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 $
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/fcntl.h>
71 #include <sys/malloc.h>
73 #include <sys/domain.h>
74 #include <sys/file.h> /* for struct knote */
75 #include <sys/kernel.h>
76 #include <sys/event.h>
78 #include <sys/protosw.h>
79 #include <sys/socket.h>
80 #include <sys/socketvar.h>
81 #include <sys/socketops.h>
82 #include <sys/resourcevar.h>
83 #include <sys/signalvar.h>
84 #include <sys/sysctl.h>
87 #include <vm/vm_zone.h>
89 #include <net/netmsg2.h>
90 #include <net/netisr2.h>
92 #include <sys/thread2.h>
93 #include <sys/socketvar2.h>
94 #include <sys/spinlock2.h>
96 #include <machine/limits.h>
99 extern int tcp_sosend_agglim;
100 extern int tcp_sosend_async;
101 extern int tcp_sosend_jcluster;
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 int soclose_sync(struct socket *so, int fflag);
115 static void soclose_fast(struct socket *so);
117 static struct filterops solisten_filtops =
118 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten };
119 static struct filterops soread_filtops =
120 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
121 static struct filterops sowrite_filtops =
122 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite };
123 static struct filterops soexcept_filtops =
124 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
126 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
127 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
128 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
131 static int somaxconn = SOMAXCONN;
132 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
133 &somaxconn, 0, "Maximum pending socket connection queue size");
135 static int use_soclose_fast = 1;
136 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW,
137 &use_soclose_fast, 0, "Fast socket close");
139 int use_soaccept_pred_fast = 1;
140 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW,
141 &use_soaccept_pred_fast, 0, "Fast socket accept predication");
143 int use_sendfile_async = 1;
144 SYSCTL_INT(_kern_ipc, OID_AUTO, sendfile_async, CTLFLAG_RW,
145 &use_sendfile_async, 0, "sendfile uses asynchronized pru_send");
147 int use_soconnect_async = 1;
148 SYSCTL_INT(_kern_ipc, OID_AUTO, soconnect_async, CTLFLAG_RW,
149 &use_soconnect_async, 0, "soconnect uses asynchronized pru_connect");
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, "soalloc");
182 netmsg_init(&so->so_rcvd_msg.base, so, &netisr_adone_rport,
183 MSGF_DROPABLE | MSGF_PRIORITY,
184 so->so_proto->pr_usrreqs->pru_rcvd);
185 so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC;
186 so->so_state = SS_NOFDREF;
193 socreate(int dom, struct socket **aso, int type,
194 int proto, struct thread *td)
196 struct proc *p = td->td_proc;
199 struct pru_attach_info ai;
203 prp = pffindproto(dom, proto, type);
205 prp = pffindtype(dom, type);
207 if (prp == NULL || prp->pr_usrreqs->pru_attach == 0)
208 return (EPROTONOSUPPORT);
210 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
211 prp->pr_domain->dom_family != PF_LOCAL &&
212 prp->pr_domain->dom_family != PF_INET &&
213 prp->pr_domain->dom_family != PF_INET6 &&
214 prp->pr_domain->dom_family != PF_ROUTE) {
215 return (EPROTONOSUPPORT);
218 if (prp->pr_type != type)
220 so = soalloc(p != NULL, prp);
225 * Callers of socreate() presumably will connect up a descriptor
226 * and call soclose() if they cannot. This represents our so_refs
227 * (which should be 1) from soalloc().
229 soclrstate(so, SS_NOFDREF);
232 * Set a default port for protocol processing. No action will occur
233 * on the socket on this port until an inpcb is attached to it and
234 * is able to match incoming packets, or until the socket becomes
235 * available to userland.
237 * We normally default the socket to the protocol thread on cpu 0,
238 * if protocol does not provide its own method to initialize the
241 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
242 * thread and all pr_*()/pru_*() calls are executed synchronously.
244 if (prp->pr_flags & PR_SYNC_PORT)
245 so->so_port = &netisr_sync_port;
246 else if (prp->pr_initport != NULL)
247 so->so_port = prp->pr_initport();
249 so->so_port = netisr_cpuport(0);
251 TAILQ_INIT(&so->so_incomp);
252 TAILQ_INIT(&so->so_comp);
254 so->so_cred = crhold(p->p_ucred);
255 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
256 ai.p_ucred = p->p_ucred;
257 ai.fd_rdir = p->p_fd->fd_rdir;
260 * Auto-sizing of socket buffers is managed by the protocols and
261 * the appropriate flags must be set in the pru_attach function.
263 error = so_pru_attach(so, proto, &ai);
265 sosetstate(so, SS_NOFDREF);
266 sofree(so); /* from soalloc */
271 * NOTE: Returns referenced socket.
278 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
282 error = so_pru_bind(so, nam, td);
287 sodealloc(struct socket *so)
289 if (so->so_rcv.ssb_hiwat)
290 (void)chgsbsize(so->so_cred->cr_uidinfo,
291 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
292 if (so->so_snd.ssb_hiwat)
293 (void)chgsbsize(so->so_cred->cr_uidinfo,
294 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
296 /* remove accept filter if present */
297 if (so->so_accf != NULL)
298 do_setopt_accept_filter(so, NULL);
301 if (so->so_faddr != NULL)
302 kfree(so->so_faddr, M_SONAME);
307 solisten(struct socket *so, int backlog, struct thread *td)
309 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
312 lwkt_gettoken(&so->so_rcv.ssb_token);
313 if (TAILQ_EMPTY(&so->so_comp))
314 so->so_options |= SO_ACCEPTCONN;
315 lwkt_reltoken(&so->so_rcv.ssb_token);
316 if (backlog < 0 || backlog > somaxconn)
318 so->so_qlimit = backlog;
319 return so_pru_listen(so, td);
323 * Destroy a disconnected socket. This routine is a NOP if entities
324 * still have a reference on the socket:
326 * so_pcb - The protocol stack still has a reference
327 * SS_NOFDREF - There is no longer a file pointer reference
330 sofree(struct socket *so)
335 * This is a bit hackish at the moment. We need to interlock
336 * any accept queue we are on before we potentially lose the
337 * last reference to avoid races against a re-reference from
338 * someone operating on the queue.
340 while ((head = so->so_head) != NULL) {
341 lwkt_getpooltoken(head);
342 if (so->so_head == head)
344 lwkt_relpooltoken(head);
348 * Arbitrage the last free.
350 KKASSERT(so->so_refs > 0);
351 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
353 lwkt_relpooltoken(head);
357 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
358 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
361 * We're done, remove ourselves from the accept queue we are
362 * on, if we are on one.
365 if (so->so_state & SS_INCOMP) {
366 TAILQ_REMOVE(&head->so_incomp, so, so_list);
368 } else if (so->so_state & SS_COMP) {
370 * We must not decommission a socket that's
371 * on the accept(2) queue. If we do, then
372 * accept(2) may hang after select(2) indicated
373 * that the listening socket was ready.
375 lwkt_relpooltoken(head);
378 panic("sofree: not queued");
380 soclrstate(so, SS_INCOMP);
382 lwkt_relpooltoken(head);
384 ssb_release(&so->so_snd, so);
390 * Close a socket on last file table reference removal.
391 * Initiate disconnect if connected.
392 * Free socket when disconnect complete.
395 soclose(struct socket *so, int fflag)
399 funsetown(&so->so_sigio);
400 sosetstate(so, SS_ISCLOSING);
401 if (!use_soclose_fast ||
402 (so->so_proto->pr_flags & PR_SYNC_PORT) ||
403 ((so->so_state & SS_ISCONNECTED) &&
404 (so->so_options & SO_LINGER))) {
405 error = soclose_sync(so, fflag);
414 sodiscard(struct socket *so)
416 lwkt_getpooltoken(so);
417 if (so->so_options & SO_ACCEPTCONN) {
420 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
421 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
423 soclrstate(sp, SS_INCOMP);
427 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
428 TAILQ_REMOVE(&so->so_comp, sp, so_list);
430 soclrstate(sp, SS_COMP);
435 lwkt_relpooltoken(so);
437 if (so->so_state & SS_NOFDREF)
438 panic("soclose: NOFDREF");
439 sosetstate(so, SS_NOFDREF); /* take ref */
443 soinherit(struct socket *so, struct socket *so_inh)
445 TAILQ_HEAD(, socket) comp, incomp;
449 KASSERT(so->so_options & SO_ACCEPTCONN,
450 ("so does not accept connection"));
451 KASSERT(so_inh->so_options & SO_ACCEPTCONN,
452 ("so_inh does not accept connection"));
457 lwkt_getpooltoken(so);
458 lwkt_getpooltoken(so_inh);
461 * Save completed queue and incompleted queue
463 TAILQ_CONCAT(&comp, &so->so_comp, so_list);
467 TAILQ_CONCAT(&incomp, &so->so_incomp, so_list);
468 incqlen = so->so_incqlen;
472 * Append the saved completed queue and incompleted
473 * queue to the socket inherits them.
476 * This may temporarily break the inheriting socket's
479 TAILQ_FOREACH(sp, &comp, so_list) {
480 sp->so_head = so_inh;
482 sp->so_cred = crhold(so_inh->so_cred);
485 TAILQ_FOREACH(sp, &incomp, so_list) {
486 sp->so_head = so_inh;
488 sp->so_cred = crhold(so_inh->so_cred);
491 TAILQ_CONCAT(&so_inh->so_comp, &comp, so_list);
492 so_inh->so_qlen += qlen;
494 TAILQ_CONCAT(&so_inh->so_incomp, &incomp, so_list);
495 so_inh->so_incqlen += incqlen;
497 lwkt_relpooltoken(so_inh);
498 lwkt_relpooltoken(so);
502 * "New" connections have arrived
505 wakeup(&so_inh->so_timeo);
510 soclose_sync(struct socket *so, int fflag)
514 if (so->so_pcb == NULL)
516 if (so->so_state & SS_ISCONNECTED) {
517 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
518 error = sodisconnect(so);
522 if (so->so_options & SO_LINGER) {
523 if ((so->so_state & SS_ISDISCONNECTING) &&
526 while (so->so_state & SS_ISCONNECTED) {
527 error = tsleep(&so->so_timeo, PCATCH,
528 "soclos", so->so_linger * hz);
538 error2 = so_pru_detach(so);
539 if (error2 == EJUSTRETURN) {
541 * Protocol will call sodiscard()
542 * and sofree() for us.
551 so_pru_sync(so); /* unpend async sending */
552 sofree(so); /* dispose of ref */
558 soclose_sofree_async_handler(netmsg_t msg)
560 sofree(msg->base.nm_so);
564 soclose_sofree_async(struct socket *so)
566 struct netmsg_base *base = &so->so_clomsg;
568 netmsg_init(base, so, &netisr_apanic_rport, 0,
569 soclose_sofree_async_handler);
570 lwkt_sendmsg(so->so_port, &base->lmsg);
574 soclose_disconn_async_handler(netmsg_t msg)
576 struct socket *so = msg->base.nm_so;
578 if ((so->so_state & SS_ISCONNECTED) &&
579 (so->so_state & SS_ISDISCONNECTING) == 0)
580 so_pru_disconnect_direct(so);
585 error = so_pru_detach_direct(so);
586 if (error == EJUSTRETURN) {
588 * Protocol will call sodiscard()
589 * and sofree() for us.
600 soclose_disconn_async(struct socket *so)
602 struct netmsg_base *base = &so->so_clomsg;
604 netmsg_init(base, so, &netisr_apanic_rport, 0,
605 soclose_disconn_async_handler);
606 lwkt_sendmsg(so->so_port, &base->lmsg);
610 soclose_detach_async_handler(netmsg_t msg)
612 struct socket *so = msg->base.nm_so;
617 error = so_pru_detach_direct(so);
618 if (error == EJUSTRETURN) {
620 * Protocol will call sodiscard()
621 * and sofree() for us.
632 soclose_detach_async(struct socket *so)
634 struct netmsg_base *base = &so->so_clomsg;
636 netmsg_init(base, so, &netisr_apanic_rport, 0,
637 soclose_detach_async_handler);
638 lwkt_sendmsg(so->so_port, &base->lmsg);
642 soclose_fast(struct socket *so)
644 if (so->so_pcb == NULL)
647 if ((so->so_state & SS_ISCONNECTED) &&
648 (so->so_state & SS_ISDISCONNECTING) == 0) {
649 soclose_disconn_async(so);
654 soclose_detach_async(so);
660 soclose_sofree_async(so);
664 * Abort and destroy a socket. Only one abort can be in progress
665 * at any given moment.
668 soabort_async(struct socket *so)
671 so_pru_abort_async(so);
675 soabort_oncpu(struct socket *so)
678 so_pru_abort_direct(so);
682 * so is passed in ref'd, which becomes owned by
683 * the cleared SS_NOFDREF flag.
686 soaccept_generic(struct socket *so)
688 if ((so->so_state & SS_NOFDREF) == 0)
689 panic("soaccept: !NOFDREF");
690 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */
694 soaccept(struct socket *so, struct sockaddr **nam)
698 soaccept_generic(so);
699 error = so_pru_accept(so, nam);
704 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td,
709 if (so->so_options & SO_ACCEPTCONN)
712 * If protocol is connection-based, can only connect once.
713 * Otherwise, if connected, try to disconnect first.
714 * This allows user to disconnect by connecting to, e.g.,
717 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
718 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
719 (error = sodisconnect(so)))) {
723 * Prevent accumulated error from previous connection
727 if (!sync && so->so_proto->pr_usrreqs->pru_preconnect)
728 error = so_pru_connect_async(so, nam, td);
730 error = so_pru_connect(so, nam, td);
736 soconnect2(struct socket *so1, struct socket *so2)
740 error = so_pru_connect2(so1, so2);
745 sodisconnect(struct socket *so)
749 if ((so->so_state & SS_ISCONNECTED) == 0) {
753 if (so->so_state & SS_ISDISCONNECTING) {
757 error = so_pru_disconnect(so);
762 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
765 * If send must go all at once and message is larger than
766 * send buffering, then hard error.
767 * Lock against other senders.
768 * If must go all at once and not enough room now, then
769 * inform user that this would block and do nothing.
770 * Otherwise, if nonblocking, send as much as possible.
771 * The data to be sent is described by "uio" if nonzero,
772 * otherwise by the mbuf chain "top" (which must be null
773 * if uio is not). Data provided in mbuf chain must be small
774 * enough to send all at once.
776 * Returns nonzero on error, timeout or signal; callers
777 * must check for short counts if EINTR/ERESTART are returned.
778 * Data and control buffers are freed on return.
781 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
782 struct mbuf *top, struct mbuf *control, int flags,
789 int clen = 0, error, dontroute, mlen;
790 int atomic = sosendallatonce(so) || top;
794 resid = uio->uio_resid;
796 resid = (size_t)top->m_pkthdr.len;
799 for (m = top; m; m = m->m_next)
801 KKASSERT(top->m_pkthdr.len == len);
806 * WARNING! resid is unsigned, space and len are signed. space
807 * can wind up negative if the sockbuf is overcommitted.
809 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
810 * type sockets since that's an error.
812 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
818 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
819 (so->so_proto->pr_flags & PR_ATOMIC);
820 if (td->td_lwp != NULL)
821 td->td_lwp->lwp_ru.ru_msgsnd++;
823 clen = control->m_len;
824 #define gotoerr(errcode) { error = errcode; goto release; }
827 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
832 if (so->so_state & SS_CANTSENDMORE)
835 error = so->so_error;
839 if ((so->so_state & SS_ISCONNECTED) == 0) {
841 * `sendto' and `sendmsg' is allowed on a connection-
842 * based socket if it supports implied connect.
843 * Return ENOTCONN if not connected and no address is
846 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
847 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
848 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
849 !(resid == 0 && clen != 0))
851 } else if (addr == NULL)
852 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
853 ENOTCONN : EDESTADDRREQ);
855 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
856 clen > so->so_snd.ssb_hiwat) {
859 space = ssb_space(&so->so_snd);
862 if ((space < 0 || (size_t)space < resid + clen) && uio &&
863 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
864 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
865 gotoerr(EWOULDBLOCK);
866 ssb_unlock(&so->so_snd);
867 error = ssb_wait(&so->so_snd);
877 * Data is prepackaged in "top".
881 top->m_flags |= M_EOR;
885 m = m_getl((int)resid, M_WAITOK, MT_DATA,
886 top == NULL ? M_PKTHDR : 0, &mlen);
889 m->m_pkthdr.rcvif = NULL;
891 len = imin((int)szmin(mlen, resid), space);
892 if (resid < MINCLSIZE) {
894 * For datagram protocols, leave room
895 * for protocol headers in first mbuf.
897 if (atomic && top == NULL && len < mlen)
901 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
902 resid = uio->uio_resid;
905 top->m_pkthdr.len += len;
911 top->m_flags |= M_EOR;
914 } while (space > 0 && atomic);
916 so->so_options |= SO_DONTROUTE;
917 if (flags & MSG_OOB) {
918 pru_flags = PRUS_OOB;
919 } else if ((flags & MSG_EOF) &&
920 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
923 * If the user set MSG_EOF, the protocol
924 * understands this flag and nothing left to
925 * send then use PRU_SEND_EOF instead of PRU_SEND.
927 pru_flags = PRUS_EOF;
928 } else if (resid > 0 && space > 0) {
929 /* If there is more to send, set PRUS_MORETOCOME */
930 pru_flags = PRUS_MORETOCOME;
935 * XXX all the SS_CANTSENDMORE checks previously
936 * done could be out of date. We could have recieved
937 * a reset packet in an interrupt or maybe we slept
938 * while doing page faults in uiomove() etc. We could
939 * probably recheck again inside the splnet() protection
940 * here, but there are probably other places that this
941 * also happens. We must rethink this.
943 error = so_pru_send(so, pru_flags, top, addr, control, td);
945 so->so_options &= ~SO_DONTROUTE;
952 } while (resid && space > 0);
956 ssb_unlock(&so->so_snd);
967 * A specialization of sosend() for UDP based on protocol-specific knowledge:
968 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
969 * sosendallatonce() returns true,
970 * the "atomic" variable is true,
971 * and sosendudp() blocks until space is available for the entire send.
972 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
973 * PR_IMPLOPCL flags set.
974 * UDP has no out-of-band data.
975 * UDP has no control data.
976 * UDP does not support MSG_EOR.
979 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
980 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
983 int error, pru_flags = 0;
986 if (td->td_lwp != NULL)
987 td->td_lwp->lwp_ru.ru_msgsnd++;
991 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
992 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
995 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
999 if (so->so_state & SS_CANTSENDMORE)
1002 error = so->so_error;
1006 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
1007 gotoerr(EDESTADDRREQ);
1008 if (resid > so->so_snd.ssb_hiwat)
1010 space = ssb_space(&so->so_snd);
1011 if (uio && (space < 0 || (size_t)space < resid)) {
1012 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1013 gotoerr(EWOULDBLOCK);
1014 ssb_unlock(&so->so_snd);
1015 error = ssb_wait(&so->so_snd);
1022 int hdrlen = max_hdr;
1025 * We try to optimize out the additional mbuf
1026 * allocations in M_PREPEND() on output path, e.g.
1027 * - udp_output(), when it tries to prepend protocol
1029 * - Link layer output function, when it tries to
1030 * prepend link layer header.
1032 * This probably will not benefit any data that will
1033 * be fragmented, so this optimization is only performed
1034 * when the size of data and max size of protocol+link
1035 * headers fit into one mbuf cluster.
1037 if (uio->uio_resid > MCLBYTES - hdrlen ||
1038 !udp_sosend_prepend) {
1039 top = m_uiomove(uio);
1045 top = m_getl(uio->uio_resid + hdrlen, M_WAITOK,
1046 MT_DATA, M_PKTHDR, &nsize);
1047 KASSERT(nsize >= uio->uio_resid + hdrlen,
1048 ("sosendudp invalid nsize %d, "
1049 "resid %zu, hdrlen %d",
1050 nsize, uio->uio_resid, hdrlen));
1052 top->m_len = uio->uio_resid;
1053 top->m_pkthdr.len = uio->uio_resid;
1054 top->m_data += hdrlen;
1056 error = uiomove(mtod(top, caddr_t), top->m_len, uio);
1062 if (flags & MSG_DONTROUTE)
1063 pru_flags |= PRUS_DONTROUTE;
1065 if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
1066 so_pru_send_async(so, pru_flags, top, addr, NULL, td);
1069 error = so_pru_send(so, pru_flags, top, addr, NULL, td);
1071 top = NULL; /* sent or freed in lower layer */
1074 ssb_unlock(&so->so_snd);
1082 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1083 struct mbuf *top, struct mbuf *control, int flags,
1095 KKASSERT(top == NULL);
1097 resid = uio->uio_resid;
1100 resid = (size_t)top->m_pkthdr.len;
1103 for (m = top; m; m = m->m_next)
1105 KKASSERT(top->m_pkthdr.len == len);
1110 * WARNING! resid is unsigned, space and len are signed. space
1111 * can wind up negative if the sockbuf is overcommitted.
1113 * Also check to make sure that MSG_EOR isn't used on TCP
1115 if (flags & MSG_EOR) {
1121 /* TCP doesn't do control messages (rights, creds, etc) */
1122 if (control->m_len) {
1126 m_freem(control); /* empty control, just free it */
1130 if (td->td_lwp != NULL)
1131 td->td_lwp->lwp_ru.ru_msgsnd++;
1133 #define gotoerr(errcode) { error = errcode; goto release; }
1136 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1141 if (so->so_state & SS_CANTSENDMORE)
1144 error = so->so_error;
1148 if ((so->so_state & SS_ISCONNECTED) == 0 &&
1149 (so->so_state & SS_ISCONFIRMING) == 0)
1151 if (allatonce && resid > so->so_snd.ssb_hiwat)
1154 space = ssb_space_prealloc(&so->so_snd);
1155 if (flags & MSG_OOB)
1157 if ((space < 0 || (size_t)space < resid) && !allatonce &&
1158 space < so->so_snd.ssb_lowat) {
1159 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1160 gotoerr(EWOULDBLOCK);
1161 ssb_unlock(&so->so_snd);
1162 error = ssb_wait(&so->so_snd);
1169 int cnt = 0, async = 0;
1173 * Data is prepackaged in "top".
1177 if (resid > INT_MAX)
1179 if (tcp_sosend_jcluster) {
1180 m = m_getlj((int)resid, M_WAITOK, MT_DATA,
1181 top == NULL ? M_PKTHDR : 0, &mlen);
1183 m = m_getl((int)resid, M_WAITOK, MT_DATA,
1184 top == NULL ? M_PKTHDR : 0, &mlen);
1187 m->m_pkthdr.len = 0;
1188 m->m_pkthdr.rcvif = NULL;
1190 len = imin((int)szmin(mlen, resid), space);
1192 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1193 resid = uio->uio_resid;
1196 top->m_pkthdr.len += len;
1203 } while (space > 0 && cnt < tcp_sosend_agglim);
1205 if (tcp_sosend_async)
1208 if (flags & MSG_OOB) {
1209 pru_flags = PRUS_OOB;
1211 } else if ((flags & MSG_EOF) && resid == 0) {
1212 pru_flags = PRUS_EOF;
1213 } else if (resid > 0 && space > 0) {
1214 /* If there is more to send, set PRUS_MORETOCOME */
1215 pru_flags = PRUS_MORETOCOME;
1221 if (flags & MSG_SYNC)
1225 * XXX all the SS_CANTSENDMORE checks previously
1226 * done could be out of date. We could have recieved
1227 * a reset packet in an interrupt or maybe we slept
1228 * while doing page faults in uiomove() etc. We could
1229 * probably recheck again inside the splnet() protection
1230 * here, but there are probably other places that this
1231 * also happens. We must rethink this.
1233 for (m = top; m; m = m->m_next)
1234 ssb_preallocstream(&so->so_snd, m);
1236 error = so_pru_send(so, pru_flags, top,
1239 so_pru_send_async(so, pru_flags, top,
1248 } while (resid && space > 0);
1252 ssb_unlock(&so->so_snd);
1263 * Implement receive operations on a socket.
1265 * We depend on the way that records are added to the signalsockbuf
1266 * by sbappend*. In particular, each record (mbufs linked through m_next)
1267 * must begin with an address if the protocol so specifies,
1268 * followed by an optional mbuf or mbufs containing ancillary data,
1269 * and then zero or more mbufs of data.
1271 * Although the signalsockbuf is locked, new data may still be appended.
1272 * A token inside the ssb_lock deals with MP issues and still allows
1273 * the network to access the socket if we block in a uio.
1275 * The caller may receive the data as a single mbuf chain by supplying
1276 * an mbuf **mp0 for use in returning the chain. The uio is then used
1277 * only for the count in uio_resid.
1280 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1281 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1284 struct mbuf *free_chain = NULL;
1285 int flags, len, error, offset;
1286 struct protosw *pr = so->so_proto;
1288 size_t resid, orig_resid;
1291 resid = uio->uio_resid;
1293 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1301 flags = *flagsp &~ MSG_EOR;
1304 if (flags & MSG_OOB) {
1305 m = m_get(M_WAITOK, MT_DATA);
1308 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1314 KKASSERT(resid >= (size_t)m->m_len);
1315 resid -= (size_t)m->m_len;
1316 } while (resid > 0 && m);
1319 uio->uio_resid = resid;
1320 error = uiomove(mtod(m, caddr_t),
1321 (int)szmin(resid, m->m_len),
1323 resid = uio->uio_resid;
1325 } while (uio->uio_resid && error == 0 && m);
1332 if ((so->so_state & SS_ISCONFIRMING) && resid)
1336 * The token interlocks against the protocol thread while
1337 * ssb_lock is a blocking lock against other userland entities.
1339 lwkt_gettoken(&so->so_rcv.ssb_token);
1341 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1345 m = so->so_rcv.ssb_mb;
1347 * If we have less data than requested, block awaiting more
1348 * (subject to any timeout) if:
1349 * 1. the current count is less than the low water mark, or
1350 * 2. MSG_WAITALL is set, and it is possible to do the entire
1351 * receive operation at once if we block (resid <= hiwat).
1352 * 3. MSG_DONTWAIT is not set
1353 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1354 * we have to do the receive in sections, and thus risk returning
1355 * a short count if a timeout or signal occurs after we start.
1357 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1358 (size_t)so->so_rcv.ssb_cc < resid) &&
1359 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1360 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1361 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1362 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1366 error = so->so_error;
1367 if ((flags & MSG_PEEK) == 0)
1371 if (so->so_state & SS_CANTRCVMORE) {
1377 for (; m; m = m->m_next) {
1378 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1379 m = so->so_rcv.ssb_mb;
1383 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1384 (pr->pr_flags & PR_CONNREQUIRED)) {
1390 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1391 error = EWOULDBLOCK;
1394 ssb_unlock(&so->so_rcv);
1395 error = ssb_wait(&so->so_rcv);
1401 if (uio && uio->uio_td && uio->uio_td->td_proc)
1402 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1405 * note: m should be == sb_mb here. Cache the next record while
1406 * cleaning up. Note that calling m_free*() will break out critical
1409 KKASSERT(m == so->so_rcv.ssb_mb);
1412 * Skip any address mbufs prepending the record.
1414 if (pr->pr_flags & PR_ADDR) {
1415 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1418 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1419 if (flags & MSG_PEEK)
1422 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1426 * Skip any control mbufs prepending the record.
1428 while (m && m->m_type == MT_CONTROL && error == 0) {
1429 if (flags & MSG_PEEK) {
1431 *controlp = m_copy(m, 0, m->m_len);
1432 m = m->m_next; /* XXX race */
1435 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1436 if (pr->pr_domain->dom_externalize &&
1437 mtod(m, struct cmsghdr *)->cmsg_type ==
1439 error = (*pr->pr_domain->dom_externalize)(m);
1443 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1446 if (controlp && *controlp) {
1448 controlp = &(*controlp)->m_next;
1457 if (type == MT_OOBDATA)
1462 * Copy to the UIO or mbuf return chain (*mp).
1466 while (m && resid > 0 && error == 0) {
1467 if (m->m_type == MT_OOBDATA) {
1468 if (type != MT_OOBDATA)
1470 } else if (type == MT_OOBDATA)
1473 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1475 soclrstate(so, SS_RCVATMARK);
1476 len = (resid > INT_MAX) ? INT_MAX : resid;
1477 if (so->so_oobmark && len > so->so_oobmark - offset)
1478 len = so->so_oobmark - offset;
1479 if (len > m->m_len - moff)
1480 len = m->m_len - moff;
1483 * Copy out to the UIO or pass the mbufs back to the SIO.
1484 * The SIO is dealt with when we eat the mbuf, but deal
1485 * with the resid here either way.
1488 uio->uio_resid = resid;
1489 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1490 resid = uio->uio_resid;
1494 resid -= (size_t)len;
1498 * Eat the entire mbuf or just a piece of it
1500 if (len == m->m_len - moff) {
1501 if (m->m_flags & M_EOR)
1503 if (flags & MSG_PEEK) {
1508 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1512 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1516 if (flags & MSG_PEEK) {
1520 n = m_copym(m, 0, len, M_WAITOK);
1526 so->so_rcv.ssb_cc -= len;
1529 if (so->so_oobmark) {
1530 if ((flags & MSG_PEEK) == 0) {
1531 so->so_oobmark -= len;
1532 if (so->so_oobmark == 0) {
1533 sosetstate(so, SS_RCVATMARK);
1538 if (offset == so->so_oobmark)
1542 if (flags & MSG_EOR)
1545 * If the MSG_WAITALL flag is set (for non-atomic socket),
1546 * we must not quit until resid == 0 or an error
1547 * termination. If a signal/timeout occurs, return
1548 * with a short count but without error.
1549 * Keep signalsockbuf locked against other readers.
1551 while ((flags & MSG_WAITALL) && m == NULL &&
1552 resid > 0 && !sosendallatonce(so) &&
1553 so->so_rcv.ssb_mb == NULL) {
1554 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1557 * The window might have closed to zero, make
1558 * sure we send an ack now that we've drained
1559 * the buffer or we might end up blocking until
1560 * the idle takes over (5 seconds).
1562 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1563 so_pru_rcvd(so, flags);
1564 error = ssb_wait(&so->so_rcv);
1566 ssb_unlock(&so->so_rcv);
1570 m = so->so_rcv.ssb_mb;
1575 * If an atomic read was requested but unread data still remains
1576 * in the record, set MSG_TRUNC.
1578 if (m && pr->pr_flags & PR_ATOMIC)
1582 * Cleanup. If an atomic read was requested drop any unread data.
1584 if ((flags & MSG_PEEK) == 0) {
1585 if (m && (pr->pr_flags & PR_ATOMIC))
1586 sbdroprecord(&so->so_rcv.sb);
1587 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1588 so_pru_rcvd(so, flags);
1591 if (orig_resid == resid && orig_resid &&
1592 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1593 ssb_unlock(&so->so_rcv);
1600 ssb_unlock(&so->so_rcv);
1602 lwkt_reltoken(&so->so_rcv.ssb_token);
1604 m_freem(free_chain);
1609 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1610 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1613 struct mbuf *free_chain = NULL;
1614 int flags, len, error, offset;
1615 struct protosw *pr = so->so_proto;
1618 size_t resid, orig_resid, restmp;
1621 resid = uio->uio_resid;
1623 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1631 flags = *flagsp &~ MSG_EOR;
1634 if (flags & MSG_OOB) {
1635 m = m_get(M_WAITOK, MT_DATA);
1638 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1644 KKASSERT(resid >= (size_t)m->m_len);
1645 resid -= (size_t)m->m_len;
1646 } while (resid > 0 && m);
1649 uio->uio_resid = resid;
1650 error = uiomove(mtod(m, caddr_t),
1651 (int)szmin(resid, m->m_len),
1653 resid = uio->uio_resid;
1655 } while (uio->uio_resid && error == 0 && m);
1664 * The token interlocks against the protocol thread while
1665 * ssb_lock is a blocking lock against other userland entities.
1667 * Lock a limited number of mbufs (not all, so sbcompress() still
1668 * works well). The token is used as an interlock for sbwait() so
1669 * release it afterwords.
1672 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1676 lwkt_gettoken(&so->so_rcv.ssb_token);
1677 m = so->so_rcv.ssb_mb;
1680 * If we have less data than requested, block awaiting more
1681 * (subject to any timeout) if:
1682 * 1. the current count is less than the low water mark, or
1683 * 2. MSG_WAITALL is set, and it is possible to do the entire
1684 * receive operation at once if we block (resid <= hiwat).
1685 * 3. MSG_DONTWAIT is not set
1686 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1687 * we have to do the receive in sections, and thus risk returning
1688 * a short count if a timeout or signal occurs after we start.
1690 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1691 (size_t)so->so_rcv.ssb_cc < resid) &&
1692 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1693 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1694 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1698 lwkt_reltoken(&so->so_rcv.ssb_token);
1699 error = so->so_error;
1700 if ((flags & MSG_PEEK) == 0)
1704 if (so->so_state & SS_CANTRCVMORE) {
1707 lwkt_reltoken(&so->so_rcv.ssb_token);
1710 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1711 (pr->pr_flags & PR_CONNREQUIRED)) {
1712 lwkt_reltoken(&so->so_rcv.ssb_token);
1717 lwkt_reltoken(&so->so_rcv.ssb_token);
1720 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1721 lwkt_reltoken(&so->so_rcv.ssb_token);
1722 error = EWOULDBLOCK;
1725 ssb_unlock(&so->so_rcv);
1726 error = ssb_wait(&so->so_rcv);
1727 lwkt_reltoken(&so->so_rcv.ssb_token);
1739 while (n && restmp < resid) {
1740 n->m_flags |= M_SOLOCKED;
1742 if (n->m_next == NULL)
1749 * Release token for loop
1751 lwkt_reltoken(&so->so_rcv.ssb_token);
1752 if (uio && uio->uio_td && uio->uio_td->td_proc)
1753 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1756 * note: m should be == sb_mb here. Cache the next record while
1757 * cleaning up. Note that calling m_free*() will break out critical
1760 KKASSERT(m == so->so_rcv.ssb_mb);
1763 * Copy to the UIO or mbuf return chain (*mp).
1765 * NOTE: Token is not held for loop
1771 while (m && (m->m_flags & M_SOLOCKED) && resid > 0 && error == 0) {
1772 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1775 soclrstate(so, SS_RCVATMARK);
1776 len = (resid > INT_MAX) ? INT_MAX : resid;
1777 if (so->so_oobmark && len > so->so_oobmark - offset)
1778 len = so->so_oobmark - offset;
1779 if (len > m->m_len - moff)
1780 len = m->m_len - moff;
1783 * Copy out to the UIO or pass the mbufs back to the SIO.
1784 * The SIO is dealt with when we eat the mbuf, but deal
1785 * with the resid here either way.
1788 uio->uio_resid = resid;
1789 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1790 resid = uio->uio_resid;
1794 resid -= (size_t)len;
1798 * Eat the entire mbuf or just a piece of it
1801 if (len == m->m_len - moff) {
1811 if (so->so_oobmark && offset == so->so_oobmark) {
1818 * Synchronize sockbuf with data we read.
1820 * NOTE: (m) is junk on entry (it could be left over from the
1823 if ((flags & MSG_PEEK) == 0) {
1824 lwkt_gettoken(&so->so_rcv.ssb_token);
1825 m = so->so_rcv.ssb_mb;
1826 while (m && offset >= m->m_len) {
1827 if (so->so_oobmark) {
1828 so->so_oobmark -= m->m_len;
1829 if (so->so_oobmark == 0) {
1830 sosetstate(so, SS_RCVATMARK);
1836 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1840 m = sbunlinkmbuf(&so->so_rcv.sb,
1847 n = m_copym(m, 0, offset, M_WAITOK);
1851 m->m_data += offset;
1853 so->so_rcv.ssb_cc -= offset;
1854 if (so->so_oobmark) {
1855 so->so_oobmark -= offset;
1856 if (so->so_oobmark == 0) {
1857 sosetstate(so, SS_RCVATMARK);
1863 lwkt_reltoken(&so->so_rcv.ssb_token);
1867 * If the MSG_WAITALL flag is set (for non-atomic socket),
1868 * we must not quit until resid == 0 or an error termination.
1870 * If a signal/timeout occurs, return with a short count but without
1873 * Keep signalsockbuf locked against other readers.
1875 * XXX if MSG_PEEK we currently do quit.
1877 if ((flags & MSG_WAITALL) && !(flags & MSG_PEEK) &&
1878 didoob == 0 && resid > 0 &&
1879 !sosendallatonce(so)) {
1880 lwkt_gettoken(&so->so_rcv.ssb_token);
1882 while ((m = so->so_rcv.ssb_mb) == NULL) {
1883 if (so->so_error || (so->so_state & SS_CANTRCVMORE)) {
1884 error = so->so_error;
1888 * The window might have closed to zero, make
1889 * sure we send an ack now that we've drained
1890 * the buffer or we might end up blocking until
1891 * the idle takes over (5 seconds).
1894 so_pru_rcvd_async(so);
1895 if (so->so_rcv.ssb_mb == NULL)
1896 error = ssb_wait(&so->so_rcv);
1898 lwkt_reltoken(&so->so_rcv.ssb_token);
1899 ssb_unlock(&so->so_rcv);
1904 if (m && error == 0)
1906 lwkt_reltoken(&so->so_rcv.ssb_token);
1910 * Token not held here.
1912 * Cleanup. If an atomic read was requested drop any unread data XXX
1914 if ((flags & MSG_PEEK) == 0) {
1916 so_pru_rcvd_async(so);
1919 if (orig_resid == resid && orig_resid &&
1920 (so->so_state & SS_CANTRCVMORE) == 0) {
1921 ssb_unlock(&so->so_rcv);
1928 ssb_unlock(&so->so_rcv);
1931 m_freem(free_chain);
1936 * Shut a socket down. Note that we do not get a frontend lock as we
1937 * want to be able to shut the socket down even if another thread is
1938 * blocked in a read(), thus waking it up.
1941 soshutdown(struct socket *so, int how)
1943 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1946 if (how != SHUT_WR) {
1947 /*ssb_lock(&so->so_rcv, M_WAITOK);*/
1949 /*ssb_unlock(&so->so_rcv);*/
1952 return (so_pru_shutdown(so));
1957 sorflush(struct socket *so)
1959 struct signalsockbuf *ssb = &so->so_rcv;
1960 struct protosw *pr = so->so_proto;
1961 struct signalsockbuf asb;
1963 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1965 lwkt_gettoken(&ssb->ssb_token);
1970 * Can't just blow up the ssb structure here
1972 bzero(&ssb->sb, sizeof(ssb->sb));
1977 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1979 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1980 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1981 ssb_release(&asb, so);
1983 lwkt_reltoken(&ssb->ssb_token);
1988 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1990 struct accept_filter_arg *afap = NULL;
1991 struct accept_filter *afp;
1992 struct so_accf *af = so->so_accf;
1995 /* do not set/remove accept filters on non listen sockets */
1996 if ((so->so_options & SO_ACCEPTCONN) == 0) {
2001 /* removing the filter */
2004 if (af->so_accept_filter != NULL &&
2005 af->so_accept_filter->accf_destroy != NULL) {
2006 af->so_accept_filter->accf_destroy(so);
2008 if (af->so_accept_filter_str != NULL) {
2009 kfree(af->so_accept_filter_str, M_ACCF);
2014 so->so_options &= ~SO_ACCEPTFILTER;
2017 /* adding a filter */
2018 /* must remove previous filter first */
2023 /* don't put large objects on the kernel stack */
2024 afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
2025 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
2026 afap->af_name[sizeof(afap->af_name)-1] = '\0';
2027 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
2030 afp = accept_filt_get(afap->af_name);
2035 af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
2036 if (afp->accf_create != NULL) {
2037 if (afap->af_name[0] != '\0') {
2038 int len = strlen(afap->af_name) + 1;
2040 af->so_accept_filter_str = kmalloc(len, M_ACCF,
2042 strcpy(af->so_accept_filter_str, afap->af_name);
2044 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
2045 if (af->so_accept_filter_arg == NULL) {
2046 kfree(af->so_accept_filter_str, M_ACCF);
2053 af->so_accept_filter = afp;
2055 so->so_options |= SO_ACCEPTFILTER;
2058 kfree(afap, M_TEMP);
2064 * Perhaps this routine, and sooptcopyout(), below, ought to come in
2065 * an additional variant to handle the case where the option value needs
2066 * to be some kind of integer, but not a specific size.
2067 * In addition to their use here, these functions are also called by the
2068 * protocol-level pr_ctloutput() routines.
2071 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2073 return soopt_to_kbuf(sopt, buf, len, minlen);
2077 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2081 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2082 KKASSERT(kva_p(buf));
2085 * If the user gives us more than we wanted, we ignore it,
2086 * but if we don't get the minimum length the caller
2087 * wants, we return EINVAL. On success, sopt->sopt_valsize
2088 * is set to however much we actually retrieved.
2090 if ((valsize = sopt->sopt_valsize) < minlen)
2093 sopt->sopt_valsize = valsize = len;
2095 bcopy(sopt->sopt_val, buf, valsize);
2101 sosetopt(struct socket *so, struct sockopt *sopt)
2107 struct signalsockbuf *sotmp;
2110 sopt->sopt_dir = SOPT_SET;
2111 if (sopt->sopt_level != SOL_SOCKET) {
2112 if (so->so_proto && so->so_proto->pr_ctloutput) {
2113 return (so_pr_ctloutput(so, sopt));
2115 error = ENOPROTOOPT;
2117 switch (sopt->sopt_name) {
2119 case SO_ACCEPTFILTER:
2120 error = do_setopt_accept_filter(so, sopt);
2126 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2130 so->so_linger = l.l_linger;
2132 so->so_options |= SO_LINGER;
2134 so->so_options &= ~SO_LINGER;
2140 case SO_USELOOPBACK:
2147 error = sooptcopyin(sopt, &optval, sizeof optval,
2152 so->so_options |= sopt->sopt_name;
2154 so->so_options &= ~sopt->sopt_name;
2161 error = sooptcopyin(sopt, &optval, sizeof optval,
2167 * Values < 1 make no sense for any of these
2168 * options, so disallow them.
2175 switch (sopt->sopt_name) {
2178 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2179 &so->so_snd : &so->so_rcv, (u_long)optval,
2181 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2185 sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2186 &so->so_snd : &so->so_rcv;
2187 atomic_clear_int(&sotmp->ssb_flags,
2192 * Make sure the low-water is never greater than
2196 so->so_snd.ssb_lowat =
2197 (optval > so->so_snd.ssb_hiwat) ?
2198 so->so_snd.ssb_hiwat : optval;
2199 atomic_clear_int(&so->so_snd.ssb_flags,
2203 so->so_rcv.ssb_lowat =
2204 (optval > so->so_rcv.ssb_hiwat) ?
2205 so->so_rcv.ssb_hiwat : optval;
2206 atomic_clear_int(&so->so_rcv.ssb_flags,
2214 error = sooptcopyin(sopt, &tv, sizeof tv,
2219 /* assert(hz > 0); */
2220 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2221 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2225 /* assert(tick > 0); */
2226 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2227 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2228 if (val > INT_MAX) {
2232 if (val == 0 && tv.tv_usec != 0)
2235 switch (sopt->sopt_name) {
2237 so->so_snd.ssb_timeo = val;
2240 so->so_rcv.ssb_timeo = val;
2245 error = ENOPROTOOPT;
2248 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2249 (void) so_pr_ctloutput(so, sopt);
2256 /* Helper routine for getsockopt */
2258 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2260 soopt_from_kbuf(sopt, buf, len);
2265 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2270 sopt->sopt_valsize = 0;
2274 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2275 KKASSERT(kva_p(buf));
2278 * Documented get behavior is that we always return a value,
2279 * possibly truncated to fit in the user's buffer.
2280 * Traditional behavior is that we always tell the user
2281 * precisely how much we copied, rather than something useful
2282 * like the total amount we had available for her.
2283 * Note that this interface is not idempotent; the entire answer must
2284 * generated ahead of time.
2286 valsize = szmin(len, sopt->sopt_valsize);
2287 sopt->sopt_valsize = valsize;
2288 if (sopt->sopt_val != 0) {
2289 bcopy(buf, sopt->sopt_val, valsize);
2294 sogetopt(struct socket *so, struct sockopt *sopt)
2301 struct accept_filter_arg *afap;
2305 sopt->sopt_dir = SOPT_GET;
2306 if (sopt->sopt_level != SOL_SOCKET) {
2307 if (so->so_proto && so->so_proto->pr_ctloutput) {
2308 return (so_pr_ctloutput(so, sopt));
2310 return (ENOPROTOOPT);
2312 switch (sopt->sopt_name) {
2314 case SO_ACCEPTFILTER:
2315 if ((so->so_options & SO_ACCEPTCONN) == 0)
2317 afap = kmalloc(sizeof(*afap), M_TEMP,
2319 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2320 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2321 if (so->so_accf->so_accept_filter_str != NULL)
2322 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2324 error = sooptcopyout(sopt, afap, sizeof(*afap));
2325 kfree(afap, M_TEMP);
2330 l.l_onoff = so->so_options & SO_LINGER;
2331 l.l_linger = so->so_linger;
2332 error = sooptcopyout(sopt, &l, sizeof l);
2335 case SO_USELOOPBACK:
2345 optval = so->so_options & sopt->sopt_name;
2347 error = sooptcopyout(sopt, &optval, sizeof optval);
2351 optval = so->so_type;
2355 optval = so->so_error;
2360 optval = so->so_snd.ssb_hiwat;
2364 optval = so->so_rcv.ssb_hiwat;
2368 optval = so->so_snd.ssb_lowat;
2372 optval = so->so_rcv.ssb_lowat;
2377 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2378 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2380 tv.tv_sec = optval / hz;
2381 tv.tv_usec = (optval % hz) * ustick;
2382 error = sooptcopyout(sopt, &tv, sizeof tv);
2386 optval_l = ssb_space(&so->so_snd);
2387 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2391 optval = -1; /* no hint */
2395 error = ENOPROTOOPT;
2398 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput)
2399 so_pr_ctloutput(so, sopt);
2404 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2406 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2408 struct mbuf *m, *m_prev;
2409 int sopt_size = sopt->sopt_valsize, msize;
2411 m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA,
2415 m->m_len = min(msize, sopt_size);
2416 sopt_size -= m->m_len;
2420 while (sopt_size > 0) {
2421 m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT,
2422 MT_DATA, 0, &msize);
2427 m->m_len = min(msize, sopt_size);
2428 sopt_size -= m->m_len;
2435 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2437 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2439 soopt_to_mbuf(sopt, m);
2444 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2449 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2451 if (sopt->sopt_val == NULL)
2453 val = sopt->sopt_val;
2454 valsize = sopt->sopt_valsize;
2455 while (m != NULL && valsize >= m->m_len) {
2456 bcopy(val, mtod(m, char *), m->m_len);
2457 valsize -= m->m_len;
2458 val = (caddr_t)val + m->m_len;
2461 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2462 panic("ip6_sooptmcopyin");
2465 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2467 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2469 return soopt_from_mbuf(sopt, m);
2473 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2475 struct mbuf *m0 = m;
2480 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2482 if (sopt->sopt_val == NULL)
2484 val = sopt->sopt_val;
2485 maxsize = sopt->sopt_valsize;
2486 while (m != NULL && maxsize >= m->m_len) {
2487 bcopy(mtod(m, char *), val, m->m_len);
2488 maxsize -= m->m_len;
2489 val = (caddr_t)val + m->m_len;
2490 valsize += m->m_len;
2494 /* enough soopt buffer should be given from user-land */
2498 sopt->sopt_valsize = valsize;
2503 sohasoutofband(struct socket *so)
2505 if (so->so_sigio != NULL)
2506 pgsigio(so->so_sigio, SIGURG, 0);
2507 KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
2511 sokqfilter(struct file *fp, struct knote *kn)
2513 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2514 struct signalsockbuf *ssb;
2516 switch (kn->kn_filter) {
2518 if (so->so_options & SO_ACCEPTCONN)
2519 kn->kn_fop = &solisten_filtops;
2521 kn->kn_fop = &soread_filtops;
2525 kn->kn_fop = &sowrite_filtops;
2529 kn->kn_fop = &soexcept_filtops;
2533 return (EOPNOTSUPP);
2536 knote_insert(&ssb->ssb_kq.ki_note, kn);
2537 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2542 filt_sordetach(struct knote *kn)
2544 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2546 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2547 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2548 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2553 filt_soread(struct knote *kn, long hint)
2555 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2557 if (kn->kn_sfflags & NOTE_OOB) {
2558 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2559 kn->kn_fflags |= NOTE_OOB;
2564 kn->kn_data = so->so_rcv.ssb_cc;
2566 if (so->so_state & SS_CANTRCVMORE) {
2568 * Only set NODATA if all data has been exhausted.
2570 if (kn->kn_data == 0)
2571 kn->kn_flags |= EV_NODATA;
2572 kn->kn_flags |= EV_EOF;
2573 kn->kn_fflags = so->so_error;
2576 if (so->so_error) /* temporary udp error */
2578 if (kn->kn_sfflags & NOTE_LOWAT)
2579 return (kn->kn_data >= kn->kn_sdata);
2580 return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2581 !TAILQ_EMPTY(&so->so_comp));
2585 filt_sowdetach(struct knote *kn)
2587 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2589 knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2590 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2591 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2596 filt_sowrite(struct knote *kn, long hint)
2598 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2600 kn->kn_data = ssb_space(&so->so_snd);
2601 if (so->so_state & SS_CANTSENDMORE) {
2602 kn->kn_flags |= (EV_EOF | EV_NODATA);
2603 kn->kn_fflags = so->so_error;
2606 if (so->so_error) /* temporary udp error */
2608 if (((so->so_state & SS_ISCONNECTED) == 0) &&
2609 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2611 if (kn->kn_sfflags & NOTE_LOWAT)
2612 return (kn->kn_data >= kn->kn_sdata);
2613 return (kn->kn_data >= so->so_snd.ssb_lowat);
2618 filt_solisten(struct knote *kn, long hint)
2620 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2622 kn->kn_data = so->so_qlen;
2623 return (! TAILQ_EMPTY(&so->so_comp));