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);
424 soabort_async(sp, TRUE);
426 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
427 TAILQ_REMOVE(&so->so_comp, sp, so_list);
429 soclrstate(sp, SS_COMP);
430 soabort_async(sp, TRUE);
433 lwkt_relpooltoken(so);
435 if (so->so_state & SS_NOFDREF)
436 panic("soclose: NOFDREF");
437 sosetstate(so, SS_NOFDREF); /* take ref */
441 * Append the completed queue of head to head_inh (inherting listen socket).
444 soinherit(struct socket *head, struct socket *head_inh)
446 boolean_t do_wakeup = FALSE;
448 KASSERT(head->so_options & SO_ACCEPTCONN,
449 ("head does not accept connection"));
450 KASSERT(head_inh->so_options & SO_ACCEPTCONN,
451 ("head_inh does not accept connection"));
453 lwkt_getpooltoken(head);
454 lwkt_getpooltoken(head_inh);
456 if (head->so_qlen > 0)
459 while (!TAILQ_EMPTY(&head->so_comp)) {
460 struct ucred *old_cr;
463 sp = TAILQ_FIRST(&head->so_comp);
466 * Remove this socket from the current listen socket
469 TAILQ_REMOVE(&head->so_comp, sp, so_list);
472 /* Save the old ucred for later free. */
473 old_cr = sp->so_cred;
476 * Install this socket to the inheriting listen socket
479 sp->so_cred = crhold(head_inh->so_cred); /* non-blocking */
480 sp->so_head = head_inh;
482 TAILQ_INSERT_TAIL(&head_inh->so_comp, sp, so_list);
487 * crfree() may block and release the tokens temporarily.
488 * However, we are fine here, since the transition is done.
493 lwkt_relpooltoken(head_inh);
494 lwkt_relpooltoken(head);
498 * "New" connections have arrived
501 wakeup(&head_inh->so_timeo);
506 soclose_sync(struct socket *so, int fflag)
510 if (so->so_pcb == NULL)
512 if (so->so_state & SS_ISCONNECTED) {
513 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
514 error = sodisconnect(so);
518 if (so->so_options & SO_LINGER) {
519 if ((so->so_state & SS_ISDISCONNECTING) &&
522 while (so->so_state & SS_ISCONNECTED) {
523 error = tsleep(&so->so_timeo, PCATCH,
524 "soclos", so->so_linger * hz);
534 error2 = so_pru_detach(so);
535 if (error2 == EJUSTRETURN) {
537 * Protocol will call sodiscard()
538 * and sofree() for us.
547 so_pru_sync(so); /* unpend async sending */
548 sofree(so); /* dispose of ref */
554 soclose_sofree_async_handler(netmsg_t msg)
556 sofree(msg->base.nm_so);
560 soclose_sofree_async(struct socket *so)
562 struct netmsg_base *base = &so->so_clomsg;
564 netmsg_init(base, so, &netisr_apanic_rport, 0,
565 soclose_sofree_async_handler);
566 lwkt_sendmsg(so->so_port, &base->lmsg);
570 soclose_disconn_async_handler(netmsg_t msg)
572 struct socket *so = msg->base.nm_so;
574 if ((so->so_state & SS_ISCONNECTED) &&
575 (so->so_state & SS_ISDISCONNECTING) == 0)
576 so_pru_disconnect_direct(so);
581 error = so_pru_detach_direct(so);
582 if (error == EJUSTRETURN) {
584 * Protocol will call sodiscard()
585 * and sofree() for us.
596 soclose_disconn_async(struct socket *so)
598 struct netmsg_base *base = &so->so_clomsg;
600 netmsg_init(base, so, &netisr_apanic_rport, 0,
601 soclose_disconn_async_handler);
602 lwkt_sendmsg(so->so_port, &base->lmsg);
606 soclose_detach_async_handler(netmsg_t msg)
608 struct socket *so = msg->base.nm_so;
613 error = so_pru_detach_direct(so);
614 if (error == EJUSTRETURN) {
616 * Protocol will call sodiscard()
617 * and sofree() for us.
628 soclose_detach_async(struct socket *so)
630 struct netmsg_base *base = &so->so_clomsg;
632 netmsg_init(base, so, &netisr_apanic_rport, 0,
633 soclose_detach_async_handler);
634 lwkt_sendmsg(so->so_port, &base->lmsg);
638 soclose_fast(struct socket *so)
640 if (so->so_pcb == NULL)
643 if ((so->so_state & SS_ISCONNECTED) &&
644 (so->so_state & SS_ISDISCONNECTING) == 0) {
645 soclose_disconn_async(so);
650 soclose_detach_async(so);
656 soclose_sofree_async(so);
660 * Abort and destroy a socket. Only one abort can be in progress
661 * at any given moment.
664 soabort_async(struct socket *so, boolean_t clr_head)
667 * Keep a reference before clearing the so_head
668 * to avoid racing socket close in netisr.
673 so_pru_abort_async(so);
677 soabort_oncpu(struct socket *so)
680 so_pru_abort_direct(so);
684 * so is passed in ref'd, which becomes owned by
685 * the cleared SS_NOFDREF flag.
688 soaccept_generic(struct socket *so)
690 if ((so->so_state & SS_NOFDREF) == 0)
691 panic("soaccept: !NOFDREF");
692 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */
696 soaccept(struct socket *so, struct sockaddr **nam)
700 soaccept_generic(so);
701 error = so_pru_accept(so, nam);
706 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td,
711 if (so->so_options & SO_ACCEPTCONN)
714 * If protocol is connection-based, can only connect once.
715 * Otherwise, if connected, try to disconnect first.
716 * This allows user to disconnect by connecting to, e.g.,
719 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
720 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
721 (error = sodisconnect(so)))) {
725 * Prevent accumulated error from previous connection
729 if (!sync && so->so_proto->pr_usrreqs->pru_preconnect)
730 error = so_pru_connect_async(so, nam, td);
732 error = so_pru_connect(so, nam, td);
738 soconnect2(struct socket *so1, struct socket *so2)
742 error = so_pru_connect2(so1, so2);
747 sodisconnect(struct socket *so)
751 if ((so->so_state & SS_ISCONNECTED) == 0) {
755 if (so->so_state & SS_ISDISCONNECTING) {
759 error = so_pru_disconnect(so);
764 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
767 * If send must go all at once and message is larger than
768 * send buffering, then hard error.
769 * Lock against other senders.
770 * If must go all at once and not enough room now, then
771 * inform user that this would block and do nothing.
772 * Otherwise, if nonblocking, send as much as possible.
773 * The data to be sent is described by "uio" if nonzero,
774 * otherwise by the mbuf chain "top" (which must be null
775 * if uio is not). Data provided in mbuf chain must be small
776 * enough to send all at once.
778 * Returns nonzero on error, timeout or signal; callers
779 * must check for short counts if EINTR/ERESTART are returned.
780 * Data and control buffers are freed on return.
783 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
784 struct mbuf *top, struct mbuf *control, int flags,
791 int clen = 0, error, dontroute, mlen;
792 int atomic = sosendallatonce(so) || top;
796 resid = uio->uio_resid;
798 resid = (size_t)top->m_pkthdr.len;
801 for (m = top; m; m = m->m_next)
803 KKASSERT(top->m_pkthdr.len == len);
808 * WARNING! resid is unsigned, space and len are signed. space
809 * can wind up negative if the sockbuf is overcommitted.
811 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
812 * type sockets since that's an error.
814 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
820 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
821 (so->so_proto->pr_flags & PR_ATOMIC);
822 if (td->td_lwp != NULL)
823 td->td_lwp->lwp_ru.ru_msgsnd++;
825 clen = control->m_len;
826 #define gotoerr(errcode) { error = errcode; goto release; }
829 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
834 if (so->so_state & SS_CANTSENDMORE)
837 error = so->so_error;
841 if ((so->so_state & SS_ISCONNECTED) == 0) {
843 * `sendto' and `sendmsg' is allowed on a connection-
844 * based socket if it supports implied connect.
845 * Return ENOTCONN if not connected and no address is
848 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
849 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
850 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
851 !(resid == 0 && clen != 0))
853 } else if (addr == NULL)
854 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
855 ENOTCONN : EDESTADDRREQ);
857 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
858 clen > so->so_snd.ssb_hiwat) {
861 space = ssb_space(&so->so_snd);
864 if ((space < 0 || (size_t)space < resid + clen) && uio &&
865 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
866 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
867 gotoerr(EWOULDBLOCK);
868 ssb_unlock(&so->so_snd);
869 error = ssb_wait(&so->so_snd);
879 * Data is prepackaged in "top".
883 top->m_flags |= M_EOR;
887 m = m_getl((int)resid, M_WAITOK, MT_DATA,
888 top == NULL ? M_PKTHDR : 0, &mlen);
891 m->m_pkthdr.rcvif = NULL;
893 len = imin((int)szmin(mlen, resid), space);
894 if (resid < MINCLSIZE) {
896 * For datagram protocols, leave room
897 * for protocol headers in first mbuf.
899 if (atomic && top == NULL && len < mlen)
903 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
904 resid = uio->uio_resid;
907 top->m_pkthdr.len += len;
913 top->m_flags |= M_EOR;
916 } while (space > 0 && atomic);
918 so->so_options |= SO_DONTROUTE;
919 if (flags & MSG_OOB) {
920 pru_flags = PRUS_OOB;
921 } else if ((flags & MSG_EOF) &&
922 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
925 * If the user set MSG_EOF, the protocol
926 * understands this flag and nothing left to
927 * send then use PRU_SEND_EOF instead of PRU_SEND.
929 pru_flags = PRUS_EOF;
930 } else if (resid > 0 && space > 0) {
931 /* If there is more to send, set PRUS_MORETOCOME */
932 pru_flags = PRUS_MORETOCOME;
937 * XXX all the SS_CANTSENDMORE checks previously
938 * done could be out of date. We could have recieved
939 * a reset packet in an interrupt or maybe we slept
940 * while doing page faults in uiomove() etc. We could
941 * probably recheck again inside the splnet() protection
942 * here, but there are probably other places that this
943 * also happens. We must rethink this.
945 error = so_pru_send(so, pru_flags, top, addr, control, td);
947 so->so_options &= ~SO_DONTROUTE;
954 } while (resid && space > 0);
958 ssb_unlock(&so->so_snd);
969 * A specialization of sosend() for UDP based on protocol-specific knowledge:
970 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
971 * sosendallatonce() returns true,
972 * the "atomic" variable is true,
973 * and sosendudp() blocks until space is available for the entire send.
974 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
975 * PR_IMPLOPCL flags set.
976 * UDP has no out-of-band data.
977 * UDP has no control data.
978 * UDP does not support MSG_EOR.
981 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
982 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
985 int error, pru_flags = 0;
988 if (td->td_lwp != NULL)
989 td->td_lwp->lwp_ru.ru_msgsnd++;
993 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
994 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
997 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1001 if (so->so_state & SS_CANTSENDMORE)
1004 error = so->so_error;
1008 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
1009 gotoerr(EDESTADDRREQ);
1010 if (resid > so->so_snd.ssb_hiwat)
1012 space = ssb_space(&so->so_snd);
1013 if (uio && (space < 0 || (size_t)space < resid)) {
1014 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1015 gotoerr(EWOULDBLOCK);
1016 ssb_unlock(&so->so_snd);
1017 error = ssb_wait(&so->so_snd);
1024 int hdrlen = max_hdr;
1027 * We try to optimize out the additional mbuf
1028 * allocations in M_PREPEND() on output path, e.g.
1029 * - udp_output(), when it tries to prepend protocol
1031 * - Link layer output function, when it tries to
1032 * prepend link layer header.
1034 * This probably will not benefit any data that will
1035 * be fragmented, so this optimization is only performed
1036 * when the size of data and max size of protocol+link
1037 * headers fit into one mbuf cluster.
1039 if (uio->uio_resid > MCLBYTES - hdrlen ||
1040 !udp_sosend_prepend) {
1041 top = m_uiomove(uio);
1047 top = m_getl(uio->uio_resid + hdrlen, M_WAITOK,
1048 MT_DATA, M_PKTHDR, &nsize);
1049 KASSERT(nsize >= uio->uio_resid + hdrlen,
1050 ("sosendudp invalid nsize %d, "
1051 "resid %zu, hdrlen %d",
1052 nsize, uio->uio_resid, hdrlen));
1054 top->m_len = uio->uio_resid;
1055 top->m_pkthdr.len = uio->uio_resid;
1056 top->m_data += hdrlen;
1058 error = uiomove(mtod(top, caddr_t), top->m_len, uio);
1064 if (flags & MSG_DONTROUTE)
1065 pru_flags |= PRUS_DONTROUTE;
1067 if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
1068 so_pru_send_async(so, pru_flags, top, addr, NULL, td);
1071 error = so_pru_send(so, pru_flags, top, addr, NULL, td);
1073 top = NULL; /* sent or freed in lower layer */
1076 ssb_unlock(&so->so_snd);
1084 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1085 struct mbuf *top, struct mbuf *control, int flags,
1097 KKASSERT(top == NULL);
1099 resid = uio->uio_resid;
1102 resid = (size_t)top->m_pkthdr.len;
1105 for (m = top; m; m = m->m_next)
1107 KKASSERT(top->m_pkthdr.len == len);
1112 * WARNING! resid is unsigned, space and len are signed. space
1113 * can wind up negative if the sockbuf is overcommitted.
1115 * Also check to make sure that MSG_EOR isn't used on TCP
1117 if (flags & MSG_EOR) {
1123 /* TCP doesn't do control messages (rights, creds, etc) */
1124 if (control->m_len) {
1128 m_freem(control); /* empty control, just free it */
1132 if (td->td_lwp != NULL)
1133 td->td_lwp->lwp_ru.ru_msgsnd++;
1135 #define gotoerr(errcode) { error = errcode; goto release; }
1138 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1143 if (so->so_state & SS_CANTSENDMORE)
1146 error = so->so_error;
1150 if ((so->so_state & SS_ISCONNECTED) == 0 &&
1151 (so->so_state & SS_ISCONFIRMING) == 0)
1153 if (allatonce && resid > so->so_snd.ssb_hiwat)
1156 space = ssb_space_prealloc(&so->so_snd);
1157 if (flags & MSG_OOB)
1159 if ((space < 0 || (size_t)space < resid) && !allatonce &&
1160 space < so->so_snd.ssb_lowat) {
1161 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1162 gotoerr(EWOULDBLOCK);
1163 ssb_unlock(&so->so_snd);
1164 error = ssb_wait(&so->so_snd);
1171 int cnt = 0, async = 0;
1175 * Data is prepackaged in "top".
1179 if (resid > INT_MAX)
1181 if (tcp_sosend_jcluster) {
1182 m = m_getlj((int)resid, M_WAITOK, MT_DATA,
1183 top == NULL ? M_PKTHDR : 0, &mlen);
1185 m = m_getl((int)resid, M_WAITOK, MT_DATA,
1186 top == NULL ? M_PKTHDR : 0, &mlen);
1189 m->m_pkthdr.len = 0;
1190 m->m_pkthdr.rcvif = NULL;
1192 len = imin((int)szmin(mlen, resid), space);
1194 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1195 resid = uio->uio_resid;
1198 top->m_pkthdr.len += len;
1205 } while (space > 0 && cnt < tcp_sosend_agglim);
1207 if (tcp_sosend_async)
1210 if (flags & MSG_OOB) {
1211 pru_flags = PRUS_OOB;
1213 } else if ((flags & MSG_EOF) && resid == 0) {
1214 pru_flags = PRUS_EOF;
1215 } else if (resid > 0 && space > 0) {
1216 /* If there is more to send, set PRUS_MORETOCOME */
1217 pru_flags = PRUS_MORETOCOME;
1223 if (flags & MSG_SYNC)
1227 * XXX all the SS_CANTSENDMORE checks previously
1228 * done could be out of date. We could have recieved
1229 * a reset packet in an interrupt or maybe we slept
1230 * while doing page faults in uiomove() etc. We could
1231 * probably recheck again inside the splnet() protection
1232 * here, but there are probably other places that this
1233 * also happens. We must rethink this.
1235 for (m = top; m; m = m->m_next)
1236 ssb_preallocstream(&so->so_snd, m);
1238 error = so_pru_send(so, pru_flags, top,
1241 so_pru_send_async(so, pru_flags, top,
1250 } while (resid && space > 0);
1254 ssb_unlock(&so->so_snd);
1265 * Implement receive operations on a socket.
1267 * We depend on the way that records are added to the signalsockbuf
1268 * by sbappend*. In particular, each record (mbufs linked through m_next)
1269 * must begin with an address if the protocol so specifies,
1270 * followed by an optional mbuf or mbufs containing ancillary data,
1271 * and then zero or more mbufs of data.
1273 * Although the signalsockbuf is locked, new data may still be appended.
1274 * A token inside the ssb_lock deals with MP issues and still allows
1275 * the network to access the socket if we block in a uio.
1277 * The caller may receive the data as a single mbuf chain by supplying
1278 * an mbuf **mp0 for use in returning the chain. The uio is then used
1279 * only for the count in uio_resid.
1282 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1283 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1286 struct mbuf *free_chain = NULL;
1287 int flags, len, error, offset;
1288 struct protosw *pr = so->so_proto;
1290 size_t resid, orig_resid;
1293 resid = uio->uio_resid;
1295 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1303 flags = *flagsp &~ MSG_EOR;
1306 if (flags & MSG_OOB) {
1307 m = m_get(M_WAITOK, MT_DATA);
1310 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1316 KKASSERT(resid >= (size_t)m->m_len);
1317 resid -= (size_t)m->m_len;
1318 } while (resid > 0 && m);
1321 uio->uio_resid = resid;
1322 error = uiomove(mtod(m, caddr_t),
1323 (int)szmin(resid, m->m_len),
1325 resid = uio->uio_resid;
1327 } while (uio->uio_resid && error == 0 && m);
1334 if ((so->so_state & SS_ISCONFIRMING) && resid)
1338 * The token interlocks against the protocol thread while
1339 * ssb_lock is a blocking lock against other userland entities.
1341 lwkt_gettoken(&so->so_rcv.ssb_token);
1343 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1347 m = so->so_rcv.ssb_mb;
1349 * If we have less data than requested, block awaiting more
1350 * (subject to any timeout) if:
1351 * 1. the current count is less than the low water mark, or
1352 * 2. MSG_WAITALL is set, and it is possible to do the entire
1353 * receive operation at once if we block (resid <= hiwat).
1354 * 3. MSG_DONTWAIT is not set
1355 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1356 * we have to do the receive in sections, and thus risk returning
1357 * a short count if a timeout or signal occurs after we start.
1359 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1360 (size_t)so->so_rcv.ssb_cc < resid) &&
1361 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1362 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1363 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1364 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1368 error = so->so_error;
1369 if ((flags & MSG_PEEK) == 0)
1373 if (so->so_state & SS_CANTRCVMORE) {
1379 for (; m; m = m->m_next) {
1380 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1381 m = so->so_rcv.ssb_mb;
1385 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1386 (pr->pr_flags & PR_CONNREQUIRED)) {
1392 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1393 error = EWOULDBLOCK;
1396 ssb_unlock(&so->so_rcv);
1397 error = ssb_wait(&so->so_rcv);
1403 if (uio && uio->uio_td && uio->uio_td->td_proc)
1404 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1407 * note: m should be == sb_mb here. Cache the next record while
1408 * cleaning up. Note that calling m_free*() will break out critical
1411 KKASSERT(m == so->so_rcv.ssb_mb);
1414 * Skip any address mbufs prepending the record.
1416 if (pr->pr_flags & PR_ADDR) {
1417 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1420 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1421 if (flags & MSG_PEEK)
1424 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1428 * Skip any control mbufs prepending the record.
1430 while (m && m->m_type == MT_CONTROL && error == 0) {
1431 if (flags & MSG_PEEK) {
1433 *controlp = m_copy(m, 0, m->m_len);
1434 m = m->m_next; /* XXX race */
1437 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1438 if (pr->pr_domain->dom_externalize &&
1439 mtod(m, struct cmsghdr *)->cmsg_type ==
1441 error = (*pr->pr_domain->dom_externalize)(m);
1445 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1448 if (controlp && *controlp) {
1450 controlp = &(*controlp)->m_next;
1459 if (type == MT_OOBDATA)
1464 * Copy to the UIO or mbuf return chain (*mp).
1468 while (m && resid > 0 && error == 0) {
1469 if (m->m_type == MT_OOBDATA) {
1470 if (type != MT_OOBDATA)
1472 } else if (type == MT_OOBDATA)
1475 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1477 soclrstate(so, SS_RCVATMARK);
1478 len = (resid > INT_MAX) ? INT_MAX : resid;
1479 if (so->so_oobmark && len > so->so_oobmark - offset)
1480 len = so->so_oobmark - offset;
1481 if (len > m->m_len - moff)
1482 len = m->m_len - moff;
1485 * Copy out to the UIO or pass the mbufs back to the SIO.
1486 * The SIO is dealt with when we eat the mbuf, but deal
1487 * with the resid here either way.
1490 uio->uio_resid = resid;
1491 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1492 resid = uio->uio_resid;
1496 resid -= (size_t)len;
1500 * Eat the entire mbuf or just a piece of it
1502 if (len == m->m_len - moff) {
1503 if (m->m_flags & M_EOR)
1505 if (flags & MSG_PEEK) {
1510 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1514 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1518 if (flags & MSG_PEEK) {
1522 n = m_copym(m, 0, len, M_WAITOK);
1528 so->so_rcv.ssb_cc -= len;
1531 if (so->so_oobmark) {
1532 if ((flags & MSG_PEEK) == 0) {
1533 so->so_oobmark -= len;
1534 if (so->so_oobmark == 0) {
1535 sosetstate(so, SS_RCVATMARK);
1540 if (offset == so->so_oobmark)
1544 if (flags & MSG_EOR)
1547 * If the MSG_WAITALL flag is set (for non-atomic socket),
1548 * we must not quit until resid == 0 or an error
1549 * termination. If a signal/timeout occurs, return
1550 * with a short count but without error.
1551 * Keep signalsockbuf locked against other readers.
1553 while ((flags & MSG_WAITALL) && m == NULL &&
1554 resid > 0 && !sosendallatonce(so) &&
1555 so->so_rcv.ssb_mb == NULL) {
1556 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1559 * The window might have closed to zero, make
1560 * sure we send an ack now that we've drained
1561 * the buffer or we might end up blocking until
1562 * the idle takes over (5 seconds).
1564 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1565 so_pru_rcvd(so, flags);
1566 error = ssb_wait(&so->so_rcv);
1568 ssb_unlock(&so->so_rcv);
1572 m = so->so_rcv.ssb_mb;
1577 * If an atomic read was requested but unread data still remains
1578 * in the record, set MSG_TRUNC.
1580 if (m && pr->pr_flags & PR_ATOMIC)
1584 * Cleanup. If an atomic read was requested drop any unread data.
1586 if ((flags & MSG_PEEK) == 0) {
1587 if (m && (pr->pr_flags & PR_ATOMIC))
1588 sbdroprecord(&so->so_rcv.sb);
1589 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1590 so_pru_rcvd(so, flags);
1593 if (orig_resid == resid && orig_resid &&
1594 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1595 ssb_unlock(&so->so_rcv);
1602 ssb_unlock(&so->so_rcv);
1604 lwkt_reltoken(&so->so_rcv.ssb_token);
1606 m_freem(free_chain);
1611 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1612 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1615 struct mbuf *free_chain = NULL;
1616 int flags, len, error, offset;
1617 struct protosw *pr = so->so_proto;
1620 size_t resid, orig_resid, restmp;
1623 resid = uio->uio_resid;
1625 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1633 flags = *flagsp &~ MSG_EOR;
1636 if (flags & MSG_OOB) {
1637 m = m_get(M_WAITOK, MT_DATA);
1640 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1646 KKASSERT(resid >= (size_t)m->m_len);
1647 resid -= (size_t)m->m_len;
1648 } while (resid > 0 && m);
1651 uio->uio_resid = resid;
1652 error = uiomove(mtod(m, caddr_t),
1653 (int)szmin(resid, m->m_len),
1655 resid = uio->uio_resid;
1657 } while (uio->uio_resid && error == 0 && m);
1666 * The token interlocks against the protocol thread while
1667 * ssb_lock is a blocking lock against other userland entities.
1669 * Lock a limited number of mbufs (not all, so sbcompress() still
1670 * works well). The token is used as an interlock for sbwait() so
1671 * release it afterwords.
1674 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1678 lwkt_gettoken(&so->so_rcv.ssb_token);
1679 m = so->so_rcv.ssb_mb;
1682 * If we have less data than requested, block awaiting more
1683 * (subject to any timeout) if:
1684 * 1. the current count is less than the low water mark, or
1685 * 2. MSG_WAITALL is set, and it is possible to do the entire
1686 * receive operation at once if we block (resid <= hiwat).
1687 * 3. MSG_DONTWAIT is not set
1688 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1689 * we have to do the receive in sections, and thus risk returning
1690 * a short count if a timeout or signal occurs after we start.
1692 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1693 (size_t)so->so_rcv.ssb_cc < resid) &&
1694 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1695 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1696 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1700 lwkt_reltoken(&so->so_rcv.ssb_token);
1701 error = so->so_error;
1702 if ((flags & MSG_PEEK) == 0)
1706 if (so->so_state & SS_CANTRCVMORE) {
1709 lwkt_reltoken(&so->so_rcv.ssb_token);
1712 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1713 (pr->pr_flags & PR_CONNREQUIRED)) {
1714 lwkt_reltoken(&so->so_rcv.ssb_token);
1719 lwkt_reltoken(&so->so_rcv.ssb_token);
1722 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1723 lwkt_reltoken(&so->so_rcv.ssb_token);
1724 error = EWOULDBLOCK;
1727 ssb_unlock(&so->so_rcv);
1728 error = ssb_wait(&so->so_rcv);
1729 lwkt_reltoken(&so->so_rcv.ssb_token);
1741 while (n && restmp < resid) {
1742 n->m_flags |= M_SOLOCKED;
1744 if (n->m_next == NULL)
1751 * Release token for loop
1753 lwkt_reltoken(&so->so_rcv.ssb_token);
1754 if (uio && uio->uio_td && uio->uio_td->td_proc)
1755 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1758 * note: m should be == sb_mb here. Cache the next record while
1759 * cleaning up. Note that calling m_free*() will break out critical
1762 KKASSERT(m == so->so_rcv.ssb_mb);
1765 * Copy to the UIO or mbuf return chain (*mp).
1767 * NOTE: Token is not held for loop
1773 while (m && (m->m_flags & M_SOLOCKED) && resid > 0 && error == 0) {
1774 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1777 soclrstate(so, SS_RCVATMARK);
1778 len = (resid > INT_MAX) ? INT_MAX : resid;
1779 if (so->so_oobmark && len > so->so_oobmark - offset)
1780 len = so->so_oobmark - offset;
1781 if (len > m->m_len - moff)
1782 len = m->m_len - moff;
1785 * Copy out to the UIO or pass the mbufs back to the SIO.
1786 * The SIO is dealt with when we eat the mbuf, but deal
1787 * with the resid here either way.
1790 uio->uio_resid = resid;
1791 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1792 resid = uio->uio_resid;
1796 resid -= (size_t)len;
1800 * Eat the entire mbuf or just a piece of it
1803 if (len == m->m_len - moff) {
1813 if (so->so_oobmark && offset == so->so_oobmark) {
1820 * Synchronize sockbuf with data we read.
1822 * NOTE: (m) is junk on entry (it could be left over from the
1825 if ((flags & MSG_PEEK) == 0) {
1826 lwkt_gettoken(&so->so_rcv.ssb_token);
1827 m = so->so_rcv.ssb_mb;
1828 while (m && offset >= m->m_len) {
1829 if (so->so_oobmark) {
1830 so->so_oobmark -= m->m_len;
1831 if (so->so_oobmark == 0) {
1832 sosetstate(so, SS_RCVATMARK);
1838 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1842 m = sbunlinkmbuf(&so->so_rcv.sb,
1849 n = m_copym(m, 0, offset, M_WAITOK);
1853 m->m_data += offset;
1855 so->so_rcv.ssb_cc -= offset;
1856 if (so->so_oobmark) {
1857 so->so_oobmark -= offset;
1858 if (so->so_oobmark == 0) {
1859 sosetstate(so, SS_RCVATMARK);
1865 lwkt_reltoken(&so->so_rcv.ssb_token);
1869 * If the MSG_WAITALL flag is set (for non-atomic socket),
1870 * we must not quit until resid == 0 or an error termination.
1872 * If a signal/timeout occurs, return with a short count but without
1875 * Keep signalsockbuf locked against other readers.
1877 * XXX if MSG_PEEK we currently do quit.
1879 if ((flags & MSG_WAITALL) && !(flags & MSG_PEEK) &&
1880 didoob == 0 && resid > 0 &&
1881 !sosendallatonce(so)) {
1882 lwkt_gettoken(&so->so_rcv.ssb_token);
1884 while ((m = so->so_rcv.ssb_mb) == NULL) {
1885 if (so->so_error || (so->so_state & SS_CANTRCVMORE)) {
1886 error = so->so_error;
1890 * The window might have closed to zero, make
1891 * sure we send an ack now that we've drained
1892 * the buffer or we might end up blocking until
1893 * the idle takes over (5 seconds).
1896 so_pru_rcvd_async(so);
1897 if (so->so_rcv.ssb_mb == NULL)
1898 error = ssb_wait(&so->so_rcv);
1900 lwkt_reltoken(&so->so_rcv.ssb_token);
1901 ssb_unlock(&so->so_rcv);
1906 if (m && error == 0)
1908 lwkt_reltoken(&so->so_rcv.ssb_token);
1912 * Token not held here.
1914 * Cleanup. If an atomic read was requested drop any unread data XXX
1916 if ((flags & MSG_PEEK) == 0) {
1918 so_pru_rcvd_async(so);
1921 if (orig_resid == resid && orig_resid &&
1922 (so->so_state & SS_CANTRCVMORE) == 0) {
1923 ssb_unlock(&so->so_rcv);
1930 ssb_unlock(&so->so_rcv);
1933 m_freem(free_chain);
1938 * Shut a socket down. Note that we do not get a frontend lock as we
1939 * want to be able to shut the socket down even if another thread is
1940 * blocked in a read(), thus waking it up.
1943 soshutdown(struct socket *so, int how)
1945 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1948 if (how != SHUT_WR) {
1949 /*ssb_lock(&so->so_rcv, M_WAITOK);*/
1951 /*ssb_unlock(&so->so_rcv);*/
1954 return (so_pru_shutdown(so));
1959 sorflush(struct socket *so)
1961 struct signalsockbuf *ssb = &so->so_rcv;
1962 struct protosw *pr = so->so_proto;
1963 struct signalsockbuf asb;
1965 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1967 lwkt_gettoken(&ssb->ssb_token);
1972 * Can't just blow up the ssb structure here
1974 bzero(&ssb->sb, sizeof(ssb->sb));
1979 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1981 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1982 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1983 ssb_release(&asb, so);
1985 lwkt_reltoken(&ssb->ssb_token);
1990 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1992 struct accept_filter_arg *afap = NULL;
1993 struct accept_filter *afp;
1994 struct so_accf *af = so->so_accf;
1997 /* do not set/remove accept filters on non listen sockets */
1998 if ((so->so_options & SO_ACCEPTCONN) == 0) {
2003 /* removing the filter */
2006 if (af->so_accept_filter != NULL &&
2007 af->so_accept_filter->accf_destroy != NULL) {
2008 af->so_accept_filter->accf_destroy(so);
2010 if (af->so_accept_filter_str != NULL) {
2011 kfree(af->so_accept_filter_str, M_ACCF);
2016 so->so_options &= ~SO_ACCEPTFILTER;
2019 /* adding a filter */
2020 /* must remove previous filter first */
2025 /* don't put large objects on the kernel stack */
2026 afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
2027 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
2028 afap->af_name[sizeof(afap->af_name)-1] = '\0';
2029 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
2032 afp = accept_filt_get(afap->af_name);
2037 af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
2038 if (afp->accf_create != NULL) {
2039 if (afap->af_name[0] != '\0') {
2040 int len = strlen(afap->af_name) + 1;
2042 af->so_accept_filter_str = kmalloc(len, M_ACCF,
2044 strcpy(af->so_accept_filter_str, afap->af_name);
2046 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
2047 if (af->so_accept_filter_arg == NULL) {
2048 kfree(af->so_accept_filter_str, M_ACCF);
2055 af->so_accept_filter = afp;
2057 so->so_options |= SO_ACCEPTFILTER;
2060 kfree(afap, M_TEMP);
2066 * Perhaps this routine, and sooptcopyout(), below, ought to come in
2067 * an additional variant to handle the case where the option value needs
2068 * to be some kind of integer, but not a specific size.
2069 * In addition to their use here, these functions are also called by the
2070 * protocol-level pr_ctloutput() routines.
2073 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2075 return soopt_to_kbuf(sopt, buf, len, minlen);
2079 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2083 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2084 KKASSERT(kva_p(buf));
2087 * If the user gives us more than we wanted, we ignore it,
2088 * but if we don't get the minimum length the caller
2089 * wants, we return EINVAL. On success, sopt->sopt_valsize
2090 * is set to however much we actually retrieved.
2092 if ((valsize = sopt->sopt_valsize) < minlen)
2095 sopt->sopt_valsize = valsize = len;
2097 bcopy(sopt->sopt_val, buf, valsize);
2103 sosetopt(struct socket *so, struct sockopt *sopt)
2109 struct signalsockbuf *sotmp;
2112 sopt->sopt_dir = SOPT_SET;
2113 if (sopt->sopt_level != SOL_SOCKET) {
2114 if (so->so_proto && so->so_proto->pr_ctloutput) {
2115 return (so_pr_ctloutput(so, sopt));
2117 error = ENOPROTOOPT;
2119 switch (sopt->sopt_name) {
2121 case SO_ACCEPTFILTER:
2122 error = do_setopt_accept_filter(so, sopt);
2128 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2132 so->so_linger = l.l_linger;
2134 so->so_options |= SO_LINGER;
2136 so->so_options &= ~SO_LINGER;
2142 case SO_USELOOPBACK:
2149 error = sooptcopyin(sopt, &optval, sizeof optval,
2154 so->so_options |= sopt->sopt_name;
2156 so->so_options &= ~sopt->sopt_name;
2163 error = sooptcopyin(sopt, &optval, sizeof optval,
2169 * Values < 1 make no sense for any of these
2170 * options, so disallow them.
2177 switch (sopt->sopt_name) {
2180 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2181 &so->so_snd : &so->so_rcv, (u_long)optval,
2183 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2187 sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2188 &so->so_snd : &so->so_rcv;
2189 atomic_clear_int(&sotmp->ssb_flags,
2194 * Make sure the low-water is never greater than
2198 so->so_snd.ssb_lowat =
2199 (optval > so->so_snd.ssb_hiwat) ?
2200 so->so_snd.ssb_hiwat : optval;
2201 atomic_clear_int(&so->so_snd.ssb_flags,
2205 so->so_rcv.ssb_lowat =
2206 (optval > so->so_rcv.ssb_hiwat) ?
2207 so->so_rcv.ssb_hiwat : optval;
2208 atomic_clear_int(&so->so_rcv.ssb_flags,
2216 error = sooptcopyin(sopt, &tv, sizeof tv,
2221 /* assert(hz > 0); */
2222 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2223 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2227 /* assert(tick > 0); */
2228 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2229 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2230 if (val > INT_MAX) {
2234 if (val == 0 && tv.tv_usec != 0)
2237 switch (sopt->sopt_name) {
2239 so->so_snd.ssb_timeo = val;
2242 so->so_rcv.ssb_timeo = val;
2247 error = ENOPROTOOPT;
2250 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2251 (void) so_pr_ctloutput(so, sopt);
2258 /* Helper routine for getsockopt */
2260 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2262 soopt_from_kbuf(sopt, buf, len);
2267 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2272 sopt->sopt_valsize = 0;
2276 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2277 KKASSERT(kva_p(buf));
2280 * Documented get behavior is that we always return a value,
2281 * possibly truncated to fit in the user's buffer.
2282 * Traditional behavior is that we always tell the user
2283 * precisely how much we copied, rather than something useful
2284 * like the total amount we had available for her.
2285 * Note that this interface is not idempotent; the entire answer must
2286 * generated ahead of time.
2288 valsize = szmin(len, sopt->sopt_valsize);
2289 sopt->sopt_valsize = valsize;
2290 if (sopt->sopt_val != 0) {
2291 bcopy(buf, sopt->sopt_val, valsize);
2296 sogetopt(struct socket *so, struct sockopt *sopt)
2303 struct accept_filter_arg *afap;
2307 sopt->sopt_dir = SOPT_GET;
2308 if (sopt->sopt_level != SOL_SOCKET) {
2309 if (so->so_proto && so->so_proto->pr_ctloutput) {
2310 return (so_pr_ctloutput(so, sopt));
2312 return (ENOPROTOOPT);
2314 switch (sopt->sopt_name) {
2316 case SO_ACCEPTFILTER:
2317 if ((so->so_options & SO_ACCEPTCONN) == 0)
2319 afap = kmalloc(sizeof(*afap), M_TEMP,
2321 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2322 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2323 if (so->so_accf->so_accept_filter_str != NULL)
2324 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2326 error = sooptcopyout(sopt, afap, sizeof(*afap));
2327 kfree(afap, M_TEMP);
2332 l.l_onoff = so->so_options & SO_LINGER;
2333 l.l_linger = so->so_linger;
2334 error = sooptcopyout(sopt, &l, sizeof l);
2337 case SO_USELOOPBACK:
2347 optval = so->so_options & sopt->sopt_name;
2349 error = sooptcopyout(sopt, &optval, sizeof optval);
2353 optval = so->so_type;
2357 optval = so->so_error;
2362 optval = so->so_snd.ssb_hiwat;
2366 optval = so->so_rcv.ssb_hiwat;
2370 optval = so->so_snd.ssb_lowat;
2374 optval = so->so_rcv.ssb_lowat;
2379 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2380 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2382 tv.tv_sec = optval / hz;
2383 tv.tv_usec = (optval % hz) * ustick;
2384 error = sooptcopyout(sopt, &tv, sizeof tv);
2388 optval_l = ssb_space(&so->so_snd);
2389 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2393 optval = -1; /* no hint */
2397 error = ENOPROTOOPT;
2400 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput)
2401 so_pr_ctloutput(so, sopt);
2406 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2408 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2410 struct mbuf *m, *m_prev;
2411 int sopt_size = sopt->sopt_valsize, msize;
2413 m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA,
2417 m->m_len = min(msize, sopt_size);
2418 sopt_size -= m->m_len;
2422 while (sopt_size > 0) {
2423 m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT,
2424 MT_DATA, 0, &msize);
2429 m->m_len = min(msize, sopt_size);
2430 sopt_size -= m->m_len;
2437 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2439 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2441 soopt_to_mbuf(sopt, m);
2446 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2451 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2453 if (sopt->sopt_val == NULL)
2455 val = sopt->sopt_val;
2456 valsize = sopt->sopt_valsize;
2457 while (m != NULL && valsize >= m->m_len) {
2458 bcopy(val, mtod(m, char *), m->m_len);
2459 valsize -= m->m_len;
2460 val = (caddr_t)val + m->m_len;
2463 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2464 panic("ip6_sooptmcopyin");
2467 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2469 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2471 return soopt_from_mbuf(sopt, m);
2475 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2477 struct mbuf *m0 = m;
2482 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2484 if (sopt->sopt_val == NULL)
2486 val = sopt->sopt_val;
2487 maxsize = sopt->sopt_valsize;
2488 while (m != NULL && maxsize >= m->m_len) {
2489 bcopy(mtod(m, char *), val, m->m_len);
2490 maxsize -= m->m_len;
2491 val = (caddr_t)val + m->m_len;
2492 valsize += m->m_len;
2496 /* enough soopt buffer should be given from user-land */
2500 sopt->sopt_valsize = valsize;
2505 sohasoutofband(struct socket *so)
2507 if (so->so_sigio != NULL)
2508 pgsigio(so->so_sigio, SIGURG, 0);
2509 KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
2513 sokqfilter(struct file *fp, struct knote *kn)
2515 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2516 struct signalsockbuf *ssb;
2518 switch (kn->kn_filter) {
2520 if (so->so_options & SO_ACCEPTCONN)
2521 kn->kn_fop = &solisten_filtops;
2523 kn->kn_fop = &soread_filtops;
2527 kn->kn_fop = &sowrite_filtops;
2531 kn->kn_fop = &soexcept_filtops;
2535 return (EOPNOTSUPP);
2538 knote_insert(&ssb->ssb_kq.ki_note, kn);
2539 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2544 filt_sordetach(struct knote *kn)
2546 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2548 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2549 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2550 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2555 filt_soread(struct knote *kn, long hint)
2557 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2559 if (kn->kn_sfflags & NOTE_OOB) {
2560 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2561 kn->kn_fflags |= NOTE_OOB;
2566 kn->kn_data = so->so_rcv.ssb_cc;
2568 if (so->so_state & SS_CANTRCVMORE) {
2570 * Only set NODATA if all data has been exhausted.
2572 if (kn->kn_data == 0)
2573 kn->kn_flags |= EV_NODATA;
2574 kn->kn_flags |= EV_EOF;
2575 kn->kn_fflags = so->so_error;
2578 if (so->so_error) /* temporary udp error */
2580 if (kn->kn_sfflags & NOTE_LOWAT)
2581 return (kn->kn_data >= kn->kn_sdata);
2582 return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2583 !TAILQ_EMPTY(&so->so_comp));
2587 filt_sowdetach(struct knote *kn)
2589 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2591 knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2592 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2593 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2598 filt_sowrite(struct knote *kn, long hint)
2600 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2602 kn->kn_data = ssb_space(&so->so_snd);
2603 if (so->so_state & SS_CANTSENDMORE) {
2604 kn->kn_flags |= (EV_EOF | EV_NODATA);
2605 kn->kn_fflags = so->so_error;
2608 if (so->so_error) /* temporary udp error */
2610 if (((so->so_state & SS_ISCONNECTED) == 0) &&
2611 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2613 if (kn->kn_sfflags & NOTE_LOWAT)
2614 return (kn->kn_data >= kn->kn_sdata);
2615 return (kn->kn_data >= so->so_snd.ssb_lowat);
2620 filt_solisten(struct knote *kn, long hint)
2622 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2624 kn->kn_data = so->so_qlen;
2625 return (! TAILQ_EMPTY(&so->so_comp));