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");
155 static int soavailconn = 32;
156 SYSCTL_INT(_kern_ipc, OID_AUTO, soavailconn, CTLFLAG_RW,
157 &soavailconn, 0, "Maximum available socket connection queue size");
160 * Socket operation routines.
161 * These routines are called by the routines in
162 * sys_socket.c or from a system process, and
163 * implement the semantics of socket operations by
164 * switching out to the protocol specific routines.
168 * Get a socket structure, and initialize it.
169 * Note that it would probably be better to allocate socket
170 * and PCB at the same time, but I'm not convinced that all
171 * the protocols can be easily modified to do this.
174 soalloc(int waitok, struct protosw *pr)
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;
201 socreate(int dom, struct socket **aso, int type,
202 int proto, struct thread *td)
204 struct proc *p = td->td_proc;
207 struct pru_attach_info ai;
211 prp = pffindproto(dom, proto, type);
213 prp = pffindtype(dom, type);
215 if (prp == NULL || prp->pr_usrreqs->pru_attach == 0)
216 return (EPROTONOSUPPORT);
218 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
219 prp->pr_domain->dom_family != PF_LOCAL &&
220 prp->pr_domain->dom_family != PF_INET &&
221 prp->pr_domain->dom_family != PF_INET6 &&
222 prp->pr_domain->dom_family != PF_ROUTE) {
223 return (EPROTONOSUPPORT);
226 if (prp->pr_type != type)
228 so = soalloc(p != NULL, prp);
233 * Callers of socreate() presumably will connect up a descriptor
234 * and call soclose() if they cannot. This represents our so_refs
235 * (which should be 1) from soalloc().
237 soclrstate(so, SS_NOFDREF);
240 * Set a default port for protocol processing. No action will occur
241 * on the socket on this port until an inpcb is attached to it and
242 * is able to match incoming packets, or until the socket becomes
243 * available to userland.
245 * We normally default the socket to the protocol thread on cpu 0,
246 * if protocol does not provide its own method to initialize the
249 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
250 * thread and all pr_*()/pru_*() calls are executed synchronously.
252 if (prp->pr_flags & PR_SYNC_PORT)
253 so->so_port = &netisr_sync_port;
254 else if (prp->pr_initport != NULL)
255 so->so_port = prp->pr_initport();
257 so->so_port = netisr_cpuport(0);
259 TAILQ_INIT(&so->so_incomp);
260 TAILQ_INIT(&so->so_comp);
262 so->so_cred = crhold(p->p_ucred);
263 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
264 ai.p_ucred = p->p_ucred;
265 ai.fd_rdir = p->p_fd->fd_rdir;
268 * Auto-sizing of socket buffers is managed by the protocols and
269 * the appropriate flags must be set in the pru_attach function.
271 if (use_socreate_fast && prp->pr_usrreqs->pru_preattach)
272 error = so_pru_attach_fast(so, proto, &ai);
274 error = so_pru_attach(so, proto, &ai);
276 sosetstate(so, SS_NOFDREF);
277 sofree(so); /* from soalloc */
282 * NOTE: Returns referenced socket.
289 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
293 error = so_pru_bind(so, nam, td);
298 sodealloc(struct socket *so)
300 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 0);
303 if (so->so_options & SO_ACCEPTCONN) {
304 KASSERT(TAILQ_EMPTY(&so->so_comp), ("so_comp is not empty"));
305 KASSERT(TAILQ_EMPTY(&so->so_incomp),
306 ("so_incomp is not empty"));
310 if (so->so_rcv.ssb_hiwat)
311 (void)chgsbsize(so->so_cred->cr_uidinfo,
312 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
313 if (so->so_snd.ssb_hiwat)
314 (void)chgsbsize(so->so_cred->cr_uidinfo,
315 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
317 /* remove accept filter if present */
318 if (so->so_accf != NULL)
319 do_setopt_accept_filter(so, NULL);
322 if (so->so_faddr != NULL)
323 kfree(so->so_faddr, M_SONAME);
328 solisten(struct socket *so, int backlog, struct thread *td)
330 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
333 lwkt_gettoken(&so->so_rcv.ssb_token);
334 if (TAILQ_EMPTY(&so->so_comp))
335 so->so_options |= SO_ACCEPTCONN;
336 lwkt_reltoken(&so->so_rcv.ssb_token);
337 if (backlog < 0 || backlog > somaxconn)
339 so->so_qlimit = backlog;
340 return so_pru_listen(so, td);
344 soqflush(struct socket *so)
346 lwkt_getpooltoken(so);
347 if (so->so_options & SO_ACCEPTCONN) {
350 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
351 KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
353 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
355 soclrstate(sp, SS_INCOMP);
356 soabort_async(sp, TRUE);
358 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
359 KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
361 TAILQ_REMOVE(&so->so_comp, sp, so_list);
363 soclrstate(sp, SS_COMP);
364 soabort_async(sp, TRUE);
367 lwkt_relpooltoken(so);
371 * Destroy a disconnected socket. This routine is a NOP if entities
372 * still have a reference on the socket:
374 * so_pcb - The protocol stack still has a reference
375 * SS_NOFDREF - There is no longer a file pointer reference
378 sofree(struct socket *so)
383 * This is a bit hackish at the moment. We need to interlock
384 * any accept queue we are on before we potentially lose the
385 * last reference to avoid races against a re-reference from
386 * someone operating on the queue.
388 while ((head = so->so_head) != NULL) {
389 lwkt_getpooltoken(head);
390 if (so->so_head == head)
392 lwkt_relpooltoken(head);
396 * Arbitrage the last free.
398 KKASSERT(so->so_refs > 0);
399 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
401 lwkt_relpooltoken(head);
405 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
406 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
410 * We're done, remove ourselves from the accept queue we are
411 * on, if we are on one.
413 if (so->so_state & SS_INCOMP) {
414 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) ==
416 TAILQ_REMOVE(&head->so_incomp, so, so_list);
418 } else if (so->so_state & SS_COMP) {
420 * We must not decommission a socket that's
421 * on the accept(2) queue. If we do, then
422 * accept(2) may hang after select(2) indicated
423 * that the listening socket was ready.
425 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) ==
427 lwkt_relpooltoken(head);
430 panic("sofree: not queued");
432 soclrstate(so, SS_INCOMP);
434 lwkt_relpooltoken(head);
436 /* Flush accept queues, if we are accepting. */
439 ssb_release(&so->so_snd, so);
445 * Close a socket on last file table reference removal.
446 * Initiate disconnect if connected.
447 * Free socket when disconnect complete.
450 soclose(struct socket *so, int fflag)
454 funsetown(&so->so_sigio);
455 sosetstate(so, SS_ISCLOSING);
456 if (!use_soclose_fast ||
457 (so->so_proto->pr_flags & PR_SYNC_PORT) ||
458 ((so->so_state & SS_ISCONNECTED) &&
459 (so->so_options & SO_LINGER) &&
460 so->so_linger != 0)) {
461 error = soclose_sync(so, fflag);
470 sodiscard(struct socket *so)
472 if (so->so_state & SS_NOFDREF)
473 panic("soclose: NOFDREF");
474 sosetstate(so, SS_NOFDREF); /* take ref */
478 * Append the completed queue of head to head_inh (inherting listen socket).
481 soinherit(struct socket *head, struct socket *head_inh)
483 boolean_t do_wakeup = FALSE;
485 KASSERT(head->so_options & SO_ACCEPTCONN,
486 ("head does not accept connection"));
487 KASSERT(head_inh->so_options & SO_ACCEPTCONN,
488 ("head_inh does not accept connection"));
490 lwkt_getpooltoken(head);
491 lwkt_getpooltoken(head_inh);
493 if (head->so_qlen > 0)
496 while (!TAILQ_EMPTY(&head->so_comp)) {
497 struct ucred *old_cr;
500 sp = TAILQ_FIRST(&head->so_comp);
501 KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP);
504 * Remove this socket from the current listen socket
507 TAILQ_REMOVE(&head->so_comp, sp, so_list);
510 /* Save the old ucred for later free. */
511 old_cr = sp->so_cred;
514 * Install this socket to the inheriting listen socket
517 sp->so_cred = crhold(head_inh->so_cred); /* non-blocking */
518 sp->so_head = head_inh;
520 TAILQ_INSERT_TAIL(&head_inh->so_comp, sp, so_list);
525 * crfree() may block and release the tokens temporarily.
526 * However, we are fine here, since the transition is done.
531 lwkt_relpooltoken(head_inh);
532 lwkt_relpooltoken(head);
536 * "New" connections have arrived
539 wakeup(&head_inh->so_timeo);
544 soclose_sync(struct socket *so, int fflag)
548 if ((so->so_proto->pr_flags & PR_SYNC_PORT) == 0)
549 so_pru_sync(so); /* unpend async prus */
551 if (so->so_pcb == NULL)
554 if (so->so_state & SS_ISCONNECTED) {
555 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
556 error = sodisconnect(so);
560 if (so->so_options & SO_LINGER) {
561 if ((so->so_state & SS_ISDISCONNECTING) &&
564 while (so->so_state & SS_ISCONNECTED) {
565 error = tsleep(&so->so_timeo, PCATCH,
566 "soclos", so->so_linger * hz);
576 error2 = so_pru_detach(so);
577 if (error2 == EJUSTRETURN) {
579 * Protocol will call sodiscard()
580 * and sofree() for us.
589 sofree(so); /* dispose of ref */
595 soclose_fast_handler(netmsg_t msg)
597 struct socket *so = msg->base.nm_so;
599 if (so->so_pcb == NULL)
602 if ((so->so_state & SS_ISCONNECTED) &&
603 (so->so_state & SS_ISDISCONNECTING) == 0)
604 so_pru_disconnect_direct(so);
609 error = so_pru_detach_direct(so);
610 if (error == EJUSTRETURN) {
612 * Protocol will call sodiscard()
613 * and sofree() for us.
624 soclose_fast(struct socket *so)
626 struct netmsg_base *base = &so->so_clomsg;
628 netmsg_init(base, so, &netisr_apanic_rport, 0,
629 soclose_fast_handler);
630 if (so->so_port == netisr_curport())
631 lwkt_sendmsg_oncpu(so->so_port, &base->lmsg);
633 lwkt_sendmsg(so->so_port, &base->lmsg);
637 * Abort and destroy a socket. Only one abort can be in progress
638 * at any given moment.
641 soabort_async(struct socket *so, boolean_t clr_head)
644 * Keep a reference before clearing the so_head
645 * to avoid racing socket close in netisr.
650 so_pru_abort_async(so);
654 soabort_direct(struct socket *so)
657 so_pru_abort_direct(so);
661 * so is passed in ref'd, which becomes owned by
662 * the cleared SS_NOFDREF flag.
665 soaccept_generic(struct socket *so)
667 if ((so->so_state & SS_NOFDREF) == 0)
668 panic("soaccept: !NOFDREF");
669 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */
673 soaccept(struct socket *so, struct sockaddr **nam)
677 soaccept_generic(so);
678 error = so_pru_accept(so, nam);
683 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td,
688 if (so->so_options & SO_ACCEPTCONN)
691 * If protocol is connection-based, can only connect once.
692 * Otherwise, if connected, try to disconnect first.
693 * This allows user to disconnect by connecting to, e.g.,
696 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
697 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
698 (error = sodisconnect(so)))) {
702 * Prevent accumulated error from previous connection
706 if (!sync && so->so_proto->pr_usrreqs->pru_preconnect)
707 error = so_pru_connect_async(so, nam, td);
709 error = so_pru_connect(so, nam, td);
715 soconnect2(struct socket *so1, struct socket *so2)
719 error = so_pru_connect2(so1, so2);
724 sodisconnect(struct socket *so)
728 if ((so->so_state & SS_ISCONNECTED) == 0) {
732 if (so->so_state & SS_ISDISCONNECTING) {
736 error = so_pru_disconnect(so);
741 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
744 * If send must go all at once and message is larger than
745 * send buffering, then hard error.
746 * Lock against other senders.
747 * If must go all at once and not enough room now, then
748 * inform user that this would block and do nothing.
749 * Otherwise, if nonblocking, send as much as possible.
750 * The data to be sent is described by "uio" if nonzero,
751 * otherwise by the mbuf chain "top" (which must be null
752 * if uio is not). Data provided in mbuf chain must be small
753 * enough to send all at once.
755 * Returns nonzero on error, timeout or signal; callers
756 * must check for short counts if EINTR/ERESTART are returned.
757 * Data and control buffers are freed on return.
760 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
761 struct mbuf *top, struct mbuf *control, int flags,
768 int clen = 0, error, dontroute, mlen;
769 int atomic = sosendallatonce(so) || top;
773 resid = uio->uio_resid;
775 resid = (size_t)top->m_pkthdr.len;
778 for (m = top; m; m = m->m_next)
780 KKASSERT(top->m_pkthdr.len == len);
785 * WARNING! resid is unsigned, space and len are signed. space
786 * can wind up negative if the sockbuf is overcommitted.
788 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
789 * type sockets since that's an error.
791 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
797 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
798 (so->so_proto->pr_flags & PR_ATOMIC);
799 if (td->td_lwp != NULL)
800 td->td_lwp->lwp_ru.ru_msgsnd++;
802 clen = control->m_len;
803 #define gotoerr(errcode) { error = errcode; goto release; }
806 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
811 if (so->so_state & SS_CANTSENDMORE)
814 error = so->so_error;
818 if ((so->so_state & SS_ISCONNECTED) == 0) {
820 * `sendto' and `sendmsg' is allowed on a connection-
821 * based socket if it supports implied connect.
822 * Return ENOTCONN if not connected and no address is
825 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
826 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
827 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
828 !(resid == 0 && clen != 0))
830 } else if (addr == NULL)
831 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
832 ENOTCONN : EDESTADDRREQ);
834 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
835 clen > so->so_snd.ssb_hiwat) {
838 space = ssb_space(&so->so_snd);
841 if ((space < 0 || (size_t)space < resid + clen) && uio &&
842 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
843 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
844 gotoerr(EWOULDBLOCK);
845 ssb_unlock(&so->so_snd);
846 error = ssb_wait(&so->so_snd);
856 * Data is prepackaged in "top".
860 top->m_flags |= M_EOR;
864 m = m_getl((int)resid, M_WAITOK, MT_DATA,
865 top == NULL ? M_PKTHDR : 0, &mlen);
868 m->m_pkthdr.rcvif = NULL;
870 len = imin((int)szmin(mlen, resid), space);
871 if (resid < MINCLSIZE) {
873 * For datagram protocols, leave room
874 * for protocol headers in first mbuf.
876 if (atomic && top == NULL && len < mlen)
880 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
881 resid = uio->uio_resid;
884 top->m_pkthdr.len += len;
890 top->m_flags |= M_EOR;
893 } while (space > 0 && atomic);
895 so->so_options |= SO_DONTROUTE;
896 if (flags & MSG_OOB) {
897 pru_flags = PRUS_OOB;
898 } else if ((flags & MSG_EOF) &&
899 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
902 * If the user set MSG_EOF, the protocol
903 * understands this flag and nothing left to
904 * send then use PRU_SEND_EOF instead of PRU_SEND.
906 pru_flags = PRUS_EOF;
907 } else if (resid > 0 && space > 0) {
908 /* If there is more to send, set PRUS_MORETOCOME */
909 pru_flags = PRUS_MORETOCOME;
914 * XXX all the SS_CANTSENDMORE checks previously
915 * done could be out of date. We could have recieved
916 * a reset packet in an interrupt or maybe we slept
917 * while doing page faults in uiomove() etc. We could
918 * probably recheck again inside the splnet() protection
919 * here, but there are probably other places that this
920 * also happens. We must rethink this.
922 error = so_pru_send(so, pru_flags, top, addr, control, td);
924 so->so_options &= ~SO_DONTROUTE;
931 } while (resid && space > 0);
935 ssb_unlock(&so->so_snd);
946 * A specialization of sosend() for UDP based on protocol-specific knowledge:
947 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
948 * sosendallatonce() returns true,
949 * the "atomic" variable is true,
950 * and sosendudp() blocks until space is available for the entire send.
951 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
952 * PR_IMPLOPCL flags set.
953 * UDP has no out-of-band data.
954 * UDP has no control data.
955 * UDP does not support MSG_EOR.
958 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
959 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
962 int error, pru_flags = 0;
965 if (td->td_lwp != NULL)
966 td->td_lwp->lwp_ru.ru_msgsnd++;
970 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
971 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
974 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
978 if (so->so_state & SS_CANTSENDMORE)
981 error = so->so_error;
985 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
986 gotoerr(EDESTADDRREQ);
987 if (resid > so->so_snd.ssb_hiwat)
989 space = ssb_space(&so->so_snd);
990 if (uio && (space < 0 || (size_t)space < resid)) {
991 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
992 gotoerr(EWOULDBLOCK);
993 ssb_unlock(&so->so_snd);
994 error = ssb_wait(&so->so_snd);
1001 int hdrlen = max_hdr;
1004 * We try to optimize out the additional mbuf
1005 * allocations in M_PREPEND() on output path, e.g.
1006 * - udp_output(), when it tries to prepend protocol
1008 * - Link layer output function, when it tries to
1009 * prepend link layer header.
1011 * This probably will not benefit any data that will
1012 * be fragmented, so this optimization is only performed
1013 * when the size of data and max size of protocol+link
1014 * headers fit into one mbuf cluster.
1016 if (uio->uio_resid > MCLBYTES - hdrlen ||
1017 !udp_sosend_prepend) {
1018 top = m_uiomove(uio);
1024 top = m_getl(uio->uio_resid + hdrlen, M_WAITOK,
1025 MT_DATA, M_PKTHDR, &nsize);
1026 KASSERT(nsize >= uio->uio_resid + hdrlen,
1027 ("sosendudp invalid nsize %d, "
1028 "resid %zu, hdrlen %d",
1029 nsize, uio->uio_resid, hdrlen));
1031 top->m_len = uio->uio_resid;
1032 top->m_pkthdr.len = uio->uio_resid;
1033 top->m_data += hdrlen;
1035 error = uiomove(mtod(top, caddr_t), top->m_len, uio);
1041 if (flags & MSG_DONTROUTE)
1042 pru_flags |= PRUS_DONTROUTE;
1044 if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
1045 so_pru_send_async(so, pru_flags, top, addr, NULL, td);
1048 error = so_pru_send(so, pru_flags, top, addr, NULL, td);
1050 top = NULL; /* sent or freed in lower layer */
1053 ssb_unlock(&so->so_snd);
1061 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1062 struct mbuf *top, struct mbuf *control, int flags,
1074 KKASSERT(top == NULL);
1076 resid = uio->uio_resid;
1079 resid = (size_t)top->m_pkthdr.len;
1082 for (m = top; m; m = m->m_next)
1084 KKASSERT(top->m_pkthdr.len == len);
1089 * WARNING! resid is unsigned, space and len are signed. space
1090 * can wind up negative if the sockbuf is overcommitted.
1092 * Also check to make sure that MSG_EOR isn't used on TCP
1094 if (flags & MSG_EOR) {
1100 /* TCP doesn't do control messages (rights, creds, etc) */
1101 if (control->m_len) {
1105 m_freem(control); /* empty control, just free it */
1109 if (td->td_lwp != NULL)
1110 td->td_lwp->lwp_ru.ru_msgsnd++;
1112 #define gotoerr(errcode) { error = errcode; goto release; }
1115 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1120 if (so->so_state & SS_CANTSENDMORE)
1123 error = so->so_error;
1127 if ((so->so_state & SS_ISCONNECTED) == 0 &&
1128 (so->so_state & SS_ISCONFIRMING) == 0)
1130 if (allatonce && resid > so->so_snd.ssb_hiwat)
1133 space = ssb_space_prealloc(&so->so_snd);
1134 if (flags & MSG_OOB)
1136 if ((space < 0 || (size_t)space < resid) && !allatonce &&
1137 space < so->so_snd.ssb_lowat) {
1138 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1139 gotoerr(EWOULDBLOCK);
1140 ssb_unlock(&so->so_snd);
1141 error = ssb_wait(&so->so_snd);
1148 int cnt = 0, async = 0;
1152 * Data is prepackaged in "top".
1156 if (resid > INT_MAX)
1158 if (tcp_sosend_jcluster) {
1159 m = m_getlj((int)resid, M_WAITOK, MT_DATA,
1160 top == NULL ? M_PKTHDR : 0, &mlen);
1162 m = m_getl((int)resid, M_WAITOK, MT_DATA,
1163 top == NULL ? M_PKTHDR : 0, &mlen);
1166 m->m_pkthdr.len = 0;
1167 m->m_pkthdr.rcvif = NULL;
1169 len = imin((int)szmin(mlen, resid), space);
1171 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1172 resid = uio->uio_resid;
1175 top->m_pkthdr.len += len;
1182 } while (space > 0 && cnt < tcp_sosend_agglim);
1184 if (tcp_sosend_async)
1187 if (flags & MSG_OOB) {
1188 pru_flags = PRUS_OOB;
1190 } else if ((flags & MSG_EOF) && resid == 0) {
1191 pru_flags = PRUS_EOF;
1192 } else if (resid > 0 && space > 0) {
1193 /* If there is more to send, set PRUS_MORETOCOME */
1194 pru_flags = PRUS_MORETOCOME;
1200 if (flags & MSG_SYNC)
1204 * XXX all the SS_CANTSENDMORE checks previously
1205 * done could be out of date. We could have recieved
1206 * a reset packet in an interrupt or maybe we slept
1207 * while doing page faults in uiomove() etc. We could
1208 * probably recheck again inside the splnet() protection
1209 * here, but there are probably other places that this
1210 * also happens. We must rethink this.
1212 for (m = top; m; m = m->m_next)
1213 ssb_preallocstream(&so->so_snd, m);
1215 error = so_pru_send(so, pru_flags, top,
1218 so_pru_send_async(so, pru_flags, top,
1227 } while (resid && space > 0);
1231 ssb_unlock(&so->so_snd);
1242 * Implement receive operations on a socket.
1244 * We depend on the way that records are added to the signalsockbuf
1245 * by sbappend*. In particular, each record (mbufs linked through m_next)
1246 * must begin with an address if the protocol so specifies,
1247 * followed by an optional mbuf or mbufs containing ancillary data,
1248 * and then zero or more mbufs of data.
1250 * Although the signalsockbuf is locked, new data may still be appended.
1251 * A token inside the ssb_lock deals with MP issues and still allows
1252 * the network to access the socket if we block in a uio.
1254 * The caller may receive the data as a single mbuf chain by supplying
1255 * an mbuf **mp0 for use in returning the chain. The uio is then used
1256 * only for the count in uio_resid.
1259 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1260 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1263 struct mbuf *free_chain = NULL;
1264 int flags, len, error, offset;
1265 struct protosw *pr = so->so_proto;
1267 size_t resid, orig_resid;
1268 boolean_t free_rights = FALSE;
1271 resid = uio->uio_resid;
1273 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1281 flags = *flagsp &~ MSG_EOR;
1284 if (flags & MSG_OOB) {
1285 m = m_get(M_WAITOK, MT_DATA);
1288 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1294 KKASSERT(resid >= (size_t)m->m_len);
1295 resid -= (size_t)m->m_len;
1296 } while (resid > 0 && m);
1299 uio->uio_resid = resid;
1300 error = uiomove(mtod(m, caddr_t),
1301 (int)szmin(resid, m->m_len),
1303 resid = uio->uio_resid;
1305 } while (uio->uio_resid && error == 0 && m);
1312 if ((so->so_state & SS_ISCONFIRMING) && resid)
1316 * The token interlocks against the protocol thread while
1317 * ssb_lock is a blocking lock against other userland entities.
1319 lwkt_gettoken(&so->so_rcv.ssb_token);
1321 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1325 m = so->so_rcv.ssb_mb;
1327 * If we have less data than requested, block awaiting more
1328 * (subject to any timeout) if:
1329 * 1. the current count is less than the low water mark, or
1330 * 2. MSG_WAITALL is set, and it is possible to do the entire
1331 * receive operation at once if we block (resid <= hiwat).
1332 * 3. MSG_DONTWAIT is not set
1333 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1334 * we have to do the receive in sections, and thus risk returning
1335 * a short count if a timeout or signal occurs after we start.
1337 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1338 (size_t)so->so_rcv.ssb_cc < resid) &&
1339 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1340 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1341 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1342 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1346 error = so->so_error;
1347 if ((flags & MSG_PEEK) == 0)
1351 if (so->so_state & SS_CANTRCVMORE) {
1357 for (; m; m = m->m_next) {
1358 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1359 m = so->so_rcv.ssb_mb;
1363 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1364 (pr->pr_flags & PR_CONNREQUIRED)) {
1370 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1371 error = EWOULDBLOCK;
1374 ssb_unlock(&so->so_rcv);
1375 error = ssb_wait(&so->so_rcv);
1381 if (uio && uio->uio_td && uio->uio_td->td_proc)
1382 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1385 * note: m should be == sb_mb here. Cache the next record while
1386 * cleaning up. Note that calling m_free*() will break out critical
1389 KKASSERT(m == so->so_rcv.ssb_mb);
1392 * Skip any address mbufs prepending the record.
1394 if (pr->pr_flags & PR_ADDR) {
1395 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1398 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1399 if (flags & MSG_PEEK)
1402 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1406 * Skip any control mbufs prepending the record.
1408 while (m && m->m_type == MT_CONTROL && error == 0) {
1409 if (flags & MSG_PEEK) {
1411 *controlp = m_copy(m, 0, m->m_len);
1412 m = m->m_next; /* XXX race */
1414 const struct cmsghdr *cm = mtod(m, struct cmsghdr *);
1417 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1418 if (pr->pr_domain->dom_externalize &&
1419 cm->cmsg_level == SOL_SOCKET &&
1420 cm->cmsg_type == SCM_RIGHTS) {
1421 error = pr->pr_domain->dom_externalize
1427 if (cm->cmsg_level == SOL_SOCKET &&
1428 cm->cmsg_type == SCM_RIGHTS)
1430 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1433 if (controlp && *controlp) {
1435 controlp = &(*controlp)->m_next;
1444 if (type == MT_OOBDATA)
1449 * Copy to the UIO or mbuf return chain (*mp).
1453 while (m && resid > 0 && error == 0) {
1454 if (m->m_type == MT_OOBDATA) {
1455 if (type != MT_OOBDATA)
1457 } else if (type == MT_OOBDATA)
1460 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1462 soclrstate(so, SS_RCVATMARK);
1463 len = (resid > INT_MAX) ? INT_MAX : resid;
1464 if (so->so_oobmark && len > so->so_oobmark - offset)
1465 len = so->so_oobmark - offset;
1466 if (len > m->m_len - moff)
1467 len = m->m_len - moff;
1470 * Copy out to the UIO or pass the mbufs back to the SIO.
1471 * The SIO is dealt with when we eat the mbuf, but deal
1472 * with the resid here either way.
1475 uio->uio_resid = resid;
1476 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1477 resid = uio->uio_resid;
1481 resid -= (size_t)len;
1485 * Eat the entire mbuf or just a piece of it
1487 if (len == m->m_len - moff) {
1488 if (m->m_flags & M_EOR)
1490 if (flags & MSG_PEEK) {
1495 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1499 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1503 if (flags & MSG_PEEK) {
1507 n = m_copym(m, 0, len, M_WAITOK);
1513 so->so_rcv.ssb_cc -= len;
1516 if (so->so_oobmark) {
1517 if ((flags & MSG_PEEK) == 0) {
1518 so->so_oobmark -= len;
1519 if (so->so_oobmark == 0) {
1520 sosetstate(so, SS_RCVATMARK);
1525 if (offset == so->so_oobmark)
1529 if (flags & MSG_EOR)
1532 * If the MSG_WAITALL flag is set (for non-atomic socket),
1533 * we must not quit until resid == 0 or an error
1534 * termination. If a signal/timeout occurs, return
1535 * with a short count but without error.
1536 * Keep signalsockbuf locked against other readers.
1538 while ((flags & MSG_WAITALL) && m == NULL &&
1539 resid > 0 && !sosendallatonce(so) &&
1540 so->so_rcv.ssb_mb == NULL) {
1541 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1544 * The window might have closed to zero, make
1545 * sure we send an ack now that we've drained
1546 * the buffer or we might end up blocking until
1547 * the idle takes over (5 seconds).
1549 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1550 so_pru_rcvd(so, flags);
1551 error = ssb_wait(&so->so_rcv);
1553 ssb_unlock(&so->so_rcv);
1557 m = so->so_rcv.ssb_mb;
1562 * If an atomic read was requested but unread data still remains
1563 * in the record, set MSG_TRUNC.
1565 if (m && pr->pr_flags & PR_ATOMIC)
1569 * Cleanup. If an atomic read was requested drop any unread data.
1571 if ((flags & MSG_PEEK) == 0) {
1572 if (m && (pr->pr_flags & PR_ATOMIC))
1573 sbdroprecord(&so->so_rcv.sb);
1574 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1575 so_pru_rcvd(so, flags);
1578 if (orig_resid == resid && orig_resid &&
1579 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1580 ssb_unlock(&so->so_rcv);
1587 ssb_unlock(&so->so_rcv);
1589 lwkt_reltoken(&so->so_rcv.ssb_token);
1591 if (free_rights && (pr->pr_flags & PR_RIGHTS) &&
1592 pr->pr_domain->dom_dispose)
1593 pr->pr_domain->dom_dispose(free_chain);
1594 m_freem(free_chain);
1600 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1601 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1604 struct mbuf *free_chain = NULL;
1605 int flags, len, error, offset;
1606 struct protosw *pr = so->so_proto;
1609 size_t resid, orig_resid, restmp;
1612 resid = uio->uio_resid;
1614 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1622 flags = *flagsp &~ MSG_EOR;
1625 if (flags & MSG_OOB) {
1626 m = m_get(M_WAITOK, MT_DATA);
1629 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1635 KKASSERT(resid >= (size_t)m->m_len);
1636 resid -= (size_t)m->m_len;
1637 } while (resid > 0 && m);
1640 uio->uio_resid = resid;
1641 error = uiomove(mtod(m, caddr_t),
1642 (int)szmin(resid, m->m_len),
1644 resid = uio->uio_resid;
1646 } while (uio->uio_resid && error == 0 && m);
1655 * The token interlocks against the protocol thread while
1656 * ssb_lock is a blocking lock against other userland entities.
1658 * Lock a limited number of mbufs (not all, so sbcompress() still
1659 * works well). The token is used as an interlock for sbwait() so
1660 * release it afterwords.
1663 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1667 lwkt_gettoken(&so->so_rcv.ssb_token);
1668 m = so->so_rcv.ssb_mb;
1671 * If we have less data than requested, block awaiting more
1672 * (subject to any timeout) if:
1673 * 1. the current count is less than the low water mark, or
1674 * 2. MSG_WAITALL is set, and it is possible to do the entire
1675 * receive operation at once if we block (resid <= hiwat).
1676 * 3. MSG_DONTWAIT is not set
1677 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1678 * we have to do the receive in sections, and thus risk returning
1679 * a short count if a timeout or signal occurs after we start.
1681 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1682 (size_t)so->so_rcv.ssb_cc < resid) &&
1683 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1684 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1685 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1689 lwkt_reltoken(&so->so_rcv.ssb_token);
1690 error = so->so_error;
1691 if ((flags & MSG_PEEK) == 0)
1695 if (so->so_state & SS_CANTRCVMORE) {
1698 lwkt_reltoken(&so->so_rcv.ssb_token);
1701 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1702 (pr->pr_flags & PR_CONNREQUIRED)) {
1703 lwkt_reltoken(&so->so_rcv.ssb_token);
1708 lwkt_reltoken(&so->so_rcv.ssb_token);
1711 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1712 lwkt_reltoken(&so->so_rcv.ssb_token);
1713 error = EWOULDBLOCK;
1716 ssb_unlock(&so->so_rcv);
1717 error = ssb_wait(&so->so_rcv);
1718 lwkt_reltoken(&so->so_rcv.ssb_token);
1730 while (n && restmp < resid) {
1731 n->m_flags |= M_SOLOCKED;
1733 if (n->m_next == NULL)
1740 * Release token for loop
1742 lwkt_reltoken(&so->so_rcv.ssb_token);
1743 if (uio && uio->uio_td && uio->uio_td->td_proc)
1744 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1747 * note: m should be == sb_mb here. Cache the next record while
1748 * cleaning up. Note that calling m_free*() will break out critical
1751 KKASSERT(m == so->so_rcv.ssb_mb);
1754 * Copy to the UIO or mbuf return chain (*mp).
1756 * NOTE: Token is not held for loop
1762 while (m && (m->m_flags & M_SOLOCKED) && resid > 0 && error == 0) {
1763 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1766 soclrstate(so, SS_RCVATMARK);
1767 len = (resid > INT_MAX) ? INT_MAX : resid;
1768 if (so->so_oobmark && len > so->so_oobmark - offset)
1769 len = so->so_oobmark - offset;
1770 if (len > m->m_len - moff)
1771 len = m->m_len - moff;
1774 * Copy out to the UIO or pass the mbufs back to the SIO.
1775 * The SIO is dealt with when we eat the mbuf, but deal
1776 * with the resid here either way.
1779 uio->uio_resid = resid;
1780 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1781 resid = uio->uio_resid;
1785 resid -= (size_t)len;
1789 * Eat the entire mbuf or just a piece of it
1792 if (len == m->m_len - moff) {
1802 if (so->so_oobmark && offset == so->so_oobmark) {
1809 * Synchronize sockbuf with data we read.
1811 * NOTE: (m) is junk on entry (it could be left over from the
1814 if ((flags & MSG_PEEK) == 0) {
1815 lwkt_gettoken(&so->so_rcv.ssb_token);
1816 m = so->so_rcv.ssb_mb;
1817 while (m && offset >= m->m_len) {
1818 if (so->so_oobmark) {
1819 so->so_oobmark -= m->m_len;
1820 if (so->so_oobmark == 0) {
1821 sosetstate(so, SS_RCVATMARK);
1827 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1831 m = sbunlinkmbuf(&so->so_rcv.sb,
1838 n = m_copym(m, 0, offset, M_WAITOK);
1842 m->m_data += offset;
1844 so->so_rcv.ssb_cc -= offset;
1845 if (so->so_oobmark) {
1846 so->so_oobmark -= offset;
1847 if (so->so_oobmark == 0) {
1848 sosetstate(so, SS_RCVATMARK);
1854 lwkt_reltoken(&so->so_rcv.ssb_token);
1858 * If the MSG_WAITALL flag is set (for non-atomic socket),
1859 * we must not quit until resid == 0 or an error termination.
1861 * If a signal/timeout occurs, return with a short count but without
1864 * Keep signalsockbuf locked against other readers.
1866 * XXX if MSG_PEEK we currently do quit.
1868 if ((flags & MSG_WAITALL) && !(flags & MSG_PEEK) &&
1869 didoob == 0 && resid > 0 &&
1870 !sosendallatonce(so)) {
1871 lwkt_gettoken(&so->so_rcv.ssb_token);
1873 while ((m = so->so_rcv.ssb_mb) == NULL) {
1874 if (so->so_error || (so->so_state & SS_CANTRCVMORE)) {
1875 error = so->so_error;
1879 * The window might have closed to zero, make
1880 * sure we send an ack now that we've drained
1881 * the buffer or we might end up blocking until
1882 * the idle takes over (5 seconds).
1885 so_pru_rcvd_async(so);
1886 if (so->so_rcv.ssb_mb == NULL)
1887 error = ssb_wait(&so->so_rcv);
1889 lwkt_reltoken(&so->so_rcv.ssb_token);
1890 ssb_unlock(&so->so_rcv);
1895 if (m && error == 0)
1897 lwkt_reltoken(&so->so_rcv.ssb_token);
1901 * Token not held here.
1903 * Cleanup. If an atomic read was requested drop any unread data XXX
1905 if ((flags & MSG_PEEK) == 0) {
1907 so_pru_rcvd_async(so);
1910 if (orig_resid == resid && orig_resid &&
1911 (so->so_state & SS_CANTRCVMORE) == 0) {
1912 ssb_unlock(&so->so_rcv);
1919 ssb_unlock(&so->so_rcv);
1922 m_freem(free_chain);
1927 * Shut a socket down. Note that we do not get a frontend lock as we
1928 * want to be able to shut the socket down even if another thread is
1929 * blocked in a read(), thus waking it up.
1932 soshutdown(struct socket *so, int how)
1934 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1937 if (how != SHUT_WR) {
1938 /*ssb_lock(&so->so_rcv, M_WAITOK);*/
1940 /*ssb_unlock(&so->so_rcv);*/
1943 return (so_pru_shutdown(so));
1948 sorflush(struct socket *so)
1950 struct signalsockbuf *ssb = &so->so_rcv;
1951 struct protosw *pr = so->so_proto;
1952 struct signalsockbuf asb;
1954 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1956 lwkt_gettoken(&ssb->ssb_token);
1961 * Can't just blow up the ssb structure here
1963 bzero(&ssb->sb, sizeof(ssb->sb));
1968 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1970 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1971 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1972 ssb_release(&asb, so);
1974 lwkt_reltoken(&ssb->ssb_token);
1979 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1981 struct accept_filter_arg *afap = NULL;
1982 struct accept_filter *afp;
1983 struct so_accf *af = so->so_accf;
1986 /* do not set/remove accept filters on non listen sockets */
1987 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1992 /* removing the filter */
1995 if (af->so_accept_filter != NULL &&
1996 af->so_accept_filter->accf_destroy != NULL) {
1997 af->so_accept_filter->accf_destroy(so);
1999 if (af->so_accept_filter_str != NULL) {
2000 kfree(af->so_accept_filter_str, M_ACCF);
2005 so->so_options &= ~SO_ACCEPTFILTER;
2008 /* adding a filter */
2009 /* must remove previous filter first */
2014 /* don't put large objects on the kernel stack */
2015 afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
2016 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
2017 afap->af_name[sizeof(afap->af_name)-1] = '\0';
2018 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
2021 afp = accept_filt_get(afap->af_name);
2026 af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
2027 if (afp->accf_create != NULL) {
2028 if (afap->af_name[0] != '\0') {
2029 int len = strlen(afap->af_name) + 1;
2031 af->so_accept_filter_str = kmalloc(len, M_ACCF,
2033 strcpy(af->so_accept_filter_str, afap->af_name);
2035 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
2036 if (af->so_accept_filter_arg == NULL) {
2037 kfree(af->so_accept_filter_str, M_ACCF);
2044 af->so_accept_filter = afp;
2046 so->so_options |= SO_ACCEPTFILTER;
2049 kfree(afap, M_TEMP);
2055 * Perhaps this routine, and sooptcopyout(), below, ought to come in
2056 * an additional variant to handle the case where the option value needs
2057 * to be some kind of integer, but not a specific size.
2058 * In addition to their use here, these functions are also called by the
2059 * protocol-level pr_ctloutput() routines.
2062 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2064 return soopt_to_kbuf(sopt, buf, len, minlen);
2068 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2072 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2073 KKASSERT(kva_p(buf));
2076 * If the user gives us more than we wanted, we ignore it,
2077 * but if we don't get the minimum length the caller
2078 * wants, we return EINVAL. On success, sopt->sopt_valsize
2079 * is set to however much we actually retrieved.
2081 if ((valsize = sopt->sopt_valsize) < minlen)
2084 sopt->sopt_valsize = valsize = len;
2086 bcopy(sopt->sopt_val, buf, valsize);
2092 sosetopt(struct socket *so, struct sockopt *sopt)
2098 struct signalsockbuf *sotmp;
2101 sopt->sopt_dir = SOPT_SET;
2102 if (sopt->sopt_level != SOL_SOCKET) {
2103 if (so->so_proto && so->so_proto->pr_ctloutput) {
2104 return (so_pr_ctloutput(so, sopt));
2106 error = ENOPROTOOPT;
2108 switch (sopt->sopt_name) {
2110 case SO_ACCEPTFILTER:
2111 error = do_setopt_accept_filter(so, sopt);
2117 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2121 so->so_linger = l.l_linger;
2123 so->so_options |= SO_LINGER;
2125 so->so_options &= ~SO_LINGER;
2131 case SO_USELOOPBACK:
2138 error = sooptcopyin(sopt, &optval, sizeof optval,
2143 so->so_options |= sopt->sopt_name;
2145 so->so_options &= ~sopt->sopt_name;
2152 error = sooptcopyin(sopt, &optval, sizeof optval,
2158 * Values < 1 make no sense for any of these
2159 * options, so disallow them.
2166 switch (sopt->sopt_name) {
2169 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2170 &so->so_snd : &so->so_rcv, (u_long)optval,
2172 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2176 sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2177 &so->so_snd : &so->so_rcv;
2178 atomic_clear_int(&sotmp->ssb_flags,
2183 * Make sure the low-water is never greater than
2187 so->so_snd.ssb_lowat =
2188 (optval > so->so_snd.ssb_hiwat) ?
2189 so->so_snd.ssb_hiwat : optval;
2190 atomic_clear_int(&so->so_snd.ssb_flags,
2194 so->so_rcv.ssb_lowat =
2195 (optval > so->so_rcv.ssb_hiwat) ?
2196 so->so_rcv.ssb_hiwat : optval;
2197 atomic_clear_int(&so->so_rcv.ssb_flags,
2205 error = sooptcopyin(sopt, &tv, sizeof tv,
2210 /* assert(hz > 0); */
2211 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2212 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2216 /* assert(tick > 0); */
2217 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2218 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2219 if (val > INT_MAX) {
2223 if (val == 0 && tv.tv_usec != 0)
2226 switch (sopt->sopt_name) {
2228 so->so_snd.ssb_timeo = val;
2231 so->so_rcv.ssb_timeo = val;
2236 error = ENOPROTOOPT;
2239 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2240 (void) so_pr_ctloutput(so, sopt);
2247 /* Helper routine for getsockopt */
2249 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2251 soopt_from_kbuf(sopt, buf, len);
2256 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2261 sopt->sopt_valsize = 0;
2265 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2266 KKASSERT(kva_p(buf));
2269 * Documented get behavior is that we always return a value,
2270 * possibly truncated to fit in the user's buffer.
2271 * Traditional behavior is that we always tell the user
2272 * precisely how much we copied, rather than something useful
2273 * like the total amount we had available for her.
2274 * Note that this interface is not idempotent; the entire answer must
2275 * generated ahead of time.
2277 valsize = szmin(len, sopt->sopt_valsize);
2278 sopt->sopt_valsize = valsize;
2279 if (sopt->sopt_val != 0) {
2280 bcopy(buf, sopt->sopt_val, valsize);
2285 sogetopt(struct socket *so, struct sockopt *sopt)
2292 struct accept_filter_arg *afap;
2296 sopt->sopt_dir = SOPT_GET;
2297 if (sopt->sopt_level != SOL_SOCKET) {
2298 if (so->so_proto && so->so_proto->pr_ctloutput) {
2299 return (so_pr_ctloutput(so, sopt));
2301 return (ENOPROTOOPT);
2303 switch (sopt->sopt_name) {
2305 case SO_ACCEPTFILTER:
2306 if ((so->so_options & SO_ACCEPTCONN) == 0)
2308 afap = kmalloc(sizeof(*afap), M_TEMP,
2310 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2311 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2312 if (so->so_accf->so_accept_filter_str != NULL)
2313 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2315 error = sooptcopyout(sopt, afap, sizeof(*afap));
2316 kfree(afap, M_TEMP);
2321 l.l_onoff = so->so_options & SO_LINGER;
2322 l.l_linger = so->so_linger;
2323 error = sooptcopyout(sopt, &l, sizeof l);
2326 case SO_USELOOPBACK:
2336 optval = so->so_options & sopt->sopt_name;
2338 error = sooptcopyout(sopt, &optval, sizeof optval);
2342 optval = so->so_type;
2346 optval = so->so_error;
2351 optval = so->so_snd.ssb_hiwat;
2355 optval = so->so_rcv.ssb_hiwat;
2359 optval = so->so_snd.ssb_lowat;
2363 optval = so->so_rcv.ssb_lowat;
2368 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2369 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2371 tv.tv_sec = optval / hz;
2372 tv.tv_usec = (optval % hz) * ustick;
2373 error = sooptcopyout(sopt, &tv, sizeof tv);
2377 optval_l = ssb_space(&so->so_snd);
2378 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2382 optval = -1; /* no hint */
2386 error = ENOPROTOOPT;
2389 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput)
2390 so_pr_ctloutput(so, sopt);
2395 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2397 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2399 struct mbuf *m, *m_prev;
2400 int sopt_size = sopt->sopt_valsize, msize;
2402 m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA,
2406 m->m_len = min(msize, sopt_size);
2407 sopt_size -= m->m_len;
2411 while (sopt_size > 0) {
2412 m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT,
2413 MT_DATA, 0, &msize);
2418 m->m_len = min(msize, sopt_size);
2419 sopt_size -= m->m_len;
2426 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2428 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2430 soopt_to_mbuf(sopt, m);
2435 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2440 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2442 if (sopt->sopt_val == NULL)
2444 val = sopt->sopt_val;
2445 valsize = sopt->sopt_valsize;
2446 while (m != NULL && valsize >= m->m_len) {
2447 bcopy(val, mtod(m, char *), m->m_len);
2448 valsize -= m->m_len;
2449 val = (caddr_t)val + m->m_len;
2452 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2453 panic("ip6_sooptmcopyin");
2456 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2458 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2460 return soopt_from_mbuf(sopt, m);
2464 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2466 struct mbuf *m0 = m;
2471 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2473 if (sopt->sopt_val == NULL)
2475 val = sopt->sopt_val;
2476 maxsize = sopt->sopt_valsize;
2477 while (m != NULL && maxsize >= m->m_len) {
2478 bcopy(mtod(m, char *), val, m->m_len);
2479 maxsize -= m->m_len;
2480 val = (caddr_t)val + m->m_len;
2481 valsize += m->m_len;
2485 /* enough soopt buffer should be given from user-land */
2489 sopt->sopt_valsize = valsize;
2494 sohasoutofband(struct socket *so)
2496 if (so->so_sigio != NULL)
2497 pgsigio(so->so_sigio, SIGURG, 0);
2500 * There is no need to use NOTE_OOB as KNOTE hint here:
2501 * soread filter depends on so_oobmark and SS_RCVATMARK
2502 * so_state. NOTE_OOB would cause unnecessary penalty
2503 * in KNOTE, if there was knote processing contention.
2505 KNOTE(&so->so_rcv.ssb_kq.ki_note, 0);
2509 sokqfilter(struct file *fp, struct knote *kn)
2511 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2512 struct signalsockbuf *ssb;
2514 switch (kn->kn_filter) {
2516 if (so->so_options & SO_ACCEPTCONN)
2517 kn->kn_fop = &solisten_filtops;
2519 kn->kn_fop = &soread_filtops;
2523 kn->kn_fop = &sowrite_filtops;
2527 kn->kn_fop = &soexcept_filtops;
2531 return (EOPNOTSUPP);
2534 knote_insert(&ssb->ssb_kq.ki_note, kn);
2535 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2540 filt_sordetach(struct knote *kn)
2542 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2544 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2545 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2546 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2551 filt_soread(struct knote *kn, long hint __unused)
2553 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2555 if (kn->kn_sfflags & NOTE_OOB) {
2556 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2557 kn->kn_fflags |= NOTE_OOB;
2562 kn->kn_data = so->so_rcv.ssb_cc;
2564 if (so->so_state & SS_CANTRCVMORE) {
2566 * Only set NODATA if all data has been exhausted.
2568 if (kn->kn_data == 0)
2569 kn->kn_flags |= EV_NODATA;
2570 kn->kn_flags |= EV_EOF;
2571 kn->kn_fflags = so->so_error;
2574 if (so->so_error) /* temporary udp error */
2576 if (kn->kn_sfflags & NOTE_LOWAT)
2577 return (kn->kn_data >= kn->kn_sdata);
2578 return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2579 !TAILQ_EMPTY(&so->so_comp));
2583 filt_sowdetach(struct knote *kn)
2585 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2587 knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2588 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2589 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2594 filt_sowrite(struct knote *kn, long hint __unused)
2596 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2598 if (so->so_snd.ssb_flags & SSB_PREALLOC)
2599 kn->kn_data = ssb_space_prealloc(&so->so_snd);
2601 kn->kn_data = ssb_space(&so->so_snd);
2603 if (so->so_state & SS_CANTSENDMORE) {
2604 kn->kn_flags |= (EV_EOF | EV_NODATA);
2605 kn->kn_fflags = so->so_error;
2608 if (so->so_error) /* temporary udp error */
2610 if (((so->so_state & SS_ISCONNECTED) == 0) &&
2611 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2613 if (kn->kn_sfflags & NOTE_LOWAT)
2614 return (kn->kn_data >= kn->kn_sdata);
2615 return (kn->kn_data >= so->so_snd.ssb_lowat);
2620 filt_solisten(struct knote *kn, long hint __unused)
2622 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2623 int qlen = so->so_qlen;
2625 if (soavailconn > 0 && qlen > soavailconn)
2629 return (!TAILQ_EMPTY(&so->so_comp));