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_dispose_so(struct socket *so);
282 static void unp_shutdown(struct unpcb *);
283 static void unp_drop(struct unpcb *);
284 static void unp_gc(__unused void *, int);
285 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
286 static void unp_discard(struct file *);
287 static void unp_freerights(struct filedescent **, int);
288 static void unp_init(void);
289 static int unp_internalize(struct mbuf **, struct thread *);
290 static void unp_internalize_fp(struct file *);
291 static int unp_externalize(struct mbuf *, struct mbuf **, int);
292 static int unp_externalize_fp(struct file *);
293 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
294 static void unp_process_defers(void * __unused, int);
297 * Definitions of protocols supported in the LOCAL domain.
299 static struct domain localdomain;
300 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
301 static struct pr_usrreqs uipc_usrreqs_seqpacket;
302 static struct protosw localsw[] = {
304 .pr_type = SOCK_STREAM,
305 .pr_domain = &localdomain,
306 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
307 .pr_ctloutput = &uipc_ctloutput,
308 .pr_usrreqs = &uipc_usrreqs_stream
311 .pr_type = SOCK_DGRAM,
312 .pr_domain = &localdomain,
313 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
314 .pr_ctloutput = &uipc_ctloutput,
315 .pr_usrreqs = &uipc_usrreqs_dgram
318 .pr_type = SOCK_SEQPACKET,
319 .pr_domain = &localdomain,
322 * XXXRW: For now, PR_ADDR because soreceive will bump into them
323 * due to our use of sbappendaddr. A new sbappend variants is needed
324 * that supports both atomic record writes and control data.
326 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
328 .pr_ctloutput = &uipc_ctloutput,
329 .pr_usrreqs = &uipc_usrreqs_seqpacket,
333 static struct domain localdomain = {
334 .dom_family = AF_LOCAL,
336 .dom_init = unp_init,
337 .dom_externalize = unp_externalize,
338 .dom_dispose = unp_dispose_so,
339 .dom_protosw = localsw,
340 .dom_protoswNPROTOSW = &localsw[nitems(localsw)]
345 uipc_abort(struct socket *so)
347 struct unpcb *unp, *unp2;
350 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
354 unp2 = unp->unp_conn;
358 UNP_PCB_UNLOCK(unp2);
365 uipc_accept(struct socket *so, struct sockaddr **nam)
367 struct unpcb *unp, *unp2;
368 const struct sockaddr *sa;
371 * Pass back name of connected socket, if it was bound and we are
372 * still connected (our peer may have closed already!).
375 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
377 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
379 unp2 = unp->unp_conn;
380 if (unp2 != NULL && unp2->unp_addr != NULL) {
382 sa = (struct sockaddr *) unp2->unp_addr;
383 bcopy(sa, *nam, sa->sa_len);
384 UNP_PCB_UNLOCK(unp2);
387 bcopy(sa, *nam, sa->sa_len);
394 uipc_attach(struct socket *so, int proto, struct thread *td)
396 u_long sendspace, recvspace;
400 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
401 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
402 switch (so->so_type) {
404 sendspace = unpst_sendspace;
405 recvspace = unpst_recvspace;
409 sendspace = unpdg_sendspace;
410 recvspace = unpdg_recvspace;
414 sendspace = unpsp_sendspace;
415 recvspace = unpsp_recvspace;
419 panic("uipc_attach");
421 error = soreserve(so, sendspace, recvspace);
425 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
428 LIST_INIT(&unp->unp_refs);
429 UNP_PCB_LOCK_INIT(unp);
430 unp->unp_socket = so;
432 unp->unp_refcount = 1;
435 unp->unp_gencnt = ++unp_gencnt;
437 switch (so->so_type) {
439 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
443 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
447 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
451 panic("uipc_attach");
459 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
461 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
471 if (nam->sa_family != AF_UNIX)
472 return (EAFNOSUPPORT);
475 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
477 if (soun->sun_len > sizeof(struct sockaddr_un))
479 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
484 * We don't allow simultaneous bind() calls on a single UNIX domain
485 * socket, so flag in-progress operations, and return an error if an
486 * operation is already in progress.
488 * Historically, we have not allowed a socket to be rebound, so this
489 * also returns an error. Not allowing re-binding simplifies the
490 * implementation and avoids a great many possible failure modes.
493 if (unp->unp_vnode != NULL) {
497 if (unp->unp_flags & UNP_BINDING) {
501 unp->unp_flags |= UNP_BINDING;
504 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
505 bcopy(soun->sun_path, buf, namelen);
509 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
510 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
511 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
516 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
517 NDFREE(&nd, NDF_ONLY_PNBUF);
527 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
533 vattr.va_type = VSOCK;
534 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
536 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
540 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
541 NDFREE(&nd, NDF_ONLY_PNBUF);
544 vn_finished_write(mp);
548 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
549 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
553 VOP_UNP_BIND(vp, unp->unp_socket);
555 unp->unp_addr = soun;
556 unp->unp_flags &= ~UNP_BINDING;
560 vn_finished_write(mp);
566 unp->unp_flags &= ~UNP_BINDING;
573 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
576 return (uipc_bindat(AT_FDCWD, so, nam, td));
580 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
584 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
586 error = unp_connect(so, nam, td);
592 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
597 KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
599 error = unp_connectat(fd, so, nam, td);
605 uipc_close(struct socket *so)
607 struct unpcb *unp, *unp2;
610 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
614 unp2 = unp->unp_conn;
617 unp_disconnect(unp, unp2);
618 UNP_PCB_UNLOCK(unp2);
625 uipc_connect2(struct socket *so1, struct socket *so2)
627 struct unpcb *unp, *unp2;
632 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
635 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
637 error = unp_connect2(so1, so2, PRU_CONNECT2);
638 UNP_PCB_UNLOCK(unp2);
645 uipc_detach(struct socket *so)
647 struct unpcb *unp, *unp2;
648 struct sockaddr_un *saved_unp_addr;
650 int freeunp, local_unp_rights;
653 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
658 LIST_REMOVE(unp, unp_link);
659 unp->unp_gencnt = ++unp_gencnt;
664 * XXXRW: Should assert vp->v_socket == so.
666 if ((vp = unp->unp_vnode) != NULL) {
668 unp->unp_vnode = NULL;
670 unp2 = unp->unp_conn;
673 unp_disconnect(unp, unp2);
674 UNP_PCB_UNLOCK(unp2);
678 * We hold the linkage lock exclusively, so it's OK to acquire
679 * multiple pcb locks at a time.
681 while (!LIST_EMPTY(&unp->unp_refs)) {
682 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
688 local_unp_rights = unp_rights;
690 unp->unp_socket->so_pcb = NULL;
691 saved_unp_addr = unp->unp_addr;
692 unp->unp_addr = NULL;
694 freeunp = (unp->unp_refcount == 0);
695 if (saved_unp_addr != NULL)
696 free(saved_unp_addr, M_SONAME);
698 UNP_PCB_LOCK_DESTROY(unp);
699 uma_zfree(unp_zone, unp);
704 if (local_unp_rights)
705 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
709 uipc_disconnect(struct socket *so)
711 struct unpcb *unp, *unp2;
714 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
718 unp2 = unp->unp_conn;
721 unp_disconnect(unp, unp2);
722 UNP_PCB_UNLOCK(unp2);
730 uipc_listen(struct socket *so, int backlog, struct thread *td)
736 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
739 if (unp->unp_vnode == NULL) {
740 /* Already connected or not bound to an address. */
741 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
747 error = solisten_proto_check(so);
749 cru2x(td->td_ucred, &unp->unp_peercred);
750 unp->unp_flags |= UNP_HAVEPCCACHED;
751 solisten_proto(so, backlog);
759 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
761 struct unpcb *unp, *unp2;
762 const struct sockaddr *sa;
765 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
767 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
770 * XXX: It seems that this test always fails even when connection is
771 * established. So, this else clause is added as workaround to
772 * return PF_LOCAL sockaddr.
774 unp2 = unp->unp_conn;
777 if (unp2->unp_addr != NULL)
778 sa = (struct sockaddr *) unp2->unp_addr;
781 bcopy(sa, *nam, sa->sa_len);
782 UNP_PCB_UNLOCK(unp2);
785 bcopy(sa, *nam, sa->sa_len);
792 uipc_rcvd(struct socket *so, int flags)
794 struct unpcb *unp, *unp2;
799 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
800 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
801 ("%s: socktype %d", __func__, so->so_type));
804 * Adjust backpressure on sender and wakeup any waiting to write.
806 * The unp lock is acquired to maintain the validity of the unp_conn
807 * pointer; no lock on unp2 is required as unp2->unp_socket will be
808 * static as long as we don't permit unp2 to disconnect from unp,
809 * which is prevented by the lock on unp. We cache values from
810 * so_rcv to avoid holding the so_rcv lock over the entire
811 * transaction on the remote so_snd.
813 SOCKBUF_LOCK(&so->so_rcv);
814 mbcnt = so->so_rcv.sb_mbcnt;
815 sbcc = sbavail(&so->so_rcv);
816 SOCKBUF_UNLOCK(&so->so_rcv);
818 * There is a benign race condition at this point. If we're planning to
819 * clear SB_STOP, but uipc_send is called on the connected socket at
820 * this instant, it might add data to the sockbuf and set SB_STOP. Then
821 * we would erroneously clear SB_STOP below, even though the sockbuf is
822 * full. The race is benign because the only ill effect is to allow the
823 * sockbuf to exceed its size limit, and the size limits are not
824 * strictly guaranteed anyway.
827 unp2 = unp->unp_conn;
832 so2 = unp2->unp_socket;
833 SOCKBUF_LOCK(&so2->so_snd);
834 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
835 so2->so_snd.sb_flags &= ~SB_STOP;
836 sowwakeup_locked(so2);
842 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
843 struct mbuf *control, struct thread *td)
845 struct unpcb *unp, *unp2;
851 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
852 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
853 so->so_type == SOCK_SEQPACKET,
854 ("%s: socktype %d", __func__, so->so_type));
856 if (flags & PRUS_OOB) {
860 if (control != NULL && (error = unp_internalize(&control, td)))
862 if ((nam != NULL) || (flags & PRUS_EOF))
866 switch (so->so_type) {
869 const struct sockaddr *from;
871 unp2 = unp->unp_conn;
873 UNP_LINK_WLOCK_ASSERT();
878 error = unp_connect(so, nam, td);
881 unp2 = unp->unp_conn;
885 * Because connect() and send() are non-atomic in a sendto()
886 * with a target address, it's possible that the socket will
887 * have disconnected before the send() can run. In that case
888 * return the slightly counter-intuitive but otherwise
889 * correct error that the socket is not connected.
896 if (unp2->unp_flags & UNP_WANTCRED)
897 control = unp_addsockcred(td, control);
899 if (unp->unp_addr != NULL)
900 from = (struct sockaddr *)unp->unp_addr;
903 so2 = unp2->unp_socket;
904 SOCKBUF_LOCK(&so2->so_rcv);
905 if (sbappendaddr_locked(&so2->so_rcv, from, m,
907 sorwakeup_locked(so2);
911 SOCKBUF_UNLOCK(&so2->so_rcv);
915 UNP_LINK_WLOCK_ASSERT();
917 unp_disconnect(unp, unp2);
918 UNP_PCB_UNLOCK(unp2);
926 if ((so->so_state & SS_ISCONNECTED) == 0) {
928 UNP_LINK_WLOCK_ASSERT();
929 error = unp_connect(so, nam, td);
939 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
945 * Because connect() and send() are non-atomic in a sendto()
946 * with a target address, it's possible that the socket will
947 * have disconnected before the send() can run. In that case
948 * return the slightly counter-intuitive but otherwise
949 * correct error that the socket is not connected.
951 * Locking here must be done carefully: the linkage lock
952 * prevents interconnections between unpcbs from changing, so
953 * we can traverse from unp to unp2 without acquiring unp's
954 * lock. Socket buffer locks follow unpcb locks, so we can
955 * acquire both remote and lock socket buffer locks.
957 unp2 = unp->unp_conn;
962 so2 = unp2->unp_socket;
964 SOCKBUF_LOCK(&so2->so_rcv);
965 if (unp2->unp_flags & UNP_WANTCRED) {
967 * Credentials are passed only once on SOCK_STREAM
968 * and SOCK_SEQPACKET.
970 unp2->unp_flags &= ~UNP_WANTCRED;
971 control = unp_addsockcred(td, control);
974 * Send to paired receive port, and then reduce send buffer
975 * hiwater marks to maintain backpressure. Wake up readers.
977 switch (so->so_type) {
979 if (control != NULL) {
980 if (sbappendcontrol_locked(&so2->so_rcv, m,
984 sbappend_locked(&so2->so_rcv, m, flags);
987 case SOCK_SEQPACKET: {
988 const struct sockaddr *from;
992 * Don't check for space available in so2->so_rcv.
993 * Unix domain sockets only check for space in the
994 * sending sockbuf, and that check is performed one
995 * level up the stack.
997 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1004 mbcnt = so2->so_rcv.sb_mbcnt;
1005 sbcc = sbavail(&so2->so_rcv);
1007 sorwakeup_locked(so2);
1009 SOCKBUF_UNLOCK(&so2->so_rcv);
1012 * The PCB lock on unp2 protects the SB_STOP flag. Without it,
1013 * it would be possible for uipc_rcvd to be called at this
1014 * point, drain the receiving sockbuf, clear SB_STOP, and then
1015 * we would set SB_STOP below. That could lead to an empty
1016 * sockbuf having SB_STOP set
1018 SOCKBUF_LOCK(&so->so_snd);
1019 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1020 so->so_snd.sb_flags |= SB_STOP;
1021 SOCKBUF_UNLOCK(&so->so_snd);
1022 UNP_PCB_UNLOCK(unp2);
1028 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1030 if (flags & PRUS_EOF) {
1034 UNP_PCB_UNLOCK(unp);
1037 if ((nam != NULL) || (flags & PRUS_EOF))
1042 if (control != NULL && error != 0)
1043 unp_dispose(control);
1046 if (control != NULL)
1054 uipc_ready(struct socket *so, struct mbuf *m, int count)
1056 struct unpcb *unp, *unp2;
1060 unp = sotounpcb(so);
1063 unp2 = unp->unp_conn;
1065 so2 = unp2->unp_socket;
1067 SOCKBUF_LOCK(&so2->so_rcv);
1068 if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1069 sorwakeup_locked(so2);
1071 SOCKBUF_UNLOCK(&so2->so_rcv);
1073 UNP_PCB_UNLOCK(unp2);
1080 uipc_sense(struct socket *so, struct stat *sb)
1084 unp = sotounpcb(so);
1085 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1087 sb->st_blksize = so->so_snd.sb_hiwat;
1090 if (unp->unp_ino == 0)
1091 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1092 sb->st_ino = unp->unp_ino;
1093 UNP_PCB_UNLOCK(unp);
1098 uipc_shutdown(struct socket *so)
1102 unp = sotounpcb(so);
1103 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1109 UNP_PCB_UNLOCK(unp);
1115 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1118 const struct sockaddr *sa;
1120 unp = sotounpcb(so);
1121 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1123 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1125 if (unp->unp_addr != NULL)
1126 sa = (struct sockaddr *) unp->unp_addr;
1129 bcopy(sa, *nam, sa->sa_len);
1130 UNP_PCB_UNLOCK(unp);
1134 static struct pr_usrreqs uipc_usrreqs_dgram = {
1135 .pru_abort = uipc_abort,
1136 .pru_accept = uipc_accept,
1137 .pru_attach = uipc_attach,
1138 .pru_bind = uipc_bind,
1139 .pru_bindat = uipc_bindat,
1140 .pru_connect = uipc_connect,
1141 .pru_connectat = uipc_connectat,
1142 .pru_connect2 = uipc_connect2,
1143 .pru_detach = uipc_detach,
1144 .pru_disconnect = uipc_disconnect,
1145 .pru_listen = uipc_listen,
1146 .pru_peeraddr = uipc_peeraddr,
1147 .pru_rcvd = uipc_rcvd,
1148 .pru_send = uipc_send,
1149 .pru_sense = uipc_sense,
1150 .pru_shutdown = uipc_shutdown,
1151 .pru_sockaddr = uipc_sockaddr,
1152 .pru_soreceive = soreceive_dgram,
1153 .pru_close = uipc_close,
1156 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1157 .pru_abort = uipc_abort,
1158 .pru_accept = uipc_accept,
1159 .pru_attach = uipc_attach,
1160 .pru_bind = uipc_bind,
1161 .pru_bindat = uipc_bindat,
1162 .pru_connect = uipc_connect,
1163 .pru_connectat = uipc_connectat,
1164 .pru_connect2 = uipc_connect2,
1165 .pru_detach = uipc_detach,
1166 .pru_disconnect = uipc_disconnect,
1167 .pru_listen = uipc_listen,
1168 .pru_peeraddr = uipc_peeraddr,
1169 .pru_rcvd = uipc_rcvd,
1170 .pru_send = uipc_send,
1171 .pru_sense = uipc_sense,
1172 .pru_shutdown = uipc_shutdown,
1173 .pru_sockaddr = uipc_sockaddr,
1174 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1175 .pru_close = uipc_close,
1178 static struct pr_usrreqs uipc_usrreqs_stream = {
1179 .pru_abort = uipc_abort,
1180 .pru_accept = uipc_accept,
1181 .pru_attach = uipc_attach,
1182 .pru_bind = uipc_bind,
1183 .pru_bindat = uipc_bindat,
1184 .pru_connect = uipc_connect,
1185 .pru_connectat = uipc_connectat,
1186 .pru_connect2 = uipc_connect2,
1187 .pru_detach = uipc_detach,
1188 .pru_disconnect = uipc_disconnect,
1189 .pru_listen = uipc_listen,
1190 .pru_peeraddr = uipc_peeraddr,
1191 .pru_rcvd = uipc_rcvd,
1192 .pru_send = uipc_send,
1193 .pru_ready = uipc_ready,
1194 .pru_sense = uipc_sense,
1195 .pru_shutdown = uipc_shutdown,
1196 .pru_sockaddr = uipc_sockaddr,
1197 .pru_soreceive = soreceive_generic,
1198 .pru_close = uipc_close,
1202 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1208 if (sopt->sopt_level != 0)
1211 unp = sotounpcb(so);
1212 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1214 switch (sopt->sopt_dir) {
1216 switch (sopt->sopt_name) {
1217 case LOCAL_PEERCRED:
1219 if (unp->unp_flags & UNP_HAVEPC)
1220 xu = unp->unp_peercred;
1222 if (so->so_type == SOCK_STREAM)
1227 UNP_PCB_UNLOCK(unp);
1229 error = sooptcopyout(sopt, &xu, sizeof(xu));
1233 /* Unlocked read. */
1234 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1235 error = sooptcopyout(sopt, &optval, sizeof(optval));
1238 case LOCAL_CONNWAIT:
1239 /* Unlocked read. */
1240 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1241 error = sooptcopyout(sopt, &optval, sizeof(optval));
1251 switch (sopt->sopt_name) {
1253 case LOCAL_CONNWAIT:
1254 error = sooptcopyin(sopt, &optval, sizeof(optval),
1259 #define OPTSET(bit) do { \
1260 UNP_PCB_LOCK(unp); \
1262 unp->unp_flags |= bit; \
1264 unp->unp_flags &= ~bit; \
1265 UNP_PCB_UNLOCK(unp); \
1268 switch (sopt->sopt_name) {
1270 OPTSET(UNP_WANTCRED);
1273 case LOCAL_CONNWAIT:
1274 OPTSET(UNP_CONNWAIT);
1283 error = ENOPROTOOPT;
1296 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1299 return (unp_connectat(AT_FDCWD, so, nam, td));
1303 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1306 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1308 struct socket *so2, *so3;
1309 struct unpcb *unp, *unp2, *unp3;
1310 struct nameidata nd;
1311 char buf[SOCK_MAXADDRLEN];
1312 struct sockaddr *sa;
1313 cap_rights_t rights;
1316 if (nam->sa_family != AF_UNIX)
1317 return (EAFNOSUPPORT);
1319 UNP_LINK_WLOCK_ASSERT();
1321 unp = sotounpcb(so);
1322 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1324 if (nam->sa_len > sizeof(struct sockaddr_un))
1326 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1329 bcopy(soun->sun_path, buf, len);
1333 if (unp->unp_flags & UNP_CONNECTING) {
1334 UNP_PCB_UNLOCK(unp);
1338 unp->unp_flags |= UNP_CONNECTING;
1339 UNP_PCB_UNLOCK(unp);
1341 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1342 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1343 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1349 ASSERT_VOP_LOCKED(vp, "unp_connect");
1350 NDFREE(&nd, NDF_ONLY_PNBUF);
1354 if (vp->v_type != VSOCK) {
1359 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1363 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1367 unp = sotounpcb(so);
1368 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1371 * Lock linkage lock for two reasons: make sure v_socket is stable,
1372 * and to protect simultaneous locking of multiple pcbs.
1375 VOP_UNP_CONNECT(vp, &so2);
1377 error = ECONNREFUSED;
1380 if (so->so_type != so2->so_type) {
1384 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1385 if (so2->so_options & SO_ACCEPTCONN) {
1386 CURVNET_SET(so2->so_vnet);
1387 so3 = sonewconn(so2, 0);
1392 error = ECONNREFUSED;
1395 unp = sotounpcb(so);
1396 unp2 = sotounpcb(so2);
1397 unp3 = sotounpcb(so3);
1401 if (unp2->unp_addr != NULL) {
1402 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1403 unp3->unp_addr = (struct sockaddr_un *) sa;
1408 * The connector's (client's) credentials are copied from its
1409 * process structure at the time of connect() (which is now).
1411 cru2x(td->td_ucred, &unp3->unp_peercred);
1412 unp3->unp_flags |= UNP_HAVEPC;
1415 * The receiver's (server's) credentials are copied from the
1416 * unp_peercred member of socket on which the former called
1417 * listen(); uipc_listen() cached that process's credentials
1418 * at that time so we can use them now.
1420 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1421 ("unp_connect: listener without cached peercred"));
1422 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1423 sizeof(unp->unp_peercred));
1424 unp->unp_flags |= UNP_HAVEPC;
1425 if (unp2->unp_flags & UNP_WANTCRED)
1426 unp3->unp_flags |= UNP_WANTCRED;
1427 UNP_PCB_UNLOCK(unp3);
1428 UNP_PCB_UNLOCK(unp2);
1429 UNP_PCB_UNLOCK(unp);
1431 mac_socketpeer_set_from_socket(so, so3);
1432 mac_socketpeer_set_from_socket(so3, so);
1437 unp = sotounpcb(so);
1438 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1439 unp2 = sotounpcb(so2);
1440 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1443 error = unp_connect2(so, so2, PRU_CONNECT);
1444 UNP_PCB_UNLOCK(unp2);
1445 UNP_PCB_UNLOCK(unp);
1454 unp->unp_flags &= ~UNP_CONNECTING;
1455 UNP_PCB_UNLOCK(unp);
1460 unp_connect2(struct socket *so, struct socket *so2, int req)
1465 unp = sotounpcb(so);
1466 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1467 unp2 = sotounpcb(so2);
1468 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1470 UNP_LINK_WLOCK_ASSERT();
1471 UNP_PCB_LOCK_ASSERT(unp);
1472 UNP_PCB_LOCK_ASSERT(unp2);
1474 if (so2->so_type != so->so_type)
1475 return (EPROTOTYPE);
1476 unp->unp_conn = unp2;
1478 switch (so->so_type) {
1480 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1485 case SOCK_SEQPACKET:
1486 unp2->unp_conn = unp;
1487 if (req == PRU_CONNECT &&
1488 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1496 panic("unp_connect2");
1502 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1506 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1508 UNP_LINK_WLOCK_ASSERT();
1509 UNP_PCB_LOCK_ASSERT(unp);
1510 UNP_PCB_LOCK_ASSERT(unp2);
1512 unp->unp_conn = NULL;
1513 switch (unp->unp_socket->so_type) {
1515 LIST_REMOVE(unp, unp_reflink);
1516 so = unp->unp_socket;
1518 so->so_state &= ~SS_ISCONNECTED;
1523 case SOCK_SEQPACKET:
1524 soisdisconnected(unp->unp_socket);
1525 unp2->unp_conn = NULL;
1526 soisdisconnected(unp2->unp_socket);
1532 * unp_pcblist() walks the global list of struct unpcb's to generate a
1533 * pointer list, bumping the refcount on each unpcb. It then copies them out
1534 * sequentially, validating the generation number on each to see if it has
1535 * been detached. All of this is necessary because copyout() may sleep on
1539 unp_pcblist(SYSCTL_HANDLER_ARGS)
1543 struct unpcb *unp, **unp_list;
1545 struct xunpgen *xug;
1546 struct unp_head *head;
1549 switch ((intptr_t)arg1) {
1558 case SOCK_SEQPACKET:
1563 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1567 * The process of preparing the PCB list is too time-consuming and
1568 * resource-intensive to repeat twice on every request.
1570 if (req->oldptr == NULL) {
1572 req->oldidx = 2 * (sizeof *xug)
1573 + (n + n/8) * sizeof(struct xunpcb);
1577 if (req->newptr != NULL)
1581 * OK, now we're committed to doing something.
1583 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1585 gencnt = unp_gencnt;
1589 xug->xug_len = sizeof *xug;
1591 xug->xug_gen = gencnt;
1592 xug->xug_sogen = so_gencnt;
1593 error = SYSCTL_OUT(req, xug, sizeof *xug);
1599 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1602 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1603 unp = LIST_NEXT(unp, unp_link)) {
1605 if (unp->unp_gencnt <= gencnt) {
1606 if (cr_cansee(req->td->td_ucred,
1607 unp->unp_socket->so_cred)) {
1608 UNP_PCB_UNLOCK(unp);
1611 unp_list[i++] = unp;
1612 unp->unp_refcount++;
1614 UNP_PCB_UNLOCK(unp);
1617 n = i; /* In case we lost some during malloc. */
1620 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1621 for (i = 0; i < n; i++) {
1624 unp->unp_refcount--;
1625 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1626 xu->xu_len = sizeof *xu;
1629 * XXX - need more locking here to protect against
1630 * connect/disconnect races for SMP.
1632 if (unp->unp_addr != NULL)
1633 bcopy(unp->unp_addr, &xu->xu_addr,
1634 unp->unp_addr->sun_len);
1635 if (unp->unp_conn != NULL &&
1636 unp->unp_conn->unp_addr != NULL)
1637 bcopy(unp->unp_conn->unp_addr,
1639 unp->unp_conn->unp_addr->sun_len);
1640 bcopy(unp, &xu->xu_unp, sizeof *unp);
1641 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1642 UNP_PCB_UNLOCK(unp);
1643 error = SYSCTL_OUT(req, xu, sizeof *xu);
1645 freeunp = (unp->unp_refcount == 0);
1646 UNP_PCB_UNLOCK(unp);
1648 UNP_PCB_LOCK_DESTROY(unp);
1649 uma_zfree(unp_zone, unp);
1656 * Give the user an updated idea of our state. If the
1657 * generation differs from what we told her before, she knows
1658 * that something happened while we were processing this
1659 * request, and it might be necessary to retry.
1661 xug->xug_gen = unp_gencnt;
1662 xug->xug_sogen = so_gencnt;
1663 xug->xug_count = unp_count;
1664 error = SYSCTL_OUT(req, xug, sizeof *xug);
1666 free(unp_list, M_TEMP);
1671 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1672 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1673 "List of active local datagram sockets");
1674 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1675 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1676 "List of active local stream sockets");
1677 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1678 CTLTYPE_OPAQUE | CTLFLAG_RD,
1679 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1680 "List of active local seqpacket sockets");
1683 unp_shutdown(struct unpcb *unp)
1688 UNP_LINK_WLOCK_ASSERT();
1689 UNP_PCB_LOCK_ASSERT(unp);
1691 unp2 = unp->unp_conn;
1692 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1693 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1694 so = unp2->unp_socket;
1701 unp_drop(struct unpcb *unp)
1703 struct socket *so = unp->unp_socket;
1706 UNP_LINK_WLOCK_ASSERT();
1707 UNP_PCB_LOCK_ASSERT(unp);
1710 * Regardless of whether the socket's peer dropped the connection
1711 * with this socket by aborting or disconnecting, POSIX requires
1712 * that ECONNRESET is returned.
1714 so->so_error = ECONNRESET;
1715 unp2 = unp->unp_conn;
1719 unp_disconnect(unp, unp2);
1720 UNP_PCB_UNLOCK(unp2);
1724 unp_freerights(struct filedescent **fdep, int fdcount)
1729 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1731 for (i = 0; i < fdcount; i++) {
1732 fp = fdep[i]->fde_file;
1733 filecaps_free(&fdep[i]->fde_caps);
1736 free(fdep[0], M_FILECAPS);
1740 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1742 struct thread *td = curthread; /* XXX */
1743 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1746 struct filedesc *fdesc = td->td_proc->p_fd;
1747 struct filedescent **fdep;
1749 socklen_t clen = control->m_len, datalen;
1753 UNP_LINK_UNLOCK_ASSERT();
1756 if (controlp != NULL) /* controlp == NULL => free control messages */
1758 while (cm != NULL) {
1759 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1763 data = CMSG_DATA(cm);
1764 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1765 if (cm->cmsg_level == SOL_SOCKET
1766 && cm->cmsg_type == SCM_RIGHTS) {
1767 newfds = datalen / sizeof(*fdep);
1772 /* If we're not outputting the descriptors free them. */
1773 if (error || controlp == NULL) {
1774 unp_freerights(fdep, newfds);
1777 FILEDESC_XLOCK(fdesc);
1780 * Now change each pointer to an fd in the global
1781 * table to an integer that is the index to the local
1782 * fd table entry that we set up to point to the
1783 * global one we are transferring.
1785 newlen = newfds * sizeof(int);
1786 *controlp = sbcreatecontrol(NULL, newlen,
1787 SCM_RIGHTS, SOL_SOCKET);
1788 if (*controlp == NULL) {
1789 FILEDESC_XUNLOCK(fdesc);
1791 unp_freerights(fdep, newfds);
1796 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1797 if (fdallocn(td, 0, fdp, newfds) != 0) {
1798 FILEDESC_XUNLOCK(fdesc);
1800 unp_freerights(fdep, newfds);
1805 for (i = 0; i < newfds; i++, fdp++) {
1806 _finstall(fdesc, fdep[i]->fde_file, *fdp,
1807 (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
1808 &fdep[i]->fde_caps);
1809 unp_externalize_fp(fdep[i]->fde_file);
1811 FILEDESC_XUNLOCK(fdesc);
1812 free(fdep[0], M_FILECAPS);
1814 /* We can just copy anything else across. */
1815 if (error || controlp == NULL)
1817 *controlp = sbcreatecontrol(NULL, datalen,
1818 cm->cmsg_type, cm->cmsg_level);
1819 if (*controlp == NULL) {
1824 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1827 controlp = &(*controlp)->m_next;
1830 if (CMSG_SPACE(datalen) < clen) {
1831 clen -= CMSG_SPACE(datalen);
1832 cm = (struct cmsghdr *)
1833 ((caddr_t)cm + CMSG_SPACE(datalen));
1845 unp_zone_change(void *tag)
1848 uma_zone_set_max(unp_zone, maxsockets);
1856 if (!IS_DEFAULT_VNET(curvnet))
1859 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1860 NULL, NULL, UMA_ALIGN_PTR, 0);
1861 if (unp_zone == NULL)
1863 uma_zone_set_max(unp_zone, maxsockets);
1864 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1865 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1866 NULL, EVENTHANDLER_PRI_ANY);
1867 LIST_INIT(&unp_dhead);
1868 LIST_INIT(&unp_shead);
1869 LIST_INIT(&unp_sphead);
1870 SLIST_INIT(&unp_defers);
1871 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1872 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1873 UNP_LINK_LOCK_INIT();
1874 UNP_LIST_LOCK_INIT();
1875 UNP_DEFERRED_LOCK_INIT();
1879 unp_internalize(struct mbuf **controlp, struct thread *td)
1881 struct mbuf *control = *controlp;
1882 struct proc *p = td->td_proc;
1883 struct filedesc *fdesc = p->p_fd;
1885 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1886 struct cmsgcred *cmcred;
1887 struct filedescent *fde, **fdep, *fdev;
1892 socklen_t clen = control->m_len, datalen;
1896 UNP_LINK_UNLOCK_ASSERT();
1900 while (cm != NULL) {
1901 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1902 || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
1906 data = CMSG_DATA(cm);
1907 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1909 switch (cm->cmsg_type) {
1911 * Fill in credential information.
1914 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1915 SCM_CREDS, SOL_SOCKET);
1916 if (*controlp == NULL) {
1920 cmcred = (struct cmsgcred *)
1921 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1922 cmcred->cmcred_pid = p->p_pid;
1923 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1924 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1925 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1926 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1928 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1929 cmcred->cmcred_groups[i] =
1930 td->td_ucred->cr_groups[i];
1934 oldfds = datalen / sizeof (int);
1938 * Check that all the FDs passed in refer to legal
1939 * files. If not, reject the entire operation.
1942 FILEDESC_SLOCK(fdesc);
1943 for (i = 0; i < oldfds; i++, fdp++) {
1944 fp = fget_locked(fdesc, *fdp);
1946 FILEDESC_SUNLOCK(fdesc);
1950 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1951 FILEDESC_SUNLOCK(fdesc);
1959 * Now replace the integer FDs with pointers to the
1960 * file structure and capability rights.
1962 newlen = oldfds * sizeof(fdep[0]);
1963 *controlp = sbcreatecontrol(NULL, newlen,
1964 SCM_RIGHTS, SOL_SOCKET);
1965 if (*controlp == NULL) {
1966 FILEDESC_SUNLOCK(fdesc);
1971 fdep = (struct filedescent **)
1972 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1973 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1975 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1976 fde = &fdesc->fd_ofiles[*fdp];
1978 fdep[i]->fde_file = fde->fde_file;
1979 filecaps_copy(&fde->fde_caps,
1980 &fdep[i]->fde_caps, true);
1981 unp_internalize_fp(fdep[i]->fde_file);
1983 FILEDESC_SUNLOCK(fdesc);
1987 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1988 SCM_TIMESTAMP, SOL_SOCKET);
1989 if (*controlp == NULL) {
1993 tv = (struct timeval *)
1994 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1999 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2000 SCM_BINTIME, SOL_SOCKET);
2001 if (*controlp == NULL) {
2005 bt = (struct bintime *)
2006 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2015 controlp = &(*controlp)->m_next;
2016 if (CMSG_SPACE(datalen) < clen) {
2017 clen -= CMSG_SPACE(datalen);
2018 cm = (struct cmsghdr *)
2019 ((caddr_t)cm + CMSG_SPACE(datalen));
2031 static struct mbuf *
2032 unp_addsockcred(struct thread *td, struct mbuf *control)
2034 struct mbuf *m, *n, *n_prev;
2035 struct sockcred *sc;
2036 const struct cmsghdr *cm;
2040 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2041 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2045 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2046 sc->sc_uid = td->td_ucred->cr_ruid;
2047 sc->sc_euid = td->td_ucred->cr_uid;
2048 sc->sc_gid = td->td_ucred->cr_rgid;
2049 sc->sc_egid = td->td_ucred->cr_gid;
2050 sc->sc_ngroups = ngroups;
2051 for (i = 0; i < sc->sc_ngroups; i++)
2052 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2055 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2056 * created SCM_CREDS control message (struct sockcred) has another
2059 if (control != NULL)
2060 for (n = control, n_prev = NULL; n != NULL;) {
2061 cm = mtod(n, struct cmsghdr *);
2062 if (cm->cmsg_level == SOL_SOCKET &&
2063 cm->cmsg_type == SCM_CREDS) {
2065 control = n->m_next;
2067 n_prev->m_next = n->m_next;
2075 /* Prepend it to the head. */
2076 m->m_next = control;
2080 static struct unpcb *
2081 fptounp(struct file *fp)
2085 if (fp->f_type != DTYPE_SOCKET)
2087 if ((so = fp->f_data) == NULL)
2089 if (so->so_proto->pr_domain != &localdomain)
2091 return sotounpcb(so);
2095 unp_discard(struct file *fp)
2097 struct unp_defer *dr;
2099 if (unp_externalize_fp(fp)) {
2100 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2102 UNP_DEFERRED_LOCK();
2103 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2104 UNP_DEFERRED_UNLOCK();
2105 atomic_add_int(&unp_defers_count, 1);
2106 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2108 (void) closef(fp, (struct thread *)NULL);
2112 unp_process_defers(void *arg __unused, int pending)
2114 struct unp_defer *dr;
2115 SLIST_HEAD(, unp_defer) drl;
2120 UNP_DEFERRED_LOCK();
2121 if (SLIST_FIRST(&unp_defers) == NULL) {
2122 UNP_DEFERRED_UNLOCK();
2125 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2126 UNP_DEFERRED_UNLOCK();
2128 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2129 SLIST_REMOVE_HEAD(&drl, ud_link);
2130 closef(dr->ud_fp, NULL);
2134 atomic_add_int(&unp_defers_count, -count);
2139 unp_internalize_fp(struct file *fp)
2144 if ((unp = fptounp(fp)) != NULL) {
2146 unp->unp_msgcount++;
2154 unp_externalize_fp(struct file *fp)
2160 if ((unp = fptounp(fp)) != NULL) {
2161 unp->unp_msgcount--;
2171 * unp_defer indicates whether additional work has been defered for a future
2172 * pass through unp_gc(). It is thread local and does not require explicit
2175 static int unp_marked;
2176 static int unp_unreachable;
2179 unp_accessable(struct filedescent **fdep, int fdcount)
2185 for (i = 0; i < fdcount; i++) {
2186 fp = fdep[i]->fde_file;
2187 if ((unp = fptounp(fp)) == NULL)
2189 if (unp->unp_gcflag & UNPGC_REF)
2191 unp->unp_gcflag &= ~UNPGC_DEAD;
2192 unp->unp_gcflag |= UNPGC_REF;
2198 unp_gc_process(struct unpcb *unp)
2204 /* Already processed. */
2205 if (unp->unp_gcflag & UNPGC_SCANNED)
2210 * Check for a socket potentially in a cycle. It must be in a
2211 * queue as indicated by msgcount, and this must equal the file
2212 * reference count. Note that when msgcount is 0 the file is NULL.
2214 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2215 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2216 unp->unp_gcflag |= UNPGC_DEAD;
2222 * Mark all sockets we reference with RIGHTS.
2224 so = unp->unp_socket;
2225 if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2226 SOCKBUF_LOCK(&so->so_rcv);
2227 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2228 SOCKBUF_UNLOCK(&so->so_rcv);
2232 * Mark all sockets in our accept queue.
2235 TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2236 if ((sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS) != 0)
2238 SOCKBUF_LOCK(&soa->so_rcv);
2239 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2240 SOCKBUF_UNLOCK(&soa->so_rcv);
2243 unp->unp_gcflag |= UNPGC_SCANNED;
2246 static int unp_recycled;
2247 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2248 "Number of unreachable sockets claimed by the garbage collector.");
2250 static int unp_taskcount;
2251 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2252 "Number of times the garbage collector has run.");
2255 unp_gc(__unused void *arg, int pending)
2257 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2259 struct unp_head **head;
2260 struct file *f, **unref;
2267 * First clear all gc flags from previous runs, apart from
2268 * UNPGC_IGNORE_RIGHTS.
2270 for (head = heads; *head != NULL; head++)
2271 LIST_FOREACH(unp, *head, unp_link)
2273 (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
2276 * Scan marking all reachable sockets with UNPGC_REF. Once a socket
2277 * is reachable all of the sockets it references are reachable.
2278 * Stop the scan once we do a complete loop without discovering
2279 * a new reachable socket.
2282 unp_unreachable = 0;
2284 for (head = heads; *head != NULL; head++)
2285 LIST_FOREACH(unp, *head, unp_link)
2286 unp_gc_process(unp);
2287 } while (unp_marked);
2289 if (unp_unreachable == 0)
2293 * Allocate space for a local list of dead unpcbs.
2295 unref = malloc(unp_unreachable * sizeof(struct file *),
2299 * Iterate looking for sockets which have been specifically marked
2300 * as as unreachable and store them locally.
2304 for (total = 0, head = heads; *head != NULL; head++)
2305 LIST_FOREACH(unp, *head, unp_link)
2306 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2308 if (unp->unp_msgcount == 0 || f == NULL ||
2309 f->f_count != unp->unp_msgcount)
2313 KASSERT(total <= unp_unreachable,
2314 ("unp_gc: incorrect unreachable count."));
2320 * Now flush all sockets, free'ing rights. This will free the
2321 * struct files associated with these sockets but leave each socket
2322 * with one remaining ref.
2324 for (i = 0; i < total; i++) {
2327 so = unref[i]->f_data;
2328 CURVNET_SET(so->so_vnet);
2334 * And finally release the sockets so they can be reclaimed.
2336 for (i = 0; i < total; i++)
2337 fdrop(unref[i], NULL);
2338 unp_recycled += total;
2339 free(unref, M_TEMP);
2343 unp_dispose(struct mbuf *m)
2347 unp_scan(m, unp_freerights);
2351 * Synchronize against unp_gc, which can trip over data as we are freeing it.
2354 unp_dispose_so(struct socket *so)
2358 unp = sotounpcb(so);
2360 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2362 unp_dispose(so->so_rcv.sb_mb);
2366 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2371 socklen_t clen, datalen;
2373 while (m0 != NULL) {
2374 for (m = m0; m; m = m->m_next) {
2375 if (m->m_type != MT_CONTROL)
2378 cm = mtod(m, struct cmsghdr *);
2381 while (cm != NULL) {
2382 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2385 data = CMSG_DATA(cm);
2386 datalen = (caddr_t)cm + cm->cmsg_len
2389 if (cm->cmsg_level == SOL_SOCKET &&
2390 cm->cmsg_type == SCM_RIGHTS) {
2391 (*op)(data, datalen /
2392 sizeof(struct filedescent *));
2395 if (CMSG_SPACE(datalen) < clen) {
2396 clen -= CMSG_SPACE(datalen);
2397 cm = (struct cmsghdr *)
2398 ((caddr_t)cm + CMSG_SPACE(datalen));
2410 * A helper function called by VFS before socket-type vnode reclamation.
2411 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2415 vfs_unp_reclaim(struct vnode *vp)
2421 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2422 KASSERT(vp->v_type == VSOCK,
2423 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2427 VOP_UNP_CONNECT(vp, &so);
2430 unp = sotounpcb(so);
2434 if (unp->unp_vnode == vp) {
2436 unp->unp_vnode = NULL;
2439 UNP_PCB_UNLOCK(unp);
2448 db_print_indent(int indent)
2452 for (i = 0; i < indent; i++)
2457 db_print_unpflags(int unp_flags)
2462 if (unp_flags & UNP_HAVEPC) {
2463 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2466 if (unp_flags & UNP_HAVEPCCACHED) {
2467 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2470 if (unp_flags & UNP_WANTCRED) {
2471 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2474 if (unp_flags & UNP_CONNWAIT) {
2475 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2478 if (unp_flags & UNP_CONNECTING) {
2479 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2482 if (unp_flags & UNP_BINDING) {
2483 db_printf("%sUNP_BINDING", comma ? ", " : "");
2489 db_print_xucred(int indent, struct xucred *xu)
2493 db_print_indent(indent);
2494 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
2495 xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2496 db_print_indent(indent);
2497 db_printf("cr_groups: ");
2499 for (i = 0; i < xu->cr_ngroups; i++) {
2500 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2507 db_print_unprefs(int indent, struct unp_head *uh)
2513 LIST_FOREACH(unp, uh, unp_reflink) {
2514 if (counter % 4 == 0)
2515 db_print_indent(indent);
2516 db_printf("%p ", unp);
2517 if (counter % 4 == 3)
2521 if (counter != 0 && counter % 4 != 0)
2525 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2530 db_printf("usage: show unpcb <addr>\n");
2533 unp = (struct unpcb *)addr;
2535 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2538 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2541 db_printf("unp_refs:\n");
2542 db_print_unprefs(2, &unp->unp_refs);
2544 /* XXXRW: Would be nice to print the full address, if any. */
2545 db_printf("unp_addr: %p\n", unp->unp_addr);
2547 db_printf("unp_gencnt: %llu\n",
2548 (unsigned long long)unp->unp_gencnt);
2550 db_printf("unp_flags: %x (", unp->unp_flags);
2551 db_print_unpflags(unp->unp_flags);
2554 db_printf("unp_peercred:\n");
2555 db_print_xucred(2, &unp->unp_peercred);
2557 db_printf("unp_refcount: %u\n", unp->unp_refcount);