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. Neither the name of the University nor the names of its contributors
47 * may be used to endorse or promote products derived from this software
48 * without specific prior written permission.
50 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
63 * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.24 2003/11/11 17:18:18 silby Exp $
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/fcntl.h>
72 #include <sys/malloc.h>
74 #include <sys/domain.h>
75 #include <sys/file.h> /* for struct knote */
76 #include <sys/kernel.h>
77 #include <sys/event.h>
79 #include <sys/protosw.h>
80 #include <sys/socket.h>
81 #include <sys/socketvar.h>
82 #include <sys/socketops.h>
83 #include <sys/resourcevar.h>
84 #include <sys/signalvar.h>
85 #include <sys/sysctl.h>
88 #include <vm/vm_zone.h>
90 #include <net/netmsg2.h>
91 #include <net/netisr2.h>
93 #include <sys/thread2.h>
94 #include <sys/socketvar2.h>
95 #include <sys/spinlock2.h>
97 #include <machine/limits.h>
100 extern int tcp_sosend_agglim;
101 extern int tcp_sosend_async;
102 extern int udp_sosend_async;
103 extern int udp_sosend_prepend;
105 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
108 static void filt_sordetach(struct knote *kn);
109 static int filt_soread(struct knote *kn, long hint);
110 static void filt_sowdetach(struct knote *kn);
111 static int filt_sowrite(struct knote *kn, long hint);
112 static int filt_solisten(struct knote *kn, long hint);
114 static void sodiscard(struct socket *so);
115 static int soclose_sync(struct socket *so, int fflag);
116 static void soclose_fast(struct socket *so);
118 static struct filterops solisten_filtops =
119 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten };
120 static struct filterops soread_filtops =
121 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
122 static struct filterops sowrite_filtops =
123 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite };
124 static struct filterops soexcept_filtops =
125 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
127 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
128 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
129 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
132 static int somaxconn = SOMAXCONN;
133 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
134 &somaxconn, 0, "Maximum pending socket connection queue size");
136 static int use_soclose_fast = 1;
137 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW,
138 &use_soclose_fast, 0, "Fast socket close");
140 int use_soaccept_pred_fast = 1;
141 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW,
142 &use_soaccept_pred_fast, 0, "Fast socket accept predication");
144 int use_sendfile_async = 1;
145 SYSCTL_INT(_kern_ipc, OID_AUTO, sendfile_async, CTLFLAG_RW,
146 &use_sendfile_async, 0, "sendfile uses asynchronized pru_send");
148 int use_soconnect_async = 1;
149 SYSCTL_INT(_kern_ipc, OID_AUTO, soconnect_async, CTLFLAG_RW,
150 &use_soconnect_async, 0, "soconnect uses asynchronized pru_connect");
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 | MSGF_PRIORITY,
185 so->so_proto->pr_usrreqs->pru_rcvd);
186 so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC;
187 so->so_state = SS_NOFDREF;
194 socreate(int dom, struct socket **aso, int type,
195 int proto, struct thread *td)
197 struct proc *p = td->td_proc;
200 struct pru_attach_info ai;
204 prp = pffindproto(dom, proto, type);
206 prp = pffindtype(dom, type);
208 if (prp == NULL || prp->pr_usrreqs->pru_attach == 0)
209 return (EPROTONOSUPPORT);
211 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
212 prp->pr_domain->dom_family != PF_LOCAL &&
213 prp->pr_domain->dom_family != PF_INET &&
214 prp->pr_domain->dom_family != PF_INET6 &&
215 prp->pr_domain->dom_family != PF_ROUTE) {
216 return (EPROTONOSUPPORT);
219 if (prp->pr_type != type)
221 so = soalloc(p != NULL, prp);
226 * Callers of socreate() presumably will connect up a descriptor
227 * and call soclose() if they cannot. This represents our so_refs
228 * (which should be 1) from soalloc().
230 soclrstate(so, SS_NOFDREF);
233 * Set a default port for protocol processing. No action will occur
234 * on the socket on this port until an inpcb is attached to it and
235 * is able to match incoming packets, or until the socket becomes
236 * available to userland.
238 * We normally default the socket to the protocol thread on cpu 0.
239 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
240 * thread and all pr_*()/pru_*() calls are executed synchronously.
242 if (prp->pr_flags & PR_SYNC_PORT)
243 so->so_port = &netisr_sync_port;
245 so->so_port = netisr_cpuport(0);
247 TAILQ_INIT(&so->so_incomp);
248 TAILQ_INIT(&so->so_comp);
250 so->so_cred = crhold(p->p_ucred);
251 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
252 ai.p_ucred = p->p_ucred;
253 ai.fd_rdir = p->p_fd->fd_rdir;
256 * Auto-sizing of socket buffers is managed by the protocols and
257 * the appropriate flags must be set in the pru_attach function.
259 error = so_pru_attach(so, proto, &ai);
261 sosetstate(so, SS_NOFDREF);
262 sofree(so); /* from soalloc */
267 * NOTE: Returns referenced socket.
274 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
278 error = so_pru_bind(so, nam, td);
283 sodealloc(struct socket *so)
285 if (so->so_rcv.ssb_hiwat)
286 (void)chgsbsize(so->so_cred->cr_uidinfo,
287 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
288 if (so->so_snd.ssb_hiwat)
289 (void)chgsbsize(so->so_cred->cr_uidinfo,
290 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
292 /* remove accept filter if present */
293 if (so->so_accf != NULL)
294 do_setopt_accept_filter(so, NULL);
297 if (so->so_faddr != NULL)
298 kfree(so->so_faddr, M_SONAME);
303 solisten(struct socket *so, int backlog, struct thread *td)
307 short oldopt, oldqlimit;
310 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
314 oldopt = so->so_options;
315 oldqlimit = so->so_qlimit;
318 lwkt_gettoken(&so->so_rcv.ssb_token);
319 if (TAILQ_EMPTY(&so->so_comp))
320 so->so_options |= SO_ACCEPTCONN;
321 lwkt_reltoken(&so->so_rcv.ssb_token);
322 if (backlog < 0 || backlog > somaxconn)
324 so->so_qlimit = backlog;
325 /* SCTP needs to look at tweak both the inbound backlog parameter AND
326 * the so_options (UDP model both connect's and gets inbound
327 * connections .. implicitly).
329 error = so_pru_listen(so, td);
332 /* Restore the params */
333 so->so_options = oldopt;
334 so->so_qlimit = oldqlimit;
342 * Destroy a disconnected socket. This routine is a NOP if entities
343 * still have a reference on the socket:
345 * so_pcb - The protocol stack still has a reference
346 * SS_NOFDREF - There is no longer a file pointer reference
349 sofree(struct socket *so)
354 * This is a bit hackish at the moment. We need to interlock
355 * any accept queue we are on before we potentially lose the
356 * last reference to avoid races against a re-reference from
357 * someone operating on the queue.
359 while ((head = so->so_head) != NULL) {
360 lwkt_getpooltoken(head);
361 if (so->so_head == head)
363 lwkt_relpooltoken(head);
367 * Arbitrage the last free.
369 KKASSERT(so->so_refs > 0);
370 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
372 lwkt_relpooltoken(head);
376 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
377 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
380 * We're done, remove ourselves from the accept queue we are
381 * on, if we are on one.
384 if (so->so_state & SS_INCOMP) {
385 TAILQ_REMOVE(&head->so_incomp, so, so_list);
387 } else if (so->so_state & SS_COMP) {
389 * We must not decommission a socket that's
390 * on the accept(2) queue. If we do, then
391 * accept(2) may hang after select(2) indicated
392 * that the listening socket was ready.
394 lwkt_relpooltoken(head);
397 panic("sofree: not queued");
399 soclrstate(so, SS_INCOMP);
401 lwkt_relpooltoken(head);
403 ssb_release(&so->so_snd, so);
409 * Close a socket on last file table reference removal.
410 * Initiate disconnect if connected.
411 * Free socket when disconnect complete.
414 soclose(struct socket *so, int fflag)
418 funsetown(&so->so_sigio);
419 if (!use_soclose_fast ||
420 (so->so_proto->pr_flags & PR_SYNC_PORT) ||
421 ((so->so_state & SS_ISCONNECTED) &&
422 (so->so_options & SO_LINGER))) {
423 error = soclose_sync(so, fflag);
432 sodiscard(struct socket *so)
434 lwkt_getpooltoken(so);
435 if (so->so_options & SO_ACCEPTCONN) {
438 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
439 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
440 soclrstate(sp, SS_INCOMP);
445 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
446 TAILQ_REMOVE(&so->so_comp, sp, so_list);
447 soclrstate(sp, SS_COMP);
453 lwkt_relpooltoken(so);
455 if (so->so_state & SS_NOFDREF)
456 panic("soclose: NOFDREF");
457 sosetstate(so, SS_NOFDREF); /* take ref */
461 soinherit(struct socket *so, struct socket *so_inh)
463 TAILQ_HEAD(, socket) comp, incomp;
467 KASSERT(so->so_options & SO_ACCEPTCONN,
468 ("so does not accept connection"));
469 KASSERT(so_inh->so_options & SO_ACCEPTCONN,
470 ("so_inh does not accept connection"));
475 lwkt_getpooltoken(so);
476 lwkt_getpooltoken(so_inh);
479 * Save completed queue and incompleted queue
481 TAILQ_CONCAT(&comp, &so->so_comp, so_list);
485 TAILQ_CONCAT(&incomp, &so->so_incomp, so_list);
486 incqlen = so->so_incqlen;
490 * Append the saved completed queue and incompleted
491 * queue to the socket inherits them.
494 * This may temporarily break the inheriting socket's
497 TAILQ_FOREACH(sp, &comp, so_list) {
498 sp->so_head = so_inh;
500 sp->so_cred = crhold(so_inh->so_cred);
503 TAILQ_FOREACH(sp, &incomp, so_list) {
504 sp->so_head = so_inh;
506 sp->so_cred = crhold(so_inh->so_cred);
509 TAILQ_CONCAT(&so_inh->so_comp, &comp, so_list);
510 so_inh->so_qlen += qlen;
512 TAILQ_CONCAT(&so_inh->so_incomp, &incomp, so_list);
513 so_inh->so_incqlen += incqlen;
515 lwkt_relpooltoken(so_inh);
516 lwkt_relpooltoken(so);
520 * "New" connections have arrived
523 wakeup(&so_inh->so_timeo);
528 soclose_sync(struct socket *so, int fflag)
532 if (so->so_pcb == NULL)
534 if (so->so_state & SS_ISCONNECTED) {
535 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
536 error = sodisconnect(so);
540 if (so->so_options & SO_LINGER) {
541 if ((so->so_state & SS_ISDISCONNECTING) &&
544 while (so->so_state & SS_ISCONNECTED) {
545 error = tsleep(&so->so_timeo, PCATCH,
546 "soclos", so->so_linger * hz);
556 error2 = so_pru_detach(so);
562 so_pru_sync(so); /* unpend async sending */
563 sofree(so); /* dispose of ref */
569 soclose_sofree_async_handler(netmsg_t msg)
571 sofree(msg->base.nm_so);
575 soclose_sofree_async(struct socket *so)
577 struct netmsg_base *base = &so->so_clomsg;
579 netmsg_init(base, so, &netisr_apanic_rport, 0,
580 soclose_sofree_async_handler);
581 lwkt_sendmsg(so->so_port, &base->lmsg);
585 soclose_disconn_async_handler(netmsg_t msg)
587 struct socket *so = msg->base.nm_so;
589 if ((so->so_state & SS_ISCONNECTED) &&
590 (so->so_state & SS_ISDISCONNECTING) == 0)
591 so_pru_disconnect_direct(so);
594 so_pru_detach_direct(so);
601 soclose_disconn_async(struct socket *so)
603 struct netmsg_base *base = &so->so_clomsg;
605 netmsg_init(base, so, &netisr_apanic_rport, 0,
606 soclose_disconn_async_handler);
607 lwkt_sendmsg(so->so_port, &base->lmsg);
611 soclose_detach_async_handler(netmsg_t msg)
613 struct socket *so = msg->base.nm_so;
616 so_pru_detach_direct(so);
623 soclose_detach_async(struct socket *so)
625 struct netmsg_base *base = &so->so_clomsg;
627 netmsg_init(base, so, &netisr_apanic_rport, 0,
628 soclose_detach_async_handler);
629 lwkt_sendmsg(so->so_port, &base->lmsg);
633 soclose_fast(struct socket *so)
635 if (so->so_pcb == NULL)
638 if ((so->so_state & SS_ISCONNECTED) &&
639 (so->so_state & SS_ISDISCONNECTING) == 0) {
640 soclose_disconn_async(so);
645 soclose_detach_async(so);
651 soclose_sofree_async(so);
655 * Abort and destroy a socket. Only one abort can be in progress
656 * at any given moment.
659 soabort(struct socket *so)
666 soaborta(struct socket *so)
673 soabort_oncpu(struct socket *so)
676 so_pru_abort_oncpu(so);
680 * so is passed in ref'd, which becomes owned by
681 * the cleared SS_NOFDREF flag.
684 soaccept_generic(struct socket *so)
686 if ((so->so_state & SS_NOFDREF) == 0)
687 panic("soaccept: !NOFDREF");
688 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */
692 soaccept(struct socket *so, struct sockaddr **nam)
696 soaccept_generic(so);
697 error = so_pru_accept(so, nam);
702 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td,
707 if (so->so_options & SO_ACCEPTCONN)
710 * If protocol is connection-based, can only connect once.
711 * Otherwise, if connected, try to disconnect first.
712 * This allows user to disconnect by connecting to, e.g.,
715 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
716 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
717 (error = sodisconnect(so)))) {
721 * Prevent accumulated error from previous connection
725 if (!sync && so->so_proto->pr_usrreqs->pru_preconnect)
726 error = so_pru_connect_async(so, nam, td);
728 error = so_pru_connect(so, nam, td);
734 soconnect2(struct socket *so1, struct socket *so2)
738 error = so_pru_connect2(so1, so2);
743 sodisconnect(struct socket *so)
747 if ((so->so_state & SS_ISCONNECTED) == 0) {
751 if (so->so_state & SS_ISDISCONNECTING) {
755 error = so_pru_disconnect(so);
760 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
763 * If send must go all at once and message is larger than
764 * send buffering, then hard error.
765 * Lock against other senders.
766 * If must go all at once and not enough room now, then
767 * inform user that this would block and do nothing.
768 * Otherwise, if nonblocking, send as much as possible.
769 * The data to be sent is described by "uio" if nonzero,
770 * otherwise by the mbuf chain "top" (which must be null
771 * if uio is not). Data provided in mbuf chain must be small
772 * enough to send all at once.
774 * Returns nonzero on error, timeout or signal; callers
775 * must check for short counts if EINTR/ERESTART are returned.
776 * Data and control buffers are freed on return.
779 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
780 struct mbuf *top, struct mbuf *control, int flags,
787 int clen = 0, error, dontroute, mlen;
788 int atomic = sosendallatonce(so) || top;
792 resid = uio->uio_resid;
794 resid = (size_t)top->m_pkthdr.len;
797 for (m = top; m; m = m->m_next)
799 KKASSERT(top->m_pkthdr.len == len);
804 * WARNING! resid is unsigned, space and len are signed. space
805 * can wind up negative if the sockbuf is overcommitted.
807 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
808 * type sockets since that's an error.
810 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
816 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
817 (so->so_proto->pr_flags & PR_ATOMIC);
818 if (td->td_lwp != NULL)
819 td->td_lwp->lwp_ru.ru_msgsnd++;
821 clen = control->m_len;
822 #define gotoerr(errcode) { error = errcode; goto release; }
825 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
830 if (so->so_state & SS_CANTSENDMORE)
833 error = so->so_error;
837 if ((so->so_state & SS_ISCONNECTED) == 0) {
839 * `sendto' and `sendmsg' is allowed on a connection-
840 * based socket if it supports implied connect.
841 * Return ENOTCONN if not connected and no address is
844 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
845 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
846 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
847 !(resid == 0 && clen != 0))
849 } else if (addr == NULL)
850 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
851 ENOTCONN : EDESTADDRREQ);
853 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
854 clen > so->so_snd.ssb_hiwat) {
857 space = ssb_space(&so->so_snd);
860 if ((space < 0 || (size_t)space < resid + clen) && uio &&
861 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
862 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
863 gotoerr(EWOULDBLOCK);
864 ssb_unlock(&so->so_snd);
865 error = ssb_wait(&so->so_snd);
875 * Data is prepackaged in "top".
879 top->m_flags |= M_EOR;
883 m = m_getl((int)resid, MB_WAIT, MT_DATA,
884 top == NULL ? M_PKTHDR : 0, &mlen);
887 m->m_pkthdr.rcvif = NULL;
889 len = imin((int)szmin(mlen, resid), space);
890 if (resid < MINCLSIZE) {
892 * For datagram protocols, leave room
893 * for protocol headers in first mbuf.
895 if (atomic && top == NULL && len < mlen)
899 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
900 resid = uio->uio_resid;
903 top->m_pkthdr.len += len;
909 top->m_flags |= M_EOR;
912 } while (space > 0 && atomic);
914 so->so_options |= SO_DONTROUTE;
915 if (flags & MSG_OOB) {
916 pru_flags = PRUS_OOB;
917 } else if ((flags & MSG_EOF) &&
918 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
921 * If the user set MSG_EOF, the protocol
922 * understands this flag and nothing left to
923 * send then use PRU_SEND_EOF instead of PRU_SEND.
925 pru_flags = PRUS_EOF;
926 } else if (resid > 0 && space > 0) {
927 /* If there is more to send, set PRUS_MORETOCOME */
928 pru_flags = PRUS_MORETOCOME;
933 * XXX all the SS_CANTSENDMORE checks previously
934 * done could be out of date. We could have recieved
935 * a reset packet in an interrupt or maybe we slept
936 * while doing page faults in uiomove() etc. We could
937 * probably recheck again inside the splnet() protection
938 * here, but there are probably other places that this
939 * also happens. We must rethink this.
941 error = so_pru_send(so, pru_flags, top, addr, control, td);
943 so->so_options &= ~SO_DONTROUTE;
950 } while (resid && space > 0);
954 ssb_unlock(&so->so_snd);
965 * A specialization of sosend() for UDP based on protocol-specific knowledge:
966 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
967 * sosendallatonce() returns true,
968 * the "atomic" variable is true,
969 * and sosendudp() blocks until space is available for the entire send.
970 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
971 * PR_IMPLOPCL flags set.
972 * UDP has no out-of-band data.
973 * UDP has no control data.
974 * UDP does not support MSG_EOR.
977 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
978 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
981 int error, pru_flags = 0;
984 if (td->td_lwp != NULL)
985 td->td_lwp->lwp_ru.ru_msgsnd++;
989 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
990 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
993 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
997 if (so->so_state & SS_CANTSENDMORE)
1000 error = so->so_error;
1004 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
1005 gotoerr(EDESTADDRREQ);
1006 if (resid > so->so_snd.ssb_hiwat)
1008 space = ssb_space(&so->so_snd);
1009 if (uio && (space < 0 || (size_t)space < resid)) {
1010 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1011 gotoerr(EWOULDBLOCK);
1012 ssb_unlock(&so->so_snd);
1013 error = ssb_wait(&so->so_snd);
1020 int hdrlen = max_hdr;
1023 * We try to optimize out the additional mbuf
1024 * allocations in M_PREPEND() on output path, e.g.
1025 * - udp_output(), when it tries to prepend protocol
1027 * - Link layer output function, when it tries to
1028 * prepend link layer header.
1030 * This probably will not benefit any data that will
1031 * be fragmented, so this optimization is only performed
1032 * when the size of data and max size of protocol+link
1033 * headers fit into one mbuf cluster.
1035 if (uio->uio_resid > MCLBYTES - hdrlen ||
1036 !udp_sosend_prepend) {
1037 top = m_uiomove(uio);
1043 top = m_getl(uio->uio_resid + hdrlen, MB_WAIT,
1044 MT_DATA, M_PKTHDR, &nsize);
1045 KASSERT(nsize >= uio->uio_resid + hdrlen,
1046 ("sosendudp invalid nsize %d, "
1047 "resid %zu, hdrlen %d",
1048 nsize, uio->uio_resid, hdrlen));
1050 top->m_len = uio->uio_resid;
1051 top->m_pkthdr.len = uio->uio_resid;
1052 top->m_data += hdrlen;
1054 error = uiomove(mtod(top, caddr_t), top->m_len, uio);
1060 if (flags & MSG_DONTROUTE)
1061 pru_flags |= PRUS_DONTROUTE;
1063 if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
1064 so_pru_send_async(so, pru_flags, top, addr, NULL, td);
1067 error = so_pru_send(so, pru_flags, top, addr, NULL, td);
1069 top = NULL; /* sent or freed in lower layer */
1072 ssb_unlock(&so->so_snd);
1080 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1081 struct mbuf *top, struct mbuf *control, int flags,
1093 KKASSERT(top == NULL);
1095 resid = uio->uio_resid;
1098 resid = (size_t)top->m_pkthdr.len;
1101 for (m = top; m; m = m->m_next)
1103 KKASSERT(top->m_pkthdr.len == len);
1108 * WARNING! resid is unsigned, space and len are signed. space
1109 * can wind up negative if the sockbuf is overcommitted.
1111 * Also check to make sure that MSG_EOR isn't used on TCP
1113 if (flags & MSG_EOR) {
1119 /* TCP doesn't do control messages (rights, creds, etc) */
1120 if (control->m_len) {
1124 m_freem(control); /* empty control, just free it */
1128 if (td->td_lwp != NULL)
1129 td->td_lwp->lwp_ru.ru_msgsnd++;
1131 #define gotoerr(errcode) { error = errcode; goto release; }
1134 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1139 if (so->so_state & SS_CANTSENDMORE)
1142 error = so->so_error;
1146 if ((so->so_state & SS_ISCONNECTED) == 0 &&
1147 (so->so_state & SS_ISCONFIRMING) == 0)
1149 if (allatonce && resid > so->so_snd.ssb_hiwat)
1152 space = ssb_space_prealloc(&so->so_snd);
1153 if (flags & MSG_OOB)
1155 if ((space < 0 || (size_t)space < resid) && !allatonce &&
1156 space < so->so_snd.ssb_lowat) {
1157 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1158 gotoerr(EWOULDBLOCK);
1159 ssb_unlock(&so->so_snd);
1160 error = ssb_wait(&so->so_snd);
1167 int cnt = 0, async = 0;
1171 * Data is prepackaged in "top".
1175 if (resid > INT_MAX)
1177 m = m_getl((int)resid, MB_WAIT, MT_DATA,
1178 top == NULL ? M_PKTHDR : 0, &mlen);
1180 m->m_pkthdr.len = 0;
1181 m->m_pkthdr.rcvif = NULL;
1183 len = imin((int)szmin(mlen, resid), space);
1185 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1186 resid = uio->uio_resid;
1189 top->m_pkthdr.len += len;
1196 } while (space > 0 && cnt < tcp_sosend_agglim);
1198 if (tcp_sosend_async)
1201 if (flags & MSG_OOB) {
1202 pru_flags = PRUS_OOB;
1204 } else if ((flags & MSG_EOF) && resid == 0) {
1205 pru_flags = PRUS_EOF;
1206 } else if (resid > 0 && space > 0) {
1207 /* If there is more to send, set PRUS_MORETOCOME */
1208 pru_flags = PRUS_MORETOCOME;
1214 if (flags & MSG_SYNC)
1218 * XXX all the SS_CANTSENDMORE checks previously
1219 * done could be out of date. We could have recieved
1220 * a reset packet in an interrupt or maybe we slept
1221 * while doing page faults in uiomove() etc. We could
1222 * probably recheck again inside the splnet() protection
1223 * here, but there are probably other places that this
1224 * also happens. We must rethink this.
1226 for (m = top; m; m = m->m_next)
1227 ssb_preallocstream(&so->so_snd, m);
1229 error = so_pru_send(so, pru_flags, top,
1232 so_pru_send_async(so, pru_flags, top,
1241 } while (resid && space > 0);
1245 ssb_unlock(&so->so_snd);
1256 * Implement receive operations on a socket.
1258 * We depend on the way that records are added to the signalsockbuf
1259 * by sbappend*. In particular, each record (mbufs linked through m_next)
1260 * must begin with an address if the protocol so specifies,
1261 * followed by an optional mbuf or mbufs containing ancillary data,
1262 * and then zero or more mbufs of data.
1264 * Although the signalsockbuf is locked, new data may still be appended.
1265 * A token inside the ssb_lock deals with MP issues and still allows
1266 * the network to access the socket if we block in a uio.
1268 * The caller may receive the data as a single mbuf chain by supplying
1269 * an mbuf **mp0 for use in returning the chain. The uio is then used
1270 * only for the count in uio_resid.
1273 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1274 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1277 struct mbuf *free_chain = NULL;
1278 int flags, len, error, offset;
1279 struct protosw *pr = so->so_proto;
1281 size_t resid, orig_resid;
1284 resid = uio->uio_resid;
1286 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1294 flags = *flagsp &~ MSG_EOR;
1297 if (flags & MSG_OOB) {
1298 m = m_get(MB_WAIT, MT_DATA);
1301 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1307 KKASSERT(resid >= (size_t)m->m_len);
1308 resid -= (size_t)m->m_len;
1309 } while (resid > 0 && m);
1312 uio->uio_resid = resid;
1313 error = uiomove(mtod(m, caddr_t),
1314 (int)szmin(resid, m->m_len),
1316 resid = uio->uio_resid;
1318 } while (uio->uio_resid && error == 0 && m);
1325 if ((so->so_state & SS_ISCONFIRMING) && resid)
1329 * The token interlocks against the protocol thread while
1330 * ssb_lock is a blocking lock against other userland entities.
1332 lwkt_gettoken(&so->so_rcv.ssb_token);
1334 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1338 m = so->so_rcv.ssb_mb;
1340 * If we have less data than requested, block awaiting more
1341 * (subject to any timeout) if:
1342 * 1. the current count is less than the low water mark, or
1343 * 2. MSG_WAITALL is set, and it is possible to do the entire
1344 * receive operation at once if we block (resid <= hiwat).
1345 * 3. MSG_DONTWAIT is not set
1346 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1347 * we have to do the receive in sections, and thus risk returning
1348 * a short count if a timeout or signal occurs after we start.
1350 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1351 (size_t)so->so_rcv.ssb_cc < resid) &&
1352 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1353 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1354 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1355 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1359 error = so->so_error;
1360 if ((flags & MSG_PEEK) == 0)
1364 if (so->so_state & SS_CANTRCVMORE) {
1370 for (; m; m = m->m_next) {
1371 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1372 m = so->so_rcv.ssb_mb;
1376 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1377 (pr->pr_flags & PR_CONNREQUIRED)) {
1383 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1384 error = EWOULDBLOCK;
1387 ssb_unlock(&so->so_rcv);
1388 error = ssb_wait(&so->so_rcv);
1394 if (uio && uio->uio_td && uio->uio_td->td_proc)
1395 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1398 * note: m should be == sb_mb here. Cache the next record while
1399 * cleaning up. Note that calling m_free*() will break out critical
1402 KKASSERT(m == so->so_rcv.ssb_mb);
1405 * Skip any address mbufs prepending the record.
1407 if (pr->pr_flags & PR_ADDR) {
1408 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1411 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1412 if (flags & MSG_PEEK)
1415 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1419 * Skip any control mbufs prepending the record.
1422 if (pr->pr_flags & PR_ADDR_OPT) {
1424 * For SCTP we may be getting a
1425 * whole message OR a partial delivery.
1427 if (m && m->m_type == MT_SONAME) {
1430 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1431 if (flags & MSG_PEEK)
1434 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1438 while (m && m->m_type == MT_CONTROL && error == 0) {
1439 if (flags & MSG_PEEK) {
1441 *controlp = m_copy(m, 0, m->m_len);
1442 m = m->m_next; /* XXX race */
1445 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1446 if (pr->pr_domain->dom_externalize &&
1447 mtod(m, struct cmsghdr *)->cmsg_type ==
1449 error = (*pr->pr_domain->dom_externalize)(m);
1453 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1456 if (controlp && *controlp) {
1458 controlp = &(*controlp)->m_next;
1467 if (type == MT_OOBDATA)
1472 * Copy to the UIO or mbuf return chain (*mp).
1476 while (m && resid > 0 && error == 0) {
1477 if (m->m_type == MT_OOBDATA) {
1478 if (type != MT_OOBDATA)
1480 } else if (type == MT_OOBDATA)
1483 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1485 soclrstate(so, SS_RCVATMARK);
1486 len = (resid > INT_MAX) ? INT_MAX : resid;
1487 if (so->so_oobmark && len > so->so_oobmark - offset)
1488 len = so->so_oobmark - offset;
1489 if (len > m->m_len - moff)
1490 len = m->m_len - moff;
1493 * Copy out to the UIO or pass the mbufs back to the SIO.
1494 * The SIO is dealt with when we eat the mbuf, but deal
1495 * with the resid here either way.
1498 uio->uio_resid = resid;
1499 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1500 resid = uio->uio_resid;
1504 resid -= (size_t)len;
1508 * Eat the entire mbuf or just a piece of it
1510 if (len == m->m_len - moff) {
1511 if (m->m_flags & M_EOR)
1514 if (m->m_flags & M_NOTIFICATION)
1515 flags |= MSG_NOTIFICATION;
1517 if (flags & MSG_PEEK) {
1522 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1526 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1530 if (flags & MSG_PEEK) {
1534 n = m_copym(m, 0, len, MB_WAIT);
1540 so->so_rcv.ssb_cc -= len;
1543 if (so->so_oobmark) {
1544 if ((flags & MSG_PEEK) == 0) {
1545 so->so_oobmark -= len;
1546 if (so->so_oobmark == 0) {
1547 sosetstate(so, SS_RCVATMARK);
1552 if (offset == so->so_oobmark)
1556 if (flags & MSG_EOR)
1559 * If the MSG_WAITALL flag is set (for non-atomic socket),
1560 * we must not quit until resid == 0 or an error
1561 * termination. If a signal/timeout occurs, return
1562 * with a short count but without error.
1563 * Keep signalsockbuf locked against other readers.
1565 while ((flags & MSG_WAITALL) && m == NULL &&
1566 resid > 0 && !sosendallatonce(so) &&
1567 so->so_rcv.ssb_mb == NULL) {
1568 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1571 * The window might have closed to zero, make
1572 * sure we send an ack now that we've drained
1573 * the buffer or we might end up blocking until
1574 * the idle takes over (5 seconds).
1576 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1577 so_pru_rcvd(so, flags);
1578 error = ssb_wait(&so->so_rcv);
1580 ssb_unlock(&so->so_rcv);
1584 m = so->so_rcv.ssb_mb;
1589 * If an atomic read was requested but unread data still remains
1590 * in the record, set MSG_TRUNC.
1592 if (m && pr->pr_flags & PR_ATOMIC)
1596 * Cleanup. If an atomic read was requested drop any unread data.
1598 if ((flags & MSG_PEEK) == 0) {
1599 if (m && (pr->pr_flags & PR_ATOMIC))
1600 sbdroprecord(&so->so_rcv.sb);
1601 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1602 so_pru_rcvd(so, flags);
1605 if (orig_resid == resid && orig_resid &&
1606 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1607 ssb_unlock(&so->so_rcv);
1614 ssb_unlock(&so->so_rcv);
1616 lwkt_reltoken(&so->so_rcv.ssb_token);
1618 m_freem(free_chain);
1623 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1624 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1627 struct mbuf *free_chain = NULL;
1628 int flags, len, error, offset;
1629 struct protosw *pr = so->so_proto;
1631 size_t resid, orig_resid;
1634 resid = uio->uio_resid;
1636 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1644 flags = *flagsp &~ MSG_EOR;
1647 if (flags & MSG_OOB) {
1648 m = m_get(MB_WAIT, MT_DATA);
1651 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1657 KKASSERT(resid >= (size_t)m->m_len);
1658 resid -= (size_t)m->m_len;
1659 } while (resid > 0 && m);
1662 uio->uio_resid = resid;
1663 error = uiomove(mtod(m, caddr_t),
1664 (int)szmin(resid, m->m_len),
1666 resid = uio->uio_resid;
1668 } while (uio->uio_resid && error == 0 && m);
1677 * The token interlocks against the protocol thread while
1678 * ssb_lock is a blocking lock against other userland entities.
1680 lwkt_gettoken(&so->so_rcv.ssb_token);
1682 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1686 m = so->so_rcv.ssb_mb;
1688 * If we have less data than requested, block awaiting more
1689 * (subject to any timeout) if:
1690 * 1. the current count is less than the low water mark, or
1691 * 2. MSG_WAITALL is set, and it is possible to do the entire
1692 * receive operation at once if we block (resid <= hiwat).
1693 * 3. MSG_DONTWAIT is not set
1694 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1695 * we have to do the receive in sections, and thus risk returning
1696 * a short count if a timeout or signal occurs after we start.
1698 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1699 (size_t)so->so_rcv.ssb_cc < resid) &&
1700 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1701 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1702 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1706 error = so->so_error;
1707 if ((flags & MSG_PEEK) == 0)
1711 if (so->so_state & SS_CANTRCVMORE) {
1717 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1718 (pr->pr_flags & PR_CONNREQUIRED)) {
1724 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1725 error = EWOULDBLOCK;
1728 ssb_unlock(&so->so_rcv);
1729 error = ssb_wait(&so->so_rcv);
1735 if (uio && uio->uio_td && uio->uio_td->td_proc)
1736 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1739 * note: m should be == sb_mb here. Cache the next record while
1740 * cleaning up. Note that calling m_free*() will break out critical
1743 KKASSERT(m == so->so_rcv.ssb_mb);
1746 * Copy to the UIO or mbuf return chain (*mp).
1750 while (m && resid > 0 && error == 0) {
1751 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1754 soclrstate(so, SS_RCVATMARK);
1755 len = (resid > INT_MAX) ? INT_MAX : resid;
1756 if (so->so_oobmark && len > so->so_oobmark - offset)
1757 len = so->so_oobmark - offset;
1758 if (len > m->m_len - moff)
1759 len = m->m_len - moff;
1762 * Copy out to the UIO or pass the mbufs back to the SIO.
1763 * The SIO is dealt with when we eat the mbuf, but deal
1764 * with the resid here either way.
1767 uio->uio_resid = resid;
1768 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1769 resid = uio->uio_resid;
1773 resid -= (size_t)len;
1777 * Eat the entire mbuf or just a piece of it
1779 if (len == m->m_len - moff) {
1780 if (flags & MSG_PEEK) {
1785 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1789 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1793 if (flags & MSG_PEEK) {
1797 n = m_copym(m, 0, len, MB_WAIT);
1803 so->so_rcv.ssb_cc -= len;
1806 if (so->so_oobmark) {
1807 if ((flags & MSG_PEEK) == 0) {
1808 so->so_oobmark -= len;
1809 if (so->so_oobmark == 0) {
1810 sosetstate(so, SS_RCVATMARK);
1815 if (offset == so->so_oobmark)
1820 * If the MSG_WAITALL flag is set (for non-atomic socket),
1821 * we must not quit until resid == 0 or an error
1822 * termination. If a signal/timeout occurs, return
1823 * with a short count but without error.
1824 * Keep signalsockbuf locked against other readers.
1826 while ((flags & MSG_WAITALL) && m == NULL &&
1827 resid > 0 && !sosendallatonce(so) &&
1828 so->so_rcv.ssb_mb == NULL) {
1829 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1832 * The window might have closed to zero, make
1833 * sure we send an ack now that we've drained
1834 * the buffer or we might end up blocking until
1835 * the idle takes over (5 seconds).
1838 so_pru_rcvd_async(so);
1839 error = ssb_wait(&so->so_rcv);
1841 ssb_unlock(&so->so_rcv);
1845 m = so->so_rcv.ssb_mb;
1850 * Cleanup. If an atomic read was requested drop any unread data.
1852 if ((flags & MSG_PEEK) == 0) {
1854 so_pru_rcvd_async(so);
1857 if (orig_resid == resid && orig_resid &&
1858 (so->so_state & SS_CANTRCVMORE) == 0) {
1859 ssb_unlock(&so->so_rcv);
1866 ssb_unlock(&so->so_rcv);
1868 lwkt_reltoken(&so->so_rcv.ssb_token);
1870 m_freem(free_chain);
1875 * Shut a socket down. Note that we do not get a frontend lock as we
1876 * want to be able to shut the socket down even if another thread is
1877 * blocked in a read(), thus waking it up.
1880 soshutdown(struct socket *so, int how)
1882 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1885 if (how != SHUT_WR) {
1886 /*ssb_lock(&so->so_rcv, M_WAITOK);*/
1888 /*ssb_unlock(&so->so_rcv);*/
1891 return (so_pru_shutdown(so));
1896 sorflush(struct socket *so)
1898 struct signalsockbuf *ssb = &so->so_rcv;
1899 struct protosw *pr = so->so_proto;
1900 struct signalsockbuf asb;
1902 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1904 lwkt_gettoken(&ssb->ssb_token);
1909 * Can't just blow up the ssb structure here
1911 bzero(&ssb->sb, sizeof(ssb->sb));
1916 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1918 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1919 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1920 ssb_release(&asb, so);
1922 lwkt_reltoken(&ssb->ssb_token);
1927 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1929 struct accept_filter_arg *afap = NULL;
1930 struct accept_filter *afp;
1931 struct so_accf *af = so->so_accf;
1934 /* do not set/remove accept filters on non listen sockets */
1935 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1940 /* removing the filter */
1943 if (af->so_accept_filter != NULL &&
1944 af->so_accept_filter->accf_destroy != NULL) {
1945 af->so_accept_filter->accf_destroy(so);
1947 if (af->so_accept_filter_str != NULL) {
1948 kfree(af->so_accept_filter_str, M_ACCF);
1953 so->so_options &= ~SO_ACCEPTFILTER;
1956 /* adding a filter */
1957 /* must remove previous filter first */
1962 /* don't put large objects on the kernel stack */
1963 afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
1964 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1965 afap->af_name[sizeof(afap->af_name)-1] = '\0';
1966 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1969 afp = accept_filt_get(afap->af_name);
1974 af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1975 if (afp->accf_create != NULL) {
1976 if (afap->af_name[0] != '\0') {
1977 int len = strlen(afap->af_name) + 1;
1979 af->so_accept_filter_str = kmalloc(len, M_ACCF,
1981 strcpy(af->so_accept_filter_str, afap->af_name);
1983 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1984 if (af->so_accept_filter_arg == NULL) {
1985 kfree(af->so_accept_filter_str, M_ACCF);
1992 af->so_accept_filter = afp;
1994 so->so_options |= SO_ACCEPTFILTER;
1997 kfree(afap, M_TEMP);
2003 * Perhaps this routine, and sooptcopyout(), below, ought to come in
2004 * an additional variant to handle the case where the option value needs
2005 * to be some kind of integer, but not a specific size.
2006 * In addition to their use here, these functions are also called by the
2007 * protocol-level pr_ctloutput() routines.
2010 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2012 return soopt_to_kbuf(sopt, buf, len, minlen);
2016 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2020 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2021 KKASSERT(kva_p(buf));
2024 * If the user gives us more than we wanted, we ignore it,
2025 * but if we don't get the minimum length the caller
2026 * wants, we return EINVAL. On success, sopt->sopt_valsize
2027 * is set to however much we actually retrieved.
2029 if ((valsize = sopt->sopt_valsize) < minlen)
2032 sopt->sopt_valsize = valsize = len;
2034 bcopy(sopt->sopt_val, buf, valsize);
2040 sosetopt(struct socket *so, struct sockopt *sopt)
2046 struct signalsockbuf *sotmp;
2049 sopt->sopt_dir = SOPT_SET;
2050 if (sopt->sopt_level != SOL_SOCKET) {
2051 if (so->so_proto && so->so_proto->pr_ctloutput) {
2052 return (so_pr_ctloutput(so, sopt));
2054 error = ENOPROTOOPT;
2056 switch (sopt->sopt_name) {
2058 case SO_ACCEPTFILTER:
2059 error = do_setopt_accept_filter(so, sopt);
2065 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2069 so->so_linger = l.l_linger;
2071 so->so_options |= SO_LINGER;
2073 so->so_options &= ~SO_LINGER;
2079 case SO_USELOOPBACK:
2086 error = sooptcopyin(sopt, &optval, sizeof optval,
2091 so->so_options |= sopt->sopt_name;
2093 so->so_options &= ~sopt->sopt_name;
2100 error = sooptcopyin(sopt, &optval, sizeof optval,
2106 * Values < 1 make no sense for any of these
2107 * options, so disallow them.
2114 switch (sopt->sopt_name) {
2117 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2118 &so->so_snd : &so->so_rcv, (u_long)optval,
2120 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2124 sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2125 &so->so_snd : &so->so_rcv;
2126 atomic_clear_int(&sotmp->ssb_flags,
2131 * Make sure the low-water is never greater than
2135 so->so_snd.ssb_lowat =
2136 (optval > so->so_snd.ssb_hiwat) ?
2137 so->so_snd.ssb_hiwat : optval;
2138 atomic_clear_int(&so->so_snd.ssb_flags,
2142 so->so_rcv.ssb_lowat =
2143 (optval > so->so_rcv.ssb_hiwat) ?
2144 so->so_rcv.ssb_hiwat : optval;
2145 atomic_clear_int(&so->so_rcv.ssb_flags,
2153 error = sooptcopyin(sopt, &tv, sizeof tv,
2158 /* assert(hz > 0); */
2159 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2160 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2164 /* assert(tick > 0); */
2165 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2166 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2167 if (val > INT_MAX) {
2171 if (val == 0 && tv.tv_usec != 0)
2174 switch (sopt->sopt_name) {
2176 so->so_snd.ssb_timeo = val;
2179 so->so_rcv.ssb_timeo = val;
2184 error = ENOPROTOOPT;
2187 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2188 (void) so_pr_ctloutput(so, sopt);
2195 /* Helper routine for getsockopt */
2197 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2199 soopt_from_kbuf(sopt, buf, len);
2204 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2209 sopt->sopt_valsize = 0;
2213 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2214 KKASSERT(kva_p(buf));
2217 * Documented get behavior is that we always return a value,
2218 * possibly truncated to fit in the user's buffer.
2219 * Traditional behavior is that we always tell the user
2220 * precisely how much we copied, rather than something useful
2221 * like the total amount we had available for her.
2222 * Note that this interface is not idempotent; the entire answer must
2223 * generated ahead of time.
2225 valsize = szmin(len, sopt->sopt_valsize);
2226 sopt->sopt_valsize = valsize;
2227 if (sopt->sopt_val != 0) {
2228 bcopy(buf, sopt->sopt_val, valsize);
2233 sogetopt(struct socket *so, struct sockopt *sopt)
2240 struct accept_filter_arg *afap;
2244 sopt->sopt_dir = SOPT_GET;
2245 if (sopt->sopt_level != SOL_SOCKET) {
2246 if (so->so_proto && so->so_proto->pr_ctloutput) {
2247 return (so_pr_ctloutput(so, sopt));
2249 return (ENOPROTOOPT);
2251 switch (sopt->sopt_name) {
2253 case SO_ACCEPTFILTER:
2254 if ((so->so_options & SO_ACCEPTCONN) == 0)
2256 afap = kmalloc(sizeof(*afap), M_TEMP,
2258 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2259 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2260 if (so->so_accf->so_accept_filter_str != NULL)
2261 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2263 error = sooptcopyout(sopt, afap, sizeof(*afap));
2264 kfree(afap, M_TEMP);
2269 l.l_onoff = so->so_options & SO_LINGER;
2270 l.l_linger = so->so_linger;
2271 error = sooptcopyout(sopt, &l, sizeof l);
2274 case SO_USELOOPBACK:
2284 optval = so->so_options & sopt->sopt_name;
2286 error = sooptcopyout(sopt, &optval, sizeof optval);
2290 optval = so->so_type;
2294 optval = so->so_error;
2299 optval = so->so_snd.ssb_hiwat;
2303 optval = so->so_rcv.ssb_hiwat;
2307 optval = so->so_snd.ssb_lowat;
2311 optval = so->so_rcv.ssb_lowat;
2316 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2317 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2319 tv.tv_sec = optval / hz;
2320 tv.tv_usec = (optval % hz) * ustick;
2321 error = sooptcopyout(sopt, &tv, sizeof tv);
2325 optval_l = ssb_space(&so->so_snd);
2326 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2330 error = ENOPROTOOPT;
2337 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2339 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2341 struct mbuf *m, *m_prev;
2342 int sopt_size = sopt->sopt_valsize, msize;
2344 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
2348 m->m_len = min(msize, sopt_size);
2349 sopt_size -= m->m_len;
2353 while (sopt_size > 0) {
2354 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
2355 MT_DATA, 0, &msize);
2360 m->m_len = min(msize, sopt_size);
2361 sopt_size -= m->m_len;
2368 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2370 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2372 soopt_to_mbuf(sopt, m);
2377 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2382 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2384 if (sopt->sopt_val == NULL)
2386 val = sopt->sopt_val;
2387 valsize = sopt->sopt_valsize;
2388 while (m != NULL && valsize >= m->m_len) {
2389 bcopy(val, mtod(m, char *), m->m_len);
2390 valsize -= m->m_len;
2391 val = (caddr_t)val + m->m_len;
2394 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2395 panic("ip6_sooptmcopyin");
2398 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2400 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2402 return soopt_from_mbuf(sopt, m);
2406 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2408 struct mbuf *m0 = m;
2413 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2415 if (sopt->sopt_val == NULL)
2417 val = sopt->sopt_val;
2418 maxsize = sopt->sopt_valsize;
2419 while (m != NULL && maxsize >= m->m_len) {
2420 bcopy(mtod(m, char *), val, m->m_len);
2421 maxsize -= m->m_len;
2422 val = (caddr_t)val + m->m_len;
2423 valsize += m->m_len;
2427 /* enough soopt buffer should be given from user-land */
2431 sopt->sopt_valsize = valsize;
2436 sohasoutofband(struct socket *so)
2438 if (so->so_sigio != NULL)
2439 pgsigio(so->so_sigio, SIGURG, 0);
2440 KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
2444 sokqfilter(struct file *fp, struct knote *kn)
2446 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2447 struct signalsockbuf *ssb;
2449 switch (kn->kn_filter) {
2451 if (so->so_options & SO_ACCEPTCONN)
2452 kn->kn_fop = &solisten_filtops;
2454 kn->kn_fop = &soread_filtops;
2458 kn->kn_fop = &sowrite_filtops;
2462 kn->kn_fop = &soexcept_filtops;
2466 return (EOPNOTSUPP);
2469 knote_insert(&ssb->ssb_kq.ki_note, kn);
2470 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2475 filt_sordetach(struct knote *kn)
2477 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2479 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2480 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2481 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2486 filt_soread(struct knote *kn, long hint)
2488 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2490 if (kn->kn_sfflags & NOTE_OOB) {
2491 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2492 kn->kn_fflags |= NOTE_OOB;
2497 kn->kn_data = so->so_rcv.ssb_cc;
2499 if (so->so_state & SS_CANTRCVMORE) {
2501 * Only set NODATA if all data has been exhausted.
2503 if (kn->kn_data == 0)
2504 kn->kn_flags |= EV_NODATA;
2505 kn->kn_flags |= EV_EOF;
2506 kn->kn_fflags = so->so_error;
2509 if (so->so_error) /* temporary udp error */
2511 if (kn->kn_sfflags & NOTE_LOWAT)
2512 return (kn->kn_data >= kn->kn_sdata);
2513 return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2514 !TAILQ_EMPTY(&so->so_comp));
2518 filt_sowdetach(struct knote *kn)
2520 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2522 knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2523 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2524 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2529 filt_sowrite(struct knote *kn, long hint)
2531 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2533 kn->kn_data = ssb_space(&so->so_snd);
2534 if (so->so_state & SS_CANTSENDMORE) {
2535 kn->kn_flags |= (EV_EOF | EV_NODATA);
2536 kn->kn_fflags = so->so_error;
2539 if (so->so_error) /* temporary udp error */
2541 if (((so->so_state & SS_ISCONNECTED) == 0) &&
2542 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2544 if (kn->kn_sfflags & NOTE_LOWAT)
2545 return (kn->kn_data >= kn->kn_sdata);
2546 return (kn->kn_data >= so->so_snd.ssb_lowat);
2551 filt_solisten(struct knote *kn, long hint)
2553 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2555 kn->kn_data = so->so_qlen;
2556 return (! TAILQ_EMPTY(&so->so_comp));