2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2004-2009 Robert N. M. Watson
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 4. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
35 * UNIX Domain (Local) Sockets
37 * This is an implementation of UNIX (local) domain sockets. Each socket has
38 * an associated struct unpcb (UNIX protocol control block). Stream sockets
39 * may be connected to 0 or 1 other socket. Datagram sockets may be
40 * connected to 0, 1, or many other sockets. Sockets may be created and
41 * connected in pairs (socketpair(2)), or bound/connected to using the file
42 * system name space. For most purposes, only the receive socket buffer is
43 * used, as sending on one socket delivers directly to the receive socket
44 * buffer of a second socket.
46 * The implementation is substantially complicated by the fact that
47 * "ancillary data", such as file descriptors or credentials, may be passed
48 * across UNIX domain sockets. The potential for passing UNIX domain sockets
49 * over other UNIX domain sockets requires the implementation of a simple
50 * garbage collector to find and tear down cycles of disconnected sockets.
54 * distinguish datagram size limits from flow control limits in SEQPACKET
55 * rethink name space problems
56 * need a proper out-of-band
59 #include <sys/cdefs.h>
60 __FBSDID("$FreeBSD$");
64 #include <sys/param.h>
65 #include <sys/capability.h>
66 #include <sys/domain.h>
67 #include <sys/fcntl.h>
68 #include <sys/malloc.h> /* XXX must be before <sys/file.h> */
69 #include <sys/eventhandler.h>
71 #include <sys/filedesc.h>
72 #include <sys/kernel.h>
75 #include <sys/mount.h>
76 #include <sys/mutex.h>
77 #include <sys/namei.h>
79 #include <sys/protosw.h>
80 #include <sys/queue.h>
81 #include <sys/resourcevar.h>
82 #include <sys/rwlock.h>
83 #include <sys/socket.h>
84 #include <sys/socketvar.h>
85 #include <sys/signalvar.h>
88 #include <sys/sysctl.h>
89 #include <sys/systm.h>
90 #include <sys/taskqueue.h>
92 #include <sys/unpcb.h>
93 #include <sys/vnode.h>
101 #include <security/mac/mac_framework.h>
105 MALLOC_DECLARE(M_FILECAPS);
109 * (l) Locked using list lock
110 * (g) Locked using linkage lock
113 static uma_zone_t unp_zone;
114 static unp_gen_t unp_gencnt; /* (l) */
115 static u_int unp_count; /* (l) Count of local sockets. */
116 static ino_t unp_ino; /* Prototype for fake inode numbers. */
117 static int unp_rights; /* (g) File descriptors in flight. */
118 static struct unp_head unp_shead; /* (l) List of stream sockets. */
119 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */
120 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */
123 SLIST_ENTRY(unp_defer) ud_link;
126 static SLIST_HEAD(, unp_defer) unp_defers;
127 static int unp_defers_count;
129 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
132 * Garbage collection of cyclic file descriptor/socket references occurs
133 * asynchronously in a taskqueue context in order to avoid recursion and
134 * reentrance in the UNIX domain socket, file descriptor, and socket layer
135 * code. See unp_gc() for a full description.
137 static struct timeout_task unp_gc_task;
140 * The close of unix domain sockets attached as SCM_RIGHTS is
141 * postponed to the taskqueue, to avoid arbitrary recursion depth.
142 * The attached sockets might have another sockets attached.
144 static struct task unp_defer_task;
147 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
148 * stream sockets, although the total for sender and receiver is actually
151 * Datagram sockets really use the sendspace as the maximum datagram size,
152 * and don't really want to reserve the sendspace. Their recvspace should be
153 * large enough for at least one max-size datagram plus address.
158 static u_long unpst_sendspace = PIPSIZ;
159 static u_long unpst_recvspace = PIPSIZ;
160 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
161 static u_long unpdg_recvspace = 4*1024;
162 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */
163 static u_long unpsp_recvspace = PIPSIZ;
165 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
166 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
168 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
169 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
172 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
173 &unpst_sendspace, 0, "Default stream send space.");
174 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
175 &unpst_recvspace, 0, "Default stream receive space.");
176 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
177 &unpdg_sendspace, 0, "Default datagram send space.");
178 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
179 &unpdg_recvspace, 0, "Default datagram receive space.");
180 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
181 &unpsp_sendspace, 0, "Default seqpacket send space.");
182 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
183 &unpsp_recvspace, 0, "Default seqpacket receive space.");
184 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
185 "File descriptors in flight.");
186 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
187 &unp_defers_count, 0,
188 "File descriptors deferred to taskqueue for close.");
191 * Locking and synchronization:
193 * Three types of locks exit in the local domain socket implementation: a
194 * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the
195 * global locks, the list lock protects the socket count, global generation
196 * number, and stream/datagram global lists. The linkage lock protects the
197 * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
198 * held exclusively over the acquisition of multiple unpcb locks to prevent
201 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
202 * allocated in pru_attach() and freed in pru_detach(). The validity of that
203 * pointer is an invariant, so no lock is required to dereference the so_pcb
204 * pointer if a valid socket reference is held by the caller. In practice,
205 * this is always true during operations performed on a socket. Each unpcb
206 * has a back-pointer to its socket, unp_socket, which will be stable under
207 * the same circumstances.
209 * This pointer may only be safely dereferenced as long as a valid reference
210 * to the unpcb is held. Typically, this reference will be from the socket,
211 * or from another unpcb when the referring unpcb's lock is held (in order
212 * that the reference not be invalidated during use). For example, to follow
213 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
214 * as unp_socket remains valid as long as the reference to unp_conn is valid.
216 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual
217 * atomic reads without the lock may be performed "lockless", but more
218 * complex reads and read-modify-writes require the mutex to be held. No
219 * lock order is defined between unpcb locks -- multiple unpcb locks may be
220 * acquired at the same time only when holding the linkage rwlock
221 * exclusively, which prevents deadlocks.
223 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
224 * protocols, bind() is a non-atomic operation, and connect() requires
225 * potential sleeping in the protocol, due to potentially waiting on local or
226 * distributed file systems. We try to separate "lookup" operations, which
227 * may sleep, and the IPC operations themselves, which typically can occur
228 * with relative atomicity as locks can be held over the entire operation.
230 * Another tricky issue is simultaneous multi-threaded or multi-process
231 * access to a single UNIX domain socket. These are handled by the flags
232 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
233 * binding, both of which involve dropping UNIX domain socket locks in order
234 * to perform namei() and other file system operations.
236 static struct rwlock unp_link_rwlock;
237 static struct mtx unp_list_lock;
238 static struct mtx unp_defers_lock;
240 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
243 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
245 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
248 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
249 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
250 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
251 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
252 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
255 #define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \
256 "unp_list_lock", NULL, MTX_DEF)
257 #define UNP_LIST_LOCK() mtx_lock(&unp_list_lock)
258 #define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock)
260 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
261 "unp_defer", NULL, MTX_DEF)
262 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
263 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
265 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
266 "unp_mtx", "unp_mtx", \
267 MTX_DUPOK|MTX_DEF|MTX_RECURSE)
268 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
269 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
270 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
271 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
273 static int uipc_connect2(struct socket *, struct socket *);
274 static int uipc_ctloutput(struct socket *, struct sockopt *);
275 static int unp_connect(struct socket *, struct sockaddr *,
277 static int unp_connectat(int, struct socket *, struct sockaddr *,
279 static int unp_connect2(struct socket *so, struct socket *so2, int);
280 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
281 static void unp_dispose(struct mbuf *);
282 static void unp_shutdown(struct unpcb *);
283 static void unp_drop(struct unpcb *, int);
284 static void unp_gc(__unused void *, int);
285 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
286 static void unp_discard(struct file *);
287 static void unp_freerights(struct filedescent **, int);
288 static void unp_init(void);
289 static int unp_internalize(struct mbuf **, struct thread *);
290 static void unp_internalize_fp(struct file *);
291 static int unp_externalize(struct mbuf *, struct mbuf **, int);
292 static int unp_externalize_fp(struct file *);
293 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
294 static void unp_process_defers(void * __unused, int);
297 * Definitions of protocols supported in the LOCAL domain.
299 static struct domain localdomain;
300 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
301 static struct pr_usrreqs uipc_usrreqs_seqpacket;
302 static struct protosw localsw[] = {
304 .pr_type = SOCK_STREAM,
305 .pr_domain = &localdomain,
306 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
307 .pr_ctloutput = &uipc_ctloutput,
308 .pr_usrreqs = &uipc_usrreqs_stream
311 .pr_type = SOCK_DGRAM,
312 .pr_domain = &localdomain,
313 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
314 .pr_ctloutput = &uipc_ctloutput,
315 .pr_usrreqs = &uipc_usrreqs_dgram
318 .pr_type = SOCK_SEQPACKET,
319 .pr_domain = &localdomain,
322 * XXXRW: For now, PR_ADDR because soreceive will bump into them
323 * due to our use of sbappendaddr. A new sbappend variants is needed
324 * that supports both atomic record writes and control data.
326 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
328 .pr_ctloutput = &uipc_ctloutput,
329 .pr_usrreqs = &uipc_usrreqs_seqpacket,
333 static struct domain localdomain = {
334 .dom_family = AF_LOCAL,
336 .dom_init = unp_init,
337 .dom_externalize = unp_externalize,
338 .dom_dispose = unp_dispose,
339 .dom_protosw = localsw,
340 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])]
345 uipc_abort(struct socket *so)
347 struct unpcb *unp, *unp2;
350 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
354 unp2 = unp->unp_conn;
357 unp_drop(unp2, ECONNABORTED);
358 UNP_PCB_UNLOCK(unp2);
365 uipc_accept(struct socket *so, struct sockaddr **nam)
367 struct unpcb *unp, *unp2;
368 const struct sockaddr *sa;
371 * Pass back name of connected socket, if it was bound and we are
372 * still connected (our peer may have closed already!).
375 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
377 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
379 unp2 = unp->unp_conn;
380 if (unp2 != NULL && unp2->unp_addr != NULL) {
382 sa = (struct sockaddr *) unp2->unp_addr;
383 bcopy(sa, *nam, sa->sa_len);
384 UNP_PCB_UNLOCK(unp2);
387 bcopy(sa, *nam, sa->sa_len);
394 uipc_attach(struct socket *so, int proto, struct thread *td)
396 u_long sendspace, recvspace;
400 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
401 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
402 switch (so->so_type) {
404 sendspace = unpst_sendspace;
405 recvspace = unpst_recvspace;
409 sendspace = unpdg_sendspace;
410 recvspace = unpdg_recvspace;
414 sendspace = unpsp_sendspace;
415 recvspace = unpsp_recvspace;
419 panic("uipc_attach");
421 error = soreserve(so, sendspace, recvspace);
425 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
428 LIST_INIT(&unp->unp_refs);
429 UNP_PCB_LOCK_INIT(unp);
430 unp->unp_socket = so;
432 unp->unp_refcount = 1;
435 unp->unp_gencnt = ++unp_gencnt;
437 switch (so->so_type) {
439 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
443 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
447 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
451 panic("uipc_attach");
459 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
461 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
472 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
474 if (soun->sun_len > sizeof(struct sockaddr_un))
476 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
481 * We don't allow simultaneous bind() calls on a single UNIX domain
482 * socket, so flag in-progress operations, and return an error if an
483 * operation is already in progress.
485 * Historically, we have not allowed a socket to be rebound, so this
486 * also returns an error. Not allowing re-binding simplifies the
487 * implementation and avoids a great many possible failure modes.
490 if (unp->unp_vnode != NULL) {
494 if (unp->unp_flags & UNP_BINDING) {
498 unp->unp_flags |= UNP_BINDING;
501 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
502 bcopy(soun->sun_path, buf, namelen);
506 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME,
507 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
508 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
513 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
514 NDFREE(&nd, NDF_ONLY_PNBUF);
524 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
530 vattr.va_type = VSOCK;
531 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
533 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
537 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
538 NDFREE(&nd, NDF_ONLY_PNBUF);
541 vn_finished_write(mp);
545 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
546 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
550 VOP_UNP_BIND(vp, unp->unp_socket);
552 unp->unp_addr = soun;
553 unp->unp_flags &= ~UNP_BINDING;
557 vn_finished_write(mp);
563 unp->unp_flags &= ~UNP_BINDING;
570 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
573 return (uipc_bindat(AT_FDCWD, so, nam, td));
577 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
581 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
583 error = unp_connect(so, nam, td);
589 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
594 KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
596 error = unp_connectat(fd, so, nam, td);
602 uipc_close(struct socket *so)
604 struct unpcb *unp, *unp2;
607 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
611 unp2 = unp->unp_conn;
614 unp_disconnect(unp, unp2);
615 UNP_PCB_UNLOCK(unp2);
622 uipc_connect2(struct socket *so1, struct socket *so2)
624 struct unpcb *unp, *unp2;
629 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
632 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
634 error = unp_connect2(so1, so2, PRU_CONNECT2);
635 UNP_PCB_UNLOCK(unp2);
642 uipc_detach(struct socket *so)
644 struct unpcb *unp, *unp2;
645 struct sockaddr_un *saved_unp_addr;
647 int freeunp, local_unp_rights;
650 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
655 LIST_REMOVE(unp, unp_link);
656 unp->unp_gencnt = ++unp_gencnt;
661 * XXXRW: Should assert vp->v_socket == so.
663 if ((vp = unp->unp_vnode) != NULL) {
665 unp->unp_vnode = NULL;
667 unp2 = unp->unp_conn;
670 unp_disconnect(unp, unp2);
671 UNP_PCB_UNLOCK(unp2);
675 * We hold the linkage lock exclusively, so it's OK to acquire
676 * multiple pcb locks at a time.
678 while (!LIST_EMPTY(&unp->unp_refs)) {
679 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
682 unp_drop(ref, ECONNRESET);
685 local_unp_rights = unp_rights;
687 unp->unp_socket->so_pcb = NULL;
688 saved_unp_addr = unp->unp_addr;
689 unp->unp_addr = NULL;
691 freeunp = (unp->unp_refcount == 0);
692 if (saved_unp_addr != NULL)
693 free(saved_unp_addr, M_SONAME);
695 UNP_PCB_LOCK_DESTROY(unp);
696 uma_zfree(unp_zone, unp);
701 if (local_unp_rights)
702 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
706 uipc_disconnect(struct socket *so)
708 struct unpcb *unp, *unp2;
711 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
715 unp2 = unp->unp_conn;
718 unp_disconnect(unp, unp2);
719 UNP_PCB_UNLOCK(unp2);
727 uipc_listen(struct socket *so, int backlog, struct thread *td)
733 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
736 if (unp->unp_vnode == NULL) {
742 error = solisten_proto_check(so);
744 cru2x(td->td_ucred, &unp->unp_peercred);
745 unp->unp_flags |= UNP_HAVEPCCACHED;
746 solisten_proto(so, backlog);
754 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
756 struct unpcb *unp, *unp2;
757 const struct sockaddr *sa;
760 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
762 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
765 * XXX: It seems that this test always fails even when connection is
766 * established. So, this else clause is added as workaround to
767 * return PF_LOCAL sockaddr.
769 unp2 = unp->unp_conn;
772 if (unp2->unp_addr != NULL)
773 sa = (struct sockaddr *) unp2->unp_addr;
776 bcopy(sa, *nam, sa->sa_len);
777 UNP_PCB_UNLOCK(unp2);
780 bcopy(sa, *nam, sa->sa_len);
787 uipc_rcvd(struct socket *so, int flags)
789 struct unpcb *unp, *unp2;
795 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
797 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
798 panic("uipc_rcvd socktype %d", so->so_type);
801 * Adjust backpressure on sender and wakeup any waiting to write.
803 * The unp lock is acquired to maintain the validity of the unp_conn
804 * pointer; no lock on unp2 is required as unp2->unp_socket will be
805 * static as long as we don't permit unp2 to disconnect from unp,
806 * which is prevented by the lock on unp. We cache values from
807 * so_rcv to avoid holding the so_rcv lock over the entire
808 * transaction on the remote so_snd.
810 SOCKBUF_LOCK(&so->so_rcv);
811 mbcnt = so->so_rcv.sb_mbcnt;
812 sbcc = so->so_rcv.sb_cc;
813 SOCKBUF_UNLOCK(&so->so_rcv);
815 unp2 = unp->unp_conn;
820 so2 = unp2->unp_socket;
821 SOCKBUF_LOCK(&so2->so_snd);
822 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
823 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
824 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
825 newhiwat, RLIM_INFINITY);
826 sowwakeup_locked(so2);
827 unp->unp_mbcnt = mbcnt;
834 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
835 struct mbuf *control, struct thread *td)
837 struct unpcb *unp, *unp2;
839 u_int mbcnt_delta, sbcc;
844 KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
846 if (flags & PRUS_OOB) {
850 if (control != NULL && (error = unp_internalize(&control, td)))
852 if ((nam != NULL) || (flags & PRUS_EOF))
856 switch (so->so_type) {
859 const struct sockaddr *from;
861 unp2 = unp->unp_conn;
863 UNP_LINK_WLOCK_ASSERT();
868 error = unp_connect(so, nam, td);
871 unp2 = unp->unp_conn;
875 * Because connect() and send() are non-atomic in a sendto()
876 * with a target address, it's possible that the socket will
877 * have disconnected before the send() can run. In that case
878 * return the slightly counter-intuitive but otherwise
879 * correct error that the socket is not connected.
886 if (unp2->unp_flags & UNP_WANTCRED)
887 control = unp_addsockcred(td, control);
889 if (unp->unp_addr != NULL)
890 from = (struct sockaddr *)unp->unp_addr;
893 so2 = unp2->unp_socket;
894 SOCKBUF_LOCK(&so2->so_rcv);
895 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
896 sorwakeup_locked(so2);
900 SOCKBUF_UNLOCK(&so2->so_rcv);
904 UNP_LINK_WLOCK_ASSERT();
906 unp_disconnect(unp, unp2);
907 UNP_PCB_UNLOCK(unp2);
915 if ((so->so_state & SS_ISCONNECTED) == 0) {
917 UNP_LINK_WLOCK_ASSERT();
918 error = unp_connect(so, nam, td);
928 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
934 * Because connect() and send() are non-atomic in a sendto()
935 * with a target address, it's possible that the socket will
936 * have disconnected before the send() can run. In that case
937 * return the slightly counter-intuitive but otherwise
938 * correct error that the socket is not connected.
940 * Locking here must be done carefully: the linkage lock
941 * prevents interconnections between unpcbs from changing, so
942 * we can traverse from unp to unp2 without acquiring unp's
943 * lock. Socket buffer locks follow unpcb locks, so we can
944 * acquire both remote and lock socket buffer locks.
946 unp2 = unp->unp_conn;
951 so2 = unp2->unp_socket;
953 SOCKBUF_LOCK(&so2->so_rcv);
954 if (unp2->unp_flags & UNP_WANTCRED) {
956 * Credentials are passed only once on SOCK_STREAM
957 * and SOCK_SEQPACKET.
959 unp2->unp_flags &= ~UNP_WANTCRED;
960 control = unp_addsockcred(td, control);
963 * Send to paired receive port, and then reduce send buffer
964 * hiwater marks to maintain backpressure. Wake up readers.
966 switch (so->so_type) {
968 if (control != NULL) {
969 if (sbappendcontrol_locked(&so2->so_rcv, m,
973 sbappend_locked(&so2->so_rcv, m);
976 case SOCK_SEQPACKET: {
977 const struct sockaddr *from;
980 if (sbappendaddr_locked(&so2->so_rcv, from, m,
988 * XXXRW: While fine for SOCK_STREAM, this conflates maximum
989 * datagram size and back-pressure for SOCK_SEQPACKET, which
990 * can lead to undesired return of EMSGSIZE on send instead
991 * of more desirable blocking.
993 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
994 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
995 sbcc = so2->so_rcv.sb_cc;
996 sorwakeup_locked(so2);
998 SOCKBUF_LOCK(&so->so_snd);
999 if ((int)so->so_snd.sb_hiwat >= (int)(sbcc - unp2->unp_cc))
1000 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
1003 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
1004 newhiwat, RLIM_INFINITY);
1005 so->so_snd.sb_mbmax -= mbcnt_delta;
1006 SOCKBUF_UNLOCK(&so->so_snd);
1007 unp2->unp_cc = sbcc;
1008 UNP_PCB_UNLOCK(unp2);
1013 panic("uipc_send unknown socktype");
1017 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1019 if (flags & PRUS_EOF) {
1023 UNP_PCB_UNLOCK(unp);
1026 if ((nam != NULL) || (flags & PRUS_EOF))
1031 if (control != NULL && error != 0)
1032 unp_dispose(control);
1035 if (control != NULL)
1043 uipc_sense(struct socket *so, struct stat *sb)
1045 struct unpcb *unp, *unp2;
1048 unp = sotounpcb(so);
1049 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1051 sb->st_blksize = so->so_snd.sb_hiwat;
1054 unp2 = unp->unp_conn;
1055 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) &&
1057 so2 = unp2->unp_socket;
1058 sb->st_blksize += so2->so_rcv.sb_cc;
1061 if (unp->unp_ino == 0)
1062 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1063 sb->st_ino = unp->unp_ino;
1064 UNP_PCB_UNLOCK(unp);
1070 uipc_shutdown(struct socket *so)
1074 unp = sotounpcb(so);
1075 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1081 UNP_PCB_UNLOCK(unp);
1087 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1090 const struct sockaddr *sa;
1092 unp = sotounpcb(so);
1093 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1095 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1097 if (unp->unp_addr != NULL)
1098 sa = (struct sockaddr *) unp->unp_addr;
1101 bcopy(sa, *nam, sa->sa_len);
1102 UNP_PCB_UNLOCK(unp);
1106 static struct pr_usrreqs uipc_usrreqs_dgram = {
1107 .pru_abort = uipc_abort,
1108 .pru_accept = uipc_accept,
1109 .pru_attach = uipc_attach,
1110 .pru_bind = uipc_bind,
1111 .pru_bindat = uipc_bindat,
1112 .pru_connect = uipc_connect,
1113 .pru_connectat = uipc_connectat,
1114 .pru_connect2 = uipc_connect2,
1115 .pru_detach = uipc_detach,
1116 .pru_disconnect = uipc_disconnect,
1117 .pru_listen = uipc_listen,
1118 .pru_peeraddr = uipc_peeraddr,
1119 .pru_rcvd = uipc_rcvd,
1120 .pru_send = uipc_send,
1121 .pru_sense = uipc_sense,
1122 .pru_shutdown = uipc_shutdown,
1123 .pru_sockaddr = uipc_sockaddr,
1124 .pru_soreceive = soreceive_dgram,
1125 .pru_close = uipc_close,
1128 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1129 .pru_abort = uipc_abort,
1130 .pru_accept = uipc_accept,
1131 .pru_attach = uipc_attach,
1132 .pru_bind = uipc_bind,
1133 .pru_bindat = uipc_bindat,
1134 .pru_connect = uipc_connect,
1135 .pru_connectat = uipc_connectat,
1136 .pru_connect2 = uipc_connect2,
1137 .pru_detach = uipc_detach,
1138 .pru_disconnect = uipc_disconnect,
1139 .pru_listen = uipc_listen,
1140 .pru_peeraddr = uipc_peeraddr,
1141 .pru_rcvd = uipc_rcvd,
1142 .pru_send = uipc_send,
1143 .pru_sense = uipc_sense,
1144 .pru_shutdown = uipc_shutdown,
1145 .pru_sockaddr = uipc_sockaddr,
1146 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1147 .pru_close = uipc_close,
1150 static struct pr_usrreqs uipc_usrreqs_stream = {
1151 .pru_abort = uipc_abort,
1152 .pru_accept = uipc_accept,
1153 .pru_attach = uipc_attach,
1154 .pru_bind = uipc_bind,
1155 .pru_bindat = uipc_bindat,
1156 .pru_connect = uipc_connect,
1157 .pru_connectat = uipc_connectat,
1158 .pru_connect2 = uipc_connect2,
1159 .pru_detach = uipc_detach,
1160 .pru_disconnect = uipc_disconnect,
1161 .pru_listen = uipc_listen,
1162 .pru_peeraddr = uipc_peeraddr,
1163 .pru_rcvd = uipc_rcvd,
1164 .pru_send = uipc_send,
1165 .pru_sense = uipc_sense,
1166 .pru_shutdown = uipc_shutdown,
1167 .pru_sockaddr = uipc_sockaddr,
1168 .pru_soreceive = soreceive_generic,
1169 .pru_close = uipc_close,
1173 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1179 if (sopt->sopt_level != 0)
1182 unp = sotounpcb(so);
1183 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1185 switch (sopt->sopt_dir) {
1187 switch (sopt->sopt_name) {
1188 case LOCAL_PEERCRED:
1190 if (unp->unp_flags & UNP_HAVEPC)
1191 xu = unp->unp_peercred;
1193 if (so->so_type == SOCK_STREAM)
1198 UNP_PCB_UNLOCK(unp);
1200 error = sooptcopyout(sopt, &xu, sizeof(xu));
1204 /* Unlocked read. */
1205 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1206 error = sooptcopyout(sopt, &optval, sizeof(optval));
1209 case LOCAL_CONNWAIT:
1210 /* Unlocked read. */
1211 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1212 error = sooptcopyout(sopt, &optval, sizeof(optval));
1222 switch (sopt->sopt_name) {
1224 case LOCAL_CONNWAIT:
1225 error = sooptcopyin(sopt, &optval, sizeof(optval),
1230 #define OPTSET(bit) do { \
1231 UNP_PCB_LOCK(unp); \
1233 unp->unp_flags |= bit; \
1235 unp->unp_flags &= ~bit; \
1236 UNP_PCB_UNLOCK(unp); \
1239 switch (sopt->sopt_name) {
1241 OPTSET(UNP_WANTCRED);
1244 case LOCAL_CONNWAIT:
1245 OPTSET(UNP_CONNWAIT);
1254 error = ENOPROTOOPT;
1267 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1270 return (unp_connectat(AT_FDCWD, so, nam, td));
1274 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1277 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1279 struct socket *so2, *so3;
1280 struct unpcb *unp, *unp2, *unp3;
1281 struct nameidata nd;
1282 char buf[SOCK_MAXADDRLEN];
1283 struct sockaddr *sa;
1284 cap_rights_t rights;
1287 UNP_LINK_WLOCK_ASSERT();
1289 unp = sotounpcb(so);
1290 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1292 if (nam->sa_len > sizeof(struct sockaddr_un))
1294 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1297 bcopy(soun->sun_path, buf, len);
1301 if (unp->unp_flags & UNP_CONNECTING) {
1302 UNP_PCB_UNLOCK(unp);
1306 unp->unp_flags |= UNP_CONNECTING;
1307 UNP_PCB_UNLOCK(unp);
1309 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1310 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1311 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1317 ASSERT_VOP_LOCKED(vp, "unp_connect");
1318 NDFREE(&nd, NDF_ONLY_PNBUF);
1322 if (vp->v_type != VSOCK) {
1327 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1331 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1335 unp = sotounpcb(so);
1336 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1339 * Lock linkage lock for two reasons: make sure v_socket is stable,
1340 * and to protect simultaneous locking of multiple pcbs.
1343 VOP_UNP_CONNECT(vp, &so2);
1345 error = ECONNREFUSED;
1348 if (so->so_type != so2->so_type) {
1352 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1353 if (so2->so_options & SO_ACCEPTCONN) {
1354 CURVNET_SET(so2->so_vnet);
1355 so3 = sonewconn(so2, 0);
1360 error = ECONNREFUSED;
1363 unp = sotounpcb(so);
1364 unp2 = sotounpcb(so2);
1365 unp3 = sotounpcb(so3);
1369 if (unp2->unp_addr != NULL) {
1370 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1371 unp3->unp_addr = (struct sockaddr_un *) sa;
1376 * The connector's (client's) credentials are copied from its
1377 * process structure at the time of connect() (which is now).
1379 cru2x(td->td_ucred, &unp3->unp_peercred);
1380 unp3->unp_flags |= UNP_HAVEPC;
1383 * The receiver's (server's) credentials are copied from the
1384 * unp_peercred member of socket on which the former called
1385 * listen(); uipc_listen() cached that process's credentials
1386 * at that time so we can use them now.
1388 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1389 ("unp_connect: listener without cached peercred"));
1390 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1391 sizeof(unp->unp_peercred));
1392 unp->unp_flags |= UNP_HAVEPC;
1393 if (unp2->unp_flags & UNP_WANTCRED)
1394 unp3->unp_flags |= UNP_WANTCRED;
1395 UNP_PCB_UNLOCK(unp3);
1396 UNP_PCB_UNLOCK(unp2);
1397 UNP_PCB_UNLOCK(unp);
1399 mac_socketpeer_set_from_socket(so, so3);
1400 mac_socketpeer_set_from_socket(so3, so);
1405 unp = sotounpcb(so);
1406 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1407 unp2 = sotounpcb(so2);
1408 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1411 error = unp_connect2(so, so2, PRU_CONNECT);
1412 UNP_PCB_UNLOCK(unp2);
1413 UNP_PCB_UNLOCK(unp);
1422 unp->unp_flags &= ~UNP_CONNECTING;
1423 UNP_PCB_UNLOCK(unp);
1428 unp_connect2(struct socket *so, struct socket *so2, int req)
1433 unp = sotounpcb(so);
1434 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1435 unp2 = sotounpcb(so2);
1436 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1438 UNP_LINK_WLOCK_ASSERT();
1439 UNP_PCB_LOCK_ASSERT(unp);
1440 UNP_PCB_LOCK_ASSERT(unp2);
1442 if (so2->so_type != so->so_type)
1443 return (EPROTOTYPE);
1444 unp->unp_conn = unp2;
1446 switch (so->so_type) {
1448 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1453 case SOCK_SEQPACKET:
1454 unp2->unp_conn = unp;
1455 if (req == PRU_CONNECT &&
1456 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1464 panic("unp_connect2");
1470 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1474 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1476 UNP_LINK_WLOCK_ASSERT();
1477 UNP_PCB_LOCK_ASSERT(unp);
1478 UNP_PCB_LOCK_ASSERT(unp2);
1480 unp->unp_conn = NULL;
1481 switch (unp->unp_socket->so_type) {
1483 LIST_REMOVE(unp, unp_reflink);
1484 so = unp->unp_socket;
1486 so->so_state &= ~SS_ISCONNECTED;
1491 case SOCK_SEQPACKET:
1492 soisdisconnected(unp->unp_socket);
1493 unp2->unp_conn = NULL;
1494 soisdisconnected(unp2->unp_socket);
1500 * unp_pcblist() walks the global list of struct unpcb's to generate a
1501 * pointer list, bumping the refcount on each unpcb. It then copies them out
1502 * sequentially, validating the generation number on each to see if it has
1503 * been detached. All of this is necessary because copyout() may sleep on
1507 unp_pcblist(SYSCTL_HANDLER_ARGS)
1511 struct unpcb *unp, **unp_list;
1513 struct xunpgen *xug;
1514 struct unp_head *head;
1517 switch ((intptr_t)arg1) {
1526 case SOCK_SEQPACKET:
1531 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1535 * The process of preparing the PCB list is too time-consuming and
1536 * resource-intensive to repeat twice on every request.
1538 if (req->oldptr == NULL) {
1540 req->oldidx = 2 * (sizeof *xug)
1541 + (n + n/8) * sizeof(struct xunpcb);
1545 if (req->newptr != NULL)
1549 * OK, now we're committed to doing something.
1551 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1553 gencnt = unp_gencnt;
1557 xug->xug_len = sizeof *xug;
1559 xug->xug_gen = gencnt;
1560 xug->xug_sogen = so_gencnt;
1561 error = SYSCTL_OUT(req, xug, sizeof *xug);
1567 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1570 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1571 unp = LIST_NEXT(unp, unp_link)) {
1573 if (unp->unp_gencnt <= gencnt) {
1574 if (cr_cansee(req->td->td_ucred,
1575 unp->unp_socket->so_cred)) {
1576 UNP_PCB_UNLOCK(unp);
1579 unp_list[i++] = unp;
1580 unp->unp_refcount++;
1582 UNP_PCB_UNLOCK(unp);
1585 n = i; /* In case we lost some during malloc. */
1588 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1589 for (i = 0; i < n; i++) {
1592 unp->unp_refcount--;
1593 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1594 xu->xu_len = sizeof *xu;
1597 * XXX - need more locking here to protect against
1598 * connect/disconnect races for SMP.
1600 if (unp->unp_addr != NULL)
1601 bcopy(unp->unp_addr, &xu->xu_addr,
1602 unp->unp_addr->sun_len);
1603 if (unp->unp_conn != NULL &&
1604 unp->unp_conn->unp_addr != NULL)
1605 bcopy(unp->unp_conn->unp_addr,
1607 unp->unp_conn->unp_addr->sun_len);
1608 bcopy(unp, &xu->xu_unp, sizeof *unp);
1609 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1610 UNP_PCB_UNLOCK(unp);
1611 error = SYSCTL_OUT(req, xu, sizeof *xu);
1613 freeunp = (unp->unp_refcount == 0);
1614 UNP_PCB_UNLOCK(unp);
1616 UNP_PCB_LOCK_DESTROY(unp);
1617 uma_zfree(unp_zone, unp);
1624 * Give the user an updated idea of our state. If the
1625 * generation differs from what we told her before, she knows
1626 * that something happened while we were processing this
1627 * request, and it might be necessary to retry.
1629 xug->xug_gen = unp_gencnt;
1630 xug->xug_sogen = so_gencnt;
1631 xug->xug_count = unp_count;
1632 error = SYSCTL_OUT(req, xug, sizeof *xug);
1634 free(unp_list, M_TEMP);
1639 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1640 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1641 "List of active local datagram sockets");
1642 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1643 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1644 "List of active local stream sockets");
1645 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1646 CTLTYPE_OPAQUE | CTLFLAG_RD,
1647 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1648 "List of active local seqpacket sockets");
1651 unp_shutdown(struct unpcb *unp)
1656 UNP_LINK_WLOCK_ASSERT();
1657 UNP_PCB_LOCK_ASSERT(unp);
1659 unp2 = unp->unp_conn;
1660 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1661 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1662 so = unp2->unp_socket;
1669 unp_drop(struct unpcb *unp, int errno)
1671 struct socket *so = unp->unp_socket;
1674 UNP_LINK_WLOCK_ASSERT();
1675 UNP_PCB_LOCK_ASSERT(unp);
1677 so->so_error = errno;
1678 unp2 = unp->unp_conn;
1682 unp_disconnect(unp, unp2);
1683 UNP_PCB_UNLOCK(unp2);
1687 unp_freerights(struct filedescent **fdep, int fdcount)
1692 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1694 for (i = 0; i < fdcount; i++) {
1695 fp = fdep[i]->fde_file;
1696 filecaps_free(&fdep[i]->fde_caps);
1699 free(fdep[0], M_FILECAPS);
1703 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1705 struct thread *td = curthread; /* XXX */
1706 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1709 struct filedesc *fdesc = td->td_proc->p_fd;
1710 struct filedescent *fde, **fdep;
1712 socklen_t clen = control->m_len, datalen;
1716 UNP_LINK_UNLOCK_ASSERT();
1719 if (controlp != NULL) /* controlp == NULL => free control messages */
1721 while (cm != NULL) {
1722 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1726 data = CMSG_DATA(cm);
1727 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1728 if (cm->cmsg_level == SOL_SOCKET
1729 && cm->cmsg_type == SCM_RIGHTS) {
1730 newfds = datalen / sizeof(*fdep);
1735 /* If we're not outputting the descriptors free them. */
1736 if (error || controlp == NULL) {
1737 unp_freerights(fdep, newfds);
1740 FILEDESC_XLOCK(fdesc);
1743 * Now change each pointer to an fd in the global
1744 * table to an integer that is the index to the local
1745 * fd table entry that we set up to point to the
1746 * global one we are transferring.
1748 newlen = newfds * sizeof(int);
1749 *controlp = sbcreatecontrol(NULL, newlen,
1750 SCM_RIGHTS, SOL_SOCKET);
1751 if (*controlp == NULL) {
1752 FILEDESC_XUNLOCK(fdesc);
1754 unp_freerights(fdep, newfds);
1759 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1760 if (fdallocn(td, 0, fdp, newfds) != 0) {
1761 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1763 unp_freerights(fdep, newfds);
1768 for (i = 0; i < newfds; i++, fdp++) {
1769 fde = &fdesc->fd_ofiles[*fdp];
1770 fde->fde_file = fdep[i]->fde_file;
1771 filecaps_move(&fdep[i]->fde_caps,
1773 if ((flags & MSG_CMSG_CLOEXEC) != 0)
1774 fde->fde_flags |= UF_EXCLOSE;
1775 unp_externalize_fp(fde->fde_file);
1777 FILEDESC_XUNLOCK(fdesc);
1778 free(fdep[0], M_FILECAPS);
1780 /* We can just copy anything else across. */
1781 if (error || controlp == NULL)
1783 *controlp = sbcreatecontrol(NULL, datalen,
1784 cm->cmsg_type, cm->cmsg_level);
1785 if (*controlp == NULL) {
1790 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1793 controlp = &(*controlp)->m_next;
1796 if (CMSG_SPACE(datalen) < clen) {
1797 clen -= CMSG_SPACE(datalen);
1798 cm = (struct cmsghdr *)
1799 ((caddr_t)cm + CMSG_SPACE(datalen));
1811 unp_zone_change(void *tag)
1814 uma_zone_set_max(unp_zone, maxsockets);
1822 if (!IS_DEFAULT_VNET(curvnet))
1825 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1826 NULL, NULL, UMA_ALIGN_PTR, 0);
1827 if (unp_zone == NULL)
1829 uma_zone_set_max(unp_zone, maxsockets);
1830 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1831 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1832 NULL, EVENTHANDLER_PRI_ANY);
1833 LIST_INIT(&unp_dhead);
1834 LIST_INIT(&unp_shead);
1835 LIST_INIT(&unp_sphead);
1836 SLIST_INIT(&unp_defers);
1837 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1838 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1839 UNP_LINK_LOCK_INIT();
1840 UNP_LIST_LOCK_INIT();
1841 UNP_DEFERRED_LOCK_INIT();
1845 unp_internalize(struct mbuf **controlp, struct thread *td)
1847 struct mbuf *control = *controlp;
1848 struct proc *p = td->td_proc;
1849 struct filedesc *fdesc = p->p_fd;
1851 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1852 struct cmsgcred *cmcred;
1853 struct filedescent *fde, **fdep, *fdev;
1858 socklen_t clen = control->m_len, datalen;
1862 UNP_LINK_UNLOCK_ASSERT();
1866 while (cm != NULL) {
1867 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1868 || cm->cmsg_len > clen) {
1872 data = CMSG_DATA(cm);
1873 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1875 switch (cm->cmsg_type) {
1877 * Fill in credential information.
1880 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1881 SCM_CREDS, SOL_SOCKET);
1882 if (*controlp == NULL) {
1886 cmcred = (struct cmsgcred *)
1887 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1888 cmcred->cmcred_pid = p->p_pid;
1889 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1890 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1891 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1892 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1894 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1895 cmcred->cmcred_groups[i] =
1896 td->td_ucred->cr_groups[i];
1900 oldfds = datalen / sizeof (int);
1904 * Check that all the FDs passed in refer to legal
1905 * files. If not, reject the entire operation.
1908 FILEDESC_SLOCK(fdesc);
1909 for (i = 0; i < oldfds; i++) {
1911 if (fget_locked(fdesc, fd) == NULL) {
1912 FILEDESC_SUNLOCK(fdesc);
1916 fp = fdesc->fd_ofiles[fd].fde_file;
1917 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1918 FILEDESC_SUNLOCK(fdesc);
1926 * Now replace the integer FDs with pointers to the
1927 * file structure and capability rights.
1929 newlen = oldfds * sizeof(fdep[0]);
1930 *controlp = sbcreatecontrol(NULL, newlen,
1931 SCM_RIGHTS, SOL_SOCKET);
1932 if (*controlp == NULL) {
1933 FILEDESC_SUNLOCK(fdesc);
1938 fdep = (struct filedescent **)
1939 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1940 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1942 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1943 fde = &fdesc->fd_ofiles[*fdp];
1945 fdep[i]->fde_file = fde->fde_file;
1946 filecaps_copy(&fde->fde_caps,
1947 &fdep[i]->fde_caps);
1948 unp_internalize_fp(fdep[i]->fde_file);
1950 FILEDESC_SUNLOCK(fdesc);
1954 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1955 SCM_TIMESTAMP, SOL_SOCKET);
1956 if (*controlp == NULL) {
1960 tv = (struct timeval *)
1961 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1966 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
1967 SCM_BINTIME, SOL_SOCKET);
1968 if (*controlp == NULL) {
1972 bt = (struct bintime *)
1973 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1982 controlp = &(*controlp)->m_next;
1983 if (CMSG_SPACE(datalen) < clen) {
1984 clen -= CMSG_SPACE(datalen);
1985 cm = (struct cmsghdr *)
1986 ((caddr_t)cm + CMSG_SPACE(datalen));
1998 static struct mbuf *
1999 unp_addsockcred(struct thread *td, struct mbuf *control)
2001 struct mbuf *m, *n, *n_prev;
2002 struct sockcred *sc;
2003 const struct cmsghdr *cm;
2007 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2008 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2012 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2013 sc->sc_uid = td->td_ucred->cr_ruid;
2014 sc->sc_euid = td->td_ucred->cr_uid;
2015 sc->sc_gid = td->td_ucred->cr_rgid;
2016 sc->sc_egid = td->td_ucred->cr_gid;
2017 sc->sc_ngroups = ngroups;
2018 for (i = 0; i < sc->sc_ngroups; i++)
2019 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2022 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2023 * created SCM_CREDS control message (struct sockcred) has another
2026 if (control != NULL)
2027 for (n = control, n_prev = NULL; n != NULL;) {
2028 cm = mtod(n, struct cmsghdr *);
2029 if (cm->cmsg_level == SOL_SOCKET &&
2030 cm->cmsg_type == SCM_CREDS) {
2032 control = n->m_next;
2034 n_prev->m_next = n->m_next;
2042 /* Prepend it to the head. */
2043 m->m_next = control;
2047 static struct unpcb *
2048 fptounp(struct file *fp)
2052 if (fp->f_type != DTYPE_SOCKET)
2054 if ((so = fp->f_data) == NULL)
2056 if (so->so_proto->pr_domain != &localdomain)
2058 return sotounpcb(so);
2062 unp_discard(struct file *fp)
2064 struct unp_defer *dr;
2066 if (unp_externalize_fp(fp)) {
2067 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2069 UNP_DEFERRED_LOCK();
2070 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2071 UNP_DEFERRED_UNLOCK();
2072 atomic_add_int(&unp_defers_count, 1);
2073 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2075 (void) closef(fp, (struct thread *)NULL);
2079 unp_process_defers(void *arg __unused, int pending)
2081 struct unp_defer *dr;
2082 SLIST_HEAD(, unp_defer) drl;
2087 UNP_DEFERRED_LOCK();
2088 if (SLIST_FIRST(&unp_defers) == NULL) {
2089 UNP_DEFERRED_UNLOCK();
2092 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2093 UNP_DEFERRED_UNLOCK();
2095 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2096 SLIST_REMOVE_HEAD(&drl, ud_link);
2097 closef(dr->ud_fp, NULL);
2101 atomic_add_int(&unp_defers_count, -count);
2106 unp_internalize_fp(struct file *fp)
2111 if ((unp = fptounp(fp)) != NULL) {
2113 unp->unp_msgcount++;
2121 unp_externalize_fp(struct file *fp)
2127 if ((unp = fptounp(fp)) != NULL) {
2128 unp->unp_msgcount--;
2138 * unp_defer indicates whether additional work has been defered for a future
2139 * pass through unp_gc(). It is thread local and does not require explicit
2142 static int unp_marked;
2143 static int unp_unreachable;
2146 unp_accessable(struct filedescent **fdep, int fdcount)
2152 for (i = 0; i < fdcount; i++) {
2153 fp = fdep[i]->fde_file;
2154 if ((unp = fptounp(fp)) == NULL)
2156 if (unp->unp_gcflag & UNPGC_REF)
2158 unp->unp_gcflag &= ~UNPGC_DEAD;
2159 unp->unp_gcflag |= UNPGC_REF;
2165 unp_gc_process(struct unpcb *unp)
2171 /* Already processed. */
2172 if (unp->unp_gcflag & UNPGC_SCANNED)
2177 * Check for a socket potentially in a cycle. It must be in a
2178 * queue as indicated by msgcount, and this must equal the file
2179 * reference count. Note that when msgcount is 0 the file is NULL.
2181 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2182 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2183 unp->unp_gcflag |= UNPGC_DEAD;
2189 * Mark all sockets we reference with RIGHTS.
2191 so = unp->unp_socket;
2192 SOCKBUF_LOCK(&so->so_rcv);
2193 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2194 SOCKBUF_UNLOCK(&so->so_rcv);
2197 * Mark all sockets in our accept queue.
2200 TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2201 SOCKBUF_LOCK(&soa->so_rcv);
2202 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2203 SOCKBUF_UNLOCK(&soa->so_rcv);
2206 unp->unp_gcflag |= UNPGC_SCANNED;
2209 static int unp_recycled;
2210 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2211 "Number of unreachable sockets claimed by the garbage collector.");
2213 static int unp_taskcount;
2214 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2215 "Number of times the garbage collector has run.");
2218 unp_gc(__unused void *arg, int pending)
2220 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2222 struct unp_head **head;
2223 struct file *f, **unref;
2230 * First clear all gc flags from previous runs.
2232 for (head = heads; *head != NULL; head++)
2233 LIST_FOREACH(unp, *head, unp_link)
2234 unp->unp_gcflag = 0;
2237 * Scan marking all reachable sockets with UNPGC_REF. Once a socket
2238 * is reachable all of the sockets it references are reachable.
2239 * Stop the scan once we do a complete loop without discovering
2240 * a new reachable socket.
2243 unp_unreachable = 0;
2245 for (head = heads; *head != NULL; head++)
2246 LIST_FOREACH(unp, *head, unp_link)
2247 unp_gc_process(unp);
2248 } while (unp_marked);
2250 if (unp_unreachable == 0)
2254 * Allocate space for a local list of dead unpcbs.
2256 unref = malloc(unp_unreachable * sizeof(struct file *),
2260 * Iterate looking for sockets which have been specifically marked
2261 * as as unreachable and store them locally.
2265 for (total = 0, head = heads; *head != NULL; head++)
2266 LIST_FOREACH(unp, *head, unp_link)
2267 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2269 if (unp->unp_msgcount == 0 || f == NULL ||
2270 f->f_count != unp->unp_msgcount)
2274 KASSERT(total <= unp_unreachable,
2275 ("unp_gc: incorrect unreachable count."));
2281 * Now flush all sockets, free'ing rights. This will free the
2282 * struct files associated with these sockets but leave each socket
2283 * with one remaining ref.
2285 for (i = 0; i < total; i++) {
2288 so = unref[i]->f_data;
2289 CURVNET_SET(so->so_vnet);
2295 * And finally release the sockets so they can be reclaimed.
2297 for (i = 0; i < total; i++)
2298 fdrop(unref[i], NULL);
2299 unp_recycled += total;
2300 free(unref, M_TEMP);
2304 unp_dispose(struct mbuf *m)
2308 unp_scan(m, unp_freerights);
2312 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2317 socklen_t clen, datalen;
2319 while (m0 != NULL) {
2320 for (m = m0; m; m = m->m_next) {
2321 if (m->m_type != MT_CONTROL)
2324 cm = mtod(m, struct cmsghdr *);
2327 while (cm != NULL) {
2328 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2331 data = CMSG_DATA(cm);
2332 datalen = (caddr_t)cm + cm->cmsg_len
2335 if (cm->cmsg_level == SOL_SOCKET &&
2336 cm->cmsg_type == SCM_RIGHTS) {
2337 (*op)(data, datalen /
2338 sizeof(struct filedescent *));
2341 if (CMSG_SPACE(datalen) < clen) {
2342 clen -= CMSG_SPACE(datalen);
2343 cm = (struct cmsghdr *)
2344 ((caddr_t)cm + CMSG_SPACE(datalen));
2356 * A helper function called by VFS before socket-type vnode reclamation.
2357 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2361 vfs_unp_reclaim(struct vnode *vp)
2367 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2368 KASSERT(vp->v_type == VSOCK,
2369 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2373 VOP_UNP_CONNECT(vp, &so);
2376 unp = sotounpcb(so);
2380 if (unp->unp_vnode == vp) {
2382 unp->unp_vnode = NULL;
2385 UNP_PCB_UNLOCK(unp);
2394 db_print_indent(int indent)
2398 for (i = 0; i < indent; i++)
2403 db_print_unpflags(int unp_flags)
2408 if (unp_flags & UNP_HAVEPC) {
2409 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2412 if (unp_flags & UNP_HAVEPCCACHED) {
2413 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2416 if (unp_flags & UNP_WANTCRED) {
2417 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2420 if (unp_flags & UNP_CONNWAIT) {
2421 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2424 if (unp_flags & UNP_CONNECTING) {
2425 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2428 if (unp_flags & UNP_BINDING) {
2429 db_printf("%sUNP_BINDING", comma ? ", " : "");
2435 db_print_xucred(int indent, struct xucred *xu)
2439 db_print_indent(indent);
2440 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
2441 xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2442 db_print_indent(indent);
2443 db_printf("cr_groups: ");
2445 for (i = 0; i < xu->cr_ngroups; i++) {
2446 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2453 db_print_unprefs(int indent, struct unp_head *uh)
2459 LIST_FOREACH(unp, uh, unp_reflink) {
2460 if (counter % 4 == 0)
2461 db_print_indent(indent);
2462 db_printf("%p ", unp);
2463 if (counter % 4 == 3)
2467 if (counter != 0 && counter % 4 != 0)
2471 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2476 db_printf("usage: show unpcb <addr>\n");
2479 unp = (struct unpcb *)addr;
2481 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2484 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2487 db_printf("unp_refs:\n");
2488 db_print_unprefs(2, &unp->unp_refs);
2490 /* XXXRW: Would be nice to print the full address, if any. */
2491 db_printf("unp_addr: %p\n", unp->unp_addr);
2493 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n",
2494 unp->unp_cc, unp->unp_mbcnt,
2495 (unsigned long long)unp->unp_gencnt);
2497 db_printf("unp_flags: %x (", unp->unp_flags);
2498 db_print_unpflags(unp->unp_flags);
2501 db_printf("unp_peercred:\n");
2502 db_print_xucred(2, &unp->unp_peercred);
2504 db_printf("unp_refcount: %u\n", unp->unp_refcount);