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 * distinguish datagram size limits from flow control limits in SEQPACKET
55 * rethink name space problems
56 * need a proper out-of-band
59 #include <sys/cdefs.h>
60 __FBSDID("$FreeBSD$");
64 #include <sys/param.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>
106 * (l) Locked using list lock
107 * (g) Locked using linkage lock
110 static uma_zone_t unp_zone;
111 static unp_gen_t unp_gencnt; /* (l) */
112 static u_int unp_count; /* (l) Count of local sockets. */
113 static ino_t unp_ino; /* Prototype for fake inode numbers. */
114 static int unp_rights; /* (g) File descriptors in flight. */
115 static struct unp_head unp_shead; /* (l) List of stream sockets. */
116 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */
117 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */
120 SLIST_ENTRY(unp_defer) ud_link;
123 static SLIST_HEAD(, unp_defer) unp_defers;
124 static int unp_defers_count;
126 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
129 * Garbage collection of cyclic file descriptor/socket references occurs
130 * asynchronously in a taskqueue context in order to avoid recursion and
131 * reentrance in the UNIX domain socket, file descriptor, and socket layer
132 * code. See unp_gc() for a full description.
134 static struct timeout_task unp_gc_task;
137 * The close of unix domain sockets attached as SCM_RIGHTS is
138 * postponed to the taskqueue, to avoid arbitrary recursion depth.
139 * The attached sockets might have another sockets attached.
141 static struct task unp_defer_task;
144 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
145 * stream sockets, although the total for sender and receiver is actually
148 * Datagram sockets really use the sendspace as the maximum datagram size,
149 * and don't really want to reserve the sendspace. Their recvspace should be
150 * large enough for at least one max-size datagram plus address.
155 static u_long unpst_sendspace = PIPSIZ;
156 static u_long unpst_recvspace = PIPSIZ;
157 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
158 static u_long unpdg_recvspace = 4*1024;
159 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */
160 static u_long unpsp_recvspace = PIPSIZ;
162 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
163 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
165 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
166 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
169 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
170 &unpst_sendspace, 0, "Default stream send space.");
171 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
172 &unpst_recvspace, 0, "Default stream receive space.");
173 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
174 &unpdg_sendspace, 0, "Default datagram send space.");
175 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
176 &unpdg_recvspace, 0, "Default datagram receive space.");
177 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
178 &unpsp_sendspace, 0, "Default seqpacket send space.");
179 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
180 &unpsp_recvspace, 0, "Default seqpacket receive space.");
181 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
182 "File descriptors in flight.");
183 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
184 &unp_defers_count, 0,
185 "File descriptors deferred to taskqueue for close.");
188 * Locking and synchronization:
190 * Three types of locks exit in the local domain socket implementation: a
191 * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the
192 * global locks, the list lock protects the socket count, global generation
193 * number, and stream/datagram global lists. The linkage lock protects the
194 * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
195 * held exclusively over the acquisition of multiple unpcb locks to prevent
198 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
199 * allocated in pru_attach() and freed in pru_detach(). The validity of that
200 * pointer is an invariant, so no lock is required to dereference the so_pcb
201 * pointer if a valid socket reference is held by the caller. In practice,
202 * this is always true during operations performed on a socket. Each unpcb
203 * has a back-pointer to its socket, unp_socket, which will be stable under
204 * the same circumstances.
206 * This pointer may only be safely dereferenced as long as a valid reference
207 * to the unpcb is held. Typically, this reference will be from the socket,
208 * or from another unpcb when the referring unpcb's lock is held (in order
209 * that the reference not be invalidated during use). For example, to follow
210 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
211 * as unp_socket remains valid as long as the reference to unp_conn is valid.
213 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual
214 * atomic reads without the lock may be performed "lockless", but more
215 * complex reads and read-modify-writes require the mutex to be held. No
216 * lock order is defined between unpcb locks -- multiple unpcb locks may be
217 * acquired at the same time only when holding the linkage rwlock
218 * exclusively, which prevents deadlocks.
220 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
221 * protocols, bind() is a non-atomic operation, and connect() requires
222 * potential sleeping in the protocol, due to potentially waiting on local or
223 * distributed file systems. We try to separate "lookup" operations, which
224 * may sleep, and the IPC operations themselves, which typically can occur
225 * with relative atomicity as locks can be held over the entire operation.
227 * Another tricky issue is simultaneous multi-threaded or multi-process
228 * access to a single UNIX domain socket. These are handled by the flags
229 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
230 * binding, both of which involve dropping UNIX domain socket locks in order
231 * to perform namei() and other file system operations.
233 static struct rwlock unp_link_rwlock;
234 static struct mtx unp_list_lock;
235 static struct mtx unp_defers_lock;
237 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
240 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
242 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
245 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
246 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
247 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
248 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
249 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
252 #define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \
253 "unp_list_lock", NULL, MTX_DEF)
254 #define UNP_LIST_LOCK() mtx_lock(&unp_list_lock)
255 #define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock)
257 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
258 "unp_defer", NULL, MTX_DEF)
259 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
260 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
262 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
263 "unp_mtx", "unp_mtx", \
264 MTX_DUPOK|MTX_DEF|MTX_RECURSE)
265 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
266 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
267 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
268 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
270 static int uipc_connect2(struct socket *, struct socket *);
271 static int uipc_ctloutput(struct socket *, struct sockopt *);
272 static int unp_connect(struct socket *, struct sockaddr *,
274 static int unp_connect2(struct socket *so, struct socket *so2, int);
275 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
276 static void unp_dispose(struct mbuf *);
277 static void unp_shutdown(struct unpcb *);
278 static void unp_drop(struct unpcb *, int);
279 static void unp_gc(__unused void *, int);
280 static void unp_scan(struct mbuf *, void (*)(struct file *));
281 static void unp_discard(struct file *);
282 static void unp_freerights(struct file **, int);
283 static void unp_init(void);
284 static int unp_internalize(struct mbuf **, struct thread *);
285 static void unp_internalize_fp(struct file *);
286 static int unp_externalize(struct mbuf *, struct mbuf **);
287 static int unp_externalize_fp(struct file *);
288 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
289 static void unp_process_defers(void * __unused, int);
292 * Definitions of protocols supported in the LOCAL domain.
294 static struct domain localdomain;
295 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
296 static struct pr_usrreqs uipc_usrreqs_seqpacket;
297 static struct protosw localsw[] = {
299 .pr_type = SOCK_STREAM,
300 .pr_domain = &localdomain,
301 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
302 .pr_ctloutput = &uipc_ctloutput,
303 .pr_usrreqs = &uipc_usrreqs_stream
306 .pr_type = SOCK_DGRAM,
307 .pr_domain = &localdomain,
308 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
309 .pr_ctloutput = &uipc_ctloutput,
310 .pr_usrreqs = &uipc_usrreqs_dgram
313 .pr_type = SOCK_SEQPACKET,
314 .pr_domain = &localdomain,
317 * XXXRW: For now, PR_ADDR because soreceive will bump into them
318 * due to our use of sbappendaddr. A new sbappend variants is needed
319 * that supports both atomic record writes and control data.
321 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
323 .pr_ctloutput = &uipc_ctloutput,
324 .pr_usrreqs = &uipc_usrreqs_seqpacket,
328 static struct domain localdomain = {
329 .dom_family = AF_LOCAL,
331 .dom_init = unp_init,
332 .dom_externalize = unp_externalize,
333 .dom_dispose = unp_dispose,
334 .dom_protosw = localsw,
335 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])]
340 uipc_abort(struct socket *so)
342 struct unpcb *unp, *unp2;
345 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
349 unp2 = unp->unp_conn;
352 unp_drop(unp2, ECONNABORTED);
353 UNP_PCB_UNLOCK(unp2);
360 uipc_accept(struct socket *so, struct sockaddr **nam)
362 struct unpcb *unp, *unp2;
363 const struct sockaddr *sa;
366 * Pass back name of connected socket, if it was bound and we are
367 * still connected (our peer may have closed already!).
370 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
372 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
374 unp2 = unp->unp_conn;
375 if (unp2 != NULL && unp2->unp_addr != NULL) {
377 sa = (struct sockaddr *) unp2->unp_addr;
378 bcopy(sa, *nam, sa->sa_len);
379 UNP_PCB_UNLOCK(unp2);
382 bcopy(sa, *nam, sa->sa_len);
389 uipc_attach(struct socket *so, int proto, struct thread *td)
391 u_long sendspace, recvspace;
395 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
396 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
397 switch (so->so_type) {
399 sendspace = unpst_sendspace;
400 recvspace = unpst_recvspace;
404 sendspace = unpdg_sendspace;
405 recvspace = unpdg_recvspace;
409 sendspace = unpsp_sendspace;
410 recvspace = unpsp_recvspace;
414 panic("uipc_attach");
416 error = soreserve(so, sendspace, recvspace);
420 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
423 LIST_INIT(&unp->unp_refs);
424 UNP_PCB_LOCK_INIT(unp);
425 unp->unp_socket = so;
427 unp->unp_refcount = 1;
430 unp->unp_gencnt = ++unp_gencnt;
432 switch (so->so_type) {
434 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
438 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
442 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
446 panic("uipc_attach");
454 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
456 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
458 int error, namelen, vfslocked;
466 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
468 if (soun->sun_len > sizeof(struct sockaddr_un))
470 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
475 * We don't allow simultaneous bind() calls on a single UNIX domain
476 * socket, so flag in-progress operations, and return an error if an
477 * operation is already in progress.
479 * Historically, we have not allowed a socket to be rebound, so this
480 * also returns an error. Not allowing re-binding simplifies the
481 * implementation and avoids a great many possible failure modes.
484 if (unp->unp_vnode != NULL) {
488 if (unp->unp_flags & UNP_BINDING) {
492 unp->unp_flags |= UNP_BINDING;
495 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
496 bcopy(soun->sun_path, buf, namelen);
501 NDINIT(&nd, CREATE, MPSAFE | NOFOLLOW | LOCKPARENT | SAVENAME,
502 UIO_SYSSPACE, buf, td);
503 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
508 vfslocked = NDHASGIANT(&nd);
509 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
510 NDFREE(&nd, NDF_ONLY_PNBUF);
520 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
523 VFS_UNLOCK_GIANT(vfslocked);
527 vattr.va_type = VSOCK;
528 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
530 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
534 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
535 NDFREE(&nd, NDF_ONLY_PNBUF);
538 vn_finished_write(mp);
542 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
543 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
547 VOP_UNP_BIND(vp, unp->unp_socket);
549 unp->unp_addr = soun;
550 unp->unp_flags &= ~UNP_BINDING;
554 vn_finished_write(mp);
555 VFS_UNLOCK_GIANT(vfslocked);
560 VFS_UNLOCK_GIANT(vfslocked);
562 unp->unp_flags &= ~UNP_BINDING;
569 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
573 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
575 error = unp_connect(so, nam, td);
581 uipc_close(struct socket *so)
583 struct unpcb *unp, *unp2;
586 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
590 unp2 = unp->unp_conn;
593 unp_disconnect(unp, unp2);
594 UNP_PCB_UNLOCK(unp2);
601 uipc_connect2(struct socket *so1, struct socket *so2)
603 struct unpcb *unp, *unp2;
608 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
611 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
613 error = unp_connect2(so1, so2, PRU_CONNECT2);
614 UNP_PCB_UNLOCK(unp2);
621 uipc_detach(struct socket *so)
623 struct unpcb *unp, *unp2;
624 struct sockaddr_un *saved_unp_addr;
626 int freeunp, local_unp_rights;
629 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
634 LIST_REMOVE(unp, unp_link);
635 unp->unp_gencnt = ++unp_gencnt;
640 * XXXRW: Should assert vp->v_socket == so.
642 if ((vp = unp->unp_vnode) != NULL) {
644 unp->unp_vnode = NULL;
646 unp2 = unp->unp_conn;
649 unp_disconnect(unp, unp2);
650 UNP_PCB_UNLOCK(unp2);
654 * We hold the linkage lock exclusively, so it's OK to acquire
655 * multiple pcb locks at a time.
657 while (!LIST_EMPTY(&unp->unp_refs)) {
658 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
661 unp_drop(ref, ECONNRESET);
664 local_unp_rights = unp_rights;
666 unp->unp_socket->so_pcb = NULL;
667 saved_unp_addr = unp->unp_addr;
668 unp->unp_addr = NULL;
670 freeunp = (unp->unp_refcount == 0);
671 if (saved_unp_addr != NULL)
672 free(saved_unp_addr, M_SONAME);
674 UNP_PCB_LOCK_DESTROY(unp);
675 uma_zfree(unp_zone, unp);
681 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
683 VFS_UNLOCK_GIANT(vfslocked);
685 if (local_unp_rights)
686 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
690 uipc_disconnect(struct socket *so)
692 struct unpcb *unp, *unp2;
695 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
699 unp2 = unp->unp_conn;
702 unp_disconnect(unp, unp2);
703 UNP_PCB_UNLOCK(unp2);
711 uipc_listen(struct socket *so, int backlog, struct thread *td)
717 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
720 if (unp->unp_vnode == NULL) {
726 error = solisten_proto_check(so);
728 cru2x(td->td_ucred, &unp->unp_peercred);
729 unp->unp_flags |= UNP_HAVEPCCACHED;
730 solisten_proto(so, backlog);
738 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
740 struct unpcb *unp, *unp2;
741 const struct sockaddr *sa;
744 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
746 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
749 * XXX: It seems that this test always fails even when connection is
750 * established. So, this else clause is added as workaround to
751 * return PF_LOCAL sockaddr.
753 unp2 = unp->unp_conn;
756 if (unp2->unp_addr != NULL)
757 sa = (struct sockaddr *) unp2->unp_addr;
760 bcopy(sa, *nam, sa->sa_len);
761 UNP_PCB_UNLOCK(unp2);
764 bcopy(sa, *nam, sa->sa_len);
771 uipc_rcvd(struct socket *so, int flags)
773 struct unpcb *unp, *unp2;
779 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
781 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
782 panic("uipc_rcvd socktype %d", so->so_type);
785 * Adjust backpressure on sender and wakeup any waiting to write.
787 * The unp lock is acquired to maintain the validity of the unp_conn
788 * pointer; no lock on unp2 is required as unp2->unp_socket will be
789 * static as long as we don't permit unp2 to disconnect from unp,
790 * which is prevented by the lock on unp. We cache values from
791 * so_rcv to avoid holding the so_rcv lock over the entire
792 * transaction on the remote so_snd.
794 SOCKBUF_LOCK(&so->so_rcv);
795 mbcnt = so->so_rcv.sb_mbcnt;
796 sbcc = so->so_rcv.sb_cc;
797 SOCKBUF_UNLOCK(&so->so_rcv);
799 unp2 = unp->unp_conn;
804 so2 = unp2->unp_socket;
805 SOCKBUF_LOCK(&so2->so_snd);
806 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
807 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
808 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
809 newhiwat, RLIM_INFINITY);
810 sowwakeup_locked(so2);
811 unp->unp_mbcnt = mbcnt;
818 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
819 struct mbuf *control, struct thread *td)
821 struct unpcb *unp, *unp2;
823 u_int mbcnt_delta, sbcc;
828 KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
830 if (flags & PRUS_OOB) {
834 if (control != NULL && (error = unp_internalize(&control, td)))
836 if ((nam != NULL) || (flags & PRUS_EOF))
840 switch (so->so_type) {
843 const struct sockaddr *from;
845 unp2 = unp->unp_conn;
847 UNP_LINK_WLOCK_ASSERT();
852 error = unp_connect(so, nam, td);
855 unp2 = unp->unp_conn;
859 * Because connect() and send() are non-atomic in a sendto()
860 * with a target address, it's possible that the socket will
861 * have disconnected before the send() can run. In that case
862 * return the slightly counter-intuitive but otherwise
863 * correct error that the socket is not connected.
870 if (unp2->unp_flags & UNP_WANTCRED)
871 control = unp_addsockcred(td, control);
873 if (unp->unp_addr != NULL)
874 from = (struct sockaddr *)unp->unp_addr;
877 so2 = unp2->unp_socket;
878 SOCKBUF_LOCK(&so2->so_rcv);
879 if (sbappendaddr_locked(&so2->so_rcv, from, m,
881 sorwakeup_locked(so2);
885 SOCKBUF_UNLOCK(&so2->so_rcv);
889 UNP_LINK_WLOCK_ASSERT();
891 unp_disconnect(unp, unp2);
892 UNP_PCB_UNLOCK(unp2);
900 if ((so->so_state & SS_ISCONNECTED) == 0) {
902 UNP_LINK_WLOCK_ASSERT();
903 error = unp_connect(so, nam, td);
913 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
919 * Because connect() and send() are non-atomic in a sendto()
920 * with a target address, it's possible that the socket will
921 * have disconnected before the send() can run. In that case
922 * return the slightly counter-intuitive but otherwise
923 * correct error that the socket is not connected.
925 * Locking here must be done carefully: the linkage lock
926 * prevents interconnections between unpcbs from changing, so
927 * we can traverse from unp to unp2 without acquiring unp's
928 * lock. Socket buffer locks follow unpcb locks, so we can
929 * acquire both remote and lock socket buffer locks.
931 unp2 = unp->unp_conn;
936 so2 = unp2->unp_socket;
938 SOCKBUF_LOCK(&so2->so_rcv);
939 if (unp2->unp_flags & UNP_WANTCRED) {
941 * Credentials are passed only once on SOCK_STREAM
942 * and SOCK_SEQPACKET.
944 unp2->unp_flags &= ~UNP_WANTCRED;
945 control = unp_addsockcred(td, control);
948 * Send to paired receive port, and then reduce send buffer
949 * hiwater marks to maintain backpressure. Wake up readers.
951 switch (so->so_type) {
953 if (control != NULL) {
954 if (sbappendcontrol_locked(&so2->so_rcv, m,
958 sbappend_locked(&so2->so_rcv, m);
961 case SOCK_SEQPACKET: {
962 const struct sockaddr *from;
966 * Don't check for space available in so2->so_rcv.
967 * Unix domain sockets only check for space in the
968 * sending sockbuf, and that check is performed one
969 * level up the stack.
971 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
979 * XXXRW: While fine for SOCK_STREAM, this conflates maximum
980 * datagram size and back-pressure for SOCK_SEQPACKET, which
981 * can lead to undesired return of EMSGSIZE on send instead
982 * of more desirable blocking.
984 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
985 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
986 sbcc = so2->so_rcv.sb_cc;
987 sorwakeup_locked(so2);
989 SOCKBUF_LOCK(&so->so_snd);
990 if ((int)so->so_snd.sb_hiwat >= (int)(sbcc - unp2->unp_cc))
991 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
994 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
995 newhiwat, RLIM_INFINITY);
996 so->so_snd.sb_mbmax -= mbcnt_delta;
997 SOCKBUF_UNLOCK(&so->so_snd);
999 UNP_PCB_UNLOCK(unp2);
1004 panic("uipc_send unknown socktype");
1008 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1010 if (flags & PRUS_EOF) {
1014 UNP_PCB_UNLOCK(unp);
1017 if ((nam != NULL) || (flags & PRUS_EOF))
1022 if (control != NULL && error != 0)
1023 unp_dispose(control);
1026 if (control != NULL)
1034 uipc_sense(struct socket *so, struct stat *sb)
1036 struct unpcb *unp, *unp2;
1039 unp = sotounpcb(so);
1040 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1042 sb->st_blksize = so->so_snd.sb_hiwat;
1045 unp2 = unp->unp_conn;
1046 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) &&
1048 so2 = unp2->unp_socket;
1049 sb->st_blksize += so2->so_rcv.sb_cc;
1052 if (unp->unp_ino == 0)
1053 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1054 sb->st_ino = unp->unp_ino;
1055 UNP_PCB_UNLOCK(unp);
1061 uipc_shutdown(struct socket *so)
1065 unp = sotounpcb(so);
1066 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1072 UNP_PCB_UNLOCK(unp);
1078 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1081 const struct sockaddr *sa;
1083 unp = sotounpcb(so);
1084 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1086 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1088 if (unp->unp_addr != NULL)
1089 sa = (struct sockaddr *) unp->unp_addr;
1092 bcopy(sa, *nam, sa->sa_len);
1093 UNP_PCB_UNLOCK(unp);
1097 static struct pr_usrreqs uipc_usrreqs_dgram = {
1098 .pru_abort = uipc_abort,
1099 .pru_accept = uipc_accept,
1100 .pru_attach = uipc_attach,
1101 .pru_bind = uipc_bind,
1102 .pru_connect = uipc_connect,
1103 .pru_connect2 = uipc_connect2,
1104 .pru_detach = uipc_detach,
1105 .pru_disconnect = uipc_disconnect,
1106 .pru_listen = uipc_listen,
1107 .pru_peeraddr = uipc_peeraddr,
1108 .pru_rcvd = uipc_rcvd,
1109 .pru_send = uipc_send,
1110 .pru_sense = uipc_sense,
1111 .pru_shutdown = uipc_shutdown,
1112 .pru_sockaddr = uipc_sockaddr,
1113 .pru_soreceive = soreceive_dgram,
1114 .pru_close = uipc_close,
1117 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1118 .pru_abort = uipc_abort,
1119 .pru_accept = uipc_accept,
1120 .pru_attach = uipc_attach,
1121 .pru_bind = uipc_bind,
1122 .pru_connect = uipc_connect,
1123 .pru_connect2 = uipc_connect2,
1124 .pru_detach = uipc_detach,
1125 .pru_disconnect = uipc_disconnect,
1126 .pru_listen = uipc_listen,
1127 .pru_peeraddr = uipc_peeraddr,
1128 .pru_rcvd = uipc_rcvd,
1129 .pru_send = uipc_send,
1130 .pru_sense = uipc_sense,
1131 .pru_shutdown = uipc_shutdown,
1132 .pru_sockaddr = uipc_sockaddr,
1133 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1134 .pru_close = uipc_close,
1137 static struct pr_usrreqs uipc_usrreqs_stream = {
1138 .pru_abort = uipc_abort,
1139 .pru_accept = uipc_accept,
1140 .pru_attach = uipc_attach,
1141 .pru_bind = uipc_bind,
1142 .pru_connect = uipc_connect,
1143 .pru_connect2 = uipc_connect2,
1144 .pru_detach = uipc_detach,
1145 .pru_disconnect = uipc_disconnect,
1146 .pru_listen = uipc_listen,
1147 .pru_peeraddr = uipc_peeraddr,
1148 .pru_rcvd = uipc_rcvd,
1149 .pru_send = uipc_send,
1150 .pru_sense = uipc_sense,
1151 .pru_shutdown = uipc_shutdown,
1152 .pru_sockaddr = uipc_sockaddr,
1153 .pru_soreceive = soreceive_generic,
1154 .pru_close = uipc_close,
1158 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1164 if (sopt->sopt_level != 0)
1167 unp = sotounpcb(so);
1168 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1170 switch (sopt->sopt_dir) {
1172 switch (sopt->sopt_name) {
1173 case LOCAL_PEERCRED:
1175 if (unp->unp_flags & UNP_HAVEPC)
1176 xu = unp->unp_peercred;
1178 if (so->so_type == SOCK_STREAM)
1183 UNP_PCB_UNLOCK(unp);
1185 error = sooptcopyout(sopt, &xu, sizeof(xu));
1189 /* Unlocked read. */
1190 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1191 error = sooptcopyout(sopt, &optval, sizeof(optval));
1194 case LOCAL_CONNWAIT:
1195 /* Unlocked read. */
1196 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1197 error = sooptcopyout(sopt, &optval, sizeof(optval));
1207 switch (sopt->sopt_name) {
1209 case LOCAL_CONNWAIT:
1210 error = sooptcopyin(sopt, &optval, sizeof(optval),
1215 #define OPTSET(bit) do { \
1216 UNP_PCB_LOCK(unp); \
1218 unp->unp_flags |= bit; \
1220 unp->unp_flags &= ~bit; \
1221 UNP_PCB_UNLOCK(unp); \
1224 switch (sopt->sopt_name) {
1226 OPTSET(UNP_WANTCRED);
1229 case LOCAL_CONNWAIT:
1230 OPTSET(UNP_CONNWAIT);
1239 error = ENOPROTOOPT;
1252 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1254 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1256 struct socket *so2, *so3;
1257 struct unpcb *unp, *unp2, *unp3;
1258 int error, len, vfslocked;
1259 struct nameidata nd;
1260 char buf[SOCK_MAXADDRLEN];
1261 struct sockaddr *sa;
1263 UNP_LINK_WLOCK_ASSERT();
1265 unp = sotounpcb(so);
1266 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1268 if (nam->sa_len > sizeof(struct sockaddr_un))
1270 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1273 bcopy(soun->sun_path, buf, len);
1277 if (unp->unp_flags & UNP_CONNECTING) {
1278 UNP_PCB_UNLOCK(unp);
1282 unp->unp_flags |= UNP_CONNECTING;
1283 UNP_PCB_UNLOCK(unp);
1285 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1286 NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKSHARED | LOCKLEAF,
1287 UIO_SYSSPACE, buf, td);
1293 ASSERT_VOP_LOCKED(vp, "unp_connect");
1294 vfslocked = NDHASGIANT(&nd);
1295 NDFREE(&nd, NDF_ONLY_PNBUF);
1299 if (vp->v_type != VSOCK) {
1304 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1308 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1311 VFS_UNLOCK_GIANT(vfslocked);
1313 unp = sotounpcb(so);
1314 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1317 * Lock linkage lock for two reasons: make sure v_socket is stable,
1318 * and to protect simultaneous locking of multiple pcbs.
1321 VOP_UNP_CONNECT(vp, &so2);
1323 error = ECONNREFUSED;
1326 if (so->so_type != so2->so_type) {
1330 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1331 if (so2->so_options & SO_ACCEPTCONN) {
1332 CURVNET_SET(so2->so_vnet);
1333 so3 = sonewconn(so2, 0);
1338 error = ECONNREFUSED;
1341 unp = sotounpcb(so);
1342 unp2 = sotounpcb(so2);
1343 unp3 = sotounpcb(so3);
1347 if (unp2->unp_addr != NULL) {
1348 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1349 unp3->unp_addr = (struct sockaddr_un *) sa;
1354 * The connector's (client's) credentials are copied from its
1355 * process structure at the time of connect() (which is now).
1357 cru2x(td->td_ucred, &unp3->unp_peercred);
1358 unp3->unp_flags |= UNP_HAVEPC;
1361 * The receiver's (server's) credentials are copied from the
1362 * unp_peercred member of socket on which the former called
1363 * listen(); uipc_listen() cached that process's credentials
1364 * at that time so we can use them now.
1366 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1367 ("unp_connect: listener without cached peercred"));
1368 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1369 sizeof(unp->unp_peercred));
1370 unp->unp_flags |= UNP_HAVEPC;
1371 if (unp2->unp_flags & UNP_WANTCRED)
1372 unp3->unp_flags |= UNP_WANTCRED;
1373 UNP_PCB_UNLOCK(unp3);
1374 UNP_PCB_UNLOCK(unp2);
1375 UNP_PCB_UNLOCK(unp);
1377 mac_socketpeer_set_from_socket(so, so3);
1378 mac_socketpeer_set_from_socket(so3, so);
1383 unp = sotounpcb(so);
1384 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1385 unp2 = sotounpcb(so2);
1386 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1389 error = unp_connect2(so, so2, PRU_CONNECT);
1390 UNP_PCB_UNLOCK(unp2);
1391 UNP_PCB_UNLOCK(unp);
1396 * Giant has been previously acquired. This means filesystem
1397 * isn't MPSAFE. Do it once again.
1403 VFS_UNLOCK_GIANT(vfslocked);
1407 unp->unp_flags &= ~UNP_CONNECTING;
1408 UNP_PCB_UNLOCK(unp);
1413 unp_connect2(struct socket *so, struct socket *so2, int req)
1418 unp = sotounpcb(so);
1419 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1420 unp2 = sotounpcb(so2);
1421 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1423 UNP_LINK_WLOCK_ASSERT();
1424 UNP_PCB_LOCK_ASSERT(unp);
1425 UNP_PCB_LOCK_ASSERT(unp2);
1427 if (so2->so_type != so->so_type)
1428 return (EPROTOTYPE);
1429 unp->unp_conn = unp2;
1431 switch (so->so_type) {
1433 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1438 case SOCK_SEQPACKET:
1439 unp2->unp_conn = unp;
1440 if (req == PRU_CONNECT &&
1441 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1449 panic("unp_connect2");
1455 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1459 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1461 UNP_LINK_WLOCK_ASSERT();
1462 UNP_PCB_LOCK_ASSERT(unp);
1463 UNP_PCB_LOCK_ASSERT(unp2);
1465 unp->unp_conn = NULL;
1466 switch (unp->unp_socket->so_type) {
1468 LIST_REMOVE(unp, unp_reflink);
1469 so = unp->unp_socket;
1471 so->so_state &= ~SS_ISCONNECTED;
1476 case SOCK_SEQPACKET:
1477 soisdisconnected(unp->unp_socket);
1478 unp2->unp_conn = NULL;
1479 soisdisconnected(unp2->unp_socket);
1485 * unp_pcblist() walks the global list of struct unpcb's to generate a
1486 * pointer list, bumping the refcount on each unpcb. It then copies them out
1487 * sequentially, validating the generation number on each to see if it has
1488 * been detached. All of this is necessary because copyout() may sleep on
1492 unp_pcblist(SYSCTL_HANDLER_ARGS)
1496 struct unpcb *unp, **unp_list;
1498 struct xunpgen *xug;
1499 struct unp_head *head;
1502 switch ((intptr_t)arg1) {
1511 case SOCK_SEQPACKET:
1516 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1520 * The process of preparing the PCB list is too time-consuming and
1521 * resource-intensive to repeat twice on every request.
1523 if (req->oldptr == NULL) {
1525 req->oldidx = 2 * (sizeof *xug)
1526 + (n + n/8) * sizeof(struct xunpcb);
1530 if (req->newptr != NULL)
1534 * OK, now we're committed to doing something.
1536 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1538 gencnt = unp_gencnt;
1542 xug->xug_len = sizeof *xug;
1544 xug->xug_gen = gencnt;
1545 xug->xug_sogen = so_gencnt;
1546 error = SYSCTL_OUT(req, xug, sizeof *xug);
1552 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1555 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1556 unp = LIST_NEXT(unp, unp_link)) {
1558 if (unp->unp_gencnt <= gencnt) {
1559 if (cr_cansee(req->td->td_ucred,
1560 unp->unp_socket->so_cred)) {
1561 UNP_PCB_UNLOCK(unp);
1564 unp_list[i++] = unp;
1565 unp->unp_refcount++;
1567 UNP_PCB_UNLOCK(unp);
1570 n = i; /* In case we lost some during malloc. */
1573 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1574 for (i = 0; i < n; i++) {
1577 unp->unp_refcount--;
1578 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1579 xu->xu_len = sizeof *xu;
1582 * XXX - need more locking here to protect against
1583 * connect/disconnect races for SMP.
1585 if (unp->unp_addr != NULL)
1586 bcopy(unp->unp_addr, &xu->xu_addr,
1587 unp->unp_addr->sun_len);
1588 if (unp->unp_conn != NULL &&
1589 unp->unp_conn->unp_addr != NULL)
1590 bcopy(unp->unp_conn->unp_addr,
1592 unp->unp_conn->unp_addr->sun_len);
1593 bcopy(unp, &xu->xu_unp, sizeof *unp);
1594 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1595 UNP_PCB_UNLOCK(unp);
1596 error = SYSCTL_OUT(req, xu, sizeof *xu);
1598 freeunp = (unp->unp_refcount == 0);
1599 UNP_PCB_UNLOCK(unp);
1601 UNP_PCB_LOCK_DESTROY(unp);
1602 uma_zfree(unp_zone, unp);
1609 * Give the user an updated idea of our state. If the
1610 * generation differs from what we told her before, she knows
1611 * that something happened while we were processing this
1612 * request, and it might be necessary to retry.
1614 xug->xug_gen = unp_gencnt;
1615 xug->xug_sogen = so_gencnt;
1616 xug->xug_count = unp_count;
1617 error = SYSCTL_OUT(req, xug, sizeof *xug);
1619 free(unp_list, M_TEMP);
1624 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1625 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1626 "List of active local datagram sockets");
1627 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1628 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1629 "List of active local stream sockets");
1630 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1631 CTLTYPE_OPAQUE | CTLFLAG_RD,
1632 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1633 "List of active local seqpacket sockets");
1636 unp_shutdown(struct unpcb *unp)
1641 UNP_LINK_WLOCK_ASSERT();
1642 UNP_PCB_LOCK_ASSERT(unp);
1644 unp2 = unp->unp_conn;
1645 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1646 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1647 so = unp2->unp_socket;
1654 unp_drop(struct unpcb *unp, int errno)
1656 struct socket *so = unp->unp_socket;
1659 UNP_LINK_WLOCK_ASSERT();
1660 UNP_PCB_LOCK_ASSERT(unp);
1662 so->so_error = errno;
1663 unp2 = unp->unp_conn;
1667 unp_disconnect(unp, unp2);
1668 UNP_PCB_UNLOCK(unp2);
1672 unp_freerights(struct file **rp, int fdcount)
1677 for (i = 0; i < fdcount; i++) {
1685 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1687 struct thread *td = curthread; /* XXX */
1688 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1694 socklen_t clen = control->m_len, datalen;
1698 UNP_LINK_UNLOCK_ASSERT();
1701 if (controlp != NULL) /* controlp == NULL => free control messages */
1703 while (cm != NULL) {
1704 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1708 data = CMSG_DATA(cm);
1709 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1710 if (cm->cmsg_level == SOL_SOCKET
1711 && cm->cmsg_type == SCM_RIGHTS) {
1712 newfds = datalen / sizeof(struct file *);
1715 /* If we're not outputting the descriptors free them. */
1716 if (error || controlp == NULL) {
1717 unp_freerights(rp, newfds);
1720 FILEDESC_XLOCK(td->td_proc->p_fd);
1722 * Now change each pointer to an fd in the global
1723 * table to an integer that is the index to the local
1724 * fd table entry that we set up to point to the
1725 * global one we are transferring.
1727 newlen = newfds * sizeof(int);
1728 *controlp = sbcreatecontrol(NULL, newlen,
1729 SCM_RIGHTS, SOL_SOCKET);
1730 if (*controlp == NULL) {
1731 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1733 unp_freerights(rp, newfds);
1738 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1739 if (fdallocn(td, 0, fdp, newfds) != 0) {
1740 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1742 unp_freerights(rp, newfds);
1747 for (i = 0; i < newfds; i++) {
1749 td->td_proc->p_fd->fd_ofiles[fdp[i]] = fp;
1750 unp_externalize_fp(fp);
1752 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1754 /* We can just copy anything else across. */
1755 if (error || controlp == NULL)
1757 *controlp = sbcreatecontrol(NULL, datalen,
1758 cm->cmsg_type, cm->cmsg_level);
1759 if (*controlp == NULL) {
1764 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1767 controlp = &(*controlp)->m_next;
1770 if (CMSG_SPACE(datalen) < clen) {
1771 clen -= CMSG_SPACE(datalen);
1772 cm = (struct cmsghdr *)
1773 ((caddr_t)cm + CMSG_SPACE(datalen));
1785 unp_zone_change(void *tag)
1788 uma_zone_set_max(unp_zone, maxsockets);
1796 if (!IS_DEFAULT_VNET(curvnet))
1799 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1800 NULL, NULL, UMA_ALIGN_PTR, 0);
1801 if (unp_zone == NULL)
1803 uma_zone_set_max(unp_zone, maxsockets);
1804 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1805 NULL, EVENTHANDLER_PRI_ANY);
1806 LIST_INIT(&unp_dhead);
1807 LIST_INIT(&unp_shead);
1808 LIST_INIT(&unp_sphead);
1809 SLIST_INIT(&unp_defers);
1810 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1811 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1812 UNP_LINK_LOCK_INIT();
1813 UNP_LIST_LOCK_INIT();
1814 UNP_DEFERRED_LOCK_INIT();
1818 unp_internalize(struct mbuf **controlp, struct thread *td)
1820 struct mbuf *control = *controlp;
1821 struct proc *p = td->td_proc;
1822 struct filedesc *fdescp = p->p_fd;
1824 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1825 struct cmsgcred *cmcred;
1831 socklen_t clen = control->m_len, datalen;
1835 UNP_LINK_UNLOCK_ASSERT();
1839 while (cm != NULL) {
1840 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1841 || cm->cmsg_len > clen) {
1845 data = CMSG_DATA(cm);
1846 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1848 switch (cm->cmsg_type) {
1850 * Fill in credential information.
1853 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1854 SCM_CREDS, SOL_SOCKET);
1855 if (*controlp == NULL) {
1859 cmcred = (struct cmsgcred *)
1860 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1861 cmcred->cmcred_pid = p->p_pid;
1862 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1863 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1864 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1865 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1867 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1868 cmcred->cmcred_groups[i] =
1869 td->td_ucred->cr_groups[i];
1873 oldfds = datalen / sizeof (int);
1875 * Check that all the FDs passed in refer to legal
1876 * files. If not, reject the entire operation.
1879 FILEDESC_SLOCK(fdescp);
1880 for (i = 0; i < oldfds; i++) {
1882 if ((unsigned)fd >= fdescp->fd_nfiles ||
1883 fdescp->fd_ofiles[fd] == NULL) {
1884 FILEDESC_SUNLOCK(fdescp);
1888 fp = fdescp->fd_ofiles[fd];
1889 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1890 FILEDESC_SUNLOCK(fdescp);
1898 * Now replace the integer FDs with pointers to the
1899 * associated global file table entry..
1901 newlen = oldfds * sizeof(struct file *);
1902 *controlp = sbcreatecontrol(NULL, newlen,
1903 SCM_RIGHTS, SOL_SOCKET);
1904 if (*controlp == NULL) {
1905 FILEDESC_SUNLOCK(fdescp);
1910 rp = (struct file **)
1911 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1912 for (i = 0; i < oldfds; i++) {
1913 fp = fdescp->fd_ofiles[*fdp++];
1915 unp_internalize_fp(fp);
1917 FILEDESC_SUNLOCK(fdescp);
1921 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1922 SCM_TIMESTAMP, SOL_SOCKET);
1923 if (*controlp == NULL) {
1927 tv = (struct timeval *)
1928 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1933 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
1934 SCM_BINTIME, SOL_SOCKET);
1935 if (*controlp == NULL) {
1939 bt = (struct bintime *)
1940 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1949 controlp = &(*controlp)->m_next;
1950 if (CMSG_SPACE(datalen) < clen) {
1951 clen -= CMSG_SPACE(datalen);
1952 cm = (struct cmsghdr *)
1953 ((caddr_t)cm + CMSG_SPACE(datalen));
1965 static struct mbuf *
1966 unp_addsockcred(struct thread *td, struct mbuf *control)
1968 struct mbuf *m, *n, *n_prev;
1969 struct sockcred *sc;
1970 const struct cmsghdr *cm;
1974 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1975 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1979 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1980 sc->sc_uid = td->td_ucred->cr_ruid;
1981 sc->sc_euid = td->td_ucred->cr_uid;
1982 sc->sc_gid = td->td_ucred->cr_rgid;
1983 sc->sc_egid = td->td_ucred->cr_gid;
1984 sc->sc_ngroups = ngroups;
1985 for (i = 0; i < sc->sc_ngroups; i++)
1986 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1989 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
1990 * created SCM_CREDS control message (struct sockcred) has another
1993 if (control != NULL)
1994 for (n = control, n_prev = NULL; n != NULL;) {
1995 cm = mtod(n, struct cmsghdr *);
1996 if (cm->cmsg_level == SOL_SOCKET &&
1997 cm->cmsg_type == SCM_CREDS) {
1999 control = n->m_next;
2001 n_prev->m_next = n->m_next;
2009 /* Prepend it to the head. */
2010 m->m_next = control;
2014 static struct unpcb *
2015 fptounp(struct file *fp)
2019 if (fp->f_type != DTYPE_SOCKET)
2021 if ((so = fp->f_data) == NULL)
2023 if (so->so_proto->pr_domain != &localdomain)
2025 return sotounpcb(so);
2029 unp_discard(struct file *fp)
2031 struct unp_defer *dr;
2033 if (unp_externalize_fp(fp)) {
2034 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2036 UNP_DEFERRED_LOCK();
2037 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2038 UNP_DEFERRED_UNLOCK();
2039 atomic_add_int(&unp_defers_count, 1);
2040 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2042 (void) closef(fp, (struct thread *)NULL);
2046 unp_process_defers(void *arg __unused, int pending)
2048 struct unp_defer *dr;
2049 SLIST_HEAD(, unp_defer) drl;
2054 UNP_DEFERRED_LOCK();
2055 if (SLIST_FIRST(&unp_defers) == NULL) {
2056 UNP_DEFERRED_UNLOCK();
2059 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2060 UNP_DEFERRED_UNLOCK();
2062 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2063 SLIST_REMOVE_HEAD(&drl, ud_link);
2064 closef(dr->ud_fp, NULL);
2068 atomic_add_int(&unp_defers_count, -count);
2073 unp_internalize_fp(struct file *fp)
2078 if ((unp = fptounp(fp)) != NULL) {
2080 unp->unp_msgcount++;
2088 unp_externalize_fp(struct file *fp)
2094 if ((unp = fptounp(fp)) != NULL) {
2095 unp->unp_msgcount--;
2105 * unp_defer indicates whether additional work has been defered for a future
2106 * pass through unp_gc(). It is thread local and does not require explicit
2109 static int unp_marked;
2110 static int unp_unreachable;
2113 unp_accessable(struct file *fp)
2117 if ((unp = fptounp(fp)) == NULL)
2119 if (unp->unp_gcflag & UNPGC_REF)
2121 unp->unp_gcflag &= ~UNPGC_DEAD;
2122 unp->unp_gcflag |= UNPGC_REF;
2127 unp_gc_process(struct unpcb *unp)
2133 /* Already processed. */
2134 if (unp->unp_gcflag & UNPGC_SCANNED)
2139 * Check for a socket potentially in a cycle. It must be in a
2140 * queue as indicated by msgcount, and this must equal the file
2141 * reference count. Note that when msgcount is 0 the file is NULL.
2143 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2144 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2145 unp->unp_gcflag |= UNPGC_DEAD;
2151 * Mark all sockets we reference with RIGHTS.
2153 so = unp->unp_socket;
2154 SOCKBUF_LOCK(&so->so_rcv);
2155 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2156 SOCKBUF_UNLOCK(&so->so_rcv);
2159 * Mark all sockets in our accept queue.
2162 TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2163 SOCKBUF_LOCK(&soa->so_rcv);
2164 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2165 SOCKBUF_UNLOCK(&soa->so_rcv);
2168 unp->unp_gcflag |= UNPGC_SCANNED;
2171 static int unp_recycled;
2172 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2173 "Number of unreachable sockets claimed by the garbage collector.");
2175 static int unp_taskcount;
2176 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2177 "Number of times the garbage collector has run.");
2180 unp_gc(__unused void *arg, int pending)
2182 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2184 struct unp_head **head;
2185 struct file *f, **unref;
2192 * First clear all gc flags from previous runs.
2194 for (head = heads; *head != NULL; head++)
2195 LIST_FOREACH(unp, *head, unp_link)
2196 unp->unp_gcflag = 0;
2199 * Scan marking all reachable sockets with UNPGC_REF. Once a socket
2200 * is reachable all of the sockets it references are reachable.
2201 * Stop the scan once we do a complete loop without discovering
2202 * a new reachable socket.
2205 unp_unreachable = 0;
2207 for (head = heads; *head != NULL; head++)
2208 LIST_FOREACH(unp, *head, unp_link)
2209 unp_gc_process(unp);
2210 } while (unp_marked);
2212 if (unp_unreachable == 0)
2216 * Allocate space for a local list of dead unpcbs.
2218 unref = malloc(unp_unreachable * sizeof(struct file *),
2222 * Iterate looking for sockets which have been specifically marked
2223 * as as unreachable and store them locally.
2227 for (total = 0, head = heads; *head != NULL; head++)
2228 LIST_FOREACH(unp, *head, unp_link)
2229 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2231 if (unp->unp_msgcount == 0 || f == NULL ||
2232 f->f_count != unp->unp_msgcount)
2236 KASSERT(total <= unp_unreachable,
2237 ("unp_gc: incorrect unreachable count."));
2243 * Now flush all sockets, free'ing rights. This will free the
2244 * struct files associated with these sockets but leave each socket
2245 * with one remaining ref.
2247 for (i = 0; i < total; i++) {
2250 so = unref[i]->f_data;
2251 CURVNET_SET(so->so_vnet);
2257 * And finally release the sockets so they can be reclaimed.
2259 for (i = 0; i < total; i++)
2260 fdrop(unref[i], NULL);
2261 unp_recycled += total;
2262 free(unref, M_TEMP);
2266 unp_dispose(struct mbuf *m)
2270 unp_scan(m, unp_discard);
2274 unp_scan(struct mbuf *m0, void (*op)(struct file *))
2281 socklen_t clen, datalen;
2284 while (m0 != NULL) {
2285 for (m = m0; m; m = m->m_next) {
2286 if (m->m_type != MT_CONTROL)
2289 cm = mtod(m, struct cmsghdr *);
2292 while (cm != NULL) {
2293 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2296 data = CMSG_DATA(cm);
2297 datalen = (caddr_t)cm + cm->cmsg_len
2300 if (cm->cmsg_level == SOL_SOCKET &&
2301 cm->cmsg_type == SCM_RIGHTS) {
2302 qfds = datalen / sizeof (struct file *);
2304 for (i = 0; i < qfds; i++)
2308 if (CMSG_SPACE(datalen) < clen) {
2309 clen -= CMSG_SPACE(datalen);
2310 cm = (struct cmsghdr *)
2311 ((caddr_t)cm + CMSG_SPACE(datalen));
2323 * A helper function called by VFS before socket-type vnode reclamation.
2324 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2328 vfs_unp_reclaim(struct vnode *vp)
2334 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2335 KASSERT(vp->v_type == VSOCK,
2336 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2340 VOP_UNP_CONNECT(vp, &so);
2343 unp = sotounpcb(so);
2347 if (unp->unp_vnode == vp) {
2349 unp->unp_vnode = NULL;
2352 UNP_PCB_UNLOCK(unp);
2361 db_print_indent(int indent)
2365 for (i = 0; i < indent; i++)
2370 db_print_unpflags(int unp_flags)
2375 if (unp_flags & UNP_HAVEPC) {
2376 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2379 if (unp_flags & UNP_HAVEPCCACHED) {
2380 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2383 if (unp_flags & UNP_WANTCRED) {
2384 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2387 if (unp_flags & UNP_CONNWAIT) {
2388 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2391 if (unp_flags & UNP_CONNECTING) {
2392 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2395 if (unp_flags & UNP_BINDING) {
2396 db_printf("%sUNP_BINDING", comma ? ", " : "");
2402 db_print_xucred(int indent, struct xucred *xu)
2406 db_print_indent(indent);
2407 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
2408 xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2409 db_print_indent(indent);
2410 db_printf("cr_groups: ");
2412 for (i = 0; i < xu->cr_ngroups; i++) {
2413 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2420 db_print_unprefs(int indent, struct unp_head *uh)
2426 LIST_FOREACH(unp, uh, unp_reflink) {
2427 if (counter % 4 == 0)
2428 db_print_indent(indent);
2429 db_printf("%p ", unp);
2430 if (counter % 4 == 3)
2434 if (counter != 0 && counter % 4 != 0)
2438 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2443 db_printf("usage: show unpcb <addr>\n");
2446 unp = (struct unpcb *)addr;
2448 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2451 db_printf("unp_ino: %d unp_conn: %p\n", unp->unp_ino,
2454 db_printf("unp_refs:\n");
2455 db_print_unprefs(2, &unp->unp_refs);
2457 /* XXXRW: Would be nice to print the full address, if any. */
2458 db_printf("unp_addr: %p\n", unp->unp_addr);
2460 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n",
2461 unp->unp_cc, unp->unp_mbcnt,
2462 (unsigned long long)unp->unp_gencnt);
2464 db_printf("unp_flags: %x (", unp->unp_flags);
2465 db_print_unpflags(unp->unp_flags);
2468 db_printf("unp_peercred:\n");
2469 db_print_xucred(2, &unp->unp_peercred);
2471 db_printf("unp_refcount: %u\n", unp->unp_refcount);