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>
103 extern int tcp_sosend_agglim;
104 extern int tcp_sosend_async;
105 extern int udp_sosend_async;
106 extern int udp_sosend_prepend;
108 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
111 static void filt_sordetach(struct knote *kn);
112 static int filt_soread(struct knote *kn, long hint);
113 static void filt_sowdetach(struct knote *kn);
114 static int filt_sowrite(struct knote *kn, long hint);
115 static int filt_solisten(struct knote *kn, long hint);
117 static void sodiscard(struct socket *so);
118 static int soclose_sync(struct socket *so, int fflag);
119 static void soclose_fast(struct socket *so);
121 static struct filterops solisten_filtops =
122 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten };
123 static struct filterops soread_filtops =
124 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
125 static struct filterops sowrite_filtops =
126 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite };
127 static struct filterops soexcept_filtops =
128 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
130 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
131 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
132 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
135 static int somaxconn = SOMAXCONN;
136 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
137 &somaxconn, 0, "Maximum pending socket connection queue size");
139 static int use_soclose_fast = 1;
140 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW,
141 &use_soclose_fast, 0, "Fast socket close");
143 int use_soaccept_pred_fast = 1;
144 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW,
145 &use_soaccept_pred_fast, 0, "Fast socket accept predication");
147 int use_sendfile_async = 1;
148 SYSCTL_INT(_kern_ipc, OID_AUTO, sendfile_async, CTLFLAG_RW,
149 &use_sendfile_async, 0, "sendfile uses asynchronized pru_send");
152 * Socket operation routines.
153 * These routines are called by the routines in
154 * sys_socket.c or from a system process, and
155 * implement the semantics of socket operations by
156 * switching out to the protocol specific routines.
160 * Get a socket structure, and initialize it.
161 * Note that it would probably be better to allocate socket
162 * and PCB at the same time, but I'm not convinced that all
163 * the protocols can be easily modified to do this.
166 soalloc(int waitok, struct protosw *pr)
171 waitmask = waitok ? M_WAITOK : M_NOWAIT;
172 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
174 /* XXX race condition for reentrant kernel */
176 TAILQ_INIT(&so->so_aiojobq);
177 TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist);
178 TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist);
179 lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
180 lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
181 spin_init(&so->so_rcvd_spin);
182 netmsg_init(&so->so_rcvd_msg.base, so, &netisr_adone_rport,
183 MSGF_DROPABLE, so->so_proto->pr_usrreqs->pru_rcvd);
184 so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC;
185 so->so_state = SS_NOFDREF;
192 socreate(int dom, struct socket **aso, int type,
193 int proto, struct thread *td)
195 struct proc *p = td->td_proc;
198 struct pru_attach_info ai;
202 prp = pffindproto(dom, proto, type);
204 prp = pffindtype(dom, type);
206 if (prp == NULL || prp->pr_usrreqs->pru_attach == 0)
207 return (EPROTONOSUPPORT);
209 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
210 prp->pr_domain->dom_family != PF_LOCAL &&
211 prp->pr_domain->dom_family != PF_INET &&
212 prp->pr_domain->dom_family != PF_INET6 &&
213 prp->pr_domain->dom_family != PF_ROUTE) {
214 return (EPROTONOSUPPORT);
217 if (prp->pr_type != type)
219 so = soalloc(p != NULL, prp);
224 * Callers of socreate() presumably will connect up a descriptor
225 * and call soclose() if they cannot. This represents our so_refs
226 * (which should be 1) from soalloc().
228 soclrstate(so, SS_NOFDREF);
231 * Set a default port for protocol processing. No action will occur
232 * on the socket on this port until an inpcb is attached to it and
233 * is able to match incoming packets, or until the socket becomes
234 * available to userland.
236 * We normally default the socket to the protocol thread on cpu 0.
237 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
238 * thread and all pr_*()/pru_*() calls are executed synchronously.
240 if (prp->pr_flags & PR_SYNC_PORT)
241 so->so_port = &netisr_sync_port;
243 so->so_port = netisr_portfn(0);
245 TAILQ_INIT(&so->so_incomp);
246 TAILQ_INIT(&so->so_comp);
248 so->so_cred = crhold(p->p_ucred);
249 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
250 ai.p_ucred = p->p_ucred;
251 ai.fd_rdir = p->p_fd->fd_rdir;
254 * Auto-sizing of socket buffers is managed by the protocols and
255 * the appropriate flags must be set in the pru_attach function.
257 error = so_pru_attach(so, proto, &ai);
259 sosetstate(so, SS_NOFDREF);
260 sofree(so); /* from soalloc */
265 * NOTE: Returns referenced socket.
272 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
276 error = so_pru_bind(so, nam, td);
281 sodealloc(struct socket *so)
283 if (so->so_rcv.ssb_hiwat)
284 (void)chgsbsize(so->so_cred->cr_uidinfo,
285 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
286 if (so->so_snd.ssb_hiwat)
287 (void)chgsbsize(so->so_cred->cr_uidinfo,
288 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
290 /* remove accept filter if present */
291 if (so->so_accf != NULL)
292 do_setopt_accept_filter(so, NULL);
295 if (so->so_faddr != NULL)
296 kfree(so->so_faddr, M_SONAME);
301 solisten(struct socket *so, int backlog, struct thread *td)
305 short oldopt, oldqlimit;
308 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
312 oldopt = so->so_options;
313 oldqlimit = so->so_qlimit;
316 lwkt_gettoken(&so->so_rcv.ssb_token);
317 if (TAILQ_EMPTY(&so->so_comp))
318 so->so_options |= SO_ACCEPTCONN;
319 lwkt_reltoken(&so->so_rcv.ssb_token);
320 if (backlog < 0 || backlog > somaxconn)
322 so->so_qlimit = backlog;
323 /* SCTP needs to look at tweak both the inbound backlog parameter AND
324 * the so_options (UDP model both connect's and gets inbound
325 * connections .. implicitly).
327 error = so_pru_listen(so, td);
330 /* Restore the params */
331 so->so_options = oldopt;
332 so->so_qlimit = oldqlimit;
340 * Destroy a disconnected socket. This routine is a NOP if entities
341 * still have a reference on the socket:
343 * so_pcb - The protocol stack still has a reference
344 * SS_NOFDREF - There is no longer a file pointer reference
347 sofree(struct socket *so)
352 * This is a bit hackish at the moment. We need to interlock
353 * any accept queue we are on before we potentially lose the
354 * last reference to avoid races against a re-reference from
355 * someone operating on the queue.
357 while ((head = so->so_head) != NULL) {
358 lwkt_getpooltoken(head);
359 if (so->so_head == head)
361 lwkt_relpooltoken(head);
365 * Arbitrage the last free.
367 KKASSERT(so->so_refs > 0);
368 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
370 lwkt_relpooltoken(head);
374 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
375 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
378 * We're done, remove ourselves from the accept queue we are
379 * on, if we are on one.
382 if (so->so_state & SS_INCOMP) {
383 TAILQ_REMOVE(&head->so_incomp, so, so_list);
385 } else if (so->so_state & SS_COMP) {
387 * We must not decommission a socket that's
388 * on the accept(2) queue. If we do, then
389 * accept(2) may hang after select(2) indicated
390 * that the listening socket was ready.
392 lwkt_relpooltoken(head);
395 panic("sofree: not queued");
397 soclrstate(so, SS_INCOMP);
399 lwkt_relpooltoken(head);
401 ssb_release(&so->so_snd, so);
407 * Close a socket on last file table reference removal.
408 * Initiate disconnect if connected.
409 * Free socket when disconnect complete.
412 soclose(struct socket *so, int fflag)
416 funsetown(&so->so_sigio);
417 if (!use_soclose_fast ||
418 (so->so_proto->pr_flags & PR_SYNC_PORT) ||
419 (so->so_options & SO_LINGER)) {
420 error = soclose_sync(so, fflag);
429 sodiscard(struct socket *so)
431 lwkt_getpooltoken(so);
432 if (so->so_options & SO_ACCEPTCONN) {
435 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
436 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
437 soclrstate(sp, SS_INCOMP);
442 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
443 TAILQ_REMOVE(&so->so_comp, sp, so_list);
444 soclrstate(sp, SS_COMP);
450 lwkt_relpooltoken(so);
452 if (so->so_state & SS_NOFDREF)
453 panic("soclose: NOFDREF");
454 sosetstate(so, SS_NOFDREF); /* take ref */
458 soclose_sync(struct socket *so, int fflag)
462 if (so->so_pcb == NULL)
464 if (so->so_state & SS_ISCONNECTED) {
465 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
466 error = sodisconnect(so);
470 if (so->so_options & SO_LINGER) {
471 if ((so->so_state & SS_ISDISCONNECTING) &&
474 while (so->so_state & SS_ISCONNECTED) {
475 error = tsleep(&so->so_timeo, PCATCH,
476 "soclos", so->so_linger * hz);
486 error2 = so_pru_detach(so);
492 so_pru_sync(so); /* unpend async sending */
493 sofree(so); /* dispose of ref */
499 soclose_sofree_async_handler(netmsg_t msg)
501 sofree(msg->base.nm_so);
505 soclose_sofree_async(struct socket *so)
507 struct netmsg_base *base = &so->so_clomsg;
509 netmsg_init(base, so, &netisr_apanic_rport, 0,
510 soclose_sofree_async_handler);
511 lwkt_sendmsg(so->so_port, &base->lmsg);
515 soclose_disconn_async_handler(netmsg_t msg)
517 struct socket *so = msg->base.nm_so;
519 if ((so->so_state & SS_ISCONNECTED) &&
520 (so->so_state & SS_ISDISCONNECTING) == 0)
521 so_pru_disconnect_direct(so);
524 so_pru_detach_direct(so);
531 soclose_disconn_async(struct socket *so)
533 struct netmsg_base *base = &so->so_clomsg;
535 netmsg_init(base, so, &netisr_apanic_rport, 0,
536 soclose_disconn_async_handler);
537 lwkt_sendmsg(so->so_port, &base->lmsg);
541 soclose_detach_async_handler(netmsg_t msg)
543 struct socket *so = msg->base.nm_so;
546 so_pru_detach_direct(so);
553 soclose_detach_async(struct socket *so)
555 struct netmsg_base *base = &so->so_clomsg;
557 netmsg_init(base, so, &netisr_apanic_rport, 0,
558 soclose_detach_async_handler);
559 lwkt_sendmsg(so->so_port, &base->lmsg);
563 soclose_fast(struct socket *so)
565 if (so->so_pcb == NULL)
568 if ((so->so_state & SS_ISCONNECTED) &&
569 (so->so_state & SS_ISDISCONNECTING) == 0) {
570 soclose_disconn_async(so);
575 soclose_detach_async(so);
581 soclose_sofree_async(so);
585 * Abort and destroy a socket. Only one abort can be in progress
586 * at any given moment.
589 soabort(struct socket *so)
596 soaborta(struct socket *so)
603 soabort_oncpu(struct socket *so)
606 so_pru_abort_oncpu(so);
610 * so is passed in ref'd, which becomes owned by
611 * the cleared SS_NOFDREF flag.
614 soaccept_generic(struct socket *so)
616 if ((so->so_state & SS_NOFDREF) == 0)
617 panic("soaccept: !NOFDREF");
618 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */
622 soaccept(struct socket *so, struct sockaddr **nam)
626 soaccept_generic(so);
627 error = so_pru_accept(so, nam);
632 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
636 if (so->so_options & SO_ACCEPTCONN)
639 * If protocol is connection-based, can only connect once.
640 * Otherwise, if connected, try to disconnect first.
641 * This allows user to disconnect by connecting to, e.g.,
644 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
645 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
646 (error = sodisconnect(so)))) {
650 * Prevent accumulated error from previous connection
654 error = so_pru_connect(so, nam, td);
660 soconnect2(struct socket *so1, struct socket *so2)
664 error = so_pru_connect2(so1, so2);
669 sodisconnect(struct socket *so)
673 if ((so->so_state & SS_ISCONNECTED) == 0) {
677 if (so->so_state & SS_ISDISCONNECTING) {
681 error = so_pru_disconnect(so);
686 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
689 * If send must go all at once and message is larger than
690 * send buffering, then hard error.
691 * Lock against other senders.
692 * If must go all at once and not enough room now, then
693 * inform user that this would block and do nothing.
694 * Otherwise, if nonblocking, send as much as possible.
695 * The data to be sent is described by "uio" if nonzero,
696 * otherwise by the mbuf chain "top" (which must be null
697 * if uio is not). Data provided in mbuf chain must be small
698 * enough to send all at once.
700 * Returns nonzero on error, timeout or signal; callers
701 * must check for short counts if EINTR/ERESTART are returned.
702 * Data and control buffers are freed on return.
705 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
706 struct mbuf *top, struct mbuf *control, int flags,
713 int clen = 0, error, dontroute, mlen;
714 int atomic = sosendallatonce(so) || top;
718 resid = uio->uio_resid;
720 resid = (size_t)top->m_pkthdr.len;
723 for (m = top; m; m = m->m_next)
725 KKASSERT(top->m_pkthdr.len == len);
730 * WARNING! resid is unsigned, space and len are signed. space
731 * can wind up negative if the sockbuf is overcommitted.
733 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
734 * type sockets since that's an error.
736 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
742 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
743 (so->so_proto->pr_flags & PR_ATOMIC);
744 if (td->td_lwp != NULL)
745 td->td_lwp->lwp_ru.ru_msgsnd++;
747 clen = control->m_len;
748 #define gotoerr(errcode) { error = errcode; goto release; }
751 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
756 if (so->so_state & SS_CANTSENDMORE)
759 error = so->so_error;
763 if ((so->so_state & SS_ISCONNECTED) == 0) {
765 * `sendto' and `sendmsg' is allowed on a connection-
766 * based socket if it supports implied connect.
767 * Return ENOTCONN if not connected and no address is
770 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
771 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
772 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
773 !(resid == 0 && clen != 0))
775 } else if (addr == NULL)
776 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
777 ENOTCONN : EDESTADDRREQ);
779 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
780 clen > so->so_snd.ssb_hiwat) {
783 space = ssb_space(&so->so_snd);
786 if ((space < 0 || (size_t)space < resid + clen) && uio &&
787 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
788 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
789 gotoerr(EWOULDBLOCK);
790 ssb_unlock(&so->so_snd);
791 error = ssb_wait(&so->so_snd);
801 * Data is prepackaged in "top".
805 top->m_flags |= M_EOR;
809 m = m_getl((int)resid, MB_WAIT, MT_DATA,
810 top == NULL ? M_PKTHDR : 0, &mlen);
813 m->m_pkthdr.rcvif = NULL;
815 len = imin((int)szmin(mlen, resid), space);
816 if (resid < MINCLSIZE) {
818 * For datagram protocols, leave room
819 * for protocol headers in first mbuf.
821 if (atomic && top == NULL && len < mlen)
825 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
826 resid = uio->uio_resid;
829 top->m_pkthdr.len += len;
835 top->m_flags |= M_EOR;
838 } while (space > 0 && atomic);
840 so->so_options |= SO_DONTROUTE;
841 if (flags & MSG_OOB) {
842 pru_flags = PRUS_OOB;
843 } else if ((flags & MSG_EOF) &&
844 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
847 * If the user set MSG_EOF, the protocol
848 * understands this flag and nothing left to
849 * send then use PRU_SEND_EOF instead of PRU_SEND.
851 pru_flags = PRUS_EOF;
852 } else if (resid > 0 && space > 0) {
853 /* If there is more to send, set PRUS_MORETOCOME */
854 pru_flags = PRUS_MORETOCOME;
859 * XXX all the SS_CANTSENDMORE checks previously
860 * done could be out of date. We could have recieved
861 * a reset packet in an interrupt or maybe we slept
862 * while doing page faults in uiomove() etc. We could
863 * probably recheck again inside the splnet() protection
864 * here, but there are probably other places that this
865 * also happens. We must rethink this.
867 error = so_pru_send(so, pru_flags, top, addr, control, td);
869 so->so_options &= ~SO_DONTROUTE;
876 } while (resid && space > 0);
880 ssb_unlock(&so->so_snd);
891 * A specialization of sosend() for UDP based on protocol-specific knowledge:
892 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
893 * sosendallatonce() returns true,
894 * the "atomic" variable is true,
895 * and sosendudp() blocks until space is available for the entire send.
896 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
897 * PR_IMPLOPCL flags set.
898 * UDP has no out-of-band data.
899 * UDP has no control data.
900 * UDP does not support MSG_EOR.
903 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
904 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
907 int error, pru_flags = 0;
910 if (td->td_lwp != NULL)
911 td->td_lwp->lwp_ru.ru_msgsnd++;
915 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
916 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
919 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
923 if (so->so_state & SS_CANTSENDMORE)
926 error = so->so_error;
930 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
931 gotoerr(EDESTADDRREQ);
932 if (resid > so->so_snd.ssb_hiwat)
934 space = ssb_space(&so->so_snd);
935 if (uio && (space < 0 || (size_t)space < resid)) {
936 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
937 gotoerr(EWOULDBLOCK);
938 ssb_unlock(&so->so_snd);
939 error = ssb_wait(&so->so_snd);
946 int hdrlen = max_hdr;
949 * We try to optimize out the additional mbuf
950 * allocations in M_PREPEND() on output path, e.g.
951 * - udp_output(), when it tries to prepend protocol
953 * - Link layer output function, when it tries to
954 * prepend link layer header.
956 * This probably will not benefit any data that will
957 * be fragmented, so this optimization is only performed
958 * when the size of data and max size of protocol+link
959 * headers fit into one mbuf cluster.
961 if (uio->uio_resid > MCLBYTES - hdrlen ||
962 !udp_sosend_prepend) {
963 top = m_uiomove(uio);
969 top = m_getl(uio->uio_resid + hdrlen, MB_WAIT,
970 MT_DATA, M_PKTHDR, &nsize);
971 KASSERT(nsize >= uio->uio_resid + hdrlen,
972 ("sosendudp invalid nsize %d, "
973 "resid %zu, hdrlen %d",
974 nsize, uio->uio_resid, hdrlen));
976 top->m_len = uio->uio_resid;
977 top->m_pkthdr.len = uio->uio_resid;
978 top->m_data += hdrlen;
980 error = uiomove(mtod(top, caddr_t), top->m_len, uio);
986 if (flags & MSG_DONTROUTE)
987 pru_flags |= PRUS_DONTROUTE;
989 if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
990 so_pru_send_async(so, pru_flags, top, addr, NULL, td);
993 error = so_pru_send(so, pru_flags, top, addr, NULL, td);
995 top = NULL; /* sent or freed in lower layer */
998 ssb_unlock(&so->so_snd);
1006 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1007 struct mbuf *top, struct mbuf *control, int flags,
1019 KKASSERT(top == NULL);
1021 resid = uio->uio_resid;
1024 resid = (size_t)top->m_pkthdr.len;
1027 for (m = top; m; m = m->m_next)
1029 KKASSERT(top->m_pkthdr.len == len);
1034 * WARNING! resid is unsigned, space and len are signed. space
1035 * can wind up negative if the sockbuf is overcommitted.
1037 * Also check to make sure that MSG_EOR isn't used on TCP
1039 if (flags & MSG_EOR) {
1045 /* TCP doesn't do control messages (rights, creds, etc) */
1046 if (control->m_len) {
1050 m_freem(control); /* empty control, just free it */
1054 if (td->td_lwp != NULL)
1055 td->td_lwp->lwp_ru.ru_msgsnd++;
1057 #define gotoerr(errcode) { error = errcode; goto release; }
1060 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1065 if (so->so_state & SS_CANTSENDMORE)
1068 error = so->so_error;
1072 if ((so->so_state & SS_ISCONNECTED) == 0 &&
1073 (so->so_state & SS_ISCONFIRMING) == 0)
1075 if (allatonce && resid > so->so_snd.ssb_hiwat)
1078 space = ssb_space_prealloc(&so->so_snd);
1079 if (flags & MSG_OOB)
1081 if ((space < 0 || (size_t)space < resid) && !allatonce &&
1082 space < so->so_snd.ssb_lowat) {
1083 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1084 gotoerr(EWOULDBLOCK);
1085 ssb_unlock(&so->so_snd);
1086 error = ssb_wait(&so->so_snd);
1093 int cnt = 0, async = 0;
1097 * Data is prepackaged in "top".
1101 if (resid > INT_MAX)
1103 m = m_getl((int)resid, MB_WAIT, MT_DATA,
1104 top == NULL ? M_PKTHDR : 0, &mlen);
1106 m->m_pkthdr.len = 0;
1107 m->m_pkthdr.rcvif = NULL;
1109 len = imin((int)szmin(mlen, resid), space);
1111 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1112 resid = uio->uio_resid;
1115 top->m_pkthdr.len += len;
1122 } while (space > 0 && cnt < tcp_sosend_agglim);
1124 if (tcp_sosend_async)
1127 if (flags & MSG_OOB) {
1128 pru_flags = PRUS_OOB;
1130 } else if ((flags & MSG_EOF) && resid == 0) {
1131 pru_flags = PRUS_EOF;
1132 } else if (resid > 0 && space > 0) {
1133 /* If there is more to send, set PRUS_MORETOCOME */
1134 pru_flags = PRUS_MORETOCOME;
1140 if (flags & MSG_SYNC)
1144 * XXX all the SS_CANTSENDMORE checks previously
1145 * done could be out of date. We could have recieved
1146 * a reset packet in an interrupt or maybe we slept
1147 * while doing page faults in uiomove() etc. We could
1148 * probably recheck again inside the splnet() protection
1149 * here, but there are probably other places that this
1150 * also happens. We must rethink this.
1152 for (m = top; m; m = m->m_next)
1153 ssb_preallocstream(&so->so_snd, m);
1155 error = so_pru_send(so, pru_flags, top,
1158 so_pru_send_async(so, pru_flags, top,
1167 } while (resid && space > 0);
1171 ssb_unlock(&so->so_snd);
1182 * Implement receive operations on a socket.
1184 * We depend on the way that records are added to the signalsockbuf
1185 * by sbappend*. In particular, each record (mbufs linked through m_next)
1186 * must begin with an address if the protocol so specifies,
1187 * followed by an optional mbuf or mbufs containing ancillary data,
1188 * and then zero or more mbufs of data.
1190 * Although the signalsockbuf is locked, new data may still be appended.
1191 * A token inside the ssb_lock deals with MP issues and still allows
1192 * the network to access the socket if we block in a uio.
1194 * The caller may receive the data as a single mbuf chain by supplying
1195 * an mbuf **mp0 for use in returning the chain. The uio is then used
1196 * only for the count in uio_resid.
1199 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1200 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1203 struct mbuf *free_chain = NULL;
1204 int flags, len, error, offset;
1205 struct protosw *pr = so->so_proto;
1207 size_t resid, orig_resid;
1210 resid = uio->uio_resid;
1212 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1220 flags = *flagsp &~ MSG_EOR;
1223 if (flags & MSG_OOB) {
1224 m = m_get(MB_WAIT, MT_DATA);
1227 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1233 KKASSERT(resid >= (size_t)m->m_len);
1234 resid -= (size_t)m->m_len;
1235 } while (resid > 0 && m);
1238 uio->uio_resid = resid;
1239 error = uiomove(mtod(m, caddr_t),
1240 (int)szmin(resid, m->m_len),
1242 resid = uio->uio_resid;
1244 } while (uio->uio_resid && error == 0 && m);
1251 if ((so->so_state & SS_ISCONFIRMING) && resid)
1255 * The token interlocks against the protocol thread while
1256 * ssb_lock is a blocking lock against other userland entities.
1258 lwkt_gettoken(&so->so_rcv.ssb_token);
1260 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1264 m = so->so_rcv.ssb_mb;
1266 * If we have less data than requested, block awaiting more
1267 * (subject to any timeout) if:
1268 * 1. the current count is less than the low water mark, or
1269 * 2. MSG_WAITALL is set, and it is possible to do the entire
1270 * receive operation at once if we block (resid <= hiwat).
1271 * 3. MSG_DONTWAIT is not set
1272 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1273 * we have to do the receive in sections, and thus risk returning
1274 * a short count if a timeout or signal occurs after we start.
1276 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1277 (size_t)so->so_rcv.ssb_cc < resid) &&
1278 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1279 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1280 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1281 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1285 error = so->so_error;
1286 if ((flags & MSG_PEEK) == 0)
1290 if (so->so_state & SS_CANTRCVMORE) {
1296 for (; m; m = m->m_next) {
1297 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1298 m = so->so_rcv.ssb_mb;
1302 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1303 (pr->pr_flags & PR_CONNREQUIRED)) {
1309 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1310 error = EWOULDBLOCK;
1313 ssb_unlock(&so->so_rcv);
1314 error = ssb_wait(&so->so_rcv);
1320 if (uio && uio->uio_td && uio->uio_td->td_proc)
1321 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1324 * note: m should be == sb_mb here. Cache the next record while
1325 * cleaning up. Note that calling m_free*() will break out critical
1328 KKASSERT(m == so->so_rcv.ssb_mb);
1331 * Skip any address mbufs prepending the record.
1333 if (pr->pr_flags & PR_ADDR) {
1334 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1337 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1338 if (flags & MSG_PEEK)
1341 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1345 * Skip any control mbufs prepending the record.
1348 if (pr->pr_flags & PR_ADDR_OPT) {
1350 * For SCTP we may be getting a
1351 * whole message OR a partial delivery.
1353 if (m && m->m_type == MT_SONAME) {
1356 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1357 if (flags & MSG_PEEK)
1360 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1364 while (m && m->m_type == MT_CONTROL && error == 0) {
1365 if (flags & MSG_PEEK) {
1367 *controlp = m_copy(m, 0, m->m_len);
1368 m = m->m_next; /* XXX race */
1371 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1372 if (pr->pr_domain->dom_externalize &&
1373 mtod(m, struct cmsghdr *)->cmsg_type ==
1375 error = (*pr->pr_domain->dom_externalize)(m);
1379 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1382 if (controlp && *controlp) {
1384 controlp = &(*controlp)->m_next;
1393 if (type == MT_OOBDATA)
1398 * Copy to the UIO or mbuf return chain (*mp).
1402 while (m && resid > 0 && error == 0) {
1403 if (m->m_type == MT_OOBDATA) {
1404 if (type != MT_OOBDATA)
1406 } else if (type == MT_OOBDATA)
1409 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1411 soclrstate(so, SS_RCVATMARK);
1412 len = (resid > INT_MAX) ? INT_MAX : resid;
1413 if (so->so_oobmark && len > so->so_oobmark - offset)
1414 len = so->so_oobmark - offset;
1415 if (len > m->m_len - moff)
1416 len = m->m_len - moff;
1419 * Copy out to the UIO or pass the mbufs back to the SIO.
1420 * The SIO is dealt with when we eat the mbuf, but deal
1421 * with the resid here either way.
1424 uio->uio_resid = resid;
1425 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1426 resid = uio->uio_resid;
1430 resid -= (size_t)len;
1434 * Eat the entire mbuf or just a piece of it
1436 if (len == m->m_len - moff) {
1437 if (m->m_flags & M_EOR)
1440 if (m->m_flags & M_NOTIFICATION)
1441 flags |= MSG_NOTIFICATION;
1443 if (flags & MSG_PEEK) {
1448 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1452 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1456 if (flags & MSG_PEEK) {
1460 n = m_copym(m, 0, len, MB_WAIT);
1466 so->so_rcv.ssb_cc -= len;
1469 if (so->so_oobmark) {
1470 if ((flags & MSG_PEEK) == 0) {
1471 so->so_oobmark -= len;
1472 if (so->so_oobmark == 0) {
1473 sosetstate(so, SS_RCVATMARK);
1478 if (offset == so->so_oobmark)
1482 if (flags & MSG_EOR)
1485 * If the MSG_WAITALL flag is set (for non-atomic socket),
1486 * we must not quit until resid == 0 or an error
1487 * termination. If a signal/timeout occurs, return
1488 * with a short count but without error.
1489 * Keep signalsockbuf locked against other readers.
1491 while ((flags & MSG_WAITALL) && m == NULL &&
1492 resid > 0 && !sosendallatonce(so) &&
1493 so->so_rcv.ssb_mb == NULL) {
1494 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1497 * The window might have closed to zero, make
1498 * sure we send an ack now that we've drained
1499 * the buffer or we might end up blocking until
1500 * the idle takes over (5 seconds).
1502 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1503 so_pru_rcvd(so, flags);
1504 error = ssb_wait(&so->so_rcv);
1506 ssb_unlock(&so->so_rcv);
1510 m = so->so_rcv.ssb_mb;
1515 * If an atomic read was requested but unread data still remains
1516 * in the record, set MSG_TRUNC.
1518 if (m && pr->pr_flags & PR_ATOMIC)
1522 * Cleanup. If an atomic read was requested drop any unread data.
1524 if ((flags & MSG_PEEK) == 0) {
1525 if (m && (pr->pr_flags & PR_ATOMIC))
1526 sbdroprecord(&so->so_rcv.sb);
1527 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1528 so_pru_rcvd(so, flags);
1531 if (orig_resid == resid && orig_resid &&
1532 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1533 ssb_unlock(&so->so_rcv);
1540 ssb_unlock(&so->so_rcv);
1542 lwkt_reltoken(&so->so_rcv.ssb_token);
1544 m_freem(free_chain);
1549 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1550 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1553 struct mbuf *free_chain = NULL;
1554 int flags, len, error, offset;
1555 struct protosw *pr = so->so_proto;
1557 size_t resid, orig_resid;
1560 resid = uio->uio_resid;
1562 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1570 flags = *flagsp &~ MSG_EOR;
1573 if (flags & MSG_OOB) {
1574 m = m_get(MB_WAIT, MT_DATA);
1577 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1583 KKASSERT(resid >= (size_t)m->m_len);
1584 resid -= (size_t)m->m_len;
1585 } while (resid > 0 && m);
1588 uio->uio_resid = resid;
1589 error = uiomove(mtod(m, caddr_t),
1590 (int)szmin(resid, m->m_len),
1592 resid = uio->uio_resid;
1594 } while (uio->uio_resid && error == 0 && m);
1603 * The token interlocks against the protocol thread while
1604 * ssb_lock is a blocking lock against other userland entities.
1606 lwkt_gettoken(&so->so_rcv.ssb_token);
1608 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1612 m = so->so_rcv.ssb_mb;
1614 * If we have less data than requested, block awaiting more
1615 * (subject to any timeout) if:
1616 * 1. the current count is less than the low water mark, or
1617 * 2. MSG_WAITALL is set, and it is possible to do the entire
1618 * receive operation at once if we block (resid <= hiwat).
1619 * 3. MSG_DONTWAIT is not set
1620 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1621 * we have to do the receive in sections, and thus risk returning
1622 * a short count if a timeout or signal occurs after we start.
1624 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1625 (size_t)so->so_rcv.ssb_cc < resid) &&
1626 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1627 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1628 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1632 error = so->so_error;
1633 if ((flags & MSG_PEEK) == 0)
1637 if (so->so_state & SS_CANTRCVMORE) {
1643 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1644 (pr->pr_flags & PR_CONNREQUIRED)) {
1650 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1651 error = EWOULDBLOCK;
1654 ssb_unlock(&so->so_rcv);
1655 error = ssb_wait(&so->so_rcv);
1661 if (uio && uio->uio_td && uio->uio_td->td_proc)
1662 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1665 * note: m should be == sb_mb here. Cache the next record while
1666 * cleaning up. Note that calling m_free*() will break out critical
1669 KKASSERT(m == so->so_rcv.ssb_mb);
1672 * Copy to the UIO or mbuf return chain (*mp).
1676 while (m && resid > 0 && error == 0) {
1677 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1680 soclrstate(so, SS_RCVATMARK);
1681 len = (resid > INT_MAX) ? INT_MAX : resid;
1682 if (so->so_oobmark && len > so->so_oobmark - offset)
1683 len = so->so_oobmark - offset;
1684 if (len > m->m_len - moff)
1685 len = m->m_len - moff;
1688 * Copy out to the UIO or pass the mbufs back to the SIO.
1689 * The SIO is dealt with when we eat the mbuf, but deal
1690 * with the resid here either way.
1693 uio->uio_resid = resid;
1694 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1695 resid = uio->uio_resid;
1699 resid -= (size_t)len;
1703 * Eat the entire mbuf or just a piece of it
1705 if (len == m->m_len - moff) {
1706 if (flags & MSG_PEEK) {
1711 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1715 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1719 if (flags & MSG_PEEK) {
1723 n = m_copym(m, 0, len, MB_WAIT);
1729 so->so_rcv.ssb_cc -= len;
1732 if (so->so_oobmark) {
1733 if ((flags & MSG_PEEK) == 0) {
1734 so->so_oobmark -= len;
1735 if (so->so_oobmark == 0) {
1736 sosetstate(so, SS_RCVATMARK);
1741 if (offset == so->so_oobmark)
1746 * If the MSG_WAITALL flag is set (for non-atomic socket),
1747 * we must not quit until resid == 0 or an error
1748 * termination. If a signal/timeout occurs, return
1749 * with a short count but without error.
1750 * Keep signalsockbuf locked against other readers.
1752 while ((flags & MSG_WAITALL) && m == NULL &&
1753 resid > 0 && !sosendallatonce(so) &&
1754 so->so_rcv.ssb_mb == NULL) {
1755 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1758 * The window might have closed to zero, make
1759 * sure we send an ack now that we've drained
1760 * the buffer or we might end up blocking until
1761 * the idle takes over (5 seconds).
1764 so_pru_rcvd_async(so);
1765 error = ssb_wait(&so->so_rcv);
1767 ssb_unlock(&so->so_rcv);
1771 m = so->so_rcv.ssb_mb;
1776 * Cleanup. If an atomic read was requested drop any unread data.
1778 if ((flags & MSG_PEEK) == 0) {
1780 so_pru_rcvd_async(so);
1783 if (orig_resid == resid && orig_resid &&
1784 (so->so_state & SS_CANTRCVMORE) == 0) {
1785 ssb_unlock(&so->so_rcv);
1792 ssb_unlock(&so->so_rcv);
1794 lwkt_reltoken(&so->so_rcv.ssb_token);
1796 m_freem(free_chain);
1801 * Shut a socket down. Note that we do not get a frontend lock as we
1802 * want to be able to shut the socket down even if another thread is
1803 * blocked in a read(), thus waking it up.
1806 soshutdown(struct socket *so, int how)
1808 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1811 if (how != SHUT_WR) {
1812 /*ssb_lock(&so->so_rcv, M_WAITOK);*/
1814 /*ssb_unlock(&so->so_rcv);*/
1817 return (so_pru_shutdown(so));
1822 sorflush(struct socket *so)
1824 struct signalsockbuf *ssb = &so->so_rcv;
1825 struct protosw *pr = so->so_proto;
1826 struct signalsockbuf asb;
1828 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1830 lwkt_gettoken(&ssb->ssb_token);
1835 * Can't just blow up the ssb structure here
1837 bzero(&ssb->sb, sizeof(ssb->sb));
1842 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1844 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1845 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1846 ssb_release(&asb, so);
1848 lwkt_reltoken(&ssb->ssb_token);
1853 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1855 struct accept_filter_arg *afap = NULL;
1856 struct accept_filter *afp;
1857 struct so_accf *af = so->so_accf;
1860 /* do not set/remove accept filters on non listen sockets */
1861 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1866 /* removing the filter */
1869 if (af->so_accept_filter != NULL &&
1870 af->so_accept_filter->accf_destroy != NULL) {
1871 af->so_accept_filter->accf_destroy(so);
1873 if (af->so_accept_filter_str != NULL) {
1874 kfree(af->so_accept_filter_str, M_ACCF);
1879 so->so_options &= ~SO_ACCEPTFILTER;
1882 /* adding a filter */
1883 /* must remove previous filter first */
1888 /* don't put large objects on the kernel stack */
1889 afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
1890 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1891 afap->af_name[sizeof(afap->af_name)-1] = '\0';
1892 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1895 afp = accept_filt_get(afap->af_name);
1900 af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1901 if (afp->accf_create != NULL) {
1902 if (afap->af_name[0] != '\0') {
1903 int len = strlen(afap->af_name) + 1;
1905 af->so_accept_filter_str = kmalloc(len, M_ACCF,
1907 strcpy(af->so_accept_filter_str, afap->af_name);
1909 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1910 if (af->so_accept_filter_arg == NULL) {
1911 kfree(af->so_accept_filter_str, M_ACCF);
1918 af->so_accept_filter = afp;
1920 so->so_options |= SO_ACCEPTFILTER;
1923 kfree(afap, M_TEMP);
1929 * Perhaps this routine, and sooptcopyout(), below, ought to come in
1930 * an additional variant to handle the case where the option value needs
1931 * to be some kind of integer, but not a specific size.
1932 * In addition to their use here, these functions are also called by the
1933 * protocol-level pr_ctloutput() routines.
1936 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1938 return soopt_to_kbuf(sopt, buf, len, minlen);
1942 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1946 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1947 KKASSERT(kva_p(buf));
1950 * If the user gives us more than we wanted, we ignore it,
1951 * but if we don't get the minimum length the caller
1952 * wants, we return EINVAL. On success, sopt->sopt_valsize
1953 * is set to however much we actually retrieved.
1955 if ((valsize = sopt->sopt_valsize) < minlen)
1958 sopt->sopt_valsize = valsize = len;
1960 bcopy(sopt->sopt_val, buf, valsize);
1966 sosetopt(struct socket *so, struct sockopt *sopt)
1972 struct signalsockbuf *sotmp;
1975 sopt->sopt_dir = SOPT_SET;
1976 if (sopt->sopt_level != SOL_SOCKET) {
1977 if (so->so_proto && so->so_proto->pr_ctloutput) {
1978 return (so_pr_ctloutput(so, sopt));
1980 error = ENOPROTOOPT;
1982 switch (sopt->sopt_name) {
1984 case SO_ACCEPTFILTER:
1985 error = do_setopt_accept_filter(so, sopt);
1991 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1995 so->so_linger = l.l_linger;
1997 so->so_options |= SO_LINGER;
1999 so->so_options &= ~SO_LINGER;
2005 case SO_USELOOPBACK:
2012 error = sooptcopyin(sopt, &optval, sizeof optval,
2017 so->so_options |= sopt->sopt_name;
2019 so->so_options &= ~sopt->sopt_name;
2026 error = sooptcopyin(sopt, &optval, sizeof optval,
2032 * Values < 1 make no sense for any of these
2033 * options, so disallow them.
2040 switch (sopt->sopt_name) {
2043 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2044 &so->so_snd : &so->so_rcv, (u_long)optval,
2046 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2050 sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2051 &so->so_snd : &so->so_rcv;
2052 atomic_clear_int(&sotmp->ssb_flags,
2057 * Make sure the low-water is never greater than
2061 so->so_snd.ssb_lowat =
2062 (optval > so->so_snd.ssb_hiwat) ?
2063 so->so_snd.ssb_hiwat : optval;
2064 atomic_clear_int(&so->so_snd.ssb_flags,
2068 so->so_rcv.ssb_lowat =
2069 (optval > so->so_rcv.ssb_hiwat) ?
2070 so->so_rcv.ssb_hiwat : optval;
2071 atomic_clear_int(&so->so_rcv.ssb_flags,
2079 error = sooptcopyin(sopt, &tv, sizeof tv,
2084 /* assert(hz > 0); */
2085 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2086 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2090 /* assert(tick > 0); */
2091 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2092 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2093 if (val > INT_MAX) {
2097 if (val == 0 && tv.tv_usec != 0)
2100 switch (sopt->sopt_name) {
2102 so->so_snd.ssb_timeo = val;
2105 so->so_rcv.ssb_timeo = val;
2110 error = ENOPROTOOPT;
2113 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2114 (void) so_pr_ctloutput(so, sopt);
2121 /* Helper routine for getsockopt */
2123 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2125 soopt_from_kbuf(sopt, buf, len);
2130 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2135 sopt->sopt_valsize = 0;
2139 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2140 KKASSERT(kva_p(buf));
2143 * Documented get behavior is that we always return a value,
2144 * possibly truncated to fit in the user's buffer.
2145 * Traditional behavior is that we always tell the user
2146 * precisely how much we copied, rather than something useful
2147 * like the total amount we had available for her.
2148 * Note that this interface is not idempotent; the entire answer must
2149 * generated ahead of time.
2151 valsize = szmin(len, sopt->sopt_valsize);
2152 sopt->sopt_valsize = valsize;
2153 if (sopt->sopt_val != 0) {
2154 bcopy(buf, sopt->sopt_val, valsize);
2159 sogetopt(struct socket *so, struct sockopt *sopt)
2166 struct accept_filter_arg *afap;
2170 sopt->sopt_dir = SOPT_GET;
2171 if (sopt->sopt_level != SOL_SOCKET) {
2172 if (so->so_proto && so->so_proto->pr_ctloutput) {
2173 return (so_pr_ctloutput(so, sopt));
2175 return (ENOPROTOOPT);
2177 switch (sopt->sopt_name) {
2179 case SO_ACCEPTFILTER:
2180 if ((so->so_options & SO_ACCEPTCONN) == 0)
2182 afap = kmalloc(sizeof(*afap), M_TEMP,
2184 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2185 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2186 if (so->so_accf->so_accept_filter_str != NULL)
2187 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2189 error = sooptcopyout(sopt, afap, sizeof(*afap));
2190 kfree(afap, M_TEMP);
2195 l.l_onoff = so->so_options & SO_LINGER;
2196 l.l_linger = so->so_linger;
2197 error = sooptcopyout(sopt, &l, sizeof l);
2200 case SO_USELOOPBACK:
2210 optval = so->so_options & sopt->sopt_name;
2212 error = sooptcopyout(sopt, &optval, sizeof optval);
2216 optval = so->so_type;
2220 optval = so->so_error;
2225 optval = so->so_snd.ssb_hiwat;
2229 optval = so->so_rcv.ssb_hiwat;
2233 optval = so->so_snd.ssb_lowat;
2237 optval = so->so_rcv.ssb_lowat;
2242 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2243 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2245 tv.tv_sec = optval / hz;
2246 tv.tv_usec = (optval % hz) * ustick;
2247 error = sooptcopyout(sopt, &tv, sizeof tv);
2251 optval_l = ssb_space(&so->so_snd);
2252 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2256 error = ENOPROTOOPT;
2263 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2265 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2267 struct mbuf *m, *m_prev;
2268 int sopt_size = sopt->sopt_valsize, msize;
2270 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
2274 m->m_len = min(msize, sopt_size);
2275 sopt_size -= m->m_len;
2279 while (sopt_size > 0) {
2280 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
2281 MT_DATA, 0, &msize);
2286 m->m_len = min(msize, sopt_size);
2287 sopt_size -= m->m_len;
2294 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2296 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2298 soopt_to_mbuf(sopt, m);
2303 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2308 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2310 if (sopt->sopt_val == NULL)
2312 val = sopt->sopt_val;
2313 valsize = sopt->sopt_valsize;
2314 while (m != NULL && valsize >= m->m_len) {
2315 bcopy(val, mtod(m, char *), m->m_len);
2316 valsize -= m->m_len;
2317 val = (caddr_t)val + m->m_len;
2320 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2321 panic("ip6_sooptmcopyin");
2324 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2326 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2328 return soopt_from_mbuf(sopt, m);
2332 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2334 struct mbuf *m0 = m;
2339 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2341 if (sopt->sopt_val == NULL)
2343 val = sopt->sopt_val;
2344 maxsize = sopt->sopt_valsize;
2345 while (m != NULL && maxsize >= m->m_len) {
2346 bcopy(mtod(m, char *), val, m->m_len);
2347 maxsize -= m->m_len;
2348 val = (caddr_t)val + m->m_len;
2349 valsize += m->m_len;
2353 /* enough soopt buffer should be given from user-land */
2357 sopt->sopt_valsize = valsize;
2362 sohasoutofband(struct socket *so)
2364 if (so->so_sigio != NULL)
2365 pgsigio(so->so_sigio, SIGURG, 0);
2366 KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
2370 sokqfilter(struct file *fp, struct knote *kn)
2372 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2373 struct signalsockbuf *ssb;
2375 switch (kn->kn_filter) {
2377 if (so->so_options & SO_ACCEPTCONN)
2378 kn->kn_fop = &solisten_filtops;
2380 kn->kn_fop = &soread_filtops;
2384 kn->kn_fop = &sowrite_filtops;
2388 kn->kn_fop = &soexcept_filtops;
2392 return (EOPNOTSUPP);
2395 knote_insert(&ssb->ssb_kq.ki_note, kn);
2396 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2401 filt_sordetach(struct knote *kn)
2403 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2405 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2406 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2407 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2412 filt_soread(struct knote *kn, long hint)
2414 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2416 if (kn->kn_sfflags & NOTE_OOB) {
2417 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2418 kn->kn_fflags |= NOTE_OOB;
2423 kn->kn_data = so->so_rcv.ssb_cc;
2425 if (so->so_state & SS_CANTRCVMORE) {
2427 * Only set NODATA if all data has been exhausted.
2429 if (kn->kn_data == 0)
2430 kn->kn_flags |= EV_NODATA;
2431 kn->kn_flags |= EV_EOF;
2432 kn->kn_fflags = so->so_error;
2435 if (so->so_error) /* temporary udp error */
2437 if (kn->kn_sfflags & NOTE_LOWAT)
2438 return (kn->kn_data >= kn->kn_sdata);
2439 return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2440 !TAILQ_EMPTY(&so->so_comp));
2444 filt_sowdetach(struct knote *kn)
2446 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2448 knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2449 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2450 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2455 filt_sowrite(struct knote *kn, long hint)
2457 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2459 kn->kn_data = ssb_space(&so->so_snd);
2460 if (so->so_state & SS_CANTSENDMORE) {
2461 kn->kn_flags |= (EV_EOF | EV_NODATA);
2462 kn->kn_fflags = so->so_error;
2465 if (so->so_error) /* temporary udp error */
2467 if (((so->so_state & SS_ISCONNECTED) == 0) &&
2468 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2470 if (kn->kn_sfflags & NOTE_LOWAT)
2471 return (kn->kn_data >= kn->kn_sdata);
2472 return (kn->kn_data >= so->so_snd.ssb_lowat);
2477 filt_solisten(struct knote *kn, long hint)
2479 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2481 kn->kn_data = so->so_qlen;
2482 return (! TAILQ_EMPTY(&so->so_comp));