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) {
744 error = solisten_proto_check(so);
746 cru2x(td->td_ucred, &unp->unp_peercred);
747 unp->unp_flags |= UNP_HAVEPCCACHED;
748 solisten_proto(so, backlog);
756 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
758 struct unpcb *unp, *unp2;
759 const struct sockaddr *sa;
762 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
764 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
767 * XXX: It seems that this test always fails even when connection is
768 * established. So, this else clause is added as workaround to
769 * return PF_LOCAL sockaddr.
771 unp2 = unp->unp_conn;
774 if (unp2->unp_addr != NULL)
775 sa = (struct sockaddr *) unp2->unp_addr;
778 bcopy(sa, *nam, sa->sa_len);
779 UNP_PCB_UNLOCK(unp2);
782 bcopy(sa, *nam, sa->sa_len);
789 uipc_rcvd(struct socket *so, int flags)
791 struct unpcb *unp, *unp2;
796 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
797 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
798 ("%s: socktype %d", __func__, so->so_type));
801 * Adjust backpressure on sender and wakeup any waiting to write.
803 * The unp lock is acquired to maintain the validity of the unp_conn
804 * pointer; no lock on unp2 is required as unp2->unp_socket will be
805 * static as long as we don't permit unp2 to disconnect from unp,
806 * which is prevented by the lock on unp. We cache values from
807 * so_rcv to avoid holding the so_rcv lock over the entire
808 * transaction on the remote so_snd.
810 SOCKBUF_LOCK(&so->so_rcv);
811 mbcnt = so->so_rcv.sb_mbcnt;
812 sbcc = sbavail(&so->so_rcv);
813 SOCKBUF_UNLOCK(&so->so_rcv);
815 * There is a benign race condition at this point. If we're planning to
816 * clear SB_STOP, but uipc_send is called on the connected socket at
817 * this instant, it might add data to the sockbuf and set SB_STOP. Then
818 * we would erroneously clear SB_STOP below, even though the sockbuf is
819 * full. The race is benign because the only ill effect is to allow the
820 * sockbuf to exceed its size limit, and the size limits are not
821 * strictly guaranteed anyway.
824 unp2 = unp->unp_conn;
829 so2 = unp2->unp_socket;
830 SOCKBUF_LOCK(&so2->so_snd);
831 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
832 so2->so_snd.sb_flags &= ~SB_STOP;
833 sowwakeup_locked(so2);
839 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
840 struct mbuf *control, struct thread *td)
842 struct unpcb *unp, *unp2;
848 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
849 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
850 so->so_type == SOCK_SEQPACKET,
851 ("%s: socktype %d", __func__, so->so_type));
853 if (flags & PRUS_OOB) {
857 if (control != NULL && (error = unp_internalize(&control, td)))
859 if ((nam != NULL) || (flags & PRUS_EOF))
863 switch (so->so_type) {
866 const struct sockaddr *from;
868 unp2 = unp->unp_conn;
870 UNP_LINK_WLOCK_ASSERT();
875 error = unp_connect(so, nam, td);
878 unp2 = unp->unp_conn;
882 * Because connect() and send() are non-atomic in a sendto()
883 * with a target address, it's possible that the socket will
884 * have disconnected before the send() can run. In that case
885 * return the slightly counter-intuitive but otherwise
886 * correct error that the socket is not connected.
893 if (unp2->unp_flags & UNP_WANTCRED)
894 control = unp_addsockcred(td, control);
896 if (unp->unp_addr != NULL)
897 from = (struct sockaddr *)unp->unp_addr;
900 so2 = unp2->unp_socket;
901 SOCKBUF_LOCK(&so2->so_rcv);
902 if (sbappendaddr_locked(&so2->so_rcv, from, m,
904 sorwakeup_locked(so2);
908 SOCKBUF_UNLOCK(&so2->so_rcv);
912 UNP_LINK_WLOCK_ASSERT();
914 unp_disconnect(unp, unp2);
915 UNP_PCB_UNLOCK(unp2);
923 if ((so->so_state & SS_ISCONNECTED) == 0) {
925 UNP_LINK_WLOCK_ASSERT();
926 error = unp_connect(so, nam, td);
936 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
942 * Because connect() and send() are non-atomic in a sendto()
943 * with a target address, it's possible that the socket will
944 * have disconnected before the send() can run. In that case
945 * return the slightly counter-intuitive but otherwise
946 * correct error that the socket is not connected.
948 * Locking here must be done carefully: the linkage lock
949 * prevents interconnections between unpcbs from changing, so
950 * we can traverse from unp to unp2 without acquiring unp's
951 * lock. Socket buffer locks follow unpcb locks, so we can
952 * acquire both remote and lock socket buffer locks.
954 unp2 = unp->unp_conn;
959 so2 = unp2->unp_socket;
961 SOCKBUF_LOCK(&so2->so_rcv);
962 if (unp2->unp_flags & UNP_WANTCRED) {
964 * Credentials are passed only once on SOCK_STREAM
965 * and SOCK_SEQPACKET.
967 unp2->unp_flags &= ~UNP_WANTCRED;
968 control = unp_addsockcred(td, control);
971 * Send to paired receive port, and then reduce send buffer
972 * hiwater marks to maintain backpressure. Wake up readers.
974 switch (so->so_type) {
976 if (control != NULL) {
977 if (sbappendcontrol_locked(&so2->so_rcv, m,
981 sbappend_locked(&so2->so_rcv, m);
984 case SOCK_SEQPACKET: {
985 const struct sockaddr *from;
989 * Don't check for space available in so2->so_rcv.
990 * Unix domain sockets only check for space in the
991 * sending sockbuf, and that check is performed one
992 * level up the stack.
994 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1001 mbcnt = so2->so_rcv.sb_mbcnt;
1002 sbcc = sbavail(&so2->so_rcv);
1004 sorwakeup_locked(so2);
1006 SOCKBUF_UNLOCK(&so2->so_rcv);
1009 * The PCB lock on unp2 protects the SB_STOP flag. Without it,
1010 * it would be possible for uipc_rcvd to be called at this
1011 * point, drain the receiving sockbuf, clear SB_STOP, and then
1012 * we would set SB_STOP below. That could lead to an empty
1013 * sockbuf having SB_STOP set
1015 SOCKBUF_LOCK(&so->so_snd);
1016 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1017 so->so_snd.sb_flags |= SB_STOP;
1018 SOCKBUF_UNLOCK(&so->so_snd);
1019 UNP_PCB_UNLOCK(unp2);
1025 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1027 if (flags & PRUS_EOF) {
1031 UNP_PCB_UNLOCK(unp);
1034 if ((nam != NULL) || (flags & PRUS_EOF))
1039 if (control != NULL && error != 0)
1040 unp_dispose(control);
1043 if (control != NULL)
1051 uipc_ready(struct socket *so, struct mbuf *m, int count)
1053 struct unpcb *unp, *unp2;
1057 unp = sotounpcb(so);
1060 unp2 = unp->unp_conn;
1062 so2 = unp2->unp_socket;
1064 SOCKBUF_LOCK(&so2->so_rcv);
1065 if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1066 sorwakeup_locked(so2);
1068 SOCKBUF_UNLOCK(&so2->so_rcv);
1070 UNP_PCB_UNLOCK(unp2);
1077 uipc_sense(struct socket *so, struct stat *sb)
1081 unp = sotounpcb(so);
1082 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1084 sb->st_blksize = so->so_snd.sb_hiwat;
1087 if (unp->unp_ino == 0)
1088 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1089 sb->st_ino = unp->unp_ino;
1090 UNP_PCB_UNLOCK(unp);
1095 uipc_shutdown(struct socket *so)
1099 unp = sotounpcb(so);
1100 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1106 UNP_PCB_UNLOCK(unp);
1112 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1115 const struct sockaddr *sa;
1117 unp = sotounpcb(so);
1118 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1120 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1122 if (unp->unp_addr != NULL)
1123 sa = (struct sockaddr *) unp->unp_addr;
1126 bcopy(sa, *nam, sa->sa_len);
1127 UNP_PCB_UNLOCK(unp);
1131 static struct pr_usrreqs uipc_usrreqs_dgram = {
1132 .pru_abort = uipc_abort,
1133 .pru_accept = uipc_accept,
1134 .pru_attach = uipc_attach,
1135 .pru_bind = uipc_bind,
1136 .pru_bindat = uipc_bindat,
1137 .pru_connect = uipc_connect,
1138 .pru_connectat = uipc_connectat,
1139 .pru_connect2 = uipc_connect2,
1140 .pru_detach = uipc_detach,
1141 .pru_disconnect = uipc_disconnect,
1142 .pru_listen = uipc_listen,
1143 .pru_peeraddr = uipc_peeraddr,
1144 .pru_rcvd = uipc_rcvd,
1145 .pru_send = uipc_send,
1146 .pru_sense = uipc_sense,
1147 .pru_shutdown = uipc_shutdown,
1148 .pru_sockaddr = uipc_sockaddr,
1149 .pru_soreceive = soreceive_dgram,
1150 .pru_close = uipc_close,
1153 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1154 .pru_abort = uipc_abort,
1155 .pru_accept = uipc_accept,
1156 .pru_attach = uipc_attach,
1157 .pru_bind = uipc_bind,
1158 .pru_bindat = uipc_bindat,
1159 .pru_connect = uipc_connect,
1160 .pru_connectat = uipc_connectat,
1161 .pru_connect2 = uipc_connect2,
1162 .pru_detach = uipc_detach,
1163 .pru_disconnect = uipc_disconnect,
1164 .pru_listen = uipc_listen,
1165 .pru_peeraddr = uipc_peeraddr,
1166 .pru_rcvd = uipc_rcvd,
1167 .pru_send = uipc_send,
1168 .pru_sense = uipc_sense,
1169 .pru_shutdown = uipc_shutdown,
1170 .pru_sockaddr = uipc_sockaddr,
1171 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1172 .pru_close = uipc_close,
1175 static struct pr_usrreqs uipc_usrreqs_stream = {
1176 .pru_abort = uipc_abort,
1177 .pru_accept = uipc_accept,
1178 .pru_attach = uipc_attach,
1179 .pru_bind = uipc_bind,
1180 .pru_bindat = uipc_bindat,
1181 .pru_connect = uipc_connect,
1182 .pru_connectat = uipc_connectat,
1183 .pru_connect2 = uipc_connect2,
1184 .pru_detach = uipc_detach,
1185 .pru_disconnect = uipc_disconnect,
1186 .pru_listen = uipc_listen,
1187 .pru_peeraddr = uipc_peeraddr,
1188 .pru_rcvd = uipc_rcvd,
1189 .pru_send = uipc_send,
1190 .pru_ready = uipc_ready,
1191 .pru_sense = uipc_sense,
1192 .pru_shutdown = uipc_shutdown,
1193 .pru_sockaddr = uipc_sockaddr,
1194 .pru_soreceive = soreceive_generic,
1195 .pru_close = uipc_close,
1199 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1205 if (sopt->sopt_level != 0)
1208 unp = sotounpcb(so);
1209 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1211 switch (sopt->sopt_dir) {
1213 switch (sopt->sopt_name) {
1214 case LOCAL_PEERCRED:
1216 if (unp->unp_flags & UNP_HAVEPC)
1217 xu = unp->unp_peercred;
1219 if (so->so_type == SOCK_STREAM)
1224 UNP_PCB_UNLOCK(unp);
1226 error = sooptcopyout(sopt, &xu, sizeof(xu));
1230 /* Unlocked read. */
1231 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1232 error = sooptcopyout(sopt, &optval, sizeof(optval));
1235 case LOCAL_CONNWAIT:
1236 /* Unlocked read. */
1237 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1238 error = sooptcopyout(sopt, &optval, sizeof(optval));
1248 switch (sopt->sopt_name) {
1250 case LOCAL_CONNWAIT:
1251 error = sooptcopyin(sopt, &optval, sizeof(optval),
1256 #define OPTSET(bit) do { \
1257 UNP_PCB_LOCK(unp); \
1259 unp->unp_flags |= bit; \
1261 unp->unp_flags &= ~bit; \
1262 UNP_PCB_UNLOCK(unp); \
1265 switch (sopt->sopt_name) {
1267 OPTSET(UNP_WANTCRED);
1270 case LOCAL_CONNWAIT:
1271 OPTSET(UNP_CONNWAIT);
1280 error = ENOPROTOOPT;
1293 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1296 return (unp_connectat(AT_FDCWD, so, nam, td));
1300 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1303 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1305 struct socket *so2, *so3;
1306 struct unpcb *unp, *unp2, *unp3;
1307 struct nameidata nd;
1308 char buf[SOCK_MAXADDRLEN];
1309 struct sockaddr *sa;
1310 cap_rights_t rights;
1313 if (nam->sa_family != AF_UNIX)
1314 return (EAFNOSUPPORT);
1316 UNP_LINK_WLOCK_ASSERT();
1318 unp = sotounpcb(so);
1319 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1321 if (nam->sa_len > sizeof(struct sockaddr_un))
1323 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1326 bcopy(soun->sun_path, buf, len);
1330 if (unp->unp_flags & UNP_CONNECTING) {
1331 UNP_PCB_UNLOCK(unp);
1335 unp->unp_flags |= UNP_CONNECTING;
1336 UNP_PCB_UNLOCK(unp);
1338 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1339 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1340 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1346 ASSERT_VOP_LOCKED(vp, "unp_connect");
1347 NDFREE(&nd, NDF_ONLY_PNBUF);
1351 if (vp->v_type != VSOCK) {
1356 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1360 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1364 unp = sotounpcb(so);
1365 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1368 * Lock linkage lock for two reasons: make sure v_socket is stable,
1369 * and to protect simultaneous locking of multiple pcbs.
1372 VOP_UNP_CONNECT(vp, &so2);
1374 error = ECONNREFUSED;
1377 if (so->so_type != so2->so_type) {
1381 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1382 if (so2->so_options & SO_ACCEPTCONN) {
1383 CURVNET_SET(so2->so_vnet);
1384 so3 = sonewconn(so2, 0);
1389 error = ECONNREFUSED;
1392 unp = sotounpcb(so);
1393 unp2 = sotounpcb(so2);
1394 unp3 = sotounpcb(so3);
1398 if (unp2->unp_addr != NULL) {
1399 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1400 unp3->unp_addr = (struct sockaddr_un *) sa;
1405 * The connector's (client's) credentials are copied from its
1406 * process structure at the time of connect() (which is now).
1408 cru2x(td->td_ucred, &unp3->unp_peercred);
1409 unp3->unp_flags |= UNP_HAVEPC;
1412 * The receiver's (server's) credentials are copied from the
1413 * unp_peercred member of socket on which the former called
1414 * listen(); uipc_listen() cached that process's credentials
1415 * at that time so we can use them now.
1417 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1418 ("unp_connect: listener without cached peercred"));
1419 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1420 sizeof(unp->unp_peercred));
1421 unp->unp_flags |= UNP_HAVEPC;
1422 if (unp2->unp_flags & UNP_WANTCRED)
1423 unp3->unp_flags |= UNP_WANTCRED;
1424 UNP_PCB_UNLOCK(unp3);
1425 UNP_PCB_UNLOCK(unp2);
1426 UNP_PCB_UNLOCK(unp);
1428 mac_socketpeer_set_from_socket(so, so3);
1429 mac_socketpeer_set_from_socket(so3, so);
1434 unp = sotounpcb(so);
1435 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1436 unp2 = sotounpcb(so2);
1437 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1440 error = unp_connect2(so, so2, PRU_CONNECT);
1441 UNP_PCB_UNLOCK(unp2);
1442 UNP_PCB_UNLOCK(unp);
1451 unp->unp_flags &= ~UNP_CONNECTING;
1452 UNP_PCB_UNLOCK(unp);
1457 unp_connect2(struct socket *so, struct socket *so2, int req)
1462 unp = sotounpcb(so);
1463 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1464 unp2 = sotounpcb(so2);
1465 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1467 UNP_LINK_WLOCK_ASSERT();
1468 UNP_PCB_LOCK_ASSERT(unp);
1469 UNP_PCB_LOCK_ASSERT(unp2);
1471 if (so2->so_type != so->so_type)
1472 return (EPROTOTYPE);
1473 unp->unp_conn = unp2;
1475 switch (so->so_type) {
1477 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1482 case SOCK_SEQPACKET:
1483 unp2->unp_conn = unp;
1484 if (req == PRU_CONNECT &&
1485 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1493 panic("unp_connect2");
1499 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1503 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1505 UNP_LINK_WLOCK_ASSERT();
1506 UNP_PCB_LOCK_ASSERT(unp);
1507 UNP_PCB_LOCK_ASSERT(unp2);
1509 unp->unp_conn = NULL;
1510 switch (unp->unp_socket->so_type) {
1512 LIST_REMOVE(unp, unp_reflink);
1513 so = unp->unp_socket;
1515 so->so_state &= ~SS_ISCONNECTED;
1520 case SOCK_SEQPACKET:
1521 soisdisconnected(unp->unp_socket);
1522 unp2->unp_conn = NULL;
1523 soisdisconnected(unp2->unp_socket);
1529 * unp_pcblist() walks the global list of struct unpcb's to generate a
1530 * pointer list, bumping the refcount on each unpcb. It then copies them out
1531 * sequentially, validating the generation number on each to see if it has
1532 * been detached. All of this is necessary because copyout() may sleep on
1536 unp_pcblist(SYSCTL_HANDLER_ARGS)
1540 struct unpcb *unp, **unp_list;
1542 struct xunpgen *xug;
1543 struct unp_head *head;
1546 switch ((intptr_t)arg1) {
1555 case SOCK_SEQPACKET:
1560 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1564 * The process of preparing the PCB list is too time-consuming and
1565 * resource-intensive to repeat twice on every request.
1567 if (req->oldptr == NULL) {
1569 req->oldidx = 2 * (sizeof *xug)
1570 + (n + n/8) * sizeof(struct xunpcb);
1574 if (req->newptr != NULL)
1578 * OK, now we're committed to doing something.
1580 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1582 gencnt = unp_gencnt;
1586 xug->xug_len = sizeof *xug;
1588 xug->xug_gen = gencnt;
1589 xug->xug_sogen = so_gencnt;
1590 error = SYSCTL_OUT(req, xug, sizeof *xug);
1596 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1599 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1600 unp = LIST_NEXT(unp, unp_link)) {
1602 if (unp->unp_gencnt <= gencnt) {
1603 if (cr_cansee(req->td->td_ucred,
1604 unp->unp_socket->so_cred)) {
1605 UNP_PCB_UNLOCK(unp);
1608 unp_list[i++] = unp;
1609 unp->unp_refcount++;
1611 UNP_PCB_UNLOCK(unp);
1614 n = i; /* In case we lost some during malloc. */
1617 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1618 for (i = 0; i < n; i++) {
1621 unp->unp_refcount--;
1622 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1623 xu->xu_len = sizeof *xu;
1626 * XXX - need more locking here to protect against
1627 * connect/disconnect races for SMP.
1629 if (unp->unp_addr != NULL)
1630 bcopy(unp->unp_addr, &xu->xu_addr,
1631 unp->unp_addr->sun_len);
1632 if (unp->unp_conn != NULL &&
1633 unp->unp_conn->unp_addr != NULL)
1634 bcopy(unp->unp_conn->unp_addr,
1636 unp->unp_conn->unp_addr->sun_len);
1637 bcopy(unp, &xu->xu_unp, sizeof *unp);
1638 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1639 UNP_PCB_UNLOCK(unp);
1640 error = SYSCTL_OUT(req, xu, sizeof *xu);
1642 freeunp = (unp->unp_refcount == 0);
1643 UNP_PCB_UNLOCK(unp);
1645 UNP_PCB_LOCK_DESTROY(unp);
1646 uma_zfree(unp_zone, unp);
1653 * Give the user an updated idea of our state. If the
1654 * generation differs from what we told her before, she knows
1655 * that something happened while we were processing this
1656 * request, and it might be necessary to retry.
1658 xug->xug_gen = unp_gencnt;
1659 xug->xug_sogen = so_gencnt;
1660 xug->xug_count = unp_count;
1661 error = SYSCTL_OUT(req, xug, sizeof *xug);
1663 free(unp_list, M_TEMP);
1668 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1669 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1670 "List of active local datagram sockets");
1671 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1672 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1673 "List of active local stream sockets");
1674 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1675 CTLTYPE_OPAQUE | CTLFLAG_RD,
1676 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1677 "List of active local seqpacket sockets");
1680 unp_shutdown(struct unpcb *unp)
1685 UNP_LINK_WLOCK_ASSERT();
1686 UNP_PCB_LOCK_ASSERT(unp);
1688 unp2 = unp->unp_conn;
1689 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1690 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1691 so = unp2->unp_socket;
1698 unp_drop(struct unpcb *unp, int errno)
1700 struct socket *so = unp->unp_socket;
1703 UNP_LINK_WLOCK_ASSERT();
1704 UNP_PCB_LOCK_ASSERT(unp);
1706 so->so_error = errno;
1707 unp2 = unp->unp_conn;
1711 unp_disconnect(unp, unp2);
1712 UNP_PCB_UNLOCK(unp2);
1716 unp_freerights(struct filedescent **fdep, int fdcount)
1721 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1723 for (i = 0; i < fdcount; i++) {
1724 fp = fdep[i]->fde_file;
1725 filecaps_free(&fdep[i]->fde_caps);
1728 free(fdep[0], M_FILECAPS);
1732 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1734 struct thread *td = curthread; /* XXX */
1735 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1738 struct filedesc *fdesc = td->td_proc->p_fd;
1739 struct filedescent *fde, **fdep;
1741 socklen_t clen = control->m_len, datalen;
1745 UNP_LINK_UNLOCK_ASSERT();
1748 if (controlp != NULL) /* controlp == NULL => free control messages */
1750 while (cm != NULL) {
1751 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1755 data = CMSG_DATA(cm);
1756 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1757 if (cm->cmsg_level == SOL_SOCKET
1758 && cm->cmsg_type == SCM_RIGHTS) {
1759 newfds = datalen / sizeof(*fdep);
1764 /* If we're not outputting the descriptors free them. */
1765 if (error || controlp == NULL) {
1766 unp_freerights(fdep, newfds);
1769 FILEDESC_XLOCK(fdesc);
1772 * Now change each pointer to an fd in the global
1773 * table to an integer that is the index to the local
1774 * fd table entry that we set up to point to the
1775 * global one we are transferring.
1777 newlen = newfds * sizeof(int);
1778 *controlp = sbcreatecontrol(NULL, newlen,
1779 SCM_RIGHTS, SOL_SOCKET);
1780 if (*controlp == NULL) {
1781 FILEDESC_XUNLOCK(fdesc);
1783 unp_freerights(fdep, newfds);
1788 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1789 if (fdallocn(td, 0, fdp, newfds) != 0) {
1790 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1792 unp_freerights(fdep, newfds);
1797 for (i = 0; i < newfds; i++, fdp++) {
1798 fde = &fdesc->fd_ofiles[*fdp];
1799 fde->fde_file = fdep[i]->fde_file;
1800 filecaps_move(&fdep[i]->fde_caps,
1802 if ((flags & MSG_CMSG_CLOEXEC) != 0)
1803 fde->fde_flags |= UF_EXCLOSE;
1804 unp_externalize_fp(fde->fde_file);
1806 FILEDESC_XUNLOCK(fdesc);
1807 free(fdep[0], M_FILECAPS);
1809 /* We can just copy anything else across. */
1810 if (error || controlp == NULL)
1812 *controlp = sbcreatecontrol(NULL, datalen,
1813 cm->cmsg_type, cm->cmsg_level);
1814 if (*controlp == NULL) {
1819 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1822 controlp = &(*controlp)->m_next;
1825 if (CMSG_SPACE(datalen) < clen) {
1826 clen -= CMSG_SPACE(datalen);
1827 cm = (struct cmsghdr *)
1828 ((caddr_t)cm + CMSG_SPACE(datalen));
1840 unp_zone_change(void *tag)
1843 uma_zone_set_max(unp_zone, maxsockets);
1851 if (!IS_DEFAULT_VNET(curvnet))
1854 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1855 NULL, NULL, UMA_ALIGN_PTR, 0);
1856 if (unp_zone == NULL)
1858 uma_zone_set_max(unp_zone, maxsockets);
1859 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1860 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1861 NULL, EVENTHANDLER_PRI_ANY);
1862 LIST_INIT(&unp_dhead);
1863 LIST_INIT(&unp_shead);
1864 LIST_INIT(&unp_sphead);
1865 SLIST_INIT(&unp_defers);
1866 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1867 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1868 UNP_LINK_LOCK_INIT();
1869 UNP_LIST_LOCK_INIT();
1870 UNP_DEFERRED_LOCK_INIT();
1874 unp_internalize(struct mbuf **controlp, struct thread *td)
1876 struct mbuf *control = *controlp;
1877 struct proc *p = td->td_proc;
1878 struct filedesc *fdesc = p->p_fd;
1880 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1881 struct cmsgcred *cmcred;
1882 struct filedescent *fde, **fdep, *fdev;
1887 socklen_t clen = control->m_len, datalen;
1891 UNP_LINK_UNLOCK_ASSERT();
1895 while (cm != NULL) {
1896 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1897 || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
1901 data = CMSG_DATA(cm);
1902 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1904 switch (cm->cmsg_type) {
1906 * Fill in credential information.
1909 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1910 SCM_CREDS, SOL_SOCKET);
1911 if (*controlp == NULL) {
1915 cmcred = (struct cmsgcred *)
1916 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1917 cmcred->cmcred_pid = p->p_pid;
1918 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1919 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1920 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1921 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1923 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1924 cmcred->cmcred_groups[i] =
1925 td->td_ucred->cr_groups[i];
1929 oldfds = datalen / sizeof (int);
1933 * Check that all the FDs passed in refer to legal
1934 * files. If not, reject the entire operation.
1937 FILEDESC_SLOCK(fdesc);
1938 for (i = 0; i < oldfds; i++, fdp++) {
1939 fp = fget_locked(fdesc, *fdp);
1941 FILEDESC_SUNLOCK(fdesc);
1945 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1946 FILEDESC_SUNLOCK(fdesc);
1954 * Now replace the integer FDs with pointers to the
1955 * file structure and capability rights.
1957 newlen = oldfds * sizeof(fdep[0]);
1958 *controlp = sbcreatecontrol(NULL, newlen,
1959 SCM_RIGHTS, SOL_SOCKET);
1960 if (*controlp == NULL) {
1961 FILEDESC_SUNLOCK(fdesc);
1966 fdep = (struct filedescent **)
1967 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1968 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1970 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1971 fde = &fdesc->fd_ofiles[*fdp];
1973 fdep[i]->fde_file = fde->fde_file;
1974 filecaps_copy(&fde->fde_caps,
1975 &fdep[i]->fde_caps);
1976 unp_internalize_fp(fdep[i]->fde_file);
1978 FILEDESC_SUNLOCK(fdesc);
1982 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1983 SCM_TIMESTAMP, SOL_SOCKET);
1984 if (*controlp == NULL) {
1988 tv = (struct timeval *)
1989 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1994 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
1995 SCM_BINTIME, SOL_SOCKET);
1996 if (*controlp == NULL) {
2000 bt = (struct bintime *)
2001 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2010 controlp = &(*controlp)->m_next;
2011 if (CMSG_SPACE(datalen) < clen) {
2012 clen -= CMSG_SPACE(datalen);
2013 cm = (struct cmsghdr *)
2014 ((caddr_t)cm + CMSG_SPACE(datalen));
2026 static struct mbuf *
2027 unp_addsockcred(struct thread *td, struct mbuf *control)
2029 struct mbuf *m, *n, *n_prev;
2030 struct sockcred *sc;
2031 const struct cmsghdr *cm;
2035 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2036 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2040 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2041 sc->sc_uid = td->td_ucred->cr_ruid;
2042 sc->sc_euid = td->td_ucred->cr_uid;
2043 sc->sc_gid = td->td_ucred->cr_rgid;
2044 sc->sc_egid = td->td_ucred->cr_gid;
2045 sc->sc_ngroups = ngroups;
2046 for (i = 0; i < sc->sc_ngroups; i++)
2047 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2050 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2051 * created SCM_CREDS control message (struct sockcred) has another
2054 if (control != NULL)
2055 for (n = control, n_prev = NULL; n != NULL;) {
2056 cm = mtod(n, struct cmsghdr *);
2057 if (cm->cmsg_level == SOL_SOCKET &&
2058 cm->cmsg_type == SCM_CREDS) {
2060 control = n->m_next;
2062 n_prev->m_next = n->m_next;
2070 /* Prepend it to the head. */
2071 m->m_next = control;
2075 static struct unpcb *
2076 fptounp(struct file *fp)
2080 if (fp->f_type != DTYPE_SOCKET)
2082 if ((so = fp->f_data) == NULL)
2084 if (so->so_proto->pr_domain != &localdomain)
2086 return sotounpcb(so);
2090 unp_discard(struct file *fp)
2092 struct unp_defer *dr;
2094 if (unp_externalize_fp(fp)) {
2095 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2097 UNP_DEFERRED_LOCK();
2098 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2099 UNP_DEFERRED_UNLOCK();
2100 atomic_add_int(&unp_defers_count, 1);
2101 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2103 (void) closef(fp, (struct thread *)NULL);
2107 unp_process_defers(void *arg __unused, int pending)
2109 struct unp_defer *dr;
2110 SLIST_HEAD(, unp_defer) drl;
2115 UNP_DEFERRED_LOCK();
2116 if (SLIST_FIRST(&unp_defers) == NULL) {
2117 UNP_DEFERRED_UNLOCK();
2120 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2121 UNP_DEFERRED_UNLOCK();
2123 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2124 SLIST_REMOVE_HEAD(&drl, ud_link);
2125 closef(dr->ud_fp, NULL);
2129 atomic_add_int(&unp_defers_count, -count);
2134 unp_internalize_fp(struct file *fp)
2139 if ((unp = fptounp(fp)) != NULL) {
2141 unp->unp_msgcount++;
2149 unp_externalize_fp(struct file *fp)
2155 if ((unp = fptounp(fp)) != NULL) {
2156 unp->unp_msgcount--;
2166 * unp_defer indicates whether additional work has been defered for a future
2167 * pass through unp_gc(). It is thread local and does not require explicit
2170 static int unp_marked;
2171 static int unp_unreachable;
2174 unp_accessable(struct filedescent **fdep, int fdcount)
2180 for (i = 0; i < fdcount; i++) {
2181 fp = fdep[i]->fde_file;
2182 if ((unp = fptounp(fp)) == NULL)
2184 if (unp->unp_gcflag & UNPGC_REF)
2186 unp->unp_gcflag &= ~UNPGC_DEAD;
2187 unp->unp_gcflag |= UNPGC_REF;
2193 unp_gc_process(struct unpcb *unp)
2199 /* Already processed. */
2200 if (unp->unp_gcflag & UNPGC_SCANNED)
2205 * Check for a socket potentially in a cycle. It must be in a
2206 * queue as indicated by msgcount, and this must equal the file
2207 * reference count. Note that when msgcount is 0 the file is NULL.
2209 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2210 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2211 unp->unp_gcflag |= UNPGC_DEAD;
2217 * Mark all sockets we reference with RIGHTS.
2219 so = unp->unp_socket;
2220 SOCKBUF_LOCK(&so->so_rcv);
2221 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2222 SOCKBUF_UNLOCK(&so->so_rcv);
2225 * Mark all sockets in our accept queue.
2228 TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2229 SOCKBUF_LOCK(&soa->so_rcv);
2230 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2231 SOCKBUF_UNLOCK(&soa->so_rcv);
2234 unp->unp_gcflag |= UNPGC_SCANNED;
2237 static int unp_recycled;
2238 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2239 "Number of unreachable sockets claimed by the garbage collector.");
2241 static int unp_taskcount;
2242 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2243 "Number of times the garbage collector has run.");
2246 unp_gc(__unused void *arg, int pending)
2248 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2250 struct unp_head **head;
2251 struct file *f, **unref;
2258 * First clear all gc flags from previous runs.
2260 for (head = heads; *head != NULL; head++)
2261 LIST_FOREACH(unp, *head, unp_link)
2262 unp->unp_gcflag = 0;
2265 * Scan marking all reachable sockets with UNPGC_REF. Once a socket
2266 * is reachable all of the sockets it references are reachable.
2267 * Stop the scan once we do a complete loop without discovering
2268 * a new reachable socket.
2271 unp_unreachable = 0;
2273 for (head = heads; *head != NULL; head++)
2274 LIST_FOREACH(unp, *head, unp_link)
2275 unp_gc_process(unp);
2276 } while (unp_marked);
2278 if (unp_unreachable == 0)
2282 * Allocate space for a local list of dead unpcbs.
2284 unref = malloc(unp_unreachable * sizeof(struct file *),
2288 * Iterate looking for sockets which have been specifically marked
2289 * as as unreachable and store them locally.
2293 for (total = 0, head = heads; *head != NULL; head++)
2294 LIST_FOREACH(unp, *head, unp_link)
2295 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2297 if (unp->unp_msgcount == 0 || f == NULL ||
2298 f->f_count != unp->unp_msgcount)
2302 KASSERT(total <= unp_unreachable,
2303 ("unp_gc: incorrect unreachable count."));
2309 * Now flush all sockets, free'ing rights. This will free the
2310 * struct files associated with these sockets but leave each socket
2311 * with one remaining ref.
2313 for (i = 0; i < total; i++) {
2316 so = unref[i]->f_data;
2317 CURVNET_SET(so->so_vnet);
2323 * And finally release the sockets so they can be reclaimed.
2325 for (i = 0; i < total; i++)
2326 fdrop(unref[i], NULL);
2327 unp_recycled += total;
2328 free(unref, M_TEMP);
2332 unp_dispose(struct mbuf *m)
2336 unp_scan(m, unp_freerights);
2340 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2345 socklen_t clen, datalen;
2347 while (m0 != NULL) {
2348 for (m = m0; m; m = m->m_next) {
2349 if (m->m_type != MT_CONTROL)
2352 cm = mtod(m, struct cmsghdr *);
2355 while (cm != NULL) {
2356 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2359 data = CMSG_DATA(cm);
2360 datalen = (caddr_t)cm + cm->cmsg_len
2363 if (cm->cmsg_level == SOL_SOCKET &&
2364 cm->cmsg_type == SCM_RIGHTS) {
2365 (*op)(data, datalen /
2366 sizeof(struct filedescent *));
2369 if (CMSG_SPACE(datalen) < clen) {
2370 clen -= CMSG_SPACE(datalen);
2371 cm = (struct cmsghdr *)
2372 ((caddr_t)cm + CMSG_SPACE(datalen));
2384 * A helper function called by VFS before socket-type vnode reclamation.
2385 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2389 vfs_unp_reclaim(struct vnode *vp)
2395 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2396 KASSERT(vp->v_type == VSOCK,
2397 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2401 VOP_UNP_CONNECT(vp, &so);
2404 unp = sotounpcb(so);
2408 if (unp->unp_vnode == vp) {
2410 unp->unp_vnode = NULL;
2413 UNP_PCB_UNLOCK(unp);
2422 db_print_indent(int indent)
2426 for (i = 0; i < indent; i++)
2431 db_print_unpflags(int unp_flags)
2436 if (unp_flags & UNP_HAVEPC) {
2437 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2440 if (unp_flags & UNP_HAVEPCCACHED) {
2441 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2444 if (unp_flags & UNP_WANTCRED) {
2445 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2448 if (unp_flags & UNP_CONNWAIT) {
2449 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2452 if (unp_flags & UNP_CONNECTING) {
2453 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2456 if (unp_flags & UNP_BINDING) {
2457 db_printf("%sUNP_BINDING", comma ? ", " : "");
2463 db_print_xucred(int indent, struct xucred *xu)
2467 db_print_indent(indent);
2468 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
2469 xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2470 db_print_indent(indent);
2471 db_printf("cr_groups: ");
2473 for (i = 0; i < xu->cr_ngroups; i++) {
2474 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2481 db_print_unprefs(int indent, struct unp_head *uh)
2487 LIST_FOREACH(unp, uh, unp_reflink) {
2488 if (counter % 4 == 0)
2489 db_print_indent(indent);
2490 db_printf("%p ", unp);
2491 if (counter % 4 == 3)
2495 if (counter != 0 && counter % 4 != 0)
2499 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2504 db_printf("usage: show unpcb <addr>\n");
2507 unp = (struct unpcb *)addr;
2509 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2512 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2515 db_printf("unp_refs:\n");
2516 db_print_unprefs(2, &unp->unp_refs);
2518 /* XXXRW: Would be nice to print the full address, if any. */
2519 db_printf("unp_addr: %p\n", unp->unp_addr);
2521 db_printf("unp_gencnt: %llu\n",
2522 (unsigned long long)unp->unp_gencnt);
2524 db_printf("unp_flags: %x (", unp->unp_flags);
2525 db_print_unpflags(unp->unp_flags);
2528 db_printf("unp_peercred:\n");
2529 db_print_xucred(2, &unp->unp_peercred);
2531 db_printf("unp_refcount: %u\n", unp->unp_refcount);