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 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 SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
163 SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM");
164 SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
165 SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
168 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
169 &unpst_sendspace, 0, "Default stream send space.");
170 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
171 &unpst_recvspace, 0, "Default stream receive space.");
172 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
173 &unpdg_sendspace, 0, "Default datagram send space.");
174 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
175 &unpdg_recvspace, 0, "Default datagram receive space.");
176 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
177 &unpsp_sendspace, 0, "Default seqpacket send space.");
178 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
179 &unpsp_recvspace, 0, "Default seqpacket receive space.");
180 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
181 "File descriptors in flight.");
182 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
183 &unp_defers_count, 0,
184 "File descriptors deferred to taskqueue for close.");
187 * Locking and synchronization:
189 * Three types of locks exit in the local domain socket implementation: a
190 * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the
191 * global locks, the list lock protects the socket count, global generation
192 * number, and stream/datagram global lists. The linkage lock protects the
193 * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
194 * held exclusively over the acquisition of multiple unpcb locks to prevent
197 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
198 * allocated in pru_attach() and freed in pru_detach(). The validity of that
199 * pointer is an invariant, so no lock is required to dereference the so_pcb
200 * pointer if a valid socket reference is held by the caller. In practice,
201 * this is always true during operations performed on a socket. Each unpcb
202 * has a back-pointer to its socket, unp_socket, which will be stable under
203 * the same circumstances.
205 * This pointer may only be safely dereferenced as long as a valid reference
206 * to the unpcb is held. Typically, this reference will be from the socket,
207 * or from another unpcb when the referring unpcb's lock is held (in order
208 * that the reference not be invalidated during use). For example, to follow
209 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
210 * as unp_socket remains valid as long as the reference to unp_conn is valid.
212 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual
213 * atomic reads without the lock may be performed "lockless", but more
214 * complex reads and read-modify-writes require the mutex to be held. No
215 * lock order is defined between unpcb locks -- multiple unpcb locks may be
216 * acquired at the same time only when holding the linkage rwlock
217 * exclusively, which prevents deadlocks.
219 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
220 * protocols, bind() is a non-atomic operation, and connect() requires
221 * potential sleeping in the protocol, due to potentially waiting on local or
222 * distributed file systems. We try to separate "lookup" operations, which
223 * may sleep, and the IPC operations themselves, which typically can occur
224 * with relative atomicity as locks can be held over the entire operation.
226 * Another tricky issue is simultaneous multi-threaded or multi-process
227 * access to a single UNIX domain socket. These are handled by the flags
228 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
229 * binding, both of which involve dropping UNIX domain socket locks in order
230 * to perform namei() and other file system operations.
232 static struct rwlock unp_link_rwlock;
233 static struct mtx unp_list_lock;
234 static struct mtx unp_defers_lock;
236 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
239 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
241 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
244 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
245 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
246 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
247 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
248 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
251 #define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \
252 "unp_list_lock", NULL, MTX_DEF)
253 #define UNP_LIST_LOCK() mtx_lock(&unp_list_lock)
254 #define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock)
256 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
257 "unp_defer", NULL, MTX_DEF)
258 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
259 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
261 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
262 "unp_mtx", "unp_mtx", \
263 MTX_DUPOK|MTX_DEF|MTX_RECURSE)
264 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
265 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
266 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
267 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
269 static int uipc_connect2(struct socket *, struct socket *);
270 static int uipc_ctloutput(struct socket *, struct sockopt *);
271 static int unp_connect(struct socket *, struct sockaddr *,
273 static int unp_connect2(struct socket *so, struct socket *so2, int);
274 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
275 static void unp_dispose(struct mbuf *);
276 static void unp_shutdown(struct unpcb *);
277 static void unp_drop(struct unpcb *, int);
278 static void unp_gc(__unused void *, int);
279 static void unp_scan(struct mbuf *, void (*)(struct file *));
280 static void unp_discard(struct file *);
281 static void unp_freerights(struct file **, int);
282 static void unp_init(void);
283 static int unp_internalize(struct mbuf **, struct thread *);
284 static void unp_internalize_fp(struct file *);
285 static int unp_externalize(struct mbuf *, struct mbuf **);
286 static int unp_externalize_fp(struct file *);
287 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
288 static void unp_process_defers(void * __unused, int);
291 * Definitions of protocols supported in the LOCAL domain.
293 static struct domain localdomain;
294 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
295 static struct pr_usrreqs uipc_usrreqs_seqpacket;
296 static struct protosw localsw[] = {
298 .pr_type = SOCK_STREAM,
299 .pr_domain = &localdomain,
300 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
301 .pr_ctloutput = &uipc_ctloutput,
302 .pr_usrreqs = &uipc_usrreqs_stream
305 .pr_type = SOCK_DGRAM,
306 .pr_domain = &localdomain,
307 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
308 .pr_usrreqs = &uipc_usrreqs_dgram
311 .pr_type = SOCK_SEQPACKET,
312 .pr_domain = &localdomain,
315 * XXXRW: For now, PR_ADDR because soreceive will bump into them
316 * due to our use of sbappendaddr. A new sbappend variants is needed
317 * that supports both atomic record writes and control data.
319 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
321 .pr_usrreqs = &uipc_usrreqs_seqpacket,
325 static struct domain localdomain = {
326 .dom_family = AF_LOCAL,
328 .dom_init = unp_init,
329 .dom_externalize = unp_externalize,
330 .dom_dispose = unp_dispose,
331 .dom_protosw = localsw,
332 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])]
337 uipc_abort(struct socket *so)
339 struct unpcb *unp, *unp2;
342 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
346 unp2 = unp->unp_conn;
349 unp_drop(unp2, ECONNABORTED);
350 UNP_PCB_UNLOCK(unp2);
357 uipc_accept(struct socket *so, struct sockaddr **nam)
359 struct unpcb *unp, *unp2;
360 const struct sockaddr *sa;
363 * Pass back name of connected socket, if it was bound and we are
364 * still connected (our peer may have closed already!).
367 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
369 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
371 unp2 = unp->unp_conn;
372 if (unp2 != NULL && unp2->unp_addr != NULL) {
374 sa = (struct sockaddr *) unp2->unp_addr;
375 bcopy(sa, *nam, sa->sa_len);
376 UNP_PCB_UNLOCK(unp2);
379 bcopy(sa, *nam, sa->sa_len);
386 uipc_attach(struct socket *so, int proto, struct thread *td)
388 u_long sendspace, recvspace;
392 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
393 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
394 switch (so->so_type) {
396 sendspace = unpst_sendspace;
397 recvspace = unpst_recvspace;
401 sendspace = unpdg_sendspace;
402 recvspace = unpdg_recvspace;
406 sendspace = unpsp_sendspace;
407 recvspace = unpsp_recvspace;
411 panic("uipc_attach");
413 error = soreserve(so, sendspace, recvspace);
417 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
420 LIST_INIT(&unp->unp_refs);
421 UNP_PCB_LOCK_INIT(unp);
422 unp->unp_socket = so;
424 unp->unp_refcount = 1;
427 unp->unp_gencnt = ++unp_gencnt;
429 switch (so->so_type) {
431 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
435 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
439 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
443 panic("uipc_attach");
451 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
453 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
455 int error, namelen, vfslocked;
463 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
465 if (soun->sun_len > sizeof(struct sockaddr_un))
467 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
472 * We don't allow simultaneous bind() calls on a single UNIX domain
473 * socket, so flag in-progress operations, and return an error if an
474 * operation is already in progress.
476 * Historically, we have not allowed a socket to be rebound, so this
477 * also returns an error. Not allowing re-binding simplifies the
478 * implementation and avoids a great many possible failure modes.
481 if (unp->unp_vnode != NULL) {
485 if (unp->unp_flags & UNP_BINDING) {
489 unp->unp_flags |= UNP_BINDING;
492 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
493 bcopy(soun->sun_path, buf, namelen);
498 NDINIT(&nd, CREATE, MPSAFE | NOFOLLOW | LOCKPARENT | SAVENAME,
499 UIO_SYSSPACE, buf, td);
500 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
505 vfslocked = NDHASGIANT(&nd);
506 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
507 NDFREE(&nd, NDF_ONLY_PNBUF);
517 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
520 VFS_UNLOCK_GIANT(vfslocked);
524 vattr.va_type = VSOCK;
525 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
527 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
531 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
532 NDFREE(&nd, NDF_ONLY_PNBUF);
535 vn_finished_write(mp);
539 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
540 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
544 vp->v_socket = unp->unp_socket;
546 unp->unp_addr = soun;
547 unp->unp_flags &= ~UNP_BINDING;
551 vn_finished_write(mp);
552 VFS_UNLOCK_GIANT(vfslocked);
557 VFS_UNLOCK_GIANT(vfslocked);
559 unp->unp_flags &= ~UNP_BINDING;
566 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
570 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
572 error = unp_connect(so, nam, td);
578 uipc_close(struct socket *so)
580 struct unpcb *unp, *unp2;
583 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
587 unp2 = unp->unp_conn;
590 unp_disconnect(unp, unp2);
591 UNP_PCB_UNLOCK(unp2);
598 uipc_connect2(struct socket *so1, struct socket *so2)
600 struct unpcb *unp, *unp2;
605 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
608 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
610 error = unp_connect2(so1, so2, PRU_CONNECT2);
611 UNP_PCB_UNLOCK(unp2);
618 uipc_detach(struct socket *so)
620 struct unpcb *unp, *unp2;
621 struct sockaddr_un *saved_unp_addr;
623 int freeunp, local_unp_rights;
626 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
631 LIST_REMOVE(unp, unp_link);
632 unp->unp_gencnt = ++unp_gencnt;
637 * XXXRW: Should assert vp->v_socket == so.
639 if ((vp = unp->unp_vnode) != NULL) {
640 unp->unp_vnode->v_socket = NULL;
641 unp->unp_vnode = NULL;
643 unp2 = unp->unp_conn;
646 unp_disconnect(unp, unp2);
647 UNP_PCB_UNLOCK(unp2);
651 * We hold the linkage lock exclusively, so it's OK to acquire
652 * multiple pcb locks at a time.
654 while (!LIST_EMPTY(&unp->unp_refs)) {
655 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
658 unp_drop(ref, ECONNRESET);
661 local_unp_rights = unp_rights;
663 unp->unp_socket->so_pcb = NULL;
664 saved_unp_addr = unp->unp_addr;
665 unp->unp_addr = NULL;
667 freeunp = (unp->unp_refcount == 0);
668 if (saved_unp_addr != NULL)
669 free(saved_unp_addr, M_SONAME);
671 UNP_PCB_LOCK_DESTROY(unp);
672 uma_zfree(unp_zone, unp);
678 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
680 VFS_UNLOCK_GIANT(vfslocked);
682 if (local_unp_rights)
683 taskqueue_enqueue(taskqueue_thread, &unp_gc_task);
687 uipc_disconnect(struct socket *so)
689 struct unpcb *unp, *unp2;
692 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
696 unp2 = unp->unp_conn;
699 unp_disconnect(unp, unp2);
700 UNP_PCB_UNLOCK(unp2);
708 uipc_listen(struct socket *so, int backlog, struct thread *td)
714 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
717 if (unp->unp_vnode == NULL) {
723 error = solisten_proto_check(so);
725 cru2x(td->td_ucred, &unp->unp_peercred);
726 unp->unp_flags |= UNP_HAVEPCCACHED;
727 solisten_proto(so, backlog);
735 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
737 struct unpcb *unp, *unp2;
738 const struct sockaddr *sa;
741 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
743 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
746 * XXX: It seems that this test always fails even when connection is
747 * established. So, this else clause is added as workaround to
748 * return PF_LOCAL sockaddr.
750 unp2 = unp->unp_conn;
753 if (unp2->unp_addr != NULL)
754 sa = (struct sockaddr *) unp2->unp_addr;
757 bcopy(sa, *nam, sa->sa_len);
758 UNP_PCB_UNLOCK(unp2);
761 bcopy(sa, *nam, sa->sa_len);
768 uipc_rcvd(struct socket *so, int flags)
770 struct unpcb *unp, *unp2;
776 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
778 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
779 panic("uipc_rcvd socktype %d", so->so_type);
782 * Adjust backpressure on sender and wakeup any waiting to write.
784 * The unp lock is acquired to maintain the validity of the unp_conn
785 * pointer; no lock on unp2 is required as unp2->unp_socket will be
786 * static as long as we don't permit unp2 to disconnect from unp,
787 * which is prevented by the lock on unp. We cache values from
788 * so_rcv to avoid holding the so_rcv lock over the entire
789 * transaction on the remote so_snd.
791 SOCKBUF_LOCK(&so->so_rcv);
792 mbcnt = so->so_rcv.sb_mbcnt;
793 sbcc = so->so_rcv.sb_cc;
794 SOCKBUF_UNLOCK(&so->so_rcv);
796 unp2 = unp->unp_conn;
801 so2 = unp2->unp_socket;
802 SOCKBUF_LOCK(&so2->so_snd);
803 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
804 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
805 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
806 newhiwat, RLIM_INFINITY);
807 sowwakeup_locked(so2);
808 unp->unp_mbcnt = mbcnt;
815 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
816 struct mbuf *control, struct thread *td)
818 struct unpcb *unp, *unp2;
820 u_int mbcnt_delta, sbcc;
825 KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
827 if (flags & PRUS_OOB) {
831 if (control != NULL && (error = unp_internalize(&control, td)))
833 if ((nam != NULL) || (flags & PRUS_EOF))
837 switch (so->so_type) {
840 const struct sockaddr *from;
842 unp2 = unp->unp_conn;
844 UNP_LINK_WLOCK_ASSERT();
849 error = unp_connect(so, nam, td);
852 unp2 = unp->unp_conn;
856 * Because connect() and send() are non-atomic in a sendto()
857 * with a target address, it's possible that the socket will
858 * have disconnected before the send() can run. In that case
859 * return the slightly counter-intuitive but otherwise
860 * correct error that the socket is not connected.
867 if (unp2->unp_flags & UNP_WANTCRED)
868 control = unp_addsockcred(td, control);
870 if (unp->unp_addr != NULL)
871 from = (struct sockaddr *)unp->unp_addr;
874 so2 = unp2->unp_socket;
875 SOCKBUF_LOCK(&so2->so_rcv);
876 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
877 sorwakeup_locked(so2);
881 SOCKBUF_UNLOCK(&so2->so_rcv);
885 UNP_LINK_WLOCK_ASSERT();
887 unp_disconnect(unp, unp2);
888 UNP_PCB_UNLOCK(unp2);
896 if ((so->so_state & SS_ISCONNECTED) == 0) {
898 UNP_LINK_WLOCK_ASSERT();
899 error = unp_connect(so, nam, td);
909 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
915 * Because connect() and send() are non-atomic in a sendto()
916 * with a target address, it's possible that the socket will
917 * have disconnected before the send() can run. In that case
918 * return the slightly counter-intuitive but otherwise
919 * correct error that the socket is not connected.
921 * Locking here must be done carefully: the linkage lock
922 * prevents interconnections between unpcbs from changing, so
923 * we can traverse from unp to unp2 without acquiring unp's
924 * lock. Socket buffer locks follow unpcb locks, so we can
925 * acquire both remote and lock socket buffer locks.
927 unp2 = unp->unp_conn;
932 so2 = unp2->unp_socket;
934 SOCKBUF_LOCK(&so2->so_rcv);
935 if (unp2->unp_flags & UNP_WANTCRED) {
937 * Credentials are passed only once on SOCK_STREAM.
939 unp2->unp_flags &= ~UNP_WANTCRED;
940 control = unp_addsockcred(td, control);
943 * Send to paired receive port, and then reduce send buffer
944 * hiwater marks to maintain backpressure. Wake up readers.
946 switch (so->so_type) {
948 if (control != NULL) {
949 if (sbappendcontrol_locked(&so2->so_rcv, m,
953 sbappend_locked(&so2->so_rcv, m);
956 case SOCK_SEQPACKET: {
957 const struct sockaddr *from;
960 if (sbappendaddr_locked(&so2->so_rcv, from, m,
968 * XXXRW: While fine for SOCK_STREAM, this conflates maximum
969 * datagram size and back-pressure for SOCK_SEQPACKET, which
970 * can lead to undesired return of EMSGSIZE on send instead
971 * of more desirable blocking.
973 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
974 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
975 sbcc = so2->so_rcv.sb_cc;
976 sorwakeup_locked(so2);
978 SOCKBUF_LOCK(&so->so_snd);
979 if ((int)so->so_snd.sb_hiwat >= (int)(sbcc - unp2->unp_cc))
980 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
983 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
984 newhiwat, RLIM_INFINITY);
985 so->so_snd.sb_mbmax -= mbcnt_delta;
986 SOCKBUF_UNLOCK(&so->so_snd);
988 UNP_PCB_UNLOCK(unp2);
993 panic("uipc_send unknown socktype");
997 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
999 if (flags & PRUS_EOF) {
1003 UNP_PCB_UNLOCK(unp);
1006 if ((nam != NULL) || (flags & PRUS_EOF))
1011 if (control != NULL && error != 0)
1012 unp_dispose(control);
1015 if (control != NULL)
1023 uipc_sense(struct socket *so, struct stat *sb)
1025 struct unpcb *unp, *unp2;
1028 unp = sotounpcb(so);
1029 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1031 sb->st_blksize = so->so_snd.sb_hiwat;
1034 unp2 = unp->unp_conn;
1035 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) &&
1037 so2 = unp2->unp_socket;
1038 sb->st_blksize += so2->so_rcv.sb_cc;
1041 if (unp->unp_ino == 0)
1042 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1043 sb->st_ino = unp->unp_ino;
1044 UNP_PCB_UNLOCK(unp);
1050 uipc_shutdown(struct socket *so)
1054 unp = sotounpcb(so);
1055 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1061 UNP_PCB_UNLOCK(unp);
1067 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1070 const struct sockaddr *sa;
1072 unp = sotounpcb(so);
1073 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1075 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1077 if (unp->unp_addr != NULL)
1078 sa = (struct sockaddr *) unp->unp_addr;
1081 bcopy(sa, *nam, sa->sa_len);
1082 UNP_PCB_UNLOCK(unp);
1086 static struct pr_usrreqs uipc_usrreqs_dgram = {
1087 .pru_abort = uipc_abort,
1088 .pru_accept = uipc_accept,
1089 .pru_attach = uipc_attach,
1090 .pru_bind = uipc_bind,
1091 .pru_connect = uipc_connect,
1092 .pru_connect2 = uipc_connect2,
1093 .pru_detach = uipc_detach,
1094 .pru_disconnect = uipc_disconnect,
1095 .pru_listen = uipc_listen,
1096 .pru_peeraddr = uipc_peeraddr,
1097 .pru_rcvd = uipc_rcvd,
1098 .pru_send = uipc_send,
1099 .pru_sense = uipc_sense,
1100 .pru_shutdown = uipc_shutdown,
1101 .pru_sockaddr = uipc_sockaddr,
1102 .pru_soreceive = soreceive_dgram,
1103 .pru_close = uipc_close,
1106 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1107 .pru_abort = uipc_abort,
1108 .pru_accept = uipc_accept,
1109 .pru_attach = uipc_attach,
1110 .pru_bind = uipc_bind,
1111 .pru_connect = uipc_connect,
1112 .pru_connect2 = uipc_connect2,
1113 .pru_detach = uipc_detach,
1114 .pru_disconnect = uipc_disconnect,
1115 .pru_listen = uipc_listen,
1116 .pru_peeraddr = uipc_peeraddr,
1117 .pru_rcvd = uipc_rcvd,
1118 .pru_send = uipc_send,
1119 .pru_sense = uipc_sense,
1120 .pru_shutdown = uipc_shutdown,
1121 .pru_sockaddr = uipc_sockaddr,
1122 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1123 .pru_close = uipc_close,
1126 static struct pr_usrreqs uipc_usrreqs_stream = {
1127 .pru_abort = uipc_abort,
1128 .pru_accept = uipc_accept,
1129 .pru_attach = uipc_attach,
1130 .pru_bind = uipc_bind,
1131 .pru_connect = uipc_connect,
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,
1143 .pru_close = uipc_close,
1147 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1153 if (sopt->sopt_level != 0)
1156 unp = sotounpcb(so);
1157 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1159 switch (sopt->sopt_dir) {
1161 switch (sopt->sopt_name) {
1162 case LOCAL_PEERCRED:
1164 if (unp->unp_flags & UNP_HAVEPC)
1165 xu = unp->unp_peercred;
1167 if (so->so_type == SOCK_STREAM)
1172 UNP_PCB_UNLOCK(unp);
1174 error = sooptcopyout(sopt, &xu, sizeof(xu));
1178 /* Unlocked read. */
1179 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1180 error = sooptcopyout(sopt, &optval, sizeof(optval));
1183 case LOCAL_CONNWAIT:
1184 /* Unlocked read. */
1185 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1186 error = sooptcopyout(sopt, &optval, sizeof(optval));
1196 switch (sopt->sopt_name) {
1198 case LOCAL_CONNWAIT:
1199 error = sooptcopyin(sopt, &optval, sizeof(optval),
1204 #define OPTSET(bit) do { \
1205 UNP_PCB_LOCK(unp); \
1207 unp->unp_flags |= bit; \
1209 unp->unp_flags &= ~bit; \
1210 UNP_PCB_UNLOCK(unp); \
1213 switch (sopt->sopt_name) {
1215 OPTSET(UNP_WANTCRED);
1218 case LOCAL_CONNWAIT:
1219 OPTSET(UNP_CONNWAIT);
1228 error = ENOPROTOOPT;
1241 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1243 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1245 struct socket *so2, *so3;
1246 struct unpcb *unp, *unp2, *unp3;
1247 int error, len, vfslocked;
1248 struct nameidata nd;
1249 char buf[SOCK_MAXADDRLEN];
1250 struct sockaddr *sa;
1252 UNP_LINK_WLOCK_ASSERT();
1254 unp = sotounpcb(so);
1255 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1257 if (nam->sa_len > sizeof(struct sockaddr_un))
1259 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1262 bcopy(soun->sun_path, buf, len);
1266 if (unp->unp_flags & UNP_CONNECTING) {
1267 UNP_PCB_UNLOCK(unp);
1271 unp->unp_flags |= UNP_CONNECTING;
1272 UNP_PCB_UNLOCK(unp);
1274 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1275 NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKSHARED | LOCKLEAF,
1276 UIO_SYSSPACE, buf, td);
1282 ASSERT_VOP_LOCKED(vp, "unp_connect");
1283 vfslocked = NDHASGIANT(&nd);
1284 NDFREE(&nd, NDF_ONLY_PNBUF);
1288 if (vp->v_type != VSOCK) {
1293 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1297 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1300 VFS_UNLOCK_GIANT(vfslocked);
1302 unp = sotounpcb(so);
1303 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1306 * Lock linkage lock for two reasons: make sure v_socket is stable,
1307 * and to protect simultaneous locking of multiple pcbs.
1312 error = ECONNREFUSED;
1315 if (so->so_type != so2->so_type) {
1319 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1320 if (so2->so_options & SO_ACCEPTCONN) {
1321 CURVNET_SET(so2->so_vnet);
1322 so3 = sonewconn(so2, 0);
1327 error = ECONNREFUSED;
1330 unp = sotounpcb(so);
1331 unp2 = sotounpcb(so2);
1332 unp3 = sotounpcb(so3);
1336 if (unp2->unp_addr != NULL) {
1337 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1338 unp3->unp_addr = (struct sockaddr_un *) sa;
1343 * The connecter's (client's) credentials are copied from its
1344 * process structure at the time of connect() (which is now).
1346 cru2x(td->td_ucred, &unp3->unp_peercred);
1347 unp3->unp_flags |= UNP_HAVEPC;
1350 * The receiver's (server's) credentials are copied from the
1351 * unp_peercred member of socket on which the former called
1352 * listen(); uipc_listen() cached that process's credentials
1353 * at that time so we can use them now.
1355 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1356 ("unp_connect: listener without cached peercred"));
1357 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1358 sizeof(unp->unp_peercred));
1359 unp->unp_flags |= UNP_HAVEPC;
1360 if (unp2->unp_flags & UNP_WANTCRED)
1361 unp3->unp_flags |= UNP_WANTCRED;
1362 UNP_PCB_UNLOCK(unp3);
1363 UNP_PCB_UNLOCK(unp2);
1364 UNP_PCB_UNLOCK(unp);
1366 mac_socketpeer_set_from_socket(so, so3);
1367 mac_socketpeer_set_from_socket(so3, so);
1372 unp = sotounpcb(so);
1373 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1374 unp2 = sotounpcb(so2);
1375 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1378 error = unp_connect2(so, so2, PRU_CONNECT);
1379 UNP_PCB_UNLOCK(unp2);
1380 UNP_PCB_UNLOCK(unp);
1385 * Giant has been previously acquired. This means filesystem
1386 * isn't MPSAFE. Do it once again.
1392 VFS_UNLOCK_GIANT(vfslocked);
1396 unp->unp_flags &= ~UNP_CONNECTING;
1397 UNP_PCB_UNLOCK(unp);
1402 unp_connect2(struct socket *so, struct socket *so2, int req)
1407 unp = sotounpcb(so);
1408 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1409 unp2 = sotounpcb(so2);
1410 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1412 UNP_LINK_WLOCK_ASSERT();
1413 UNP_PCB_LOCK_ASSERT(unp);
1414 UNP_PCB_LOCK_ASSERT(unp2);
1416 if (so2->so_type != so->so_type)
1417 return (EPROTOTYPE);
1418 unp->unp_conn = unp2;
1420 switch (so->so_type) {
1422 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1427 case SOCK_SEQPACKET:
1428 unp2->unp_conn = unp;
1429 if (req == PRU_CONNECT &&
1430 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1438 panic("unp_connect2");
1444 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1448 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1450 UNP_LINK_WLOCK_ASSERT();
1451 UNP_PCB_LOCK_ASSERT(unp);
1452 UNP_PCB_LOCK_ASSERT(unp2);
1454 unp->unp_conn = NULL;
1455 switch (unp->unp_socket->so_type) {
1457 LIST_REMOVE(unp, unp_reflink);
1458 so = unp->unp_socket;
1460 so->so_state &= ~SS_ISCONNECTED;
1465 case SOCK_SEQPACKET:
1466 soisdisconnected(unp->unp_socket);
1467 unp2->unp_conn = NULL;
1468 soisdisconnected(unp2->unp_socket);
1474 * unp_pcblist() walks the global list of struct unpcb's to generate a
1475 * pointer list, bumping the refcount on each unpcb. It then copies them out
1476 * sequentially, validating the generation number on each to see if it has
1477 * been detached. All of this is necessary because copyout() may sleep on
1481 unp_pcblist(SYSCTL_HANDLER_ARGS)
1485 struct unpcb *unp, **unp_list;
1487 struct xunpgen *xug;
1488 struct unp_head *head;
1491 switch ((intptr_t)arg1) {
1500 case SOCK_SEQPACKET:
1505 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1509 * The process of preparing the PCB list is too time-consuming and
1510 * resource-intensive to repeat twice on every request.
1512 if (req->oldptr == NULL) {
1514 req->oldidx = 2 * (sizeof *xug)
1515 + (n + n/8) * sizeof(struct xunpcb);
1519 if (req->newptr != NULL)
1523 * OK, now we're committed to doing something.
1525 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1527 gencnt = unp_gencnt;
1531 xug->xug_len = sizeof *xug;
1533 xug->xug_gen = gencnt;
1534 xug->xug_sogen = so_gencnt;
1535 error = SYSCTL_OUT(req, xug, sizeof *xug);
1541 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1544 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1545 unp = LIST_NEXT(unp, unp_link)) {
1547 if (unp->unp_gencnt <= gencnt) {
1548 if (cr_cansee(req->td->td_ucred,
1549 unp->unp_socket->so_cred)) {
1550 UNP_PCB_UNLOCK(unp);
1553 unp_list[i++] = unp;
1554 unp->unp_refcount++;
1556 UNP_PCB_UNLOCK(unp);
1559 n = i; /* In case we lost some during malloc. */
1562 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1563 for (i = 0; i < n; i++) {
1566 unp->unp_refcount--;
1567 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1568 xu->xu_len = sizeof *xu;
1571 * XXX - need more locking here to protect against
1572 * connect/disconnect races for SMP.
1574 if (unp->unp_addr != NULL)
1575 bcopy(unp->unp_addr, &xu->xu_addr,
1576 unp->unp_addr->sun_len);
1577 if (unp->unp_conn != NULL &&
1578 unp->unp_conn->unp_addr != NULL)
1579 bcopy(unp->unp_conn->unp_addr,
1581 unp->unp_conn->unp_addr->sun_len);
1582 bcopy(unp, &xu->xu_unp, sizeof *unp);
1583 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1584 UNP_PCB_UNLOCK(unp);
1585 error = SYSCTL_OUT(req, xu, sizeof *xu);
1587 freeunp = (unp->unp_refcount == 0);
1588 UNP_PCB_UNLOCK(unp);
1590 UNP_PCB_LOCK_DESTROY(unp);
1591 uma_zfree(unp_zone, unp);
1598 * Give the user an updated idea of our state. If the
1599 * generation differs from what we told her before, she knows
1600 * that something happened while we were processing this
1601 * request, and it might be necessary to retry.
1603 xug->xug_gen = unp_gencnt;
1604 xug->xug_sogen = so_gencnt;
1605 xug->xug_count = unp_count;
1606 error = SYSCTL_OUT(req, xug, sizeof *xug);
1608 free(unp_list, M_TEMP);
1613 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1614 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1615 "List of active local datagram sockets");
1616 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1617 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1618 "List of active local stream sockets");
1619 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1620 CTLTYPE_OPAQUE | CTLFLAG_RD,
1621 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1622 "List of active local seqpacket sockets");
1625 unp_shutdown(struct unpcb *unp)
1630 UNP_LINK_WLOCK_ASSERT();
1631 UNP_PCB_LOCK_ASSERT(unp);
1633 unp2 = unp->unp_conn;
1634 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1635 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1636 so = unp2->unp_socket;
1643 unp_drop(struct unpcb *unp, int errno)
1645 struct socket *so = unp->unp_socket;
1648 UNP_LINK_WLOCK_ASSERT();
1649 UNP_PCB_LOCK_ASSERT(unp);
1651 so->so_error = errno;
1652 unp2 = unp->unp_conn;
1656 unp_disconnect(unp, unp2);
1657 UNP_PCB_UNLOCK(unp2);
1661 unp_freerights(struct file **rp, int fdcount)
1666 for (i = 0; i < fdcount; i++) {
1674 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1676 struct thread *td = curthread; /* XXX */
1677 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1683 socklen_t clen = control->m_len, datalen;
1688 UNP_LINK_UNLOCK_ASSERT();
1691 if (controlp != NULL) /* controlp == NULL => free control messages */
1693 while (cm != NULL) {
1694 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1698 data = CMSG_DATA(cm);
1699 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1700 if (cm->cmsg_level == SOL_SOCKET
1701 && cm->cmsg_type == SCM_RIGHTS) {
1702 newfds = datalen / sizeof(struct file *);
1705 /* If we're not outputting the descriptors free them. */
1706 if (error || controlp == NULL) {
1707 unp_freerights(rp, newfds);
1710 FILEDESC_XLOCK(td->td_proc->p_fd);
1711 /* if the new FD's will not fit free them. */
1712 if (!fdavail(td, newfds)) {
1713 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1715 unp_freerights(rp, newfds);
1720 * Now change each pointer to an fd in the global
1721 * table to an integer that is the index to the local
1722 * fd table entry that we set up to point to the
1723 * global one we are transferring.
1725 newlen = newfds * sizeof(int);
1726 *controlp = sbcreatecontrol(NULL, newlen,
1727 SCM_RIGHTS, SOL_SOCKET);
1728 if (*controlp == NULL) {
1729 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1731 unp_freerights(rp, newfds);
1736 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1737 for (i = 0; i < newfds; i++) {
1738 if (fdalloc(td, 0, &f))
1739 panic("unp_externalize fdalloc failed");
1741 td->td_proc->p_fd->fd_ofiles[f] = fp;
1742 unp_externalize_fp(fp);
1745 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1747 /* We can just copy anything else across. */
1748 if (error || controlp == NULL)
1750 *controlp = sbcreatecontrol(NULL, datalen,
1751 cm->cmsg_type, cm->cmsg_level);
1752 if (*controlp == NULL) {
1757 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1760 controlp = &(*controlp)->m_next;
1763 if (CMSG_SPACE(datalen) < clen) {
1764 clen -= CMSG_SPACE(datalen);
1765 cm = (struct cmsghdr *)
1766 ((caddr_t)cm + CMSG_SPACE(datalen));
1778 unp_zone_change(void *tag)
1781 uma_zone_set_max(unp_zone, maxsockets);
1789 if (!IS_DEFAULT_VNET(curvnet))
1792 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1793 NULL, NULL, UMA_ALIGN_PTR, 0);
1794 if (unp_zone == NULL)
1796 uma_zone_set_max(unp_zone, maxsockets);
1797 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1798 NULL, EVENTHANDLER_PRI_ANY);
1799 LIST_INIT(&unp_dhead);
1800 LIST_INIT(&unp_shead);
1801 LIST_INIT(&unp_sphead);
1802 SLIST_INIT(&unp_defers);
1803 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL);
1804 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1805 UNP_LINK_LOCK_INIT();
1806 UNP_LIST_LOCK_INIT();
1807 UNP_DEFERRED_LOCK_INIT();
1811 unp_internalize(struct mbuf **controlp, struct thread *td)
1813 struct mbuf *control = *controlp;
1814 struct proc *p = td->td_proc;
1815 struct filedesc *fdescp = p->p_fd;
1816 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1817 struct cmsgcred *cmcred;
1823 socklen_t clen = control->m_len, datalen;
1827 UNP_LINK_UNLOCK_ASSERT();
1831 while (cm != NULL) {
1832 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1833 || cm->cmsg_len > clen) {
1837 data = CMSG_DATA(cm);
1838 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1840 switch (cm->cmsg_type) {
1842 * Fill in credential information.
1845 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1846 SCM_CREDS, SOL_SOCKET);
1847 if (*controlp == NULL) {
1851 cmcred = (struct cmsgcred *)
1852 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1853 cmcred->cmcred_pid = p->p_pid;
1854 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1855 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1856 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1857 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1859 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1860 cmcred->cmcred_groups[i] =
1861 td->td_ucred->cr_groups[i];
1865 oldfds = datalen / sizeof (int);
1867 * Check that all the FDs passed in refer to legal
1868 * files. If not, reject the entire operation.
1871 FILEDESC_SLOCK(fdescp);
1872 for (i = 0; i < oldfds; i++) {
1874 if ((unsigned)fd >= fdescp->fd_nfiles ||
1875 fdescp->fd_ofiles[fd] == NULL) {
1876 FILEDESC_SUNLOCK(fdescp);
1880 fp = fdescp->fd_ofiles[fd];
1881 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1882 FILEDESC_SUNLOCK(fdescp);
1890 * Now replace the integer FDs with pointers to the
1891 * associated global file table entry..
1893 newlen = oldfds * sizeof(struct file *);
1894 *controlp = sbcreatecontrol(NULL, newlen,
1895 SCM_RIGHTS, SOL_SOCKET);
1896 if (*controlp == NULL) {
1897 FILEDESC_SUNLOCK(fdescp);
1902 rp = (struct file **)
1903 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1904 for (i = 0; i < oldfds; i++) {
1905 fp = fdescp->fd_ofiles[*fdp++];
1907 unp_internalize_fp(fp);
1909 FILEDESC_SUNLOCK(fdescp);
1913 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1914 SCM_TIMESTAMP, SOL_SOCKET);
1915 if (*controlp == NULL) {
1919 tv = (struct timeval *)
1920 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1929 controlp = &(*controlp)->m_next;
1930 if (CMSG_SPACE(datalen) < clen) {
1931 clen -= CMSG_SPACE(datalen);
1932 cm = (struct cmsghdr *)
1933 ((caddr_t)cm + CMSG_SPACE(datalen));
1945 static struct mbuf *
1946 unp_addsockcred(struct thread *td, struct mbuf *control)
1948 struct mbuf *m, *n, *n_prev;
1949 struct sockcred *sc;
1950 const struct cmsghdr *cm;
1954 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1955 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1959 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1960 sc->sc_uid = td->td_ucred->cr_ruid;
1961 sc->sc_euid = td->td_ucred->cr_uid;
1962 sc->sc_gid = td->td_ucred->cr_rgid;
1963 sc->sc_egid = td->td_ucred->cr_gid;
1964 sc->sc_ngroups = ngroups;
1965 for (i = 0; i < sc->sc_ngroups; i++)
1966 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1969 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
1970 * created SCM_CREDS control message (struct sockcred) has another
1973 if (control != NULL)
1974 for (n = control, n_prev = NULL; n != NULL;) {
1975 cm = mtod(n, struct cmsghdr *);
1976 if (cm->cmsg_level == SOL_SOCKET &&
1977 cm->cmsg_type == SCM_CREDS) {
1979 control = n->m_next;
1981 n_prev->m_next = n->m_next;
1989 /* Prepend it to the head. */
1990 m->m_next = control;
1994 static struct unpcb *
1995 fptounp(struct file *fp)
1999 if (fp->f_type != DTYPE_SOCKET)
2001 if ((so = fp->f_data) == NULL)
2003 if (so->so_proto->pr_domain != &localdomain)
2005 return sotounpcb(so);
2009 unp_discard(struct file *fp)
2011 struct unp_defer *dr;
2013 if (unp_externalize_fp(fp)) {
2014 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2016 UNP_DEFERRED_LOCK();
2017 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2018 UNP_DEFERRED_UNLOCK();
2019 atomic_add_int(&unp_defers_count, 1);
2020 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2022 (void) closef(fp, (struct thread *)NULL);
2026 unp_process_defers(void *arg __unused, int pending)
2028 struct unp_defer *dr;
2029 SLIST_HEAD(, unp_defer) drl;
2034 UNP_DEFERRED_LOCK();
2035 if (SLIST_FIRST(&unp_defers) == NULL) {
2036 UNP_DEFERRED_UNLOCK();
2039 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2040 UNP_DEFERRED_UNLOCK();
2042 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2043 SLIST_REMOVE_HEAD(&drl, ud_link);
2044 closef(dr->ud_fp, NULL);
2048 atomic_add_int(&unp_defers_count, -count);
2053 unp_internalize_fp(struct file *fp)
2058 if ((unp = fptounp(fp)) != NULL) {
2060 unp->unp_msgcount++;
2068 unp_externalize_fp(struct file *fp)
2074 if ((unp = fptounp(fp)) != NULL) {
2075 unp->unp_msgcount--;
2085 * unp_defer indicates whether additional work has been defered for a future
2086 * pass through unp_gc(). It is thread local and does not require explicit
2089 static int unp_marked;
2090 static int unp_unreachable;
2093 unp_accessable(struct file *fp)
2097 if ((unp = fptounp(fp)) == NULL)
2099 if (unp->unp_gcflag & UNPGC_REF)
2101 unp->unp_gcflag &= ~UNPGC_DEAD;
2102 unp->unp_gcflag |= UNPGC_REF;
2107 unp_gc_process(struct unpcb *unp)
2113 /* Already processed. */
2114 if (unp->unp_gcflag & UNPGC_SCANNED)
2119 * Check for a socket potentially in a cycle. It must be in a
2120 * queue as indicated by msgcount, and this must equal the file
2121 * reference count. Note that when msgcount is 0 the file is NULL.
2123 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2124 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2125 unp->unp_gcflag |= UNPGC_DEAD;
2131 * Mark all sockets we reference with RIGHTS.
2133 so = unp->unp_socket;
2134 SOCKBUF_LOCK(&so->so_rcv);
2135 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2136 SOCKBUF_UNLOCK(&so->so_rcv);
2139 * Mark all sockets in our accept queue.
2142 TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2143 SOCKBUF_LOCK(&soa->so_rcv);
2144 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2145 SOCKBUF_UNLOCK(&soa->so_rcv);
2148 unp->unp_gcflag |= UNPGC_SCANNED;
2151 static int unp_recycled;
2152 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2153 "Number of unreachable sockets claimed by the garbage collector.");
2155 static int unp_taskcount;
2156 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2157 "Number of times the garbage collector has run.");
2160 unp_gc(__unused void *arg, int pending)
2162 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2164 struct unp_head **head;
2165 struct file *f, **unref;
2172 * First clear all gc flags from previous runs.
2174 for (head = heads; *head != NULL; head++)
2175 LIST_FOREACH(unp, *head, unp_link)
2176 unp->unp_gcflag = 0;
2179 * Scan marking all reachable sockets with UNPGC_REF. Once a socket
2180 * is reachable all of the sockets it references are reachable.
2181 * Stop the scan once we do a complete loop without discovering
2182 * a new reachable socket.
2185 unp_unreachable = 0;
2187 for (head = heads; *head != NULL; head++)
2188 LIST_FOREACH(unp, *head, unp_link)
2189 unp_gc_process(unp);
2190 } while (unp_marked);
2192 if (unp_unreachable == 0)
2196 * Allocate space for a local list of dead unpcbs.
2198 unref = malloc(unp_unreachable * sizeof(struct file *),
2202 * Iterate looking for sockets which have been specifically marked
2203 * as as unreachable and store them locally.
2207 for (total = 0, head = heads; *head != NULL; head++)
2208 LIST_FOREACH(unp, *head, unp_link)
2209 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2211 if (unp->unp_msgcount == 0 || f == NULL ||
2212 f->f_count != unp->unp_msgcount)
2216 KASSERT(total <= unp_unreachable,
2217 ("unp_gc: incorrect unreachable count."));
2223 * Now flush all sockets, free'ing rights. This will free the
2224 * struct files associated with these sockets but leave each socket
2225 * with one remaining ref.
2227 for (i = 0; i < total; i++) {
2230 so = unref[i]->f_data;
2231 CURVNET_SET(so->so_vnet);
2237 * And finally release the sockets so they can be reclaimed.
2239 for (i = 0; i < total; i++)
2240 fdrop(unref[i], NULL);
2241 unp_recycled += total;
2242 free(unref, M_TEMP);
2246 unp_dispose(struct mbuf *m)
2250 unp_scan(m, unp_discard);
2254 unp_scan(struct mbuf *m0, void (*op)(struct file *))
2261 socklen_t clen, datalen;
2264 while (m0 != NULL) {
2265 for (m = m0; m; m = m->m_next) {
2266 if (m->m_type != MT_CONTROL)
2269 cm = mtod(m, struct cmsghdr *);
2272 while (cm != NULL) {
2273 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2276 data = CMSG_DATA(cm);
2277 datalen = (caddr_t)cm + cm->cmsg_len
2280 if (cm->cmsg_level == SOL_SOCKET &&
2281 cm->cmsg_type == SCM_RIGHTS) {
2282 qfds = datalen / sizeof (struct file *);
2284 for (i = 0; i < qfds; i++)
2288 if (CMSG_SPACE(datalen) < clen) {
2289 clen -= CMSG_SPACE(datalen);
2290 cm = (struct cmsghdr *)
2291 ((caddr_t)cm + CMSG_SPACE(datalen));
2304 db_print_indent(int indent)
2308 for (i = 0; i < indent; i++)
2313 db_print_unpflags(int unp_flags)
2318 if (unp_flags & UNP_HAVEPC) {
2319 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2322 if (unp_flags & UNP_HAVEPCCACHED) {
2323 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2326 if (unp_flags & UNP_WANTCRED) {
2327 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2330 if (unp_flags & UNP_CONNWAIT) {
2331 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2334 if (unp_flags & UNP_CONNECTING) {
2335 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2338 if (unp_flags & UNP_BINDING) {
2339 db_printf("%sUNP_BINDING", comma ? ", " : "");
2345 db_print_xucred(int indent, struct xucred *xu)
2349 db_print_indent(indent);
2350 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
2351 xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2352 db_print_indent(indent);
2353 db_printf("cr_groups: ");
2355 for (i = 0; i < xu->cr_ngroups; i++) {
2356 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2363 db_print_unprefs(int indent, struct unp_head *uh)
2369 LIST_FOREACH(unp, uh, unp_reflink) {
2370 if (counter % 4 == 0)
2371 db_print_indent(indent);
2372 db_printf("%p ", unp);
2373 if (counter % 4 == 3)
2377 if (counter != 0 && counter % 4 != 0)
2381 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2386 db_printf("usage: show unpcb <addr>\n");
2389 unp = (struct unpcb *)addr;
2391 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2394 db_printf("unp_ino: %d unp_conn: %p\n", unp->unp_ino,
2397 db_printf("unp_refs:\n");
2398 db_print_unprefs(2, &unp->unp_refs);
2400 /* XXXRW: Would be nice to print the full address, if any. */
2401 db_printf("unp_addr: %p\n", unp->unp_addr);
2403 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n",
2404 unp->unp_cc, unp->unp_mbcnt,
2405 (unsigned long long)unp->unp_gencnt);
2407 db_printf("unp_flags: %x (", unp->unp_flags);
2408 db_print_unpflags(unp->unp_flags);
2411 db_printf("unp_peercred:\n");
2412 db_print_xucred(2, &unp->unp_peercred);
2414 db_printf("unp_refcount: %u\n", unp->unp_refcount);