2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
30 * $FreeBSD: src/sys/kern/uipc_usrreq.c,v 1.54.2.10 2003/03/04 17:28:09 nectar Exp $
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/kernel.h>
36 #include <sys/domain.h>
37 #include <sys/fcntl.h>
38 #include <sys/malloc.h> /* XXX must be before <sys/file.h> */
41 #include <sys/filedesc.h>
43 #include <sys/nlookup.h>
44 #include <sys/protosw.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/resourcevar.h>
49 #include <sys/mount.h>
50 #include <sys/sysctl.h>
52 #include <sys/unpcb.h>
53 #include <sys/vnode.h>
55 #include <sys/file2.h>
56 #include <sys/spinlock2.h>
57 #include <sys/socketvar2.h>
58 #include <sys/msgport2.h>
60 #define UNP_DETACHED UNP_PRIVATE1
61 #define UNP_CONNECTING UNP_PRIVATE2
62 #define UNP_DROPPED UNP_PRIVATE3
64 #define UNP_ISATTACHED(unp) \
65 ((unp) != NULL && ((unp)->unp_flags & UNP_DETACHED) == 0)
68 #define UNP_ASSERT_TOKEN_HELD(unp) \
69 ASSERT_LWKT_TOKEN_HELD(lwkt_token_pool_lookup((unp)))
70 #else /* !INVARIANTS */
71 #define UNP_ASSERT_TOKEN_HELD(unp)
72 #endif /* INVARIANTS */
74 typedef struct unp_defdiscard {
75 struct unp_defdiscard *next;
79 static MALLOC_DEFINE(M_UNPCB, "unpcb", "unpcb struct");
80 static unp_gen_t unp_gencnt;
81 static u_int unp_count;
83 static struct unp_head unp_shead, unp_dhead;
85 static struct lwkt_token unp_token = LWKT_TOKEN_INITIALIZER(unp_token);
86 static int unp_defdiscard_nest;
87 static unp_defdiscard_t unp_defdiscard_base;
90 * Unix communications domain.
94 * rethink name space problems
95 * need a proper out-of-band
98 static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
99 static ino_t unp_ino = 1; /* prototype for fake inode numbers */
100 static struct spinlock unp_ino_spin = SPINLOCK_INITIALIZER(&unp_ino_spin, "unp_ino_spin");
102 static int unp_attach (struct socket *, struct pru_attach_info *);
103 static void unp_detach (struct unpcb *);
104 static int unp_bind (struct unpcb *,struct sockaddr *, struct thread *);
105 static int unp_connect (struct socket *,struct sockaddr *,
107 static void unp_disconnect(struct unpcb *, int);
108 static void unp_shutdown (struct unpcb *);
109 static void unp_gc (void);
110 static int unp_gc_clearmarks(struct file *, void *);
111 static int unp_gc_checkmarks(struct file *, void *);
112 static int unp_gc_checkrefs(struct file *, void *);
113 static int unp_revoke_gc_check(struct file *, void *);
114 static void unp_scan (struct mbuf *, void (*)(struct file *, void *),
116 static void unp_mark (struct file *, void *data);
117 static void unp_discard (struct file *, void *);
118 static int unp_internalize (struct mbuf *, struct thread *);
119 static int unp_listen (struct unpcb *, struct thread *);
120 static void unp_fp_externalize(struct lwp *lp, struct file *fp, int fd);
121 static int unp_find_lockref(struct sockaddr *nam, struct thread *td,
122 short type, struct unpcb **unp_ret);
123 static int unp_connect_pair(struct unpcb *unp, struct unpcb *unp2);
124 static void unp_drop(struct unpcb *unp, int error);
127 * SMP Considerations:
129 * Since unp_token will be automaticly released upon execution of
130 * blocking code, we need to reference unp_conn before any possible
131 * blocking code to prevent it from being ripped behind our back.
133 * Any adjustment to unp->unp_conn requires both the global unp_token
134 * AND the per-unp token (lwkt_token_pool_lookup(unp)) to be held.
136 * Any access to so_pcb to obtain unp requires the pool token for
140 /* NOTE: unp_token MUST be held */
142 unp_reference(struct unpcb *unp)
144 atomic_add_int(&unp->unp_refcnt, 1);
147 /* NOTE: unp_token MUST be held */
149 unp_free(struct unpcb *unp)
151 KKASSERT(unp->unp_refcnt > 0);
152 if (atomic_fetchadd_int(&unp->unp_refcnt, -1) == 1)
156 static __inline struct unpcb *
157 unp_getsocktoken(struct socket *so)
162 * The unp pointer is invalid until we verify that it is
163 * good by re-checking so_pcb AFTER obtaining the token.
165 while ((unp = so->so_pcb) != NULL) {
166 lwkt_getpooltoken(unp);
167 if (unp == so->so_pcb)
169 lwkt_relpooltoken(unp);
175 unp_reltoken(struct unpcb *unp)
178 lwkt_relpooltoken(unp);
182 unp_setflags(struct unpcb *unp, int flags)
184 atomic_set_int(&unp->unp_flags, flags);
188 unp_clrflags(struct unpcb *unp, int flags)
190 atomic_clear_int(&unp->unp_flags, flags);
194 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
195 * will sofree() it when we return.
198 uipc_abort(netmsg_t msg)
203 lwkt_gettoken(&unp_token);
204 unp = unp_getsocktoken(msg->base.nm_so);
206 if (UNP_ISATTACHED(unp)) {
207 unp_setflags(unp, UNP_DETACHED);
208 unp_drop(unp, ECONNABORTED);
216 lwkt_reltoken(&unp_token);
218 lwkt_replymsg(&msg->lmsg, error);
222 uipc_accept(netmsg_t msg)
227 lwkt_gettoken(&unp_token);
228 unp = unp_getsocktoken(msg->base.nm_so);
230 if (!UNP_ISATTACHED(unp)) {
233 struct unpcb *unp2 = unp->unp_conn;
236 * Pass back name of connected socket,
237 * if it was bound and we are still connected
238 * (our peer may have closed already!).
240 if (unp2 && unp2->unp_addr) {
242 *msg->accept.nm_nam = dup_sockaddr(
243 (struct sockaddr *)unp2->unp_addr);
246 *msg->accept.nm_nam = dup_sockaddr(&sun_noname);
252 lwkt_reltoken(&unp_token);
254 lwkt_replymsg(&msg->lmsg, error);
258 uipc_attach(netmsg_t msg)
262 lwkt_gettoken(&unp_token);
264 KASSERT(msg->base.nm_so->so_pcb == NULL, ("double unp attach"));
265 error = unp_attach(msg->base.nm_so, msg->attach.nm_ai);
267 lwkt_reltoken(&unp_token);
268 lwkt_replymsg(&msg->lmsg, error);
272 uipc_bind(netmsg_t msg)
277 lwkt_gettoken(&unp_token);
278 unp = unp_getsocktoken(msg->base.nm_so);
280 if (UNP_ISATTACHED(unp))
281 error = unp_bind(unp, msg->bind.nm_nam, msg->bind.nm_td);
286 lwkt_reltoken(&unp_token);
288 lwkt_replymsg(&msg->lmsg, error);
292 uipc_connect(netmsg_t msg)
296 error = unp_connect(msg->base.nm_so, msg->connect.nm_nam,
298 lwkt_replymsg(&msg->lmsg, error);
302 uipc_connect2(netmsg_t msg)
306 error = unp_connect2(msg->connect2.nm_so1, msg->connect2.nm_so2);
307 lwkt_replymsg(&msg->lmsg, error);
310 /* control is EOPNOTSUPP */
313 uipc_detach(netmsg_t msg)
318 lwkt_gettoken(&unp_token);
319 unp = unp_getsocktoken(msg->base.nm_so);
321 if (UNP_ISATTACHED(unp)) {
322 unp_setflags(unp, UNP_DETACHED);
331 lwkt_reltoken(&unp_token);
333 lwkt_replymsg(&msg->lmsg, error);
337 uipc_disconnect(netmsg_t msg)
342 lwkt_gettoken(&unp_token);
343 unp = unp_getsocktoken(msg->base.nm_so);
345 if (UNP_ISATTACHED(unp)) {
346 unp_disconnect(unp, 0);
353 lwkt_reltoken(&unp_token);
355 lwkt_replymsg(&msg->lmsg, error);
359 uipc_listen(netmsg_t msg)
364 lwkt_gettoken(&unp_token);
365 unp = unp_getsocktoken(msg->base.nm_so);
367 if (!UNP_ISATTACHED(unp) || unp->unp_vnode == NULL)
370 error = unp_listen(unp, msg->listen.nm_td);
373 lwkt_reltoken(&unp_token);
375 lwkt_replymsg(&msg->lmsg, error);
379 uipc_peeraddr(netmsg_t msg)
384 lwkt_gettoken(&unp_token);
385 unp = unp_getsocktoken(msg->base.nm_so);
387 if (!UNP_ISATTACHED(unp)) {
389 } else if (unp->unp_conn && unp->unp_conn->unp_addr) {
390 struct unpcb *unp2 = unp->unp_conn;
393 *msg->peeraddr.nm_nam = dup_sockaddr(
394 (struct sockaddr *)unp2->unp_addr);
399 * XXX: It seems that this test always fails even when
400 * connection is established. So, this else clause is
401 * added as workaround to return PF_LOCAL sockaddr.
403 *msg->peeraddr.nm_nam = dup_sockaddr(&sun_noname);
408 lwkt_reltoken(&unp_token);
410 lwkt_replymsg(&msg->lmsg, error);
414 uipc_rcvd(netmsg_t msg)
416 struct unpcb *unp, *unp2;
422 * so_pcb is only modified with both the global and the unp
425 so = msg->base.nm_so;
426 unp = unp_getsocktoken(so);
428 if (!UNP_ISATTACHED(unp)) {
433 switch (so->so_type) {
435 panic("uipc_rcvd DGRAM?");
439 if (unp->unp_conn == NULL)
441 unp2 = unp->unp_conn; /* protected by pool token */
444 * Because we are transfering mbufs directly to the
445 * peer socket we have to use SSB_STOP on the sender
446 * to prevent it from building up infinite mbufs.
448 * As in several places in this module w ehave to ref unp2
449 * to ensure that it does not get ripped out from under us
450 * if we block on the so2 token or in sowwakeup().
452 so2 = unp2->unp_socket;
454 lwkt_gettoken(&so2->so_rcv.ssb_token);
455 if (so->so_rcv.ssb_cc < so2->so_snd.ssb_hiwat &&
456 so->so_rcv.ssb_mbcnt < so2->so_snd.ssb_mbmax
458 atomic_clear_int(&so2->so_snd.ssb_flags, SSB_STOP);
462 lwkt_reltoken(&so2->so_rcv.ssb_token);
466 panic("uipc_rcvd unknown socktype");
472 lwkt_replymsg(&msg->lmsg, error);
475 /* pru_rcvoob is EOPNOTSUPP */
478 uipc_send(netmsg_t msg)
480 struct unpcb *unp, *unp2;
483 struct mbuf *control;
487 so = msg->base.nm_so;
488 control = msg->send.nm_control;
492 * so_pcb is only modified with both the global and the unp
495 so = msg->base.nm_so;
496 unp = unp_getsocktoken(so);
498 if (!UNP_ISATTACHED(unp)) {
503 if (msg->send.nm_flags & PRUS_OOB) {
508 wakeup_start_delayed();
510 if (control && (error = unp_internalize(control, msg->send.nm_td)))
513 switch (so->so_type) {
516 struct sockaddr *from;
518 if (msg->send.nm_addr) {
523 error = unp_find_lockref(msg->send.nm_addr,
524 msg->send.nm_td, so->so_type, &unp2);
529 * unp2 is locked and referenced.
531 * We could unlock unp2 now, since it was checked
536 if (unp->unp_conn == NULL) {
541 unp2 = unp->unp_conn;
544 /* NOTE: unp2 is referenced. */
545 so2 = unp2->unp_socket;
548 from = (struct sockaddr *)unp->unp_addr;
552 lwkt_gettoken(&so2->so_rcv.ssb_token);
553 if (ssb_appendaddr(&so2->so_rcv, from, m, control)) {
560 lwkt_reltoken(&so2->so_rcv.ssb_token);
568 /* Connect if not connected yet. */
570 * Note: A better implementation would complain
571 * if not equal to the peer's address.
573 if (!(so->so_state & SS_ISCONNECTED)) {
574 if (msg->send.nm_addr) {
575 error = unp_connect(so,
586 if (so->so_state & SS_CANTSENDMORE) {
590 if (unp->unp_conn == NULL)
591 panic("uipc_send connected but no connection?");
592 unp2 = unp->unp_conn;
593 so2 = unp2->unp_socket;
598 * Send to paired receive port, and then reduce
599 * send buffer hiwater marks to maintain backpressure.
602 lwkt_gettoken(&so2->so_rcv.ssb_token);
604 if (ssb_appendcontrol(&so2->so_rcv, m, control)) {
608 } else if (so->so_type == SOCK_SEQPACKET) {
609 sbappendrecord(&so2->so_rcv.sb, m);
612 sbappend(&so2->so_rcv.sb, m);
617 * Because we are transfering mbufs directly to the
618 * peer socket we have to use SSB_STOP on the sender
619 * to prevent it from building up infinite mbufs.
621 if (so2->so_rcv.ssb_cc >= so->so_snd.ssb_hiwat ||
622 so2->so_rcv.ssb_mbcnt >= so->so_snd.ssb_mbmax
624 atomic_set_int(&so->so_snd.ssb_flags, SSB_STOP);
626 lwkt_reltoken(&so2->so_rcv.ssb_token);
633 panic("uipc_send unknown socktype");
637 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN.
639 if (msg->send.nm_flags & PRUS_EOF) {
644 if (control && error != 0)
645 unp_dispose(control);
648 wakeup_end_delayed();
654 lwkt_replymsg(&msg->lmsg, error);
661 uipc_sense(netmsg_t msg)
668 so = msg->base.nm_so;
669 sb = msg->sense.nm_stat;
672 * so_pcb is only modified with both the global and the unp
675 unp = unp_getsocktoken(so);
677 if (!UNP_ISATTACHED(unp)) {
682 sb->st_blksize = so->so_snd.ssb_hiwat;
684 if (unp->unp_ino == 0) { /* make up a non-zero inode number */
685 spin_lock(&unp_ino_spin);
686 unp->unp_ino = unp_ino++;
687 spin_unlock(&unp_ino_spin);
689 sb->st_ino = unp->unp_ino;
693 lwkt_replymsg(&msg->lmsg, error);
697 uipc_shutdown(netmsg_t msg)
704 * so_pcb is only modified with both the global and the unp
707 so = msg->base.nm_so;
708 unp = unp_getsocktoken(so);
710 if (UNP_ISATTACHED(unp)) {
719 lwkt_replymsg(&msg->lmsg, error);
723 uipc_sockaddr(netmsg_t msg)
729 * so_pcb is only modified with both the global and the unp
732 unp = unp_getsocktoken(msg->base.nm_so);
734 if (UNP_ISATTACHED(unp)) {
736 *msg->sockaddr.nm_nam =
737 dup_sockaddr((struct sockaddr *)unp->unp_addr);
745 lwkt_replymsg(&msg->lmsg, error);
748 struct pr_usrreqs uipc_usrreqs = {
749 .pru_abort = uipc_abort,
750 .pru_accept = uipc_accept,
751 .pru_attach = uipc_attach,
752 .pru_bind = uipc_bind,
753 .pru_connect = uipc_connect,
754 .pru_connect2 = uipc_connect2,
755 .pru_control = pr_generic_notsupp,
756 .pru_detach = uipc_detach,
757 .pru_disconnect = uipc_disconnect,
758 .pru_listen = uipc_listen,
759 .pru_peeraddr = uipc_peeraddr,
760 .pru_rcvd = uipc_rcvd,
761 .pru_rcvoob = pr_generic_notsupp,
762 .pru_send = uipc_send,
763 .pru_sense = uipc_sense,
764 .pru_shutdown = uipc_shutdown,
765 .pru_sockaddr = uipc_sockaddr,
766 .pru_sosend = sosend,
767 .pru_soreceive = soreceive
771 uipc_ctloutput(netmsg_t msg)
774 struct sockopt *sopt;
778 so = msg->base.nm_so;
779 sopt = msg->ctloutput.nm_sopt;
781 lwkt_gettoken(&unp_token);
782 unp = unp_getsocktoken(so);
784 if (!UNP_ISATTACHED(unp)) {
789 switch (sopt->sopt_dir) {
791 switch (sopt->sopt_name) {
793 if (unp->unp_flags & UNP_HAVEPC)
794 soopt_from_kbuf(sopt, &unp->unp_peercred,
795 sizeof(unp->unp_peercred));
797 if (so->so_type == SOCK_STREAM)
799 else if (so->so_type == SOCK_SEQPACKET)
818 lwkt_reltoken(&unp_token);
820 lwkt_replymsg(&msg->lmsg, error);
824 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
825 * for stream sockets, although the total for sender and receiver is
826 * actually only PIPSIZ.
828 * Datagram sockets really use the sendspace as the maximum datagram size,
829 * and don't really want to reserve the sendspace. Their recvspace should
830 * be large enough for at least one max-size datagram plus address.
832 * We want the local send/recv space to be significant larger then lo0's
838 static u_long unpst_sendspace = PIPSIZ;
839 static u_long unpst_recvspace = PIPSIZ;
840 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
841 static u_long unpdg_recvspace = 4*1024;
843 static int unp_rights; /* file descriptors in flight */
844 static struct spinlock unp_spin = SPINLOCK_INITIALIZER(&unp_spin, "unp_spin");
846 SYSCTL_DECL(_net_local_seqpacket);
847 SYSCTL_DECL(_net_local_stream);
848 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
849 &unpst_sendspace, 0, "Size of stream socket send buffer");
850 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
851 &unpst_recvspace, 0, "Size of stream socket receive buffer");
853 SYSCTL_DECL(_net_local_dgram);
854 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
855 &unpdg_sendspace, 0, "Max datagram socket size");
856 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
857 &unpdg_recvspace, 0, "Size of datagram socket receive buffer");
859 SYSCTL_DECL(_net_local);
860 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
861 "File descriptors in flight");
864 unp_attach(struct socket *so, struct pru_attach_info *ai)
869 lwkt_gettoken(&unp_token);
871 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
872 switch (so->so_type) {
875 error = soreserve(so, unpst_sendspace, unpst_recvspace,
880 error = soreserve(so, unpdg_sendspace, unpdg_recvspace,
892 * In order to support sendfile we have to set either SSB_STOPSUPP
893 * or SSB_PREALLOC. Unix domain sockets use the SSB_STOP flow
896 if (so->so_type == SOCK_STREAM) {
897 atomic_set_int(&so->so_rcv.ssb_flags, SSB_STOPSUPP);
898 atomic_set_int(&so->so_snd.ssb_flags, SSB_STOPSUPP);
901 unp = kmalloc(sizeof(*unp), M_UNPCB, M_WAITOK | M_ZERO | M_NULLOK);
907 unp->unp_gencnt = ++unp_gencnt;
909 LIST_INIT(&unp->unp_refs);
910 unp->unp_socket = so;
911 unp->unp_rvnode = ai->fd_rdir; /* jail cruft XXX JH */
912 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
913 : &unp_shead, unp, unp_link);
914 so->so_pcb = (caddr_t)unp;
918 lwkt_reltoken(&unp_token);
923 unp_detach(struct unpcb *unp)
927 lwkt_gettoken(&unp_token);
928 lwkt_getpooltoken(unp);
930 LIST_REMOVE(unp, unp_link); /* both tokens required */
931 unp->unp_gencnt = ++unp_gencnt;
933 if (unp->unp_vnode) {
934 unp->unp_vnode->v_socket = NULL;
935 vrele(unp->unp_vnode);
936 unp->unp_vnode = NULL;
938 soisdisconnected(unp->unp_socket);
939 so = unp->unp_socket;
940 soreference(so); /* for delayed sorflush */
941 KKASSERT(so->so_pcb == unp);
942 so->so_pcb = NULL; /* both tokens required */
943 unp->unp_socket = NULL;
944 sofree(so); /* remove pcb ref */
948 * Normally the receive buffer is flushed later,
949 * in sofree, but if our receive buffer holds references
950 * to descriptors that are now garbage, we will dispose
951 * of those descriptor references after the garbage collector
952 * gets them (resulting in a "panic: closef: count < 0").
958 lwkt_relpooltoken(unp);
959 lwkt_reltoken(&unp_token);
961 KASSERT(unp->unp_conn == NULL, ("unp is still connected"));
962 KASSERT(LIST_EMPTY(&unp->unp_refs), ("unp still has references"));
965 kfree(unp->unp_addr, M_SONAME);
970 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
972 struct proc *p = td->td_proc;
973 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
977 struct nlookupdata nd;
978 char buf[SOCK_MAXADDRLEN];
980 ASSERT_LWKT_TOKEN_HELD(&unp_token);
981 UNP_ASSERT_TOKEN_HELD(unp);
983 if (unp->unp_vnode != NULL)
986 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
989 strncpy(buf, soun->sun_path, namelen);
990 buf[namelen] = 0; /* null-terminate the string */
991 error = nlookup_init(&nd, buf, UIO_SYSSPACE,
992 NLC_LOCKVP | NLC_CREATE | NLC_REFDVP);
994 error = nlookup(&nd);
995 if (error == 0 && nd.nl_nch.ncp->nc_vp != NULL)
1001 vattr.va_type = VSOCK;
1002 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask);
1003 error = VOP_NCREATE(&nd.nl_nch, nd.nl_dvp, &vp, nd.nl_cred, &vattr);
1005 if (unp->unp_vnode == NULL) {
1006 vp->v_socket = unp->unp_socket;
1007 unp->unp_vnode = vp;
1008 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam);
1011 vput(vp); /* late race */
1021 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1023 struct unpcb *unp, *unp2;
1024 int error, flags = 0;
1026 lwkt_gettoken(&unp_token);
1028 unp = unp_getsocktoken(so);
1029 if (!UNP_ISATTACHED(unp)) {
1034 if ((unp->unp_flags & UNP_CONNECTING) || unp->unp_conn != NULL) {
1039 flags = UNP_CONNECTING;
1040 unp_setflags(unp, flags);
1042 error = unp_find_lockref(nam, td, so->so_type, &unp2);
1047 * unp2 is locked and referenced.
1050 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1051 struct socket *so2, *so3;
1054 so2 = unp2->unp_socket;
1055 if (!(so2->so_options & SO_ACCEPTCONN) ||
1056 (so3 = sonewconn_faddr(so2, 0, NULL,
1057 TRUE /* keep ref */)) == NULL) {
1058 error = ECONNREFUSED;
1061 /* so3 has a socket reference. */
1063 unp3 = unp_getsocktoken(so3);
1064 if (!UNP_ISATTACHED(unp3)) {
1067 * Already aborted; we only need to drop the
1068 * socket reference held by sonewconn_faddr().
1071 error = ECONNREFUSED;
1074 unp_reference(unp3);
1077 * unp3 is locked and referenced.
1081 * Release so3 socket reference held by sonewconn_faddr().
1082 * Since we have referenced unp3, neither unp3 nor so3 will
1083 * be destroyed here.
1087 if (unp2->unp_addr != NULL) {
1088 unp3->unp_addr = (struct sockaddr_un *)
1089 dup_sockaddr((struct sockaddr *)unp2->unp_addr);
1093 * unp_peercred management:
1095 * The connecter's (client's) credentials are copied
1096 * from its process structure at the time of connect()
1099 cru2x(td->td_proc->p_ucred, &unp3->unp_peercred);
1100 unp_setflags(unp3, UNP_HAVEPC);
1102 * The receiver's (server's) credentials are copied
1103 * from the unp_peercred member of socket on which the
1104 * former called listen(); unp_listen() cached that
1105 * process's credentials at that time so we can use
1108 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1109 ("unp_connect: listener without cached peercred"));
1110 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1111 sizeof(unp->unp_peercred));
1112 unp_setflags(unp, UNP_HAVEPC);
1114 error = unp_connect_pair(unp, unp3);
1116 /* XXX we need a better name */
1120 /* Done with unp3 */
1124 error = unp_connect_pair(unp, unp2);
1131 unp_clrflags(unp, flags);
1134 lwkt_reltoken(&unp_token);
1139 * Connect two unix domain sockets together.
1141 * NOTE: Semantics for any change to unp_conn requires that the per-unp
1142 * pool token also be held.
1145 unp_connect2(struct socket *so, struct socket *so2)
1147 struct unpcb *unp, *unp2;
1150 lwkt_gettoken(&unp_token);
1151 if (so2->so_type != so->so_type) {
1152 lwkt_reltoken(&unp_token);
1153 return (EPROTOTYPE);
1155 unp = unp_getsocktoken(so);
1156 unp2 = unp_getsocktoken(so2);
1158 if (!UNP_ISATTACHED(unp)) {
1162 if (!UNP_ISATTACHED(unp2)) {
1163 error = ECONNREFUSED;
1167 if (unp->unp_conn != NULL) {
1171 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) &&
1172 unp2->unp_conn != NULL) {
1177 error = unp_connect_pair(unp, unp2);
1181 lwkt_reltoken(&unp_token);
1186 * Disconnect a unix domain socket pair.
1188 * NOTE: Semantics for any change to unp_conn requires that the per-unp
1189 * pool token also be held.
1192 unp_disconnect(struct unpcb *unp, int error)
1194 struct socket *so = unp->unp_socket;
1197 ASSERT_LWKT_TOKEN_HELD(&unp_token);
1198 UNP_ASSERT_TOKEN_HELD(unp);
1201 so->so_error = error;
1203 while ((unp2 = unp->unp_conn) != NULL) {
1204 lwkt_getpooltoken(unp2);
1205 if (unp2 == unp->unp_conn)
1207 lwkt_relpooltoken(unp2);
1211 /* unp2 is locked. */
1213 KASSERT((unp2->unp_flags & UNP_DROPPED) == 0, ("unp2 was dropped"));
1215 unp->unp_conn = NULL;
1217 switch (so->so_type) {
1219 LIST_REMOVE(unp, unp_reflink);
1220 soclrstate(so, SS_ISCONNECTED);
1224 case SOCK_SEQPACKET:
1226 * Keep a reference before clearing the unp_conn
1227 * to avoid racing uipc_detach()/uipc_abort() in
1230 unp_reference(unp2);
1231 KASSERT(unp2->unp_conn == unp, ("unp_conn mismatch"));
1232 unp2->unp_conn = NULL;
1234 soisdisconnected(so);
1235 soisdisconnected(unp2->unp_socket);
1241 lwkt_relpooltoken(unp2);
1246 unp_abort(struct unpcb *unp)
1248 lwkt_gettoken(&unp_token);
1250 lwkt_reltoken(&unp_token);
1255 prison_unpcb(struct thread *td, struct unpcb *unp)
1261 if ((p = td->td_proc) == NULL)
1263 if (!p->p_ucred->cr_prison)
1265 if (p->p_fd->fd_rdir == unp->unp_rvnode)
1271 unp_pcblist(SYSCTL_HANDLER_ARGS)
1274 struct unpcb *unp, **unp_list;
1276 struct unp_head *head;
1278 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1280 KKASSERT(curproc != NULL);
1283 * The process of preparing the PCB list is too time-consuming and
1284 * resource-intensive to repeat twice on every request.
1286 if (req->oldptr == NULL) {
1288 req->oldidx = (n + n/8) * sizeof(struct xunpcb);
1292 if (req->newptr != NULL)
1295 lwkt_gettoken(&unp_token);
1298 * OK, now we're committed to doing something.
1300 gencnt = unp_gencnt;
1303 unp_list = kmalloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1305 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1306 unp = LIST_NEXT(unp, unp_link)) {
1307 if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->td, unp))
1308 unp_list[i++] = unp;
1310 n = i; /* in case we lost some during malloc */
1313 for (i = 0; i < n; i++) {
1315 if (unp->unp_gencnt <= gencnt) {
1317 xu.xu_len = sizeof xu;
1320 * XXX - need more locking here to protect against
1321 * connect/disconnect races for SMP.
1324 bcopy(unp->unp_addr, &xu.xu_addr,
1325 unp->unp_addr->sun_len);
1326 if (unp->unp_conn && unp->unp_conn->unp_addr)
1327 bcopy(unp->unp_conn->unp_addr,
1329 unp->unp_conn->unp_addr->sun_len);
1330 bcopy(unp, &xu.xu_unp, sizeof *unp);
1331 sotoxsocket(unp->unp_socket, &xu.xu_socket);
1332 error = SYSCTL_OUT(req, &xu, sizeof xu);
1335 lwkt_reltoken(&unp_token);
1336 kfree(unp_list, M_TEMP);
1341 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1342 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1343 "List of active local datagram sockets");
1344 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1345 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1346 "List of active local stream sockets");
1347 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLFLAG_RD,
1348 (caddr_t)(long)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1349 "List of active local seqpacket stream sockets");
1352 unp_shutdown(struct unpcb *unp)
1356 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1357 unp->unp_socket->so_type == SOCK_SEQPACKET) &&
1358 unp->unp_conn != NULL && (so = unp->unp_conn->unp_socket)) {
1367 lwkt_gettoken(&unp_token);
1368 lwkt_reltoken(&unp_token);
1373 unp_externalize(struct mbuf *rights)
1375 struct thread *td = curthread;
1376 struct proc *p = td->td_proc; /* XXX */
1377 struct lwp *lp = td->td_lwp;
1378 struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
1383 int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm))
1384 / sizeof (struct file *);
1387 lwkt_gettoken(&unp_token);
1390 * if the new FD's will not fit, then we free them all
1392 if (!fdavail(p, newfds)) {
1393 rp = (struct file **)CMSG_DATA(cm);
1394 for (i = 0; i < newfds; i++) {
1397 * zero the pointer before calling unp_discard,
1398 * since it may end up in unp_gc()..
1401 unp_discard(fp, NULL);
1403 lwkt_reltoken(&unp_token);
1408 * now change each pointer to an fd in the global table to
1409 * an integer that is the index to the local fd table entry
1410 * that we set up to point to the global one we are transferring.
1411 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1412 * then do it in forward order. In that case, an integer will
1413 * always come in the same place or before its corresponding
1414 * struct file pointer.
1415 * If sizeof (struct file *) is smaller than sizeof int, then
1416 * do it in reverse order.
1418 if (sizeof (struct file *) >= sizeof (int)) {
1419 fdp = (int *)CMSG_DATA(cm);
1420 rp = (struct file **)CMSG_DATA(cm);
1421 for (i = 0; i < newfds; i++) {
1422 if (fdalloc(p, 0, &f))
1423 panic("unp_externalize");
1425 unp_fp_externalize(lp, fp, f);
1429 fdp = (int *)CMSG_DATA(cm) + newfds - 1;
1430 rp = (struct file **)CMSG_DATA(cm) + newfds - 1;
1431 for (i = 0; i < newfds; i++) {
1432 if (fdalloc(p, 0, &f))
1433 panic("unp_externalize");
1435 unp_fp_externalize(lp, fp, f);
1441 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1444 cm->cmsg_len = CMSG_LEN(newfds * sizeof(int));
1445 rights->m_len = cm->cmsg_len;
1447 lwkt_reltoken(&unp_token);
1452 unp_fp_externalize(struct lwp *lp, struct file *fp, int fd)
1457 lwkt_gettoken(&unp_token);
1461 if (fp->f_flag & FREVOKED) {
1462 kprintf("Warning: revoked fp exiting unix socket\n");
1464 error = falloc(lp, &fx, NULL);
1466 fsetfd(lp->lwp_proc->p_fd, fx, fd);
1468 fsetfd(lp->lwp_proc->p_fd, NULL, fd);
1471 fsetfd(lp->lwp_proc->p_fd, fp, fd);
1474 spin_lock(&unp_spin);
1477 spin_unlock(&unp_spin);
1480 lwkt_reltoken(&unp_token);
1487 LIST_INIT(&unp_dhead);
1488 LIST_INIT(&unp_shead);
1489 spin_init(&unp_spin, "unpinit");
1493 unp_internalize(struct mbuf *control, struct thread *td)
1495 struct proc *p = td->td_proc;
1496 struct filedesc *fdescp;
1497 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1501 struct cmsgcred *cmcred;
1507 lwkt_gettoken(&unp_token);
1510 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) ||
1511 cm->cmsg_level != SOL_SOCKET ||
1512 CMSG_ALIGN(cm->cmsg_len) != control->m_len) {
1518 * Fill in credential information.
1520 if (cm->cmsg_type == SCM_CREDS) {
1521 cmcred = (struct cmsgcred *)CMSG_DATA(cm);
1522 cmcred->cmcred_pid = p->p_pid;
1523 cmcred->cmcred_uid = p->p_ucred->cr_ruid;
1524 cmcred->cmcred_gid = p->p_ucred->cr_rgid;
1525 cmcred->cmcred_euid = p->p_ucred->cr_uid;
1526 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups,
1528 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1529 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i];
1535 * cmsghdr may not be aligned, do not allow calculation(s) to
1538 if (cm->cmsg_len < CMSG_LEN(0)) {
1543 oldfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof (int);
1546 * check that all the FDs passed in refer to legal OPEN files
1547 * If not, reject the entire operation.
1549 fdp = (int *)CMSG_DATA(cm);
1550 for (i = 0; i < oldfds; i++) {
1552 if ((unsigned)fd >= fdescp->fd_nfiles ||
1553 fdescp->fd_files[fd].fp == NULL) {
1557 if (fdescp->fd_files[fd].fp->f_type == DTYPE_KQUEUE) {
1563 * Now replace the integer FDs with pointers to
1564 * the associated global file table entry..
1565 * Allocate a bigger buffer as necessary. But if an cluster is not
1566 * enough, return E2BIG.
1568 newlen = CMSG_LEN(oldfds * sizeof(struct file *));
1569 if (newlen > MCLBYTES) {
1573 if (newlen - control->m_len > M_TRAILINGSPACE(control)) {
1574 if (control->m_flags & M_EXT) {
1578 MCLGET(control, M_WAITOK);
1579 if (!(control->m_flags & M_EXT)) {
1584 /* copy the data to the cluster */
1585 memcpy(mtod(control, char *), cm, cm->cmsg_len);
1586 cm = mtod(control, struct cmsghdr *);
1590 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1593 cm->cmsg_len = newlen;
1594 control->m_len = CMSG_ALIGN(newlen);
1597 * Transform the file descriptors into struct file pointers.
1598 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1599 * then do it in reverse order so that the int won't get until
1601 * If sizeof (struct file *) is smaller than sizeof int, then
1602 * do it in forward order.
1604 if (sizeof (struct file *) >= sizeof (int)) {
1605 fdp = (int *)CMSG_DATA(cm) + oldfds - 1;
1606 rp = (struct file **)CMSG_DATA(cm) + oldfds - 1;
1607 for (i = 0; i < oldfds; i++) {
1608 fp = fdescp->fd_files[*fdp--].fp;
1611 spin_lock(&unp_spin);
1614 spin_unlock(&unp_spin);
1617 fdp = (int *)CMSG_DATA(cm);
1618 rp = (struct file **)CMSG_DATA(cm);
1619 for (i = 0; i < oldfds; i++) {
1620 fp = fdescp->fd_files[*fdp++].fp;
1623 spin_lock(&unp_spin);
1626 spin_unlock(&unp_spin);
1631 lwkt_reltoken(&unp_token);
1636 * Garbage collect in-transit file descriptors that get lost due to
1637 * loops (i.e. when a socket is sent to another process over itself,
1638 * and more complex situations).
1640 * NOT MPSAFE - TODO socket flush code and maybe closef. Rest is MPSAFE.
1643 struct unp_gc_info {
1644 struct file **extra_ref;
1645 struct file *locked_fp;
1654 struct unp_gc_info info;
1655 static boolean_t unp_gcing;
1660 * Only one gc can be in-progress at any given moment
1662 spin_lock(&unp_spin);
1664 spin_unlock(&unp_spin);
1668 spin_unlock(&unp_spin);
1670 lwkt_gettoken(&unp_token);
1673 * Before going through all this, set all FDs to be NOT defered
1674 * and NOT externally accessible (not marked). During the scan
1675 * a fd can be marked externally accessible but we may or may not
1676 * be able to immediately process it (controlled by FDEFER).
1678 * If we loop sleep a bit. The complexity of the topology can cause
1679 * multiple loops. Also failure to acquire the socket's so_rcv
1680 * token can cause us to loop.
1682 allfiles_scan_exclusive(unp_gc_clearmarks, NULL);
1685 allfiles_scan_exclusive(unp_gc_checkmarks, &info);
1687 tsleep(&info, 0, "gcagain", 1);
1688 } while (info.defer);
1691 * We grab an extra reference to each of the file table entries
1692 * that are not otherwise accessible and then free the rights
1693 * that are stored in messages on them.
1695 * The bug in the orginal code is a little tricky, so I'll describe
1696 * what's wrong with it here.
1698 * It is incorrect to simply unp_discard each entry for f_msgcount
1699 * times -- consider the case of sockets A and B that contain
1700 * references to each other. On a last close of some other socket,
1701 * we trigger a gc since the number of outstanding rights (unp_rights)
1702 * is non-zero. If during the sweep phase the gc code un_discards,
1703 * we end up doing a (full) closef on the descriptor. A closef on A
1704 * results in the following chain. Closef calls soo_close, which
1705 * calls soclose. Soclose calls first (through the switch
1706 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1707 * returns because the previous instance had set unp_gcing, and
1708 * we return all the way back to soclose, which marks the socket
1709 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush
1710 * to free up the rights that are queued in messages on the socket A,
1711 * i.e., the reference on B. The sorflush calls via the dom_dispose
1712 * switch unp_dispose, which unp_scans with unp_discard. This second
1713 * instance of unp_discard just calls closef on B.
1715 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1716 * which results in another closef on A. Unfortunately, A is already
1717 * being closed, and the descriptor has already been marked with
1718 * SS_NOFDREF, and soclose panics at this point.
1720 * Here, we first take an extra reference to each inaccessible
1721 * descriptor. Then, we call sorflush ourself, since we know
1722 * it is a Unix domain socket anyhow. After we destroy all the
1723 * rights carried in messages, we do a last closef to get rid
1724 * of our extra reference. This is the last close, and the
1725 * unp_detach etc will shut down the socket.
1727 * 91/09/19, bsy@cs.cmu.edu
1729 info.extra_ref = kmalloc(256 * sizeof(struct file *), M_FILE, M_WAITOK);
1730 info.maxindex = 256;
1737 allfiles_scan_exclusive(unp_gc_checkrefs, &info);
1740 * For each FD on our hit list, do the following two things
1742 for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) {
1743 struct file *tfp = *fpp;
1744 if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL)
1745 sorflush((struct socket *)(tfp->f_data));
1747 for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp)
1749 } while (info.index == info.maxindex);
1751 lwkt_reltoken(&unp_token);
1753 kfree((caddr_t)info.extra_ref, M_FILE);
1758 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1761 unp_gc_checkrefs(struct file *fp, void *data)
1763 struct unp_gc_info *info = data;
1765 if (fp->f_count == 0)
1767 if (info->index == info->maxindex)
1771 * If all refs are from msgs, and it's not marked accessible
1772 * then it must be referenced from some unreachable cycle
1773 * of (shut-down) FDs, so include it in our
1774 * list of FDs to remove
1776 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1777 info->extra_ref[info->index++] = fp;
1784 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1787 unp_gc_clearmarks(struct file *fp, void *data __unused)
1789 atomic_clear_int(&fp->f_flag, FMARK | FDEFER);
1794 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1797 unp_gc_checkmarks(struct file *fp, void *data)
1799 struct unp_gc_info *info = data;
1803 * If the file is not open, skip it. Make sure it isn't marked
1804 * defered or we could loop forever, in case we somehow race
1807 if (fp->f_count == 0) {
1808 if (fp->f_flag & FDEFER)
1809 atomic_clear_int(&fp->f_flag, FDEFER);
1813 * If we already marked it as 'defer' in a
1814 * previous pass, then try process it this time
1817 if (fp->f_flag & FDEFER) {
1818 atomic_clear_int(&fp->f_flag, FDEFER);
1821 * if it's not defered, then check if it's
1822 * already marked.. if so skip it
1824 if (fp->f_flag & FMARK)
1827 * If all references are from messages
1828 * in transit, then skip it. it's not
1829 * externally accessible.
1831 if (fp->f_count == fp->f_msgcount)
1834 * If it got this far then it must be
1835 * externally accessible.
1837 atomic_set_int(&fp->f_flag, FMARK);
1841 * either it was defered, or it is externally
1842 * accessible and not already marked so.
1843 * Now check if it is possibly one of OUR sockets.
1845 if (fp->f_type != DTYPE_SOCKET ||
1846 (so = (struct socket *)fp->f_data) == NULL) {
1849 if (so->so_proto->pr_domain != &localdomain ||
1850 !(so->so_proto->pr_flags & PR_RIGHTS)) {
1855 * So, Ok, it's one of our sockets and it IS externally accessible
1856 * (or was defered). Now we look to see if we hold any file
1857 * descriptors in its message buffers. Follow those links and mark
1858 * them as accessible too.
1860 * We are holding multiple spinlocks here, if we cannot get the
1861 * token non-blocking defer until the next loop.
1863 info->locked_fp = fp;
1864 if (lwkt_trytoken(&so->so_rcv.ssb_token)) {
1865 unp_scan(so->so_rcv.ssb_mb, unp_mark, info);
1866 lwkt_reltoken(&so->so_rcv.ssb_token);
1868 atomic_set_int(&fp->f_flag, FDEFER);
1875 * Scan all unix domain sockets and replace any revoked file pointers
1876 * found with the dummy file pointer fx. We don't worry about races
1877 * against file pointers being read out as those are handled in the
1881 #define REVOKE_GC_MAXFILES 32
1883 struct unp_revoke_gc_info {
1885 struct file *fary[REVOKE_GC_MAXFILES];
1890 unp_revoke_gc(struct file *fx)
1892 struct unp_revoke_gc_info info;
1895 lwkt_gettoken(&unp_token);
1899 allfiles_scan_exclusive(unp_revoke_gc_check, &info);
1900 for (i = 0; i < info.fcount; ++i)
1901 unp_fp_externalize(NULL, info.fary[i], -1);
1902 } while (info.fcount == REVOKE_GC_MAXFILES);
1903 lwkt_reltoken(&unp_token);
1907 * Check for and replace revoked descriptors.
1909 * WARNING: This routine is not allowed to block.
1912 unp_revoke_gc_check(struct file *fps, void *vinfo)
1914 struct unp_revoke_gc_info *info = vinfo;
1925 * Is this a unix domain socket with rights-passing abilities?
1927 if (fps->f_type != DTYPE_SOCKET)
1929 if ((so = (struct socket *)fps->f_data) == NULL)
1931 if (so->so_proto->pr_domain != &localdomain)
1933 if ((so->so_proto->pr_flags & PR_RIGHTS) == 0)
1937 * Scan the mbufs for control messages and replace any revoked
1938 * descriptors we find.
1940 lwkt_gettoken(&so->so_rcv.ssb_token);
1941 m0 = so->so_rcv.ssb_mb;
1943 for (m = m0; m; m = m->m_next) {
1944 if (m->m_type != MT_CONTROL)
1946 if (m->m_len < sizeof(*cm))
1948 cm = mtod(m, struct cmsghdr *);
1949 if (cm->cmsg_level != SOL_SOCKET ||
1950 cm->cmsg_type != SCM_RIGHTS) {
1953 qfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof(void *);
1954 rp = (struct file **)CMSG_DATA(cm);
1955 for (i = 0; i < qfds; i++) {
1957 if (fp->f_flag & FREVOKED) {
1958 kprintf("Warning: Removing revoked fp from unix domain socket queue\n");
1960 info->fx->f_msgcount++;
1963 info->fary[info->fcount++] = fp;
1965 if (info->fcount == REVOKE_GC_MAXFILES)
1968 if (info->fcount == REVOKE_GC_MAXFILES)
1972 if (info->fcount == REVOKE_GC_MAXFILES)
1975 lwkt_reltoken(&so->so_rcv.ssb_token);
1978 * Stop the scan if we filled up our array.
1980 if (info->fcount == REVOKE_GC_MAXFILES)
1986 * Dispose of the fp's stored in a mbuf.
1988 * The dds loop can cause additional fps to be entered onto the
1989 * list while it is running, flattening out the operation and avoiding
1990 * a deep kernel stack recursion.
1993 unp_dispose(struct mbuf *m)
1995 unp_defdiscard_t dds;
1997 lwkt_gettoken(&unp_token);
1998 ++unp_defdiscard_nest;
2000 unp_scan(m, unp_discard, NULL);
2002 if (unp_defdiscard_nest == 1) {
2003 while ((dds = unp_defdiscard_base) != NULL) {
2004 unp_defdiscard_base = dds->next;
2005 closef(dds->fp, NULL);
2006 kfree(dds, M_UNPCB);
2009 --unp_defdiscard_nest;
2010 lwkt_reltoken(&unp_token);
2014 unp_listen(struct unpcb *unp, struct thread *td)
2016 struct proc *p = td->td_proc;
2018 ASSERT_LWKT_TOKEN_HELD(&unp_token);
2019 UNP_ASSERT_TOKEN_HELD(unp);
2022 cru2x(p->p_ucred, &unp->unp_peercred);
2023 unp_setflags(unp, UNP_HAVEPCCACHED);
2028 unp_scan(struct mbuf *m0, void (*op)(struct file *, void *), void *data)
2037 for (m = m0; m; m = m->m_next) {
2038 if (m->m_type == MT_CONTROL &&
2039 m->m_len >= sizeof(*cm)) {
2040 cm = mtod(m, struct cmsghdr *);
2041 if (cm->cmsg_level != SOL_SOCKET ||
2042 cm->cmsg_type != SCM_RIGHTS)
2044 qfds = (cm->cmsg_len - CMSG_LEN(0)) /
2046 rp = (struct file **)CMSG_DATA(cm);
2047 for (i = 0; i < qfds; i++)
2049 break; /* XXX, but saves time */
2057 * Mark visibility. info->defer is recalculated on every pass.
2060 unp_mark(struct file *fp, void *data)
2062 struct unp_gc_info *info = data;
2064 if ((fp->f_flag & FMARK) == 0) {
2066 atomic_set_int(&fp->f_flag, FMARK | FDEFER);
2067 } else if (fp->f_flag & FDEFER) {
2073 * Discard a fp previously held in a unix domain socket mbuf. To
2074 * avoid blowing out the kernel stack due to contrived chain-reactions
2075 * we may have to defer the operation to a higher procedural level.
2077 * Caller holds unp_token
2080 unp_discard(struct file *fp, void *data __unused)
2082 unp_defdiscard_t dds;
2084 spin_lock(&unp_spin);
2087 spin_unlock(&unp_spin);
2089 if (unp_defdiscard_nest) {
2090 dds = kmalloc(sizeof(*dds), M_UNPCB, M_WAITOK|M_ZERO);
2092 dds->next = unp_defdiscard_base;
2093 unp_defdiscard_base = dds;
2100 unp_find_lockref(struct sockaddr *nam, struct thread *td, short type,
2101 struct unpcb **unp_ret)
2103 struct proc *p = td->td_proc;
2104 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
2105 struct vnode *vp = NULL;
2109 struct nlookupdata nd;
2110 char buf[SOCK_MAXADDRLEN];
2114 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
2119 strncpy(buf, soun->sun_path, len);
2122 error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW);
2124 error = nlookup(&nd);
2126 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2133 if (vp->v_type != VSOCK) {
2137 error = VOP_EACCESS(vp, VWRITE, p->p_ucred);
2142 error = ECONNREFUSED;
2145 if (so->so_type != type) {
2150 /* Lock this unp. */
2151 unp = unp_getsocktoken(so);
2152 if (!UNP_ISATTACHED(unp)) {
2154 error = ECONNREFUSED;
2157 /* And keep this unp referenced. */
2170 unp_connect_pair(struct unpcb *unp, struct unpcb *unp2)
2172 struct socket *so = unp->unp_socket;
2173 struct socket *so2 = unp2->unp_socket;
2175 ASSERT_LWKT_TOKEN_HELD(&unp_token);
2176 UNP_ASSERT_TOKEN_HELD(unp);
2177 UNP_ASSERT_TOKEN_HELD(unp2);
2179 KASSERT(so->so_type == so2->so_type,
2180 ("socket type mismatch, so %d, so2 %d", so->so_type, so2->so_type));
2182 if (!UNP_ISATTACHED(unp))
2184 if (!UNP_ISATTACHED(unp2))
2185 return ECONNREFUSED;
2187 KASSERT(unp->unp_conn == NULL, ("unp is already connected"));
2188 unp->unp_conn = unp2;
2190 switch (so->so_type) {
2192 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
2197 case SOCK_SEQPACKET:
2198 KASSERT(unp2->unp_conn == NULL, ("unp2 is already connected"));
2199 unp2->unp_conn = unp;
2205 panic("unp_connect_pair: unknown socket type %d", so->so_type);
2211 unp_drop(struct unpcb *unp, int error)
2215 ASSERT_LWKT_TOKEN_HELD(&unp_token);
2216 UNP_ASSERT_TOKEN_HELD(unp);
2217 KASSERT(unp->unp_flags & UNP_DETACHED, ("unp is not detached"));
2219 unp_disconnect(unp, error);
2221 while ((unp2 = LIST_FIRST(&unp->unp_refs)) != NULL) {
2222 lwkt_getpooltoken(unp2);
2223 unp_disconnect(unp2, ECONNRESET);
2224 lwkt_relpooltoken(unp2);
2226 unp_setflags(unp, UNP_DROPPED);