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/capability.h>
66 #include <sys/domain.h>
67 #include <sys/fcntl.h>
68 #include <sys/malloc.h> /* XXX must be before <sys/file.h> */
69 #include <sys/eventhandler.h>
71 #include <sys/filedesc.h>
72 #include <sys/kernel.h>
75 #include <sys/mount.h>
76 #include <sys/mutex.h>
77 #include <sys/namei.h>
79 #include <sys/protosw.h>
80 #include <sys/queue.h>
81 #include <sys/resourcevar.h>
82 #include <sys/rwlock.h>
83 #include <sys/socket.h>
84 #include <sys/socketvar.h>
85 #include <sys/signalvar.h>
88 #include <sys/sysctl.h>
89 #include <sys/systm.h>
90 #include <sys/taskqueue.h>
92 #include <sys/unpcb.h>
93 #include <sys/vnode.h>
101 #include <security/mac/mac_framework.h>
107 * (l) Locked using list lock
108 * (g) Locked using linkage lock
111 static uma_zone_t unp_zone;
112 static unp_gen_t unp_gencnt; /* (l) */
113 static u_int unp_count; /* (l) Count of local sockets. */
114 static ino_t unp_ino; /* Prototype for fake inode numbers. */
115 static int unp_rights; /* (g) File descriptors in flight. */
116 static struct unp_head unp_shead; /* (l) List of stream sockets. */
117 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */
118 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */
121 SLIST_ENTRY(unp_defer) ud_link;
124 static SLIST_HEAD(, unp_defer) unp_defers;
125 static int unp_defers_count;
127 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
130 * Garbage collection of cyclic file descriptor/socket references occurs
131 * asynchronously in a taskqueue context in order to avoid recursion and
132 * reentrance in the UNIX domain socket, file descriptor, and socket layer
133 * code. See unp_gc() for a full description.
135 static struct timeout_task unp_gc_task;
138 * The close of unix domain sockets attached as SCM_RIGHTS is
139 * postponed to the taskqueue, to avoid arbitrary recursion depth.
140 * The attached sockets might have another sockets attached.
142 static struct task unp_defer_task;
145 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
146 * stream sockets, although the total for sender and receiver is actually
149 * Datagram sockets really use the sendspace as the maximum datagram size,
150 * and don't really want to reserve the sendspace. Their recvspace should be
151 * large enough for at least one max-size datagram plus address.
156 static u_long unpst_sendspace = PIPSIZ;
157 static u_long unpst_recvspace = PIPSIZ;
158 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
159 static u_long unpdg_recvspace = 4*1024;
160 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */
161 static u_long unpsp_recvspace = PIPSIZ;
163 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
164 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
166 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
167 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
170 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
171 &unpst_sendspace, 0, "Default stream send space.");
172 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
173 &unpst_recvspace, 0, "Default stream receive space.");
174 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
175 &unpdg_sendspace, 0, "Default datagram send space.");
176 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
177 &unpdg_recvspace, 0, "Default datagram receive space.");
178 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
179 &unpsp_sendspace, 0, "Default seqpacket send space.");
180 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
181 &unpsp_recvspace, 0, "Default seqpacket receive space.");
182 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
183 "File descriptors in flight.");
184 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
185 &unp_defers_count, 0,
186 "File descriptors deferred to taskqueue for close.");
189 * Locking and synchronization:
191 * Three types of locks exit in the local domain socket implementation: a
192 * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the
193 * global locks, the list lock protects the socket count, global generation
194 * number, and stream/datagram global lists. The linkage lock protects the
195 * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
196 * held exclusively over the acquisition of multiple unpcb locks to prevent
199 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
200 * allocated in pru_attach() and freed in pru_detach(). The validity of that
201 * pointer is an invariant, so no lock is required to dereference the so_pcb
202 * pointer if a valid socket reference is held by the caller. In practice,
203 * this is always true during operations performed on a socket. Each unpcb
204 * has a back-pointer to its socket, unp_socket, which will be stable under
205 * the same circumstances.
207 * This pointer may only be safely dereferenced as long as a valid reference
208 * to the unpcb is held. Typically, this reference will be from the socket,
209 * or from another unpcb when the referring unpcb's lock is held (in order
210 * that the reference not be invalidated during use). For example, to follow
211 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
212 * as unp_socket remains valid as long as the reference to unp_conn is valid.
214 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual
215 * atomic reads without the lock may be performed "lockless", but more
216 * complex reads and read-modify-writes require the mutex to be held. No
217 * lock order is defined between unpcb locks -- multiple unpcb locks may be
218 * acquired at the same time only when holding the linkage rwlock
219 * exclusively, which prevents deadlocks.
221 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
222 * protocols, bind() is a non-atomic operation, and connect() requires
223 * potential sleeping in the protocol, due to potentially waiting on local or
224 * distributed file systems. We try to separate "lookup" operations, which
225 * may sleep, and the IPC operations themselves, which typically can occur
226 * with relative atomicity as locks can be held over the entire operation.
228 * Another tricky issue is simultaneous multi-threaded or multi-process
229 * access to a single UNIX domain socket. These are handled by the flags
230 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
231 * binding, both of which involve dropping UNIX domain socket locks in order
232 * to perform namei() and other file system operations.
234 static struct rwlock unp_link_rwlock;
235 static struct mtx unp_list_lock;
236 static struct mtx unp_defers_lock;
238 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
241 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
243 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
246 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
247 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
248 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
249 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
250 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
253 #define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \
254 "unp_list_lock", NULL, MTX_DEF)
255 #define UNP_LIST_LOCK() mtx_lock(&unp_list_lock)
256 #define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock)
258 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
259 "unp_defer", NULL, MTX_DEF)
260 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
261 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
263 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
264 "unp_mtx", "unp_mtx", \
265 MTX_DUPOK|MTX_DEF|MTX_RECURSE)
266 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
267 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
268 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
269 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
271 static int uipc_connect2(struct socket *, struct socket *);
272 static int uipc_ctloutput(struct socket *, struct sockopt *);
273 static int unp_connect(struct socket *, struct sockaddr *,
275 static int unp_connectat(int, struct socket *, struct sockaddr *,
277 static int unp_connect2(struct socket *so, struct socket *so2, int);
278 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
279 static void unp_dispose(struct mbuf *);
280 static void unp_shutdown(struct unpcb *);
281 static void unp_drop(struct unpcb *, int);
282 static void unp_gc(__unused void *, int);
283 static void unp_scan(struct mbuf *, void (*)(struct file *));
284 static void unp_discard(struct file *);
285 static void unp_freerights(struct filedescent *, int);
286 static void unp_init(void);
287 static int unp_internalize(struct mbuf **, struct thread *);
288 static void unp_internalize_fp(struct file *);
289 static int unp_externalize(struct mbuf *, struct mbuf **);
290 static int unp_externalize_fp(struct file *);
291 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
292 static void unp_process_defers(void * __unused, int);
295 * Definitions of protocols supported in the LOCAL domain.
297 static struct domain localdomain;
298 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
299 static struct pr_usrreqs uipc_usrreqs_seqpacket;
300 static struct protosw localsw[] = {
302 .pr_type = SOCK_STREAM,
303 .pr_domain = &localdomain,
304 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
305 .pr_ctloutput = &uipc_ctloutput,
306 .pr_usrreqs = &uipc_usrreqs_stream
309 .pr_type = SOCK_DGRAM,
310 .pr_domain = &localdomain,
311 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
312 .pr_ctloutput = &uipc_ctloutput,
313 .pr_usrreqs = &uipc_usrreqs_dgram
316 .pr_type = SOCK_SEQPACKET,
317 .pr_domain = &localdomain,
320 * XXXRW: For now, PR_ADDR because soreceive will bump into them
321 * due to our use of sbappendaddr. A new sbappend variants is needed
322 * that supports both atomic record writes and control data.
324 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
326 .pr_usrreqs = &uipc_usrreqs_seqpacket,
330 static struct domain localdomain = {
331 .dom_family = AF_LOCAL,
333 .dom_init = unp_init,
334 .dom_externalize = unp_externalize,
335 .dom_dispose = unp_dispose,
336 .dom_protosw = localsw,
337 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])]
342 uipc_abort(struct socket *so)
344 struct unpcb *unp, *unp2;
347 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
351 unp2 = unp->unp_conn;
354 unp_drop(unp2, ECONNABORTED);
355 UNP_PCB_UNLOCK(unp2);
362 uipc_accept(struct socket *so, struct sockaddr **nam)
364 struct unpcb *unp, *unp2;
365 const struct sockaddr *sa;
368 * Pass back name of connected socket, if it was bound and we are
369 * still connected (our peer may have closed already!).
372 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
374 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
376 unp2 = unp->unp_conn;
377 if (unp2 != NULL && unp2->unp_addr != NULL) {
379 sa = (struct sockaddr *) unp2->unp_addr;
380 bcopy(sa, *nam, sa->sa_len);
381 UNP_PCB_UNLOCK(unp2);
384 bcopy(sa, *nam, sa->sa_len);
391 uipc_attach(struct socket *so, int proto, struct thread *td)
393 u_long sendspace, recvspace;
397 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
398 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
399 switch (so->so_type) {
401 sendspace = unpst_sendspace;
402 recvspace = unpst_recvspace;
406 sendspace = unpdg_sendspace;
407 recvspace = unpdg_recvspace;
411 sendspace = unpsp_sendspace;
412 recvspace = unpsp_recvspace;
416 panic("uipc_attach");
418 error = soreserve(so, sendspace, recvspace);
422 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
425 LIST_INIT(&unp->unp_refs);
426 UNP_PCB_LOCK_INIT(unp);
427 unp->unp_socket = so;
429 unp->unp_refcount = 1;
432 unp->unp_gencnt = ++unp_gencnt;
434 switch (so->so_type) {
436 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
440 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
444 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
448 panic("uipc_attach");
456 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
458 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
468 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
470 if (soun->sun_len > sizeof(struct sockaddr_un))
472 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
477 * We don't allow simultaneous bind() calls on a single UNIX domain
478 * socket, so flag in-progress operations, and return an error if an
479 * operation is already in progress.
481 * Historically, we have not allowed a socket to be rebound, so this
482 * also returns an error. Not allowing re-binding simplifies the
483 * implementation and avoids a great many possible failure modes.
486 if (unp->unp_vnode != NULL) {
490 if (unp->unp_flags & UNP_BINDING) {
494 unp->unp_flags |= UNP_BINDING;
497 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
498 bcopy(soun->sun_path, buf, namelen);
502 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME,
503 UIO_SYSSPACE, buf, fd, CAP_BINDAT, td);
504 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
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);
526 vattr.va_type = VSOCK;
527 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
529 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
533 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
534 NDFREE(&nd, NDF_ONLY_PNBUF);
537 vn_finished_write(mp);
541 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
542 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
546 VOP_UNP_BIND(vp, unp->unp_socket);
548 unp->unp_addr = soun;
549 unp->unp_flags &= ~UNP_BINDING;
553 vn_finished_write(mp);
559 unp->unp_flags &= ~UNP_BINDING;
566 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
569 return (uipc_bindat(AT_FDCWD, so, nam, td));
573 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
577 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
579 error = unp_connect(so, nam, td);
585 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
590 KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
592 error = unp_connectat(fd, so, nam, td);
598 uipc_close(struct socket *so)
600 struct unpcb *unp, *unp2;
603 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
607 unp2 = unp->unp_conn;
610 unp_disconnect(unp, unp2);
611 UNP_PCB_UNLOCK(unp2);
618 uipc_connect2(struct socket *so1, struct socket *so2)
620 struct unpcb *unp, *unp2;
625 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
628 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
630 error = unp_connect2(so1, so2, PRU_CONNECT2);
631 UNP_PCB_UNLOCK(unp2);
638 uipc_detach(struct socket *so)
640 struct unpcb *unp, *unp2;
641 struct sockaddr_un *saved_unp_addr;
643 int freeunp, local_unp_rights;
646 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
651 LIST_REMOVE(unp, unp_link);
652 unp->unp_gencnt = ++unp_gencnt;
657 * XXXRW: Should assert vp->v_socket == so.
659 if ((vp = unp->unp_vnode) != NULL) {
661 unp->unp_vnode = NULL;
663 unp2 = unp->unp_conn;
666 unp_disconnect(unp, unp2);
667 UNP_PCB_UNLOCK(unp2);
671 * We hold the linkage lock exclusively, so it's OK to acquire
672 * multiple pcb locks at a time.
674 while (!LIST_EMPTY(&unp->unp_refs)) {
675 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
678 unp_drop(ref, ECONNRESET);
681 local_unp_rights = unp_rights;
683 unp->unp_socket->so_pcb = NULL;
684 saved_unp_addr = unp->unp_addr;
685 unp->unp_addr = NULL;
687 freeunp = (unp->unp_refcount == 0);
688 if (saved_unp_addr != NULL)
689 free(saved_unp_addr, M_SONAME);
691 UNP_PCB_LOCK_DESTROY(unp);
692 uma_zfree(unp_zone, unp);
697 if (local_unp_rights)
698 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
702 uipc_disconnect(struct socket *so)
704 struct unpcb *unp, *unp2;
707 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
711 unp2 = unp->unp_conn;
714 unp_disconnect(unp, unp2);
715 UNP_PCB_UNLOCK(unp2);
723 uipc_listen(struct socket *so, int backlog, struct thread *td)
729 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
732 if (unp->unp_vnode == NULL) {
738 error = solisten_proto_check(so);
740 cru2x(td->td_ucred, &unp->unp_peercred);
741 unp->unp_flags |= UNP_HAVEPCCACHED;
742 solisten_proto(so, backlog);
750 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
752 struct unpcb *unp, *unp2;
753 const struct sockaddr *sa;
756 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
758 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
761 * XXX: It seems that this test always fails even when connection is
762 * established. So, this else clause is added as workaround to
763 * return PF_LOCAL sockaddr.
765 unp2 = unp->unp_conn;
768 if (unp2->unp_addr != NULL)
769 sa = (struct sockaddr *) unp2->unp_addr;
772 bcopy(sa, *nam, sa->sa_len);
773 UNP_PCB_UNLOCK(unp2);
776 bcopy(sa, *nam, sa->sa_len);
783 uipc_rcvd(struct socket *so, int flags)
785 struct unpcb *unp, *unp2;
791 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
793 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
794 panic("uipc_rcvd socktype %d", so->so_type);
797 * Adjust backpressure on sender and wakeup any waiting to write.
799 * The unp lock is acquired to maintain the validity of the unp_conn
800 * pointer; no lock on unp2 is required as unp2->unp_socket will be
801 * static as long as we don't permit unp2 to disconnect from unp,
802 * which is prevented by the lock on unp. We cache values from
803 * so_rcv to avoid holding the so_rcv lock over the entire
804 * transaction on the remote so_snd.
806 SOCKBUF_LOCK(&so->so_rcv);
807 mbcnt = so->so_rcv.sb_mbcnt;
808 sbcc = so->so_rcv.sb_cc;
809 SOCKBUF_UNLOCK(&so->so_rcv);
811 unp2 = unp->unp_conn;
816 so2 = unp2->unp_socket;
817 SOCKBUF_LOCK(&so2->so_snd);
818 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
819 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
820 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
821 newhiwat, RLIM_INFINITY);
822 sowwakeup_locked(so2);
823 unp->unp_mbcnt = mbcnt;
830 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
831 struct mbuf *control, struct thread *td)
833 struct unpcb *unp, *unp2;
835 u_int mbcnt_delta, sbcc;
840 KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
842 if (flags & PRUS_OOB) {
846 if (control != NULL && (error = unp_internalize(&control, td)))
848 if ((nam != NULL) || (flags & PRUS_EOF))
852 switch (so->so_type) {
855 const struct sockaddr *from;
857 unp2 = unp->unp_conn;
859 UNP_LINK_WLOCK_ASSERT();
864 error = unp_connect(so, nam, td);
867 unp2 = unp->unp_conn;
871 * Because connect() and send() are non-atomic in a sendto()
872 * with a target address, it's possible that the socket will
873 * have disconnected before the send() can run. In that case
874 * return the slightly counter-intuitive but otherwise
875 * correct error that the socket is not connected.
882 if (unp2->unp_flags & UNP_WANTCRED)
883 control = unp_addsockcred(td, control);
885 if (unp->unp_addr != NULL)
886 from = (struct sockaddr *)unp->unp_addr;
889 so2 = unp2->unp_socket;
890 SOCKBUF_LOCK(&so2->so_rcv);
891 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
892 sorwakeup_locked(so2);
896 SOCKBUF_UNLOCK(&so2->so_rcv);
900 UNP_LINK_WLOCK_ASSERT();
902 unp_disconnect(unp, unp2);
903 UNP_PCB_UNLOCK(unp2);
911 if ((so->so_state & SS_ISCONNECTED) == 0) {
913 UNP_LINK_WLOCK_ASSERT();
914 error = unp_connect(so, nam, td);
924 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
930 * Because connect() and send() are non-atomic in a sendto()
931 * with a target address, it's possible that the socket will
932 * have disconnected before the send() can run. In that case
933 * return the slightly counter-intuitive but otherwise
934 * correct error that the socket is not connected.
936 * Locking here must be done carefully: the linkage lock
937 * prevents interconnections between unpcbs from changing, so
938 * we can traverse from unp to unp2 without acquiring unp's
939 * lock. Socket buffer locks follow unpcb locks, so we can
940 * acquire both remote and lock socket buffer locks.
942 unp2 = unp->unp_conn;
947 so2 = unp2->unp_socket;
949 SOCKBUF_LOCK(&so2->so_rcv);
950 if (unp2->unp_flags & UNP_WANTCRED) {
952 * Credentials are passed only once on SOCK_STREAM
953 * and SOCK_SEQPACKET.
955 unp2->unp_flags &= ~UNP_WANTCRED;
956 control = unp_addsockcred(td, control);
959 * Send to paired receive port, and then reduce send buffer
960 * hiwater marks to maintain backpressure. Wake up readers.
962 switch (so->so_type) {
964 if (control != NULL) {
965 if (sbappendcontrol_locked(&so2->so_rcv, m,
969 sbappend_locked(&so2->so_rcv, m);
972 case SOCK_SEQPACKET: {
973 const struct sockaddr *from;
976 if (sbappendaddr_locked(&so2->so_rcv, from, m,
984 * XXXRW: While fine for SOCK_STREAM, this conflates maximum
985 * datagram size and back-pressure for SOCK_SEQPACKET, which
986 * can lead to undesired return of EMSGSIZE on send instead
987 * of more desirable blocking.
989 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
990 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
991 sbcc = so2->so_rcv.sb_cc;
992 sorwakeup_locked(so2);
994 SOCKBUF_LOCK(&so->so_snd);
995 if ((int)so->so_snd.sb_hiwat >= (int)(sbcc - unp2->unp_cc))
996 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
999 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
1000 newhiwat, RLIM_INFINITY);
1001 so->so_snd.sb_mbmax -= mbcnt_delta;
1002 SOCKBUF_UNLOCK(&so->so_snd);
1003 unp2->unp_cc = sbcc;
1004 UNP_PCB_UNLOCK(unp2);
1009 panic("uipc_send unknown socktype");
1013 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1015 if (flags & PRUS_EOF) {
1019 UNP_PCB_UNLOCK(unp);
1022 if ((nam != NULL) || (flags & PRUS_EOF))
1027 if (control != NULL && error != 0)
1028 unp_dispose(control);
1031 if (control != NULL)
1039 uipc_sense(struct socket *so, struct stat *sb)
1041 struct unpcb *unp, *unp2;
1044 unp = sotounpcb(so);
1045 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1047 sb->st_blksize = so->so_snd.sb_hiwat;
1050 unp2 = unp->unp_conn;
1051 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) &&
1053 so2 = unp2->unp_socket;
1054 sb->st_blksize += so2->so_rcv.sb_cc;
1057 if (unp->unp_ino == 0)
1058 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1059 sb->st_ino = unp->unp_ino;
1060 UNP_PCB_UNLOCK(unp);
1066 uipc_shutdown(struct socket *so)
1070 unp = sotounpcb(so);
1071 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1077 UNP_PCB_UNLOCK(unp);
1083 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1086 const struct sockaddr *sa;
1088 unp = sotounpcb(so);
1089 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1091 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1093 if (unp->unp_addr != NULL)
1094 sa = (struct sockaddr *) unp->unp_addr;
1097 bcopy(sa, *nam, sa->sa_len);
1098 UNP_PCB_UNLOCK(unp);
1102 static struct pr_usrreqs uipc_usrreqs_dgram = {
1103 .pru_abort = uipc_abort,
1104 .pru_accept = uipc_accept,
1105 .pru_attach = uipc_attach,
1106 .pru_bind = uipc_bind,
1107 .pru_bindat = uipc_bindat,
1108 .pru_connect = uipc_connect,
1109 .pru_connectat = uipc_connectat,
1110 .pru_connect2 = uipc_connect2,
1111 .pru_detach = uipc_detach,
1112 .pru_disconnect = uipc_disconnect,
1113 .pru_listen = uipc_listen,
1114 .pru_peeraddr = uipc_peeraddr,
1115 .pru_rcvd = uipc_rcvd,
1116 .pru_send = uipc_send,
1117 .pru_sense = uipc_sense,
1118 .pru_shutdown = uipc_shutdown,
1119 .pru_sockaddr = uipc_sockaddr,
1120 .pru_soreceive = soreceive_dgram,
1121 .pru_close = uipc_close,
1124 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1125 .pru_abort = uipc_abort,
1126 .pru_accept = uipc_accept,
1127 .pru_attach = uipc_attach,
1128 .pru_bind = uipc_bind,
1129 .pru_bindat = uipc_bindat,
1130 .pru_connect = uipc_connect,
1131 .pru_connectat = uipc_connectat,
1132 .pru_connect2 = uipc_connect2,
1133 .pru_detach = uipc_detach,
1134 .pru_disconnect = uipc_disconnect,
1135 .pru_listen = uipc_listen,
1136 .pru_peeraddr = uipc_peeraddr,
1137 .pru_rcvd = uipc_rcvd,
1138 .pru_send = uipc_send,
1139 .pru_sense = uipc_sense,
1140 .pru_shutdown = uipc_shutdown,
1141 .pru_sockaddr = uipc_sockaddr,
1142 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1143 .pru_close = uipc_close,
1146 static struct pr_usrreqs uipc_usrreqs_stream = {
1147 .pru_abort = uipc_abort,
1148 .pru_accept = uipc_accept,
1149 .pru_attach = uipc_attach,
1150 .pru_bind = uipc_bind,
1151 .pru_bindat = uipc_bindat,
1152 .pru_connect = uipc_connect,
1153 .pru_connectat = uipc_connectat,
1154 .pru_connect2 = uipc_connect2,
1155 .pru_detach = uipc_detach,
1156 .pru_disconnect = uipc_disconnect,
1157 .pru_listen = uipc_listen,
1158 .pru_peeraddr = uipc_peeraddr,
1159 .pru_rcvd = uipc_rcvd,
1160 .pru_send = uipc_send,
1161 .pru_sense = uipc_sense,
1162 .pru_shutdown = uipc_shutdown,
1163 .pru_sockaddr = uipc_sockaddr,
1164 .pru_soreceive = soreceive_generic,
1165 .pru_close = uipc_close,
1169 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1175 if (sopt->sopt_level != 0)
1178 unp = sotounpcb(so);
1179 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1181 switch (sopt->sopt_dir) {
1183 switch (sopt->sopt_name) {
1184 case LOCAL_PEERCRED:
1186 if (unp->unp_flags & UNP_HAVEPC)
1187 xu = unp->unp_peercred;
1189 if (so->so_type == SOCK_STREAM)
1194 UNP_PCB_UNLOCK(unp);
1196 error = sooptcopyout(sopt, &xu, sizeof(xu));
1200 /* Unlocked read. */
1201 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1202 error = sooptcopyout(sopt, &optval, sizeof(optval));
1205 case LOCAL_CONNWAIT:
1206 /* Unlocked read. */
1207 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1208 error = sooptcopyout(sopt, &optval, sizeof(optval));
1218 switch (sopt->sopt_name) {
1220 case LOCAL_CONNWAIT:
1221 error = sooptcopyin(sopt, &optval, sizeof(optval),
1226 #define OPTSET(bit) do { \
1227 UNP_PCB_LOCK(unp); \
1229 unp->unp_flags |= bit; \
1231 unp->unp_flags &= ~bit; \
1232 UNP_PCB_UNLOCK(unp); \
1235 switch (sopt->sopt_name) {
1237 OPTSET(UNP_WANTCRED);
1240 case LOCAL_CONNWAIT:
1241 OPTSET(UNP_CONNWAIT);
1250 error = ENOPROTOOPT;
1263 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1266 return (unp_connectat(AT_FDCWD, so, nam, td));
1270 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1273 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1275 struct socket *so2, *so3;
1276 struct unpcb *unp, *unp2, *unp3;
1278 struct nameidata nd;
1279 char buf[SOCK_MAXADDRLEN];
1280 struct sockaddr *sa;
1282 UNP_LINK_WLOCK_ASSERT();
1284 unp = sotounpcb(so);
1285 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1287 if (nam->sa_len > sizeof(struct sockaddr_un))
1289 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1292 bcopy(soun->sun_path, buf, len);
1296 if (unp->unp_flags & UNP_CONNECTING) {
1297 UNP_PCB_UNLOCK(unp);
1301 unp->unp_flags |= UNP_CONNECTING;
1302 UNP_PCB_UNLOCK(unp);
1304 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1305 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1306 UIO_SYSSPACE, buf, fd, CAP_CONNECTAT, td);
1312 ASSERT_VOP_LOCKED(vp, "unp_connect");
1313 NDFREE(&nd, NDF_ONLY_PNBUF);
1317 if (vp->v_type != VSOCK) {
1322 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1326 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1330 unp = sotounpcb(so);
1331 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1334 * Lock linkage lock for two reasons: make sure v_socket is stable,
1335 * and to protect simultaneous locking of multiple pcbs.
1338 VOP_UNP_CONNECT(vp, &so2);
1340 error = ECONNREFUSED;
1343 if (so->so_type != so2->so_type) {
1347 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1348 if (so2->so_options & SO_ACCEPTCONN) {
1349 CURVNET_SET(so2->so_vnet);
1350 so3 = sonewconn(so2, 0);
1355 error = ECONNREFUSED;
1358 unp = sotounpcb(so);
1359 unp2 = sotounpcb(so2);
1360 unp3 = sotounpcb(so3);
1364 if (unp2->unp_addr != NULL) {
1365 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1366 unp3->unp_addr = (struct sockaddr_un *) sa;
1371 * The connecter's (client's) credentials are copied from its
1372 * process structure at the time of connect() (which is now).
1374 cru2x(td->td_ucred, &unp3->unp_peercred);
1375 unp3->unp_flags |= UNP_HAVEPC;
1378 * The receiver's (server's) credentials are copied from the
1379 * unp_peercred member of socket on which the former called
1380 * listen(); uipc_listen() cached that process's credentials
1381 * at that time so we can use them now.
1383 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1384 ("unp_connect: listener without cached peercred"));
1385 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1386 sizeof(unp->unp_peercred));
1387 unp->unp_flags |= UNP_HAVEPC;
1388 if (unp2->unp_flags & UNP_WANTCRED)
1389 unp3->unp_flags |= UNP_WANTCRED;
1390 UNP_PCB_UNLOCK(unp3);
1391 UNP_PCB_UNLOCK(unp2);
1392 UNP_PCB_UNLOCK(unp);
1394 mac_socketpeer_set_from_socket(so, so3);
1395 mac_socketpeer_set_from_socket(so3, so);
1400 unp = sotounpcb(so);
1401 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1402 unp2 = sotounpcb(so2);
1403 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1406 error = unp_connect2(so, so2, PRU_CONNECT);
1407 UNP_PCB_UNLOCK(unp2);
1408 UNP_PCB_UNLOCK(unp);
1417 unp->unp_flags &= ~UNP_CONNECTING;
1418 UNP_PCB_UNLOCK(unp);
1423 unp_connect2(struct socket *so, struct socket *so2, int req)
1428 unp = sotounpcb(so);
1429 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1430 unp2 = sotounpcb(so2);
1431 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1433 UNP_LINK_WLOCK_ASSERT();
1434 UNP_PCB_LOCK_ASSERT(unp);
1435 UNP_PCB_LOCK_ASSERT(unp2);
1437 if (so2->so_type != so->so_type)
1438 return (EPROTOTYPE);
1439 unp->unp_conn = unp2;
1441 switch (so->so_type) {
1443 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1448 case SOCK_SEQPACKET:
1449 unp2->unp_conn = unp;
1450 if (req == PRU_CONNECT &&
1451 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1459 panic("unp_connect2");
1465 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1469 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1471 UNP_LINK_WLOCK_ASSERT();
1472 UNP_PCB_LOCK_ASSERT(unp);
1473 UNP_PCB_LOCK_ASSERT(unp2);
1475 unp->unp_conn = NULL;
1476 switch (unp->unp_socket->so_type) {
1478 LIST_REMOVE(unp, unp_reflink);
1479 so = unp->unp_socket;
1481 so->so_state &= ~SS_ISCONNECTED;
1486 case SOCK_SEQPACKET:
1487 soisdisconnected(unp->unp_socket);
1488 unp2->unp_conn = NULL;
1489 soisdisconnected(unp2->unp_socket);
1495 * unp_pcblist() walks the global list of struct unpcb's to generate a
1496 * pointer list, bumping the refcount on each unpcb. It then copies them out
1497 * sequentially, validating the generation number on each to see if it has
1498 * been detached. All of this is necessary because copyout() may sleep on
1502 unp_pcblist(SYSCTL_HANDLER_ARGS)
1506 struct unpcb *unp, **unp_list;
1508 struct xunpgen *xug;
1509 struct unp_head *head;
1512 switch ((intptr_t)arg1) {
1521 case SOCK_SEQPACKET:
1526 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1530 * The process of preparing the PCB list is too time-consuming and
1531 * resource-intensive to repeat twice on every request.
1533 if (req->oldptr == NULL) {
1535 req->oldidx = 2 * (sizeof *xug)
1536 + (n + n/8) * sizeof(struct xunpcb);
1540 if (req->newptr != NULL)
1544 * OK, now we're committed to doing something.
1546 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1548 gencnt = unp_gencnt;
1552 xug->xug_len = sizeof *xug;
1554 xug->xug_gen = gencnt;
1555 xug->xug_sogen = so_gencnt;
1556 error = SYSCTL_OUT(req, xug, sizeof *xug);
1562 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1565 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1566 unp = LIST_NEXT(unp, unp_link)) {
1568 if (unp->unp_gencnt <= gencnt) {
1569 if (cr_cansee(req->td->td_ucred,
1570 unp->unp_socket->so_cred)) {
1571 UNP_PCB_UNLOCK(unp);
1574 unp_list[i++] = unp;
1575 unp->unp_refcount++;
1577 UNP_PCB_UNLOCK(unp);
1580 n = i; /* In case we lost some during malloc. */
1583 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1584 for (i = 0; i < n; i++) {
1587 unp->unp_refcount--;
1588 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1589 xu->xu_len = sizeof *xu;
1592 * XXX - need more locking here to protect against
1593 * connect/disconnect races for SMP.
1595 if (unp->unp_addr != NULL)
1596 bcopy(unp->unp_addr, &xu->xu_addr,
1597 unp->unp_addr->sun_len);
1598 if (unp->unp_conn != NULL &&
1599 unp->unp_conn->unp_addr != NULL)
1600 bcopy(unp->unp_conn->unp_addr,
1602 unp->unp_conn->unp_addr->sun_len);
1603 bcopy(unp, &xu->xu_unp, sizeof *unp);
1604 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1605 UNP_PCB_UNLOCK(unp);
1606 error = SYSCTL_OUT(req, xu, sizeof *xu);
1608 freeunp = (unp->unp_refcount == 0);
1609 UNP_PCB_UNLOCK(unp);
1611 UNP_PCB_LOCK_DESTROY(unp);
1612 uma_zfree(unp_zone, unp);
1619 * Give the user an updated idea of our state. If the
1620 * generation differs from what we told her before, she knows
1621 * that something happened while we were processing this
1622 * request, and it might be necessary to retry.
1624 xug->xug_gen = unp_gencnt;
1625 xug->xug_sogen = so_gencnt;
1626 xug->xug_count = unp_count;
1627 error = SYSCTL_OUT(req, xug, sizeof *xug);
1629 free(unp_list, M_TEMP);
1634 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1635 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1636 "List of active local datagram sockets");
1637 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1638 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1639 "List of active local stream sockets");
1640 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1641 CTLTYPE_OPAQUE | CTLFLAG_RD,
1642 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1643 "List of active local seqpacket sockets");
1646 unp_shutdown(struct unpcb *unp)
1651 UNP_LINK_WLOCK_ASSERT();
1652 UNP_PCB_LOCK_ASSERT(unp);
1654 unp2 = unp->unp_conn;
1655 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1656 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1657 so = unp2->unp_socket;
1664 unp_drop(struct unpcb *unp, int errno)
1666 struct socket *so = unp->unp_socket;
1669 UNP_LINK_WLOCK_ASSERT();
1670 UNP_PCB_LOCK_ASSERT(unp);
1672 so->so_error = errno;
1673 unp2 = unp->unp_conn;
1677 unp_disconnect(unp, unp2);
1678 UNP_PCB_UNLOCK(unp2);
1682 unp_freerights(struct filedescent *fde, int fdcount)
1687 for (i = 0; i < fdcount; i++, fde++) {
1689 bzero(fde, sizeof(*fde));
1695 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1697 struct thread *td = curthread; /* XXX */
1698 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1701 struct filedesc *fdesc = td->td_proc->p_fd;
1702 struct filedescent *fde, *fdep;
1704 socklen_t clen = control->m_len, datalen;
1709 UNP_LINK_UNLOCK_ASSERT();
1712 if (controlp != NULL) /* controlp == NULL => free control messages */
1714 while (cm != NULL) {
1715 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1719 data = CMSG_DATA(cm);
1720 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1721 if (cm->cmsg_level == SOL_SOCKET
1722 && cm->cmsg_type == SCM_RIGHTS) {
1723 newfds = datalen / sizeof(*fdep);
1726 /* If we're not outputting the descriptors free them. */
1727 if (error || controlp == NULL) {
1728 unp_freerights(fdep, newfds);
1731 FILEDESC_XLOCK(fdesc);
1732 /* if the new FD's will not fit free them. */
1733 if (!fdavail(td, newfds)) {
1734 FILEDESC_XUNLOCK(fdesc);
1736 unp_freerights(fdep, newfds);
1741 * Now change each pointer to an fd in the global
1742 * table to an integer that is the index to the local
1743 * fd table entry that we set up to point to the
1744 * global one we are transferring.
1746 newlen = newfds * sizeof(int);
1747 *controlp = sbcreatecontrol(NULL, newlen,
1748 SCM_RIGHTS, SOL_SOCKET);
1749 if (*controlp == NULL) {
1750 FILEDESC_XUNLOCK(fdesc);
1752 unp_freerights(fdep, newfds);
1757 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1758 for (i = 0; i < newfds; i++, fdep++, fdp++) {
1759 if (fdalloc(td, 0, &f))
1760 panic("unp_externalize fdalloc failed");
1761 fde = &fdesc->fd_ofiles[f];
1762 fde->fde_file = fdep->fde_file;
1763 filecaps_copy(&fdep->fde_caps, &fde->fde_caps);
1764 unp_externalize_fp(fde->fde_file);
1767 FILEDESC_XUNLOCK(fdesc);
1769 /* We can just copy anything else across. */
1770 if (error || controlp == NULL)
1772 *controlp = sbcreatecontrol(NULL, datalen,
1773 cm->cmsg_type, cm->cmsg_level);
1774 if (*controlp == NULL) {
1779 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1782 controlp = &(*controlp)->m_next;
1785 if (CMSG_SPACE(datalen) < clen) {
1786 clen -= CMSG_SPACE(datalen);
1787 cm = (struct cmsghdr *)
1788 ((caddr_t)cm + CMSG_SPACE(datalen));
1800 unp_zone_change(void *tag)
1803 uma_zone_set_max(unp_zone, maxsockets);
1811 if (!IS_DEFAULT_VNET(curvnet))
1814 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1815 NULL, NULL, UMA_ALIGN_PTR, 0);
1816 if (unp_zone == NULL)
1818 uma_zone_set_max(unp_zone, maxsockets);
1819 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1820 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1821 NULL, EVENTHANDLER_PRI_ANY);
1822 LIST_INIT(&unp_dhead);
1823 LIST_INIT(&unp_shead);
1824 LIST_INIT(&unp_sphead);
1825 SLIST_INIT(&unp_defers);
1826 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1827 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1828 UNP_LINK_LOCK_INIT();
1829 UNP_LIST_LOCK_INIT();
1830 UNP_DEFERRED_LOCK_INIT();
1834 unp_internalize(struct mbuf **controlp, struct thread *td)
1836 struct mbuf *control = *controlp;
1837 struct proc *p = td->td_proc;
1838 struct filedesc *fdesc = p->p_fd;
1840 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1841 struct cmsgcred *cmcred;
1842 struct filedescent *fde, *fdep;
1847 socklen_t clen = control->m_len, datalen;
1851 UNP_LINK_UNLOCK_ASSERT();
1855 while (cm != NULL) {
1856 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1857 || cm->cmsg_len > clen) {
1861 data = CMSG_DATA(cm);
1862 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1864 switch (cm->cmsg_type) {
1866 * Fill in credential information.
1869 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1870 SCM_CREDS, SOL_SOCKET);
1871 if (*controlp == NULL) {
1875 cmcred = (struct cmsgcred *)
1876 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1877 cmcred->cmcred_pid = p->p_pid;
1878 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1879 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1880 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1881 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1883 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1884 cmcred->cmcred_groups[i] =
1885 td->td_ucred->cr_groups[i];
1889 oldfds = datalen / sizeof (int);
1891 * Check that all the FDs passed in refer to legal
1892 * files. If not, reject the entire operation.
1895 FILEDESC_SLOCK(fdesc);
1896 for (i = 0; i < oldfds; i++) {
1898 if (fget_locked(fdesc, fd) == NULL) {
1899 FILEDESC_SUNLOCK(fdesc);
1903 fp = fdesc->fd_ofiles[fd].fde_file;
1904 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1905 FILEDESC_SUNLOCK(fdesc);
1913 * Now replace the integer FDs with pointers to the
1914 * file structure and capability rights.
1916 newlen = oldfds * sizeof(*fdep);
1917 *controlp = sbcreatecontrol(NULL, newlen,
1918 SCM_RIGHTS, SOL_SOCKET);
1919 if (*controlp == NULL) {
1920 FILEDESC_SUNLOCK(fdesc);
1925 fdep = (struct filedescent *)
1926 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1927 for (i = 0; i < oldfds; i++, fdep++, fdp++) {
1928 fde = &fdesc->fd_ofiles[*fdp];
1929 fdep->fde_file = fde->fde_file;
1930 filecaps_copy(&fde->fde_caps, &fdep->fde_caps);
1931 unp_internalize_fp(fdep->fde_file);
1933 FILEDESC_SUNLOCK(fdesc);
1937 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1938 SCM_TIMESTAMP, SOL_SOCKET);
1939 if (*controlp == NULL) {
1943 tv = (struct timeval *)
1944 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1949 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
1950 SCM_BINTIME, SOL_SOCKET);
1951 if (*controlp == NULL) {
1955 bt = (struct bintime *)
1956 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1965 controlp = &(*controlp)->m_next;
1966 if (CMSG_SPACE(datalen) < clen) {
1967 clen -= CMSG_SPACE(datalen);
1968 cm = (struct cmsghdr *)
1969 ((caddr_t)cm + CMSG_SPACE(datalen));
1981 static struct mbuf *
1982 unp_addsockcred(struct thread *td, struct mbuf *control)
1984 struct mbuf *m, *n, *n_prev;
1985 struct sockcred *sc;
1986 const struct cmsghdr *cm;
1990 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1991 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1995 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1996 sc->sc_uid = td->td_ucred->cr_ruid;
1997 sc->sc_euid = td->td_ucred->cr_uid;
1998 sc->sc_gid = td->td_ucred->cr_rgid;
1999 sc->sc_egid = td->td_ucred->cr_gid;
2000 sc->sc_ngroups = ngroups;
2001 for (i = 0; i < sc->sc_ngroups; i++)
2002 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2005 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2006 * created SCM_CREDS control message (struct sockcred) has another
2009 if (control != NULL)
2010 for (n = control, n_prev = NULL; n != NULL;) {
2011 cm = mtod(n, struct cmsghdr *);
2012 if (cm->cmsg_level == SOL_SOCKET &&
2013 cm->cmsg_type == SCM_CREDS) {
2015 control = n->m_next;
2017 n_prev->m_next = n->m_next;
2025 /* Prepend it to the head. */
2026 m->m_next = control;
2030 static struct unpcb *
2031 fptounp(struct file *fp)
2035 if (fp->f_type != DTYPE_SOCKET)
2037 if ((so = fp->f_data) == NULL)
2039 if (so->so_proto->pr_domain != &localdomain)
2041 return sotounpcb(so);
2045 unp_discard(struct file *fp)
2047 struct unp_defer *dr;
2049 if (unp_externalize_fp(fp)) {
2050 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2052 UNP_DEFERRED_LOCK();
2053 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2054 UNP_DEFERRED_UNLOCK();
2055 atomic_add_int(&unp_defers_count, 1);
2056 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2058 (void) closef(fp, (struct thread *)NULL);
2062 unp_process_defers(void *arg __unused, int pending)
2064 struct unp_defer *dr;
2065 SLIST_HEAD(, unp_defer) drl;
2070 UNP_DEFERRED_LOCK();
2071 if (SLIST_FIRST(&unp_defers) == NULL) {
2072 UNP_DEFERRED_UNLOCK();
2075 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2076 UNP_DEFERRED_UNLOCK();
2078 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2079 SLIST_REMOVE_HEAD(&drl, ud_link);
2080 closef(dr->ud_fp, NULL);
2084 atomic_add_int(&unp_defers_count, -count);
2089 unp_internalize_fp(struct file *fp)
2094 if ((unp = fptounp(fp)) != NULL) {
2096 unp->unp_msgcount++;
2104 unp_externalize_fp(struct file *fp)
2110 if ((unp = fptounp(fp)) != NULL) {
2111 unp->unp_msgcount--;
2121 * unp_defer indicates whether additional work has been defered for a future
2122 * pass through unp_gc(). It is thread local and does not require explicit
2125 static int unp_marked;
2126 static int unp_unreachable;
2129 unp_accessable(struct file *fp)
2133 if ((unp = fptounp(fp)) == NULL)
2135 if (unp->unp_gcflag & UNPGC_REF)
2137 unp->unp_gcflag &= ~UNPGC_DEAD;
2138 unp->unp_gcflag |= UNPGC_REF;
2143 unp_gc_process(struct unpcb *unp)
2149 /* Already processed. */
2150 if (unp->unp_gcflag & UNPGC_SCANNED)
2155 * Check for a socket potentially in a cycle. It must be in a
2156 * queue as indicated by msgcount, and this must equal the file
2157 * reference count. Note that when msgcount is 0 the file is NULL.
2159 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2160 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2161 unp->unp_gcflag |= UNPGC_DEAD;
2167 * Mark all sockets we reference with RIGHTS.
2169 so = unp->unp_socket;
2170 SOCKBUF_LOCK(&so->so_rcv);
2171 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2172 SOCKBUF_UNLOCK(&so->so_rcv);
2175 * Mark all sockets in our accept queue.
2178 TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2179 SOCKBUF_LOCK(&soa->so_rcv);
2180 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2181 SOCKBUF_UNLOCK(&soa->so_rcv);
2184 unp->unp_gcflag |= UNPGC_SCANNED;
2187 static int unp_recycled;
2188 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2189 "Number of unreachable sockets claimed by the garbage collector.");
2191 static int unp_taskcount;
2192 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2193 "Number of times the garbage collector has run.");
2196 unp_gc(__unused void *arg, int pending)
2198 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2200 struct unp_head **head;
2201 struct file *f, **unref;
2208 * First clear all gc flags from previous runs.
2210 for (head = heads; *head != NULL; head++)
2211 LIST_FOREACH(unp, *head, unp_link)
2212 unp->unp_gcflag = 0;
2215 * Scan marking all reachable sockets with UNPGC_REF. Once a socket
2216 * is reachable all of the sockets it references are reachable.
2217 * Stop the scan once we do a complete loop without discovering
2218 * a new reachable socket.
2221 unp_unreachable = 0;
2223 for (head = heads; *head != NULL; head++)
2224 LIST_FOREACH(unp, *head, unp_link)
2225 unp_gc_process(unp);
2226 } while (unp_marked);
2228 if (unp_unreachable == 0)
2232 * Allocate space for a local list of dead unpcbs.
2234 unref = malloc(unp_unreachable * sizeof(struct file *),
2238 * Iterate looking for sockets which have been specifically marked
2239 * as as unreachable and store them locally.
2243 for (total = 0, head = heads; *head != NULL; head++)
2244 LIST_FOREACH(unp, *head, unp_link)
2245 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2247 if (unp->unp_msgcount == 0 || f == NULL ||
2248 f->f_count != unp->unp_msgcount)
2252 KASSERT(total <= unp_unreachable,
2253 ("unp_gc: incorrect unreachable count."));
2259 * Now flush all sockets, free'ing rights. This will free the
2260 * struct files associated with these sockets but leave each socket
2261 * with one remaining ref.
2263 for (i = 0; i < total; i++) {
2266 so = unref[i]->f_data;
2267 CURVNET_SET(so->so_vnet);
2273 * And finally release the sockets so they can be reclaimed.
2275 for (i = 0; i < total; i++)
2276 fdrop(unref[i], NULL);
2277 unp_recycled += total;
2278 free(unref, M_TEMP);
2282 unp_dispose(struct mbuf *m)
2286 unp_scan(m, unp_discard);
2290 unp_scan(struct mbuf *m0, void (*op)(struct file *))
2293 struct filedescent *fdep;
2297 socklen_t clen, datalen;
2300 while (m0 != NULL) {
2301 for (m = m0; m; m = m->m_next) {
2302 if (m->m_type != MT_CONTROL)
2305 cm = mtod(m, struct cmsghdr *);
2308 while (cm != NULL) {
2309 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2312 data = CMSG_DATA(cm);
2313 datalen = (caddr_t)cm + cm->cmsg_len
2316 if (cm->cmsg_level == SOL_SOCKET &&
2317 cm->cmsg_type == SCM_RIGHTS) {
2318 qfds = datalen / sizeof(*fdep);
2320 for (i = 0; i < qfds; i++, fdep++)
2321 (*op)(fdep->fde_file);
2324 if (CMSG_SPACE(datalen) < clen) {
2325 clen -= CMSG_SPACE(datalen);
2326 cm = (struct cmsghdr *)
2327 ((caddr_t)cm + CMSG_SPACE(datalen));
2339 * A helper function called by VFS before socket-type vnode reclamation.
2340 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2344 vfs_unp_reclaim(struct vnode *vp)
2350 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2351 KASSERT(vp->v_type == VSOCK,
2352 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2356 VOP_UNP_CONNECT(vp, &so);
2359 unp = sotounpcb(so);
2363 if (unp->unp_vnode == vp) {
2365 unp->unp_vnode = NULL;
2368 UNP_PCB_UNLOCK(unp);
2377 db_print_indent(int indent)
2381 for (i = 0; i < indent; i++)
2386 db_print_unpflags(int unp_flags)
2391 if (unp_flags & UNP_HAVEPC) {
2392 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2395 if (unp_flags & UNP_HAVEPCCACHED) {
2396 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2399 if (unp_flags & UNP_WANTCRED) {
2400 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2403 if (unp_flags & UNP_CONNWAIT) {
2404 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2407 if (unp_flags & UNP_CONNECTING) {
2408 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2411 if (unp_flags & UNP_BINDING) {
2412 db_printf("%sUNP_BINDING", comma ? ", " : "");
2418 db_print_xucred(int indent, struct xucred *xu)
2422 db_print_indent(indent);
2423 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
2424 xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2425 db_print_indent(indent);
2426 db_printf("cr_groups: ");
2428 for (i = 0; i < xu->cr_ngroups; i++) {
2429 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2436 db_print_unprefs(int indent, struct unp_head *uh)
2442 LIST_FOREACH(unp, uh, unp_reflink) {
2443 if (counter % 4 == 0)
2444 db_print_indent(indent);
2445 db_printf("%p ", unp);
2446 if (counter % 4 == 3)
2450 if (counter != 0 && counter % 4 != 0)
2454 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2459 db_printf("usage: show unpcb <addr>\n");
2462 unp = (struct unpcb *)addr;
2464 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2467 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2470 db_printf("unp_refs:\n");
2471 db_print_unprefs(2, &unp->unp_refs);
2473 /* XXXRW: Would be nice to print the full address, if any. */
2474 db_printf("unp_addr: %p\n", unp->unp_addr);
2476 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n",
2477 unp->unp_cc, unp->unp_mbcnt,
2478 (unsigned long long)unp->unp_gencnt);
2480 db_printf("unp_flags: %x (", unp->unp_flags);
2481 db_print_unpflags(unp->unp_flags);
2484 db_printf("unp_peercred:\n");
2485 db_print_xucred(2, &unp->unp_peercred);
2487 db_printf("unp_refcount: %u\n", unp->unp_refcount);