt->sopt_valsize = valsize = len;
1961 bcopy(sopt->sopt_val, buf, valsize);
1967 sosetopt(struct socket *so, struct sockopt *sopt)
1973 struct signalsockbuf *sotmp;
1976 sopt->sopt_dir = SOPT_SET;
1977 if (sopt->sopt_level != SOL_SOCKET) {
1978 if (so->so_proto && so->so_proto->pr_ctloutput) {
1979 return (so_pr_ctloutput(so, sopt));
1981 error = ENOPROTOOPT;
1983 switch (sopt->sopt_name) {
1985 case SO_ACCEPTFILTER:
1986 error = do_setopt_accept_filter(so, sopt);
1992 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1996 so->so_linger = l.l_linger;
1998 so->so_options |= SO_LINGER;
2000 so->so_options &= ~SO_LINGER;
2006 case SO_USELOOPBACK:
2013 error = sooptcopyin(sopt, &optval, sizeof optval,
2018 so->so_options |= sopt->sopt_name;
2020 so->so_options &= ~sopt->sopt_name;
2027 error = sooptcopyin(sopt, &optval, sizeof optval,
2033 * Values < 1 make no sense for any of these
2034 * options, so disallow them.
2041 switch (sopt->sopt_name) {
2044 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2045 &so->so_snd : &so->so_rcv, (u_long)optval,
2047 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2051 sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2052 &so->so_snd : &so->so_rcv;
2053 atomic_clear_int(&sotmp->ssb_flags,
2058 * Make sure the low-water is never greater than
2062 so->so_snd.ssb_lowat =
2063 (optval > so->so_snd.ssb_hiwat) ?
2064 so->so_snd.ssb_hiwat : optval;
2065 atomic_clear_int(&so->so_snd.ssb_flags,
2069 so->so_rcv.ssb_lowat =
2070 (optval > so->so_rcv.ssb_hiwat) ?
2071 so->so_rcv.ssb_hiwat : optval;
2072 atomic_clear_int(&so->so_rcv.ssb_flags,
2080 error = sooptcopyin(sopt, &tv, sizeof tv,
2085 /* assert(hz > 0); */
2086 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2087 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2091 /* assert(tick > 0); */
2092 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2093 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2094 if (val > INT_MAX) {
2098 if (val == 0 && tv.tv_usec != 0)
2101 switch (sopt->sopt_name) {
2103 so->so_snd.ssb_timeo = val;
2106 so->so_rcv.ssb_timeo = val;
2111 error = ENOPROTOOPT;
2114 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2115 (void) so_pr_ctloutput(so, sopt);
2122 /* Helper routine for getsockopt */
2124 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2126 soopt_from_kbuf(sopt, buf, len);
2131 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2136 sopt->sopt_valsize = 0;
2140 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2141 KKASSERT(kva_p(buf));
2144 * Documented get behavior is that we always return a value,
2145 * possibly truncated to fit in the user's buffer.
2146 * Traditional behavior is that we always tell the user
2147 * precisely how much we copied, rather than something useful
2148 * like the total amount we had available for her.
2149 * Note that this interface is not idempotent; the entire answer must
2150 * generated ahead of time.
2152 valsize = szmin(len, sopt->sopt_valsize);
2153 sopt->sopt_valsize = valsize;
2154 if (sopt->sopt_val != 0) {
2155 bcopy(buf, sopt->sopt_val, valsize);
2160 sogetopt(struct socket *so, struct sockopt *sopt)
2167 struct accept_filter_arg *afap;
2171 sopt->sopt_dir = SOPT_GET;
2172 if (sopt->sopt_level != SOL_SOCKET) {
2173 if (so->so_proto && so->so_proto->pr_ctloutput) {
2174 return (so_pr_ctloutput(so, sopt));
2176 return (ENOPROTOOPT);
2178 switch (sopt->sopt_name) {
2180 case SO_ACCEPTFILTER:
2181 if ((so->so_options & SO_ACCEPTCONN) == 0)
2183 afap = kmalloc(sizeof(*afap), M_TEMP,
2185 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2186 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2187 if (so->so_accf->so_accept_filter_str != NULL)
2188 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2190 error = sooptcopyout(sopt, afap, sizeof(*afap));
2191 kfree(afap, M_TEMP);
2196 l.l_onoff = so->so_options & SO_LINGER;
2197 l.l_linger = so->so_linger;
2198 error = sooptcopyout(sopt, &l, sizeof l);
2201 case SO_USELOOPBACK:
2211 optval = so->so_options & sopt->sopt_name;
2213 error = sooptcopyout(sopt, &optval, sizeof optval);
2217 optval = so->so_type;
2221 optval = so->so_error;
2226 optval = so->so_snd.ssb_hiwat;
2230 optval = so->so_rcv.ssb_hiwat;
2234 optval = so->so_snd.ssb_lowat;
2238 optval = so->so_rcv.ssb_lowat;
2243 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2244 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2246 tv.tv_sec = optval / hz;
2247 tv.tv_usec = (optval % hz) * ustick;
2248 error = sooptcopyout(sopt, &tv, sizeof tv);
2252 optval_l = ssb_space(&so->so_snd);
2253 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2257 error = ENOPROTOOPT;
2264 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2266 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2268 struct mbuf *m, *m_prev;
2269 int sopt_size = sopt->sopt_valsize, msize;
2271 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
2275 m->m_len = min(msize, sopt_size);
2276 sopt_size -= m->m_len;
2280 while (sopt_size > 0) {
2281 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
2282 MT_DATA, 0, &msize);
2287 m->m_len = min(msize, sopt_size);
2288 sopt_size -= m->m_len;
2295 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2297 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2299 soopt_to_mbuf(sopt, m);
2304 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2309 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2311 if (sopt->sopt_val == NULL)
2313 val = sopt->sopt_val;
2314 valsize = sopt->sopt_valsize;
2315 while (m != NULL && valsize >= m->m_len) {
2316 bcopy(val, mtod(m, char *), m->m_len);
2317 valsize -= m->m_len;
2318 val = (caddr_t)val + m->m_len;
2321 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2322 panic("ip6_sooptmcopyin");
2325 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2327 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2329 return soopt_from_mbuf(sopt, m);
2333 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2335 struct mbuf *m0 = m;
2340 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2342 if (sopt->sopt_val == NULL)
2344 val = sopt->sopt_val;
2345 maxsize = sopt->sopt_valsize;
2346 while (m != NULL && maxsize >= m->m_len) {
2347 bcopy(mtod(m, char *), val, m->m_len);
2348 maxsize -= m->m_len;
2349 val = (caddr_t)val + m->m_len;
2350 valsize += m->m_len;
2354 /* enough soopt buffer should be given from user-land */
2358 sopt->sopt_valsize = valsize;
2363 sohasoutofband(struct socket *so)
2365 if (so->so_sigio != NULL)
2366 pgsigio(so->so_sigio, SIGURG, 0);
2367 KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
2371 sokqfilter(struct file *fp, struct knote *kn)
2373 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2374 struct signalsockbuf *ssb;
2376 switch (kn->kn_filter) {
2378 if (so->so_options & SO_ACCEPTCONN)
2379 kn->kn_fop = &solisten_filtops;
2381 kn->kn_fop = &soread_filtops;
2385 kn->kn_fop = &sowrite_filtops;
2389 kn->kn_fop = &soexcept_filtops;
2393 return (EOPNOTSUPP);
2396 knote_insert(&ssb->ssb_kq.ki_note, kn);
2397 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2402 filt_sordetach(struct knote *kn)
2404 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2406 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2407 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2408 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2413 filt_soread(struct knote *kn, long hint)
2415 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2417 if (kn->kn_sfflags & NOTE_OOB) {
2418 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2419 kn->kn_fflags |= NOTE_OOB;
2424 kn->kn_data = so->so_rcv.ssb_cc;
2426 if (so->so_state & SS_CANTRCVMORE) {
2428 * Only set NODATA if all data has been exhausted.
2430 if (kn->kn_data == 0)
2431 kn->kn_flags |= EV_NODATA;
2432 kn->kn_flags |= EV_EOF;
2433 kn->kn_fflags = so->so_error;
2436 if (so->so_error) /* temporary udp error */
2438 if (kn->kn_sfflags & NOTE_LOWAT)
2439 return (kn->kn_data >= kn->kn_sdata);
2440 return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2441 !TAILQ_EMPTY(&so->so_comp));
2445 filt_sowdetach(struct knote *kn)
2447 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2449 knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2450 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2451 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2456 filt_sowrite(struct knote *kn, long hint)
2458 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2460 kn->kn_data = ssb_space(&so->so_snd);
2461 if (so->so_state & SS_CANTSENDMORE) {
2462 kn->kn_flags |= (EV_EOF | EV_NODATA);
2463 kn->kn_fflags = so->so_error;
2466 if (so->so_error) /* temporary udp error */
2468 if (((so->so_state & SS_ISCONNECTED) == 0) &&
2469 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2471 if (kn->kn_sfflags & NOTE_LOWAT)
2472 return (kn->kn_data >= kn->kn_sdata);
2473 return (kn->kn_data >= so->so_snd.ssb_lowat);
2478 filt_solisten(struct knote *kn, long hint)
2480 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2482 kn->kn_data = so->so_qlen;
2483 return (! TAILQ_EMPTY(&so->so_comp));