2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2004 The FreeBSD Foundation
5 * Copyright (c) 2004-2007 Robert N. M. Watson
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 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
36 * Comments on the socket life cycle:
38 * soalloc() sets of socket layer state for a socket, called only by
39 * socreate() and sonewconn(). Socket layer private.
41 * sodealloc() tears down socket layer state for a socket, called only by
42 * sofree() and sonewconn(). Socket layer private.
44 * pru_attach() associates protocol layer state with an allocated socket;
45 * called only once, may fail, aborting socket allocation. This is called
46 * from socreate() and sonewconn(). Socket layer private.
48 * pru_detach() disassociates protocol layer state from an attached socket,
49 * and will be called exactly once for sockets in which pru_attach() has
50 * been successfully called. If pru_attach() returned an error,
51 * pru_detach() will not be called. Socket layer private.
53 * pru_abort() and pru_close() notify the protocol layer that the last
54 * consumer of a socket is starting to tear down the socket, and that the
55 * protocol should terminate the connection. Historically, pru_abort() also
56 * detached protocol state from the socket state, but this is no longer the
59 * socreate() creates a socket and attaches protocol state. This is a public
60 * interface that may be used by socket layer consumers to create new
63 * sonewconn() creates a socket and attaches protocol state. This is a
64 * public interface that may be used by protocols to create new sockets when
65 * a new connection is received and will be available for accept() on a
68 * soclose() destroys a socket after possibly waiting for it to disconnect.
69 * This is a public interface that socket consumers should use to close and
70 * release a socket when done with it.
72 * soabort() destroys a socket without waiting for it to disconnect (used
73 * only for incoming connections that are already partially or fully
74 * connected). This is used internally by the socket layer when clearing
75 * listen socket queues (due to overflow or close on the listen socket), but
76 * is also a public interface protocols may use to abort connections in
77 * their incomplete listen queues should they no longer be required. Sockets
78 * placed in completed connection listen queues should not be aborted for
79 * reasons described in the comment above the soclose() implementation. This
80 * is not a general purpose close routine, and except in the specific
81 * circumstances described here, should not be used.
83 * sofree() will free a socket and its protocol state if all references on
84 * the socket have been released, and is the public interface to attempt to
85 * free a socket when a reference is removed. This is a socket layer private
88 * NOTE: In addition to socreate() and soclose(), which provide a single
89 * socket reference to the consumer to be managed as required, there are two
90 * calls to explicitly manage socket references, soref(), and sorele().
91 * Currently, these are generally required only when transitioning a socket
92 * from a listen queue to a file descriptor, in order to prevent garbage
93 * collection of the socket at an untimely moment. For a number of reasons,
94 * these interfaces are not preferred, and should be avoided.
97 #include <sys/cdefs.h>
98 __FBSDID("$FreeBSD$");
100 #include "opt_inet.h"
102 #include "opt_zero.h"
103 #include "opt_compat.h"
105 #include <sys/param.h>
106 #include <sys/systm.h>
107 #include <sys/fcntl.h>
108 #include <sys/limits.h>
109 #include <sys/lock.h>
111 #include <sys/malloc.h>
112 #include <sys/mbuf.h>
113 #include <sys/mutex.h>
114 #include <sys/domain.h>
115 #include <sys/file.h> /* for struct knote */
116 #include <sys/kernel.h>
117 #include <sys/event.h>
118 #include <sys/eventhandler.h>
119 #include <sys/poll.h>
120 #include <sys/proc.h>
121 #include <sys/protosw.h>
122 #include <sys/socket.h>
123 #include <sys/socketvar.h>
124 #include <sys/resourcevar.h>
125 #include <sys/signalvar.h>
126 #include <sys/stat.h>
128 #include <sys/sysctl.h>
130 #include <sys/jail.h>
132 #include <security/mac/mac_framework.h>
137 #include <sys/mount.h>
138 #include <compat/freebsd32/freebsd32.h>
140 extern struct sysentvec ia32_freebsd_sysvec;
143 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
146 static void filt_sordetach(struct knote *kn);
147 static int filt_soread(struct knote *kn, long hint);
148 static void filt_sowdetach(struct knote *kn);
149 static int filt_sowrite(struct knote *kn, long hint);
150 static int filt_solisten(struct knote *kn, long hint);
152 static struct filterops solisten_filtops =
153 { 1, NULL, filt_sordetach, filt_solisten };
154 static struct filterops soread_filtops =
155 { 1, NULL, filt_sordetach, filt_soread };
156 static struct filterops sowrite_filtops =
157 { 1, NULL, filt_sowdetach, filt_sowrite };
159 uma_zone_t socket_zone;
160 so_gen_t so_gencnt; /* generation count for sockets */
164 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
165 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
167 static int somaxconn = SOMAXCONN;
168 static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS);
169 /* XXX: we dont have SYSCTL_USHORT */
170 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
171 0, sizeof(int), sysctl_somaxconn, "I", "Maximum pending socket connection "
173 static int numopensockets;
174 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
175 &numopensockets, 0, "Number of open sockets");
176 #ifdef ZERO_COPY_SOCKETS
177 /* These aren't static because they're used in other files. */
178 int so_zero_copy_send = 1;
179 int so_zero_copy_receive = 1;
180 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
181 "Zero copy controls");
182 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
183 &so_zero_copy_receive, 0, "Enable zero copy receive");
184 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
185 &so_zero_copy_send, 0, "Enable zero copy send");
186 #endif /* ZERO_COPY_SOCKETS */
189 * accept_mtx locks down per-socket fields relating to accept queues. See
190 * socketvar.h for an annotation of the protected fields of struct socket.
192 struct mtx accept_mtx;
193 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
196 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
199 static struct mtx so_global_mtx;
200 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
203 * General IPC sysctl name space, used by sockets and a variety of other IPC
206 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
209 * Sysctl to get and set the maximum global sockets limit. Notify protocols
210 * of the change so that they can update their dependent limits as required.
213 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
215 int error, newmaxsockets;
217 newmaxsockets = maxsockets;
218 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
219 if (error == 0 && req->newptr) {
220 if (newmaxsockets > maxsockets) {
221 maxsockets = newmaxsockets;
222 if (maxsockets > ((maxfiles / 4) * 3)) {
223 maxfiles = (maxsockets * 5) / 4;
224 maxfilesperproc = (maxfiles * 9) / 10;
226 EVENTHANDLER_INVOKE(maxsockets_change);
233 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
234 &maxsockets, 0, sysctl_maxsockets, "IU",
235 "Maximum number of sockets avaliable");
238 * Initialise maxsockets.
240 static void init_maxsockets(void *ignored)
242 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
243 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
245 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
248 * Socket operation routines. These routines are called by the routines in
249 * sys_socket.c or from a system process, and implement the semantics of
250 * socket operations by switching out to the protocol specific routines.
254 * Get a socket structure from our zone, and initialize it. Note that it
255 * would probably be better to allocate socket and PCB at the same time, but
256 * I'm not convinced that all the protocols can be easily modified to do
259 * soalloc() returns a socket with a ref count of 0.
261 static struct socket *
266 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
270 if (mac_socket_init(so, M_NOWAIT) != 0) {
271 uma_zfree(socket_zone, so);
275 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
276 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
277 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
278 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
279 TAILQ_INIT(&so->so_aiojobq);
280 mtx_lock(&so_global_mtx);
281 so->so_gencnt = ++so_gencnt;
283 mtx_unlock(&so_global_mtx);
288 * Free the storage associated with a socket at the socket layer, tear down
289 * locks, labels, etc. All protocol state is assumed already to have been
290 * torn down (and possibly never set up) by the caller.
293 sodealloc(struct socket *so)
296 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
297 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
299 mtx_lock(&so_global_mtx);
300 so->so_gencnt = ++so_gencnt;
301 --numopensockets; /* Could be below, but faster here. */
302 mtx_unlock(&so_global_mtx);
303 if (so->so_rcv.sb_hiwat)
304 (void)chgsbsize(so->so_cred->cr_uidinfo,
305 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
306 if (so->so_snd.sb_hiwat)
307 (void)chgsbsize(so->so_cred->cr_uidinfo,
308 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
310 /* remove acccept filter if one is present. */
311 if (so->so_accf != NULL)
312 do_setopt_accept_filter(so, NULL);
315 mac_socket_destroy(so);
318 sx_destroy(&so->so_snd.sb_sx);
319 sx_destroy(&so->so_rcv.sb_sx);
320 SOCKBUF_LOCK_DESTROY(&so->so_snd);
321 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
322 uma_zfree(socket_zone, so);
326 * socreate returns a socket with a ref count of 1. The socket should be
327 * closed with soclose().
330 socreate(int dom, struct socket **aso, int type, int proto,
331 struct ucred *cred, struct thread *td)
338 prp = pffindproto(dom, proto, type);
340 prp = pffindtype(dom, type);
342 if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
343 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
344 return (EPROTONOSUPPORT);
346 if (jailed(cred) && jail_socket_unixiproute_only &&
347 prp->pr_domain->dom_family != PF_LOCAL &&
348 prp->pr_domain->dom_family != PF_INET &&
349 prp->pr_domain->dom_family != PF_ROUTE) {
350 return (EPROTONOSUPPORT);
353 if (prp->pr_type != type)
359 TAILQ_INIT(&so->so_incomp);
360 TAILQ_INIT(&so->so_comp);
362 so->so_cred = crhold(cred);
365 mac_socket_create(cred, so);
367 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
369 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
373 * Auto-sizing of socket buffers is managed by the protocols and
374 * the appropriate flags must be set in the pru_attach function.
376 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
378 KASSERT(so->so_count == 1, ("socreate: so_count %d",
389 static int regression_sonewconn_earlytest = 1;
390 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
391 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
395 * When an attempt at a new connection is noted on a socket which accepts
396 * connections, sonewconn is called. If the connection is possible (subject
397 * to space constraints, etc.) then we allocate a new structure, propoerly
398 * linked into the data structure of the original socket, and return this.
399 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
401 * Note: the ref count on the socket is 0 on return.
404 sonewconn(struct socket *head, int connstatus)
410 over = (head->so_qlen > 3 * head->so_qlimit / 2);
413 if (regression_sonewconn_earlytest && over)
421 if ((head->so_options & SO_ACCEPTFILTER) != 0)
424 so->so_type = head->so_type;
425 so->so_options = head->so_options &~ SO_ACCEPTCONN;
426 so->so_linger = head->so_linger;
427 so->so_state = head->so_state | SS_NOFDREF;
428 so->so_proto = head->so_proto;
429 so->so_cred = crhold(head->so_cred);
432 mac_socket_newconn(head, so);
435 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
437 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
439 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
440 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
444 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
445 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
446 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
447 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
448 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
449 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
450 so->so_state |= connstatus;
453 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
454 so->so_qstate |= SQ_COMP;
458 * Keep removing sockets from the head until there's room for
459 * us to insert on the tail. In pre-locking revisions, this
460 * was a simple if(), but as we could be racing with other
461 * threads and soabort() requires dropping locks, we must
462 * loop waiting for the condition to be true.
464 while (head->so_incqlen > head->so_qlimit) {
466 sp = TAILQ_FIRST(&head->so_incomp);
467 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
469 sp->so_qstate &= ~SQ_INCOMP;
475 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
476 so->so_qstate |= SQ_INCOMP;
482 wakeup_one(&head->so_timeo);
488 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
491 return ((*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td));
495 * solisten() transitions a socket from a non-listening state to a listening
496 * state, but can also be used to update the listen queue depth on an
497 * existing listen socket. The protocol will call back into the sockets
498 * layer using solisten_proto_check() and solisten_proto() to check and set
499 * socket-layer listen state. Call backs are used so that the protocol can
500 * acquire both protocol and socket layer locks in whatever order is required
503 * Protocol implementors are advised to hold the socket lock across the
504 * socket-layer test and set to avoid races at the socket layer.
507 solisten(struct socket *so, int backlog, struct thread *td)
510 return ((*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td));
514 solisten_proto_check(struct socket *so)
517 SOCK_LOCK_ASSERT(so);
519 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
526 solisten_proto(struct socket *so, int backlog)
529 SOCK_LOCK_ASSERT(so);
531 if (backlog < 0 || backlog > somaxconn)
533 so->so_qlimit = backlog;
534 so->so_options |= SO_ACCEPTCONN;
538 * Attempt to free a socket. This should really be sotryfree().
540 * sofree() will succeed if:
542 * - There are no outstanding file descriptor references or related consumers
545 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
547 * - The protocol does not have an outstanding strong reference on the socket
550 * - The socket is not in a completed connection queue, so a process has been
551 * notified that it is present. If it is removed, the user process may
552 * block in accept() despite select() saying the socket was ready.
554 * Otherwise, it will quietly abort so that a future call to sofree(), when
555 * conditions are right, can succeed.
558 sofree(struct socket *so)
560 struct protosw *pr = so->so_proto;
563 ACCEPT_LOCK_ASSERT();
564 SOCK_LOCK_ASSERT(so);
566 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
567 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
575 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
576 (so->so_qstate & SQ_INCOMP) != 0,
577 ("sofree: so_head != NULL, but neither SQ_COMP nor "
579 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
580 (so->so_qstate & SQ_INCOMP) == 0,
581 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
582 TAILQ_REMOVE(&head->so_incomp, so, so_list);
584 so->so_qstate &= ~SQ_INCOMP;
587 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
588 (so->so_qstate & SQ_INCOMP) == 0,
589 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
590 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
591 if (so->so_options & SO_ACCEPTCONN) {
592 KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated"));
593 KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_comp populated"));
598 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
599 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
600 if (pr->pr_usrreqs->pru_detach != NULL)
601 (*pr->pr_usrreqs->pru_detach)(so);
604 * From this point on, we assume that no other references to this
605 * socket exist anywhere else in the stack. Therefore, no locks need
606 * to be acquired or held.
608 * We used to do a lot of socket buffer and socket locking here, as
609 * well as invoke sorflush() and perform wakeups. The direct call to
610 * dom_dispose() and sbrelease_internal() are an inlining of what was
611 * necessary from sorflush().
613 * Notice that the socket buffer and kqueue state are torn down
614 * before calling pru_detach. This means that protocols shold not
615 * assume they can perform socket wakeups, etc, in their detach code.
617 sbdestroy(&so->so_snd, so);
618 sbdestroy(&so->so_rcv, so);
619 knlist_destroy(&so->so_rcv.sb_sel.si_note);
620 knlist_destroy(&so->so_snd.sb_sel.si_note);
625 * Close a socket on last file table reference removal. Initiate disconnect
626 * if connected. Free socket when disconnect complete.
628 * This function will sorele() the socket. Note that soclose() may be called
629 * prior to the ref count reaching zero. The actual socket structure will
630 * not be freed until the ref count reaches zero.
633 soclose(struct socket *so)
637 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
639 funsetown(&so->so_sigio);
640 if (so->so_state & SS_ISCONNECTED) {
641 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
642 error = sodisconnect(so);
646 if (so->so_options & SO_LINGER) {
647 if ((so->so_state & SS_ISDISCONNECTING) &&
648 (so->so_state & SS_NBIO))
650 while (so->so_state & SS_ISCONNECTED) {
651 error = tsleep(&so->so_timeo,
652 PSOCK | PCATCH, "soclos", so->so_linger * hz);
660 if (so->so_proto->pr_usrreqs->pru_close != NULL)
661 (*so->so_proto->pr_usrreqs->pru_close)(so);
662 if (so->so_options & SO_ACCEPTCONN) {
665 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
666 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
668 sp->so_qstate &= ~SQ_INCOMP;
674 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
675 TAILQ_REMOVE(&so->so_comp, sp, so_list);
677 sp->so_qstate &= ~SQ_COMP;
687 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
688 so->so_state |= SS_NOFDREF;
694 * soabort() is used to abruptly tear down a connection, such as when a
695 * resource limit is reached (listen queue depth exceeded), or if a listen
696 * socket is closed while there are sockets waiting to be accepted.
698 * This interface is tricky, because it is called on an unreferenced socket,
699 * and must be called only by a thread that has actually removed the socket
700 * from the listen queue it was on, or races with other threads are risked.
702 * This interface will call into the protocol code, so must not be called
703 * with any socket locks held. Protocols do call it while holding their own
704 * recursible protocol mutexes, but this is something that should be subject
705 * to review in the future.
708 soabort(struct socket *so)
712 * In as much as is possible, assert that no references to this
713 * socket are held. This is not quite the same as asserting that the
714 * current thread is responsible for arranging for no references, but
715 * is as close as we can get for now.
717 KASSERT(so->so_count == 0, ("soabort: so_count"));
718 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
719 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
720 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
721 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
723 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
724 (*so->so_proto->pr_usrreqs->pru_abort)(so);
731 soaccept(struct socket *so, struct sockaddr **nam)
736 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
737 so->so_state &= ~SS_NOFDREF;
739 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
744 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
748 if (so->so_options & SO_ACCEPTCONN)
751 * If protocol is connection-based, can only connect once.
752 * Otherwise, if connected, try to disconnect first. This allows
753 * user to disconnect by connecting to, e.g., a null address.
755 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
756 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
757 (error = sodisconnect(so)))) {
761 * Prevent accumulated error from previous connection from
765 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
772 soconnect2(struct socket *so1, struct socket *so2)
775 return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
779 sodisconnect(struct socket *so)
783 if ((so->so_state & SS_ISCONNECTED) == 0)
785 if (so->so_state & SS_ISDISCONNECTING)
787 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
791 #ifdef ZERO_COPY_SOCKETS
792 struct so_zerocopy_stats{
797 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
798 #include <netinet/in.h>
799 #include <net/route.h>
800 #include <netinet/in_pcb.h>
802 #include <vm/vm_page.h>
803 #include <vm/vm_object.h>
806 * sosend_copyin() is only used if zero copy sockets are enabled. Otherwise
807 * sosend_dgram() and sosend_generic() use m_uiotombuf().
809 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
810 * all of the data referenced by the uio. If desired, it uses zero-copy.
811 * *space will be updated to reflect data copied in.
813 * NB: If atomic I/O is requested, the caller must already have checked that
814 * space can hold resid bytes.
816 * NB: In the event of an error, the caller may need to free the partial
817 * chain pointed to by *mpp. The contents of both *uio and *space may be
818 * modified even in the case of an error.
821 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
824 struct mbuf *m, **mp, *top;
827 #ifdef ZERO_COPY_SOCKETS
834 resid = uio->uio_resid;
837 #ifdef ZERO_COPY_SOCKETS
839 #endif /* ZERO_COPY_SOCKETS */
840 if (resid >= MINCLSIZE) {
841 #ifdef ZERO_COPY_SOCKETS
843 m = m_gethdr(M_WAITOK, MT_DATA);
845 m->m_pkthdr.rcvif = NULL;
847 m = m_get(M_WAITOK, MT_DATA);
848 if (so_zero_copy_send &&
851 uio->uio_iov->iov_len>=PAGE_SIZE) {
852 so_zerocp_stats.size_ok++;
853 so_zerocp_stats.align_ok++;
854 cow_send = socow_setup(m, uio);
858 m_clget(m, M_WAITOK);
859 len = min(min(MCLBYTES, resid), *space);
861 #else /* ZERO_COPY_SOCKETS */
863 m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
865 m->m_pkthdr.rcvif = NULL;
867 m = m_getcl(M_WAIT, MT_DATA, 0);
868 len = min(min(MCLBYTES, resid), *space);
869 #endif /* ZERO_COPY_SOCKETS */
872 m = m_gethdr(M_WAIT, MT_DATA);
874 m->m_pkthdr.rcvif = NULL;
876 len = min(min(MHLEN, resid), *space);
878 * For datagram protocols, leave room
879 * for protocol headers in first mbuf.
881 if (atomic && m && len < MHLEN)
884 m = m_get(M_WAIT, MT_DATA);
885 len = min(min(MLEN, resid), *space);
894 #ifdef ZERO_COPY_SOCKETS
898 #endif /* ZERO_COPY_SOCKETS */
899 error = uiomove(mtod(m, void *), (int)len, uio);
900 resid = uio->uio_resid;
903 top->m_pkthdr.len += len;
909 top->m_flags |= M_EOR;
912 } while (*space > 0 && atomic);
917 #endif /*ZERO_COPY_SOCKETS*/
919 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
922 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
923 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
926 int clen = 0, error, dontroute;
927 #ifdef ZERO_COPY_SOCKETS
928 int atomic = sosendallatonce(so) || top;
931 KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
932 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
933 ("sodgram_send: !PR_ATOMIC"));
936 resid = uio->uio_resid;
938 resid = top->m_pkthdr.len;
940 * In theory resid should be unsigned. However, space must be
941 * signed, as it might be less than 0 if we over-committed, and we
942 * must use a signed comparison of space and resid. On the other
943 * hand, a negative resid causes us to loop sending 0-length
944 * segments to the protocol.
946 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
947 * type sockets since that's an error.
955 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
957 td->td_ru.ru_msgsnd++;
959 clen = control->m_len;
961 SOCKBUF_LOCK(&so->so_snd);
962 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
963 SOCKBUF_UNLOCK(&so->so_snd);
968 error = so->so_error;
970 SOCKBUF_UNLOCK(&so->so_snd);
973 if ((so->so_state & SS_ISCONNECTED) == 0) {
975 * `sendto' and `sendmsg' is allowed on a connection-based
976 * socket if it supports implied connect. Return ENOTCONN if
977 * not connected and no address is supplied.
979 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
980 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
981 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
982 !(resid == 0 && clen != 0)) {
983 SOCKBUF_UNLOCK(&so->so_snd);
987 } else if (addr == NULL) {
988 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
991 error = EDESTADDRREQ;
992 SOCKBUF_UNLOCK(&so->so_snd);
998 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
999 * problem and need fixing.
1001 space = sbspace(&so->so_snd);
1002 if (flags & MSG_OOB)
1005 SOCKBUF_UNLOCK(&so->so_snd);
1006 if (resid > space) {
1012 if (flags & MSG_EOR)
1013 top->m_flags |= M_EOR;
1015 #ifdef ZERO_COPY_SOCKETS
1016 error = sosend_copyin(uio, &top, atomic, &space, flags);
1021 * Copy the data from userland into a mbuf chain.
1022 * If no data is to be copied in, a single empty mbuf
1025 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1026 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1028 error = EFAULT; /* only possible error */
1031 space -= resid - uio->uio_resid;
1033 resid = uio->uio_resid;
1035 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1037 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1042 so->so_options |= SO_DONTROUTE;
1046 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1047 * of date. We could have recieved a reset packet in an interrupt or
1048 * maybe we slept while doing page faults in uiomove() etc. We could
1049 * probably recheck again inside the locking protection here, but
1050 * there are probably other places that this also happens. We must
1053 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1054 (flags & MSG_OOB) ? PRUS_OOB :
1056 * If the user set MSG_EOF, the protocol understands this flag and
1057 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1059 ((flags & MSG_EOF) &&
1060 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1063 /* If there is more to send set PRUS_MORETOCOME */
1064 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1065 top, addr, control, td);
1068 so->so_options &= ~SO_DONTROUTE;
1077 if (control != NULL)
1083 * Send on a socket. If send must go all at once and message is larger than
1084 * send buffering, then hard error. Lock against other senders. If must go
1085 * all at once and not enough room now, then inform user that this would
1086 * block and do nothing. Otherwise, if nonblocking, send as much as
1087 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1088 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1089 * in mbuf chain must be small enough to send all at once.
1091 * Returns nonzero on error, timeout or signal; callers must check for short
1092 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1096 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1097 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1100 int clen = 0, error, dontroute;
1101 int atomic = sosendallatonce(so) || top;
1104 resid = uio->uio_resid;
1106 resid = top->m_pkthdr.len;
1108 * In theory resid should be unsigned. However, space must be
1109 * signed, as it might be less than 0 if we over-committed, and we
1110 * must use a signed comparison of space and resid. On the other
1111 * hand, a negative resid causes us to loop sending 0-length
1112 * segments to the protocol.
1114 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1115 * type sockets since that's an error.
1117 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1123 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1124 (so->so_proto->pr_flags & PR_ATOMIC);
1126 td->td_ru.ru_msgsnd++;
1127 if (control != NULL)
1128 clen = control->m_len;
1130 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1136 SOCKBUF_LOCK(&so->so_snd);
1137 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1138 SOCKBUF_UNLOCK(&so->so_snd);
1143 error = so->so_error;
1145 SOCKBUF_UNLOCK(&so->so_snd);
1148 if ((so->so_state & SS_ISCONNECTED) == 0) {
1150 * `sendto' and `sendmsg' is allowed on a connection-
1151 * based socket if it supports implied connect.
1152 * Return ENOTCONN if not connected and no address is
1155 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1156 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1157 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1158 !(resid == 0 && clen != 0)) {
1159 SOCKBUF_UNLOCK(&so->so_snd);
1163 } else if (addr == NULL) {
1164 SOCKBUF_UNLOCK(&so->so_snd);
1165 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1168 error = EDESTADDRREQ;
1172 space = sbspace(&so->so_snd);
1173 if (flags & MSG_OOB)
1175 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1176 clen > so->so_snd.sb_hiwat) {
1177 SOCKBUF_UNLOCK(&so->so_snd);
1181 if (space < resid + clen &&
1182 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1183 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1184 SOCKBUF_UNLOCK(&so->so_snd);
1185 error = EWOULDBLOCK;
1188 error = sbwait(&so->so_snd);
1189 SOCKBUF_UNLOCK(&so->so_snd);
1194 SOCKBUF_UNLOCK(&so->so_snd);
1199 if (flags & MSG_EOR)
1200 top->m_flags |= M_EOR;
1202 #ifdef ZERO_COPY_SOCKETS
1203 error = sosend_copyin(uio, &top, atomic,
1209 * Copy the data from userland into a mbuf
1210 * chain. If no data is to be copied in,
1211 * a single empty mbuf is returned.
1213 top = m_uiotombuf(uio, M_WAITOK, space,
1214 (atomic ? max_hdr : 0),
1215 (atomic ? M_PKTHDR : 0) |
1216 ((flags & MSG_EOR) ? M_EOR : 0));
1218 error = EFAULT; /* only possible error */
1221 space -= resid - uio->uio_resid;
1223 resid = uio->uio_resid;
1227 so->so_options |= SO_DONTROUTE;
1231 * XXX all the SBS_CANTSENDMORE checks previously
1232 * done could be out of date. We could have recieved
1233 * a reset packet in an interrupt or maybe we slept
1234 * while doing page faults in uiomove() etc. We
1235 * could probably recheck again inside the locking
1236 * protection here, but there are probably other
1237 * places that this also happens. We must rethink
1240 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1241 (flags & MSG_OOB) ? PRUS_OOB :
1243 * If the user set MSG_EOF, the protocol understands
1244 * this flag and nothing left to send then use
1245 * PRU_SEND_EOF instead of PRU_SEND.
1247 ((flags & MSG_EOF) &&
1248 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1251 /* If there is more to send set PRUS_MORETOCOME. */
1252 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1253 top, addr, control, td);
1256 so->so_options &= ~SO_DONTROUTE;
1264 } while (resid && space > 0);
1268 sbunlock(&so->so_snd);
1272 if (control != NULL)
1278 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1279 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1282 /* XXXRW: Temporary debugging. */
1283 KASSERT(so->so_proto->pr_usrreqs->pru_sosend != sosend,
1284 ("sosend: protocol calls sosend"));
1286 return (so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1287 control, flags, td));
1291 * The part of soreceive() that implements reading non-inline out-of-band
1292 * data from a socket. For more complete comments, see soreceive(), from
1293 * which this code originated.
1295 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1296 * unable to return an mbuf chain to the caller.
1299 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1301 struct protosw *pr = so->so_proto;
1305 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1307 m = m_get(M_WAIT, MT_DATA);
1308 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1312 #ifdef ZERO_COPY_SOCKETS
1313 if (so_zero_copy_receive) {
1316 if ((m->m_flags & M_EXT)
1317 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1322 error = uiomoveco(mtod(m, void *),
1323 min(uio->uio_resid, m->m_len),
1326 #endif /* ZERO_COPY_SOCKETS */
1327 error = uiomove(mtod(m, void *),
1328 (int) min(uio->uio_resid, m->m_len), uio);
1330 } while (uio->uio_resid && error == 0 && m);
1338 * Following replacement or removal of the first mbuf on the first mbuf chain
1339 * of a socket buffer, push necessary state changes back into the socket
1340 * buffer so that other consumers see the values consistently. 'nextrecord'
1341 * is the callers locally stored value of the original value of
1342 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1343 * NOTE: 'nextrecord' may be NULL.
1345 static __inline void
1346 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1349 SOCKBUF_LOCK_ASSERT(sb);
1351 * First, update for the new value of nextrecord. If necessary, make
1352 * it the first record.
1354 if (sb->sb_mb != NULL)
1355 sb->sb_mb->m_nextpkt = nextrecord;
1357 sb->sb_mb = nextrecord;
1360 * Now update any dependent socket buffer fields to reflect the new
1361 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1362 * addition of a second clause that takes care of the case where
1363 * sb_mb has been updated, but remains the last record.
1365 if (sb->sb_mb == NULL) {
1366 sb->sb_mbtail = NULL;
1367 sb->sb_lastrecord = NULL;
1368 } else if (sb->sb_mb->m_nextpkt == NULL)
1369 sb->sb_lastrecord = sb->sb_mb;
1374 * Implement receive operations on a socket. We depend on the way that
1375 * records are added to the sockbuf by sbappend. In particular, each record
1376 * (mbufs linked through m_next) must begin with an address if the protocol
1377 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1378 * data, and then zero or more mbufs of data. In order to allow parallelism
1379 * between network receive and copying to user space, as well as avoid
1380 * sleeping with a mutex held, we release the socket buffer mutex during the
1381 * user space copy. Although the sockbuf is locked, new data may still be
1382 * appended, and thus we must maintain consistency of the sockbuf during that
1385 * The caller may receive the data as a single mbuf chain by supplying an
1386 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1387 * the count in uio_resid.
1390 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1391 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1393 struct mbuf *m, **mp;
1394 int flags, len, error, offset;
1395 struct protosw *pr = so->so_proto;
1396 struct mbuf *nextrecord;
1398 int orig_resid = uio->uio_resid;
1403 if (controlp != NULL)
1406 flags = *flagsp &~ MSG_EOR;
1409 if (flags & MSG_OOB)
1410 return (soreceive_rcvoob(so, uio, flags));
1413 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1415 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1417 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1422 SOCKBUF_LOCK(&so->so_rcv);
1423 m = so->so_rcv.sb_mb;
1425 * If we have less data than requested, block awaiting more (subject
1426 * to any timeout) if:
1427 * 1. the current count is less than the low water mark, or
1428 * 2. MSG_WAITALL is set, and it is possible to do the entire
1429 * receive operation at once if we block (resid <= hiwat).
1430 * 3. MSG_DONTWAIT is not set
1431 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1432 * we have to do the receive in sections, and thus risk returning a
1433 * short count if a timeout or signal occurs after we start.
1435 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1436 so->so_rcv.sb_cc < uio->uio_resid) &&
1437 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1438 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1439 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1440 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1441 ("receive: m == %p so->so_rcv.sb_cc == %u",
1442 m, so->so_rcv.sb_cc));
1446 error = so->so_error;
1447 if ((flags & MSG_PEEK) == 0)
1449 SOCKBUF_UNLOCK(&so->so_rcv);
1452 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1453 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1455 SOCKBUF_UNLOCK(&so->so_rcv);
1460 for (; m != NULL; m = m->m_next)
1461 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1462 m = so->so_rcv.sb_mb;
1465 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1466 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1467 SOCKBUF_UNLOCK(&so->so_rcv);
1471 if (uio->uio_resid == 0) {
1472 SOCKBUF_UNLOCK(&so->so_rcv);
1475 if ((so->so_state & SS_NBIO) ||
1476 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1477 SOCKBUF_UNLOCK(&so->so_rcv);
1478 error = EWOULDBLOCK;
1481 SBLASTRECORDCHK(&so->so_rcv);
1482 SBLASTMBUFCHK(&so->so_rcv);
1483 error = sbwait(&so->so_rcv);
1484 SOCKBUF_UNLOCK(&so->so_rcv);
1491 * From this point onward, we maintain 'nextrecord' as a cache of the
1492 * pointer to the next record in the socket buffer. We must keep the
1493 * various socket buffer pointers and local stack versions of the
1494 * pointers in sync, pushing out modifications before dropping the
1495 * socket buffer mutex, and re-reading them when picking it up.
1497 * Otherwise, we will race with the network stack appending new data
1498 * or records onto the socket buffer by using inconsistent/stale
1499 * versions of the field, possibly resulting in socket buffer
1502 * By holding the high-level sblock(), we prevent simultaneous
1503 * readers from pulling off the front of the socket buffer.
1505 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1507 uio->uio_td->td_ru.ru_msgrcv++;
1508 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1509 SBLASTRECORDCHK(&so->so_rcv);
1510 SBLASTMBUFCHK(&so->so_rcv);
1511 nextrecord = m->m_nextpkt;
1512 if (pr->pr_flags & PR_ADDR) {
1513 KASSERT(m->m_type == MT_SONAME,
1514 ("m->m_type == %d", m->m_type));
1517 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1519 if (flags & MSG_PEEK) {
1522 sbfree(&so->so_rcv, m);
1523 so->so_rcv.sb_mb = m_free(m);
1524 m = so->so_rcv.sb_mb;
1525 sockbuf_pushsync(&so->so_rcv, nextrecord);
1530 * Process one or more MT_CONTROL mbufs present before any data mbufs
1531 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1532 * just copy the data; if !MSG_PEEK, we call into the protocol to
1533 * perform externalization (or freeing if controlp == NULL).
1535 if (m != NULL && m->m_type == MT_CONTROL) {
1536 struct mbuf *cm = NULL, *cmn;
1537 struct mbuf **cme = &cm;
1540 if (flags & MSG_PEEK) {
1541 if (controlp != NULL) {
1542 *controlp = m_copy(m, 0, m->m_len);
1543 controlp = &(*controlp)->m_next;
1547 sbfree(&so->so_rcv, m);
1548 so->so_rcv.sb_mb = m->m_next;
1551 cme = &(*cme)->m_next;
1552 m = so->so_rcv.sb_mb;
1554 } while (m != NULL && m->m_type == MT_CONTROL);
1555 if ((flags & MSG_PEEK) == 0)
1556 sockbuf_pushsync(&so->so_rcv, nextrecord);
1557 while (cm != NULL) {
1560 if (pr->pr_domain->dom_externalize != NULL) {
1561 SOCKBUF_UNLOCK(&so->so_rcv);
1562 error = (*pr->pr_domain->dom_externalize)
1564 SOCKBUF_LOCK(&so->so_rcv);
1565 } else if (controlp != NULL)
1569 if (controlp != NULL) {
1571 while (*controlp != NULL)
1572 controlp = &(*controlp)->m_next;
1577 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1579 nextrecord = so->so_rcv.sb_mb;
1583 if ((flags & MSG_PEEK) == 0) {
1584 KASSERT(m->m_nextpkt == nextrecord,
1585 ("soreceive: post-control, nextrecord !sync"));
1586 if (nextrecord == NULL) {
1587 KASSERT(so->so_rcv.sb_mb == m,
1588 ("soreceive: post-control, sb_mb!=m"));
1589 KASSERT(so->so_rcv.sb_lastrecord == m,
1590 ("soreceive: post-control, lastrecord!=m"));
1594 if (type == MT_OOBDATA)
1597 if ((flags & MSG_PEEK) == 0) {
1598 KASSERT(so->so_rcv.sb_mb == nextrecord,
1599 ("soreceive: sb_mb != nextrecord"));
1600 if (so->so_rcv.sb_mb == NULL) {
1601 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1602 ("soreceive: sb_lastercord != NULL"));
1606 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1607 SBLASTRECORDCHK(&so->so_rcv);
1608 SBLASTMBUFCHK(&so->so_rcv);
1611 * Now continue to read any data mbufs off of the head of the socket
1612 * buffer until the read request is satisfied. Note that 'type' is
1613 * used to store the type of any mbuf reads that have happened so far
1614 * such that soreceive() can stop reading if the type changes, which
1615 * causes soreceive() to return only one of regular data and inline
1616 * out-of-band data in a single socket receive operation.
1620 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1622 * If the type of mbuf has changed since the last mbuf
1623 * examined ('type'), end the receive operation.
1625 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1626 if (m->m_type == MT_OOBDATA) {
1627 if (type != MT_OOBDATA)
1629 } else if (type == MT_OOBDATA)
1632 KASSERT(m->m_type == MT_DATA,
1633 ("m->m_type == %d", m->m_type));
1634 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1635 len = uio->uio_resid;
1636 if (so->so_oobmark && len > so->so_oobmark - offset)
1637 len = so->so_oobmark - offset;
1638 if (len > m->m_len - moff)
1639 len = m->m_len - moff;
1641 * If mp is set, just pass back the mbufs. Otherwise copy
1642 * them out via the uio, then free. Sockbuf must be
1643 * consistent here (points to current mbuf, it points to next
1644 * record) when we drop priority; we must note any additions
1645 * to the sockbuf when we block interrupts again.
1648 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1649 SBLASTRECORDCHK(&so->so_rcv);
1650 SBLASTMBUFCHK(&so->so_rcv);
1651 SOCKBUF_UNLOCK(&so->so_rcv);
1652 #ifdef ZERO_COPY_SOCKETS
1653 if (so_zero_copy_receive) {
1656 if ((m->m_flags & M_EXT)
1657 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1662 error = uiomoveco(mtod(m, char *) + moff,
1666 #endif /* ZERO_COPY_SOCKETS */
1667 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1668 SOCKBUF_LOCK(&so->so_rcv);
1671 * The MT_SONAME mbuf has already been removed
1672 * from the record, so it is necessary to
1673 * remove the data mbufs, if any, to preserve
1674 * the invariant in the case of PR_ADDR that
1675 * requires MT_SONAME mbufs at the head of
1678 if (m && pr->pr_flags & PR_ATOMIC &&
1679 ((flags & MSG_PEEK) == 0))
1680 (void)sbdroprecord_locked(&so->so_rcv);
1681 SOCKBUF_UNLOCK(&so->so_rcv);
1685 uio->uio_resid -= len;
1686 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1687 if (len == m->m_len - moff) {
1688 if (m->m_flags & M_EOR)
1690 if (flags & MSG_PEEK) {
1694 nextrecord = m->m_nextpkt;
1695 sbfree(&so->so_rcv, m);
1699 so->so_rcv.sb_mb = m = m->m_next;
1702 so->so_rcv.sb_mb = m_free(m);
1703 m = so->so_rcv.sb_mb;
1705 sockbuf_pushsync(&so->so_rcv, nextrecord);
1706 SBLASTRECORDCHK(&so->so_rcv);
1707 SBLASTMBUFCHK(&so->so_rcv);
1710 if (flags & MSG_PEEK)
1716 if (flags & MSG_DONTWAIT)
1717 copy_flag = M_DONTWAIT;
1720 if (copy_flag == M_WAIT)
1721 SOCKBUF_UNLOCK(&so->so_rcv);
1722 *mp = m_copym(m, 0, len, copy_flag);
1723 if (copy_flag == M_WAIT)
1724 SOCKBUF_LOCK(&so->so_rcv);
1727 * m_copym() couldn't
1728 * allocate an mbuf. Adjust
1729 * uio_resid back (it was
1730 * adjusted down by len
1731 * bytes, which we didn't end
1732 * up "copying" over).
1734 uio->uio_resid += len;
1740 so->so_rcv.sb_cc -= len;
1743 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1744 if (so->so_oobmark) {
1745 if ((flags & MSG_PEEK) == 0) {
1746 so->so_oobmark -= len;
1747 if (so->so_oobmark == 0) {
1748 so->so_rcv.sb_state |= SBS_RCVATMARK;
1753 if (offset == so->so_oobmark)
1757 if (flags & MSG_EOR)
1760 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1761 * must not quit until "uio->uio_resid == 0" or an error
1762 * termination. If a signal/timeout occurs, return with a
1763 * short count but without error. Keep sockbuf locked
1764 * against other readers.
1766 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1767 !sosendallatonce(so) && nextrecord == NULL) {
1768 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1769 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1772 * Notify the protocol that some data has been
1773 * drained before blocking.
1775 if (pr->pr_flags & PR_WANTRCVD) {
1776 SOCKBUF_UNLOCK(&so->so_rcv);
1777 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1778 SOCKBUF_LOCK(&so->so_rcv);
1780 SBLASTRECORDCHK(&so->so_rcv);
1781 SBLASTMBUFCHK(&so->so_rcv);
1782 error = sbwait(&so->so_rcv);
1784 SOCKBUF_UNLOCK(&so->so_rcv);
1787 m = so->so_rcv.sb_mb;
1789 nextrecord = m->m_nextpkt;
1793 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1794 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1796 if ((flags & MSG_PEEK) == 0)
1797 (void) sbdroprecord_locked(&so->so_rcv);
1799 if ((flags & MSG_PEEK) == 0) {
1802 * First part is an inline SB_EMPTY_FIXUP(). Second
1803 * part makes sure sb_lastrecord is up-to-date if
1804 * there is still data in the socket buffer.
1806 so->so_rcv.sb_mb = nextrecord;
1807 if (so->so_rcv.sb_mb == NULL) {
1808 so->so_rcv.sb_mbtail = NULL;
1809 so->so_rcv.sb_lastrecord = NULL;
1810 } else if (nextrecord->m_nextpkt == NULL)
1811 so->so_rcv.sb_lastrecord = nextrecord;
1813 SBLASTRECORDCHK(&so->so_rcv);
1814 SBLASTMBUFCHK(&so->so_rcv);
1816 * If soreceive() is being done from the socket callback,
1817 * then don't need to generate ACK to peer to update window,
1818 * since ACK will be generated on return to TCP.
1820 if (!(flags & MSG_SOCALLBCK) &&
1821 (pr->pr_flags & PR_WANTRCVD)) {
1822 SOCKBUF_UNLOCK(&so->so_rcv);
1823 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1824 SOCKBUF_LOCK(&so->so_rcv);
1827 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1828 if (orig_resid == uio->uio_resid && orig_resid &&
1829 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1830 SOCKBUF_UNLOCK(&so->so_rcv);
1833 SOCKBUF_UNLOCK(&so->so_rcv);
1838 sbunlock(&so->so_rcv);
1843 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1844 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1847 /* XXXRW: Temporary debugging. */
1848 KASSERT(so->so_proto->pr_usrreqs->pru_soreceive != soreceive,
1849 ("soreceive: protocol calls soreceive"));
1851 return (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
1856 soshutdown(struct socket *so, int how)
1858 struct protosw *pr = so->so_proto;
1860 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1862 if (pr->pr_usrreqs->pru_flush != NULL) {
1863 (*pr->pr_usrreqs->pru_flush)(so, how);
1868 return ((*pr->pr_usrreqs->pru_shutdown)(so));
1873 sorflush(struct socket *so)
1875 struct sockbuf *sb = &so->so_rcv;
1876 struct protosw *pr = so->so_proto;
1880 * In order to avoid calling dom_dispose with the socket buffer mutex
1881 * held, and in order to generally avoid holding the lock for a long
1882 * time, we make a copy of the socket buffer and clear the original
1883 * (except locks, state). The new socket buffer copy won't have
1884 * initialized locks so we can only call routines that won't use or
1885 * assert those locks.
1887 * Dislodge threads currently blocked in receive and wait to acquire
1888 * a lock against other simultaneous readers before clearing the
1889 * socket buffer. Don't let our acquire be interrupted by a signal
1890 * despite any existing socket disposition on interruptable waiting.
1893 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
1896 * Invalidate/clear most of the sockbuf structure, but leave selinfo
1897 * and mutex data unchanged.
1900 bzero(&asb, offsetof(struct sockbuf, sb_startzero));
1901 bcopy(&sb->sb_startzero, &asb.sb_startzero,
1902 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1903 bzero(&sb->sb_startzero,
1904 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1909 * Dispose of special rights and flush the socket buffer. Don't call
1910 * any unsafe routines (that rely on locks being initialized) on asb.
1912 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1913 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
1914 sbrelease_internal(&asb, so);
1918 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
1919 * additional variant to handle the case where the option value needs to be
1920 * some kind of integer, but not a specific size. In addition to their use
1921 * here, these functions are also called by the protocol-level pr_ctloutput()
1925 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1930 * If the user gives us more than we wanted, we ignore it, but if we
1931 * don't get the minimum length the caller wants, we return EINVAL.
1932 * On success, sopt->sopt_valsize is set to however much we actually
1935 if ((valsize = sopt->sopt_valsize) < minlen)
1938 sopt->sopt_valsize = valsize = len;
1940 if (sopt->sopt_td != NULL)
1941 return (copyin(sopt->sopt_val, buf, valsize));
1943 bcopy(sopt->sopt_val, buf, valsize);
1948 * Kernel version of setsockopt(2).
1950 * XXX: optlen is size_t, not socklen_t
1953 so_setsockopt(struct socket *so, int level, int optname, void *optval,
1956 struct sockopt sopt;
1958 sopt.sopt_level = level;
1959 sopt.sopt_name = optname;
1960 sopt.sopt_dir = SOPT_SET;
1961 sopt.sopt_val = optval;
1962 sopt.sopt_valsize = optlen;
1963 sopt.sopt_td = NULL;
1964 return (sosetopt(so, &sopt));
1968 sosetopt(struct socket *so, struct sockopt *sopt)
1979 if (sopt->sopt_level != SOL_SOCKET) {
1980 if (so->so_proto && so->so_proto->pr_ctloutput)
1981 return ((*so->so_proto->pr_ctloutput)
1983 error = ENOPROTOOPT;
1985 switch (sopt->sopt_name) {
1987 case SO_ACCEPTFILTER:
1988 error = do_setopt_accept_filter(so, sopt);
1994 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1999 so->so_linger = l.l_linger;
2001 so->so_options |= SO_LINGER;
2003 so->so_options &= ~SO_LINGER;
2010 case SO_USELOOPBACK:
2018 error = sooptcopyin(sopt, &optval, sizeof optval,
2024 so->so_options |= sopt->sopt_name;
2026 so->so_options &= ~sopt->sopt_name;
2034 error = sooptcopyin(sopt, &optval, sizeof optval,
2040 * Values < 1 make no sense for any of these options,
2048 switch (sopt->sopt_name) {
2051 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2052 &so->so_snd : &so->so_rcv, (u_long)optval,
2053 so, curthread) == 0) {
2057 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2058 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2062 * Make sure the low-water is never greater than the
2066 SOCKBUF_LOCK(&so->so_snd);
2067 so->so_snd.sb_lowat =
2068 (optval > so->so_snd.sb_hiwat) ?
2069 so->so_snd.sb_hiwat : optval;
2070 SOCKBUF_UNLOCK(&so->so_snd);
2073 SOCKBUF_LOCK(&so->so_rcv);
2074 so->so_rcv.sb_lowat =
2075 (optval > so->so_rcv.sb_hiwat) ?
2076 so->so_rcv.sb_hiwat : optval;
2077 SOCKBUF_UNLOCK(&so->so_rcv);
2085 if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
2086 struct timeval32 tv32;
2088 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2090 CP(tv32, tv, tv_sec);
2091 CP(tv32, tv, tv_usec);
2094 error = sooptcopyin(sopt, &tv, sizeof tv,
2099 /* assert(hz > 0); */
2100 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2101 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2105 /* assert(tick > 0); */
2106 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2107 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2108 if (val > INT_MAX) {
2112 if (val == 0 && tv.tv_usec != 0)
2115 switch (sopt->sopt_name) {
2117 so->so_snd.sb_timeo = val;
2120 so->so_rcv.sb_timeo = val;
2127 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2131 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2139 error = ENOPROTOOPT;
2142 if (error == 0 && so->so_proto != NULL &&
2143 so->so_proto->pr_ctloutput != NULL) {
2144 (void) ((*so->so_proto->pr_ctloutput)
2153 * Helper routine for getsockopt.
2156 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2164 * Documented get behavior is that we always return a value, possibly
2165 * truncated to fit in the user's buffer. Traditional behavior is
2166 * that we always tell the user precisely how much we copied, rather
2167 * than something useful like the total amount we had available for
2168 * her. Note that this interface is not idempotent; the entire
2169 * answer must generated ahead of time.
2171 valsize = min(len, sopt->sopt_valsize);
2172 sopt->sopt_valsize = valsize;
2173 if (sopt->sopt_val != NULL) {
2174 if (sopt->sopt_td != NULL)
2175 error = copyout(buf, sopt->sopt_val, valsize);
2177 bcopy(buf, sopt->sopt_val, valsize);
2183 sogetopt(struct socket *so, struct sockopt *sopt)
2193 if (sopt->sopt_level != SOL_SOCKET) {
2194 if (so->so_proto && so->so_proto->pr_ctloutput) {
2195 return ((*so->so_proto->pr_ctloutput)
2198 return (ENOPROTOOPT);
2200 switch (sopt->sopt_name) {
2202 case SO_ACCEPTFILTER:
2203 error = do_getopt_accept_filter(so, sopt);
2208 l.l_onoff = so->so_options & SO_LINGER;
2209 l.l_linger = so->so_linger;
2211 error = sooptcopyout(sopt, &l, sizeof l);
2214 case SO_USELOOPBACK:
2226 optval = so->so_options & sopt->sopt_name;
2228 error = sooptcopyout(sopt, &optval, sizeof optval);
2232 optval = so->so_type;
2237 optval = so->so_error;
2243 optval = so->so_snd.sb_hiwat;
2247 optval = so->so_rcv.sb_hiwat;
2251 optval = so->so_snd.sb_lowat;
2255 optval = so->so_rcv.sb_lowat;
2260 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2261 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2263 tv.tv_sec = optval / hz;
2264 tv.tv_usec = (optval % hz) * tick;
2266 if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
2267 struct timeval32 tv32;
2269 CP(tv, tv32, tv_sec);
2270 CP(tv, tv32, tv_usec);
2271 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2274 error = sooptcopyout(sopt, &tv, sizeof tv);
2279 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2283 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2287 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2295 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2299 error = mac_getsockopt_peerlabel(
2300 sopt->sopt_td->td_ucred, so, &extmac);
2303 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2309 case SO_LISTENQLIMIT:
2310 optval = so->so_qlimit;
2314 optval = so->so_qlen;
2317 case SO_LISTENINCQLEN:
2318 optval = so->so_incqlen;
2322 error = ENOPROTOOPT;
2329 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2331 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2333 struct mbuf *m, *m_prev;
2334 int sopt_size = sopt->sopt_valsize;
2336 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2339 if (sopt_size > MLEN) {
2340 MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
2341 if ((m->m_flags & M_EXT) == 0) {
2345 m->m_len = min(MCLBYTES, sopt_size);
2347 m->m_len = min(MLEN, sopt_size);
2349 sopt_size -= m->m_len;
2354 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2359 if (sopt_size > MLEN) {
2360 MCLGET(m, sopt->sopt_td != NULL ? M_WAIT :
2362 if ((m->m_flags & M_EXT) == 0) {
2367 m->m_len = min(MCLBYTES, sopt_size);
2369 m->m_len = min(MLEN, sopt_size);
2371 sopt_size -= m->m_len;
2378 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2380 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2382 struct mbuf *m0 = m;
2384 if (sopt->sopt_val == NULL)
2386 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2387 if (sopt->sopt_td != NULL) {
2390 error = copyin(sopt->sopt_val, mtod(m, char *),
2397 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2398 sopt->sopt_valsize -= m->m_len;
2399 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2402 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2403 panic("ip6_sooptmcopyin");
2407 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2409 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2411 struct mbuf *m0 = m;
2414 if (sopt->sopt_val == NULL)
2416 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2417 if (sopt->sopt_td != NULL) {
2420 error = copyout(mtod(m, char *), sopt->sopt_val,
2427 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2428 sopt->sopt_valsize -= m->m_len;
2429 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2430 valsize += m->m_len;
2434 /* enough soopt buffer should be given from user-land */
2438 sopt->sopt_valsize = valsize;
2443 * sohasoutofband(): protocol notifies socket layer of the arrival of new
2444 * out-of-band data, which will then notify socket consumers.
2447 sohasoutofband(struct socket *so)
2450 if (so->so_sigio != NULL)
2451 pgsigio(&so->so_sigio, SIGURG, 0);
2452 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2456 sopoll(struct socket *so, int events, struct ucred *active_cred,
2460 /* XXXRW: Temporary debugging. */
2461 KASSERT(so->so_proto->pr_usrreqs->pru_sopoll != sopoll,
2462 ("sopoll: protocol calls sopoll"));
2464 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2469 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2474 SOCKBUF_LOCK(&so->so_snd);
2475 SOCKBUF_LOCK(&so->so_rcv);
2476 if (events & (POLLIN | POLLRDNORM))
2478 revents |= events & (POLLIN | POLLRDNORM);
2480 if (events & POLLINIGNEOF)
2481 if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
2482 !TAILQ_EMPTY(&so->so_comp) || so->so_error)
2483 revents |= POLLINIGNEOF;
2485 if (events & (POLLOUT | POLLWRNORM))
2486 if (sowriteable(so))
2487 revents |= events & (POLLOUT | POLLWRNORM);
2489 if (events & (POLLPRI | POLLRDBAND))
2490 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2491 revents |= events & (POLLPRI | POLLRDBAND);
2495 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
2497 selrecord(td, &so->so_rcv.sb_sel);
2498 so->so_rcv.sb_flags |= SB_SEL;
2501 if (events & (POLLOUT | POLLWRNORM)) {
2502 selrecord(td, &so->so_snd.sb_sel);
2503 so->so_snd.sb_flags |= SB_SEL;
2507 SOCKBUF_UNLOCK(&so->so_rcv);
2508 SOCKBUF_UNLOCK(&so->so_snd);
2513 soo_kqfilter(struct file *fp, struct knote *kn)
2515 struct socket *so = kn->kn_fp->f_data;
2518 switch (kn->kn_filter) {
2520 if (so->so_options & SO_ACCEPTCONN)
2521 kn->kn_fop = &solisten_filtops;
2523 kn->kn_fop = &soread_filtops;
2527 kn->kn_fop = &sowrite_filtops;
2535 knlist_add(&sb->sb_sel.si_note, kn, 1);
2536 sb->sb_flags |= SB_KNOTE;
2542 * Some routines that return EOPNOTSUPP for entry points that are not
2543 * supported by a protocol. Fill in as needed.
2546 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
2553 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
2560 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2567 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2574 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
2581 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
2582 struct ifnet *ifp, struct thread *td)
2589 pru_disconnect_notsupp(struct socket *so)
2596 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
2603 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
2610 pru_rcvd_notsupp(struct socket *so, int flags)
2617 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
2624 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
2625 struct sockaddr *addr, struct mbuf *control, struct thread *td)
2632 * This isn't really a ``null'' operation, but it's the default one and
2633 * doesn't do anything destructive.
2636 pru_sense_null(struct socket *so, struct stat *sb)
2639 sb->st_blksize = so->so_snd.sb_hiwat;
2644 pru_shutdown_notsupp(struct socket *so)
2651 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
2658 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
2659 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
2666 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
2667 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2674 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
2682 filt_sordetach(struct knote *kn)
2684 struct socket *so = kn->kn_fp->f_data;
2686 SOCKBUF_LOCK(&so->so_rcv);
2687 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
2688 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
2689 so->so_rcv.sb_flags &= ~SB_KNOTE;
2690 SOCKBUF_UNLOCK(&so->so_rcv);
2695 filt_soread(struct knote *kn, long hint)
2699 so = kn->kn_fp->f_data;
2700 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2702 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
2703 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2704 kn->kn_flags |= EV_EOF;
2705 kn->kn_fflags = so->so_error;
2707 } else if (so->so_error) /* temporary udp error */
2709 else if (kn->kn_sfflags & NOTE_LOWAT)
2710 return (kn->kn_data >= kn->kn_sdata);
2712 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
2716 filt_sowdetach(struct knote *kn)
2718 struct socket *so = kn->kn_fp->f_data;
2720 SOCKBUF_LOCK(&so->so_snd);
2721 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
2722 if (knlist_empty(&so->so_snd.sb_sel.si_note))
2723 so->so_snd.sb_flags &= ~SB_KNOTE;
2724 SOCKBUF_UNLOCK(&so->so_snd);
2729 filt_sowrite(struct knote *kn, long hint)
2733 so = kn->kn_fp->f_data;
2734 SOCKBUF_LOCK_ASSERT(&so->so_snd);
2735 kn->kn_data = sbspace(&so->so_snd);
2736 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2737 kn->kn_flags |= EV_EOF;
2738 kn->kn_fflags = so->so_error;
2740 } else if (so->so_error) /* temporary udp error */
2742 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2743 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2745 else if (kn->kn_sfflags & NOTE_LOWAT)
2746 return (kn->kn_data >= kn->kn_sdata);
2748 return (kn->kn_data >= so->so_snd.sb_lowat);
2753 filt_solisten(struct knote *kn, long hint)
2755 struct socket *so = kn->kn_fp->f_data;
2757 kn->kn_data = so->so_qlen;
2758 return (! TAILQ_EMPTY(&so->so_comp));
2762 socheckuid(struct socket *so, uid_t uid)
2767 if (so->so_cred->cr_uid != uid)
2773 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
2779 error = sysctl_handle_int(oidp, &val, 0, req);
2780 if (error || !req->newptr )
2783 if (val < 1 || val > USHRT_MAX)
2791 * These functions are used by protocols to notify the socket layer (and its
2792 * consumers) of state changes in the sockets driven by protocol-side events.
2796 * Procedures to manipulate state flags of socket and do appropriate wakeups.
2798 * Normal sequence from the active (originating) side is that
2799 * soisconnecting() is called during processing of connect() call, resulting
2800 * in an eventual call to soisconnected() if/when the connection is
2801 * established. When the connection is torn down soisdisconnecting() is
2802 * called during processing of disconnect() call, and soisdisconnected() is
2803 * called when the connection to the peer is totally severed. The semantics
2804 * of these routines are such that connectionless protocols can call
2805 * soisconnected() and soisdisconnected() only, bypassing the in-progress
2806 * calls when setting up a ``connection'' takes no time.
2808 * From the passive side, a socket is created with two queues of sockets:
2809 * so_incomp for connections in progress and so_comp for connections already
2810 * made and awaiting user acceptance. As a protocol is preparing incoming
2811 * connections, it creates a socket structure queued on so_incomp by calling
2812 * sonewconn(). When the connection is established, soisconnected() is
2813 * called, and transfers the socket structure to so_comp, making it available
2816 * If a socket is closed with sockets on either so_incomp or so_comp, these
2817 * sockets are dropped.
2819 * If higher-level protocols are implemented in the kernel, the wakeups done
2820 * here will sometimes cause software-interrupt process scheduling.
2823 soisconnecting(struct socket *so)
2827 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
2828 so->so_state |= SS_ISCONNECTING;
2833 soisconnected(struct socket *so)
2835 struct socket *head;
2839 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
2840 so->so_state |= SS_ISCONNECTED;
2842 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
2843 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
2845 TAILQ_REMOVE(&head->so_incomp, so, so_list);
2847 so->so_qstate &= ~SQ_INCOMP;
2848 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
2850 so->so_qstate |= SQ_COMP;
2853 wakeup_one(&head->so_timeo);
2857 head->so_accf->so_accept_filter->accf_callback;
2858 so->so_upcallarg = head->so_accf->so_accept_filter_arg;
2859 so->so_rcv.sb_flags |= SB_UPCALL;
2860 so->so_options &= ~SO_ACCEPTFILTER;
2862 so->so_upcall(so, so->so_upcallarg, M_DONTWAIT);
2868 wakeup(&so->so_timeo);
2874 soisdisconnecting(struct socket *so)
2878 * Note: This code assumes that SOCK_LOCK(so) and
2879 * SOCKBUF_LOCK(&so->so_rcv) are the same.
2881 SOCKBUF_LOCK(&so->so_rcv);
2882 so->so_state &= ~SS_ISCONNECTING;
2883 so->so_state |= SS_ISDISCONNECTING;
2884 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
2885 sorwakeup_locked(so);
2886 SOCKBUF_LOCK(&so->so_snd);
2887 so->so_snd.sb_state |= SBS_CANTSENDMORE;
2888 sowwakeup_locked(so);
2889 wakeup(&so->so_timeo);
2893 soisdisconnected(struct socket *so)
2897 * Note: This code assumes that SOCK_LOCK(so) and
2898 * SOCKBUF_LOCK(&so->so_rcv) are the same.
2900 SOCKBUF_LOCK(&so->so_rcv);
2901 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
2902 so->so_state |= SS_ISDISCONNECTED;
2903 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
2904 sorwakeup_locked(so);
2905 SOCKBUF_LOCK(&so->so_snd);
2906 so->so_snd.sb_state |= SBS_CANTSENDMORE;
2907 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
2908 sowwakeup_locked(so);
2909 wakeup(&so->so_timeo);
2913 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
2916 sodupsockaddr(const struct sockaddr *sa, int mflags)
2918 struct sockaddr *sa2;
2920 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
2922 bcopy(sa, sa2, sa->sa_len);
2927 * Create an external-format (``xsocket'') structure using the information in
2928 * the kernel-format socket structure pointed to by so. This is done to
2929 * reduce the spew of irrelevant information over this interface, to isolate
2930 * user code from changes in the kernel structure, and potentially to provide
2931 * information-hiding if we decide that some of this information should be
2932 * hidden from users.
2935 sotoxsocket(struct socket *so, struct xsocket *xso)
2938 xso->xso_len = sizeof *xso;
2940 xso->so_type = so->so_type;
2941 xso->so_options = so->so_options;
2942 xso->so_linger = so->so_linger;
2943 xso->so_state = so->so_state;
2944 xso->so_pcb = so->so_pcb;
2945 xso->xso_protocol = so->so_proto->pr_protocol;
2946 xso->xso_family = so->so_proto->pr_domain->dom_family;
2947 xso->so_qlen = so->so_qlen;
2948 xso->so_incqlen = so->so_incqlen;
2949 xso->so_qlimit = so->so_qlimit;
2950 xso->so_timeo = so->so_timeo;
2951 xso->so_error = so->so_error;
2952 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
2953 xso->so_oobmark = so->so_oobmark;
2954 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
2955 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
2956 xso->so_uid = so->so_cred->cr_uid;