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 * rethink name space problems
55 * need a proper out-of-band
58 #include <sys/cdefs.h>
59 __FBSDID("$FreeBSD$");
63 #include <sys/param.h>
64 #include <sys/capsicum.h>
65 #include <sys/domain.h>
66 #include <sys/fcntl.h>
67 #include <sys/malloc.h> /* XXX must be before <sys/file.h> */
68 #include <sys/eventhandler.h>
70 #include <sys/filedesc.h>
71 #include <sys/kernel.h>
74 #include <sys/mount.h>
75 #include <sys/mutex.h>
76 #include <sys/namei.h>
78 #include <sys/protosw.h>
79 #include <sys/queue.h>
80 #include <sys/resourcevar.h>
81 #include <sys/rwlock.h>
82 #include <sys/socket.h>
83 #include <sys/socketvar.h>
84 #include <sys/signalvar.h>
87 #include <sys/sysctl.h>
88 #include <sys/systm.h>
89 #include <sys/taskqueue.h>
91 #include <sys/unpcb.h>
92 #include <sys/vnode.h>
100 #include <security/mac/mac_framework.h>
104 MALLOC_DECLARE(M_FILECAPS);
108 * (l) Locked using list lock
109 * (g) Locked using linkage lock
112 static uma_zone_t unp_zone;
113 static unp_gen_t unp_gencnt; /* (l) */
114 static u_int unp_count; /* (l) Count of local sockets. */
115 static ino_t unp_ino; /* Prototype for fake inode numbers. */
116 static int unp_rights; /* (g) File descriptors in flight. */
117 static struct unp_head unp_shead; /* (l) List of stream sockets. */
118 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */
119 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */
122 SLIST_ENTRY(unp_defer) ud_link;
125 static SLIST_HEAD(, unp_defer) unp_defers;
126 static int unp_defers_count;
128 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
131 * Garbage collection of cyclic file descriptor/socket references occurs
132 * asynchronously in a taskqueue context in order to avoid recursion and
133 * reentrance in the UNIX domain socket, file descriptor, and socket layer
134 * code. See unp_gc() for a full description.
136 static struct timeout_task unp_gc_task;
139 * The close of unix domain sockets attached as SCM_RIGHTS is
140 * postponed to the taskqueue, to avoid arbitrary recursion depth.
141 * The attached sockets might have another sockets attached.
143 static struct task unp_defer_task;
146 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
147 * stream sockets, although the total for sender and receiver is actually
150 * Datagram sockets really use the sendspace as the maximum datagram size,
151 * and don't really want to reserve the sendspace. Their recvspace should be
152 * large enough for at least one max-size datagram plus address.
157 static u_long unpst_sendspace = PIPSIZ;
158 static u_long unpst_recvspace = PIPSIZ;
159 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
160 static u_long unpdg_recvspace = 4*1024;
161 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */
162 static u_long unpsp_recvspace = PIPSIZ;
164 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
165 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
167 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
168 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
171 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
172 &unpst_sendspace, 0, "Default stream send space.");
173 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
174 &unpst_recvspace, 0, "Default stream receive space.");
175 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
176 &unpdg_sendspace, 0, "Default datagram send space.");
177 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
178 &unpdg_recvspace, 0, "Default datagram receive space.");
179 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
180 &unpsp_sendspace, 0, "Default seqpacket send space.");
181 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
182 &unpsp_recvspace, 0, "Default seqpacket receive space.");
183 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
184 "File descriptors in flight.");
185 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
186 &unp_defers_count, 0,
187 "File descriptors deferred to taskqueue for close.");
190 * Locking and synchronization:
192 * Three types of locks exit in the local domain socket implementation: a
193 * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the
194 * global locks, the list lock protects the socket count, global generation
195 * number, and stream/datagram global lists. The linkage lock protects the
196 * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
197 * held exclusively over the acquisition of multiple unpcb locks to prevent
200 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
201 * allocated in pru_attach() and freed in pru_detach(). The validity of that
202 * pointer is an invariant, so no lock is required to dereference the so_pcb
203 * pointer if a valid socket reference is held by the caller. In practice,
204 * this is always true during operations performed on a socket. Each unpcb
205 * has a back-pointer to its socket, unp_socket, which will be stable under
206 * the same circumstances.
208 * This pointer may only be safely dereferenced as long as a valid reference
209 * to the unpcb is held. Typically, this reference will be from the socket,
210 * or from another unpcb when the referring unpcb's lock is held (in order
211 * that the reference not be invalidated during use). For example, to follow
212 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
213 * as unp_socket remains valid as long as the reference to unp_conn is valid.
215 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual
216 * atomic reads without the lock may be performed "lockless", but more
217 * complex reads and read-modify-writes require the mutex to be held. No
218 * lock order is defined between unpcb locks -- multiple unpcb locks may be
219 * acquired at the same time only when holding the linkage rwlock
220 * exclusively, which prevents deadlocks.
222 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
223 * protocols, bind() is a non-atomic operation, and connect() requires
224 * potential sleeping in the protocol, due to potentially waiting on local or
225 * distributed file systems. We try to separate "lookup" operations, which
226 * may sleep, and the IPC operations themselves, which typically can occur
227 * with relative atomicity as locks can be held over the entire operation.
229 * Another tricky issue is simultaneous multi-threaded or multi-process
230 * access to a single UNIX domain socket. These are handled by the flags
231 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
232 * binding, both of which involve dropping UNIX domain socket locks in order
233 * to perform namei() and other file system operations.
235 static struct rwlock unp_link_rwlock;
236 static struct mtx unp_list_lock;
237 static struct mtx unp_defers_lock;
239 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
242 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
244 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
247 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
248 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
249 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
250 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
251 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
254 #define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \
255 "unp_list_lock", NULL, MTX_DEF)
256 #define UNP_LIST_LOCK() mtx_lock(&unp_list_lock)
257 #define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock)
259 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
260 "unp_defer", NULL, MTX_DEF)
261 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
262 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
264 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
265 "unp_mtx", "unp_mtx", \
266 MTX_DUPOK|MTX_DEF|MTX_RECURSE)
267 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
268 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
269 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
270 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
272 static int uipc_connect2(struct socket *, struct socket *);
273 static int uipc_ctloutput(struct socket *, struct sockopt *);
274 static int unp_connect(struct socket *, struct sockaddr *,
276 static int unp_connectat(int, struct socket *, struct sockaddr *,
278 static int unp_connect2(struct socket *so, struct socket *so2, int);
279 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
280 static void unp_dispose(struct mbuf *);
281 static void unp_shutdown(struct unpcb *);
282 static void unp_drop(struct unpcb *, int);
283 static void unp_gc(__unused void *, int);
284 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
285 static void unp_discard(struct file *);
286 static void unp_freerights(struct filedescent **, int);
287 static void unp_init(void);
288 static int unp_internalize(struct mbuf **, struct thread *);
289 static void unp_internalize_fp(struct file *);
290 static int unp_externalize(struct mbuf *, struct mbuf **, int);
291 static int unp_externalize_fp(struct file *);
292 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
293 static void unp_process_defers(void * __unused, int);
296 * Definitions of protocols supported in the LOCAL domain.
298 static struct domain localdomain;
299 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
300 static struct pr_usrreqs uipc_usrreqs_seqpacket;
301 static struct protosw localsw[] = {
303 .pr_type = SOCK_STREAM,
304 .pr_domain = &localdomain,
305 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
306 .pr_ctloutput = &uipc_ctloutput,
307 .pr_usrreqs = &uipc_usrreqs_stream
310 .pr_type = SOCK_DGRAM,
311 .pr_domain = &localdomain,
312 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
313 .pr_ctloutput = &uipc_ctloutput,
314 .pr_usrreqs = &uipc_usrreqs_dgram
317 .pr_type = SOCK_SEQPACKET,
318 .pr_domain = &localdomain,
321 * XXXRW: For now, PR_ADDR because soreceive will bump into them
322 * due to our use of sbappendaddr. A new sbappend variants is needed
323 * that supports both atomic record writes and control data.
325 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
327 .pr_ctloutput = &uipc_ctloutput,
328 .pr_usrreqs = &uipc_usrreqs_seqpacket,
332 static struct domain localdomain = {
333 .dom_family = AF_LOCAL,
335 .dom_init = unp_init,
336 .dom_externalize = unp_externalize,
337 .dom_dispose = unp_dispose,
338 .dom_protosw = localsw,
339 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])]
344 uipc_abort(struct socket *so)
346 struct unpcb *unp, *unp2;
349 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
353 unp2 = unp->unp_conn;
356 unp_drop(unp2, ECONNABORTED);
357 UNP_PCB_UNLOCK(unp2);
364 uipc_accept(struct socket *so, struct sockaddr **nam)
366 struct unpcb *unp, *unp2;
367 const struct sockaddr *sa;
370 * Pass back name of connected socket, if it was bound and we are
371 * still connected (our peer may have closed already!).
374 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
376 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
378 unp2 = unp->unp_conn;
379 if (unp2 != NULL && unp2->unp_addr != NULL) {
381 sa = (struct sockaddr *) unp2->unp_addr;
382 bcopy(sa, *nam, sa->sa_len);
383 UNP_PCB_UNLOCK(unp2);
386 bcopy(sa, *nam, sa->sa_len);
393 uipc_attach(struct socket *so, int proto, struct thread *td)
395 u_long sendspace, recvspace;
399 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
400 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
401 switch (so->so_type) {
403 sendspace = unpst_sendspace;
404 recvspace = unpst_recvspace;
408 sendspace = unpdg_sendspace;
409 recvspace = unpdg_recvspace;
413 sendspace = unpsp_sendspace;
414 recvspace = unpsp_recvspace;
418 panic("uipc_attach");
420 error = soreserve(so, sendspace, recvspace);
424 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
427 LIST_INIT(&unp->unp_refs);
428 UNP_PCB_LOCK_INIT(unp);
429 unp->unp_socket = so;
431 unp->unp_refcount = 1;
432 if (so->so_head != NULL)
433 unp->unp_flags |= UNP_NASCENT;
436 unp->unp_gencnt = ++unp_gencnt;
438 switch (so->so_type) {
440 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
444 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
448 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
452 panic("uipc_attach");
460 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
462 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
472 if (nam->sa_family != AF_UNIX)
473 return (EAFNOSUPPORT);
476 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
478 if (soun->sun_len > sizeof(struct sockaddr_un))
480 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
485 * We don't allow simultaneous bind() calls on a single UNIX domain
486 * socket, so flag in-progress operations, and return an error if an
487 * operation is already in progress.
489 * Historically, we have not allowed a socket to be rebound, so this
490 * also returns an error. Not allowing re-binding simplifies the
491 * implementation and avoids a great many possible failure modes.
494 if (unp->unp_vnode != NULL) {
498 if (unp->unp_flags & UNP_BINDING) {
502 unp->unp_flags |= UNP_BINDING;
505 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
506 bcopy(soun->sun_path, buf, namelen);
510 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
511 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
512 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
517 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
518 NDFREE(&nd, NDF_ONLY_PNBUF);
528 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
534 vattr.va_type = VSOCK;
535 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
537 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
541 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
542 NDFREE(&nd, NDF_ONLY_PNBUF);
545 vn_finished_write(mp);
549 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
550 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
554 VOP_UNP_BIND(vp, unp->unp_socket);
556 unp->unp_addr = soun;
557 unp->unp_flags &= ~UNP_BINDING;
561 vn_finished_write(mp);
567 unp->unp_flags &= ~UNP_BINDING;
574 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
577 return (uipc_bindat(AT_FDCWD, so, nam, td));
581 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
585 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
587 error = unp_connect(so, nam, td);
593 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
598 KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
600 error = unp_connectat(fd, so, nam, td);
606 uipc_close(struct socket *so)
608 struct unpcb *unp, *unp2;
611 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
615 unp2 = unp->unp_conn;
618 unp_disconnect(unp, unp2);
619 UNP_PCB_UNLOCK(unp2);
626 uipc_connect2(struct socket *so1, struct socket *so2)
628 struct unpcb *unp, *unp2;
633 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
636 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
638 error = unp_connect2(so1, so2, PRU_CONNECT2);
639 UNP_PCB_UNLOCK(unp2);
646 uipc_detach(struct socket *so)
648 struct unpcb *unp, *unp2;
649 struct sockaddr_un *saved_unp_addr;
651 int freeunp, local_unp_rights;
654 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
657 local_unp_rights = 0;
660 LIST_REMOVE(unp, unp_link);
661 unp->unp_gencnt = ++unp_gencnt;
665 if ((unp->unp_flags & UNP_NASCENT) != 0) {
673 * XXXRW: Should assert vp->v_socket == so.
675 if ((vp = unp->unp_vnode) != NULL) {
677 unp->unp_vnode = NULL;
679 unp2 = unp->unp_conn;
682 unp_disconnect(unp, unp2);
683 UNP_PCB_UNLOCK(unp2);
687 * We hold the linkage lock exclusively, so it's OK to acquire
688 * multiple pcb locks at a time.
690 while (!LIST_EMPTY(&unp->unp_refs)) {
691 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
694 unp_drop(ref, ECONNRESET);
697 local_unp_rights = unp_rights;
700 unp->unp_socket->so_pcb = NULL;
701 saved_unp_addr = unp->unp_addr;
702 unp->unp_addr = NULL;
704 freeunp = (unp->unp_refcount == 0);
705 if (saved_unp_addr != NULL)
706 free(saved_unp_addr, M_SONAME);
708 UNP_PCB_LOCK_DESTROY(unp);
709 uma_zfree(unp_zone, unp);
714 if (local_unp_rights)
715 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
719 uipc_disconnect(struct socket *so)
721 struct unpcb *unp, *unp2;
724 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
728 unp2 = unp->unp_conn;
731 unp_disconnect(unp, unp2);
732 UNP_PCB_UNLOCK(unp2);
740 uipc_listen(struct socket *so, int backlog, struct thread *td)
746 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
749 if (unp->unp_vnode == NULL) {
755 error = solisten_proto_check(so);
757 cru2x(td->td_ucred, &unp->unp_peercred);
758 unp->unp_flags |= UNP_HAVEPCCACHED;
759 solisten_proto(so, backlog);
767 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
769 struct unpcb *unp, *unp2;
770 const struct sockaddr *sa;
773 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
775 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
778 * XXX: It seems that this test always fails even when connection is
779 * established. So, this else clause is added as workaround to
780 * return PF_LOCAL sockaddr.
782 unp2 = unp->unp_conn;
785 if (unp2->unp_addr != NULL)
786 sa = (struct sockaddr *) unp2->unp_addr;
789 bcopy(sa, *nam, sa->sa_len);
790 UNP_PCB_UNLOCK(unp2);
793 bcopy(sa, *nam, sa->sa_len);
800 uipc_rcvd(struct socket *so, int flags)
802 struct unpcb *unp, *unp2;
807 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
809 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
810 panic("uipc_rcvd socktype %d", so->so_type);
813 * Adjust backpressure on sender and wakeup any waiting to write.
815 * The unp lock is acquired to maintain the validity of the unp_conn
816 * pointer; no lock on unp2 is required as unp2->unp_socket will be
817 * static as long as we don't permit unp2 to disconnect from unp,
818 * which is prevented by the lock on unp. We cache values from
819 * so_rcv to avoid holding the so_rcv lock over the entire
820 * transaction on the remote so_snd.
822 SOCKBUF_LOCK(&so->so_rcv);
823 mbcnt = so->so_rcv.sb_mbcnt;
824 sbcc = so->so_rcv.sb_cc;
825 SOCKBUF_UNLOCK(&so->so_rcv);
827 * There is a benign race condition at this point. If we're planning to
828 * clear SB_STOP, but uipc_send is called on the connected socket at
829 * this instant, it might add data to the sockbuf and set SB_STOP. Then
830 * we would erroneously clear SB_STOP below, even though the sockbuf is
831 * full. The race is benign because the only ill effect is to allow the
832 * sockbuf to exceed its size limit, and the size limits are not
833 * strictly guaranteed anyway.
836 unp2 = unp->unp_conn;
841 so2 = unp2->unp_socket;
842 SOCKBUF_LOCK(&so2->so_snd);
843 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
844 so2->so_snd.sb_flags &= ~SB_STOP;
845 sowwakeup_locked(so2);
851 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
852 struct mbuf *control, struct thread *td)
854 struct unpcb *unp, *unp2;
860 KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
862 if (flags & PRUS_OOB) {
866 if (control != NULL && (error = unp_internalize(&control, td)))
868 if ((nam != NULL) || (flags & PRUS_EOF))
872 switch (so->so_type) {
875 const struct sockaddr *from;
877 unp2 = unp->unp_conn;
879 UNP_LINK_WLOCK_ASSERT();
884 error = unp_connect(so, nam, td);
887 unp2 = unp->unp_conn;
891 * Because connect() and send() are non-atomic in a sendto()
892 * with a target address, it's possible that the socket will
893 * have disconnected before the send() can run. In that case
894 * return the slightly counter-intuitive but otherwise
895 * correct error that the socket is not connected.
902 if (unp2->unp_flags & UNP_WANTCRED)
903 control = unp_addsockcred(td, control);
905 if (unp->unp_addr != NULL)
906 from = (struct sockaddr *)unp->unp_addr;
909 so2 = unp2->unp_socket;
910 SOCKBUF_LOCK(&so2->so_rcv);
911 if (sbappendaddr_locked(&so2->so_rcv, from, m,
913 sorwakeup_locked(so2);
917 SOCKBUF_UNLOCK(&so2->so_rcv);
921 UNP_LINK_WLOCK_ASSERT();
923 unp_disconnect(unp, unp2);
924 UNP_PCB_UNLOCK(unp2);
932 if ((so->so_state & SS_ISCONNECTED) == 0) {
934 UNP_LINK_WLOCK_ASSERT();
935 error = unp_connect(so, nam, td);
945 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
951 * Because connect() and send() are non-atomic in a sendto()
952 * with a target address, it's possible that the socket will
953 * have disconnected before the send() can run. In that case
954 * return the slightly counter-intuitive but otherwise
955 * correct error that the socket is not connected.
957 * Locking here must be done carefully: the linkage lock
958 * prevents interconnections between unpcbs from changing, so
959 * we can traverse from unp to unp2 without acquiring unp's
960 * lock. Socket buffer locks follow unpcb locks, so we can
961 * acquire both remote and lock socket buffer locks.
963 unp2 = unp->unp_conn;
968 so2 = unp2->unp_socket;
970 SOCKBUF_LOCK(&so2->so_rcv);
971 if (unp2->unp_flags & UNP_WANTCRED) {
973 * Credentials are passed only once on SOCK_STREAM
974 * and SOCK_SEQPACKET.
976 unp2->unp_flags &= ~UNP_WANTCRED;
977 control = unp_addsockcred(td, control);
980 * Send to paired receive port, and then reduce send buffer
981 * hiwater marks to maintain backpressure. Wake up readers.
983 switch (so->so_type) {
985 if (control != NULL) {
986 if (sbappendcontrol_locked(&so2->so_rcv, m,
990 sbappend_locked(&so2->so_rcv, m);
993 case SOCK_SEQPACKET: {
994 const struct sockaddr *from;
998 * Don't check for space available in so2->so_rcv.
999 * Unix domain sockets only check for space in the
1000 * sending sockbuf, and that check is performed one
1001 * level up the stack.
1003 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1010 mbcnt = so2->so_rcv.sb_mbcnt;
1011 sbcc = so2->so_rcv.sb_cc;
1012 sorwakeup_locked(so2);
1015 * The PCB lock on unp2 protects the SB_STOP flag. Without it,
1016 * it would be possible for uipc_rcvd to be called at this
1017 * point, drain the receiving sockbuf, clear SB_STOP, and then
1018 * we would set SB_STOP below. That could lead to an empty
1019 * sockbuf having SB_STOP set
1021 SOCKBUF_LOCK(&so->so_snd);
1022 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1023 so->so_snd.sb_flags |= SB_STOP;
1024 SOCKBUF_UNLOCK(&so->so_snd);
1025 UNP_PCB_UNLOCK(unp2);
1030 panic("uipc_send unknown socktype");
1034 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1036 if (flags & PRUS_EOF) {
1040 UNP_PCB_UNLOCK(unp);
1043 if ((nam != NULL) || (flags & PRUS_EOF))
1048 if (control != NULL && error != 0)
1049 unp_dispose(control);
1052 if (control != NULL)
1060 uipc_sense(struct socket *so, struct stat *sb)
1064 unp = sotounpcb(so);
1065 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1067 sb->st_blksize = so->so_snd.sb_hiwat;
1070 if (unp->unp_ino == 0)
1071 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1072 sb->st_ino = unp->unp_ino;
1073 UNP_PCB_UNLOCK(unp);
1078 uipc_shutdown(struct socket *so)
1082 unp = sotounpcb(so);
1083 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1089 UNP_PCB_UNLOCK(unp);
1095 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1098 const struct sockaddr *sa;
1100 unp = sotounpcb(so);
1101 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1103 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1105 if (unp->unp_addr != NULL)
1106 sa = (struct sockaddr *) unp->unp_addr;
1109 bcopy(sa, *nam, sa->sa_len);
1110 UNP_PCB_UNLOCK(unp);
1114 static struct pr_usrreqs uipc_usrreqs_dgram = {
1115 .pru_abort = uipc_abort,
1116 .pru_accept = uipc_accept,
1117 .pru_attach = uipc_attach,
1118 .pru_bind = uipc_bind,
1119 .pru_bindat = uipc_bindat,
1120 .pru_connect = uipc_connect,
1121 .pru_connectat = uipc_connectat,
1122 .pru_connect2 = uipc_connect2,
1123 .pru_detach = uipc_detach,
1124 .pru_disconnect = uipc_disconnect,
1125 .pru_listen = uipc_listen,
1126 .pru_peeraddr = uipc_peeraddr,
1127 .pru_rcvd = uipc_rcvd,
1128 .pru_send = uipc_send,
1129 .pru_sense = uipc_sense,
1130 .pru_shutdown = uipc_shutdown,
1131 .pru_sockaddr = uipc_sockaddr,
1132 .pru_soreceive = soreceive_dgram,
1133 .pru_close = uipc_close,
1136 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1137 .pru_abort = uipc_abort,
1138 .pru_accept = uipc_accept,
1139 .pru_attach = uipc_attach,
1140 .pru_bind = uipc_bind,
1141 .pru_bindat = uipc_bindat,
1142 .pru_connect = uipc_connect,
1143 .pru_connectat = uipc_connectat,
1144 .pru_connect2 = uipc_connect2,
1145 .pru_detach = uipc_detach,
1146 .pru_disconnect = uipc_disconnect,
1147 .pru_listen = uipc_listen,
1148 .pru_peeraddr = uipc_peeraddr,
1149 .pru_rcvd = uipc_rcvd,
1150 .pru_send = uipc_send,
1151 .pru_sense = uipc_sense,
1152 .pru_shutdown = uipc_shutdown,
1153 .pru_sockaddr = uipc_sockaddr,
1154 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1155 .pru_close = uipc_close,
1158 static struct pr_usrreqs uipc_usrreqs_stream = {
1159 .pru_abort = uipc_abort,
1160 .pru_accept = uipc_accept,
1161 .pru_attach = uipc_attach,
1162 .pru_bind = uipc_bind,
1163 .pru_bindat = uipc_bindat,
1164 .pru_connect = uipc_connect,
1165 .pru_connectat = uipc_connectat,
1166 .pru_connect2 = uipc_connect2,
1167 .pru_detach = uipc_detach,
1168 .pru_disconnect = uipc_disconnect,
1169 .pru_listen = uipc_listen,
1170 .pru_peeraddr = uipc_peeraddr,
1171 .pru_rcvd = uipc_rcvd,
1172 .pru_send = uipc_send,
1173 .pru_sense = uipc_sense,
1174 .pru_shutdown = uipc_shutdown,
1175 .pru_sockaddr = uipc_sockaddr,
1176 .pru_soreceive = soreceive_generic,
1177 .pru_close = uipc_close,
1181 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1187 if (sopt->sopt_level != 0)
1190 unp = sotounpcb(so);
1191 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1193 switch (sopt->sopt_dir) {
1195 switch (sopt->sopt_name) {
1196 case LOCAL_PEERCRED:
1198 if (unp->unp_flags & UNP_HAVEPC)
1199 xu = unp->unp_peercred;
1201 if (so->so_type == SOCK_STREAM)
1206 UNP_PCB_UNLOCK(unp);
1208 error = sooptcopyout(sopt, &xu, sizeof(xu));
1212 /* Unlocked read. */
1213 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1214 error = sooptcopyout(sopt, &optval, sizeof(optval));
1217 case LOCAL_CONNWAIT:
1218 /* Unlocked read. */
1219 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1220 error = sooptcopyout(sopt, &optval, sizeof(optval));
1230 switch (sopt->sopt_name) {
1232 case LOCAL_CONNWAIT:
1233 error = sooptcopyin(sopt, &optval, sizeof(optval),
1238 #define OPTSET(bit) do { \
1239 UNP_PCB_LOCK(unp); \
1241 unp->unp_flags |= bit; \
1243 unp->unp_flags &= ~bit; \
1244 UNP_PCB_UNLOCK(unp); \
1247 switch (sopt->sopt_name) {
1249 OPTSET(UNP_WANTCRED);
1252 case LOCAL_CONNWAIT:
1253 OPTSET(UNP_CONNWAIT);
1262 error = ENOPROTOOPT;
1275 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1278 return (unp_connectat(AT_FDCWD, so, nam, td));
1282 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1285 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1287 struct socket *so2, *so3;
1288 struct unpcb *unp, *unp2, *unp3;
1289 struct nameidata nd;
1290 char buf[SOCK_MAXADDRLEN];
1291 struct sockaddr *sa;
1292 cap_rights_t rights;
1295 if (nam->sa_family != AF_UNIX)
1296 return (EAFNOSUPPORT);
1298 UNP_LINK_WLOCK_ASSERT();
1300 unp = sotounpcb(so);
1301 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1303 if (nam->sa_len > sizeof(struct sockaddr_un))
1305 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1308 bcopy(soun->sun_path, buf, len);
1312 if (unp->unp_flags & UNP_CONNECTING) {
1313 UNP_PCB_UNLOCK(unp);
1317 unp->unp_flags |= UNP_CONNECTING;
1318 UNP_PCB_UNLOCK(unp);
1320 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1321 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1322 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1328 ASSERT_VOP_LOCKED(vp, "unp_connect");
1329 NDFREE(&nd, NDF_ONLY_PNBUF);
1333 if (vp->v_type != VSOCK) {
1338 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1342 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1346 unp = sotounpcb(so);
1347 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1350 * Lock linkage lock for two reasons: make sure v_socket is stable,
1351 * and to protect simultaneous locking of multiple pcbs.
1354 VOP_UNP_CONNECT(vp, &so2);
1356 error = ECONNREFUSED;
1359 if (so->so_type != so2->so_type) {
1363 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1364 if (so2->so_options & SO_ACCEPTCONN) {
1365 CURVNET_SET(so2->so_vnet);
1366 so3 = sonewconn(so2, 0);
1371 error = ECONNREFUSED;
1374 unp = sotounpcb(so);
1375 unp2 = sotounpcb(so2);
1376 unp3 = sotounpcb(so3);
1380 if (unp2->unp_addr != NULL) {
1381 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1382 unp3->unp_addr = (struct sockaddr_un *) sa;
1387 * The connector's (client's) credentials are copied from its
1388 * process structure at the time of connect() (which is now).
1390 cru2x(td->td_ucred, &unp3->unp_peercred);
1391 unp3->unp_flags |= UNP_HAVEPC;
1394 * The receiver's (server's) credentials are copied from the
1395 * unp_peercred member of socket on which the former called
1396 * listen(); uipc_listen() cached that process's credentials
1397 * at that time so we can use them now.
1399 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1400 ("unp_connect: listener without cached peercred"));
1401 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1402 sizeof(unp->unp_peercred));
1403 unp->unp_flags |= UNP_HAVEPC;
1404 if (unp2->unp_flags & UNP_WANTCRED)
1405 unp3->unp_flags |= UNP_WANTCRED;
1406 UNP_PCB_UNLOCK(unp3);
1407 UNP_PCB_UNLOCK(unp2);
1408 UNP_PCB_UNLOCK(unp);
1410 mac_socketpeer_set_from_socket(so, so3);
1411 mac_socketpeer_set_from_socket(so3, so);
1416 unp = sotounpcb(so);
1417 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1418 unp2 = sotounpcb(so2);
1419 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1422 error = unp_connect2(so, so2, PRU_CONNECT);
1423 UNP_PCB_UNLOCK(unp2);
1424 UNP_PCB_UNLOCK(unp);
1433 unp->unp_flags &= ~UNP_CONNECTING;
1434 UNP_PCB_UNLOCK(unp);
1439 unp_connect2(struct socket *so, struct socket *so2, int req)
1444 unp = sotounpcb(so);
1445 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1446 unp2 = sotounpcb(so2);
1447 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1449 UNP_LINK_WLOCK_ASSERT();
1450 UNP_PCB_LOCK_ASSERT(unp);
1451 UNP_PCB_LOCK_ASSERT(unp2);
1453 if (so2->so_type != so->so_type)
1454 return (EPROTOTYPE);
1455 unp2->unp_flags &= ~UNP_NASCENT;
1456 unp->unp_conn = unp2;
1458 switch (so->so_type) {
1460 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1465 case SOCK_SEQPACKET:
1466 unp2->unp_conn = unp;
1467 if (req == PRU_CONNECT &&
1468 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1476 panic("unp_connect2");
1482 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1486 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1488 UNP_LINK_WLOCK_ASSERT();
1489 UNP_PCB_LOCK_ASSERT(unp);
1490 UNP_PCB_LOCK_ASSERT(unp2);
1492 unp->unp_conn = NULL;
1493 switch (unp->unp_socket->so_type) {
1495 LIST_REMOVE(unp, unp_reflink);
1496 so = unp->unp_socket;
1498 so->so_state &= ~SS_ISCONNECTED;
1503 case SOCK_SEQPACKET:
1504 soisdisconnected(unp->unp_socket);
1505 unp2->unp_conn = NULL;
1506 soisdisconnected(unp2->unp_socket);
1512 * unp_pcblist() walks the global list of struct unpcb's to generate a
1513 * pointer list, bumping the refcount on each unpcb. It then copies them out
1514 * sequentially, validating the generation number on each to see if it has
1515 * been detached. All of this is necessary because copyout() may sleep on
1519 unp_pcblist(SYSCTL_HANDLER_ARGS)
1523 struct unpcb *unp, **unp_list;
1525 struct xunpgen *xug;
1526 struct unp_head *head;
1529 switch ((intptr_t)arg1) {
1538 case SOCK_SEQPACKET:
1543 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1547 * The process of preparing the PCB list is too time-consuming and
1548 * resource-intensive to repeat twice on every request.
1550 if (req->oldptr == NULL) {
1552 req->oldidx = 2 * (sizeof *xug)
1553 + (n + n/8) * sizeof(struct xunpcb);
1557 if (req->newptr != NULL)
1561 * OK, now we're committed to doing something.
1563 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1565 gencnt = unp_gencnt;
1569 xug->xug_len = sizeof *xug;
1571 xug->xug_gen = gencnt;
1572 xug->xug_sogen = so_gencnt;
1573 error = SYSCTL_OUT(req, xug, sizeof *xug);
1579 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1582 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1583 unp = LIST_NEXT(unp, unp_link)) {
1585 if (unp->unp_gencnt <= gencnt) {
1586 if (cr_cansee(req->td->td_ucred,
1587 unp->unp_socket->so_cred)) {
1588 UNP_PCB_UNLOCK(unp);
1591 unp_list[i++] = unp;
1592 unp->unp_refcount++;
1594 UNP_PCB_UNLOCK(unp);
1597 n = i; /* In case we lost some during malloc. */
1600 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1601 for (i = 0; i < n; i++) {
1604 unp->unp_refcount--;
1605 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1606 xu->xu_len = sizeof *xu;
1609 * XXX - need more locking here to protect against
1610 * connect/disconnect races for SMP.
1612 if (unp->unp_addr != NULL)
1613 bcopy(unp->unp_addr, &xu->xu_addr,
1614 unp->unp_addr->sun_len);
1615 if (unp->unp_conn != NULL &&
1616 unp->unp_conn->unp_addr != NULL)
1617 bcopy(unp->unp_conn->unp_addr,
1619 unp->unp_conn->unp_addr->sun_len);
1620 bcopy(unp, &xu->xu_unp, sizeof *unp);
1621 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1622 UNP_PCB_UNLOCK(unp);
1623 error = SYSCTL_OUT(req, xu, sizeof *xu);
1625 freeunp = (unp->unp_refcount == 0);
1626 UNP_PCB_UNLOCK(unp);
1628 UNP_PCB_LOCK_DESTROY(unp);
1629 uma_zfree(unp_zone, unp);
1636 * Give the user an updated idea of our state. If the
1637 * generation differs from what we told her before, she knows
1638 * that something happened while we were processing this
1639 * request, and it might be necessary to retry.
1641 xug->xug_gen = unp_gencnt;
1642 xug->xug_sogen = so_gencnt;
1643 xug->xug_count = unp_count;
1644 error = SYSCTL_OUT(req, xug, sizeof *xug);
1646 free(unp_list, M_TEMP);
1651 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1652 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1653 "List of active local datagram sockets");
1654 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1655 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1656 "List of active local stream sockets");
1657 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1658 CTLTYPE_OPAQUE | CTLFLAG_RD,
1659 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1660 "List of active local seqpacket sockets");
1663 unp_shutdown(struct unpcb *unp)
1668 UNP_LINK_WLOCK_ASSERT();
1669 UNP_PCB_LOCK_ASSERT(unp);
1671 unp2 = unp->unp_conn;
1672 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1673 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1674 so = unp2->unp_socket;
1681 unp_drop(struct unpcb *unp, int errno)
1683 struct socket *so = unp->unp_socket;
1686 UNP_LINK_WLOCK_ASSERT();
1687 UNP_PCB_LOCK_ASSERT(unp);
1689 so->so_error = errno;
1690 unp2 = unp->unp_conn;
1694 unp_disconnect(unp, unp2);
1695 UNP_PCB_UNLOCK(unp2);
1699 unp_freerights(struct filedescent **fdep, int fdcount)
1704 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1706 for (i = 0; i < fdcount; i++) {
1707 fp = fdep[i]->fde_file;
1708 filecaps_free(&fdep[i]->fde_caps);
1711 free(fdep[0], M_FILECAPS);
1715 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1717 struct thread *td = curthread; /* XXX */
1718 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1721 struct filedesc *fdesc = td->td_proc->p_fd;
1722 struct filedescent *fde, **fdep;
1724 socklen_t clen = control->m_len, datalen;
1728 UNP_LINK_UNLOCK_ASSERT();
1731 if (controlp != NULL) /* controlp == NULL => free control messages */
1733 while (cm != NULL) {
1734 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1738 data = CMSG_DATA(cm);
1739 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1740 if (cm->cmsg_level == SOL_SOCKET
1741 && cm->cmsg_type == SCM_RIGHTS) {
1742 newfds = datalen / sizeof(*fdep);
1747 /* If we're not outputting the descriptors free them. */
1748 if (error || controlp == NULL) {
1749 unp_freerights(fdep, newfds);
1752 FILEDESC_XLOCK(fdesc);
1755 * Now change each pointer to an fd in the global
1756 * table to an integer that is the index to the local
1757 * fd table entry that we set up to point to the
1758 * global one we are transferring.
1760 newlen = newfds * sizeof(int);
1761 *controlp = sbcreatecontrol(NULL, newlen,
1762 SCM_RIGHTS, SOL_SOCKET);
1763 if (*controlp == NULL) {
1764 FILEDESC_XUNLOCK(fdesc);
1766 unp_freerights(fdep, newfds);
1771 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1772 if (fdallocn(td, 0, fdp, newfds) != 0) {
1773 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1775 unp_freerights(fdep, newfds);
1780 for (i = 0; i < newfds; i++, fdp++) {
1781 fde = &fdesc->fd_ofiles[*fdp];
1782 fde->fde_file = fdep[i]->fde_file;
1783 filecaps_move(&fdep[i]->fde_caps,
1785 if ((flags & MSG_CMSG_CLOEXEC) != 0)
1786 fde->fde_flags |= UF_EXCLOSE;
1787 unp_externalize_fp(fde->fde_file);
1789 FILEDESC_XUNLOCK(fdesc);
1790 free(fdep[0], M_FILECAPS);
1792 /* We can just copy anything else across. */
1793 if (error || controlp == NULL)
1795 *controlp = sbcreatecontrol(NULL, datalen,
1796 cm->cmsg_type, cm->cmsg_level);
1797 if (*controlp == NULL) {
1802 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1805 controlp = &(*controlp)->m_next;
1808 if (CMSG_SPACE(datalen) < clen) {
1809 clen -= CMSG_SPACE(datalen);
1810 cm = (struct cmsghdr *)
1811 ((caddr_t)cm + CMSG_SPACE(datalen));
1823 unp_zone_change(void *tag)
1826 uma_zone_set_max(unp_zone, maxsockets);
1834 if (!IS_DEFAULT_VNET(curvnet))
1837 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1838 NULL, NULL, UMA_ALIGN_PTR, 0);
1839 if (unp_zone == NULL)
1841 uma_zone_set_max(unp_zone, maxsockets);
1842 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1843 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1844 NULL, EVENTHANDLER_PRI_ANY);
1845 LIST_INIT(&unp_dhead);
1846 LIST_INIT(&unp_shead);
1847 LIST_INIT(&unp_sphead);
1848 SLIST_INIT(&unp_defers);
1849 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1850 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1851 UNP_LINK_LOCK_INIT();
1852 UNP_LIST_LOCK_INIT();
1853 UNP_DEFERRED_LOCK_INIT();
1857 unp_internalize(struct mbuf **controlp, struct thread *td)
1859 struct mbuf *control = *controlp;
1860 struct proc *p = td->td_proc;
1861 struct filedesc *fdesc = p->p_fd;
1863 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1864 struct cmsgcred *cmcred;
1865 struct filedescent *fde, **fdep, *fdev;
1870 socklen_t clen = control->m_len, datalen;
1874 UNP_LINK_UNLOCK_ASSERT();
1878 while (cm != NULL) {
1879 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1880 || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
1884 data = CMSG_DATA(cm);
1885 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1887 switch (cm->cmsg_type) {
1889 * Fill in credential information.
1892 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1893 SCM_CREDS, SOL_SOCKET);
1894 if (*controlp == NULL) {
1898 cmcred = (struct cmsgcred *)
1899 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1900 cmcred->cmcred_pid = p->p_pid;
1901 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1902 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1903 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1904 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1906 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1907 cmcred->cmcred_groups[i] =
1908 td->td_ucred->cr_groups[i];
1912 oldfds = datalen / sizeof (int);
1916 * Check that all the FDs passed in refer to legal
1917 * files. If not, reject the entire operation.
1920 FILEDESC_SLOCK(fdesc);
1921 for (i = 0; i < oldfds; i++, fdp++) {
1922 fp = fget_locked(fdesc, *fdp);
1924 FILEDESC_SUNLOCK(fdesc);
1928 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1929 FILEDESC_SUNLOCK(fdesc);
1937 * Now replace the integer FDs with pointers to the
1938 * file structure and capability rights.
1940 newlen = oldfds * sizeof(fdep[0]);
1941 *controlp = sbcreatecontrol(NULL, newlen,
1942 SCM_RIGHTS, SOL_SOCKET);
1943 if (*controlp == NULL) {
1944 FILEDESC_SUNLOCK(fdesc);
1949 fdep = (struct filedescent **)
1950 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1951 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1953 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1954 fde = &fdesc->fd_ofiles[*fdp];
1956 fdep[i]->fde_file = fde->fde_file;
1957 filecaps_copy(&fde->fde_caps,
1958 &fdep[i]->fde_caps);
1959 unp_internalize_fp(fdep[i]->fde_file);
1961 FILEDESC_SUNLOCK(fdesc);
1965 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1966 SCM_TIMESTAMP, SOL_SOCKET);
1967 if (*controlp == NULL) {
1971 tv = (struct timeval *)
1972 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1977 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
1978 SCM_BINTIME, SOL_SOCKET);
1979 if (*controlp == NULL) {
1983 bt = (struct bintime *)
1984 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1993 controlp = &(*controlp)->m_next;
1994 if (CMSG_SPACE(datalen) < clen) {
1995 clen -= CMSG_SPACE(datalen);
1996 cm = (struct cmsghdr *)
1997 ((caddr_t)cm + CMSG_SPACE(datalen));
2009 static struct mbuf *
2010 unp_addsockcred(struct thread *td, struct mbuf *control)
2012 struct mbuf *m, *n, *n_prev;
2013 struct sockcred *sc;
2014 const struct cmsghdr *cm;
2018 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2019 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2023 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2024 sc->sc_uid = td->td_ucred->cr_ruid;
2025 sc->sc_euid = td->td_ucred->cr_uid;
2026 sc->sc_gid = td->td_ucred->cr_rgid;
2027 sc->sc_egid = td->td_ucred->cr_gid;
2028 sc->sc_ngroups = ngroups;
2029 for (i = 0; i < sc->sc_ngroups; i++)
2030 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2033 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2034 * created SCM_CREDS control message (struct sockcred) has another
2037 if (control != NULL)
2038 for (n = control, n_prev = NULL; n != NULL;) {
2039 cm = mtod(n, struct cmsghdr *);
2040 if (cm->cmsg_level == SOL_SOCKET &&
2041 cm->cmsg_type == SCM_CREDS) {
2043 control = n->m_next;
2045 n_prev->m_next = n->m_next;
2053 /* Prepend it to the head. */
2054 m->m_next = control;
2058 static struct unpcb *
2059 fptounp(struct file *fp)
2063 if (fp->f_type != DTYPE_SOCKET)
2065 if ((so = fp->f_data) == NULL)
2067 if (so->so_proto->pr_domain != &localdomain)
2069 return sotounpcb(so);
2073 unp_discard(struct file *fp)
2075 struct unp_defer *dr;
2077 if (unp_externalize_fp(fp)) {
2078 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2080 UNP_DEFERRED_LOCK();
2081 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2082 UNP_DEFERRED_UNLOCK();
2083 atomic_add_int(&unp_defers_count, 1);
2084 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2086 (void) closef(fp, (struct thread *)NULL);
2090 unp_process_defers(void *arg __unused, int pending)
2092 struct unp_defer *dr;
2093 SLIST_HEAD(, unp_defer) drl;
2098 UNP_DEFERRED_LOCK();
2099 if (SLIST_FIRST(&unp_defers) == NULL) {
2100 UNP_DEFERRED_UNLOCK();
2103 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2104 UNP_DEFERRED_UNLOCK();
2106 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2107 SLIST_REMOVE_HEAD(&drl, ud_link);
2108 closef(dr->ud_fp, NULL);
2112 atomic_add_int(&unp_defers_count, -count);
2117 unp_internalize_fp(struct file *fp)
2122 if ((unp = fptounp(fp)) != NULL) {
2124 unp->unp_msgcount++;
2132 unp_externalize_fp(struct file *fp)
2138 if ((unp = fptounp(fp)) != NULL) {
2139 unp->unp_msgcount--;
2149 * unp_defer indicates whether additional work has been defered for a future
2150 * pass through unp_gc(). It is thread local and does not require explicit
2153 static int unp_marked;
2154 static int unp_unreachable;
2157 unp_accessable(struct filedescent **fdep, int fdcount)
2163 for (i = 0; i < fdcount; i++) {
2164 fp = fdep[i]->fde_file;
2165 if ((unp = fptounp(fp)) == NULL)
2167 if (unp->unp_gcflag & UNPGC_REF)
2169 unp->unp_gcflag &= ~UNPGC_DEAD;
2170 unp->unp_gcflag |= UNPGC_REF;
2176 unp_gc_process(struct unpcb *unp)
2182 /* Already processed. */
2183 if (unp->unp_gcflag & UNPGC_SCANNED)
2188 * Check for a socket potentially in a cycle. It must be in a
2189 * queue as indicated by msgcount, and this must equal the file
2190 * reference count. Note that when msgcount is 0 the file is NULL.
2192 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2193 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2194 unp->unp_gcflag |= UNPGC_DEAD;
2200 * Mark all sockets we reference with RIGHTS.
2202 so = unp->unp_socket;
2203 if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2204 SOCKBUF_LOCK(&so->so_rcv);
2205 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2206 SOCKBUF_UNLOCK(&so->so_rcv);
2210 * Mark all sockets in our accept queue.
2213 TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2214 if ((sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS) != 0)
2216 SOCKBUF_LOCK(&soa->so_rcv);
2217 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2218 SOCKBUF_UNLOCK(&soa->so_rcv);
2221 unp->unp_gcflag |= UNPGC_SCANNED;
2224 static int unp_recycled;
2225 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2226 "Number of unreachable sockets claimed by the garbage collector.");
2228 static int unp_taskcount;
2229 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2230 "Number of times the garbage collector has run.");
2233 unp_gc(__unused void *arg, int pending)
2235 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2237 struct unp_head **head;
2238 struct file *f, **unref;
2245 * First clear all gc flags from previous runs, apart from
2246 * UNPGC_IGNORE_RIGHTS.
2248 for (head = heads; *head != NULL; head++)
2249 LIST_FOREACH(unp, *head, unp_link)
2251 (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
2254 * Scan marking all reachable sockets with UNPGC_REF. Once a socket
2255 * is reachable all of the sockets it references are reachable.
2256 * Stop the scan once we do a complete loop without discovering
2257 * a new reachable socket.
2260 unp_unreachable = 0;
2262 for (head = heads; *head != NULL; head++)
2263 LIST_FOREACH(unp, *head, unp_link)
2264 unp_gc_process(unp);
2265 } while (unp_marked);
2267 if (unp_unreachable == 0)
2271 * Allocate space for a local list of dead unpcbs.
2273 unref = malloc(unp_unreachable * sizeof(struct file *),
2277 * Iterate looking for sockets which have been specifically marked
2278 * as as unreachable and store them locally.
2282 for (total = 0, head = heads; *head != NULL; head++)
2283 LIST_FOREACH(unp, *head, unp_link)
2284 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2286 if (unp->unp_msgcount == 0 || f == NULL ||
2287 f->f_count != unp->unp_msgcount)
2291 KASSERT(total <= unp_unreachable,
2292 ("unp_gc: incorrect unreachable count."));
2298 * Now flush all sockets, free'ing rights. This will free the
2299 * struct files associated with these sockets but leave each socket
2300 * with one remaining ref.
2302 for (i = 0; i < total; i++) {
2305 so = unref[i]->f_data;
2306 CURVNET_SET(so->so_vnet);
2312 * And finally release the sockets so they can be reclaimed.
2314 for (i = 0; i < total; i++)
2315 fdrop(unref[i], NULL);
2316 unp_recycled += total;
2317 free(unref, M_TEMP);
2321 unp_dispose(struct mbuf *m)
2325 unp_scan(m, unp_freerights);
2329 * Synchronize against unp_gc, which can trip over data as we are freeing it.
2332 unp_dispose_so(struct socket *so)
2336 unp = sotounpcb(so);
2338 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2340 unp_dispose(so->so_rcv.sb_mb);
2344 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2349 socklen_t clen, datalen;
2351 while (m0 != NULL) {
2352 for (m = m0; m; m = m->m_next) {
2353 if (m->m_type != MT_CONTROL)
2356 cm = mtod(m, struct cmsghdr *);
2359 while (cm != NULL) {
2360 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2363 data = CMSG_DATA(cm);
2364 datalen = (caddr_t)cm + cm->cmsg_len
2367 if (cm->cmsg_level == SOL_SOCKET &&
2368 cm->cmsg_type == SCM_RIGHTS) {
2369 (*op)(data, datalen /
2370 sizeof(struct filedescent *));
2373 if (CMSG_SPACE(datalen) < clen) {
2374 clen -= CMSG_SPACE(datalen);
2375 cm = (struct cmsghdr *)
2376 ((caddr_t)cm + CMSG_SPACE(datalen));
2388 * A helper function called by VFS before socket-type vnode reclamation.
2389 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2393 vfs_unp_reclaim(struct vnode *vp)
2399 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2400 KASSERT(vp->v_type == VSOCK,
2401 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2405 VOP_UNP_CONNECT(vp, &so);
2408 unp = sotounpcb(so);
2412 if (unp->unp_vnode == vp) {
2414 unp->unp_vnode = NULL;
2417 UNP_PCB_UNLOCK(unp);
2426 db_print_indent(int indent)
2430 for (i = 0; i < indent; i++)
2435 db_print_unpflags(int unp_flags)
2440 if (unp_flags & UNP_HAVEPC) {
2441 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2444 if (unp_flags & UNP_HAVEPCCACHED) {
2445 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2448 if (unp_flags & UNP_WANTCRED) {
2449 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2452 if (unp_flags & UNP_CONNWAIT) {
2453 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2456 if (unp_flags & UNP_CONNECTING) {
2457 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2460 if (unp_flags & UNP_BINDING) {
2461 db_printf("%sUNP_BINDING", comma ? ", " : "");
2467 db_print_xucred(int indent, struct xucred *xu)
2471 db_print_indent(indent);
2472 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
2473 xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2474 db_print_indent(indent);
2475 db_printf("cr_groups: ");
2477 for (i = 0; i < xu->cr_ngroups; i++) {
2478 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2485 db_print_unprefs(int indent, struct unp_head *uh)
2491 LIST_FOREACH(unp, uh, unp_reflink) {
2492 if (counter % 4 == 0)
2493 db_print_indent(indent);
2494 db_printf("%p ", unp);
2495 if (counter % 4 == 3)
2499 if (counter != 0 && counter % 4 != 0)
2503 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2508 db_printf("usage: show unpcb <addr>\n");
2511 unp = (struct unpcb *)addr;
2513 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2516 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2519 db_printf("unp_refs:\n");
2520 db_print_unprefs(2, &unp->unp_refs);
2522 /* XXXRW: Would be nice to print the full address, if any. */
2523 db_printf("unp_addr: %p\n", unp->unp_addr);
2525 db_printf("unp_gencnt: %llu\n",
2526 (unsigned long long)unp->unp_gencnt);
2528 db_printf("unp_flags: %x (", unp->unp_flags);
2529 db_print_unpflags(unp->unp_flags);
2532 db_printf("unp_peercred:\n");
2533 db_print_xucred(2, &unp->unp_peercred);
2535 db_printf("unp_refcount: %u\n", unp->unp_refcount);