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/capability.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;
471 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
473 if (soun->sun_len > sizeof(struct sockaddr_un))
475 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
480 * We don't allow simultaneous bind() calls on a single UNIX domain
481 * socket, so flag in-progress operations, and return an error if an
482 * operation is already in progress.
484 * Historically, we have not allowed a socket to be rebound, so this
485 * also returns an error. Not allowing re-binding simplifies the
486 * implementation and avoids a great many possible failure modes.
489 if (unp->unp_vnode != NULL) {
493 if (unp->unp_flags & UNP_BINDING) {
497 unp->unp_flags |= UNP_BINDING;
500 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
501 bcopy(soun->sun_path, buf, namelen);
505 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME,
506 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
507 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
512 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
513 NDFREE(&nd, NDF_ONLY_PNBUF);
523 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
529 vattr.va_type = VSOCK;
530 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
532 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
536 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
537 NDFREE(&nd, NDF_ONLY_PNBUF);
540 vn_finished_write(mp);
544 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
545 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
549 VOP_UNP_BIND(vp, unp->unp_socket);
551 unp->unp_addr = soun;
552 unp->unp_flags &= ~UNP_BINDING;
556 vn_finished_write(mp);
562 unp->unp_flags &= ~UNP_BINDING;
569 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
572 return (uipc_bindat(AT_FDCWD, so, nam, td));
576 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
580 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
582 error = unp_connect(so, nam, td);
588 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
593 KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
595 error = unp_connectat(fd, so, nam, td);
601 uipc_close(struct socket *so)
603 struct unpcb *unp, *unp2;
606 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
610 unp2 = unp->unp_conn;
613 unp_disconnect(unp, unp2);
614 UNP_PCB_UNLOCK(unp2);
621 uipc_connect2(struct socket *so1, struct socket *so2)
623 struct unpcb *unp, *unp2;
628 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
631 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
633 error = unp_connect2(so1, so2, PRU_CONNECT2);
634 UNP_PCB_UNLOCK(unp2);
641 uipc_detach(struct socket *so)
643 struct unpcb *unp, *unp2;
644 struct sockaddr_un *saved_unp_addr;
646 int freeunp, local_unp_rights;
649 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
654 LIST_REMOVE(unp, unp_link);
655 unp->unp_gencnt = ++unp_gencnt;
660 * XXXRW: Should assert vp->v_socket == so.
662 if ((vp = unp->unp_vnode) != NULL) {
664 unp->unp_vnode = NULL;
666 unp2 = unp->unp_conn;
669 unp_disconnect(unp, unp2);
670 UNP_PCB_UNLOCK(unp2);
674 * We hold the linkage lock exclusively, so it's OK to acquire
675 * multiple pcb locks at a time.
677 while (!LIST_EMPTY(&unp->unp_refs)) {
678 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
681 unp_drop(ref, ECONNRESET);
684 local_unp_rights = unp_rights;
686 unp->unp_socket->so_pcb = NULL;
687 saved_unp_addr = unp->unp_addr;
688 unp->unp_addr = NULL;
690 freeunp = (unp->unp_refcount == 0);
691 if (saved_unp_addr != NULL)
692 free(saved_unp_addr, M_SONAME);
694 UNP_PCB_LOCK_DESTROY(unp);
695 uma_zfree(unp_zone, unp);
700 if (local_unp_rights)
701 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
705 uipc_disconnect(struct socket *so)
707 struct unpcb *unp, *unp2;
710 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
714 unp2 = unp->unp_conn;
717 unp_disconnect(unp, unp2);
718 UNP_PCB_UNLOCK(unp2);
726 uipc_listen(struct socket *so, int backlog, struct thread *td)
732 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
735 if (unp->unp_vnode == NULL) {
741 error = solisten_proto_check(so);
743 cru2x(td->td_ucred, &unp->unp_peercred);
744 unp->unp_flags |= UNP_HAVEPCCACHED;
745 solisten_proto(so, backlog);
753 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
755 struct unpcb *unp, *unp2;
756 const struct sockaddr *sa;
759 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
761 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
764 * XXX: It seems that this test always fails even when connection is
765 * established. So, this else clause is added as workaround to
766 * return PF_LOCAL sockaddr.
768 unp2 = unp->unp_conn;
771 if (unp2->unp_addr != NULL)
772 sa = (struct sockaddr *) unp2->unp_addr;
775 bcopy(sa, *nam, sa->sa_len);
776 UNP_PCB_UNLOCK(unp2);
779 bcopy(sa, *nam, sa->sa_len);
786 uipc_rcvd(struct socket *so, int flags)
788 struct unpcb *unp, *unp2;
793 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
795 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
796 panic("uipc_rcvd socktype %d", so->so_type);
799 * Adjust backpressure on sender and wakeup any waiting to write.
801 * The unp lock is acquired to maintain the validity of the unp_conn
802 * pointer; no lock on unp2 is required as unp2->unp_socket will be
803 * static as long as we don't permit unp2 to disconnect from unp,
804 * which is prevented by the lock on unp. We cache values from
805 * so_rcv to avoid holding the so_rcv lock over the entire
806 * transaction on the remote so_snd.
808 SOCKBUF_LOCK(&so->so_rcv);
809 mbcnt = so->so_rcv.sb_mbcnt;
810 sbcc = so->so_rcv.sb_cc;
811 SOCKBUF_UNLOCK(&so->so_rcv);
813 * There is a benign race condition at this point. If we're planning to
814 * clear SB_STOP, but uipc_send is called on the connected socket at
815 * this instant, it might add data to the sockbuf and set SB_STOP. Then
816 * we would erroneously clear SB_STOP below, even though the sockbuf is
817 * full. The race is benign because the only ill effect is to allow the
818 * sockbuf to exceed its size limit, and the size limits are not
819 * strictly guaranteed anyway.
822 unp2 = unp->unp_conn;
827 so2 = unp2->unp_socket;
828 SOCKBUF_LOCK(&so2->so_snd);
829 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
830 so2->so_snd.sb_flags &= ~SB_STOP;
831 sowwakeup_locked(so2);
837 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
838 struct mbuf *control, struct thread *td)
840 struct unpcb *unp, *unp2;
846 KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
848 if (flags & PRUS_OOB) {
852 if (control != NULL && (error = unp_internalize(&control, td)))
854 if ((nam != NULL) || (flags & PRUS_EOF))
858 switch (so->so_type) {
861 const struct sockaddr *from;
863 unp2 = unp->unp_conn;
865 UNP_LINK_WLOCK_ASSERT();
870 error = unp_connect(so, nam, td);
873 unp2 = unp->unp_conn;
877 * Because connect() and send() are non-atomic in a sendto()
878 * with a target address, it's possible that the socket will
879 * have disconnected before the send() can run. In that case
880 * return the slightly counter-intuitive but otherwise
881 * correct error that the socket is not connected.
888 if (unp2->unp_flags & UNP_WANTCRED)
889 control = unp_addsockcred(td, control);
891 if (unp->unp_addr != NULL)
892 from = (struct sockaddr *)unp->unp_addr;
895 so2 = unp2->unp_socket;
896 SOCKBUF_LOCK(&so2->so_rcv);
897 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, from, m,
899 sorwakeup_locked(so2);
903 SOCKBUF_UNLOCK(&so2->so_rcv);
907 UNP_LINK_WLOCK_ASSERT();
909 unp_disconnect(unp, unp2);
910 UNP_PCB_UNLOCK(unp2);
918 if ((so->so_state & SS_ISCONNECTED) == 0) {
920 UNP_LINK_WLOCK_ASSERT();
921 error = unp_connect(so, nam, td);
931 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
937 * Because connect() and send() are non-atomic in a sendto()
938 * with a target address, it's possible that the socket will
939 * have disconnected before the send() can run. In that case
940 * return the slightly counter-intuitive but otherwise
941 * correct error that the socket is not connected.
943 * Locking here must be done carefully: the linkage lock
944 * prevents interconnections between unpcbs from changing, so
945 * we can traverse from unp to unp2 without acquiring unp's
946 * lock. Socket buffer locks follow unpcb locks, so we can
947 * acquire both remote and lock socket buffer locks.
949 unp2 = unp->unp_conn;
954 so2 = unp2->unp_socket;
956 SOCKBUF_LOCK(&so2->so_rcv);
957 if (unp2->unp_flags & UNP_WANTCRED) {
959 * Credentials are passed only once on SOCK_STREAM
960 * and SOCK_SEQPACKET.
962 unp2->unp_flags &= ~UNP_WANTCRED;
963 control = unp_addsockcred(td, control);
966 * Send to paired receive port, and then reduce send buffer
967 * hiwater marks to maintain backpressure. Wake up readers.
969 switch (so->so_type) {
971 if (control != NULL) {
972 if (sbappendcontrol_locked(&so2->so_rcv, m,
976 sbappend_locked(&so2->so_rcv, m);
979 case SOCK_SEQPACKET: {
980 const struct sockaddr *from;
984 * Don't check for space available in so2->so_rcv.
985 * Unix domain sockets only check for space in the
986 * sending sockbuf, and that check is performed one
987 * level up the stack.
989 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
996 mbcnt = so2->so_rcv.sb_mbcnt;
997 sbcc = so2->so_rcv.sb_cc;
998 sorwakeup_locked(so2);
1001 * The PCB lock on unp2 protects the SB_STOP flag. Without it,
1002 * it would be possible for uipc_rcvd to be called at this
1003 * point, drain the receiving sockbuf, clear SB_STOP, and then
1004 * we would set SB_STOP below. That could lead to an empty
1005 * sockbuf having SB_STOP set
1007 SOCKBUF_LOCK(&so->so_snd);
1008 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1009 so->so_snd.sb_flags |= SB_STOP;
1010 SOCKBUF_UNLOCK(&so->so_snd);
1011 UNP_PCB_UNLOCK(unp2);
1016 panic("uipc_send unknown socktype");
1020 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1022 if (flags & PRUS_EOF) {
1026 UNP_PCB_UNLOCK(unp);
1029 if ((nam != NULL) || (flags & PRUS_EOF))
1034 if (control != NULL && error != 0)
1035 unp_dispose(control);
1038 if (control != NULL)
1046 uipc_sense(struct socket *so, struct stat *sb)
1050 unp = sotounpcb(so);
1051 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1053 sb->st_blksize = so->so_snd.sb_hiwat;
1056 if (unp->unp_ino == 0)
1057 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1058 sb->st_ino = unp->unp_ino;
1059 UNP_PCB_UNLOCK(unp);
1064 uipc_shutdown(struct socket *so)
1068 unp = sotounpcb(so);
1069 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1075 UNP_PCB_UNLOCK(unp);
1081 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1084 const struct sockaddr *sa;
1086 unp = sotounpcb(so);
1087 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1089 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1091 if (unp->unp_addr != NULL)
1092 sa = (struct sockaddr *) unp->unp_addr;
1095 bcopy(sa, *nam, sa->sa_len);
1096 UNP_PCB_UNLOCK(unp);
1100 static struct pr_usrreqs uipc_usrreqs_dgram = {
1101 .pru_abort = uipc_abort,
1102 .pru_accept = uipc_accept,
1103 .pru_attach = uipc_attach,
1104 .pru_bind = uipc_bind,
1105 .pru_bindat = uipc_bindat,
1106 .pru_connect = uipc_connect,
1107 .pru_connectat = uipc_connectat,
1108 .pru_connect2 = uipc_connect2,
1109 .pru_detach = uipc_detach,
1110 .pru_disconnect = uipc_disconnect,
1111 .pru_listen = uipc_listen,
1112 .pru_peeraddr = uipc_peeraddr,
1113 .pru_rcvd = uipc_rcvd,
1114 .pru_send = uipc_send,
1115 .pru_sense = uipc_sense,
1116 .pru_shutdown = uipc_shutdown,
1117 .pru_sockaddr = uipc_sockaddr,
1118 .pru_soreceive = soreceive_dgram,
1119 .pru_close = uipc_close,
1122 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1123 .pru_abort = uipc_abort,
1124 .pru_accept = uipc_accept,
1125 .pru_attach = uipc_attach,
1126 .pru_bind = uipc_bind,
1127 .pru_bindat = uipc_bindat,
1128 .pru_connect = uipc_connect,
1129 .pru_connectat = uipc_connectat,
1130 .pru_connect2 = uipc_connect2,
1131 .pru_detach = uipc_detach,
1132 .pru_disconnect = uipc_disconnect,
1133 .pru_listen = uipc_listen,
1134 .pru_peeraddr = uipc_peeraddr,
1135 .pru_rcvd = uipc_rcvd,
1136 .pru_send = uipc_send,
1137 .pru_sense = uipc_sense,
1138 .pru_shutdown = uipc_shutdown,
1139 .pru_sockaddr = uipc_sockaddr,
1140 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1141 .pru_close = uipc_close,
1144 static struct pr_usrreqs uipc_usrreqs_stream = {
1145 .pru_abort = uipc_abort,
1146 .pru_accept = uipc_accept,
1147 .pru_attach = uipc_attach,
1148 .pru_bind = uipc_bind,
1149 .pru_bindat = uipc_bindat,
1150 .pru_connect = uipc_connect,
1151 .pru_connectat = uipc_connectat,
1152 .pru_connect2 = uipc_connect2,
1153 .pru_detach = uipc_detach,
1154 .pru_disconnect = uipc_disconnect,
1155 .pru_listen = uipc_listen,
1156 .pru_peeraddr = uipc_peeraddr,
1157 .pru_rcvd = uipc_rcvd,
1158 .pru_send = uipc_send,
1159 .pru_sense = uipc_sense,
1160 .pru_shutdown = uipc_shutdown,
1161 .pru_sockaddr = uipc_sockaddr,
1162 .pru_soreceive = soreceive_generic,
1163 .pru_close = uipc_close,
1167 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1173 if (sopt->sopt_level != 0)
1176 unp = sotounpcb(so);
1177 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1179 switch (sopt->sopt_dir) {
1181 switch (sopt->sopt_name) {
1182 case LOCAL_PEERCRED:
1184 if (unp->unp_flags & UNP_HAVEPC)
1185 xu = unp->unp_peercred;
1187 if (so->so_type == SOCK_STREAM)
1192 UNP_PCB_UNLOCK(unp);
1194 error = sooptcopyout(sopt, &xu, sizeof(xu));
1198 /* Unlocked read. */
1199 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1200 error = sooptcopyout(sopt, &optval, sizeof(optval));
1203 case LOCAL_CONNWAIT:
1204 /* Unlocked read. */
1205 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1206 error = sooptcopyout(sopt, &optval, sizeof(optval));
1216 switch (sopt->sopt_name) {
1218 case LOCAL_CONNWAIT:
1219 error = sooptcopyin(sopt, &optval, sizeof(optval),
1224 #define OPTSET(bit) do { \
1225 UNP_PCB_LOCK(unp); \
1227 unp->unp_flags |= bit; \
1229 unp->unp_flags &= ~bit; \
1230 UNP_PCB_UNLOCK(unp); \
1233 switch (sopt->sopt_name) {
1235 OPTSET(UNP_WANTCRED);
1238 case LOCAL_CONNWAIT:
1239 OPTSET(UNP_CONNWAIT);
1248 error = ENOPROTOOPT;
1261 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1264 return (unp_connectat(AT_FDCWD, so, nam, td));
1268 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1271 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1273 struct socket *so2, *so3;
1274 struct unpcb *unp, *unp2, *unp3;
1275 struct nameidata nd;
1276 char buf[SOCK_MAXADDRLEN];
1277 struct sockaddr *sa;
1278 cap_rights_t rights;
1281 UNP_LINK_WLOCK_ASSERT();
1283 unp = sotounpcb(so);
1284 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1286 if (nam->sa_len > sizeof(struct sockaddr_un))
1288 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1291 bcopy(soun->sun_path, buf, len);
1295 if (unp->unp_flags & UNP_CONNECTING) {
1296 UNP_PCB_UNLOCK(unp);
1300 unp->unp_flags |= UNP_CONNECTING;
1301 UNP_PCB_UNLOCK(unp);
1303 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1304 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1305 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1311 ASSERT_VOP_LOCKED(vp, "unp_connect");
1312 NDFREE(&nd, NDF_ONLY_PNBUF);
1316 if (vp->v_type != VSOCK) {
1321 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1325 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1329 unp = sotounpcb(so);
1330 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1333 * Lock linkage lock for two reasons: make sure v_socket is stable,
1334 * and to protect simultaneous locking of multiple pcbs.
1337 VOP_UNP_CONNECT(vp, &so2);
1339 error = ECONNREFUSED;
1342 if (so->so_type != so2->so_type) {
1346 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1347 if (so2->so_options & SO_ACCEPTCONN) {
1348 CURVNET_SET(so2->so_vnet);
1349 so3 = sonewconn(so2, 0);
1354 error = ECONNREFUSED;
1357 unp = sotounpcb(so);
1358 unp2 = sotounpcb(so2);
1359 unp3 = sotounpcb(so3);
1363 if (unp2->unp_addr != NULL) {
1364 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1365 unp3->unp_addr = (struct sockaddr_un *) sa;
1370 * The connector's (client's) credentials are copied from its
1371 * process structure at the time of connect() (which is now).
1373 cru2x(td->td_ucred, &unp3->unp_peercred);
1374 unp3->unp_flags |= UNP_HAVEPC;
1377 * The receiver's (server's) credentials are copied from the
1378 * unp_peercred member of socket on which the former called
1379 * listen(); uipc_listen() cached that process's credentials
1380 * at that time so we can use them now.
1382 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1383 ("unp_connect: listener without cached peercred"));
1384 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1385 sizeof(unp->unp_peercred));
1386 unp->unp_flags |= UNP_HAVEPC;
1387 if (unp2->unp_flags & UNP_WANTCRED)
1388 unp3->unp_flags |= UNP_WANTCRED;
1389 UNP_PCB_UNLOCK(unp3);
1390 UNP_PCB_UNLOCK(unp2);
1391 UNP_PCB_UNLOCK(unp);
1393 mac_socketpeer_set_from_socket(so, so3);
1394 mac_socketpeer_set_from_socket(so3, so);
1399 unp = sotounpcb(so);
1400 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1401 unp2 = sotounpcb(so2);
1402 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1405 error = unp_connect2(so, so2, PRU_CONNECT);
1406 UNP_PCB_UNLOCK(unp2);
1407 UNP_PCB_UNLOCK(unp);
1416 unp->unp_flags &= ~UNP_CONNECTING;
1417 UNP_PCB_UNLOCK(unp);
1422 unp_connect2(struct socket *so, struct socket *so2, int req)
1427 unp = sotounpcb(so);
1428 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1429 unp2 = sotounpcb(so2);
1430 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1432 UNP_LINK_WLOCK_ASSERT();
1433 UNP_PCB_LOCK_ASSERT(unp);
1434 UNP_PCB_LOCK_ASSERT(unp2);
1436 if (so2->so_type != so->so_type)
1437 return (EPROTOTYPE);
1438 unp->unp_conn = unp2;
1440 switch (so->so_type) {
1442 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1447 case SOCK_SEQPACKET:
1448 unp2->unp_conn = unp;
1449 if (req == PRU_CONNECT &&
1450 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1458 panic("unp_connect2");
1464 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1468 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1470 UNP_LINK_WLOCK_ASSERT();
1471 UNP_PCB_LOCK_ASSERT(unp);
1472 UNP_PCB_LOCK_ASSERT(unp2);
1474 unp->unp_conn = NULL;
1475 switch (unp->unp_socket->so_type) {
1477 LIST_REMOVE(unp, unp_reflink);
1478 so = unp->unp_socket;
1480 so->so_state &= ~SS_ISCONNECTED;
1485 case SOCK_SEQPACKET:
1486 soisdisconnected(unp->unp_socket);
1487 unp2->unp_conn = NULL;
1488 soisdisconnected(unp2->unp_socket);
1494 * unp_pcblist() walks the global list of struct unpcb's to generate a
1495 * pointer list, bumping the refcount on each unpcb. It then copies them out
1496 * sequentially, validating the generation number on each to see if it has
1497 * been detached. All of this is necessary because copyout() may sleep on
1501 unp_pcblist(SYSCTL_HANDLER_ARGS)
1505 struct unpcb *unp, **unp_list;
1507 struct xunpgen *xug;
1508 struct unp_head *head;
1511 switch ((intptr_t)arg1) {
1520 case SOCK_SEQPACKET:
1525 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1529 * The process of preparing the PCB list is too time-consuming and
1530 * resource-intensive to repeat twice on every request.
1532 if (req->oldptr == NULL) {
1534 req->oldidx = 2 * (sizeof *xug)
1535 + (n + n/8) * sizeof(struct xunpcb);
1539 if (req->newptr != NULL)
1543 * OK, now we're committed to doing something.
1545 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1547 gencnt = unp_gencnt;
1551 xug->xug_len = sizeof *xug;
1553 xug->xug_gen = gencnt;
1554 xug->xug_sogen = so_gencnt;
1555 error = SYSCTL_OUT(req, xug, sizeof *xug);
1561 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1564 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1565 unp = LIST_NEXT(unp, unp_link)) {
1567 if (unp->unp_gencnt <= gencnt) {
1568 if (cr_cansee(req->td->td_ucred,
1569 unp->unp_socket->so_cred)) {
1570 UNP_PCB_UNLOCK(unp);
1573 unp_list[i++] = unp;
1574 unp->unp_refcount++;
1576 UNP_PCB_UNLOCK(unp);
1579 n = i; /* In case we lost some during malloc. */
1582 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1583 for (i = 0; i < n; i++) {
1586 unp->unp_refcount--;
1587 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1588 xu->xu_len = sizeof *xu;
1591 * XXX - need more locking here to protect against
1592 * connect/disconnect races for SMP.
1594 if (unp->unp_addr != NULL)
1595 bcopy(unp->unp_addr, &xu->xu_addr,
1596 unp->unp_addr->sun_len);
1597 if (unp->unp_conn != NULL &&
1598 unp->unp_conn->unp_addr != NULL)
1599 bcopy(unp->unp_conn->unp_addr,
1601 unp->unp_conn->unp_addr->sun_len);
1602 bcopy(unp, &xu->xu_unp, sizeof *unp);
1603 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1604 UNP_PCB_UNLOCK(unp);
1605 error = SYSCTL_OUT(req, xu, sizeof *xu);
1607 freeunp = (unp->unp_refcount == 0);
1608 UNP_PCB_UNLOCK(unp);
1610 UNP_PCB_LOCK_DESTROY(unp);
1611 uma_zfree(unp_zone, unp);
1618 * Give the user an updated idea of our state. If the
1619 * generation differs from what we told her before, she knows
1620 * that something happened while we were processing this
1621 * request, and it might be necessary to retry.
1623 xug->xug_gen = unp_gencnt;
1624 xug->xug_sogen = so_gencnt;
1625 xug->xug_count = unp_count;
1626 error = SYSCTL_OUT(req, xug, sizeof *xug);
1628 free(unp_list, M_TEMP);
1633 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1634 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1635 "List of active local datagram sockets");
1636 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1637 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1638 "List of active local stream sockets");
1639 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1640 CTLTYPE_OPAQUE | CTLFLAG_RD,
1641 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1642 "List of active local seqpacket sockets");
1645 unp_shutdown(struct unpcb *unp)
1650 UNP_LINK_WLOCK_ASSERT();
1651 UNP_PCB_LOCK_ASSERT(unp);
1653 unp2 = unp->unp_conn;
1654 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1655 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1656 so = unp2->unp_socket;
1663 unp_drop(struct unpcb *unp, int errno)
1665 struct socket *so = unp->unp_socket;
1668 UNP_LINK_WLOCK_ASSERT();
1669 UNP_PCB_LOCK_ASSERT(unp);
1671 so->so_error = errno;
1672 unp2 = unp->unp_conn;
1676 unp_disconnect(unp, unp2);
1677 UNP_PCB_UNLOCK(unp2);
1681 unp_freerights(struct filedescent **fdep, int fdcount)
1686 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1688 for (i = 0; i < fdcount; i++) {
1689 fp = fdep[i]->fde_file;
1690 filecaps_free(&fdep[i]->fde_caps);
1693 free(fdep[0], M_FILECAPS);
1697 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1699 struct thread *td = curthread; /* XXX */
1700 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1703 struct filedesc *fdesc = td->td_proc->p_fd;
1704 struct filedescent *fde, **fdep;
1706 socklen_t clen = control->m_len, datalen;
1710 UNP_LINK_UNLOCK_ASSERT();
1713 if (controlp != NULL) /* controlp == NULL => free control messages */
1715 while (cm != NULL) {
1716 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1720 data = CMSG_DATA(cm);
1721 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1722 if (cm->cmsg_level == SOL_SOCKET
1723 && cm->cmsg_type == SCM_RIGHTS) {
1724 newfds = datalen / sizeof(*fdep);
1729 /* If we're not outputting the descriptors free them. */
1730 if (error || controlp == NULL) {
1731 unp_freerights(fdep, newfds);
1734 FILEDESC_XLOCK(fdesc);
1737 * Now change each pointer to an fd in the global
1738 * table to an integer that is the index to the local
1739 * fd table entry that we set up to point to the
1740 * global one we are transferring.
1742 newlen = newfds * sizeof(int);
1743 *controlp = sbcreatecontrol(NULL, newlen,
1744 SCM_RIGHTS, SOL_SOCKET);
1745 if (*controlp == NULL) {
1746 FILEDESC_XUNLOCK(fdesc);
1748 unp_freerights(fdep, newfds);
1753 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1754 if (fdallocn(td, 0, fdp, newfds) != 0) {
1755 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1757 unp_freerights(fdep, newfds);
1762 for (i = 0; i < newfds; i++, fdp++) {
1763 fde = &fdesc->fd_ofiles[*fdp];
1764 fde->fde_file = fdep[i]->fde_file;
1765 filecaps_move(&fdep[i]->fde_caps,
1767 if ((flags & MSG_CMSG_CLOEXEC) != 0)
1768 fde->fde_flags |= UF_EXCLOSE;
1769 unp_externalize_fp(fde->fde_file);
1771 FILEDESC_XUNLOCK(fdesc);
1772 free(fdep[0], M_FILECAPS);
1774 /* We can just copy anything else across. */
1775 if (error || controlp == NULL)
1777 *controlp = sbcreatecontrol(NULL, datalen,
1778 cm->cmsg_type, cm->cmsg_level);
1779 if (*controlp == NULL) {
1784 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1787 controlp = &(*controlp)->m_next;
1790 if (CMSG_SPACE(datalen) < clen) {
1791 clen -= CMSG_SPACE(datalen);
1792 cm = (struct cmsghdr *)
1793 ((caddr_t)cm + CMSG_SPACE(datalen));
1805 unp_zone_change(void *tag)
1808 uma_zone_set_max(unp_zone, maxsockets);
1816 if (!IS_DEFAULT_VNET(curvnet))
1819 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1820 NULL, NULL, UMA_ALIGN_PTR, 0);
1821 if (unp_zone == NULL)
1823 uma_zone_set_max(unp_zone, maxsockets);
1824 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1825 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1826 NULL, EVENTHANDLER_PRI_ANY);
1827 LIST_INIT(&unp_dhead);
1828 LIST_INIT(&unp_shead);
1829 LIST_INIT(&unp_sphead);
1830 SLIST_INIT(&unp_defers);
1831 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1832 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1833 UNP_LINK_LOCK_INIT();
1834 UNP_LIST_LOCK_INIT();
1835 UNP_DEFERRED_LOCK_INIT();
1839 unp_internalize(struct mbuf **controlp, struct thread *td)
1841 struct mbuf *control = *controlp;
1842 struct proc *p = td->td_proc;
1843 struct filedesc *fdesc = p->p_fd;
1845 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1846 struct cmsgcred *cmcred;
1847 struct filedescent *fde, **fdep, *fdev;
1852 socklen_t clen = control->m_len, datalen;
1856 UNP_LINK_UNLOCK_ASSERT();
1860 while (cm != NULL) {
1861 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1862 || cm->cmsg_len > clen) {
1866 data = CMSG_DATA(cm);
1867 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1869 switch (cm->cmsg_type) {
1871 * Fill in credential information.
1874 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1875 SCM_CREDS, SOL_SOCKET);
1876 if (*controlp == NULL) {
1880 cmcred = (struct cmsgcred *)
1881 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1882 cmcred->cmcred_pid = p->p_pid;
1883 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1884 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1885 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1886 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1888 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1889 cmcred->cmcred_groups[i] =
1890 td->td_ucred->cr_groups[i];
1894 oldfds = datalen / sizeof (int);
1898 * Check that all the FDs passed in refer to legal
1899 * files. If not, reject the entire operation.
1902 FILEDESC_SLOCK(fdesc);
1903 for (i = 0; i < oldfds; i++) {
1905 if (fget_locked(fdesc, fd) == NULL) {
1906 FILEDESC_SUNLOCK(fdesc);
1910 fp = fdesc->fd_ofiles[fd].fde_file;
1911 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1912 FILEDESC_SUNLOCK(fdesc);
1920 * Now replace the integer FDs with pointers to the
1921 * file structure and capability rights.
1923 newlen = oldfds * sizeof(fdep[0]);
1924 *controlp = sbcreatecontrol(NULL, newlen,
1925 SCM_RIGHTS, SOL_SOCKET);
1926 if (*controlp == NULL) {
1927 FILEDESC_SUNLOCK(fdesc);
1932 fdep = (struct filedescent **)
1933 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1934 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1936 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1937 fde = &fdesc->fd_ofiles[*fdp];
1939 fdep[i]->fde_file = fde->fde_file;
1940 filecaps_copy(&fde->fde_caps,
1941 &fdep[i]->fde_caps);
1942 unp_internalize_fp(fdep[i]->fde_file);
1944 FILEDESC_SUNLOCK(fdesc);
1948 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1949 SCM_TIMESTAMP, SOL_SOCKET);
1950 if (*controlp == NULL) {
1954 tv = (struct timeval *)
1955 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1960 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
1961 SCM_BINTIME, SOL_SOCKET);
1962 if (*controlp == NULL) {
1966 bt = (struct bintime *)
1967 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1976 controlp = &(*controlp)->m_next;
1977 if (CMSG_SPACE(datalen) < clen) {
1978 clen -= CMSG_SPACE(datalen);
1979 cm = (struct cmsghdr *)
1980 ((caddr_t)cm + CMSG_SPACE(datalen));
1992 static struct mbuf *
1993 unp_addsockcred(struct thread *td, struct mbuf *control)
1995 struct mbuf *m, *n, *n_prev;
1996 struct sockcred *sc;
1997 const struct cmsghdr *cm;
2001 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2002 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2006 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2007 sc->sc_uid = td->td_ucred->cr_ruid;
2008 sc->sc_euid = td->td_ucred->cr_uid;
2009 sc->sc_gid = td->td_ucred->cr_rgid;
2010 sc->sc_egid = td->td_ucred->cr_gid;
2011 sc->sc_ngroups = ngroups;
2012 for (i = 0; i < sc->sc_ngroups; i++)
2013 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2016 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2017 * created SCM_CREDS control message (struct sockcred) has another
2020 if (control != NULL)
2021 for (n = control, n_prev = NULL; n != NULL;) {
2022 cm = mtod(n, struct cmsghdr *);
2023 if (cm->cmsg_level == SOL_SOCKET &&
2024 cm->cmsg_type == SCM_CREDS) {
2026 control = n->m_next;
2028 n_prev->m_next = n->m_next;
2036 /* Prepend it to the head. */
2037 m->m_next = control;
2041 static struct unpcb *
2042 fptounp(struct file *fp)
2046 if (fp->f_type != DTYPE_SOCKET)
2048 if ((so = fp->f_data) == NULL)
2050 if (so->so_proto->pr_domain != &localdomain)
2052 return sotounpcb(so);
2056 unp_discard(struct file *fp)
2058 struct unp_defer *dr;
2060 if (unp_externalize_fp(fp)) {
2061 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2063 UNP_DEFERRED_LOCK();
2064 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2065 UNP_DEFERRED_UNLOCK();
2066 atomic_add_int(&unp_defers_count, 1);
2067 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2069 (void) closef(fp, (struct thread *)NULL);
2073 unp_process_defers(void *arg __unused, int pending)
2075 struct unp_defer *dr;
2076 SLIST_HEAD(, unp_defer) drl;
2081 UNP_DEFERRED_LOCK();
2082 if (SLIST_FIRST(&unp_defers) == NULL) {
2083 UNP_DEFERRED_UNLOCK();
2086 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2087 UNP_DEFERRED_UNLOCK();
2089 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2090 SLIST_REMOVE_HEAD(&drl, ud_link);
2091 closef(dr->ud_fp, NULL);
2095 atomic_add_int(&unp_defers_count, -count);
2100 unp_internalize_fp(struct file *fp)
2105 if ((unp = fptounp(fp)) != NULL) {
2107 unp->unp_msgcount++;
2115 unp_externalize_fp(struct file *fp)
2121 if ((unp = fptounp(fp)) != NULL) {
2122 unp->unp_msgcount--;
2132 * unp_defer indicates whether additional work has been defered for a future
2133 * pass through unp_gc(). It is thread local and does not require explicit
2136 static int unp_marked;
2137 static int unp_unreachable;
2140 unp_accessable(struct filedescent **fdep, int fdcount)
2146 for (i = 0; i < fdcount; i++) {
2147 fp = fdep[i]->fde_file;
2148 if ((unp = fptounp(fp)) == NULL)
2150 if (unp->unp_gcflag & UNPGC_REF)
2152 unp->unp_gcflag &= ~UNPGC_DEAD;
2153 unp->unp_gcflag |= UNPGC_REF;
2159 unp_gc_process(struct unpcb *unp)
2165 /* Already processed. */
2166 if (unp->unp_gcflag & UNPGC_SCANNED)
2171 * Check for a socket potentially in a cycle. It must be in a
2172 * queue as indicated by msgcount, and this must equal the file
2173 * reference count. Note that when msgcount is 0 the file is NULL.
2175 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2176 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2177 unp->unp_gcflag |= UNPGC_DEAD;
2183 * Mark all sockets we reference with RIGHTS.
2185 so = unp->unp_socket;
2186 SOCKBUF_LOCK(&so->so_rcv);
2187 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2188 SOCKBUF_UNLOCK(&so->so_rcv);
2191 * Mark all sockets in our accept queue.
2194 TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2195 SOCKBUF_LOCK(&soa->so_rcv);
2196 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2197 SOCKBUF_UNLOCK(&soa->so_rcv);
2200 unp->unp_gcflag |= UNPGC_SCANNED;
2203 static int unp_recycled;
2204 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2205 "Number of unreachable sockets claimed by the garbage collector.");
2207 static int unp_taskcount;
2208 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2209 "Number of times the garbage collector has run.");
2212 unp_gc(__unused void *arg, int pending)
2214 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2216 struct unp_head **head;
2217 struct file *f, **unref;
2224 * First clear all gc flags from previous runs.
2226 for (head = heads; *head != NULL; head++)
2227 LIST_FOREACH(unp, *head, unp_link)
2228 unp->unp_gcflag = 0;
2231 * Scan marking all reachable sockets with UNPGC_REF. Once a socket
2232 * is reachable all of the sockets it references are reachable.
2233 * Stop the scan once we do a complete loop without discovering
2234 * a new reachable socket.
2237 unp_unreachable = 0;
2239 for (head = heads; *head != NULL; head++)
2240 LIST_FOREACH(unp, *head, unp_link)
2241 unp_gc_process(unp);
2242 } while (unp_marked);
2244 if (unp_unreachable == 0)
2248 * Allocate space for a local list of dead unpcbs.
2250 unref = malloc(unp_unreachable * sizeof(struct file *),
2254 * Iterate looking for sockets which have been specifically marked
2255 * as as unreachable and store them locally.
2259 for (total = 0, head = heads; *head != NULL; head++)
2260 LIST_FOREACH(unp, *head, unp_link)
2261 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2263 if (unp->unp_msgcount == 0 || f == NULL ||
2264 f->f_count != unp->unp_msgcount)
2268 KASSERT(total <= unp_unreachable,
2269 ("unp_gc: incorrect unreachable count."));
2275 * Now flush all sockets, free'ing rights. This will free the
2276 * struct files associated with these sockets but leave each socket
2277 * with one remaining ref.
2279 for (i = 0; i < total; i++) {
2282 so = unref[i]->f_data;
2283 CURVNET_SET(so->so_vnet);
2289 * And finally release the sockets so they can be reclaimed.
2291 for (i = 0; i < total; i++)
2292 fdrop(unref[i], NULL);
2293 unp_recycled += total;
2294 free(unref, M_TEMP);
2298 unp_dispose(struct mbuf *m)
2302 unp_scan(m, unp_freerights);
2306 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2311 socklen_t clen, datalen;
2313 while (m0 != NULL) {
2314 for (m = m0; m; m = m->m_next) {
2315 if (m->m_type != MT_CONTROL)
2318 cm = mtod(m, struct cmsghdr *);
2321 while (cm != NULL) {
2322 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2325 data = CMSG_DATA(cm);
2326 datalen = (caddr_t)cm + cm->cmsg_len
2329 if (cm->cmsg_level == SOL_SOCKET &&
2330 cm->cmsg_type == SCM_RIGHTS) {
2331 (*op)(data, datalen /
2332 sizeof(struct filedescent *));
2335 if (CMSG_SPACE(datalen) < clen) {
2336 clen -= CMSG_SPACE(datalen);
2337 cm = (struct cmsghdr *)
2338 ((caddr_t)cm + CMSG_SPACE(datalen));
2350 * A helper function called by VFS before socket-type vnode reclamation.
2351 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2355 vfs_unp_reclaim(struct vnode *vp)
2361 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2362 KASSERT(vp->v_type == VSOCK,
2363 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2367 VOP_UNP_CONNECT(vp, &so);
2370 unp = sotounpcb(so);
2374 if (unp->unp_vnode == vp) {
2376 unp->unp_vnode = NULL;
2379 UNP_PCB_UNLOCK(unp);
2388 db_print_indent(int indent)
2392 for (i = 0; i < indent; i++)
2397 db_print_unpflags(int unp_flags)
2402 if (unp_flags & UNP_HAVEPC) {
2403 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2406 if (unp_flags & UNP_HAVEPCCACHED) {
2407 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2410 if (unp_flags & UNP_WANTCRED) {
2411 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2414 if (unp_flags & UNP_CONNWAIT) {
2415 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2418 if (unp_flags & UNP_CONNECTING) {
2419 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2422 if (unp_flags & UNP_BINDING) {
2423 db_printf("%sUNP_BINDING", comma ? ", " : "");
2429 db_print_xucred(int indent, struct xucred *xu)
2433 db_print_indent(indent);
2434 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
2435 xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2436 db_print_indent(indent);
2437 db_printf("cr_groups: ");
2439 for (i = 0; i < xu->cr_ngroups; i++) {
2440 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2447 db_print_unprefs(int indent, struct unp_head *uh)
2453 LIST_FOREACH(unp, uh, unp_reflink) {
2454 if (counter % 4 == 0)
2455 db_print_indent(indent);
2456 db_printf("%p ", unp);
2457 if (counter % 4 == 3)
2461 if (counter != 0 && counter % 4 != 0)
2465 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2470 db_printf("usage: show unpcb <addr>\n");
2473 unp = (struct unpcb *)addr;
2475 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2478 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2481 db_printf("unp_refs:\n");
2482 db_print_unprefs(2, &unp->unp_refs);
2484 /* XXXRW: Would be nice to print the full address, if any. */
2485 db_printf("unp_addr: %p\n", unp->unp_addr);
2487 db_printf("unp_gencnt: %llu\n",
2488 (unsigned long long)unp->unp_gencnt);
2490 db_printf("unp_flags: %x (", unp->unp_flags);
2491 db_print_unpflags(unp->unp_flags);
2494 db_printf("unp_peercred:\n");
2495 db_print_xucred(2, &unp->unp_peercred);
2497 db_printf("unp_refcount: %u\n", unp->unp_refcount);