2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All Rights Reserved.
6 * Copyright (c) 2004-2009 Robert N. M. Watson All Rights Reserved.
7 * Copyright (c) 2018 Matthew Macy
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
37 * UNIX Domain (Local) Sockets
39 * This is an implementation of UNIX (local) domain sockets. Each socket has
40 * an associated struct unpcb (UNIX protocol control block). Stream sockets
41 * may be connected to 0 or 1 other socket. Datagram sockets may be
42 * connected to 0, 1, or many other sockets. Sockets may be created and
43 * connected in pairs (socketpair(2)), or bound/connected to using the file
44 * system name space. For most purposes, only the receive socket buffer is
45 * used, as sending on one socket delivers directly to the receive socket
46 * buffer of a second socket.
48 * The implementation is substantially complicated by the fact that
49 * "ancillary data", such as file descriptors or credentials, may be passed
50 * across UNIX domain sockets. The potential for passing UNIX domain sockets
51 * over other UNIX domain sockets requires the implementation of a simple
52 * garbage collector to find and tear down cycles of disconnected sockets.
56 * rethink name space problems
57 * need a proper out-of-band
60 #include <sys/cdefs.h>
61 __FBSDID("$FreeBSD$");
65 #include <sys/param.h>
66 #include <sys/capsicum.h>
67 #include <sys/domain.h>
68 #include <sys/eventhandler.h>
69 #include <sys/fcntl.h>
71 #include <sys/filedesc.h>
72 #include <sys/kernel.h>
74 #include <sys/malloc.h>
76 #include <sys/mount.h>
77 #include <sys/mutex.h>
78 #include <sys/namei.h>
80 #include <sys/protosw.h>
81 #include <sys/queue.h>
82 #include <sys/resourcevar.h>
83 #include <sys/rwlock.h>
84 #include <sys/socket.h>
85 #include <sys/socketvar.h>
86 #include <sys/signalvar.h>
89 #include <sys/sysctl.h>
90 #include <sys/systm.h>
91 #include <sys/taskqueue.h>
93 #include <sys/unpcb.h>
94 #include <sys/vnode.h>
102 #include <security/mac/mac_framework.h>
106 MALLOC_DECLARE(M_FILECAPS);
109 * See unpcb.h for the locking key.
112 static uma_zone_t unp_zone;
113 static unp_gen_t unp_gencnt; /* (l) */
114 static u_int unp_count; /* (l) Count of local sockets. */
115 static ino_t unp_ino; /* Prototype for fake inode numbers. */
116 static int unp_rights; /* (g) File descriptors in flight. */
117 static struct unp_head unp_shead; /* (l) List of stream sockets. */
118 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */
119 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */
122 SLIST_ENTRY(unp_defer) ud_link;
125 static SLIST_HEAD(, unp_defer) unp_defers;
126 static int unp_defers_count;
128 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
131 * Garbage collection of cyclic file descriptor/socket references occurs
132 * asynchronously in a taskqueue context in order to avoid recursion and
133 * reentrance in the UNIX domain socket, file descriptor, and socket layer
134 * code. See unp_gc() for a full description.
136 static struct timeout_task unp_gc_task;
139 * The close of unix domain sockets attached as SCM_RIGHTS is
140 * postponed to the taskqueue, to avoid arbitrary recursion depth.
141 * The attached sockets might have another sockets attached.
143 static struct task unp_defer_task;
146 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
147 * stream sockets, although the total for sender and receiver is actually
150 * Datagram sockets really use the sendspace as the maximum datagram size,
151 * and don't really want to reserve the sendspace. Their recvspace should be
152 * large enough for at least one max-size datagram plus address.
157 static u_long unpst_sendspace = PIPSIZ;
158 static u_long unpst_recvspace = PIPSIZ;
159 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
160 static u_long unpdg_recvspace = 16*1024; /* support 8KB syslog msgs */
161 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */
162 static u_long unpsp_recvspace = PIPSIZ;
164 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
166 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream,
167 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
169 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram,
170 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
172 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket,
173 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
176 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
177 &unpst_sendspace, 0, "Default stream send space.");
178 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
179 &unpst_recvspace, 0, "Default stream receive space.");
180 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
181 &unpdg_sendspace, 0, "Default datagram send space.");
182 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
183 &unpdg_recvspace, 0, "Default datagram receive space.");
184 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
185 &unpsp_sendspace, 0, "Default seqpacket send space.");
186 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
187 &unpsp_recvspace, 0, "Default seqpacket receive space.");
188 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
189 "File descriptors in flight.");
190 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
191 &unp_defers_count, 0,
192 "File descriptors deferred to taskqueue for close.");
195 * Locking and synchronization:
197 * Several types of locks exist in the local domain socket implementation:
198 * - a global linkage lock
199 * - a global connection list lock
201 * - per-unpcb mutexes
203 * The linkage lock protects the global socket lists, the generation number
204 * counter and garbage collector state.
206 * The connection list lock protects the list of referring sockets in a datagram
207 * socket PCB. This lock is also overloaded to protect a global list of
208 * sockets whose buffers contain socket references in the form of SCM_RIGHTS
209 * messages. To avoid recursion, such references are released by a dedicated
212 * The mtxpool lock protects the vnode from being modified while referenced.
213 * Lock ordering rules require that it be acquired before any PCB locks.
215 * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the
216 * unpcb. This includes the unp_conn field, which either links two connected
217 * PCBs together (for connected socket types) or points at the destination
218 * socket (for connectionless socket types). The operations of creating or
219 * destroying a connection therefore involve locking multiple PCBs. To avoid
220 * lock order reversals, in some cases this involves dropping a PCB lock and
221 * using a reference counter to maintain liveness.
223 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
224 * allocated in pru_attach() and freed in pru_detach(). The validity of that
225 * pointer is an invariant, so no lock is required to dereference the so_pcb
226 * pointer if a valid socket reference is held by the caller. In practice,
227 * this is always true during operations performed on a socket. Each unpcb
228 * has a back-pointer to its socket, unp_socket, which will be stable under
229 * the same circumstances.
231 * This pointer may only be safely dereferenced as long as a valid reference
232 * to the unpcb is held. Typically, this reference will be from the socket,
233 * or from another unpcb when the referring unpcb's lock is held (in order
234 * that the reference not be invalidated during use). For example, to follow
235 * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee
236 * that detach is not run clearing unp_socket.
238 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
239 * protocols, bind() is a non-atomic operation, and connect() requires
240 * potential sleeping in the protocol, due to potentially waiting on local or
241 * distributed file systems. We try to separate "lookup" operations, which
242 * may sleep, and the IPC operations themselves, which typically can occur
243 * with relative atomicity as locks can be held over the entire operation.
245 * Another tricky issue is simultaneous multi-threaded or multi-process
246 * access to a single UNIX domain socket. These are handled by the flags
247 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
248 * binding, both of which involve dropping UNIX domain socket locks in order
249 * to perform namei() and other file system operations.
251 static struct rwlock unp_link_rwlock;
252 static struct mtx unp_defers_lock;
254 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
257 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
259 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
262 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
263 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
264 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
265 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
266 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
268 #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock)
270 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
271 "unp_defer", NULL, MTX_DEF)
272 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
273 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
275 #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK();
276 #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK();
278 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
281 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
282 #define UNP_PCB_LOCKPTR(unp) (&(unp)->unp_mtx)
283 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
284 #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx)
285 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
286 #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx)
287 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
288 #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED)
290 static int uipc_connect2(struct socket *, struct socket *);
291 static int uipc_ctloutput(struct socket *, struct sockopt *);
292 static int unp_connect(struct socket *, struct sockaddr *,
294 static int unp_connectat(int, struct socket *, struct sockaddr *,
296 static int unp_connect2(struct socket *so, struct socket *so2, int);
297 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
298 static void unp_dispose(struct socket *so);
299 static void unp_dispose_mbuf(struct mbuf *);
300 static void unp_shutdown(struct unpcb *);
301 static void unp_drop(struct unpcb *);
302 static void unp_gc(__unused void *, int);
303 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
304 static void unp_discard(struct file *);
305 static void unp_freerights(struct filedescent **, int);
306 static int unp_internalize(struct mbuf **, struct thread *);
307 static void unp_internalize_fp(struct file *);
308 static int unp_externalize(struct mbuf *, struct mbuf **, int);
309 static int unp_externalize_fp(struct file *);
310 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *, int);
311 static void unp_process_defers(void * __unused, int);
314 unp_pcb_hold(struct unpcb *unp)
318 old = refcount_acquire(&unp->unp_refcount);
319 KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp));
322 static __result_use_check bool
323 unp_pcb_rele(struct unpcb *unp)
327 UNP_PCB_LOCK_ASSERT(unp);
329 if ((ret = refcount_release(&unp->unp_refcount))) {
331 UNP_PCB_LOCK_DESTROY(unp);
332 uma_zfree(unp_zone, unp);
338 unp_pcb_rele_notlast(struct unpcb *unp)
342 ret = refcount_release(&unp->unp_refcount);
343 KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp));
347 unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2)
349 UNP_PCB_UNLOCK_ASSERT(unp);
350 UNP_PCB_UNLOCK_ASSERT(unp2);
354 } else if ((uintptr_t)unp2 > (uintptr_t)unp) {
364 unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2)
368 UNP_PCB_UNLOCK(unp2);
372 * Try to lock the connected peer of an already locked socket. In some cases
373 * this requires that we unlock the current socket. The pairbusy counter is
374 * used to block concurrent connection attempts while the lock is dropped. The
375 * caller must be careful to revalidate PCB state.
377 static struct unpcb *
378 unp_pcb_lock_peer(struct unpcb *unp)
382 UNP_PCB_LOCK_ASSERT(unp);
383 unp2 = unp->unp_conn;
386 if (__predict_false(unp == unp2))
389 UNP_PCB_UNLOCK_ASSERT(unp2);
391 if (__predict_true(UNP_PCB_TRYLOCK(unp2)))
393 if ((uintptr_t)unp2 > (uintptr_t)unp) {
403 KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL,
404 ("%s: socket %p was reconnected", __func__, unp));
405 if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) {
406 unp->unp_flags &= ~UNP_WAITING;
409 if (unp_pcb_rele(unp2)) {
410 /* unp2 is unlocked. */
413 if (unp->unp_conn == NULL) {
414 UNP_PCB_UNLOCK(unp2);
421 * Definitions of protocols supported in the LOCAL domain.
423 static struct domain localdomain;
424 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
425 static struct pr_usrreqs uipc_usrreqs_seqpacket;
426 static struct protosw localsw[] = {
428 .pr_type = SOCK_STREAM,
429 .pr_domain = &localdomain,
430 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS|
432 .pr_ctloutput = &uipc_ctloutput,
433 .pr_usrreqs = &uipc_usrreqs_stream
436 .pr_type = SOCK_DGRAM,
437 .pr_domain = &localdomain,
438 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS|PR_CAPATTACH,
439 .pr_ctloutput = &uipc_ctloutput,
440 .pr_usrreqs = &uipc_usrreqs_dgram
443 .pr_type = SOCK_SEQPACKET,
444 .pr_domain = &localdomain,
447 * XXXRW: For now, PR_ADDR because soreceive will bump into them
448 * due to our use of sbappendaddr. A new sbappend variants is needed
449 * that supports both atomic record writes and control data.
451 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|
452 PR_WANTRCVD|PR_RIGHTS|PR_CAPATTACH,
453 .pr_ctloutput = &uipc_ctloutput,
454 .pr_usrreqs = &uipc_usrreqs_seqpacket,
458 static struct domain localdomain = {
459 .dom_family = AF_LOCAL,
461 .dom_externalize = unp_externalize,
462 .dom_dispose = unp_dispose,
463 .dom_protosw = localsw,
464 .dom_protoswNPROTOSW = &localsw[nitems(localsw)]
469 uipc_abort(struct socket *so)
471 struct unpcb *unp, *unp2;
474 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
475 UNP_PCB_UNLOCK_ASSERT(unp);
478 unp2 = unp->unp_conn;
488 uipc_accept(struct socket *so, struct sockaddr **nam)
490 struct unpcb *unp, *unp2;
491 const struct sockaddr *sa;
494 * Pass back name of connected socket, if it was bound and we are
495 * still connected (our peer may have closed already!).
498 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
500 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
502 unp2 = unp_pcb_lock_peer(unp);
503 if (unp2 != NULL && unp2->unp_addr != NULL)
504 sa = (struct sockaddr *)unp2->unp_addr;
507 bcopy(sa, *nam, sa->sa_len);
509 unp_pcb_unlock_pair(unp, unp2);
516 uipc_attach(struct socket *so, int proto, struct thread *td)
518 u_long sendspace, recvspace;
523 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
524 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
525 switch (so->so_type) {
527 sendspace = unpst_sendspace;
528 recvspace = unpst_recvspace;
532 sendspace = unpdg_sendspace;
533 recvspace = unpdg_recvspace;
537 sendspace = unpsp_sendspace;
538 recvspace = unpsp_recvspace;
542 panic("uipc_attach");
544 error = soreserve(so, sendspace, recvspace);
548 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
551 LIST_INIT(&unp->unp_refs);
552 UNP_PCB_LOCK_INIT(unp);
553 unp->unp_socket = so;
555 refcount_init(&unp->unp_refcount, 1);
557 if ((locked = UNP_LINK_WOWNED()) == false)
560 unp->unp_gencnt = ++unp_gencnt;
561 unp->unp_ino = ++unp_ino;
563 switch (so->so_type) {
565 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
569 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
573 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
577 panic("uipc_attach");
587 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
589 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
599 if (nam->sa_family != AF_UNIX)
600 return (EAFNOSUPPORT);
603 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
605 if (soun->sun_len > sizeof(struct sockaddr_un))
607 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
612 * We don't allow simultaneous bind() calls on a single UNIX domain
613 * socket, so flag in-progress operations, and return an error if an
614 * operation is already in progress.
616 * Historically, we have not allowed a socket to be rebound, so this
617 * also returns an error. Not allowing re-binding simplifies the
618 * implementation and avoids a great many possible failure modes.
621 if (unp->unp_vnode != NULL) {
625 if (unp->unp_flags & UNP_BINDING) {
629 unp->unp_flags |= UNP_BINDING;
632 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
633 bcopy(soun->sun_path, buf, namelen);
637 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
638 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT));
639 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
644 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
645 NDFREE(&nd, NDF_ONLY_PNBUF);
655 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
661 vattr.va_type = VSOCK;
662 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_pd->pd_cmask);
664 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
668 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
669 NDFREE(&nd, NDF_ONLY_PNBUF);
671 VOP_VPUT_PAIR(nd.ni_dvp, NULL, true);
672 vn_finished_write(mp);
673 if (error == ERELOOKUP)
678 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
679 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
682 VOP_UNP_BIND(vp, unp);
684 unp->unp_addr = soun;
685 unp->unp_flags &= ~UNP_BINDING;
688 VOP_VPUT_PAIR(nd.ni_dvp, &vp, true);
689 vn_finished_write(mp);
695 unp->unp_flags &= ~UNP_BINDING;
702 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
705 return (uipc_bindat(AT_FDCWD, so, nam, td));
709 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
713 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
714 error = unp_connect(so, nam, td);
719 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
724 KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
725 error = unp_connectat(fd, so, nam, td);
730 uipc_close(struct socket *so)
732 struct unpcb *unp, *unp2;
733 struct vnode *vp = NULL;
737 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
740 if ((vp = unp->unp_vnode) != NULL) {
741 vplock = mtx_pool_find(mtxpool_sleep, vp);
745 if (vp && unp->unp_vnode == NULL) {
751 unp->unp_vnode = NULL;
753 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
754 unp_disconnect(unp, unp2);
764 uipc_connect2(struct socket *so1, struct socket *so2)
766 struct unpcb *unp, *unp2;
770 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
772 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
773 unp_pcb_lock_pair(unp, unp2);
774 error = unp_connect2(so1, so2, PRU_CONNECT2);
775 unp_pcb_unlock_pair(unp, unp2);
780 uipc_detach(struct socket *so)
782 struct unpcb *unp, *unp2;
785 int local_unp_rights;
788 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
794 if (!SOLISTENING(so)) {
796 * Once the socket is removed from the global lists,
797 * uipc_ready() will not be able to locate its socket buffer, so
798 * clear the buffer now. At this point internalized rights have
799 * already been disposed of.
801 sbrelease(&so->so_rcv, so);
806 LIST_REMOVE(unp, unp_link);
807 if (unp->unp_gcflag & UNPGC_DEAD)
808 LIST_REMOVE(unp, unp_dead);
809 unp->unp_gencnt = ++unp_gencnt;
813 UNP_PCB_UNLOCK_ASSERT(unp);
815 if ((vp = unp->unp_vnode) != NULL) {
816 vplock = mtx_pool_find(mtxpool_sleep, vp);
820 if (unp->unp_vnode != vp && unp->unp_vnode != NULL) {
826 if ((vp = unp->unp_vnode) != NULL) {
828 unp->unp_vnode = NULL;
830 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
831 unp_disconnect(unp, unp2);
836 while (!LIST_EMPTY(&unp->unp_refs)) {
837 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
840 UNP_REF_LIST_UNLOCK();
843 UNP_PCB_UNLOCK_ASSERT(ref);
847 UNP_REF_LIST_UNLOCK();
850 local_unp_rights = unp_rights;
851 unp->unp_socket->so_pcb = NULL;
852 unp->unp_socket = NULL;
853 free(unp->unp_addr, M_SONAME);
854 unp->unp_addr = NULL;
855 if (!unp_pcb_rele(unp))
861 if (local_unp_rights)
862 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
866 uipc_disconnect(struct socket *so)
868 struct unpcb *unp, *unp2;
871 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
874 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
875 unp_disconnect(unp, unp2);
882 uipc_listen(struct socket *so, int backlog, struct thread *td)
887 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
891 * Synchronize with concurrent connection attempts.
896 if (unp->unp_conn != NULL || (unp->unp_flags & UNP_CONNECTING) != 0)
898 else if (unp->unp_vnode == NULL)
899 error = EDESTADDRREQ;
906 error = solisten_proto_check(so);
908 cru2xt(td, &unp->unp_peercred);
909 solisten_proto(so, backlog);
917 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
919 struct unpcb *unp, *unp2;
920 const struct sockaddr *sa;
923 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
925 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
928 * XXX: It seems that this test always fails even when connection is
929 * established. So, this else clause is added as workaround to
930 * return PF_LOCAL sockaddr.
932 unp2 = unp->unp_conn;
935 if (unp2->unp_addr != NULL)
936 sa = (struct sockaddr *) unp2->unp_addr;
939 bcopy(sa, *nam, sa->sa_len);
940 UNP_PCB_UNLOCK(unp2);
943 bcopy(sa, *nam, sa->sa_len);
950 uipc_rcvd(struct socket *so, int flags)
952 struct unpcb *unp, *unp2;
957 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
958 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
959 ("%s: socktype %d", __func__, so->so_type));
962 * Adjust backpressure on sender and wakeup any waiting to write.
964 * The unp lock is acquired to maintain the validity of the unp_conn
965 * pointer; no lock on unp2 is required as unp2->unp_socket will be
966 * static as long as we don't permit unp2 to disconnect from unp,
967 * which is prevented by the lock on unp. We cache values from
968 * so_rcv to avoid holding the so_rcv lock over the entire
969 * transaction on the remote so_snd.
971 SOCKBUF_LOCK(&so->so_rcv);
972 mbcnt = so->so_rcv.sb_mbcnt;
973 sbcc = sbavail(&so->so_rcv);
974 SOCKBUF_UNLOCK(&so->so_rcv);
976 * There is a benign race condition at this point. If we're planning to
977 * clear SB_STOP, but uipc_send is called on the connected socket at
978 * this instant, it might add data to the sockbuf and set SB_STOP. Then
979 * we would erroneously clear SB_STOP below, even though the sockbuf is
980 * full. The race is benign because the only ill effect is to allow the
981 * sockbuf to exceed its size limit, and the size limits are not
982 * strictly guaranteed anyway.
985 unp2 = unp->unp_conn;
990 so2 = unp2->unp_socket;
991 SOCKBUF_LOCK(&so2->so_snd);
992 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
993 so2->so_snd.sb_flags &= ~SB_STOP;
994 sowwakeup_locked(so2);
1000 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
1001 struct mbuf *control, struct thread *td)
1003 struct unpcb *unp, *unp2;
1008 unp = sotounpcb(so);
1009 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
1010 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
1011 so->so_type == SOCK_SEQPACKET,
1012 ("%s: socktype %d", __func__, so->so_type));
1015 if (flags & PRUS_OOB) {
1019 if (control != NULL && (error = unp_internalize(&control, td)))
1023 switch (so->so_type) {
1026 const struct sockaddr *from;
1029 error = unp_connect(so, nam, td);
1036 * Because connect() and send() are non-atomic in a sendto()
1037 * with a target address, it's possible that the socket will
1038 * have disconnected before the send() can run. In that case
1039 * return the slightly counter-intuitive but otherwise
1040 * correct error that the socket is not connected.
1042 unp2 = unp_pcb_lock_peer(unp);
1044 UNP_PCB_UNLOCK(unp);
1049 if (unp2->unp_flags & UNP_WANTCRED_MASK)
1050 control = unp_addsockcred(td, control,
1052 if (unp->unp_addr != NULL)
1053 from = (struct sockaddr *)unp->unp_addr;
1056 so2 = unp2->unp_socket;
1057 SOCKBUF_LOCK(&so2->so_rcv);
1058 if (sbappendaddr_locked(&so2->so_rcv, from, m,
1060 sorwakeup_locked(so2);
1064 soroverflow_locked(so2);
1065 error = (so->so_state & SS_NBIO) ? EAGAIN : ENOBUFS;
1068 unp_disconnect(unp, unp2);
1070 unp_pcb_unlock_pair(unp, unp2);
1074 case SOCK_SEQPACKET:
1076 if ((so->so_state & SS_ISCONNECTED) == 0) {
1078 error = unp_connect(so, nam, td);
1088 if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) {
1089 UNP_PCB_UNLOCK(unp);
1092 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1093 unp_pcb_unlock_pair(unp, unp2);
1097 UNP_PCB_UNLOCK(unp);
1098 if ((so2 = unp2->unp_socket) == NULL) {
1099 UNP_PCB_UNLOCK(unp2);
1103 SOCKBUF_LOCK(&so2->so_rcv);
1104 if (unp2->unp_flags & UNP_WANTCRED_MASK) {
1106 * Credentials are passed only once on SOCK_STREAM and
1107 * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or
1108 * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS).
1110 control = unp_addsockcred(td, control, unp2->unp_flags);
1111 unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT;
1115 * Send to paired receive port and wake up readers. Don't
1116 * check for space available in the receive buffer if we're
1117 * attaching ancillary data; Unix domain sockets only check
1118 * for space in the sending sockbuf, and that check is
1119 * performed one level up the stack. At that level we cannot
1120 * precisely account for the amount of buffer space used
1121 * (e.g., because control messages are not yet internalized).
1123 switch (so->so_type) {
1125 if (control != NULL) {
1126 sbappendcontrol_locked(&so2->so_rcv, m,
1130 sbappend_locked(&so2->so_rcv, m, flags);
1133 case SOCK_SEQPACKET:
1134 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1135 &sun_noname, m, control))
1140 mbcnt = so2->so_rcv.sb_mbcnt;
1141 sbcc = sbavail(&so2->so_rcv);
1143 sorwakeup_locked(so2);
1145 SOCKBUF_UNLOCK(&so2->so_rcv);
1148 * The PCB lock on unp2 protects the SB_STOP flag. Without it,
1149 * it would be possible for uipc_rcvd to be called at this
1150 * point, drain the receiving sockbuf, clear SB_STOP, and then
1151 * we would set SB_STOP below. That could lead to an empty
1152 * sockbuf having SB_STOP set
1154 SOCKBUF_LOCK(&so->so_snd);
1155 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1156 so->so_snd.sb_flags |= SB_STOP;
1157 SOCKBUF_UNLOCK(&so->so_snd);
1158 UNP_PCB_UNLOCK(unp2);
1164 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1166 if (flags & PRUS_EOF) {
1170 UNP_PCB_UNLOCK(unp);
1172 if (control != NULL && error != 0)
1173 unp_dispose_mbuf(control);
1176 if (control != NULL)
1179 * In case of PRUS_NOTREADY, uipc_ready() is responsible
1180 * for freeing memory.
1182 if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1188 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp)
1190 struct mbuf *mb, *n;
1194 if (SOLISTENING(so)) {
1201 if (sb->sb_fnrdy != NULL) {
1202 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) {
1204 *errorp = sbready(sb, m, count);
1217 return (mb != NULL);
1221 uipc_ready(struct socket *so, struct mbuf *m, int count)
1223 struct unpcb *unp, *unp2;
1227 unp = sotounpcb(so);
1229 KASSERT(so->so_type == SOCK_STREAM,
1230 ("%s: unexpected socket type for %p", __func__, so));
1233 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
1234 UNP_PCB_UNLOCK(unp);
1235 so2 = unp2->unp_socket;
1236 SOCKBUF_LOCK(&so2->so_rcv);
1237 if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1238 sorwakeup_locked(so2);
1240 SOCKBUF_UNLOCK(&so2->so_rcv);
1241 UNP_PCB_UNLOCK(unp2);
1244 UNP_PCB_UNLOCK(unp);
1247 * The receiving socket has been disconnected, but may still be valid.
1248 * In this case, the now-ready mbufs are still present in its socket
1249 * buffer, so perform an exhaustive search before giving up and freeing
1253 LIST_FOREACH(unp, &unp_shead, unp_link) {
1254 if (uipc_ready_scan(unp->unp_socket, m, count, &error))
1260 for (i = 0; i < count; i++)
1268 uipc_sense(struct socket *so, struct stat *sb)
1272 unp = sotounpcb(so);
1273 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1275 sb->st_blksize = so->so_snd.sb_hiwat;
1277 sb->st_ino = unp->unp_ino;
1282 uipc_shutdown(struct socket *so)
1286 unp = sotounpcb(so);
1287 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1292 UNP_PCB_UNLOCK(unp);
1297 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1300 const struct sockaddr *sa;
1302 unp = sotounpcb(so);
1303 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1305 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1307 if (unp->unp_addr != NULL)
1308 sa = (struct sockaddr *) unp->unp_addr;
1311 bcopy(sa, *nam, sa->sa_len);
1312 UNP_PCB_UNLOCK(unp);
1316 static struct pr_usrreqs uipc_usrreqs_dgram = {
1317 .pru_abort = uipc_abort,
1318 .pru_accept = uipc_accept,
1319 .pru_attach = uipc_attach,
1320 .pru_bind = uipc_bind,
1321 .pru_bindat = uipc_bindat,
1322 .pru_connect = uipc_connect,
1323 .pru_connectat = uipc_connectat,
1324 .pru_connect2 = uipc_connect2,
1325 .pru_detach = uipc_detach,
1326 .pru_disconnect = uipc_disconnect,
1327 .pru_listen = uipc_listen,
1328 .pru_peeraddr = uipc_peeraddr,
1329 .pru_rcvd = uipc_rcvd,
1330 .pru_send = uipc_send,
1331 .pru_sense = uipc_sense,
1332 .pru_shutdown = uipc_shutdown,
1333 .pru_sockaddr = uipc_sockaddr,
1334 .pru_soreceive = soreceive_dgram,
1335 .pru_close = uipc_close,
1338 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1339 .pru_abort = uipc_abort,
1340 .pru_accept = uipc_accept,
1341 .pru_attach = uipc_attach,
1342 .pru_bind = uipc_bind,
1343 .pru_bindat = uipc_bindat,
1344 .pru_connect = uipc_connect,
1345 .pru_connectat = uipc_connectat,
1346 .pru_connect2 = uipc_connect2,
1347 .pru_detach = uipc_detach,
1348 .pru_disconnect = uipc_disconnect,
1349 .pru_listen = uipc_listen,
1350 .pru_peeraddr = uipc_peeraddr,
1351 .pru_rcvd = uipc_rcvd,
1352 .pru_send = uipc_send,
1353 .pru_sense = uipc_sense,
1354 .pru_shutdown = uipc_shutdown,
1355 .pru_sockaddr = uipc_sockaddr,
1356 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1357 .pru_close = uipc_close,
1360 static struct pr_usrreqs uipc_usrreqs_stream = {
1361 .pru_abort = uipc_abort,
1362 .pru_accept = uipc_accept,
1363 .pru_attach = uipc_attach,
1364 .pru_bind = uipc_bind,
1365 .pru_bindat = uipc_bindat,
1366 .pru_connect = uipc_connect,
1367 .pru_connectat = uipc_connectat,
1368 .pru_connect2 = uipc_connect2,
1369 .pru_detach = uipc_detach,
1370 .pru_disconnect = uipc_disconnect,
1371 .pru_listen = uipc_listen,
1372 .pru_peeraddr = uipc_peeraddr,
1373 .pru_rcvd = uipc_rcvd,
1374 .pru_send = uipc_send,
1375 .pru_ready = uipc_ready,
1376 .pru_sense = uipc_sense,
1377 .pru_shutdown = uipc_shutdown,
1378 .pru_sockaddr = uipc_sockaddr,
1379 .pru_soreceive = soreceive_generic,
1380 .pru_close = uipc_close,
1384 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1390 if (sopt->sopt_level != SOL_LOCAL)
1393 unp = sotounpcb(so);
1394 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1396 switch (sopt->sopt_dir) {
1398 switch (sopt->sopt_name) {
1399 case LOCAL_PEERCRED:
1401 if (unp->unp_flags & UNP_HAVEPC)
1402 xu = unp->unp_peercred;
1404 if (so->so_type == SOCK_STREAM)
1409 UNP_PCB_UNLOCK(unp);
1411 error = sooptcopyout(sopt, &xu, sizeof(xu));
1415 /* Unlocked read. */
1416 optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0;
1417 error = sooptcopyout(sopt, &optval, sizeof(optval));
1420 case LOCAL_CREDS_PERSISTENT:
1421 /* Unlocked read. */
1422 optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0;
1423 error = sooptcopyout(sopt, &optval, sizeof(optval));
1426 case LOCAL_CONNWAIT:
1427 /* Unlocked read. */
1428 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1429 error = sooptcopyout(sopt, &optval, sizeof(optval));
1439 switch (sopt->sopt_name) {
1441 case LOCAL_CREDS_PERSISTENT:
1442 case LOCAL_CONNWAIT:
1443 error = sooptcopyin(sopt, &optval, sizeof(optval),
1448 #define OPTSET(bit, exclusive) do { \
1449 UNP_PCB_LOCK(unp); \
1451 if ((unp->unp_flags & (exclusive)) != 0) { \
1452 UNP_PCB_UNLOCK(unp); \
1456 unp->unp_flags |= (bit); \
1458 unp->unp_flags &= ~(bit); \
1459 UNP_PCB_UNLOCK(unp); \
1462 switch (sopt->sopt_name) {
1464 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS);
1467 case LOCAL_CREDS_PERSISTENT:
1468 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT);
1471 case LOCAL_CONNWAIT:
1472 OPTSET(UNP_CONNWAIT, 0);
1481 error = ENOPROTOOPT;
1494 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1497 return (unp_connectat(AT_FDCWD, so, nam, td));
1501 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1505 struct sockaddr_un *soun;
1508 struct unpcb *unp, *unp2, *unp3;
1509 struct nameidata nd;
1510 char buf[SOCK_MAXADDRLEN];
1511 struct sockaddr *sa;
1512 cap_rights_t rights;
1516 if (nam->sa_family != AF_UNIX)
1517 return (EAFNOSUPPORT);
1518 if (nam->sa_len > sizeof(struct sockaddr_un))
1520 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1523 soun = (struct sockaddr_un *)nam;
1524 bcopy(soun->sun_path, buf, len);
1528 unp = sotounpcb(so);
1532 * Wait for connection state to stabilize. If a connection
1533 * already exists, give up. For datagram sockets, which permit
1534 * multiple consecutive connect(2) calls, upper layers are
1535 * responsible for disconnecting in advance of a subsequent
1536 * connect(2), but this is not synchronized with PCB connection
1539 * Also make sure that no threads are currently attempting to
1540 * lock the peer socket, to ensure that unp_conn cannot
1541 * transition between two valid sockets while locks are dropped.
1543 if (SOLISTENING(so))
1545 else if (unp->unp_conn != NULL)
1547 else if ((unp->unp_flags & UNP_CONNECTING) != 0) {
1551 UNP_PCB_UNLOCK(unp);
1554 if (unp->unp_pairbusy > 0) {
1555 unp->unp_flags |= UNP_WAITING;
1556 mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0);
1561 unp->unp_flags |= UNP_CONNECTING;
1562 UNP_PCB_UNLOCK(unp);
1564 connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0;
1566 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1569 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1570 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT));
1576 ASSERT_VOP_LOCKED(vp, "unp_connect");
1577 NDFREE_NOTHING(&nd);
1581 if (vp->v_type != VSOCK) {
1586 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1590 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1594 unp = sotounpcb(so);
1595 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1597 vplock = mtx_pool_find(mtxpool_sleep, vp);
1599 VOP_UNP_CONNECT(vp, &unp2);
1601 error = ECONNREFUSED;
1604 so2 = unp2->unp_socket;
1605 if (so->so_type != so2->so_type) {
1610 if (SOLISTENING(so2)) {
1611 CURVNET_SET(so2->so_vnet);
1612 so2 = sonewconn(so2, 0);
1617 error = ECONNREFUSED;
1620 unp3 = sotounpcb(so2);
1621 unp_pcb_lock_pair(unp2, unp3);
1622 if (unp2->unp_addr != NULL) {
1623 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1624 unp3->unp_addr = (struct sockaddr_un *) sa;
1628 unp_copy_peercred(td, unp3, unp, unp2);
1630 UNP_PCB_UNLOCK(unp2);
1634 * It is safe to block on the PCB lock here since unp2 is
1635 * nascent and cannot be connected to any other sockets.
1639 mac_socketpeer_set_from_socket(so, so2);
1640 mac_socketpeer_set_from_socket(so2, so);
1643 unp_pcb_lock_pair(unp, unp2);
1645 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
1646 sotounpcb(so2) == unp2,
1647 ("%s: unp2 %p so2 %p", __func__, unp2, so2));
1648 error = unp_connect2(so, so2, PRU_CONNECT);
1649 unp_pcb_unlock_pair(unp, unp2);
1658 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0,
1659 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp));
1660 unp->unp_flags &= ~UNP_CONNECTING;
1661 UNP_PCB_UNLOCK(unp);
1666 * Set socket peer credentials at connection time.
1668 * The client's PCB credentials are copied from its process structure. The
1669 * server's PCB credentials are copied from the socket on which it called
1670 * listen(2). uipc_listen cached that process's credentials at the time.
1673 unp_copy_peercred(struct thread *td, struct unpcb *client_unp,
1674 struct unpcb *server_unp, struct unpcb *listen_unp)
1676 cru2xt(td, &client_unp->unp_peercred);
1677 client_unp->unp_flags |= UNP_HAVEPC;
1679 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred,
1680 sizeof(server_unp->unp_peercred));
1681 server_unp->unp_flags |= UNP_HAVEPC;
1682 client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK);
1686 unp_connect2(struct socket *so, struct socket *so2, int req)
1691 unp = sotounpcb(so);
1692 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1693 unp2 = sotounpcb(so2);
1694 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1696 UNP_PCB_LOCK_ASSERT(unp);
1697 UNP_PCB_LOCK_ASSERT(unp2);
1698 KASSERT(unp->unp_conn == NULL,
1699 ("%s: socket %p is already connected", __func__, unp));
1701 if (so2->so_type != so->so_type)
1702 return (EPROTOTYPE);
1703 unp->unp_conn = unp2;
1706 switch (so->so_type) {
1708 UNP_REF_LIST_LOCK();
1709 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1710 UNP_REF_LIST_UNLOCK();
1715 case SOCK_SEQPACKET:
1716 KASSERT(unp2->unp_conn == NULL,
1717 ("%s: socket %p is already connected", __func__, unp2));
1718 unp2->unp_conn = unp;
1719 if (req == PRU_CONNECT &&
1720 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1728 panic("unp_connect2");
1734 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1736 struct socket *so, *so2;
1738 struct unpcb *unptmp;
1741 UNP_PCB_LOCK_ASSERT(unp);
1742 UNP_PCB_LOCK_ASSERT(unp2);
1743 KASSERT(unp->unp_conn == unp2,
1744 ("%s: unpcb %p is not connected to %p", __func__, unp, unp2));
1746 unp->unp_conn = NULL;
1747 so = unp->unp_socket;
1748 so2 = unp2->unp_socket;
1749 switch (unp->unp_socket->so_type) {
1751 UNP_REF_LIST_LOCK();
1753 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) {
1757 KASSERT(unptmp != NULL,
1758 ("%s: %p not found in reflist of %p", __func__, unp, unp2));
1760 LIST_REMOVE(unp, unp_reflink);
1761 UNP_REF_LIST_UNLOCK();
1764 so->so_state &= ~SS_ISCONNECTED;
1770 case SOCK_SEQPACKET:
1772 soisdisconnected(so);
1773 MPASS(unp2->unp_conn == unp);
1774 unp2->unp_conn = NULL;
1776 soisdisconnected(so2);
1781 unp_pcb_rele_notlast(unp);
1782 if (!unp_pcb_rele(unp))
1783 UNP_PCB_UNLOCK(unp);
1785 if (!unp_pcb_rele(unp))
1786 UNP_PCB_UNLOCK(unp);
1787 if (!unp_pcb_rele(unp2))
1788 UNP_PCB_UNLOCK(unp2);
1793 * unp_pcblist() walks the global list of struct unpcb's to generate a
1794 * pointer list, bumping the refcount on each unpcb. It then copies them out
1795 * sequentially, validating the generation number on each to see if it has
1796 * been detached. All of this is necessary because copyout() may sleep on
1800 unp_pcblist(SYSCTL_HANDLER_ARGS)
1802 struct unpcb *unp, **unp_list;
1804 struct xunpgen *xug;
1805 struct unp_head *head;
1810 switch ((intptr_t)arg1) {
1819 case SOCK_SEQPACKET:
1824 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1828 * The process of preparing the PCB list is too time-consuming and
1829 * resource-intensive to repeat twice on every request.
1831 if (req->oldptr == NULL) {
1833 req->oldidx = 2 * (sizeof *xug)
1834 + (n + n/8) * sizeof(struct xunpcb);
1838 if (req->newptr != NULL)
1842 * OK, now we're committed to doing something.
1844 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO);
1846 gencnt = unp_gencnt;
1850 xug->xug_len = sizeof *xug;
1852 xug->xug_gen = gencnt;
1853 xug->xug_sogen = so_gencnt;
1854 error = SYSCTL_OUT(req, xug, sizeof *xug);
1860 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1863 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1864 unp = LIST_NEXT(unp, unp_link)) {
1866 if (unp->unp_gencnt <= gencnt) {
1867 if (cr_cansee(req->td->td_ucred,
1868 unp->unp_socket->so_cred)) {
1869 UNP_PCB_UNLOCK(unp);
1872 unp_list[i++] = unp;
1875 UNP_PCB_UNLOCK(unp);
1878 n = i; /* In case we lost some during malloc. */
1881 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1882 for (i = 0; i < n; i++) {
1885 if (unp_pcb_rele(unp))
1888 if (unp->unp_gencnt <= gencnt) {
1889 xu->xu_len = sizeof *xu;
1890 xu->xu_unpp = (uintptr_t)unp;
1892 * XXX - need more locking here to protect against
1893 * connect/disconnect races for SMP.
1895 if (unp->unp_addr != NULL)
1896 bcopy(unp->unp_addr, &xu->xu_addr,
1897 unp->unp_addr->sun_len);
1899 bzero(&xu->xu_addr, sizeof(xu->xu_addr));
1900 if (unp->unp_conn != NULL &&
1901 unp->unp_conn->unp_addr != NULL)
1902 bcopy(unp->unp_conn->unp_addr,
1904 unp->unp_conn->unp_addr->sun_len);
1906 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr));
1907 xu->unp_vnode = (uintptr_t)unp->unp_vnode;
1908 xu->unp_conn = (uintptr_t)unp->unp_conn;
1909 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs);
1910 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink);
1911 xu->unp_gencnt = unp->unp_gencnt;
1912 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1913 UNP_PCB_UNLOCK(unp);
1914 error = SYSCTL_OUT(req, xu, sizeof *xu);
1916 UNP_PCB_UNLOCK(unp);
1922 * Give the user an updated idea of our state. If the
1923 * generation differs from what we told her before, she knows
1924 * that something happened while we were processing this
1925 * request, and it might be necessary to retry.
1927 xug->xug_gen = unp_gencnt;
1928 xug->xug_sogen = so_gencnt;
1929 xug->xug_count = unp_count;
1930 error = SYSCTL_OUT(req, xug, sizeof *xug);
1932 free(unp_list, M_TEMP);
1937 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist,
1938 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1939 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1940 "List of active local datagram sockets");
1941 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist,
1942 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1943 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1944 "List of active local stream sockets");
1945 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1946 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1947 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1948 "List of active local seqpacket sockets");
1951 unp_shutdown(struct unpcb *unp)
1956 UNP_PCB_LOCK_ASSERT(unp);
1958 unp2 = unp->unp_conn;
1959 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1960 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1961 so = unp2->unp_socket;
1968 unp_drop(struct unpcb *unp)
1974 * Regardless of whether the socket's peer dropped the connection
1975 * with this socket by aborting or disconnecting, POSIX requires
1976 * that ECONNRESET is returned.
1980 so = unp->unp_socket;
1982 so->so_error = ECONNRESET;
1983 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
1984 /* Last reference dropped in unp_disconnect(). */
1985 unp_pcb_rele_notlast(unp);
1986 unp_disconnect(unp, unp2);
1987 } else if (!unp_pcb_rele(unp)) {
1988 UNP_PCB_UNLOCK(unp);
1993 unp_freerights(struct filedescent **fdep, int fdcount)
1998 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
2000 for (i = 0; i < fdcount; i++) {
2001 fp = fdep[i]->fde_file;
2002 filecaps_free(&fdep[i]->fde_caps);
2005 free(fdep[0], M_FILECAPS);
2009 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
2011 struct thread *td = curthread; /* XXX */
2012 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
2015 struct filedesc *fdesc = td->td_proc->p_fd;
2016 struct filedescent **fdep;
2018 socklen_t clen = control->m_len, datalen;
2022 UNP_LINK_UNLOCK_ASSERT();
2025 if (controlp != NULL) /* controlp == NULL => free control messages */
2027 while (cm != NULL) {
2028 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
2032 data = CMSG_DATA(cm);
2033 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2034 if (cm->cmsg_level == SOL_SOCKET
2035 && cm->cmsg_type == SCM_RIGHTS) {
2036 newfds = datalen / sizeof(*fdep);
2041 /* If we're not outputting the descriptors free them. */
2042 if (error || controlp == NULL) {
2043 unp_freerights(fdep, newfds);
2046 FILEDESC_XLOCK(fdesc);
2049 * Now change each pointer to an fd in the global
2050 * table to an integer that is the index to the local
2051 * fd table entry that we set up to point to the
2052 * global one we are transferring.
2054 newlen = newfds * sizeof(int);
2055 *controlp = sbcreatecontrol(NULL, newlen,
2056 SCM_RIGHTS, SOL_SOCKET);
2057 if (*controlp == NULL) {
2058 FILEDESC_XUNLOCK(fdesc);
2060 unp_freerights(fdep, newfds);
2065 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2066 if (fdallocn(td, 0, fdp, newfds) != 0) {
2067 FILEDESC_XUNLOCK(fdesc);
2069 unp_freerights(fdep, newfds);
2074 for (i = 0; i < newfds; i++, fdp++) {
2075 _finstall(fdesc, fdep[i]->fde_file, *fdp,
2076 (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0,
2077 &fdep[i]->fde_caps);
2078 unp_externalize_fp(fdep[i]->fde_file);
2082 * The new type indicates that the mbuf data refers to
2083 * kernel resources that may need to be released before
2084 * the mbuf is freed.
2086 m_chtype(*controlp, MT_EXTCONTROL);
2087 FILEDESC_XUNLOCK(fdesc);
2088 free(fdep[0], M_FILECAPS);
2090 /* We can just copy anything else across. */
2091 if (error || controlp == NULL)
2093 *controlp = sbcreatecontrol(NULL, datalen,
2094 cm->cmsg_type, cm->cmsg_level);
2095 if (*controlp == NULL) {
2100 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
2103 controlp = &(*controlp)->m_next;
2106 if (CMSG_SPACE(datalen) < clen) {
2107 clen -= CMSG_SPACE(datalen);
2108 cm = (struct cmsghdr *)
2109 ((caddr_t)cm + CMSG_SPACE(datalen));
2121 unp_zone_change(void *tag)
2124 uma_zone_set_max(unp_zone, maxsockets);
2129 unp_zdtor(void *mem, int size __unused, void *arg __unused)
2135 KASSERT(LIST_EMPTY(&unp->unp_refs),
2136 ("%s: unpcb %p has lingering refs", __func__, unp));
2137 KASSERT(unp->unp_socket == NULL,
2138 ("%s: unpcb %p has socket backpointer", __func__, unp));
2139 KASSERT(unp->unp_vnode == NULL,
2140 ("%s: unpcb %p has vnode references", __func__, unp));
2141 KASSERT(unp->unp_conn == NULL,
2142 ("%s: unpcb %p is still connected", __func__, unp));
2143 KASSERT(unp->unp_addr == NULL,
2144 ("%s: unpcb %p has leaked addr", __func__, unp));
2149 unp_init(void *arg __unused)
2158 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor,
2159 NULL, NULL, UMA_ALIGN_CACHE, 0);
2160 uma_zone_set_max(unp_zone, maxsockets);
2161 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
2162 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
2163 NULL, EVENTHANDLER_PRI_ANY);
2164 LIST_INIT(&unp_dhead);
2165 LIST_INIT(&unp_shead);
2166 LIST_INIT(&unp_sphead);
2167 SLIST_INIT(&unp_defers);
2168 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
2169 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
2170 UNP_LINK_LOCK_INIT();
2171 UNP_DEFERRED_LOCK_INIT();
2173 SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL);
2176 unp_internalize_cleanup_rights(struct mbuf *control)
2183 for (m = control; m != NULL; m = m->m_next) {
2184 cp = mtod(m, struct cmsghdr *);
2185 if (cp->cmsg_level != SOL_SOCKET ||
2186 cp->cmsg_type != SCM_RIGHTS)
2188 data = CMSG_DATA(cp);
2189 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data;
2190 unp_freerights(data, datalen / sizeof(struct filedesc *));
2195 unp_internalize(struct mbuf **controlp, struct thread *td)
2197 struct mbuf *control, **initial_controlp;
2199 struct filedesc *fdesc;
2202 struct cmsgcred *cmcred;
2203 struct filedescent *fde, **fdep, *fdev;
2206 struct timespec *ts;
2208 socklen_t clen, datalen;
2209 int i, j, error, *fdp, oldfds;
2212 UNP_LINK_UNLOCK_ASSERT();
2217 control = *controlp;
2218 clen = control->m_len;
2220 initial_controlp = controlp;
2221 for (cm = mtod(control, struct cmsghdr *); cm != NULL;) {
2222 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
2223 || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
2227 data = CMSG_DATA(cm);
2228 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2230 switch (cm->cmsg_type) {
2232 * Fill in credential information.
2235 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
2236 SCM_CREDS, SOL_SOCKET);
2237 if (*controlp == NULL) {
2241 cmcred = (struct cmsgcred *)
2242 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2243 cmcred->cmcred_pid = p->p_pid;
2244 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
2245 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
2246 cmcred->cmcred_euid = td->td_ucred->cr_uid;
2247 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
2249 for (i = 0; i < cmcred->cmcred_ngroups; i++)
2250 cmcred->cmcred_groups[i] =
2251 td->td_ucred->cr_groups[i];
2255 oldfds = datalen / sizeof (int);
2259 * Check that all the FDs passed in refer to legal
2260 * files. If not, reject the entire operation.
2263 FILEDESC_SLOCK(fdesc);
2264 for (i = 0; i < oldfds; i++, fdp++) {
2265 fp = fget_noref(fdesc, *fdp);
2267 FILEDESC_SUNLOCK(fdesc);
2271 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
2272 FILEDESC_SUNLOCK(fdesc);
2279 * Now replace the integer FDs with pointers to the
2280 * file structure and capability rights.
2282 newlen = oldfds * sizeof(fdep[0]);
2283 *controlp = sbcreatecontrol(NULL, newlen,
2284 SCM_RIGHTS, SOL_SOCKET);
2285 if (*controlp == NULL) {
2286 FILEDESC_SUNLOCK(fdesc);
2291 for (i = 0; i < oldfds; i++, fdp++) {
2292 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) {
2294 for (j = 0; j < i; j++, fdp++) {
2295 fdrop(fdesc->fd_ofiles[*fdp].
2298 FILEDESC_SUNLOCK(fdesc);
2304 fdep = (struct filedescent **)
2305 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2306 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
2308 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
2309 fde = &fdesc->fd_ofiles[*fdp];
2311 fdep[i]->fde_file = fde->fde_file;
2312 filecaps_copy(&fde->fde_caps,
2313 &fdep[i]->fde_caps, true);
2314 unp_internalize_fp(fdep[i]->fde_file);
2316 FILEDESC_SUNLOCK(fdesc);
2320 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2321 SCM_TIMESTAMP, SOL_SOCKET);
2322 if (*controlp == NULL) {
2326 tv = (struct timeval *)
2327 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2332 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2333 SCM_BINTIME, SOL_SOCKET);
2334 if (*controlp == NULL) {
2338 bt = (struct bintime *)
2339 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2344 *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2345 SCM_REALTIME, SOL_SOCKET);
2346 if (*controlp == NULL) {
2350 ts = (struct timespec *)
2351 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2356 *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2357 SCM_MONOTONIC, SOL_SOCKET);
2358 if (*controlp == NULL) {
2362 ts = (struct timespec *)
2363 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2372 if (*controlp != NULL)
2373 controlp = &(*controlp)->m_next;
2374 if (CMSG_SPACE(datalen) < clen) {
2375 clen -= CMSG_SPACE(datalen);
2376 cm = (struct cmsghdr *)
2377 ((caddr_t)cm + CMSG_SPACE(datalen));
2385 if (error != 0 && initial_controlp != NULL)
2386 unp_internalize_cleanup_rights(*initial_controlp);
2391 static struct mbuf *
2392 unp_addsockcred(struct thread *td, struct mbuf *control, int mode)
2394 struct mbuf *m, *n, *n_prev;
2395 const struct cmsghdr *cm;
2396 int ngroups, i, cmsgtype;
2399 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2400 if (mode & UNP_WANTCRED_ALWAYS) {
2401 ctrlsz = SOCKCRED2SIZE(ngroups);
2402 cmsgtype = SCM_CREDS2;
2404 ctrlsz = SOCKCREDSIZE(ngroups);
2405 cmsgtype = SCM_CREDS;
2408 m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET);
2412 if (mode & UNP_WANTCRED_ALWAYS) {
2413 struct sockcred2 *sc;
2415 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
2417 sc->sc_pid = td->td_proc->p_pid;
2418 sc->sc_uid = td->td_ucred->cr_ruid;
2419 sc->sc_euid = td->td_ucred->cr_uid;
2420 sc->sc_gid = td->td_ucred->cr_rgid;
2421 sc->sc_egid = td->td_ucred->cr_gid;
2422 sc->sc_ngroups = ngroups;
2423 for (i = 0; i < sc->sc_ngroups; i++)
2424 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2426 struct sockcred *sc;
2428 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
2429 sc->sc_uid = td->td_ucred->cr_ruid;
2430 sc->sc_euid = td->td_ucred->cr_uid;
2431 sc->sc_gid = td->td_ucred->cr_rgid;
2432 sc->sc_egid = td->td_ucred->cr_gid;
2433 sc->sc_ngroups = ngroups;
2434 for (i = 0; i < sc->sc_ngroups; i++)
2435 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2439 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2440 * created SCM_CREDS control message (struct sockcred) has another
2443 if (control != NULL && cmsgtype == SCM_CREDS)
2444 for (n = control, n_prev = NULL; n != NULL;) {
2445 cm = mtod(n, struct cmsghdr *);
2446 if (cm->cmsg_level == SOL_SOCKET &&
2447 cm->cmsg_type == SCM_CREDS) {
2449 control = n->m_next;
2451 n_prev->m_next = n->m_next;
2459 /* Prepend it to the head. */
2460 m->m_next = control;
2464 static struct unpcb *
2465 fptounp(struct file *fp)
2469 if (fp->f_type != DTYPE_SOCKET)
2471 if ((so = fp->f_data) == NULL)
2473 if (so->so_proto->pr_domain != &localdomain)
2475 return sotounpcb(so);
2479 unp_discard(struct file *fp)
2481 struct unp_defer *dr;
2483 if (unp_externalize_fp(fp)) {
2484 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2486 UNP_DEFERRED_LOCK();
2487 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2488 UNP_DEFERRED_UNLOCK();
2489 atomic_add_int(&unp_defers_count, 1);
2490 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2492 closef_nothread(fp);
2496 unp_process_defers(void *arg __unused, int pending)
2498 struct unp_defer *dr;
2499 SLIST_HEAD(, unp_defer) drl;
2504 UNP_DEFERRED_LOCK();
2505 if (SLIST_FIRST(&unp_defers) == NULL) {
2506 UNP_DEFERRED_UNLOCK();
2509 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2510 UNP_DEFERRED_UNLOCK();
2512 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2513 SLIST_REMOVE_HEAD(&drl, ud_link);
2514 closef_nothread(dr->ud_fp);
2518 atomic_add_int(&unp_defers_count, -count);
2523 unp_internalize_fp(struct file *fp)
2528 if ((unp = fptounp(fp)) != NULL) {
2530 unp->unp_msgcount++;
2537 unp_externalize_fp(struct file *fp)
2543 if ((unp = fptounp(fp)) != NULL) {
2544 unp->unp_msgcount--;
2554 * unp_defer indicates whether additional work has been defered for a future
2555 * pass through unp_gc(). It is thread local and does not require explicit
2558 static int unp_marked;
2561 unp_remove_dead_ref(struct filedescent **fdep, int fdcount)
2568 * This function can only be called from the gc task.
2570 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2571 ("%s: not on gc callout", __func__));
2572 UNP_LINK_LOCK_ASSERT();
2574 for (i = 0; i < fdcount; i++) {
2575 fp = fdep[i]->fde_file;
2576 if ((unp = fptounp(fp)) == NULL)
2578 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
2585 unp_restore_undead_ref(struct filedescent **fdep, int fdcount)
2592 * This function can only be called from the gc task.
2594 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2595 ("%s: not on gc callout", __func__));
2596 UNP_LINK_LOCK_ASSERT();
2598 for (i = 0; i < fdcount; i++) {
2599 fp = fdep[i]->fde_file;
2600 if ((unp = fptounp(fp)) == NULL)
2602 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
2610 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int))
2612 struct socket *so, *soa;
2614 so = unp->unp_socket;
2616 if (SOLISTENING(so)) {
2618 * Mark all sockets in our accept queue.
2620 TAILQ_FOREACH(soa, &so->sol_comp, so_list) {
2621 if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
2623 SOCKBUF_LOCK(&soa->so_rcv);
2624 unp_scan(soa->so_rcv.sb_mb, op);
2625 SOCKBUF_UNLOCK(&soa->so_rcv);
2629 * Mark all sockets we reference with RIGHTS.
2631 if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2632 SOCKBUF_LOCK(&so->so_rcv);
2633 unp_scan(so->so_rcv.sb_mb, op);
2634 SOCKBUF_UNLOCK(&so->so_rcv);
2640 static int unp_recycled;
2641 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2642 "Number of unreachable sockets claimed by the garbage collector.");
2644 static int unp_taskcount;
2645 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2646 "Number of times the garbage collector has run.");
2648 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0,
2649 "Number of active local sockets.");
2652 unp_gc(__unused void *arg, int pending)
2654 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2656 struct unp_head **head;
2657 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */
2658 struct file *f, **unref;
2659 struct unpcb *unp, *unptmp;
2660 int i, total, unp_unreachable;
2662 LIST_INIT(&unp_deadhead);
2666 * First determine which sockets may be in cycles.
2668 unp_unreachable = 0;
2670 for (head = heads; *head != NULL; head++)
2671 LIST_FOREACH(unp, *head, unp_link) {
2672 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0,
2673 ("%s: unp %p has unexpected gc flags 0x%x",
2674 __func__, unp, (unsigned int)unp->unp_gcflag));
2679 * Check for an unreachable socket potentially in a
2680 * cycle. It must be in a queue as indicated by
2681 * msgcount, and this must equal the file reference
2682 * count. Note that when msgcount is 0 the file is
2685 if (f != NULL && unp->unp_msgcount != 0 &&
2686 refcount_load(&f->f_count) == unp->unp_msgcount) {
2687 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead);
2688 unp->unp_gcflag |= UNPGC_DEAD;
2689 unp->unp_gcrefs = unp->unp_msgcount;
2695 * Scan all sockets previously marked as potentially being in a cycle
2696 * and remove the references each socket holds on any UNPGC_DEAD
2697 * sockets in its queue. After this step, all remaining references on
2698 * sockets marked UNPGC_DEAD should not be part of any cycle.
2700 LIST_FOREACH(unp, &unp_deadhead, unp_dead)
2701 unp_gc_scan(unp, unp_remove_dead_ref);
2704 * If a socket still has a non-negative refcount, it cannot be in a
2705 * cycle. In this case increment refcount of all children iteratively.
2706 * Stop the scan once we do a complete loop without discovering
2707 * a new reachable socket.
2711 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp)
2712 if (unp->unp_gcrefs > 0) {
2713 unp->unp_gcflag &= ~UNPGC_DEAD;
2714 LIST_REMOVE(unp, unp_dead);
2715 KASSERT(unp_unreachable > 0,
2716 ("%s: unp_unreachable underflow.",
2719 unp_gc_scan(unp, unp_restore_undead_ref);
2721 } while (unp_marked);
2725 if (unp_unreachable == 0)
2729 * Allocate space for a local array of dead unpcbs.
2730 * TODO: can this path be simplified by instead using the local
2731 * dead list at unp_deadhead, after taking out references
2732 * on the file object and/or unpcb and dropping the link lock?
2734 unref = malloc(unp_unreachable * sizeof(struct file *),
2738 * Iterate looking for sockets which have been specifically marked
2739 * as unreachable and store them locally.
2743 LIST_FOREACH(unp, &unp_deadhead, unp_dead) {
2744 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0,
2745 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp));
2746 unp->unp_gcflag &= ~UNPGC_DEAD;
2748 if (unp->unp_msgcount == 0 || f == NULL ||
2749 refcount_load(&f->f_count) != unp->unp_msgcount ||
2753 KASSERT(total <= unp_unreachable,
2754 ("%s: incorrect unreachable count.", __func__));
2759 * Now flush all sockets, free'ing rights. This will free the
2760 * struct files associated with these sockets but leave each socket
2761 * with one remaining ref.
2763 for (i = 0; i < total; i++) {
2766 so = unref[i]->f_data;
2767 CURVNET_SET(so->so_vnet);
2773 * And finally release the sockets so they can be reclaimed.
2775 for (i = 0; i < total; i++)
2776 fdrop(unref[i], NULL);
2777 unp_recycled += total;
2778 free(unref, M_TEMP);
2782 unp_dispose_mbuf(struct mbuf *m)
2786 unp_scan(m, unp_freerights);
2790 * Synchronize against unp_gc, which can trip over data as we are freeing it.
2793 unp_dispose(struct socket *so)
2797 unp = sotounpcb(so);
2799 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2801 if (!SOLISTENING(so))
2802 unp_dispose_mbuf(so->so_rcv.sb_mb);
2806 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2811 socklen_t clen, datalen;
2813 while (m0 != NULL) {
2814 for (m = m0; m; m = m->m_next) {
2815 if (m->m_type != MT_CONTROL)
2818 cm = mtod(m, struct cmsghdr *);
2821 while (cm != NULL) {
2822 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2825 data = CMSG_DATA(cm);
2826 datalen = (caddr_t)cm + cm->cmsg_len
2829 if (cm->cmsg_level == SOL_SOCKET &&
2830 cm->cmsg_type == SCM_RIGHTS) {
2831 (*op)(data, datalen /
2832 sizeof(struct filedescent *));
2835 if (CMSG_SPACE(datalen) < clen) {
2836 clen -= CMSG_SPACE(datalen);
2837 cm = (struct cmsghdr *)
2838 ((caddr_t)cm + CMSG_SPACE(datalen));
2850 * A helper function called by VFS before socket-type vnode reclamation.
2851 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2855 vfs_unp_reclaim(struct vnode *vp)
2861 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2862 KASSERT(vp->v_type == VSOCK,
2863 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2866 vplock = mtx_pool_find(mtxpool_sleep, vp);
2868 VOP_UNP_CONNECT(vp, &unp);
2872 if (unp->unp_vnode == vp) {
2874 unp->unp_vnode = NULL;
2877 UNP_PCB_UNLOCK(unp);
2886 db_print_indent(int indent)
2890 for (i = 0; i < indent; i++)
2895 db_print_unpflags(int unp_flags)
2900 if (unp_flags & UNP_HAVEPC) {
2901 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2904 if (unp_flags & UNP_WANTCRED_ALWAYS) {
2905 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : "");
2908 if (unp_flags & UNP_WANTCRED_ONESHOT) {
2909 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : "");
2912 if (unp_flags & UNP_CONNWAIT) {
2913 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2916 if (unp_flags & UNP_CONNECTING) {
2917 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2920 if (unp_flags & UNP_BINDING) {
2921 db_printf("%sUNP_BINDING", comma ? ", " : "");
2927 db_print_xucred(int indent, struct xucred *xu)
2931 db_print_indent(indent);
2932 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n",
2933 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups);
2934 db_print_indent(indent);
2935 db_printf("cr_groups: ");
2937 for (i = 0; i < xu->cr_ngroups; i++) {
2938 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2945 db_print_unprefs(int indent, struct unp_head *uh)
2951 LIST_FOREACH(unp, uh, unp_reflink) {
2952 if (counter % 4 == 0)
2953 db_print_indent(indent);
2954 db_printf("%p ", unp);
2955 if (counter % 4 == 3)
2959 if (counter != 0 && counter % 4 != 0)
2963 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2968 db_printf("usage: show unpcb <addr>\n");
2971 unp = (struct unpcb *)addr;
2973 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2976 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2979 db_printf("unp_refs:\n");
2980 db_print_unprefs(2, &unp->unp_refs);
2982 /* XXXRW: Would be nice to print the full address, if any. */
2983 db_printf("unp_addr: %p\n", unp->unp_addr);
2985 db_printf("unp_gencnt: %llu\n",
2986 (unsigned long long)unp->unp_gencnt);
2988 db_printf("unp_flags: %x (", unp->unp_flags);
2989 db_print_unpflags(unp->unp_flags);
2992 db_printf("unp_peercred:\n");
2993 db_print_xucred(2, &unp->unp_peercred);
2995 db_printf("unp_refcount: %u\n", unp->unp_refcount);