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/socketvar2.h>
93 #include <sys/spinlock2.h>
95 #include <machine/limits.h>
98 extern int tcp_sosend_agglim;
99 extern int tcp_sosend_async;
100 extern int tcp_sosend_jcluster;
101 extern int udp_sosend_async;
102 extern int udp_sosend_prepend;
104 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
107 static void filt_sordetach(struct knote *kn);
108 static int filt_soread(struct knote *kn, long hint);
109 static void filt_sowdetach(struct knote *kn);
110 static int filt_sowrite(struct knote *kn, long hint);
111 static int filt_solisten(struct knote *kn, long hint);
113 static int soclose_sync(struct socket *so, int fflag);
114 static void soclose_fast(struct socket *so);
116 static struct filterops solisten_filtops =
117 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten };
118 static struct filterops soread_filtops =
119 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
120 static struct filterops sowrite_filtops =
121 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite };
122 static struct filterops soexcept_filtops =
123 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
125 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
126 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
127 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
130 static int somaxconn = SOMAXCONN;
131 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
132 &somaxconn, 0, "Maximum pending socket connection queue size");
134 static int use_soclose_fast = 1;
135 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW,
136 &use_soclose_fast, 0, "Fast socket close");
138 int use_soaccept_pred_fast = 1;
139 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW,
140 &use_soaccept_pred_fast, 0, "Fast socket accept predication");
142 int use_sendfile_async = 1;
143 SYSCTL_INT(_kern_ipc, OID_AUTO, sendfile_async, CTLFLAG_RW,
144 &use_sendfile_async, 0, "sendfile uses asynchronized pru_send");
146 int use_soconnect_async = 1;
147 SYSCTL_INT(_kern_ipc, OID_AUTO, soconnect_async, CTLFLAG_RW,
148 &use_soconnect_async, 0, "soconnect uses asynchronized pru_connect");
150 static int use_socreate_fast = 1;
151 SYSCTL_INT(_kern_ipc, OID_AUTO, socreate_fast, CTLFLAG_RW,
152 &use_socreate_fast, 0, "Fast socket creation");
154 static int soavailconn = 32;
155 SYSCTL_INT(_kern_ipc, OID_AUTO, soavailconn, CTLFLAG_RW,
156 &soavailconn, 0, "Maximum available socket connection queue size");
159 * Socket operation routines.
160 * These routines are called by the routines in
161 * sys_socket.c or from a system process, and
162 * implement the semantics of socket operations by
163 * switching out to the protocol specific routines.
167 * Get a socket structure, and initialize it.
168 * Note that it would probably be better to allocate socket
169 * and PCB at the same time, but I'm not convinced that all
170 * the protocols can be easily modified to do this.
173 soalloc(int waitok, struct protosw *pr)
175 globaldata_t gd = mycpu;
179 waitmask = waitok ? M_WAITOK : M_NOWAIT;
180 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
182 /* XXX race condition for reentrant kernel */
184 TAILQ_INIT(&so->so_aiojobq);
185 TAILQ_INIT(&so->so_rcv.ssb_mlist);
186 TAILQ_INIT(&so->so_snd.ssb_mlist);
187 lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
188 lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
189 spin_init(&so->so_rcvd_spin, "soalloc");
190 netmsg_init(&so->so_rcvd_msg.base, so, &netisr_adone_rport,
191 MSGF_DROPABLE | MSGF_PRIORITY,
192 so->so_proto->pr_usrreqs->pru_rcvd);
193 so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC;
194 so->so_state = SS_NOFDREF;
196 so->so_inum = gd->gd_anoninum++ * ncpus + gd->gd_cpuid + 2;
202 socreate(int dom, struct socket **aso, int type,
203 int proto, struct thread *td)
205 struct proc *p = td->td_proc;
208 struct pru_attach_info ai;
212 prp = pffindproto(dom, proto, type);
214 prp = pffindtype(dom, type);
216 if (prp == NULL || prp->pr_usrreqs->pru_attach == 0)
217 return (EPROTONOSUPPORT);
219 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
220 prp->pr_domain->dom_family != PF_LOCAL &&
221 prp->pr_domain->dom_family != PF_INET &&
222 prp->pr_domain->dom_family != PF_INET6 &&
223 prp->pr_domain->dom_family != PF_ROUTE) {
224 return (EPROTONOSUPPORT);
227 if (prp->pr_type != type)
229 so = soalloc(p != NULL, prp);
234 * Callers of socreate() presumably will connect up a descriptor
235 * and call soclose() if they cannot. This represents our so_refs
236 * (which should be 1) from soalloc().
238 soclrstate(so, SS_NOFDREF);
241 * Set a default port for protocol processing. No action will occur
242 * on the socket on this port until an inpcb is attached to it and
243 * is able to match incoming packets, or until the socket becomes
244 * available to userland.
246 * We normally default the socket to the protocol thread on cpu 0,
247 * if protocol does not provide its own method to initialize the
250 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
251 * thread and all pr_*()/pru_*() calls are executed synchronously.
253 if (prp->pr_flags & PR_SYNC_PORT)
254 so->so_port = &netisr_sync_port;
255 else if (prp->pr_initport != NULL)
256 so->so_port = prp->pr_initport();
258 so->so_port = netisr_cpuport(0);
260 TAILQ_INIT(&so->so_incomp);
261 TAILQ_INIT(&so->so_comp);
263 so->so_cred = crhold(p->p_ucred);
264 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
265 ai.p_ucred = p->p_ucred;
266 ai.fd_rdir = p->p_fd->fd_rdir;
269 * Auto-sizing of socket buffers is managed by the protocols and
270 * the appropriate flags must be set in the pru_attach function.
272 if (use_socreate_fast && prp->pr_usrreqs->pru_preattach)
273 error = so_pru_attach_fast(so, proto, &ai);
275 error = so_pru_attach(so, proto, &ai);
277 sosetstate(so, SS_NOFDREF);
278 sofree(so); /* from soalloc */
283 * NOTE: Returns referenced socket.
290 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
294 error = so_pru_bind(so, nam, td);
299 sodealloc(struct socket *so)
301 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 0);
304 if (so->so_options & SO_ACCEPTCONN) {
305 KASSERT(TAILQ_EMPTY(&so->so_comp), ("so_comp is not empty"));
306 KASSERT(TAILQ_EMPTY(&so->so_incomp),
307 ("so_incomp is not empty"));
311 if (so->so_rcv.ssb_hiwat)
312 (void)chgsbsize(so->so_cred->cr_uidinfo,
313 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
314 if (so->so_snd.ssb_hiwat)
315 (void)chgsbsize(so->so_cred->cr_uidinfo,
316 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
318 /* remove accept filter if present */
319 if (so->so_accf != NULL)
320 do_setopt_accept_filter(so, NULL);
323 if (so->so_faddr != NULL)
324 kfree(so->so_faddr, M_SONAME);
329 solisten(struct socket *so, int backlog, struct thread *td)
331 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
334 lwkt_gettoken(&so->so_rcv.ssb_token);
335 if (TAILQ_EMPTY(&so->so_comp))
336 so->so_options |= SO_ACCEPTCONN;
337 lwkt_reltoken(&so->so_rcv.ssb_token);
338 if (backlog < 0 || backlog > somaxconn)
340 so->so_qlimit = backlog;
341 return so_pru_listen(so, td);
345 soqflush(struct socket *so)
347 lwkt_getpooltoken(so);
348 if (so->so_options & SO_ACCEPTCONN) {
351 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
352 KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
354 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
356 soclrstate(sp, SS_INCOMP);
357 soabort_async(sp, TRUE);
359 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
360 KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
362 TAILQ_REMOVE(&so->so_comp, sp, so_list);
364 soclrstate(sp, SS_COMP);
365 soabort_async(sp, TRUE);
368 lwkt_relpooltoken(so);
372 * Destroy a disconnected socket. This routine is a NOP if entities
373 * still have a reference on the socket:
375 * so_pcb - The protocol stack still has a reference
376 * SS_NOFDREF - There is no longer a file pointer reference
379 sofree(struct socket *so)
384 * This is a bit hackish at the moment. We need to interlock
385 * any accept queue we are on before we potentially lose the
386 * last reference to avoid races against a re-reference from
387 * someone operating on the queue.
389 while ((head = so->so_head) != NULL) {
390 lwkt_getpooltoken(head);
391 if (so->so_head == head)
393 lwkt_relpooltoken(head);
397 * Arbitrage the last free.
399 KKASSERT(so->so_refs > 0);
400 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
402 lwkt_relpooltoken(head);
406 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
407 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
411 * We're done, remove ourselves from the accept queue we are
412 * on, if we are on one.
414 if (so->so_state & SS_INCOMP) {
415 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) ==
417 TAILQ_REMOVE(&head->so_incomp, so, so_list);
419 } else if (so->so_state & SS_COMP) {
421 * We must not decommission a socket that's
422 * on the accept(2) queue. If we do, then
423 * accept(2) may hang after select(2) indicated
424 * that the listening socket was ready.
426 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) ==
428 lwkt_relpooltoken(head);
431 panic("sofree: not queued");
433 soclrstate(so, SS_INCOMP);
435 lwkt_relpooltoken(head);
437 /* Flush accept queues, if we are accepting. */
440 ssb_release(&so->so_snd, so);
446 * Close a socket on last file table reference removal.
447 * Initiate disconnect if connected.
448 * Free socket when disconnect complete.
451 soclose(struct socket *so, int fflag)
455 funsetown(&so->so_sigio);
456 sosetstate(so, SS_ISCLOSING);
457 if (!use_soclose_fast ||
458 (so->so_proto->pr_flags & PR_SYNC_PORT) ||
459 ((so->so_state & SS_ISCONNECTED) &&
460 (so->so_options & SO_LINGER) &&
461 so->so_linger != 0)) {
462 error = soclose_sync(so, fflag);
471 sodiscard(struct socket *so)
473 if (so->so_state & SS_NOFDREF)
474 panic("soclose: NOFDREF");
475 sosetstate(so, SS_NOFDREF); /* take ref */
479 * Append the completed queue of head to head_inh (inherting listen socket).
482 soinherit(struct socket *head, struct socket *head_inh)
484 boolean_t do_wakeup = FALSE;
486 KASSERT(head->so_options & SO_ACCEPTCONN,
487 ("head does not accept connection"));
488 KASSERT(head_inh->so_options & SO_ACCEPTCONN,
489 ("head_inh does not accept connection"));
491 lwkt_getpooltoken(head);
492 lwkt_getpooltoken(head_inh);
494 if (head->so_qlen > 0)
497 while (!TAILQ_EMPTY(&head->so_comp)) {
498 struct ucred *old_cr;
501 sp = TAILQ_FIRST(&head->so_comp);
502 KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP);
505 * Remove this socket from the current listen socket
508 TAILQ_REMOVE(&head->so_comp, sp, so_list);
511 /* Save the old ucred for later free. */
512 old_cr = sp->so_cred;
515 * Install this socket to the inheriting listen socket
518 sp->so_cred = crhold(head_inh->so_cred); /* non-blocking */
519 sp->so_head = head_inh;
521 TAILQ_INSERT_TAIL(&head_inh->so_comp, sp, so_list);
526 * crfree() may block and release the tokens temporarily.
527 * However, we are fine here, since the transition is done.
532 lwkt_relpooltoken(head_inh);
533 lwkt_relpooltoken(head);
537 * "New" connections have arrived
540 wakeup(&head_inh->so_timeo);
545 soclose_sync(struct socket *so, int fflag)
549 if ((so->so_proto->pr_flags & PR_SYNC_PORT) == 0)
550 so_pru_sync(so); /* unpend async prus */
552 if (so->so_pcb == NULL)
555 if (so->so_state & SS_ISCONNECTED) {
556 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
557 error = sodisconnect(so);
561 if (so->so_options & SO_LINGER) {
562 if ((so->so_state & SS_ISDISCONNECTING) &&
565 while (so->so_state & SS_ISCONNECTED) {
566 error = tsleep(&so->so_timeo, PCATCH,
567 "soclos", so->so_linger * hz);
577 error2 = so_pru_detach(so);
578 if (error2 == EJUSTRETURN) {
580 * Protocol will call sodiscard()
581 * and sofree() for us.
590 sofree(so); /* dispose of ref */
596 soclose_fast_handler(netmsg_t msg)
598 struct socket *so = msg->base.nm_so;
600 if (so->so_pcb == NULL)
603 if ((so->so_state & SS_ISCONNECTED) &&
604 (so->so_state & SS_ISDISCONNECTING) == 0)
605 so_pru_disconnect_direct(so);
610 error = so_pru_detach_direct(so);
611 if (error == EJUSTRETURN) {
613 * Protocol will call sodiscard()
614 * and sofree() for us.
625 soclose_fast(struct socket *so)
627 struct netmsg_base *base = &so->so_clomsg;
629 netmsg_init(base, so, &netisr_apanic_rport, 0,
630 soclose_fast_handler);
631 if (so->so_port == netisr_curport())
632 lwkt_sendmsg_oncpu(so->so_port, &base->lmsg);
634 lwkt_sendmsg(so->so_port, &base->lmsg);
638 * Abort and destroy a socket. Only one abort can be in progress
639 * at any given moment.
642 soabort_async(struct socket *so, boolean_t clr_head)
645 * Keep a reference before clearing the so_head
646 * to avoid racing socket close in netisr.
651 so_pru_abort_async(so);
655 soabort_direct(struct socket *so)
658 so_pru_abort_direct(so);
662 * so is passed in ref'd, which becomes owned by
663 * the cleared SS_NOFDREF flag.
666 soaccept_generic(struct socket *so)
668 if ((so->so_state & SS_NOFDREF) == 0)
669 panic("soaccept: !NOFDREF");
670 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */
674 soaccept(struct socket *so, struct sockaddr **nam)
678 soaccept_generic(so);
679 error = so_pru_accept(so, nam);
684 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td,
689 if (so->so_options & SO_ACCEPTCONN)
692 * If protocol is connection-based, can only connect once.
693 * Otherwise, if connected, try to disconnect first.
694 * This allows user to disconnect by connecting to, e.g.,
697 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
698 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
699 (error = sodisconnect(so)))) {
703 * Prevent accumulated error from previous connection
707 if (!sync && so->so_proto->pr_usrreqs->pru_preconnect)
708 error = so_pru_connect_async(so, nam, td);
710 error = so_pru_connect(so, nam, td);
716 soconnect2(struct socket *so1, struct socket *so2)
720 error = so_pru_connect2(so1, so2);
725 sodisconnect(struct socket *so)
729 if ((so->so_state & SS_ISCONNECTED) == 0) {
733 if (so->so_state & SS_ISDISCONNECTING) {
737 error = so_pru_disconnect(so);
742 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
745 * If send must go all at once and message is larger than
746 * send buffering, then hard error.
747 * Lock against other senders.
748 * If must go all at once and not enough room now, then
749 * inform user that this would block and do nothing.
750 * Otherwise, if nonblocking, send as much as possible.
751 * The data to be sent is described by "uio" if nonzero,
752 * otherwise by the mbuf chain "top" (which must be null
753 * if uio is not). Data provided in mbuf chain must be small
754 * enough to send all at once.
756 * Returns nonzero on error, timeout or signal; callers
757 * must check for short counts if EINTR/ERESTART are returned.
758 * Data and control buffers are freed on return.
761 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
762 struct mbuf *top, struct mbuf *control, int flags,
769 int clen = 0, error, dontroute, mlen;
770 int atomic = sosendallatonce(so) || top;
774 resid = uio->uio_resid;
776 resid = (size_t)top->m_pkthdr.len;
779 for (m = top; m; m = m->m_next)
781 KKASSERT(top->m_pkthdr.len == len);
786 * WARNING! resid is unsigned, space and len are signed. space
787 * can wind up negative if the sockbuf is overcommitted.
789 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
790 * type sockets since that's an error.
792 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
798 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
799 (so->so_proto->pr_flags & PR_ATOMIC);
800 if (td->td_lwp != NULL)
801 td->td_lwp->lwp_ru.ru_msgsnd++;
803 clen = control->m_len;
804 #define gotoerr(errcode) { error = errcode; goto release; }
807 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
812 if (so->so_state & SS_CANTSENDMORE)
815 error = so->so_error;
819 if ((so->so_state & SS_ISCONNECTED) == 0) {
821 * `sendto' and `sendmsg' is allowed on a connection-
822 * based socket if it supports implied connect.
823 * Return ENOTCONN if not connected and no address is
826 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
827 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
828 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
829 !(resid == 0 && clen != 0))
831 } else if (addr == NULL)
832 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
833 ENOTCONN : EDESTADDRREQ);
835 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
836 clen > so->so_snd.ssb_hiwat) {
839 space = ssb_space(&so->so_snd);
842 if ((space < 0 || (size_t)space < resid + clen) && uio &&
843 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
844 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
845 gotoerr(EWOULDBLOCK);
846 ssb_unlock(&so->so_snd);
847 error = ssb_wait(&so->so_snd);
857 * Data is prepackaged in "top".
861 top->m_flags |= M_EOR;
865 m = m_getl((int)resid, M_WAITOK, MT_DATA,
866 top == NULL ? M_PKTHDR : 0, &mlen);
869 m->m_pkthdr.rcvif = NULL;
871 len = imin((int)szmin(mlen, resid), space);
872 if (resid < MINCLSIZE) {
874 * For datagram protocols, leave room
875 * for protocol headers in first mbuf.
877 if (atomic && top == NULL && len < mlen)
881 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
882 resid = uio->uio_resid;
885 top->m_pkthdr.len += len;
891 top->m_flags |= M_EOR;
894 } while (space > 0 && atomic);
896 so->so_options |= SO_DONTROUTE;
897 if (flags & MSG_OOB) {
898 pru_flags = PRUS_OOB;
899 } else if ((flags & MSG_EOF) &&
900 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
903 * If the user set MSG_EOF, the protocol
904 * understands this flag and nothing left to
905 * send then use PRU_SEND_EOF instead of PRU_SEND.
907 pru_flags = PRUS_EOF;
908 } else if (resid > 0 && space > 0) {
909 /* If there is more to send, set PRUS_MORETOCOME */
910 pru_flags = PRUS_MORETOCOME;
915 * XXX all the SS_CANTSENDMORE checks previously
916 * done could be out of date. We could have recieved
917 * a reset packet in an interrupt or maybe we slept
918 * while doing page faults in uiomove() etc. We could
919 * probably recheck again inside the splnet() protection
920 * here, but there are probably other places that this
921 * also happens. We must rethink this.
923 error = so_pru_send(so, pru_flags, top, addr, control, td);
925 so->so_options &= ~SO_DONTROUTE;
932 } while (resid && space > 0);
936 ssb_unlock(&so->so_snd);
947 * A specialization of sosend() for UDP based on protocol-specific knowledge:
948 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
949 * sosendallatonce() returns true,
950 * the "atomic" variable is true,
951 * and sosendudp() blocks until space is available for the entire send.
952 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
953 * PR_IMPLOPCL flags set.
954 * UDP has no out-of-band data.
955 * UDP has no control data.
956 * UDP does not support MSG_EOR.
959 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
960 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
963 int error, pru_flags = 0;
966 if (td->td_lwp != NULL)
967 td->td_lwp->lwp_ru.ru_msgsnd++;
971 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
972 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
975 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
979 if (so->so_state & SS_CANTSENDMORE)
982 error = so->so_error;
986 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
987 gotoerr(EDESTADDRREQ);
988 if (resid > so->so_snd.ssb_hiwat)
990 space = ssb_space(&so->so_snd);
991 if (uio && (space < 0 || (size_t)space < resid)) {
992 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
993 gotoerr(EWOULDBLOCK);
994 ssb_unlock(&so->so_snd);
995 error = ssb_wait(&so->so_snd);
1002 int hdrlen = max_hdr;
1005 * We try to optimize out the additional mbuf
1006 * allocations in M_PREPEND() on output path, e.g.
1007 * - udp_output(), when it tries to prepend protocol
1009 * - Link layer output function, when it tries to
1010 * prepend link layer header.
1012 * This probably will not benefit any data that will
1013 * be fragmented, so this optimization is only performed
1014 * when the size of data and max size of protocol+link
1015 * headers fit into one mbuf cluster.
1017 if (uio->uio_resid > MCLBYTES - hdrlen ||
1018 !udp_sosend_prepend) {
1019 top = m_uiomove(uio);
1025 top = m_getl(uio->uio_resid + hdrlen, M_WAITOK,
1026 MT_DATA, M_PKTHDR, &nsize);
1027 KASSERT(nsize >= uio->uio_resid + hdrlen,
1028 ("sosendudp invalid nsize %d, "
1029 "resid %zu, hdrlen %d",
1030 nsize, uio->uio_resid, hdrlen));
1032 top->m_len = uio->uio_resid;
1033 top->m_pkthdr.len = uio->uio_resid;
1034 top->m_data += hdrlen;
1036 error = uiomove(mtod(top, caddr_t), top->m_len, uio);
1042 if (flags & MSG_DONTROUTE)
1043 pru_flags |= PRUS_DONTROUTE;
1045 if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
1046 so_pru_send_async(so, pru_flags, top, addr, NULL, td);
1049 error = so_pru_send(so, pru_flags, top, addr, NULL, td);
1051 top = NULL; /* sent or freed in lower layer */
1054 ssb_unlock(&so->so_snd);
1062 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1063 struct mbuf *top, struct mbuf *control, int flags,
1075 KKASSERT(top == NULL);
1077 resid = uio->uio_resid;
1080 resid = (size_t)top->m_pkthdr.len;
1083 for (m = top; m; m = m->m_next)
1085 KKASSERT(top->m_pkthdr.len == len);
1090 * WARNING! resid is unsigned, space and len are signed. space
1091 * can wind up negative if the sockbuf is overcommitted.
1093 * Also check to make sure that MSG_EOR isn't used on TCP
1095 if (flags & MSG_EOR) {
1101 /* TCP doesn't do control messages (rights, creds, etc) */
1102 if (control->m_len) {
1106 m_freem(control); /* empty control, just free it */
1110 if (td->td_lwp != NULL)
1111 td->td_lwp->lwp_ru.ru_msgsnd++;
1113 #define gotoerr(errcode) { error = errcode; goto release; }
1116 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1121 if (so->so_state & SS_CANTSENDMORE)
1124 error = so->so_error;
1128 if ((so->so_state & SS_ISCONNECTED) == 0 &&
1129 (so->so_state & SS_ISCONFIRMING) == 0)
1131 if (allatonce && resid > so->so_snd.ssb_hiwat)
1134 space = ssb_space_prealloc(&so->so_snd);
1135 if (flags & MSG_OOB)
1137 if ((space < 0 || (size_t)space < resid) && !allatonce &&
1138 space < so->so_snd.ssb_lowat) {
1139 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1140 gotoerr(EWOULDBLOCK);
1141 ssb_unlock(&so->so_snd);
1142 error = ssb_wait(&so->so_snd);
1149 int cnt = 0, async = 0;
1153 * Data is prepackaged in "top".
1157 if (resid > INT_MAX)
1159 if (tcp_sosend_jcluster) {
1160 m = m_getlj((int)resid, M_WAITOK, MT_DATA,
1161 top == NULL ? M_PKTHDR : 0, &mlen);
1163 m = m_getl((int)resid, M_WAITOK, MT_DATA,
1164 top == NULL ? M_PKTHDR : 0, &mlen);
1167 m->m_pkthdr.len = 0;
1168 m->m_pkthdr.rcvif = NULL;
1170 len = imin((int)szmin(mlen, resid), space);
1172 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1173 resid = uio->uio_resid;
1176 top->m_pkthdr.len += len;
1183 } while (space > 0 && cnt < tcp_sosend_agglim);
1185 if (tcp_sosend_async)
1188 if (flags & MSG_OOB) {
1189 pru_flags = PRUS_OOB;
1191 } else if ((flags & MSG_EOF) && resid == 0) {
1192 pru_flags = PRUS_EOF;
1193 } else if (resid > 0 && space > 0) {
1194 /* If there is more to send, set PRUS_MORETOCOME */
1195 pru_flags = PRUS_MORETOCOME;
1201 if (flags & MSG_SYNC)
1205 * XXX all the SS_CANTSENDMORE checks previously
1206 * done could be out of date. We could have recieved
1207 * a reset packet in an interrupt or maybe we slept
1208 * while doing page faults in uiomove() etc. We could
1209 * probably recheck again inside the splnet() protection
1210 * here, but there are probably other places that this
1211 * also happens. We must rethink this.
1213 for (m = top; m; m = m->m_next)
1214 ssb_preallocstream(&so->so_snd, m);
1216 error = so_pru_send(so, pru_flags, top,
1219 so_pru_send_async(so, pru_flags, top,
1228 } while (resid && space > 0);
1232 ssb_unlock(&so->so_snd);
1243 * Implement receive operations on a socket.
1245 * We depend on the way that records are added to the signalsockbuf
1246 * by sbappend*. In particular, each record (mbufs linked through m_next)
1247 * must begin with an address if the protocol so specifies,
1248 * followed by an optional mbuf or mbufs containing ancillary data,
1249 * and then zero or more mbufs of data.
1251 * Although the signalsockbuf is locked, new data may still be appended.
1252 * A token inside the ssb_lock deals with MP issues and still allows
1253 * the network to access the socket if we block in a uio.
1255 * The caller may receive the data as a single mbuf chain by supplying
1256 * an mbuf **mp0 for use in returning the chain. The uio is then used
1257 * only for the count in uio_resid.
1260 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1261 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1264 struct mbuf *free_chain = NULL;
1265 int flags, len, error, offset;
1266 struct protosw *pr = so->so_proto;
1268 size_t resid, orig_resid;
1269 boolean_t free_rights = FALSE;
1272 resid = uio->uio_resid;
1274 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1282 flags = *flagsp &~ MSG_EOR;
1285 if (flags & MSG_OOB) {
1286 m = m_get(M_WAITOK, MT_DATA);
1289 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1295 KKASSERT(resid >= (size_t)m->m_len);
1296 resid -= (size_t)m->m_len;
1297 } while (resid > 0 && m);
1300 uio->uio_resid = resid;
1301 error = uiomove(mtod(m, caddr_t),
1302 (int)szmin(resid, m->m_len),
1304 resid = uio->uio_resid;
1306 } while (uio->uio_resid && error == 0 && m);
1313 if ((so->so_state & SS_ISCONFIRMING) && resid)
1317 * The token interlocks against the protocol thread while
1318 * ssb_lock is a blocking lock against other userland entities.
1320 lwkt_gettoken(&so->so_rcv.ssb_token);
1322 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1326 m = so->so_rcv.ssb_mb;
1328 * If we have less data than requested, block awaiting more
1329 * (subject to any timeout) if:
1330 * 1. the current count is less than the low water mark, or
1331 * 2. MSG_WAITALL is set, and it is possible to do the entire
1332 * receive operation at once if we block (resid <= hiwat).
1333 * 3. MSG_DONTWAIT is not set
1334 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1335 * we have to do the receive in sections, and thus risk returning
1336 * a short count if a timeout or signal occurs after we start.
1338 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1339 (size_t)so->so_rcv.ssb_cc < resid) &&
1340 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1341 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1342 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1343 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1347 error = so->so_error;
1348 if ((flags & MSG_PEEK) == 0)
1352 if (so->so_state & SS_CANTRCVMORE) {
1358 for (; m; m = m->m_next) {
1359 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1360 m = so->so_rcv.ssb_mb;
1364 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1365 (pr->pr_flags & PR_CONNREQUIRED)) {
1371 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1372 error = EWOULDBLOCK;
1375 ssb_unlock(&so->so_rcv);
1376 error = ssb_wait(&so->so_rcv);
1382 if (uio && uio->uio_td && uio->uio_td->td_proc)
1383 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1386 * note: m should be == sb_mb here. Cache the next record while
1387 * cleaning up. Note that calling m_free*() will break out critical
1390 KKASSERT(m == so->so_rcv.ssb_mb);
1393 * Skip any address mbufs prepending the record.
1395 if (pr->pr_flags & PR_ADDR) {
1396 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1399 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1400 if (flags & MSG_PEEK)
1403 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1407 * Skip any control mbufs prepending the record.
1409 while (m && m->m_type == MT_CONTROL && error == 0) {
1410 if (flags & MSG_PEEK) {
1412 *controlp = m_copy(m, 0, m->m_len);
1413 m = m->m_next; /* XXX race */
1415 const struct cmsghdr *cm = mtod(m, struct cmsghdr *);
1418 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1419 if (pr->pr_domain->dom_externalize &&
1420 cm->cmsg_level == SOL_SOCKET &&
1421 cm->cmsg_type == SCM_RIGHTS) {
1422 error = pr->pr_domain->dom_externalize
1428 if (cm->cmsg_level == SOL_SOCKET &&
1429 cm->cmsg_type == SCM_RIGHTS)
1431 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1434 if (controlp && *controlp) {
1436 controlp = &(*controlp)->m_next;
1445 if (type == MT_OOBDATA)
1450 * Copy to the UIO or mbuf return chain (*mp).
1454 while (m && resid > 0 && error == 0) {
1455 if (m->m_type == MT_OOBDATA) {
1456 if (type != MT_OOBDATA)
1458 } else if (type == MT_OOBDATA)
1461 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1463 soclrstate(so, SS_RCVATMARK);
1464 len = (resid > INT_MAX) ? INT_MAX : resid;
1465 if (so->so_oobmark && len > so->so_oobmark - offset)
1466 len = so->so_oobmark - offset;
1467 if (len > m->m_len - moff)
1468 len = m->m_len - moff;
1471 * Copy out to the UIO or pass the mbufs back to the SIO.
1472 * The SIO is dealt with when we eat the mbuf, but deal
1473 * with the resid here either way.
1476 uio->uio_resid = resid;
1477 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1478 resid = uio->uio_resid;
1482 resid -= (size_t)len;
1486 * Eat the entire mbuf or just a piece of it
1488 if (len == m->m_len - moff) {
1489 if (m->m_flags & M_EOR)
1491 if (flags & MSG_PEEK) {
1496 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1500 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1504 if (flags & MSG_PEEK) {
1508 n = m_copym(m, 0, len, M_WAITOK);
1514 so->so_rcv.ssb_cc -= len;
1517 if (so->so_oobmark) {
1518 if ((flags & MSG_PEEK) == 0) {
1519 so->so_oobmark -= len;
1520 if (so->so_oobmark == 0) {
1521 sosetstate(so, SS_RCVATMARK);
1526 if (offset == so->so_oobmark)
1530 if (flags & MSG_EOR)
1533 * If the MSG_WAITALL flag is set (for non-atomic socket),
1534 * we must not quit until resid == 0 or an error
1535 * termination. If a signal/timeout occurs, return
1536 * with a short count but without error.
1537 * Keep signalsockbuf locked against other readers.
1539 while ((flags & MSG_WAITALL) && m == NULL &&
1540 resid > 0 && !sosendallatonce(so) &&
1541 so->so_rcv.ssb_mb == NULL) {
1542 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1545 * The window might have closed to zero, make
1546 * sure we send an ack now that we've drained
1547 * the buffer or we might end up blocking until
1548 * the idle takes over (5 seconds).
1550 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1551 so_pru_rcvd(so, flags);
1552 error = ssb_wait(&so->so_rcv);
1554 ssb_unlock(&so->so_rcv);
1558 m = so->so_rcv.ssb_mb;
1563 * If an atomic read was requested but unread data still remains
1564 * in the record, set MSG_TRUNC.
1566 if (m && pr->pr_flags & PR_ATOMIC)
1570 * Cleanup. If an atomic read was requested drop any unread data.
1572 if ((flags & MSG_PEEK) == 0) {
1573 if (m && (pr->pr_flags & PR_ATOMIC))
1574 sbdroprecord(&so->so_rcv.sb);
1575 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1576 so_pru_rcvd(so, flags);
1579 if (orig_resid == resid && orig_resid &&
1580 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1581 ssb_unlock(&so->so_rcv);
1588 ssb_unlock(&so->so_rcv);
1590 lwkt_reltoken(&so->so_rcv.ssb_token);
1592 if (free_rights && (pr->pr_flags & PR_RIGHTS) &&
1593 pr->pr_domain->dom_dispose)
1594 pr->pr_domain->dom_dispose(free_chain);
1595 m_freem(free_chain);
1601 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1602 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1605 struct mbuf *free_chain = NULL;
1606 int flags, len, error, offset;
1607 struct protosw *pr = so->so_proto;
1610 size_t resid, orig_resid, restmp;
1613 resid = uio->uio_resid;
1615 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1623 flags = *flagsp &~ MSG_EOR;
1626 if (flags & MSG_OOB) {
1627 m = m_get(M_WAITOK, MT_DATA);
1630 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1636 KKASSERT(resid >= (size_t)m->m_len);
1637 resid -= (size_t)m->m_len;
1638 } while (resid > 0 && m);
1641 uio->uio_resid = resid;
1642 error = uiomove(mtod(m, caddr_t),
1643 (int)szmin(resid, m->m_len),
1645 resid = uio->uio_resid;
1647 } while (uio->uio_resid && error == 0 && m);
1656 * The token interlocks against the protocol thread while
1657 * ssb_lock is a blocking lock against other userland entities.
1659 * Lock a limited number of mbufs (not all, so sbcompress() still
1660 * works well). The token is used as an interlock for sbwait() so
1661 * release it afterwords.
1664 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1668 lwkt_gettoken(&so->so_rcv.ssb_token);
1669 m = so->so_rcv.ssb_mb;
1672 * If we have less data than requested, block awaiting more
1673 * (subject to any timeout) if:
1674 * 1. the current count is less than the low water mark, or
1675 * 2. MSG_WAITALL is set, and it is possible to do the entire
1676 * receive operation at once if we block (resid <= hiwat).
1677 * 3. MSG_DONTWAIT is not set
1678 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1679 * we have to do the receive in sections, and thus risk returning
1680 * a short count if a timeout or signal occurs after we start.
1682 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1683 (size_t)so->so_rcv.ssb_cc < resid) &&
1684 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1685 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1686 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1690 lwkt_reltoken(&so->so_rcv.ssb_token);
1691 error = so->so_error;
1692 if ((flags & MSG_PEEK) == 0)
1696 if (so->so_state & SS_CANTRCVMORE) {
1699 lwkt_reltoken(&so->so_rcv.ssb_token);
1702 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1703 (pr->pr_flags & PR_CONNREQUIRED)) {
1704 lwkt_reltoken(&so->so_rcv.ssb_token);
1709 lwkt_reltoken(&so->so_rcv.ssb_token);
1712 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1713 lwkt_reltoken(&so->so_rcv.ssb_token);
1714 error = EWOULDBLOCK;
1717 ssb_unlock(&so->so_rcv);
1718 error = ssb_wait(&so->so_rcv);
1719 lwkt_reltoken(&so->so_rcv.ssb_token);
1731 while (n && restmp < resid) {
1732 n->m_flags |= M_SOLOCKED;
1734 if (n->m_next == NULL)
1741 * Release token for loop
1743 lwkt_reltoken(&so->so_rcv.ssb_token);
1744 if (uio && uio->uio_td && uio->uio_td->td_proc)
1745 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1748 * note: m should be == sb_mb here. Cache the next record while
1749 * cleaning up. Note that calling m_free*() will break out critical
1752 KKASSERT(m == so->so_rcv.ssb_mb);
1755 * Copy to the UIO or mbuf return chain (*mp).
1757 * NOTE: Token is not held for loop
1763 while (m && (m->m_flags & M_SOLOCKED) && resid > 0 && error == 0) {
1764 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1767 soclrstate(so, SS_RCVATMARK);
1768 len = (resid > INT_MAX) ? INT_MAX : resid;
1769 if (so->so_oobmark && len > so->so_oobmark - offset)
1770 len = so->so_oobmark - offset;
1771 if (len > m->m_len - moff)
1772 len = m->m_len - moff;
1775 * Copy out to the UIO or pass the mbufs back to the SIO.
1776 * The SIO is dealt with when we eat the mbuf, but deal
1777 * with the resid here either way.
1780 uio->uio_resid = resid;
1781 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1782 resid = uio->uio_resid;
1786 resid -= (size_t)len;
1790 * Eat the entire mbuf or just a piece of it
1793 if (len == m->m_len - moff) {
1803 if (so->so_oobmark && offset == so->so_oobmark) {
1810 * Synchronize sockbuf with data we read.
1812 * NOTE: (m) is junk on entry (it could be left over from the
1815 if ((flags & MSG_PEEK) == 0) {
1816 lwkt_gettoken(&so->so_rcv.ssb_token);
1817 m = so->so_rcv.ssb_mb;
1818 while (m && offset >= m->m_len) {
1819 if (so->so_oobmark) {
1820 so->so_oobmark -= m->m_len;
1821 if (so->so_oobmark == 0) {
1822 sosetstate(so, SS_RCVATMARK);
1828 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1832 m = sbunlinkmbuf(&so->so_rcv.sb,
1839 n = m_copym(m, 0, offset, M_WAITOK);
1843 m->m_data += offset;
1845 so->so_rcv.ssb_cc -= offset;
1846 if (so->so_oobmark) {
1847 so->so_oobmark -= offset;
1848 if (so->so_oobmark == 0) {
1849 sosetstate(so, SS_RCVATMARK);
1855 lwkt_reltoken(&so->so_rcv.ssb_token);
1859 * If the MSG_WAITALL flag is set (for non-atomic socket),
1860 * we must not quit until resid == 0 or an error termination.
1862 * If a signal/timeout occurs, return with a short count but without
1865 * Keep signalsockbuf locked against other readers.
1867 * XXX if MSG_PEEK we currently do quit.
1869 if ((flags & MSG_WAITALL) && !(flags & MSG_PEEK) &&
1870 didoob == 0 && resid > 0 &&
1871 !sosendallatonce(so)) {
1872 lwkt_gettoken(&so->so_rcv.ssb_token);
1874 while ((m = so->so_rcv.ssb_mb) == NULL) {
1875 if (so->so_error || (so->so_state & SS_CANTRCVMORE)) {
1876 error = so->so_error;
1880 * The window might have closed to zero, make
1881 * sure we send an ack now that we've drained
1882 * the buffer or we might end up blocking until
1883 * the idle takes over (5 seconds).
1886 so_pru_rcvd_async(so);
1887 if (so->so_rcv.ssb_mb == NULL)
1888 error = ssb_wait(&so->so_rcv);
1890 lwkt_reltoken(&so->so_rcv.ssb_token);
1891 ssb_unlock(&so->so_rcv);
1896 if (m && error == 0)
1898 lwkt_reltoken(&so->so_rcv.ssb_token);
1902 * Token not held here.
1904 * Cleanup. If an atomic read was requested drop any unread data XXX
1906 if ((flags & MSG_PEEK) == 0) {
1908 so_pru_rcvd_async(so);
1911 if (orig_resid == resid && orig_resid &&
1912 (so->so_state & SS_CANTRCVMORE) == 0) {
1913 ssb_unlock(&so->so_rcv);
1920 ssb_unlock(&so->so_rcv);
1923 m_freem(free_chain);
1928 * Shut a socket down. Note that we do not get a frontend lock as we
1929 * want to be able to shut the socket down even if another thread is
1930 * blocked in a read(), thus waking it up.
1933 soshutdown(struct socket *so, int how)
1935 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1938 if (how != SHUT_WR) {
1939 /*ssb_lock(&so->so_rcv, M_WAITOK);*/
1941 /*ssb_unlock(&so->so_rcv);*/
1944 return (so_pru_shutdown(so));
1949 sorflush(struct socket *so)
1951 struct signalsockbuf *ssb = &so->so_rcv;
1952 struct protosw *pr = so->so_proto;
1953 struct signalsockbuf asb;
1955 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1957 lwkt_gettoken(&ssb->ssb_token);
1962 * Can't just blow up the ssb structure here
1964 bzero(&ssb->sb, sizeof(ssb->sb));
1969 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1971 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1972 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1973 ssb_release(&asb, so);
1975 lwkt_reltoken(&ssb->ssb_token);
1980 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1982 struct accept_filter_arg *afap = NULL;
1983 struct accept_filter *afp;
1984 struct so_accf *af = so->so_accf;
1987 /* do not set/remove accept filters on non listen sockets */
1988 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1993 /* removing the filter */
1996 if (af->so_accept_filter != NULL &&
1997 af->so_accept_filter->accf_destroy != NULL) {
1998 af->so_accept_filter->accf_destroy(so);
2000 if (af->so_accept_filter_str != NULL) {
2001 kfree(af->so_accept_filter_str, M_ACCF);
2006 so->so_options &= ~SO_ACCEPTFILTER;
2009 /* adding a filter */
2010 /* must remove previous filter first */
2015 /* don't put large objects on the kernel stack */
2016 afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
2017 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
2018 afap->af_name[sizeof(afap->af_name)-1] = '\0';
2019 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
2022 afp = accept_filt_get(afap->af_name);
2027 af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
2028 if (afp->accf_create != NULL) {
2029 if (afap->af_name[0] != '\0') {
2030 int len = strlen(afap->af_name) + 1;
2032 af->so_accept_filter_str = kmalloc(len, M_ACCF,
2034 strcpy(af->so_accept_filter_str, afap->af_name);
2036 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
2037 if (af->so_accept_filter_arg == NULL) {
2038 kfree(af->so_accept_filter_str, M_ACCF);
2045 af->so_accept_filter = afp;
2047 so->so_options |= SO_ACCEPTFILTER;
2050 kfree(afap, M_TEMP);
2056 * Perhaps this routine, and sooptcopyout(), below, ought to come in
2057 * an additional variant to handle the case where the option value needs
2058 * to be some kind of integer, but not a specific size.
2059 * In addition to their use here, these functions are also called by the
2060 * protocol-level pr_ctloutput() routines.
2063 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2065 return soopt_to_kbuf(sopt, buf, len, minlen);
2069 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2073 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2074 KKASSERT(kva_p(buf));
2077 * If the user gives us more than we wanted, we ignore it,
2078 * but if we don't get the minimum length the caller
2079 * wants, we return EINVAL. On success, sopt->sopt_valsize
2080 * is set to however much we actually retrieved.
2082 if ((valsize = sopt->sopt_valsize) < minlen)
2085 sopt->sopt_valsize = valsize = len;
2087 bcopy(sopt->sopt_val, buf, valsize);
2093 sosetopt(struct socket *so, struct sockopt *sopt)
2099 struct signalsockbuf *sotmp;
2102 sopt->sopt_dir = SOPT_SET;
2103 if (sopt->sopt_level != SOL_SOCKET) {
2104 if (so->so_proto && so->so_proto->pr_ctloutput) {
2105 return (so_pr_ctloutput(so, sopt));
2107 error = ENOPROTOOPT;
2109 switch (sopt->sopt_name) {
2111 case SO_ACCEPTFILTER:
2112 error = do_setopt_accept_filter(so, sopt);
2118 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2122 so->so_linger = l.l_linger;
2124 so->so_options |= SO_LINGER;
2126 so->so_options &= ~SO_LINGER;
2132 case SO_USELOOPBACK:
2139 error = sooptcopyin(sopt, &optval, sizeof optval,
2144 so->so_options |= sopt->sopt_name;
2146 so->so_options &= ~sopt->sopt_name;
2153 error = sooptcopyin(sopt, &optval, sizeof optval,
2159 * Values < 1 make no sense for any of these
2160 * options, so disallow them.
2167 switch (sopt->sopt_name) {
2170 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2171 &so->so_snd : &so->so_rcv, (u_long)optval,
2173 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2177 sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2178 &so->so_snd : &so->so_rcv;
2179 atomic_clear_int(&sotmp->ssb_flags,
2184 * Make sure the low-water is never greater than
2188 so->so_snd.ssb_lowat =
2189 (optval > so->so_snd.ssb_hiwat) ?
2190 so->so_snd.ssb_hiwat : optval;
2191 atomic_clear_int(&so->so_snd.ssb_flags,
2195 so->so_rcv.ssb_lowat =
2196 (optval > so->so_rcv.ssb_hiwat) ?
2197 so->so_rcv.ssb_hiwat : optval;
2198 atomic_clear_int(&so->so_rcv.ssb_flags,
2206 error = sooptcopyin(sopt, &tv, sizeof tv,
2211 /* assert(hz > 0); */
2212 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2213 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2217 /* assert(tick > 0); */
2218 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2219 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2220 if (val > INT_MAX) {
2224 if (val == 0 && tv.tv_usec != 0)
2227 switch (sopt->sopt_name) {
2229 so->so_snd.ssb_timeo = val;
2232 so->so_rcv.ssb_timeo = val;
2237 error = ENOPROTOOPT;
2240 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2241 (void) so_pr_ctloutput(so, sopt);
2248 /* Helper routine for getsockopt */
2250 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2252 soopt_from_kbuf(sopt, buf, len);
2257 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2262 sopt->sopt_valsize = 0;
2266 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2267 KKASSERT(kva_p(buf));
2270 * Documented get behavior is that we always return a value,
2271 * possibly truncated to fit in the user's buffer.
2272 * Traditional behavior is that we always tell the user
2273 * precisely how much we copied, rather than something useful
2274 * like the total amount we had available for her.
2275 * Note that this interface is not idempotent; the entire answer must
2276 * generated ahead of time.
2278 valsize = szmin(len, sopt->sopt_valsize);
2279 sopt->sopt_valsize = valsize;
2280 if (sopt->sopt_val != 0) {
2281 bcopy(buf, sopt->sopt_val, valsize);
2286 sogetopt(struct socket *so, struct sockopt *sopt)
2293 struct accept_filter_arg *afap;
2297 sopt->sopt_dir = SOPT_GET;
2298 if (sopt->sopt_level != SOL_SOCKET) {
2299 if (so->so_proto && so->so_proto->pr_ctloutput) {
2300 return (so_pr_ctloutput(so, sopt));
2302 return (ENOPROTOOPT);
2304 switch (sopt->sopt_name) {
2306 case SO_ACCEPTFILTER:
2307 if ((so->so_options & SO_ACCEPTCONN) == 0)
2309 afap = kmalloc(sizeof(*afap), M_TEMP,
2311 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2312 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2313 if (so->so_accf->so_accept_filter_str != NULL)
2314 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2316 error = sooptcopyout(sopt, afap, sizeof(*afap));
2317 kfree(afap, M_TEMP);
2322 l.l_onoff = so->so_options & SO_LINGER;
2323 l.l_linger = so->so_linger;
2324 error = sooptcopyout(sopt, &l, sizeof l);
2327 case SO_USELOOPBACK:
2337 optval = so->so_options & sopt->sopt_name;
2339 error = sooptcopyout(sopt, &optval, sizeof optval);
2343 optval = so->so_type;
2347 optval = so->so_error;
2352 optval = so->so_snd.ssb_hiwat;
2356 optval = so->so_rcv.ssb_hiwat;
2360 optval = so->so_snd.ssb_lowat;
2364 optval = so->so_rcv.ssb_lowat;
2369 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2370 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2372 tv.tv_sec = optval / hz;
2373 tv.tv_usec = (optval % hz) * ustick;
2374 error = sooptcopyout(sopt, &tv, sizeof tv);
2378 optval_l = ssb_space(&so->so_snd);
2379 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2383 optval = -1; /* no hint */
2387 error = ENOPROTOOPT;
2390 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput)
2391 so_pr_ctloutput(so, sopt);
2396 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2398 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2400 struct mbuf *m, *m_prev;
2401 int sopt_size = sopt->sopt_valsize, msize;
2403 m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA,
2407 m->m_len = min(msize, sopt_size);
2408 sopt_size -= m->m_len;
2412 while (sopt_size > 0) {
2413 m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT,
2414 MT_DATA, 0, &msize);
2419 m->m_len = min(msize, sopt_size);
2420 sopt_size -= m->m_len;
2427 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2429 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2431 soopt_to_mbuf(sopt, m);
2436 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2441 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2443 if (sopt->sopt_val == NULL)
2445 val = sopt->sopt_val;
2446 valsize = sopt->sopt_valsize;
2447 while (m != NULL && valsize >= m->m_len) {
2448 bcopy(val, mtod(m, char *), m->m_len);
2449 valsize -= m->m_len;
2450 val = (caddr_t)val + m->m_len;
2453 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2454 panic("ip6_sooptmcopyin");
2457 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2459 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2461 return soopt_from_mbuf(sopt, m);
2465 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2467 struct mbuf *m0 = m;
2472 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2474 if (sopt->sopt_val == NULL)
2476 val = sopt->sopt_val;
2477 maxsize = sopt->sopt_valsize;
2478 while (m != NULL && maxsize >= m->m_len) {
2479 bcopy(mtod(m, char *), val, m->m_len);
2480 maxsize -= m->m_len;
2481 val = (caddr_t)val + m->m_len;
2482 valsize += m->m_len;
2486 /* enough soopt buffer should be given from user-land */
2490 sopt->sopt_valsize = valsize;
2495 sohasoutofband(struct socket *so)
2497 if (so->so_sigio != NULL)
2498 pgsigio(so->so_sigio, SIGURG, 0);
2501 * There is no need to use NOTE_OOB as KNOTE hint here:
2502 * soread filter depends on so_oobmark and SS_RCVATMARK
2503 * so_state. NOTE_OOB would cause unnecessary penalty
2504 * in KNOTE, if there was knote processing contention.
2506 KNOTE(&so->so_rcv.ssb_kq.ki_note, 0);
2510 sokqfilter(struct file *fp, struct knote *kn)
2512 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2513 struct signalsockbuf *ssb;
2515 switch (kn->kn_filter) {
2517 if (so->so_options & SO_ACCEPTCONN)
2518 kn->kn_fop = &solisten_filtops;
2520 kn->kn_fop = &soread_filtops;
2524 kn->kn_fop = &sowrite_filtops;
2528 kn->kn_fop = &soexcept_filtops;
2532 return (EOPNOTSUPP);
2535 knote_insert(&ssb->ssb_kq.ki_note, kn);
2536 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2541 filt_sordetach(struct knote *kn)
2543 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2545 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2546 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2547 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2552 filt_soread(struct knote *kn, long hint __unused)
2554 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2556 if (kn->kn_sfflags & NOTE_OOB) {
2557 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2558 kn->kn_fflags |= NOTE_OOB;
2563 kn->kn_data = so->so_rcv.ssb_cc;
2565 if (so->so_state & SS_CANTRCVMORE) {
2567 * Only set NODATA if all data has been exhausted.
2569 if (kn->kn_data == 0)
2570 kn->kn_flags |= EV_NODATA;
2571 kn->kn_flags |= EV_EOF;
2572 kn->kn_fflags = so->so_error;
2575 if (so->so_error) /* temporary udp error */
2577 if (kn->kn_sfflags & NOTE_LOWAT)
2578 return (kn->kn_data >= kn->kn_sdata);
2579 return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2580 !TAILQ_EMPTY(&so->so_comp));
2584 filt_sowdetach(struct knote *kn)
2586 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2588 knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2589 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2590 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2595 filt_sowrite(struct knote *kn, long hint __unused)
2597 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2599 if (so->so_snd.ssb_flags & SSB_PREALLOC)
2600 kn->kn_data = ssb_space_prealloc(&so->so_snd);
2602 kn->kn_data = ssb_space(&so->so_snd);
2604 if (so->so_state & SS_CANTSENDMORE) {
2605 kn->kn_flags |= (EV_EOF | EV_NODATA);
2606 kn->kn_fflags = so->so_error;
2609 if (so->so_error) /* temporary udp error */
2611 if (((so->so_state & SS_ISCONNECTED) == 0) &&
2612 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2614 if (kn->kn_sfflags & NOTE_LOWAT)
2615 return (kn->kn_data >= kn->kn_sdata);
2616 return (kn->kn_data >= so->so_snd.ssb_lowat);
2621 filt_solisten(struct knote *kn, long hint __unused)
2623 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2624 int qlen = so->so_qlen;
2626 if (soavailconn > 0 && qlen > soavailconn)
2630 return (!TAILQ_EMPTY(&so->so_comp));