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;
434 unp->unp_gencnt = ++unp_gencnt;
436 switch (so->so_type) {
438 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
442 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
446 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
450 panic("uipc_attach");
458 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
460 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
470 if (nam->sa_family != AF_UNIX)
471 return (EAFNOSUPPORT);
474 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
476 if (soun->sun_len > sizeof(struct sockaddr_un))
478 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
483 * We don't allow simultaneous bind() calls on a single UNIX domain
484 * socket, so flag in-progress operations, and return an error if an
485 * operation is already in progress.
487 * Historically, we have not allowed a socket to be rebound, so this
488 * also returns an error. Not allowing re-binding simplifies the
489 * implementation and avoids a great many possible failure modes.
492 if (unp->unp_vnode != NULL) {
496 if (unp->unp_flags & UNP_BINDING) {
500 unp->unp_flags |= UNP_BINDING;
503 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
504 bcopy(soun->sun_path, buf, namelen);
508 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
509 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
510 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
515 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
516 NDFREE(&nd, NDF_ONLY_PNBUF);
526 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
532 vattr.va_type = VSOCK;
533 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
535 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
539 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
540 NDFREE(&nd, NDF_ONLY_PNBUF);
543 vn_finished_write(mp);
547 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
548 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
552 VOP_UNP_BIND(vp, unp->unp_socket);
554 unp->unp_addr = soun;
555 unp->unp_flags &= ~UNP_BINDING;
559 vn_finished_write(mp);
565 unp->unp_flags &= ~UNP_BINDING;
572 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
575 return (uipc_bindat(AT_FDCWD, so, nam, td));
579 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
583 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
585 error = unp_connect(so, nam, td);
591 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
596 KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
598 error = unp_connectat(fd, so, nam, td);
604 uipc_close(struct socket *so)
606 struct unpcb *unp, *unp2;
609 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
613 unp2 = unp->unp_conn;
616 unp_disconnect(unp, unp2);
617 UNP_PCB_UNLOCK(unp2);
624 uipc_connect2(struct socket *so1, struct socket *so2)
626 struct unpcb *unp, *unp2;
631 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
634 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
636 error = unp_connect2(so1, so2, PRU_CONNECT2);
637 UNP_PCB_UNLOCK(unp2);
644 uipc_detach(struct socket *so)
646 struct unpcb *unp, *unp2;
647 struct sockaddr_un *saved_unp_addr;
649 int freeunp, local_unp_rights;
652 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
657 LIST_REMOVE(unp, unp_link);
658 unp->unp_gencnt = ++unp_gencnt;
663 * XXXRW: Should assert vp->v_socket == so.
665 if ((vp = unp->unp_vnode) != NULL) {
667 unp->unp_vnode = NULL;
669 unp2 = unp->unp_conn;
672 unp_disconnect(unp, unp2);
673 UNP_PCB_UNLOCK(unp2);
677 * We hold the linkage lock exclusively, so it's OK to acquire
678 * multiple pcb locks at a time.
680 while (!LIST_EMPTY(&unp->unp_refs)) {
681 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
684 unp_drop(ref, ECONNRESET);
687 local_unp_rights = unp_rights;
689 unp->unp_socket->so_pcb = NULL;
690 saved_unp_addr = unp->unp_addr;
691 unp->unp_addr = NULL;
693 freeunp = (unp->unp_refcount == 0);
694 if (saved_unp_addr != NULL)
695 free(saved_unp_addr, M_SONAME);
697 UNP_PCB_LOCK_DESTROY(unp);
698 uma_zfree(unp_zone, unp);
703 if (local_unp_rights)
704 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
708 uipc_disconnect(struct socket *so)
710 struct unpcb *unp, *unp2;
713 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
717 unp2 = unp->unp_conn;
720 unp_disconnect(unp, unp2);
721 UNP_PCB_UNLOCK(unp2);
729 uipc_listen(struct socket *so, int backlog, struct thread *td)
735 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
738 if (unp->unp_vnode == NULL) {
739 /* Already connected or not bound to an address. */
740 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
746 error = solisten_proto_check(so);
748 cru2x(td->td_ucred, &unp->unp_peercred);
749 unp->unp_flags |= UNP_HAVEPCCACHED;
750 solisten_proto(so, backlog);
758 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
760 struct unpcb *unp, *unp2;
761 const struct sockaddr *sa;
764 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
766 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
769 * XXX: It seems that this test always fails even when connection is
770 * established. So, this else clause is added as workaround to
771 * return PF_LOCAL sockaddr.
773 unp2 = unp->unp_conn;
776 if (unp2->unp_addr != NULL)
777 sa = (struct sockaddr *) unp2->unp_addr;
780 bcopy(sa, *nam, sa->sa_len);
781 UNP_PCB_UNLOCK(unp2);
784 bcopy(sa, *nam, sa->sa_len);
791 uipc_rcvd(struct socket *so, int flags)
793 struct unpcb *unp, *unp2;
798 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
799 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
800 ("%s: socktype %d", __func__, so->so_type));
803 * Adjust backpressure on sender and wakeup any waiting to write.
805 * The unp lock is acquired to maintain the validity of the unp_conn
806 * pointer; no lock on unp2 is required as unp2->unp_socket will be
807 * static as long as we don't permit unp2 to disconnect from unp,
808 * which is prevented by the lock on unp. We cache values from
809 * so_rcv to avoid holding the so_rcv lock over the entire
810 * transaction on the remote so_snd.
812 SOCKBUF_LOCK(&so->so_rcv);
813 mbcnt = so->so_rcv.sb_mbcnt;
814 sbcc = sbavail(&so->so_rcv);
815 SOCKBUF_UNLOCK(&so->so_rcv);
817 * There is a benign race condition at this point. If we're planning to
818 * clear SB_STOP, but uipc_send is called on the connected socket at
819 * this instant, it might add data to the sockbuf and set SB_STOP. Then
820 * we would erroneously clear SB_STOP below, even though the sockbuf is
821 * full. The race is benign because the only ill effect is to allow the
822 * sockbuf to exceed its size limit, and the size limits are not
823 * strictly guaranteed anyway.
826 unp2 = unp->unp_conn;
831 so2 = unp2->unp_socket;
832 SOCKBUF_LOCK(&so2->so_snd);
833 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
834 so2->so_snd.sb_flags &= ~SB_STOP;
835 sowwakeup_locked(so2);
841 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
842 struct mbuf *control, struct thread *td)
844 struct unpcb *unp, *unp2;
850 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
851 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
852 so->so_type == SOCK_SEQPACKET,
853 ("%s: socktype %d", __func__, so->so_type));
855 if (flags & PRUS_OOB) {
859 if (control != NULL && (error = unp_internalize(&control, td)))
861 if ((nam != NULL) || (flags & PRUS_EOF))
865 switch (so->so_type) {
868 const struct sockaddr *from;
870 unp2 = unp->unp_conn;
872 UNP_LINK_WLOCK_ASSERT();
877 error = unp_connect(so, nam, td);
880 unp2 = unp->unp_conn;
884 * Because connect() and send() are non-atomic in a sendto()
885 * with a target address, it's possible that the socket will
886 * have disconnected before the send() can run. In that case
887 * return the slightly counter-intuitive but otherwise
888 * correct error that the socket is not connected.
895 if (unp2->unp_flags & UNP_WANTCRED)
896 control = unp_addsockcred(td, control);
898 if (unp->unp_addr != NULL)
899 from = (struct sockaddr *)unp->unp_addr;
902 so2 = unp2->unp_socket;
903 SOCKBUF_LOCK(&so2->so_rcv);
904 if (sbappendaddr_locked(&so2->so_rcv, from, m,
906 sorwakeup_locked(so2);
910 SOCKBUF_UNLOCK(&so2->so_rcv);
914 UNP_LINK_WLOCK_ASSERT();
916 unp_disconnect(unp, unp2);
917 UNP_PCB_UNLOCK(unp2);
925 if ((so->so_state & SS_ISCONNECTED) == 0) {
927 UNP_LINK_WLOCK_ASSERT();
928 error = unp_connect(so, nam, td);
938 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
944 * Because connect() and send() are non-atomic in a sendto()
945 * with a target address, it's possible that the socket will
946 * have disconnected before the send() can run. In that case
947 * return the slightly counter-intuitive but otherwise
948 * correct error that the socket is not connected.
950 * Locking here must be done carefully: the linkage lock
951 * prevents interconnections between unpcbs from changing, so
952 * we can traverse from unp to unp2 without acquiring unp's
953 * lock. Socket buffer locks follow unpcb locks, so we can
954 * acquire both remote and lock socket buffer locks.
956 unp2 = unp->unp_conn;
961 so2 = unp2->unp_socket;
963 SOCKBUF_LOCK(&so2->so_rcv);
964 if (unp2->unp_flags & UNP_WANTCRED) {
966 * Credentials are passed only once on SOCK_STREAM
967 * and SOCK_SEQPACKET.
969 unp2->unp_flags &= ~UNP_WANTCRED;
970 control = unp_addsockcred(td, control);
973 * Send to paired receive port, and then reduce send buffer
974 * hiwater marks to maintain backpressure. Wake up readers.
976 switch (so->so_type) {
978 if (control != NULL) {
979 if (sbappendcontrol_locked(&so2->so_rcv, m,
983 sbappend_locked(&so2->so_rcv, m);
986 case SOCK_SEQPACKET: {
987 const struct sockaddr *from;
991 * Don't check for space available in so2->so_rcv.
992 * Unix domain sockets only check for space in the
993 * sending sockbuf, and that check is performed one
994 * level up the stack.
996 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1003 mbcnt = so2->so_rcv.sb_mbcnt;
1004 sbcc = sbavail(&so2->so_rcv);
1006 sorwakeup_locked(so2);
1008 SOCKBUF_UNLOCK(&so2->so_rcv);
1011 * The PCB lock on unp2 protects the SB_STOP flag. Without it,
1012 * it would be possible for uipc_rcvd to be called at this
1013 * point, drain the receiving sockbuf, clear SB_STOP, and then
1014 * we would set SB_STOP below. That could lead to an empty
1015 * sockbuf having SB_STOP set
1017 SOCKBUF_LOCK(&so->so_snd);
1018 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1019 so->so_snd.sb_flags |= SB_STOP;
1020 SOCKBUF_UNLOCK(&so->so_snd);
1021 UNP_PCB_UNLOCK(unp2);
1027 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1029 if (flags & PRUS_EOF) {
1033 UNP_PCB_UNLOCK(unp);
1036 if ((nam != NULL) || (flags & PRUS_EOF))
1041 if (control != NULL && error != 0)
1042 unp_dispose(control);
1045 if (control != NULL)
1053 uipc_ready(struct socket *so, struct mbuf *m, int count)
1055 struct unpcb *unp, *unp2;
1059 unp = sotounpcb(so);
1062 unp2 = unp->unp_conn;
1064 so2 = unp2->unp_socket;
1066 SOCKBUF_LOCK(&so2->so_rcv);
1067 if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1068 sorwakeup_locked(so2);
1070 SOCKBUF_UNLOCK(&so2->so_rcv);
1072 UNP_PCB_UNLOCK(unp2);
1079 uipc_sense(struct socket *so, struct stat *sb)
1083 unp = sotounpcb(so);
1084 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1086 sb->st_blksize = so->so_snd.sb_hiwat;
1089 if (unp->unp_ino == 0)
1090 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1091 sb->st_ino = unp->unp_ino;
1092 UNP_PCB_UNLOCK(unp);
1097 uipc_shutdown(struct socket *so)
1101 unp = sotounpcb(so);
1102 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1108 UNP_PCB_UNLOCK(unp);
1114 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1117 const struct sockaddr *sa;
1119 unp = sotounpcb(so);
1120 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1122 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1124 if (unp->unp_addr != NULL)
1125 sa = (struct sockaddr *) unp->unp_addr;
1128 bcopy(sa, *nam, sa->sa_len);
1129 UNP_PCB_UNLOCK(unp);
1133 static struct pr_usrreqs uipc_usrreqs_dgram = {
1134 .pru_abort = uipc_abort,
1135 .pru_accept = uipc_accept,
1136 .pru_attach = uipc_attach,
1137 .pru_bind = uipc_bind,
1138 .pru_bindat = uipc_bindat,
1139 .pru_connect = uipc_connect,
1140 .pru_connectat = uipc_connectat,
1141 .pru_connect2 = uipc_connect2,
1142 .pru_detach = uipc_detach,
1143 .pru_disconnect = uipc_disconnect,
1144 .pru_listen = uipc_listen,
1145 .pru_peeraddr = uipc_peeraddr,
1146 .pru_rcvd = uipc_rcvd,
1147 .pru_send = uipc_send,
1148 .pru_sense = uipc_sense,
1149 .pru_shutdown = uipc_shutdown,
1150 .pru_sockaddr = uipc_sockaddr,
1151 .pru_soreceive = soreceive_dgram,
1152 .pru_close = uipc_close,
1155 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1156 .pru_abort = uipc_abort,
1157 .pru_accept = uipc_accept,
1158 .pru_attach = uipc_attach,
1159 .pru_bind = uipc_bind,
1160 .pru_bindat = uipc_bindat,
1161 .pru_connect = uipc_connect,
1162 .pru_connectat = uipc_connectat,
1163 .pru_connect2 = uipc_connect2,
1164 .pru_detach = uipc_detach,
1165 .pru_disconnect = uipc_disconnect,
1166 .pru_listen = uipc_listen,
1167 .pru_peeraddr = uipc_peeraddr,
1168 .pru_rcvd = uipc_rcvd,
1169 .pru_send = uipc_send,
1170 .pru_sense = uipc_sense,
1171 .pru_shutdown = uipc_shutdown,
1172 .pru_sockaddr = uipc_sockaddr,
1173 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1174 .pru_close = uipc_close,
1177 static struct pr_usrreqs uipc_usrreqs_stream = {
1178 .pru_abort = uipc_abort,
1179 .pru_accept = uipc_accept,
1180 .pru_attach = uipc_attach,
1181 .pru_bind = uipc_bind,
1182 .pru_bindat = uipc_bindat,
1183 .pru_connect = uipc_connect,
1184 .pru_connectat = uipc_connectat,
1185 .pru_connect2 = uipc_connect2,
1186 .pru_detach = uipc_detach,
1187 .pru_disconnect = uipc_disconnect,
1188 .pru_listen = uipc_listen,
1189 .pru_peeraddr = uipc_peeraddr,
1190 .pru_rcvd = uipc_rcvd,
1191 .pru_send = uipc_send,
1192 .pru_ready = uipc_ready,
1193 .pru_sense = uipc_sense,
1194 .pru_shutdown = uipc_shutdown,
1195 .pru_sockaddr = uipc_sockaddr,
1196 .pru_soreceive = soreceive_generic,
1197 .pru_close = uipc_close,
1201 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1207 if (sopt->sopt_level != 0)
1210 unp = sotounpcb(so);
1211 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1213 switch (sopt->sopt_dir) {
1215 switch (sopt->sopt_name) {
1216 case LOCAL_PEERCRED:
1218 if (unp->unp_flags & UNP_HAVEPC)
1219 xu = unp->unp_peercred;
1221 if (so->so_type == SOCK_STREAM)
1226 UNP_PCB_UNLOCK(unp);
1228 error = sooptcopyout(sopt, &xu, sizeof(xu));
1232 /* Unlocked read. */
1233 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1234 error = sooptcopyout(sopt, &optval, sizeof(optval));
1237 case LOCAL_CONNWAIT:
1238 /* Unlocked read. */
1239 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1240 error = sooptcopyout(sopt, &optval, sizeof(optval));
1250 switch (sopt->sopt_name) {
1252 case LOCAL_CONNWAIT:
1253 error = sooptcopyin(sopt, &optval, sizeof(optval),
1258 #define OPTSET(bit) do { \
1259 UNP_PCB_LOCK(unp); \
1261 unp->unp_flags |= bit; \
1263 unp->unp_flags &= ~bit; \
1264 UNP_PCB_UNLOCK(unp); \
1267 switch (sopt->sopt_name) {
1269 OPTSET(UNP_WANTCRED);
1272 case LOCAL_CONNWAIT:
1273 OPTSET(UNP_CONNWAIT);
1282 error = ENOPROTOOPT;
1295 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1298 return (unp_connectat(AT_FDCWD, so, nam, td));
1302 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1305 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1307 struct socket *so2, *so3;
1308 struct unpcb *unp, *unp2, *unp3;
1309 struct nameidata nd;
1310 char buf[SOCK_MAXADDRLEN];
1311 struct sockaddr *sa;
1312 cap_rights_t rights;
1315 if (nam->sa_family != AF_UNIX)
1316 return (EAFNOSUPPORT);
1318 UNP_LINK_WLOCK_ASSERT();
1320 unp = sotounpcb(so);
1321 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1323 if (nam->sa_len > sizeof(struct sockaddr_un))
1325 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1328 bcopy(soun->sun_path, buf, len);
1332 if (unp->unp_flags & UNP_CONNECTING) {
1333 UNP_PCB_UNLOCK(unp);
1337 unp->unp_flags |= UNP_CONNECTING;
1338 UNP_PCB_UNLOCK(unp);
1340 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1341 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1342 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1348 ASSERT_VOP_LOCKED(vp, "unp_connect");
1349 NDFREE(&nd, NDF_ONLY_PNBUF);
1353 if (vp->v_type != VSOCK) {
1358 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1362 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1366 unp = sotounpcb(so);
1367 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1370 * Lock linkage lock for two reasons: make sure v_socket is stable,
1371 * and to protect simultaneous locking of multiple pcbs.
1374 VOP_UNP_CONNECT(vp, &so2);
1376 error = ECONNREFUSED;
1379 if (so->so_type != so2->so_type) {
1383 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1384 if (so2->so_options & SO_ACCEPTCONN) {
1385 CURVNET_SET(so2->so_vnet);
1386 so3 = sonewconn(so2, 0);
1391 error = ECONNREFUSED;
1394 unp = sotounpcb(so);
1395 unp2 = sotounpcb(so2);
1396 unp3 = sotounpcb(so3);
1400 if (unp2->unp_addr != NULL) {
1401 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1402 unp3->unp_addr = (struct sockaddr_un *) sa;
1407 * The connector's (client's) credentials are copied from its
1408 * process structure at the time of connect() (which is now).
1410 cru2x(td->td_ucred, &unp3->unp_peercred);
1411 unp3->unp_flags |= UNP_HAVEPC;
1414 * The receiver's (server's) credentials are copied from the
1415 * unp_peercred member of socket on which the former called
1416 * listen(); uipc_listen() cached that process's credentials
1417 * at that time so we can use them now.
1419 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1420 ("unp_connect: listener without cached peercred"));
1421 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1422 sizeof(unp->unp_peercred));
1423 unp->unp_flags |= UNP_HAVEPC;
1424 if (unp2->unp_flags & UNP_WANTCRED)
1425 unp3->unp_flags |= UNP_WANTCRED;
1426 UNP_PCB_UNLOCK(unp3);
1427 UNP_PCB_UNLOCK(unp2);
1428 UNP_PCB_UNLOCK(unp);
1430 mac_socketpeer_set_from_socket(so, so3);
1431 mac_socketpeer_set_from_socket(so3, so);
1436 unp = sotounpcb(so);
1437 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1438 unp2 = sotounpcb(so2);
1439 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1442 error = unp_connect2(so, so2, PRU_CONNECT);
1443 UNP_PCB_UNLOCK(unp2);
1444 UNP_PCB_UNLOCK(unp);
1453 unp->unp_flags &= ~UNP_CONNECTING;
1454 UNP_PCB_UNLOCK(unp);
1459 unp_connect2(struct socket *so, struct socket *so2, int req)
1464 unp = sotounpcb(so);
1465 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1466 unp2 = sotounpcb(so2);
1467 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1469 UNP_LINK_WLOCK_ASSERT();
1470 UNP_PCB_LOCK_ASSERT(unp);
1471 UNP_PCB_LOCK_ASSERT(unp2);
1473 if (so2->so_type != so->so_type)
1474 return (EPROTOTYPE);
1475 unp->unp_conn = unp2;
1477 switch (so->so_type) {
1479 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1484 case SOCK_SEQPACKET:
1485 unp2->unp_conn = unp;
1486 if (req == PRU_CONNECT &&
1487 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1495 panic("unp_connect2");
1501 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1505 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1507 UNP_LINK_WLOCK_ASSERT();
1508 UNP_PCB_LOCK_ASSERT(unp);
1509 UNP_PCB_LOCK_ASSERT(unp2);
1511 unp->unp_conn = NULL;
1512 switch (unp->unp_socket->so_type) {
1514 LIST_REMOVE(unp, unp_reflink);
1515 so = unp->unp_socket;
1517 so->so_state &= ~SS_ISCONNECTED;
1522 case SOCK_SEQPACKET:
1523 soisdisconnected(unp->unp_socket);
1524 unp2->unp_conn = NULL;
1525 soisdisconnected(unp2->unp_socket);
1531 * unp_pcblist() walks the global list of struct unpcb's to generate a
1532 * pointer list, bumping the refcount on each unpcb. It then copies them out
1533 * sequentially, validating the generation number on each to see if it has
1534 * been detached. All of this is necessary because copyout() may sleep on
1538 unp_pcblist(SYSCTL_HANDLER_ARGS)
1542 struct unpcb *unp, **unp_list;
1544 struct xunpgen *xug;
1545 struct unp_head *head;
1548 switch ((intptr_t)arg1) {
1557 case SOCK_SEQPACKET:
1562 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1566 * The process of preparing the PCB list is too time-consuming and
1567 * resource-intensive to repeat twice on every request.
1569 if (req->oldptr == NULL) {
1571 req->oldidx = 2 * (sizeof *xug)
1572 + (n + n/8) * sizeof(struct xunpcb);
1576 if (req->newptr != NULL)
1580 * OK, now we're committed to doing something.
1582 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1584 gencnt = unp_gencnt;
1588 xug->xug_len = sizeof *xug;
1590 xug->xug_gen = gencnt;
1591 xug->xug_sogen = so_gencnt;
1592 error = SYSCTL_OUT(req, xug, sizeof *xug);
1598 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1601 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1602 unp = LIST_NEXT(unp, unp_link)) {
1604 if (unp->unp_gencnt <= gencnt) {
1605 if (cr_cansee(req->td->td_ucred,
1606 unp->unp_socket->so_cred)) {
1607 UNP_PCB_UNLOCK(unp);
1610 unp_list[i++] = unp;
1611 unp->unp_refcount++;
1613 UNP_PCB_UNLOCK(unp);
1616 n = i; /* In case we lost some during malloc. */
1619 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1620 for (i = 0; i < n; i++) {
1623 unp->unp_refcount--;
1624 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1625 xu->xu_len = sizeof *xu;
1628 * XXX - need more locking here to protect against
1629 * connect/disconnect races for SMP.
1631 if (unp->unp_addr != NULL)
1632 bcopy(unp->unp_addr, &xu->xu_addr,
1633 unp->unp_addr->sun_len);
1634 if (unp->unp_conn != NULL &&
1635 unp->unp_conn->unp_addr != NULL)
1636 bcopy(unp->unp_conn->unp_addr,
1638 unp->unp_conn->unp_addr->sun_len);
1639 bcopy(unp, &xu->xu_unp, sizeof *unp);
1640 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1641 UNP_PCB_UNLOCK(unp);
1642 error = SYSCTL_OUT(req, xu, sizeof *xu);
1644 freeunp = (unp->unp_refcount == 0);
1645 UNP_PCB_UNLOCK(unp);
1647 UNP_PCB_LOCK_DESTROY(unp);
1648 uma_zfree(unp_zone, unp);
1655 * Give the user an updated idea of our state. If the
1656 * generation differs from what we told her before, she knows
1657 * that something happened while we were processing this
1658 * request, and it might be necessary to retry.
1660 xug->xug_gen = unp_gencnt;
1661 xug->xug_sogen = so_gencnt;
1662 xug->xug_count = unp_count;
1663 error = SYSCTL_OUT(req, xug, sizeof *xug);
1665 free(unp_list, M_TEMP);
1670 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1671 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1672 "List of active local datagram sockets");
1673 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1674 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1675 "List of active local stream sockets");
1676 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1677 CTLTYPE_OPAQUE | CTLFLAG_RD,
1678 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1679 "List of active local seqpacket sockets");
1682 unp_shutdown(struct unpcb *unp)
1687 UNP_LINK_WLOCK_ASSERT();
1688 UNP_PCB_LOCK_ASSERT(unp);
1690 unp2 = unp->unp_conn;
1691 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1692 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1693 so = unp2->unp_socket;
1700 unp_drop(struct unpcb *unp, int errno)
1702 struct socket *so = unp->unp_socket;
1705 UNP_LINK_WLOCK_ASSERT();
1706 UNP_PCB_LOCK_ASSERT(unp);
1708 so->so_error = errno;
1709 unp2 = unp->unp_conn;
1713 unp_disconnect(unp, unp2);
1714 UNP_PCB_UNLOCK(unp2);
1718 unp_freerights(struct filedescent **fdep, int fdcount)
1723 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1725 for (i = 0; i < fdcount; i++) {
1726 fp = fdep[i]->fde_file;
1727 filecaps_free(&fdep[i]->fde_caps);
1730 free(fdep[0], M_FILECAPS);
1734 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1736 struct thread *td = curthread; /* XXX */
1737 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1740 struct filedesc *fdesc = td->td_proc->p_fd;
1741 struct filedescent **fdep;
1743 socklen_t clen = control->m_len, datalen;
1747 UNP_LINK_UNLOCK_ASSERT();
1750 if (controlp != NULL) /* controlp == NULL => free control messages */
1752 while (cm != NULL) {
1753 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1757 data = CMSG_DATA(cm);
1758 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1759 if (cm->cmsg_level == SOL_SOCKET
1760 && cm->cmsg_type == SCM_RIGHTS) {
1761 newfds = datalen / sizeof(*fdep);
1766 /* If we're not outputting the descriptors free them. */
1767 if (error || controlp == NULL) {
1768 unp_freerights(fdep, newfds);
1771 FILEDESC_XLOCK(fdesc);
1774 * Now change each pointer to an fd in the global
1775 * table to an integer that is the index to the local
1776 * fd table entry that we set up to point to the
1777 * global one we are transferring.
1779 newlen = newfds * sizeof(int);
1780 *controlp = sbcreatecontrol(NULL, newlen,
1781 SCM_RIGHTS, SOL_SOCKET);
1782 if (*controlp == NULL) {
1783 FILEDESC_XUNLOCK(fdesc);
1785 unp_freerights(fdep, newfds);
1790 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1791 if (fdallocn(td, 0, fdp, newfds) != 0) {
1792 FILEDESC_XUNLOCK(fdesc);
1794 unp_freerights(fdep, newfds);
1799 for (i = 0; i < newfds; i++, fdp++) {
1800 _finstall(fdesc, fdep[i]->fde_file, *fdp,
1801 (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
1802 &fdep[i]->fde_caps);
1803 unp_externalize_fp(fdep[i]->fde_file);
1805 FILEDESC_XUNLOCK(fdesc);
1806 free(fdep[0], M_FILECAPS);
1808 /* We can just copy anything else across. */
1809 if (error || controlp == NULL)
1811 *controlp = sbcreatecontrol(NULL, datalen,
1812 cm->cmsg_type, cm->cmsg_level);
1813 if (*controlp == NULL) {
1818 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1821 controlp = &(*controlp)->m_next;
1824 if (CMSG_SPACE(datalen) < clen) {
1825 clen -= CMSG_SPACE(datalen);
1826 cm = (struct cmsghdr *)
1827 ((caddr_t)cm + CMSG_SPACE(datalen));
1839 unp_zone_change(void *tag)
1842 uma_zone_set_max(unp_zone, maxsockets);
1850 if (!IS_DEFAULT_VNET(curvnet))
1853 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1854 NULL, NULL, UMA_ALIGN_PTR, 0);
1855 if (unp_zone == NULL)
1857 uma_zone_set_max(unp_zone, maxsockets);
1858 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1859 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1860 NULL, EVENTHANDLER_PRI_ANY);
1861 LIST_INIT(&unp_dhead);
1862 LIST_INIT(&unp_shead);
1863 LIST_INIT(&unp_sphead);
1864 SLIST_INIT(&unp_defers);
1865 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1866 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1867 UNP_LINK_LOCK_INIT();
1868 UNP_LIST_LOCK_INIT();
1869 UNP_DEFERRED_LOCK_INIT();
1873 unp_internalize(struct mbuf **controlp, struct thread *td)
1875 struct mbuf *control = *controlp;
1876 struct proc *p = td->td_proc;
1877 struct filedesc *fdesc = p->p_fd;
1879 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1880 struct cmsgcred *cmcred;
1881 struct filedescent *fde, **fdep, *fdev;
1886 socklen_t clen = control->m_len, datalen;
1890 UNP_LINK_UNLOCK_ASSERT();
1894 while (cm != NULL) {
1895 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1896 || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
1900 data = CMSG_DATA(cm);
1901 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1903 switch (cm->cmsg_type) {
1905 * Fill in credential information.
1908 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1909 SCM_CREDS, SOL_SOCKET);
1910 if (*controlp == NULL) {
1914 cmcred = (struct cmsgcred *)
1915 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1916 cmcred->cmcred_pid = p->p_pid;
1917 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1918 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1919 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1920 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1922 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1923 cmcred->cmcred_groups[i] =
1924 td->td_ucred->cr_groups[i];
1928 oldfds = datalen / sizeof (int);
1932 * Check that all the FDs passed in refer to legal
1933 * files. If not, reject the entire operation.
1936 FILEDESC_SLOCK(fdesc);
1937 for (i = 0; i < oldfds; i++, fdp++) {
1938 fp = fget_locked(fdesc, *fdp);
1940 FILEDESC_SUNLOCK(fdesc);
1944 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1945 FILEDESC_SUNLOCK(fdesc);
1953 * Now replace the integer FDs with pointers to the
1954 * file structure and capability rights.
1956 newlen = oldfds * sizeof(fdep[0]);
1957 *controlp = sbcreatecontrol(NULL, newlen,
1958 SCM_RIGHTS, SOL_SOCKET);
1959 if (*controlp == NULL) {
1960 FILEDESC_SUNLOCK(fdesc);
1965 fdep = (struct filedescent **)
1966 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1967 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1969 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1970 fde = &fdesc->fd_ofiles[*fdp];
1972 fdep[i]->fde_file = fde->fde_file;
1973 filecaps_copy(&fde->fde_caps,
1974 &fdep[i]->fde_caps);
1975 unp_internalize_fp(fdep[i]->fde_file);
1977 FILEDESC_SUNLOCK(fdesc);
1981 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1982 SCM_TIMESTAMP, SOL_SOCKET);
1983 if (*controlp == NULL) {
1987 tv = (struct timeval *)
1988 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1993 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
1994 SCM_BINTIME, SOL_SOCKET);
1995 if (*controlp == NULL) {
1999 bt = (struct bintime *)
2000 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2009 controlp = &(*controlp)->m_next;
2010 if (CMSG_SPACE(datalen) < clen) {
2011 clen -= CMSG_SPACE(datalen);
2012 cm = (struct cmsghdr *)
2013 ((caddr_t)cm + CMSG_SPACE(datalen));
2025 static struct mbuf *
2026 unp_addsockcred(struct thread *td, struct mbuf *control)
2028 struct mbuf *m, *n, *n_prev;
2029 struct sockcred *sc;
2030 const struct cmsghdr *cm;
2034 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2035 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2039 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2040 sc->sc_uid = td->td_ucred->cr_ruid;
2041 sc->sc_euid = td->td_ucred->cr_uid;
2042 sc->sc_gid = td->td_ucred->cr_rgid;
2043 sc->sc_egid = td->td_ucred->cr_gid;
2044 sc->sc_ngroups = ngroups;
2045 for (i = 0; i < sc->sc_ngroups; i++)
2046 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2049 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2050 * created SCM_CREDS control message (struct sockcred) has another
2053 if (control != NULL)
2054 for (n = control, n_prev = NULL; n != NULL;) {
2055 cm = mtod(n, struct cmsghdr *);
2056 if (cm->cmsg_level == SOL_SOCKET &&
2057 cm->cmsg_type == SCM_CREDS) {
2059 control = n->m_next;
2061 n_prev->m_next = n->m_next;
2069 /* Prepend it to the head. */
2070 m->m_next = control;
2074 static struct unpcb *
2075 fptounp(struct file *fp)
2079 if (fp->f_type != DTYPE_SOCKET)
2081 if ((so = fp->f_data) == NULL)
2083 if (so->so_proto->pr_domain != &localdomain)
2085 return sotounpcb(so);
2089 unp_discard(struct file *fp)
2091 struct unp_defer *dr;
2093 if (unp_externalize_fp(fp)) {
2094 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2096 UNP_DEFERRED_LOCK();
2097 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2098 UNP_DEFERRED_UNLOCK();
2099 atomic_add_int(&unp_defers_count, 1);
2100 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2102 (void) closef(fp, (struct thread *)NULL);
2106 unp_process_defers(void *arg __unused, int pending)
2108 struct unp_defer *dr;
2109 SLIST_HEAD(, unp_defer) drl;
2114 UNP_DEFERRED_LOCK();
2115 if (SLIST_FIRST(&unp_defers) == NULL) {
2116 UNP_DEFERRED_UNLOCK();
2119 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2120 UNP_DEFERRED_UNLOCK();
2122 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2123 SLIST_REMOVE_HEAD(&drl, ud_link);
2124 closef(dr->ud_fp, NULL);
2128 atomic_add_int(&unp_defers_count, -count);
2133 unp_internalize_fp(struct file *fp)
2138 if ((unp = fptounp(fp)) != NULL) {
2140 unp->unp_msgcount++;
2148 unp_externalize_fp(struct file *fp)
2154 if ((unp = fptounp(fp)) != NULL) {
2155 unp->unp_msgcount--;
2165 * unp_defer indicates whether additional work has been defered for a future
2166 * pass through unp_gc(). It is thread local and does not require explicit
2169 static int unp_marked;
2170 static int unp_unreachable;
2173 unp_accessable(struct filedescent **fdep, int fdcount)
2179 for (i = 0; i < fdcount; i++) {
2180 fp = fdep[i]->fde_file;
2181 if ((unp = fptounp(fp)) == NULL)
2183 if (unp->unp_gcflag & UNPGC_REF)
2185 unp->unp_gcflag &= ~UNPGC_DEAD;
2186 unp->unp_gcflag |= UNPGC_REF;
2192 unp_gc_process(struct unpcb *unp)
2198 /* Already processed. */
2199 if (unp->unp_gcflag & UNPGC_SCANNED)
2204 * Check for a socket potentially in a cycle. It must be in a
2205 * queue as indicated by msgcount, and this must equal the file
2206 * reference count. Note that when msgcount is 0 the file is NULL.
2208 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2209 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2210 unp->unp_gcflag |= UNPGC_DEAD;
2216 * Mark all sockets we reference with RIGHTS.
2218 so = unp->unp_socket;
2219 SOCKBUF_LOCK(&so->so_rcv);
2220 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2221 SOCKBUF_UNLOCK(&so->so_rcv);
2224 * Mark all sockets in our accept queue.
2227 TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2228 SOCKBUF_LOCK(&soa->so_rcv);
2229 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2230 SOCKBUF_UNLOCK(&soa->so_rcv);
2233 unp->unp_gcflag |= UNPGC_SCANNED;
2236 static int unp_recycled;
2237 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2238 "Number of unreachable sockets claimed by the garbage collector.");
2240 static int unp_taskcount;
2241 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2242 "Number of times the garbage collector has run.");
2245 unp_gc(__unused void *arg, int pending)
2247 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2249 struct unp_head **head;
2250 struct file *f, **unref;
2257 * First clear all gc flags from previous runs.
2259 for (head = heads; *head != NULL; head++)
2260 LIST_FOREACH(unp, *head, unp_link)
2261 unp->unp_gcflag = 0;
2264 * Scan marking all reachable sockets with UNPGC_REF. Once a socket
2265 * is reachable all of the sockets it references are reachable.
2266 * Stop the scan once we do a complete loop without discovering
2267 * a new reachable socket.
2270 unp_unreachable = 0;
2272 for (head = heads; *head != NULL; head++)
2273 LIST_FOREACH(unp, *head, unp_link)
2274 unp_gc_process(unp);
2275 } while (unp_marked);
2277 if (unp_unreachable == 0)
2281 * Allocate space for a local list of dead unpcbs.
2283 unref = malloc(unp_unreachable * sizeof(struct file *),
2287 * Iterate looking for sockets which have been specifically marked
2288 * as as unreachable and store them locally.
2292 for (total = 0, head = heads; *head != NULL; head++)
2293 LIST_FOREACH(unp, *head, unp_link)
2294 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2296 if (unp->unp_msgcount == 0 || f == NULL ||
2297 f->f_count != unp->unp_msgcount)
2301 KASSERT(total <= unp_unreachable,
2302 ("unp_gc: incorrect unreachable count."));
2308 * Now flush all sockets, free'ing rights. This will free the
2309 * struct files associated with these sockets but leave each socket
2310 * with one remaining ref.
2312 for (i = 0; i < total; i++) {
2315 so = unref[i]->f_data;
2316 CURVNET_SET(so->so_vnet);
2322 * And finally release the sockets so they can be reclaimed.
2324 for (i = 0; i < total; i++)
2325 fdrop(unref[i], NULL);
2326 unp_recycled += total;
2327 free(unref, M_TEMP);
2331 unp_dispose(struct mbuf *m)
2335 unp_scan(m, unp_freerights);
2339 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2344 socklen_t clen, datalen;
2346 while (m0 != NULL) {
2347 for (m = m0; m; m = m->m_next) {
2348 if (m->m_type != MT_CONTROL)
2351 cm = mtod(m, struct cmsghdr *);
2354 while (cm != NULL) {
2355 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2358 data = CMSG_DATA(cm);
2359 datalen = (caddr_t)cm + cm->cmsg_len
2362 if (cm->cmsg_level == SOL_SOCKET &&
2363 cm->cmsg_type == SCM_RIGHTS) {
2364 (*op)(data, datalen /
2365 sizeof(struct filedescent *));
2368 if (CMSG_SPACE(datalen) < clen) {
2369 clen -= CMSG_SPACE(datalen);
2370 cm = (struct cmsghdr *)
2371 ((caddr_t)cm + CMSG_SPACE(datalen));
2383 * A helper function called by VFS before socket-type vnode reclamation.
2384 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2388 vfs_unp_reclaim(struct vnode *vp)
2394 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2395 KASSERT(vp->v_type == VSOCK,
2396 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2400 VOP_UNP_CONNECT(vp, &so);
2403 unp = sotounpcb(so);
2407 if (unp->unp_vnode == vp) {
2409 unp->unp_vnode = NULL;
2412 UNP_PCB_UNLOCK(unp);
2421 db_print_indent(int indent)
2425 for (i = 0; i < indent; i++)
2430 db_print_unpflags(int unp_flags)
2435 if (unp_flags & UNP_HAVEPC) {
2436 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2439 if (unp_flags & UNP_HAVEPCCACHED) {
2440 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2443 if (unp_flags & UNP_WANTCRED) {
2444 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2447 if (unp_flags & UNP_CONNWAIT) {
2448 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2451 if (unp_flags & UNP_CONNECTING) {
2452 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2455 if (unp_flags & UNP_BINDING) {
2456 db_printf("%sUNP_BINDING", comma ? ", " : "");
2462 db_print_xucred(int indent, struct xucred *xu)
2466 db_print_indent(indent);
2467 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
2468 xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2469 db_print_indent(indent);
2470 db_printf("cr_groups: ");
2472 for (i = 0; i < xu->cr_ngroups; i++) {
2473 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2480 db_print_unprefs(int indent, struct unp_head *uh)
2486 LIST_FOREACH(unp, uh, unp_reflink) {
2487 if (counter % 4 == 0)
2488 db_print_indent(indent);
2489 db_printf("%p ", unp);
2490 if (counter % 4 == 3)
2494 if (counter != 0 && counter % 4 != 0)
2498 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2503 db_printf("usage: show unpcb <addr>\n");
2506 unp = (struct unpcb *)addr;
2508 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2511 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2514 db_printf("unp_refs:\n");
2515 db_print_unprefs(2, &unp->unp_refs);
2517 /* XXXRW: Would be nice to print the full address, if any. */
2518 db_printf("unp_addr: %p\n", unp->unp_addr);
2520 db_printf("unp_gencnt: %llu\n",
2521 (unsigned long long)unp->unp_gencnt);
2523 db_printf("unp_flags: %x (", unp->unp_flags);
2524 db_print_unpflags(unp->unp_flags);
2527 db_printf("unp_peercred:\n");
2528 db_print_xucred(2, &unp->unp_peercred);
2530 db_printf("unp_refcount: %u\n", unp->unp_refcount);