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. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
34 * $FreeBSD: src/sys/kern/uipc_usrreq.c,v 1.54.2.10 2003/03/04 17:28:09 nectar Exp $
35 * $DragonFly: src/sys/kern/uipc_usrreq.c,v 1.29 2006/09/05 00:55:45 dillon Exp $
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/domain.h>
42 #include <sys/fcntl.h>
43 #include <sys/malloc.h> /* XXX must be before <sys/file.h> */
46 #include <sys/filedesc.h>
48 #include <sys/nlookup.h>
49 #include <sys/protosw.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/resourcevar.h>
54 #include <sys/mount.h>
55 #include <sys/sysctl.h>
57 #include <sys/unpcb.h>
58 #include <sys/vnode.h>
59 #include <sys/file2.h>
60 #include <sys/spinlock2.h>
62 #include <vm/vm_zone.h>
64 static struct vm_zone *unp_zone;
65 static unp_gen_t unp_gencnt;
66 static u_int unp_count;
68 static struct unp_head unp_shead, unp_dhead;
71 * Unix communications domain.
75 * rethink name space problems
76 * need a proper out-of-band
79 static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
80 static ino_t unp_ino; /* prototype for fake inode numbers */
82 static int unp_attach (struct socket *, struct pru_attach_info *);
83 static void unp_detach (struct unpcb *);
84 static int unp_bind (struct unpcb *,struct sockaddr *, struct thread *);
85 static int unp_connect (struct socket *,struct sockaddr *,
87 static void unp_disconnect (struct unpcb *);
88 static void unp_shutdown (struct unpcb *);
89 static void unp_drop (struct unpcb *, int);
90 static void unp_gc (void);
91 static int unp_gc_clearmarks(struct file *, void *);
92 static int unp_gc_checkmarks(struct file *, void *);
93 static int unp_gc_checkrefs(struct file *, void *);
94 static void unp_scan (struct mbuf *, void (*)(struct file *, void *),
96 static void unp_mark (struct file *, void *data);
97 static void unp_discard (struct file *, void *);
98 static int unp_internalize (struct mbuf *, struct thread *);
99 static int unp_listen (struct unpcb *, struct thread *);
102 uipc_abort(struct socket *so)
104 struct unpcb *unp = so->so_pcb;
108 unp_drop(unp, ECONNABORTED);
115 uipc_accept(struct socket *so, struct sockaddr **nam)
117 struct unpcb *unp = so->so_pcb;
123 * Pass back name of connected socket,
124 * if it was bound and we are still connected
125 * (our peer may have closed already!).
127 if (unp->unp_conn && unp->unp_conn->unp_addr) {
128 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr);
130 *nam = dup_sockaddr((struct sockaddr *)&sun_noname);
136 uipc_attach(struct socket *so, int proto, struct pru_attach_info *ai)
138 struct unpcb *unp = so->so_pcb;
142 return unp_attach(so, ai);
146 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
148 struct unpcb *unp = so->so_pcb;
152 return unp_bind(unp, nam, td);
156 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
158 struct unpcb *unp = so->so_pcb;
162 return unp_connect(so, nam, td);
166 uipc_connect2(struct socket *so1, struct socket *so2)
168 struct unpcb *unp = so1->so_pcb;
173 return unp_connect2(so1, so2);
176 /* control is EOPNOTSUPP */
179 uipc_detach(struct socket *so)
181 struct unpcb *unp = so->so_pcb;
191 uipc_disconnect(struct socket *so)
193 struct unpcb *unp = so->so_pcb;
202 uipc_listen(struct socket *so, struct thread *td)
204 struct unpcb *unp = so->so_pcb;
206 if (unp == NULL || unp->unp_vnode == NULL)
208 return unp_listen(unp, td);
212 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
214 struct unpcb *unp = so->so_pcb;
218 if (unp->unp_conn && unp->unp_conn->unp_addr)
219 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr);
222 * XXX: It seems that this test always fails even when
223 * connection is established. So, this else clause is
224 * added as workaround to return PF_LOCAL sockaddr.
226 *nam = dup_sockaddr((struct sockaddr *)&sun_noname);
232 uipc_rcvd(struct socket *so, int flags)
234 struct unpcb *unp = so->so_pcb;
240 switch (so->so_type) {
242 panic("uipc_rcvd DGRAM?");
246 if (unp->unp_conn == NULL)
248 so2 = unp->unp_conn->unp_socket;
250 * Adjust backpressure on sender
251 * and wakeup any waiting to write.
253 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt;
254 unp->unp_mbcnt = so->so_rcv.sb_mbcnt;
256 so2->so_snd.sb_hiwat + unp->unp_cc - so->so_rcv.sb_cc;
257 chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
258 newhiwat, RLIM_INFINITY);
259 unp->unp_cc = so->so_rcv.sb_cc;
264 panic("uipc_rcvd unknown socktype");
269 /* pru_rcvoob is EOPNOTSUPP */
272 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
273 struct mbuf *control, struct thread *td)
276 struct unpcb *unp = so->so_pcb;
284 if (flags & PRUS_OOB) {
289 if (control && (error = unp_internalize(control, td)))
292 switch (so->so_type) {
295 struct sockaddr *from;
302 error = unp_connect(so, nam, td);
306 if (unp->unp_conn == NULL) {
311 so2 = unp->unp_conn->unp_socket;
313 from = (struct sockaddr *)unp->unp_addr;
316 if (sbappendaddr(&so2->so_rcv, from, m, control)) {
328 /* Connect if not connected yet. */
330 * Note: A better implementation would complain
331 * if not equal to the peer's address.
333 if (!(so->so_state & SS_ISCONNECTED)) {
335 error = unp_connect(so, nam, td);
344 if (so->so_state & SS_CANTSENDMORE) {
348 if (unp->unp_conn == NULL)
349 panic("uipc_send connected but no connection?");
350 so2 = unp->unp_conn->unp_socket;
352 * Send to paired receive port, and then reduce
353 * send buffer hiwater marks to maintain backpressure.
357 if (sbappendcontrol(&so2->so_rcv, m, control))
360 sbappend(&so2->so_rcv, m);
361 so->so_snd.sb_mbmax -=
362 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt;
363 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt;
364 newhiwat = so->so_snd.sb_hiwat -
365 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc);
366 chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
367 newhiwat, RLIM_INFINITY);
368 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc;
374 panic("uipc_send unknown socktype");
378 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN.
380 if (flags & PRUS_EOF) {
385 if (control && error != 0)
386 unp_dispose(control);
397 uipc_sense(struct socket *so, struct stat *sb)
399 struct unpcb *unp = so->so_pcb;
404 sb->st_blksize = so->so_snd.sb_hiwat;
405 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) {
406 so2 = unp->unp_conn->unp_socket;
407 sb->st_blksize += so2->so_rcv.sb_cc;
410 if (unp->unp_ino == 0) /* make up a non-zero inode number */
411 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
412 sb->st_ino = unp->unp_ino;
417 uipc_shutdown(struct socket *so)
419 struct unpcb *unp = so->so_pcb;
429 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
431 struct unpcb *unp = so->so_pcb;
436 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr);
440 struct pr_usrreqs uipc_usrreqs = {
441 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect,
442 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect,
443 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp,
444 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr,
445 sosend, soreceive, sopoll
449 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
451 struct unpcb *unp = so->so_pcb;
454 switch (sopt->sopt_dir) {
456 switch (sopt->sopt_name) {
458 if (unp->unp_flags & UNP_HAVEPC)
459 error = sooptcopyout(sopt, &unp->unp_peercred,
460 sizeof(unp->unp_peercred));
462 if (so->so_type == SOCK_STREAM)
482 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
483 * for stream sockets, although the total for sender and receiver is
484 * actually only PIPSIZ.
485 * Datagram sockets really use the sendspace as the maximum datagram size,
486 * and don't really want to reserve the sendspace. Their recvspace should
487 * be large enough for at least one max-size datagram plus address.
492 static u_long unpst_sendspace = PIPSIZ;
493 static u_long unpst_recvspace = PIPSIZ;
494 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
495 static u_long unpdg_recvspace = 4*1024;
497 static int unp_rights; /* file descriptors in flight */
498 static struct spinlock unp_spin = SPINLOCK_INITIALIZER(&unp_spin);
500 SYSCTL_DECL(_net_local_stream);
501 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
502 &unpst_sendspace, 0, "");
503 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
504 &unpst_recvspace, 0, "");
506 SYSCTL_DECL(_net_local_dgram);
507 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
508 &unpdg_sendspace, 0, "");
509 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
510 &unpdg_recvspace, 0, "");
512 SYSCTL_DECL(_net_local);
513 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
516 unp_attach(struct socket *so, struct pru_attach_info *ai)
521 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
522 switch (so->so_type) {
525 error = soreserve(so, unpst_sendspace, unpst_recvspace,
530 error = soreserve(so, unpdg_sendspace, unpdg_recvspace,
540 unp = zalloc(unp_zone);
543 bzero(unp, sizeof *unp);
544 unp->unp_gencnt = ++unp_gencnt;
546 LIST_INIT(&unp->unp_refs);
547 unp->unp_socket = so;
548 unp->unp_rvnode = ai->fd_rdir; /* jail cruft XXX JH */
549 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
550 : &unp_shead, unp, unp_link);
551 so->so_pcb = (caddr_t)unp;
556 unp_detach(struct unpcb *unp)
558 LIST_REMOVE(unp, unp_link);
559 unp->unp_gencnt = ++unp_gencnt;
561 if (unp->unp_vnode) {
562 unp->unp_vnode->v_socket = NULL;
563 vrele(unp->unp_vnode);
564 unp->unp_vnode = NULL;
568 while (!LIST_EMPTY(&unp->unp_refs))
569 unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET);
570 soisdisconnected(unp->unp_socket);
571 unp->unp_socket->so_pcb = NULL;
574 * Normally the receive buffer is flushed later,
575 * in sofree, but if our receive buffer holds references
576 * to descriptors that are now garbage, we will dispose
577 * of those descriptor references after the garbage collector
578 * gets them (resulting in a "panic: closef: count < 0").
580 sorflush(unp->unp_socket);
584 kfree(unp->unp_addr, M_SONAME);
585 zfree(unp_zone, unp);
589 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
591 struct proc *p = td->td_proc;
592 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
596 struct nlookupdata nd;
597 char buf[SOCK_MAXADDRLEN];
599 if (unp->unp_vnode != NULL)
601 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
604 strncpy(buf, soun->sun_path, namelen);
605 buf[namelen] = 0; /* null-terminate the string */
606 error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_LOCKVP|NLC_CREATE);
608 error = nlookup(&nd);
609 if (error == 0 && nd.nl_ncp->nc_vp != NULL)
615 vattr.va_type = VSOCK;
616 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask);
617 error = VOP_NCREATE(nd.nl_ncp, &vp, nd.nl_cred, &vattr);
619 vp->v_socket = unp->unp_socket;
621 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam);
630 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
632 struct proc *p = td->td_proc;
633 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
635 struct socket *so2, *so3;
636 struct unpcb *unp, *unp2, *unp3;
638 struct nlookupdata nd;
639 char buf[SOCK_MAXADDRLEN];
643 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
646 strncpy(buf, soun->sun_path, len);
650 error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW);
652 error = nlookup(&nd);
654 error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
659 if (vp->v_type != VSOCK) {
663 error = VOP_ACCESS(vp, VWRITE, p->p_ucred);
668 error = ECONNREFUSED;
671 if (so->so_type != so2->so_type) {
675 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
676 if (!(so2->so_options & SO_ACCEPTCONN) ||
677 (so3 = sonewconn(so2, 0)) == NULL) {
678 error = ECONNREFUSED;
685 unp3->unp_addr = (struct sockaddr_un *)
686 dup_sockaddr((struct sockaddr *)unp2->unp_addr);
689 * unp_peercred management:
691 * The connecter's (client's) credentials are copied
692 * from its process structure at the time of connect()
695 cru2x(p->p_ucred, &unp3->unp_peercred);
696 unp3->unp_flags |= UNP_HAVEPC;
698 * The receiver's (server's) credentials are copied
699 * from the unp_peercred member of socket on which the
700 * former called listen(); unp_listen() cached that
701 * process's credentials at that time so we can use
704 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
705 ("unp_connect: listener without cached peercred"));
706 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
707 sizeof(unp->unp_peercred));
708 unp->unp_flags |= UNP_HAVEPC;
712 error = unp_connect2(so, so2);
719 unp_connect2(struct socket *so, struct socket *so2)
721 struct unpcb *unp = so->so_pcb;
724 if (so2->so_type != so->so_type)
727 unp->unp_conn = unp2;
728 switch (so->so_type) {
731 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
736 unp2->unp_conn = unp;
742 panic("unp_connect2");
748 unp_disconnect(struct unpcb *unp)
750 struct unpcb *unp2 = unp->unp_conn;
755 unp->unp_conn = NULL;
757 switch (unp->unp_socket->so_type) {
759 LIST_REMOVE(unp, unp_reflink);
760 unp->unp_socket->so_state &= ~SS_ISCONNECTED;
763 soisdisconnected(unp->unp_socket);
764 unp2->unp_conn = NULL;
765 soisdisconnected(unp2->unp_socket);
772 unp_abort(struct unpcb *unp)
780 prison_unpcb(struct thread *td, struct unpcb *unp)
786 if ((p = td->td_proc) == NULL)
788 if (!p->p_ucred->cr_prison)
790 if (p->p_fd->fd_rdir == unp->unp_rvnode)
796 unp_pcblist(SYSCTL_HANDLER_ARGS)
799 struct unpcb *unp, **unp_list;
801 struct unp_head *head;
803 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
805 KKASSERT(curproc != NULL);
808 * The process of preparing the PCB list is too time-consuming and
809 * resource-intensive to repeat twice on every request.
811 if (req->oldptr == NULL) {
813 req->oldidx = (n + n/8) * sizeof(struct xunpcb);
817 if (req->newptr != NULL)
821 * OK, now we're committed to doing something.
826 unp_list = kmalloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
827 if (unp_list == NULL)
830 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
831 unp = LIST_NEXT(unp, unp_link)) {
832 if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->td, unp))
835 n = i; /* in case we lost some during malloc */
838 for (i = 0; i < n; i++) {
840 if (unp->unp_gencnt <= gencnt) {
842 xu.xu_len = sizeof xu;
845 * XXX - need more locking here to protect against
846 * connect/disconnect races for SMP.
849 bcopy(unp->unp_addr, &xu.xu_addr,
850 unp->unp_addr->sun_len);
851 if (unp->unp_conn && unp->unp_conn->unp_addr)
852 bcopy(unp->unp_conn->unp_addr,
854 unp->unp_conn->unp_addr->sun_len);
855 bcopy(unp, &xu.xu_unp, sizeof *unp);
856 sotoxsocket(unp->unp_socket, &xu.xu_socket);
857 error = SYSCTL_OUT(req, &xu, sizeof xu);
860 kfree(unp_list, M_TEMP);
864 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
865 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
866 "List of active local datagram sockets");
867 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
868 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
869 "List of active local stream sockets");
872 unp_shutdown(struct unpcb *unp)
876 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn != NULL &&
877 (so = unp->unp_conn->unp_socket))
882 unp_drop(struct unpcb *unp, int errno)
884 struct socket *so = unp->unp_socket;
886 so->so_error = errno;
899 unp_externalize(struct mbuf *rights)
901 struct proc *p = curproc; /* XXX */
903 struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
907 int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm))
908 / sizeof (struct file *);
912 * if the new FD's will not fit, then we free them all
914 if (!fdavail(p, newfds)) {
915 rp = (struct file **)CMSG_DATA(cm);
916 for (i = 0; i < newfds; i++) {
919 * zero the pointer before calling unp_discard,
920 * since it may end up in unp_gc()..
923 unp_discard(fp, NULL);
928 * now change each pointer to an fd in the global table to
929 * an integer that is the index to the local fd table entry
930 * that we set up to point to the global one we are transferring.
931 * If sizeof (struct file *) is bigger than or equal to sizeof int,
932 * then do it in forward order. In that case, an integer will
933 * always come in the same place or before its corresponding
934 * struct file pointer.
935 * If sizeof (struct file *) is smaller than sizeof int, then
936 * do it in reverse order.
938 if (sizeof (struct file *) >= sizeof (int)) {
939 fdp = (int *)(cm + 1);
940 rp = (struct file **)CMSG_DATA(cm);
941 for (i = 0; i < newfds; i++) {
942 if (fdalloc(p, 0, &f))
943 panic("unp_externalize");
947 spin_lock_wr(&unp_spin);
950 spin_unlock_wr(&unp_spin);
954 fdp = (int *)(cm + 1) + newfds - 1;
955 rp = (struct file **)CMSG_DATA(cm) + newfds - 1;
956 for (i = 0; i < newfds; i++) {
957 if (fdalloc(p, 0, &f))
958 panic("unp_externalize");
962 spin_lock_wr(&unp_spin);
965 spin_unlock_wr(&unp_spin);
971 * Adjust length, in case sizeof(struct file *) and sizeof(int)
974 cm->cmsg_len = CMSG_LEN(newfds * sizeof(int));
975 rights->m_len = cm->cmsg_len;
982 unp_zone = zinit("unpcb", sizeof(struct unpcb), nmbclusters, 0, 0);
983 if (unp_zone == NULL)
985 LIST_INIT(&unp_dhead);
986 LIST_INIT(&unp_shead);
987 spin_init(&unp_spin);
991 unp_internalize(struct mbuf *control, struct thread *td)
993 struct proc *p = td->td_proc;
994 struct filedesc *fdescp;
995 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
999 struct cmsgcred *cmcred;
1005 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) ||
1006 cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len)
1010 * Fill in credential information.
1012 if (cm->cmsg_type == SCM_CREDS) {
1013 cmcred = (struct cmsgcred *)(cm + 1);
1014 cmcred->cmcred_pid = p->p_pid;
1015 cmcred->cmcred_uid = p->p_ucred->cr_ruid;
1016 cmcred->cmcred_gid = p->p_ucred->cr_rgid;
1017 cmcred->cmcred_euid = p->p_ucred->cr_uid;
1018 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups,
1020 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1021 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i];
1025 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
1027 * check that all the FDs passed in refer to legal OPEN files
1028 * If not, reject the entire operation.
1030 fdp = (int *)(cm + 1);
1031 for (i = 0; i < oldfds; i++) {
1033 if ((unsigned)fd >= fdescp->fd_nfiles ||
1034 fdescp->fd_files[fd].fp == NULL)
1036 if (fdescp->fd_files[fd].fp->f_type == DTYPE_KQUEUE)
1037 return (EOPNOTSUPP);
1040 * Now replace the integer FDs with pointers to
1041 * the associated global file table entry..
1042 * Allocate a bigger buffer as necessary. But if an cluster is not
1043 * enough, return E2BIG.
1045 newlen = CMSG_LEN(oldfds * sizeof(struct file *));
1046 if (newlen > MCLBYTES)
1048 if (newlen - control->m_len > M_TRAILINGSPACE(control)) {
1049 if (control->m_flags & M_EXT)
1051 MCLGET(control, MB_WAIT);
1052 if (!(control->m_flags & M_EXT))
1055 /* copy the data to the cluster */
1056 memcpy(mtod(control, char *), cm, cm->cmsg_len);
1057 cm = mtod(control, struct cmsghdr *);
1061 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1064 control->m_len = cm->cmsg_len = newlen;
1067 * Transform the file descriptors into struct file pointers.
1068 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1069 * then do it in reverse order so that the int won't get until
1071 * If sizeof (struct file *) is smaller than sizeof int, then
1072 * do it in forward order.
1074 if (sizeof (struct file *) >= sizeof (int)) {
1075 fdp = (int *)(cm + 1) + oldfds - 1;
1076 rp = (struct file **)CMSG_DATA(cm) + oldfds - 1;
1077 for (i = 0; i < oldfds; i++) {
1078 fp = fdescp->fd_files[*fdp--].fp;
1081 spin_lock_wr(&unp_spin);
1084 spin_unlock_wr(&unp_spin);
1087 fdp = (int *)(cm + 1);
1088 rp = (struct file **)CMSG_DATA(cm);
1089 for (i = 0; i < oldfds; i++) {
1090 fp = fdescp->fd_files[*fdp++].fp;
1093 spin_lock_wr(&unp_spin);
1096 spin_unlock_wr(&unp_spin);
1103 * Garbage collect in-transit file descriptors that get lost due to
1104 * loops (i.e. when a socket is sent to another process over itself,
1105 * and more complex situations).
1107 * NOT MPSAFE - TODO socket flush code and maybe closef. Rest is MPSAFE.
1110 struct unp_gc_info {
1111 struct file **extra_ref;
1112 struct file *locked_fp;
1121 struct unp_gc_info info;
1122 static boolean_t unp_gcing;
1126 spin_lock_wr(&unp_spin);
1128 spin_unlock_wr(&unp_spin);
1132 spin_unlock_wr(&unp_spin);
1135 * before going through all this, set all FDs to
1136 * be NOT defered and NOT externally accessible
1139 allfiles_scan_exclusive(unp_gc_clearmarks, NULL);
1141 allfiles_scan_exclusive(unp_gc_checkmarks, &info);
1142 } while (info.defer);
1145 * We grab an extra reference to each of the file table entries
1146 * that are not otherwise accessible and then free the rights
1147 * that are stored in messages on them.
1149 * The bug in the orginal code is a little tricky, so I'll describe
1150 * what's wrong with it here.
1152 * It is incorrect to simply unp_discard each entry for f_msgcount
1153 * times -- consider the case of sockets A and B that contain
1154 * references to each other. On a last close of some other socket,
1155 * we trigger a gc since the number of outstanding rights (unp_rights)
1156 * is non-zero. If during the sweep phase the gc code un_discards,
1157 * we end up doing a (full) closef on the descriptor. A closef on A
1158 * results in the following chain. Closef calls soo_close, which
1159 * calls soclose. Soclose calls first (through the switch
1160 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1161 * returns because the previous instance had set unp_gcing, and
1162 * we return all the way back to soclose, which marks the socket
1163 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush
1164 * to free up the rights that are queued in messages on the socket A,
1165 * i.e., the reference on B. The sorflush calls via the dom_dispose
1166 * switch unp_dispose, which unp_scans with unp_discard. This second
1167 * instance of unp_discard just calls closef on B.
1169 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1170 * which results in another closef on A. Unfortunately, A is already
1171 * being closed, and the descriptor has already been marked with
1172 * SS_NOFDREF, and soclose panics at this point.
1174 * Here, we first take an extra reference to each inaccessible
1175 * descriptor. Then, we call sorflush ourself, since we know
1176 * it is a Unix domain socket anyhow. After we destroy all the
1177 * rights carried in messages, we do a last closef to get rid
1178 * of our extra reference. This is the last close, and the
1179 * unp_detach etc will shut down the socket.
1181 * 91/09/19, bsy@cs.cmu.edu
1183 info.extra_ref = kmalloc(256 * sizeof(struct file *), M_FILE, M_WAITOK);
1184 info.maxindex = 256;
1191 allfiles_scan_exclusive(unp_gc_checkrefs, &info);
1194 * For each FD on our hit list, do the following two things
1196 for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) {
1197 struct file *tfp = *fpp;
1198 if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL)
1199 sorflush((struct socket *)(tfp->f_data));
1201 for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp)
1203 } while (info.index == info.maxindex);
1204 kfree((caddr_t)info.extra_ref, M_FILE);
1209 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1212 unp_gc_checkrefs(struct file *fp, void *data)
1214 struct unp_gc_info *info = data;
1216 if (fp->f_count == 0)
1218 if (info->index == info->maxindex)
1222 * If all refs are from msgs, and it's not marked accessible
1223 * then it must be referenced from some unreachable cycle
1224 * of (shut-down) FDs, so include it in our
1225 * list of FDs to remove
1227 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1228 info->extra_ref[info->index++] = fp;
1235 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1238 unp_gc_clearmarks(struct file *fp, void *data __unused)
1240 fp->f_flag &= ~(FMARK|FDEFER);
1245 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1248 unp_gc_checkmarks(struct file *fp, void *data)
1250 struct unp_gc_info *info = data;
1254 * If the file is not open, skip it
1256 if (fp->f_count == 0)
1259 * If we already marked it as 'defer' in a
1260 * previous pass, then try process it this time
1263 if (fp->f_flag & FDEFER) {
1264 fp->f_flag &= ~FDEFER;
1268 * if it's not defered, then check if it's
1269 * already marked.. if so skip it
1271 if (fp->f_flag & FMARK)
1274 * If all references are from messages
1275 * in transit, then skip it. it's not
1276 * externally accessible.
1278 if (fp->f_count == fp->f_msgcount)
1281 * If it got this far then it must be
1282 * externally accessible.
1284 fp->f_flag |= FMARK;
1287 * either it was defered, or it is externally
1288 * accessible and not already marked so.
1289 * Now check if it is possibly one of OUR sockets.
1291 if (fp->f_type != DTYPE_SOCKET ||
1292 (so = (struct socket *)fp->f_data) == NULL)
1294 if (so->so_proto->pr_domain != &localdomain ||
1295 !(so->so_proto->pr_flags & PR_RIGHTS))
1298 XXX note: exclusive fp->f_spin lock held
1299 if (so->so_rcv.sb_flags & SB_LOCK) {
1301 * This is problematical; it's not clear
1302 * we need to wait for the sockbuf to be
1303 * unlocked (on a uniprocessor, at least),
1304 * and it's also not clear what to do
1305 * if sbwait returns an error due to receipt
1306 * of a signal. If sbwait does return
1307 * an error, we'll go into an infinite
1308 * loop. Delete all of this for now.
1310 sbwait(&so->so_rcv);
1315 * So, Ok, it's one of our sockets and it IS externally
1316 * accessible (or was defered). Now we look
1317 * to see if we hold any file descriptors in its
1318 * message buffers. Follow those links and mark them
1319 * as accessible too.
1321 info->locked_fp = fp;
1322 /* spin_lock_wr(&so->so_rcv.sb_spin); */
1323 unp_scan(so->so_rcv.sb_mb, unp_mark, info);
1324 /* spin_unlock_wr(&so->so_rcv.sb_spin);*/
1329 unp_dispose(struct mbuf *m)
1332 unp_scan(m, unp_discard, NULL);
1336 unp_listen(struct unpcb *unp, struct thread *td)
1338 struct proc *p = td->td_proc;
1341 cru2x(p->p_ucred, &unp->unp_peercred);
1342 unp->unp_flags |= UNP_HAVEPCCACHED;
1347 unp_scan(struct mbuf *m0, void (*op)(struct file *, void *), void *data)
1356 for (m = m0; m; m = m->m_next) {
1357 if (m->m_type == MT_CONTROL &&
1358 m->m_len >= sizeof(*cm)) {
1359 cm = mtod(m, struct cmsghdr *);
1360 if (cm->cmsg_level != SOL_SOCKET ||
1361 cm->cmsg_type != SCM_RIGHTS)
1363 qfds = (cm->cmsg_len -
1364 (CMSG_DATA(cm) - (u_char *)cm))
1365 / sizeof (struct file *);
1366 rp = (struct file **)CMSG_DATA(cm);
1367 for (i = 0; i < qfds; i++)
1369 break; /* XXX, but saves time */
1377 unp_mark(struct file *fp, void *data)
1379 struct unp_gc_info *info = data;
1381 if (info->locked_fp != fp)
1382 spin_lock_wr(&fp->f_spin);
1383 if ((fp->f_flag & FMARK) == 0) {
1385 fp->f_flag |= (FMARK|FDEFER);
1387 if (info->locked_fp != fp)
1388 spin_unlock_wr(&fp->f_spin);
1392 unp_discard(struct file *fp, void *data __unused)
1394 spin_lock_wr(&unp_spin);
1397 spin_unlock_wr(&unp_spin);