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 1.42 2008/06/19 00:03:45 aggelos 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
;
239 switch (so
->so_type
) {
241 panic("uipc_rcvd DGRAM?");
246 if (unp
->unp_conn
== NULL
)
249 * Because we are transfering mbufs directly to the
250 * peer socket we have to use SSB_STOP on the sender
251 * to prevent it from building up infinite mbufs.
253 so2
= unp
->unp_conn
->unp_socket
;
254 if (so
->so_rcv
.ssb_cc
< so2
->so_snd
.ssb_hiwat
&&
255 so
->so_rcv
.ssb_mbcnt
< so2
->so_snd
.ssb_mbmax
257 so2
->so_snd
.ssb_flags
&= ~SSB_STOP
;
263 panic("uipc_rcvd unknown socktype");
268 /* pru_rcvoob is EOPNOTSUPP */
271 uipc_send(struct socket
*so
, int flags
, struct mbuf
*m
, struct sockaddr
*nam
,
272 struct mbuf
*control
, struct thread
*td
)
275 struct unpcb
*unp
= so
->so_pcb
;
282 if (flags
& PRUS_OOB
) {
287 if (control
&& (error
= unp_internalize(control
, td
)))
290 switch (so
->so_type
) {
293 struct sockaddr
*from
;
300 error
= unp_connect(so
, nam
, td
);
304 if (unp
->unp_conn
== NULL
) {
309 so2
= unp
->unp_conn
->unp_socket
;
311 from
= (struct sockaddr
*)unp
->unp_addr
;
314 if (ssb_appendaddr(&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 (ssb_appendcontrol(&so2
->so_rcv
, m
, control
)) {
361 } else if (so
->so_type
== SOCK_SEQPACKET
) {
362 sbappendrecord(&so2
->so_rcv
.sb
, m
);
365 sbappend(&so2
->so_rcv
.sb
, m
);
370 * Because we are transfering mbufs directly to the
371 * peer socket we have to use SSB_STOP on the sender
372 * to prevent it from building up infinite mbufs.
374 if (so2
->so_rcv
.ssb_cc
>= so
->so_snd
.ssb_hiwat
||
375 so2
->so_rcv
.ssb_mbcnt
>= so
->so_snd
.ssb_mbmax
377 so
->so_snd
.ssb_flags
|= SSB_STOP
;
383 panic("uipc_send unknown socktype");
387 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN.
389 if (flags
& PRUS_EOF
) {
394 if (control
&& error
!= 0)
395 unp_dispose(control
);
406 uipc_sense(struct socket
*so
, struct stat
*sb
)
408 struct unpcb
*unp
= so
->so_pcb
;
412 sb
->st_blksize
= so
->so_snd
.ssb_hiwat
;
414 if (unp
->unp_ino
== 0) /* make up a non-zero inode number */
415 unp
->unp_ino
= (++unp_ino
== 0) ? ++unp_ino
: unp_ino
;
416 sb
->st_ino
= unp
->unp_ino
;
421 uipc_shutdown(struct socket
*so
)
423 struct unpcb
*unp
= so
->so_pcb
;
433 uipc_sockaddr(struct socket
*so
, struct sockaddr
**nam
)
435 struct unpcb
*unp
= so
->so_pcb
;
440 *nam
= dup_sockaddr((struct sockaddr
*)unp
->unp_addr
);
444 struct pr_usrreqs uipc_usrreqs
= {
445 .pru_abort
= uipc_abort
,
446 .pru_accept
= uipc_accept
,
447 .pru_attach
= uipc_attach
,
448 .pru_bind
= uipc_bind
,
449 .pru_connect
= uipc_connect
,
450 .pru_connect2
= uipc_connect2
,
451 .pru_control
= pru_control_notsupp
,
452 .pru_detach
= uipc_detach
,
453 .pru_disconnect
= uipc_disconnect
,
454 .pru_listen
= uipc_listen
,
455 .pru_peeraddr
= uipc_peeraddr
,
456 .pru_rcvd
= uipc_rcvd
,
457 .pru_rcvoob
= pru_rcvoob_notsupp
,
458 .pru_send
= uipc_send
,
459 .pru_sense
= uipc_sense
,
460 .pru_shutdown
= uipc_shutdown
,
461 .pru_sockaddr
= uipc_sockaddr
,
462 .pru_sosend
= sosend
,
463 .pru_soreceive
= soreceive
,
468 uipc_ctloutput(struct socket
*so
, struct sockopt
*sopt
)
470 struct unpcb
*unp
= so
->so_pcb
;
473 switch (sopt
->sopt_dir
) {
475 switch (sopt
->sopt_name
) {
477 if (unp
->unp_flags
& UNP_HAVEPC
)
478 soopt_from_kbuf(sopt
, &unp
->unp_peercred
,
479 sizeof(unp
->unp_peercred
));
481 if (so
->so_type
== SOCK_STREAM
)
483 else if (so
->so_type
== SOCK_SEQPACKET
)
503 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
504 * for stream sockets, although the total for sender and receiver is
505 * actually only PIPSIZ.
506 * Datagram sockets really use the sendspace as the maximum datagram size,
507 * and don't really want to reserve the sendspace. Their recvspace should
508 * be large enough for at least one max-size datagram plus address.
513 static u_long unpst_sendspace
= PIPSIZ
;
514 static u_long unpst_recvspace
= PIPSIZ
;
515 static u_long unpdg_sendspace
= 2*1024; /* really max datagram size */
516 static u_long unpdg_recvspace
= 4*1024;
518 static int unp_rights
; /* file descriptors in flight */
519 static struct spinlock unp_spin
= SPINLOCK_INITIALIZER(&unp_spin
);
521 SYSCTL_DECL(_net_local_seqpacket
);
522 SYSCTL_DECL(_net_local_stream
);
523 SYSCTL_INT(_net_local_stream
, OID_AUTO
, sendspace
, CTLFLAG_RW
,
524 &unpst_sendspace
, 0, "");
525 SYSCTL_INT(_net_local_stream
, OID_AUTO
, recvspace
, CTLFLAG_RW
,
526 &unpst_recvspace
, 0, "");
528 SYSCTL_DECL(_net_local_dgram
);
529 SYSCTL_INT(_net_local_dgram
, OID_AUTO
, maxdgram
, CTLFLAG_RW
,
530 &unpdg_sendspace
, 0, "");
531 SYSCTL_INT(_net_local_dgram
, OID_AUTO
, recvspace
, CTLFLAG_RW
,
532 &unpdg_recvspace
, 0, "");
534 SYSCTL_DECL(_net_local
);
535 SYSCTL_INT(_net_local
, OID_AUTO
, inflight
, CTLFLAG_RD
, &unp_rights
, 0, "");
538 unp_attach(struct socket
*so
, struct pru_attach_info
*ai
)
543 if (so
->so_snd
.ssb_hiwat
== 0 || so
->so_rcv
.ssb_hiwat
== 0) {
544 switch (so
->so_type
) {
548 error
= soreserve(so
, unpst_sendspace
, unpst_recvspace
,
553 error
= soreserve(so
, unpdg_sendspace
, unpdg_recvspace
,
563 unp
= zalloc(unp_zone
);
566 bzero(unp
, sizeof *unp
);
567 unp
->unp_gencnt
= ++unp_gencnt
;
569 LIST_INIT(&unp
->unp_refs
);
570 unp
->unp_socket
= so
;
571 unp
->unp_rvnode
= ai
->fd_rdir
; /* jail cruft XXX JH */
572 LIST_INSERT_HEAD(so
->so_type
== SOCK_DGRAM
? &unp_dhead
573 : &unp_shead
, unp
, unp_link
);
574 so
->so_pcb
= (caddr_t
)unp
;
579 unp_detach(struct unpcb
*unp
)
581 LIST_REMOVE(unp
, unp_link
);
582 unp
->unp_gencnt
= ++unp_gencnt
;
584 if (unp
->unp_vnode
) {
585 unp
->unp_vnode
->v_socket
= NULL
;
586 vrele(unp
->unp_vnode
);
587 unp
->unp_vnode
= NULL
;
591 while (!LIST_EMPTY(&unp
->unp_refs
))
592 unp_drop(LIST_FIRST(&unp
->unp_refs
), ECONNRESET
);
593 soisdisconnected(unp
->unp_socket
);
594 unp
->unp_socket
->so_pcb
= NULL
;
597 * Normally the receive buffer is flushed later,
598 * in sofree, but if our receive buffer holds references
599 * to descriptors that are now garbage, we will dispose
600 * of those descriptor references after the garbage collector
601 * gets them (resulting in a "panic: closef: count < 0").
603 sorflush(unp
->unp_socket
);
607 kfree(unp
->unp_addr
, M_SONAME
);
608 zfree(unp_zone
, unp
);
612 unp_bind(struct unpcb
*unp
, struct sockaddr
*nam
, struct thread
*td
)
614 struct proc
*p
= td
->td_proc
;
615 struct sockaddr_un
*soun
= (struct sockaddr_un
*)nam
;
619 struct nlookupdata nd
;
620 char buf
[SOCK_MAXADDRLEN
];
622 if (unp
->unp_vnode
!= NULL
)
624 namelen
= soun
->sun_len
- offsetof(struct sockaddr_un
, sun_path
);
627 strncpy(buf
, soun
->sun_path
, namelen
);
628 buf
[namelen
] = 0; /* null-terminate the string */
629 error
= nlookup_init(&nd
, buf
, UIO_SYSSPACE
,
630 NLC_LOCKVP
| NLC_CREATE
| NLC_REFDVP
);
632 error
= nlookup(&nd
);
633 if (error
== 0 && nd
.nl_nch
.ncp
->nc_vp
!= NULL
)
639 vattr
.va_type
= VSOCK
;
640 vattr
.va_mode
= (ACCESSPERMS
& ~p
->p_fd
->fd_cmask
);
641 error
= VOP_NCREATE(&nd
.nl_nch
, nd
.nl_dvp
, &vp
, nd
.nl_cred
, &vattr
);
643 vp
->v_socket
= unp
->unp_socket
;
645 unp
->unp_addr
= (struct sockaddr_un
*)dup_sockaddr(nam
);
654 unp_connect(struct socket
*so
, struct sockaddr
*nam
, struct thread
*td
)
656 struct proc
*p
= td
->td_proc
;
657 struct sockaddr_un
*soun
= (struct sockaddr_un
*)nam
;
659 struct socket
*so2
, *so3
;
660 struct unpcb
*unp
, *unp2
, *unp3
;
662 struct nlookupdata nd
;
663 char buf
[SOCK_MAXADDRLEN
];
667 len
= nam
->sa_len
- offsetof(struct sockaddr_un
, sun_path
);
670 strncpy(buf
, soun
->sun_path
, len
);
674 error
= nlookup_init(&nd
, buf
, UIO_SYSSPACE
, NLC_FOLLOW
);
676 error
= nlookup(&nd
);
678 error
= cache_vget(&nd
.nl_nch
, nd
.nl_cred
, LK_EXCLUSIVE
, &vp
);
683 if (vp
->v_type
!= VSOCK
) {
687 error
= VOP_ACCESS(vp
, VWRITE
, p
->p_ucred
);
692 error
= ECONNREFUSED
;
695 if (so
->so_type
!= so2
->so_type
) {
699 if (so
->so_proto
->pr_flags
& PR_CONNREQUIRED
) {
700 if (!(so2
->so_options
& SO_ACCEPTCONN
) ||
701 (so3
= sonewconn(so2
, 0)) == NULL
) {
702 error
= ECONNREFUSED
;
709 unp3
->unp_addr
= (struct sockaddr_un
*)
710 dup_sockaddr((struct sockaddr
*)unp2
->unp_addr
);
713 * unp_peercred management:
715 * The connecter's (client's) credentials are copied
716 * from its process structure at the time of connect()
719 cru2x(p
->p_ucred
, &unp3
->unp_peercred
);
720 unp3
->unp_flags
|= UNP_HAVEPC
;
722 * The receiver's (server's) credentials are copied
723 * from the unp_peercred member of socket on which the
724 * former called listen(); unp_listen() cached that
725 * process's credentials at that time so we can use
728 KASSERT(unp2
->unp_flags
& UNP_HAVEPCCACHED
,
729 ("unp_connect: listener without cached peercred"));
730 memcpy(&unp
->unp_peercred
, &unp2
->unp_peercred
,
731 sizeof(unp
->unp_peercred
));
732 unp
->unp_flags
|= UNP_HAVEPC
;
736 error
= unp_connect2(so
, so2
);
743 unp_connect2(struct socket
*so
, struct socket
*so2
)
745 struct unpcb
*unp
= so
->so_pcb
;
748 if (so2
->so_type
!= so
->so_type
)
751 unp
->unp_conn
= unp2
;
752 switch (so
->so_type
) {
755 LIST_INSERT_HEAD(&unp2
->unp_refs
, unp
, unp_reflink
);
761 unp2
->unp_conn
= unp
;
767 panic("unp_connect2");
773 unp_disconnect(struct unpcb
*unp
)
775 struct unpcb
*unp2
= unp
->unp_conn
;
780 unp
->unp_conn
= NULL
;
782 switch (unp
->unp_socket
->so_type
) {
784 LIST_REMOVE(unp
, unp_reflink
);
785 unp
->unp_socket
->so_state
&= ~SS_ISCONNECTED
;
789 soisdisconnected(unp
->unp_socket
);
790 unp2
->unp_conn
= NULL
;
791 soisdisconnected(unp2
->unp_socket
);
798 unp_abort(struct unpcb
*unp
)
806 prison_unpcb(struct thread
*td
, struct unpcb
*unp
)
812 if ((p
= td
->td_proc
) == NULL
)
814 if (!p
->p_ucred
->cr_prison
)
816 if (p
->p_fd
->fd_rdir
== unp
->unp_rvnode
)
822 unp_pcblist(SYSCTL_HANDLER_ARGS
)
825 struct unpcb
*unp
, **unp_list
;
827 struct unp_head
*head
;
829 head
= ((intptr_t)arg1
== SOCK_DGRAM
? &unp_dhead
: &unp_shead
);
831 KKASSERT(curproc
!= NULL
);
834 * The process of preparing the PCB list is too time-consuming and
835 * resource-intensive to repeat twice on every request.
837 if (req
->oldptr
== NULL
) {
839 req
->oldidx
= (n
+ n
/8) * sizeof(struct xunpcb
);
843 if (req
->newptr
!= NULL
)
847 * OK, now we're committed to doing something.
852 unp_list
= kmalloc(n
* sizeof *unp_list
, M_TEMP
, M_WAITOK
);
854 for (unp
= LIST_FIRST(head
), i
= 0; unp
&& i
< n
;
855 unp
= LIST_NEXT(unp
, unp_link
)) {
856 if (unp
->unp_gencnt
<= gencnt
&& !prison_unpcb(req
->td
, unp
))
859 n
= i
; /* in case we lost some during malloc */
862 for (i
= 0; i
< n
; i
++) {
864 if (unp
->unp_gencnt
<= gencnt
) {
866 xu
.xu_len
= sizeof xu
;
869 * XXX - need more locking here to protect against
870 * connect/disconnect races for SMP.
873 bcopy(unp
->unp_addr
, &xu
.xu_addr
,
874 unp
->unp_addr
->sun_len
);
875 if (unp
->unp_conn
&& unp
->unp_conn
->unp_addr
)
876 bcopy(unp
->unp_conn
->unp_addr
,
878 unp
->unp_conn
->unp_addr
->sun_len
);
879 bcopy(unp
, &xu
.xu_unp
, sizeof *unp
);
880 sotoxsocket(unp
->unp_socket
, &xu
.xu_socket
);
881 error
= SYSCTL_OUT(req
, &xu
, sizeof xu
);
884 kfree(unp_list
, M_TEMP
);
888 SYSCTL_PROC(_net_local_dgram
, OID_AUTO
, pcblist
, CTLFLAG_RD
,
889 (caddr_t
)(long)SOCK_DGRAM
, 0, unp_pcblist
, "S,xunpcb",
890 "List of active local datagram sockets");
891 SYSCTL_PROC(_net_local_stream
, OID_AUTO
, pcblist
, CTLFLAG_RD
,
892 (caddr_t
)(long)SOCK_STREAM
, 0, unp_pcblist
, "S,xunpcb",
893 "List of active local stream sockets");
894 SYSCTL_PROC(_net_local_seqpacket
, OID_AUTO
, pcblist
, CTLFLAG_RD
,
895 (caddr_t
)(long)SOCK_SEQPACKET
, 0, unp_pcblist
, "S,xunpcb",
896 "List of active local seqpacket stream sockets");
899 unp_shutdown(struct unpcb
*unp
)
903 if ((unp
->unp_socket
->so_type
== SOCK_STREAM
||
904 unp
->unp_socket
->so_type
== SOCK_SEQPACKET
) &&
905 unp
->unp_conn
!= NULL
&& (so
= unp
->unp_conn
->unp_socket
)) {
911 unp_drop(struct unpcb
*unp
, int err
)
913 struct socket
*so
= unp
->unp_socket
;
928 unp_externalize(struct mbuf
*rights
)
930 struct proc
*p
= curproc
; /* XXX */
932 struct cmsghdr
*cm
= mtod(rights
, struct cmsghdr
*);
936 int newfds
= (cm
->cmsg_len
- (CMSG_DATA(cm
) - (u_char
*)cm
))
937 / sizeof (struct file
*);
941 * if the new FD's will not fit, then we free them all
943 if (!fdavail(p
, newfds
)) {
944 rp
= (struct file
**)CMSG_DATA(cm
);
945 for (i
= 0; i
< newfds
; i
++) {
948 * zero the pointer before calling unp_discard,
949 * since it may end up in unp_gc()..
952 unp_discard(fp
, NULL
);
957 * now change each pointer to an fd in the global table to
958 * an integer that is the index to the local fd table entry
959 * that we set up to point to the global one we are transferring.
960 * If sizeof (struct file *) is bigger than or equal to sizeof int,
961 * then do it in forward order. In that case, an integer will
962 * always come in the same place or before its corresponding
963 * struct file pointer.
964 * If sizeof (struct file *) is smaller than sizeof int, then
965 * do it in reverse order.
967 if (sizeof (struct file
*) >= sizeof (int)) {
968 fdp
= (int *)(cm
+ 1);
969 rp
= (struct file
**)CMSG_DATA(cm
);
970 for (i
= 0; i
< newfds
; i
++) {
971 if (fdalloc(p
, 0, &f
))
972 panic("unp_externalize");
976 spin_lock_wr(&unp_spin
);
979 spin_unlock_wr(&unp_spin
);
983 fdp
= (int *)(cm
+ 1) + newfds
- 1;
984 rp
= (struct file
**)CMSG_DATA(cm
) + newfds
- 1;
985 for (i
= 0; i
< newfds
; i
++) {
986 if (fdalloc(p
, 0, &f
))
987 panic("unp_externalize");
991 spin_lock_wr(&unp_spin
);
994 spin_unlock_wr(&unp_spin
);
1000 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1003 cm
->cmsg_len
= CMSG_LEN(newfds
* sizeof(int));
1004 rights
->m_len
= cm
->cmsg_len
;
1011 unp_zone
= zinit("unpcb", sizeof(struct unpcb
), nmbclusters
, 0, 0);
1012 if (unp_zone
== NULL
)
1014 LIST_INIT(&unp_dhead
);
1015 LIST_INIT(&unp_shead
);
1016 spin_init(&unp_spin
);
1020 unp_internalize(struct mbuf
*control
, struct thread
*td
)
1022 struct proc
*p
= td
->td_proc
;
1023 struct filedesc
*fdescp
;
1024 struct cmsghdr
*cm
= mtod(control
, struct cmsghdr
*);
1028 struct cmsgcred
*cmcred
;
1034 if ((cm
->cmsg_type
!= SCM_RIGHTS
&& cm
->cmsg_type
!= SCM_CREDS
) ||
1035 cm
->cmsg_level
!= SOL_SOCKET
|| cm
->cmsg_len
!= control
->m_len
)
1039 * Fill in credential information.
1041 if (cm
->cmsg_type
== SCM_CREDS
) {
1042 cmcred
= (struct cmsgcred
*)(cm
+ 1);
1043 cmcred
->cmcred_pid
= p
->p_pid
;
1044 cmcred
->cmcred_uid
= p
->p_ucred
->cr_ruid
;
1045 cmcred
->cmcred_gid
= p
->p_ucred
->cr_rgid
;
1046 cmcred
->cmcred_euid
= p
->p_ucred
->cr_uid
;
1047 cmcred
->cmcred_ngroups
= MIN(p
->p_ucred
->cr_ngroups
,
1049 for (i
= 0; i
< cmcred
->cmcred_ngroups
; i
++)
1050 cmcred
->cmcred_groups
[i
] = p
->p_ucred
->cr_groups
[i
];
1054 oldfds
= (cm
->cmsg_len
- sizeof (*cm
)) / sizeof (int);
1056 * check that all the FDs passed in refer to legal OPEN files
1057 * If not, reject the entire operation.
1059 fdp
= (int *)(cm
+ 1);
1060 for (i
= 0; i
< oldfds
; i
++) {
1062 if ((unsigned)fd
>= fdescp
->fd_nfiles
||
1063 fdescp
->fd_files
[fd
].fp
== NULL
)
1065 if (fdescp
->fd_files
[fd
].fp
->f_type
== DTYPE_KQUEUE
)
1066 return (EOPNOTSUPP
);
1069 * Now replace the integer FDs with pointers to
1070 * the associated global file table entry..
1071 * Allocate a bigger buffer as necessary. But if an cluster is not
1072 * enough, return E2BIG.
1074 newlen
= CMSG_LEN(oldfds
* sizeof(struct file
*));
1075 if (newlen
> MCLBYTES
)
1077 if (newlen
- control
->m_len
> M_TRAILINGSPACE(control
)) {
1078 if (control
->m_flags
& M_EXT
)
1080 MCLGET(control
, MB_WAIT
);
1081 if (!(control
->m_flags
& M_EXT
))
1084 /* copy the data to the cluster */
1085 memcpy(mtod(control
, char *), cm
, cm
->cmsg_len
);
1086 cm
= mtod(control
, struct cmsghdr
*);
1090 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1093 control
->m_len
= cm
->cmsg_len
= newlen
;
1096 * Transform the file descriptors into struct file pointers.
1097 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1098 * then do it in reverse order so that the int won't get until
1100 * If sizeof (struct file *) is smaller than sizeof int, then
1101 * do it in forward order.
1103 if (sizeof (struct file
*) >= sizeof (int)) {
1104 fdp
= (int *)(cm
+ 1) + oldfds
- 1;
1105 rp
= (struct file
**)CMSG_DATA(cm
) + oldfds
- 1;
1106 for (i
= 0; i
< oldfds
; i
++) {
1107 fp
= fdescp
->fd_files
[*fdp
--].fp
;
1110 spin_lock_wr(&unp_spin
);
1113 spin_unlock_wr(&unp_spin
);
1116 fdp
= (int *)(cm
+ 1);
1117 rp
= (struct file
**)CMSG_DATA(cm
);
1118 for (i
= 0; i
< oldfds
; i
++) {
1119 fp
= fdescp
->fd_files
[*fdp
++].fp
;
1122 spin_lock_wr(&unp_spin
);
1125 spin_unlock_wr(&unp_spin
);
1132 * Garbage collect in-transit file descriptors that get lost due to
1133 * loops (i.e. when a socket is sent to another process over itself,
1134 * and more complex situations).
1136 * NOT MPSAFE - TODO socket flush code and maybe closef. Rest is MPSAFE.
1139 struct unp_gc_info
{
1140 struct file
**extra_ref
;
1141 struct file
*locked_fp
;
1150 struct unp_gc_info info
;
1151 static boolean_t unp_gcing
;
1155 spin_lock_wr(&unp_spin
);
1157 spin_unlock_wr(&unp_spin
);
1161 spin_unlock_wr(&unp_spin
);
1164 * before going through all this, set all FDs to
1165 * be NOT defered and NOT externally accessible
1168 allfiles_scan_exclusive(unp_gc_clearmarks
, NULL
);
1170 allfiles_scan_exclusive(unp_gc_checkmarks
, &info
);
1171 } while (info
.defer
);
1174 * We grab an extra reference to each of the file table entries
1175 * that are not otherwise accessible and then free the rights
1176 * that are stored in messages on them.
1178 * The bug in the orginal code is a little tricky, so I'll describe
1179 * what's wrong with it here.
1181 * It is incorrect to simply unp_discard each entry for f_msgcount
1182 * times -- consider the case of sockets A and B that contain
1183 * references to each other. On a last close of some other socket,
1184 * we trigger a gc since the number of outstanding rights (unp_rights)
1185 * is non-zero. If during the sweep phase the gc code un_discards,
1186 * we end up doing a (full) closef on the descriptor. A closef on A
1187 * results in the following chain. Closef calls soo_close, which
1188 * calls soclose. Soclose calls first (through the switch
1189 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1190 * returns because the previous instance had set unp_gcing, and
1191 * we return all the way back to soclose, which marks the socket
1192 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush
1193 * to free up the rights that are queued in messages on the socket A,
1194 * i.e., the reference on B. The sorflush calls via the dom_dispose
1195 * switch unp_dispose, which unp_scans with unp_discard. This second
1196 * instance of unp_discard just calls closef on B.
1198 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1199 * which results in another closef on A. Unfortunately, A is already
1200 * being closed, and the descriptor has already been marked with
1201 * SS_NOFDREF, and soclose panics at this point.
1203 * Here, we first take an extra reference to each inaccessible
1204 * descriptor. Then, we call sorflush ourself, since we know
1205 * it is a Unix domain socket anyhow. After we destroy all the
1206 * rights carried in messages, we do a last closef to get rid
1207 * of our extra reference. This is the last close, and the
1208 * unp_detach etc will shut down the socket.
1210 * 91/09/19, bsy@cs.cmu.edu
1212 info
.extra_ref
= kmalloc(256 * sizeof(struct file
*), M_FILE
, M_WAITOK
);
1213 info
.maxindex
= 256;
1220 allfiles_scan_exclusive(unp_gc_checkrefs
, &info
);
1223 * For each FD on our hit list, do the following two things
1225 for (i
= info
.index
, fpp
= info
.extra_ref
; --i
>= 0; ++fpp
) {
1226 struct file
*tfp
= *fpp
;
1227 if (tfp
->f_type
== DTYPE_SOCKET
&& tfp
->f_data
!= NULL
)
1228 sorflush((struct socket
*)(tfp
->f_data
));
1230 for (i
= info
.index
, fpp
= info
.extra_ref
; --i
>= 0; ++fpp
)
1232 } while (info
.index
== info
.maxindex
);
1233 kfree((caddr_t
)info
.extra_ref
, M_FILE
);
1238 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1241 unp_gc_checkrefs(struct file
*fp
, void *data
)
1243 struct unp_gc_info
*info
= data
;
1245 if (fp
->f_count
== 0)
1247 if (info
->index
== info
->maxindex
)
1251 * If all refs are from msgs, and it's not marked accessible
1252 * then it must be referenced from some unreachable cycle
1253 * of (shut-down) FDs, so include it in our
1254 * list of FDs to remove
1256 if (fp
->f_count
== fp
->f_msgcount
&& !(fp
->f_flag
& FMARK
)) {
1257 info
->extra_ref
[info
->index
++] = fp
;
1264 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1267 unp_gc_clearmarks(struct file
*fp
, void *data __unused
)
1269 fp
->f_flag
&= ~(FMARK
|FDEFER
);
1274 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1277 unp_gc_checkmarks(struct file
*fp
, void *data
)
1279 struct unp_gc_info
*info
= data
;
1283 * If the file is not open, skip it
1285 if (fp
->f_count
== 0)
1288 * If we already marked it as 'defer' in a
1289 * previous pass, then try process it this time
1292 if (fp
->f_flag
& FDEFER
) {
1293 fp
->f_flag
&= ~FDEFER
;
1297 * if it's not defered, then check if it's
1298 * already marked.. if so skip it
1300 if (fp
->f_flag
& FMARK
)
1303 * If all references are from messages
1304 * in transit, then skip it. it's not
1305 * externally accessible.
1307 if (fp
->f_count
== fp
->f_msgcount
)
1310 * If it got this far then it must be
1311 * externally accessible.
1313 fp
->f_flag
|= FMARK
;
1316 * either it was defered, or it is externally
1317 * accessible and not already marked so.
1318 * Now check if it is possibly one of OUR sockets.
1320 if (fp
->f_type
!= DTYPE_SOCKET
||
1321 (so
= (struct socket
*)fp
->f_data
) == NULL
)
1323 if (so
->so_proto
->pr_domain
!= &localdomain
||
1324 !(so
->so_proto
->pr_flags
& PR_RIGHTS
))
1327 XXX note
: exclusive fp
->f_spin lock held
1328 if (so
->so_rcv
.sb_flags
& SB_LOCK
) {
1330 * This is problematical; it's not clear
1331 * we need to wait for the sockbuf to be
1332 * unlocked (on a uniprocessor, at least),
1333 * and it's also not clear what to do
1334 * if sbwait returns an error due to receipt
1335 * of a signal. If sbwait does return
1336 * an error, we'll go into an infinite
1337 * loop. Delete all of this for now.
1339 sbwait(&so
->so_rcv
);
1344 * So, Ok, it's one of our sockets and it IS externally
1345 * accessible (or was defered). Now we look
1346 * to see if we hold any file descriptors in its
1347 * message buffers. Follow those links and mark them
1348 * as accessible too.
1350 info
->locked_fp
= fp
;
1351 /* spin_lock_wr(&so->so_rcv.sb_spin); */
1352 unp_scan(so
->so_rcv
.ssb_mb
, unp_mark
, info
);
1353 /* spin_unlock_wr(&so->so_rcv.sb_spin);*/
1358 unp_dispose(struct mbuf
*m
)
1361 unp_scan(m
, unp_discard
, NULL
);
1365 unp_listen(struct unpcb
*unp
, struct thread
*td
)
1367 struct proc
*p
= td
->td_proc
;
1370 cru2x(p
->p_ucred
, &unp
->unp_peercred
);
1371 unp
->unp_flags
|= UNP_HAVEPCCACHED
;
1376 unp_scan(struct mbuf
*m0
, void (*op
)(struct file
*, void *), void *data
)
1385 for (m
= m0
; m
; m
= m
->m_next
) {
1386 if (m
->m_type
== MT_CONTROL
&&
1387 m
->m_len
>= sizeof(*cm
)) {
1388 cm
= mtod(m
, struct cmsghdr
*);
1389 if (cm
->cmsg_level
!= SOL_SOCKET
||
1390 cm
->cmsg_type
!= SCM_RIGHTS
)
1392 qfds
= (cm
->cmsg_len
-
1393 (CMSG_DATA(cm
) - (u_char
*)cm
))
1394 / sizeof (struct file
*);
1395 rp
= (struct file
**)CMSG_DATA(cm
);
1396 for (i
= 0; i
< qfds
; i
++)
1398 break; /* XXX, but saves time */
1406 unp_mark(struct file
*fp
, void *data
)
1408 struct unp_gc_info
*info
= data
;
1410 if (info
->locked_fp
!= fp
)
1411 spin_lock_wr(&fp
->f_spin
);
1412 if ((fp
->f_flag
& FMARK
) == 0) {
1414 fp
->f_flag
|= (FMARK
|FDEFER
);
1416 if (info
->locked_fp
!= fp
)
1417 spin_unlock_wr(&fp
->f_spin
);
1421 unp_discard(struct file
*fp
, void *data __unused
)
1423 spin_lock_wr(&unp_spin
);
1426 spin_unlock_wr(&unp_spin
);