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");
151 static int use_socreate_fast = 1;
152 SYSCTL_INT(_kern_ipc, OID_AUTO, socreate_fast, CTLFLAG_RW,
153 &use_socreate_fast, 0, "Fast socket creation");
156 * Socket operation routines.
157 * These routines are called by the routines in
158 * sys_socket.c or from a system process, and
159 * implement the semantics of socket operations by
160 * switching out to the protocol specific routines.
164 * Get a socket structure, and initialize it.
165 * Note that it would probably be better to allocate socket
166 * and PCB at the same time, but I'm not convinced that all
167 * the protocols can be easily modified to do this.
170 soalloc(int waitok, struct protosw *pr)
175 waitmask = waitok ? M_WAITOK : M_NOWAIT;
176 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
178 /* XXX race condition for reentrant kernel */
180 TAILQ_INIT(&so->so_aiojobq);
181 TAILQ_INIT(&so->so_rcv.ssb_mlist);
182 TAILQ_INIT(&so->so_snd.ssb_mlist);
183 lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
184 lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
185 spin_init(&so->so_rcvd_spin, "soalloc");
186 netmsg_init(&so->so_rcvd_msg.base, so, &netisr_adone_rport,
187 MSGF_DROPABLE | MSGF_PRIORITY,
188 so->so_proto->pr_usrreqs->pru_rcvd);
189 so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC;
190 so->so_state = SS_NOFDREF;
197 socreate(int dom, struct socket **aso, int type,
198 int proto, struct thread *td)
200 struct proc *p = td->td_proc;
203 struct pru_attach_info ai;
207 prp = pffindproto(dom, proto, type);
209 prp = pffindtype(dom, type);
211 if (prp == NULL || prp->pr_usrreqs->pru_attach == 0)
212 return (EPROTONOSUPPORT);
214 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
215 prp->pr_domain->dom_family != PF_LOCAL &&
216 prp->pr_domain->dom_family != PF_INET &&
217 prp->pr_domain->dom_family != PF_INET6 &&
218 prp->pr_domain->dom_family != PF_ROUTE) {
219 return (EPROTONOSUPPORT);
222 if (prp->pr_type != type)
224 so = soalloc(p != NULL, prp);
229 * Callers of socreate() presumably will connect up a descriptor
230 * and call soclose() if they cannot. This represents our so_refs
231 * (which should be 1) from soalloc().
233 soclrstate(so, SS_NOFDREF);
236 * Set a default port for protocol processing. No action will occur
237 * on the socket on this port until an inpcb is attached to it and
238 * is able to match incoming packets, or until the socket becomes
239 * available to userland.
241 * We normally default the socket to the protocol thread on cpu 0,
242 * if protocol does not provide its own method to initialize the
245 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
246 * thread and all pr_*()/pru_*() calls are executed synchronously.
248 if (prp->pr_flags & PR_SYNC_PORT)
249 so->so_port = &netisr_sync_port;
250 else if (prp->pr_initport != NULL)
251 so->so_port = prp->pr_initport();
253 so->so_port = netisr_cpuport(0);
255 TAILQ_INIT(&so->so_incomp);
256 TAILQ_INIT(&so->so_comp);
258 so->so_cred = crhold(p->p_ucred);
259 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
260 ai.p_ucred = p->p_ucred;
261 ai.fd_rdir = p->p_fd->fd_rdir;
264 * Auto-sizing of socket buffers is managed by the protocols and
265 * the appropriate flags must be set in the pru_attach function.
267 if (use_socreate_fast && prp->pr_usrreqs->pru_preattach)
268 error = so_pru_attach_fast(so, proto, &ai);
270 error = so_pru_attach(so, proto, &ai);
272 sosetstate(so, SS_NOFDREF);
273 sofree(so); /* from soalloc */
278 * NOTE: Returns referenced socket.
285 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
289 error = so_pru_bind(so, nam, td);
294 sodealloc(struct socket *so)
296 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 0);
299 if (so->so_options & SO_ACCEPTCONN) {
300 KASSERT(TAILQ_EMPTY(&so->so_comp), ("so_comp is not empty"));
301 KASSERT(TAILQ_EMPTY(&so->so_incomp),
302 ("so_incomp is not empty"));
306 if (so->so_rcv.ssb_hiwat)
307 (void)chgsbsize(so->so_cred->cr_uidinfo,
308 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
309 if (so->so_snd.ssb_hiwat)
310 (void)chgsbsize(so->so_cred->cr_uidinfo,
311 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
313 /* remove accept filter if present */
314 if (so->so_accf != NULL)
315 do_setopt_accept_filter(so, NULL);
318 if (so->so_faddr != NULL)
319 kfree(so->so_faddr, M_SONAME);
324 solisten(struct socket *so, int backlog, struct thread *td)
326 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
329 lwkt_gettoken(&so->so_rcv.ssb_token);
330 if (TAILQ_EMPTY(&so->so_comp))
331 so->so_options |= SO_ACCEPTCONN;
332 lwkt_reltoken(&so->so_rcv.ssb_token);
333 if (backlog < 0 || backlog > somaxconn)
335 so->so_qlimit = backlog;
336 return so_pru_listen(so, td);
340 soqflush(struct socket *so)
342 lwkt_getpooltoken(so);
343 if (so->so_options & SO_ACCEPTCONN) {
346 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
347 KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
349 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
351 soclrstate(sp, SS_INCOMP);
352 soabort_async(sp, TRUE);
354 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
355 KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
357 TAILQ_REMOVE(&so->so_comp, sp, so_list);
359 soclrstate(sp, SS_COMP);
360 soabort_async(sp, TRUE);
363 lwkt_relpooltoken(so);
367 * Destroy a disconnected socket. This routine is a NOP if entities
368 * still have a reference on the socket:
370 * so_pcb - The protocol stack still has a reference
371 * SS_NOFDREF - There is no longer a file pointer reference
374 sofree(struct socket *so)
379 * This is a bit hackish at the moment. We need to interlock
380 * any accept queue we are on before we potentially lose the
381 * last reference to avoid races against a re-reference from
382 * someone operating on the queue.
384 while ((head = so->so_head) != NULL) {
385 lwkt_getpooltoken(head);
386 if (so->so_head == head)
388 lwkt_relpooltoken(head);
392 * Arbitrage the last free.
394 KKASSERT(so->so_refs > 0);
395 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
397 lwkt_relpooltoken(head);
401 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
402 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
406 * We're done, remove ourselves from the accept queue we are
407 * on, if we are on one.
409 if (so->so_state & SS_INCOMP) {
410 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) ==
412 TAILQ_REMOVE(&head->so_incomp, so, so_list);
414 } else if (so->so_state & SS_COMP) {
416 * We must not decommission a socket that's
417 * on the accept(2) queue. If we do, then
418 * accept(2) may hang after select(2) indicated
419 * that the listening socket was ready.
421 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) ==
423 lwkt_relpooltoken(head);
426 panic("sofree: not queued");
428 soclrstate(so, SS_INCOMP);
430 lwkt_relpooltoken(head);
432 /* Flush accept queues, if we are accepting. */
435 ssb_release(&so->so_snd, so);
441 * Close a socket on last file table reference removal.
442 * Initiate disconnect if connected.
443 * Free socket when disconnect complete.
446 soclose(struct socket *so, int fflag)
450 funsetown(&so->so_sigio);
451 sosetstate(so, SS_ISCLOSING);
452 if (!use_soclose_fast ||
453 (so->so_proto->pr_flags & PR_SYNC_PORT) ||
454 ((so->so_state & SS_ISCONNECTED) &&
455 (so->so_options & SO_LINGER))) {
456 error = soclose_sync(so, fflag);
465 sodiscard(struct socket *so)
467 if (so->so_state & SS_NOFDREF)
468 panic("soclose: NOFDREF");
469 sosetstate(so, SS_NOFDREF); /* take ref */
473 * Append the completed queue of head to head_inh (inherting listen socket).
476 soinherit(struct socket *head, struct socket *head_inh)
478 boolean_t do_wakeup = FALSE;
480 KASSERT(head->so_options & SO_ACCEPTCONN,
481 ("head does not accept connection"));
482 KASSERT(head_inh->so_options & SO_ACCEPTCONN,
483 ("head_inh does not accept connection"));
485 lwkt_getpooltoken(head);
486 lwkt_getpooltoken(head_inh);
488 if (head->so_qlen > 0)
491 while (!TAILQ_EMPTY(&head->so_comp)) {
492 struct ucred *old_cr;
495 sp = TAILQ_FIRST(&head->so_comp);
496 KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP);
499 * Remove this socket from the current listen socket
502 TAILQ_REMOVE(&head->so_comp, sp, so_list);
505 /* Save the old ucred for later free. */
506 old_cr = sp->so_cred;
509 * Install this socket to the inheriting listen socket
512 sp->so_cred = crhold(head_inh->so_cred); /* non-blocking */
513 sp->so_head = head_inh;
515 TAILQ_INSERT_TAIL(&head_inh->so_comp, sp, so_list);
520 * crfree() may block and release the tokens temporarily.
521 * However, we are fine here, since the transition is done.
526 lwkt_relpooltoken(head_inh);
527 lwkt_relpooltoken(head);
531 * "New" connections have arrived
534 wakeup(&head_inh->so_timeo);
539 soclose_sync(struct socket *so, int fflag)
543 if ((so->so_proto->pr_flags & PR_SYNC_PORT) == 0)
544 so_pru_sync(so); /* unpend async prus */
546 if (so->so_pcb == NULL)
549 if (so->so_state & SS_ISCONNECTED) {
550 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
551 error = sodisconnect(so);
555 if (so->so_options & SO_LINGER) {
556 if ((so->so_state & SS_ISDISCONNECTING) &&
559 while (so->so_state & SS_ISCONNECTED) {
560 error = tsleep(&so->so_timeo, PCATCH,
561 "soclos", so->so_linger * hz);
571 error2 = so_pru_detach(so);
572 if (error2 == EJUSTRETURN) {
574 * Protocol will call sodiscard()
575 * and sofree() for us.
584 sofree(so); /* dispose of ref */
590 soclose_fast_handler(netmsg_t msg)
592 struct socket *so = msg->base.nm_so;
594 if (so->so_pcb == NULL)
597 if ((so->so_state & SS_ISCONNECTED) &&
598 (so->so_state & SS_ISDISCONNECTING) == 0)
599 so_pru_disconnect_direct(so);
604 error = so_pru_detach_direct(so);
605 if (error == EJUSTRETURN) {
607 * Protocol will call sodiscard()
608 * and sofree() for us.
619 soclose_fast(struct socket *so)
621 struct netmsg_base *base = &so->so_clomsg;
623 netmsg_init(base, so, &netisr_apanic_rport, 0,
624 soclose_fast_handler);
625 lwkt_sendmsg(so->so_port, &base->lmsg);
629 * Abort and destroy a socket. Only one abort can be in progress
630 * at any given moment.
633 soabort_async(struct socket *so, boolean_t clr_head)
636 * Keep a reference before clearing the so_head
637 * to avoid racing socket close in netisr.
642 so_pru_abort_async(so);
646 soabort_direct(struct socket *so)
649 so_pru_abort_direct(so);
653 * so is passed in ref'd, which becomes owned by
654 * the cleared SS_NOFDREF flag.
657 soaccept_generic(struct socket *so)
659 if ((so->so_state & SS_NOFDREF) == 0)
660 panic("soaccept: !NOFDREF");
661 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */
665 soaccept(struct socket *so, struct sockaddr **nam)
669 soaccept_generic(so);
670 error = so_pru_accept(so, nam);
675 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td,
680 if (so->so_options & SO_ACCEPTCONN)
683 * If protocol is connection-based, can only connect once.
684 * Otherwise, if connected, try to disconnect first.
685 * This allows user to disconnect by connecting to, e.g.,
688 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
689 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
690 (error = sodisconnect(so)))) {
694 * Prevent accumulated error from previous connection
698 if (!sync && so->so_proto->pr_usrreqs->pru_preconnect)
699 error = so_pru_connect_async(so, nam, td);
701 error = so_pru_connect(so, nam, td);
707 soconnect2(struct socket *so1, struct socket *so2)
711 error = so_pru_connect2(so1, so2);
716 sodisconnect(struct socket *so)
720 if ((so->so_state & SS_ISCONNECTED) == 0) {
724 if (so->so_state & SS_ISDISCONNECTING) {
728 error = so_pru_disconnect(so);
733 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
736 * If send must go all at once and message is larger than
737 * send buffering, then hard error.
738 * Lock against other senders.
739 * If must go all at once and not enough room now, then
740 * inform user that this would block and do nothing.
741 * Otherwise, if nonblocking, send as much as possible.
742 * The data to be sent is described by "uio" if nonzero,
743 * otherwise by the mbuf chain "top" (which must be null
744 * if uio is not). Data provided in mbuf chain must be small
745 * enough to send all at once.
747 * Returns nonzero on error, timeout or signal; callers
748 * must check for short counts if EINTR/ERESTART are returned.
749 * Data and control buffers are freed on return.
752 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
753 struct mbuf *top, struct mbuf *control, int flags,
760 int clen = 0, error, dontroute, mlen;
761 int atomic = sosendallatonce(so) || top;
765 resid = uio->uio_resid;
767 resid = (size_t)top->m_pkthdr.len;
770 for (m = top; m; m = m->m_next)
772 KKASSERT(top->m_pkthdr.len == len);
777 * WARNING! resid is unsigned, space and len are signed. space
778 * can wind up negative if the sockbuf is overcommitted.
780 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
781 * type sockets since that's an error.
783 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
789 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
790 (so->so_proto->pr_flags & PR_ATOMIC);
791 if (td->td_lwp != NULL)
792 td->td_lwp->lwp_ru.ru_msgsnd++;
794 clen = control->m_len;
795 #define gotoerr(errcode) { error = errcode; goto release; }
798 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
803 if (so->so_state & SS_CANTSENDMORE)
806 error = so->so_error;
810 if ((so->so_state & SS_ISCONNECTED) == 0) {
812 * `sendto' and `sendmsg' is allowed on a connection-
813 * based socket if it supports implied connect.
814 * Return ENOTCONN if not connected and no address is
817 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
818 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
819 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
820 !(resid == 0 && clen != 0))
822 } else if (addr == NULL)
823 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
824 ENOTCONN : EDESTADDRREQ);
826 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
827 clen > so->so_snd.ssb_hiwat) {
830 space = ssb_space(&so->so_snd);
833 if ((space < 0 || (size_t)space < resid + clen) && uio &&
834 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
835 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
836 gotoerr(EWOULDBLOCK);
837 ssb_unlock(&so->so_snd);
838 error = ssb_wait(&so->so_snd);
848 * Data is prepackaged in "top".
852 top->m_flags |= M_EOR;
856 m = m_getl((int)resid, M_WAITOK, MT_DATA,
857 top == NULL ? M_PKTHDR : 0, &mlen);
860 m->m_pkthdr.rcvif = NULL;
862 len = imin((int)szmin(mlen, resid), space);
863 if (resid < MINCLSIZE) {
865 * For datagram protocols, leave room
866 * for protocol headers in first mbuf.
868 if (atomic && top == NULL && len < mlen)
872 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
873 resid = uio->uio_resid;
876 top->m_pkthdr.len += len;
882 top->m_flags |= M_EOR;
885 } while (space > 0 && atomic);
887 so->so_options |= SO_DONTROUTE;
888 if (flags & MSG_OOB) {
889 pru_flags = PRUS_OOB;
890 } else if ((flags & MSG_EOF) &&
891 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
894 * If the user set MSG_EOF, the protocol
895 * understands this flag and nothing left to
896 * send then use PRU_SEND_EOF instead of PRU_SEND.
898 pru_flags = PRUS_EOF;
899 } else if (resid > 0 && space > 0) {
900 /* If there is more to send, set PRUS_MORETOCOME */
901 pru_flags = PRUS_MORETOCOME;
906 * XXX all the SS_CANTSENDMORE checks previously
907 * done could be out of date. We could have recieved
908 * a reset packet in an interrupt or maybe we slept
909 * while doing page faults in uiomove() etc. We could
910 * probably recheck again inside the splnet() protection
911 * here, but there are probably other places that this
912 * also happens. We must rethink this.
914 error = so_pru_send(so, pru_flags, top, addr, control, td);
916 so->so_options &= ~SO_DONTROUTE;
923 } while (resid && space > 0);
927 ssb_unlock(&so->so_snd);
938 * A specialization of sosend() for UDP based on protocol-specific knowledge:
939 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
940 * sosendallatonce() returns true,
941 * the "atomic" variable is true,
942 * and sosendudp() blocks until space is available for the entire send.
943 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
944 * PR_IMPLOPCL flags set.
945 * UDP has no out-of-band data.
946 * UDP has no control data.
947 * UDP does not support MSG_EOR.
950 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
951 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
954 int error, pru_flags = 0;
957 if (td->td_lwp != NULL)
958 td->td_lwp->lwp_ru.ru_msgsnd++;
962 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
963 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
966 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
970 if (so->so_state & SS_CANTSENDMORE)
973 error = so->so_error;
977 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
978 gotoerr(EDESTADDRREQ);
979 if (resid > so->so_snd.ssb_hiwat)
981 space = ssb_space(&so->so_snd);
982 if (uio && (space < 0 || (size_t)space < resid)) {
983 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
984 gotoerr(EWOULDBLOCK);
985 ssb_unlock(&so->so_snd);
986 error = ssb_wait(&so->so_snd);
993 int hdrlen = max_hdr;
996 * We try to optimize out the additional mbuf
997 * allocations in M_PREPEND() on output path, e.g.
998 * - udp_output(), when it tries to prepend protocol
1000 * - Link layer output function, when it tries to
1001 * prepend link layer header.
1003 * This probably will not benefit any data that will
1004 * be fragmented, so this optimization is only performed
1005 * when the size of data and max size of protocol+link
1006 * headers fit into one mbuf cluster.
1008 if (uio->uio_resid > MCLBYTES - hdrlen ||
1009 !udp_sosend_prepend) {
1010 top = m_uiomove(uio);
1016 top = m_getl(uio->uio_resid + hdrlen, M_WAITOK,
1017 MT_DATA, M_PKTHDR, &nsize);
1018 KASSERT(nsize >= uio->uio_resid + hdrlen,
1019 ("sosendudp invalid nsize %d, "
1020 "resid %zu, hdrlen %d",
1021 nsize, uio->uio_resid, hdrlen));
1023 top->m_len = uio->uio_resid;
1024 top->m_pkthdr.len = uio->uio_resid;
1025 top->m_data += hdrlen;
1027 error = uiomove(mtod(top, caddr_t), top->m_len, uio);
1033 if (flags & MSG_DONTROUTE)
1034 pru_flags |= PRUS_DONTROUTE;
1036 if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
1037 so_pru_send_async(so, pru_flags, top, addr, NULL, td);
1040 error = so_pru_send(so, pru_flags, top, addr, NULL, td);
1042 top = NULL; /* sent or freed in lower layer */
1045 ssb_unlock(&so->so_snd);
1053 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1054 struct mbuf *top, struct mbuf *control, int flags,
1066 KKASSERT(top == NULL);
1068 resid = uio->uio_resid;
1071 resid = (size_t)top->m_pkthdr.len;
1074 for (m = top; m; m = m->m_next)
1076 KKASSERT(top->m_pkthdr.len == len);
1081 * WARNING! resid is unsigned, space and len are signed. space
1082 * can wind up negative if the sockbuf is overcommitted.
1084 * Also check to make sure that MSG_EOR isn't used on TCP
1086 if (flags & MSG_EOR) {
1092 /* TCP doesn't do control messages (rights, creds, etc) */
1093 if (control->m_len) {
1097 m_freem(control); /* empty control, just free it */
1101 if (td->td_lwp != NULL)
1102 td->td_lwp->lwp_ru.ru_msgsnd++;
1104 #define gotoerr(errcode) { error = errcode; goto release; }
1107 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1112 if (so->so_state & SS_CANTSENDMORE)
1115 error = so->so_error;
1119 if ((so->so_state & SS_ISCONNECTED) == 0 &&
1120 (so->so_state & SS_ISCONFIRMING) == 0)
1122 if (allatonce && resid > so->so_snd.ssb_hiwat)
1125 space = ssb_space_prealloc(&so->so_snd);
1126 if (flags & MSG_OOB)
1128 if ((space < 0 || (size_t)space < resid) && !allatonce &&
1129 space < so->so_snd.ssb_lowat) {
1130 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1131 gotoerr(EWOULDBLOCK);
1132 ssb_unlock(&so->so_snd);
1133 error = ssb_wait(&so->so_snd);
1140 int cnt = 0, async = 0;
1144 * Data is prepackaged in "top".
1148 if (resid > INT_MAX)
1150 if (tcp_sosend_jcluster) {
1151 m = m_getlj((int)resid, M_WAITOK, MT_DATA,
1152 top == NULL ? M_PKTHDR : 0, &mlen);
1154 m = m_getl((int)resid, M_WAITOK, MT_DATA,
1155 top == NULL ? M_PKTHDR : 0, &mlen);
1158 m->m_pkthdr.len = 0;
1159 m->m_pkthdr.rcvif = NULL;
1161 len = imin((int)szmin(mlen, resid), space);
1163 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1164 resid = uio->uio_resid;
1167 top->m_pkthdr.len += len;
1174 } while (space > 0 && cnt < tcp_sosend_agglim);
1176 if (tcp_sosend_async)
1179 if (flags & MSG_OOB) {
1180 pru_flags = PRUS_OOB;
1182 } else if ((flags & MSG_EOF) && resid == 0) {
1183 pru_flags = PRUS_EOF;
1184 } else if (resid > 0 && space > 0) {
1185 /* If there is more to send, set PRUS_MORETOCOME */
1186 pru_flags = PRUS_MORETOCOME;
1192 if (flags & MSG_SYNC)
1196 * XXX all the SS_CANTSENDMORE checks previously
1197 * done could be out of date. We could have recieved
1198 * a reset packet in an interrupt or maybe we slept
1199 * while doing page faults in uiomove() etc. We could
1200 * probably recheck again inside the splnet() protection
1201 * here, but there are probably other places that this
1202 * also happens. We must rethink this.
1204 for (m = top; m; m = m->m_next)
1205 ssb_preallocstream(&so->so_snd, m);
1207 error = so_pru_send(so, pru_flags, top,
1210 so_pru_send_async(so, pru_flags, top,
1219 } while (resid && space > 0);
1223 ssb_unlock(&so->so_snd);
1234 * Implement receive operations on a socket.
1236 * We depend on the way that records are added to the signalsockbuf
1237 * by sbappend*. In particular, each record (mbufs linked through m_next)
1238 * must begin with an address if the protocol so specifies,
1239 * followed by an optional mbuf or mbufs containing ancillary data,
1240 * and then zero or more mbufs of data.
1242 * Although the signalsockbuf is locked, new data may still be appended.
1243 * A token inside the ssb_lock deals with MP issues and still allows
1244 * the network to access the socket if we block in a uio.
1246 * The caller may receive the data as a single mbuf chain by supplying
1247 * an mbuf **mp0 for use in returning the chain. The uio is then used
1248 * only for the count in uio_resid.
1251 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1252 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1255 struct mbuf *free_chain = NULL;
1256 int flags, len, error, offset;
1257 struct protosw *pr = so->so_proto;
1259 size_t resid, orig_resid;
1260 boolean_t free_rights = FALSE;
1263 resid = uio->uio_resid;
1265 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1273 flags = *flagsp &~ MSG_EOR;
1276 if (flags & MSG_OOB) {
1277 m = m_get(M_WAITOK, MT_DATA);
1280 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1286 KKASSERT(resid >= (size_t)m->m_len);
1287 resid -= (size_t)m->m_len;
1288 } while (resid > 0 && m);
1291 uio->uio_resid = resid;
1292 error = uiomove(mtod(m, caddr_t),
1293 (int)szmin(resid, m->m_len),
1295 resid = uio->uio_resid;
1297 } while (uio->uio_resid && error == 0 && m);
1304 if ((so->so_state & SS_ISCONFIRMING) && resid)
1308 * The token interlocks against the protocol thread while
1309 * ssb_lock is a blocking lock against other userland entities.
1311 lwkt_gettoken(&so->so_rcv.ssb_token);
1313 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1317 m = so->so_rcv.ssb_mb;
1319 * If we have less data than requested, block awaiting more
1320 * (subject to any timeout) if:
1321 * 1. the current count is less than the low water mark, or
1322 * 2. MSG_WAITALL is set, and it is possible to do the entire
1323 * receive operation at once if we block (resid <= hiwat).
1324 * 3. MSG_DONTWAIT is not set
1325 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1326 * we have to do the receive in sections, and thus risk returning
1327 * a short count if a timeout or signal occurs after we start.
1329 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1330 (size_t)so->so_rcv.ssb_cc < resid) &&
1331 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1332 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1333 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1334 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1338 error = so->so_error;
1339 if ((flags & MSG_PEEK) == 0)
1343 if (so->so_state & SS_CANTRCVMORE) {
1349 for (; m; m = m->m_next) {
1350 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1351 m = so->so_rcv.ssb_mb;
1355 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1356 (pr->pr_flags & PR_CONNREQUIRED)) {
1362 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1363 error = EWOULDBLOCK;
1366 ssb_unlock(&so->so_rcv);
1367 error = ssb_wait(&so->so_rcv);
1373 if (uio && uio->uio_td && uio->uio_td->td_proc)
1374 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1377 * note: m should be == sb_mb here. Cache the next record while
1378 * cleaning up. Note that calling m_free*() will break out critical
1381 KKASSERT(m == so->so_rcv.ssb_mb);
1384 * Skip any address mbufs prepending the record.
1386 if (pr->pr_flags & PR_ADDR) {
1387 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1390 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1391 if (flags & MSG_PEEK)
1394 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1398 * Skip any control mbufs prepending the record.
1400 while (m && m->m_type == MT_CONTROL && error == 0) {
1401 if (flags & MSG_PEEK) {
1403 *controlp = m_copy(m, 0, m->m_len);
1404 m = m->m_next; /* XXX race */
1406 const struct cmsghdr *cm = mtod(m, struct cmsghdr *);
1409 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1410 if (pr->pr_domain->dom_externalize &&
1411 cm->cmsg_level == SOL_SOCKET &&
1412 cm->cmsg_type == SCM_RIGHTS) {
1413 error = pr->pr_domain->dom_externalize
1419 if (cm->cmsg_level == SOL_SOCKET &&
1420 cm->cmsg_type == SCM_RIGHTS)
1422 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1425 if (controlp && *controlp) {
1427 controlp = &(*controlp)->m_next;
1436 if (type == MT_OOBDATA)
1441 * Copy to the UIO or mbuf return chain (*mp).
1445 while (m && resid > 0 && error == 0) {
1446 if (m->m_type == MT_OOBDATA) {
1447 if (type != MT_OOBDATA)
1449 } else if (type == MT_OOBDATA)
1452 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1454 soclrstate(so, SS_RCVATMARK);
1455 len = (resid > INT_MAX) ? INT_MAX : resid;
1456 if (so->so_oobmark && len > so->so_oobmark - offset)
1457 len = so->so_oobmark - offset;
1458 if (len > m->m_len - moff)
1459 len = m->m_len - moff;
1462 * Copy out to the UIO or pass the mbufs back to the SIO.
1463 * The SIO is dealt with when we eat the mbuf, but deal
1464 * with the resid here either way.
1467 uio->uio_resid = resid;
1468 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1469 resid = uio->uio_resid;
1473 resid -= (size_t)len;
1477 * Eat the entire mbuf or just a piece of it
1479 if (len == m->m_len - moff) {
1480 if (m->m_flags & M_EOR)
1482 if (flags & MSG_PEEK) {
1487 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1491 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1495 if (flags & MSG_PEEK) {
1499 n = m_copym(m, 0, len, M_WAITOK);
1505 so->so_rcv.ssb_cc -= len;
1508 if (so->so_oobmark) {
1509 if ((flags & MSG_PEEK) == 0) {
1510 so->so_oobmark -= len;
1511 if (so->so_oobmark == 0) {
1512 sosetstate(so, SS_RCVATMARK);
1517 if (offset == so->so_oobmark)
1521 if (flags & MSG_EOR)
1524 * If the MSG_WAITALL flag is set (for non-atomic socket),
1525 * we must not quit until resid == 0 or an error
1526 * termination. If a signal/timeout occurs, return
1527 * with a short count but without error.
1528 * Keep signalsockbuf locked against other readers.
1530 while ((flags & MSG_WAITALL) && m == NULL &&
1531 resid > 0 && !sosendallatonce(so) &&
1532 so->so_rcv.ssb_mb == NULL) {
1533 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1536 * The window might have closed to zero, make
1537 * sure we send an ack now that we've drained
1538 * the buffer or we might end up blocking until
1539 * the idle takes over (5 seconds).
1541 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1542 so_pru_rcvd(so, flags);
1543 error = ssb_wait(&so->so_rcv);
1545 ssb_unlock(&so->so_rcv);
1549 m = so->so_rcv.ssb_mb;
1554 * If an atomic read was requested but unread data still remains
1555 * in the record, set MSG_TRUNC.
1557 if (m && pr->pr_flags & PR_ATOMIC)
1561 * Cleanup. If an atomic read was requested drop any unread data.
1563 if ((flags & MSG_PEEK) == 0) {
1564 if (m && (pr->pr_flags & PR_ATOMIC))
1565 sbdroprecord(&so->so_rcv.sb);
1566 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1567 so_pru_rcvd(so, flags);
1570 if (orig_resid == resid && orig_resid &&
1571 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1572 ssb_unlock(&so->so_rcv);
1579 ssb_unlock(&so->so_rcv);
1581 lwkt_reltoken(&so->so_rcv.ssb_token);
1583 if (free_rights && (pr->pr_flags & PR_RIGHTS) &&
1584 pr->pr_domain->dom_dispose)
1585 pr->pr_domain->dom_dispose(free_chain);
1586 m_freem(free_chain);
1592 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1593 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1596 struct mbuf *free_chain = NULL;
1597 int flags, len, error, offset;
1598 struct protosw *pr = so->so_proto;
1601 size_t resid, orig_resid, restmp;
1604 resid = uio->uio_resid;
1606 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1614 flags = *flagsp &~ MSG_EOR;
1617 if (flags & MSG_OOB) {
1618 m = m_get(M_WAITOK, MT_DATA);
1621 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1627 KKASSERT(resid >= (size_t)m->m_len);
1628 resid -= (size_t)m->m_len;
1629 } while (resid > 0 && m);
1632 uio->uio_resid = resid;
1633 error = uiomove(mtod(m, caddr_t),
1634 (int)szmin(resid, m->m_len),
1636 resid = uio->uio_resid;
1638 } while (uio->uio_resid && error == 0 && m);
1647 * The token interlocks against the protocol thread while
1648 * ssb_lock is a blocking lock against other userland entities.
1650 * Lock a limited number of mbufs (not all, so sbcompress() still
1651 * works well). The token is used as an interlock for sbwait() so
1652 * release it afterwords.
1655 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1659 lwkt_gettoken(&so->so_rcv.ssb_token);
1660 m = so->so_rcv.ssb_mb;
1663 * If we have less data than requested, block awaiting more
1664 * (subject to any timeout) if:
1665 * 1. the current count is less than the low water mark, or
1666 * 2. MSG_WAITALL is set, and it is possible to do the entire
1667 * receive operation at once if we block (resid <= hiwat).
1668 * 3. MSG_DONTWAIT is not set
1669 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1670 * we have to do the receive in sections, and thus risk returning
1671 * a short count if a timeout or signal occurs after we start.
1673 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1674 (size_t)so->so_rcv.ssb_cc < resid) &&
1675 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1676 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1677 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1681 lwkt_reltoken(&so->so_rcv.ssb_token);
1682 error = so->so_error;
1683 if ((flags & MSG_PEEK) == 0)
1687 if (so->so_state & SS_CANTRCVMORE) {
1690 lwkt_reltoken(&so->so_rcv.ssb_token);
1693 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1694 (pr->pr_flags & PR_CONNREQUIRED)) {
1695 lwkt_reltoken(&so->so_rcv.ssb_token);
1700 lwkt_reltoken(&so->so_rcv.ssb_token);
1703 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1704 lwkt_reltoken(&so->so_rcv.ssb_token);
1705 error = EWOULDBLOCK;
1708 ssb_unlock(&so->so_rcv);
1709 error = ssb_wait(&so->so_rcv);
1710 lwkt_reltoken(&so->so_rcv.ssb_token);
1722 while (n && restmp < resid) {
1723 n->m_flags |= M_SOLOCKED;
1725 if (n->m_next == NULL)
1732 * Release token for loop
1734 lwkt_reltoken(&so->so_rcv.ssb_token);
1735 if (uio && uio->uio_td && uio->uio_td->td_proc)
1736 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1739 * note: m should be == sb_mb here. Cache the next record while
1740 * cleaning up. Note that calling m_free*() will break out critical
1743 KKASSERT(m == so->so_rcv.ssb_mb);
1746 * Copy to the UIO or mbuf return chain (*mp).
1748 * NOTE: Token is not held for loop
1754 while (m && (m->m_flags & M_SOLOCKED) && resid > 0 && error == 0) {
1755 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1758 soclrstate(so, SS_RCVATMARK);
1759 len = (resid > INT_MAX) ? INT_MAX : resid;
1760 if (so->so_oobmark && len > so->so_oobmark - offset)
1761 len = so->so_oobmark - offset;
1762 if (len > m->m_len - moff)
1763 len = m->m_len - moff;
1766 * Copy out to the UIO or pass the mbufs back to the SIO.
1767 * The SIO is dealt with when we eat the mbuf, but deal
1768 * with the resid here either way.
1771 uio->uio_resid = resid;
1772 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1773 resid = uio->uio_resid;
1777 resid -= (size_t)len;
1781 * Eat the entire mbuf or just a piece of it
1784 if (len == m->m_len - moff) {
1794 if (so->so_oobmark && offset == so->so_oobmark) {
1801 * Synchronize sockbuf with data we read.
1803 * NOTE: (m) is junk on entry (it could be left over from the
1806 if ((flags & MSG_PEEK) == 0) {
1807 lwkt_gettoken(&so->so_rcv.ssb_token);
1808 m = so->so_rcv.ssb_mb;
1809 while (m && offset >= m->m_len) {
1810 if (so->so_oobmark) {
1811 so->so_oobmark -= m->m_len;
1812 if (so->so_oobmark == 0) {
1813 sosetstate(so, SS_RCVATMARK);
1819 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1823 m = sbunlinkmbuf(&so->so_rcv.sb,
1830 n = m_copym(m, 0, offset, M_WAITOK);
1834 m->m_data += offset;
1836 so->so_rcv.ssb_cc -= offset;
1837 if (so->so_oobmark) {
1838 so->so_oobmark -= offset;
1839 if (so->so_oobmark == 0) {
1840 sosetstate(so, SS_RCVATMARK);
1846 lwkt_reltoken(&so->so_rcv.ssb_token);
1850 * If the MSG_WAITALL flag is set (for non-atomic socket),
1851 * we must not quit until resid == 0 or an error termination.
1853 * If a signal/timeout occurs, return with a short count but without
1856 * Keep signalsockbuf locked against other readers.
1858 * XXX if MSG_PEEK we currently do quit.
1860 if ((flags & MSG_WAITALL) && !(flags & MSG_PEEK) &&
1861 didoob == 0 && resid > 0 &&
1862 !sosendallatonce(so)) {
1863 lwkt_gettoken(&so->so_rcv.ssb_token);
1865 while ((m = so->so_rcv.ssb_mb) == NULL) {
1866 if (so->so_error || (so->so_state & SS_CANTRCVMORE)) {
1867 error = so->so_error;
1871 * The window might have closed to zero, make
1872 * sure we send an ack now that we've drained
1873 * the buffer or we might end up blocking until
1874 * the idle takes over (5 seconds).
1877 so_pru_rcvd_async(so);
1878 if (so->so_rcv.ssb_mb == NULL)
1879 error = ssb_wait(&so->so_rcv);
1881 lwkt_reltoken(&so->so_rcv.ssb_token);
1882 ssb_unlock(&so->so_rcv);
1887 if (m && error == 0)
1889 lwkt_reltoken(&so->so_rcv.ssb_token);
1893 * Token not held here.
1895 * Cleanup. If an atomic read was requested drop any unread data XXX
1897 if ((flags & MSG_PEEK) == 0) {
1899 so_pru_rcvd_async(so);
1902 if (orig_resid == resid && orig_resid &&
1903 (so->so_state & SS_CANTRCVMORE) == 0) {
1904 ssb_unlock(&so->so_rcv);
1911 ssb_unlock(&so->so_rcv);
1914 m_freem(free_chain);
1919 * Shut a socket down. Note that we do not get a frontend lock as we
1920 * want to be able to shut the socket down even if another thread is
1921 * blocked in a read(), thus waking it up.
1924 soshutdown(struct socket *so, int how)
1926 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1929 if (how != SHUT_WR) {
1930 /*ssb_lock(&so->so_rcv, M_WAITOK);*/
1932 /*ssb_unlock(&so->so_rcv);*/
1935 return (so_pru_shutdown(so));
1940 sorflush(struct socket *so)
1942 struct signalsockbuf *ssb = &so->so_rcv;
1943 struct protosw *pr = so->so_proto;
1944 struct signalsockbuf asb;
1946 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1948 lwkt_gettoken(&ssb->ssb_token);
1953 * Can't just blow up the ssb structure here
1955 bzero(&ssb->sb, sizeof(ssb->sb));
1960 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1962 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1963 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1964 ssb_release(&asb, so);
1966 lwkt_reltoken(&ssb->ssb_token);
1971 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1973 struct accept_filter_arg *afap = NULL;
1974 struct accept_filter *afp;
1975 struct so_accf *af = so->so_accf;
1978 /* do not set/remove accept filters on non listen sockets */
1979 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1984 /* removing the filter */
1987 if (af->so_accept_filter != NULL &&
1988 af->so_accept_filter->accf_destroy != NULL) {
1989 af->so_accept_filter->accf_destroy(so);
1991 if (af->so_accept_filter_str != NULL) {
1992 kfree(af->so_accept_filter_str, M_ACCF);
1997 so->so_options &= ~SO_ACCEPTFILTER;
2000 /* adding a filter */
2001 /* must remove previous filter first */
2006 /* don't put large objects on the kernel stack */
2007 afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
2008 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
2009 afap->af_name[sizeof(afap->af_name)-1] = '\0';
2010 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
2013 afp = accept_filt_get(afap->af_name);
2018 af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
2019 if (afp->accf_create != NULL) {
2020 if (afap->af_name[0] != '\0') {
2021 int len = strlen(afap->af_name) + 1;
2023 af->so_accept_filter_str = kmalloc(len, M_ACCF,
2025 strcpy(af->so_accept_filter_str, afap->af_name);
2027 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
2028 if (af->so_accept_filter_arg == NULL) {
2029 kfree(af->so_accept_filter_str, M_ACCF);
2036 af->so_accept_filter = afp;
2038 so->so_options |= SO_ACCEPTFILTER;
2041 kfree(afap, M_TEMP);
2047 * Perhaps this routine, and sooptcopyout(), below, ought to come in
2048 * an additional variant to handle the case where the option value needs
2049 * to be some kind of integer, but not a specific size.
2050 * In addition to their use here, these functions are also called by the
2051 * protocol-level pr_ctloutput() routines.
2054 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2056 return soopt_to_kbuf(sopt, buf, len, minlen);
2060 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2064 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2065 KKASSERT(kva_p(buf));
2068 * If the user gives us more than we wanted, we ignore it,
2069 * but if we don't get the minimum length the caller
2070 * wants, we return EINVAL. On success, sopt->sopt_valsize
2071 * is set to however much we actually retrieved.
2073 if ((valsize = sopt->sopt_valsize) < minlen)
2076 sopt->sopt_valsize = valsize = len;
2078 bcopy(sopt->sopt_val, buf, valsize);
2084 sosetopt(struct socket *so, struct sockopt *sopt)
2090 struct signalsockbuf *sotmp;
2093 sopt->sopt_dir = SOPT_SET;
2094 if (sopt->sopt_level != SOL_SOCKET) {
2095 if (so->so_proto && so->so_proto->pr_ctloutput) {
2096 return (so_pr_ctloutput(so, sopt));
2098 error = ENOPROTOOPT;
2100 switch (sopt->sopt_name) {
2102 case SO_ACCEPTFILTER:
2103 error = do_setopt_accept_filter(so, sopt);
2109 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2113 so->so_linger = l.l_linger;
2115 so->so_options |= SO_LINGER;
2117 so->so_options &= ~SO_LINGER;
2123 case SO_USELOOPBACK:
2130 error = sooptcopyin(sopt, &optval, sizeof optval,
2135 so->so_options |= sopt->sopt_name;
2137 so->so_options &= ~sopt->sopt_name;
2144 error = sooptcopyin(sopt, &optval, sizeof optval,
2150 * Values < 1 make no sense for any of these
2151 * options, so disallow them.
2158 switch (sopt->sopt_name) {
2161 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2162 &so->so_snd : &so->so_rcv, (u_long)optval,
2164 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2168 sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2169 &so->so_snd : &so->so_rcv;
2170 atomic_clear_int(&sotmp->ssb_flags,
2175 * Make sure the low-water is never greater than
2179 so->so_snd.ssb_lowat =
2180 (optval > so->so_snd.ssb_hiwat) ?
2181 so->so_snd.ssb_hiwat : optval;
2182 atomic_clear_int(&so->so_snd.ssb_flags,
2186 so->so_rcv.ssb_lowat =
2187 (optval > so->so_rcv.ssb_hiwat) ?
2188 so->so_rcv.ssb_hiwat : optval;
2189 atomic_clear_int(&so->so_rcv.ssb_flags,
2197 error = sooptcopyin(sopt, &tv, sizeof tv,
2202 /* assert(hz > 0); */
2203 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2204 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2208 /* assert(tick > 0); */
2209 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2210 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2211 if (val > INT_MAX) {
2215 if (val == 0 && tv.tv_usec != 0)
2218 switch (sopt->sopt_name) {
2220 so->so_snd.ssb_timeo = val;
2223 so->so_rcv.ssb_timeo = val;
2228 error = ENOPROTOOPT;
2231 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2232 (void) so_pr_ctloutput(so, sopt);
2239 /* Helper routine for getsockopt */
2241 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2243 soopt_from_kbuf(sopt, buf, len);
2248 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2253 sopt->sopt_valsize = 0;
2257 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2258 KKASSERT(kva_p(buf));
2261 * Documented get behavior is that we always return a value,
2262 * possibly truncated to fit in the user's buffer.
2263 * Traditional behavior is that we always tell the user
2264 * precisely how much we copied, rather than something useful
2265 * like the total amount we had available for her.
2266 * Note that this interface is not idempotent; the entire answer must
2267 * generated ahead of time.
2269 valsize = szmin(len, sopt->sopt_valsize);
2270 sopt->sopt_valsize = valsize;
2271 if (sopt->sopt_val != 0) {
2272 bcopy(buf, sopt->sopt_val, valsize);
2277 sogetopt(struct socket *so, struct sockopt *sopt)
2284 struct accept_filter_arg *afap;
2288 sopt->sopt_dir = SOPT_GET;
2289 if (sopt->sopt_level != SOL_SOCKET) {
2290 if (so->so_proto && so->so_proto->pr_ctloutput) {
2291 return (so_pr_ctloutput(so, sopt));
2293 return (ENOPROTOOPT);
2295 switch (sopt->sopt_name) {
2297 case SO_ACCEPTFILTER:
2298 if ((so->so_options & SO_ACCEPTCONN) == 0)
2300 afap = kmalloc(sizeof(*afap), M_TEMP,
2302 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2303 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2304 if (so->so_accf->so_accept_filter_str != NULL)
2305 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2307 error = sooptcopyout(sopt, afap, sizeof(*afap));
2308 kfree(afap, M_TEMP);
2313 l.l_onoff = so->so_options & SO_LINGER;
2314 l.l_linger = so->so_linger;
2315 error = sooptcopyout(sopt, &l, sizeof l);
2318 case SO_USELOOPBACK:
2328 optval = so->so_options & sopt->sopt_name;
2330 error = sooptcopyout(sopt, &optval, sizeof optval);
2334 optval = so->so_type;
2338 optval = so->so_error;
2343 optval = so->so_snd.ssb_hiwat;
2347 optval = so->so_rcv.ssb_hiwat;
2351 optval = so->so_snd.ssb_lowat;
2355 optval = so->so_rcv.ssb_lowat;
2360 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2361 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2363 tv.tv_sec = optval / hz;
2364 tv.tv_usec = (optval % hz) * ustick;
2365 error = sooptcopyout(sopt, &tv, sizeof tv);
2369 optval_l = ssb_space(&so->so_snd);
2370 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2374 optval = -1; /* no hint */
2378 error = ENOPROTOOPT;
2381 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput)
2382 so_pr_ctloutput(so, sopt);
2387 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2389 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2391 struct mbuf *m, *m_prev;
2392 int sopt_size = sopt->sopt_valsize, msize;
2394 m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA,
2398 m->m_len = min(msize, sopt_size);
2399 sopt_size -= m->m_len;
2403 while (sopt_size > 0) {
2404 m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT,
2405 MT_DATA, 0, &msize);
2410 m->m_len = min(msize, sopt_size);
2411 sopt_size -= m->m_len;
2418 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2420 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2422 soopt_to_mbuf(sopt, m);
2427 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2432 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2434 if (sopt->sopt_val == NULL)
2436 val = sopt->sopt_val;
2437 valsize = sopt->sopt_valsize;
2438 while (m != NULL && valsize >= m->m_len) {
2439 bcopy(val, mtod(m, char *), m->m_len);
2440 valsize -= m->m_len;
2441 val = (caddr_t)val + m->m_len;
2444 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2445 panic("ip6_sooptmcopyin");
2448 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2450 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2452 return soopt_from_mbuf(sopt, m);
2456 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2458 struct mbuf *m0 = m;
2463 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2465 if (sopt->sopt_val == NULL)
2467 val = sopt->sopt_val;
2468 maxsize = sopt->sopt_valsize;
2469 while (m != NULL && maxsize >= m->m_len) {
2470 bcopy(mtod(m, char *), val, m->m_len);
2471 maxsize -= m->m_len;
2472 val = (caddr_t)val + m->m_len;
2473 valsize += m->m_len;
2477 /* enough soopt buffer should be given from user-land */
2481 sopt->sopt_valsize = valsize;
2486 sohasoutofband(struct socket *so)
2488 if (so->so_sigio != NULL)
2489 pgsigio(so->so_sigio, SIGURG, 0);
2492 * There is no need to use NOTE_OOB as KNOTE hint here:
2493 * soread filter depends on so_oobmark and SS_RCVATMARK
2494 * so_state. NOTE_OOB would cause unnecessary penalty
2495 * in KNOTE, if there was knote processing contention.
2497 KNOTE(&so->so_rcv.ssb_kq.ki_note, 0);
2501 sokqfilter(struct file *fp, struct knote *kn)
2503 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2504 struct signalsockbuf *ssb;
2506 switch (kn->kn_filter) {
2508 if (so->so_options & SO_ACCEPTCONN)
2509 kn->kn_fop = &solisten_filtops;
2511 kn->kn_fop = &soread_filtops;
2515 kn->kn_fop = &sowrite_filtops;
2519 kn->kn_fop = &soexcept_filtops;
2523 return (EOPNOTSUPP);
2526 knote_insert(&ssb->ssb_kq.ki_note, kn);
2527 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2532 filt_sordetach(struct knote *kn)
2534 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2536 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2537 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2538 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2543 filt_soread(struct knote *kn, long hint __unused)
2545 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2547 if (kn->kn_sfflags & NOTE_OOB) {
2548 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2549 kn->kn_fflags |= NOTE_OOB;
2554 kn->kn_data = so->so_rcv.ssb_cc;
2556 if (so->so_state & SS_CANTRCVMORE) {
2558 * Only set NODATA if all data has been exhausted.
2560 if (kn->kn_data == 0)
2561 kn->kn_flags |= EV_NODATA;
2562 kn->kn_flags |= EV_EOF;
2563 kn->kn_fflags = so->so_error;
2566 if (so->so_error) /* temporary udp error */
2568 if (kn->kn_sfflags & NOTE_LOWAT)
2569 return (kn->kn_data >= kn->kn_sdata);
2570 return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2571 !TAILQ_EMPTY(&so->so_comp));
2575 filt_sowdetach(struct knote *kn)
2577 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2579 knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2580 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2581 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2586 filt_sowrite(struct knote *kn, long hint __unused)
2588 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2590 if (so->so_snd.ssb_flags & SSB_PREALLOC)
2591 kn->kn_data = ssb_space_prealloc(&so->so_snd);
2593 kn->kn_data = ssb_space(&so->so_snd);
2595 if (so->so_state & SS_CANTSENDMORE) {
2596 kn->kn_flags |= (EV_EOF | EV_NODATA);
2597 kn->kn_fflags = so->so_error;
2600 if (so->so_error) /* temporary udp error */
2602 if (((so->so_state & SS_ISCONNECTED) == 0) &&
2603 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2605 if (kn->kn_sfflags & NOTE_LOWAT)
2606 return (kn->kn_data >= kn->kn_sdata);
2607 return (kn->kn_data >= so->so_snd.ssb_lowat);
2612 filt_solisten(struct knote *kn, long hint __unused)
2614 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2616 kn->kn_data = so->so_qlen;
2617 return (! TAILQ_EMPTY(&so->so_comp));